Introduction
For a centrally symmetric convex disk D ⊂ R 2 D \subset\mathbb{R}^{2} D ⊂ R 2 with nonempty interior, let
δ L ( D ) = sup L ∣ D ∣ det L , \delta_{L}(D)=\sup_{\mathcal{L}}\frac{|D|}{\det\mathcal{L}}, δ L ( D ) = L sup det L ∣ D ∣ ,
where the supremum is over lattices whose translates of int D \operatorname{int} D int D are pairwise disjoint. The Reinhardt problem asks for the minimum of δ L ( D ) \delta_{L}(D) δ L ( D ) . Nonsingular affine transformations preserve this density.
In 1934, Reinhardt proposed the smoothed regular octagon as a minimizer [6]. It is obtained by rounding the corners of a regular octagon with arcs of hyperbolas. The problem is unusual even among planar packing questions: the disk that minimizes its best lattice packing density is neither a polygon nor an ellipse.
Mahler developed a variational approach in 1947 [4]. His programme separates two questions. The first is whether every minimizing disk is a smoothed polygon, with finitely many straight and hyperbolic boundary pieces. The second is whether that polygon must be the smoothed regular octagon. These are commonly called Mahler’s first conjecture and the Reinhardt conjecture, respectively. Nazarov proved local extremality of the Reinhardt octagon in 1986; the English translation appeared in 1988 [5]. This established the conjectured disk as a local minimizer, while leaving the global problem unresolved.
Hales gave a variational formulation [1] and subsequently formulated the problem as an optimal-control problem [2]. In this formulation, a constant vertex control traces one arc of a smoothed polygon. An obstruction to a finite description is the possible accumulation of infinitely many switches in finite time. Hales and Vajjha excluded this behavior and proved Mahler’s first conjecture [3, Theorem 16.3.1]. Their theorem applies to every original global minimizer itself, and supplies a finite positive control representation. It does not supply a numerical bound on the number of its arcs.
We prove the second conjecture, including the full equality statement.
Theorem 1.1 (The Reinhardt conjecture). Let D o c t D_{\mathrm{oct}} D oct be the smoothed regular octagon. Every centrally symmetric convex disk D ⊂ R 2 D \subset\mathbb{R}^{2} D ⊂ R 2 with nonempty interior satisfies
δ L ( D ) ≥ δ L ( D o c t ) = 8 − 4 2 − log 2 2 2 − 1 . \delta_{L}(D) \ge\delta_{L}(D_{\mathrm{oct}}) = \frac{8-4\sqrt{2}-\log2}{2\sqrt{2}-1}. δ L ( D ) ≥ δ L ( D oct ) = 2 2 − 1 8 − 4 2 − log 2 . Equality holds precisely when D D D is a nonsingular affine image of D o c t D_{\mathrm{oct}} D oct .
The proof begins by making qualitative finiteness quantitative. In § 3.6 , a uniform positive cost for every thirteen consecutive control arcs gives the explicit bound N < 26 ⋅ 2 160 < 2 165 N < 26 \cdot2^{160} < 2^{165} N < 26 ⋅ 2 160 < 2 165 for the number N N N of positive maximal pieces at a prescribed fundamental cut. The estimate uses the geometric bounds and switching equations of the original minimizing problem. It precedes the sharper inequalities adapted to particular switching words.
The local estimates of Part I provide lower costs for forward pairs, five-arc sources, returned supports and larger cores. The capacity theorem, Theorem 6.2 , combines these estimates while charging each physical arc with total weight at most one. Applied to maximal runs of positive and negative increments, they first give B ≤ 20 B \le20 B ≤ 20 , hence N ≤ 21 N \le21 N ≤ 21 , in § 4.2 ; here B B B is the number of essential switches. The twist and switching restrictions then leave 1355 necessary cyclic words. Proposition 17.1 excludes 1334 by exact rational combinations of local costs, leaving 21 words and the sharper bounds B ≤ 13 B \le13 B ≤ 13 , N ≤ 14 N \le14 N ≤ 14 . The all-length run argument in § 4.6 –§ 4.10 obtains the same list directly. The numbers 1355 and 21 count candidate words, rather than arcs.
The passage from those twenty-one words to six types uses further geometry. § 4.13 –§ 4.23 combine returned and crossed-core cost bounds with comparisons at the actual arc endpoints, a conditional bound for the cost of a central triple, and exhaustive alternatives for the necessary equations. Theorem 24.1 leaves precisely R01–R06 as necessary types: three have eight essential arcs and three have nine. Parts III and IV exclude them individually, using their closure equations, geometric reductions and rigorous finite covers. The short-chain rigidity argument in Appendix A then identifies the smoothed regular octagon and gives the equality statement. Appendix H specifies the mathematical inputs and checking procedures for the computational steps; the accompanying reader archive supplies their algorithms and fixed data.
Acknowledgments
The author proposed the argument giving an explicit finite upper bound for the number of control arcs. ChatGPT dot carried out the sustained exploration of local cost lower bounds. ChatGPT 6.1 Sol carried out the proofs of the six remaining types R01–R06 and the organization of the manuscript. The author retains responsibility for the mathematical claims and their presentation.
The author thanks Chuanming Zong and Thomas Hales for the seminar discussions around 2015 through which they introduced him to this problem. For more than a decade, leaving this conjecture unresolved made him restless. The purpose of this paper is to show that the conjecture can be resolved. The present proof is long, and the author hopes that further work will lead to a shorter and more beautiful proof.
Part I. Geometry, sources, and local costs
The geometric problem and its normalization
The finite reduction is applied to a global minimizer of the original geometric problem. We first specify the trajectory, the full lift, and the changes of coordinates under which the local estimates will be used. The foundational interfaces are stated here; their detailed proofs appear after the finite assembly, and are referenced explicitly below.
The original minimizing class
For a centrally symmetric convex disk D D D with nonempty interior, write
δ L ( D ) = sup L ∣ D ∣ det L . \delta_L(D)=\sup_{\mathcal{L}}\frac{|D|}{\det\mathcal{L}}. δ L ( D ) = L sup det L ∣ D ∣ .
where the supremum is over lattices whose translates of D D D have pairwise disjoint interiors. An original global minimizer is a disk attaining the infimum of this quantity over all centrally symmetric convex disks. The density is unchanged by nonsingular affine transformations. The normalization used here fixes the area of a minimum-area circumscribed centrally symmetric hexagon to 2 3 2\sqrt{3} 2 3 and represents its moving frame in S L 2 ( R ) \mathrm{SL}_2(\mathbb{R}) SL 2 ( R ) .
The geometric formulation, existence and reconstruction results are the inputs of [2, Sections 2.2–2.3] and [3, Chapters 2–3]. In particular, the regular representation of an arbitrary original minimizer uses [3, Lemma 2.2.5, Lemmas 2.4.2–2.4.3, and Proposition 3.1.9]. The statements below concern that minimizing disk and its actual trajectory, rather than merely an equal-cost representative of a relaxed problem.
Set
D ∗ = diag ( 3 1 / 4 , 3 − 1 / 4 ) , J = ( 0 − 1 1 0 ) , R ∗ = 1 2 ( 1 − 3 1 1 ) = D ∗ e π J / 3 D ∗ − 1 . D_*=\operatorname{diag}(3^{1/4},3^{-1/4}),\qquad
J=\begin{pmatrix}0&-1\\1&0\end{pmatrix},\qquad
R_*=\frac{1}{2}\begin{pmatrix}1&-3\\1&1\end{pmatrix}=D_*e^{\pi J/3}D_*^{-1}. D ∗ = diag ( 3 1/4 , 3 − 1/4 ) , J = ( 0 1 − 1 0 ) , R ∗ = 2 1 ( 1 1 − 3 1 ) = D ∗ e π J /3 D ∗ − 1 .
For W = ( u , v ) W=(u,v) W = ( u , v ) , put
Ω = { ( u , v ) : − 1 < u < 1 , v > 0 , u 2 + v 2 > 1 } , S = u 2 + v 2 − 1. (1) \Omega=\{(u,v):-1<u<1,\ v>0,\ u^2+v^2>1\},\qquad S=u^2+v^2-1. \tag*{(1)} Ω = {( u , v ) : − 1 < u < 1 , v > 0 , u 2 + v 2 > 1 } , S = u 2 + v 2 − 1. (1)
The three vertex fields and the frame velocity are
F 0 = ( v , v 2 1 + u ) , F 1 = ( − v ( 1 − u 2 + v 2 ) S , 2 u v 2 S ) , F 2 = ( v , v 2 u − 1 ) , F_0=\left(v,\frac{v^2}{1+u}\right),\qquad
F_1=\left(-\frac{v(1-u^2+v^2)}{S},\frac{2uv^2}{S}\right),\qquad
F_2=\left(v,\frac{v^2}{u-1}\right), F 0 = ( v , 1 + u v 2 ) , F 1 = ( − S v ( 1 − u 2 + v 2 ) , S 2 u v 2 ) , F 2 = ( v , u − 1 v 2 ) ,
X ( W ) = 1 v ( u − u 2 − v 2 1 − u ) , A ∗ = ( 0 3 / 2 − 1 / 2 0 ) . X(W)=\frac{1}{v}\begin{pmatrix}u&-u^2-v^2\\1&-u\end{pmatrix},\qquad
A_*=\begin{pmatrix}0&3/2\\-1/2&0\end{pmatrix}. X ( W ) = v 1 ( u 1 − u 2 − v 2 − u ) , A ∗ = ( 0 − 1/2 3/2 0 ) .
The admissible controls are all measurable maps into the full triangle Δ = { ρ ∈ R 3 : ρ j ≥ 0 , ∑ j ρ j = 1 } \Delta=\{\rho\in\mathbb{R}^3:\rho_j\geq0,\ \sum_j\rho_j=1\} Δ = { ρ ∈ R 3 : ρ j ≥ 0 , ∑ j ρ j = 1 } . The equations and physical cost are
W ′ = ∑ j = 0 2 ρ j F j ( W ) , G ′ = G X ( W ) , G ( t ) ∈ S L 2 ( R ) , (2) W'=\sum_{j=0}^{2}\rho_jF_j(W),\qquad G'=GX(W),\qquad G(t)\in\mathrm{SL}_2(\mathbb{R}), \tag*{(2)} W ′ = j = 0 ∑ 2 ρ j F j ( W ) , G ′ = GX ( W ) , G ( t ) ∈ SL 2 ( R ) , (2)
ℓ ( W ) = tr ( A ∗ X ( W ) ) = 3 + u 2 + v 2 2 v , C = ∫ 0 T ℓ ( W ( t ) ) d t , ∣ D ∣ = 3 C . (3) \ell(W)=\operatorname{tr}(A_*X(W))=\frac{3+u^2+v^2}{2v},\qquad
C=\int_0^T\ell(W(t))\,\mathrm{d}t,\qquad|D|=\sqrt{3}C. \tag*{(3)} ℓ ( W ) = tr ( A ∗ X ( W )) = 2 v 3 + u 2 + v 2 , C = ∫ 0 T ℓ ( W ( t )) d t , ∣ D ∣ = 3 C . (3)
Here t t t is physical time; the identity X 2 = − I X^2=-I X 2 = − I fixes its scale. The six curves
G ( t ) q j , q j = D ∗ ( cos ( j π / 3 ) , sin ( j π / 3 ) ) t , 0 ≤ j ≤ 5 , G(t)q_j,\qquad q_j=D_*\left(\cos(j\pi/3),\sin(j\pi/3)\right)^t,\qquad0\le j\le5, G ( t ) q j , q j = D ∗ ( cos ( j π /3 ) , sin ( j π /3 ) ) t , 0 ≤ j ≤ 5 ,
reconstruct the normalized boundary. The group and state endpoints are
G ( 0 ) = I , G ( T ) = R ∗ , W ( T ) = R ∗ − 1 ⋅ W ( 0 ) , (4) G(0)=I,\qquad G(T)=R_*,\qquad W(T)=R_*^{-1}\cdot W(0), \tag*{(4)} G ( 0 ) = I , G ( T ) = R ∗ , W ( T ) = R ∗ − 1 ⋅ W ( 0 ) , (4)
with the usual fractional linear action on u + i v u+iv u + i v . There is also a real winding condition: the path D ∗ − 1 G ( t ) D ∗ D_*^{-1}G(t)D_* D ∗ − 1 G ( t ) D ∗ is homotopic with endpoints fixed to e π J t / ( 3 T ) e^{\pi Jt/(3T)} e π J t / ( 3 T ) . Thus its lifted rotation has increment π / 3 \pi/3 π /3 , with no additional whole turn. Closing only W W W , closing only a matrix trace, or imposing the angle merely modulo 2 π 2\pi 2 π would change the problem.
For clarity, the weights in (2.3) preserve the original measurable-control class. If μ \mu μ denotes the original barycentric control, then
( d 0 , d 1 , d 2 ) = ( 1 + u v , S 2 v , 1 − u v ) , ρ j = μ j d j ∑ k μ k d k , μ j = ρ j / d j ∑ k ρ k / d k . (5) (d_0,d_1,d_2)=\left(\frac{1+u}{v},\frac{S}{2v},\frac{1-u}{v}\right),\qquad
\rho_j=\frac{\mu_jd_j}{\sum_k\mu_kd_k},\qquad
\mu_j=\frac{\rho_j/d_j}{\sum_k\rho_k/d_k}. \tag*{(5)} ( d 0 , d 1 , d 2 ) = ( v 1 + u , 2 v S , v 1 − u ) , ρ j = ∑ k μ k d k μ j d j , μ j = ∑ k ρ k / d k ρ j / d j . (5)
All d j d_j d j are positive on Ω \Omega Ω , so these maps are measurable inverses and preserve the vertices. The half-plane coordinates of [3] are ( u , v ) / 3 (u,v)/\sqrt{3} ( u , v ) / 3 ; its control α \alpha α corresponds to μ = ( α 2 , α 0 , α 1 ) \mu=(\alpha_2,\alpha_0,\alpha_1) μ = ( α 2 , α 0 , α 1 ) . The matrix conjugation and the commutator identity establishing this convention are given later in equation (A.1). No change of physical time is involved.
Compactness and admissible replacements
Theorem 2.1 (Geometric compactness). Every normalized original global minimizer has its state trajectory in the compact subset K K K of Ω \Omega Ω specified by the following six inequalities:
v ≤ 15 ( 1 + u ) , v ≤ 15 ( 1 − u ) , S ≥ 2 v / 15 , v ≤ 9 3 / 2 , v\le15(1+u),\qquad v\le15(1-u),\qquad S\ge2v/15,\qquad v\le9\sqrt{3}/2, v ≤ 15 ( 1 + u ) , v ≤ 15 ( 1 − u ) , S ≥ 2 v /15 , v ≤ 9 3 /2 , ( u + 1 ) 2 + v 2 ≥ 8 v 9 3 , ( u − 1 ) 2 + v 2 ≥ 8 v 9 3 . (6) (u+1)^2+v^2\ge\frac{8v}{9\sqrt{3}},\qquad(u-1)^2+v^2\ge\frac{8v}{9\sqrt{3}}. \tag*{(6)} ( u + 1 ) 2 + v 2 ≥ 9 3 8 v , ( u − 1 ) 2 + v 2 ≥ 9 3 8 v . (6) On this set, the rates a = v / ( 1 + u ) a=v/(1+u) a = v / ( 1 + u ) , b = v / ( 1 − u ) b=v/(1-u) b = v / ( 1 − u ) and c = 2 v / S c=2v/S c = 2 v / S obey
0 < a , b , c ≤ 15 , a + b , a + c , b + c ≤ 750 37 , 99 200 ≤ v ≤ 39 5 . (7) 0<a,b,c\le15,\qquad a+b,a+c,b+c\le\frac{750}{37},\qquad\frac{99}{200}\le v\le\frac{39}{5}. \tag*{(7)} 0 < a , b , c ≤ 15 , a + b , a + c , b + c ≤ 37 750 , 200 99 ≤ v ≤ 5 39 . (7) The compactness assertion is exactly [3, Definition 5.0.1 and Theorem 5.0.2], written in the present coordinates; the elementary consequences (2.8) are proved in Lemma A.1. These inequalities constrain the original minimizing trajectory. They are not additional restrictions on a local competitor: the admissible replacement problem remains the full control triangle in Ω \Omega Ω .
Lemma 2.2 (Five-endpoint replacement). Let [ a , b ] [a,b] [ a , b ] be a window of an admissible trajectory. A sufficiently small family of solutions of (2.3), with states in Ω \Omega Ω and controls in Δ \Delta Δ , can replace this window if it has the same entrance and exit values of both W W W and G G G and converges uniformly to the original window after orientation-preserving reparametrization. Its duration may vary and tend to b − a b-a b − a . The resulting curve is an admissible convex competitor, and its area change is 3 \sqrt{3} 3 times the window-cost change.
The five endpoint coordinates are the three group coordinates and the two state coordinates. The proof in Section A.1 checks curvature, real winding, and variable-duration attachment. The five-constraint viewpoint also appears in [1, Section 7.6]. Durations are varied about positive arcs with positive caps; signed analytic extensions used for differentiation are not asserted to be admissible competitors.
A full maximizing lift and its normalization.
Definition 2.3. A full maximizing lift on a genuine minimizing interval consists of the constant cost multiplier η ≥ 0 \eta\ge0 η ≥ 0 , both state covectors p = ( p u , p v ) t p=(p_u,p_v)^t p = ( p u , p v ) t , and the entire traceless matrix adjoint Λ \Lambda Λ , not all identically zero, supplied by the free-terminal-time maximum principle for that interval. Put
Z = Λ − η A ∗ , U = tr ( Z X ) , H j = p t F j + U . Z=\Lambda-\eta A_*, \qquad U=\operatorname{tr}(ZX), \qquad H_j=p^tF_j+U. Z = Λ − η A ∗ , U = tr ( Z X ) , H j = p t F j + U . The equations and maximizing conditions used below are
p ′ = − ∑ j ρ j ( D F j ) t p − ∇ W U , Λ ′ = [ Λ , X ] , max j H j = 0 , ρ j > 0 ⟹ H j = 0 almost everywhere. (8) \begin{aligned}
p'&=-\sum_j\rho_j(DF_j)^tp-\nabla_WU,\qquad\Lambda'=[\Lambda,X],\\
\max_j H_j&=0,\qquad\rho_j>0\Longrightarrow H_j=0\quad\text{almost everywhere.} \tag*{(8)}
\end{aligned} p ′ j max H j = − j ∑ ρ j ( D F j ) t p − ∇ W U , Λ ′ = [ Λ , X ] , = 0 , ρ j > 0 ⟹ H j = 0 almost everywhere. (8) The lift is normal when η > 0 \eta>0 η > 0 . Its normalization means dividing η \eta η , both components of p p p , and every entry of Λ \Lambda Λ by that same positive number, so that η = 1 \eta=1 η = 1 .
The source is [3, Sections 6.2–6.8], with matrix pairing tr ( Λ G − 1 δ G ) \operatorname{tr}(\Lambda G^{-1}\delta G) tr ( Λ G − 1 δ G ) . There is no extra endpoint condition fixing T T T . In particular an isolated binary arc does not supply a freely selectable substitute for the full lift.
It is useful to fix the switching convention. With g 0 = 1 − u 2 g_0=1-u^2 g 0 = 1 − u 2 , set
α 0 = ( 0 , S / v ) , \alpha_0=(0,S/v), α 0 = ( 0 , S / v ) ,
α 1 = ( − g 0 , ( 1 − u ) ( ( 1 + u ) 2 − v 2 ) 2 v ) , (9) \alpha_1=\left(-g_0,\frac{(1-u)((1+u)^2-v^2)}{2v}\right), \tag*{(9)} α 1 = ( − g 0 , 2 v ( 1 − u ) (( 1 + u ) 2 − v 2 ) ) , (9)
α 2 = ( g 0 , ( 1 + u ) ( ( 1 − u ) 2 − v 2 ) 2 v ) , σ i = α i p . \alpha_2=\left(g_0,\frac{(1+u)((1-u)^2-v^2)}{2v}\right),\qquad\sigma_i=\alpha_i p. α 2 = ( g 0 , 2 v ( 1 + u ) (( 1 − u ) 2 − v 2 ) ) , σ i = α i p .
Their sum vanishes, and with β = g 0 S / ( 2 v 3 ) > 0 \beta=g_0S/(2v^3)>0 β = g 0 S / ( 2 v 3 ) > 0 ,
H 0 − H 2 = σ 0 / β , H 1 − H 0 = σ 1 / β , H 2 − H 1 = σ 2 / β . (10) H_0-H_2=\sigma_0/\beta,\qquad H_1-H_0=\sigma_1/\beta,\qquad H_2-H_1=\sigma_2/\beta. \tag*{(10)} H 0 − H 2 = σ 0 / β , H 1 − H 0 = σ 1 / β , H 2 − H 1 = σ 2 / β . (10)
For measurable maximizing controls the σ i \sigma_i σ i are C 1 C^1 C 1 with absolutely continuous first derivatives (Lemma A.2 ). Thus the contact derivatives in later arguments belong to the same continuous switching response on the two sides of an actual switch.
Proposition 2.4 (Switching restrictions and normality). The following statements hold for the actual positive maximal control arcs.
(1) The words 2020, 0101, and 1212 cannot occur on an original global minimizer, without a normality assumption.
(2) On an interval containing six consecutive positive maximal arcs and avoiding those three words, every selected nonzero full free-time maximizing lift is normal.
(3) The words 202101, 010212, and 121020 cannot occur on a full maximizing lift with positive arcs and caps. Their durations may be any positive numbers.
These are Lemma A.4 , Theorem A.8 , and Lemma A.9 , respectively. The first proof uses an independent edge problem; it does not presume that its edge multiplier is a projection of a chosen full lift. The second proves normality on the very interval later used, without a compactness or numerical cost hypothesis. In all applications with B ≥ 8 B\ge8 B ≥ 8 below, the genuine containing fundamental interval contains six complete positive arcs, so a selected full lift can be jointly normalized before any normal identity is used.
The determinant
C = det Λ = − 1 2 tr ( Λ 2 ) (11) \mathcal{C}=\det\Lambda=-\frac{1}{2}\operatorname{tr}(\Lambda^2) \tag*{(11)} C = det Λ = − 2 1 tr ( Λ 2 ) (11)
is constant along one lift. It is distinct from the physical cost C C C . The harmonic estimate in Theorem A.12 has the following precise scope: for a normal full lift on a genuine fundamental interval, jointly normalized to η = 1 \eta=1 η = 1 , if C > 0 \mathcal{C}>0 C > 0 and tr ( Λ X ) > 0 \operatorname{tr}(\Lambda X)>0 tr ( Λ X ) > 0 at one actual point, then
C ≤ 3 4 ( 2 ∣ D ∣ π − 1 ) 2 < 29025933167887355473205403 39478417604337498218323684 . (12) \mathcal{C}\leq\frac{3}{4}\left(\frac{2|D|}{\pi}-1\right)^2<\frac{29025933167887355473205403}{39478417604337498218323684}. \tag*{(12)} C ≤ 4 3 ( π 2∣ D ∣ − 1 ) 2 < 39478417604337498218323684 29025933167887355473205403 . (12)
The last inequality uses the global competitor comparison below. Normality, positive determinant and positive trace must all concern the same selected lift. Neither positive determinant nor the future-sheet condition is presumed for an arbitrary normal extremal. The proof uses only the lawful three-copy transport of the matrix adjoint and never assumes a periodic state covector.
Whole symmetries and genuine changes of cut
Proposition 2.5 (Lawful transport). Cyclic relabeling, and time reversal combined with j ↦ k − j j\mapsto k-j j ↦ k − j , preserve the problem when applied to the state, frame and whole lift together. An odd permutation of the labels alone is not a permitted transformation.
Here are explicit formulas fixing this convention. Put Φ ( w ) = ( w + 3 ) / ( 1 − w ) = R ∗ − 1 ⋅ w \Phi(w)=(w+3)/(1-w)=R_*^{-1}\cdot w Φ ( w ) = ( w + 3 ) / ( 1 − w ) = R ∗ − 1 ⋅ w . A cyclic phase acts by
W n e w = Φ W , G n e w = R ∗ − 1 G R ∗ , p n e w = ( D Φ ) − t p , Λ n e w = R ∗ − 1 Λ R ∗ , j n e w = j + 1. W_{\mathrm{new}}=\Phi W,\qquad G_{\mathrm{new}}=R_*^{-1}GR_*,\qquad p_{\mathrm{new}}=(D\Phi)^{-t}p,\qquad\Lambda_{\mathrm{new}}=R_*^{-1}\Lambda R_*,\qquad j_{\mathrm{new}}=j+1. W new = Φ W , G new = R ∗ − 1 G R ∗ , p new = ( D Φ ) − t p , Λ new = R ∗ − 1 Λ R ∗ , j new = j + 1.
It preserves η \eta η , ℓ \ell ℓ , the Hamiltonians and physical time, since D Φ F j = F j + 1 ∘ Φ D\Phi F_j=F_{j+1}\circ\Phi D Φ F j = F j + 1 ∘ Φ and X ( Φ W ) = R ∗ − 1 X ( W ) R ∗ X(\Phi W)=R_*^{-1}X(W)R_* X ( Φ W ) = R ∗ − 1 X ( W ) R ∗ . For a full reflection and time reversal let ψ ( u , v ) = ( − u , v ) \psi(u,v)=(-u,v) ψ ( u , v ) = ( − u , v ) and Q = diag ( 1 , − 1 ) Q=\operatorname{diag}(1,-1) Q = diag ( 1 , − 1 ) . With every old quantity on the right evaluated at T − t T-t T − t , set
W n e w ( t ) = ψ W ( T − t ) , G n e w ( t ) = Q G ( T ) − 1 G ( T − t ) Q , W_{\mathrm{new}}(t)=\psi W(T-t),\qquad G_{\mathrm{new}}(t)=QG(T)^{-1}G(T-t)Q, W new ( t ) = ψ W ( T − t ) , G new ( t ) = QG ( T ) − 1 G ( T − t ) Q ,
p n e w ( t ) = − ( D ψ ) − t p ( T − t ) , Λ n e w ( t ) = − Q Λ ( T − t ) Q , j n e w ( t ) = 2 − j ( T − t ) . (13) p_{\mathrm{new}}(t)=-(D\psi)^{-t}p(T-t),\qquad\Lambda_{\mathrm{new}}(t)=-Q\Lambda(T-t)Q,\qquad j_{\mathrm{new}}(t)=2-j(T-t). \tag*{(13)} p new ( t ) = − ( D ψ ) − t p ( T − t ) , Λ new ( t ) = − Q Λ ( T − t ) Q , j new ( t ) = 2 − j ( T − t ) . (13)
These formulas retain η \eta η and physical cost. The identities X ( ψ W ) = − Q X ( W ) Q X(\psi W)=-QX(W)Q X ( ψ W ) = − QX ( W ) Q , Q R ∗ Q = R ∗ − 1 QR_*Q=R_*^{-1} Q R ∗ Q = R ∗ − 1 , and ψ Φ ψ = Φ − 1 \psi\Phi\psi=\Phi^{-1} ψ Φ ψ = Φ − 1 verify the equations and endpoints. Reflection and reversal also preserve the prescribed real homotopy class. Composing with a cyclic phase gives each j ↦ k − j j\mapsto k-j j ↦ k − j . The same formulas, with the appropriate entrance normalization of the frame, apply to a local window. Each cap, return, contact and charged arc is transported along with the complete source.
The original trajectory has the twisted extension
G e ( t + k T ) = G ( t ) R ∗ k , W e ( t + k T ) = Φ k W ( t ) , G_e(t+kT)=G(t)R_*^k,\qquad W_e(t+kT)=\Phi^kW(t), G e ( t + k T ) = G ( t ) R ∗ k , W e ( t + k T ) = Φ k W ( t ) ,
ρ j + k , e ( t + k T ) = ρ j ( t ) , 0 ≤ t < T , k ∈ Z . (14) \rho_{j+k,e}(t+kT)=\rho_j(t),\qquad0\leq t<T,\quad k\in\mathbb{Z}. \tag*{(14)} ρ j + k , e ( t + k T ) = ρ j ( t ) , 0 ≤ t < T , k ∈ Z . (14)
No state costate is extended by this definition.
Lemma 2.6 (Genuine reanchoring). For any s ∈ R s\in\mathbb{R} s ∈ R , the trajectory
G s ( t ) = G e ( s ) − 1 G e ( s + t ) , W s ( t ) = W e ( s + t ) , 0 ≤ t ≤ T , G_s(t)=G_e(s)^{-1}G_e(s+t),\qquad W_s(t)=W_e(s+t),\qquad0\leq t\leq T, G s ( t ) = G e ( s ) − 1 G e ( s + t ) , W s ( t ) = W e ( s + t ) , 0 ≤ t ≤ T , is an original global minimizer with the same physical cost and the original endpoints and real winding. It therefore has its own nonzero full maximizing lift.
The proof in Section A.1 identifies the same boundary under an area-preserving transformation and verifies the real homotopy. One must select the lift on the new genuine interval; independently selected state covectors are not identified across a change of cut.
Physical arc counts and the global budget
The qualitative finiteness theorem [3, Theorem 16.3.1] states that every normalized original global minimizer itself is a vertex control with finitely many switches, up to a null set. Merge adjacent equal controls and discard zero-duration pieces. Let B B B be the number of essential switches on the twisted physical circle. It is also the number of complete physical arcs in that circle. Writing their successive labels in the extension as j i j_i j i , we have
j i + B = j i + 1 ( m o d 3 ) . (15) j_{i+B}=j_i+1 \pmod{3}. \tag*{(15)} j i + B = j i + 1 ( mod 3 ) . (15)
If N N N counts the positive control arcs after a prescribed fundamental cut, that cut splits at most one complete physical arc. Consequently
N ≤ B + 1. (16) N \le B+1. \tag*{(16)} N ≤ B + 1. (16)
The upper argument first bounds B B B ; it then uses this inequality to bound N N N , keeping the prescribed cut.
Let c i > 0 c_i>0 c i > 0 be the cost of complete physical arc i i i , with indices modulo B B B . Cost invariance under Φ \Phi Φ gives c i + B = c i c_{i+B}=c_i c i + B = c i , even though the labels have the twist (15) . Therefore
C = ∑ i = 0 B − 1 c i . (17) C=\sum_{i=0}^{B-1}c_i. \tag*{(17)} C = i = 0 ∑ B − 1 c i . (17)
The smoothed-octagon competitor of [3, Section 7.5] gives
C ≤ C o c t ≔ 2 ( 8 − 4 2 − log 2 ) 2 2 − 1 < C ∗ ≔ 180483 100000 . (18) C\le C_{\mathrm{oct}}\coloneqq\frac{2(8-4\sqrt{2}-\log2)}{2\sqrt{2}-1}<C_*\coloneqq\frac{180483}{100000}. \tag*{(18)} C ≤ C oct : = 2 2 − 1 2 ( 8 − 4 2 − log 2 ) < C ∗ : = 100000 180483 . (18)
The exact rational comparison, including elementary error bounds for 2 \sqrt{2} 2 , 3 \sqrt{3} 3 , log 2 \log2 log 2 and π \pi π , is supplied with (A.4) below. Positivity of every c i c_i c i and the strict upper budget are the global ingredients in the assembly argument below.
Finally put ε i = 1 \varepsilon_i=1 ε i = 1 or − 1 -1 − 1 according as j i + 1 − j i ≡ 1 j_{i+1}-j_i\equiv1 j i + 1 − j i ≡ 1 or − 1 ( m o d 3 ) -1\pmod{3} − 1 ( mod 3 ) . These increments are B B B -periodic, and their integer walk J ( 0 ) = 0 J(0)=0 J ( 0 ) = 0 , J ( i + 1 ) = J ( i ) + ε i J(i+1)=J(i)+\varepsilon_i J ( i + 1 ) = J ( i ) + ε i satisfies
J ( B ) ≡ 1 ( m o d 3 ) , J ( q B + r ) = q J ( B ) + J ( r ) , 0 ≤ r < B . (19) J(B)\equiv1\pmod{3},\qquad J(qB+r)=qJ(B)+J(r),\qquad0\le r<B. \tag*{(19)} J ( B ) ≡ 1 ( mod 3 ) , J ( q B + r ) = q J ( B ) + J ( r ) , 0 ≤ r < B . (19)
Cyclic rotation changes the chosen first switch. Our type convention uses the least rotation with − < + -\ <\ + − < + , without additionally identifying a word with its reversal or sign complement. The first and third restrictions of Proposition 2.4 exclude the increment patterns + − + +-+ + − + and + − − − + +---+ + − − − + , including occurrences crossing the seam. Source words are always reconstructed from the twisted walk, never from a periodic continuation of the control labels.
Table 1. Global notation. Local comparison coordinates are defined at their first use.
A uniform explicit arc bound
Theorem 3.1. (An explicit uniform arc bound). For every original global minimizer, the number N N N of positive maximal vertex-control pieces in any prescribed fundamental interval satisfies N < 26 ⋅ 2 160 < 2 165 N < 26 \cdot2^{160} < 2^{165} N < 26 ⋅ 2 160 < 2 165 .
This is the first quantitative step after qualitative finiteness. It gives a numerical bound before any word-dependent cost table is used. We retain a deliberately coarse constant: the improvement will come from using the geometry of particular words, rather than optimizing the following powers of two.
We use the state, full lift and switching functions of § 3.1 . All intervals below lie in one genuine fundamental interval, with its original signed endpoint and real winding. The compact set is the same K K K as in Theorem 2.1 .
Lemma 3.2. (Elementary bounds on the minimizing state). On K K K ,
∣ u ∣ < 1 , 1 4 < v < 8 , 1 + u , 1 − u ≥ 1 60 , S ≥ 1 30 . (20) |u| < 1,\qquad\frac{1}{4} < v < 8,\qquad1+u,1-u \ge\frac{1}{60},\qquad S \ge\frac{1}{30}. \tag*{(20)} ∣ u ∣ < 1 , 4 1 < v < 8 , 1 + u , 1 − u ≥ 60 1 , S ≥ 30 1 . (20) max j ∥ F j ∥ ∞ < 128. (21) \max_j \lVert F_j\rVert_\infty< 128. \tag*{(21)} j max ∥ F j ∥ ∞ < 128. (21)
Proof. Let r = 1 − ∣ u ∣ r=1-|u| r = 1 − ∣ u ∣ . The side cuts give r ≥ v / 15 r \ge v/15 r ≥ v /15 , while g 0 = 1 − u 2 = r ( 2 − r ) g_0=1-u^2=r(2-r) g 0 = 1 − u 2 = r ( 2 − r ) and 0 ≤ r ≤ 1 0 \le r \le1 0 ≤ r ≤ 1 . Therefore
v 2 = g 0 + S ≥ v 15 ( 2 − v 15 ) + 2 v 15 , v ≥ 30 113 > 1 4 . v^2=g_0+S \ge\frac{v}{15}\left(2-\frac{v}{15}\right)+\frac{2v}{15},\qquad v \ge\frac{30}{113}>\frac{1}{4}. v 2 = g 0 + S ≥ 15 v ( 2 − 15 v ) + 15 2 v , v ≥ 113 30 > 4 1 .
The height cut gives v ≤ 9 3 / 2 < 8 v \le9\sqrt{3}/2 < 8 v ≤ 9 3 /2 < 8 , and the remaining bounds in (20) follow from the side cuts. Set a = v / ( 1 + u ) a=v/(1+u) a = v / ( 1 + u ) , b = v / ( 1 − u ) b=v/(1-u) b = v / ( 1 − u ) and c = 2 v / S c=2v/S c = 2 v / S ; each is at most 15. Thus the vertical components of F 0 , F 2 F_0,F_2 F 0 , F 2 have absolute value at most 15 v < 120 15v<120 15 v < 120 . For F 1 F_1 F 1 ,
∣ ( F 1 ) u ∣ = v + c ( 1 − u 2 ) < 23 , ∣ ( F 1 ) v ∣ = ∣ u ∣ v c < 120. |(F_1)_u|=v+c(1-u^2)<23,\qquad|(F_1)_v|=|u|vc<120. ∣ ( F 1 ) u ∣ = v + c ( 1 − u 2 ) < 23 , ∣ ( F 1 ) v ∣ = ∣ u ∣ v c < 120.
The other horizontal components are v < 8 v<8 v < 8 , proving the claim. □
Lemma 3.3. (Thirteen arcs provide two repeated projections). Among the twelve interior switches of any thirteen consecutive positive maximal arcs, two distinct unordered control pairs each occur at least twice.
Proof. Four consecutive switches of one unordered pair would lie on a binary interval and contradict Lemma A.4 . Thus every consecutive run of one switch-pair type has length at most three. If at most one of the three pair types occurred more than once, the other two types would occupy at most two positions. They divide the remaining positions into at most three runs of the repeated type. There could therefore be at most 3 ⋅ 3 + 2 = 11 3\cdot3+2=11 3 ⋅ 3 + 2 = 11 switches, whereas there are twelve. This contradiction proves the claim. □
A lower bound for switching windows
We now obtain the local cost estimate used in the finite reduction.
Proposition 4.1. Every block of thirteen consecutive positive maximal arcs of a minimizing trajectory has duration and cost greater than h = 2 − 160 h=2^{-160} h = 2 − 160 .
Fix a nonzero full maximizing lift on the prescribed fundamental interval. All windows in this section are restrictions of this one lift. For a vector, ∥ ⋅ ∥ ∞ \lVert\cdot\rVert_\infty ∥ ⋅ ∥ ∞ is the maximum norm; for a linear map it is the induced row-sum norm. For a traceless matrix Z = ( z 1 z 2 z 3 − z 1 ) Z=\begin{pmatrix}z_1&z_2\\z_3&-z_1\end{pmatrix} Z = ( z 1 z 3 z 2 − z 1 ) , write ∣ Z ∣ e n t = max ( ∣ z 1 ∣ , ∣ z 2 ∣ , ∣ z 3 ∣ ) |Z|_{\mathrm{ent}}=\max(|z_1|,|z_2|,|z_3|) ∣ Z ∣ ent = max ( ∣ z 1 ∣ , ∣ z 2 ∣ , ∣ z 3 ∣ ) .
Uniform derivative bounds
The compact state bounds give explicit constants for the local estimate.
Lemma 4.2 (Uniform derivative bounds). On an open vertex-control arc with mode j j j , put P = ∥ p ∥ ∞ P=\lVert p\rVert_{\infty} P = ∥ p ∥ ∞ and Z 0 = ∣ Z ∣ e n t Z_0=\lvert Z\rvert_{\mathrm{ent}} Z 0 = ∣ Z ∣ ent . For i = 0 , 1 , 2 i=0,1,2 i = 0 , 1 , 2 ,
∣ σ i ′ ′ ∣ ≤ 2 128 ( P + Z 0 + η ) . (22) \lvert\sigma_i''\rvert\le2^{128}(P+Z_0+\eta). \tag*{(22)} ∣ σ i ′′ ∣ ≤ 2 128 ( P + Z 0 + η ) . (22) Moreover,
σ i ′ = R 1 , i , j p − α i ∇ U , ∥ R 1 , i , j ∥ 1 ≤ 2 83 . \sigma_i'=R_{1,i,j}p-\alpha_i\nabla U,\qquad\lVert R_{1,i,j}\rVert_1\le2^{83}. σ i ′ = R 1 , i , j p − α i ∇ U , ∥ R 1 , i , j ∥ 1 ≤ 2 83 . Proof. Each component of F j , α i , X F_j,\alpha_i,X F j , α i , X is a polynomial or a quotient n / d n/d n / d , up to a constant factor of absolute value at most one, with d ∈ { v , 1 + u , 1 − u , S } d\in\{v,1+u,1-u,S\} d ∈ { v , 1 + u , 1 − u , S } . On (3.1), the numerators and their spatial derivatives through order two are bounded by 2 10 2^{10} 2 10 , while d ≥ 1 / 64 d\ge1/64 d ≥ 1/64 , ∣ d k ∣ ≤ 16 \lvert d_k\rvert\le16 ∣ d k ∣ ≤ 16 and ∣ d k l ∣ ≤ 2 \lvert d_{kl}\rvert\le2 ∣ d k l ∣ ≤ 2 . Thus the derivatives of d − 1 d^{-1} d − 1 through order two are bounded by 2 28 2^{28} 2 28 ; the largest bound is
∣ ( d − 1 ) k l ∣ ≤ 2 ⋅ 64 3 ⋅ 16 2 + 64 2 ⋅ 2 < 2 28 . \lvert(d^{-1})_{kl}\rvert\le2\cdot64^3\cdot16^2+64^2\cdot2<2^{28}. ∣( d − 1 ) k l ∣ ≤ 2 ⋅ 6 4 3 ⋅ 1 6 2 + 6 4 2 ⋅ 2 < 2 28 . The product rule therefore bounds all components of F j , α i , X F_j,\alpha_i,X F j , α i , X and their derivatives through order two by L : = 4 ⋅ 2 10 ⋅ 2 28 = 2 40 L:=4\cdot2^{10}\cdot2^{28}=2^{40} L := 4 ⋅ 2 10 ⋅ 2 28 = 2 40 .
Since ∥ ∇ U ∥ ∞ ≤ 4 L Z 0 \lVert\nabla U\rVert_\infty\le4LZ_0 ∥ ∇ U ∥ ∞ ≤ 4 L Z 0 , the adjoint equations
p ′ = − ( D F j ) t p − ∇ U , Z ′ = [ Z + η A ∗ , X , X ] p'=-(DF_j)^{\mathsf{t}}p-\nabla U,\qquad Z'=[Z+\eta A_{*,X},X] p ′ = − ( D F j ) t p − ∇ U , Z ′ = [ Z + η A ∗ , X , X ] and the differentiated identity
p ′ ′ = − D 2 F j [ F j ′ ] t p − ( D F j ) t p ′ − tr ( Z ′ ∇ X ) − tr ( Z D ( ∇ X ) F j ) p''=-D^2F_j[F_j']^{\mathsf{t}}p-(DF_j)^{\mathsf{t}}p'-\operatorname{tr}(Z'\nabla X)-\operatorname{tr}(ZD(\nabla X)F_j) p ′′ = − D 2 F j [ F j ′ ] t p − ( D F j ) t p ′ − tr ( Z ′ ∇ X ) − tr ( Z D ( ∇ X ) F j ) give the bounds
∥ p ′ ∥ ∞ ≤ 2 L P + 4 L Z 0 , ∣ Z ′ ∣ e n t ≤ 4 L Z 0 + 8 L η , \lVert p'\rVert_\infty\le2LP+4LZ_0,\qquad\lvert Z'\rvert_{\mathrm{ent}}\le4LZ_0+8L\eta, ∥ p ′ ∥ ∞ ≤ 2 L P + 4 L Z 0 , ∣ Z ′ ∣ ent ≤ 4 L Z 0 + 8 L η , ∥ p ′ ′ ∥ ∞ ≤ 8 L 2 P + 32 L 2 Z 0 + 32 L 2 η . \lVert p''\rVert_\infty\le8L^2P+32L^2Z_0+32L^2\eta. ∥ p ′′ ∥ ∞ ≤ 8 L 2 P + 32 L 2 Z 0 + 32 L 2 η . Differentiating σ i ′ = α i p \sigma_i'=\alpha_i p σ i ′ = α i p twice gives
σ i ′ ′ = D 2 α i [ F j , F j ] p + D α i ( D F j F j ) p + 2 ( D α i F j ) p ′ + α i p ′ ′ , \sigma_i''=D^2\alpha_i[F_j,F_j]p+D\alpha_i(DF_jF_j)p+2(D\alpha_iF_j)p'+\alpha_i p'', σ i ′′ = D 2 α i [ F j , F j ] p + D α i ( D F j F j ) p + 2 ( D α i F j ) p ′ + α i p ′′ , and hence
∣ σ i ′ ′ ∣ ≤ 48 L 3 P + 96 L 3 Z 0 + 64 L 3 η ≤ 2 128 ( P + Z 0 + η ) . \lvert\sigma_i''\rvert\le48L^3P+96L^3Z_0+64L^3\eta\le2^{128}(P+Z_0+\eta). ∣ σ i ′′ ∣ ≤ 48 L 3 P + 96 L 3 Z 0 + 64 L 3 η ≤ 2 128 ( P + Z 0 + η ) . Finally, the first derivative is
σ i ′ = ( D α i [ F j ] − α i ( D F j ) t ) p − α i ∇ U , \sigma_i'=(D\alpha_i[F_j]-\alpha_i(DF_j)^{\mathsf{t}})p-\alpha_i\nabla U, σ i ′ = ( D α i [ F j ] − α i ( D F j ) t ) p − α i ∇ U , so
( R 1 , i , j ) k = ∑ l = 1 2 ( ( ∂ l α i , k ) F j , l − α i , l ∂ l F j , k ) . (R_{1,i,j})_k=\sum_{l=1}^{2}\left((\partial_l\alpha_{i,k})F_{j,l}-\alpha_{i,l}\partial_lF_{j,k}\right). ( R 1 , i , j ) k = l = 1 ∑ 2 ( ( ∂ l α i , k ) F j , l − α i , l ∂ l F j , k ) . Each component is bounded by 4 L 2 4L^2 4 L 2 , so the row-sum norm is at most 8 L 2 = 2 83 8L^2=2^{83} 8 L 2 = 2 83 . □
Recovering the full adjoint from repeated ties
Two switching projections determine p p p . Their first derivatives then determine ∇ U \nabla U ∇ U , while the active zero Hamiltonian determines U U U . Finally, these three quantities recover the three entries of Z Z Z .
Lemma 4.3 (Recovery from two switching projections). Let [ a , b ] [a,b] [ a , b ] be a window of duration H = b − a ≤ 2 − 90 H=b-a\le2^{-90} H = b − a ≤ 2 − 90 . Suppose that two distinct projections each have two distinct zeros in this window. Put
R = max i = 0 , 1 , 2 ess sup [ a , b ] ∣ σ i ′ ′ ∣ . R=\max_{i=0,1,2}\operatorname*{ess\,sup}_{[a,b]}\lvert\sigma_i''\rvert. R = i = 0 , 1 , 2 max [ a , b ] ess sup ∣ σ i ′′ ∣ . Then, throughout the window,
∥ p ∥ ∞ ≤ 2 7 H 2 R , ∣ Z ∣ e n t ≤ 2 14 H R . (23) \lVert p\rVert_\infty\le2^7H^2R,\qquad\lvert Z\rvert_{\mathrm{ent}}\le2^{14}HR. \tag*{(23)} ∥ p ∥ ∞ ≤ 2 7 H 2 R , ∣ Z ∣ ent ≤ 2 14 H R . (23)
Proof. The finite control representative and Lemma A.2 give C 1 C^1 C 1 projections with absolutely continuous first derivatives and finite R R R . For each selected projection, Rolle’s theorem supplies a derivative zero. Integrating from it and from a zero of the projection gives on the entire window
∥ s ′ ∥ ∞ ≤ H R , ∥ s ∥ ∞ ≤ H 2 R , (24) \|s'\|_{\infty} \le HR,\qquad\|s\|_{\infty} \le H^2R, \tag*{(24)} ∥ s ′ ∥ ∞ ≤ H R , ∥ s ∥ ∞ ≤ H 2 R , (24)
where s s s is the vector of the two selected projections. Neither estimate divides by a distance between zeros.
Let B B B be the inverse of the matrix of the two selected rows. With g 0 = 1 − u 2 g_0=1-u^2 g 0 = 1 − u 2 and
h + = u S − 1 2 ( 1 + u ) , h − = u S + 1 2 ( 1 − u ) , h_+ = \frac{u}{S}-\frac{1}{2(1+u)},\qquad h_-=\frac{u}{S}+\frac{1}{2(1-u)}, h + = S u − 2 ( 1 + u ) 1 , h − = S u + 2 ( 1 − u ) 1 ,
the three possible inverses are
B 01 = ( h + − 1 / g 0 v / S 0 ) , B 02 = ( h − 1 / g 0 v / S 0 ) , B 12 = ( − h − − h + − v / S − v / S ) . B_{01}=\begin{pmatrix} h_+ & -1/g_0 \\ v/S & 0 \end{pmatrix},\qquad B_{02}=\begin{pmatrix} h_- & 1/g_0 \\ v/S & 0 \end{pmatrix},\qquad B_{12}=\begin{pmatrix} -h_- & -h_+ \\ -v/S & -v/S \end{pmatrix}. B 01 = ( h + v / S − 1/ g 0 0 ) , B 02 = ( h − v / S 1/ g 0 0 ) , B 12 = ( − h − − v / S − h + − v / S ) .
The side cuts and (20) give
∣ h ± ∣ ≤ 60 , 1 / g 0 ≤ 60 , v / S ≤ 15 / 2. |h_\pm|\le60,\qquad1/g_0\le60,\qquad v/S\le15/2. ∣ h ± ∣ ≤ 60 , 1/ g 0 ≤ 60 , v / S ≤ 15/2.
Indeed g 0 = r ( 2 − r ) ≥ r ≥ 1 / 60 g_0=r(2-r)\ge r\ge1/60 g 0 = r ( 2 − r ) ≥ r ≥ 1/60 , with r = 1 − ∣ u ∣ r=1-|u| r = 1 − ∣ u ∣ . Therefore ∥ B ∥ ∞ < 128 \|B\|_\infty<128 ∥ B ∥ ∞ < 128 , for every pair everywhere on K K K , including at ( u , v ) = ( 0 , 3 ) (u,v)=(0,\sqrt{3}) ( u , v ) = ( 0 , 3 ) . This proves the first estimate in (23) by p = B s p=Bs p = B s .
Let R 1 R_1 R 1 be the matrix obtained by stacking the two corresponding rows R 1 , i , j R_{1,i,j} R 1 , i , j . The first-derivative equations and the active zero Hamiltonian give
∇ U = B ( R 1 p − s ′ ) , U = − p t F j . \nabla U=B(R_1p-s'),\qquad U=-p^tF_j. ∇ U = B ( R 1 p − s ′ ) , U = − p t F j .
Using Lemma 4.2 , we obtain
∥ ∇ U ∥ ∞ ≤ 2 7 ( 2 83 ∥ p ∥ ∞ + H R ) ≤ 2 97 H 2 R + 2 7 H R ≤ 2 8 H R , \|\nabla U\|_\infty\le2^7(2^{83}\|p\|_\infty+HR)\le2^{97}H^2R+2^7HR\le2^8HR, ∥∇ U ∥ ∞ ≤ 2 7 ( 2 83 ∥ p ∥ ∞ + H R ) ≤ 2 97 H 2 R + 2 7 H R ≤ 2 8 H R ,
∣ U ∣ ≤ 256 ∥ p ∥ ∞ ≤ 2 15 H 2 R ≤ 2 8 H R . |U|\le256\|p\|_\infty\le2^{15}H^2R\le2^8HR. ∣ U ∣ ≤ 256∥ p ∥ ∞ ≤ 2 15 H 2 R ≤ 2 8 H R .
The assumption H ≤ 2 − 90 H\le2^{-90} H ≤ 2 − 90 justifies both last inequalities. These identities and bounds extend to switches by continuity.
The three matrix entries are recovered from U U U and its gradient by
z 3 = − 1 2 ( U v + U / v ) , z 1 = 1 2 v U u + u z 3 , z_3=-\frac{1}{2}(Uv+U/v),\qquad z_1=\frac{1}{2}vU_u+uz_3, z 3 = − 2 1 ( U v + U / v ) , z 1 = 2 1 v U u + u z 3 ,
z 2 = v U − u v U u + ( v 2 − u 2 ) z 3 . (25) z_2=vU-uvU_u+(v^2-u^2)z_3. \tag*{(25)} z 2 = v U − uv U u + ( v 2 − u 2 ) z 3 . (25)
As a linear map from ( U , U u , U v ) (U,U_u,U_v) ( U , U u , U v ) , its row-sum norm is less than 64. For example, after expanding z 3 z_3 z 3 , the largest row is bounded by
v 2 + u 2 2 v + ∣ u v ∣ + ∣ v 2 − u 2 ∣ 2 < 47 ; \frac{v}{2}+\frac{u^2}{2v}+|uv|+\frac{|v^2-u^2|}{2}<47; 2 v + 2 v u 2 + ∣ uv ∣ + 2 ∣ v 2 − u 2 ∣ < 47 ;
the first two rows have bounds 5 / 2 5/2 5/2 and 13 / 2 13/2 13/2 . Thus ∣ Z ∣ e n t ≤ 2 14 H R |Z|_{\mathrm{ent}}\le2^{14}HR ∣ Z ∣ ent ≤ 2 14 H R , as asserted. □
Excluding short windows.
Proposition 4.4 (Short-window exclusion). On the chosen full lift, no window of duration 0 < H ≤ h : = 2 − 160 0<H\le h:=2^{-160} 0 < H ≤ h := 2 − 160 contains two distinct switching projections each with two distinct zeros.
Proof. Suppose such a window exists, and use R R R from Lemma 4.3 . The bounds there imply
sup ( ∥ p ∥ ∞ + ∣ Z ∣ e n t ) ≤ 2 15 H R . \sup(\|p\|_\infty+|Z|_{\mathrm{ent}})\le2^{15}HR. sup ( ∥ p ∥ ∞ + ∣ Z ∣ ent ) ≤ 2 15 H R .
Taking essential suprema in (22) gives
R ≤ 2 128 ( 2 15 H R + η ) , 2 143 H ≤ 2 − 17 < 1 2 . R\le2^{128}(2^{15}HR+\eta),\qquad2^{143}H\le2^{-17}<\frac{1}{2}. R ≤ 2 128 ( 2 15 H R + η ) , 2 143 H ≤ 2 − 17 < 2 1 .
Absorption yields R ≤ 2 129 η R \le2^{129}\eta R ≤ 2 129 η . If η = 0 \eta= 0 η = 0 , the recovery identities force p = Z = 0 p = Z = 0 p = Z = 0 on the window, and hence Λ = 0 \Lambda= 0 Λ = 0 there. Uniqueness for the full homogeneous adjoint equation propagates this zero lift to the containing fundamental interval, contradicting the nonzero multiplier. Thus η > 0 \eta> 0 η > 0 .
Divide η \eta η , both components of p p p , and every entry of Λ \Lambda Λ by this same positive number. The normalized whole lift has η = 1 \eta= 1 η = 1 , and
∣ Z ∣ e n t ≤ 2 143 H ≤ δ : = 2 − 17 . (26) \lvert Z\rvert_{\mathrm{ent}} \le2^{143}H \le\delta:= 2^{-17}. \tag*{(26)} ∣ Z ∣ ent ≤ 2 143 H ≤ δ := 2 − 17 . (26)
At any point of the window,
det Λ = det ( A ∗ + Z ) = 3 4 + z 2 2 − 3 z 3 2 − z 1 2 − z 2 z 3 ≥ 3 4 − 2 δ − 2 δ 2 > 0.74. \det\Lambda= \det(A_* + Z) = \frac{3}{4} + \frac{z_2}{2} - \frac{3z_3}{2} - z_1^2 - z_2z_3 \ge\frac{3}{4} - 2\delta- 2\delta^2 > 0.74. det Λ = det ( A ∗ + Z ) = 4 3 + 2 z 2 − 2 3 z 3 − z 1 2 − z 2 z 3 ≥ 4 3 − 2 δ − 2 δ 2 > 0.74.
In particular the determinant is positive. Also ℓ ≥ 3 > 1 \ell\ge\sqrt{3} > 1 ℓ ≥ 3 > 1 , and the explicit formula for X X X on (20) gives
∣ X 11 ∣ , ∣ X 22 ∣ , ∣ X 21 ∣ < 4 , ∣ X 12 ∣ < 260. \lvert X_{11}\rvert, \lvert X_{22}\rvert, \lvert X_{21}\rvert< 4,\qquad\lvert X_{12}\rvert< 260. ∣ X 11 ∣ , ∣ X 22 ∣ , ∣ X 21 ∣ < 4 , ∣ X 12 ∣ < 260.
Consequently ∣ tr ( Z X ) ∣ < 272 δ < 1 \lvert\operatorname{tr}(ZX)\rvert< 272\delta< 1 ∣ tr ( Z X )∣ < 272 δ < 1 , and
tr ( Λ X ) = ℓ + tr ( Z X ) > 0. \operatorname{tr}(\Lambda X) = \ell+ \operatorname{tr}(ZX) > 0. tr ( Λ X ) = ℓ + tr ( Z X ) > 0.
Theorem A.12, applied to this normalized full lift, now gives det Λ < 0.74 \det\Lambda< 0.74 det Λ < 0.74 , a contradiction.
Proof of Proposition 4.1. Lemma 3.3 and (10) supply two distinct switching functions, each vanishing at two switch times in the block. Positive intervening arc durations make these times distinct. Proposition 4.4 therefore forces the block’s duration to exceed h h h . Finally,
ℓ ( W ) = 3 + u 2 + v 2 2 v ≥ 3 + u 2 ≥ 3 > 1 , \ell(W) = \frac{3 + u^2 + v^2}{2v} \ge\sqrt{3 + u^2} \ge\sqrt{3} > 1, ℓ ( W ) = 2 v 3 + u 2 + v 2 ≥ 3 + u 2 ≥ 3 > 1 ,
so its cost exceeds its duration.
The explicit finite reduction
Proof of Theorem 3.1. Use the finite vertex-control representative from Theorem A.3, merge equal neighbors, and keep the prescribed cut. Write N = 13 q + r N = 13q + r N = 13 q + r , where 0 ≤ r ≤ 12 0 \le r \le12 0 ≤ r ≤ 12 , and divide the first 13 q 13q 13 q arcs into q q q disjoint blocks of thirteen arcs. The cut may truncate the first or last arc of the fundamental interval; each block still has thirteen positive arcs and twelve interior switches.
Proposition 4.1 and the total cost bound give q h < C < 2 qh < C < 2 q h < C < 2 . Since h = 2 − 160 h = 2^{-160} h = 2 − 160 and q q q is an integer,
q ≤ 2 161 − 1 , N ≤ 13 ( 2 161 − 1 ) + 12 = 26 ⋅ 2 160 − 1 < 2 165 . q \le2^{161} - 1,\qquad N \le13(2^{161} - 1) + 12 = 26 \cdot2^{160} - 1 < 2^{165}. q ≤ 2 161 − 1 , N ≤ 13 ( 2 161 − 1 ) + 12 = 26 ⋅ 2 160 − 1 < 2 165 .
There are only finitely many words over three labels with this bounded length. Hence only finitely many switching words, and therefore only finitely many of their symmetry classes, can occur.
For a fixed word, the initial state and arc durations remain continuous parameters. The theorem bounds the combinatorial possibilities.
Assembly of local physical-cost inequalities
The interface below separates complete geometric sources from physical charges and supplies the common rule for all later scalar certificates.
Supporting windows and physical charges.
Fix the twisted period of an original global minimizer and its complete physical costs c 0 , … , c B − 1 > 0 c_0,\ldots,c_{B-1}>0 c 0 , … , c B − 1 > 0 . A local support is an ordered interval of actual positive maximal arcs in the unwrapped trajectory, together with any specified endpoints of that interval. A charge specifies which of those arc costs, and with which nonnegative coefficients, occur on the left side of the local inequality. Support and charge need not coincide.
Uncharged caps and returns remain hypotheses: deleting them may destroy an actual switch sign or circle condition.
Lemma 6.1 (Placement of a proper source). A consecutive source of L < B L<B L < B complete physical arcs can be placed inside a genuine minimizing fundamental interval while retaining every source arc complete and positive. A full lift can then be chosen on that whole interval. If B ≥ 8 B\ge8 B ≥ 8 , it is normal and can be jointly normalized.
Proof. Place the fundamental cut strictly inside the positive arc immediately preceding the source. Since L < B L<B L < B , the interval ends after the source, before returning to that cut. Lemma 2.6 supplies the genuine minimizing trajectory with its original endpoints and real winding. There are B − 1 ≥ 7 B-1\ge7 B − 1 ≥ 7 complete arcs between the two pieces of the cut arc. The first assertion of Proposition 2.4 applies to this minimizer, and its second assertion proves normality of any selected nonzero full lift on the containing interval. Only then normalize it. □
The strict inequality L < B L<B L < B is a sufficient placement condition, not a substitute for the reanchoring argument. If a theorem uses a larger union of supports on one common lift, the containing interval must retain that whole union. Checking properness separately for its constituent pieces does not establish the common-lift hypothesis.
A qualified application records its complete unwrapped source (including positive caps, returns and auxiliary arcs), its whole lawful phase or reflected-time transformation, a genuine containing interval and the required selected jointly normalized full lift, and every theorem-specific root, contact, sign, mean or circle hypothesis. If identities link several pieces, their common-lift requirement is part of these data. Finally it records the charged offsets, coefficients, lower bound and strictness. A word match alone supplies neither an endpoint sign nor an adjoint.
If local positions k k k have coefficients q r k ≥ 0 q_{rk}\ge0 q r k ≥ 0 and map to physical arc π r ( k ) ∈ Z / B Z \pi_r(k)\in\mathbb{Z}/B\mathbb{Z} π r ( k ) ∈ Z / B Z , their physical row is
a r i = ∑ π r ( k ) = i q r k , a r ⋅ c = ∑ i = 0 B − 1 a r i c i . (27) a_{ri}=\sum_{\pi_r(k)=i}q_{rk},\qquad a_r\cdot c=\sum_{i=0}^{B-1}a_{ri}c_i. \tag*{(27)} a r i = π r ( k ) = i ∑ q r k , a r ⋅ c = i = 0 ∑ B − 1 a r i c i . (27)
Thus repeated charged occurrences of the same physical arc count with multiplicity. One cannot replace that sum by a set union unless a separate union theorem proves the once-counted charge. On a proper consecutive source the positions themselves have distinct physical images, but different sources can still charge the same arc.
The capacity theorem, including equality.
Theorem 6.2 (Physical-cost assembly). Let c i ≥ 0 c_i\ge0 c i ≥ 0 , 0 ≤ i < B 0\le i<B 0 ≤ i < B , and C = ∑ i c i C=\sum_i c_i C = ∑ i c i . Suppose finitely many valid local conclusions on this trajectory have the form
a r ⋅ c ≥ b r , a r ∈ R B , (28) a_r\cdot c\ge b_r,\qquad a_r\in\mathbb{R}^{B}, \tag*{(28)} a r ⋅ c ≥ b r , a r ∈ R B , (28) with specified rows known to be strict. The algebra also allows real row coefficients; physical charge rows have a r i ≥ 0 a_{ri}\ge0 a r i ≥ 0 as in (6.1). If numbers λ r ≥ 0 \lambda_r\ge0 λ r ≥ 0 satisfy
γ i : = ∑ r λ r a r i ≤ 1 ( 0 ≤ i < B ) , (29) \gamma_i:=\sum_r\lambda_r a_{ri}\le1\qquad(0\le i<B), \tag*{(29)} γ i := r ∑ λ r a r i ≤ 1 ( 0 ≤ i < B ) , (29)
then
C ≥ L ≔ ∑ r λ r b r . (30) C \ge L \coloneqq\sum_r \lambda_r b_r. \tag*{(30)} C ≥ L : = r ∑ λ r b r . (30)
The conclusion is strict if a strict row has positive weight, or if γ i < 1 \gamma_i < 1 γ i < 1 and c i > 0 c_i > 0 c i > 0 for some i i i . Equality C = L C = L C = L holds precisely when ( 1 − γ i ) c i = 0 (1-\gamma_i)c_i = 0 ( 1 − γ i ) c i = 0 for every i i i and every positively weighted row is an equality. In particular, positive physical costs force all capacities to be one in the equality case.
Proof. There is an exact identity
C − L = ∑ i = 0 B − 1 ( 1 − γ i ) c i + ∑ r λ r ( a r ⋅ c − b r ) . (31) C-L = \sum_{i=0}^{B-1}(1-\gamma_i)c_i+\sum_r\lambda_r(a_r\cdot c-b_r). \tag*{(31)} C − L = i = 0 ∑ B − 1 ( 1 − γ i ) c i + r ∑ λ r ( a r ⋅ c − b r ) . (31)
Every summand is nonnegative. A strict source row with positive weight makes a summand in the second sum positive. An unsaturated capacity at a positive cost makes a summand in the first sum positive. This proves the inequality and both strictness claims. Equality is equivalent to every term in the two sums vanishing. □
Corollary 6.3 (Exclusion at the competitor budget). For an original global minimizer, a valid capacity combination with L ≥ C ∗ L \ge C_* L ≥ C ∗ is impossible. If only the weaker comparison C ≤ C ∗ C \le C_* C ≤ C ∗ is used, the same conclusion holds when L > C ∗ L > C_* L > C ∗ , or when L = C ∗ L = C_* L = C ∗ and either strictness condition in Theorem 6.2 holds.
Proof. The actual budget is C < C ∗ C < C_* C < C ∗ by (18) , whereas Theorem 6.2 gives C ≥ L C \ge L C ≥ L . The assertions with a weak budget follow from its strict version. □
A weak source row remains weak; strictness is supplied by the stated positive-weight or residual-capacity condition. Under the actual strict budget, a weak lower sum exactly C ∗ C_* C ∗ already excludes the type.
Only physical costs occur in (31) . Supporting windows may overlap arbitrarily, and independently established scalar inequalities may use different genuine containing intervals. Their full lifts need not be identified. By contrast, a relation between two means, two switching derivatives, or a common determinant is not a scalar cost row until its stipulated common-lift argument has been completed. It cannot be assembled by choosing the most favorable component from independent lifts.
Exhaustive alternatives and conditional rows
For example,
max ( c i , c j ) > h (32) \max(c_i,c_j)>h \tag*{(32)} max ( c i , c j ) > h (32)
is the alternative c i > h c_i > h c i > h or c j > h c_j > h c j > h , not two simultaneous rows.
Theorem 6.4 (Branchwise physical-cost assembly). Fix a twisted increment word w w w and let M ( w ) \mathcal{M}(w) M ( w ) denote the original minimizing trajectories with that word, equipped with any full lifts required by the source theorems. Suppose finitely many predicates P ν P_\nu P ν , ν ∈ V \nu\in V ν ∈ V , cover M ( w ) \mathcal{M}(w) M ( w ) : every such trajectory, with the stipulated choices of lifts, satisfies at least one P ν P_\nu P ν . For each ν \nu ν , suppose that under P ν P_\nu P ν the complete source theorems prove rows
a ν r ⋅ c ≥ b ν r , a ν r i ≥ 0 , a_{\nu r}\cdot c \ge b_{\nu r}, \qquad a_{\nu r i}\ge0, a ν r ⋅ c ≥ b ν r , a ν r i ≥ 0 , with their stated strictness. If that branch has nonnegative weights λ ν r \lambda_{\nu r} λ ν r satisfying
∑ r λ ν r a ν r i ≤ 1 for every i , L ν ≔ ∑ r λ ν r b ν r ≥ C ∗ , \sum_r \lambda_{\nu r}a_{\nu r i}\le1 \quad\text{for every }i, \qquad L_\nu\coloneqq\sum_r\lambda_{\nu r}b_{\nu r}\ge C_*, r ∑ λ ν r a ν r i ≤ 1 for every i , L ν : = r ∑ λ ν r b ν r ≥ C ∗ , then M ( w ) \mathcal{M}(w) M ( w ) is empty. With a weak global budget C ≤ C ∗ C \le C_* C ≤ C ∗ , replace the last condition by L ν > C ∗ L_\nu> C_* L ν > C ∗ , or by L ν = C ∗ L_\nu= C_* L ν = C ∗ together with a strictness condition from Theorem 6.2 , in every branch.
Proof. Suppose a trajectory in M ( w ) \mathcal{M}(w) M ( w ) exists. Choose its lifts as required by the source hypotheses and select a predicate P ν P_{\nu} P ν that it satisfies. All rows in that branch now hold on its one physical cost vector. Corollary 6.3 gives a contradiction. The proof uses no row from a different branch. The predicates need not be disjoint, since choosing any one satisfied branch suffices. □
Predicates involving means or adjoints must use the same lifts as their rows; independently proved scalar costs may be included in every branch. A finite branch tree yields the predicates by conjunction along paths. Each node’s children must cover its entire condition, including equality boundaries unless separately excluded. The predicates may overlap; a maximum such as (32) still requires a sufficient complementary payment in each branch.
An explicit ten-switch example
We illustrate the source and charge bookkeeping with two actual exclusions used in the upper reduction. This example uses three local estimates proved later in this paper:
The complete seven-arc source 2021020 charges its central position 3. We use the sufficient consequence that its cost is greater than 461/500 of Theorem 8.10 .
The right one-return source 1020212 charges positions 1, 2, 3, 4; the left source 1210102 charges positions 2, 3, 4, 5. Each cost is greater than
E ∗ : = 4154032208439 8759490110000 E_* := \frac{4154032208439}{8759490110000} E ∗ := 8759490110000 4154032208439
(Theorem 8.2 ).
The complete six-arc sources 101202, 212010, and 020121 charge their four interior arcs with cost greater than 41/100 (the four-own-arc estimate 8.5).
These statements include their whole lawful transforms and require all source arcs to be positive and complete on a genuine containing interval. In the last statement the two actual returns supply the endpoint signs; a four-letter substring alone would not supply them.
Proposition 6.5. The cyclic increment words ----+--+++ and ----+++--+ cannot occur on an original global minimizer.
Proof. For each word reconstruct j i = J ( i ) j_i = J(i) j i = J ( i ) (mod 3) using (19) , starting with j 0 = 0 j_0 = 0 j 0 = 0 . In both cases the source [ref-9 , ref-15 ] is 0102101, whose whole cyclic phase +2 is 2021020. Its centre is arc 12, or physical arc 2. The actual incoming and outgoing switches and all three terminal arcs are retained. A cut inside arc 8 puts this entire source in a genuine containing interval, and the first source theorem gives
c 2 > 461 / 500. c_2 > 461/500. c 2 > 461/500.
For ----+--+++, the source [ref-2 , ref-8 ] is 1020212. It charges {3, 4, 5, 6}, with caps 2, 7 and returned arc 8. The source [ref-6 , ref-11 ] is 212010 and charges physical {7, 8, 9, 0}; its two exterior arcs remain part of the source. Thus the three inequalities charge disjoint sets, leaving physical arc 1 unpaid.
For ----+++--+, the source [ref-6 , ref-12 ] is 1210102. It charges physical {8, 9, 0, 1}, with caps 7, 12 and returned arc 6. The source [ref-3 , ref-8 ] is 020121 and charges {4, 5, 6, 7}. Together with the central charge {2} these sets are disjoint, leaving physical arc 3 unpaid.
Each source contains at most seven complete arcs, fewer than the period of ten. Lemma 6.1 supplies a genuine interval for each one. The local theorem’s internal contact and return arguments use one selected normal lift on that interval. Only their scalar costs are added between intervals. Giving all three rows weight one satisfies every physical capacity constraint. In either case their exact sum is
C > E ∗ + 41 100 + 461 500 = 15821673034959 8759490110000 = C ∗ + 122824997277 87594901100000 > C ∗ . C > E_* + \frac{41}{100} + \frac{461}{500}
= \frac{15821673034959}{8759490110000}
= C_* + \frac{122824997277}{87594901100000} > C_*. C > E ∗ + 100 41 + 500 461 = 8759490110000 15821673034959 = C ∗ + 87594901100000 122824997277 > C ∗ .
This contradicts (2.19). □
The supports in this example overlap, including at arcs that are charged by another row. This is harmless: capacity counts coefficients in the charges, not the number of supporting windows through an arc.
6.5. What a finite certificate proves. A finite certificate consists of the qualified applications above, their physical rows (6.1), exhaustive branch predicates and nonnegative rational weights satisfying (6.3) with the required lower sum. Discovery may use optimization; verification uses only exact arithmetic after source qualification. A complete reduction must also cover every excluded word. The differential theorems establish geometric eligibility; the finite check establishes source maps, branches, weights and capacities. Neither approximate optima nor reported counts replace these obligations. A feasible surviving scalar vector is not a trajectory without state and frame evolution, original endpoints, real winding and a full maximizing lift.
Cap transport and the cost of an own pair
The hypotheses in this section always include the original global minimizer framework, genuine full endpoints and real winding. A “proper source” means fewer than B B B complete positive maximal arcs. A genuine cut in an omitted positive arc retains the source. The selected nonzero maximizing lift is first proved normal and only then jointly normalized. All means in an identity are actual means on that same full lift. Algebraic comparison points need not be realizable.
We use the original harmonic ceiling
H ∗ : = 29025933167887355473205403 39478417604337498218323684 . H_* := \frac{29025933167887355473205403}{39478417604337498218323684}. H ∗ := 39478417604337498218323684 29025933167887355473205403 .
7.1. The positive-cap identity. Use an actual returned leaf on each side of a positive shared cap. Write u , v u,v u , v for their positive outer-root magnitudes, h 2 , ν h^2,\nu h 2 , ν for the corresponding positive actual weighted means, and x x x for the shared reciprocal squared. Put
A = 1 + x , k = 4 x 1 + x , t = k ( 1 − u ) ( 1 − v ) , Φ ( t ) = t − 1 − log t . A = 1+x,\qquad k = \frac{4x}{1+x},\qquad t = \frac{k}{(1-u)(1-v)},\qquad\Phi(t) = t-1-\log t. A = 1 + x , k = 1 + x 4 x , t = ( 1 − u ) ( 1 − v ) k , Φ ( t ) = t − 1 − log t .
The actual domain is
0 < h < u < 1 , 0 < ν < v 2 , 0 < v < 1 , 0 < t < 1. 0<h<u<1,\qquad0<\nu<v^2,\qquad0<v<1,\qquad0<t<1. 0 < h < u < 1 , 0 < ν < v 2 , 0 < v < 1 , 0 < t < 1.
Let λ \lambda λ be the oriented integration constant of this actual cap, and put E = λ + 3 x / 2 E=\lambda+3x/2 E = λ + 3 x /2 . The positive-cap identity is
E = A 2 { Φ ( t ) + ( 1 − t ) ( u + v ) + t u v } + A 4 { u 2 − h 2 1 − u + v 2 − ν 1 − v } . (33) E = \frac{A}{2}\{\Phi(t)+(1-t)(u+v)+tuv\}+\frac{A}{4}\left\{\frac{u^2-h^2}{1-u}+\frac{v^2-\nu}{1-v}\right\}. \tag*{(33)} E = 2 A { Φ ( t ) + ( 1 − t ) ( u + v ) + t uv } + 4 A { 1 − u u 2 − h 2 + 1 − v v 2 − ν } . (33)
Here E E E is a matrix-transport term. To derive the identity, set K = ( 3 + h 2 ) / 2 K=(3+h^2)/2 K = ( 3 + h 2 ) /2 , K ^ = ( 3 + ν ) / 2 \widehat{K}=(3+\nu)/2 K = ( 3 + ν ) /2 and β = 1 − t ( 1 − u ) \beta=1-t(1-u) β = 1 − t ( 1 − u ) . The actual binary primitive and cap equality of Section C.1.2 read
χ β = x { h 2 − 1 1 − u ( t − 1 − 1 ) − 2 log t + ( 1 − u ) ( t − 1 ) } \chi_\beta=x\left\{\frac{h^2-1}{1-u}(t^{-1}-1)-2\log t+(1-u)(t-1)\right\} χ β = x { 1 − u h 2 − 1 ( t − 1 − 1 ) − 2 log t + ( 1 − u ) ( t − 1 ) }
f β = ( h 2 − β 2 ) x 2 ( 1 − β ) 2 − A x β 2 ( 1 − β ) , f_\beta=\frac{(h^2-\beta^2)x}{2(1-\beta)^2}-\frac{Ax_\beta}{2(1-\beta)}, f β = 2 ( 1 − β ) 2 ( h 2 − β 2 ) x − 2 ( 1 − β ) A x β ,
0 = 1 2 − K ^ A 2 4 − 2 λ x 1 − β − K x 2 ( 1 − β ) 2 − f β . 0=\frac{1}{2}-\frac{\widehat{K}A^2}{4}-\frac{2\lambda x}{1-\beta}-\frac{Kx^2}{(1-\beta)^2}-f_\beta. 0 = 2 1 − 4 K A 2 − 1 − β 2 λ x − ( 1 − β ) 2 K x 2 − f β .
Solving for λ \lambda λ , using 4 x / A = t ( 1 − u ) ( 1 − v ) 4x/A=t(1-u)(1-v) 4 x / A = t ( 1 − u ) ( 1 − v ) , gives (33) . In particular the identity preserves the oriented constant; neither its square nor a separately normalized cap has been substituted. No physical cap cost is added to the charge.
Lemma 7.1 (Root-width contribution of one cap). In the preceding actual domain,
E + x ( h 2 + 3 u ) 2 ( 1 − u ) > u 2 + ( 1 − x ) ( u 2 − h 2 ) 4 ( 1 − u ) > u 2 . E+\frac{x(h^2+3u)}{2(1-u)}>\frac{u}{2}+\frac{(1-x)(u^2-h^2)}{4(1-u)}>\frac{u}{2}. E + 2 ( 1 − u ) x ( h 2 + 3 u ) > 2 u + 4 ( 1 − u ) ( 1 − x ) ( u 2 − h 2 ) > 2 u .
Proof. Set t 0 = k / ( 1 − u ) t_0=k/(1-u) t 0 = k / ( 1 − u ) . The actual positive other root gives v = 1 − t 0 / t > 0 v=1-t_0/t>0 v = 1 − t 0 / t > 0 , so t 0 < t < 1 t_0<t<1 t 0 < t < 1 . Substitution yields
Φ ( t ) + ( 1 − t ) ( u + v ) + t u v = u + k − log t − k t ( 1 − u ) . \Phi(t)+(1-t)(u+v)+tuv=u+k-\log t-\frac{k}{t(1-u)}. Φ ( t ) + ( 1 − t ) ( u + v ) + t uv = u + k − log t − t ( 1 − u ) k .
For fixed actual x x x , u u u , the derivative of the last two terms is ( t 0 − t ) / t 2 < 0 (t_0-t)/t^2<0 ( t 0 − t ) / t 2 < 0 . Their value is therefore strictly larger than its comparison value at t = 1 t=1 t = 1 . Dropping only the positive other-leaf deficit in (33) gives
E > A ( 2 u − u 2 − h 2 ) 4 ( 1 − u ) − 2 x u 1 − u . E>\frac{A(2u-u^2-h^2)}{4(1-u)}-\frac{2xu}{1-u}. E > 4 ( 1 − u ) A ( 2 u − u 2 − h 2 ) − 1 − u 2 xu .
Adding the displayed density correction gives the first inequality. The actual positive shared arc and positive outer roots imply x < 1 / 3 x<1/3 x < 1/3 , so the final remainder is strictly positive. No physical zero-length cap or separate calibrated comparison lift has been introduced. □
The own-pair theorem
Consider the complete proper source 01020212. Let − ξ < z < ρ -\xi<z<\rho − ξ < z < ρ be the three actual roots of its own pair, with common weighted mean h 2 h^2 h 2 , K ^ = ( 3 + h 2 ) / 2 \widehat{K}=(3+h^2)/2 K = ( 3 + h 2 ) /2 , and positive shear T T T . Set w = ξ + ρ w=\xi+\rho w = ξ + ρ and let x L , x R x_L,x_R x L , x R be the two actual anchor densities. The physical cost on local arcs { 3 , 4 } \{3,4\} { 3 , 4 } is exactly
Q = w 2 + K T . (34) Q=\frac{w}{2}+KT. \tag*{(34)} Q = 2 w + K T . (34)
The original full-matrix transport gives
K T = E L + E R + x L ( h 2 + 3 ξ ) 2 ( 1 − ξ ) + x R ( h 2 + 3 ρ ) 2 ( 1 − ρ ) . (35) KT=E_L+E_R+\frac{x_L(h^2+3\xi)}{2(1-\xi)}+\frac{x_R(h^2+3\rho)}{2(1-\rho)}. \tag*{(35)} K T = E L + E R + 2 ( 1 − ξ ) x L ( h 2 + 3 ξ ) + 2 ( 1 − ρ ) x R ( h 2 + 3 ρ ) . (35)
The two cap defects in this identity are not two physical payments.
Theorem 7.2 (Analytic own-pair bound). On this proper source,
K T > w 2 + ( 1 − x L ) ( ξ 2 − h 2 ) 4 ( 1 − ξ ) + ( 1 − x R ) ( ρ 2 − h 2 ) 4 ( 1 − ρ ) , Q > w . KT>\frac{w}{2}+\frac{(1-x_L)(\xi^2-h^2)}{4(1-\xi)}+\frac{(1-x_R)(\rho^2-h^2)}{4(1-\rho)},\qquad Q>w. K T > 2 w + 4 ( 1 − ξ ) ( 1 − x L ) ( ξ 2 − h 2 ) + 4 ( 1 − ρ ) ( 1 − x R ) ( ρ 2 − h 2 ) , Q > w . Consequently the original source-qualified ceilings x L , x R < 47 / 250 x_L,x_R<47/250 x L , x R < 47/250 imply Q > 9 / 25 Q>9/25 Q > 9/25 . If two additional qualified sources give x L , x R < 1 / 6 x_L,x_R<1/6 x L , x R < 1/6 , then Q > 3 / 7 Q>3/7 Q > 3/7 . The latter density hypothesis is not a consequence of the bare eight-letter source.
Proof. Apply Lemma 7.1 to the two actual caps in (35) . To obtain the rational constants, put a = x L a=\sqrt{x_L} a = x L , b = x R b=\sqrt{x_R} b = x R and
U = 1 + ξ 1 − ξ , V = 1 + ρ 1 − ρ . U=\frac{1+\xi}{1-\xi},\qquad V=\frac{1+\rho}{1-\rho}. U = 1 − ξ 1 + ξ , V = 1 − ρ 1 + ρ .
Direct integration of the two physical shear profiles gives
T = a b U V − a 2 U + b 2 V 2 . T=abUV-\frac{a^2U+b^2V}{2}. T = ab U V − 2 a 2 U + b 2 V .
If a , b < C a,b<\sqrt{C} a , b < C , scale both by the same factor until one reaches C \sqrt{C} C . This strictly increases T T T , because the actual T T T is positive and the expression is homogeneous of degree two. On the resulting boundary a = C a=\sqrt{C} a = C , the derivative in b b b is V ( C U − b ) > 0 V(\sqrt{C}U-b)>0 V ( C U − b ) > 0 ; the symmetric argument handles the other boundary. Hence, with P 0 = ξ ρ P_0=\xi\rho P 0 = ξ ρ ,
T < C w + 2 P 0 1 − w + P 0 . (36) T<C\frac{w+2P_0}{1-w+P_0}. \tag*{(36)} T < C 1 − w + P 0 w + 2 P 0 . (36)
This does not assert coordinatewise monotonicity throughout the unrestricted anchor square. It is a comparison using positivity and homogeneity first.
Suppose 0 < w ≤ w 0 < 2 / 3 0<w\le w_0<2/3 0 < w ≤ w 0 < 2/3 . The derivative of the fraction in (36) with respect to P 0 P_0 P 0 is ( 2 − 3 w ) / ( 1 − w + P 0 ) 2 > 0 (2-3w)/(1-w+P_0)^2>0 ( 2 − 3 w ) / ( 1 − w + P 0 ) 2 > 0 . Since P 0 ≤ w 2 / 4 P_0\le w^2/4 P 0 ≤ w 2 /4 ,
T < C w ( 1 + w / 2 ) ( 1 − w / 2 ) 2 . T<C\frac{w(1+w/2)}{(1-w/2)^2}. T < C ( 1 − w /2 ) 2 w ( 1 + w /2 ) .
The actual two leaf-quadratic inequalities give w > 12 h 2 − 3 z 2 ≥ 3 h w>\sqrt{12h^2-3z^2}\ge3h w > 12 h 2 − 3 z 2 ≥ 3 h ; thus K < ( 3 + w 2 / 9 ) / 2 K<(3+w^2/9)/2 K < ( 3 + w 2 /9 ) /2 . Combining this with K T > w / 2 KT>w/2 K T > w /2 forces
1 < 2 K T w < C ( 3 + w 2 / 9 ) ( 1 + w / 2 ) ( 1 − w / 2 ) 2 . 1<\frac{2KT}{w}<C\frac{(3+w^2/9)(1+w/2)}{(1-w/2)^2}. 1 < w 2 K T < C ( 1 − w /2 ) 2 ( 3 + w 2 /9 ) ( 1 + w /2 ) .
The last expression increases in w w w on the relevant interval. For ( C , w 0 ) = ( 47 / 250 , 9 / 25 ) (C,w_0)=(47/250,9/25) ( C , w 0 ) = ( 47/250 , 9/25 ) its endpoint is 5224332 / 5253125 < 1 5224332/5253125<1 5224332/5253125 < 1 . For ( C , w 0 ) = ( 1 / 6 , 3 / 7 ) (C,w_0)=(1/6,3/7) ( C , w 0 ) = ( 1/6 , 3/7 ) it is 2516 / 2541 < 1 2516/2541<1 2516/2541 < 1 . Both assumptions w ≤ w 0 w\le w_0 w ≤ w 0 , including equality, are impossible. (34) then proves the claims.
Corollary 7.3 (Two adjacent own pairs). A proper complete source 01020212101 satisfies
c 3 + c 4 + c 5 + c 6 + c 7 > 18 25 . c_3+c_4+c_5+c_6+c_7>\frac{18}{25}. c 3 + c 4 + c 5 + c 6 + c 7 > 25 18 .
Proof. Its source on [ 0 , 7 ] [0,7] [ 0 , 7 ] is 01020212 and charges { 3 , 4 } \{3,4\} { 3 , 4 } . Its source on [ 3 , 10 ] [3,10] [ 3 , 10 ] , after whole phase + 1 +1 + 1 , is the same source and charges { 6 , 7 } \{6,7\} { 6 , 7 } . Both are proper and retain all their caps and actual returns. Theorem 7.2 pays more than 9 / 25 9/25 9/25 on each pair. Adding the two disjoint physical charges and the positive cost of arc 5 proves the assertion. This addition requires no comparison of the two source multipliers.
A six-arc density source.
Theorem 7.4 (Shortened density support). A complete proper 201020 source, or a lawful whole image , gives x < 1 / 6 x<1/6 x < 1/6 for the own reciprocal squared of its local arc 3. The conclusion is a physical state inequality; it adds no cost charge.
Proof. The first five arcs are a cyclic image of 12021. Its binary core has two distinct internal ties and both third-mode caps; full-lift rigidity excludes abnormality. The last five arcs are 01020 and supply the two actual returned means, their roots and both midpoint ties. Thus the normality witness and all three matrix rows below belong to the stated six-arc support.
Put R = 1 + x − 1 R = 1 + x^{-1} R = 1 + x − 1 , A = 1 + x A = 1 + x A = 1 + x , and use the actual root variables p , q ∈ ( 0 , 1 ) p, q \in(0,1) p , q ∈ ( 0 , 1 ) and means 0 < μ L < ( 1 − p ) 2 0 < \mu_L < (1-p)^2 0 < μ L < ( 1 − p ) 2 , 0 < μ R < ( 1 − q ) 2 0 < \mu_R < (1-q)^2 0 < μ R < ( 1 − q ) 2 . The positive shared arc gives R p q > 4 Rpq > 4 R pq > 4 . The same matrix gives
V 0 = log ( R / 4 ) + g ( p ) + g ( q ) , g ( z ) = log z + ( z − 1 − z ) / 2 , V_0 = \log(R/4) + g(p) + g(q), \qquad g(z) = \log z + (z^{-1} - z)/2, V 0 = log ( R /4 ) + g ( p ) + g ( q ) , g ( z ) = log z + ( z − 1 − z ) /2 ,
B 0 = ( 1 − A V 0 ) / 2 + A ( μ L / p + μ R / q ) / 4 , B_0 = (1 - AV_0)/2 + A(\mu_L/p + \mu_R/q)/4, B 0 = ( 1 − A V 0 ) /2 + A ( μ L / p + μ R / q ) /4 ,
Γ = A 2 ( 3 + μ L ) ( 3 + μ R ) / 16 − B 0 2 . \Gamma= A^2(3+\mu_L)(3+\mu_R)/16 - B_0^2. Γ = A 2 ( 3 + μ L ) ( 3 + μ R ) /16 − B 0 2 .
At fixed R , p , q R,p,q R , p , q , both mean derivatives of Γ \Gamma Γ are positive on the entire closed mean rectangle. Indeed the first, multiplied by 16 p / A 2 16p/A^2 16 p / A 2 , has its minimum at the upper mean corner. With k = 4 / R k = 4/R k = 4/ R , t = p q t = pq t = pq , that minimum is
( 1 − q ) ( 4 − t ) + k − t + 4 log ( t / k ) > 0 ; (1-q)(4-t) + k - t + 4\log(t/k) > 0; ( 1 − q ) ( 4 − t ) + k − t + 4 log ( t / k ) > 0 ;
the last two terms vanish at t = k t = k t = k and their derivative in t t t is − 1 + 4 / t > 0 -1 + 4/t > 0 − 1 + 4/ t > 0 . The other derivative is symmetric. Therefore
Γ ≥ 9 A 2 16 − ( 1 − A V 0 ) 2 4 . \Gamma\ge\frac{9A^2}{16} - \frac{(1-AV_0)^2}{4}. Γ ≥ 16 9 A 2 − 4 ( 1 − A V 0 ) 2 .
On 4 < R ≤ 7 4 < R \le7 4 < R ≤ 7 , the elementary bounds
log ( R / 4 ) ≤ V 0 ≤ R / 8 − 2 / R , A V 0 ≤ 11 / 16 < 1 \log(R/4) \le V_0 \le R/8 - 2/R, \qquad AV_0 \le11/16 < 1 log ( R /4 ) ≤ V 0 ≤ R /8 − 2/ R , A V 0 ≤ 11/16 < 1
permit squaring with the correct sign, giving Γ ≥ ϕ ( R ) \Gamma\ge\phi(R) Γ ≥ ϕ ( R ) , where
ϕ ( R ) = 9 R 2 16 ( R − 1 ) 2 − 1 4 ( 1 − R log ( R / 4 ) R − 1 ) 2 . \phi(R) = \frac{9R^2}{16(R-1)^2} - \frac{1}{4}\left(1-\frac{R\log(R/4)}{R-1}\right)^2. ϕ ( R ) = 16 ( R − 1 ) 2 9 R 2 − 4 1 ( 1 − R − 1 R log ( R /4 ) ) 2 .
Writing z = log ( R / 4 ) z = \log(R/4) z = log ( R /4 ) , one obtains
8 ( R − 1 ) 3 ϕ ′ ( R ) = − 9 R + 4 ( R − 1 − R z ) ( R − 1 − z ) . 8(R-1)^3\phi'(R) = -9R + 4(R-1-Rz)(R-1-z). 8 ( R − 1 ) 3 ϕ ′ ( R ) = − 9 R + 4 ( R − 1 − R z ) ( R − 1 − z ) .
Both factors are positive: the preceding logarithm bound gives R − 1 − R z ≥ R + 1 − R 2 / 8 ≥ 15 / 8 R-1-Rz \ge R+1-R^2/8 \ge15/8 R − 1 − R z ≥ R + 1 − R 2 /8 ≥ 15/8 and R − 1 − z > 2 R-1-z > 2 R − 1 − z > 2 . Substitute the lower bound z ≥ 2 ( R − 4 ) / ( R + 4 ) z \ge2(R-4)/(R+4) z ≥ 2 ( R − 4 ) / ( R + 4 ) to bound the right side by − ( R − 4 ) 2 ( 4 R 2 + R + 4 ) / ( R + 4 ) 2 ≤ 0 -(R-4)^2(4R^2+R+4)/(R+4)^2 \le0 − ( R − 4 ) 2 ( 4 R 2 + R + 4 ) / ( R + 4 ) 2 ≤ 0 . Four positive atanh terms give
log ( 7 / 4 ) > 381685152 682050985 , ϕ ( 7 ) > 446891067930101121 607599733733185600 > H ∗ . \log(7/4) > \frac{381685152}{682050985}, \qquad\phi(7) > \frac{446891067930101121}{607599733733185600} > H_*. log ( 7/4 ) > 682050985 381685152 , ϕ ( 7 ) > 607599733733185600 446891067930101121 > H ∗ .
Thus this domain would have actual Γ > H ∗ > 0 \Gamma> H_* > 0 Γ > H ∗ > 0 . An actual positive binary invariant supplies the future sheet on the same full lift; the original harmonic theorem gives Γ < H ∗ \Gamma< H_* Γ < H ∗ . This contradiction excludes 4 < R ≤ 7 4 < R \le7 4 < R ≤ 7 , so x < 1 / 6 x < 1/6 x < 1/6 . Only the actual lift is calibrated; the zero-mean comparison corner is not. All closed comparison boundaries used above are retained. □
Theorem 7.2 with 9 / 25 9/25 9/25 applies on every qualified proper returned eight-source. The stronger 3 / 7 3/7 3/7 version requires both additional density guards. They can be proved on separate genuine cuts and exported as physical state facts, but this does not identify the matrices or covectors of those separate lifts. Joint matrix arguments still need the full common-lift theorem.
Local physical-cost inequalities
The local estimates fall into four families: capped binary cores, forward pairs and their physical arms, returned cores and shared-arc unions, and central arcs coupled by actual switching responses. We state their usable forms here. § 3.17 and B.6 prove the cap-transport, four-arc, opposite-centre and nonpositive-tail estimates. The remaining local derivations are in Sections B.1–B.5, including finite rational or interval witnesses for compact residual domains. This distinction is important: the global argument adds physical costs, whereas the proofs of several local inequalities compare quantities belonging to one full lift.
The source and the charge
A source j 0 ⋯ j L − 1 j_0\cdots j_{L-1} j 0 ⋯ j L − 1 means L L L consecutive complete positive maximal arcs of the original minimizing trajectory, in chronological order. Its positions are numbered from zero. A charge I I I means c ( I ) = ∑ i ∈ I c i c(I)=\sum_{i\in I}c_i c ( I ) = ∑ i ∈ I c i , where c i = ∫ arc i ℓ d t c_i=\int_{\operatorname{arc} i}\ell\,\mathrm{d}t c i = ∫ arc i ℓ d t is the entire physical cost. Thus a positive cap or a returned arc may be required in the source without belonging to its charge. An interval [ a , b ] [a,b] [ a , b ] in a charge denotes the integer set { a , a + 1 , … , b } \{a,a+1,\ldots,b\} { a , a + 1 , … , b } .
Each result retains its stated scope: explicit local hypotheses take precedence over the complete-source convention below. Statements and proof steps on partial arcs or algebraic, analytic, comparison or relaxed domains require only their stated data; comparison points carry no actual-source or lift hypothesis unless specified. For complete sources we retain the original state and frame endpoints, real winding, compact geometric restrictions and measurable-control maximum principle of § 3.1 . A source is first placed in a genuine containing fundamental interval, and a nonzero full maximizing lift in the sense of Definition 2.3 is selected there. Normality is established before jointly dividing the objective multiplier, the matrix adjoint and both state covectors by its positive value. These operations are never performed separately on the individual leaves of a source. For a proper source, L < B L<B L < B , a cut inside the preceding positive arc gives the needed containment. Where two sources must share a lift across different cuts, that extra assertion is proved explicitly below.
A lawful image means a whole cyclic symmetry, or a whole reflection combined with reversal of physical time. It acts on the state, frame, matrix adjoint and both state covectors, and transports the charge with its source. A bare permutation of control names is not an additional symmetry. In particular, a displayed source is not a statement about an isolated control word with freely assigned endpoint data.
Capped cores and their common binary algebra
The binary family uses actual roots, exterior maximizing jets and the same matrix adjoint throughout each source. The derivation is given in Section B.1; it includes the order in which normality, positive determinant, positive trace and harmonic calibration are proved. The bounds below charge complete physical arcs rather than a scalar relaxation of a middle-root mean.
Theorem 8.1 (Capped core costs). Under the complete-source hypotheses above, the following costs have the stated bounds. Every listed source includes its uncharged exterior arcs.
(i) The sources 02010, 10121, 21202 have c ( [ 1 , 3 ] ) > 49 / 200 c([1,3])>49/200 c ([ 1 , 3 ]) > 49/200 .
(ii) The sources 102021, 210102, 021210 have c ( [ 1 , 4 ] ) > 389 / 1000 c([1,4])>389/1000 c ([ 1 , 4 ]) > 389/1000 .
(iii) The source 10210, and its lawful images, has c ( [ 1 , 3 ] ) ≥ 3 / 5 c([1,3])\ge3/5 c ([ 1 , 3 ]) ≥ 3/5 .
(iv) The source 102102, and its lawful images, has c ( [ 1 , 4 ] ) > 3 / 5 > 62 / 125 c([1,4])>3/5>62/125 c ([ 1 , 4 ]) > 3/5 > 62/125 .
(v) The source 012021, and its lawful images, has c ( [ 1 , 4 ] ) > 81 / 200 c([1,4])>81/200 c ([ 1 , 4 ]) > 81/200 .
The third assertion is weak. Its equality case is not discarded.
Proof: Theorems B.20 , B.14 , B.47, Proposition B.40, and Theorem B.83.
Theorem 8.2 (One returned four-core). Put
E ∗ : = 4154032208439 8759490110000 . E_* := \frac{4154032208439}{8759490110000}. E ∗ := 8759490110000 4154032208439 . The right sources 1020212, 2101020, 0212101 satisfy c ( [ 1 , 4 ] ) > E ∗ c([1,4])>E_* c ([ 1 , 4 ]) > E ∗ . The left sources 0102021, 1210102, 2021210 satisfy $c([2,5])>E_. The two exterior caps and the returned arc are positive and uncharged.
Proof: Theorem B.16 .
Theorem 8.3 (A four-core with two returns). For the complete source 01020212 on an original minimizer, with both actual returned mean leaves and both exterior caps retained in a genuine fundamental interval, the charge on arcs 2, 3, 4, 5 satisfies
c ( [ 2 , 5 ] ) > 74 125 > 11 20 > 261 500 . c([2,5]) > \frac{74}{125} > \frac{11}{20} > \frac{261}{500}. c ([ 2 , 5 ]) > 125 74 > 20 11 > 500 261 . The same statement holds for whole lawful cyclic images. Such a source is automatically proper by Lemma B.1 .
Proof: Section C.6.2 and Theorem B.19 . The correlated shear envelope is stated independently in Lemma C.3.
Forward pairs and four physical arms
A forward pair needs its complete surrounding caps even when only its two central arcs are paid. The two-arm estimate has a different support, although some of its numerical consequences coincide. Their proofs are grouped in Section B.2.
Theorem 8.4 (The forward central pair). In the complete positive source 012012, and its lawful images,
c 2 + c 3 > 180483 500000 > 20901 62500 . c_2+c_3 > \frac{180483}{500000} > \frac{20901}{62500}. c 2 + c 3 > 500000 180483 > 62500 20901 . The four supporting arcs 0, 1, 4, 5 remain complete and uncharged.
Proof: Proposition B.37.
Theorem 8.5 (Four own arcs). The complete six-sources 101202, 212010, 020121 satisfy
c ( [ 1 , 4 ] ) > 103 250 . c([1,4]) > \frac{103}{250}. c ([ 1 , 4 ]) > 250 103 . The seven-source 1012012 has this bound on [ 1 , 4 ] [1,4] [ 1 , 4 ] , and 0120121 has it on [ 2 , 5 ] [2,5] [ 2 , 5 ] , with all supporting arcs retained.
Proof: Theorem B.122, with the actual return and capped-forward endpoint signs established there.
Proposition 8.6 (Scalar costs used in the finite catalogue). On the complete positive source 01201, and either whole cyclic image,
c ( [ 1 , 3 ] ) > 47 400 . c([1,3]) > \frac{47}{400}. c ([ 1 , 3 ]) > 400 47 . On the complete positive source 10120121, and either whole cyclic image,
c ( [ 1 , 6 ] ) > 609 1000 . c([1,6]) > \frac{609}{1000}. c ([ 1 , 6 ]) > 1000 609 . All supporting arcs and the genuine full-lift conditions above are retained.
Proofs: Proposition B.36 and Theorem B.32. These are the exact thresholds used by the finite source catalogue.
Returns, entrance geometry, and shared unions
A returned tail retains more information than a generic three-core. It is essential to say which extra geometric inequality is supplied by its source.
Theorem 8.7 (Nonpositive and returned tails). Consider the complete capped core 1 ∣ 202 ∣ 1, whose actual entrance is u = − a ≤ 0 u=-a\leq0 u = − a ≤ 0 .
(i) On 12021, the charge [ 1 , 3 ] [1,3] [ 1 , 3 ] exceeds 61 200 \frac{61}{200} 200 61 .
(ii) On 120212, the same charge exceeds 379 1000 \frac{379}{1000} 1000 379 .
The return in (ii) is the complete arc at position 5. No entrance-circle or LOW-mean condition is required for (ii). In particular it implies both weaker bounds 361/1000 and 3/8. The full source 012021 gives c ( [ 2 , 4 ] ) > 61 / 200 c([2,4])>61/200 c ([ 2 , 4 ]) > 61/200 ; 0120212 gives c ( [ 2 , 4 ] ) > 379 / 1000 c([2,4])>379/1000 c ([ 2 , 4 ]) > 379/1000 . The full source 202120212 gives c ( [ 4 , 6 ] ) > 379 / 1000 c([4,6])>379/1000 c ([ 4 , 6 ]) > 379/1000 ; its complete reflected-time image 010201020 gives c ( [ 2 , 4 ] ) > 379 / 1000 c([2,4])>379/1000 c ([ 2 , 4 ]) > 379/1000 . In all cases the entire sign-producing context, both caps and the stated actual return are retained.
Proofs: Theorem B.127, Theorem B.73, and Corollaries B.94 and B.96.
Theorem 8.8 (Shared-arc and shared-derivative unions). The following are once-counted physical charges of complete sources:
(i) 02120102 has c ( [ 1 , 6 ] ) > 71 / 100 c([1,6])>71/100 c ([ 1 , 6 ]) > 71/100 (hence > 63 / 100 >63/100 > 63/100 );
(ii) 20212101020 has c ( [ 2 , 8 ] ) > 153 / 200 c([2,8])>153/200 c ([ 2 , 8 ]) > 153/200 ;
(iii) 0102010201020 has c ( [ 2 , 8 ] ) > 79483 / 100000 c([2,8])>79483/100000 c ([ 2 , 8 ]) > 79483/100000 .
All means, switching derivatives and matrix comparisons used within an individual assertion belong to one selected full normal lift.
Proofs: Theorems B.128, B.98, and B.102.
Actual central contacts and the terminal comparison
The complete-terminal and prefix estimates have different sufficient interfaces. Let a complete mode-1 arc be entered at 2 → 1 2 \to1 2 → 1 and exited at 1 → 0 1 \to0 1 → 0 on one full normal lift. Its rate k = 2 v u 2 + v 2 − 1 > 0 k = \frac{2v}{u^2+v^2-1}>0 k = u 2 + v 2 − 1 2 v > 0 is constant. In the whole cyclic chart making it mode 0, put ξ = 1 + u n e w \xi=1+u_{\mathrm{new}} ξ = 1 + u new , v n e w = k ξ v_{\mathrm{new}}=k\xi v new = k ξ , ξ L = A \xi_L=A ξ L = A , ξ R = A E \xi_R=AE ξ R = A E , E = e k d > 1 E=e^{kd}>1 E = e k d > 1 and τ = 1 + k 2 \tau=\sqrt{1+k^2} τ = 1 + k 2 . The actual domain is A E < 2 AE<2 A E < 2 , A τ 2 > 2 A\tau^2>2 A τ 2 > 2 . The raw switching response q q q is given by the complete solution (B.162)–(B.163). The scaled switching functions U , V U,V U , V in that solution satisfy the actual contact conditions
V ( A ) = U ( A E ) = 0 , V ξ ( A ) ≤ 0 , U ξ ( A E ) ≤ 0. (37) V(A)=U(AE)=0,\qquad V_{\xi}(A)\le0,\qquad U_{\xi}(AE)\le0. \tag*{(37)} V ( A ) = U ( A E ) = 0 , V ξ ( A ) ≤ 0 , U ξ ( A E ) ≤ 0. (37)
They are one-sided derivatives from the central arc, in the increasing coordinate ξ \xi ξ ; U , V U,V U , V are not physical state coordinates.
An actual preceding mode-2 tail of full duration d 0 > 0 d_0>0 d 0 > 0 supplies strict ingress if
u L > 0 , q L < 0 , q L ′ + k q L < b L 1 − e − b L d 0 q L , b L = v L 1 − u L > 0. (38) u_L>0,\qquad q_L<0,\qquad q'_L+kq_L<\frac{b_L}{1-e^{-b_Ld_0}}q_L,\qquad b_L=\frac{v_L}{1-u_L}>0. \tag*{(38)} u L > 0 , q L < 0 , q L ′ + k q L < 1 − e − b L d 0 b L q L , b L = 1 − u L v L > 0. (38)
A complete initial 202 supplies these data. Regeneration must prove them again at the new actual entrance, with its own rate and duration. The central proofs are collected in Section B.4.
Lemma 8.9 (The actual terminal inequality). Suppose this central exit is followed by a complete positive 020 on the same lift, with u R < 0 u_R<0 u R < 0 . Then
q R > 0 , q R ′ < ( k − a R ) q R , a R = v R 1 + u R > 0. q_R>0,\qquad q'_R<(k-a_R)q_R,\qquad a_R=\frac{v_R}{1+u_R}>0. q R > 0 , q R ′ < ( k − a R ) q R , a R = 1 + u R v R > 0. Proof: Lemma B.108. Its weighting has the variable rate w ′ / w = 2 v / ( u 2 + v 2 − 1 ) w'/w=2v/(u^2+v^2-1) w ′ / w = 2 v / ( u 2 + v 2 − 1 ) ; only its actual junction value equals k k k . Identity (B.184) identifies the same raw derivative across the switch.
Theorem 8.10 (A complete seven-arc center). A properly retained complete positive source 2021020 satisfies
c 3 > 299 / 250. c_3>299/250. c 3 > 299/250. Only its complete central arc is charged. More generally, the same conclusion holds for a complete central arc satisfying (8.2), both actual contact jets (8.1), and the complete same-lift terminal 020 in Lemma 8.9.
Proof: Section C.8.2, using the common central equations and terminal identities of Section B.4.
Prefix propagation and disjunctive central costs.
Theorem 8.11 (Actual prefix propagation). For a complete actual central arc entered at 2 → 1 2 \to1 2 → 1 and exited at 1 → 0 1 \to0 1 → 0 , retain one full normal maximizing lift and the actual strict ingress ((38) ). If its physical cost satisfies C l o c ≤ 13 / 20 C_{\mathrm{loc}} \le13/20 C loc ≤ 13/20 , then
u R < 0 , q R ≥ 0 , q R ′ > k q R . u_R < 0,\qquad q_R \ge0,\qquad q'_R > kq_R. u R < 0 , q R ≥ 0 , q R ′ > k q R . The actual contacts supply both jets ((37) ) by full maximization. No terminal word is assumed. The same conclusion holds, in particular, under the weaker budget C l o c ≤ 74 / 125 C_{\mathrm{loc}} \le74/125 C loc ≤ 74/125 . If a complete actual terminal 020 is available on the same lift, its comparison contradicts either small budget.
Proof: Theorem B.107; the automatic contact conditions and the weaker-budget consequence are recorded in Theorem B.113.
Corollary 8.12 (A regenerated alternative). For the complete proper ten-source 2021021020,
max ( c 3 , c 6 ) > 13 / 20. \max(c_3,c_6) > 13/20. max ( c 3 , c 6 ) > 13/20. If its two reverse four-core sources retain the hypotheses of Theorem 8.1 (iv), then
c ( [ 2 , 7 ] ) > 13 / 20 + 62 / 125 = 573 / 500. c([2,7]) > 13/20 + 62/125 = 573/500. c ([ 2 , 7 ]) > 13/20 + 62/125 = 573/500.
Proof: Lemma B.110 and Corollary B.111. If c 3 > 13 / 20 c_3 > 13/20 c 3 > 13/20 , pay the disjoint reverse-four charge [ 4 , 7 ] [4,7] [ 4 , 7 ] ; if c 6 > 13 / 20 c_6 > 13/20 c 6 > 13/20 , pay [ 2 , 5 ] [2,5] [ 2 , 5 ] . The branches are not added.
A paired argument requires one common lift for the two rooted tails. The following criterion supplies it from genuine fundamental intervals; it is stronger information than the independent validity of two scalar cost inequalities.
Lemma 8.13 (A sufficient common-lift criterion). Let a < b < a + T a < b < a + T a < b < a + T be two actual cuts. Suppose [ b , a + T ] [b,a+T] [ b , a + T ] contains a complete positive capped-binary core with two internal switching roots, and ( a , b ) (a,b) ( a , b ) contains a switch of distinct modes. The full lifts on the two genuine fundamental intervals can be jointly normalized, agree on their overlap, and extend to one normalized full reduced maximizing lift on the actual unfolding.
Proof: Lemma B.3 and Proposition B.114.
Theorem 8.14 (Paired rooted responses). Suppose one normal full maximizing lift contains the complete positive source 21021, with middle mode-0 connector. Take its right rooted mode-2 tail, and take the corresponding opposite rooted tail after the lawful full reversed-time transformation. For a tail whose entrance and exit horizontal coordinates are U U U , V V V , define
Ψ ( U , V ) = 2 ( 1 − U ) log 1 − U 1 − V − ( V − U ) ( 2 + V − U ) . \Psi(U,V)=2(1-U)\log\frac{1-U}{1-V}-(V-U)(2+V-U). Ψ ( U , V ) = 2 ( 1 − U ) log 1 − V 1 − U − ( V − U ) ( 2 + V − U ) . At least one of the two full-duration responses is strictly positive. If both corresponding ordered extensions are complete proper 021020 sources, their two central physical costs satisfy
max ( c − , c + ) > 299 / 250. \max(c_-,c_+) > 299/250. max ( c − , c + ) > 299/250. The extensions retain their actual endpoints, compact geometry and full maximizing lifts. If reanchoring is needed for the selected extension, its positive response is state-determined and the ingress is rederived on that complete lift. The other center has no separate lower bound asserted by this theorem.
Proof: Theorem B.117, Lemma B.118, and Corollary B.119.
8.6.0.1. Proof order and finite arithmetic. Sections B.1–B.4, placed after the global assembly, prove these statements from the full equations. The remaining closed-domain sign comparisons are in Section B.5. Polynomial coefficients or interval cells are used only after the actual source has been shown to satisfy their full domain. They cannot supply a missing cap, a missing return, or a common lift. The weak proper-five inequality remains weak, and every maximum inequality is used branch by branch.
Reusable strengthened estimates
Each numerical bound is used with its complete support, physical charge and lift hypotheses. We collect the stronger forms that enter the disjoint-charge assemblies.
Returned sources and shared charges
Theorem 8.3 pays more than 74/125 on the four interior arcs of the complete two-return source 01020212. Theorem 8.7 pays more than 379/1000 on the capped returned triple with nonpositive entrance. The latter does not assume an entrance circle or a LOW mean. When both actual circles and the full returned-tail interface are available, Proposition 10.1 improves the three-arc payment to 2/5. These hypotheses cannot be inferred from a sign word alone.
The complete shared source 02120102 pays more than 71/100 on its six interior physical arcs by Theorem B.128. With both actual returns in the longer proper source 2021201020, Proposition C.12 pays more than 807/1000 on the same six physical positions. Its proof retains a common switching jet, selects one actual LOW mean, and counts each triple once. The larger number is not available on the shorter support without those returns.
A stronger returned-terminal union
The proper complete nine-source 012021201 satisfies c 2 + c 3 + c 4 + c 5 + c 6 ≥ 180483 / 250000 c_2 + c_3 + c_4 + c_5 + c_6 \ge180483/250000 c 2 + c 3 + c 4 + c 5 + c 6 ≥ 180483/250000 by the local transfer in Lemma D.3 and its first-contact contradiction. Both returned contexts, both complete forward entrance witnesses, all outgoing leaves and ties, and one genuine containing interval are required. The bound is weak; its equality case is not excluded.
Corollary B.92 places the same physical charge on positions 3, …, 7 of 01201020120. It implies the strict 71/125 + k k k /100 consequences for $k = 0, 1, 2 without a LOW predicate. The structural bridge and mean comparisons are proved in Lemma B.91. The local H3 theorem retains all its analytic and finite prerequisites.
A half-unit charge from two crossed circles
Theorem E.17 applies to the middle triple of a complete positive 02010 source or a lawful whole image. Its original unsorted parameters must satisfy
E a > max ( 1 , b ) , E b > max ( 1 , a ) , Ea > \max(1,b), \qquad Eb > \max(1,a), E a > max ( 1 , b ) , E b > max ( 1 , a ) ,
and its two caps, both true leaf means and all maximizing conditions must belong to one full normal lift on a genuine containing interval. It gives α + β + γ > 1 / 2 \alpha+ \beta+ \gamma> 1/2 α + β + γ > 1/2 . Its reconstruction, scalar boundary proofs and final actual-lift calibration are supplied in Sections E.2 and E.3.
The complete nine-source of Lemma 12.2 retains both seven-source witnesses on one selected full lift and supplies these conditions. A bare five-source does not supply them.
A two-arc payment from bilateral prefix forcing
Theorem E.2 gives more than 13/20 on local arcs 3, 4 of the complete source 202101212. All eight arcs must lie in one genuine fundamental interval. This includes a proper occurrence and an eight-arc occurrence retained in a genuine switch-cut period. The two strict prefix forcings act on the same actual capped pair. The proof obtains positive determinant first, then the future sheet from a true binary mean, and only then uses harmonic calibration. It neither identifies independent lifts nor calibrates an algebraic comparison endpoint.
This source theorem supplies a general two-arc row for the uniform capacity reduction and the qualified pair charges in the length-five negative-run family.
Own pairs and disjoint adjacent pairs
Theorem 7.2 gives Q > 9 / 25 Q > 9/25 Q > 9/25 on a complete proper 01020212 source. Both caps and the actual chronological returns are retained. Its proof combines the two actual cap contributions through one full-matrix transport identity. Two additional qualified six-sources giving both anchor densities below 1 / 6 1/6 1/6 improve the same physical charge to Q > 3 / 7 Q > 3/7 Q > 3/7 .
The weaker 283/1000 threshold in fixed source assemblies is a consequence of this theorem; its structural mean, density and shear identities remain available independently. On the proper eleven-source 01020212101, two disjoint own pairs give the five-arc charge > 18 / 25 > 18/25 > 18/25 in Corollary 7.3.
Theorem B.122 also gives 103/250 on the complete four-arc charges of 101202 and 1012012, and on the corresponding charge of 0120121. The actual return and forward-forcing contexts supply its endpoint signs. Its proof, the opposite-centre payment 63/500, and the nonpositive-tail payment 61/200 are given in Section B.6; the last uses two displayed polynomial minorants in addition to analytic comparisons.
From the fees to the exclusions
The applications use each bound only after matching its complete support and physical charge. The weak 3/5 proper-five estimate retains its weak inequality. Section H.2 specifies the finite proof obligations and their arithmetic contracts.
Two estimates for complete returned supports
These local statements use the independently established returned-tail interface, loss obstruction, LOW bound and HIGH absorption. They do not use the returned-word exclusions or the finite census conclusion. Their individual source dependencies remain those stated below.
The endpoint information of a returned support is stronger than that of an isolated binary core. We first use both endpoint circles to estimate a returned triple, and then retain the density dependence of the two cap defects of a returned own pair. The inequalities in this section are statements about complete supports with the interfaces specified below; arbitrary formal triples do not acquire these hypotheses merely by having the same labels.
The returned triple interface
Consider the complete positive source 120212 in a genuine containing interval. In its capped 202 triple write the entrance as − a ≤ 0 -a \le0 − a ≤ 0 , the internal roots as r < s r < s r < s , the exit as b b b , and the actual middle mean as μ \mu μ . The actual root/mean geometry gives r 2 ≤ μ < b 2 r^2 \le\mu< b^2 r 2 ≤ μ < b 2 and s > μ > 0 s > \sqrt{\mu} > 0 s > μ > 0 Lemma B.67 and its root identities. Its physical cost is T T T . Put
q ( u ) = v ( u ) − 2 , σ = ∫ − a b q ( u ) d u , Z = 3 + μ 2 σ , X = f ( − a ) σ , Y = − f ( b ) σ , q(u)=v(u)^{-2},\qquad\sigma=\int_{-a}^{b}q(u)\,\mathrm{d}u,\qquad Z=\frac{3+\mu}{2\sigma},\qquad X=\frac{f(-a)}{\sigma},\qquad Y=-\frac{f(b)}{\sigma}, q ( u ) = v ( u ) − 2 , σ = ∫ − a b q ( u ) d u , Z = 2 σ 3 + μ , X = σ f ( − a ) , Y = − σ f ( b ) ,
L = ( X + Y + Z ) 2 − 4 Z Y − μ , Δ 0 = 59 1000 . \mathcal{L}=(X+Y+Z)^2-4ZY-\mu,\qquad\Delta_0=\frac{59}{1000}. L = ( X + Y + Z ) 2 − 4 Z Y − μ , Δ 0 = 1000 59 .
The function f f f is the binary switching response in the original normalized lift. We use the returned-tail interface and loss obstruction proved in Appendices B–E: the actual signed strip has Y , Z > 0 Y,Z>0 Y , Z > 0 , − Z < X + Y < Z -Z<X+Y<Z − Z < X + Y < Z , and L ≥ Δ 0 \mathcal{L}\ge\Delta_0 L ≥ Δ 0 . Its LOW estimate gives T > 407 / 1000 T>407/1000 T > 407/1000 when μ ≤ a 2 \mu\le a^2 μ ≤ a 2 . On the HIGH branch a 2 < μ < b 2 a^2<\mu<b^2 a 2 < μ < b 2 , the returned-HIGH absorption result states that w = a + b ≤ 1 / 2 w=a+b\le1/2 w = a + b ≤ 1/2 and Z < 1 / 4 Z<1/4 Z < 1/4 imply L < Z 2 \mathcal{L}<Z^2 L < Z 2 . This last implication includes the exact polynomial comparison described in Appendix D.§ 3.30 ; it is not being replaced by a numerical plot or a sample of mean values.
We additionally retain the two actual circles
v ( − a ) 2 > ( 1 − a ) ( 3 + a ) , v ( b ) 2 > ( 1 − b ) ( 3 + b ) . (39) v(-a)^2>(1-a)(3+a),\qquad v(b)^2>(1-b)(3+b). \tag*{(39)} v ( − a ) 2 > ( 1 − a ) ( 3 + a ) , v ( b ) 2 > ( 1 − b ) ( 3 + b ) . (39)
The first is an entrance hypothesis; the last follows from the complete return. They are not interchangeable endpoint assertions.
Proposition 10.1. A complete returned triple satisfying this interface and (10.1) has T > 2 / 5 T > 2/5 T > 2/5 .
Proof. Suppose T ≤ 2 / 5 T \le2/5 T ≤ 2/5 . The LOW estimate leaves only the HIGH branch. Write w = a + b w=a+b w = a + b . The exact physical moment identity gives
0 ≤ a < b , 0 < w < 4 / 5 , Z < ( 3 + μ ) ( 4 / 5 − w ) 6 . (40) 0 \le a < b,\qquad0 < w < 4/5,\qquad Z < \frac{(3+\mu)(4/5-w)}{6}. \tag*{(40)} 0 ≤ a < b , 0 < w < 4/5 , Z < 6 ( 3 + μ ) ( 4/5 − w ) . (40)
Both logarithmic density slopes belong to [ − 2 / ( 1 + u ) , 2 / ( 1 − u ) ] [-2/(1+u),2/(1-u)] [ − 2/ ( 1 + u ) , 2/ ( 1 − u )] . Forward comparison from the entrance and backward comparison from the exit, using the same actual density, give
q ( u ) < min { A ( 1 − u ) 2 , B ( 1 + u ) 2 } , A = ( 1 + a ) 2 ( 1 − a ) ( 3 + a ) , B = ( 1 + b ) 2 ( 1 − b ) ( 3 + b ) . q(u)<\min\left\{\frac{A}{(1-u)^2},\frac{B}{(1+u)^2}\right\},\qquad A=\frac{(1+a)^2}{(1-a)(3+a)},\qquad B=\frac{(1+b)^2}{(1-b)(3+b)}. q ( u ) < min { ( 1 − u ) 2 A , ( 1 + u ) 2 B } , A = ( 1 − a ) ( 3 + a ) ( 1 + a ) 2 , B = ( 1 − b ) ( 3 + b ) ( 1 + b ) 2 .
The crossing u ∗ = ( B − A ) / ( B + A ) u_*=(\sqrt{B}-\sqrt{A})/(\sqrt{B}+\sqrt{A}) u ∗ = ( B − A ) / ( B + A ) lies in ( − a , b ) (-a,b) ( − a , b ) . At the left endpoint this follows from ( 1 − a ) ( 3 + a ) ≥ 3 ( 1 − a ) 2 (1-a)(3+a)\ge3(1-a)^2 ( 1 − a ) ( 3 + a ) ≥ 3 ( 1 − a ) 2 ; the right endpoint is the reflected comparison. The excluded case a = b = 0 a=b=0 a = b = 0 is the only simultaneous equality. Integrating on the two sides of u ∗ u_* u ∗ yields
σ < A B − 1 + a 2 ( 3 + a ) − 1 + b 2 ( 3 + b ) = : S ( a , b ) . (41) \sigma<\sqrt{AB}-\frac{1+a}{2(3+a)}-\frac{1+b}{2(3+b)}=:S(a,b). \tag*{(41)} σ < A B − 2 ( 3 + a ) 1 + a − 2 ( 3 + b ) 1 + b =: S ( a , b ) . (41)
For 0 < w ≤ 1 / 3 0<w\le1/3 0 < w ≤ 1/3 put p = a b ∈ [ 0 , w 2 / 4 ] p=ab\in[0,w^2/4] p = ab ∈ [ 0 , w 2 /4 ] . Then
S = 1 + w + p F − 3 + 2 w + p 9 + 3 w + p , F = 9 − 6 w − 3 w 2 + ( 10 + 2 w ) p + p 2 , S=\frac{1+w+p}{\sqrt{F}}-\frac{3+2w+p}{9+3w+p},\qquad F=9-6w-3w^2+(10+2w)p+p^2, S = F 1 + w + p − 9 + 3 w + p 3 + 2 w + p , F = 9 − 6 w − 3 w 2 + ( 10 + 2 w ) p + p 2 ,
S p = 4 ( 1 − 3 w − w 2 + p ) F 3 / 2 − 6 + w ( 9 + 3 w + p ) 2 ≤ 4 27 . S_p=\frac{4(1-3w-w^2+p)}{F^{3/2}}-\frac{6+w}{(9+3w+p)^2}\le\frac{4}{27}. S p = F 3/2 4 ( 1 − 3 w − w 2 + p ) − ( 9 + 3 w + p ) 2 6 + w ≤ 27 4 .
Indeed, if the first numerator is positive, use 1 − 3 w − w 2 + p ≤ 1 − 3 w 1-3w-w^2+p\le1-3w 1 − 3 w − w 2 + p ≤ 1 − 3 w , F ≥ 9 ( 1 − w ) F\ge9(1-w) F ≥ 9 ( 1 − w ) and 1 − 3 w ≤ ( 1 − w ) 3 / 2 1-3w\le(1-w)^{3/2} 1 − 3 w ≤ ( 1 − w ) 3/2 . The last inequality follows by squaring: ( 1 − w ) 3 − ( 1 − 3 w ) 2 = w ( 3 − 6 w − w 2 ) ≥ 0 (1-w)^3-(1-3w)^2=w(3-6w-w^2)\ge0 ( 1 − w ) 3 − ( 1 − 3 w ) 2 = w ( 3 − 6 w − w 2 ) ≥ 0 . If that numerator is nonpositive the claimed derivative bound is immediate. At p = 0 p=0 p = 0 the integral representation in (10.3) is increasing in w w w : one endpoint coefficient is fixed, the other increases, and the positive integration interval extends. Hence
S ( w , p ) ≤ S ( 1 / 3 , 0 ) + 1 / 243 = 2 15 − 11 30 + 1 243 < 749 4860 < 31 200 . S(w,p)\le S(1/3,0)+1/243=\frac{2}{\sqrt{15}}-\frac{11}{30}+\frac{1}{243}<\frac{749}{4860}<\frac{31}{200}. S ( w , p ) ≤ S ( 1/3 , 0 ) + 1/243 = 15 2 − 30 11 + 243 1 < 4860 749 < 200 31 .
Here ( 31 / 60 ) 2 − 4 / 15 = 1 / 3600 (31/60)^2-4/15=1/3600 ( 31/60 ) 2 − 4/15 = 1/3600 and the final gap is 43 / 48600 43/48600 43/48600 . Since μ < w 2 ≤ 1 / 9 \mu<w^2\le1/9 μ < w 2 ≤ 1/9 , this gives
Z < 217 900 < 1 4 , Δ 0 − ( 217 900 ) 2 = 701 810000 > 0. Z<\frac{217}{900}<\frac14,\qquad\Delta_0-\left(\frac{217}{900}\right)^2=\frac{701}{810000}>0. Z < 900 217 < 4 1 , Δ 0 − ( 900 217 ) 2 = 810000 701 > 0.
For w ≥ 1 / 3 w\ge1/3 w ≥ 1/3 , the right side of (10.2) is bounded by h ( w ) = ( 3 + w 2 ) ( 4 / 5 − w ) / 6 h(w)=(3+w^2)(4/5-w)/6 h ( w ) = ( 3 + w 2 ) ( 4/5 − w ) /6 , whose derivative is negative. Therefore
Z < 98 405 < 1 4 , Δ 0 − ( 98 405 ) 2 = 2939 6561000 > 0. Z<\frac{98}{405}<\frac14,\qquad\Delta_0-\left(\frac{98}{405}\right)^2=\frac{2939}{6561000}>0. Z < 405 98 < 4 1 , Δ 0 − ( 405 98 ) 2 = 6561000 2939 > 0.
In particular the entire range w ≤ 1 / 2 w\le1/2 w ≤ 1/2 contradicts HIGH absorption.
It remains 1 / 2 ≤ w < 4 / 5 1/2\le w<4/5 1/2 ≤ w < 4/5 . Put I 0 = ( r + a ) / ( ( 1 − r ) ( 1 + a ) ) I_0=(r+a)/((1-r)(1+a)) I 0 = ( r + a ) / (( 1 − r ) ( 1 + a )) . The actual entrance-response inequality of Lemma B.62 is
X ≤ μ − a 2 2 L 0 + 2 G 0 μ 3 + a , G 0 = I 0 ( 1 − a 2 ) D 0 < 1 1 − a 2 , X\le\frac{\mu-a^2}{2L_0}+\frac{2G_0\mu}{3+a},\qquad G_0=\frac{I_0}{(1-a^2)D_0}<\frac{1}{1-a^2}, X ≤ 2 L 0 μ − a 2 + 3 + a 2 G 0 μ , G 0 = ( 1 − a 2 ) D 0 I 0 < 1 − a 2 1 ,
L 0 = ( 1 + a ) 2 D 0 , L_0=(1+a)^2D_0, L 0 = ( 1 + a ) 2 D 0 ,
D 0 = r + a ( 1 − r ) ( 1 + a ) + ( 1 + r ) ( b − r ) ( 1 − r ) 2 ( 1 + b ) . D_0=\frac{r+a}{(1-r)(1+a)}+\frac{(1+r)(b-r)}{(1-r)^2(1+b)}. D 0 = ( 1 − r ) ( 1 + a ) r + a + ( 1 − r ) 2 ( 1 + b ) ( 1 + r ) ( b − r ) .
At fixed a , b a,b a , b , ( D 0 ) r = 4 ( b − r ) / [ ( 1 + b ) ( 1 − r ) 3 ] ≥ 0 (D_0)_r=4(b-r)/[(1+b)(1-r)^3]\ge0 ( D 0 ) r = 4 ( b − r ) / [( 1 + b ) ( 1 − r ) 3 ] ≥ 0 . Since r > − a r>-a r > − a , we obtain L 0 ≥ w ( 1 − a ) / ( 1 + b ) L_0\ge w(1-a)/(1+b) L 0 ≥ w ( 1 − a ) / ( 1 + b ) . Now a < w / 2 < 2 / 5 a<w/2<2/5 a < w /2 < 2/5 , b < 4 / 5 b<4/5 b < 4/5 , and consequently
X / μ < 3 + 50 / 63 = 239 / 63 < 4. (42) X/\mu<3+50/63=239/63<4. \tag*{(42)} X / μ < 3 + 50/63 = 239/63 < 4. (42)
Also (40) gives Z < ( 3 + μ ) / 20 Z<(3+\mu)/20 Z < ( 3 + μ ) /20 . For μ ≥ 1 / 25 \mu\ge1/25 μ ≥ 1/25 , the clipped signed-strip bound gives
L < 4 Z 2 − μ < ( 3 + μ ) 2 100 − μ ≤ 3276 62500 < Δ 0 , \mathcal{L}<4Z^2-\mu<\frac{(3+\mu)^2}{100}-\mu\le\frac{3276}{62500}<\Delta_0, L < 4 Z 2 − μ < 100 ( 3 + μ ) 2 − μ ≤ 62500 3276 < Δ 0 ,
where the last gap is 823 / 125000 823/125000 823/125000 ; the comparison function decreases on [ 1 / 25 , 16 / 25 ] [1/25,16/25] [ 1/25 , 16/25 ] . For 0 < μ ≤ 1 / 25 0<\mu\le1/25 0 < μ ≤ 1/25 , (42) and Y > 0 Y>0 Y > 0 give the unclipped bound
L < ( Z + 4 μ ) 2 − μ < ( 3 + 81 μ 20 ) 2 − μ . \mathcal{L}<(Z+4\mu)^2-\mu<\left(\frac{3+81\mu}{20}\right)^2-\mu. L < ( Z + 4 μ ) 2 − μ < ( 20 3 + 81 μ ) 2 − μ .
The last expression is convex. Its endpoint values 9 / 400 9/400 9/400 and 896 / 15625 896/15625 896/15625 are below Δ 0 \Delta_0 Δ 0 , with gaps 73 / 2000 73/2000 73/2000 and 207 / 125000 207/125000 207/125000 , respectively. Every remaining HIGH point is therefore impossible. The width seams were included throughout. □
Density-dependent payment for an own pair
Consider a proper complete source 01020212 with both actual returned mean leaves and both exterior caps. Charge only its local arcs 3, 4. Write their cost as Q Q Q , their genuine height as h > 0 h>0 h > 0 , their shear as T > 0 T>0 T > 0 , and put K = ( 3 + h 2 ) / 2 K=(3+h^2)/2 K = ( 3 + h 2 ) /2 . The original same-lift identities are
Q = w / 2 + K T , w > 3 h , N = e L + e R = K ( T − E ) − h 2 2 ( x L + x R ) , (43) Q=w/2+KT,\qquad w>3h,\qquad N=e_L+e_R=K(T-E)-\frac{h^2}{2}(x_L+x_R), \tag*{(43)} Q = w /2 + K T , w > 3 h , N = e L + e R = K ( T − E ) − 2 h 2 ( x L + x R ) , (43)
where 0 < x L , x R < 47 / 250 0<x_L,x_R<47/250 0 < x L , x R < 47/250 and e L , e R ≥ 0 e_L,e_R\ge0 e L , e R ≥ 0 . Writing δ = 3 / 4 − Γ \delta=3/4-\Gamma δ = 3/4 − Γ for the determinant gap of this same harmonically qualified lift, the cap gaps obey
29 2000 < δ ≤ 3 4 ( 1 − 3 x j ) e j + e j 2 . \frac{29}{2000}<\delta\le\frac{3}{4}(1-3x_j)e_j+e_j^2. 2000 29 < δ ≤ 4 3 ( 1 − 3 x j ) e j + e j 2 .
Here E = x L a / ( 1 − a ) + x R b / ( 1 − b ) E=x_La/(1-a)+x_Rb/(1-b) E = x L a / ( 1 − a ) + x R b / ( 1 − b ) is the cap-density term, where − a -a − a and b b b are the two exterior roots of the actual own pair. For 0 < h < 1 / 3 0<h<1/3 0 < h < 1/3 the actual leaf-radius and density correlation give
R h = 8 h 4 − 3 h < 1 , k = 1 − R h 1 + R h , E > k T , x L + x R ≥ L T , R_h=\frac{8h}{4-3h}<1,\qquad k=\frac{1-R_h}{1+R_h},\qquad E>kT,\qquad x_L+x_R\ge LT, R h = 4 − 3 h 8 h < 1 , k = 1 + R h 1 − R h , E > k T , x L + x R ≥ L T ,
L = ( 1 − R h ) 2 R h ( 1 + R h ) = ( 4 − 11 h ) 2 8 h ( 4 + 5 h ) . L=\frac{(1-R_h)^2}{R_h(1+R_h)}=\frac{(4-11h)^2}{8h(4+5h)}. L = R h ( 1 + R h ) ( 1 − R h ) 2 = 8 h ( 4 + 5 h ) ( 4 − 11 h ) 2 .
These identities use no numerical own-pair estimate. The exact transport is (C.11); the chronological density ceilings and cap gaps are (C.10) and Section C.1.8. The two actual middle means give Q = w / 2 + K T Q=w/2+KT Q = w /2 + K T , and Lemma B.8 gives w > 3 h w>3h w > 3 h . For h < 1 / 3 h<1/3 h < 1/3 , apply Lemma C.1 with ( a , b , R , q ) = ( ξ , r , R h , q z ) (a,b,R,q)=(\xi,r,R_h,q_z) ( a , b , R , q ) = ( ξ , r , R h , q z ) , q z = v ( z ) − 2 q_z=v(z)^{-2} q z = v ( z ) − 2 . It gives E > k T E>kT E > k T and T ≤ 2 q z R h / ( 1 + R h ) T\le2q_zR_h/(1+R_h) T ≤ 2 q z R h / ( 1 + R h ) . The same actual density identities give
x L + x R ≥ 2 q z k 2 ( 1 + z 2 ) ≥ 2 q z k 2 ≥ ( 1 − R h ) 2 R h ( 1 + R h ) T = L T . x_L+x_R\ge2q_zk^2(1+z^2)\ge2q_zk^2\ge\frac{(1-R_h)^2}{R_h(1+R_h)}T=LT. x L + x R ≥ 2 q z k 2 ( 1 + z 2 ) ≥ 2 q z k 2 ≥ R h ( 1 + R h ) ( 1 − R h ) 2 T = L T .
All these quantities belong to the one selected full lift and its continuous physical profile. In particular, no conclusion of Proposition C.2 is used to establish this interface. The following numerical consequence depends only on the displayed interface. It also applies when that interface has been recovered by a one-sided-leaf construction; it does not require an additional unproved returned source.
Proposition 10.2. Suppose the identities and inequalities of the preceding interface hold on one selected actual lift. Then Q > 283 / 1000 Q > 283/1000 Q > 283/1000 .
Proof. Put e 0 = 47 / 2500 e_0 = 47/2500 e 0 = 47/2500 . At e = e 0 + x / 20 e = e_0 + x/20 e = e 0 + x /20 , the cap expression minus 29 / 2000 29/2000 29/2000 is exactly
− 291 6250000 − 73 x 25000 − 11 x 2 100 < 0. -\frac{291}{6250000}-\frac{73x}{25000}-\frac{11x^2}{100}<0. − 6250000 291 − 25000 73 x − 100 11 x 2 < 0.
It increases strictly with e ≥ 0 e \ge0 e ≥ 0 for the stated density range. Thus e j > e 0 + x j / 20 e_j > e_0 + x_j/20 e j > e 0 + x j /20 , and N > 47 / 1250 + ( x L + x R ) / 20 N > 47/1250 + (x_L + x_R)/20 N > 47/1250 + ( x L + x R ) /20 . For h ≥ 1 / 3 h \ge1/3 h ≥ 1/3 , (10.5) already gives Q > 1 / 2 Q > 1/2 Q > 1/2 . For h < 1 / 3 h < 1/3 h < 1/3 , set
A = K ( 1 − k ) − h 2 L / 2 = h ( 368 + 88 h + 7 h 2 ) 16 ( 4 + 5 h ) , H = A − L / 20. A=K(1-k)-h^2L/2=\frac{h(368+88h+7h^2)}{16(4+5h)},\qquad H=A-L/20. A = K ( 1 − k ) − h 2 L /2 = 16 ( 4 + 5 h ) h ( 368 + 88 h + 7 h 2 ) , H = A − L /20.
The same identities give N < A T N < AT N < A T , so
H T > 47 / 1250. (44) HT>47/1250. \tag*{(44)} H T > 47/1250. (44)
In particular H > 0 H>0 H > 0 is a conclusion, not an assumption used in division. If h ≥ 283 / 1500 h \ge283/1500 h ≥ 283/1500 , the strict width inequality proves the assertion. Otherwise a contrary Q ≤ 283 / 1000 Q \le283/1000 Q ≤ 283/1000 implies
47 1250 K < H ( 283 1000 − 3 h 2 ) . \frac{47}{1250}K < H\left(\frac{283}{1000}-\frac{3h}{2}\right). 1250 47 K < H ( 1000 283 − 2 3 h ) .
The opposite difference, multiplied by 16 h ( 4 + 5 h ) > 0 16h(4+5h)>0 16 h ( 4 + 5 h ) > 0 , is
F = 21 2 h 5 + 131523 1000 h 4 + 1275373 2500 h 3 − 830077 10000 h 2 − 1601 1250 h + 283 625 . F=\frac{21}{2}h^5+\frac{131523}{1000}h^4+\frac{1275373}{2500}h^3-\frac{830077}{10000}h^2-\frac{1601}{1250}h+\frac{283}{625}. F = 2 21 h 5 + 1000 131523 h 4 + 2500 1275373 h 3 − 10000 830077 h 2 − 1250 1601 h + 625 283 .
Its exact decomposition is
F = 510 h ( h − 11 / 100 ) 2 + 131 ( h 2 − 121 / 10000 ) 2 + 3217 100 h 2 − 373 50 h + 541 1250 + 21 2 h 5 + 523 1000 h 4 + 373 2500 h 3 + 77 400 h 2 + 41 5000 h + 82029 100000000 . \begin{aligned}
F={}&510h(h-11/100)^2+131(h^2-121/10000)^2\\
&+\frac{3217}{100}h^2-\frac{373}{50}h+\frac{541}{1250}\\
&+\frac{21}{2}h^5+\frac{523}{1000}h^4+\frac{373}{2500}h^3+\frac{77}{400}h^2+\frac{41}{5000}h+\frac{82029}{100000000}.
\end{aligned} F = 510 h ( h − 11/100 ) 2 + 131 ( h 2 − 121/10000 ) 2 + 100 3217 h 2 − 50 373 h + 1250 541 + 2 21 h 5 + 1000 523 h 4 + 2500 373 h 3 + 400 77 h 2 + 5000 41 h + 100000000 82029 .
The nonsquare quadratic has positive leading coefficient and discriminant − 2569 / 62500 < 0 -2569/62500 < 0 − 2569/62500 < 0 . All other terms are nonnegative, with a positive constant. Hence F > 0 F>0 F > 0 , a contradiction. □
A conditional fee for a central triple
The following estimate keeps state information from the neighboring sources. It will be used both to exclude R15 and to obtain a lower bound for R06. In this section put
C 0 = 80483 125000 < 2 3 . C_0=\frac{80483}{125000}<\frac{2}{3}. C 0 = 125000 80483 < 3 2 .
All costs retain the original physical normalization.
Theorem 11.1 (Central three-arc fee). Let a complete proper six-source be a lawful whole image of 102010 on an original globally minimizing trajectory. Its last five arcs are 02010, with successive own reciprocals p p p , A A A , y y y , B B B , q q q . Select one full maximizing lift on a genuine interval containing the six-source. Suppose qualified sources on the same trajectory give
E a > max ( 1 , b ) , E b > max ( 1 , a ) , E a , E b < 2 , (45) E_a>\max(1,b),\qquad E_b>\max(1,a),\qquad E_a,E_b<2, \tag*{(45)} E a > max ( 1 , b ) , E b > max ( 1 , a ) , E a , E b < 2 , (45) p < A , q < B , p 2 + 2 p A < 1 , q 2 + 2 q B < 1 , p 2 , q 2 < 1 / 6 , p<A,\qquad q<B,\qquad p^2+2pA<1,\qquad q^2+2qB<1,\qquad p^2,q^2<1/6, p < A , q < B , p 2 + 2 p A < 1 , q 2 + 2 q B < 1 , p 2 , q 2 < 1/6 , where
a = 2 y y + A , b = 2 y y + B , E = ( y + A ) ( y + B ) 1 + y 2 . a=\frac{2y}{y+A},\qquad b=\frac{2y}{y+B},\qquad E=\frac{(y+A)(y+B)}{1+y^2}. a = y + A 2 y , b = y + B 2 y , E = 1 + y 2 ( y + A ) ( y + B ) . Then the complete physical cost F = α + A c + β F=\alpha+A_c+\beta F = α + A c + β of the three interior arcs of 02010 satisfies F > C 0 F>C_0 F > C 0 .
Proof. The first five arcs of 102010 102010 102010 , reversed with the whole label map j ↦ 2 − j j \mapsto2-j j ↦ 2 − j , are 12021 12021 12021 . Their capped binary core excludes abnormality. Jointly normalize the objective multiplier, entire matrix and both state covectors on the selected full lift. All central means and matrix identities below belong to this lift. The additional sources in (45) export only physical state inequalities; their multipliers need not be identified with the central lift. The original harmonic bound gives Γ < H ∗ : \Gamma< H_*: Γ < H ∗ : if Γ ≤ 0 \Gamma\le0 Γ ≤ 0 this is immediate; if Γ > 0 \Gamma> 0 Γ > 0 , the actual binary invariant K > 3 / 2 K > 3/2 K > 3/2 gives the future sheet on this same lift before Corollary A.13 is applied.
Assume, for contradiction, that F ≤ C 0 F \le C_0 F ≤ C 0 . We first exclude two exterior ranges without finite computation. Put
f ( s , x ) = 1 − s x s + x , H ( t ) = 1 − 3 t − t log 1 + t 4 t , h ( y ) = 1 + y 2 − y . f(s,x)=\frac{1-sx}{s+x}, \qquad\mathcal{H}(t)=1-3t-t\log\frac{1+t}{4t}, \qquad h(y)=\sqrt{1+y^2}-y. f ( s , x ) = s + x 1 − s x , H ( t ) = 1 − 3 t − t log 4 t 1 + t , h ( y ) = 1 + y 2 − y .
The two crossings imply A , B > g ( y ) : = max { h ( y ) , 1 − y 2 2 y } A,B>g(y):=\max\left\{h(y),\frac{1-y^2}{2y}\right\} A , B > g ( y ) := max { h ( y ) , 2 y 1 − y 2 } . The exact physical leaf and half-center integrals are
L ( s ; l , u ) = s ( u − l ) + s ( 1 + s 2 ) ( 1 s + l − 1 s + u ) − s 2 log s + u s + l , L(s;l,u)=s(u-l)+s(1+s^2)\left(\frac{1}{s+l}-\frac{1}{s+u}\right)-s^2\log\frac{s+u}{s+l}, L ( s ; l , u ) = s ( u − l ) + s ( 1 + s 2 ) ( s + l 1 − s + u 1 ) − s 2 log s + l s + u ,
Z y ( s ) = y [ s − f ( y , s ) ] − y 2 log y + s 1 + y 2 . Z_y(s)=y[s-f(y,s)]-y^2\log\frac{y+s}{\sqrt{1+y^2}}. Z y ( s ) = y [ s − f ( y , s )] − y 2 log 1 + y 2 y + s .
Thus F = L ( A ; p , y ) + Z y ( A ) + Z y ( B ) + L ( B ; q , y ) F=L(A;p,y)+Z_y(A)+Z_y(B)+L(B;q,y) F = L ( A ; p , y ) + Z y ( A ) + Z y ( B ) + L ( B ; q , y ) . For y ≤ 1 / 3 y\le1/\sqrt{3} y ≤ 1/ 3 , the mandatory central interval gives F > H ( y 2 ) F>\mathcal{H}(y^2) F > H ( y 2 ) . For y ≥ 1 / 3 y\ge1/\sqrt{3} y ≥ 1/ 3 , define
D y ( s ) = { Z y ( s ) + L ( s ; s , y ) , s ≤ y , Z y ( s ) , s ≥ y . D_y(s)=
\begin{cases}
Z_y(s)+L(s;s,y), & s\le y,\\
Z_y(s), & s\ge y.
\end{cases} D y ( s ) = { Z y ( s ) + L ( s ; s , y ) , Z y ( s ) , s ≤ y , s ≥ y .
The actual leaves contain these mandatory portions, so F > D y ( A ) + D y ( B ) F>D_y(A)+D_y(B) F > D y ( A ) + D y ( B ) . For s < y s<y s < y ,
D y ′ ( s ) = s ( 2 z − 3 + 2 z 1 + z − 2 log 1 + z 2 ) , z = y / s > 1. D_y'(s)=s\left(2z-3+\frac{2z}{1+z}-2\log\frac{1+z}{2}\right), \qquad z=y/s>1. D y ′ ( s ) = s ( 2 z − 3 + 1 + z 2 z − 2 log 2 1 + z ) , z = y / s > 1.
The bracket vanishes at z = 1 z=1 z = 1 and has derivative 2 − 2 z / ( 1 + z ) 2 > 0 2-2z/(1+z)^2>0 2 − 2 z / ( 1 + z ) 2 > 0 . For s ≥ y s\ge y s ≥ y , Z y ′ ( s ) = y [ 1 + ( 1 − y s ) / ( y + s ) 2 ] > 0 Z_y'(s)=y[1+(1-ys)/(y+s)^2]>0 Z y ′ ( s ) = y [ 1 + ( 1 − y s ) / ( y + s ) 2 ] > 0 . Consequently D y D_y D y is increasing and F > 2 D y ( h ( y ) ) = 2 H ( h ( y ) 2 ) F>2D_y(h(y))=2\mathcal{H}(h(y)^2) F > 2 D y ( h ( y )) = 2 H ( h ( y ) 2 ) . These are comparisons of the complete center and mandatory leaf portions, with their common state preserved.
The function H \mathcal{H} H decreases on ( 0 , 1 / 3 ) (0,1/3) ( 0 , 1/3 ) . The positive exponential partial sums give log ( 53 / 16 ) < 5 / 4 \log(53/16)<5/4 log ( 53/16 ) < 5/4 and log ( 97 / 64 ) < 5 / 12 \log(97/64)<5/12 log ( 97/64 ) < 5/12 . Since h ( 65 / 72 ) = 4 / 9 h(65/72)=4/9 h ( 65/72 ) = 4/9 , the two closed ranges satisfy
y ≤ 2 / 7 ⇒ F > 32 / 49 = C 0 + 56333 6125000 , y\le2/7 \quad\Rightarrow\quad F>32/49=C_0+\frac{56333}{6125000}, y ≤ 2/7 ⇒ F > 32/49 = C 0 + 6125000 56333 ,
y ≥ 65 / 72 ⇒ F > 158 / 243 = C 0 + 192631 30375000 . y\ge65/72 \quad\Rightarrow\quad F>158/243=C_0+\frac{192631}{30375000}. y ≥ 65/72 ⇒ F > 158/243 = C 0 + 30375000 192631 .
It remains to treat 2 / 7 < y < 65 / 72 2/7<y<65/72 2/7 < y < 65/72 .
The exact central matrix and leaf floors
Set
k = 4 y 2 1 + y 2 , D = 4 − k , P 0 = a b , E = k P 0 , k=\frac{4y^2}{1+y^2}, \qquad D=4-k, \qquad P_0=ab, \qquad E=\frac{k}{P_0}, k = 1 + y 2 4 y 2 , D = 4 − k , P 0 = ab , E = P 0 k ,
z = log E , t = 2 z D , r = 1 − a , s = 1 − b . z=\log E, \qquad t=\frac{2z}{D}, \qquad r=1-a, \qquad s=1-b. z = log E , t = D 2 z , r = 1 − a , s = 1 − b .
The crossing guards give
E a = k / b , E b = k / a , k / 2 < a , b < min { k , k } . Ea=k/b, \qquad Eb=k/a, \qquad k/2<a,b<\min\{k,\sqrt{k}\}. E a = k / b , E b = k / a , k /2 < a , b < min { k , k } .
Indeed E a , E b < 2 Ea,Eb<2 E a , E b < 2 give the lower bounds; E a , E b > 1 Ea,Eb>1 E a , E b > 1 give a , b < k a,b<k a , b < k , and E a > b Ea>b E a > b , E b > a Eb>a E b > a give a 2 , b 2 < k a^2,b^2<k a 2 , b 2 < k . These last two inequalities also imply E > 1 E>1 E > 1 .
The actual mean excesses are X = ( μ A − r 2 ) / ( a D ) > 0 X=(\mu_A-r^2)/(aD)>0 X = ( μ A − r 2 ) / ( a D ) > 0 and Y = ( μ B − s 2 ) / ( b D ) > 0 Y=(\mu_B-s^2)/(bD)>0 Y = ( μ B − s 2 ) / ( b D ) > 0 . The original binary matrix and physical cost identities give
K A = 3 + r 2 + a D X 2 , K B = 3 + s 2 + b D Y 2 , (46) K_A=\frac{3+r^2+aDX}{2},\qquad K_B=\frac{3+s^2+bDY}{2}, \tag*{(46)} K A = 2 3 + r 2 + a D X , K B = 2 3 + s 2 + b D Y , (46)
J m = X + Y + 2 z D + 4 ( a + b ) − 4 − k 2 D , Γ ( X , Y ) = 4 K A K B D 2 − J m 2 , J_m=X+Y+\frac{2z}{D}+\frac{4(a+b)-4-k}{2D},\qquad\Gamma(X,Y)=\frac{4K_AK_B}{D^2}-J_m^2, J m = X + Y + D 2 z + 2 D 4 ( a + b ) − 4 − k , Γ ( X , Y ) = D 2 4 K A K B − J m 2 ,
A c = 2 ( a + b ) ( E − 1 ) − k z D , J a = 2 a + 2 k a − k , J b = 2 b + 2 k b − k . A_c=\frac{2(a+b)(E-1)-kz}{D},\qquad J_a=2a+\frac{2k}{a}-k,\qquad J_b=2b+\frac{2k}{b}-k. A c = D 2 ( a + b ) ( E − 1 ) − k z , J a = 2 a + a 2 k − k , J b = 2 b + b 2 k − k .
For the left leaf put Q = 2 − a Q=2-a Q = 2 − a and d = ( A − p ) / ( A + p ) − ( A − y ) / ( A + y ) > 0 d=(A-p)/(A+p)-(A-y)/(A+y)>0 d = ( A − p ) / ( A + p ) − ( A − y ) / ( A + y ) > 0 . The terminal circle and the strict leaf-gap identity imply
α > m a t e r m : = J a ( 2 − a ) ( k − a ) 2 k D . \alpha>m_a^{\mathrm{term}}:=\frac{J_a(2-a)(k-a)}{2kD}. α > m a term := 2 k D J a ( 2 − a ) ( k − a ) .
Indeed p < 1 + A 2 − A p<\sqrt{1+A^2}-A p < 1 + A 2 − A implies d / ( 1 + d / Q ) > Q a ( k − a ) / ( 2 k ) d/(1+d/Q)>Qa(k-a)/(2k) d / ( 1 + d / Q ) > Q a ( k − a ) / ( 2 k ) , while the leaf-gap cost exceeds J a d / [ a D ( 1 + d / Q ) ] J_ad/[aD(1+d/Q)] J a d / [ a D ( 1 + d / Q )] . When k ≥ 4 / 7 k\ge4/7 k ≥ 4/7 , the additional density p < 1 / 6 p<1/\sqrt{6} p < 1/ 6 gives
α > m a d e n s : = J a ( 2 − a ) 2 D ( 1 − D 6 k ) . \alpha>m_a^{\mathrm{dens}}:=\frac{J_a(2-a)}{2D}\left(1-\sqrt{\frac{D}{6k}}\right). α > m a dens := 2 D J a ( 2 − a ) ( 1 − 6 k D ) .
Use m a = m a t e r m m_a=m_a^{\mathrm{term}} m a = m a term for k ≤ 4 / 7 k\le4/7 k ≤ 4/7 and m a = max { m a t e r m , m a d e n s } m_a=\max\{m_a^{\mathrm{term}},m_a^{\mathrm{dens}}\} m a = max { m a term , m a dens } thereafter. The right leaf gives β > m b \beta>m_b β > m b by the identical actual calculation. The two floors for one leaf are combined by maximum, not by addition.
A signed cost-to-mean comparison
For comparison cost c ∈ [ 0 , C 0 ] c\in[0,C_0] c ∈ [ 0 , C 0 ] , define
λ a = a D J a ( 2 − a ) , w = λ a c , \lambda_a=\frac{aD}{J_a(2-a)},\qquad w=\lambda_a c, λ a = J a ( 2 − a ) a D , w = λ a c ,
T a ( c ) = c J a { r 1 − w + ( 2 − a ) w 3 ( 1 − w ) 2 } , T_a(c)=\frac{c}{J_a}\left\{\frac{r}{1-w}+\frac{(2-a)w}{3(1-w)^2}\right\}, T a ( c ) = J a c { 1 − w r + 3 ( 1 − w ) 2 ( 2 − a ) w } ,
T a + ( c ) = max { 0 , T a ( c ) } . T_a^+(c)=\max\{0,T_a(c)\}. T a + ( c ) = max { 0 , T a ( c )} .
The denominator is uniformly positive. For k ≤ 1 k\le1 k ≤ 1 , λ a ≤ 1 \lambda_a\le1 λ a ≤ 1 . On the high domain λ a ≤ 4 / 3 \lambda_a\le4/3 λ a ≤ 4/3 , since
a [ 4 J a ( 2 − a ) − 3 a D ] ≥ a 2 ( a − 2 ) ( 7 a − 10 ) > 0 a[4J_a(2-a)-3aD]\ge a^2(a-2)(7a-10)>0 a [ 4 J a ( 2 − a ) − 3 a D ] ≥ a 2 ( a − 2 ) ( 7 a − 10 ) > 0
for 1 ≤ k < 2 1\le k<2 1 ≤ k < 2 , a ≤ k < 2 < 10 / 7 a\le\sqrt{k}<\sqrt{2}<10/7 a ≤ k < 2 < 10/7 . Hence w ≤ 8 / 9 w\le8/9 w ≤ 8/9 even on the larger cost interval [ 0 , 2 / 3 ] [0,2/3] [ 0 , 2/3 ] .
The actual leaf gap gives d < ( 2 − a ) w / ( 1 − w ) d<(2-a)w/(1-w) d < ( 2 − a ) w / ( 1 − w ) at c = α c=\alpha c = α . For r ≥ 0 r\ge0 r ≥ 0 , decreasing-weight covariance gives μ A − r 2 ≤ r d + d 2 / 3 \mu_A-r^2\le rd+d^2/3 μ A − r 2 ≤ r d + d 2 /3 . For r < 0 r<0 r < 0 , integration by parts gives the same bound with the right side replaced by its positive part: for 0 ≤ v ≤ d 0\le v\le d 0 ≤ v ≤ d ,
v d ( r d + d 2 3 ) − ( r v + v 2 3 ) = v ( d − v ) 3 ≥ 0. \frac{v}{d}\left(rd+\frac{d^2}{3}\right)-\left(rv+\frac{v^2}{3}\right)=\frac{v(d-v)}{3}\ge0. d v ( r d + 3 d 2 ) − ( r v + 3 v 2 ) = 3 v ( d − v ) ≥ 0.
The actual excess is positive, so d > − 3 r d>-3r d > − 3 r ; the quadratic is increasing on the branch used in the substitution. Thus X < T a + ( α ) X<T_a^+(\alpha) X < T a + ( α ) in either sign. This argument does not assume a ≤ 1 a\le1 a ≤ 1 .
The function T a + T_a^+ T a + is increasing and convex on the entire comparison cost interval. For r ≥ 0 r\ge0 r ≥ 0 this follows from its two nonnegative power series. For − 1 < r < 0 -1<r<0 − 1 < r < 0 , its positive branch starts at w 0 = − 3 r / ( 1 − 2 r ) w_0=-3r/(1-2r) w 0 = − 3 r / ( 1 − 2 r ) ; for f r ( w ) = w [ r / ( 1 − w ) + ( 1 + r ) w / ( 3 ( 1 − w ) 2 ) ] f_r(w)=w[r/(1-w)+(1+r)w/(3(1-w)^2)] f r ( w ) = w [ r / ( 1 − w ) + ( 1 + r ) w / ( 3 ( 1 − w ) 2 )] ,
f r ′ ( w ) = 3 r + ( 2 − r ) w 3 ( 1 − w ) 3 , f r ′ ′ ( w ) = 2 [ 1 + 4 r + ( 2 − r ) w ] 3 ( 1 − w ) 4 . f_r'(w)=\frac{3r+(2-r)w}{3(1-w)^3},\qquad f_r''(w)=\frac{2[1+4r+(2-r)w]}{3(1-w)^4}. f r ′ ( w ) = 3 ( 1 − w ) 3 3 r + ( 2 − r ) w , f r ′′ ( w ) = 3 ( 1 − w ) 4 2 [ 1 + 4 r + ( 2 − r ) w ] .
Both numerators are positive at w 0 w_0 w 0 and increase thereafter. Joining this branch to zero retains convexity, including a = 1 a=1 a = 1 .
Eliminating the allocation.
The Hessian of (46) in X , Y X,Y X , Y has eigenvalues − P 0 -P_0 − P 0 and P 0 − 4 P_0-4 P 0 − 4 , both negative. Its coordinate derivatives are nonpositive on the comparison quadrant. To check the latter, the coefficient of X X X in its expanded quadratic is
A X = s − 2 t + ( k − P 0 ) ( 2 − b ) D . A_X=s-2t+\frac{(k-P_0)(2-b)}{D}. A X = s − 2 t + D ( k − P 0 ) ( 2 − b ) .
For k ≤ 1 k\le1 k ≤ 1 , its maximum is at a = b = k a=b=k a = b = k , where D A X ≤ ( 1 − k ) ( 4 + k − k 2 ) + 4 log k ≤ 0 D A_X\le(1-k)(4+k-k^2)+4\log k\le0 D A X ≤ ( 1 − k ) ( 4 + k − k 2 ) + 4 log k ≤ 0 . For 1 ≤ k < 2 1\le k<2 1 ≤ k < 2 , its maximum is at a = k , b = 1 a=\sqrt{k},b=1 a = k , b = 1 , where D A X ≤ k − k − 2 log k ≤ 0 D A_X\le k-\sqrt{k}-2\log k\le0 D A X ≤ k − k − 2 log k ≤ 0 . The first one-variable expression increases to zero at 1, and the second decreases from zero. The assertion for Y Y Y is symmetric.
Therefore Γ ( T a + ( α ) , T b + ( β ) ) \Gamma(T_a^+(\alpha),T_b^+(\beta)) Γ ( T a + ( α ) , T b + ( β )) is decreasing and concave in the two comparison costs. Put U = C 0 − A c U=C_0-A_c U = C 0 − A c . The actual inequalities give α > m a \alpha>m_a α > m a , β > m b \beta>m_b β > m b and α + β ≤ U \alpha+\beta\le U α + β ≤ U . If A c + m a + m b > C 0 A_c+m_a+m_b>C_0 A c + m a + m b > C 0 the budget is impossible. Otherwise increasing total comparison cost to U U U and using the chord of the remaining allocation segment gives
Γ a c t u a l ≥ min { Γ ( T a + ( m a ) , T b + ( U − m a ) ) , Γ ( T a + ( U − m b ) , T b + ( m b ) ) } . (47) \Gamma_{\mathrm{actual}}\ge\min\{\Gamma(T_a^+(m_a),T_b^+(U-m_a)),\ \Gamma(T_a^+(U-m_b),T_b^+(m_b))\}. \tag*{(47)} Γ actual ≥ min { Γ ( T a + ( m a ) , T b + ( U − m a )) , Γ ( T a + ( U − m b ) , T b + ( m b ))} . (47)
Both endpoints are required. If α + β = U \alpha+\beta=U α + β = U , no enlargement is needed; the same chord argument applies. Thus this reduction retains the boundary F = C 0 F=C_0 F = C 0 . Artificial allocation endpoints need not represent trajectories and are not assigned calibrated lifts.
All actual data with 2 / 7 < y < 65 / 72 2/7<y<65/72 2/7 < y < 65/72 lie in the closed domains of Lemma 11.2 below by (45) . That lemma and (47) give Γ a c t u a l ≥ H ∗ \Gamma_{\mathrm{actual}}\ge H_* Γ actual ≥ H ∗ , contradicting Γ a c t u a l < H ∗ \Gamma_{\mathrm{actual}}<H_* Γ actual < H ∗ on the selected full lift. Together with the two exterior ranges this proves F > C 0 F>C_0 F > C 0 .
Lemma 11.2 (Three closed comparison covers). Every point in the following closed domains satisfying A c + m a + m b ≤ C 0 A_c+m_a+m_b\le C_0 A c + m a + m b ≤ C 0 has both lower bounds in (11.3) at least H ∗ H_* H ∗ :
Table 11.2.
Computer-assisted proof. Here A c ≥ 0 A_c\ge0 A c ≥ 0 : E ≥ 1 E\ge1 E ≥ 1 , a + b ≥ k a+b\ge k a + b ≥ k and log E ≤ E − 1 \log E\le E-1 log E ≤ E − 1 give D A c ≥ [ 2 ( a + b ) − k ] ( E − 1 ) ≥ 0 D A_c\ge[2(a+b)-k](E-1)\ge0 D A c ≥ [ 2 ( a + b ) − k ] ( E − 1 ) ≥ 0 . Thus the stated budget condition and nonnegative leaf floors give the comparison cost interval used in (11.3). Write a = k x a=kx a = k x , b = k v b=kv b = k v and start from ( x , v ) ∈ [ 1 / 2 , 1 ] 2 (x,v)\in[1/2,1]^2 ( x , v ) ∈ [ 1/2 , 1 ] 2 . The high-domain curved inequalities are retained. The witnesses have respectively 390193, 13434119 and 9922659 nodes. Their interval evaluators use 48, 46 and 46 binary fractional bits, with 72, 64 and 64 positive logarithm terms. For E ≥ 1 E\ge1 E ≥ 1 , ζ = ( E − 1 ) / ( E + 1 ) \zeta=(E-1)/(E+1) ζ = ( E − 1 ) / ( E + 1 ) , the logarithm tail after N N N terms is bounded by 2 ζ 2 N + 1 ( 2 N + 1 ) ( 1 − ζ 2 ) \frac{2\zeta^{2N+1}}{(2N+1)(1-\zeta^2)} ( 2 N + 1 ) ( 1 − ζ 2 ) 2 ζ 2 N + 1 . Integer square roots enclose radicals by exact squared comparisons. All products and quotients round outward by signed integer floor and ceiling; stored endpoints have bounds 2 60 2^{60} 2 60 in the first cover and 2 61 2^{61} 2 61 in the other two. The corresponding intermediate products fit in signed 128-bit integers. A denominator or range failure cannot accept a leaf.
