Introduction
The channel patrol problem
A searcher who commits to a patrol can block a channel only if an evader who knows its schedule cannot time a crossing. Washburn posed this question for a single searcher and compared it with two solvable restrictions: joining the walls into a tube, and imposing a transverse back-and-forth patrol [7 Sec. 2.C.] , [8 pp. 36–37] . In the latter case the critical width and the optimal transverse half-amplitude are
W b f ( q ) = r 4 + q − 2 , a 0 = r 2 q 1 + 4 q 2 , (1) W_{\mathrm{bf}}(q)=r\sqrt{4+q^{-2}}, \qquad a_0=\frac{r}{2q\sqrt{1+4q^2}}, \tag*{(1)} W bf ( q ) = r 4 + q − 2 , a 0 = 2 q 1 + 4 q 2 r , (1)
where q q q is the ratio of the intruder’s speed to the guard’s. In the tube approximation the critical width for q < 1 q<1 q < 1 is 2 r / q 2r/q 2 r / q . Even the transverse value in (1) exceeds this when 3 / 2 < q < 1 \sqrt{3}/2<q<1 3 /2 < q < 1 , giving a counterexample to the conjecture that joining the walls always favors the searcher [7 Sec. 2.C.] .
Bowtie patrols have a longer history. Koopman’s crossover patrol uses a lead angle chosen for a target traveling on a straight course [1 Sec. 7.1 and Fig. 6] , [4 pp. 200–203] . Balinski, Lucas, Owen, and Sand [2 App. I–II] and Washburn [3 ] optimize detection probabilities for straight transitors. Galyaev and Maslov [9 ] also study fixed-course intruders. Hew [10 , 11 ] studies prescribed patrols against maneuvering evaders, and Rowling [12 ] treats effective sweep width with counter-detection. In the present game every admissible transit must be intercepted, at any starting phase.
Washburn’s area estimate for closed search paths against straight targets [5 eqs. (2)–(3)] gives
W o p t ( q ) ≤ W a r e a ( q ) : = 2 r 1 + q − 2 . (2) W_{\mathrm{opt}}(q)\le W_{\mathrm{area}}(q):=2r\sqrt{1+q^{-2}}. \tag*{(2)} W opt ( q ) ≤ W area ( q ) := 2 r 1 + q − 2 . (2)
For completeness, set the guard speed to one. In the straight runners’ coordinates ( x , q t − y ) (x,qt-y) ( x , q t − y ) , the swept disk on [ 0 , L ] [0,L] [ 0 , L ] has area at most π r 2 + 2 r ∫ 0 L X ′ 2 + ( q − Y ′ ) 2 d t \pi r^2+2r\int_0^L\sqrt{X'^2+(q-Y')^2}\,\mathrm{d}t π r 2 + 2 r ∫ 0 L X ′2 + ( q − Y ′ ) 2 d t by polygonal approximation.
The integrand is at most 1 + q 2 − q Y ′ / 1 + q 2 \sqrt{1+q^2}-qY'/\sqrt{1+q^2} 1 + q 2 − q Y ′ / 1 + q 2 by the speed bound and concavity of the square root. If ∣ Y ∣ ≤ H |Y|\le H ∣ Y ∣ ≤ H , all straight phases in [ H + r , q L − H − r ] [H+r,qL-H-r] [ H + r , q L − H − r ] on every lane must be detected in this window. Comparing areas, dividing by L L L , and sending L L L to infinity proves (2) .
Exact Mathematical Model and Conventions
Fix capture radius r > 0 r>0 r > 0 , maximum guard speed V > 0 V>0 V > 0 , maximum intruder speed U > 0 U>0 U > 0 , and dimensionless speed ratio q : = U / V > 0 q:=U/V>0 q := U / V > 0 . By rescaling space by r r r and time by r / V r/V r / V , we often work in normalized units r = 1 r=1 r = 1 , V = 1 V=1 V = 1 , U = q U=q U = q , restoring explicit factors of r r r in theorem statements. In these units a unit-speed patrol means a patrol with speed bounded by one; waiting is allowed.
Definition 1.1 (Corridor, Patrols, and Evasion). For half-width b > 0 b>0 b > 0 (corridor width W = 2 b W=2b W = 2 b ), let C b : = [ − b , b ] × R C_b:=[-b,b]\times\mathbb{R} C b := [ − b , b ] × R .
A prescribed patrol is an absolutely continuous trajectory g = ( X , Y ) : R → C b g=(X,Y):\mathbb{R}\to C_b g = ( X , Y ) : R → C b satisfying the speed bound ess sup t ∈ R X ′ ( t ) 2 + Y ′ ( t ) 2 ≤ V \operatorname*{ess\,sup}_{t\in\mathbb{R}}\sqrt{X'(t)^2+Y'(t)^2}\le V ess sup t ∈ R X ′ ( t ) 2 + Y ′ ( t ) 2 ≤ V . We say g g g has bounded longitudinal excursion if H ( g ) : = sup t ∈ R ∣ Y ( t ) ∣ < ∞ H(g):=\sup_{t\in\mathbb{R}}|Y(t)|<\infty H ( g ) := sup t ∈ R ∣ Y ( t ) ∣ < ∞ , and is T T T -periodic if g ( t + T ) = g ( t ) g(t+T)=g(t) g ( t + T ) = g ( t ) for all t ∈ R t\in\mathbb{R} t ∈ R (which implies H ( g ) < ∞ H(g)<\infty H ( g ) < ∞ ).
A (one-way upward) transit trajectory is an absolutely continuous path p = ( x , y ) : R → C b p=(x,y):\mathbb{R}\to C_b p = ( x , y ) : R → C b satisfying ess sup t x ′ ( t ) 2 + y ′ ( t ) 2 ≤ U \operatorname*{ess\,sup}_{t}\sqrt{x'(t)^2+y'(t)^2}\le U ess sup t x ′ ( t ) 2 + y ′ ( t ) 2 ≤ U that traverses the corridor upward in Lab coordinates: lim t → − ∞ y ( t ) = − ∞ \lim_{t\to-\infty}y(t)=-\infty lim t → − ∞ y ( t ) = − ∞ and lim t → + ∞ y ( t ) = + ∞ \lim_{t\to+\infty}y(t)=+\infty lim t → + ∞ y ( t ) = + ∞ . (When H ( g ) ≤ H < ∞ H(g)\le H<\infty H ( g ) ≤ H < ∞ , any finite-horizon segment p : [ t 0 , t 1 ] → C b p:[t_0,t_1]\to C_b p : [ t 0 , t 1 ] → C b with y ( t 0 ) ≤ − H − 2 r y(t_0)\le-H-2r y ( t 0 ) ≤ − H − 2 r and y ( t 1 ) ≥ + H + 2 r y(t_1)\ge+H+2r y ( t 1 ) ≥ + H + 2 r extends outside [ t 0 , t 1 ] [t_0,t_1] [ t 0 , t 1 ] at velocity ( 0 , + U ) (0,+U) ( 0 , + U ) without meeting B r ‾ ( g ( t ) ) \overline{B_r}(g(t)) B r ( g ( t )) , so finite-zone and full Lab transit are equivalent.)
Closed-disk capture: The patrol g g g captures p p p if there exists t ∈ R t\in\mathbb{R} t ∈ R such that
∥ p ( t ) − g ( t ) ∥ 2 = ( x ( t ) − X ( t ) ) 2 + ( y ( t ) − Y ( t ) ) 2 ≤ r . \|p(t)-g(t)\|_2=\sqrt{(x(t)-X(t))^2+(y(t)-Y(t))^2}\le r. ∥ p ( t ) − g ( t ) ∥ 2 = ( x ( t ) − X ( t ) ) 2 + ( y ( t ) − Y ( t ) ) 2 ≤ r . The patrol g g g successfully guards C b C_b C b if it captures every admissible upward transit trajectory p p p .
Definition 1.2 (Minimax Guarded Corridor Widths). For q = U / V > 0 q=U/V>0 q = U / V > 0 and H ≥ 0 H\ge0 H ≥ 0 , define the excursion-constrained minimax width
W o p t ( q , H ) : = sup { 2 b > 0 : ∃ patrol g on C b with sup t ∈ R ∣ Y ( t ) ∣ ≤ H guarding C b } , W_{\mathrm{opt}}(q,H):=\sup\{2b>0:\exists\text{ patrol }g\text{ on }C_b\text{ with }\sup_{t\in\mathbb{R}}|Y(t)|\le H\text{ guarding }C_b\}, W opt ( q , H ) := sup { 2 b > 0 : ∃ patrol g on C b with t ∈ R sup ∣ Y ( t ) ∣ ≤ H guarding C b } , and the global bounded-excursion minimax width W o p t ( q ) : = sup H ≥ 0 W o p t ( q , H ) W_{\mathrm{opt}}(q):=\sup_{H\ge0}W_{\mathrm{opt}}(q,H) W opt ( q ) := sup H ≥ 0 W opt ( q , H ) . Define W 2 w a y ( q , H ) W_{\mathrm{2way}}(q,H) W 2way ( q , H ) and W 2 w a y ( q ) W_{\mathrm{2way}}(q) W 2way ( q ) analogously when g g g must simultaneously block both upward ( y ( ∓ ∞ ) = ∓ ∞ ) (y(\mp\infty)=\mp\infty) ( y ( ∓ ∞ ) = ∓ ∞ ) and downward ( y ( ∓ ∞ ) = ± ∞ ) (y(\mp\infty)=\pm\infty) ( y ( ∓ ∞ ) = ± ∞ ) speed-U U U intruders. Because any speed-q 2 q_2 q 2 trajectory is admissible at speed q 1 ≥ q 2 q_1\ge q_2 q 1 ≥ q 2 , W o p t ( q ) W_{\mathrm{opt}}(q) W opt ( q ) is nonincreasing in q q q ; W o p t ( q , H ) W_{\mathrm{opt}}(q,H) W opt ( q , H ) is nondecreasing in H H H . The feasible widths are downward closed: projecting a wider patrol horizontally onto a narrower strip preserves the speed bound and decreases its distance from every point of that strip.
if g g g guards C b C_b C b with closed disk B ‾ r \overline{B}_r B r , then for every ε ∈ ( 0 , 1 ) \varepsilon\in(0,1) ε ∈ ( 0 , 1 ) the contracted trajectory g ε ( t ) : = ( 1 − ε ) g ( t / ( 1 − ε ) ) g_\varepsilon(t) := (1-\varepsilon)g(t/(1-\varepsilon)) g ε ( t ) := ( 1 − ε ) g ( t / ( 1 − ε )) has speed ≤ V \le V ≤ V and guards C ( 1 − ε ) b C_{(1-\varepsilon)b} C ( 1 − ε ) b with closed radius ( 1 − ε ) r < r (1-\varepsilon)r < r ( 1 − ε ) r < r (hence with open disk B r B_r B r ). Letting ε ↓ 0 \varepsilon\downarrow0 ε ↓ 0 shows that open-disk and closed-disk suprema have the exact same numerical value W o p t ( q ) W_{\mathrm{opt}}(q) W opt ( q ) .
One-way vs. two-way transit: Definition 1.1 specifies known-direction (upward) transit. In the two-way variant (§ 7.1 ), a single patrol g g g blocks both directions; transverse sweeping ((1) ) is direction-symmetric, whereas tilted bowties exploit directional asymmetry.
Results
For 1 / 3 < q < 1 1/3 < q < 1 1/3 < q < 1 , just two families of straight intruders, one along each wall, force the width below or equal to
F ( q ) = 2 r 1 + ( 3 q 2 + 2 q − 1 ) − 1 . F(q) = 2r\sqrt{1 + (3q^2 + 2q - 1)^{-1}}. F ( q ) = 2 r 1 + ( 3 q 2 + 2 q − 1 ) − 1 .
The patrol that uses this wall budget most efficiently descends along an inset segment on each side and rises on each diagonal crossing. Its angle with the transverse direction satisfies sin α = ( 1 − q ) / ( 2 q ) \sin\alpha= (1-q)/(2q) sin α = ( 1 − q ) / ( 2 q ) . At q = 1 / 2 q = 1/2 q = 1/2 it guards width 2 r 7 / 3 2r\sqrt{7/3} 2 r 7/3 , exceeding the transverse value 2 r 2 2r\sqrt{2} 2 r 2 . For q a ≤ q < 1 q_a \le q < 1 q a ≤ q < 1 a moving separator certifies this bowtie against every admissible intruder (Theorem 3.6 ). For q ≥ 1 q \ge1 q ≥ 1 the transverse value is exact (Theorem 6.1 ).
The distinction between failure of one patrol and a strict bound on the supremum matters below q a q_a q a . An escaping intruder against the wall-optimal bowtie alone would leave open the possibility of other patrols approaching the wall ceiling. Theorems 5.1 and 5.2 exclude that possibility even for sequences with increasing periods or excursions. Near the ceiling, a patrol must devote almost all its time to the wall contacts and crossings of a bowtie. The proof finds consecutive crossings where this is true of the original patrol in physical time, and where visits from outside that time interval cannot intercept the evader. The resulting gap has quadratic order at q a q_a q a . Theorem 4.1 allows the period to vary in taking the supremum.
For two-way transit the transverse sweep is optimal already at q = 1 / 2 q = 1/2 q = 1/2 and gains are possible below this speed (§ 7.1 ). Exact rational certificates give quantitative one-way brackets at lower speeds (§ 8 ). As q ↓ 0 q \downarrow0 q ↓ 0 , the one-way optimum is asymptotic to 2 r / q 2r/q 2 r / q (Theorem 9.1 ). For q ≤ 1 / 40 q \le1/40 q ≤ 1/40 an explicit ten-leg patrol keeps the difference from 2 r / q 2r/q 2 r / q below 1.235353 r 1.235353r 1.235353 r (Theorem 9.2 ).
The straight-wall ceiling
In this section normalize V = 1 V = 1 V = 1 , take 1 / 3 < q < 1 1/3 < q < 1 1/3 < q < 1 , and set
F ( q ) : = 2 r 1 + 1 3 q 2 + 2 q − 1 . (3) F(q) := 2r\sqrt{1 + \frac{1}{3q^2 + 2q - 1}}. \tag*{(3)} F ( q ) := 2 r 1 + 3 q 2 + 2 q − 1 1 . (3)
Notice 3 q 2 + 2 q − 1 = ( 3 q − 1 ) ( q + 1 ) > 0 3q^2 + 2q - 1 = (3q - 1)(q + 1) > 0 3 q 2 + 2 q − 1 = ( 3 q − 1 ) ( q + 1 ) > 0 precisely when q > 1 / 3 q > 1/3 q > 1/3 , and F ( q ) > 2 r F(q) > 2r F ( q ) > 2 r for q ∈ ( 1 / 3 , 1 ) q \in(1/3,1) q ∈ ( 1/3 , 1 ) .
Theorem 2.1 (Wall-phase ceiling). Let 1 / 3 < q < 1 1/3 < q < 1 1/3 < q < 1 , and let g = ( X , Y ) : R → [ − b , b ] × R g = (X,Y) : \mathbb{R} \to[-b,b] \times\mathbb{R} g = ( X , Y ) : R → [ − b , b ] × R be any unit-speed patrol with bounded longitudinal excursion sup t ∈ R ∣ Y ( t ) ∣ ≤ H < ∞ \sup_{t \in\mathbb{R}} |Y(t)| \le H < \infty sup t ∈ R ∣ Y ( t ) ∣ ≤ H < ∞ (including every periodic patrol). If g g g blocks all upward straight-line wall runners p ± , ζ ( t ) = ( ± b , q t − ζ ) p_{\pm,\zeta}(t) = (\pm b, qt-\zeta) p ± , ζ ( t ) = ( ± b , q t − ζ ) (ζ ∈ R \zeta\in\mathbb{R} ζ ∈ R ), then W = 2 b ≤ F ( q ) W = 2b \le F(q) W = 2 b ≤ F ( q ) .
Rather than summing potentially overlapping phase spans from arbitrary visits, we map the patrol onto a Euclidean phase cylinder, uncross self-intersections without increasing Euclidean length, and extract nondecreasing lifted contact order by Jordan separation.
Figure 1. Left: the four-leg optimal patrol at q = 1 / 2 q = 1/2 q = 1/2 . Both vertical legs point down and both diagonals point up. Right: one-way widths. The solid curve gives the optimum for q ≥ q a q \ge q_a q ≥ q a . The dashed gray curve gives the straight-wall ceiling below q a q_a q a , where the strict gap theorem applies. Vertical bars show the certified lower and upper bounds at q = 1 / 4 , 1 / 3 , 37 / 100 q = 1/4, 1/3, 37/100 q = 1/4 , 1/3 , 37/100 from § 8 .
Lemma 2.2 (Phase-Cylinder Speed Budget and Inverse Parameterization). Fix q ∈ ( 0 , 1 ) q \in(0,1) q ∈ ( 0 , 1 ) , α : = 1 − q 2 \alpha:= \sqrt{1-q^2} α := 1 − q 2 , half-width b > 0 b > 0 b > 0 , and period T > 0 T > 0 T > 0 with P : = q T P := qT P := q T .
For any T T T -periodic unit-speed patrol g = ( X , Y ) g = (X,Y) g = ( X , Y ) in [ − b , b ] × R [-b,b] \times\mathbb{R} [ − b , b ] × R , the curve Γ ( t ) : = ( α X ( t ) , q t − Y ( t ) ) \Gamma(t) := (\alpha X(t), qt - Y(t)) Γ ( t ) := ( α X ( t ) , q t − Y ( t )) projects to a winding-number-+ 1 +1 + 1 loop on the cylinder [ − α b , α b ] × ( R / P Z ) [-\alpha b,\alpha b] \times(\mathbb{R}/P\mathbb{Z}) [ − α b , α b ] × ( R / P Z ) of Euclidean length L ( Γ ) ≤ T = P / q L(\Gamma) \le T = P/q L ( Γ ) ≤ T = P / q , and g g g detects the straight wall runner ( ± b , q t − ζ ) (\pm b,qt-\zeta) ( ± b , q t − ζ ) iff Γ \Gamma Γ enters the elliptical disk
( u ∓ α b ) 2 α 2 + min k ∈ Z ( z − ζ − k P ) 2 ≤ r 2 . \frac{(u \mp\alpha b)^2}{\alpha^2} + \min_{k \in\mathbb{Z}}(z-\zeta-kP)^2 \le r^2. α 2 ( u ∓ α b ) 2 + k ∈ Z min ( z − ζ − k P ) 2 ≤ r 2 . Conversely, every rectifiable winding-+ 1 +1 + 1 loop Γ 0 ( s ) = ( u ( s ) , z ( s ) ) \Gamma_0(s) = (u(s),z(s)) Γ 0 ( s ) = ( u ( s ) , z ( s )) (s ∈ [ 0 , L ] s \in[0,L] s ∈ [ 0 , L ] parameterized by Euclidean arclength, z ( L ) = z ( 0 ) + P z(L) = z(0) + P z ( L ) = z ( 0 ) + P , ∣ u ( s ) ∣ ≤ α b |u(s)| \le\alpha b ∣ u ( s ) ∣ ≤ α b ) with length L ≤ P / q L \le P/q L ≤ P / q lifts to a T T T -periodic unit-speed patrol g 0 g_0 g 0 (T = P / q T = P/q T = P / q ) tracing the exact same cylinder image.
Proof. For (1), since X ′ 2 + Y ′ 2 ≤ 1 X'^2 + Y'^2 \le1 X ′2 + Y ′2 ≤ 1 a.e. and 1 − q Y ′ ≥ 1 − q > 0 1-qY' \ge1-q > 0 1 − q Y ′ ≥ 1 − q > 0 , the pointwise algebraic identity
( 1 − q Y ′ ) 2 − [ α 2 X ′ 2 + ( q − Y ′ ) 2 ] = ( 1 − q 2 ) ( 1 − X ′ 2 − Y ′ 2 ) ≥ 0 (1-qY')^2 - [\alpha^2X'^2 + (q-Y')^2] = (1-q^2)(1-X'^2-Y'^2) \ge0 ( 1 − q Y ′ ) 2 − [ α 2 X ′2 + ( q − Y ′ ) 2 ] = ( 1 − q 2 ) ( 1 − X ′2 − Y ′2 ) ≥ 0
yields ∥ Γ ′ ( t ) ∥ 2 = α 2 X ′ 2 + ( q − Y ′ ) 2 ≤ 1 − q Y ′ ( t ) \|\Gamma'(t)\|_2 = \sqrt{\alpha^2X'^2 + (q-Y')^2} \le1-qY'(t) ∥ Γ ′ ( t ) ∥ 2 = α 2 X ′2 + ( q − Y ′ ) 2 ≤ 1 − q Y ′ ( t ) a.e. Integrating over [ 0 , T ] [0,T] [ 0 , T ] gives L ( Γ ) ≤ T − q ( Y ( T ) − Y ( 0 ) ) = T = P / q L(\Gamma) \le T-q(Y(T)-Y(0))=T=P/q L ( Γ ) ≤ T − q ( Y ( T ) − Y ( 0 )) = T = P / q , and ( X ( t ) ∓ b ) 2 + ( q t − ζ − Y ( t ) ) 2 ≤ r 2 (X(t)\mp b)^2+(qt-\zeta-Y(t))^2\le r^2 ( X ( t ) ∓ b ) 2 + ( q t − ζ − Y ( t ) ) 2 ≤ r 2 translates directly into the stated ellipse condition.
