Introduction
Let
Ξ ( z ) = ξ ( 1 2 + i z ) \Xi(z)=\xi\left(\frac{1}{2}+iz\right) Ξ ( z ) = ξ ( 2 1 + i z )
be the Riemann Ξ \Xi Ξ -function, where ξ ( s ) = 1 2 s ( s − 1 ) π − s / 2 Γ ( s / 2 ) ζ ( s ) \xi(s)=\frac{1}{2}s(s-1)\pi^{-s/2}\Gamma(s/2)\zeta(s) ξ ( s ) = 2 1 s ( s − 1 ) π − s /2 Γ ( s /2 ) ζ ( s ) is the completed zeta function, ζ \zeta ζ is the Riemann zeta function and Γ \Gamma Γ is Euler’s gamma function. Classical approaches to the Riemann hypothesis are organized around the zeros, growth, positivity, and Fourier representation of Ξ \Xi Ξ ; see, for example, [1 , 2 , 3 ] . Here we isolate a different object: a spectral double cover naturally attached to the transverse product
τ x ( T ) = Ξ ( x + i T ) Ξ ( x − i T ) Ξ ( x ) 2 . (1) \tau_x(T)=\frac{\Xi\left(x+i\sqrt{T}\right)\Xi\left(x-i\sqrt{T}\right)}{\Xi(x)^2}. \tag*{(1)} τ x ( T ) = Ξ ( x ) 2 Ξ ( x + i T ) Ξ ( x − i T ) . (1)
The square root disappears from the product, so τ x \tau_x τ x is locally analytic in T T T wherever Ξ ( x ) ≠ 0 \Xi(x)\ne0 Ξ ( x ) = 0 . The construction arose from an RH-motivated P I I I ( D 6 ) P_{\mathrm{III}}(D_6) P III ( D 6 ) and Stieltjes program [4 , 5 , 6 ] , but the present paper has a deliberately narrower objective: to determine, with certification, the geometry and arithmetic of the first compact genus-two core at the arithmetic fixed point x = 0 x=0 x = 0 .
The fixed-point discriminant admits a regular theta-moment numerator. Put
D ( z ) = 1 4 Ξ ( z ) = ∫ 0 ∞ Φ ( u ) cos ( z u ) d u , Y ( T ) = D ( i T ) = ∑ n ≥ 0 M 2 n ( 2 n ) ! T n , D(z)=\frac{1}{4}\Xi(z)=\int_0^\infty\Phi(u)\cos(zu)\,\mathrm{d}u,\qquad Y(T)=D(i\sqrt{T})=\sum_{n\ge0}\frac{M_{2n}}{(2n)!}T^n, D ( z ) = 4 1 Ξ ( z ) = ∫ 0 ∞ Φ ( u ) cos ( z u ) d u , Y ( T ) = D ( i T ) = n ≥ 0 ∑ ( 2 n )! M 2 n T n ,
where Φ \Phi Φ is the standard even theta kernel. The normalized spectral discriminant takes the form
Δ 0 ( T ) = K ( T ) Y ( T ) 6 , \Delta_0(T) = \frac{K(T)}{Y(T)^6}, Δ 0 ( T ) = Y ( T ) 6 K ( T ) ,
with K K K analytic through T = 0 T = 0 T = 0 . This regularization converts branch-point location into the certified study of the zeros of an explicitly controlled entire function.
The first result is geometric. Directed theta-moment and Rouché/Krawczyk bounds isolate exactly six simple branch values inside a conjugation-invariant domain: two real values and two nonreal conjugate pairs. The normalized double cover over this domain is canonically identified with the restriction of a compact hyperelliptic sextic of genus two and real topological type
( g , k , a ) = ( 2 , 1 , 1 ) . (g,k,a) = (2,1,1). ( g , k , a ) = ( 2 , 1 , 1 ) .
Here g g g is the genus, k k k the number of connected components of the real locus, and a = 0 a = 0 a = 0 or 1 1 1 according as that locus divides the complex curve into two components or does not. The same branch sextic has two additional geometric interpretations. First, it admits an orientation-differential realization with the ( 2 , 4 , 4 ) (2,4,4) ( 2 , 4 , 4 ) pole signature associated with P I I I ( D 6 ) P_{\mathrm{III}}(D_6) P III ( D 6 ) . Second, and more substantively, it is certified to coincide, after a real affine change of spectral coordinate, with the sextic numerator arising from the scalar normal form of a real D 6 D_6 D 6 Lax phase point. We call the latter hyperelliptic double the scalar-normal-form orientation curve ; it should not be confused with the invariant Hitchin characteristic spectral curve. The Jacobian is the abelian variety of degree-zero divisor classes of the curve. Its quotient by the sign involution, followed by resolution of the sixteen nodes, is the associated Kummer K3 surface. A K3 surface is a smooth simply connected compact complex surface with trivial canonical bundle. Explicit Kummer and Shioda–Inose models then make the Humbert and Néron–Severi questions computationally accessible.
The second result is arithmetic. The global Humbert analysis is most naturally written in a canonical Vinnikov basis with H = I 2 H = I_2 H = I 2 , whereas the certified action integrals use the natural three-cut basis described below. In that cut basis ( α 1 , α 2 , β 1 , β 2 ) (\alpha_1,\alpha_2,\beta_1,\beta_2) ( α 1 , α 2 , β 1 , β 2 ) the real involution has matrix
τ ∗ α = α , τ ∗ β = H c u t α − β , H c u t = ( 1 1 1 0 ) . \tau_\ast\alpha= \alpha,\qquad\tau_\ast\beta= H_{\mathrm{cut}}\alpha- \beta,\qquad H_{\mathrm{cut}} =
\begin{pmatrix}
1 & 1\\
1 & 0
\end{pmatrix}. τ ∗ α = α , τ ∗ β = H cut α − β , H cut = ( 1 1 1 0 ) .
For the action differential ϖ 0 = Δ 0 ( T ) d T \varpi_0 = \sqrt{\Delta_0(T)}\,\mathrm{d}T ϖ 0 = Δ 0 ( T ) d T , write
u j = ∮ α j ϖ 0 , v j = ∮ β j ϖ 0 , v = 1 2 H c u t u + i r . u_j = \oint_{\alpha_j} \varpi_0,\qquad v_j = \oint_{\beta_j} \varpi_0,\qquad v = \frac{1}{2}H_{\mathrm{cut}}u + ir. u j = ∮ α j ϖ 0 , v j = ∮ β j ϖ 0 , v = 2 1 H cut u + i r .
The four calibrated cut cycles lie inside the certified normalized core and form a symplectic rank-four lattice Γ c o r e \Gamma_{\mathrm{core}} Γ core (its intersection pairing is integral, alternating and nondegenerate), abstractly identified with H 1 ( X 2 , Z ) H_1(X_2,\mathbb{Z}) H 1 ( X 2 , Z ) . The action homomorphism is evaluated on these actual core representatives, and we define
Λ 2 = ker ( A c o r e : Γ c o r e → C , γ ↦ ∮ γ ϖ 0 ) . \Lambda_2 = \ker\left(\mathcal{A}_{\mathrm{core}}:\Gamma_{\mathrm{core}}\to\mathbb{C},\ \gamma\mapsto\oint_\gamma\varpi_0\right). Λ 2 = ker ( A core : Γ core → C , γ ↦ ∮ γ ϖ 0 ) .
This calibrated zero-action lattice is trivial exactly when both real ratios u 1 / u 2 u_1/u_2 u 1 / u 2 and r 1 / r 2 r_1/r_2 r 1 / r 2 are irrational. We do not prove those irrationalities. Instead, fully directed contour integration and exact Farey-neighbor arithmetic give
13749 419336 < u 1 u 2 < 20219 616667 , 16579 8678 < − r 1 r 2 < 252355 132091 , \frac{13749}{419336} < \frac{u_1}{u_2} < \frac{20219}{616667},\qquad
\frac{16579}{8678} < -\frac{r_1}{r_2} < \frac{252355}{132091}, 419336 13749 < u 2 u 1 < 616667 20219 , 8678 16579 < − r 2 r 1 < 132091 252355 ,
and therefore the denominator barriers
q u ≥ 1,036,003 , q r ≥ 140,769. q_u \ge1{,}036{,}003,\qquad q_r \ge140{,}769. q u ≥ 1 , 036 , 003 , q r ≥ 140 , 769.
for any rational value in lowest terms. This is a finite-height nonresonance theorem, not an irrationality proof.
This motivates the structural conjecture
Conjecture 1.1 (Spectral nonresonance at the arithmetic fixed point). For the calibrated x = 0 x = 0 x = 0 genus-two core,
u 1 u 2 ∉ Q , r 1 r 2 ∉ Q . \frac{u_1}{u_2} \notin\mathbb{Q}, \qquad\frac{r_1}{r_2} \notin\mathbb{Q}. u 2 u 1 ∈ / Q , r 2 r 1 ∈ / Q . Equivalently, Λ 2 = 0 \Lambda_2 = 0 Λ 2 = 0 .
The conjecture is deliberately separated from the certified Farey theorem: finite-width interval enclosures can exclude rational values only up to finite denominator height.
The Kummer/Humbert analysis provides a complementary measure of arithmetic genericity. A Humbert surface H D H_D H D is the locus of principally polarized abelian surfaces admitting an additional non-scalar endomorphism symmetric for the polarization, of Humbert discriminant D D D . The Néron–Severi group consists of divisor classes modulo algebraic equivalence; its rank is the Picard number. We exclude the nine loci
H 4 , H 5 , H 8 , H 9 , H 12 , H 13 , H 16 , H 17 , H 20 , H_4, H_5, H_8, H_9, H_{12}, H_{13}, H_{16}, H_{17}, H_{20}, H 4 , H 5 , H 8 , H 9 , H 12 , H 13 , H 16 , H 17 , H 20 ,
obtain a Mordell–Weil height barrier h ^ ( P ) ≥ 21 / 2 \hat{h}(P) \ge21/2 h ^ ( P ) ≥ 21/2 for any hypothetical nonzero section on the Shioda–Inose model, and reduce the full real-slice Humbert problem to two exact arithmetic statements about the imaginary part of the Vinnikov period matrix: irrationality of its determinant and Q \mathbb{Q} Q -linear independence of its three entries. Here the Mordell–Weil group is the group of sections of the elliptic fibration, and h ^ \hat{h} h ^ is its canonical quadratic height in Shioda’s normalization. The union of Humbert divisors is dense on this real slice, so finite neighborhood exclusion cannot settle the global Picard-rank question.
The heat time is denoted by κ \kappa κ ; the Polymath time is t = 4 κ t = 4\kappa t = 4 κ , as fixed in (90) . The de Bruijn–Newman constant Λ \Lambda Λ is the threshold for reality of all zeros of H t H_t H t : they are all real exactly when t ≥ Λ t \ge\Lambda t ≥ Λ . This real number is distinct from the zero-action lattice Λ 2 \Lambda_2 Λ 2 . The heat deformation connects the geometry to the arithmetic. It transports the selected branch points and calibrated contours, so the same theta-derived data determine both the motion of the curve and the motion of its action periods. The resulting positive ratio velocities make rational crossings transverse: each fixed nonzero cycle can resonate at most once locally, and almost every nearby heat time has quantitative action separation. Uniform bounds also preserve the displayed Farey windows on an explicit heat interval. These generic statements do not settle the irrationality conjecture at the distinguished time κ = 0 \kappa= 0 κ = 0 .
The phase–transverse identity supplies a second connection, from the same discriminant to global zero control. A uniform off-axis phase sign gives a zero strip, which de Bruijn contraction converts into an upper bound for the Newman constant. Effective approximation, exterior estimates and a finite rectangular nonvanishing cover implement this route. The final bound Λ ≤ 0.1732 \Lambda\le0.1732 Λ ≤ 0.1732 combines that cover with the published finite-RH and rectangular time-barrier verifications. It is logically independent of the Humbert exclusions and the period-irrationality conjecture. The companion studies [7 , 8 ] develop separate collective-contraction and classical-strip comparisons, with their own stated premises.
The distinction between local arithmetic and global zero reality is essential. Almost every sufficiently small negative heat time has a nonresonant core and nonreal heat zeros simultaneously. Thus generic core nonresonance alone cannot characterize reality of the full heat family. At κ = 0 \kappa= 0 κ = 0 , no necessity or sufficiency of Λ 2 = 0 \Lambda_2 = 0 Λ 2 = 0 for RH is known.
The paper is organized as follows. Section § 2 derives the regular theta-moment discriminant and its phase interpretation. Section § 3 constructs the real genus-two core, its D 6 D_6 D 6 realization and K 3 K_3 K 3 models, and isolates the Humbert/Picard question. Section § 4 studies the moving branch geometry and its relation to isomonodromic directions. Sections 5 and 6 identify and differentiate the calibrated action lattice. Section 7 gives finite-height arithmetic certificates and the separate global Newman-bound argument. Section 8 explains the numerical architecture; Section 9 returns to the unresolved arithmetic questions. Table 1 locates the principal results in this order.
Table 1. Guide to the principal results and open questions.
For clarity, the status labels used in a few intermediate statements have the following meaning: status [A] denotes an analytic statement or a computer-assisted statement backed by a directed/exact certificate in the supplement, whereas status [N] denotes a numerical observation not used as theorem input.
Theta moments and the fixed-point discriminant
We use the classical theta kernel
Φ ( u ) = ∑ n = 1 ∞ ( 2 π 2 n 4 e 9 u / 2 − 3 π n 2 e 5 u / 2 ) e − π n 2 e 2 u . (2) \Phi(u) = \sum_{n=1}^{\infty} \left(2\pi^{2}n^{4}e^{9u/2} - 3\pi n^{2}e^{5u/2}\right)e^{-\pi n^{2}e^{2u}}. \tag*{(2)} Φ ( u ) = n = 1 ∑ ∞ ( 2 π 2 n 4 e 9 u /2 − 3 π n 2 e 5 u /2 ) e − π n 2 e 2 u . (2)
The Jacobi transformation makes Φ \Phi Φ even; it is real analytic, strictly positive, and doubly exponentially decaying together with all derivatives. With
D ( z ) = ∫ 0 ∞ Φ ( u ) cos ( z u ) d u , Ξ ( z ) = 4 D ( z ) , (3) D(z)=\int_{0}^{\infty}\Phi(u)\cos(zu)\,\mathrm{d}u,\qquad\Xi(z)=4D(z), \tag*{(3)} D ( z ) = ∫ 0 ∞ Φ ( u ) cos ( z u ) d u , Ξ ( z ) = 4 D ( z ) , (3)
put
M m = ∫ 0 ∞ Φ ( u ) u m d u , Y ( T ) = D ( i T ) = ∑ n ≥ 0 M 2 n ( 2 n ) ! T n . (4) M_m=\int_{0}^{\infty}\Phi(u)u^m\,\mathrm{d}u,\qquad Y(T)=D(i\sqrt{T})=\sum_{n\geq0}\frac{M_{2n}}{(2n)!}T^n. \tag*{(4)} M m = ∫ 0 ∞ Φ ( u ) u m d u , Y ( T ) = D ( i T ) = n ≥ 0 ∑ ( 2 n )! M 2 n T n . (4)
Thus Y Y Y is entire and
4 Y ( T ) = ξ ( 1 2 + T ) . 4Y(T)=\xi\left(\frac{1}{2}+\sqrt{T}\right). 4 Y ( T ) = ξ ( 2 1 + T ) .
For real x x x with Ξ ( x ) ≠ 0 \Xi(x)\ne0 Ξ ( x ) = 0 , define
τ x ( T ) = Ξ ( x + i T ) Ξ ( x − i T ) Ξ ( x ) 2 , A x ( T ) = d d T log τ x ( T ) , σ x ( T ) = T A x ( T ) . (5) \tau_x(T)=\frac{\Xi(x+i\sqrt{T})\Xi(x-i\sqrt{T})}{\Xi(x)^2},\qquad\mathcal{A}_x(T)=\frac{\mathrm{d}}{\mathrm{d}T}\log\tau_x(T),\qquad\sigma_x(T)=T\mathcal{A}_x(T). \tag*{(5)} τ x ( T ) = Ξ ( x ) 2 Ξ ( x + i T ) Ξ ( x − i T ) , A x ( T ) = d T d log τ x ( T ) , σ x ( T ) = T A x ( T ) . (5)
At x = 0 x=0 x = 0 , evenness gives
A 0 ( T ) = 2 Y ′ ( T ) Y ( T ) . (6) \mathcal{A}_0(T)=2\frac{Y'(T)}{Y(T)}. \tag*{(6)} A 0 ( T ) = 2 Y ( T ) Y ′ ( T ) . (6)
A phase–transverse-amplitude interpretation
The logarithmic derivative in (5) has a direct phase interpretation that is useful independently of the later branch-point construction. For T > 0 T>0 T > 0 at which Ξ ( x ± i T ) ≠ 0 \Xi(x\pm i\sqrt{T})\ne0 Ξ ( x ± i T ) = 0 , define the unnormalized transverse amplitude
B x ( T ) : = ∂ T log { Ξ ( x + i T ) Ξ ( x − i T ) } . (7) \mathcal{B}_x(T):=\partial_T\log\left\{\Xi(x+i\sqrt{T})\Xi(x-i\sqrt{T})\right\}. \tag*{(7)} B x ( T ) := ∂ T log { Ξ ( x + i T ) Ξ ( x − i T ) } . (7)
For real x x x the product in braces is ∣ Ξ ( x + i T ) ∣ 2 > 0 |\Xi(x+i\sqrt{T})|^2>0 ∣Ξ ( x + i T ) ∣ 2 > 0 . Whenever Ξ ( x ) ≠ 0 \Xi(x)\ne0 Ξ ( x ) = 0 , the normalization in (5) is independent of T T T , so B x = A x \mathcal{B}_x=\mathcal{A}_x B x = A x .
Proposition 2.1 (Exact phase–transverse-amplitude bridge). Fix η > 0 \eta>0 η > 0 and a real x x x for which Ξ ( x ± i η ) ≠ 0 \Xi(x\pm i\eta)\ne0 Ξ ( x ± i η ) = 0 . Choose a continuous local branch
P η ( s ) : = arg Ξ ( s − i η ) P_\eta(s):=\arg\Xi(s-i\eta) P η ( s ) := arg Ξ ( s − i η ) near s = x s=x s = x . Then
d d s P η ( s ) ∣ s = x = − Im Ξ ′ ( x + i η ) Ξ ( x + i η ) = η B x ( η 2 ) . (8) \left.\frac{\mathrm{d}}{\mathrm{d}s}P_\eta(s)\right|_{s=x}=-\operatorname{Im}\frac{\Xi'(x+i\eta)}{\Xi(x+i\eta)}=\eta\mathcal{B}_x(\eta^2). \tag*{(8)} d s d P η ( s ) s = x = − Im Ξ ( x + i η ) Ξ ′ ( x + i η ) = η B x ( η 2 ) . (8) If also Ξ ( x ) ≠ 0 \Xi(x)\ne0 Ξ ( x ) = 0 , the last term is η A x ( η 2 ) \eta\mathcal{A}_x(\eta^2) η A x ( η 2 ) . In particular, at the arithmetic fixed point,
d d s arg Ξ ( s − i η ) ∣ s = 0 = 2 η Y ′ ( η 2 ) Y ( η 2 ) . (9) \left.\frac{\mathrm{d}}{\mathrm{d}s}\arg\Xi(s-i\eta)\right|_{s=0}=2\eta\frac{Y'(\eta^2)}{Y(\eta^2)}. \tag*{(9)} d s d arg Ξ ( s − i η ) s = 0 = 2 η Y ( η 2 ) Y ′ ( η 2 ) . (9) Proof. Reality of Ξ \Xi Ξ on the real axis gives Ξ ′ ( x − i η ) / Ξ ( x − i η ) = q ‾ \Xi'(x-i\eta)/\Xi(x-i\eta)=\overline{q} Ξ ′ ( x − i η ) /Ξ ( x − i η ) = q , where q = Ξ ′ ( x + i η ) / Ξ ( x + i η ) q=\Xi'(x+i\eta)/\Xi(x+i\eta) q = Ξ ′ ( x + i η ) /Ξ ( x + i η ) . Therefore
d d s arg Ξ ( s − i η ) ∣ s = x = Im q ‾ = − Im q . \left.\frac{\mathrm{d}}{\mathrm{d}s}\arg\Xi(s-i\eta)\right|_{s=x}=\operatorname{Im}\overline{q}=-\operatorname{Im}q. d s d arg Ξ ( s − i η ) s = x = Im q = − Im q . On the other hand, differentiating (7) at T = η 2 T=\eta^2 T = η 2 gives
B x ( η 2 ) = i q − i q ‾ 2 η = − Im q η . \mathcal{B}_x(\eta^2)=\frac{iq-i\overline{q}}{2\eta}=-\frac{\operatorname{Im}q}{\eta}. B x ( η 2 ) = 2 η i q − i q = − η Im q . This proves (8) ; (9) follows from (6) . □
Remark 2.2 (Relation to the heat-flow companion). The identity above is static and requires neither RH nor any Newman barrier estimate. Under the de Bruijn–Newman deformation the same calculation identifies the off-axis logarithmic-derivative probe used in the collective-contraction argument with the heat-evolved transverse amplitude. § 4 makes this local evolution explicit. Global localisation, source transfer, and contact arguments remain in the companion heat-flow studies [7 , 8 ] ; they are not used in the fixed-point genus-two, Humbert, or period arguments here.
Let A ∗ = A 0 ( 0 ) = M 2 / M 0 A_*=\mathcal{A}_0(0)=M_2/M_0 A ∗ = A 0 ( 0 ) = M 2 / M 0 and set
P ( T ) = σ 0 ′ ( T ) = A 0 ( T ) + T A 0 ′ ( T ) . P(T)=\sigma_0'(T)=\mathcal{A}_0(T)+T\mathcal{A}_0'(T). P ( T ) = σ 0 ′ ( T ) = A 0 ( T ) + T A 0 ′ ( T ) .
The normalized discriminant used here is
Δ 0 ( T ) = { σ 0 ′ ′ ( T ) } 2 − 4 { σ 0 ′ ( T ) − A ∗ T } 2 { P ( T ) − T 2 A 0 ′ ( T ) } . (10) \Delta_0(T)=\{\sigma_0''(T)\}^2-4\left\{\frac{\sigma_0'(T)-A_*}{T}\right\}^2\{P(T)-T^2\mathcal{A}_0'(T)\}. \tag*{(10)} Δ 0 ( T ) = { σ 0 ′′ ( T ) } 2 − 4 { T σ 0 ′ ( T ) − A ∗ } 2 { P ( T ) − T 2 A 0 ′ ( T )} . (10)
The quotient in the middle is analytic at T = 0 T=0 T = 0 .
To remove all apparent singular behavior, write Y j = Y ( j ) Y_j=Y^{(j)} Y j = Y ( j ) and
N 2 = Y 2 Y − Y 1 2 , N 3 = Y 3 Y 2 − 3 Y 2 Y 1 Y + 2 Y 1 3 , N_2=Y_2Y-Y_1^2,\qquad N_3=Y_3Y^2-3Y_2Y_1Y+2Y_1^3, N 2 = Y 2 Y − Y 1 2 , N 3 = Y 3 Y 2 − 3 Y 2 Y 1 Y + 2 Y 1 3 ,
and define
S = 4 N 2 Y + 2 T N 3 , S=4N_2Y+2TN_3, S = 4 N 2 Y + 2 T N 3 ,
E = 2 Y 1 Y + 2 T N 2 − A ∗ Y 2 T , \mathcal{E}=\frac{2Y_1Y+2TN_2-A_*Y^2}{T}, E = T 2 Y 1 Y + 2 T N 2 − A ∗ Y 2 ,
R = 2 Y 1 Y + 2 ( T − T 2 ) N 2 . (11) \mathcal{R}=2Y_1Y+2(T-T^2)N_2. \tag*{(11)} R = 2 Y 1 Y + 2 ( T − T 2 ) N 2 . (11)
The numerator defining E \mathcal{E} E vanishes at T = 0 T=0 T = 0 .
Proposition 2.3 (Regular entire numerator). On every domain on which Y ≠ 0 Y\ne0 Y = 0 ,
Δ 0 ( T ) = K ( T ) Y ( T ) 6 , K ( T ) = S ( T ) 2 − 4 E ( T ) 2 R ( T ) . (12) \Delta_0(T)=\frac{K(T)}{Y(T)^6},\qquad K(T)=S(T)^2-4\mathcal{E}(T)^2\mathcal{R}(T). \tag*{(12)} Δ 0 ( T ) = Y ( T ) 6 K ( T ) , K ( T ) = S ( T ) 2 − 4 E ( T ) 2 R ( T ) . (12) In particular, K K K is analytic through T = 0 T=0 T = 0 .
Proof. From (6) ,
A 0 ′ = 2 N 2 Y 2 , A 0 ′ ′ = 2 N 3 Y 3 . \mathcal{A}_0'=2\frac{N_2}{Y^2},\qquad\mathcal{A}_0''=2\frac{N_3}{Y^3}. A 0 ′ = 2 Y 2 N 2 , A 0 ′′ = 2 Y 3 N 3 .
A direct substitution gives
P ′ = S Y 3 , P − A ∗ T = E Y 2 , P − T 2 A 0 ′ = R Y 2 , P'=\frac{S}{Y^3},\qquad\frac{P-A_*}{T}=\frac{\mathcal{E}}{Y^2},\qquad P-T^2\mathcal{A}_0'=\frac{\mathcal{R}}{Y^2}, P ′ = Y 3 S , T P − A ∗ = Y 2 E , P − T 2 A 0 ′ = Y 2 R ,
and (12) follows from (10) .
The spectral curve is the normalization of
Σ 0 : y 2 = Δ 0 ( T ) , (13) \Sigma_0:\qquad y^2=\Delta_0(T), \tag*{(13)} Σ 0 : y 2 = Δ 0 ( T ) , (13)
with action differential
ϖ 0 = y d T = Δ 0 ( T ) d T . (14) \varpi_0=y\,dT=\sqrt{\Delta_0(T)}\,dT. \tag*{(14)} ϖ 0 = y d T = Δ 0 ( T ) d T . (14)
Every simple zero of K K K away from Y = 0 Y=0 Y = 0 is a simple ramification point of (13) .
Certified six-point geometry
All branch-point statements in this section are directed interval results. Directed interval arithmetic encloses exact quantities with outward-rounded bounds. Rouché tests count zeros by comparing an analytic function with a dominant term on a contour; Krawczyk tests use an interval Newton-type inclusion to certify existence and uniqueness of a solution. The computations use the theta representation (2) , moment intervals from (4) , analytic tails, and Rouché tests for the complex zeros. Details of the certification architecture are summarized in § 8 .
Theorem 3.1 (Real branch pair). There are unique simple real zeros a , b a,b a , b of K K K in
− 65.85 < a < − 65.70 , 46.3829 < b < 46.3830. (15) -65.85 < a < -65.70,\qquad46.3829 < b < 46.3830. \tag*{(15)} − 65.85 < a < − 65.70 , 46.3829 < b < 46.3830. (15) Moreover Y Y Y is nonzero on the connecting real corridor and Δ 0 ( T ) > 0 \Delta_0(T)>0 Δ 0 ( T ) > 0 for a < T < b a<T<b a < T < b .
Proof. The directed moment certificate gives
K ( − 65.85 ) < − 2.72 × 10 − 20 , K ( − 65.70 ) > 4.31 × 10 − 20 , K(-65.85)<-2.72\times10^{-20},\qquad K(-65.70)>4.31\times10^{-20}, K ( − 65.85 ) < − 2.72 × 1 0 − 20 , K ( − 65.70 ) > 4.31 × 1 0 − 20 , with K ′ > 7.03 × 10 − 19 K'>7.03\times10^{-19} K ′ > 7.03 × 1 0 − 19 throughout the left interval. On the right interval it gives
K ( 46.3829 ) > 5.5399 × 10 − 17 , K ( 46.3830 ) < − 1.3335 × 10 − 18 , K(46.3829)>5.5399\times10^{-17},\qquad K(46.3830)<-1.3335\times10^{-18}, K ( 46.3829 ) > 5.5399 × 1 0 − 17 , K ( 46.3830 ) < − 1.3335 × 1 0 − 18 , with K ′ < − 5.6734 × 10 − 13 K'<-5.6734\times10^{-13} K ′ < − 5.6734 × 1 0 − 13 . Hence each interval contains exactly one simple zero. Directed polynomial evaluation gives Y > 0 Y>0 Y > 0 on a covering of the whole corridor and K > 0 K>0 K > 0 on a finite rational-cell covering of the interior, with monotonicity handling the two endpoint boxes. Equation (12) then yields Δ 0 > 0 \Delta_0>0 Δ 0 > 0 on ( a , b ) (a,b) ( a , b ) . □
Theorem 3.2 (First nonreal branch pair). Let
c ^ 1 = − 231.0846402275628298879620029048 + 59.4949465030781144742436618558 i . \widehat{c}_1=-231.0846402275628298879620029048+59.4949465030781144742436618558i. c 1 = − 231.0846402275628298879620029048 + 59.4949465030781144742436618558 i . Then K K K has exactly one simple zero c 1 c_1 c 1 in
∣ T − c ^ 1 ∣ < 10 − 9 , (16) |T-\widehat{c}_1|<10^{-9}, \tag*{(16)} ∣ T − c 1 ∣ < 1 0 − 9 , (16) and exactly one simple zero c ‾ 1 \overline{c}_1 c 1 in the conjugate disk. The function Y Y Y is nonzero on both disks.
Proof. For numerical stability the directed certificate applies Rouché’s theorem to Q = T 2 K Q=T^2K Q = T 2 K . On ∣ T − c ^ 1 ∣ = 10 − 9 |T-\widehat{c}_1|=10^{-9} ∣ T − c 1 ∣ = 1 0 − 9 , the finite Taylor remainder and the analytic entire-series tail give
∣ Q ( c ^ 1 ) ∣ + R n l + R t a i l < 5.352016461 × 10 − 32 < 9.054459459 × 10 − 32 ≤ ∣ Q ′ ( c ^ 1 ) ∣ 10 − 9 . |Q(\widehat{c}_1)|+R_{\mathrm{nl}}+R_{\mathrm{tail}}<5.352016461\times10^{-32}<9.054459459\times10^{-32}\leq|Q'(\widehat{c}_1)|10^{-9}. ∣ Q ( c 1 ) ∣ + R nl + R tail < 5.352016461 × 1 0 − 32 < 9.054459459 × 1 0 − 32 ≤ ∣ Q ′ ( c 1 ) ∣1 0 − 9 . It also gives ∣ Y ∣ > 4.356747335 × 10 − 4 |Y|>4.356747335\times10^{-4} ∣ Y ∣ > 4.356747335 × 1 0 − 4 throughout the disk. Hence the disk contains exactly one zero of Q Q Q , counted with multiplicity; because it is away from T = 0 T=0 T = 0 , this is exactly one simple zero of K K K . Conjugation gives the lower-half-plane zero. □
Theorem 3.3 (Second nonreal branch pair). Let
c ^ 2 = − 269.8852991848204115111948309919 + 150.2255429937459843575350632811 i . \widehat{c}_2=-269.8852991848204115111948309919+150.2255429937459843575350632811i. c 2 = − 269.8852991848204115111948309919 + 150.2255429937459843575350632811 i . Then K K K has exactly one simple zero c 2 c_2 c 2 in
∣ T − c ^ 2 ∣ < 10 − 9 , (17) |T-\widehat{c}_2|<10^{-9}, \tag*{(17)} ∣ T − c 2 ∣ < 1 0 − 9 , (17) and exactly one simple zero c ‾ 2 \overline{c}_2 c 2 in the conjugate disk. The function Y Y Y is nonzero on both disks.
Proof. Again use the regularized numerator Q = T 2 K Q=T^2K Q = T 2 K . On ∣ T − c ^ 2 ∣ = 10 − 9 |T-\widehat{c}_2|=10^{-9} ∣ T − c 2 ∣ = 1 0 − 9 the outward bounds give
∣ Q ( c ^ 2 ) ∣ + R n l + R t a i l < 1.603287091 × 10 − 32 < 1.746001014 × 10 − 32 ≤ ∣ Q ′ ( c ^ 2 ) ∣ 10 − 9 . |Q(\widehat{c}_2)|+R_{\mathrm{nl}}+R_{\mathrm{tail}}<1.603287091\times10^{-32}<1.746001014\times10^{-32}\leq|Q'(\widehat{c}_2)|10^{-9}. ∣ Q ( c 2 ) ∣ + R nl + R tail < 1.603287091 × 1 0 − 32 < 1.746001014 × 1 0 − 32 ≤ ∣ Q ′ ( c 2 ) ∣1 0 − 9 . The same disk computation gives ∣ Y ∣ > 6.323600546 × 10 − 4 |Y|>6.323600546\times10^{-4} ∣ Y ∣ > 6.323600546 × 1 0 − 4 . Rouché’s theorem therefore gives exactly one simple zero of K K K in the disk and one in its conjugate. □
Corollary 3.5 (Genus-two core). Let Ω 2 \Omega_2 Ω 2 be such a Jordan domain. The normalization of the two-sheeted cover (13) over Ω 2 \Omega_2 Ω 2 has genus two and two boundary components.
Proof. Riemann–Hurwitz gives
χ ( Ω ~ 2 ) = 2 χ ( Ω 2 ) − 6 = − 4. \chi(\widetilde{\Omega}_2)=2\chi(\Omega_2)-6=-4. χ ( Ω 2 ) = 2 χ ( Ω 2 ) − 6 = − 4. Because the branch number is even, the boundary of Ω 2 \Omega_2 Ω 2 has two lifts. Thus − 4 = 2 − 2 g − 2 -4=2-2g-2 − 4 = 2 − 2 g − 2 , hence g = 2 g=2 g = 2 . □
The six selected roots define the monic real sextic
P X ( T ) = ( T − a ) ( T − b ) ( T − c 1 ) ( T − c ‾ 1 ) ( T − c 2 ) ( T − c ‾ 2 ) . (18) P_X(T)=(T-a)(T-b)(T-c_1)(T-\overline{c}_1)(T-c_2)(T-\overline{c}_2). \tag*{(18)} P X ( T ) = ( T − a ) ( T − b ) ( T − c 1 ) ( T − c 1 ) ( T − c 2 ) ( T − c 2 ) . (18)
On Ω 2 \Omega_2 Ω 2 write K = H P X K=HP_X K = H P X . The quotient H H H extends holomorphically and never vanishes there, because the six selected zeros are simple and there are no other zeros of K K K in Ω 2 \Omega_2 Ω 2 . Since Ω 2 \Omega_2 Ω 2 is simply connected, choose S S S with S 2 = H S^2=H S 2 = H . At the level of function fields, the substitution U = y Y 3 U=yY^3 U = y Y 3 turns y 2 = K / Y 6 y^2=K/Y^6 y 2 = K / Y 6 into U 2 = K U^2=K U 2 = K ; after normalization it therefore extends across any removable points of the original affine presentation. The further change W X = U / S W_X=U/S W X = U / S identifies the normalized core with the restriction of
X 2 : W X 2 = P X ( T ) (19) X_2:\quad W_X^2=P_X(T) \tag*{(19)} X 2 : W X 2 = P X ( T ) (19)
to Ω 2 \Omega_2 Ω 2 . We henceforth define X 2 X_2 X 2 to be the smooth projective completion of (19) . This gives an intrinsic compact model for the certified core and does not assert that the full noncompact spectral curve has no further branch points.
Complex conjugation acts on this model by
τ ( T , W X ) = ( T ‾ , W ‾ X ) . (20) \tau(T,W_X)=(\overline{T},\overline{W}_X). \tag*{(20)} τ ( T , W X ) = ( T , W X ) . (20)
Proposition 3.7 (Real topological type). The compact real surface ( X 2 , τ ) (X_2,\tau) ( X 2 , τ ) has topological type
( g , k , a ) = ( 2 , 1 , 1 ) . (g,k,a)=(2,1,1). ( g , k , a ) = ( 2 , 1 , 1 ) . Proof. The smooth projective completion of a sextic double cover with six simple branch values has genus two. For real T T T , the two conjugate quadratic factors in (18) are strictly positive, so the sign of P X ( T ) P_X(T) P X ( T ) is the sign of ( T − a ) ( T − b ) (T-a)(T-b) ( T − a ) ( T − b ) . Thus the compact real locus consists of one oval, hence k = 1 k=1 k = 1 . To determine the dividing invariant, use the hyperelliptic analysis of Gross–Harris [9 ] , §6, proof of Prop. 6.3: if a real hyperelliptic curve is dividing, then in a real even-degree model its branch roots are either all real or all nonreal. Here the branch configuration is mixed (exactly two real roots and two conjugate pairs), so X 2 X_2 X 2 is nondividing and a = 1 a=1 a = 1 . □
Corollary 3.8 (Refined real boxes used in the Lax realization). The real branch points satisfy
− 65.78507454 < a < − 65.78507452 , 46.38299764 < b < 46.38299766. (21) -65.78507454 < a < -65.78507452,\qquad46.38299764 < b < 46.38299766. \tag*{(21)} − 65.78507454 < a < − 65.78507452 , 46.38299764 < b < 46.38299766. (21) Together with the 10 − 9 10^{-9} 1 0 − 9 complex disks in theorems 3.2 and 3.3, these are the branch boxes used in the certified D 6 D_6 D 6 coefficient match.
Proof. The coarse real-root theorem gives exactly one simple zero in each of [ − 65.85 , − 65.70 ] [-65.85,-65.70] [ − 65.85 , − 65.70 ] and [ 46.3829 , 46.3830 ] [46.3829,46.3830] [ 46.3829 , 46.3830 ] . The refined M 80 M_{80} M 80 moment certificate gives opposite directed signs of Δ 0 \Delta_0 Δ 0 at the two endpoints of each interval in (21) ; hence the unique coarse roots lie in the displayed subintervals. □
A PIII(D6) orientation-double bridge
The genus-two value also has an independent interpretation from the decorated P I I I ( D 6 ) P_{\mathrm{III}}(D_6) P III ( D 6 ) geometry. Chekhov–Mazzocco–Rubtsov associate to P I I I ( D 6 ) P_{\mathrm{III}}(D_6) P III ( D 6 ) the cusp/Katz signature
( 0 , 2 , 2 ) , ( 0 , 1 , 1 ) , (0,2,2),\qquad(0,1,1), ( 0 , 2 , 2 ) , ( 0 , 1 , 1 ) ,
and, on the Hitchin side, the pole-order pattern
( 2 , 4 , 4 ) (22) (2,4,4) \tag*{(22)} ( 2 , 4 , 4 ) (22)
for the corresponding meromorphic quadratic differential ϕ \phi ϕ ; see Appendix A, Table 2 of [10 ] . The following elementary observation gives a more precise bridge to the present six-branch-point core than a Schottky-double analogy.
Table 2. Certified coefficients of the conditioned Xi sextic (54) .
Proposition 3.9 (D 6 D_6 D 6 orientation double and the six-point core). Let ϕ \phi ϕ be a meromorphic quadratic differential on P 1 \mathbb{P}^1 P 1 with pole orders ( 2 , 4 , 4 ) (2,4,4) ( 2 , 4 , 4 ) , and suppose its zeros are simple. Then ϕ \phi ϕ has exactly six zeros, and its canonical orientation double
π D 6 : C D 6 ⟶ P 1 , λ 2 = ϕ , (23) \pi_{D_6}: C_{D_6} \longrightarrow\mathbb{P}^1,\qquad\lambda^2=\phi, \tag*{(23)} π D 6 : C D 6 ⟶ P 1 , λ 2 = ϕ , (23) is a genus-two hyperelliptic curve. After a Möbius normalization placing the two fourth-order poles at 0 , 1 0,1 0 , 1 and the second-order pole at ∞ \infty ∞ , one may write
ϕ ( z ) = C P D 6 ( z ) z 4 ( z − 1 ) 4 d z 2 , P D 6 ( z ) = ∏ j = 1 6 ( z − z j ) , (24) \phi(z)=C\frac{P_{D_6}(z)}{z^4(z-1)^4}\,\mathrm{d}z^2,\qquad P_{D_6}(z)=\prod_{j=1}^{6}(z-z_j), \tag*{(24)} ϕ ( z ) = C z 4 ( z − 1 ) 4 P D 6 ( z ) d z 2 , P D 6 ( z ) = j = 1 ∏ 6 ( z − z j ) , (24) and the orientation double has the affine hyperelliptic model
W 2 = C P D 6 ( z ) . (25) W^2=CP_{D_6}(z). \tag*{(25)} W 2 = C P D 6 ( z ) . (25) Let
B X = { a , b , c 1 , c ‾ 1 , c 2 , c ‾ 2 } B_X=\{a,b,c_1,\overline{c}_1,c_2,\overline{c}_2\} B X = { a , b , c 1 , c 1 , c 2 , c 2 } be the branch locus of the compact model X 2 → P 1 X_2\to\mathbb{P}^1 X 2 → P 1 in (19) . For any labelling Z ( ϕ ) = { z 1 , … , z 6 } Z(\phi)=\{z_1,\ldots,z_6\} Z ( ϕ ) = { z 1 , … , z 6 } and any orientation-preserving homeomorphism
h : P 1 ⟶ P 1 , h ( Z ( ϕ ) ) = B X , h:\mathbb{P}^1\longrightarrow\mathbb{P}^1,\qquad h(Z(\phi))=B_X, h : P 1 ⟶ P 1 , h ( Z ( ϕ )) = B X , the map h h h lifts to a homeomorphism
h ~ : C D 6 ⟶ X 2 (26) \widetilde{h}:C_{D_6}\longrightarrow X_2 \tag*{(26)} h : C D 6 ⟶ X 2 (26) unique up to the hyperelliptic deck involution. If the six points are paired into three cuts and h h h is chosen to carry those cuts to Γ 0 , Γ 1 , Γ 2 \Gamma_0,\Gamma_1,\Gamma_2 Γ 0 , Γ 1 , Γ 2 , the lift identifies the associated cut-cycle and connector symplectic bases.
Proof. For a quadratic differential on P 1 \mathbb{P}^{1} P 1 , deg div ( ϕ ) = deg K P 1 ⊗ 2 = − 4 \deg\operatorname{div}(\phi)=\deg K_{\mathbb{P}^{1}}^{\otimes2}=-4 deg div ( ϕ ) = deg K P 1 ⊗ 2 = − 4 . The three poles in (22) have total order 10, hence the total zero order is 10 − 4 = 6 10-4=6 10 − 4 = 6 . Under the stated genericity assumption these are six simple zeros. The orientation double branches at the odd-order points of div ( ϕ ) \operatorname{div}(\phi) div ( ϕ ) . All three pole orders are even, so the branch locus consists exactly of the six simple zeros. Riemann–Hurwitz gives
2 g ( C D 6 ) − 2 = 2 ( − 2 ) + 6 = 2 , 2g(C_{D_{6}})-2=2(-2)+6=2, 2 g ( C D 6 ) − 2 = 2 ( − 2 ) + 6 = 2 ,
therefore g ( C D 6 ) = 2 g(C_{D_{6}})=2 g ( C D 6 ) = 2 .
With the pole normalization in (24) , writing λ = W d z / [ z 2 ( z − 1 ) 2 ] \lambda=W\,\mathrm{d}z/[z^{2}(z-1)^{2}] λ = W d z / [ z 2 ( z − 1 ) 2 ] gives (25) directly.
By (19) , the compact Xi core is likewise a two-sheeted cover of the sphere branched exactly at B X B_{X} B X . On the complement of the branch locus, both covers are classified by the homomorphism to Z / 2 Z \mathbb{Z}/2\mathbb{Z} Z /2 Z that sends each small branch meridian to the nontrivial element. Consequently an orientation-preserving homeomorphism carrying one labelled branch set to the other preserves the covering monodromy and therefore lifts, uniquely up to deck transformation. Choosing compatible cut systems gives the final homological statement. □
Corollary 3.10 (A D 6 D_{6} D 6 -signature orientation differential on the Xi core). Let
P X ( T ) = ( T − a ) ( T − b ) ( T − c 1 ) ( T − c ‾ 1 ) ( T − c 2 ) ( T − c ‾ 2 ) P_{X}(T)=(T-a)(T-b)(T-c_{1})(T-\overline{c}_{1})(T-c_{2})(T-\overline{c}_{2}) P X ( T ) = ( T − a ) ( T − b ) ( T − c 1 ) ( T − c 1 ) ( T − c 2 ) ( T − c 2 ) be the certified Xi branch sextic, and choose distinct points p , q ∈ P 1 ∖ ( B X ∪ { ∞ } ) p,q\in\mathbb{P}^{1}\setminus(B_{X}\cup\{\infty\}) p , q ∈ P 1 ∖ ( B X ∪ { ∞ }) . Then, for every κ ∈ C × \kappa\in\mathbb{C}^{\times} κ ∈ C × ,
ϕ X D 6 ( T ) = κ P X ( T ) ( T − p ) 4 ( T − q ) 4 d T 2 (27) \phi_{X}^{D_{6}}(T)=\kappa\frac{P_{X}(T)}{(T-p)^{4}(T-q)^{4}}\,\mathrm{d}T^{2} \tag*{(27)} ϕ X D 6 ( T ) = κ ( T − p ) 4 ( T − q ) 4 P X ( T ) d T 2 (27) has six simple zeros at B X B_{X} B X and pole orders ( 4 , 4 , 2 ) (4,4,2) ( 4 , 4 , 2 ) at p , q , ∞ p,q,\infty p , q , ∞ . Hence its cusp/Katz data are ( 2 , 2 , 0 ) (2,2,0) ( 2 , 2 , 0 ) and ( 1 , 1 , 0 ) (1,1,0) ( 1 , 1 , 0 ) , respectively, i.e. the Chekhov–Mazzocco–Rubtsov D 6 D_{6} D 6 data ( 0 , 2 , 2 ) (0,2,2) ( 0 , 2 , 2 ) and ( 0 , 1 , 1 ) (0,1,1) ( 0 , 1 , 1 ) up to ordering. Its canonical orientation double is biholomorphic to X 2 X_{2} X 2 . In the sextic coordinate of (19) , the identification is explicit:
λ X = κ W X ( T − p ) 2 ( T − q ) 2 d T , λ X 2 = ϕ X D 6 . (28) \lambda_{X}=\sqrt{\kappa}\frac{W_{X}}{(T-p)^{2}(T-q)^{2}}\,\mathrm{d}T,\qquad\lambda_{X}^{2}=\phi_{X}^{D_{6}}. \tag*{(28)} λ X = κ ( T − p ) 2 ( T − q ) 2 W X d T , λ X 2 = ϕ X D 6 . (28) If p p p and q q q are chosen outside Ω ‾ 2 \overline{\Omega}_{2} Ω 2 , then on the spectral core the D 6 D_{6} D 6 orientation differential and the Xi action differential are related by the nowhere-vanishing holomorphic factor
λ X = G p , q ( T ) ϖ 0 , G p , q ( T ) = κ Y ( T ) 3 S ( T ) ( T − p ) 2 ( T − q ) 2 , (29) \lambda_{X}=G_{p,q}(T)\,\varpi_{0},\qquad G_{p,q}(T)=\sqrt{\kappa}\frac{Y(T)^{3}}{S(T)(T-p)^{2}(T-q)^{2}}, \tag*{(29)} λ X = G p , q ( T ) ϖ 0 , G p , q ( T ) = κ S ( T ) ( T − p ) 2 ( T − q ) 2 Y ( T ) 3 , (29) where S 2 = H S^{2}=H S 2 = H in the factorization K = H P X K=HP_{X} K = H P X used in the construction of (19) . This pole-signature construction alone is not a Lax realization; the certified van der Put–Saito phase-point realization is proved in theorem 3.12.