Each terminal box either violates the curved domain, has A c + m a + m b > C 0 A_c+m_a+m_b>C_0 A c + m a + m b > C 0 throughout, or proves both expressions in (11.3) at least the outward upper enclosure of H ∗ H_* H ∗ . The finite trees use closed children sharing the split face and have no unresolved box. Comparison costs are clipped only to [ 0 , 2 / 3 ] [0,2/3] [ 0 , 2/3 ] , which contains every feasible allocation; clipping does not discard the equality F = C 0 F=C_0 F = C 0 or the faces a = 1 , b = 1 a=1,b=1 a = 1 , b = 1 . Finite induction over these trees proves the lemma.
Costs assigned to positive and negative runs
A mixed increment cycle has alternating maximal positive and negative runs. Write their lengths in cyclic order as
( a 1 , b 1 ) , … , ( a M , b M ) , a i ≥ 1 , b i ≥ 2 , b i ≠ 3. (a_1,b_1),\ldots,(a_M,b_M),\qquad a_i\ge1,\quad b_i\ge2,\quad b_i\ne3. ( a 1 , b 1 ) , … , ( a M , b M ) , a i ≥ 1 , b i ≥ 2 , b i = 3.
The two restrictions on b i b_i b i are exactly the forbidden increment factors +-+ and +---+. The lengths count increments, while the costs below belong to complete physical arcs. No quotient by reversing this cyclic list is taken.
If a positive run starts at increment s s s and has length a a a , assign its own arcs s , s + 1 , … , s + a s,s+1,\ldots,s+a s , s + 1 , … , s + a to that run. A negative run of length b b b , between positive runs ending at t t t and starting at t + b t+b t + b , has the interior arcs t + 1 , … , t + b − 1 t+1,\ldots,t+b-1 t + 1 , … , t + b − 1 . These own sets and interiors partition the physical period. The two arcs facing the positive run at s s s are
s − 1 , s + a + 1 ; s-1,\qquad s+a+1; s − 1 , s + a + 1 ;
we call them its left and right ports . The two ports belonging to opposite ends of a negative separator coincide when b = 2 b=2 b = 2 and are distinct when b ≥ 4 b\ge4 b ≥ 4 .
A local cost profile records a proved lower payment together with its coefficients on the own arcs and its two ports. Supports retain all complete caps, returns and other hypotheses of the source theorem. The port coefficients describe how neighboring profiles can be added; they are not additional costs. Whenever formal ports represent the same physical arc, their coefficients are added, as in (27) . All indices are unfolded before taking physical positions modulo B B B .
Short positive runs without port charges
The new own-pair payment has a useful consequence for this assignment: a singleton positive run can pay for its own arcs while leaving its ports available to its neighbors.
Lemma 12.1 (A singleton between short negative runs). Suppose B > 8 B>8 B > 8 , a i = 1 a_i=1 a i = 1 and b i − 1 = b i = 2 b_{i-1}=b_i=2 b i − 1 = b i = 2 . The cost of the two own arcs of this positive run is strictly greater than 9 / 25 9/25 9/25 . If also a i − 1 , a i + 1 ≥ 2 a_{i-1},a_{i+1}\ge2 a i − 1 , a i + 1 ≥ 2 , it is strictly greater than 3 / 7 3/7 3/7 . Both assertions have zero port loads.
Proof. Let s s s be the starting increment and use a whole cyclic chart so that j s = 0 j_s=0 j s = 0 . The complete source [ s − 3 , s + 4 ] [s-3,s+4] [ s − 3 , s + 4 ] , with whole label map j ↦ j + 2 j\mapsto j+2 j ↦ j + 2 , is 01020212. It has charge { s , s + 1 } \{s,s+1\} { s , s + 1 } and anchor owners s − 1 , s + 2 s-1,s+2 s − 1 , s + 2 . Its length eight is less than B B B , so Lemma 6.1 retains the whole source on a genuine containing interval. Theorem 7.2 gives the 9 / 25 9/25 9/25 bound. No cost on either anchor is charged.
For the stronger conclusion the actual sources and maps are
Table 1 (PDF p. 32).
The ascending witness uses the increment factor +---+; the other comes from the chronological factor +--+ by whole reflection and time reversal. The hypotheses on the two neighboring positive lengths retain the additional positive increments. Each six-source is proper and Theorem 7.4 gives the corresponding physical anchor square less than 1 / 6 1/6 1/6 . Apply these two state facts to the single lift selected for the eight-source, and then apply Theorem 7.2 again.
The six-sources need not have a common containing interval or identical multipliers. Only their physical state conclusions are used. In particular, when M = 2 M=2 M = 2 the two neighboring positive runs are the same run, but the two density owners are distinct. The two six-sources may have a union of length ten even at B = 9 B=9 B = 9 ; this does not identify their covectors or charge their shared support twice. ∎
Lemma 12.2 (A complete nine-source for a crossed core). A complete proper source 010201020 in an original global minimizer satisfies
c 3 + c 4 + c 5 > 1 2 , c_{3}+c_{4}+c_{5}>\frac{1}{2}, c 3 + c 4 + c 5 > 2 1 , where the indices are local to this nine-source.
Proof. Choose one full maximizing lift on a genuine containing interval and normalize it jointly after normality, as in Lemma 6.1 . Its middle five-source [ 2 , 6 ] [2,6] [ 2 , 6 ] is 02010. The two actual seven-sources
[ 2 , 8 ] = 0201020 , ( 6 , 5 , … , 0 ) → j ↦ − j 0201020 [2,8]=0201020,\qquad(6,5,\ldots,0)\xrightarrow{j\mapsto-j}0201020 [ 2 , 8 ] = 0201020 , ( 6 , 5 , … , 0 ) j ↦ − j 0201020
are retained on this same lift, with the whole reflected transport in the second case. For the central own reciprocals ( p , A , y , B 0 , q ) (p,A,y,B_{0},q) ( p , A , y , B 0 , q ) , use the same coordinates as in § 3.32 :
a = 2 y y + A , b = 2 y y + B 0 , E = ( y + A ) ( y + B 0 ) 1 + y 2 . a=\frac{2y}{y+A},\qquad b=\frac{2y}{y+B_{0}},\qquad E=\frac{(y+A)(y+B_{0})}{1+y^{2}}. a = y + A 2 y , b = y + B 0 2 y , E = 1 + y 2 ( y + A ) ( y + B 0 ) .
Here a , b a,b a , b are local comparison coordinates, not run lengths. Lemma B.53 gives
E a > max ( 1 , b ) , E b > max ( 1 , a ) . Ea>\max(1,b),\qquad Eb>\max(1,a). E a > max ( 1 , b ) , E b > max ( 1 , a ) .
Whole reversal exchanges the two analytical sides and does not sort these coordinates. The positive central duration and actual state identities give E > 1 E>1 E > 1 , a , b > 0 a,b>0 a , b > 0 and E a , E b < 2 Ea,Eb<2 E a , E b < 2 . The two positive caps, actual leaf means and all maximizing equations are those of the selected full lift. Thus every hypothesis of Definition E.3 holds, and Theorem E.17 gives the assertion. This implication uses the complete nine-source, not an unqualified five-letter occurrence of 02010. □
Corollary 12.3 (A positive run of length two). If B > 9 B>9 B > 9 , a i = 2 a_i=2 a i = 2 and b i − 1 = b i = 2 b_{i-1}=b_i=2 b i − 1 = b i = 2 , its three own arcs have cost greater than 1 / 2 1/2 1/2 , with zero port loads.
Proof. For a run beginning at s s s , set j s = 0 j_s=0 j s = 0 . The complete source [ s − 3 , s + 5 ] [s-3,s+5] [ s − 3 , s + 5 ] , under j ↦ j + 2 j\mapsto j+2 j ↦ j + 2 , is 010201020. Its local positions 3, 4, 5 are exactly s s s , s + 1 s+1 s + 1 , s + 2 s+2 s + 2 . Apply Lemma 12.2 . □
Terminal payments and available port capacity
Put t 0 = 379 / 1000 t_0=379/1000 t 0 = 379/1000 , j 0 = 63 / 500 j_0=63/500 j 0 = 63/500 and P = 180483 / 500000 P=180483/500000 P = 180483/500000 . For a positive run of length a ≥ 3 a\geq3 a ≥ 3 , with both adjacent negative runs of length two, retain the two returned sources
As above j s = 0 j_s=0 j s = 0 . Each forward prefix supplies its actual nonpositive entrance and the last arc is its actual return. Theorem 8.7 gives more than t 0 t_0 t 0 for each charged triple, without an additional entrance-circle or LOW-mean premise.
The table below records usable positive-run profiles. Every own arc has coefficient at most one. The pair in the last column gives the left and right formal port coefficients; all supports are proper under the displayed restrictions.
Table 12.2.
The first three rows were just proved. For a = 3 a=3 a = 3 the two terminal triples have disjoint own charges. At a = 4 a=4 a = 4 , the complete source [ s − 2 , s + 6 ] + 2 [s-2,s+6]+2 [ s − 2 , s + 6 ] + 2 is 102012021 and charges its middle position, namely s + 2 s+2 s + 2 , above j 0 j_0 j 0 by Theorem C.6. Its internal forward contexts [ 2 , 5 ] = 2012 [2,5]=2012 [ 2 , 5 ] = 2012 and [ 3 , 6 ] = 0120 [3,6]=0120 [ 3 , 6 ] = 0120 supply both actual outward negative entrances, as required there. For a ≥ 5 a\geq5 a ≥ 5 , insert the disjoint internal pairs
{ s + 2 + 2 r , s + 3 + 2 r } , 0 ≤ r < ⌊ a − 3 2 ⌋ . \{s+2+2r,s+3+2r\},\qquad0\leq r<\left\lfloor\frac{a-3}{2}\right\rfloor. { s + 2 + 2 r , s + 3 + 2 r } , 0 ≤ r < ⌊ 2 a − 3 ⌋ .
Their complete six-sources [ s + 2 r , s + 5 + 2 r ] [s+2r,s+5+2r] [ s + 2 r , s + 5 + 2 r ] lie in the positive run and phase to 012012; Proposition B.37 pays more than P P P for each. These pairs avoid both terminal charges. All supports used in these constructions have at most nine arcs.
There is also a half-port choice: the positive and negative profiles of Lemmas C.13 and C.14 have load at most 1 / 2 1/2 1/2 on each incident formal port. Their complete source proof applies for B ≥ 10 B\geq10 B ≥ 10 . The choices ( 0 , 0 ) (0,0) ( 0 , 0 ) , the half-port choices, and ( 1 , 1 ) (1,1) ( 1 , 1 ) are alternatives; their lower values cannot all be added on the same own arcs. Their benefit is that a zero-port choice on one run can leave full capacity for a terminal payment next to it.
For a separator of length two the two facing ports are a single physical arc. If a negative profile has coefficient n n n there and the two incident positive profiles have coefficients r , l r,l r , l , their compatibility condition is n + r + l ≤ 1 n+r+l\leq1 n + r + l ≤ 1 . At a longer separator the two ports are different, so each has its own capacity condition. This applies also to repeated incidences when M = 1 M=1 M = 1 or 2. For example, if the two formal ports of a ( 1 , 1 ) (1,1) ( 1 , 1 ) profile coincide, its physical row has coefficient two there and cannot be selected with weight one in a capacity-one assembly. Theorem 6.2 adds precisely the simultaneously compatible profiles.
Two uniform three-run exclusions
The following applications show how short-run payments release the ports needed by a neighboring longer run. They exclude whole families without first assigning individual names to their words.
Proposition 12.4 (Positive lengths 1, 2, a a a ). Suppose a mixed increment cycle has exactly three negative runs, all of length two, and positive lengths (1, 2, a a a ) or (1, a a a , 2) in cyclic order, where a ≥ 4 a\geq4 a ≥ 4 . It cannot be the switching type of an original global minimizer.
Proof. Here B = a + 9 ≥ 13 B=a+9\geq13 B = a + 9 ≥ 13 , so every source in the profiles above is proper. The length-one run has neighboring positive lengths 2, a ≥ 2 a\geq2 a ≥ 2 , and Lemma 12.1 pays more than 3 / 7 3/7 3/7 on its own two arcs. Corollary 12.3 pays more than 1 / 2 1/2 1/2 on the own three arcs of the length-two run. Neither uses a port. At a = 4 a=4 a = 4 the remaining positive run uses its two terminal payments and middle-center payment. At a ≥ 5 a\geq5 a ≥ 5 it uses the two terminal payments and at least one complete forward pair. Since P > j 0 P>j_0 P > j 0 , its payment in either case is greater than 2 t 0 + j 0 2t_0+j_0 2 t 0 + j 0 . The three own sets are disjoint, and only the long run charges its two incident ports. The third separator need not be charged. Consequently
C > 3 7 + 1 2 + 2 379 1000 + 63 500 = 1586 875 = C ∗ + 5419 700000 > C ∗ . C>\frac{3}{7}+\frac{1}{2}+2\frac{379}{1000}+\frac{63}{500}=\frac{1586}{875}=C_*+\frac{5419}{700000}>C_*. C > 7 3 + 2 1 + 2 1000 379 + 500 63 = 875 1586 = C ∗ + 700000 5419 > C ∗ .
The twist requires a − 3 ≡ 1 ( m o d 3 ) a-3\equiv1\pmod{3} a − 3 ≡ 1 ( mod 3 ) , hence a ≡ 1 ( m o d 3 ) a\equiv1\pmod{3} a ≡ 1 ( mod 3 ) ; no larger allowed value changes the source argument or weakens its bound. Both cyclic orders are covered by their whole lawful source placements. At a = 4 a=4 a = 4 the two orders give the words − − + − + + − + + - - + - + + - + + − − + − + + − + + and − − + − + + + − + - - + - + + + - + − − + − + + + − + . □ \square □
Corollary 12.5 (Positive lengths 1, 1, a a a ). A cycle with exactly three negative runs of length two and positive lengths (1, 1, a a a ), where a ≥ 5 a\geq5 a ≥ 5 , cannot occur in an original global minimizer.
Proof. Its two singleton positive runs are different physical runs. Each pays more than 9 / 25 9/25 9/25 on its own arcs and leaves both ports unused. The remaining run pays more than 2 t 0 + P 2t_0+P 2 t 0 + P . The charges are compatible, so
C > 2 9 25 + 2 379 1000 + P = 919483 500000 > C ∗ . C>2\frac{9}{25}+2\frac{379}{1000}+P=\frac{919483}{500000}>C_*. C > 2 25 9 + 2 1000 379 + P = 500000 919483 > C ∗ .
The twist here requires a ≡ 2 ( m o d 3 ) a \equiv2 \pmod{3} a ≡ 2 ( mod 3 ) . The first case a = 5 a=5 a = 5 gives
− − + − − + − − + + + + + . - - + - - + - - + + + + +. − − + − − + − − + + + + + .
Distinctness of the two singleton runs is essential; if M = 2 M=2 M = 2 one physical run cannot be charged twice. □
Complete-source costs for the run arguments
The following costs restrict the number and lengths of the runs. They are statements about original global minimizers. A supporting source contains all positive caps, returns and sign-producing arcs used in its proof. Its charge is the complete physical cost of the displayed local positions. Positions start at zero, and r : s r:s r : s denotes { r , r + 1 , … , s } \{r,r+1,\ldots,s\} { r , r + 1 , … , s } . Every source must be proper. A lawful change of chart transports the whole source, its full lift and its charge. Put P = C ∗ / 5 = 180483 / 500000 P=C_*/5=180483/500000 P = C ∗ /5 = 180483/500000 .
Table 2. Complete-source costs for the all-length classification. Two payments are weak, as indicated.
The first group pays for positive own arcs, possibly with their ports. The second pays terminal triples or joins the two sides of a short negative separator. The last supplies negative interiors, central arcs and internal forward pairs. The table gives sufficient values; several of the cited proofs give stronger constants.
Lemma 13.1 (Complete-source qualifications). Each row of Table 2 has its stated physical payment on its complete proper source. In the 3/7 row, each of local anchor positions 2, 5 must also be the owner at position 3 of a complete proper 201020 source in its own whole lawful chart.
Proof. Lemma 6.1 gives a genuine containing fundamental interval, retaining the original state, signed frame endpoints and real winding. Select a nonzero full maximizing lift there, prove normality, and then normalize its objective multiplier, matrix adjoint and both state covectors jointly. A cost theorem retains this one lift wherever it couples means or switching responses. Independently proved scalar costs may be added without identifying their selected covectors.
The capped-core, one-return and two-return estimates retain all exterior arcs. The 74/125 row uses both returns of 01020212 and charges four arcs; the 9/25 row charges only its own pair. For 3/7, the two proper six-sources give physical anchor-density bounds x 2 , x 5 < 1 / 6 x_2, x_5 < 1/6 x 2 , x 5 < 1/6 by Theorem 7.4. These state facts apply to the selected own-pair lift even if the six-sources used different containing intervals. Neither guard is charged. Lemma 12.2 retains both crossing contexts on one nine-source lift before giving 1/2; a bare 02010 core does not qualify that row. The four-own-arc theorem gives 103/250 > 41/100, and the six-own-arc theorem retains its entire eighth supporting arc.
The ordinary source 012021 has the nonpositive entrance proved in Corollary B.76. Its full return 0120212 gives 379/1000 without an additional entrance-circle or mean hypothesis. For 10201020, apply Lemma B.103 to positions 1 : 7. It gives E a > 1 E_a > 1 E a > 1 in the actual unsorted chart. The reversed positions (4, 3, 2, 1, 0), under the whole map j ↦ 2 − j j \mapsto2-j j ↦ 2 − j , form 12021 with entrance 1 − E a < 0 1-E_a < 0 1 − E a < 0 and physical charge {1, 2, 3}. The nonpositive-tail theorem therefore gives the displayed payment.
The nine-source 012021201 keeps both returned contexts, internal forward entrance witnesses, outgoing leaves and ties of Lemma D.3. Its payment remains weak. The ten-source 2021201020 keeps the common switching jet and both actual returns of Proposition C.12; the two entrance and terminal circles are retained on its selected full lift. The returns are part of the source and are not charged again.
The seven-source 2021020 contains the rooted initial 202, both actual contact jets and the full terminal 020. The nine-source 202102101 retains the strict rooted response, regenerated prefix and complete terminal used in Lemma C.5. That lemma proves separate singleton inequalities, not two simultaneous consequences of a maximum. The reverse-pair source 20210212 has the complete caps of Theorem E.2.
For 102012021, its contexts 2012 at 2 : 5 and 0120 at 3 : 6 supply both outward negative entrances by Lemma B.24. The two capped triples, crossing and endpoint jets remain on one lift, so the opposite-center theorem applies without a HIGH predicate. The complete ten-source 012021202 retains the returned four-core and internal H7 suffix of Lemma C.8; they establish the cap endpoint qualifications without a later exterior cap. The forward-pair and weak-five rows keep their uncharged complete caps. □
We next combine these costs for arbitrary run lengths. The classification will then follow from length bounds, the twist congruence and explicit source placements. Finite calculations in some cost proofs concern those local inequalities rather than an enumeration of increment words.
Part II. Finite bounds and classification by increment runs
From a uniform bound to twenty essential arcs
The uniform bound of Theorem 3.1 makes the problem finite, but is too large for a useful word list. We now replace its common thirteen-arc payment by local payments adapted to the increment runs. This first sharper stage gives B ≤ 20 B \le20 B ≤ 20 , hence N ≤ 21 N \le21 N ≤ 21 at a prescribed cut.
For a mixed cycle write its positive and negative run lengths as ( a i , b i ) (a_i,b_i) ( a i , b i ) , 1 ≤ i ≤ M 1 \le i \le M 1 ≤ i ≤ M . Then a i ≥ 1 a_i \ge1 a i ≥ 1 , b i ≥ 2 b_i \ge2 b i ≥ 2 and b i ≠ 3 b_i \ne3 b i = 3 . Let d i d_i d i count the adjacent negative runs of length two. The complete-source profiles of Lemmas C.13 and C.14 give the following own payments with at most half capacity on either facing port:
Table 1 (PDF p. 37).
P = 180483 500000 P = \frac{180483}{500000} P = 500000 180483 .
The positive-run payment is strictly greater than u ( a i ) + v ( a i ) d i u(a_i) + v(a_i)d_i u ( a i ) + v ( a i ) d i . The negative payment is V ( 2 ) = 0 V(2)=0 V ( 2 ) = 0 , V ( 4 ) = 299 / 250 V(4)=299/250 V ( 4 ) = 299/250 , V ( 6 ) = 13 / 10 V(6)=13/10 V ( 6 ) = 13/10 , and for b ≥ 5 b \ge5 b ≥ 5 , b ≠ 6 b \ne6 b = 6 ,
V ( b ) = { b / 5 − 1 / 2 , b ≡ 1 ( m o d 3 ) , b / 5 − 2 / 5 , b ≡ 2 ( m o d 3 ) , b / 5 − 3 / 5 , b ≡ 0 ( m o d 3 ) . V(b) =
\begin{cases}
b/5 - 1/2, & b \equiv1 \pmod{3},\\
b/5 - 2/5, & b \equiv2 \pmod{3},\\
b/5 - 3/5, & b \equiv0 \pmod{3}.
\end{cases} V ( b ) = ⎩ ⎨ ⎧ b /5 − 1/2 , b /5 − 2/5 , b /5 − 3/5 , b ≡ 1 ( mod 3 ) , b ≡ 2 ( mod 3 ) , b ≡ 0 ( mod 3 ) .
The full-source proofs in Appendix C justify these profiles for every B ≥ 10 B \ge10 B ≥ 10 , independently of an upper bound on B B B . At a short separator the two incident positive half-port loads add to at most one; elsewhere they meet a negative half-port load. Thus the capacity theorem applies simultaneously to all runs. In the next margin comparison p = P p=P p = P and C C C denotes total physical cost.
Lemma 14.1 (The only zero-margin incidence). Put c = 167 / 2000 c=167/2000 c = 167/2000 and
F ( a , b ) = u ( a ) + V ( b ) + 2 v ( a ) [ b = 2 ] , Δ ( a , b ) = F ( a , b ) − c ( a + b ) . F(a,b)=u(a)+V(b)+2v(a)[b=2], \qquad\Delta(a,b)=F(a,b)-c(a+b). F ( a , b ) = u ( a ) + V ( b ) + 2 v ( a ) [ b = 2 ] , Δ ( a , b ) = F ( a , b ) − c ( a + b ) . For every allowed pair a , b a,b a , b , Δ ( a , b ) ≥ 0 \Delta(a,b) \ge0 Δ ( a , b ) ≥ 0 , with equality only at ( 4 , 2 ) (4,2) ( 4 , 2 ) . More precisely,
Δ ( a , 2 ) ≥ 7 250 ( a ≠ 4 ) , Δ ( a , b ) ≥ 261 2000 ( b ≥ 4 ) . \Delta(a,2) \ge\frac{7}{250} \quad(a \ne4), \qquad\Delta(a,b) \ge\frac{261}{2000} \quad(b \ge4). Δ ( a , 2 ) ≥ 250 7 ( a = 4 ) , Δ ( a , b ) ≥ 2000 261 ( b ≥ 4 ) . Also Δ ( a , 4 ) ≥ 81 / 100 \Delta(a,4) \ge81/100 Δ ( a , 4 ) ≥ 81/100 .
Proof. For a ≥ 5 a \ge5 a ≥ 5 , increasing a a a by two increases every margin by p − 2 c = 96983 / 500000 > 0 p-2c=96983/500000>0 p − 2 c = 96983/500000 > 0 . In each stable negative residue class, increasing b b b by three increases it by 3 / 5 − 3 c = 699 / 2000 > 0 3/5-3c=699/2000>0 3/5 − 3 c = 699/2000 > 0 . The exceptional rows b = 4 , 6 b=4,6 b = 4 , 6 are checked separately and are not recurrence bases. It therefore suffices to check a = 1 , … , 6 a=1,\ldots,6 a = 1 , … , 6 and b = 2 , 4 , 5 , 6 , 7 , 9 b=2,4,5,6,7,9 b = 2 , 4 , 5 , 6 , 7 , 9 . The minima below are exact:
Table 2 (PDF p. 37).
The two positive recurrences prove the claims for all lengths, with no finite-cutoff inference.
Proposition 14.2 (All large mixed periods). No minimizing mixed period has B ≥ 21 B \ge21 B ≥ 21 .
Proof. Combine all positive and negative profiles. For a short separator the two incident positive half-port loads sum to at most one; for every other separator each positive half-port load meets at most one negative half load. All remaining charged arcs belong to the disjoint own sets or separator interiors. Consequently Theorem 6.2 gives
C > ∑ i { u ( a i ) + v ( a i ) ( [ b i − 1 = 2 ] + [ b i = 2 ] ) + V ( b i ) } = 1 2 ∑ i { F ( a i , b i ) + F ( a i + 1 , b i ) } = c B + 1 2 ∑ i { Δ ( a i , b i ) + Δ ( a i + 1 , b i ) } . \begin{aligned}
C > \sum_i \{u(a_i) + v(a_i)([b_{i-1}=2] + [b_i=2]) + V(b_i)\} \\
&= \frac{1}{2}\sum_i \{F(a_i,b_i) + F(a_{i+1},b_i)\} \\
&= cB + \frac{1}{2}\sum_i \{\Delta(a_i,b_i) + \Delta(a_{i+1},b_i)\}.
\end{aligned} C > i ∑ { u ( a i ) + v ( a i ) ([ b i − 1 = 2 ] + [ b i = 2 ]) + V ( b i )} = 2 1 i ∑ { F ( a i , b i ) + F ( a i + 1 , b i )} = c B + 2 1 i ∑ { Δ ( a i , b i ) + Δ ( a i + 1 , b i )} .
Strictness comes from the positive profiles even when some negative rows are weak. For B ≥ 22 B \ge22 B ≥ 22 the right side is at least 22 c = 1837 / 1000 > 180483 100000 22c = 1837/1000 > \frac{180483}{100000} 22 c = 1837/1000 > 100000 180483 .
Suppose B = 21 B=21 B = 21 . A separator of length at least four contributes its two incident margins, giving C > 21 c + 261 / 2000 = 471 / 250 > 180483 100000 C > 21c + 261/2000 = 471/250 > \frac{180483}{100000} C > 21 c + 261/2000 = 471/250 > 100000 180483 . Otherwise all separators have length two. If at least two positive runs differ from four, their margins give C > 21 c + 7 / 125 = 3619 / 2000 > 180483 100000 C > 21c + 7/125 = 3619/2000 > \frac{180483}{100000} C > 21 c + 7/125 = 3619/2000 > 100000 180483 . If all positive runs equal four, B = 6 M B=6M B = 6 M cannot equal 21. With exactly one exceptional positive length a a a , the length identity is a = 25 − 6 M > 0 a=25-6M>0 a = 25 − 6 M > 0 , so 1 ≤ M ≤ 4 1 \le M \le4 1 ≤ M ≤ 4 . The actual twist requires
∑ i ( a i − b i ) = B − 4 M ≡ 1 ( m o d 3 ) . \sum_i(a_i-b_i)=B-4M\equiv1\pmod{3}. i ∑ ( a i − b i ) = B − 4 M ≡ 1 ( mod 3 ) .
At B = 21 B=21 B = 21 this forces M ≡ 2 ( m o d 3 ) M\equiv2\pmod{3} M ≡ 2 ( mod 3 ) , hence M = 2 M=2 M = 2 and a = 13 a=13 a = 13 . The length-thirteen run contains the five disjoint forward pairs constructed above, with complete six-sources retained. Their strict sum is 5 p = 180483 100000 5p=\frac{180483}{100000} 5 p = 100000 180483 , again a contradiction. □
Corollary 14.3 (A uniform length-four separator exclusion). A minimizing mixed period with B ≥ 16 B \ge16 B ≥ 16 and a maximal negative separator of length four cannot satisfy the competitor budget.
Proof. All profiles remain proper for these B B B . The two incident margins at that separator give C > c B + 81 / 100 ≥ 1073 / 500 > 180483 100000 C > cB + 81/100 \ge1073/500 > \frac{180483}{100000} C > c B + 81/100 ≥ 1073/500 > 100000 180483 . □
Proposition 14.4 (Pure periods). No minimizing pure positive or pure negative period with B ≥ 10 B \ge10 B ≥ 10 satisfies the competitor budget.
Proof. In a pure positive period use the complete six-source at every physical start with coefficient 1 / 2 1/2 1/2 . Each phases to 012012 and charges its two central arcs; every physical arc therefore has load one. Theorem C.10 gives C > B p / 2 = ( B / 10 ) 180483 100000 ≥ 180483 100000 C > Bp/2 = (B/10)\frac{180483}{100000} \ge\frac{180483}{100000} C > B p /2 = ( B /10 ) 100000 180483 ≥ 100000 180483 . In a pure negative period use each proper complete five-source with coefficient 1 / 3 1/3 1/3 . Its three central arcs give every physical arc load one, so Theorem B.47 gives C ≥ B / 5 ≥ 2 > 180483 100000 C \ge B/5 \ge2 > \frac{180483}{100000} C ≥ B /5 ≥ 2 > 100000 180483 . All sources are proper, with complete caps and full containing lifts. A window crossing the twisted seam merely changes its lawful cyclic phase. Only physical scalar costs are assembled. □
The coarse bound of twenty essential switches
Theorem 15.1 (Coarse finite window). Every normalized global minimizer has B ≤ 20 B \le20 B ≤ 20 .
Proof. The qualitative finiteness theorem gives a finite essential-switch count B B B for the original minimizer. Proposition 14.2 excludes every mixed B ≥ 21 B \ge21 B ≥ 21 , and Proposition 14.4 excludes every pure period in that range. Thus B ≤ 20 B \le20 B ≤ 20 . The original cut relation N ≤ B + 1 N \le B+1 N ≤ B + 1 then gives N ≤ 21 N \le21 N ≤ 21 without changing the prescribed cut; the extra arc is exactly a possible split of one physical arc. No bound on B B B is inferred from a bound on N N N . □
Complete sources for the finite reduction
The fixed payments below are statements about original global minimizers. A source includes every positive cap, return and sign-producing arc in its displayed word. Its charge is the complete physical cost of the specified local positions. All positions start at zero; r : s r:s r : s denotes the integer interval { r , r + 1 , … , s } \{r,r+1,\ldots,s\} { r , r + 1 , … , s } . A lawful image transports the entire source, its selected full lift and its charge.
Put
P = 180483 500000 , E ∗ = 4154032208439 8759490110000 . P=\frac{180483}{500000},\qquad E_*=\frac{4154032208439}{8759490110000}. P = 500000 180483 , E ∗ = 8759490110000 4154032208439 .
Every source in Table 3 is required to be proper. Row 16 additionally requires both qualified six-source density witnesses specified below. Weak inequalities in rows 3 and 17 remain weak.
Table 3. Complete-source physical payments used by the single reduction. All positions are local and start at zero.
Lemma 16.1 (Source qualifications). *Each row of Table 3 has its stated physical payment on the indicated complete source and legal placement. In row 16, each of local anchor
positions 2, 5 must also be the owner at position 3 of a complete proper source 201020, in its own whole lawful chart.
Proof. For a proper source, Lemma 6.1 gives a genuine containing fundamental interval, with the original state and signed frame endpoints and real winding. Select a nonzero full maximizing lift there; prove normality before jointly normalizing the objective multiplier, whole matrix adjoint and both state covectors. Each row uses that complete lift wherever its proof couples means or switching responses. Separately proved scalar rows need not have the same selected lift.
Rows 1–10 are the capped-core, returned-core, own-arm and forward conclusions cited in the table, with their uncharged exterior arcs retained. In particular, the 74/125 four-arc estimate uses both actual returns of 01020212; it is distinct from its own-pair payment. The two 103/250 rows have the identical source and charge hypotheses of Theorem B.122. The six-own-arc bound retains the complete eighth supporting arc.
For rows 11 and 12, Corollary B.76 proves the nonpositive entrance required by Theorem B.127. For row 13, apply Lemma B.103 to positions 1, …, 7 of 10201020. It gives E a > 1 E_a > 1 E a > 1 in the actual unsorted chart. Positions (4, 3, 2, 1, 0) under the whole map j ↦ 2 − j j \mapsto2-j j ↦ 2 − j are 12021, with entrance 1 − E a < 0 1-E_a < 0 1 − E a < 0 and physical triple { 1 , 2 , 3 } \{1, 2, 3\} { 1 , 2 , 3 } . Thus the nonpositive-tail theorem applies. Row 14 retains the full forward entrance witness and final return of 0120212; Lemma B.77, on the full prefix, supplies its actual negative entrance and entrance circle, and Proposition 10.1 uses that circle and the terminal circle on the selected lift. The stronger tail payment is not inferred from the bare three-arc word. The complete nine-source 202120212 likewise contains its seven-arc sign and circle witness at positions 0, …, 6 and its full returned context at 3, …, 8; thus its positions 4, 5, 6 also admit the two-circle payment. Its conservative 379/1000 consequence is the attachment used in the coarse bound.
For row 15, Theorem 7.2 uses the complete proper eight-source and its actual anchors. For row 16, the two proper six-sources give physical anchor-density bounds x 2 , x 5 < 1 / 6 x_2, x_5 < 1/6 x 2 , x 5 < 1/6 by Theorem 7.4. These are state facts of the original trajectory. They may be exported from their own containing intervals without identifying their covectors with the selected own-pair lift. Applying the stronger clause of Theorem 7.2 on that lift gives 3/7. The guards themselves carry no charge.
The proper nine-source in row 17 has precisely the two returned contexts, internal forward entrance witnesses, outgoing leaves and ties of Lemma D.3. Its bound is ≥ 2 P \ge2P ≥ 2 P , not a strict inequality. Its proof retains the first-contact reduction and local transfer; no Casimir bound coming from exterior payments of a special period is imported. Row 18 is Proposition B.104: its full eleven-source supplies both actual nonpositive entrances, the returned tail and the complete bridge, on one selected lift.
Rows 19–22 retain the entire support of their union theorems, including the shared switching derivative or actual means used there. The two exterior returns in row 20 supply the two terminal circles required for the six-arc 807/1000 payment. Neither return is charged again. Row 21 retains both complete one-return cores, their shared physical arc and the same selected matrix adjoint. Row 22 retains the two original capped contexts on its complete nine-source.
The seven-source of row 23 contains the rooted initial 202, both actual 2 → 1 → 0 2 \to1 \to0 2 → 1 → 0 contact jets and the full terminal 020. It therefore qualifies the direct center bound. The nine-source of row 24 retains the strict rooted response, regenerated prefix and complete terminal used in Lemma C.5. That lemma proves two separate singleton inequalities; the table uses its position-3 consequence at 13/20. It is not a bound on a maximum asserted simultaneously for both arcs.
For row 25, 102012021 contains the two outward forcing contexts 2012 at positions 2, …, 5 and 0120 at positions 3, …, 6. Lemma B.24 , in the respective whole charts, supplies both negative outward entrances. The two capped triples, central crossing and actual endpoint jets lie on the same selected lift, so Theorem B.126 applies without a HIGH predicate. The complete reverse-pair source and positive caps in row 26 are exactly Theorem E.2. Row 27 is the complete proper ten-source of Lemma C.8; its returned four-core and internal H7 suffix establish the two cap endpoint qualifications without a later exterior cap. Row 28 is Lemma 12.2, which supplies both crossings on one complete nine-source lift before applying the half-unit estimate.
All rows in this reduction use strictly fewer than B B B source arcs. The rule therefore needs no full-period source extension or identification of state covectors across a seam. The broader scope of a source theorem is not used to enlarge this finite menu. □
The table omits several consequences that would give redundant counting interfaces. Two disjoint proper five-sources yield the weak six-arc 6/5 payment on 10210210. The nine-source at positions 1, …, 9 of 01201020120, after phase +2, gives the same five-arc union as row 17. Two proper own pairs in 01020212101 give 18/25 without charging their intervening arc. Finally, the five-source prefix of 102102 already gives 3/5 on its first three charged arcs. Using these shorter payments preserves the lower bound and can only decrease the physical load. Thus neither a longer supporting word nor an unused positive arc needs a separate table row.
A single finite reduction to three mixed families
The payments of Table 3 define one fixed classification rule. We first enumerate a finite necessary domain, then apply that same rule to each word. The words left by this rule fall into the three mixed run families below, together with the all-minus word. No abstract increment word is asserted to be realizable by a minimizing disk.
The complete necessary domain
Take an essential switch as origin and let j i ∈ Z / 3 Z j_i \in\mathbb{Z}/3\mathbb{Z} j i ∈ Z /3 Z be the successive controls. Their increments ε i ∈ { − 1 , 1 } \varepsilon_i \in\{-1,1\} ε i ∈ { − 1 , 1 } satisfy
J ( 0 ) = 0 , J ( r ) = ∑ i = 0 r − 1 ε i ( 1 ≤ r ≤ B ) , J ( q B + r ) = q J ( B ) + J ( r ) ( 0 ≤ r < B ) . (48) J(0)=0,\qquad J(r)=\sum_{i=0}^{r-1}\varepsilon_i\quad(1\le r\le B),\qquad J(qB+r)=qJ(B)+J(r)\quad(0\le r<B). \tag*{(48)} J ( 0 ) = 0 , J ( r ) = i = 0 ∑ r − 1 ε i ( 1 ≤ r ≤ B ) , J ( q B + r ) = q J ( B ) + J ( r ) ( 0 ≤ r < B ) . (48)
The last formula holds for every integer q q q . The controls and the necessary real-twist congruence are
j i = J ( i ) ( m o d 3 ) , J ( B ) ≡ 1 ( m o d 3 ) , j i + B = j i + 1 ( m o d 3 ) . (49) j_i=J(i)\pmod3,\qquad J(B)\equiv1\pmod3,\qquad j_{i+B}=j_i+1\pmod3. \tag*{(49)} j i = J ( i ) ( mod 3 ) , J ( B ) ≡ 1 ( mod 3 ) , j i + B = j i + 1 ( mod 3 ) . (49)
Only the increments are periodic. In particular, a source crossing the seam is reconstructed from the integer walk, not by periodically repeating the control labels.
Changing the first switch cyclically rotates the increment word. We take its least rotation in the order − < + -<+ − < + . We make no further quotient under reversal or sign complementation. A local source with L L L controls has L − 1 L-1 L − 1 internal increments, whereas the cycle has B B B increments including its seam.
The actual switching exclusions of Proposition 2.4 forbid the cyclic increment patterns + − + +-+ + − + and + − − − + +---+ + − − − + . The all-length coarse bound, Theorem 15.1, is proved independently of the finite enumeration and gives B ≤ 20 B\le20 B ≤ 20 . Thus every minimizer with B ≥ 8 B\ge8 B ≥ 8 belongs to
W = ⋃ B = 8 20 { least rotations of ε ∈ { − 1 , 1 } B : ∑ i ε i ≡ 1 ( m o d 3 ) , + − + and + − − − + are absent cyclically } . (50) \mathcal{W}=\bigcup_{B=8}^{20}\left\{\text{least rotations of }\varepsilon\in\{-1,1\}^{B}:\ \sum_i\varepsilon_i\equiv1\pmod3,\quad+-+\text{ and }+---+\text{ are absent cyclically}\right\}. \tag*{(50)} W = B = 8 ⋃ 20 { least rotations of ε ∈ { − 1 , 1 } B : i ∑ ε i ≡ 1 ( mod 3 ) , + − + and + − − − + are absent cyclically } . (50)
This is a necessary domain only. The long-type length bound to be obtained below is not used to construct it.
For completeness, enumerate all 2 B 2^B 2 B binary strings for each 8 ≤ B ≤ 20 8\le B\le20 8 ≤ B ≤ 20 , imposing the congruence and the two forbidden patterns, then retain each least rotation once. Appending the first four symbols tests every seam occurrence, because each forbidden pattern has length at most five. The process visits every binary string, and taking the least rotation identifies exactly the stated cyclic equivalence. This proves exhaustion without an optimization routine.
The single payment rule
For each w ∈ W w \in\mathcal{W} w ∈ W reconstruct its controls by (48) . For every row of Table 3 , test every chronological start and both whole orientations. A canonical source v 0 ⋯ v L − 1 v_0 \cdots v_{L-1} v 0 ⋯ v L − 1 matches when, for one d ∈ { 0 , 1 , 2 } d \in\{0,1,2\} d ∈ { 0 , 1 , 2 } , either
v k = d + j t + k ( m o d 3 ) ( 0 ≤ k < L ) , or v k = d − j t − k ( m o d 3 ) ( 0 ≤ k < L ) . (51) v_k = d + j_{t+k} \pmod{3} \quad(0 \le k < L), \qquad\text{or} \qquad v_k = d - j_{t-k} \pmod{3} \quad(0 \le k < L). \tag*{(51)} v k = d + j t + k ( mod 3 ) ( 0 ≤ k < L ) , or v k = d − j t − k ( mod 3 ) ( 0 ≤ k < L ) . (51)
The second match means full reflection with reversal of time. Its state, frame and both covectors are transported with the control; it is not a bare renaming of labels.
Require L < B L < B L < B for every row, including every supporting density witness. For row 16, reconstruct both additional proper 201020 sources and check that their local position-3 owners are exactly the two actual own anchors. These guards supply density facts and contribute no cost. The same 3/7 payment is unavailable without both witnesses.
Transport every charged offset to its physical arc modulo B B B , after applying the true seam twist. This yields nonnegative rows
∑ i = 0 B − 1 a r i c i ≥ b r . \sum_{i=0}^{B-1} a_{ri} c_i \ge b_r. i = 0 ∑ B − 1 a r i c i ≥ b r .
with the stated strictness. Their supports may overlap. A payment witness is a finite list of nonnegative rational λ r \lambda_r λ r with
γ i ≔ ∑ r λ r a r i ≤ 1 ( 0 ≤ i < B ) , L ( w ) ≔ ∑ r λ r b r ≥ C ∗ . (52) \gamma_i \coloneqq\sum_r \lambda_r a_{ri} \le1 \quad(0 \le i < B), \qquad L(w) \coloneqq\sum_r \lambda_r b_r \ge C_*. \tag*{(52)} γ i : = r ∑ λ r a r i ≤ 1 ( 0 ≤ i < B ) , L ( w ) : = r ∑ λ r b r ≥ C ∗ . (52)
Theorem 6.2 then gives C ≥ L ( w ) ≥ C ∗ C \ge L(w) \ge C_* C ≥ L ( w ) ≥ C ∗ , contradicting the actual strict budget C < C ∗ C < C_* C < C ∗ . Thus equality in the five-source or nine-source payment causes no difficulty. We do not silently make those weak bounds strict.
Proposition 17.1 (Exact finite payment reduction). The domain W \mathcal{W} W has 1355 members. The rule above has rational payment witnesses excluding 1334 of them. The remaining 21 words are exactly Table 4 . They consist of the all-minus word and mixed families of sizes 6, 7, 7.
Table 4: The twenty-one necessary increment types before the subsequent exclusions, grouped by the number M M M of positive runs. The subsequent comparisons leave the six types in Table 5 , all excluded later. A pair ( a , b ) (a,b) ( a , b ) records a positive run of length a a a followed by a negative run of length b b b . Only cyclic rotation is used to choose a representative.
Table 4. The twenty-one necessary increment types before the subsequent exclusions, grouped by the number M M M of positive runs. The subsequent comparisons leave the six types in Table 5 , all excluded later. A pair ( a , b ) (a,b) ( a , b ) records a positive run of length a a a followed by a negative run of length b b b . Only cyclic rotation is used to choose a representative.
Table 5. The six long types left by the qualified local comparisons.
Exact finite proof. The following counts give the initial and remaining words by length. The remaining sets themselves are printed in Table 4 ; counts are not used in place of equality of those sets. The finite witness and its standard-library reader are supplied as capacity_witnesses.json.gz and verify.py in the reader archive’s directory mathematical_inputs/development/run_reduction/. The reader independently enumerates all 2 096 896 binary strings of lengths eight through twenty and checks equality of the resulting necessary domain with the disjoint union of the 1334 excluded and 21 retained strings.
Table 8.
For each excluded word, every witness entry records its table row, nonnegative rational weight, complete unwrapped ordered support, orientation and whole phase. A strengthened own entry also records both complete six-source guards and their anchor owners. The reader reconstructs each source by (51) , checks strict properness, reconstructs the charge, and verifies (52) by integer and rational arithmetic. The record has 8266 positive-weight source applications, 256 density witnesses and 24 501 physical-arc capacity checks. All pass the stated exact conditions. These are finite rational sums; no floating-point minimum is an input to the proof.
The reader also decomposes every retained word into the maximal runs of § 3.33 . Exactly six have all negative runs of length two; seven have one length-five negative run and otherwise length-two negative runs; seven have one negative run of length seven, eight or nine and otherwise length-two negative runs. The sole pure word is all-minus. This proves the claimed set partition. □
The witnesses require only feasible rational combinations, not the optimal value of any linear program. For each retained word the supplement also gives a nonnegative comparison cost vector of sum less than C ∗ C_{*} C ∗ satisfying the weak versions of every placement of the chosen menu. Those vectors show the scope of this fixed scalar pass. They are not trajectories, and they neither establish realization nor preclude stronger inequalities or the conditional geometric arguments used next.
Theorem 17.2 (Three mixed families and the all-minus word). Every original global minimizer with B ≥ 8 B \ge8 B ≥ 8 has one of the twenty-one essential increment types in Table 4 . Its mixed word has one of the three run forms displayed there. In particular every original global minimizer has B ≤ 13 B \le13 B ≤ 13 , and any prescribed fundamental cut has at most N ≤ 14 N \le14 N ≤ 14 positive pieces.
Proof. The independent coarse bound and actual switching restrictions put every minimizer with B ≥ 8 B \ge8 B ≥ 8 in (50) . Lemma 16.1 gives each eligible physical payment used by Proposition 17.1. Any word outside Table 4 would therefore contradict C < C ∗ C < C_* C < C ∗ . Its longest retained word has length thirteen. A minimizer with B ≤ 7 B \le7 B ≤ 7 already satisfies that bound, and a prescribed cut splits at most one positive physical arc. Thus N ≤ B + 1 ≤ 14 N \le B + 1 \le14 N ≤ B + 1 ≤ 14 . □
The following all-length argument derives the same twenty-one words directly from the run inequalities, without the preceding numerical cutoff or word enumeration. The later geometric exclusions are organized by these run forms. The uniform family of Proposition 12.4 illustrates why the boundary capacity matters: the two short positive runs pay on their own arcs, leaving both boundary arcs available to the longer positive run. Other candidates require the actual rooted responses, exhaustive alternatives or central state information specified in their proofs.
All-length structure of the increment runs
The local costs now bound both the number and the lengths of the increment runs. We use the assignment of own arcs and ports in § 3.33 . For a mixed cycle write
( a 1 , b 1 ) , … , ( a M , b M ) , a i ≥ 1 , b i ≥ 2 , b i ≠ 3 , (a_1,b_1),\ldots,(a_M,b_M),\qquad a_i\ge1,\qquad b_i\ge2,\qquad b_i\ne3, ( a 1 , b 1 ) , … , ( a M , b M ) , a i ≥ 1 , b i ≥ 2 , b i = 3 ,
and put
B = ∑ i ( a i + b i ) , L = # { i : b i ≥ 4 } . B=\sum_i(a_i+b_i),\qquad L=\#\{i:b_i\ge4\}. B = i ∑ ( a i + b i ) , L = # { i : b i ≥ 4 } .
Thus L L L counts the negative runs longer than two. The original twist (19) gives
∑ i ( a i − b i ) ≡ 1 ( m o d 3 ) . (53) \sum_i(a_i-b_i)\equiv1\pmod{3}. \tag*{(53)} i ∑ ( a i − b i ) ≡ 1 ( mod 3 ) . (53)
Throughout, C C C is the total physical cost and
P = 180483 500000 , C ∗ = 5 P = 180483 100000 , C < C ∗ . P=\frac{180483}{500000},\qquad C_* = 5P=\frac{180483}{100000},\qquad C<C_*. P = 500000 180483 , C ∗ = 5 P = 100000 180483 , C < C ∗ .
Theorem 18.1 (All-length run structure). For an original global minimizer with a mixed increment cycle and B ≥ 8 B\ge8 B ≥ 8 ,
M ≤ 3 , a i ≤ 8 , b i ∈ { 2 , 5 , 7 , 8 , 9 } . M\le3,\qquad a_i\le8,\qquad b_i\in\{2,5,7,8,9\}. M ≤ 3 , a i ≤ 8 , b i ∈ { 2 , 5 , 7 , 8 , 9 } . At most one negative run has length greater than two.
We first bound M M M by a uniform half-port payment, then bound the negative lengths and their multiplicity. The few remaining lengths four and six are excluded by explicit compatible charges. The last step bounds the positive lengths. A source is always retained with all its complete supporting arcs. Separately proved scalar costs are added by Theorem 6.2 ; their selected full lifts need not be identified.
A uniform payment for every positive run.
Lemma 18.2. Suppose B ≥ 10 B\ge10 B ≥ 10 . If d ∈ { 0 , 1 , 2 } d\in\{0,1,2\} d ∈ { 0 , 1 , 2 } counts the incident negative runs of length two, every positive run has a strict charged profile of value at least p d p_d p d , where
p 0 = 389 2000 , p 1 = 473 2000 , p 2 = 119 250 . p_0=\frac{389}{2000},\qquad p_1=\frac{473}{2000},\qquad p_2=\frac{119}{250}. p 0 = 2000 389 , p 1 = 2000 473 , p 2 = 250 119 . The load is at most one on each own arc and at most one half on each formal port.
Proof. For a positive run of length one, half the complete 102021 or 1020212 four-arc profile gives p 0 p_0 p 0 or p 1 p_1 p 1 . If d = 2 d=2 d = 2 , the complete eight-source 01020212 pays more than 74 / 125 74/125 74/125 on its four-arc core at local positions 2, …, 5 and more than 9 / 25 9/25 9/25 on its two own arcs at positions 3, 4. Giving each inequality weight one half yields
1 2 ( 74 125 + 9 25 ) = 119 250 . \frac{1}{2}\left(\frac{74}{125}+\frac{9}{25}\right)=\frac{119}{250}. 2 1 ( 125 74 + 25 9 ) = 250 119 .
Each own arc has load one and each port load one half. The four-arc and two-arc bounds are Theorem 8.3 and Lemma 12.1, respectively.
For a run of length two, the proper five-source 02010 pays 49/200 on its three own arcs when d ≤ 1 d \le1 d ≤ 1 . When d = 2 d = 2 d = 2 , Corollary 12.3 pays 1/2 on the same own set. Both choices have zero port loads and dominate the stated values.
For a ≥ 3 a \ge3 a ≥ 3 , Lemma C.13 gives
caption:
:::
a strict half-port payment 3 51 / 100 + 7 d / 200 4 431 / 1000 + 7 d / 200 a ≥ 5 61 / 200 + P ⌊ ( a − 3 ) / 2 ⌋ + 7 d / 200 (54) \begin{array}{c|c}
a & \text{strict half-port payment} \\
3 & 51/100 + 7d/200 \\
4 & 431/1000 + 7d/200 \\
a \ge5 & 61/200 + P\lfloor(a-3)/2\rfloor+ 7d/200 \tag*{(54)}
\end{array} a 3 4 a ≥ 5 strict half-port payment 51/100 + 7 d /200 431/1000 + 7 d /200 61/200 + P ⌊( a − 3 ) /2 ⌋ + 7 d /200 (54)
For every d d d its minimum occurs at a = 4 a = 4 a = 4 and dominates p d p_d p d . Indeed the a ≥ 5 a \ge5 a ≥ 5 value increases by P P P every two increments, and its first two values already exceed the a = 4 a = 4 a = 4 value. All supporting sources have at most nine arcs, so they are proper under B ≥ 10 B \ge10 B ≥ 10 .
Lemma 18.3. Every mixed minimizing increment cycle has M ≤ 3 M \le3 M ≤ 3 .
Proof. If M ≥ 4 M \ge4 M ≥ 4 , then B ≥ 3 M ≥ 12 B \ge3M \ge12 B ≥ 3 M ≥ 12 , so the preceding profiles are proper. They satisfy
p d ≥ 119 250 − 479 2000 ( 2 − d ) . p_d \ge\frac{119}{250} - \frac{479}{2000}(2-d). p d ≥ 250 119 − 2000 479 ( 2 − d ) .
There is equality at d = 1 , 2 d = 1, 2 d = 1 , 2 ; at d = 0 d = 0 d = 0 the right side is − 3 / 1000 -3/1000 − 3/1000 . The negative half-port profile of Lemma C.14 has value zero at b = 2 b = 2 b = 2 and at least 3 / 5 3/5 3/5 at every allowed b ≥ 4 b \ge4 b ≥ 4 . Each long negative run has two incident slots, so ∑ i ( 2 − d i ) = 2 L \sum_i(2-d_i)=2L ∑ i ( 2 − d i ) = 2 L . Adding these profiles gives
C > 119 250 M + ( 3 5 − 479 1000 ) L = 119 250 M + 121 1000 L . (55) C > \frac{119}{250}M + \left(\frac{3}{5} - \frac{479}{1000}\right)L = \frac{119}{250}M + \frac{121}{1000}L. \tag*{(55)} C > 250 119 M + ( 5 3 − 1000 479 ) L = 250 119 M + 1000 121 L . (55)
At a length-two negative run the positive half-port loads add to at most one on the shared physical arc. At a longer negative run, each positive half-load meets at most one negative half-load. All other charges lie in disjoint own sets or negative interiors. Thus the physical capacity is at most one, also at repeated incidences. The positive profiles make the sum strict. For M ≥ 4 M \ge4 M ≥ 4 the right side is at least 238 / 125 > C ∗ 238/125 > C_* 238/125 > C ∗ , a contradiction.
The negative lengths and their multiplicity
Lemma 18.4. Every negative run of a mixed minimizing cycle has length at most nine.
Proof. A negative length b ≥ 10 b \ge10 b ≥ 10 implies B ≥ 11 B \ge11 B ≥ 11 , so all half-port sources are proper. Let V ( b ) V(b) V ( b ) be the negative payment of Lemma C.14. For b ≥ 11 b \ge11 b ≥ 11 , V ( b ) ≥ 9 / 5 V(b) \ge9/5 V ( b ) ≥ 9/5 : this holds at b = 11 , 12 , 13 b = 11,12,13 b = 11 , 12 , 13 and V ( b + 3 ) − V ( b ) = 3 / 5 V(b+3)-V(b)=3/5 V ( b + 3 ) − V ( b ) = 3/5 thereafter. One positive profile therefore gives C > 9 / 5 + 389 / 2000 = 3989 / 2000 > C ∗ . C > 9/5 + 389/2000 = 3989/2000 > C_*. C > 9/5 + 389/2000 = 3989/2000 > C ∗ .
At b = 10 b = 10 b = 10 , V ( b ) = 3 / 2 V(b) = 3/2 V ( b ) = 3/2 . If M ≥ 2 M \ge2 M ≥ 2 , two positive profiles give C > 3 / 2 + 2 ( 389 / 2000 ) = 1889 / 1000 > C ∗ C > 3/2 + 2(389/2000) = 1889/1000 > C_* C > 3/2 + 2 ( 389/2000 ) = 1889/1000 > C ∗ . For M = 1 M = 1 M = 1 , the twist gives a ≡ 2 ( m o d 3 ) a \equiv2 \pmod{3} a ≡ 2 ( mod 3 ) , so a = 2 a = 2 a = 2 or a ≥ 5 a \ge5 a ≥ 5 . Number the negative run’s arcs 0 , … , 10 0,\ldots,10 0 , … , 10 . Its nine interior arcs split into
[ 1 , 3 ] , [ 4 , 6 ] , [ 7 , 9 ] . [1,3],\quad[4,6],\quad[7,9]. [ 1 , 3 ] , [ 4 , 6 ] , [ 7 , 9 ] .
with complete five-sources [ 0 , 4 ] [0,4] [ 0 , 4 ] , [ 3 , 7 ] [3,7] [ 3 , 7 ] , [ 6 , 10 ] [6,10] [ 6 , 10 ] . These proper lawful images of 10210 give the weak total 9 / 5 9/5 9/5 , with unit load even at the two ports. At a = 2 a = 2 a = 2 add the zero-port 49/200 own payment from 02010. At a ≥ 5 a \ge5 a ≥ 5 a complete 012012 source inside the positive run gives a disjoint own pair above P > 49 / 200 P > 49/200 P > 49/200 . Hence in either case
C > 9 5 + 49 200 = 409 200 > C ∗ . C > \frac{9}{5} + \frac{49}{200} = \frac{409}{200} > C_*. C > 5 9 + 200 49 = 200 409 > C ∗ .
Here the twist supplies exactly the zero-port positive payment needed after both negative ports have been used.
Lemma 18.5. A mixed minimizing cycle has at most one negative run of length greater than two. Proof. Suppose L ≥ 2 L \ge2 L ≥ 2 . Then B ≥ 10 B \ge10 B ≥ 10 , so all half-port profiles are proper, and M ≤ 3 M \le3 M ≤ 3 . If M = 3 M=3 M = 3 , L = 2 L=2 L = 2 , the positive incidences have d = 0 , 1 , 1 d=0,1,1 d = 0 , 1 , 1 . Thus
C > 2 3 5 + 389 2000 + 2 473 2000 = 747 400 > C ∗ . C > 2\frac{3}{5}+\frac{389}{2000}+2\frac{473}{2000}=\frac{747}{400}>C_*. C > 2 5 3 + 2000 389 + 2 2000 473 = 400 747 > C ∗ .
If M = L = 3 M=L=3 M = L = 3 , the lower bound is larger: C > 9 / 5 + 3 ( 389 / 2000 ) > C ∗ . C>9/5+3(389/2000)>C_*. C > 9/5 + 3 ( 389/2000 ) > C ∗ .
It remains M = L = 2 M=L=2 M = L = 2 . Every allowed negative length other than five has V ( b ) ≥ 9 / 10 V(b)\ge9/10 V ( b ) ≥ 9/10 ; this follows from the exceptional values at four and six and the three increasing residue classes. If either negative length differs from five, then C > 9 / 10 + 3 / 5 + 2 ( 389 / 2000 ) > C ∗ . C>9/10+3/5+2(389/2000)>C_*. C > 9/10 + 3/5 + 2 ( 389/2000 ) > C ∗ . If both are five and a positive length is at least three, (18.2) gives
C > 6 5 + 431 1000 + 389 2000 = 3651 2000 > C ∗ . C>\frac{6}{5}+\frac{431}{1000}+\frac{389}{2000}=\frac{3651}{2000}>C_*. C > 5 6 + 1000 431 + 2000 389 = 2000 3651 > C ∗ .
Both positive lengths must therefore belong to { 1 , 2 } \{1,2\} { 1 , 2 } . The twist a 1 + a 2 − 10 ≡ 1 ( m o d 3 ) a_1+a_2-10\equiv1\pmod{3} a 1 + a 2 − 10 ≡ 1 ( mod 3 ) leaves only ( a 1 , a 2 ) = ( 1 , 1 ) (a_1,a_2)=(1,1) ( a 1 , a 2 ) = ( 1 , 1 ) .
In this configuration − 5 + + − 5 + -5^+ + -5^+ − 5 + + − 5 + , each negative run has complete eight-source 20210212. Theorem E.2 pays more than 13 / 20 13/20 13/20 on its two middle interior arcs, avoiding both ports. Each positive singleton uses its full 102021 profile above 389 / 1000 389/1000 389/1000 , charging its own pair and its two ports. These four charged sets are disjoint: the two ports of a length-five negative run are distinct. Every complete source is proper at B = 12 B=12 B = 12 , and
C > 2 13 20 + 2 389 1000 = 1039 500 > C ∗ . C>2\frac{13}{20}+2\frac{389}{1000}=\frac{1039}{500}>C_*. C > 2 20 13 + 2 1000 389 = 500 1039 > C ∗ .
Lemma 18.6. A mixed minimizing cycle with B ≥ 8 B\ge8 B ≥ 8 has no negative run of length four or six.
Proof. There is at most one such run, and all other negative lengths are two. Write its length as b = 4 b=4 b = 4 or 6.
If M = 3 M=3 M = 3 , the positive incidences have d = 1 , 1 , 2 d=1,1,2 d = 1 , 1 , 2 and pay at least 2 p 1 + p 2 = 949 / 1000 2p_1+p_2=949/1000 2 p 1 + p 2 = 949/1000 . Adding the negative payment 299 / 250 299/250 299/250 or 13 / 10 13/10 13/10 exceeds C ∗ C_* C ∗ . Here B ≥ 11 B\ge11 B ≥ 11 , so all sources are proper.
Suppose M = 2 M=2 M = 2 . At b = 4 b=4 b = 4 the twist gives a 1 + a 2 ≡ 1 ( m o d 3 ) a_1+a_2\equiv1\pmod{3} a 1 + a 2 ≡ 1 ( mod 3 ) , hence B ≥ 10 B\ge10 B ≥ 10 ; at b = 6 b=6 b = 6 it gives a 1 + a 2 ≡ 0 ( m o d 3 ) a_1+a_2\equiv0\pmod{3} a 1 + a 2 ≡ 0 ( mod 3 ) , hence B ≥ 11 B\ge11 B ≥ 11 . If a positive length is at least three, its half-port value with d = 1 d=1 d = 1 is at least 233 / 500 233/500 233/500 , the minimum in (18.2). The other positive run pays at least 473 / 2000 473/2000 473/2000 . Adding even the smaller negative payment gives C > 299 / 250 + 233 / 500 + 473 / 2000 > C ∗ C>299/250+233/500+473/2000>C_* C > 299/250 + 233/500 + 473/2000 > C ∗ . Both positive lengths are therefore at most two. The twist leaves ( 2 , 2 ) (2,2) ( 2 , 2 ) at b = 4 b=4 b = 4 , and ( 1 , 2 ) (1,2) ( 1 , 2 ) or ( 2 , 1 ) (2,1) ( 2 , 1 ) at b = 6 b=6 b = 6 .
For b = 6 b=6 b = 6 , the two separate negative singleton payments total 13 / 10 13/10 13/10 and avoid the ports. Use the singleton positive run’s full 1020212 profile above 473 / 1000 473/1000 473/1000 , and the other positive run’s zero-port 49 / 200 49/200 49/200 own profile. Only the singleton uses the shared short-separator port. These charges give
C > 13 10 + 473 1000 + 49 200 = 1009 500 > C ∗ . C>\frac{13}{10}+\frac{473}{1000}+\frac{49}{200}=\frac{1009}{500}>C_*. C > 10 13 + 1000 473 + 200 49 = 500 1009 > C ∗ . For ( b , a 1 , a 2 ) = ( 4 , 2 , 2 ) (b,a_1,a_2)=(4,2,2) ( b , a 1 , a 2 ) = ( 4 , 2 , 2 ) , start the word at − − − − + + − − + + ----++--++ − − − − + + − − + + and set j 0 = 0 j_0=0 j 0 = 0 . Reconstruct all labels from the integer walk in (2.20), including negative indices. The following whole-source maps give three proper placements:
ordered support map source physical charge ( 2 , 1 , … , − 5 ) 2 − j 10201020 { 9 , 0 , 1 } ( 5 , 4 , … , − 1 ) 2 − j 2021020 { 2 } ( 2 , 3 , … , 9 ) j 10201020 { 3 , 4 , 5 } (56) \begin{array}{c|c|c|c}
\text{ordered support} & \text{map} & \text{source} & \text{physical charge}\\
(2,1,\ldots,-5) & 2-j & 10201020 & \{9,0,1\}\\
(5,4,\ldots,-1) & 2-j & 2021020 & \{2\}\\
(2,3,\ldots,9) & j & 10201020 & \{3,4,5\} \tag*{(56)}
\end{array} ordered support ( 2 , 1 , … , − 5 ) ( 5 , 4 , … , − 1 ) ( 2 , 3 , … , 9 ) map 2 − j 2 − j j source 10201020 2021020 10201020 physical charge { 9 , 0 , 1 } { 2 } { 3 , 4 , 5 } (56)
The eight-sources retain the complete seven-arc sign witnesses required for the nonpositive-tail transfer of Corollary B.76; their triples each exceed 61 / 200 61/200 61/200 . The seven-source retains the two contact jets and terminal 020 required for its 299 / 250 299/250 299/250 center payment. Their physical charges are disjoint, so
C > 2 61 200 + 299 250 = 903 500 > C ∗ . C > 2\frac{61}{200}+\frac{299}{250}=\frac{903}{500}>C_*. C > 2 200 61 + 250 299 = 500 903 > C ∗ .
Finally let M = 1 M=1 M = 1 . At b = 4 b=4 b = 4 the twist gives a ≡ 2 ( m o d 3 ) a\equiv2\pmod{3} a ≡ 2 ( mod 3 ) . The value a = 2 a=2 a = 2 has B = 6 B=6 B = 6 . Every a ≥ 5 a\geq5 a ≥ 5 has positive half-port payment at least 61 / 200 + P 61/200+P 61/200 + P , which together with 299 / 250 299/250 299/250 exceeds C ∗ C_* C ∗ . At a = 5 a=5 a = 5 , B = 9 B=9 B = 9 this construction still uses only proper six- and seven-sources. At b = 6 b=6 b = 6 the twist gives a ≡ 1 ( m o d 3 ) a\equiv1\pmod{3} a ≡ 1 ( mod 3 ) . The value a = 1 a=1 a = 1 has B = 7 B=7 B = 7 . For a ≥ 7 a\geq7 a ≥ 7 , the positive half-port value is at least 61 / 200 + 2 P 61/200+2P 61/200 + 2 P and its sum with 13 / 10 13/10 13/10 exceeds C ∗ C_* C ∗ . At a = 4 a=4 a = 4 , B = 10 B=10 B = 10 , use the two full ordinary terminal triples above 61 / 200 61/200 61/200 each, the middle own arc above 63 / 500 63/500 63/500 , and the two negative singleton payments totaling 13 / 10 13/10 13/10 . The negative charges avoid both ports, so
C > 2 61 200 + 63 500 + 13 10 = 509 250 > C ∗ . C > 2\frac{61}{200}+\frac{63}{500}+\frac{13}{10}=\frac{509}{250}>C_*. C > 2 200 61 + 500 63 + 10 13 = 250 509 > C ∗ .
Positive lengths and pure cycles.
Lemma 18.7. Every positive run of a mixed minimizing cycle has length at most eight.
Proof. Suppose a ≥ 9 a\geq9 a ≥ 9 . Then B ≥ 11 B\geq11 B ≥ 11 , so the sources of at most nine arcs used here are proper. If a negative run is longer than two, its half-port profile pays at least 3 / 5 3/5 3/5 . The positive run’s half-port value is at least 61 / 200 + 3 P 61/200+3P 61/200 + 3 P , so C > 61 / 200 + 3 P + 3 / 5 > C ∗ C>61/200+3P+3/5>C_* C > 61/200 + 3 P + 3/5 > C ∗ . If all negative lengths are two and M ≥ 2 M\geq2 M ≥ 2 , this positive run pays at least 3 / 8 + 3 P 3/8+3P 3/8 + 3 P and another positive run pays at least 119 / 250 119/250 119/250 . Their compatible half-port sum again exceeds C ∗ C_* C ∗ . It remains M = 1 M=1 M = 1 , b = 2 b=2 b = 2 . The two formal ports now coincide, so we instead use the complete nine-source 012021201 across the short separator. Its incremental pattern + + + − − + + + +++\mathord{-}\mathord{-}+\mathord{++} + + + − − + + + is supplied by the end and beginning of the same positive run. Lemma D.3 pays the weak value 2 P 2P 2 P on the last two own arcs, the shared port, and the first two own arcs. Number the own arcs 0 , … , a 0,\ldots,a 0 , … , a . The full support is [ a − 3 , a + 5 ] [a-3,a+5] [ a − 3 , a + 5 ] , under the whole phase j ↦ j − ( a − 3 ) j\mapsto j-(a-3) j ↦ j − ( a − 3 ) ; its physical charge is { a − 1 , a , a + 1 , 0 , 1 } \{a-1,a,a+1,0,1\} { a − 1 , a , a + 1 , 0 , 1 } modulo B = a + 2 B=a+2 B = a + 2 . These five arcs are distinct. The three pairs
{ 2 , 3 } , { 4 , 5 } , { 6 , 7 } \{2,3\},\qquad\{4,5\},\qquad\{6,7\} { 2 , 3 } , { 4 , 5 } , { 6 , 7 } have complete forward sources [ 0 , 5 ] [0,5] [ 0 , 5 ] , [ 2 , 7 ] [2,7] [ 2 , 7 ] , [ 4 , 9 ] [4,9] [ 4 , 9 ] inside the positive run and each cost more than P P P . They avoid the first and last own pairs because a ≥ 9 a\geq9 a ≥ 9 . Thus C > 2 P + 3 P = C ∗ C>2P+3P=C_* C > 2 P + 3 P = C ∗ , where strictness comes from the three forward pairs.
Proof of Theorem 18.1. Lemmas 18.3, 18.4 and 18.5 bound the number of positive runs, the negative lengths and their multiplicity. Lemma 18.6 removes the negative lengths four and six for B ≥ 8 B\geq8 B ≥ 8 . Lemma 18.7 gives the positive-length bound.
Proposition 18.8 (Pure increment cycles). Among pure increment cycles with B ≥ 8 B\geq8 B ≥ 8 , only the all-minus cycle of length eight can belong to an original global minimizer.
Proof. For a pure positive cycle, the twist gives B ≡ 1 ( m o d 3 ) B\equiv1\pmod{3} B ≡ 1 ( mod 3 ) , so B ≥ 8 B\geq8 B ≥ 8 implies B ≥ 10 B\geq10 B ≥ 10 . Use the proper complete 012012 source at every physical start with coefficient one half. Each arc belongs to two charged pairs, so its physical load is one. The strict forward-pair bound gives C > B / 2 ≥ 5 P = C ∗ . C> B/2\geq5P=C_*. C > B /2 ≥ 5 P = C ∗ .
For a pure negative cycle, B ≡ 2 ( m o d 3 ) B\equiv2\pmod{3} B ≡ 2 ( mod 3 ) . Use every proper complete five-source with coefficient one third. Each arc belongs to three charged triples, giving load one and the weak bound C ≥ B / 5 C\geq B/5 C ≥ B /5 . Apart from B = 8 B=8 B = 8 , the first possible length is eleven, which already
exceeds the budget. In both averages a source crossing the seam uses the original twisted labels and its lawful phase. Only physical scalar costs are summed. □
Three positive runs
We now determine the mixed cycles with three positive runs. Theorem 18.1 leaves either three negative runs of length two or one longer negative run and two of length two. All charges below are original physical costs. We retain the complete proper sources and their whole lawful charts from § 3.36 ; each source keeps the full maximizing data required by its proof. Independently established scalar payments may be added without identifying their selected lifts.
Proposition 19.1. Suppose a mixed minimizing cycle has three positive runs and all its negative runs have length two. Its positive lengths must be ( 1 , 1 , 2 ) (1,1,2) ( 1 , 1 , 2 ) in cyclic order.
Proof. Put A = a 1 + a 2 + a 3 A=a_1+a_2+a_3 A = a 1 + a 2 + a 3 . The twist gives A ≡ 1 ( m o d 3 ) A \equiv1 \pmod{3} A ≡ 1 ( mod 3 ) and hence B = A + 6 ≥ 10 B=A+6 \ge10 B = A + 6 ≥ 10 . A useful charge across a short separator is
012021201 , charge 2 : 6 , payment ≥ 2 P . 012021201,\qquad\text{charge }2:6,\qquad\text{payment }\ge2P. 012021201 , charge 2 : 6 , payment ≥ 2 P .
Its increment factor is + + + − − + + + +++--+++ + + + − − + + + . Between positive runs of length at least three, it charges the last two own arcs on the left, the shared port, and the first two own arcs on the right. The complete nine-source retains both returned contexts and their internal forward witnesses, as required by Lemma D.3. The two terminal own pairs within any such positive run are disjoint.
First suppose none of the positive lengths is one. If all three are at least three, use this nine-source at every separator. Their physical charges are disjoint, so C ≥ 6 P > C ∗ C \ge6P > C_* C ≥ 6 P > C ∗ . If exactly one length is two, pay 1 / 2 1/2 1/2 on its three own arcs by Corollary 12.3. Between the other two runs use the nine-source, and at their two other ends use the returned terminal triples. The charges are disjoint and give
C > 1 2 + 2 P + 2 379 1000 > C ∗ . C > \frac{1}{2}+2P+2\frac{379}{1000}>C_*. C > 2 1 + 2 P + 2 1000 379 > C ∗ .
Otherwise write the lengths as ( 2 , 2 , a ) (2,2,a) ( 2 , 2 , a ) . The twist forces a ≡ 0 ( m o d 3 ) a \equiv0 \pmod{3} a ≡ 0 ( mod 3 ) , so a = 3 a=3 a = 3 or a ≥ 6 a \ge6 a ≥ 6 . For a = 3 a=3 a = 3 , let s s s be the start of that run and set j s = 0 j_s=0 j s = 0 . The complete source [ s − 3 , s + 6 ] = 1210120212 [s-3,s+6]=1210120212 [ s − 3 , s + 6 ] = 1210120212 , under the whole phase j ↦ j + 1 j\mapsto j+1 j ↦ j + 1 , is 2021201020 2021201020 2021201020 . Its charge 2 : 7 2:7 2 : 7 is [ s − 1 , s + 4 ] [s-1,s+4] [ s − 1 , s + 4 ] , the four own arcs and both ports of the length-three run. Both actual exterior returns are retained. The strict 807 / 1000 807/1000 807/1000 bound of Proposition C.12, together with the two zero-port half-unit payments, gives
C > 1 2 + 1 2 + 807 1000 = 1807 1000 > C ∗ . C>\frac{1}{2}+\frac{1}{2}+\frac{807}{1000}=\frac{1807}{1000}>C_*. C > 2 1 + 2 1 + 1000 807 = 1000 1807 > C ∗ .
For a ≥ 6 a \ge6 a ≥ 6 , the two returned terminal triples and one disjoint internal forward pair instead give C > 1 + 2 ( 379 / 1000 ) + P > C ∗ C>1+2(379/1000)+P>C_* C > 1 + 2 ( 379/1000 ) + P > C ∗ . Here B ≥ 13 B\ge13 B ≥ 13 , so even the ten-source used at a = 3 a=3 a = 3 is proper.
Now suppose exactly one positive length is one. If its two neighbors are at least three, Lemma 12.1 pays 3 / 7 3/7 3/7 on its own pair, with zero port loads. The two proper six-source witnesses give the required physical anchor-density facts; they do not identify covectors across their containing intervals. Use the nine-source between the two neighbors and one returned triple at each of their other ends. This yields
C > 3 7 + 2 P + 2 379 1000 > C ∗ . C>\frac{3}{7}+2P+2\frac{379}{1000}>C_*. C > 7 3 + 2 P + 2 1000 379 > C ∗ .
Otherwise the two possible cyclic orders are ( 1 , 2 , a ) (1,2,a) ( 1 , 2 , a ) and ( 1 , a , 2 ) (1,a,2) ( 1 , a , 2 ) . The twist gives a ≡ 1 ( m o d 3 ) a\equiv1\pmod{3} a ≡ 1 ( mod 3 ) , hence a ≥ 4 a\ge4 a ≥ 4 . The own payments 3 / 7 3/7 3/7 and 1 / 2 1/2 1/2 leave every port available to the longer run. Its two returned triples and either its middle-center payment (a = 4 a = 4 a = 4 ) or an internal forward pair (a ≥ 5 a \ge5 a ≥ 5 ) give, in either order,
C > 3 7 + 1 2 + 2 379 1000 + 63 500 = 1586 875 > C ∗ . C > \frac{3}{7} + \frac{1}{2} + 2\frac{379}{1000} + \frac{63}{500} = \frac{1586}{875} > C_*. C > 7 3 + 2 1 + 2 1000 379 + 500 63 = 875 1586 > C ∗ .
These are the full source placements of Proposition 12.4; the two whole lawful placements do not identify the two cyclic types.
Finally, with at least two singleton runs write the lengths as ( 1 , 1 , a ) (1,1,a) ( 1 , 1 , a ) . The twist gives a ≡ 2 ( m o d 3 ) a \equiv2 \pmod{3} a ≡ 2 ( mod 3 ) . If a ≥ 5 a \ge5 a ≥ 5 , each singleton pays 9 / 25 9/25 9/25 on its own pair, while the other run pays its two returned triples and one internal pair. Thus
C > 2 9 25 + 2 379 1000 + P = 919483 500000 > C ∗ . C > 2\frac{9}{25} + 2\frac{379}{1000} + P = \frac{919483}{500000} > C_*. C > 2 25 9 + 2 1000 379 + P = 500000 919483 > C ∗ .
The only value left is a = 2 a = 2 a = 2 . The cases above exhaust the numbers of positive lengths equal to one and two. Every nine-source used is proper, and the sole ten-source was used only at B ≥ 13 B \ge13 B ≥ 13 .
Proposition 19.2. A mixed minimizing cycle with three positive runs has no negative run longer than two.
Proof. There is at most one long negative run. If its length is seven, eight or nine, its half-port value is at least 9 / 10 9/10 9/10 . The three positive incidences have d = 1 , 1 , 2 d = 1,1,2 d = 1 , 1 , 2 , so Lemma 18.2 gives
C > 9 10 + 2 473 2000 + 119 250 = 1849 1000 > C ∗ . C > \frac{9}{10} + 2\frac{473}{2000} + \frac{119}{250} = \frac{1849}{1000} > C_*. C > 10 9 + 2 2000 473 + 250 119 = 1000 1849 > C ∗ .
It remains to consider negative lengths ( 5 , 2 , 2 ) (5,2,2) ( 5 , 2 , 2 ) . Write the positive lengths, starting after the length-five run, as ( x , y , z ) (x,y,z) ( x , y , z ) . The end runs have d = 1 d = 1 d = 1 and the middle run has d = 2 d = 2 d = 2 . The complete eight-source 20210212 pays more than 13 / 20 13/20 13/20 on the two middle interior arcs of the length-five run, with zero port loads, by Theorem E.2.
If x x x or z z z is at least three, its half-port payment is at least 233 / 500 233/500 233/500 , the minimum of Lemma C.13 at d = 1 d = 1 d = 1 . The other two positive profiles therefore give
C > 13 20 + 233 500 + 473 2000 + 119 250 = 3657 2000 > C ∗ . C > \frac{13}{20} + \frac{233}{500} + \frac{473}{2000} + \frac{119}{250} = \frac{3657}{2000} > C_*. C > 20 13 + 500 233 + 2000 473 + 250 119 = 2000 3657 > C ∗ .
Thus x , z ≤ 2 x,z \le2 x , z ≤ 2 . If y ≥ 5 y \ge5 y ≥ 5 , its half-port payment is at least 3 / 8 + P 3/8 + P 3/8 + P , giving C > 13 / 20 + 2 ( 473 / 2000 ) + 3 / 8 + P > C ∗ C > 13/20 + 2(473/2000) + 3/8 + P > C_* C > 13/20 + 2 ( 473/2000 ) + 3/8 + P > C ∗ . Consequently y ≤ 4 y \le4 y ≤ 4 . The twist is x + y + z ≡ 1 ( m o d 3 ) x + y + z \equiv1 \pmod{3} x + y + z ≡ 1 ( mod 3 ) . For x + z = 2 , 3 , 4 x + z = 2, 3, 4 x + z = 2 , 3 , 4 , respectively, this leaves y = 2 y = 2 y = 2 , y ∈ { 1 , 4 } y \in\{1,4\} y ∈ { 1 , 4 } , and y = 3 y = 3 y = 3 . The six possible triples are therefore
( 1 , 2 , 1 ) (1,2,1) ( 1 , 2 , 1 ) , ( 1 , 1 , 2 ) (1,1,2) ( 1 , 1 , 2 ) , ( 2 , 1 , 1 ) (2,1,1) ( 2 , 1 , 1 ) , ( 1 , 4 , 2 ) (1,4,2) ( 1 , 4 , 2 ) , ( 2 , 4 , 1 ) (2,4,1) ( 2 , 4 , 1 ) , ( 2 , 3 , 2 ) (2,3,2) ( 2 , 3 , 2 ) .
Their lengths are at least thirteen, so every source used below is proper.
At ( 1 , 2 , 1 ) (1,2,1) ( 1 , 2 , 1 ) give each end singleton its full one-return four-arc profile, above 473/1000 by Corollary B.17 . The middle length-two run pays 1 / 2 1/2 1/2 on its own arcs and the negative run pays its zero-port pair. The two long-separator ports are distinct; each short-separator port is charged by only one singleton. Hence
C > 2 473 1000 + 1 2 + 13 20 = 262 125 > C ∗ . C > 2\frac{473}{1000} + \frac{1}{2} + \frac{13}{20} = \frac{262}{125} > C_*. C > 2 1000 473 + 2 1 + 20 13 = 125 262 > C ∗ .
At ( 2 , 3 , 2 ) (2,3,2) ( 2 , 3 , 2 ) use the middle run’s complete W 807 W807 W 807 profile, including both actual exterior returns. The end runs use their zero-port 49 / 200 49/200 49/200 own payments from the complete five-source 02010. All these charges avoid the negative pair, so
C > 807 1000 + 2 49 200 + 13 20 = 1947 1000 > C ∗ . C > \frac{807}{1000} + 2\frac{49}{200} + \frac{13}{20} = \frac{1947}{1000} > C_*. C > 1000 807 + 2 200 49 + 20 13 = 1000 1947 > C ∗ .
At ( 1 , 4 , 2 ) (1,4,2) ( 1 , 4 , 2 ) give coefficient one half to the singleton’s full 473/1000 profile and to the adjacent returned terminal triple of the middle run. Their common port then has load one. Give coefficient one to the middle run’s other returned triple and its middle 63 / 500 63/500 63/500 payment, to the length-two run’s zero-port 49 / 200 49/200 49/200 own payment, and to the negative pair. The middle own charges are disjoint, and every other physical load is at most one. Thus
C > 1 2 473 1000 + 3 2 379 1000 + 63 500 + 49 200 + 13 20 = 913 500 > C ∗ . C > \frac{1}{2}\frac{473}{1000}+\frac{3}{2}\frac{379}{1000}+\frac{63}{500}+\frac{49}{200}+\frac{13}{20}=\frac{913}{500}>C_*. C > 2 1 1000 473 + 2 3 1000 379 + 500 63 + 200 49 + 20 13 = 500 913 > C ∗ .
For ( 2 , 4 , 1 ) (2,4,1) ( 2 , 4 , 1 ) interchange left and right by whole lawful reflection and time reversal, retaining the complete sources. This gives a separate compatible allocation for that cyclic type.
For the last two triples we record every source and charge. Take j 0 = 0 j_0=0 j 0 = 0 and reconstruct the native labels from the printed increment word, with j i + 13 = j i + 1 j_{i+13}=j_i+1 j i + 13 = j i + 1 modulo three. The maps in the tables act on the entire ordered source and its full lift; only physical costs are reduced modulo thirteen. The final column is a strict lower payment.
For ( x , y , z ) = ( 1 , 1 , 2 ) (x,y,z)=(1,1,2) ( x , y , z ) = ( 1 , 1 , 2 ) the word is − − − − − + − − + − − + + -\ -\ -\ -\ -+\ -\ -+\ -\ -+\ + − − − − − + − − + − − + + , and the placements are
support map source charge payment ( 2 , 1 , … , − 5 ) 2 − j 10201020 { 12 , 0 , 1 } 61 / 200 ( 6 , 5 , … , − 1 ) 1 − j 20210212 { 2 , 3 } 13 / 20 [ 3 , 9 ] 1 + j 1020212 { 4 , 5 , 6 , 7 } 473 / 1000 [ 5 , 12 ] 2 + j 01020212 { 8 , 9 } 9 / 25 [ 5 , 14 ] 2 + j 0102021202 { 11 } 3 / 40 (57) \begin{array}{c|c|c|c|c}
\text{support} & \text{map} & \text{source} & \text{charge} & \text{payment}\\
(2,1,\ldots,-5) & 2-j & 10201020 & \{12,0,1\} & 61/200\\
(6,5,\ldots,-1) & 1-j & 20210212 & \{2,3\} & 13/20\\
[3,9] & 1+j & 1020212 & \{4,5,6,7\} & 473/1000\\
[5,12] & 2+j & 01020212 & \{8,9\} & 9/25\\
[5,14] & 2+j & 0102021202 & \{11\} & 3/40 \tag*{(57)}
\end{array} support ( 2 , 1 , … , − 5 ) ( 6 , 5 , … , − 1 ) [ 3 , 9 ] [ 5 , 12 ] [ 5 , 14 ] map 2 − j 1 − j 1 + j 2 + j 2 + j source 10201020 20210212 1020212 01020212 0102021202 charge { 12 , 0 , 1 } { 2 , 3 } { 4 , 5 , 6 , 7 } { 8 , 9 } { 11 } payment 61/200 13/20 473/1000 9/25 3/40 (57)
For ( x , y , z ) = ( 2 , 1 , 1 ) (x,y,z)=(2,1,1) ( x , y , z ) = ( 2 , 1 , 1 ) the distinct cyclic word is − − − − − + + − − + − − + -\ -\ -\ -\ -+\ +\ -\ -+\ -\ -+ − − − − − + + − − + − − + . Its placements are
support map source charge payment [ 3 , 10 ] 1 + j 10201020 { 4 , 5 , 6 } 61 / 200 ( 6 , 5 , … , − 1 ) 1 − j 20210212 { 2 , 3 } 13 / 20 ( 2 , 1 , … , − 4 ) 2 − j 1020212 { 1 , 0 , 12 , 11 } 473 / 1000 ( 0 , − 1 , … , − 7 ) − j 01020212 { 10 , 9 } 9 / 25 ( 0 , − 1 , … , − 9 ) − j 0102021202 { 7 } 3 / 40 (58) \begin{array}{c|c|c|c|c}
\text{support} & \text{map} & \text{source} & \text{charge} & \text{payment}\\
[3,10] & 1+j & 10201020 & \{4,5,6\} & 61/200\\
(6,5,\ldots,-1) & 1-j & 20210212 & \{2,3\} & 13/20\\
(2,1,\ldots,-4) & 2-j & 1020212 & \{1,0,12,11\} & 473/1000\\
(0,-1,\ldots,-7) & -j & 01020212 & \{10,9\} & 9/25\\
(0,-1,\ldots,-9) & -j & 0102021202 & \{7\} & 3/40 \tag*{(58)}
\end{array} support [ 3 , 10 ] ( 6 , 5 , … , − 1 ) ( 2 , 1 , … , − 4 ) ( 0 , − 1 , … , − 7 ) ( 0 , − 1 , … , − 9 ) map 1 + j 1 − j 2 − j − j − j source 10201020 20210212 1020212 01020212 0102021202 charge { 4 , 5 , 6 } { 2 , 3 } { 1 , 0 , 12 , 11 } { 10 , 9 } { 7 } payment 61/200 13/20 473/1000 9/25 3/40 (58)
In both tables the eight-source for 61 / 200 61/200 61/200 keeps its internal seven-arc sign witness and the full nonpositive-tail transfer of Corollary B.76. The ten-source keeps its returned four-core, internal seven-arc sign and circle witness, and uncharged final cap, as required by Lemma C.8. The 9 / 25 9/25 9/25 row charges only its own pair; its two anchors remain uncharged by that row. Each table has five pairwise disjoint physical charged sets. Therefore both give
C > 61 200 + 13 20 + 473 1000 + 9 25 + 3 40 = 1863 1000 > C ∗ . C>\frac{61}{200}+\frac{13}{20}+\frac{473}{1000}+\frac{9}{25}+\frac{3}{40}=\frac{1863}{1000}>C_*. C > 200 61 + 20 13 + 1000 473 + 25 9 + 40 3 = 1000 1863 > C ∗ .
All six triples have been excluded. □
Corollary 19.3. A mixed minimizing cycle with three positive runs has B = 10 B=10 B = 10 , negative lengths ( 2 , 2 , 2 ) (2,2,2) ( 2 , 2 , 2 ) and positive lengths ( 1 , 1 , 2 ) (1,1,2) ( 1 , 1 , 2 ) in cyclic order. Its cyclic increment word is − − + − − + − − + + -\ -+\ -\ -+\ -\ -++ − − + − − + − − + + .
Proof. Combine Propositions 19.1 and 19.2 . □
One and two positive runs
Theorem 18.1 and Lemma 18.7 reduce the mixed case to at most three positive runs, with
1 ≤ a i ≤ 8 , b i ∈ { 2 , 5 , 7 , 8 , 9 } , # { i : b i > 2 } ≤ 1. 1\le a_i\le8,\qquad b_i\in\{2,5,7,8,9\},\qquad\#\{i:b_i>2\}\le1. 1 ≤ a i ≤ 8 , b i ∈ { 2 , 5 , 7 , 8 , 9 } , # { i : b i > 2 } ≤ 1.
The three-run case was settled in Corollary 19.3. We now determine the one-run and two-run alternatives directly from the twist and compatible local payments. No upper bound on the period is assumed in this argument.
The payments used at a separator
We retain the physical own arcs and ports of § 3.33 . Put
P = 180483 500000 , t 0 = 379 1000 , C ∗ = 5 P . P = \frac{180483}{500000}, \qquad t_0 = \frac{379}{1000}, \qquad C_* = 5P. P = 500000 180483 , t 0 = 1000 379 , C ∗ = 5 P .
The following placements explain the combinations used below. Their complete-source qualifications are those of § 3.36 . Number the own arcs of a positive run of length a a a as 0 , … , a 0,\ldots,a 0 , … , a . For a ≥ 3 a \ge3 a ≥ 3 , the ordinary terminal payments charge
{ − 1 , 0 , 1 } , { a − 1 , a , a + 1 } , \{-1,0,1\}, \qquad\{a-1,a,a+1\}, { − 1 , 0 , 1 } , { a − 1 , a , a + 1 } ,
respectively, above 61 / 200 61/200 61/200 . Their complete six-sources are ( 3 , 2 , … , − 2 ) (3,2,\ldots,-2) ( 3 , 2 , … , − 2 ) , with whole label map j ↦ − j j \mapsto-j j ↦ − j , and [ a − 3 , a + 2 ] [a-3,a+2] [ a − 3 , a + 2 ] , with whole phase 3 − a 3-a 3 − a . Set the control label at own arc zero to zero; both sources then become 012021. If the facing negative run has length two, the actual return extends the corresponding source to seven arcs, giving 0120212 and improving the same charge to more than t 0 t_0 t 0 by Theorem 8.7. These are alternative payments on the same arcs, rather than additional charges.
When B > 9 B > 9 B > 9 , a negative run of length two whose neighboring positive runs both have length at least three admits the complete nine-source 012021201, which pays weakly 2 P 2P 2 P . It charges the terminal own pair of the first run, the shared separator port, and the initial own pair of the second. Indeed, if the first run ends at own arc t t t , the source is [ t − 3 , t + 5 ] [t-3,t+5] [ t − 3 , t + 5 ] , phased so that the label at t t t is zero, and its charge is [ t − 1 , t + 3 ] [t-1,t+3] [ t − 1 , t + 3 ] . Within a positive run of length a ≥ 5 a \ge5 a ≥ 5 , the pair { 2 , 3 } \{2,3\} { 2 , 3 } has the complete source [ 0 , 5 ] = 012012 [0,5] = 012012 [ 0 , 5 ] = 012012 and pays more than P P P . More generally the pairs
{ 2 + 2 r , 3 + 2 r } , 0 ≤ r < ⌊ a − 3 2 ⌋ , \{2+2r,3+2r\}, \qquad0 \le r < \left\lfloor\frac{a-3}{2}\right\rfloor, { 2 + 2 r , 3 + 2 r } , 0 ≤ r < ⌊ 2 a − 3 ⌋ ,
are disjoint from both terminal own pairs and from one another.
A positive run of length two pays more than 49 / 200 49/200 49/200 on its three own arcs, with zero port loads: its complete five-source [ − 1 , 3 ] [-1,3] [ − 1 , 3 ] , phased by + 2 +2 + 2 , is 02010. A negative run of length five pays more than 13 / 20 13/20 13/20 on its two middle interior arcs, also with zero port loads. With its starting arc numbered zero, its complete eight-source is [ − 1 , 6 ] = 20210212 [-1,6] = 20210212 [ − 1 , 6 ] = 20210212 , after setting the initial control label to zero. These two consequences are Theorems B.20 and E.2.
We also use the half-port profiles of Lemmas C.13 and C.14. If a positive run has exactly one adjacent negative run of length two, its payment is at least
473 2000 ; 233 500 ( a ≥ 3 ) ; 17 50 + P ⌊ a − 3 2 ⌋ ( a ≥ 5 ) . \frac{473}{2000}; \qquad\frac{233}{500}\quad(a \ge3); \qquad\frac{17}{50}+P\left\lfloor\frac{a-3}{2}\right\rfloor\quad(a \ge5). 2000 473 ; 500 233 ( a ≥ 3 ) ; 50 17 + P ⌊ 2 a − 3 ⌋ ( a ≥ 5 ) .
At a = 3 a=3 a = 3 the value is at least 109 / 200 109/200 109/200 . The first bound for a run of length two follows from its stronger zero-port 49 / 200 49/200 49/200 payment. The negative half-port values needed here are
V ( 5 ) = 3 5 , V ( 7 ) = 9 10 , V ( 8 ) = V ( 9 ) = 6 5 . V(5)=\frac{3}{5}, \qquad V(7)=\frac{9}{10}, \qquad V(8)=V(9)=\frac{6}{5}. V ( 5 ) = 5 3 , V ( 7 ) = 10 9 , V ( 8 ) = V ( 9 ) = 5 6 .
At each separator these profiles have total physical load at most one. When full terminal payments are used instead, their charged sets will be kept disjoint explicitly. All exclusions below have B ≥ 12 B \ge12 B ≥ 12 ; all their complete sources have at most nine arcs and are therefore proper. In particular, these payments impose no new promise on the short candidates of periods eight and nine.
Proposition 20.1 (One positive run). If an original global minimizer has B ≥ 8 B \ge8 B ≥ 8 and exactly one positive run, its run pair is one of
( 6 , 2 ) , ( 3 , 5 ) , ( 6 , 5 ) , ( 2 , 7 ) , ( 3 , 8 ) , ( 1 , 9 ) . (6,2), \qquad(3,5), \qquad(6,5), \qquad(2,7), \qquad(3,8), \qquad(1,9). ( 6 , 2 ) , ( 3 , 5 ) , ( 6 , 5 ) , ( 2 , 7 ) , ( 3 , 8 ) , ( 1 , 9 ) .
Proof. Write the pair as ( a , b ) (a,b) ( a , b ) . The bound a ≤ 8 a \le8 a ≤ 8 and the twist a − b ≡ 1 ( m o d 3 ) a-b \equiv1 \pmod{3} a − b ≡ 1 ( mod 3 ) leave the following possibilities:
Table 9.
Here the condition a + b ≥ 8 a+b \ge8 a + b ≥ 8 has already removed the shorter pairs. The half-port profiles exclude ( 8 , 7 ) (8,7) ( 8 , 7 ) , ( 6 , 8 ) (6,8) ( 6 , 8 ) , ( 7 , 9 ) (7,9) ( 7 , 9 ) , respectively, by
61 200 + 2 P + 9 10 > C ∗ , \frac{61}{200}+2P+\frac{9}{10}>C_*, 200 61 + 2 P + 10 9 > C ∗ ,
61 200 + P + 6 5 > C ∗ , \frac{61}{200}+P+\frac{6}{5}>C_*, 200 61 + P + 5 6 > C ∗ ,
61 200 + 2 P + 6 5 > C ∗ . \frac{61}{200}+2P+\frac{6}{5}>C_*. 200 61 + 2 P + 5 6 > C ∗ .
For ( a , b ) = ( 5 , 7 ) (a,b)=(5,7) ( a , b ) = ( 5 , 7 ) , the complete eight-source 10120121 charges the six positive own arcs above 609/1000 by Theorem B.32 . Its support is [ − 1 , 6 ] [-1,6] [ − 1 , 6 ] , with own arc zero labeled zero. The six negative interior arcs split into the disjoint triples [ 1 , 3 ] [1,3] [ 1 , 3 ] and [ 4 , 6 ] [4,6] [ 4 , 6 ] , each retaining its complete weak five-source. They pay 6 / 5 6/5 6/5 and are disjoint from the positive own arcs. Thus
C > 609 1000 + 6 5 = 1809 1000 > C ∗ . C>\frac{609}{1000}+\frac{6}{5}=\frac{1809}{1000}>C_*. C > 1000 609 + 5 6 = 1000 1809 > C ∗ .
For ( a , b ) = ( 4 , 9 ) (a,b)=(4,9) ( a , b ) = ( 4 , 9 ) , the positive run uses its two ordinary terminal triples and its middle opposite-center payment 63/500. The latter charges own arc 2, on the complete source [ − 2 , 6 ] [-2,6] [ − 2 , 6 ] , sent to 102012021 by whole phase + 2 +2 + 2 , of Theorem C.6. Number the negative interior arcs 1 , … , 8 1,\ldots,8 1 , … , 8 . Two complete weak five-sources charge [ 2 , 4 ] [2,4] [ 2 , 4 ] and [ 5 , 7 ] [5,7] [ 5 , 7 ] , leaving both ports 1, 8 free. All five charged sets are disjoint, so
C > 2 61 200 + 63 500 + 6 5 = 242 125 > C ∗ . C>2\frac{61}{200}+\frac{63}{500}+\frac{6}{5}=\frac{242}{125}>C_*. C > 2 200 61 + 500 63 + 5 6 = 125 242 > C ∗ .
These exclusions leave precisely the six stated necessary alternatives. □
Proposition 20.2 (Two positive runs). If an original global minimizer has B ≥ 8 B \ge8 B ≥ 8 and exactly two positive runs, there are thirteen necessary run configurations. When both negative lengths are two, the unordered positive pair is
{ a 1 , a 2 } = { 1 , 4 } , { 2 , 3 } , { 2 , 6 } , { 4 , 4 } . \{a_1,a_2\}=\{1,4\},\{2,3\},\{2,6\},\{4,4\}. { a 1 , a 2 } = { 1 , 4 } , { 2 , 3 } , { 2 , 6 } , { 4 , 4 } . Otherwise write the cyclic increments as + x − b + y − 2 +x\ -b\ +y-2 + x − b + y − 2 , where b > 2 b>2 b > 2 is the distinguished negative length. The ordered pair ( x , y ) (x,y) ( x , y ) is one of
Table 10.
In particular b = 9 b=9 b = 9 is impossible.
Proof. Suppose first that both negative lengths are two. If both positive lengths are at least three and one is at least five, use the fused nine-source at each separator and an internal forward pair in the longer positive run. The two fusions use opposite terminal own pairs in each run, and the internal pair avoids them. Their disjoint charges give C > 2 P + 2 P + P = C ∗ C>2P+2P+P=C_* C > 2 P + 2 P + P = C ∗ . Thus if both lengths are at least three, they lie in { 3 , 4 } \{3,4\} { 3 , 4 } ; the twist a 1 + a 2 ≡ 2 ( m o d 3 ) a_1+a_2\equiv2\pmod{3} a 1 + a 2 ≡ 2 ( mod 3 ) leaves only ( 4 , 4 ) (4,4) ( 4 , 4 ) .
If one positive length is one, the other is 1, 4 or 7. The first choice has B = 6 B=6 B = 6 . For the choice seven, Lemma 12.1 pays more than 3/7 on the singleton’s own arcs. The longer run uses its two returned terminal triples and the internal pairs { 2 , 3 } \{2, 3\} { 2 , 3 } , { 4 , 5 } \{4, 5\} { 4 , 5 } , giving
C > 3 7 + 2 t 0 + 2 P > C ∗ . C > \frac{3}{7} + 2t_0 + 2P > C_*. C > 7 3 + 2 t 0 + 2 P > C ∗ .
The two neighboring positive occurrences in the singleton lemma are here the same physical run; only its two distinct density owners are used. The longer run is charged once. Finally, if one length is two and the other is at least two, the twist and a i ≤ 8 a_i \le8 a i ≤ 8 leave the other length three or six. This proves the four short-negative alternatives.
Now let b > 2 b > 2 b > 2 be the unique long negative length. Each positive run has exactly one short-negative incidence. For b = 5 b = 5 b = 5 , if both positive lengths are at least three, use the zero-port negative middle pair, the fused nine-source at the short separator, and an ordinary terminal triple at each end facing the long separator. These four charges are disjoint and give
C > 13 20 + 2 P + 2 61 200 > C ∗ . C > \frac{13}{20} + 2P + 2\frac{61}{200} > C_*. C > 20 13 + 2 P + 2 200 61 > C ∗ .
One positive length is therefore at most two. If it is one, the twist x + y ≡ 2 ( m o d 3 ) x+y \equiv2 \pmod{3} x + y ≡ 2 ( mod 3 ) leaves the other length 1, 4 or 7. At seven the compatible half-port profiles give
C > 3 5 + 473 2000 + 17 50 + 2 P > C ∗ . C > \frac{3}{5} + \frac{473}{2000} + \frac{17}{50} + 2P > C_*. C > 5 3 + 2000 473 + 50 17 + 2 P > C ∗ .
If the shorter length is two and neither length is one, the other is three or six. At six, combine the shorter run’s zero-port 49 / 200 49/200 49/200 payment, both terminal triples of the longer run, one of its internal forward pairs, and the zero-port negative middle pair. The ordinary triple faces the long separator, while the returned triple faces the short one. Their ports and own pairs are disjoint from the other charges, so
C > 49 200 + 61 200 + t 0 + P + 13 20 > C ∗ . C > \frac{49}{200} + \frac{61}{200} + t_0 + P + \frac{13}{20} > C_*. C > 200 49 + 200 61 + t 0 + P + 20 13 > C ∗ .
The five displayed ordered pairs for b = 5 b = 5 b = 5 remain.
For b = 7 b = 7 b = 7 , any positive length at least five gives the compatible half-port bound
C > 17 50 + P + 473 2000 + 9 10 > C ∗ . C > \frac{17}{50} + P + \frac{473}{2000} + \frac{9}{10} > C_*. C > 50 17 + P + 2000 473 + 10 9 > C ∗ .
Thus x , y ≤ 4 x,y \le4 x , y ≤ 4 . The twist x + y ≡ 1 ( m o d 3 ) x+y \equiv1 \pmod{3} x + y ≡ 1 ( mod 3 ) leaves totals four and seven. Total seven consists of lengths three and four, whose half-port payments give
C > 109 200 + 233 500 + 9 10 > C ∗ . C > \frac{109}{200} + \frac{233}{500} + \frac{9}{10} > C_*. C > 200 109 + 500 233 + 10 9 > C ∗ .
Total four gives exactly (1, 3), (2, 2), (3, 1).
For b = 8 b = 8 b = 8 or 9, a positive length at least three would give
C > 233 500 + 473 2000 + 6 5 > C ∗ . C > \frac{233}{500} + \frac{473}{2000} + \frac{6}{5} > C_*. C > 500 233 + 2000 473 + 5 6 > C ∗ .
Hence x , y ≤ 2 x,y \le2 x , y ≤ 2 . At b = 8 b = 8 b = 8 the twist leaves only (1, 1). At b = 9 b = 9 b = 9 it leaves (1, 2), (2, 1). For either of the latter configurations, the negative run admits weak five-source charges [ref-2 , ref-4 ] and [ref-5 , ref-7 ], of total 6 / 5 6/5 6/5 , avoiding both ports. The singleton positive run uses its complete one-return four-arc payment above 473 / 1000 473/1000 473/1000 , while the length-two run pays more than 49 / 200 49/200 49/200 on its own arcs. Their charged sets are disjoint: the shared short-separator port is used only by the singleton. Therefore
C > 6 5 + 473 1000 + 49 200 = 959 500 > C ∗ . C > \frac{6}{5} + \frac{473}{1000} + \frac{49}{200} = \frac{959}{500} > C_*. C > 5 6 + 1000 473 + 200 49 = 500 959 > C ∗ .
The singleton payment is Corollary B.17 ; its full seven-source retains the actual short return.
All cases are now exhausted. When both negative lengths are two, cyclic rotation exchanges the two positive lengths. When b > 2 b > 2 b > 2 , its distinguished position fixes the ordered pair ( x , y ) (x,y) ( x , y ) ; the proof covers both orders and does not identify them by reflection.
The twenty-one necessary types
Theorem 20.3 (The run classification). Every original global minimizer with B ≥ 8 B \ge8 B ≥ 8 has one of the twenty-one essential increment types in Table 4 : six types with one positive run, thirteen with two, one with three, and the all-minus eight-arc type. In particular every original global minimizer has B ≤ 13 B \le13 B ≤ 13 , and any prescribed fundamental cut has at most N ≤ 14 N \le14 N ≤ 14 positive pieces.
Proof. Theorem 18.1 puts every mixed type into M = 1 , 2 , 3 M = 1, 2, 3 M = 1 , 2 , 3 . Propositions 20.1 and 20.2 give the first nineteen types, while Corollary 19.3 leaves the single cyclic pattern ( a 1 , a 2 , a 3 ) = ( 1 , 1 , 2 ) (a_1,a_2,a_3) = (1,1,2) ( a 1 , a 2 , a 3 ) = ( 1 , 1 , 2 ) with all b i = 2 b_i = 2 b i = 2 . Proposition 18.8 leaves only the all-minus word with B = 8 B = 8 B = 8 in the pure case. These configurations translate directly into Table 4 ; the maximal listed length is thirteen. The types with B ≤ 7 B \le7 B ≤ 7 already satisfy this bound. A prescribed cut splits at most one physical positive arc, so N ≤ B + 1 ≤ 14 N \le B + 1 \le14 N ≤ B + 1 ≤ 14 .
The same twenty-one strings were obtained by the finite capacity argument in Theorem 17.2 ; the present proof establishes them directly for all lengths.
The table records necessary alternatives, without asserting their minimizing realization. Its derivation uses the local physical-cost inequalities and elementary run cases; some of those local inequalities have computer-assisted proofs. The later exclusions use further geometric information within these run families. Part III combines the three eight-arc exclusions while retaining their different charts and predicates.
Cycles with all negative runs of length two
The analytic classification in Theorem 20.3 leaves six words whose negative runs all have length two: R03, R05, R06, R07, R15 and R17. Their positive lengths are respectively (6), (1, 4), (2, 3), (1, 1, 2), (2, 6) and (4, 4), with only cyclic rotation identified.
The cost of a short positive run is controlled by Section 12. The last three types require more than a fixed lower payment on each separate run: R07 keeps correlated means, R15 uses the conditional central fee, and R17 retains entrance terms already contained in the disjoint physical costs.
Theorem 21.1 (The short-negative family). An original global minimizer with B ≥ 8 B \ge8 B ≥ 8 and a mixed increment cycle whose negative runs all have length two has one of the three necessary types R03, R05 and R06.
The proof occupies the remainder of this section. The complete finite comparisons retain their closed boundaries and original arithmetic contracts. They are applied after the source, common-lift and physical cost reductions, not to independently assigned formal parameters.
21.1. Full-lift matching across the true seam. We apply Proposition B.114 and Corollary B.115 on the following actual supports. A complete capped binary core with two distinct internal binary ties and its two third-mode cap ties excludes abnormality and identifies the entire normalized lift by equality-only rigidity. This includes the matrix and both state covectors. On the overlap of two genuine periods, G A G − 1 GAG^{-1} G A G − 1 is constant; it first gives the matrix twist. At an actual switch between distinct modes, the difference of the state covectors annihilates two independent control fields. It therefore vanishes in both coordinates. Adjoint uniqueness then extends the full lift. Matrix twist alone would not prove this step.
For R07, the actual support [ 0 , 5 ] = 0 ∣ 212 ∣ 0 [0,5] = 0|212|0 [ 0 , 5 ] = 0∣212∣0 , phased by + 1 +1 + 1 , is 1 ∣ 0202 ∣ 1 1|0202|1 1∣0202∣1 . Cuts inside arcs − 5 -5 − 5 and 0 retain its closed core [ τ 1 , τ 5 ] [\tau_1,\tau_5] [ τ 1 , τ 5 ] and positive caps on their overlap, and the distinct switch at τ 0 \tau_0 τ 0 lies between the cuts. A second source [6, 10] = 1 ∣ 020 ∣ 1 =1|020|1 = 1∣020∣1 has closed core [ τ 7 , τ 10 ] [\tau_7,\tau_{10}] [ τ 7 , τ 10 ] disjoint from the first on the period circle. Hence every genuine cut retains at least one complete closed core, which supplies the stronger compatibility of all required source lifts. For R17, use [ 4 , 8 ] + 1 = 1 ∣ 202 ∣ 1 [4,8]+1=1|202|1 [ 4 , 8 ] + 1 = 1∣202∣1 , with core [ τ 5 , τ 8 ] [\tau_5,\tau_8] [ τ 5 , τ 8 ] , cuts inside − 4 -4 − 4 and 4 4 4 , and switch τ 4 \tau_4 τ 4 . The second source [ 10 , 14 ] + 2 = 1 ∣ 202 ∣ 1 [10,14]+2=1|202|1 [ 10 , 14 ] + 2 = 1∣202∣1 has disjoint closed core [ τ 11 , τ 14 ] [\tau_{11},\tau_{14}] [ τ 11 , τ 14 ] . The same argument identifies all required source lifts, including both state covectors, before any cross-seam mean, signed entrance or determinant is compared.
The positive lengths ( 1 , 1 , 2 ) (1,1,2) ( 1 , 1 , 2 ) : four exhaustive ranges
R07 has native word 0212101020 and own reciprocals
( w , p , q L , r L , s , r R , q R , v , b , z ) . (w,p,q_L,r_L,s,r_R,q_R,v,b,z). ( w , p , q L , r L , s , r R , q R , v , b , z ) .
Use the common full lift of Section 21.1. Put x = s 2 x=s^2 x = s 2 , H = ( 1 + x ) / 2 H=(1+x)/2 H = ( 1 + x ) /2 and t L = s + r L t_L=s+r_L t L = s + r L , t R = s + r R t_R=s+r_R t R = s + r R . The positive shared arc gives t L , t R > 2 s t_L,t_R>2s t L , t R > 2 s , t L t R < 2 H t_Lt_R<2H t L t R < 2 H and x < 1 / 3 x<1/3 x < 1/3 . Both actual mean factors are K i = ( 3 + μ i ) / 2 K_i=(3+\mu_i)/2 K i = ( 3 + μ i ) /2 .
21.2.0.1. The low singleton range s ≤ 1 / 5 s\le1/5 s ≤ 1/5 . The two actual binary mean integrals, extended to [ − 1 , 0 ] [-1,0] [ − 1 , 0 ] or [ 0 , 1 ] [0,1] [ 0 , 1 ] according to the sign of their middle root, give μ i < 3 − 4 log 2 \mu_i<3-4\log2 μ i < 3 − 4 log 2 . Indeed the extended weighted mean is ( 3 / 2 − 2 log 2 ) / ( 1 / 2 ) = 3 − 4 log 2 (3/2-2\log2)/(1/2)=3-4\log2 ( 3/2 − 2 log 2 ) / ( 1/2 ) = 3 − 4 log 2 , whereas the removed outer portion has strictly positive mean excess. Thus K i < K ‾ : = 3 − 2 log 2 < 13 / 8 K_i<\overline{K}:=3-2\log2<13/8 K i < K := 3 − 2 log 2 < 13/8 . Define the positive frame deficits
L = ( p + s ) 2 2 − ( r L − p ) ( q L − s ) , U = ( v + s ) 2 2 − ( r R − v ) ( q R − s ) . L=\frac{(p+s)^2}{2}-(r_L-p)(q_L-s),\qquad U=\frac{(v+s)^2}{2}-(r_R-v)(q_R-s). L = 2 ( p + s ) 2 − ( r L − p ) ( q L − s ) , U = 2 ( v + s ) 2 − ( r R − v ) ( q R − s ) .
Put A = ( 1 + p 2 ) / 2 A=(1+p^2)/2 A = ( 1 + p 2 ) /2 , B = ( 1 + v 2 ) / 2 B=(1+v^2)/2 B = ( 1 + v 2 ) /2 and Δ = H 2 − L U > 0 \Delta=H^2-LU>0 Δ = H 2 − LU > 0 . The outer cap denominators and the inner cap coefficients are
Y L = ( 1 + p 2 ) p 2 ( p + w ) , Y R = ( 1 + v 2 ) v 2 ( v + b ) , Y_L=\frac{(1+p^2)p}{2(p+w)},\qquad Y_R=\frac{(1+v^2)v}{2(v+b)}, Y L = 2 ( p + w ) ( 1 + p 2 ) p , Y R = 2 ( v + b ) ( 1 + v 2 ) v ,
Z L = Y L − L , Z R = Y R − U , B L = s H t R , B R = s H t L . Z_L=Y_L-L,\qquad Z_R=Y_R-U,\qquad B_L=\frac{sH}{t_R},\qquad B_R=\frac{sH}{t_L}. Z L = Y L − L , Z R = Y R − U , B L = t R sH , B R = t L sH .
All four actual outward cap values are strictly positive before squaring. They use the same unsquared constant λ \lambda λ , whose sign is not prescribed. With d i = 1 / 2 − H 2 K o t h e r > 0 d_i=1/2-H^2K_{\mathrm{other}}>0 d i = 1/2 − H 2 K other > 0 , the two cap equalities for owner i i i read
K i Z i 2 − 2 λ Z i = d i + E i , o u t , K i B i 2 + 2 λ B i = d i + E i , i n , E i , o u t , E i , i n > 0. K_iZ_i^2-2\lambda Z_i=d_i+E_{i,\mathrm{out}},\qquad K_iB_i^2+2\lambda B_i=d_i+E_{i,\mathrm{in}},\qquad E_{i,\mathrm{out}},E_{i,\mathrm{in}}>0. K i Z i 2 − 2 λ Z i = d i + E i , out , K i B i 2 + 2 λ B i = d i + E i , in , E i , out , E i , in > 0.
Here H ≤ 13 / 25 H\le13/25 H ≤ 13/25 and H 2 K i < 1 / 2 H^2K_i<1/2 H 2 K i < 1/2 . If Z i ≤ 0 Z_i\le0 Z i ≤ 0 , positivity of the outward cap value P i , o u t = K i Z i − λ P_{i,\mathrm{out}}=K_iZ_i-\lambda P i , out = K i Z i − λ gives 0 < P i , o u t ≤ − λ 0<P_{i,\mathrm{out}}\le-\lambda 0 < P i , out ≤ − λ ; the same determinant identity then forces H 2 K o t h e r > 1 / 2 H^2K_{\mathrm{other}}>1/2 H 2 K other > 1/2 , a contradiction. Hence Z i > 0 Z_i>0 Z i > 0 . Multiply the two displayed rows by B i , Z i B_i,Z_i B i , Z i and add. The shared constant cancels, giving K i Z i B i > d i K_iZ_iB_i>d_i K i Z i B i > d i . The full frame gives
( p + w ) Z L = − ( A − L ) ( z − p ) 2 − A 2 H ( H − B ) Δ ( z − p ) , (p+w)Z_L=-\frac{(A-L)(z-p)}{2}-\frac{A^2H(H-B)}{\Delta(z-p)}, ( p + w ) Z L = − 2 ( A − L ) ( z − p ) − Δ ( z − p ) A 2 H ( H − B ) ,
and its right counterpart. Here A − L A-L A − L , B − U > 0 B-U>0 B − U > 0 follow from shared positive time. Since Z L , Z R > 0 Z_L,Z_R>0 Z L , Z R > 0 , these rows force v > s v>s v > s , p > s p>s p > s . Positivity of the two exterior numerators further gives U ( A − L ) > H ( B − H ) U(A-L)>H(B-H) U ( A − L ) > H ( B − H ) and L ( B − U ) > H ( A − H ) L(B-U)>H(A-H) L ( B − U ) > H ( A − H ) . If either L − ( A − H ) L-(A-H) L − ( A − H ) or U − ( B − H ) U-(B-H) U − ( B − H ) were nonpositive, equality would contradict the two strict rows immediately; in the negative case both would be negative and their product would give H 2 < L U H^2<LU H 2 < LU , contrary to Δ > 0 \Delta>0 Δ > 0 . Hence L > ( p 2 − x ) / 2 L>(p^2-x)/2 L > ( p 2 − x ) /2 , U > ( v 2 − x ) / 2 U>(v^2-x)/2 U > ( v 2 − x ) /2 . The cap-product floors consequently give
Z L > c 0 ( v + s ) , Z R > c 0 ( p + s ) , c 0 = 1 / ( 2 K ‾ ) − H 2 s H . Z_L>c_0(v+s),\qquad Z_R>c_0(p+s),\qquad c_0=\frac{1/(2\overline{K})-H^2}{sH}. Z L > c 0 ( v + s ) , Z R > c 0 ( p + s ) , c 0 = sH 1/ ( 2 K ) − H 2 .
The last positive native arc gives ( z + w ) ( z + b ) > 1 + z 2 (z+w)(z+b)>1+z^2 ( z + w ) ( z + b ) > 1 + z 2 . The same frame has ( z + w ) 2 = ( p + w ) 2 + 2 A 2 U / Δ (z+w)^2=(p+w)^2+2A^2U/\Delta ( z + w ) 2 = ( p + w ) 2 + 2 A 2 U /Δ and ( z + b ) 2 = ( v + b ) 2 + 2 B 2 L / Δ (z+b)^2=(v+b)^2+2B^2L/\Delta ( z + b ) 2 = ( v + b ) 2 + 2 B 2 L /Δ . Substituting these identities and cancelling only positive factors gives the exact product inequality
F L F R < H 2 , F L = 2 Y L 2 p 2 − L , F R = 2 Y R 2 v 2 − U . F_LF_R < H^2,\qquad F_L=\frac{2Y_L^2}{p^2}-L,\qquad F_R=\frac{2Y_R^2}{v^2}-U. F L F R < H 2 , F L = p 2 2 Y L 2 − L , F R = v 2 2 Y R 2 − U .
For c 0 ≥ κ ∈ { 1 / 2 , 1 / 3 } c_0\geq\kappa\in\{1/2,1/3\} c 0 ≥ κ ∈ { 1/2 , 1/3 } the preceding deficit floors and an increasing quadratic give
F L > 2 κ ( v + s ) − x 2 + [ 2 κ ( v + s ) − x ] 2 2 p 2 , F_L>2\kappa(v+s)-\frac{x}{2}+\frac{[2\kappa(v+s)-x]^2}{2p^2}, F L > 2 κ ( v + s ) − 2 x + 2 p 2 [ 2 κ ( v + s ) − x ] 2 ,
and the symmetric right inequality. The endpoint rational comparisons 4 / 13 − 1591 / 5184 > 0 4/13-1591/5184>0 4/13 − 1591/5184 > 0 and 4 / 13 − 572 / 1875 > 0 4/13-572/1875>0 4/13 − 572/1875 > 0 give c 0 > 1 / 2 c_0>1/2 c 0 > 1/2 for s ≤ 1 / 6 s\leq1/6 s ≤ 1/6 and c 0 > 1 / 3 c_0>1/3 c 0 > 1/3 for s ≤ 1 / 5 s\leq1/5 s ≤ 1/5 . In the first range, the two factors are positive and their product exceeds
( v + v 2 2 p 2 ) ( p + p 2 2 v 2 ) ≥ ( p + 1 2 ) ( v + 1 2 ) > H 2 . \left(v+\frac{v^2}{2p^2}\right)\left(p+\frac{p^2}{2v^2}\right)\geq\left(p+\frac12\right)\left(v+\frac12\right)>H^2. ( v + 2 p 2 v 2 ) ( p + 2 v 2 p 2 ) ≥ ( p + 2 1 ) ( v + 2 1 ) > H 2 .
In the second range put d = s ( 2 / 3 − s ) ≥ 1 / 12 > 1 / 16 d=s(2/3-s)\geq1/12>1/16 d = s ( 2/3 − s ) ≥ 1/12 > 1/16 . Cauchy’s inequality and t + d / t ≥ 2 d t+d/t\geq2\sqrt{d} t + d / t ≥ 2 d , t = p v t=\sqrt{pv} t = p v , give F L F R > 2 / 9 + ( 4 / 3 ) d > 5 / 9 > 13 / 25 ≥ H \sqrt{F_LF_R}>2/9+(4/3)\sqrt{d}>5/9>13/25\geq H F L F R > 2/9 + ( 4/3 ) d > 5/9 > 13/25 ≥ H . Both conclusions contradict the last-arc product. This excludes the entire low range, without an anchor-order assumption.
21.2.0.2. The high singleton range x ≥ 1 / 6 x\geq1/6 x ≥ 1/6 . Keep the original unsquared shared law
λ = H 2 s { ( 2 − K L ) t L + ( 2 − K R ) t R − 2 x ( 1 t L + 1 t R ) } + H log 2 H t L t R − 1 2 . \lambda=\frac{H}{2s}\left\{(2-K_L)t_L+(2-K_R)t_R-2x\left(\frac{1}{t_L}+\frac{1}{t_R}\right)\right\}+H\log\frac{2H}{t_Lt_R}-\frac12. λ = 2 s H { ( 2 − K L ) t L + ( 2 − K R ) t R − 2 x ( t L 1 + t R 1 ) } + H log t L t R 2 H − 2 1 .
The actual root bounds first give λ > − 3 x / 2 \lambda>-3x/2 λ > − 3 x /2 . The determinant H 2 K L K R − λ 2 H^2K_LK_R-\lambda^2 H 2 K L K R − λ 2 increases in each mean factor throughout the rectangle [ 3 / 2 , K L ] × [ 3 / 2 , K R ] [3/2,K_L]\times[3/2,K_R] [ 3/2 , K L ] × [ 3/2 , K R ] : for example its first derivative is greater than ( 3 H / 2 ) ( H − s t L ) > 0 (3H/2)(H-st_L)>0 ( 3 H /2 ) ( H − s t L ) > 0 . At the zero-mean comparison, the shared law gives a negative value λ 0 \lambda_0 λ 0 , with
λ 00 : = H log 1 + x 4 x − 1 2 < λ 0 < 1 − 6 x − 15 x 2 16 x < 0. \lambda_{00}:=H\log\frac{1+x}{4x}-\frac12<\lambda_0<\frac{1-6x-15x^2}{16x}<0. λ 00 := H log 4 x 1 + x − 2 1 < λ 0 < 16 x 1 − 6 x − 15 x 2 < 0.
Only after these signs are proved may one square, obtaining Γ > 9 H 2 / 4 − λ 00 2 \Gamma>9H^2/4-\lambda_{00}^2 Γ > 9 H 2 /4 − λ 00 2 . Set z 0 = ( 1 − 3 x ) / ( 1 + 5 x ) ∈ ( 0 , 3 / 11 ] z_0=(1-3x)/(1+5x)\in(0,3/11] z 0 = ( 1 − 3 x ) / ( 1 + 5 x ) ∈ ( 0 , 3/11 ] and ℓ = 2 z 0 + 2 z 0 3 / 3 < log ( ( 1 + x ) / ( 4 x ) ) \ell=2z_0+2z_0^3/3<\log((1+x)/(4x)) ℓ = 2 z 0 + 2 z 0 3 /3 < log (( 1 + x ) / ( 4 x )) . Then Γ > 3 / 4 − D 2 ( z 0 ) \Gamma>3/4-D_2(z_0) Γ > 3/4 − D 2 ( z 0 ) , where
D 2 ( z ) = 8 z 3 + 16 z 4 + ( 44 / 3 ) z 5 + ( 112 / 9 ) z 6 + ( 32 / 9 ) z 7 + ( 16 / 9 ) z 8 ( 3 + 5 z ) 2 . D_2(z)=\frac{8z^3+16z^4+(44/3)z^5+(112/9)z^6+(32/9)z^7+(16/9)z^8}{(3+5z)^2}. D 2 ( z ) = ( 3 + 5 z ) 2 8 z 3 + 16 z 4 + ( 44/3 ) z 5 + ( 112/9 ) z 6 + ( 32/9 ) z 7 + ( 16/9 ) z 8 .
Every z n / ( 3 + 5 z ) 2 z^n/(3+5z)^2 z n / ( 3 + 5 z ) 2 , n ≥ 3 n\geq3 n ≥ 3 , increases. The endpoint gives
Γ > 83376585 113379904 > 1470471 2000000 > H ∗ , \Gamma>\frac{83376585}{113379904}>\frac{1470471}{2000000}>H_*, Γ > 113379904 83376585 > 2000000 1470471 > H ∗ ,
contrary to harmonic comparison on this actual lift.
21.2.0.3. The double-lower range p , v ≤ s p,v\leq s p , v ≤ s . For a lower-anchor side let R = r i / s R=r_i/s R = r i / s , T = 1 + R T=1+R T = 1 + R , e = ( R − 1 ) / ( R + 1 ) e=(R-1)/(R+1) e = ( R − 1 ) / ( R + 1 ) and
P i = ( 1 − μ i ) T / 2 − 2 / T , F i = P i − 2 log T . P_i=(1-\mu_i)T/2-2/T,\qquad F_i=P_i-2\log T. P i = ( 1 − μ i ) T /2 − 2/ T , F i = P i − 2 log T .
If s + r i ≤ 1 + x s+r_i\leq\sqrt{1+x} s + r i ≤ 1 + x , substitution in its true outer-cap slope gives F i − F o t h e r > 2 B i ∗ F_i-F_{\mathrm{other}}>2B_i^* F i − F other > 2 B i ∗ , where
B i ∗ = e 2 ( 1 + 3 e + e 2 + e 3 ) 6 ( 1 − e ) > 0. B_i^*=\frac{e^2(1+3e+e^2+e^3)}{6(1-e)}>0. B i ∗ = 6 ( 1 − e ) e 2 ( 1 + 3 e + e 2 + e 3 ) > 0.
This follows by replacing the formal mean by its upper bound on that same side; its coefficient in the slope is negative. The two unsquared correction terms cancel when both sides are short, giving a contradiction. Both cannot be long because t L t R < 1 + x t_Lt_R<1+x t L t R < 1 + x . Thus precisely one is short and one is long. Name their normalized widths T < V T<V T < V and keep e e e from the short side. The same slope comparison, retaining the long-side mean coefficient, gives
V < e 3 + 2 e 2 + 5 e + 12 2 ( 3 − 2 e ) , e 3 + e 2 + 3 e − 1 < 0. V<\frac{e^3+2e^2+5e+12}{2(3-2e)},\qquad e^3+e^2+3e-1<0. V < 2 ( 3 − 2 e ) e 3 + 2 e 2 + 5 e + 12 , e 3 + e 2 + 3 e − 1 < 0.
whence, using x < 1 / 6 x < 1/6 x < 1/6 ,
50 / 19 < T < V < 125 / 44 , 1936 / 13689 < x < 1 / 6. 50/19 < T < V < 125/44,\qquad1936/13689 < x < 1/6. 50/19 < T < V < 125/44 , 1936/13689 < x < 1/6.
The full-frame positive deficits give p / s , v / s > 3 / 11 p/s,v/s > 3/11 p / s , v / s > 3/11 . The remaining candidate is therefore contained in [ 3 / 11 , 1 ] 2 × [ 3 / 2 , 2 ] 2 × [ 1 / 8 , 1 / 6 ] [3/11,1]^2 \times[3/2,2]^2 \times[1/8,1/6] [ 3/11 , 1 ] 2 × [ 3/2 , 2 ] 2 × [ 1/8 , 1/6 ] in the coordinates ( p / s , v / s , r L / s , r R / s , x ) (p/s,v/s,r_L/s,r_R/s,x) ( p / s , v / s , r L / s , r R / s , x ) , after naming the short side first. Its exact finite witness has 32965 nodes and 16483 terminal boxes. The evaluator uses unbounded integers with 128 fractional bits, sign-certified mean roots and a positive logarithm lower series. Closed children retain the common split face. Every leaf contradicts a strict collar, positive duration, full-frame or actual cap-slope condition, and the terminal leaves cover the entire root.
The remaining mixed and upper ranges
Put
a = p / s , c = v / s , R = r L / s , S = r R / s , Q = q L / s , V = q R / s . a=p/s,\qquad c=v/s,\qquad R=r_L/s,\qquad S=r_R/s,\qquad Q=q_L/s,\qquad V=q_R/s. a = p / s , c = v / s , R = r L / s , S = r R / s , Q = q L / s , V = q R / s .
The two proper returned eight-sources give p 2 , s 2 , v 2 < 47 / 250 p^2,s^2,v^2 < 47/250 p 2 , s 2 , v 2 < 47/250 . The low and high exclusions already give 1 / 25 < x < 1 / 6 1/25 < x < 1/6 1/25 < x < 1/6 . The full frame gives a , c > 3 / 13 a,c > 3/13 a , c > 3/13 : using its positive deficits,
x ( 1 + a ) 2 ( 1 + c ) 1 + x > 1 − c , ( s + p ) 2 1 + x < 5 / 8 , \frac{x(1+a)^2(1+c)}{1+x} > 1-c,\qquad\frac{(s+p)^2}{1+x} < 5/8, 1 + x x ( 1 + a ) 2 ( 1 + c ) > 1 − c , 1 + x ( s + p ) 2 < 5/8 ,
where the latter has the rational numerator reserve $29/2000. The actual calibrated root satisfies Q ≥ min ( a , 1 ) R Q \ge\min(a,1)R Q ≥ min ( a , 1 ) R . For R ≥ 6 R \ge6 R ≥ 6 this contradicts the left positive deficit, since
( 6 − a ) ( 6 a − 1 ) > ( 1 + a ) 2 / 2 ( 3 / 13 ≤ a < 1 ) , (6-a)(6a-1) > (1+a)^2/2\qquad(3/13 \le a < 1), ( 6 − a ) ( 6 a − 1 ) > ( 1 + a ) 2 /2 ( 3/13 ≤ a < 1 ) ,
and 5 ( 6 − a ) > ( 1 + a ) 2 / 2 5(6-a) > (1+a)^2/2 5 ( 6 − a ) > ( 1 + a ) 2 /2 for 1 ≤ a ≤ 11 / 5 1 \le a \le11/5 1 ≤ a ≤ 11/5 . The upper density bounds also give a , c < 11 / 5 a,c < 11/5 a , c < 11/5 . Hence every remaining candidate lies in the closed outward enlargement
( a , c ) ∈ [ 3 / 13 , 11 / 5 ] 2 , ( R , S ) ∈ [ 1 , 6 ] 2 , x ∈ [ 1 / 25 , 1 / 6 ] . (59) (a,c) \in[3/13,11/5]^2,\qquad(R,S) \in[1,6]^2,\qquad x \in[1/25,1/6]. \tag*{(59)} ( a , c ) ∈ [ 3/13 , 11/5 ] 2 , ( R , S ) ∈ [ 1 , 6 ] 2 , x ∈ [ 1/25 , 1/6 ] . (59)
The root enclosure used here is analytic. Define the smooth divided mean
M ( d ; b , h ) = ∫ 0 1 ( d − b − ( h − b ) t d + b + ( h − b ) t ) 2 d t , F ( q ; a , R ) = M ( q ; a , R ) − M ( R ; 1 , q ) . M(d;b,h)=\int_0^1\left(\frac{d-b-(h-b)t}{d+b+(h-b)t}\right)^2\,dt,\qquad F(q;a,R)=M(q;a,R)-M(R;1,q). M ( d ; b , h ) = ∫ 0 1 ( d + b + ( h − b ) t d − b − ( h − b ) t ) 2 d t , F ( q ; a , R ) = M ( q ; a , R ) − M ( R ; 1 , q ) .
At a zero on the full physical ray q > max ( 1 , a ) q>\max(1,a) q > max ( 1 , a ) , let μ \mu μ be the common mean, f a = ( ( q − a ) / ( q + a ) ) 2 f_a=((q-a)/(q+a))^2 f a = (( q − a ) / ( q + a ) ) 2 and f R = ( ( q − R ) / ( q + R ) ) 2 f_R=((q-R)/(q+R))^2 f R = (( q − R ) / ( q + R ) ) 2 . If q < R q<R q < R , the second integrand is decreasing, so μ > f R \mu>f_R μ > f R ; the first has no interior maximum, so f a > μ f_a>\mu f a > μ . If q > R q>R q > R , the first is decreasing and gives f a > μ > f R f_a>\mu>f_R f a > μ > f R directly. The endpoint q = R q=R q = R is a root only when a = 1 a=1 a = 1 . Homogeneity gives
q M d ( q ; a , R ) = R ( μ − f R ) + a ( f a − μ ) R − a > 0 , ∂ q M ( R ; 1 , q ) = f R − μ q − 1 < 0. qM_d(q;a,R)=\frac{R(\mu-f_R)+a(f_a-\mu)}{R-a}>0,\qquad\partial_qM(R;1,q)=\frac{f_R-\mu}{q-1}<0. q M d ( q ; a , R ) = R − a R ( μ − f R ) + a ( f a − μ ) > 0 , ∂ q M ( R ; 1 , q ) = q − 1 f R − μ < 0.
Thus every zero is an upward crossing, so there is at most one. For a < 1 a<1 a < 1 , it lies in ( 1 , R ) (1,R) ( 1 , R ) when F ( 1 ) < 0 F(1)<0 F ( 1 ) < 0 ; otherwise use the clipped value 1 1 1 . For a > 1 a>1 a > 1 , the endpoint signs give the unique physical root in ( R , a R ) (R,aR) ( R , a R ) ; the preceding argument also excludes roots below or above that bracket. At a = 1 a=1 a = 1 it is R R R . For completeness, at a nonendpoint root
F a = μ − f a R − a < 0 , F R = f R − μ R − a − M d ( R ; 1 , q ) < 0. F_a=\frac{\mu-f_a}{R-a}<0,\qquad F_R=\frac{f_R-\mu}{R-a}-M_d(R;1,q)<0. F a = R − a μ − f a < 0 , F R = R − a f R − μ − M d ( R ; 1 , q ) < 0.
The second homogeneity identity gives M d ( R ; 1 , q ) > 0 M_d(R;1,q)>0 M d ( R ; 1 , q ) > 0 by the same endpoint comparison. Thus implicit differentiation makes the root increasing in both a , R a,R a , R ; clipping preserves this order. The point-mean limit R = a > 1 R=a>1 R = a > 1 and the corner ( 1 , 1 ) (1,1) ( 1 , 1 ) are retained. Consequently any actual root in a parameter box obeys
q ^ ( a − , max ( a − , R − ) ) ≤ q a c t u a l ≤ q ^ ( min ( a + , R + ) , R + ) . \widehat q(a_-,\max(a_-,R_-))\le q_{\mathrm{actual}}\le\widehat q(\min(a_+,R_+),R_+). q ( a − , max ( a − , R − )) ≤ q actual ≤ q ( min ( a + , R + ) , R + ) .
Bisection changes a root endpoint only at a certified strict sign; an uncertain sign retains the whole current bracket.
Lemma 21.2 (R07 closed necessary-system cover). No actual original minimizing R07 candidate lies in (21.1) after the double-lower exclusion.
Computer-assisted proof. We specify the correlated rows which carry the finite contradiction.
Put
l = ( 1 + a ) 2 / 2 − ( R − a ) ( Q − 1 ) , u = ( 1 + c ) 2 / 2 − ( S − c ) ( V − 1 ) , L = x l , U = x u . l=(1+a)^2/2-(R-a)(Q-1),\qquad u=(1+c)^2/2-(S-c)(V-1),\qquad L=xl,\qquad U=xu. l = ( 1 + a ) 2 /2 − ( R − a ) ( Q − 1 ) , u = ( 1 + c ) 2 /2 − ( S − c ) ( V − 1 ) , L = x l , U = xu .
A = ( 1 + a 2 x ) / 2 , B = ( 1 + c 2 x ) / 2 , K L = ( 3 + μ L ) / 2 , K R = ( 3 + μ R ) / 2 , A=(1+a^2x)/2,\qquad B=(1+c^2x)/2,\qquad K_L=(3+\mu_L)/2,\qquad K_R=(3+\mu_R)/2, A = ( 1 + a 2 x ) /2 , B = ( 1 + c 2 x ) /2 , K L = ( 3 + μ L ) /2 , K R = ( 3 + μ R ) /2 ,
with the two calibrated means. The shared law above computes the same unsquared λ \lambda λ . Set D 0 = H 2 K L K R D_0=H^2K_LK_R D 0 = H 2 K L K R , k = K L L k=K_LL k = K L L , j = K R U j=K_RU j = K R U , b ( y ) = 3 ( 1 + y ) / 4 b(y)=3(1+y)/4 b ( y ) = 3 ( 1 + y ) /4 and c ( y ) = 9 y ( 1 − y ) / 8 − 3 / 4 c(y)=9y(1-y)/8-3/4 c ( y ) = 9 y ( 1 − y ) /8 − 3/4 . The determinant is
Γ = D 0 − λ 2 < 1471 / 2000. \Gamma=D_0-\lambda^2<1471/2000. Γ = D 0 − λ 2 < 1471/2000.
The three actual defects are λ + 3 x / 2 \lambda+3x/2 λ + 3 x /2 , − k − λ + 3 a 2 x / 2 -k-\lambda+3a^2x/2 − k − λ + 3 a 2 x /2 and − j − λ + 3 c 2 x / 2 -j-\lambda+3c^2x/2 − j − λ + 3 c 2 x /2 . At their respective densities y y y they are strictly greater than
ϕ ( y ) = 29 / 1000 [ 3 ( 1 − 3 y ) / 4 ] 2 + 29 / 500 + 3 ( 1 − 3 y ) / 4 . \phi(y)=\frac{29/1000}{\sqrt{[3(1-3y)/4]^2+29/500+3(1-3y)/4}}. ϕ ( y ) = [ 3 ( 1 − 3 y ) /4 ] 2 + 29/500 + 3 ( 1 − 3 y ) /4 29/1000 .
Retain these strict floors and the closed actual-determinant rows
b ( x ) λ + D 0 + c ( x ) ≥ 0 , b(x)\lambda+D_0+c(x)\ge0, b ( x ) λ + D 0 + c ( x ) ≥ 0 ,
[ 2 k − b ( a 2 x ) ] λ + D 0 + k 2 − b ( a 2 x ) k + c ( a 2 x ) ≥ 0 , [2k-b(a^2x)]\lambda+D_0+k^2-b(a^2x)k+c(a^2x)\ge0, [ 2 k − b ( a 2 x )] λ + D 0 + k 2 − b ( a 2 x ) k + c ( a 2 x ) ≥ 0 ,
[ 2 j − b ( c 2 x ) ] λ + D 0 + j 2 − b ( c 2 x ) j + c ( c 2 x ) ≥ 0. [2j-b(c^2x)]\lambda+D_0+j^2-b(c^2x)j+c(c^2x)\ge0. [ 2 j − b ( c 2 x )] λ + D 0 + j 2 − b ( c 2 x ) j + c ( c 2 x ) ≥ 0.
These affine rows are never divided by a coefficient of unknown sign. The actual cap slopes are strictly negative; for the left slope the formula is
η + ( η 2 − μ L ) ( 1 − a 2 x ) 2 ( 1 − η ) − ( 3 + μ L ) x { a ( a + Q ) / 2 − l } + 2 λ , η = ( Q − a ) / ( Q + a ) . \eta+\frac{(\eta^2-\mu_L)(1-a^2x)}{2(1-\eta)}-(3+\mu_L)x\{a(a+Q)/2-l\}+2\lambda,\qquad\eta=(Q-a)/(Q+a). η + 2 ( 1 − η ) ( η 2 − μ L ) ( 1 − a 2 x ) − ( 3 + μ L ) x { a ( a + Q ) /2 − l } + 2 λ , η = ( Q − a ) / ( Q + a ) .
The right formula exchanges ( a , R , Q , l , μ L ) (a,R,Q,l,\mu_L) ( a , R , Q , l , μ L ) with ( c , S , V , u , μ R ) (c,S,V,u,\mu_R) ( c , S , V , u , μ R ) .
The full signed frame determines the exterior reciprocals:
Δ = H 2 − L U > 0 , J = A B H / Δ , Z = z / s = 2 J − 1 / x , \Delta=H^2-LU>0,\qquad J=ABH/\Delta,\qquad Z=z/s=\sqrt{2J-1}/x, Δ = H 2 − LU > 0 , J = A B H /Δ , Z = z / s = 2 J − 1 / x ,
W = w / s = A 2 u / Δ − ( J − A ) / x Z − a , B o = b / s = B 2 l / Δ − ( J − B ) / x Z − c . W=w/s=\frac{A^2u/\Delta-(J-A)/x}{Z-a},\qquad B_o=b/s=\frac{B^2l/\Delta-(J-B)/x}{Z-c}. W = w / s = Z − a A 2 u /Δ − ( J − A ) / x , B o = b / s = Z − c B 2 l /Δ − ( J − B ) / x .
Their actual THREE-core means are μ w = M ( W ; a , Z ) \mu_w=M(W;a,Z) μ w = M ( W ; a , Z ) , μ b = M ( B o ; c , Z ) \mu_b=M(B_o;c,Z) μ b = M ( B o ; c , Z ) and satisfy
μ w ≥ ( Z − W Z + W ) 2 , μ b ≥ ( Z − B o Z + B o ) 2 , W > a , B o > c . \mu_w\ge\left(\frac{Z-W}{Z+W}\right)^2,\qquad\mu_b\ge\left(\frac{Z-B_o}{Z+B_o}\right)^2,\qquad W>a,\qquad B_o>c. μ w ≥ ( Z + W Z − W ) 2 , μ b ≥ ( Z + B o Z − B o ) 2 , W > a , B o > c .
Only one endpoint at each THREE-core is strict. The same matrix also requires
A 2 ( 3 + μ w ) / 2 − Δ K R − L ( k + j + 2 λ ) = 0 , A^2(3+\mu_w)/2-\Delta K_R-L(k+j+2\lambda)=0, A 2 ( 3 + μ w ) /2 − Δ K R − L ( k + j + 2 λ ) = 0 ,
and its right counterpart. The positive cap durations, neighboring products and final native duration are retained as well.
The finite witness has 6994079 nodes and 3497040 leaves. A GMP integer evaluator uses 96 fractional bits on the exact 48-bit partition. For the divided mean set η = ( 2 d − b − h ) / ( 2 d + b + h ) \eta=(2d-b-h)/(2d+b+h) η = ( 2 d − b − h ) / ( 2 d + b + h ) and ρ = ( h − b ) / ( 2 d + b + h ) \rho=(h-b)/(2d+b+h) ρ = ( h − b ) / ( 2 d + b + h ) . Its even-moment coefficient is η 2 + 4 n η / ( 2 n + 1 ) + ( 2 n − 1 ) / ( 2 n + 1 ) \eta^2+4n\eta/(2n+1)+(2n-1)/(2n+1) η 2 + 4 n η / ( 2 n + 1 ) + ( 2 n − 1 ) / ( 2 n + 1 ) , of absolute value at most 4; after 24 terms the tail is bounded by 4 ρ 50 / ( 1 − ρ 2 ) 4\rho^{50}/(1-\rho^2) 4 ρ 50 / ( 1 − ρ 2 ) . The shared logarithm uses 32 positive atanh terms with its geometric tail. All divisions, radicals and signs are outward enclosed. The verifier reconstructs both closed children at every split and recomputes the assigned terminal reason at every leaf; syntax and stack exhaustion certify that no box is lost. A strict defect or slope condition is rejected by the corresponding non-strict endpoint test. A closed affine row is rejected only when its entire interval is strictly negative. Unknown domains cannot accept a leaf. This proves the necessary-system contradiction on the whole root. ∎
The low, high, double-lower and mixed/upper ranges exhaust all actual candidates. Their boundaries have been retained throughout, so R07 is excluded.
The positive lengths (2, 6): the conditional central fee
Proposition 21.3. R 15 R15 R 15 cannot be the essential type of an original global minimizer.
Proof. Its native word is 021202120120. The complete proper six-source [ 0 , 5 ] + 1 = 102010 [0,5]+1=102010 [ 0 , 5 ] + 1 = 102010 retains the central [ 1 , 5 ] + 1 = 02010 [1,5]+1=02010 [ 1 , 5 ] + 1 = 02010 and the capped binary normality witness [ 0 , 4 ] [0,4] [ 0 , 4 ] . Select its full nonzero lift before restriction. The forward source [ 1 , 7 ] + 1 [1,7]+1 [ 1 , 7 ] + 1 and the whole reflected source ( 5 , 4 , … , − 1 ) (5,4,\ldots,-1) ( 5 , 4 , … , − 1 ) with map j ↦ 2 − j j\mapsto2-j j ↦ 2 − j are 0201020. They give both actual unsorted H150 crossings. The returned sources [ 9 , 15 ] + 2 [9,15]+2 [ 9 , 15 ] + 2 and ( 9 , 8 , … , 3 ) (9,8,\ldots,3) ( 9 , 8 , … , 3 ) with map j ↦ 1 − j j\mapsto1-j j ↦ 1 − j are 0120212. They give the terminal circles for the central cap reciprocals p = r 1 p=r_1 p = r 1 and q = r 5 q=r_5 q = r 5 . The six-sources [ 10 , 15 ] [10,15] [ 10 , 15 ] and [ 2 , 7 ] + 1 [2,7]+1 [ 2 , 7 ] + 1 are 201020 with physical owners 1 and 5, respectively; hence p 2 , q 2 < 1 / 6 p^2,q^2<1/6 p 2 , q 2 < 1/6 . All these supports are proper, and their state conclusions give precisely (11.1). Theorem 11.1 therefore gives
F = c 2 + c 3 + c 4 > C 0 = 80483 125000 . (60) F=c_2+c_3+c_4>C_0=\frac{80483}{125000}. \tag*{(60)} F = c 2 + c 3 + c 4 > C 0 = 125000 80483 . (60) The two returned triples just used cost more than 2 / 5 2/5 2/5 on { 11 , 0 , 1 } \{11,0,1\} { 11 , 0 , 1 } and { 5 , 6 , 7 } \{5,6,7\} { 5 , 6 , 7 } . The forward source [ 6 , 11 ] + 2 = 012012 [6,11]+2=012012 [ 6 , 11 ] + 2 = 012012 gives c 8 + c 9 > P c_8+c_9>P c 8 + c 9 > P , and c 10 > 0 c_{10}>0 c 10 > 0 . These charges and F F F are disjoint. Thus
C > C 0 + 2 2 5 + P = 180483 100000 = C ∗ , C>C_0+2\frac{2}{5}+P=\frac{180483}{100000}=C_*, C > C 0 + 2 5 2 + P = 100000 180483 = C ∗ , contradicting the strict original budget.
The positive lengths (4, 4): retained entrance terms
The native word is 021201210120, with shared owners 1, 7 and opposite centers 4, 10. Section 21.1 identifies the entire full lift across its true twist. Define the physical unions
U 1 = c 11 + c 0 + c 1 + c 2 + c 3 , U 7 = c 5 + c 6 + c 7 + c 8 + c 9 . U_1=c_{11}+c_0+c_1+c_2+c_3,\qquad U_7=c_5+c_6+c_7+c_8+c_9. U 1 = c 11 + c 0 + c 1 + c 2 + c 3 , U 7 = c 5 + c 6 + c 7 + c 8 + c 9 .
There is the exact partition
C = U 1 + U 7 + c 4 + c 10 . (61) C=U_1+U_7+c_4+c_{10}. \tag*{(61)} C = U 1 + U 7 + c 4 + c 10 . (61)
The proper H3 nine-sources inside the actual eleven-sources give U 1 , U 7 ≥ U 0 = 180483 / 250000 U_1,U_7\ge U_0=180483/250000 U 1 , U 7 ≥ U 0 = 180483/250000 . The ordinary complete opposite centers give c 4 , c 10 > 63 / 500 c_4,c_{10}>63/500 c 4 , c 10 > 63/500 . Therefore each center separately satisfies
63 500 < c i < 117483 500000 ( i = 4 , 10 ) . (62) \frac{63}{500}<c_i<\frac{117483}{500000}\qquad(i=4,10). \tag*{(62)} 500 63 < c i < 500000 117483 ( i = 4 , 10 ) . (62)
For each upper bound, the other center contributes its independent 63 / 500 63/500 63/500 floor; the two centers need not have equal costs.
21.4.0.1. Exclusive costs and the four actual means. The actual pair { 11 , 0 } \{11,0\} { 11 , 0 } carries mean μ 0 \mu_0 μ 0 and entrance λ 10 R \lambda_{10}R λ 10 R ; { 2 , 3 } \{2,3\} { 2 , 3 } carries μ 2 , λ 4 L \mu_2,\lambda_{4L} μ 2 , λ 4 L ; { 5 , 6 } \{5,6\} { 5 , 6 } carries μ 6 , λ 4 R \mu_6,\lambda_{4R} μ 6 , λ 4 R ; and { 8 , 9 } \{8,9\} { 8 , 9 } carries μ 8 , λ 10 L \mu_8,\lambda_{10L} μ 8 , λ 10 L . All four entrance slopes are strictly positive by the complete signed-strip source theorem. The exact original exclusive integral contains K Σ = L e x i t + λ K\Sigma=L_{\mathrm{exit}}+\lambda K Σ = L exit + λ , where K = ( 3 + μ ) / 2 K=(3+\mu)/2 K = ( 3 + μ ) /2 . Retaining this equality, instead of dropping its entrance term, gives the same physical exclusive pair
E > α 2 + 1 − p 2 + 3 x 2 p − 3 B 3 + μ + 3 λ 3 + μ . E>\frac{\alpha}{2}+\frac{1-p}{2}+\frac{3x}{2p}-\frac{3B}{3+\mu}+\frac{3\lambda}{3+\mu}. E > 2 α + 2 1 − p + 2 p 3 x − 3 + μ 3 B + 3 + μ 3 λ .
The strictness comes from the actual positive shear-square integral; α ≥ 0 \alpha\ge0 α ≥ 0 is the original source entrance width. Adding the two exclusive pairs and the shared physical arc gives the original J J J cancellation with both entrance terms retained. Hence set
G 4 = c 4 + 3 λ 4 L 3 + μ 2 + 3 λ 4 R 3 + μ 6 , G_4=c_4+\frac{3\lambda_{4L}}{3+\mu_2}+\frac{3\lambda_{4R}}{3+\mu_6}, G 4 = c 4 + 3 + μ 2 3 λ 4 L + 3 + μ 6 3 λ 4 R ,
G 10 = c 10 + 3 λ 10 L 3 + μ 8 + 3 λ 10 R 3 + μ 0 . G_{10}=c_{10}+\frac{3\lambda_{10L}}{3+\mu_8}+\frac{3\lambda_{10R}}{3+\mu_0}. G 10 = c 10 + 3 + μ 8 3 λ 10 L + 3 + μ 0 3 λ 10 R .
Equation (61) and the exact cost algebra imply
C > J 1 + J 7 + G 4 + G 10 . (63) C>J_1+J_7+G_4+G_{10}. \tag*{(63)} C > J 1 + J 7 + G 4 + G 10 . (63)
The G i G_i G i are regroupings of terms already present in disjoint physical costs. They are not new fees, nor bounds for the center arc c i c_i c i alone. The discarded four α / 2 \alpha/2 α /2 terms are nonnegative.
21.4.0.2. The outer bound. The actual normal capped sources at owners 1 and 7 imply x < 1 / 6 x<1/6 x < 1/6 . Their normality is established on the required full lift; the density conclusion is not an extra payment. For either outer profile the complete retained-cost expression is J J J . The scalar argument below excludes x ≤ 1 / 10 x\le1/10 x ≤ 1/10 , where J > 511 / 880 > 14 / 25 J>511/880>14/25 J > 511/880 > 14/25 . The remaining closed root is
x ∈ [ 1 / 10 , 1 / 6 ] , p , q ∈ [ 4 / 11 , 1 ] . x\in[1/10,1/6],\qquad p,q\in[4/11,1]. x ∈ [ 1/10 , 1/6 ] , p , q ∈ [ 4/11 , 1 ] .
For clarity, the exact function to be bounded is the returned-union quantity in (B.155). Put A 0 = 1 + x A_0=1+x A 0 = 1 + x and, for its two actual means m , n m,n m , n ,
0 < m < ( 1 − p ) 2 , 0 < n < ( 1 − q ) 2 , p q > 4 x / A 0 , 0<m<(1-p)^2,\qquad0<n<(1-q)^2,\qquad pq>4x/A_0, 0 < m < ( 1 − p ) 2 , 0 < n < ( 1 − q ) 2 , pq > 4 x / A 0 ,
V = log A 0 4 x + g ( p ) + g ( q ) , g ( t ) = log t + ( t − 1 − t ) / 2 , M = m / p + n / q , \mathcal{V}=\log\frac{A_0}{4x}+g(p)+g(q),\qquad g(t)=\log t+(t^{-1}-t)/2,\qquad M=m/p+n/q, V = log 4 x A 0 + g ( p ) + g ( q ) , g ( t ) = log t + ( t − 1 − t ) /2 , M = m / p + n / q ,
B = ( 1 − A 0 V ) / 2 + A 0 M / 4 , δ = m 3 + m + n 3 + n , \mathcal{B}=(1-A_0\mathcal{V})/2+A_0M/4,\qquad\delta=\frac{m}{3+m}+\frac{n}{3+n}, B = ( 1 − A 0 V ) /2 + A 0 M /4 , δ = 3 + m m + 3 + n n ,
Γ = A 0 2 ( 3 + m ) ( 3 + n ) 16 − B 2 , J = V − A 0 M / 2 + B δ = 1 A 0 − x M 2 − B ( 2 / A 0 − δ ) . \Gamma=\frac{A_0^2(3+m)(3+n)}{16}-\mathcal{B}^2,\qquad J=\mathcal{V}-A_0M/2+\mathcal{B}\delta=\frac{1}{A_0}-\frac{xM}{2}-\mathcal{B}(2/A_0-\delta). Γ = 16 A 0 2 ( 3 + m ) ( 3 + n ) − B 2 , J = V − A 0 M /2 + B δ = A 0 1 − 2 x M − B ( 2/ A 0 − δ ) .
Here B \mathcal{B} B is the same actual transport constant retained in each exclusive cost. The mean rectangle gives M ≤ ( 1 − p q ) 2 / ( p q ) M\le(1-pq)^2/(pq) M ≤ ( 1 − pq ) 2 / ( pq ) and therefore 1 / A 0 − x M / 2 ≥ 93 / 112 > 14 / 25 1/A_0-xM/2\ge93/112>14/25 1/ A 0 − x M /2 ≥ 93/112 > 14/25 . Thus J ≤ 14 / 25 J\le14/25 J ≤ 14/25 forces B > 0 \mathcal{B}>0 B > 0 and J > V − A 0 M / 2 J>\mathcal{V}-A_0M/2 J > V − A 0 M /2 . For f x ( t ) = log t + A 0 − x / ( 2 t ) − ( 2 + x ) t / 2 f_x(t)=\log t+A_0-x/(2t)-(2+x)t/2 f x ( t ) = log t + A 0 − x / ( 2 t ) − ( 2 + x ) t /2 ,
f x ′ ( t ) = ( 1 − t ) ( ( 2 + x ) t + x ) 2 t 2 ≥ 0 , f'_x(t)=\frac{(1-t)((2+x)t+x)}{2t^2}\ge0, f x ′ ( t ) = 2 t 2 ( 1 − t ) (( 2 + x ) t + x ) ≥ 0 ,
f x ( p ) + f x ( q ) − f x ( p q ) = ( 1 − p ) ( 1 − q ) ( x 2 p q + 2 + x 2 ) ≥ 0. f_x(p)+f_x(q)-f_x(pq)=(1-p)(1-q)\left(\frac{x}{2pq}+\frac{2+x}{2}\right)\ge0. f x ( p ) + f x ( q ) − f x ( pq ) = ( 1 − p ) ( 1 − q ) ( 2 pq x + 2 2 + x ) ≥ 0.
It follows that J > ( 7 − 18 x − 9 x 2 ) / [ 8 ( 1 + x ) ] J>(7-18x-9x^2)/[8(1+x)] J > ( 7 − 18 x − 9 x 2 ) / [ 8 ( 1 + x )] , proving the stated exclusion of x ≤ 1 / 10 x\le1/10 x ≤ 1/10 . Also p , q ≥ 4 / 11 p,q\ge4/11 p , q ≥ 4/11 on the remaining root.
For the actual two-mean rectangle consider
L = J + 60 x ( Γ − H ∗ ) . \mathcal{L}=J+60x(\Gamma-H_*). L = J + 60 x ( Γ − H ∗ ) .
Its Hessian is strictly negative definite throughout that rectangle. Put
K = − diag ( p , q ) ∇ 2 L diag ( p , q ) . K=-\operatorname{diag}(p,q)\nabla^2\mathcal{L}\operatorname{diag}(p,q). K = − diag ( p , q ) ∇ 2 L diag ( p , q ) .
Direct differentiation gives K 12 ≥ 1867 / 7200 > 0 K_{12}\ge1867/7200>0 K 12 ≥ 1867/7200 > 0 and K 11 − K 12 , K 22 − K 12 ≥ 119 / 3600 > 0 K_{11}-K_{12},K_{22}-K_{12}\ge119/3600>0 K 11 − K 12 , K 22 − K 12 ≥ 119/3600 > 0 . The bound A 0 V ≤ 21 / 16 A_0\mathcal{V}\le21/16 A 0 V ≤ 21/16 gives B ≥ − 5 / 32 \mathcal{B}\ge-5/32 B ≥ − 5/32 on the entire comparison rectangle; its positivity at the actual counterexample is not imposed at its corners. The minimum of L \mathcal{L} L is therefore bounded below by the minimum at the four corners. Only the actual determinant is calibrated. A four-corner lower bound L > 14 / 25 \mathcal{L}>14/25 L > 14/25 contradicts J ≤ 14 / 25 J\le14/25 J ≤ 14/25 and Γ < H ∗ \Gamma<H_* Γ < H ∗ at the actual point.
The directed-integer cover uses 50 fractional bits and 56 logarithm-series terms. Its finite tree has 165373 nodes and 82687 leaves, with all four corners checked at each accepted leaf and no unresolved box. It proves
J 1 , J 7 > 14 / 25. (64) J_1,J_7 > 14/25. \tag*{(64)} J 1 , J 7 > 14/25. (64)
Indeed the four-corner inequality gives L a c t u a l > 14 / 25 \mathcal{L}_{\mathrm{actual}} > 14/25 L actual > 14/25 , whereas Γ a c t u a l < H ∗ \Gamma_{\mathrm{actual}} < H_* Γ actual < H ∗ gives L a c t u a l < J a c t u a l ≤ 14 / 25 \mathcal{L}_{\mathrm{actual}} < J_{\mathrm{actual}} \le14/25 L actual < J actual ≤ 14/25 .
21.4.0.3. The center bound and its compact root. The center supports are [ 0 , 8 ] + 1 [0,8]+1 [ 0 , 8 ] + 1 and [ 6 , 14 ] + 2 [6,14]+2 [ 6 , 14 ] + 2 , both 102012021. In the mode-1 coordinates of Section B.3.7, let a > b > 0 a>b>0 a > b > 0 be the actual endpoint reciprocals, y y y the central own reciprocal and q = V 2 q=V^2 q = V 2 its crossing height squared. Put
y = q − 1 2 q , h = q + 1 2 q , r = a + y h , s = b + y h , ρ = y ( a − b ) , θ = r − 1 r − s . y=\frac{q-1}{2\sqrt{q}},\qquad h=\frac{q+1}{2\sqrt{q}},\qquad r=\frac{a+y}{h},\qquad s=\frac{b+y}{h},\qquad\rho=y(a-b),\qquad\theta=\frac{r-1}{r-s}. y = 2 q q − 1 , h = 2 q q + 1 , r = h a + y , s = h b + y , ρ = y ( a − b ) , θ = r − s r − 1 .
The complete forward contexts give r > 1 > s > 0 r>1>s>0 r > 1 > s > 0 and both strict signed entrance slopes. The full matrix is reconstructed from H ( a ) − 1 N ( y ( a − b ) / 2 ) H ( b ) H(a)^{-1}N(y(a-b)/2)H(b) H ( a ) − 1 N ( y ( a − b ) /2 ) H ( b ) and the actual state-adjoint cost row ((B.136)). The endpoint cap equalities ((B.132)) and maximizing rows ((B.133)) remain affine in the two actual returned means.
Original compactness gives q ≤ 243 / 4 q\le243/4 q ≤ 243/4 at this actual crossing. For q ≤ 5 / 2 q\le5/2 q ≤ 5/2 its mandatory cost exceeds 39 / 160 39/160 39/160 , contradicting ((62) ). Whole reflection preserves the physical cost, determinant and G G G , exchanges the two means and slopes, and sends ρ \rho ρ to ρ / ( r s ) \rho/(rs) ρ / ( r s ) ; it does not preserve ρ \rho ρ . Choose the reflected orientation so that r s ≥ 1 rs\ge1 r s ≥ 1 . The exact physical cost in these coordinates is
c = ρ { 1 + 1 r s − q − 1 q + 1 log ( r / s ) r − s } . c=\rho\left\{1+\frac{1}{rs}-\frac{q-1}{q+1}\frac{\log(r/s)}{r-s}\right\}. c = ρ { 1 + r s 1 − q + 1 q − 1 r − s log ( r / s ) } .
Its positive density gives c > ρ c>\rho c > ρ , and r s ≥ 1 rs\ge1 r s ≥ 1 gives c < 2 ρ c<2\rho c < 2 ρ . The strict positive center floor supplies the closed comparison root
q ∈ [ 5 / 2 , 243 / 4 ] , ρ ∈ [ 63 / 1000 , 117483 / 500000 ] , θ ∈ [ 1 / 2 , 1 ] , ( μ L , μ R ) ∈ [ 0 , 1 ] 2 . (65) q\in[5/2,243/4],\qquad\rho\in[63/1000,117483/500000],\qquad\theta\in[1/2,1],\qquad(\mu_L,\mu_R)\in[0,1]^2. \tag*{(65)} q ∈ [ 5/2 , 243/4 ] , ρ ∈ [ 63/1000 , 117483/500000 ] , θ ∈ [ 1/2 , 1 ] , ( μ L , μ R ) ∈ [ 0 , 1 ] 2 . (65)
There is no arbitrary small-arc cutoff. The denominators, signs and logarithm arguments are checked on the larger rectangle ρ ∈ [ 1 / 16 , 6 / 25 ] \rho\in[1/16,6/25] ρ ∈ [ 1/16 , 6/25 ] , which includes midpoint comparison segments. The explicit transport and affine-cost solution in Section B.3.7 have positive state and cap denominators on this rectangle; in that chart u C + 1 > 0 u_C+1>0 u C + 1 > 0 , u D − 1 < 0 u_D-1<0 u D − 1 < 0 and β C > 0 > β D \beta_C>0>\beta_D β C > 0 > β D . Thus the relaxed boxes contain no hidden pole.