For (2), since u ′ ( s ) 2 + z ′ ( s ) 2 = 1 u'(s)^2+z'(s)^2=1 u ′ ( s ) 2 + z ′ ( s ) 2 = 1 a.e. and q ∈ ( 0 , 1 ) q\in(0,1) q ∈ ( 0 , 1 ) , the minimum time-rate density τ min ( s ) : = 1 − q z ′ ( s ) 1 − q 2 ≥ 1 − q 1 − q 2 = 1 1 + q > 0 \tau_{\min}(s):=\frac{1-qz'(s)}{1-q^2}\ge\frac{1-q}{1-q^2}=\frac{1}{1+q}>0 τ m i n ( s ) := 1 − q 2 1 − q z ′ ( s ) ≥ 1 − q 2 1 − q = 1 + q 1 > 0 satisfies ∫ 0 L τ min ( s ) d s = L − q P 1 − q 2 ≤ P / q − q P 1 − q 2 = P q = T \int_0^L\tau_{\min}(s)\,ds=\frac{L-qP}{1-q^2}\le\frac{P/q-qP}{1-q^2}=\frac{P}{q}=T ∫ 0 L τ m i n ( s ) d s = 1 − q 2 L − q P ≤ 1 − q 2 P / q − q P = q P = T . Setting t ′ ( s ) : = τ min ( s ) + c 0 > 0 t'(s):=\tau_{\min}(s)+c_0>0 t ′ ( s ) := τ m i n ( s ) + c 0 > 0 with constant c 0 : = T − ∫ 0 L τ min L ≥ 0 c_0:=\frac{T-\int_0^L\tau_{\min}}{L}\ge0 c 0 := L T − ∫ 0 L τ m i n ≥ 0 defines a bi-Lipschitz homeomorphism t : [ 0 , L ] → [ 0 , T ] t:[0,L]\to[0,T] t : [ 0 , L ] → [ 0 , T ] . Setting X ( t ( s ) ) : = u ( s ) / α X(t(s)):=u(s)/\alpha X ( t ( s )) := u ( s ) / α and Y ( t ( s ) ) : = q t ( s ) − z ( s ) Y(t(s)):=qt(s)-z(s) Y ( t ( s )) := q t ( s ) − z ( s ) gives Y ( T ) = Y ( 0 ) Y(T)=Y(0) Y ( T ) = Y ( 0 ) , X ′ ( t ) = u ′ ( s ) α t ′ ( s ) X'(t)=\frac{u'(s)}{\alpha t'(s)} X ′ ( t ) = α t ′ ( s ) u ′ ( s ) , and Y ′ ( t ) = q − z ′ ( s ) t ′ ( s ) Y'(t)=q-\frac{z'(s)}{t'(s)} Y ′ ( t ) = q − t ′ ( s ) z ′ ( s ) . Using u ′ ( s ) 2 = 1 − z ′ ( s ) 2 u'(s)^2=1-z'(s)^2 u ′ ( s ) 2 = 1 − z ′ ( s ) 2 and t ′ ( s ) = τ min ( s ) + c 0 t'(s)=\tau_{\min}(s)+c_0 t ′ ( s ) = τ m i n ( s ) + c 0 , direct expansion yields
α 2 t ′ ( s ) 2 ( 1 − X ′ ( t ) 2 − Y ′ ( t ) 2 ) = α 2 c 0 ( 2 + α 2 c 0 ) ≥ 0 \alpha^2t'(s)^2(1-X'(t)^2-Y'(t)^2)=\alpha^2c_0(2+\alpha^2c_0)\ge0 α 2 t ′ ( s ) 2 ( 1 − X ′ ( t ) 2 − Y ′ ( t ) 2 ) = α 2 c 0 ( 2 + α 2 c 0 ) ≥ 0
almost everywhere, since c 0 ≥ 0 c_0\ge0 c 0 ≥ 0 and α 2 > 0 \alpha^2>0 α 2 > 0 . Thus X ′ ( t ) 2 + Y ′ ( t ) 2 ≤ 1 X'(t)^2+Y'(t)^2\le1 X ′ ( t ) 2 + Y ′ ( t ) 2 ≤ 1 a.e. □ \square □
Lemma 2.3 (Cylindrical Uncrossing with Arbitrarily Small Coverage Loss). Let Γ \Gamma Γ be a rectifiable loop of winding number + 1 +1 + 1 and Euclidean length L L L in [ − α b , α b ] × ( R / P Z ) [-\alpha b,\alpha b] \times(\mathbb{R}/P\mathbb{Z}) [ − α b , α b ] × ( R / P Z ) with b > r b > r b > r , whose elliptical neighborhoods of radius r r r cover both wall circles { ± α b } × ( R / P Z ) \{\pm\alpha b\} \times(\mathbb{R}/P\mathbb{Z}) { ± α b } × ( R / P Z ) . Then for every ε > 0 \varepsilon> 0 ε > 0 , there exists a simple (embedded) winding-+ 1 +1 + 1 loop Γ ~ \widetilde{\Gamma} Γ in [ − α b , α b ] × ( R / P Z ) [-\alpha b,\alpha b] \times(\mathbb{R}/P\mathbb{Z}) [ − α b , α b ] × ( R / P Z ) of Euclidean length at most L L L whose elliptical neighborhoods of radius r + ε r + \varepsilon r + ε cover both wall circles.
Proof. First compress horizontally via u ↦ ( 1 − η ) u u \mapsto(1-\eta)u u ↦ ( 1 − η ) u for small η ∈ ( 0 , 1 ) \eta\in(0,1) η ∈ ( 0 , 1 ) . This moves points by at most η b \eta b η b in the physical detection metric of Lemma 2.2 (1), leaves a positive boundary margin η α b \eta\alpha b η α b inside [ − α b , α b ] [-\alpha b,\alpha b] [ − α b , α b ] , and strictly reduces Euclidean length: since b > r b > r b > r , covering both walls forces positive horizontal variation ∫ ∣ u ′ ∣ d s > 0 \int|u'|\,\mathrm{d}s > 0 ∫ ∣ u ′ ∣ d s > 0 , so the line element strictly shrinks on a set of positive measure. Next, approximate the compressed curve by an inscribed polygon and then perturb its vertices within the strict length and boundary margins. This gives a generic polygonal immersion with the same winding number + 1 +1 + 1 , transverse double crossings only (no triple points or overlapping edges), length strictly less than L L L , image inside [ − α b , α b ] [-\alpha b,\alpha b] [ − α b , α b ] , and Hausdorff error small enough that η b \eta b η b plus polygonal error is less than ε / 2 \varepsilon/2 ε /2 .
Place pairwise disjoint disks of radius δ \delta δ with 2 δ / α < ε / 2 2\delta/\alpha< \varepsilon/2 2 δ / α < ε /2 inside ( − α b , α b ) × ( R / P Z ) (-\alpha b,\alpha b) \times(\mathbb{R}/P\mathbb{Z}) ( − α b , α b ) × ( R / P Z ) around the finitely many double points, avoiding all vertices and non-incident edges. Inside each disk, replacing the four radial half-edges by either of the two noncrossing adjacent chord pairs does not increase Euclidean length (by the triangle inequality) and keeps every replaced point within 2 δ 2\delta 2 δ of a new chord. Any choice of noncrossing pairings decomposes the graph into finitely many pairwise disjoint simple closed curves. If more than one component results, connectedness of the original immersion implies that two distinct components meet at some common crossing disk; switching the noncrossing pairing inside that single disk merges those two components into one without introducing intersections (compare the A-trail formulation in [6 ] ). Iterating yields a single simple closed curve. Because local smoothing preserves mod-2 homology on the cylinder, the resulting simple loop has odd winding number; since every simple closed curve on a cylinder has winding number in { 0 , ± 1 } \{0,\pm1\} { 0 , ± 1 } , orienting it upward gives winding number + 1 +1 + 1 , length ≤ L \le L ≤ L , and wall coverage at radius r + ε r+\varepsilon r + ε . □
Proof of Theorem 2.1 . Assume b > r b > r b > r (otherwise 2 b ≤ 2 r < F ( q ) 2b \le2r < F(q) 2 b ≤ 2 r < F ( q ) ).
Step 1: Periodic patrols. Suppose g g g is T T T -periodic. Fix 0 < ε < b − r 0 < \varepsilon< b-r 0 < ε < b − r . Applying Lemma 2.2 (1), Lemma 2.3 , and Lemma 2.2 (2) yields a T T T -periodic unit-speed patrol g ~ = ( X ~ , Y ~ ) \widetilde{g} = (\widetilde{X},\widetilde{Y}) g = ( X , Y ) in [ − b , b ] × R [-b,b] \times\mathbb{R} [ − b , b ] × R whose phase-cylinder curve Γ ~ \widetilde{\Gamma} Γ is simple and covers both wall circles with capture radius r ε : = r + ε < b r_\varepsilon:= r+\varepsilon< b r ε := r + ε < b . Lift Γ ~ \widetilde{\Gamma} Γ to an infinite simple curve Γ ^ \widehat{\Gamma} Γ in [ − α b , α b ] × R [-\alpha b,\alpha b] \times\mathbb{R} [ − α b , α b ] × R . For each lifted phase z ∈ R z \in\mathbb{R} z ∈ R , let q + ( z ) q_+(z) q + ( z ) be the rightmost point of Γ ^ ∩ ( ( − α b , α b ) × { z } ) \widehat{\Gamma} \cap((-\alpha b,\alpha b) \times\{z\}) Γ ∩ (( − α b , α b ) × { z }) and q − ( z ) q_-(z) q − ( z ) the leftmost; retaining q ± ( z ) q_\pm(z) q ± ( z ) preserves all wall coverage because horizontal distance to ± α b \pm\alpha b ± α b only decreases at fixed phase z z z . Along the oriented simple curve Γ ^ \widehat{\Gamma} Γ , the selected points { q + ( z ) } \{q_+(z)\} { q + ( z )} (and likewise { q − ( z ) } ) \{q_-(z)\}) { q − ( z )}) appear in nondecreasing lifted phase order : horizontal segments from q + ( z 1 ) , q + ( z 2 ) q_+(z_1),q_+(z_2) q + ( z 1 ) , q + ( z 2 ) (z 1 < z 2 z_1<z_2 z 1 < z 2 ) to a vertical line { u = − α b + 1 } \{u=-\alpha b+1\} { u = − α b + 1 } are disjoint crosscuts of the strip to the right of Γ ^ \widehat{\Gamma} Γ , so one cutting along one crosscut forms a topological rectangle whose opposite sides must share endpoint order by Jordan separation. More explicitly, truncate the lift between two horizontal lines far below and above the chosen phase levels, taking the last visit to the bottom line before the two contacts and the first visit to the top line after them. Properness and the limits z ( t ) → ± ∞ z(t)\to\pm\infty z ( t ) → ± ∞ at the two ends allow the bottom line to lie below all phases after the earlier contact and the top line above all phases before the later contact. Thus the truncated arc stays between the two lines. The resulting simple crosscut joins the bottom to the top of a rectangle and the rightward horizontal connectors miss the crosscut except at their endpoints. Disjoint connectors preserve the boundary order along its right-hand component. The left wall is identical. Passing to closures preserves this weak order.
Take the closure of the rightmost selections satisfying X ~ ≥ b − r ε \widetilde{X} \ge b-r_\varepsilon X ≥ b − r ε , and the closure of the leftmost selections satisfying X ~ ≤ − b + r ε \widetilde{X} \le-b+r_{\varepsilon} X ≤ − b + r ε ; because b > r ε b>r_{\varepsilon} b > r ε , the left and right zones are separated by horizontal gap 2 ( b − r ε ) > 0 2(b-r_{\varepsilon})>0 2 ( b − r ε ) > 0 . By rectifiability, one period contains finitely many alternating left- and right-wall contact groups (2 N 2N 2 N groups total for some integer N ≥ 1 N\ge1 N ≥ 1 ). In group j ∈ { 1 , … , 2 N } j\in\{1,\ldots,2N\} j ∈ { 1 , … , 2 N } , let u j , v j u_j,v_j u j , v j be points of the group attaining the maximum of p + R ε p+R_{\varepsilon} p + R ε and minimum of p − R ε p-R_{\varepsilon} p − R ε respectively, where p = q t − Y ~ p=qt-\widetilde{Y} p = q t − Y is the center phase and R ε ( X ~ ) = r ε 2 − ( b − ∣ X ~ ∣ ) 2 R_{\varepsilon}(\widetilde{X})=\sqrt{r_{\varepsilon}^{2}-(b-\lvert\widetilde{X}\rvert)^{2}} R ε ( X ) = r ε 2 − ( b − ∣ X ∣ ) 2 . We claim v j v_j v j can always be chosen no later along Γ ~ \widetilde{\Gamma} Γ than u j u_j u j : if u j u_j u j precedes v j v_j v j , nondecreasing lifted phase order gives p ( u j ) ≤ p ( v j ) p(u_j)\le p(v_j) p ( u j ) ≤ p ( v j ) , while extremality gives p ( v j ) + R ε ( v j ) ≤ p ( u j ) + R ε ( u j ) p(v_j)+R_{\varepsilon}(v_j)\le p(u_j)+R_{\varepsilon}(u_j) p ( v j ) + R ε ( v j ) ≤ p ( u j ) + R ε ( u j ) and p ( u j ) − R ε ( u j ) ≥ p ( v j ) − R ε ( v j ) p(u_j)-R_{\varepsilon}(u_j)\ge p(v_j)-R_{\varepsilon}(v_j) p ( u j ) − R ε ( u j ) ≥ p ( v j ) − R ε ( v j ) , i.e., R ε ( u j ) − R ε ( v j ) ≥ p ( v j ) − p ( u j ) ≥ 0 R_{\varepsilon}(u_j)-R_{\varepsilon}(v_j)\ge p(v_j)-p(u_j)\ge0 R ε ( u j ) − R ε ( v j ) ≥ p ( v j ) − p ( u j ) ≥ 0 and R ε ( v j ) − R ε ( u j ) ≥ p ( v j ) − p ( u j ) ≥ 0 R_{\varepsilon}(v_j)-R_{\varepsilon}(u_j)\ge p(v_j)-p(u_j)\ge0 R ε ( v j ) − R ε ( u j ) ≥ p ( v j ) − p ( u j ) ≥ 0 . Hence p ( u j ) = p ( v j ) p(u_j)=p(v_j) p ( u j ) = p ( v j ) and R ε ( u j ) = R ε ( v j ) R_{\varepsilon}(u_j)=R_{\varepsilon}(v_j) R ε ( u j ) = R ε ( v j ) , so swapping ( u j , v j ) (u_j,v_j) ( u j , v j ) puts the minimizer before the maximizer.
Thus each group j ∈ { 1 , … , 2 N } j\in\{1,\ldots,2N\} j ∈ { 1 , … , 2 N } has chronologically ordered marker times t j − ≤ t j + t_j^{-}\le t_j^{+} t j − ≤ t j + covering phase length q ( t j + − t j − ) + ( Y ~ ( t j − ) − Y ~ ( t j + ) ) + R ε ( t j − ) + R ε ( t j + ) q(t_j^{+}-t_j^{-})+(\widetilde{Y}(t_j^{-})-\widetilde{Y}(t_j^{+}))+R_{\varepsilon}(t_j^{-})+R_{\varepsilon}(t_j^{+}) q ( t j + − t j − ) + ( Y ( t j − ) − Y ( t j + )) + R ε ( t j − ) + R ε ( t j + ) . Let A : = ∑ j = 1 2 N ( t j + − t j − ) ≥ 0 A:=\sum_{j=1}^{2N}(t_j^{+}-t_j^{-})\ge0 A := ∑ j = 1 2 N ( t j + − t j − ) ≥ 0 , B : = T − A = ∑ j = 1 2 N ( t j + 1 − − t j + ) ≥ 0 B:=T-A=\sum_{j=1}^{2N}(t_{j+1}^{-}-t_j^{+})\ge0 B := T − A = ∑ j = 1 2 N ( t j + 1 − − t j + ) ≥ 0 (with t 2 N + 1 − : = t 1 − + T t_{2N+1}^{-}:=t_1^{-}+T t 2 N + 1 − := t 1 − + T ), Z : = ∑ j = 1 2 N ( Y ~ ( t j − ) − Y ~ ( t j + ) ) Z:=\sum_{j=1}^{2N}(\widetilde{Y}(t_j^{-})-\widetilde{Y}(t_j^{+})) Z := ∑ j = 1 2 N ( Y ( t j − ) − Y ( t j + )) , and a : = 1 4 N ∑ j = 1 2 N ( ∣ X ~ ( t j − ) ∣ + ∣ X ~ ( t j + ) ∣ ) ∈ [ b − r ε , b ] a:=\frac{1}{4N}\sum_{j=1}^{2N}(\lvert\widetilde{X}(t_j^{-})\rvert+\lvert\widetilde{X}(t_j^{+})\rvert)\in[b-r_{\varepsilon},b] a := 4 N 1 ∑ j = 1 2 N (∣ X ( t j − )∣ + ∣ X ( t j + )∣) ∈ [ b − r ε , b ] . Concavity of d ↦ r ε 2 − d 2 d\mapsto\sqrt{r_{\varepsilon}^{2}-d^{2}} d ↦ r ε 2 − d 2 bounds the sum of the 4 N 4N 4 N marker radii by 4 N R 4NR 4 N R , where R : = r ε 2 − ( b − a ) 2 R:=\sqrt{r_{\varepsilon}^{2}-(b-a)^{2}} R := r ε 2 − ( b − a ) 2 . On each inter-group transit [ t j + , t j + 1 − ] [t_j^{+},t_{j+1}^{-}] [ t j + , t j + 1 − ] , X ~ ( t j + ) \widetilde{X}(t_j^{+}) X ( t j + ) and X ~ ( t j + 1 − ) \widetilde{X}(t_{j+1}^{-}) X ( t j + 1 − ) have opposite signs, so horizontal travel is ∣ X ~ ( t j + ) ∣ + ∣ X ~ ( t j + 1 − ) ∣ \lvert\widetilde{X}(t_j^{+})\rvert+\lvert\widetilde{X}(t_{j+1}^{-})\rvert ∣ X ( t j + )∣ + ∣ X ( t j + 1 − )∣ , summing over j = 1 , … , 2 N j=1,\ldots,2N j = 1 , … , 2 N to 4 N a 4Na 4 N a , while vertical shifts telescope to ∑ j = 1 2 N ( Y ~ ( t j + 1 − ) − Y ~ ( t j + ) ) = Z \sum_{j=1}^{2N}(\widetilde{Y}(t_{j+1}^{-})-\widetilde{Y}(t_j^{+}))=Z ∑ j = 1 2 N ( Y ( t j + 1 − ) − Y ( t j + )) = Z . By the Euclidean triangle inequality over [ t j + , t j + 1 − ] [t_j^{+},t_{j+1}^{-}] [ t j + , t j + 1 − ] and unit speed on [ t j − , t j + ] [t_j^{-},t_j^{+}] [ t j − , t j + ] , full coverage of both length-q T qT q T wall circles implies
2 q T ≤ q A + Z + 4 N R , A ≥ ∣ Z ∣ , T − A = B ≥ ( 4 N a ) 2 + Z 2 . (4) 2qT\le qA+Z+4NR,\qquad A\ge\lvert Z\rvert,\qquad T-A=B\ge\sqrt{(4Na)^{2}+Z^{2}}. \tag*{(4)} 2 q T ≤ q A + Z + 4 N R , A ≥ ∣ Z ∣ , T − A = B ≥ ( 4 N a ) 2 + Z 2 . (4)
Since 2 q T − q A − Z = q A + 2 q B − Z ≥ q ∣ Z ∣ + 2 q ( 4 N a ) 2 + Z 2 − Z 2qT-qA-Z=qA+2qB-Z\ge q\lvert Z\rvert+2q\sqrt{(4Na)^{2}+Z^{2}}-Z 2 q T − q A − Z = q A + 2 q B − Z ≥ q ∣ Z ∣ + 2 q ( 4 N a ) 2 + Z 2 − Z , setting h : = Z / ( 4 N ) ∈ R h:=Z/(4N)\in\mathbb{R} h := Z / ( 4 N ) ∈ R and dividing by 4 N 4N 4 N yields
2 q a 2 + h 2 + q ∣ h ∣ − h ≤ R . 2q\sqrt{a^{2}+h^{2}}+q\lvert h\rvert-h\le R. 2 q a 2 + h 2 + q ∣ h ∣ − h ≤ R .
If h < 0 h<0 h < 0 , replacing h h h by 0 0 0 strictly decreases the left-hand side; for h ≥ 0 h\ge0 h ≥ 0 , ∣ h ∣ = h \lvert h\rvert=h ∣ h ∣ = h gives Φ q ( a , h ) : = 2 q a 2 + h 2 − ( 1 − q ) h ≤ R \Phi_q(a,h):=2q\sqrt{a^{2}+h^{2}}-(1-q)h\le R Φ q ( a , h ) := 2 q a 2 + h 2 − ( 1 − q ) h ≤ R . Because 0 < 1 − q < 2 q 0<1-q<2q 0 < 1 − q < 2 q when 1 / 3 < q < 1 1/3<q<1 1/3 < q < 1 , Φ q ( a , ⋅ ) \Phi_q(a,\cdot) Φ q ( a , ⋅ ) is strictly convex on [ 0 , ∞ ) [0,\infty) [ 0 , ∞ ) and attains its global minimum at
h ∗ ( a ) = ( 1 − q ) a 3 q 2 + 2 q − 1 , min h ≥ 0 Φ q ( a , h ) = a 3 q 2 + 2 q − 1 . (5) h_*(a)=\frac{(1-q)a}{\sqrt{3q^{2}+2q-1}},\qquad\min_{h\ge0}\Phi_q(a,h)=a\sqrt{3q^{2}+2q-1}. \tag*{(5)} h ∗ ( a ) = 3 q 2 + 2 q − 1 ( 1 − q ) a , h ≥ 0 min Φ q ( a , h ) = a 3 q 2 + 2 q − 1 . (5)
Writing c : = 3 q 2 + 2 q − 1 > 0 c:=3q^{2}+2q-1>0 c := 3 q 2 + 2 q − 1 > 0 , a c ≤ r ε 2 − ( b − a ) 2 a\sqrt{c}\le\sqrt{r_{\varepsilon}^{2}-(b-a)^{2}} a c ≤ r ε 2 − ( b − a ) 2 gives c a 2 + ( b − a ) 2 ≤ r ε 2 ca^{2}+(b-a)^{2}\le r_{\varepsilon}^{2} c a 2 + ( b − a ) 2 ≤ r ε 2 . By Cauchy–Schwarz on ( 1 / c , 1 ) ⋅ ( c a , b − a ) (1/\sqrt{c},1)\cdot(\sqrt{c}a,b-a) ( 1/ c , 1 ) ⋅ ( c a , b − a ) , b = a + ( b − a ) ≤ r ε 1 + c − 1 b=a+(b-a)\le r_{\varepsilon}\sqrt{1+c^{-1}} b = a + ( b − a ) ≤ r ε 1 + c − 1 . Letting ε ↓ 0 \varepsilon\downarrow0 ε ↓ 0 proves 2 b ≤ 2 r 1 + c − 1 = F ( q ) 2b\le2r\sqrt{1+c^{-1}}=F(q) 2 b ≤ 2 r 1 + c − 1 = F ( q ) .
Step 2: Nonperiodic bounded-excursion patrols. Suppose sup t ∈ R ∣ Y ( t ) ∣ ≤ H \sup_{t\in\mathbb{R}}\lvert Y(t)\rvert\le H sup t ∈ R ∣ Y ( t )∣ ≤ H without periodicity. Put K : = H + r K:=H+r K := H + r and examine [ 0 , L ] [0,L] [ 0 , L ] for large L L L with P : = q L > 2 K P:=qL>2K P := q L > 2 K . Any wall phase ζ ∈ ( K , P − K ) \zeta\in(K,P-K) ζ ∈ ( K , P − K ) satisfies ∣ q t − ζ ∣ ≤ H + r \lvert qt-\zeta\rvert\le H+r ∣ q t − ζ ∣ ≤ H + r for all t ∉ [ 0 , L ] t\notin[0,L] t ∈ / [ 0 , L ] (since ∣ Y ∣ ≤ H \lvert Y\rvert\le H ∣ Y ∣ ≤ H ), so Γ [ 0 , L ] \Gamma_{[0,L]} Γ [ 0 , L ] of Euclidean length ≤ L + q ∣ Y ( L ) − Y ( 0 ) ∣ ≤ L + 2 q H \le L+q\lvert Y(L)-Y(0)\rvert\le L+2qH ≤ L + q ∣ Y ( L ) − Y ( 0 )∣ ≤ L + 2 q H covers both walls across ( K , P − K ) (K,P-K) ( K , P − K ) . Close Γ [ 0 , L ] \Gamma_{[0,L]} Γ [ 0 , L ] on the circumference-P P P cylinder by joining ( α X ( L ) , P − Y ( L ) ) (\alpha X(L),P-Y(L)) ( α X ( L ) , P − Y ( L )) to ( α X ( 0 ) , P − Y ( 0 ) ) (\alpha X(0),P-Y(0)) ( α X ( 0 ) , P − Y ( 0 )) (length ≤ 2 α 2 b 2 + H 2 \le2\sqrt{\alpha^{2}b^{2}+H^{2}} ≤ 2 α 2 b 2 + H 2 ), and attach at its base point two contractible polygonal detours inside [ − α b , α b ] × ( R / P Z ) [-\alpha b,\alpha b]\times(\mathbb{R}/P\mathbb{Z}) [ − α b , α b ] × ( R / P Z ) visiting the vertical seam segments { ± α b } × [ − K , K ] \{\pm\alpha b\}\times[-K,K] { ± α b } × [ − K , K ] and returning (total length ≤ 8 α b + 8 K + 4 H \le8\alpha b+8K+4H ≤ 8 α b + 8 K + 4 H ). The resulting closed loop has winding number + 1 +1 + 1 , full wall coverage at radius r r r , and Euclidean length L g e o m ≤ L + C 0 L_{\mathrm{geom}}\le L+C_0 L geom ≤ L + C 0 , where C 0 C_0 C 0 depends only on ( q , b , H , r ) (q,b,H,r) ( q , b , H , r ) .