Proof. The finite poles in (27) have order four because p , q ∉ B X p,q\notin B_{X} p , q ∈ / B X . Since P X P_{X} P X has degree six, its coefficient behaves as T − 2 T^{-2} T − 2 at infinity; after the coordinate change T = 1 / w T=1/w T = 1/ w and d T 2 = w − 4 d w 2 \mathrm{d}T^{2}=w^{-4}\mathrm{d}w^{2} d T 2 = w − 4 d w 2 , this is a pole of order two. All poles have even order and all six zeros have odd order one, so the orientation double branches exactly at B X B_{X} B X . Substituting (28) gives λ X 2 = ϕ X D 6 \lambda_{X}^{2}=\phi_{X}^{D_{6}} λ X 2 = ϕ X D 6 and leaves the underlying sextic equation W X 2 = P X ( T ) W_{X}^{2}=P_{X}(T) W X 2 = P X ( T ) unchanged. This proves the biholomorphic identification with X 2 X_{2} X 2 . Finally, W X = y Y 3 / S W_{X}=yY^{3}/S W X = y Y 3 / S and ϖ 0 = y d T \varpi_{0}=y\,\mathrm{d}T ϖ 0 = y d T , so (29) follows directly. If p , q ∉ Ω ‾ 2 p,q\notin\overline{\Omega}_{2} p , q ∈ / Ω 2 , every factor in G p , q G_{p,q} G p , q is holomorphic and nonzero on Ω 2 \Omega_{2} Ω 2 . □
A Lax connection is a matrix differential system whose compatible deformation encodes a Painlevé equation. An isomonodromic deformation preserves its monodromy and Stokes data, the latter describing changes of asymptotic solution bases near irregular singularities.
Proposition 3.11 (Scalar normal form of the P I I I ( D 6 ) P_{\mathrm{III}}(D_{6}) P III ( D 6 ) Lax connection). Use the P I I I ( D 6 ) P_{\mathrm{III}}(D_{6}) P III ( D 6 ) rank-two connection of van der Put and Saito [11 §4.5] , written as
∇ d / d z = d d z + 1 z 2 A ( z ) , A ( z ) = ( A 11 ( z ) z − q L ( z ) z − q − A 11 ( z ) ) \nabla_{\mathrm{d}/\mathrm{d}z}=\frac{\mathrm{d}}{\mathrm{d}z}+\frac{1}{z^{2}}A(z),\qquad A(z)=\begin{pmatrix}\dfrac{A_{11}(z)}{z-q} & L(z)\\ z-q & -A_{11}(z)\end{pmatrix} ∇ d / d z = d z d + z 2 1 A ( z ) , A ( z ) = z − q A 11 ( z ) z − q L ( z ) − A 11 ( z ) where p , q p,q p , q are phase coordinates, t t t is the isomonodromic deformation parameter, and θ 0 , θ ∞ \theta_{0},\theta_{\infty} θ 0 , θ ∞ are formal monodromy exponent parameters at 0 0 0 and ∞ \infty ∞ , respectively. Here t t t is unrelated to heat time until a separate comparison is specified. With this convention,
A 11 ( z ) = p + 1 2 ( z − q ) ( q t + z t + θ ∞ ) A_{11}(z)=p+\frac{1}{2}(z-q)(qt+zt+\theta_{\infty}) A 11 ( z ) = p + 2 1 ( z − q ) ( q t + z t + θ ∞ ) and L ( z ) L(z) L ( z ) is the explicit affine function displayed in [11 §4.5, Eq. (12)] . For q t ≠ 0 qt\ne0 q t = 0 , eliminate the first component of a horizontal section and remove the first-derivative term. The resulting scalar normal form is
ψ ′ ′ ( z ) = V D 6 ( z ) ψ ( z ) , V D 6 ( z ) = N 6 ( z ) 4 q 2 z 4 ( z − q ) 2 (30) \psi^{\prime\prime}(z)=V_{D_{6}}(z)\psi(z),\qquad V_{D_{6}}(z)=\frac{N_{6}(z)}{4q^{2}z^{4}(z-q)^{2}} \tag*{(30)} ψ ′′ ( z ) = V D 6 ( z ) ψ ( z ) , V D 6 ( z ) = 4 q 2 z 4 ( z − q ) 2 N 6 ( z ) (30) where
N 6 ( z ) = q 2 t 2 z 6 + 2 q 2 t ( θ ∞ + 1 − q t ) z 5 + ( 4 p 2 − 6 q 3 t ( θ ∞ + 1 ) − q 2 − 2 q t θ 0 − t 2 ) z 4 + ( − 8 p 2 q − 4 p q 2 + 2 q 5 t 2 + 6 q 4 t ( θ ∞ + 1 ) + 4 q 3 + 6 q 2 t θ 0 + 2 q t 2 ) z 3 + ( 4 p 2 q 2 + 4 p q 3 − q 6 t 2 − 2 q 5 t ( θ ∞ + 1 ) − 6 q 3 t θ 0 ) z 2 + ( 2 q 4 t θ 0 − 2 q 3 t 2 ) z + q 4 t 2 . (31) \begin{aligned}
N_{6}(z)={}&q^{2}t^{2}z^{6}+2q^{2}t(\theta_{\infty}+1-qt)z^{5}\\
&+\left(4p^{2}-6q^{3}t(\theta_{\infty}+1)-q^{2}-2qt\theta_{0}-t^{2}\right)z^{4}\\
&+\left(-8p^{2}q-4pq^{2}+2q^{5}t^{2}+6q^{4}t(\theta_{\infty}+1)+4q^{3}+6q^{2}t\theta_{0}+2qt^{2}\right)z^{3}\\
&+\left(4p^{2}q^{2}+4pq^{3}-q^{6}t^{2}-2q^{5}t(\theta_{\infty}+1)-6q^{3}t\theta_{0}\right)z^{2}\\
&+\left(2q^{4}t\theta_{0}-2q^{3}t^{2}\right)z+q^{4}t^{2}. \tag*{(31)}
\end{aligned} N 6 ( z ) = q 2 t 2 z 6 + 2 q 2 t ( θ ∞ + 1 − q t ) z 5 + ( 4 p 2 − 6 q 3 t ( θ ∞ + 1 ) − q 2 − 2 q t θ 0 − t 2 ) z 4 + ( − 8 p 2 q − 4 p q 2 + 2 q 5 t 2 + 6 q 4 t ( θ ∞ + 1 ) + 4 q 3 + 6 q 2 t θ 0 + 2 q t 2 ) z 3 + ( 4 p 2 q 2 + 4 p q 3 − q 6 t 2 − 2 q 5 t ( θ ∞ + 1 ) − 6 q 3 t θ 0 ) z 2 + ( 2 q 4 t θ 0 − 2 q 3 t 2 ) z + q 4 t 2 . (31) In particular
N 6 ( q ) = 3 q 6 . (32) N_{6}(q)=3q^{6}. \tag*{(32)} N 6 ( q ) = 3 q 6 . (32) Hence, relative to the standard projective coordinate on P 1 \mathbb{P}^{1} P 1 , the scalar-normal-form potential V D 6 ( z ) d z 2 V_{D_{6}}(z)\,\mathrm{d}z^{2} V D 6 ( z ) d z 2 has pole orders ( 4 , 2 , 4 ) (4,2,4) ( 4 , 2 , 4 ) at 0 , q , ∞ 0,q,\infty 0 , q , ∞ , respectively. Its pole-order multiset therefore agrees with the ( 2 , 4 , 4 ) (2,4,4) ( 2 , 4 , 4 ) pattern associated with P I I I ( D 6 ) P_{\mathrm{III}}(D_{6}) P III ( D 6 ) in Appendix A, Table 2 of [10 ] . The pole at q q q is, however, the apparent singularity introduced by scalarization, so this numerical agreement is not by itself an identification with the invariant Hitchin quadratic differential. We call the canonical double of the scalar-normal-form potential the scalar-normal-form orientation curve. When the zeros are simple, denote its smooth projective completion by C D 6 s c a l C_{D_{6}}^{\mathrm{scal}} C D 6 scal . It has affine model
η 2 = N 6 ( z ) , (33) \eta^{2}=N_{6}(z), \tag*{(33)} η 2 = N 6 ( z ) , (33) and is a genus-two hyperelliptic curve.
Proof. For a horizontal section write the coefficient matrix as M = − A / z 2 = ( c − a a − c ) M=-A/z^{2}=\begin{pmatrix}c&-a\\ a&-c\end{pmatrix} M = − A / z 2 = ( c a − a − c ) , so c = − ( z − q ) / z 2 c=-(z-q)/z^{2} c = − ( z − q ) / z 2 . Eliminating the first component gives
y 2 ′ ′ − c ′ c y 2 ′ − ( c ′ c a + a 2 + b c − a ′ ) y 2 = 0. y_{2}^{\prime\prime}-\frac{c^{\prime}}{c}y_{2}^{\prime}-\left(\frac{c^{\prime}}{c}a+a^{2}+bc-a^{\prime}\right)y_{2}=0. y 2 ′′ − c c ′ y 2 ′ − ( c c ′ a + a 2 + b c − a ′ ) y 2 = 0. The substitution y 2 = c ψ y_{2}=\sqrt{c}\,\psi y 2 = c ψ removes the first derivative. Substitution of the explicit entries of [11 §4.5, Eq. (12)] and simplification gives (30) –(31) . The identity (32) is immediate from the sextic.
Also N 6 ( 0 ) = q 4 t 2 N_6(0)=q^4t^2 N 6 ( 0 ) = q 4 t 2 and the leading coefficient is q 2 t 2 q^2t^2 q 2 t 2 , so the pole orders of V D 6 d z 2 V_{D_6}\,dz^2 V D 6 d z 2 are 4 at 0, 2 at the apparent scalar singularity q q q , and 4 at infinity. The six simple zeros of N 6 N_6 N 6 are the branch points of its orientation double, so Riemann–Hurwitz gives genus two. □
Theorem 3.12 (Certified real P I I I ( D 6 ) P_{\mathrm{III}}(D_6) P III ( D 6 ) scalar-normal-form realization of the Xi core). Let P X ( T ) P_X(T) P X ( T ) be the exact Xi branch sextic of (19) . There exist real parameters with q = 1 q=1 q = 1 and a real affine coordinate
such that
N 6 ( s T + r ) t 2 s 6 = P X ( T ) . (34) \frac{N_6(sT+r)}{t^2s^6}=P_X(T). \tag*{(34)} t 2 s 6 N 6 ( s T + r ) = P X ( T ) . (34) More precisely, a solution lies in the directed box
2.3749166077 < p < 2.3749366078 , − 1.7097537396 < t < − 1.7097337395 , − 5.7723733561 < θ 0 < − 5.7723333560 , 0.7484294656 < θ ∞ < 0.7484494657 , − 0.008025896829 < s < − 0.008025876828 , − 0.6919860012 < r < − 0.6919840011. (35) \begin{aligned}
2.3749166077 &< p < 2.3749366078,\\
-1.7097537396 &< t < -1.7097337395,\\
-5.7723733561 &< \theta_0 < -5.7723333560,\\
0.7484294656 &< \theta_\infty< 0.7484494657,\\
-0.008025896829 &< s < -0.008025876828,\\
-0.6919860012 &< r < -0.6919840011. \tag*{(35)}
\end{aligned} 2.3749166077 − 1.7097537396 − 5.7723733561 0.7484294656 − 0.008025896829 − 0.6919860012 < p < 2.3749366078 , < t < − 1.7097337395 , < θ 0 < − 5.7723333560 , < θ ∞ < 0.7484494657 , < s < − 0.008025876828 , < r < − 0.6919840011. (35) Consequently the real genus-two curves are explicitly biholomorphic by
z = s T + r , η = t s 3 W X , (36) z=sT+r,\qquad\eta=ts^3W_X, \tag*{(36)} z = s T + r , η = t s 3 W X , (36) and this map commutes with complex conjugation. The certified Xi real type ( 2 , 1 , 1 ) (2,1,1) ( 2 , 1 , 1 ) therefore occurs as the scalar-normal-form orientation curve of an actual P I I I ( D 6 ) P_{\mathrm{III}}(D_6) P III ( D 6 ) Lax phase point.
Proof. Use the refined branch boxes of theorem 3.8 to form an outward interval box for the six nonleading coefficients of the monic sextic P X P_X P X . Fix q = 1 q=1 q = 1 and consider the six-dimensional real coefficient map
( p , t , θ 0 , θ ∞ , s , r ) ⟼ monic ( N 6 ( s T + r ) ) . (p,t,\theta_0,\theta_\infty,s,r)\longmapsto\operatorname{monic}(N_6(sT+r)). ( p , t , θ 0 , θ ∞ , s , r ) ⟼ monic ( N 6 ( s T + r )) .
At the numerical center
( p , t , θ 0 , θ ∞ , s , r ) ≈ ( 2.37492660773014 , − 1.70974373954814 , − 5.77235335600993 , 0.748439465636624 , − 0.00802588682829644 , − 0.691985001154100 ) , \begin{aligned}
(p,t,\theta_0,\theta_\infty,s,r) \approx(&2.37492660773014,-1.70974373954814,\\
&-5.77235335600993,0.748439465636624,\\
&-0.00802588682829644,-0.691985001154100),
\end{aligned} ( p , t , θ 0 , θ ∞ , s , r ) ≈ ( 2.37492660773014 , − 1.70974373954814 , − 5.77235335600993 , 0.748439465636624 , − 0.00802588682829644 , − 0.691985001154100 ) ,
a parametric Krawczyk operator, with the Xi coefficient box retained as an interval parameter, satisfies
K ( X ) ⊂ int X . K(X)\subset\operatorname{int}X. K ( X ) ⊂ int X .
for the box X X X in (35) . Hence for every coefficient vector in the certified Xi box there is a unique real parameter vector in X X X solving the six coefficient equations. In particular this holds for the exact Xi sextic. This proves (34) . Since the leading coefficient of N 6 N_6 N 6 is t 2 t^2 t 2 when q = 1 q=1 q = 1 , the fibre rescaling in (36) follows immediately. All coefficients and the affine map are real, so conjugation is preserved. □
Corollary 3.13 (The local isomonodromic trajectory in the chosen normalization). Every point in the box (35) has q t ≠ 0 qt\ne0 q t = 0 and therefore lies in the affine phase chart of the van der Put–Saito
P I I I ( D 6 ) P_{\mathrm{III}}(D_6) P III ( D 6 ) Hamiltonian system. The certified realization of theorem 3.12 therefore lies on a unique local isomonodromic trajectory satisfying
d q d t = 4 p + q t , d p d t = 4 p 2 q t + p t + t q 3 + q 2 − t q − θ 0 + q 2 θ ∞ . (37) \frac{\mathrm{d}q}{\mathrm{d}t}=\frac{4p+q}{t},\qquad\frac{\mathrm{d}p}{\mathrm{d}t}=\frac{4p^2}{qt}+\frac{p}{t}+tq^3+q^2-\frac{t}{q}-\theta_0+q^2\theta_\infty. \tag*{(37)} d t d q = t 4 p + q , d t d p = q t 4 p 2 + t p + t q 3 + q 2 − q t − θ 0 + q 2 θ ∞ . (37)
At the midpoint realization one has, for orientation only,
d q d t ≈ − 6.14110 , d p d t ≈ − 7.06387. \frac{\mathrm{d}q}{\mathrm{d}t}\approx-6.14110,\qquad\frac{\mathrm{d}p}{\mathrm{d}t}\approx-7.06387. d t d q ≈ − 6.14110 , d t d p ≈ − 7.06387.
Thus, within the chosen q = 1 q=1 q = 1 affine normalization, the certified X i X_i X i match determines a local D 6 D_6 D 6 isomonodromic direction through the corresponding Lax phase point.
There is also a canonical conceptual route from decorated/Stokes data to this genus-two curve. On the generic locus it has the form
X D 6 d e c ⇢ M B w i l d ⟷ R H M d R ⟷ w i l d N A H M D o l → det Φ ϕ ⟼ C D 6 . (38) \mathcal{X}_{D_6}^{\mathrm{dec}}\dashrightarrow\mathcal{M}_{B}^{\mathrm{wild}}\mathrel{\mathop{\longleftrightarrow}^{\mathrm{RH}}}\mathcal{M}_{\mathrm{dR}}\mathrel{\mathop{\longleftrightarrow}^{\mathrm{wild\ NAH}}}\mathcal{M}_{\mathrm{Dol}}\xrightarrow{\det\Phi}\phi\longmapsto C_{D_6}. \tag*{(38)} X D 6 dec ⇢ M B wild ⟷ RH M dR ⟷ wild NAH M Dol d e t Φ ϕ ⟼ C D 6 . (38)
Here X D 6 d e c \mathcal{X}_{D_6}^{\mathrm{dec}} X D 6 dec is the decorated character space, whose extra choices of local flags or framings provide coordinates on monodromy data. The wild Betti moduli space M B w i l d \mathcal{M}_{B}^{\mathrm{wild}} M B wild records monodromy, formal monodromy and Stokes matrices; “wild” refers to irregular singularities. The dashed arrow forgets the decorations where that map is defined. The space M d R \mathcal{M}_{\mathrm{dR}} M dR parametrizes meromorphic flat connections, while M D o l \mathcal{M}_{\mathrm{Dol}} M Dol parametrizes the corresponding stable meromorphic Higgs bundles. A Higgs field Φ \Phi Φ is an endomorphism-valued meromorphic one-form; for the trace-free rank-two system, det Φ \det\Phi det Φ is the associated quadratic differential ϕ \phi ϕ in the convention of the diagram. The curve C D 6 C_{D_6} C D 6 is its orientation double, on which a square root of ϕ \phi ϕ becomes single-valued.
The label RH in (38) means the irregular Riemann–Hilbert correspondence , not the Riemann hypothesis. The label wild NAH means wild nonabelian Hodge correspondence [12 ] . The spaces are understood with compatible singularity, stability and parabolic data fixed; the generic locus excludes unstable or degenerate points and has simple branch zeros. With these choices, the diagram describes correspondences at the moduli level, not an elementary rational formula in the decorated cluster/λ \lambda λ -length coordinates (edge coordinates on the decorated surface).
There are now two levels of comparison. If one starts from an independently specified D 6 D_6 D 6 Higgs/isomonodromic point, then a holomorphic identification C D 6 ≃ X 2 C_{D_6}\simeq X_2 C D 6 ≃ X 2 compatible with the hyperelliptic projections exists precisely when the two labelled branch sextuples agree modulo PGL 2 ( C ) \operatorname{PGL}_2(\mathbb{C}) PGL 2 ( C ) , equivalently when their genus-two moduli (for example Rosenhain data or absolute Igusa invariants) agree after relabelling. Here PGL 2 ( C ) \operatorname{PGL}_2(\mathbb{C}) PGL 2 ( C ) is the group of Möbius transformations; Rosenhain data are the remaining three branch coordinates after three labelled branch points have been sent to 0 , 1 , ∞ 0,1,\infty 0 , 1 , ∞ . Absolute Igusa invariants are scale-invariant functions of the sextic coefficients that describe the generic genus-two moduli point. Indeed, if
M ( z ) = α z + β γ z + δ M(z) = \frac{\alpha z + \beta}{\gamma z + \delta} M ( z ) = γ z + δ α z + β
sends the six D 6 D_6 D 6 branch values to B X B_X B X , then for a nonzero constant μ \mu μ
P X ( M ( z ) ) = μ P D 6 ( z ) ( γ z + δ ) 6 , P_X(M(z)) = \mu\frac{P_{D_6}(z)}{(\gamma z + \delta)^6}, P X ( M ( z )) = μ ( γ z + δ ) 6 P D 6 ( z ) ,
and the lift is explicitly
T = M ( z ) , W X = μ W ( γ z + δ ) 3 . (39) T = M(z), \qquad W_X = \sqrt{\mu}\frac{W}{(\gamma z + \delta)^3}. \tag*{(39)} T = M ( z ) , W X = μ ( γ z + δ ) 3 W . (39)
On the other hand, Corollary 3.10 shows that no moduli matching is needed merely to realize the Xi curve in the same quadratic-differential stratum: one can construct the D 6 D_6 D 6 -signature differential directly from P X P_X P X . Theorem 3.12 solves the corresponding scalar-projective isomonodromic realization problem: the Xi sextic occurs at a certified real phase point of a standard P I I I ( D 6 ) \mathrm{P}_{\mathrm{III}}(D_6) P III ( D 6 ) Lax connection after cyclic scalarization. What remains is sharper and genuinely nonabelian: identify this projective quadratic differential with a preferred Dolbeault/Hitchin determinant under wild nonabelian Hodge theory, and compute the decorated Stokes/cluster coordinates of the certified phase point.
A real-equivariant realization must reproduce the chamber with two real branch values and two nonreal conjugate pairs, hence the real type ( 2 , 1 , 1 ) (2,1,1) ( 2 , 1 , 1 ) . Finally, even an identification of the curves does not by itself identify the action periods. Formula (29) isolates the discrepancy exactly: the Xi action differential is a weighted D 6 D_6 D 6 orientation differential. Thus the remaining tests are now polar/isomonodromic realization, real chamber, and differential normalization, rather than genus or branch-cover topology.
One caution remains important. The quadratic differential in theorem 3.11 is the projective scalar-normal-form potential of the Lax connection; the apparent pole at z = q z=q z = q is created by cyclic scalarization. It should therefore not be identified silently with every matrix-gauge representative of det A ( z ) d z 2 \det A(z)\,\mathrm{d}z^2 det A ( z ) d z 2 . The agreement with the Chekhov–Mazzocco–Rubtsov ( 0 , 2 , 2 ) (0,2,2) ( 0 , 2 , 2 ) pole signature is exact, while the direct Dolbeault/Hitchin interpretation of this particular projective representative is a further comparison problem.
Deformation obstruction and the canonical Kummer surface
The fixed-point realization of theorem 3.12 raises a sharper question: does the Xi family itself follow the D 6 D_6 D 6 isomonodromic flow? The answer is already negative at first order for every regular identification of the two deformation parameters.
For four ordered branch values put
λ ( z 4 ; z 1 , z 2 , z 3 ) : = ( z 4 − z 1 ) ( z 2 − z 3 ) ( z 4 − z 3 ) ( z 2 − z 1 ) . (40) \lambda(z_4;z_1,z_2,z_3) := \frac{(z_4-z_1)(z_2-z_3)}{(z_4-z_3)(z_2-z_1)}. \tag*{(40)} λ ( z 4 ; z 1 , z 2 , z 3 ) := ( z 4 − z 3 ) ( z 2 − z 1 ) ( z 4 − z 1 ) ( z 2 − z 3 ) . (40)
This is invariant under an arbitrary, parameter-dependent element of P G L 2 ( C ) \mathrm{PGL}_2(\mathbb{C}) PGL 2 ( C ) .
Theorem 3.15 (First-order isomonodromic deformation obstruction). Let
λ Ξ ( x ) = λ ( c ˉ 1 ( x ) ; a ( x ) , b ( x ) , c 1 ( x ) ) \lambda_{\Xi}(x)=\lambda(\bar{c}_1(x);a(x),b(x),c_1(x)) λ Ξ ( x ) = λ ( c ˉ 1 ( x ) ; a ( x ) , b ( x ) , c 1 ( x )) for the locally transported Xi branch values, and let λ D 6 ( t ) \lambda_{D_6}(t) λ D 6 ( t ) be the corresponding cross ratio of the four matched branch values of the certified P I I I ( D 6 ) \mathrm{P}_{\mathrm{III}}(D_6) P III ( D 6 ) scalar spectral curve. Then
λ Ξ ′ ( 0 ) = 0 , (41) \lambda_{\Xi}'(0)=0, \tag*{(41)} λ Ξ ′ ( 0 ) = 0 , (41)
whereas along the D 6 D_{6} D 6 isomonodromic vector field at the certified phase box,
∣ d λ D 6 d t ∣ > 0.4. (42) \left|\frac{\mathrm{d}\lambda_{D_{6}}}{\mathrm{d}t}\right| > 0.4. \tag*{(42)} d t d λ D 6 > 0.4. (42)
Consequently there is no local identity of Ξ \Xi Ξ and D 6 D_{6} D 6 spectral curves, even modulo a parameter-dependent projective change of spectral coordinate, with a time reparametrization t = t ( x ) t=t(x) t = t ( x ) satisfying t ′ ( 0 ) ≠ 0 t'(0)\ne0 t ′ ( 0 ) = 0 . The statement is status [A].
Proof. The normalized Xi discriminant Δ x ( T ) \Delta_x(T) Δ x ( T ) is even in x x x . Every simple branch value therefore has a locally unique even continuation by the implicit function theorem; hence every projective cross ratio of the six branch values is even, proving (41) .
For the D 6 D_{6} D 6 side, differentiate the sextic N 6 ( z ) N_{6}(z) N 6 ( z ) along the Hamiltonian flow (37) . If z j ( t ) z_j(t) z j ( t ) is a simple zero, then
z ˙ j = − ∂ t N 6 + p ˙ ∂ p N 6 + q ˙ ∂ q N 6 ∂ z N 6 ( z j ) . \dot{z}_j=-\frac{\partial_tN_{6}+\dot{p}\,\partial_pN_{6}+\dot{q}\,\partial_qN_{6}}{\partial_zN_{6}}(z_j). z ˙ j = − ∂ z N 6 ∂ t N 6 + p ˙ ∂ p N 6 + q ˙ ∂ q N 6 ( z j ) .
Substitution into the logarithmic derivative of (40) and outward interval evaluation over the certified parameter and root boxes gives
d λ D 6 d t ∈ [ − 0.729124 , 0.729128 ] + i [ − 1.821216 , − 0.435579 ] , \frac{\mathrm{d}\lambda_{D_{6}}}{\mathrm{d}t}\in[-0.729124,0.729128]+\mathrm{i}[-1.821216,-0.435579], d t d λ D 6 ∈ [ − 0.729124 , 0.729128 ] + i [ − 1.821216 , − 0.435579 ] ,
so its modulus is > 0.435579 >0.435579 > 0.435579 . This proves (42) . Because cross ratios are projective invariants, any local curve identity with t ′ ( 0 ) ≠ 0 t'(0)\ne0 t ′ ( 0 ) = 0 would give 0 = λ Ξ ′ ( 0 ) = λ D 6 , t t ′ ( 0 ) 0=\lambda_{\Xi}'(0)=\lambda_{D_{6},t}t'(0) 0 = λ Ξ ′ ( 0 ) = λ D 6 , t t ′ ( 0 ) , a contradiction. □
Remark 3.16 (The even reparametrization remains a second-order question). The theorem does not exclude a singular/even identification t − t 0 = β x 2 + O ( x 4 ) t-t_0=\beta x^2+O(x^4) t − t 0 = β x 2 + O ( x 4 ) . Writing τ = x 2 \tau=x^2 τ = x 2 , the Xi branch velocity is
d T j d τ ∣ 0 = − Δ [ 2 ] ( T j ) 2 ∂ T Δ 0 ( T j ) . \left.\frac{\mathrm{d}T_j}{\mathrm{d}\tau}\right|_0=-\frac{\Delta_{[2]}(T_j)}{2\,\partial_T\Delta_0(T_j)}. d τ d T j 0 = − 2 ∂ T Δ 0 ( T j ) Δ [ 2 ] ( T j ) .
A midpoint theta-moment computation gives three independent cross-ratio velocity comparisons. The first would require β ≈ − 0.06287647 \beta\approx-0.06287647 β ≈ − 0.06287647 , while the other two give incompatible complex ratios. Thus the quadratic D 6 D_{6} D 6 tangency also fails numerically, but this second statement is status [N] until the x 2 x^2 x 2 branch velocities are placed in directed boxes.
The genus-two curve itself carries a canonical surface of complex dimension two. Put
J X : = Jac ( X 2 ) , Kum s i n g ( X 2 ) : = J X / { ± 1 } , J_X:=\operatorname{Jac}(X_2),\qquad\operatorname{Kum}^{\mathrm{sing}}(X_2):=J_X/\{\pm1\}, J X := Jac ( X 2 ) , Kum sing ( X 2 ) := J X / { ± 1 } ,
and denote by Kum ( X 2 ) \operatorname{Kum}(X_2) Kum ( X 2 ) its minimal resolution.
Proposition 3.17 (Kummer lift of the Xi–D 6 D_{6} D 6 realization). The singular Kummer surface Kum s i n g ( X 2 ) \operatorname{Kum}^{\mathrm{sing}}(X_2) Kum sing ( X 2 ) has exactly 16 ordinary double points, indexed by
J X [ 2 ] ≃ ( Z / 2 Z ) 4 , J_X[2]\simeq(\mathbb{Z}/2\mathbb{Z})^4, J X [ 2 ] ≃ ( Z /2 Z ) 4 , and Kum ( X 2 ) \operatorname{Kum}(X_2) Kum ( X 2 ) is a K3 surface. The biholomorphism (36) identifies X 2 X_2 X 2 with the specific certified scalar-normal-form curve C D 6 s c a l C_{D_{6}}^{\mathrm{scal}} C D 6 scal and induces an isomorphism of principally polarized abelian surfaces and therefore canonical isomorphisms
J X ≃ J ( C D 6 s c a l ) , Kum ( X 2 ) ≃ Kum ( C D 6 s c a l ) . (43) J_X\simeq J(C_{D_{6}}^{\mathrm{scal}}),\qquad\operatorname{Kum}(X_2)\simeq\operatorname{Kum}(C_{D_{6}}^{\mathrm{scal}}). \tag*{(43)} J X ≃ J ( C D 6 scal ) , Kum ( X 2 ) ≃ Kum ( C D 6 scal ) . (43) All statements are status [A].
Proof. For a genus-two curve the Jacobian is a principally polarized abelian surface, and its two-torsion group has order 2 2 g = 16 2^{2g}=16 2 2 g = 16 . The fixed points of the involution u ↦ − u u\mapsto-u u ↦ − u are exactly these two-torsion points, which become the 16 nodes of the quotient. Resolving them gives the classical Kummer K3 surface. A curve biholomorphism induces the principally polarized Jacobian isomorphism by functoriality of the Abel–Jacobi construction; it commutes with [ − 1 ] [-1] [ − 1 ] , hence descends to the quotient and lifts to the minimal resolutions. □
The real chamber found in theorem 3.7 leaves a particularly small real skeleton inside the sixteen-node configuration.
Proposition 3.18 (Real Kummer nodes). Let P a P_a P a , P b P_b P b , P c 1 P_{c_1} P c 1 , P c ‾ 1 P_{\overline{c}_1} P c 1 , P c 2 P_{c_2} P c 2 , P c ‾ 2 P_{\overline{c}_2} P c 2 be the six Weierstrass points above the certified branch values . The conjugation-fixed subgroup of two-torsion is
J X [ 2 ] τ = { 0 , e 0 , e 1 , e 2 } ≃ ( Z / 2 Z ) 2 (44) J_X[2]^\tau=\{0,e_0,e_1,e_2\}\simeq(\mathbb{Z}/2\mathbb{Z})^2 \tag*{(44)} J X [ 2 ] τ = { 0 , e 0 , e 1 , e 2 } ≃ ( Z /2 Z ) 2 (44) where
e 0 = [ P a − P b ] , e 1 = [ P c 1 − P c ‾ 1 ] , e 2 = [ P c 2 − P c ‾ 2 ] , e 0 + e 1 + e 2 = 0. (45) e_0=[P_a-P_b],\qquad e_1=[P_{c_1}-P_{\overline{c}_1}],\qquad e_2=[P_{c_2}-P_{\overline{c}_2}],\qquad e_0+e_1+e_2=0. \tag*{(45)} e 0 = [ P a − P b ] , e 1 = [ P c 1 − P c 1 ] , e 2 = [ P c 2 − P c 2 ] , e 0 + e 1 + e 2 = 0. (45) Thus the induced real structure on the singular Kummer surface has exactly four real nodes: the origin and three nonzero nodes canonically associated with the three spectral cuts Γ 0 , Γ 1 , Γ 2 \Gamma_0,\Gamma_1,\Gamma_2 Γ 0 , Γ 1 , Γ 2 . This statement is status [A].
Proof. For a genus-two hyperelliptic curve, J [ 2 ] J[2] J [ 2 ] is represented by even subsets of the six Weierstrass points modulo replacement by the complementary subset. Every nonzero class has a unique two-element representative. Conjugation fixes a , b a,b a , b individually and interchanges the two pairs c j ↔ c ‾ j c_j\leftrightarrow\overline{c}_j c j ↔ c j . Hence the only invariant two-element subsets are
{ a , b } , { c 1 , c ‾ 1 } , { c 2 , c ‾ 2 } . \{a,b\},\qquad\{c_1,\overline{c}_1\},\qquad\{c_2,\overline{c}_2\}. { a , b } , { c 1 , c 1 } , { c 2 , c 2 } .
Together with the zero class these give (44) . Addition is symmetric difference modulo complement, so the symmetric difference of the first two pairs is the complement of the third; therefore e 0 + e 1 = e 2 e_0+e_1=e_2 e 0 + e 1 = e 2 . The other relations follow cyclically. □
Remark 3.19 (Relation with earlier Kummer constructions and the (0, 2, 2) channels). The genus-two-to-Kummer mechanism is the same classical construction used in Planat–Chester–Aschheim–Amaral–Fang–Irwin [13 ] : six Weierstrass branch points determine a Jacobian, its sixteen two-torsion points become the nodes of the Kummer quotient, and the minimal resolution is a K3 surface. The present Xi chamber is nevertheless a different real form from the explicit fully real sextic examples considered there: here the branch configuration is two real points plus two conjugate pairs and has real type (2, 1, 1).
The three nonzero real Kummer nodes in (45) give a precise finite skeleton attached to the three cuts. It is tempting to view the split real channel e 0 e_0 e 0 and the two conjugate channels e 1 , e 2 e_1,e_2 e 1 , e 2 as a Kummer shadow of the Chekhov–Mazzocco–Rubtsov (0, 2, 2) organization. This last correspondence is an interpretation, not yet a theorem about decorated Stokes coordinates; proving it requires computing those coordinates at the certified D 6 D_6 D 6 phase point.
Remark 3.20 (Kummer deformation inherits the first-order obstruction). The principally polarized Torelli map sends the genus-two curve to its Jacobian and hence to its polarized Kummer surface. Therefore the nonzero D 6 D_6 D 6 cross-ratio tangent in theorem 3.15 also gives a nontrivial deformation of the associated polarized Kummer surface, whereas the Xi x x x -deformation has zero first jet at the fixed point. Thus the same regular-time obstruction persists after passing from X 2 X_2 X 2 to its Kummer surface. A directed second-order comparison in τ = x 2 \tau=x^2 τ = x 2 is the natural next test.
Low-discriminant Humbert exclusions
The Kummer construction is universal for a genus-two Jacobian, whereas a Humbert condition records an additional algebraic class of the principally polarized abelian surface. We now locate the Xi point relative to a collection of low-discriminant Humbert loci. Write I 2 , I 4 , I 6 , I 10 I_2,I_4,I_6,I_{10} I 2 , I 4 , I 6 , I 10 for the Igusa–Clebsch polynomial invariants of the branch sextic, of weights 2, 4, 6, 10; I 10 I_{10} I 10 is a discriminant invariant. On the chart I 2 ≠ 0 I_2 \ne0 I 2 = 0 , we use the absolute Igusa invariants
i 1 = 144 I 4 I 2 2 , i 2 = − 1728 ( I 2 I 4 − 3 I 6 ) I 2 3 , i 3 = 486 I 10 I 2 5 . (46) i_1 = \frac{144I_4}{I_2^2}, \qquad i_2 = -\frac{1728(I_2I_4 - 3I_6)}{I_2^3}, \qquad i_3 = \frac{486I_{10}}{I_2^5}. \tag*{(46)} i 1 = I 2 2 144 I 4 , i 2 = − I 2 3 1728 ( I 2 I 4 − 3 I 6 ) , i 3 = I 2 5 486 I 10 . (46)
The branch-disk certificate gives the following deliberately inflated boxes:
∣ i 1 − 23.41189486771366 ∣ < 10 − 6 , ∣ i 2 + 76.70307117848387 ∣ < 10 − 5 , ∣ i 3 + 0.0008995087060813192 ∣ < 10 − 10 . (47) \begin{aligned}
|i_1 - 23.41189486771366| &< 10^{-6},\\
|i_2 + 76.70307117848387| &< 10^{-5},\\
|i_3 + 0.0008995087060813192| &< 10^{-10}. \tag*{(47)}
\end{aligned} ∣ i 1 − 23.41189486771366∣ ∣ i 2 + 76.70307117848387∣ ∣ i 3 + 0.0008995087060813192∣ < 1 0 − 6 , < 1 0 − 5 , < 1 0 − 10 . (47)
For the higher-discriminant Satake tests we also propagate the branch disks without this deliberate inflation. An independent exact-rational audit gives
∣ i 1 − 23.41189486771366 ∣ < 2.75 × 10 − 8 , ∣ i 2 + 76.70307117848387 ∣ < 5.90 × 10 − 7 , ∣ i 3 + 0.0008995087060813192 ∣ < 2.82 × 10 − 12 . (48) \begin{aligned}
|i_1 - 23.41189486771366| &< 2.75 \times10^{-8},\\
|i_2 + 76.70307117848387| &< 5.90 \times10^{-7},\\
|i_3 + 0.0008995087060813192| &< 2.82 \times10^{-12}. \tag*{(48)}
\end{aligned} ∣ i 1 − 23.41189486771366∣ ∣ i 2 + 76.70307117848387∣ ∣ i 3 + 0.0008995087060813192∣ < 2.75 × 1 0 − 8 , < 5.90 × 1 0 − 7 , < 2.82 × 1 0 − 12 . (48)
The required outward radii before rounding are 2.74224155 × 10 − 8 2.74224155 \times10^{-8} 2.74224155 × 1 0 − 8 , 5.89208525 × 10 − 7 5.89208525 \times10^{-7} 5.89208525 × 1 0 − 7 , and 2.81746158 × 10 − 12 2.81746158 \times10^{-12} 2.81746158 × 1 0 − 12 , respectively.
Theorem 3.21 (Nine certified Humbert exclusions). Let J X = Jac ( X 2 ) J_X = \operatorname{Jac}(X_2) J X = Jac ( X 2 ) with its canonical principal polarization. Then
J X ∉ H 4 ∪ H 5 ∪ H 8 ∪ H 9 ∪ H 12 ∪ H 13 ∪ H 16 ∪ H 17 ∪ H 20 . (49) J_X \notin H_4 \cup H_5 \cup H_8 \cup H_9 \cup H_{12} \cup H_{13} \cup H_{16} \cup H_{17} \cup H_{20}. \tag*{(49)} J X ∈ / H 4 ∪ H 5 ∪ H 8 ∪ H 9 ∪ H 12 ∪ H 13 ∪ H 16 ∪ H 17 ∪ H 20 . (49) In particular, J X J_X J X admits no optimal ( 2 , 2 ) (2,2) ( 2 , 2 ) -, ( 3 , 3 ) (3,3) ( 3 , 3 ) -, or ( 4 , 4 ) (4,4) ( 4 , 4 ) -split elliptic quotient of the square-discriminant types 4, 9, 16. All nine exclusions are status [A].
Proof. We separate the loci because the most economical certificates are different.
For H 4 H_4 H 4 , a degree-two elliptic quotient of a genus-two hyperelliptic curve is equivalent to a nonhyperelliptic involution; equivalently, this is the square-discriminant-4 elliptic-subcover locus in the Humbert description [14 ] . Such an involution descends to a Möbius involution preserving the six branch values. Its permutation on the branch set has cycle type 2 3 2^3 2 3 or 1 2 2 2 1^2 2^2 1 2 2 2 . Directed complex-disk arithmetic excludes all 15 pairings of the first type and all 45 choices of the second type. Thus X 2 X_2 X 2 has no nonhyperelliptic involution and J X ∉ H 4 J_X \notin H_4 J X ∈ / H 4 .
For H 5 H_5 H 5 we use the standard absolute-invariant equation of the Humbert surface, equivalently the discriminant-5 Hilbert-modular parametrization of Elkies–Kumar [15 ] . Substitution of (47) , with exact rational polydisc bounds, gives
∣ F 5 ( i 1 , i 2 , i 3 ) ∣ > 3.8210230 × 10 8 , |F_5(i_1,i_2,i_3)| > 3.8210230 \times10^8, ∣ F 5 ( i 1 , i 2 , i 3 ) ∣ > 3.8210230 × 1 0 8 ,
so J X ∉ H 5 J_X \notin H_5 J X ∈ / H 5 .
For H 8 H_8 H 8 and H 9 H_9 H 9 we use the explicit two-parameter Igusa–Clebsch parametrizations of Elkies–Kumar and Kumar [15 , 14 ] . Let f 1 = f 2 = f 3 = 0 f_1=f_2=f_3=0 f 1 = f 2 = f 3 = 0 be the three polynomial equations obtained by equating the parametrized absolute invariants to ( i 1 , i 2 , i 3 ) (i_1,i_2,i_3) ( i 1 , i 2 , i 3 ) , and define
R 12 ( s ) = Res r ( f 1 , f 2 ) , R 13 ( s ) = Res r ( f 1 , f 3 ) . R_{12}(s) = \operatorname{Res}_r(f_1,f_2), \qquad R_{13}(s) = \operatorname{Res}_r(f_1,f_3). R 12 ( s ) = Res r ( f 1 , f 2 ) , R 13 ( s ) = Res r ( f 1 , f 3 ) .
For H 8 H_8 H 8 , exact-rational Rouché bounds isolate all six zeros of R 12 R_{12} R 12 in disjoint disks; the largest Rouché ratio is 0.151379, and on every such disk ∣ R 13 ∣ > 8.50614 × 10 5 |R_{13}| > 8.50614 \times10^5 ∣ R 13 ∣ > 8.50614 × 1 0 5 . For H 9 H_9 H 9 , all eight zeros are isolated with largest Rouché ratio 0.054486, and ∣ R 13 ∣ > 1.85052 |R_{13}| > 1.85052 ∣ R 13 ∣ > 1.85052 on every disk. Therefore neither surface meets the full Xi invariant box.
The same elimination strategy applied to the discriminant-12 model of Elkies–Kumar gives nondegenerate resultants of degrees 12 and 20 in the remaining parameter. Twelve disjoint disks of radius 10 − 4 10^{-4} 1 0 − 4 exhaust the first resultant over (47) ; the largest Rouché ratio is 0.003025, and the second resultant has the uniform lower bound
∣ R 13 ∣ > 39.8557. |R_{13}| > 39.8557. ∣ R 13 ∣ > 39.8557.
Thus J X ∉ H 12 J_X \notin H_{12} J X ∈ / H 12 .
Satake coordinates here are six auxiliary modular coordinates, permuted by relabelling and represented as the roots of the Satake sextic; they are not the six branch coordinates of X 2 X_2 X 2 . Their symmetric functions are determined by the Igusa data in the cited parametrizations. For H 13 H_{13} H 13 we use Gruenewald’s level-2 Satake description [16 ] . Eliminating the free component parameter gives a weighted-homogeneous necessary equation G 13 = 0 G_{13}=0 G 13 = 0 of weight 20. The six roots of the Xi Satake sextic persist in six disjoint disks of radius 10 − 4 10^{-4} 1 0 − 4 over the entire inflated box (47) ; the largest Rouché ratio is 0.005636. Testing each possible distinguished Satake root gives
∣ G 13 ∣ > 355.45 , |G_{13}| > 355.45, ∣ G 13 ∣ > 355.45 ,
so J X ∉ H 13 J_X \notin H_{13} J X ∈ / H 13 .
The H 16 H_{16} H 16 exclusion is more delicate. We use Kumar’s square-discriminant parametrization, equivalently the ( 4 , 4 ) (4,4) ( 4 , 4 ) -split locus studied by Bruin–Doerksen [14 , 17 ] . For this local elimination we use the certified enlargement
∣ i 1 − 23.41189486771366 ∣ < 3 × 10 − 8 , ∣ i 2 + 76.70307117848387 ∣ < 6 × 10 − 7 , ∣ i 3 + 0.0008995087060813192 ∣ < 3 × 10 − 12 . (50) \begin{aligned}
|i_1-23.41189486771366| &< 3 \times10^{-8},\\
|i_2+76.70307117848387| &< 6 \times10^{-7},\\
|i_3+0.0008995087060813192| &< 3 \times10^{-12}. \tag*{(50)}
\end{aligned} ∣ i 1 − 23.41189486771366∣ ∣ i 2 + 76.70307117848387∣ ∣ i 3 + 0.0008995087060813192∣ < 3 × 1 0 − 8 , < 6 × 1 0 − 7 , < 3 × 1 0 − 12 . (50)
It strictly contains the independently certified tight Xi box (48) . With X = i 1 X=i_1 X = i 1 and Y = i 2 Y=i_2 Y = i 2 , exact elimination gives
Res r ( f 1 , f 2 ) = 2 40 s 162 R 16 ( s ; X , Y ) , deg s R 16 = 18. (51) \operatorname{Res}_r(f_1,f_2)=2^{40}s^{162}R_{16}(s;X,Y),\qquad\deg_s R_{16}=18. \tag*{(51)} Res r ( f 1 , f 2 ) = 2 40 s 162 R 16 ( s ; X , Y ) , deg s R 16 = 18. (51)
The coefficient degrees satisfy deg X R 16 ≤ 12 \deg_X R_{16}\le12 deg X R 16 ≤ 12 and deg Y R 16 ≤ 8 \deg_Y R_{16}\le8 deg Y R 16 ≤ 8 . Hence a 13 × 9 13\times9 13 × 9 tensor interpolation determines R 16 R_{16} R 16 exactly; independent exact off-grid evaluations agree with the reconstructed polynomial. Over the entire box (50) , Rouché estimates place the eighteen zeros of R 16 R_{16} R 16 in eighteen disjoint disks of common radius 10 − 3 10^{-3} 1 0 − 3 , with largest Rouché ratio 0.033482. Since R 16 R_{16} R 16 has degree 18, this list is exhaustive. The factor s 162 s^{162} s 162 corresponds to the degenerate chart contribution; at s = 0 s=0 s = 0 the common root r = 0 r=0 r = 0 has I 2 I_2 I 2 -denominator data A = D = 0 A=D=0 A = D = 0 and does not define absolute invariants, while the nondegenerate residual factor does not vanish there.
Inside each of the eighteen exhaustive s s s -disks, an exact-rational complex-ball Krawczyk map certifies a unique transverse ( r , s ) (r,s) ( r , s ) branch of f 1 = f 2 = 0 f_1=f_2=0 f 1 = f 2 = 0 ; the local s s s -box is checked to lie strictly inside the matching Rouché disk. Direct ball evaluation of the third invariant equation gives
∣ i 3 ( 16 ) − ı ˉ 3 ∣ > 7.93 × 10 − 4 (52) \left|i_3^{(16)}-\bar{\imath}_3\right|>7.93\times10^{-4} \tag*{(52)} i 3 ( 16 ) − ˉ 3 > 7.93 × 1 0 − 4 (52)
on every certified branch. Thus J X ∉ H 16 J_X\notin H_{16} J X ∈ / H 16 .
It remains to record the two new height-selected tests. Gruenewald’s level-2 equation for H 17 H_{17} H 17 distinguishes a 3-subset of the six Satake coordinates. Over the tight invariant box (48) , all six roots of the Satake sextic persist as disjoint analytic branches in disks of radius 10 − 8 10^{-8} 1 0 − 8 ; the largest exact-rational Rouché ratio is 0.552755. All ( 6 3 ) = 20 \binom{6}{3}=20 ( 3 6 ) = 20 triples are tested. Direct interval substitution is unnecessarily ill-conditioned, so the certificate retains the root-invariant correlation: implicit differentiation of the Satake sextic gives uniform rational bounds on the root derivatives, and hence on the three derivatives of Gruenewald’s 67-term component equation. The smallest exact-center separation is 1658.56649, while the largest mean-value variation budget is 0.166232; branch by branch the minimum final separation is
∣ f 17 ∣ > 1658.5644. \lvert f_{17}\rvert> 1658.5644. ∣ f 17 ∣ > 1658.5644.
Therefore J X ∉ H 17 J_X \notin H_{17} J X ∈ / H 17 .
For H 20 H_{20} H 20 a level-2 component distinguishes a pair of Satake coordinates. The same six analytic root disks exhaust the possibilities, and all ( 6 2 ) = 15 \binom{6}{2}=15 ( 2 6 ) = 15 pairs can now be handled directly by exact complex-ball arithmetic. Gruenewald’s 24-term component equation satisfies
∣ f 20 ∣ > 158.6128 \lvert f_{20}\rvert> 158.6128 ∣ f 20 ∣ > 158.6128
on the worst pair over the full tight invariant box. Hence J X ∉ H 20 J_X \notin H_{20} J X ∈ / H 20 , completing (49) .