For a fixed geometric box, enclose coefficients from above in the six affine necessary cap, switching and slope rows. Because the means are nonnegative, these rows define a single polygon containing every actual mean pair in that box. Intersections of every pair of nonparallel supporting lines are outward enclosed. A candidate intersection is discarded only by a strict violation; exact parallelism is recognized by exact mantissa products, while ambiguous determinants force subdivision. This includes zero-dimensional and line-segment polygons through their supporting-line intersections.
Let h L , h R h_L,h_R h L , h R be the maximal coordinates of this fixed containing polygon and put α = 3 / ( 3 + h L ) \alpha=3/(3+h_L) α = 3/ ( 3 + h L ) , β = 3 / ( 3 + h R ) \beta=3/(3+h_R) β = 3/ ( 3 + h R ) . At the actual point, positivity of the actual slopes gives
G ≥ W : = c + α λ L + β λ R . G\ge W:=c+\alpha\lambda_L+\beta\lambda_R. G ≥ W := c + α λ L + β λ R .
The right side is affine in the two means. The determinant concavity follows from the full cap and transport rows. In the whole mode-0 chart let ( u − , v − ) (u_-,v_-) ( u − , v − ) and ( u + , v + ) (u_+,v_+) ( u + , v + ) be the central entrance and exit states, and let K c , ℓ , m K_c,\ell,m K c , ℓ , m be the central invariant, outgoing slope and outgoing lower-left matrix entry. The outgoing cap equation has
m = D c K c − Z c , D c = v + − 2 + 2 β + Σ > 0 , m=D_cK_c-Z_c,\qquad D_c=v_+^{-2}+2\beta_+\Sigma>0, m = D c K c − Z c , D c = v + − 2 + 2 β + Σ > 0 ,
where Σ > 0 \Sigma> 0 Σ > 0 is the central shear, β + = [ ( 1 + u + ) / v + 2 + ( 1 + u + ) − 1 ] / 2 > 0 \beta_{+} = [(1 + u_{+})/v_{+}^{2} + (1 + u_{+})^{-1}]/2 > 0 β + = [( 1 + u + ) / v + 2 + ( 1 + u + ) − 1 ] /2 > 0 , and
Z c = 3 + u + 2 2 v + 2 + β + [ 2 c − ( u + − u − ) ] + 1 2 . Z_{c} = \frac{3 + u_{+}^{2}}{2v_{+}^{2}} + \beta_{+}[2c - (u_{+} - u_{-})] + \frac{1}{2}. Z c = 2 v + 2 3 + u + 2 + β + [ 2 c − ( u + − u − )] + 2 1 .
Thus
Γ = Z c 2 4 D c − D c ( K c − Z c 2 D c ) 2 − ℓ 2 . \Gamma= \frac{Z_{c}^{2}}{4D_{c}} - D_{c}\left(K_{c} - \frac{Z_{c}}{2D_{c}}\right)^{2} - \ell^{2}. Γ = 4 D c Z c 2 − D c ( K c − 2 D c Z c ) 2 − ℓ 2 .
Let d , e d,e d , e be the preceding and following own reciprocals in this whole chart, Q = ( y + d ) ( y + e ) / ( 1 + y 2 ) > 1 Q = (y + d)(y + e)/(1 + y^{2}) > 1 Q = ( y + d ) ( y + e ) / ( 1 + y 2 ) > 1 and b 0 = ( 1 − y e ) / ( y + e ) > 0 b_{0} = (1 - ye)/(y + e) > 0 b 0 = ( 1 − y e ) / ( y + e ) > 0 . Put
t L = d ( y + d ) / 2 + Σ , t R = ( 1 − u + ) / v + 2 , h L = ( 1 + d 2 ) / 2 , h R = ( 1 + b 0 2 ) / 2. t_{L} = d(y + d)/2 + \Sigma,\qquad t_{R} = (1 - u_{+})/v_{+}^{2},\qquad\mathfrak{h}_{L} = (1 + d^{2})/2,\qquad\mathfrak{h}_{R} = (1 + b_{0}^{2})/2. t L = d ( y + d ) /2 + Σ , t R = ( 1 − u + ) / v + 2 , h L = ( 1 + d 2 ) /2 , h R = ( 1 + b 0 2 ) /2.
The returned-invariant rows are
h i 2 K i = ( t i 2 − D c ) K c − 2 t i ℓ + Z c ( i = L , R ) . \mathfrak{h}_{i}^{2}K_{i} = (t_{i}^{2} - D_{c})K_{c} - 2t_{i}\ell+ Z_{c}\qquad(i = L,R). h i 2 K i = ( t i 2 − D c ) K c − 2 t i ℓ + Z c ( i = L , R ) .
They have a nonsingular coefficient matrix, since
t L − t R = ( 1 + y 2 ) 2 ( Q 2 − 1 ) 2 ( y + e ) 2 > 0 , ( t L − t R ) ( t L t R + D c ) > 0. t_{L} - t_{R} = \frac{(1 + y^{2})^{2}(Q^{2} - 1)}{2(y + e)^{2}} > 0,\qquad(t_{L} - t_{R})(t_{L}t_{R} + D_{c}) > 0. t L − t R = 2 ( y + e ) 2 ( 1 + y 2 ) 2 ( Q 2 − 1 ) > 0 , ( t L − t R ) ( t L t R + D c ) > 0.
Thus the two means map affinely and nonsingularly to ( K c , ℓ ) (K_{c},\ell) ( K c , ℓ ) . With J c J_{c} J c denoting the Jacobian, the determinant Hessian is − 2 J c T diag ( D c , 1 ) J c < 0 -2J_{c}^{T}\operatorname{diag}(D_{c},1)J_{c} < 0 − 2 J c T diag ( D c , 1 ) J c < 0 at fixed geometry and resulting actual cost. The central invariant K c K_{c} K c is distinct from the returned factors ( 3 + μ i ) / 2 (3+\mu_{i})/2 ( 3 + μ i ) /2 . Therefore both Γ \Gamma Γ and W + 8 ( Γ − H ‾ ) W + 8(\Gamma- \overline{H}) W + 8 ( Γ − H ) can be bounded below on the whole polygon by their vertex minima, where H ‾ = 1470471 / 2000000 > H ∗ \overline{H} = 1470471/2000000 > H_{*} H = 1470471/2000000 > H ∗ . The polygon, its weights and vertex enclosures are kept fixed in the geometric mean-value estimate. No concavity claim for the original rational G G G is used. Comparison-vertex slopes need not be positive; positivity is used only at the actual point to obtain G ≥ W G \ge W G ≥ W .
Lemma 21.4 (Complete weighted-center certificate). Under the actual complete-source hypotheses, the cost window (62) , and the same-lift harmonic ceiling,
G > 343 / 1000. G > 343/1000. G > 343/1000.
Computer-assisted proof. The binary80 interval program covers the entire root (65) by the six closed q q q slabs [ 5 / 2 , 7 / 2 ] [5/2, 7/2] [ 5/2 , 7/2 ] , [ 7 / 2 , 5 ] [7/2, 5] [ 7/2 , 5 ] , [ 5 , 8 ] [5, 8] [ 5 , 8 ] , [ 8 , 16 ] [8, 16] [ 8 , 16 ] , [ 16 , 32 ] [16, 32] [ 16 , 32 ] and [ 32 , 243 / 4 ] [32, 243/4] [ 32 , 243/4 ] . It verifies the preceding necessary rows and the fixed-polygon vertex tests. The whole-root analytic guards justify all interval denominators and analytic extensions. Arithmetic uses 64 significand bits with directed nextafterl endpoint enlargement; exact products of the 64-bit mantissas fit in unsigned 128-bit integers for the parallel-line test. The centered logarithm kernel atanh ( z ) / z \operatorname{atanh}(z)/z atanh ( z ) / z uses terms through index 24 with value remainder ∣ z ∣ 50 / [ 51 ( 1 − z 2 ) ] |z|^{50}/[51(1-z^{2})] ∣ z ∣ 50 / [ 51 ( 1 − z 2 )] and derivative remainder ∣ z ∣ 49 / ( 1 − z 2 ) |z|^{49}/(1-z^{2}) ∣ z ∣ 49 / ( 1 − z 2 ) . Unsafe intervals become unresolved or wide; they cannot certify exclusion.
The six closed trees total 5237850 nodes, 955835 accepted leaves and 1663093 excluded leaves, with zero unresolved boxes. Their exact forest identity is nodes = 2 ( accepted + excluded ) − 6 = 2(\text{accepted} + \text{excluded}) - 6 = 2 ( accepted + excluded ) − 6 . Every accepted reserve is strictly positive. The root union, endpoint inclusion, fixed-polygon logic and arithmetic contract are checked together with the finite witness. The program explicitly requires its binary80 platform contract; it does not silently treat ordinary binary64 arithmetic as equivalent.
The actual opposite centers 4 and 10 both have the required complete sources and cost windows. Equations (63) , (64) and Lemma 21.4 yield
C > 2 14 25 + 2 343 1000 = 903 500 = C ∗ + 117 100000 . C > 2\frac{14}{25} + 2\frac{343}{1000} = \frac{903}{500} = C_{*} + \frac{117}{100000}. C > 2 25 14 + 2 1000 343 = 500 903 = C ∗ + 100000 117 .
This contradicts the original budget and excludes R17. The full maximizing-PMP and source hypotheses remain essential.
Proof of Theorem 21.1 . The six types above exhaust this family by Theorem 20.3 . The low, high, double-lower and mixed/upper ranges in § 4.14 , with Lemma 21.2 , exclude R07. Proposition 21.3 excludes R15 by its conditional central fee and three complementary physical payments. For R17, (63) , (64) and Lemma 21.4 give C > 903 / 500 > C ∗ C > 903/500 > C_* C > 903/500 > C ∗ . Only R03, R05 and R06 remain. □
One negative run of length five
The one- and two-positive-run propositions leave seven words with a negative run of length five. With one positive run they are − 5 + 3 -5+3 − 5 + 3 and − 5 + 6 -5+6 − 5 + 6 , namely R02 and R08. With two positive runs, start immediately before the length-five negative run and write
The remaining pairs are
( a , b ) = ( 1 , 1 ) , ( 1 , 4 ) , ( 2 , 3 ) , ( 3 , 2 ) , ( 4 , 1 ) , (a,b)=(1,1),(1,4),(2,3),(3,2),(4,1), ( a , b ) = ( 1 , 1 ) , ( 1 , 4 ) , ( 2 , 3 ) , ( 3 , 2 ) , ( 4 , 1 ) ,
corresponding to R04 and R10–R13. These are exactly the seven members of this family in Theorem 20.3 ; the bounds on all run lengths have already been proved in § 4.6 . The length-five negative run supplies a rooted prefix and its adjacent one-sided leaf. The decisive information is the actual endpoint sign and its transfer to the neighboring capped core. The following three arguments retain that common geometry and give the full conditional cost comparisons for R08, the pair R10/R13, and the pair R11/R12. Their reusable one-sided construction is proved in Section C.9.
Exclusion of R08 by the reverse-pair payment.
Theorem 22.1. R08 does not occur as the positive-maximal-arc word of an original global minimizer.
Proof. Use native word 02102120120, B = 11 B=11 B = 11 . Set x = c 2 + c 3 x=c_2+c_3 x = c 2 + c 3 , Q = c 4 + c 5 + c 6 Q=c_4+c_5+c_6 Q = c 4 + c 5 + c 6 , R = c 10 + c 0 + c 1 R=c_{10}+c_0+c_1 R = c 10 + c 0 + c 1 , M = c 7 + c 8 + c 9 M=c_7+c_8+c_9 M = c 7 + c 8 + c 9 , P ∗ = 180483 / 500000 P_*=180483/500000 P ∗ = 180483/500000 .
The descending six-source ( 8 , 7 , 6 , 5 , 4 , 3 ) (8,7,6,5,4,3) ( 8 , 7 , 6 , 5 , 4 , 3 ) under full reflection j ↦ 1 − j j\mapsto1-j j ↦ 1 − j is 012021; it gives Q > 61 / 200 Q>61/200 Q > 61/200 and Q + c 7 > 81 / 200 Q+c_7>81/200 Q + c 7 > 81/200 . Native [ − 3 , 2 ] = 012021 [-3,2]=012021 [ − 3 , 2 ] = 012021 similarly gives R > 61 / 200 R>61/200 R > 61/200 and R + c 9 > 81 / 200 R+c_9>81/200 R + c 9 > 81/200 . The proper forward-six sources [ 5 , 10 ] [5,10] [ 5 , 10 ] and [ 6 , 11 ] [6,11] [ 6 , 11 ] give c 7 + c 8 > P ∗ c_7+c_8>P_* c 7 + c 8 > P ∗ and c 8 + c 9 > P ∗ c_8+c_9>P_* c 8 + c 9 > P ∗ .
The prefix on [ − 1 , 4 ] [-1,4] [ − 1 , 4 ] with c 2 ≤ 13 / 20 c_2\le13/20 c 2 ≤ 13/20 gives the positive terminal root of its one-sided leaf exactly as in Lemma C.7. In Q’s reversed chart, its endpoint w = b + i v w=b+iv w = b + i v maps back by F ( w ) = ( − w − 3 ) / ( 1 − w ) F(w)=(-w-3)/(1-w) F ( w ) = ( − w − 3 ) / ( 1 − w ) . Positivity of its real part is
b 2 + 2 b − 3 + v 2 ( 1 − b ) 2 + v 2 > 0 , \frac{b^2+2b-3+v^2}{(1-b)^2+v^2}>0, ( 1 − b ) 2 + v 2 b 2 + 2 b − 3 + v 2 > 0 ,
which supplies Q’s missing terminal circle. Together with its entrance circle this proves Q > 2 / 5 Q>2/5 Q > 2/5 by the circle-only tail comparison. More explicitly, its determinant parameter satisfies Γ = 3 / 4 − L / 4 \Gamma=3/4-\mathcal{L}/4 Γ = 3/4 − L /4 . Were L < 59 / 1000 \mathcal{L}<59/1000 L < 59/1000 , one would have Γ > 2941 / 4000 > 0 \Gamma>2941/4000>0 Γ > 2941/4000 > 0 ; together with the actual K > 0 K>0 K > 0 , this contradicts Lemma B.67. Thus the loss bound required by Proposition 10.1 is obtained without assuming harmonic calibration in advance. The combined support is [ − 1 , 8 ] [-1,8] [ − 1 , 8 ] , ten arcs, so one genuine period suffices. The reflected prefix similarly gives c 3 ≤ 13 / 20 ⇒ R > 2 / 5 c_3\le13/20\Rightarrow R>2/5 c 3 ≤ 13/20 ⇒ R > 2/5 , on the proper support [ − 3 , 6 ] [-3,6] [ − 3 , 6 ] . In these applications determinant positivity follows from the tail-loss comparison before the positive mean supplies the future sheet and harmonic calibration.
If c 2 > 13 / 20 c_2>13/20 c 2 > 13/20 , the unconditional tail bounds imply c 3 < 54483 / 100000 < 13 / 20 c_3<54483/100000<13/20 c 3 < 54483/100000 < 13/20 . Then the disjoint masks { 2 } , { 4 , 5 , 6 , 7 } , { 8 , 9 } , { 10 , 0 , 1 } \{2\},\{4,5,6,7\},\{8,9\},\{10,0,1\} { 2 } , { 4 , 5 , 6 , 7 } , { 8 , 9 } , { 10 , 0 , 1 } give
C > 13 20 + 81 200 + P ∗ + 2 5 = 180483 100000 + 1392 125000 . (66) C>\frac{13}{20}+\frac{81}{200}+P_*+\frac{2}{5}
=\frac{180483}{100000}+\frac{1392}{125000}. \tag*{(66)} C > 20 13 + 200 81 + P ∗ + 5 2 = 100000 180483 + 125000 1392 . (66)
The full reflected self-map j i ′ = 1 − j 5 − i j'_i=1-j_{5-i} j i ′ = 1 − j 5 − i gives the same contradiction for c 3 > 13 / 20 c_3>13/20 c 3 > 13/20 . Hence any survivor has both centres cheap, Q , R > 2 / 5 Q,R>2/5 Q , R > 2/5 , and M > P ∗ M>P_* M > P ∗ . The disjoint partition then gives
x < 180483 100000 − 4 5 − P ∗ = 80483 125000 < 13 20 . x<\frac{180483}{100000}-\frac{4}{5}-P_*=\frac{80483}{125000}<\frac{13}{20}. x < 100000 180483 − 5 4 − P ∗ = 125000 80483 < 20 13 .
The proper source [ − 1 , 6 ] = 20210212 [-1,6]=20210212 [ − 1 , 6 ] = 20210212 charges exactly the pair { 2 , 3 } \{2,3\} { 2 , 3 } . Theorem E.2 gives x > 13 / 20 x>13/20 x > 13/20 , contradicting the preceding strict upper bound. Equivalently, the disjoint charges { 2 , 3 } \{2,3\} { 2 , 3 } , Q Q Q , R R R and either complete forward pair in M M M give
C > 13 20 + 4 5 + 180483 500000 = 905483 500000 = 180483 100000 + 767 125000 . C>\frac{13}{20}+\frac{4}{5}+\frac{180483}{500000}=\frac{905483}{500000}=\frac{180483}{100000}+\frac{767}{125000}. C > 20 13 + 5 4 + 500000 180483 = 500000 905483 = 100000 180483 + 125000 767 .
No source matrix is identified with the lift used for a different scalar charge; the two prefix forcings and final calibration are already proved together inside Theorem E.2. □
Exclusion of the two twelve-arc classes
Throughout this section the geometric hypotheses, full endpoint conditions, and switching-type convention are those of the original problem. Put
C ∗ = 180483 100000 , δ 0 = 29529 2000000 , E ∗ = 4154032208439 8759490110000 . C_*=\frac{180483}{100000},\qquad\delta_0=\frac{29529}{2000000},\qquad E_*=\frac{4154032208439}{8759490110000}. C ∗ = 100000 180483 , δ 0 = 2000000 29529 , E ∗ = 8759490110000 4154032208439 .
Physical arc indices are taken modulo 12, whereas source intervals are read on the actual twisted lift. For R 10 R10 R 10 that lift has
( j 0 , … , j 11 ) = 021021210120 , j i + 12 = j i + 1 ( m o d 3 ) . (j_0,\ldots,j_{11})=021021210120,\qquad j_{i+12}=j_i+1\pmod{3}. ( j 0 , … , j 11 ) = 021021210120 , j i + 12 = j i + 1 ( mod 3 ) .
Write c i > 0 c_i>0 c i > 0 for the cost of its complete maximal arc i i i and C = ∑ i = 0 11 c i C=\sum_{i=0}^{11}c_i C = ∑ i = 0 11 c i . A surviving minimizing owner would satisfy C < C ∗ C<C_* C < C ∗ . All phase changes and reflections below are the full transformations of Proposition 2.5 ; they transport the state, frame, matrix adjoint, both state covectors, and objective multiplier.
Source placement and the conditional interfaces
Every coupled calculation below is made on a genuine containing period, with its original state and frame endpoints and real winding. Lemma 2.6 supplies the reanchoring. The complete capped four-core [ 3 , 8 ] = 021210 [3,8]=021210 [ 3 , 8 ] = 021210 , transformed by phase + 1 +1 + 1 to 102021, supplies normality and the automatic calibration of Corollary B.9 . Joint normalization of that full lift, followed by the actual apex and future-sheet conclusions of that corollary, gives
0 < Γ < H ∗ < 1470471 2000000 , δ = 3 4 − Γ > δ 0 . (67) 0<\Gamma<H_*<\frac{1470471}{2000000},\qquad\delta=\frac{3}{4}-\Gamma>\delta_0. \tag*{(67)} 0 < Γ < H ∗ < 2000000 1470471 , δ = 4 3 − Γ > δ 0 . (67)
In particular, no state covector is extended periodically.
Lemma 22.2 (Transferred own-pair interface). On the complete source [ − 1 , 9 ] = 20210212101 [-1,9]=20210212101 [ − 1 , 9 ] = 20210212101 , if c 2 ≤ 13 / 20 c_2\le13/20 c 2 ≤ 13/20 , the actual E E E own pair { 5 , 6 } \{5,6\} { 5 , 6 } has both chronological density ceilings 0 < x L , x R < 47 / 250 0<x_L,x_R<47/250 0 < x L , x R < 47/250 and both cap gaps
e j ≥ 0 , δ ≤ 3 ( 1 − 3 x j ) 4 e j + e j 2 , e j > 47 2500 + x j 20 . (68) e_j\ge0,\qquad\delta\le\frac{3(1-3x_j)}{4}e_j+e_j^2,\qquad e_j>\frac{47}{2500}+\frac{x_j}{20}. \tag*{(68)} e j ≥ 0 , δ ≤ 4 3 ( 1 − 3 x j ) e j + e j 2 , e j > 2500 47 + 20 x j . (68) Its two actual zero-to-zero mean leaves and their common invariant are unchanged. In particular c 5 + c 6 > 283 / 1000 c_5+c_6>283/1000 c 5 + c 6 > 283/1000 .
Proof. The eleven-source is identical to the complete source used in the one-sided construction in the proof of Lemma C.7. Here retain it in the genuine twelve-arc period [ − 1 , 10 ] [-1,10] [ − 1 , 10 ] , together with the capped calibration source [ 3 , 8 ] [3,8] [ 3 , 8 ] . The prefix [ − 1 , 4 ] [-1,4] [ − 1 , 4 ] has centre 2, its actual outgoing arc is 3, the shared incoming arc of E E E is 4, the true own leaves are 5, 6, and the right return is 7, 8, 9. Thus that construction applies without any change of its local hypotheses. In particular its cap equation retains the nonnegative interval x x x , its one-sided integral yields the missing midpoint gap, and its actual full matrix conjugacy gives (22.2). The one-sided
invariant on arc 3 is not identified with the own invariant of E E E . Proposition 10.2, with these recovered interfaces, gives the stated own-pair cost. Only this local construction is used, not any exclusion of another switching type. □
Lemma 22.3 (Circle transfer). For R 10 R10 R 10 , put T = c 11 + c 0 + c 1 T=c_{11}+c_0+c_1 T = c 11 + c 0 + c 1 . Then T > 61 / 200 T>61/200 T > 61/200 always. If c 3 ≤ 13 / 20 c_3\le13/20 c 3 ≤ 13/20 , then T > 2 / 5 T>2/5 T > 2/5 . More generally the latter conclusion holds whenever the actual prefix read on ( 6 , 5 , 4 , 3 , 2 , 1 ) (6,5,4,3,2,1) ( 6 , 5 , 4 , 3 , 2 , 1 ) under j ↦ 1 − j j\mapsto1-j j ↦ 1 − j has the strict exit sign of Theorem B.107.
Proof. The complete source [ − 3 , 2 ] = 012021 [-3,2]=012021 [ − 3 , 2 ] = 012021 gives the negative entrance and entrance circle for the capped 202 tail on [ − 1 , 1 ] [-1,1] [ − 1 , 1 ] , whose two caps are retained in [ − 2 , 2 ] = 12021 [-2,2]=12021 [ − 2 , 2 ] = 12021 ; see Lemma B.77. Lemma B.75 gives the unconditional payment. The stated reverse prefix is literally 202102, with centre 3. Its exit sign, together with (B.184), gives the one-sided leaf argument in the proof of Lemma C.7: at the actual junction 1 ∣ 2 1|2 1∣2 its reflected horizontal coordinate is strictly positive. In the original tail chart, at w = b + i v w=b+iv w = b + i v , the whole transformation is
F ( w ) = − w − 3 1 − w . F(w)=\frac{-w-3}{1-w}. F ( w ) = 1 − w − w − 3 .
The inequality Re F ( w ) > 0 \operatorname{Re} F(w)>0 Re F ( w ) > 0 is exactly v 2 > ( 1 − b ) ( 3 + b ) v^2>(1-b)(3+b) v 2 > ( 1 − b ) ( 3 + b ) , the terminal circle. This is intrinsic state data. For the tail comparison choose the genuine period [ − 3 , 8 ] [-3,8] [ − 3 , 8 ] , of exactly 12 complete arcs. It retains the whole coupled source [ − 3 , 6 ] [-3,6] [ − 3 , 6 ] and the capped calibration source [ 3 , 8 ] [3,8] [ 3 , 8 ] . Thus both circles, both actual caps and the calibration are available on one full lift. If the exit sign was established on [ − 1 , 10 ] [-1,10] [ − 1 , 10 ] , only its state-circle consequence is carried to this period; no external normalized costate is imported. Proposition 10.1, in the circle-only form proved in Lemma C.7, now gives T > 2 / 5 T>2/5 T > 2/5 . For c 3 ≤ 13 / 20 c_3\le13/20 c 3 ≤ 13/20 the required exit sign follows from Theorem B.107, including equality at the cost threshold. □
An exclusive payment at the right tail.
Lemma 22.4 (Two exclusive arcs). Every surviving R 10 R10 R 10 owner satisfies c 8 + c 9 > 31 / 200 c_8+c_9>31/200 c 8 + c 9 > 31/200 .
Proof. Retain the proper nine-source [ 3 , 11 ] = 021210120 [3,11]=021210120 [ 3 , 11 ] = 021210120 on the genuine period [ 3 , 14 ] [3,14] [ 3 , 14 ] . The one-return source [ 3 , 9 ] [3,9] [ 3 , 9 ] , in phase + 1 +1 + 1 , is 1020212; its 0202 core is { 4 , 5 , 6 , 7 } \{4,5,6,7\} { 4 , 5 , 6 , 7 } , with own means 5, 6 and true right returned mean 8. The capped source [ 6 , 10 ] [6,10] [ 6 , 10 ] , in phase + 2 +2 + 2 , is 10201; its 020 core is { 7 , 8 , 9 } \{7,8,9\} { 7 , 8 , 9 } and its middle mean is arc 8. Their shared physical arc is 7, the last mode-2 arc of the first chart and the first mode-0 arc of the second. Their adjacent actual mean leaves are 6 and 8. The capped four-core supplies (22.1) on this one full lift.
In the first chart write K L = ( 3 + h L 2 ) / 2 K_L=(3+h_L^2)/2 K L = ( 3 + h L 2 ) /2 , and let r > h L r>h_L r > h L be its positive outer root. In the second chart put
K = 3 + μ 2 , − s ≤ − μ , ∣ z ∣ ≤ μ . K=\frac{3+\mu}{2},\qquad-s\le-\sqrt{\mu},\qquad|z|\le\sqrt{\mu}. K = 2 3 + μ , − s ≤ − μ , ∣ z ∣ ≤ μ .
where − s , z -s,z − s , z are the two internal roots. Let β > z \beta>z β > z be the terminal cap coordinate. Lemma B.58 and (C.4), (C.7), applied to the actual means 6, 8, give, for the shared density x x x and constant λ \lambda λ ,
Γ = K L K ( 1 + x ) 2 4 − λ 2 , 0 < x < 1 / 3 , e = λ + 3 x / 2 ≥ 0 , \Gamma=\frac{K_LK(1+x)^2}{4}-\lambda^2,\qquad0<x<1/3,\qquad e=\lambda+3x/2\ge0, Γ = 4 K L K ( 1 + x ) 2 − λ 2 , 0 < x < 1/3 , e = λ + 3 x /2 ≥ 0 ,
δ ≤ 3 ( 1 − 3 x ) 4 e + e 2 ≤ 3 e / 4 + e 2 . (69) \delta\le\frac{3(1-3x)}{4}e+e^2\le3e/4+e^2. \tag*{(69)} δ ≤ 4 3 ( 1 − 3 x ) e + e 2 ≤ 3 e /4 + e 2 . (69)
The local scope of these identities is precisely one true adjacent mean on each side and their two midpoint gaps, as in Section C.1.7; a right four-core is not required. Since
29 2000 − 3 4 47 2500 − ( 47 2500 ) 2 = 291 6250000 > 0 , \frac{29}{2000}-\frac{3}{4}\frac{47}{2500}-\left(\frac{47}{2500}\right)^2=\frac{291}{6250000}>0, 2000 29 − 4 3 2500 47 − ( 2500 47 ) 2 = 6250000 291 > 0 ,
(22.1) and monotonicity for e ≥ 0 e\ge0 e ≥ 0 imply
e > 47 / 2500. (70) e>47/2500. \tag*{(70)} e > 47/2500. (70)
No high-density ceiling or second determinant gap is needed here.
Descending arcs ( 11 , 10 , 9 , 8 , 7 , 6 , 5 ) (11, 10, 9, 8, 7, 6, 5) ( 11 , 10 , 9 , 8 , 7 , 6 , 5 ) under j ↦ − j j \mapsto-j j ↦ − j form 0120212. The entrance conclusion of Lemma B.77 for this actual source becomes, in the forward 020 chart,
β > 0 , q ( β ) < 1 ( 1 − β ) ( 3 + β ) , q = v − 2 . (71) \beta> 0,\qquad q(\beta) < \frac{1}{(1-\beta)(3+\beta)},\qquad q=v^{-2}. \tag*{(71)} β > 0 , q ( β ) < ( 1 − β ) ( 3 + β ) 1 , q = v − 2 . (71)
Indeed the reverse chart is exactly the whole binary reflection of the phase-+ 2 +2 + 2 chart, since 2 − ( j + 2 ) = − j 2-(j+2)=-j 2 − ( j + 2 ) = − j modulo 3. This uses an actual entrance circle in reverse chronology; it does not insert a terminal returned mean into the forward 020 core.
Write P = c 8 + c 9 P=c_8+c_9 P = c 8 + c 9 , w = s + β w=s+\beta w = s + β , and let T > 0 T>0 T > 0 be the shear over precisely these two arcs, with d s = d t / v = q d u \mathrm{d}s=\mathrm{d}t/v=q\,\mathrm{d}u d s = d t / v = q d u . The mean and physical-cost identities give
0 < μ < s 2 < w 2 , P = w 2 + 1 2 ∫ ( 3 + u 2 ) d s > w / 2 + 3 T / 2 , P > 2 T . 0<\mu<s^2<w^2,\qquad P=\frac{w}{2}+\frac{1}{2}\int(3+u^2)\,\mathrm{d}s>w/2+3T/2,\qquad P>2T. 0 < μ < s 2 < w 2 , P = 2 w + 2 1 ∫ ( 3 + u 2 ) d s > w /2 + 3 T /2 , P > 2 T .
Suppose P ≤ 31 / 200 P\le31/200 P ≤ 31/200 , and put V = 31 / 100 V=31/100 V = 31/100 . Then
0 < w < V , μ < w 2 < V 2 < 1 / 10 , T < min { 31 / 400 , ( V − w ) / 3 } , K < 30961 / 20000. (72) 0<w<V,\qquad\mu<w^2<V^2<1/10,\qquad T<\min\{31/400,(V-w)/3\},\qquad K<30961/20000. \tag*{(72)} 0 < w < V , μ < w 2 < V 2 < 1/10 , T < min { 31/400 , ( V − w ) /3 } , K < 30961/20000. (72)
In particular K T < 959791 / 8000000 < 1 / 2 KT<959791/8000000<1/2 K T < 959791/8000000 < 1/2 .
We first determine the sign of the actual terminal matrix entry L L L . At the last mode-0 cap put d = 1 + β d=1+\beta d = 1 + β and abbreviate terminal values by q , χ , f , L q,\chi,f,L q , χ , f , L . The actual maximizing equations give
χ ≥ 0 , g = m + 1 / 2 − f = 0 , f = ( μ − β 2 ) q 2 + d q + 1 / d 2 χ , \chi\ge0,\qquad g=m+1/2-f=0,\qquad f=\frac{(\mu-\beta^2)q}{2}+\frac{dq+1/d}{2}\chi, χ ≥ 0 , g = m + 1/2 − f = 0 , f = 2 ( μ − β 2 ) q + 2 d q + 1/ d χ ,
g s = − 2 L − f s ≤ 0 , f s = − β − f d q + ( μ − β 2 ) q d 2 . g_s=-2L-f_s\le0,\qquad f_s=-\beta-\frac{f}{dq}+\frac{(\mu-\beta^2)qd}{2}. g s = − 2 L − f s ≤ 0 , f s = − β − d q f + 2 ( μ − β 2 ) q d .
These are the mode-0 cap identities used in Theorem B.61; the inequality follows from g ≥ 0 g\ge0 g ≥ 0 on the inward side. By (71) and β < V < 2 − 1 \beta<V<\sqrt{2}-1 β < V < 2 − 1 , one has q d 2 < 1 qd^2<1 q d 2 < 1 . Hence
f s ≤ − β + μ − β 2 2 ( q d − 1 / d ) < 0. f_s\le-\beta+\frac{\mu-\beta^2}{2}(qd-1/d)<0. f s ≤ − β + 2 μ − β 2 ( q d − 1/ d ) < 0.
For μ ≥ β 2 \mu\ge\beta^2 μ ≥ β 2 the second term is nonpositive; for μ < β 2 \mu<\beta^2 μ < β 2 it is strictly less than β 2 / ( 2 d ) < β \beta^2/(2d)<\beta β 2 / ( 2 d ) < β . It follows that L ≥ − f s / 2 > 0 L\ge-f_s/2>0 L ≥ − f s /2 > 0 , before any square root is taken.
The real root z z z has χ ( z ) = 0 \chi(z)=0 χ ( z ) = 0 , so integration of χ s = μ − u 2 \chi_s=\mu-u^2 χ s = μ − u 2 on the last arc gives 0 ≤ χ ≤ μ T 0\le\chi\le\mu T 0 ≤ χ ≤ μ T . Retain this possibly nonzero terminal value. The cap determinant is
L 2 = δ + μ / 4 − K f . (73) L^2=\delta+\mu/4-Kf. \tag*{(73)} L 2 = δ + μ /4 − K f . (73)
Put D = ( 1 − β ) ( 3 + β ) D=(1-\beta)(3+\beta) D = ( 1 − β ) ( 3 + β ) and B = μ − β 2 + d μ T B=\mu-\beta^2+d\mu T B = μ − β 2 + d μ T . If B ≤ 0 B\le0 B ≤ 0 , the formula for f f f gives f ≤ μ T / ( 2 d ) f\le\mu T/(2d) f ≤ μ T / ( 2 d ) , and therefore
L 2 ≥ δ + μ { 1 / 4 − K T / ( 2 d ) } > δ > δ 0 . L^2\ge\delta+\mu\{1/4-KT/(2d)\}>\delta>\delta_0. L 2 ≥ δ + μ { 1/4 − K T / ( 2 d )} > δ > δ 0 .
If B > 0 B>0 B > 0 , the same terminal circle gives f ≤ q B / 2 + μ T / ( 2 d ) < B / ( 2 D ) + μ T / ( 2 d ) f\le qB/2+\mu T/(2d)<B/(2D)+\mu T/(2d) f ≤ q B /2 + μ T / ( 2 d ) < B / ( 2 D ) + μ T / ( 2 d ) . Substitution into (73) gives exactly
L 2 > δ + 3 β 2 − 2 μ β − μ 2 4 D − ( 3 + μ ) μ T d D . L^2>\delta+\frac{3\beta^2-2\mu\beta-\mu^2}{4D}-\frac{(3+\mu)\mu T}{dD}. L 2 > δ + 4 D 3 β 2 − 2 μ β − μ 2 − d D ( 3 + μ ) μ T .
Here d 2 + D = 4 d^2+D=4 d 2 + D = 4 , so the last term includes the full contribution of χ \chi χ . Complete the square as 3 β 2 − 2 μ β − μ 2 = 3 ( β − μ / 3 ) 2 − 4 μ 2 / 3 3\beta^2-2\mu\beta-\mu^2=3(\beta-\mu/3)^2-4\mu^2/3 3 β 2 − 2 μ β − μ 2 = 3 ( β − μ /3 ) 2 − 4 μ 2 /3 . Moreover
D > 22839 10000 > 9 4 , d D = 3 + β − 3 β 2 − β 3 ≥ min { 3 , 3 + V − 3 V 2 − V 3 } = 2991909 1000000 > 299 100 . D>\frac{22839}{10000}>\frac{9}{4},\qquad dD=3+\beta-3\beta^2-\beta^3\ge\min\{3,3+V-3V^2-V^3\}=\frac{2991909}{1000000}>\frac{299}{100}. D > 10000 22839 > 4 9 , d D = 3 + β − 3 β 2 − β 3 ≥ min { 3 , 3 + V − 3 V 2 − V 3 } = 1000000 2991909 > 100 299 .
The minimum assertion follows from concavity on [ 0 , V ] [0,V] [ 0 , V ] . Thus ( 3 + μ ) / ( d D ) < 310 / 299 < 21 / 20 (3+\mu)/(dD)<310/299<21/20 ( 3 + μ ) / ( d D ) < 310/299 < 21/20 . Both signs of B B B yield
L 2 > δ 0 − 4 w 4 27 − 21 20 w 2 T . (74) L^2>\delta_0-\frac{4w^4}{27}-\frac{21}{20}w^2T. \tag*{(74)} L 2 > δ 0 − 27 4 w 4 − 20 21 w 2 T . (74)
By (72) and w 2 ( V − w ) ≤ 4 V 3 / 27 w^2(V-w)\le4V^3/27 w 2 ( V − w ) ≤ 4 V 3 /27 ,
4 w 4 27 + 21 20 w 2 T < 4 V 4 27 + 7 V 3 135 = 7864824 2700000000 . \frac{4w^4}{27}+\frac{21}{20}w^2T<\frac{4V^4}{27}+\frac{7V^3}{135}=\frac{7864824}{2700000000}. 27 4 w 4 + 20 21 w 2 T < 27 4 V 4 + 135 7 V 3 = 2700000000 7864824 .
The exact reserve
δ 0 − ( 21 / 200 ) 2 − 7864824 2700000000 = 2231826 2700000000 > 0 \delta_0-(21/200)^2-\frac{7864824}{2700000000}=\frac{2231826}{2700000000}>0 δ 0 − ( 21/200 ) 2 − 2700000000 7864824 = 2700000000 2231826 > 0
therefore proves L > 21 / 200 L>21/200 L > 21/200 .
Finally, the full shared transform (B.103) and the actual equation L s = K L_s=K L s = K give
L ( − s ) = − λ − K x 1 − s , K T = L − L ( − s ) = L + e + x ( μ + 3 s ) 2 ( 1 − s ) > L + e > 619 5000 . L(-s)=-\lambda-\frac{Kx}{1-s},\qquad KT=L-L(-s)=L+e+\frac{x(\mu+3s)}{2(1-s)}>L+e>\frac{619}{5000}. L ( − s ) = − λ − 1 − s K x , K T = L − L ( − s ) = L + e + 2 ( 1 − s ) x ( μ + 3 s ) > L + e > 5000 619 .
This contradicts (72) , since
619 5000 − 959791 8000000 = 30609 8000000 > 0. \frac{619}{5000}-\frac{959791}{8000000}=\frac{30609}{8000000}>0. 5000 619 − 8000000 959791 = 8000000 30609 > 0.
Thus P > 31 / 200 P>31/200 P > 31/200 . Its charge excludes shared arc 7 and both caps; every coupled identity used only the common nine-source. □
A height floor for the actual own pair.
Lemma 22.5 (The refined height). If c 2 ≤ 13 / 20 c_2\le13/20 c 2 ≤ 13/20 , the actual height h E h_E h E of the own mean on arcs 5, 6 satisfies h E > 1 / 12 h_E>1/12 h E > 1/12 .
Proof. Use the same full lift and densities supplied by Lemma 22.2 . For either density x x x and defect e e e , put F x ( e ) = 3 ( 1 − 3 x ) e / 4 + e 2 F_x(e)=3(1-3x)e/4+e^2 F x ( e ) = 3 ( 1 − 3 x ) e /4 + e 2 . It increases for e ≥ 0 e\ge0 e ≥ 0 and 0 < x < 1 / 3 0<x<1/3 0 < x < 1/3 , and
δ 0 − F x ( 19 / 1000 + x / 20 ) = 307 2000000 + 67 x 20000 + 11 x 2 100 > 0. \delta_0-F_x(19/1000+x/20)=\frac{307}{2000000}+\frac{67x}{20000}+\frac{11x^2}{100}>0. δ 0 − F x ( 19/1000 + x /20 ) = 2000000 307 + 20000 67 x + 100 11 x 2 > 0.
Thus e > 19 / 1000 + x / 20 e>19/1000+x/20 e > 19/1000 + x /20 .
Suppose 0 < h = h E ≤ 1 / 12 0<h=h_E\le1/12 0 < h = h E ≤ 1/12 . Write the actual roots as − a , z , b -a,z,b − a , z , b , and put
R = 8 h 4 − 3 h , k = 1 − R 1 + R , X = max ( x L , x R ) , S = x L + x R , K = 3 + h 2 2 . R=\frac{8h}{4-3h},\qquad k=\frac{1-R}{1+R},\qquad X=\max(x_L,x_R),\qquad S=x_L+x_R,\qquad K=\frac{3+h^2}{2}. R = 4 − 3 h 8 h , k = 1 + R 1 − R , X = max ( x L , x R ) , S = x L + x R , K = 2 3 + h 2 .
Lemma B.8 gives h < a , b < R h<a,b<R h < a , b < R , ∣ z ∣ ≤ h |z|\le h ∣ z ∣ ≤ h . For M ( t ) = ( 1 + t ) / ( 1 − t ) M(t)=(1+t)/(1-t) M ( t ) = ( 1 + t ) / ( 1 − t ) the exact density ratio is
x R x L = ( M ( a ) M ( z ) M ( b ) ) 2 . \frac{x_R}{x_L}=\left(\frac{M(a)M(z)}{M(b)}\right)^2. x L x R = ( M ( b ) M ( a ) M ( z ) ) 2 .
Since z ≤ h < b z\le h<b z ≤ h < b this ratio is less than M ( R ) 2 M(R)^2 M ( R ) 2 ; using − z ≤ h < a -z\le h<a − z ≤ h < a proves the reciprocal bound. Consequently S > ( 1 + k 2 ) X S>(1+k^2)X S > ( 1 + k 2 ) X .
Let T T T be the actual own shear and D = x L a / ( 1 − a ) + x R b / ( 1 − b ) D=x_La/(1-a)+x_Rb/(1-b) D = x L a / ( 1 − a ) + x R b / ( 1 − b ) . The exact transport and Lemma C.1 give
N : = e L + e R = K ( T − D ) − h 2 S / 2 ≤ K ( 1 − k ) T − h 2 S / 2 ≤ A ( h ) X − h 2 S / 2 , N:=e_L+e_R=K(T-D)-h^2S/2\le K(1-k)T-h^2S/2\le A(h)X-h^2S/2, N := e L + e R = K ( T − D ) − h 2 S /2 ≤ K ( 1 − k ) T − h 2 S /2 ≤ A ( h ) X − h 2 S /2 ,
where the last inequality uses T ≤ 2 q ( z ) R / ( 1 + R ) T\le2q(z)R/(1+R) T ≤ 2 q ( z ) R / ( 1 + R ) , q ( z ) ≤ X / k 2 q(z)\le X/k^2 q ( z ) ≤ X / k 2 , and
A ( h ) = 4 K R 2 ( 1 − R ) 2 = 128 h 2 ( 3 + h 2 ) ( 4 − 11 h ) 2 . A(h)=\frac{4KR^2}{(1-R)^2}=\frac{128h^2(3+h^2)}{(4-11h)^2}. A ( h ) = ( 1 − R ) 2 4 K R 2 = ( 4 − 11 h ) 2 128 h 2 ( 3 + h 2 ) .
The refined defects also give N > 19 / 500 + S / 20 N > 19/500 + S/20 N > 19/500 + S /20 , so
19 500 < A ( h ) X − ( 1 / 20 + h 2 / 2 ) S < F ( h ) X , F ( h ) = A ( h ) − ( 1 / 20 + h 2 / 2 ) ( 1 + k 2 ) . (75) \frac{19}{500} < A(h)X - (1/20 + h^2/2)S < F(h)X,\qquad F(h) = A(h) - (1/20 + h^2/2)(1 + k^2). \tag*{(75)} 500 19 < A ( h ) X − ( 1/20 + h 2 /2 ) S < F ( h ) X , F ( h ) = A ( h ) − ( 1/20 + h 2 /2 ) ( 1 + k 2 ) . (75)
This function increases on [ 0 , 1 / 12 ] [0,1/12] [ 0 , 1/12 ] . Indeed A ( h ) = h 2 a ( h ) A(h) = h^2a(h) A ( h ) = h 2 a ( h ) , where a ( h ) a(h) a ( h ) is positive, increasing, and at least 24, so A ′ ( h ) ≥ 48 h A'(h) \ge48h A ′ ( h ) ≥ 48 h . Also σ ( h ) = 1 + k 2 \sigma(h) = 1 + k^2 σ ( h ) = 1 + k 2 decreases and σ ≤ 2 \sigma\le2 σ ≤ 2 . Thus F ′ = A ′ − h σ − ( 1 / 20 + h 2 / 2 ) σ ′ ≥ 46 h ≥ 0 F' = A' - h\sigma- (1/20 + h^2/2)\sigma' \ge46h \ge0 F ′ = A ′ − hσ − ( 1/20 + h 2 /2 ) σ ′ ≥ 46 h ≥ 0 . At h 0 = 1 / 12 h_0 = 1/12 h 0 = 1/12 one has
R 0 = 8 45 , A ( h 0 ) = 3464 12321 , σ ( h 0 ) = 4178 2809 , F ( h 0 ) = 558222323 2768775120 , R_0 = \frac{8}{45},\qquad A(h_0) = \frac{3464}{12321},\qquad\sigma(h_0) = \frac{4178}{2809},\qquad F(h_0) = \frac{558222323}{2768775120}, R 0 = 45 8 , A ( h 0 ) = 12321 3464 , σ ( h 0 ) = 2809 4178 , F ( h 0 ) = 2768775120 558222323 ,
19 500 − 47 250 F ( h 0 ) = 133828918 1384387560000 > 0. \frac{19}{500} - \frac{47}{250}F(h_0) = \frac{133828918}{1384387560000} > 0. 500 19 − 250 47 F ( h 0 ) = 1384387560000 133828918 > 0.
Since X < 47 / 250 X < 47/250 X < 47/250 , this contradicts (75) , including its endpoint h = h 0 h = h_0 h = h 0 . □
The strengthened central-prefix estimate.
Lemma 22.6 (Exit under the refined ingress). Suppose an actual contact-qualified 202102 prefix has central cost c < 69083 / 100000 c < 69083/100000 c < 69083/100000 and ingress u L > 1 / 12 u_L > 1/12 u L > 1/12 . Then its actual exit satisfies
u R < 0 , q R ≥ 0 , q R ′ > k q R . u_R < 0,\qquad q_R \ge0,\qquad q'_R > kq_R. u R < 0 , q R ≥ 0 , q R ′ > k q R . Here q q q is the raw switching response of (B.163), not the reciprocal state density used above.
Proof. Use the actual full central response in (B.160)–(B.167), with A > 0 A > 0 A > 0 , E > 1 E > 1 E > 1 , z = log E z = \log E z = log E and τ 2 = 1 + k 2 \tau^2 = 1 + k^2 τ 2 = 1 + k 2 . The strict ingress, both contact jets and physical cost imply
A E < 2 , A τ 2 > 2 , A > A t = 2 ( E z − E + 1 ) ( E − 1 ) 2 , a : = A ( E − 1 ) < 2 c < 7 / 5. (76) AE < 2,\qquad A\tau^2 > 2,\qquad A > A_t = \frac{2(Ez - E + 1)}{(E - 1)^2},\qquad a := A(E - 1) < 2c < 7/5. \tag*{(76)} A E < 2 , A τ 2 > 2 , A > A t = ( E − 1 ) 2 2 ( E z − E + 1 ) , a := A ( E − 1 ) < 2 c < 7/5. (76)
Lemma B.105 already gives u R < 0 u_R < 0 u R < 0 , without a cost restriction, and the actual outgoing tie gives q R ≥ 0 q_R \ge0 q R ≥ 0 . Since
u L = 4 − τ 2 A 2 4 − 4 A + τ 2 A 2 > 1 12 , τ 2 < 4 ( 11 + A ) 13 A 2 , u_L = \frac{4 - \tau^2 A^2}{4 - 4A + \tau^2 A^2} > \frac{1}{12},\qquad\tau^2 < \frac{4(11 + A)}{13A^2}, u L = 4 − 4 A + τ 2 A 2 4 − τ 2 A 2 > 12 1 , τ 2 < 13 A 2 4 ( 11 + A ) ,
the monotonicity of (B.173) in τ 2 \tau^2 τ 2 gives
c > C 12 ( A , E ) : = 2 A ( ( 11 + A ) ( E − 1 ) + 13 ( 1 − E − 1 ) ) − 13 A 2 log E 44 + 4 A − 13 A 2 . (77) c > C_{12}(A,E) := \frac{2A((11 + A)(E - 1) + 13(1 - E^{-1})) - 13A^2 \log E}{44 + 4A - 13A^2}. \tag*{(77)} c > C 12 ( A , E ) := 44 + 4 A − 13 A 2 2 A (( 11 + A ) ( E − 1 ) + 13 ( 1 − E − 1 )) − 13 A 2 log E . (77)
The denominator is ( 2 − A ) ( 13 A + 22 ) > 0 (2 - A)(13A + 22) > 0 ( 2 − A ) ( 13 A + 22 ) > 0 .
First C 12 C_{12} C 12 increases strictly in A A A on 0 < A < 2 / E 0 < A < 2/E 0 < A < 2/ E . Put e = E − 1 e = E - 1 e = E − 1 , a 1 = 22 e + 26 e / E a_1 = 22e + 26e/E a 1 = 22 e + 26 e / E , and a 2 = 2 e − 13 z a_2 = 2e - 13z a 2 = 2 e − 13 z . Its A A A -derivative has positive denominator and numerator
44 a 1 + 88 a 2 A + ( 13 a 1 + 4 a 2 ) A 2 . 44a_1 + 88a_2A + (13a_1 + 4a_2)A^2. 44 a 1 + 88 a 2 A + ( 13 a 1 + 4 a 2 ) A 2 .
If a 2 ≥ 0 a_2 \ge0 a 2 ≥ 0 this is positive. Otherwise z < e z < e z < e gives a 2 > − 11 e a_2 > -11e a 2 > − 11 e , whence 13 a 1 + 4 a 2 > 0 13a_1 + 4a_2 > 0 13 a 1 + 4 a 2 > 0 , and, using A < 2 / E A < 2/E A < 2/ E , the first two terms exceed 968 e − 792 e / E > 0 968e - 792e/E > 0 968 e − 792 e / E > 0 .
The whole prefix has E < 47 / 20 E < 47/20 E < 47/20 . Write D ( A ) = 44 + 4 A − 13 A 2 D(A) = 44 + 4A - 13A^2 D ( A ) = 44 + 4 A − 13 A 2 and W ( E ) = E z / ( E − 1 ) − 1 W(E) = Ez/(E - 1) - 1 W ( E ) = E z / ( E − 1 ) − 1 . At the contact comparator A = A t A = A_t A = A t direct substitution gives
C 12 ( A t , E ) = W ( E ) { 1 + 2 R ( A t ) / E } , R ( A ) = 13 ( A 2 − 2 A + 4 ) 2 D ( A ) . C_{12}(A_t,E) = W(E)\{1 + 2R(A_t)/E\},\qquad R(A) = \frac{13(A^2 - 2A + 4)}{2D(A)}. C 12 ( A t , E ) = W ( E ) { 1 + 2 R ( A t ) / E } , R ( A ) = 2 D ( A ) 13 ( A 2 − 2 A + 4 ) .
For 0 < A < 2 0 < A < 2 0 < A < 2 one has R ( A ) > 491 / 1000 R(A) > 491/1000 R ( A ) > 491/1000 , since
1000 D ( A ) ( R ( A ) − 491 / 1000 ) = 12883 A 2 − 14964 A + 4396 , 1000D(A)(R(A) - 491/1000) = 12883A^2 - 14964A + 4396, 1000 D ( A ) ( R ( A ) − 491/1000 ) = 12883 A 2 − 14964 A + 4396 ,
whose leading coefficient is positive and whose negative-discriminant reserve is 4 ⋅ 12883 ⋅ 4396 − 14964 2 = 2613376 > 0 4 \cdot12883 \cdot4396 - 14964^2 = 2613376 > 0 4 ⋅ 12883 ⋅ 4396 − 1496 4 2 = 2613376 > 0 . With b = 491 / 500 b = 491/500 b = 491/500 , the lower envelope W ( E ) ( 1 + b / E ) W(E)(1 + b/E) W ( E ) ( 1 + b / E ) is strictly increasing for E > 1 E > 1 E > 1 . Its derivative, multiplied by E 2 ( E − 1 ) 2 E^2(E - 1)^2 E 2 ( E − 1 ) 2 , is
( E − 1 ) { E 2 + b ( 2 E − 1 ) } − ( 1 + b ) E 2 log E . (E - 1)\{E^2 + b(2E - 1)\} - (1 + b)E^2\log E. ( E − 1 ) { E 2 + b ( 2 E − 1 )} − ( 1 + b ) E 2 log E .
The trapezoid bound log E < ( E − 1 ) ( E + 1 ) / ( 2 E ) \log E < (E-1)(E+1)/(2E) log E < ( E − 1 ) ( E + 1 ) / ( 2 E ) makes this greater than ( E − 1 ) 2 { ( 1 − b ) E + 2 b } / 2 > 0 (E-1)^2\{(1-b)E+2b\}/2>0 ( E − 1 ) 2 {( 1 − b ) E + 2 b } /2 > 0 .
For clarity the needed logarithm comparisons are entirely rational. For t = ( E − 1 ) / ( E + 1 ) t=(E-1)/(E+1) t = ( E − 1 ) / ( E + 1 ) let
L N ( E ) = 2 ∑ m = 0 N − 1 t 2 m + 1 2 m + 1 , U N ( E ) = L N ( E ) + 2 t 2 N + 1 ( 2 N + 1 ) ( 1 − t 2 ) . L_N(E)=2\sum_{m=0}^{N-1}\frac{t^{2m+1}}{2m+1},\qquad U_N(E)=L_N(E)+\frac{2t^{2N+1}}{(2N+1)(1-t^2)}. L N ( E ) = 2 m = 0 ∑ N − 1 2 m + 1 t 2 m + 1 , U N ( E ) = L N ( E ) + ( 2 N + 1 ) ( 1 − t 2 ) 2 t 2 N + 1 .
The positive series for log E \log E log E gives L N < log E < U N L_N<\log E<U_N L N < log E < U N . At E 0 = 47 / 20 E_0=47/20 E 0 = 47/20 , t = 27 / 67 t=27/67 t = 27/67 and
t > 402985 1000000 , t 2 > 1623969 10000000 , t 3 > 654435 10000000 , t>\frac{402985}{1000000},\qquad t^2>\frac{1623969}{10000000},\qquad t^3>\frac{654435}{10000000}, t > 1000000 402985 , t 2 > 10000000 1623969 , t 3 > 10000000 654435 ,
t 5 > 106278 10000000 , t 7 > 17259 10000000 , t 9 > 2802 10000000 . t^5>\frac{106278}{10000000},\qquad t^7>\frac{17259}{10000000},\qquad t^9>\frac{2802}{10000000}. t 5 > 10000000 106278 , t 7 > 10000000 17259 , t 9 > 10000000 2802 .
Multiplication of the first two bounds successively proves the remaining power bounds. Substitution gives L 5 ( E 0 ) > 854405 / 1000000 > 534 / 625 L_5(E_0)>854405/1000000>534/625 L 5 ( E 0 ) > 854405/1000000 > 534/625 . Consequently
W ( E 0 ) > 2741 5625 , W ( E 0 ) ( 1 + b / E 0 ) > 4566506 6609375 > 69083 100000 , W(E_0)>\frac{2741}{5625},\qquad W(E_0)(1+b/E_0)>\frac{4566506}{6609375}>\frac{69083}{100000}, W ( E 0 ) > 5625 2741 , W ( E 0 ) ( 1 + b / E 0 ) > 6609375 4566506 > 100000 69083 ,
with last difference 55146875 / 660937500000 > 0 55146875/660937500000>0 55146875/660937500000 > 0 . Since A > A t A>A_t A > A t , the two monotonicities prove
E < 47 / 20. (78) E<47/20. \tag*{(78)} E < 47/20. (78)
In particular E < 31 / 10 E<31/10 E < 31/10 , so the original Lemma B.106 gives Q = f − + f + > 0 Q=f_-+f_+>0 Q = f − + f + > 0 . No extension of its domain is required. This Q Q Q is an affine response coefficient, unrelated to the cost c 8 + c 9 c_8+c_9 c 8 + c 9 .
The remaining affine coefficient is positive. Use the notation of (B.166), and put G = Q − ρ f − G=Q-\rho f_- G = Q − ρ f − . Only ρ < 0 \rho<0 ρ < 0 needs further work, since f − < 0 f_-<0 f − < 0 . In this sector the argument preceding (B.179) gives
H = − A ( E − 1 ) ( E 2 − E + 1 ) + 2 E ( E + 1 ) z − 4 E 2 + 6 E − 2 < 0. H=-A(E-1)(E^2-E+1)+2E(E+1)z-4E^2+6E-2<0. H = − A ( E − 1 ) ( E 2 − E + 1 ) + 2 E ( E + 1 ) z − 4 E 2 + 6 E − 2 < 0.
The exact coefficient of τ 2 \tau^2 τ 2 in G G G is A H / ( 2 − A ) AH/(2-A) A H / ( 2 − A ) . The strengthened upper bound on τ 2 \tau^2 τ 2 , compared with 4 / A 2 4/A^2 4/ A 2 , therefore gives
G > 2 E ( 2 J / A − W 0 ) − 4 H 13 A . (79) G>\frac{2}{E}(2J/A-W_0)-\frac{4H}{13A}. \tag*{(79)} G > E 2 ( 2 J / A − W 0 ) − 13 A 4 H . (79)
where
J = ( E + 1 ) ( E 2 + 1 ) z − ( E − 1 ) ( 2 E 2 − E + 3 ) , W 0 = ( E − 1 ) 3 ( E + 1 ) . J=(E+1)(E^2+1)z-(E-1)(2E^2-E+3),\qquad W_0=(E-1)^3(E+1). J = ( E + 1 ) ( E 2 + 1 ) z − ( E − 1 ) ( 2 E 2 − E + 3 ) , W 0 = ( E − 1 ) 3 ( E + 1 ) .
Multiplying the right side of (22.13) by 13 E A / 4 13EA/4 13 E A /4 gives exactly J 12 − a P 11 ( E ) / 2 J_{12}-aP_{11}(E)/2 J 12 − a P 11 ( E ) /2 , where
J 12 = ( E + 1 ) ( 11 E 2 + 13 ) z − ( E − 1 ) ( 22 E 2 − 11 E + 39 ) , J_{12}=(E+1)(11E^2+13)z-(E-1)(22E^2-11E+39), J 12 = ( E + 1 ) ( 11 E 2 + 13 ) z − ( E − 1 ) ( 22 E 2 − 11 E + 39 ) ,
P 11 ( E ) = 11 E 3 − 11 E 2 − 15 E + 13. P_{11}(E)=11E^3-11E^2-15E+13. P 11 ( E ) = 11 E 3 − 11 E 2 − 15 E + 13.
For 1 < E ≤ 3 / 2 1<E\leq3/2 1 < E ≤ 3/2 the weaker original comparison (B.179) already gives G > 0 G>0 G > 0 . In fact A < 2 / E A<2/E A < 2/ E , and
J > ( E − 1 ) 2 ( 2 E 3 + E 2 + 8 E + 1 ) 3 ( E + 1 ) 2 J>\frac{(E-1)^2(2E^3+E^2+8E+1)}{3(E+1)^2} J > 3 ( E + 1 ) 2 ( E − 1 ) 2 ( 2 E 3 + E 2 + 8 E + 1 )
by the lower logarithm L 2 L_2 L 2 . Hence E J > W 0 EJ>W_0 E J > W 0 follows from
P 0 ( E ) = − E 4 − 5 E 3 + 8 E 2 + 7 E + 3 > 0. P_0(E)=-E^4-5E^3+8E^2+7E+3>0. P 0 ( E ) = − E 4 − 5 E 3 + 8 E 2 + 7 E + 3 > 0.
This polynomial is strictly concave on [ 1 , 3 / 2 ] [1,3/2] [ 1 , 3/2 ] , with endpoint values 12 12 12 and 153 / 16 153/16 153/16 . Assume now G ≤ 0 G \le0 G ≤ 0 . Then 3 / 2 < E < 47 / 20 3/2 < E < 47/20 3/2 < E < 47/20 . On this interval P 11 > 0 P_{11} > 0 P 11 > 0 : its value at 3 / 2 3/2 3/2 is 23 / 8 23/8 23/8 and its derivative is positive and increasing. Put t = ( E − 1 ) / ( E + 1 ) < 27 / 67 < 81 / 200 t=(E-1)/(E+1)<27/67<81/200 t = ( E − 1 ) / ( E + 1 ) < 27/67 < 81/200 . Substitution of L 3 L_3 L 3 gives the exact identity
120 ( 1 − t ) 3 { J 12 ( L 3 ) − ( 39 / 80 ) P 11 ( E ) } = P ( t ) , 120(1-t)^3\left\{J_{12}(L_3)-(39/80)P_{11}(E)\right\}=P(t), 120 ( 1 − t ) 3 { J 12 ( L 3 ) − ( 39/80 ) P 11 ( E ) } = P ( t ) ,
P ( t ) = 117 − 363 t + 507 t 2 − 1569 t 3 − 640 t 4 + 6144 t 5 − 384 t 6 + 2304 t 7 . P(t)=117-363t+507t^2-1569t^3-640t^4+6144t^5-384t^6+2304t^7. P ( t ) = 117 − 363 t + 507 t 2 − 1569 t 3 − 640 t 4 + 6144 t 5 − 384 t 6 + 2304 t 7 .
We prove P > 0 P>0 P > 0 on the whole interval [ 0 , 81 / 200 ] [0,81/200] [ 0 , 81/200 ] . Its derivative is
P ′ = − 363 + 1014 t − 4707 t 2 − 2560 t 3 + 30720 t 4 − 2304 t 5 + 16128 t 6 . P'=-363+1014t-4707t^2-2560t^3+30720t^4-2304t^5+16128t^6. P ′ = − 363 + 1014 t − 4707 t 2 − 2560 t 3 + 30720 t 4 − 2304 t 5 + 16128 t 6 .
For 0 ≤ t ≤ 1 / 3 0\le t\le1/3 0 ≤ t ≤ 1/3 , the positive higher terms are at most ( 10240 / 3 + 1792 / 9 ) t 2 (10240/3+1792/9)t^2 ( 10240/3 + 1792/9 ) t 2 ; the remaining quadratic coefficient is negative, so P ′ ≤ − 363 + 1014 / 3 = − 25 P'\le-363+1014/3=-25 P ′ ≤ − 363 + 1014/3 = − 25 . For 1 / 3 ≤ t ≤ 3 / 8 1/3\le t\le3/8 1/3 ≤ t ≤ 3/8 use
30720 t 4 ≤ 4320 t 2 , 16128 t 6 ≤ ( 5103 / 16 ) t 2 , − 2560 t 3 ≤ − ( 2560 / 3 ) t 2 . 30720t^4\le4320t^2,\qquad16128t^6\le(5103/16)t^2,\qquad-2560t^3\le-(2560/3)t^2. 30720 t 4 ≤ 4320 t 2 , 16128 t 6 ≤ ( 5103/16 ) t 2 , − 2560 t 3 ≤ − ( 2560/3 ) t 2 .
Dropping the other negative term gives
P ′ ≤ − 363 + 1014 t − ( 44227 / 48 ) t 2 ≤ 69 / 4 − 44227 / 432 < 0. P'\le-363+1014t-(44227/48)t^2\le69/4-44227/432<0. P ′ ≤ − 363 + 1014 t − ( 44227/48 ) t 2 ≤ 69/4 − 44227/432 < 0.
For 3 / 8 ≤ t ≤ 81 / 200 3/8\le t\le81/200 3/8 ≤ t ≤ 81/200 use
30720 t 4 ≤ ( 629856 / 125 ) t 2 , 16128 t 6 < 434 t 2 , 30720t^4\le(629856/125)t^2,\qquad16128t^6<434t^2, 30720 t 4 ≤ ( 629856/125 ) t 2 , 16128 t 6 < 434 t 2 ,
− 2560 t 3 ≤ − 960 t 2 , − 2304 t 5 ≤ − ( 243 / 2 ) t 2 . -2560t^3\le-960t^2,\qquad-2304t^5\le-(243/2)t^2. − 2560 t 3 ≤ − 960 t 2 , − 2304 t 5 ≤ − ( 243/2 ) t 2 .
Their total quadratic coefficient is less than − 315 -315 − 315 . The resulting quadratic − 363 + 1014 t − 315 t 2 -363+1014t-315t^2 − 363 + 1014 t − 315 t 2 increases on this interval and at 81 / 200 81/200 81/200 equals − 159915 / 40000 < 0 -159915/40000<0 − 159915/40000 < 0 . Thus P ′ < 0 P'<0 P ′ < 0 everywhere. At t = 81 / 200 t=81/200 t = 81/200 the exact rational bounds
507 t 2 > 83.16 , 6144 t 5 > 66.94 , 2304 t 7 > 4.11 , 507t^2>83.16,\qquad6144t^5>66.94,\qquad2304t^7>4.11, 507 t 2 > 83.16 , 6144 t 5 > 66.94 , 2304 t 7 > 4.11 ,
363 t = 147.015 , 1569 t 3 < 104.232 , 640 t 4 < 17.219 , 384 t 6 < 1.695 363t=147.015,\qquad1569t^3<104.232,\qquad640t^4<17.219,\qquad384t^6<1.695 363 t = 147.015 , 1569 t 3 < 104.232 , 640 t 4 < 17.219 , 384 t 6 < 1.695
prove P ( 81 / 200 ) > 1.049 > 0 P(81/200)>1.049>0 P ( 81/200 ) > 1.049 > 0 ; all terminating decimals here are exact rationals. This proves P > 0 P>0 P > 0 throughout. Since z > L 3 z>L_3 z > L 3 and its coefficient in J 12 J_{12} J 12 is positive, J 12 > ( 39 / 80 ) P 11 J_{12}>(39/80)P_{11} J 12 > ( 39/80 ) P 11 . From G ≤ 0 G\le0 G ≤ 0 and the strict lower comparison (79) it follows that
a > 2 J 12 P 11 > 39 / 40. (80) a>\frac{2J_{12}}{P_{11}}>39/40. \tag*{(80)} a > P 11 2 J 12 > 39/40. (80)
The denominator was shown positive before this division.
The forced value of a a a contradicts the cost. At fixed a = A ( E − 1 ) a=A(E-1) a = A ( E − 1 ) put e = E − 1 e=E-1 e = E − 1 and
D a = 44 e 2 + 4 a e − 13 a 2 , B a = a 2 − 2 a log E + 4 e 2 / E . D_a=44e^2+4ae-13a^2,\qquad B_a=a^2-2a\log E+4e^2/E. D a = 44 e 2 + 4 a e − 13 a 2 , B a = a 2 − 2 a log E + 4 e 2 / E .
Then
C 12 ( a / e , E ) = a 2 + 13 a 2 B a D a . C_{12}(a/e,E)=\frac{a}{2}+\frac{13a}{2}\frac{B_a}{D_a}. C 12 ( a / e , E ) = 2 a + 2 13 a D a B a .
For 39 / 40 ≤ a < 7 / 5 39/40\le a<7/5 39/40 ≤ a < 7/5 and E ≥ 3 / 2 E\ge3/2 E ≥ 3/2 , this decreases strictly in E E E wherever D a > 0 D_a>0 D a > 0 . Here is the derivative comparison. For F = D a ′ B a − D a B a ′ F=D'_aB_a-D_aB'_a F = D a ′ B a − D a B a ′ , the logarithm trapezoid bound gives
E 2 F 4 ≥ − 11 a 3 2 + ( 48 a 2 − 9 a 3 / 2 ) e + ( a 3 + 56 a 2 − 26 a ) e 2 + ( 21 a 2 − 44 a ) e 3 + ( 44 − 22 a ) e 4 . \frac{E^2F}{4}\ge-\frac{11a^3}{2}+(48a^2-9a^3/2)e+(a^3+56a^2-26a)e^2
+(21a^2-44a)e^3+(44-22a)e^4. 4 E 2 F ≥ − 2 11 a 3 + ( 48 a 2 − 9 a 3 /2 ) e + ( a 3 + 56 a 2 − 26 a ) e 2 + ( 21 a 2 − 44 a ) e 3 + ( 44 − 22 a ) e 4 .
The first two terms are at least ( 263 / 20 ) a 2 > 0 (263/20)a^2>0 ( 263/20 ) a 2 > 0 , using e ≥ 1 / 2 e\ge1/2 e ≥ 1/2 and a ≤ 7 / 5 a\le7/5 a ≤ 7/5 . On the same a a a interval,
a 3 + 56 a 2 − 26 a > 28 , a ( 44 − 21 a ) ≤ 484 / 21 < 24 , 44 − 22 a ≥ 66 / 5. a^3+56a^2-26a>28,\qquad a(44-21a)\le484/21<24,\qquad44-22a\ge66/5. a 3 + 56 a 2 − 26 a > 28 , a ( 44 − 21 a ) ≤ 484/21 < 24 , 44 − 22 a ≥ 66/5.
The remaining bracket therefore exceeds 28 − 24 e + ( 66 / 5 ) e 2 > 0 28-24e+(66/5)e^2>0 28 − 24 e + ( 66/5 ) e 2 > 0 , whose negative-discriminant reserve is 4512 / 5 4512/5 4512/5 . Hence F > 0 F>0 F > 0 , as asserted.
First decrease a a a to 39 / 40 39/40 39/40 at fixed E E E , using the already proved A A A monotonicity. This preserves 0 < A < 2 / E 0 < A < 2/E 0 < A < 2/ E . Then increase E E E to 47 / 20 47/20 47/20 with a = 39 / 40 a = 39/40 a = 39/40 fixed; both A = a / ( E − 1 ) A = a/(E-1) A = a / ( E − 1 ) and A E AE A E decrease, so the denominator remains positive. Thus
c > C 12 ( A , E ) > C 12 ( ( 39 / 40 ) / ( E − 1 ) , E ) > C 12 ( 13 / 18 , 47 / 20 ) . c > C_{12}(A,E) > C_{12}\left((39/40)/(E-1),E\right) > C_{12}(13/18,47/20). c > C 12 ( A , E ) > C 12 ( ( 39/40 ) / ( E − 1 ) , E ) > C 12 ( 13/18 , 47/20 ) .
For the upper logarithm at E 0 E_0 E 0 , use t = 27 / 67 < 403 / 1000 t = 27/67 < 403/1000 t = 27/67 < 403/1000 , t 3 < 0.065451 t^3 < 0.065451 t 3 < 0.065451 , t 5 < 0.010631 t^5 < 0.010631 t 5 < 0.010631 , t 7 < 0.001727 t^7 < 0.001727 t 7 < 0.001727 and 1 − t 2 > 0.837 1-t^2 > 0.837 1 − t 2 > 0.837 . Then
U 3 ( E 0 ) < 0.806 + 0.043634 + 0.0042524 + 0.0006 < 171 / 200. U_3(E_0) < 0.806 + 0.043634 + 0.0042524 + 0.0006 < 171/200. U 3 ( E 0 ) < 0.806 + 0.043634 + 0.0042524 + 0.0006 < 171/200.
Since the cost decreases in its logarithm argument, direct rational substitution gives
C 12 ( 13 / 18 , 47 / 20 ) > 39 80 + 454753143 2198754000 > 69083 100000 . C_{12}(13/18,47/20) > \frac{39}{80} + \frac{454753143}{2198754000} > \frac{69083}{100000}. C 12 ( 13/18 , 47/20 ) > 80 39 + 2198754000 454753143 > 100000 69083 .
with last difference 768049218000 / 219875400000000 > 0 768049218000/219875400000000 > 0 768049218000/219875400000000 > 0 . This contradicts the assumed cost. Therefore G > 0 G > 0 G > 0 whenever ρ < 0 \rho< 0 ρ < 0 .
Finally retain the original affine exit row (B.181). For completeness set X − = − k 4 q L > 0 X_- = -k^4q_L > 0 X − = − k 4 q L > 0 , Y + = k 4 q R ≥ 0 Y_+ = k^4q_R \ge0 Y + = k 4 q R ≥ 0 , with m 0 , n , a 0 > 0 m_0,n,a_0 > 0 m 0 , n , a 0 > 0 as in (B.166) and (B.169). Then
k 3 ( q R ′ − k q R ) = f + + ( ρ − 1 ) m 0 Y + + n X − . k^3(q'_R-kq_R)=f_+ + (\rho-1)m_0Y_+ + nX_-. k 3 ( q R ′ − k q R ) = f + + ( ρ − 1 ) m 0 Y + + n X − .
For ρ ≥ 1 \rho\ge1 ρ ≥ 1 , f + = Q − f − > 0 f_+ = Q-f_- > 0 f + = Q − f − > 0 proves positivity. For ρ < 1 \rho< 1 ρ < 1 , put ε = − f − − a 0 X − − m 0 Y + > 0 \varepsilon= -f_- - a_0X_- - m_0Y_+ > 0 ε = − f − − a 0 X − − m 0 Y + > 0 . The same expression is exactly
G + { n + ( 1 − ρ ) a 0 } X − + ( 1 − ρ ) ε > 0. G + \{n+(1-\rho)a_0\}X_- + (1-\rho)\varepsilon> 0. G + { n + ( 1 − ρ ) a 0 } X − + ( 1 − ρ ) ε > 0.
For 0 ≤ ρ < 1 0 \le\rho< 1 0 ≤ ρ < 1 use G = Q − ρ f − > 0 G = Q-\rho f_- > 0 G = Q − ρ f − > 0 ; for ρ < 0 \rho< 0 ρ < 0 use the preceding argument. This proves the strict exit assertion in every case, including ρ = 0 \rho= 0 ρ = 0 . □
The complete physical partitions and reflection.
Theorem 22.7. Neither R 10 R10 R 10 nor R 13 R13 R 13 is the positive-maximal-arc word of an original global minimizer.
Proof. For R 10 R10 R 10 retain the following unconditional physical charges:
K = c 7 + c 8 + c 9 > 2 / 5 , K = c_7+c_8+c_9 > 2/5, K = c 7 + c 8 + c 9 > 2/5 , J = c 10 > 63 / 500 , T = c 11 + c 0 + c 1 > 61 / 200. J = c_{10} > 63/500,\qquad T = c_{11}+c_0+c_1 > 61/200. J = c 10 > 63/500 , T = c 11 + c 0 + c 1 > 61/200. The sources and orientations are as follows. Descending ( 11 , 10 , 9 , 8 , 7 , 6 , 5 ) (11,10,9,8,7,6,5) ( 11 , 10 , 9 , 8 , 7 , 6 , 5 ) under j ↦ − j j \mapsto-j j ↦ − j is 0120212. Its actual 202 core is ( 9 , 8 , 7 ) (9,8,7) ( 9 , 8 , 7 ) , its caps are 10, 6, and its return is 5. Its forward prefix supplies the negative entrance and entrance circle, while the true return supplies the terminal circle. Proposition 10.1 gives K > 2 / 5 K > 2/5 K > 2/5 . The complete nine-source [ 6 , 14 ] = 210120102 [6,14] = 210120102 [ 6 , 14 ] = 210120102 becomes 102012021 under phase + 2 +2 + 2 , retaining both actual outward forward-four witnesses. Theorem C.6 gives J > 63 / 500 J > 63/500 J > 63/500 without a HIGH predicate. The source of the ordinary T T T bound is that of Lemma 22.3 .
There are two exhaustive cases, according as c 3 ≤ 13 / 20 c_3 \le13/20 c 3 ≤ 13/20 or c 3 > 13 / 20 c_3 > 13/20 c 3 > 13/20 .
Suppose first c 3 ≤ 13 / 20 c_3 \le13/20 c 3 ≤ 13/20 , including equality. Lemma 22.3 gives T > 2 / 5 T > 2/5 T > 2/5 . The complete proper source [ − 1 , 6 ] = 20210212 [-1,6] = 20210212 [ − 1 , 6 ] = 20210212 charges its local positions 3, 4, namely native arcs 2, 3. Theorem E.2 gives c 2 + c 3 > 13 / 20 c_2+c_3 > 13/20 c 2 + c 3 > 13/20 , on a genuine containing interval obtained by a cut inside arc − 2 -2 − 2 . The original refined one-return Theorem B.16 , on the actual source [ 3 , 9 ] = 0212101 [3,9] = 0212101 [ 3 , 9 ] = 0212101 , gives
c 4 + c 5 + c 6 + c 7 > E ∗ > 237 500 , E ∗ − 237 500 = 2033896299 8759490110000 > 0. c_4+c_5+c_6+c_7 > E_* > \frac{237}{500},\qquad E_*-\frac{237}{500}=\frac{2033896299}{8759490110000}>0. c 4 + c 5 + c 6 + c 7 > E ∗ > 500 237 , E ∗ − 500 237 = 8759490110000 2033896299 > 0. Together with Lemma 22.4 , the disjoint masks
{ 2 , 3 } , { 4 , 5 , 6 , 7 } , { 8 , 9 } , { 10 } , { 11 , 0 , 1 } \{2,3\},\qquad\{4,5,6,7\},\qquad\{8,9\},\qquad\{10\},\qquad\{11,0,1\} { 2 , 3 } , { 4 , 5 , 6 , 7 } , { 8 , 9 } , { 10 } , { 11 , 0 , 1 }
partition all twelve physical arcs. In particular shared arc 7 is paid only by the four-core. Consequently
C > 13 20 + E ∗ + 31 200 + 63 500 + 2 5 > 361 200 = C ∗ + 17 100000 , C > \frac{13}{20} + E_* + \frac{31}{200} + \frac{63}{500} + \frac{2}{5} > \frac{361}{200} = C_* + \frac{17}{100000}, C > 20 13 + E ∗ + 200 31 + 500 63 + 5 2 > 200 361 = C ∗ + 100000 17 ,
a contradiction. The exact E ∗ E_* E ∗ is retained; the last inequality merely displays a smaller rational margin. No separate expensive c 2 c_2 c 2 branch or weak-five charge is needed.
If c 3 > 13 / 20 c_3 > 13/20 c 3 > 13/20 , the same unconditional rows give
c 2 + c 4 + c 5 + c 6 < 32383 100000 < 13 20 . c_2 + c_4 + c_5 + c_6 < \frac{32383}{100000} < \frac{13}{20}. c 2 + c 4 + c 5 + c 6 < 100000 32383 < 20 13 .
Thus c 2 < 13 / 20 c_2 < 13/20 c 2 < 13/20 from unconditional estimates, and Lemma 22.2 gives c 5 + c 6 > 283 / 1000 c_5 + c_6 > 283/1000 c 5 + c 6 > 283/1000 . The resulting disjoint budgets are
c 2 + c 4 < 4083 100000 , 13 20 < c 3 < 69083 100000 . c_2 + c_4 < \frac{4083}{100000}, \qquad\frac{13}{20} < c_3 < \frac{69083}{100000}. c 2 + c 4 < 100000 4083 , 20 13 < c 3 < 100000 69083 .
Lemma 22.5 gives the actual own height h E > 1 / 12 h_E > 1/12 h E > 1/12 . The reverse prefix on ( 6 , 5 , 4 , 3 , 2 , 1 ) (6,5,4,3,2,1) ( 6 , 5 , 4 , 3 , 2 , 1 ) under j ↦ 1 − j j \mapsto1-j j ↦ 1 − j is 202102, with centre 3. Its initial 202 middle mean is exactly the transformed own mean leaf 5, not the one-sided mean on arc 3 of the incoming construction. Its positive outer root U U U satisfies U > h E U > h_E U > h E , and its following positive mode-2 tail increases u u u . Thus this actual prefix has u L > U > h E > 1 / 12 u_L > U > h_E > 1/12 u L > U > h E > 1/12 . Lemma 22.6 restores its strict exit sign. Lemma 22.3, on the genuine period [ − 3 , 8 ] [-3,8] [ − 3 , 8 ] , then gives T > 2 / 5 T > 2/5 T > 2/5 even though c 3 > 13 / 20 c_3 > 13/20 c 3 > 13/20 . The disjoint masks are now
{ 3 } , { 5 , 6 } , { 7 , 8 , 9 } , { 10 } , { 11 , 0 , 1 } , \{3\}, \quad\{5,6\}, \quad\{7,8,9\}, \quad\{10\}, \quad\{11,0,1\}, { 3 } , { 5 , 6 } , { 7 , 8 , 9 } , { 10 } , { 11 , 0 , 1 } ,
leaving only positive arcs 2, 4 uncharged. Consequently
C > 13 20 + 283 1000 + 2 5 + 63 500 + 2 5 = 1859 1000 = C ∗ + 5417 100000 . C > \frac{13}{20} + \frac{283}{1000} + \frac{2}{5} + \frac{63}{500} + \frac{2}{5} = \frac{1859}{1000} = C_* + \frac{5417}{100000}. C > 20 13 + 1000 283 + 5 2 + 500 63 + 5 2 = 1000 1859 = C ∗ + 100000 5417 .
This excludes both cases. Internal identities coupling means or matrices used the full common lifts specified above. The final additions concern only physical costs, so different scalar estimates may be established on different genuine containing periods.
To transfer the entire conclusion to R 13 R13 R 13 , use the lawful whole reflected-time transformation and phase with label map
j i ′ = 1 − j 5 − i , r ↦ i = 5 − r ( m o d 12 ) . j'_i = 1 - j_{5-i}, \qquad r \mapsto i = 5-r \pmod{12}. j i ′ = 1 − j 5 − i , r ↦ i = 5 − r ( mod 12 ) .
Its native word is 021021201210. Its twist is still + 1 +1 + 1 , since
j i + 12 ′ = 1 − j 5 − i − 12 = j i ′ + 1 ( m o d 3 ) . j'_{i+12} = 1 - j_{5-i-12} = j'_i + 1 \pmod{3}. j i + 12 ′ = 1 − j 5 − i − 12 = j i ′ + 1 ( mod 3 ) .
The transformation carries the entire path, full state and frame endpoints, real winding, both state covectors, matrix adjoint, normal multiplier, and every positive maximal support arc. It neither presumes reflection symmetry of a trajectory nor identifies the two cyclic words by a reflection quotient. The predicates c 2 , c 3 c_2,c_3 c 2 , c 3 of R 10 R10 R 10 become respectively c 3 ′ , c 2 ′ c'_3,c'_2 c 3 ′ , c 2 ′ of R 13 R13 R 13 .
For c 2 ′ ≤ 13 / 20 c'_2 \le13/20 c 2 ′ ≤ 13/20 , the reflected first-case masks are
{ 2 , 3 } , { 0 , 1 , 10 , 11 } , { 8 , 9 } , { 7 } , { 4 , 5 , 6 } . \{2,3\}, \quad\{0,1,10,11\}, \quad\{8,9\}, \quad\{7\}, \quad\{4,5,6\}. { 2 , 3 } , { 0 , 1 , 10 , 11 } , { 8 , 9 } , { 7 } , { 4 , 5 , 6 } .
They partition all twelve arcs and have total greater than 361 / 200 361/200 361/200 . For c 2 ′ > 13 / 20 c'_2 > 13/20 c 2 ′ > 13/20 , the reflected second case has masks
{ 2 } , { 0 , 11 } , { 8 , 9 , 10 } , { 7 } , { 4 , 5 , 6 } , \{2\}, \quad\{0,11\}, \quad\{8,9,10\}, \quad\{7\}, \quad\{4,5,6\}, { 2 } , { 0 , 11 } , { 8 , 9 , 10 } , { 7 } , { 4 , 5 , 6 } ,
omits only arcs 1, 3, and has total greater than $1859/1000. The common nine-source [ 3 , 11 ] [3,11] [ 3 , 11 ] becomes the complete reversed source [ − 6 , 2 ] [-6,2] [ − 6 , 2 ] , and the eleven-source [ − 1 , 9 ] [-1,9] [ − 1 , 9 ] becomes [ − 4 , 6 ] [-4,6] [ − 4 , 6 ] .
Their genuine containing periods transform as whole periods; source length, cap completeness and calibration are preserved at the seam. The same two strict rational margins therefore exclude R13 as well. □
A bilateral-pair exclusion of R11 and R12
Throughout this section the trajectory is an original global minimizer, with positive maximal arcs and the original group and state endpoints and real winding. Its cost satisfies C < 180483 100000 C < \frac{180483}{100000} C < 100000 180483 . The native word of R11 is 021021202120, with j i + 12 = j i + 1 j_{i+12}=j_i+1 j i + 12 = j i + 1 modulo three. An interval [ i , k ] [i,k] [ i , k ] denotes the complete arcs i , … , k i,\ldots,k i , … , k on this unfolded word. Write
P = c 8 + c 9 + c 10 , Q = c 4 + c 5 + c 6 , R = c 11 + c 0 + c 1 . (81) P=c_8+c_9+c_{10}, \qquad Q=c_4+c_5+c_6, \qquad R=c_{11}+c_0+c_1. \tag*{(81)} P = c 8 + c 9 + c 10 , Q = c 4 + c 5 + c 6 , R = c 11 + c 0 + c 1 . (81)
These triples are disjoint, and C = c 2 + c 3 + c 7 + P + Q + R C=c_2+c_3+c_7+P+Q+R C = c 2 + c 3 + c 7 + P + Q + R .
Physical sources and calibration
Lemma 22.8. The genuine switch-cut period [ 3 , 14 ] [3,14] [ 3 , 14 ] admits a jointly normalized full lift with
0 < Γ < H ∗ < 1470471 2000000 . 0<\Gamma<H_*<\frac{1470471}{2000000}. 0 < Γ < H ∗ < 2000000 1470471 . No improved numerical bound on the remaining three-arc charge is needed for this qualification.
Proof. The source [ 1 , 5 ] = 21021 [1,5]=21021 [ 1 , 5 ] = 21021 becomes 10210 under whole phase + 2 +2 + 2 . The independently qualified weak-five theorem B.47 gives V : = c 2 + c 3 + c 4 ≥ 3 5 V:=c_2+c_3+c_4\ge\frac{3}{5} V := c 2 + c 3 + c 4 ≥ 5 3 . Put W = P + R W=P+R W = P + R . Positivity of c 5 + c 6 + c 7 c_5+c_6+c_7 c 5 + c 6 + c 7 alone therefore gives
W < C − 3 5 < 180483 100000 − 3 5 < 5 4 . W<C-\frac{3}{5}<\frac{180483}{100000}-\frac{3}{5}<\frac{5}{4}. W < C − 5 3 < 100000 180483 − 5 3 < 4 5 .
The full source [ 7 , 14 ] = 02120102 [7,14]=02120102 [ 7 , 14 ] = 02120102 is retained in [ 3 , 14 ] [3,14] [ 3 , 14 ] . This interval has exactly twelve complete arcs and is a fundamental period cut at a switch. Reanchoring preserves the original endpoints and real winding.
The actual capped source [ 7 , 11 ] = 02120 [7,11]=02120 [ 7 , 11 ] = 02120 , read backwards under j ↦ 1 − j j\mapsto1-j j ↦ 1 − j , is 12021. Its two roots and two caps establish normality before the matrix and both state covectors are jointly normalized. The central forward-four [ 9 , 12 ] = 1201 [9,12]=1201 [ 9 , 12 ] = 1201 in W W W supplies both negative entrances, without harmonic calibration. Choose the cheaper actual W W W tail on this same selected lift. Its cost satisfies T ≤ W / 2 < 5 / 8 T\le W/2<5/8 T ≤ W /2 < 5/8 . At its terminal point the signed strip of Lemma B.11 gives
3 2 < K < 2 , m < − 1 2 , 0 < L < K σ < 2 σ < T . \frac{3}{2}<K<2,\qquad m<-\frac{1}{2},\qquad0<L<K\sigma<2\sigma<T. 2 3 < K < 2 , m < − 2 1 , 0 < L < K σ < 2 σ < T .
Consequently
Γ = − L 2 − K m > 3 4 − T 2 > 23 64 > 0. \Gamma=-L^2-Km>\frac{3}{4}-T^2>\frac{23}{64}>0. Γ = − L 2 − K m > 4 3 − T 2 > 64 23 > 0.
The positive actual invariant K K K supplies the future sign in Corollary A.13 . The future-sheet and harmonic results A.10 and A.12 now apply to this same genuine period. No costate is continued periodically across a seam. □
Lemma 22.9. The physical charges in (81) satisfy
P > 2 5 , P + c 7 > 91 200 , Q , R > 61 200 . P>\frac{2}{5},\qquad P+c_7>\frac{91}{200},\qquad Q,R>\frac{61}{200}. P > 5 2 , P + c 7 > 200 91 , Q , R > 200 61 . Proof. On the full reflected-time chart j ↦ 1 − j j\mapsto1-j j ↦ 1 − j , the actual chain is
( 10 , 9 , 8 ) ∣ 7 ∣ ( 6 , 5 , 4 ) = 202 ∣ 1 ∣ 202. (10,9,8)\mid7\mid(6,5,4)=202\mid1\mid202. ( 10 , 9 , 8 ) ∣ 7 ∣ ( 6 , 5 , 4 ) = 202 ∣ 1 ∣ 202. The first triple, of cost P P P , has caps 11 and 7 and return 6. Its entrance circle is furnished by [ 9 , 12 ] = 1201 [9,12]=1201 [ 9 , 12 ] = 1201 , and its terminal circle by its return. The second triple, of cost Q Q Q , has
caps 7 and 3 and its negative entrance and entrance circle from the seven-source (10, …, 4), by Lemma B.95. The complete support [ref-3 , ref-12 ] and the full W W W source both lie in the period [ref-3 , ref-14 ]. Apply Lemma 22.8 on that period.
Proposition 10.1 gives P > 2 / 5 P > 2/5 P > 2/5 . The complete reversed source (11, 10, 9, 8, 7, 6) under 1 − j 1-j 1 − j is 120212. Its charged triple is P P P , its following complete mode-one bridge is arc 7, and its actual return is arc 6. The bridge ends at the negative entrance of Q Q Q , as proved by the retained H7 source. Thus Theorem B.101, on this same qualified interval, gives P + c 7 > 91 / 200 P+c_7 > 91/200 P + c 7 > 91/200 . Lemma B.75 gives Q > 61 / 200 Q > 61/200 Q > 61/200 and, on the other actual capped W W W triple, R > 61 / 200 R > 61/200 R > 61/200 .
The two cheap centre implications.
Lemma 22.10. For the actual R11 trajectory,
c 2 ≤ 13 20 ⟹ Q > 2 5 , c 3 ≤ 13 20 ⟹ R > 2 5 . c_2 \le\frac{13}{20} \Longrightarrow Q > \frac{2}{5}, \qquad c_3 \le\frac{13}{20} \Longrightarrow R > \frac{2}{5}. c 2 ≤ 20 13 ⟹ Q > 5 2 , c 3 ≤ 20 13 ⟹ R > 5 2 . Proof. Suppose first c 2 ≤ 13 / 20 c_2 \le13/20 c 2 ≤ 13/20 . The actual prefix [ − 1 , 4 ] = 202102 [-1,4] = 202102 [ − 1 , 4 ] = 202102 has centre 2. By B.107 and B.184, its outgoing native mode-zero arc 3 starts at z < 0 z < 0 z < 0 with χ ( z ) ≥ 0 \chi(z) \ge0 χ ( z ) ≥ 0 and χ t ( z ) > 0 \chi_t(z) > 0 χ t ( z ) > 0 . Write μ c = 2 K c − 3 \mu_c = 2K_c - 3 μ c = 2 K c − 3 for its binary invariant. Since χ s = μ c − u 2 \chi_s = \mu_c-u^2 χ s = μ c − u 2 , we have μ c > z 2 \mu_c > z^2 μ c > z 2 . At its actual binary tie 3 ∣ 4 3|4 3∣4 , let the horizontal coordinate be r 0 r_0 r 0 . The switching function first increases and can reach zero only after u > μ c u > \sqrt{\mu_c} u > μ c , so
0 < μ c < r 0 < 1. (82) 0 < \sqrt{\mu_c} < r_0 < 1. \tag*{(82)} 0 < μ c < r 0 < 1. (82) The exact one-sided relation is ∫ z r 0 ( u 2 − μ c ) ( 1 + u ) − 2 d u ≥ 0 \int_z^{r_0}(u^2-\mu_c)(1+u)^{-2}\,du \ge0 ∫ z r 0 ( u 2 − μ c ) ( 1 + u ) − 2 d u ≥ 0 . In particular μ c \mu_c μ c has not been identified with a two-zero mean.
The reflected Q Q Q triple (6, 5, 4) already has its entrance circle and caps 7 and 3. Its terminal junction is the same physical 3 ∣ 4 3|4 3∣4 . The full state involution between its chart and the native chart is
T ( w ) = − w ‾ − 3 1 − w ‾ , ℜ T ( b + i v ) = b 2 + 2 b − 3 + v 2 ( 1 − b ) 2 + v 2 = r 0 > 0. \mathcal{T}(w)=\frac{-\overline{w}-3}{1-\overline{w}}, \qquad\Re\mathcal{T}(b+iv)=\frac{b^2+2b-3+v^2}{(1-b)^2+v^2}=r_0>0. T ( w ) = 1 − w − w − 3 , ℜ T ( b + i v ) = ( 1 − b ) 2 + v 2 b 2 + 2 b − 3 + v 2 = r 0 > 0. Thus Q Q Q has its terminal circle as well. The union of this H7 source and the prefix is [ − 1 , 10 ] [-1,10] [ − 1 , 10 ] , exactly one genuine twelve-arc period. On this selected genuine period, the capped-three-core argument proves normality before joint normalization. Its actual loss and determinant satisfy Γ = 3 / 4 − L / 4 \Gamma=3/4-\mathcal{L}/4 Γ = 3/4 − L /4 . If L < 59 / 1000 \mathcal{L}<59/1000 L < 59/1000 , then Γ > 2941 / 4000 > 0 \Gamma>2941/4000>0 Γ > 2941/4000 > 0 . The actual positive binary invariant K = ( 3 + μ ) / 2 K=(3+\mu)/2 K = ( 3 + μ ) /2 supplies the future sheet, and the harmonic ceiling contradicts this determinant bound. Thus L ≥ 59 / 1000 \mathcal{L}\ge59/1000 L ≥ 59/1000 is established on this lift independently of the calibration on [ref-3 , ref-14 ]. The circle-only transfer in the proof of Lemma C.7 now gives Q > 2 / 5 Q>2/5 Q > 2/5 .
If c 3 ≤ 13 / 20 c_3\le13/20 c 3 ≤ 13/20 , instead use descending native arcs (6, …, 1) under full reflected time reversal j ↦ 1 − j j\mapsto1-j j ↦ 1 − j . The centre is 3; the outgoing reversed native arc 2 gives a positive coordinate at 1 ∣ 2 1|2 1∣2 . The same involution supplies the terminal circle for [ − 1 , 1 ] = 202 [-1,1]=202 [ − 1 , 1 ] = 202 , which represents R R R . Its capped context is [ − 2 , 2 ] = 12021 [-2,2]=12021 [ − 2 , 2 ] = 12021 , and the forward-four [ − 3 , 0 ] = 0120 [-3,0]=0120 [ − 3 , 0 ] = 0120 gives its entrance circle. This full union [ − 3 , 6 ] [-3,6] [ − 3 , 6 ] has ten arcs. On its own selected normal lift the same loss-obstruction, future-sheet and circle-only argument gives R > 2 / 5 R>2/5 R > 2/5 .
The disjoint partition.
Theorem 22.11. Neither R11 nor R12 occurs as the positive-maximal-arc word of an original global minimizer.
Proof. First suppose c 2 > 13 / 20 c_2 > 13/20 c 2 > 13/20 . Lemma 22.9 and (81) give
c 3 < C ∗ − 13 20 − 91 200 − 2 61 200 − 8983 100000 = C ∗ + 13 20 . c_3 < C_* - \frac{13}{20} - \frac{91}{200} - 2\frac{61}{200} - \frac{8983}{100000} = C_* + \frac{13}{20}. c 3 < C ∗ − 20 13 − 200 91 − 2 200 61 − 100000 8983 = C ∗ + 20 13 .
Thus Lemma 22.10 gives R > 2 / 5 R > 2/5 R > 2/5 . The disjoint masks { 2 } \{2\} { 2 } , { 7 , 8 , 9 , 10 } \{7,8,9,10\} { 7 , 8 , 9 , 10 } , { 4 , 5 , 6 } \{4,5,6\} { 4 , 5 , 6 } and { 11 , 0 , 1 } \{11,0,1\} { 11 , 0 , 1 } now imply
C > 13 20 + 91 200 + 61 200 + 2 5 = 181 100 = C ∗ + 517 100000 , C > \frac{13}{20} + \frac{91}{200} + \frac{61}{200} + \frac{2}{5} = \frac{181}{100} = C_* + \frac{517}{100000}, C > 20 13 + 200 91 + 200 61 + 5 2 = 100 181 = C ∗ + 100000 517 ,
a contradiction.
It remains c 2 ≤ 13 / 20 c_2 \le13/20 c 2 ≤ 13/20 , including equality. Then Lemma 22.10 gives Q > 2 / 5 Q > 2/5 Q > 2/5 . The unconditional bridge payment and R > 61 / 200 R > 61/200 R > 61/200 force
c 3 < C ∗ − 91 200 − 2 5 − 61 200 = 64483 100000 < 13 20 . c_3 < C_* - \frac{91}{200} - \frac{2}{5} - \frac{61}{200} = \frac{64483}{100000} < \frac{13}{20}. c 3 < C ∗ − 200 91 − 5 2 − 200 61 = 100000 64483 < 20 13 .
Consequently P , Q , R > 2 / 5 P,Q,R > 2/5 P , Q , R > 2/5 . The complete proper source [ − 1 , 6 ] = 20210212 [-1,6] = 20210212 [ − 1 , 6 ] = 20210212 has eight arcs; its local charged positions 3, 4 are exactly native arcs 2, 3. Theorem E.2 applies on a genuine containing interval and gives c 2 + c 3 > 13 / 20 c_2+c_3 > 13/20 c 2 + c 3 > 13/20 . These are scalar costs on the same physical trajectory; their full lifts are not identified. The resulting disjoint partition gives
C > c 2 + c 3 + P + Q + R > 13 20 + 3 2 5 = 37 20 = C ∗ + 4517 100000 , C > c_2+c_3+P+Q+R > \frac{13}{20}+3\frac{2}{5} = \frac{37}{20} = C_* + \frac{4517}{100000}, C > c 2 + c 3 + P + Q + R > 20 13 + 3 5 2 = 20 37 = C ∗ + 100000 4517 ,
again a contradiction. Arc 7 is left unpaid in this final partition. No bridge or overlapping triple is charged twice.
Finally the full reflected reanchoring
j i ′ = 1 − j 5 − i j_i' = 1-j_{5-i} j i ′ = 1 − j 5 − i
takes R11 to R12, whose native word is 021021201020, and preserves the + 1 +1 + 1 twist. It carries the state, group, matrix, both state covectors, endpoints and real winding. The physical index map is r ↦ 5 − r r \mapsto5-r r ↦ 5 − r modulo twelve. The centres 2 and 3 are interchanged, and the three charged triples become
P ′ = { 7 , 8 , 9 } , Q ′ = { 11 , 0 , 1 } , R ′ = { 4 , 5 , 6 } . P' = \{7,8,9\}, \qquad Q' = \{11,0,1\}, \qquad R' = \{4,5,6\}. P ′ = { 7 , 8 , 9 } , Q ′ = { 11 , 0 , 1 } , R ′ = { 4 , 5 , 6 } .
The bridge 7 maps to arc 10, and the pair { 2 , 3 } \{2,3\} { 2 , 3 } is retained. All strict costs and surpluses therefore transport to R12. □
Theorem 22.12. An original minimizing word in the length-five-negative-run family must have type R02 or R04.
Proof. The family consists of the seven listed words. The preceding source-qualified prefix and bilateral comparisons exclude R08 and R10–R13. The remaining two words are R02 and R04. □
A single negative run of length seven, eight, or nine
The remaining mixed group in Theorem 20.3 has one negative run of length 7, 8 or 9; every other negative run has length 2. Its three one-positive-run and four two-positive-run words are specified in Table 4. Reflection is used only by transporting the full geometric construction, not by identifying cyclic types.
Theorem 23.1 (Exclusion of the long-negative-run family). No original global minimizer belongs to the seven-word long-negative-run family of Theorem 20.3 .
We exclude the first two by resolving actual rooted responses, the next two by paying one expensive centre and a disjoint complement, and the last three by a centre alternative with a separately chosen weak-five payment. This last choice depends on the successful centre; it does not turn a maximum into two simultaneous bounds.
Throughout this section, the native controls are reconstructed from the printed increment word by j 0 = 0 j_0=0 j 0 = 0 and j i + B = j i + 1 j_{i+B}=j_i+1 j i + B = j i + 1 modulo three. A source a : b a:b a : b lists all indices from a a a to b b b in their indicated order. On a decreasing source the displayed label reflection accompanies full time reversal. Physical costs alone are reduced modulo B B B . All the sources below are complete and proper. The full geometric endpoints, real winding and selected normal lifts are those of § 3.1 .
Rooted alternatives and their actual continuations
Let U , V U,V U , V be the horizontal endpoints of an actual whole mode-2 tail rooted at a 0 → 2 0 \to2 0 → 2 switch in its lawful chart. Its rooted response is
Ψ ( U , V ) = 2 ( 1 − U ) log 1 − U 1 − V − ( V − U ) ( 2 + V − U ) . \Psi(U,V)=2(1-U)\log\frac{1-U}{1-V}-(V-U)(2+V-U). Ψ ( U , V ) = 2 ( 1 − U ) log 1 − V 1 − U − ( V − U ) ( 2 + V − U ) .
This is the full-duration response of (xref)(B.198), not an assigned adjoint at a comparison endpoint. Theorem B.117 says that a complete 21021 context forces a positive response in at least one of the two lawful orientations at its middle connector. The two alternatives must first belong to the stipulated common full lift. A short rigid capped source supplies such a lift by Corollary B.115; its covector twist is proved there, not assumed.
Lemma 23.2 (A blocked continuation). A complete actual 02101 source on a qualified full normal maximizing lift has nonpositive rooted response at its first 0 → 2 0 \to2 0 → 2 switch.
Proof. The response identity (xref)(C.21) converts a positive response into the strict actual ingress of (xref)(B.168). The retained terminal 101 gives the opposite reflected-tail inequality on this same lift, contradicting (xref)(A.16). The full binary derivation is in Section C.8.1. Equality of the response is retained. □
Lemma 23.3 (Two successful continuations). A complete 021020 source with positive rooted response has central cost greater than 299 / 250 299/250 299/250 . A complete 02102101 source with positive rooted response has central cost greater than 13 / 20 13/20 13/20 . Both statements use a qualified common full lift, and the charged arc is local position 2.
Proof. The positive response gives strict ingress at the actual central arc. For 021020, both central contact jets and the full terminal 020 are retained, so Theorem 8.10 applies. For 02102101, a contrary central cost at most 13 / 20 13/20 13/20 gives the prefix exit inequality. Lemma B.110 propagates it along the two following complete arcs to a fresh strict ingress; the actual terminal 101 then contradicts the preceding blocked-continuation argument. Only the prefix estimate is iterated. Section C.8.1 gives the equations and raw-switching regularity at both joins. □
For the next two words the rigid sources, in native indexing, are [ 7 , 11 ] + 2 = 12021 [7,11]+2=12021 [ 7 , 11 ] + 2 = 12021 at B = 9 B=9 B = 9 and [ 7 , 12 ] + 2 = 102021 [7,12]+2=102021 [ 7 , 12 ] + 2 = 102021 at B = 10 B=10 B = 10 . Corollary B.115, using genuine cuts inside arcs 2 and 7 and their translates, supplies one normalized lift on the unfolding. Its restrictions retain every complete response source used below. The scalar weak-five payments may use separate genuine lifts; no covector identification is needed when their physical costs are added.
Proposition 23.4. The word − − − − − + + -----++ − − − − − + + , of length 9, is excluded.
Proof. The complete five-source [ 1 , 5 ] = 21021 [1,5]=21021 [ 1 , 5 ] = 21021 has connector 3. Its reverse source ( 3 , 2 , 1 , 0 , − 1 ) (3,2,1,0,-1) ( 3 , 2 , 1 , 0 , − 1 ) with whole map j ↦ − j j\mapsto-j j ↦ − j is 02101, so Lemma 23.2 blocks that alternative. The response in the forward source [ 3 , 8 ] = 021020 [3,8]=021020 [ 3 , 8 ] = 021020 is therefore positive, giving c 5 > 299 / 250 c_5>299/250 c 5 > 299/250 .
Resolve a second disjunction at connector 4, using [ 2 , 6 ] + 1 = 21021 [2,6]+1=21021 [ 2 , 6 ] + 1 = 21021 . Its forward [ 4 , 8 ] + 1 = 02101 [4,8]+1=02101 [ 4 , 8 ] + 1 = 02101 is blocked. The reverse source ( 4 , 3 , 2 , 1 , 0 , − 1 ) (4,3,2,1,0,-1) ( 4 , 3 , 2 , 1 , 0 , − 1 ) with whole map j ↦ 2 − j j\mapsto2-j j ↦ 2 − j is 021020, giving c 2 > 299 / 250 c_2>299/250 c 2 > 299/250 . All response supports have five or six distinct physical arcs and are retained complete in the indicated genuine periods. Each singleton bound follows from its own resolved disjunction. Since the two charges are distinct,
C > c 2 + c 5 > 299 125 > C ∗ . C > c_2 + c_5 > \frac{299}{125} > C_*. C > c 2 + c 5 > 125 299 > C ∗ .
□
Proposition 23.5. The word − − − − − − − − −+, of length 10, is excluded.
Proof. Again [ 1 , 5 ] = 21021 [1,5]=21021 [ 1 , 5 ] = 21021 has connector 3. The reverse source ( 3 , 2 , 1 , 0 , − 1 ) (3,2,1,0,-1) ( 3 , 2 , 1 , 0 , − 1 ) with whole map j ↦ − j j \mapsto-j j ↦ − j is 02101 and is blocked. The forward source [ 3 , 10 ] = 02102101 [3,10]=02102101 [ 3 , 10 ] = 02102101 has positive response, so c 5 > 13 / 20 c_5 > 13/20 c 5 > 13/20 . The two proper five-sources [ 0 , 4 ] + 1 = 10210 [0,4]+1=10210 [ 0 , 4 ] + 1 = 10210 and [ 5 , 9 ] = 10210 [5,9]=10210 [ 5 , 9 ] = 10210 charge { 1 , 2 , 3 } \{1,2,3\} { 1 , 2 , 3 } and { 6 , 7 , 8 } \{6,7,8\} { 6 , 7 , 8 } at least 3 / 5 3/5 3/5 each by Lemma B.39 . Cuts in physical arcs 5 and 0 respectively retain these sources. Their charges and { 5 } \{5\} { 5 } are disjoint, so
C > 3 5 + 13 20 + 3 5 = 37 20 > C ∗ . C > \frac{3}{5}+\frac{13}{20}+\frac{3}{5}=\frac{37}{20}>C_*. C > 5 3 + 20 13 + 5 3 = 20 37 > C ∗ . Every response continuation has at most eight arcs, fewer than 10; its caps and terminal arcs have not been deleted. □
The two length-eight negative runs
The remaining words with a negative run of length eight use the same paired-centre geometry. Here a single maximum bound suffices because one common scalar complement avoids both possible centres.
Proposition 23.6 (A paired centre and its complement). The following two words are excluded:
− − − − − − − − + + +, − − − − − − − − + − −+.
Their lengths are 11 and 12, respectively.
Proof. Use the following rigid sources and genuine cuts:
Table 1 (PDF p. 78).
Corollary B.115 gives a common normalized lift and its proved twist. The rigidity overlap needs only positive cap portions; each cost comparison below still retains its full proper source. At B = 11 B=11 B = 11 the resulting periods retain [ 3 , 12 ] [3,12] [ 3 , 12 ] , [ 10 , 19 ] [10,19] [ 10 , 19 ] , and, after translating the cut 9 by − 11 -11 − 11 , [ 2 , 6 ] [2,6] [ 2 , 6 ] . At B = 12 B=12 B = 12 they retain [ 0 , 10 ] [0,10] [ 0 , 10 ] , [ 7 , 17 ] [7,17] [ 7 , 17 ] and [ 2 , 6 ] [2,6] [ 2 , 6 ] . These are genuine minimizing periods with the prescribed endpoints and real winding.
In both words [ 2 , 6 ] + 1 = 21021 [2,6]+1=21021 [ 2 , 6 ] + 1 = 21021 , with connector 4. The ordered sources ( 4 , 5 , 6 , 7 , 8 , 9 ) (4,5,6,7,8,9) ( 4 , 5 , 6 , 7 , 8 , 9 ) , j ↦ j + 1 j \mapsto j+1 j ↦ j + 1 ; ( B + 4 , B + 3 , B + 2 , B + 1 , B , B − 1 ) (B+4,B+3,B+2,B+1,B,B-1) ( B + 4 , B + 3 , B + 2 , B + 1 , B , B − 1 ) , j ↦ − j j \mapsto-j j ↦ − j are 021020. Their ingress switches are 4 ∣ 5 4|5 4∣5 and ( B + 4 ) ∣ ( B + 3 ) (B+4)|(B+3) ( B + 4 ) ∣ ( B + 3 ) ; their centres are physical arcs 6 and 2. Their full terminal triples are ( 7 , 8 , 9 ) (7,8,9) ( 7 , 8 , 9 ) and ( B + 1 , B , B − 1 ) (B+1,B,B-1) ( B + 1 , B , B − 1 ) . The proved twist identifies connector B + 4 B+4 B + 4 with 4 on the common lift. Theorem 8.14 therefore gives
max ( c 2 , c 6 ) > 299 / 250. \max(c_2,c_6) > 299/250. max ( c 2 , c 6 ) > 299/250. It does not give two singleton inequalities.
The complementary complete scalar sources are
Table 2 (PDF p. 78).
Their costs are greater than 71 / 100 71/100 71/100 by Theorem B.97, and at least 16 / 25 16/25 16/25 by Theorem B.59, respectively. Both sources are proper and retain their two exterior caps. Each complement
avoids both possible centres 2, 6; its separately chosen scalar lift need not be identified with the paired lift. Paying only a successful centre gives
C > 299 250 + { 71 100 , B = 11 , 16 25 , B = 12 , = { 953 500 , B = 11 , 459 250 , B = 12. C > \frac{299}{250}+\begin{cases}\frac{71}{100}, & B=11,\\ \frac{16}{25}, & B=12,\end{cases}=\begin{cases}\frac{953}{500}, & B=11,\\ \frac{459}{250}, & B=12.\end{cases} C > 250 299 + { 100 71 , 25 16 , B = 11 , B = 12 , = { 500 953 , 250 459 , B = 11 , B = 12.
Both values exceed C ∗ C_* C ∗ . The other centre and physical arcs 3, 4, 5 remain unpaid. □
A length-seven run with one short separator
All three remaining thirteen-arc words have one negative run of length seven and one of length two. Their three positive-length pairs are ( 1 , 3 ) (1,3) ( 1 , 3 ) , ( 2 , 2 ) (2,2) ( 2 , 2 ) and ( 3 , 1 ) (3,1) ( 3 , 1 ) . They share the proper ten-source
[ 12 , 21 ] + 2 = 2021021020 , [12,21]+2=2021021020, [ 12 , 21 ] + 2 = 2021021020 ,
retained by a cut inside physical arc 11. Corollary B.111, including its regeneration arcs and terminal 020, yields the alternative
max ( c 2 , c 5 ) > 13 / 20. (83) \max(c_2,c_5)>13/20. \tag*{(83)} max ( c 2 , c 5 ) > 13/20. (83)
Every branch below uses just one successful centre.
Proposition 23.7. The following two words are excluded:
− − − − − − − + − − + + + -\ -\ -\ -\ -\ -\ -\ +\ -\ -\ +\ +\ + − − − − − − − + − − + + + , − − − − − − − + + + − − + -\ -\ -\ -\ -\ -\ -\ +\ +\ +\ -\ -\ + − − − − − − − + + + − − + .
Proof. For w − : = − − − − − − − + − − + + + w_- :=-\ -\ -\ -\ -\ -\ -\ +\ -\ -\ +\ +\ + w − := − − − − − − − + − − + + + and w + : = − − − − − − − + + + − − + w_+ :=-\ -\ -\ -\ -\ -\ -\ +\ +\ +\ -\ -\ + w + := − − − − − − − + + + − − + , the following two scalar rows are available independently of that alternative:
The two costs for either word exceed 103 / 250 103/250 103/250 and 473 / 1000 473/1000 473/1000 by Theorem B.122 and Corollary B.17. Their cuts are respectively 8, 4 for w − w_- w − and 5, 8 for w + w_+ w + . If c 2 > 13 / 20 c_2>13/20 c 2 > 13/20 , use [ 2 , 6 ] = 10210 [2,6]=10210 [ 2 , 6 ] = 10210 in w − w_- w − , charging { 3 , 4 , 5 } \{3,4,5\} { 3 , 4 , 5 } , or [ 3 , 7 ] + 1 = 10210 [3,7]+1=10210 [ 3 , 7 ] + 1 = 10210 in w + w_+ w + , charging { 4 , 5 , 6 } \{4,5,6\} { 4 , 5 , 6 } . Otherwise (23.1) forces c 5 > 13 / 20 c_5>13/20 c 5 > 13/20 ; use [ 1 , 5 ] + 2 = 10210 [1,5]+2=10210 [ 1 , 5 ] + 2 = 10210 in either word, charging { 2 , 3 , 4 } \{2,3,4\} { 2 , 3 , 4 } . The cuts for these five-sources are 1, 2 and 0, respectively. Each chosen triple costs at least 3 / 5 3/5 3/5 . The successful centre, selected triple and two fixed scalar charges are pairwise disjoint in each branch. Hence both words satisfy
C > 103 250 + 473 1000 + 3 5 + 13 20 = 427 200 > C ∗ . C>\frac{103}{250}+\frac{473}{1000}+\frac{3}{5}+\frac{13}{20}=\frac{427}{200}>C_*. C > 250 103 + 1000 473 + 5 3 + 20 13 = 200 427 > C ∗ . The branch with c 2 = 13 / 20 c_2=13/20 c 2 = 13/20 belongs to the second alternative, so the boundary is included.
Proposition 23.8. The word − − − − − − − + + − − + + -\ -\ -\ -\ -\ -\ -\ +\ +\ -\ -\ +\ + − − − − − − − + + − − + + is excluded.
Proof. Use the same ten-source and centre alternative (23.1). The complete ordered sources [ 5 , 14 ] [5,14] [ 5 , 14 ] and ( 15 , 14 , … , 6 ) (15,14,\ldots,6) ( 15 , 14 , … , 6 ) , with maps j j j and − j -j − j and cuts 4, 5, are 1020102010. The prefix 10201020 contains the seven-source at positions 1, …, 7 used by Lemma B.103. Its actual unsorted chart gives E a > 1 Ea>1 E a > 1 . Reading prefix positions (4, 3, 2, 1, 0) with whole map j ↦ 2 − j j\mapsto2-j j ↦ 2 − j gives the complete capped source 12021 with entrance 1 − E a < 0 1-Ea<0 1 − E a < 0 . Lemma B.75 therefore charges local positions 1, 2, 3 strictly above 61/200. The two physical charges are { 6 , 7 , 8 } \{6,7,8\} { 6 , 7 , 8 } and { 12 , 0 , 1 } \{12,0,1\} { 12 , 0 , 1 } . Their supporting caps and sign witnesses remain in the two proper complete ten-sources.
If c 2 > 13 / 20 c_2 > 13/20 c 2 > 13/20 , the proper five-source [ref-2 , ref-6 ] = 10210, cut 1, pays at least 3 / 5 3/5 3/5 on { 3 , 4 , 5 } \{3,4,5\} { 3 , 4 , 5 } . Otherwise c 5 > 13 / 20 c_5 > 13/20 c 5 > 13/20 , and [ref-1 , ref-5 ] + 2 = 10210, cut 0, pays at least 3 / 5 3/5 3/5 on { 2 , 3 , 4 } \{2,3,4\} { 2 , 3 , 4 } . The four selected physical charges are disjoint in either branch, so
C > 2 61 200 + 3 5 + 13 20 = 93 50 > C ∗ . C > 2\frac{61}{200}+\frac{3}{5}+\frac{13}{20}=\frac{93}{50}>C_*. C > 2 200 61 + 5 3 + 20 13 = 50 93 > C ∗ .
The weak five-source remains weak; strictness is supplied by the selected centre and both tails. Equality of a centre with 13 / 20 13/20 13/20 is not lost.
Proof of Theorem 23.1. The single-run cases of lengths 9 and 10 are excluded by Propositions 23.4 and 23.5. Proposition 23.6 excludes the two words of lengths 11 and 12. Proposition 23.7 excludes the two asymmetric thirteen-arc words, and Proposition 23.8 excludes the equal positive-length pair (2, 2). These are all seven words in this family, as classified in Theorem 20.3.
The uniform six-type conclusion
Theorem 24.1 (The mixed alternatives and the all-minus word). Every original global minimizer with B ≥ 8 B \ge8 B ≥ 8 has one of the essential increment types R01–R06. These have eight or nine positive maximal arcs.
Proof. Theorem 20.3 gives six one-positive-run, thirteen two-positive-run and one three-positive-run candidates, together with R01. Among the one-run candidates, Theorem 22.1 and the rooted comparisons of Theorem 23.1 leave only R02 and R03. Among the two-run candidates, Theorems 22.7, 22.11 and 23.1, Proposition 21.3 and the R17 conclusion in Theorem 21.1 leave only R04, R05 and R06. The sole three-run candidate is R07, excluded by the exhaustive four-range argument of Section 21.2. Thus only R01–R06 remain as necessary alternatives.
R01 has B = 8 B = 8 B = 8 and increment word --------. Part III excludes R01, R02 and R03 using full closure and their respective finite covers. Part IV excludes R04, R05 and R06. All six identifiers are retained here to state the intermediate necessary classification; their final exclusions are assembled in Section 39.
An intermediate lower bound for R06
The source hypotheses of Theorem 11.1 can hold on a nine-arc type even when its ordinary scalar menu contains no proper returned own pair. The following consequence uses the central fee itself rather than a new unconditional entry for the bare word 02010.
Corollary 25.1. If an original global minimizer has type R06, then
c 2 + c 3 + c 4 > 80483 125000 , C > 180483 125000 . c_2+c_3+c_4>\frac{80483}{125000},\qquad C>\frac{180483}{125000}. c 2 + c 3 + c 4 > 125000 80483 , C > 125000 180483 .
Proof. The native word is 021202120, with B = 9 B=9 B = 9 and genuine j i + 9 = j i + 1 j_{i+9}=j_i+1 j i + 9 = j i + 1 . The proper six-source [ 0 , 5 ] + 1 = 102010 [0,5]+1=102010 [ 0 , 5 ] + 1 = 102010 retains the central [ 1 , 5 ] + 1 = 02010 [1,5]+1=02010 [ 1 , 5 ] + 1 = 02010 . Its first five arcs, read in the order ( 4 , 3 , 2 , 1 , 0 ) (4,3,2,1,0) ( 4 , 3 , 2 , 1 , 0 ) with whole reflected map j ↦ 1 − j j\mapsto1-j j ↦ 1 − j , are 12021 and prove normality before joint normalization. A cut inside omitted positive arc 8 retains this six-source. All central means, the full matrix and both state covectors belong to this one selected full lift.
The additional physical state facts are supplied by the following proper sources:
Table 11.
Their lengths are six or seven, strictly less than 9. Lemma B.53 applies to both complete crossing sources. They retain the original unsorted pair ( a , b , E ) (a,b,E) ( a , b , E ) ; whole reversal exchanges the two sides. The returned-tail sources supply the terminal circles for p = r 1 p=r_1 p = r 1 , q = r 5 q=r_5 q = r 5 and the orders p < r 2 p<r_2 p < r 2 , q < r 4 q<r_4 q < r 4 . The six-sources have shared physical owners 1 and 5 and give p 2 , q 2 < 1 / 6 p^2,q^2<1/6 p 2 , q 2 < 1/6 . These are exactly the state conditions (11.1). They do not require the different source lifts to agree. Theorem 11.1 yields the first assertion.
The left and right returned triples are { 8 , 0 , 1 } \{8,0,1\} { 8 , 0 , 1 } and { 5 , 6 , 7 } \{5,6,7\} { 5 , 6 , 7 } . Each costs more than 2 / 5 2/5 2/5 by Proposition 10.1 . They are disjoint from the central triple and from each other, so
C > 80483 125000 + 2 2 5 = 180483 125000 . C>\frac{80483}{125000}+\frac{2^2}{5}=\frac{180483}{125000}. C > 125000 80483 + 5 2 2 = 125000 180483 .
The bound remains below C ∗ C_* C ∗ by 180483/500000; it is a local fee. The complete R06 exclusion in Theorem 37.1 uses the additional same-point equations and full finite cover.
Part III. Exclusion of the three eight-arc types
The common geometric and covering principle
The run argument has left three eight-arc possibilities. In the native label convention they are
R01 02102102 --------, R02 02102120 -----+++, R03 02120120 --++++++.
In each row the labels continue by j i + 8 = j i + 1 j_{i+8}=j_i+1 j i + 8 = j i + 1 . Thus the final increment is read across the true twisted seam. The common feature is a full signed matrix closure, which reduces eight complete arc multipliers to five coordinates. The three resulting charts need not have the same domain or the same rejection predicates.
The physical subset of a closed parameter domain
Throughout this part a candidate is an original global minimizer in § 3.1 , with the indicated eight positive maximal arcs. Its state lies in K K K , its physical cost satisfies C ≤ C o c t < C ∗ C\le C_{\mathrm{oct}}<C_* C ≤ C oct < C ∗ , and its frame and state satisfy (4) with the prescribed real winding. The complete multiplier of each arc is q i > 1 q_i>1 q i > 1 ; its physical duration is a j log q i a_j\log q_i a j log q i . In particular q i q_i q i is neither a duration nor a square-root parameter for a half-return.
Each chart below is obtained from the full state and frame equations. A lawful change of cut or whole reflection transports the physical trajectory and its complete source qualifications. It can select a representative in a smaller closed root domain. The root may also contain points with a zero limiting duration, extra winding, or no physical reconstruction. Inclusion of such points enlarges the search domain and is harmless: the required implication is from every strict physical candidate to at least one root box. No converse is used.
Every local cost or state inequality is applied with its complete supporting source, on a genuine containing interval. Proposition 2.4 permits joint normalization of the selected full maximizing lift. A scalar physical conclusion can subsequently be combined with another such conclusion. This operation does not identify independently selected covectors on different windows.
Lemma 26.1 (Finite exclusion of the physical subset). Let S \mathcal{S} S be a set of qualified physical candidates, and let each candidate have a coordinate representative in a finite union of closed root boxes. Suppose a finite covering construction has the following properties. Every split covers its parent, including the common boundary; every contraction retains the representatives of all candidates in the box; and every terminal box has a verified strict contradiction to a necessary condition for its represented candidates. Then S \mathcal{S} S is empty.
Proof. If a candidate existed, choose one root box containing its representative. A contraction retains that point and a covering split places it in at least one child. Since the construction is finite and all branches terminate, the point reaches a terminal box. Its verified contradiction is incompatible with the candidate. Overlaps of closed boxes can repeat the point but cannot remove it. □
The geometric inclusion, the finite covering identities and the numerical signs are separate obligations in this lemma. File identities and counts bind the actual inputs to a checking record; they supply neither a geometric inclusion nor a sign proof. Unknown arithmetic outcomes and unfinished branches cannot serve as terminal contradictions.
For R01, some contradictions use a necessary local-minimum condition for the complete moving-state cost. For R02 and R03, the accepted terminal predicates use only necessary domains, physical costs, state inequalities and full closure values. Their Taylor or polynomial bounds may differentiate the explicit functions to enclose their values on a box. Such differentiation does not add a stationarity equation or a positive-semidefinite Hessian condition to those two candidate sets. We now give the three arguments with these distinctions retained.
The all-minus type R01 and its geometric class
We exclude the strict essential eight-arc control word 02102102 from global minimizers in the full Reinhardt admissible class. Full signed frame closure admits a smooth five-dimensional Gram chart. A four-duration inequality makes its state return elliptic and gives a stable expression for the complete moving-state cost. Physical changes of cut and reflection, together with the compactness bound for minimizers, reduce the problem to a finite rational cover. Rigorous interval bounds then contradict the comparison bound or a necessary local-minimum condition throughout this cover.
We use the original geometric minimizing class and normalization of § 3.1 , with its cited control correspondence and reconstruction inputs. The following exclusion concerns that class before imposing a switching word or compact-set restriction.
Theorem 27.1. An original globally minimizing Reinhardt control cannot have eight positive maximal arcs with successive labels
0 , 2 , 1 , 0 , 2 , 1 , 0 , 2 , 0,2,1,0,2,1,0,2, 0 , 2 , 1 , 0 , 2 , 1 , 0 , 2 , where the twisted continuation has label j i + 8 = j i + 1 ( m o d 3 ) j_{i+8}=j_i+1 \pmod{3} j i + 8 = j i + 1 ( mod 3 ) . Equivalently, its strict essential type cannot be B = 8 B=8 B = 8 with all eight cyclic increments negative. We call this type R01.
The assertion is about eight essential control arcs, not eight sides of a polygon. We prove it directly, without using a census of other types or an upper bound on their number of arcs. Its role in the present classification is to remove R01 from the six-type conclusion of Theorem 24.1 . None of the other whole-class exclusions is a premise of this argument.
Coordinates and physical time
Use (2.1)–(2.6), with the local notation
w = u + i v , A = A ∗ , R = R ∗ , cost = C . (84) w=u+iv,\qquad A=A_{*},\qquad R=R_{*},\qquad\operatorname{cost}=C. \tag*{(84)} w = u + i v , A = A ∗ , R = R ∗ , cost = C . (84)
Thus the full measurable control triangle, physical time (X 2 = − I X^{2}=-I X 2 = − I ), state and signed frame endpoints, and prescribed real winding are unchanged. The reconstructed area is 3 cost \sqrt{3}\operatorname{cost} 3 cost ; neither trace closure nor an angle only modulo 2 π 2\pi 2 π suffices. The inverse control map (5) retains the original barycentric class without a time change, with ( x , y ) = ( u , v ) / 3 (x,y)=(u,v)/\sqrt{3} ( x , y ) = ( u , v ) / 3 and μ = ( α 2 , α 0 , α 1 ) \mu=(\alpha_{2},\alpha_{0},\alpha_{1}) μ = ( α 2 , α 0 , α 1 ) in [3].
The finite-vertex-control theorem [3, Theorem 16.3.1] permits the usual essential arc count for original minimizers. The present theorem only needs its stated eight-arc hypothesis. Zero-duration pieces are deleted and adjacent equal controls are merged in the twisted extension, including its seam. The remaining arcs are maximal and have positive time. A cut at a switch gives exactly B B B arcs; a cut inside an arc may give B + 1 B+1 B + 1 pieces.
Two imported bounds and their consequences
The imported compactness theorem and smoothed-octagon comparison are Theorem 2.1 and (18) . In this notation
C ∗ = 180483 100000 C_{*}=\frac{180483}{100000} C ∗ = 100000 180483
and cost ≤ C o c t < C ∗ \operatorname{cost}\leq C_{\mathrm{oct}}<C_{*} cost ≤ C oct < C ∗ . The cuts (6) constrain original minimizers, not local competitors. The rational check of the strict budget is given with (A.4).
Lemma 27.2. On an original minimizing trajectory, 99 / 200 ≤ v ≤ 39 / 5 99/200\leq v\leq39/5 99/200 ≤ v ≤ 39/5 . Every complete vertex arc has multiplier 1 < q ≤ Q ≔ 520 / 33 1<q\leq Q\coloneqq520/33 1 < q ≤ Q : = 520/33 , where its physical duration is d j log q d_{j}\log q d j log q .
Proof. The height bounds are (7) , proved in Lemma A.1. Rotate a complete arc cyclically into mode 0. Invariance of K K K makes the same height bounds valid at both endpoints. Its exact flow is w ( p ) = p ( w ( 1 ) + 1 ) − 1 w(p)=p(w(1)+1)-1 w ( p ) = p ( w ( 1 ) + 1 ) − 1 , 1 ≤ p ≤ q 1\leq p\leq q 1 ≤ p ≤ q , so v o u t = q v i n v_{\mathrm{out}}=qv_{\mathrm{in}} v out = q v in and q ≤ ( 39 / 5 ) / ( 99 / 200 ) = 520 / 33 q\leq(39/5)/(99/200)=520/33 q ≤ ( 39/5 ) / ( 99/200 ) = 520/33 . Since d 0 = ( 1 + u ) / v d_{0}=(1+u)/v d 0 = ( 1 + u ) / v is constant, d log p / d t = 1 / d 0 d\log p/dt=1/d_{0} d log p / d t = 1/ d 0 ; its elapsed time is d 0 log q d_{0}\log q d 0 log q , not log q \log q log q .
Full closure in duration variables
Write B j = B j B_{j}=B_{j} B j = B j for the three control matrices in Section A.1, so d j = tr ( B j X ) d_{j}=\operatorname{tr}(B_{j}X) d j = tr ( B j X ) as in (A.1). On a vertex-j j j arc, d = tr ( B j X ) d=\operatorname{tr}(B_{j}X) d = tr ( B j X ) is constant and positive. For h = log q h=\log q h = log q , let E = exp ( h B j ) E=\exp(hB_{j}) E = exp ( h B j ) and Q j ( q ) = E − 2 Q_{j}(q)=E^{-2} Q j ( q ) = E − 2 . Direct integration gives
X o u t = E X i n E − 1 , M = I + d sinh h X i n − 2 d ( cosh h − 1 ) X i n B j = X i n Q j ( q ) X o u t − 1 . (85) X_{\mathrm{out}}=EX_{\mathrm{in}}E^{-1},\qquad M=I+d\sinh h\,X_{\mathrm{in}}-2d(\cosh h-1)X_{\mathrm{in}}B_{j}=X_{\mathrm{in}}Q_{j}(q)X_{\mathrm{out}}^{-1}. \tag*{(85)} X out = E X in E − 1 , M = I + d sinh h X in − 2 d ( cosh h − 1 ) X in B j = X in Q j ( q ) X out − 1 . (85)
The formulas can also be checked by differentiation: p d M / d p = d M X ( w ( p ) ) p\,dM/dp=dMX(w(p)) p d M / d p = d M X ( w ( p )) and M ( 1 ) = I M(1)=I M ( 1 ) = I . Thus the physical-time and frame scales agree.
For the R01 labels j i = − i ( m o d 3 ) j_{i}=-i\pmod{3} j i = − i ( mod 3 ) , put
N ( q ) = ( q − q q q − 1 − q ) , P = ( 1 − 2 1 0 ) . (86) N(q)=\begin{pmatrix}q&-q\\q&q^{-1}-q\end{pmatrix},\qquad P=\begin{pmatrix}1&-2\\1&0\end{pmatrix}. \tag*{(86)} N ( q ) = ( q q − q q − 1 − q ) , P = ( 1 1 − 2 0 ) . (86)
Since B − i = R i B 0 R − i B_{-i}=R^{i}B_{0}R^{-i} B − i = R i B 0 R − i and Q 0 ( q ) R = P N ( q ) P − 1 Q_{0}(q)R=PN(q)P^{-1} Q 0 ( q ) R = P N ( q ) P − 1 , multiplication gives
∏ i = 0 7 Q j i ( q i ) = P ( ∏ i = 0 7 N ( q i ) ) P − 1 R − 8 . \prod_{i=0}^{7}Q_{j_{i}}(q_{i})=P\left(\prod_{i=0}^{7}N(q_{i})\right)P^{-1}R^{-8}. i = 0 ∏ 7 Q j i ( q i ) = P ( i = 0 ∏ 7 N ( q i ) ) P − 1 R − 8 .
The frame telescope in (85) , together with the twisted state endpoint, makes G ( T ) = R G(T)=R G ( T ) = R equivalent to ∏ Q j i = R \prod Q_{j_{i}}=R ∏ Q j i = R . As R 3 = − I R^{3}=-I R 3 = − I , this is exactly
N ( q 0 ) ⋯ N ( q 7 ) = − I . (87) N(q_{0})\cdots N(q_{7})=-I. \tag*{(87)} N ( q 0 ) ⋯ N ( q 7 ) = − I . (87)
In particular the sign in this equation is not dispensable.
A complete Gram chart
All duration indices in this section are modulo eight and all pair indices modulo four. Full closure has a stronger consequence than a single scalar equation.
Lemma 28.1. Suppose q j ≥ 1 q_j \ge1 q j ≥ 1 and (87) holds. Set α i = q 2 i 2 + q 2 i + 1 2 − 1 \alpha_i=q_{2i}^{2}+q_{2i+1}^{2}-1 α i = q 2 i 2 + q 2 i + 1 2 − 1 and
C = ( q 0 0 0 − q 7 q 1 q 2 0 0 0 q 3 q 4 0 0 0 q 5 q 6 ) . C=\begin{pmatrix}
q_0 & 0 & 0 & -q_7\\
q_1 & q_2 & 0 & 0\\
0 & q_3 & q_4 & 0\\
0 & 0 & q_5 & q_6
\end{pmatrix}. C = q 0 q 1 0 0 0 q 2 q 3 0 0 0 q 4 q 5 − q 7 0 0 q 6 . Then C t C − I C^{\mathsf{t}}C-I C t C − I is positive semidefinite of rank two. After congruence by diag ( α i − 1 / 2 ) \operatorname{diag}(\alpha_i^{-1/2}) diag ( α i − 1/2 ) , it has diagonal entries one and consecutive off-diagonal entries c c c , s s s , c c c , − s -s − s , where
c = q 1 q 2 α 0 α 1 = q 5 q 6 α 2 α 3 > 0 , s = q 3 q 4 α 1 α 2 = q 7 q 0 α 3 α 0 > 0 , c 2 + s 2 = 1. (88) c=\frac{q_1q_2}{\sqrt{\alpha_0\alpha_1}}=\frac{q_5q_6}{\sqrt{\alpha_2\alpha_3}}>0,\qquad
s=\frac{q_3q_4}{\sqrt{\alpha_1\alpha_2}}=\frac{q_7q_0}{\sqrt{\alpha_3\alpha_0}}>0,\qquad
c^2+s^2=1. \tag*{(88)} c = α 0 α 1 q 1 q 2 = α 2 α 3 q 5 q 6 > 0 , s = α 1 α 2 q 3 q 4 = α 3 α 0 q 7 q 0 > 0 , c 2 + s 2 = 1. (88) Proof. Consider the antiperiodic recurrence q j − 1 z j − 1 + q j z j + 1 = z j q_{j-1}z_{j-1}+q_jz_{j+1}=z_j q j − 1 z j − 1 + q j z j + 1 = z j , z j + 8 = − z j z_{j+8}=-z_j z j + 8 = − z j . Its transfer on ( z j , q j − 1 z j − 1 ) t (z_j,q_{j-1}z_{j-1})^{\mathsf{t}} ( z j , q j − 1 z j − 1 ) t is
T ( q ) = ( 1 / q − 1 / q q 0 ) = ( S 0 − 1 N ( q ) S 0 ) t , S 0 = ( 1 1 1 0 ) . T(q)=\begin{pmatrix}1/q&-1/q\\q&0\end{pmatrix}=(S_0^{-1}N(q)S_0)^{\mathsf{t}},\qquad
S_0=\begin{pmatrix}1&1\\1&0\end{pmatrix}. T ( q ) = ( 1/ q q − 1/ q 0 ) = ( S 0 − 1 N ( q ) S 0 ) t , S 0 = ( 1 1 1 0 ) . By full signed closure its antiperiodic solution space has dimension two. Splitting odd and even coordinates gives the equations C v = u Cv=u C v = u , C t u = v C^{\mathsf{t}}u=v C t u = v . Consequently ker ( C t C − I ) \ker(C^{\mathsf{t}}C-I) ker ( C t C − I ) has dimension two. Its opposite-coordinate principal block diag ( α 0 , α 2 ) \operatorname{diag}(\alpha_0,\alpha_2) diag ( α 0 , α 2 ) is positive definite. The Schur complement must vanish by rank two, proving positive semidefiniteness. In the order ( 0 , 2 , 1 , 3 ) (0,2,1,3) ( 0 , 2 , 1 , 3 ) the normalized matrix is
( I U U t I ) ; \begin{pmatrix}I&U\\U^{\mathsf{t}}&I\end{pmatrix}; ( I U t U I ) ; its vanishing Schur complement says U t U = I U^{\mathsf{t}}U=I U t U = I . The signs of its four nonzero entries make U U U exactly
( c − s s c ) . \begin{pmatrix}c&-s\\s&c\end{pmatrix}. ( c s − s c ) . This proves every assertion. □
Theorem 28.2. The full positive locus q j > 1 q_j>1 q j > 1 satisfying (87) has a smooth bijective chart of differential rank five,
( t , x 0 , x 1 , x 2 , x 3 ) ∈ ( 0 , 1 ) 5 . (t,x_0,x_1,x_2,x_3)\in(0,1)^5. ( t , x 0 , x 1 , x 2 , x 3 ) ∈ ( 0 , 1 ) 5 . Its forward formulas are
a i = { t i even , 1 − t i odd , u i = 1 − a i x i , v i = a i u i + 1 , a_i=
\begin{cases}
t & i\ \text{even},\\
1-t & i\ \text{odd},
\end{cases}
\qquad
u_i=1-a_ix_i,\qquad
v_i=\frac{a_i}{u_{i+1}}, a i = { t 1 − t i even , i odd , u i = 1 − a i x i , v i = u i + 1 a i , D i = 1 − x i + ( 1 − a i ) x i x i + 1 , z i = a i D i u i + 1 . (89) D_i=1-x_i+(1-a_i)x_ix_{i+1},\qquad
z_i=\frac{a_iD_i}{u_{i+1}}. \tag*{(89)} D i = 1 − x i + ( 1 − a i ) x i x i + 1 , z i = u i + 1 a i D i . (89) q 2 i 2 = u i z i = 1 + ( 1 − a i ) ( 1 − x i + 1 ) a i D i , q 2 i + 1 2 = v i z i = 1 D i . (90) q_{2i}^{2}=\frac{u_i}{z_i}=1+\frac{(1-a_i)(1-x_{i+1})}{a_iD_i},
\qquad
q_{2i+1}^{2}=\frac{v_i}{z_i}=\frac{1}{D_i}. \tag*{(90)} q 2 i 2 = z i u i = 1 + a i D i ( 1 − a i ) ( 1 − x i + 1 ) , q 2 i + 1 2 = z i v i = D i 1 . (90) Every square root is positive. The inverse is
t = q 1 2 q 2 2 α 0 α 1 , x i = 1 − q 2 i 2 / α i a i . (91) t=\frac{q_1^2q_2^2}{\alpha_0\alpha_1},
\qquad
x_i=\frac{1-q_{2i}^2/\alpha_i}{a_i}. \tag*{(91)} t = α 0 α 1 q 1 2 q 2 2 , x i = a i 1 − q 2 i 2 / α i . (91) Proof. On the cube, u i > 1 − a i > 0 u_i>1-a_i>0 u i > 1 − a i > 0 and D i = 1 − x i u i + 1 ∈ ( 0 , 1 ) D_i=1-x_iu_{i+1}\in(0,1) D i = 1 − x i u i + 1 ∈ ( 0 , 1 ) . Also z i = u i + v i − 1 z_i=u_i+v_i-1 z i = u i + v i − 1 and
α i = z i − 1 , q 2 i 2 α i = u i , q 2 i + 1 2 α i = v i , v i u i + 1 = a i . \alpha_i=z_i^{-1},\qquad
\frac{q_{2i}^2}{\alpha_i}=u_i,\qquad
\frac{q_{2i+1}^2}{\alpha_i}=v_i,\qquad
v_iu_{i+1}=a_i. α i = z i − 1 , α i q 2 i 2 = u i , α i q 2 i + 1 2 = v i , v i u i + 1 = a i . These identities make the normalized Gram matrix the rank-two matrix of Lemma 28.1, with c = t c=\sqrt{t} c = t and s = 1 − t s=\sqrt{1-t} s = 1 − t . The same parity splitting gives a two-dimensional space of antiperiodic solutions of the recurrence. Since its unrestricted solution space also has dimension two, its monodromy is exactly − I -I − I . The transpose conjugacy above now proves (87) . This argument proves full closure, rather than merely its trace.
Conversely, Lemma 28.1 gives 0 < t = c 2 < 1 0<t=c^2<1 0 < t = c 2 < 1 . Set u i = q 2 i 2 / α i u_i=q_{2i}^2/\alpha_i u i = q 2 i 2 / α i , v i = q 2 i + 1 2 / α i v_i=q_{2i+1}^2/\alpha_i v i = q 2 i + 1 2 / α i . Strict durations give u i , v i < 1 u_i,v_i<1 u i , v i < 1 , and v i − 1 u i = 1 − a i v_{i-1}u_i=1-a_i v i − 1 u i = 1 − a i gives u i > 1 − a i u_i>1-a_i u i > 1 − a i . Thus the inverse in (28.4) lies
in the open cube. Substitution recovers (28.2)–(28.3) uniquely. Both the squared forward map and this inverse are rational and smooth near the image, with nonzero denominators. Differentiating their composition gives a five-dimensional left inverse of the forward Jacobian. Positive square roots preserve its rank. □
The same argument covers finite q j ≥ 1 q_j \ge1 q j ≥ 1 by allowing 0 ≤ x i ≤ 1 0 \le x_i \le1 0 ≤ x i ≤ 1 while retaining 0 < t < 1 0 < t < 1 0 < t < 1 and D i > 0 D_i > 0 D i > 0 . On this domain,
q 2 i + 1 = 1 ⟺ x i = 0 , q 2 i = 1 ⟺ x i + 1 = 1. (92) q_{2i+1}=1 \Longleftrightarrow x_i=0,\qquad q_{2i}=1 \Longleftrightarrow x_{i+1}=1. \tag*{(92)} q 2 i + 1 = 1 ⟺ x i = 0 , q 2 i = 1 ⟺ x i + 1 = 1. (92)
A face D i = 0 D_i=0 D i = 0 , namely x i = 1 x_i=1 x i = 1 , x i + 1 = 0 x_{i+1}=0 x i + 1 = 0 , is a duration pole, not a finite zero-time arc. Our use of closed boxes will not impose interior stationarity on the finite zero-time faces.
The gap inequality and ellipticity
The state return involves half durations. Its discriminant cannot be inferred from the rank of the full Gram matrix alone.
Lemma 29.1 (Four-duration gap). For a , b , c , d ≥ 1 a,b,c,d \ge1 a , b , c , d ≥ 1 , suppose
R 0 = b 2 − ( a 2 − 1 ) ( c 2 − 1 ) ≥ 0 , U 0 = c 2 − ( d 2 − 1 ) ( b 2 − 1 ) ≥ 0. R_0=b^2-(a^2-1)(c^2-1)\ge0,\qquad U_0=c^2-(d^2-1)(b^2-1)\ge0. R 0 = b 2 − ( a 2 − 1 ) ( c 2 − 1 ) ≥ 0 , U 0 = c 2 − ( d 2 − 1 ) ( b 2 − 1 ) ≥ 0. Put
H = ( a 2 + b 2 − 1 ) ( c 2 + d 2 − 1 ) − b 2 c 2 , E = ( a + b − 1 ) ( c + d − 1 ) − b c . H=(a^2+b^2-1)(c^2+d^2-1)-b^2c^2,\qquad E=(a+b-1)(c+d-1)-bc. H = ( a 2 + b 2 − 1 ) ( c 2 + d 2 − 1 ) − b 2 c 2 , E = ( a + b − 1 ) ( c + d − 1 ) − b c . Then E ≥ 0 E\ge0 E ≥ 0 and H − E 2 ≥ 0 H-E^2\ge0 H − E 2 ≥ 0 , with equality in the latter precisely when a = d = 1 a=d=1 a = d = 1 .
Proof. For increments A = a − 1 A=a-1 A = a − 1 , B = b − 1 B=b-1 B = b − 1 , C = c − 1 C=c-1 C = c − 1 , D = d − 1 D=d-1 D = d − 1 , one has E = A C + A D + A + B D + D ≥ 0 E=AC+AD+A+BD+D\ge0 E = A C + A D + A + B D + D ≥ 0 . First let b , c > 1 b,c>1 b , c > 1 . The guards put ( a , d ) (a,d) ( a , d ) in the rectangle
1 ≤ a ≤ a ∗ : = 1 + b 2 c 2 − 1 , 1 ≤ d ≤ d ∗ : = 1 + c 2 b 2 − 1 . 1\le a\le a_*:=\sqrt{1+\frac{b^2}{c^2-1}},\qquad1\le d\le d_*:=\sqrt{1+\frac{c^2}{b^2-1}}. 1 ≤ a ≤ a ∗ := 1 + c 2 − 1 b 2 , 1 ≤ d ≤ d ∗ := 1 + b 2 − 1 c 2 . Writing P 0 = H − E 2 P_0=H-E^2 P 0 = H − E 2 , direct differentiation gives ∂ a 2 P 0 = − 4 ( c − 1 ) ( d − 1 ) ≤ 0 \partial_a^2P_0=-4(c-1)(d-1)\le0 ∂ a 2 P 0 = − 4 ( c − 1 ) ( d − 1 ) ≤ 0 and ∂ d 2 P 0 = − 4 ( a − 1 ) ( b − 1 ) ≤ 0 \partial_d^2P_0=-4(a-1)(b-1)\le0 ∂ d 2 P 0 = − 4 ( a − 1 ) ( b − 1 ) ≤ 0 . Its lower bound is therefore the bilinear interpolation of its four corner values. Three values are 0 0 0 , 2 c 2 ( a ∗ − 1 ) 2c^2(a_*-1) 2 c 2 ( a ∗ − 1 ) and 2 b 2 ( d ∗ − 1 ) 2b^2(d_*-1) 2 b 2 ( d ∗ − 1 ) . To treat the fourth, set
x = b 2 − 1 , y = c 2 − 1 , h = 1 + x 2 + y 2 . x=\sqrt{b^2-1},\qquad y=\sqrt{c^2-1},\qquad h=\sqrt{1+x^2+y^2}. x = b 2 − 1 , y = c 2 − 1 , h = 1 + x 2 + y 2 . At that corner a ∗ = h / y a_*=h/y a ∗ = h / y , d ∗ = h / x d_*=h/x d ∗ = h / x and H = b c h / ( x y ) \sqrt{H}=bch/(xy) H = b c h / ( x y ) . Expansion gives
x y ( H − E ) = b c ( h + x y ) − [ h + ( b − 1 ) y ] [ h + ( c − 1 ) x ] xy(\sqrt{H}-E)=bc(h+xy)-[h+(b-1)y][h+(c-1)x] x y ( H − E ) = b c ( h + x y ) − [ h + ( b − 1 ) y ] [ h + ( c − 1 ) x ] = h [ x + y + ( b − x ) ( c − y ) − h ] + x y [ b + c − 1 − h ] . =h[x+y+(b-x)(c-y)-h]+xy[b+c-1-h]. = h [ x + y + ( b − x ) ( c − y ) − h ] + x y [ b + c − 1 − h ] . Both brackets are positive. For the second, square and use ( b + c − 1 ) 2 − h 2 = 2 ( b − 1 ) ( c − 1 ) > 0 (b+c-1)^2-h^2=2(b-1)(c-1)>0 ( b + c − 1 ) 2 − h 2 = 2 ( b − 1 ) ( c − 1 ) > 0 . For the first put r = b − x r=b-x r = b − x , s = c − y s=c-y s = c − y , so 0 < r , s < 1 0<r,s<1 0 < r , s < 1 and x = ( 1 − r 2 ) / ( 2 r ) x=(1-r^2)/(2r) x = ( 1 − r 2 ) / ( 2 r ) , y = ( 1 − s 2 ) / ( 2 s ) y=(1-s^2)/(2s) y = ( 1 − s 2 ) / ( 2 s ) . Then
( x + y + r s ) 2 − h 2 = ( 1 − r ) ( 1 − s ) ( 2 r 2 s 2 − r s + r + s + 1 ) 2 r s > 0. (x+y+rs)^2-h^2=\frac{(1-r)(1-s)(2r^2s^2-rs+r+s+1)}{2rs}>0. ( x + y + r s ) 2 − h 2 = 2 r s ( 1 − r ) ( 1 − s ) ( 2 r 2 s 2 − r s + r + s + 1 ) > 0. The fourth corner is strictly positive. Interpolation is strictly positive unless a = d = 1 a=d=1 a = d = 1 .
If b = 1 b=1 b = 1 , direct expansion instead gives P 0 / 2 = A c 2 + D a ( 1 − A C ) P_0/2=Ac^2+Da(1-AC) P 0 /2 = A c 2 + D a ( 1 − A C ) . The guard implies A ( A + 2 ) C ( C + 2 ) ≤ 1 A(A+2)C(C+2)\le1 A ( A + 2 ) C ( C + 2 ) ≤ 1 , hence A C ≤ 1 / 4 AC\le1/4 A C ≤ 1/4 . This expression vanishes precisely when A = D = 0 A=D=0 A = D = 0 . If c = 1 c=1 c = 1 , use P 0 / 2 = D b 2 + A d ( 1 − B D ) P_0/2=Db^2+Ad(1-BD) P 0 /2 = D b 2 + A d ( 1 − B D ) and B D ≤ 1 / 4 BD\le1/4 B D ≤ 1/4 . These formulas include b = c = 1 b=c=1 b = c = 1 and finish the proof. □
Lemma 29.2. Every cyclic prefix ( a , b , c , d ) (a,b,c,d) ( a , b , c , d ) of a signed-closed tuple q j ≥ 1 q_j\ge1 q j ≥ 1 satisfies the guards of Lemma 29.1 and H > 0 H>0 H > 0 . Hence δ : = H − E > 0 \delta:=\sqrt{H}-E>0 δ := H − E > 0 at all eight cuts.
Proof. A four-factor product has the exact form
N ( a ) N ( b ) N ( c ) N ( d ) = ( − p p + s − p − r p + r + s − τ ) , (93) \mathrm{N}(a)\mathrm{N}(b)\mathrm{N}(c)\mathrm{N}(d)=
\begin{pmatrix}
-p & p+s\\
-p-r & p+r+s-\tau
\end{pmatrix}, \tag*{(93)} N ( a ) N ( b ) N ( c ) N ( d ) = ( − p − p − r p + s p + r + s − τ ) , (93)
where p = a d / ( b c ) > 0 p=ad/(bc)>0 p = a d / ( b c ) > 0 , r = d ( b 2 − 1 ) / ( a b c ) ≥ 0 r=d(b^2-1)/(abc)\geq0 r = d ( b 2 − 1 ) / ( ab c ) ≥ 0 , s = a ( c 2 − 1 ) / ( b c d ) ≥ 0 s=a(c^2-1)/(bcd)\geq0 s = a ( c 2 − 1 ) / ( b c d ) ≥ 0 and τ = ( b 2 + c 2 − 1 ) / ( a b c d ) \tau=(b^2+c^2-1)/(abcd) τ = ( b 2 + c 2 − 1 ) / ( ab c d ) . The complementary four-factor product equals minus its inverse and has the same form, with p ′ > 0 p'>0 p ′ > 0 , r ′ , s ′ ≥ 0 r',s'\geq0 r ′ , s ′ ≥ 0 . Entry comparison gives p ′ = H / ( a b c d ) p'=H/(abcd) p ′ = H / ( ab c d ) , r ′ = R 0 / ( a b c d ) r'=R_0/(abcd) r ′ = R 0 / ( ab c d ) , s ′ = U 0 / ( a b c d ) s'=U_0/(abcd) s ′ = U 0 / ( ab c d ) . Finally a = d = 1 a=d=1 a = d = 1 would give H = 0 H=0 H = 0 , so the equality case of Lemma 29.1 is impossible. □
Theorem 29.3. For every finite q j ≥ 1 q_j\geq1 q j ≥ 1 satisfying full signed closure, the normalized half-return B = N ( q 0 ) ⋯ N ( q 7 ) \mathrm{B}=\mathrm{N}(\sqrt{q_0})\cdots\mathrm{N}(\sqrt{q_7}) B = N ( q 0 ) ⋯ N ( q 7 ) satisfies
− 2 < tr B < 0. (94) -2<\operatorname{tr}\mathrm{B}<0. \tag*{(94)} − 2 < tr B < 0. (94)
Proof. Congruence of the half-duration Gram matrix by diag ( α i − 1 / 4 ) \operatorname{diag}(\alpha_i^{-1/4}) diag ( α i − 1/4 ) gives
M = ( b 0 c 0 − s c b 1 s 0 0 s b 2 c − s 0 c b 3 ) , b i = q 2 i + q 2 i + 1 − 1 α i ∈ [ 1 , 2 ] . M=
\begin{pmatrix}
b_0 & \sqrt{c} & 0 & -\sqrt{s}\\
\sqrt{c} & b_1 & \sqrt{s} & 0\\
0 & \sqrt{s} & b_2 & \sqrt{c}\\
-\sqrt{s} & 0 & \sqrt{c} & b_3
\end{pmatrix},
\qquad
b_i=\frac{q_{2i}+q_{2i+1}-1}{\sqrt{\alpha_i}}\in[1,\sqrt{2}]. M = b 0 c 0 − s c b 1 s 0 0 s b 2 c − s 0 c b 3 , b i = α i q 2 i + q 2 i + 1 − 1 ∈ [ 1 , 2 ] .
Put k = c + s k=c+s k = c + s and g i j = k − b i b j g_{ij}=k-b_i b_j g ij = k − b i b j for adjacent pairs 01, 12, 23, 03. Lemma 29.2 gives
g 01 = δ 0 α 0 α 1 , g 12 = δ 2 α 1 α 2 , g 23 = δ 4 α 2 α 3 , g 03 = δ 6 α 3 α 0 > 0. g_{01}=\frac{\delta_0}{\sqrt{\alpha_0\alpha_1}},\qquad
g_{12}=\frac{\delta_2}{\sqrt{\alpha_1\alpha_2}},\qquad
g_{23}=\frac{\delta_4}{\sqrt{\alpha_2\alpha_3}},\qquad
g_{03}=\frac{\delta_6}{\sqrt{\alpha_3\alpha_0}}>0. g 01 = α 0 α 1 δ 0 , g 12 = α 1 α 2 δ 2 , g 23 = α 2 α 3 δ 4 , g 03 = α 3 α 0 δ 6 > 0.
Direct determinant expansion yields
det M = c g 01 g 23 + s g 03 g 12 c + s > 0. (95) \det M=\frac{c g_{01}g_{23}+s g_{03}g_{12}}{c+s}>0. \tag*{(95)} det M = c + s c g 01 g 23 + s g 03 g 12 > 0. (95)
For arbitrary positive durations r j r_j r j the transfer determinant identity is
det ( C ( r ) t C ( r ) − I ) = ( ∏ j r j ) ( tr ( N ( r 0 ) ⋯ N ( r 7 ) ) + 2 ) . \det\left(\mathrm{C}(r)^{\mathsf{t}}\mathrm{C}(r)-I\right)
=\left(\prod_j r_j\right)\left(\operatorname{tr}\left(\mathrm{N}(r_0)\cdots\mathrm{N}(r_7)\right)+2\right). det ( C ( r ) t C ( r ) − I ) = ( j ∏ r j ) ( tr ( N ( r 0 ) ⋯ N ( r 7 ) ) + 2 ) .
It follows either by expansion of the cyclic tridiagonal determinant, or by the two-by-two recurrence used in Lemma 28.1. Apply it to r j = q j r_j=\sqrt{q_j} r j = q j and use the congruence determinant. Writing Π = ∏ q j \Pi=\prod q_j Π = ∏ q j , we obtain
tr B + 2 = ∏ i α i Π det M > 0. (96) \operatorname{tr}\mathrm{B}+2=\sqrt{\frac{\prod_i\alpha_i}{\Pi}}\det M>0. \tag*{(96)} tr B + 2 = Π ∏ i α i det M > 0. (96)
For the opposite inequality, split B \mathrm{B} B into two four-factor half-products of the form (29.1), writing them as
( − p p + s P − p − r P D ) and ( − p ′ p ′ + s Q − p ′ − r Q D ′ ) . \begin{pmatrix}
-p & p+s_P\\
-p-r_P & D
\end{pmatrix}
\qquad\text{and}\qquad
\begin{pmatrix}
-p' & p'+s_Q\\
-p'-r_Q & D'
\end{pmatrix}. ( − p − p − r P p + s P D ) and ( − p ′ − p ′ − r Q p ′ + s Q D ′ ) .
All r , s r,s r , s are nonnegative. At the first full-duration cut, D = E 0 / q 0 q 1 q 2 q 3 ≥ 0 D=E_0/\sqrt{q_0q_1q_2q_3}\geq0 D = E 0 / q 0 q 1 q 2 q 3 ≥ 0 , whereas the complementary full product gives D ′ = H 0 / q 0 q 1 q 2 q 3 > D D'=\sqrt{H_0}/\sqrt{q_0q_1q_2q_3}>D D ′ = H 0 / q 0 q 1 q 2 q 3 > D . The opposite cut gives p > D ′ ≥ 0 p>D'\geq0 p > D ′ ≥ 0 . Thus
tr B = D D ′ − p p ′ − p ( r Q + s Q ) − p ′ ( r P + s P ) − s P r Q − r P s Q < 0. \operatorname{tr}\mathrm{B}
=DD'-pp'-p(r_Q+s_Q)-p'(r_P+s_P)-s_P r_Q-r_P s_Q<0. tr B = D D ′ − p p ′ − p ( r Q + s Q ) − p ′ ( r P + s P ) − s P r Q − r P s Q < 0.
In cube coordinates ∏ i u i v i = t 2 ( 1 − t ) 2 \prod_i u_i v_i=t^2(1-t)^2 ∏ i u i v i = t 2 ( 1 − t ) 2 , and
Π = ∏ i u i t ( 1 − t ) ∏ i D i , ∏ i α i Π = 1 t ( 1 − t ) . (97) \Pi=\frac{\prod_i u_i}{t(1-t)\prod_i D_i},
\qquad
\sqrt{\frac{\prod_i\alpha_i}{\Pi}}=\frac{1}{\sqrt{t(1-t)}}. \tag*{(97)} Π = t ( 1 − t ) ∏ i D i ∏ i u i , Π ∏ i α i = t ( 1 − t ) 1 . (97)
Consequently a positive discriminant circuit is
L : = tr B + 2 = c g 01 g 23 + s g 03 g 12 ( c + s ) c s , Δ : = 4 − ( tr B ) 2 = L ( 4 − L ) , 0 < L < 2. (98) L:=\operatorname{tr}\mathrm{B}+2
=\frac{c g_{01}g_{23}+s g_{03}g_{12}}{(c+s)cs},
\qquad
\Delta:=4-(\operatorname{tr}\mathrm{B})^2=L(4-L),
\qquad
0<L<2. \tag*{(98)} L := tr B + 2 = ( c + s ) cs c g 01 g 23 + s g 03 g 12 , Δ := 4 − ( tr B ) 2 = L ( 4 − L ) , 0 < L < 2. (98)
These are pointwise identities and positivity statements. An interval division by Δ \Delta Δ still requires a numerical lower endpoint above zero.
The moving state and the complete cost
At each cut rotate the current label to zero. The canonical state update is the fractional linear action of
F ( q ) = 1 2 ( q q + 2 − q 2 − q ) , H j = F ( q j + 7 ) ⋯ F ( q j ) . (99) F(q)=\frac{1}{2}\begin{pmatrix}q & q+2 \\ -q & 2-q\end{pmatrix}, \qquad H_j=F(q_{j+7})\cdots F(q_j). \tag*{(99)} F ( q ) = 2 1 ( q − q q + 2 2 − q ) , H j = F ( q j + 7 ) ⋯ F ( q j ) . (99)
Indeed F ( q ) F(q) F ( q ) represents R − 1 E 0 ( q ) R^{-1}E_0(q) R − 1 E 0 ( q ) , with its harmless positive scalar chosen so that det F ( q ) = q \det F(q)=q det F ( q ) = q . Let B j B_j B j be the cyclic half-product starting with q j q_j q j and put K 0 = ( 2 1 0 1 ) K_0=\begin{pmatrix}2 & 1 \\ 0 & 1\end{pmatrix} K 0 = ( 2 0 1 1 ) . Direct multiplication gives the exact bridge
H j Π = K 0 B j t K 0 − 1 , V : = 4 det H j − ( tr H j ) 2 = Π Δ > 0. (100) \frac{H_j}{\sqrt{\Pi}}=K_0B_j^tK_0^{-1}, \qquad V:=4\det H_j-(\operatorname{tr}H_j)^2=\Pi\Delta>0. \tag*{(100)} Π H j = K 0 B j t K 0 − 1 , V := 4 det H j − ( tr H j ) 2 = ΠΔ > 0. (100)
Thus the upper fixed state is unique and smooth. For a real return H = ( H r s ) H=(H_{rs}) H = ( H r s ) it is
u = H 00 − H 11 2 H 10 , v = 4 det H − ( tr H ) 2 2 ∣ H 10 ∣ . (101) u=\frac{H_{00}-H_{11}}{2H_{10}}, \qquad v=\frac{\sqrt{4\det H-(\operatorname{tr}H)^2}}{2\lvert H_{10}\rvert}. \tag*{(101)} u = 2 H 10 H 00 − H 11 , v = 2 ∣ H 10 ∣ 4 det H − ( tr H ) 2 . (101)
Ellipticity makes H 10 ≠ 0 H_{10}\ne0 H 10 = 0 . This formula recomputes the state at every point of the duration chart; no initial state is frozen when taking cost derivatives.
A normalized cost formula.
For B j = ( A j B j ′ C j ′ D j ′ ) B_j=\begin{pmatrix}A_j & B'_j \\ C'_j & D'_j\end{pmatrix} B j = ( A j C j ′ B j ′ D j ′ ) , set
σ j = A j + 2 B j ′ − D j ′ , G = 3 ∑ j = 0 7 σ j σ j + 1 + ∑ j = 0 7 σ j 2 log q j . (102) \sigma_j=A_j+2B'_j-D'_j,\qquad G=3\sum_{j=0}^{7}\sigma_j\sigma_{j+1}+\sum_{j=0}^{7}\sigma_j^2\log q_j. \tag*{(102)} σ j = A j + 2 B j ′ − D j ′ , G = 3 j = 0 ∑ 7 σ j σ j + 1 + j = 0 ∑ 7 σ j 2 log q j . (102)
(The matrices B j B_j B j of § 5.2 are distinct from the entry B j ′ B'_j B j ′ .)