Set q L : = P / max ( L , L g e o m ) ≤ q q_L:=P/\max(L,L_{\mathrm{geom}})\le q q L := P / max ( L , L geom ) ≤ q and α L : = 1 − q L 2 ≥ α \alpha_L:=\sqrt{1-q_L^{2}}\ge\alpha α L := 1 − q L 2 ≥ α ; then q L → q q_L\to q q L → q as L → ∞ L\to\infty L → ∞ . Read the exact same cylinder loop as data for speed ratio q L q_L q L , half-width b L : = α b / α L b_L:=\alpha b/\alpha_L b L := α b / α L , and capture radius r r r : since α L ≥ α \alpha_L\ge\alpha α L ≥ α , the old α \alpha α -ellipses fit inside the new α L \alpha_L α L -ellipses, so full wall coverage persists and L g e o m ≤ P / q L L_{\mathrm{geom}}\le P/q_L L geom ≤ P / q L . Lemma 2.2(2) supplies a periodic unit-speed patrol at ( q L , b L ) (q_L,b_L) ( q L , b L ) ; for all large L L L , q L > 1 / 3 q_L>1/3 q L > 1/3 , so Step 1 yields 2 ( α b / α L ) ≤ F ( q L ) 2(\alpha b/\alpha_L)\le F(q_L) 2 ( α b / α L ) ≤ F ( q L ) . Letting L → ∞ L\to\infty L → ∞ proves 2 b ≤ F ( q ) 2b\le F(q) 2 b ≤ F ( q ) . ∎
Bowtie patrols and attainment of the ceiling
Theorem 2.1 shows that no bounded-excursion patrol can exceed F ( q ) F(q) F ( q ) even against straight wall runs alone. To prove W o p t ( q ) = F ( q ) W_{\mathrm{opt}}(q)=F(q) W opt ( q ) = F ( q ) across [ q a , 1 ) [q_{\mathrm{a}},1) [ q a , 1 ) —and simultaneously establish Washburn’s back-and-forth width (1) for all q > 0 q>0 q > 0 , strict suboptimality W o p t ( q ) > W b f ( q ) W_{\mathrm{opt}}(q)>W_{\mathrm{bf}}(q) W opt ( q ) > W bf ( q ) for all 0 < q < 1 0<q<1 0 < q < 1 , and our slow-intruder constructions in § 9 —we prove two complementary all-path bowtie sufficiency theorems: an elementary two-level separator that handles q ∈ [ q 0 , 1 ) q\in[q_{0},1) q ∈ [ q 0 , 1 ) (q 0 ≈ 0.471959 q_{0}\approx0.471959 q 0 ≈ 0.471959 , including q > 1 / 2 q>1/2 q > 1/2 where corner diffraction does not occur), and a continuous obstacle-wavefront separator that extends attainment down to the exact diffraction root q a ≈ 0.457892036651 q_{\mathrm{a}}\approx0.457892036651 q a ≈ 0.457892036651 .
General Two-Level Separator for Bowtie and Transverse Patrols
Consider a general symmetric four-leg bowtie in [ − b , b ] × R [-b,b]\times\mathbb{R} [ − b , b ] × R (b = a + δ b=a+\delta b = a + δ , δ ≥ 0 \delta\geq0 δ ≥ 0 , a > 0 a>0 a > 0 , h ≥ 0 h\geq0 h ≥ 0 ) tracing
( − a , + h ) → vertical 2 h ( − a , − h ) → diagonal 2 ℓ ( + a , + h ) → vertical 2 h ( + a , − h ) → diagonal 2 ℓ ( − a , + h ) (-a,+h)\xrightarrow{\text{vertical }2h}(-a,-h)\xrightarrow{\text{diagonal }2\ell}(+a,+h)\xrightarrow{\text{vertical }2h}(+a,-h)\xrightarrow{\text{diagonal }2\ell}(-a,+h) ( − a , + h ) vertical 2 h ( − a , − h ) diagonal 2 ℓ ( + a , + h ) vertical 2 h ( + a , − h ) diagonal 2 ℓ ( − a , + h )
at unit speed V = 1 V=1 V = 1 , where ℓ : = a 2 + h 2 \ell:=\sqrt{a^{2}+h^{2}} ℓ := a 2 + h 2 and period T = 4 h + 4 ℓ T=4h+4\ell T = 4 h + 4 ℓ (when h = 0 h=0 h = 0 , vertical legs have zero duration and the patrol reduces to horizontal back-and-forth oscillation between ± a \pm a ± a ). Set
s : = h ( 1 + q ) , G : = 2 q ℓ − ( 1 − q ) h . s:=h(1+q),\qquad G:=2q\ell-(1-q)h. s := h ( 1 + q ) , G := 2 q ℓ − ( 1 − q ) h .
Theorem 3.1 (General Two-Level Separator). Let q > 0 q>0 q > 0 , r > 0 r>0 r > 0 , a > 0 a>0 a > 0 , h ≥ 0 h\geq0 h ≥ 0 , and b = a + δ b=a+\delta b = a + δ (δ ≥ 0 \delta\geq0 δ ≥ 0 ). If
s = h ( 1 + q ) ≤ r and δ 2 + G 2 ≤ r 2 , s=h(1+q)\leq r\qquad\text{and}\qquad\delta^{2}+G^{2}\leq r^{2}, s = h ( 1 + q ) ≤ r and δ 2 + G 2 ≤ r 2 , then the four-leg bowtie ( a , h ) (a,h) ( a , h ) captures every speed-q q q upward transit trajectory (including curved, waiting, and reversing paths) in [ − b , b ] × R [-b,b]\times\mathbb{R} [ − b , b ] × R . No sign restriction on G G G or lower bound on h / ℓ − q h/\ell-q h / ℓ − q is required.
Proof. Construct a bounded T T T -periodic moving graph y = f ( t , x ) y=f(t,x) y = f ( t , x ) spanning x ∈ [ − b , b ] x\in[-b,b] x ∈ [ − b , b ] according to three barrier rules: (i) on every unlit (non-captured) horizontal flat piece, f f f rises at speed ∂ t f = + q \partial_t f=+q ∂ t f = + q , so along any trajectory p ( t ) = ( x ( t ) , y ( t ) ) p(t)=(x(t),y(t)) p ( t ) = ( x ( t ) , y ( t )) with ∥ p ′ ∥ ≤ q \lVert p'\rVert\leq q ∥ p ′ ∥ ≤ q ,
d d t ( y ( t ) − f ( t , x ( t ) ) ) = y ′ ( t ) − q ≤ 0 , \frac{\mathrm{d}}{\mathrm{d}t}\bigl(y(t)-f(t,x(t))\bigr)=y'(t)-q\leq0, d t d ( y ( t ) − f ( t , x ( t )) ) = y ′ ( t ) − q ≤ 0 ,
preventing below-to-above crossings; (ii) every connecting chord or vertical step lies entirely inside the current closed capture disk B ‾ r ( g ( t ) ) \overline{B}_{r}(g(t)) B r ( g ( t )) (by convexity of B ‾ r ( g ( t ) ) \overline{B}_{r}(g(t)) B r ( g ( t )) , it suffices to check segment endpoints); and (iii) instantaneous resets are either upward (expanding the region below f f f ) or have their entire swept polygonal region between old and new graphs contained inside B ‾ r ( g ( t ) ) \overline{B}_{r}(g(t)) B r ( g ( t )) ).
At the start of the right-hand downward wall leg ( + a , + h ) → ( + a , − h ) (+a,+h)\to(+a,-h) ( + a , + h ) → ( + a , − h ) , set the flat level on [ − b , a ] [-b,a] [ − b , a ] to Y − s Y-s Y − s and join ( a , Y − s ) (a,Y-s) ( a , Y − s ) to the right-wall point ( b , Y − G ) (b,Y-G) ( b , Y − G ) by a chord. During the downward leg of duration 2 h 2h 2 h (Y ′ = − 1 Y'=-1 Y ′ = − 1 ), the flat level rises at speed + q +q + q , so its vertical offset relative to Y ( t ) Y(t) Y ( t ) increases linearly from − s -s − s to − s + 2 h ( 1 + q ) = + s ∈ [ − r , r ] -s+2h(1+q)=+s\in[-r,r] − s + 2 h ( 1 + q ) = + s ∈ [ − r , r ] ; at every instant, ( a , Y ( t ) + offset ) ∈ B ‾ r ( g ( t ) ) (a,Y(t)+\text{offset})\in\overline{B}_{r}(g(t)) ( a , Y ( t ) + offset ) ∈ B r ( g ( t )) since ∣ offset ∣ ≤ s ≤ r |\text{offset}|\leq s\leq r ∣ offset ∣ ≤ s ≤ r , and ( b , Y ( t ) − G ) ∈ B ‾ r ( g ( t ) ) (b,Y(t)-G)\in\overline{B}_{r}(g(t)) ( b , Y ( t ) − G ) ∈ B r ( g ( t )) since δ 2 + G 2 ≤ r 2 \delta^{2}+G^{2}\leq r^{2} δ 2 + G 2 ≤ r 2 .
At departure from ( + a , − h ) (+a,-h) ( + a , − h ) (Y 0 : = − h Y_{0}:=-h Y 0 := − h , t = 0 t=0 t = 0 ), reset the right strip [ a , b ] [a,b] [ a , b ] downward to the flat level Y 0 − G Y_{0}-G Y 0 − G : the swept cut is the triangle with vertices ( a , Y 0 + s ) (a,Y_{0}+s) ( a , Y 0 + s ) , ( a , Y 0 − G ) (a,Y_{0}-G) ( a , Y 0 − G ) , and ( b , Y 0 − G ) (b,Y_{0}-G) ( b , Y 0 − G ) , all three inside B ‾ r ( a , Y 0 ) \overline{B}_{r}(a,Y_{0}) B r ( a , Y 0 ) . During the leftward diagonal transfer t ∈ [ 0 , 2 ℓ ] t\in[0,2\ell] t ∈ [ 0 , 2 ℓ ] , X ( t ) = a − ( a / ℓ ) t X(t)=a-(a/\ell)t X ( t ) = a − ( a / ℓ ) t , Y ( t ) = Y 0 + ( h / ℓ ) t Y(t)=Y_{0}+(h/\ell)t Y ( t ) = Y 0 + ( h / ℓ ) t , place two rising flat levels separated at the guard column X ( t ) X(t) X ( t ) :
F a h e a d ( t ) = Y 0 + s + q t on [ − b , X ( t ) ] , F b e h i n d ( t ) = Y 0 − G + q t on [ X ( t ) , b ] , F_{\mathrm{ahead}}(t)=Y_{0}+s+qt\quad\text{on }[-b,X(t)],\qquad F_{\mathrm{behind}}(t)=Y_{0}-G+qt\quad\text{on }[X(t),b], F ahead ( t ) = Y 0 + s + q t on [ − b , X ( t )] , F behind ( t ) = Y 0 − G + q t on [ X ( t ) , b ] ,
joined by a vertical step at x = X ( t ) x=X(t) x = X ( t ) . Since 2 ℓ ( a − h / ℓ ) = 2 q ℓ − 2 h = G − s 2\ell(a-h/\ell)=2q\ell-2h=G-s 2 ℓ ( a − h / ℓ ) = 2 q ℓ − 2 h = G − s , as t t t runs from 0 0 0 to 2 ℓ 2\ell 2 ℓ , the vertical offset F a h e a d ( t ) − Y ( t ) F_{\mathrm{ahead}}(t)-Y(t) F ahead ( t ) − Y ( t ) changes linearly from + s +s + s to + G +G + G , and F b e h i n d ( t ) − Y ( t ) F_{\mathrm{behind}}(t)-Y(t) F behind ( t ) − Y ( t ) changes linearly from − G -G − G to − s -s − s . Because ∣ s ∣ ≤ r |s| \le r ∣ s ∣ ≤ r and ∣ G ∣ ≤ r |G| \le r ∣ G ∣ ≤ r , both step endpoints at x = X ( t ) x = X(t) x = X ( t ) stay inside B ‾ r ( g ( t ) ) \overline{B}_r(g(t)) B r ( g ( t )) throughout [ 0 , 2 ℓ ] [0,2\ell] [ 0 , 2 ℓ ] , with F a h e a d − F b e h i n d = s + G = 2 q ( h + ℓ ) > 0 F_{\mathrm{ahead}} - F_{\mathrm{behind}} = s + G = 2q(h+\ell) > 0 F ahead − F behind = s + G = 2 q ( h + ℓ ) > 0 .
Upon arrival at ( − a , + h ) (-a,+h) ( − a , + h ) (Y 1 = + h Y_1=+h Y 1 = + h ), F b e h i n d F_{\mathrm{behind}} F behind has offset − s -s − s on [ − a , b ] [-a,b] [ − a , b ] , while F a h e a d F_{\mathrm{ahead}} F ahead on [ − b , − a ] [-b,-a] [ − b , − a ] has offset + G +G + G . Reset [ − b , − a ] [-b,-a] [ − b , − a ] to the chord joining ( − b , Y 1 − G ) (-b,Y_1-G) ( − b , Y 1 − G ) to ( − a , Y 1 − s ) (-a,Y_1-s) ( − a , Y 1 − s ) : the swept region lies inside the convex hull of ( − b , Y 1 ± G ) (-b,Y_1\pm G) ( − b , Y 1 ± G ) and ( − a , Y 1 + { G , − s } ) (-a,Y_1+\{G,-s\}) ( − a , Y 1 + { G , − s }) , all four inside B ‾ r ( − a , Y 1 ) \overline{B}_r(-a,Y_1) B r ( − a , Y 1 ) (if G < 0 G<0 G < 0 , any non-illuminated part of the reset is upward and hence safe). Reflecting x ↔ − x x\leftrightarrow-x x ↔ − x across the second half-period closes a bounded T T T -periodic barrier across [ − b , b ] × R [-b,b]\times\mathbb{R} [ − b , b ] × R that no uncaptured transit can cross from − ∞ -\infty − ∞ to + ∞ +\infty + ∞ .
Corollary 3.2 (Transverse Sufficiency, Small-H H H Gain, and Midrange Exactness). Theorem 3.1 implies:
All-speed transverse sufficiency: Setting h = 0 h=0 h = 0 (s = 0 s=0 s = 0 , G = 2 q a G=2qa G = 2 q a ) and maximizing b = a + r 2 − 4 q 2 a 2 b=a+\sqrt{r^2-4q^2a^2} b = a + r 2 − 4 q 2 a 2 at a 0 = r / ( 2 q 1 + 4 q 2 ) a_0=r/(2q\sqrt{1+4q^2}) a 0 = r / ( 2 q 1 + 4 q 2 ) proves W o p t ( q , 0 ) ≥ W b f ( q ) = r 4 + q − 2 W_{\mathrm{opt}}(q,0)\ge W_{\mathrm{bf}}(q)=r\sqrt{4+q^{-2}} W opt ( q , 0 ) ≥ W bf ( q ) = r 4 + q − 2 against all maneuvering paths for every q > 0 q>0 q > 0 .
Strict suboptimality of W b f ( q ) W_{\mathrm{bf}}(q) W bf ( q ) for all 0 < q < 1 0<q<1 0 < q < 1 : For any 0 < q < 1 0<q<1 0 < q < 1 and all sufficiently small H > 0 H>0 H > 0 , setting h = H h=H h = H (s ≤ r s\le r s ≤ r ) and a = a 0 a=a_0 a = a 0 gives G = R 0 − ( 1 − q ) H + O ( H 2 / r ) G=R_0-(1-q)H+O(H^2/r) G = R 0 − ( 1 − q ) H + O ( H 2 / r ) (R 0 = r / 1 + 4 q 2 R_0=r/\sqrt{1+4q^2} R 0 = r / 1 + 4 q 2 , δ 0 = 2 q R 0 \delta_0=2qR_0 δ 0 = 2 q R 0 ), hence r 2 − G 2 = δ 0 + 1 − q 2 q H − O ( H 2 / r ) \sqrt{r^2-G^2}=\delta_0+\frac{1-q}{2q}H-O(H^2/r) r 2 − G 2 = δ 0 + 2 q 1 − q H − O ( H 2 / r ) and W o p t ( q , H ) ≥ W b f ( q ) + 1 − q q H − O ( H 2 / r ) > W b f ( q ) W_{\mathrm{opt}}(q,H)\ge W_{\mathrm{bf}}(q)+\frac{1-q}{q}H-O(H^2/r)>W_{\mathrm{bf}}(q) W opt ( q , H ) ≥ W bf ( q ) + q 1 − q H − O ( H 2 / r ) > W bf ( q ) . Every larger positive cap allows a sufficiently small successful stroke as well.
Wall-optimal attainment on [ q 0 , 1 ) [q_0,1) [ q 0 , 1 ) : For q ∈ ( 1 / 3 , 1 ) q\in(1/3,1) q ∈ ( 1/3 , 1 ) and c : = 3 q 2 + 2 q − 1 > 0 c:=3q^2+2q-1>0 c := 3 q 2 + 2 q − 1 > 0 , define the wall-optimal bowtie parameters (normalized to r = 1 r=1 r = 1 )
b ( q ) = 1 + 1 c , a ( q ) = 1 c ( 1 + c ) , h ( q ) = 1 − q c 1 + c , R ( q ) = 1 1 + c , ℓ ( q ) = 2 q c 1 + c . (6) b(q)=\sqrt{1+\frac{1}{c}},\qquad a(q)=\frac{1}{\sqrt{c(1+c)}},\qquad h(q)=\frac{1-q}{c\sqrt{1+c}},\qquad R(q)=\frac{1}{\sqrt{1+c}},\qquad\ell(q)=\frac{2q}{c\sqrt{1+c}}. \tag*{(6)} b ( q ) = 1 + c 1 , a ( q ) = c ( 1 + c ) 1 , h ( q ) = c 1 + c 1 − q , R ( q ) = 1 + c 1 , ℓ ( q ) = c 1 + c 2 q . (6) Then G = R ( q ) G=R(q) G = R ( q ) , δ = b ( q ) − a ( q ) = c / ( 1 + c ) \delta=b(q)-a(q)=\sqrt{c/(1+c)} δ = b ( q ) − a ( q ) = c / ( 1 + c ) , δ 2 + G 2 = 1 \delta^2+G^2=1 δ 2 + G 2 = 1 , and s ( q ) = 1 − q 2 c 1 + c ≤ 1 s(q)=\frac{1-q^2}{c\sqrt{1+c}}\le1 s ( q ) = c 1 + c 1 − q 2 ≤ 1 iff
c 2 ( 1 + c ) − ( 1 − q 2 ) 2 = ( 1 + q ) 2 [ q ( 3 q − 1 ) 2 ( 3 q + 2 ) − ( 1 − q ) 2 ] = ( 1 + q ) 2 ( 27 q 4 − 10 q 2 + 4 q − 1 ) ≥ 0. c^2(1+c)-(1-q^2)^2=(1+q)^2[q(3q-1)^2(3q+2)-(1-q)^2]=(1+q)^2(27q^4-10q^2+4q-1)\ge0. c 2 ( 1 + c ) − ( 1 − q 2 ) 2 = ( 1 + q ) 2 [ q ( 3 q − 1 ) 2 ( 3 q + 2 ) − ( 1 − q ) 2 ] = ( 1 + q ) 2 ( 27 q 4 − 10 q 2 + 4 q − 1 ) ≥ 0. Hence W o p t ( q ) = F ( q ) W_{\mathrm{opt}}(q)=F(q) W opt ( q ) = F ( q ) for all q ∈ [ q 0 , 1 ) q\in[q_0,1) q ∈ [ q 0 , 1 ) , where q 0 ≈ 0.4719588902 q_0\approx0.4719588902 q 0 ≈ 0.4719588902 is the unique root of 27 q 4 − 10 q 2 + 4 q − 1 = 0 27q^4-10q^2+4q-1=0 27 q 4 − 10 q 2 + 4 q − 1 = 0 in ( 1 / 3 , 1 ) (1/3,1) ( 1/3 , 1 ) ; in particular, since 27 ( 59 / 125 ) 4 − 10 ( 59 / 125 ) 2 + 4 ( 59 / 125 ) − 1 = 59372 244140625 > 0 27(59/125)^4-10(59/125)^2+4(59/125)-1=\frac{59372}{244140625}>0 27 ( 59/125 ) 4 − 10 ( 59/125 ) 2 + 4 ( 59/125 ) − 1 = 244140625 59372 > 0 , Theorem 3.1 proves W o p t ( q ) = F ( q ) W_{\mathrm{opt}}(q)=F(q) W opt ( q ) = F ( q ) on [ 59 / 125 , 1 ) = [ 0.472 , 1 ) [59/125,1)=[0.472,1) [ 59/125 , 1 ) = [ 0.472 , 1 ) .
For every 1 / 3 < q < 1 1/3<q<1 1/3 < q < 1 , the parameters in (6) attain the wall ceiling against straight wall runners alone. On each downward leg the wall chord has half-length R ( q ) R(q) R ( q ) and its center phase moves forward by 2 ( 1 + q ) h ( q ) 2(1+q)h(q) 2 ( 1 + q ) h ( q ) . The swept interval therefore has length 2 R ( q ) + 2 ( 1 + q ) h ( q ) = q T 2R(q)+2(1+q)h(q)=qT 2 R ( q ) + 2 ( 1 + q ) h ( q ) = q T , a whole period. The distinction below q a q_a q a comes from runners allowed to leave a wall.