The directed and exact-rational audits are recorded in the eleven principal programs of certificates/humbert/, listed in the computational index.
Remark 3.22 (Kummer interpretation). The theorem does not say that the Néron–Severi group of J X J_X J X has minimal rank, nor that Kum ( X 2 ) \operatorname{Kum}(X_2) Kum ( X 2 ) has Picard number exactly 17: there are infinitely many Humbert surfaces. It does show that the certified Ξ / D 6 \Xi/D_6 Ξ/ D 6 point avoids nine low-complexity or Mordell–Weil-height-selected enhancements tested here. For square discriminants 4, 9, 16, the excluded loci are precisely the optimal degree-2, 3, 4 elliptic-subcover loci. The contrast with the earlier curve
y 2 = x 6 − 6 x 4 + 9 x 2 − 1 y^2=x^6-6x^4+9x^2-1 y 2 = x 6 − 6 x 4 + 9 x 2 − 1
from [13 ] is concrete: its even sextic has the extra involution x ↦ − x x\mapsto-x x ↦ − x and therefore lies on H 4 H_4 H 4 , whereas the Xi curve is rigorously off H 4 H_4 H 4 . Thus the common sixteen-node Kummer mechanism does not force the two constructions into the same arithmetic chamber.
Explicit Kummer and Shioda–Inose models
The preceding Humbert exclusions are moduli statements. We now make the associated K3 geometry explicit. This is useful for two reasons: it gives a concrete projective equation for the Xi Kummer surface, and it exhibits the rank-17 lattice which is present before any possible Noether–Lefschetz jump, meaning an increase of Picard number through an additional algebraic divisor class. A Shioda–Inose partner is a K3 surface with a degree-two rational map to the Kummer surface and a compatible Hodge isometry of transcendental lattices. All numerical boxes in this subsection are status [A].
A conditioned sextic and the Cassels–Flynn Kummer quartic
For numerical conditioning only, make the fixed real affine change
x = T − m 0 h 0 , m 0 = − 9.70103844 , h 0 = 56.08403609. (53) x=\frac{T-m_0}{h_0},\qquad m_0=-9.70103844,\qquad h_0=56.08403609. \tag*{(53)} x = h 0 T − m 0 , m 0 = − 9.70103844 , h 0 = 56.08403609. (53)
The exact six certified branch values then define the monic sextic
C X : y 2 = f X ( x ) = ∑ j = 0 6 f j x j = ∏ ν = 1 6 ( x − e ν ) , f 6 = 1. (54) C_X:\qquad y^2=f_X(x)=\sum_{j=0}^{6}f_jx^j=\prod_{\nu=1}^{6}(x-e_\nu),\qquad f_6=1. \tag*{(54)} C X : y 2 = f X ( x ) = j = 0 ∑ 6 f j x j = ν = 1 ∏ 6 ( x − e ν ) , f 6 = 1. (54)
Exact rational complex-ball propagation from the branch disks gives the following outward coefficient enclosures. Let ( κ 1 : κ 2 : κ 3 : κ 4 ) (\kappa_1:\kappa_2:\kappa_3:\kappa_4) ( κ 1 : κ 2 : κ 3 : κ 4 ) be the classical Kummer coordinates of Cassels–Flynn [18 ] . For a sextic f = ∑ j = 0 6 f j x j f=\sum_{j=0}^{6}f_jx^j f = ∑ j = 0 6 f j x j , the singular Kummer quartic is
K X : K 2 κ 4 2 + K 1 κ 4 + K 0 = 0. (55) \mathcal{K}_X:\qquad K_2\kappa_4^2+K_1\kappa_4+K_0=0. \tag*{(55)} K X : K 2 κ 4 2 + K 1 κ 4 + K 0 = 0. (55)
where
K 2 = κ 2 2 − 4 κ 1 κ 3 (56) K_2 = \kappa_2^2 - 4\kappa_1\kappa_3 \tag*{(56)} K 2 = κ 2 2 − 4 κ 1 κ 3 (56)
K 1 = − 2 ( 2 f 0 κ 1 3 + f 1 κ 1 2 κ 2 + 2 f 2 κ 1 2 κ 3 + f 3 κ 1 κ 2 κ 3 + 2 f 4 κ 1 κ 3 2 + f 5 κ 2 κ 3 2 + 2 f 6 κ 3 3 ) , (57) \begin{aligned}
K_1 ={}& -2(2f_0\kappa_1^3 + f_1\kappa_1^2\kappa_2 + 2f_2\kappa_1^2\kappa_3 + f_3\kappa_1\kappa_2\kappa_3 \\
&\quad+ 2f_4\kappa_1\kappa_3^2 + f_5\kappa_2\kappa_3^2 + 2f_6\kappa_3^3), \tag*{(57)}
\end{aligned} K 1 = − 2 ( 2 f 0 κ 1 3 + f 1 κ 1 2 κ 2 + 2 f 2 κ 1 2 κ 3 + f 3 κ 1 κ 2 κ 3 + 2 f 4 κ 1 κ 3 2 + f 5 κ 2 κ 3 2 + 2 f 6 κ 3 3 ) , (57)
and
K 0 = ( f 1 2 − 4 f 0 f 2 ) κ 1 4 − 4 f 0 f 3 κ 1 3 κ 2 − 2 f 1 f 3 κ 1 3 κ 3 − 4 f 0 f 4 κ 1 2 κ 2 2 + 4 ( f 0 f 5 − f 1 f 4 ) κ 1 2 κ 2 κ 3 + ( f 3 2 + 2 f 1 f 5 − 4 f 2 f 4 − 4 f 0 f 6 ) κ 1 2 κ 3 2 − 4 f 0 f 5 κ 1 κ 2 3 + 4 ( 2 f 0 f 6 − f 1 f 5 ) κ 1 κ 2 2 κ 3 + 4 ( f 1 f 6 − f 2 f 5 ) κ 1 κ 2 κ 3 2 − 2 f 3 f 5 κ 1 κ 3 3 − 4 f 0 f 6 κ 2 4 − 4 f 1 f 6 κ 2 3 κ 3 − 4 f 2 f 6 κ 2 2 κ 3 2 − 4 f 3 f 6 κ 2 κ 3 3 + ( f 5 2 − 4 f 4 f 6 ) κ 3 4 . (58) \begin{aligned}
K_0 ={}& (f_1^2 - 4f_0f_2)\kappa_1^4 - 4f_0f_3\kappa_1^3\kappa_2 - 2f_1f_3\kappa_1^3\kappa_3 - 4f_0f_4\kappa_1^2\kappa_2^2 \\
&+ 4(f_0f_5 - f_1f_4)\kappa_1^2\kappa_2\kappa_3 + (f_3^2 + 2f_1f_5 - 4f_2f_4 - 4f_0f_6)\kappa_1^2\kappa_3^2 \\
&- 4f_0f_5\kappa_1\kappa_2^3 + 4(2f_0f_6 - f_1f_5)\kappa_1\kappa_2^2\kappa_3 + 4(f_1f_6 - f_2f_5)\kappa_1\kappa_2\kappa_3^2 \\
&- 2f_3f_5\kappa_1\kappa_3^3 - 4f_0f_6\kappa_2^4 - 4f_1f_6\kappa_2^3\kappa_3 - 4f_2f_6\kappa_2^2\kappa_3^2 \\
&- 4f_3f_6\kappa_2\kappa_3^3 + (f_5^2 - 4f_4f_6)\kappa_3^4. \tag*{(58)}
\end{aligned} K 0 = ( f 1 2 − 4 f 0 f 2 ) κ 1 4 − 4 f 0 f 3 κ 1 3 κ 2 − 2 f 1 f 3 κ 1 3 κ 3 − 4 f 0 f 4 κ 1 2 κ 2 2 + 4 ( f 0 f 5 − f 1 f 4 ) κ 1 2 κ 2 κ 3 + ( f 3 2 + 2 f 1 f 5 − 4 f 2 f 4 − 4 f 0 f 6 ) κ 1 2 κ 3 2 − 4 f 0 f 5 κ 1 κ 2 3 + 4 ( 2 f 0 f 6 − f 1 f 5 ) κ 1 κ 2 2 κ 3 + 4 ( f 1 f 6 − f 2 f 5 ) κ 1 κ 2 κ 3 2 − 2 f 3 f 5 κ 1 κ 3 3 − 4 f 0 f 6 κ 2 4 − 4 f 1 f 6 κ 2 3 κ 3 − 4 f 2 f 6 κ 2 2 κ 3 2 − 4 f 3 f 6 κ 2 κ 3 3 + ( f 5 2 − 4 f 4 f 6 ) κ 3 4 . (58)
Proposition 3.23 (Explicit Xi Kummer quartic). Equations (54) –(58) , with the exact sextic coefficients determined by the six certified Xi branch values, define the singular Kummer surface
K X = J X / { ± 1 } . \mathcal{K}_{X} = J_{X}/\{\pm1\}. K X = J X / { ± 1 } . Every coefficient of the quartic is enclosed by directed rational arithmetic; the largest absolute coefficient radius is 8.93 × 10 − 4 8.93 \times10^{-4} 8.93 × 1 0 − 4 , compared with a largest coefficient of size 8.37 × 10 5 8.37 \times10^{5} 8.37 × 1 0 5 . The minimal resolution is the K3 surface Kum ( X 2 ) \operatorname{Kum}(X_2) Kum ( X 2 ) .
As an exact convention check, the same symbolic formula applied to the curve
y 2 = x 6 − 6 x 4 + 9 x 2 − 1 y^2 = x^6 - 6x^4 + 9x^2 - 1 y 2 = x 6 − 6 x 4 + 9 x 2 − 1
from [13 ] reproduces term-for-term the Kummer quartic printed there. Thus the present Xi construction and the earlier biological Kummer construction use exactly the same classical quartic normalization, although theorem 3.21 shows that they occupy different arithmetic chambers.
A certified elliptic fibration on the Xi Kummer surface
Normalize the weighted Igusa–Clebsch coordinates by I 2 = 1 I_2 = 1 I 2 = 1 and put
A = i 1 1728 , B = − i 2 186624 , C = i 3 1944 , J = i 3 486 = 4 C , D = 1 24 . (59) A = \frac{i_1}{1728}, \qquad B = -\frac{i_2}{186624}, \qquad C = \frac{i_3}{1944}, \qquad J = \frac{i_3}{486} = 4C, \qquad D = \frac{1}{24}. \tag*{(59)} A = 1728 i 1 , B = − 186624 i 2 , C = 1944 i 3 , J = 486 i 3 = 4 C , D = 24 1 . (59)
From (47) ,
A = 0.013548550270667627 ± 5.79 × 10 − 10 , B = 0.0004110032534855317 ± 5.36 × 10 − 11 , C = − 4.62710239753765 × 10 − 7 ± 5.15 × 10 − 14 , J = − 1.85084095901506 × 10 − 6 ± 2.06 × 10 − 13 . (60) \begin{aligned}
A &= 0.013548550270667627 \pm5.79 \times10^{-10},\\
B &= 0.0004110032534855317 \pm5.36 \times10^{-11},\\
C &= -4.62710239753765 \times10^{-7} \pm5.15 \times10^{-14},\\
J &= -1.85084095901506 \times10^{-6} \pm2.06 \times10^{-13}. \tag*{(60)}
\end{aligned} A B C J = 0.013548550270667627 ± 5.79 × 1 0 − 10 , = 0.0004110032534855317 ± 5.36 × 1 0 − 11 , = − 4.62710239753765 × 1 0 − 7 ± 5.15 × 1 0 − 14 , = − 1.85084095901506 × 1 0 − 6 ± 2.06 × 1 0 − 13 . (60)
Kumar’s fibration 23 on the Jacobian Kummer surface [19 , 20 ] becomes
Y 2 = X 3 − 2 P ( t ) X 2 + ( P ( t ) 2 + J ( t − D ) ) X , P ( t ) = t 3 − A t + B . (61) Y^2 = X^3 - 2P(t)X^2 + \left(P(t)^2 + J(t-D)\right)X,\qquad P(t)=t^3-At+B. \tag*{(61)} Y 2 = X 3 − 2 P ( t ) X 2 + ( P ( t ) 2 + J ( t − D ) ) X , P ( t ) = t 3 − A t + B . (61)
The section ( X , Y ) = ( 0 , 0 ) (X,Y)=(0,0) ( X , Y ) = ( 0 , 0 ) is visibly two-torsion. Its discriminant factors as
Δ K ( t ) = − 64 J ( t − D ) S 6 ( t ) 2 , S 6 ( t ) = P ( t ) 2 + J ( t − D ) . (62) \Delta_{\mathrm{K}}(t)=-64J(t-D)S_6(t)^2,\qquad S_6(t)=P(t)^2+J(t-D). \tag*{(62)} Δ K ( t ) = − 64 J ( t − D ) S 6 ( t ) 2 , S 6 ( t ) = P ( t ) 2 + J ( t − D ) . (62)
Theorem 3.24 (Certified Kummer elliptic fibers). The polynomial S 6 S_6 S 6 has exactly six simple zeros, one in each disk of radius 10 − 4 10^{-4} 1 0 − 4 centered at
0.06072249283374816 , 0.11477131965968400 , 0.06072249283374816,\qquad0.11477131965968400, 0.06072249283374816 , 0.11477131965968400 , 0.04329613590809414 ± 0.01244029536308813 i , 0.04329613590809414 \pm0.01244029536308813i, 0.04329613590809414 ± 0.01244029536308813 i , − 0.13104304215481022 ± 0.01499370760980722 i . (63) -0.13104304215481022 \pm0.01499370760980722i. \tag*{(63)} − 0.13104304215481022 ± 0.01499370760980722 i . (63) Moreover ∣ P ( 1 / 24 ) ∣ > 8.11816 × 10 − 5 |P(1/24)|>8.11816\times10^{-5} ∣ P ( 1/24 ) ∣ > 8.11816 × 1 0 − 5 . Hence the elliptic K3 (61) has fiber configuration
I 5 ∗ + 6 I 2 + I 1 , (64) I_5^*+6I_2+I_1, \tag*{(64)} I 5 ∗ + 6 I 2 + I 1 , (64) with the I 5 ∗ I_5^* I 5 ∗ fiber at infinity and the I 1 I_1 I 1 fiber at t = 1 / 24 t=1/24 t = 1/24 .
Proof. The coefficient boxes in (60) are propagated directly into S 6 S_6 S 6 . Rouché estimates on the six disks in (63) have maximal ratio 0.00987 < 1 0.00987<1 0.00987 < 1 ; the disks are disjoint and deg S 6 = 6 \deg S_6=6 deg S 6 = 6 , so the list is exhaustive. The displayed lower bound for P ( D ) P(D) P ( D ) makes the factor t − D t-D t − D a simple discriminant zero; there c 4 ≠ 0 c_4\ne0 c 4 = 0 , so the fiber is I 1 I_1 I 1 . At a zero of S 6 S_6 S 6 , the disk is disjoint from D D D and J ≠ 0 J\ne0 J = 0 , hence P ( t ) 2 = − J ( t − D ) ≠ 0 P(t)^2=-J(t-D)\ne0 P ( t ) 2 = − J ( t − D ) = 0 ; moreover c 4 = 64 P ( t ) 2 c_4=64P(t)^2 c 4 = 64 P ( t ) 2 on S 6 = 0 S_6=0 S 6 = 0 , so the double discriminant zero is an I 2 I_2 I 2 fiber. At infinity, with u = 1 / t u=1/t u = 1/ t and the standard K3 Weierstrass scaling, exact symbolic algebra gives
ord u = 0 ( c 4 , c 6 , Δ ) = ( 2 , 3 , 11 ) , \operatorname{ord}_{u=0}(c_4,c_6,\Delta)=(2,3,11), ord u = 0 ( c 4 , c 6 , Δ ) = ( 2 , 3 , 11 ) ,
which is type I 5 ∗ I_5^* I 5 ∗ by Tate’s algorithm [21 ] . The Euler numbers 11 + 6 ⋅ 2 + 1 = 24 11+6\cdot2+1=24 11 + 6 ⋅ 2 + 1 = 24 provide the expected K3 check. □
The Shioda–Inose partner and its normal quartic
The Shioda–Inose K3 surface associated with J X J_X J X has Kumar’s normalized Weierstrass model [22 ]
y 2 = x 3 − t 3 ( 1 + A t ) x + t 5 ( D + B t + C t 2 ) . (65) y^2=x^3-t^3(1+At)x+t^5(D+Bt+Ct^2). \tag*{(65)} y 2 = x 3 − t 3 ( 1 + A t ) x + t 5 ( D + B t + C t 2 ) . (65)
Its discriminant is
Δ S I ( t ) = − 16 t 9 R 5 ( t ) , R 5 ( t ) = − 4 ( 1 + A t ) 3 + 27 t ( D + B t + C t 2 ) 2 . (66) \Delta_{\mathrm{SI}}(t)=-16t^9R_5(t),\qquad R_5(t)=-4(1+At)^3+27t(D+Bt+Ct^2)^2. \tag*{(66)} Δ SI ( t ) = − 16 t 9 R 5 ( t ) , R 5 ( t ) = − 4 ( 1 + A t ) 3 + 27 t ( D + B t + C t 2 ) 2 . (66)
Theorem 3.25 (Certified Shioda–Inose fiber pattern). The five zeros of R 5 R_5 R 5 are simple and lie, one each, in disks of radius 0.1 0.1 0.1 centered at
− 246.0708096953812 , − 86.13686784034077 , 2302.762231755577 , -246.0708096953812,\qquad-86.13686784034077,\qquad2302.762231755577, − 246.0708096953812 , − 86.13686784034077 , 2302.762231755577 , − 97.02536873579858 ± 69.01396595582464 i . (67) -97.02536873579858\pm69.01396595582464i. \tag*{(67)} − 97.02536873579858 ± 69.01396595582464 i . (67) On all five disks one has t ( 1 + A t ) ≠ 0 t(1+At)\ne0 t ( 1 + A t ) = 0 . Therefore (65) has fiber configuration
I I ∗ + I I I ∗ + 5 I 1 , (68) II^*+III^*+5I_1, \tag*{(68)} I I ∗ + I I I ∗ + 5 I 1 , (68)
where I I I ∗ III^{*} I I I ∗ is at t = 0 t = 0 t = 0 and I I ∗ II^{*} I I ∗ at t = ∞ t = \infty t = ∞ .
Proof. Directed Rouché bounds for R 5 R_5 R 5 on the five disks have maximal ratio 0.01266 < 1 0.01266 < 1 0.01266 < 1 ; since deg R 5 = 5 \deg R_5 = 5 deg R 5 = 5 , the list is exhaustive. The certified lower bounds on the residual disks are
∣ t ∣ > 86.03 , ∣ 1 + A t ∣ > 0.1656 , |t| > 86.03, \qquad|1 + At| > 0.1656, ∣ t ∣ > 86.03 , ∣1 + A t ∣ > 0.1656 ,
so the residual simple discriminant zeros have c 4 ≠ 0 c_4 \ne0 c 4 = 0 and are of type I 1 I_1 I 1 . At t = 0 t = 0 t = 0 the exact vanishing orders are ( 3 , 5 , 9 ) (3,5,9) ( 3 , 5 , 9 ) , giving I I I ∗ III^{*} I I I ∗ . After the standard K3 coordinate change at infinity the exact orders are ( 4 , 5 , 10 ) (4,5,10) ( 4 , 5 , 10 ) , giving I I ∗ II^{*} I I ∗ by Tate’s algorithm [21 ] . Again 9 + 10 + 5 = 24 9 + 10 + 5 = 24 9 + 10 + 5 = 24 . □
The same Shioda–Inose surface admits the Clingher–Doran N N N -polarized quartic normal form [23 ]
Y 2 Z W − 4 X 3 Z + 3 α X Z W 2 + β Z W 3 + γ X Z 2 W − 1 2 ( δ Z 2 W 2 + W 4 ) = 0 , (69) Y^2ZW - 4X^3Z + 3\alpha XZW^2 + \beta ZW^3 + \gamma XZ^2W - \frac{1}{2}\left(\delta Z^2W^2 + W^4\right) = 0, \tag*{(69)} Y 2 Z W − 4 X 3 Z + 3 α X Z W 2 + β Z W 3 + γ X Z 2 W − 2 1 ( δ Z 2 W 2 + W 4 ) = 0 , (69)
with, in the gauge I 2 = 1 I_2 = 1 I 2 = 1 ,
α = i 1 576 , β = − i 2 13824 , γ = − i 3 8 , δ = i 3 64 . (70) \alpha= \frac{i_1}{576}, \qquad\beta= -\frac{i_2}{13824}, \qquad\gamma= -\frac{i_3}{8}, \qquad\delta= \frac{i_3}{64}. \tag*{(70)} α = 576 i 1 , β = − 13824 i 2 , γ = − 8 i 3 , δ = 64 i 3 . (70)
The certified parameter boxes are
α = 0.04064565081200288 ± 1.74 × 10 − 9 , β = 0.005548543922054678 ± 7.24 × 10 − 10 , γ = 1.124385882601649 × 10 − 4 ± 1.25 × 10 − 11 , δ = − 1.405482353252061 × 10 − 5 ± 1.57 × 10 − 12 . (71) \begin{aligned}
\alpha&= 0.04064565081200288 \pm1.74 \times10^{-9},\\
\beta&= 0.005548543922054678 \pm7.24 \times10^{-10},\\
\gamma&= 1.124385882601649 \times10^{-4} \pm1.25 \times10^{-11},\\
\delta&= -1.405482353252061 \times10^{-5} \pm1.57 \times10^{-12}. \tag*{(71)}
\end{aligned} α β γ δ = 0.04064565081200288 ± 1.74 × 1 0 − 9 , = 0.005548543922054678 ± 7.24 × 1 0 − 10 , = 1.124385882601649 × 1 0 − 4 ± 1.25 × 1 0 − 11 , = − 1.405482353252061 × 1 0 − 5 ± 1.57 × 1 0 − 12 . (71)
Thus (69) is a second explicit projective model attached to exactly the same Xi genus-two moduli point.
Néron–Severi lattice frontier
The lattice U U U is the rank-two hyperbolic plane; A n A_n A n , D n D_n D n , E n E_n E n denote the usual positive-definite root lattices, and L ( − 1 ) L(-1) L ( − 1 ) means reversal of the bilinear form of L L L . A sublattice is primitive when the quotient of its ambient lattice by that sublattice is torsion-free; the transcendental lattice is the orthogonal complement of the Néron–Severi lattice in integral second cohomology. A lattice polarization is a primitive embedding of the specified lattice into that divisor lattice.
The fiber configuration (68) exhibits a primitive lattice polarization
N = U ⊕ E 8 ( − 1 ) ⊕ E 7 ( − 1 ) ↪ NS ( SI ( J X ) ) , (72) N = U \oplus E_8(-1) \oplus E_7(-1) \hookrightarrow\operatorname{NS}(\operatorname{SI}(J_X)), \tag*{(72)} N = U ⊕ E 8 ( − 1 ) ⊕ E 7 ( − 1 ) ↪ NS ( SI ( J X )) , (72)
of rank 17 and discriminant 2; this is precisely the Clingher–Doran/Kumar N N N -polarized moduli realization [22 , 23 ] . On the Kummer side, (64) gives the rank-17 trivial lattice
U ⊕ D 9 ( − 1 ) ⊕ A 1 ( − 1 ) ⊕ 6 ⊆ NS ( Kum ( X 2 ) ) , (73) U \oplus D_9(-1) \oplus A_1(-1)^{\oplus6} \subseteq\operatorname{NS}(\operatorname{Kum}(X_2)), \tag*{(73)} U ⊕ D 9 ( − 1 ) ⊕ A 1 ( − 1 ) ⊕ 6 ⊆ NS ( Kum ( X 2 )) , (73)
together with the visible two-torsion section.
Proposition 3.26 (Picard-rank reduction to the remaining Humbert problem). Let ρ X = rank NS ( J X ) \rho_X = \operatorname{rank}\operatorname{NS}(J_X) ρ X = rank NS ( J X ) . Then
ρ ( Kum ( X 2 ) ) = ρ ( SI ( J X ) ) = 16 + ρ X . (74) \rho(\operatorname{Kum}(X_2)) = \rho(\operatorname{SI}(J_X)) = 16 + \rho_X. \tag*{(74)} ρ ( Kum ( X 2 )) = ρ ( SI ( J X )) = 16 + ρ X . (74) Equivalently, for the two displayed elliptic fibrations,
ρ ( SI ( J X ) ) = 17 + rank MW ( π S I ) , ρ ( Kum ( X 2 ) ) = 17 + rank MW ( π K ) . (75) \begin{aligned}
\rho(\operatorname{SI}(J_X)) &= 17 + \operatorname{rank}\operatorname{MW}(\pi_{\mathrm{SI}}),\\
\rho(\operatorname{Kum}(X_2)) &= 17 + \operatorname{rank}\operatorname{MW}(\pi_{\mathrm{K}}). \tag*{(75)}
\end{aligned} ρ ( SI ( J X )) ρ ( Kum ( X 2 )) = 17 + rank MW ( π SI ) , = 17 + rank MW ( π K ) . (75)
Consequently the minimal possibility is Picard rank 17. If it occurs, then
NS ( SI ( J X ) ) = N , \operatorname{NS}(\operatorname{SI}(J_X)) = N, NS ( SI ( J X )) = N ,
and the Kummer Néron–Severi lattice is the standard generic (16, 6) Jacobian-Kummer lattice Λ ( 16 , 6 ) \Lambda_{(16,6)} Λ ( 16 , 6 ) of signature ( 1 , 16 ) (1, 16) ( 1 , 16 ) and discriminant 2 6 2^6 2 6 .
Proof. For an abelian surface A A A , the transcendental lattice has rank 6 − ρ ( A ) 6-\rho(A) 6 − ρ ( A ) . A Shioda–Inose structure preserves its transcendental lattice up to Hodge isometry, while passage to the Kummer surface rescales the intersection form by 2 without changing its rank. Since a K3 surface has second Betti number 22, (74) follows. The two Shioda–Tate formulas follow independently from the root ranks 8 + 7 = 15 8+7=15 8 + 7 = 15 and 9 + 6 = 15 9+6=15 9 + 6 = 15 . If the first surface has rank 17, the primitive rank-17 lattice N N N already fills its Néron–Severi group; the generic rank-17 Kummer lattice is the classical node–trope lattice Λ ( 16 , 6 ) \Lambda_{(16,6)} Λ ( 16 , 6 ) [22 ] . □
Proposition 3.27 (Certified Mordell–Weil height barrier). Let P ≠ O P \ne O P = O be a section of the Shioda–Inose fibration π S I \pi_{\mathrm{SI}} π SI and put n = ( P ⋅ O ) ≥ 0 n=(P\mathbin{\cdot}O)\ge0 n = ( P ⋅ O ) ≥ 0 . Then
h ^ ( P ) = 4 + 2 n − contr I I I ∗ ( P ) , contr I I I ∗ ( P ) ∈ { 0 , 3 2 } . (76) \hat{h}(P)=4+2n-\operatorname{contr}_{\mathrm{III}^{*}}(P),\qquad\operatorname{contr}_{\mathrm{III}^{*}}(P)\in\left\{0,\frac{3}{2}\right\}. \tag*{(76)} h ^ ( P ) = 4 + 2 n − contr III ∗ ( P ) , contr III ∗ ( P ) ∈ { 0 , 2 3 } . (76) Consequently MW ( π S I ) t o r s = 0 \operatorname{MW}(\pi_{\mathrm{SI}})_{\mathrm{tors}}=0 MW ( π SI ) tors = 0 , and the nine Humbert exclusions of theorem 3.21 imply
P ≠ O ⟹ h ^ ( P ) ≥ 21 2 . (77) P\ne O\quad\Longrightarrow\quad\hat{h}(P)\ge\frac{21}{2}. \tag*{(77)} P = O ⟹ h ^ ( P ) ≥ 2 21 . (77) Equivalently, any hypothetical section has ( P ⋅ O ) ≥ 4 (P\mathbin{\cdot}O)\ge4 ( P ⋅ O ) ≥ 4 .
Proof. Shioda’s height pairing on an elliptic surface [24 ] gives h ^ ( P ) = 4 + 2 ( P ⋅ O ) − ∑ v contr v ( P ) \hat{h}(P)=4+2(P\mathbin{\cdot}O)-\sum_v\operatorname{contr}_v(P) h ^ ( P ) = 4 + 2 ( P ⋅ O ) − ∑ v contr v ( P ) . The I I ∗ \mathrm{II}^{*} II ∗ fiber contributes zero, while at the I I I ∗ \mathrm{III}^{*} III ∗ fiber the identity and nonidentity simple components contribute 0 and 3 / 2 3/2 3/2 , respectively. Thus the two height ladders are
h ^ ( P ) = 4 + 2 n , h ^ ( P ) = 5 2 + 2 n . \hat{h}(P)=4+2n,\qquad\hat{h}(P)=\frac{5}{2}+2n. h ^ ( P ) = 4 + 2 n , h ^ ( P ) = 2 5 + 2 n .
In particular every nonzero section has height at least 5 / 2 5/2 5/2 , so the Mordell–Weil torsion subgroup is trivial. Consequently the trivial lattice N = U ⊕ E 8 ( − 1 ) ⊕ E 7 ( − 1 ) N=U\oplus E_8(-1)\oplus E_7(-1) N = U ⊕ E 8 ( − 1 ) ⊕ E 7 ( − 1 ) is primitive in the Néron–Severi lattice: Shioda’s exact sequence identifies the quotient by the trivial lattice with the Mordell–Weil group, and there is no torsion left to produce a nontrivial saturation.
Now let P P P be primitive in the free Mordell–Weil group, and let L P L_P L P be the inverse image of Z P \mathbb{Z}P Z P in the Néron–Severi lattice. Then L P L_P L P is a primitive rank-18 extension of N N N . Shioda’s discriminant formula for the Mordell–Weil lattice gives
disc ( L P ) = − disc ( N ) h ^ ( P ) = − 2 h ^ ( P ) , \operatorname{disc}(L_P)=-\operatorname{disc}(N)\hat{h}(P)=-2\hat{h}(P), disc ( L P ) = − disc ( N ) h ^ ( P ) = − 2 h ^ ( P ) ,
where the sign is dictated by signature ( 1 , 17 ) (1,17) ( 1 , 17 ) . Elkies–Kumar describe the Humbert surface H D H_D H D as the image, in the N N N -polarized moduli space, of K3 surfaces polarized by a rank-18 lattice L D L_D L D of signature ( 1 , 17 ) (1,17) ( 1 , 17 ) and discriminant − D -D − D [15 ] . Thus the primitive extension L P L_P L P above has Humbert discriminant equal to its absolute discriminant. Hence
D = 2 h ^ ( P ) . (78) D=2\hat{h}(P). \tag*{(78)} D = 2 h ^ ( P ) . (78)
For n = 0 , 1 , 2 , 3 n=0,1,2,3 n = 0 , 1 , 2 , 3 the two ladders therefore give
Every section of height < 10 <10 < 10 is primitive: if P = m Q P=mQ P = m Q , m ≥ 2 m\ge2 m ≥ 2 , then h ^ ( Q ) = h ^ ( P ) / m 2 < 5 / 2 \hat{h}(Q)=\hat{h}(P)/m^2<5/2 h ^ ( Q ) = h ^ ( P ) / m 2 < 5/2 , a contradiction. The only boundary case is h ^ ( P ) = 10 \hat{h}(P)=10 h ^ ( P ) = 10 ; if P = 2 Q P=2Q P = 2 Q then h ^ ( Q ) = 5 / 2 \hat{h}(Q)=5/2 h ^ ( Q ) = 5/2 , which would put the Xi point on the already excluded H 5 H_5 H 5 , while m ≥ 3 m \ge3 m ≥ 3 is again impossible. Hence all eight entries in the table are valid primitive Humbert obstructions. Since H 5 H_5 H 5 , H 8 H_8 H 8 , H 9 H_9 H 9 , H 12 H_{12} H 12 , H 13 H_{13} H 13 , H 16 H_{16} H 16 , H 17 H_{17} H 17 , H 20 H_{20} H 20 are excluded by theorem 3.21, one must have n ≥ 4 n \ge4 n ≥ 4 . The smaller of the two heights at n = 4 n = 4 n = 4 is 5 / 2 + 8 = 21 / 2 5/2 + 8 = 21/2 5/2 + 8 = 21/2 , proving (77) .
Remark 3.28 (What is and is not proved about the lattice). We do not claim that the Picard rank is exactly 17. By Theorem 3.21 and Proposition 3.27 , the first four rungs of both Mordell–Weil height ladders are absent, but infinitely many Humbert divisors remain. Thus the explicit K3 models turn the previous vague lattice question into a precise one: decide whether either Mordell–Weil rank in (75) is positive, equivalently whether the Xi moduli point lies on some as-yet-unexcluded Humbert/Noether–Lefschetz divisor. The next formal ladder entries would be H 21 H_{21} H 21 and H 24 H_{24} H 24 , but continuing this finite elimination indefinitely cannot prove rank zero. The structural target is instead an exact Hodge/arithmetic argument showing that the Xi period point lies on no Humbert divisor at all.
For this real type the Vinnikov–Weichold normal form has H = I 2 H = I_2 H = I 2 ; see [25 , 26 ] . Hence one can choose an integral symplectic basis
A 1 , A 2 , B 1 , B 2 A_1, A_2, B_1, B_2 A 1 , A 2 , B 1 , B 2
with
τ ∗ A j = A j , τ ∗ B j = A j − B j . (79) \tau_* A_j = A_j,\qquad\tau_* B_j = A_j - B_j. \tag*{(79)} τ ∗ A j = A j , τ ∗ B j = A j − B j . (79)
Global Humbert reduction on the real period slice
Let η 1 \eta_1 η 1 , η 2 \eta_2 η 2 be the normalized holomorphic differentials, ∫ A j η k = δ j k \int_{A_j} \eta_k = \delta_{jk} ∫ A j η k = δ j k , and let
Ω j k = ∫ B j η k \Omega_{jk} = \int_{B_j} \eta_k Ω j k = ∫ B j η k
be the normalized genus-two period matrix. The real involution and (79) force a particularly rigid real slice of Siegel space.
Proposition 3.29 (Vinnikov form of the holomorphic period matrix). There is a real symmetric positive-definite matrix
Y = ( y 1 y 12 y 12 y 2 ) > 0 Y = \begin{pmatrix} y_1 & y_{12} \\ y_{12} & y_2 \end{pmatrix} > 0 Y = ( y 1 y 12 y 12 y 2 ) > 0 such that
Ω = 1 2 I 2 + i Y . (80) \Omega= \frac{1}{2} I_2 + iY. \tag*{(80)} Ω = 2 1 I 2 + iY . (80) Proof. The holomorphic vector τ ∗ η \tau^*\eta τ ∗ η has the same A A A -periods as η \eta η , hence equals η \eta η by normalization. Therefore
Ω ‾ j k = ∫ B j η ‾ k = ∫ τ ∗ B j η k = δ j k − Ω j k . \overline{\Omega}_{jk} = \int_{B_j} \overline{\eta}_k = \int_{\tau_*B_j} \eta_k = \delta_{jk} - \Omega_{jk}. Ω j k = ∫ B j η k = ∫ τ ∗ B j η k = δ j k − Ω j k . Thus 2 Re Ω = I 2 2\operatorname{Re}\Omega= I_2 2 Re Ω = I 2 . Symmetry and positive definiteness of Y = Im Ω Y = \operatorname{Im}\Omega Y = Im Ω are the Riemann bilinear relations.
Recall Humbert’s singular relation for a principally polarized abelian surface,
a Ω 11 + b Ω 12 + c Ω 22 + d ( Ω 12 2 − Ω 11 Ω 22 ) + e = 0 , a , b , c , d , e ∈ Z , (81) a\Omega_{11} + b\Omega_{12} + c\Omega_{22} + d(\Omega_{12}^{2} - \Omega_{11}\Omega_{22}) + e = 0,\qquad a,b,c,d,e \in\mathbb{Z}, \tag*{(81)} a Ω 11 + b Ω 12 + c Ω 22 + d ( Ω 12 2 − Ω 11 Ω 22 ) + e = 0 , a , b , c , d , e ∈ Z , (81)
with primitive discriminant
Δ = b 2 − 4 a c − 4 d e > 0. (82) \Delta= b^2 - 4ac - 4de > 0. \tag*{(82)} Δ = b 2 − 4 a c − 4 d e > 0. (82)
The polarization induces the Rosati involution f ↦ f † f \mapsto f^\dagger f ↦ f † on rational endomorphisms; “Rosati-symmetric” means f = f † f = f^\dagger f = f † . Such a relation is equivalent to an additional Rosati-symmetric endomorphism of the principally polarized abelian surface; its moduli image lies on the Humbert surface H Δ H_{\Delta} H Δ [16 ] .
Theorem 3.30 (Global real-slice Humbert criterion). Put
q = det Y = y 1 y 2 − y 12 2 > 0. q=\det Y=y_1y_2-y_{12}^{2}>0. q = det Y = y 1 y 2 − y 12 2 > 0. Then the following are equivalent:
(i) J X J_X J X lies on no Humbert divisor;
(ii) NS ( J X ) = Z [ Θ ] \operatorname{NS}(J_X)=\mathbb{Z}[\Theta] NS ( J X ) = Z [ Θ ] and hence ρ ( J X ) = 1 \rho(J_X)=1 ρ ( J X ) = 1 ;
(iii)
q ∉ Q , dim Q ⟨ y 1 , y 12 , y 2 ⟩ Q = 3. (83) q\notin\mathbb{Q},\qquad\dim_{\mathbb{Q}}\langle y_1,y_{12},y_2\rangle_{\mathbb{Q}}=3. \tag*{(83)} q ∈ / Q , dim Q ⟨ y 1 , y 12 , y 2 ⟩ Q = 3. (83) Consequently
ρ ( SI ( J X ) ) = 17 ⟺ q ∉ Q and y 1 , y 12 , y 2 are Q -linearly independent . (84) \rho(\operatorname{SI}(J_X))=17\Longleftrightarrow q\notin\mathbb{Q}\ \text{and}\ y_1,y_{12},y_2\ \text{are }\mathbb{Q}\text{-linearly independent}. \tag*{(84)} ρ ( SI ( J X )) = 17 ⟺ q ∈ / Q and y 1 , y 12 , y 2 are Q -linearly independent . (84)
Proof. Substitute (80) into (81) . Its imaginary and real parts are, respectively,
( 2 a − d ) y 1 + 2 b y 12 + ( 2 c − d ) y 2 = 0 , (85) (2a-d)y_1+2by_{12}+(2c-d)y_2=0, \tag*{(85)} ( 2 a − d ) y 1 + 2 b y 12 + ( 2 c − d ) y 2 = 0 , (85)
4 d q + 2 a + 2 c − d + 4 e = 0. (86) 4dq+2a+2c-d+4e=0. \tag*{(86)} 4 d q + 2 a + 2 c − d + 4 e = 0. (86)
If d ≠ 0 d\ne0 d = 0 , (86) gives q ∈ Q q\in\mathbb{Q} q ∈ Q . If d = 0 d=0 d = 0 , (85) is a nontrivial rational linear relation among the three entries of Y Y Y . Hence (83) excludes every singular relation at once.
Conversely suppose first that q = P / S ∈ Q > 0 q=P/S\in\mathbb{Q}_{>0} q = P / S ∈ Q > 0 in lowest terms. The integer quintuple
( a , b , c , d , e ) = ( 2 S , 0 , 2 S , 4 S , − ( 4 P + S ) ) (87) (a,b,c,d,e)=(2S,0,2S,4S,-(4P+S)) \tag*{(87)} ( a , b , c , d , e ) = ( 2 S , 0 , 2 S , 4 S , − ( 4 P + S )) (87)
satisfies (81) ; before primitive division its discriminant is
Δ = 64 P S > 0. \Delta=64PS>0. Δ = 64 P S > 0.
Thus rationality of det Y \det Y det Y itself forces membership in a Humbert divisor.
Now suppose A y 1 + B y 12 + C y 2 = 0 Ay_1+By_{12}+Cy_2=0 A y 1 + B y 12 + C y 2 = 0 for nonzero ( A , B , C ) ∈ Z 3 (A,B,C)\in\mathbb{Z}^{3} ( A , B , C ) ∈ Z 3 . If A + C A+C A + C is even, take ( a , b , c , d , e ) = ( A , B , C , 0 , − ( A + C ) / 2 ) (a,b,c,d,e)=(A,B,C,0,-(A+C)/2) ( a , b , c , d , e ) = ( A , B , C , 0 , − ( A + C ) /2 ) . If A + C A+C A + C is odd, replace ( A , B , C ) (A,B,C) ( A , B , C ) by ( 2 A , 2 B , 2 C ) (2A,2B,2C) ( 2 A , 2 B , 2 C ) and take ( a , b , c , d , e ) = ( 2 A , 2 B , 2 C , 0 , − ( A + C ) ) (a,b,c,d,e)=(2A,2B,2C,0,-(A+C)) ( a , b , c , d , e ) = ( 2 A , 2 B , 2 C , 0 , − ( A + C )) . In either case this gives an integral singular relation. Its discriminant is positive: for the original relation, with
M = ( A B / 2 B / 2 C ) M=\begin{pmatrix} A & B/2 \\ B/2 & C \end{pmatrix} M = ( A B /2 B /2 C )
one has tr ( M Y ) = 0 \operatorname{tr}(MY)=0 tr ( M Y ) = 0 . Since Y > 0 Y>0 Y > 0 , the nonzero real symmetric matrix Y 1 / 2 M Y 1 / 2 Y^{1/2}MY^{1/2} Y 1/2 M Y 1/2 has trace zero and hence eigenvalues λ , − λ \lambda,-\lambda λ , − λ with λ ≠ 0 \lambda\ne0 λ = 0 . Therefore det M < 0 \det M<0 det M < 0 , i.e.
B 2 − 4 A C > 0. B^{2}-4AC>0. B 2 − 4 A C > 0.
After dividing the five coefficients by their common divisor one obtains a primitive Humbert relation. This proves the equivalence between (i) and (iii). The equivalence with (ii) is the standard identification of extra Néron–Severi classes with non-scalar Rosati-symmetric endomorphisms of a principally polarized abelian surface. Finally (84) follows from (74) . □
Corollary 3.31 (Exact endpoint of the finite Humbert ladder). The nine certified exclusions in theorem 3.21 are finite-height evidence for the two arithmetic conditions in (83) , but no continuation through H 21 , H 24 , … H_{21},H_{24},\ldots H 21 , H 24 , … can replace them. The remaining Picard-rank problem is exactly the pair of irrationality statements in (83) .
Proposition 3.33 (Density obstruction on the Vinnikov slice). Inside the real slice
V = { 1 2 I 2 + i Y : Y = Y T > 0 } ⊂ H 2 , \mathcal{V}=\left\{\frac{1}{2}I_{2}+iY:Y=Y^{\mathsf{T}}>0\right\}\subset\mathfrak{H}_{2}, V = { 2 1 I 2 + iY : Y = Y T > 0 } ⊂ H 2 , the union of all Humbert divisors is dense. Consequently no open interval or ball enclosure of the Xi matrix Y Y Y —however small—can by itself prove that J X J_{X} J X lies on no Humbert divisor.
Proof. Fix Y = ( y j k ) > 0 Y=(y_{jk})>0 Y = ( y j k ) > 0 and put q = det Y > 0 q=\det Y>0 q = det Y > 0 . Choose any sequence q n ∈ Q > 0 q_{n}\in\mathbb{Q}_{>0} q n ∈ Q > 0 with q n → q q_{n}\to q q n → q and set
Y n = ( y 1 y 12 y 12 ( q n + y 12 2 ) / y 1 ) . Y_{n}=\begin{pmatrix} y_{1} & y_{12} \\ y_{12} & (q_{n}+y_{12}^{2})/y_{1} \end{pmatrix}. Y n = ( y 1 y 12 y 12 ( q n + y 12 2 ) / y 1 ) .
Then Y n > 0 Y_{n}>0 Y n > 0 , Y n → Y Y_{n}\to Y Y n → Y , and det Y n = q n ∈ Q \det Y_{n}=q_{n}\in\mathbb{Q} det Y n = q n ∈ Q . By the converse part of theorem 3.30, every 1 2 I 2 + i Y n \frac{1}{2}I_{2}+iY_{n} 2 1 I 2 + i Y n satisfies a positive-discriminant singular relation and therefore lies on a Humbert divisor. This proves density. The final statement follows because every nonempty neighborhood of the Xi period point contains such a Humbert point. □
Proposition 3.35 (Numerical holomorphic period matrix). Using the certified branch centers in the conditioned coordinate (53) , endpoint-regularized Gauss–Legendre integration of d x / y \mathrm{d}x/y d x / y and x d x / y x\,\mathrm{d}x/y x d x / y gives, at status [N], the Vinnikov-gauge matrix
Y ≈ ( 1.23522084149017 − 1.52074137328966 − 1.52074137328966 2.55233871684607 ) , det Y ≈ 0.840047653155622. (88) Y\approx\begin{pmatrix}1.23522084149017 & -1.52074137328966 \\ -1.52074137328966 & 2.55233871684607\end{pmatrix},\qquad\det Y\approx0.840047653155622. \tag*{(88)} Y ≈ ( 1.23522084149017 − 1.52074137328966 − 1.52074137328966 2.55233871684607 ) , det Y ≈ 0.840047653155622. (88) The eigenvalues are approximately 0.2365669990 0.2365669990 0.2365669990 and 3.5509925594 3.5509925594 3.5509925594 .
Numerical construction. In the natural three-cut basis the midpoint computation gives
2 Re Ω c u t = H c u t = ( 1 1 1 0 ) . 2\operatorname{Re}\Omega_{\mathrm{cut}}=H_{\mathrm{cut}}=\begin{pmatrix}1 & 1 \\ 1 & 0\end{pmatrix}. 2 Re Ω cut = H cut = ( 1 1 1 0 ) . With
U = ( − 1 − 1 0 − 1 ) , U − 1 = ( − 1 1 0 − 1 ) , U=\begin{pmatrix}-1 & -1 \\ 0 & -1\end{pmatrix},\qquad U^{-1}=\begin{pmatrix}-1 & 1 \\ 0 & -1\end{pmatrix}, U = ( − 1 0 − 1 − 1 ) , U − 1 = ( − 1 0 1 − 1 ) , one has ( U − 1 ) T H c u t U − 1 = diag ( 1 , − 1 ) (U^{-1})^{\mathsf{T}}H_{\mathrm{cut}}U^{-1}=\operatorname{diag}(1,-1) ( U − 1 ) T H cut U − 1 = diag ( 1 , − 1 ) . The integral symmetric shear S = diag ( 0 , 1 ) S=\operatorname{diag}(0,1) S = diag ( 0 , 1 ) then changes this to I 2 I_{2} I 2 without altering the imaginary part. Applying the same integral congruence to the numerical Riemann matrix gives (88) . The endpoint substitution z = r a + ( r b − r a ) ( 1 − cos θ ) / 2 z=r_{a}+(r_{b}-r_{a})(1-\cos\theta)/2 z = r a + ( r b − r a ) ( 1 − cos θ ) /2 removes the square-root endpoint singularities analytically. Orders 100 and 140 of the midpoint Gauss–Legendre calculation differ by less than 3.4 × 10 − 24 3.4\times10^{-24} 3.4 × 1 0 − 24 . This last number measures quadrature convergence for the midpoint curve only; it does not include propagation of the certified branch disks, hence the status remains [N]. □
Conjecture 3.36 (Minimal Picard rank for the Xi core). For the compactified Xi genus-two core,
N S ( J X ) = Z [ Θ ] , ρ ( SI ( J X ) ) = 17. \mathrm{NS}(J_X)=\mathbb{Z}[\Theta],\qquad\rho(\operatorname{SI}(J_X))=17. NS ( J X ) = Z [ Θ ] , ρ ( SI ( J X )) = 17. Equivalently, the two conditions in (83) hold.
Natural heat deformation and the isomonodromic tangent
The fixed-point realization in theorem 3.12 is an identity of curves at one phase point. A comparison of deformation directions requires a specified continuation of the Xi discriminant. We use the continuation in which both the theta moments and the central normalization A ∗ A_\ast A ∗ evolve under the same heat flow. The isomonodromic time below is denoted t D t_D t D ; it is the variable t t t of (37) , and is distinct from heat time.