Proposition 30.1. At every physical R01 chain, and by analytic continuation in its regular Gram chart, the complete moving-state cost is
cost = 8 − G Δ . (103) \operatorname{cost}=8-\frac{G}{\Delta}. \tag*{(103)} cost = 8 − Δ G . (103)
Proof. Let a = d j a=\mathfrak{d}_j a = d j , b = d j − 1 b=\mathfrak{d}_{j-1} b = d j − 1 and c = d j + 1 c=\mathfrak{d}_{j+1} c = d j + 1 be the positive active reciprocals at three successive canonical cuts of a physical chain. The canonical state transport is
d 0 , n e x t = d 2 − ( q − 1 ) d 0 q , d 1 , n e x t = d 0 , d 2 , n e x t = q d 1 + ( q − 1 ) d 0 . (104) d_{0,\mathrm{next}}=\frac{d_2-(q-1)d_0}{q},\qquad d_{1,\mathrm{next}}=d_0,\qquad d_{2,\mathrm{next}}=qd_1+(q-1)d_0. \tag*{(104)} d 0 , next = q d 2 − ( q − 1 ) d 0 , d 1 , next = d 0 , d 2 , next = q d 1 + ( q − 1 ) d 0 . (104)
The identity d 0 d 1 + d 0 d 2 + d 1 d 2 = 1 d_0d_1+d_0d_2+d_1d_2=1 d 0 d 1 + d 0 d 2 + d 1 d 2 = 1 gives
d 2 = 1 − a b a + b , q ( a + b ) ( a + c ) = 1 + a 2 , v = 2 ( a + b ) 1 + a 2 , u = a v − 1. d_2=\frac{1-ab}{a+b},\qquad q(a+b)(a+c)=1+a^2,\qquad v=\frac{2(a+b)}{1+a^2},\qquad u=av-1. d 2 = a + b 1 − ab , q ( a + b ) ( a + c ) = 1 + a 2 , v = 1 + a 2 2 ( a + b ) , u = a v − 1.
Along the mode-zero arc v ( p ) = p v v(p)=pv v ( p ) = p v , u ( p ) = a p v − 1 u(p)=apv-1 u ( p ) = a p v − 1 and d t = a d p / p \mathrm{d}t=a\,\mathrm{d}p/p d t = a d p / p . Integrating (3) therefore gives its exact cost
c j = a [ 2 v ( 1 − q − 1 ) + ( 1 + a 2 ) v 2 ( q − 1 ) ] − a 2 log q = a ( q − 1 ) ( 2 a + b + c ) − a 2 log q . \begin{aligned}
c_j=a\left[\frac{2}{v}(1-q^{-1})+\frac{(1+a^2)v}{2}(q-1)\right]-a^2\log q \\
&= a(q-1)(2a+b+c)-a^2\log q.
\end{aligned} c j = a [ v 2 ( 1 − q − 1 ) + 2 ( 1 + a 2 ) v ( q − 1 ) ] − a 2 log q = a ( q − 1 ) ( 2 a + b + c ) − a 2 log q .
The rational term can be written
a [ 1 − a b a + b + 1 − a c a + c − b − c ] . a\left[\frac{1-ab}{a+b}+\frac{1-ac}{a+c}-b-c\right]. a [ a + b 1 − ab + a + c 1 − a c − b − c ] .
In the sum over eight arcs, each neighboring pair ( a , c ) (a,c) ( a , c ) contributes 1 − a c 1-ac 1 − a c from the two fractions and − 2 a c -2ac − 2 a c from the remaining terms. Hence
cost = 8 − 3 ∑ j d j d j + 1 − ∑ j d j 2 log q j . (105) \operatorname{cost}=8-3\sum_j d_jd_{j+1}-\sum_j d_j^2\log q_j. \tag*{(105)} cost = 8 − 3 j ∑ d j d j + 1 − j ∑ d j 2 log q j . (105)
From (101) –(100) , the active reciprocal is the raw numerator H 00 − H 11 + 2 H 10 H_{00}-H_{11}+2H_{10} H 00 − H 11 + 2 H 10 divided by V \sqrt{V} V , with sign ε j = sgn ( H j , 10 ) \varepsilon_j=\operatorname{sgn}(H_{j,10}) ε j = sgn ( H j , 10 ) . These signs agree at all cuts: if H H H fixes w = u + i v w=u+iv w = u + i v , then H = ( tr H / 2 ) I + ( H 10 v ) X ( w ) H=(\operatorname{tr}H/2)I+(H_{10}v)X(w) H = ( tr H /2 ) I + ( H 10 v ) X ( w ) ; conjugation by the positive-determinant state map carries X ( w ) X(w) X ( w ) to X ( F ⋅ w ) X(F\mathbin{\cdot}w) X ( F ⋅ w ) and preserves this scalar coefficient. Since all upper heights are positive, its sign is unchanged. The numerator is Π σ j \sqrt{\Pi}\,\sigma_j Π σ j , so d j = ε σ j / Δ d_j=\varepsilon\sigma_j/\sqrt{\Delta} d j = ε σ j / Δ for one common ε ∈ { 1 , − 1 } \varepsilon\in\{1,-1\} ε ∈ { 1 , − 1 } . The sign cancels in both quadratic sums, proving (103) . Every step is an algebraic identity followed by a logarithm on positive durations, so the same formula defines the analytic continuation where the circuit is regular. □
For a small, reproducible arithmetic circuit define
b i = u i + a i x i u i + 1 1 + D i . (106) b_i=\sqrt{u_i}+\frac{\sqrt{a_i}\,x_i\sqrt{u_{i+1}}}{1+\sqrt{D_i}}. \tag*{(106)} b i = u i + 1 + D i a i x i u i + 1 . (106)
This equals the b i b_i b i of Theorem 29.3 . The pair matrix
C i = ( 0 − u i v i − b i ) C_i=\begin{pmatrix}0&-\sqrt{u_i}\\ \sqrt{v_i}&-b_i\end{pmatrix} C i = ( 0 v i − u i − b i )
satisfies N ( q 2 i ) N ( q 2 i + 1 ) = C i / ( u i v i ) 1 / 4 N(\sqrt{q_{2i}})N(\sqrt{q_{2i+1}})=C_i/(u_iv_i)^{1/4} N ( q 2 i ) N ( q 2 i + 1 ) = C i / ( u i v i ) 1/4 . Thus B 2 i = C i C i + 1 C i + 2 C i + 3 / ( c s ) B_{2i}=C_iC_{i+1}C_{i+2}C_{i+3}/(cs) B 2 i = C i C i + 1 C i + 2 C i + 3 / ( cs ) ; odd cuts follow by
B j + 1 = T 0 ( q j ) − 1 B j T 0 ( q j ) , T 0 ( q ) = ( q − q q 1 − q ) , det T 0 ( q ) = q . (107) B_{j+1}=T_0(q_j)^{-1}B_jT_0(q_j),\qquad T_0(q)=\begin{pmatrix}q&-q\\q&1-q\end{pmatrix},\qquad\det T_0(q)=q. \tag*{(107)} B j + 1 = T 0 ( q j ) − 1 B j T 0 ( q j ) , T 0 ( q ) = ( q q − q 1 − q ) , det T 0 ( q ) = q . (107)
Equations (106) , (98) and (102) therefore compute all quantities without the possibly large scale Π \Pi Π . A matrix-free equivalent and the function-value bounds used in the certificates are proved in Appendix § 5.16 .
Legitimate five-dimensional variations
Lemma 30.2. At any strict physical R01 chain which is an original local minimum, all five partial derivatives of (30.5) in the Gram coordinates vanish. In particular this holds at an original global minimizer, including one on an equality face of a compactness, cap, or canonical representative condition.
Proof. Theorem 28.2 gives a full open neighborhood of signed-closed positive durations of rank five. Theorem 29.3 and (101) give its smoothly moving upper initial state. Its canonical return closes exactly, so the original state has the prescribed twist. The frame telescope (85) and (87) then give G ( T ) = R G(T)=R G ( T ) = R exactly, not just a trace condition.
At the given physical chain the whole finite trajectory is a compact subset of the open set Ω \Omega Ω . Parameterize each arc on a fixed unit interval by p = 1 + s ( q − 1 ) p=1+s(q-1) p = 1 + s ( q − 1 ) . State and frame formulas depend smoothly on the five parameters, uniformly on these finitely many intervals. A sufficiently small neighborhood therefore retains Ω \Omega Ω , positive reciprocals, and positive physical durations. The endpoints of the frame paths stay I , R I,R I , R throughout the deformation. This is a fixed-point homotopy and preserves the required real winding. The geometric reconstruction consequently gives genuine competitors. Equivalently, its nonzero tangents have nonnegative curvature: with B = ∑ μ j B j B=\sum\mu_jB_j B = ∑ μ j B j , d = tr ( B X ) > 0 d=\operatorname{tr}(BX)>0 d = tr ( B X ) > 0 , the identity X B X = B + d X XBX=B+dX X B X = B + d X gives X + X − 1 X ˙ = − 2 B / d X+X^{-1}\dot X=-2B/d X + X − 1 X ˙ = − 2 B / d ; the three independent boundary curvature determinants are 3 ( μ 1 , μ 2 , μ 0 ) / d ≥ 0 \sqrt{3}(\mu_1,\mu_2,\mu_0)/d\geq0 3 ( μ 1 , μ 2 , μ 0 ) / d ≥ 0 . The fixed winding makes the resulting boundary convex and traversed once.
::: The total time is free in the original problem. Thus each coordinate can vary in both directions, with the initial state recomputed rather than held fixed. Fermat’s theorem proves all five equations. The compact set K K K , the cap and the choice of representative were never imposed on these competitors. □
No theorem asserting that every point of the entire open Gram cube is physical is needed for this local argument. It starts at the actual physical candidate and uses openness. In contrast, the Gram, gap and ellipticity statements above are whole-domain statements; they also justify the interval value bounds below.
For a chart coordinate y r y_r y r define the cleared quantities
P = ( 8 − C ∗ ) Δ − G = ( cost − C ∗ ) Δ , N r = G ∂ r Δ − Δ ∂ r G = Δ 2 ∂ r cost . (108) P=(8-C_*)\Delta-G=(\operatorname{cost}-C_*)\Delta,\qquad N_r=G\partial_r\Delta-\Delta\partial_rG=\Delta^2\partial_r\operatorname{cost}. \tag*{(108)} P = ( 8 − C ∗ ) Δ − G = ( cost − C ∗ ) Δ , N r = G ∂ r Δ − Δ ∂ r G = Δ 2 ∂ r cost . (108)
Since Δ > 0 \Delta>0 Δ > 0 at candidates, P > 0 P>0 P > 0 contradicts the budget and N r ≠ 0 N_r\ne0 N r = 0 contradicts stationarity. A nonzero linear combination of these derivatives works as well. These two kinds of exclusion have different meanings: a derivative sign is not a pointwise cost bound.
A finite cover of all possible candidates
Physical symmetries
The duration permutations used here are changes of a physical chain, not numerical sorting operations. Extend a trajectory by
G e ( s + T ) = G e ( s ) R , X e ( s + T ) = R − 1 X e ( s ) R , j i + 8 = j i + 1. G_e(s+T)=G_e(s)R,\qquad X_e(s+T)=R^{-1}X_e(s)R,\qquad j_{i+8}=j_i+1. G e ( s + T ) = G e ( s ) R , X e ( s + T ) = R − 1 X e ( s ) R , j i + 8 = j i + 1.
For a switch time s s s , the new frame G e ( s ) − 1 G e ( s + τ ) G_e(s)^{-1}G_e(s+\tau) G e ( s ) − 1 G e ( s + τ ) starts at I I I and ends at R R R . Its state is X e ( s + τ ) X_e(s+\tau) X e ( s + τ ) . Varying s s s continuously gives a fixed-endpoint homotopy. A cyclic phase X ↦ R − 1 X R X\mapsto R^{-1}XR X ↦ R − 1 X R , G ↦ R − 1 G R G\mapsto R^{-1}GR G ↦ R − 1 GR changes label j j j to j + 1 j+1 j + 1 and preserves tr ( A X ) \operatorname{tr}(AX) tr ( A X ) . A one-arc recut followed by this phase therefore implements
S ( q ) j = q j + 1 . (109) S(q)_j=q_{j+1}. \tag*{(109)} S ( q ) j = q j + 1 . (109)
For reflection put J 0 = diag ( 1 , − 1 ) J_0=\operatorname{diag}(1,-1) J 0 = diag ( 1 , − 1 ) and choose the old cut s s s at the beginning of arc 2. The formulas
G ~ ( τ ) = J 0 G e ( s ) − 1 G e ( s − τ ) J 0 , X ~ ( τ ) = − J 0 X e ( s − τ ) J 0 (110) \widetilde{G}(\tau)=J_0G_e(s)^{-1}G_e(s-\tau)J_0,\qquad\widetilde{X}(\tau)=-J_0X_e(s-\tau)J_0 \tag*{(110)} G ( τ ) = J 0 G e ( s ) − 1 G e ( s − τ ) J 0 , X ( τ ) = − J 0 X e ( s − τ ) J 0 (110)
implement V ( q ) j = q 1 − j V(q)_j=q_{1-j} V ( q ) j = q 1 − j . Indeed J 0 R − 1 J 0 = R J_0R^{-1}J_0=R J 0 R − 1 J 0 = R , J 0 A J 0 = − A J_0AJ_0=-A J 0 A J 0 = − A and − J 0 B j J 0 = B 2 − j -J_0B_jJ_0=B_{2-j} − J 0 B j J 0 = B 2 − j . They give the original equations, both endpoints, the same positive reciprocal on each reversed arc, and the same cost. The state mirror is ( u , v ) ↦ ( − u , v ) (u,v)\mapsto(-u,v) ( u , v ) ↦ ( − u , v ) and preserves Ω \Omega Ω . After genuine reanchoring, reversal has the form G ( τ ) ↦ J 0 R − 1 G ( T − τ ) J 0 G(\tau)\mapsto J_0R^{-1}G(T-\tau)J_0 G ( τ ) ↦ J 0 R − 1 G ( T − τ ) J 0 . Apply it to the fixed-endpoint homotopy to D ∗ exp ( π J τ / ( 3 T ) ) D ∗ − 1 D_*\exp(\pi J\tau/(3T))D_*^{-1} D ∗ exp ( π J τ / ( 3 T )) D ∗ − 1 . It fixes this reference path, because J 0 J_0 J 0 commutes with D ∗ D_* D ∗ and reverses J J J . Thus the required real winding is retained. An odd label permutation alone would not prove these statements.
These operations generate all rotations and reflections of the eight duration indices. Choose a least duration as q 0 q_0 q 0 and, if necessary, reflect about index zero. Every physical chain has a cost-equivalent representative with
q 0 ≤ q j ( 1 ≤ j ≤ 7 ) , q 1 ≤ q 7 . (111) q_0\le q_j\ (1\le j\le7),\qquad q_1\le q_7. \tag*{(111)} q 0 ≤ q j ( 1 ≤ j ≤ 7 ) , q 1 ≤ q 7 . (111)
An original global minimizer remains one after these operations, so Lemma 27.2 applies anew. Ties are retained. Selection of a representative is not differentiated and does not lower the local five-dimensional variation space.
For use in interval filters, (111) is exactly the following eight polynomial inequalities on the strict cube:
t D 0 − u 0 u 1 D i ≥ 0 ( 0 ≤ i ≤ 3 ) , t D 0 u i u i + 1 − a i D i u 0 u 1 ≥ 0 ( 1 ≤ i ≤ 3 ) , D 0 − D 3 ≥ 0. (112) \begin{aligned}
tD_0-u_0u_1D_i&\ge0 &&(0\le i\le3),\\
tD_0u_i u_{i+1}-a_iD_i u_0u_1&\ge0 &&(1\le i\le3),\\
D_0-D_3&\ge0. \tag*{(112)}
\end{aligned} t D 0 − u 0 u 1 D i t D 0 u i u i + 1 − a i D i u 0 u 1 D 0 − D 3 ≥ 0 ≥ 0 ≥ 0. ( 0 ≤ i ≤ 3 ) , ( 1 ≤ i ≤ 3 ) , (112)
Their denominators before clearing are positive. Similarly the cap is exactly
D i ≥ e ≔ Q − 2 = 1089 270400 , ( 1 − a i ) ( 1 − x i + 1 ) ≤ ( Q 2 − 1 ) a i D i . (113) D_i \ge e \coloneqq Q^{-2} = \frac{1089}{270400}, \qquad(1-a_i)(1-x_{i+1}) \le(Q^2-1)a_iD_i. \tag*{(113)} D i ≥ e : = Q − 2 = 270400 1089 , ( 1 − a i ) ( 1 − x i + 1 ) ≤ ( Q 2 − 1 ) a i D i . (113)
The angular bounds.
Lemma 31.1. A strict canonical capped candidate satisfies
T 0 < t < 1 − T 0 , T 0 = 1 1 + Q 2 = 1089 271489 . (114) T_0 < t < 1-T_0, \qquad T_0 = \frac{1}{1+Q^2} = \frac{1089}{271489}. \tag*{(114)} T 0 < t < 1 − T 0 , T 0 = 1 + Q 2 1 = 271489 1089 . (114) On a lower angular band [ t l , t h ] ⊂ [ T 0 , 1 / 5 ] [t_l,t_h] \subset[T_0,1/5] [ t l , t h ] ⊂ [ T 0 , 1/5 ] it also satisfies x 1 ≥ 1 − t h 2 / ( 1 − t h ) 2 x_1 \ge1-t_h^2/(1-t_h)^2 x 1 ≥ 1 − t h 2 / ( 1 − t h ) 2 ; on an upper band [ t l , t h ] ⊂ [ 4 / 5 , 1 − T 0 ] [t_l,t_h] \subset[4/5,1-T_0] [ t l , t h ] ⊂ [ 4/5 , 1 − T 0 ] it satisfies x 0 ≤ 1 / ( 2 t l ) x_0 \le1/(2t_l) x 0 ≤ 1/ ( 2 t l ) .
Proof. The comparison q 0 ≤ q 1 q_0 \le q_1 q 0 ≤ q 1 says u 0 u 1 ≤ t u_0u_1 \le t u 0 u 1 ≤ t . Strict shapes imply D 1 < u 1 ≤ t / u 0 < t / ( 1 − t ) D_1 < u_1 \le t/u_0 < t/(1-t) D 1 < u 1 ≤ t / u 0 < t / ( 1 − t ) , so q 3 2 > ( 1 − t ) / t q_3^2 > (1-t)/t q 3 2 > ( 1 − t ) / t . Its cap gives t > T 0 t > T_0 t > T 0 . Writing s = 1 − t s=1-t s = 1 − t , the exact identity
u 3 u 0 − t D 3 = t ( D 0 − D 3 ) + s ( 1 − x 3 ) u 0 + t s x 0 ( 1 − x 1 ) u_3u_0-tD_3=t(D_0-D_3)+s(1-x_3)u_0+tsx_0(1-x_1) u 3 u 0 − t D 3 = t ( D 0 − D 3 ) + s ( 1 − x 3 ) u 0 + t s x 0 ( 1 − x 1 )
has positive right side, since q 1 ≤ q 7 q_1 \le q_7 q 1 ≤ q 7 means D 0 ≥ D 3 D_0 \ge D_3 D 0 ≥ D 3 . Thus q 6 2 = u 3 u 0 / ( s D 3 ) > t / s q_6^2=u_3u_0/(sD_3)>t/s q 6 2 = u 3 u 0 / ( s D 3 ) > t / s , and t < 1 − T 0 t<1-T_0 t < 1 − T 0 .
Solving u 0 u 1 ≤ t u_0u_1 \le t u 0 u 1 ≤ t gives 1 − x 1 ≤ t 2 x 0 / [ ( 1 − t ) ( 1 − t x 0 ) ] < t 2 / ( 1 − t ) 2 1-x_1 \le t^2x_0/[(1-t)(1-tx_0)] < t^2/(1-t)^2 1 − x 1 ≤ t 2 x 0 / [( 1 − t ) ( 1 − t x 0 )] < t 2 / ( 1 − t ) 2 , which proves the lower restriction. For the upper restriction use D 0 ≥ D 3 D_0 \ge D_3 D 0 ≥ D 3 and x 3 ≤ 1 x_3 \le1 x 3 ≤ 1 : x 0 u 1 ≤ x 3 u 0 ≤ u 0 x_0u_1 \le x_3u_0 \le u_0 x 0 u 1 ≤ x 3 u 0 ≤ u 0 . As u 1 ≥ t u_1 \ge t u 1 ≥ t , this implies 2 t x 0 ≤ 1 2tx_0 \le1 2 t x 0 ≤ 1 . □
The three closed intervals
[ T 0 , 1 / 5 ] , [ 1 / 5 , 4 / 5 ] , [ 4 / 5 , 1 − T 0 ] (115) [T_0,1/5], \qquad[1/5,4/5], \qquad[4/5,1-T_0] \tag*{(115)} [ T 0 , 1/5 ] , [ 1/5 , 4/5 ] , [ 4/5 , 1 − T 0 ] (115)
cover the candidates, with shared endpoints. In particular T 0 < e T_0<e T 0 < e ; the short interval below e e e is not discarded.
The middle grid and the endpoint roots.
In the middle interval use the Cartesian grid with cuts
t : ( 1 / 5 , 1 / 2 , 4 / 5 ) , x i : ( 0 , 1 / 5 , 3 / 10 , 7 / 10 , 4 / 5 , 1 ) . (116) t:(1/5,1/2,4/5), \qquad x_i:(0,1/5,3/10,7/10,4/5,1). \tag*{(116)} t : ( 1/5 , 1/2 , 4/5 ) , x i : ( 0 , 1/5 , 3/10 , 7/10 , 4/5 , 1 ) . (116)
It consists of 2 ⋅ 5 4 = 1250 2\cdot5^4=1250 2 ⋅ 5 4 = 1250 closed rational boxes. Its serial numbers are the lexicographic ordering of { 0 , 1 } × { 0 , 1 , 2 , 3 , 4 } 4 \{0,1\}\times\{0,1,2,3,4\}^4 { 0 , 1 } × { 0 , 1 , 2 , 3 , 4 } 4 , starting at zero.
For the endpoints, partition the lower interval at
T 0 , 1 / 128 , 1 / 64 , 1 / 32 , 1 / 16 , 1 / 8 , 1 / 5 , (117) T_0,\ 1/128,\ 1/64,\ 1/32,\ 1/16,\ 1/8,\ 1/5, \tag*{(117)} T 0 , 1/128 , 1/64 , 1/32 , 1/16 , 1/8 , 1/5 , (117)
and reflect these six bands about 1 / 2 1/2 1/2 for the upper interval. Put H = 1 − e / 2 H=1-e/2 H = 1 − e /2 . On each band choose, independently for every x i x_i x i , either [ 0 , H ] [0,H] [ 0 , H ] or [ H , 1 ] [H,1] [ H , 1 ] . Whenever the latter is chosen, also require
x i + 1 ≥ e ( 2 − e ) b h , b = 1 − a i , (118) x_{i+1} \ge\frac{e}{(2-e)b_h}, \qquad b=1-a_i, \tag*{(118)} x i + 1 ≥ ( 2 − e ) b h e , b = 1 − a i , (118)
where b h b_h b h is its band maximum. Finally intersect with the shape restrictions of Lemma 31.1 and discard empty intersections.
Lemma 31.2. This construction retains every canonical capped endpoint candidate. Exactly 120 boxes remain, 72 lower and 48 upper. On each entire retained box, all u i > 0 u_i>0 u i > 0 and D i ≥ e / 4 > 0 D_i\ge e/4>0 D i ≥ e /4 > 0 .
Proof. For a high coordinate, the candidate inequality D i ≥ e D_i \ge e D i ≥ e implies
b x i + 1 ≥ e − 1 + x i x i ≥ e − 1 + H H = e 2 − e , bx_{i+1}\ge\frac{e-1+x_i}{x_i}\ge\frac{e-1+H}{H}=\frac{e}{2-e}, b x i + 1 ≥ x i e − 1 + x i ≥ H e − 1 + H = 2 − e e ,
@page 90/200 starts_mid=1 ends_mid=1 Their denominators before clearing are positive. Similarly the cap is exactly
D i ≥ e ≔ Q − 2 = 1089 270400 , ( 1 − a i ) ( 1 − x i + 1 ) ≤ ( Q 2 − 1 ) a i D i . (119) D_i \ge e \coloneqq Q^{-2} = \frac{1089}{270400}, \qquad(1-a_i)(1-x_{i+1}) \le(Q^2-1)a_iD_i. \tag*{(119)} D i ≥ e : = Q − 2 = 270400 1089 , ( 1 − a i ) ( 1 − x i + 1 ) ≤ ( Q 2 − 1 ) a i D i . (119)
The angular bounds.
Lemma 31.1. A strict canonical capped candidate satisfies
T 0 < t < 1 − T 0 , T 0 = 1 1 + Q 2 = 1089 271489 . (120) T_0 < t < 1-T_0, \qquad T_0 = \frac{1}{1+Q^2} = \frac{1089}{271489}. \tag*{(120)} T 0 < t < 1 − T 0 , T 0 = 1 + Q 2 1 = 271489 1089 . (120) On a lower angular band [ t l , t h ] ⊂ [ T 0 , 1 / 5 ] [t_l,t_h] \subset[T_0,1/5] [ t l , t h ] ⊂ [ T 0 , 1/5 ] it also satisfies x 1 ≥ 1 − t h 2 / ( 1 − t h ) 2 x_1 \ge1-t_h^2/(1-t_h)^2 x 1 ≥ 1 − t h 2 / ( 1 − t h ) 2 ; on an upper band [ t l , t h ] ⊂ [ 4 / 5 , 1 − T 0 ] [t_l,t_h] \subset[4/5,1-T_0] [ t l , t h ] ⊂ [ 4/5 , 1 − T 0 ] it satisfies x 0 ≤ 1 / ( 2 t l ) x_0 \le1/(2t_l) x 0 ≤ 1/ ( 2 t l ) .
Proof. The comparison q 0 ≤ q 1 q_0 \le q_1 q 0 ≤ q 1 says u 0 u 1 ≤ t u_0u_1 \le t u 0 u 1 ≤ t . Strict shapes imply D 1 < u 1 ≤ t / u 0 < t / ( 1 − t ) D_1 < u_1 \le t/u_0 < t/(1-t) D 1 < u 1 ≤ t / u 0 < t / ( 1 − t ) , so q 3 2 > ( 1 − t ) / t q_3^2 > (1-t)/t q 3 2 > ( 1 − t ) / t . Its cap gives t > T 0 t > T_0 t > T 0 . Writing s = 1 − t s=1-t s = 1 − t , the exact identity
u 3 u 0 − t D 3 = t ( D 0 − D 3 ) + s ( 1 − x 3 ) u 0 + t s x 0 ( 1 − x 1 ) u_3u_0-tD_3=t(D_0-D_3)+s(1-x_3)u_0+tsx_0(1-x_1) u 3 u 0 − t D 3 = t ( D 0 − D 3 ) + s ( 1 − x 3 ) u 0 + t s x 0 ( 1 − x 1 )
has positive right side, since q 1 ≤ q 7 q_1 \le q_7 q 1 ≤ q 7 means D 0 ≥ D 3 D_0 \ge D_3 D 0 ≥ D 3 . Thus q 6 2 = u 3 u 0 / ( s D 3 ) > t / s q_6^2=u_3u_0/(sD_3)>t/s q 6 2 = u 3 u 0 / ( s D 3 ) > t / s , and t < 1 − T 0 t<1-T_0 t < 1 − T 0 .
Solving u 0 u 1 ≤ t u_0u_1 \le t u 0 u 1 ≤ t gives 1 − x 1 ≤ t 2 x 0 / [ ( 1 − t ) ( 1 − t x 0 ) ] < t 2 / ( 1 − t ) 2 1-x_1 \le t^2x_0/[(1-t)(1-tx_0)] < t^2/(1-t)^2 1 − x 1 ≤ t 2 x 0 / [( 1 − t ) ( 1 − t x 0 )] < t 2 / ( 1 − t ) 2 , which proves the lower restriction. For the upper restriction use D 0 ≥ D 3 D_0 \ge D_3 D 0 ≥ D 3 and x 3 ≤ 1 x_3 \le1 x 3 ≤ 1 : x 0 u 1 ≤ x 3 u 0 ≤ u 0 x_0u_1 \le x_3u_0 \le u_0 x 0 u 1 ≤ x 3 u 0 ≤ u 0 . As u 1 ≥ t u_1 \ge t u 1 ≥ t , this implies 2 t x 0 ≤ 1 2tx_0 \le1 2 t x 0 ≤ 1 . □
The three closed intervals
[ T 0 , 1 / 5 ] , [ 1 / 5 , 4 / 5 ] , [ 4 / 5 , 1 − T 0 ] (121) [T_0,1/5], \qquad[1/5,4/5], \qquad[4/5,1-T_0] \tag*{(121)} [ T 0 , 1/5 ] , [ 1/5 , 4/5 ] , [ 4/5 , 1 − T 0 ] (121)
cover the candidates, with shared endpoints. In particular T 0 < e T_0<e T 0 < e ; the short interval below e e e is not discarded.
The middle grid and the endpoint roots.
In the middle interval use the Cartesian grid with cuts
t : ( 1 / 5 , 1 / 2 , 4 / 5 ) , x i : ( 0 , 1 / 5 , 3 / 10 , 7 / 10 , 4 / 5 , 1 ) . (122) t:(1/5,1/2,4/5), \qquad x_i:(0,1/5,3/10,7/10,4/5,1). \tag*{(122)} t : ( 1/5 , 1/2 , 4/5 ) , x i : ( 0 , 1/5 , 3/10 , 7/10 , 4/5 , 1 ) . (122)
It consists of 2 ⋅ 5 4 = 1250 2\cdot5^4=1250 2 ⋅ 5 4 = 1250 closed rational boxes. Its serial numbers are the lexicographic ordering of { 0 , 1 } × { 0 , 1 , 2 , 3 , 4 } 4 \{0,1\}\times\{0,1,2,3,4\}^4 { 0 , 1 } × { 0 , 1 , 2 , 3 , 4 } 4 , starting at zero.
For the endpoints, partition the lower interval at
T 0 , 1 / 128 , 1 / 64 , 1 / 32 , 1 / 16 , 1 / 8 , 1 / 5 , (123) T_0,\ 1/128,\ 1/64,\ 1/32,\ 1/16,\ 1/8,\ 1/5, \tag*{(123)} T 0 , 1/128 , 1/64 , 1/32 , 1/16 , 1/8 , 1/5 , (123)
and reflect these six bands about 1 / 2 1/2 1/2 for the upper interval. Put H = 1 − e / 2 H=1-e/2 H = 1 − e /2 . On each band choose, independently for every x i x_i x i , either [ 0 , H ] [0,H] [ 0 , H ] or [ H , 1 ] [H,1] [ H , 1 ] . Whenever the latter is chosen, also require
x i + 1 ≥ e ( 2 − e ) b h , b = 1 − a i , (124) x_{i+1} \ge\frac{e}{(2-e)b_h}, \qquad b=1-a_i, \tag*{(124)} x i + 1 ≥ ( 2 − e ) b h e , b = 1 − a i , (124)
where b h b_h b h is its band maximum. Finally intersect with the shape restrictions of Lemma 31.1 and discard empty intersections.
Lemma 31.2. This construction retains every canonical capped endpoint candidate. Exactly 120 boxes remain, 72 lower and 48 upper. On each entire retained box, all u i > 0 u_i>0 u i > 0 and D i ≥ e / 4 > 0 D_i\ge e/4>0 D i ≥ e /4 > 0 .
Proof. For a high coordinate, the candidate inequality D i ≥ e D_i \ge e D i ≥ e implies
b x i + 1 ≥ e − 1 + x i x i ≥ e − 1 + H H = e 2 − e , bx_{i+1}\ge\frac{e-1+x_i}{x_i}\ge\frac{e-1+H}{H}=\frac{e}{2-e}, b x i + 1 ≥ x i e − 1 + x i ≥ H e − 1 + H = 2 − e e ,
so (31.10) is necessary. This proves coverage. For a low coordinate D i ≥ 1 − x i ≥ e / 2 D_i \ge1-x_i \ge e/2 D i ≥ 1 − x i ≥ e /2 throughout its box. For a high coordinate every point of the retained rectangle satisfies
D i ≥ b l H e ( 2 − e ) b h = e 2 b l b h ≥ e 4 . D_i \ge b_l H\frac{e}{(2-e)b_h}=\frac{e}{2}\frac{b_l}{b_h}\ge\frac{e}{4}. D i ≥ b l H ( 2 − e ) b h e = 2 e b h b l ≥ 4 e .
The last inequality follows directly from (31.9) for both b = t b=t b = t and b = 1 − t b=1-t b = 1 − t on each band. Also u i ≥ T 0 u_i \ge T_0 u i ≥ T 0 . The number of boxes is a finite rational intersection calculation: the 12 ⋅ 16 = 192 12\cdot16=192 12 ⋅ 16 = 192 band-mask choices leave 72 lower, 48 upper and 72 empty intersections. These counts are obtained by the displayed bounds, without interval cost evaluation. □ \square □
The endpoint root indices 0 , … , 119 0,\ldots,119 0 , … , 119 are a separate namespace from the middle serials. The lower bound on D i D_i D i is a whole-box primitive bound. It does not authorize an invented uniform numerical lower bound for Δ \Delta Δ on those closed boxes.
The finite interval verification
The remaining step is finite. Here a certificate means a collection of rational boxes, exact covering records, and rigorously outward bounds for specified analytic functions. It does not mean a collection of floating-point samples or a successful search log. The arithmetic and the precise interpretation of these records are given in Appendices 33–34.
Elementary exclusion rules
A box is removed only by one of the following rules.
A necessary canonical or cap inequality is strictly impossible on the whole box. For polynomial inequalities an exact rational Bernstein bound may be used.
A rigorous range enclosure of P P P in (30.10), or an explicitly stated positive multiple of it, has lower endpoint greater than zero.
The whole box maps into a previously verified region of the cost or descent lemma below.
An exact candidate-preserving contraction has empty intersection.
For (iii), containment concerns the entire image enclosure and the precise hypotheses of the earlier result. For (iv), the contractor is proved from necessary candidate inequalities; it is not used to restrict the domain of a derivative calculation. Every other box is subdivided or retained as unresolved. An inconclusive interval is never an exclusion.
The Bernstein rule in (i) has an elementary justification. Affinely rescale a rational box to [ 0 , 1 ] 5 [0,1]^5 [ 0 , 1 ] 5 and expand a necessary polynomial in the tensor-product Bernstein basis. The basis polynomials are nonnegative and sum to one. Thus its value lies between the smallest and largest coefficients. If the largest coefficient is strictly negative, a necessary nonnegative inequality is impossible. Exact rational coefficients, not rounded signs near zero, decide this test. Equality faces are retained.
Lemma 32.1 (Previously bounded regions). The cost satisfies cost > C ∗ \operatorname{cost}>C_* cost > C ∗ in each of the following regions, where it denotes the regular analytic cost of (30.5):
0 < t < 1 , 3 10 ≤ x i ≤ 7 10 ( 0 ≤ i ≤ 3 ) ; 0<t<1,\qquad\frac{3}{10}\le x_i\le\frac{7}{10}\quad(0\le i\le3); 0 < t < 1 , 10 3 ≤ x i ≤ 10 7 ( 0 ≤ i ≤ 3 ) ; 0 < t ≤ 3 5 , 1 5 ≤ x 0 ≤ 3 10 , 3 10 ≤ x i ≤ 7 10 ( i > 0 ) ; 0<t\le\frac{3}{5},\qquad\frac{1}{5}\le x_0\le\frac{3}{10},\qquad\frac{3}{10}\le x_i\le\frac{7}{10}\quad(i>0); 0 < t ≤ 5 3 , 5 1 ≤ x 0 ≤ 10 3 , 10 3 ≤ x i ≤ 10 7 ( i > 0 ) ; 9 20 ≤ t ≤ 11 20 , 7 10 ≤ x i ≤ 3 4 ( 0 ≤ i ≤ 3 ) . \frac{9}{20}\le t\le\frac{11}{20},\qquad\frac{7}{10}\le x_i\le\frac{3}{4}\quad(0\le i\le3). 20 9 ≤ t ≤ 20 11 , 10 7 ≤ x i ≤ 4 3 ( 0 ≤ i ≤ 3 ) . On the cube 0 ≤ x i ≤ 1 / 5 0\le x_i\le1/5 0 ≤ x i ≤ 1/5 , it satisfies
{ cost > C ∗ , 0 < t ≤ 3 / 8 or 5 / 8 ≤ t < 1 , ∑ i = 0 3 ∂ x i cost > 1 / 250 , 3 / 8 ≤ t ≤ 5 / 8. (125) \begin{cases}
\operatorname{cost}>C_*, & 0<t\le3/8\ \text{or}\ 5/8\le t<1,\\
\sum_{i=0}^{3}\partial_{x_i}\operatorname{cost}>1/250, & 3/8\le t\le5/8. \tag*{(125)}
\end{cases} { cost > C ∗ , ∑ i = 0 3 ∂ x i cost > 1/250 , 0 < t ≤ 3/8 or 5/8 ≤ t < 1 , 3/8 ≤ t ≤ 5/8. (125)
Consequently no strict original local minimum of cost at most C ∗ C_* C ∗ lies in any of these regions or in a lawful symmetry image of them.
Interval proof. The cost assertions are verified by outward enclosures of the cleared cost, using the normalized circuit or the factored endpoint circuit of Appendix § 5.18 . Each record checks its entire primitive domain before evaluation. The interior central calculation, for example, consists of 184 positive-cost terminal boxes; it uses no stationary-root or compactness test. The edge cost region is another finite cost-only covering with exact rational joins. The upper-shape region is instead transported into the central one by a single recut. To see this without an additional cost calculation, set
F i ∗ = 1 − a i + a i ( 1 − x i ) ( 1 − x i + 2 ) , t ′ = t D 0 D 2 F 0 ∗ F 1 ∗ , x i ′ = ( 1 − x i + 2 ) F i + 1 ∗ u i + 1 D i + 2 . F_i^* = 1-a_i+a_i(1-x_i)(1-x_{i+2}), \qquad t'=\frac{tD_0D_2}{F_0^*F_1^*}, \qquad x_i'=\frac{(1-x_{i+2})F_{i+1}^*}{u_{i+1}D_{i+2}}. F i ∗ = 1 − a i + a i ( 1 − x i ) ( 1 − x i + 2 ) , t ′ = F 0 ∗ F 1 ∗ t D 0 D 2 , x i ′ = u i + 1 D i + 2 ( 1 − x i + 2 ) F i + 1 ∗ .
Substitution in the inverse chart proves q j ′ = q j + 1 q_j'=q_{j+1} q j ′ = q j + 1 . Positivity and F 0 ∗ F 1 ∗ = t D 0 D 2 + ( 1 − t ) D 1 D 3 F_0^*F_1^*=tD_0D_2+(1-t)D_1D_3 F 0 ∗ F 1 ∗ = t D 0 D 2 + ( 1 − t ) D 1 D 3 give 0 < t ′ < 1 0<t'<1 0 < t ′ < 1 . On the stated upper box,
47 80 ≤ u i ≤ 137 200 , 389 800 ≤ D i ≤ 471 800 , 31 64 ≤ F i ∗ ≤ 1181 2000 . \frac{47}{80}\leq u_i\leq\frac{137}{200},\qquad\frac{389}{800}\leq D_i\leq\frac{471}{800},\qquad\frac{31}{64}\leq F_i^*\leq\frac{1181}{2000}. 80 47 ≤ u i ≤ 200 137 , 800 389 ≤ D i ≤ 800 471 , 64 31 ≤ F i ∗ ≤ 2000 1181 .
The positive quotient therefore gives 19375 / 64527 ≤ x i ′ ≤ 56688 / 91415 19375/64527\leq x_i'\leq56688/91415 19375/64527 ≤ x i ′ ≤ 56688/91415 , strictly inside [ 3 / 10 , 7 / 10 ] [3/10,7/10] [ 3/10 , 7/10 ] . This proves the claimed transport and cost bound.
For specificity, the larger small-shape region in (125) uses the following complete 21-strip half-angle cover, without shape subdivision. Here c = t c=\sqrt{t} c = t in the first row.
Table 12 (PDF p. 92).
The first eleven records have a strictly positive cost target. For each of the last ten records the exact represented interval endpoints prove
H : = G ∑ i Δ x i − Δ ∑ i G x i > 7 / 1000 , 0 < Δ < 13 / 10. H:=G\sum_i\Delta_{x_i}-\Delta\sum_iG_{x_i}>7/1000,\qquad0<\Delta<13/10. H := G i ∑ Δ x i − Δ i ∑ G x i > 7/1000 , 0 < Δ < 13/10.
Since H = Δ 2 ∑ i cost x i H=\Delta^2\sum_i\operatorname{cost}_{x_i} H = Δ 2 ∑ i cost x i , their ratio is greater than 7 / 1690 > 1 / 250 7/1690>1/250 7/1690 > 1/250 . Pair recutting,
( t , x 0 , x 1 , x 2 , x 3 ) ↦ ( 1 − t , x 1 , x 2 , x 3 , x 0 ) , (t,x_0,x_1,x_2,x_3)\mapsto(1-t,x_1,x_2,x_3,x_0), ( t , x 0 , x 1 , x 2 , x 3 ) ↦ ( 1 − t , x 1 , x 2 , x 3 , x 0 ) ,
preserves the cost and summed shape derivative and supplies the other half of the angle interval. The endpoint record touching c = 0 c=0 c = 0 uses a continuous value enclosure for c log c c\log c c log c , never its derivative at zero. The whole-box Taylor rule is proved below.
At a strict candidate the simultaneous negative shape direction is a legitimate variation by Lemma 30.2 . This proves the last conclusion, but does not exclude a zero-duration point on the central faces of the collar. □
The exact rational coordinate formula for the reversal used to transport these regions is
t ′ = 1 − t , x i ′ = 1 − x 1 − i 1 − a 1 − i x 1 − i . (126) t'=1-t,\qquad x_i'=\frac{1-x_{1-i}}{1-a_{1-i}x_{1-i}}. \tag*{(126)} t ′ = 1 − t , x i ′ = 1 − a 1 − i x 1 − i 1 − x 1 − i . (126)
It follows by applying (91) to q j ′ = q 1 − j q_j'=q_{1-j} q j ′ = q 1 − j . All denominators are positive, and 1 − ( 1 − x ) / ( 1 − a x ) = ( 1 − a ) x / ( 1 − a x ) > 0 1-(1-x)/(1-ax)=(1-a)x/(1-ax)>0 1 − ( 1 − x ) / ( 1 − a x ) = ( 1 − a ) x / ( 1 − a x ) > 0 for strict shapes. The map is an involution. Together with the four pair recuts it gives the eight whole-box transports used in the inherited-region tests.
Proposition 32.2 (Finite covering result). No point in the middle grid (31.8) or the 120 endpoint roots of Lemma 31.2 can simultaneously represent a strict original global minimizer, obey (31.3)–(31.5), and have cost at most C ∗ C_* C ∗ .
Interval proof. The computation applies only the preceding rules and the exact subdivision and coordinate-change rules of the appendices. Its complete original-root accounting is as follows:
Table 13.
Table 14.
The middle groups are disjoint as sets of original serials. All 1250 serials and all 120 endpoint indices occur; the two namespaces are not identified. Shared geometric box faces cause no gap.
The two largest endpoint completions retain their original source rectangles and the exact candidate-preserving chart maps. Their saved-frontier decompositions are
Table 15.
Root 44 uses a parameter frontier; root 60 retains its original broad-coordinate frontier. These are counts of covered frontier parents, not counts of arbitrary physical boxes or independent whole roots. The adaptive children cover their parents by exact rational splits, and every resulting terminal cost bound is strictly positive with its required domain guards. Earlier cost-positive branches, canonical omissions and inherited regions remain part of the same original-root proof. The final uncovered sets are empty.
The checking is in two parts. Numerical checking recomputes the claimed bounds with outward arithmetic and verifies their strict signs and primitive domains. Exact checking reconstructs every root, contractor, change of chart and binary split, and verifies the union from the original roots to those terminal conclusions. Reused records are identified by their original source box, coordinate chart, target, arithmetic order and immutable source files; a matching local index alone is insufficient. The independent checks reproduce both parts, including the original-root joins. Appendix 34 specifies where these finite records and checking programs are located and separates a numerical replay from an integrity-only check.
Thus induction over the finite subdivision trees, with the four sound terminal rules, proves the proposition. Neither unresolved nodes nor exploratory numerical records are part of its accepted terminal cover. □
Proof of Theorem 27.1. Suppose an original minimizer has the strict R01 type. Cut at one of its eight switches and normalize its first label. Equations (27.2)–(27.4) give full signed closure, and Theorem 28.2 places it in the open Gram cube. Its physical reconstruction is the unique elliptic moving-state branch of § 5.8 . Lemma 30.2 supplies all five necessary derivative equations. Choose a lawful canonical representative. It is still an original minimizer, so it obeys the cap of Lemma 27.2 , the strict budget (2.19), and the angular bounds of Lemma 31.1. The closed middle and endpoint covers contain it. This contradicts Proposition 32.2. □ \square □
A zero multiplier increment q j − 1 = 0 q_{j-1}=0 q j − 1 = 0 has zero physical time: Q j ( 1 ) = I Q_j(1)=I Q j ( 1 ) = I and Q j ( a ) Q j ( b ) = Q j ( a b ) Q_j(a)Q_j(b)=Q_j(ab) Q j ( a ) Q j ( b ) = Q j ( ab ) . Deleting such an arc and merging equal neighbors preserves the original state, frame and winding, and leaves at most seven essential arcs. One must not instead delete N ( 1 ) N(1) N ( 1 ) , which is not the identity. These lower essential types are outside the theorem. In particular, the two alternating finite boundary points of the R01 chart can encode the four-arc smoothed-octagon competitor; strict interior descent does not contradict its role in (2.19).
Arithmetic identities and rigorous range bounds
This appendix states the analytic rules behind the finite computation. They concern values and derivatives on the actual rectangular domain of an evaluation, not just the subset containing possible minimizers.
Matrix free numerators and value bounds
At pair cut i i i , write the cyclic b b b ’s as b 0 , b 1 , b 2 , b 3 b_0,b_1,b_2,b_3 b 0 , b 1 , b 2 , b 3 and put k = a i k=\sqrt{a_i} k = a i , l = 1 − a i l=\sqrt{1-a_i} l = 1 − a i and
W = l b 3 g 12 + k b 1 g 23 > 0 , d = 2 g 12 / k − L , e ′ = − 2 u i W ( k + l ) k l . W=lb_3g_{12}+kb_1g_{23}>0,\qquad d=2g_{12}/k-L,\qquad e'=-\frac{2\sqrt{u_i}W}{(k+l)kl}. W = l b 3 g 12 + k b 1 g 23 > 0 , d = 2 g 12 / k − L , e ′ = − ( k + l ) k l 2 u i W .
The exact formulas used for the normalized numerators are
σ 2 i = d + e ′ , σ 2 i + 1 = − d + ( q 2 i − 1 − 1 ) e ′ = − σ 2 i + e ′ / q 2 i , e ′ q 2 i = − 2 D i W u i + 1 ( k + l ) l . (127) \sigma_{2i}=d+e',\qquad\sigma_{2i+1}=-d+(q_{2i}^{-1}-1)e'=-\sigma_{2i}+e'/q_{2i},\qquad\frac{e'}{q_{2i}}=-\frac{2\sqrt{D_i}W}{\sqrt{u_{i+1}}(k+l)l}. \tag*{(127)} σ 2 i = d + e ′ , σ 2 i + 1 = − d + ( q 2 i − 1 − 1 ) e ′ = − σ 2 i + e ′ / q 2 i , q 2 i e ′ = − u i + 1 ( k + l ) l 2 D i W . (127)
To verify them, conjugate C i C_i C i by S i = diag ( u i , 1 ) S_i=\operatorname{diag}(\sqrt{u_i},1) S i = diag ( u i , 1 ) :
S i − 1 C i S i + 1 = ( 0 − 1 a i − b i ) . S_i^{-1}C_iS_{i+1}=
\begin{pmatrix}
0 & -1\\
\sqrt{a_i} & -b_i
\end{pmatrix}. S i − 1 C i S i + 1 = ( 0 a i − 1 − b i ) .
The product of four such matrices has entries P 00 = l ( l − b 1 b 2 ) P_{00}=l(l-b_1b_2) P 00 = l ( l − b 1 b 2 ) and P 01 = b 1 b 2 b 3 − k b 1 − l b 3 = − W / ( k + l ) P_{01}=b_1b_2b_3-kb_1-lb_3=-W/(k+l) P 01 = b 1 b 2 b 3 − k b 1 − l b 3 = − W / ( k + l ) . Since B 2 i = S i P S i − 1 / ( k l ) B_{2i}=S_iPS_i^{-1}/(kl) B 2 i = S i P S i − 1 / ( k l ) , the first formula follows from its trace; (30.9) gives the second. These equalities also provide an independent matrix-product check of the scalar circuit.
For fixed a ∈ ( 0 , 1 ) a\in(0,1) a ∈ ( 0 , 1 ) write u = 1 − ( 1 − a ) y u=1-(1-a)y u = 1 − ( 1 − a ) y , D = 1 − u x D=1-ux D = 1 − ux and
b ( x , y ; a ) = 1 − a x + a / u ( 1 − D ) . b(x,y;a)=\sqrt{1-ax}+\sqrt{a/u}(1-\sqrt{D}). b ( x , y ; a ) = 1 − a x + a / u ( 1 − D ) .
Throughout the finite closed-shape domain,
∂ x b = 1 2 ( a u 1 − u x − a 1 − a x ) ≥ 0 , (128) \partial_x b=\frac{1}{2}\left(\frac{\sqrt{au}}{\sqrt{1-ux}}-\frac{a}{\sqrt{1-ax}}\right)\ge0, \tag*{(128)} ∂ x b = 2 1 ( 1 − ux a u − 1 − a x a ) ≥ 0 , (128)
∂ y b = − ( 1 − a ) a x 2 u D ( 1 + D ) ≤ 0. \partial_y b=-\frac{(1-a)\sqrt{ax}}{2\sqrt{u}\sqrt{D}(1+\sqrt{D})}\le0. ∂ y b = − 2 u D ( 1 + D ) ( 1 − a ) a x ≤ 0.
For the first sign, the difference of the squared nonnegative terms is a ( u − a ) / [ ( 1 − u x ) ( 1 − a x ) ] ≥ 0 a(u-a)/[(1-ux)(1-ax)]\ge0 a ( u − a ) / [( 1 − ux ) ( 1 − a x )] ≥ 0 . The second formula is direct differentiation. Consequently
1 ≤ b ≤ a + 1 − a , b ( x l , y h ; a ) ≤ b ( x , y ; a ) ≤ b ( x h , y l ; a ) . (129) 1\le b\le\sqrt{a}+\sqrt{1-a},\qquad b(x_l,y_h;a)\le b(x,y;a)\le b(x_h,y_l;a). \tag*{(129)} 1 ≤ b ≤ a + 1 − a , b ( x l , y h ; a ) ≤ b ( x , y ; a ) ≤ b ( x h , y l ; a ) . (129)
Evaluate both endpoint expressions with the same unchanged interval for a a a . No monotonicity in a a a is assumed. A continuous endpoint value at x = 1 , y = 0 x=1,y=0 x = 1 , y = 0 may be used for this value bound, but cannot regularize a duration circuit whose D D D vanishes.
The function k ( t ) = t + 1 − t k(t)=\sqrt{t}+\sqrt{1-t} k ( t ) = t + 1 − t is strictly concave; its maximum is 2 \sqrt{2} 2 if the interval contains 1 / 2 1/2 1/2 , and otherwise is the larger endpoint value. Its minimum is the smaller endpoint value. Together with the positive gaps proved above this gives the whole-domain bounds
0 < g i j ≤ k − 1 , 0 < L < 2 , 0 < Δ ≤ 4. (130) 0<g_{ij}\leq k-1,\qquad0<L<2,\qquad0<\Delta\leq4. \tag*{(130)} 0 < g ij ≤ k − 1 , 0 < L < 2 , 0 < Δ ≤ 4. (130)
If an interval enclosure for L L L has been validly tightened to [ l , h ] ⊂ [ 0 , 2 ] [l,h]\subset[0,2] [ l , h ] ⊂ [ 0 , 2 ] , monotonicity of L ( 4 − L ) L(4-L) L ( 4 − L ) gives the value range [ l ( 4 − l ) , h ( 4 − h ) ] [l(4-l),h(4-h)] [ l ( 4 − l ) , h ( 4 − h )] with outward endpoints. A lower bound of zero stays zero unless the arithmetic separates it; abstract strict positivity is not a numerical positive floor.
Every intersection in (129) –(130) concerns the zeroth coefficient only. All higher coefficients remain enclosures of the true derivatives of the original analytic expression. In particular one never differentiates an endpoint envelope, a minimum, a constant clipped bound, or a candidate restriction. Valid value bounds may of course be used as value inputs in subsequent exact chain rules.
True Taylor coefficients
For a smooth scalar function f f f on a rectangle X X X with center m m m , let f α = ∂ α f / α ! f_\alpha=\partial^\alpha f/\alpha! f α = ∂ α f / α ! . For an integer n ≥ 1 n\geq1 n ≥ 1 , a valid centered enclosure is
f ( X ) ⊆ ∑ ∣ α ∣ < n [ f α ( m ) ] ( X − m ) α + ∑ ∣ α ∣ = n [ f α ( X ) ] ( X − m ) α . (131) f(X)\subseteq\sum_{|\alpha|<n}[f_\alpha(m)](X-m)^\alpha+\sum_{|\alpha|=n}[f_\alpha(X)](X-m)^\alpha. \tag*{(131)} f ( X ) ⊆ ∣ α ∣ < n ∑ [ f α ( m )] ( X − m ) α + ∣ α ∣ = n ∑ [ f α ( X )] ( X − m ) α . (131)
Brackets denote outward enclosures. To prove it, apply one-variable Taylor’s theorem with integral remainder to f ( m + θ ( x − m ) ) f(m+\theta(x-m)) f ( m + θ ( x − m )) and expand each directional derivative by the multinomial theorem. The integral remainder has total weight one with this factorial normalization. Center coefficients themselves must be rigorously enclosed. A degree-n n n center polynomial without a higher-order remainder is not this rule.
The implementations carry normalized multivariate jets for rational operations, positive square roots, and true logarithms. Multiplication uses convolution; division and square roots use the coefficient recurrences obtained from f g = 1 fg=1 f g = 1 and g 2 = f g^2=f g 2 = f . Before either recurrence, the full-box constant term is separated from zero with the required sign. Composition uses the actual normalized derivatives of the scalar primitive. The proof of enclosure is induction through this finite straight-line expression, followed by (131) . A second centered tightening, used in descending coefficient order, is
[ f α ( X ) ] ∩ ( [ f α ( m ) ] + ∑ j ( α j + 1 ) [ f α + e j ( X ) ] ( X j − m j ) ) . [f_\alpha(X)]\cap\left([f_\alpha(m)]+\sum_j(\alpha_j+1)[f_{\alpha+e_j}(X)](X_j-m_j)\right). [ f α ( X )] ∩ ( [ f α ( m )] + j ∑ ( α j + 1 ) [ f α + e j ( X )] ( X j − m j ) ) .
It follows from the mean-value integral along the same segment and ∂ j f α = ( α j + 1 ) f α + e j \partial_jf_\alpha=(\alpha_j+1)f_{\alpha+e_j} ∂ j f α = ( α j + 1 ) f α + e j . For a gradient target, order n n n coefficients of G G G , Δ \Delta Δ yield coefficients through order n − 1 n-1 n − 1 of N r N_r N r . Natural and centered ranges may be intersected only after each independently encloses the same whole-box function.
The implemented logarithm value primitive uses rational Padé bounds. For r > 0 r>0 r > 0 put
P 4 ( r ) = r 4 + 10 r 3 + 20 r 2 + 10 r + 1 , P_4(r)=r^4+10r^3+20r^2+10r+1, P 4 ( r ) = r 4 + 10 r 3 + 20 r 2 + 10 r + 1 ,
Q 4 ( r ) = r 4 + 16 r 3 + 36 r 2 + 16 r + 1 , Q_4(r)=r^4+16r^3+36r^2+16r+1, Q 4 ( r ) = r 4 + 16 r 3 + 36 r 2 + 16 r + 1 ,
ℓ ( r ) = 5 ( r − 1 ) ( r + 1 ) ( 5 r 2 + 32 r + 5 ) 12 Q 4 ( r ) , \ell(r)=\frac{5(r-1)(r+1)(5r^2+32r+5)}{12Q_4(r)}, ℓ ( r ) = 12 Q 4 ( r ) 5 ( r − 1 ) ( r + 1 ) ( 5 r 2 + 32 r + 5 ) ,
u ( r ) = ( r − 1 ) ( r + 1 ) ( r 4 + 101 r 3 + 426 r 2 + 101 r + 1 ) 60 r P 4 ( r ) . u(r)=\frac{(r-1)(r+1)(r^4+101r^3+426r^2+101r+1)}{60rP_4(r)}. u ( r ) = 60 r P 4 ( r ) ( r − 1 ) ( r + 1 ) ( r 4 + 101 r 3 + 426 r 2 + 101 r + 1 ) .
Their exact error derivatives are
1 2 r − ℓ ′ ( r ) = ( r − 1 ) 8 2 r Q 4 ( r ) 2 , u ′ ( r ) − 1 2 r = ( r − 1 ) 10 60 r 2 P 4 ( r ) 2 . \frac{1}{2r}-\ell'(r)=\frac{(r-1)^8}{2rQ_4(r)^2},\qquad u'(r)-\frac{1}{2r}=\frac{(r-1)^{10}}{60r^2P_4(r)^2}. 2 r 1 − ℓ ′ ( r ) = 2 r Q 4 ( r ) 2 ( r − 1 ) 8 , u ′ ( r ) − 2 r 1 = 60 r 2 P 4 ( r ) 2 ( r − 1 ) 10 .
All three functions vanish at r = 1 r=1 r = 1 . Integration proves ℓ ( r ) ≤ 1 2 log r ≤ u ( r ) \ell(r)\leq\frac{1}{2}\log r\leq u(r) ℓ ( r ) ≤ 2 1 log r ≤ u ( r ) for r ≥ 1 r\geq1 r ≥ 1 , with the inequalities reversed for 0 < r ≤ 1 0<r\leq1 0 < r ≤ 1 . The code evaluates a hull of both expressions at r = q 2 r=q^2 r = q 2 , so it also handles a positive interval crossing q = 1 q=1 q = 1 . Repeated positive square roots and log q = 256 log ( q 1 / 256 ) \log q=256\log(q^{1/256}) log q = 256 log ( q 1/256 ) tighten the value bound where used. Higher logarithm coefficients are always ( − 1 ) k − 1 / ( k q k ) (-1)^{k-1}/(kq^k) ( − 1 ) k − 1 / ( k q k ) , k ≥ 1 k\geq1 k ≥ 1 , not derivatives of either rational bound. The same distinction applies to the primitive ϕ \phi ϕ below.
Binary input endpoints are proved to contain the exact rational box; square-root coordinate changes are checked by exact rational squaring. The arithmetic uses IEEE binary64 with each elementary rounded result extended outward by nextafter. The adopted compiler flags exclude fast-math transformations and fused contraction. The runtime checks round-to-nearest, the floating evaluation method, and the absence of flush-to-zero or denormals-are-zero modes. Nonfinite results or unseparated domains are rejected, not converted into a sign conclusion. The serialized binary endpoints are compared as their exact dyadic rational values. These conditions are part of the verification contract.
Factoring the angular endpoint
For 0 < c ≤ 1 / 2 0<c\leq1/2 0 < c ≤ 1/2 , set t = c 2 t=c^2 t = c 2 , s = 1 − c 2 s=\sqrt{1-c^2} s = 1 − c 2 , r i = u i r_i=\sqrt{u_i} r i = u i and h = − 1 / ( 1 + s ) h=-1/(1+s) h = − 1/ ( 1 + s ) . On any whole box with 0 ≤ x i ≤ 1 0\leq x_i\leq1 0 ≤ x i ≤ 1 and D i > 0 D_i>0 D i > 0 the following analytic identities are valid. Define
J i = 2 x i ( 1 − x i + 1 ) r i + 1 ( b i + 1 ) ( r i + z i ) ( 1 + D i ) , z i = a i D i r i + 1 . J_i=\frac{2x_i(1-x_{i+1})}{r_{i+1}(b_i+1)(r_i+\sqrt{z_i})(1+\sqrt{D_i})},\qquad
\sqrt{z_i}=\frac{\sqrt{a_i}\sqrt{D_i}}{r_{i+1}}. J i = r i + 1 ( b i + 1 ) ( r i + z i ) ( 1 + D i ) 2 x i ( 1 − x i + 1 ) , z i = r i + 1 a i D i .
Rationalizing ( b i − 1 ) ( b i + 1 ) (b_i-1)(b_i+1) ( b i − 1 ) ( b i + 1 ) gives b i − 1 = a i ( 1 − a i ) J i b_i-1=\sqrt{a_i}(1-a_i)J_i b i − 1 = a i ( 1 − a i ) J i . For even i i i put β i = s 2 J i \beta_i=s^2J_i β i = s 2 J i , and for odd i i i put η i = s J i \eta_i=sJ_i η i = s J i . Thus b i = 1 + c β i b_i=1+c\beta_i b i = 1 + c β i or 1 + c 2 η i 1+c^2\eta_i 1 + c 2 η i , respectively. For adjacent even E E E and odd O O O put
G E O = 1 − β E + c ( h − η O ) − c 2 β E η O , g E O = c G E O . \mathcal{G}_{EO}=1-\beta_E+c(h-\eta_O)-c^2\beta_E\eta_O,\qquad g_{EO}=c\mathcal{G}_{EO}. G E O = 1 − β E + c ( h − η O ) − c 2 β E η O , g E O = c G E O .
Write G i \mathcal{G}_i G i for the gap between i , i + 1 i,i+1 i , i + 1 . Then
L ‾ = c G 0 G 2 + s G 3 G 1 ( c + s ) s , D ‾ = L ‾ ( 4 − c L ‾ ) , L = c L ‾ , Δ = c D ‾ . (132) \overline{L}=\frac{c\mathcal{G}_0\mathcal{G}_2+s\mathcal{G}_3\mathcal{G}_1}{(c+s)s},\qquad
\overline{D}=\overline{L}(4-c\overline{L}),\qquad
L=c\overline{L},\qquad
\Delta=c\overline{D}. \tag*{(132)} L = ( c + s ) s c G 0 G 2 + s G 3 G 1 , D = L ( 4 − c L ) , L = c L , Δ = c D . (132)
These last two positive scales must not be identified.
For completeness the factored numerator circuit is as follows; local subscripts on b b b start at pair i i i , and g 12 ′ = G i + 1 g'_{12}=\mathcal{G}_{i+1} g 12 ′ = G i + 1 , g 23 ′ = G i + 2 g'_{23}=\mathcal{G}_{i+2} g 23 ′ = G i + 2 . For even i i i set ρ = − x i / ( 1 + r i ) \rho=-x_i/(1+r_i) ρ = − x i / ( 1 + r i ) and J ′ = 1 + c ( h − ρ − η i + 3 ) − c 3 ρ η i + 3 J'=1+c(h-\rho-\eta_{i+3})-c^3\rho\eta_{i+3} J ′ = 1 + c ( h − ρ − η i + 3 ) − c 3 ρ η i + 3 , then
τ 2 i = 2 g 12 ′ J ′ c + s − 2 r i b 1 g 23 ′ ( c + s ) s − L ‾ , \tau_{2i}=\frac{2g'_{12}J'}{c+s}-\frac{2r_i b_1g'_{23}}{(c+s)s}-\overline{L}, τ 2 i = c + s 2 g 12 ′ J ′ − ( c + s ) s 2 r i b 1 g 23 ′ − L ,
τ 2 i + 1 = − τ 2 i − 2 D i ( s b 3 g 12 ′ + c b 1 g 23 ′ ) r i + 1 ( c + s ) s . \tau_{2i+1}=-\tau_{2i}-\frac{2\sqrt{D_i}(s b_3g'_{12}+c b_1g'_{23})}{r_{i+1}(c+s)s}. τ 2 i + 1 = − τ 2 i − r i + 1 ( c + s ) s 2 D i ( s b 3 g 12 ′ + c b 1 g 23 ′ ) .
For odd i i i set
e ′ = − 2 r i ( c b 3 g 12 ′ + s b 1 g 23 ′ ) ( c + s ) s , H i = 1 − x i + 1 s 2 D i , q = 1 + c 2 H i , M i = H i q ( q + 1 ) , e'=-\frac{2r_i(c b_3g'_{12}+s b_1g'_{23})}{(c+s)s},\qquad
H_i=\frac{1-x_{i+1}}{s^2D_i},\qquad
q=\sqrt{1+c^2H_i},\qquad
M_i=\frac{H_i}{q(q+1)}, e ′ = − ( c + s ) s 2 r i ( c b 3 g 12 ′ + s b 1 g 23 ′ ) , H i = s 2 D i 1 − x i + 1 , q = 1 + c 2 H i , M i = q ( q + 1 ) H i ,
κ 2 i = e ′ + c ( 2 g 12 ′ / s − L ‾ ) , τ 2 i + 1 = L ‾ − 2 g 12 ′ / s − c M i e ′ . \kappa_{2i}=e'+c(2g'_{12}/s-\overline{L}),\qquad
\tau_{2i+1}=\overline{L}-2g'_{12}/s-cM_i e'. κ 2 i = e ′ + c ( 2 g 12 ′ / s − L ) , τ 2 i + 1 = L − 2 g 12 ′ / s − c M i e ′ .
Substitution into (33.1) proves σ 2 i = c τ 2 i \sigma_{2i}=c\tau_{2i} σ 2 i = c τ 2 i for even i i i , σ 2 i = κ 2 i \sigma_{2i}=\kappa_{2i} σ 2 i = κ 2 i for odd i i i , and σ 2 i + 1 = c τ 2 i + 1 \sigma_{2i+1}=c\tau_{2i+1} σ 2 i + 1 = c τ 2 i + 1 for all $i.
Put K 2 i = r i r i + 1 / D i K_{2i}=r_i r_{i+1}/\sqrt{D_i} K 2 i = r i r i + 1 / D i for even i i i , and define
A = τ 0 τ 1 + τ 3 τ 4 + τ 4 τ 5 + τ 7 τ 0 − τ 1 τ 3 − τ 5 τ 7 , \mathcal{A}=\tau_0\tau_1+\tau_3\tau_4+\tau_4\tau_5+\tau_7\tau_0-\tau_1\tau_3-\tau_5\tau_7, A = τ 0 τ 1 + τ 3 τ 4 + τ 4 τ 5 + τ 7 τ 0 − τ 1 τ 3 − τ 5 τ 7 ,
R s = τ 0 2 log K 0 + τ 4 2 log K 4 + ∑ j o d d τ j 2 log q j , R_s=\tau_0^2\log K_0+\tau_4^2\log K_4+\sum_{j\ \mathrm{odd}}\tau_j^2\log q_j, R s = τ 0 2 log K 0 + τ 4 2 log K 4 + j odd ∑ τ j 2 log q j ,
Z i = 3 q 2 i ( q 2 i + 1 ) − 1 2 ϕ ( c 2 H i ) ( i = 1 , 3 ) , Z_i=\frac{3}{q_{2i}(q_{2i}+1)}-\frac{1}{2}\phi(c^2H_i)\quad(i=1,3), Z i = q 2 i ( q 2 i + 1 ) 3 − 2 1 ϕ ( c 2 H i ) ( i = 1 , 3 ) ,
W = 3 A + R s − H 1 κ 2 2 Z 1 − H 3 κ 6 2 Z 3 − 3 c D ‾ ( M 1 + M 3 ) , T = τ 0 2 + τ 4 2 , \mathcal{W}=3\mathcal{A}+R_s-H_1\kappa_2^2Z_1-H_3\kappa_6^2Z_3-3c\overline{D}(M_1+M_3),\qquad\mathcal{T}=\tau_0^2+\tau_4^2, W = 3 A + R s − H 1 κ 2 2 Z 1 − H 3 κ 6 2 Z 3 − 3 c D ( M 1 + M 3 ) , T = τ 0 2 + τ 4 2 ,
where ϕ ( z ) = log ( 1 + z ) / z \phi(z)=\log(1+z)/z ϕ ( z ) = log ( 1 + z ) / z , ϕ ( 0 ) = 1 \phi(0)=1 ϕ ( 0 ) = 1 . The resulting cost identity is
cost = 2 − c W D ‾ + c log c T D ‾ , \operatorname{cost}=2-\frac{c\mathcal{W}}{\overline{D}}+\frac{c\log c\,\mathcal{T}}{\overline{D}}, cost = 2 − D c W + D c log c T ,
( cost − C ∗ ) D ‾ = ( 2 − C ∗ ) D ‾ − c W + c log c T . (133) (\operatorname{cost}-C_*)\overline{D}=(2-C_*)\overline{D}-c\mathcal{W}+c\log c\,\mathcal{T}. \tag*{(133)} ( cost − C ∗ ) D = ( 2 − C ∗ ) D − c W + c log c T . (133)
Here is a short check of its cancellation. The local reciprocal identity in the cost proof, multiplied by Δ \Delta Δ , becomes
Δ = ( q j − 1 ) σ j 2 + q j { σ j ( σ j − 1 + σ j + 1 ) + σ j − 1 σ j + 1 } . \Delta=(q_j-1)\sigma_j^2+q_j\{\sigma_j(\sigma_{j-1}+\sigma_{j+1})+\sigma_{j-1}\sigma_{j+1}\}. Δ = ( q j − 1 ) σ j 2 + q j { σ j ( σ j − 1 + σ j + 1 ) + σ j − 1 σ j + 1 } .
At j = 2 , 6 j=2,6 j = 2 , 6 , use q 2 i − 1 = c 2 H i / ( q 2 i + 1 ) q_{2i}-1=c^2H_i/(q_{2i}+1) q 2 i − 1 = c 2 H i / ( q 2 i + 1 ) and the factored numerators above. It gives
κ 2 i ( τ 2 i − 1 + τ 2 i + 1 ) = D ‾ q 2 i − c ( M i κ 2 i 2 + τ 2 i − 1 τ 2 i + 1 ) . \kappa_{2i}(\tau_{2i-1}+\tau_{2i+1})=\frac{\overline{D}}{q_{2i}}-c(M_i\kappa_{2i}^2+\tau_{2i-1}\tau_{2i+1}). κ 2 i ( τ 2 i − 1 + τ 2 i + 1 ) = q 2 i D − c ( M i κ 2 i 2 + τ 2 i − 1 τ 2 i + 1 ) .
Substitute these two equations into the adjacent sum in G G G , and split log q 0 = log K 0 − log c \log q_0=\log K_0-\log c log q 0 = log K 0 − log c , log q 4 = log K 4 − log c \log q_4=\log K_4-\log c log q 4 = log K 4 − log c . The odd-pair logs are c 2 H i ϕ ( c 2 H i ) / 2 c^2H_i\phi(c^2H_i)/2 c 2 H i ϕ ( c 2 H i ) /2 . This proves (33.7) by exact algebra.
Thus a positive enclosure for E + c log c T E+c\log c\,\mathcal{T} E + c log c T has the same cost meaning as P > 0 P>0 P > 0 , because D ‾ = Δ / c > 0 \overline{D}=\Delta/c>0 D = Δ/ c > 0 at candidates. If a calculation claims a positive whole-box discriminant interval, it separately checks the lower endpoint of the scale actually returned. No division by a zero-containing interval is permitted. The broad endpoint program’s output named “delta” is D ‾ \overline{D} D ; the normalized program returns Δ \Delta Δ .
The formula extends to c = 0 c=0 c = 0 on the compact shape subdomains used by the inherited cost and collar results: their u i , D i u_i,D_i u i , D i have positive lower bounds, and the displayed analytic factors remain regular. Set c log c = 0 c\log c=0 c log c = 0 there. Only the continuous value hull of this last factor is used on a box touching zero. On positive-c c c boxes its true logarithm jet is available. These are different evaluation cases.
For z ≥ 0 z\ge0 z ≥ 0 , differentiation under the integral sign gives
( − 1 ) k ϕ ( k ) ( z ) k ! = A k ( z ) : = ∫ 0 1 v k ( 1 + z v ) k + 1 d v > 0 , A k ′ ( z ) < 0. \frac{(-1)^k\phi^{(k)}(z)}{k!}=A_k(z):=\int_0^1\frac{v^k}{(1+zv)^{k+1}}\,dv>0,\qquad A_k'(z)<0. k ! ( − 1 ) k ϕ ( k ) ( z ) = A k ( z ) := ∫ 0 1 ( 1 + z v ) k + 1 v k d v > 0 , A k ′ ( z ) < 0.
With r = z / ( 1 + z ) r=z/(1+z) r = z / ( 1 + z ) , substitution yields the positive series
A k ( z ) = ( 1 + z ) − k − 1 ∑ n ≥ 0 r n n + k + 1 . A_k(z)=(1+z)^{-k-1}\sum_{n\ge0}\frac{r^n}{n+k+1}. A k ( z ) = ( 1 + z ) − k − 1 n ≥ 0 ∑ n + k + 1 r n .
After n = 0 , … , N n=0,\ldots,N n = 0 , … , N , the tail lies between zero and ( 1 + z ) − k − 1 r N + 1 / [ ( N + k + 2 ) ( 1 − r ) ] (1+z)^{-k-1}r^{N+1}/[(N+k+2)(1-r)] ( 1 + z ) − k − 1 r N + 1 / [( N + k + 2 ) ( 1 − r )] . Monotonicity supplies whole-argument coefficient enclosures from the two endpoints. For larger positive arguments the equivalent exact recurrence
z ϕ k ( z ) + ϕ k − 1 ( z ) = ( − 1 ) k − 1 k ( 1 + z ) k , k ≥ 1 , ϕ k = ϕ ( k ) k ! z\phi_k(z)+\phi_{k-1}(z)=\frac{(-1)^{k-1}}{k(1+z)^k},\qquad k\ge1,\qquad\phi_k=\frac{\phi^{(k)}}{k!} z ϕ k ( z ) + ϕ k − 1 ( z ) = k ( 1 + z ) k ( − 1 ) k − 1 , k ≥ 1 , ϕ k = k ! ϕ ( k )
starts from a true logarithm value bound. The code uses the series through N = 48 N=48 N = 48 for small arguments (with stored binary64 threshold 0.8 0.8 0.8 ) and the recurrence above that threshold. Both formulas are valid on their checked domains. This computes enclosures of actual derivatives, not derivatives of a logarithm approximation.
Candidate preserving interval contractions
For one adjacent shape pair put x = x i x=x_i x = x i , y = x i + 1 y=x_{i+1} y = x i + 1 , b = 1 − a i b=1-a_i b = 1 − a i and K = Q 2 − 1 K=Q^2-1 K = Q 2 − 1 . The cap inequalities are
1 − x + b x y ≥ e , K a i ( 1 − x + b x y ) ≥ b ( 1 − y ) . 1-x+bxy\ge e,\qquad Ka_i(1-x+bxy)\ge b(1-y). 1 − x + b x y ≥ e , K a i ( 1 − x + b x y ) ≥ b ( 1 − y ) .
Suppose a current rectangle has endpoints a l a_l a l , a h a_h a h , x l x_l x l , x h x_h x h , y l y_l y l , y h y_h y h , and write b h = 1 − a l b_h=1-a_l b h = 1 − a l , b 0 = 1 − a h b_0=1-a_h b 0 = 1 − a h . Solving these inequalities and taking monotone endpoint bounds gives
x ≤ 1 − e 1 − b h y h , y ≥ e − 1 + x l b h x l , x\le\frac{1-e}{1-b_hy_h},\qquad y\ge\frac{e-1+x_l}{b_hx_l}, x ≤ 1 − b h y h 1 − e , y ≥ b h x l e − 1 + x l ,
y ≥ b 0 − K a h ( 1 − x l ) b 0 ( 1 + K a h x l ) , x ≤ 1 − b 0 ( 1 − y h ) / ( K a h ) 1 − b 0 y h . y\ge\frac{b_0-Ka_h(1-x_l)}{b_0(1+Ka_hx_l)},\qquad x\le\frac{1-b_0(1-y_h)/(Ka_h)}{1-b_0y_h}. y ≥ b 0 ( 1 + K a h x l ) b 0 − K a h ( 1 − x l ) , x ≤ 1 − b 0 y h 1 − b 0 ( 1 − y h ) / ( K a h ) .
Each update is used only when its displayed denominator is strictly positive; a lower bound with nonpositive numerator need not tighten y ≥ 0 y\ge0 y ≥ 0 . For example the nonconstant part of the even-duration square is b ( 1 − y ) / [ a ( 1 − x + b x y ) ] b(1-y)/[a(1-x+bxy)] b ( 1 − y ) / [ a ( 1 − x + b x y )] , increasing in x x x and decreasing in a , y a,y a , y . These signs verify the endpoint choices in the second row.
The two basic canonical comparisons similarly give
x 1 ≥ 1 − t h 2 x 0 , h ( 1 − t h ) ( 1 − t h x 0 , h ) , x_1\ge1-\frac{t_h^2x_{0,h}}{(1-t_h)(1-t_hx_{0,h})}, x 1 ≥ 1 − ( 1 − t h ) ( 1 − t h x 0 , h ) t h 2 x 0 , h ,
x 0 ≤ x 3 , h 1 − x 1 , h + t l ( x 1 , h + x 3 , h ) , x 3 ≥ x 0 , l [ 1 − ( 1 − t l ) x 1 , h ] 1 − t l x 0 , l . x_0\le\frac{x_{3,h}}{1-x_{1,h}+t_l(x_{1,h}+x_{3,h})},\qquad x_3\ge\frac{x_{0,l}[1-(1-t_l)x_{1,h}]}{1-t_lx_{0,l}}. x 0 ≤ 1 − x 1 , h + t l ( x 1 , h + x 3 , h ) x 3 , h , x 3 ≥ 1 − t l x 0 , l x 0 , l [ 1 − ( 1 − t l ) x 1 , h ] .
The last two follow from x 0 u 1 ≤ x 3 u 0 x_0u_1\le x_3u_0 x 0 u 1 ≤ x 3 u 0 ; differentiation of the positive quotients supplies the stated monotone endpoint choices. For i = 1 , 2 , 3 i=1,2,3 i = 1 , 2 , 3 , the remaining even comparisons imply
1 − x 1 ≤ f i ( t ) ( 1 − x i + 1 ) D 0 / D i , f 1 = f 3 = ( t / ( 1 − t ) ) 2 , f 2 = 1. 1-x_1\le f_i(t)(1-x_{i+1})D_0/D_i,\qquad f_1=f_3=(t/(1-t))^2,\qquad f_2=1. 1 − x 1 ≤ f i ( t ) ( 1 − x i + 1 ) D 0 / D i , f 1 = f 3 = ( t / ( 1 − t ) ) 2 , f 2 = 1.
A whole-box upper bound for this positive right side gives another candidate lower bound for x 1 x_1 x 1 . All such contractions are rounded outward on the rational lattice 2 − 36 Z 2^{-36}\mathbb{Z} 2 − 36 Z . Each application contains every previously retained candidate; their composition does also, regardless of whether it finds the best possible hull. The numerical function and its derivatives are then evaluated on the actual resulting rectangle, not clipped during evaluation to the unrecorded candidate subset. Algebraically equivalent contractors in a rechart follow by substituting its exact formulas and checking the same positive-denominator conditions.
Recharts and source qualified boxes
The broad chart uses c = t c=\sqrt{t} c = t . An upper-end box is first pair-recut so its new angle is 1 − t 1-t 1 − t , then enclosed by an outward c c c rectangle. All jets are newly seeded in these coordinates. The thin chart actually used in the endpoint coverings has coordinates ( c , x 0 , z , x 2 , x 3 ) (c,x_0,z,x_2,x_3) ( c , x 0 , z , x 2 , x 3 ) and
t = c 2 , x 1 = 1 − t 2 x 0 z ( 1 − t ) ( 1 − t x 0 ) . (134) t=c^2,\qquad x_1=1-\frac{t^2x_0z}{(1-t)(1-tx_0)}. \tag*{(134)} t = c 2 , x 1 = 1 − ( 1 − t ) ( 1 − t x 0 ) t 2 x 0 z . (134)
For a strict canonical point its inverse satisfies z = ( 1 − x 1 ) ( 1 − t ) ( 1 − t x 0 ) / ( t 2 x 0 ) ∈ [ 0 , 1 ] z=(1-x_1)(1-t)(1-tx_0)/(t^2x_0)\in[0,1] z = ( 1 − x 1 ) ( 1 − t ) ( 1 − t x 0 ) / ( t 2 x 0 ) ∈ [ 0 , 1 ] . Its shape derivative with respect to z z z is strictly negative at that point. The canonical face z = 1 z=1 z = 1 is not a feasibility boundary for local competitors. With A = x 0 z / [ ( 1 − t ) ( 1 − t x 0 ) ] A=x_0z/[(1-t)(1-tx_0)] A = x 0 z / [( 1 − t ) ( 1 − t x 0 )] , the exact substitutions are
u 1 = t V , D 1 = t W , V = 1 + t x 0 z 1 − t x 0 , W = x 2 + t A ( 1 − t x 2 ) . u_1=tV,\qquad D_1=tW,\qquad V=1+\frac{tx_0z}{1-tx_0},\qquad W=x_2+tA(1-tx_2). u 1 = t V , D 1 = t W , V = 1 + 1 − t x 0 t x 0 z , W = x 2 + t A ( 1 − t x 2 ) .
Hence u 1 = c V \sqrt{u_1}=c\sqrt{V} u 1 = c V and D 1 = c W \sqrt{D_1}=c\sqrt{W} D 1 = c W on its positive-c c c domain. The square roots of V , W V,W V , W still require full-box positive bounds.
A further cap chart, used on marked disjoint shape pairs, puts b = 1 − a i b=1-a_i b = 1 − a i and
x i = 1 − e z , U = b + ( 1 − b ) e z , x i + 1 = e − e z + ( U − e ) h b ( 1 − e z ) , D i = e + ( U − e ) h . (135) x_i=1-ez,\qquad U=b+(1-b)ez,\qquad x_{i+1}=\frac{e-ez+(U-e)h}{b(1-ez)},\qquad D_i=e+(U-e)h. \tag*{(135)} x i = 1 − ez , U = b + ( 1 − b ) ez , x i + 1 = b ( 1 − ez ) e − ez + ( U − e ) h , D i = e + ( U − e ) h . (135)
For the marked pair used in the accepted cap-chart cover, i = 1 i=1 i = 1 , t ∈ [ 1 / 5 , 4 / 5 ] t\in[1/5,4/5] t ∈ [ 1/5 , 4/5 ] , x 1 ∈ [ 1 − e , 1 ] x_1\in[1-e,1] x 1 ∈ [ 1 − e , 1 ] , and the other shapes are at most 1 − e 1-e 1 − e . Thus b = t > e b=t>e b = t > e . Set A 0 = e ( 1 − z ) / [ b ( 1 − e z ) ] ∈ [ 0 , 1 ) A_0=e(1-z)/[b(1-ez)]\in[0,1) A 0 = e ( 1 − z ) / [ b ( 1 − ez )] ∈ [ 0 , 1 ) ; then x i + 1 = A 0 + ( 1 − A 0 ) h x_{i+1}=A_0+(1-A_0)h x i + 1 = A 0 + ( 1 − A 0 ) h . At a retained candidate,
z = ( 1 − x i ) / e ∈ [ 0 , 1 ] , h = ( D i − e ) / ( U − e ) = ( x i + 1 − A 0 ) / ( 1 − A 0 ) ∈ [ 0 , x i + 1 ] ⊂ [ 0 , 1 ) . z=(1-x_i)/e\in[0,1],\qquad h=(D_i-e)/(U-e)=(x_{i+1}-A_0)/(1-A_0)\in[0,x_{i+1}]\subset[0,1). z = ( 1 − x i ) / e ∈ [ 0 , 1 ] , h = ( D i − e ) / ( U − e ) = ( x i + 1 − A 0 ) / ( 1 − A 0 ) ∈ [ 0 , x i + 1 ] ⊂ [ 0 , 1 ) .
Here U − e > 0 U-e>0 U − e > 0 and the shape Jacobian is − e ( 1 − A 0 ) ≠ 0 -e(1-A_0)\ne0 − e ( 1 − A 0 ) = 0 . The whole binary parameter box independently checks b l > e h b_l>e h b l > e h , h h < 1 h_h<1 h h < 1 and all unmarked D i > 0 D_i>0 D i > 0 . The generic disjoint-pair formula does not assert that every possible mask was used numerically. A source may constrain its image further. In particular, not every point of a parameter rectangle need map into its recorded original source box. Coverage therefore carries both the parameter box and that source condition.
If an independently enclosed whole image I ( B ) I(B) I ( B ) is intersected with an original source box A A A , the intersection contains every relevant candidate image. A normalized value cost test on this intersection can exclude those candidates, once its own complete domain checks pass. It does not replace a parameter jet by derivatives restricted to I ( B ) ∩ A I(B)\cap A I ( B ) ∩ A . Parameter derivatives are always computed on the unchanged parameter rectangle through the exact map. Empty intersections are candidate exclusions, not pointwise statements about all parameter values. This distinction is essential for the endpoint roots 44 and 60.
For comparison with older raw circuits, if G r a w = Π G G_{\mathrm{raw}}=\Pi G G raw = Π G and V r a w = Π Δ V_{\mathrm{raw}}=\Pi\Delta V raw = ΠΔ , then
( 8 − C ∗ ) V r a w − G r a w = Π P , G r a w ∂ r V r a w − V r a w ∂ r G r a w = Π 2 N r . (8-C_*)V_{\mathrm{raw}}-G_{\mathrm{raw}}=\Pi P,\qquad G_{\mathrm{raw}}\partial_rV_{\mathrm{raw}}-V_{\mathrm{raw}}\partial_rG_{\mathrm{raw}}=\Pi^2N_r. ( 8 − C ∗ ) V raw − G raw = Π P , G raw ∂ r V raw − V raw ∂ r G raw = Π 2 N r .
The derivatives of Π \Pi Π cancel in the cleared gradient. The raw derivatives themselves do not simply scale by Π \Pi Π .
The regularized thin expression
Some accepted thin-chart trees use a further algebraic regularization, without changing the five independent variables ( c , x 0 , z , x 2 , x 3 ) (c,x_0,z,x_2,x_3) ( c , x 0 , z , x 2 , x 3 ) . Its whole-box contract requires 0 < c l ≤ c h ≤ 1 / 2 0<c_l\le c_h\le1/2 0 < c l ≤ c h ≤ 1/2 , all four other intervals in [ 0 , 1 ] [0,1] [ 0 , 1 ] , x 2 , l > 0 x_{2,l}>0 x 2 , l > 0 , and positive lower bounds
D 0 ≥ 1 − t h / ( 1 − t h ) ≥ 2 / 3 , D 1 ≥ t l x 2 , l , D 2 ≥ 1 − x 2 , h + ( 1 − t h ) x 2 , l x 3 , l , D 3 ≥ 1 − x 3 , h + t l x 3 , l x 0 , l . \begin{aligned}
D_0&\ge1-t_h/(1-t_h)\ge2/3, & D_1&\ge t_lx_{2,l},\\
D_2&\ge1-x_{2,h}+(1-t_h)x_{2,l}x_{3,l}, & D_3&\ge1-x_{3,h}+t_lx_{3,l}x_{0,l}.
\end{aligned} D 0 D 2 ≥ 1 − t h / ( 1 − t h ) ≥ 2/3 , ≥ 1 − x 2 , h + ( 1 − t h ) x 2 , l x 3 , l , D 1 D 3 ≥ t l x 2 , l , ≥ 1 − x 3 , h + t l x 3 , l x 0 , l .
where t l = c l 2 t_l=c_l^2 t l = c l 2 , t h = c h 2 t_h=c_h^2 t h = c h 2 . The external checker verifies these conditions as well as the runtime positivity checks. They cannot be replaced by assuming that candidate points alone have positive D i D_i D i .
With V , W V,W V , W from (134) and the following lines, set
Y 3 = ( 1 − x 3 ) / t , R 1 = V , R 3 = x 3 + Y 3 , S 1 = W , S 3 = x 3 x 0 + Y 3 . Y_3=(1-x_3)/t,\qquad R_1=\sqrt{V},\qquad R_3=\sqrt{x_3+Y_3},\qquad S_1=\sqrt{W},\qquad S_3=\sqrt{x_3x_0+Y_3}. Y 3 = ( 1 − x 3 ) / t , R 1 = V , R 3 = x 3 + Y 3 , S 1 = W , S 3 = x 3 x 0 + Y 3 .
Then r i = c R i r_i=cR_i r i = c R i and D i = c S i \sqrt{D_i}=cS_i D i = c S i for odd i i i . The bounds give R i ≥ 1 R_i\ge1 R i ≥ 1 , S i > 0 S_i>0 S i > 0 , r 0 , r 2 , s ≥ 3 / 4 r_0,r_2,s\ge\sqrt{3}/4 r 0 , r 2 , s ≥ 3 /4 , and 8 / 9 ≤ x 1 ≤ 1 8/9\le x_1\le1 8/9 ≤ x 1 ≤ 1 . Thus all added denominators and radicals are strictly positive on the whole rectangle. There is no extension through c = 0 c=0 c = 0 or through x 2 , l = 0 x_{2,l}=0 x 2 , l = 0 in this contract.
Define rescaled quantities, still as functions of the same five variables,
T 1 = c τ 1 , T 5 = c τ 5 , κ ˉ 2 i = κ 2 i / c , H ˉ i = t H i ( i = 1 , 3 ) , A ˉ = c A . T_1=c\tau_1,\qquad T_5=c\tau_5,\qquad\bar{\kappa}_{2i}=\kappa_{2i}/c,\qquad\bar{H}_i=tH_i\quad(i=1,3),\qquad\bar{A}=cA. T 1 = c τ 1 , T 5 = c τ 5 , κ ˉ 2 i = κ 2 i / c , H ˉ i = t H i ( i = 1 , 3 ) , A ˉ = c A .
In the actual evaluation these quotients are cancelled algebraically before the jet is formed: substitute r i = c R i r_i=cR_i r i = c R i , D i = c S i \sqrt{D_i}=cS_i D i = c S i in the preceding displayed circuit, and use η ˉ i = c η i \bar{\eta}_i=c\eta_i η ˉ i = c η i . For example g E O / c = 1 − β E + c h − η O − c β E η O g_{EO}/c=1-\beta_E+ch-\eta_O-c\beta_E\eta_O g E O / c = 1 − β E + c h − η O − c β E η O , so D ˉ \bar{D} D ˉ is unchanged. Two equivalent expressions for the adjacency term are
A ˉ = T 1 ( τ 0 − τ 3 ) + T 5 ( τ 4 − τ 7 ) + c ( τ 3 τ 4 + τ 7 τ 0 ) = 2 c τ 0 τ 4 − ( T 1 − c τ 4 ) ( τ 3 − τ 0 ) − ( T 5 − c τ 0 ) ( τ 7 − τ 4 ) . \begin{aligned}
\bar{A}&=T_1(\tau_0-\tau_3)+T_5(\tau_4-\tau_7)+c(\tau_3\tau_4+\tau_7\tau_0)\\
&=2c\tau_0\tau_4-(T_1-c\tau_4)(\tau_3-\tau_0)-(T_5-c\tau_0)(\tau_7-\tau_4).
\end{aligned} A ˉ = T 1 ( τ 0 − τ 3 ) + T 5 ( τ 4 − τ 7 ) + c ( τ 3 τ 4 + τ 7 τ 0 ) = 2 c τ 0 τ 4 − ( T 1 − c τ 4 ) ( τ 3 − τ 0 ) − ( T 5 − c τ 0 ) ( τ 7 − τ 4 ) .
For even i i i put B i ∗ = x i R i + 1 2 B_i^*=x_iR_{i+1}^2 B i ∗ = x i R i + 1 2 and
z 0 ∗ = t x 0 z ( 1 − t x 0 ) D 0 , z 4 ∗ = ( 1 − t ) Y 3 D 2 . z_0^*=\frac{tx_0z}{(1-tx_0)D_0},\qquad z_4^*=\frac{(1-t)Y_3}{D_2}. z 0 ∗ = ( 1 − t x 0 ) D 0 t x 0 z , z 4 ∗ = D 2 ( 1 − t ) Y 3 .
The exact identities are D i = 1 − t B i ∗ D_i=1-tB_i^* D i = 1 − t B i ∗ and q 0 2 − 1 = z 0 ∗ q_0^2-1=z_0^* q 0 2 − 1 = z 0 ∗ , q 4 2 − 1 = z 4 ∗ q_4^2-1=z_4^* q 4 2 − 1 = z 4 ∗ . Thus the arguments below are nonnegative, and positive logarithms give
log q 2 i + 1 t = B i ∗ 2 D i ϕ ( t B i ∗ / D i ) , log q 2 i = 1 2 z 2 i ∗ ϕ ( z 2 i ∗ ) ( i = 0 , 2 ) . \frac{\log q_{2i+1}}{t}=\frac{B_i^*}{2D_i}\phi(tB_i^*/D_i),\qquad\log q_{2i}=\frac{1}{2}z_{2i}^*\phi(z_{2i}^*)\quad(i=0,2). t log q 2 i + 1 = 2 D i B i ∗ ϕ ( t B i ∗ / D i ) , log q 2 i = 2 1 z 2 i ∗ ϕ ( z 2 i ∗ ) ( i = 0 , 2 ) .
Their zero-argument limits follow from analyticity of ϕ \phi ϕ . Set
R r e g = ∑ i = 0 , 2 { T 2 i + 1 2 B i ∗ 2 D i ϕ ( t B i ∗ / D i ) + T 2 i 2 z 2 i ∗ 2 ϕ ( z 2 i ∗ ) } − τ 3 2 log S 1 − τ 7 2 log S 3 , R_{\mathrm{reg}}=\sum_{i=0,2}\left\{\frac{T_{2i+1}^2B_i^*}{2D_i}\phi(tB_i^*/D_i)+\frac{T_{2i}^2z_{2i}^*}{2}\phi(z_{2i}^*)\right\}-\tau_3^2\log S_1-\tau_7^2\log S_3, R reg = i = 0 , 2 ∑ { 2 D i T 2 i + 1 2 B i ∗ ϕ ( t B i ∗ / D i ) + 2 T 2 i 2 z 2 i ∗ ϕ ( z 2 i ∗ ) } − τ 3 2 log S 1 − τ 7 2 log S 3 ,
H Z = ∑ i = 1 , 3 H ‾ i − 2 i { 3 q 2 i ( q 2 i + 1 ) − 1 2 ϕ ( H ‾ i ) } , HZ=\sum_{i=1,3}\overline{H}_{i}^{-2i}\left\{\frac{3}{q_{2i}(q_{2i}+1)}-\frac{1}{2}\phi(\overline{H}_i)\right\}, H Z = i = 1 , 3 ∑ H i − 2 i { q 2 i ( q 2 i + 1 ) 3 − 2 1 ϕ ( H i ) } ,
M s = ∑ i = 1 , 3 H ‾ i q 2 i ( q 2 i + 1 ) . M_s=\sum_{i=1,3}\frac{\overline{H}_i}{q_{2i}(q_{2i}+1)}. M s = i = 1 , 3 ∑ q 2 i ( q 2 i + 1 ) H i .
The regularized cleared cost is exactly
( 2 − C ∗ ) D ‾ − 3 A ‾ − c R r e g + c H Z + 3 D ‾ M s + c log c ( τ 3 2 + τ 7 2 ) . (136) (2-C_*)\overline{D}-3\overline{\mathcal{A}}-cR_{\mathrm{reg}}+cHZ+3\overline{D}M_s+c\log c(\tau_3^2+\tau_7^2). \tag*{(136)} ( 2 − C ∗ ) D − 3 A − c R reg + cH Z + 3 D M s + c log c ( τ 3 2 + τ 7 2 ) . (136)
To verify equality to ( c o s t − C ∗ ) D ‾ (\mathrm{cost}-C_*)\overline{D} ( cost − C ∗ ) D , observe that K 0 = c q 0 K_0=cq_0 K 0 = c q 0 , K 4 = c q 4 K_4=cq_4 K 4 = c q 4 , q 3 = 1 / ( c S 1 ) q_3=1/(cS_1) q 3 = 1/ ( c S 1 ) and q 7 = 1 / ( c S 3 ) q_7=1/(cS_3) q 7 = 1/ ( c S 3 ) . Hence
R s = R r e g + log c ( τ 0 2 + τ 4 2 − τ 3 2 − τ 7 2 ) . R_s=R_{\mathrm{reg}}+\log c(\tau_0^2+\tau_4^2-\tau_3^2-\tau_7^2). R s = R reg + log c ( τ 0 2 + τ 4 2 − τ 3 2 − τ 7 2 ) .
Substitution in (33.7) cancels the two even logarithm terms and gives (33.10), with the displayed positive sign on its final term. The other substitutions are H ‾ i κ 2 i 2 = H ‾ i κ 2 i 2 \overline{H}_i\kappa_{2i}^2=\overline{H}_i\kappa_{2i}^2 H i κ 2 i 2 = H i κ 2 i 2 and c 2 ∑ M i = M s c^2\sum M_i=M_s c 2 ∑ M i = M s . A terminal use of this circuit separately requires positive lower bounds for its margin and D ‾ \overline{D} D .
All five variables remain live jets; in particular Y 3 = ( 1 − x 3 ) / c 2 Y_3=(1-x_3)/c^2 Y 3 = ( 1 − x 3 ) / c 2 is not an independent coordinate. Nonnegativity intersections for Y 3 Y_3 Y 3 , H ‾ i \overline{H}_i H i and the ϕ \phi ϕ arguments change only their value slots. The two identical expressions for A ‾ \overline{\mathcal{A}} A may be intersected coefficientwise because every derivative is of the same analytic function.
Exact tree coverage
A binary tree record specifies its original source, chart, root rectangle, coordinate and rational split point, and two children. The checker requires the children to agree with the parent in every other coordinate and their split intervals to cover the parent’s full closed interval. Contractions require the candidate-preserving inequality and outward rounding described above. Terminal records specify the precise evaluation rectangle, program source, arithmetic order, target and bounds, or an inherited-region containment with its complete hypotheses. Starting from sound terminal exclusions, finite induction proves exclusion at the root. The final union check matches these root identities against the complete initial manifests and requires an empty uncovered set. Terminal counts alone do not establish coverage.
Verification data and reproducibility
The compact R01 verification component, identified by its complete extracted runtime and source manifest in the accompanying dependency register, separates the final proof object from the search history. Its runtime/ directory contains the typed proof forests, exact covering and chart data, retained analytic sources, and frozen interval kernels. The original five-volume R01proofbundle20261030v1 remains the unchanged provenance archive; it is not a runtime input to the compact verifier. The finite mathematical predicates and their domains are those used in § 5.15 .
The checker reconstructs rational roots, subdivisions, candidate-preserving contractions, chart maps and source-qualified frontier attachments. It regenerates the binary numerical inputs and checks the exact necessary-condition exclusions and covering joins. Finite central and lower-edge coverings and all 21 small-shape strips remain explicit dependencies. Every numerical terminal retains its whole-box domain, positive-scale and target requirements. A matching local index does not identify a source-qualified obligation. Neither an archived PASS nor a file hash proves a mathematical predicate; hashes identify files. Missing inputs or uncovered obligations are failures, not accepted exceptions.
File integrity, exact structural checking and numerical evaluation have distinct roles. The exact traversal reconstructs all 120 endpoint roots, 1 250 middle-grid cells and 1 392 finite-lemma links. A complete replay then evaluates all 955 790 typed numerical inputs across 28 profiles with the original kernels and applies their original acceptance predicates. The accompanying execution records include this complete replay, the raw mathematical outputs, the source-qualified input hashes and a relocated saved-output check. A further 54 serial evaluations compare selected mathematical output fields. Timing diagnostics are recorded separately from those fields.
The replay uses binary64 arithmetic under the contract of Appendix § 5.16 : preserved subnormal inputs and outputs, round-to-nearest, and no implicit fused multiply-add. The tested compiler is Apple clang 14.0.3 on arm64 macOS, with C++17, -O2, -fno-fast-math and -ffp-contract=off. The retained rational logarithm bounds and square-root endpoint checks provide value enclosures; a floating-point library logarithm is not accepted as an enclosure. Platform checks and a full replay establish the stated execution contract, not formal verification of the compiler or hardware.
The companion’s structural entry reconstructs its complete covering records; its full-evaluation entry additionally compiles the original kernels and evaluates every numerical obligation, including the finite-lemma components. The separate verification guide records these entry points, their required working directory, the tested environment and the exact distinction between the two outputs. Each actual execution record identifies the method used. An interrupted execution is not a completed run.
This remains a computer-assisted proof. Its explicit imported premises include the geometric correspondence and reconstruction, the invariant-K K K theorem and smoothed-octagon comparison cited in § 5.2 . The local stationarity, chart, recut, true-log and arithmetic arguments remain essential, as do conforming checkers, compiler, runtime and hardware. Compact packaging supplies neither a new purely analytic proof nor a proof-assistant formalization. Historical comparison-text and executable exceptions belong to the provenance archive and do not excuse missing compact-runtime dependencies.
The two mixed eight-arc types
The two remaining eight-arc words require different closure charts from the Gram chart of R01. Their complete-source estimates also differ. We keep the common geometric minimizing class of § 5.1.1 : the physical state lies in Ω \Omega Ω , the frame has its full signed endpoint, and the real winding is retained. In particular, the compactness bound, the strict comparison budget and every local source qualification apply to the original global minimizer. The finite exclusion below uses necessary state and cost values on this class.
Theorem 35.1. An original global minimizer cannot have eight positive maximal arcs with native labels 02102120 and twisted continuation j i + 8 = j i + 1 j_{i+8}=j_i+1 j i + 8 = j i + 1 . Thus the strict essential type R02, with increments − − − − + + - - - - + + − − − − + + , is excluded.
Theorem 35.2. An original global minimizer cannot have eight positive maximal arcs with native labels 02120120 and twisted continuation j i + 8 = j i + 1 j_{i+8}=j_i+1 j i + 8 = j i + 1 . Thus the strict essential type R03, with increments − − + + + + + + - - + + + + + + − − + + + + + + , is excluded.
All arcs in these statements have positive physical duration. A zero-duration face is retained when enclosing the candidate set, but the theorem does not identify that face with a strict eight-arc trajectory. The proofs use no five-variable Fermat equation, positive-semidefinite objective Hessian, or Krawczyk exclusion. Derivatives will enter only in rigorous bounds for function values.
Full closure and two positive charts
The multiplier q i > 1 q_i>1 q i > 1 of a complete arc has physical duration d j i log q i d_{j_i}\log q_i d j i log q i , as in Lemma 27.2. Put
N − ( q ) = ( q − q q − 1 − q ) , N + ( q ) = ( 0 q − q − 1 q ) . (137) N_-(q)=\begin{pmatrix}q&-q\\q^{-1}&-q\end{pmatrix},\qquad N_+(q)=\begin{pmatrix}0&q\\-q^{-1}&q\end{pmatrix}. \tag*{(137)} N − ( q ) = ( q q − 1 − q − q ) , N + ( q ) = ( 0 − q − 1 q q ) . (137)
The arc-frame telescope (85) , with the label conjugations transported at each switch, gives
∏ i = 0 7 N ε i ( q i ) = − I (138) \prod_{i=0}^{7}N_{\varepsilon_i}(q_i)=-I \tag*{(138)} i = 0 ∏ 7 N ε i ( q i ) = − I (138)
for both native words. Here the sum of the increments is − 2 -2 − 2 for R02 and 4 4 4 for R03, and the actual + 1 +1 + 1 seam fixes the displayed sign. These are full multipliers in a full matrix equation. Replacing them by their square roots gives the state half-return, not the frame equation (138) .
For R02 take q 0 , … , q 4 q_0,\ldots,q_4 q 0 , … , q 4 as free variables and write
u i = q i 2 − 1 ( 0 ≤ i ≤ 4 ) , u_i=q_i^2-1\qquad(0\leq i\leq4), u i = q i 2 − 1 ( 0 ≤ i ≤ 4 ) ,
K 1 = 1 + u 1 − u 0 u 2 , K 2 = 1 + u 3 − u 2 u 4 , K 3 = 1 + u 2 − u 1 u 3 , (139) K_1=1+u_1-u_0u_2,\qquad K_2=1+u_3-u_2u_4,\qquad K_3=1+u_2-u_1u_3, \tag*{(139)} K 1 = 1 + u 1 − u 0 u 2 , K 2 = 1 + u 3 − u 2 u 4 , K 3 = 1 + u 2 − u 1 u 3 , (139)
α = u 0 ( 1 + u 2 ) + u 3 ( 1 + u 0 + u 1 ) , V = ∏ i = 0 4 ( 1 + u i ) . \alpha=u_0(1+u_2)+u_3(1+u_0+u_1),\qquad V=\prod_{i=0}^{4}(1+u_i). α = u 0 ( 1 + u 2 ) + u 3 ( 1 + u 0 + u 1 ) , V = i = 0 ∏ 4 ( 1 + u i ) .
Lemma 35.3. The strict positive R02 solutions of (138) are exactly u i > 0 u_i>0 u i > 0 , K 1 , K 2 , K 3 > 0 K_1,K_2,K_3>0 K 1 , K 2 , K 3 > 0 , with the last three multipliers recovered by the positive roots
q 5 2 = 1 + ( 1 + u 4 ) K 1 V , q 6 2 = 1 + ( 1 + u 0 ) K 2 α V , q 7 2 = 1 + K 3 α . (140) q_5^2=1+\frac{(1+u_4)K_1}{V},\qquad q_6^2=1+\frac{(1+u_0)K_2\alpha}{V},\qquad q_7^2=1+\frac{K_3}{\alpha}. \tag*{(140)} q 5 2 = 1 + V ( 1 + u 4 ) K 1 , q 6 2 = 1 + V ( 1 + u 0 ) K 2 α , q 7 2 = 1 + α K 3 . (140) In particular these coordinates are one-to-one and have rank five.
Proof. Put P = ∏ i = 0 4 q i P=\prod_{i=0}^{4}q_i P = ∏ i = 0 4 q i . Exact multiplication gives
N − ( q 0 ) ⋯ N − ( q 4 ) = P − 1 ( A B C D ) , N_-(q_0)\cdots N_-(q_4)=P^{-1}\begin{pmatrix}A&B\\C&D\end{pmatrix}, N − ( q 0 ) ⋯ N − ( q 4 ) = P − 1 ( A C B D ) ,
where
A = ( 1 + u 0 ) ( 1 + u 4 ) u 2 , B = ( 1 + u 0 ) K 2 , C = − ( 1 + u 4 ) K 1 , A=(1+u_0)(1+u_4)u_2,\qquad B=(1+u_0)K_2,\qquad C=-(1+u_4)K_1, A = ( 1 + u 0 ) ( 1 + u 4 ) u 2 , B = ( 1 + u 0 ) K 2 , C = − ( 1 + u 4 ) K 1 ,
D + C = α > 0 , A + B = α + K 3 , A D − B C = P 2 = V . D+C=\alpha>0,\qquad A+B=\alpha+K_3,\qquad AD-BC=P^2=V. D + C = α > 0 , A + B = α + K 3 , A D − B C = P 2 = V .
Inverting the three positive factors in (138) gives
q 5 2 = D D + C , q 6 2 = D ( A + B ) P 2 , q 7 2 = A + B D + C . q_5^2=\frac{D}{D+C},\qquad q_6^2=\frac{D(A+B)}{P^2},\qquad q_7^2=\frac{A+B}{D+C}. q 5 2 = D + C D , q 6 2 = P 2 D ( A + B ) , q 7 2 = D + C A + B .
The identity D ( A + B ) = P 2 + B ( D + C ) D(A+B)=P^2+B(D+C) D ( A + B ) = P 2 + B ( D + C ) shows that these three quantities exceed one precisely when K 1 , K 2 , K 3 > 0 K_1,K_2,K_3>0 K 1 , K 2 , K 3 > 0 . This proves (140) . Conversely, substituting its positive roots gives the first three entries of
N + ( q 5 ) N + ( q 6 ) N + ( q 7 ) = − P ( A B C D ) − 1 ; N_+(q_5)N_+(q_6)N_+(q_7)=-P\begin{pmatrix}A&B\\C&D\end{pmatrix}^{-1}; N + ( q 5 ) N + ( q 6 ) N + ( q 7 ) = − P ( A C B D ) − 1 ;
determinant one fixes the last entry. Thus the full matrix closure holds, with no additional sign branch.
The free multipliers are retained among the output coordinates, which proves the rank assertion. □
For R03 the free variables are instead q 3 , … , q 7 q_3,\ldots,q_7 q 3 , … , q 7 . To describe their strict positive domain, set
( A , B , C , D , E ) = ( q 3 2 − 1 , q 4 2 − 1 , q 5 2 − 1 , q 6 2 − 1 , q 7 2 − 1 ) , (A,B,C,D,E)=(q_3^2-1,q_4^2-1,q_5^2-1,q_6^2-1,q_7^2-1), ( A , B , C , D , E ) = ( q 3 2 − 1 , q 4 2 − 1 , q 5 2 − 1 , q 6 2 − 1 , q 7 2 − 1 ) ,
G = 1 − B C , H = 1 − C D , K = 1 + B + D G , G=1-BC,\qquad H=1-CD,\qquad K=1+B+DG, G = 1 − B C , H = 1 − C D , K = 1 + B + D G ,
R 1 = 1 + C + A G , L = A ( 1 + B ) + D R 1 , J = B ( 1 + C ) + E K , R_1=1+C+AG,\qquad L=A(1+B)+DR_1,\qquad J=B(1+C)+EK, R 1 = 1 + C + A G , L = A ( 1 + B ) + D R 1 , J = B ( 1 + C ) + E K ,
F = ( 1 + C ) ( 1 − A B ) − E L . (141) F=(1+C)(1-AB)-EL. \tag*{(141)} F = ( 1 + C ) ( 1 − A B ) − E L . (141)
Lemma 35.4. The strict positive R03 solutions of (138) are exactly
A , B , C , D , E > 0 , A B < 1 , B C < 1 , C D < 1 , F > 0 , (142) A,B,C,D,E>0,\qquad AB<1,\qquad BC<1,\qquad CD<1,\qquad F>0, \tag*{(142)} A , B , C , D , E > 0 , A B < 1 , B C < 1 , C D < 1 , F > 0 , (142) with the first three multipliers given by the positive roots
q 0 2 = 1 + ( 1 + D ) G J , q 1 2 = 1 + K F L J , q 2 2 = 1 + ( 1 + B ) H L . (143) q_0^2=1+\frac{(1+D)G}{J},\qquad q_1^2=1+\frac{KF}{LJ},\qquad q_2^2=1+\frac{(1+B)H}{L}. \tag*{(143)} q 0 2 = 1 + J ( 1 + D ) G , q 1 2 = 1 + L J K F , q 2 2 = 1 + L ( 1 + B ) H . (143) All denominators are strictly positive, and the chart is one-to-one of rank five.
Proof. Factor the native frame product after its first 021 triple, and let M M M be the required inverse tail matrix. If a = q 3 , … , e = q 7 a=q_3,\ldots,e=q_7 a = q 3 , … , e = q 7 , write r − = M 22 − M 21 r_-=M_{22}-M_{21} r − = M 22 − M 21 , r + = M 22 + M 21 r_+=M_{22}+M_{21} r + = M 22 + M 21 and Σ = ∑ M i j \Sigma=\sum M_{ij} Σ = ∑ M ij , using indices 1, 2 in this calculation. Direct multiplication of the five tail factors gives
r − = e L a b c d , r + = a e K b c d , Σ = 2 a J b c d e . r_-=\frac{eL}{abcd},\qquad r_+=\frac{aeK}{bcd},\qquad\Sigma=\frac{2aJ}{bcde}. r − = ab c d e L , r + = b c d a eK , Σ = b c d e 2 a J .
The positive 021 inverse is
q 0 2 = 2 r + Σ = ( 1 + E ) K J , q 1 2 = 2 r − Σ = ( 1 + B ) ( 1 + C ) ( 1 + D ) L J , q 2 2 = r + r − = ( 1 + A ) K L . q_0^2=\frac{2r_+}{\Sigma}=\frac{(1+E)K}{J},\qquad q_1^2=\frac{2}{r_-\Sigma}=\frac{(1+B)(1+C)(1+D)}{LJ},\qquad q_2^2=\frac{r_+}{r_-}=\frac{(1+A)K}{L}. q 0 2 = Σ 2 r + = J ( 1 + E ) K , q 1 2 = r − Σ 2 = L J ( 1 + B ) ( 1 + C ) ( 1 + D ) , q 2 2 = r − r + = L ( 1 + A ) K .
Subtracting one uses the exact numerator identities
( 1 + E ) K − J = ( 1 + D ) ( 1 − B C ) , (1+E)K-J=(1+D)(1-BC), ( 1 + E ) K − J = ( 1 + D ) ( 1 − B C ) ,
( 1 + A ) K − L = ( 1 + B ) ( 1 − C D ) , (1+A)K-L=(1+B)(1-CD), ( 1 + A ) K − L = ( 1 + B ) ( 1 − C D ) ,
( 1 + B ) ( 1 + C ) ( 1 + D ) − L J = K F . (1+B)(1+C)(1+D)-LJ=KF. ( 1 + B ) ( 1 + C ) ( 1 + D ) − L J = K F .
For an actual positive inverse, L , K , J > 0 L,K,J>0 L , K , J > 0 ; the first and third recovered multipliers force B C , C D < 1 BC,CD<1 B C , C D < 1 , and the middle one forces F > 0 F>0 F > 0 . Since E , L > 0 E,L>0 E , L > 0 , this also implies A B < 1 AB<1 A B < 1 . Conversely, (142) makes K , R 1 , L , J K,R_1,L,J K , R 1 , L , J positive and all three roots in (143) greater than one. They recover r − , r + , Σ r_-,r_+,\Sigma r − , r + , Σ exactly. These three quantities determine the lower row and upper-row sum of a determinant-one matrix; r + > 0 r_+>0 r + > 0 and the determinant then determine the remaining entry. Thus the recovered triple is the entire matrix M M M . Retention of the five free tail variables proves injectivity and rank five. □
The finite cover uses reciprocal variables r i = q i − 1 r_i=q_i^{-1} r i = q i − 1 . For R03 put a = r 3 2 a=r_3^2 a = r 3 2 , b = r 4 2 b=r_4^2 b = r 4 2 , c = r 5 2 c=r_5^2 c = r 5 2 , d = r 6 2 d=r_6^2 d = r 6 2 , e = r 7 2 e=r_7^2 e = r 7 2 and define
T = b + c + d − b d − 1 , L r = ( 1 − a ) T + a b ( 1 − d ) , J r = ( 1 − e ) T + d e ( 1 − b ) , V r = a b c d e . (144) T=b+c+d-bd-1,\qquad L_r=(1-a)T+ab(1-d),\qquad J_r=(1-e)T+de(1-b),\qquad V_r=abcde. \tag*{(144)} T = b + c + d − b d − 1 , L r = ( 1 − a ) T + ab ( 1 − d ) , J r = ( 1 − e ) T + d e ( 1 − b ) , V r = ab c d e . (144)
Then the same positive chart is
( r 0 2 , r 1 2 , r 2 2 ) = ( J r T , L r J r V r , L r T ) . (145) (r_0^2,r_1^2,r_2^2)=\left(\frac{J_r}{T},\frac{L_rJ_r}{V_r},\frac{L_r}{T}\right). \tag*{(145)} ( r 0 2 , r 1 2 , r 2 2 ) = ( T J r , V r L r J r , T L r ) . (145)
These identities follow by multiplying (143) by the positive powers of a , b , c , d , e a,b,c,d,e a , b , c , d , e ; in particular they recover the same full matrix, not merely its trace. The necessary positive guards and multiplier caps provide conservative intersections for candidate values. Whole-box derivatives are taken from the unmodified rational and positive-root formulas on every regular box.
A common physical state and cost circuit
The half-return associated with (138) is noncentral at every strict positive closed tuple of either type. The short verification below is useful because a scalar return would not determine its moving initial state.
Lemma 35.5. For R 02 R02 R 02 and R 03 R03 R 03 , B = ∏ i N ε i ( q i ) ≠ ± I \mathcal{B} = \prod_i N_{\varepsilon_i}(\sqrt{q_i}) \ne\pm I B = ∏ i N ε i ( q i ) = ± I whenever q i > 1 q_i > 1 q i > 1 and (138) holds. At an actual physical candidate its canonical state return is therefore strictly elliptic.
Proof. For R02 split the full product into four negative factors with multipliers ( a , b , c , d ) (a,b,c,d) ( a , b , c , d ) and the remaining mixed block. The mixed block has the form
( − p ′ p ′ + s ′ − p ′ − r ′ D ′ ) , p ′ = q 4 q 6 q 5 q 7 , r ′ = q 6 ( q 5 2 − 1 ) q 4 q 5 q 7 > 0 , s ′ = q 4 q 6 ( q 7 2 − 1 ) q 5 q 7 > 0. \begin{pmatrix}
-p' & p' + s' \\
-p' - r' & D'
\end{pmatrix},
\qquad p' = \frac{q_4q_6}{q_5q_7}, \qquad r' = \frac{q_6(q_5^2-1)}{q_4q_5q_7} > 0, \qquad s' = \frac{q_4q_6(q_7^2-1)}{q_5q_7} > 0. ( − p ′ − p ′ − r ′ p ′ + s ′ D ′ ) , p ′ = q 5 q 7 q 4 q 6 , r ′ = q 4 q 5 q 7 q 6 ( q 5 2 − 1 ) > 0 , s ′ = q 5 q 7 q 4 q 6 ( q 7 2 − 1 ) > 0. Full closure gives p ′ = H / ( a b c d ) p' = H/(abcd) p ′ = H / ( ab c d ) , r ′ = R 0 / ( a b c d ) r' = R_0/(abcd) r ′ = R 0 / ( ab c d ) , s ′ = U 0 / ( a b c d ) s' = U_0/(abcd) s ′ = U 0 / ( ab c d ) in the notation of Lemma 29.1. Hence 0 < E < H 0 < E < \sqrt{H} 0 < E < H . For the corresponding half blocks L h , C h L_h,C_h L h , C h , the lower-right entry of L h L_h L h is E / a b c d E/\sqrt{abcd} E / ab c d , whereas − ( C h ) 00 = p ′ = H / ( a b c d ) -(C_h)_{00} = \sqrt{p'} = \sqrt{H}/(abcd) − ( C h ) 00 = p ′ = H / ( ab c d ) is larger. If L h C h = I L_hC_h = I L h C h = I , their inverse relation would require ( C h ) 00 = ( L h ) 11 > 0 (C_h)_{00} = (L_h)_{11} > 0 ( C h ) 00 = ( L h ) 11 > 0 , contrary to its sign. If L h C h = − I L_hC_h = -I L h C h = − I , it would require − ( C h ) 00 = ( L h ) 11 -(C_h)_{00} = (L_h)_{11} − ( C h ) 00 = ( L h ) 11 , contrary to the strict gap.
For R03 write its six-positive full tail as P P P . The two-negative prefix gives
P 10 = q 1 / q 0 > 0 , P 11 = 0. P_{10} = q_1/q_0 > 0, \qquad P_{11} = 0. P 10 = q 1 / q 0 > 0 , P 11 = 0. Put U k = q k + 2 2 − 1 U_k = q_{k+2}^2 - 1 U k = q k + 2 2 − 1 for 0 ≤ k ≤ 5 0 \le k \le5 0 ≤ k ≤ 5 and
G = ( U 0 U 1 − 1 ) ( U 2 U 3 − 1 ) − ( 1 + U 0 ) ( 1 + U 3 ) , \mathcal{G} = (U_0U_1 - 1)(U_2U_3 - 1) - (1 + U_0)(1 + U_3), G = ( U 0 U 1 − 1 ) ( U 2 U 3 − 1 ) − ( 1 + U 0 ) ( 1 + U 3 ) , E ( U ) = ( 1 − U 0 U 1 ) ( 1 − U 3 U 4 ) ( 1 + U 2 ) − U 0 U 4 ( 1 + U 1 ) ( 1 + U 3 ) . \mathcal{E}(U) = (1 - U_0U_1)(1 - U_3U_4)(1 + U_2) - U_0U_4(1 + U_1)(1 + U_3). E ( U ) = ( 1 − U 0 U 1 ) ( 1 − U 3 U 4 ) ( 1 + U 2 ) − U 0 U 4 ( 1 + U 1 ) ( 1 + U 3 ) . Multiplication of
( 0 1 + U k − 1 1 + U k ) \begin{pmatrix}
0 & 1+U_k \\
-1 & 1+U_k
\end{pmatrix} ( 0 − 1 1 + U k 1 + U k ) gives its lower-left entry − ( 1 + U 4 ) G -(1 + U_4)\mathcal{G} − ( 1 + U 4 ) G and lower-right entry ( 1 + U 5 ) E ( U ) (1 + U_5)\mathcal{E}(U) ( 1 + U 5 ) E ( U ) . Also
E ( U ) = U 4 G + ( 1 + U 2 ) ( 1 − U 0 U 1 ) . \mathcal{E}(U) = U_4\mathcal{G} + (1 + U_2)(1 - U_0U_1). E ( U ) = U 4 G + ( 1 + U 2 ) ( 1 − U 0 U 1 ) . Thus G < 0 \mathcal{G} < 0 G < 0 , E ( U ) = 0 \mathcal{E}(U) = 0 E ( U ) = 0 , U 0 U 1 < 1 U_0U_1 < 1 U 0 U 1 < 1 and U 3 U 4 < 1 U_3U_4 < 1 U 3 U 4 < 1 . Set v k = q k + 2 − 1 v_k = q_{k+2} - 1 v k = q k + 2 − 1 , so U k = v k ( v k + 2 ) U_k = v_k(v_k + 2) U k = v k ( v k + 2 ) . For v , w > 0 v,w > 0 v , w > 0 and V = v ( v + 2 ) V = v(v+2) V = v ( v + 2 ) , W = w ( w + 2 ) W = w(w+2) W = w ( w + 2 ) with V W < 1 VW < 1 V W < 1 ,
( 1 − v w ) 2 − ( 1 − V W ) = 2 v w ( 1 + v ) ( 1 + w ) > 0. (1-vw)^2 - (1-VW) = 2vw(1+v)(1+w) > 0. ( 1 − v w ) 2 − ( 1 − V W ) = 2 v w ( 1 + v ) ( 1 + w ) > 0. Take the positive square root of E ( U ) = 0 \mathcal{E}(U) = 0 E ( U ) = 0 , apply this strict inequality to the pairs ( 0 , 1 ) (0,1) ( 0 , 1 ) and ( 3 , 4 ) (3,4) ( 3 , 4 ) , and use U 0 U 4 > v 0 v 4 \sqrt{U_0U_4} > v_0v_4 U 0 U 4 > v 0 v 4 . It follows that E ( v ) > 0 \mathcal{E}(v) > 0 E ( v ) > 0 . The six-positive half-tail consequently has positive lower-right entry. If B = ± I \mathcal{B} = \pm I B = ± I , it would be plus or minus the inverse two-negative half-prefix, whose lower-right entry is zero. This is a contradiction.
A real determinant-one noncentral matrix fixing an upper-half-plane point is elliptic. The actual physical return has that fixed point, proving the final assertion; no ellipticity claim for all nonphysical tuples is needed.
Here is an explicit common circuit for the values used below. Define matrices affine in r = q − 1 r = q^{-1} r = q − 1 by
F ‾ − ( r ) = 1 2 ( 1 1 + 2 r − 1 2 r − 1 ) , F ‾ + ( r ) = 1 2 ( 1 1 − 4 r 1 1 ) . (146) \overline{F}_{-}(r) = \frac{1}{2}
\begin{pmatrix}
1 & 1+2r \\
-1 & 2r-1
\end{pmatrix},
\qquad
\overline{F}_{+}(r) = \frac{1}{2}
\begin{pmatrix}
1 & 1-4r \\
1 & 1
\end{pmatrix}. \tag*{(146)} F − ( r ) = 2 1 ( 1 − 1 1 + 2 r 2 r − 1 ) , F + ( r ) = 2 1 ( 1 1 1 − 4 r 1 ) . (146)
Each has determinant r r r . For cut i i i take the cyclic product in state-transport order, H i = F ‾ i − 1 ⋯ F ‾ i H_i=\overline{F}_{i-1}\cdots\overline{F}_i H i = F i − 1 ⋯ F i , where all eight factors occur once. Set
δ = 4 ∏ i = 0 7 r i − ( tr H 0 ) 2 , \delta=4\prod_{i=0}^{7}r_i-(\operatorname{tr}H_0)^2, δ = 4 i = 0 ∏ 7 r i − ( tr H 0 ) 2 ,
n i 0 = ( H i ) 00 − ( H i ) 11 + 2 ( H i ) 10 , n_{i0}=(H_i)_{00}-(H_i)_{11}+2(H_i)_{10}, n i 0 = ( H i ) 00 − ( H i ) 11 + 2 ( H i ) 10 ,
n i 1 = − ( H i ) 01 − ( H i ) 10 , n i 2 = − ( H i ) 00 + ( H i ) 11 + 2 ( H i ) 10 . (147) n_{i1}=-(H_i)_{01}-(H_i)_{10},\qquad n_{i2}=-(H_i)_{00}+(H_i)_{11}+2(H_i)_{10}. \tag*{(147)} n i 1 = − ( H i ) 01 − ( H i ) 10 , n i 2 = − ( H i ) 00 + ( H i ) 11 + 2 ( H i ) 10 . (147)
At a physical candidate δ > 0 \delta>0 δ > 0 . Formula (30.3) gives its canonical reciprocals as
d k ( i ) = κ n i k δ > 0 , κ = sgn ( H i ) 10 ∈ { − 1 , 1 } . (148) d_k(i)=\frac{\kappa n_{ik}}{\sqrt{\delta}}>0,\qquad\kappa=\operatorname{sgn}(H_i)_{10}\in\{-1,1\}. \tag*{(148)} d k ( i ) = δ κ n ik > 0 , κ = sgn ( H i ) 10 ∈ { − 1 , 1 } . (148)
The sign is common to all cuts: cyclic conjugation transports the positive state matrix and preserves the scalar coefficient of its traceless part. The Lorentz identity reads
n i 0 n i 1 + n i 0 n i 2 + n i 1 n i 2 = δ . (149) n_{i0}n_{i1}+n_{i0}n_{i2}+n_{i1}n_{i2}=\delta. \tag*{(149)} n i 0 n i 1 + n i 0 n i 2 + n i 1 n i 2 = δ . (149)
Compactness gives 1 / 15 ≤ d k ( i ) ≤ 400 / 99 1/15\le d_k(i)\le400/99 1/15 ≤ d k ( i ) ≤ 400/99 . These are necessary candidate bounds, rather than restrictions on every point of the closure chart.
Lemma 35.6. For the physical R02 and R03 branches in (35.11), κ = − 1 \kappa=-1 κ = − 1 and all twenty-four n i k n_{ik} n ik are strictly negative.
Proof. For R02 write its half-product as B = L h C h \mathcal{B}=L_hC_h B = L h C h , as in Lemma 35.5, with D = ( L h ) 11 > 0 D=(L_h)_{11}>0 D = ( L h ) 11 > 0 and γ = − ( C h ) 00 − D > 0 \gamma=-(C_h)_{00}-D>0 γ = − ( C h ) 00 − D > 0 . The full closure guard U 0 > 0 U_0>0 U 0 > 0 gives
( L h ) 01 − D = c − ( b − 1 ) ( d − 1 ) a b c d > 0. (L_h)_{01}-D=\frac{c-(b-1)(d-1)}{\sqrt{abcd}}>0. ( L h ) 01 − D = ab c d c − ( b − 1 ) ( d − 1 ) > 0.
The bridge with (35.10) is H 0 = ( ∏ q i − 1 / 2 ) S B t S − 1 H_0=\left(\prod q_i^{-1/2}\right)S\mathcal{B}^tS^{-1} H 0 = ( ∏ q i − 1/2 ) S B t S − 1 , where S = ( 2 1 0 1 ) S=\begin{pmatrix}2&1\\0&1\end{pmatrix} S = ( 2 0 1 1 ) . If w H w_H w H and w B w_{\mathcal{B}} w B are their upper fixed points, w H = 1 − 2 / w B w_H=1-2/w_{\mathcal{B}} w H = 1 − 2/ w B . The physical inequalities − 1 < ℜ w H < 1 -1<\Re w_H<1 − 1 < ℜ w H < 1 and ∣ w H ∣ > 1 |w_H|>1 ∣ w H ∣ > 1 give x : = ℜ w B ∈ ( 0 , 1 ) x:=\Re w_{\mathcal{B}}\in(0,1) x := ℜ w B ∈ ( 0 , 1 ) . Comparing the 00 entry of C h = L h − 1 B C_h=L_h^{-1}\mathcal{B} C h = L h − 1 B yields
γ = B 10 ( ( L h ) 01 − D x ) − D 2 ( tr B + 2 ) . \gamma=\mathcal{B}_{10}\bigl((L_h)_{01}-Dx\bigr)-\frac{D}{2}(\operatorname{tr}\mathcal{B}+2). γ = B 10 ( ( L h ) 01 − D x ) − 2 D ( tr B + 2 ) .
Strict ellipticity makes tr B + 2 > 0 \operatorname{tr}\mathcal{B}+2>0 tr B + 2 > 0 . Thus B 10 > 0 \mathcal{B}_{10}>0 B 10 > 0 ; ellipticity then gives B 01 < 0 \mathcal{B}_{01}<0 B 01 < 0 , so ( H 0 ) 10 < 0 (H_0)_{10}<0 ( H 0 ) 10 < 0 .
For R03 put x = r 4 x=r_4 x = r 4 , y = r 5 y=r_5 y = r 5 , z = r 6 z=r_6 z = r 6 ,
T = x 2 + y 2 + z 2 − x 2 z 2 − 1 , c = y 2 + z 2 − 1 T , B = y + z − 1 , h = y − ( 1 − x ) ( 1 − z ) , T=x^2+y^2+z^2-x^2z^2-1,\qquad c=\frac{y^2+z^2-1}{T},\qquad B=y+z-1,\qquad h=y-(1-x)(1-z), T = x 2 + y 2 + z 2 − x 2 z 2 − 1 , c = T y 2 + z 2 − 1 , B = y + z − 1 , h = y − ( 1 − x ) ( 1 − z ) ,
and ρ = r 3 / ( 1 + r 2 ) \rho=r_3/(1+r_2) ρ = r 3 / ( 1 + r 2 ) , ω = x y z / T \omega=xyz/T ω = x y z / T . The positive chart gives 0 < c < 1 0<c<1 0 < c < 1 , T , B , h > 0 T,B,h>0 T , B , h > 0 , r 2 2 + c r 3 2 = 1 r_2^2+cr_3^2=1 r 2 2 + c r 3 2 = 1 , and the exact block identity
( H 0 ) 10 = ( r 0 r 2 ) c h ρ − B 2 ω r 7 . (H_0)_{10}=(r_0r_2)\frac{c h\rho-B}{2\omega r_7}. ( H 0 ) 10 = ( r 0 r 2 ) 2 ω r 7 c h ρ − B .
Indeed it follows by multiplying the last four positive factors in (35.10), whose lower-left entry is − B / 2 -B/2 − B /2 and lower-row sum is − h -h − h , by the remaining four factors. From r 3 = 2 ρ / ( 1 + c ρ 2 ) < 1 r_3=2\rho/(1+c\rho^2)<1 r 3 = 2 ρ / ( 1 + c ρ 2 ) < 1 and c ρ 2 < 1 c\rho^2<1 c ρ 2 < 1 we obtain c ρ < 1 − 1 − c c\rho<1-\sqrt{1-c} c ρ < 1 − 1 − c . Moreover 1 − c = x 2 ( 1 − z 2 ) / T 1-c=x^2(1-z^2)/T 1 − c = x 2 ( 1 − z 2 ) / T , and
( 1 + z ) h 2 − ( 1 − z ) T = 2 { z y 2 − ( 1 − z 2 ) ( 1 − x ) B } > 0. (1+z)h^2-(1-z)T=2\{zy^2-(1-z^2)(1-x)B\}>0. ( 1 + z ) h 2 − ( 1 − z ) T = 2 { z y 2 − ( 1 − z 2 ) ( 1 − x ) B } > 0.
The last sign follows from y 2 > 1 − z 2 y^2>1-z^2 y 2 > 1 − z 2 and 0 < B < z 0<B<z 0 < B < z . Hence 1 − c h > x ( 1 − z ) = h − B \sqrt{1-c}\,h>x(1-z)=h-B 1 − c h > x ( 1 − z ) = h − B , and c ρ h < B c\rho h<B c ρ h < B . Thus ( H 0 ) 10 < 0 (H_0)_{10}<0 ( H 0 ) 10 < 0 also for R03.
At each physical cut the three numerator formulas have the sign of ( H i ) 10 (H_i)_{10} ( H i ) 10 . The exact transport identity n i + 1 , k = n i 0 n_{i+1,k}=n_{i0} n i + 1 , k = n i 0 , with k = 1 k=1 k = 1 for a negative increment and k = 2 k=2 k = 2 for a positive increment, propagates this negative sign through all eight cuts. □
Integrating the physical cost along a mode-zero arc with entrance reciprocals ( a , b , c ) (a,b,c) ( a , b , c ) gives
c i = a 2 ( q i − q i − 1 − log q i ) + a b ( q i − 1 ) + a c ( 1 − q i − 1 ) . c_i = a^2(q_i-q_i^{-1}-\log q_i)+ab(q_i-1)+ac(1-q_i^{-1}). c i = a 2 ( q i − q i − 1 − log q i ) + ab ( q i − 1 ) + a c ( 1 − q i − 1 ) .
Indeed d t = a d log p dt=a\,d\log p d t = a d log p , the two passive reciprocals are p b + ( p − 1 ) a pb+(p-1)a p b + ( p − 1 ) a and ( c + a ) / p − a (c+a)/p-a ( c + a ) / p − a , and the integrand is their sum with a a a . Thus the common-scale cost numerators are
N i = n i 0 2 ( q i − q i − 1 − log q i ) + n i 0 n i 1 ( q i − 1 ) + n i 0 n i 2 ( 1 − q i − 1 ) , c i = N i / δ . (150) N_i=n_{i0}^2(q_i-q_i^{-1}-\log q_i)+n_{i0}n_{i1}(q_i-1)+n_{i0}n_{i2}(1-q_i^{-1}),\qquad c_i=N_i/\delta. \tag*{(150)} N i = n i 0 2 ( q i − q i − 1 − log q i ) + n i 0 n i 1 ( q i − 1 ) + n i 0 n i 2 ( 1 − q i − 1 ) , c i = N i / δ . (150)
The passive components in this identity are the entrance components in their stated order. The two signs in the next switch change their subsequent positions, not the physical integral. A cancellation-preserving polynomial form is
o i = { n i + 1 , 2 − n i + 1 , 0 , ε i = − 1 , n i + 1 , 0 − n i + 1 , 1 , ε i = 1 , o_i=\begin{cases}
n_{i+1,2}-n_{i+1,0},&\varepsilon_i=-1,\\
n_{i+1,0}-n_{i+1,1},&\varepsilon_i=1,
\end{cases} o i = { n i + 1 , 2 − n i + 1 , 0 , n i + 1 , 0 − n i + 1 , 1 , ε i = − 1 , ε i = 1 ,
P i = n i 0 ( o i − n i 1 + n i 2 ) , A i = n i 0 2 , N i = P i + A i log r i . (151) P_i=n_{i0}(o_i-n_{i1}+n_{i2}),\qquad A_i=n_{i0}^2,\qquad N_i=P_i+A_i\log r_i. \tag*{(151)} P i = n i 0 ( o i − n i 1 + n i 2 ) , A i = n i 0 2 , N i = P i + A i log r i . (151)
The transport formulas prove the equality of (150) and (151) . Every n i k n_{ik} n ik is multiaffine in the eight formal r i r_i r i , and P i P_i P i , A i A_i A i , δ \delta δ have degree at most two in each variable. This eight-variable extension is used for value bounds on an enclosing box; the physical closure chart still has five free variables.
The candidate inequalities in a common positive scale are
B 0 = C ∗ δ − ∑ i N i ≥ 0 , F j = ∑ i a j i N i − b j δ ≥ 0. (152) \mathcal{B}_0=C_*\delta-\sum_i N_i\geq0,\qquad\mathcal{F}_j=\sum_i a_{ji}N_i-b_j\delta\geq0. \tag*{(152)} B 0 = C ∗ δ − i ∑ N i ≥ 0 , F j = i ∑ a j i N i − b j δ ≥ 0. (152)
The second family consists of nine rows for R02 and thirteen for R03. We now specify their complete physical supports.
R02: complete sources and a compact root
Throughout the source tables, [ s , t ] [s,t] [ s , t ] denotes the increasing list of unwrapped arc indices when s ≤ t s\leq t s ≤ t and the decreasing list when s > t s>t s > t . Labels at unwrapped indices use j i + 8 = j i + 1 j_{i+8}=j_i+1 j i + 8 = j i + 1 ; physical charges are reduced modulo eight. The column ℓ ( j ) \ell(j) ℓ ( j ) gives the label map on the entire ordered source. A decreasing list is transported by whole reflection with time reversal, including its state, frame, covectors and charges.
Rows 0–1 are Theorem B.47; rows 2–3 are Theorem B.83; row 4 is Theorem B.122. Rows 5–6 retain the actual nonpositive entrance required by Corollary B.76. Their five- or six-arc supports are proper. Rows 7 and 8 have eight-arc supports and use the whole-period permissions of Theorems B.128 and E.2, respectively. For row 7 the two tails and common switching derivative belong to one selected complete lift; for row 8 both caps and rooted contacts lie inside its genuine switch-cut interval. Neither application continues a state covector periodically or claims strict properness. All rows retain the normality and joint normalization stipulated by their individual theorems. They are scalar facts about the same physical candidate; their addition does not glue covectors selected on different intervals.
Whole reflection and reanchoring give the exact R02 involution
j r ′ = 1 − j 5 − r = j r ( m o d 3 ) , q r ′ = q 5 − r , c r ′ = c 5 − r . (153) j'_r = 1-j_{5-r}=j_r \pmod{3}, \qquad q'_r=q_{5-r}, \qquad c'_r=c_{5-r}. \tag*{(153)} j r ′ = 1 − j 5 − r = j r ( mod 3 ) , q r ′ = q 5 − r , c r ′ = c 5 − r . (153)
The twisted seam is unchanged, as j r + 8 ′ = j r ′ + 1 j'_{r+8}=j'_r+1 j r + 8 ′ = j r ′ + 1 . The lawful symmetry of § 3.1 transports the full endpoint and real winding. Thus we choose c 2 ≤ c 3 c_2\leq c_3 c 2 ≤ c 3 . Row 7 and the budget give
c 2 + c 3 < 109483 100000 , c 2 < 109483 200000 < 13 20 , c 3 > 13 40 , (154) c_2+c_3<\frac{109483}{100000}, \qquad c_2<\frac{109483}{200000}<\frac{13}{20}, \qquad c_3>\frac{13}{40}, \tag*{(154)} c 2 + c 3 < 100000 109483 , c 2 < 200000 109483 < 20 13 , c 3 > 40 13 , (154)
the last inequality following from row 8 and the ordering.
The two actual proper sources [ − 1 , 4 ] [-1,4] [ − 1 , 4 ] and [ 6 , 1 ] [6,1] [ 6 , 1 ] , with maps j j j and 1 − j 1-j 1 − j , are both 202102. Their respective centres are physical arcs 2 and 3. The unconditional crossing estimate of Lemma B.105 and the contact floor (B.174) give
d 0 ( 2 ) > d 2 ( 2 ) , d 2 ( 3 ) > d 0 ( 3 ) , d 2 ( 3 ) > d 1 ( 3 ) , d 1 ( 4 ) > d 2 ( 4 ) , (155) d_0(2)>d_2(2), \qquad d_2(3)>d_0(3), \qquad d_2(3)>d_1(3), \qquad d_1(4)>d_2(4), \tag*{(155)} d 0 ( 2 ) > d 2 ( 2 ) , d 2 ( 3 ) > d 0 ( 3 ) , d 2 ( 3 ) > d 1 ( 3 ) , d 1 ( 4 ) > d 2 ( 4 ) , (155)
and
c i > W ( q i ) ( i = 2 , 3 ) , W ( x ) = x log x x − 1 − 1. (156) c_i>W(q_i)\quad(i=2,3), \qquad W(x)=\frac{x\log x}{x-1}-1. \tag*{(156)} c i > W ( q i ) ( i = 2 , 3 ) , W ( x ) = x − 1 x log x − 1. (156)
These crossing statements precede the cheap-centre assumption in the local proof. Its additional cheap-centre conclusion is justified for arc 2 by (35.18): the actual outgoing switching function starts nonnegative with positive derivative, has derivative μ − u 2 \mu-u^2 μ − u 2 , and cannot return to zero before u > μ > 0 u>\sqrt{\mu}>0 u > μ > 0 . Lemma B.107 therefore gives d 0 ( 4 ) > d 2 ( 4 ) d_0(4)>d_2(4) d 0 ( 4 ) > d 2 ( 4 ) . The analogous extra conclusion for arc 3 is used only when c 3 ≤ 13 / 20 c_3\leq13/20 c 3 ≤ 13/20 . The proper source [ 5 , 8 ] = 1201 [5,8]=1201 [ 5 , 8 ] = 1201 gives d 1 ( 7 ) < d 0 ( 7 ) , d 2 ( 7 ) d_1(7)<d_0(7),d_2(7) d 1 ( 7 ) < d 0 ( 7 ) , d 2 ( 7 ) by Lemma B.24 .
Since W ′ ( x ) = ( x − 1 − log x ) / ( x − 1 ) 2 > 0 W'(x)=(x-1-\log x)/(x-1)^2>0 W ′ ( x ) = ( x − 1 − log x ) / ( x − 1 ) 2 > 0 , the positive logarithm series at 13 / 5 13/5 13/5 and 6 gives
W ( 13 / 5 ) > 162628 295245 > 109483 200000 , W ( 6 ) > 18405 16807 > 109483 100000 . W(13/5)>\frac{162628}{295245}>\frac{109483}{200000}, \qquad W(6)>\frac{18405}{16807}>\frac{109483}{100000}. W ( 13/5 ) > 295245 162628 > 200000 109483 , W ( 6 ) > 16807 18405 > 100000 109483 .
Equations (35.18)–(35.20) imply q 2 < 13 / 5 q_2<13/5 q 2 < 13/5 and q 3 < 6 q_3<6 q 3 < 6 . For the opposite bound on q 3 q_3 q 3 , let a i a_i a i be the active physical reciprocal on arc i i i . At cut 3, (35.19) and the Lorentz identity give a 3 2 + 2 a 2 a 3 < 1 a_3^2+2a_2a_3<1 a 3 2 + 2 a 2 a 3 < 1 . Since a 2 ≥ 1 / 15 a_2\geq1/15 a 2 ≥ 1/15 and ( 15 / 16 ) 2 + 2 ( 15 / 16 ) / 15 = 257 / 256 > 1 (15/16)^2+2(15/16)/15=257/256>1 ( 15/16 ) 2 + 2 ( 15/16 ) /15 = 257/256 > 1 , we have a 3 < 15 / 16 a_3<15/16 a 3 < 15/16 . Along every positive physical arc,
d c d log p = a ( a + b ( p ) + c ( p ) ) = 1 + a 2 − b ( p ) c ( p ) < 1 + a 2 . (157) \frac{dc}{d\log p}=a(a+b(p)+c(p))=1+a^2-b(p)c(p)<1+a^2. \tag*{(157)} d log p d c = a ( a + b ( p ) + c ( p )) = 1 + a 2 − b ( p ) c ( p ) < 1 + a 2 . (157)
Consequently c 3 < ( 481 / 256 ) log q 3 c_3<(481/256)\log q_3 c 3 < ( 481/256 ) log q 3 . If q 3 ≤ 19 / 16 q_3\leq19/16 q 3 ≤ 19/16 , then log x < ( x − x − 1 ) / 2 \log x<(x-x^{-1})/2 log x < ( x − x − 1 ) /2 gives
c 3 < 481 512 ( 19 16 − 16 19 ) = 50505 155648 < 13 40 , c_3<\frac{481}{512}\left(\frac{19}{16}-\frac{16}{19}\right)=\frac{50505}{155648}<\frac{13}{40}, c 3 < 512 481 ( 16 19 − 19 16 ) = 155648 50505 < 40 13 ,
contrary to (35.18). Thus q 3 > 19 / 16 q_3>19/16 q 3 > 19/16 and α > u 3 > 105 / 256 \alpha>u_3>105/256 α > u 3 > 105/256 .
The same estimate gives q 2 q 3 > 7 / 5 q_2q_3>7/5 q 2 q 3 > 7/5 without an unresolved cheap-centre branch. If c 3 ≤ 13 / 20 c_3\leq13/20 c 3 ≤ 13/20 , the two cheap-centre conclusions give a 2 , a 3 < 15 / 16 a_2,a_3<15/16 a 2 , a 3 < 15/16 , and q 2 q 3 ≤ 7 / 5 q_2q_3\leq7/5 q 2 q 3 ≤ 7/5 would imply
c 2 + c 3 < 481 256 log ( 7 / 5 ) < 1443 2240 < 13 20 , c_2+c_3<\frac{481}{256}\log(7/5)<\frac{1443}{2240}<\frac{13}{20}, c 2 + c 3 < 256 481 log ( 7/5 ) < 2240 1443 < 20 13 ,
contrary to row 8. If c 3 > 13 / 20 c_3>13/20 c 3 > 13/20 , the same bound on c 3 c_3 c 3 alone forces q 3 > 7 / 5 q_3>7/5 q 3 > 7/5 , which suffices since q 2 > 1 q_2>1 q 2 > 1 .
We have proved the root inclusion
( r 0 , r 1 , r 2 , r 3 , r 4 ) ∈ [ 33 / 520 , 1 ] 2 × [ 5 / 13 , 1 ] × [ 1 / 6 , 16 / 19 ] × [ 33 / 520 , 1 ] . (158) (r_0,r_1,r_2,r_3,r_4)\in[33/520,1]^2\times[5/13,1]\times[1/6,16/19]\times[33/520,1]. \tag*{(158)} ( r 0 , r 1 , r 2 , r 3 , r 4 ) ∈ [ 33/520 , 1 ] 2 × [ 5/13 , 1 ] × [ 1/6 , 16/19 ] × [ 33/520 , 1 ] . (158)
The actual root uses outward binary endpoint enclosures of this exact box. Weakening its faces retains every strict candidate. The positivity guards in Lemma 35.3, the cap q i ≤ 520 / 33 q_i \le520/33 q i ≤ 520/33 , (35.19), the conditional cheap-centre comparison and the nine fees give only necessary intersections in this root. In particular the R03 density owner used next is not imported into R02.
R03: proper sources and the positive-tail root.
Table 7 uses the same support and label conventions as Table 6 . Every support has length five, six or seven, strictly less than eight.
Table 6. The nine R02 inequalities ∑ i c h a r g e d c i ≥ b j \sum_{i\ {\rm charged}}c_i\geq b_j ∑ i charged c i ≥ b j . Only rows 0 and 1 are intrinsically weak; the other strict bounds are safely weakened in (35.16).
Table 7. The thirteen proper-source R03 inequalities. Their original strict lower bounds are weakened to closed necessary inequalities in (35.16).
Rows 0–1 use Theorem B.83. For rows 2–3 the complete seven-source preserves the uncharged support arc as well as both caps; its 41 / 100 41/100 41/100 estimate is Theorem B.31 , also weaker than the own-four bound in Theorem B.122. Rows 4–5 use Theorem C.10. Rows 6–7 use Proposition 10.1 : the complete forward prefix supplies the negative entrance and entrance circle by Lemma B.77, and the actual final return supplies its terminal circle. This 2 / 5 2/5 2/5 payment is not a theorem about an isolated 0212 string. Rows 8–10 use the complete capped estimate of Proposition B.36 ; rows 11–12 use Corollary B.38 . All retain their actual returns, contacts, boundary ties and one selected complete lift wherever means or switching rows are coupled. Independently established scalar consequences may be used together without identifying the covectors of different sources.
The complete proper source [6, 11] = 201020 has local density owner 3, physical arc 1. Theorem 7.4 therefore gives
d 0 ( 1 ) 2 < 1 / 6 , δ / 6 − n 10 2 > 0. (159) d_0(1)^2 < 1/6,\qquad\delta/6-n_{10}^2 > 0. \tag*{(159)} d 0 ( 1 ) 2 < 1/6 , δ /6 − n 10 2 > 0. (159)
For cuts i = 4 , 5 , 6 , 7 i=4,5,6,7 i = 4 , 5 , 6 , 7 , the proper source [i − 2, i + 1] is a whole phase of 1201. Lemma B.24 says that its canonical second reciprocal is strictly smaller than both others.
These state comparisons improve the duration-product cuts to
q 4 q 5 > 5 / 4 , q 5 q 6 > 5 / 4. (160) q_4q_5 > 5/4,\qquad q_5q_6 > 5/4. \tag*{(160)} q 4 q 5 > 5/4 , q 5 q 6 > 5/4. (160)
We include the state estimate underlying this strengthening. In canonical coordinates a positive increment transports
Φ q ( a , b , c ) = ( ( q − 1 ) a + q b , c − ( q − 1 ) a q , a ) , a b + a c + b c = 1. \Phi_q(a,b,c)=\left((q-1)a+qb,\frac{c-(q-1)a}{q},a\right),\qquad ab+ac+bc=1. Φ q ( a , b , c ) = ( ( q − 1 ) a + q b , q c − ( q − 1 ) a , a ) , ab + a c + b c = 1.
Suppose q , r ≥ 1 q,r \ge1 q , r ≥ 1 , q r ≤ 5 / 4 qr \le5/4 q r ≤ 5/4 , all three successive states are positive, and their first and last second reciprocals are minimal. Then
a 2 < 3 / 5 , ( a ′ ) 2 < 3 / 5. (161) a^2 < 3/5,\qquad(a')^2 < 3/5. \tag*{(161)} a 2 < 3/5 , ( a ′ ) 2 < 3/5. (161)
where a ′ a' a ′ is active after the first step. To prove the first inequality, divide the initial coordinates by b b b , writing t = a / b ≥ 1 t=a/b \ge1 t = a / b ≥ 1 , s = c / b ≥ 1 s=c/b \ge1 s = c / b ≥ 1 . The last second-versus-third comparison gives
t ≤ T ( q , p ) = q ( 2 p − q ) q 2 − 2 p q + 2 p , p = q r . t \le T(q,p)=\frac{q(2p-q)}{q^2-2pq+2p},\qquad p=qr. t ≤ T ( q , p ) = q 2 − 2 pq + 2 p q ( 2 p − q ) , p = q r .
The denominator is positive for 1 ≤ q , r ≤ 5 / 4 1 \le q,r \le5/4 1 ≤ q , r ≤ 5/4 . For q ≤ 15 / 14 q \le15/14 q ≤ 15/14 , differentiation in p p p and then q q q gives t ≤ T ( 15 / 14 , 5 / 4 ) = 30 / 19 t \le T(15/14,5/4)=30/19 t ≤ T ( 15/14 , 5/4 ) = 30/19 , whence
a 2 = t 2 t + s + t s ≤ t 2 1 + 2 t ≤ 900 1501 < 3 / 5. a^2=\frac{t^2}{t+s+ts}\le\frac{t^2}{1+2t}\le\frac{900}{1501}<3/5. a 2 = t + s + t s t 2 ≤ 1 + 2 t t 2 ≤ 1501 900 < 3/5.
For q ≥ 15 / 14 q \ge15/14 q ≥ 15/14 one has r ≤ 7 / 6 r \le7/6 r ≤ 7/6 . The other terminal comparison gives s ≥ H t − J s \ge Ht-J s ≥ H t − J , where
H = q 2 r 2 − ( q − 1 ) 2 , J = q 2 ( 1 − r − 2 ) . H=\frac{q^2}{r^2}-(q-1)^2,\qquad J=q^2(1-r^{-2}). H = r 2 q 2 − ( q − 1 ) 2 , J = q 2 ( 1 − r − 2 ) .
Here H − J ≥ 593 / 1372 > 0 H-J \ge593/1372>0 H − J ≥ 593/1372 > 0 , and R ( t ) = t 2 / [ t + ( 1 + t ) ( H t − J ) ] R(t)=t^2/[t+(1+t)(Ht-J)] R ( t ) = t 2 / [ t + ( 1 + t ) ( H t − J )] is increasing for t ≥ 1 t \ge1 t ≥ 1 : with u = 1 − r − 2 u=1-r^{-2} u = 1 − r − 2 , its derivative has positive numerator 2 q t { ( 1 − q u ) t − q u } 2qt\{(1-qu)t-qu\} 2 q t {( 1 − q u ) t − q u } , since q u ≤ 65 / 196 < 1 / 2 qu \le65/196<1/2 q u ≤ 65/196 < 1/2 . At the endpoint,
a 2 ≤ R ( T ) = ( 2 r − 1 ) 2 3 ≤ 16 27 < 3 / 5. a^2\le R(T)=\frac{(2r-1)^2}{3}\le\frac{16}{27}<3/5. a 2 ≤ R ( T ) = 3 ( 2 r − 1 ) 2 ≤ 27 16 < 3/5.
Finally S ( a , b , c ) = ( c , b , a ) S(a,b,c)=(c,b,a) S ( a , b , c ) = ( c , b , a ) satisfies Φ q S Φ q = S \Phi_qS\Phi_q=S Φ q S Φ q = S . Applying the first inequality to the reversed two-step state chain, with multipliers r , q r,q r , q , gives the second inequality in (35.25). This is an identity for physical states and does not reverse or join local covectors.
If q r ≤ 5 / 4 qr \le5/4 q r ≤ 5/4 , (35.21) and (35.25) now give
c i + c i + 1 < 8 5 log ( q r ) ≤ 8 5 log ( 5 / 4 ) < 9 25 < 180483 500000 , c_i+c_{i+1}<\frac{8}{5}\log(qr)\le\frac{8}{5}\log(5/4)<\frac{9}{25}<\frac{180483}{500000}, c i + c i + 1 < 5 8 log ( q r ) ≤ 5 8 log ( 5/4 ) < 25 9 < 500000 180483 ,
contrary to the proper forward-pair fee. For pair ( 4 , 5 ) (4,5) ( 4 , 5 ) the forcing cuts are 4 and 6 and its forward-pair source is [ 2 , 7 ] = 120120 [2,7]=120120 [ 2 , 7 ] = 120120 ; for pair ( 5 , 6 ) (5,6) ( 5 , 6 ) they are 5 and 7 with source [ 3 , 8 ] = 201201 [3,8]=201201 [ 3 , 8 ] = 201201 . These are whole phases of 012012, with their charged local middle pair and both caps retained. This proves (35.24).
The exact whole reflection for R03 is
j r ′ = 1 − j 2 − r = j r ( m o d 3 ) , q r ′ = q 2 − r . (162) j'_r=1-j_{2-r}=j_r\pmod3,\qquad q'_r=q_{2-r}. \tag*{(162)} j r ′ = 1 − j 2 − r = j r ( mod 3 ) , q r ′ = q 2 − r . (162)
It preserves the full return and the +1 seam. For example the matrix C = ( − 1 − 3 − 1 1 ) C=\begin{pmatrix}-1&-3\\-1&1\end{pmatrix} C = ( − 1 − 1 − 3 1 ) satisfies C 2 = 4 I C^2=4I C 2 = 4 I , C B j C − 1 = − B 1 − j CB_jC^{-1}=-B_{1-j} C B j C − 1 = − B 1 − j and C R C − 1 = R − 1 CRC^{-1}=R^{-1} C R C − 1 = R − 1 ; hence the reversed inverse frame product again equals R R R . It fixes density owner 1 and permutes the thirteen fee rows in pairs ( 0 , 1 ) (0,1) ( 0 , 1 ) , ( 2 , 3 ) (2,3) ( 2 , 3 ) , ( 4 , 5 ) (4,5) ( 4 , 5 ) , ( 6 , 7 ) (6,7) ( 6 , 7 ) , ( 8 , 10 ) (8,10) ( 8 , 10 ) , ( 11 , 12 ) (11,12) ( 11 , 12 ) , with row 9 fixed. Thus we choose q 3 ≤ q 7 q_3\le q_7 q 3 ≤ q 7 , or A ≤ E A\le E A ≤ E in (35.5). Its exact symmetric identity is
F = ( 1 + C ) ( 1 − A B − D E ) − A E K > 0. F=(1+C)(1-AB-DE)-AEK>0. F = ( 1 + C ) ( 1 − A B − D E ) − A E K > 0.
Since K ≥ 1 K\ge1 K ≥ 1 , this yields
A 2 K ≤ A E K < ( 1 + C ) ( 1 − A B − D E ) ≤ 1 + C = q 5 2 ≤ Q 2 , A^2K\le AEK<(1+C)(1-AB-DE)\le1+C=q_5^2\le Q^2, A 2 K ≤ A E K < ( 1 + C ) ( 1 − A B − D E ) ≤ 1 + C = q 5 2 ≤ Q 2 ,
so q 3 2 < 1 + Q = 553 / 33 q_3^2<1+Q=553/33 q 3 2 < 1 + Q = 553/33 . Lemma 27.2 now proves the root inclusion
( r 3 , r 4 , r 5 , r 6 , r 7 ) ∈ [ 33 / 553 , 1 ] × [ 33 / 520 , 1 ] 4 . (163) (r_3,r_4,r_5,r_6,r_7)\in[\sqrt{33/553},1]\times[33/520,1]^4. \tag*{(163)} ( r 3 , r 4 , r 5 , r 6 , r 7 ) ∈ [ 33/553 , 1 ] × [ 33/520 , 1 ] 4 . (163)
The first lower endpoint, like all rational endpoints, is rounded outward in the actual closed root. The exact positivity guards of Lemma 35.4, the cap, (35.23), the four forward conditions, (35.24) and the thirteen rows are necessary cuts on this root. Their use does not require every formal point of the box to be physical.
Value bounds and the verified closed covers
Let S k ≥ 0 \mathcal{S}_k \ge0 S k ≥ 0 denote any of the qualified state or domain inequalities above, and let Z l = 0 \mathcal{Z}_l = 0 Z l = 0 be a polynomial consequence of the four entries of full signed closure after clearing positive multiplier denominators. At every candidate,
Q = λ 0 B 0 + ∑ j λ j F j + ∑ k η k S k + ∑ l μ l Z l ≥ 0 , λ 0 , λ j , η k ≥ 0 , (164) Q = \lambda_0\mathcal{B}_0 + \sum_j \lambda_j\mathcal{F}_j + \sum_k \eta_k\mathcal{S}_k + \sum_l \mu_l\mathcal{Z}_l \ge0,\qquad\lambda_0,\lambda_j,\eta_k \ge0, \tag*{(164)} Q = λ 0 B 0 + j ∑ λ j F j + k ∑ η k S k + l ∑ μ l Z l ≥ 0 , λ 0 , λ j , η k ≥ 0 , (164)
while μ l \mu_l μ l may have either sign. This is a nonnegative combination of established scalar necessities. Overlapping fee supports are permitted here: the combination is tested as a function of the actual eight costs, rather than asserted to be a new capacity-one payment.
The value predicates have three forms. First, outward intervals can make a necessary intersection empty, or give a strict wrong sign for one of its necessary functions. Second, the true analytic function in (35.28) can be bounded above by its natural interval range or by
f ( x 0 ) + D f ( x 0 ) ( x − x 0 ) + 1 2 ( x − x 0 ) t D 2 f ( X ) ( x − x 0 ) , x ∈ X . (165) f(x_0) + Df(x_0)(x-x_0) + \tfrac{1}{2}(x-x_0)^{t}D^2f(X)(x-x_0),\qquad x \in X. \tag*{(165)} f ( x 0 ) + D f ( x 0 ) ( x − x 0 ) + 2 1 ( x − x 0 ) t D 2 f ( X ) ( x − x 0 ) , x ∈ X . (165)
Here X X X is the whole current five-dimensional rectangle. The centre value, centre derivative and whole-box Hessian belong to the same unmodified function. No derivative is set to zero. Candidate-only value intersections, including positive guards, compactness bounds and fee propagation, are not substituted into the whole-box derivative calculation. If a normalized circuit is used, it is differentiated only when the whole-box lower bound for δ \delta δ is positive, and the derivatives of 1 / δ 1/\delta 1/ δ are retained.
Third, (35.15) permits a polynomial upper envelope before any division by δ \delta δ . Fix an eight-dimensional box J = ∏ i [ l i , h i ] J = \prod_i [l_i,h_i] J = ∏ i [ l i , h i ] enclosing all recovered reciprocal values. For one variable, with the other variables fixed, A i = ( u + v r i ) 2 A_i = (u+vr_i)^2 A i = ( u + v r i ) 2 and
d 3 d r i 3 { A i log r i } = 2 ( u 2 − u v r i + v 2 r i 2 ) r i 3 ≥ 0. (166) \frac{d^3}{dr_i^3}\{A_i\log r_i\} = \frac{2(u^2-uvr_i+v^2r_i^2)}{r_i^3} \ge0. \tag*{(166)} d r i 3 d 3 { A i log r i } = r i 3 2 ( u 2 − uv r i + v 2 r i 2 ) ≥ 0. (166)
The quadratic Hermite interpolant with a double right endpoint is a lower bound; the one with a double left endpoint is an upper bound. Adding the same P i P_i P i gives polynomials L i ≤ N i ≤ U i L_i \le N_i \le U_i L i ≤ N i ≤ U i , of degree at most two in each variable. The fixed-sign remainder in (35.30), including the repeated-node case by continuity, proves their direction.
For a cost combination ∑ i w i N i + c δ \sum_i w_iN_i+c\delta ∑ i w i N i + cδ , one may choose the upper or lower envelope according to the sign of w i w_i w i . The adaptive form instead chooses finite e i ≥ max ( 0 , w i ) e_i \ge\max(0,w_i) e i ≥ max ( 0 , w i ) and uses
w i L i + e i ( U i − L i ) − w i N i = e i ( U i − N i ) + ( e i − w i ) ( N i − L i ) ≥ 0. (167) w_iL_i + e_i(U_i-L_i)-w_iN_i = e_i(U_i-N_i)+(e_i-w_i)(N_i-L_i) \ge0. \tag*{(167)} w i L i + e i ( U i − L i ) − w i N i = e i ( U i − N i ) + ( e i − w i ) ( N i − L i ) ≥ 0. (167)
Thus no sign assumption on w i w_i w i , N i N_i N i or L i L_i L i is needed. After all actual fee weights have been chosen, their exact combined coefficients are enclosed again and the e i e_i e i are increased if necessary to satisfy the displayed inequality. A numerical search bound on proposed weights is never used to truncate this correction.