Continuous Obstacle-Wavefront Separator Criterion
Below q 0 ≈ 0.471959 q_0\approx0.471959 q 0 ≈ 0.471959 , h ( q ) ( 1 + q ) > 1 h(q)(1+q)>1 h ( q ) ( 1 + q ) > 1 on the wall-optimal bowtie (6) , so a flat horizontal front resets outside B ‾ 1 \overline{B}_1 B 1 . Instead, we attach a curved eikonal diffraction fan directly to the descending disk boundary and hand it off into the convex spacetime capture tube of the succeeding diagonal leg. For any general bowtie ( a , h ) (a,h) ( a , h ) (r = 1 r=1 r = 1 , V = 1 V=1 V = 1 , ℓ = a 2 + h 2 \ell=\sqrt{a^2+h^2} ℓ = a 2 + h 2 , diagonal leg length L d i a g : = 2 ℓ L_{\mathrm{diag}}:=2\ell L diag := 2 ℓ ), set
α = 1 − q 2 , c l e g = a ℓ , d l e g = h ℓ , m = q d l e g + α c l e g , κ = α d l e g − q c l e g . (7) \alpha=\sqrt{1-q^2},\qquad c_{\mathrm{leg}}=\frac{a}{\ell},\qquad d_{\mathrm{leg}}=\frac{h}{\ell},\qquad m=qd_{\mathrm{leg}}+\alpha c_{\mathrm{leg}},\qquad\kappa=\alpha d_{\mathrm{leg}}-qc_{\mathrm{leg}}. \tag*{(7)} α = 1 − q 2 , c leg = ℓ a , d leg = ℓ h , m = q d leg + α c leg , κ = α d leg − q c leg . (7)
Notice that 0 < q < d l e g < 1 ⟺ h 2 > q 2 ( a 2 + h 2 ) ⟺ α d l e g > q c l e g ⟺ κ > 0 0<q<d_{\mathrm{leg}}<1 \Longleftrightarrow h^2>q^2(a^2+h^2) \Longleftrightarrow\alpha d_{\mathrm{leg}}>q c_{\mathrm{leg}} \Longleftrightarrow\kappa>0 0 < q < d leg < 1 ⟺ h 2 > q 2 ( a 2 + h 2 ) ⟺ α d leg > q c leg ⟺ κ > 0 . Define the terminal diffraction angle θ ∗ \theta_* θ ∗ and elliptic-type fan integral p p p by
θ ∗ : = arctan ( q κ 1 + q m ) , p : = ∫ 0 θ ∗ d θ cos θ + q 2 − sin 2 θ , (8) \theta_*:=\arctan\left(\frac{q\kappa}{1+qm}\right),\qquad p:=\int_0^{\theta_*}\frac{d\theta}{\cos\theta+\sqrt{q^2-\sin^2\theta}}, \tag*{(8)} θ ∗ := arctan ( 1 + q m q κ ) , p := ∫ 0 θ ∗ cos θ + q 2 − sin 2 θ d θ , (8)
and the rounded-escape ratio
R r o u n d ( a , h ; q ) : = h ( m + q ) 1 + κ cos θ ∗ − m sin θ ∗ + ( m + q ) p . (9) \mathcal{R}_{\mathrm{round}}(a,h;q):=\frac{h(m+q)}{1+\kappa\cos\theta_*-m\sin\theta_*+(m+q)p}. \tag*{(9)} R round ( a , h ; q ) := 1 + κ cos θ ∗ − m sin θ ∗ + ( m + q ) p h ( m + q ) . (9)
Lemma 3.3 (Continuous Attached Diffraction Fan). Suppose 0 < q < d l e g < 1 0<q<d_{\mathrm{leg}}<1 0 < q < d leg < 1 (κ > 0 \kappa>0 κ > 0 ), and let the guard move downward along ( a , − t ) (a,-t) ( a , − t ) for t ∈ [ 0 , p ] t\in[0,p] t ∈ [ 0 , p ] . Define θ ( s ) ∈ [ 0 , θ ∗ ] \theta(s)\in[0,\theta_*] θ ( s ) ∈ [ 0 , θ ∗ ] implicitly for s ∈ [ 0 , p ] s\in[0,p] s ∈ [ 0 , p ] by ∫ 0 θ ( s ) d θ cos θ + q 2 − sin 2 θ = s \int_0^{\theta(s)}\frac{d\theta}{\cos\theta+\sqrt{q^2-\sin^2\theta}}=s ∫ 0 θ ( s ) c o s θ + q 2 − s i n 2 θ d θ = s , and set X f a n ( s ) : = ( a − cos θ ( s ) , sin θ ( s ) − s ) X_{\mathrm{fan}}(s):=(a-\cos\theta(s),\sin\theta(s)-s) X fan ( s ) := ( a − cos θ ( s ) , sin θ ( s ) − s ) with velocity X f a n ′ ( s ) = ( u ( s ) , w ( s ) ) X'_{\mathrm{fan}}(s)=(u(s),w(s)) X fan ′ ( s ) = ( u ( s ) , w ( s )) . Then u 2 + w 2 = q 2 u^2+w^2=q^2 u 2 + w 2 = q 2 , u ′ > 0 u'>0 u ′ > 0 , w ≥ q m > 0 w\ge qm>0 w ≥ q m > 0 , ( u ( 0 ) , w ( 0 ) ) = ( 0 , q ) (u(0),w(0))=(0,q) ( u ( 0 ) , w ( 0 )) = ( 0 , q ) , ( u ( p ) , w ( p ) ) = q ( κ , m ) (u(p),w(p))=q(\kappa,m) ( u ( p ) , w ( p )) = q ( κ , m ) , and for each t ∈ [ 0 , p ] t\in[0,p] t ∈ [ 0 , p ] the tangent family
P s ( t ) : = X f a n ( s ) + ( t − s ) X f a n ′ ( s ) , 0 ≤ s ≤ t , P_s(t):=X_{\mathrm{fan}}(s)+(t-s)X'_{\mathrm{fan}}(s),\qquad0\le s\le t, P s ( t ) := X fan ( s ) + ( t − s ) X fan ′ ( s ) , 0 ≤ s ≤ t , forms a spatial graph y = f ( t , x ) y=f(t,x) y = f ( t , x ) , smooth on its unlit interior with continuous first derivatives up to the contact endpoint, over x ∈ [ a − 1 , a − cos θ ( t ) ] x\in[a-1,a-\cos\theta(t)] x ∈ [ a − 1 , a − cos θ ( t )] satisfying the eikonal identity f t = q 1 + f x 2 f_t=q\sqrt{1+f_x^2} f t = q 1 + f x 2 outside B 1 ( a , − t ) B_1(a,-t) B 1 ( a , − t ) , joining continuously to the flat front y = q t y=qt y = q t at x = a − 1 x=a-1 x = a − 1 and ending at X f a n ( t ) ∈ ∂ B 1 ( a , − t ) X_{\mathrm{fan}}(t)\in\partial B_1(a,-t) X fan ( t ) ∈ ∂ B 1 ( a , − t ) .
Proof. Differentiating θ ′ ( s ) = cos θ + q 2 − sin 2 θ > 0 \theta'(s)=\cos\theta+\sqrt{q^2-\sin^2\theta}>0 θ ′ ( s ) = cos θ + q 2 − sin 2 θ > 0 gives u = sin θ θ ′ u=\sin\theta\,\theta' u = sin θ θ ′ , w = cos θ θ ′ − 1 w=\cos\theta\,\theta'-1 w = cos θ θ ′ − 1 , and u 2 + w 2 = θ ′ 2 − 2 cos θ θ ′ + 1 = q 2 u^2+w^2=\theta'^2-2\cos\theta\,\theta'+1=q^2 u 2 + w 2 = θ ′2 − 2 cos θ θ ′ + 1 = q 2 . Writing ( u , w ) = ( q sin ψ , q cos ψ ) (u,w)=(q\sin\psi,q\cos\psi) ( u , w ) = ( q sin ψ , q cos ψ ) yields tan θ = q sin ψ 1 + q cos ψ \tan\theta=\frac{q\sin\psi}{1+q\cos\psi} tan θ = 1 + q c o s ψ q s i n ψ with d θ d ψ = q ( q + cos ψ ) 1 + 2 q cos ψ + q 2 > 0 \frac{d\theta}{d\psi}=\frac{q(q+\cos\psi)}{1+2q\cos\psi+q^2}>0 d ψ d θ = 1 + 2 q c o s ψ + q 2 q ( q + c o s ψ ) > 0 ; as ψ \psi ψ increases from 0 0 0 to arcsin κ = arccos m ∈ ( 0 , π / 2 ) \arcsin\kappa=\arccos m\in(0,\pi/2) arcsin κ = arccos m ∈ ( 0 , π /2 ) (note κ 2 + m 2 = 1 \kappa^2+m^2=1 κ 2 + m 2 = 1 ), θ \theta θ increases strictly from 0 0 0 to θ ∗ \theta_* θ ∗ , with L : = 1 + 2 q cos ψ + q 2 L:=\sqrt{1+2q\cos\psi+q^2} L := 1 + 2 q cos ψ + q 2 matching cos θ + q 2 − sin 2 θ \cos\theta+\sqrt{q^2-\sin^2\theta} cos θ + q 2 − sin 2 θ since L − cos θ = q 2 + q cos ψ L > 0 L-\cos\theta=\frac{q^2+q\cos\psi}{L}>0 L − cos θ = L q 2 + q c o s ψ > 0 . Thus w ≥ q m > 0 w\ge qm>0 w ≥ q m > 0 and u ′ > 0 u'>0 u ′ > 0 . For 0 ≤ s < t 0\le s<t 0 ≤ s < t , ∂ s P s ( t ) = ( t − s ) u ′ ( s ) ( 1 , − u ( s ) / w ( s ) ) \partial_sP_s(t)=(t-s)u'(s)(1,-u(s)/w(s)) ∂ s P s ( t ) = ( t − s ) u ′ ( s ) ( 1 , − u ( s ) / w ( s )) has strictly positive x x x -component and slope f x = − u / w f_x=-u/w f x = − u / w , so f t = w − f x u = ( w 2 + u 2 ) / w = q 2 / w = q 1 + f x 2 f_t=w-f_xu=(w^2+u^2)/w=q^2/w=q\sqrt{1+f_x^2} f t = w − f x u = ( w 2 + u 2 ) / w = q 2 / w = q 1 + f x 2 . Relative to the current guard ( a , − t ) (a,-t) ( a , − t ) , P s ( t ) − ( a , − t ) = ( − cos θ ( s ) , sin θ ( s ) ) + ( t − s ) ( u ( s ) , w ( s ) + 1 ) P_s(t)-(a,-t)=(-\cos\theta(s),\sin\theta(s))+(t-s)(u(s),w(s)+1) P s ( t ) − ( a , − t ) = ( − cos θ ( s ) , sin θ ( s )) + ( t − s ) ( u ( s ) , w ( s ) + 1 ) ; these two vectors are orthogonal (θ ′ ( cos θ + q 2 − sin 2 θ ) ⟹ − cos θ u + sin θ ( w + 1 ) = 0 \theta'\bigl(\cos\theta+\sqrt{q^2-\sin^2\theta}\bigr)\Longrightarrow-\cos\theta\,u+\sin\theta(w+1)=0 θ ′ ( cos θ + q 2 − sin 2 θ ) ⟹ − cos θ u + sin θ ( w + 1 ) = 0 ), so ∥ P s ( t ) − ( a , − t ) ∥ 2 > 1 \|P_s(t)-(a,-t)\|^2>1 ∥ P s ( t ) − ( a , − t ) ∥ 2 > 1 for s < t s<t s < t . By Cauchy–Schwarz and f t = q 1 + f x 2 f_t=q\sqrt{1+f_x^2} f t = q 1 + f x 2 , no speed-q q q trajectory can cross y = f ( t , x ) y=f(t,x) y = f ( t , x ) upward outside the captured endpoint X f a n ( t ) X_{\mathrm{fan}}(t) X fan ( t ) .
Theorem 3.4 (Obstacle-Wavefront Sufficient Criterion). Normalize r = 1 r=1 r = 1 , V = 1 V=1 V = 1 . Suppose a four-leg bowtie ( a , h ) (a,h) ( a , h ) (b = a + δ b=a+\delta b = a + δ , ℓ = 2 + h 2 \ell=\sqrt{2+h^2} ℓ = 2 + h 2 , L d i a g = 2 ℓ L_{\mathrm{diag}}=2\ell L diag = 2 ℓ , G = 2 q ℓ − ( 1 − q ) h G=2q\ell-(1-q)h G = 2 q ℓ − ( 1 − q ) h ) satisfies 0 < q < d l e g < 1 0<q<d_{\mathrm{leg}}<1 0 < q < d leg < 1 , δ ≥ 0 \delta\ge0 δ ≥ 0 , 0 < G < 1 0<G<1 0 < G < 1 , and δ 2 + G 2 ≤ 1 \delta^2+G^2\le1 δ 2 + G 2 ≤ 1 . Put C : = cos θ ∗ C:=\cos\theta_* C := cos θ ∗ , S : = sin θ ∗ S:=\sin\theta_* S := sin θ ∗ , K : = κ C − m S + ( m + q ) p K:=\kappa C-mS+(m+q)p K := κ C − m S + ( m + q ) p , v : = h − p v:=h-p v := h − p , and for j ∈ { 0 , 1 } j\in\{0,1\} j ∈ { 0 , 1 } define ( x 0 , y 0 ) : = ( − 1 , h + q p ) (x_0,y_0):=(-1,h+qp) ( x 0 , y 0 ) := ( − 1 , h + q p ) , ( x 1 , y 1 ) : = ( − C , h − p + S ) (x_1,y_1):=(-C,h-p+S) ( x 1 , y 1 ) := ( − C , h − p + S ) , and
N j : = d l e g x j + c l e g y j , U j : = − m x j + κ y j α . N_j:=d_{\mathrm{leg}}x_j+c_{\mathrm{leg}}y_j,\qquad U_j:=\frac{-mx_j+\kappa y_j}{\alpha}. N j := d leg x j + c leg y j , U j := α − m x j + κ y j . If all four condition groups hold:
h ( m + q ) ≤ 1 + K (i.e., R r o u n d ( a , h ; q ) ≤ 1 ), (10) h(m+q)\le1+K\qquad\text{(i.e., }\mathcal{R}_{\mathrm{round}}(a,h;q)\le1\text{),} \tag*{(10)} h ( m + q ) ≤ 1 + K (i.e., R round ( a , h ; q ) ≤ 1 ), (10) 0 < p < h , a + b ≥ 1 , (11) 0<p<h,\qquad a+b\ge1, \tag*{(11)} 0 < p < h , a + b ≥ 1 , (11) q v ≤ 1 , α 2 v ≤ 2 ( G − q ) , (12) qv\le1,\qquad\alpha^2v\le2(G-q), \tag*{(12)} q v ≤ 1 , α 2 v ≤ 2 ( G − q ) , (12) N j ≥ 0 , 0 ≤ U j ≤ L d i a g ( j = 0 , 1 ) , (13) N_j\ge0,\qquad0\le U_j\le L_{\mathrm{diag}}\qquad(j=0,1), \tag*{(13)} N j ≥ 0 , 0 ≤ U j ≤ L diag ( j = 0 , 1 ) , (13) then ( a , h ) (a,h) ( a , h ) captures every speed-q q q upward transit trajectory in [ − b , b ] × R [-b,b]\times\mathbb{R} [ − b , b ] × R .
Proof. Parameterize the right-hand downward leg ( a , − τ ) (a,-\tau) ( a , − τ ) by centered time τ ∈ [ − h , h ] \tau\in[-h,h] τ ∈ [ − h , h ] . During τ ∈ [ 0 , p ] \tau\in[0,p] τ ∈ [ 0 , p ] , use Lemma 3.3 ’s fan P s ( τ ) P_s(\tau) P s ( τ ) on [ a − 1 , a − cos θ ( τ ) ] [a-1,a-\cos\theta(\tau)] [ a − 1 , a − cos θ ( τ )] , extended leftward by the flat front y = q τ y=q\tau y = q τ on [ − b , a − 1 ] [-b,a-1] [ − b , a − 1 ] (inside [ − b , b ] [-b,b] [ − b , b ] since a − 1 ≥ − b a-1\ge-b a − 1 ≥ − b by (11) ) and rightward by the chord from X f a n ( τ ) X_{\mathrm{fan}}(\tau) X fan ( τ ) to W − ( τ ) : = ( b , − τ − G ) W_-(\tau):=(b,-\tau-G) W − ( τ ) := ( b , − τ − G ) (inside B 1 ( a , − τ ) B_1(a,-\tau) B 1 ( a , − τ ) by convexity since δ 2 + G 2 ≤ 1 \delta^2+G^2\le1 δ 2 + G 2 ≤ 1 ). During τ ∈ [ − p , 0 ] \tau\in[-p,0] τ ∈ [ − p , 0 ] , use the vertically reflected incoming fan at time − τ -\tau − τ , joined to the same wall point W − ( τ ) W_-(\tau) W − ( τ ) . The graph coincides with y = 0 y=0 y = 0 on [ − b , a − 1 ] [-b,a-1] [ − b , a − 1 ] and the chord ( a − 1 , 0 ) → ( b , − G ) (a-1,0)\to(b,-G) ( a − 1 , 0 ) → ( b , − G ) at τ = 0 \tau=0 τ = 0 , giving a continuous eikonal barrier across [ − p , p ] [-p,p] [ − p , p ] .
Next, shift to departure time t : = τ − h t:=\tau-h t := τ − h , so the succeeding leftward diagonal leg runs over u ∈ [ 0 , L d i a g ] u\in[0,L_{\mathrm{diag}}] u ∈ [ 0 , L diag ] as g d i a g ( u ) = ( a − c l e g u , − h + d l e g u ) g_{\mathrm{diag}}(u)=(a-c_{\mathrm{leg}}u,-h+d_{\mathrm{leg}}u) g diag ( u ) = ( a − c leg u , − h + d leg u ) , and handoff τ = p \tau=p τ = p is t = − v t=-v t = − v . The single-leg spacetime capture tube C d i a g : = { ( t , P ) : ∃ u ∈ [ 0 , L d i a g ] , ∥ P − g d i a g ( u ) ∥ 2 + q ∣ u − t ∣ ≤ 1 } C_{\mathrm{diag}}:=\{(t,P):\exists u\in[0,L_{\mathrm{diag}}],\ \lVert P-g_{\mathrm{diag}}(u)\rVert_2+q\lvert u-t\rvert\le1\} C diag := {( t , P ) : ∃ u ∈ [ 0 , L diag ] , ∥ P − g diag ( u ) ∥ 2 + q ∣ u − t ∣ ≤ 1 } is convex in ( t , P ) (t,P) ( t , P ) (by convexity of ∥ ⋅ ∥ 2 \lVert\cdot\rVert_2 ∥ ⋅ ∥ 2 and ∣ ⋅ ∣ \lvert\cdot\rvert ∣ ⋅ ∣ under joint interpolation of ( t , P , u ) (t,P,u) ( t , P , u ) ), and any intruder meeting C d i a g C_{\mathrm{diag}} C diag cannot avoid B 1 ( g d i a g ( u ) ) B_1(g_{\mathrm{diag}}(u)) B 1 ( g diag ( u )) at time u u u by ∥ p ( u ) − p ( t ) ∥ 2 ≤ q ∣ u − t ∣ \lVert p(u)-p(t)\rVert_2\le q\lvert u-t\rvert ∥ p ( u ) − p ( t ) ∥ 2 ≤ q ∣ u − t ∣ . For any point P = ( a + x , − h + y ) P=(a+x,-h+y) P = ( a + x , − h + y ) at time t = − v t=-v t = − v (v > 0 v>0 v > 0 ), the witness U : = ( − m x + κ y ) / α U:=(-mx+\kappa y)/\alpha U := ( − m x + κ y ) / α satisfies U − ( − v ) ≥ 0 U-(-v)\ge0 U − ( − v ) ≥ 0 when U ≥ 0 U\ge0 U ≥ 0 , and direct orthogonal expansion along ( c l e g , d l e g ) (c_{\mathrm{leg}},d_{\mathrm{leg}}) ( c leg , d leg ) gives ∥ P − g d i a g ( U ) ∥ 2 = N / α \lVert P-g_{\mathrm{diag}}(U)\rVert_2=N/\alpha ∥ P − g diag ( U ) ∥ 2 = N / α whenever N : = d l e g x + c l e g y ≥ 0 N:=d_{\mathrm{leg}}x+c_{\mathrm{leg}}y\ge0 N := d leg x + c leg y ≥ 0 , yielding the exact affine identity
∥ ( a + x , − h + y ) − g d i a g ( U ) ∥ 2 + q ( U + v ) = q v + κ x + m y . (14) \lVert(a+x,-h+y)-g_{\mathrm{diag}}(U)\rVert_2+q(U+v)=qv+\kappa x+my. \tag*{(14)} ∥( a + x , − h + y ) − g diag ( U ) ∥ 2 + q ( U + v ) = q v + κ x + m y . (14)
At handoff τ = p \tau=p τ = p (t = − v t=-v t = − v ), the fan s ↦ P s ( p ) = ( a + x s , − h + y s ) s\mapsto P_s(p)=(a+x_s,-h+y_s) s ↦ P s ( p ) = ( a + x s , − h + y s ) runs from ( x 0 , y 0 ) = ( − 1 , h + q p ) (x_0,y_0)=(-1,h+qp) ( x 0 , y 0 ) = ( − 1 , h + q p ) (s = 0 s=0 s = 0 ) to ( x 1 , y 1 ) = ( − C , h − p + S ) (x_1,y_1)=(-C,h-p+S) ( x 1 , y 1 ) = ( − C , h − p + S ) (s = p s=p s = p ). Differentiating along s ∈ [ 0 , p ] s\in[0,p] s ∈ [ 0 , p ] using ∂ s P s ( p ) = ( p − s ) u ′ ( s ) ( 1 , − u / w ) \partial_sP_s(p)=(p-s)u'(s)(1,-u/w) ∂ s P s ( p ) = ( p − s ) u ′ ( s ) ( 1 , − u / w ) with 0 ≤ u / w ≤ κ / m 0\le u/w\le\kappa/m 0 ≤ u / w ≤ κ / m and d l e g m − c l e g κ = q > 0 d_{\mathrm{leg}}m-c_{\mathrm{leg}}\kappa=q>0 d leg m − c leg κ = q > 0 gives:
d d s ( q v + κ x s + m y s ) = ( p − s ) u ′ ( κ − m u w ) ≥ 0 , d N s d s = ( p − s ) u ′ ( d l e g − c l e g u w ) ≥ 0 , \frac{d}{ds}(qv+\kappa x_s+my_s)=(p-s)u'\left(\kappa-m\frac{u}{w}\right)\ge0,
\qquad
\frac{dN_s}{ds}=(p-s)u'\left(d_{\mathrm{leg}}-c_{\mathrm{leg}}\frac{u}{w}\right)\ge0, d s d ( q v + κ x s + m y s ) = ( p − s ) u ′ ( κ − m w u ) ≥ 0 , d s d N s = ( p − s ) u ′ ( d leg − c leg w u ) ≥ 0 ,
d U s d s = − ( p − s ) u ′ α ( m + κ u w ) ≤ 0. \frac{dU_s}{ds}=-\frac{(p-s)u'}{\alpha}\left(m+\kappa\frac{u}{w}\right)\le0. d s d U s = − α ( p − s ) u ′ ( m + κ w u ) ≤ 0.