Normalization and exact evolution
Define
Ξ κ ( z ) = 4 ∫ 0 ∞ e − κ u 2 Φ ( u ) cos ( z u ) d u , Y κ ( T ) = 1 4 Ξ κ ( i T ) . (89) \Xi_\kappa(z)=4\int_0^\infty e^{-\kappa u^2}\Phi(u)\cos(zu)\,\mathrm{d}u,\qquad Y_\kappa(T)=\frac{1}{4}\Xi_\kappa(i\sqrt{T}). \tag*{(89)} Ξ κ ( z ) = 4 ∫ 0 ∞ e − κ u 2 Φ ( u ) cos ( z u ) d u , Y κ ( T ) = 4 1 Ξ κ ( i T ) . (89)
The double-exponential decay of Φ \Phi Φ makes the integral locally uniformly convergent in ( κ , z ) ∈ C 2 (\kappa,z)\in\mathbb{C}^2 ( κ , z ) ∈ C 2 . Evenness removes the square root from Y κ Y_\kappa Y κ , which is entire in ( κ , T ) (\kappa,T) ( κ , T ) . In the Polymath normalization [29 ] , substitution u = 2 v u=2v u = 2 v gives
Ξ κ ( z ) = 8 H 4 κ ( 2 z ) , t P o l y m a t h = 4 κ . (90) \Xi_\kappa(z)=8H_{4\kappa}(2z),\qquad t_{\mathrm{Polymath}}=4\kappa. \tag*{(90)} Ξ κ ( z ) = 8 H 4 κ ( 2 z ) , t Polymath = 4 κ . (90)
In particular Ξ 0 = Ξ \Xi_0=\Xi Ξ 0 = Ξ and ∂ κ Ξ κ = − ∂ z 2 Ξ κ \partial_\kappa\Xi_\kappa=-\partial_z^2\Xi_\kappa ∂ κ Ξ κ = − ∂ z 2 Ξ κ .
Put M 2 n ( κ ) = ∫ 0 ∞ e − κ u 2 Φ ( u ) u 2 n d u M_{2n}(\kappa)=\int_0^\infty e^{-\kappa u^2}\Phi(u)u^{2n}\,\mathrm{d}u M 2 n ( κ ) = ∫ 0 ∞ e − κ u 2 Φ ( u ) u 2 n d u , and, where Y κ ≠ 0 Y_\kappa\ne0 Y κ = 0 , put A κ = 2 ∂ T log Y κ \mathcal{A}_\kappa=2\partial_T\log Y_\kappa A κ = 2 ∂ T log Y κ and A ∗ ( κ ) = A κ ( 0 ) A_\ast(\kappa)=\mathcal{A}_\kappa(0) A ∗ ( κ ) = A κ ( 0 ) . Define K κ K_\kappa K κ and Δ κ = K κ / Y κ 6 \Delta_\kappa=K_\kappa/Y_\kappa^6 Δ κ = K κ / Y κ 6 by the formulas of theorem 2.3, replacing Y Y Y and A ∗ A_\ast A ∗ by Y κ Y_\kappa Y κ and A ∗ ( κ ) A_\ast(\kappa) A ∗ ( κ ) . Thus the central normalization is updated, rather than frozen at its time-zero value.
Proposition 4.1 (Exact transverse heat evolution). With primes denoting T T T -derivatives,
∂ κ M 2 n ( κ ) = M 2 n + 2 ( κ ) , \partial_\kappa M_{2n}(\kappa)=M_{2n+2}(\kappa), ∂ κ M 2 n ( κ ) = M 2 n + 2 ( κ ) , ∂ κ Y κ = 2 Y κ ′ + 4 T Y κ ′ ′ . (91) \partial_\kappa Y_\kappa=2Y_\kappa'+4T Y_\kappa''. \tag*{(91)} ∂ κ Y κ = 2 Y κ ′ + 4 T Y κ ′′ . (91) ∂ κ A κ = 4 T A κ ′ ′ + ( 6 + 4 T A κ ) A κ ′ + 2 A κ 2 . (92) \partial_\kappa\mathcal{A}_\kappa=4T\mathcal{A}_\kappa''+(6+4T\mathcal{A}_\kappa)\mathcal{A}_\kappa'+2\mathcal{A}_\kappa^2. \tag*{(92)} ∂ κ A κ = 4 T A κ ′′ + ( 6 + 4 T A κ ) A κ ′ + 2 A κ 2 . (92) For real κ \kappa κ ,
∂ κ A ∗ ( κ ) = M 4 ( κ ) M 0 ( κ ) − M 2 ( κ ) 2 M 0 ( κ ) 2 > 0. (93) \partial_\kappa A_\ast(\kappa)=\frac{M_4(\kappa)M_0(\kappa)-M_2(\kappa)^2}{M_0(\kappa)^2}>0. \tag*{(93)} ∂ κ A ∗ ( κ ) = M 0 ( κ ) 2 M 4 ( κ ) M 0 ( κ ) − M 2 ( κ ) 2 > 0. (93) If T j ( κ ) T_j(\kappa) T j ( κ ) is a simple branch value with Y κ ( T j ) ≠ 0 Y_\kappa(T_j)\ne0 Y κ ( T j ) = 0 , then
d T j d κ = − ∂ κ K κ ( T j ) ∂ T K κ ( T j ) . (94) \frac{\mathrm{d}T_j}{\mathrm{d}\kappa}=-\frac{\partial_\kappa K_\kappa(T_j)}{\partial_T K_\kappa(T_j)}. \tag*{(94)} d κ d T j = − ∂ T K κ ( T j ) ∂ κ K κ ( T j ) . (94)
Proof. Differentiation under the integral gives the moment shift. From Ξ κ ( i T ) = 4 Y κ ( T ) \Xi_\kappa(i\sqrt{T})=4Y_\kappa(T) Ξ κ ( i T ) = 4 Y κ ( T ) , the chain rule gives ∂ z 2 Ξ κ ( i T ) = − 4 ( 2 Y κ ′ + 4 T Y κ ′ ′ ) \partial_z^2\Xi_\kappa(i\sqrt{T})=-4(2Y'_\kappa+4TY''_\kappa) ∂ z 2 Ξ κ ( i T ) = − 4 ( 2 Y κ ′ + 4 T Y κ ′′ ) , proving (91) ; analyticity extends it through T = 0 T=0 T = 0 . Writing L = log Y κ L=\log Y_\kappa L = log Y κ gives ∂ κ L = A κ + 2 T A κ ′ + T A κ 2 \partial_\kappa L=A_\kappa+2T A'_\kappa+T A_\kappa^2 ∂ κ L = A κ + 2 T A κ ′ + T A κ 2 . Differentiating this identity in T T T proves (92) . Equation (93) is the variance of u 2 u^2 u 2 under the strictly positive normalized density e κ u 2 Φ ( u ) / M 0 ( κ ) e^{\kappa u^2}\Phi(u)/M_0(\kappa) e κ u 2 Φ ( u ) / M 0 ( κ ) , so it is strictly positive. Finally, implicit differentiation of K κ ( T j ( κ ) ) = 0 K_\kappa(T_j(\kappa))=0 K κ ( T j ( κ )) = 0 proves (94) .
Proposition 4.2 (Local persistence of the calibrated core). There is ε > 0 \varepsilon>0 ε > 0 such that, for real ∣ κ ∣ < ε |\kappa|<\varepsilon ∣ κ ∣ < ε , the six selected branch values have simple analytic continuations, retain two real values and two nonreal conjugate pairs, and avoid the zeros of Y κ Y_\kappa Y κ . Their compact sextic model has real type ( 2 , 1 , 1 ) (2,1,1) ( 2 , 1 , 1 ) . The calibrated action cycles and periods continue locally. After decreasing ε \varepsilon ε if necessary, the same two Farey intervals and the same nine finite Humbert exclusions remain valid. This is a qualitative local statement; no numerical heat interval is asserted.
Proof. The analytic implicit-function theorem applies to each simple selected root. The roots remain distinct and the two nonreal pairs remain nonreal for sufficiently small κ \kappa κ . Reality and uniqueness keep the real roots real and preserve conjugate pairing. On the compact closure of a slightly adjusted core domain, the time-zero boundary lower bounds for ∣ K ∣ |K| ∣ K ∣ and the lower bound for ∣ Y ∣ |Y| ∣ Y ∣ persist by uniform continuity. The boundary Rouché comparison keeps the branch count equal to six. The local normalization and its cut cycles therefore transport continuously; their action integrals are continuous, as can also be seen on fixed zero-free contour representatives. The mixed branch configuration retains the real type proved in theorem 3.7. The strict Farey inequalities of § 7 and the finite Humbert nonvanishing inequalities are open conditions on the transported periods and the sextic coefficients. Their finite number gives a common sufficiently small ε \varepsilon ε .
Positive contraction of the six selected branch values
Order the transported values ( T 1 , T 2 , T 3 , T 4 , T 5 , T 6 ) = ( a , b , c 1 , c ˉ 1 , c 2 , c ˉ 2 ) (T_1,T_2,T_3,T_4,T_5,T_6)=(a,b,c_1,\bar{c}_1,c_2,\bar{c}_2) ( T 1 , T 2 , T 3 , T 4 , T 5 , T 6 ) = ( a , b , c 1 , c ˉ 1 , c 2 , c ˉ 2 ) and define their logarithmic contraction rates in the natural T T T coordinate by
c j k ( κ ) : = − Re T j ′ ( κ ) − T k ′ ( κ ) T j ( κ ) − T k ( κ ) = − ∂ κ log ∣ T j ( κ ) − T k ( κ ) ∣ . (95) c_{jk}(\kappa):=-\operatorname{Re}\frac{T'_j(\kappa)-T'_k(\kappa)}{T_j(\kappa)-T_k(\kappa)}=-\partial_\kappa\log\lvert T_j(\kappa)-T_k(\kappa)\rvert. \tag*{(95)} c j k ( κ ) := − Re T j ( κ ) − T k ( κ ) T j ′ ( κ ) − T k ′ ( κ ) = − ∂ κ log ∣ T j ( κ ) − T k ( κ )∣ . (95)
Here the prime denotes a heat-time derivative of a branch value.
Theorem 4.3 (Certified pairwise branch contraction). At heat time zero, all fifteen rates satisfy the strict directed bound
c j k ( 0 ) > 1 9 ( 1 ≤ j < k ≤ 6 ) . (96) c_{jk}(0)>\frac{1}{9}\qquad(1\leq j<k\leq6). \tag*{(96)} c j k ( 0 ) > 9 1 ( 1 ≤ j < k ≤ 6 ) . (96) Two representative enclosures are
c a b ( 0 ) ∈ ( 0.15178 , 0.15179 ) , c c 1 c 2 ( 0 ) ∈ ( 0.12007 , 0.12008 ) . (97) c_{ab}(0)\in(0.15178,0.15179),\qquad c_{c_1c_2}(0)\in(0.12007,0.12008). \tag*{(97)} c ab ( 0 ) ∈ ( 0.15178 , 0.15179 ) , c c 1 c 2 ( 0 ) ∈ ( 0.12007 , 0.12008 ) . (97) There is a sufficiently small ε > 0 \varepsilon>0 ε > 0 such that c j k ( κ ) > 1 / 9 c_{jk}(\kappa)>1/9 c j k ( κ ) > 1/9 for real ∣ κ ∣ < ε |\kappa|<\varepsilon ∣ κ ∣ < ε , and hence
∣ T j ( κ ) − T k ( κ ) ∣ ≤ e − κ / 9 ∣ T j ( 0 ) − T k ( 0 ) ∣ ( 0 ≤ κ < ε ) . (98) \lvert T_j(\kappa)-T_k(\kappa)\rvert\leq e^{-\kappa/9}\lvert T_j(0)-T_k(0)\rvert\qquad(0\leq\kappa<\varepsilon). \tag*{(98)} ∣ T j ( κ ) − T k ( κ )∣ ≤ e − κ /9 ∣ T j ( 0 ) − T k ( 0 )∣ ( 0 ≤ κ < ε ) . (98) This is status [A]; the interval ε \varepsilon ε is qualitative.
Proof. The certificate in certificates/heat/ refines the four real and upper-half-plane roots to certified disks of radius 10 − 18 10^{-18} 1 0 − 18 , checks Y 0 ≠ 0 Y_0\ne0 Y 0 = 0 there, and encloses their heat velocities using (94) with the evolving A ∗ ( κ ) A_*(\kappa) A ∗ ( κ ) . Conjugation supplies the remaining two roots and velocities. Directed evaluation of (95) excludes every gap denominator from zero and proves (96) and (97) . All fifteen comparisons pass at both 192 and 256 bits. Analytic root continuation makes the rates continuous. Their finite number and the strict initial margin give a common local interval with c j k > 1 / 9 c_{jk} > 1/9 c j k > 1/9 . Integrating (95) from 0 0 0 to κ ≥ 0 \kappa\ge0 κ ≥ 0 proves (98) . □ \square □
The six-point configuration therefore contracts with unequal rates: the real gap closes faster in relative terms than the upper connector c 1 c 2 c_1c_2 c 1 c 2 . This describes a change of shape as well as size in the natural T T T coordinate. Euclidean distances depend on that coordinate; the cross-ratio certificate below supplies the projectively invariant comparison with the matched isomonodromic direction. The contraction theorem concerns these six discriminant branch values and a local heat interval. A global heat assertion about all zeros would require additional input.
A directed obstruction for the matched D 6 D_6 D 6 trajectory
Use the cross ratio of (40) and set
λ 1 = λ ( c ˉ 1 ; a , b , c 1 ) , λ 2 = λ ( c 2 ; a , b , c 1 ) . \lambda_1 = \lambda(\bar{c}_1; a,b,c_1), \qquad\lambda_2 = \lambda(c_2; a,b,c_1). λ 1 = λ ( c ˉ 1 ; a , b , c 1 ) , λ 2 = λ ( c 2 ; a , b , c 1 ) .
Let h j h_j h j be their derivatives along the natural heat deformation at κ = 0 \kappa= 0 κ = 0 , and let d j d_j d j be their derivatives along the D 6 D_6 D 6 isomonodromic vector field at the phase point of theorem 3.12. These derivatives are invariant under a parameter-dependent projective change of spectral coordinate.
Theorem 4.4 (Certified heat–isomonodromic tangent obstruction). For the specific real D 6 D_6 D 6 realization of theorem 3.12, the tangent determinant satisfies
D = h 1 d 2 − h 2 d 1 ∈ ( 7 , 9 ) + i ( − 9 , − 7 ) , ∣ D ∣ > 10. (99) D = h_1d_2 - h_2d_1 \in(7,9) + i(-9,-7), \qquad|D| > 10. \tag*{(99)} D = h 1 d 2 − h 2 d 1 ∈ ( 7 , 9 ) + i ( − 9 , − 7 ) , ∣ D ∣ > 10. (99) Consequently the natural heat tangent is not proportional, even over C \mathbb{C} C , to this isomonodromic tangent. There is no differentiable local identification of these curve families through the stated static match, even modulo moving projective coordinates and a change t D = t D ( κ ) t_D = t_D(\kappa) t D = t D ( κ ) . The statement is status [A] and uses no zero-free hypothesis H α H_\alpha H α .
Proof. The directed program certificates/heat/ uses Arb complex Taylor arithmetic [30 ] to evaluate
Y 0 ( T ) = 1 4 ξ ( 1 2 + T ) . Y_0(T) = \frac{1}{4}\xi\left(\frac{1}{2}+\sqrt{T}\right). Y 0 ( T ) = 4 1 ξ ( 2 1 + T ) .
It refines the selected roots to disks of radius 10 − 18 10^{-18} 1 0 − 18 about the rational centers listed in its source. For each center c c c and radius r = 10 − 18 r = 10^{-18} r = 1 0 − 18 it verifies
∣ K 0 ( c ) ∣ + 1 2 sup ∣ T − c ∣ ≤ r ∣ K 0 ′ ′ ( T ) ∣ r 2 < ∣ K 0 ′ ( c ) ∣ r , |K_0(c)| + \frac{1}{2}\sup_{|T-c|\le r}|K_0^{\prime\prime}(T)|r^2 < |K_0^{\prime}(c)|r, ∣ K 0 ( c ) ∣ + 2 1 ∣ T − c ∣ ≤ r sup ∣ K 0 ′′ ( T ) ∣ r 2 < ∣ K 0 ′ ( c ) ∣ r ,
with the ratio of the left to right side less than 1 / 20000 1/20000 1/20000 . The refined disks lie inside the previously certified branch enclosures, and Y 0 Y_0 Y 0 is excluded from zero there. Rouché’s theorem identifies one simple branch root in each disk. Conjugation supplies the lower nonreal pair and the reality of the first two roots.
At time zero, (91) gives the exact jet shift
∂ κ Y κ ( j ) ∣ 0 = ( 2 + 4 j ) Y 0 ( j + 1 ) + 4 T Y 0 ( j + 2 ) . \left.\partial_\kappa Y_\kappa^{(j)}\right|_0 = (2+4j)Y_0^{(j+1)} + 4T Y_0^{(j+2)}. ∂ κ Y κ ( j ) 0 = ( 2 + 4 j ) Y 0 ( j + 1 ) + 4 T Y 0 ( j + 2 ) .
The central coefficient and its heat derivative are enclosed using M 2 n ( 0 ) = ξ ( 2 n ) ( 1 / 2 ) / 4 M_{2n}(0)=\xi^{(2n)}(1/2)/4 M 2 n ( 0 ) = ξ ( 2 n ) ( 1/2 ) /4 . Differentiation of the regular numerator then gives (94) on each refined disk. For the D 6 D_6 D 6 tangent, the program retains the complete parameter box (35) , fixes θ 0 , θ ∞ \theta_0,\theta_\infty θ 0 , θ ∞ , and differentiates N 6 N_6 N 6 along (37) . The resulting root velocities are divided by the affine scale s s s to express them in the T T T coordinate; this does not affect cross-ratio derivatives.
All derivative and cross-ratio denominators are excluded from zero before division. Substitution in the logarithmic derivative of (40) proves the rational rectangle and the strict modulus bound in (99) . The full calculation passes at both 192 and 256 bits, with separate saved logs. These two runs use the same directed library and are not independent implementations. If a local family identification existed, projective invariance and the chain rule would give h j = d j t D ′ ( 0 ) h_j=d_jt'_D(0) h j = d j t D ′ ( 0 ) for both j j j , forcing D = 0 D=0 D = 0 , a contradiction. □
For reference, the directed heat velocities are recorded in Table 3 . The component errors in the table are verified explicitly by the program, rather than inferred from agreement of decimal outputs.
Table 3. Heat branch velocities at κ = 0 \kappa=0 κ = 0 . Each displayed real or imaginary component has absolute error less than 10 − 7 10^{-7} 1 0 − 7 .
Remark 4.5 (Scope of the obstruction). The theorem concerns the natural heat normalization and the specified static Lax realization with fixed isomonodromic exponents. It does not exclude other D 6 D_6 D 6 realizations or a larger deformation in which the exponents themselves vary. Together with theorem 3.15, it distinguishes both the ordinary x x x -deformation and the natural heat deformation from the chosen isomonodromic trajectory. The fixed-point curve identity and its Kummer lift retain their geometric meaning.
An enlarged real-moduli rank test
The obstruction above concerns one isomonodromic direction. We now test a specified enlargement, allowing the two formal exponents to vary. This concerns the branch curve, not the full irregular connection or its Stokes data.
Write τ \tau τ for the Lax time in (37) , to distinguish it from heat time. Near the matched phase point let
m = ( Im λ 1 , Re λ 2 , Im λ 2 ) . m=(\operatorname{Im}\lambda_1,\operatorname{Re}\lambda_2,\operatorname{Im}\lambda_2). m = ( Im λ 1 , Re λ 2 , Im λ 2 ) .
These are three real local coordinates on the labelled real branch-moduli space. Indeed Re λ 1 = 1 / 2 \operatorname{Re}\lambda_1=1/2 Re λ 1 = 1/2 and the cross ratio for c ‾ 2 \overline{c}_2 c 2 is
λ 3 = λ 1 λ ‾ 2 λ 2 + λ 1 − 1 ; \lambda_3=\frac{\lambda_1\overline{\lambda}_2}{\lambda_2+\lambda_1-1}; λ 3 = λ 2 + λ 1 − 1 λ 1 λ 2 ;
the denominator is nonzero for the distinct labelled branch values. Let v i s o v_{\mathrm{iso}} v iso be the derivative of m m m along (37) , and let v 0 , v ∞ v_0,v_\infty v 0 , v ∞ be its partial derivatives with respect to θ 0 , θ ∞ \theta_0,\theta_\infty θ 0 , θ ∞ , holding p , q , τ p,q,\tau p , q , τ fixed.
Proposition 4.6 (Certified enlarged tangent rank). At the realization of theorem 3.12, the real matrix
A = ( v i s o v 0 v ∞ ) A=(v_{\mathrm{iso}}\ v_0\ v_\infty) A = ( v iso v 0 v ∞ ) satisfies
600 < det A < 800. (100) 600<\det A<800. \tag*{(100)} 600 < det A < 800. (100) There is therefore a unique real solution of
m ˙ h e a t = τ ˙ v i s o + θ ˙ 0 v 0 + θ ˙ ∞ v ∞ . \dot{m}_{\mathrm{heat}}=\dot{\tau}v_{\mathrm{iso}}+\dot{\theta}_0v_0+\dot{\theta}_\infty v_\infty. m ˙ heat = τ ˙ v iso + θ ˙ 0 v 0 + θ ˙ ∞ v ∞ . With derivatives taken with respect to the natural heat time κ \kappa κ , its components satisfy
0.0007 < τ ˙ < 0.0009 , − 0.084 < θ ˙ 0 < − 0.069 , 0.020 < θ ˙ ∞ < 0.027. (101) 0.0007<\dot{\tau}<0.0009,\qquad-0.084<\dot{\theta}_0<-0.069,\qquad0.020<\dot{\theta}_\infty<0.027. \tag*{(101)} 0.0007 < τ ˙ < 0.0009 , − 0.084 < θ ˙ 0 < − 0.069 , 0.020 < θ ˙ ∞ < 0.027. (101)
Proof. For a simple root z j z_j z j of (31) , the exponent velocities are
∂ θ ν z j = − ∂ θ ν N 6 ( z j ) N 6 ′ ( z j ) , ν ∈ { 0 , ∞ } . \partial_{\theta_\nu} z_j=-\frac{\partial_{\theta_\nu}N_6(z_j)}{N'_6(z_j)},\qquad\nu\in\{0,\infty\}. ∂ θ ν z j = − N 6 ′ ( z j ) ∂ θ ν N 6 ( z j ) , ν ∈ { 0 , ∞ } .
The program certificates/lax_enlarged/cert_enlarged_rank.py retains the full realization box (35) , recertifies the four independent branch disks by the Rouché bounds used above, and excludes every derivative denominator from zero. It differentiates the displayed sextic by Arb Taylor arithmetic, transforms velocities by the affine scale, and differentiates the cross ratios. The heat jet is recomputed from (91) . Direct interval expansion of the real 3 × 3 3 \times3 3 × 3 determinant proves (100) . Cramer’s rule, with this separated denominator, proves (101) . All strict rational bounds pass at 192 and 256 bits. The full phase box, rather than a numerical midpoint realization, is used in both runs.
Corollary 4.7 (Local moving-exponent curve lift). Through the specified realization, the heat-deformed genus-two branch curves admit a real-analytic local representation by scalar-normal-form D 6 D_6 D 6 curves with moving exponents, modulo real projective changes of spectral coordinate. No global continuation or preservation of monodromy or Stokes data is asserted.
Proof. For each fixed nearby exponent pair θ \theta θ , solve (37) with initial phase values ( p ∗ , q ∗ ) (p_*,q_*) ( p ∗ , q ∗ ) at the same initial time τ ∗ \tau_* τ ∗ . This gives a real-analytic three-parameter family m ( τ , θ ) m(\tau,\theta) m ( τ , θ ) , since q ∗ τ ∗ ≠ 0 q_*\tau_*\ne0 q ∗ τ ∗ = 0 . At ( τ ∗ , θ ∗ ) (\tau_*,\theta_*) ( τ ∗ , θ ∗ ) its three derivative columns are exactly v i s o , v 0 , v ∞ v_{\mathrm{iso}},v_0,v_\infty v iso , v 0 , v ∞ : the initial values p ∗ , q ∗ p_*,q_* p ∗ , q ∗ do not depend on θ \theta θ . The real inverse-function theorem and theorem 4.6 now lift the real-analytic heat-moduli path uniquely within this specified family. Equality of the labelled branch moduli gives the local real projective equivalence of the orientation doubles.
This local lift does not identify the heat deformation with (37) at fixed exponents. Along the lifted path the exponents have nonzero derivatives, so the preceding obstruction remains valid. Nor does matching the branch curve identify its full scalar potential, choose a flat heat-time connection, or determine Stokes evolution. Those are additional deformation problems beyond this rank test.
A conditional zero strip and uniform phase positivity
For 1 / 2 < α ≤ 1 1/2 < \alpha\le1 1/2 < α ≤ 1 , let H α H_\alpha H α mean that every nontrivial zeta zero has real part at most α \alpha α , with boundary zeros permitted. The functional equation gives
H α ⟹ ∣ Im z ∣ ≤ α − 1 2 for every zero of Ξ . (102) H_\alpha\Longrightarrow|\operatorname{Im} z| \le\alpha-\frac{1}{2}\quad\text{for every zero of }\Xi. \tag*{(102)} H α ⟹ ∣ Im z ∣ ≤ α − 2 1 for every zero of Ξ. (102)
The 7/8 result reported in the QRH release and its formalization scope [31 , 32 ] is retained here as the explicit external input H 7 / 8 H_{7/8} H 7/8 . The present computational supplement does not replay a Lean build or audit that release’s full analytic argument.
Proposition 4.8 (Zero-strip phase domain). If all zeros of a real heat deformation Ξ κ \Xi_\kappa Ξ κ lie in ∣ Im z ∣ ≤ h |\operatorname{Im} z|\le h ∣ Im z ∣ ≤ h , then for every real x x x and η > h \eta>h η > h ,
− Im Ξ κ ′ Ξ κ ( x + i η ) = d d s arg Ξ κ ( s − i η ) ∣ s = x > 0. (103) -\operatorname{Im}\frac{\Xi'_\kappa}{\Xi_\kappa}(x+\mathrm{i}\eta)=\left.\frac{\mathrm{d}}{\mathrm{d}s}\arg\Xi_\kappa(s-\mathrm{i}\eta)\right|_{s=x}>0. \tag*{(103)} − Im Ξ κ Ξ κ ′ ( x + i η ) = d s d arg Ξ κ ( s − i η ) s = x > 0. (103) In particular, at heat time zero, H α H_\alpha H α supplies this uniform domain for η > α − 1 / 2 \eta>\alpha-1/2 η > α − 1/2 ; H 7 / 8 H_{7/8} H 7/8 gives η > 3 / 8 \eta>3/8 η > 3/8 .
Proof. The real even canonical-product representation of the heat deformation [29 ] permits conjugate-pair grouping of its logarithmic derivative. A pair a ± i b a\pm\mathrm{i}b a ± i b , 0 ≤ b ≤ h 0\le b\le h 0 ≤ b ≤ h , contributes
η − b ( x − a ) 2 + ( η − b ) 2 + η + b ( x − a ) 2 + ( η + b ) 2 > 0. \frac{\eta-b}{(x-a)^2+(\eta-b)^2}+\frac{\eta+b}{(x-a)^2+(\eta+b)^2}>0. ( x − a ) 2 + ( η − b ) 2 η − b + ( x − a ) 2 + ( η + b ) 2 η + b > 0.
A real zero contributes its single positive summand. The imaginary parts converge after pairing, and any real constant in the logarithmic derivative has zero imaginary part. The even order-one function is nonconstant, so its canonical product contains a zero and at least one summand is present. This proves strict positivity. The phase identity is the conjugation calculation of theorem 2.1, which also applies to Ξ κ \Xi_\kappa Ξ κ . □
At x = 0 x=0 x = 0 , positivity for every real κ \kappa κ and T ≥ 0 T \ge0 T ≥ 0 already follows from the positive theta integral: Y κ ( T ) > 0 Y_\kappa(T)>0 Y κ ( T ) > 0 and Y κ ′ ( T ) > 0 Y_\kappa'(T)>0 Y κ ′ ( T ) > 0 . The role of H α H_\alpha H α is to extend the phase statement uniformly over all real x x x above a narrower zero strip. It supplies no irrationality statement for the action ratios and no all-discriminant Humbert exclusion. In particular, the local heat obstruction above and the finite-height arithmetic theorems remain independent of this external input.
Period normal form and the zero-action lattice
The canonical Vinnikov basis used in the Humbert discussion is convenient for normalized holomorphic periods, but it is not the basis in which the directed action integrals are evaluated. The latter use the natural three-cut symplectic basis ( α 1 , α 2 , β 1 , β 2 ) (\alpha_1,\alpha_2,\beta_1,\beta_2) ( α 1 , α 2 , β 1 , β 2 ) described in the next section. With the connector winding fixed there, direct conjugation of the oriented cut cycles gives
τ ∗ α = α , τ ∗ β = H c u t α − β , H c u t = ( 1 1 1 0 ) . (104) \tau_*\alpha=\alpha,\qquad\tau_*\beta=H_{\mathrm{cut}}\alpha-\beta,\qquad H_{\mathrm{cut}}=\begin{pmatrix}1&1\\1&0\end{pmatrix}. \tag*{(104)} τ ∗ α = α , τ ∗ β = H cut α − β , H cut = ( 1 1 1 0 ) . (104)
This matrix is an integral representative of the same nondividing ( g , k , a ) = ( 2 , 1 , 1 ) (g,k,a)=(2,1,1) ( g , k , a ) = ( 2 , 1 , 1 ) real type as the canonical Vinnikov matrix I 2 I_2 I 2 ; the two bases are related by an integral symplectic change of basis; the explicit U U U -congruence and symmetric shear are displayed in the proof of theorem 3.35. In particular, there is no conflict with (79) or (80) .
Define the calibrated core cycle lattice
Γ c o r e ≔ Z α 1 ⊕ Z α 2 ⊕ Z β 1 ⊕ Z β 2 . (105) \Gamma_{\mathrm{core}}\coloneqq\mathbb{Z}\alpha_1\oplus\mathbb{Z}\alpha_2\oplus\mathbb{Z}\beta_1\oplus\mathbb{Z}\beta_2. \tag*{(105)} Γ core : = Z α 1 ⊕ Z α 2 ⊕ Z β 1 ⊕ Z β 2 . (105)
The cycles are contained in the normalized cover over Ω 2 \Omega_2 Ω 2 and their images under the core identification form a symplectic basis of H 1 ( X 2 , Z ) H_1(X_2,\mathbb{Z}) H 1 ( X 2 , Z ) . In accordance with theorem 3.6, define the action homomorphism only on these calibrated representatives,
A c o r e ( γ ) = ∮ γ ϖ 0 , γ ∈ Γ c o r e . \mathcal{A}_{\mathrm{core}}(\gamma)=\oint_\gamma\varpi_0,\qquad\gamma\in\Gamma_{\mathrm{core}}. A core ( γ ) = ∮ γ ϖ 0 , γ ∈ Γ core .
and put
u j = A c o r e ( α j ) , v j = A c o r e ( β j ) . (106) u_j=\mathcal{A}_{\mathrm{core}}(\alpha_j),\qquad v_j=\mathcal{A}_{\mathrm{core}}(\beta_j). \tag*{(106)} u j = A core ( α j ) , v j = A core ( β j ) . (106)
Proposition 5.1 (Real normal form in the cut gauge). There exist r 1 , r 2 ∈ R r_1,r_2\in\mathbb{R} r 1 , r 2 ∈ R such that, with u = ( u 1 , u 2 ) T u=(u_1,u_2)^{\mathsf T} u = ( u 1 , u 2 ) T and r = ( r 1 , r 2 ) T r=(r_1,r_2)^{\mathsf T} r = ( r 1 , r 2 ) T ,
u ∈ R 2 , v = 1 2 H c u t u + i r . (107) u\in\mathbb{R}^2,\qquad v=\frac{1}{2}H_{\mathrm{cut}}u+ir. \tag*{(107)} u ∈ R 2 , v = 2 1 H cut u + i r . (107) Equivalently,
v 1 = u 1 + u 2 2 + i r 1 , v 2 = u 1 2 + i r 2 . v_1=\frac{u_1+u_2}{2}+ir_1,\qquad v_2=\frac{u_1}{2}+ir_2. v 1 = 2 u 1 + u 2 + i r 1 , v 2 = 2 u 1 + i r 2 .
Proof. Since τ ∗ ϖ 0 = ϖ 0 ‾ \tau^*\varpi_0=\overline{\varpi_0} τ ∗ ϖ 0 = ϖ 0 , (104) gives
u ‾ = u , v ‾ = H c u t u − v . \overline{u}=u,\qquad\overline{v}=H_{\mathrm{cut}}u-v. u = u , v = H cut u − v .
Thus u u u is real and 2 Re v = H c u t u 2\operatorname{Re}v=H_{\mathrm{cut}}u 2 Re v = H cut u . □
Put
Λ u = { a ∈ Z 2 : a ⋅ u = 0 } , Λ r = { b ∈ Z 2 : b ⋅ r = 0 } . (108) \Lambda_u = \{a \in\mathbb{Z}^2 : a \cdot u = 0\}, \qquad\Lambda_r = \{b \in\mathbb{Z}^2 : b \cdot r = 0\}. \tag*{(108)} Λ u = { a ∈ Z 2 : a ⋅ u = 0 } , Λ r = { b ∈ Z 2 : b ⋅ r = 0 } . (108)
The calibrated zero-action lattice is
Λ 2 = ker A c o r e ⊂ Γ c o r e . \Lambda_2 = \ker A_{\mathrm{core}} \subset\Gamma_{\mathrm{core}}. Λ 2 = ker A core ⊂ Γ core .
Theorem 5.2 (Kernel factorization). In the three-cut basis,
Λ 2 = { ( m , n ) ∈ Z 2 ⊕ Z 2 : ( 2 m + H c u t n ) ⋅ u = 0 , n ⋅ r = 0 } . (109) \Lambda_2 = \left\{(m,n) \in\mathbb{Z}^2 \oplus\mathbb{Z}^2 : (2m + H_{\mathrm{cut}}n) \cdot u = 0,\quad n \cdot r = 0\right\}. \tag*{(109)} Λ 2 = { ( m , n ) ∈ Z 2 ⊕ Z 2 : ( 2 m + H cut n ) ⋅ u = 0 , n ⋅ r = 0 } . (109)
Consequently
Λ 2 = 0 ⟺ Λ u = 0 and Λ r = 0. (110) \Lambda_2 = 0 \quad\Longleftrightarrow\quad\Lambda_u = 0 \text{ and } \Lambda_r = 0. \tag*{(110)} Λ 2 = 0 ⟺ Λ u = 0 and Λ r = 0. (110)
Proof. For
γ = m 1 α 1 + m 2 α 2 + n 1 β 1 + n 2 β 2 , \gamma= m_1\alpha_1 + m_2\alpha_2 + n_1\beta_1 + n_2\beta_2, γ = m 1 α 1 + m 2 α 2 + n 1 β 1 + n 2 β 2 ,
theorem 5.1 gives
∮ γ ϖ 0 = ( m + 1 2 H c u t n ) ⋅ u + i n ⋅ r , \oint_\gamma\varpi_0 = \left(m + \frac{1}{2}H_{\mathrm{cut}}n\right) \cdot u + i\,n \cdot r, ∮ γ ϖ 0 = ( m + 2 1 H cut n ) ⋅ u + i n ⋅ r ,
because H c u t H_{\mathrm{cut}} H cut is symmetric. Vanishing of real and imaginary parts is exactly (109) . If both real relation lattices vanish, then n = 0 n = 0 n = 0 and then m = 0 m = 0 m = 0 . Conversely, 0 ≠ a ∈ Λ u 0 \ne a \in\Lambda_u 0 = a ∈ Λ u gives ( a , 0 ) ∈ Λ 2 (a,0) \in\Lambda_2 ( a , 0 ) ∈ Λ 2 , while 0 ≠ b ∈ Λ r 0 \ne b \in\Lambda_r 0 = b ∈ Λ r gives
( m , n ) = ( − H c u t b , 2 b ) ∈ Λ 2 . (m,n) = (-H_{\mathrm{cut}}b,2b) \in\Lambda_2. ( m , n ) = ( − H cut b , 2 b ) ∈ Λ 2 .
□
Corollary 5.3 (Two-irrationality criterion). If u 1 u 2 r 1 r 2 ≠ 0 u_1u_2r_1r_2 \ne0 u 1 u 2 r 1 r 2 = 0 , then
Λ 2 = 0 ⟺ u 1 u 2 ∉ Q and r 1 r 2 ∉ Q . \Lambda_2 = 0 \quad\Longleftrightarrow\quad\frac{u_1}{u_2} \notin\mathbb{Q} \text{ and } \frac{r_1}{r_2} \notin\mathbb{Q}. Λ 2 = 0 ⟺ u 2 u 1 ∈ / Q and r 2 r 1 ∈ / Q .
Corollary 5.4 (Explicit resonant cycles). If u 1 / u 2 = p / q ∈ Q u_1/u_2 = p/q \in\mathbb{Q} u 1 / u 2 = p / q ∈ Q in lowest terms, then
q α 1 − p α 2 ∈ Λ 2 ∖ { 0 } . q\alpha_1 - p\alpha_2 \in\Lambda_2 \setminus\{0\}. q α 1 − p α 2 ∈ Λ 2 ∖ { 0 } . If r 1 / r 2 = p / q ∈ Q r_1/r_2 = p/q \in\mathbb{Q} r 1 / r 2 = p / q ∈ Q in lowest terms, then
( p − q ) α 1 − q α 2 + 2 q β 1 − 2 p β 2 ∈ Λ 2 ∖ { 0 } . (p-q)\alpha_1 - q\alpha_2 + 2q\beta_1 - 2p\beta_2 \in\Lambda_2 \setminus\{0\}. ( p − q ) α 1 − q α 2 + 2 q β 1 − 2 p β 2 ∈ Λ 2 ∖ { 0 } . Both displayed cycles are primitive.
Proof. The first statement is immediate. For the second take b = ( q , − p ) b = (q,-p) b = ( q , − p ) , so b ⋅ r = 0 b \cdot r = 0 b ⋅ r = 0 , and use ( m , n ) = ( − H c u t b , 2 b ) (m,n) = (-H_{\mathrm{cut}}b,2b) ( m , n ) = ( − H cut b , 2 b ) from the proof of theorem 5.2. Since H c u t b = ( q − p , q ) H_{\mathrm{cut}}b = (q-p,q) H cut b = ( q − p , q ) , the displayed cycle follows. Its coefficients have greatest common divisor one because gcd ( p , q ) = 1 \gcd(p,q) = 1 g cd( p , q ) = 1 .
□
This gives a precise answer to what the failure of nonresonance would mean. It would not invalidate the spectral curve; it would select an explicit primitive calibrated core-cycle direction on which the additive action closes. In the wider monodromy framework, that direction is exactly where a residual multiplicative character must then be evaluated.
Calibrated action coordinates
Choose three disjoint cuts joining the branch pairs
Γ 0 : a ↔ b , Γ 1 : c 1 ↔ c ‾ 1 , Γ 2 : c 2 ↔ c ‾ 2 . \Gamma_0:a\leftrightarrow b,\qquad\Gamma_1:c_1\leftrightarrow\overline{c}_1,\qquad\Gamma_2:c_2\leftrightarrow\overline{c}_2. Γ 0 : a ↔ b , Γ 1 : c 1 ↔ c 1 , Γ 2 : c 2 ↔ c 2 .
Let α 1 , α 2 \alpha_1,\alpha_2 α 1 , α 2 be the positive circuits around Γ 0 , Γ 1 \Gamma_0,\Gamma_1 Γ 0 , Γ 1 , and let β 1 , β 2 \beta_1,\beta_2 β 1 , β 2 be connecting cycles from the first two cuts to the reference cut Γ 2 \Gamma_2 Γ 2 . The connector winding is fixed in the conjugation-symmetric gauge so that this cut basis realizes (104) .
Put
y + ( T ) = K ( T ) Y ( T ) 3 (111) y_+(T)=\frac{\sqrt{K(T)}}{Y(T)^3} \tag*{(111)} y + ( T ) = Y ( T ) 3 K ( T ) (111)
with square root continued on the selected sheet of the cut complement. Let γ 0 \gamma_0 γ 0 and γ 1 \gamma_1 γ 1 denote selected positive sides of Γ 0 \Gamma_0 Γ 0 and Γ 1 \Gamma_1 Γ 1 , and let ℓ 1 , ℓ 2 \ell_1,\ell_2 ℓ 1 , ℓ 2 be the calibrated connecting arcs.
Proposition 6.1 (Explicit theta-moment periods). The real period coordinates are
u 1 = 2 ∫ γ 0 y + ( T ) d T , u 2 = 2 ∫ γ 1 y + ( T ) d T , (112) u_1=2\int_{\gamma_0}y_+(T)\,\mathrm{d}T,\qquad u_2=2\int_{\gamma_1}y_+(T)\,\mathrm{d}T, \tag*{(112)} u 1 = 2 ∫ γ 0 y + ( T ) d T , u 2 = 2 ∫ γ 1 y + ( T ) d T , (112) r j = 2 Im ∫ ℓ j y + ( T ) d T , j = 1 , 2. (113) r_j=2\operatorname{Im}\int_{\ell_j}y_+(T)\,\mathrm{d}T,\qquad j=1,2. \tag*{(113)} r j = 2 Im ∫ ℓ j y + ( T ) d T , j = 1 , 2. (113) For later arithmetic aggregation set simply
d : = r 1 − r 2 . (114) d:=r_1-r_2. \tag*{(114)} d := r 1 − r 2 . (114) Proof. An α j \alpha_j α j -cycle goes once along the chosen positive side of its cut and returns along the opposite sheet. Sheet exchange changes the sign of y + y_+ y + while reversal of orientation changes the sign of d T \mathrm{d}T d T , so the two contributions agree and give the factor 2 in (112) . The calibrated β j \beta_j β j -cycle is obtained from the connector ℓ j \ell_j ℓ j and its conjugate return path, together with the cut-side pieces fixed by the chosen winding. Using τ ∗ ϖ 0 = ϖ ‾ 0 \tau^*\varpi_0=\overline{\varpi}_0 τ ∗ ϖ 0 = ϖ 0 and the homological action (104) , those cut-side pieces account exactly for 1 2 ( H c u t u ) j \frac{1}{2}(H_{\mathrm{cut}}u)_j 2 1 ( H cut u ) j , while the connector and conjugate connector give 2 i Im ∫ ℓ j y + ( T ) d T 2i\operatorname{Im}\int_{\ell_j}y_+(T)\,\mathrm{d}T 2 i Im ∫ ℓ j y + ( T ) d T . Comparing with theorem 5.1 proves (113) . The definition of d d d is then immediate.
The formulas make all four real quantities functions of the common theta-moment data. This shared dependence is important computationally: independent interval boxes discard correlations, while a future affine/Taylor-model treatment of a linear combination can retain them.
Exact heat variation of the calibrated actions
Transport the cut cycles locally as in theorem 4.2, and set ϖ κ = Δ κ ( T ) d T \varpi_\kappa=\sqrt{\Delta_\kappa(T)}\,\mathrm{d}T ϖ κ = Δ κ ( T ) d T on the continued sheet. In this subsection a dot means ∂ κ ∣ κ = 0 \partial_\kappa|_{\kappa=0} ∂ κ ∣ κ = 0 at fixed T T T , with the central normalization A ∗ ( κ ) A_*(\kappa) A ∗ ( κ ) updated. Thus
ϖ ˙ = Δ ˙ 2 Δ 0 d T , Δ ˙ = K ˙ Y 6 − 6 K Y ˙ Y 7 , Y ˙ = 2 Y ′ + 4 T Y ′ ′ . (115) \dot{\varpi}=\frac{\dot{\Delta}}{2\sqrt{\Delta_0}}\,\mathrm{d}T,\qquad\dot{\Delta}=\frac{\dot{K}}{Y^6}-\frac{6K\dot{Y}}{Y^7},\qquad\dot{Y}=2Y'+4TY''. \tag*{(115)} ϖ ˙ = 2 Δ 0 Δ ˙ d T , Δ ˙ = Y 6 K ˙ − Y 7 6 K Y ˙ , Y ˙ = 2 Y ′ + 4 T Y ′′ . (115)
Proposition 6.2 (Differentiation of transported actions). The calibrated action coordinates are real analytic in real heat time near zero. For a locally transported calibrated cycle γ \gamma γ ,
d d κ ∫ γ ( κ ) ϖ κ ∣ 0 = ∫ γ ( 0 ) Δ ˙ ( T ) 2 Δ 0 ( T ) d T . (116) \left.\frac{\mathrm{d}}{\mathrm{d}\kappa}\int_{\gamma(\kappa)}\varpi_\kappa\right|_0=\int_{\gamma(0)}\frac{\dot{\Delta}(T)}{2\sqrt{\Delta_0(T)}}\,\mathrm{d}T. \tag*{(116)} d κ d ∫ γ ( κ ) ϖ κ 0 = ∫ γ ( 0 ) 2 Δ 0 ( T ) Δ ˙ ( T ) d T . (116) The corresponding open-cut formulas are differentiated with the same fixed sheet and orientation. Moving simple branch endpoints contribute no boundary term, and their inverse-square-root singularities in ϖ ˙ \dot{\varpi} ϖ ˙ are integrable.
Proof. On fixed representatives away from the branch values, the continued square root is analytic in the heat parameter and differentiation under the integral proves (116) . For a parametrized moving path T ( κ , s ) T(\kappa,s) T ( κ , s ) , the contribution from its motion is the total derivative ∂ s ( Δ κ ∂ κ T ) \partial_s(\sqrt{\Delta_\kappa}\,\partial_\kappa T) ∂ s ( Δ κ ∂ κ T ) ; it cancels on a closed cycle. At a simple branch endpoint T j ( κ ) T_j(\kappa) T j ( κ ) , the local factorization
Δ κ ( T ) = ( T − T j ( κ ) ) g κ ( T ) , g κ ( T j ( κ ) ) ≠ 0 , \Delta_\kappa(T)=(T-T_j(\kappa))g_\kappa(T),\qquad g_\kappa(T_j(\kappa))\ne0, Δ κ ( T ) = ( T − T j ( κ )) g κ ( T ) , g κ ( T j ( κ )) = 0 ,
shows that Δ κ \sqrt{\Delta_\kappa} Δ κ vanishes there and that its fixed-T T T heat derivative is O ( ( T − T j ) − 1 / 2 ) O((T-T_j)^{-1/2}) O (( T − T j ) − 1/2 ) . Hence the open-cut endpoint term vanishes and the differentiated integrand is integrable. In the local coordinate T − T j ( κ ) = w 2 T-T_j(\kappa)=w^2 T − T j ( κ ) = w 2 , the action differential is analytic in ( κ , w ) (\kappa,w) ( κ , w ) , proving analytic dependence of the actions. Reality of the transported differential and the unchanged conjugation action give real analyticity of u 1 , u 2 , d , r 2 u_1,u_2,d,r_2 u 1 , u 2 , d , r 2 . □
Write
ρ u ( κ ) = u 1 ( κ ) u 2 ( κ ) , ρ r ( κ ) = − r 1 ( κ ) r 2 ( κ ) = − 1 − d ( κ ) r 2 ( κ ) . (117) \rho_u(\kappa)=\frac{u_1(\kappa)}{u_2(\kappa)},\qquad\rho_r(\kappa)=-\frac{r_1(\kappa)}{r_2(\kappa)}=-1-\frac{d(\kappa)}{r_2(\kappa)}. \tag*{(117)} ρ u ( κ ) = u 2 ( κ ) u 1 ( κ ) , ρ r ( κ ) = − r 2 ( κ ) r 1 ( κ ) = − 1 − r 2 ( κ ) d ( κ ) . (117)
The exact quotient derivatives are
ρ ˙ u = u 2 u ˙ 1 − u 1 u ˙ 2 u 2 2 , ρ ˙ r = d r ˙ 2 − r 2 d ˙ r 2 2 . (118) \dot{\rho}_u=\frac{u_2\dot{u}_1-u_1\dot{u}_2}{u_2^2},\qquad\dot{\rho}_r=\frac{d\dot{r}_2-r_2\dot{d}}{r_2^2}. \tag*{(118)} ρ ˙ u = u 2 2 u 2 u ˙ 1 − u 1 u ˙ 2 , ρ ˙ r = r 2 2 d r ˙ 2 − r 2 d ˙ . (118)
These are ratios of the calibrated action integrals of ϖ κ \varpi_\kappa ϖ κ . The normalized holomorphic Jacobian period matrix of theorem 3.35 is a separate object.