The resulting polynomial majorant is expanded on J J J in the tensor product of quadratic Bernstein bases. For c 0 + c 1 x + c 2 x 2 c_0+c_1x+c_2x^2 c 0 + c 1 x + c 2 x 2 on [ l , h ] [l,h] [ l , h ] , the three coefficients are
c 0 + l c 1 + l 2 c 2 , c 0 + 1 2 ( l + h ) c 1 + l h c 2 , c 0 + h c 1 + h 2 c 2 . c_0+lc_1+l^2c_2,\qquad c_0+\tfrac{1}{2}(l+h)c_1+lhc_2,\qquad c_0+hc_1+h^2c_2. c 0 + l c 1 + l 2 c 2 , c 0 + 2 1 ( l + h ) c 1 + l h c 2 , c 0 + h c 1 + h 2 c 2 .
The basis functions are nonnegative and sum to one. Hence a strictly negative upper endpoint for every one of the 3 8 = 6561 3^8=6561 3 8 = 6561 coefficients proves that the majorant, and therefore Q Q Q , is negative throughout J J J . Multiaffine state predicates analogously use all their vertices. The coefficients are combined before taking bounds, preserving cancellation among the common state, discriminant, fee and closure terms.
Floating-point linear optimization only proposes weights or a menu entry. Each adopted finite binary weight is treated as its exact dyadic value; its finiteness and required non-negativity are checked, and the original nonlinear combination or all Bernstein coefficients are recomputed with outward intervals. The logarithm primitive uses the rational Padé bounds proved in Section 33, after six positive square roots. More explicitly, for q ≥ 1 q \ge1 q ≥ 1 , take v = q 1 / 64 v = q^{1/64} v = q 1/64 and put t = v 2 = q 1 / 32 t = v^2 = q^{1/32} t = v 2 = q 1/32 and use the polynomials P 4 P_4 P 4 , Q 4 Q_4 Q 4 and upper function u u u defined there. Their exact gap is
g ( t ) = u ( t ) − ℓ ( t ) = ( t − 1 ) 9 ( t + 1 ) 60 t P 4 ( t ) Q 4 ( t ) , 64 { u ( t ) − g ( t ) } ≤ log q ≤ 64 u ( t ) . g(t) = u(t) - \ell(t) = \frac{(t - 1)^9(t + 1)}{60tP_4(t)Q_4(t)}, \qquad64\{u(t) - g(t)\} \le\log q \le64u(t). g ( t ) = u ( t ) − ℓ ( t ) = 60 t P 4 ( t ) Q 4 ( t ) ( t − 1 ) 9 ( t + 1 ) , 64 { u ( t ) − g ( t )} ≤ log q ≤ 64 u ( t ) .
For q < 1 q < 1 q < 1 use log q = − log ( q − 1 ) \log q = -\log(q^{-1}) log q = − log ( q − 1 ) . Derivatives in (165) are those of the true logarithm, not of a rational bound. The stable Hermite formulas separately bound their positive-series tails. The frozen binary64 routines widen basic-operation and square-root bounds outward, preserve their stated round-to-nearest and subnormal-operation contract, and exclude implicit fused multiply-add. Unsupported square-root ranges remain unresolved. Rational constants are constructed from their integer numerators and denominators. The arithmetic and source identities of each adopted kernel are checked with its own terminal input; a newer kernel is not silently substituted for an older terminal’s source.
Any undefined operation, unsuccessful envelope, unrecognized predicate or exhausted search remains unresolved. Only a necessary empty intersection or a strict, whole-box contradiction is a terminal exclusion. The final R02 and R03 terminal inventories contain respectively 34 and 32 predicate labels, all covered by the domain, fee, state, full-closure and value-envelope mechanisms just described. Neither inventory contains a five-dimensional stationarity, objective-Hessian or Krawczyk terminal.
Proposition 35.7 (Verified mixed eight-arc covers). The outward root enclosures of (158) and (163) have complete finite closed covers with respectively 15 310 123 and 10 100 416 terminal exclusions. Every actual terminal has been checked against its own immutable kernel, root and ordered input. No unresolved terminal remains in either cover.
Proof. The structural records rebuild every closed split and every conservative contraction from the original root. A split uses two closed child intervals with the same splitting endpoint. When an unresolved frontier is handed to another kernel, the ordered incoming and outgoing boxes are matched exactly; terminal boxes retain their actual source. Thus the covering induction of Lemma 26.1 applies without dropping shared faces or previously unresolved branches.
For R02 the six complete mathematical evaluations match the final root cover with the following disjoint terminal counts:
2 234 792 + 2 103 709 + 6 225 555 + 1 694 964 + 2 494 082 + 557 021 = 15 310 123. 2\,234\,792 + 2\,103\,709 + 6\,225\,555 + 1\,694\,964 + 2\,494\,082 + 557\,021 = 15\,310\,123. 2 234 792 + 2 103 709 + 6 225 555 + 1 694 964 + 2 494 082 + 557 021 = 15 310 123.
For R03 the final root structure has 2001 original batches and twelve immutable kernel sources. Its ten initial source evaluations account for 9 549 789 terminal inputs. The eleventh source has 234 545 exactly matched inputs, comprising its original 168 811-input evaluation and a complete 65 734-input evaluation of the remaining ordered suffix. The twelfth source supplies a complete 316 082-input evaluation. Hence
9 549 789 + 168 811 + 65 734 + 316 082 = 10 100 416. 9\,549\,789 + 168\,811 + 65\,734 + 316\,082 = 10\,100\,416. 9 549 789 + 168 811 + 65 734 + 316 082 = 10 100 416.
The saved evaluations are adopted only after their actual kernel, root endpoints, complete input body, order, count and output are matched. The final mathematical records and the original-root structural records both have empty unresolved sets. The accompanying verification index for Section H.6 identifies these exact records; the displayed sums summarize their scope but are not a replacement for the root and input correspondences. □
Proofs of Theorems 35.1 and 35.2. Suppose an original global minimizer has the stated strict type. The frame integration gives (138) with the full + 1 +1 + 1 seam. The positive chart of Lemma 35.3 or 35.4, the original compactness cap and the lawful whole reflection place a representative in (158) or (163) . Its noncentral physical return gives the state and common-scale physical costs of (147) –(150) . Its complete sources qualify all the rows and state inequalities specified for that type; no new periodic covector or ambient stationarity condition is imposed.
By Proposition 35.7, its root point belongs to at least one verified terminal. Every contraction and split retains this same candidate, whereas that terminal gives a strict contradiction to one of its necessary domain, state, cost or full-closure conditions. This is impossible. Both strict eight-arc types are therefore excluded from the original globally minimizing class. □
Part IV. Exclusion of the three nine-arc types
The nine-arc type R05
The remaining two-positive-run types require information coupling the complete frame and the actual means. We give the completed application for R05. The same general device—an enclosing domain, auxiliary values at one candidate, necessary contractions, and a closed cover—can be used for other words. The source placements and necessary equations below are specific to R05; no exclusion of R04 or R06 follows from them.
Theorem 36.1 (Exclusion of R05). An original global minimizer cannot have nine essential positive arcs with native labels 021210120 and increments − − + − − + + + + - - + - - + + + + − − + − − + + + + , with the genuine extension j i + 9 = j i + 1 j_{i+9}=j_i+1 j i + 9 = j i + 1 .
The proof first identifies these original minimizers with the candidates to which the source reduction applies. It then excludes their necessary values on a complete finite cover. It does not assert that every positive nine-tuple, or every zero of an auxiliary residual system, is absent.
Reciprocal states and the original source
For W = ( u , v ) W=(u,v) W = ( u , v ) in the state domain of § 3.1 , put
d 0 = 1 + u v , d 1 = u 2 + v 2 − 1 2 v , d 2 = 1 − u v . d_0=\frac{1+u}{v},\qquad d_1=\frac{u^2+v^2-1}{2v},\qquad d_2=\frac{1-u}{v}. d 0 = v 1 + u , d 1 = 2 v u 2 + v 2 − 1 , d 2 = v 1 − u .
All three numbers are positive and satisfy d 0 d 1 + d 1 d 2 + d 2 d 0 = 1 d_0d_1+d_1d_2+d_2d_0=1 d 0 d 1 + d 1 d 2 + d 2 d 0 = 1 . Conversely this relation, with positivity, gives v = 2 / ( d 0 + d 2 ) v=2/(d_0+d_2) v = 2/ ( d 0 + d 2 ) and u = ( d 0 − d 2 ) / ( d 0 + d 2 ) u=(d_0-d_2)/(d_0+d_2) u = ( d 0 − d 2 ) / ( d 0 + d 2 ) . At a switch, two assigned active reciprocals x , z x,z x , z determine the third as
κ ( x , z ) = 1 − x z x + z > 0. \kappa(x,z)=\frac{1-xz}{x+z}>0. κ ( x , z ) = x + z 1 − x z > 0.
The assignment to the three control labels remains part of the state. Write the actual active reciprocals of R05 in native order as
( r 0 , … , r 8 ) = ( E , a , Q , R , b , Z , c , y , f ) , r i + 9 = r i . (168) (r_0,\ldots,r_8)=(E,a,Q,R,b,Z,c,y,f),\qquad r_{i+9}=r_i. \tag*{(168)} ( r 0 , … , r 8 ) = ( E , a , Q , R , b , Z , c , y , f ) , r i + 9 = r i . (168)
Here Q Q Q is a reciprocal coordinate, not the multiplier bound 520 33 \frac{520}{33} 33 520 . The outer mean owners are E , Z E,Z E , Z , the density owners are a , b a,b a , b , and the central own pair is Q , R Q,R Q , R .
To check the state and physical-time convention, let x = r i x=r_i x = r i , z = r i − 1 z=r_{i-1} z = r i − 1 , w = r i + 1 w=r_{i+1} w = r i + 1 , α = ε i − 1 \alpha=\varepsilon_{i-1} α = ε i − 1 and β = ε i \beta=\varepsilon_i β = ε i . The exact multiplier and physical duration are
q i = ( x + z ) α ( x + w ) β ( 1 + x 2 ) ( α + β ) / 2 , τ i = x log q i > 0. (169) q_i=\frac{(x+z)^\alpha(x+w)^\beta}{(1+x^2)^{(\alpha+\beta)/2}},\qquad\tau_i=x\log q_i>0. \tag*{(169)} q i = ( 1 + x 2 ) ( α + β ) /2 ( x + z ) α ( x + w ) β , τ i = x log q i > 0. (169)
Indeed, in the mode-0 chart d 0 = x d_0=x d 0 = x is constant, d 1 o u t + x = q i ( d 1 i n + x ) d_1^{\mathrm{out}}+x=q_i(d_1^{\mathrm{in}}+x) d 1 out + x = q i ( d 1 in + x ) and d 2 o u t + x = ( d 2 i n + x ) / q i d_2^{\mathrm{out}}+x=(d_2^{\mathrm{in}}+x)/q_i d 2 out + x = ( d 2 in + x ) / q i . At either end use the assigned neighbor or κ ( x , ⋅ ) \kappa(x,\cdot) κ ( x , ⋅ ) ; the identity x + κ ( x , z ) = ( 1 + x 2 ) / ( x + z ) x+\kappa(x,z)=(1+x^2)/(x+z) x + κ ( x , z ) = ( 1 + x 2 ) / ( x + z ) gives all four cases of (169) . Positive endpoint states remain positive along the arc. These formulas give a necessary coordinate description of the actual trajectory, while full signed frame closure and real winding are retained separately.
Lemma 36.2 (Qualification of the R05 sources). Every original R05 global minimizer admits one jointly normalized full maximizing lift on the actual unfolding, with the matrix twist and both state-covector twists. Its proper returned source is native [ − 1 , 6 ] [-1,6] [ − 1 , 6 ] , with whole phase + 1 +1 + 1 equal to 01020212. On this source the density owners are native 1, 4, the exterior mean owners are native 0, 5, and the own pair is native 2, 3. Moreover a 2 , b 2 < 1 / 6 a^2,b^2<1/6 a 2 , b 2 < 1/6 .
Proof. Take two cuts strictly inside native arcs 7 and 8, in that order. Their overlap contains the next twisted translate [ 9 , 14 ] = 102021 = 1 ∣ 0202 ∣ 1 [9,14]=102021=1|0202|1 [ 9 , 14 ] = 102021 = 1∣0202∣1 of the complete capped source [ 0 , 5 ] = 021210 = 0 ∣ 2121 ∣ 0 [0,5]=021210=0|2121|0 [ 0 , 5 ] = 021210 = 0∣2121∣0 . Its four-core has the two internal switching roots and both positive caps required by Lemma B.3 . Between the two cuts is the actual switch from label 2 to label 0. Proposition B.114 therefore applies: first it proves normality, then identifies the whole jointly normalized lifts on the overlap and proves their matrix and state-covector twists. This supplies the full lift, rather than an identification of independently selected determinants.
The twisted continuation gives j − 1 ⋯ j 6 = 20212101 j_{-1}\cdots j_6=20212101 j − 1 ⋯ j 6 = 20212101 , hence the stated phase+1 1 1 eight-source. It is proper since 8 < 9, so Lemma 6.1 retains it in a genuine fundamental interval. In its local numbering the four-core is [ 2 , 5 ] [2,5] [ 2 , 5 ] , the density owners are 2, 5, and the own pair is 3, 4; subtracting one gives exactly the native dictionary in the statement. Both actual returns [ − 1 , 1 ] [-1,1] [ − 1 , 1 ] and [ 4 , 6 ] [4,6] [ 4 , 6 ] , and both exterior caps, remain present. Section C.1.7 supplies the chronological root jets and density profiles on this same lift.
For the sharper density bounds use the proper six-source [ 7 , 12 ] = 201020 [7,12]=201020 [ 7 , 12 ] = 201020 , whose local arc 3 is physical arc 1, and the reversed source [ 7 , 6 , 5 , 4 , 3 , 2 ] [7,6,5,4,3,2] [ 7 , 6 , 5 , 4 , 3 , 2 ] with the whole reflected label map j ↦ 1 − j j\mapsto1-j j ↦ 1 − j . The latter is also 201020, and its local arc 3 is physical arc 4. Theorem 7.4 gives a 2 , b 2 < 1 / 6 a^2,b^2<1/6 a 2 , b 2 < 1/6 . Each six-source is placed on its own genuine interval if necessary; only these physical state conclusions are exported. Their covectors are not spliced into the common returned-source lift. □
Define the actual uniform reciprocal-shear mean by
M ( x ; k , z ) = ∫ 0 1 ( x − ( ( 1 − t ) k + t z ) x + ( ( 1 − t ) k + t z ) ) 2 d t . (170) M(x;k,z)=\int_0^1\left(\frac{x-((1-t)k+tz)}{x+((1-t)k+tz)}\right)^2\,\mathrm{d}t. \tag*{(170)} M ( x ; k , z ) = ∫ 0 1 ( x + (( 1 − t ) k + t z ) x − (( 1 − t ) k + t z ) ) 2 d t . (170)
On an actual positive window this is the uniform average over [ k , z ] [k,z] [ k , z ] . It is not a physical-time average. The source identifies three values at the same candidate:
μ = M ( Q ; a , R ) = M ( R ; b , Q ) , ν a = M ( E ; a , f ) , ν b = M ( Z ; b , c ) . (171) \mu=M(Q;a,R)=M(R;b,Q),\qquad\nu_a=M(E;a,f),\qquad\nu_b=M(Z;b,c). \tag*{(171)} μ = M ( Q ; a , R ) = M ( R ; b , Q ) , ν a = M ( E ; a , f ) , ν b = M ( Z ; b , c ) . (171)
In particular, the two exterior functions retain their different owners even when their values happen to coincide.
The source reduction and its closed domain
We use three previously established inputs from the complete-source proof collection described in Section H.10. The first consists of the continuous source, calibration, cap and domain comparisons; we call it D1. Its statements retain the actual owners, selected roots, complete lift and permitted limits. The second, D2, is the closed-root, continuation and half-F F F covering theorem. The third, D3, derives the physical necessary-value equations and their smooth auxiliary extensions. Section H.10 gives their mathematical interfaces; the electronic collection supplies the original physical-graph texts, the 29 domain sources, both complete coefficient proofs, and the finite covers. D1 is a geometric implication and D2 is a covering theorem; neither follows from the final finite table being empty.
Set
H = Q + R , p = R − a H , t = Q − b H , s = p + t , η s = p − t , e = η s s , δ = R − Q H . (172) H = Q + R,\quad p = \frac{R-a}{H},\quad t = \frac{Q-b}{H},\quad s=p+t,\quad\eta_s=p-t,\quad e=\frac{\eta_s}{s},\quad\delta=\frac{R-Q}{H}. \tag*{(172)} H = Q + R , p = H R − a , t = H Q − b , s = p + t , η s = p − t , e = s η s , δ = H R − Q . (172)
The positive inner windows make s > 0 s>0 s > 0 . The actual calibration equality in (171) selects its original positive branch; no alternate root of its cleared quadratic is admitted. In these coordinates
R = H 2 ( 1 + δ ) , Q = H 2 ( 1 − δ ) , a = H 2 ( 1 − s + δ − η s ) , b = H 2 ( 1 − s − δ + η s ) , R=\frac{H}{2}(1+\delta),\quad Q=\frac{H}{2}(1-\delta),\quad a=\frac{H}{2}(1-s+\delta-\eta_s),\quad b=\frac{H}{2}(1-s-\delta+\eta_s), R = 2 H ( 1 + δ ) , Q = 2 H ( 1 − δ ) , a = 2 H ( 1 − s + δ − η s ) , b = 2 H ( 1 − s − δ + η s ) ,
D = ( a + b ) 2 2 − ( R − a ) ( Q − b ) = H 2 ( 1 2 − s + s 2 + η s 2 4 ) > 0. (173) D=\frac{(a+b)^2}{2}-(R-a)(Q-b)=H^2\left(\frac{1}{2}-s+\frac{s^2+\eta_s^2}{4}\right)>0. \tag*{(173)} D = 2 ( a + b ) 2 − ( R − a ) ( Q − b ) = H 2 ( 2 1 − s + 4 s 2 + η s 2 ) > 0. (173)
The symbol D D D in this section denotes this area parameter. With h k = ( 1 + k 2 ) / 2 h_k=(1+k^2)/2 h k = ( 1 + k 2 ) /2 , put F = D / ( h a h b ) F=D/(h_a h_b) F = D / ( h a h b ) and r = ( h a − h b ) / D r=(h_a-h_b)/D r = ( h a − h b ) / D .
Proposition 36.3 (Retained source-to-domain input). Every original global minimizer of type R05, qualified as in Lemma 36.2, either contradicts one of the retained source comparisons or has its coordinates ( s , e , H , c , f ) (s,e,H,c,f) ( s , e , H , c , f ) in the closed box
B 05 = [ 49 125 , 2341 4000 ] × [ − 327 1000 , 327 1000 ] × [ 13 10 , 2 ] × [ 3 20 , 4 3 ] 2 . (174) B_{05}=\left[\frac{49}{125},\frac{2341}{4000}\right]\times\left[-\frac{327}{1000},\frac{327}{1000}\right]\times\left[\frac{13}{10},2\right]\times\left[\frac{3}{20},\frac{4}{3}\right]^2. \tag*{(174)} B 05 = [ 125 49 , 4000 2341 ] × [ − 1000 327 , 1000 327 ] × [ 10 13 , 2 ] × [ 20 3 , 3 4 ] 2 . (174) The same candidate retains its complete caps, original normal full lift, all maximizing conditions, selected calibration and return roots, actual means, signed endpoints and real winding. The accepted closed-root and continuation covers reduce it further to at least one of the fixed 46 383 complete original cells.
Identification and composition of the retained inputs. The source comparisons in D1 apply to the actual owners identified in Lemma 36.2. We spell out their dependencies and the inclusions used here. The complete state compactness theorem, transported by whole cyclic phases, bounds every endpoint reciprocal in [ 1 / 15 , 400 / 99 ] [1/15,400/99] [ 1/15 , 400/99 ] and each multiplier by 520 33 \frac{520}{33} 33 520 . It therefore applies also to the third components at the initial and terminal switches, rather than only to the five displayed coordinates. The two proper six-sources give the sharpened density bounds. The returned eight-source supplies its two genuine return jets, the same determinant and the unsquared signed sum of the two constants. The harmonic estimate uses that selected normal lift after positivity of its determinant and trace has been established. The global cost comparison is exactly (2.19), with every physical arc charged once.
The two-cap and selected-calibration comparisons give
49 125 < s < 2341 4000 , ∣ η s ∣ < 327 1000 s , 13 10 < H < 2. \frac{49}{125}<s<\frac{2341}{4000},\qquad|\eta_s|<\frac{327}{1000}s,\qquad\frac{13}{10}<H<2. 125 49 < s < 4000 2341 , ∣ η s ∣ < 1000 327 s , 10 13 < H < 2.
Their shape bounds are whole-domain comparisons, including both signs of imbalance and balance. The lower scale bounds come from the actual own-pair payment and the two-cap floor, not a scale chosen independently of the candidate. In particular a , b > 3 / 20 a,b>3/20 a , b > 3/20 and a + b > 5 / 8 a+b>5/8 a + b > 5/8 . Positive returns give f > a f>a f > a and c > b c>b c > b ; the selected-return and state-product comparisons give c , f < 4 / 3 c,f<4/3 c , f < 4/3 . These are the five required coordinate inclusions. Replacing their strict endpoints by the closed endpoints of (36.7) only enlarges the domain.
The additional necessary conditions used during evaluation retain those same premises. They include the selected calibration cone, ∣ r ∣ ≤ 3 / 7 |r|\leq3/7 ∣ r ∣ ≤ 3/7 , F > 1 / 20 F>1/20 F > 1/20 , the whole-sector exclusion F ≥ 1 / 2 F\geq1/2 F ≥ 1/2 , the return collars f − a , c − b > 1 / 1000 f-a,c-b>1/1000 f − a , c − b > 1/1000 , and the same-side owner and state-product bounds. The return collars are proved from the undivided frame equations on the full closed thickness [ 0 , 1 / 1000 ] [0,1/1000] [ 0 , 1/1000 ] ; they are not inferred by extending a zero-width derivative test. The owner bounds use their own selected mean roots and true leaves. Thus a zero exterior leaf is not confused with a zero return. All omitted maximizing or winding conditions remain conjuncts even when an evaluator uses only a subset of them.
The accepted root cover of (36.7) has 31 142 complete terminal cells; 18 934 have necessary contradictions and 12 208 are retained. The continuation begins from exactly those 12 208 cells. Its 114 055 closed splits have 67 989 excluded and 58 274 retained terminal cells. The half-F F F overlay then excludes 11 891 of these complete cells and retains exactly 46 383, without changing an original endpoint. The half-F F F theorem includes equality F = 1 / 2 F = 1/2 F = 1/2 ; evaluating on the larger collar [ 1 / 20 , 1 / 2 ] [1/20,1/2] [ 1/20 , 1/2 ] is conservative. Each root or continuation split has two closed children with union equal to the parent, and every excluded cell has its stated necessary contradiction. Induction on these accepted trees, followed by the complete-cell overlay, proves the final inclusion. The counts check the assembly; the recorded paths, shared faces and predicates prove coverage. □
The original source comparisons and these upstream covers are used at their previously verified scopes. Their proof texts, full coefficient arrays, closed forests and terminal data are supplied as separate inputs in the electronic collection. In particular, the continuation’s complete frozen forest carries the covering and exclusion obligations; intermediate scheduling records are not additional mathematical premises. The final 46 383-cell verification described below independently reconstructs the required values and contradictions on that fixed table. It does not claim a new replay of the earlier root or continuation forests. The electronic source index records these two layers separately.
The complete caps and necessary value equations.
The finite argument works with an outer box of the following thirteen values at one candidate:
z = ( a , b , Q , R , c , y , f , E , Z , D , μ , ν a , ν b ) . (175) z = (a,b,Q,R,c,y,f,E,Z,D,\mu,\nu_a,\nu_b). \tag*{(175)} z = ( a , b , Q , R , c , y , f , E , Z , D , μ , ν a , ν b ) . (175)
The three parameters θ = ( s , e , H ) \theta= (s,e,H) θ = ( s , e , H ) range over their entire original cell intervals. They are fixed when differentiating in z z z . On the auxiliary open domain
Ω 05 = ( 0 , ∞ ) 10 × R 3 \Omega_{05} = (0,\infty)^{10} \times\mathbb{R}^{3} Ω 05 = ( 0 , ∞ ) 10 × R 3
all formulas below are smooth. The last three coordinates are unrestricted real variables for this analytic extension. They still represent the actual means (36.4) at a physical candidate. An auxiliary coordinate is not an additional physical degree of freedom.
Put K = ( 3 + μ ) / 2 K = (3+\mu)/2 K = ( 3 + μ ) /2 , K a = ( 3 + ν a ) / 2 K_a = (3+\nu_a)/2 K a = ( 3 + ν a ) /2 and K b = ( 3 + ν b ) / 2 K_b = (3+\nu_b)/2 K b = ( 3 + ν b ) /2 . The original common signed lift gives
λ a = − K D 2 − h a 2 K a − h b 2 K b 2 D , λ b = − K D 2 + h a 2 K a − h b 2 K b 2 D . (176) \lambda_a = -\frac{KD}{2} - \frac{h_a^2K_a-h_b^2K_b}{2D}, \qquad
\lambda_b = -\frac{KD}{2} + \frac{h_a^2K_a-h_b^2K_b}{2D}. \tag*{(176)} λ a = − 2 K D − 2 D h a 2 K a − h b 2 K b , λ b = − 2 K D + 2 D h a 2 K a − h b 2 K b . (176)
Indeed its unsquared equations are λ a + λ b = − K D \lambda_a+\lambda_b=-KD λ a + λ b = − K D and K h a 2 K a − λ a 2 = K h b 2 K b − λ b 2 K h_a^2 K_a-\lambda_a^2=K h_b^2 K_b-\lambda_b^2 K h a 2 K a − λ a 2 = K h b 2 K b − λ b 2 . Since K D > 0 KD>0 K D > 0 , subtracting the latter equations determines the difference and gives (36.9). There is no independent square-root choice on either side. The determinant is
Γ = K 2 ( h a 2 K a + h b 2 K b ) − K 2 D 2 4 − ( h a 2 K a − h b 2 K b ) 2 4 D 2 . \Gamma= \frac{K}{2}(h_a^2K_a+h_b^2K_b)-\frac{K^2D^2}{4}-\frac{(h_a^2K_a-h_b^2K_b)^2}{4D^2}. Γ = 2 K ( h a 2 K a + h b 2 K b ) − 4 K 2 D 2 − 4 D 2 ( h a 2 K a − h b 2 K b ) 2 .
Both negative squares remain in every use of this expression; Γ \Gamma Γ is not a physical cost.
For one side ( k , P , O , ν , λ ) (k,P,O,\nu,\lambda) ( k , P , O , ν , λ ) define
ν k = k ( k + P ) 2 , ω k = k h k k + O , ρ k = 1 − k 2 k 2 , \nu_k = \frac{k(k+P)}{2}, \qquad\omega_k = \frac{k h_k}{k+O}, \qquad\rho_k = 1-\frac{k^2}{k^2}, ν k = 2 k ( k + P ) , ω k = k + O k h k , ρ k = 1 − k 2 k 2 ,
I k = ( 1 − μ ) ( ω k − ν k ) − 2 k 2 log ( ω k / ν k ) + k 4 ( 1 / ν k − 1 / ω k ) . I_k = (1-\mu)(\omega_k-\nu_k)-2k^2\log(\omega_k/\nu_k)+k^4(1/\nu_k-1/\omega_k). I k = ( 1 − μ ) ( ω k − ν k ) − 2 k 2 log ( ω k / ν k ) + k 4 ( 1/ ν k − 1/ ω k ) .
The complete cap residual is
P ( k , P , O , ν , λ ) = − h k 2 3 + ν 2 − 2 λ ω k − K ω k 2 + 1 2 + ( ρ k 2 − μ ) ω k 2 2 k 2 − ( 1 + k 2 ) ω k I k 2 k 2 . (177) \mathcal{P}(k,P,O,\nu,\lambda)=-h_k^2\frac{3+\nu}{2}-2\lambda\omega_k-K\omega_k^2+\frac{1}{2}
+\frac{(\rho_k^2-\mu)\omega_k^2}{2k^2}
-\frac{(1+k^2)\omega_k I_k}{2k^2}. \tag*{(177)} P ( k , P , O , ν , λ ) = − h k 2 2 3 + ν − 2 λ ω k − K ω k 2 + 2 1 + 2 k 2 ( ρ k 2 − μ ) ω k 2 − 2 k 2 ( 1 + k 2 ) ω k I k . (177)
Use ( a , Q , E , ν a , λ a ) (a,Q,E,\nu_a,\lambda_a) ( a , Q , E , ν a , λ a ) and ( b , R , Z , ν b , λ b ) (b,R,Z,\nu_b,\lambda_b) ( b , R , Z , ν b , λ b ) to obtain P a , P b \mathcal{P}_a,\mathcal{P}_b P a , P b . The logarithmic integral term is part of each cap. For example, replacing ω k \omega_k ω k by a variable w w w gives I k ′ ( w ) = ( 1 − k 2 / w ) 2 − μ I_k'(w)=(1-k^2/w)^2-\mu I k ′ ( w ) = ( 1 − k 2 / w ) 2 − μ and I k ( ν k ) = 0 I_k(\nu_k)=0 I k ( ν k ) = 0 , which recover that term from the original primitive without truncation.
For the central functions set
U = c f − 1 + ( c + f ) y , Y = 1 + y 2 , V = U + 2 Y = c f + ( c + f ) y + 1 + 2 y 2 , U=cf-1+(c+f)y,\qquad Y=1+y^2,\qquad V=U+2Y=cf+(c+f)y+1+2y^2, U = c f − 1 + ( c + f ) y , Y = 1 + y 2 , V = U + 2 Y = c f + ( c + f ) y + 1 + 2 y 2 ,
F = 4 U + 2 U 2 / Y ( 1 + c 2 ) ( 1 + f 2 ) , v a = 2 y ( c + y ) ( 1 + f 2 ) V , v b = 2 y ( f + y ) ( 1 + c 2 ) V . (178) \mathcal{F}=\frac{4U+2U^2/Y}{(1+c^2)(1+f^2)},\qquad v_a=\frac{2y(c+y)(1+f^2)}{V},\qquad v_b=\frac{2y(f+y)(1+c^2)}{V}. \tag*{(178)} F = ( 1 + c 2 ) ( 1 + f 2 ) 4 U + 2 U 2 / Y , v a = V 2 y ( c + y ) ( 1 + f 2 ) , v b = V 2 y ( f + y ) ( 1 + c 2 ) . (178)
Here F i \mathcal{F}_i F i denotes its partial derivative with respect to i i i . Full frame closure supplies F = F = D / ( h a h b ) \mathcal{F}=F=D/(h_a h_b) F = F = D / ( h a h b ) and the two undivided equations in the following list. With d = h a − h b d=h_a-h_b d = h a − h b , the first ten necessary zero rows are
D − ( a + b ) 2 2 + ( R − a ) ( Q − b ) = 0 , D-\frac{(a+b)^2}{2}+(R-a)(Q-b)=0, D − 2 ( a + b ) 2 + ( R − a ) ( Q − b ) = 0 ,
M ( Q ; a , R ) − μ = M ( R ; b , Q ) − μ = 0 , M(Q;a,R)-\mu=M(R;b,Q)-\mu=0, M ( Q ; a , R ) − μ = M ( R ; b , Q ) − μ = 0 ,
M ( E ; a , f ) − ν a = M ( Z ; b , c ) − ν b = 0 , M(E;a,f)-\nu_a=M(Z;b,c)-\nu_b=0, M ( E ; a , f ) − ν a = M ( Z ; b , c ) − ν b = 0 ,
F − D h a h b = 0 , \mathcal{F}-\frac{D}{h_a h_b}=0, F − h a h b D = 0 ,
A a ≔ 2 E ( f − a ) − ( 1 − d D ) ( 1 + a 2 ) + v a = 0 , A_a\coloneqq2E(f-a)-\left(1-\frac{d}{D}\right)(1+a^2)+v_a=0, A a : = 2 E ( f − a ) − ( 1 − D d ) ( 1 + a 2 ) + v a = 0 ,
A b ≔ 2 Z ( c − b ) − ( 1 + d D ) ( 1 + b 2 ) + v b = 0 , A_b\coloneqq2Z(c-b)-\left(1+\frac{d}{D}\right)(1+b^2)+v_b=0, A b : = 2 Z ( c − b ) − ( 1 + D d ) ( 1 + b 2 ) + v b = 0 ,
P a = P b = 0. (179) \mathcal{P}_a=\mathcal{P}_b=0. \tag*{(179)} P a = P b = 0. (179)
The original full signed matrix equation implies these rows. Its nonzero central off-diagonal entry and determinant one recover the remaining entry on the strict source graph. This algebra does not recover real winding from a matrix equality; the actual path supplies winding throughout.
We give the remaining three rows explicitly, to distinguish physical stationarity already proved on the original source graph from differentiation of an arbitrary auxiliary system. Define the symmetric physical-cost potential
ϕ ( x , t ) = 3 x t − 1 − ( x 2 + t 2 ) log ( x + t ) + x 2 log ( 1 + x 2 ) + t 2 log ( 1 + t 2 ) 2 , (180) \phi(x,t)=3xt-1-(x^2+t^2)\log(x+t)+\frac{x^2\log(1+x^2)+t^2\log(1+t^2)}{2}, \tag*{(180)} ϕ ( x , t ) = 3 x t − 1 − ( x 2 + t 2 ) log ( x + t ) + 2 x 2 log ( 1 + x 2 ) + t 2 log ( 1 + t 2 ) , (180)
whose first derivative is
g 1 ( x , t ) = 2 t + 2 x t x + t − x 1 + x 2 − x log ( x + t ) 2 1 + x 2 . g_1(x,t)=2t+\frac{2xt}{x+t}-\frac{x}{1+x^2}-x\log\frac{(x+t)^2}{1+x^2}. g 1 ( x , t ) = 2 t + x + t 2 x t − 1 + x 2 x − x log 1 + x 2 ( x + t ) 2 .
Integrating the physical density on one complete arc gives
x { d 2 i n − d 2 o u t + d 1 o u t − d 1 i n } − x 2 log q . x\{d_2^{\mathrm{in}}-d_2^{\mathrm{out}}+d_1^{\mathrm{out}}-d_1^{\mathrm{in}}\}-x^2\log q. x { d 2 in − d 2 out + d 1 out − d 1 in } − x 2 log q .
At a switch, the paired endpoint contributions equal ε ( 3 x t − 1 ) \varepsilon(3xt-1) ε ( 3 x t − 1 ) , since x [ κ ( x , t ) − t ] + t [ κ ( x , t ) − x ] = 1 − 3 x t x[\kappa(x,t)-t]+t[\kappa(x,t)-x]=1-3xt x [ κ ( x , t ) − t ] + t [ κ ( x , t ) − x ] = 1 − 3 x t . Summing over the twisted circle and using (36.2) gives the once-counted cost
C 05 = − ϕ ( E , a ) − ϕ ( a , Q ) + ϕ ( Q , R ) − ϕ ( R , b ) − ϕ ( b , Z ) + ϕ ( Z , c ) + ϕ ( c , y ) + ϕ ( y , f ) + ϕ ( f , E ) . (181) C_{05}=-\phi(E,a)-\phi(a,Q)+\phi(Q,R)-\phi(R,b)-\phi(b,Z)
+\phi(Z,c)+\phi(c,y)+\phi(y,f)+\phi(f,E). \tag*{(181)} C 05 = − ϕ ( E , a ) − ϕ ( a , Q ) + ϕ ( Q , R ) − ϕ ( R , b ) − ϕ ( b , Z ) + ϕ ( Z , c ) + ϕ ( c , y ) + ϕ ( y , f ) + ϕ ( f , E ) . (181)
In particular the closing edge occurs once, and the negative terms cannot be discarded or optimized independently.
The retained source chain-rule theorem D3 differentiates this complete physical cost on its strictly positive whole-period frame graph, before restricting the resulting values to the actual means and complete caps. Its geometric justification is the local calibrated competitor family: the selected calibration and central roots vary smoothly at the actual positive point; three signed frame entries, including a nonzero central off-diagonal entry, and determinant one give the complete frame equation; and positivity of all nine durations and state coordinates persists in a neighbourhood. The resulting frame paths have fixed signed endpoints and the same real winding by continuity. The whole-period geometric reconstruction premise of Section 2 then makes these genuine competitors. This step uses more than the injectivity of a necessary five-coordinate description. Minimizer-only cap or density inequalities are not imposed on the nearby competitors. The original mean–frame proof and the physical chain rule are supplied in the electronic source collection, with the interface summarized in Section H.10. At the actual minimizing point its three central necessary values are R i = 0 R_i=0 R i = 0 , i = y , c , f i=y,c,f i = y , c , f , where
Q c = h a h b [ K − 3 d 2 2 D 2 + d ( h b ν b − h a ν a ) 2 D 2 ] , Q_c=h_a h_b\left[K-\frac{3d^2}{2D^2}+\frac{d(h_b\nu_b-h_a\nu_a)}{2D^2}\right], Q c = h a h b [ K − 2 D 2 3 d 2 + 2 D 2 d ( h b ν b − h a ν a ) ] ,
B y = g 1 ( y , c ) + g 1 ( y , f ) − 3 + ν a 4 ( v a ) y − 3 + ν b 4 ( v b ) y , B_y=g_1(y,c)+g_1(y,f)-\frac{3+\nu_a}{4}(v_a)_y-\frac{3+\nu_b}{4}(v_b)_y, B y = g 1 ( y , c ) + g 1 ( y , f ) − 4 3 + ν a ( v a ) y − 4 3 + ν b ( v b ) y ,
B c = g 1 ( c , Z ) + g 1 ( c , y ) − 3 + ν a 4 ( v a ) c − 3 + ν b 4 ( v b ) c − Z K b , B_c=g_1(c,Z)+g_1(c,y)-\frac{3+\nu_a}{4}(v_a)_c-\frac{3+\nu_b}{4}(v_b)_c-ZK_b, B c = g 1 ( c , Z ) + g 1 ( c , y ) − 4 3 + ν a ( v a ) c − 4 3 + ν b ( v b ) c − Z K b ,
B f = g 1 ( f , E ) + g 1 ( f , y ) − 3 + ν a 4 ( v a ) f − 3 + ν b 4 ( v b ) f − E K a , B_f=g_1(f,E)+g_1(f,y)-\frac{3+\nu_a}{4}(v_a)_f-\frac{3+\nu_b}{4}(v_b)_f-EK_a, B f = g 1 ( f , E ) + g 1 ( f , y ) − 4 3 + ν a ( v a ) f − 4 3 + ν b ( v b ) f − E K a ,
R i = B i − F i Q c . (182) R_i=B_i-\mathcal{F}_iQ_c. \tag*{(182)} R i = B i − F i Q c . (182)
All central derivatives here are ordinary partial derivatives of the explicit rational functions (178) . The c c c equation uses Z , ν b Z,\nu_b Z , ν b and the f f f equation uses E , ν a E,\nu_a E , ν a . The signed contrast in Q c Q_c Q c and both owner terms are essential. The source derivation of these physical necessary values is an input distinct from the analytic extension used below: freezing μ , ν a , ν b \mu,\nu_a,\nu_b μ , ν a , ν b in (181) would not prove (182) .
Three further zero rows retain the original parameters:
Q + R − H = 0 , a + b − ( 1 − s ) ( Q + R ) = 0 , ( 1 − e ) ( R − a ) − ( 1 + e ) ( Q − b ) = 0. (183) Q+R-H=0,\qquad a+b-(1-s)(Q+R)=0,\qquad(1-e)(R-a)-(1+e)(Q-b)=0. \tag*{(183)} Q + R − H = 0 , a + b − ( 1 − s ) ( Q + R ) = 0 , ( 1 − e ) ( R − a ) − ( 1 + e ) ( Q − b ) = 0. (183)
Thus (179) –(183) give sixteen necessary equalities. They are used as a necessary subsystem of the original physical problem, with all original source conditions still retained.
Additional identities and derivatives on the full auxiliary domain.
The cap identities below improve interval enclosures without introducing a new physical payment. For one side put
L ( k , P , O , ν ) = h k [ ( 1 − ν ) ( 1 + O / k ) + ( 1 − μ ) ( 1 + P / k ) − 4 k k + O − 4 k k + P + 4 log 1 + k 2 ( k + O ) ( k + P ) ] − 2. L(k,P,O,\nu)=h_k\left[(1-\nu)(1+O/k)+(1-\mu)(1+P/k)-\frac{4k}{k+O}-\frac{4k}{k+P}+4\log\frac{1+k^2}{(k+O)(k+P)}\right]-2. L ( k , P , O , ν ) = h k [ ( 1 − ν ) ( 1 + O / k ) + ( 1 − μ ) ( 1 + P / k ) − k + O 4 k − k + P 4 k + 4 log ( k + O ) ( k + P ) 1 + k 2 ] − 2.
Let L a , L b L_a,L_b L a , L b be its two actual-side expressions and put Δ = h a 2 ( 3 + ν a ) − h b 2 ( 3 + ν b ) \Delta=h_a^2(3+\nu_a)-h_b^2(3+\nu_b) Δ = h a 2 ( 3 + ν a ) − h b 2 ( 3 + ν b ) . Substitution into the complete expression (177) , using ω k / υ k = ( 1 + k 2 ) / [ ( k + O ) ( k + P ) ] \omega_k/\upsilon_k=(1+k^2)/[(k+O)(k+P)] ω k / υ k = ( 1 + k 2 ) / [( k + O ) ( k + P )] , gives identities on all of Ω 05 \Omega_{05} Ω 05 :
2 P a ω a = L a + ( 3 + μ ) D + Δ D , 2 P b ω b = L b + ( 3 + μ ) D − Δ D . (184) \frac{2\mathcal{P}_a}{\omega_a}=L_a+(3+\mu)D+\frac{\Delta}{D},\qquad\frac{2\mathcal{P}_b}{\omega_b}=L_b+(3+\mu)D-\frac{\Delta}{D}. \tag*{(184)} ω a 2 P a = L a + ( 3 + μ ) D + D Δ , ω b 2 P b = L b + ( 3 + μ ) D − D Δ . (184)
Consequently the following two zero rows can be added, giving eighteen:
S = L a + L b + 2 ( 3 + μ ) D = 0 , T = L a + L b + ( 3 + μ ) [ ( a + b ) 2 − 2 ( R − a ) ( Q − b ) ] = 0. S=L_a+L_b+2(3+\mu)D=0,\qquad T=L_a+L_b+(3+\mu)\left[(a+b)^2-2(R-a)(Q-b)\right]=0. S = L a + L b + 2 ( 3 + μ ) D = 0 , T = L a + L b + ( 3 + μ ) [ ( a + b ) 2 − 2 ( R − a ) ( Q − b ) ] = 0.
Indeed S = 2 P a / ω a + 2 P b / ω b S=2\mathcal{P}_a/\omega_a+2\mathcal{P}_b/\omega_b S = 2 P a / ω a + 2 P b / ω b , while T T T subtracts 2 ( 3 + μ ) 2(3+\mu) 2 ( 3 + μ ) times the area row from S S S . No cap equality is imposed before differentiating these functions.
The final extended systems also use six redundant zero rows:
D ( L a − L b ) + 2 Δ = 0 , D A a = D A b = 0 , D(L_a-L_b)+2\Delta=0,\qquad DA_a=DA_b=0, D ( L a − L b ) + 2Δ = 0 , D A a = D A b = 0 ,
D 2 B i − F i Q ~ c = 0 ( i = y , c , f ) , D^2B_i-\mathcal{F}_i\widetilde{Q}_c=0\qquad(i=y,c,f), D 2 B i − F i Q c = 0 ( i = y , c , f ) ,
Q ~ c = h a h b [ K D 2 − 3 d 2 2 + d ( h b ν b − h a ν a ) 2 ] . (185) \widetilde{Q}_c=h_a h_b\left[KD^2-\frac{3d^2}{2}+\frac{d(h_b\nu_b-h_a\nu_a)}{2}\right]. \tag*{(185)} Q c = h a h b [ K D 2 − 2 3 d 2 + 2 d ( h b ν b − h a ν a ) ] . (185)
The first equals D ( 2 P a / ω a − 2 P b / ω b ) D(2\mathcal{P}_a/\omega_a-2\mathcal{P}_b/\omega_b) D ( 2 P a / ω a − 2 P b / ω b ) ; the last three equal D 2 R i D^2R_i D 2 R i . These identities hold away from the candidate zero set as well. Their derivatives therefore retain all product terms. For a residual R R R and the coordinate vector e D e_D e D ,
∇ 2 ( D 2 R ) = D 2 ∇ 2 R + 2 D ( e D ∇ R t + ∇ R e D t ) + 2 R e D e D t . \nabla^2(D^2R)=D^2\nabla^2R+2D(e_D\nabla R^t+\nabla R e_D^t)+2Re_De_D^t. ∇ 2 ( D 2 R ) = D 2 ∇ 2 R + 2 D ( e D ∇ R t + ∇ R e D t ) + 2 R e D e D t .
The analogous formula for D R DR D R contains D ∇ 2 R + e D ∇ R t + ∇ R e D t D\nabla^2R+e_D\nabla R^t+\nabla R e_D^t D ∇ 2 R + e D ∇ R t + ∇ R e D t . Setting R = 0 R=0 R = 0 before differentiation would lose necessary terms.
For completeness, the integral means have controlled derivatives on the full auxiliary boxes. With g ( x , w ) = ( ( x − w ) / ( x + w ) ) 2 g(x,w)=((x-w)/(x+w))^2 g ( x , w ) = (( x − w ) / ( x + w ) ) 2 ,
g x x = 8 w ( 2 w − x ) ( x + w ) 4 , g x w = 4 ( x 2 − 4 x w + w 2 ) ( x + w ) 4 , g w w = 8 x ( 2 x − w ) ( x + w ) 4 . g_{xx}=\frac{8w(2w-x)}{(x+w)^4},\qquad g_{xw}=\frac{4(x^2-4xw+w^2)}{(x+w)^4},\qquad g_{ww}=\frac{8x(2x-w)}{(x+w)^4}. g xx = ( x + w ) 4 8 w ( 2 w − x ) , g x w = ( x + w ) 4 4 ( x 2 − 4 x w + w 2 ) , g w w = ( x + w ) 4 8 x ( 2 x − w ) .
On a compact positive box all these functions are continuous and uniformly bounded for 0 ≤ t ≤ 1 0\leq t\leq1 0 ≤ t ≤ 1 . Differentiating (36.3) under the integral gives the six Hessian entries as integrals of
g x x , ( 1 − t ) g x w , t g x w , ( 1 − t ) 2 g w w , t ( 1 − t ) g w w , t 2 g w w . g_{xx},\quad(1-t)g_{xw},\quad tg_{xw},\quad(1-t)^2g_{ww},\quad t(1-t)g_{ww},\quad t^2g_{ww}. g xx , ( 1 − t ) g x w , t g x w , ( 1 − t ) 2 g w w , t ( 1 − t ) g w w , t 2 g w w .
Enclosing the affine w = ( 1 − t ) k + t z w=(1-t)k+tz w = ( 1 − t ) k + t z on each of sixteen closed equal subintervals, and integrating these nonnegative weights exactly, gives full-box enclosures. This is interval inclusion, not quadrature. It divides by neither z − k z-k z − k nor an imbalance. Analytic smoothness at coincident or reversed positive endpoints grants no additional physical return or boundary stationarity qualification.
Necessary contractions and exact rational separation
Fix one complete original cell. The accepted value constructor starts from its full five-dimensional endpoints, the selected calibration branch, complete frame, actual means and original source bounds. It either proves a necessary contradiction or encloses the same candidate in a positive box X ⊂ Ω 05 X\subset\Omega_{05} X ⊂ Ω 05 . In the latter case it initially retains the entire parameter box for θ = ( s , e , H ) \theta=(s,e,H) θ = ( s , e , H ) . A later closed chart split restricts this box only by covering it with both closed children. A subsequent contraction always intersects the old box. It cannot replace one candidate’s mean, root or owner by a separately favorable choice.
Here is the inclusion principle used for the first-order contractions. For a residual T T T and fixed θ \theta θ , with center m ∈ X m\in X m ∈ X ,
0 = T ( z ; θ ) = T ( m ; θ ) + ∑ j ( ∫ 0 1 ∂ j T ( m + t ( z − m ) ; θ ) d t ) ( z j − m j ) . 0=T(z;\theta)=T(m;\theta)+\sum_j\left(\int_0^1\partial_jT(m+t(z-m);\theta)\,dt\right)(z_j-m_j). 0 = T ( z ; θ ) = T ( m ; θ ) + j ∑ ( ∫ 0 1 ∂ j T ( m + t ( z − m ) ; θ ) d t ) ( z j − m j ) .
at every candidate zero. Whole-box derivative intervals therefore give necessary linear interval rows. If a coefficient omits zero, isolating that coordinate and intersecting its existing interval retains the same zero. Sequential use remains valid because each previous step has preserved it. A sign-separated value residual or a strictly empty necessary intersection gives a contradiction. Failure to certify a denominator, a disjoint pair of purported enclosures of the same function, or an unresolved calculation is not an exclusion.
For the stronger rows, twice applying the fundamental theorem of calculus gives
T ( z ; θ ) = T ( m ; θ ) + ∇ T ( m ; θ ) ⋅ ξ + ∫ 0 1 ( 1 − t ) ξ t ∇ 2 T ( m + t ξ ; θ ) ξ d t , ξ = z − m . (186) T(z;\theta)=T(m;\theta)+\nabla T(m;\theta)\cdot\xi+\int_0^1(1-t)\xi^t\nabla^2T(m+t\xi;\theta)\xi\,dt,\qquad\xi=z-m. \tag*{(186)} T ( z ; θ ) = T ( m ; θ ) + ∇ T ( m ; θ ) ⋅ ξ + ∫ 0 1 ( 1 − t ) ξ t ∇ 2 T ( m + t ξ ; θ ) ξ d t , ξ = z − m . (186)
The line segment remains in X X X , and every parameter value in the current retained parameter box is quantified; initially this is the full original parameter box. If H i j H_{ij} H ij encloses each Hessian entry throughout this product domain, the remainder is enclosed by
R = 1 2 ∑ i H i i [ ξ i 2 ] + ∑ i < j H i j [ ξ i ] [ ξ j ] . (187) \mathcal{R}=\frac{1}{2}\sum_i H_{ii}[\xi_i^2]+\sum_{i<j}H_{ij}[\xi_i][\xi_j]. \tag*{(187)} R = 2 1 i ∑ H ii [ ξ i 2 ] + i < j ∑ H ij [ ξ i ] [ ξ j ] . (187)
The square interval in the diagonal term has lower endpoint zero. The factor 1 / 2 1/2 1/2 on the diagonal and its cancellation on the symmetric off-diagonal terms follow directly from the integral in (186) ; no common intermediate point for all entries is asserted.
Put ρ j = max ( m j − inf X j , sup X j − m j ) \rho_j = \max(m_j-\inf X_j,\sup X_j-m_j) ρ j = max ( m j − inf X j , sup X j − m j ) and ξ j = ρ j u j \xi_j=\rho_j u_j ξ j = ρ j u j , ∣ u j ∣ ≤ 1 |u_j|\le1 ∣ u j ∣ ≤ 1 . A zero-radius coordinate uses u j = 0 u_j=0 u j = 0 , without division. Suppose T 0 \mathbb{T}_0 T 0 encloses T ( m ; θ ) T(m;\theta) T ( m ; θ ) , J j \mathbb{J}_j J j encloses ∂ j T ( m ; θ ) \partial_jT(m;\theta) ∂ j T ( m ; θ ) , and write
A j = ρ j mid J j , E J = ∑ j ρ j rad J j . A_j=\rho_j\operatorname{mid}\mathbb{J}_j,\qquad E_J=\sum_j\rho_j\operatorname{rad}\mathbb{J}_j. A j = ρ j mid J j , E J = j ∑ ρ j rad J j .
Then a common zero satisfies the two rational inequalities
A ⋅ u ≤ − inf T 0 − inf R + E J , − A ⋅ u ≤ sup T 0 + sup R + E J . (188) A\cdot u\le-\inf\mathbb{T}_0-\inf\mathbb{R}+E_J,\qquad-A\cdot u\le\sup\mathbb{T}_0+\sup\mathbb{R}+E_J. \tag*{(188)} A ⋅ u ≤ − inf T 0 − inf R + E J , − A ⋅ u ≤ sup T 0 + sup R + E J . (188)
The same construction with whole-box first derivatives and zero explicit remainder gives the first-order necessary rows. Eighteen equalities supply thirty-six such rows, and the original cost condition C 05 − C ∗ ≤ 0 C_{05}-C_*\le0 C 05 − C ∗ ≤ 0 supplies one further row, used once. Adding the two second-order rows for each of eighteen or twenty-four equalities gives respectively 73 or 85 rows. These are linear necessary inequalities at a candidate, rather than a positive-semidefiniteness requirement on the Hessian of its physical cost.
Lemma 36.4 (Rational support and coordinate contraction). Suppose the necessary rows are a j ⋅ u ≤ b j a_j\cdot u\le b_j a j ⋅ u ≤ b j , with ∣ u i ∣ ≤ 1 |u_i|\le1 ∣ u i ∣ ≤ 1 . For nonnegative rational weights λ j \lambda_j λ j , put a = ∑ j λ j a j a=\sum_j\lambda_j a_j a = ∑ j λ j a j and b = ∑ j λ j b j b=\sum_j\lambda_j b_j b = ∑ j λ j b j . If
− ∑ i ∣ a i ∣ − b > 0 , (189) -\sum_i|a_i|-b>0, \tag*{(189)} − i ∑ ∣ a i ∣ − b > 0 , (189) there is no candidate in the box. Otherwise, if a k > 0 a_k>0 a k > 0 ,
u k ≤ b + ∑ i ≠ k ∣ a i ∣ a k ; u_k\le\frac{b+\sum_{i\ne k}|a_i|}{a_k}; u k ≤ a k b + ∑ i = k ∣ a i ∣ ; if a k < 0 a_k<0 a k < 0 the same quotient is a lower bound.
Proof. Nonnegative combination gives a ⋅ u ≤ b a\cdot u\le b a ⋅ u ≤ b , whereas a ⋅ u ≥ − ∑ i ∣ a i ∣ a\cdot u\ge-\sum_i|a_i| a ⋅ u ≥ − ∑ i ∣ a i ∣ . This proves the strict contradiction. Isolating a k u k a_ku_k a k u k and using − ∑ i ≠ k a i u i ≤ ∑ i ≠ k ∣ a i ∣ -\sum_{i\ne k}a_i u_i\le\sum_{i\ne k}|a_i| − ∑ i = k a i u i ≤ ∑ i = k ∣ a i ∣ gives the contraction, with the inequality reversed when dividing by a negative coefficient. □
Every adopted support recomputes its nonlinear values, derivative coefficients, nonnegative rational weights and strict margin. Floating-point optimization proposes weights only. Rational arithmetic checks their actual use; outward floor and ceiling round the reconstructed coordinate endpoints. A zero margin is retained. The nonlinear residuals may exclude all auxiliary zeros in a box, but cost and known-domain filters exclude only the qualified physical subset to which their theorems apply. The proof uses the latter, narrower statement.
Arithmetic and the complete original-cell cover
The final original-cell arithmetic uses arbitrary-precision integers and intervals with endpoints on the scale 2 − 128 2^{-128} 2 − 128 . The earlier whole-sector half-F F F theorem has its own 2 − 160 2^{-160} 2 − 160 interval implementation and 96-term logarithmic bound; it is an input to the overlay and is not re-evaluated by the final-cell kernel. Rational inputs, multiplication and division are rounded outward; square roots use integer square roots with an upward correction when needed. Positive logarithms are reduced by exact powers of two to [ 1 , 2 ] [1,2] [ 1 , 2 ] . For z = ( x − 1 ) / ( x + 1 ) z=(x-1)/(x+1) z = ( x − 1 ) / ( x + 1 ) the eighty-term expansion
log x = 2 ∑ j = 0 79 z 2 j + 1 2 j + 1 + R , ∣ R ∣ ≤ 2 ∣ z ∣ 161 161 ( 1 − ∣ z ∣ 2 ) \log x=2\sum_{j=0}^{79}\frac{z^{2j+1}}{2j+1}+R,\qquad|R|\le\frac{2|z|^{161}}{161(1-|z|^2)} log x = 2 j = 0 ∑ 79 2 j + 1 z 2 j + 1 + R , ∣ R ∣ ≤ 161 ( 1 − ∣ z ∣ 2 ) 2∣ z ∣ 161
encloses the complete tail, with the endpoint range of z z z itself rounded outward. The same method encloses log 2 \log2 log 2 . The calibrated roots and actual means retain their separately proved convergent-series tails and positive-root interval bounds; all operations preserve the same actual values.
One can also enclose the complete mean and its first derivatives by writing
q = z − k 2 x + k + z , r = 2 x − k − z 2 x + k + z , M ( x ; k , z ) = r 2 + ∑ n ≥ 1 q 2 n [ r 2 + 4 n r 2 n + 1 + 2 n − 1 2 n + 1 ] . q=\frac{z-k}{2x+k+z},\qquad r=\frac{2x-k-z}{2x+k+z},\qquad M(x;k,z)=r^2+\sum_{n\geq1}q^{2n}\left[r^2+\frac{4nr}{2n+1}+\frac{2n-1}{2n+1}\right]. q = 2 x + k + z z − k , r = 2 x + k + z 2 x − k − z , M ( x ; k , z ) = r 2 + n ≥ 1 ∑ q 2 n [ r 2 + 2 n + 1 4 n r + 2 n + 1 2 n − 1 ] .
This follows from the uniformly convergent expansion of ( 1 + t q ) − 2 (1+tq)^{-2} ( 1 + tq ) − 2 on [ − 1 , 1 ] [-1,1] [ − 1 , 1 ] . If ∣ q ∣ ≤ v < 1 |q|\leq v<1 ∣ q ∣ ≤ v < 1 and ∣ r ∣ ≤ R 0 |r|\leq R_0 ∣ r ∣ ≤ R 0 , the tails after n = 40 n=40 n = 40 of the value, r r r derivative and q q q derivative are bounded respectively by
( 1 + R 0 ) 2 v 82 1 − v 2 , 2 ( 1 + R 0 ) v 82 1 − v 2 , 2 ( 1 + R 0 ) 2 v 81 ( 41 − 40 v 2 ) ( 1 − v 2 ) 2 . \frac{(1+R_0)^2v^{82}}{1-v^2},\qquad\frac{2(1+R_0)v^{82}}{1-v^2},\qquad\frac{2(1+R_0)^2v^{81}(41-40v^2)}{(1-v^2)^2}. 1 − v 2 ( 1 + R 0 ) 2 v 82 , 1 − v 2 2 ( 1 + R 0 ) v 82 , ( 1 − v 2 ) 2 2 ( 1 + R 0 ) 2 v 81 ( 41 − 40 v 2 ) .
They can be intersected with the independent compact-integral enclosures because both contain the same functions. An empty intersection between two such enclosures signals an unsafe evaluation, not absence of a physical candidate.
The fixed original-cell list has exact identities consisting of its seed, continuation address, five-dimensional endpoints and original record. Their fixed evaluation profiles partition these identities as follows:
Table 16 (PDF p. 120).
Each original identity occurs once in this partition. The final verification reconstructs each complete original record and its mathematical contradiction using the appropriate accepted equations and arithmetic. Fifteen cells use separate complete-original support plans: thirteen use their full nonlinear calculations, and two use complete closed internal covers. Each such plan is checked on its entire original cell; an inherited acceptance label is not a mathematical terminal.
Internal subdivision is allowed only as a full cover of the current enclosing box: an interval [ l , h ] [l,h] [ l , h ] is replaced by [ l , m ] [l,m] [ l , m ] and [ m , h ] [m,h] [ m , h ] at its exact stored midpoint. For a native-coordinate split, both children retain the entire current parameter box. For a chart-coordinate split, both retain the entire current native enclosure, while their closed parameter intervals cover the current parameter box. If both descendants are excluded, so is the parent. Otherwise the original cell remains unresolved. Depth limits, failed guards, unknown labels, unsafe arithmetic and missing branches never count as success. Internal leaf counts are not added to the number of original cells.
Lemma 36.5 (Exhaustion of the R05 original cells). Under the source implications of Proposition 36.3 , the union of the fixed 46 383 original cells contains no original R05 minimizing candidate.
Proof. The verified original-identity index contains exactly the indices 0 , … , 46382 0,\ldots,46382 0 , … , 46382 , each with its complete original record, and no unresolved original. Every such cell has either a full-cell strict necessary contradiction or a complete internal closed cover whose every terminal has such a contradiction. The value construction encloses the lift of every qualified candidate in that cell. The contractions in § 6.6 preserve the same candidate; each internal closed split preserves coverage. Finite induction along its recorded tree would therefore place an existing candidate in a terminal where a strict contradiction holds, which is impossible.
The final acceptance also binds the original source and expression versions, the full ordered inputs and their mathematical receipts. Its evaluation-profile counts agree with the disjoint original-identity partition above; the check is on complete identities and records, not just their numbers. The accepted source-to-domain implication remains a premise of this lemma, rather than a conclusion drawn from the acceptance index. □
Proof of Theorem 36.1. Suppose an original R05 global minimizer exists. Its finite essential arcs are positive, with the genuine twist and original cost budget of Section 2. Lemma 36.2 supplies the required complete sources and one normal jointly normalized full lift, including both state-covector twists. Its actual states, caps, selected root jets, three means and real winding therefore have precisely the qualifications of Proposition 36.3. That proposition places the same candidate in one complete original cell, unless an earlier source comparison has already contradicted it. Lemma 36.5 excludes the former possibility as well. Thus no such original global minimizer exists. □
The argument concerns strict essential R05 arcs. Closed numerical boxes add harmless auxiliary boundary points, while any physical boundary use is restricted to the qualified limits in the source theorems. It grants no general stationary or Hessian condition at zero physical duration, D = 0 D=0 D = 0 , a lost selected root, or a formal tuple without the original full lift and winding. The independent short octagon class is unaffected.
The nine-arc type R06
We now exclude the second nine-arc type needed here. Its middle shared arc has the opposite cap orientation from the two exterior shared arcs. The proof retains that distinction and couples all three caps through one original full lift.
Theorem 37.1 (Exclusion of R06). An original global minimizer cannot have nine essential positive arcs with native labels 021202120 and increments − − + + − − + + + - - + + - - + + + − − + + − − + + + , with the genuine extension j i + 9 = j i + 1 j_{i+9}=j_i+1 j i + 9 = j i + 1 .
The argument has two parts. We first place every such minimizing trajectory in one fixed five-dimensional root, keeping its full signed endpoints, maximizing lift and original cost. We then exclude its necessary values on a complete closed cover. The mathematical reduction is given below and in Appendix F; the independent finite verification is specified in Section H.8.
The actual source and its positive durations
Use the reciprocal state coordinates and label assignment of Section 36.1. Write the native own reciprocals as
( r 0 , … , r 8 ) = ( d , e , b , m , D , E , B , f , c ) . (190) (r_0,\ldots,r_8)=(d,e,b,m,D,E,B,f,c). \tag*{(190)} ( r 0 , … , r 8 ) = ( d , e , b , m , D , E , B , f , c ) . (190)
Within this section and its mathematical appendix, D D D and B B B are ordinary reciprocal coordinates; in particular this B B B is distinct from the essential-arc count. Put
h ( t ) = 1 + t 2 2 , h 1 = h ( e ) , h 2 = h ( E ) , H = h ( m ) . h(t)=\frac{1+t^2}{2},\qquad h_1=h(e),\qquad h_2=h(E),\qquad H=h(m). h ( t ) = 2 1 + t 2 , h 1 = h ( e ) , h 2 = h ( E ) , H = h ( m ) .
Let S 06 \mathcal{S}_{06} S 06 denote the original global-minimizing trajectories of this type, together with their actual labelled reciprocal data on a genuine period. Thus the nine maximal arcs are positive, the state and group endpoints satisfy the original signed twist, and the real winding and budget of Section 2 are retained. The auxiliary equations introduced below do not define this source class.
For an arc with own reciprocal x x x , predecessor p p p , successor n n n and adjacent increments α , β ∈ { − 1 , 1 } \alpha,\beta\in\{-1,1\} α , β ∈ { − 1 , 1 } , the physical multiplier is
Q = ( x + p ) α ( x + n ) β ( 1 + x 2 ) ( α + β ) / 2 , τ = x log Q > 0. Q=\frac{(x+p)^\alpha(x+n)^\beta}{(1+x^2)^{(\alpha+\beta)/2}},\qquad\tau=x\log Q>0. Q = ( 1 + x 2 ) ( α + β ) /2 ( x + p ) α ( x + n ) β , τ = x log Q > 0.
The same formula applies to negative increments: the physical time is not reversed. Substitution of the R06 word gives
Q d = d + c d + e , Q e = 1 + e 2 ( e + d ) ( e + b ) , Q b = b + m b + e , Q m = ( m + b ) ( m + D ) 1 + m 2 , Q D = D + m D + E , Q E = 1 + E 2 ( E + D ) ( E + B ) , Q B = B + f B + E , Q f = ( f + B ) ( f + c ) 1 + f 2 , Q c = ( c + f ) ( c + d ) 1 + c 2 . (191) \begin{aligned}
Q_d &= \frac{d+c}{d+e}, & Q_e &= \frac{1+e^2}{(e+d)(e+b)}, & Q_b &= \frac{b+m}{b+e},\\
Q_m &= \frac{(m+b)(m+D)}{1+m^2}, & Q_D &= \frac{D+m}{D+E}, & Q_E &= \frac{1+E^2}{(E+D)(E+B)},\\
Q_B &= \frac{B+f}{B+E}, & Q_f &= \frac{(f+B)(f+c)}{1+f^2}, & Q_c &= \frac{(c+f)(c+d)}{1+c^2}. \tag*{(191)}
\end{aligned} Q d Q m Q B = d + e d + c , = 1 + m 2 ( m + b ) ( m + D ) , = B + E B + f , Q e Q D Q f = ( e + d ) ( e + b ) 1 + e 2 , = D + E D + m , = 1 + f 2 ( f + B ) ( f + c ) , Q b Q E Q c = b + e b + m , = ( E + D ) ( E + B ) 1 + E 2 , = 1 + c 2 ( c + f ) ( c + d ) . (191)
All nine are greater than one. In particular
c > e , m > e , m > E , f > E . (192) c>e,\qquad m>e,\qquad m>E,\qquad f>E. \tag*{(192)} c > e , m > e , m > E , f > E . (192)
The actual third reciprocal at the c , f c,f c , f switch is
s = 1 − c f c + f > 0 , c f + s ( c + f ) = 1. (193) s=\frac{1-cf}{c+f}>0,\qquad cf+s(c+f)=1. \tag*{(193)} s = c + f 1 − c f > 0 , c f + s ( c + f ) = 1. (193)
Since Q c = ( c + d ) / ( c + s ) Q_c=(c+d)/(c+s) Q c = ( c + d ) / ( c + s ) and Q f = ( f + B ) / ( f + s ) Q_f=(f+B)/(f+s) Q f = ( f + B ) / ( f + s ) , also d > s d>s d > s and B > s B>s B > s . Thus both actual rectangle areas
A = ( c − e ) ( d − s ) , Z = ( f − E ) ( B − s ) (194) A=(c-e)(d-s),\qquad Z=(f-E)(B-s) \tag*{(194)} A = ( c − e ) ( d − s ) , Z = ( f − E ) ( B − s ) (194)
are positive. These conclusions use positive physical arcs, without a boundary-face exclusion or an assumed numerical collar.
Lemma 37.2 (Qualification on one original period). Every member of S 06 \mathcal{S}_{06} S 06 has one jointly normalized original full maximizing lift on native [ − 1 , 7 ] [-1,7] [ − 1 , 7 ] . Its matrix and both state covectors retain the genuine endpoint relations. In a common whole phase + 1 +1 + 1 , its three five-arc supports are
[ − 1 , 3 ] : 01020 , [ 1 , 5 ] : 02010 , [ 3 , 7 ] : 01020 , [-1,3]:01020,\qquad[1,5]:02010,\qquad[3,7]:01020, [ − 1 , 3 ] : 01020 , [ 1 , 5 ] : 02010 , [ 3 , 7 ] : 01020 , with shared reciprocals e e e , m m m , E E E , respectively. All actual own and adjacent third reciprocals lie in [ 1 / 15 , 5 ] [1/15,5] [ 1/15 , 5 ] , and e 2 , E 2 < 1 / 6 e^2,E^2<1/6 e 2 , E 2 < 1/6 .
Proof. The interval [ − 1 , 7 ] [-1,7] [ − 1 , 7 ] has literal word 202120212 and contains all nine complete physical arcs once. Native [ 0 , 4 ] = 02120 [0,4]=02120 [ 0 , 4 ] = 02120 becomes 10201 under whole phase + 1 +1 + 1 ; its binary core 020 has two distinct interior switches and both actual third-mode caps. The homogeneous argument of Lemma B.3 first proves normality of a full maximizing lift selected on this very period. Dividing its objective multiplier, matrix and both state covectors by the same positive number gives the stated normalization. Appendix F proves the free-state endpoint row for that same lift; no covector is copied across a seam.
The reversed support [ 4 , 3 , 2 , 1 , 0 , − 1 ] [4,3,2,1,0,-1] [ 4 , 3 , 2 , 1 , 0 , − 1 ] with whole reflected label map j ↦ 2 − j j\mapsto2-j j ↦ 2 − j , and the forward support [ 2 , 7 ] [2,7] [ 2 , 7 ] with whole phase + 1 +1 + 1 , both become the proper six-source 201020. Their physical shared centers are e e e and E E E . Theorem 7.4 gives the two density bounds. Only physical state consequences are exported from these recuts. The compactness theorem 2.1, transported by the lawful whole phases, bounds every actual reciprocal, including the third component s s s ; its elementary bound 400 / 99 < 5 400/99<5 400/99 < 5 gives the stated convenient interval. The middle support remains inner-sharing, so no exterior density bound is imposed on m m m . □
The complete frame and a common cap system
Define
R = ( m − e ) ( b + m + e 2 ) , L = ( m − E ) ( D + m + E 2 ) , F = H 2 − R L , G = 2 h 1 H h 2 . (195) R=(m-e)\left(b+\frac{m+e}{2}\right),\qquad L=(m-E)\left(D+\frac{m+E}{2}\right),\qquad F=H^2-RL,\qquad G=2h_1Hh_2. \tag*{(195)} R = ( m − e ) ( b + 2 m + e ) , L = ( m − E ) ( D + 2 m + E ) , F = H 2 − R L , G = 2 h 1 H h 2 . (195)
The original full signed matrix equation gives
F ( 1 + s 2 ) = G , A = ( s + e ) 2 2 − h 1 2 L F , Z = ( s + E ) 2 2 − h 2 2 R F . (196) F(1+s^2)=G,\qquad A=\frac{(s+e)^2}{2}-\frac{h_1^2L}{F},\qquad Z=\frac{(s+E)^2}{2}-\frac{h_2^2R}{F}. \tag*{(196)} F ( 1 + s 2 ) = G , A = 2 ( s + e ) 2 − F h 1 2 L , Z = 2 ( s + E ) 2 − F h 2 2 R . (196)
Appendix F derives these identities from all four signed frame entries. In particular their use retains the original sign of the frame.
Use the fixed-interval mean M M M of (170) , and set
μ 0 = M ( d ; e , c ) , μ 1 = M ( b ; e , m ) , μ 2 = M ( D ; E , m ) , μ 3 = M ( B ; E , f ) , K i = 3 + μ i 2 . (197) \mu_0=M(d;e,c),\qquad\mu_1=M(b;e,m),
\qquad\mu_2=M(D;E,m),\qquad\mu_3=M(B;E,f),\qquad K_i=\frac{3+\mu_i}{2}. \tag*{(197)} μ 0 = M ( d ; e , c ) , μ 1 = M ( b ; e , m ) , μ 2 = M ( D ; E , m ) , μ 3 = M ( B ; E , f ) , K i = 2 3 + μ i . (197)
These are four different integrals at one candidate. For positive t , x , y t,x,y t , x , y define the bridge function
B ( t ; x , y ; μ , ν ) = 1 2 + h ( t ) log ( t + x ) ( t + y ) 1 + t 2 + h ( t ) 2 ( p + q + μ − 1 p + ν − 1 q ) , p = 2 t t + x , q = 2 t t + y . (198) \mathcal{B}(t;x,y;\mu,\nu)=\frac{1}{2}+h(t)\log\frac{(t+x)(t+y)}{1+t^2}
+\frac{h(t)}{2}\left(p+q+\frac{\mu-1}{p}+\frac{\nu-1}{q}\right),\qquad
p=\frac{2t}{t+x},\quad q=\frac{2t}{t+y}. \tag*{(198)} B ( t ; x , y ; μ , ν ) = 2 1 + h ( t ) log 1 + t 2 ( t + x ) ( t + y ) + 2 h ( t ) ( p + q + p μ − 1 + q ν − 1 ) , p = t + x 2 t , q = t + y 2 t . (198)
Write
B e = B ( e ; d , b ; μ 0 , μ 1 ) , B m = B ( m ; b , D ; μ 1 , μ 2 ) , B E = B ( E ; D , B ; μ 2 , μ 3 ) . B_e=\mathcal{B}(e;d,b;\mu_0,\mu_1),\qquad
B_m=\mathcal{B}(m;b,D;\mu_1,\mu_2),\qquad
B_E=\mathcal{B}(E;D,B;\mu_2,\mu_3). B e = B ( e ; d , b ; μ 0 , μ 1 ) , B m = B ( m ; b , D ; μ 1 , μ 2 ) , B E = B ( E ; D , B ; μ 2 , μ 3 ) .
The three restrictions of the same full lift have common determinant and the two unsquared overlap equations
B e − B m + R K 1 = 0 , B E − B m + L K 2 = 0. (199) B_e-B_m+RK_1=0,\qquad B_E-B_m+LK_2=0. \tag*{(199)} B e − B m + R K 1 = 0 , B E − B m + L K 2 = 0. (199)
Eliminating the common determinants, using only K 1 , K 2 > 0 K_1,K_2>0 K 1 , K 2 > 0 , gives
h 1 2 K 0 − R 2 K 1 − H 2 K 2 + 2 R B m = 0 , h 2 2 K 3 − H 2 K 1 − L 2 K 2 + 2 L B m = 0. (200) h_1^2K_0-R^2K_1-H^2K_2+2RB_m=0,\qquad
h_2^2K_3-H^2K_1-L^2K_2+2LB_m=0. \tag*{(200)} h 1 2 K 0 − R 2 K 1 − H 2 K 2 + 2 R B m = 0 , h 2 2 K 3 − H 2 K 1 − L 2 K 2 + 2 L B m = 0. (200)
The unsquared equations are retained. Equality of three determinant values alone would not identify the matrices or either state covector.
The enclosing root and its fourteen-variable system
Proposition 37.3 (Source-to-root reduction). The projection ( e , E , b , D , m ) (e,E,b,D,m) ( e , E , b , D , m ) of every member of S 06 \mathcal{S}_{06} S 06 belongs to
B 06 = [ 1 16 , 27 64 ] 2 × [ 1 16 , 5 ] 2 × [ 3 8 , 1 ] . (201) B_{06}=\left[\frac{1}{16},\frac{27}{64}\right]^2\times\left[\frac{1}{16},5\right]^2\times\left[\frac{3}{8},1\right]. \tag*{(201)} B 06 = [ 16 1 , 64 27 ] 2 × [ 16 1 , 5 ] 2 × [ 8 3 , 1 ] . (201) At the same candidate,
21 50 < m < 471 500 , b > m + e 2 , D > m + E 2 , s < m , F > h 1 h 2 . (202) \frac{21}{50}<m<\frac{471}{500},\qquad b>\frac{m+e}{2},\qquad D>\frac{m+E}{2},\qquad s<m,\qquad F>h_1h_2. \tag*{(202)} 50 21 < m < 500 471 , b > 2 m + e , D > 2 m + E , s < m , F > h 1 h 2 . (202) The complete original minimizing and full-lift qualifications remain in force.
Proof. The reserve identity in Lemma F.2 gives s < m s<m s < m and F > h 1 h 2 F>h_1h_2 F > h 1 h 2 directly on the strict source. The actual return inequalities give the two owner quadratics in Lemma F.1. The analytic Lemma F.3 then proves m < 1 m<1 m < 1 . It uses only these quadratics, the compact and density bounds, and the strict frame reserve. The midpoint comparison and the lower and upper middle estimates of Appendix F give the remaining assertions. This order is independent of any unit-sector coefficient certificate or boundary exclusion.
Finally [ 1 / 15 , 1 / 6 ] ⊂ [ 1 / 16 , 27 / 64 ] [1/15,1/\sqrt{6}]\subset[1/16,27/64] [ 1/15 , 1/ 6 ] ⊂ [ 1/16 , 27/64 ] , [ 1 / 15 , 5 ] ⊂ [ 1 / 16 , 5 ] [1/15,5]\subset[1/16,5] [ 1/15 , 5 ] ⊂ [ 1/16 , 5 ] and [ 21 / 50 , 471 / 500 ] ⊂ [ 3 / 8 , 1 ] [21/50,471/500]\subset[3/8,1] [ 21/50 , 471/500 ] ⊂ [ 3/8 , 1 ] . Replacing strict source bounds by these closed outer intervals proves the projection claim without selecting an owner ordering. □
For the finite verification use fourteen coordinates
z = ( e , E , b , D , m , s , c , d , B , f , u 0 , u 1 , u 2 , u 3 ) ∈ Ω 06 : = ( 0 , ∞ ) 10 × R 4 . (203) z=(e,E,b,D,m,s,c,d,B,f,u_0,u_1,u_2,u_3)\in\Omega_{06}:=(0,\infty)^{10}\times\mathbb{R}^4. \tag*{(203)} z = ( e , E , b , D , m , s , c , d , B , f , u 0 , u 1 , u 2 , u 3 ) ∈ Ω 06 := ( 0 , ∞ ) 10 × R 4 . (203)
At a physical candidate u i = μ i u_i=\mu_i u i = μ i ; the auxiliary mean coordinates are unrestricted real variables in this smooth extension. Put F ^ = G / ( 1 + s 2 ) \widehat{F}=G/(1+s^2) F = G / ( 1 + s 2 ) and use K i = ( 3 + u i ) / 2 K_i=(3+u_i)/2 K i = ( 3 + u i ) /2 in the bridges. Define twelve smooth residuals by
f 0 = ( H 2 − R L ) ( 1 + s 2 ) − G , f 1 = ( c − e ) ( d − s ) − ( s + e ) 2 / 2 + h 1 2 L / F ^ , f 2 = ( f − E ) ( B − s ) − ( s + E ) 2 / 2 + h 2 2 R / F ^ , f 3 = c f + s ( c + f ) − 1 , f 4 = h 1 2 K 0 − R 2 K 1 − H 2 K 2 + 2 R B m , f 5 = h 2 2 K 3 − H 2 K 1 − L 2 K 2 + 2 L B m , f 6 = B e − B m + R K 1 , f 7 = B E − B m + L K 2 , f 8 = M ( d ; e , c ) − u 0 , f 9 = M ( b ; e , m ) − u 1 , f 10 = M ( D ; E , m ) − u 2 , f 11 = M ( B ; E , f ) − u 3 . (204) \begin{aligned}
f_0 &= (H^2-RL)(1+s^2)-G,\\
f_1 &= (c-e)(d-s)-(s+e)^2/2+h_1^2L/\widehat{F},\\
f_2 &= (f-E)(B-s)-(s+E)^2/2+h_2^2R/\widehat{F},\\
f_3 &= cf+s(c+f)-1,\\
f_4 &= h_1^2K_0-R^2K_1-H^2K_2+2RB_m,\\
f_5 &= h_2^2K_3-H^2K_1-L^2K_2+2LB_m,\\
f_6 &= B_e-B_m+RK_1,\qquad f_7=BE-B_m+LK_2,\\
f_8 &= M(d;e,c)-u_0,\qquad f_9=M(b;e,m)-u_1,\\
f_{10} &= M(D;E,m)-u_2,\qquad f_{11}=M(B;E,f)-u_3. \tag*{(204)}
\end{aligned} f 0 f 1 f 2 f 3 f 4 f 5 f 6 f 8 f 10 = ( H 2 − R L ) ( 1 + s 2 ) − G , = ( c − e ) ( d − s ) − ( s + e ) 2 /2 + h 1 2 L / F , = ( f − E ) ( B − s ) − ( s + E ) 2 /2 + h 2 2 R / F , = c f + s ( c + f ) − 1 , = h 1 2 K 0 − R 2 K 1 − H 2 K 2 + 2 R B m , = h 2 2 K 3 − H 2 K 1 − L 2 K 2 + 2 L B m , = B e − B m + R K 1 , f 7 = B E − B m + L K 2 , = M ( d ; e , c ) − u 0 , f 9 = M ( b ; e , m ) − u 1 , = M ( D ; E , m ) − u 2 , f 11 = M ( B ; E , f ) − u 3 . (204)
Every source point satisfies f i = 0 f_i=0 f i = 0 . The use of F ^ \widehat{F} F , rather than substitution of H 2 − R L H^2-RL H 2 − R L off the frame, fixes the ambient functions whose derivatives and Hessians are enclosed. The mean in (170) is smooth even when its two endpoints coincide. Bounds valid only for actual returned means may intersect their values, but are not differentiated as ambient formulas.
With the potential ϕ \phi ϕ in (180) , the original cost is
C 06 = − ϕ ( d , e ) − ϕ ( e , b ) + ϕ ( b , m ) + ϕ ( m , D ) − ϕ ( D , E ) − ϕ ( E , B ) + ϕ ( B , f ) + ϕ ( f , c ) + ϕ ( c , d ) . (205) \begin{aligned}
C_{06}={}&-\phi(d,e)-\phi(e,b)+\phi(b,m)+\phi(m,D)-\phi(D,E)\\
&-\phi(E,B)+\phi(B,f)+\phi(f,c)+\phi(c,d). \tag*{(205)}
\end{aligned} C 06 = − ϕ ( d , e ) − ϕ ( e , b ) + ϕ ( b , m ) + ϕ ( m , D ) − ϕ ( D , E ) − ϕ ( E , B ) + ϕ ( B , f ) + ϕ ( f , c ) + ϕ ( c , d ) . (205)
The closing edge is included once. Set
J 8 : = C o c t = 2 ( 8 − 4 2 − log 2 ) 2 2 − 1 , J_8:=C_{\mathrm{oct}}=\frac{2(8-4\sqrt{2}-\log2)}{2\sqrt{2}-1}, J 8 := C oct = 2 2 − 1 2 ( 8 − 4 2 − log 2 ) ,
as in (18) . The thirteenth row is the one-sided inequality C 06 − J 8 ≤ 0 C_{06}-J_8\le0 C 06 − J 8 ≤ 0 . There are thus twenty-four signed equality inequalities and one cost inequality, not two directions for the cost row.
Exhaustion and conclusion
Theorem 37.4 (Complete finite exclusion). The fixed closed binary partition of B 06 B_{06} B 06 has 806 875 original leaves. Every original leaf is excluded for a point of S 06 \mathcal{S}_{06} S 06 by the necessary systems and candidate-preserving comparisons above and in Appendix F. The independently assembled cover has no retained or unvisited original leaf.
Proof. The exact midpoint parser reconstructs every original interval from the single root (201) , with both closed children retained at each split. In a surviving leaf the full enclosing c c c interval is divided into eight closed pieces, and the actual mean values lift each surviving piece to its complete fourteen-dimensional box. All contractions retain the same possible source point. A leaf is excluded only by a necessary strict sign, an empty necessary intersection, or a fully checked affine contradiction with a valid whole-box Taylor remainder. Further subdivisions likewise retain both complete closed children.
Section H.8 gives the arithmetic, row construction and exact acceptance rules, together with the reader entry points and immutable input identities. The final assembly checks every original identifier and endpoint against the reconstructed tree. The 34 838 cells covered by previously completed independent subtrees are admitted only after their original inputs and complete proof chains are matched; the remaining 772 037 original cells were independently checked in the completed full review. The present assembly rebinds these completed proof records. The disjoint union is exactly the stated 806 875 leaves, all excluded. Counts alone are not used to establish any geometric inclusion or scalar sign. □ \square □
Proof of Theorem 37.1. Suppose an original global minimizer has the stated type. Its nine essential maximal arcs are positive. Lemma 37.2 qualifies the same trajectory, its actual means and one complete normalized maximizing lift; Proposition 37.3 places its projection in B 06 B_{06} B 06 . It lies in an original leaf and in at least one member of each subsequent complete closed cover. At every candidate-preserving step its same fourteen values survive unless a necessary condition is contradicted. Theorem 37.4 leaves no such member. This contradiction proves the exclusion. □ \square □
In particular S 06 = ∅ \mathcal{S}_{06}=\varnothing S 06 = ∅ , and consequently its genuine closure in the bounded reciprocal coordinates is empty. This final implication concerns limits of that same strict original source class. It does not identify a larger weak algebraic envelope with that closure or impose new variational conditions at a zero-duration boundary.
The nine-arc type R04
The final type has two separated positive increments. Its two positive rectangles can be minimized at fixed area without changing the complete signed frame. This reduces its cost to a function of four exterior reciprocals. Its continuous boundary extension takes the octagon cost at one explicit degenerate point. We prove a strict larger-cost inequality throughout the required open domain.
Theorem 38.1 (Exclusion of R04). An original global minimizer cannot have nine essential positive arcs with native labels 021210102 and increments − − + − − + − − − - - + - - + - - - − − + − − + − − − , with the genuine extension j i + 9 = j i + 1 j_{i+9}=j_i+1 j i + 9 = j i + 1 .
38.1. The strict physical source and the four exterior coordinates. Let S 04 \mathcal{S}_{04} S 04 be the class of original global-minimizing trajectories of this type, with their actual reciprocal states, original full maximizing lift, both state covectors, complete signed endpoints and real winding. The nine physical arcs are all strictly positive. In native order put
( r 0 , … , r 8 ) = ( A , B , Q 1 , R 1 , C , Q 2 , R 2 , D , E ) , r i + 9 = r i . (206) (r_0,\ldots,r_8)=(A,B,Q_1,R_1,C,Q_2,R_2,D,E), \qquad r_{i+9}=r_i. \tag*{(206)} ( r 0 , … , r 8 ) = ( A , B , Q 1 , R 1 , C , Q 2 , R 2 , D , E ) , r i + 9 = r i . (206)
Here B B B denotes a reciprocal coordinate, not the essential count. The multiplier formula (36.2) gives
q 0 = 1 + A 2 ( A + B ) ( A + E ) ≔ p , q 1 = 1 + B 2 ( A + B ) ( B + Q 1 ) , q 2 = Q 1 + R 1 B + Q 1 , q 3 = Q 1 + R 1 C + R 1 , q 4 = 1 + C 2 ( C + R 1 ) ( C + Q 2 ) , q 5 = Q 2 + R 2 C + Q 2 , q 6 = Q 2 + R 2 D + R 2 , q 7 = 1 + D 2 ( D + R 2 ) ( D + E ) , q 8 = 1 + E 2 ( D + E ) ( A + E ) ≔ q . (207) \begin{aligned}
q_0&=\frac{1+A^2}{(A+B)(A+E)}\coloneqq p,\qquad
q_1=\frac{1+B^2}{(A+B)(B+Q_1)},\\
q_2&=\frac{Q_1+R_1}{B+Q_1},\qquad
q_3=\frac{Q_1+R_1}{C+R_1},\qquad
q_4=\frac{1+C^2}{(C+R_1)(C+Q_2)},\\
q_5&=\frac{Q_2+R_2}{C+Q_2},\qquad
q_6=\frac{Q_2+R_2}{D+R_2},\\
q_7&=\frac{1+D^2}{(D+R_2)(D+E)},\qquad
q_8=\frac{1+E^2}{(D+E)(A+E)}\coloneqq q. \tag*{(207)}
\end{aligned} q 0 q 2 q 5 q 7 = ( A + B ) ( A + E ) 1 + A 2 : = p , q 1 = ( A + B ) ( B + Q 1 ) 1 + B 2 , = B + Q 1 Q 1 + R 1 , q 3 = C + R 1 Q 1 + R 1 , q 4 = ( C + R 1 ) ( C + Q 2 ) 1 + C 2 , = C + Q 2 Q 2 + R 2 , q 6 = D + R 2 Q 2 + R 2 , = ( D + R 2 ) ( D + E ) 1 + D 2 , q 8 = ( D + E ) ( A + E ) 1 + E 2 : = q . (207)
Since τ i = r i log q i > 0 \tau_i=r_i\log q_i>0 τ i = r i log q i > 0 , every displayed multiplier is greater than one. In particular
X 1 = Q 1 − C > 0 , Y 1 = R 1 − B > 0 , X 2 = Q 2 − D > 0 , Y 2 = R 2 − C > 0. (208) X_1=Q_1-C>0,\qquad Y_1=R_1-B>0,\qquad X_2=Q_2-D>0,\qquad Y_2=R_2-C>0. \tag*{(208)} X 1 = Q 1 − C > 0 , Y 1 = R 1 − B > 0 , X 2 = Q 2 − D > 0 , Y 2 = R 2 − C > 0. (208)
Write I i = X i Y i I_i=X_iY_i I i = X i Y i . Set
H = ( 1 + B 2 ) ( 1 + D 2 ) ( A + B ) ( D + E ) , M = p 2 + q 2 − 1 , H=\frac{(1+B^2)(1+D^2)}{(A+B)(D+E)},\qquad M=p^2+q^2-1, H = ( A + B ) ( D + E ) ( 1 + B 2 ) ( 1 + D 2 ) , M = p 2 + q 2 − 1 ,
C 2 = p q H M − 1 , C^2=\frac{pqH}{M}-1, C 2 = M pq H − 1 ,
I 1 = ( B + C ) 2 2 − ( 1 + B 2 ) 2 ( q 2 − 1 ) 2 ( A + B ) 2 M , I_1=\frac{(B+C)^2}{2}-\frac{(1+B^2)^2(q^2-1)}{2(A+B)^2M}, I 1 = 2 ( B + C ) 2 − 2 ( A + B ) 2 M ( 1 + B 2 ) 2 ( q 2 − 1 ) ,
I 2 = ( C + D ) 2 2 − ( 1 + D 2 ) 2 ( p 2 − 1 ) 2 ( D + E ) 2 M . (209) I_2=\frac{(C+D)^2}{2}-\frac{(1+D^2)^2(p^2-1)}{2(D+E)^2M}. \tag*{(209)} I 2 = 2 ( C + D ) 2 − 2 ( D + E ) 2 M ( 1 + D 2 ) 2 ( p 2 − 1 ) . (209)
These are identities at the same physical candidate. Their derivation from the signed frame and the fixed-area variation is given in Section G.1. All denominators are positive since p , q > 1 p,q>1 p , q > 1 , so M > 1 M>1 M > 1 . The positive root determines C C C without a branch choice.
For positive a , b , X , Y a,b,X,Y a , b , X , Y define
k ( u , v ) = 2 − 2 u v ( u + v ) 2 , Γ ( a , b ; X , Y ) = ∫ a a + Y ∫ b b + X k ( u , v ) d v d u , k(u,v)=2-\frac{2uv}{(u+v)^2},\qquad\Gamma(a,b;X,Y)=\int_a^{a+Y}\int_b^{b+X} k(u,v)\,\mathrm{d}v\,\mathrm{d}u, k ( u , v ) = 2 − ( u + v ) 2 2 uv , Γ ( a , b ; X , Y ) = ∫ a a + Y ∫ b b + X k ( u , v ) d v d u ,
E ( a , b , I ) = min X > 0 Γ ( a , b ; X , I / X ) . (210) \mathcal{E}(a,b,I)=\min_{X>0}\Gamma(a,b;X,I/X). \tag*{(210)} E ( a , b , I ) = X > 0 min Γ ( a , b ; X , I / X ) . (210)
Lemma G.1 proves existence and uniqueness of this minimum for every a , b , I > 0 a,b,I>0 a , b , I > 0 . It also proves that the unique stationary shape is its global minimum. The actual two shapes in (38.3) are stationary by the fixed-area physical variation. Thus they realize E ( B , C , I 1 ) \mathcal{E}(B,C,I_1) E ( B , C , I 1 ) and E ( C , D , I 2 ) \mathcal{E}(C,D,I_2) E ( C , D , I 2 ) , respectively.
Define Ω 04 \Omega_{04} Ω 04 as the following open domain of z = ( A , B , D , E ) z=(A,B,D,E) z = ( A , B , D , E ) : the four coordinates are positive, p , q > 1 p,q>1 p , q > 1 , the positive root in (38.4) exists, I 1 , I 2 > 0 I_1,I_2>0 I 1 , I 2 > 0 , and the two unique minimizing shapes satisfy
X 1 < 1 + B 2 A + B − B − C , Y 2 < 1 + D 2 D + E − C − D , X_1<\frac{1+B^2}{A+B}-B-C,\qquad Y_2<\frac{1+D^2}{D+E}-C-D, X 1 < A + B 1 + B 2 − B − C , Y 2 < D + E 1 + D 2 − C − D ,
( B + C + Y 1 ) ( C + D + X 2 ) < 1 + C 2 . (211) (B+C+Y_1)(C+D+X_2)<1+C^2. \tag*{(211)} ( B + C + Y 1 ) ( C + D + X 2 ) < 1 + C 2 . (211)
The three inequalities follow exactly from q 1 , q 7 , q 4 > 1 q_1,q_7,q_4>1 q 1 , q 7 , q 4 > 1 . Consequently every member of S 04 \mathcal{S}_{04} S 04 projects into Ω 04 \Omega_{04} Ω 04 . This is an inclusion into a larger domain: none of the original winding, state or maximizing qualifications is replaced by a system of weak algebraic equations.
The exact physical cost
Use the potential ϕ \phi ϕ in (36.13); explicitly,
ϕ ( u , v ) = 3 u v − 1 − ( u 2 + v 2 ) log ( u + v ) + u 2 log ( 1 + u 2 ) + v 2 log ( 1 + v 2 ) 2 . \phi(u,v)=3uv-1-(u^2+v^2)\log(u+v)+\frac{u^2\log(1+u^2)+v^2\log(1+v^2)}{2}. ϕ ( u , v ) = 3 uv − 1 − ( u 2 + v 2 ) log ( u + v ) + 2 u 2 log ( 1 + u 2 ) + v 2 log ( 1 + v 2 ) .
Its mixed derivative is ϕ u v = k \phi_{uv}=k ϕ uv = k . For s = ( A , B , C , D , E ) s=(A,B,C,D,E) s = ( A , B , C , D , E ) with cyclic indices define
C 5 = − ∑ i = 0 4 ϕ ( s i , s i + 1 ) , G = C 5 + E ( B , C , I 1 ) + E ( C , D , I 2 ) . (212) C_5=-\sum_{i=0}^{4}\phi(s_i,s_{i+1}),\qquad G=C_5+\mathcal{E}(B,C,I_1)+\mathcal{E}(C,D,I_2). \tag*{(212)} C 5 = − i = 0 ∑ 4 ϕ ( s i , s i + 1 ) , G = C 5 + E ( B , C , I 1 ) + E ( C , D , I 2 ) . (212)
The fundamental theorem of calculus gives
Γ ( B , C ; X 1 , Y 1 ) = ϕ ( R 1 , Q 1 ) − ϕ ( B , Q 1 ) − ϕ ( R 1 , C ) + ϕ ( B , C ) , \Gamma(B,C;X_1,Y_1)=\phi(R_1,Q_1)-\phi(B,Q_1)-\phi(R_1,C)+\phi(B,C), Γ ( B , C ; X 1 , Y 1 ) = ϕ ( R 1 , Q 1 ) − ϕ ( B , Q 1 ) − ϕ ( R 1 , C ) + ϕ ( B , C ) ,
Γ ( C , D ; X 2 , Y 2 ) = ϕ ( R 2 , Q 2 ) − ϕ ( C , Q 2 ) − ϕ ( R 2 , D ) + ϕ ( C , D ) . \Gamma(C,D;X_2,Y_2)=\phi(R_2,Q_2)-\phi(C,Q_2)-\phi(R_2,D)+\phi(C,D). Γ ( C , D ; X 2 , Y 2 ) = ϕ ( R 2 , Q 2 ) − ϕ ( C , Q 2 ) − ϕ ( R 2 , D ) + ϕ ( C , D ) .
Adding these to C 5 C_5 C 5 cancels precisely the compressed edges ( B , C ) (B,C) ( B , C ) and ( C , D ) (C,D) ( C , D ) . Hence the once-counted physical nine-arc cost is
C 04 = G = − ϕ ( A , B ) − ϕ ( B , Q 1 ) + ϕ ( Q 1 , R 1 ) − ϕ ( R 1 , C ) − ϕ ( C , Q 2 ) + ϕ ( Q 2 , R 2 ) − ϕ ( R 2 , D ) − ϕ ( D , E ) − ϕ ( E , A ) . (213) \begin{aligned}
C_{04}=G={}&-\phi(A,B)-\phi(B,Q_1)+\phi(Q_1,R_1)-\phi(R_1,C)\\
&-\phi(C,Q_2)+\phi(Q_2,R_2)-\phi(R_2,D)\\
&-\phi(D,E)-\phi(E,A). \tag*{(213)}
\end{aligned} C 04 = G = − ϕ ( A , B ) − ϕ ( B , Q 1 ) + ϕ ( Q 1 , R 1 ) − ϕ ( R 1 , C ) − ϕ ( C , Q 2 ) + ϕ ( Q 2 , R 2 ) − ϕ ( R 2 , D ) − ϕ ( D , E ) − ϕ ( E , A ) . (213)
at every member of S 04 \mathcal{S}_{04} S 04 . In particular no additional 3 I i / 2 3I_i/2 3 I i /2 is charged after the rectangle integral has been inserted.
Put
s = 2 , b 0 = ( 5 + 4 s ) − 1 / 2 , a 0 = ( 1 + s ) b 0 , z 0 = ( a 0 , b 0 , b 0 , a 0 ) , s=\sqrt{2},\quad b_0=(5+4s)^{-1/2},\quad a_0=(1+s)b_0,\quad z_0=(a_0,b_0,b_0,a_0), s = 2 , b 0 = ( 5 + 4 s ) − 1/2 , a 0 = ( 1 + s ) b 0 , z 0 = ( a 0 , b 0 , b 0 , a 0 ) ,
m 0 = 16 + 8 s − ( 6 + 4 s ) log 2 5 + 4 s . (214) m_0=\frac{16+8s-(6+4s)\log2}{5+4s}. \tag*{(214)} m 0 = 5 + 4 s 16 + 8 s − ( 6 + 4 s ) log 2 . (214)
The exact identity
m 0 = J 8 = 2 ( 8 − 4 2 − log 2 ) 2 2 − 1 (215) m_0=J_8=\frac{2(8-4\sqrt{2}-\log2)}{2\sqrt{2}-1} \tag*{(215)} m 0 = J 8 = 2 2 − 1 2 ( 8 − 4 2 − log 2 ) (215)
follows by cross multiplication, using s 2 = 2 s^2=2 s 2 = 2 : both cross products are 16 + 24 s − ( 10 + 8 s ) log 2 16+24s-(10+8s)\log2 16 + 24 s − ( 10 + 8 s ) log 2 .
Theorem 38.2 (Strict cost bound on the four-dimensional domain). For every z ∈ Ω 04 z\in\Omega_{04} z ∈ Ω 04 , the function in (212) satisfies G ( z ) > m 0 G(z)>m_0 G ( z ) > m 0 .
The proof occupies Appendix G. A rectangle comparison and shape estimates give explicit lower functions and necessary conditions. An analytic tail estimate reduces A , E A,E A , E to ( 0 , 5 ) (0,5) ( 0 , 5 ) . A quantitative local argument treats a cube of radius 11 / 10000 11/10000 11/10000 about z 0 z_0 z 0 . The complement is excluded by a complete directed interval cover beginning from sixteen exact closed boxes in [ 0 , 5 ] × [ 0 , 1 ] 2 × [ 0 , 5 ] [0,5]\times[0,1]^2\times[0,5] [ 0 , 5 ] × [ 0 , 1 ] 2 × [ 0 , 5 ] . Section H.9 specifies the frozen numerical kernel, arithmetic rules and independent coverage checks.
Proof of Theorem 38.1. A member of S 04 \mathcal{S}_{04} S 04 would project into Ω 04 \Omega_{04} Ω 04 and, by (213) , have C 04 = G C_{04}=G C 04 = G . The competitor budget (18) , Theorem 38.2 and (215) give
J 8 ≥ C 04 = G > m 0 = J 8 , J_8\geq C_{04}=G>m_0=J_8, J 8 ≥ C 04 = G > m 0 = J 8 ,
a contradiction. Thus S 04 = ∅ \mathcal{S}_{04}=\varnothing S 04 = ∅ and its qualified global closure S ‾ 04 \overline{\mathcal{S}}_{04} S 04 is empty as well. □ \square □
At z 0 z_0 z 0 the continuous extension has p = q = 1 p=q=1 p = q = 1 , C = b 0 C=b_0 C = b 0 , I 1 = I 2 = 2 b 0 2 I_1=I_2=2b_0^2 I 1 = I 2 = 2 b 0 2 and X i = Y i = 2 b 0 X_i=Y_i=\sqrt{2}b_0 X i = Y i = 2 b 0 . Its cost is m 0 m_0 m 0 . These zero-duration arcs place it on the boundary of the relaxed domain, outside S 04 \mathcal{S}_{04} S 04 . The strict pointwise bound requires no uniform positive gap from J 8 J_8 J 8 on the open domain and does not classify arbitrary weak nine-tuples as original minimizing trajectories.
Completion of the proof
Theorem 39.1 (Exclusion of all long minimizing types). No original global minimizer has essential count B ≥ 8 B \ge8 B ≥ 8 .
Proof. Theorem 24.1 leaves precisely R01–R06. Theorem 27.1 excludes R01; Theorems 35.1 and 35.2 exclude R02 and R03. The source interface of § 6.2 and Theorem 36.1 exclude R05; Theorem 37.1 excludes R06. These statements retain the original signed trajectory, true twist, real winding, required common lift and physical cost counted once.
For R04, § 6.12 maps the original strict nine-arc source into Ω 04 \Omega_{04} Ω 04 with G = C 9 G=C_9 G = C 9 . Theorem 38.2 and (215) give C 9 = G > J 8 C_9=G>J_8 C 9 = G > J 8 , contradicting the octagonal budget C 9 ≤ J 8 = C o c t C_9\le J_8=C_{\mathrm{oct}} C 9 ≤ J 8 = C oct . Thus the strict source is empty, and so is its qualified global closure, as stated in Theorem 38.1 . Every long necessary type has been excluded. □ \square □
Proof of Theorem 1.1 . An original global minimizer exists and has a finite positive representation, by the geometric inputs recalled in § 3.1 . The preceding theorem excludes B ≥ 8 B\ge8 B ≥ 8 . Proposition A.15 , with the domain identification of Lemma A.14 , therefore identifies every global minimizer with a nonsingular affine image of the smoothed regular octagon. The minimum density is C o c t / 2 C_{\mathrm{oct}}/2 C oct /2 : the normalization fixes the minimum circumscribed hexagon area at 2 3 2\sqrt{3} 2 3 and gives disk area 3 C \sqrt{3}C 3 C . Substitution of (18) gives the stated constant. Conversely, nonsingular affine transformations preserve δ L \delta_L δ L , so every such image attains the same minimum. A disk has equality in the density inequality exactly when it is itself a global minimizer; this proves the asserted equality range. The octagon has B = 4 B=4 B = 4 , and (16) gives N ≤ B + 1 = 5 N\le B+1=5 N ≤ B + 1 = 5 for every prescribed cut. □ \square □
Appendix A. Geometric and switching proofs
The geometric problem and its full lift
Normalization and admissible trajectories
We use the geometric class, matrices, physical-time equations, cost and full endpoints of § 3.1 , in particular (1) –(4) . The prescribed real winding and six-curve reconstruction remain part of that class. We verify the control convention and the local assertions stated there.
To verify the control convention, put
B 0 = ( 1 / 2 1 0 − 1 / 2 ) , B 1 = ( 0 − 1 / 2 − 1 / 2 0 ) , B 2 = ( − 1 / 2 1 0 1 / 2 ) . B_0=\begin{pmatrix}1/2&1\\0&-1/2\end{pmatrix},\quad
B_1=\begin{pmatrix}0&-1/2\\-1/2&0\end{pmatrix},\quad
B_2=\begin{pmatrix}-1/2&1\\0&1/2\end{pmatrix}. B 0 = ( 1/2 0 1 − 1/2 ) , B 1 = ( 0 − 1/2 − 1/2 0 ) , B 2 = ( − 1/2 0 1 1/2 ) .
Then d j = tr ( B j X ) d_j=\operatorname{tr}(B_jX) d j = tr ( B j X ) are the positive quantities
( d 0 , d 1 , d 2 ) = ( 1 + u v , S 2 v , 1 − u v ) , D X ( W ) F j = [ B j , X ] d j . (216) (d_0,d_1,d_2)=\left(\frac{1+u}{v},\frac{S}{2v},\frac{1-u}{v}\right),\qquad
DX(W)F_j=\frac{[B_j,X]}{d_j}. \tag*{(216)} ( d 0 , d 1 , d 2 ) = ( v 1 + u , 2 v S , v 1 − u ) , D X ( W ) F j = d j [ B j , X ] . (216)
The positive d j d_j d j make the control changes (5) inverse, measurable and vertex-preserving. In the coordinates of [3], ( x , y ) = ( u , v ) / 3 (x,y)=(u,v)/\sqrt{3} ( x , y ) = ( u , v ) / 3 . For its control α \alpha α , take μ = ( α 2 , α 0 , α 1 ) \mu=(\alpha_2,\alpha_0,\alpha_1) μ = ( α 2 , α 0 , α 1 ) ; conjugation by D ∗ D_* D ∗ sends its control matrix to − 2 ∑ j μ j B j / 3 -2\sum_j\mu_jB_j/\sqrt{3} − 2 ∑ j μ j B j / 3 . The common factor cancels in the commutator-to-trace quotient. Thus the complete measurable control class and the physical time are unchanged.
The original compact set
Let R ( w ) = ( w − 3 ) / ( w + 1 ) \mathcal{R}(w)=(w-3)/(w+1) R ( w ) = ( w − 3 ) / ( w + 1 ) . The compactness theorem of [3], Definition 5.0.1 and Theorem 5.0.2, places every normalized global minimizer in
K = Ω ∖ ⋃ j = 0 2 R j ( { v > 9 3 / 2 } ∪ { v > 15 ( 1 − u ) } ) . K=\Omega\setminus\bigcup_{j=0}^{2}\mathcal{R}^{j}\left(\{v>9\sqrt{3}/2\}\cup\{v>15(1-u)\}\right). K = Ω ∖ j = 0 ⋃ 2 R j ( { v > 9 3 /2 } ∪ { v > 15 ( 1 − u )} ) .
Its six cuts are (6) ; they constrain the minimizing trajectory, not a competing local replacement.
Lemma A.1. The rates on K K K satisfy (7) .
Proof. The side cuts give the three individual bounds. Put c 0 = 513 / 1000 c_0=513/1000 c 0 = 513/1000 . The cusp cuts imply v < 39 / 5 v<39/5 v < 39/5 and ( u ± 1 ) 2 + v 2 ≥ c 0 v (u\pm1)^2+v^2\ge c_0v ( u ± 1 ) 2 + v 2 ≥ c 0 v . If v < 1 / 2 v<1/2 v < 1/2 , the side cut gives u 2 ≥ q 0 = 1 + 2 v / 15 − v 2 > 0 u^2\ge q_0=1+2v/15-v^2>0 u 2 ≥ q 0 = 1 + 2 v /15 − v 2 > 0 . Hence, with δ = 1 − ∣ u ∣ \delta=1-|u| δ = 1 − ∣ u ∣ ,
c 0 v ≤ δ 2 + v 2 ≤ ( 1 − q 0 ) 2 + v 2 . c_0v\le\delta^2+v^2\le(1-\sqrt{q_0})^2+v^2. c 0 v ≤ δ 2 + v 2 ≤ ( 1 − q 0 ) 2 + v 2 .
Writing d = c 0 / 2 − 1 / 15 = 1139 / 6000 d=c_0/2-1/15=1139/6000 d = c 0 /2 − 1/15 = 1139/6000 , this implies q 0 ≤ 1 − d v \sqrt{q_0}\le1-dv q 0 ≤ 1 − d v , whose positive right side may be squared. Consequently v ≥ c 0 / ( 1 + d 2 ) = 18468000 / 37297321 > 99 / 200 v\ge c_0/(1+d^2)=18468000/37297321>99/200 v ≥ c 0 / ( 1 + d 2 ) = 18468000/37297321 > 99/200 . For any two rates x , y x,y x , y , the three identities
2 a b a + b = v , 2 a c a + c = 4 v ( u + 1 ) 2 + v 2 , 2 b c b + c = 4 v ( u − 1 ) 2 + v 2 \frac{2ab}{a+b}=v,\qquad
\frac{2ac}{a+c}=\frac{4v}{(u+1)^2+v^2},\qquad
\frac{2bc}{b+c}=\frac{4v}{(u-1)^2+v^2} a + b 2 ab = v , a + c 2 a c = ( u + 1 ) 2 + v 2 4 v , b + c 2 b c = ( u − 1 ) 2 + v 2 4 v
give x y ≤ ( 39 / 10 ) ( x + y ) xy\le(39/10)(x+y) x y ≤ ( 39/10 ) ( x + y ) . Also ( 15 − x ) ( 15 − y ) ≥ 0 (15-x)(15-y)\ge0 ( 15 − x ) ( 15 − y ) ≥ 0 , so x y ≥ 15 ( x + y ) − 225 xy\ge15(x+y)-225 x y ≥ 15 ( x + y ) − 225 . Combining them gives the asserted pairwise bound. □
Five-endpoint replacements
Proof of Lemma 2.2. Translate the original tail in time and attach it to the new window. Matching G , W G,W G , W matches all five endpoint coordinates: three group coordinates and two state coordinates. The global endpoints and twist are unchanged. Each tangent G X q j GXq_j GX q j is nonzero because det X = 1 \det X=1 det X = 1 , and the matching endpoints match these tangents. For completeness, convexity follows directly from the complete control triangle. Set B = ∑ j μ j B j B = \sum_j \mu_j B_j B = ∑ j μ j B j and d = tr ( B X ) > 0 d = \operatorname{tr}(BX) > 0 d = tr ( B X ) > 0 . The identity X B X = B + d X XBX = B + dX X B X = B + d X gives
X + X − 1 X ′ = − 2 B / d . X + X^{-1}X' = -2B/d. X + X − 1 X ′ = − 2 B / d .
Thus for γ j = G q j \gamma_j = Gq_j γ j = G q j ,
( det ( γ 0 ′ , γ 0 ′ ′ ) , det ( γ 2 ′ , γ 2 ′ ′ ) , det ( γ 4 ′ , γ 4 ′ ′ ) ) = 3 d ( μ 1 , μ 2 , μ 0 ) ≥ 0 \left(\det(\gamma'_0,\gamma''_0),\det(\gamma'_2,\gamma''_2),\det(\gamma'_4,\gamma''_4)\right) = \frac{\sqrt{3}}{d}(\mu_1,\mu_2,\mu_0) \ge0 ( det ( γ 0 ′ , γ 0 ′′ ) , det ( γ 2 ′ , γ 2 ′′ ) , det ( γ 4 ′ , γ 4 ′′ ) ) = d 3 ( μ 1 , μ 2 , μ 0 ) ≥ 0
almost everywhere. The opposite arcs have the same curvatures. Uniform convergence of the nonzero tangents preserves their winding number, which is one. Their absolutely continuous angle is nondecreasing and changes by 2 π 2\pi 2 π . Every oriented tangent is consequently a supporting line: relative to that tangent, the normal coordinate first increases and then decreases to zero. Constant-angle portions cannot reverse because the speed is positive. The concatenation therefore traces a convex boundary once.
The matrix homotopy also persists. After extending the comparison reparametrization to the whole interval, the matrix quotient of the new and old paths lies in a real logarithm neighborhood of I I I . Its logarithm is traceless, continuous, and zero at both endpoints. Exponentiating a scalar multiple of this logarithm gives a fixed-endpoint homotopy. Reparametrization is used only for this comparison; the replacement still solves the physical-time equation. Green’s formula for the six arcs gives
∣ D ∣ = 1 2 ∑ j = 0 5 ∫ det ( γ j , γ j ′ ) d t = 3 ∫ ℓ ( W ) d t . |D| = \frac{1}{2}\sum_{j=0}^{5}\int\det(\gamma_j,\gamma'_j)\,dt = \sqrt{3}\int\ell(W)\,dt. ∣ D ∣ = 2 1 j = 0 ∑ 5 ∫ det ( γ j , γ j ′ ) d t = 3 ∫ ℓ ( W ) d t .
All terms outside the replacement cancel. This proves the assertion. □
All duration variations below are taken about positive arcs with positive caps. Small variations preserve positivity and remain in the open set Ω \Omega Ω . Signed analytic extensions of durations may be used for differentiation, but are not themselves competing geometric trajectories.
The full free-time maximum principle
A.1.4. The full free-time maximum principle. Use the full lift of Definition 2.3, with the matrix pairing tr ( Λ G − 1 δ G ) \operatorname{tr}(\Lambda G^{-1}\delta G) tr ( Λ G − 1 δ G ) and the free-time equations (8) . The switching rows and their positive factor β \beta β are (9) –(10) ; costates from independently selected lifts are never combined. Thus a maximizing 02, 01, 12 tie has signs respectively ( 0 , ≤ 0 , ≥ 0 ) (0,\le0,\ge0) ( 0 , ≤ 0 , ≥ 0 ) , ( ≥ 0 , 0 , ≤ 0 ) (\ge0,0,\le0) ( ≥ 0 , 0 , ≤ 0 ) , ( ≤ 0 , ≥ 0 , 0 ) (\le0,\ge0,0) ( ≤ 0 , ≥ 0 , 0 ) .
Lemma A.2. Along every measurable maximizing lift, each σ i \sigma_i σ i is C 1 C^1 C 1 with absolutely continuous first derivative. Its second derivative is the control-weighted sum of the supported vertex equations almost everywhere.
Proof. Let L j L_j L j be differentiation along the full equations with mode j j j . From α 0 = β ( F 0 − F 2 ) \alpha_0 = \beta(F_0-F_2) α 0 = β ( F 0 − F 2 ) and its cyclic versions,
L j σ i − L k σ i = f i ( H j − H k ) , f i = α i ∇ β β , L j H j = 0. L_j\sigma_i-L_k\sigma_i=f_i(H_j-H_k),\qquad f_i=\frac{\alpha_i\nabla\beta}{\beta},\qquad L_jH_j=0. L j σ i − L k σ i = f i ( H j − H k ) , f i = β α i ∇ β , L j H j = 0.
Hence Ψ i = L j σ i − f i H j \Psi_i=L_j\sigma_i-f_iH_j Ψ i = L j σ i − f i H j is independent of j j j . On the maximizing trajectory σ i ′ = Ψ i \sigma'_i=\Psi_i σ i ′ = Ψ i almost everywhere. The right side is continuous and absolutely continuous. Differentiating once more on a supported mode, where H j = 0 H_j=0 H j = 0 , gives L j Ψ i = L j 2 σ i L_j\Psi_i=L_j^2\sigma_i L j Ψ i = L j 2 σ i . Averaging gives the assertion without differentiating the measurable control. □
Symmetry and genuine reanchoring
The whole-state, frame and lift transports are those of Proposition 2.5; the twisted state and group extension is (14) , with no state-costate extension.
Proof of Lemma 2.6. The equations, endpoints, and periodicity of the cost follow from the covariance identities. The family G e ( λ s ) − 1 G e ( λ s + t ) G_e(\lambda s)^{-1}G_e(\lambda s+t) G e ( λ s ) − 1 G e ( λ s + t ) , 0 ≤ λ ≤ 1 0\leq\lambda\leq1 0 ≤ λ ≤ 1 , has constant endpoints I I I , R ∗ R_* R ∗ and preserves the real homotopy class. More geometrically, with Γ ( t ) = G e ( t ) q 0 \Gamma(t)=G_e(t)q_0 Γ ( t ) = G e ( t ) q 0 ,
G s ( t ) q j = G e ( s ) − 1 Γ ( s + t + j T ) . G_s(t)q_j=G_e(s)^{-1}\Gamma(s+t+jT). G s ( t ) q j = G e ( s ) − 1 Γ ( s + t + j T ) .
These are the same convex boundary, with a new starting point and an area-preserving linear transformation. Its area and circumscribed-hexagon normalization are unchanged. The original maximum principle applies afresh. □
A complete local support using fewer than B B B physical arcs can be put inside such a genuine interval by placing the cut in a remaining positive arc. The numerical properness condition guarantees room for the support; it does not replace the geometric reanchoring argument.
Finiteness and the competitor budget
We use the following qualitative input.
Theorem A.3 (Hales–Vajjha). Every normalized global minimizer of the original problem is, up to a null set, a vertex control with finitely many switches.
This is [3, Theorem 16.3.1]; it applies to the chosen minimizer, not merely to an equal-cost representative. After adjacent equal labels are merged, all arcs are positive.
Elementary scalar enclosures
For integers N ≥ 1 N\geq1 N ≥ 1 , n ≥ 0 n\geq0 n ≥ 0 , x ≥ 1 x\geq1 x ≥ 1 and 0 ≤ w < n + 2 0\leq w<n+2 0 ≤ w < n + 2 , put v x = ( x − 1 ) / ( x + 1 ) v_x=(x-1)/(x+1) v x = ( x − 1 ) / ( x + 1 ) and
L N ( x ) = 2 ∑ j = 0 N − 1 v x 2 j + 1 2 j + 1 , U N ( x ) = L N ( x ) + 2 v x 2 N + 1 ( 2 N + 1 ) ( 1 − v x 2 ) . (217) L_N(x)=2\sum_{j=0}^{N-1}\frac{v_x^{2j+1}}{2j+1},\qquad U_N(x)=L_N(x)+\frac{2v_x^{2N+1}}{(2N+1)(1-v_x^2)}. \tag*{(217)} L N ( x ) = 2 j = 0 ∑ N − 1 2 j + 1 v x 2 j + 1 , U N ( x ) = L N ( x ) + ( 2 N + 1 ) ( 1 − v x 2 ) 2 v x 2 N + 1 . (217)
Then L N ( x ) ≤ log x ≤ U N ( x ) L_N(x)\leq\log x\leq U_N(x) L N ( x ) ≤ log x ≤ U N ( x ) , strictly for x > 1 x>1 x > 1 . Evaluate U N − L N U_N-L_N U N − L N as its displayed tail, without rounded subtraction. As functions of v x v_x v x , the lower and upper error derivatives are 2 v x 2 N / ( 1 − v x 2 ) 2v_x^{2N}/(1-v_x^2) 2 v x 2 N / ( 1 − v x 2 ) and 4 v x 2 N + 2 / ( ( 2 N + 1 ) ( 1 − v x 2 ) 2 ) 4v_x^{2N+2}/((2N+1)(1-v_x^2)^2) 4 v x 2 N + 2 / (( 2 N + 1 ) ( 1 − v x 2 ) 2 ) ; both errors vanish at zero. Also
E n ( w ) = ∑ j = 0 n w j j ! + w n + 1 ( n + 1 ) ! [ 1 − w / ( n + 2 ) ] ≥ e w , (218) \mathcal{E}_n(w)=\sum_{j=0}^{n}\frac{w^j}{j!}+\frac{w^{n+1}}{(n+1)![1-w/(n+2)]}\geq e^w, \tag*{(218)} E n ( w ) = j = 0 ∑ n j ! w j + ( n + 1 )! [ 1 − w / ( n + 2 )] w n + 1 ≥ e w , (218)
strictly for w > 0 w>0 w > 0 : bound ( n + 1 + j ) ! ≥ ( n + 1 ) ! ( n + 2 ) j (n+1+j)!\geq(n+1)!(n+2)^j ( n + 1 + j )! ≥ ( n + 1 )! ( n + 2 ) j .
The competitor cost C o c t C_{\mathrm{oct}} C oct in (18) has area A o c t = 3 C o c t \mathcal{A}_{\mathrm{oct}}=\sqrt{3}C_{\mathrm{oct}} A oct = 3 C oct . It gives the exact bounds
C < C ∗ : = 180483 100000 , A < A 0 : = 781513607211 250000000000 , π > p 0 : = 3141592653589 10 12 . (219) C<C_*:=\frac{180483}{100000},\qquad\mathcal{A}<A_0:=\frac{781513607211}{250000000000},\qquad\pi>p_0:=\frac{3141592653589}{10^{12}}. \tag*{(219)} C < C ∗ := 100000 180483 , A < A 0 := 250000000000 781513607211 , π > p 0 := 1 0 12 3141592653589 . (219)
Here are exact bounds sufficient to verify the constants. Squaring the consecutive rationals 1414213562373095 / 10 15 1414213562373095/10^{15} 1414213562373095/1 0 15 and 1414213562373096 / 10 15 1414213562373096/10^{15} 1414213562373096/1 0 15 brackets 2 \sqrt{2} 2 ; the analogous numerators 1732050807568877 1732050807568877 1732050807568877 , 1732050807568878 1732050807568878 1732050807568878 bracket 3 \sqrt{3} 3 . Moreover L 16 ( 2 ) < log 2 < U 16 ( 2 ) L_{16}(2)<\log2<U_{16}(2) L 16 ( 2 ) < log 2 < U 16 ( 2 ) by (A.2). The quotient ( 8 − 4 h − l ) / ( 2 h − 1 ) (8-4h-l)/(2h-1) ( 8 − 4 h − l ) / ( 2 h − 1 ) decreases in both h h h and l l l on these intervals. Its rational upper bound proves the first two claims. Machin’s identity π = 16 arctan ( 1 / 5 ) − 4 arctan ( 1 / 239 ) \pi=16\arctan(1/5)-4\arctan(1/239) π = 16 arctan ( 1/5 ) − 4 arctan ( 1/239 ) , with respectively 26 and 10 alternating-series terms and their next-frame remainders, proves the third. The branch of the identity follows from tan ( 4 arctan ( 1 / 5 ) ) = 120 / 119 \tan(4\arctan(1/5))=120/119 tan ( 4 arctan ( 1/5 )) = 120/119 : subtracting arctan ( 1 / 239 ) \arctan(1/239) arctan ( 1/239 ) gives the angle π / 4 \pi/4 π /4 in ( 0 , 1 ) (0,1) ( 0 , 1 ) .
Switching restrictions and normality
Every letter in this section denotes an actual positive maximal control arc. The two exterior letters of a local word remain positive caps. Roots and crossing derivatives always belong to the chosen lift of the whole containing interval.
An independent edge argument.
Lemma A.4. The words 2020, 0101, 1212 cannot occur on a global minimizer. No normality hypothesis is needed.
Proof. A minimizing subinterval on a control edge is also an extremal of the independent edge problem: an edge competitor fixing the state and the group increment is a competitor in Lemma 2.2. Use the edge coordinates of [3, Proposition 8.1.3], with x = u / 3 x=u/\sqrt{3} x = u / 3 , y = v / 3 y=v/\sqrt{3} y = v / 3 , and the group coordinate s s s . The equations and free-time Hamiltonian are
x ′ = y > 0 , s ′ = 1 / y , y ′ = f ( x , y , r ) = 2 3 y 2 r − 1 + 2 3 x r , ∣ r ∣ ≤ 1 2 . x'=y>0,\qquad s'=1/y,\qquad y'=f(x,y,r)=\frac{2\sqrt{3}y^2r}{-1+2\sqrt{3}xr},\qquad\lvert r\rvert\le\frac{1}{2}. x ′ = y > 0 , s ′ = 1/ y , y ′ = f ( x , y , r ) = − 1 + 2 3 x r 2 3 y 2 r , ∣ r ∣ ≤ 2 1 . H = λ 1 y + λ 2 f + Q ( x ) / y = 0 , Q ( x ) = λ 3 + λ c o s t x 2 , λ c o s t ≤ 0. H=\lambda_1y+\lambda_2f+Q(x)/y=0,\qquad Q(x)=\lambda_3+\lambda_{\mathrm{cost}}x^2,\qquad\lambda_{\mathrm{cost}}\le0. H = λ 1 y + λ 2 f + Q ( x ) / y = 0 , Q ( x ) = λ 3 + λ cost x 2 , λ cost ≤ 0. Here λ 2 \lambda_2 λ 2 and λ c o s t \lambda_{\mathrm{cost}} λ cost are constant; this independent edge lift need not be the projection of the full lift. Since f y = 2 f / y f_y=2f/y f y = 2 f / y , its equations give
χ : = y λ 2 , χ ′ = 2 Q / y . \chi:=y\lambda_2,\qquad\chi'=2Q/y. χ := y λ 2 , χ ′ = 2 Q / y . If λ c o s t < 0 \lambda_{\mathrm{cost}}<0 λ cost < 0 , the right side changes sign at most twice and χ \chi χ has at most three zeros. If λ c o s t = 0 \lambda_{\mathrm{cost}}=0 λ cost = 0 , then χ \chi χ is either strictly monotone or a nonzero constant; a zero constant would annihilate the full edge multiplier. Also f r < 0 f_r<0 f r < 0 , so the maximizing control is an endpoint except at the finitely many zeros.
The labels 0, 2 correspond respectively to r = − 1 / 2 , 1 / 2 r=-1/2,1/2 r = − 1/2 , 1/2 , and f ( − 1 / 2 ) − f ( 1 / 2 ) = 2 3 y 2 / ( 1 − 3 x 2 ) > 0 f(-1/2)-f(1/2)=2\sqrt{3}y^2/(1-3x^2)>0 f ( − 1/2 ) − f ( 1/2 ) = 2 3 y 2 / ( 1 − 3 x 2 ) > 0 . At the three switches of 2020, located at x 1 < x 2 < x 3 x_1<x_2<x_3 x 1 < x 2 < x 3 , maximization therefore gives Q ( x 1 ) ≥ 0 Q(x_1)\ge0 Q ( x 1 ) ≥ 0 , Q ( x 2 ) ≤ 0 Q(x_2)\le0 Q ( x 2 ) ≤ 0 , Q ( x 3 ) ≥ 0 Q(x_3)\ge0 Q ( x 3 ) ≥ 0 . Strict concavity contradicts these inequalities if λ c o s t < 0 \lambda_{\mathrm{cost}}<0 λ cost < 0 . Otherwise Q = 0 Q=0 Q = 0 , and the three switching zeros again annihilate the edge lift. Cyclic relabeling proves the other exclusions. □
Binary identities and actual signs.
Write
Λ = ( l b m − l ) , K e = 2 u l + b − u 2 m , L = l − u m , χ = v p v , d s = d t / v . \Lambda=\begin{pmatrix}l&b\\m&-l\end{pmatrix},\qquad K_e=2ul+b-u^2m,\qquad L=l-um,\qquad\chi=vp_v,\qquad ds=dt/v. Λ = ( l m b − l ) , K e = 2 u l + b − u 2 m , L = l − u m , χ = v p v , d s = d t / v .
On either mode 0 or mode 2, substitution in the full equations gives
( K e ) s = 0 , L s = K e , m s = − 2 L , χ s = 2 K e − η ( 3 + u 2 ) , u s = v 2 > 0. (220) (K_e)_s=0,\qquad L_s=K_e,\qquad m_s=-2L,\qquad\chi_s=2K_e-\eta(3+u^2),\qquad u_s=v^2>0. \tag*{(220)} ( K e ) s = 0 , L s = K e , m s = − 2 L , χ s = 2 K e − η ( 3 + u 2 ) , u s = v 2 > 0. (220)
At a 02 tie, χ = 0 \chi=0 χ = 0 ; its maximizing signs are χ ≥ 0 \chi\ge0 χ ≥ 0 on mode 0 and χ ≤ 0 \chi\le0 χ ≤ 0 on mode 2. Set
B e = K e − η ( 3 + u 2 ) / 2 , w = S / v 2 . B_e=K_e-\eta(3+u^2)/2,\qquad w=S/v^2. B e = K e − η ( 3 + u 2 ) /2 , w = S / v 2 .
Then σ 0 = w χ \sigma_0=w\chi σ 0 = w χ , B e ′ = − η u v B'_e=-\eta uv B e ′ = − η uv , and w ′ = 2 / v w'=2/v w ′ = 2/ v on mode 0, w ′ = − 2 / v w'=-2/v w ′ = − 2/ v on mode 2.
Lemma A.5 (Complete binary action and transport). On a binary block of one full active-zero lift, retain its objective multiplier η \eta η , write K = K e K=K_e K = K e , and put
P η = η u + 2 L − χ , B = ( L K m − L ) , N ( S ) = ( 1 0 S 1 ) . P_\eta=\eta u+2L-\chi,\qquad\mathrm{B}=\begin{pmatrix}L&K\\m&-L\end{pmatrix},\qquad N(S)=\begin{pmatrix}1&0\\S&1\end{pmatrix}. P η = η u + 2 L − χ , B = ( L m K − L ) , N ( S ) = ( 1 S 0 1 ) . For physical cost A A A and shear S S S from a fixed initial point,
2 η A = Δ P η = η Δ u + 2 K S − Δ χ , B ( S ) = N ( S ) − 1 B ( 0 ) N ( S ) . (221) 2\eta A=\Delta P_\eta=\eta\Delta u+2KS-\Delta\chi,\qquad\mathrm{B}(S)=N(S)^{-1}\mathrm{B}(0)N(S). \tag*{(221)} 2 η A = Δ P η = η Δ u + 2 K S − Δ χ , B ( S ) = N ( S ) − 1 B ( 0 ) N ( S ) . (221) At an active mode, set e = 1 + u e=1+u e = 1 + u for mode 0 and e = u − 1 e=u-1 e = u − 1 for mode 2. The complete reconstruction is
l = L + u m , b = K − 2 u L − u 2 m , p v = χ / v , l=L+um,\qquad b=K-2uL-u^2m,\qquad p_v=\chi/v, l = L + u m , b = K − 2 uL − u 2 m , p v = χ / v , p u = − χ / e − tr ( ( Λ − η A ∗ ) X ) / v . (222) p_u=-\chi/e-\operatorname{tr}((\Lambda-\eta A_*)X)/v. \tag*{(222)} p u = − χ / e − tr (( Λ − η A ∗ ) X ) / v . (222)
On a single mode, put ε = 1 \varepsilon=1 ε = 1 for mode 0 and − 1 -1 − 1 for mode 2; e = u + ε e=u+\varepsilon e = u + ε and the constant κ = v / e ≠ 0 \kappa=v/e\ne0 κ = v / e = 0 are its ray data. With arbitrary constants K , B , M , C K,B,M,C K , B , M , C , the same solution is
L = B − K κ 2 e , m = M + 2 B κ 2 e − K κ 4 e 2 , χ = C + ( 4 η − 2 K ) / e + 2 η ε log ∣ e ∣ − η e κ 2 . (223) \begin{aligned}
L&=B-\frac{K}{\kappa^{2}e},\qquad m=M+\frac{2B}{\kappa^{2}e}-\frac{K}{\kappa^{4}e^{2}},\\
\chi&=C+\frac{(4\eta-2K)/e+2\eta\varepsilon\log\lvert e\rvert-\eta e}{\kappa^{2}}. \tag*{(223)}
\end{aligned} L χ = B − κ 2 e K , m = M + κ 2 e 2 B − κ 4 e 2 K , = C + κ 2 ( 4 η − 2 K ) / e + 2 η ε log ∣ e ∣ − η e . (223)
On one constant-mode arc the four arbitrary initial data ( K , L , m , χ ) (K,L,m,\chi) ( K , L , m , χ ) , equivalently ( K , B , M , C ) (K,B,M,C) ( K , B , M , C ) , give every active-zero full adjoint, including singular matrix adjoints. Across a binary switch the actual tie χ = 0 \chi=0 χ = 0 is additionally retained. These identities neither prove normality nor identify lifts across a global seam.
If the lift is normal, after joint normalization set P = P 1 P=P_1 P = P 1 . It is continuous at actual binary switches. Under a lawful chart change evaluate it on the whole transported lift; under reversal use P γ = ϵ γ ( u γ + 2 L γ − χ γ ) P_\gamma=\epsilon_\gamma(u_\gamma+2L_\gamma-\chi_\gamma) P γ = ϵ γ ( u γ + 2 L γ − χ γ ) , with ϵ γ = − 1 \epsilon_\gamma=-1 ϵ γ = − 1 for reversal and + 1 +1 + 1 otherwise. For consecutive blocks with actual joins q i q_i q i ,
2 ∑ i = 1 n A i = P γ n ( q n ) − P γ 1 ( q 0 ) + ∑ i = 1 n − 1 [ P γ i ( q i ) − P γ i + 1 ( q i ) ] . (224) 2\sum_{i=1}^{n}A_i=P_{\gamma_n}(q_n)-P_{\gamma_1}(q_0)+\sum_{i=1}^{n-1}\left[P_{\gamma_i}(q_i)-P_{\gamma_{i+1}}(q_i)\right]. \tag*{(224)} 2 i = 1 ∑ n A i = P γ n ( q n ) − P γ 1 ( q 0 ) + i = 1 ∑ n − 1 [ P γ i ( q i ) − P γ i + 1 ( q i ) ] . (224)
Thus for two overlapping blocks 2 J = Δ P L + Δ P R − 2 c ( I ) 2J=\Delta P_L+\Delta P_R-2c(I) 2 J = Δ P L + Δ P R − 2 c ( I ) , where I I I is their shared physical arc. Different-chart endpoint potentials need not agree.
Proof. Equation (A.5) gives ( P η ) s = η ( 3 + u 2 + v 2 ) = 2 η v ℓ (P_\eta)_s=\eta(3+u^2+v^2)=2\eta v\ell ( P η ) s = η ( 3 + u 2 + v 2 ) = 2 η v ℓ and integrates the matrix entries to ( L + K S , K , m − 2 L S − K S 2 ) (L+KS,K,m-2LS-KS^2) ( L + K S , K , m − 2 L S − K S 2 ) . For a single ray, substitute d s = d e / ( κ 2 e 2 ) ds=de/(\kappa^2e^2) d s = d e / ( κ 2 e 2 ) to obtain (A.8). Reconstruction is invertible since v > 0 v>0 v > 0 . Conversely these formulas give the matrix adjoint and the p v p_v p v adjoint equation; differentiating the imposed active zero energy then gives v ( p u ′ + H u ) = 0 v(p'_u+H_u)=0 v ( p u ′ + H u ) = 0 , hence the remaining state-adjoint equation. At a binary switch χ = 0 \chi=0 χ = 0 makes the two reconstructions of p u p_u p u agree. Continuity, oriented telescoping and inclusion–exclusion prove the remaining assertions. □
Lemma A.6. On a normal lift normalized to η = 1 \eta=1 η = 1 , the second switch of 202 has u > 0 u>0 u > 0 , and the first switch of 020 has u < 0 u<0 u < 0 . Their cyclic versions give the corresponding strict signs of
u 2 + v 2 − 2 u − 3 and 3 − u 2 − v 2 − 2 u . u^2+v^2-2u-3\qquad\text{and}\qquad3-u^2-v^2-2u. u 2 + v 2 − 2 u − 3 and 3 − u 2 − v 2 − 2 u . Proof. At the two switches of 202, u 1 < u 2 u_1<u_2 u 1 < u 2 and B e ( u 1 ) ≥ 0 ≥ B e ( u 2 ) B_e(u_1)\ge0\ge B_e(u_2) B e ( u 1 ) ≥ 0 ≥ B e ( u 2 ) . If u 2 ≤ 0 u_2\le0 u 2 ≤ 0 , then B e ( u 2 ) − B e ( u 1 ) = − ( u 2 − u 1 ) ( u 2 + u 1 ) / 2 > 0 B_e(u_2)-B_e(u_1)=-(u_2-u_1)(u_2+u_1)/2>0 B e ( u 2 ) − B e ( u 1 ) = − ( u 2 − u 1 ) ( u 2 + u 1 ) /2 > 0 , a contradiction. The other assertion follows with the signs reversed. Finally
R R ( W ) = u 2 + v 2 − 2 u − 3 ( 1 + u ) 2 + v 2 , R R 2 ( W ) = 3 − u 2 − v 2 − 2 u ( 1 − u ) 2 + v 2 \mathfrak{R}\mathfrak{R}(W)=\frac{u^2+v^2-2u-3}{(1+u)^2+v^2},\qquad\mathfrak{R}\mathfrak{R}^{2}(W)=\frac{3-u^2-v^2-2u}{(1-u)^2+v^2} RR ( W ) = ( 1 + u ) 2 + v 2 u 2 + v 2 − 2 u − 3 , R R 2 ( W ) = ( 1 − u ) 2 + v 2 3 − u 2 − v 2 − 2 u has positive denominators and transports the result legally. □
Lemma A.7 (Signed capped-pair comparison). On one selected nonzero full free-time maximizing lift, retain an actual positive source 1 ∣ 02 ∣ 1 1\mid02\mid1 1 ∣ 02 ∣ 1 or 1 ∣ 20 ∣ 1 1\mid20\mid1 1 ∣ 20 ∣ 1 , with complete interior arcs and both actual third-mode ties. Set δ = 1 \delta=1 δ = 1 or − 1 -1 − 1 , respectively. At its middle put
x = 1 + u v , y = u 2 + v 2 − 1 2 v , z = 1 − u v , x y + x z + y z = 1. x=\frac{1+u}{v},\qquad y=\frac{u^2+v^2-1}{2v},\qquad z=\frac{1-u}{v},\qquad xy+xz+yz=1. x = v 1 + u , y = 2 v u 2 + v 2 − 1 , z = v 1 − u , x y + x z + y z = 1. For the positive physical mode-0,2 durations d x , d z d_x,d_z d x , d z , set P = e δ d x / x P=e^{\delta d_x/x} P = e δ d x / x , R = e δ d z / z R=e^{\delta d_z/z} R = e δ d z / z . These are signed endpoint parameters of the actual durations remain positive. With U = tr ( ( Λ − η A ∗ ) X ) U=\operatorname{tr}((\Lambda-\eta A_*)X) U = tr (( Λ − η A ∗ ) X ) , put ∇ H = v ∇ \nabla_H=v\nabla ∇ H = v ∇ and V = ∇ H U + J ( v p ) V=\nabla_HU+J(vp) V = ∇ H U + J ( v p ) , fixing Λ , η \Lambda,\eta Λ , η in the gradient, with J ( a , b ) = ( − b , a ) J(a,b)=(-b,a) J ( a , b ) = ( − b , a ) . At the middle v p = ( h , 0 ) vp=(h,0) v p = ( h , 0 ) , h ≥ 0 h\ge0 h ≥ 0 , U = − h U=-h U = − h and δ V y ≥ 0 \delta V_y\ge0 δ V y ≥ 0 . Define
ν x = ( x 2 − 1 , 2 x ) 1 + x 2 , ν z = ( z 2 − 1 , − 2 z ) 1 + z 2 , f ( g ) = g log g g − 1 − 1 , \nu_x=\frac{(x^2-1,2x)}{1+x^2},\qquad\nu_z=\frac{(z^2-1,-2z)}{1+z^2},\qquad f(g)=\frac{g\log g}{g-1}-1, ν x = 1 + x 2 ( x 2 − 1 , 2 x ) , ν z = 1 + z 2 ( z 2 − 1 , − 2 z ) , f ( g ) = g − 1 g log g − 1 ,
A x = ( P − 1 ) ( 1 + x 2 ) 2 , β x = x [ x ( P − 1 ) x + y − f ( P ) ] , b x = 2 x [ log P P − 1 − z + x / P x + z ] , A_x=\frac{(P-1)(1+x^2)}{2},\qquad\beta_x=x\left[\frac{x(P-1)}{x+y}-f(P)\right],\qquad b_x=2x\left[\frac{\log P}{P-1}-\frac{z+x/P}{x+z}\right], A x = 2 ( P − 1 ) ( 1 + x 2 ) , β x = x [ x + y x ( P − 1 ) − f ( P ) ] , b x = 2 x [ P − 1 log P − x + z z + x / P ] ,
and define A z , β z , b z A_z,\beta_z,b_z A z , β z , b z by ( x , z , P ) ↔ ( z , x , R ) (x,z,P)\leftrightarrow(z,x,R) ( x , z , P ) ↔ ( z , x , R ) . The unnormalized actual cap rows are
h = A x ( V ⋅ J ν x + 2 η β x ) = A z ( V ⋅ J ν z + 2 η β z ) . (225) h=A_x(V\cdot J\nu_x+2\eta\beta_x)=A_z(V\cdot J\nu_z+2\eta\beta_z). \tag*{(225)} h = A x ( V ⋅ J ν x + 2 η β x ) = A z ( V ⋅ J ν z + 2 η β z ) . (225)
Put
D = x ( P − 1 ) + z ( R − 1 ) − ( x + y ) f ( P ) − ( y + z ) f ( R ) , D=x(P-1)+z(R-1)-(x+y)f(P)-(y+z)f(R), D = x ( P − 1 ) + z ( R − 1 ) − ( x + y ) f ( P ) − ( y + z ) f ( R ) ,
q 0 = 2 x z D y ( x + z ) , r = 2 { z / ( P − 1 ) + x / ( R − 1 ) } y ( x + z ) 2 . q_0=\frac{2xzD}{y(x+z)},\qquad r=\frac{2\{z/(P-1)+x/(R-1)\}}{y(x+z)^2}. q 0 = y ( x + z ) 2 x z D , r = y ( x + z ) 2 2 { z / ( P − 1 ) + x / ( R − 1 )} .
Then V y = η q 0 − r h V_y=\eta q_0-rh V y = η q 0 − r h , with r > 0 r>0 r > 0 if δ = 1 \delta=1 δ = 1 and r < − 2 / ( 1 − x z ) < − 2 r<-2/(1-xz)<-2 r < − 2/ ( 1 − x z ) < − 2 if δ = − 1 \delta=-1 δ = − 1 . The selected lift is normal. After joint normalization, its actual cap lies in the algebraic segment
0 ≤ h ≤ h ˉ : = δ q 0 − M δ r ≤ h ∗ : = q 0 r , M = max ( 0 , δ b x , δ b z ) , δ q 0 ≥ M . (226) 0\le h\le\bar h:=\frac{\delta q_0-M}{\delta r}\le h_*:=\frac{q_0}{r},\qquad M=\max(0,\delta b_x,\delta b_z),\qquad\delta q_0\ge M. \tag*{(226)} 0 ≤ h ≤ h ˉ := δ r δ q 0 − M ≤ h ∗ := r q 0 , M = max ( 0 , δ b x , δ b z ) , δ q 0 ≥ M . (226)
Writing ℓ = x + y + z \ell=x+y+z ℓ = x + y + z and N = ( ( x − z ) / ( x + z ) , y − ( x 2 + z 2 ) / ( x + z ) ) N=((x-z)/(x+z),y-(x^2+z^2)/(x+z)) N = (( x − z ) / ( x + z ) , y − ( x 2 + z 2 ) / ( x + z )) , the exact polarization is
C = 3 4 − N ⋅ V 2 − ( ℓ − N y ) h 2 − ∣ V ∣ 2 4 + h V y 2 , tr ( Λ X ) = ℓ − h . (227) \mathcal{C}=\frac{3}{4}-\frac{N\cdot V}{2}-\frac{(\ell-N_y)h}{2}-\frac{|V|^2}{4}+\frac{hV_y}{2},\qquad\operatorname{tr}(\Lambda X)=\ell-h. \tag*{(227)} C = 4 3 − 2 N ⋅ V − 2 ( ℓ − N y ) h − 4 ∣ V ∣ 2 + 2 h V y , tr ( Λ X ) = ℓ − h . (227)
It is strictly concave on the cap line, so the actual determinant is bounded below by the minimum of its values at 0 , h ˉ 0,\bar h 0 , h ˉ , or more coarsely by the minimum at 0 , h ∗ 0,h_* 0 , h ∗ . In the latter pair h V y = 0 hV_y=0 h V y = 0 . These are algebraic comparison endpoints, not independently normalized or calibrated trajectories. The sum I I I of the two complete physical costs is
I = δ [ ( x + z ) D + f ( P ) + f ( R ) + x 2 f ( P ) P + x y ( 1 − P − 1 ) + z 2 f ( R ) R + z y ( 1 − R − 1 ) ] . (228) I=\delta\left[(x+z)D+f(P)+f(R)+\frac{x^2f(P)}{P}+xy(1-P^{-1})
+\frac{z^2f(R)}{R}+zy(1-R^{-1})\right]. \tag*{(228)} I = δ [ ( x + z ) D + f ( P ) + f ( R ) + P x 2 f ( P ) + x y ( 1 − P − 1 ) + R z 2 f ( R ) + z y ( 1 − R − 1 ) ] . (228)
After multiplication by δ \delta δ every term is nonnegative and the six non-residual terms are positive.
Proof. The binary tie gives p v = 0 p_v=0 p v = 0 and the active zero energy gives U = − h U=-h U = − h . The third Hamiltonian inequality gives 2 h v 2 / ( u 2 + v 2 − 1 ) ≥ 0 2hv^2/(u^2+v^2-1)\ge0 2 h v 2 / ( u 2 + v 2 − 1 ) ≥ 0 . Write K = K e K=K_e K = K e . Differentiating U U U at fixed Λ , η \Lambda,\eta Λ , η and using that energy yields
K = η ( 3 + u 2 ) − v V y 2 , L = V x + η u 2 , m = − η 2 + h − V y / 2 v , K=\frac{\eta(3+u^2)-vV_y}{2},\qquad L=\frac{V_x+\eta u}{2},\qquad m=-\frac{\eta}{2}+\frac{h-V_y/2}{v}, K = 2 η ( 3 + u 2 ) − v V y , L = 2 V x + η u , m = − 2 η + v h − V y /2 ,
l = L + u m , b = K − 2 u L − u 2 m , p = ( h / v , 0 ) . l=L+um,\qquad b=K-2uL-u^2m,\qquad p=(h/v,0). l = L + u m , b = K − 2 uL − u 2 m , p = ( h / v , 0 ) .
Thus χ t = − V y \chi_t=-V_y χ t = − V y ; its actual maximizing crossing gives δ V y ≥ 0 \delta V_y\ge0 δ V y ≥ 0 . The mode-zero and mode-two cap endpoints and signed shears from the middle are
( u A , v A ) = ( ( 1 + u ) / P − 1 , v / P ) , s A = − x ( P − 1 ) / v , (u_A,v_A)=((1+u)/P-1,v/P),\qquad s_A=-x(P-1)/v, ( u A , v A ) = (( 1 + u ) / P − 1 , v / P ) , s A = − x ( P − 1 ) / v ,
( u B , v B ) = ( 1 − ( 1 − u ) / R , v / R ) , s B = z ( R − 1 ) / v . (u_B,v_B)=(1-(1-u)/R,v/R),\qquad s_B=z(R-1)/v. ( u B , v B ) = ( 1 − ( 1 − u ) / R , v / R ) , s B = z ( R − 1 ) / v .
Integrating (A.5) with d s = x 2 d u / ( 1 + u ) 2 ds=x^2du/(1+u)^2 d s = x 2 d u / ( 1 + u ) 2 or z 2 d u / ( 1 − u ) 2 z^2du/(1-u)^2 z 2 d u / ( 1 − u ) 2 gives
m j = m − 2 L s j − K s j 2 , χ A = x ( P − 1 ) ( V y − η b x ) , χ B = − z ( R − 1 ) ( V y − η b z ) . m_j=m-2Ls_j-Ks_j^2,\qquad\chi_A=x(P-1)(V_y-\eta b_x),\qquad\chi_B=-z(R-1)(V_y-\eta b_z). m j = m − 2 L s j − K s j 2 , χ A = x ( P − 1 ) ( V y − η b x ) , χ B = − z ( R − 1 ) ( V y − η b z ) .
Eliminating p u p_u p u from the active energy and the third-mode tie gives
m j + η 2 = 2 K − η ( 3 + u j 2 ) 2 v j 2 + 1 2 ( e j / v j 2 + e j − 1 ) χ j , e A = 1 + u A , e B = u B − 1. m_j+\frac{\eta}{2}=\frac{2K-\eta(3+u_j^2)}{2v_j^2}+\frac{1}{2}(e_j/v_j^2+e_j^{-1})\chi_j,\qquad e_A=1+u_A,\quad e_B=u_B-1. m j + 2 η = 2 v j 2 2 K − η ( 3 + u j 2 ) + 2 1 ( e j / v j 2 + e j − 1 ) χ j , e A = 1 + u A , e B = u B − 1.
Substitution reduces the two residuals to ( x + z ) / 2 (x+z)/2 ( x + z ) /2 times the corresponding row of (A.10). No multiplier was normalized in this calculation. The normal determinant is 2 y ( x + z ) 2 / [ ( 1 + x 2 ) ( 1 + z 2 ) ] > 0 2y(x+z)^2/[(1+x^2)(1+z^2)]>0 2 y ( x + z ) 2 / [( 1 + x 2 ) ( 1 + z 2 )] > 0 . Using
y ( x + z ) 2 ( 1 , 0 ) + z ( 1 + x 2 ) ν x + x ( 1 + z 2 ) ν z = 0 y(x+z)^2(1,0)+z(1+x^2)\nu_x+x(1+z^2)\nu_z=0 y ( x + z ) 2 ( 1 , 0 ) + z ( 1 + x 2 ) ν x + x ( 1 + z 2 ) ν z = 0
therefore gives the stated solved line and signs of r r r . If η = 0 \eta=0 η = 0 , the signs h ≥ 0 h\geq0 h ≥ 0 , δ V y ≥ 0 \delta V_y\geq0 δ V y ≥ 0 force h = V y = 0 h=V_y=0 h = V y = 0 ; either cap row gives V x = 0 V_x=0 V x = 0 . The displayed reconstruction annihilates all matrix and state covectors. Homogeneous adjoint uniqueness propagates zero on the original interval, contradicting nontriviality. Normalize jointly. The actual endpoint signs χ A ≥ 0 \chi_A\geq0 χ A ≥ 0 , χ B ≤ 0 \chi_B\leq0 χ B ≤ 0 now give (A.11), including its zero-height cases without division by q 0 q_0 q 0 .
Direct matrix expansion gives the general identity
det Λ = 3 η 2 4 + η ( ℓ U − N ⋅ ∇ H U ) 2 + U 2 − ∣ ∇ H U ∣ 2 4 . \det\Lambda=\frac{3\eta^2}{4}+\frac{\eta(\ell U-N\cdot\nabla_HU)}{2}+\frac{U^2-\lvert\nabla_HU\rvert^2}{4}. det Λ = 4 3 η 2 + 2 η ( ℓ U − N ⋅ ∇ H U ) + 4 U 2 − ∣ ∇ H U ∣ 2 .
Use U = − h U=-h U = − h , ∇ H U = V − ( 0 , h ) \nabla_HU=V-(0,h) ∇ H U = V − ( 0 , h ) after normalization to obtain (A.12). If V x = p 0 + α h V_x=p_0+\alpha h V x = p 0 + α h , its quadratic coefficient is − ( α 2 + r 2 + 2 r ) / 4 < 0 -(\alpha^2+r^2+2r)/4<0 − ( α 2 + r 2 + 2 r ) /4 < 0 , proving both endpoint comparisons. Finally, integrating the physical density gives
I / δ = x [ ( x + y ) ( 1 − P − 1 ) + ( x + z ) ( P − 1 ) − x log P ] + z [ ( y + z ) ( 1 − R − 1 ) + ( x + z ) ( R − 1 ) − z log R ] . I/\delta=x[(x+y)(1-P^{-1})+(x+z)(P-1)-x\log P]
+z[(y+z)(1-R^{-1})+(x+z)(R-1)-z\log R]. I / δ = x [( x + y ) ( 1 − P − 1 ) + ( x + z ) ( P − 1 ) − x log P ] + z [( y + z ) ( 1 − R − 1 ) + ( x + z ) ( R − 1 ) − z log R ] .
For δ = − 1 \delta=-1 δ = − 1 both integrals have reversed endpoint order; multiplication by δ \delta δ restores the actual positive physical costs. Expanding D D D and using x y + x z + y z = 1 xy+xz+yz=1 x y + x z + y z = 1 proves (A.13). Since f ( 1 ) = 0 f(1)=0 f ( 1 ) = 0 by continuity and f ′ ( g ) = ( g − 1 − log g ) / ( g − 1 ) 2 > 0 f'(g)=(g-1-\log g)/(g-1)^2>0 f ′ ( g ) = ( g − 1 − log g ) / ( g − 1 ) 2 > 0 for g ≠ 1 g\ne1 g = 1 , f ( P ) f(P) f ( P ) , f ( R ) f(R) f ( R ) , 1 − P − 1 1-P^{-1} 1 − P − 1 , 1 − R − 1 1-R^{-1} 1 − R − 1 all have sign δ \delta δ ; the residual sign follows from (A.11). □
Normality of the selected six-arc lift.
Theorem A.8. If an interval contains six consecutive positive maximal arcs and avoids the three words of Lemma A.4 , every selected nonzero full free-time maximizing lift on that interval is normal.
Proof. Suppose η = 0 \eta=0 η = 0 . On a binary interval, (A.5) reduces to χ s = 2 K e \chi_s=2K_e χ s = 2 K e , with K e K_e K e constant. A tie with a third-mode cap gives, after eliminating p u p_u p u from the active Hamiltonian,
m = K e v 2 + A j χ , A 0 = 1 2 ( 1 + u v 2 + 1 1 + u ) , A 2 = − 1 2 ( 1 − u v 2 + 1 1 − u ) . (229) m=\frac{K_e}{v^2}+A_j\chi,\qquad A_0=\frac{1}{2}\left(\frac{1+u}{v^2}+\frac{1}{1+u}\right),\qquad A_2=-\frac{1}{2}\left(\frac{1-u}{v^2}+\frac{1}{1-u}\right). \tag*{(229)} m = v 2 K e + A j χ , A 0 = 2 1 ( v 2 1 + u + 1 + u 1 ) , A 2 = − 2 1 ( v 2 1 − u + 1 − u 1 ) . (229) The signed capped-pair comparison, Lemma A.7, excludes every abnormal four-letter A B C A ABCA A B C A with three distinct labels: a cyclic relabeling makes it 1021 or 1201, both covered before normalization.
Two other short arguments suffice. In every A B A ABA A B A block, the two distinct zeros of affine χ \chi χ imply K e = χ = 0 K_e=\chi=0 K e = χ = 0 on the full block. In A B A C A ABACA A B A C A , the A B A ABA A B A and A C A ACA A C A blocks overlap on a positive A A A arc, where two independent Hamiltonian differences vanish. Indeed
det ( F 0 − F 2 , F 1 − F 2 ) = − 4 v 5 ( u − 1 ) ( u + 1 ) S ≠ 0. \det(F_0-F_2,F_1-F_2)=\frac{-4v^5}{(u-1)(u+1)S}\ne0. det ( F 0 − F 2 , F 1 − F 2 ) = ( u − 1 ) ( u + 1 ) S − 4 v 5 = 0. Thus p = 0 p=0 p = 0 on that arc. Its adjoint and active Hamiltonian give U = U u = U v = 0 U=U_u=U_v=0 U = U u = U v = 0 . The linear map Λ ↦ ( U , U u , U v ) \Lambda\mapsto(U,U_u,U_v) Λ ↦ ( U , U u , U v ) has determinant 4 / v 2 4/v^2 4/ v 2 , so Λ = 0 \Lambda=0 Λ = 0 as well. Finally, for a binary core with at least two switches and third-mode caps at both ends, K e = χ = 0 K_e=\chi=0 K e = χ = 0 throughout the core; the two cap equations give m = 0 m=0 m = 0 at distinct ends. Since L L L is constant and m s = − 2 L m_s=-2L m s = − 2 L , m = L = 0 m=L=0 m = L = 0 . This excludes A B C B A ABCBA A B C B A and A B C B C A ABCBC A A B C B C A .
It remains a short literal classification. After adjacent equal letters, the three hard four-letter words, and the patterns A B C A ABCA A B C A , A B A C A ABACA A B A C A , A B C B A ABCBA A B C B A are excluded, the possible five-letter words are
02021 , 02121 , 10102 , 10202 , 21010 , 21210. 02021,\quad02121,\quad10102,\quad10202,\quad21010,\quad21210. 02021 , 02121 , 10102 , 10202 , 21010 , 21210. The first, third, and sixth have no extension satisfying the same restrictions. The others extend only to 021210, 102021, and 210102, respectively. Each is an A B C B C A ABCBC A A B C B C A word. This list can also be checked by fixing the first letter and distinguishing whether the third
equals the first; each subsequent letter has two possibilities and the displayed shorter patterns remove all omitted branches. Thus six positive arcs are impossible for an abnormal lift. Homogeneous uniqueness propagates the vanishing found in each case, contradicting nontriviality. □ \square □
No compactness or quantitative cost bound enters this theorem. In particular, it supplies normality on the very interval whose lift is later used.
A six-letter exclusion without a span bound.
Lemma A.9. The words 202101, 010212, 121020 are impossible on a full maximizing lift with positive arcs and caps. The conclusion holds for arbitrary positive durations.
Proof. Consider 202101 and its switches t 1 < ⋯ < t 5 t_1 < \cdots< t_5 t 1 < ⋯ < t 5 . Normality follows from Theorem A.8 ; normalize jointly. On the mode 2 tail after the left 202, Lemma A.6 gives u > 0 u > 0 u > 0 . Let y = − χ y = -\chi y = − χ and z = y ′ = − 2 B e / v z = y' = -2B_e/v z = y ′ = − 2 B e / v . The rate b = v / ( 1 − u ) > 0 b = v/(1-u) > 0 b = v / ( 1 − u ) > 0 is constant there, and
z ′ − b z = 2 u > 0 , y ( t 2 ) = 0 , z ( t 2 ) ≥ 0. z' - bz = 2u > 0,\qquad y(t_2) = 0,\qquad z(t_2) \ge0. z ′ − b z = 2 u > 0 , y ( t 2 ) = 0 , z ( t 2 ) ≥ 0.
For d − = t 3 − t 2 > 0 d_- = t_3 - t_2 > 0 d − = t 3 − t 2 > 0 , integration of the strict differential inequality yields
0 < y ( t 3 ) < 1 − e − b d − b z ( t 3 ) . 0 < y(t_3) < \frac{1-e^{-bd_-}}{b}z(t_3). 0 < y ( t 3 ) < b 1 − e − b d − z ( t 3 ) .
Since w ′ / w = − c w'/w = -c w ′ / w = − c on mode 2, where c = 2 v / S c = 2v/S c = 2 v / S , at the central tie
σ 0 < 0 , σ 0 ′ < ( c − b 1 − e − b d − ) ∣ σ 0 ∣ . (230) \sigma_0 < 0,\qquad\sigma'_0 < \left(c-\frac{b}{1-e^{-bd_-}}\right)|\sigma_0|. \tag*{(230)} σ 0 < 0 , σ 0 ′ < ( c − 1 − e − b d − b ) ∣ σ 0 ∣. (230)
For the right tail use time reversal together with
ψ ( W ) = 3 − W ‾ 1 + W ‾ , Q = 1 2 ( − 1 3 1 1 ) , Q 2 = I . \psi(W)=\frac{3-\overline{W}}{1+\overline{W}},\qquad Q=\frac{1}{2}\begin{pmatrix}-1&3\\1&1\end{pmatrix},\qquad Q^2=I. ψ ( W ) = 1 + W 3 − W , Q = 2 1 ( − 1 1 3 1 ) , Q 2 = I .
The transported lift is p n e w = − ( D ψ ) − t p p_{\mathrm{new}}=-(D\psi)^{-t}p p new = − ( D ψ ) − t p , Λ n e w = − Q Λ Q \Lambda_{\mathrm{new}}=-Q\Lambda Q Λ new = − Q Λ Q , η n e w = η \eta_{\mathrm{new}}=\eta η new = η . Substitution gives the label permutation ( 1 2 ) (1\ 2) ( 1 2 ) , b ( ψ W ) = c ( W ) b(\psi W)=c(W) b ( ψ W ) = c ( W ) , c ( ψ W ) = b ( W ) c(\psi W)=b(W) c ( ψ W ) = b ( W ) , and
( σ 0 , σ 1 , σ 2 ) n e w = ( − σ 1 , − σ 0 , − σ 2 ) o l d . (\sigma_0,\sigma_1,\sigma_2)_{\mathrm{new}}=(-\sigma_1,-\sigma_0,-\sigma_2)_{\mathrm{old}}. ( σ 0 , σ 1 , σ 2 ) new = ( − σ 1 , − σ 0 , − σ 2 ) old .
The reversal cancels the derivative sign. Applying (A.15) to the reversed right 101 with d + = t 4 − t 3 d_+ = t_4-t_3 d + = t 4 − t 3 gives
σ 1 ′ < ( b − c 1 − e − c d + ) ∣ σ 1 ∣ . \sigma'_1 < \left(b-\frac{c}{1-e^{-cd_+}}\right)|\sigma_1|. σ 1 ′ < ( b − 1 − e − c d + c ) ∣ σ 1 ∣.
At the central 21 tie, q = σ 1 = − σ 0 > 0 q=\sigma_1=-\sigma_0>0 q = σ 1 = − σ 0 > 0 . Therefore
σ 0 ′ + σ 1 ′ < − ( b e − b d − 1 − e − b d − + c e − c d + 1 − e − c d + ) q < 0. (231) \sigma'_0+\sigma'_1<-\left(\frac{be^{-bd_-}}{1-e^{-bd_-}}+\frac{ce^{-cd_+}}{1-e^{-cd_+}}\right)q<0. \tag*{(231)} σ 0 ′ + σ 1 ′ < − ( 1 − e − b d − b e − b d − + 1 − e − c d + c e − c d + ) q < 0. (231)
But the actual crossing has σ 2 ′ ≤ 0 \sigma'_2\le0 σ 2 ′ ≤ 0 , whereas the zero-sum identity gives σ 0 ′ + σ 1 ′ = − σ 2 ′ ≥ 0 \sigma'_0+\sigma'_1=-\sigma'_2\ge0 σ 0 ′ + σ 1 ′ = − σ 2 ′ ≥ 0 . Cyclic transport proves the two remaining words. □ \square □
The final two-tail contradiction applies at any actual 2 → 1 2\to1 2 → 1 tie of one full maximizing lift with q = σ 1 = − σ 0 > 0 q=\sigma_1=-\sigma_0>0 q = σ 1 = − σ 0 > 0 and the two displayed tail inequalities for positive rates and durations; it does not require either tail to have arisen from an A B A ABA A B A block.