Hence (13) implies N s ≥ N 0 ≥ 0 N_s\ge N_0\ge0 N s ≥ N 0 ≥ 0 and 0 ≤ U 1 ≤ U s ≤ U 0 ≤ L d i a g 0\le U_1\le U_s\le U_0\le L_{\mathrm{diag}} 0 ≤ U 1 ≤ U s ≤ U 0 ≤ L diag for all s ∈ [ 0 , p ] s\in[0,p] s ∈ [ 0 , p ] , and (10) bounds the maximum cost at s = p s=p s = p by q v − κ C + m ( h − p + S ) = h ( m + q ) − K ≤ 1 qv-\kappa C+m(h-p+S)=h(m+q)-K\le1 q v − κ C + m ( h − p + S ) = h ( m + q ) − K ≤ 1 . Thus the entire handoff fan lies inside C d i a g C_{\mathrm{diag}} C diag . Likewise, for t = − w ∈ [ − v , 0 ] t=-w\in[-v,0] t = − w ∈ [ − v , 0 ] (w ∈ [ 0 , v ] w\in[0,v] w ∈ [ 0 , v ] ), the wall point W − ( h − w ) = ( b , − h + w − G ) W_-(h-w)=(b,-h+w-G) W − ( h − w ) = ( b , − h + w − G ) satisfies δ 2 + ( w − G ) 2 ≤ 1 − 2 G w + w 2 ≤ ( 1 − q w ) 2 \delta^2+(w-G)^2\le1-2Gw+w^2\le(1-qw)^2 δ 2 + ( w − G ) 2 ≤ 1 − 2 Gw + w 2 ≤ ( 1 − q w ) 2 and 1 − q w ≥ 0 1-qw\ge0 1 − q w ≥ 0 by (12) (since ( 1 − q w ) 2 − ( 1 − 2 G w + w 2 ) = w [ 2 ( G − q ) − α 2 w ] ≥ w [ 2 ( G − q ) − α 2 v ] ≥ 0 (1-qw)^2-(1-2Gw+w^2)=w[2(G-q)-\alpha^2w]\ge w[2(G-q)-\alpha^2v]\ge0 ( 1 − q w ) 2 − ( 1 − 2 Gw + w 2 ) = w [ 2 ( G − q ) − α 2 w ] ≥ w [ 2 ( G − q ) − α 2 v ] ≥ 0 ), so ( − w , W − ( h − w ) ) ∈ C d i a g (-w,W_-(h-w))\in C_{\mathrm{diag}} ( − w , W − ( h − w )) ∈ C diag via witness u = 0 u=0 u = 0 . By convexity of the t = − v t=-v t = − v slice of C d i a g C_{\mathrm{diag}} C diag , replacing the fan-and-chord at t = − v t=-v t = − v by the straight chord from ( a − 1 , q p ) (a-1,qp) ( a − 1 , q p ) to W − ( p ) W_-(p) W − ( p ) sweeps a region wholly inside C d i a g C_{\mathrm{diag}} C diag .
Finally, interpolate in spacetime from ( − v , ( a − 1 , q p ) ) ∈ C d i a g (-v,(a-1,qp))\in C_{\mathrm{diag}} ( − v , ( a − 1 , q p )) ∈ C diag to ( L d i a g , ( − a , h + G ) ) ∈ C d i a g (L_{\mathrm{diag}},(-a,h+G))\in C_{\mathrm{diag}} ( L diag , ( − a , h + G )) ∈ C diag (the latter has ∥ ( − a , h + G ) − g d i a g ( L d i a g ) ∥ 2 = G ≤ 1 \lVert(-a,h+G)-g_{\mathrm{diag}}(L_{\mathrm{diag}})\rVert_2=G\le1 ∥( − a , h + G ) − g diag ( L diag ) ∥ 2 = G ≤ 1 at u = L d i a g u=L_{\mathrm{diag}} u = L diag ). Because q ( h + L d i a g ) = q h + 2 q ℓ = h + G q(h+L_{\mathrm{diag}})=qh+2q\ell=h+G q ( h + L diag ) = q h + 2 q ℓ = h + G , the interpolated point U ( t ) U(t) U ( t ) lies on the rising flat level F + ( t ) = q ( h + t ) F_+(t)=q(h+t) F + ( t ) = q ( h + t ) throughout t ∈ [ − v , L d i a g ] t\in[-v,L_{\mathrm{diag}}] t ∈ [ − v , L diag ] , with x U ( 0 ) ≤ a x_U(0)\le a x U ( 0 ) ≤ a . The spacetime reflection L ( t ) : = − U ( L d i a g − t ) L(t):=-U(L_{\mathrm{diag}}-t) L ( t ) := − U ( L diag − t ) lies in the same diagonal tube on the lower level F − ( t ) = − h − G + q t F_-(t)=-h-G+qt F − ( t ) = − h − G + q t . For 0 ≤ t ≤ L d i a g 0\le t\le L_{\mathrm{diag}} 0 ≤ t ≤ L diag , the affine horizontal coordinates obey x U ( t ) − x L ( t ) = x U ( 0 ) − a ≤ 0 x_U(t)-x_L(t)=x_U(0)-a\le0 x U ( t ) − x L ( t ) = x U ( 0 ) − a ≤ 0 . Use F + F_+ F + on [ − b , x U ( t ) ] [-b,x_U(t)] [ − b , x U ( t )] , the captured chord U ( t ) L ( t ) U(t)L(t) U ( t ) L ( t ) , and F − F_- F − on [ x L ( t ) , b ] [x_L(t),b] [ x L ( t ) , b ] .
During − v ≤ t ≤ 0 -v\le t\le0 − v ≤ t ≤ 0 , use F + F_+ F + to the left of U ( t ) U(t) U ( t ) and join U ( t ) U(t) U ( t ) to W − ( h + t ) W_-(h+t) W − ( h + t ) in the same convex tube. At departure the swept triangle has vertices U ( 0 ) U(0) U ( 0 ) , L ( 0 ) = ( a , − h − G ) L(0)=(a,-h-G) L ( 0 ) = ( a , − h − G ) and ( b , − h − G ) (b,-h-G) ( b , − h − G ) , all in the tube. After arrival use the spacetime reflection of this construction, joining the left wall at offset + G +G + G to L ( t ) L(t) L ( t ) ; the arrival triangle is the reflection of the departure triangle. At the late handoff τ = − p \tau=-p τ = − p on the next downward leg, reverse the fan-to-chord replacement in the tube. Switch the wall offset from + G +G + G to − G -G − G across the triangle formed by the disk contact point and the two wall points at offsets ± G \pm G ± G . This triangle lies in the current closed disk. The next incoming fan now has the required wall endpoint. Repeating with left and right interchanged gives a bounded periodic separator throughout the corridor.
The critical speed and attainment
Specialize (7)–(9) to the wall-optimal bowtie (6) , where c l e g = q 2 q c_{\mathrm{leg}}=\frac{\sqrt{q}}{2q} c leg = 2 q q and d l e g = 1 − q 2 q d_{\mathrm{leg}}=\frac{1-q}{2q} d leg = 2 q 1 − q , and write R r o u n d ( q ) : = R r o u n d ( a ( q ) , h ( q ) ; q ) \mathcal{R}_{\mathrm{round}}(q):=\mathcal{R}_{\mathrm{round}}(a(q),h(q);q) R round ( q ) := R round ( a ( q ) , h ( q ) ; q ) . Observe the exact factorization
4 q 2 ( q 2 d l e g 2 − q 2 c l e g 2 ) = ( 1 − q 2 ) ( 1 − q ) 2 − q 2 ( 3 q 2 + 2 q − 1 ) = ( 1 + q ) ( 1 − 2 q ) ( 1 − q + 2 q 2 ) . 4q^{2}\left(q^{2}d_{\mathrm{leg}}^{2}-q^{2}c_{\mathrm{leg}}^{2}\right)=(1-q^{2})(1-q)^{2}-q^{2}(3q^{2}+2q-1)=(1+q)(1-2q)(1-q+2q^{2}). 4 q 2 ( q 2 d leg 2 − q 2 c leg 2 ) = ( 1 − q 2 ) ( 1 − q ) 2 − q 2 ( 3 q 2 + 2 q − 1 ) = ( 1 + q ) ( 1 − 2 q ) ( 1 − q + 2 q 2 ) .
Because 1 + q > 0 1+q>0 1 + q > 0 and 1 − q + 2 q 2 = 2 ( q − 1 4 ) 2 + 7 8 > 0 1-q+2q^{2}=2\left(q-\frac{1}{4}\right)^{2}+\frac{7}{8}>0 1 − q + 2 q 2 = 2 ( q − 4 1 ) 2 + 8 7 > 0 , we have κ ( q ) > 0 ⟺ q ∈ ( 1 / 3 , 1 / 2 ) \kappa(q)>0\Longleftrightarrow q\in(1/3,1/2) κ ( q ) > 0 ⟺ q ∈ ( 1/3 , 1/2 ) , with κ ( 1 / 2 ) = 0 \kappa(1/2)=0 κ ( 1/2 ) = 0 (above q = 1 / 2 q=1/2 q = 1/2 , d l e g < q d_{\mathrm{leg}}<q d leg < q so the obstacle-fan regime does not arise and Corollary 3.2 (3) already applies).
Lemma 3.5 (Critical Speed q a q_a q a and Wavefront Certification on [ q a , 59 / 125 ] [q_a,59/125] [ q a , 59/125 ] ). On ( 1 / 3 , 1 / 2 ] (1/3,1/2] ( 1/3 , 1/2 ] , where κ ( q ) ≥ 0 \kappa(q)\ge0 κ ( q ) ≥ 0 , the function R r o u n d ( q ) \mathcal{R}_{\mathrm{round}}(q) R round ( q ) is continuous, satisfies lim q ↓ 1 / 3 R r o u n d ( q ) = + ∞ \lim_{q\downarrow1/3}\mathcal{R}_{\mathrm{round}}(q)=+\infty lim q ↓ 1/3 R round ( q ) = + ∞ and R r o u n d ( 1 / 2 ) = 2 / 7 < 1 \mathcal{R}_{\mathrm{round}}(1/2)=2/\sqrt{7}<1 R round ( 1/2 ) = 2/ 7 < 1 , and crosses 1 at a unique root q a q_a q a enclosed in
0.457892036651 < q a < 0.457892036652 , F ( q a ) ∈ [ 3.36785112514 , 3.36785112516 ] r . (15) 0.457892036651<q_a<0.457892036652,\qquad F(q_a)\in[3.36785112514,3.36785112516]\,r. \tag*{(15)} 0.457892036651 < q a < 0.457892036652 , F ( q a ) ∈ [ 3.36785112514 , 3.36785112516 ] r . (15) Moreover, for every q ∈ [ q a , 59 / 125 ] = [ q a , 0.472 ] q\in[q_a,59/125]=[q_a,0.472] q ∈ [ q a , 59/125 ] = [ q a , 0.472 ] , the wall-optimal bowtie (6) satisfies all hypotheses (10)–(13) of Theorem 3.4 .
Proof. Continuity on ( 1 / 3 , 1 / 2 ] (1/3,1/2] ( 1/3 , 1/2 ] is immediate from sin θ ∗ ( q ) ≤ q q 1 + q m < q \sin\theta_*(q)\le\frac{q^q}{1+qm}<q sin θ ∗ ( q ) ≤ 1 + q m q q < q , and as q ↓ 1 / 3 q\downarrow1/3 q ↓ 1/3 , c ↓ 0 c\downarrow0 c ↓ 0 sends h ( q ) → + ∞ h(q)\to+\infty h ( q ) → + ∞ while p ( q ) p(q) p ( q ) stays bounded, giving R r o u n d ( q ) → + ∞ \mathcal{R}_{\mathrm{round}}(q)\to+\infty R round ( q ) → + ∞ ; at q = 1 / 2 q=1/2 q = 1/2 , κ = 0 ⟹ θ ∗ = 0 \kappa=0\Longrightarrow\theta_*=0 κ = 0 ⟹ θ ∗ = 0 , p = 0 p=0 p = 0 , c ( 1 / 2 ) = 3 / 4 c(1/2)=3/4 c ( 1/2 ) = 3/4 , h ( 1 / 2 ) = 4 3 7 h(1/2)=\frac{4}{3\sqrt{7}} h ( 1/2 ) = 3 7 4 , and m ( 1 / 2 ) + 1 / 2 = 3 2 m(1/2)+1/2=\frac{3}{2} m ( 1/2 ) + 1/2 = 2 3 , giving R r o u n d ( 1 / 2 ) = 2 7 < 1 \mathcal{R}_{\mathrm{round}}(1/2)=\frac{2}{\sqrt{7}}<1 R round ( 1/2 ) = 7 2 < 1 . Rigorous degree-6 rational Taylor-polynomial enclosures of arctan and ( 1 − t ) − 1 / 2 (1-t)^{-1/2} ( 1 − t ) − 1/2 (machine-checked over R \mathbb{R} R in AttainmentRoot.lean and verified independently in anc/verify_certificates.py) prove R r o u n d ( 0.457892036651 ) > 1 > R r o u n d ( 0.457892036652 ) \mathcal{R}_{\mathrm{round}}(0.457892036651)>1>\mathcal{R}_{\mathrm{round}}(0.457892036652) R round ( 0.457892036651 ) > 1 > R round ( 0.457892036652 ) . Furthermore, outward-rounded rational interval arithmetic in anc/verify_certificates.py certifies 1 + K − h ( m + q ) < 0 1+K-h(m+q)<0 1 + K − h ( m + q ) < 0 on ( 1 / 3 , 0.456 ] (1/3,0.456] ( 1/3 , 0.456 ] , d d q [ 1 + K − h ( m + q ) ] > 0 \frac{d}{dq}[1+K-h(m+q)]>0 d q d [ 1 + K − h ( m + q )] > 0 on [ 0.456 , 0.460 ] [0.456,0.460] [ 0.456 , 0.460 ] , and 1 + K − h ( m + q ) > 0 1+K-h(m+q)>0 1 + K − h ( m + q ) > 0 on [ 0.460 , 1 / 2 ] [0.460,1/2] [ 0.460 , 1/2 ] (global uniqueness of q a q_a q a on ( 1 / 3 , 1 / 2 ] (1/3,1/2] ( 1/3 , 1/2 ] ), together with strict auxiliary inequalities (11)–(13) across [ q a , 59 / 125 ] [q_a,59/125] [ q a , 59/125 ] . Inequality (10) is an equality at q a q_a q a and holds strictly to its right.
Theorem 3.6 (Exact Global Optimum for q a ≤ q < 1 q_a\le q<1 q a ≤ q < 1 ). For every speed ratio q ∈ [ q a , 1 ) q\in[q_a,1) q ∈ [ q a , 1 ) , the wall-optimal four-leg bowtie (6) captures every speed-q q q upward transit trajectory across [ − b ( q ) , b ( q ) ] × R [-b(q),b(q)]\times\mathbb{R} [ − b ( q ) , b ( q )] × R . Combining this with Theorem 2.1 , the exact minimax corridor width over all bounded-excursion patrols is
W o p t ( q ) = F ( q ) = 2 r 1 + 1 3 q 2 + 2 q − 1 , ∀ q ∈ [ q a , 1 ) . (16) W_{\mathrm{opt}}(q)=F(q)=2r\sqrt{1+\frac{1}{3q^{2}+2q-1}},\qquad\forall q\in[q_a,1). \tag*{(16)} W opt ( q ) = F ( q ) = 2 r 1 + 3 q 2 + 2 q − 1 1 , ∀ q ∈ [ q a , 1 ) . (16)
Proof. Theorem 2.1 gives W o p t ( q ) ≤ F ( q ) W_{\mathrm{opt}}(q)\le F(q) W opt ( q ) ≤ F ( q ) on ( 1 / 3 , 1 ) (1/3,1) ( 1/3 , 1 ) . On [ q a , 59 / 125 ] [q_a,59/125] [ q a , 59/125 ] , Lemma 3.5 and Theorem 3.4 prove that (6) guards [ − b ( q ) , b ( q ) ] × R [-b(q),b(q)]\times\mathbb{R} [ − b ( q ) , b ( q )] × R (where each right-wall downward visit covers straight-wall phase 2 R ( q ) + 2 ( 1 + q ) h ( q ) = 4 q ( h ( q ) + ℓ ( q ) ) = q T 2R(q)+2(1+q)h(q)=4q(h(q)+\ell(q))=qT 2 R ( q ) + 2 ( 1 + q ) h ( q ) = 4 q ( h ( q ) + ℓ ( q )) = q T with zero slack, while the interior fan-and-tube barrier blocks all diffracting paths). On [ 59 / 125 , 1 ) [59/125,1) [ 59/125 , 1 ) , Corollary 3.2 (3) proves that the same bowtie guards [ − b ( q ) , b ( q ) ] × R [-b(q),b(q)]\times\mathbb{R} [ − b ( q ) , b ( q )] × R via Theorem 3.1 . Since [ q a , 59 / 125 ] ∪ [ 59 / 125 , 1 ) = [ q a , 1 ) [q_a,59/125]\cup[59/125,1)=[q_a,1) [ q a , 59/125 ] ∪ [ 59/125 , 1 ) = [ q a , 1 ) , (16) holds throughout [ q a , 1 ) [q_a,1) [ q a , 1 ) .
Periodic approximation and escape from the canonical bowtie
Periodic approximation at fixed excursion height
Work in units with guard speed one and keep the capture radius r > 0 r>0 r > 0 explicit. Height is measured from the same fixed horizontal axis as in Definition 1.2 .
Theorem 4.1 (Periodic polygonal approximation). Suppose a prescribed patrol of height at most H < ∞ H<\infty H < ∞ guards a corridor of width W W W against every upward transit of speed at most q > 0 q>0 q > 0 . For every 0 < λ < 1 0<\lambda<1 0 < λ < 1 , there is a finite-piece periodic polygonal patrol of height at most H H H , with the same speed bounds and capture radius, guarding a corridor of width greater than λ W \lambda W λW . Consequently, periodic polygonal patrols give the same supremum at fixed H H H , and also after taking the supremum over H H H . The assertion holds for two-way guarding as well.
Proof. Let g g g be the given guard and choose r < s < R 0 < R < r / λ r<s<R_0<R<r/\lambda r < s < R 0 < R < r / λ . First, increasing the radius to s s s allows a positive increase in intruder speed. To see this, fix L > H + R L>H+R L > H + R . Choose c < 1 c<1 c < 1 sufficiently close to one that c L > H + r cL>H+r c L > H + r and ( 1 − c ) ( W / 2 ) 2 + L 2 < s − r (1-c)\sqrt{(W/2)^2+L^2}<s-r ( 1 − c ) ( W /2 ) 2 + L 2 < s − r . If a speed-q 1 = q / c q_1=q/c q 1 = q / c intruder escaped the radius-s s s guard, a piece of its path would travel from − L -L − L to L L L within the closed slab between these levels (take the last visit to − L -L − L before a first subsequent visit to L L L ). Scale this piece about the origin by c c c , keeping its time parameter. It has speed at most q q q , moves by less than s − r s-r s − r at each time, and stays strictly beyond the radius-r r r disk. Its endpoints at ± c L \pm cL ± c L admit safe vertical extensions to a full transit, contradicting guarding.
Choose 1 < A < q 1 / q 1<A<q_1/q 1 < A < q 1 / q . Take a time step Δ > 0 \Delta>0 Δ > 0 and then a spatial net radius η > 0 \eta>0 η > 0 so that
Δ / 2 + η < R 0 − s , A q Δ / 2 + η < R − R 0 , 1 + 2 η / Δ ≤ A , A q + 2 η / Δ ≤ q 1 . \Delta/2+\eta<R_0-s,\qquad A q\Delta/2+\eta<R-R_0,\qquad1+2\eta/\Delta\le A,\qquad A q+2\eta/\Delta\le q_1. Δ/2 + η < R 0 − s , A q Δ/2 + η < R − R 0 , 1 + 2 η /Δ ≤ A , A q + 2 η /Δ ≤ q 1 .
Use rectangular product nets in the guard box and in [ − W / 2 , W / 2 ] × [ − L , L ] [-W/2,W/2]\times[-L,L] [ − W /2 , W /2 ] × [ − L , L ] . Include the boundary rows and round each coordinate to a nearest grid coordinate. Denote the finite nets by G G G , E E E and the bottom and top rows of E E E by B , C B,C B , C . For i ≥ 0 i\ge0 i ≥ 0 let z i ∈ G z_i\in G z i ∈ G be the rounded g ( i Δ ) g(i\Delta) g ( i Δ ) . The interpolant of these vertices has speed at most A A A and stays within Δ / 2 + η \Delta/2+\eta Δ/2 + η of g g g . Indeed, at fractional time v v v the deviation of a speed-K K K path from the chord between its true endpoints is at most 2 v ( 1 − v ) K Δ 2v(1-v)K\Delta 2 v ( 1 − v ) K Δ .
Given consecutive guard vertices z , z ′ z,z' z , z ′ , allow an evader step e → e ′ e\to e' e → e ′ in E E E if ∣ e − e ′ ∣ ≤ q 1 Δ |e-e'|\le q_1\Delta ∣ e − e ′ ∣ ≤ q 1 Δ and the simultaneous affine interpolants maintain distance strictly greater than R 0 R_0 R 0 throughout the step. Write Post z , z ′ ( S ) \operatorname{Post}_{z,z'}(S) Post z , z ′ ( S ) for all destinations from S S S . Define S 0 = B S_0=B S 0 = B and
S i + 1 = B ∪ Post z i , z i + 1 ( S i ) . S_{i+1}=B\cup\operatorname{Post}_{z_i,z_{i+1}}(S_i). S i + 1 = B ∪ Post z i , z i + 1 ( S i ) .
A path reaching C C C in any S i S_i S i interpolates to a speed-q 1 q_1 q 1 transit avoiding the radius-s s s guard. The endpoints admit safe extensions since L > H + s L>H+s L > H + s . Thus every S i S_i S i avoids C C C . Some pair ( z i , S i ) (z_i,S_i) ( z i , S i ) repeats because G × 2 E G\times2^E G × 2 E is finite. Repeat the polygon between the repeated states on the whole real line. The intervening sets now give a periodic invariant containing B B B at every step and disjoint from C C C , including across the seam.
The periodic polygon g P g_P g P guards at radius R R R against speed A q Aq A q . Otherwise take a putative evader piece from − L -L − L to L L L , and extend its endpoints constantly to grid times. Round its samples to E E E . The rounded steps have length at most A q Δ + 2 η Aq\Delta+2\eta A q Δ + 2 η , and their interpolants differ from the evader by at most A q Δ / 2 + η < R − R 0 Aq\Delta/2+\eta<R-R_0 A q Δ/2 + η < R − R 0 . This produces a forbidden discrete path from B B B to C C C . Finally let β = r / R \beta=r/R β = r / R and put g ∗ ( t ) = β g P ( t / ( A β ) ) g_*(t)=\beta g_P(t/(A\beta)) g ∗ ( t ) = β g P ( t / ( A β )) . This scales the two speed bounds A , A q A,Aq A , A q to 1 , q 1,q 1 , q , the radius R R R to r r r , and the height to at most β H \beta H β H . The guarded width is β W > λ W \beta W>\lambda W β W > λW . For two-way transit, include in each finite state both the set reachable from B B B and the set reachable from C C C , with the opposite row excluded in each case.