Theorem 6.3 (Certified positive motion of both action ratios). At heat time zero, the derivatives satisfy the strict rational enclosures in Table 4 . In particular,
ρ ˙ u ∈ ( 0.0033792688 , 0.0033792692 ) , ρ ˙ r ∈ ( 0.0021198 , 0.0021200 ) . (119) \dot{\rho}_u\in(0.0033792688,0.0033792692),\qquad\dot{\rho}_r\in(0.0021198,0.0021200). \tag*{(119)} ρ ˙ u ∈ ( 0.0033792688 , 0.0033792692 ) , ρ ˙ r ∈ ( 0.0021198 , 0.0021200 ) . (119) Both ratios are strictly increasing on a sufficiently small real heat interval about zero. The statement is status [A] and uses no QRH input.
Table 4. Strict directed heat-derivative enclosures at κ = 0 \kappa=0 κ = 0 . Here d = r 1 − r 2 d=r_1-r_2 d = r 1 − r 2 and t P o l y m a t h = 4 κ t_{\mathrm{Polymath}}=4\kappa t Polymath = 4 κ .
Proof. The certificate in certificates/period_heat/ and its accompanying Taylor engine integrate (115) on the original calibrated contours: the real segment a a a to b b b , the straight cut c 1 c_1 c 1 to c ˉ 1 \bar{c}_1 c ˉ 1 , the connector 10 10 10 to c 1 c_1 c 1 for d d d , and the upper connector c 1 c_1 c 1 to c 2 c_2 c 2 for r 2 r_2 r 2 . The real or imaginary projection and its sign are fixed by the original gauge and transported unchanged. The new base period boxes lie strictly inside all four retained enclosures (128) –(129) ; none of the sign choices meets zero.
The engine encloses the even central coefficients of ξ \xi ξ at 2048 bits. It uses a degree-85 polynomial for Y Y Y and degree-80 polynomials for the entire functions E E E and E ˙ \dot{E} E ˙ of theorem 2.3. Their removable quotients by T T T are performed by coefficient shifting, using the exact identities for A ∗ A_* A ∗ and A ˙ ∗ \dot{A}_* A ˙ ∗ .
Analytic bounds on ∣ T ∣ = 4096 |T| = 4096 ∣ T ∣ = 4096 control every omitted coefficient by Cauchy’s estimate. For Y Y Y and its derivatives these bounds follow from the positive theta-moment coefficients; for E E E and E ˙ \dot{E} E ˙ they follow from their regular numerators and triangle inequalities on that circle.
On each interior panel, an outer disk excludes zeros of Y Y Y and bounds Δ 0 \Delta_0 Δ 0 and Δ ˙ \dot{\Delta} Δ ˙ ; a nested inner disk excludes zeros of Δ 0 \Delta_0 Δ 0 . Taylor remainders are bounded, and the same square-root sheet is propagated by overlapping image disks that exclude the opposite sheet. Order-24 symmetric Taylor integration encloses both the base integral and its heat derivative. Path subdivisions use exact rational parameters. The endpoint cutoff is a fraction 10 − 18 10^{-18} 1 0 − 18 of each segment, and the actual endpoints lie in the refined radius-10 − 18 10^{-18} 1 0 − 18 root disks. The omitted integrals use rigorous bounds of orders ℓ 3 / 2 \ell^{3/2} ℓ 3/2 and ℓ 1 / 2 \ell^{1/2} ℓ 1/2 , respectively, as detailed in § 8.6 .
Directed aggregation in (118) , with u 2 u_2 u 2 and r 2 r_2 r 2 excluded from zero, proves every strict decimal enclosure in the table. Both the 512-bit and 768-bit replays pass these comparisons. Real analyticity and the strictly positive initial margins give local strict monotonicity of both ratios. No numerical length for this local interval is inferred from the time-zero certificate.
The geometry and the actions thus move in different senses: all fifteen branch distances shrink, while both calibrated action ratios increase. In particular, r ˙ 1 = d ˙ + r ˙ 2 < 0 \dot{r}_1 = \dot{d} + \dot{r}_2 < 0 r ˙ 1 = d ˙ + r ˙ 2 < 0 and r ˙ 2 > 0 \dot{r}_2 > 0 r ˙ 2 > 0 , so the second ratio records a small positive imbalance between the growth of ∣ r 1 ∣ |r_1| ∣ r 1 ∣ and that of r 2 r_2 r 2 . Its sign is certified after the cancellation in (118) .
Local generic nonresonance and dense resonant times
For real heat time near zero, let Λ 2 ( κ ) \Lambda_2(\kappa) Λ 2 ( κ ) be the kernel of the calibrated action homomorphism on the transported core cycles. A G δ G_\delta G δ set is a countable intersection of open sets; “full measure” below refers to Lebesgue measure in heat time.
Corollary 6.4 (Generic local heat nonresonance). There is an interval I = ( − ε , ε ) I = (-\varepsilon,\varepsilon) I = ( − ε , ε ) on which all four coordinates u 1 , u 2 , r 1 , r 2 u_1,u_2,r_1,r_2 u 1 , u 2 , r 1 , r 2 are nonzero and both ratios in (117) are strictly increasing. On this interval,
{ κ ∈ I : Λ 2 ( κ ) ≠ 0 } = ρ u − 1 ( Q ) ∪ ρ r − 1 ( Q ) . (120) \{\kappa\in I : \Lambda_2(\kappa) \ne0\} = \rho_u^{-1}(\mathbb{Q}) \cup\rho_r^{-1}(\mathbb{Q}). \tag*{(120)} { κ ∈ I : Λ 2 ( κ ) = 0 } = ρ u − 1 ( Q ) ∪ ρ r − 1 ( Q ) . (120) This resonant set is countable and dense in I I I . Its complement is a dense G δ G_\delta G δ set of full Lebesgue measure, and Λ 2 ( κ ) = 0 \Lambda_2(\kappa) = 0 Λ 2 ( κ ) = 0 at every point of that complement. No membership assertion for the specified heat time κ = 0 \kappa= 0 κ = 0 follows.
Proof. The time-zero period boxes give nonzero coordinates, and theorem 6.3 gives local strict monotonicity. Reduce ε \varepsilon ε to preserve both properties. The conjugation matrix H c u t H_{\mathrm{cut}} H cut is unchanged under local cycle transport, so theorem 5.3 applies at each real heat time in I I I and proves (120) . A continuous strictly monotone ratio maps every nonempty open subinterval of I I I onto an interval with nonempty interior. Rational values are dense in that image, and each rational value has at most one preimage. Each of the two preimage sets is therefore countable and dense. Their union has the same properties. Removing a countable set from an interval gives a dense G δ G_\delta G δ set of full Lebesgue measure.
This is an all-denominator statement for generic local heat parameters, while the arithmetic specialization at zero remains theorem 1.1. The dense resonant times are compatible with the locally persistent Farey windows: any rational values occurring inside those windows must satisfy the same large denominator barriers. The result concerns the action lattice; it supplies no irrationality certificate for the holomorphic Jacobian periods or the Picard rank.
Simple resonances and generic Diophantine bounds
Shrink the interval I I I of theorem 6.4 if necessary so that its closure lies in the analytic continuation domain, u 2 , r 2 > 0 u_2,r_2>0 u 2 , r 2 > 0 , and both ratio derivatives have positive lower bounds on I I I . This is possible by theorem 6.3 and continuity. Continue to write I = ( − ε , ε ) I=(-\varepsilon,\varepsilon) I = ( − ε , ε ) for this smaller interval. Identify the transported core lattice with the fixed integer lattice using the calibrated cycles. For
γ = m ⋅ α + n ⋅ β , m , n ∈ Z , \gamma=m\cdot\alpha+n\cdot\beta,\qquad m,n\in\mathbb{Z}, γ = m ⋅ α + n ⋅ β , m , n ∈ Z ,
write its analytic action as
A γ ( κ ) = ( m + 1 2 H c u t n ) ⋅ u ( κ ) + i n ⋅ r ( κ ) . (121) A_\gamma(\kappa)=\left(m+\frac{1}{2}H_{\mathrm{cut}}n\right)\cdot u(\kappa)+i\,n\cdot r(\kappa). \tag*{(121)} A γ ( κ ) = ( m + 2 1 H cut n ) ⋅ u ( κ ) + i n ⋅ r ( κ ) . (121)
Proposition 6.5 (At most one simple resonance for each fixed cycle). For each fixed γ ≠ 0 \gamma\ne0 γ = 0 , the equation A γ ( κ ) = 0 A_\gamma(\kappa)=0 A γ ( κ ) = 0 has at most one real solution in I I I . At any such solution κ 0 \kappa_0 κ 0 ,
A γ ′ ( κ 0 ) ≠ 0. A_\gamma'(\kappa_0)\ne0. A γ ′ ( κ 0 ) = 0. In particular, no fixed nonzero core cycle is resonant throughout an interval. The rational-ratio crossings in the two real channels are simple: for q > 0 q>0 q > 0 ,
F p , q = q u 1 − p u 2 , F p , q ′ ∣ ρ u = p / q = q u 2 ρ u ′ > 0. (122) F_{p,q}=qu_1-pu_2,\qquad F'_{p,q}\big|_{\rho_u=p/q}=qu_2\rho'_u>0. \tag*{(122)} F p , q = q u 1 − p u 2 , F p , q ′ ρ u = p / q = q u 2 ρ u ′ > 0. (122) G p , q = q r 1 + p r 2 , G p , q ′ ∣ ρ r = p / q = − q r 2 ρ r ′ < 0. (123) G_{p,q}=qr_1+pr_2,\qquad G'_{p,q}\big|_{\rho_r=p/q}=-qr_2\rho'_r<0. \tag*{(123)} G p , q = q r 1 + p r 2 , G p , q ′ ρ r = p / q = − q r 2 ρ r ′ < 0. (123)
Proof. If n ≠ 0 n\ne0 n = 0 , vanishing of the imaginary part of Eq. (121) forces n 1 ≠ 0 n_1\ne0 n 1 = 0 and
n ⋅ r = r 2 ( n 2 − n 1 ρ r ) = 0 , ρ r = n 2 n 1 . n\cdot r=r_2(n_2-n_1\rho_r)=0,\qquad\rho_r=\frac{n_2}{n_1}. n ⋅ r = r 2 ( n 2 − n 1 ρ r ) = 0 , ρ r = n 1 n 2 .
Strict monotonicity gives at most one such time. At that time,
∂ κ ( n ⋅ r ) = − n 1 r 2 ρ r ′ ≠ 0 , \partial_\kappa(n\cdot r)=-n_1r_2\rho'_r\ne0, ∂ κ ( n ⋅ r ) = − n 1 r 2 ρ r ′ = 0 ,
so the full complex action has nonzero derivative whenever it vanishes. If n = 0 n=0 n = 0 , then m ≠ 0 m\ne0 m = 0 and A γ = u 2 ( m 1 ρ u + m 2 ) A_\gamma=u_2(m_1\rho_u+m_2) A γ = u 2 ( m 1 ρ u + m 2 ) . A zero forces m 1 ≠ 0 m_1\ne0 m 1 = 0 and ρ u = − m 2 / m 1 \rho_u=-m_2/m_1 ρ u = − m 2 / m 1 , again at most once, with derivative m 1 u 2 ρ u ′ ≠ 0 m_1u_2\rho'_u\ne0 m 1 u 2 ρ u ′ = 0 . Finally, F p , q = u 2 ( q ρ u − p ) F_{p,q}=u_2(q\rho_u-p) F p , q = u 2 ( q ρ u − p ) and G p , q = r 2 ( p − q ρ r ) G_{p,q}=r_2(p-q\rho_r) G p , q = r 2 ( p − q ρ r ) give (122) –(123) . □
Thus the countable dense resonant set is assembled from simple zeros of different integer-cycle actions. Its density does not require repeated resonance of any one fixed cycle.
For γ ≠ 0 \gamma\ne0 γ = 0 , define its coefficient size in this fixed calibration by
N ( γ ) = max { ∣ m 1 ∣ , ∣ m 2 ∣ , ∣ n 1 ∣ , ∣ n 2 ∣ } ≥ 1. (124) N(\gamma)=\max\{|m_1|,|m_2|,|n_1|,|n_2|\}\ge1. \tag*{(124)} N ( γ ) = max { ∣ m 1 ∣ , ∣ m 2 ∣ , ∣ n 1 ∣ , ∣ n 2 ∣ } ≥ 1. (124)
Theorem 6.6 (Generic Diophantine separation of the calibrated actions). There is a full-Lebesgue-measure subset I D i o ⊂ I I_{\mathrm{Dio}}\subset I I Dio ⊂ I such that, for every κ ∈ I D i o \kappa\in I_{\mathrm{Dio}} κ ∈ I Dio and every δ > 0 \delta>0 δ > 0 , there are constants c κ , δ , C κ , δ > 0 c_{\kappa,\delta},C_{\kappa,\delta}>0 c κ , δ , C κ , δ > 0 with
∣ ρ j ( κ ) − p q ∣ ≥ c κ , δ q 2 + δ ( j = u , r ; p ∈ Z , q ∈ Z ≥ 1 ) (125) \left|\rho_j(\kappa)-\frac{p}{q}\right|\ge\frac{c_{\kappa,\delta}}{q^{2+\delta}}\qquad(j=u,r;\ p\in\mathbb{Z},\ q\in\mathbb{Z}_{\ge1}) \tag*{(125)} ρ j ( κ ) − q p ≥ q 2 + δ c κ , δ ( j = u , r ; p ∈ Z , q ∈ Z ≥ 1 ) (125) and
∣ A γ ( κ ) ∣ ≥ C κ , δ N ( γ ) 1 + δ ( γ ∈ Γ c o r e ∖ { 0 } ) . (126) |A_\gamma(\kappa)|\ge\frac{C_{\kappa,\delta}}{N(\gamma)^{1+\delta}}\qquad(\gamma\in\Gamma_{\mathrm{core}}\setminus\{0\}). \tag*{(126)} ∣ A γ ( κ ) ∣ ≥ N ( γ ) 1 + δ C κ , δ ( γ ∈ Γ core ∖ { 0 }) . (126)
The quantifier is almost every heat time, with constants depending on that time; no assertion that 0 ∈ I D i o 0\in I_{\mathrm{Dio}} 0 ∈ I Dio is made.
Proof. Both ratios have bounded images on I I I and derivatives bounded below by a common b > 0 b > 0 b > 0 . Fix δ > 0 \delta> 0 δ > 0 . For each j = u , r j = u,r j = u , r and q ≥ 1 q \ge1 q ≥ 1 , consider
B j , q ( δ ) = ⋃ p ∈ Z { κ ∈ I : ∣ ρ j ( κ ) − p q ∣ < q − 2 − δ } . B_{j,q}(\delta)=\bigcup_{p\in\mathbb{Z}}\left\{\kappa\in I:\left|\rho_j(\kappa)-\frac{p}{q}\right|<q^{-2-\delta}\right\}. B j , q ( δ ) = p ∈ Z ⋃ { κ ∈ I : ρ j ( κ ) − q p < q − 2 − δ } .
Only O ( q ) O(q) O ( q ) numerators meet the bounded image enlarged by one. By monotonicity and the derivative lower bound, each preimage interval has length at most 2 b − 1 q − 2 − δ 2b^{-1}q^{-2-\delta} 2 b − 1 q − 2 − δ . Therefore
meas B j , q ( δ ) = O ( q − 1 − δ ) , ∑ q ≥ 1 meas B j , q ( δ ) < ∞ . \operatorname{meas} B_{j,q}(\delta)=O(q^{-1-\delta}),\qquad\sum_{q\geq1}\operatorname{meas} B_{j,q}(\delta)<\infty. meas B j , q ( δ ) = O ( q − 1 − δ ) , q ≥ 1 ∑ meas B j , q ( δ ) < ∞.
The measure of the union over q ≥ Q q \geq Q q ≥ Q tends to zero as Q → ∞ Q \to\infty Q → ∞ . Consequently the limsup set has measure zero: almost every heat time belongs to only finitely many of these approximation sets. Apply this argument to both ratios and the countable exponents δ = 1 / ℓ \delta= 1/\ell δ = 1/ ℓ , ℓ ≥ 1 \ell\geq1 ℓ ≥ 1 , and also remove the countable resonant set of theorem 6.4. The resulting set I D i o I_{\mathrm{Dio}} I Dio has full measure. For any κ \kappa κ in this set and any δ > 0 \delta> 0 δ > 0 , choose 1 / ℓ < δ 1/\ell< \delta 1/ ℓ < δ . The estimate with exponent 2 + 1 / ℓ 2 + 1/\ell 2 + 1/ ℓ gives (125) for all large q q q with constant one. For the finitely many remaining denominators, irrationality makes the distance to the nearest p / q p/q p / q strictly positive. Reducing the constant, and taking the smaller constant for the two ratios, proves the bound for all p , q p,q p , q .
For the action estimate, if n 1 ≠ 0 n_1\ne0 n 1 = 0 , the imaginary part gives
∣ A γ ∣ ≥ ∣ n ⋅ r ∣ = r 2 ∣ n 2 − n 1 ρ r ∣ ≥ r 2 c κ , δ ∣ n 1 ∣ − 1 − δ . |\mathcal{A}_\gamma|\geq|n\cdot r|=r_2|n_2-n_1\rho_r|\geq r_2c_{\kappa,\delta}|n_1|^{-1-\delta}. ∣ A γ ∣ ≥ ∣ n ⋅ r ∣ = r 2 ∣ n 2 − n 1 ρ r ∣ ≥ r 2 c κ , δ ∣ n 1 ∣ − 1 − δ .
If n 1 = 0 n_1=0 n 1 = 0 but n 2 ≠ 0 n_2\ne0 n 2 = 0 , it instead gives ∣ A γ ∣ ≥ r 2 ∣ n 2 ∣ ≥ r 2 |\mathcal{A}_\gamma|\geq r_2|n_2|\geq r_2 ∣ A γ ∣ ≥ r 2 ∣ n 2 ∣ ≥ r 2 . If n = 0 n=0 n = 0 and m 1 ≠ 0 m_1\ne0 m 1 = 0 , the real part gives
∣ A γ ∣ ≥ u 2 c κ , δ ∣ m 1 ∣ − 1 − δ ; |\mathcal{A}_\gamma|\geq u_2c_{\kappa,\delta}|m_1|^{-1-\delta}; ∣ A γ ∣ ≥ u 2 c κ , δ ∣ m 1 ∣ − 1 − δ ;
the remaining case m 1 = 0 , m 2 ≠ 0 m_1=0,m_2\ne0 m 1 = 0 , m 2 = 0 gives ∣ A γ ∣ ≥ u 2 |\mathcal{A}_\gamma|\geq u_2 ∣ A γ ∣ ≥ u 2 . Since each coefficient is at most N ( γ ) N(\gamma) N ( γ ) in absolute value, all cases imply (126) with
C κ , δ = min { u 2 ( κ ) , r 2 ( κ ) } min { 1 , c κ , δ } > 0. C_{\kappa,\delta}=\min\{u_2(\kappa),r_2(\kappa)\}\min\{1,c_{\kappa,\delta}\}>0. C κ , δ = min { u 2 ( κ ) , r 2 ( κ )} min { 1 , c κ , δ } > 0.
These constants cannot have a common positive lower bound as κ \kappa κ varies through I D i o I_{\mathrm{Dio}} I Dio in any nonempty open subinterval of I I I . Such a subinterval contains a resonant time and a fixed cycle whose action vanishes there. Full measure makes I D i o I_{\mathrm{Dio}} I Dio dense, so that action tends to zero along nearby points of I D i o I_{\mathrm{Dio}} I Dio , while its coefficient size remains fixed. This also explains why generic quantitative separation cannot simply be continued to zero.
Arithmetic specialization and separation from zero reality
Theorem 6.4 proves a generic local heat version of theorem 1.1; the conjecture asks whether the specified time κ = 0 \kappa= 0 κ = 0 lies outside the exceptional set. A countable exceptional set can contain that point. Even positive analytic derivatives do not exclude this possibility: f ( κ ) = 1 + κ f(\kappa)=1+\kappa f ( κ ) = 1 + κ is irrational for almost every real κ \kappa κ , yet f ( 0 ) = 1 f(0)=1 f ( 0 ) = 1 . A fixed-point bound of the form (125) at zero would prove the conjecture, but would be stronger than mere irrationality and is not established here.
The difference from the global zero-reality question can be made rigorous using the established lower bound for the Bruijn–Newman constant.
Corollary 6.7 (Nonresonant cores with nonreal heat zeros). For almost every κ ∈ ( − ε , 0 ) \kappa\in(-\varepsilon,0) κ ∈ ( − ε , 0 ) in the local interval above,
Λ 2 ( κ ) = 0 and Ξ κ has a nonreal zero . (127) \boxed{\Lambda_2(\kappa)=0\quad\text{and}\quad\Xi_\kappa\text{ has a nonreal zero}.} \tag*{(127)} Λ 2 ( κ ) = 0 and Ξ κ has a nonreal zero . (127) At almost every such negative time the stronger action bound (126) also holds for every $\delta > 0. Thus local action nonresonance, even with generic Diophantine separation, does not imply reality of all zeros throughout this heat family.
Proof. In the Polymath normalization, H t H_t H t has only real zeros precisely when t ≥ Λ t \geq\Lambda t ≥ Λ [29 ] , and Rodgers–Tao proved Λ ≥ 0 \Lambda\geq0 Λ ≥ 0 [33 ] . If κ < 0 \kappa< 0 κ < 0 , then 4 κ < 0 ≤ Λ 4\kappa< 0 \leq\Lambda 4 κ < 0 ≤ Λ , so H 4 κ H_{4\kappa} H 4 κ has a nonreal zero. Equation (90) , Ξ κ ( z ) = 8 H 4 κ ( 2 z ) \Xi_\kappa(z) = 8H_{4\kappa}(2z) Ξ κ ( z ) = 8 H 4 κ ( 2 z ) , preserves the reality or nonreality of zeros. This gives a nonreal zero for every negative κ \kappa κ . Intersecting the negative local interval with the full-measure sets of theorems 6.4 and 6.6 proves the two conclusions. □
This is a statement about the heat-deformed family. It decides neither theorem 1.1 at κ = 0 \kappa= 0 κ = 0 nor RH, and establishes no logical equivalence between them. If RH holds, the same zero-reality criterion and Λ ≥ 0 \Lambda\geq0 Λ ≥ 0 give Λ = 0 \Lambda= 0 Λ = 0 , so every positive heat time has only real zeros. Nevertheless, theorem 6.4 still supplies a countable dense set of positive local heat times with resonant core cycles. Hence, conditional on RH, reality of all zeros also coexists with local action resonance.
The QRH input H α H_\alpha H α in theorem 4.8 has a different role: it narrows the zero strip and gives uniform phase positivity for η > α − 1 / 2 \eta> \alpha- 1/2 η > α − 1/2 , including η > 3 / 8 \eta> 3/8 η > 3/8 under H 7 / 8 H_{7/8} H 7/8 . No estimate of the form (125) at the specified arithmetic point follows from the arguments of that proposition. An implication from RH or QRH to fixed-point action irrationality would require an additional theorem connecting zero geometry to these particular calibrated integrals.
Deep directed Farey separation
Proposition 7.1 (Directed period and linear-form boxes). The calibrated action coordinates satisfy the outward enclosures
0.0282613617668264 < u 1 < 0.0282613617671469 , 0.8619540619015512 < u 2 < 0.8619540619074953 (128) 0.0282613617668264 < u_1 < 0.0282613617671469, \\
0.8619540619015512 < u_2 < 0.8619540619074953 \tag*{(128)} 0.0282613617668264 < u 1 < 0.0282613617671469 , 0.8619540619015512 < u 2 < 0.8619540619074953 (128) − 0.3351510384592472 < d < − 0.3351510384566369 , 0.1151538468901364 < r 2 < 0.1151538468923606 (129) -0.3351510384592472 < d < -0.3351510384566369, \\
0.1151538468901364 < r_2 < 0.1151538468923606 \tag*{(129)} − 0.3351510384592472 < d < − 0.3351510384566369 , 0.1151538468901364 < r 2 < 0.1151538468923606 (129) The directly aggregated linear forms satisfy
419336 u 1 − 13749 u 2 ∈ [ 6.87762 × 10 − 7 , 9.03886 × 10 − 7 ] , 616667 u 1 − 20219 u 2 ∈ [ − 1.04412 × 10 − 6 , − 7.26286 × 10 − 7 ] (130) 419336u_1 - 13749u_2 \in[6.87762 \times10^{-7}, 9.03886 \times10^{-7}], \\
616667u_1 - 20219u_2 \in[-1.04412 \times10^{-6}, -7.26286 \times10^{-7}] \tag*{(130)} 419336 u 1 − 13749 u 2 ∈ [ 6.87762 × 1 0 − 7 , 9.03886 × 1 0 − 7 ] , 616667 u 1 − 20219 u 2 ∈ [ − 1.04412 × 1 0 − 6 , − 7.26286 × 1 0 − 7 ] (130) 8678 d + 25257 r 2 ∈ [ − 8.45173 × 10 − 7 , − 7.66343 × 10 − 7 ] , 132091 d + 384446 r 2 ∈ [ 4.04956 × 10 − 7 , 1.60484 × 10 − 6 ] (131) 8678d + 25257r_2 \in[-8.45173 \times10^{-7}, -7.66343 \times10^{-7}], \\
132091d + 384446r_2 \in[4.04956 \times10^{-7}, 1.60484 \times10^{-6}] \tag*{(131)} 8678 d + 25257 r 2 ∈ [ − 8.45173 × 1 0 − 7 , − 7.66343 × 1 0 − 7 ] , 132091 d + 384446 r 2 ∈ [ 4.04956 × 1 0 − 7 , 1.60484 × 1 0 − 6 ] (131) Proof. The common theta moments are enclosed with outward-rounded symmetric Taylor quadrature and analytic theta/outer tails. The removable coordinate at T = 0 T = 0 T = 0 is integrated from the regular form K / Y 6 K/Y^6 K / Y 6 . On the remaining contour pieces, the selected square root is propagated by Taylor panels inside certified zero-free disks, with simple-root endpoint tails treated separately. The nonreal endpoints use the 10 − 9 10^{-9} 1 0 − 9 disks of theorems 3.2 and 3.3. All conversions from interval endpoints to multiprecision centers and radii are performed directly from the interval endpoint tuples, with no intermediate binary float narrowing. The deep endpoint cut is 10 − 6 10^{-6} 1 0 − 6 . The direct certificates first produce boxes strictly inside (128)–(129); the displayed decimal intervals are outward-rounded supersets and are then inserted into the integer forms by exact decimal arithmetic. The supplied period engine in certificates/period_heat/ independently checks containment in the displayed calibration boxes, and the uniform engine in certificates/finite_heat/ verifies the Farey inequalities throughout the stated heat interval.
Theorem 7.2 (Certified Farey windows). The period ratios satisfy
13749 419336 < u 1 u 2 < 20219 616667 (132) \frac{13749}{419336} < \frac{u_1}{u_2} < \frac{20219}{616667} \tag*{(132)} 419336 13749 < u 2 u 1 < 616667 20219 (132)
and
16579 8678 < − r 1 r 2 < 252355 132091 . (133) \frac{16579}{8678}<-\frac{r_1}{r_2}<\frac{252355}{132091}. \tag*{(133)} 8678 16579 < − r 2 r 1 < 132091 252355 . (133)
Proof. Since u 2 > 0 u_2>0 u 2 > 0 , (130) is exactly equivalent to (132) . Since r 1 = d + r 2 r_1=d+r_2 r 1 = d + r 2 ,
− r 1 r 2 = − 1 − d r 2 . -\frac{r_1}{r_2}=-1-\frac{d}{r_2}. − r 2 r 1 = − 1 − r 2 d .
For positive integers p , q p,q p , q ,
− r 1 r 2 > p q ⟺ q d + ( p + q ) r 2 < 0. -\frac{r_1}{r_2}>\frac{p}{q}\quad\Longleftrightarrow\quad qd+(p+q)r_2<0. − r 2 r 1 > q p ⟺ q d + ( p + q ) r 2 < 0.
The two inequalities in (131) therefore give (133) .
Corollary 7.3 (Denominator barriers). If u 1 / u 2 = p / q u_1/u_2=p/q u 1 / u 2 = p / q is reduced with q > 0 q>0 q > 0 , then
q ≥ 1,036,003 , q\ge1{,}036{,}003, q ≥ 1 , 036 , 003 , and equality forces
p q = 33968 1036003 . \frac{p}{q}=\frac{33968}{1036003}. q p = 1036003 33968 . If − r 1 / r 2 = p / q -r_1/r_2=p/q − r 1 / r 2 = p / q is reduced with q > 0 q>0 q > 0 , then
q ≥ 140,769 , q\ge140{,}769, q ≥ 140 , 769 , and equality forces
p q = 268934 140769 . \frac{p}{q}=\frac{268934}{140769}. q p = 140769 268934 .
Proof. For the first interval,
20219 ⋅ 419336 − 13749 ⋅ 616667 = 1 , 20219\cdot419336-13749\cdot616667=1, 20219 ⋅ 419336 − 13749 ⋅ 616667 = 1 ,
so the endpoints are Farey neighbors. Every reduced fraction strictly between them has denominator at least
419336 + 616667 = 1,036,003 , 419336+616667=1{,}036{,}003, 419336 + 616667 = 1 , 036 , 003 ,
with equality only for the mediant. Likewise
252355 ⋅ 8678 − 16579 ⋅ 132091 = 1 , 252355\cdot8678-16579\cdot132091=1, 252355 ⋅ 8678 − 16579 ⋅ 132091 = 1 ,
which gives q ≥ 8678 + 132091 = 140769 q\ge8678+132091=140769 q ≥ 8678 + 132091 = 140769 and the displayed mediant.
The next two mediant forms are not separated by the present independent boxes:
− 3.56350 × 10 − 7 < 1036003 u 1 − 33968 u 2 < 1.77599 × 10 − 7 (134) -3.56350\times10^{-7}<1036003u_1-33968u_2<1.77599\times10^{-7} \tag*{(134)} − 3.56350 × 1 0 − 7 < 1036003 u 1 − 33968 u 2 < 1.77599 × 1 0 − 7 (134)
− 4.40216 × 10 − 7 < 140769 d + 409703 r 2 < 8.38496 × 10 − 7 . (135) -4.40216\times10^{-7}<140769d+409703r_2<8.38496\times10^{-7}. \tag*{(135)} − 4.40216 × 1 0 − 7 < 140769 d + 409703 r 2 < 8.38496 × 1 0 − 7 . (135)
Both intervals contain zero, marking the first unresolved mediants for these independent period boxes.
Proposition 7.4 (Finite-width limitation). No finite-width interval enclosure of either period ratio can by itself prove irrationality. It can exclude rational resonances only up to a finite denominator height.
Proof. Every nonempty open interval contains reduced rational numbers with arbitrarily large denominators. For each fixed Q Q Q , only finitely many rationals of denominator at most Q Q Q lie in a bounded interval; removing them leaves a nonempty open set containing rationals of larger denominator.
Thus the natural frontier is not simply “more digits.” One can try to decide (134) –(135) by preserving the common theta-moment dependence directly in the linear forms, but even success would move only one more finite Farey rung. A proof of theorem 1.1 requires an all-denominator mechanism.
An explicit heat interval for both Farey windows
Theorem 7.5 (Certified heat persistence). Put h = 3 × 10 − 10 h = 3 \times10^{-10} h = 3 × 1 0 − 10 and I h = ( − h , h ) I_h = (-h,h) I h = ( − h , h ) . The six selected simple branch values and their calibrated core cycles continue throughout [ − h , h ] [-h,h] [ − h , h ] , with u 1 , u 2 , r 2 > 0 u_1,u_2,r_2 > 0 u 1 , u 2 , r 2 > 0 and r 1 = d + r 2 < 0 r_1 = d + r_2 < 0 r 1 = d + r 2 < 0 . For every real ∣ κ ∣ ≤ h |\kappa| \le h ∣ κ ∣ ≤ h , both Farey windows (132) –(133) hold with the transported action periods. In particular, any rational values there have reduced denominators at least 1,036,003 and 140,769, respectively.
On this same interval, directed uniform bounds give
0.0033 < ρ μ ′ ( κ ) < 0.0035 , 0.0001 < ρ r ′ ( κ ) < 0.0045. (136) 0.0033 < \rho'_{\mu}(\kappa) < 0.0035,\qquad0.0001 < \rho'_{r}(\kappa) < 0.0045. \tag*{(136)} 0.0033 < ρ μ ′ ( κ ) < 0.0035 , 0.0001 < ρ r ′ ( κ ) < 0.0045. (136) Thus theorems 6.4 to 6.6 apply with the explicit interval I h I_h I h , and theorem 6.7 applies on ( − h , 0 ) (-h,0) ( − h , 0 ) . This statement is status [A] and uses no RH or QRH input.
Proof. The certificate in certificates/finite_heat/ uses directed Taylor models uniform in the real parameter ∣ κ ∣ ≤ 10 − 9 |\kappa| \le10^{-9} ∣ κ ∣ ≤ 1 0 − 9 . Its Rouché bounds isolate one simple root in each radius-10 − 7 10^{-7} 1 0 − 7 disk about the four refined centers in theorem 4.3; conjugation supplies the other two. The four Rouché ratios are all less than 0.419. The disks are disjoint, and Y κ Y_\kappa Y κ is excluded from zero there. Uniform zero-free panel disks along the calibrated contours continue the original square-root sheets, so the original cut calibration and conjugation matrix are transported without a change of orientation.
The engine keeps the heat parameter shared through the construction of K K K and ∂ κ K \partial_\kappa K ∂ κ K before interval evaluation. It integrates (115) on the same four contours as theorem 6.3, with fixed interior panels and moving-root endpoint bounds (186) –(187) . The uniform endpoint cutoff is a fraction 10 − 8 10^{-8} 1 0 − 8 of each segment; the root-location allowance is the full radius 10 − 7 10^{-7} 1 0 − 7 . Fresh time-zero integrations reproduce all four original calibration boxes and all retained time-zero derivative enclosures.
For each period coordinate P P P , the mean-value enclosure P ( κ ) ∈ P ( 0 ) + [ − 10 − 9 , 10 − 9 ] ∂ κ P P(\kappa) \in P(0) + [-10^{-9},10^{-9}]\partial_\kappa P P ( κ ) ∈ P ( 0 ) + [ − 1 0 − 9 , 1 0 − 9 ] ∂ κ P proves the required signs and excludes the ratio denominators from zero. Inserting these boxes and the uniform period derivatives into (118) proves (136) . For each of the four Farey forms, integration of its combined heat derivative gives an enclosure on [ − h , h ] [-h,h] [ − h , h ] . The following strict outward supersets are verified directly:
419336 u 1 − 13749 u 2 ∈ ( 10 − 7 , 2 × 10 − 6 ) , 616667 u 1 − 20219 u 2 ∈ ( − 2 × 10 − 6 , − 10 − 7 ) , 8678 d + 25257 r 2 ∈ ( − 10 − 6 , − 5 × 10 − 7 ) , 132091 d + 384446 r 2 ∈ ( 5 × 10 − 7 , 2 × 10 − 6 ) . (137) \begin{aligned}
419336u_1 - 13749u_2 &\in(10^{-7},2 \times10^{-6}),\\
616667u_1 - 20219u_2 &\in(-2 \times10^{-6},-10^{-7}),\\
8678d + 25257r_2 &\in(-10^{-6},-5 \times10^{-7}),\\
132091d + 384446r_2 &\in(5 \times10^{-7},2 \times10^{-6}). \tag*{(137)}
\end{aligned} 419336 u 1 − 13749 u 2 616667 u 1 − 20219 u 2 8678 d + 25257 r 2 132091 d + 384446 r 2 ∈ ( 1 0 − 7 , 2 × 1 0 − 6 ) , ∈ ( − 2 × 1 0 − 6 , − 1 0 − 7 ) , ∈ ( − 1 0 − 6 , − 5 × 1 0 − 7 ) , ∈ ( 5 × 1 0 − 7 , 2 × 1 0 − 6 ) . (137)
The same sign algebra and exact neighbor determinants used in theorems 7.2 and 7.3 now apply at every heat time in the closed interval. The 512-bit and 768-bit replays both check every displayed rational box. The generic conclusions follow from the already proved kernel criterion, strict monotonicity, and measure estimates on the explicit open interval.
The interval is a sufficient certified one; maximality is not asserted. It quantifies the selected core, its calibrated actions, and the two Farey windows. It does not quantify the separate persistence of the nine Humbert exclusions, or the time-zero contraction inequalities. Countably many resonant times remain dense in I h I_h I h : the result excludes their small denominators rather than excluding all resonance.
Farey gaps and heat residence times
The denominator barrier has a quantitative interpretation along a monotone heat trajectory. It measures an arithmetic gap; conversion of that gap into heat time also requires a velocity bound.
Proposition 7.6 (Farey residence time). Let a / b < c / d a/b<c/d a / b < c / d be reduced Farey neighbors with b c − a d = 1 bc-ad=1 b c − a d = 1 , and let ρ \rho ρ be a differentiable period ratio on a heat interval J J J , with 0 < m ≤ ρ ′ ≤ M 0<m\leq\rho'\leq M 0 < m ≤ ρ ′ ≤ M . If two times in J J J satisfy ρ ( κ L ) = a / b \rho(\kappa_L)=a/b ρ ( κ L ) = a / b and ρ ( κ R ) = c / d \rho(\kappa_R)=c/d ρ ( κ R ) = c / d , then
1 b d M ≤ κ R − κ L ≤ 1 b d m . (138) \frac{1}{bdM}\leq\kappa_R-\kappa_L\leq\frac{1}{bdm}. \tag*{(138)} b d M 1 ≤ κ R − κ L ≤ b d m 1 . (138) Every reduced rational value strictly between these crossings has denominator at least D = b + d D=b+d D = b + d . In particular, b d ≤ D 2 / 4 bd\leq D^2/4 b d ≤ D 2 /4 gives κ R − κ L ≥ 4 / ( D 2 M ) \kappa_R-\kappa_L\geq4/(D^2M) κ R − κ L ≥ 4/ ( D 2 M ) .
More generally, suppose 0 ∈ J 0\in J 0 ∈ J , ρ ( 0 ) ∈ ( a / b , c / d ) \rho(0)\in(a/b,c/d) ρ ( 0 ) ∈ ( a / b , c / d ) , and ∣ ρ ′ ∣ ≤ M |\rho'|\leq M ∣ ρ ′ ∣ ≤ M on [ − h , h ] ⊂ J [-h,h]\subset J [ − h , h ] ⊂ J . If
h M < min { ρ ( 0 ) − a / b , c / d − ρ ( 0 ) } , (139) hM<\min\{\rho(0)-a/b,c/d-\rho(0)\}, \tag*{(139)} h M < min { ρ ( 0 ) − a / b , c / d − ρ ( 0 )} , (139) the same Farey window holds throughout [ − h , h ] [-h,h] [ − h , h ] .
Proof. The neighbor identity gives c / d − a / b = 1 / ( b d ) c/d-a/b=1/(bd) c / d − a / b = 1/ ( b d ) . Integrating m ≤ ρ ′ ≤ M m\leq\rho'\leq M m ≤ ρ ′ ≤ M between the two crossing times proves (138) ; the denominator assertion is the same exact Farey argument as in theorem 7.3. The second statement follows from ∣ ρ ( κ ) − ρ ( 0 ) ∣ ≤ M ∣ κ ∣ |\rho(\kappa)-\rho(0)|\leq M|\kappa| ∣ ρ ( κ ) − ρ ( 0 ) ∣ ≤ M ∣ κ ∣ .
This proposition does not assert that either endpoint crossing occurs in the present continuation domain. It gives a conditional residence time when both crossings are known, and a sufficient persistence test that does not require locating them. The exact gap depends on the product b d bd b d , while the excluded denominator height depends on the sum b + d b+d b + d . A denominator cutoff alone therefore does not specify a heat clock.
Global zero strips and the Newman threshold
The scalar Newman constant concerns all zeros, and a separate global input is needed to convert heat time into a bound for that constant.
Proposition 7.7 (Zero-strip input for a Newman bound). Suppose all zeros of Ξ κ 0 \Xi_{\kappa_0} Ξ κ 0 lie in ∣ Im z ∣ ≤ h 0 |\operatorname{Im}z|\leq h_0 ∣ Im z ∣ ≤ h 0 , with h 0 ≥ 0 h_0\geq0 h 0 ≥ 0 . For every κ ≥ κ 0 \kappa\geq\kappa_0 κ ≥ κ 0 , all zeros of Ξ κ \Xi_\kappa Ξ κ lie in
∣ Im z ∣ ≤ max { h 0 2 − 2 ( κ − κ 0 ) , 0 } . (140) |\operatorname{Im}z|\leq\sqrt{\max\{h_0^2-2(\kappa-\kappa_0),0\}}. \tag*{(140)} ∣ Im z ∣ ≤ max { h 0 2 − 2 ( κ − κ 0 ) , 0 } . (140) Consequently, in the Polymath normalization,
Λ ≤ 4 κ 0 + 2 h 0 2 . (141) \Lambda\leq4\kappa_0+2h_0^2. \tag*{(141)} Λ ≤ 4 κ 0 + 2 h 0 2 . (141) Under the external hypothesis H α H_\alpha H α at κ 0 = 0 \kappa_0=0 κ 0 = 0 , this gives Λ ≤ 2 ( α − 1 2 ) 2 \Lambda\leq2(\alpha-\frac{1}{2})^2 Λ ≤ 2 ( α − 2 1 ) 2 ; for H 7 / 8 H_{7/8} H 7/8 the corresponding classical conditional ceiling is 9 / 32 9/32 9/32 .
Proof. Use Ξ κ ( z ) = 8 H 4 κ ( 2 z ) \Xi_\kappa(z)=8H_{4\kappa}(2z) Ξ κ ( z ) = 8 H 4 κ ( 2 z ) . The initial strip for H 4 κ 0 H_{4\kappa_0} H 4 κ 0 has height 2 h 0 2h_0 2 h 0 . De Bruijn’s strip theorem, in the formulation of [29 Theorem 3.2] , bounds its later height by max { 4 h 0 2 − 8 ( κ − κ 0 ) , 0 } \sqrt{\max\{4h_0^2-8(\kappa-\kappa_0),0\}} max { 4 h 0 2 − 8 ( κ − κ 0 ) , 0 } . Dividing by two proves (140) . At κ = κ 0 + h 0 2 / 2 \kappa=\kappa_0+h_0^2/2 κ = κ 0 + h 0 2 /2 , every zero is real, so the Newman threshold is at most 4 κ 0 + 2 h 0 2 4\kappa_0+2h_0^2 4 κ 0 + 2 h 0 2 . If h 0 = 0 h_0=0 h 0 = 0 , this last conclusion already follows at the initial time. Under H α H_\alpha H α , the functional equation and the asserted bound on zero real parts give h 0 = α − 1 / 2 h_0=\alpha-1/2 h 0 = α − 1/2 , proving the final statement.
This is the established strip mechanism expressed in the present normalization, not a new upper bound. The classical critical strip gives h 0 = 1 / 2 h_0 = 1/2 h 0 = 1/2 at zero and hence Λ ≤ 1 / 2 \Lambda\le1/2 Λ ≤ 1/2 . Merely advancing that same strip by positive heat does not improve the ceiling: before it collapses, the later-time substitution gives
4 κ + 2 ( 1 / 4 − 2 κ ) = 1 / 2. 4\kappa+ 2(1/4 - 2\kappa) = 1/2. 4 κ + 2 ( 1/4 − 2 κ ) = 1/2.
Corollary 7.8 (Polymath comparison). Suppose t 0 , X > 0 t_0, X > 0 t 0 , X > 0 and 0 < y 0 ≤ 1 0 < y_0 \le1 0 < y 0 ≤ 1 satisfy the three hypotheses of [29 Theorem 1.2] . Put κ 0 = t 0 / 4 \kappa_0 = t_0/4 κ 0 = t 0 /4 and h 0 = y 0 / 2 h_0 = y_0/2 h 0 = y 0 /2 . Then all zeros of Ξ κ 0 \Xi_{\kappa_0} Ξ κ 0 lie in ∣ Im z ∣ ≤ h 0 |\operatorname{Im} z| \le h_0 ∣ Im z ∣ ≤ h 0 , and the corresponding sufficient time for reality of all zeros is
κ b o u n d = κ 0 + h 0 2 2 = 1 4 ( t 0 + y 0 2 2 ) . (142) \kappa_{\mathrm{bound}} = \kappa_0 + \frac{h_0^2}{2} = \frac{1}{4}\left(t_0 + \frac{y_0^2}{2}\right). \tag*{(142)} κ bound = κ 0 + 2 h 0 2 = 4 1 ( t 0 + 2 y 0 2 ) . (142) In particular,
Λ ≤ t b o u n d : = 4 κ b o u n d = t 0 + y 0 2 2 . (143) \Lambda\le t_{\mathrm{bound}} := 4\kappa_{\mathrm{bound}} = t_0 + \frac{y_0^2}{2}. \tag*{(143)} Λ ≤ t bound := 4 κ bound = t 0 + 2 y 0 2 . (143)
Proof. The three hypotheses are initial finite-height zero exclusion for ζ \zeta ζ , a final-time zero-free region at large horizontal coordinate, and a zero-free barrier at every intermediate heat time. Their combination gives the global strip for H t 0 H_{t_0} H t 0 in [29 Proposition 3.3] ; evenness and conjugation cover both horizontal directions and both signs of the imaginary part. Equation (90) divides its half-width by two. Theorem 7.7 then gives (142) and (143) . □
For the published Polymath choice t 0 = y 0 = 1 / 5 t_0 = y_0 = 1/5 t 0 = y 0 = 1/5 , [29 Theorems 1.1–1.2] therefore gives
κ 0 = 1 20 , h 0 = 1 10 , κ b o u n d = 11 200 , t b o u n d = 11 50 = 0.22. \kappa_0 = \frac{1}{20}, \qquad h_0 = \frac{1}{10}, \qquad\kappa_{\mathrm{bound}} = \frac{11}{200}, \qquad t_{\mathrm{bound}} = \frac{11}{50} = 0.22. κ 0 = 20 1 , h 0 = 10 1 , κ bound = 200 11 , t bound = 50 11 = 0.22.
Table 5 compares this with the classical input. Both use de Bruijn’s contraction theorem. Polymath improves the input strip: at κ 0 = 1 / 20 \kappa_0 = 1/20 κ 0 = 1/20 , contraction of the classical strip alone gives h 0 2 = 3 / 20 h_0^2 = 3/20 h 0 2 = 3/20 , whereas the published certificate gives h 0 2 = 1 / 100 h_0^2 = 1/100 h 0 2 = 1/100 . The resulting ceilings are respectively 1 / 2 1/2 1/2 and 11 / 50 11/50 11/50 .
Table 5. Two global inputs to the same strip-contraction mechanism. The Polymath strip is supplied by the published external certificate.
The phase interpretation gives an equivalent analytic target for the global input.