Twisted counts and necessary run data
On the extended trajectory let w i w_i w i be the consecutive maximal-arc labels. A physical period has B B B essential switches and
w i + B = w i + 1 ( m o d 3 ) . (232) w_{i+B}=w_i+1 \pmod{3}. \tag*{(232)} w i + B = w i + 1 ( mod 3 ) . (232)
This counts arcs intrinsically on the twisted circle. Cutting can split at most one such arc, proving (16) . The increments δ i = w i + 1 − w i \delta_i=w_{i+1}-w_i δ i = w i + 1 − w i lie in { 1 , 2 } \{1,2\} { 1 , 2 } . If both occur, write the alternating positive and reverse run lengths as a i , b i a_i,b_i a i , b i , 1 ≤ i ≤ M 1\le i\le M 1 ≤ i ≤ M . The hard four-letter words forbid a reverse run of length one: its surrounding increments 121 are exactly a forbidden four-letter pattern. The six-letter exclusion forbids a reverse run of length three: its surrounding increments 12221 are exactly the corresponding six-letter pattern. Thus
a i ≥ 1 , b i ≥ 2 , b i ≠ 3 , ∑ i ( a i + b i ) = B . (233) a_i\ge1,\qquad b_i\ge2,\qquad b_i\ne3,\qquad\sum_i(a_i+b_i)=B. \tag*{(233)} a i ≥ 1 , b i ≥ 2 , b i = 3 , i ∑ ( a i + b i ) = B . (233)
If L = ∑ i a i L=\sum_i a_i L = ∑ i a i , the twist also gives L + 2 ( B − L ) ≡ 1 ( m o d 3 ) L+2(B-L)\equiv1\pmod{3} L + 2 ( B − L ) ≡ 1 ( mod 3 ) . Sorting a i a_i a i and b i b_i b i separately gives an unordered necessary key ( a a , b b ) (aa,bb) ( aa , bb ) . Actual incidence information is forgotten, so this is an upper cover, not a realization theorem. Bounds used for a key must hold for all of its actual occurrence orders.
A harmonic calibration of the matrix costate
Fix one normal full lift on a genuine fundamental interval, normalized jointly to η = 1 \eta=1 η = 1 . Its Casimir
C = det Λ = − 1 2 tr ( Λ 2 ) \mathcal{C}=\det\Lambda=-\frac{1}{2}\operatorname{tr}(\Lambda^2) C = det Λ = − 2 1 tr ( Λ 2 )
is constant by the Lax equation. In this section assume C > 0 \mathcal{C}>0 C > 0 and tr ( Λ X ) > 0 \operatorname{tr}(\Lambda X)>0 tr ( Λ X ) > 0 at one actual point. Both hypotheses must be established before the following bound is used.
Lemma A.10 (The future sheet). Set g = D ∗ − 1 G D ∗ g=D_*^{-1}GD_* g = D ∗ − 1 G D ∗ , X ^ = D ∗ − 1 X D ∗ \widehat{X}=D_*^{-1}XD_* X = D ∗ − 1 X D ∗ , and Λ ^ = D ∗ − 1 Λ D ∗ \widehat{\Lambda}=D_*^{-1}\Lambda D_* Λ = D ∗ − 1 Λ D ∗ . With H = diag ( 1 , − 1 ) H=\operatorname{diag}(1,-1) H = diag ( 1 , − 1 ) and S 0 = ( 0 1 1 0 ) S_0=\begin{pmatrix}0&1\\1&0\end{pmatrix} S 0 = ( 0 1 1 0 ) , write
X ^ = x J + r 1 H + r 2 S 0 , Y = − Λ ^ C = z J + y 1 H + y 2 S 0 . \widehat{X}=xJ+r_1H+r_2S_0,\qquad Y=-\frac{\widehat{\Lambda}}{\sqrt{\mathcal{C}}}=zJ+y_1H+y_2S_0. X = x J + r 1 H + r 2 S 0 , Y = − C Λ = z J + y 1 H + y 2 S 0 . Then
x = 3 + u 2 + v 2 2 3 v ≥ 1 , x 2 − ∣ r ∣ 2 = z 2 − ∣ y ∣ 2 = 1 , z ≥ 1 , x=\frac{3+u^2+v^2}{2\sqrt{3}v}\ge1,\qquad x^2-\lvert r\rvert^2=z^2-\lvert y\rvert^2=1,\qquad z\ge1, x = 2 3 v 3 + u 2 + v 2 ≥ 1 , x 2 − ∣ r ∣ 2 = z 2 − ∣ y ∣ 2 = 1 , z ≥ 1 , and, with κ = 2 C \kappa=2\sqrt{\mathcal{C}} κ = 2 C ,
q = z x − y ⋅ r = tr ( Λ X ) κ ≥ 1 , h p : = max j t F j = 3 x − κ q ≥ 0. (234) q=zx-y\cdot r=\frac{\operatorname{tr}(\Lambda X)}{\kappa}\ge1,\qquad h_p:=\max_j{}^tF_j=\sqrt{3}x-\kappa q\ge0. \tag*{(234)} q = z x − y ⋅ r = κ tr ( Λ X ) ≥ 1 , h p := j max t F j = 3 x − κ q ≥ 0. (234) Proof. The determinant identities are direct. On the sheet z ≤ − 1 z\le-1 z ≤ − 1 , Lorentz Cauchy gives z x − y ⋅ r < 0 zx-y\cdot r<0 z x − y ⋅ r < 0 , contradicting the specified positive trace. Continuity preserves the sheet z ≥ 1 z\ge1 z ≥ 1 , and on this sheet Lorentz Cauchy gives q ≥ 1 q\ge1 q ≥ 1 . Explicitly write x = cosh s x=\cosh s x = cosh s , ∣ r ∣ = sinh s \lvert r\rvert=\sinh s ∣ r ∣ = sinh s , z = cosh t z=\cosh t z = cosh t , ∣ y ∣ = sinh t \lvert y\rvert=\sinh t ∣ y ∣ = sinh t ; then q ≥ cosh ( t − s ) q\ge\cosh(t-s) q ≥ cosh ( t − s ) .
The fields have the strictly positive equilibrium weights
ω 0 = ( 1 + u ) ( v 2 + ( 1 − u ) 2 ) 4 v 2 , ω 1 = v 2 2 v 2 , ω 2 = ( 1 − u ) ( v 2 + ( 1 + u ) 2 ) 4 v 2 . \omega_0=\frac{(1+u)(v^2+(1-u)^2)}{4v^2},\qquad
\omega_1=\frac{v^2}{2v^2},\qquad
\omega_2=\frac{(1-u)(v^2+(1+u)^2)}{4v^2}. ω 0 = 4 v 2 ( 1 + u ) ( v 2 + ( 1 − u ) 2 ) , ω 1 = 2 v 2 v 2 , ω 2 = 4 v 2 ( 1 − u ) ( v 2 + ( 1 + u ) 2 ) . They sum to one and ∑ j ω j F j = 0 \sum_j\omega_jF_j=0 ∑ j ω j F j = 0 . Therefore h p ≥ 0 h_p\ge0 h p ≥ 0 , and the zero maximum Hamiltonian gives the equality in (A.19).
Lemma A.11 (The real rotation identity). The legal three-copy extension of this same matrix lift satisfies
∫ 0 3 T x d t = A , π = ∫ 0 3 T x + q z + 1 d t . (235) \int_0^{3T}x\,dt=\mathcal{A},\qquad
\pi=\int_0^{3T}\frac{x+q}{z+1}\,dt. \tag*{(235)} ∫ 0 3 T x d t = A , π = ∫ 0 3 T z + 1 x + q d t . (235) No periodicity of the state costate is assumed.
Proof. The Lax equation gives ( g Λ ^ g − 1 ) ′ = 0 (g\widehat{\Lambda}g^{-1})'=0 ( g Λ g − 1 ) ′ = 0 , hence Λ ^ ( T ) = R − 1 Λ ^ ( 0 ) R \widehat{\Lambda}(T)=R^{-1}\widehat{\Lambda}(0)R Λ ( T ) = R − 1 Λ ( 0 ) R with R = e π J / 3 R=e^{\pi J/3} R = e π J /3 . Extend Λ ^ ( t + k T ) = R − k Λ ^ ( t ) R k \widehat{\Lambda}(t+kT)=R^{-k}\widehat{\Lambda}(t)R^k Λ ( t + k T ) = R − k Λ ( t ) R k . This is its continuous Lax transport on the extended state trajectory. Since R R R commutes with J J J , the scalars x , z , q x,z,q x , z , q are T T T -periodic. The scalar Hamiltonian inequality extends in this way without extending p p p . The area formula gives the first identity.
Choose S ∈ S L 2 ( R ) S\in\mathrm{SL}_{2}(\mathbb{R}) S ∈ SL 2 ( R ) with Y ( 0 ) = S J S − 1 Y(0)=SJS^{-1} Y ( 0 ) = S J S − 1 , and set f = g e − 1 S f=g_e^{-1}S f = g e − 1 S . Then Y = f J f − 1 Y=fJf^{-1} Y = f J f − 1 , f ′ = − X ^ e f f'=-\widehat{X}_ef f ′ = − X e f , and f ( 3 T ) = − f ( 0 ) f(3T)=-f(0) f ( 3 T ) = − f ( 0 ) . The original real homotopy class implies that the Iwasawa angle of f f f changes by − π -\pi − π . To see the exact integer, homotope g e g_e g e to e π J t / ( 3 T ) e^{\pi Jt/(3T)} e π J t / ( 3 T ) , invert and multiply by S S S , and then move S S S continuously to I I I . The odd multiple of π \pi π cannot change; at S = I S=I S = I it is − π -\pi − π .
Write
f = ( 1 ξ 0 1 ) ( h 0 0 1 / h ) e J θ , h > 0. f=\begin{pmatrix}1&\xi\\0&1\end{pmatrix}
\begin{pmatrix}\sqrt{h}&0\\0&1/\sqrt{h}\end{pmatrix}
e^{J\theta},\qquad h>0. f = ( 1 0 ξ 1 ) ( h 0 0 1/ h ) e J θ , h > 0.
The closed hyperbolic projection has ξ + i h = ( y 1 + i ) / ( z + y 2 ) \xi+ih=(y_1+i)/(z+y_2) ξ + ih = ( y 1 + i ) / ( z + y 2 ) . Its connection form is
− 1 2 tr ( J f − 1 d f ) = d θ − 1 2 d ξ / h . -\frac{1}{2}\operatorname{tr}(Jf^{-1}df)=d\theta-\frac{1}{2}\,d\xi/h. − 2 1 tr ( J f − 1 df ) = d θ − 2 1 d ξ / h .
On the positive hyperboloid, introduce the globally smooth one-form
α = − y 1 d y 2 − y 2 d y 1 z + 1 . \alpha=-\frac{y_1dy_2-y_2dy_1}{z+1}. α = − z + 1 y 1 d y 2 − y 2 d y 1 .
Direct differentiation gives
d ξ h − α = d ( 2 arctan y 1 1 + z + y 2 ) . \frac{d\xi}{h}-\alpha=d\left(2\arctan\frac{y_1}{1+z+y_2}\right). h d ξ − α = d ( 2 arctan 1 + z + y 2 y 1 ) .
The denominator is positive, so the potential is a single-valued real function, including at the hyperboloid’s pole. Its integral on the closed projection is zero. Since the left side of the connection formula equals − q d t -q\,dt − q d t on f ′ = − X ^ f f'=-\widehat{X}f f ′ = − X f , integration yields
∫ 0 3 T q d t = π + 1 2 ∫ 0 3 T α ( Y ′ ) d t . \int_{0}^{3T}q\,dt=\pi+\frac{1}{2}\int_{0}^{3T}\alpha(Y')\,dt. ∫ 0 3 T q d t = π + 2 1 ∫ 0 3 T α ( Y ′ ) d t .
The Lax equations in these coordinates are
z ′ = − 2 ( y 1 r 2 − y 2 r 1 ) , y 1 ′ = 2 x y 2 − 2 z r 2 , y 2 ′ = 2 z r 1 − 2 x y 1 . z'=-2(y_1r_2-y_2r_1),\qquad y_1'=2xy_2-2zr_2,\qquad y_2'=2zr_1-2xy_1. z ′ = − 2 ( y 1 r 2 − y 2 r 1 ) , y 1 ′ = 2 x y 2 − 2 z r 2 , y 2 ′ = 2 z r 1 − 2 x y 1 .
Consequently
q − 1 2 α ( Y ′ ) = x + q z + 1 . q-\frac{1}{2}\alpha(Y')=\frac{x+q}{z+1}. q − 2 1 α ( Y ′ ) = z + 1 x + q .
This proves the second identity. The signed area contribution has been retained, so self-intersection or negative oriented area causes no difficulty. □
Theorem A.12 (Harmonic Casimir bound). For the chosen lift,
C ≤ 3 4 ( 2 A π − 1 ) 2 . (236) \mathcal{C}\leq\frac{3}{4}\left(\frac{2\mathcal{A}}{\pi}-1\right)^2. \tag*{(236)} C ≤ 4 3 ( π 2 A − 1 ) 2 . (236) For a global minimizer this implies
C < H ∗ : = 29025933167887355473205403 39478417604337498218323684 < H 0 : = 1832290554767847573595563 2491868845732503577653889 . \mathcal{C}<H_*:=\frac{29025933167887355473205403}{39478417604337498218323684}<H_0:=\frac{1832290554767847573595563}{2491868845732503577653889}. C < H ∗ := 39478417604337498218323684 29025933167887355473205403 < H 0 := 2491868845732503577653889 1832290554767847573595563 . Proof. Let w = y 1 r 2 − y 2 r 1 w=y_1r_2-y_2r_1 w = y 1 r 2 − y 2 r 1 . The Euclidean identity ∣ y ∣ 2 ∣ r ∣ 2 = ( y ⋅ r ) 2 + w 2 |\boldsymbol{y}|^2|\boldsymbol{r}|^2=(\boldsymbol{y}\cdot\boldsymbol{r})^2+w^2 ∣ y ∣ 2 ∣ r ∣ 2 = ( y ⋅ r ) 2 + w 2 and the two hyperboloid equations give the exact relation
2 x q ( z + 1 ) − ( x + q ) 2 = z 2 − 1 + w 2 ≥ 0. 2xq(z+1)-(x+q)^2=z^2-1+w^2\geq0. 2 x q ( z + 1 ) − ( x + q ) 2 = z 2 − 1 + w 2 ≥ 0.
All denominators below are positive. The function q ↦ 2 x q / ( x + q ) q \mapsto2xq/(x+q) q ↦ 2 x q / ( x + q ) is increasing, and (A.19) gives q ≤ 3 x / κ q \le\sqrt{3}x/\kappa q ≤ 3 x / κ . Thus
x + q z + 1 ≤ 2 x q x + q ≤ 2 3 3 + κ x . \frac{x+q}{z+1} \le\frac{2xq}{x+q} \le\frac{2\sqrt{3}}{\sqrt{3}+\kappa}x. z + 1 x + q ≤ x + q 2 x q ≤ 3 + κ 2 3 x .
Integrating (A.20) gives π ≤ 2 3 A / ( 3 + κ ) \pi\le2\sqrt{3}\mathcal{A}/(\sqrt{3}+\kappa) π ≤ 2 3 A / ( 3 + κ ) , or κ ≤ 3 ( 2 A / π − 1 ) \kappa\le\sqrt{3}(2\mathcal{A}/\pi-1) κ ≤ 3 ( 2 A / π − 1 ) . Its right side is positive because κ > 0 \kappa> 0 κ > 0 , so squaring is legitimate. Using (A.4) then yields
C < 3 ( 2 A 0 − p 0 ) 2 4 p 0 2 = H ∗ . \mathcal{C} < \frac{3(2A_0-p_0)^2}{4p_0^2}=H_*. C < 4 p 0 2 3 ( 2 A 0 − p 0 ) 2 = H ∗ .
The final comparison with H 0 H_0 H 0 is exact rational arithmetic. For reference, the weaker bound is also obtained directly from ( x + q ) / ( z + 1 ) ≤ ( x + q ) / 2 (x+q)/(z+1) \le(x+q)/2 ( x + q ) / ( z + 1 ) ≤ ( x + q ) /2 : it gives C ≤ 3 A 2 / [ 4 ( 2 π − A ) 2 ] \mathcal{C} \le3\mathcal{A}^2/[4(2\pi-\mathcal{A})^2] C ≤ 3 A 2 / [ 4 ( 2 π − A ) 2 ] and the stated H 0 H_0 H 0 .
Corollary A.13 (One-point calibration criterion). Fix one complete normal maximizing free-time lift on a genuine fundamental interval of the original problem, with its original state and group endpoints and prescribed real winding, and normalize the whole multiplier jointly to η = 1 \eta=1 η = 1 . At one actual point, in a lawful whole chart, write Λ = ( l b m − l ) \Lambda=\begin{pmatrix} l & b \\ m & -l \end{pmatrix} Λ = ( l m b − l ) and put K = 2 u l + b − u 2 m K=2ul+b-u^2m K = 2 u l + b − u 2 m , L = l − u m L=l-um L = l − u m . If C = det Λ > 0 \mathcal{C}=\det\Lambda>0 C = det Λ > 0 , then
K > 0 ⟺ m < 0 ⟺ tr ( Λ X ) > 0. K>0 \quad\Longleftrightarrow\quad m<0 \quad\Longleftrightarrow\quad\operatorname{tr}(\Lambda X)>0. K > 0 ⟺ m < 0 ⟺ tr ( Λ X ) > 0. Any one of these signs licenses (A.21) on this same lift; for an original global minimizer it gives 0 < C < H ∗ 0<\mathcal{C}<H_* 0 < C < H ∗ . For such a global minimizer, an independently proved floor C ≥ H ∗ \mathcal{C}\ge H_* C ≥ H ∗ contradicts K > 0 K>0 K > 0 on that lift. No assertion is made for an arbitrary local extremal.
Proof. Direct expansion at the specified actual state gives C = − L 2 − K m \mathcal{C}=-L^2-Km C = − L 2 − K m and tr ( Λ X ) = ( K − v 2 m ) / v \operatorname{tr}(\Lambda X)=(K-v^2m)/v tr ( Λ X ) = ( K − v 2 m ) / v , independently of the active control. Since v > 0 v>0 v > 0 and K m = − C − L 2 < 0 Km=-\mathcal{C}-L^2<0 K m = − C − L 2 < 0 , the three signs are equivalent. Lemma A.10 and Theorem A.12 now apply on the original interval. Their matrix-only three-copy transport retains the real winding and assumes no periodic state covector.
The preceding proof is valid for measurable maximizing controls. All derivatives are identities of absolutely continuous functions and almost-everywhere equations; none differentiates a control. Every application must establish normality, positivity of the determinant, and one positive trace for the same selected lift. A scalar local cost conclusion may be transferred to a global sum, but its costate and Casimir may not be transferred to another lift.
Short-chain rigidity
The long-word classification concerns B ≥ 8 B\ge8 B ≥ 8 . We now prove that an original global minimizer with B ≤ 7 B\le7 B ≤ 7 is the affine smoothed regular octagon. This is the short alternative used in Theorem 1.1; its proof is independent of the long-run classification.
The proof has three parts. An oriented four-arc descent and a missing-label argument leave a short list of words. The entire coherent four-arc family has the octagon as its unique area minimum. Exact closure charts, local variations and complete-PMP inequalities exclude the remaining five-, six- and seven-arc words. The argument uses the original control triangle, free terminal time and the true fixed-endpoint winding class throughout. We give the full analytic proof below; its finite inequalities are specified mathematically there and their arithmetic contracts are collected in Appendix H.11.
Identification of the geometric and variational domains
The comparison constant is
J 8 = 2 ( 8 − 4 2 − log 2 ) 2 2 − 1 , U = 180483 100000 , V = 361 200 , K = 29 16 . J_8=\frac{2(8-4\sqrt{2}-\log2)}{2\sqrt{2}-1},\qquad U=\frac{180483}{100000},\qquad V=\frac{361}{200},\qquad K=\frac{29}{16}. J 8 = 2 2 − 1 2 ( 8 − 4 2 − log 2 ) , U = 100000 180483 , V = 200 361 , K = 16 29 .
The exact inequalities are J 8 < U < V < K J_8 < U < V < K J 8 < U < V < K . In the explicit closure calculations below we write J = C \mathcal{J}=C J = C and J 8 = J 8 \mathcal{J}_8=J_8 J 8 = J 8 . The edge cost J J J used only in the four-arc variation differs from area by a positive factor and endpoint-fixed terms. These scalar symbols are local and distinct from the matrix and compact set denoted by J J J and K K K elsewhere.
Lemma A.14 (Compatibility of the short-chain proof). The normalized trajectory of an original global minimizer satisfies the geometric, endpoint, winding and minimizing hypotheses of the short-chain argument. Its essential count is the same B B B in both arguments. The complete-PMP statements used in the short proof may be applied on their actual containing intervals without matching independently chosen costates through the twisted seam.
Proof. The diagonal-conjugate coordinates used in the short-chain proof give
X ( u , v ) = 1 v ( u − u 2 − v 2 1 − u ) , R ∗ = 1 2 ( 1 − 3 1 1 ) , X(u,v)=\frac{1}{v}\begin{pmatrix} u & -u^2-v^2 \\ 1 & -u \end{pmatrix}, \qquad R_*=\frac{1}{2}\begin{pmatrix} 1 & -3 \\ 1 & 1 \end{pmatrix}, X ( u , v ) = v 1 ( u 1 − u 2 − v 2 − u ) , R ∗ = 2 1 ( 1 1 − 3 1 ) ,
− 1 < u < 1 , v > 0 , u 2 + v 2 > 1. -1<u<1,\qquad v>0,\qquad u^2+v^2>1. − 1 < u < 1 , v > 0 , u 2 + v 2 > 1.
These are the data in Section A.1. The group endpoint is G ( T ) = R ∗ G(T)=R_* G ( T ) = R ∗ and the state endpoint is W ( T ) = R ∗ − 1 ⋅ W ( 0 ) W(T)=R_*^{-1}\cdot W(0) W ( T ) = R ∗ − 1 ⋅ W ( 0 ) , rather than merely a trace condition or a state-return condition. Both use the fixed-endpoint real winding class. The scalar normalization follows from (3) . The physical circle-coordinate labels 0, 2, 4 become the reduced labels 0, 1, 2 by division by two. Only the edge variation and the octagon geometry below retain physical labels; the closure charts use reduced labels.
The finite positive bang representation is supplied by Theorem A.3 . Zero-duration pieces are deleted and equal adjacent controls are merged, including the twisted seam, before either argument counts B B B . Both extend the reduced labels by j i + B = j i + 1 j_{i+B}=j_i+1 j i + B = j i + 1 modulo three. Consequently their increment strings have the same orientation and satisfy ∑ i ϵ i ≡ 1 \sum_i\epsilon_i\equiv1 ∑ i ϵ i ≡ 1 modulo three. Only cyclic rotation is used to identify the short representatives below.
The admissible competitors are those of the original full control triangle, with free terminal time and variable entrance state. No upper bound on the number of arcs is imposed on that competitor class. Lemmas 2.2 and 2.6 apply to fixed-endpoint local windows and genuine reanchoring, respectively. For the six-arc arguments the variation is instead a whole-cycle family satisfying exactly
G ( 0 ) = I , G ( T ) = R ∗ , W ( T ) = R ∗ − 1 ⋅ W ( 0 ) , G(0)=I,\qquad G(T)=R_*,\qquad W(T)=R_*^{-1}\cdot W(0), G ( 0 ) = I , G ( T ) = R ∗ , W ( T ) = R ∗ − 1 ⋅ W ( 0 ) ,
with W ( 0 ) W(0) W ( 0 ) allowed to vary as its formulas prescribe. These are the original endpoint conditions; they are not a claim that both state endpoint values remain fixed. After rescaling the time interval, the pointwise quotient of sufficiently close group paths has a continuous real matrix logarithm, providing an endpoint-fixed homotopy of the group paths. Thus the real winding class is preserved. Positive durations and strict-star admissibility persist by openness. The same curvature and winding reconstruction used in Lemma 2.2, now with the twist providing tangent matching around the whole boundary, gives an admissible nearby convex disk. No application of a fixed-window lemma to varying window endpoint data is needed.
The oriented four-arc variation proved in Theorem A.16 concerns exactly the reduced words 0101, 1212, 2020. The six-letter exclusion is Lemma A.9 . For the pure-forward state filter we use Lemma B.24 on an actual containing interval. At the actual middle switch, its reciprocals are
d 0 = a i , d 2 = a i − 1 , d 1 = 1 − a i a i − 1 a i + a i − 1 . d_0=a_i,\qquad d_2=a_{i-1},\qquad d_1=\frac{1-a_i a_{i-1}}{a_i+a_{i-1}}. d 0 = a i , d 2 = a i − 1 , d 1 = a i + a i − 1 1 − a i a i − 1 .
Thus d 1 < min ( d 0 , d 2 ) d_1<\min(d_0,d_2) d 1 < min ( d 0 , d 2 ) is exactly
a i 2 + 2 a i a i − 1 > 1 , a i − 1 2 + 2 a i a i − 1 > 1. a_i^2+2a_i a_{i-1}>1,\qquad a_{i-1}^2+2a_i a_{i-1}>1. a i 2 + 2 a i a i − 1 > 1 , a i − 1 2 + 2 a i a i − 1 > 1.
Strict star gives 0 < a i a i − 1 < 1 0 < a_i a_{i-1} < 1 0 < a i a i − 1 < 1 . Each capped 1201 support is placed in one genuine fundamental interval. Different cuts may supply different full lifts; only the resulting scalar state inequalities are combined. No periodic state-costate assumption is made. This verifies the interfaces needed below. □ \square □
Exhaustion of short words
For B ≥ 4 B \ge4 B ≥ 4 , any cyclic factor + − + +-+ + − + uses four distinct physical arcs and is prohibited by Lemma A.4. For B = 3 B = 3 B = 3 the formal factor repeats an arc and must not be used as a local exclusion. Keeping that case for a separate geometric argument gives the following complete list, up to rotation:
Table 17.
Indeed if k k k signs are negative, B − 2 k ≡ 1 ( m o d 3 ) B - 2k \equiv1 \pmod{3} B − 2 k ≡ 1 ( mod 3 ) . For B ≥ 4 B \ge4 B ≥ 4 distribute the negative signs into cyclic runs, with no singleton negative run between positive signs. Listing the possibilities gives the table. The marked-string counts before forbidden factors are 1, 1, 3, 5, 11, 21, 43. The exact elementary enumerator supplied as short/check_short_words.py checks these counts, all representatives, and the six-arc placement below. It does not purport to prove geometric feasibility of a word.
Proposition A.15 (Short-chain rigidity). An original Reinhardt global minimizer with B ≤ 7 B \le7 B ≤ 7 is an affine image of the smoothed regular octagon.
Proof. By Lemma A.14, all the variations and complete-PMP conditions below apply in the original geometric class. We first assemble the cases; the complete calculations follow.
At most four arcs. Lemma A.21 excludes the one-, two- and three-arc words and the mixed four-cycle. Its missing-label argument uses the straight boundary components and the strict positive determinant of their tangents; it also handles the repeated-arc issue at B = 3 B = 3 B = 3 . For the coherent four-cycle, Proposition A.22 classifies the whole family rather than imposing symmetry. The balanced support octagon is an intersection of two congruent squares. Its hyperbolic rounding gives a one-parameter area function, strictly increasing away from the regular octagon. Thus J ≥ J 8 \mathcal{J} \ge\mathcal{J}_8 J ≥ J 8 , with equality precisely for its affine images.
Five arcs. For the word 01202, Lemmas A.23 and A.24 give the exact two-parameter cost and exclude all its interior critical points. Proposition A.25 then checks every boundary and escaping sequence, giving J > J 8 \mathcal{J} > \mathcal{J}_8 J > J 8 . For 02102, Proposition A.26 gives the stronger bound J > K \mathcal{J} > K J > K . Both compare all actual strict chains of the stated words.
Six arcs. Proposition A.27 gives an exact three-parameter closure family about every actual 010202 chain and proves that its cost has no critical point. For 021020, Lemma A.28 gives a local exact closure family with parameters ( a , b , t ) (a,b,t) ( a , b , t ) . Lemma A.29 proves
J ≤ K ⟹ ∂ J ∂ b ∣ a , t > 0. \mathcal{J} \le K \quad\Longrightarrow\quad\left.\frac{\partial\mathcal{J}}{\partial b}\right|_{a,t} > 0. J ≤ K ⟹ ∂ b ∂ J a , t > 0. The octagonal competitor gives J ≤ J 8 < K \mathcal{J} \le\mathcal{J}_8 < K J ≤ J 8 < K , so decreasing b b b slightly improves the cost. The full endpoint conditions, strict star and real winding persist as verified in Lemma A.14.
Seven arcs. The representative + − − − − +---- + − − − − has a cut 0210202. Proposition A.30 and Lemma A.31 give a lower comparison J ≥ L > J 8 \mathcal{J} \ge L > \mathcal{J}_8 J ≥ L > J 8 on its complete necessary domain, including the noncompact tails and the octagonal boundary collar. No uniform positive gap is needed.
For +++++++, Lemmas A.34 and A.35 use complete maximization and a finite state graph at budget V V V to prove q i < 2 q_i < 2 q i < 2 and 757 / 2000 ≤ a i ≤ 67 / 50 757/2000 \leq a_i \leq67/50 757/2000 ≤ a i ≤ 67/50 . The exact full-matrix closure chart, a largest-exponent cut and Lemma A.36 then give J > U > J 8 \mathcal{J} > U > \mathcal{J}_8 J > U > J 8 , proving Proposition A.32. The prerequisite budget V > U V > U V > U keeps the state-domain reduction independent of the final cost exclusion.
For ++++---, take the actual extended arc indices 3, 4, 5, 6, 7, 8. Modulo seven these are 3, 4, 5, 6, 0, 1, all distinct, and their phased reduced word is 010212. They omit positive arc 2. Cutting in the interior of that omitted arc puts the six-arc support strictly inside one genuine fundamental interval. Lemma A.9 applies to its one complete maximizing lift and excludes this last word.
Only the coherent four-arc family remains. The octagonal competitor and its strict equality classification force the asserted affine smoothed regular octagon. □
A four-arc variation with fixed complete endpoints
The reduction of short chains begins with a variation that preserves both components of the endpoint: the position matrix and the reduced state. In the four-arc variation the physical labels of active opposite pairs are 0, 2, 4; dividing these labels by two gives the reduced labels 0, 1, 2. The sign of a transition is read in the fixed cyclic order of the controls. The reversed oriented edge is a different variational problem.
Theorem A.16 (Oriented four-arc descent). An actual positive four-arc window with the oriented physical word 0, 2, 0, 2 has an arbitrarily small, strictly area-decreasing variation with the same word and the same complete entrance and exit states. Consequently, a cyclic sign factor + − + +-+ + − + cannot occur in a minimizing chain when its four physical arcs are distinct.
Suppose that the physical word is 0–2–0–2. The two controls lie on one edge of the control simplex. After a cyclic relabeling of the three control coordinates, the word is A A A –B B B –A A A –B B B , where the edge parameter of [3, Section 8.1] takes the values
u e d g e ( A ) = 1 2 , u e d g e ( B ) = − 1 2 . u_{\mathrm{edge}}(A)=\frac{1}{2}, \qquad u_{\mathrm{edge}}(B)=-\frac{1}{2}. u edge ( A ) = 2 1 , u edge ( B ) = − 2 1 .
In the edge coordinates, put
q = 1 y , a = 3 , y > 0. (237) q=\frac{1}{y}, \qquad a=\sqrt{3}, \qquad y>0. \tag*{(237)} q = y 1 , a = 3 , y > 0. (237)
We retain throughout the full strict-star condition
∣ x ∣ < 1 3 , x 2 + y 2 > 1 3 . (238) \lvert x\rvert<\frac{1}{\sqrt{3}}, \qquad x^2+y^2>\frac{1}{3}. \tag*{(238)} ∣ x ∣ < 3 1 , x 2 + y 2 > 3 1 . (238)
Only the first inequality is needed for the signs below; the second is part of admissibility and persists by openness. Fix the full entrance state z − = ( g − , X − ) z_-=(g_-,X_-) z − = ( g − , X − ) , and write x 0 + i y 0 x_0+iy_0 x 0 + i y 0 for the upper-half-plane coordinate of X − X_- X − . Thus g − g_- g − , x 0 x_0 x 0 , and
q ( x 0 ) = q 0 : = y 0 − 1 q(x_0)=q_0:=y_0^{-1} q ( x 0 ) = q 0 := y 0 − 1
are fixed throughout the variation. Since x ′ = y > 0 x'=y>0 x ′ = y > 0 , we use x x x as the independent variable. The two bang equations become
d log q d x = p A ( x ) = a 1 − a x ′ , d log q d x = p B ( x ) = − a 1 + a x ′ . (239) \frac{d\log q}{dx}=p_A(x)=\frac{a}{1-ax'}, \qquad\frac{d\log q}{dx}=p_B(x)=-\frac{a}{1+ax'}. \tag*{(239)} d x d log q = p A ( x ) = 1 − a x ′ a , d x d log q = p B ( x ) = − 1 + a x ′ a . (239)
Define
ρ = q 2 , h ( x ) = p A ( x ) − p B ( x ) = 2 a 1 − a 2 x 2 > 0. (240) \rho=q^2, \qquad h(x)=p_A(x)-p_B(x)=\frac{2a}{1-a^2x^2}>0. \tag*{(240)} ρ = q 2 , h ( x ) = p A ( x ) − p B ( x ) = 1 − a 2 x 2 2 a > 0. (240)
Write s s s for the shear state and
S : = Δ s = s ( x 4 ) − s ( x 0 ) S:=\Delta s=s(x_4)-s(x_0) S := Δ s = s ( x 4 ) − s ( x 0 )
for its increment across the four-link window. Since x ′ = y x'=y x ′ = y and q = 1 / y q=1/y q = 1/ y , we have
d s = ρ d x , S = ∫ x 0 x 4 ρ d x , d J = x 2 ρ d x . (241) ds=\rho\,dx,\qquad S=\int_{x_0}^{x_4}\rho\,dx,\qquad dJ=x^2\rho\,dx. \tag*{(241)} d s = ρ d x , S = ∫ x 0 x 4 ρ d x , dJ = x 2 ρ d x . (241)
Here J J J is the edge cost after removal of a boundary term fixed by the endpoint data; it differs from the Reinhardt area by a positive constant factor and fixed endpoint terms [3, Eqs. (8.1.3)–(8.1.5)].
Fix entrance and exit coordinates x 0 < x 4 x_0<x_4 x 0 < x 4 , and let
x 0 < x 1 < x 2 < x 3 < x 4 x_0<x_1<x_2<x_3<x_4 x 0 < x 1 < x 2 < x 3 < x 4
be the switching positions. For the unperturbed trajectory, put
R i = ∫ x i x 4 ρ d x , m L = R 1 − R 2 , m R = R 2 − R 3 . (242) R_i=\int_{x_i}^{x_4}\rho\,dx,\qquad m_L=R_1-R_2,\qquad m_R=R_2-R_3. \tag*{(242)} R i = ∫ x i x 4 ρ d x , m L = R 1 − R 2 , m R = R 2 − R 3 . (242)
Thus m L , m R > 0 m_L,m_R>0 m L , m R > 0 are the ρ \rho ρ -masses of the two middle links.
Let ξ i = x ˙ i \xi_i=\dot{x}_i ξ i = x ˙ i , h i = h ( x i ) h_i=h(x_i) h i = h ( x i ) , and ρ i = ρ ( x i ) \rho_i=\rho(x_i) ρ i = ρ ( x i ) . Moving a switch changes log q \log q log q on the entire downstream tail. The three downstream increments are
( b 1 , b 2 , b 3 ) = ( h 1 ξ 1 , − h 2 ξ 2 , h 3 ξ 3 ) . (243) (b_1,b_2,b_3)=(h_1\xi_1,-h_2\xi_2,h_3\xi_3). \tag*{(243)} ( b 1 , b 2 , b 3 ) = ( h 1 ξ 1 , − h 2 ξ 2 , h 3 ξ 3 ) . (243)
The linearized conditions that the terminal values of q q q and S S S remain fixed are
b 1 + b 2 + b 3 = 0 , R 1 b 1 + R 2 b 2 + R 3 b 3 = 0. (244) b_1+b_2+b_3=0,\qquad R_1b_1+R_2b_2+R_3b_3=0. \tag*{(244)} b 1 + b 2 + b 3 = 0 , R 1 b 1 + R 2 b 2 + R 3 b 3 = 0. (244)
Because R 1 > R 2 > R 3 R_1>R_2>R_3 R 1 > R 2 > R 3 , these equations have rank two. Their kernel is
( b 1 , b 2 , b 3 ) = c 0 ( m R , − ( m L + m R ) , m L ) , (245) (b_1,b_2,b_3)=c_0\left(m_R,-(m_L+m_R),m_L\right), \tag*{(245)} ( b 1 , b 2 , b 3 ) = c 0 ( m R , − ( m L + m R ) , m L ) , (245)
or, equivalently,
( ξ 1 , ξ 2 , ξ 3 ) = c 0 ( m R h 1 , m L + m R h 2 , m L h 3 ) . (246) (\xi_1,\xi_2,\xi_3)=c_0\left(\frac{m_R}{h_1},\frac{m_L+m_R}{h_2},\frac{m_L}{h_3}\right). \tag*{(246)} ( ξ 1 , ξ 2 , ξ 3 ) = c 0 ( h 1 m R , h 2 m L + m R , h 3 m L ) . (246)
Lemma A.17 (Exact edge fiber). Near the given four-link chain, the A A A –B B B –A A A –B B B trajectories with the fixed full entrance state and the same exit data ( S , x 4 , q 4 ) (S,x_4,q_4) ( S , x 4 , q 4 ) form a one-dimensional analytic manifold. Equivalently, they have the same full exit state ( g 4 , X 4 ) (g_4,X_4) ( g 4 , X 4 ) . The tangent is (A.31), and both local directions keep all four links positive.
Proof. Let x = ( x 1 , x 2 , x 3 ) \mathbf{x}=(x_1,x_2,x_3) x = ( x 1 , x 2 , x 3 ) , and let q x q_{\mathbf{x}} q x solve the four bang equations with q x ( x 0 ) = q 0 q_{\mathbf{x}}(x_0)=q_0 q x ( x 0 ) = q 0 . On the ordered open set
Ω = { x 0 < x 1 < x 2 < x 3 < x 4 } , \Omega=\{x_0<x_1<x_2<x_3<x_4\}, Ω = { x 0 < x 1 < x 2 < x 3 < x 4 } ,
define the exact endpoint map
F ( x ) = ( log q x ( x 4 ) , ∫ x 0 x 4 q x ( x ) 2 d x ) . F(\mathbf{x})=\left(\log q_{\mathbf{x}}(x_4),\int_{x_0}^{x_4}q_{\mathbf{x}}(x)^2\,dx\right). F ( x ) = ( log q x ( x 4 ) , ∫ x 0 x 4 q x ( x ) 2 d x ) .
It is analytic. If d = ( h 1 , − h 2 , h 3 ) d=(h_1,-h_2,h_3) d = ( h 1 , − h 2 , h 3 ) , direct differentiation gives
D F ( x ) = ( d 1 d 2 d 3 2 R 1 d 1 2 R 2 d 2 2 R 3 d 3 ) . (247) DF(\mathbf{x})=
\begin{pmatrix}
d_1 & d_2 & d_3\\
2R_1d_1 & 2R_2d_2 & 2R_3d_3
\end{pmatrix}. \tag*{(247)} D F ( x ) = ( d 1 2 R 1 d 1 d 2 2 R 2 d 2 d 3 2 R 3 d 3 ) . (247)
Since R 1 > R 2 > R 3 R_1>R_2>R_3 R 1 > R 2 > R 3 , this matrix has rank two. Hence the analytic implicit function theorem makes F − 1 ( F ( x ∗ ) ) F^{-1}(F(\mathbf{x}^*)) F − 1 ( F ( x ∗ )) an exact one-dimensional analytic manifold, not merely a linearized constraint set. Its tangent is (A.31). Because Ω \Omega Ω is open, both signs of a sufficiently small parameter retain all four positive links. The passage from the four flow durations to the four successive x x x -endpoints is triangular with positive diagonal entries x ′ = y x'=y x ′ = y ; it is therefore a local analytic change of coordinates.
It remains to check that these edge data are precisely the full-state endpoint data. In the notation of [3, Eq. (8.1.2)],
g ( t ) = g ( t 0 ) h ( t 0 ) − 1 h ( t ) , h ( s , x ) = ( 1 − x s 1 − s x ) . g(t)=g(t_0)h(t_0)^{-1}h(t),\qquad h(s,x)=\begin{pmatrix}1&-x\\s&1-sx\end{pmatrix}. g ( t ) = g ( t 0 ) h ( t 0 ) − 1 h ( t ) , h ( s , x ) = ( 1 s − x 1 − s x ) .
and the additive constant in s s s is a gauge. More explicitly, if s i = s ( x i ) s_i=s(x_i) s i = s ( x i ) and S = s 4 − s 0 S=s_4-s_0 S = s 4 − s 0 , then
h ( s 0 , x 0 ) − 1 h ( s 4 , x 4 ) = ( 1 + x 0 S x 0 − x 4 − x 0 x 4 S S 1 − x 4 S ) . (248) h(s_0,x_0)^{-1}h(s_4,x_4)=\begin{pmatrix}1+x_0S&x_0-x_4-x_0x_4S\\S&1-x_4S\end{pmatrix}. \tag*{(248)} h ( s 0 , x 0 ) − 1 h ( s 4 , x 4 ) = ( 1 + x 0 S S x 0 − x 4 − x 0 x 4 S 1 − x 4 S ) . (248)
Thus
g 4 = g − h ( s 0 , x 0 ) − 1 h ( s 4 , x 4 ) g_4=g_-h(s_0,x_0)^{-1}h(s_4,x_4) g 4 = g − h ( s 0 , x 0 ) − 1 h ( s 4 , x 4 )
is determined exactly by the fixed entrance data and ( S , x 4 ) (S,x_4) ( S , x 4 ) . The reduced state X 4 X_4 X 4 is determined by x 4 + i y 4 = x 4 + i / q 4 x_4+iy_4=x_4+i/q_4 x 4 + i y 4 = x 4 + i / q 4 . Consequently, after the full entrance state is fixed,
( S , x 4 , q 4 ) ⟷ ( g 4 , X 4 ) (S,x_4,q_4)\longleftrightarrow(g_4,X_4) ( S , x 4 , q 4 ) ⟷ ( g 4 , X 4 )
are equivalent endpoint data on this fixed control edge. The endpoint image of edge-confined trajectories is three-dimensional; its four-duration fixed-endpoint fiber therefore has dimension one. □ \square □
We now compute the area along this exact fiber. Define the weighted means
μ L = 1 m L ∫ x 1 x 2 x 2 ρ d x , μ R = 1 m R ∫ x 2 x 3 x 2 ρ d x . (249) \mu_L=\frac{1}{m_L}\int_{x_1}^{x_2}x^2\rho\,dx,\qquad\mu_R=\frac{1}{m_R}\int_{x_2}^{x_3}x^2\rho\,dx. \tag*{(249)} μ L = m L 1 ∫ x 1 x 2 x 2 ρ d x , μ R = m R 1 ∫ x 2 x 3 x 2 ρ d x . (249)
Lemma A.18 (First and second edge variations). Along the tangent (A.31),
J ˙ = 2 c 0 m L m R ( μ L − μ R ) . (250) \dot J=2c_0m_Lm_R(\mu_L-\mu_R). \tag*{(250)} J ˙ = 2 c 0 m L m R ( μ L − μ R ) . (250) If this derivative vanishes, write μ L = μ R = λ \mu_L=\mu_R=\lambda μ L = μ R = λ . The constrained second variation is
Q = − 2 c 0 2 [ ρ 1 h 1 ( x 1 2 − λ ) m R 2 + ρ 2 h 2 ( λ − x 2 2 ) ( m L + m R ) 2 + ρ 3 h 3 ( x 3 2 − λ ) m L 2 ] < 0. (251) Q=-2c_0^2\left[\frac{\rho_1}{h_1}(x_1^2-\lambda)m_R^2+\frac{\rho_2}{h_2}(\lambda-x_2^2)(m_L+m_R)^2+\frac{\rho_3}{h_3}(x_3^2-\lambda)m_L^2\right]<0. \tag*{(251)} Q = − 2 c 0 2 [ h 1 ρ 1 ( x 1 2 − λ ) m R 2 + h 2 ρ 2 ( λ − x 2 2 ) ( m L + m R ) 2 + h 3 ρ 3 ( x 3 2 − λ ) m L 2 ] < 0. (251)
Proof. Let
T i = ∫ x i x 4 x 2 ρ d x . T_i=\int_{x_i}^{x_4}x^2\rho\,dx. T i = ∫ x i x 4 x 2 ρ d x .
A switch increment b i b_i b i changes ρ = q 2 \rho=q^2 ρ = q 2 on its downstream tail by 2 b i ρ 2b_i\rho 2 b i ρ to first order. Consequently
J ˙ = 2 ∑ i = 1 3 b i T i . \dot J=2\sum_{i=1}^{3}b_iT_i. J ˙ = 2 i = 1 ∑ 3 b i T i .
Substitution of (A.30) and grouping adjacent tails gives
∑ i = 1 3 b i T i = c 0 m R ( T 1 − T 2 ) − c 0 m L ( T 2 − T 3 ) = c 0 m L m R ( μ L − μ R ) , \sum_{i=1}^{3}b_iT_i=c_0m_R(T_1-T_2)-c_0m_L(T_2-T_3)=c_0m_Lm_R(\mu_L-\mu_R), i = 1 ∑ 3 b i T i = c 0 m R ( T 1 − T 2 ) − c 0 m L ( T 2 − T 3 ) = c 0 m L m R ( μ L − μ R ) ,
which is (A.35).
Suppose now that μ L = μ R = λ \mu_L=\mu_R=\lambda μ L = μ R = λ . For completeness, we derive the quadratic form rather than differentiating a chosen parametrization of the fiber. Set
L = J − λ S − ν log q ( x 4 ) , U i = ∫ x i x 4 ( x 2 − λ ) ρ d x . \mathcal{L}=J-\lambda S-\nu\log q(x_4),\qquad U_i=\int_{x_i}^{x_4}(x^2-\lambda)\rho\,dx. L = J − λ S − ν log q ( x 4 ) , U i = ∫ x i x 4 ( x 2 − λ ) ρ d x .
Switch stationarity is 2 U i = ν 2U_i=\nu 2 U i = ν for i = 1 , 2 , 3 i=1,2,3 i = 1 , 2 , 3 . If σ = ( 1 , − 1 , 1 ) \sigma=(1,-1,1) σ = ( 1 , − 1 , 1 ) and d i = σ i h i d_i=\sigma_i h_i d i = σ i h i , direct differentiation gives the Hessian at a stationary switch configuration as
D 2 L = 2 ν d d T − 2 diag ( d i ( x i 2 − λ ) ρ i ) i = 1 3 . (252) D^2\mathcal{L}=2\nu\,d d^T-2\operatorname{diag}\left(d_i(x_i^2-\lambda)\rho_i\right)_{i=1}^{3}. \tag*{(252)} D 2 L = 2 ν d d T − 2 diag ( d i ( x i 2 − λ ) ρ i ) i = 1 3 . (252)
The equality μ L = μ R = λ \mu_L=\mu_R=\lambda μ L = μ R = λ gives
U 1 − U 2 = m L ( μ L − λ ) = 0 , U 2 − U 3 = m R ( μ R − λ ) = 0. U_1-U_2=m_L(\mu_L-\lambda)=0,\qquad U_2-U_3=m_R(\mu_R-\lambda)=0. U 1 − U 2 = m L ( μ L − λ ) = 0 , U 2 − U 3 = m R ( μ R − λ ) = 0.
Hence U 1 = U 2 = U 3 U_1=U_2=U_3 U 1 = U 2 = U 3 , and choosing ν = 2 U i \nu=2U_i ν = 2 U i makes D L = 0 D\mathcal{L}=0 D L = 0 in all three switch variables. Let x ( r ) \mathbf{x}(r) x ( r ) be an analytic parametrization of the exact fiber with x ′ ( 0 ) = ξ \mathbf{x}'(0)=\xi x ′ ( 0 ) = ξ . Both S S S and log q ( x 4 ) \log q(x_4) log q ( x 4 ) are constant on this curve, so L ( x ( r ) ) \mathcal{L}(\mathbf{x}(r)) L ( x ( r )) differs from J ( x ( r ) ) J(\mathbf{x}(r)) J ( x ( r )) by a constant. Since D L = 0 D\mathcal{L}=0 D L = 0 ,
d 2 d r 2 J ( x ( r ) ) ∣ r = 0 = D 2 L [ ξ , ξ ] . (253) \left.\frac{d^2}{dr^2}J(\mathbf{x}(r))\right|_{r=0}=D^2\mathcal{L}[\xi,\xi]. \tag*{(253)} d r 2 d 2 J ( x ( r )) r = 0 = D 2 L [ ξ , ξ ] . (253)
Thus the quadratic form below is the actual second derivative along an exact fixed-endpoint curve; no acceleration or exterior correction remains. On the terminal-q q q tangent space, d ⋅ ξ = ∑ b i = 0 d\cdot\xi=\sum b_i=0 d ⋅ ξ = ∑ b i = 0 , so the rank-one term vanishes. Inserting (A.30) into the diagonal term yields precisely (A.36).
It remains to prove the sign. Equality of the two means forces
x 1 2 > λ > x 2 2 , x 3 2 > λ . (254) x_1^2>\lambda>x_2^2,\qquad x_3^2>\lambda. \tag*{(254)} x 1 2 > λ > x 2 2 , x 3 2 > λ . (254)
If x 2 ≤ 0 x_2\leq0 x 2 ≤ 0 , then x 2 x^2 x 2 is strictly decreasing on ( x 1 , x 2 ) (x_1,x_2) ( x 1 , x 2 ) , giving the first two inequalities. If x 3 2 ≤ λ x_3^2\leq\lambda x 3 2 ≤ λ , convexity of x 2 x^2 x 2 would make x 2 < λ x^2<\lambda x 2 < λ throughout ( x 2 , x 3 ) (x_2,x_3) ( x 2 , x 3 ) , contradicting its mean. If x 2 ≥ 0 x_2\geq0 x 2 ≥ 0 , the right interval first gives x 2 2 < λ < x 3 2 x_2^2<\lambda<x_3^2 x 2 2 < λ < x 3 2 ; if x 1 2 ≤ λ x_1^2\leq\lambda x 1 2 ≤ λ , the same endpoint-maximum argument on the left interval gives a contradiction. Thus every term in brackets in (A.36) is strictly positive. □
Proof of Theorem A.16. Lemma A.17 gives an exact two-sided variation fixing the complete endpoint. If (A.35) is nonzero, choose its decreasing sign. If it vanishes, Lemma A.18 gives a strictly negative constrained second variation. The four-link replacement may therefore be inserted into any longer chain without moving its complement.
To spell out the concatenation, let z i n z_{\mathrm{in}} z in and z o u t z_{\mathrm{out}} z out be the full states at the ends of the window. The exact fiber keeps both fixed. The exterior flow therefore restarts from precisely the same state z o u t z_{\mathrm{out}} z out , and uniqueness of the state equation makes the entire downstream trajectory, including its final full state, unchanged. The window duration may change, but terminal time is free, so the exterior is only translated in time. For a twisted-closed trajectory this preserves both the position condition on g g g and the twisted terminal condition on X X X . Strict-star admissibility and embeddedness persist for a sufficiently small C 1 C^1 C 1 replacement. No additional global closure equation or second-order compensator is needed.
A global minimizer must minimize on every edge-restricted subinterval with the same endpoint data [3, Prop. 8.1.3]; hence it cannot contain the factor. Cyclic label symmetry supplies the other two physical shifts. □
The finite list of short sign cycles
Let B B B denote the number of essential positive arcs, after zero arcs have been deleted and equal adjacent controls have been merged, including at the twisted seam. If a i ∈ Z / 3 Z a_i \in\mathbb{Z}/3\mathbb{Z} a i ∈ Z /3 Z is the active label, extend it by a i + B = a i + 1 a_{i+B}=a_i+1 a i + B = a i + 1 . Then
ϵ i = a i + 1 − a i ∈ { 1 , − 1 } , ∑ i = 0 B − 1 ϵ i = 1 ( m o d 3 ) . (255) \epsilon_i=a_{i+1}-a_i\in\{1,-1\},\qquad\sum_{i=0}^{B-1}\epsilon_i=1\pmod3. \tag*{(255)} ϵ i = a i + 1 − a i ∈ { 1 , − 1 } , i = 0 ∑ B − 1 ϵ i = 1 ( mod 3 ) . (255)
For B ≥ 4 B\ge4 B ≥ 4 , a cyclic occurrence of + − + +-+ + − + uses four different physical arcs. Cutting immediately before its first arc puts it inside one fundamental interval. Theorem A.16 then applies to that interval, with its actual twisted labels. Its complement remains fixed. No continuation of a state costate through the seam is involved.
The complete list for B ≤ 7 B\le7 B ≤ 7 is given in Section A.4.2. The three-arc representative is retained there until the following geometric argument: a formal + − − + +--+ + − − + at B = 3 B=3 B = 3 repeats a physical arc and cannot be excluded by the four-arc variation.
Lemma A.21. A regular strict-star twisted-closed chain cannot omit one active label. The mixed four-arc sign cycle + − − − +--- + − − − has no such realization.
Proof. Write the six boundary curves in the circle representation as σ j = g e j \sigma_j=ge_j σ j = g e j , where e j = ( cos ( j π / 3 ) , sin ( j π / 3 ) ) e_j=(\cos(j\pi/3),\sin(j\pi/3)) e j = ( cos ( j π /3 ) , sin ( j π /3 )) . If physical label 4 is omitted, σ 4 \sigma_4 σ 4 is straight throughout a fundamental interval. Regular C 1 C^1 C 1 matching puts its pieces on one line. The twist gives σ 4 ′ ( T ) = σ 5 ′ ( 0 ) = − σ 2 ′ ( 0 ) \sigma_4'(T)=\sigma_5'(0)=-\sigma_2'(0) σ 4 ′ ( T ) = σ 5 ′ ( 0 ) = − σ 2 ′ ( 0 ) , whereas strict star gives
det ( σ 2 ′ , σ 4 ′ ) = det X det ( e 2 , e 4 ) = 3 2 det X > 0. \det(\sigma_2',\sigma_4')=\det X\det(e_2,e_4)=\frac{\sqrt{3}}{2}\det X>0. det ( σ 2 ′ , σ 4 ′ ) = det X det ( e 2 , e 4 ) = 2 3 det X > 0.
This is impossible. The classes + + + and − − -- − − omit labels. For the three-arc class, choose the cut giving the reduced word 121; it omits label 0.
For the mixed four-arc class use physical labels ( 0 , 4 , 2 , 0 ) (0,4,2,0) ( 0 , 4 , 2 , 0 ) and extend by a r + 4 = a r + 2 a_{r+4}=a_r+2 a r + 4 = a r + 2 . Around the seam the labels are
r − 2 − 1 0 1 2 3 4 5 a r 0 4 0 4 2 0 2 0 \begin{array}{c|rrrrrrrr}
r & -2 & -1 & 0 & 1 & 2 & 3 & 4 & 5\\
a_r & 0 & 4 & 0 & 4 & 2 & 0 & 2 & 0
\end{array} r a r − 2 0 − 1 4 0 0 1 4 2 2 3 0 4 2 5 0
Thus σ 2 \sigma_2 σ 2 is straight on ( q − 2 , q 2 ) (q_{-2},q_2) ( q − 2 , q 2 ) , and σ 4 \sigma_4 σ 4 is straight on ( q 2 , q 6 ) = ( q − 2 , q 2 ) + T (q_2,q_6)=(q_{-2},q_2)+T ( q 2 , q 6 ) = ( q − 2 , q 2 ) + T . The identity σ 4 ( t + T ) = − σ 2 ( t ) \sigma_4(t+T)=-\sigma_2(t) σ 4 ( t + T ) = − σ 2 ( t ) makes their carrying lines parallel. Taking the common tangent limits at q 2 q_2 q 2 contradicts the same positive determinant as above. □
All coherent four-arc chains
We give the entire four-arc comparison, because comparison on a symmetric subfamily would not suffice for equality rigidity. A coherent four-arc chain has physical labels ( 0 , 2 , 4 , 0 ) (0,2,4,0) ( 0 , 2 , 4 , 0 ) and eight curved boundary arcs. Between them lie eight positive straight segments: each active opposite pair contributes two curved arcs under the sixfold twist, and maximality of the control intervals separates consecutive curved arcs by a positive inactive interval. Consecutive inactive pieces are collinear by C 1 C^1 C 1 matching. Distinct maximal faces lie on distinct support lines, since a face of a convex disk is connected. The four opposite support pairs therefore bound a proper centrally symmetric octagon P P P .
Let S 0 , S 1 , S 2 , S 3 S_0,S_1,S_2,S_3 S 0 , S 1 , S 2 , S 3 be its four strips in cyclic order. At each switch the three tangent support pairs are three consecutive strips; thus the four balanced switch hexagons are H r = ⋂ j ≠ r S j H_r=\bigcap_{j\ne r}S_j H r = ⋂ j = r S j . Omitting a strip adds two opposite exterior triangles. The hexagons have the same area, so those four triangles have the same positive area.
Normalize alternating strip pairs to ∣ x ∣ , ∣ y ∣ ≤ 1 |x|,|y|\le1 ∣ x ∣ , ∣ y ∣ ≤ 1 , and write four successive vertices as
( 1 , a ) , ( b , 1 ) , ( − c , 1 ) , ( − 1 , d ) . (1,a),\quad(b,1),\quad(-c,1),\quad(-1,d). ( 1 , a ) , ( b , 1 ) , ( − c , 1 ) , ( − 1 , d ) .
Set α = 1 − a \alpha=1-a α = 1 − a , β = 1 − b \beta=1-b β = 1 − b , γ = 1 − c \gamma=1-c γ = 1 − c , η = 1 − d \eta=1-d η = 1 − d and D 0 = α γ + β η D_0=\alpha\gamma+\beta\eta D 0 = α γ + β η . Twice the four exterior triangle areas are
α β , γ η , ( a + d ) 2 β γ D 0 , ( b + c ) 2 α η D 0 . (256) \alpha\beta,\qquad\gamma\eta,\qquad\frac{(a+d)^2\beta\gamma}{D_0},\qquad\frac{(b+c)^2\alpha\eta}{D_0}. \tag*{(256)} α β , γ η , D 0 ( a + d ) 2 β γ , D 0 ( b + c ) 2 α η . (256)
Their equality, with all factors positive, gives α β = γ η \alpha\beta=\gamma\eta α β = γ η and ( a + d ) β = ( b + c ) η (a+d)\beta=(b+c)\eta ( a + d ) β = ( b + c ) η . Adding the latter identity to ( 1 − a ) β = ( 1 − c ) η (1-a)\beta=(1-c)\eta ( 1 − a ) β = ( 1 − c ) η yields b = d b=d b = d , and then a = c a=c a = c . The remaining equality is
( 1 − a ) 2 + ( 1 − b ) 2 = ( a + b ) 2 , 1 = a + b + a b . (1-a)^2+(1-b)^2=(a+b)^2,\qquad1=a+b+ab. ( 1 − a ) 2 + ( 1 − b ) 2 = ( a + b ) 2 , 1 = a + b + ab .
Consequently P P P is linearly equivalent to the intersection of two congruent squares,
P p = { ( x , y ) : ∣ x ∣ , ∣ y ∣ , ∣ c x + d y ∣ , ∣ − d x + c y ∣ ≤ 1 } , c 2 + d 2 = 1 , c , d > 0 , p = c − d . (257) P_p=\{(x,y): |x|,|y|,|cx+dy|,|-dx+cy|\le1\},\qquad c^2+d^2=1,\quad c,d>0,\quad p=c-d. \tag*{(257)} P p = {( x , y ) : ∣ x ∣ , ∣ y ∣ , ∣ c x + d y ∣ , ∣ − d x + cy ∣ ≤ 1 } , c 2 + d 2 = 1 , c , d > 0 , p = c − d . (257)
Changing the marking replaces p p p by − p -p − p , so take 0 ≤ p < 1 0\le p<1 0 ≤ p < 1 and put s = c + d = 2 − p 2 s=c+d=\sqrt{2-p^2} s = c + d = 2 − p 2 . The rounding is also forced. At the corner x = 1 x=1 x = 1 , c x + d y = 1 cx+dy=1 c x + d y = 1 , the inactive support lines prescribe asymptotes y = 1 y=1 y = 1 and d x − c y = 1 dx-cy=1 d x − cy = 1 . In coordinates
U = 1 − y , V = 1 − d x + c y , A c = 1 + c − d , U=1-y,\qquad V=1-dx+cy,\qquad A_c=1+c-d, U = 1 − y , V = 1 − d x + cy , A c = 1 + c − d ,
the incident sides become V + c U = A c V+cU=A_c V + c U = A c and U + c V = A c U+cV=A_c U + c V = A c . A hyperbola U V = h UV=h U V = h is tangent to both precisely when
h = A c 2 4 c , ( U , V ) = ( A c 2 c , A c c 2 ) , ( A c 2 , A c 2 c ) . (258) h=\frac{A_c^2}{4c},\qquad(U,V)=\left(\frac{A_c}{2c},\frac{A_c c}{2}\right),\qquad\left(\frac{A_c}{2},\frac{A_c}{2c}\right). \tag*{(258)} h = 4 c A c 2 , ( U , V ) = ( 2 c A c , 2 A c c ) , ( 2 A c , 2 c A c ) . (258)
Both tangencies lie inside their sides. The other type interchanges c c c and d d d . The two neighboring tangencies on x = 1 x=1 x = 1 have ordinates ( s − 1 ) / ( 2 c ) (s-1)/(2c) ( s − 1 ) / ( 2 c ) and − ( s − 1 ) / ( 2 d ) -(s-1)/(2d) − ( s − 1 ) / ( 2 d ) , so every intervening straight segment is positive. The graph V = h / U V=h/U V = h / U is strictly convex and tangent to both incident lines, hence clips exactly the prescribed corner. These eight arcs form a convex regular C 1 C^1 C 1 boundary.
Its unnormalized balanced determinant is
D p = 1 − h c d = 1 − h d c = s + 3 2 ( s + 1 ) . D_p=\frac{1-h_c}{d}=\frac{1-h_d}{c}=\frac{s+3}{2(s+1)}. D p = d 1 − h c = c 1 − h d = 2 ( s + 1 ) s + 3 .
Scaling by λ p 2 = 3 ( s + 1 ) / ( s + 3 ) \lambda_p^2=\sqrt{3}(s+1)/(s+3) λ p 2 = 3 ( s + 1 ) / ( s + 3 ) restores determinant 3 / 2 \sqrt{3}/2 3 /2 . Integration of h / U h/U h / U between the endpoints in (A.42) shows that a c c c -type corner removes area
E c = ( 1 + c − d ) 2 4 c d ( log 1 c − 2 ( 1 − c ) 1 + c ) , E_c=\frac{(1+c-d)^2}{4cd}\left(\log\frac{1}{c}-\frac{2(1-c)}{1+c}\right), E c = 4 c d ( 1 + c − d ) 2 ( log c 1 − 1 + c 2 ( 1 − c ) ) ,
and E d E_d E d is obtained by exchange. The polygon has area 8 / ( s + 1 ) 8/(s+1) 8/ ( s + 1 ) ; there are four corners of each type. Multiplication by λ p 2 / 3 \lambda_p^2/\sqrt{3} λ p 2 / 3 gives the normalized cost
J ( p ) = 2 δ ( p ) , δ ( p ) = 4 s ( s − 1 ) − ( 1 + p 2 ) log 2 1 − p 2 + 4 p artanh ( p / s ) ( s − 1 ) ( s + 3 ) . (259) \mathcal{J}(p)=2\delta(p),\qquad\delta(p)=\frac{4s(s-1)-(1+p^2)\log\frac{2}{1-p^2}+4p\operatorname{artanh}(p/s)}{(s-1)(s+3)}. \tag*{(259)} J ( p ) = 2 δ ( p ) , δ ( p ) = ( s − 1 ) ( s + 3 ) 4 s ( s − 1 ) − ( 1 + p 2 ) log 1 − p 2 2 + 4 p artanh ( p / s ) . (259)
The positive determinant reconstruction also proves that the regular member p = 0 p=0 p = 0 is an admissible competitor. Uniqueness of the prescribed hyperbolas means that no further rounding parameter has been suppressed.
Proposition A.22. Every coherent four-arc chain has J ≥ δ 8 \mathcal{J}\ge\delta_8 J ≥ δ 8 . Equality holds precisely for an affine image of the smoothed regular octagon.
Proof. The function δ \delta δ is even. For 0 < p < 1 0<p<1 0 < p < 1 set λ = log ( 2 / ( 1 − p 2 ) ) \lambda=\log(2/(1-p^2)) λ = log ( 2/ ( 1 − p 2 )) , a = artanh ( p / s ) a=\operatorname{artanh}(p/s) a = artanh ( p / s ) and D = ( s − 1 ) ( s + 3 ) D=(s-1)(s+3) D = ( s − 1 ) ( s + 3 ) . Differentiation gives
δ ′ ( p ) = 2 E s D 2 , E = 2 ( 4 + s − s 3 ) a − p ( s 2 + 3 ) λ + p s ( s − 1 ) 3 s + 1 . \delta'(p)=\frac{2E}{sD^2},\qquad E=2(4+s-s^3)a-p(s^2+3)\lambda+\frac{ps(s-1)^3}{s+1}. δ ′ ( p ) = s D 2 2 E , E = 2 ( 4 + s − s 3 ) a − p ( s 2 + 3 ) λ + s + 1 p s ( s − 1 ) 3 .
If
B 0 = ( 1 − c ) ( 1 + d + c d ) log ( 1 / d ) − ( 1 − d ) ( 1 + c + c d ) log ( 1 / c ) , B_0=(1-c)(1+d+cd)\log(1/d)-(1-d)(1+c+cd)\log(1/c), B 0 = ( 1 − c ) ( 1 + d + c d ) log ( 1/ d ) − ( 1 − d ) ( 1 + c + c d ) log ( 1/ c ) ,
then E = 4 B 0 + p s ( s − 1 ) 3 / ( s + 1 ) E = 4B_0 + ps(s-1)^3/(s+1) E = 4 B 0 + p s ( s − 1 ) 3 / ( s + 1 ) . Put t = d / c ∈ ( 0 , 1 ) t = d/c \in(0,1) t = d / c ∈ ( 0 , 1 ) and q = 1 + t 2 q = \sqrt{1+t^2} q = 1 + t 2 , so c = 1 / q c = 1/q c = 1/ q , d = t / q d = t/q d = t / q . The assertion B 0 > 0 B_0 > 0 B 0 > 0 is
log ( 1 + t 2 ) log ( 1 / t ) < 2 ( q 3 − q 2 + t 3 ) ( 1 − t ) ( 2 q 2 + t ) . \frac{\log(1+t^2)}{\log(1/t)} < \frac{2(q^3-q^2+t^3)}{(1-t)(2q^2+t)}. log ( 1/ t ) log ( 1 + t 2 ) < ( 1 − t ) ( 2 q 2 + t ) 2 ( q 3 − q 2 + t 3 ) .
The positive series for artanh \operatorname{artanh} artanh gives
log ( 1 / t ) > 2 ( 1 − t ) 1 + t , log ( 1 + t 2 ) < U ( t ) ≔ t 2 ( t 4 + 12 t 2 + 12 ) 6 ( 1 + t 2 ) ( 2 + t 2 ) . \log(1/t) > \frac{2(1-t)}{1+t}, \qquad\log(1+t^2) < U(t) \coloneqq\frac{t^2(t^4+12t^2+12)}{6(1+t^2)(2+t^2)}. log ( 1/ t ) > 1 + t 2 ( 1 − t ) , log ( 1 + t 2 ) < U ( t ) : = 6 ( 1 + t 2 ) ( 2 + t 2 ) t 2 ( t 4 + 12 t 2 + 12 ) .
Also q 3 − q 2 + t 3 ≥ t 2 ( t + 2 ( 1 + t 2 ) / ( 4 + t 2 ) ) q^3-q^2+t^3 \ge t^2(t+2(1+t^2)/(4+t^2)) q 3 − q 2 + t 3 ≥ t 2 ( t + 2 ( 1 + t 2 ) / ( 4 + t 2 )) . Subtract U ( t ) ( 1 + t ) / ( 2 ( 1 − t ) ) U(t)(1+t)/(2(1-t)) U ( t ) ( 1 + t ) / ( 2 ( 1 − t )) from the resulting lower bound for the right side. The difference is
t 3 P ( t ) 12 ( 1 + t 2 ) ( 2 + t 2 ) ( 1 − t ) ( 2 + t + 2 t 2 ) ( 4 + t 2 ) . \frac{t^3P(t)}{12(1+t^2)(2+t^2)(1-t)(2+t+2t^2)(4+t^2)}. 12 ( 1 + t 2 ) ( 2 + t 2 ) ( 1 − t ) ( 2 + t + 2 t 2 ) ( 4 + t 2 ) t 3 P ( t ) .
where
P ( t ) = 48 − 24 t + 60 t 2 − 20 t 3 − 2 t 5 − 11 t 6 − 3 t 7 − 2 t 8 > 24 + 22 t 2 > 0. P(t) = 48-24t+60t^2-20t^3-2t^5-11t^6-3t^7-2t^8 > 24+22t^2 > 0. P ( t ) = 48 − 24 t + 60 t 2 − 20 t 3 − 2 t 5 − 11 t 6 − 3 t 7 − 2 t 8 > 24 + 22 t 2 > 0.
Thus B 0 > 0 B_0 > 0 B 0 > 0 , E > 0 E > 0 E > 0 and δ ′ ( p ) > 0 \delta'(p) > 0 δ ′ ( p ) > 0 . At p = 0 p = 0 p = 0 , δ ( 0 ) = ( 8 − 4 2 − log 2 ) / ( 2 2 − 1 ) \delta(0) = (8-4\sqrt{2}-\log2)/(2\sqrt{2}-1) δ ( 0 ) = ( 8 − 4 2 − log 2 ) / ( 2 2 − 1 ) . The equality case follows from the complete polygon and rounding classification above. □
The two five-arc classes
For the algebra below it is convenient to use a fixed diagonal conjugate of the circle coordinates. The reduced state and twist are
X ( u , v ) = 1 v ( u − u 2 − v 2 1 − u ) , R ∗ = 1 2 ( 1 − 3 1 1 ) , − 1 < u < 1 , v > 0 , u 2 + v 2 > 1. (260) X(u,v) = \frac{1}{v}\begin{pmatrix} u & -u^2-v^2 \\ 1 & -u \end{pmatrix}, \qquad R_* = \frac{1}{2}\begin{pmatrix} 1 & -3 \\ 1 & 1 \end{pmatrix}, \qquad-1 < u < 1,\quad v > 0,\quad u^2+v^2 > 1. \tag*{(260)} X ( u , v ) = v 1 ( u 1 − u 2 − v 2 − u ) , R ∗ = 2 1 ( 1 1 − 3 1 ) , − 1 < u < 1 , v > 0 , u 2 + v 2 > 1. (260)
The change is a conjugacy, so group closure, normalized area and the fixed-endpoint homotopy class are unchanged. Put
P ( s ) = 1 + s 2 1 − s 2 , Q ( s ) = 1 + s 2 2 s , L ( s , t ) = 4 s t ( 1 − s 2 ) ( 1 − t 2 ) . P(s) = \frac{1+s^2}{1-s^2}, \qquad Q(s) = \frac{1+s^2}{2s}, \qquad L(s,t) = \frac{4st}{(1-s^2)(1-t^2)}. P ( s ) = 1 − s 2 1 + s 2 , Q ( s ) = 2 s 1 + s 2 , L ( s , t ) = ( 1 − s 2 ) ( 1 − t 2 ) 4 s t .
All arc exponents below are the positive real exponents in the exact vertex-control flow. They exceed one on positive arcs.
Lemma A.23. Every actual strict five-arc chain of reduced word 01202 is represented by 0 < a , r < 1 0 < a,r < 1 0 < a , r < 1 , ( 1 + a ) ( 1 + r ) > 2 (1+a)(1+r) > 2 ( 1 + a ) ( 1 + r ) > 2 , with exponents
( P ( a ) , Q ( a ) , Q ( r ) , P ( r ) , L ( a , r ) ) . (P(a), Q(a), Q(r), P(r), L(a,r)). ( P ( a ) , Q ( a ) , Q ( r ) , P ( r ) , L ( a , r )) . Putting x = 2 a / ( 1 − a ) x = 2a/(1-a) x = 2 a / ( 1 − a ) and y = 2 r / ( 1 − r ) y = 2r/(1-r) y = 2 r / ( 1 − r ) , its cost is
J = 2 − G ( x ) + G ( y ) x y + 2 x + 2 y + 3 , x y > 2. (261) \mathcal{J} = 2 - \frac{G(x)+G(y)}{xy+2x+2y+3}, \qquad xy > 2. \tag*{(261)} J = 2 − x y + 2 x + 2 y + 3 G ( x ) + G ( y ) , x y > 2. (261) where, for t = x + 1 t = x+1 t = x + 1 ,
G ( x ) = 2 ( t 2 + 1 ) ( t − 1 ) 2 log t 2 + 1 2 t + ( t 2 − 1 ) log t 2 + 1 t 2 − 1 − 3. G(x) = \frac{2(t^2+1)}{(t-1)^2}\log\frac{t^2+1}{2t} + (t^2-1)\log\frac{t^2+1}{t^2-1} - 3. G ( x ) = ( t − 1 ) 2 2 ( t 2 + 1 ) log 2 t t 2 + 1 + ( t 2 − 1 ) log t 2 − 1 t 2 + 1 − 3.
Proof. Let Q u = ( 1 u 0 1 ) Q_u = \begin{pmatrix}1 & u \\ 0 & 1\end{pmatrix} Q u = ( 1 0 u 1 ) and N b = ( 1 0 b 1 ) N_b = \begin{pmatrix}1 & 0 \\ b & 1\end{pmatrix} N b = ( 1 b 0 1 ) . On a 0 or 2 arc,
G o u t Q u o u t = G i n Q u i n N b , b = ∫ d u v 2 . (262) G_{\mathrm{out}}Q_{u_{\mathrm{out}}} = G_{\mathrm{in}}Q_{u_{\mathrm{in}}}N_b,\qquad b = \int\frac{\mathrm{d}u}{v^2}. \tag*{(262)} G out Q u out = G in Q u in N b , b = ∫ v 2 d u . (262)
In particular a suffix using only these two modes fixes the second column of G Q u GQ_u G Q u . Let p , q > 1 p,q > 1 p , q > 1 be the first two exponents, w = v 0 2 w = v_0^2 w = v 0 2 , Δ 0 = ( 1 − u 0 ) 2 + w \Delta_0 = (1-u_0)^2+w Δ 0 = ( 1 − u 0 ) 2 + w , and
D 2 = p 2 ( q − 1 ) 2 ( ( u 0 + 1 ) 2 + w ) − 4 p q ( q − 1 ) ( u 0 + 1 ) + 4 q 2 . D_2 = p^2(q-1)^2((u_0+1)^2+w)-4pq(q-1)(u_0+1)+4q^2. D 2 = p 2 ( q − 1 ) 2 (( u 0 + 1 ) 2 + w ) − 4 pq ( q − 1 ) ( u 0 + 1 ) + 4 q 2 .
The second-column closure equations have numerators N 1 , N 2 N_1,N_2 N 1 , N 2 satisfying
N 2 + 4 w ( p 2 q 2 − p 2 − q 2 ) = ( 1 − u 0 ) N 1 . N_2+4w(p^2q^2-p^2-q^2)=(1-u_0)N_1. N 2 + 4 w ( p 2 q 2 − p 2 − q 2 ) = ( 1 − u 0 ) N 1 .
Here v 2 = 4 p q v 0 / D 2 > 0 v_2=4pqv_0/D_2>0 v 2 = 4 pq v 0 / D 2 > 0 , so all divided quantities are nonzero. Thus p 2 q 2 = p 2 + q 2 p^2q^2=p^2+q^2 p 2 q 2 = p 2 + q 2 . With a = q ( 1 − 1 / p ) a=q(1-1/p) a = q ( 1 − 1/ p ) this gives p = P ( a ) p=P(a) p = P ( a ) , q = Q ( a ) q=Q(a) q = Q ( a ) and w = ( u 0 + 1 ) ( 1 + 2 a − u 0 ) w=(u_0+1)(1+2a-u_0) w = ( u 0 + 1 ) ( 1 + 2 a − u 0 ) . Moreover v 2 / ( 1 + u 2 ) = v T / ( 1 − u T ) = : B 0 > 0 v_2/(1+u_2)=v_T/(1-u_T)=:B_0>0 v 2 / ( 1 + u 2 ) = v T / ( 1 − u T ) =: B 0 > 0 .
Write the three suffix lines as v = B 1 ( 1 − u ) v=B_1(1-u) v = B 1 ( 1 − u ) , v = A 0 ( 1 + u ) v=A_0(1+u) v = A 0 ( 1 + u ) , v = B 0 ( 1 − u ) v=B_0(1-u) v = B 0 ( 1 − u ) . Their shear residual is exactly
b a c t u a l − b r e q u i r e d = − 2 v 2 ( u 2 A 0 + 1 B 0 ) . b_{\mathrm{actual}}-b_{\mathrm{required}}=-\frac{2}{v_2}\left(\frac{u_2}{A_0}+\frac{1}{B_0}\right). b actual − b required = − v 2 2 ( A 0 u 2 + B 0 1 ) .
It vanishes precisely when A 0 = − u 2 B 0 A_0=-u_2B_0 A 0 = − u 2 B 0 , and forces u 2 < 0 u_2<0 u 2 < 0 . Set r = − u 2 ∈ ( 0 , 1 ) r=-u_2\in(0,1) r = − u 2 ∈ ( 0 , 1 ) . Solving the linear equation for the entrance state gives, with D = ( r − 1 ) ( a 3 − a + 1 ) − a 2 ( r + 3 ) < 0 D=(r-1)(a^3-a+1)-a^2(r+3)<0 D = ( r − 1 ) ( a 3 − a + 1 ) − a 2 ( r + 3 ) < 0 ,
u 0 = − a 3 r − a 3 + a 2 r − 5 a 2 − a r + a − r + 1 D , u_0=-\frac{a^3r-a^3+a^2r-5a^2-ar+a-r+1}{D}, u 0 = − D a 3 r − a 3 + a 2 r − 5 a 2 − a r + a − r + 1 ,
v 0 2 = − 4 a 2 ( a − 1 ) ( a + 1 ) 2 ( r − 1 ) ( a r − a − r − 3 ) D 2 . v_0^2=-\frac{4a^2(a-1)(a+1)^2(r-1)(ar-a-r-3)}{D^2}. v 0 2 = − D 2 4 a 2 ( a − 1 ) ( a + 1 ) 2 ( r − 1 ) ( a r − a − r − 3 ) .
The suffix intersections now give its three displayed exponents. Their positivity reduces to ( 1 + a ) ( 1 + r ) > 2 (1+a)(1+r)>2 ( 1 + a ) ( 1 + r ) > 2 . Integrating the original arc cost, and putting z = ( 1 − a ) ( 1 − r ) z=(1-a)(1-r) z = ( 1 − a ) ( 1 − r ) , gives
J = 4 ( z + 2 ) 4 − z − z 4 − z [ f 0 ( a ) + f 0 ( r ) + log L ( a , r ) ] , f 0 ( s ) = log P ( s ) s 2 + ( 1 + s 1 − s ) 2 log Q ( s ) . \mathcal{J}=\frac{4(z+2)}{4-z}-\frac{z}{4-z}\left[f_0(a)+f_0(r)+\log L(a,r)\right],\qquad f_0(s)=\frac{\log P(s)}{s^2}+\left(\frac{1+s}{1-s}\right)^2\log Q(s). J = 4 − z 4 ( z + 2 ) − 4 − z z [ f 0 ( a ) + f 0 ( r ) + log L ( a , r ) ] , f 0 ( s ) = s 2 log P ( s ) + ( 1 − s 1 + s ) 2 log Q ( s ) .
Substitution of x , y x,y x , y gives (A.45). The argument uses necessary identities of the full group closure; it never eliminates an exceptional return by a nonsingularity assumption. □
Lemma A.24. The function h ( x ) = ( x + 2 ) G ′ ( x ) h(x)=(x+2)G'(x) h ( x ) = ( x + 2 ) G ′ ( x ) satisfies
h ( x ) > 2 ( 0 < x < 2 ) , h ( 2 ) = 2 , 0 < h ( x ) < 2 ( x > 2 ) . h(x)>2\quad(0<x<\sqrt{2}),\qquad h(\sqrt{2})=2,\qquad0<h(x)<2\quad(x>\sqrt{2}). h ( x ) > 2 ( 0 < x < 2 ) , h ( 2 ) = 2 , 0 < h ( x ) < 2 ( x > 2 ) . Proof. Put t = x + 1 t=x+1 t = x + 1 , L 0 = log ( ( t 2 + 1 ) / ( 2 t ) ) L_0=\log((t^2+1)/(2t)) L 0 = log (( t 2 + 1 ) / ( 2 t )) , M 0 = log ( ( t 2 + 1 ) / ( t 2 − 1 ) ) M_0=\log((t^2+1)/(t^2-1)) M 0 = log (( t 2 + 1 ) / ( t 2 − 1 )) and K 0 = t 2 − 2 t − 1 K_0=t^2-2t-1 K 0 = t 2 − 2 t − 1 . Direct differentiation yields
h − 2 = A L 0 + B M 0 + C , A = − 4 ( t + 1 ) 2 ( t − 1 ) 3 , B = 2 t ( t + 1 ) , C = − 2 K 0 ( 2 t 2 + t + 1 ) t ( t − 1 ) ( t 2 + 1 ) . h-2=AL_0+BM_0+C,\qquad A=-\frac{4(t+1)^2}{(t-1)^3},\qquad B=2t(t+1),\qquad C=-\frac{2K_0(2t^2+t+1)}{t(t-1)(t^2+1)}. h − 2 = A L 0 + B M 0 + C , A = − ( t − 1 ) 3 4 ( t + 1 ) 2 , B = 2 t ( t + 1 ) , C = − t ( t − 1 ) ( t 2 + 1 ) 2 K 0 ( 2 t 2 + t + 1 ) . The bounds L 0 < ( t − 1 ) 2 / ( 2 t ) L_0<(t-1)^2/(2t) L 0 < ( t − 1 ) 2 / ( 2 t ) and M 0 > 2 / t 2 M_0>2/t^2 M 0 > 2/ t 2 give h > 4 ( t + 1 ) / ( t ( t 2 + 1 ) ) > 0 h>4(t+1)/(t(t^2+1))>0 h > 4 ( t + 1 ) / ( t ( t 2 + 1 )) > 0 . For 1 < t ≤ 2 1<t\le2 1 < t ≤ 2 , set b = ( t − 1 ) / ( t + 1 ) b=(t-1)/(t+1) b = ( t − 1 ) / ( t + 1 ) and use L 0 < 2 b 2 + 2 b 6 / ( 3 ( 1 − b 4 ) ) L_0<2b^2+2b^6/(3(1-b^4)) L 0 < 2 b 2 + 2 b 6 / ( 3 ( 1 − b 4 )) . Then
h − 2 > − t 3 + 3 t 2 − 9 t − 7 3 t ( t 2 + 1 ) > 0 , h-2>-\frac{t^3+3t^2-9t-7}{3t(t^2+1)}>0, h − 2 > − 3 t ( t 2 + 1 ) t 3 + 3 t 2 − 9 t − 7 > 0 , since the numerator is increasing on [ 1 , 2 ] [1,2] [ 1 , 2 ] and equals − 5 -5 − 5 at 2 2 2 . For t ≥ 2 t\ge2 t ≥ 2 use R = ( t 2 − 1 ) / ( 2 t ) R=(t^2-1)/(2t) R = ( t 2 − 1 ) / ( 2 t ) and
log R K 0 > 2 t 2 + 2 t − 1 , M 0 < 2 t 2 + 2 3 t 2 ( t 4 − 1 ) . \frac{\log R}{K_0}>\frac{2}{t^2+2t-1},\qquad M_0<\frac{2}{t^2}+\frac{2}{3t^2(t^4-1)}. K 0 log R > t 2 + 2 t − 1 2 , M 0 < t 2 2 + 3 t 2 ( t 4 − 1 ) 2 . The first inequality is the integral inequality log R / ( R − 1 ) > 2 / ( R + 1 ) \log R/(R-1)>2/(R+1) log R / ( R − 1 ) > 2/ ( R + 1 ) . Substitution into
h − 2 K 0 = A log R K 0 + 2 ( t + 1 ) ( t 2 − t + 2 ) ( t − 1 ) 3 M 0 − 2 ( 2 t 2 + t + 1 ) t ( t − 1 ) ( t 2 + 1 ) \frac{h-2}{K_0}=A\frac{\log R}{K_0}+\frac{2(t+1)(t^2-t+2)}{(t-1)^3}M_0-\frac{2(2t^2+t+1)}{t(t-1)(t^2+1)} K 0 h − 2 = A K 0 log R + ( t − 1 ) 3 2 ( t + 1 ) ( t 2 − t + 2 ) M 0 − t ( t − 1 ) ( t 2 + 1 ) 2 ( 2 t 2 + t + 1 ) gives a strictly negative upper bound whose numerator is − 2 p ( t ) -2p(t) − 2 p ( t ) and whose denominator is positive. The positivity of p p p follows from
p ( v + 2 ) = 3 v 7 + 36 v 6 + 183 v 5 + 508 v 4 + 831 v 3 + 782 v 2 + 363 v + 46. p(v+2)=3v^7+36v^6+183v^5+508v^4+831v^3+782v^2+363v+46. p ( v + 2 ) = 3 v 7 + 36 v 6 + 183 v 5 + 508 v 4 + 831 v 3 + 782 v 2 + 363 v + 46. Thus ( h − 2 ) / K 0 < 0 (h-2)/K_0<0 ( h − 2 ) / K 0 < 0 . At the sole zero t = 1 + 2 t=1+\sqrt{2} t = 1 + 2 of K 0 K_0 K 0 , L 0 = M 0 = log 2 L_0=M_0=\log\sqrt{2} L 0 = M 0 = log 2 and A + B = C = 0 A+B=C=0 A + B = C = 0 . This proves the assertion. □
Proposition A.25. Every strict five-arc chain of word 01202 satisfies J > J 8 \mathcal{J}>\mathcal{J}_8 J > J 8 .
Proof. An interior stationary point of (A.45) must have h ( x ) = h ( y ) h(x)=h(y) h ( x ) = h ( y ) . Since x y > 2 xy>2 x y > 2 , Lemma A.24 then forces x , y > 2 x,y>\sqrt{2} x , y > 2 . But
G ( 2 ) = ( 3 + 2 2 ) log 2 − 3 > 1 , G(\sqrt{2})=(3+2\sqrt{2})\log2-3>1, G ( 2 ) = ( 3 + 2 2 ) log 2 − 3 > 1 ,
using log 2 > 56 / 81 \log2>56/81 log 2 > 56/81 and 2 > 39 / 28 \sqrt{2}>39/28 2 > 39/28 . Since G ′ > 0 G'>0 G ′ > 0 , the numerator of J x \mathcal{J}_x J x is
( y + 2 ) ( G ( x ) + G ( y ) − h ( x ) ) + G ′ ( x ) > 0 , (y+2)(G(x)+G(y)-h(x))+G'(x)>0, ( y + 2 ) ( G ( x ) + G ( y ) − h ( x )) + G ′ ( x ) > 0 ,
a contradiction. On the boundary x y = 2 xy=2 x y = 2 , put a = x / ( x + 2 ) a=x/(x+2) a = x / ( x + 2 ) , r = ( 1 − a ) / ( 1 + a ) r=(1-a)/(1+a) r = ( 1 − a ) / ( 1 + a ) and
c = 1 − a 2 1 + a 2 , d = 2 a 1 + a 2 , p = c − d . c=\frac{1-a^2}{1+a^2},\qquad d=\frac{2a}{1+a^2},\qquad p=c-d. c = 1 + a 2 1 − a 2 , d = 1 + a 2 2 a , p = c − d .
Substitution gives J ( x , 2 / x ) = 2 δ ( p ) \mathcal{J}(x,2/x)=2\delta(p) J ( x , 2/ x ) = 2 δ ( p ) , with the exact function in (A.43). Hence the boundary value is at least J 8 \mathcal{J}_8 J 8 , with equality only at x = y = 2 x=y=\sqrt{2} x = y = 2 . Finally G ( x ) → − 1 G(x)\to-1 G ( x ) → − 1 as x ↓ 0 x\downarrow0 x ↓ 0 , and − 1 < G ( x ) < 10 log ( x + 1 ) − 1 -1<G(x)<10\log(x+1)-1 − 1 < G ( x ) < 10 log ( x + 1 ) − 1 for x ≥ 1 x\ge1 x ≥ 1 . For an escaping sequence with x y ≥ 2 xy\ge2 x y ≥ 2 , Q = max ( x , y ) → ∞ Q=\max(x,y)\to\infty Q = max ( x , y ) → ∞ and
∣ G ( x ) + G ( y ) ∣ x y + 2 x + 2 y + 3 ≤ 2 + 20 log ( Q + 1 ) 2 Q + 3 ⟶ 0. \frac{\lvert G(x)+G(y)\rvert}{xy+2x+2y+3}\le\frac{2+20\log(Q+1)}{2Q+3}\longrightarrow0. x y + 2 x + 2 y + 3 ∣ G ( x ) + G ( y )∣ ≤ 2 Q + 3 2 + 20 log ( Q + 1 ) ⟶ 0.
Thus the cost tends to 2 2 2 at infinity. Any interior point with cost at most J 8 \mathcal{J}_8 J 8 would give an interior attained minimum, already excluded by the stationary equations. □
Proposition A.26. Every strict five-arc chain of word 02102 satisfies J > 29 / 16 > J 8 \mathcal{J}>29/16>\mathcal{J}_8 J > 29/16 > J 8 .
Proof. Write the word as 02|1|02 and use (A.46) on both sides. If the middle mode-1 arc has entrance ( x , V 0 ) (x,V_0) ( x , V 0 ) , exit ( z , W 0 ) (z,W_0) ( z , W 0 ) and exponent q q q , its matrix in the Q Q Q frame has
C 12 = − V 0 W 0 sinh log q , C 22 = ( W 0 / V 0 ) ( cosh log q − x sinh log q ) . C_{12}=-V_0W_0\sinh\log q,\qquad C_{22}=(W_0/V_0)(\cosh\log q-x\sinh\log q). C 12 = − V 0 W 0 sinh log q , C 22 = ( W 0 / V 0 ) ( cosh log q − x sinh log q ) . The equations for the upper-right and lower-right entries of the full matrix closure give q i 2 = ( q i − 1 2 − 1 ) ( q i + 1 2 − 1 ) q_i^2=(q_{i-1}^2-1)(q_{i+1}^2-1) q i 2 = ( q i − 1 2 − 1 ) ( q i + 1 2 − 1 ) at every cyclic cut. For example, putting e = ( q − 1 ) / ( q + 1 ) e=(q-1)/(q+1) e = ( q − 1 ) / ( q + 1 ) , k = V 0 / v 0 k=V_0/v_0 k = V 0 / v 0 and K 0 = 2 / q 1 K_0=2/q_1 K 0 = 2/ q 1 , the second equation after the first is used is 4 k 2 − K 0 2 − ( 1 − e ) 2 / e = 0 4k^2-K_0^2-(1-e)^2/e=0 4 k 2 − K 0 2 − ( 1 − e ) 2 / e = 0 . Its divisors are positive because all arcs are positive. With p i = q i 2 − 1 p_i=q_i^2-1 p i = q i 2 − 1 we obtain
p i − 1 p i + 1 = 1 + p i , ( p 0 , … , p 4 ) = ( A , B , 1 + B A , 1 + A + B A B , 1 + A B ) . p_{i-1}p_{i+1}=1+p_i,\qquad(p_0,\ldots,p_4)=\left(A,B,\frac{1+B}{A},\frac{1+A+B}{AB},\frac{1+A}{B}\right). p i − 1 p i + 1 = 1 + p i , ( p 0 , … , p 4 ) = ( A , B , A 1 + B , A B 1 + A + B , B 1 + A ) . Let S = ∑ q i S=\sum q_i S = ∑ q i , T = ∑ q i q i + 2 T=\sum q_iq_{i+2} T = ∑ q i q i + 2 and K 0 = 2 + S + T K_0=2+S+T K 0 = 2 + S + T . The state transfer is
T − ( q ) = 1 2 ( q q + 2 − q 2 − q ) . T_-(q)=\frac{1}{2}\begin{pmatrix}q&q+2\\-q&2-q\end{pmatrix}. T − ( q ) = 2 1 ( q − q q + 2 2 − q ) . For its fivefold product, direct multiplication using the displayed recurrence gives
Δ = 4 ∏ q i − ( tr ∏ T − ( q i ) ) 2 = 4 K 0 ∏ ( q i − 1 ) > 0 , d i 2 = q i + 1 K 0 ( q i − 1 ) . \Delta=4\prod q_i-\left(\operatorname{tr}\prod T_-(q_i)\right)^2=4K_0\prod(q_i-1)>0,\qquad d_i^2=\frac{q_i+1}{K_0(q_i-1)}. Δ = 4 ∏ q i − ( tr ∏ T − ( q i ) ) 2 = 4 K 0 ∏ ( q i − 1 ) > 0 , d i 2 = K 0 ( q i − 1 ) q i + 1 . Thus the scalar-return case is excluded by an identity, rather than by an assumed regularity of the endpoint map. The original single-arc rational cost is ( q i − 1 ) d i ( 2 d i + d i − 1 + d i + 1 ) (q_i-1)d_i(2d_i+d_{i-1}+d_{i+1}) ( q i − 1 ) d i ( 2 d i + d i − 1 + d i + 1 ) . Its sum is 2 + 6 / K 0 2+6/K_0 2 + 6/ K 0 . One verifies the cross terms by writing q = ( p , r , c / a , p r / ( a b ) , c / b ) q=(p,r,c/a,pr/(ab),c/b) q = ( p , r , c / a , p r / ( ab ) , c / b ) , where a 2 = p 2 − 1 a^2=p^2-1 a 2 = p 2 − 1 , b 2 = r 2 − 1 b^2=r^2-1 b 2 = r 2 − 1 , c 2 = p 2 + r 2 − 1 c^2=p^2+r^2-1 c 2 = p 2 + r 2 − 1 , and taking all roots positive. The logarithmic coefficient on arc i i i is − d i 2 -d_i^2 − d i 2 , so
J = 2 − ∑ i f ( q i ) − 6 2 + S + T , f ( q ) = q + 1 q − 1 log q . \mathcal{J}=2-\frac{\sum_i f(q_i)-6}{2+S+T},\qquad f(q)=\frac{q+1}{q-1}\log q. J = 2 − 2 + S + T ∑ i f ( q i ) − 6 , f ( q ) = q − 1 q + 1 log q . The positive logarithm series gives f ( q ) < ( q + q − 1 + 10 ) / 6 f(q)<(q+q^{-1}+10)/6 f ( q ) < ( q + q − 1 + 10 ) /6 . Multiplying the recurrence gives ∏ ( q i − q i − 1 ) = 1 \prod(q_i-q_i^{-1})=1 ∏ ( q i − q i − 1 ) = 1 , hence S − ∑ q i − 1 ≥ 5 S-\sum q_i^{-1}\ge5 S − ∑ q i − 1 ≥ 5 . If Q = ( ∏ q i ) 1 / 5 Q=(\prod q_i)^{1/5} Q = ( ∏ q i ) 1/5 , convexity of z ↦ log ( 1 + e z ) z\mapsto\log(1+e^z) z ↦ log ( 1 + e z ) gives Q 2 ≥ Q + 1 Q^2\ge Q+1 Q 2 ≥ Q + 1 , and therefore T ≥ 5 Q 2 > 13 T\ge5Q^2>13 T ≥ 5 Q 2 > 13 . Also T > 2 S − 5 T>2S-5 T > 2 S − 5 by expansion of ∑ ( q i − 1 ) ( q i + 2 − 1 ) > 0 \sum(q_i-1)(q_{i+2}-1)>0 ∑ ( q i − 1 ) ( q i + 2 − 1 ) > 0 . For S ≤ 9 S\le9 S ≤ 9 use the first bound, and for S ≥ 9 S\ge9 S ≥ 9 the second, to get 9 T ≥ 7 S + 54 9T\ge7S+54 9 T ≥ 7 S + 54 . Consequently
S + ∑ q i − 1 + 14 6 ( 2 + S + T ) < 3 16 , \frac{S+\sum q_i^{-1}+14}{6(2+S+T)}<\frac{3}{16}, 6 ( 2 + S + T ) S + ∑ q i − 1 + 14 < 16 3 ,
which proves the cost bound. Finally 186 2 + 32 log 2 − 285 > 209 / 2349 > 0 186\sqrt{2}+32\log2-285>209/2349>0 186 2 + 32 log 2 − 285 > 209/2349 > 0 , using 2 > 41 / 29 \sqrt{2}>41/29 2 > 41/29 and log 2 > 56 / 81 \log2>56/81 log 2 > 56/81 , is exactly 29 / 16 > J 8 29/16>\mathcal{J}_8 29/16 > J 8 . □
The two six-arc classes
The following use of a parameter family is local. At an actual strict chain, positive durations and strict star are open conditions. An exact nearby family with the same full group and twisted state endpoints preserves the fixed-endpoint homotopy class of the group path: for close paths their pointwise quotient has a real matrix logarithm, which gives an endpoint-fixed homotopy. The curvature identities in the original model then give a regular convex boundary with the same winding. Consequently, a nonzero derivative of the original cost in such a family excludes local minimality. This argument requires neither a globally connected parameter domain nor a generic rank assumption.
Proposition A.27. An actual strict six-arc chain of word 010202 is not a local area minimum, even among chains with the same six controls.
Proof. The first two arcs have the same second-column calculation as in Lemma A.23 . Their exponents are P ( a ) , Q ( a ) P(a),Q(a) P ( a ) , Q ( a ) , 0 < a < 1 0<a<1 0 < a < 1 . The suffix lines can be written
v = B 0 ( 1 + u ) , v = s B 0 ( 1 − u ) , v = r B 0 ( 1 + u ) , v = B 0 ( 1 − u ) . v=B_0(1+u),\qquad v=sB_0(1-u),\qquad v=rB_0(1+u),\qquad v=B_0(1-u). v = B 0 ( 1 + u ) , v = s B 0 ( 1 − u ) , v = r B 0 ( 1 + u ) , v = B 0 ( 1 − u ) . where r , s > 0 r,s>0 r , s > 0 . Its shear, multiplied by B 0 2 B_0^2 B 0 2 , is
1 1 + u 2 + 1 1 − u T + 1 r s − 1 r − 1 s − 1. \frac{1}{1+u_2}+\frac{1}{1-u_T}+\frac{1}{rs}-\frac{1}{r}-\frac{1}{s}-1. 1 + u 2 1 + 1 − u T 1 + r s 1 − r 1 − s 1 − 1. The required shear consists of the first two terms. Thus 1 − r − s − r s = 0 1-r-s-rs=0 1 − r − s − r s = 0 , so s = ( 1 − r ) / ( 1 + r ) s=(1-r)/(1+r) s = ( 1 − r ) / ( 1 + r ) and 0 < r < 1 0<r<1 0 < r < 1 . The first suffix switch is u 3 = − r u_3=-r u 3 = − r . Positive duration gives u 2 = − z u_2=-z u 2 = − z with r < z < 1 r<z<1 r < z < 1 . Set
k = ( 1 − a 2 ) ( 1 − z ) 2 a , B 0 2 = 3 + a + ( 1 − a ) z ( 1 − a ) ( 1 − z ) . k=\frac{(1-a^2)(1-z)}{2a},\qquad B_0^2=\frac{3+a+(1-a)z}{(1-a)(1-z)}. k = 2 a ( 1 − a 2 ) ( 1 − z ) , B 0 2 = ( 1 − a ) ( 1 − z ) 3 + a + ( 1 − a ) z . The six exponents are
( P ( a ) , Q ( a ) , 1 − r 1 − z , Q ( r ) , P ( r ) , 2 r ( 1 + r ) k ) . \left(P(a),Q(a),\frac{1-r}{1-z},Q(r),P(r),\frac{2r}{(1+r)k}\right). ( P ( a ) , Q ( a ) , 1 − z 1 − r , Q ( r ) , P ( r ) , ( 1 + r ) k 2 r ) . Their strict domain is 0 < a , z , r < 1 0<a,z,r<1 0 < a , z , r < 1 , r < z r<z r < z , k ( 1 + r ) < 2 r k(1+r)<2r k ( 1 + r ) < 2 r , with the additional suffix star condition r s B 0 2 > 1 rsB_0^2>1 r s B 0 2 > 1 . The entrance formulas are those of Lemma A.23 with r r r there replaced by z z z . Thus these expressions are an exact three-parameter full-closure family around every actual point. The mode-1 star check follows from u θ ′ = ( u 2 − v 2 − 1 ) / 2 u'_\theta=(u^2-v^2-1)/2 u θ ′ = ( u 2 − v 2 − 1 ) /2 : at u = ± 1 u=\pm1 u = ± 1 this is − v 2 / 2 < 0 -v^2/2<0 − v 2 /2 < 0 , so a trajectory with both endpoints in ( − 1 , 1 ) (-1,1) ( − 1 , 1 ) cannot leave and return. The 0 and 2 checks follow from their monotone horizontal coordinates; the displayed suffix condition checks their smallest star value.
Define
F ( s ) = 1 + s 2 s 2 log P ( s ) + 4 s ( 1 − s ) 2 log Q ( s ) , F(s)=\frac{1+s^2}{s^2}\log P(s)+\frac{4s}{(1-s)^2}\log Q(s), F ( s ) = s 2 1 + s 2 log P ( s ) + ( 1 − s ) 2 4 s log Q ( s ) , T ( s ) = F ( s ) + 2 log ( 1 − s ) , ω = ( 1 − a ) ( 1 − z ) 4 − ( 1 − a ) ( 1 − z ) . T(s)=F(s)+2\log(1-s),\qquad\omega=\frac{(1-a)(1-z)}{4-(1-a)(1-z)}. T ( s ) = F ( s ) + 2 log ( 1 − s ) , ω = 4 − ( 1 − a ) ( 1 − z ) ( 1 − a ) ( 1 − z ) . The full original arc integral simplifies to
J = 2 + ω [ 6 − F ( a ) − T ( r ) + 2 log ( 1 − z ) ] . (263) \mathcal{J}=2+\omega[6-F(a)-T(r)+2\log(1-z)]. \tag*{(263)} J = 2 + ω [ 6 − F ( a ) − T ( r ) + 2 log ( 1 − z )] . (263) For a direct check, its rational part is 2 + 6 ω 2+6\omega 2 + 6 ω and the six logarithmic weights are − ω ( a − 2 , ( ( 1 + a ) / ( 1 − a ) ) 2 , 1 , ( ( 1 + r ) / ( 1 − r ) ) 2 , r − 2 , 1 ) -\omega(a^{-2},((1+a)/(1-a))^2,1,((1+r)/(1-r))^2,r^{-2},1) − ω ( a − 2 , (( 1 + a ) / ( 1 − a ) ) 2 , 1 , (( 1 + r ) / ( 1 − r ) ) 2 , r − 2 , 1 ) . Their combination uses only
Q ( a ) Q ( r ) E 2 E 5 P ( a ) P ( r ) = ( 1 − r ) 2 ( 1 − z ) 2 . \frac{Q(a)Q(r)E_2E_5}{P(a)P(r)}=\frac{(1-r)^2}{(1-z)^2}. P ( a ) P ( r ) Q ( a ) Q ( r ) E 2 E 5 = ( 1 − z ) 2 ( 1 − r ) 2 .
Since G ( x ) = F ( x / ( x + 2 ) ) − 3 G(x)=F(x/(x+2))-3 G ( x ) = F ( x / ( x + 2 )) − 3 , Lemma A.24 gives
F ′ ( s ) = h ( 2 s / ( 1 − s ) ) 1 − s , T ′ ( s ) = h ( 2 s / ( 1 − s ) ) − 2 1 − s . F'(s)=\frac{h(2s/(1-s))}{1-s},\qquad T'(s)=\frac{h(2s/(1-s))-2}{1-s}. F ′ ( s ) = 1 − s h ( 2 s / ( 1 − s )) , T ′ ( s ) = 1 − s h ( 2 s / ( 1 − s )) − 2 .
Thus T ′ T' T ′ vanishes only at ρ = 2 − 1 \rho=\sqrt{2}-1 ρ = 2 − 1 . At a stationary point, (A.47) first forces r = ρ r=\rho r = ρ . Put t = 1 − a t=1-a t = 1 − a , q = 1 − z q=1-z q = 1 − z , H = 4 − t q H=4-tq H = 4 − tq and W = 6 − F ( a ) − T ( r ) + 2 log q W=6-F(a)-T(r)+2\log q W = 6 − F ( a ) − T ( r ) + 2 log q . Then
J z = − 2 t ( 2 W + H ) H 2 , J a = − 4 q W H 2 − t q H F ′ ( a ) . \mathcal{J}_z=-\frac{2t(2W+H)}{H^2},\qquad\mathcal{J}_a=-\frac{4qW}{H^2}-\frac{tq}{H}F'(a). J z = − H 2 2 t ( 2 W + H ) , J a = − H 2 4 q W − H tq F ′ ( a ) .
The first equation forces W = − H / 2 W=-H/2 W = − H /2 , and the second then forces a = ρ a=\rho a = ρ . Since z > r = ρ z>r=\rho z > r = ρ , 0 < q < 1 − ρ 0<q<1-\rho 0 < q < 1 − ρ . The remaining equation would be
Ψ ( q ) = 16 − 4 F ( ρ ) + 4 log q 1 − ρ − ( 1 − ρ ) q = 0. \Psi(q)=16-4F(\rho)+4\log\frac{q}{1-\rho}-(1-\rho)q=0. Ψ ( q ) = 16 − 4 F ( ρ ) + 4 log 1 − ρ q − ( 1 − ρ ) q = 0.
But Ψ ′ ( q ) > 0 \Psi'(q)>0 Ψ ′ ( q ) > 0 and Ψ ( q ) < Ψ ( 1 − ρ ) = 16 − 4 F ( ρ ) − ( 1 − ρ ) 2 < 0 \Psi(q)<\Psi(1-\rho)=16-4F(\rho)-(1-\rho)^2<0 Ψ ( q ) < Ψ ( 1 − ρ ) = 16 − 4 F ( ρ ) − ( 1 − ρ ) 2 < 0 , because F ( ρ ) = G ( 2 ) + 3 > 4 F(\rho)=G(\sqrt{2})+3>4 F ( ρ ) = G ( 2 ) + 3 > 4 . There is no stationary point in the open actual domain. □
The other six-arc word requires a finite inequality. We state its polynomial and interval witnesses explicitly, so that the computational input is a finite assertion about specified functions and domains.
Lemma A.28. Every actual strict six-arc chain of word 021020 belongs locally to the following exact full-closure family, with a , b , t ∈ ( 0 , 1 ) a,b,t\in(0,1) a , b , t ∈ ( 0 , 1 ) :
E = ( P ( a ) , L ( a , b ) , P ( b ) , L ( b , t ) , P ( t ) , Q ( a ) Q ( t ) P ( b ) ) , (264) E=\left(P(a),L(a,b),P(b),L(b,t),P(t),\frac{Q(a)Q(t)}{P(b)}\right), \tag*{(264)} E = ( P ( a ) , L ( a , b ) , P ( b ) , L ( b , t ) , P ( t ) , P ( b ) Q ( a ) Q ( t ) ) , (264) R = 1 − a b − a t − b t , (265) R=1-ab-at-bt, \tag*{(265)} R = 1 − ab − a t − b t , (265) S 1 = b + t − a + a b t , S 3 = a + b − t + a b t , S 5 = a + b + t − a b t , (266) S_1=b+t-a+abt,\qquad S_3=a+b-t+abt,\qquad S_5=a+b+t-abt, \tag*{(266)} S 1 = b + t − a + ab t , S 3 = a + b − t + ab t , S 5 = a + b + t − ab t , (266) Z = 2 + b 2 − b ( a + t ) + b 2 a t , κ = ( 1 − b 2 ) R 2 Z 2 − S 5 2 , (267) Z=2+b^2-b(a+t)+b^2at,\qquad\kappa=\frac{(1-b^2)R^2}{Z^2-S_5^2}, \tag*{(267)} Z = 2 + b 2 − b ( a + t ) + b 2 a t , κ = Z 2 − S 5 2 ( 1 − b 2 ) R 2 , (267) ( g 0 , … , g 5 ) = ( a − 1 , S 1 R , b − 1 , S 3 R , t − 1 , S 5 R ) , (268) (g_0,\ldots,g_5)=\left(a^{-1},\frac{S_1}{R},b^{-1},\frac{S_3}{R},t^{-1},\frac{S_5}{R}\right), \tag*{(268)} ( g 0 , … , g 5 ) = ( a − 1 , R S 1 , b − 1 , R S 3 , t − 1 , R S 5 ) , (268) J = 2 + κ ( 6 − ∑ i = 0 5 g i 2 log E i ) . (269) \mathcal{J}=2+\kappa\left(6-\sum_{i=0}^{5}g_i^2\log E_i\right). \tag*{(269)} J = 2 + κ ( 6 − i = 0 ∑ 5 g i 2 log E i ) . (269) At an actual point R , κ , g i > 0 R,\kappa,g_i>0 R , κ , g i > 0 , E i > 1 E_i>1 E i > 1 and κ g i g i − 1 < 1 \kappa g_i g_{i-1}<1 κ g i g i − 1 < 1 for all cyclic i i i .
Proof. Apply (A.46) to the suffix 020. The prefix 021 upper-right and lower-right group equations, exactly as in the 02|1|02 calculation, give E 1 2 = ( E 0 2 − 1 ) ( E 2 2 − 1 ) E_1^2=(E_0^2-1)(E_2^2-1) E 1 2 = ( E 0 2 − 1 ) ( E 2 2 − 1 ) . Write E 0 = P ( a ) E_0=P(a) E 0 = P ( a ) , E 2 = P ( b ) E_2=P(b) E 2 = P ( b ) . If u u u is the entrance coordinate, c = ( 1 − u ) / ( 1 + u ) c=(1-u)/(1+u) c = ( 1 − u ) / ( 1 + u ) , H = a + b H=a+b H = a + b and M = 1 − a b M=1-ab M = 1 − ab , put
t = M c − a H H c + a M , c = a ( H + M t ) M − H t . t=\frac{Mc-aH}{Hc+aM},\qquad c=\frac{a(H+Mt)}{M-Ht}. t = H c + a M M c − a H , c = M − H t a ( H + M t ) . The suffix shear determines its middle line uniquely: the coefficient of its reciprocal slope is positive because the two outer 0 arcs are positive. Its three exponents are the last three entries in E E E . The entrance state is
u 0 = ( 1 − a 2 ) ( 1 − b t ) − 2 a ( b + t ) ( 1 + a 2 ) ( 1 − b t ) , v 0 2 = 4 a 2 R 2 κ ( 1 + a 2 ) ( 1 − b t ) 2 . u_0=\frac{(1-a^2)(1-bt)-2a(b+t)}{(1+a^2)(1-bt)},\qquad v_0^2=\frac{4a^2R^2}{\kappa(1+a^2)(1-bt)^2}. u 0 = ( 1 + a 2 ) ( 1 − b t ) ( 1 − a 2 ) ( 1 − b t ) − 2 a ( b + t ) , v 0 2 = κ ( 1 + a 2 ) ( 1 − b t ) 2 4 a 2 R 2 . Substitution in the exact arc flows gives the stated projections and both endpoint conditions. The rational parts of the six arc integrals sum to 2 + 6 κ 2+6\kappa 2 + 6 κ , and their logarithmic weights are − κ g i 2 -\kappa g_i^2 − κ g i 2 . Every divided quantity is strictly nonzero at an actual point; positivity and star therefore persist under a sufficiently small change of all three parameters. The preceding local homotopy argument preserves winding. □
Lemma A.29 (Six-arc derivative inequality). On the actual domain of Lemma A.28, J ≤ 29 / 16 ⇒ ∂ J ∂ b > 0 \mathcal{J} \le29/16 \Rightarrow\frac{\partial\mathcal{J}}{\partial b} > 0 J ≤ 29/16 ⇒ ∂ b ∂ J > 0 , where a , t a,t a , t are held fixed.
Proof. The formula is symmetric in a , t a,t a , t , so take a ≥ t a \ge t a ≥ t and put
v = 1 − t 1 + t , r = 1 − a ( 1 + a ) v , b = v y . v=\frac{1-t}{1+t},\qquad r=\frac{1-a}{(1+a)v},\qquad b=vy. v = 1 + t 1 − t , r = ( 1 + a ) v 1 − a , b = v y .
Then 0 < v < 1 0<v<1 0 < v < 1 , 0 < r ≤ 1 0<r\le1 0 < r ≤ 1 . The positive exponents E 3 , E 5 > 1 E_3,E_5>1 E 3 , E 5 > 1 imply respectively
1 < y < Y : = 1 + r 2 1 + r 2 v 4 . (270) 1<y<Y:=\sqrt{\frac{1+r^2}{1+r^2v^4}}. \tag*{(270)} 1 < y < Y := 1 + r 2 v 4 1 + r 2 . (270)
We estimate on the larger closed y y y interval.
First, the region r ≤ 3 / 5 r\le3/5 r ≤ 3/5 has J > 29 / 16 \mathcal{J}>29/16 J > 29/16 by a polynomial witness. For q ≥ 1 q\ge1 q ≥ 1 ,
log q ≤ U ( q ) : = ( q − 1 ) ( q 2 + 10 q + 1 ) 6 q ( q + 1 ) , \log q\le\mathcal{U}(q):=\frac{(q-1)(q^2+10q+1)}{6q(q+1)}, log q ≤ U ( q ) := 6 q ( q + 1 ) ( q − 1 ) ( q 2 + 10 q + 1 ) ,
while log q ≤ q − 1 \log q\le q-1 log q ≤ q − 1 for all q > 0 q>0 q > 0 . Replace the first five logarithms in (A.53) by U ( E i ) \mathcal{U}(E_i) U ( E i ) and the last by E 5 − 1 E_5-1 E 5 − 1 . Parameterize a rational outer cover of (A.54) by
y = 1 + r 2 ( 1 − v 4 ) 2 ( 1 + r 2 v 4 ) ζ , 0 ≤ ζ ≤ 1. y=1+\frac{r^2(1-v^4)}{2(1+r^2v^4)}\zeta,\qquad0\le\zeta\le1. y = 1 + 2 ( 1 + r 2 v 4 ) r 2 ( 1 − v 4 ) ζ , 0 ≤ ζ ≤ 1.
Here v y < 1 vy<1 v y < 1 , y < 1 + r y<1+r y < 1 + r and all six weights are positive. The resulting rational lower bound for J − 29 / 16 \mathcal{J}-29/16 J − 29/16 , after positive factors are removed and r = 3 x / 5 r=3x/5 r = 3 x /5 , is a polynomial of coordinate degrees ( 66 , 32 , 12 ) (66,32,12) ( 66 , 32 , 12 ) on [ 0 , 1 ] 3 [0,1]^3 [ 0 , 1 ] 3 . Its 67 ⋅ 33 ⋅ 13 = 28 , 743 67\cdot33\cdot13=28,743 67 ⋅ 33 ⋅ 13 = 28 , 743 scaled integer Bernstein coefficients are all strictly positive. The finite witness consists of this array, the cleared-denominator identity, and the inverse Bernstein transform back to the rational bound. This defines a check using only integer arithmetic; in particular, the list is checked against the cost formula rather than accepted as an unexplained table of signs.
It remains to take 1 / 2 ≤ r ≤ 1 1/2\le r\le1 1/2 ≤ r ≤ 1 ; this slightly larger interval simplifies the remaining witnesses. Define
T 0 = 1 + r − y + r y v 2 , T_0=1+r-y+ryv^2, T 0 = 1 + r − y + r y v 2 ,
U 1 = y + r − 1 + r y v 2 , U 3 = y + 1 − r + r y v 2 , U_1=y+r-1+ryv^2,\qquad U_3=y+1-r+ryv^2, U 1 = y + r − 1 + r y v 2 , U 3 = y + 1 − r + r y v 2 ,
V 0 = 1 + v 2 [ y ( 1 + r ) − r ] , V_0=1+v^2[y(1+r)-r], V 0 = 1 + v 2 [ y ( 1 + r ) − r ] ,
A 0 = T 0 + v [ ( y − 1 ) ( y − r ) + r ] + r y 2 v 3 , A_0=T_0+v[(y-1)(y-r)+r]+ry^2v^3, A 0 = T 0 + v [( y − 1 ) ( y − r ) + r ] + r y 2 v 3 ,
B 0 = 2 + v ( 1 + r − y ) + v 2 [ y 2 + y ( 1 + r ) ] + r y v 3 + r y 2 v 4 , B_0=2+v(1+r-y)+v^2[y^2+y(1+r)]+ryv^3+ry^2v^4, B 0 = 2 + v ( 1 + r − y ) + v 2 [ y 2 + y ( 1 + r )] + r y v 3 + r y 2 v 4 ,
C 0 = 1 − v 2 y 2 A 0 B 0 . C_0=\frac{1-v^2y^2}{A_0B_0}. C 0 = A 0 B 0 1 − v 2 y 2 .
All their required denominators are positive. Set
f ( s ) = log P ( s ) s 2 , M = log ( P ( r v ) P ( v ) / P ( v y ) ) v 2 , K 0 = 6 − log E 0 a 2 − log E 4 t 2 − f ( v y ) , f(s)=\frac{\log P(s)}{s^2},\qquad M=\frac{\log(P(rv)P(v)/P(vy))}{v^2},\qquad K_0=6-\frac{\log E_0}{a^2}-\frac{\log E_4}{t^2}-f(vy), f ( s ) = s 2 log P ( s ) , M = v 2 log ( P ( r v ) P ( v ) / P ( v y )) , K 0 = 6 − a 2 log E 0 − t 2 log E 4 − f ( v y ) ,
where E 0 = ( 1 + r 2 v 2 ) / ( 2 r v ) E_0=(1+r^2v^2)/(2rv) E 0 = ( 1 + r 2 v 2 ) / ( 2 r v ) and E 4 = ( 1 + v 2 ) / ( 2 v ) E_4=(1+v^2)/(2v) E 4 = ( 1 + v 2 ) / ( 2 v ) . Then the exact stable cost is
J = 2 + v C 0 H , H = T 0 2 K 0 − U 1 2 log E 1 − U 3 2 log E 3 − V 0 2 M , (271) \mathcal{J}=2+vC_0H,\qquad H=T_0^2K_0-U_1^2\log E_1-U_3^2\log E_3-V_0^2M, \tag*{(271)} J = 2 + v C 0 H , H = T 0 2 K 0 − U 1 2 log E 1 − U 3 2 log E 3 − V 0 2 M , (271)
with E 1 = y ( 1 − r 2 v 2 ) / ( r ( 1 − v 2 y 2 ) ) E_1=y(1-r^2v^2)/(r(1-v^2y^2)) E 1 = y ( 1 − r 2 v 2 ) / ( r ( 1 − v 2 y 2 )) and E 3 = y ( 1 − v 2 ) / ( 1 − v 2 y 2 ) E_3=y(1-v^2)/(1-v^2y^2) E 3 = y ( 1 − v 2 ) / ( 1 − v 2 y 2 ) . Holding a , t a,t a , t fixed is the same as holding v , r v,r v , r fixed. Thus
J b = ( C 0 H ) y , M y = − 4 y 1 − v 4 y 4 . (272) \mathcal{J}_b=(C_0H)_y,\qquad M_y=-\frac{4y}{1-v^4y^4}. \tag*{(272)} J b = ( C 0 H ) y , M y = − 1 − v 4 y 4 4 y . (272)
The two noncompact ends are analytic. For v ≤ 1 / 100 v \le1/100 v ≤ 1/100 , the five bounds on the logarithmic contributions, grouping terms 0, 4 together, are respectively
11 200 , 93 2500 , 27 1000 , 103 10000 , 11 200 . \frac{11}{200},\quad\frac{93}{2500},\quad\frac{27}{1000},\quad\frac{103}{10000},\quad\frac{11}{200}. 200 11 , 2500 93 , 1000 27 , 10000 103 , 200 11 .
They follow from T 0 > 3 / 8 T_0 > 3/8 T 0 > 3/8 , κ < 51 v / 100 \kappa< 51v/100 κ < 51 v /100 , a , t > 49 / 50 a,t > 49/50 a , t > 49/50 , log P ( x ) ≤ 2 x 2 / ( 1 − x 2 ) \log P(x) \le2x^2/(1-x^2) log P ( x ) ≤ 2 x 2 / ( 1 − x 2 ) and the monotonicity of v log ( c / v ) v\log(c/v) v log ( c / v ) in this interval. Their sum is less than 369 / 2000 369/2000 369/2000 , so J > 3631 / 2000 > K \mathcal{J} > 3631/2000 > K J > 3631/2000 > K . For v ≥ 299 / 300 v \ge299/300 v ≥ 299/300 put ε = 1 − v \varepsilon= 1-v ε = 1 − v and δ = 1 − b \delta= 1-b δ = 1 − b . Then 0 < δ ≤ ε ≤ 1 / 300 0 < \delta\le\varepsilon\le1/300 0 < δ ≤ ε ≤ 1/300 , κ < 2 δ / 5 \kappa< 2\delta/5 κ < 2 δ /5 and g 1 , g 3 , g 5 < 21 / 10 g_1,g_3,g_5 < 21/10 g 1 , g 3 , g 5 < 21/10 . The same logarithm bound gives
2 − J < ε [ 42 25 + 1889 250 log 2 ε ] < 13643 75000 . 2-\mathcal{J}<\varepsilon\left[\frac{42}{25}+\frac{1889}{250}\log\frac{2}{\varepsilon}\right]<\frac{13643}{75000}. 2 − J < ε [ 25 42 + 250 1889 log ε 2 ] < 75000 13643 .
using log 600 < 7 \log600 < 7 log 600 < 7 . Hence J > 136357 / 75000 > K \mathcal{J} > 136357/75000 > K J > 136357/75000 > K . These estimates are uniform in the other parameters.
The remaining compact domain is parameterized by
( v , r , z ) ∈ [ 1 / 100 , 299 / 300 ] × [ 1 / 2 , 1 ] × [ 0 , 1 ] , y = 1 + z ( Y − 1 ) . (273) (v,r,z)\in[1/100,299/300]\times[1/2,1]\times[0,1],\qquad y=1+z(Y-1). \tag*{(273)} ( v , r , z ) ∈ [ 1/100 , 299/300 ] × [ 1/2 , 1 ] × [ 0 , 1 ] , y = 1 + z ( Y − 1 ) . (273)
Its finite witness is a partition into 41,716 rational closed boxes. Each box satisfies one of the following explicitly checked alternatives:
J > K , J b > 0 , J b + 8 ( J − K ) > 0 , ( a , b , t ) ∈ [ 9 / 25 , 47 / 100 ] 3 . \mathcal{J}>K,\qquad\mathcal{J}_b>0,\qquad\mathcal{J}_b+8(\mathcal{J}-K)>0,\qquad(a,b,t)\in[9/25,47/100]^3. J > K , J b > 0 , J b + 8 ( J − K ) > 0 , ( a , b , t ) ∈ [ 9/25 , 47/100 ] 3 .
On the last cube a separate rational interval witness gives J b > 1 / 1000 \mathcal{J}_b>1/1000 J b > 1/1000 . Each alternative therefore proves the desired implication. The partition identity is checked on the common 66 × 38 × 33 66\times38\times33 66 × 38 × 33 nonuniform coordinate endpoint grid: every one of its 65 ⋅ 37 ⋅ 32 = 76 , 960 65\cdot37\cdot32=76,960 65 ⋅ 37 ⋅ 32 = 76 , 960 elementary open cells is covered exactly once, and the closed boxes cover their boundary faces. Every box carries its rational coordinates, source index and acceptance inequality. The outward arithmetic rule, stable logarithm evaluation and whole-box derivative enclosures are specified in Appendix H.11.2. Every recorded lower bound is rational and strictly positive. Together with the analytic tails and the polynomial region these witnesses exhaust the necessary domain, proving the implication. □
At an actual point with cost at most J 8 < K \mathcal{J}_8<K J 8 < K , Lemma A.29 gives a strict decrease by decreasing b b b slightly in the exact family of Lemma A.28 . Combined with Proposition A.27 and the short-word table, this already proves Proposition A.15 for B ≤ 6 B\le6 B ≤ 6 . No compactness assertion for a class of arbitrary fixed-word competitors is needed for this conclusion.
A complete cost comparison for the reverse seven-arc word.
Proposition A.30. Every actual strict seven-arc chain of word 0210202 satisfies J > J 8 \mathcal{J}>\mathcal{J}_8 J > J 8 .
We supply the reduction, the noncompact estimates and the finite verification separately. The proof concerns every feasible chain of this word; complete PMP is not used. The prefix 021 has exponents P ( a ) , d , P ( b ) P(a),d,P(b) P ( a ) , d , P ( b ) , where
0 < a , b < 1 , B 0 = ( 1 − a 2 ) ( 1 − b 2 ) , d = 4 a b B 0 > 1 , 0<a,b<1,\qquad B_0=(1-a^2)(1-b^2),\qquad d=\frac{4ab}{B_0}>1, 0 < a , b < 1 , B 0 = ( 1 − a 2 ) ( 1 − b 2 ) , d = B 0 4 ab > 1 , D = 4 a b − B 0 > 0 , U 0 = 4 a b + B 0 . D=4ab-B_0>0,\qquad U_0=4ab+B_0. D = 4 ab − B 0 > 0 , U 0 = 4 ab + B 0 . To see necessity without a state-return assumption, introduce the full group transfers
N ( q ) = ( q − q q q − 1 − q ) , M ( q ) = ( 0 q − q − 1 q ) . N(q)=\begin{pmatrix}q&-q\\q&q^{-1}-q\end{pmatrix},\qquad M(q)=\begin{pmatrix}0&q\\-q^{-1}&q\end{pmatrix}. N ( q ) = ( q q − q q − 1 − q ) , M ( q ) = ( 0 − q − 1 q q ) . In the correct twisted frame the complete product is
N ( p ) N ( d ) N ( q ) N ( E 3 ) M ( E 4 ) N ( E 5 ) N ( E 6 ) = I . N(p)N(d)N(q)N(E_3)M(E_4)N(E_5)N(E_6)=I. N ( p ) N ( d ) N ( q ) N ( E 3 ) M ( E 4 ) N ( E 5 ) N ( E 6 ) = I .
The first-column entries of M ( e ) N ( f ) N ( g ) M(e)N(f)N(g) M ( e ) N ( f ) N ( g ) agree. The corresponding condition on the inverse prefix is
( p 2 − 1 ) ( q 2 − 1 ) − d 2 p d q E 3 = 0. \frac{(p^2-1)(q^2-1)-d^2}{pdqE_3}=0. p d q E 3 ( p 2 − 1 ) ( q 2 − 1 ) − d 2 = 0.
Every denominator is positive, so it gives the asserted prefix relation. The upper-right group equation determines the entrance height, including possible scalar state returns. In terms of c = ( 1 − u 0 ) / ( 1 + u 0 ) c=(1-u_0)/(1+u_0) c = ( 1 − u 0 ) / ( 1 + u 0 ) set
τ = ( 1 − a 2 ) c − 2 a 2 a ( 1 + a 2 ) . \tau=\frac{(1-a^2)c-2a^2}{a(1+a^2)}. τ = a ( 1 + a 2 ) ( 1 − a 2 ) c − 2 a 2 .
The four positive suffix arcs, using (A.46), imply
1 < τ < U 0 / D . (274) 1<\tau<\sqrt{U_0/D}. \tag*{(274)} 1 < τ < U 0 / D . (274)
For completeness, writing the suffix slopes as A B 1 , s B 1 , r B 1 , B 1 AB_1,sB_1,rB_1,B_1 A B 1 , s B 1 , r B 1 , B 1 reduces the shear equation to s = ( A − r ) / ( A + β r ) s=(A-r)/(A+\beta r) s = ( A − r ) / ( A + β r ) with − 1 < β < 1 -1<\beta<1 − 1 < β < 1 . The outer positive lengths give τ > 1 \tau>1 τ > 1 ; the quadratic numerator of β + 1 \beta+1 β + 1 has discriminant 4 a 2 ( 1 + a 2 ) 2 D U 0 4a^2(1+a^2)^2DU_0 4 a 2 ( 1 + a 2 ) 2 D U 0 and vertex c = 2 a 2 / ( 1 − a 2 ) c=2a^2/(1-a^2) c = 2 a 2 / ( 1 − a 2 ) . On τ > 1 \tau>1 τ > 1 its permitted side is exactly (A.58). These are necessary conditions; no geometric realization of every point in their enlarged domain is required.
Put
κ = B 0 4 + 4 ( a + b ) τ + D τ 2 , σ = 2 b ( 1 + a 2 ) + D τ ( 1 − a 2 ) ( 1 + b 2 ) , ε = 2 a ( 1 + b 2 ) + D τ ( 1 + a 2 ) ( 1 − b 2 ) , h = 2 ( 1 − a b ) 2 ( U 0 − D τ 2 ) ( 1 − a 4 ) ( 1 − b 4 ) , V 0 = 1 + a b + ( a + b ) τ , C = σ + ε 2 , f σ = ( 1 − b 2 ) [ 2 b ( 1 + a 2 ) + D τ ] 2 b ( 1 − a b ) V 0 , f ε = ( 1 − a 2 ) [ 2 a ( 1 + b 2 ) + D τ ] 2 a ( 1 − a b ) V 0 . \begin{aligned}
\kappa&=\frac{B_0}{4+4(a+b)\tau+D\tau^2},&
\sigma&=\frac{2b(1+a^2)+D\tau}{(1-a^2)(1+b^2)},&
\varepsilon&=\frac{2a(1+b^2)+D\tau}{(1+a^2)(1-b^2)},\\
h&=\frac{2(1-ab)^2(U_0-D\tau^2)}{(1-a^4)(1-b^4)},&
V_0&=1+ab+(a+b)\tau,&
C&=\frac{\sigma+\varepsilon}{2},\\
f_\sigma&=\frac{(1-b^2)[2b(1+a^2)+D\tau]}{2b(1-ab)V_0},&
f_\varepsilon&=\frac{(1-a^2)[2a(1+b^2)+D\tau]}{2a(1-ab)V_0}.
\end{aligned} κ h f σ = 4 + 4 ( a + b ) τ + D τ 2 B 0 , = ( 1 − a 4 ) ( 1 − b 4 ) 2 ( 1 − ab ) 2 ( U 0 − D τ 2 ) , = 2 b ( 1 − ab ) V 0 ( 1 − b 2 ) [ 2 b ( 1 + a 2 ) + D τ ] , σ V 0 f ε = ( 1 − a 2 ) ( 1 + b 2 ) 2 b ( 1 + a 2 ) + D τ , = 1 + ab + ( a + b ) τ , = 2 a ( 1 − ab ) V 0 ( 1 − a 2 ) [ 2 a ( 1 + b 2 ) + D τ ] . ε C = ( 1 + a 2 ) ( 1 − b 2 ) 2 a ( 1 + b 2 ) + D τ , = 2 σ + ε ,
Φ ( s , e ) = s 2 log 2 s s + e + e 2 log 2 e s + e . \Phi(s,e)=s^2\log\frac{2s}{s+e}+e^2\log\frac{2e}{s+e}. Φ ( s , e ) = s 2 log s + e 2 s + e 2 log s + e 2 e .
The original cost has the lower bound
L ( a , b , τ ) = 2 + κ [ 6 − log P ( a ) a 2 − log P ( b ) b 2 − τ 2 log d + σ 2 log f σ + ε 2 log f ε − Φ ( σ , ε ) − 2 Φ ( C , C + h ) ] ≤ J . (275) L(a,b,\tau)=2+\kappa\left[6-\frac{\log P(a)}{a^2}-\frac{\log P(b)}{b^2}-\tau^2\log d+\sigma^2\log f_\sigma+\varepsilon^2\log f_\varepsilon-\Phi(\sigma,\varepsilon)-2\Phi(C,C+\sqrt{h})\right]\leq\mathcal{J}. \tag*{(275)} L ( a , b , τ ) = 2 + κ [ 6 − a 2 log P ( a ) − b 2 log P ( b ) − τ 2 log d + σ 2 log f σ + ε 2 log f ε − Φ ( σ , ε ) − 2Φ ( C , C + h ) ] ≤ J . (275)
Here the rational parts of the seven original arc integrals sum to 2 + 6 κ 2+6\kappa 2 + 6 κ . We explain the sole inequality in this reduction. For s 0 , e 0 , p , q > 0 s_0,e_0,p,q>0 s 0 , e 0 , p , q > 0 , H = p q H=pq H = pq and s 0 + e 0 ≥ H s_0+e_0\geq\sqrt{H} s 0 + e 0 ≥ H , the rectangle integral
∫ s 0 s 0 + p ∫ e 0 e 0 + q s e ( s + e ) 2 d e d s \int_{s_0}^{s_0+p}\int_{e_0}^{e_0+q}\frac{se}{(s+e)^2}\,de\,ds ∫ s 0 s 0 + p ∫ e 0 e 0 + q ( s + e ) 2 se d e d s
is at most its value on the square with outer endpoints ( s 0 + e 0 ) / 2 (s_0+e_0)/2 ( s 0 + e 0 ) /2 and side H \sqrt{H} H . Indeed pair ( u , v ) (u,v) ( u , v ) with ( v , u ) (v,u) ( v , u ) on the unit square, put T = u + v T=u+v T = u + v , Δ = u − v \Delta=u-v Δ = u − v , S = p + q S=p+q S = p + q , δ = p − q \delta=p-q δ = p − q , and
D 0 = s 0 − e 0 + δ T / 2 , D 1 = S Δ / 2 , L 0 = s 0 + e 0 + S T / 2 , L 1 = δ Δ / 2. D_0=s_0-e_0+\delta T/2,\qquad D_1=S\Delta/2,\qquad L_0=s_0+e_0+ST/2,\qquad L_1=\delta\Delta/2. D 0 = s 0 − e 0 + δ T /2 , D 1 = S Δ/2 , L 0 = s 0 + e 0 + S T /2 , L 1 = δ Δ/2.
Completing the square gives
1 2 ( ( D 0 + D 1 ) 2 ( L 0 + L 1 ) 2 + ( D 0 − D 1 ) 2 ( L 0 − L 1 ) 2 ) ≥ D 0 2 L 0 2 + D 1 2 L 1 2 . \frac{1}{2}\left(\frac{(D_0+D_1)^2}{(L_0+L_1)^2}+\frac{(D_0-D_1)^2}{(L_0-L_1)^2}\right)\geq\frac{D_0^2}{L_0^2}+\frac{D_1^2}{L_1^2}. 2 1 ( ( L 0 + L 1 ) 2 ( D 0 + D 1 ) 2 + ( L 0 − L 1 ) 2 ( D 0 − D 1 ) 2 ) ≥ L 0 2 D 0 2 + L 1 2 D 1 2 .
With η = S / ( 2 H ) ≥ 1 \eta=S/(2\sqrt{H})\geq1 η = S / ( 2 H ) ≥ 1 and c = ( s 0 + e 0 ) / H ≥ 1 c=(s_0+e_0)/\sqrt{H}\geq1 c = ( s 0 + e 0 ) / H ≥ 1 , the latter is at least Δ 2 / ( c + T ) 2 \Delta^2/(c+T)^2 Δ 2 / ( c + T ) 2 , because the difference of cross-multiplied denominators is ( η 2 − 1 ) ( c 2 − Δ 2 ) + 2 c T η ( η − 1 ) ≥ 0 (\eta^2-1)(c^2-\Delta^2)+2cT\eta(\eta-1)\geq0 ( η 2 − 1 ) ( c 2 − Δ 2 ) + 2 c T η ( η − 1 ) ≥ 0 . Use s e / ( s + e ) 2 = 1 / 4 − ( s − e ) 2 / ( 4 ( s + e ) 2 ) se/(s+e)^2=1/4-(s-e)^2/(4(s+e)^2) se / ( s + e ) 2 = 1/4 − ( s − e ) 2 / ( 4 ( s + e ) 2 ) and integrate. Since Φ s e = − 1 − 2 s e / ( s + e ) 2 \Phi_{se}=-1-2se/(s+e)^2 Φ se = − 1 − 2 se / ( s + e ) 2 , this bounds precisely the suffix logarithmic contribution. In the present suffix H / ( s 0 + e 0 ) 2 = A ( β + 1 ) / ( A + 1 ) 2 < 1 / 2 H/(s_0+e_0)^2=A(\beta+1)/(A+1)^2<1/2 H / ( s 0 + e 0 ) 2 = A ( β + 1 ) / ( A + 1 ) 2 < 1/2 , so the required comparison hypothesis holds. This proves (A.59) with its stated direction.
Lemma A.31. The function in (A.59) satisfies L > J 8 L>\mathcal{J}_8 L > J 8 throughout (A.58) and D > 0 D>0 D > 0 .
Proof. The symmetry in a , b a,b a , b will always be retained. The entropy function has the convergent expansion
Φ ( s , e ) = ( s − e ) 2 4 ( 3 − ∑ k ≥ 1 w k k ( k + 1 ) ( 2 k + 1 ) ) , w = ( s − e ) 2 ( s + e ) 2 < 1. (276) \Phi(s,e)=\frac{(s-e)^2}{4}\left(3-\sum_{k\geq1}\frac{w^k}{k(k+1)(2k+1)}\right),\qquad w=\frac{(s-e)^2}{(s+e)^2}<1. \tag*{(276)} Φ ( s , e ) = 4 ( s − e ) 2 ( 3 − k ≥ 1 ∑ k ( k + 1 ) ( 2 k + 1 ) w k ) , w = ( s + e ) 2 ( s − e ) 2 < 1. (276)
This follows by twice integrating [ ( 1 + u ) 2 log ( 1 + u ) + ( 1 − u ) 2 log ( 1 − u ) ] ′ ′ = 6 + 2 log ( 1 − u 2 ) [(1+u)^2\log(1+u)+(1-u)^2\log(1-u)]''=6+2\log(1-u^2) [( 1 + u ) 2 log ( 1 + u ) + ( 1 − u ) 2 log ( 1 − u ) ] ′′ = 6 + 2 log ( 1 − u 2 ) . In particular 0 ≤ Φ ≤ 3 ( s − e ) 2 / 4 0\leq\Phi\leq3(s-e)^2/4 0 ≤ Φ ≤ 3 ( s − e ) 2 /4 . The stable combined logarithms are
σ 2 log f σ + ε 2 log f ε − Φ ( σ , ε ) = σ 2 log ( 1 − u s ) + ε 2 log ( 1 − u e ) , (277) \sigma^2\log f_\sigma+\varepsilon^2\log f_\varepsilon-\Phi(\sigma,\varepsilon)=\sigma^2\log(1-u_s)+\varepsilon^2\log(1-u_e), \tag*{(277)} σ 2 log f σ + ε 2 log f ε − Φ ( σ , ε ) = σ 2 log ( 1 − u s ) + ε 2 log ( 1 − u e ) , (277)
where
u s = ( 1 − a 2 ) ( 1 + b 2 ) [ ( 1 − a b ) τ − a + b ] 2 b ( 1 + a 2 ) V 0 , u e = ( 1 + a 2 ) ( 1 − b 2 ) [ ( 1 − a b ) τ + a − b ] 2 a ( 1 + b 2 ) V 0 . u_s=\frac{(1-a^2)(1+b^2)[(1-ab)\tau-a+b]}{2b(1+a^2)V_0},\qquad u_e=\frac{(1+a^2)(1-b^2)[(1-ab)\tau+a-b]}{2a(1+b^2)V_0}. u s = 2 b ( 1 + a 2 ) V 0 ( 1 − a 2 ) ( 1 + b 2 ) [( 1 − ab ) τ − a + b ] , u e = 2 a ( 1 + b 2 ) V 0 ( 1 + a 2 ) ( 1 − b 2 ) [( 1 − ab ) τ + a − b ] .
Both belong to ( 0 , 1 ) (0,1) ( 0 , 1 ) on the necessary domain. This exact combination is essential in the noncompact estimates: estimating its original summands separately would lose their cancellation.
We first list and prove the uniform tail estimates. For τ ≥ 2 \tau\geq2 τ ≥ 2 use the coordinates
ξ 2 = U 0 − D τ 2 2 B 0 , d = 1 + 2 ( 1 − ξ 2 ) τ 2 − 1 , b = d ( 1 − a 2 ) 4 a 2 + d 2 ( 1 − a 2 ) 2 + 2 a , 0 ≤ ξ ≤ 1. (278) \xi^2=\frac{U_0-D\tau^2}{2B_0},\qquad d=1+\frac{2(1-\xi^2)}{\tau^2-1},\qquad b=\frac{d(1-a^2)}{\sqrt{4a^2+d^2(1-a^2)^2}+2a},\qquad0\leq\xi\leq1. \tag*{(278)} ξ 2 = 2 B 0 U 0 − D τ 2 , d = 1 + τ 2 − 1 2 ( 1 − ξ 2 ) , b = 4 a 2 + d 2 ( 1 − a 2 ) 2 + 2 a d ( 1 − a 2 ) , 0 ≤ ξ ≤ 1. (278)
They include both finite boundary faces. If 1 / 3 ≤ a ≤ 1 / 2 1/3\leq a\leq1/2 1/3 ≤ a ≤ 1/2 and τ ≥ 128 \tau\geq128 τ ≥ 128 , then 1 / 3 ≤ b < 3 / 5 1/3\leq b<3/5 1/3 ≤ b < 3/5 , κ ≤ 8 / ( 27 τ ) \kappa\leq8/(27\tau) κ ≤ 8/ ( 27 τ ) , σ , ε < 2 \sigma,\varepsilon<2 σ , ε < 2 , f σ , f ε > 1 / ( 3 ( 1 + τ ) ) f_\sigma,f_\varepsilon>1/(3(1+\tau)) f σ , f ε > 1/ ( 3 ( 1 + τ )) , f ( a ) + f ( b ) < 9 / 2 f(a)+f(b)<9/2 f ( a ) + f ( b ) < 9/2 , and τ 2 log d < 3 \tau^2\log d<3 τ 2 log d < 3 , where f ( s ) = log P ( s ) / s 2 f(s)=\log P(s)/s^2 f ( s ) = log P ( s ) / s 2 . The two entropy bounds total at most 6, cancelling the positive 6. Consequently
L ≥ 2 − 8 27 τ ( 15 2 + 8 log ( 3 ( 1 + τ ) ) ) > 539 288 > K . L\geq2-\frac{8}{27\tau}\left(\frac{15}{2}+8\log(3(1+\tau))\right)>\frac{539}{288}>K. L ≥ 2 − 27 τ 8 ( 2 15 + 8 log ( 3 ( 1 + τ )) ) > 288 539 > K .
The logarithmic error decreases with τ \tau τ ; log 387 < 6 \log387<6 log 387 < 6 completes the rational comparison. For 0 < a ≤ 1 / 3 0<a\leq1/3 0 < a ≤ 1/3 and τ ≥ 2 \tau\geq2 τ ≥ 2 , the same coordinates give κ ≤ a / ( 1 + τ ) \kappa\leq a/(1+\tau) κ ≤ a / ( 1 + τ ) and 1 ≤ σ , ε ≤ 1 + 2 / ( τ − 1 ) 1\leq\sigma,\varepsilon\leq1+2/(\tau-1) 1 ≤ σ , ε ≤ 1 + 2/ ( τ − 1 ) . If a ≤ 1 / 64 a\leq1/64 a ≤ 1/64 , the sharper σ < 17 / 16 \sigma<17/16 σ < 17/16 yields
L ≥ 2 − a 1 + τ [ 5 + 8431 3840 log 1 a + 2593 256 log ( 2 ( 1 + τ ) ) ] > 374983 204800 > K . L\geq2-\frac{a}{1+\tau}\left[5+\frac{8431}{3840}\log\frac{1}{a}+\frac{2593}{256}\log(2(1+\tau))\right]>\frac{374983}{204800}>K. L ≥ 2 − 1 + τ a [ 5 + 3840 8431 log a 1 + 256 2593 log ( 2 ( 1 + τ )) ] > 204800 374983 > K .
The error is increasing in a a a and decreasing in τ \tau τ , so use a = 1 / 64 a=1/64 a = 1/64 , τ = 2 \tau=2 τ = 2 , log 64 < 21 / 5 \log64<21/5 log 64 < 21/5 , log 6 < 9 / 5 \log6<9/5 log 6 < 9/5 . For all a ≤ 1 / 3 a\leq1/3 a ≤ 1/3 and τ ≥ 128 \tau\geq128 τ ≥ 128 , instead use
L ≥ 2 − a 1 + τ [ 81 40 + 3 + 4 log 5 3 a + 81 64 log 6 ( 1 + τ ) 2 a ] > 238247 123840 > K . L\geq2-\frac{a}{1+\tau}\left[\frac{81}{40}+3+4\log\frac{5}{3a}+\frac{81}{64}\log\frac{6(1+\tau)^2}{a}\right]>\frac{238247}{123840}>K. L ≥ 2 − 1 + τ a [ 40 81 + 3 + 4 log 3 a 5 + 64 81 log a 6 ( 1 + τ ) 2 ] > 123840 238247 > K .
The same monotonicities reduce this to a = 1 / 3 a=1/3 a = 1/3 , τ = 128 \tau=128 τ = 128 , log 5 < 2 \log5<2 log 5 < 2 and log 299538 < 13 \log299538<13 log 299538 < 13 .
For 1 / 3 ≤ a ≤ 1 / 2 1/3\leq a\leq1/2 1/3 ≤ a ≤ 1/2 and b ≥ 99 / 100 b\geq99/100 b ≥ 99/100 , put s = 1 − b 2 < 1 / 50 s=1-b^2<1/50 s = 1 − b 2 < 1/50 . Equation (A.61) gives σ < 4 \sigma<4 σ < 4 , ε < 4 / s \varepsilon<4/s ε < 4/ s , u s < 1 / 2 u_s<1/2 u s < 1/2 , u e < s 2 u_e<s^2 u e < s 2 and κ ≤ s / 8 \kappa\leq s/8 κ ≤ s /8 . Hence
L > 2 − s 8 [ 39 + 11 10 log 2 s + 2 log 8 3 s ] > 1491 800 > K . L>2-\frac{s}{8}\left[39+\frac{11}{10}\log\frac{2}{s}+2\log\frac{8}{3s}\right]>\frac{1491}{800}>K. L > 2 − 8 s [ 39 + 10 11 log s 2 + 2 log 3 s 8 ] > 800 1491 > K .
Here s log ( c / s ) s\log(c/s) s log ( c / s ) increases on the stated range, and both required logarithms at s = 1 / 50 s=1/50 s = 1/50 are less than 5. For a ≥ 1 / 2 a\geq1/2 a ≥ 1/2 , b ≥ 39 / 40 b\geq39/40 b ≥ 39/40 , let A = 1 − a 2 ≤ 3 / 4 A=1-a^2\leq3/4 A = 1 − a 2 ≤ 3/4 and s = 1 − b 2 < 1 / 20 s=1-b^2<1/20 s = 1 − b 2 < 1/20 . Cancelling the factors in σ 2 u s \sigma^{2}u_s σ 2 u s and ε 2 u e \varepsilon^{2}u_e ε 2 u e before estimating gives respectively bounds 7 and 12, while u s < 1 / 4 u_s < 1/4 u s < 1/4 , u e < s 2 u_e < s^2 u e < s 2 and κ < A s / 11 \kappa< As/11 κ < A s /11 . Thus
L > 2 − A s 11 [ 26 + 4 log 2 A + 11 10 log 2 s + 2 log 4 A s ] > 4067 2200 > K . L > 2-\frac{As}{11}\left[26+4\log\frac{2}{A}+\frac{11}{10}\log\frac{2}{s}+2\log\frac{4}{As}\right]>\frac{4067}{2200}>K. L > 2 − 11 A s [ 26 + 4 log A 2 + 10 11 log s 2 + 2 log A s 4 ] > 2200 4067 > K .
The error increases separately in A , s A,s A , s ; use ( A , s ) = ( 3 / 4 , 1 / 20 ) (A,s)=(3/4,1/20) ( A , s ) = ( 3/4 , 1/20 ) and log ( 8 / 3 ) < 1 \log(8/3)<1 log ( 8/3 ) < 1 , log 40 < 4 \log40<4 log 40 < 4 , log ( 320 / 3 ) < 5 \log(320/3)<5 log ( 320/3 ) < 5 . All logarithmic constants just used follow, for example, by comparing the stated argument with a finite positive Taylor sum for the exponential.
It remains to control the low-angle corner with a ≤ 1 / 3 a\leq1/3 a ≤ 1/3 , τ ≤ 2 \tau\leq2 τ ≤ 2 . Put
r = 1 − b a ( 1 + b ) , b = 1 − a r 1 + a r , τ = 1 + r z . (279) r=\frac{1-b}{a(1+b)},\qquad b=\frac{1-ar}{1+ar},\qquad\tau=1+rz. \tag*{(279)} r = a ( 1 + b ) 1 − b , b = 1 + a r 1 − a r , τ = 1 + r z . (279)
Then 0 < r < 1 0<r<1 0 < r < 1 , and D τ 2 ≤ U 0 D\tau^2\leq U_0 D τ 2 ≤ U 0 gives
τ 2 − 1 τ 2 + 1 ≤ r ( 1 − a 2 ) 1 − a 2 r 2 ≤ r , 0 ≤ z ≤ τ 2 + 1 τ + 1 ≤ 5 3 . \frac{\tau^2-1}{\tau^2+1}\leq\frac{r(1-a^2)}{1-a^2r^2}\leq r,\qquad0\leq z\leq\frac{\tau^2+1}{\tau+1}\leq\frac{5}{3}. τ 2 + 1 τ 2 − 1 ≤ 1 − a 2 r 2 r ( 1 − a 2 ) ≤ r , 0 ≤ z ≤ τ + 1 τ 2 + 1 ≤ 3 5 .
This covers all coupled approaches to a = 0 a=0 a = 0 , b = 1 b=1 b = 1 and τ = 1 \tau=1 τ = 1 . After cancellation in (A.61), direct positive-factor estimates give, uniformly in the other two variables,
a ≤ 1 / 256 : L > 2 − a r 2 [ 60 + 11 10 log 1 a r + 4 log 9 8 r + 121 100 log 3 a ] > 94493 51200 > K , a\leq1/256:\qquad L>2-\frac{ar}{2}\left[60+\frac{11}{10}\log\frac{1}{ar}+4\log\frac{9}{8r}+\frac{121}{100}\log\frac{3}{a}\right]>\frac{94493}{51200}>K, a ≤ 1/256 : L > 2 − 2 a r [ 60 + 10 11 log a r 1 + 4 log 8 r 9 + 100 121 log a 3 ] > 51200 94493 > K ,
r ≤ 1 / 100 : L > 2 − a r 2 [ 18 + 11 10 log 1 a r + 2 log 9 8 r + 441 100 log 3 a ] > 115217 60000 > K . r\leq1/100:\qquad L>2-\frac{ar}{2}\left[18+\frac{11}{10}\log\frac{1}{ar}+2\log\frac{9}{8r}+\frac{441}{100}\log\frac{3}{a}\right]>\frac{115217}{60000}>K. r ≤ 1/100 : L > 2 − 2 a r [ 18 + 10 11 log a r 1 + 2 log 8 r 9 + 100 441 log a 3 ] > 60000 115217 > K .
For the first inequality one uses κ ≤ a r / 2 \kappa\leq ar/2 κ ≤ a r /2 , σ < 11 / 10 \sigma<11/10 σ < 11/10 , r ε ≤ 3 r\varepsilon\leq3 r ε ≤ 3 , u e / r 2 < 17 / 20 u_e/r^2<17/20 u e / r 2 < 17/20 , and ( 1 − u s ) / a > 1 / 3 (1-u_s)/a>1/3 ( 1 − u s ) / a > 1/3 . For the second, τ ≤ 61 / 60 \tau\leq61/60 τ ≤ 61/60 , σ , r ε < 21 / 10 \sigma,r\varepsilon<21/10 σ , r ε < 21/10 and u e / r 2 < 8 / 5 u_e/r^2<8/5 u e / r 2 < 8/5 suffice. The bracket times a r ar a r increases in both a , r a,r a , r . Evaluation at ( 1 / 256 , 1 ) (1/256,1) ( 1/256 , 1 ) and ( 1 / 3 , 1 / 100 ) (1/3,1/100) ( 1/3 , 1/100 ) , respectively, gives the displayed rational bounds.
We now specify the finite part. Put b 0 ( a ) = ( 1 − a ) / ( 1 + a ) b_0(a)=(1-a)/(1+a) b 0 ( a ) = ( 1 − a ) / ( 1 + a ) , δ = b − b 0 ( a ) \delta=b-b_0(a) δ = b − b 0 ( a ) and ρ = 2 − 1 \rho=\sqrt{2}-1 ρ = 2 − 1 . The three derivative witnesses on
1 / 3 ≤ a ≤ 1 / 2 , 0 ≤ δ ≤ 1 / 100 , 1 ≤ τ ≤ 2 1/3\leq a\leq1/2,\qquad0\leq\delta\leq1/100,\qquad1\leq\tau\leq2 1/3 ≤ a ≤ 1/2 , 0 ≤ δ ≤ 1/100 , 1 ≤ τ ≤ 2
prove L τ > 1 / 1000 L_\tau>1/1000 L τ > 1/1000 , L δ ∣ τ = 1 > 1 / 1000 L_\delta|_{\tau=1}>1/1000 L δ ∣ τ = 1 > 1/1000 and F ′ ′ ( a ) > 1 / 10 F''(a)>1/10 F ′′ ( a ) > 1/10 , where F ( a ) = L ( a , b 0 ( a ) , 1 ) F(a)=L(a,b_0(a),1) F ( a ) = L ( a , b 0 ( a ) , 1 ) . These are respectively three-, two- and one-dimensional inequalities. The identities F ( ρ ) = J 8 F(\rho)=\mathcal{J}_8 F ( ρ ) = J 8 and F ′ ( ρ ) = 0 F'(\rho)=0 F ′ ( ρ ) = 0 follow by substitution and the symmetry F ( a ) = F ( b 0 ( a ) ) F(a)=F(b_0(a)) F ( a ) = F ( b 0 ( a )) . Integrating in the order a a a , then δ \delta δ at τ = 1 \tau=1 τ = 1 , then τ \tau τ , gives
L ≥ J 8 + ( a − ρ ) 2 20 + δ 1000 + τ − 1 1000 . (280) L\geq\mathcal{J}_8+\frac{(a-\rho)^2}{20}+\frac{\delta}{1000}+\frac{\tau-1}{1000}. \tag*{(280)} L ≥ J 8 + 20 ( a − ρ ) 2 + 1000 δ + 1000 τ − 1 . (280)
This is a comparison of costs, not an assertion that L L L has the same feasible derivative as the original cost.
The other compact witnesses are listed below. An entry “cost” means a strict lower bound for L L L (or a stated smaller expression); the three collar rows retain their derivative meanings.
Table 1 (PDF p. 157).
In rows with ( a , b , ξ ) (a,b,\xi) ( a , b , ξ ) use τ = 1 + 2 B 0 ( 1 − ξ 2 ) / D \tau=\sqrt{1+2B_0(1-\xi^2)/D} τ = 1 + 2 B 0 ( 1 − ξ 2 ) / D ; the variable h \sqrt{h} h is then 2 ( 1 − a b ) ξ / ( 1 + a 2 ) ( 1 + b 2 ) 2(1-ab)\xi/\sqrt{(1+a^2)(1+b^2)} 2 ( 1 − ab ) ξ / ( 1 + a 2 ) ( 1 + b 2 ) , with no endpoint singularity. The variable interval for b b b in the seventh row is affinely parameterized by [ 0 , 1 ] [0,1] [ 0 , 1 ] . For the eighth row replace the final entropy term by 3 h / 2 3h/2 3 h /2 . For the last row use its sharper upper bound
2 Φ ( C , C + h ) ≤ 3 h 2 − h 2 48 ( C + 1 ) 2 , 2\Phi(C,C+\sqrt{h})\le\frac{3h}{2}-\frac{h^2}{48(C+1)^2}, 2Φ ( C , C + h ) ≤ 2 3 h − 48 ( C + 1 ) 2 h 2 ,
which follows from the first subtracted term of (A.60) and h < 2 \sqrt{h}<2 h < 2 . Boxes with a strictly negative upper bound for the numerator of h h h have no necessary-domain points. Boxes wholly in τ ≥ 2 \tau\ge2 τ ≥ 2 may invoke the already verified fifth and sixth rows and their tails. No such reference is used in the fifth or sixth row, so the checking order is acyclic.
The finite checking rule and the entropy remainder bounds are specified in Appendix H.11.3. The witnesses are exact binary covers and every leaf proves its stated strict inequality; the completed cover has no unresolved leaf.
Finally the domains exhaust the necessary set. If τ ≥ 2 \tau\ge2 τ ≥ 2 and neither variable lies in [ 1 / 3 , 1 / 2 ] [1/3,1/2] [ 1/3 , 1/2 ] , they cannot both be at least 1 / 2 1/2 1/2 , since then U 0 / D ≤ 25 / 7 < 4 U_0/D\le25/7<4 U 0 / D ≤ 25/7 < 4 . The remaining asymmetric case is covered by the sixth row and its tails. For τ ≤ 2 \tau\le2 τ ≤ 2 , a variable in [ 1 / 3 , 1 / 2 ] [1/3,1/2] [ 1/3 , 1/2 ] is handled by the collar, the seventh row or its b b b tail. If both variables exceed 1 / 2 1/2 1/2 , use the first or eighth row and the high tail. Otherwise exchange them and apply the last row with its two uniform tails. All interfaces overlap as closed sets. Strictness in the collar follows from δ > 0 \delta>0 δ > 0 and τ > 1 \tau>1 τ > 1 for an actual chain. This proves the lemma. □
Completion of the proof. The full-closure reduction gives (A.58) and J ≥ L \mathcal{J}\ge L J ≥ L . Lemma A.31 supplies the strict comparison. □
Pure-forward seven-arc minimizers and the complete PMP filter
The pure-forward case is different from Proposition A.30: we use inequalities supplied by maximization over the full control triangle. Their role is to separate the candidate set from the zero-duration octagonal boundary and to reduce a genuinely four-dimensional domain to a finite verified inequality.
Proposition A.32. A positive pure-forward seven-arc trajectory satisfying the original complete free-time maximizing PMP conditions cannot be an original global minimizer.
We first record the exact chart and state formula used in the finite proof. The matrix M ( q ) M(q) M ( q ) defined above is the group transfer in the pure-forward gauge. This is a full matrix identity:
M ( q 0 ) ⋯ M ( q 6 ) = I , q i > 1. M(q_0)\cdots M(q_6)=I,\qquad q_i>1. M ( q 0 ) ⋯ M ( q 6 ) = I , q i > 1.
The sign on the right is fixed by the original twist, rather than by a projective trace equation. Put t i = q i 2 − 1 t_i=q_i^2-1 t i = q i 2 − 1 . Multiplication and elimination of the last three positive exponents give
F = 1 + t 0 + t 2 − t 0 t 2 , E = t 0 ( 1 + t 1 ) + t 3 F , F=1+t_0+t_2-t_0t_2,\qquad E=t_0(1+t_1)+t_3F, F = 1 + t 0 + t 2 − t 0 t 2 , E = t 0 ( 1 + t 1 ) + t 3 F ,
t 4 = ( 1 + t 2 ) ( 1 − t 0 t 1 ) E , t_4=\frac{(1+t_2)(1-t_0t_1)}{E}, t 4 = E ( 1 + t 2 ) ( 1 − t 0 t 1 ) ,
t 5 = ( 1 − t 1 t 2 ) E ( 1 + t 1 ) ( 1 + t 2 ) , t 6 = ( 1 + t 1 ) ( 1 − t 2 t 3 ) E . (281) t_5=\frac{(1-t_1t_2)E}{(1+t_1)(1+t_2)},\qquad
t_6=\frac{(1+t_1)(1-t_2t_3)}{E}. \tag*{(281)} t 5 = ( 1 + t 1 ) ( 1 + t 2 ) ( 1 − t 1 t 2 ) E , t 6 = E ( 1 + t 1 ) ( 1 − t 2 t 3 ) . (281)
The complete positive chart has t 0 , … , t 3 > 0 t_0,\ldots,t_3>0 t 0 , … , t 3 > 0 and t 0 t 1 , t 1 t 2 , t 2 t 3 < 1 t_0t_1,t_1t_2,t_2t_3<1 t 0 t 1 , t 1 t 2 , t 2 t 3 < 1 . The denominators are positive, and the same relations hold at every cyclic cut. In particular,
t i + 3 t i + 4 = ( 1 − t i − 1 t i ) ( 1 − t i t i + 1 ) 1 + t i . (282) t_{i+3}t_{i+4}=\frac{(1-t_{i-1}t_i)(1-t_it_{i+1})}{1+t_i}. \tag*{(282)} t i + 3 t i + 4 = 1 + t i ( 1 − t i − 1 t i ) ( 1 − t i t i + 1 ) . (282)
These identities are verified by multiplying the seven matrices; conversely substitution of (A.65) gives I I I in all four entries. In logarithmic t t t coordinates the three defining inequalities are affine half-spaces, so the positive chart is connected.
For the state return use the unnormalized matrices
T ( q ) = 1 2 ( q q − 4 q q ) , P i = T ( q i − 1 ) ⋯ T ( q 0 ) T ( q 6 ) ⋯ T ( q i ) , (283) T(q)=\frac{1}{2}\begin{pmatrix}q & q-4 \\ q & q\end{pmatrix}, \qquad P_i=T(q_{i-1})\cdots T(q_0)T(q_6)\cdots T(q_i), \tag*{(283)} T ( q ) = 2 1 ( q q q − 4 q ) , P i = T ( q i − 1 ) ⋯ T ( q 0 ) T ( q 6 ) ⋯ T ( q i ) , (283)
with cyclic interpretations at the endpoints. Define
τ 0 = tr P 0 , Δ = 4 ∏ i q i − τ 0 2 , n i = 2 ( P i ) 21 + ( P i ) 11 − ( P i ) 22 . \tau_0=\operatorname{tr}P_0,\qquad\Delta=4\prod_i q_i-\tau_0^2,\qquad n_i=2(P_i)_{21}+(P_i)_{11}-(P_i)_{22}. τ 0 = tr P 0 , Δ = 4 i ∏ q i − τ 0 2 , n i = 2 ( P i ) 21 + ( P i ) 11 − ( P i ) 22 .