Strict escape at the width of the wall-optimal bowtie
Use r = V = 1 r=V=1 r = V = 1 . For the canonical bowtie (6) write D = 3 q 2 + 2 q − 1 D=3q^{2}+2q-1 D = 3 q 2 + 2 q − 1 , c = a / ℓ c=a/\ell c = a / ℓ , d = h / ℓ d=h/\ell d = h / ℓ , and G = 2 q ℓ − ( 1 − q ) h G=2q\ell-(1-q)h G = 2 q ℓ − ( 1 − q ) h . Set time zero at the middle of the downward return at x = a x=a x = a . For t ≥ 0 t\geq0 t ≥ 0 the remaining right return ends at h h h , the next diagonal ends at h + 2 ℓ h+2\ell h + 2 ℓ , and the left return ends at 3 h + 2 ℓ 3h+2\ell 3 h + 2 ℓ . The guard satisfies g ( − t ) = ( g x ( t ) , − g y ( t ) ) g(-t)=(g_x(t),-g_y(t)) g ( − t ) = ( g x ( t ) , − g y ( t )) .
Theorem 4.2 (Escape with positive clearance). For every 1 / 3 < q < q a 1/3<q<q_a 1/3 < q < q a there are a speed-q q q upward transit in [ − b , b ] × R [-b,b]\times\mathbb{R} [ − b , b ] × R and a number ρ ( q ) > 1 \rho(q)>1 ρ ( q ) > 1 such that ∣ p ( t ) − g ( t ) ∣ ≥ ρ ( q ) \lvert p(t)-g(t)\rvert\geq\rho(q) ∣ p ( t ) − g ( t )∣ ≥ ρ ( q ) at every real t t t . The path can be taken vertical outside a finite time interval. Here g g g is the canonical wall-optimal bowtie at this q q q .
Proof. We use three explicit paths. In the first range 1 / 3 < q ≤ 0.4563 1/3<q\leq0.4563 1/3 < q ≤ 0.4563 let s = h ( 1 + q ) s=h(1+q) s = h ( 1 + q ) , k = ( d − q ) / c > 0 k=(d-q)/c>0 k = ( d − q ) / c > 0 , n = 1 + k 2 n=\sqrt{1+k^{2}} n = 1 + k 2 , and L = s / ( n + k ) L=s/(n+k) L = s / ( n + k ) . In the ( X , q t − Y ) (X,qt-Y) ( X , q t − Y ) plane, the right return centered at phase zero is the segment from ( a , − s ) (a,-s) ( a , − s ) to ( a , s ) (a,s) ( a , s ) . The adjacent diagonal lines have equations z = ± ( s − k ( a − x ) ) z=\pm(s-k(a-x)) z = ± ( s − k ( a − x )) ; their distance from ( a − L , 0 ) (a-L,0) ( a − L , 0 ) equals ( s − k L ) / n = L (s-kL)/n=L ( s − k L ) / n = L . The right return is distance L L L , and the left column is at least 2 a − L 2a-L 2 a − L away. Consecutive copies of each diagonal are translated vertically by P = 4 q ( h + ℓ ) = 2 ( s + G ) P=4q(h+\ell)=2(s+G) P = 4 q ( h + ℓ ) = 2 ( s + G ) . Their ordinates on [ − a , a ] [-a,a] [ − a , a ] have magnitude at most s s s before translation, since s − 2 a k = G > 0 s-2ak=G>0 s − 2 ak = G > 0 . Hence all other copies are at least P − s > s P-s>s P − s > s away. It follows that the straight lane p ( t ) = ( a − L , q t ) p(t)=(a-L,qt) p ( t ) = ( a − L , q t ) has clearance at least L L L whenever L ≤ a L\leq a L ≤ a . To check 1 < L < a 1<L<a 1 < L < a on this whole range, set w = q / ( 1 + q ) w=\sqrt{q}/(1+q) w = q / ( 1 + q ) . Substitution gives
L a = 1 − q 1 − 2 q + 2 q w < 1. \frac{L}{a}=\frac{1-q}{1-2q+2qw}<1. a L = 1 − 2 q + 2 q w 1 − q < 1.
The denominator is positive and w > 1 / 2 w>1/2 w > 1/2 . The derivative of this ratio has the sign of 1 − 2 w − 2 q ( 1 − q ) w ′ < 0 1-2w-2q(1-q)w'<0 1 − 2 w − 2 q ( 1 − q ) w ′ < 0 . Since a a a decreases and L ( 0.4563 ) > 1 L(0.4563)>1 L ( 0.4563 ) > 1 by rational radical enclosure, the assertion holds throughout.
For 0.4563 ≤ q ≤ 0.4578 0.4563\leq q\leq0.4578 0.4563 ≤ q ≤ 0.4578 , fix e = 39 / 500 e=39/500 e = 39/500 and ( u , v ) = q ( 2 e , 1 − e 2 ) / ( 1 + e 2 ) (u,v)=q(2e,1-e^{2})/(1+e^{2}) ( u , v ) = q ( 2 e , 1 − e 2 ) / ( 1 + e 2 ) , so u 2 + v 2 = q 2 u^{2}+v^{2}=q^{2} u 2 + v 2 = q 2 . Define
w = 1 + v , z = d − v , j = c + u , M = u 2 + w 2 , N = j 2 + z 2 , w=1+v,\qquad z=d-v,\qquad j=c+u,\qquad M=\sqrt{u^{2}+w^{2}},\qquad N=\sqrt{j^{2}+z^{2}}, w = 1 + v , z = d − v , j = c + u , M = u 2 + w 2 , N = j 2 + z 2 ,
R = h ( ( 1 + d ) u + c w ) N + z M / w , x 0 = a − R M / w . R=\frac{h((1+d)u+cw)}{N+zM/w},\qquad x_{0}=a-RM/w. R = N + z M / w h (( 1 + d ) u + c w ) , x 0 = a − R M / w .
Use p ( t ) = ( x 0 + u min { ∣ t ∣ , S } , v t ) p(t)=(x_{0}+u\min\{\lvert t\rvert,S\},vt) p ( t ) = ( x 0 + u min {∣ t ∣ , S } , v t ) , where S = 21 / ( 10 v ) S=21/(10v) S = 21/ ( 10 v ) . On the right return the complete relative line is ( x 0 − a + u t , w t ) (x_{0}-a+ut,wt) ( x 0 − a + u t , w t ) for t ≥ 0 t\geq0 t ≥ 0 , at perpendicular distance ( a − x 0 ) w / M = R (a-x_{0})w/M=R ( a − x 0 ) w / M = R from the origin. At departure its displacement is ( x 0 − a + u h , w h ) (x_{0}-a+uh,wh) ( x 0 − a + u h , w h ) ; the succeeding relative line has velocity ( j , − z ) (j,-z) ( j , − z ) . Its signed normal distance is [ h ( ( 1 + d ) u + c w ) − ( a − x 0 ) z ] / N = R [h((1+d)u+cw)-(a-x_{0})z]/N=R [ h (( 1 + d ) u + c w ) − ( a − x 0 ) z ] / N = R . The exact inequalities for the remaining pieces are
R > 1 + 1 / 3000 , − b < x 0 , x 0 + u S < b , a + x 0 > R , R>1+1/3000,\qquad-b<x_{0},\qquad x_{0}+uS<b,\qquad a+x_{0}>R, R > 1 + 1/3000 , − b < x 0 , x 0 + u S < b , a + x 0 > R ,
h + 2 ℓ < S < 3 h + 2 ℓ , 21 / 10 − h > R . h+2\ell<S<3h+2\ell,\qquad21/10-h>R. h + 2 ℓ < S < 3 h + 2 ℓ , 21/10 − h > R .
These hold by rational interval substitution on 64 equal subintervals of [ 0.4563 , 0.4578 ] [0.4563,0.4578] [ 0.4563 , 0.4578 ] . Until the tail starts, the left return is horizontally at least a + x 0 a+x_{0} a + x 0 away. Thereafter the vertical separation is at least 21 / 10 − h 21/10-h 21/10 − h . Reflection supplies negative times.
Finally take 0.4578 ≤ q < q a 0.4578\leq q<q_a 0.4578 ≤ q < q a . Put α = 1 − q 2 \alpha=\sqrt{1-q^{2}} α = 1 − q 2 , m = q d + α c m=qd+\alpha c m = q d + α c , κ = α d − q c \kappa=\alpha d-qc κ = α d − q c , ( u , v ) = q ( κ , m ) (u,v)=q(\kappa,m) ( u , v ) = q ( κ , m ) , and
θ ∗ = arctan u 1 + v , J ( θ ) = ∫ 0 θ d ϕ cos ϕ + q 2 − sin 2 ϕ , K = κ cos θ ∗ − m sin θ ∗ + ( m + q ) J ( θ ∗ ) . \theta_{*}=\arctan\frac{u}{1+v},\qquad J(\theta)=\int_{0}^{\theta}\frac{d\phi}{\cos\phi+\sqrt{q^{2}-\sin^{2}\phi}},\qquad K=\kappa\cos\theta_{*}-m\sin\theta_{*}+(m+q)J(\theta_{*}). θ ∗ = arctan 1 + v u , J ( θ ) = ∫ 0 θ cos ϕ + q 2 − sin 2 ϕ d ϕ , K = κ cos θ ∗ − m sin θ ∗ + ( m + q ) J ( θ ∗ ) .
Take R = h ( m + q ) / ( 1 + K ) = R r o u n d ( q ) > 1 R = h(m+q)/(1+K) = R_{\mathrm{round}}(q) > 1 R = h ( m + q ) / ( 1 + K ) = R round ( q ) > 1 by Lemma 3.5 , and put τ = R J ( θ ∗ ) \tau= RJ(\theta_*) τ = R J ( θ ∗ ) . For 0 ≤ t ≤ τ 0 \le t \le\tau 0 ≤ t ≤ τ let
t = R J ( θ ) , p ( t ) = ( a − R cos θ , R sin θ − t ) . t = RJ(\theta), \qquad p(t) = (a-R\cos\theta,R\sin\theta-t). t = R J ( θ ) , p ( t ) = ( a − R cos θ , R sin θ − t ) .
This arc has constant distance R R R from the guard. Setting w = R θ ˙ = cos θ + q 2 − sin 2 θ w=R\dot{\theta}=\cos\theta+\sqrt{q^2-\sin^2\theta} w = R θ ˙ = cos θ + q 2 − sin 2 θ gives
∣ p ′ ∣ 2 = w 2 − 2 w cos θ + 1 = q 2 . \lvert p'\rvert^2=w^2-2w\cos\theta+1=q^2. ∣ p ′ ∣ 2 = w 2 − 2 w cos θ + 1 = q 2 .
At release the velocity is ( u , v ) (u,v) ( u , v ) : the vector ( u , 1 + v ) (u,1+v) ( u , 1 + v ) has direction ( sin θ ∗ , cos θ ∗ ) (\sin\theta_*,\cos\theta_*) ( sin θ ∗ , cos θ ∗ ) and length equal to the positive branch w w w . Continue on its absolute tangent
p ( t ) = ( a − R cos θ ∗ + u ( t − τ ) , v t + B ) , B = R sin θ ∗ − ( 1 + v ) τ . p(t)=(a-R\cos\theta_*+u(t-\tau),vt+B), \qquad B=R\sin\theta_*-(1+v)\tau. p ( t ) = ( a − R cos θ ∗ + u ( t − τ ) , v t + B ) , B = R sin θ ∗ − ( 1 + v ) τ .
This is tangent to the return circle in relative coordinates. On the next diagonal the normal ( κ , m ) (\kappa,m) ( κ , m ) is unit and − c κ + d m = q -c\kappa+dm=q − c κ + d m = q ; both guard and intruder have normal velocity q q q . Their constant signed separation is h ( m + q ) − R K = R h(m+q)-RK=R h ( m + q ) − R K = R . At S = ( 21 / 10 − B ) / v S=(21/10-B)/v S = ( 21/10 − B ) / v freeze the horizontal coordinate and continue vertically. The following checks hold on the closed interval [ 0.4578 , 0.4579 ] [0.4578,0.4579] [ 0.4578 , 0.4579 ] , which contains the required range:
0 < τ < h , a > R , h + 2 ℓ < S < 3 h + 2 ℓ , 21 / 10 − h > R , 0<\tau<h,\qquad a>R,\qquad h+2\ell<S<3h+2\ell,\qquad21/10-h>R, 0 < τ < h , a > R , h + 2 ℓ < S < 3 h + 2 ℓ , 21/10 − h > R ,
− b < a − R , a − R cos θ ∗ + u ( S − τ ) < b . -b<a-R,\qquad a-R\cos\theta_*+u(S-\tau)<b. − b < a − R , a − R cos θ ∗ + u ( S − τ ) < b .
The horizontal coordinate increases up to the tail. On the left return the horizontal separation before the tail is at least 2 a − R > R 2a-R>R 2 a − R > R ; after the tail the vertical separation suffices. Reflect the whole path in time and height to complete the transit.
For a reproducible enclosure of the rounded-arc time let σ = sin θ ∗ \sigma=\sin\theta_* σ = sin θ ∗ . Substitution sin ϕ = σ t \sin\phi=\sigma t sin ϕ = σ t gives
J ( θ ∗ ) = σ 1 − q 2 ( 1 − q ∫ 0 1 1 − ( σ 2 / q 2 ) t 2 1 − σ 2 t 2 d t ) . J(\theta_*)=\frac{\sigma}{1-q^2}\left(1-q\int_0^1\frac{\sqrt{1-(\sigma^2/q^2)t^2}}{\sqrt{1-\sigma^2t^2}}\,dt\right). J ( θ ∗ ) = 1 − q 2 σ ( 1 − q ∫ 0 1 1 − σ 2 t 2 1 − ( σ 2 / q 2 ) t 2 d t ) .
For both binomial factors the parameter x x x lies in [ 0 , 1 / 10 ) [0,1/10) [ 0 , 1/10 ) . Truncate each after degree nine in t 2 t^2 t 2 , multiply, and integrate the polynomial term by term. The absolute coefficient bounds by one give each factor a uniform remainder at most x 10 / ( 1 − x ) x^{10}/(1-x) x 10 / ( 1 − x ) . Both factors and truncated factors have magnitude below two, so twice the sum of the two remainders bounds the product error. All radicals use outward integer-square-root enclosures. The read-only ancillary checker anc/escape/verify_escape.py replays every range, including the seven V inequalities and finite-piece rounded margins.
For each fixed q q q the positive clearance gives a robust finite crossing as well. Truncate the vertical tails at heights beyond ± ( h + 2 ) \pm(h+2) ± ( h + 2 ) . On this compact interval small uniform changes of the guard preserve avoidance. If the wall positions move slightly, projection of the intruder’s horizontal coordinate onto the new corridor preserves its speed bound and costs at most the wall displacement in clearance. Safe tail extensions are available for uniformly bounded guards when the truncation heights are outside their full sensing slab.
Strict loss below the diffraction threshold
At the straight-wall ceiling the optimal bowtie admits an escaping transit for every 1 / 3 < q < q a 1/3<q<q_a 1/3 < q < q a by Theorem 4.2 . To pass from this single patrol to a statement about the supremum, we must also rule out patrols with increasing periods or increasing excursion.
Theorem 5.1 (Gap uniform over all periods). For every 1 / 3 < q < q a 1/3<q<q_a 1/3 < q < q a there is an ϵ ( q ) > 0 \epsilon(q)>0 ϵ ( q ) > 0 such that
W o p t ( q ) ≤ F ( q ) − r ϵ ( q ) . W_{\mathrm{opt}}(q)\le F(q)-r\epsilon(q). W opt ( q ) ≤ F ( q ) − r ϵ ( q ) . Together with Theorem 3.6 , this identifies [ q a , 1 ) [q_a,1) [ q a , 1 ) as the full set of speeds in ( 1 / 3 , 1 ) (1/3,1) ( 1/3 , 1 ) where the straight-wall ceiling is attained as a supremum.
Theorem 5.2 (Quadratic loss at the threshold). There are constants k 1 , k 2 , η > 0 k_1,k_2,\eta> 0 k 1 , k 2 , η > 0 such that
k 1 r ( q a − q ) 2 ≤ F ( q ) − W o p t ( q ) ≤ k 2 r ( q a − q ) 2 ( q a − η < q < q a ) . k_1r(q_a-q)^2 \le F(q)-W_{\mathrm{opt}}(q) \le k_2r(q_a-q)^2 \qquad(q_a-\eta<q<q_a). k 1 r ( q a − q ) 2 ≤ F ( q ) − W opt ( q ) ≤ k 2 r ( q a − q ) 2 ( q a − η < q < q a ) . The upper inequality holds with k 2 = 175 k_2=175 k 2 = 175 throughout [ 0.4578 , q a ] [0.4578,q_a] [ 0.4578 , q a ] .
Control of ordered wall contacts
Use units r = V = 1 r=V=1 r = V = 1 and restrict q q q to a compact interval J ⊂ ( 1 / 3 , 1 / 2 ) J\subset(1/3,1/2) J ⊂ ( 1/3 , 1/2 ) . Write
D = 3 q 2 + 2 q − 1 , b ∗ = 1 + D − 1 , a ∗ = 1 D ( 1 + D ) , R ∗ = 1 1 + D , D=3q^2+2q-1,\qquad b_*=\sqrt{1+D^{-1}},\qquad a_*=\frac{1}{\sqrt{D(1+D)}},\qquad R_*=\frac{1}{\sqrt{1+D}}, D = 3 q 2 + 2 q − 1 , b ∗ = 1 + D − 1 , a ∗ = D ( 1 + D ) 1 , R ∗ = 1 + D 1 ,
c = D 2 q , d = 1 − q 2 q , ℓ ∗ = a ∗ / c , h ∗ = d ℓ ∗ , T ∗ = 4 ( h ∗ + ℓ ∗ ) . c=\frac{\sqrt{D}}{2q},\qquad d=\frac{1-q}{2q},\qquad\ell_* = a_*/c,\qquad h_*=d\ell_*,\qquad T_*=4(h_*+\ell_*). c = 2 q D , d = 2 q 1 − q , ℓ ∗ = a ∗ / c , h ∗ = d ℓ ∗ , T ∗ = 4 ( h ∗ + ℓ ∗ ) .
These are the parameters of (6) , and c 2 + d 2 = 1 c^2+d^2=1 c 2 + d 2 = 1 . In each half-cycle the bowtie descends at one pole for 2 h ∗ 2h_* 2 h ∗ seconds and transfers to the other pole at velocity ( ± c , d ) (\pm c,d) ( ± c , d ) for 2 ℓ ∗ 2\ell_* 2 ℓ ∗ seconds.
Let g = ( X , Y ) g=(X,Y) g = ( X , Y ) be a periodic guard detecting both straight-wall families, with period T T T , half-width b b b , P = q T P=qT P = q T , α = 1 − q 2 \alpha=\sqrt{1-q^2} α = 1 − q 2 and Γ = ( α X , z ) \Gamma=(\alpha X,z) Γ = ( α X , z ) , where z = q t − Y z=qt-Y z = q t − Y . Theorem 2.1 gives Δ = b ∗ − b ≥ 0 \Delta=b_*-b\ge0 Δ = b ∗ − b ≥ 0 . We work with Δ \Delta Δ small enough that b > 1 b>1 b > 1 .
Lemma 5.3 (Ordered comparison at the original radius). There is a speed-bounded periodic comparison patrol with the same entire lifted phase image as g g g , up to a phase translation. It has an even number N N N of alternating wall groups, with chronological markers t i − ≤ t i + ≤ t i + 1 − t_i^-\le t_i^+\le t_{i+1}^- t i − ≤ t i + ≤ t i + 1 − . Write
p i ± = z ( t i ± ) , x i ± = ∣ X ( t i ± ) ∣ , r i ± = 1 − ( b − x i ± ) 2 p_i^\pm=z(t_i^\pm),\qquad x_i^\pm=|X(t_i^\pm)|,\qquad r_i^\pm=\sqrt{1-(b-x_i^\pm)^2} p i ± = z ( t i ± ) , x i ± = ∣ X ( t i ± ) ∣ , r i ± = 1 − ( b − x i ± ) 2 for the comparison patrol. Then p i − ≤ p i + ≤ p i + 2 − p_i^-\le p_i^+\le p_{i+2}^- p i − ≤ p i + ≤ p i + 2 − , p i + N − = p i − + P p_{i+N}^-=p_i^-+P p i + N − = p i − + P , and the periodically extended intervals [ p i − − r i − , p i + + r i + ] [p_i^- - r_i^-,p_i^+ + r_i^+] [ p i − − r i − , p i + + r i + ] cover each wall’s phase line.
Proof. Fix the original period. Apply the construction of Lemma 2.3 with excess radius tending to zero. Horizontal compression, polygonal approximation and smoothing in finitely many disks may each have arbitrarily small two-sided Hausdorff error. Thus the embedded comparison images converge to the entire original cylinder image. Their lengths are at most the original length, so Lemma 2.2 lifts each to a speed-bounded patrol of period T T T . Choose the outermost contacts, groups and chronological markers as in the periodic proof of Theorem 2.1 . Start the period at a left marker and subtract the initial physical height. The resulting translations of phase modulo P P P have a convergent subsequence, and the normalized patrols converge uniformly along a further subsequence by equicontinuity. At excess radius ϵ \epsilon ϵ the opposite active zones are separated by 2 ( b − 1 − ϵ ) 2(b-1-\epsilon) 2 ( b − 1 − ϵ ) , so the number of groups is bounded for this fixed period. Extract a constant count and convergent marker times. Closed interval coverage and the lifted order pass to the limit. Its cylinder image agrees with the translated original image. Each patrol advances by P P P per period, so its entire lifted image consists of all the translates of its image over one period. This yields equality of the lifted images as well.
On the comparison patrol put u i = t i + − t i − u_i=t_i^+-t_i^- u i = t i + − t i − , v i = t i + 1 − − t i + v_i=t_{i+1}^--t_i^+ v i = t i + 1 − − t i + and
z i = Y ( t i − ) − Y ( t i + ) , w i = Y ( t i + 1 − ) − Y ( t i + ) , l i = x i + + x i + 1 − , A = ∑ i u i , Z = ∑ i z i = ∑ i w i . z_i=Y(t_i^-)-Y(t_i^+),\qquad w_i=Y(t_{i+1}^-)-Y(t_i^+),\qquad l_i=x_i^++x_{i+1}^-,\qquad A=\sum_i u_i,\qquad Z=\sum_i z_i=\sum_i w_i. z i = Y ( t i − ) − Y ( t i + ) , w i = Y ( t i + 1 − ) − Y ( t i + ) , l i = x i + + x i + 1 − , A = i ∑ u i , Z = i ∑ z i = i ∑ w i .
Use j j j to index all 2 N 2N 2 N marker endpoints. Define the coverage excess O O O , the marker defects L j L_j L j , and the two speed defects f i , e i f_i,e_i f i , e i by
O = q A + Z + ∑ j r j − 2 q T , L j = D ( x j + Δ ) − r j , O=qA+Z+\sum_j r_j-2qT,\qquad L_j=\sqrt{D}(x_j+\Delta)-r_j, O = q A + Z + j ∑ r j − 2 q T , L j = D ( x j + Δ ) − r j ,
f i = u i − z i , e i = v i − c l i − d w i . f_i=u_i-z_i,\qquad e_i=v_i-cl_i-dw_i. f i = u i − z i , e i = v i − c l i − d w i .