Proposition 7.9 (Uniform phase criterion for a global strip). Fix real κ 0 \kappa_0 κ 0 and h 0 ≥ 0 h_0 \ge0 h 0 ≥ 0 . All zeros of Ξ κ 0 \Xi_{\kappa_0} Ξ κ 0 lie in ∣ Im z ∣ ≤ h 0 |\operatorname{Im} z| \le h_0 ∣ Im z ∣ ≤ h 0 if and only if, at every point with x ∈ R x \in\mathbb{R} x ∈ R , η > h 0 \eta> h_0 η > h 0 , and Ξ κ 0 ( x + i η ) ≠ 0 \Xi_{\kappa_0}(x + i\eta) \ne0 Ξ κ 0 ( x + i η ) = 0 ,
P κ 0 ( x , η ) : = − Im Ξ κ 0 ′ ( x + i η ) Ξ κ 0 ( x + i η ) ≥ 0. (144) P_{\kappa_0}(x,\eta) := -\operatorname{Im}\frac{\Xi'_{\kappa_0}(x+i\eta)}{\Xi_{\kappa_0}(x+i\eta)} \ge0. \tag*{(144)} P κ 0 ( x , η ) := − Im Ξ κ 0 ( x + i η ) Ξ κ 0 ′ ( x + i η ) ≥ 0. (144) Under either condition the inequality is strict at every such point, and Λ ≤ 4 κ 0 + 2 h 0 2 \Lambda\le4\kappa_0 + 2h_0^2 Λ ≤ 4 κ 0 + 2 h 0 2 .
Proof. The strip implies strict positivity by theorem 4.8. Conversely, suppose the sign condition holds and z ∗ = a + i b z_* = a + ib z ∗ = a + ib is a zero with b > h 0 b > h_0 b > h 0 , of multiplicity m ≥ 1 m \ge1 m ≥ 1 . Local factorization gives
Ξ κ 0 ′ ( z ) Ξ κ 0 ( z ) = m z − z ∗ + O ( 1 ) . \frac{\Xi'_{\kappa_0}(z)}{\Xi_{\kappa_0}(z)} = \frac{m}{z-z_*} + O(1). Ξ κ 0 ( z ) Ξ κ 0 ′ ( z ) = z − z ∗ m + O ( 1 ) .
For sufficiently small ε > 0 \varepsilon> 0 ε > 0 , the point z = z ∗ − i ε z = z_* - i\varepsilon z = z ∗ − i ε lies above h 0 h_0 h 0 and is not a zero, but
P κ 0 ( a , b − ε ) = − m ε + O ( 1 ) < 0 , \mathcal{P}_{\kappa_0}(a,b-\varepsilon)=-\frac{m}{\varepsilon}+O(1)<0, P κ 0 ( a , b − ε ) = − ε m + O ( 1 ) < 0 ,
a contradiction. Conjugation excludes zeros below − h 0 -h_0 − h 0 . The Newman ceiling follows from theorem 7.7.
By the phase–transverse-amplitude identity of theorem 2.1, extended to the heat family, P κ ( x , η ) \mathcal{P}_{\kappa}(x,\eta) P κ ( x , η ) is the horizontal phase velocity and equals η B κ , x ( η 2 ) \eta B_{\kappa,x}(\eta^2) η B κ , x ( η 2 ) , where
B κ , x ( T ) : = ∂ T log { Ξ κ ( x + i T ) Ξ κ ( x − i T ) ‾ } . B_{\kappa,x}(T):=\partial_T\log\{\Xi_\kappa(x+i\sqrt{T})\overline{\Xi_\kappa(x-i\sqrt{T})}\}. B κ , x ( T ) := ∂ T log { Ξ κ ( x + i T ) Ξ κ ( x − i T ) } .
Thus theorem 7.9 specifies an analytic bridge from the transverse quantity used in the spectral construction to a Newman bound: its sign must be controlled uniformly at every horizontal coordinate and every η > h 0 \eta>h_0 η > h 0 . Polymath supplies the global strip by its three-region argument; the proposition states an equivalent target in the phase language.
A Farey window constrains two calibrated integrals over a bounded core. It supplies neither the initial all-height zero strip in theorem 7.7 nor an improvement to its height. Thus the denominator–velocity relation in theorem 7.6 is a local arithmetic heat scale; no formula for Λ \Lambda Λ in terms of these denominator cutoffs follows. The certified interval ∣ κ ∣ ≤ 3 × 10 − 10 |\kappa|\leq3\times10^{-10} ∣ κ ∣ ≤ 3 × 1 0 − 10 also does not reach Polymath’s κ 0 = 1 / 20 \kappa_0=1/20 κ 0 = 1/20 . Extending the spectral/Farey program towards a Newman ceiling therefore requires an additional estimate establishing (144) uniformly, or certifying the hypotheses of theorem 7.8.
A certified exterior phase domain
The global sign condition can be formulated without division by Ξ κ \Xi_\kappa Ξ κ . Define its continuous numerator
J κ ( x , η ) : = − Im { Ξ κ ′ ( x + i η ) Ξ κ ( x + i η ) ‾ } = 1 2 ∂ η ∣ Ξ κ ( x + i η ) ∣ 2 . (145) J_\kappa(x,\eta):=-\operatorname{Im}\{\Xi'_\kappa(x+i\eta)\overline{\Xi_\kappa(x+i\eta)}\}=\frac{1}{2}\partial_\eta|\Xi_\kappa(x+i\eta)|^2. \tag*{(145)} J κ ( x , η ) := − Im { Ξ κ ′ ( x + i η ) Ξ κ ( x + i η ) } = 2 1 ∂ η ∣ Ξ κ ( x + i η ) ∣ 2 . (145)
At a regular point it equals ∣ Ξ κ ( x + i η ) ∣ 2 P κ ( x , η ) |\Xi_\kappa(x+i\eta)|^2\mathcal{P}_\kappa(x,\eta) ∣ Ξ κ ( x + i η ) ∣ 2 P κ ( x , η ) ; at a zero it is still defined and equals zero. This gives a continuous quantity for finite interval tests. The following error estimate also makes the Polymath approximation available to such tests.
Proposition 7.10 (Phase-numerator error transfer). Write H t = B t g t H_t=B_tg_t H t = B t g t on a domain where B t B_t B t is holomorphic and nonvanishing, and put b B = − Im ( B t ′ / B t ) b_B=-\operatorname{Im}(B'_t/B_t) b B = − Im ( B t ′ / B t ) . Suppose a holomorphic model ψ \psi ψ has pointwise errors ∣ g t − ψ ∣ ≤ e 0 |g_t-\psi|\leq e_0 ∣ g t − ψ ∣ ≤ e 0 and ∣ g t ′ − ψ ′ ∣ ≤ e 1 |g'_t-\psi'|\leq e_1 ∣ g t ′ − ψ ′ ∣ ≤ e 1 . Define
N ψ = b B ∣ ψ ∣ 2 − Im ( ψ ′ ψ ‾ ) , (146) N_\psi=b_B|\psi|^2-\operatorname{Im}(\psi'\overline{\psi}), \tag*{(146)} N ψ = b B ∣ ψ ∣ 2 − Im ( ψ ′ ψ ) , (146) R ψ = ∣ b B ∣ e 0 ( 2 ∣ ψ ∣ + e 0 ) + e 1 ∣ ψ ∣ + e 0 ∣ ψ ′ ∣ + e 0 e 1 . (147) R_\psi=|b_B|e_0(2|\psi|+e_0)+e_1|\psi|+e_0|\psi'|+e_0e_1. \tag*{(147)} R ψ = ∣ b B ∣ e 0 ( 2∣ ψ ∣ + e 0 ) + e 1 ∣ ψ ∣ + e 0 ∣ ψ ′ ∣ + e 0 e 1 . (147) Then
∣ − Im ( H t ′ H t ‾ ) ∣ B t ∣ 2 − N ψ ∣ ≤ R ψ . (148) \left|\frac{-\operatorname{Im}(H'_t\overline{H_t})}{|B_t|^2}-N_\psi\right|\leq R_\psi. \tag*{(148)} ∣ B t ∣ 2 − Im ( H t ′ H t ) − N ψ ≤ R ψ . (148) In particular , N ψ > R ψ N_\psi>R_\psi N ψ > R ψ proves both H t ≠ 0 H_t\neq0 H t = 0 and − Im ( H t ′ / H t ) > 0 -\operatorname{Im}(H'_t/H_t)>0 − Im ( H t ′ / H t ) > 0 at that point.
Alternatively , if g t g_t g t is holomorphic on a neighborhood of the closed disk D ‾ ( w , r ) \overline{D}(w,r) D ( w , r ) and ∣ g t − 1 ∣ ≤ ε < 1 |g_t-1|\leq\varepsilon<1 ∣ g t − 1∣ ≤ ε < 1 throughout that disk, then
− Im H t ′ H t ( w ) ≥ b B ( w ) − ε r ( 1 − ε ) . (149) -\operatorname{Im}\frac{H'_t}{H_t}(w)\geq b_B(w)-\frac{\varepsilon}{r(1-\varepsilon)}. \tag*{(149)} − Im H t H t ′ ( w ) ≥ b B ( w ) − r ( 1 − ε ) ε . (149) Proof. Differentiating H t = B t g t H_t=B_tg_t H t = B t g t gives
− Im ( H t ′ H t ‾ ) ∣ B t ∣ 2 = b B ∣ g t ∣ 2 − Im ( g t ′ g t ‾ ) . \frac{-\operatorname{Im}(H'_t\overline{H_t})}{|B_t|^2}=b_B|g_t|^2-\operatorname{Im}(g'_t\overline{g_t}). ∣ B t ∣ 2 − Im ( H t ′ H t ) = b B ∣ g t ∣ 2 − Im ( g t ′ g t ) . :::`
The squared-modulus error is at most e 0 ( 2 ∣ ψ ∣ + e 0 ) e_0(2|\psi|+e_0) e 0 ( 2∣ ψ ∣ + e 0 ) . Expanding the second product gives the three remaining terms in (147) . Strict positivity of the numerator excludes a zero, at which that numerator would vanish. For the disk estimate, Cauchy’s formula applied to g t − 1 g_t-1 g t − 1 gives ∣ g t ′ ( w ) ∣ ≤ ε / r |g_t'(w)|\leq\varepsilon/r ∣ g t ′ ( w ) ∣ ≤ ε / r , while ∣ g t ( w ) ∣ ≥ 1 − ε |g_t(w)|\geq1-\varepsilon ∣ g t ( w ) ∣ ≥ 1 − ε . Substitution in H t ′ / H t = B t ′ / B t + g t ′ / g t H_t'/H_t=B_t'/B_t+g_t'/g_t H t ′ / H t = B t ′ / B t + g t ′ / g t proves (149) .
In [29 Theorem 1.3] , g t = H t / B t g_t=H_t/B_t g t = H t / B t has an effective model f t f_t f t and explicit amplitude errors. A derivative error follows from Cauchy only when the error bound holds on the whole surrounding circle and the compared functions are holomorphic there. The cutoff defining f t f_t f t changes with the horizontal coordinate. The exterior calculation below therefore applies Cauchy to the exact function g t − 1 g_t-1 g t − 1 , whose holomorphy does not depend on that cutoff.
Theorem 7.11 (Uniform exterior phase positivity). For every 1 25 ≤ κ ≤ 1 20 \frac{1}{25}\leq\kappa\leq\frac{1}{20} 25 1 ≤ κ ≤ 20 1 , ∣ x ∣ ≥ 5 × 10 24 |x|\geq5\times10^{24} ∣ x ∣ ≥ 5 × 1 0 24 , 1 10 ≤ η ≤ 1 4 − 2 κ \frac{1}{10}\leq\eta\leq\sqrt{\frac{1}{4}-2\kappa} 10 1 ≤ η ≤ 4 1 − 2 κ , the heat function is nonzero at x + i η x+i\eta x + i η , and
P κ ( x , η ) > 38 3 . (150) \mathcal{P}_{\kappa}(x,\eta)>\frac{38}{3}. \tag*{(150)} P κ ( x , η ) > 3 38 . (150) Consequently P κ ( x , η ) > 0 \mathcal{P}_{\kappa}(x,\eta)>0 P κ ( x , η ) > 0 for every ∣ x ∣ ≥ 5 × 10 24 |x|\geq5\times10^{24} ∣ x ∣ ≥ 5 × 1 0 24 and every η ≥ 1 / 10 \eta\geq1/10 η ≥ 1/10 . The displayed quantitative lower bound is asserted only on the stated capped range of η \eta η .
Proof. We work first in the Polymath coordinates w = u + i y w=u+iy w = u + i y , t = 4 κ t=4\kappa t = 4 κ . Put
t − = 4 25 , t + = 1 5 , X = 10 25 , A = X − 1 , r = β = 1 10 , y + = 19 20 . t_-=\frac{4}{25},\qquad t_+=\frac{1}{5},\qquad X=10^{25},\qquad A=X-1,\qquad r=\beta=\frac{1}{10},\qquad y_+=\frac{19}{20}. t − = 25 4 , t + = 5 1 , X = 1 0 25 , A = X − 1 , r = β = 10 1 , y + = 20 19 .
If u ≥ X u\geq X u ≥ X and 1 / 5 ≤ y ≤ 1 − 2 t 1/5\leq y\leq\sqrt{1-2t} 1/5 ≤ y ≤ 1 − 2 t , the closed disk of radius r r r lies in ℜ w ≥ A \Re w\geq A ℜ w ≥ A and β ≤ ℑ w ≤ y + < 1 \beta\leq\Im w\leq y_+<1 β ≤ ℑ w ≤ y + < 1 : indeed 1 − 2 t − + r < 19 / 20 \sqrt{1-2t_-}+r<19/20 1 − 2 t − + r < 19/20 . The prefactor B t B_t B t is holomorphic and nonzero on a neighborhood of each such disk.
Here are explicit majorants for the exact function g t = H t / B t g_t=H_t/B_t g t = H t / B t . For a point in this expanded region, write
L = log ℜ w 4 π , L 0 = log A 4 π , δ = 10 − 20 , c = t − 4 , a = 1 + β 2 − δ , b = a − β , U ( L ) = L 2 + δ . L=\log\frac{\Re w}{4\pi},\qquad L_0=\log\frac{A}{4\pi},\qquad\delta=10^{-20},\qquad c=\frac{t_-}{4},\qquad a=\frac{1+\beta}{2}-\delta,\qquad b=a-\beta,\qquad U(L)=\frac{L}{2}+\delta. L = log 4 π ℜ w , L 0 = log 4 π A , δ = 1 0 − 20 , c = 4 t − , a = 2 1 + β − δ , b = a − β , U ( L ) = 2 L + δ .
The bounds (20)–(22) of [29 ] give ℜ s ∗ ≥ ( 1 + y ) / 2 + t L / 4 − δ \Re s_*\geq(1+y)/2+tL/4-\delta ℜ s ∗ ≥ ( 1 + y ) /2 + t L /4 − δ , ∣ κ R S ∣ ≤ δ y |\kappa_{\mathrm{RS}}|\leq\delta y ∣ κ RS ∣ ≤ δ y , and log N ≤ U ( L ) \log N\leq U(L) log N ≤ U ( L ) ; here κ R S \kappa_{\mathrm{RS}} κ RS denotes the small shift in the Riemann–Siegel model, not the heat time. The three losses are bounded by δ \delta δ already at ℜ w = A \Re w=A ℜ w = A and decrease thereafter. Also log N < L \log N<L log N < L . Thus increasing t t t from t − t_- t − decreases each absolute coefficient after the lower bound for ℜ s ∗ \Re s_* ℜ s ∗ is inserted. The combined dual coefficient decreases as y y y increases from β \beta β : its logarithmic y y y -derivative is bounded by 1 / 50 − L / 2 + ( 1 / 2 + δ ) U ( L ) < 0 1/50-L/2+(1/2+\delta)U(L)<0 1/50 − L /2 + ( 1/2 + δ ) U ( L ) < 0 .
Put ℓ = log 100 \ell=\log100 ℓ = log 100 and define
T v ( L ) = U ( L ) { e ( 1 − v − c L ) ℓ + c ℓ 2 + e ( 1 − v − c L ) U ( L ) + c U ( L ) 2 } . (151) T_v(L)=U(L)\left\{e^{(1-v-cL)\ell+c\ell^2}+e^{(1-v-cL)U(L)+cU(L)^2}\right\}. \tag*{(151)} T v ( L ) = U ( L ) { e ( 1 − v − c L ) ℓ + c ℓ 2 + e ( 1 − v − c L ) U ( L ) + c U ( L ) 2 } . (151)
γ ∗ ( L ) = exp { β ( 1 50 − L 2 + δ U ( L ) ) } . (152) \gamma_*(L)=\exp\left\{\beta\left(\frac{1}{50}-\frac{L}{2}+\delta U(L)\right)\right\}. \tag*{(152)} γ ∗ ( L ) = exp { β ( 50 1 − 2 L + δ U ( L ) ) } . (152)
The decreasing-sum integral test and the convexity argument of [29 Lemma 8.2] , with head length 100, give
∣ f t − 1 ∣ ≤ M ( L ) : = ∑ n = 2 100 e c log 2 n − ( a + c L ) log n + T a ( L ) + γ ∗ ( L ) { ∑ n = 1 100 e c log 2 n − ( b + c L ) log n + T b ( L ) } . (153) \begin{aligned}
|f_t-1|&\leq M(L):=\sum_{n=2}^{100}e^{c\log^2 n-(a+cL)\log n}+T_a(L)\\
&\quad+\gamma_*(L)\left\{\sum_{n=1}^{100}e^{c\log^2 n-(b+cL)\log n}+T_b(L)\right\}. \tag*{(153)}
\end{aligned} ∣ f t − 1∣ ≤ M ( L ) := n = 2 ∑ 100 e c l o g 2 n − ( a + c L ) l o g n + T a ( L ) + γ ∗ ( L ) { n = 1 ∑ 100 e c l o g 2 n − ( b + c L ) l o g n + T b ( L ) } . (153)
The dual factor includes N ∣ κ R S ∣ N^{|\kappa_{RS}|} N ∣ κ R S ∣ , so the same majorant also controls the sum multiplying the exponential error in (23) of [29 ] , after restoring the primary n = 1 n=1 n = 1 term. Replacing the maximum of the two endpoint values by their sum gives the displayed T v T_v T v . The endpoint with log N \log N log N is bounded by its value at U ( L ) U(L) U ( L ) : its derivative in that endpoint is positive on L / 2 − 2 e − L 0 / 2 ≤ log N ≤ U ( L ) L/2-2e^{-L_0/2}\leq\log N\leq U(L) L /2 − 2 e − L 0 /2 ≤ log N ≤ U ( L ) .
Every term in M M M decreases for L ≥ L 0 L\geq L_0 L ≥ L 0 . For the two primary endpoint tails, the logarithmic derivatives are 1 / ( 2 U ) − c ℓ 1/(2U)-c\ell 1/ ( 2 U ) − c ℓ and 1 / ( 2 U ) + ( 1 − a ) / 2 − c L / 2 1/(2U)+(1-a)/2-cL/2 1/ ( 2 U ) + ( 1 − a ) /2 − c L /2 . The dual versions replace a a a by b b b and subtract β ( 1 − δ ) / 2 \beta(1-\delta)/2 β ( 1 − δ ) /2 . These expressions are negative at L 0 L_0 L 0 and decrease thereafter; the finite-head terms also decrease. All these endpoint and monotonicity guards are strict directed comparisons in the supplementary program.
For completeness, let ν = e L 0 / 2 − 1 \nu=e^{L_0/2}-1 ν = e L 0 /2 − 1 and put
d 1 = t + 2 L 0 2 / 16 + 0.626 A − 6.66 , (154) d_1=\frac{t_+^2L_0^2/16+0.626}{A-6.66}, \tag*{(154)} d 1 = A − 6.66 t + 2 L 0 2 /16 + 0.626 , (154)
E C = exp { − 1 + β 4 L 0 − t − 16 L 0 2 + 1.24 ( 3 y + + 3 − y + ) ν − 1 / 8 + 3 L 0 2 + π 2 / 4 + 10.44 A − 12 } . (155) E_C=\exp\left\{-\frac{1+\beta}{4}L_0-\frac{t_-}{16}L_0^2+\frac{1.24(3^{y_+}+3^{-y_+})}{\nu-1/8}+\frac{3\sqrt{L_0^2+\pi^2/4+10.44}}{A-12}\right\}. \tag*{(155)} E C = exp { − 4 1 + β L 0 − 16 t − L 0 2 + ν − 1/8 1.24 ( 3 y + + 3 − y + ) + A − 12 3 L 0 2 + π 2 /4 + 10.44 } . (155)
Equations (23) –(24) of [29 ] imply throughout the expanded region
∣ g t − 1 ∣ ≤ ε ∗ : = M ( L 0 ) + ( e d 1 − 1 ) ( 1 + M ( L 0 ) ) + E C < 2 5 . (156) |g_t-1|\leq\varepsilon_*:=M(L_0)+(e^{d_1}-1)(1+M(L_0))+E_C<\frac{2}{5}. \tag*{(156)} ∣ g t − 1∣ ≤ ε ∗ := M ( L 0 ) + ( e d 1 − 1 ) ( 1 + M ( L 0 )) + E C < 5 2 . (156)
The error majorants decrease with the horizontal coordinate; the logarithmic square in (23) is bounded by L 2 L^2 L 2 because − 2 δ ≤ L − 2 log n ≤ L -2\delta\leq L-2\log n\leq L − 2 δ ≤ L − 2 log n ≤ L . The decreasing factors in (24) are evaluated at L 0 L_0 L 0 , using N ≥ e L 0 / 2 − 1 N\geq e^{L_0/2}-1 N ≥ e L 0 /2 − 1 . The strict numerical inequality in (156) is checked at both 160 and 256 bits, together with every guard used above. The enclosures give 0.38094 < ε ∗ < 0.38095 0.38094<\varepsilon_*<0.38095 0.38094 < ε ∗ < 0.38095 .
The prefactor itself supplies a positive phase reserve. With s = ( 1 + y − i u ) / 2 s=(1+y-iu)/2 s = ( 1 + y − i u ) /2 and
α ( s ) = 1 2 s + 1 s − 1 + 1 2 log s 2 π , \alpha(s)=\frac{1}{2s}+\frac{1}{s-1}+\frac{1}{2}\log\frac{s}{2\pi}, α ( s ) = 2 s 1 + s − 1 1 + 2 1 log 2 π s ,
its exact logarithmic derivative is
b B = 1 2 Re { α ( s ) ( 1 + t + 2 α ′ ( s ) ) } . b_B=\frac{1}{2}\operatorname{Re}\left\{\alpha(s)\left(1+\frac{t_+}{2}\alpha'(s)\right)\right\}. b B = 2 1 Re { α ( s ) ( 1 + 2 t + α ′ ( s ) ) } .
The estimates ∣ s ∣ , ∣ s − 1 ∣ ≥ u / 2 |s|,|s-1|\geq u/2 ∣ s ∣ , ∣ s − 1∣ ≥ u /2 and 0 < ℜ s ≤ 1 0<\Re s\leq1 0 < ℜ s ≤ 1 give
b B ≥ L 4 − 3 2 u − t + 4 ( L 2 + π 4 + 4 u ) ( 1 u + 6 u 2 ) > 13. (157) b_B\geq\frac{L}{4}-\frac{3}{2u}-\frac{t_+}{4}\left(\frac{L}{2}+\frac{\pi}{4}+\frac{4}{u}\right)\left(\frac{1}{u}+\frac{6}{u^2}\right)>13. \tag*{(157)} b B ≥ 4 L − 2 u 3 − 4 t + ( 2 L + 4 π + u 4 ) ( u 1 + u 2 6 ) > 13. (157)
The negative error terms decrease for u ≥ A u\geq A u ≥ A , while L / 4 L/4 L /4 increases. At A A A the directed lower enclosure is greater than 13.7584. Theorem 7.10, with r = 1 / 10 r=1/10 r = 1/10 , now yields
− Im H t ′ H t ( u + i y ) > 13 − 2 / 5 ( 1 / 10 ) ( 3 / 5 ) = 19 3 . -\operatorname{Im}\frac{H'_t}{H_t}(u+iy)>13-\frac{2/5}{(1/10)(3/5)}=\frac{19}{3}. − Im H t H t ′ ( u + i y ) > 13 − ( 1/10 ) ( 3/5 ) 2/5 = 3 19 .
The scaling (90) multiplies this phase velocity by two and halves both coordinates, proving (150) . Evenness and reality symmetry cover negative x x x . For η > 1 / 4 − 2 κ \eta>\sqrt{1/4-2\kappa} η > 1/4 − 2 κ , the classical strip contraction and theorem 4.8 already give strict positivity, which proves the final qualitative assertion. □
The preceding estimate differentiates the whole amplitude error by Cauchy. A sharper estimate retains the logarithmic generators of the finite model and applies Cauchy only to its small remainder. Centring those generators also preserves part of the correlation between the model and its derivative.
Theorem 7.12 (Exterior phase via centred moments). For every 1 25 ≤ κ ≤ 1 20 \frac{1}{25} \le\kappa\le\frac{1}{20} 25 1 ≤ κ ≤ 20 1 , ∣ x ∣ ≥ 5 × 10 15 |x| \ge5 \times10^{15} ∣ x ∣ ≥ 5 × 1 0 15 , 1 10 ≤ η ≤ 1 4 − 2 κ \frac{1}{10} \le\eta\le\sqrt{\frac{1}{4}-2\kappa} 10 1 ≤ η ≤ 4 1 − 2 κ , the heat function is nonzero at x + i η x+i\eta x + i η , and
P κ ( x , η ) > 1 2 . (158) \mathcal{P}_{\kappa}(x,\eta)>\frac{1}{2}. \tag*{(158)} P κ ( x , η ) > 2 1 . (158) Consequently its phase is positive for every ∣ x ∣ ≥ 5 × 10 15 |x| \ge5 \times10^{15} ∣ x ∣ ≥ 5 × 1 0 15 and every η ≥ 1 / 10 \eta\ge1/10 η ≥ 1/10 . The quantitative margin is asserted only in the displayed capped range.
Proof. Use w = u + i y w=u+iy w = u + i y , t = 4 κ t=4\kappa t = 4 κ , and put
X = 10 16 , A = X − 1 , r = 1 20 , t − = 4 25 , t + = 1 5 , δ = 10 − 12 , c = 1 25 . X=10^{16},\qquad A=X-1,\qquad r=\frac{1}{20},\qquad t_{-}=\frac{4}{25},\qquad t_{+}=\frac{1}{5},\qquad\delta=10^{-12},\qquad c=\frac{1}{25}. X = 1 0 16 , A = X − 1 , r = 20 1 , t − = 25 4 , t + = 5 1 , δ = 1 0 − 12 , c = 25 1 .
For u ≥ X u \ge X u ≥ X and 1 / 5 ≤ y ≤ 1 − 2 t 1/5 \le y \le\sqrt{1-2t} 1/5 ≤ y ≤ 1 − 2 t the surrounding disk lies in u ≥ A u \ge A u ≥ A , 3 / 20 ≤ y ≤ 9 / 10 < 1 3/20 \le y \le9/10 < 1 3/20 ≤ y ≤ 9/10 < 1 . Write L = log ( u / ( 4 π ) ) L=\log(u/(4\pi)) L = log ( u / ( 4 π )) , L 0 = log ( A / ( 4 π ) ) L_{0}=\log(A/(4\pi)) L 0 = log ( A / ( 4 π )) , U = L / 2 + δ U=L/2+\delta U = L /2 + δ , n 0 = 1000 n_{0}=1000 n 0 = 1000 , ℓ = log n 0 \ell=\log n_{0} ℓ = log n 0 , and λ = 4 \lambda=4 λ = 4 . The coefficient and loss comparisons in the preceding proof remain valid with this A , δ A,\delta A , δ ; in particular log N ≤ U < L \log N \le U < L log N ≤ U < L .
For β ∈ { 1 / 5 , 3 / 20 } \beta\in\{1/5,3/20\} β ∈ { 1/5 , 3/20 } define
a β = 1 + β 2 − δ , b β = 1 − β 2 − δ , Γ β ( L ) = exp { β ( 1 50 − L 2 + δ U ) } . a_{\beta}=\frac{1+\beta}{2}-\delta,\qquad b_{\beta}=\frac{1-\beta}{2}-\delta,\qquad\Gamma_{\beta}(L)=\exp\left\{\beta\left(\frac{1}{50}-\frac{L}{2}+\delta U\right)\right\}. a β = 2 1 + β − δ , b β = 2 1 − β − δ , Γ β ( L ) = exp { β ( 50 1 − 2 L + δ U ) } .
For v = a β , b β v=a_{\beta},b_{\beta} v = a β , b β let
I j v ( L ) = ∫ ℓ U q j e ( 1 − v − c L ) q + c q 2 d q , j = 0 , 1 , (159) I_{j}^{v}(L)=\int_{\ell}^{U}q^{j}e^{(1-v-cL)q+cq^{2}}\,dq,\qquad j=0,1, \tag*{(159)} I j v ( L ) = ∫ ℓ U q j e ( 1 − v − c L ) q + c q 2 d q , j = 0 , 1 , (159)
P j β ( L ) = ∑ n = 2 n 0 ( log n ) j e c log 2 n − ( a β + c L ) log n + I j a β ( L ) , P_{j}^{\beta}(L)=\sum_{n=2}^{n_{0}}(\log n)^{j}e^{c\log^{2}n-(a_{\beta}+cL)\log n}+I_{j}^{a_{\beta}}(L), P j β ( L ) = n = 2 ∑ n 0 ( log n ) j e c l o g 2 n − ( a β + c L ) l o g n + I j a β ( L ) ,
Q j β ( L ) = Γ β ( L ) { ∑ n = 1 n 0 ( log n ) j e c log 2 n − ( b β + c L ) log n + I j b β ( L ) } . (160) Q_{j}^{\beta}(L)=\Gamma_{\beta}(L)\left\{\sum_{n=1}^{n_{0}}(\log n)^{j}e^{c\log^{2}n-(b_{\beta}+cL)\log n}+I_{j}^{b_{\beta}}(L)\right\}. \tag*{(160)} Q j β ( L ) = Γ β ( L ) { n = 1 ∑ n 0 ( log n ) j e c l o g 2 n − ( b β + c L ) l o g n + I j b β ( L ) } . (160)
In the j = 0 j=0 j = 0 sums the weight is 1, including at n = 1 n=1 n = 1 . The decreasing-sum integral test bounds the model mass and its logarithmic moments by these quantities. The logarithmic and centred weights also pass the decreasing-sum guards. The integrals are evaluated by the exact erfi antiderivative, where erfi ( z ) = ( 2 / π ) ∫ 0 z e v 2 d v \operatorname{erfi}(z)=(2/\sqrt{\pi})\int_{0}^{z}e^{v^{2}}\,dv erfi ( z ) = ( 2/ π ) ∫ 0 z e v 2 d v , as given in [29 Remark 8.3] ; integration by parts gives I 1 I_{1} I 1 . At the centres use β = 1 / 5 \beta=1/5 β = 1/5 and omit that superscript. Define the centred envelopes
P λ ( L ) = ∑ n = 2 n 0 ∣ log n − λ ∣ e c log 2 n − ( a 1 / 5 + c L ) log n + I 1 a 1 / 5 − λ I 0 a 1 / 5 , (161) P_{\lambda}(L)=\sum_{n=2}^{n_{0}}|\log n-\lambda|e^{c\log^{2}n-(a_{1/5}+cL)\log n}+I_{1}^{a_{1/5}}-\lambda I_{0}^{a_{1/5}}, \tag*{(161)} P λ ( L ) = n = 2 ∑ n 0 ∣ log n − λ ∣ e c l o g 2 n − ( a 1/5 + c L ) l o g n + I 1 a 1/5 − λ I 0 a 1/5 , (161)
Q λ ( L ) = Γ 1 / 5 ( L ) { ∑ n = 1 n 0 ( L − log n − λ ) e c log 2 n − ( b 1 / 5 + c L ) log n + ( L − λ ) I 0 b 1 / 5 − I 1 b 1 / 5 } . (162) Q_{\lambda}(L)=\Gamma_{1/5}(L)\left\{\sum_{n=1}^{n_{0}}(L-\log n-\lambda)e^{c\log^{2}n-(b_{1/5}+cL)\log n}+(L-\lambda)I_{0}^{b_{1/5}}-I_{1}^{b_{1/5}}\right\}. \tag*{(162)} Q λ ( L ) = Γ 1/5 ( L ) { n = 1 ∑ n 0 ( L − log n − λ ) e c l o g 2 n − ( b 1/5 + c L ) l o g n + ( L − λ ) I 0 b 1/5 − I 1 b 1/5 } . (162)
Here ℓ > λ \ell>\lambda ℓ > λ and L − U − λ > 0 L-U-\lambda>0 L − U − λ > 0 , so the tail weights and all dual weights have the indicated signs. We justify freezing every envelope at L 0 L_{0} L 0 despite the growing endpoint U U U . Put d = 3 d=3 d = 3 , Z = U − d Z=U-d Z = U − d , and
K v = 1 − v + 2 c δ , F v = 1 − e − K v d K v . K_{v}=1-v+2c\delta,\qquad F_{v}=\frac{1-e^{-K_{v}d}}{K_{v}}. K v = 1 − v + 2 cδ , F v = K v 1 − e − K v d .
Since ∂ q { ( 1 − v − c L ) q + c q 2 } ≤ K v \partial_{q}\{(1-v-cL)q+cq^{2}\}\le K_{v} ∂ q {( 1 − v − c L ) q + c q 2 } ≤ K v for q ≤ U q\le U q ≤ U , the integral over [ U − d , U ] [U-d,U] [ U − d , U ] of q j q^{j} q j times the integrand is at least Z j F v Z^{j}F_{v} Z j F v times its value at U U U . Differentiating (159) with respect to L L L , the primary mass and logarithmic tails decrease provided
c Z F a > 1 2 , c Z 2 F a > U 2 , 2 c Z F a > 1 2 . cZ F_a > \frac{1}{2}, \qquad cZ^2 F_a > \frac{U}{2}, \qquad2cZ F_a > \frac{1}{2}. c Z F a > 2 1 , c Z 2 F a > 2 U , 2 c Z F a > 2 1 .
For the dual tails, put ρ = β ( 1 − δ ) / 2 \rho= \beta(1-\delta)/2 ρ = β ( 1 − δ ) /2 and replace these guards respectively by
( ρ + c Z ) F b > 1 2 , ( ρ + c Z ) Z F b > U 2 , ( ρ + 2 c Z ) F b > 1 2 . (\rho+cZ)F_b > \frac{1}{2}, \qquad(\rho+cZ)ZF_b > \frac{U}{2}, \qquad(\rho+2cZ)F_b > \frac{1}{2}. ( ρ + c Z ) F b > 2 1 , ( ρ + c Z ) Z F b > 2 U , ( ρ + 2 c Z ) F b > 2 1 .
The final guard in each group makes the preceding quadratic margin increase with U U U . These guards hold at L 0 L_0 L 0 for both choices of β \beta β , and their margins thereafter increase. For the centred primary tail the corresponding guards are
c Z ( Z − λ ) F a > U − λ 2 , c ( 2 Z − λ ) F a > 1 2 . cZ(Z-\lambda)F_a > \frac{U-\lambda}{2}, \qquad c(2Z-\lambda)F_a > \frac{1}{2}. c Z ( Z − λ ) F a > 2 U − λ , c ( 2 Z − λ ) F a > 2 1 .
For the centred dual tail, with β = 1 / 5 \beta=1/5 β = 1/5 and V ∗ = L − U − λ V_* = L-U-\lambda V ∗ = L − U − λ , they are
( ρ + c ℓ ) V ∗ > 1 , (\rho+c\ell)V_* > 1, ( ρ + c ℓ ) V ∗ > 1 ,
{ ( ρ + c Z ) V ∗ − 1 } F b > V ∗ 2 , \{(\rho+cZ)V_* - 1\}F_b > \frac{V_*}{2}, {( ρ + c Z ) V ∗ − 1 } F b > 2 V ∗ ,
( ρ + c Z + c V ∗ ) F b > 1 2 . (\rho+cZ+cV_*)F_b > \frac{1}{2}. ( ρ + c Z + c V ∗ ) F b > 2 1 .
Indeed its differentiated integrand is { 1 − ( ρ + c q ) ( L − q − λ ) } \{1-(\rho+cq)(L-q-\lambda)\} { 1 − ( ρ + c q ) ( L − q − λ )} times the original exponential. The first guard makes it negative throughout; the last two control the positive moving-endpoint term and the increase of its margin. The finite primary heads decrease term by term. The centred dual head decreases since 1 / ( L 0 − ℓ − λ ) < ρ 1/(L_0-\ell-\lambda)<\rho 1/ ( L 0 − ℓ − λ ) < ρ ; the other dual heads decrease as well. All these strict guards, the decreasing-sum guards, and Z > ℓ Z>\ell Z > ℓ are directed comparisons in the certificate. Thus every mass, logarithmic, and centred envelope used below is bounded by its value at L 0 L_0 L 0 , uniformly for all L ≥ L 0 L\ge L_0 L ≥ L 0 .
For a fixed centre let N c N_c N c be its integer cutoff. The holomorphic fixed-cutoff model of [29 Theorem 1.3] can be written
ψ N c = ∑ n ≤ N c a n + ∑ n ≤ N c d n , a n = b n t n − p + , d n = γ b n t n − p − , (163) \psi_{N_c}=\sum_{n\le N_c}a_n+\sum_{n\le N_c}d_n,\qquad a_n=b_n^t n^{-p_+},\qquad d_n=\gamma b_n^t n^{-p_-}, \tag*{(163)} ψ N c = n ≤ N c ∑ a n + n ≤ N c ∑ d n , a n = b n t n − p + , d n = γ b n t n − p − , (163)
where s ± = ( 1 ∓ i w ) / 2 s_\pm=(1\mp iw)/2 s ± = ( 1 ∓ i w ) /2 , p ± = s ± + t α ( s ± ) / 2 p_\pm=s_\pm+t\alpha(s_\pm)/2 p ± = s ± + t α ( s ± ) /2 , b n t = exp ( t log 2 n / 4 ) b_n^t=\exp(t\log^2 n/4) b n t = exp ( t log 2 n /4 ) , and γ = M t ( s − ) / M t ( s + ) \gamma=M_t(s_-)/M_t(s_+) γ = M t ( s − ) / M t ( s + ) . This is their formula (14) after combining its conjugate exponent, n γ n^\gamma n γ , and the small shift. In particular it does not conjugate w w w when differentiated.
Let d ± = 1 + t α ′ ( s ± ) / 2 d_\pm=1+t\alpha'(s_\pm)/2 d ± = 1 + t α ′ ( s ± ) /2 and A ± = α ( s ± ) d ± A_\pm=\alpha(s_\pm)d_\pm A ± = α ( s ± ) d ± . Direct differentiation gives
2 ψ N c ′ i = ∑ n ≤ N c d + log n a n + ∑ n ≤ N c ( A + + A − − d − log n ) d n . \frac{2\psi'_{N_c}}{i}=\sum_{n\le N_c}d_+\log n\,a_n+\sum_{n\le N_c}(A_++A_--d_-\log n)d_n. i 2 ψ N c ′ = n ≤ N c ∑ d + log n a n + n ≤ N c ∑ ( A + + A − − d − log n ) d n .
Introduce
V = ∑ n ≤ N c log n a n + ∑ n ≤ N c ( L − log n ) d n . V=\sum_{n\le N_c}\log n\,a_n+\sum_{n\le N_c}(L-\log n)d_n. V = n ≤ N c ∑ log n a n + n ≤ N c ∑ ( L − log n ) d n .
Then ∣ V − λ ( ψ N c − 1 ) ∣ ≤ P λ ( L 0 ) + Q λ ( L 0 ) |V-\lambda(\psi_{N_c}-1)|\le P_\lambda(L_0)+Q_\lambda(L_0) ∣ V − λ ( ψ N c − 1 ) ∣ ≤ P λ ( L 0 ) + Q λ ( L 0 ) . The logarithmic contribution to α ( s + ) + α ( s − ) \alpha(s_+)+\alpha(s_-) α ( s + ) + α ( s − ) is 1 2 log ( ( 1 + w 2 ) / ( 16 π 2 ) ) \frac{1}{2}\log((1+w^2)/(16\pi^2)) 2 1 log (( 1 + w 2 ) / ( 16 π 2 )) on the principal branches. With
A ∗ ′ = 1 A + 6 A 2 , A ∗ = L 0 2 + π 4 + 4 A , ϵ log = 2 A + 2 A 2 , C A = 3 A + ϵ log 4 ( 1 − ϵ log ) + t + 2 A ∗ A ∗ ′ , (164) A_*'=\frac{1}{A}+\frac{6}{A^2},\qquad A_*=\frac{L_0}{2}+\frac{\pi}{4}+\frac{4}{A},\qquad\epsilon_{\log}=\frac{2}{A}+\frac{2}{A^2},\qquad C_A=\frac{3}{A}+\frac{\epsilon_{\log}}{4(1-\epsilon_{\log})}+\frac{t_+}{2}A_*A_*', \tag*{(164)} A ∗ ′ = A 1 + A 2 6 , A ∗ = 2 L 0 + 4 π + A 4 , ϵ l o g = A 2 + A 2 2 , C A = A 3 + 4 ( 1 − ϵ l o g ) ϵ l o g + 2 t + A ∗ A ∗ ′ , (164)
the same rational and logarithmic estimates as before give ∣ A + + A − − L ∣ ≤ 2 C A |A_+ + A_- - L| \le2C_A ∣ A + + A − − L ∣ ≤ 2 C A and ∣ d ± − 1 ∣ ≤ t + A ∗ ′ / 2 |d_\pm- 1| \le t_+A_*'/2 ∣ d ± − 1∣ ≤ t + A ∗ ′ /2 . Their error terms decrease with u u u , so
∣ 2 ψ N c ′ i − V ∣ ≤ E g e n : = t + 2 A ∗ ′ ( P 1 ( L 0 ) + Q 1 ( L 0 ) ) + 2 C A Q 0 ( L 0 ) . (165) \left|\frac{2\psi'_{N_c}}{i}-V\right| \le E_{\mathrm{gen}} := \frac{t_+}{2}A_*'\left(P_1(L_0)+Q_1(L_0)\right)+2C_AQ_0(L_0). \tag*{(165)} i 2 ψ N c ′ − V ≤ E gen := 2 t + A ∗ ′ ( P 1 ( L 0 ) + Q 1 ( L 0 ) ) + 2 C A Q 0 ( L 0 ) . (165)
It remains to bound the derivative of the remainder across possible cutoff changes. Let d 1 , E C d_1,E_C d 1 , E C be (154) –(155) with the present A , L 0 , β = 3 / 20 A,L_0,\beta=3/20 A , L 0 , β = 3/20 , and y + = 9 / 10 y_+=9/10 y + = 9/10 , and set
M d i s k = P 0 3 / 20 ( L 0 ) + Q 0 3 / 20 ( L 0 ) , E 0 = ( e d 1 − 1 ) ( 1 + M d i s k ) + E C . M_{\mathrm{disk}}=P_0^{3/20}(L_0)+Q_0^{3/20}(L_0),\qquad E_0=(e^{d_1}-1)(1+M_{\mathrm{disk}})+E_C. M disk = P 0 3/20 ( L 0 ) + Q 0 3/20 ( L 0 ) , E 0 = ( e d 1 − 1 ) ( 1 + M disk ) + E C .
This bounds ∣ g t − f t ∣ |g_t-f_t| ∣ g t − f t ∣ throughout the disk. No bound M d i s k < 1 M_{\mathrm{disk}}<1 M disk < 1 is needed. The continuous cutoff u / ( 4 π ) + t / 16 \sqrt{u/(4\pi)}+t/16 u / ( 4 π ) + t /16 varies by less than one on the disk, so its floor differs from N c N_c N c by at most one. For the boundary pair, put h = δ + 4 e − L 0 / 2 h=\delta+4e^{-L_0/2} h = δ + 4 e − L 0 /2 . Its index satisfies ∣ log n − L / 2 ∣ ≤ h |\log n-L/2|\le h ∣ log n − L /2∣ ≤ h . Since log n < L \log n<L log n < L , the primary weight is at most exp ( − a 3 / 20 L / 2 − c L 2 / 4 + a 3 / 20 h + c h 2 ) \exp(-a_{3/20}L/2-cL^2/4+a_{3/20}h+ch^2) exp ( − a 3/20 L /2 − c L 2 /4 + a 3/20 h + c h 2 ) . The ratio of the dual weight to this primary bound is at most exp ( 1 / 50 + h ( 1 + δ ) + δ L / 2 ) \exp(1/50+h(1+\delta)+\delta L/2) exp ( 1/50 + h ( 1 + δ ) + δ L /2 ) . Consequently a uniform bound for the single possible jump is
J c u t : = ( 1 + e 1 / 50 + h ( 1 + δ ) ) exp { − a 3 / 20 − δ 2 L 0 − c 4 L 0 2 + a 3 / 20 h + c h 2 } . (166) J_{\mathrm{cut}}:=\left(1+e^{1/50+h(1+\delta)}\right)\exp\left\{-\frac{a_{3/20}-\delta}{2}L_0-\frac{c}{4}L_0^2+a_{3/20}h+ch^2\right\}. \tag*{(166)} J cut := ( 1 + e 1/50 + h ( 1 + δ ) ) exp { − 2 a 3/20 − δ L 0 − 4 c L 0 2 + a 3/20 h + c h 2 } . (166)
The bound decreases for L ≥ L 0 L\ge L_0 L ≥ L 0 . Cauchy is now applied to the holomorphic function g t − ψ N c g_t-\psi_{N_c} g t − ψ N c , whose disk amplitude is at most E 0 + J c u t E_0+J_{\mathrm{cut}} E 0 + J cut , giving
∣ g t ′ − ψ N c ′ ∣ ≤ E 1 : = E 0 + J c u t r . |g'_t-\psi'_{N_c}|\le E_1:=\frac{E_0+J_{\mathrm{cut}}}{r}. ∣ g t ′ − ψ N c ′ ∣ ≤ E 1 := r E 0 + J cut .
At the centre the cutoffs agree, so ∣ g t − ψ N c ∣ ≤ E 0 |g_t-\psi_{N_c}|\le E_0 ∣ g t − ψ N c ∣ ≤ E 0 .
Put
m = P 0 ( L 0 ) + Q 0 ( L 0 ) + E 0 , D = P λ ( L 0 ) + Q λ ( L 0 ) + λ E 0 , τ = E g e n / 2 + E 1 . m=P_0(L_0)+Q_0(L_0)+E_0,\qquad D=P_\lambda(L_0)+Q_\lambda(L_0)+\lambda E_0,\qquad\tau=E_{\mathrm{gen}}/2+E_1. m = P 0 ( L 0 ) + Q 0 ( L 0 ) + E 0 , D = P λ ( L 0 ) + Q λ ( L 0 ) + λ E 0 , τ = E gen /2 + E 1 .
Thus ∣ g t − 1 ∣ ≤ m < 1 |g_t-1|\le m<1 ∣ g t − 1∣ ≤ m < 1 , ∣ V − λ ( g t − 1 ) ∣ ≤ D |V-\lambda(g_t-1)|\le D ∣ V − λ ( g t − 1 ) ∣ ≤ D , and ∣ g t ′ − i V / 2 ∣ ≤ τ |g'_t-iV/2|\le\tau ∣ g t ′ − iV /2∣ ≤ τ . These imply the centred phase bound
− Im H t ′ H t ≥ b B − λ 4 − D 2 4 λ ( 1 − m 2 ) − τ 1 − m . (167) -\operatorname{Im}\frac{H'_t}{H_t}\ge b_B-\frac{\lambda}{4}-\frac{D^2}{4\lambda(1-m^2)}-\frac{\tau}{1-m}. \tag*{(167)} − Im H t H t ′ ≥ b B − 4 λ − 4 λ ( 1 − m 2 ) D 2 − 1 − m τ . (167)
To see this without independent bounds on numerator and denominator, write V = λ ( g t − 1 ) + R V=\lambda(g_t-1)+R V = λ ( g t − 1 ) + R , ∣ R ∣ ≤ D |R|\le D ∣ R ∣ ≤ D . From ∣ g t − 1 ∣ ≤ m |g_t-1|\le m ∣ g t − 1∣ ≤ m , Re g t ≥ ( ∣ g t ∣ 2 + 1 − m 2 ) / 2 \operatorname{Re}g_t\ge(|g_t|^2+1-m^2)/2 Re g t ≥ ( ∣ g t ∣ 2 + 1 − m 2 ) /2 , whence
Re ( V / g t ) ≤ λ 2 − λ ( 1 − m 2 ) 2 ∣ g t ∣ 2 + D ∣ g t ∣ ≤ λ 2 + D 2 2 λ ( 1 − m 2 ) . \operatorname{Re}(V/g_t)\le\frac{\lambda}{2}-\frac{\lambda(1-m^2)}{2|g_t|^2}+\frac{D}{|g_t|}\le\frac{\lambda}{2}+\frac{D^2}{2\lambda(1-m^2)}. Re ( V / g t ) ≤ 2 λ − 2∣ g t ∣ 2 λ ( 1 − m 2 ) + ∣ g t ∣ D ≤ 2 λ + 2 λ ( 1 − m 2 ) D 2 .