The original positive state has active projection a i = n i / Δ a_i=n_i/\sqrt{\Delta} a i = n i / Δ . The square root is positive, and its sign is fixed by strict star; it is not a choice made independently on each arc. The exact state recursion and cost are
q i = ( a i + a i − 1 ) ( a i + a i + 1 ) 1 + a i 2 , J = 3 ∑ i a i a i + 1 − 7 − ∑ i a i 2 log q i . (284) q_i=\frac{(a_i+a_{i-1})(a_i+a_{i+1})}{1+a_i^2},\qquad\mathcal{J}=3\sum_i a_i a_{i+1}-7-\sum_i a_i^2\log q_i. \tag*{(284)} q i = 1 + a i 2 ( a i + a i − 1 ) ( a i + a i + 1 ) , J = 3 i ∑ a i a i + 1 − 7 − i ∑ a i 2 log q i . (284)
Equivalently, n i = 2 q i ∂ q i τ 0 − τ 0 n_i=2q_i\partial_{q_i}\tau_0-\tau_0 n i = 2 q i ∂ q i τ 0 − τ 0 before substitution of the closure chart. This identity is obtained by differentiating the product in (A.67); it verifies the normalization of the cost numerator used below.
Lemma A.33. On every finite closed chart obtained from (A.65) with t i ≥ 0 t_i\ge0 t i ≥ 0 , adjacent products at most one, and E > 0 E>0 E > 0 , the state return has Δ > 0 \Delta>0 Δ > 0 .
Proof. Put r i = 1 / q i > 0 r_i=1/q_i>0 r i = 1/ q i > 0 . The full group recurrence is the recurrence for a seven-cycle Jacobi matrix with edge weights r i r_i r i , at eigenvalue one. The identity monodromy gives a two-dimensional eigenspace. Thus its six-point principal and adjacent cofactors vanish. The half-state matrix has edge weights r i \sqrt{r_i} r i ; its monodromy has the same trace as P 0 / ∏ i q i P_0/\sqrt{\prod_i q_i} P 0 / ∏ i q i . The adjacent constraints say r i 2 + r i + 1 2 ≥ 1 r_i^2+r_{i+1}^2\ge1 r i 2 + r i + 1 2 ≥ 1 .
For a six-point path, write its five squared full edges as a , b , c , d , e ∈ ( 0 , 1 ] a,b,c,d,e\in(0,1] a , b , c , d , e ∈ ( 0 , 1 ] . Its zero determinant is
D 6 = 1 − a − b − c − d − e + a c + a d + a e + b d + b e + c e − a c e = 0. D_6=1-a-b-c-d-e+ac+ad+ae+bd+be+ce-ace=0. D 6 = 1 − a − b − c − d − e + a c + a d + a e + b d + b e + ce − a ce = 0.
If X = a + b − 1 X=a+b-1 X = a + b − 1 , then X d = ( 1 − e ) [ X + c ( 1 − a ) ] Xd=(1-e)[X+c(1-a)] X d = ( 1 − e ) [ X + c ( 1 − a )] . Write A = a A=\sqrt{a} A = a , B = b B=\sqrt{b} B = b , C = c C=\sqrt{c} C = c , D = d D=\sqrt{d} D = d , E 0 = e E_0=\sqrt{e} E 0 = e , and X h = A + B − 1 > 0 X_h=A+B-1>0 X h = A + B − 1 > 0 . If X = 0 X=0 X = 0 then e = 1 e=1 e = 1 , and the half-path determinant is X h D > 0 X_hD>0 X h D > 0 . If X > 0 X>0 X > 0 and a = 1 a=1 a = 1 , it is X h ( D + E 0 − 1 ) > 0 X_h(D+E_0-1)>0 X h ( D + E 0 − 1 ) > 0 , since d + e = 1 d+e=1 d + e = 1 . Otherwise set α = ( 1 − a ) / X \alpha=(1-a)/X α = ( 1 − a ) / X and α h = ( 1 − A ) / X h \alpha_h=(1-A)/X_h α h = ( 1 − A ) / X h . The identity X h 2 − X = 2 ( 1 − A ) ( 1 − B ) ≥ 0 X_h^2-X=2(1-A)(1-B)\ge0 X h 2 − X = 2 ( 1 − A ) ( 1 − B ) ≥ 0 gives α h < α \alpha_h<\sqrt{\alpha} α h < α . With s = c α s=c\alpha s = c α , d = ( 1 − e ) ( 1 + s ) ≤ 1 d=(1-e)(1+s)\le1 d = ( 1 − e ) ( 1 + s ) ≤ 1 , so
D 1 − E 0 ≥ 1 + s + s ( 1 + s ) > 1 + C α h . \frac{D}{1-E_0}\ge1+s+\sqrt{s(1+s)}>1+C\alpha_h. 1 − E 0 D ≥ 1 + s + s ( 1 + s ) > 1 + C α h .
The half-path determinant factors as [ X h + C ( 1 − A ) ] ( E 0 − 1 + D / ( 1 + C α h ) ) [X_h+C(1-A)](E_0-1+D/(1+C\alpha_h)) [ X h + C ( 1 − A )] ( E 0 − 1 + D / ( 1 + C α h )) and is positive.
For the remaining five-point path put X = a + b − 1 X=a+b-1 X = a + b − 1 , Y = c + d − 1 Y=c+d-1 Y = c + d − 1 and R = b c − X Y > 0 R=bc-XY>0 R = b c − X Y > 0 , the last strict inequality following from the full adjacent cofactor identity R = ∏ j ≠ i r j / r i R=\prod_{j\ne i}r_j/r_i R = ∏ j = i r j / r i . Its half determinant is D 5 , h = X h Y h − b c < 0 D_{5,h}=X_hY_h-\sqrt{bc}<0 D 5 , h = X h Y h − b c < 0 , and D 5 , h + R > 0 D_{5,h}+\sqrt{R}>0 D 5 , h + R > 0 . Indeed X h ≥ X X_h\ge\sqrt{X} X h ≥ X , Y h ≥ Y Y_h\ge\sqrt{Y} Y h ≥ Y , and, when X Y > 0 XY>0 X Y > 0 , X Y + b c − X Y > b c \sqrt{XY}+\sqrt{bc-XY}>\sqrt{bc} X Y + b c − X Y > b c ; the case X Y = 0 XY=0 X Y = 0 is immediate. Thus the periodic half adjacent cofactors are positive; the antiperiodic ones have sign opposite the corresponding signed edge.
If the half return were scalar ± I \pm I ± I , all six-point cofactors would vanish. If its trace were 2 2 2 and it were nonscalar, its adjugate would be η v v T \eta vv^{\mathsf T} η v v T , with η > 0 \eta>0 η > 0 and all adjacent coordinates of v v v having the same sign. Eigenvalue one would then be the Perron root, contradicting the three-point principal-path root r i + r i + 1 > 1 \sqrt{r_i}+\sqrt{r_{i+1}}>1 r i + r i + 1 > 1 . At trace − 2 -2 − 2 , the antiperiodic signs instead make every signed edge term of v T H v v^{\mathsf T}Hv v T H v negative, contradicting H v = v Hv=v H v = v . The positive chart is connected and its regular point q i = 2 cos ( 2 π / 7 ) q_i=2\cos(2\pi/7) q i = 2 cos ( 2 π /7 ) has strictly elliptic half return. All finite closed-chart points are limits of positive points (shrink the t i t_i t i and then add a smaller positive increment). Continuity and the excluded parabolic cases prove strict ellipticity throughout the stated closed chart. □
Lemma A.34. For an actual pure-forward seven-arc global-minimizer candidate ,
a i 2 + 2 a i a i − 1 > 1 , a i − 1 2 + 2 a i a i − 1 > 1 , 0 < a i a i − 1 < 1. (285) a_i^2+2a_i a_{i-1}>1,\qquad a_{i-1}^2+2a_i a_{i-1}>1,\qquad0<a_i a_{i-1}<1. \tag*{(285)} a i 2 + 2 a i a i − 1 > 1 , a i − 1 2 + 2 a i a i − 1 > 1 , 0 < a i a i − 1 < 1. (285) Moreover 1 / 5 < a i < 11 / 4 1/5<a_i<11/4 1/5 < a i < 11/4 .
Proof. At each switch choose a genuine fundamental interval containing positive pieces of the capped forward word 1201. Apply the complete maximizing-PMP forward-force lemma, Lemma B.24, on that interval. In its normalized middle-switch coordinates,
d 0 = a i , d 2 = a i − 1 , d 1 = 1 − a i a i − 1 a i + a i − 1 . d_0=a_i,\qquad d_2=a_{i-1},\qquad d_1=\frac{1-a_i a_{i-1}}{a_i+a_{i-1}}. d 0 = a i , d 2 = a i − 1 , d 1 = a i + a i − 1 1 − a i a i − 1 .
The conclusion d 1 < min ( d 0 , d 2 ) d_1<\min(d_0,d_2) d 1 < min ( d 0 , d 2 ) is precisely the first two inequalities. The last is strict star. Different legitimate cuts may have different lifts; only their scalar state conclusions are combined.
Put α i = arctan a i \alpha_i=\arctan a_i α i = arctan a i and x i = 2 α i / π x_i=2\alpha_i/\pi x i = 2 α i / π . The pair inequalities say P i : = x i + x i + 1 < 1 P_i:=x_i+x_{i+1}<1 P i := x i + x i + 1 < 1 and F i j : = 2 x i + x j > 1 F_{ij}:=2x_i+x_j>1 F ij := 2 x i + x j > 1 for each oriented adjacent pair. The state phase is
ϕ = 3 π 2 − ∑ i α i ∈ ( 0 , π ) . (286) \phi=\frac{3\pi}{2}-\sum_i\alpha_i\in(0,\pi). \tag*{(286)} ϕ = 2 3 π − i ∑ α i ∈ ( 0 , π ) . (286)
Here the branch can be fixed on the entire connected exponent chart; we do not assume that its geometrically feasible subset is connected. Indeed, in the coordinates b j = log t j b_j=\log t_j b j = log t j , 0 ≤ j ≤ 3 0\leq j\leq3 0 ≤ j ≤ 3 , the full positive group chart is exactly
b 0 + b 1 < 0 , b 1 + b 2 < 0 , b 2 + b 3 < 0. b_0+b_1<0,\qquad b_1+b_2<0,\qquad b_2+b_3<0. b 0 + b 1 < 0 , b 1 + b 2 < 0 , b 2 + b 3 < 0.
It is convex and contains the regular point. By Lemma A.33, the functions
ϕ = arccos τ 0 2 ∏ i q i ∈ ( 0 , π ) , a i = n i Δ \phi=\arccos\frac{\tau_0}{2\sqrt{\prod_i q_i}}\in(0,\pi),\qquad a_i=\frac{n_i}{\sqrt{\Delta}} ϕ = arccos 2 ∏ i q i τ 0 ∈ ( 0 , π ) , a i = Δ n i
are single-valued real analytic functions on this entire chart. For this continuation the a i a_i a i need not initially be declared positive. The state identities extend throughout the chart in the denominator-free form
q i ( Δ + n i 2 ) = ( n i + n i − 1 ) ( n i + n i + 1 ) . q_i(\Delta+n_i^2)=(n_i+n_{i-1})(n_i+n_{i+1}). q i ( Δ + n i 2 ) = ( n i + n i − 1 ) ( n i + n i + 1 ) .
They follow directly from the cyclic state-return products; equivalently, they hold near the regular point by the original state recursion and then everywhere by the identity principle for real analytic functions on a connected domain. The regular trajectory is strictly in the star domain, so that neighborhood is nonempty. As q i > 0 q_i>0 q i > 0 and Δ > 0 \Delta>0 Δ > 0 , no adjacent sum n i + n i + 1 n_i+n_{i+1} n i + n i + 1 vanishes. Its sign is positive at the regular point and hence is positive throughout the connected chart. Consequently the logarithmic recursion in (A.68) has no hidden logarithm branch there.
Differentiating the trace product gives d ϕ = − 1 2 ∑ i a i d log q i d\phi=-\frac{1}{2}\sum_i a_i\,d\log q_i d ϕ = − 2 1 ∑ i a i d log q i . If K = ∂ log q / ∂ a K=\partial\log q/\partial a K = ∂ log q / ∂ a , the recursion gives a symmetric matrix with
K i i = 1 a i + a i − 1 + 1 a i + a i + 1 − 2 a i 1 + a i 2 , K i , i + 1 = 1 a i + a i + 1 , K_{ii}=\frac{1}{a_i+a_{i-1}}+\frac{1}{a_i+a_{i+1}}-\frac{2a_i}{1+a_i^2},\qquad K_{i,i+1}=\frac{1}{a_i+a_{i+1}}, K ii = a i + a i − 1 1 + a i + a i + 1 1 − 1 + a i 2 2 a i , K i , i + 1 = a i + a i + 1 1 ,
and K a = ( 2 / ( 1 + a i 2 ) ) i K a=(2/(1+a_i^2))_i K a = ( 2/ ( 1 + a i 2 ) ) i . Thus, with the ordinary real branch of arctan,
d ( ϕ + ∑ i arctan a i ) = 0 d\left(\phi+\sum_i\arctan a_i\right)=0 d ( ϕ + i ∑ arctan a i ) = 0
on the whole connected exponent chart. At the regular point q ∗ = 2 cos ( 2 π / 7 ) q_* = 2\cos(2\pi/7) q ∗ = 2 cos ( 2 π /7 ) one has a i = q ∗ / ( 4 − q ∗ ) a_i = \sqrt{q_*}/(4-q_*) a i = q ∗ / ( 4 − q ∗ ) . Write θ = arccos ( q ∗ / 2 ) \theta= \arccos(\sqrt{q_*}/2) θ = arccos ( q ∗ /2 ) , so that arctan a i = π / 2 − θ \arctan a_i = \pi/2-\theta arctan a i = π /2 − θ and 2 π < 7 θ < 7 π / 3 2\pi<7\theta<7\pi/3 2 π < 7 θ < 7 π /3 . Therefore its principal half-state phase is ϕ = 7 θ − 2 π \phi=7\theta-2\pi ϕ = 7 θ − 2 π , and the constant is exactly 3 π / 2 3\pi/2 3 π /2 . This proves (A.70) for every actual candidate, regardless of the connected components of the additional geometric constraints. In particular S : = ∑ x i < 3 S:=\sum x_i<3 S := ∑ x i < 3 . The two exact linear identities
− 7 x 0 = P 1 + P 5 + S − 2 F 01 − F 23 − F 54 − 2 F 06 < − 1 , -7x_0=P_1+P_5+S-2F_{01}-F_{23}-F_{54}-2F_{06}<-1, − 7 x 0 = P 1 + P 5 + S − 2 F 01 − F 23 − F 54 − 2 F 06 < − 1 ,
9 x 0 = 3 P 0 + 3 P 6 + 3 S − 3 F 12 − F 34 − F 43 − 3 F 65 < 7 9x_0=3P_0+3P_6+3S-3F_{12}-F_{34}-F_{43}-3F_{65}<7 9 x 0 = 3 P 0 + 3 P 6 + 3 S − 3 F 12 − F 34 − F 43 − 3 F 65 < 7
give π / 14 < α i < 7 π / 18 \pi/14<\alpha_i<7\pi/18 π /14 < α i < 7 π /18 cyclically. The lower bound implies a i > 1 / 5 a_i>1/5 a i > 1/5 . For the upper bound put y = arctan ( 4 / 11 ) < π / 6 y=\arctan(4/11)<\pi/6 y = arctan ( 4/11 ) < π /6 ; then tan ( 3 y ) = 1388 / 803 < 3 \tan(3y)=1388/803<\sqrt{3} tan ( 3 y ) = 1388/803 < 3 , because 3 ⋅ 803 2 − 1388 2 = 7883 > 0 3\cdot803^2-1388^2=7883>0 3 ⋅ 80 3 2 − 138 8 2 = 7883 > 0 . Hence y < π / 9 y<\pi/9 y < π /9 and a i < cot ( π / 9 ) < 11 / 4 a_i<\cot(\pi/9)<11/4 a i < cot ( π /9 ) < 11/4 . □
Lemma A.35. The extension of the cost in (A.68) to all positive state tuples is strictly increasing in each state variable.
Proof. Write J e x t + 7 = ∑ i F ( a i , a i + 1 ) \mathcal{J}_{\mathrm{ext}}+7=\sum_i F(a_i,a_{i+1}) J ext + 7 = ∑ i F ( a i , a i + 1 ) , where
F ( a , b ) = 3 a b − ( a 2 + b 2 ) log ( a + b ) + a 2 2 log ( 1 + a 2 ) + b 2 2 log ( 1 + b 2 ) . F(a,b)=3ab-(a^2+b^2)\log(a+b)+\frac{a^2}{2}\log(1+a^2)+\frac{b^2}{2}\log(1+b^2). F ( a , b ) = 3 ab − ( a 2 + b 2 ) log ( a + b ) + 2 a 2 log ( 1 + a 2 ) + 2 b 2 log ( 1 + b 2 ) .
For r = b / a > 0 r=b/a>0 r = b / a > 0 and v = 1 / a 2 > 0 v=1/a^2>0 v = 1/ a 2 > 0 ,
F a a = 3 r − 1 + r 2 1 + r − 2 log ( 1 + r ) + log ( 1 + v ) + 1 1 + v > 2 ( r + r 1 + r − log ( 1 + r ) ) > 0. \frac{F_a}{a}=3r-\frac{1+r^2}{1+r}-2\log(1+r)+\log(1+v)+\frac{1}{1+v}>2\left(r+\frac{r}{1+r}-\log(1+r)\right)>0. a F a = 3 r − 1 + r 1 + r 2 − 2 log ( 1 + r ) + log ( 1 + v ) + 1 + v 1 > 2 ( r + 1 + r r − log ( 1 + r ) ) > 0.
The first inequality uses ( log ( 1 + v ) + ( 1 + v ) − 1 ) ′ = v / ( 1 + v ) 2 > 0 (\log(1+v)+(1+v)^{-1})'=v/(1+v)^2>0 ( log ( 1 + v ) + ( 1 + v ) − 1 ) ′ = v / ( 1 + v ) 2 > 0 . The last bracket vanishes at zero and has derivative ( 1 + r + r 2 ) / ( 1 + r ) 2 > 0 (1+r+r^2)/(1+r)^2>0 ( 1 + r + r 2 ) / ( 1 + r ) 2 > 0 . Symmetry gives F b > 0 F_b>0 F b > 0 . □
The following finite state-graph calculation is a domain bound, and is kept separate from the final cost exclusion. Assume only the closed versions of (A.69), a i ∈ [ 1 / 5 , 11 / 4 ] a_i\in[1/5,11/4] a i ∈ [ 1/5 , 11/4 ] , and J e x t ≤ V \mathcal{J}_{\mathrm{ext}}\le V J ext ≤ V . Divide the interval into closed cells C j = [ l j / D , u j / D ] C_j=[l_j/D,u_j/D] C j = [ l j / D , u j / D ] , where u j = l j + 1 u_j=l_j+1 u j = l j + 1 . Retain a directed pair ( j , k ) (j,k) ( j , k ) only if
u j 2 + 2 u j u k ≥ D 2 , u k 2 + 2 u j u k ≥ D 2 , l j l k ≤ D 2 . (287) u_j^2+2u_ju_k\ge D^2,\qquad u_k^2+2u_ju_k\ge D^2,\qquad l_jl_k\le D^2. \tag*{(287)} u j 2 + 2 u j u k ≥ D 2 , u k 2 + 2 u j u k ≥ D 2 , l j l k ≤ D 2 . (287)
Every discarded pair is impossible. On a retained pair take a directed lower bound for F ( l j / D , l k / D ) F(l_j/D,l_k/D) F ( l j / D , l k / D ) , multiply by 10 9 10^9 1 0 9 and round down to an integer W j k W_{jk} W j k ; otherwise set W j k = + ∞ W_{jk}=+\infty W j k = + ∞ . Lemma A.35 proves that these are lower bounds on entire cells, although their lower corners need not satisfy group closure.
Let ⊗ \otimes ⊗ denote minimum-plus multiplication and write W m W^m W m for its powers. For a cycle whose zeroth cell is i i i ,
B i = min j { ( W 2 ) i j + ( W 5 ) j i } − 7 ⋅ 10 9 B_i=\min_j\{(W^2)_{ij}+(W^5)_{ji}\}-7\cdot10^9 B i = j min {( W 2 ) ij + ( W 5 ) j i } − 7 ⋅ 1 0 9
is a necessary cost lower bound. At D = 100 D=100 D = 100 on the full state interval, there are 255 cells and 13,896 retained pairs; the inequalities B i ≤ 1805000000 B_i\le1805000000 B i ≤ 1805000000 restrict every state to [ 9 / 25 , 7 / 5 ] [9/25,7/5] [ 9/25 , 7/5 ] . On this already established interval repeat with D = 2000 D=2000 D = 2000 . There are 2,080 cells and 2,514,283 retained directed pairs, and the necessary state range becomes
757 / 2000 ≤ a i ≤ 67 / 50. (288) 757/2000\le a_i\le67/50. \tag*{(288)} 757/2000 ≤ a i ≤ 67/50. (288)
For a proposed q 0 ≥ 2 q_0\ge2 q 0 ≥ 2 , retain a central triple of cells ( i , j , k ) (i,j,k) ( i , j , k ) only when
( u i + u j ) ( u i + u k ) D 2 + l i 2 ≥ 2. \frac{(u_i+u_j)(u_i+u_k)}{D^2+l_i^2}\ge2. D 2 + l i 2 ( u i + u j ) ( u i + u k ) ≥ 2.
This is a necessary upper bound, with no monotonicity assumption on q 0 q_0 q 0 as a function of a 0 a_0 a 0 . The exact minimum of W i j + W i k + ( W 5 ) j k − 7 ⋅ 10 9 W_{ij}+W_{ik}+(W^5)_{jk}-7\cdot10^9 W ij + W ik + ( W 5 ) j k − 7 ⋅ 1 0 9 over the remaining triples is 1811660735. Dividing by 10 9 10^{9} 1 0 9 gives a bound strictly larger than V = 1.805 V = 1.805 V = 1.805 . Thus every low-cost candidate has
q i < 2 for every i . (289) q_i < 2 \quad\text{for every } i. \tag*{(289)} q i < 2 for every i . (289)
All edge enclosures are completed before these integer minima are taken; the remaining recurrence is finite integer arithmetic. The retained triple count is 731,743,421. It describes a finite enlarged-state proof, not a sampling of geometric trajectories.
Cut at a largest exponent and write T 0 = max i t i = t 0 T_0 = \max_i t_i = t_0 T 0 = max i t i = t 0 . If T 0 ≥ 1 T_0 \ge1 T 0 ≥ 1 , then T 0 > 11 / 20 T_0 > 11/20 T 0 > 11/20 immediately. Otherwise (A.66) gives
T 0 2 ≥ ( 1 − T 0 2 ) 2 1 + T 0 , T 0 3 − 2 T 0 2 − T 0 + 1 ≤ 0. T_0^2 \ge\frac{(1-T_0^2)^2}{1+T_0}, \qquad T_0^3 - 2T_0^2 - T_0 + 1 \le0. T 0 2 ≥ 1 + T 0 ( 1 − T 0 2 ) 2 , T 0 3 − 2 T 0 2 − T 0 + 1 ≤ 0.
The cubic is decreasing on [ 0 , 11 / 20 ] [0,11/20] [ 0 , 11/20 ] and has value 91 / 8000 91/8000 91/8000 at its right endpoint. Hence q 0 > 31 / 20 q_0 > \sqrt{31}/20 q 0 > 31 /20 in either case, and E > 11 / 20 E > 11/20 E > 11/20 . This avoids a singular tail denominator at a zero-duration boundary.
For the interval divisions one may use a uniform rational discriminant floor on the filtered candidate set. Here is its quantitative proof. Put ρ i = r i \rho_i = \sqrt{r_i} ρ i = r i and let C i C_i C i be the periodic half-state adjacent cofactor in Lemma A.33. Since q i ≤ 2 q_i \le2 q i ≤ 2 , r i ≥ 1 / 2 r_i \ge1/2 r i ≥ 1/2 . More generally, if r i ≥ δ r_i \ge\delta r i ≥ δ , the notation in that proof gives X h , Y h ≥ δ 2 / 3 X_h, Y_h \ge\delta^2/3 X h , Y h ≥ δ 2 /3 and R ≥ δ 6 R \ge\delta^6 R ≥ δ 6 . Set z = X Y z = \sqrt{XY} z = X Y , w = R w = \sqrt{R} w = R and A = D 5 , h + R A = D_{5,h} + \sqrt{R} A = D 5 , h + R . Then
A ≥ δ 4 / 9 − z , A ≥ z + w − z 2 + w 2 ≥ 2 z w / 3. A \ge\delta^4/9-z, \qquad A \ge z+w-\sqrt{z^2+w^2} \ge2zw/3. A ≥ δ 4 /9 − z , A ≥ z + w − z 2 + w 2 ≥ 2 z w /3.
The last bound uses z 2 + w 2 = b c ≤ 1 z^2+w^2 = bc \le1 z 2 + w 2 = b c ≤ 1 , so that the rationalizing denominator is at most 1 + 2 < 3 1+\sqrt{2}<3 1 + 2 < 3 . Splitting at z = δ 4 / 18 z=\delta^4/18 z = δ 4 /18 gives A ≥ δ 7 / 27 A\ge\delta^7/27 A ≥ δ 7 /27 and hence
ρ i C i ∏ j ρ j ≥ δ 8 27 . \frac{\rho_i C_i}{\prod_j \rho_j} \ge\frac{\delta^8}{27}. ∏ j ρ j ρ i C i ≥ 27 δ 8 .
Take c = 1 / 6912 c=1/6912 c = 1/6912 and write b = tr ( P 0 / ∏ q i ) = 2 cos ϕ b=\operatorname{tr}(P_0/\sqrt{\prod q_i})=2\cos\phi b = tr ( P 0 / ∏ q i ) = 2 cos ϕ , e = 2 − b e=2-b e = 2 − b . Summing the angular PMP inequalities gives ∑ α i > 7 π / 6 \sum\alpha_i>7\pi/6 ∑ α i > 7 π /6 ; by (A.70), 0 < ϕ < π / 3 0<\phi<\pi/3 0 < ϕ < π /3 , so 0 < e < 1 0<e<1 0 < e < 1 . Differentiating the seven-cycle characteristic identity gives
a i sin ϕ = ρ i C i ∏ j ρ j − e 2 ≥ c − e 2 . a_i\sin\phi= \frac{\rho_i C_i}{\prod_j \rho_j}-\frac{e}{2} \ge c-\frac{e}{2}. a i sin ϕ = ∏ j ρ j ρ i C i − 2 e ≥ c − 2 e .
Since a i ≤ 7 / 5 a_i\le7/5 a i ≤ 7/5 and sin ϕ ≤ e \sin\phi\le\sqrt{e} sin ϕ ≤ e , either e ≥ c e\ge c e ≥ c , or ( 7 / 5 ) e > c / 2 (7/5)\sqrt{e}>c/2 ( 7/5 ) e > c /2 . In both cases the normalized discriminant obeys
Δ ‾ : = Δ ∏ q i = e ( 4 − e ) ≥ 25 3121348608 > 1 125000000 . \overline{\Delta}:=\frac{\Delta}{\prod q_i}=e(4-e)\ge\frac{25}{3121348608}>\frac{1}{125000000}. Δ := ∏ q i Δ = e ( 4 − e ) ≥ 3121348608 25 > 125000000 1 .
Moreover n ‾ i : = n i / ∏ q i > 1 / 32000 \overline{n}_i:=n_i/\sqrt{\prod q_i}>1/32000 n i := n i / ∏ q i > 1/32000 , using a i ≥ 9 / 25 a_i\ge9/25 a i ≥ 9/25 . The same weaker floors hold for the raw Δ , n i \Delta,n_i Δ , n i . These are necessary candidate bounds and may be intersected with value enclosures after the filter is imposed. They are not bounds on unrelated points of a raw parameter rectangle and are not used to clip its Taylor derivatives.
Lemma A.36 (Finite pure-forward inequality). On the chart (A.65), every point satisfying the necessary low-cost filters (A.69), (A.72), (A.73), the maximum cut and the positive state branch has J > U \mathcal{J}>U J > U .
Proof. It suffices to prove positivity of the explicit numerator
Φ 0 = 3 ∑ i n i n i + 1 − ∑ i n i 2 log q i − ( 7 + U ) Δ . (290) \Phi_0=3\sum_i n_i n_{i+1}-\sum_i n_i^2\log q_i-(7+U)\Delta. \tag*{(290)} Φ 0 = 3 i ∑ n i n i + 1 − i ∑ n i 2 log q i − ( 7 + U ) Δ. (290)
Use the closed root in ( q 0 , q 1 , q 2 , q 3 ) (q_0,q_1,q_2,q_3) ( q 0 , q 1 , q 2 , q 3 ) with q 0 ∈ [ q ∗ , 2 ] q_0\in[q_*,2] q 0 ∈ [ q ∗ , 2 ] , q 1 , q 2 , q 3 ∈ [ 1 , 2 ] q_1,q_2,q_3\in[1,2] q 1 , q 2 , q 3 ∈ [ 1 , 2 ] , where q ∗ = 0 x 1.3 e b 7 a 9792 f 3 c 5 p + 0 q_* = 0x1.3eb7a9792f3c5p+0 q ∗ = 0 x 1.3 e b 7 a 9792 f 3 c 5 p + 0 is an exact binary rational below 31 / 20 \sqrt{31}/20 31 /20 . Impose q j ≤ q 0 q_j\le q_0 q j ≤ q 0 , the tail nonnegativity and all cyclic adjacent-product inequalities. In numerator coordinates the state filters are
n i 2 + 2 n i n i − 1 − Δ ≥ 0 , n i − 1 2 + 2 n i n i − 1 − Δ ≥ 0 , Δ − n i n i − 1 ≥ 0 , 3 n i n i − 1 − Δ ≥ 0 , n_i^2 + 2n_i n_{i-1} - \Delta\ge0,\qquad n_{i-1}^2 + 2n_i n_{i-1} - \Delta\ge0,\qquad\Delta- n_i n_{i-1} \ge0,\qquad3n_i n_{i-1} - \Delta\ge0, n i 2 + 2 n i n i − 1 − Δ ≥ 0 , n i − 1 2 + 2 n i n i − 1 − Δ ≥ 0 , Δ − n i n i − 1 ≥ 0 , 3 n i n i − 1 − Δ ≥ 0 ,
( 4489 / 2500 ) Δ − n i 2 ≥ 0 , n i 2 − ( 573049 / 4000000 ) Δ ≥ 0 , n i > 0. (4489/2500)\Delta- n_i^2 \ge0,\qquad n_i^2 - (573049/4000000)\Delta\ge0,\qquad n_i > 0. ( 4489/2500 ) Δ − n i 2 ≥ 0 , n i 2 − ( 573049/4000000 ) Δ ≥ 0 , n i > 0.
Products of the matrices (A.67) define all the polynomials; there is no fitted polynomial kernel.
For reproducibility we describe the finite witness and its checking rules. Its 16 root slabs share exact binary endpoints and partition the root in the q 0 q_0 q 0 direction. Each node records its rational box, any necessary-domain contraction and either a two-child split or one of the strict terminal tests below. There are 139,838 nodes, 69,911 splits and 69,927 terminal leaves; all 16 trees end without a pending box. A terminal box must satisfy one of the following:
a necessary nonnegative constraint has a strictly negative upper bound, or a positive denominator/branch requirement is impossible;
a directed lower bound for (A.74) is strictly positive;
for specified nonnegative rational numbers λ k \lambda_k λ k and necessary constraints g k ≥ 0 g_k \ge0 g k ≥ 0 , a directed lower bound for Φ 0 − ∑ k λ k g k \Phi_0 - \sum_k \lambda_k g_k Φ 0 − ∑ k λ k g k is strictly positive.
The last test is valid because it bounds a function no larger than Φ 0 \Phi_0 Φ 0 at every candidate. A heuristic may suggest the numbers λ k \lambda_k λ k but does not supply any exclusion.
Besides elementary interval contractions, the exact cyclic identity allows the following one. If upper bounds for t i − 1 , t i + 1 , t i + 3 , t i + 4 t_{i-1},t_{i+1},t_{i+3},t_{i+4} t i − 1 , t i + 1 , t i + 3 , t i + 4 are A A A , B B B , C C C , D D D and P = C D < 1 P=CD<1 P = C D < 1 , then
t i ≥ 2 ( 1 − P ) A + B + P + ( A − B ) 2 + P ( 2 ( A + B ) + P + 4 A B ) . t_i \ge\frac{2(1-P)}{A+B+P+\sqrt{(A-B)^2+P(2(A+B)+P+4AB)}}. t i ≥ A + B + P + ( A − B ) 2 + P ( 2 ( A + B ) + P + 4 A B ) 2 ( 1 − P ) .
This is the smaller root of ( 1 − A u ) ( 1 − B u ) − P ( 1 + u ) (1-Au)(1-Bu)-P(1+u) ( 1 − A u ) ( 1 − B u ) − P ( 1 + u ) , rationalized. It lies below both finite reciprocals 1 / A , 1 / B 1/A,1/B 1/ A , 1/ B ; a smaller actual u u u contradicts (A.66). The degenerate zero-denominator case is infeasible. These contractions preserve candidates; they do not license clipping derivatives of a function on the full uncontracted box.
The directed arithmetic and complete tree-replay rules are given in Appendix H.11.4. They verify every terminal test on its whole box and the exact coverage of all sixteen roots. The state-graph inequalities and matrix identities above are separate prerequisites. Positivity at all terminal leaves, together with exact tree coverage, proves (A.74) on the necessary set. Since Δ > 0 \Delta>0 Δ > 0 by Lemma A.33, the stated cost inequality follows. □
Proof of Proposition A.32. If an actual candidate has cost greater than V V V , it cannot minimize. Otherwise the state filter and finite state graph apply. The maximum cut puts it in the complete chart used by Lemma A.36 , which gives J > U > J 8 \mathcal{J}>\mathcal{U}>\mathcal{J}_8 J > U > J 8 , contradicting the octagonal competitor. The use of the larger prerequisite budget V V V ensures that the domain reduction does not assume the conclusion at budget U \mathcal{U} U . □
Remark A.37. The proof combines several kinds of result. The reverse five- and seven-arc inequalities compare all feasible trajectories in those word classes. The six-arc arguments give actual decreases within their closed word families. The mixed and pure-forward seven-arc arguments use complete maximization for an original global minimizer. Thus Proposition A.15 supplies the lower end of the global candidate classification. It does not assert the cost inequality J ≥ J 8 \mathcal{J}\ge\mathcal{J}_8 J ≥ J 8 for every feasible seven-arc chain.
Appendix B. Fundamental local estimates
The assembly charges three- and four-arc cores in many overlapping sources. We first derive the common binary equations, root geometry and cap losses; the separate scalar bounds then differ only in which actual returns are available. This order makes the source of each extra term explicit.
We use the source convention of Section 8.1. Containment remains a geometric hypothesis: changes of cut use Lemma 2.6, and whole changes of chart use Proposition 2.5.
Binary cores and their exterior maximizing jets
For complete-source cost estimates, each source uses one full lift on its containing interval; no state covector is continued periodically across the seam.
Lemma B.1 (Automatic properness of two complete sources). Suppose one genuine fundamental interval of an original global minimizer contains L L L consecutive complete positive maximal arcs. Then L ≤ B L \le B L ≤ B . If their source is 012012 or 01020212, or a whole lawful cyclic or reflected-time image, then L < B L < B L < B .
Proof. A cut at a switch leaves exactly B B B complete maximal arcs in one period. An interior cut splits one physical arc and leaves B − 1 B-1 B − 1 complete arcs. Consequently L ≤ B L \le B L ≤ B , and equality requires a switch cut with the source occupying the entire switch-to-switch period. The actual twist then gives j i + B = j i + 1 j_{i+B}=j_i+1 j i + B = j i + 1 (mod 3) at the closing switch.
For 012012 with B = 6 B=6 B = 6 , the resulting cyclic increment word is + + + + + − +++++ - + + + + + − . Its last positive increment, closing negative increment and first positive increment form + , − , + +,-,+ + , − , + . For 01020212 with B = 8 B=8 B = 8 , the closed increment word is + − − + − − + − +--+--+- + − − + − − + − , with the same forbidden triple across the seam. Each such triple is a cyclic phase of the complete four-arc word 0101, excluded by Lemma A.4 without a normality hypothesis. These four-arc witnesses are proper since 4 < 6 , 8 4<6,8 4 < 6 , 8 . A whole cyclic phase leaves the increments unchanged. A full reflection with reversed physical time reverses their cyclic order and preserves the + 1 +1 + 1 twist and the forbidden triple. Thus equality is impossible in every stated image. □
The containing interval and complete maximal arcs are essential here. This lemma makes no assertion for arbitrary endpoint-constrained extremals, for fragments of maximal arcs, or for a multi-period interval. It neither continues a costate across the seam nor assumes its periodicity.
Lemma B.2 (Binary reduction). On a connected sequence of modes 0 and 2, write
Λ = ( l b m − l ) , k = 2 u l + b − u 2 m , L = l − u m , \Lambda=\begin{pmatrix} l & b \\ m & -l \end{pmatrix},\qquad k=2ul+b-u^2m,\qquad L=l-um, Λ = ( l m b − l ) , k = 2 u l + b − u 2 m , L = l − u m , χ = v p v , q = v − 2 , d s = d t / v = q d u . \chi=vp_v,\qquad q=v^{-2},\qquad ds=dt/v=q\,du. χ = v p v , q = v − 2 , d s = d t / v = q d u . For an active mode j ∈ { 0 , 2 } j\in\{0,2\} j ∈ { 0 , 2 } , the complete equations imply
k s = 0 , L s = k , m s = − 2 L , χ s = 2 k − η ( 3 + u 2 ) − v H j (291) k_s=0,\qquad L_s=k,\qquad m_s=-2L,\qquad\chi_s=2k-\eta(3+u^2)-vH_j \tag*{(291)} k s = 0 , L s = k , m s = − 2 L , χ s = 2 k − η ( 3 + u 2 ) − v H j (291) det Λ = − L 2 − k m , tr ( Λ X ) = k − v 2 m v . (292) \det\Lambda=-L^2-km,\qquad\operatorname{tr}(\Lambda X)=\frac{k-v^2m}{v}. \tag*{(292)} det Λ = − L 2 − k m , tr ( Λ X ) = v k − v 2 m . (292) Let e = 1 + u e=1+u e = 1 + u on mode 0 and e = u − 1 e=u-1 e = u − 1 on mode 2, and set
f = ( k − η 2 ( 3 + u 2 ) ) q + 1 2 ( e q + e − 1 ) χ . f=\left(k-\frac{\eta}{2}(3+u^2)\right)q+\frac{1}{2}(eq+e^{-1})\chi. f = ( k − 2 η ( 3 + u 2 ) ) q + 2 1 ( e q + e − 1 ) χ . The inactive Hamiltonian identity is
H 1 = 2 v 3 u 2 + v 2 − 1 ( f − m − η / 2 ) − 1 − u 2 + v 2 u 2 + v 2 − 1 H j . (293) H_1=\frac{2v^3}{u^2+v^2-1}\left(f-m-\eta/2\right)-\frac{1-u^2+v^2}{u^2+v^2-1}H_j. \tag*{(293)} H 1 = u 2 + v 2 − 1 2 v 3 ( f − m − η /2 ) − u 2 + v 2 − 1 1 − u 2 + v 2 H j . (293)
In particular, on the actual core H j = 0 H_j=0 H j = 0 , the function g = m + η / 2 − f g=m+\eta/2-f g = m + η /2 − f is nonnegative. It vanishes at an exterior mode-1 cap. Moreover
q u = − 2 q 1 + u ( j = 0 ) , q u = 2 q 1 − u ( j = 2 ) , H 0 − H 2 = 2 v χ 1 − u 2 . (294) q_u=-\frac{2q}{1+u}\quad(j=0),\qquad q_u=\frac{2q}{1-u}\quad(j=2),\qquad H_0-H_2=\frac{2v\chi}{1-u^2}. \tag*{(294)} q u = − 1 + u 2 q ( j = 0 ) , q u = 1 − u 2 q ( j = 2 ) , H 0 − H 2 = 1 − u 2 2 v χ . (294)
Proof. The two active state fields both have u t = v u_t=v u t = v , so u s = v 2 u_s=v^2 u s = v 2 and d s = q d u ds=q\,du d s = q d u . The Lax equation gives
l t = b + m ( u 2 + v 2 ) v , b t = − 2 ( l ( u 2 + v 2 ) + u b ) v , m t = 2 ( u m − l ) v . l_t=\frac{b+m(u^2+v^2)}{v},\qquad b_t=-\frac{2(l(u^2+v^2)+ub)}{v},\qquad m_t=\frac{2(um-l)}{v}. l t = v b + m ( u 2 + v 2 ) , b t = − v 2 ( l ( u 2 + v 2 ) + u b ) , m t = v 2 ( u m − l ) .
Substitution into the definitions of k , L k,L k , L proves their three differential identities and direct expansion proves (B.2). For the state adjoint, differentiate v p v vp_v v p v using p t = − ∇ W H j p_t=-\nabla_W H_j p t = − ∇ W H j , with l , b , m l,b,m l , b , m held fixed in this gradient. The terms involving p u p_u p u are eliminated using the displayed expression for H j H_j H j , leaving the fourth identity of (B.1). The same elimination in H 1 H_1 H 1 gives (B.3). Finally, differentiating v − 2 v^{-2} v − 2 using d v / d u = v / e dv/du=v/e d v / d u = v / e proves the first two identities of (B.4); subtraction of the two original Hamiltonians proves the third. These computations use the active Hamiltonian only; the inactive Hamiltonian is never set to zero in the core interior. Since u 2 + v 2 − 1 > 0 u^2+v^2-1>0 u 2 + v 2 − 1 > 0 , full maximization gives g ≥ 0 g\geq0 g ≥ 0 . □
Lemma B.3 (Equality-only capped-binary rigidity). Suppose that two full lifts with the same objective multiplier follow the same actual trajectory on a connected interval containing a complete positive binary core, two distinct internal binary switches, and positive portions of its actual exterior arcs of the third mode. Both core-endpoint third-mode ties belong to this interval; the far endpoints of the exterior arcs need not. In a lawful whole chart the binary modes are 0, 2. Then the two lifts agree on the interval. In particular, a full nonzero maximizing lift on a containing genuine period is normal. The contexts 12021 and 102021 both satisfy these hypotheses.
Proof. Subtract the full adjoint equations and active zero energies. Their homogeneous binary reduction, Lemma B.2, gives
K s = 0 , χ s = 2 K , L s = K , m s = − 2 L . (295) K_s=0,\qquad\chi_s=2K,\qquad L_s=K,\qquad m_s=-2L. \tag*{(295)} K s = 0 , χ s = 2 K , L s = K , m s = − 2 L . (295) The two distinct actual internal ties force K = χ = 0 K=\chi=0 K = χ = 0 . The two actual third-mode ties in (B.3) then force m = 0 m=0 m = 0 at both core endpoints; hence L = m = 0 L=m=0 L = m = 0 throughout. Reconstruction and the active energy give Δ Λ = Δ p = 0 \Delta\Lambda=\Delta p=0 ΔΛ = Δ p = 0 . Explicitly, the homogeneous solution is χ = c + 2 K s \chi=c+2Ks χ = c + 2 K s , L = L 0 + K s L=L_0+Ks L = L 0 + K s , m = m 0 − 2 L 0 s − K s 2 m=m_0-2L_0s-Ks^2 m = m 0 − 2 L 0 s − K s 2 . The two root rows and two cap rows have determinant 4 S c ( s 1 − s 2 ) ≠ 0 4S_c(s_1-s_2)\ne0 4 S c ( s 1 − s 2 ) = 0 in ( K , c , L 0 , m 0 ) (K,c,L_0,m_0) ( K , c , L 0 , m 0 ) , where 0 < s 1 < s 2 < S c 0<s_1<s_2<S_c 0 < s 1 < s 2 < S c . No cap coefficient or matrix entry is divided by, and the difference need not maximize. Uniqueness for the full linear adjoint equation propagates zero through all actual switches on the connected interval. Applying this argument to an abnormal lift annihilates it, contrary to nontriviality. Thus normality precedes joint normalization of every multiplier. □
Corollary B.4 (Reconstruction from a rigid capped core). For the actual state path in Lemma B.3, its normalized full reduced lift is reconstructed as follows. Let s 1 < s 2 s_1<s_2 s 1 < s 2 be two internal binary ties, a , b a,b a , b the core endpoints, and S = s b − s a > 0 S=s_b-s_a>0 S = s b − s a > 0 . In the notation of Lemma B.2,
k = 1 2 ( 3 + ∫ s 1 s 2 u ( s ) 2 d s s 2 − s 1 ) , χ ( s ) = ∫ s 1 s ( 2 k − 3 − u ( t ) 2 ) d t . k=\frac{1}{2}\left(3+\frac{\int_{s_1}^{s_2}u(s)^2\,ds}{s_2-s_1}\right),\qquad\chi(s)=\int_{s_1}^{s}(2k-3-u(t)^2)\,dt. k = 2 1 ( 3 + s 2 − s 1 ∫ s 1 s 2 u ( s ) 2 d s ) , χ ( s ) = ∫ s 1 s ( 2 k − 3 − u ( t ) 2 ) d t . Compute f a , f b f_a,f_b f a , f b from the displayed definition of f f f with η = 1 \eta=1 η = 1 , and put
m a = f a − 1 2 , m b = f b − 1 2 , L a = m a − m b − k S 2 2 S . m_a=f_a-\frac{1}{2},\qquad m_b=f_b-\frac{1}{2},\qquad L_a=\frac{m_a-m_b-kS^2}{2S}. m a = f a − 2 1 , m b = f b − 2 1 , L a = 2 S m a − m b − k S 2 . For τ = s − s a \tau=s-s_a τ = s − s a , the full reconstruction is
L = L a + k τ , m = m a − 2 L a τ − k τ 2 , l = L + u m , b = k − 2 u L − u 2 m , L=L_a+k\tau,\qquad m=m_a-2L_a\tau-k\tau^2,\qquad l=L+um,\qquad b=k-2uL-u^2m, L = L a + k τ , m = m a − 2 L a τ − k τ 2 , l = L + u m , b = k − 2 uL − u 2 m ,
p v = χ / v , p u = − tr ( ( Λ − A ∗ ) X ) − F j v p v v . p_v = \chi/v,\qquad p_u = \frac{-\operatorname{tr}((\Lambda-A_*)X)-F_j^v p_v}{v}. p v = χ / v , p u = v − tr (( Λ − A ∗ ) X ) − F j v p v .
Here s s s is shear time, and j j j is the actual adjacent binary mode.
Proof. Integrate (B.1) from the two internal zeros and use (B.3) at the two actual caps. The final row is the active zero energy, with F j v = v > 0 F_j^v=v>0 F j v = v > 0 . All formulas therefore concern the unique actual lift, including both state covectors. □
Lemma B.5 (Normality and actual roots of a capped four-core). For the positive context 102021, the selected complete lift is normal. After jointly normalizing it to η = 1 \eta=1 η = 1 , its charged 0202 core has endpoints − α , β -\alpha,\beta − α , β and internal switches − ξ , z , ζ -\xi,z,\zeta − ξ , z , ζ satisfying
− α < − ξ < z < ζ < β , α > ξ > h > 0 , β > ζ > h , -\alpha<-\xi<z<\zeta<\beta,\qquad\alpha>\xi>h>0,\qquad\beta>\zeta>h, − α < − ξ < z < ζ < β , α > ξ > h > 0 , β > ζ > h , ∣ z ∣ ≤ h , h 2 = 2 k − 3 , 3 2 < k < 2. (296) |z|\le h,\qquad h^2=2k-3,\qquad\frac{3}{2}<k<2. \tag*{(296)} ∣ z ∣ ≤ h , h 2 = 2 k − 3 , 2 3 < k < 2. (296) If S S S is the total core shear and A A A its physical cost, then
2 A = d + 3 S + M , d = α + β , M = ∫ u 2 d s , A > 2 S . (297) 2A=d+3S+M,\qquad d=\alpha+\beta,\qquad M=\int u^2\,\mathrm{d}s,\qquad A>2S. \tag*{(297)} 2 A = d + 3 S + M , d = α + β , M = ∫ u 2 d s , A > 2 S . (297)
Proof. At each of the three internal switches χ = 0 \chi=0 χ = 0 , while full maximization requires χ ≥ 0 \chi\ge0 χ ≥ 0 on mode 0 and χ ≤ 0 \chi\le0 χ ≤ 0 on mode 2. Lemma B.3 proves normality; normalize the complete lift jointly.
Put μ = 2 k − 3 \mu=2k-3 μ = 2 k − 3 . On each of the two middle arcs the zero-to-zero equation is
μ = ∫ u 2 q d u ∫ q d u > 0. \mu=\frac{\int u^2q\,\mathrm{d}u}{\int q\,\mathrm{d}u}>0. μ = ∫ q d u ∫ u 2 q d u > 0.
The weak one-sided maximizing jets give ξ 2 ≥ μ \xi^2\ge\mu ξ 2 ≥ μ , z 2 ≤ μ z^2\le\mu z 2 ≤ μ , ζ 2 ≥ μ \zeta^2\ge\mu ζ 2 ≥ μ . Ordering implies − ξ < 0 < ζ -\xi<0<\zeta − ξ < 0 < ζ . Equality ξ = μ \xi=\sqrt{\mu} ξ = μ would make ∫ − ξ z ( μ − u 2 ) q d u \int_{-\xi}^{z}(\mu-u^2)q\,\mathrm{d}u ∫ − ξ z ( μ − u 2 ) q d u strictly positive, since this interval has positive length; equality on the right is excluded in the same way. The open state domain gives ∣ u ∣ < 1 |u|<1 ∣ u ∣ < 1 , proving (B.6). On the first and last arcs ∣ u ∣ > h |u|>h ∣ u ∣ > h , so their terminal switching values give χ ( − α ) > 0 > χ ( β ) \chi(-\alpha)>0>\chi(\beta) χ ( − α ) > 0 > χ ( β ) . Finally, v 2 d s = d u v^2\mathrm{d}s=\mathrm{d}u v 2 d s = d u gives (B.7); its last inequality follows from ℓ d t = ( 3 + u 2 + v 2 ) d s / 2 > 2 d s \ell\,\mathrm{d}t=(3+u^2+v^2)\mathrm{d}s/2>2\mathrm{d}s ℓ d t = ( 3 + u 2 + v 2 ) d s /2 > 2 d s . □
Lemma B.6 (One exterior-leaf calculation). In one jointly normalized full normal binary lift, consider an actual first mode-0 leaf [ − t , − r ] [-t,-r] [ − t , − r ] or last mode-2 leaf [ r , t ] [r,t] [ r , t ] , with 0 < h < r < t < 1 0<h<r<t<1 0 < h < r < t < 1 , h 2 = 2 k − 3 h^2=2k-3 h 2 = 2 k − 3 , and an internal tie at ∓ r \mp r ∓ r . Write q t q_t q t for its outer density, f t f_t f t for its outer cap function, and D o u t D_{\mathrm{out}} D out for differentiation in shear directed towards the outer endpoint. Then
( h 2 − t 2 ) q t 2 < f t < 0 , D o u t f t < − t + t 2 − h 2 2 ( 1 + t ) < 0. \frac{(h^2-t^2)q_t}{2}<f_t<0,\qquad D_{\mathrm{out}}f_t<-t+\frac{t^2-h^2}{2(1+t)}<0. 2 ( h 2 − t 2 ) q t < f t < 0 , D out f t < − t + 2 ( 1 + t ) t 2 − h 2 < 0. With E t = − f t E_t=-f_t E t = − f t one also has
E t q t > e ( t , h ) ≔ t 2 − h 2 2 − ( t − h ) 2 ( t + 2 h ) 3 ( 1 − t 2 ) . \frac{E_t}{q_t}>e(t,h)\coloneqq\frac{t^2-h^2}{2}-\frac{(t-h)^2(t+2h)}{3(1-t^2)}. q t E t > e ( t , h ) : = 2 t 2 − h 2 − 3 ( 1 − t 2 ) ( t − h ) 2 ( t + 2 h ) . If the actual endpoint additionally satisfies q t < [ ( 1 − t ) ( 3 + t ) ] − 1 q_t<[(1-t)(3+t)]^{-1} q t < [( 1 − t ) ( 3 + t ) ] − 1 , then
E t q t > e 2 ( t , h ) ≔ t 2 − h 2 2 − 2 ( t − h ) 2 ( t + 2 h ) 3 ( 1 − t ) ( 3 + t ) . \frac{E_t}{q_t}>e_2(t,h)\coloneqq\frac{t^2-h^2}{2}-\frac{2(t-h)^2(t+2h)}{3(1-t)(3+t)}. q t E t > e 2 ( t , h ) : = 2 t 2 − h 2 − 3 ( 1 − t ) ( 3 + t ) 2 ( t − h ) 2 ( t + 2 h ) . These lower bounds are positive respectively for h < t ≤ 259 / 500 h<t\le259/500 h < t ≤ 259/500 and h < t ≤ 349 / 500 h<t\le349/500 h < t ≤ 349/500 .
Proof. Use the outward coordinate x = − u x=-u x = − u on the first leaf and x = u x=u x = u on the last. This is an integral substitution on the same lift. In both cases,
q ( x ) = κ − 2 ( 1 − x ) 2 , ψ ( x ) = ∫ r x ( y 2 − h 2 ) q ( y ) d y , q(x)=\frac{\kappa^{-2}}{(1-x)^2},\qquad\psi(x)=\int_{r}^{x}(y^2-h^2)q(y)\,\mathrm{d}y, q ( x ) = ( 1 − x ) 2 κ − 2 , ψ ( x ) = ∫ r x ( y 2 − h 2 ) q ( y ) d y ,
f ( x ) = ( h 2 − x 2 ) q ( x ) 2 + B ( x ) ψ ( x ) , B ( x ) = q ( x ) ( 1 − x ) + ( 1 − x ) − 1 2 . f(x)=\frac{(h^2-x^2)q(x)}{2}+B(x)\psi(x),\qquad B(x)=\frac{q(x)(1-x)+(1-x)^{-1}}{2}. f ( x ) = 2 ( h 2 − x 2 ) q ( x ) + B ( x ) ψ ( x ) , B ( x ) = 2 q ( x ) ( 1 − x ) + ( 1 − x ) − 1 .
If σ \sigma σ is the leaf shear, then 0 < ψ ( t ) < ( t 2 − h 2 ) σ 0<\psi(t)<(t^2-h^2)\sigma 0 < ψ ( t ) < ( t 2 − h 2 ) σ and
2 σ B ( t ) q t = ( t − r ) ( 1 + κ − 2 ) 1 − r < 1. \frac{2\sigma B(t)}{q_t}=\frac{(t-r)(1+\kappa^{-2})}{1-r}<1. q t 2 σ B ( t ) = 1 − r ( t − r ) ( 1 + κ − 2 ) < 1.
Indeed Ω \Omega Ω gives κ − 2 < ( 1 − t ) / ( 1 + t ) \kappa^{-2}<(1-t)/(1+t) κ − 2 < ( 1 − t ) / ( 1 + t ) , and the remaining positive difference is ( 1 − t ) ( 1 + r ) (1-t)(1+r) ( 1 − t ) ( 1 + r ) . Hence the stated bounds on f t f_t f t . Differentiating gives
D o u t f = − x + f q ( 1 − x ) + ( x 2 − h 2 ) q ( 1 − x ) 2 , D_{\mathrm{out}}f=-x+\frac{f}{q(1-x)}+\frac{(x^2-h^2)q(1-x)}{2}, D out f = − x + q ( 1 − x ) f + 2 ( x 2 − h 2 ) q ( 1 − x ) ,
so f t < 0 f_t<0 f t < 0 and q t ( 1 − t ) < ( 1 + t ) − 1 q_t(1-t)<(1+t)^{-1} q t ( 1 − t ) < ( 1 + t ) − 1 prove the strict jet bound. Finally, with I = ∫ r t ( x 2 − h 2 ) ( 1 − x ) − 2 d x I=\int_r^t(x^2-h^2)(1-x)^{-2}\,\mathrm{d}x I = ∫ r t ( x 2 − h 2 ) ( 1 − x ) − 2 d x ,
E t q t = t 2 − h 2 2 − 1 − t 2 { 1 + ( 1 − t ) 2 q t } I , 0 < I < ( t − h ) 2 ( t + 2 h ) 3 ( 1 − t ) 2 . \frac{E_t}{q_t}=\frac{t^2-h^2}{2}-\frac{1-t}{2}\{1+(1-t)^2q_t\}I,\qquad0<I<\frac{(t-h)^2(t+2h)}{3(1-t)^2}. q t E t = 2 t 2 − h 2 − 2 1 − t { 1 + ( 1 − t ) 2 q t } I , 0 < I < 3 ( 1 − t ) 2 ( t − h ) 2 ( t + 2 h ) .
The coefficient of I I I is smaller than ( 1 − t ) / ( 1 + t ) (1-t)/(1+t) ( 1 − t ) / ( 1 + t ) by Ω \Omega Ω , and smaller than 2 ( 1 − t ) / ( 3 + t ) 2(1-t)/(3+t) 2 ( 1 − t ) / ( 3 + t ) under the additional circle hypothesis. These give e e e , e 2 e_2 e 2 . Their positivity follows from
e ( t , h ) t 2 − h 2 = 1 2 − t 3 ( 1 − t 2 ) + 2 h 2 3 ( 1 − t 2 ) ( t + h ) > 0 \frac{e(t,h)}{t^2-h^2}=\frac{1}{2}-\frac{t}{3(1-t^2)}+\frac{2h^2}{3(1-t^2)(t+h)}>0 t 2 − h 2 e ( t , h ) = 2 1 − 3 ( 1 − t 2 ) t + 3 ( 1 − t 2 ) ( t + h ) 2 h 2 > 0
on the first range, and from ( t − h ) ( t + 2 h ) ≤ t ( t + h ) (t-h)(t+2h)\leq t(t+h) ( t − h ) ( t + 2 h ) ≤ t ( t + h ) and 3 ( 1 − t ) ( 3 + t ) − 4 t ≥ 139597 / 250000 > 0 3(1-t)(3+t)-4t\geq139597/250000>0 3 ( 1 − t ) ( 3 + t ) − 4 t ≥ 139597/250000 > 0 on the second. □
Lemma B.7 (The actual matrix apex). In the preceding notation let E a = − f ( − α ) E_a=-f(-\alpha) E a = − f ( − α ) , E b = − f ( β ) E_b=-f(\beta) E b = − f ( β ) , E = E a + E b E=E_a+E_b E = E a + E b , and C = det Λ \mathcal{C}=\det\Lambda C = det Λ . Then E a , E b > 0 E_a,E_b>0 E a , E b > 0 and
∣ E a − E b ∣ < k S 2 , (298) |E_a-E_b|<kS^2, \tag*{(298)} ∣ E a − E b ∣ < k S 2 , (298) C = k 2 + k E 2 − ( E a − E b ) 2 4 S 2 − k 2 S 2 4 , (299) \mathcal{C}=\frac{k}{2}+\frac{kE}{2}-\frac{(E_a-E_b)^2}{4S^2}-\frac{k^2S^2}{4}, \tag*{(299)} C = 2 k + 2 k E − 4 S 2 ( E a − E b ) 2 − 4 k 2 S 2 , (299) C > k 2 − k 2 S 2 4 , (300) \mathcal{C}>\frac{k}{2}-\frac{k^2S^2}{4}, \tag*{(300)} C > 2 k − 4 k 2 S 2 , (300) C > 3 4 − 9 A 2 64 ( A ≤ 1 ) . (301) \mathcal{C}>\frac{3}{4}-\frac{9A^2}{64}\qquad(A\leq1). \tag*{(301)} C > 4 3 − 64 9 A 2 ( A ≤ 1 ) . (301) Whenever the harmonic cap C < C 0 < 3 / 4 \mathcal{C}<C_0<3/4 C < C 0 < 3/4 applies to this same lift,
S > 2 k − 4 C 0 + E / S k . (302) S>\frac{\sqrt{2k-4C_0}+E/S}{k}. \tag*{(302)} S > k 2 k − 4 C 0 + E / S . (302) For the value C 0 = 29025933167887355473205403 39478417604337498218323684 C_0=\frac{29025933167887355473205403}{39478417604337498218323684} C 0 = 39478417604337498218323684 29025933167887355473205403 , it also follows that S > 81 / 500 S>81/500 S > 81/500 .
Proof. Apply Lemma B.6 with ( t , r ) = ( α , ξ ) (t,r)=(\alpha,\xi) ( t , r ) = ( α , ξ ) and ( β , ζ ) (\beta,\zeta) ( β , ζ ) . It gives E a , E b > 0 E_a,E_b>0 E a , E b > 0 and f s ( 0 ) > 0 > f s ( S ) f_s(0)>0>f_s(S) f s ( 0 ) > 0 > f s ( S ) on the same lift. The maximizing gap g = m + 1 / 2 − f ≥ 0 g=m+1/2-f\geq0 g = m + 1/2 − f ≥ 0 vanishes at the actual exterior ties, so its inward jets imply m s ( 0 ) ≥ f s ( 0 ) > 0 > f s ( S ) ≥ m s ( S ) m_s(0)\geq f_s(0)>0>f_s(S)\geq m_s(S) m s ( 0 ) ≥ f s ( 0 ) > 0 > f s ( S ) ≥ m s ( S ) . Thus L ( 0 ) < 0 < L ( S ) L(0)<0<L(S) L ( 0 ) < 0 < L ( S ) . The unique zero of L L L is an actual interior point, the apex of the concave quadratic m m m . Its interpolation is
m ( s ) = ( 1 − s / S ) ( − 1 / 2 − E a ) + ( s / S ) ( − 1 / 2 − E b ) + k s ( S − s ) . m(s)=(1-s/S)(-1/2-E_a)+(s/S)(-1/2-E_b)+ks(S-s). m ( s ) = ( 1 − s / S ) ( − 1/2 − E a ) + ( s / S ) ( − 1/2 − E b ) + k s ( S − s ) .
The endpoint derivatives give (B.8); evaluating − L 2 − k m -L^2-km − L 2 − k m at an endpoint gives (B.9). At the actual apex, m < − 1 / 2 + k S 2 / 4 m<-1/2+kS^2/4 m < − 1/2 + k S 2 /4 , proving the first strict lower bound. Using S < A / 2 S<A/2 S < A /2 and the fact that k / 2 − k 2 A 2 / 16 k/2-k^2A^2/16 k /2 − k 2 A 2 /16 increases on 3 / 2 ≤ k ≤ 2 3/2\le k\le2 3/2 ≤ k ≤ 2 , A ≤ 1 A\le1 A ≤ 1 , proves the second.
If C < C 0 < k / 2 \mathcal C<C_0<k/2 C < C 0 < k /2 , equations (B.8) and (B.9) first imply E < k S 2 E<kS^2 E < k S 2 . Thus k S − E / S > 0 kS-E/S>0 k S − E / S > 0 . The exact identity
C − { k 2 − 1 4 ( k S − E / S ) 2 } = E a E b S 2 > 0 \mathcal C-\left\{\frac{k}{2}-\frac{1}{4}(kS-E/S)^2\right\}=\frac{E_aE_b}{S^2}>0 C − { 2 k − 4 1 ( k S − E / S ) 2 } = S 2 E a E b > 0
proves (B.12) on its positive square-root branch. If S ≤ 81 / 500 S\le81/500 S ≤ 81/500 , the unconditional bound (B.10), increasing in k k k on this range, gives
C > 3 4 − 9 16 ( 81 / 500 ) 2 > C 0 . \mathcal C>\frac{3}{4}-\frac{9}{16}(81/500)^2>C_0. C > 4 3 − 16 9 ( 81/500 ) 2 > C 0 .
The last inequality is an exact rational comparison for C 0 = 29025933167887355473205403 39478417604337498218323684 C_0=\dfrac{29025933167887355473205403}{39478417604337498218323684} C 0 = 39478417604337498218323684 29025933167887355473205403 . □
Lemma B.8 (Leaf correlations and endpoint densities). For an actual four-core,
ξ 2 − ξ ζ + ζ 2 > 3 h 2 , ζ 2 + ζ z + z 2 > 3 h 2 , (303) \xi^2-\xi\zeta+\zeta^2>3h^2,\qquad\zeta^2+\zeta z+z^2>3h^2, \tag*{(303)} ξ 2 − ξ ζ + ζ 2 > 3 h 2 , ζ 2 + ζ z + z 2 > 3 h 2 , (303) d > 3 h , α β > 2 h 2 , ( α − β ) 2 > 4 h 2 − d 2 / 3 , M > h 2 S , (304) d>3h,\qquad\alpha\beta>2h^2,\qquad(\alpha-\beta)^2>4h^2-d^2/3,\qquad M>h^2S, \tag*{(304)} d > 3 h , α β > 2 h 2 , ( α − β ) 2 > 4 h 2 − d 2 /3 , M > h 2 S , (304) q ( − α ) S ≥ 1 − β d ( 1 + α ) , q ( β ) S ≥ 1 − α d ( 1 + β ) . (305) \frac{q(-\alpha)}{S}\ge\frac{1-\beta}{d(1+\alpha)},\qquad\frac{q(\beta)}{S}\ge\frac{1-\alpha}{d(1+\beta)}. \tag*{(305)} S q ( − α ) ≥ d ( 1 + α ) 1 − β , S q ( β ) ≥ d ( 1 + β ) 1 − α . (305) Moreover, for every 0 < h < 1 0<h<1 0 < h < 1 , one has
ξ , ζ < ρ ( h ) : = 8 h 4 − 3 h . (306) \xi,\zeta<\rho(h):=\frac{8h}{4-3h}. \tag*{(306)} ξ , ζ < ρ ( h ) := 4 − 3 h 8 h . (306)
Proof. The two exact middle-leaf equations are
∫ − ξ z u 2 − h 2 ( 1 − u ) 2 d u = 0 , ∫ z ζ u 2 − h 2 ( 1 + u ) 2 d u = 0. \int_{-\xi}^{z}\frac{u^2-h^2}{(1-u)^2}\,du=0,\qquad\int_{z}^{\zeta}\frac{u^2-h^2}{(1+u)^2}\,du=0. ∫ − ξ z ( 1 − u ) 2 u 2 − h 2 d u = 0 , ∫ z ζ ( 1 + u ) 2 u 2 − h 2 d u = 0.
Subtract the comparison weight ( 1 + h ) − 2 (1+h)^{-2} ( 1 + h ) − 2 . The differences are
( u − h ) ( u + h ) 2 ( 2 + h − u ) ( 1 − u ) 2 ( 1 + h ) 2 , − ( u − h ) 2 ( u + h ) ( 2 + h + u ) ( 1 + u ) 2 ( 1 + h ) 2 , \frac{(u-h)(u+h)^2(2+h-u)}{(1-u)^2(1+h)^2},\qquad
-\frac{(u-h)^2(u+h)(2+h+u)}{(1+u)^2(1+h)^2}, ( 1 − u ) 2 ( 1 + h ) 2 ( u − h ) ( u + h ) 2 ( 2 + h − u ) , − ( 1 + u ) 2 ( 1 + h ) 2 ( u − h ) 2 ( u + h ) ( 2 + h + u ) ,
respectively. Their integrals are strictly negative, since ∣ z ∣ ≤ h |z|\le h ∣ z ∣ ≤ h and both exterior portions have positive length. The unweighted integrals are therefore positive, which is (B.13). Set T = 12 h 2 − 3 z 2 T=\sqrt{12h^2-3z^2} T = 12 h 2 − 3 z 2 , a 0 = ( T + z ) / 2 a_0=(T+z)/2 a 0 = ( T + z ) /2 , b 0 = ( T − z ) / 2 b_0=(T-z)/2 b 0 = ( T − z ) /2 . The positive roots give α > a 0 \alpha>a_0 α > a 0 , β > b 0 \beta>b_0 β > b 0 ; T ≥ 3 ∣ z ∣ T\ge3|z| T ≥ 3∣ z ∣ , a 0 b 0 = 3 h 2 − z 2 ≥ 2 h 2 a_0b_0=3h^2-z^2\ge2h^2 a 0 b 0 = 3 h 2 − z 2 ≥ 2 h 2 , and a 0 2 − a 0 b 0 + b 0 2 = 3 h 2 a_0^2-a_0b_0+b_0^2=3h^2 a 0 2 − a 0 b 0 + b 0 2 = 3 h 2 . Writing α = a 0 + x \alpha=a_0+x α = a 0 + x , β = b 0 + y \beta=b_0+y β = b 0 + y , with x , y > 0 x,y>0 x , y > 0 , gives
α 2 − α β + β 2 = 3 h 2 + ( 2 a 0 − b 0 ) x + ( 2 b 0 − a 0 ) y + x 2 − x y + y 2 > 3 h 2 . \alpha^2-\alpha\beta+\beta^2=3h^2+(2a_0-b_0)x+(2b_0-a_0)y+x^2-xy+y^2>3h^2. α 2 − α β + β 2 = 3 h 2 + ( 2 a 0 − b 0 ) x + ( 2 b 0 − a 0 ) y + x 2 − x y + y 2 > 3 h 2 .
These identities prove the first three consequences. Each middle arc has mean u 2 = h 2 u^2=h^2 u 2 = h 2 , and the two positive exterior arcs have u 2 > h 2 u^2>h^2 u 2 > h 2 , proving M > h 2 S M>h^2S M > h 2 S .
On both active modes − 2 / ( 1 + u ) ≤ ( log q ) u ≤ 2 / ( 1 − u ) -2/(1+u)\le(\log q)_u\le2/(1-u) − 2/ ( 1 + u ) ≤ ( log q ) u ≤ 2/ ( 1 − u ) . The upper comparison from − α -\alpha − α gives q ( u ) ≤ q ( − α ) ( 1 + α ) 2 / ( 1 − u ) 2 q(u)\le q(-\alpha)(1+\alpha)^2/(1-u)^2 q ( u ) ≤ q ( − α ) ( 1 + α ) 2 / ( 1 − u ) 2 . Integrating over the entire core proves the first inequality in (B.15); integration backwards proves the second.
For the radius bound, the actual roots satisfy h < ξ , ζ < 1 h<\xi,\zeta<1 h < ξ , ζ < 1 . After the change of variable u ↦ − u u\mapsto-u u ↦ − u on the mode-2 leaf, both middle-mean identities have the form ∫ z r ( h 2 − u 2 ) ( 1 + u ) − 2 d u = 0 \int_z^r(h^2-u^2)(1+u)^{-2}\,du=0 ∫ z r ( h 2 − u 2 ) ( 1 + u ) − 2 d u = 0 , with ∣ z ∣ ≤ h |z|\le h ∣ z ∣ ≤ h and r > h r>h r > h . Thus J h ( r ) : = ∫ − h r ( h 2 − u 2 ) ( 1 + u ) − 2 d u ≥ 0 J_h(r):=\int_{-h}^r(h^2-u^2)(1+u)^{-2}\,du\ge0 J h ( r ) := ∫ − h r ( h 2 − u 2 ) ( 1 + u ) − 2 d u ≥ 0 , and J h J_h J h strictly decreases for r > h r>h r > h . If h ≥ 4 / 11 h\ge4/11 h ≥ 4/11 , then ρ ( h ) ≥ 1 > r \rho(h)\ge1>r ρ ( h ) ≥ 1 > r already. For 0 < h < 4 / 11 0<h<4/11 0 < h < 4/11 , use the exact primitive P h ( u ) = − u + 2 log ( 1 + u ) + ( 1 − h 2 ) / ( 1 + u ) P_h(u)=-u+2\log(1+u)+(1-h^2)/(1+u) P h ( u ) = − u + 2 log ( 1 + u ) + ( 1 − h 2 ) / ( 1 + u ) . Since J h ( ρ ( h ) ) = P h ( ρ ( h ) ) − P h ( − h ) J_h(\rho(h))=P_h(\rho(h))-P_h(-h) J h ( ρ ( h )) = P h ( ρ ( h )) − P h ( − h ) vanishes at h = 0 h=0 h = 0 and
d d h J h ( ρ ( h ) ) = 54 h 4 ( h − 4 ) ( 5 h − 4 ) ( h − 1 ) ( 3 h − 4 ) 2 ( 5 h + 4 ) 2 < 0 , \frac{d}{dh}J_h(\rho(h))=\frac{54h^4(h-4)(5h-4)}{(h-1)(3h-4)^2(5h+4)^2}<0, d h d J h ( ρ ( h )) = ( h − 1 ) ( 3 h − 4 ) 2 ( 5 h + 4 ) 2 54 h 4 ( h − 4 ) ( 5 h − 4 ) < 0 ,
we have J h ( ρ ( h ) ) < 0 ≤ J h ( r ) J_h(\rho(h)) < 0 \leq J_h(r) J h ( ρ ( h )) < 0 ≤ J h ( r ) , whence r < ρ ( h ) r < \rho(h) r < ρ ( h ) . This proves both bounds without reflecting or renormalizing the lift.
Corollary B.9 (Automatic calibration of a capped four-core). Suppose a genuine fundamental interval of an original global minimizer contains the complete positive context 102021, or a lawful whole cyclic or reflected-time image. Every selected nonzero full maximizing free-time lift is normal. After joint normalization, its determinant satisfies 0 < C < H ∗ 0 < \mathcal{C} < H_* 0 < C < H ∗ , before any improved local cost bound is used.
Proof. Lemma B.5 proves normality and gives 3 / 2 < k < 2 3/2 < k < 2 3/2 < k < 2 , S < A / 2 S < A/2 S < A /2 . The positive exterior arcs make the four-core cost a strict part of the period cost, so A < C ∗ < 2 A < C_* < 2 A < C ∗ < 2 by (A.4). Its actual apex (B.10) therefore gives
C > k 2 − k 2 S 2 4 > k 2 ( 1 − k A 2 8 ) > 0. \mathcal{C} > \frac{k}{2} - \frac{k^2 S^2}{4} > \frac{k}{2}\left(1-\frac{kA^2}{8}\right)>0. C > 2 k − 4 k 2 S 2 > 2 k ( 1 − 8 k A 2 ) > 0. Apply Corollary A.13 with this actual k > 0 k > 0 k > 0 . All caps and both original endpoints remain on the chosen interval; this argument neither transfers a costate nor calibrates a larger union.
Lemma B.10 (The signed internal root of a return). On an actual 020 triple in a normal lift, the first internal switch has u < 0 u < 0 u < 0 . On an actual 202 triple, the second internal switch has u > 0 u > 0 u > 0 . The triples need not share their binary constant k k k with any other core.
Proof. Normalize the complete lift once. The middle arc has two distinct roots of χ \chi χ , so its own μ = 2 k − 3 \mu= 2k - 3 μ = 2 k − 3 is the strictly positive weighted mean of u 2 u^2 u 2 . For 020, weak jets give r 2 ≥ μ r^2 \geq\mu r 2 ≥ μ and s 2 ≤ μ s^2 \leq\mu s 2 ≤ μ for its two ordered switches r < s r < s r < s , hence r < 0 r < 0 r < 0 . Equality r = − μ r = -\sqrt{\mu} r = − μ would give a strictly positive middle integral of μ − u 2 \mu-u^2 μ − u 2 , contradicting its zero value. For 202 the two inequalities are reversed and imply s > μ > 0 s > \sqrt{\mu} > 0 s > μ > 0 by the same argument.
A signed cap strip for nonpositive-entry three-cores.
Lemma B.11 (Signed strip). Consider an actual positive capped context 12021, charging its complete 202 core, whose canonical entrance is u 0 = − a ≤ 0 u_0=-a\leq0 u 0 = − a ≤ 0 . The selected complete lift is normal. Normalize it jointly and write its core coordinates as − a < r < s < b -a<r<s<b − a < r < s < b , its shear as σ \sigma σ , and μ = 2 k − 3 \mu=2k-3 μ = 2 k − 3 . Then
0 < μ < b 2 < 1 , r 2 ≤ μ , s > μ > 0 , k = 3 + μ 2 > 0. 0 < \mu< b^2 < 1,\qquad r^2 \leq\mu,\qquad s > \sqrt{\mu} > 0,\qquad k=\frac{3+\mu}{2}>0. 0 < μ < b 2 < 1 , r 2 ≤ μ , s > μ > 0 , k = 2 3 + μ > 0. Let f f f be the cap function of Lemma B.2 and τ = m + 1 / 2 \tau=m+1/2 τ = m + 1/2 . The initial value f 0 f_0 f 0 is allowed to have either sign; the terminal value satisfies f 3 < 0 f_3<0 f 3 < 0 . The actual endpoint jets give
f s ( 0 ) > 0 > f s ( σ ) , τ s ( 0 ) > 0 > τ s ( σ ) . f_s(0)>0>f_s(\sigma),\qquad\tau_s(0)>0>\tau_s(\sigma). f s ( 0 ) > 0 > f s ( σ ) , τ s ( 0 ) > 0 > τ s ( σ ) . In the convention
X = f 0 σ (signed) , Y = − f 3 σ > 0 , Z = k σ > 0 , Q = X + Y , X=\frac{f_0}{\sigma}\quad\text{(signed)},\qquad Y=-\frac{f_3}{\sigma}>0,\qquad Z=k\sigma>0,\qquad Q=X+Y, X = σ f 0 (signed) , Y = − σ f 3 > 0 , Z = k σ > 0 , Q = X + Y , one has
− Z < Q < Z , C = 3 4 − 1 4 L , L = ( Q + Z ) 2 − 4 Z Y − μ = ( Q − Z ) 2 + 4 Z X − μ . (307) -Z<Q<Z,\qquad\mathcal{C}=\frac{3}{4}-\frac{1}{4}\mathcal{L},\qquad\mathcal{L}=(Q+Z)^2-4ZY-\mu=(Q-Z)^2+4ZX-\mu. \tag*{(307)} − Z < Q < Z , C = 4 3 − 4 1 L , L = ( Q + Z ) 2 − 4 Z Y − μ = ( Q − Z ) 2 + 4 Z X − μ . (307) No return-circle hypothesis is used. For the actual physical cost T T T ,
2 T = w + 3 σ + ∫ u 2 d s , w = a + b , T > 2 σ . (308) 2T=w+3\sigma+\int u^2\,ds,\qquad w=a+b,\qquad T>2\sigma. \tag*{(308)} 2 T = w + 3 σ + ∫ u 2 d s , w = a + b , T > 2 σ . (308) If additionally w < 1 w<1 w < 1 , then
X ≤ U ( μ ) : = ( μ − a 2 ) ( 1 + b ) 2 ( 1 − a ) w + μ 1 − a 2 . (309) X\leq U(\mu):=\frac{(\mu-a^2)(1+b)}{2(1-a)w}+\frac{\mu}{1-a^2}. \tag*{(309)} X ≤ U ( μ ) := 2 ( 1 − a ) w ( μ − a 2 ) ( 1 + b ) + 1 − a 2 μ . (309)
Proof. Lemma B.3 proves normality from the two internal roots and actual exterior caps. Normalize the complete lift jointly.
The middle-arc mean and weak jets give μ > 0 \mu> 0 μ > 0 and r 2 ≤ μ r^2 \le\mu r 2 ≤ μ . The strict root calculation in Lemma B.10 gives s > μ > 0 s > \sqrt{\mu} > 0 s > μ > 0 , hence μ < b 2 < 1 \mu< b^2 < 1 μ < b 2 < 1 .
Full maximization gives g = τ − f ≥ 0 g=\tau-f\ge0 g = τ − f ≥ 0 on the core, with zero values at the actual exterior caps. Therefore τ s ( 0 ) ≥ f s ( 0 ) \tau_s(0)\ge f_s(0) τ s ( 0 ) ≥ f s ( 0 ) and τ s ( σ ) ≤ f s ( σ ) \tau_s(\sigma)\le f_s(\sigma) τ s ( σ ) ≤ f s ( σ ) . On mode 2, differentiation of the exact expression for f f f gives
f s = − u − f q ( u − 1 ) + ( μ − u 2 ) q ( u − 1 ) 2 . f_s=-u-\frac{f}{q(u-1)}+\frac{(\mu-u^2)q(u-1)}{2}. f s = − u − q ( u − 1 ) f + 2 ( μ − u 2 ) q ( u − 1 ) .
At u = − a u=-a u = − a , put H = 1 + a H=1+a H = 1 + a , B = ( μ − a 2 ) / 2 B=(\mu-a^2)/2 B = ( μ − a 2 ) /2 . Substitution of the actual f 0 f_0 f 0 gives
f s ( 0 ) = a + B ( H − 1 − q 0 H ) − 1 2 ( 1 + ( q 0 H 2 ) − 1 ) χ 0 . (310) f_s(0)=a+B(H^{-1}-q_0H)-\frac{1}{2}\left(1+(q_0H^2)^{-1}\right)\chi_0. \tag*{(310)} f s ( 0 ) = a + B ( H − 1 − q 0 H ) − 2 1 ( 1 + ( q 0 H 2 ) − 1 ) χ 0 . (310)
Here χ 0 ≤ 0 \chi_0\le0 χ 0 ≤ 0 . The remaining expression is affine in 0 < q 0 < ( 1 − a 2 ) − 1 0<q_0<(1-a^2)^{-1} 0 < q 0 < ( 1 − a 2 ) − 1 ; its two endpoint values are
a + B 1 + a = a 2 + 2 a + μ 2 ( 1 + a ) > 0 , a ( 1 − μ ) 1 − a 2 ≥ 0. a+\frac{B}{1+a}=\frac{a^2+2a+\mu}{2(1+a)}>0,\qquad
\frac{a(1-\mu)}{1-a^2}\ge0. a + 1 + a B = 2 ( 1 + a ) a 2 + 2 a + μ > 0 , 1 − a 2 a ( 1 − μ ) ≥ 0.
Thus f s ( 0 ) > 0 f_s(0)>0 f s ( 0 ) > 0 , without a sign assumption on f 0 f_0 f 0 or a split at a = 0 a=0 a = 0 .
Apply Lemma B.6 to the final leaf with ( t , r , h ) = ( b , s , μ ) (t,r,h)=(b,s,\sqrt{\mu}) ( t , r , h ) = ( b , s , μ ) . It gives ( μ − b 2 ) q ( b ) / 2 < f 3 < 0 (\mu-b^2)q(b)/2<f_3<0 ( μ − b 2 ) q ( b ) /2 < f 3 < 0 and f s ( σ ) < − b + ( b 2 − μ ) / [ 2 ( 1 + b ) ] < 0 f_s(\sigma)<-b+(b^2-\mu)/[2(1+b)]<0 f s ( σ ) < − b + ( b 2 − μ ) / [ 2 ( 1 + b )] < 0 . The strict matrix jets follow from the actual endpoint gaps. Since τ s s = − 2 k \tau_{ss}=-2k τ ss = − 2 k , its apex is an actual interior point.
The quadratic with endpoint values τ ( 0 ) = f 0 \tau(0)=f_0 τ ( 0 ) = f 0 , τ ( σ ) = f 3 \tau(\sigma)=f_3 τ ( σ ) = f 3 has derivatives Z − Q Z-Q Z − Q and − Z − Q -Z-Q − Z − Q at its ends. The strict matrix jets give − Z < Q < Z -Z<Q<Z − Z < Q < Z . Substituting these endpoint values in − L 2 − k m -L^2-km − L 2 − k m gives both equal expressions in (B.17). Notice that X X X retains its signed initial-cap convention. The physical cost identity follows from v 2 d s = d u v^2\,\mathrm{d}s=\mathrm{d}u v 2 d s = d u , and T > 2 σ T>2\sigma T > 2 σ from the open state domain.
Finally the lower logarithmic derivative bound ( log q ) u ≥ − 2 / ( 1 + u ) (\log q)_u\ge-2/(1+u) ( log q ) u ≥ − 2/ ( 1 + u ) gives
σ ≥ q 0 ( 1 − a ) w 1 + b , q 0 2 σ ≤ 1 + b 2 ( 1 − a ) w . \sigma\ge\frac{q_0(1-a)w}{1+b},\qquad
\frac{q_0}{2\sigma}\le\frac{1+b}{2(1-a)w}. σ ≥ 1 + b q 0 ( 1 − a ) w , 2 σ q 0 ≤ 2 ( 1 − a ) w 1 + b .
The integral to the actual first root gives ∣ χ 0 ∣ ≤ μ σ |\chi_0|\le\mu\sigma ∣ χ 0 ∣ ≤ μ σ , and Ω \Omega Ω gives ∣ A − ( − a ) ∣ < ( 1 − a 2 ) − 1 |A_-(-a)|<(1-a^2)^{-1} ∣ A − ( − a ) ∣ < ( 1 − a 2 ) − 1 . Insert these bounds in f 0 = ( μ − a 2 ) q 0 / 2 + A − χ 0 f_0=(\mu-a^2)q_0/2+A_-\chi_0 f 0 = ( μ − a 2 ) q 0 /2 + A − χ 0 , discarding its negative first term only when μ < a 2 \mu<a^2 μ < a 2 . This proves (B.19). □
Lemma B.12 (Permitted signed-loss relaxation). For an actual three-core as above, if X ≤ 0 X\le0 X ≤ 0 , then L < Z 2 \mathcal{L}<Z^2 L < Z 2 . If X ≥ 0 X\ge0 X ≥ 0 , at fixed actual a a a , b b b , Z Z Z one has
L < ( U ( μ ) + Z ) 2 − μ . \mathcal{L}<(U(\mu)+Z)^2-\mu. L < ( U ( μ ) + Z ) 2 − μ . The expression on the right is convex on each of its two affine-U U U pieces in μ \mu μ . Thus its maximum on 0 ≤ μ ≤ M 0\le\mu\le M 0 ≤ μ ≤ M is attained among 0 0 0 , M M M , and the knot a 2 a^2 a 2 when that knot belongs to the interval.
Proof. If X ≤ 0 X\le0 X ≤ 0 , then Y − X ≥ ∣ Q ∣ Y-X\ge|Q| Y − X ≥ ∣ Q ∣ and ∣ Q ∣ < Z |Q|<Z ∣ Q ∣ < Z , so
L = Z 2 + Q 2 − 2 Z ( Y − X ) − μ ≤ Z 2 + Q 2 − 2 Z ∣ Q ∣ − μ < Z 2 . \mathcal{L}=Z^2+Q^2-2Z(Y-X)-\mu\le Z^2+Q^2-2Z|Q|-\mu<Z^2. L = Z 2 + Q 2 − 2 Z ( Y − X ) − μ ≤ Z 2 + Q 2 − 2 Z ∣ Q ∣ − μ < Z 2 . If X ≥ 0 X\ge0 X ≥ 0 , the derivative in Y Y Y of the first loss expression is 2 ( X + Y − Z ) < 0 2(X+Y-Z)<0 2 ( X + Y − Z ) < 0 throughout 0 ≤ Y ≤ Y a c t u a l 0\le Y\le Y_{\mathrm{actual}} 0 ≤ Y ≤ Y actual , by the actual strip. Lowering Y Y Y to zero gives L < ( X + Z ) 2 − μ ≤ ( U ( μ ) + Z ) 2 − μ \mathcal{L}<(X+Z)^2-\mu\le(U(\mu)+Z)^2-\mu L < ( X + Z ) 2 − μ ≤ ( U ( μ ) + Z ) 2 − μ . The last expression has nonnegative second derivative on each affine piece. The comparisons vary a scalar u u u after freezing the actual Z Z Z ; no equation coupling the varying scalar to the actual shear is imposed. Any upper bound for Z Z Z is substituted only after this reduction. □
The generic four-core estimate
Throughout this section put
C 0 = 29025933167887355473205403 39478417604337498218323684 . C_0=\frac{29025933167887355473205403}{39478417604337498218323684}. C 0 = 39478417604337498218323684 29025933167887355473205403 .
Lemma B.13 (A common product envelope for binary cap losses). Let a , b ≥ 0 a,b\ge0 a , b ≥ 0 , 0 < d = a + b ≤ D < 1 0<d=a+b\le D<1 0 < d = a + b ≤ D < 1 , p = a b p=ab p = ab , and 0 ≤ h ≤ 1 / 2 0\le h\le1/2 0 ≤ h ≤ 1/2 . Using the exterior loss e e e of Lemma B.6 , put
Q ( d , h , p ) = ( 1 − b ) e ( a , h ) / ( 1 + a ) + ( 1 − a ) e ( b , h ) / ( 1 + b ) d , Q(d,h,p)=\frac{(1-b)e(a,h)/(1+a)+(1-a)e(b,h)/(1+b)}{d}, Q ( d , h , p ) = d ( 1 − b ) e ( a , h ) / ( 1 + a ) + ( 1 − a ) e ( b , h ) / ( 1 + b ) , Ψ ( d , h ) = − 64 h 3 + ( 12 d 2 + 48 d − 48 ) h 2 − 3 d 4 − 4 d 3 + 12 d 2 12 d ( d + 2 ) 2 . \Psi(d,h)=\frac{-64h^3+(12d^2+48d-48)h^2-3d^4-4d^3+12d^2}{12d(d+2)^2}. Ψ ( d , h ) = 12 d ( d + 2 ) 2 − 64 h 3 + ( 12 d 2 + 48 d − 48 ) h 2 − 3 d 4 − 4 d 3 + 12 d 2 . Define
c ( D ) = ( 2 − D ) 2 12 ( 1 − D ) 2 , t ( D ) = 5 D 3 + 13 D 2 + 8 D − 4 12 ( 1 + D ) 3 , Γ ( D ) = 1 − c ( D ) − max ( t ( D ) , 0 ) . c(D)=\frac{(2-D)^2}{12(1-D)^2},\qquad t(D)=\frac{5D^3+13D^2+8D-4}{12(1+D)^3},\qquad\Gamma(D)=1-c(D)-\max(t(D),0). c ( D ) = 12 ( 1 − D ) 2 ( 2 − D ) 2 , t ( D ) = 12 ( 1 + D ) 3 5 D 3 + 13 D 2 + 8 D − 4 , Γ ( D ) = 1 − c ( D ) − max ( t ( D ) , 0 ) . On the whole scalar interval 0 ≤ p ≤ d 2 / 4 0\le p\le d^2/4 0 ≤ p ≤ d 2 /4 ,
∂ p Q ≤ − Γ ( D ) , Q ≥ Ψ ( d , h ) + Γ ( D ) d ( d 2 / 4 − p ) . (311) \partial_p Q\le-\Gamma(D),\qquad Q\ge\Psi(d,h)+\frac{\Gamma(D)}{d}(d^2/4-p). \tag*{(311)} ∂ p Q ≤ − Γ ( D ) , Q ≥ Ψ ( d , h ) + d Γ ( D ) ( d 2 /4 − p ) . (311) In particular
Table 18.
No root, mean or physical-source condition is imposed on the comparison products.
Proof. Set A h = ( 1 − h ) 2 ( 1 + 2 h ) A_h=(1-h)^2(1+2h) A h = ( 1 − h ) 2 ( 1 + 2 h ) and B h = ( 1 + h ) 2 ( 1 − 2 h ) B_h=(1+h)^2(1-2h) B h = ( 1 + h ) 2 ( 1 − 2 h ) ; both lie in [ 0 , 1 ] [0,1] [ 0 , 1 ] . Direct symmetric reduction, with l = 1 + d + p l=1+d+p l = 1 + d + p , gives
∂ p Q = − 1 + A h ( 2 − d ) 2 12 ( 1 − d + p ) 2 + B h T d ( l ) , T d ( l ) = d ( d + 2 ) 2 3 l 3 − 4 + 4 d − d 2 12 l 2 . \partial_p Q=-1+A_h\frac{(2-d)^2}{12(1-d+p)^2}+B_hT_d(l),\qquad T_d(l)=\frac{d(d+2)^2}{3l^3}-\frac{4+4d-d^2}{12l^2}. ∂ p Q = − 1 + A h 12 ( 1 − d + p ) 2 ( 2 − d ) 2 + B h T d ( l ) , T d ( l ) = 3 l 3 d ( d + 2 ) 2 − 12 l 2 4 + 4 d − d 2 . The first positive fraction is at most c ( d ) c(d) c ( d ) . Since
T d ′ ( l ) = ( 4 + 4 d − d 2 ) l − 6 d ( d + 2 ) 2 6 l 4 , T'_d(l)=\frac{(4+4d-d^2)l-6d(d+2)^2}{6l^4}, T d ′ ( l ) = 6 l 4 ( 4 + 4 d − d 2 ) l − 6 d ( d + 2 ) 2 , T d T_d T d has no interior maximum on l ≥ 1 + d l\ge1+d l ≥ 1 + d and tends to zero there; hence T d ( l ) ≤ max ( T d ( 1 + d ) , 0 ) = max ( t ( d ) , 0 ) T_d(l)\le\max(T_d(1+d),0)=\max(t(d),0) T d ( l ) ≤ max ( T d ( 1 + d ) , 0 ) = max ( t ( d ) , 0 ) . Both c c c and t t t increase on ( 0 , 1 ) (0,1) ( 0 , 1 ) , with derivatives ( 2 − d ) / ( 6 ( 1 − d ) 3 ) (2-d)/(6(1-d)^3) ( 2 − d ) / ( 6 ( 1 − d ) 3 ) and ( d 2 + 5 d + 10 ) / ( 6 ( 1 + d ) 4 ) (d^2+5d+10)/(6(1+d)^4) ( d 2 + 5 d + 10 ) / ( 6 ( 1 + d ) 4 ) . Thus the derivative bound follows even when T d T_d T d is negative. Finally Q ( d , h , d 2 / 4 ) = Ψ ( d , h ) Q(d,h,d^2/4)=\Psi(d,h) Q ( d , h , d 2 /4 ) = Ψ ( d , h ) ; integrate the derivative bound to this balanced product. The table is exact rational substitution. □
Theorem B.14. In each actual positive context 102021, 210102, or 021210, the physical cost of the four complete arcs in positions 1, 2, 3, 4 is strictly larger than 389/1000. The exterior arcs are uncharged.
Proof. It suffices to use the canonical context and the notation of Section B.1.1. Suppose A ≤ B = 389 / 1000 A\le B=389/1000 A ≤ B = 389/1000 . Corollary B.9 gives 0 < C < C 0 0<\mathcal{C}<C_0 0 < C < C 0 on this same complete lift. Lemma B.7 gives
S > 81 / 500 , d < 2 B − 3 ( 81 / 500 ) = 73 / 250 < 3 / 10 , 0 < h 2 < d 2 / 9 < 1 / 100. (312) S>81/500,\qquad d<2B-3(81/500)=73/250<3/10,\qquad0<h^2<d^2/9<1/100. \tag*{(312)} S > 81/500 , d < 2 B − 3 ( 81/500 ) = 73/250 < 3/10 , 0 < h 2 < d 2 /9 < 1/100. (312)
A correlated lower bound for the cap losses
The positive exterior losses and (B.15) give
E / S > Q ( d , h , α β ) . (313) E/S > Q(d,h,\alpha\beta). \tag*{(313)} E / S > Q ( d , h , α β ) . (313)
Apply Lemma B.13 with D = 3 / 10 D=3/10 D = 3/10 , and retain the actual leaf correlation ( α − β ) 2 > 4 h 2 − d 2 / 3 (\alpha-\beta)^2 > 4h^2-d^2/3 ( α − β ) 2 > 4 h 2 − d 2 /3 . Since d 2 / 4 − α β = ( α − β ) 2 / 4 d^2/4-\alpha\beta=(\alpha-\beta)^2/4 d 2 /4 − α β = ( α − β ) 2 /4 , this gives, with μ = h 2 \mu=h^2 μ = h 2 and R ( d ) = ( 12 − 4 d − 3 d 2 ) / ( 3 ( d + 2 ) 2 ) R(d)=(12-4d-3d^2)/(3(d+2)^2) R ( d ) = ( 12 − 4 d − 3 d 2 ) / ( 3 ( d + 2 ) 2 ) ,
E S > R ( d ) ( d / 4 − μ / d ) + max ( 0 , 4 μ − d 2 / 3 ) 8 d + 8 μ ( 1 − 2 μ / d ) 3 ( d + 2 ) 2 . (314) \frac{E}{S} > R(d)(d/4-\mu/d)+\frac{\max(0,4\mu-d^2/3)}{8d}+\frac{8\mu(1-2\sqrt{\mu}/d)}{3(d+2)^2}. \tag*{(314)} S E > R ( d ) ( d /4 − μ / d ) + 8 d max ( 0 , 4 μ − d 2 /3 ) + 3 ( d + 2 ) 2 8 μ ( 1 − 2 μ / d ) . (314)
The comparison keeps the actual common d , μ d,\mu d , μ throughout.
Minimizing the shear envelope
For 0 ≤ μ ≤ d 2 / 12 0\leq\mu\leq d^2/12 0 ≤ μ ≤ d 2 /12 use μ / d ≤ 1 / 12 < 3 / 10 \sqrt{\mu}/d\leq1/\sqrt{12}<3/10 μ / d ≤ 1/ 12 < 3/10 in the last term of (B.24); it is at least 16 μ / [ 15 ( d + 2 ) 2 ] 16\mu/[15(d+2)^2] 16 μ / [ 15 ( d + 2 ) 2 ] . For d 2 / 12 ≤ μ ≤ d 2 / 9 d^2/12\leq\mu\leq d^2/9 d 2 /12 ≤ μ ≤ d 2 /9 , the function μ ( 1 − 2 μ / d ) \mu(1-2\sqrt{\mu}/d) μ ( 1 − 2 μ / d ) increases, and the last term is at least T ( d ) = 4 d 2 / [ 45 ( d + 2 ) 2 ] T(d)=4d^2/[45(d+2)^2] T ( d ) = 4 d 2 / [ 45 ( d + 2 ) 2 ] . Insert these two continuous lower envelopes in (B.12). The derivative with respect to μ \mu μ , multiplied by the positive k 2 k^2 k 2 , is
4 C 0 − k 2 3 + μ − 4 C 0 + H l o ( d ) or 4 C 0 − k 2 3 + μ − 4 C 0 + H h i ( d ) , \frac{4C_0-k}{2\sqrt{3+\mu-4C_0}}+H_{\mathrm{lo}}(d)
\quad\text{or}\quad
\frac{4C_0-k}{2\sqrt{3+\mu-4C_0}}+H_{\mathrm{hi}}(d), 2 3 + μ − 4 C 0 4 C 0 − k + H lo ( d ) or 2 3 + μ − 4 C 0 4 C 0 − k + H hi ( d ) ,
where
H l o ( d ) = 15 d 4 + 20 d 3 + 120 d 2 + 432 d − 720 120 d ( d + 2 ) 2 , H_{\mathrm{lo}}(d)=\frac{15d^4+20d^3+120d^2+432d-720}{120d(d+2)^2}, H lo ( d ) = 120 d ( d + 2 ) 2 15 d 4 + 20 d 3 + 120 d 2 + 432 d − 720 ,
H h i ( d ) = 105 d 4 + 148 d 3 + 1320 d 2 + 3600 d − 2160 720 d ( d + 2 ) 2 . H_{\mathrm{hi}}(d)=\frac{105d^4+148d^3+1320d^2+3600d-2160}{720d(d+2)^2}. H hi ( d ) = 720 d ( d + 2 ) 2 105 d 4 + 148 d 3 + 1320 d 2 + 3600 d − 2160 .
Both functions increase on 0 < d < 3 / 10 0<d<3/10 0 < d < 3/10 . After multiplying by positive factors their derivative numerators are
15 d 5 + 90 d 4 − 40 d 3 − 624 d 2 + 2160 d + 1440 , 105 d 5 + 630 d 4 − 728 d 3 − 4560 d 2 + 6480 d + 4320 ; 15d^5+90d^4-40d^3-624d^2+2160d+1440,\quad
105d^5+630d^4-728d^3-4560d^2+6480d+4320; 15 d 5 + 90 d 4 − 40 d 3 − 624 d 2 + 2160 d + 1440 , 105 d 5 + 630 d 4 − 728 d 3 − 4560 d 2 + 6480 d + 4320 ;
positivity follows by dropping the positive nonconstant terms and bounding the negative terms at 3 / 10 3/10 3/10 . The rational bounds
147 / 200 < C 0 < 1471 / 2000 , 243 / 1000 < 3 − 4 C 0 k < 301 / 200 , 3 + μ − 4 C 0 < 27 / 100 147/200<C_0<1471/2000,\qquad243/1000<\sqrt{3}-\frac{4C_0}{k}<301/200,\qquad\sqrt{3+\mu-4C_0}<27/100 147/200 < C 0 < 1471/2000 , 243/1000 < 3 − k 4 C 0 < 301/200 , 3 + μ − 4 C 0 < 27/100
give the two derivative signs
721 243 + H l o ( 3 / 10 ) = 721 243 − 128653 42320 < 0 , 287 108 + H h i ( 9 / 50 ) = 287 108 − 67974937 28514400 > 0. \frac{721}{243}+H_{\mathrm{lo}}(3/10)=\frac{721}{243}-\frac{128653}{42320}<0,\qquad
\frac{287}{108}+H_{\mathrm{hi}}(9/50)=\frac{287}{108}-\frac{67974937}{28514400}>0. 243 721 + H lo ( 3/10 ) = 243 721 − 42320 128653 < 0 , 108 287 + H hi ( 9/50 ) = 108 287 − 28514400 67974937 > 0.
The second sign is used only after d > 9 / 50 d>9/50 d > 9/50 has been proved below. Thus, on that range, the minimum is at μ = d 2 / 12 \mu=d^2/12 μ = d 2 /12 and
S > J ( d ) ≔ 3 + d 2 / 12 − 4 C 0 + R ( d ) d / 6 + T ( d ) ( 3 + d 2 / 12 ) / 2 . (315) S>J(d)\coloneqq\frac{\sqrt{3+d^2/12-4C_0}+R(d)d/6+T(d)}{(3+d^2/12)/2}. \tag*{(315)} S > J ( d ) : = ( 3 + d 2 /12 ) /2 3 + d 2 /12 − 4 C 0 + R ( d ) d /6 + T ( d ) . (315)
The radical quotient increases there. The derivative of the remaining rational term has the sign of 5 d 5 − 2 d 4 − 240 d 3 − 1120 d 2 − 1296 d + 1440 > 0 5d^5-2d^4-240d^3-1120d^2-1296d+1440>0 5 d 5 − 2 d 4 − 240 d 3 − 1120 d 2 − 1296 d + 1440 > 0 on [ 0 , 3 / 10 ] [0,3/10] [ 0 , 3/10 ] , again by the endpoint bound on the negative terms. Hence J J J increases.
The full-profile reserve
The original compact set gives ( u 2 + v 2 − 1 ) / v ≥ 2 / 15 (u^2+v^2-1)/v \ge2/15 ( u 2 + v 2 − 1 ) / v ≥ 2/15 . If d < D < 1 d < D < 1 d < D < 1 and V 2 − 2 V / 15 + D 2 − 1 < 0 V^2-2V/15+D^2-1<0 V 2 − 2 V /15 + D 2 − 1 < 0 , this implies v > V v>V v > V everywhere in the core. Put y = ( u 2 + v 2 − 1 ) / v y=(u^2+v^2-1)/v y = ( u 2 + v 2 − 1 ) / v . On mode 0, y s = v ( 1 + u ) + v 3 / ( 1 + u ) ≥ 2 v 2 y_s=v(1+u)+v^3/(1+u)\ge2v^2 y s = v ( 1 + u ) + v 3 / ( 1 + u ) ≥ 2 v 2 ; on mode 2 the corresponding derivative is at most − 2 v 2 -2v^2 − 2 v 2 . On each of the four positive arcs integrate from the endpoint where y y y is smaller. If its shear length is s i s_i s i , this gives
∫ ( u 2 + v 2 − 1 ) d s > 2 V S / 15 + V 3 ∑ s i 2 ≥ 2 V S / 15 + V 3 S 2 / 4. \int(u^2+v^2-1)\,\mathrm{d}s > 2VS/15+V^3\sum s_i^2 \ge2VS/15+V^3S^2/4. ∫ ( u 2 + v 2 − 1 ) d s > 2 V S /15 + V 3 ∑ s i 2 ≥ 2 V S /15 + V 3 S 2 /4.
Consequently
A > ( 2 + V / 15 ) S + V 3 S 2 / 8. (316) A>(2+V/15)S+V^3S^2/8. \tag*{(316)} A > ( 2 + V /15 ) S + V 3 S 2 /8. (316)
Since q < V − 2 q<V^{-2} q < V − 2 throughout the actual core, direct integration gives
M = ∫ − α β u 2 q ( u ) d u < α 3 + β 3 3 V 2 < d 3 3 V 2 . M=\int_{-\alpha}^{\beta}u^2q(u)\,\mathrm{d}u<\frac{\alpha^3+\beta^3}{3V^2}<\frac{d^3}{3V^2}. M = ∫ − α β u 2 q ( u ) d u < 3 V 2 α 3 + β 3 < 3 V 2 d 3 .
Combining this with d = S − M + ∫ ( u 2 + v 2 − 1 ) d s d=S-M+\int(u^2+v^2-1)\,\mathrm{d}s d = S − M + ∫ ( u 2 + v 2 − 1 ) d s gives
d > ( 1 + 2 V / 15 ) S + V 3 S 2 / 4 − D 3 / ( 3 V 2 ) . (317) d>(1+2V/15)S+V^3S^2/4-D^3/(3V^2). \tag*{(317)} d > ( 1 + 2 V /15 ) S + V 3 S 2 /4 − D 3 / ( 3 V 2 ) . (317)
The following three rational rows finish the proof:
Table 19.
For each row, direct rational arithmetic gives
1 + 2 V / 15 − D 2 − V 2 > 0 , ( 1 + 2 V / 15 ) s + V 3 s 2 / 4 − D 3 / ( 3 V 2 ) > d 0 . 1+2V/15-D^2-V^2>0,\qquad(1+2V/15)s+V^3s^2/4-D^3/(3V^2)>d_0. 1 + 2 V /15 − D 2 − V 2 > 0 , ( 1 + 2 V /15 ) s + V 3 s 2 /4 − D 3 / ( 3 V 2 ) > d 0 .
Writing k 0 = ( 3 + d 0 2 / 12 ) / 2 k_0=(3+d_0^2/12)/2 k 0 = ( 3 + d 0 2 /12 ) /2 and T 0 = k 0 s ′ − R ( d 0 ) d 0 / 6 − T ( d 0 ) T_0=k_0s'-R(d_0)d_0/6-T(d_0) T 0 = k 0 s ′ − R ( d 0 ) d 0 /6 − T ( d 0 ) , one also has T 0 > 0 T_0>0 T 0 > 0 and 3 + d 0 2 / 12 − 4 C 0 − T 0 2 > 0 3+d_0^2/12-4C_0-T_0^2>0 3 + d 0 2 /12 − 4 C 0 − T 0 2 > 0 . These are explicit rational certificates for J ( d 0 ) > s ′ J(d_0)>s' J ( d 0 ) > s ′ . The first row establishes d > 4583 / 25000 > 9 / 50 d>4583/25000>9/50 d > 4583/25000 > 9/50 before the high-μ \mu μ derivative argument is used. Each row therefore proves the next lower bound for S S S without circularity. Finally, at V = 5219 / 5000 V=5219/5000 V = 5219/5000 , s = 18563 / 100000 s=18563/100000 s = 18563/100000 ,
( 2 + V / 15 ) s + V 3 s 2 / 8 − 389 / 1000 = 2274532511176128313 30000000000000000000000 > 0. (2+V/15)s+V^3s^2/8-389/1000
=\frac{2274532511176128313}{30000000000000000000000}>0. ( 2 + V /15 ) s + V 3 s 2 /8 − 389/1000 = 30000000000000000000000 2274532511176128313 > 0.
Equation (B.26) contradicts the closed contrary budget. Legal cyclic transport gives the other two literal contexts. □
Lemma B.15 (A scalar endpoint principle). Let Ψ ( d 0 , h ) = c 0 − a h 2 − b h 3 \Psi(d_0,h)=c_0-ah^2-bh^3 Ψ ( d 0 , h ) = c 0 − a h 2 − b h 3 , where c 0 , a , b > 0 c_0,a,b>0 c 0 , a , b > 0 , and put
g ( k ) = 2 k − 4 C 0 + Ψ ( d 0 , 2 k − 3 ) k . g(k)=\frac{\sqrt{2k-4C_0}+\Psi(d_0,\sqrt{2k-3})}{k}. g ( k ) = k 2 k − 4 C 0 + Ψ ( d 0 , 2 k − 3 ) . On any closed interval where k ≥ 3 / 2 k\ge3/2 k ≥ 3/2 and 2 k − 4 C 0 > 0 2k-4C_0>0 2 k − 4 C 0 > 0 , the minimum of g g g is at an endpoint. The same is true after adding a nonnegative constant to Ψ \Psi Ψ .
Proof. For h = 2 k − 3 > 0 h=\sqrt{2k-3}>0 h = 2 k − 3 > 0 direct differentiation gives
k 2 g ′ ( k ) = 4 C 0 − k 2 k − 4 C 0 − ( c 0 + 3 a ) − b h ( k + 3 ) . k^2g'(k)=\frac{4C_0-k}{\sqrt{2k-4C_0}}-(c_0+3a)-bh(k+3). k 2 g ′ ( k ) = 2 k − 4 C 0 4 C 0 − k − ( c 0 + 3 a ) − bh ( k + 3 ) . The derivative of the right side is − k / ( 2 k − 4 C 0 ) 3 / 2 − b [ ( k + 3 ) / h + h ] < 0 -k/(2k-4C_0)^{3/2}-b[(k+3)/h+h]<0 − k / ( 2 k − 4 C 0 ) 3/2 − b [( k + 3 ) / h + h ] < 0 . Thus g ′ g' g ′ can change sign only from positive to negative, and g g g has no interior minimum. Continuity includes the endpoint k = 3 / 2 k=3/2 k = 3/2 . □
One-return four-core and circle-qualified returned-core estimates
Costs are measured in physical time. A context consists of complete positive maximal arcs on a genuine containing interval. Select one full lift there; its actual capped core proves normality before joint normalization. No binary invariant is continued through a third-mode cap without its lawful full transport. We use
C 0 = 29025933167887355473205403 39478417604337498218323684 . C_0 = \frac{29025933167887355473205403}{39478417604337498218323684}. C 0 = 39478417604337498218323684 29025933167887355473205403 .
The binary identities, endpoint jets, actual four-core apex and leaf correlations are those proved in Lemmas B.2 –B.8.
The one-return four-core cost
Theorem B.16 (Refined one-return estimate). For the actual right contexts 1020212, 2101020, 0212101, charge positions 1, 2, 3, 4; for the actual left contexts 0102021, 1210102, 2021210, charge positions 2, 3, 4, 5. Positions start at zero. Under the complete-source hypotheses above, the charged cost satisfies
A > B E ≔ 4154032208439 8759490110000 . A > B_E \coloneqq\frac{4154032208439}{8759490110000}. A > B E : = 8759490110000 4154032208439 . The two exterior caps and the one returned arc are positive and uncharged.
Proof. We give the right-context proof and then explain the left transport. Put
s ∗ = 983 5000 , r ∗ = 8 47 , x ∗ = 1 + r ∗ , K ∗ = 4 − x ∗ 2 , D ∗ = s ∗ K ∗ x ∗ x ∗ + s ∗ K ∗ = 314171715 875949011 . s_* = \frac{983}{5000}, \qquad r_* = \frac{8}{47}, \qquad x_* = 1+r_*, \qquad K_* = 4-x_*^2, \qquad D_* = \frac{s_*K_*x_*}{x_*+s_*K_*} = \frac{314171715}{875949011}. s ∗ = 5000 983 , r ∗ = 47 8 , x ∗ = 1 + r ∗ , K ∗ = 4 − x ∗ 2 , D ∗ = x ∗ + s ∗ K ∗ s ∗ K ∗ x ∗ = 875949011 314171715 .
Thus 2 B E = D ∗ + 3 s ∗ 2B_E=D_*+3s_* 2 B E = D ∗ + 3 s ∗ exactly. Suppose A ≤ B E < 1 A\le B_E<1 A ≤ B E < 1 . Use the actual variables of the capped four-core: endpoints − α , β -\alpha,\beta − α , β , total shear S S S , d = α + β d=\alpha+\beta d = α + β , and h 2 = 2 k − 3 h^2=2k-3 h 2 = 2 k − 3 . Corollary B.9 gives 0 < C < C 0 0<\mathcal{C}<C_0 0 < C < C 0 on this selected lift. The exact comparison 3 / 4 − ( 9 / 16 ) ( 81 / 500 ) 2 > C 0 3/4-(9/16)(81/500)^2>C_0 3/4 − ( 9/16 ) ( 81/500 ) 2 > C 0 gives S > 81 / 500 S>81/500 S > 81/500 . The physical identity 2 A = d + 3 S + M 2A=d+3S+M 2 A = d + 3 S + M , with M = ∫ u 2 d s > 0 M=\int u^2\,\mathrm{d}s>0 M = ∫ u 2 d s > 0 , gives the fresh initial domain
0 < α , β < d < I = 2025476111709 4379745055000 < 259 500 < 1 , 0 < h < d / 3 , p = α β ≤ d 2 / 4. 0<\alpha,\beta<d<I=\frac{2025476111709}{4379745055000}<\frac{259}{500}<1,\qquad0<h<d/3,\qquad p=\alpha\beta\le d^2/4. 0 < α , β < d < I = 4379745055000 2025476111709 < 500 259 < 1 , 0 < h < d /3 , p = α β ≤ d 2 /4.
This width bound follows directly from the stated contrary budget.
Cap loss and endpoint comparison
Let E a = − f ( − α ) E_a=-f(-\alpha) E a = − f ( − α ) , E b = − f ( β ) E_b=-f(\beta) E b = − f ( β ) and E = E a + E b E=E_a+E_b E = E a + E b . Lemma B.6 gives the positive endpoint losses e ( α , h ) , e ( β , h ) e(\alpha,h),e(\beta,h) e ( α , h ) , e ( β , h ) on the fresh domain. Thus (B.15) gives
E S > Q = ( 1 − β ) e ( α , h ) / ( 1 + α ) + ( 1 − α ) e ( β , h ) / ( 1 + β ) d . (318) \frac{E}{S}>Q=\frac{(1-\beta)e(\alpha,h)/(1+\alpha)+(1-\alpha)e(\beta,h)/(1+\beta)}{d}. \tag*{(318)} S E > Q = d ( 1 − β ) e ( α , h ) / ( 1 + α ) + ( 1 − α ) e ( β , h ) / ( 1 + β ) . (318)
Lemma B.13 , with D = 13 / 25 > 259 / 500 D=13/25>259/500 D = 13/25 > 259/500 , gives E / S > Ψ ( d , h ) E/S>\Psi(d,h) E / S > Ψ ( d , h ) throughout the fresh domain. Direct differentiation gives
12 d 2 ( d + 2 ) 3 Ψ d = d 2 ( 24 − 28 d − 18 d 2 − 3 d 3 ) + 12 h 2 ( 8 + 12 d − 6 d 2 − d 3 ) + 64 h 3 ( 2 + 3 d ) > 0 12d^2(d+2)^3\Psi_d=d^2(24-28d-18d^2-3d^3)+12h^2(8+12d-6d^2-d^3)+64h^3(2+3d)>0 12 d 2 ( d + 2 ) 3 Ψ d = d 2 ( 24 − 28 d − 18 d 2 − 3 d 3 ) + 12 h 2 ( 8 + 12 d − 6 d 2 − d 3 ) + 64 h 3 ( 2 + 3 d ) > 0
on the indicated box. For fixed d 0 d_0 d 0 write Ψ ( d 0 , h ) = c 0 − a 0 h 2 − b 0 h 3 \Psi(d_0,h)=c_0-a_0h^2-b_0h^3 Ψ ( d 0 , h ) = c 0 − a 0 h 2 − b 0 h 3 , where
c 0 = d 0 ( 12 − 4 d 0 − 3 d 0 2 ) 12 ( d 0 + 2 ) 2 , a 0 = 4 − 4 d 0 − d 0 2 d 0 ( d 0 + 2 ) 2 , b 0 = 16 3 d 0 ( d 0 + 2 ) 2 > 0. c_0=\frac{d_0(12-4d_0-3d_0^2)}{12(d_0+2)^2},\qquad a_0=\frac{4-4d_0-d_0^2}{d_0(d_0+2)^2},\qquad b_0=\frac{16}{3d_0(d_0+2)^2}>0. c 0 = 12 ( d 0 + 2 ) 2 d 0 ( 12 − 4 d 0 − 3 d 0 2 ) , a 0 = d 0 ( d 0 + 2 ) 2 4 − 4 d 0 − d 0 2 , b 0 = 3 d 0 ( d 0 + 2 ) 2 16 > 0.
All three constants are positive at the widths below. Lemma B.15 therefore puts the minimum of 2 { 3 + h 2 − 4 C 0 + Ψ ( d 0 , h ) } / ( 3 + h 2 ) 2\{\sqrt{3}+h^2-4C_0+\Psi(d_0,h)\}/(3+h^2) 2 { 3 + h 2 − 4 C 0 + Ψ ( d 0 , h )} / ( 3 + h 2 ) at a height endpoint. This is a scalar comparison, not a second physical core.
One actual return and the fresh bootstrap.
The final actual 212 transforms to 020. Its own first root is negative by Lemma B.10, giving ( 1 + β ) 2 + v ( β ) 2 > 4 (1+\beta)^2+v(\beta)^2>4 ( 1 + β ) 2 + v ( β ) 2 > 4 at the actual core exit. This uses the return’s own binary constant. Backward density integration yields
S < F ( d , 1 + β ) , F ( d , x ) = d x ( 4 − x 2 ) ( x − d ) . S<F(d,1+\beta), \qquad F(d,x)=\frac{dx}{(4-x^2)(x-d)}. S < F ( d , 1 + β ) , F ( d , x ) = ( 4 − x 2 ) ( x − d ) d x .
For 1 ≤ x ≤ 1 + d 1\le x\le1+d 1 ≤ x ≤ 1 + d , F x F_x F x has sign 2 x 3 − d x 2 − 4 d 2x^3-dx^2-4d 2 x 3 − d x 2 − 4 d , whose derivative is positive. Thus F F F has no interior maximum, and
F ( d , 1 + d ) − F ( d , 1 ) = 2 d 2 3 ( 1 − d ) ( 3 + d ) > 0 , S < G ( d ) = d ( 1 + d ) 3 − 2 d − d 2 , G ′ > 0. F(d,1+d)-F(d,1)=\frac{2d^2}{3(1-d)(3+d)}>0,\qquad S<G(d)=\frac{d(1+d)}{3-2d-d^2},\qquad G'>0. F ( d , 1 + d ) − F ( d , 1 ) = 3 ( 1 − d ) ( 3 + d ) 2 d 2 > 0 , S < G ( d ) = 3 − 2 d − d 2 d ( 1 + d ) , G ′ > 0.
Also F d = x 2 / [ ( 4 − x 2 ) ( x − d ) 2 ] > 0 F_d=x^2/[(4-x^2)(x-d)^2]>0 F d = x 2 / [( 4 − x 2 ) ( x − d ) 2 ] > 0 . Successively apply
Table 20.
These are finite rational comparisons, specified completely as follows. For each row check s − G ( d 0 ) > 0 s-G(d_0)>0 s − G ( d 0 ) > 0 and, at both h = 0 , D / 3 h=0,D/3 h = 0 , D /3 , put T h = ( 3 + h 2 ) s ′ / 2 − Ψ ( d 0 , h ) T_h=(3+h^2)s'/2-\Psi(d_0,h) T h = ( 3 + h 2 ) s ′ /2 − Ψ ( d 0 , h ) and check
T h > 0 , 3 + h 2 − 4 C 0 − T h 2 > 0. T_h>0,\qquad3+h^2-4C_0-T_h^2>0. T h > 0 , 3 + h 2 − 4 C 0 − T h 2 > 0.
The first comparison gives d > d 0 d>d_0 d > d 0 ; the latter two choose the positive square-root branch in the actual shear completion S > ( 3 + h 2 − 4 C 0 + E / S ) / k S>(\sqrt{3+h^2-4C_0+E/S})/k S > ( 3 + h 2 − 4 C 0 + E / S ) / k . Monotonicity and the endpoint principle then give S > s ′ S>s' S > s ′ throughout the height interval. Direct rational substitution verifies every displayed comparison (the smallest squared residual exceeds 16 / 100000 16/100000 16/100000 ). We obtain
S > s ∗ , 329 1000 < d < D ∗ , 0 < h < H ∗ : = D ∗ / 3. S>s_*,\qquad\frac{329}{1000}<d<D_*,\qquad0<h<H_*:=D_*/3. S > s ∗ , 1000 329 < d < D ∗ , 0 < h < H ∗ := D ∗ /3.
The width inequality is strict, even under the closed contrary budget.
Retain the asymmetry.
Since d < D ∗ < 9 / 25 d<D_*<9/25 d < D ∗ < 9/25 , the third row of Lemma B.13 gives Γ ( 9 / 25 ) / d > 1 \Gamma(9/25)/d>1 Γ ( 9/25 ) / d > 1 . Therefore
E / S > Ψ ( d , h ) + ( β − d / 2 ) 2 . E/S>\Psi(d,h)+(\beta-d/2)^2. E / S > Ψ ( d , h ) + ( β − d /2 ) 2 .
On this final domain 2 − 5 D ∗ > 0 2-5D_*>0 2 − 5 D ∗ > 0 , so F x > 0 F_x>0 F x > 0 for all x ≥ 1 x\ge1 x ≥ 1 used below. Moreover
1 k > 2 3 + H 2 > 33 50 , 2 3 + H 2 − 33 50 = 50671561585382537 14455169411838021175 > 0. \frac{1}{k}>\frac{2}{3+H^2}>\frac{33}{50},\qquad\frac{2}{3+H^2}-\frac{33}{50}=\frac{50671561585382537}{14455169411838021175}>0. k 1 > 3 + H 2 2 > 50 33 , 3 + H 2 2 − 50 33 = 14455169411838021175 50671561585382537 > 0.
The closed low-radius case.
If h ≤ 2 / 25 h\le2/25 h ≤ 2/25 , Lemma B.8 gives the actual last root r 0 < 8 h / ( 4 − 3 h ) ≤ r ∗ r_0<8h/(4-3h)\le r_* r 0 < 8 h / ( 4 − 3 h ) ≤ r ∗ . For β ≤ r ∗ \beta\le r_* β ≤ r ∗ ,
S < F ( d , 1 + β ) < F ( D ∗ , 1 + β ) ≤ F ( D ∗ , 1 + r ∗ ) = s ∗ , S<F(d,1+\beta)<F(D_*,1+\beta)\le F(D_*,1+r_*)=s_*, S < F ( d , 1 + β ) < F ( D ∗ , 1 + β ) ≤ F ( D ∗ , 1 + r ∗ ) = s ∗ ,
a contradiction. The equality in the final scalar comparison does not lose strictness. For β > r ∗ \beta>r_* β > r ∗ , retain the actual last mode-2 leaf:
S < J ( β − d , β , r 0 ) 4 − ( 1 + β ) 2 , J ( a , b , r ) = ( 1 − b ) 2 [ ( 1 + r ) 2 ( 1 − r ) 2 ( 1 1 + a − 1 1 + r ) + 1 1 − b − 1 1 − r ] . S<\frac{J(\beta-d,\beta,r_0)}{4-(1+\beta)^2},\qquad J(a,b,r)=(1-b)^2\left[\frac{(1+r)^2}{(1-r)^2}\left(\frac{1}{1+a}-\frac{1}{1+r}\right)+\frac{1}{1-b}-\frac{1}{1-r}\right]. S < 4 − ( 1 + β ) 2 J ( β − d , β , r 0 ) , J ( a , b , r ) = ( 1 − b ) 2 [ ( 1 − r ) 2 ( 1 + r ) 2 ( 1 + a 1 − 1 + r 1 ) + 1 − b 1 − 1 − r 1 ] .
Here J r = 4 ( r − a ) ( 1 − b ) 2 / [ ( 1 + a ) ( 1 − r ) 3 ] > 0 J_r=4(r-a)(1-b)^2/[(1+a)(1-r)^3]>0 J r = 4 ( r − a ) ( 1 − b ) 2 / [( 1 + a ) ( 1 − r ) 3 ] > 0 and J d = ( 1 − b ) 2 ( 1 + r ) 2 / [ ( 1 − r ) 2 ( 1 + b − d ) 2 ] > 0 J_d=(1-b)^2(1+r)^2/[(1-r)^2(1+b-d)^2]>0 J d = ( 1 − b ) 2 ( 1 + r ) 2 / [( 1 − r ) 2 ( 1 + b − d ) 2 ] > 0 after setting a = b − d a=b-d a = b − d . Enlarge r 0 r_0 r 0 to r ∗ r_* r ∗ and d d d to D ∗ D_* D ∗ , preserving all denominator signs. Multiplication of the resulting s ∗ − S s_*-S s ∗ − S bound by the positive ( b + 3 ) ( 1 − r ∗ ) 2 ( 1 + b − D ∗ ) (b+3)(1-r_*)^2(1+b-D_*) ( b + 3 ) ( 1 − r ∗ ) 2 ( 1 + b − D ∗ ) gives
q ( b ) = − 20594857 11045000 b 2 + 18296134235204047 9674856826495000 b − 323946560474377 1209357103311875 . q(b)=-\frac{20594857}{11045000}b^2+\frac{18296134235204047}{9674856826495000}b-\frac{323946560474377}{1209357103311875}. q ( b ) = − 11045000 20594857 b 2 + 9674856826495000 18296134235204047 b − 1209357103311875 323946560474377 .
It satisfies q ( r ∗ ) = 0 q(r_*)=0 q ( r ∗ ) = 0 , q ( D ∗ ) = 412481343731491 / 2418714206623750 > 0 q(D_*)=412481343731491/2418714206623750>0 q ( D ∗ ) = 412481343731491/2418714206623750 > 0 , and
q ( b ) − q ( D ∗ ) D ∗ − r ∗ = − 20594857 11045000 ( b − r ∗ ) ( b − D ∗ ) . \frac{q(b)-q(D_*)}{D_*-r_*}=-\frac{20594857}{11045000}(b-r_*)(b-D_*). D ∗ − r ∗ q ( b ) − q ( D ∗ ) = − 11045000 20594857 ( b − r ∗ ) ( b − D ∗ ) .
Thus q ( b ) > 0 q(b)>0 q ( b ) > 0 on r ∗ < b ≤ D ∗ r_*<b\leq D_* r ∗ < b ≤ D ∗ , again a contradiction.
Two closed high-radius intervals
Use the two rows
The same endpoint tests with d 0 = 329 / 1000 d_0=329/1000 d 0 = 329/1000 , now at h 0 , h 1 h_0,h_1 h 0 , h 1 , give S > s S>s S > s on each entire closed interval. Put D = 2 B E − 3 s D=2B_E-3s D = 2 B E − 3 s . Since α > h ≥ h 0 \alpha>h\geq h_0 α > h ≥ h 0 , we have β < D − h 0 \beta<D-h_0 β < D − h 0 . For β ≤ D / 2 \beta\leq D/2 β ≤ D /2 , the exact positive comparison s − F ( D , 1 + D / 2 ) > 0 s-F(D,1+D/2)>0 s − F ( D , 1 + D /2 ) > 0 is contradictory. Otherwise set
σ ( b ) = s + 33 50 ( b − D / 2 ) 2 , δ ( b ) = 2 B E − 3 σ ( b ) , L = D / 2 , U = D − h 0 . \sigma(b)=s+\frac{33}{50}(b-D/2)^2,\qquad\delta(b)=2B_E-3\sigma(b),\qquad L=D/2,\qquad U=D-h_0. σ ( b ) = s + 50 33 ( b − D /2 ) 2 , δ ( b ) = 2 B E − 3 σ ( b ) , L = D /2 , U = D − h 0 .
The retained asymmetry and β − d / 2 > β − D / 2 > 0 \beta-d/2>\beta-D/2>0 β − d /2 > β − D /2 > 0 give S > σ ( β ) S>\sigma(\beta) S > σ ( β ) and d < δ ( β ) d<\delta(\beta) d < δ ( β ) . On the full closed interval [ L , U ] [L,U] [ L , U ] , the three denominator reserves
δ ( b ) ≥ δ ( U ) > 0 , 4 − ( 1 + b ) 2 ≥ 4 − ( 1 + U ) 2 > 0 , 1 + b − δ ( b ) ≥ 1 + L − D > 0 \delta(b)\geq\delta(U)>0,\qquad4-(1+b)^2\geq4-(1+U)^2>0,\qquad1+b-\delta(b)\geq1+L-D>0 δ ( b ) ≥ δ ( U ) > 0 , 4 − ( 1 + b ) 2 ≥ 4 − ( 1 + U ) 2 > 0 , 1 + b − δ ( b ) ≥ 1 + L − D > 0
are exact rational inequalities. Therefore S < F ( δ ( β ) , 1 + β ) S<F(\delta(\beta),1+\beta) S < F ( δ ( β ) , 1 + β ) contradicts positivity of
P ( b ) = σ ( b ) [ 4 − ( 1 + b ) 2 ] [ 1 + b − δ ( b ) ] − δ ( b ) ( 1 + b ) . P(b)=\sigma(b)[4-(1+b)^2][1+b-\delta(b)]-\delta(b)(1+b). P ( b ) = σ ( b ) [ 4 − ( 1 + b ) 2 ] [ 1 + b − δ ( b )] − δ ( b ) ( 1 + b ) .
Set x = b − L x=b-L x = b − L ; in both rows 0 ≤ x ≤ U − L < 1 / 10 0\leq x\leq U-L<1/10 0 ≤ x ≤ U − L < 1/10 . Writing P ( L + x ) = ∑ i = 0 6 a i x i P(L+x)=\sum_{i=0}^{6}a_i x^i P ( L + x ) = ∑ i = 0 6 a i x i , exact rational expansion gives
( a 1 , … , a 6 ) > ( − 1117 / 5000 , 207 / 50 , 13 / 10 , 9 / 10 , − 15 / 4 , − 4 / 3 ) (a_1,\ldots,a_6)>(-1117/5000,207/50,13/10,9/10,-15/4,-4/3) ( a 1 , … , a 6 ) > ( − 1117/5000 , 207/50 , 13/10 , 9/10 , − 15/4 , − 4/3 )
componentwise, and a 0 > c i a_0>c_i a 0 > c i , where c 1 = 479 / 100000 c_1=479/100000 c 1 = 479/100000 , c 2 = 41 / 12500 c_2=41/12500 c 2 = 41/12500 . Dropping positive cubic and quartic terms, and using 207 / 50 − 15 / 4000 − 1 / 7500 > 413 / 100 207/50-15/4000-1/7500>413/100 207/50 − 15/4000 − 1/7500 > 413/100 , gives
P ( L + x ) > c i − 1117 5000 x + 413 100 x 2 ≥ c i − 1247689 413000000 = { 730581 / 413000000 , i = 1 , 106951 / 413000000 , i = 2. P(L+x)>c_i-\frac{1117}{5000}x+\frac{413}{100}x^2\geq c_i-\frac{1247689}{413000000}=
\begin{cases}
730581/413000000, & i=1,\\
106951/413000000, & i=2.
\end{cases} P ( L + x ) > c i − 5000 1117 x + 100 413 x 2 ≥ c i − 413000000 1247689 = { 730581/413000000 , 106951/413000000 , i = 1 , i = 2.
These exact positive bounds cover both whole closed intervals, including all endpoints. All height cases are now exhausted.
For a left context, its actual initial return gives its entrance circle. In the same actual integrals change variable u ↦ − u u\mapsto-u u ↦ − u and reverse limits. This exchanges α , β \alpha,\beta α , β and the two cap losses, placing that one returned endpoint in the role just proved. It neither assumes a second circle nor identifies different binary invariants. Lawful whole cyclic transport gives the other four contexts and their stated charges. □
Corollary B.17 (A weaker comparison for overlapping cores). Every one-return charge in Theorem B.16 exceeds 473 / 1000 473/1000 473/1000 .
Proof. The exact constant in Theorem B.16 exceeds 473 / 1000 473/1000 473/1000 . This corollary is used only in a few overlapping-core comparisons below. □
Lemma B.18 (Two actual return circles and cap losses). Retain the complete source 01020212 on one full normal lift, with four-core endpoints − α , β -\alpha,\beta − α , β , mean h 2 h^2 h 2 , density q = v − 2 q=v^{-2} q = v − 2 , shear S S S and losses E a , E b E_a,E_b E a , E b of Lemma B.7. Both actual returns give
q ( − α ) < 1 ( 1 − α ) ( 3 + α ) , q ( β ) < 1 ( 1 − β ) ( 3 + β ) . (319) q(-\alpha)<\frac{1}{(1-\alpha)(3+\alpha)},\qquad q(\beta)<\frac{1}{(1-\beta)(3+\beta)}. \tag*{(319)} q ( − α ) < ( 1 − α ) ( 3 + α ) 1 , q ( β ) < ( 1 − β ) ( 3 + β ) 1 . (319) Moreover,
E a q ( − α ) > e 2 ( α , h ) , E b q ( β ) > e 2 ( β , h ) , e 2 ( t , h ) = t 2 − h 2 2 − 2 ( t − h ) 2 ( t + 2 h ) 3 ( 1 − t ) ( 3 + t ) . (320) \frac{E_a}{q(-\alpha)}>e_2(\alpha,h),\qquad\frac{E_b}{q(\beta)}>e_2(\beta,h),\qquad e_2(t,h)=\frac{t^2-h^2}{2}-\frac{2(t-h)^2(t+2h)}{3(1-t)(3+t)}. \tag*{(320)} q ( − α ) E a > e 2 ( α , h ) , q ( β ) E b > e 2 ( β , h ) , e 2 ( t , h ) = 2 t 2 − h 2 − 3 ( 1 − t ) ( 3 + t ) 2 ( t − h ) 2 ( t + 2 h ) . (320)
For h < t ≤ 349 / 500 h < t \le349/500 h < t ≤ 349/500 , these lower losses are positive. In that domain, with d = α + β d = \alpha+ \beta d = α + β ,
E a S > ( 1 − β ) e 2 ( α , h ) d ( 1 + α ) , E b S > ( 1 − α ) e 2 ( β , h ) d ( 1 + β ) . \frac{E_a}{S} > \frac{(1-\beta)e_2(\alpha,h)}{d(1+\alpha)}, \qquad\frac{E_b}{S} > \frac{(1-\alpha)e_2(\beta,h)}{d(1+\beta)}. S E a > d ( 1 + α ) ( 1 − β ) e 2 ( α , h ) , S E b > d ( 1 + β ) ( 1 − α ) e 2 ( β , h ) .
No numerical four-core cost floor is used.
Proof. Apply Lemma B.10 to the actual initial 010, in its whole cyclic chart 202, and to the final 212, in its whole cyclic chart 020. Their strict roots give respectively v ( − α ) 2 > ( 1 − α ) ( 3 + α ) v(-\alpha)^2 > (1-\alpha)(3+\alpha) v ( − α ) 2 > ( 1 − α ) ( 3 + α ) and v ( β ) 2 > ( 1 − β ) ( 3 + β ) v(\beta)^2 > (1-\beta)(3+\beta) v ( β ) 2 > ( 1 − β ) ( 3 + β ) , proving (B.29). The binary constants in the return charts need not agree with the four-core constant; all restrictions belong to the same jointly normalized full lift.
Apply Lemma B.6 to both actual outer leaves, using their separately proved circles. It gives (B.30), including positivity for h < t ≤ 349 / 500 h < t \le349/500 h < t ≤ 349/500 , without identifying their return-chart constants. Only after establishing positivity multiply the two loss bounds by the corresponding endpoint-density inequalities (B.15). □
Theorem B.19 (A coarse two-return comparison). In the three actual positive contexts 01020212, 12101020, 20212101, the physical cost of positions 2, 3, 4, 5 is strictly larger than 261/500. The two caps and both returns are uncharged.
Proof. The source retains all eight complete maximal arcs in one genuine fundamental interval of the original minimizer, as required throughout this section. Lemma B.1 makes it proper. The independent two-return theorem in Section C.6.2 therefore gives the stronger bound A > 74 / 125 > 261 / 500 A > 74/125 > 261/500 A > 74/125 > 261/500 on these same four charges. Its proof uses the actual circle and cap-loss identities of Lemma B.18, not the present numerical conclusion. Whole lawful cyclic transport gives the other two contexts. Thus the original complete-source scope is retained without an additional properness assumption. □
Exact polynomial certificates for the four-core estimates
The common product envelope replaces separate polynomial symmetrizations. Its rational table and the bootstrap tables in the proofs are complete finite certificates, not samples of physical trajectories.
The bootstrap tables in the proofs constitute equally explicit rational certificates. A radical comparison R > T \sqrt{R} > T R > T is checked only after T > 0 T > 0 T > 0 , and is then exactly equivalent to R − T 2 > 0 R - T^2 > 0 R − T 2 > 0 . To make positivity readily checkable without printing large denominators, the next table gives conservative positive rational lower bounds for each group of residuals. Every bound follows by substitution in these formulas and clearing positive integer denominators.
Table 1 (PDF p. 177).
All unsquared branch thresholds in these groups exceed 1 / 10 1/10 1/10 .
The three-arc estimate
This section proves the local estimate that will be used for a charged three-arc core. The only physical contexts covered here are
02010 , 10121 , 21202. 02010,\qquad10121,\qquad21202. 02010 , 10121 , 21202.
The three middle arcs are charged; the first and last arcs are strictly positive, actual, uncharged caps. All quantities below come from one complete normal lift, with its multiplier jointly normalized once. In particular, the two leaf invariants are not independently chosen costates. The admissible reanchoring requirement of Lemma 2.6 is retained. Write
U = 49 200 , H ∗ = 29025933167887355473205403 39478417604337498218323684 . (321) U = \frac{49}{200}, \qquad H_* = \frac{29025933167887355473205403}{39478417604337498218323684}. \tag*{(321)} U = 200 49 , H ∗ = 39478417604337498218323684 29025933167887355473205403 . (321)
The symbol C \mathcal{C} C below denotes the determinant of the matrix covector, not the cost of a trajectory.
Theorem B.20 (Qualified three-arc bound). For an actual context of one of the three types above, to which the full normal PMP and the global harmonic calibration apply, the cost F F F of its three middle arcs satisfies F > 49 / 200 F > 49/200 F > 49/200 .
Reconstruction from a single full lift.
It suffices first to treat 02010. Denote its charged arcs by A , c , B A,c,B A , c , B , of modes 2, 0, 1, with costs α , A c , β \alpha,A_c,\beta α , A c , β . Put
Λ = ( λ 1 λ 2 m − λ 1 ) , K = 2 u λ 1 + λ 2 − u 2 m , L = λ 1 − u m , χ = v p v , d s = d t / v . \Lambda= \begin{pmatrix} \lambda_1 & \lambda_2 \\ m & -\lambda_1 \end{pmatrix}, \qquad K = 2u\lambda_1+\lambda_2-u^2m,\qquad L=\lambda_1-um,\qquad\chi=vp_v,\qquad ds=dt/v. Λ = ( λ 1 m λ 2 − λ 1 ) , K = 2 u λ 1 + λ 2 − u 2 m , L = λ 1 − u m , χ = v p v , d s = d t / v .
Differentiation of the full equations, using only the active Hamiltonian H j = 0 H_j=0 H j = 0 , gives on a 0/2 edge
K s = 0 , L s = K , m s = − 2 L , χ s = 2 K − 3 − u 2 , C = − L 2 − K m . (322) K_s=0,\qquad L_s=K,\qquad m_s=-2L,\qquad\chi_s=2K-3-u^2,\qquad\mathcal{C}=-L^2-Km. \tag*{(322)} K s = 0 , L s = K , m s = − 2 L , χ s = 2 K − 3 − u 2 , C = − L 2 − K m . (322)
Indeed H 0 − H 2 = 2 v 2 p v / ( 1 − u 2 ) H_0-H_2=2v^2p_v/(1-u^2) H 0 − H 2 = 2 v 2 p v / ( 1 − u 2 ) ; maximization therefore gives χ ≥ 0 \chi\geq0 χ ≥ 0 in mode 0 and χ ≤ 0 \chi\leq0 χ ≤ 0 in mode 2. At an actual 2 → 0 2\to0 2 → 0 switch, χ ′ \chi' χ ′ is strictly positive. If it vanished at u ≠ 0 u\ne0 u = 0 , its common second derivative would be − 2 u -2u − 2 u , inconsistent with the sign change. If it vanished at u = 0 u=0 u = 0 , then K = 3 / 2 K=3/2 K = 3/2 and χ ′ = − u 2 / v < 0 \chi'=-u^2/v<0 χ ′ = − u 2 / v < 0 on the neighboring nontrivial arcs, again impossible. This argument allows tied switches and does not assume transversality. The continuous maximum-zero relation at a switch follows from its a.e. version by continuity of all three Hamiltonians.
On the middle mode-0 arc set ξ = 1 + u \xi=1+u ξ = 1 + u , v = k ξ v=k\xi v = k ξ , with endpoints ξ = a , a E \xi=a,aE ξ = a , a E , where E > 1 E>1 E > 1 . Its lawful cyclic image R ( w ) = ( w − 3 ) / ( w + 1 ) \mathcal{R}(w)=(w-3)/(w+1) R ( w ) = ( w − 3 ) / ( w + 1 ) is
U s = 1 − 4 ( 1 + k 2 ) ξ , V s = 4 k ( 1 + k 2 ) ξ . U_s=1-\frac{4}{(1+k^2)\xi},\qquad V_s=\frac{4k}{(1+k^2)\xi}. U s = 1 − ( 1 + k 2 ) ξ 4 , V s = ( 1 + k 2 ) ξ 4 k .
(The subscript on U s , V s U_s,V_s U s , V s here distinguishes state coordinates.) At the exit put b = 4 / ( ( 1 + k 2 ) a E ) b=4/((1+k^2)aE) b = 4/ (( 1 + k 2 ) a E ) and define
r = 1 − a , s = 1 − b , P = a b , Q = a + b , R = r s , r=1-a,\qquad s=1-b,\qquad P=ab,\qquad Q=a+b,\qquad R=rs, r = 1 − a , s = 1 − b , P = ab , Q = a + b , R = r s ,
D = 4 − E P , z = log E , t = 2 z / D . (323) D=4-EP,\qquad z=\log E,\qquad t=2z/D. \tag*{(323)} D = 4 − E P , z = log E , t = 2 z / D . (323)
Then k 2 = D / ( E a b ) k^2=D/(Eab) k 2 = D / ( E ab ) , a , b > 0 a,b>0 a , b > 0 , a E < 2 aE<2 a E < 2 , and b E < 2 bE<2 b E < 2 . The actual positive leaf widths satisfy 0 < d A < a 0<d_A<a 0 < d A < a , 0 < d B < b 0<d_B<b 0 < d B < b . Let
e ( w , d ) = ∫ 0 d ( 2 w δ + δ 2 ) ( 1 + w + δ ) − 2 d δ ∫ 0 d ( 1 + w + δ ) − 2 d δ , e A = e ( r , d A ) , e B = e ( s , d B ) . (324) e(w,d)=\frac{\int_0^d (2w\delta+\delta^2)(1+w+\delta)^{-2}\,d\delta}{\int_0^d (1+w+\delta)^{-2}\,d\delta},\qquad e_A=e(r,d_A),\qquad e_B=e(s,d_B). \tag*{(324)} e ( w , d ) = ∫ 0 d ( 1 + w + δ ) − 2 d δ ∫ 0 d ( 2 w δ + δ 2 ) ( 1 + w + δ ) − 2 d δ , e A = e ( r , d A ) , e B = e ( s , d B ) . (324)
The two actual roots on each leaf, together with (B.32), imply
K A = ( 3 + r 2 + e A ) / 2 , K B = ( 3 + s 2 + e B ) / 2 , e A , e B > 0. K_A=(3+r^2+e_A)/2,\qquad K_B=(3+s^2+e_B)/2,\qquad e_A,e_B>0. K A = ( 3 + r 2 + e A ) /2 , K B = ( 3 + s 2 + e B ) /2 , e A , e B > 0.
The strict inequalities are the central-facing switch argument just given. Here is an explicit common-matrix reconstruction. Direct conjugation by R ∗ = 1 2 ( 1 − 3 1 1 ) R_*=\frac{1}{2}\begin{pmatrix}1&-3\\1&1\end{pmatrix} R ∗ = 2 1 ( 1 1 − 3 1 ) gives, on the central arc,
K B = 4 ( K A + 2 L k 2 ξ − k 4 m ξ 2 ) ( 1 + k 2 ) 2 ξ 2 . K_B=\frac{4(K_A+2Lk^2\xi-k^4m\xi^2)}{(1+k^2)^2\xi^2}. K B = ( 1 + k 2 ) 2 ξ 2 4 ( K A + 2 L k 2 ξ − k 4 m ξ 2 ) .
Since J m = L + K A / ( k 2 ξ ) J_m=L+K_A/(k^2\xi) J m = L + K A / ( k 2 ξ ) is constant, solving this equation gives
L = J m − K A k 2 ξ , L=J_m-\frac{K_A}{k^2\xi}, L = J m − k 2 ξ K A ,
m = 2 J m k 2 ξ − K A k 4 ξ 2 − K B ( 1 + k 2 ) 2 4 k 4 , m=\frac{2J_m}{k^2\xi}-\frac{K_A}{k^4\xi^2}-\frac{K_B(1+k^2)^2}{4k^4}, m = k 2 ξ 2 J m − k 4 ξ 2 K A − 4 k 4 K B ( 1 + k 2 ) 2 ,
C = K A K B ( 1 + k 2 ) 2 4 k 4 − J m 2 . \mathcal{C}=\frac{K_AK_B(1+k^2)^2}{4k^4}-J_m^2. C = 4 k 4 K A K B ( 1 + k 2 ) 2 − J m 2 .
Integrating χ s \chi_s χ s from the entry root gives
χ ( a E ) = k − 2 ( ( 2 K A − 4 ) ( E − 1 ) / ( a E ) − a ( E − 1 ) + 2 z ) . \chi(aE)=k^{-2}\left((2K_A-4)(E-1)/(aE)-a(E-1)+2z\right). χ ( a E ) = k − 2 ( ( 2 K A − 4 ) ( E − 1 ) / ( a E ) − a ( E − 1 ) + 2 z ) .
At the exit H 0 = H 1 = 0 H_0=H_1=0 H 0 = H 1 = 0 gives
m + 1 2 = 2 K A − 3 − u 2 2 k 2 ξ + ( 1 + k 2 ) χ 2 k 2 ξ . m+\frac{1}{2}=\frac{2K_A-3-u^2}{2k^2\xi}+\frac{(1+k^2)\chi}{2k^2\xi}. m + 2 1 = 2 k 2 ξ 2 K A − 3 − u 2 + 2 k 2 ξ ( 1 + k 2 ) χ .
Substitution determines, with x = e A / ( a D ) x=e_A/(aD) x = e A / ( a D ) and y = e B / ( b D ) y=e_B/(bD) y = e B / ( b D ) ,
J m = x + y + t + 4 Q − 4 − E P 2 D , C = 4 K A K B D 2 − J m 2 . (325) J_m=x+y+t+\frac{4Q-4-EP}{2D},\qquad\mathcal{C}=\frac{4K_AK_B}{D^2}-J_m^2. \tag*{(325)} J m = x + y + t + 2 D 4 Q − 4 − E P , C = D 2 4 K A K B − J m 2 . (325)
Consequently the full determinant expansion is
C = 3 4 + s x + r y + R ( 4 − P ) t D − ( x + y + t ) 2 + P x y + P ( E − 1 ) ( 2 − b ) x D + P ( E − 1 ) ( 2 − a ) y D + P G D 2 , G = [ 6 − 2 R − P ( E − 1 ) ] ( E − 1 − z ) + ( 2 − P ) z ( E − 1 ) . (326) \begin{aligned}
\mathcal{C}={}&\frac{3}{4}+sx+ry+\frac{R(4-P)t}{D}-(x+y+t)^2+Pxy\\
&+\frac{P(E-1)(2-b)x}{D}+\frac{P(E-1)(2-a)y}{D}+\frac{PG}{D^2},\\
G={}&[6-2R-P(E-1)](E-1-z)+(2-P)z(E-1). \tag*{(326)}
\end{aligned} C = G = 4 3 + s x + r y + D R ( 4 − P ) t − ( x + y + t ) 2 + P x y + D P ( E − 1 ) ( 2 − b ) x + D P ( E − 1 ) ( 2 − a ) y + D 2 P G , [ 6 − 2 R − P ( E − 1 )] ( E − 1 − z ) + ( 2 − P ) z ( E − 1 ) . (326)
This is an identity; in particular none of its remainder terms has yet been discarded. The actual leaf endpoint heights are V A 2 = a D / ( E b ) V_A^2=aD/(Eb) V A 2 = a D / ( E b ) and V B 2 = b D / ( E a ) V_B^2=bD/(Ea) V B 2 = b D / ( E a ) . Integration of the physical cost yields
α = d A 2 + K A d A ( 1 + r ) V A 2 ( 1 + r + d A ) , β = d B 2 + K B d B ( 1 + s ) V B 2 ( 1 + s + d B ) , A c = 2 Q ( E − 1 ) − E P z D . (327) \alpha=\frac{d_A}{2}+\frac{K_A d_A(1+r)}{V_A^2(1+r+d_A)},\qquad
\beta=\frac{d_B}{2}+\frac{K_B d_B(1+s)}{V_B^2(1+s+d_B)},\qquad
A_c=\frac{2Q(E-1)-EPz}{D}. \tag*{(327)} α = 2 d A + V A 2 ( 1 + r + d A ) K A d A ( 1 + r ) , β = 2 d B + V B 2 ( 1 + s + d B ) K B d B ( 1 + s ) , A c = D 2 Q ( E − 1 ) − E P z . (327)
Thus d A < 2 α d_A<2\alpha d A < 2 α , d B < 2 β d_B<2\beta d B < 2 β , and direct integration of ℓ d t > d u / 2 \ell\,dt>du/2 ℓ d t > d u /2 on the central arc and its cyclic image gives a ( E − 1 ) < 2 A c a(E-1)<2A_c a ( E − 1 ) < 2 A c , b ( E − 1 ) < 2 A c b(E-1)<2A_c b ( E − 1 ) < 2 A c . Put J A = 2 a + E b ( 2 − a ) J_A=2a+Eb(2-a) J A = 2 a + E b ( 2 − a ) and J B = 2 b + E a ( 2 − b ) J_B=2b+Ea(2-b) J B = 2 b + E a ( 2 − b ) . The useful strict leaf bound has the following short proof. For q = 1 + r q=1+r q = 1 + r ,
α − d A J A a D ( 1 + d A / q ) = d A 2 2 ( q + d A ) + e A d A q 2 V A 2 ( q + d A ) > 0. (328) \alpha-\frac{d_AJ_A}{aD(1+d_A/q)}
=\frac{d_A^2}{2(q+d_A)}+\frac{e_A d_Aq}{2V_A^2(q+d_A)}>0. \tag*{(328)} α − a D ( 1 + d A / q ) d A J A = 2 ( q + d A ) d A 2 + 2 V A 2 ( q + d A ) e A d A q > 0. (328)
The corresponding identity holds on the right leaf. This proof uses the actual cost and positive mean and requires no extra endpoint geometry. For w ≥ 0 w\ge0 w ≥ 0 , the decreasing-weight covariance inequality gives e ( w , d ) ≤ w d + d 2 / 3 e(w,d)\le wd+d^2/3 e ( w , d ) ≤ w d + d 2 /3 . Hence, defining
N ( w , d ) = ( max ( w , 0 ) + d 3 ) ( 1 + d 1 + w ) , N(w,d)=\left(\max(w,0)+\frac{d}{3}\right)\left(1+\frac{d}{1+w}\right), N ( w , d ) = ( max ( w , 0 ) + 3 d ) ( 1 + 1 + w d ) ,
we obtain x ≤ α N ( r , d A ) / J A x\le\alpha N(r,d_A)/J_A x ≤ α N ( r , d A ) / J A and y ≤ β N ( s , d B ) / J B y\le\beta N(s,d_B)/J_B y ≤ β N ( s , d B ) / J B ; the negative-coordinate case follows also from the signed mean proof below.
Set Φ ( E ) = E + 1 − 2 E log E / ( E − 1 ) \Phi(E)=E+1-2E\log E/(E-1) Φ ( E ) = E + 1 − 2 E log E / ( E − 1 ) , with Φ ( 1 ) = 0 \Phi(1)=0 Φ ( 1 ) = 0 . The central maximizing signs in the two actual charts give
e A ≥ a [ Φ ( E ) − r ( E − 1 ) ] , e B ≥ b [ Φ ( E ) − s ( E − 1 ) ] . (329) e_A\ge a[\Phi(E)-r(E-1)],\qquad e_B\ge b[\Phi(E)-s(E-1)]. \tag*{(329)} e A ≥ a [ Φ ( E ) − r ( E − 1 )] , e B ≥ b [ Φ ( E ) − s ( E − 1 )] . (329)
The common integral identity
Φ ( E ) = 2 ( E − 1 ) 2 ∫ 0 1 v ( 1 − v ) 1 + ( E − 1 ) v d v ≥ ( E − 1 ) 2 3 E \Phi(E)=2(E-1)^2\int_0^1\frac{v(1-v)}{1+(E-1)v}\,dv\ge\frac{(E-1)^2}{3E} Φ ( E ) = 2 ( E − 1 ) 2 ∫ 0 1 1 + ( E − 1 ) v v ( 1 − v ) d v ≥ 3 E ( E − 1 ) 2
will also be used. Both signs are derived before exchanging scalar names; no odd permutation of the physical modes is being invoked.
The necessary third-control root condition.
At an actual 0/2 tie p v = 0 p_v=0 p v = 0 . The remaining Hamiltonian difference is H 1 − H 0 = − 2 v 3 p u / ( u 2 + v 2 − 1 ) H_1-H_0=-2v^3p_u/(u^2+v^2-1) H 1 − H 0 = − 2 v 3 p u / ( u 2 + v 2 − 1 ) , so full maximization implies p u ≥ 0 p_u\geq0 p u ≥ 0 . The active Hamiltonian then gives
p u = m + 1 2 − 2 K − 3 − u 2 2 v 2 . p_u=m+\frac{1}{2}-\frac{2K-3-u^2}{2v^2}. p u = m + 2 1 − 2 v 2 2 K − 3 − u 2 .
At the two central-facing roots the same matrix has
L A = J m − K A E b / D , m A = 2 J m E b / D − K A E 2 b 2 / D 2 − 4 K B / D 2 , L_A=J_m-K_AEb/D,\qquad m_A=2J_mEb/D-K_AE^2b^2/D^2-4K_B/D^2, L A = J m − K A E b / D , m A = 2 J m E b / D − K A E 2 b 2 / D 2 − 4 K B / D 2 ,
L B = − J m + K B E a / D , m B = 2 J m E a / D − K B E 2 a 2 / D 2 − 4 K A / D 2 . L_B=-J_m+K_BEa/D,\qquad m_B=2J_mEa/D-K_BE^2a^2/D^2-4K_A/D^2. L B = − J m + K B E a / D , m B = 2 J m E a / D − K B E 2 a 2 / D 2 − 4 K A / D 2 .
The second line is obtained by the actual cyclic conjugation above. On the right leaf put q B = 2 − b q_B=2-b q B = 2 − b , d = d B d=d_B d = d B and
τ = d q B V B 2 ( q B + d ) , T B = E a D + τ = E a ( b q B + 2 d ) b D ( q B + d ) . \tau=\frac{dq_B}{V_B^2(q_B+d)},\qquad T_B=\frac{Ea}{D}+\tau=\frac{Ea(bq_B+2d)}{bD(q_B+d)}. τ = V B 2 ( q B + d ) d q B , T B = D E a + τ = b D ( q B + d ) E a ( b q B + 2 d ) .
Forward propagation gives m o u t = m B − 2 L B τ − K B τ 2 m_{\mathrm{out}}=m_B-2L_B\tau-K_B\tau^2 m out = m B − 2 L B τ − K B τ 2 , and hence the necessary inequality
p o u t = 1 2 − 4 K A D 2 + 2 J m T B − K B T B 2 − ( e B − 2 s d − d 2 ) q B 2 2 V B 2 ( q B + d ) 2 ≥ 0. (330) p_{\mathrm{out}}=\frac{1}{2}-\frac{4K_A}{D^2}+2J_mT_B-K_BT_B^2-\frac{(e_B-2sd-d^2)q_B^2}{2V_B^2(q_B+d)^2}\geq0. \tag*{(330)} p out = 2 1 − D 2 4 K A + 2 J m T B − K B T B 2 − 2 V B 2 ( q B + d ) 2 ( e B − 2 s d − d 2 ) q B 2 ≥ 0. (330)
Backward propagation on the left gives the analogous formula with m ( t 0 ) = m A + 2 L A τ A − K A τ A 2 m(t_0)=m_A+2L_A\tau_A-K_A\tau_A^2 m ( t 0 ) = m A + 2 L A τ A − K A τ A 2 and τ A = d A ( 2 − a ) / ( V A 2 ( 2 − a + d A ) ) \tau_A=d_A(2-a)/(V_A^2(2-a+d_A)) τ A = d A ( 2 − a ) / ( V A 2 ( 2 − a + d A )) . Thus both exterior tests are valid in their own charts before scalar sorting. Only the right test is needed by the finite residual proof.
An analytic treatment of negative coordinates.
Lemma B.21. If at least one of r , s r,s r , s is negative and 0 < F ≤ A ≤ U − = 129 / 500 0<F\leq A\leq U_-=129/500 0 < F ≤ A ≤ U − = 129/500 , then C > H ∗ . \mathcal{C}>H_*. C > H ∗ .
Proof. The proof uses (B.36)–(B.39), not a separate negative-coordinate theorem. All budgets in this proof refer to the same three actual arcs.
Signed means and the retained determinant remainder.
If r = − κ < 0 r=-\kappa<0 r = − κ < 0 , put q = 1 − κ > 0 q=1-\kappa>0 q = 1 − κ > 0 and M ( v ) = v 2 / 3 − κ v M(v)=v^2/3-\kappa v M ( v ) = v 2 /3 − κ v . Since
( v / d ) M ( d ) − M ( v ) = v ( d − v ) / 3 ≥ 0 , (v/d)M(d)-M(v)=v(d-v)/3\geq0, ( v / d ) M ( d ) − M ( v ) = v ( d − v ) /3 ≥ 0 ,
we have M ( v ) ≤ max ( 0 , M ( d ) ) M(v)\leq\max(0,M(d)) M ( v ) ≤ max ( 0 , M ( d )) on [ 0 , d ] [0,d] [ 0 , d ] . Integration by parts with ω ( v ) = ( q + v ) − 2 \omega(v)=(q+v)^{-2} ω ( v ) = ( q + v ) − 2 writes the mean in (B.34) as
e A = d ω ( d ) M ( d ) + ∫ 0 d [ − v ω ′ ( v ) ] M ( v ) d v d ω ( d ) + ∫ 0 d [ − v ω ′ ( v ) ] d v . e_A=\frac{d\omega(d)M(d)+\int_0^d[-v\omega'(v)]M(v)\,dv}{d\omega(d)+\int_0^d[-v\omega'(v)]\,dv}. e A = d ω ( d ) + ∫ 0 d [ − v ω ′ ( v )] d v d ω ( d ) M ( d ) + ∫ 0 d [ − v ω ′ ( v )] M ( v ) d v .
Thus e A > 0 e_A>0 e A > 0 implies d A > 3 κ d_A>3\kappa d A > 3 κ and e A ≤ d A 2 / 3 − κ d A e_A\leq d_A^2/3-\kappa d_A e A ≤ d A 2 /3 − κ d A . No monotonicity of the signed quadratic has been presumed. The central sign, the integral for Φ \Phi Φ , and ( E − 1 ) / E ≥ z (E-1)/\sqrt{E}\geq z ( E − 1 ) / E ≥ z give
e A ≥ a ( z 2 / 3 + κ z ) , e A ≥ z 2 / 3 + κ z , d A ≥ z + 3 κ . (331) e_A\geq a(z^2/3+\kappa z),\qquad e_A\geq z^2/3+\kappa z,\qquad d_A\geq z+3\kappa. \tag*{(331)} e A ≥ a ( z 2 /3 + κ z ) , e A ≥ z 2 /3 + κ z , d A ≥ z + 3 κ . (331)
The last implication follows by factoring d A 2 / 3 − κ d A − z 2 / 3 − κ z = ( d A + z ) ( d A − z − 3 κ ) / 3 d_A^2/3-\kappa d_A-z^2/3-\kappa z=(d_A+z)(d_A-z-3\kappa)/3 d A 2 /3 − κ d A − z 2 /3 − κ z = ( d A + z ) ( d A − z − 3 κ ) /3 . In particular κ < 2 α / 3 \kappa<2\alpha/3 κ < 2 α /3 and a < a m : = 1 + 2 U − / 3 = 293 / 250 a<a_m:=1+2U_-/3=293/250 a < a m := 1 + 2 U − /3 = 293/250 . The analogous facts hold on the other side.
More generally the positive mean implies − U − < r , s < 1 -U_-<r,s<1 − U − < r , s < 1 : if a coordinate is negative this follows from the stronger depth bound just proved. Therefore P < ( 1 + U − ) 2 < 2 P<(1+U_-)^2<2 P < ( 1 + U − ) 2 < 2 and P ( E − 1 ) < 2 U − ( 1 + U − ) P(E-1)<2U_-(1+U_-) P ( E − 1 ) < 2 U − ( 1 + U − ) . Every coefficient used from G G G in (B.36) is positive. Denoting its last four positive terms by Rem, the inequalities E − 1 − z ≥ z 2 / 2 E-1-z\geq z^2/2 E − 1 − z ≥ z 2 /2 and E − 1 ≥ z E-1\geq z E − 1 ≥ z give
Rem ≥ P 4 [ 5 − r s − P − P ( E − 1 ) / 2 ] t 2 . (332) \operatorname{Rem}\geq\frac{P}{4}[5-rs-P-P(E-1)/2]t^2. \tag*{(332)} Rem ≥ 4 P [ 5 − r s − P − P ( E − 1 ) /2 ] t 2 . (332)
Also the central cost has the positive-tail estimate
A c ≥ 2 Q − E P 2 t + Q D 4 t 2 + Q D 2 24 t 3 . (333) A_c \ge\frac{2Q-EP}{2}t+\frac{QD}{4}t^2+\frac{QD^2}{24}t^3. \tag*{(333)} A c ≥ 2 2 Q − E P t + 4 Q D t 2 + 24 Q D 2 t 3 . (333)
These follow by expanding e z − 1 e^z-1 e z − 1 , with z = D t / 2 z=Dt/2 z = D t /2 .