Coverage gives O ≥ 0 O\ge0 O ≥ 0 and the speed bound gives f i , e i ≥ 0 f_i,e_i\ge0 f i , e i ≥ 0 . At a ∗ a_* a ∗ the graph of x ↦ 1 − ( b ∗ − x ) 2 x\mapsto\sqrt{1-(b_*-x)^2} x ↦ 1 − ( b ∗ − x ) 2 has tangent D x \sqrt{D}x D x . Its second derivative is at most − 1 -1 − 1 on the interior of the active interval, so L j ≥ ( x j + Δ − a ∗ ) 2 / 2 L_j\ge(x_j+\Delta-a_*)^2/2 L j ≥ ( x j + Δ − a ∗ ) 2 /2 , including the endpoints by continuity. Using ∑ i l i = ∑ j x j \sum_i l_i=\sum_j x_j ∑ i l i = ∑ j x j , 2 q c = D 2qc=\sqrt{D} 2 q c = D and q + 2 q d = 1 q+2qd=1 q + 2 q d = 1 gives the exact identity
O + ∑ j L j + q ∑ i f i + 2 q ∑ i e i = 2 N D Δ . (17) O+\sum_j L_j+q\sum_i f_i+2q\sum_i e_i=2N\sqrt{D}\,\Delta. \tag*{(17)} O + j ∑ L j + q i ∑ f i + 2 q i ∑ e i = 2 N D Δ. (17)
Lemma 5.4 (Mean error on each pair of legs). With
H i = ∣ x i − − a ∗ ∣ + ∣ x i + − a ∗ ∣ + ∣ u i − 2 h ∗ ∣ + ∣ v i − 2 ℓ ∗ ∣ + u i f i + v i e i , H_i=|x_i^- -a_*|+|x_i^+ -a_*|+|u_i-2h_*|+|v_i-2\ell_*|+\sqrt{u_i f_i}+\sqrt{v_i e_i}, H i = ∣ x i − − a ∗ ∣ + ∣ x i + − a ∗ ∣ + ∣ u i − 2 h ∗ ∣ + ∣ v i − 2 ℓ ∗ ∣ + u i f i + v i e i , one has ∑ i H i ≤ C J N Δ \sum_i H_i\le C_JN\sqrt{\Delta} ∑ i H i ≤ C J N Δ for sufficiently small Δ \Delta Δ . In particular ∣ T − N T ∗ / 2 ∣ ≤ C J N Δ |T-NT_*/2|\le C_JN\sqrt{\Delta} ∣ T − N T ∗ /2∣ ≤ C J N Δ .
Proof. Constants in the proof are uniform over J J J . By (17) and Cauchy–Schwarz, ∑ j ∣ x j − a ∗ ∣ ≤ C N Δ \sum_j|x_j-a_*|\le CN\sqrt{\Delta} ∑ j ∣ x j − a ∗ ∣ ≤ C N Δ . Since r j = D ( x j + Δ ) − L j r_j=\sqrt{D}(x_j+\Delta)-L_j r j = D ( x j + Δ ) − L j and R ∗ = D a ∗ R_*=\sqrt{D}a_* R ∗ = D a ∗ , the same bound holds for ∑ j ∣ r j − R ∗ ∣ \sum_j|r_j-R_*| ∑ j ∣ r j − R ∗ ∣ . Let σ i \sigma_i σ i be the sign of the horizontal transfer displacement. The physical speed bound gives
∫ t r a n s f e r i ∣ ( σ i X ′ , Y ′ ) − ( c , d ) ∣ 2 d t ≤ 2 e i , ∫ r e t u r n i ∣ ( X ′ , Y ′ ) − ( 0 , − 1 ) ∣ 2 d t ≤ 2 f i . \int_{\mathrm{transfer}\ i}|(\sigma_iX',Y')-(c,d)|^2\,dt\le2e_i,\qquad
\int_{\mathrm{return}\ i}|(X',Y')-(0,-1)|^2\,dt\le2f_i. ∫ transfer i ∣ ( σ i X ′ , Y ′ ) − ( c , d ) ∣ 2 d t ≤ 2 e i , ∫ return i ∣ ( X ′ , Y ′ ) − ( 0 , − 1 ) ∣ 2 d t ≤ 2 f i .
Also v i ≤ ( 2 c b + e i ) / ( 1 − d ) v_i\le(2cb+e_i)/(1-d) v i ≤ ( 2 c b + e i ) / ( 1 − d ) , so ∑ i v i ≤ C N \sum_i v_i\le CN ∑ i v i ≤ C N . Applying Cauchy–Schwarz to the first integral and summing bounds ∑ i ( ∣ c v i − l i ∣ + ∣ d v i − w i ∣ ) \sum_i(|cv_i-l_i|+|dv_i-w_i|) ∑ i ( ∣ c v i − l i ∣ + ∣ d v i − w i ∣ ) by C N Δ CN\sqrt{\Delta} C N Δ . Thus for K i = q v i − w i K_i=qv_i-w_i K i = q v i − w i , K ∗ = 2 q ℓ ∗ − 2 h ∗ K_*=2q\ell_*-2h_* K ∗ = 2 q ℓ ∗ − 2 h ∗ ,
∑ i ( ∣ v i − 2 ℓ ∗ ∣ + ∣ K i − K ∗ ∣ ) ≤ C N Δ . \sum_i(|v_i-2\ell_*|+|K_i-K_*|)\le CN\sqrt{\Delta}. i ∑ ( ∣ v i − 2 ℓ ∗ ∣ + ∣ K i − K ∗ ∣ ) ≤ C N Δ .
Put G i = p i + 2 − − p i + ≥ 0 G_i=p_{i+2}^- -p_i^+\ge0 G i = p i + 2 − − p i + ≥ 0 . Replacing every endpoint radius by R ∗ R_* R ∗ uncovers total measure at most ∑ j ∣ r j − R ∗ ∣ \sum_j|r_j-R_*| ∑ j ∣ r j − R ∗ ∣ over the two phase circles. After the replacement both endpoints are ordered on each wall, so the uncovered measure equals ∑ i ( G i − 2 R ∗ ) + \sum_i(G_i-2R_*)_+ ∑ i ( G i − 2 R ∗ ) + . Telescoping on both walls gives
∑ i ( G i − 2 R ∗ ) = 2 P − q A − Z − 2 N R ∗ = ∑ j ( r j − R ∗ ) − O . \sum_i(G_i-2R_*)=2P-qA-Z-2NR_*=\sum_j(r_j-R_*)-O. i ∑ ( G i − 2 R ∗ ) = 2 P − q A − Z − 2 N R ∗ = j ∑ ( r j − R ∗ ) − O .
These two estimates imply ∑ i ∣ G i − 2 R ∗ ∣ ≤ C N Δ \sum_i|G_i-2R_*|\le CN\sqrt{\Delta} ∑ i ∣ G i − 2 R ∗ ∣ ≤ C N Δ . Finally,
G i − 1 = K i − 1 + ( 1 + q ) u i − f i + K i , R ∗ − K ∗ = ( 1 + q ) h ∗ , G_{i-1}=K_{i-1}+(1+q)u_i-f_i+K_i,\qquad R_*-K_*=(1+q)h_*, G i − 1 = K i − 1 + ( 1 + q ) u i − f i + K i , R ∗ − K ∗ = ( 1 + q ) h ∗ ,
so ∑ i ∣ u i − 2 h ∗ ∣ ≤ C N Δ \sum_i|u_i-2h_*|\le CN\sqrt{\Delta} ∑ i ∣ u i − 2 h ∗ ∣ ≤ C N Δ . In particular ∑ i u i ≤ C N \sum_i u_i\le CN ∑ i u i ≤ C N . Cauchy–Schwarz and (17) now bound the two square-root sums in ∑ i H i \sum_iH_i ∑ i H i by C N Δ CN\sqrt{\Delta} C N Δ . ∎
Figure 2. Two successive phase intervals of the ordinary bowtie at q = 0.45785 q = 0.45785 q = 0.45785 . The horizontal axis is lifted phase z = q t − Y z = qt - Y z = q t − Y , and the vertical axis is the scaled transverse position α X \alpha X α X . A transfer moves backward in phase while a downward return moves forward. Each cut through a left-return midpoint meets the curve once, and consecutive midpoint cuts differ by the period advance q T ∗ qT_\ast q T ∗ .
Recovery of the original patrol in a phase interval
The comparison may have placed points of the original image in a different temporal order. We recover the original timing over a long interval and control all visits from times outside it.
Lemma 5.5 (A synchronized crossing). Fix a positive integer m m m and a compact J ⊂ ( 1 / 3 , 1 / 2 ) J \subset(1/3,1/2) J ⊂ ( 1/3 , 1/2 ) . For Δ ≥ 0 \Delta\ge0 Δ ≥ 0 sufficiently small, the original patrol has a lifted phase interval I = [ L , U ] I = [L,U] I = [ L , U ] and a crossing [ t 0 , t 1 ] [t_0,t_1] [ t 0 , t 1 ] with the following properties. The width U − L U-L U − L differs from m q T ∗ mqT_\ast m q T ∗ by at most C Δ C\sqrt{\Delta} C Δ . The duration t 1 − t 0 t_1-t_0 t 1 − t 0 differs from m T ∗ mT_\ast m T ∗ by at most C Δ C\sqrt{\Delta} C Δ . After the changes of origin t ′ = t − t 0 t' = t-t_0 t ′ = t − t 0 and Y ′ = Y − ( q t 0 − L ) Y' = Y-(qt_0-L) Y ′ = Y − ( q t 0 − L ) the original patrol is within C Δ C\sqrt{\Delta} C Δ in physical position of the ordinary bowtie started at its left-return midpoint ( − a ∗ , 0 ) (-a_\ast,0) ( − a ∗ , 0 ) throughout the crossing. Its total time at phases in ( L , U ) (L,U) ( L , U ) outside the crossing is at most C Δ C\sqrt{\Delta} C Δ . The constant C C C depends only on J , m J,m J , m .
Proof. Number the left groups 2 j 2j 2 j . Their time and phase cuts are s j = ( t 2 j − + t 2 j + ) / 2 s_j=(t^-_{2j}+t^+_{2j})/2 s j = ( t 2 j − + t 2 j + ) /2 and C j = ( p 2 j − + p 2 j + ) / 2 C_j=(p^-_{2j}+p^+_{2j})/2 C j = ( p 2 j − + p 2 j + ) /2 . Both lists are nondecreasing. Repeat the original patrol and its ordered comparison over 6 m 6m 6 m copies of the period and use the enlarged phase circumference. Their lifted images still agree, all sums and counts scale by the same factor, and the resulting N / 2 N/2 N /2 left cycles divide exactly into blocks of m m m cycles, with at least six blocks. Use T , N , P T,N,P T , N , P henceforth for the enlarged quantities. For each block between successive chosen left midpoints, assign the time interval between its s s s cuts and the phase slab between its C C C cuts. Let H [ j ] H[j] H [ j ] be the sum of H i H_i H i over the groups touching the j j j th block, including both partial endpoint groups. Then ∑ j H [ j ] ≤ C N Δ \sum_j H[j] \le CN\sqrt{\Delta} ∑ j H [ j ] ≤ C N Δ .
Here are the relevant properties of the ideal m m m -cycle arc in phase space. Set S = 2 ( 1 + q ) h ∗ S=2(1+q)h_\ast S = 2 ( 1 + q ) h ∗ and K = K ∗ < 0 K=K_\ast<0 K = K ∗ < 0 . From the start of a left return the left and right column phases are [ 0 , S ] [0,S] [ 0 , S ] and [ S + K , 2 S + K ] [S+K,2S+K] [ S + K , 2 S + K ] . At horizontal fraction x ∈ [ 0 , 1 ] x\in[0,1] x ∈ [ 0 , 1 ] measured from the left pole, the two transfer phases are S + K x S+Kx S + K x and P 0 − K x P_0-Kx P 0 − K x , where P 0 = q T ∗ = 2 ( S + K ) P_0=qT_\ast=2(S+K) P 0 = q T ∗ = 2 ( S + K ) . The difference and its complement modulo P 0 P_0 P 0 lie between − 2 K = R ∗ > 0 -2K=R_\ast>0 − 2 K = R ∗ > 0 and S < P 0 S<P_0 S < P 0 . The finite lifted ideal curve is therefore simple. The phase line at the left midpoint meets it only there. The closest transfer is S / 2 + K = R ∗ > 0 S/2+K=R_\ast>0 S /2 + K = R ∗ > 0 away in phase. Between two such midpoint cuts the intervening arc stays inside the slab, with the return at fixed X = − a ∗ X=-a_\ast X = − a ∗ as its only part within R ∗ / 2 R_\ast/2 R ∗ /2 of either boundary.
Choose a small constant κ 0 > 0 \kappa_0>0 κ 0 > 0 . A block is regular when H [ j ] ≤ κ 0 H[j]\le\kappa_0 H [ j ] ≤ κ 0 . For each regular block, translate its initial C C C cut to zero and align the time on each of its legs linearly with the ideal leg. The comparison’s phase path is within C H [ j ] CH[j] C H [ j ] of the corresponding ideal arc in this parametrization. Indeed the integrated direction errors bound every partial displacement error by 2 u i f i \sqrt{2u_i f_i} 2 u i f i or 2 v i e i \sqrt{2v_i e_i} 2 v i e i , and aligning a leg changes the ideal displacement by at most its duration error. The initial horizontal error comes from the marker errors and the partial return displacement. Linear interpolation of z z z on a return also shows ∣ z ( s j ) − C j ∣ ≤ C H 2 j |z(s_j)-C_j|\le CH_{2j} ∣ z ( s j ) − C j ∣ ≤ C H 2 j . Summing over this fixed number of pieces proves the assertion, including a slab-width error at most C H [ j ] CH[j] C H [ j ] .
All exceptional blocks together number at most C N Δ CN\sqrt{\Delta} C N Δ . They occupy at most C N Δ CN\sqrt{\Delta} C N Δ comparison time by Lemma 5.4 , including exceptionally long legs. Project the images of their time intervals onto the phase circle and enlarge the projected intervals by one unit at each end. Their total length is at most C N Δ CN\sqrt{\Delta} C N Δ since ∣ z ′ ∣ ≤ 1 + q |z'|\le1+q ∣ z ′ ∣ ≤ 1 + q . Discard regular slabs meeting these enlarged intervals and regular slabs adjacent in the block list to an exceptional block. For small κ 0 \kappa_0 κ 0 , regular slab widths have a uniform positive lower bound l = 1 2 m min q ∈ J q T ∗ l=\frac{1}{2}m\min_{q\in J}qT_* l = 2 1 m min q ∈ J q T ∗ . Their interiors are disjoint by the ordered C C C cuts. A projected interval of length L L L can therefore meet at most L ′ / l + 4 L'/l+4 L ′ / l + 4 regular slabs, splitting at the seam if needed. Altogether O ( N Δ ) O(N\sqrt{\Delta}) O ( N Δ ) blocks are discarded. For small Δ \Delta Δ the retained count is at least a positive constant times N N N .
For a retained slab j j j let E j = H [ j − 1 ] + H [ j ] + H [ j + 1 ] E_j=H[j-1]+H[j]+H[j+1] E j = H [ j − 1 ] + H [ j ] + H [ j + 1 ] . Its entire original phase image in the closed slab lies in a C E j CE_j C E j neighborhood of its translated ideal arc. To see this, use equality of the full lifted images from Lemma 5.3, absorbing the constant translation in the choice of cuts. The discarded exceptional time blocks cannot visit the slab. Any regular time block stays within C H [ ⋅ ] CH[\cdot] C H [ ⋅ ] of its own slab. Choose κ 0 \kappa_0 κ 0 so small that this error cannot cross an intervening regular neighbor. Both immediate neighbors of a retained slab are regular. They can enter it only within C E j CE_j C E j of the shared boundary in phase, where the ideal neighbor has only its endpoint return at fixed transverse position. On this leg the distance to the endpoint equals the phase difference. Including the cut-width error gives the claimed C E j CE_j C E j confinement and places every visit to a boundary within C E j CE_j C E j of its specified ideal endpoint. The argument applies in every lifted copy.
In one lift I = [ L , U ] I=[L,U] I = [ L , U ] let t 1 t_1 t 1 be the original patrol’s first hit of U U U and t 0 t_0 t 0 the last hit of L L L before t 1 t_1 t 1 . These exist because Y Y Y is bounded and z ( t ) → ± ∞ z(t)\to\pm\infty z ( t ) → ± ∞ at the corresponding time ends. The crossing stays strictly inside I I I on ( t 0 , t 1 ) (t_0,t_1) ( t 0 , t 1 ) . Write
D I = t 1 − t 0 , M I = ∫ R 1 { L < z ( t ) < U } d t . D_I=t_1-t_0,\qquad M_I=\int_{\mathbb{R}}\mathbf{1}_{\{L<z(t)<U\}}\,dt. D I = t 1 − t 0 , M I = ∫ R 1 { L < z ( t ) < U } d t .
Translation of I I I by P P P translates all its visits by T T T , so M I M_I M I equals the time in all the open copies over one period. Hence ∑ r e t a i n e d M I ≤ T \sum_{\mathrm{retained}}M_I\le T ∑ retained M I ≤ T , even when the guard waits on a cut.
On any subinterval of an original crossing,
( 1 − q 2 ) Δ t ≥ ln ( Γ ) − q Δ z . (18) (1-q^2)\,\Delta t\ge\ln(\Gamma)-q\,\Delta z. \tag*{(18)} ( 1 − q 2 ) Δ t ≥ ln ( Γ ) − q Δ z . (18)
Indeed ∣ Γ ′ ∣ ≤ 1 − q Y ′ = 1 − q 2 + q z ′ |\Gamma'|\le1-qY'=1-q^2+qz' ∣ Γ ′ ∣ ≤ 1 − q Y ′ = 1 − q 2 + q z ′ by the physical speed bound. Equality holds along the unit-speed ideal legs, so the ideal arc length is L ∗ = m T ∗ L_*=mT_* L ∗ = m T ∗ and its forward arclength s s s corresponds to physical time t ∗ ( s ) = [ s − q ( γ z ( s ) − γ z ( 0 ) ) ] / ( 1 − q 2 ) t_*(s)=[s-q(\gamma_z(s)-\gamma_z(0))]/(1-q^2) t ∗ ( s ) = [ s − q ( γ z ( s ) − γ z ( 0 ))] / ( 1 − q 2 ) , where γ \gamma γ denotes the ideal phase polygon parametrized by arclength. This time is strictly increasing since ∣ d γ z / d s ∣ ≤ 1 |d\gamma_z/ds|\le1 ∣ d γ z / d s ∣ ≤ 1 .
The inverse arclength coordinate of the ideal polygon is Lipschitz on its image with a constant K 0 K_0 K 0 uniform over J J J . On adjacent edges this follows from the positive angle between the two outgoing rays at their common vertex. Nonadjacent edges remain a positive distance apart. There are finitely many edges of positive lengths, with fixed incidences and continuous vertices on J J J , and the simplicity margins above are uniform. Extend the coordinate to all phase points by truncating inf 0 ≤ v ≤ L ∗ ( v + K 0 ∣ x − γ ( v ) ∣ ) \inf_{0\le v\le L_*}(v+K_0|x-\gamma(v)|) inf 0 ≤ v ≤ L ∗ ( v + K 0 ∣ x − γ ( v ) ∣ ) to [ 0 , L ∗ ] [0,L_*] [ 0 , L ∗ ] , and denote the result by s ( x ) s(x) s ( x ) . It is continuous, agrees with arclength on the arc, and ∣ x − γ ( s ( x ) ) ∣ ≤ ( 1 + K 0 ) δ |x-\gamma(s(x))|\le(1+K_0)\delta ∣ x − γ ( s ( x )) ∣ ≤ ( 1 + K 0 ) δ whenever x x x lies in its δ \delta δ neighborhood.
Take δ = C E j \delta=CE_j δ = C E j on the principal crossing. Its initial and final coordinates are within C δ C\delta C δ of 0 , L ∗ 0,L_* 0 , L ∗ . For any crossing point x x x , continuity of s s s supplies visits in order to all intervening ideal vertex coordinates between the start and s ( x ) s(x) s ( x ) , choosing successive first hits. Each visit is within C δ C\delta C δ of the specified vertex. Chord lengths between consecutive hits, including the two endpoints, sum to at least s ( x ) − C δ s(x)-C\delta s ( x ) − C δ , because the ideal pieces are straight and their number is bounded. Omitting up to C δ C\delta C δ of initial arclength costs at most this length. The same argument from x x x to the end gives a suffix length at least L ∗ − s ( x ) − C δ L_* - s(x)-C\delta L ∗ − s ( x ) − C δ . Both estimates allow backtracking. Equation (18) on these prefix and suffix paths now gives
D I ≥ m T ∗ − C E j , ∣ t − t 0 − t ∗ ( s ( x ) ) ∣ ≤ ( D I − m T ∗ ) + + C E j . D_I \ge mT_* - CE_j,\qquad|t-t_0-t_*(s(x))| \le(D_I-mT_*)_+ + CE_j. D I ≥ m T ∗ − C E j , ∣ t − t 0 − t ∗ ( s ( x )) ∣ ≤ ( D I − m T ∗ ) + + C E j .
The ideal patrol is uniformly Lipschitz in physical time, also past the ends of its ideal block by periodic extension. The formulas X = Γ x / α X=\Gamma_x/\alpha X = Γ x / α and Y = q t − z Y=qt-z Y = q t − z convert these estimates into a physical position error at most C ( E j + ( D I − m T ∗ ) + ) C(E_j+(D_I-mT_*)_+) C ( E j + ( D I − m T ∗ ) + ) after the stated translations.
If n n n is the retained count, then ∣ T − n m T ∗ ∣ ≤ C N Δ |T-nmT_*| \le CN\sqrt{\Delta} ∣ T − nm T ∗ ∣ ≤ C N Δ by the duration estimate and the discarded count. Sum the preceding crossing-time bounds and use ∑ M I ≤ T \sum M_I \le T ∑ M I ≤ T , M I ≥ D I M_I \ge D_I M I ≥ D I , and ∑ r e t a i n e d E j ≤ C N Δ \sum_{\mathrm{retained}} E_j \le CN\sqrt{\Delta} ∑ retained E j ≤ C N Δ . The result is
∑ r e t a i n e d ( E j + ( D I − m T ∗ ) + + M I − D I ) ≤ C N Δ . \sum_{\mathrm{retained}}\left(E_j+(D_I-mT_*)_+ + M_I-D_I\right) \le CN\sqrt{\Delta}. retained ∑ ( E j + ( D I − m T ∗ ) + + M I − D I ) ≤ C N Δ .
As n n n is at least a positive multiple of N N N , one slab has all three errors at most C Δ C\sqrt{\Delta} C Δ . This gives every conclusion of the lemma.
Escaping the original patrol
Proof of Theorem 5.1 . Fix q ∈ ( 1 / 3 , q a ) q\in(1/3,q_a) q ∈ ( 1/3 , q a ) , and take an escaping transit of the ordinary bowtie from Theorem 4.2 . Cut it at two times at which its height is below − h ∗ − 2 -h_* - 2 − h ∗ − 2 and above h ∗ + 2 h_*+2 h ∗ + 2 , respectively. Extend the endpoints vertically at upward speed q q q on the two time tails. The tails have clearance greater than two against the ordinary bowtie. Continuity and closed-disk capture give clearance at least 1 + β 1+\beta 1 + β on the remaining compact segment for some β > 0 \beta>0 β > 0 , which we may take at most one. The evader phase ζ ( t ) = q t − y ( t ) \zeta(t)=qt-y(t) ζ ( t ) = q t − y ( t ) is nondecreasing and constant on each tail.