Completing the square proves the last inequality; the derivative error contributes at most τ / ( 1 − m ) \tau/(1-m) τ / ( 1 − m ) . The directed 160- and 256-bit replays give
m < 0.9071 , D < 4.5453 , E 0 < 10 − 9 , E 1 < 10 − 7 , E g e n < 2 × 10 − 16 . m<0.9071,\qquad D<4.5453,\qquad E_0<10^{-9},\qquad E_1<10^{-7},\qquad E_{\mathrm{gen}}<2\times10^{-16}. m < 0.9071 , D < 4.5453 , E 0 < 1 0 − 9 , E 1 < 1 0 − 7 , E gen < 2 × 1 0 − 16 .
The endpoint prefactor reserve in (157) exceeds 8.5775. Substitution of the full directed enclosures in (167) gives an H H H -phase lower bound greater than 0.29 > 1 / 4 0.29>1/4 0.29 > 1/4 . The factor-two scaling, evenness and conjugation prove (158) . Above the classical strip use theorem 4.8, as before. □
Corollary 7.13 (Finite phase-cap reduction). Fix κ 0 ∈ [ 1 / 25 , 1 / 20 ] \kappa_0\in[1/25,1/20] κ 0 ∈ [ 1/25 , 1/20 ] . If the continuous numerator (145) is nonnegative throughout the compact cap
∣ x ∣ ≤ 5 × 10 15 , 1 10 ≤ η ≤ 1 4 − 2 κ 0 , (168) |x|\le5\times10^{15},\qquad\frac{1}{10}\le\eta\le\sqrt{\frac{1}{4}-2\kappa_0}, \tag*{(168)} ∣ x ∣ ≤ 5 × 1 0 15 , 10 1 ≤ η ≤ 4 1 − 2 κ 0 , (168) then
Λ ≤ 4 κ 0 + 1 50 . (169) \Lambda\le4\kappa_0+\frac{1}{50}. \tag*{(169)} Λ ≤ 4 κ 0 + 50 1 . (169)
In particular, the finite sign condition at κ 0 = 1 / 25 \kappa_0 = 1/25 κ 0 = 1/25 would imply Λ ≤ 9 / 50 = 0.18 \Lambda\le9/50 = 0.18 Λ ≤ 9/50 = 0.18 .
Proof. At every regular point in the cap, divide the nonnegative numerator by ∣ Ξ κ 0 ∣ 2 |\Xi_{\kappa_0}|^2 ∣ Ξ κ 0 ∣ 2 . Outside its horizontal sides use theorem 7.12; above its top use the classical strip and theorem 4.8. The phase sign is therefore nonnegative at every regular point above 1 / 10 1/10 1/10 . Theorem 7.9 gives a global strip of half-width 1 / 10 1/10 1/10 , and theorem 7.7 gives (168). □
The exterior estimates are unconditional consequences of the published Polymath approximation and the directed constants above. The centred estimate reduces the initial cap radius by 10 9 10^9 1 0 9 . The next subsection closes the cap by proving a mollified zero-free canopy and combining it with published initial and time-barrier inputs. This gives the global strip, and hence the phase sign, without a separate direct phase calculation at every point of the cap. The statement concerns the entire heat function and does not assert continuation of the calibrated genus-two core to [ 1 / 25 , 1 / 20 ] [1/25,1/20] [ 1/25 , 1/20 ] .
Closing the cap with a mollified canopy
At κ 0 = 1 / 25 \kappa_0 = 1/25 κ 0 = 1/25 , the remaining cap can be closed by proving a zero-free region and then applying the global strip criterion. This avoids a separate logarithmic-derivative calculation at every point of the cap. The new numerical input is the following coefficient estimate; the initial finite-RH and time-barrier inputs are published results.
Lemma 7.14 (Positive thermal coefficients and paired tail overlap). Fix t ≥ 0 t \ge0 t ≥ 0 , ν ≥ 0 \nu\ge0 ν ≥ 0 , a positive integer N N N , and a finite set P P P of primes containing 2. Write
b n = e t log 2 n / 4 , D P = ∏ p ∈ P p , S ν ( s ) = ∑ n ≤ N b n n ν − s , μ P ( s ) = ∏ p ∈ P ( 1 − b p p − s ) , b_n=e^{t\log^2 n/4},\qquad D_P=\prod_{p\in P}p,\qquad S_\nu(s)=\sum_{n\le N}b_n n^{\nu-s},\qquad\mu_P(s)=\prod_{p\in P}(1-b_p p^{-s}), b n = e t l o g 2 n /4 , D P = p ∈ P ∏ p , S ν ( s ) = n ≤ N ∑ b n n ν − s , μ P ( s ) = p ∈ P ∏ ( 1 − b p p − s ) , where s = σ + i T s=\sigma+iT s = σ + i T and ρ p = b p p − σ < 1 \rho_p=b_p p^{-\sigma}<1 ρ p = b p p − σ < 1 . For squarefree d ∣ D P d\mid D_P d ∣ D P put ρ d = ∏ p ∣ d ρ p \rho_d=\prod_{p\mid d}\rho_p ρ d = ∏ p ∣ d ρ p and define the real positive sums
S ν ∗ = ∑ n ≤ N b n n ν − σ , T ν ( d ) = ∑ N / d < n ≤ N b n n ν − σ , C ν ( d ) = ∑ N / d < n ≤ N / 2 b n n ν − σ . S_\nu^*=\sum_{n\le N}b_n n^{\nu-\sigma},\qquad T_\nu(d)=\sum_{N/d<n\le N}b_n n^{\nu-\sigma},\qquad C_\nu(d)=\sum_{N/d<n\le N/2}b_n n^{\nu-\sigma}. S ν ∗ = n ≤ N ∑ b n n ν − σ , T ν ( d ) = N / d < n ≤ N ∑ b n n ν − σ , C ν ( d ) = N / d < n ≤ N /2 ∑ b n n ν − σ . Then
B ν : = ∏ p ∈ P ( 1 − ρ p ) S ν ∗ + 2 ∑ d ∣ D P ω ( d ) o d d ρ d T ν ( d ) − 2 ∑ e ∣ D P / 2 e > 1 ρ 2 e C ν ( 2 e ) , (170) B_\nu:=\prod_{p\in P}(1-\rho_p)S_\nu^*+2\sum_{\substack{d\mid D_P\\ \omega(d)\ \mathrm{odd}}}\rho_dT_\nu(d)-2\sum_{\substack{e\mid D_P/2\\ e>1}}\rho_{2e}C_\nu(2e), \tag*{(170)} B ν := p ∈ P ∏ ( 1 − ρ p ) S ν ∗ + 2 d ∣ D P ω ( d ) odd ∑ ρ d T ν ( d ) − 2 e ∣ D P /2 e > 1 ∑ ρ 2 e C ν ( 2 e ) , (170) ∣ μ P ( s ) S ν ( s ) ∣ ≤ B ν , ∣ μ P ( s ) S 0 ( s ) ∣ ≥ 2 − B 0 . (171) |\mu_P(s)S_\nu(s)|\le B_\nu,\qquad|\mu_P(s)S_0(s)|\ge2-B_0. \tag*{(171)} ∣ μ P ( s ) S ν ( s ) ∣ ≤ B ν , ∣ μ P ( s ) S 0 ( s ) ∣ ≥ 2 − B 0 . (171)
Proof. Expand the finite product as a Dirichlet polynomial. Its coefficient before the factor n − s n^{-s} n − s is
c n ν = ∑ d ∣ D P , d ∣ n n / d ≤ N ( − 1 ) ω ( d ) ( ∏ p ∣ d b p ) b n / d ( n / d ) ν . c_n^\nu=\sum_{\substack{d\mid D_P,\ d\mid n\\ n/d\le N}}(-1)^{\omega(d)}
\left(\prod_{p\mid d}b_p\right)b_{n/d}(n/d)^\nu. c n ν = d ∣ D P , d ∣ n n / d ≤ N ∑ ( − 1 ) ω ( d ) p ∣ d ∏ b p b n / d ( n / d ) ν .
For n ≤ N n\le N n ≤ N these coefficients are nonnegative. Indeed, let E E E be the selected primes dividing n n n and write n = m ∏ p ∈ E p n=m\prod_{p\in E}p n = m ∏ p ∈ E p , r = log m ≥ 0 r=\log m\ge0 r = log m ≥ 0 . After removing the common positive factor m ν b m ∏ p ∈ E p ν b p m^\nu b_m\prod_{p\in E}p^\nu b_p m ν b m ∏ p ∈ E p ν b p , the contribution of a removed subset A ⊂ E A\subset E A ⊂ E is
( − 1 ) ∣ A ∣ ∏ p ∈ A p − ν ∏ p ∈ E ∖ A e ( t / 2 ) r log p ∏ p , q ∈ E ∖ A p < q e ( t / 2 ) log p log q . (-1)^{|A|}\prod_{p\in A}p^{-\nu}\prod_{p\in E\setminus A}e^{(t/2)r\log p}\prod_{\substack{p,q\in E\setminus A\\p<q}}e^{(t/2)\log p\log q}. ( − 1 ) ∣ A ∣ p ∈ A ∏ p − ν p ∈ E ∖ A ∏ e ( t /2 ) r l o g p p , q ∈ E ∖ A p < q ∏ e ( t /2 ) l o g p l o g q .
Expand each of the last two products into 1 plus its nonnegative increment. If an expanded term covers a set F F F of prime vertices, the alternating sum over the remaining vertices is ∏ p ∈ E ∖ F ( 1 − p − ν ) ≥ 0 \prod_{p\in E\setminus F}(1-p^{-\nu})\ge0 ∏ p ∈ E ∖ F ( 1 − p − ν ) ≥ 0 . This proves positivity, including repeated prime factors in m m m . Also c 1 ν = 1 c_1^\nu=1 c 1 ν = 1 . The signed real sum of all coefficients weighted by n − σ n^{-\sigma} n − σ is ∏ p ( 1 − ρ p ) S ν ∗ \prod_p(1-\rho_p)S_\nu^* ∏ p ( 1 − ρ p ) S ν ∗ . Negative coefficients can occur only at n > N n>N n > N . Their total negative mass is bounded by the odd-subset tail mass in (170) , with a further overlap credit. Pair each subset e ∣ D P / 2 e\mid D_P/2 e ∣ D P /2 with 2 e 2e 2 e . For n = 2 e k > N n=2ek>N n = 2 e k > N and k ≤ N / 2 k\le N/2 k ≤ N /2 , both paired terms are present, have opposite signs, and the smaller magnitude is the 2 e 2e 2 e term, because
b 2 k ( 2 k ) ν ≥ b 2 b k k ν b_{2k}(2k)^\nu\ge b_{2b_k}k^\nu b 2 k ( 2 k ) ν ≥ b 2 b k k ν
The credited terms from different pairs use disjoint positive and negative components. Thus their sum is at most each of the total positive and negative tail masses, and it can be subtracted from the latter. It is exactly ∑ e > 1 ρ 2 e C ν ( 2 e ) \sum_{e>1}\rho_{2e}C_\nu(2e) ∑ e > 1 ρ 2 e C ν ( 2 e ) . The absolute coefficient mass is the signed mass plus twice the negative mass, proving the first inequality. The constant coefficient is 1, so the reverse triangle inequality gives the second one. □
Theorem 7.15 (Directed mollified canopy at t = 4 / 25 t=4/25 t = 4/25 ). For every u ≥ 5 × 10 12 u\ge5\times10^{12} u ≥ 5 × 1 0 12 and 1 / 5 ≤ y ≤ 17 / 25 1/5\le y\le\sqrt{17/25} 1/5 ≤ y ≤ 17/25 ,
∣ H 4 / 25 ( u + i y ) B 4 / 25 ( u + i y ) ∣ > 1 2000 . (172) \left|\frac{H_{4/25}(u+iy)}{B_{4/25}(u+iy)}\right|>\frac{1}{2000}. \tag*{(172)} B 4/25 ( u + i y ) H 4/25 ( u + i y ) > 2000 1 . (172)
Consequently Ξ 1 / 25 ( x + i η ) ≠ 0 \Xi_{1/25}(x+i\eta)\ne0 Ξ 1/25 ( x + i η ) = 0 whenever ∣ x ∣ ≥ ( 5 / 2 ) × 10 12 |x|\ge(5/2)\times10^{12} ∣ x ∣ ≥ ( 5/2 ) × 1 0 12 and 1 / 10 ≤ η ≤ 17 / 100 1/10\le\eta\le\sqrt{17/100} 1/10 ≤ η ≤ 17/100 .
Proof. Put t = 4 / 25 t=4/25 t = 4/25 , c = t / 4 = 1 / 25 c=t/4=1/25 c = t /4 = 1/25 , A = 5 × 10 12 − 1 A=5\times10^{12}-1 A = 5 × 1 0 12 − 1 , L 0 = log ( A / ( 4 π ) ) L_0=\log(A/(4\pi)) L 0 = log ( A / ( 4 π )) , and δ = 10 − 12 \delta=10^{-12} δ = 1 0 − 12 . The Polymath coefficient, shift, and cutoff losses are bounded as before. We also need an upper bound for σ = Re s ∗ \sigma=\operatorname{Re}s_* σ = Re s ∗ : the rational and logarithmic estimates for α \alpha α give
1 + y 2 + c L − δ ≤ σ ≤ 1 + y 2 + c L + δ . \frac{1+y}{2}+cL-\delta\le\sigma\le\frac{1+y}{2}+cL+\delta. 2 1 + y + c L − δ ≤ σ ≤ 2 1 + y + c L + δ .
Here 2 t / A < δ 2t/A<\delta 2 t / A < δ bounds the upper loss. The paired-logarithm estimate from the proof of theorem 7.12 supplies the sharper two-sided bound
∣ log ∣ γ ∣ + y L 2 ∣ ≤ 2 × 10 − 12 = : δ γ . (173) \left|\log|\gamma|+\frac{yL}{2}\right|\le2\times10^{-12}=:\delta_\gamma. \tag*{(173)} log ∣ γ ∣ + 2 y L ≤ 2 × 1 0 − 12 =: δ γ . (173)
Indeed ∣ γ ∣ = 1 |\gamma|=1 ∣ γ ∣ = 1 at y = 0 y=0 y = 0 , and ∂ y log γ = − ( A + + A − ) / 2 \partial_y\log\gamma=-(A_++A_-)/2 ∂ y log γ = − ( A + + A − ) /2 . The directed paired error C A C_A C A at this A , L 0 A,L_0 A , L 0 is less than δ γ \delta_\gamma δ γ ; integrating its uniform bound for 0 ≤ y ≤ 1 0\le y\le1 0 ≤ y ≤ 1 proves (173) .
We cover the finite logarithmic range by rectangles
L 0 + j 20 ≤ L ≤ L 0 + j + 1 20 , 0 ≤ j < 153 , 1 5 ≤ y ≤ 33 40 . L_0+\frac{j}{20}\le L\le L_0+\frac{j+1}{20},\qquad0\le j<153,\qquad\frac{1}{5}\le y\le\frac{33}{40}. L 0 + 20 j ≤ L ≤ L 0 + 20 j + 1 , 0 ≤ j < 153 , 5 1 ≤ y ≤ 40 33 .
The final horizontal edge exceeds log ( 10 16 / ( 4 π ) ) \log(10^{16}/(4\pi)) log ( 1 0 16 / ( 4 π )) and 33 / 40 > 17 / 25 33/40>\sqrt{17/25} 33/40 > 17/25 . Within each column the y y y interval is partitioned by exact rational subdivision. Denote a cell by [ L − , L + ] × [ β , β + ] [L_-,L_+]\times[\beta,\beta_+] [ L − , L + ] × [ β , β + ] and put
σ − = 1 + β 2 + c L − − δ , σ + = 1 + β + 2 + c L + + δ , Γ + = e − β L − / 2 + δ γ , Γ − = e − β + L + / 2 − δ γ . \sigma_-=\frac{1+\beta}{2}+cL_--\delta,\qquad\sigma_+=\frac{1+\beta_+}{2}+cL_++\delta,\qquad\Gamma_+=e^{-\beta L_-/2+\delta_\gamma},\qquad\Gamma_-=e^{-\beta_+L_+/2-\delta_\gamma}. σ − = 2 1 + β + c L − − δ , σ + = 2 1 + β + + c L + + δ , Γ + = e − β L − /2 + δ γ , Γ − = e − β + L + /2 − δ γ .
Evaluate the primary and combined dual mass bounds P , Q P,Q P , Q at L − L_- L − , with a head ending at 1000 and the exact erfi tail integral of (159) . The primary exponent is ( 1 + β ) / 2 − δ (1+\beta)/2-\delta ( 1 + β ) /2 − δ ; the dual exponent is ( 1 − β ) / 2 − δ (1-\beta)/2-\delta ( 1 − β ) /2 − δ , multiplied by Γ + \Gamma_+ Γ + . Their terminal-band monotonicity guards hold at every L − L_- L − . The combined dual weights decrease with y y y , since log n / 2 − L / 2 < 0 \log n/2-L/2<0 log n /2 − L /2 < 0 .
For each prime catalogue, form upper and lower bounds ρ p + , ρ p − \rho_p^+,\rho_p^- ρ p + , ρ p − using σ − , σ + \sigma_-,\sigma_+ σ − , σ + . The main mass in (170) is at most ∏ p ( 1 − ρ p − ) ( P + Q ) \prod_p(1-\rho_p^-)(P+Q) ∏ p ( 1 − ρ p − ) ( P + Q ) . Odd-subset spill sums are bounded by decreasing-sum integrals. Specifically, when 2 ( d + 1 ) < e L − / 2 2(d+1)<e^{L_-/2} 2 ( d + 1 ) < e L − /2 , put
h d = − log ( 1 − ( d + 1 ) e − L − / 2 ) , ℓ d = L − / 2 − log d − h d . h_d=-\log\left(1-(d+1)e^{-L_-/2}\right),\qquad\ell_d=L_-/2-\log d-h_d. h d = − log ( 1 − ( d + 1 ) e − L − /2 ) , ℓ d = L − /2 − log d − h d .
If ℓ d > 0 \ell_d>0 ℓ d > 0 , integrate from ℓ d \ell_d ℓ d to L − / 2 + δ L_-/2+\delta L − /2 + δ ; otherwise use the full mass bound. The same full bound is used when the displayed floor guard fails. This includes the case d > N d>N d > N . The floor estimate follows from ⌊ N / d ⌋ ≥ ( e L / 2 − 1 ) / d − 1 \lfloor N/d\rfloor\ge(e^{L/2}-1)/d-1 ⌊ N / d ⌋ ≥ ( e L /2 − 1 ) / d − 1 . After shifting the integral variable by L / 2 L/2 L /2 , its exponential factor is e ( 1 − v ) L / 2 − c L 2 / 4 e^{(1-v)L/2-cL^2/4} e ( 1 − v ) L /2 − c L 2 /4 . The asserted guard c L − > 1 − v cL_{-}>1-v c L − > 1 − v makes it decrease throughout the cell; the dual factor decreases as well.
For the overlap credits, the decreasing-sum lower estimate
∑ a < n ≤ b f ( n ) ≥ ∫ a + 1 b f ( v ) d v \sum_{a<n\le b} f(n) \ge\int_{a+1}^{b} f(v)\,\mathrm{d}v a < n ≤ b ∑ f ( n ) ≥ ∫ a + 1 b f ( v ) d v
gives a lower bound uniform over the changing cutoff. For even d > 2 d>2 d > 2 its log endpoints can be fixed at
a d + = log ( e L + / 2 + δ / d + 1 ) , b − = log ( ( e L − / 2 − 1 ) / 2 ) . a_d^{+}=\log(e^{L_{+}/2+\delta}/d+1),\qquad b^{-}=\log((e^{L_{-}/2}-1)/2). a d + = log ( e L + /2 + δ / d + 1 ) , b − = log (( e L − /2 − 1 ) /2 ) .
Integrate between these endpoints with exponents σ + \sigma_{+} σ + and σ + − β \sigma_{+}-\beta σ + − β for the primary and dual credits, respectively, and multiply the dual integral by Γ − \Gamma_{-} Γ − . An empty interval contributes zero. Denote the resulting combined lower credit by C d − C_d^{-} C d − and the combined upper spill by T d + T_d^{+} T d + . Thus a uniform bound for B 0 + ∣ γ ∣ B y \mathcal{B}_0+|\gamma|\mathcal{B}_y B 0 + ∣ γ ∣ B y is
B c e l l = ∏ p ∈ P ( 1 − ρ p − ) ( P + Q ) + 2 ∑ d ∣ D P ω ( d ) o d d ρ d + T d + − 2 ∑ d ∣ D P 2 ∣ d , d > 2 ρ d − C d − . (174) B_{\mathrm{cell}}=\prod_{p\in P}(1-\rho_p^{-})(P+Q)+2\sum_{\substack{d\mid D_P\\ \omega(d)\ \mathrm{odd}}}\rho_d^{+}T_d^{+}-2\sum_{\substack{d\mid D_P\\ 2\mid d,\ d>2}}\rho_d^{-}C_d^{-}. \tag*{(174)} B cell = p ∈ P ∏ ( 1 − ρ p − ) ( P + Q ) + 2 d ∣ D P ω ( d ) odd ∑ ρ d + T d + − 2 d ∣ D P 2 ∣ d , d > 2 ∑ ρ d − C d − . (174)
where in the last sum C d − C_d^{-} C d − denotes the integral credit before its ρ d − \rho_d^{-} ρ d − factor.
Dropping the small exponent shift from Polymath’s second sum gives S 0 ( s ∗ ) + γ S y ( s ∗ ) ‾ S_0(s_*)+\gamma\overline{S_y(s_*)} S 0 ( s ∗ ) + γ S y ( s ∗ ) with error at most Q ( e δ U + − 1 ) Q(e^{\delta U_{+}}-1) Q ( e δ U + − 1 ) , where U + = L + / 2 + δ U_{+}=L_{+}/2+\delta U + = L + /2 + δ . Equations (23)–(24) of [29 ] give the additional E A B , E C E_{AB},E_C E A B , E C majorants at L − L_{-} L − , using β + \beta_{+} β + for the positive 3 y + 3 − y 3^y+3^{-y} 3 y + 3 − y correction and Q e δ U + Qe^{\delta U_{+}} Q e δ U + in the A B AB A B sum. Put
E + = ∏ p ∈ P ( 1 + ρ p + ) , ε c e l l = E + { Q ( e δ U + − 1 ) + E A B + E C } . E_{+}=\prod_{p\in P}(1+\rho_p^{+}),\qquad\varepsilon_{\mathrm{cell}}=E_{+}\{Q(e^{\delta U_{+}}-1)+E_{AB}+E_C\}. E + = p ∈ P ∏ ( 1 + ρ p + ) , ε cell = E + { Q ( e δ U + − 1 ) + E A B + E C } .
Applying theorem 7.14 separately to S 0 S_0 S 0 and S y S_y S y , and then the triangle inequality, yields
∣ μ P ( s ∗ ) ∣ ∣ H t / B t ∣ ≥ Δ c e l l : = 2 − B c e l l − ε c e l l . (175) |\mu_P(s_*)||H_t/B_t|\ge\Delta_{\mathrm{cell}}:=2-B_{\mathrm{cell}}-\varepsilon_{\mathrm{cell}}. \tag*{(175)} ∣ μ P ( s ∗ ) ∣∣ H t / B t ∣ ≥ Δ cell := 2 − B cell − ε cell . (175)
The equality ∣ S y ‾ ∣ = ∣ S y ∣ |\overline{S_y}|=|S_y| ∣ S y ∣ = ∣ S y ∣ is used here; we do not multiply the conjugated sum by an incorrectly conjugated mollifier.
The directed program covers all 1993 rectangles at both 160 and 256 bits, with identical rational seams and catalogue choices. The catalogues end at 5, 11, 19, 29; their respective cell counts are 1977, 7, 6, 3. Every cell asserts
Δ c e l l > 1 1000 , Δ c e l l E + > 1 2000 . \Delta_{\mathrm{cell}}>\frac{1}{1000},\qquad\frac{\Delta_{\mathrm{cell}}}{E_{+}}>\frac{1}{2000}. Δ cell > 1000 1 , E + Δ cell > 2000 1 .
The smallest directed values exceed 0.0014164 and 0.0007914, respectively. No oscillatory phase is sampled; the interval and monotonicity bounds cover all real coordinates in each rectangle. This proves (172) up to the handoff. Beyond u = 10 16 u=10^{16} u = 1 0 16 the amplitude bound ∣ H t / B t − 1 ∣ < 0.9071 |H_t/B_t-1|<0.9071 ∣ H t / B t − 1∣ < 0.9071 in the proof of theorem 7.12 is stronger. Evenness and conjugation give the stated Xi region. □
Corollary 7.16 (Global closure with the published time barrier). Combining theorem 7.15 with the published finite-RH verification and rectangular time barrier gives
Ξ 1 / 25 ( z ) ≠ 0 for ∣ Im z ∣ ≥ 1 10 , Λ ≤ 9 50 = 0.18. (176) \Xi_{1/25}(z)\ne0\quad\text{for }|\operatorname{Im}z|\ge\frac{1}{10},\qquad\Lambda\le\frac{9}{50}=0.18. \tag*{(176)} Ξ 1/25 ( z ) = 0 for ∣ Im z ∣ ≥ 10 1 , Λ ≤ 50 9 = 0.18. (176) In particular the finite-cap phase-numerator condition of theorem 7.13 holds throughout its stated heat interval.
Proof. Use [29 Theorem 1.2] with t 0 = 4 / 25 t_0 = 4/25 t 0 = 4/25 and y 0 = 1 / 5 y_0 = 1/5 y 0 = 1/5 . Let X B X_B X B be the left edge of the unit time barrier located near 5 × 10 12 + 194858 5 \times10^{12} + 194858 5 × 1 0 12 + 194858 in the second row of [29 Section 10, Table 1] . The rectangular verification there, performed by the fixed-rectangle method of Section 8.4, covers 0 ≤ t ≤ 0.186 0 \le t \le0.186 0 ≤ t ≤ 0.186 and y B ≤ y ≤ 1 y_B \le y \le1 y B ≤ y ≤ 1 with y B < 0.17 y_B < 0.17 y B < 0.17 . Restricting it to t ≤ 4 / 25 t \le4/25 t ≤ 4/25 and y ≥ 1 / 5 y \ge1/5 y ≥ 1/5 supplies hypothesis (iii), since 1 − y 0 2 < 1 \sqrt{1-y_0^2} < 1 1 − y 0 2 < 1 . It is the rectangular verification that is reused: the original curved barrier hypothesis alone would not imply this restriction. The whole barrier lies between 5 × 10 12 5 \times10^{12} 5 × 1 0 12 and 5 × 10 12 + 200000 5 \times10^{12} + 200000 5 × 1 0 12 + 200000 ; the half-unit convention for its centre versus left edge does not affect any inclusion used here.
Platt–Trudgian’s [34 Theorem 1] verifies RH through height 3,000,175,332,800, which exceeds X B / 2 X_B/2 X B /2 . This supplies hypothesis (i). For hypothesis (ii), theorem 7.15 covers every u ≥ X B + 1 − y 0 2 u \ge X_B + \sqrt{1-y_0^2} u ≥ X B + 1 − y 0 2 and y 0 ≤ y ≤ 1 − 2 t 0 y_0 \le y \le\sqrt{1-2t_0} y 0 ≤ y ≤ 1 − 2 t 0 , with a strict overlap at the barrier. All three hypotheses are therefore satisfied, giving a global zero-free region at and above y 0 y_0 y 0 and Λ ≤ t 0 + y 0 2 / 2 = 4 / 25 + 1 / 50 = 9 / 50 \Lambda\le t_0 + y_0^2/2 = 4/25 + 1/50 = 9/50 Λ ≤ t 0 + y 0 2 /2 = 4/25 + 1/50 = 9/50 . The normalization gives the Xi strip in (176) .
The strip-product phase estimate is strict at every η ≥ 1 / 10 \eta\ge1/10 η ≥ 1/10 : there are no zeros on its boundary. Hence its continuous numerator is positive on the entire cap at κ = 1 / 25 \kappa= 1/25 κ = 1/25 . For later heat times use de Bruijn strip contraction; by κ = 9 / 200 \kappa= 9/200 κ = 9/200 all zeros are real. This proves the last assertion throughout [ 1 / 25 , 1 / 20 ] [1/25, 1/20] [ 1/25 , 1/20 ] . The finite-RH computation and the published time-barrier run are external numerical inputs, not replays of the accompanying program. □
The values 0.18 and 0.1732 are numerically larger than the companion comparison λ C = 0.148611789321711 \lambda_C = 0.148611789321711 λ C = 0.148611789321711 in [8 ] . The latter assumes the explicit model-to-function, localisation, and uniform-contact premises (A1)–(A3). The transfer here instead uses the published finite-RH and time-barrier results together with the new canopy certificate. To improve on λ C \lambda_C λ C while keeping h 0 = 1 / 10 h_0 = 1/10 h 0 = 1/10 would require
κ 0 < λ C − 1 / 50 4 = 0.03215294733042775. \kappa_0 < \frac{\lambda_C - 1/50}{4} = 0.03215294733042775. κ 0 < 4 λ C − 1/50 = 0.03215294733042775.
The present fixed-time canopy is not a certificate at those smaller times. At κ 0 = 1 / 25 \kappa_0 = 1/25 κ 0 = 1/25 , even h 0 = 0 h_0 = 0 h 0 = 0 would leave the ceiling 4 κ 0 = 0.16 4\kappa_0 = 0.16 4 κ 0 = 0.16 . Thus improving on λ C \lambda_C λ C requires reducing the heat time as well as controlling its canopy and barrier. None of this transports the calibrated genus-two core beyond its previously certified local heat interval.
Table 6 collects the successive ceilings and selected external comparisons. “Unconditional” describes the absence of an unproved zero-location hypothesis; it does not mean that a manuscript has undergone independent refereeing. The present canopy rows use the published finite-RH and rectangular time-barrier computations. Their numerical certificates are supplied here. The conditional and audit-relative rows retain their separate premises; the table is not an exhaustive record survey.
Table 6. Selected upper bounds for Λ \Lambda Λ , and the certified progression in this paper.
Joint strip-height optimization and its present limits
The preceding ceiling can be improved by varying both parameters in t 0 + y 2 / 2 t_0 + y^2/2 t 0 + y 2 /2 . A smaller final ceiling need not use a smaller heat time: a sufficiently narrower strip can compensate for an increase in t 0 t_0 t 0 .
Theorem 7.17 (Optimized directed canopy and global ceiling). Put
t ∗ = 12959 80000 = 0.1619875 , y ∗ = 37 200 = 0.185 , κ ∗ = 12959 320000 . (177) t_* = \frac{12959}{80000} = 0.1619875,\qquad y_* = \frac{37}{200} = 0.185,\qquad\kappa_* = \frac{12959}{320000}. \tag*{(177)} t ∗ = 80000 12959 = 0.1619875 , y ∗ = 200 37 = 0.185 , κ ∗ = 320000 12959 . (177) For every u ≥ 5 × 10 12 u \ge5 \times10^{12} u ≥ 5 × 1 0 12 and y ∗ ≤ y ≤ 1 − 2 t ∗ y_* \le y \le\sqrt{1 - 2t_*} y ∗ ≤ y ≤ 1 − 2 t ∗ ,
∣ H t ∗ ( u + i y ) B t ∗ ( u + i y ) ∣ > 1 30000 . (178) \left|\frac{H_{t_*}(u + iy)}{B_{t_*}(u + iy)}\right| > \frac{1}{30000}. \tag*{(178)} B t ∗ ( u + i y ) H t ∗ ( u + i y ) > 30000 1 . (178)
Together with the same published finite-RH and rectangular time-barrier verifications as in theorem 7.16, this gives
Ξ κ ∗ ( z ) ≠ 0 for ∣ Im z ∣ ≥ 37 400 , Λ ≤ 1791 10000 = 0.1791. (179) \Xi_{\kappa_*}(z) \ne0 \quad\text{for } |\operatorname{Im} z| \ge\frac{37}{400}, \qquad\Lambda\le\frac{1791}{10000}=0.1791. \tag*{(179)} Ξ κ ∗ ( z ) = 0 for ∣ Im z ∣ ≥ 400 37 , Λ ≤ 10000 1791 = 0.1791. (179)
Proof. Use exactly the finite coefficient inequality of theorem 7.14 and the rectangular error budget (174) –(175) , with t = t ∗ t=t_* t = t ∗ and c = t ∗ / 4 c=t_*/4 c = t ∗ /4 . The generalized engine is supplied in certificates/canopy_frontier/. Its domain, coefficient-shift, cutoff, paired-gamma and terminal-band guards are checked anew. The prime catalogues now end at 5 , 11 , 19 , 29 , 37 , 43 , 53 5,11,19,29,37,43,53 5 , 11 , 19 , 29 , 37 , 43 , 53 . With L 0 = log ( ( 5 × 10 12 − 1 ) / ( 4 π ) ) L_0=\log((5\times10^{12}-1)/(4\pi)) L 0 = log (( 5 × 1 0 12 − 1 ) / ( 4 π )) , the exact horizontal offsets are j / 200 j/200 j /200 , 0 ≤ j ≤ 40 0\le j\le40 0 ≤ j ≤ 40 , followed by 1 / 5 + j / 20 1/5+j/20 1/5 + j /20 , 1 ≤ j ≤ 149 1\le j\le149 1 ≤ j ≤ 149 . These give 189 columns ending at L 0 + 153 / 20 > log ( 10 16 / ( 4 π ) ) L_0+153/20>\log(10^{16}/(4\pi)) L 0 + 153/20 > log ( 1 0 16 / ( 4 π )) . An adaptive rational partition covers 37 / 200 ≤ y ≤ 33 / 40 37/200\le y\le33/40 37/200 ≤ y ≤ 33/40 ; the latter upper edge exceeds 1 − 2 t ∗ \sqrt{1-2t_*} 1 − 2 t ∗ .
At each of 160 and 256 bits, all 3591 rectangles satisfy Δ c e l l > 1 / 10000 \Delta_{\mathrm{cell}}>1/10000 Δ cell > 1/10000 and Δ c e l l / E + > 1 / 30000 \Delta_{\mathrm{cell}}/E_+>1/30000 Δ cell / E + > 1/30000 . The two precisions give identical rational seams and prime choices. The respective counts for the seven catalogues are 3324, 111, 113, 23, 12, 6, 2. The smallest margins exceed 0.0002457 and 0.0001296, respectively. Thus (178) holds throughout the finite logarithmic range.
The old exterior theorem cannot simply be used below y = 1 / 5 y=1/5 y = 1/5 . Instead, at L h = log ( ( 10 16 − 1 ) / ( 4 π ) ) L_h=\log((10^{16}-1)/(4\pi)) L h = log (( 1 0 16 − 1 ) / ( 4 π )) , put
a = 1 + y ∗ 2 − δ , b = 1 − y ∗ 2 − 2 δ , δ = 10 − 12 . a=\frac{1+y_*}{2}-\delta,\qquad b=\frac{1-y_*}{2}-2\delta,\qquad\delta=10^{-12}. a = 2 1 + y ∗ − δ , b = 2 1 − y ∗ − 2 δ , δ = 1 0 − 12 .
Let M ( v , L ) M(v,L) M ( v , L ) denote the positive head through 1000 plus its decreasing-sum integral tail, as in the preceding proof. The extra δ \delta δ in b b b absorbs the conjugate-exponent shift directly. Consequently the primary and shifted dual amplitudes are bounded by
P h = M ( a , L h ) − 1 < 0.807010 , Q h = e − y ∗ L h / 2 + 2 δ M ( b , L h ) < 0.105657. P_h=M(a,L_h)-1<0.807010,\qquad Q_h=e^{-y_*L_h/2+2\delta}M(b,L_h)<0.105657. P h = M ( a , L h ) − 1 < 0.807010 , Q h = e − y ∗ L h /2 + 2 δ M ( b , L h ) < 0.105657.
The terminal-band tests prove these bounds uniformly for L ≥ L h L\ge L_h L ≥ L h , and the combined dual weights decrease with y y y . The AB remainder is less than 4.90 × 10 − 16 4.90\times10^{-16} 4.90 × 1 0 − 16 and the C remainder less than 2.57 × 10 − 10 2.57\times10^{-10} 2.57 × 1 0 − 10 . Their endpoint bounds decrease with L L L : the AB exponent has the form ( a 1 L 2 + a 2 ) / ( 4 π e L − a 3 ) (a_1L^2+a_2)/(4\pi e^L-a_3) ( a 1 L 2 + a 2 ) / ( 4 π e L − a 3 ) with positive constants and L > 2 L>2 L > 2 , while the C exponential and its positive correction terms decrease in this range. Hence
∣ H t ∗ / B t ∗ ∣ ≥ 1 − P h − Q h − E A B − E C > 0.0873 > 1 30000 |H_{t_*}/B_{t_*}|\ge1-P_h-Q_h-E_{\mathrm{AB}}-E_{\mathrm{C}}>0.0873>\frac{1}{30000} ∣ H t ∗ / B t ∗ ∣ ≥ 1 − P h − Q h − E AB − E C > 0.0873 > 30000 1
on the infinite tail. This completes the canopy proof without extending the earlier phase theorem beyond its stated domain.
The published rectangle contains this new barrier because t ∗ < 0.186 t_*<0.186 t ∗ < 0.186 , y ∗ > 0.17 y_*>0.17 y ∗ > 0.17 , and 1 − y ∗ 2 < 1 \sqrt{1-y_*^2}<1 1 − y ∗ 2 < 1 . The initial RH height and the canopy/barrier overlap are unchanged. The same Polymath theorem therefore gives the global strip and
t ∗ + y ∗ 2 2 = 12959 80000 + 1369 80000 = 1791 10000 . t_*+\frac{y_*^2}{2}=\frac{12959}{80000}+\frac{1369}{80000}=\frac{1791}{10000}. t ∗ + 2 y ∗ 2 = 80000 12959 + 80000 1369 = 10000 1791 .
The normalization gives (179) . As before, the external finite-RH and time-barrier computations are reused published inputs, not independent replays.
The decrease from 0.18 is exactly 0.0009. Here t ∗ t_* t ∗ is larger than 0.16 by 0.0019875, whereas the strip contribution decreases from 0.02 to 0.0171125. This is why joint optimization improves on varying the time alone. No smaller ceiling is asserted by that Euler-only full-domain certificate. Directed point diagnostics with larger prime catalogues are supplied separately. They test this particular coefficient budget at specified parameter pairs; a negative reserve is not a zero and does not prove an intrinsic lower limit for the method. For example, at L = L 0 L=L_0 L = L 0 , y ∗ = 37 / 200 y_*=37/200 y ∗ = 37/200 and t = 12943 / 80000 t=12943/80000 t = 12943/80000 (which would give 0.1789), the margins for catalogues through 53 and 61 are respectively less than − 0.00492 -0.00492 − 0.00492 and − 0.00298 -0.00298 − 0.00298 at both precisions. This diagnoses a failure of the stated budget at that parameter pair, not an obstruction to every other pair or proof strategy. The diminishing gains from successive prime catalogues suggest that substantially larger progress will require stronger cancellation or a sharper treatment of the finite coefficient norm, rather than only finer rectangles. The reciprocal-head construction below supplies a sharper finite norm. The conditional comparison 0.148611789321711 0.148611789321711 0.148611789321711 remains numerically smaller; none of its premises (A1)–(A3) is used here.
A finite reciprocal head crosses the collective endpoint
The loss in the Euler-only estimate is not intrinsic to a canopy argument. Instead of bounding each signed component separately, we now retain a complete finite coefficient convolution. Put b n = e c log 2 n b_n=e^{c\log^2 n} b n = e c l o g 2 n and define the finite Dirichlet-inverse head by
a 1 = 1 , a n = − ∑ d ∣ n d < n a d b n / d , μ J ( s ) = ∑ d ≤ J a d d − s . (180) a_1=1,\qquad a_n=-\sum_{\substack{d\mid n\\d<n}}a_d b_{n/d},\qquad\mu_J(s)=\sum_{d\leq J}a_d d^{-s}. \tag*{(180)} a 1 = 1 , a n = − d ∣ n d < n ∑ a d b n / d , μ J ( s ) = d ≤ J ∑ a d d − s . (180)
For N ≥ J N\geq J N ≥ J , the coefficients of μ J ( s ) ∑ n ≤ N b n n − s \mu_J(s)\sum_{n\leq N}b_n n^{-s} μ J ( s ) ∑ n ≤ N b n n − s at indices 2 , … , J 2,\ldots,J 2 , … , J vanish exactly. No convergence or inversion of an infinite Dirichlet series is asserted.
Lemma 7.18 (Frozen finite-polynomial norm transfer). Fix c > 0 c>0 c > 0 , L r > 0 L_r>0 L r > 0 , y r > 0 y_r>0 y r > 0 , and a positive integer N r < e L r / 2 N_r<e^{L_r/2} N r < e L r /2 . Let a 1 = 1 , a 2 , … , a J a_1=1,a_2,\ldots,a_J a 1 = 1 , a 2 , … , a J be real, and put
σ r = 1 + y r 2 + c L r , G r = e − y r L r / 2 , S ν , N ( s ) = ∑ n ≤ N b n n − ν − s . \sigma_r=\frac{1+y_r}{2}+cL_r,\qquad G_r=e^{-y_rL_r/2},\qquad S_{\nu,N}(s)=\sum_{n\leq N}b_n n^{-\nu-s}. σ r = 2 1 + y r + c L r , G r = e − y r L r /2 , S ν , N ( s ) = n ≤ N ∑ b n n − ν − s . Write ∥ ⋅ ∥ 1 \|\cdot\|_1 ∥ ⋅ ∥ 1 for the absolute coefficient mass of a Dirichlet polynomial at the indicated real part. Suppose I P , I Q , ε P , ε Q , E ≥ 0 I_P,I_Q,\varepsilon_P,\varepsilon_Q,E\geq0 I P , I Q , ε P , ε Q , E ≥ 0 and
∥ μ J S 0 , N r ∥ 1 , σ r ≤ I P + ε P , G r ∥ μ J S y r , N r ∥ 1 , σ r ≤ I Q + ε Q , ε P + ε Q ≤ E . \|\mu_JS_{0,N_r}\|_{1,\sigma_r}\leq I_P+\varepsilon_P,\qquad G_r\|\mu_JS_{y_r,N_r}\|_{1,\sigma_r}\leq I_Q+\varepsilon_Q,\qquad\varepsilon_P+\varepsilon_Q\leq E. ∥ μ J S 0 , N r ∥ 1 , σ r ≤ I P + ε P , G r ∥ μ J S y r , N r ∥ 1 , σ r ≤ I Q + ε Q , ε P + ε Q ≤ E . Define
M = ∑ d ≤ J ∣ a d ∣ d − σ r , Q 1 = G r ∑ n ≤ N r b n n y r − σ r log n . M=\sum_{d\leq J}|a_d|d^{-\sigma_r},\qquad Q_1=G_r\sum_{n\leq N_r}b_n n^{y_r-\sigma_r}\log n. M = d ≤ J ∑ ∣ a d ∣ d − σ r , Q 1 = G r n ≤ N r ∑ b n n y r − σ r log n . Consider L ≥ L r L\geq L_r L ≥ L r , y r ≤ y ≤ y r + D y_r\leq y\leq y_r+D y r ≤ y ≤ y r + D , and N r ≤ N ≤ e U + N_r\leq N\leq e^{U_+} N r ≤ N ≤ e U + , with U + < L r U_+<L_r U + < L r . Assume that b v v − σ r b_vv^{-\sigma_r} b v v − σ r and b v v y r − σ r b_vv^{y_r-\sigma_r} b v v y r − σ r decrease on [ N r , e U + ] [N_r,e^{U_+}] [ N r , e U + ] , where b v = e c log 2 v b_v=e^{c\log^2v} b v = e c l o g 2 v , and that M Q 1 ≥ L r I Q / 2 MQ_1\geq L_rI_Q/2 M Q 1 ≥ L r I Q /2 . For σ = ( 1 + y ) / 2 + c L \sigma=(1+y)/2+cL σ = ( 1 + y ) /2 + c L and G = e − y L / 2 G=e^{-yL/2} G = e − y L /2 ,
∥ μ J S 0 , N ∥ 1 , σ + G ∥ μ J S y , N ∥ 1 , σ ≤ I P + e − D L r / 2 I Q + E + D M Q 1 + M T + , (181) \|\mu_JS_{0,N}\|_{1,\sigma}+G\|\mu_JS_{y,N}\|_{1,\sigma}\leq I_P+e^{-DL_r/2}I_Q+E+DMQ_1+MT_+, \tag*{(181)} ∥ μ J S 0 , N ∥ 1 , σ + G ∥ μ J S y , N ∥ 1 , σ ≤ I P + e − D L r /2 I Q + E + D M Q 1 + M T + , (181) T + = ∫ N r e U + b v ( v − σ r + G r v y r − σ r ) d v . T_+=\int_{N_r}^{e^{U_+}}b_v\left(v^{-\sigma_r}+G_rv^{y_r-\sigma_r}\right)\,dv. T + = ∫ N r e U + b v ( v − σ r + G r v y r − σ r ) d v .
Proof. For the fixed cutoff and fixed ν \nu ν , every coefficient of μ J S ν , N r \mu_JS_{\nu,N_r} μ J S ν , N r is independent of σ \sigma σ . Increasing σ \sigma σ therefore decreases its absolute coefficient mass. This immediately handles the primary term, without estimating derivatives of the two factors separately.
Put h = y − y r h=y-y_r h = y − y r . In the dual term, split n y r + h n^{y_r+h} n y r + h into n y r + n y r ( n h − 1 ) n^{y_r}+n^{y_r}(n^h-1) n y r + n y r ( n h − 1 ) . The first part has mass at most e − h L r / 2 ( I Q + ε Q ) e^{-hL_r/2}(I_Q+\varepsilon_Q) e − h L r /2 ( I Q + ε Q ) : the additional real-part reweighting and the additional heat-height factor both decrease it. For the second part use
e − h L r / 2 ( n h − 1 ) ≤ h log n , n ≤ N r < e L r / 2 , h ≥ 0. e^{-hL_r/2}(n^h-1)\leq h\log n,\qquad n\leq N_r<e^{L_r/2},\qquad h\geq0. e − h L r /2 ( n h − 1 ) ≤ h log n , n ≤ N r < e L r /2 , h ≥ 0.
The convolution inequality bounds it by h M Q 1 hMQ_1 h M Q 1 . The resulting upper bound I P + e − h L r / 2 I Q + E + h M Q 1 I_P+e^{-hL_r/2}I_Q+E+hMQ_1 I P + e − h L r /2 I Q + E + h M Q 1 increases for 0 ≤ h ≤ D 0\leq h\leq D 0 ≤ h ≤ D by the stated guard, giving its endpoint value at D D D . The added indices above N r N_r N r have total primary and dual mass at most T + T_+ T + by the decreasing-sum integral test. For them, the primary weights decrease as L , y L,y L , y increase; the dual weights do likewise because log n ≤ U + < L r ≤ L \log n \le U_{+} < L_r \le L log n ≤ U + < L r ≤ L . The mollifier mass is at most M M M . This gives the last debit in (181) . □
Theorem 7.19 (Reciprocal-head canopy and global ceiling). Put
t + = 12867 80000 = 0.1608375 , y + = 37 200 , κ + = 12867 320000 . t_{+} = \frac{12867}{80000} = 0.1608375,\qquad y_{+} = \frac{37}{200},\qquad\kappa_{+} = \frac{12867}{320000}. t + = 80000 12867 = 0.1608375 , y + = 200 37 , κ + = 320000 12867 . For every u ≥ 5 × 10 12 u \ge5 \times10^{12} u ≥ 5 × 1 0 12 and y + ≤ y ≤ 1 − 2 t + y_{+} \le y \le\sqrt{1-2t_{+}} y + ≤ y ≤ 1 − 2 t + ,
∣ H t + ( u + i y ) B t + ( u + i y ) ∣ > 1 30000 . \left|\frac{H_{t_{+}}(u+iy)}{B_{t_{+}}(u+iy)}\right| > \frac{1}{30000}. B t + ( u + i y ) H t + ( u + i y ) > 30000 1 . Using the published initial-RH and rectangular time-barrier verifications gives
Ξ κ + ( z ) ≠ 0 for ∣ Im z ∣ ≥ 37 400 , Λ ≤ 3559 20000 = 0.17795. \Xi_{\kappa_{+}}(z) \ne0 \quad\text{for } |\operatorname{Im} z| \ge\frac{37}{400},
\qquad
\Lambda\le\frac{3559}{20000} = 0.17795. Ξ κ + ( z ) = 0 for ∣ Im z ∣ ≥ 400 37 , Λ ≤ 20000 3559 = 0.17795.