Mixed signs: a uniform right-leaf coefficient.
After the two chart derivations we may name the negative coordinate r = − κ < 0 r=-\kappa<0 r = − κ < 0 and take s = 1 − b ≥ 0 s=1-b\ge0 s = 1 − b ≥ 0 . The actual exit has ξ = a E ∈ ( 1 , 2 ) \xi=aE\in(1,2) ξ = a E ∈ ( 1 , 2 ) and v s t a t e 2 ≤ 243 / 4 v_{\mathrm{state}}^2\le243/4 v state 2 ≤ 243/4 by the original compact set. Hence
b = 4 ξ ξ 2 + v s t a t e 2 ≥ 16 ξ 4 ξ 2 + 243 > b 0 : = 16 247 . b=\frac{4\xi}{\xi^2+v_{\mathrm{state}}^2}\ge\frac{16\xi}{4\xi^2+243}>b_0:=\frac{16}{247}. b = ξ 2 + v state 2 4 ξ ≥ 4 ξ 2 + 243 16 ξ > b 0 := 247 16 .
The rational function is increasing on [ 1 , 2 ] [1,2] [ 1 , 2 ] . Moreover
b D J B = b ( 4 − ξ b ) 2 b + ξ ( 2 − b ) ≤ b ( 4 − b ) 2 + b , \frac{bD}{J_B}=\frac{b(4-\xi b)}{2b+\xi(2-b)}\le\frac{b(4-b)}{2+b}, J B b D = 2 b + ξ ( 2 − b ) b ( 4 − ξ b ) ≤ 2 + b b ( 4 − b ) ,
since its ξ \xi ξ derivative is negative. The strict bound (B.38), with β ≤ U − \beta\le U_- β ≤ U − , therefore implies
d B < h ( b ) : = U − b ( 4 − b ) ( 2 − b ) 4 − b 2 − U − b ( 4 − b ) . d_B<h(b):=\frac{U_-b(4-b)(2-b)}{4-b^2-U_-b(4-b)}. d B < h ( b ) := 4 − b 2 − U − b ( 4 − b ) U − b ( 4 − b ) ( 2 − b ) .
The denominator decreases on [ 0 , 1 ] [0,1] [ 0 , 1 ] to 3 ( 1 − U − ) > 0 3(1-U_-)>0 3 ( 1 − U − ) > 0 . Since N ( 1 − b , d ) N(1-b,d) N ( 1 − b , d ) increases in d d d and J B > 2 + b J_B>2+b J B > 2 + b ,
y < β F ( b ) , F ( b ) = [ 1 − b + h ( b ) / 3 ] [ 1 + h ( b ) / ( 2 − b ) ] 2 + b . y<\beta\mathcal{F}(b),\qquad\mathcal{F}(b)=\frac{[1-b+h(b)/3][1+h(b)/(2-b)]}{2+b}. y < β F ( b ) , F ( b ) = 2 + b [ 1 − b + h ( b ) /3 ] [ 1 + h ( b ) / ( 2 − b )] .
For k 0 = 233 / 500 k_0=233/500 k 0 = 233/500 , exact reduction gives
k 0 − F ( b ) = P 4 ( b ) 500 ( 371 b 2 + 516 b − 2000 ) 2 , k_0-\mathcal{F}(b)=\frac{P_4(b)}{500(371b^2+516b-2000)^2}, k 0 − F ( b ) = 500 ( 371 b 2 + 516 b − 2000 ) 2 P 4 ( b ) ,
where
P 4 ( b ) = 135570353 b 4 − 146040824 b 3 − 770234352 b 2 + 1105088000 b − 68000000. P_4(b)=135570353b^4-146040824b^3-770234352b^2+1105088000b-68000000. P 4 ( b ) = 135570353 b 4 − 146040824 b 3 − 770234352 b 2 + 1105088000 b − 68000000.
On 0 ≤ b ≤ 1 0\le b\le1 0 ≤ b ≤ 1 , b 2 ≤ b b^2\le b b 2 ≤ b gives P 4 ′ ′ ( b ) ≤ − 789869412 < 0 P_4''(b)\le-789869412<0 P 4 ′′ ( b ) ≤ − 789869412 < 0 . Its endpoint values on the required interval are
P 4 ( 16 / 247 ) = 1173789821728512 3722098081 > 0 , P 4 ( 1 ) = 256383177 > 0. P_4(16/247)=\frac{1173789821728512}{3722098081}>0,\qquad P_4(1)=256383177>0. P 4 ( 16/247 ) = 3722098081 1173789821728512 > 0 , P 4 ( 1 ) = 256383177 > 0.
Concavity therefore proves y < k 0 β y<k_0\beta y < k 0 β on the entire closed interval.
Mixed signs: a triangular comparison region.
Put u 0 = 1 − b / 2 ∈ [ 1 / 2 , 1 ) u_0=1-b/2\in[1/2,1) u 0 = 1 − b /2 ∈ [ 1/2 , 1 ) . The actual chart bound gives D > 4 u 0 D>4u_0 D > 4 u 0 . Using (B.41) and d A < 2 α d_A<2\alpha d A < 2 α yields
α > u 0 t , 0 < κ < K 0 : = 2 ( α − u 0 t ) 3 , \alpha>u_0t,\qquad0<\kappa<K_0:=\frac{2(\alpha-u_0t)}{3}, α > u 0 t , 0 < κ < K 0 := 3 2 ( α − u 0 t ) ,
L κ : = u 0 t 2 3 + κ t 2 < x < X κ : = α 2 3 u 0 − κ α 2 u 0 . L_\kappa:=\frac{u_0t^2}{3}+\frac{\kappa t}{2}<x<X_\kappa:=\frac{\alpha^2}{3u_0}-\frac{\kappa\alpha}{2u_0}. L κ := 3 u 0 t 2 + 2 κ t < x < X κ := 3 u 0 α 2 − 2 u 0 κ α .
The bounds fit exactly:
X κ − L κ = ( α + u 0 t ) ( K 0 − κ ) 2 u 0 , X K 0 = L K 0 = α t 3 . X_\kappa-L_\kappa=\frac{(\alpha+u_0t)(K_0-\kappa)}{2u_0},\qquad X_{K_0}=L_{K_0}=\frac{\alpha t}{3}. X κ − L κ = 2 u 0 ( α + u 0 t ) ( K 0 − κ ) , X K 0 = L K 0 = 3 α t .
All compared endpoints are nonnegative. Furthermore P ( E − 1 ) < 2 U − b P(E-1)<2U_-b P ( E − 1 ) < 2 U − b , P < a m b P<a_mb P < a m b , and D > 4 − b ( 1 + 8 U − / 3 ) ≥ ( 9 − 8 U − ) / 3 D>4-b(1+8U_-/3)\ge(9-8U_-)/3 D > 4 − b ( 1 + 8 U − /3 ) ≥ ( 9 − 8 U − ) /3 imply
4 − P D < R 0 : = 1 + 6 U − 9 − 8 U − = 707 578 . \frac{4-P}{D}<R_0:=1+\frac{6U_-}{9-8U_-}=\frac{707}{578}. D 4 − P < R 0 := 1 + 9 − 8 U − 6 U − = 578 707 .
Equation (B.42), with P ≥ b P\ge b P ≥ b and r s ≤ 0 rs\le0 r s ≤ 0 , gives
Rem ≥ γ b t 2 , γ = 4 − 5 U − / 3 4 = 357 400 . \operatorname{Rem}\ge\gamma bt^2,\qquad\gamma=\frac{4-5U_-/3}{4}=\frac{357}{400}. Rem ≥ γ b t 2 , γ = 4 4 − 5 U − /3 = 400 357 .
Keeping the beneficial term s x sx s x in the determinant gives
3 4 − C ≤ ( t + x + k 0 β ) 2 + κ ( k 0 β + R 0 t ) − s x − γ b t 2 . (334) \frac{3}{4}-\mathcal{C}\le(t+x+k_0\beta)^2+\kappa(k_0\beta+R_0t)-sx-\gamma bt^2. \tag*{(334)} 4 3 − C ≤ ( t + x + k 0 β ) 2 + κ ( k 0 β + R 0 t ) − s x − γ b t 2 . (334)
Replacement of y y y is valid because the preceding loss has derivative 2 ( t + x + y ) + κ > 0 2(t+x+y)+\kappa>0 2 ( t + x + y ) + κ > 0 in y y y . At fixed κ \kappa κ the displayed expression is convex in x x x ; on either affine side x = L κ , X κ x=L_\kappa,X_\kappa x = L κ , X κ it is convex in κ \kappa κ . Applying the convex chord inequality twice reduces it to the three points
( 0 , u 0 t 2 / 3 ) , ( 0 , α 2 / ( 3 u 0 ) ) , ( K 0 , α t / 3 ) . (335) (0,u_0t^2/3),\qquad(0,\alpha^2/(3u_0)),\qquad(K_0,\alpha t/3). \tag*{(335)} ( 0 , u 0 t 2 /3 ) , ( 0 , α 2 / ( 3 u 0 )) , ( K 0 , α t /3 ) . (335)
This is a scalar comparison, not an assertion that these points are physical trajectories.
Exact allocation of the two leaf costs
Equation (B.43), Q ≥ 1 + b Q\ge1+b Q ≥ 1 + b , D > 4 u 0 D>4u_0 D > 4 u 0 , and ( 2 Q − E P ) / 2 ≥ 1 + b / 2 − b A c (2Q-EP)/2\ge1+b/2-bA_c ( 2 Q − E P ) /2 ≥ 1 + b /2 − b A c give A c ≥ g A_c\ge g A c ≥ g , where
g = ( 1 + b / 2 ) t + ( 1 + b ) u 0 t 2 + ( 1 + b ) ( 2 − b ) t 3 / 6 1 + b t , H = g + u 0 t . g=\frac{(1+b/2)t+(1+b)u_0t^2+(1+b)(2-b)t^3/6}{1+bt},\qquad H=g+u_0t. g = 1 + b t ( 1 + b /2 ) t + ( 1 + b ) u 0 t 2 + ( 1 + b ) ( 2 − b ) t 3 /6 , H = g + u 0 t .
The following two identities have nonnegative right sides:
( 1 + b t ) ( g − t ) = b t / 2 + ( 1 − b ) ( 1 + b / 2 ) t 2 + ( 1 + b ) ( 2 − b ) 2 t 3 / 6 , (1+bt)(g-t)=bt/2+(1-b)(1+b/2)t^2+(1+b)(2-b)^2t^3/6, ( 1 + b t ) ( g − t ) = b t /2 + ( 1 − b ) ( 1 + b /2 ) t 2 + ( 1 + b ) ( 2 − b ) 2 t 3 /6 ,
H − ( 2 t − t 2 / 2 ) = t 2 1 + b t [ ( 1 − b ) ( b + 3 / 2 ) + { b / 2 + ( 1 + b ) ( 2 − b ) 2 / 6 } t ] . H-(2t-t^2/2)=\frac{t^2}{1+bt}\left[(1-b)(b+3/2)+\{b/2+(1+b)(2-b)^2/6\}t\right]. H − ( 2 t − t 2 /2 ) = 1 + b t t 2 [ ( 1 − b ) ( b + 3/2 ) + { b /2 + ( 1 + b ) ( 2 − b ) 2 /6 } t ] .
Since H < A H<A H < A , first t < 2 A / 3 t<2A/3 t < 2 A /3 and then
t < A 2 − A / 3 ≤ T 0 : = 43 / 319. (336) t<\frac{A}{2-A/3}\le T_0:=43/319. \tag*{(336)} t < 2 − A /3 A ≤ T 0 := 43/319. (336)
Now fix B = α + β B=\alpha+\beta B = α + β and decrease α \alpha α to u 0 t u_0t u 0 t , keeping β = B − α ≥ 0 \beta=B-\alpha\ge0 β = B − α ≥ 0 . Each of the three losses in (B.45) decreases strictly as α \alpha α increases. For the first two corners the derivatives are respectively
− 2 k 0 ( t + u 0 t 2 / 3 + k 0 β ) , 2 W ( 2 α / ( 3 u 0 ) − k 0 ) − 2 s α / ( 3 u 0 ) , -2k_0(t+u_0t^2/3+k_0\beta),\qquad2W(2\alpha/(3u_0)-k_0)-2s\alpha/(3u_0), − 2 k 0 ( t + u 0 t 2 /3 + k 0 β ) , 2 W ( 2 α / ( 3 u 0 ) − k 0 ) − 2 s α / ( 3 u 0 ) ,
where W = t + α 2 / ( 3 u 0 ) + k 0 β > 0 W=t+\alpha^2/(3u_0)+k_0\beta>0 W = t + α 2 / ( 3 u 0 ) + k 0 β > 0 . Both are negative since 2 α / ( 3 u 0 ) ≤ 4 U − / 3 < k 0 2\alpha/(3u_0)\le4U_-/3<k_0 2 α / ( 3 u 0 ) ≤ 4 U − /3 < k 0 . For the third put ν = t / 3 \nu=t/3 ν = t /3 , κ = 2 ( α − u 0 t ) / 3 \kappa=2(\alpha-u_0t)/3 κ = 2 ( α − u 0 t ) /3 and β = B − α \beta=B-\alpha β = B − α . Its derivative is
2 t ( ν − k 0 + R 0 / 3 ) + 2 α ν ( ν − k 0 ) + 2 k 0 β ( ν − k 0 + 1 / 3 ) − k 0 κ − s ν < 0. 2t(\nu-k_0+R_0/3)+2\alpha\nu(\nu-k_0)+2k_0\beta(\nu-k_0+1/3)-k_0\kappa-s\nu<0. 2 t ( ν − k 0 + R 0 /3 ) + 2 α ν ( ν − k 0 ) + 2 k 0 β ( ν − k 0 + 1/3 ) − k 0 κ − s ν < 0.
Every term is nonpositive and the first is negative: use α , β , κ , s ≥ 0 \alpha,\beta,\kappa,s\ge0 α , β , κ , s ≥ 0 , ν < T 0 / 3 < k 0 − 1 / 3 \nu<T_0/3<k_0-1/3 ν < T 0 /3 < k 0 − 1/3 and k 0 > ( T 0 + R 0 ) / 3 k_0>(T_0+R_0)/3 k 0 > ( T 0 + R 0 ) /3 . At α = u 0 t \alpha=u_0t α = u 0 t the three corners coincide. Increasing B B B to A − g A-g A − g is then legitimate, since A − H > 0 A-H>0 A − H > 0 and the root of the square is positive. We obtain
3 4 − C < W 0 2 − s u 0 t 2 / 3 − γ b t 2 , W 0 = t + u 0 t 2 / 3 + k 0 ( A − H ) > 0. \frac{3}{4}-\mathcal{C}<W_0^2-su_0t^2/3-\gamma bt^2,\qquad W_0=t+u_0t^2/3+k_0(A-H)>0. 4 3 − C < W 0 2 − s u 0 t 2 /3 − γ b t 2 , W 0 = t + u 0 t 2 /3 + k 0 ( A − H ) > 0.
Set q 0 = 1 − 2 k 0 = 17 / 250 q_0=1-2k_0=17/250 q 0 = 1 − 2 k 0 = 17/250 and c 0 = ( 1 + 3 k 0 ) / 6 = 1199 / 3000 c_0=(1+3k_0)/6=1199/3000 c 0 = ( 1 + 3 k 0 ) /6 = 1199/3000 . Direct subtraction shows W 0 ≤ k 0 A + q 0 t + c 0 t 2 W_0\le k_0A+q_0t+c_0t^2 W 0 ≤ k 0 A + q 0 t + c 0 t 2 : after division of this gap by t 2 t^2 t 2 and multiplication by 1 + b t 1+bt 1 + b t , the result is
( 1 − b ) ( 6 k 0 b + 9 k 0 − 1 ) 6 + t [ b ( 3 k 0 − 1 + b ) 6 + k 0 ( 1 + b ) ( 2 − b ) 2 6 ] ≥ 0. \frac{(1-b)(6k_0b+9k_0-1)}{6}
+t\left[\frac{b(3k_0-1+b)}{6}+\frac{k_0(1+b)(2-b)^2}{6}\right]\ge0. 6 ( 1 − b ) ( 6 k 0 b + 9 k 0 − 1 ) + t [ 6 b ( 3 k 0 − 1 + b ) + 6 k 0 ( 1 + b ) ( 2 − b ) 2 ] ≥ 0.
Also s u 0 / 3 + γ b − 1 / 3 = b ( γ − 1 / 2 + b / 6 ) ≥ 0 su_0/3+\gamma b-1/3=b(\gamma-1/2+b/6)\ge0 s u 0 /3 + γ b − 1/3 = b ( γ − 1/2 + b /6 ) ≥ 0 . The square roots being positive, we may square this bound. Define
M 0 = 1 3 − q 0 2 − 2 k 0 U − c 0 − 2 q 0 c 0 T 0 − c 0 2 T 0 2 = 105198654743 473062500000 > 0. M_0=\frac{1}{3}-q_0^2-2k_0U_-c_0-2q_0c_0T_0-c_0^2T_0^2
=\frac{105198654743}{473062500000}>0. M 0 = 3 1 − q 0 2 − 2 k 0 U − c 0 − 2 q 0 c 0 T 0 − c 0 2 T 0 2 = 473062500000 105198654743 > 0.
Expansion and completion of the square imply
3 4 − C < ( k 0 A ) 2 + 2 k 0 A q 0 t − M 0 t 2 ≤ ( k 0 A ) 2 ( 1 + q 0 2 / M 0 ) ≤ L m : = 97015095513566129007 6574915921437500000000 . \frac{3}{4}-\mathcal{C}<(k_0A)^2+2k_0Aq_0t-M_0t^2
\le(k_0A)^2(1+q_0^2/M_0)\le L_m:=\frac{97015095513566129007}{6574915921437500000000}. 4 3 − C < ( k 0 A ) 2 + 2 k 0 A q 0 t − M 0 t 2 ≤ ( k 0 A ) 2 ( 1 + q 0 2 / M 0 ) ≤ L m := 6574915921437500000000 97015095513566129007 .
The decisive rational margin is
( 3 / 4 − H ∗ ) − L m = 3527766575724023280412656655948597000737 3839752610353803304657739339252804375000000000 > 0. (3/4-H_\ast)-L_m=\frac{3527766575724023280412656655948597000737}{3839752610353803304657739339252804375000000000}>0. ( 3/4 − H ∗ ) − L m = 3839752610353803304657739339252804375000000000 3527766575724023280412656655948597000737 > 0.
This proves the mixed case, including s = 0 s=0 s = 0 .
Both coordinates negative.
Write r = − κ r=-\kappa r = − κ , s = − λ s=-\lambda s = − λ , so 1 < a , b < a m 1<a,b<a_m 1 < a , b < a m . In (B.42), use P ≥ 1 P\ge1 P ≥ 1 , P ≤ a m 2 P\le a_m^2 P ≤ a m 2 , r s ≤ ( a m − 1 ) 2 rs\le(a_m-1)^2 r s ≤ ( a m − 1 ) 2 and P ( E − 1 ) / 2 ≤ U − a m P(E-1)/2\le U-a_m P ( E − 1 ) /2 ≤ U − a m . It follows that
Rem ≥ 411807 500000 t 2 > 2 3 t 2 . \operatorname{Rem}\ge\frac{411807}{500000}t^2>\frac{2}{3}t^2. Rem ≥ 500000 411807 t 2 > 3 2 t 2 .
For negative w w w , N ( w , d ) N(w,d) N ( w , d ) increases in d d d and decreases in w w w . At w = − U − w=-U_- w = − U − and d = 2 U − d=2U_- d = 2 U − it equals 2 n 0 2n_0 2 n 0 , where
n 0 = U − ( 1 + U − ) 3 ( 1 − U − ) = 27047 185500 . n_0=\frac{U_-(1+U_-)}{3(1-U_-)}=\frac{27047}{185500}. n 0 = 3 ( 1 − U − ) U − ( 1 + U − ) = 185500 27047 .
Since J A , J B > 2 J_A,J_B>2 J A , J B > 2 , we have x ≤ n 0 α x\le n_0\alpha x ≤ n 0 α , y ≤ n 0 β y\le n_0\beta y ≤ n 0 β . The function a + b − a b / 2 − U − min ( a , b ) a+b-ab/2-U_-\min(a,b) a + b − ab /2 − U − min ( a , b ) increases separately on either ordered half of [ 1 , a m ] 2 [1,a_m]^2 [ 1 , a m ] 2 : its derivatives are at least 1 / 2 − 4 U − / 3 1/2-4U_-/3 1/2 − 4 U − /3 and 1 / 2 − U − / 3 1/2-U_-/3 1/2 − U − /3 , both positive. Thus
A c ≥ c 1 t , c 1 = 3 / 2 − U − = 621 500 . A_c\ge c_1t,\qquad c_1=3/2-U_-=\frac{621}{500}. A c ≥ c 1 t , c 1 = 3/2 − U − = 500 621 .
Both compulsory debits d A , d B ≥ z + 3 κ , z + 3 λ d_A,d_B\ge z+3\kappa,z+3\lambda d A , d B ≥ z + 3 κ , z + 3 λ give B : = α + β > D t / 2 > ℓ 1 t B:=\alpha+\beta>Dt/2>\ell_1t B := α + β > D t /2 > ℓ 1 t , with ℓ 1 = 1 − 2 U − / 3 = 207 / 250 \ell_1=1-2U_-/3=207/250 ℓ 1 = 1 − 2 U − /3 = 207/250 . Hence
t < A c 1 + ℓ 1 , B ≤ A − c 1 t . t<\frac{A}{c_1+\ell_1},\qquad B\le A-c_1t. t < c 1 + ℓ 1 A , B ≤ A − c 1 t .
Discard the positive r s rs r s term, retain 2 t 2 / 3 2t^2/3 2 t 2 /3 of the remainder, and use κ < 2 α / 3 \kappa<2\alpha/3 κ < 2 α /3 , λ < 2 β / 3 \lambda<2\beta/3 λ < 2 β /3 . This gives
3 4 − C ≤ ( t + n 0 B ) 2 + n 0 3 B 2 − 2 3 t 2 . \frac{3}{4}-\mathcal{C}\le(t+n_0B)^2+\frac{n_0}{3}B^2-\frac{2}{3}t^2. 4 3 − C ≤ ( t + n 0 B ) 2 + 3 n 0 B 2 − 3 2 t 2 .
It increases in B ≥ 0 B\ge0 B ≥ 0 . After the residual substitution it is
Q A ( t ) = q 2 t 2 + 2 n 0 A h 1 t + ( n 0 2 + n 0 / 3 ) A 2 , Q_A(t)=q_2t^2+2n_0Ah_1t+(n_0^2+n_0/3)A^2, Q A ( t ) = q 2 t 2 + 2 n 0 A h 1 t + ( n 0 2 + n 0 /3 ) A 2 ,
where
q 2 = ( 1 − n 0 c 1 ) 2 + n 0 c 1 2 / 3 − 2 / 3 = 2036665287275407 25807687500000000 > 0 , q_2=(1-n_0c_1)^2+n_0c_1^2/3-2/3
=\frac{2036665287275407}{25807687500000000}>0, q 2 = ( 1 − n 0 c 1 ) 2 + n 0 c 1 2 /3 − 2/3 = 25807687500000000 2036665287275407 > 0 ,
h 1 = 1 − c 1 ( n 0 + 1 / 3 ) = 37555313 92750000 > 0. h_1=1-c_1(n_0+1/3)=\frac{37555313}{92750000}>0. h 1 = 1 − c 1 ( n 0 + 1/3 ) = 92750000 37555313 > 0.
Thus Q A Q_A Q A increases for t ≥ 0 t\ge0 t ≥ 0 . At t = A / ( c 1 + ℓ 1 ) t=A/(c_1+\ell_1) t = A / ( c 1 + ℓ 1 ) the residual equals 2 A / 5 > 0 2A/5>0 2 A /5 > 0 . Homogeneity yields
3 4 − C < Q A ( A c 1 + ℓ 1 ) ≤ L b : = 7428370561341219727 767940001171875000000 , \frac{3}{4}-\mathcal{C}<Q_A\left(\frac{A}{c_1+\ell_1}\right)\le L_b:=\frac{7428370561341219727}{767940001171875000000}, 4 3 − C < Q A ( c 1 + ℓ 1 A ) ≤ L b := 767940001171875000000 7428370561341219727 ,
and
( 3 / 4 − H ∗ ) − L b = 38589144975363733283697549133345658036971433 7579264015334677252965623903889516796875000000 > 0. (3/4-H_\ast)-L_b
=\frac{38589144975363733283697549133345658036971433}{7579264015334677252965623903889516796875000000}>0. ( 3/4 − H ∗ ) − L b = 7579264015334677252965623903889516796875000000 38589144975363733283697549133345658036971433 > 0.
The opposite mixed orientation follows by scalar exchange after the two actual chart derivations. Together the cases exhaust every negative coordinate, and all strict inequalities survive F = A = U − F=A=U_- F = A = U − . □
The nonnegative region: analytic outer reduction
We now suppose r , s ≥ 0 r,s \ge0 r , s ≥ 0 , and sort the already derived scalar names so 0 < b ≤ a ≤ 1 0 < b \le a \le1 0 < b ≤ a ≤ 1 . In this subsection suppose F ≤ U F \le U F ≤ U and C ≤ H ∗ \mathcal{C} \le H_* C ≤ H ∗ . All comparison inequalities precede calibration. The width bound gives P ( E − 1 ) < 2 U b P(E-1) < 2Ub P ( E − 1 ) < 2 U b . Since P ≤ 1 P \le1 P ≤ 1 , R + P ≤ 1 R+P \le1 R + P ≤ 1 , r , s ≥ R r,s \ge R r , s ≥ R , and ( 4 − P ) / D ≥ 1 (4-P)/D \ge1 ( 4 − P ) / D ≥ 1 , the remainder in (B.36) is nonnegative and
C ≥ 3 / 4 + R h − h 2 , h = x + y + t . (337) \mathcal{C} \ge3/4 + Rh-h^2,\qquad h=x+y+t. \tag*{(337)} C ≥ 3/4 + R h − h 2 , h = x + y + t . (337)
The function N ( w , d ) N(w,d) N ( w , d ) increases in d d d , and N ( w , 1 − w ) = 2 ( 1 + 2 w ) / ( 3 ( 1 + w ) ) ≤ 1 N(w,1-w)=2(1+2w)/(3(1+w)) \le1 N ( w , 1 − w ) = 2 ( 1 + 2 w ) / ( 3 ( 1 + w )) ≤ 1 for 0 ≤ w ≤ 1 0 \le w \le1 0 ≤ w ≤ 1 . Therefore α ≥ J A x \alpha\ge J_Ax α ≥ J A x , β ≥ J B y \beta\ge J_By β ≥ J B y , and α + β ≥ J 0 ( x + y ) \alpha+\beta\ge J_0(x+y) α + β ≥ J 0 ( x + y ) , where J 0 = 2 Q − P J_0=2Q-P J 0 = 2 Q − P . Put
J = J 0 − 2 U b , D ∗ = 4 − P − 2 U b , κ 0 = Q ( 1 − b / 2 ) , V 0 = J 0 / 2 + U b . J=J_0-2Ub,\qquad D_*=4-P-2Ub,\qquad\kappa_0=Q(1-b/2),\qquad V_0=J_0/2+Ub. J = J 0 − 2 U b , D ∗ = 4 − P − 2 U b , κ 0 = Q ( 1 − b /2 ) , V 0 = J 0 /2 + U b .
Then D > D ∗ > 4 − 2 b > 0 D>D_*>4-2b>0 D > D ∗ > 4 − 2 b > 0 , 2 Q − E P > J > 0 2Q-EP>J>0 2 Q − E P > J > 0 , and expansion of the positive exponential tail gives, for every integer m ≥ 2 m\ge2 m ≥ 2 ,
A c ≥ J t / 2 + ∑ n = 2 m Q D ∗ n − 1 t n 2 n − 1 n ! . (338) A_c \ge Jt/2+\sum_{n=2}^{m}\frac{QD_*^{n-1}t^n}{2^{n-1}n!}. \tag*{(338)} A c ≥ J t /2 + n = 2 ∑ m 2 n − 1 n ! Q D ∗ n − 1 t n . (338)
In particular
F ≥ J ( x + y ) + J t / 2 + κ 0 t 2 , F ≥ J 0 h − V 0 t + κ 0 t 2 . (339) F \ge J(x+y)+Jt/2+\kappa_0t^2,\qquad F \ge J_0h-V_0t+\kappa_0t^2. \tag*{(339)} F ≥ J ( x + y ) + J t /2 + κ 0 t 2 , F ≥ J 0 h − V 0 t + κ 0 t 2 . (339)
Compactness and the scalar envelopes
The original compact rate bound k ≤ 15 k\le15 k ≤ 15 gives E P = 4 / ( 1 + k 2 ) ≥ 2 / 113 EP=4/(1+k^2)\ge2/113 E P = 4/ ( 1 + k 2 ) ≥ 2/113 . Moreover, along the central ray,
ℓ / k = ( 1 + k 2 ) ξ / 2 − 1 + 2 / ξ k 2 ≥ 2 1 + k 2 − 1 k 2 > 4 31 . \ell/k=\frac{(1+k^2)\xi/2-1+2/\xi}{k^2}\ge\frac{2\sqrt{1+k^2}-1}{k^2}>\frac{4}{31}. ℓ / k = k 2 ( 1 + k 2 ) ξ /2 − 1 + 2/ ξ ≥ k 2 2 1 + k 2 − 1 > 31 4 .
For the last comparison the expression decreases with 1 + k 2 \sqrt{1+k^2} 1 + k 2 ; at k = 15 k=15 k = 15 it is equivalent to 62 226 > 931 62\sqrt{226}>931 62 226 > 931 , whose squared gap is 1983. Integration of d log ξ = k d t d\log\xi=k\,dt d log ξ = k d t gives z < 31 A c / 4 z<31A_c/4 z < 31 A c /4 . By (A.3), e z 0 < E 6 ( z 0 ) < 27 / 4 e^{z_0}<\mathcal{E}_6(z_0)<27/4 e z 0 < E 6 ( z 0 ) < 27/4 for z 0 = 1519 / 800 z_0=1519/800 z 0 = 1519/800 , with exact gap
66551571865629146457223 921257902080000000000000 . \frac{66551571865629146457223}{921257902080000000000000}. 921257902080000000000000 66551571865629146457223 .
It follows that P > p 0 : = 8 / 3051 P>p_0:=8/3051 P > p 0 := 8/3051 and Q > 1 / 10 Q>1/10 Q > 1/10 , since 4 p 0 − 1 / 100 = 149 / 305100 > 0 4p_0-1/100=149/305100>0 4 p 0 − 1/100 = 149/305100 > 0 . Put h 0 = 243 / 2000 h_0=243/2000 h 0 = 243/2000 , L 0 = h 0 2 L_0=h_0^2 L 0 = h 0 2 . The exact gap
3 / 4 − H ∗ − L 0 = 89765086136957463911195871 39478417604337498218323684000000 > 0 3/4-H_*-L_0=\frac{89765086136957463911195871}{39478417604337498218323684000000}>0 3/4 − H ∗ − L 0 = 39478417604337498218323684000000 89765086136957463911195871 > 0
and (B.47) imply
h > H : = R + R 2 + 4 L 0 2 > R . h>H:=\frac{R+\sqrt{R^2+4L_0}}{2}>R. h > H := 2 R + R 2 + 4 L 0 > R .
The increasing lower bound
W ( R ) = R + L 0 R R 2 + L 0 < H W(R)=R+\frac{L_0R}{R^2+L_0}<H W ( R ) = R + R 2 + L 0 L 0 R < H
has residual − L 0 3 / ( R 2 + L 0 ) 2 -L_0^3/(R^2+L_0)^2 − L 0 3 / ( R 2 + L 0 ) 2 in X 2 − R X − L 0 X^2-RX-L_0 X 2 − R X − L 0 ; its derivative has positive numerator R 4 + L 0 R 2 + 2 L 0 2 R^4+L_0R^2+2L_0^2 R 4 + L 0 R 2 + 2 L 0 2 . Two further global support lines are
H ≥ 4 5 ( R + h 0 ) , H ≥ 49 50 R + 7 25 h 0 . H\ge\frac{4}{5}(R+h_0),\qquad H\ge\frac{49}{50}R+\frac{7}{25}h_0. H ≥ 5 4 ( R + h 0 ) , H ≥ 50 49 R + 25 7 h 0 .
Their residuals are − ( 2 R − 3 h 0 ) 2 / 25 -(2R-3h_0)^2/25 − ( 2 R − 3 h 0 ) 2 /25 and − ( 7 R − 48 h 0 ) 2 / 2500 -(7R-48h_0)^2/2500 − ( 7 R − 48 h 0 ) 2 /2500 , respectively; both lines are positive. The further global bound H ≥ Z ( R ) : = R / 2 + h 0 ( 16 h 0 2 + 3 R 2 ) / ( 16 h 0 2 + R 2 ) H\ge Z(R):=R/2+h_0(16h_0^2+3R^2)/(16h_0^2+R^2) H ≥ Z ( R ) := R /2 + h 0 ( 16 h 0 2 + 3 R 2 ) / ( 16 h 0 2 + R 2 ) is retained: its residual is − R 6 / [ 4 ( 16 h 0 2 + R 2 ) 2 ] -R^6/[4(16h_0^2+R^2)^2] − R 6 / [ 4 ( 16 h 0 2 + R 2 ) 2 ] . Completing the square in the first bound of (B.49), over the full interval 0 ≤ t ≤ h 0\le t\le h 0 ≤ t ≤ h , gives
F ≥ g 0 ( h ) , g 0 ( v ) = { J v / 2 + κ 0 v 2 , v ≤ J / ( 4 κ 0 ) , J v − J 2 / ( 16 κ 0 ) , v ≥ J / ( 4 κ 0 ) . (340) F\ge g_0(h),\qquad
g_0(v)=
\begin{cases}
Jv/2+\kappa_0v^2, & v\le J/(4\kappa_0),\\
Jv-J^2/(16\kappa_0), & v\ge J/(4\kappa_0).
\end{cases} \tag*{(340)} F ≥ g 0 ( h ) , g 0 ( v ) = { J v /2 + κ 0 v 2 , J v − J 2 / ( 16 κ 0 ) , v ≤ J / ( 4 κ 0 ) , v ≥ J / ( 4 κ 0 ) . (340)
It is strictly increasing and its branches agree. Hence F > g 0 ( H ) F>g_0(H) F > g 0 ( H ) . Useful ratios are
2 − U ≤ J / κ 0 ≤ 3 − 2 U , J 0 ≥ 3 b , J 0 ≥ Q ( 2 − Q / 4 ) , κ 0 ≥ Q ( 1 − Q / 4 ) . (341) 2-U\leq J/\kappa_0\leq3-2U,\qquad J_0\geq3b,\qquad J_0\geq Q(2-Q/4),\qquad\kappa_0\geq Q(1-Q/4). \tag*{(341)} 2 − U ≤ J / κ 0 ≤ 3 − 2 U , J 0 ≥ 3 b , J 0 ≥ Q ( 2 − Q /4 ) , κ 0 ≥ Q ( 1 − Q /4 ) . (341)
For the lower ratio, J − ( 2 − U ) κ 0 = U ( a − b ) − U a b / 2 + ( 1 − U / 2 ) b 2 J-(2-U)\kappa_0=U(a-b)-Uab/2+(1-U/2)b^2 J − ( 2 − U ) κ 0 = U ( a − b ) − U ab /2 + ( 1 − U /2 ) b 2 increases in a a a and equals ( 1 − U ) b 2 (1-U)b^2 ( 1 − U ) b 2 at a = b a=b a = b . For the upper ratio use J − 2 κ 0 = b ( b − 2 U ) J-2\kappa_0=b(b-2U) J − 2 κ 0 = b ( b − 2 U ) and κ 0 ≥ b ( 2 − b ) \kappa_0\geq b(2-b) κ 0 ≥ b ( 2 − b ) . The other comparisons follow from a b ≤ Q 2 / 4 ab\leq Q^2/4 ab ≤ Q 2 /4 and b ≤ Q / 2 b\leq Q/2 b ≤ Q /2 .
B.1.6.11. The remaining outer inequalities. For 1 / 5 ≤ Q ≤ 9 / 16 1/5\leq Q\leq9/16 1/5 ≤ Q ≤ 9/16 set ρ = 1 − Q + p 0 \rho=1-Q+p_0 ρ = 1 − Q + p 0 , H d = ρ 2 + L 0 H_d=\rho^2+L_0 H d = ρ 2 + L 0 , and H n = ρ ( ρ 2 + 2 L 0 ) H_n=\rho(\rho^2+2L_0) H n = ρ ( ρ 2 + 2 L 0 ) . Since R > ρ > 0 R>\rho>0 R > ρ > 0 , H > H n / H d H>H_n/H_d H > H n / H d . Also put d = 1 + p 0 + 2 h 0 / 7 > 1 d=1+p_0+2h_0/7>1 d = 1 + p 0 + 2 h 0 /7 > 1 . The second support line gives H > 49 ( d − Q ) / 50 H>49(d-Q)/50 H > 49 ( d − Q ) /50 . Square completion in the second bound of (B.49) therefore gives 4 ( 4 − Q ) ( F − U ) > M ( Q ) 4(4-Q)(F-U)>M(Q) 4 ( 4 − Q ) ( F − U ) > M ( Q ) , where
M ( Q ) = 49 50 Q ( 8 − Q ) ( 4 − Q ) ( d − Q ) − 4 Q ( 1 + U / 2 ) 2 − 4 U ( 4 − Q ) . (342) M(Q)=\frac{49}{50}Q(8-Q)(4-Q)(d-Q)-4Q(1+U/2)^2-4U(4-Q). \tag*{(342)} M ( Q ) = 50 49 Q ( 8 − Q ) ( 4 − Q ) ( d − Q ) − 4 Q ( 1 + U /2 ) 2 − 4 U ( 4 − Q ) . (342)
Here 50 M ′ ′ / 49 = − 64 − 24 d + ( 72 + 6 d ) Q − 12 Q 2 ≤ − 353 / 8 < 0 50M''/49=-64-24d+(72+6d)Q-12Q^2\leq-353/8<0 50 M ′′ /49 = − 64 − 24 d + ( 72 + 6 d ) Q − 12 Q 2 ≤ − 353/8 < 0 . Thus its minimum is at an endpoint, where
M ( 1 / 5 ) = 3367552463 25425000000 > 0 , M ( 9 / 16 ) = 2712290783 5554176000 > 0. M(1/5)=\frac{3367552463}{25425000000}>0,\qquad M(9/16)=\frac{2712290783}{5554176000}>0. M ( 1/5 ) = 25425000000 3367552463 > 0 , M ( 9/16 ) = 5554176000 2712290783 > 0.
This excludes the medium-Q Q Q region.
For Q ≥ 9 / 16 Q\geq9/16 Q ≥ 9/16 in the linear branch, H ≥ H b = ( 2 − U ) / 4 = 351 / 800 H\geq H_b=(2-U)/4=351/800 H ≥ H b = ( 2 − U ) /4 = 351/800 . The identity R = H − L 0 / H R=H-L_0/H R = H − L 0 / H gives R > 2 / 5 R>2/5 R > 2/5 , since H b − L 0 / H b − 2 / 5 = 1327 / 260000 > 0 H_b-L_0/H_b-2/5=1327/260000>0 H b − L 0 / H b − 2/5 = 1327/260000 > 0 . Since R ≤ ( 1 − b ) 2 R\leq(1-b)^2 R ≤ ( 1 − b ) 2 , we have b < 2 / 5 b<2/5 b < 2/5 and
9 4 κ 0 − J 0 = ( a − b ) ( 1 − b / 2 ) + b ( 2 − 5 b ) 4 ≥ 0 , V 0 / J 0 ≤ δ 0 : = 1 / 2 + U / 3 = 349 / 600. \frac{9}{4}\kappa_0-J_0=\frac{(a-b)(1-b/2)+b(2-5b)}{4}\geq0,\qquad V_0/J_0\leq\delta_0:=1/2+U/3=349/600. 4 9 κ 0 − J 0 = 4 ( a − b ) ( 1 − b /2 ) + b ( 2 − 5 b ) ≥ 0 , V 0 / J 0 ≤ δ 0 := 1/2 + U /3 = 349/600.
Square completion yields F > J 0 ( H − 9 δ 0 2 / 16 ) F>J_0(H-9\delta_0^2/16) F > J 0 ( H − 9 δ 0 2 /16 ) . Here J 0 ≥ 1071 / 1024 J_0\geq1071/1024 J 0 ≥ 1071/1024 , and
1071 1024 ( 351 800 − 121801 640000 ) − U = 9724729 655360000 > 0. \frac{1071}{1024}\left(\frac{351}{800}-\frac{121801}{640000}\right)-U=\frac{9724729}{655360000}>0. 1024 1071 ( 800 351 − 640000 121801 ) − U = 655360000 9724729 > 0.
This excludes the linear branch, including its seam.
For 1 / 10 ≤ Q ≤ 1 / 5 1/10\leq Q\leq1/5 1/10 ≤ Q ≤ 1/5 , the actual compactness correlation gives
J B ≥ J A ≥ 2 a + 4 113 a − 2 113 ≥ 4 2 113 − 2 113 > λ 0 : = 257 / 500. J_B\geq J_A\geq2a+\frac{4}{113a}-\frac{2}{113}\geq\frac{4\sqrt{2}}{113}-\frac{2}{113}>\lambda_0:=257/500. J B ≥ J A ≥ 2 a + 113 a 4 − 113 2 ≥ 113 4 2 − 113 2 > λ 0 := 257/500.
The squared margin for the last comparison is 1538319 / 3192250000 > 0 1538319/3192250000>0 1538319/3192250000 > 0 . Thus α + β ≥ λ 0 ( x + y ) \alpha+\beta\geq\lambda_0(x+y) α + β ≥ λ 0 ( x + y ) . Define
y K = 8 − 113 Q 2 8 + 113 Q 2 , τ 0 = y K + y K 3 / 3 + y K 5 / 5 , y_K=\frac{8-113Q^2}{8+113Q^2},\qquad\tau_0=y_K+y_K^3/3+y_K^5/5, y K = 8 + 113 Q 2 8 − 113 Q 2 , τ 0 = y K + y K 3 /3 + y K 5 /5 ,
D Q = 4 − Q 2 / 4 − U Q , c Q = Q ( 2 − U − Q / 4 ) / 2 , D_Q=4-Q^2/4-UQ,\qquad c_Q=Q(2-U-Q/4)/2, D Q = 4 − Q 2 /4 − U Q , c Q = Q ( 2 − U − Q /4 ) /2 ,
C 6 ( Q , t ) = c Q t + ∑ n = 2 6 Q D Q n − 1 t n 2 n − 1 n ! . C_6(Q,t)=c_Qt+\sum_{n=2}^{6}\frac{QD_Q^{n-1}t^n}{2^{n-1}n!}. C 6 ( Q , t ) = c Q t + n = 2 ∑ 6 2 n − 1 n ! Q D Q n − 1 t n .
Since E ≥ 8 / ( 113 Q 2 ) E\geq8/(113Q^2) E ≥ 8/ ( 113 Q 2 ) and D < 4 D<4 D < 4 , we have t > log ( E ) / 2 ≥ arctanh ( y K ) > τ 0 t>\log(E)/2\geq\operatorname{arctanh}(y_K)>\tau_0 t > log ( E ) /2 ≥ arctanh ( y K ) > τ 0 . Also D > 4 − Q ≥ 19 / 5 D>4-Q\geq19/5 D > 4 − Q ≥ 19/5 and z < 31 U / 4 z<31U/4 z < 31 U /4 give t < T max = 1519 / 1520 t<T_{\max}=1519/1520 t < T m a x = 1519/1520 . The function τ 0 \tau_0 τ 0 decreases in Q Q Q , and
T max − τ 0 ( 1 / 10 ) = 54593043297215119 964267380585605360 > 0. T_{\max}-\tau_0(1/10)=\frac{54593043297215119}{964267380585605360}>0. T m a x − τ 0 ( 1/10 ) = 964267380585605360 54593043297215119 > 0.
Write t = τ 0 + ( T max − τ 0 ) θ t=\tau_0+(T_{\max}-\tau_0)\theta t = τ 0 + ( T m a x − τ 0 ) θ , 0 ≤ θ ≤ 1 0\leq\theta\leq1 0 ≤ θ ≤ 1 . Since J / 2 ≥ c Q J/2\geq c_Q J /2 ≥ c Q and D ∗ ≥ D Q D_*\geq D_Q D ∗ ≥ D Q , equation (B.48) gives F > λ 0 H n / H d + C 6 ( Q , t ) − λ 0 t F>\lambda_0H_n/H_d+C_6(Q,t)-\lambda_0t F > λ 0 H n / H d + C 6 ( Q , t ) − λ 0 t . For an explicit polynomial clearing, put
y N = 8 − 113 Q 2 , y D = 8 + 113 Q 2 , T d = 15 y D 5 , T n = 15 y N y D 4 + 5 y N 3 y D 2 + 3 y N 5 , y_N=8-113Q^2,\qquad y_D=8+113Q^2,\qquad T_d=15y_D^5,\qquad T_n=15y_Ny_D^4+5y_N^3y_D^2+3y_N^5, y N = 8 − 113 Q 2 , y D = 8 + 113 Q 2 , T d = 15 y D 5 , T n = 15 y N y D 4 + 5 y N 3 y D 2 + 3 y N 5 ,
t n = T n + ( T max T d − T n ) θ . t_n=T_n+(T_{\max}T_d-T_n)\theta. t n = T n + ( T m a x T d − T n ) θ .
The polynomial
H d T d 6 [ C 6 ( Q , t n / T d ) − U ] + λ 0 H n T d 6 − λ 0 H d t n T d 5 (343) H_dT_d^6[C_6(Q,t_n/T_d)-U]+\lambda_0H_nT_d^6-\lambda_0H_dt_nT_d^5 \tag*{(343)} H d T d 6 [ C 6 ( Q , t n / T d ) − U ] + λ 0 H n T d 6 − λ 0 H d t n T d 5 (343)
has 518 positive Bernstein coefficients, of bidegree ( 73 , 6 ) (73, 6) ( 73 , 6 ) , on [ 1 / 10 , 1 / 5 ] × [ 0 , 1 ] [1/10,1/5] \times[0,1] [ 1/10 , 1/5 ] × [ 0 , 1 ] . Its clearing denominator H d T d 6 H_dT_d^6 H d T d 6 is positive. This excludes the thin-Q Q Q region.
Finally suppose we are in the quadratic branch with R ≥ 1 / 5 R \ge1/5 R ≥ 1/5 . Then R < H ≤ J / ( 4 κ 0 ) ≤ 251 / 400 R < H \le J/(4\kappa_0) \le251/400 R < H ≤ J / ( 4 κ 0 ) ≤ 251/400 . At fixed R R R , write a = 1 − R / ( 1 − b ) a = 1 - R/(1-b) a = 1 − R / ( 1 − b ) , 0 ≤ b ≤ 1 − R 0 \le b \le1-\sqrt{R} 0 ≤ b ≤ 1 − R . The second derivative of J H / 2 + κ 0 H 2 JH/2+\kappa_0H^2 J H /2 + κ 0 H 2 in b b b is
H [ H ( R + ( 1 − b ) 3 ) + R ] ( b − 1 ) 3 < 0. \frac{H[H(R+(1-b)^3)+R]}{(b-1)^3}<0. ( b − 1 ) 3 H [ H ( R + ( 1 − b ) 3 ) + R ] < 0.
Its minimum is therefore at b = 0 b=0 b = 0 or a = b a=b a = b . At b = 0 b=0 b = 0 , use H 2 = R H + L 0 H^2=RH+L_0 H 2 = R H + L 0 and the first support line to obtain
( 1 − R ) ( H + H 2 ) ≥ f 0 ( R ) : = 4 5 ( 1 − R 2 ) ( R + h 0 ) + ( 1 − R ) L 0 . (344) (1-R)(H+H^2) \ge f_0(R) := \frac{4}{5}(1-R^2)(R+h_0)+(1-R)L_0. \tag*{(344)} ( 1 − R ) ( H + H 2 ) ≥ f 0 ( R ) := 5 4 ( 1 − R 2 ) ( R + h 0 ) + ( 1 − R ) L 0 . (344)
Since f 0 ′ ′ = − ( 24 R + 8 h 0 ) / 5 < 0 f_0''=-(24R+8h_0)/5<0 f 0 ′′ = − ( 24 R + 8 h 0 ) /5 < 0 , its minimum on [ 1 / 5 , 251 / 400 ] [1/5,251/400] [ 1/5 , 251/400 ] is at an endpoint. The endpoint excesses over U U U are 68609 / 5000000 68609/5000000 68609/5000000 and 198016309 / 1600000000 198016309/1600000000 198016309/1600000000 , respectively. At a = b a=b a = b , put w = R w=\sqrt{R} w = R , v = 1 − w v=1-w v = 1 − w , and A = ( 1 − w ) ( 2 − ( 1 − w ) / 2 − U ) + ( 1 − w 2 ) w 2 > 0 A=(1-w)(2-(1-w)/2-U)+(1-w^2)w^2>0 A = ( 1 − w ) ( 2 − ( 1 − w ) /2 − U ) + ( 1 − w 2 ) w 2 > 0 . The same identity and support line give
( 2 v − v 2 / 2 − U v ) H + ( 2 v − v 2 ) H 2 ≥ f d ( w ) : = 4 5 A ( w ) ( w 2 + h 0 ) + ( 1 − w 2 ) L 0 . (345) (2v-v^2/2-Uv)H+(2v-v^2)H^2 \ge f_d(w) := \frac{4}{5}A(w)(w^2+h_0)+(1-w^2)L_0. \tag*{(345)} ( 2 v − v 2 /2 − U v ) H + ( 2 v − v 2 ) H 2 ≥ f d ( w ) := 5 4 A ( w ) ( w 2 + h 0 ) + ( 1 − w 2 ) L 0 . (345)
On [ 447 / 1000 , 4 / 5 ] [447/1000,4/5] [ 447/1000 , 4/5 ] , f d ′ ′ ′ = − 3 ( 20000 w 3 − 1514 w + 755 ) / 625 < 0 f_d'''=-3(20000w^3-1514w+755)/625<0 f d ′′′ = − 3 ( 20000 w 3 − 1514 w + 755 ) /625 < 0 , since 20000 ( 447 / 1000 ) 2 > 1514 20000(447/1000)^2>1514 20000 ( 447/1000 ) 2 > 1514 . Moreover f d ′ ( 447 / 1000 ) = 321815473402479 / 625000000000000 > 0 f_d'(447/1000)=321815473402479/625000000000000>0 f d ′ ( 447/1000 ) = 321815473402479/625000000000000 > 0 . Concavity of f d ′ f_d' f d ′ and its positive left endpoint value preclude a negative-to-positive crossing; hence f d f_d f d has no interior minimum. Its endpoint excesses over U U U are, respectively,
22593992898613871 1250000000000000000 , 10231929 100000000 , \frac{22593992898613871}{1250000000000000000}, \qquad\frac{10231929}{100000000}, 1250000000000000000 22593992898613871 , 100000000 10231929 ,
both positive. This interval contains every required R \sqrt{R} R ; the squared endpoint gaps are 191 / 10 6 191/10^6 191/1 0 6 and 1 / 80 1/80 1/80 . All comparisons include their endpoints.
Lemma B.22 (Outer reduction). Under F ≤ 49 / 200 F \le49/200 F ≤ 49/200 , any nonnegative-coordinate point with C ≤ H ∗ \mathcal{C} \le H_\ast C ≤ H ∗ lies in
0 ≤ r < 9 / 20 , 1 / 30 < b ≤ a = 1 − r , 1 < E < 104 / 55 , R < 1 / 5. 0 \le r < 9/20,\qquad1/30 < b \le a=1-r,\qquad1<E<104/55,\qquad R<1/5. 0 ≤ r < 9/20 , 1/30 < b ≤ a = 1 − r , 1 < E < 104/55 , R < 1/5. It lies in the quadratic branch of (B.50).
Proof. The intervals above include all seams Q = 1 / 5 Q=1/5 Q = 1/5 , Q = 9 / 16 Q=9/16 Q = 9/16 , R = 1 / 5 R=1/5 R = 1/5 , and the linear/quadratic join. Thus the only region not excluded is quadratic with R < 1 / 5 R<1/5 R < 1/5 . Since ( 1 − a ) 2 ≤ R (1-a)^2\le R ( 1 − a ) 2 ≤ R and ( 1 − 11 / 20 ) 2 − 1 / 5 = 1 / 400 (1-11/20)^2-1/5=1/400 ( 1 − 11/20 ) 2 − 1/5 = 1/400 , we have a > 11 / 20 a>11/20 a > 11/20 . The actual cusp inequality is
b ≥ 16 a E 243 + 4 ( a E ) 2 . b \ge\frac{16aE}{243+4(aE)^2}. b ≥ 243 + 4 ( a E ) 2 16 a E . Its right side increases for 11 / 20 ≤ a E < 2 11/20\le aE<2 11/20 ≤ a E < 2 , and at 11 / 20 11/20 11/20 it exceeds 1 / 30 1/30 1/30 by 1979 / 732630 1979/732630 1979/732630 . Thus b > 1 / 30 b>1/30 b > 1/30 . Finally a ( E − 1 ) < 2 U a(E-1)<2U a ( E − 1 ) < 2 U gives E < 1 + 2 U / ( 11 / 20 ) = 104 / 55 E<1+2U/(11/20)=104/55 E < 1 + 2 U / ( 11/20 ) = 104/55 . □ \square □
The residual comparison and its certificate
The remaining step is computer-assisted. We give the analytic reduction, the exact meaning of every acceptance test, and a reproducible integer verification specification. The finite certificate is a cover of an auxiliary comparison domain, not a collection of physical samples. Put η = d B / b \eta=d_B/b η = d B / b and consider the closed rectangle
B = [ 0 , 9 / 20 ] × [ 1 / 30 , 1 ] × [ 1 , 104 / 55 ] × [ 0 , 1 ] (346) \mathcal{B}=[0,9/20]\times[1/30,1]\times[1,104/55]\times[0,1] \tag*{(346)} B = [ 0 , 9/20 ] × [ 1/30 , 1 ] × [ 1 , 104/55 ] × [ 0 , 1 ] (346)
in the order ( r , b , E , η ) (r,b,E,\eta) ( r , b , E , η ) . It contains every actual residual point. The zero-width faces are comparison faces only; the actual leaves remain strictly positive.
Continuous means and monotonicity
For w , d ≥ 0 w,d \ge0 w , d ≥ 0 , q = 1 + w q=1+w q = 1 + w , direct integration gives
e ( w , d ) = q ( 2 + d ) − 2 q ( q + d ) d log ( 1 + d / q ) = w d + 2 d 2 ∫ 0 1 v ( 1 − v ) q + d v d v . (347) e(w,d)=q(2+d)-\frac{2q(q+d)}{d}\log(1+d/q)=wd+2d^2\int_0^1\frac{v(1-v)}{q+dv}\,dv. \tag*{(347)} e ( w , d ) = q ( 2 + d ) − d 2 q ( q + d ) log ( 1 + d / q ) = w d + 2 d 2 ∫ 0 1 q + d v v ( 1 − v ) d v . (347)
with e ( w , 0 ) = 0 e(w,0)=0 e ( w , 0 ) = 0 . Consequently e ( w , d ) > w d e(w,d)>wd e ( w , d ) > w d for d > 0 d>0 d > 0 and the mean increases strictly with d d d . It is nondecreasing in w w w on 0 ≤ w , d ≤ 1 0\le w,d\le1 0 ≤ w , d ≤ 1 , because
∂ w e = d − 2 d 2 ∫ 0 1 v ( 1 − v ) ( q + d v ) 2 d v ≥ d − d 2 3 q 2 ≥ 0. \partial_w e=d-2d^2\int_0^1\frac{v(1-v)}{(q+dv)^2}\,dv\ge d-\frac{d^2}{3q^2}\ge0. ∂ w e = d − 2 d 2 ∫ 0 1 ( q + d v ) 2 v ( 1 − v ) d v ≥ d − 3 q 2 d 2 ≥ 0.
Let S ( γ ) = ∑ n ≥ 1 n γ 2 n / ( 2 n + 1 ) S(\gamma)=\sum_{n\ge1}n\gamma^{2n}/(2n+1) S ( γ ) = ∑ n ≥ 1 n γ 2 n / ( 2 n + 1 ) . With γ = d / ( 2 q + d ) \gamma=d/(2q+d) γ = d / ( 2 q + d ) the continuous series is
e ( w , d ) = w d + 4 q ( 1 + γ ) S ( γ ) , 0 ≤ S − S N ≤ γ 2 N + 2 2 ( 1 − γ 2 ) . (348) e(w,d)=wd+4q(1+\gamma)S(\gamma),\qquad0\le S-S_N\le\frac{\gamma^{2N+2}}{2(1-\gamma^2)}. \tag*{(348)} e ( w , d ) = w d + 4 q ( 1 + γ ) S ( γ ) , 0 ≤ S − S N ≤ 2 ( 1 − γ 2 ) γ 2 N + 2 . (348)
For γ E = ( E − 1 ) / ( E + 1 ) \gamma_E=(E-1)/(E+1) γ E = ( E − 1 ) / ( E + 1 ) , (A.2) gives z = L M ( E ) + ε z z=L_M(E)+\varepsilon_z z = L M ( E ) + ε z , 0 ≤ ε z ≤ U M ( E ) − L M ( E ) 0\le\varepsilon_z\le U_M(E)-L_M(E) 0 ≤ ε z ≤ U M ( E ) − L M ( E ) , and
Φ ( E ) = 4 ( 1 + γ E ) S ( γ E ) , rem : = E − 1 − z = 2 γ E 2 1 − γ E 2 + γ E Φ ( E ) 1 + γ E . (349) \Phi(E)=4(1+\gamma_E)S(\gamma_E),\qquad\operatorname{rem}:=E-1-z=\frac{2\gamma_E^2}{1-\gamma_E^2}+\frac{\gamma_E\Phi(E)}{1+\gamma_E}. \tag*{(349)} Φ ( E ) = 4 ( 1 + γ E ) S ( γ E ) , rem := E − 1 − z = 1 − γ E 2 2 γ E 2 + 1 + γ E γ E Φ ( E ) . (349)
No formula divides by d d d or E − 1 E-1 E − 1 in the verification.
The following corner rules enclose entire boxes. The central cost increases separately in a , b , E a,b,E a , b , E on 0 < a , b ≤ 1 0<a,b\le1 0 < a , b ≤ 1 , 1 ≤ E < 2 1\le E<2 1 ≤ E < 2 . Its a a a derivative has numerator 2 [ ( 4 + b 2 E ) ( E − 1 ) − 2 b E log E ] ≥ 0 2[(4+b^2E)(E-1)-2bE\log E]\ge0 2 [( 4 + b 2 E ) ( E − 1 ) − 2 b E log E ] ≥ 0 ; the b b b case is symmetric. Its E E E derivative has numerator 2 Q ( 4 − P ) − P ( D + 4 log E ) > 0 2Q(4-P)-P(D+4\log E)>0 2 Q ( 4 − P ) − P ( D + 4 log E ) > 0 , using D + 4 log E < E ( 4 − P ) D+4\log E<E(4-P) D + 4 log E < E ( 4 − P ) and 2 Q > E P 2Q>EP 2 Q > E P . Thus we evaluate its lower and upper corners using the noncancelling form
A c = ( 2 Q − E P ) ( E − 1 ) + E P rem D . A_c=\frac{(2Q-EP)(E-1)+EP\operatorname{rem}}{D}. A c = D ( 2 Q − E P ) ( E − 1 ) + E P rem .
For ν = e B / b \nu=e_B/b ν = e B / b the integral identity gives
ν = ( 1 − b ) η + 2 b η 2 ∫ 0 1 v ( 1 − v ) 2 − b + b η v d v . \nu=(1-b)\eta+2b\eta^2\int_0^1\frac{v(1-v)}{2-b+b\eta v}\,dv. ν = ( 1 − b ) η + 2 b η 2 ∫ 0 1 2 − b + b η v v ( 1 − v ) d v .
It increases in η \eta η and decreases in b b b , since ∂ b ν ≤ − η + 2 η 2 / [ 3 ( 2 − b ) 2 ] ≤ 0 \partial_b\nu\le-\eta+2\eta^2/[3(2-b)^2]\le0 ∂ b ν ≤ − η + 2 η 2 / [ 3 ( 2 − b ) 2 ] ≤ 0 . The remainder e B − b ( 1 − b ) η e_B-b(1-b)\eta e B − b ( 1 − b ) η increases in both b b b and η \eta η . The exact range of b ( 1 − b ) b(1-b) b ( 1 − b ) is determined by its endpoint values and its maximum 1 / 4 1/4 1/4 if the interval contains 1 / 2 1/2 1/2 . Finally V A 2 = 4 a / ( E b ) − a 2 V_A^2=4a/(Eb)-a^2 V A 2 = 4 a / ( E b ) − a 2 and a D = 4 a − E b a 2 aD=4a-Eba^2 a D = 4 a − E b a 2 increase in a a a and decrease in b , E b,E b , E , because E a b < 2 Eab<2 E ab < 2 .
Writing ξ = a E \xi=aE ξ = a E , the right cost is
β = ∫ 0 η I ( b , ξ , v ) d v , I = b 2 + ξ ( 2 − b ) 2 [ 3 + ( 1 − b + b v ) 2 ] 2 ( 4 − ξ b ) ( 2 − b + b v ) 2 . (350) \beta=\int_0^\eta I(b,\xi,v)\,dv,\qquad I=\frac{b}{2}+\frac{\xi(2-b)^2[3+(1-b+bv)^2]}{2(4-\xi b)(2-b+bv)^2}. \tag*{(350)} β = ∫ 0 η I ( b , ξ , v ) d v , I = 2 b + 2 ( 4 − ξ b ) ( 2 − b + b v ) 2 ξ ( 2 − b ) 2 [ 3 + ( 1 − b + b v ) 2 ] . (350)
It increases in ξ \xi ξ and η \eta η . Its monotonicity in b b b is the explicit polynomial identity in Section B.5.1.2. Consequently the lower and upper ( b , ξ , η ) (b,\xi,\eta) ( b , ξ , η ) corners bound the entire cost box. These rules are bounds on functions over a box, not point acceptance tests.
A strict upper bound from one residual budget
For a box let B B B be an upper bound for U − A c − β U-A_c-\beta U − A c − β , decreased to U U U if necessary. If B ≤ 0 B\le0 B ≤ 0 , an actual positive left leaf is impossible. Otherwise α ≤ B \alpha\le B α ≤ B . For w = r min w=r_{\min} w = r m i n and V 2 V^2 V 2 an upper bound for V A 2 V_A^2 V A 2 , (B.57) and the exact cost give
α > L ( w , V 2 , d A ) , L ( w , V 2 , d ) = d 2 + ( 3 + w 2 + w d ) d ( 1 + w ) 2 V 2 ( 1 + w + d ) . \alpha>L(w,V^2,d_A),\qquad L(w,V^2,d)=\frac{d}{2}+\frac{(3+w^2+wd)d(1+w)}{2V^2(1+w+d)}. α > L ( w , V 2 , d A ) , L ( w , V 2 , d ) = 2 d + 2 V 2 ( 1 + w + d ) ( 3 + w 2 + w d ) d ( 1 + w ) .
This expression increases in w w w and decreases in V 2 V^2 V 2 . Put q = 1 + w q=1+w q = 1 + w , h c = 1 + w q / V 2 h_c=1+wq/V^2 h c = 1 + w q / V 2 , and p c = q + ( 3 + w 2 ) q / V 2 − 2 B > 0 p_c=q+(3+w^2)q/V^2-2B>0 p c = q + ( 3 + w 2 ) q / V 2 − 2 B > 0 . The inequality L < B L<B L < B is equivalent to h c d 2 + p c d − 2 B q < 0 h_cd^2+p_cd-2Bq<0 h c d 2 + p c d − 2 B q < 0 , and hence
d A < δ : = 4 B q p c 2 + 8 h c B q + p c . (351) d_A<\delta:=\frac{4Bq}{\sqrt{p_c^2+8h_cBq}+p_c}. \tag*{(351)} d A < δ := p c 2 + 8 h c B q + p c 4 B q . (351)
Taking rigorous upper bounds, set d ˉ = min ( a max , δ max , 2 B ) \bar{d}=\min(a_{\max},\delta_{\max},2B) d ˉ = min ( a m a x , δ m a x , 2 B ) and u e = e ( r max , d ˉ ) max u_e=e(r_{\max},\bar{d})_{\max} u e = e ( r m a x , d ˉ ) m a x . Then
e A < u e . (352) e_A<u_e. \tag*{(352)} e A < u e . (352)
Strictness comes from the positive actual leaf and e A > r d A e_A>rd_A e A > r d A , including r = 0 r=0 r = 0 ; it is not a numerical tolerance.
B.1.6.15. Lower bounds and the continuous affine root. The central sign supplies l 0 = max ( 0 , [ a ( Φ − r ( E − 1 ) ) ] min ) ≤ e A l_0=\max(0,[a(\Phi-r(E-1))]_{\min})\le e_A l 0 = max ( 0 , [ a ( Φ − r ( E − 1 )) ] m i n ) ≤ e A . For a stronger bound define
q B = 2 − b , ϵ = b η / q B , ν = e B / b , q_B=2-b,\qquad\epsilon=b\eta/q_B,\qquad\nu=e_B/b, q B = 2 − b , ϵ = b η / q B , ν = e B / b ,
M 0 = rem − r ( E − 1 ) , M_0=\operatorname{rem}-r(E-1), M 0 = rem − r ( E − 1 ) ,
T 0 = [ 4 M 0 − ( D − 2 ) ν ] / D , T_0=[4M_0-(D-2)\nu]/D, T 0 = [ 4 M 0 − ( D − 2 ) ν ] / D ,
p i 0 = − 2 ( E − 1 ) a [ Φ − r ( E − 1 ) ] / D 2 , p_{i0}=-2(E-1)a[\Phi-r(E-1)]/D^2, p i 0 = − 2 ( E − 1 ) a [ Φ − r ( E − 1 )] / D 2 ,
I 0 = η ( 1 + ϵ ) T 0 − η e B ( 1 + ϵ / 2 ) / q B − b 2 η 2 2 q B + K B η 2 E a D , I_0=\eta(1+\epsilon)T_0-\eta e_B(1+\epsilon/2)/q_B
-\frac{b^2\eta^2}{2q_B}+\frac{K_B\eta^2Ea}{D}, I 0 = η ( 1 + ϵ ) T 0 − η e B ( 1 + ϵ /2 ) / q B − 2 q B b 2 η 2 + D K B η 2 E a ,
p 0 = p i 0 − E a I 0 D ( 1 + ϵ ) 2 , p_0=p_{i0}-\frac{EaI_0}{D(1+\epsilon)^2}, p 0 = p i 0 − D ( 1 + ϵ ) 2 E a I 0 ,
A r = 2 [ ( E − 1 ) + E η / ( 1 + ϵ ) ] / D 2 . A_r=2[(E-1)+E\eta/(1+\epsilon)]/D^2. A r = 2 [( E − 1 ) + E η / ( 1 + ϵ )] / D 2 .
Substitution in (B.40) gives the exact identity
p o u t = A r e A + p 0 . (353) p_{\mathrm{out}}=A_re_A+p_0. \tag*{(353)} p out = A r e A + p 0 . (353)
Only if the entire box has ( A r ) min > 0 (A_r)_{\min}>0 ( A r ) m i n > 0 may we divide and set l = max ( l 0 , [ − p 0 / A r ] min ) l=\max(l_0,[-p_0/A_r]_{\min}) l = max ( l 0 , [ − p 0 / A r ] m i n ) . Otherwise only l 0 l_0 l 0 is used. In particular, division is forbidden on the face E = 1 , η = 0 E=1,\eta=0 E = 1 , η = 0 , where A r = 0 A_r=0 A r = 0 .
For fixed actual a , b , E , e B a,b,E,e_B a , b , E , e B , equation (B.36) has second derivative − 2 / ( a 2 D 2 ) -2/(a^2D^2) − 2/ ( a 2 D 2 ) with respect to e A e_A e A . For l e < u e l_e<u_e l e < u e its exact chord identity is
C ( e ) = ( u e − e ) C ( l e ) + ( e − l e ) C ( u e ) u e − l e + ( e − l e ) ( u e − e ) a 2 D 2 . (354) \mathcal{C}(e)=\frac{(u_e-e)\mathcal{C}(l_e)+(e-l_e)\mathcal{C}(u_e)}{u_e-l_e}
+\frac{(e-l_e)(u_e-e)}{a^2D^2}. \tag*{(354)} C ( e ) = u e − l e ( u e − e ) C ( l e ) + ( e − l e ) C ( u e ) + a 2 D 2 ( e − l e ) ( u e − e ) . (354)
Thus l e ≥ u e l_e\ge u_e l e ≥ u e contradicts l e ≤ e A < u e l_e\le e_A<u_e l e ≤ e A < u e ; otherwise two bounds C ( l e ) , C ( u e ) ≥ H ∗ \mathcal{C}(l_e),\mathcal{C}(u_e)\ge H_* C ( l e ) , C ( u e ) ≥ H ∗ imply the actual C ≥ H ∗ \mathcal{C}\ge H_* C ≥ H ∗ . The determinant is always evaluated in full. No mean, realization, budget, or maximizing condition is imposed on an artificial endpoint l e l_e l e or u e u_e u e .
Lemma B.23 (Finite residual lemma, computer-assisted). Every box in the complete closed certificate specified in Section B.5.1 either contains no actual residual point with F ≤ U F\le U F ≤ U , or implies $\mathcal{C}\ge H_* for every such point. Consequently every actual point in (B.56) is excluded by the same-lift calibration argument.
Proof. The cover has 44 813 leaves. Its eight acceptance predicates are:
b min > a max b_{\min}>a_{\max} b m i n > a m a x , contradicting scalar ordering.
[ r ( 1 − b ) ] min ≥ 1 / 5 [r(1-b)]_{\min}\ge1/5 [ r ( 1 − b ) ] m i n ≥ 1/5 , invoking Lemma B.22, including equality.
[ b ( 243 + 4 ( a E ) 2 ) − 16 a E ] max < 0 [b(243+4(aE)^2)-16aE]_{\max}<0 [ b ( 243 + 4 ( a E ) 2 ) − 16 a E ] m a x < 0 , contradicting the actual cusp.
[ U − A c − β ] max ≤ 0 [U-A_c-\beta]_{\max}\le0 [ U − A c − β ] m a x ≤ 0 , contradicting α > 0 \alpha>0 α > 0 .
l 0 ≥ u e l_0\ge u_e l 0 ≥ u e , contradicting (B.62).
( A r ) min > 0 (A_r)_{\min}>0 ( A r ) m i n > 0 and l ≥ u e l\ge u_e l ≥ u e , giving the same contradiction.
l 0 < u e l_0<u_e l 0 < u e and both full endpoint bounds at l 0 , u e l_0,u_e l 0 , u e are at least H ∗ H_* H ∗ .
(8) ( A r ) min > 0 (A_r)_{\min} > 0 ( A r ) m i n > 0 , l < u e l < u_e l < u e , and both full endpoint bounds at l , u e l, u_e l , u e are at least H ∗ H_* H ∗ . The last two use (B.64). Every recorded predicate is checked on the entire reconstructed closed box, with directed integer bounds. The prefix-free complete binary tree covers the outward enclosure of (B.56); both children retain every common split face. Therefore the predicates exhaust every actual point. □ \square □
Completion of Theorem B.20 . Suppose F ≤ U F \le U F ≤ U . If either r r r or s s s is negative, Lemma B.21 gives C > H ∗ \mathcal{C} > H_* C > H ∗ , since U < 129 / 500 U < 129/500 U < 129/500 . Otherwise the outer reduction and the residual certificate prove C ≥ H ∗ > 0 \mathcal{C} \ge H_* > 0 C ≥ H ∗ > 0 . At an actual state K A > 0 K_A > 0 K A > 0 , so Corollary A.13 on this qualified complete lift gives C < H ∗ \mathcal{C} < H_* C < H ∗ . This contradiction proves the claim. Lawful cyclic transport covers the other two listed contexts without changing physical time or cost. □ \square □
Forward supports, homogeneous cores, and physical arm integrals
Forward runs require two kinds of payments: a central pair supported by surrounding caps, and larger disjoint physical arms. We derive the common cap support once, then use it for the pair and arm estimates. Reverse homogeneous cores use the same full-lift setup but their own signed series.
Cap forcing for forward pairs
Strict forcing by a forward pair
Lemma B.24. At the middle switch of an actual positive 1201 word on one full maximizing lift, let
x = 1 + u v , y = S 2 v , z = 1 − u v . x = \frac{1+u}{v}, \qquad y = \frac{S}{2v}, \qquad z = \frac{1-u}{v}. x = v 1 + u , y = 2 v S , z = v 1 − u . Then x y + x z + y z = 1 xy+xz+yz=1 x y + x z + y z = 1 and y < x y<x y < x , y < z y<z y < z . This conclusion has no cost threshold.
Proof. Apply Lemma A.7 on this same full lift and normalize jointly. Use its forward specialization δ = − 1 \delta=-1 δ = − 1 , P = F − 1 P=F^{-1} P = F − 1 , R = E − 1 R=E^{-1} R = E − 1 , where E = e t 2 / z > 1 E=e^{t_2/z}>1 E = e t 2 / z > 1 , F = e t 0 / x > 1 F=e^{t_0/x}>1 F = e t 0 / x > 1 . Write V = ( a , b ) V=(a,b) V = ( a , b ) , so b ≤ 0 b\le0 b ≤ 0 , and identify its ν x , ν z \nu_x,\nu_z ν x , ν z with ν 1 , ν 2 \nu_1,\nu_2 ν 1 , ν 2 , putting ν 0 = ( 1 , 0 ) \nu_0=(1,0) ν 0 = ( 1 , 0 ) and λ i = V ⋅ J ν i \lambda_i=V\cdot J\nu_i λ i = V ⋅ J ν i . With w 0 = 1 − x z = y ( x + z ) > 0 w_0=1-xz=y(x+z)>0 w 0 = 1 − x z = y ( x + z ) > 0 , the common cap rows become
h + A z λ 2 = c z , h + A x λ 1 = c x (355) h+A_z\lambda_2=c_z,\qquad h+A_x\lambda_1=c_x \tag*{(355)} h + A z λ 2 = c z , h + A x λ 1 = c x (355) where
A z = ( E − 1 ) ( 1 + z 2 ) 2 E , c z = z ( E − 1 ) E 2 ( ( 1 + z 2 ) f ( E ) − w 0 ( E − 1 ) ) , A_z=\frac{(E-1)(1+z^2)}{2E},\qquad c_z=\frac{z(E-1)}{E^2}\left((1+z^2)f(E)-w_0(E-1)\right), A z = 2 E ( E − 1 ) ( 1 + z 2 ) , c z = E 2 z ( E − 1 ) ( ( 1 + z 2 ) f ( E ) − w 0 ( E − 1 ) ) , A x = ( F − 1 ) ( 1 + x 2 ) 2 F , c x = x ( F − 1 ) F 2 ( ( 1 + x 2 ) f ( F ) − w 0 ( F − 1 ) ) . A_x=\frac{(F-1)(1+x^2)}{2F},\qquad c_x=\frac{x(F-1)}{F^2}\left((1+x^2)f(F)-w_0(F-1)\right). A x = 2 F ( F − 1 ) ( 1 + x 2 ) , c x = F 2 x ( F − 1 ) ( ( 1 + x 2 ) f ( F ) − w 0 ( F − 1 ) ) . Here the displayed A i A_i A i are positive; they are the negatives of the signed coefficients in Lemma A.7. Its solved line is
b = r h − Q , Q = r h − b ≥ 0 (356) b=rh-Q,\qquad Q=rh-b\ge0 \tag*{(356)} b = r h − Q , Q = r h − b ≥ 0 (356) where
r = 2 w 0 ( x + z ) ( z F F − 1 + x E E − 1 ) > 2 , r=\frac{2}{w_0(x+z)}\left(\frac{zF}{F-1}+\frac{xE}{E-1}\right)>2, r = w 0 ( x + z ) 2 ( F − 1 z F + E − 1 x E ) > 2 , Q = 2 x z w 0 ( x + z ) { ( 1 + x 2 ) f ( F ) F + ( 1 + z 2 ) f ( E ) E − w 0 ( F − 1 F + E − 1 E ) } . Q=\frac{2xz}{w_0(x+z)}\left\{\frac{(1+x^2)f(F)}{F}+\frac{(1+z^2)f(E)}{E}-w_0\left(\frac{F-1}{F}+\frac{E-1}{E}\right)\right\}. Q = w 0 ( x + z ) 2 x z { F ( 1 + x 2 ) f ( F ) + E ( 1 + z 2 ) f ( E ) − w 0 ( F F − 1 + E E − 1 ) } . For G > 1 G>1 G > 1 , define g ( G ) = ( G − 1 − log G ) / ( G − 1 ) g(G)=(G-1-\log G)/(G-1) g ( G ) = ( G − 1 − log G ) / ( G − 1 ) . The elementary inequality G 2 − 1 − 2 G log G > 0 G^2-1-2G\log G>0 G 2 − 1 − 2 G log G > 0 follows by differentiating twice. It gives
f ( G ) < G − 1 2 , g ( G ) > G − 1 2 G , f ( G ) G = G − 1 G − g ( G ) . f(G)<\frac{G-1}{2},\qquad g(G)>\frac{G-1}{2G},\qquad\frac{f(G)}{G}=\frac{G-1}{G}-g(G). f ( G ) < 2 G − 1 , g ( G ) > 2 G G − 1 , G f ( G ) = G G − 1 − g ( G ) .
Substitution in Q Q Q implies
0 ≤ Q < x z w 0 ( ( x − y ) F − 1 F + ( z − y ) E − 1 E ) . (357) 0 \le Q < \frac{xz}{w_0}\left((x-y)\frac{F-1}{F}+(z-y)\frac{E-1}{E}\right). \tag*{(357)} 0 ≤ Q < w 0 x z ( ( x − y ) F F − 1 + ( z − y ) E E − 1 ) . (357)
In particular, y < max ( x , z ) y<\max(x,z) y < max ( x , z ) .
The exact endpoint responses in Lemma A.7, with the same forward substitution, give the actual nonnegative cap-ray coefficients
h − = A z ( b z − b ) , h + = A x ( b x − b ) , h_- = A_z(b_z-b), \qquad h_+ = A_x(b_x-b), h − = A z ( b z − b ) , h + = A x ( b x − b ) ,
b z = 2 z ( f ( E ) − z ( E − 1 ) x + z ) , b x = 2 x ( f ( F ) − x ( F − 1 ) x + z ) . b_z=2z\left(f(E)-\frac{z(E-1)}{x+z}\right), \qquad b_x=2x\left(f(F)-\frac{x(F-1)}{x+z}\right). b z = 2 z ( f ( E ) − x + z z ( E − 1 ) ) , b x = 2 x ( f ( F ) − x + z x ( F − 1 ) ) .
Thus Q + b x = r h + h + / A x ≥ 0 Q+b_x=rh+h_+/A_x\ge0 Q + b x = r h + h + / A x ≥ 0 . The preceding strict scalar bounds give
w 0 ( Q + b x ) 2 x < F − 1 2 F { x ( z − y ) + ( F − 1 ) y ( z − x ) } + E − 1 2 E z ( z − y ) . \frac{w_0(Q+b_x)}{2x}<\frac{F-1}{2F}\{x(z-y)+(F-1)y(z-x)\}+\frac{E-1}{2E}z(z-y). 2 x w 0 ( Q + b x ) < 2 F F − 1 { x ( z − y ) + ( F − 1 ) y ( z − x )} + 2 E E − 1 z ( z − y ) .
If z ≤ min ( x , y ) z\le\min(x,y) z ≤ min ( x , y ) the right side is nonpositive, a contradiction. If z ≤ y z\le y z ≤ y still held, it would force x < z ≤ y x<z\le y x < z ≤ y , contradicting (B.67). Hence z > y z>y z > y . Interchanging ( x , F ) (x,F) ( x , F ) and ( z , E ) (z,E) ( z , E ) proves x > y x>y x > y . Positive durations supply strictness even when a cap coefficient or first switching jet vanishes.
The fixed-state support geometry of a forward pair
In this section x , y , z x,y,z x , y , z denote the positive reciprocal coordinates at the middle of an actual capped 1201 pair; by Lemma B.24 , x > y x>y x > y and z > y z>y z > y . Set ℓ = x + y + z \ell=x+y+z ℓ = x + y + z . Use the actual complete central durations d 2 , d 0 > 0 d_2,d_0>0 d 2 , d 0 > 0 , E = e d 2 / z E=e^{d_2/z} E = e d 2 / z , F = e d 0 / x F=e^{d_0/x} F = e d 0 / x , and the cap responses A i , c i A_i,c_i A i , c i from (B.65).
Lemma B.25 (Fixed-state cap family and support). Use the common affine cap line of Lemma A.7, in the forward notation V ( h ) = ( c h − p ∗ , r h − Q ) V(h)=(ch-p_*,rh-Q) V ( h ) = ( c h − p ∗ , r h − Q ) , 0 ≤ h ≤ h ∗ : = Q / r 0\le h\le h_*:=Q/r 0 ≤ h ≤ h ∗ := Q / r , r > 2 r>2 r > 2 . Its determinant, gradient N = ∇ H ℓ N=\nabla H^\ell N = ∇ H ℓ and actual trace are those of (A.12); direct state algebra gives ∣ N ∣ 2 = ℓ 2 − 3 |N|^2=\ell^2-3 ∣ N ∣ 2 = ℓ 2 − 3 . The determinant is strictly concave along this line. For i = x , z i=x,z i = x , z , with corresponding duration d i = d 0 , d 2 d_i=d_0,d_2 d i = d 0 , d 2 , let
α i = i − y i + y , β i = α i d i 2 , C x = ( x + z ) ( x − y ) 2 x , C z = ( x + z ) ( z − y ) 2 z . \alpha_i=\frac{i-y}{i+y}, \qquad\beta_i=\frac{\alpha_i d_i}{2}, \qquad C_x=\frac{(x+z)(x-y)}{2x}, \qquad C_z=\frac{(x+z)(z-y)}{2z}. α i = i + y i − y , β i = 2 α i d i , C x = 2 x ( x + z ) ( x − y ) , C z = 2 z ( x + z ) ( z − y ) . Both algebraic endpoints V ( 0 ) , V ( h ∗ ) V(0),V(h_*) V ( 0 ) , V ( h ∗ ) belong to the triangle with vertices
W 0 = ( − ( x + x − 1 ) β x , 0 ) = ( − C x d 0 , 0 ) , W_0=(-(x+x^{-1})\beta_x,0)=(-C_xd_0,0), W 0 = ( − ( x + x − 1 ) β x , 0 ) = ( − C x d 0 , 0 ) , W 2 = ( ( z + z − 1 ) β z , 0 ) = ( C z d 2 , 0 ) , W_2=((z+z^{-1})\beta_z,0)=(C_zd_2,0), W 2 = (( z + z − 1 ) β z , 0 ) = ( C z d 2 , 0 ) , W 12 = − ( y + y − 1 ) ( − β x ν 2 + β z ν 1 ) . (358) W_{12}=-(y+y^{-1})(-\beta_x\nu_2+\beta_z\nu_1). \tag*{(358)} W 12 = − ( y + y − 1 ) ( − β x ν 2 + β z ν 1 ) . (358) where ν 1 , ν 2 \nu_1,\nu_2 ν 1 , ν 2 are the normals in Lemma B.24 . In addition, V ( h ∗ ) V(h_*) V ( h ∗ ) lies on the segment [ W 0 , W 2 ] [W_0,W_2] [ W 0 , W 2 ] , and
∣ W 12 ∣ 2 = ( y + y − 1 ) 2 ( β x + β z ) 2 − 4 ( 1 + y 2 ) β x β z , (359) |W_{12}|^2=(y+y^{-1})^2(\beta_x+\beta_z)^2-4(1+y^2)\beta_x\beta_z, \tag*{(359)} ∣ W 12 ∣ 2 = ( y + y − 1 ) 2 ( β x + β z ) 2 − 4 ( 1 + y 2 ) β x β z , (359) h ∗ ≤ ( d 0 C x + d 2 C z ) d 0 d 2 d 0 + d 2 ≤ ℓ ( d 0 + d 2 ) 2 8 . (360) h_*\le(d_0C_x+d_2C_z)\frac{d_0d_2}{d_0+d_2}\le\frac{\ell(d_0+d_2)^2}{8}. \tag*{(360)} h ∗ ≤ ( d 0 C x + d 2 C z ) d 0 + d 2 d 0 d 2 ≤ 8 ℓ ( d 0 + d 2 ) 2 . (360) For L ( W ) = N ⋅ W / 2 + ∣ W ∣ 2 / 4 \mathcal{L}(W)=N\cdot W/2+|W|^2/4 L ( W ) = N ⋅ W /2 + ∣ W ∣ 2 /4 and H = ( ℓ − N y ) h ∗ / 2 H=(\ell-N_y)h_*/2 H = ( ℓ − N y ) h ∗ /2 , the actual determinant therefore satisfies
C a c t u a l ≥ 3 4 − max { L ( W 12 ) , L ( W 0 ) + H , L ( W 2 ) + H } . (361) \mathcal{C}_{\mathrm{actual}}\ge\frac{3}{4}-\max\{\mathcal{L}(W_{12}),\mathcal{L}(W_0)+H,\mathcal{L}(W_2)+H\}. \tag*{(361)} C actual ≥ 4 3 − max { L ( W 12 ) , L ( W 0 ) + H , L ( W 2 ) + H } . (361)
Proof. The common cap-line and determinant assertions follow from Lemma A.7 with the forward specialization just stated. Put γ i = c i / ( 2 A i ) \gamma_i = c_i/(2A_i) γ i = c i / ( 2 A i ) . Substituting the explicit cap responses, and writing q i = d i / i > 0 q_i = d_i/i > 0 q i = d i / i > 0 , gives the exact positive defect
β i − γ i = i { δ 0 ( q i ) + α i δ 1 ( q i ) } > 0 , δ 0 ( q ) = sinh q − q e q − 1 , δ 1 ( q ) = q − 1 + e − q 2 . \beta_i-\gamma_i=i\{\delta_0(q_i)+\alpha_i\delta_1(q_i)\}>0,\qquad\delta_0(q)=\frac{\sinh q-q}{e^q-1},\qquad\delta_1(q)=\frac{q-1+e^{-q}}{2}. β i − γ i = i { δ 0 ( q i ) + α i δ 1 ( q i )} > 0 , δ 0 ( q ) = e q − 1 sinh q − q , δ 1 ( q ) = 2 q − 1 + e − q .
Both functions are positive for q > 0 q>0 q > 0 , by differentiating their numerators. The cap rows therefore imply V ⋅ J ν i ≤ 2 γ i < 2 β i V\cdot J_{\nu_i}\leq2\gamma_i<2\beta_i V ⋅ J ν i ≤ 2 γ i < 2 β i when h ≥ 0 h\geq0 h ≥ 0 . The two algebraic endpoints also have V y ≤ 0 V_y\leq0 V y ≤ 0 . Solving the intersections of these three half-planes gives precisely (B.68); their positive normal dependence proves that the intersection is bounded. At h ∗ h_* h ∗ its vertical coordinate is zero. Direct scalar products, using x y + x z + y z = 1 xy+xz+yz=1 x y + x z + y z = 1 , prove (B.69).
The same positive normal relation, γ i < β i \gamma_i<\beta_i γ i < β i , and log G > 1 − G − 1 \log G>1-G^{-1} log G > 1 − G − 1 for G > 1 G>1 G > 1 give respectively
Q ≤ d 0 z ( x − y ) + d 2 x ( z − y ) y ( x + z ) , r > 2 x z y ( x + z ) 2 ( d 0 − 1 + d 2 − 1 ) . Q\leq\frac{d_0z(x-y)+d_2x(z-y)}{y(x+z)},\qquad r>\frac{2xz}{y(x+z)^2}(d_0^{-1}+d_2^{-1}). Q ≤ y ( x + z ) d 0 z ( x − y ) + d 2 x ( z − y ) , r > y ( x + z ) 2 2 x z ( d 0 − 1 + d 2 − 1 ) .
Division proves the first height bound when Q > 0 Q>0 Q > 0 ; if Q = 0 Q=0 Q = 0 it is immediate. Since 0 < C x , C z < ℓ / 2 0<C_x,C_z<\ell/2 0 < C x , C z < ℓ /2 and d 0 d 2 ≤ ( d 0 + d 2 ) 2 / 4 d_0d_2\leq(d_0+d_2)^2/4 d 0 d 2 ≤ ( d 0 + d 2 ) 2 /4 , the second follows.
The determinant is concave along the fixed-state cap line, so its minimum is at h = 0 h=0 h = 0 or h = h ∗ h=h_* h = h ∗ . At both endpoints h V y = 0 hV_y=0 h V y = 0 . The convex function L \mathcal{L} L is bounded above by its triangle vertex values at h = 0 h=0 h = 0 , and by its two horizontal vertex values at h = h ∗ h=h_* h = h ∗ . Since ℓ > ∣ N ∣ \ell>|N| ℓ > ∣ N ∣ , H ≥ 0 H\geq0 H ≥ 0 , so the two conclusions combine to give (B.71). The algebraic endpoints and vertices are never asserted to be physical lifts and are never calibrated. □
Lemma B.26 (A common three-vertex estimate). Write R = x / y > 1 R=x/y>1 R = x / y > 1 , t = z / y > 1 t=z/y>1 t = z / y > 1 , ρ = R − 1 R + 1 \rho=\frac{R-1}{R+1} ρ = R + 1 R − 1 , σ = t − 1 t + 1 \sigma=\frac{t-1}{t+1} σ = t + 1 t − 1 , w = ρ + σ w=\rho+\sigma w = ρ + σ , n = 3 − w − ρ σ n=3-w-\rho\sigma n = 3 − w − ρ σ .
Let k i = ℓ d i k_i=\ell d_i k i = ℓ d i , k = k 0 + k 2 k=k_0+k_2 k = k 0 + k 2 , and assume w < w 0 < 1 w<w_0<1 w < w 0 < 1 , n > n 0 > 0 n>n_0>0 n > n 0 > 0 , k < B < n 0 k<B<n_0 k < B < n 0 , and R , t < 3 R,t<3 R , t < 3 . Set
u 0 = 2 w 0 ( 1 + w 0 ) n 0 , l 0 = w 0 ( 1 + w 0 / 2 ) n 0 . u_0=\frac{2w_0}{(1+w_0)n_0},\qquad l_0=\frac{w_0}{(1+w_0/2)n_0}. u 0 = ( 1 + w 0 ) n 0 2 w 0 , l 0 = ( 1 + w 0 /2 ) n 0 w 0 . Then the three losses in (B.71) obey
L ( W 12 ) < B w 0 2 4 ( n 0 − B ) . (362) \mathcal{L}(W_{12})<\frac{Bw_0^2}{4(n_0-B)}. \tag*{(362)} L ( W 12 ) < 4 ( n 0 − B ) B w 0 2 . (362) L ( W i ) + H ≤ u 0 2 B 2 / 4 + u 0 B 2 / 6 if N ⋅ W i ≤ 0. (363) \mathcal{L}(W_i)+H\leq u_0^2B^2/4+u_0B^2/6\qquad\text{if }N\cdot W_i\leq0. \tag*{(363)} L ( W i ) + H ≤ u 0 2 B 2 /4 + u 0 B 2 /6 if N ⋅ W i ≤ 0. (363) L ( W i ) + H ≤ ( D 0 + H 0 ) 2 4 H 0 if N ⋅ W i ≥ 0. (364) \mathcal{L}(W_i)+H\leq\frac{(D_0+H_0)^2}{4H_0}\qquad\text{if }N\cdot W_i\geq0. \tag*{(364)} L ( W i ) + H ≤ 4 H 0 ( D 0 + H 0 ) 2 if N ⋅ W i ≥ 0. (364) where H 0 = 2 u 0 B 2 / 3 H_0=2u_0B^2/3 H 0 = 2 u 0 B 2 /3 and D 0 = w 0 2 B / ( 8 n 0 ) + l 0 2 B 2 / 4 D_0=w_0^2B/(8n_0)+l_0^2B^2/4 D 0 = w 0 2 B / ( 8 n 0 ) + l 0 2 B 2 /4 .
Proof. The exact scalar identities, obtained by substituting R , t R,t R , t , are
( y + y − 1 ) / ℓ = 4 / n , N ⋅ W 12 / 2 = ρ σ k / n , ∣ W 12 ∣ ≤ 2 ( ρ k 0 + σ k 2 ) / n , (y+y^{-1})/\ell=4/n,\qquad N\cdot W_{12}/2=\rho\sigma k/n,\qquad|W_{12}|\leq2(\rho k_0+\sigma k_2)/n, ( y + y − 1 ) / ℓ = 4/ n , N ⋅ W 12 /2 = ρ σ k / n , ∣ W 12 ∣ ≤ 2 ( ρ k 0 + σ k 2 ) / n ,
N ⋅ W 0 / 2 = ρ ( σ − ρ ) k 0 ( 1 + ρ ) n , N ⋅ W 2 / 2 = σ ( ρ − σ ) k 2 ( 1 + σ ) n , N\cdot W_0/2=\frac{\rho(\sigma-\rho)k_0}{(1+\rho)n},\qquad N\cdot W_2/2=\frac{\sigma(\rho-\sigma)k_2}{(1+\sigma)n}, N ⋅ W 0 /2 = ( 1 + ρ ) n ρ ( σ − ρ ) k 0 , N ⋅ W 2 /2 = ( 1 + σ ) n σ ( ρ − σ ) k 2 ,
c 0 : = C x / ℓ = 2 ρ ( 1 − ρ σ ) ( 1 + ρ ) n , c 2 : = C z / ℓ = 2 σ ( 1 − ρ σ ) ( 1 + σ ) n . c_0:=C_x/\ell=\frac{2\rho(1-\rho\sigma)}{(1+\rho)n},\qquad c_2:=C_z/\ell=\frac{2\sigma(1-\rho\sigma)}{(1+\sigma)n}. c 0 := C x / ℓ = ( 1 + ρ ) n 2 ρ ( 1 − ρ σ ) , c 2 := C z / ℓ = ( 1 + σ ) n 2 σ ( 1 − ρ σ ) .
Put δ = ∣ ρ − σ ∣ \delta=|\rho-\sigma| δ = ∣ ρ − σ ∣ . The first three formulas and (B.69) give
L ( W 12 ) ≤ B ( w 2 − δ 2 ) 4 n 0 + B 2 ( w + δ ) 2 4 n 0 2 . \mathcal{L}(W_{12})\leq\frac{B(w^2-\delta^2)}{4n_0}+\frac{B^2(w+\delta)^2}{4n_0^2}. L ( W 12 ) ≤ 4 n 0 B ( w 2 − δ 2 ) + 4 n 0 2 B 2 ( w + δ ) 2 .
For b = B / η 0 < 1 b = B/\eta_0 < 1 b = B / η 0 < 1 , the bracket is bounded by the exact square
( w 2 − δ 2 ) + b ( w + δ ) 2 = w 2 1 − b − ( 1 − b ) ( δ − b w 1 − b ) 2 . (w^2-\delta^2)+b(w+\delta)^2=\frac{w^2}{1-b}-(1-b)\left(\delta-\frac{bw}{1-b}\right)^2. ( w 2 − δ 2 ) + b ( w + δ ) 2 = 1 − b w 2 − ( 1 − b ) ( δ − 1 − b b w ) 2 .
This proves (B.72). The scalar formulas give c 0 , c 2 ≤ u 0 c_0,c_2\le u_0 c 0 , c 2 ≤ u 0 ; when a vertex has nonnegative signed term, its own coefficient is at most l 0 l_0 l 0 , since its corresponding ρ \rho ρ or σ \sigma σ is at most w / 2 w/2 w /2 . Also
ℓ − N y 2 ℓ = R 2 + R t + t 2 ( R + t ) ( R + t + 1 ) ≤ 2 / 3. \frac{\ell-N_y}{2\ell}=\frac{R^2+Rt+t^2}{(R+t)(R+t+1)}\le2/3. 2 ℓ ℓ − N y = ( R + t ) ( R + t + 1 ) R 2 + R t + t 2 ≤ 2/3.
This is equivalent to R 2 − R t + t 2 − 2 R − 2 t ≤ 0 R^2-Rt+t^2-2R-2t\le0 R 2 − R t + t 2 − 2 R − 2 t ≤ 0 on [ 1 , 3 ] 2 [1,3]^2 [ 1 , 3 ] 2 ; separate convexity bounds it by its four vertex values − 3 , − 1 , − 1 , − 3 -3,-1,-1,-3 − 3 , − 1 , − 1 , − 3 . Thus (B.70) gives H ≤ ( 2 / 3 ) ( c 0 k 0 + c 2 k 2 ) k 0 k 2 / k H\le(2/3)(c_0k_0+c_2k_2)k_0k_2/k H ≤ ( 2/3 ) ( c 0 k 0 + c 2 k 2 ) k 0 k 2 / k . Discarding a nonpositive signed term proves (B.73). For a nonnegative term put λ = k i / k \lambda=k_i/k λ = k i / k . The inequality ρ ( σ − ρ ) ≤ w 2 / 8 \rho(\sigma-\rho)\le w^2/8 ρ ( σ − ρ ) ≤ w 2 /8 and its mate give
L ( W i ) + H ≤ D 0 λ + H 0 λ ( 1 − λ ) ≤ ( D 0 + H 0 ) 2 4 H 0 , \mathcal{L}(W_i)+H\le D_0\lambda+H_0\lambda(1-\lambda)\le\frac{(D_0+H_0)^2}{4H_0}, L ( W i ) + H ≤ D 0 λ + H 0 λ ( 1 − λ ) ≤ 4 H 0 ( D 0 + H 0 ) 2 ,
by completing the square. This proves the final estimate. □
Corollary B.27 (Ratio-free support bounds). Without assuming x < 1 x<1 x < 1 , z < 1 z<1 z < 1 , or an upper ratio bound, let T = d 0 + d 2 T=d_0+d_2 T = d 0 + d 2 . Every point in the support triangle satisfies
∣ V ∣ ≤ ( y + y − 1 ) T / 2 , N ⋅ V / 2 ≤ ( x − y ) ( z − y ) 4 y T . |V|\le(y+y^{-1})T/2,\qquad N\cdot V/2\le\frac{(x-y)(z-y)}{4y}T. ∣ V ∣ ≤ ( y + y − 1 ) T /2 , N ⋅ V /2 ≤ 4 y ( x − y ) ( z − y ) T . The same cap family has h ∗ ≤ ℓ T 2 / 8 h_*\le\ell T^2/8 h ∗ ≤ ℓ T 2 /8 .
Proof. The height assertion is already (B.70). At W 12 W_{12} W 12 , (B.69) and 0 < α i < 1 0<\alpha_i<1 0 < α i < 1 give the norm bound. At W 0 , W 2 W_0,W_2 W 0 , W 2 , use C x , C z < ℓ / 2 < ( y + y − 1 ) / 2 C_x,C_z<\ell/2<(y+y^{-1})/2 C x , C z < ℓ /2 < ( y + y − 1 ) /2 ; convexity of the norm then proves it throughout the triangle. The signed bound at W 12 W_{12} W 12 is equality, since N ⋅ W 12 = ( x − y ) ( z − y ) T / ( 2 y ) N\cdot W_{12}=(x-y)(z-y)T/(2y) N ⋅ W 12 = ( x − y ) ( z − y ) T / ( 2 y ) . At W 0 W_0 W 0 its value is ( z − x ) ( x − y ) d 0 / ( 4 x ) (z-x)(x-y)d_0/(4x) ( z − x ) ( x − y ) d 0 / ( 4 x ) . If this is nonpositive the claimed bound is immediate; otherwise ( z − x ) / x < ( z − y ) / y (z-x)/x<(z-y)/y ( z − x ) / x < ( z − y ) / y since x > y x>y x > y , and the result follows. The W 2 W_2 W 2 case is the exchanged scalar computation. Linearity in V V V completes the proof on the triangle. □
Own-run and mixed-edge estimates
Lemma B.28 (The two-arm integral). Let four complete consecutive modes 1, 2, 0, 1 occupy [ t 0 , t 4 ] [t_0,t_4] [ t 0 , t 4 ] , with middle switch t 2 t_2 t 2 . Suppose their actual endpoint states satisfy
z ( t 0 ) < y ( t 0 ) , x ( t 4 ) < y ( t 4 ) . z(t_0)<y(t_0),\qquad x(t_4)<y(t_4). z ( t 0 ) < y ( t 0 ) , x ( t 4 ) < y ( t 4 ) . At the middle use ρ , σ , w \rho,\sigma,w ρ , σ , w from Lemma B.26, so 0 < ρ , σ < 1 0<\rho,\sigma<1 0 < ρ , σ < 1 and 0 < w < 2 0<w<2 0 < w < 2 . Their physical cost J J J satisfies
J > G ( ρ , σ ) = Φ ( ( 1 − ρ ) ( 1 − σ ) ) , Φ ( q ) = 4 − 4 q + q log q 4 − q . (365) J>G(\rho,\sigma)=\Phi((1-\rho)(1-\sigma)),\qquad\Phi(q)=\frac{4-4q+q\log q}{4-q}. \tag*{(365)} J > G ( ρ , σ ) = Φ (( 1 − ρ ) ( 1 − σ )) , Φ ( q ) = 4 − q 4 − 4 q + q log q . (365) At fixed w = ρ + σ w=\rho+\sigma w = ρ + σ , the exact algebraic minimum is balanced:
G ≥ Φ ( ( 1 − w / 2 ) 2 ) > G ( w ) ≔ w ( 4 − w ) ( 6 + w ) 2 ( 6 − w ) ( 2 + w ) . (366) G\ge\Phi((1-w/2)^2)>G(w)\coloneqq\frac{w(4-w)(6+w)}{2(6-w)(2+w)}. \tag*{(366)} G ≥ Φ (( 1 − w /2 ) 2 ) > G ( w ) : = 2 ( 6 − w ) ( 2 + w ) w ( 4 − w ) ( 6 + w ) . (366) The balanced comparison is strictly increasing in w ∈ ( 0 , 2 ) w\in(0,2) w ∈ ( 0 , 2 ) ; its equality point is not asserted to be an actual trajectory.
Proof. The reciprocal dynamics, obtained directly from the original fields, are
x ′ y ′ z ′ 0 0 ( x + y ) / x − ( x + z ) / x 1 − ( x + y ) / y 0 ( y + z ) / y 2 ( x + z ) / z − ( y + z ) / z 0 \begin{array}{c|ccc}
& x' & y' & z'\\
0 & 0 & (x+y)/x & -(x+z)/x\\
1 & -(x+y)/y & 0 & (y+z)/y\\
2 & (x+z)/z & -(y+z)/z & 0
\end{array} 0 1 2 x ′ 0 − ( x + y ) / y ( x + z ) / z y ′ ( x + y ) / x 0 − ( y + z ) / z z ′ − ( x + z ) / x ( y + z ) / y 0
On the actual right two arcs, q = ( x − y ) / ( x + y ) q=(x-y)/(x+y) q = ( x − y ) / ( x + y ) has q ′ = − 2 / ( x + y ) < 0 q'=-2/(x+y)<0 q ′ = − 2/ ( x + y ) < 0 . It goes from ρ > 0 \rho>0 ρ > 0 to a negative value, hence has one zero. On 0 ≤ q ≤ ρ 0\le q\le\rho 0 ≤ q ≤ ρ , the sum a ( q ) = x + y a(q)=x+y a ( q ) = x + y satisfies a ( q ) ≥ 2 y ( t 2 ) / ( 1 − q ) a(q)\ge2y(t_2)/(1-q) a ( q ) ≥ 2 y ( t 2 ) / ( 1 − q ) : mode 0 has constant x x x , and on mode 1 its constant y ( t 3 ) > y ( t 2 ) y(t_3)>y(t_2) y ( t 3 ) > y ( t 2 ) gives the same inequality. Since ℓ a = 1 + a 2 ( 3 + q 2 ) / 4 \ell a=1+a^2(3+q^2)/4 ℓ a = 1 + a 2 ( 3 + q 2 ) /4 , that portion costs at least ρ / 2 + y ( t 2 ) 2 F ( ρ ) / 2 \rho/2+y(t_2)^2F(\rho)/2 ρ /2 + y ( t 2 ) 2 F ( ρ ) /2 , where F ( b ) = b + 2 log ( 1 − b ) + 4 / ( 1 − b ) − 4 F(b)=b+2\log(1-b)+4/(1-b)-4 F ( b ) = b + 2 log ( 1 − b ) + 4/ ( 1 − b ) − 4 . The omitted cost is strictly positive by the strict endpoint sign. For the disjoint left portion use q = ( z − y ) / ( z + y ) q=(z-y)/(z+y) q = ( z − y ) / ( z + y ) , q ′ = 2 / ( z + y ) > 0 q'=2/(z+y)>0 q ′ = 2/ ( z + y ) > 0 , and read its integral backwards from the middle; this reverses an integration parameter, not the physical lift. Thus J > w / 2 + y 2 [ F ( ρ ) + F ( σ ) ] / 2 J>w/2+y^2[F(\rho)+F(\sigma)]/2 J > w /2 + y 2 [ F ( ρ ) + F ( σ )] /2 . For q = ( 1 − ρ ) ( 1 − σ ) q=(1-\rho)(1-\sigma) q = ( 1 − ρ ) ( 1 − σ ) one has y 2 = q / ( 4 − q ) y^2=q/(4-q) y 2 = q / ( 4 − q ) and
F ( ρ ) + F ( σ ) = w + 2 log q + 4 ( 2 − w ) / q − 8 , F(\rho)+F(\sigma)=w+2\log q+4(2-w)/q-8, F ( ρ ) + F ( σ ) = w + 2 log q + 4 ( 2 − w ) / q − 8 ,
which proves the exact product formula. Now Φ ′ ( q ) = ( 4 log q − 8 − q ) / ( 4 − q ) 2 < 0 \Phi'(q)=(4\log q-8-q)/(4-q)^2<0 Φ ′ ( q ) = ( 4 log q − 8 − q ) / ( 4 − q ) 2 < 0 on ( 0 , 1 ) (0,1) ( 0 , 1 ) , and q ≤ ( 1 − w / 2 ) 2 q\le(1-w/2)^2 q ≤ ( 1 − w /2 ) 2 , with equality precisely when ρ = σ \rho=\sigma ρ = σ . Finally log q > − ( 1 − q ) / q \log q>-(1-q)/\sqrt{q} log q > − ( 1 − q ) / q : with a = q < 1 a=\sqrt{q}<1 a = q < 1 , the difference log a − ( a − a − 1 ) / 2 \log a-(a-a^{-1})/2 log a − ( a − a − 1 ) /2 vanishes at 1 and has derivative − ( a − 1 ) 2 / ( 2 a 2 ) < 0 -(a-1)^2/(2a^2)<0 − ( a − 1 ) 2 / ( 2 a 2 ) < 0 . Substituting a = 1 − w / 2 a=1-w/2 a = 1 − w /2 gives G ( w ) G(w) G ( w ) .
Lemma B.29 (Four charges with endpoint state signs). Under the hypotheses of Lemma B.28, one has J > 41 / 100 J>41/100 J > 41/100 .
Proof. Assume J ≤ a = 41 / 100 J\le a=41/100 J ≤ a = 41/100 . The balanced comparison in (B.76) and the exact gap
G ( 249 / 500 ) − 41 / 100 = 2600387 / 1145333000 > 0 G(249/500)-41/100=2600387/1145333000>0 G ( 249/500 ) − 41/100 = 2600387/1145333000 > 0
force
w < 249 / 500 < 1 / 2 , R , t < 3 , n > 39 / 16. (367) w<249/500<1/2,\qquad R,t<3,\qquad n>39/16. \tag*{(367)} w < 249/500 < 1/2 , R , t < 3 , n > 39/16. (367)
The actual central two-arc cost I I I is strictly smaller than J J J . Its durations sum to T < a / 3 T<a/\sqrt{3} T < a / 3 . The exact density profiles, measured from the middle along each actual arc, are
ℓ 0 ( v ) = ( x + y ) e v / x + ( x + z ) e − v / x − x , ℓ 2 ( v ) = ( y + z ) e v / z + ( x + z ) e − v / z − z . \ell_0(v)=(x+y)e^{v/x}+(x+z)e^{-v/x}-x,\qquad\ell_2(v)=(y+z)e^{v/z}+(x+z)e^{-v/z}-z. ℓ 0 ( v ) = ( x + y ) e v / x + ( x + z ) e − v / x − x , ℓ 2 ( v ) = ( y + z ) e v / z + ( x + z ) e − v / z − z .
They are strictly convex with initial slopes ( y − z ) / x (y-z)/x ( y − z ) / x , ( y − x ) / z (y-x)/z ( y − x ) / z . Integrated tangent bounds give
I > ℓ T − 1 2 { ( z − y ) d 2 2 / x + ( x − y ) d 2 2 / z } > ℓ T − T 2 , I>\ell T-\frac{1}{2}\{(z-y)d_2^2/x+(x-y)d_2^2/z\}>\ell T-T^2, I > ℓ T − 2 1 {( z − y ) d 2 2 / x + ( x − y ) d 2 2 / z } > ℓ T − T 2 ,
using R , t < 3 R,t<3 R , t < 3 . Consequently k = ℓ T < B = a + a 2 / 3 = 13981 / 30000 k=\ell T<B=a+a^2/3=13981/30000 k = ℓ T < B = a + a 2 /3 = 13981/30000 .
Apply the support estimates with n 0 = 39 / 16 n_0=39/16 n 0 = 39/16 . For the nonhorizontal vertex retain w 0 = 249 / 500 w_0=249/500 w 0 = 249/500 ; for the two horizontal bounds use the weaker w 0 = 1 / 2 w_0=1/2 w 0 = 1/2 , so u 0 = 32 / 117 u_0=32/117 u 0 = 32/117 , l 0 = 32 / 195 l_0=32/195 l 0 = 32/195 . Their exact upper values are
L 12 = 866835981 59144000000 , L − = 2150151971 154001250000 , L + = 1864266368476801 133467750000000000 . L_{12}=\frac{866835981}{59144000000},\qquad L_-=\frac{2150151971}{154001250000},\qquad L_+=\frac{1864266368476801}{133467750000000000}. L 12 = 59144000000 866835981 , L − = 154001250000 2150151971 , L + = 133467750000000000 1864266368476801 .
The formulas for them are (B.72)–(B.74), and direct rational comparison gives L − , L + < L 12 L_-,L_+<L_{12} L − , L + < L 12 and
59 / 4000 − L 12 = 5538019 / 59144000000 > 0. 59/4000-L_{12}=5538019/59144000000>0. 59/4000 − L 12 = 5538019/59144000000 > 0.
Hence the actual determinant exceeds 2941 / 4000 > 0 2941/4000>0 2941/4000 > 0 . Independently, (B.70) gives 0 ≤ h / ℓ ≤ T 2 / 8 < a 2 / 24 < 1 0\le h/\ell\le T^2/8<a^2/24<1 0 ≤ h / ℓ ≤ T 2 /8 < a 2 /24 < 1 , so its actual trace ℓ − h \ell-h ℓ − h is positive. The same complete lift is normal by Lemma B.24. Only now apply the harmonic cap, which gives c < C 0 < 2941 / 4000 c<C_0<2941/4000 c < C 0 < 2941/4000 , a contradiction.
Corollary B.30 (Four own arcs). In the actual positive contexts 101202, 212010, 020121, the cost of the four complete interior arcs is greater than 41 / 100 41/100 41/100 .
Proof. Use the full legal cycle taking 101202 to 212010. Its four charges are 1201. The actual left return 212, transformed to 020, gives z ( t 0 ) < y ( t 0 ) z(t_0)<y(t_0) z ( t 0 ) < y ( t 0 ) by Lemma B.10; the actual right return 010, transformed to 202, gives x ( t 4 ) < y ( t 4 ) x(t_4)<y(t_4) x ( t 4 ) < y ( t 4 ) . These are precisely the two hypotheses of Lemma B.29, which proves J > 41 / 100 J>41/100 J > 41/100 . Legal cycles preserve the two other contexts and their charged positions.
Theorem B.31 (Mixed edge and long own-run resources). For each actual positive full-seven context 1012012 and its legal phases, charge the four complete arcs [1, 4]; for 0120121 and its legal phases, charge [2, 5]. Each charged cost exceeds 41 / 100 41/100 41/100 . The respective uncharged supporting arcs 5 and 1 *must be complete; the two exterior caps must remain positive. Consequently an ordinary forward run with at least seven increments has own-arc cost greater than 41 / 50 41/50 41/50 on its full eligible context.
Proof. For 1012012, the charges have modes 0120. The legal + 1 +1 + 1 cycle makes them 1201 with middle coordinates ( x n e w , y n e w , z n e w ) = ( z , x , y ) (x_{\mathrm{new}}, y_{\mathrm{new}}, z_{\mathrm{new}}) = (z, x, y) ( x new , y new , z new ) = ( z , x , y ) . The left return 101 gives x ( t 1 ) > y ( t 1 ) x(t_1) > y(t_1) x ( t 1 ) > y ( t 1 ) , so the first required canonical endpoint sign is z n e w < y n e w z_{\mathrm{new}} < y_{\mathrm{new}} z new < y new . The fully capped 2012 on arcs [3, 6] gives z ( t 5 ) < x ( t 5 ) z(t_5) < x(t_5) z ( t 5 ) < x ( t 5 ) by Lemma B.24 in its legal phase, so the second sign is x n e w < y n e w x_{\mathrm{new}} < y_{\mathrm{new}} x new < y new . Lemma B.29 applies. Its latter sign uses the complete uncharged supporting arc 5; a fragment of that arc would not suffice.
For 0120121, use the − 1 -1 − 1 cycle on the four charges 2012 to obtain canonical 1201 with coordinates ( y , z , x ) (y, z, x) ( y , z , x ) . The fully capped 0120 on [0, 3] gives the initial sign x < z x < z x < z , and the terminal return 121 gives the final sign z > y z > y z > y . In the new chart these are exactly the two endpoint inequalities. This uses the complete uncharged arc 1 and proves the second estimate.
A forward run with a a a increments has a + 1 a + 1 a + 1 own arcs, with successive labels increasing by one. Its first seven context arcs form the first mixed pattern; its last seven form a legal phase of the second. For a ≥ 7 a \ge7 a ≥ 7 , the first four and last four own arcs are disjoint. Apply both bounds on one eligible whole-run lift and add their physical costs. Their supporting contexts may overlap but their charges do not. The remaining own arcs, if any, have positive cost. The result is > 2 ( 41 / 100 ) = 41 / 50 > 2(41/100) = 41/50 > 2 ( 41/100 ) = 41/50 .
Theorem B.32 (Six own arcs). For each actual positive context 10120121, 21201202, 02012010, the physical cost of complete positions 1, …, 6 exceeds 609 / 1000 > 3 / 5 609/1000 > 3/5 609/1000 > 3/5 . All eight source arcs and the same full containing-interval hypotheses are retained.
Proof. Work with 10120121. Its prefix [0, 6] = 1012012 and suffix [1, 7] = 0120121 supply two capped middles at t 3 t_3 t 3 and t 5 t_5 t 5 , with the same full normal lift. Use canonical coordinates ( z , x , y ) (z, x, y) ( z , x , y ) at t 3 t_3 t 3 and ( y , z , x ) (y, z, x) ( y , z , x ) at t 5 t_5 t 5 . The two-arm integrals of Lemma B.28 give
J 6 > G ( ρ 3 , σ 3 ) + G ( ρ 5 , σ 5 ) . (368) J_6 > G(\rho_3, \sigma_3) + G(\rho_5, \sigma_5). \tag*{(368)} J 6 > G ( ρ 3 , σ 3 ) + G ( ρ 5 , σ 5 ) . (368)
Here the two inner arms really are disjoint: on the shared actual modes 2, 0 between t 3 , t 5 t_3, t_5 t 3 , t 5 , the quantity q = ( z − x ) / ( z + x ) q = (z - x)/(z + x) q = ( z − x ) / ( z + x ) has q ′ = − 2 / ( x + z ) < 0 q' = -2/(x + z) < 0 q ′ = − 2/ ( x + z ) < 0 . It starts positive and ends negative. The right arm at t 3 t_3 t 3 uses precisely q ≥ 0 q \ge0 q ≥ 0 ; the left arm at t 5 t_5 t 5 uses q ≤ 0 q \le0 q ≤ 0 , with canonical variable − q -q − q . They meet only at the unique zero. The outer arms lie on disjoint outer arc pairs, and their strict endpoint signs leave positive omitted cost.
Suppose J 6 ≤ A = 609 / 1000 J_6 \le A = 609/1000 J 6 ≤ A = 609/1000 . Select one actual middle at which G < A / 2 G < A/2 G < A /2 , which exists by (B.78). The same increasing balanced comparison and G ( 7 / 20 ) − A / 2 = 10451 / 10622000 > 0 G(7/20) - A/2 = 10451/10622000 > 0 G ( 7/20 ) − A /2 = 10451/10622000 > 0 give
w < w 0 = 7 / 20 , R , t < 27 / 13 , n > n 0 = 4191 / 1600. w < w_0 = 7/20,\quad R,t < 27/13,\quad n > n_0 = 4191/1600. w < w 0 = 7/20 , R , t < 27/13 , n > n 0 = 4191/1600.
The central two arcs at the selected actual middle are a subset of the six charges. The integrated tangent proof above, now using R , t < 27 / 13 R,t < 27/13 R , t < 27/13 , gives
k = ℓ T < B = A + 14 13 A 2 6 = 8782389 13000000 . k = \ell T < B = A + \frac{14}{13}\frac{A^2}{6} = \frac{8782389}{13000000}. k = ℓ T < B = A + 13 14 6 A 2 = 13000000 8782389 .
Use the support family of Lemma B.25 and put u = 2 w 0 / [ ( 1 + w 0 ) n 0 ] u = 2w_0/[(1 + w_0)n_0] u = 2 w 0 / [( 1 + w 0 ) n 0 ] , l = w 0 / [ ( 1 + w 0 / 2 ) n 0 ] l = w_0/[(1 + w_0/2)n_0] l = w 0 / [( 1 + w 0 /2 ) n 0 ] . As before c 0 , c 2 ≤ u c_0, c_2 \le u c 0 , c 2 ≤ u , and the coefficient belonging to a vertex with nonnegative signed term is at most l l l . In addition c 0 + c 2 ≤ 2 l c_0 + c_2 \le2l c 0 + c 2 ≤ 2 l : indeed ρ / ( 1 + ρ ) + σ / ( 1 + σ ) ≤ w / ( 1 + w / 2 ) \rho/(1+\rho) + \sigma/(1+\sigma) \le w/(1+w/2) ρ / ( 1 + ρ ) + σ / ( 1 + σ ) ≤ w / ( 1 + w /2 ) by concavity, and 1 − ρ σ ≤ 1 1-\rho\sigma\le1 1 − ρ σ ≤ 1 . The nonhorizontal loss satisfies
L ( W 12 ) < B w 0 2 4 ( n 0 − B ) < 59 4000 ; \mathcal{L}(W_{12})<\frac{Bw_0^2}{4(n_0-B)}<\frac{59}{4000}; L ( W 12 ) < 4 ( n 0 − B ) B w 0 2 < 4000 59 ;
the exact margin is 25154971 / 6125936000 > 0 25154971/6125936000>0 25154971/6125936000 > 0 . For a horizontal vertex with nonpositive signed term use λ 2 ( 1 − λ ) , λ ( 1 − λ ) 2 ≤ 4 / 27 \lambda^2(1-\lambda),\lambda(1-\lambda)^2\le4/27 λ 2 ( 1 − λ ) , λ ( 1 − λ ) 2 ≤ 4/27 in the height bound to obtain
L ( W i ) + H ≤ B 2 ( u 2 / 4 + 16 l / 81 ) < 59 / 4000. \mathcal{L}(W_i)+H\le B^2(u^2/4+16l/81)<59/4000. L ( W i ) + H ≤ B 2 ( u 2 /4 + 16 l /81 ) < 59/4000.
Its exact margin is 87509319798137 / 3242851377187500000 > 0 87509319798137/3242851377187500000>0 87509319798137/3242851377187500000 > 0 .
For a vertex with nonnegative signed term, let λ = k i / k \lambda=k_i/k λ = k i / k . The sharper own coefficient and the retained duration split give
L ( W i ) + H ≤ P ( λ ) : = d λ + e λ 2 + 2 3 B 2 [ l λ + u ( 1 − λ ) ] λ ( 1 − λ ) , d = w 0 2 B 8 n 0 , e = l 2 B 2 4 . \mathcal{L}(W_i)+H\le P(\lambda):=d\lambda+e\lambda^2+\frac{2}{3}B^2[l\lambda+u(1-\lambda)]\lambda(1-\lambda),\qquad d=\frac{w_0^2B}{8n_0},\quad e=\frac{l^2B^2}{4}. L ( W i ) + H ≤ P ( λ ) := d λ + e λ 2 + 3 2 B 2 [ l λ + u ( 1 − λ )] λ ( 1 − λ ) , d = 8 n 0 w 0 2 B , e = 4 l 2 B 2 .
Put a 3 = 2 B 2 ( u − l ) / 3 > 0 a_3=2B^2(u-l)/3>0 a 3 = 2 B 2 ( u − l ) /3 > 0 , p = d + B 2 ( 5 l − u ) / 6 p=d+B^2(5l-u)/6 p = d + B 2 ( 5 l − u ) /6 and q = B 2 l ( 2 / 3 − l / 4 ) > 0 q=B^2l(2/3-l/4)>0 q = B 2 l ( 2/3 − l /4 ) > 0 . Completing the square gives, for 0 ≤ λ ≤ 1 0\le\lambda\le1 0 ≤ λ ≤ 1 ,
P ( λ ) = a 3 4 + p λ − q λ 2 − a 3 ( 1 − λ ) ( λ − 1 / 2 ) 2 ≤ a 3 4 + p 2 4 q = 2918301475265239513583 205447452944760000000000 . P(\lambda)=\frac{a_3}{4}+p\lambda-q\lambda^2-a_3(1-\lambda)(\lambda-1/2)^2\le\frac{a_3}{4}+\frac{p^2}{4q}=\frac{2918301475265239513583}{205447452944760000000000}. P ( λ ) = 4 a 3 + p λ − q λ 2 − a 3 ( 1 − λ ) ( λ − 1/2 ) 2 ≤ 4 a 3 + 4 q p 2 = 205447452944760000000000 2918301475265239513583 .
Here u = 22400 / 113157 u=22400/113157 u = 22400/113157 , l = 22400 / 196977 l=22400/196977 l = 22400/196977 , and the exact margin is
59 4000 − ( a 3 4 + p 2 4 q ) = 112048455669970486417 205447452944760000000000 > 0. \frac{59}{4000}-\left(\frac{a_3}{4}+\frac{p^2}{4q}\right)=\frac{112048455669970486417}{205447452944760000000000}>0. 4000 59 − ( 4 a 3 + 4 q p 2 ) = 205447452944760000000000 112048455669970486417 > 0.
Thus the actual determinant is > 2941 / 4000 > 0 >2941/4000>0 > 2941/4000 > 0 . Independently h / ℓ ≤ T 2 / 8 < A 2 / 24 h/\ell\le T^2/8<A^2/24 h / ℓ ≤ T 2 /8 < A 2 /24 gives positive actual trace, with ratio ( ℓ − h ) / ℓ > 1 − A 2 / 24 = 7876373 / 8000000 > 0 (\ell-h)/\ell>1-A^2/24=7876373/8000000>0 ( ℓ − h ) / ℓ > 1 − A 2 /24 = 7876373/8000000 > 0 . The same-lift harmonic cap is now applicable and gives C < C 0 < 2941 / 4000 \mathcal{C}<C_0<2941/4000 C < C 0 < 2941/4000 , a contradiction.
Forward and reverse homogeneous cores
Actual reverse caps and common scalar kernels
Throughout this analytic block, C = det Λ C=\det\Lambda C = det Λ is the same-lift matrix Casimir, while A , I , J , W A,I,J,W A , I , J , W denote physical costs. An eligible actual six-context consists of six positive vertex arcs in one genuine fundamental minimizing interval, with all four charged interior arcs complete. Reanchoring retains every complete charged, return and sign arc and positive portions of the actual exterior caps; the charged span is strictly shorter than the interval. Select one full lift there and prove normality from its actual positive cap equations, as in the proof of Theorem A.8, before joint normalization. The phrase “selected-six premise” adds no numerical or selection hypothesis.
Lemma B.33 (The actual reverse-cap line). On one complete normal maximizing free-terminal-time lift, jointly normalized to η = 1 \eta=1 η = 1 , retain an actual positive capped reverse pair 1 ∣ 02 ∣ 1 1\mid02\mid1 1 ∣ 02 ∣ 1 with both charged arcs complete. Its actual cap equations, outer-ray residuals and determinant are respectively (A.10), (B.79) and (A.12). They imply a nonempty affine cap segment 0 ≤ h ≤ h ∗ 0\le h\le h_* 0 ≤ h ≤ h ∗ and strict determinant concavity on it. These identities require neither a numerical cost budget nor harmonic calibration; the algebraic endpoints are not asserted to be trajectories.
Proof. Apply Lemma A.7 with δ = 1 \delta=1 δ = 1 at the actual middle, using x = ( 1 + u ) / v x=(1+u)/v x = ( 1 + u ) / v , y = ( u 2 + v 2 − 1 ) / ( 2 v ) y=(u^2+v^2-1)/(2v) y = ( u 2 + v 2 − 1 ) / ( 2 v ) and z = ( 1 − u ) / v z=(1-u)/v z = ( 1 − u ) / v . Put P = e t 0 / x P=e^{t_0/x} P = e t 0 / x , R = e t 2 / z R=e^{t_2/z} R = e t 2 / z for the two complete physical durations. Its A x , A z , β x , β z , ν x , ν z A_x,A_z,\beta_x,\beta_z,\nu_x,\nu_z A x , A z , β x , β z , ν x , ν z give exactly the reverse cap rows (A.10). Write D c a p = D D_{\mathrm{cap}}=D D cap = D and r 0 = r > 0 r_0=r>0 r 0 = r > 0 in its solved line v y = q 0 − r 0 h v_y=q_0-r_0h v y = q 0 − r 0 h ; then D c a p = q 0 ≥ 0 D_{\mathrm{cap}=q_0\ge0} D cap = q 0 ≥ 0 and 0 ≤ h ≤ h ∗ = q 0 / r 0 0\le h\le h_*=q_0/r_0 0 ≤ h ≤ h ∗ = q 0 / r 0 . The common polarization and endpoint minimum retain the cap and both state covectors, without a cost budget or calibration. Put d ( g ) = g ( g − 1 − log g ) / ( g log g − g + 1 ) d(g)=g(g-1-\log g)/(g\log g-g+1) d ( g ) = g ( g − 1 − log g ) / ( g log g − g + 1 ) . The actual cap residual implies
y / z < max ( P d ( P ) , d ( R ) ) , y / x < max ( d ( P ) , R d ( R ) ) . (369) y/z<\max(Pd(P),d(R)),\qquad y/x<\max(d(P),Rd(R)). \tag*{(369)} y / z < max ( P d ( P ) , d ( R )) , y / x < max ( d ( P ) , R d ( R )) . (369)
Put a ( g ) = log g / ( g − 1 ) a(g)=\log g/(g-1) a ( g ) = log g / ( g − 1 ) . The actual endpoint signs in (A.11) give
V y ≥ b x = 2 x { a ( P ) − z + x / P x + z } . V_y \ge b_x=2x\left\{a(P)-\frac{z+x/P}{x+z}\right\}. V y ≥ b x = 2 x { a ( P ) − x + z z + x / P } .
Since V y = q 0 − r 0 h V_y=q_0-r_0h V y = q 0 − r 0 h and actual h ≥ 0 h\ge0 h ≥ 0 , one has q 0 ≥ b x q_0\ge b_x q 0 ≥ b x . Set k = y / z k=y/z k = y / z and r = x / z r=x/z r = x / z . Dividing this last inequality by positive factors gives the nonnegative outer-ray residual
r [ P + k P − ( P + k ) a ( P ) ] + R + 3 k − k ( P + 1 ) a ( P ) − ( k + 1 ) R a ( R ) . r\left[P+\frac{k}{P}-(P+k)a(P)\right]+R+3k-k(P+1)a(P)-(k+1)Ra(R). r [ P + P k − ( P + k ) a ( P ) ] + R + 3 k − k ( P + 1 ) a ( P ) − ( k + 1 ) R a ( R ) .
Its two components are
f ( P ) [ P d ( P ) − k ] P , f ( R ) [ d ( R ) − k ] + k [ 2 − ( P + 1 ) a ( P ) ] . \frac{f(P)[Pd(P)-k]}{P},\qquad f(R)[d(R)-k]+k[2-(P+1)a(P)]. P f ( P ) [ P d ( P ) − k ] , f ( R ) [ d ( R ) − k ] + k [ 2 − ( P + 1 ) a ( P )] .
The final bracket is strictly negative, since a ( P ) > 2 / ( P + 1 ) a(P)>2/(P+1) a ( P ) > 2/ ( P + 1 ) by Jensen’s inequality applied to a ( P ) = ∫ 0 1 [ 1 + ( P − 1 ) t ] − 1 d t a(P)=\int_0^1[1+(P-1)t]^{-1}\mathrm{d}t a ( P ) = ∫ 0 1 [ 1 + ( P − 1 ) t ] − 1 d t . This proves the first inequality in (B.79); the outgoing mode-2 cap similarly gives V y ≥ 2 z { a ( R ) − ( x + z / R ) / ( x + z ) } V_y\ge2z\{a(R)-(x+z/R)/(x+z)\} V y ≥ 2 z { a ( R ) − ( x + z / R ) / ( x + z )} , proving the second inequality by the same algebra.
The remaining cap-segment, strict concavity and trace assertions are (A.11)–(A.12) on this same selected lift. □
Lemma B.34 (A common positive kernel). Put
f ( g ) = g log g g − 1 − 1 , d ( g ) = g ( g − 1 − log g ) g log g − g + 1 . f(g)=\frac{g\log g}{g-1}-1,\qquad d(g)=\frac{g(g-1-\log g)}{g\log g-g+1}. f ( g ) = g − 1 g log g − 1 , d ( g ) = g log g − g + 1 g ( g − 1 − log g ) . For every s > 0 s>0 s > 0 ,
s 2 + 5 s 2 60 + s 2 < f ( e s ) < s 2 + s 2 12 . \frac{s}{2}+\frac{5s^2}{60+s^2}<f(e^s)<\frac{s}{2}+\frac{s^2}{12}. 2 s + 60 + s 2 5 s 2 < f ( e s ) < 2 s + 12 s 2 . For 0 < s ≤ 5 / 4 0<s\le5/4 0 < s ≤ 5/4 one also has
log d ( e s ) < 2 s 3 + s 3 360 . \log d(e^s)<\frac{2s}{3}+\frac{s^3}{360}. log d ( e s ) < 3 2 s + 360 s 3 . Consequently d ( e s ) < e c s d(e^s)<e^{cs} d ( e s ) < e cs for 0 < s ≤ B 0<s\le B 0 < s ≤ B in all three rows
( c , B ) = ( 67 / 100 , 1 ) , ( 27 / 40 , 154 / 125 ) , ( 7 / 10 , 6 / 5 ) . (c,B)=(67/100,1),\qquad(27/40,154/125),\qquad(7/10,6/5). ( c , B ) = ( 67/100 , 1 ) , ( 27/40 , 154/125 ) , ( 7/10 , 6/5 ) . The quartic lower bound f ( e s ) > s / 2 + s 2 / 12 − s 4 / 720 f(e^s)>s/2+s^2/12-s^4/720 f ( e s ) > s /2 + s 2 /12 − s 4 /720 also follows.
Proof. For t = s / 2 > 0 t=s/2>0 t = s /2 > 0 , write ϕ ( s ) = f ( e s ) − s / 2 = t coth t − 1 \phi(s)=f(e^s)-s/2=t\coth t-1 ϕ ( s ) = f ( e s ) − s /2 = t coth t − 1 . The absolutely convergent identities
( 3 + t 2 ) sinh t − 3 t cosh t = ∑ n ≥ 2 4 n ( n − 1 ) t 2 n + 1 ( 2 n + 1 ) ! > 0 , (3+t^2)\sinh t-3t\cosh t=\sum_{n\ge2}\frac{4n(n-1)t^{2n+1}}{(2n+1)!}>0, ( 3 + t 2 ) sinh t − 3 t cosh t = n ≥ 2 ∑ ( 2 n + 1 )! 4 n ( n − 1 ) t 2 n + 1 > 0 ,
( 15 + t 2 ) t cosh t − ( 15 + 6 t 2 ) sinh t = ∑ n ≥ 3 8 n ( n − 1 ) ( n − 2 ) t 2 n + 1 ( 2 n + 1 ) ! > 0 (15+t^2)t\cosh t-(15+6t^2)\sinh t=\sum_{n\ge3}\frac{8n(n-1)(n-2)t^{2n+1}}{(2n+1)!}>0 ( 15 + t 2 ) t cosh t − ( 15 + 6 t 2 ) sinh t = n ≥ 3 ∑ ( 2 n + 1 )! 8 n ( n − 1 ) ( n − 2 ) t 2 n + 1 > 0
prove the upper and lower bounds for ϕ \phi ϕ after division by 3 sinh t 3\sinh t 3 sinh t and ( 15 + t 2 ) sinh t (15+t^2)\sinh t ( 15 + t 2 ) sinh t , respectively. The quartic consequence is
5 s 2 60 + s 2 − ( s 2 12 − s 4 720 ) = s 6 720 ( 60 + s 2 ) > 0. \frac{5s^2}{60+s^2}-\left(\frac{s^2}{12}-\frac{s^4}{720}\right)=\frac{s^6}{720(60+s^2)}>0. 60 + s 2 5 s 2 − ( 12 s 2 − 720 s 4 ) = 720 ( 60 + s 2 ) s 6 > 0.
For the logarithmic bound set A = e s − 1 − s > 0 A=e^s-1-s>0 A = e s − 1 − s > 0 and B = e − s − 1 + s > 0 B=e^{-s}-1+s>0 B = e − s − 1 + s > 0 , so d ( e s ) = A / B d(e^s)=A/B d ( e s ) = A / B . Direct expansion gives
( 2 3 + s 2 120 ) A B − A ′ B + A B ′ = ∑ j ≥ 0 ( 8 j 3 + 56 j 2 + 46 j + 10 ) s 2 j + 6 60 ( 2 j + 6 ) ! > 0. \left(\frac{2}{3}+\frac{s^2}{120}\right)AB-A'B+AB'=\sum_{j\ge0}\frac{(8j^3+56j^2+46j+10)s^{2j+6}}{60(2j+6)!}>0. ( 3 2 + 120 s 2 ) A B − A ′ B + A B ′ = j ≥ 0 ∑ 60 ( 2 j + 6 )! ( 8 j 3 + 56 j 2 + 46 j + 10 ) s 2 j + 6 > 0.
Indeed the coefficient of s 2 n s^{2n} s 2 n before the low-degree cancellations is ( 8 n 3 − 16 n 2 − 74 n + 160 ) / ( 60 ( 2 n ) ! ) (8n^3-16n^2-74n+160)/(60(2n)!) ( 8 n 3 − 16 n 2 − 74 n + 160 ) / ( 60 ( 2 n )!) ; the terms through degree four vanish, and n = j + 3 n=j+3 n = j + 3 gives the displayed positive series. Dividing by A B AB A B and integrating from zero, where A / B → 1 A/B\to1 A / B → 1 , proves log d ( e s ) < 2 s / 3 + s 3 / 360 \log d(e^s) < 2s/3 + s^3/360 log d ( e s ) < 2 s /3 + s 3 /360 for every s > 0 s > 0 s > 0 . The reserves c − 2 / 3 − B 2 / 360 c - 2/3 - B^2/360 c − 2/3 − B 2 /360 for the three stated ( c , B ) (c,B) ( c , B ) rows are respectively 1 / 1800 1/1800 1/1800 , 23159 / 5625000 23159/5625000 23159/5625000 and 11 / 375 11/375 11/375 , all positive. □
For 0 ≤ s ≤ 6 / 5 0 \le s \le6/5 0 ≤ s ≤ 6/5 put
L ( s ) = ( 1 − s + s 2 / 2 − s 3 / 6 ) / 2 , P ( s ) = 1 − s / 2 + s 2 / 6 − s 3 / 24 , N ( s ) = L ( s ) + P ( s ) . L(s) = (1 - s + s^2/2 - s^3/6)/2,\qquad P(s) = 1 - s/2 + s^2/6 - s^3/24,\qquad N(s) = L(s) + P(s). L ( s ) = ( 1 − s + s 2 /2 − s 3 /6 ) /2 , P ( s ) = 1 − s /2 + s 2 /6 − s 3 /24 , N ( s ) = L ( s ) + P ( s ) .
Lemma B.35 (A two-vertex reciprocal credit). Let p , q , r > 0 p,q,r > 0 p , q , r > 0 , p q + p r + q r = 1 pq + pr + qr = 1 pq + p r + q r = 1 , with all ordered ratios at most ρ ≥ 1 \rho\ge1 ρ ≥ 1 . For L , P > 0 L,P > 0 L , P > 0 ,
L p 2 + P p q ≥ min { L + P 1 + 2 ρ , L + ρ P ρ ( ρ + 2 ) } . Lp^2 + Ppq \ge\min\left\{\frac{L + P}{1 + 2\rho},\frac{L + \rho P}{\rho(\rho+ 2)}\right\}. L p 2 + P pq ≥ min { 1 + 2 ρ L + P , ρ ( ρ + 2 ) L + ρP } . This is the exact minimum on the algebraic cone. If ρ P ≥ ( 1 + ρ ) L \rho P \ge(1 + \rho)L ρP ≥ ( 1 + ρ ) L , the first expression is the minimum. For the functions L ( s ) , P ( s ) L(s),P(s) L ( s ) , P ( s ) just defined, 0 ≤ s ≤ 6 / 5 0 \le s \le6/5 0 ≤ s ≤ 6/5 , this condition holds for every ρ ≥ 1 \rho\ge1 ρ ≥ 1 .
Proof. Write a = q / p a = q/p a = q / p , b = r / p b = r/p b = r / p . The quantity is ( L + P a ) / ( a + b + a b ) (L + Pa)/(a + b + ab) ( L + P a ) / ( a + b + ab ) and decreases with b b b . For a ≤ 1 a \le1 a ≤ 1 the largest permitted b b b is ρ a \rho a ρ a ; the resulting quotient has derivative with numerator − L ( 1 + ρ + 2 ρ a ) − ρ P a 2 < 0 -L(1 + \rho+ 2\rho a) - \rho P a^2 < 0 − L ( 1 + ρ + 2 ρ a ) − ρP a 2 < 0 . For a ≥ 1 a \ge1 a ≥ 1 the largest b b b is ρ \rho ρ ; the derivative has the constant sign of ρ P − ( 1 + ρ ) L \rho P -(1+\rho)L ρP − ( 1 + ρ ) L . It is therefore enough to test ( a , b ) = ( 1 , ρ ) , ( ρ , ρ ) (a,b) = (1,\rho),(\rho,\rho) ( a , b ) = ( 1 , ρ ) , ( ρ , ρ ) . Both are permitted, proving the minimum formula.
The indicated L , P L,P L , P decrease to L ( 6 / 5 ) = 29 / 250 > 0 L(6/5) = 29/250 > 0 L ( 6/5 ) = 29/250 > 0 , P ( 6 / 5 ) = 71 / 125 > 0 P(6/5) = 71/125 > 0 P ( 6/5 ) = 71/125 > 0 , and
P − 2 L = s ( 3 s 2 − 8 s + 12 ) / 24 ≥ 0 , P - 2L = s(3s^2 - 8s + 12)/24 \ge0, P − 2 L = s ( 3 s 2 − 8 s + 12 ) /24 ≥ 0 ,
since the quadratic has negative discriminant and positive leading term. Thus ρ P − ( 1 + ρ ) L ≥ ( ρ − 1 ) L ≥ 0 \rho P - (1 + \rho)L \ge(\rho- 1)L \ge0 ρP − ( 1 + ρ ) L ≥ ( ρ − 1 ) L ≥ 0 . □
A complete forward three-core
Proposition B.36 (Forward three-core). On an eligible full positive context 01201, or either additive phase, the sum J = c 1 + c 2 + c 3 J = c_1 + c_2 + c_3 J = c 1 + c 2 + c 3 of its three complete interior physical costs satisfies J > 47 / 400 > 29 / 250 J > 47/400 > 29/250 J > 47/400 > 29/250 . The uncharged caps 0, 4 remain positive.
Proof. Use one complete normal lift, jointly normalized as in Lemmas B.24 and B.25. Only the two genuinely capped middles of this five-arc context will be used. At either middle write
x y + x z + y z = 1 , R = x / y > 1 , t = z / y > 1 , D = R + t + R t , ℓ = ( R + t + 1 ) / D . xy + xz + yz = 1,\qquad R = x/y > 1,\qquad t = z/y > 1,\qquad D = R + t + Rt,\qquad\ell= (R + t + 1)/\sqrt{D}. x y + x z + y z = 1 , R = x / y > 1 , t = z / y > 1 , D = R + t + R t , ℓ = ( R + t + 1 ) / D .
For its adjacent own durations t x , t z t_x,t_z t x , t z , put q i = t i / i q_i = t_i/i q i = t i / i and T = t x + t z T = t_x + t_z T = t x + t z . Direct integration of the physical density gives
A x = x y ( e q x − 1 ) + x z ( 1 − e − q x ) + x 2 ( 2 sinh q x − q x ) , A_x = xy(e^{q_x} - 1) + xz(1 - e^{-q_x}) + x^2(2\sinh q_x - q_x), A x = x y ( e q x − 1 ) + x z ( 1 − e − q x ) + x 2 ( 2 sinh q x − q x ) ,
A z = x z ( 1 − e − q z ) + y z ( e q z − 1 ) + z 2 ( 2 sinh q z − q z ) . A_z = xz(1 - e^{-q_z}) + yz(e^{q_z} - 1) + z^2(2\sinh q_z - q_z). A z = x z ( 1 − e − q z ) + y z ( e q z − 1 ) + z 2 ( 2 sinh q z − q z ) .
With N = ∇ H ℓ N = \nabla_H\ell N = ∇ H ℓ , Lemma B.25 supplies the support triangle and h ∗ ≤ ℓ T 2 / 8 h_* \le\ell T^2/8 h ∗ ≤ ℓ T 2 /8 ; we use its three-vertex determinant bound directly.
First we select one of the two existing middles. Present the context as 12012. Along its shared mode-0 arc the neighboring middle is
( x , Y , Z ) = ( x , F ( x + y ) − x , ( x + z ) / F − x ) , F > 1. (x,Y,Z) = (x,F(x+y)-x,(x+z)/F-x),\qquad F > 1. ( x , Y , Z ) = ( x , F ( x + y ) − x , ( x + z ) / F − x ) , F > 1.
The unused minima at the two middles are y y y and Z Z Z . If both min ( x / y , z / y ) \min(x/y,z/y) min ( x / y , z / y ) and min ( Y / Z , x / Z ) \min(Y/Z,x/Z) min ( Y / Z , x / Z ) are at least m > 1 m > 1 m > 1 , put s = max ( y , Z ) s = \max(y,Z) s = max ( y , Z ) and choose w w w with ( x + s ) ( x + w ) = 1 + x 2 (x+s)(x+w) = 1 + x^2 ( x + s ) ( x + w ) = 1 + x 2 . Then x / s , w / s ≥ m x/s,w/s \ge m x / s , w / s ≥ m . Monotonic increase of the second coordinate and decrease of the third show that the symmetric segment ( x , s , w ) (x,s,w) ( x , s , w ) to ( x , w , s ) (x,w,s) ( x , w , s ) lies in the actual shared arc. Its factor is ( x + w ) / ( x + s ) > 1 (x+w)/(x+s) > 1 ( x + w ) / ( x + s ) > 1 . In coordinates R = x / s R = x/s R = x / s , t = w / s t = w/s t = w / s its physical cost is
A s y m = R R + t + R t [ 2 ( t − 1 ) − R log R + t R + 1 ] . A_{\mathrm{sym}} = \frac{R}{R + t + Rt}\left[2(t - 1) - R\log\frac{R + t}{R + 1}\right]. A sym = R + t + R t R [ 2 ( t − 1 ) − R log R + 1 R + t ] .
The elementary strict bound log g < g − 1 \log g < g-1 log g < g − 1 gives
A s y m > L ( R , t ) : = R ( R + 2 ) ( t − 1 ) ( R + t + R t ) ( R + 1 ) . A_{\mathrm{sym}} > L(R,t) := \frac{R(R+2)(t-1)}{(R+t+Rt)(R+1)}. A sym > L ( R , t ) := ( R + t + R t ) ( R + 1 ) R ( R + 2 ) ( t − 1 ) .
This rational function increases in t t t , since
L t = R ( R + 2 ) ( 2 R + 1 ) ( R + 1 ) ( R + t + R t ) 2 > 0 , L_t = \frac{R(R+2)(2R+1)}{(R+1)(R+t+Rt)^2} > 0, L t = ( R + 1 ) ( R + t + R t ) 2 R ( R + 2 ) ( 2 R + 1 ) > 0 ,
L ( R , m ) − m − 1 m + 1 = ( m − 1 ) ( R − m ) ( R + 1 ) ( m + 1 ) ( R + m + R m ) ≥ 0. L(R,m)-\frac{m-1}{m+1}=\frac{(m-1)(R-m)}{(R+1)(m+1)(R+m+Rm)}\geq0. L ( R , m ) − m + 1 m − 1 = ( R + 1 ) ( m + 1 ) ( R + m + R m ) ( m − 1 ) ( R − m ) ≥ 0.
Hence the actual shared cost exceeds ( m − 1 ) / ( m + 1 ) (m-1)/(m+1) ( m − 1 ) / ( m + 1 ) .
Suppose now J ≤ a ≤ 1 / 5 J\leq a\leq1/5 J ≤ a ≤ 1/5 and set M = ( 1 + a ) / ( 1 − a ) M=(1+a)/(1-a) M = ( 1 + a ) / ( 1 − a ) . Since the shared arc has cost strictly less than J J J , the preceding argument selects a genuine capped middle with min ( R , t ) < M \min(R,t)<M min ( R , t ) < M . Its two adjacent complete costs satisfy A x + A z < J ≤ a A_x+A_z<J\leq a A x + A z < J ≤ a . No third middle or extra cap has been introduced.
All terms in the two cost formulas are positive and x z > 1 / 3 xz>1/3 x z > 1/3 . Thus e − q i > d : = 1 − a D / ( R t ) > 1 − 3 a > 0 e^{-q_i}>d:=1-aD/(Rt)>1-3a>0 e − q i > d := 1 − a D / ( R t ) > 1 − 3 a > 0 . The elementary inequalities 1 − e − q > q e − q / 2 1-e^{-q}>qe^{-q/2} 1 − e − q > q e − q /2 , e q − 1 > q e^q-1>q e q − 1 > q , and 2 sinh q − q > q 2\sinh q-q>q 2 sinh q − q > q imply
ℓ T < a d , ℓ T < B : = a 1 − 3 a . \ell T<\frac{a}{\sqrt d},\qquad\ell T<B:=\frac{a}{\sqrt{1-3a}}. ℓ T < d a , ℓ T < B := 1 − 3 a a .
Define k = ( y + 1 / y ) / ℓ = ( R + 1 ) ( t + 1 ) / ( R + t + 1 ) k=(y+1/y)/\ell=(R+1)(t+1)/(R+t+1) k = ( y + 1/ y ) / ℓ = ( R + 1 ) ( t + 1 ) / ( R + t + 1 ) and g = G / ℓ g=G/\ell g = G / ℓ , where G = ( x − y ) ( z − y ) / y G=(x-y)(z-y)/y G = ( x − y ) ( z − y ) / y . The unbounded ratio sector is controlled without polynomial expansion: D / ( R t ) ≤ 2 + 1 / R D/(Rt)\leq2+1/R D / ( R t ) ≤ 2 + 1/ R and R + t + 1 ≥ R + 2 R+t+1\geq R+2 R + t + 1 ≥ R + 2 give
( R + t + 1 ) 2 d ≥ ( R + 2 ) 2 [ 1 − a ( 2 + 1 / R ) ] > ( R + 1 ) 2 ( 1 − 2 a ) . (R+t+1)^2d\geq(R+2)^2[1-a(2+1/R)]>(R+1)^2(1-2a). ( R + t + 1 ) 2 d ≥ ( R + 2 ) 2 [ 1 − a ( 2 + 1/ R )] > ( R + 1 ) 2 ( 1 − 2 a ) .
Indeed the last difference is
2 R + 3 − a ( 5 R + 10 + 4 / R ) ≥ R + 1 − 4 5 R > 0 , 2R+3-a(5R+10+4/R)\geq R+1-\frac{4}{5R}>0, 2 R + 3 − a ( 5 R + 10 + 4/ R ) ≥ R + 1 − 5 R 4 > 0 ,
and the factor 1 − a ( 2 + 1 / R ) ≥ 1 − 3 a 1-a(2+1/R)\geq1-3a 1 − a ( 2 + 1/ R ) ≥ 1 − 3 a is positive. Applying it and its exchanged version in Corollary B.27 gives, with b = a / 1 − 2 a b=a/\sqrt{1-2a} b = a / 1 − 2 a ,
∣ V ∣ < ( M + 1 ) b 2 , g ℓ T 4 < ( M − 1 ) b 4 . |V|<\frac{(M+1)b}{2},\qquad\frac{g\ell T}{4}<\frac{(M-1)b}{4}. ∣ V ∣ < 2 ( M + 1 ) b , 4 g ℓ T < 4 ( M − 1 ) b .
Here the norm bound holds throughout the support triangle. The nonhorizontal vertex loss is less than ( M − 1 ) b / 4 + ( M + 1 ) 2 b 2 / 16 (M-1)b/4+(M+1)^2b^2/16 ( M − 1 ) b /4 + ( M + 1 ) 2 b 2 /16 . For a horizontal vertex opposite in sign to N x N_x N x , discard its nonpositive linear loss and use the same norm bound. For a horizontal vertex of the same sign, retain both terms. If R ≥ t R\geq t R ≥ t , that vertex is W 2 = ( L z , 0 ) W_2=(L_z,0) W 2 = ( L z , 0 ) , and
N x L z 2 = t z ( t − 1 ) ( R − t ) 4 t D ≤ g ℓ T 4 , L z ℓ t z = ( R + t ) ( t − 1 ) 2 t ( R + t + 1 ) < t − 1 2 t < M − 1 2 . \frac{N_xL_z}{2}=\frac{t_z(t-1)(R-t)}{4t\sqrt D}\leq\frac{g\ell T}{4},\qquad\frac{L_z}{\ell t_z}=\frac{(R+t)(t-1)}{2t(R+t+1)}<\frac{t-1}{2t}<\frac{M-1}{2}. 2 N x L z = 4 t D t z ( t − 1 ) ( R − t ) ≤ 4 g ℓ T , ℓ t z L z = 2 t ( R + t + 1 ) ( R + t ) ( t − 1 ) < 2 t t − 1 < 2 M − 1 .
For t ≥ R t\geq R t ≥ R exchange the scalar names; no actual arc is reversed. The height loss H = ( ℓ − N y ) h ∗ / 2 H=(\ell-N_y)h_*/2 H = ( ℓ − N y ) h ∗ /2 is less than B 2 / 8 B^2/8 B 2 /8 , since ∣ N ∣ < ℓ |N|<\ell ∣ N ∣ < ℓ . Thus the three-vertex bound gives
C > min { L 0 , L h , o , L h , s } . C>\min\{L_0,L_{h,o},L_{h,s}\}. C > min { L 0 , L h , o , L h , s } .
where
L 0 = 3 4 − ( M − 1 ) b / 4 − ( M + 1 ) 2 b 2 / 16 , L_0=\frac{3}{4}-(M-1)b/4-(M+1)^2b^2/16, L 0 = 4 3 − ( M − 1 ) b /4 − ( M + 1 ) 2 b 2 /16 ,
L h , o = 3 4 − ( M + 1 ) 2 b 2 / 16 − B 2 / 8 , L_{h,o}=\frac{3}{4}-(M+1)^2b^2/16-B^2/8, L h , o = 4 3 − ( M + 1 ) 2 b 2 /16 − B 2 /8 ,
L h , s = 3 4 − ( M − 1 ) b / 4 − ( M − 1 ) 2 B 2 / 16 − B 2 / 8. L_{h,s}=\frac{3}{4}-(M-1)b/4-(M-1)^2B^2/16-B^2/8. L h , s = 4 3 − ( M − 1 ) b /4 − ( M − 1 ) 2 B 2 /16 − B 2 /8.
This also covers h ∗ = 0 h_*=0 h ∗ = 0 without dividing by its height.
Set a = 47 / 400 a=47/400 a = 47/400 . Then
M = 447 / 353 , B 2 = 2209 / 103600 , b 2 = 2209 / 122400. M=447/353,\qquad B^2=2209/103600,\qquad b^2=2209/122400. M = 447/353 , B 2 = 2209/103600 , b 2 = 2209/122400.
The rational bracket b < 134341 / 10 6 b < 134341/10^{6} b < 134341/1 0 6 is strict because
( 134341 / 10 6 ) 2 − b 2 = 18154993 153000000000000 > 0. (134341/10^{6})^{2} - b^{2} = \frac{18154993}{153000000000000} > 0. ( 134341/1 0 6 ) 2 − b 2 = 153000000000000 18154993 > 0.
Substituting this upper bracket in the negative linear terms in b b b , and retaining the exact b 2 b^{2} b 2 , gives the respective determinant floors
28035851215757 38130354000000 , 11717258087207 15801218697600 , 11913796512899 16136865500000 . \frac{28035851215757}{38130354000000}, \qquad\frac{11717258087207}{15801218697600}, \qquad\frac{11913796512899}{16136865500000}. 38130354000000 28035851215757 , 15801218697600 11717258087207 , 16136865500000 11913796512899 .
The first is the smallest. Its difference from the harmonic ceiling
H = 29025933167887355473205403 39478417604337498218323684 \mathcal{H} = \frac{29025933167887355473205403}{39478417604337498218323684} H = 39478417604337498218323684 29025933167887355473205403
is the positive rational number
10483829210426125893586595821272197 376331509653305185634762839376034000000 . \frac{10483829210426125893586595821272197}{376331509653305185634762839376034000000}. 376331509653305185634762839376034000000 10483829210426125893586595821272197 .
Hence the actual determinant satisfies C > H > 0 C > \mathcal{H} > 0 C > H > 0 . Independently, h / ℓ < T 2 / 8 < B 2 / 24 h/\ell< T^{2}/8 < B^{2}/24 h / ℓ < T 2 /8 < B 2 /24 , so the actual future trace has ratio ( ℓ − h ) / ℓ > 2484191 / 2486400 > 0 (\ell-h)/\ell> 2484191/2486400 > 0 ( ℓ − h ) / ℓ > 2484191/2486400 > 0 . Only now apply the harmonic calibration to this same complete lift; it gives C < H C < \mathcal{H} C < H , a contradiction. The closed contrary budget included equality, proving the strict assertion. □
The central forward-pair estimate.
Proposition B.37. In a genuine complete positive forward context 012012 on an original minimizer, or a lawful symmetry image of the entire problem, the two complete central physical arcs satisfy
c 2 + c 3 > 180483 500000 > 20901 62500 . c_{2} + c_{3} > \frac{180483}{500000} > \frac{20901}{62500}. c 2 + c 3 > 500000 180483 > 62500 20901 . All six actual arcs, including the positive exterior caps at positions 0 and 5, are retained in one genuine fundamental interval.
Proof. Lemma B.1 makes this six-arc source proper. Choose the full maximizing lift on its genuine containing interval before restriction. Lemmas B.24 and B.25 establish normality from its actual complete caps; jointly normalize the entire lift. Theorem C.10, proved independently in Section C.8.4, applies to these same two central physical charges. The exact difference between the two displayed constants is 13275 / 500000 > 0 13275/500000 > 0 13275/500000 > 0 . No extra cap, multi-period interval or freely chosen endpoint extremal has been introduced. □
Corollary B.38 (A four-charge consequence). The four interior physical arcs of a complete 012012 source under the same hypotheses have cost greater than 1 / 3 1/3 1/3 .
Proof. Proposition B.37 gives c 2 + c 3 > 20901 / 62500 > 1 / 3 c_{2} + c_{3} > 20901/62500 > 1/3 c 2 + c 3 > 20901/62500 > 1/3 . The other two interior arcs have positive cost. □
Reverse windows in a genuine interval.
The next lemma allows the three-fifths estimate to be applied without an additional omitted-arc assumption. It changes the genuine cut and selects a fresh lift; it does not enlarge the analytic domain of Theorem B.47.
Lemma B.39 (Completing the caps of a reverse five-window). Let a normalized original global minimizer of physical cost below 180483 / 100000 180483/100000 180483/100000 contain the actual consecutive pattern 1 ∣ 021 ∣ 0 1 \mid021 \mid0 1 ∣ 021 ∣ 0 , with all three interior maximal arcs complete and positive portions of both exterior maximal arcs in one genuine fundamental interval. Then its three physical interior costs satisfy c 1 + c 2 + c 3 ≥ 3 / 5 c_{1}+c_{2}+c_{3} \ge3/5 c 1 + c 2 + c 3 ≥ 3/5 . Whole lawful images have the same conclusion. No matrix adjoint is transferred from one cut to another.
Proof. The three complete charged arcs occupy a span strictly shorter than the fundamental period, since both retained cap portions have positive duration. Thus the number B B B of maximal arcs in a period satisfies B ≥ 4 B \ge4 B ≥ 4 . If B = 4 B = 4 B = 4 , the two exterior occurrences would be the same physical arc in consecutive periods. The actual twist would require j 4 = j 0 + 1 ( m o d 3 ) j_4 = j_0 + 1 \pmod{3} j 4 = j 0 + 1 ( mod 3 ) , whereas 10210 has j 0 = 1 , j 4 = 0 j_0 = 1, j_4 = 0 j 0 = 1 , j 4 = 0 . This contradiction proves B ≥ 5 B \ge5 B ≥ 5 .
Reanchor at the actual initial switch of the first exterior maximal arc, using Lemma 2.6, whose assertion is for every real cut, including a switch. In its notation G s ( T ) = G e ( s ) − 1 G e ( s ) R ∗ = R ∗ G_s(T) = G_e(s)^{-1}G_e(s)R_* = R_* G s ( T ) = G e ( s ) − 1 G e ( s ) R ∗ = R ∗ and W s ( T ) = Φ W s ( 0 ) W_s(T) = \Phi W_s(0) W s ( T ) = Φ W s ( 0 ) . The same fixed-endpoint homotopy retains the real winding. At a switch the trajectory remains absolutely continuous and the state equations hold almost everywhere; no differentiability of the control at the cut is required. Thus it is a global minimizer of the original free-terminal-time endpoint problem, not a problem with a newly fixed value of T T T . The original maximum principle therefore supplies its own full nonzero maximizing lift with the zero free-time Hamiltonian. The ensuing five entire maximal arcs are now contained in one genuine fundamental interval because 5 ≤ B 5 \le B 5 ≤ B . When B = 5 B = 5 B = 5 they fill the switch-to-switch period; no positive uncharged interval is invented. Its four source ties, between successive arcs of 10210, are still strictly inside ( 0 , T ) (0,T) ( 0 , T ) , so all cap equations use interior switch ties rather than any identification of endpoint covectors. Theorem B.47 requires five complete positive arcs in one genuine fundamental interval, so includes this equality case: neither its statement nor its proof assumes 5 < B 5 < B 5 < B . Choose its own nonzero complete maximizing lift on this reanchored original minimizing interval. Lemma B.48 proves normality before joint normalization; the proper-five proof establishes its own positive determinant and future trace before harmonic calibration. It gives c 1 + c 2 + c 3 ≥ 3 / 5 c_1 + c_2 + c_3 \ge3/5 c 1 + c 2 + c 3 ≥ 3/5 .
Reanchoring and whole lawful transport preserve complete physical costs, the original state/group endpoint problem and real winding. Only these scalar costs are identified with those of the initial interval. In particular, no Casimir or state covector is continued across its cut. □
A complete reverse four-core.
Proposition B.40 (Reverse four-core). In an eligible full positive reverse context 102102 or an additive phase, on a genuine containing full fundamental interval with the one complete lift and cap-derived normality specified in Section B.2.4.1, the four complete interior arcs have physical cost
A = c 1 + c 2 + c 3 + c 4 > 3 / 5 > 62 / 125. A = c_1 + c_2 + c_3 + c_4 > 3/5 > 62/125. A = c 1 + c 2 + c 3 + c 4 > 3/5 > 62/125. Both exterior caps remain positive.
Proof. The actual window 0 , … , 4 0,\ldots,4 0 , … , 4 is 10210 and retains the complete charged arcs 1, 2, 3. Lemma B.39 gives c 1 + c 2 + c 3 ≥ 3 / 5 c_1 + c_2 + c_3 \ge3/5 c 1 + c 2 + c 3 ≥ 3/5 , while c 4 > 0 c_4 > 0 c 4 > 0 . □
Six charged arcs in a reverse eight-context.
Proposition B.41 (Reverse six-core). On a full positive reverse eight-context 10210210, or an additive phase, charge exactly the complete arcs 1 through 6 and retain positive caps 0, 7 in one genuine fundamental minimizing interval. Assume also the separate eligibility of its contained reverse six-context on 2, …, 7. Then
A 6 = c 1 + ⋯ + c 6 ≥ 6 / 5 > 3 / 5. A_6 = c_1 + \cdots+ c_6 \ge6/5 > 3/5. A 6 = c 1 + ⋯ + c 6 ≥ 6/5 > 3/5. Proof. The windows 0 , … , 4 0,\ldots,4 0 , … , 4 and 3 , … , 7 3,\ldots,7 3 , … , 7 both read 10210. Their complete charged triples are 1, 2, 3 and 4, 5, 6. Apply Lemma B.39 to both and add their scalar inequalities. These two charge sets are disjoint; neither choice of containing lift is imposed on the other. □