Proof. Set c = t + / 4 c=t_{+}/4 c = t + /4 , J = 128 J=128 J = 128 , and L r = log ( ( 5 × 10 12 − 1 ) / ( 4 π ) ) L_r=\log((5\times10^{12}-1)/(4\pi)) L r = log (( 5 × 1 0 12 − 1 ) / ( 4 π )) . Choose N r = ⌊ e L r / 2 − 1 ⌋ = 630782 N_r=\lfloor e^{L_r/2-1}\rfloor=630782 N r = ⌊ e L r /2 − 1 ⌋ = 630782 and y r = y + y_r=y_{+} y r = y + . The program computes the coefficients (180) in Arb ball arithmetic and quantizes the weighted mollifier coefficients at scale 2 27 2^{27} 2 27 and the two weighted source arrays at scale 2 32 2^{32} 2 32 . It then enumerates their entire finite convolution through J N r = 80740096 JN_r=80740096 J N r = 80740096 , using exact signed integer multiplication and addition. Prior exact bounds exclude signed 64-bit overflow in each product and partial sum and unsigned 64-bit overflow in each absolute-value reduction. No floating-point array operation contributes to these coefficient norms.
The integer absolute masses, divided by 2 59 2^{59} 2 59 , are
I P = 827653878373035760 2 59 , I Q = 257672719430126216 2 59 . I_P=\frac{827653878373035760}{2^{59}},\qquad I_Q=\frac{257672719430126216}{2^{59}}. I P = 2 59 827653878373035760 , I Q = 2 59 257672719430126216 .
For source-array errors E P , E Q E_P,E_Q E P , E Q and weighted-coefficient error E μ E_\mu E μ , the convolution inequality gives the total quantization debit
E ≤ M ( E P + E Q ) + E μ ( ∥ P i n t ∥ 1 + ∥ Q i n t ∥ 1 ) , E \le M(E_P+E_Q)+E_\mu(\lVert P_{\mathrm{int}}\rVert_1+\lVert Q_{\mathrm{int}}\rVert_1), E ≤ M ( E P + E Q ) + E μ (∥ P int ∥ 1 + ∥ Q int ∥ 1 ) ,
where the integer arrays on the right are divided by 2 32 2^{32} 2 32 . Every rounding error is enclosed in Arb before summation. At both 160 and 256 bits, E < 0.000282 E<0.000282 E < 0.000282 and
I P + I Q + E < 1.883023 , M < 1.902432 , Q 1 < 4.731960. I_P+I_Q+E<1.883023,\qquad M<1.902432,\qquad Q_1<4.731960. I P + I Q + E < 1.883023 , M < 1.902432 , Q 1 < 4.731960.
The monotonicity guard in theorem 7.18 is strict. Both precisions produce the same integer masses and array hashes. This is a consistency check, not an independent transcendental arithmetic implementation.
Apply (181) on the short low corner. The real-part loss δ = 10 − 12 \delta=10^{-12} δ = 1 0 − 12 multiplies the ideal norm by at most exp ( δ ( log J + U + ) ) \exp(\delta(\log J+U_{+})) exp ( δ ( log J + U + )) , because the output indices are at most J e U + Je^{U_{+}} J e U + . The paired gamma loss contributes at most e δ e^{\delta} e δ ; the conjugate-exponent shift is charged separately by the positive dual mass, as in the previous canopy proof. The AB and C remainders are multiplied by the actual mollifier bound M e δ log J Me^{\delta\log J} M e δ l o g J . For completeness, the dual triangle step uses ∣ μ J S y ∣ = ∣ μ J ∣ ∣ S y ∣ = ∣ μ J S y ∣ |\mu_J S_y|=|\mu_J||S_y|=|\mu_J S_y| ∣ μ J S y ∣ = ∣ μ J ∣∣ S y ∣ = ∣ μ J S y ∣ ; it does not assert a cancellation between incorrectly conjugated polynomials.
The complete cover has 153 logarithmic columns of width 1 / 20 1/20 1/20 and adaptive exact rational height seams in [ 37 / 200 , 33 / 40 ] [37/200,33/40] [ 37/200 , 33/40 ] . All 2449 rectangles pass at both precisions. Nine use the reciprocal-head budget, all at heights at most 1 / 5 1/5 1/5 . The other 2440 use theorem 7.14, with catalogues through 5, 11, 19, 29 in respective counts 2414, 17, 8, 1. Every accepted cell has mollified margin > 1 / 10000 >1/10000 > 1/10000 and true normalized amplitude > 1 / 30000 >1/30000 > 1/30000 after all error debits. The smallest recorded margins exceed 0.000297 and 0.000161, respectively. The geometry and method choices agree at both precisions. The changing integer cutoff is covered by T + T_{+} T + , rather than by sampling selected cutoff values. At u ≥ 10 16 u \ge10^{16} u ≥ 1 0 16 , a fresh terminal-band amplitude calculation at ( t † , y † ) (t_{\dagger},y_{\dagger}) ( t † , y † ) , with the exponent shift absorbed as in theorem 7.17, gives ∣ H / B ∣ > 0.0698 |H/B| > 0.0698 ∣ H / B ∣ > 0.0698 . The last log edge overlaps this infinite tail, completing the canopy. The same published rectangle applies since t † < 0.186 t_{\dagger} < 0.186 t † < 0.186 , y † > 0.17 y_{\dagger} > 0.17 y † > 0.17 , and 1 − y † 2 < 1 \sqrt{1-y_{\dagger}^{2}} < 1 1 − y † 2 < 1 . The finite-RH height and horizontal overlap are unchanged. Polymath’s global criterion now gives
t † + y † 2 2 = 12867 + 1369 80000 = 3559 20000 . t_{\dagger}+\frac{y_{\dagger}^{2}}{2}=\frac{12867+1369}{80000}=\frac{3559}{20000}. t † + 2 y † 2 = 80000 12867 + 1369 = 20000 3559 .
The normalization yields (181). The initial-RH and time-barrier runs remain external published inputs; neither is independently replayed here. □ \square □
The new ceiling is below the audit-relative collective-contraction endpoint 0.177953254642861 … 0.177953254642861\ldots 0.177953254642861 … of [7 Corollary 4.3] by approximately 3.25464 × 10 − 6 3.25464 \times10^{-6} 3.25464 × 1 0 − 6 . The earlier endpoint consumes the stated imported Gomila frontier premises; the present transfer instead uses the newly certified finite polynomial norm and the published Polymath and Platt–Trudgian inputs. This compares different proof architectures, not independently accepted numerical records. The gain uses global coefficient information and does not continue the local genus-two core to these heat times.
Remark 7.20 (What the QRH input changes here). Under H 7 / 8 H_{7/8} H 7/8 , the functional equation places the initial H 0 H_0 H 0 zeros in ∣ Im z ∣ ≤ 3 / 4 |\operatorname{Im} z| \le3/4 ∣ Im z ∣ ≤ 3/4 . De Bruijn contraction then gives
∣ Im z ( H t ) ∣ ≤ max { 9 / 16 − 2 t , 0 } . |\operatorname{Im} z(H_t)| \le\sqrt{\max\{9/16-2t,0\}}. ∣ Im z ( H t ) ∣ ≤ max { 9/16 − 2 t , 0 } .
At t = t † t=t_{\dagger} t = t † , this lowers the cap top from 27133 / 200 ≈ 0.823605 \sqrt{27133}/200 \approx0.823605 27133 /200 ≈ 0.823605 to 9633 / 200 ≈ 0.490739 \sqrt{9633}/200 \approx0.490739 9633 /200 ≈ 0.490739 . It reduces the region requiring an additional canopy check, but does not remove any of the nine reciprocal-head rectangles, which all lie at y ≤ 0.2 y \le0.2 y ≤ 0.2 . In particular it does not improve (181) merely by truncating this cover. The classical contraction consequence of H 7 / 8 H_{7/8} H 7/8 alone is Λ ≤ 9 / 32 \Lambda\le9/32 Λ ≤ 9/32 , already weaker than the present ceiling. QRH may help a different quantitative contraction or barrier argument; no such additional saving is certified here. The directed geometry check treats H 7 / 8 H_{7/8} H 7/8 as an external hypothesis and does not audit the QRH release. It also supplies no new irrationality statement for the local period ratios.
Localized reciprocal heads: the ceiling 7/40
The preceding proof transports one reference norm across several columns. This can lose reserve even when the norm at a nearby reference is smaller. The following refinement uses the same transfer lemma with several reference pairs; no optimization of the inverse coefficients is needed.
Theorem 7.21 (Localized reciprocal-head canopy). Put
t ℓ = 12631 80000 = 0.1578875 , y ℓ = 37 200 , κ ℓ = 12631 320000 . t_{\ell}=\frac{12631}{80000}=0.1578875,\qquad y_{\ell}=\frac{37}{200},\qquad\kappa_{\ell}=\frac{12631}{320000}. t ℓ = 80000 12631 = 0.1578875 , y ℓ = 200 37 , κ ℓ = 320000 12631 . For every u ≥ 5 × 10 12 u \ge5 \times10^{12} u ≥ 5 × 1 0 12 and y ℓ ≤ y ≤ 1 − 2 t ℓ y_{\ell} \le y \le\sqrt{1-2t_{\ell}} y ℓ ≤ y ≤ 1 − 2 t ℓ ,
∣ H t ℓ ( u + i y ) B t ℓ ( u + i y ) ∣ > 1 30000 . \left|\frac{H_{t_{\ell}}(u+iy)}{B_{t_{\ell}}(u+iy)}\right|>\frac{1}{30000}. B t ℓ ( u + i y ) H t ℓ ( u + i y ) > 30000 1 . With the same published initial-RH and rectangular time-barrier inputs as theorem 7.19,
Ξ κ ℓ ( z ) ≠ 0 for ∣ Im z ∣ ≥ 37 400 , Λ ≤ 7 40 = 0.175. (182) \Xi_{\kappa_{\ell}}(z) \ne0 \quad\text{for } |\operatorname{Im} z| \ge\frac{37}{400},\qquad\Lambda\le\frac{7}{40}=0.175. \tag*{(182)} Ξ κ ℓ ( z ) = 0 for ∣ Im z ∣ ≥ 400 37 , Λ ≤ 40 7 = 0.175. (182) No QRH hypothesis is used.
Proof. Use the exact inverse head (180) with J = 512 J=512 J = 512 and c = t ℓ / 4 c=t_{\ell}/4 c = t ℓ /4 . At each reference choose
L r = log 5 × 10 12 − 1 4 π + ℓ r , N r = ⌊ e L r / 2 − 1 ⌋ , σ r = ( 1 + y r ) / 2 + c L r . L_r=\log\frac{5\times10^{12}-1}{4\pi}+\ell_r,\qquad N_r=\left\lfloor e^{L_r/2}-1\right\rfloor,\qquad\sigma_r=(1+y_r)/2+cL_r. L r = log 4 π 5 × 1 0 12 − 1 + ℓ r , N r = ⌊ e L r /2 − 1 ⌋ , σ r = ( 1 + y r ) /2 + c L r .
The eleven pairs ( ℓ r , y r ) (\ell_r,y_r) ( ℓ r , y r ) are
( 0 , 37 / 200 ) , ( 0 , 31 / 160 ) , ( 0 , 303 / 1600 ) , (0,37/200),\ (0,31/160),\ (0,303/1600), ( 0 , 37/200 ) , ( 0 , 31/160 ) , ( 0 , 303/1600 ) ,
( 1 / 10 , 37 / 200 ) , ( 1 / 10 , 1 / 5 ) , ( 1 / 10 , 313 / 1600 ) , (1/10,37/200),\ (1/10,1/5),\ (1/10,313/1600), ( 1/10 , 37/200 ) , ( 1/10 , 1/5 ) , ( 1/10 , 313/1600 ) ,
( 1 / 5 , 37 / 200 ) , ( 3 / 10 , 37 / 200 ) , ( 2 / 5 , 37 / 200 ) , (1/5,37/200),\ (3/10,37/200),\ (2/5,37/200), ( 1/5 , 37/200 ) , ( 3/10 , 37/200 ) , ( 2/5 , 37/200 ) ,
( 1 / 2 , 37 / 200 ) , ( 3 / 5 , 37 / 200 ) . (1/2,37/200),\ (3/5,37/200). ( 1/2 , 37/200 ) , ( 3/5 , 37/200 ) .
These are reference locations, not numerical samples used to infer a uniform sign. Each accepted rectangle satisfies L ≥ L r L \ge L_r L ≥ L r , y ≥ y r y \ge y_r y ≥ y r ; theorem 7.18 bounds the whole rectangle, including all intervening integer cutoffs. Different rectangles may use different mollifiers, since each proves nonvanishing of the same function.
The complete reference convolution and quantization procedure is that of theorem 7.19, with the same integer scales and overflow guards. For the first pair, the two exact integer masses are
I P = 832080961139007258 2 59 , I Q = 286385155079489980 2 59 . I_P=\frac{832080961139007258}{2^{59}},\qquad I_Q=\frac{286385155079489980}{2^{59}}. I P = 2 59 832080961139007258 , I Q = 2 59 286385155079489980 .
The total quantization debit is less than 0.000297, and I P + I Q + E < 1.940527 I_P+I_Q+E<1.940527 I P + I Q + E < 1.940527 . All eleven reference receipts retain both integer masses, array hashes, coefficient bounds and directed quantization errors.
The 153 logarithmic columns of width 1/20 are covered up to y = 17 / 20 y=17/20 y = 17/20 , which exceeds 1 − 2 t ℓ \sqrt{1-2t_\ell} 1 − 2 t ℓ . The exact-rational adaptive cover consists of 2483 rectangles: 51 reciprocal-head rectangles and 2390, 26, 11, 5 Euler rectangles with catalogues ending at 5, 11, 19, 29, respectively. The directed real-part, gamma, exponent-shift, AB and C debits are charged exactly as in theorem 7.19. Every mollified margin exceeds 0.0002248 and every normalized amplitude lower bound exceeds 0.0001194 > 1 / 30000 0.0001194>1/30000 0.0001194 > 1/30000 . At u ≥ 10 16 u\ge10^{16} u ≥ 1 0 16 the fresh terminal-band exterior bound gives ∣ H / B ∣ > 0.02185 \lvert H/B\rvert>0.02185 ∣ H / B ∣ > 0.02185 ; all its monotonicity guards are strict. The finite cover overlaps that exterior.
Both 160- and 256-bit runs pass. They give identical integer reference arrays and masses, the same rectangle assignments and seams, and overlapping directed cell intervals. The program and separate receipt auditor are in certificates/canopy_localized/. The auditor checks every cell, reference assignment and rational seam; the two precision runs are not independent implementations.
Finally t ℓ < 0.186 t_\ell<0.186 t ℓ < 0.186 , y ℓ > 0.17 y_\ell>0.17 y ℓ > 0.17 and the required barrier width 1 − y ℓ 2 < 1 \sqrt{1-y_\ell^2}<1 1 − y ℓ 2 < 1 fit the same published rectangular run. The finite-RH height and horizontal overlap are unchanged. The global criterion therefore gives
t ℓ + y ℓ 2 2 = 12631 + 1369 80000 = 7 40 . t_\ell+\frac{y_\ell^2}{2}=\frac{12631+1369}{80000}=\frac{7}{40}. t ℓ + 2 y ℓ 2 = 80000 12631 + 1369 = 40 7 .
This proves (182) . The two external numerical verifications are adopted from their published sources, not independently replayed here. □
local rectangles still pass; theorem 7.23 completes a stronger global certificate. At the improved time t ℓ t_\ell t ℓ , the same conditional QRH comparison lowers the classical cap top from 27369 / 200 \sqrt{27369}/200 27369 /200 to 9869 / 200 \sqrt{9869}/200 9869 /200 . All 51 reciprocal-head rectangles have y ≤ 43 / 200 < 9869 / 200 y \le43/200 < \sqrt{9869}/200 y ≤ 43/200 < 9869 /200 , so none is removed by this truncation.
Corollary 7.23 (Finer localization below 0.174). Set
t f = 2507 16000 = 0.1566875 , y f = 37 200 , κ f = 2507 64000 . t_f = \frac{2507}{16000} = 0.1566875,\qquad y_f = \frac{37}{200},\qquad\kappa_f = \frac{2507}{64000}. t f = 16000 2507 = 0.1566875 , y f = 200 37 , κ f = 64000 2507 . For u ≥ 5 × 10 12 u \ge5 \times10^{12} u ≥ 5 × 1 0 12 and y f ≤ y ≤ 1 − 2 t f y_f \le y \le\sqrt{1-2t_f} y f ≤ y ≤ 1 − 2 t f ,
∣ H t f ( u + i y ) B t f ( u + i y ) ∣ > 1 30000 . \left|\frac{H_{t_f}(u+iy)}{B_{t_f}(u+iy)}\right| > \frac{1}{30000}. B t f ( u + i y ) H t f ( u + i y ) > 30000 1 . With the same published external inputs as theorem 7.21,
Ξ κ f ( z ) ≠ 0 for ∣ Im z ∣ ≥ 37 400 , Λ ≤ 869 5000 = 0.1738. (183) \Xi_{\kappa_f}(z) \ne0 \quad\text{for } |\operatorname{Im} z| \ge\frac{37}{400},\qquad\Lambda\le\frac{869}{5000}=0.1738. \tag*{(183)} Ξ κ f ( z ) = 0 for ∣ Im z ∣ ≥ 400 37 , Λ ≤ 5000 869 = 0.1738. (183)
Proof. Keep J = 512 J=512 J = 512 and apply theorem 7.18 to 37 local reference pairs, with every reference location and exact integer mass retained in certificates/canopy_fine/. At the initial reference I P + I Q + E < 1.986018 I_P+I_Q+E<1.986018 I P + I Q + E < 1.986018 . The finer logarithmic grid has width 1 / 100 1/100 1/100 up to offset 1 / 5 1/5 1/5 , then width 1 / 20 1/20 1/20 through offset 199 / 20 199/20 199/20 . Its 215 columns and adaptive rational height seams cover 3571 rectangles up to y = 17 / 20 y=17/20 y = 17/20 . Of these, 199 use reciprocal heads; the Euler catalogues ending at 5, 11, 19, 29 cover 3313, 25, 27, 7 rectangles, respectively. The real-part, gamma, cutoff, conjugate-exponent and AB/C remainder allowances are those of theorem 7.19. Every mollified margin exceeds 0.0001320, and every amplitude lower bound exceeds 0.0000797 > 1 / 30000 >1/30000 > 1/30000 .
A fresh exterior calculation at t f t_f t f , now starting at u = 10 17 u=10^{17} u = 1 0 17 , gives ∣ H / B ∣ > 0.1929 |H/B|>0.1929 ∣ H / B ∣ > 0.1929 with strict monotonicity guards. The finite logarithmic endpoint overlaps this exterior. The older phase-tail statement for κ ≥ 1 / 25 \kappa\ge1/25 κ ≥ 1/25 is not extended below its domain or used at the new time. The 160- and 256-bit reference computations give identical integer masses and array hashes. Replaying the same finite plan at both precisions checks every cell and exact rational seam. These are consistency replays with the same arithmetic library.
The same published time-barrier rectangle contains t f < 0.186 t_f<0.186 t f < 0.186 , y f > 0.17 y_f>0.17 y f > 0.17 , and the required width 1 − y f 2 < 1 \sqrt{1-y_f^2}<1 1 − y f 2 < 1 . The initial finite-RH height and horizontal overlap are unchanged. Consequently the global criterion gives
t f + y f 2 2 = 12535 + 1369 80000 = 869 5000 . t_f+\frac{y_f^2}{2}=\frac{12535+1369}{80000}=\frac{869}{5000}. t f + 2 y f 2 = 80000 12535 + 1369 = 5000 869 .
The external verifications are not independently rerun, and no QRH hypothesis is used. This improves the certificate without changing the head length or using LP coefficient optimization. □
The earlier estimate near 0.174 was therefore a conservative working frontier, not a saturation theorem. The failed 0.173 reference in theorem 7.22 still concerns only its specified head, height and sufficient norm test; no exact saturation constant follows.
Corollary 7.24 (Extended reciprocal head). Set
y e = 18257 100000 = 0.18257 , t e = 3130681951 20000000000 = 0.15653409755 , κ e = t e 4 . y_e = \frac{18257}{100000} = 0.18257,\qquad t_e = \frac{3130681951}{20000000000} = 0.15653409755,\qquad\kappa_e = \frac{t_e}{4}. y e = 100000 18257 = 0.18257 , t e = 20000000000 3130681951 = 0.15653409755 , κ e = 4 t e . For u ≥ 5 × 10 12 u \ge5 \times10^{12} u ≥ 5 × 1 0 12 and y e ≤ y ≤ 1 − 2 t e y_e \le y \le\sqrt{1-2t_e} y e ≤ y ≤ 1 − 2 t e ,
∣ H t e ( u + i y ) B t e ( u + i y ) ∣ > 1 30000 . \left|\frac{H_{t_e}(u+iy)}{B_{t_e}(u+iy)}\right| > \frac{1}{30000}. B t e ( u + i y ) H t e ( u + i y ) > 30000 1 . With the same published external inputs as theorem 7.21,
Ξ κ e ( z ) ≠ 0 for ∣ Im z ∣ ≥ 18257 200000 , Λ ≤ 433 2500 = 0.1732. (184) \Xi_{\kappa_e}(z) \ne0 \quad\text{for } |\operatorname{Im} z| \ge\frac{18257}{200000},\qquad\Lambda\le\frac{433}{2500}=0.1732. \tag*{(184)} Ξ κ e ( z ) = 0 for ∣ Im z ∣ ≥ 200000 18257 , Λ ≤ 2500 433 = 0.1732. (184) Proof. Apply theorem 7.18 with head length J = 1024 J=1024 J = 1024 and 59 local reference pairs. The programs, exact integer masses, array hashes and directed receipts are retained in the ancillary certificates/canopy_1024/ package. The logarithmic grid has 215 columns: width $1/100$ up to offset 1 / 5 1/5 1/5 , then width $1/20 through offset 199 / 20 199/20 199/20 . Exact rational height seams partition the canopy from y e y_e y e to 17 / 20 17/20 17/20 into 3741 rectangles. Of these, 341 use the reciprocal head; the Euler catalogues ending at 5, 11, 19, 29 cover 3315, 45, 27, 13 rectangles, respectively. The quantization-transfer debit and every real-part, gamma, cutoff, conjugate-exponent and AB/C remainder allowance are included. The integer convolutions use explicit overflow guards. Every finite margin exceeds 0.00011010472, and every normalized amplitude lower bound exceeds 0.00005539885 > 1 / 30000 >1/30000 > 1/30000 .
A fresh exterior estimate at t e t_e t e , starting at u = 10 17 u=10^{17} u = 1 0 17 , gives ∣ H / B ∣ > 0.18596685 |H/B|>0.18596685 ∣ H / B ∣ > 0.18596685 , with all monotonicity guards strict. The finite logarithmic endpoint overlaps this exterior, and 17 / 20 > 1 − 2 t e 17/20>\sqrt{1-2t_e} 17/20 > 1 − 2 t e . The retained fresh computations at 192 and 256 bits agree on integer masses, array hashes and rectangle geometry. For this release, the receipt checker and both interval audits were rerun; these audits recompute every rectangle inequality, the exact partition and the exterior from the saved reference norms. They do not regenerate the full reference convolutions and are not independent implementations.
The published rectangular time barrier contains t e < 0.186 t_e<0.186 t e < 0.186 , y e > 0.17 y_e>0.17 y e > 0.17 and width 1 − y e 2 < 1 \sqrt{1-y_e^2}<1 1 − y e 2 < 1 . The same finite-RH height and horizontal overlap therefore close Polymath’s global criterion, exactly as in theorem 7.21. Finally,
t e + y e 2 2 = 3130681951 + 333318049 20000000000 = 433 2500 . t_e+\frac{y_e^2}{2}=\frac{3130681951+333318049}{20000000000}=\frac{433}{2500}. t e + 2 y e 2 = 20000000000 3130681951 + 333318049 = 2500 433 . The external finite-RH and time-barrier computations are adopted from the published sources, not rerun. No QRH hypothesis or LP coefficient optimization is used.
Certification architecture
We summarize the validated-numerics ingredients because they are part of the theorem statements, not merely exploratory computation.
Theta tails and directed moments
The theta sum (2) is truncated after finitely many terms and the remainder is bounded analytically using the doubly exponential majorant
Φ ( u ) < 4 π 2 exp ( 9 2 u − π e 2 u ) , u ≥ 0. (185) \Phi(u)<4\pi^2\exp\left(\frac{9}{2}u-\pi e^{2u}\right),\qquad u\ge0. \tag*{(185)} Φ ( u ) < 4 π 2 exp ( 2 9 u − π e 2 u ) , u ≥ 0. (185)
The moments M 2 n M_{2n} M 2 n are integrated on a finite interval with outward-rounded symmetric Taylor quadrature, while the outer u u u -tail is enclosed by (185) . The current high-precision run uses a common directed table through M 80 M_{80} M 80 . Since the same moment intervals feed every period, all branch and period certificates remain independent of any Riemann-zero table.
Regularization at T = 0 T = 0 T = 0
The formula (10) contains a removable quotient by T T T . Direct interval propagation across an interval containing zero would be badly conditioned. Proposition 2.3 replaces it by K / Y 6 K/Y^{6} K / Y 6 . On the small symmetric central segment, an even Taylor jet of Δ 0 ( T ) \sqrt{\Delta_0(T)} Δ 0 ( T ) is integrated analytically; odd contributions cancel. In the computation underlying theorem 7.2, the resulting central integral enclosure has width below 10 − 24 10^{-24} 1 0 − 24 , so T = 0 T = 0 T = 0 is no longer a significant error source.
Complex branch points
At a proposed center c 0 c_0 c 0 , write
K ( c 0 + w ) = k 0 + k 1 w + ∑ m ≥ 2 k m w m . K(c_0 + w) = k_0 + k_1w + \sum_{m \ge2} k_m w^m. K ( c 0 + w ) = k 0 + k 1 w + m ≥ 2 ∑ k m w m .
A directed Rouché test proves
∣ k 0 ∣ + ∑ m ≥ 2 ∣ k m ∣ ρ m < ∣ k 1 ∣ ρ |k_0| + \sum_{m \ge2} |k_m|\rho^m < |k_1|\rho ∣ k 0 ∣ + m ≥ 2 ∑ ∣ k m ∣ ρ m < ∣ k 1 ∣ ρ
on ∣ w ∣ = ρ |w| = \rho ∣ w ∣ = ρ . This certifies exactly one simple zero in the disk. A separate lower enclosure for ∣ Y ∣ |Y| ∣ Y ∣ excludes a transverse pole. For both certified nonreal pairs, the final branch-disk radius is ρ = 10 − 9 \rho= 10^{-9} ρ = 1 0 − 9 .
Heat jets, branch contraction, and the projective tangent certificate
The heat program uses the exact gamma–zeta representation of ξ \xi ξ in Arb/Acb Taylor arithmetic through python-flint 0.9.0. The identity M 2 n = ξ ( 2 n ) ( 1 / 2 ) / 4 M_{2n} = \xi^{(2n)}(1/2)/4 M 2 n = ξ ( 2 n ) ( 1/2 ) /4 relates its central jets to the theta moments; the program does not import the earlier numerical moment table. Four Rouché checks independently refine the upper and real selected roots, with conjugation supplying the remaining two. All source centers are exact decimal rationals; their role is to propose disks, and acceptance requires the displayed strict Rouché inequalities.
The shifted jets give ∂ κ K \partial_\kappa K ∂ κ K , while automatic polynomial differentiation along (37) gives the scalar Lax tangent. The input for the latter is the complete static parameter box of theorem 3.12; its existence remains the content of that earlier realization theorem. The new program does not replace its Krawczyk proof. Each derivative denominator is excluded from zero, and the rational rectangle for D \mathcal{D} D is checked by strict interval comparisons. The supplement records successful 192-bit and 256-bit replays. Agreement between these precisions is a consistency check, not independence of transcendental arithmetic implementations. The separate contraction program reuses the refined-root and heat-jet identities, evaluates all fifteen logarithmic rates in the natural T T T coordinate, and checks c j k > 1 / 9 c_{jk} > 1/9 c j k > 1/9 by strict directed comparisons. It also checks both representative rational intervals in (97) ; its 192-bit and 256-bit logs and JSON receipts are included.
Contour periods
Away from branch endpoints, Δ 0 \Delta_0 Δ 0 is expanded on certified zero-free disks. Interval Taylor coefficients for the selected branch of Δ 0 \sqrt{\Delta_0} Δ 0 are obtained recursively from
( ∑ n ≥ 0 b n w n ) 2 = ∑ n ≥ 0 d n w n . \left(\sum_{n \ge0} b_n w^n\right)^2 = \sum_{n \ge0} d_n w^n. ( n ≥ 0 ∑ b n w n ) 2 = n ≥ 0 ∑ d n w n .
Even terms are integrated exactly on symmetric panels. A Cauchy bound on a larger certified disk controls the Taylor remainder. Adjacent square-root image disks overlap on the same sheet and are disjoint from the opposite-sheet disks; this propagates the branch without repeated pointwise reevaluation. Simple-root endpoint tails are treated separately.
Directed heat derivatives of periods
The period engine constructs entire Taylor models for Y Y Y , E E E , E ˙ \dot{E} E ˙ from central gamma–zeta jets, with rigorous global Cauchy tails on ∣ T ∣ = 4096 |T|=4096 ∣ T ∣ = 4096 . This handles the removable division by T T T in E E E and E ˙ \dot{E} E ˙ before contour evaluation, including panels that cross zero. The exact shifted heat jets produce K ˙ \dot{K} K ˙ and Δ ˙ \dot{\Delta} Δ ˙ . Interior panels integrate the continued functions Δ 0 \sqrt{\Delta_0} Δ 0 and Δ ˙ / ( 2 Δ 0 ) \dot{\Delta}/(2\sqrt{\Delta_0}) Δ ˙ / ( 2 Δ 0 ) together.
For a simple endpoint T j T_j T j , write K ( T ) = ( T − T j ) k ( T ) K(T)=(T-T_j)k(T) K ( T ) = ( T − T j ) k ( T ) . On a disk containing the root and the omitted segment, the certificate proves
m ≤ ∣ k ( T ) ∣ ≤ M , ∣ Y ( T ) ∣ ≥ y > 0 , ∣ K ˙ ( T ) ∣ ≤ B , ∣ Y ˙ ( T ) ∣ ≤ H , m \leq|k(T)| \leq M,\qquad|Y(T)| \geq y>0,\qquad|\dot{K}(T)| \leq B,\qquad|\dot{Y}(T)| \leq H, m ≤ ∣ k ( T ) ∣ ≤ M , ∣ Y ( T ) ∣ ≥ y > 0 , ∣ K ˙ ( T ) ∣ ≤ B , ∣ Y ˙ ( T ) ∣ ≤ H ,
with m > 0 m>0 m > 0 . The bounds on k k k follow from a nonzero enclosure for K ′ K' K ′ at the rational root center and a Taylor–Cauchy variation bound. For an omitted segment of length at most ℓ \ell ℓ , including root-location uncertainty, the two endpoint errors are bounded by
∣ ∫ t a i l ω 0 ∣ ≤ 2 M 3 y 3 ℓ 3 / 2 , (186) \left|\int_{\mathrm{tail}}\omega_0\right| \leq\frac{2\sqrt{M}}{3y^3}\ell^{3/2}, \tag*{(186)} ∫ tail ω 0 ≤ 3 y 3 2 M ℓ 3/2 , (186)
∣ ∫ t a i l ω ˙ ∣ ≤ B y 3 m ℓ 1 / 2 + 2 M H y 4 ℓ 3 / 2 . (187) \left|\int_{\mathrm{tail}}\dot{\omega}\right| \leq\frac{B}{y^3\sqrt{m}}\ell^{1/2}+\frac{2\sqrt{M}H}{y^4}\ell^{3/2}. \tag*{(187)} ∫ tail ω ˙ ≤ y 3 m B ℓ 1/2 + y 4 2 M H ℓ 3/2 . (187)
These follow by integrating the pointwise square-root and inverse-square-root majorants. In particular, the derivative tail is bounded analytically rather than discarded because its endpoint moves. Both precision replays reproduce period boxes strictly inside the original calibration boxes and verify the six rational derivative intervals in Table 4 . The countable-dense-set conclusion of theorem 6.4 is then an analytic consequence of monotonicity and the kernel criterion, requiring no sampling of rational heat times. The simple-resonance proposition and generic Diophantine theorem are further analytic deductions from the same inputs. Their measure estimate uses a summable rational-approximation covering, rather than a new numerical certificate. The negative-heat corollary combines this local arithmetic with the independently established Rodgers–Tao lower bound.
Uniform heat models and explicit Farey persistence
The uniform heat engine constructs heat Taylor coefficients through order thirteen from exact central jets. If Y 0 ( T ) = ∑ n ≥ 0 c n T n Y_0(T)=\sum_{n\geq0}c_nT^n Y 0 ( T ) = ∑ n ≥ 0 c n T n , then
[ κ j ] Y κ ( T ) = ∑ n ≥ 0 ( 2 n + 2 j ) ! ( 2 n ) ! j ! c n + j T n . [\kappa^j]Y_\kappa(T)=\sum_{n\geq0}\frac{(2n+2j)!}{(2n)!j!}c_{n+j}T^n. [ κ j ] Y κ ( T ) = n ≥ 0 ∑ ( 2 n )! j ! ( 2 n + 2 j )! c n + j T n .
Products and the normalization A ∗ ( κ ) A_*(\kappa) A ∗ ( κ ) are formed in truncated heat series; the exact removable constant term of E E E is shifted away before evaluation. Spatial degree 85 for Y Y Y and degree 80 for E E E are retained. The spatial jets of each heat coefficient are evaluated first, and K K K is then formed in heat-series arithmetic. Only afterwards is κ \kappa κ replaced by the real ball [ − 10 − 9 , 10 − 9 ] [-10^{-9},10^{-9}] [ − 1 0 − 9 , 1 0 − 9 ] . This order retains cancellations lost by independent interval substitution into the primitive factors of K K K .
Joint Cauchy estimates use the complex heat disk ∣ κ ∣ ≤ ρ = 10 − 3 |\kappa|\leq\rho=10^{-3} ∣ κ ∣ ≤ ρ = 1 0 − 3 , the global spatial circle ∣ T ∣ = 4096 |T|=4096 ∣ T ∣ = 4096 , and the local circle ∣ T ∣ = 512 |T|=512 ∣ T ∣ = 512 . Positivity of the theta kernel bounds Y Y Y and its derivatives on these circles. Split its integral at u = 8 u=8 u = 8 : on the finite part, ∣ e κ u 2 ∣ ≤ e 64 ρ |e^{\kappa u^2}|\leq e^{64\rho} ∣ e κ u 2 ∣ ≤ e 64 ρ ; on the outer part, (185) bounds integrals with additional factors u m e ρ u 2 + v u u^m e^{\rho u^2+vu} u m e ρ u 2 + v u . The logarithmic derivative of this majorant is decreasing beyond eight and negative there, giving an explicit first-point divided-by-slope bound for the integral. Moreover,
∣ Y κ ( 0 ) ∣ ≥ ( 2 − e 64 ρ ) Y 0 ( 0 ) − ∫ 8 ∞ e ρ u 2 Φ ( u ) d u > 0. |Y_\kappa(0)|\geq(2-e^{64\rho})Y_0(0)-\int_8^\infty e^{\rho u^2}\Phi(u)\,\mathrm{d}u>0. ∣ Y κ ( 0 ) ∣ ≥ ( 2 − e 64 ρ ) Y 0 ( 0 ) − ∫ 8 ∞ e ρ u 2 Φ ( u ) d u > 0.
Thus A ∗ ( κ ) A_*(\kappa) A ∗ ( κ ) is holomorphic on this complex heat disk, and triangle inequalities also bound the entire functions E E E and K K K .
Write q = 10 − 9 / ρ q = 10^{-9}/\rho q = 1 0 − 9 / ρ . If M L M_L M L bounds an entire spatial function F κ F_\kappa F κ on ∣ T ∣ = 512 |T| = 512 ∣ T ∣ = 512 , uniformly on the complex heat disk, its heat remainder after order twelve and the corresponding derivative remainder are bounded by
∣ F κ − ∑ j = 0 12 F j κ j ∣ ≤ M L q 13 1 − q , \left|F_\kappa-\sum_{j=0}^{12}F_j\kappa^j\right|\leq\frac{M_Lq^{13}}{1-q}, F κ − j = 0 ∑ 12 F j κ j ≤ 1 − q M L q 13 ,
∣ ∂ κ F κ − ∑ j = 0 12 ( j + 1 ) F j + 1 κ j ∣ ≤ M L q 13 ( 14 − 13 q ) ρ ( 1 − q ) 2 , F j = [ κ j ] F κ . \left|\partial_\kappa F_\kappa-\sum_{j=0}^{12}(j+1)F_{j+1}\kappa^j\right|\leq\frac{M_Lq^{13}(14-13q)}{\rho(1-q)^2},\qquad F_j=[\kappa^j]F_\kappa. ∂ κ F κ − j = 0 ∑ 12 ( j + 1 ) F j + 1 κ j ≤ ρ ( 1 − q ) 2 M L q 13 ( 14 − 13 q ) , F j = [ κ j ] F κ .
Cauchy bounds on the remaining spatial radius control derivatives of these remainders; the global circle controls omitted spatial coefficients of each heat Taylor coefficient. Both error terms are included in every jet enclosure.
For the moving-root certificate, the center value is bounded with its common heat dependence:
∣ K κ ( c ) ∣ ≤ ∣ K 0 ( c ) ∣ + 10 − 9 ∣ ∂ κ K 0 ( c ) ∣ + 1 2 10 − 18 sup ∣ s ∣ ≤ 10 − 9 ∣ ∂ s 2 K s ( c ) ∣ . |K_\kappa(c)|\leq|K_0(c)|+10^{-9}|\partial_\kappa K_0(c)|+\frac{1}{2}10^{-18}\sup_{|s|\leq10^{-9}}|\partial_s^2K_s(c)|. ∣ K κ ( c ) ∣ ≤ ∣ K 0 ( c ) ∣ + 1 0 − 9 ∣ ∂ κ K 0 ( c ) ∣ + 2 1 1 0 − 18 ∣ s ∣ ≤ 1 0 − 9 sup ∣ ∂ s 2 K s ( c ) ∣.
A uniform nonzero linear spatial jet and a second-derivative bound then give the Rouché test on radius 10 − 7 10^{-7} 1 0 − 7 . The same uniform jets and endpoint majorants control the continued periods and their heat derivatives. Each Farey form is bounded by integrating its combined derivative, retaining the common heat time instead of assigning an independent time to each period coordinate. Both precision replays check the explicit interval, all four strict form boxes, and both positive velocity boxes in theorem 7.5.
Directed exterior phase constants
The certificates/phase_tail/ directory contains two standalone Arb programs with python-flint==0.9.0, a pinned requirement, and successful 160- and 256-bit logs and JSON receipts for each. The baseline evaluates the two positive heads ending at n = 100 n = 100 n = 100 , the four convex endpoint tails, the two approximation-error majorants, and the prefactor phase reserve from theorem 7.11. Every published rational bound is a strict program assertion. It also asserts each domain, loss, endpoint, and monotonicity guard used in the proof. The uniform conclusion over unbounded horizontal coordinates and the heat interval follows from the analytic monotonicity and coefficient comparisons, rather than from sampling points or times. Cauchy is applied to the exact H t / B t − 1 H_t/B_t-1 H t / B t − 1 on the expanded disks, so no derivative is taken across a change of the Riemann–Siegel cutoff. Agreement at two precisions is a consistency check within the same arithmetic implementation. The centred refinement uses heads ending at n = 1000 n = 1000 n = 1000 , exact erfi mass integrals, and their logarithmic and centred moments. Its 74 strict assertions also check the terminal-band monotonicity guards, the one-pair cutoff-jump bound, and every published constant in theorem 7.12. It differentiates a holomorphic model with the centre’s cutoff fixed; Cauchy bounds only its remainder after the possible jump is included. In particular, the expanded model’s amplitude bound need not be less than one. The exact centre amplitude bound proves nonvanishing before the phase estimate is applied. The earlier baseline program and receipts are retained unchanged. Neither exterior program checks the finite cap directly. The following canopy certificate supplies the complementary input for theorem 7.16.
Mollified canopy and published-input transfer
The certificates/finite_cap/ directory contains the directed canopy engine and covering driver, a pinned requirement, 160- and 256-bit logs and full cell receipts. All 1993 rectangles pass the true-function nonvanishing margins; exact rational seams cover the entire finite logarithmic range and the height interval. Thermal coefficient positivity and the tail-overlap formula have an analytic proof in theorem 7.14. A separate finite enumeration checks twelve small convolutions at each precision and rejects an incorrect overlap range. An exact-rational checker records the domain inclusions for the global transfer. It checks neither the original time-barrier run nor Platt–Trudgian’s large finite-RH computation. Those two published results are explicitly cited in theorem 7.16; only the canopy is newly computed here. The separate certificates/canopy_frontier/ directory generalizes the same finite coefficient budget to theorem 7.17. Its 3591 rectangles, fresh infinite-tail bound and exact transfer inclusions pass at both precisions. Separate point diagnostics up to prime 61 document failed budgets without interpreting them as zeros or optimality statements. The additional certificates/canopy_inverse/ directory supplies the finite inverse head and exact integer convolution, all 2449 covering rectangles at both precisions, a fresh infinite tail and a separate receipt/index auditor. Its QRH geometry checker assumes H 7 / 8 H_{7/8} H 7/8 only to compare cap heights; the new ceiling does not use it.
The localized extension in certificates/canopy_localized/ retains all eleven reference-pair receipts, the 2483-rectangle covers at 160 and 256 bits, a fresh exterior bound, a separate exact-seam and interval auditor, and point/transfer diagnostics. The enlarged rank test in certificates/lax_enlarged/ recertifies the branch disks and proves the rational determinant and rate bounds at 192 and 256 bits. Its inverse-function consequence concerns local branch moduli; no Stokes-data certificate is supplied.
The fine extension in certificates/canopy_fine/ retains the 37 reference pairs, full 3571-cell receipts at both precisions, a fresh 10 17 10^{17} 1 0 17 exterior, a portable frozen-plan replay and an auditor.
Arithmetic aggregation
The period intervals themselves are not the final object. Integer forms such as
419336 u 1 − 13749 u 2 419336u_1 - 13749u_2 419336 u 1 − 13749 u 2
are aggregated with outward rounding, and the sign is accepted only when the resulting interval is strictly separated from zero. Farey-neighbor determinants are exact integer identities. The supplementary certificate prints the four decisive linear-form intervals and both determinant checks.
Discussion and open arithmetic problems
The construction links three levels of description: a real genus-two core extracted from the Xi discriminant, its calibrated action arithmetic, and the global phase of the heat-deformed entire function. Table 1 records the certified connections. The principal open questions concern the distinguished arithmetic point, rather than further finite numerical refinement.
The first is theorem 1.1: are u 1 / u 2 u_1/u_2 u 1 / u 2 and r 1 / r 2 r_1/r_2 r 1 / r 2 both irrational at κ = 0 \kappa= 0 κ = 0 ? The Farey barriers exclude rational relations up to substantial denominator height, but finite-width enclosures cannot establish irrationality. Positive heat velocities give complementary generic statements, including almost-everywhere Diophantine bounds, whose constants depend on heat time and deteriorate near the dense resonant set. They do not single out zero. A direct arithmetic interpretation of the theta-moment action coordinates, or a functional identity giving fixed-point Diophantine bounds, would address this gap. Standard number-field period theorems are not directly applicable because no algebraic number-field model of this Xi-derived core is known.
The second is theorem 3.36. The finite Humbert exclusions and the Mordell–Weil height barrier constrain possible Picard enhancements, while theorem 3.30 reduces their complete exclusion to irrationality of det Im Ω \det\operatorname{Im}\Omega det Im Ω and Q \mathbb{Q} Q -linear independence of the three entries of Im Ω \operatorname{Im}\Omega Im Ω . Humbert divisors are dense on the real slice, so extending the finite exclusion ladder cannot settle this question. The numerical period matrix is consistent with minimal rank, but a suitable exact arithmetic argument remains necessary.
The D 6 D_6 D 6 correspondence clarifies the deformation problem without resolving either arithmetic conjecture. The heat tangent differs from the matched fixed-exponent isomonodromic tangent. Allowing the two exponent directions gives a local lift of the branch moduli, but does not identify the full scalar potentials, preserve Stokes data, or establish global continuation. A global positive canonical system with Weyl–Titchmarsh function
m ( z ) = − Ξ ′ ( z ) Ξ ( z ) m(z)=-\frac{\Xi'(z)}{\Xi(z)} m ( z ) = − Ξ ( z ) Ξ ′ ( z )
would require an additional Herglotz/positivity argument beyond this local curve correspondence.
A Herglotz function is a holomorphic function mapping the upper half-plane into itself; a Weyl–Titchmarsh function records the boundary response of a canonical differential system, a first-order system governed by a positive semidefinite Hamiltonian.
The global canopy certificate answers a separate analytic question: combined with the published finite-RH and rectangular time-barrier inputs, it gives Λ ≤ 0.1732 \Lambda\le0.1732 Λ ≤ 0.1732 . The bound uses no QRH assumption and makes no optimality claim. Its proof requires neither conjecture above. Conversely, the coexistence of nonresonant cores and nonreal heat zeros at negative times shows that generic action separation alone is not a zero-reality criterion. Any proposed RH interpretation at κ = 0 \kappa= 0 κ = 0 must therefore supply additional fixed-point arithmetic structure.
Supplementary computational files
The ancillary directory anc/ contains twelve certificate packages, listed in Table 7 . Directory names in that table are relative to anc/certificates/; other displayed certificates/ paths are relative to anc/. The file anc/COMPUTATIONAL_INDEX.md gives the exact Python inventory, entry points, dependencies and replay scope. Each package retains its own instructions and numerical records.
Table 7. Computational packages supplied with the manuscript.
The eleven programs in the Humbert proof are supplied under certificates/humbert/, together with the generated files required by its H 16 H_{16} H 16 chain. The extended-head package is nested under canopy_1024/head_extension_J1024/; its README distinguishes full convolution regeneration from audits using saved norms. Floating point diagnostics are identified separately and are not theorem inputs.
The archive provides the computational coverage listed above; it does not contain every standalone calculation used during development. In particular, separate static theta-moment, initial Lax-matching, Kummer-model and homology-check scripts are not included as independent replay packages. Their displayed derivations and bounds remain in the manuscript; the supplied heat and period packages also recertify selected branch and period data, within their documented scope. The computational index records these limits explicitly. The published finite-RH and time-barrier computations, and the external QRH release, are cited inputs rather than computations replayed by this supplement.
Disclosure of AI-assisted preparation
OpenAI ChatGPT 6.1 and Anthropic Claude Opus 5.5 assisted with mathematical exploration, code generation and review, manuscript organization, and language editing. The author is responsible for the mathematical claims, references, computational certificates, and final text. AI assistance and successful computational replay do not constitute independent mathematical refereeing.
References