Choose m m m large enough that a time shift of the escape by whole bowtie periods places its central segment inside an ideal m m m -cycle block and its entire phase range more than two units from the phase boundaries. This can be done simultaneously. Each period shift adds T ∗ T_* T ∗ to the time locations and q T ∗ qT_* q T ∗ to the phase locations, so first move both lower ends beyond the initial cuts, then choose m m m with room at both upper ends.
Suppose a periodic patrol fully guards half-width b = b ∗ − Δ b=b_*-\Delta b = b ∗ − Δ with Δ \Delta Δ sufficiently small. Use the selected crossing from Lemma 5.5 , translating the evader by the same time and height origins. Clamp its horizontal coordinate to [ − b , b ] [-b,b] [ − b , b ] . This changes its position by at most Δ \Delta Δ , preserves its speed bound, and preserves its phase. On the crossing the original guard and the ordinary bowtie are within C Δ C\sqrt{\Delta} C Δ in physical position. Thus the evader escapes there when C Δ + Δ < β C\sqrt{\Delta}+\Delta<\beta C Δ + Δ < β .
To exclude capture at all other times, a visit to [ L + ρ , U − ρ ] [L+\rho,U-\rho] [ L + ρ , U − ρ ] outside [ t 0 , t 1 ] [t_0,t_1] [ t 0 , t 1 ] costs at least ρ / ( 1 + q ) \rho/(1+q) ρ / ( 1 + q ) additional seconds in the open slab. Indeed a visit before t 0 t_0 t 0 entails a passage from the lower boundary to the core by z ( − ∞ ) = − ∞ z(-\infty)=-\infty z ( − ∞ ) = − ∞ , and a visit after t 1 t_1 t 1 entails a passage from the core to the upper boundary by z ( + ∞ ) = + ∞ z(+\infty)=+\infty z ( + ∞ ) = + ∞ . Take a fixed ρ > 0 \rho>0 ρ > 0 small enough, for example 1 / 2 1/2 1/2 , that the one-unit enlargement of the evader’s entire phase range lies in this core for small Δ \Delta Δ . Lemma 5.5 makes the extra occupation smaller than ρ / ( 1 + q ) \rho/(1+q) ρ / ( 1 + q ) . At any other time the phase difference then exceeds one, hence so does the physical vertical separation. The resulting full transit escapes, a contradiction. This excludes a fixed interval of widths below the ceiling, uniformly over all periods. Theorem 4.1 transfers the bound to all bounded-excursion patrols. Restoring units multiplies the width loss by r r r .
Proof of Theorem 5.2. Use J = [ 0.4578 , q a ] J=[0.4578,q_a] J = [ 0.4578 , q a ] at r = 1 r=1 r = 1 and write R ( q ) = R r o u n d ( a ∗ , h ∗ ; q ) \mathcal{R}(q)=\mathcal{R}_{\mathrm{round}}(a_*,h_*;q) R ( q ) = R round ( a ∗ , h ∗ ; q ) as in (9) . The rounded transit in the proof of Theorem 4.2 has distance at least R ( q ) \mathcal{R}(q) R ( q ) from the ordinary bowtie. It reaches absolute height 21 / 10 > h ∗ + R ( q ) 21/10>h_*+\mathcal{R}(q) 21/10 > h ∗ + R ( q ) at its two outer joins, so vertical upward speed-q q q extensions preserve this clearance. Its join times and phase range are uniformly bounded on J J J , and q T ∗ qT_* q T ∗ has a uniform positive lower bound. One choice of m m m therefore accommodates the central segments and phase buffers throughout J J J . Lemma 5.5 gives uniform constants with this m m m . For any successful periodic half-width with small Δ \Delta Δ , the preceding escape argument requires
R ( q ) − 1 ≤ C Δ + Δ . \mathcal{R}(q)-1\le C\sqrt{\Delta}+\Delta. R ( q ) − 1 ≤ C Δ + Δ.
The rational enclosure in anc/gap/check_wall_gap_upper.py gives − 7.514 < R ′ ( q ) < − 7.435 -7.514<\mathcal{R}'(q)<-7.435 − 7.514 < R ′ ( q ) < − 7.435 on [ 0.4578 , 0.4579 ] [0.4578,0.4579] [ 0.4578 , 0.4579 ] . For the derivative enclosure, a binomial tail of degree n = 9 n=9 n = 9 and parameter 0 ≤ x ≤ x 0 < 1 / 10 0\le x\le x_0<1/10 0 ≤ x ≤ x 0 < 1/10 has derivative bounded by
∣ x ′ ∣ ∑ k > n k x 0 k − 1 = ∣ x ′ ∣ x 0 n ( n + 1 ) − n x 0 ( 1 − x 0 ) 2 . |x'|\sum_{k>n}kx_0^{k-1}=|x'|x_0^n\frac{(n+1)-nx_0}{(1-x_0)^2}. ∣ x ′ ∣ k > n ∑ k x 0 k − 1 = ∣ x ′ ∣ x 0 n ( 1 − x 0 ) 2 ( n + 1 ) − n x 0 .
This follows by differentiating the uniformly convergent series on the compact parameter interval. Combining it with the value remainders for the two factors and the product rule encloses the derivative of the integral in the escape proof. Since R ( q a ) = 1 \mathcal{R}(q_a)=1 R ( q a ) = 1 by Lemma 3.5, it follows that R ( q ) − 1 ≥ 7.435 ( q a − q ) \mathcal{R}(q)-1\ge7.435(q_a-q) R ( q ) − 1 ≥ 7.435 ( q a − q ) on J J J . This proves the quadratic lower bound for small Δ \Delta Δ . Widths with larger Δ \Delta Δ satisfy the same bound upon decreasing k 1 k_1 k 1 . Taking suprema and applying Theorem 4.1 proves the global lower bound.
For the upper bound set λ = 1 / R ( q ) \lambda=1/\mathcal{R}(q) λ = 1/ R ( q ) and scale a ∗ , h ∗ a_*,h_* a ∗ , h ∗ by λ \lambda λ while holding the speed and leg direction fixed. The fan angle, integral p p p , and angular coefficients in (9) depend only on the leg direction and q q q . Therefore the handoff ratio of the scaled bowtie equals one exactly. Its wall displacement G G G is λ R ∗ \lambda R_* λ R ∗ . Choose the inset 1 − λ 2 R ∗ 2 \sqrt{1-\lambda^2R_*^2} 1 − λ 2 R ∗ 2 to satisfy the wall condition at equality. The resulting half-width is
B q ( λ ) = λ a ∗ + 1 − λ 2 R ∗ 2 . B_q(\lambda)=\lambda a_*+\sqrt{1-\lambda^2R_*^2}. B q ( λ ) = λ a ∗ + 1 − λ 2 R ∗ 2 .
The accompanying checker recomputes the remaining 19 finite-leg and fan inequalities of Theorem 3.4 on the full scaled family over [ 0.4578 , 0.4579 ] [0.4578,0.4579] [ 0.4578 , 0.4579 ] . It encloses every margin above 0.0332 0.0332 0.0332 using rational intervals and the binomial integral with its remainder. This certifies an all-transit guard at half-width B q ( λ ) B_q(\lambda) B q ( λ ) .
Since B q ( 1 ) = b ∗ B_q(1)=b_* B q ( 1 ) = b ∗ and B q ′ ( 1 ) = a ∗ − R ∗ 2 / 1 − R ∗ 2 = 0 B_q'(1)=a_*-R_*^2/\sqrt{1-R_*^2}=0 B q ′ ( 1 ) = a ∗ − R ∗ 2 / 1 − R ∗ 2 = 0 , Taylor’s theorem and B q ′ ′ ( λ ) = − R ∗ 2 ( 1 − λ 2 R ∗ 2 ) − 3 / 2 B_q''(\lambda)=-R_*^2(1-\lambda^2R_*^2)^{-3/2} B q ′′ ( λ ) = − R ∗ 2 ( 1 − λ 2 R ∗ 2 ) − 3/2 give, for 0.4578 ≤ q ≤ q a 0.4578\le q\le q_a 0.4578 ≤ q ≤ q a where λ ≤ 1 \lambda\le1 λ ≤ 1 ,
2 b ∗ − 2 B q ( λ ) ≤ 1 + D D 3 / 2 ( 1 − λ ) 2 ≤ 1 + D D 3 / 2 ( 7.514 ) 2 ( q a − q ) 2 . 2b_*-2B_q(\lambda)\le\frac{\sqrt{1+D}}{D^{3/2}}(1-\lambda)^2\le\frac{\sqrt{1+D}}{D^{3/2}}(7.514)^2(q_a-q)^2. 2 b ∗ − 2 B q ( λ ) ≤ D 3/2 1 + D ( 1 − λ ) 2 ≤ D 3/2 1 + D ( 7.514 ) 2 ( q a − q ) 2 .
Here 1 − λ = ( R − 1 ) / R ≤ R − 1 1-\lambda=(\mathcal{R}-1)/\mathcal{R}\le\mathcal{R}-1 1 − λ = ( R − 1 ) / R ≤ R − 1 . The same checker bounds the displayed coefficient by 175. Rescaling gives the announced upper estimate. At q = 0.45785 q=0.45785 q = 0.45785 the checker gives the two outer inequalities in the following bracket. Theorem 5.1 gives the middle inequality.
3.3682502470 r < W o p t ( 0.45785 ) < F ( 0.45785 ) < 3.3682505519 r . 3.3682502470r<W_{\mathrm{opt}}(0.45785)<F(0.45785)<3.3682505519r. 3.3682502470 r < W opt ( 0.45785 ) < F ( 0.45785 ) < 3.3682505519 r .
Fast intruders and bounded longitudinal excursion
When q → 1 − q \to1^{-} q → 1 − , F ( q ) → 2 r 1 + 1 / 4 = 5 r = r 4 + 1 − 2 F(q) \to2r\sqrt{1+1/4}=\sqrt{5}r=r\sqrt{4+1^{-2}} F ( q ) → 2 r 1 + 1/4 = 5 r = r 4 + 1 − 2 and h ( q ) → 0 h(q)\to0 h ( q ) → 0 , continuously merging the bowtie into Washburn’s horizontal back-and-forth patrol [1 ] . We now prove a global midline-excursion theorem valid for all q > 0 q>0 q > 0 .
Theorem 6.1 (Global Midline-Excursion Bound and Fast-Intruder Optimum). Let q > 0 q>0 q > 0 , and let g = ( X , Y ) : R → [ − b , b ] × R g=(X,Y):\mathbb{R}\to[-b,b]\times\mathbb{R} g = ( X , Y ) : R → [ − b , b ] × R be any prescribed unit-speed patrol (periodic or nonperiodic) satisfying sup t ∈ R ∣ Y ( t ) ∣ ≤ H < ∞ \sup_{t\in\mathbb{R}}|Y(t)|\le H<\infty sup t ∈ R ∣ Y ( t ) ∣ ≤ H < ∞ . If g g g successfully guards [ − b , b ] × R [-b,b]\times\mathbb{R} [ − b , b ] × R , then
W = 2 b ≤ r 4 + q − 2 + ( 1 − q ) + q H , ( 1 − q ) + : = max ( 0 , 1 − q ) . (19) W=2b\le r\sqrt{4+q^{-2}}+\frac{(1-q)_{+}}{q}H,\qquad(1-q)_{+}:=\max(0,1-q). \tag*{(19)} W = 2 b ≤ r 4 + q − 2 + q ( 1 − q ) + H , ( 1 − q ) + := max ( 0 , 1 − q ) . (19) Consequently:
For all q ≥ 1 q\ge1 q ≥ 1 , the exact minimax corridor width among all bounded-excursion patrols in both one-way and two-way transit models is
W o p t ( q ) = W 2 w a y ( q ) = r 4 + q − 2 , ∀ q ≥ 1. (20) W_{\mathrm{opt}}(q)=W_{\mathrm{2way}}(q)=r\sqrt{4+q^{-2}},\qquad\forall q\ge1. \tag*{(20)} W opt ( q ) = W 2way ( q ) = r 4 + q − 2 , ∀ q ≥ 1. (20) For every 0 < q < 1 0<q<1 0 < q < 1 , W o p t ( q , H ) = r 4 + q − 2 + 1 − q q H + O ( H 2 / r ) W_{\mathrm{opt}}(q,H)=r\sqrt{4+q^{-2}}+\frac{1-q}{q}H+O(H^{2}/r) W opt ( q , H ) = r 4 + q − 2 + q 1 − q H + O ( H 2 / r ) as H ↓ 0 H\downarrow0 H ↓ 0 .
Proof. Normalize r = 1 r=1 r = 1 , V = 1 V=1 V = 1 , U = q U=q U = q , and define the transverse reference parameters
a 0 = 1 2 q 1 + 4 q 2 , b 0 = 1 + 4 q 2 2 q = 1 2 4 + q − 2 , δ 0 = b 0 − a 0 = 2 q 1 + 4 q 2 , R 0 = 1 1 + 4 q 2 . a_{0}=\frac{1}{2q\sqrt{1+4q^{2}}},\qquad b_{0}=\frac{\sqrt{1+4q^{2}}}{2q}=\frac{1}{2}\sqrt{4+q^{-2}},\qquad\delta_{0}=b_{0}-a_{0}=\frac{2q}{\sqrt{1+4q^{2}}},\qquad R_{0}=\frac{1}{\sqrt{1+4q^{2}}}. a 0 = 2 q 1 + 4 q 2 1 , b 0 = 2 q 1 + 4 q 2 = 2 1 4 + q − 2 , δ 0 = b 0 − a 0 = 1 + 4 q 2 2 q , R 0 = 1 + 4 q 2 1 .
Note R 0 = 2 q a 0 R_{0}=2qa_{0} R 0 = 2 q a 0 and δ 0 / R 0 = 2 q \delta_{0}/R_{0}=2q δ 0 / R 0 = 2 q . Assume b > b 0 > 1 b>b_{0}>1 b > b 0 > 1 and set Δ : = b − b 0 > 0 \Delta:=b-b_{0}>0 Δ := b − b 0 > 0 .
At the right wall x = + b x=+b x = + b , detection at time t t t requires b − X ( t ) ≤ 1 b-X(t)\le1 b − X ( t ) ≤ 1 , hence X ( t ) ≥ b − 1 > b 0 − 1 > 0 X(t)\ge b-1>b_{0}-1>0 X ( t ) ≥ b − 1 > b 0 − 1 > 0 . The detected straight-run wall phases at time t t t form [ q t − Y ( t ) − R ( b − X ( t ) ) , q t − Y ( t ) + R ( b − X ( t ) ) ] [qt-Y(t)-R(b-X(t)),\,qt-Y(t)+R(b-X(t))] [ q t − Y ( t ) − R ( b − X ( t )) , q t − Y ( t ) + R ( b − X ( t ))] where R ( d ) = 1 − d 2 R(d)=\sqrt{1-d^{2}} R ( d ) = 1 − d 2 . By concavity of R R R on [ 0 , 1 ] [0,1] [ 0 , 1 ] and its supporting tangent at δ 0 \delta_{0} δ 0 (R ′ ( δ 0 ) = − δ 0 / R 0 = − 2 q R'(\delta_{0})=-\delta_{0}/R_{0}=-2q R ′ ( δ 0 ) = − δ 0 / R 0 = − 2 q ), every active right-wall contact satisfies
R ( b − X ) ≤ R 0 − 2 q [ ( b − X ) − δ 0 ] = 2 q b 0 − 2 q ( b − X ) = 2 q ( X − Δ ) . R(b-X)\le R_{0}-2q[(b-X)-\delta_{0}]=2qb_{0}-2q(b-X)=2q(X-\Delta). R ( b − X ) ≤ R 0 − 2 q [( b − X ) − δ 0 ] = 2 q b 0 − 2 q ( b − X ) = 2 q ( X − Δ ) .
Consider any complete positive midline excursion of X X X on [ 0 , D ] [0,D] [ 0 , D ] (X ( 0 ) = X ( D ) = 0 X(0)=X(D)=0 X ( 0 ) = X ( D ) = 0 and X ( t ) > 0 X(t)>0 X ( t ) > 0 on ( 0 , D ) (0,D) ( 0 , D ) ). Put A ( t ) : = q t − Y ( t ) + 2 q X ( t ) A(t):=qt-Y(t)+2qX(t) A ( t ) := q t − Y ( t ) + 2 q X ( t ) and B ( t ) : = q t − Y ( t ) − 2 q X ( t ) B(t):=qt-Y(t)-2qX(t) B ( t ) := q t − Y ( t ) − 2 q X ( t ) . The total span length I I I of right-wall phases detected during [ 0 , D ] [0,D] [ 0 , D ] satisfies I ≤ max [ 0 , D ] A − min [ 0 , D ] B − 4 q Δ I\le\max_{[0,D]}A-\min_{[0,D]}B-4q\Delta I ≤ max [ 0 , D ] A − min [ 0 , D ] B − 4 q Δ . Choose u , v ∈ [ 0 , D ] u,v\in[0,D] u , v ∈ [ 0 , D ] attaining max A \max A max A and min B \min B min B , and put d : = ∣ u − v ∣ d:=|u-v| d := ∣ u − v ∣ . Since ∣ X ′ ∣ ≤ 1 |X'|\le1 ∣ X ′ ∣ ≤ 1 and X ( 0 ) = X ( D ) = 0 X(0)=X(D)=0 X ( 0 ) = X ( D ) = 0 , X ( u ) ≤ min ( u , D − u ) X(u)\le\min(u,D-u) X ( u ) ≤ min ( u , D − u ) and X ( v ) ≤ min ( v , D − v ) X(v)\le\min(v,D-v) X ( v ) ≤ min ( v , D − v ) imply X ( u ) + X ( v ) ≤ D − d X(u)+X(v)\le D-d X ( u ) + X ( v ) ≤ D − d , while ∣ Y ∣ ≤ H |Y|\le H ∣ Y ∣ ≤ H and ∣ Y ′ ∣ ≤ 1 |Y'|\le1 ∣ Y ′ ∣ ≤ 1 give Y ( v ) − Y ( u ) ≤ min ( 2 H , d ) Y(v)-Y(u)\le\min(2H,d) Y ( v ) − Y ( u ) ≤ min ( 2 H , d ) . Therefore
A ( u ) − B ( v ) − 2 q D = q ( u − v ) + Y ( v ) − Y ( u ) + 2 q ( X ( u ) + X ( v ) − D ) ≤ { min ( 2 H , d ) − q d , u ≥ v , min ( 2 H , d ) − 3 q d , u < v . \begin{aligned}
A(u)-B(v)-2qD
&=q(u-v)+Y(v)-Y(u)+2q(X(u)+X(v)-D)\\
&\le
\begin{cases}
\min(2H,d)-qd, & u\ge v,\\
\min(2H,d)-3qd, & u<v.
\end{cases}
\end{aligned} A ( u ) − B ( v ) − 2 q D = q ( u − v ) + Y ( v ) − Y ( u ) + 2 q ( X ( u ) + X ( v ) − D ) ≤ { min ( 2 H , d ) − q d , min ( 2 H , d ) − 3 q d , u ≥ v , u < v .
Maximizing over d ≥ 0 d\ge0 d ≥ 0 yields A ( u ) − B ( v ) − 2 q D ≤ 2 H ( 1 − q ) + = : C A(u)-B(v)-2qD\le2H(1-q)_{+}=:C A ( u ) − B ( v ) − 2 q D ≤ 2 H ( 1 − q ) + =: C . By symmetry (X ↦ − X X\mapsto-X X ↦ − X ), every complete active excursion on either side of the midline covers wall-phase span at most
I ≤ 2 q D + C − 4 q Δ . (21) I\le2qD+C-4q\Delta. \tag*{(21)} I ≤ 2 q D + C − 4 q Δ. (21)
On any window [ 0 , L ] [0,L] [ 0 , L ] , at most two active excursions are truncated by 0 0 0 and L L L , each contributing span at most 2 q D t r + C + 4 q b 2qD_{\mathrm{tr}}+C+4qb 2 q D tr + C + 4 q b . Every complete active excursion reaches ∣ X ∣ ≥ b − 1 > 0 |X|\ge b-1>0 ∣ X ∣ ≥ b − 1 > 0 from 0 0 0 and returns in time ≥ 2 ( b − 1 ) > 0 \ge2(b-1)>0 ≥ 2 ( b − 1 ) > 0 , so their count N ( L ) N(L) N ( L ) is finite with ∑ D k ≤ L \sum D_k \le L ∑ D k ≤ L . Covering [ H + 2 , q L − H − 2 ] [H+2,qL-H-2] [ H + 2 , q L − H − 2 ] on both walls requires 2 q L − 4 ( H + 2 ) ≤ 2 q L + N ( L ) ( C − 4 q Δ ) + 2 ( C + 4 q b ) 2qL-4(H+2)\le2qL+N(L)(C-4q\Delta)+2(C+4qb) 2 q L − 4 ( H + 2 ) ≤ 2 q L + N ( L ) ( C − 4 q Δ ) + 2 ( C + 4 q b ) . Because late wall phases z → + ∞ z\to+\infty z → + ∞ on both walls require infinitely many sign switches, N ( L ) → ∞ N(L)\to\infty N ( L ) → ∞ as L → ∞ L\to\infty L → ∞ , forcing C − 4 q Δ ≥ 0 C-4q\Delta\ge0 C − 4 q Δ ≥ 0 , i.e., 2 Δ ≤ ( 1 − q ) + q H 2\Delta\le\frac{(1-q)_{+}}{q}H 2Δ ≤ q ( 1 − q ) + H , proving (19) . Part (1) follows from ( 1 − q ) + = 0 (1-q)_{+}=0 ( 1 − q ) + = 0 for q ≥ 1 q\ge1 q ≥ 1 combined with Corollary 3.2(1), and part (2) follows by combining (19) with Corollary 3.2(2) for each fixed 0 < q < 1 0<q<1 0 < q < 1 as H ↓ 0 H\downarrow0 H ↓ 0 .
Two-way transit
A patrol guards in both directions if it captures every upward transit in Definition 1.1 and every downward transit obtained by reflecting such a path in y = 0 y=0 y = 0 . Let W 2 w a y ( q , H ) W_{\mathrm{2way}}(q,H) W 2way ( q , H ) denote the corresponding supremum of widths under ∣ Y ∣ ≤ H |Y|\le H ∣ Y ∣ ≤ H , and set W 2 w a y ( q ) = sup H < ∞ W 2 w a y ( q , H ) W_{\mathrm{2way}}(q)=\sup_{H<\infty}W_{\mathrm{2way}}(q,H) W 2way ( q ) = sup H < ∞ W 2way ( q , H ) .
An excursion bound for both directions
Theorem 7.1 (Two-way width bound). If a patrol of height at most H H H captures every straight wall runner of speed U U U in both directions, its width satisfies
2 b ≤ W b f ( q ) + ( 1 − 2 q ) + q H , W b f ( q ) = r 4 + q − 2 . (22) 2b\le W_{\mathrm{bf}}(q)+\frac{(1-2q)_{+}}{q}H,\qquad W_{\mathrm{bf}}(q)=r\sqrt{4+q^{-2}}. \tag*{(22)} 2 b ≤ W bf ( q ) + q ( 1 − 2 q ) + H , W bf ( q ) = r 4 + q − 2 . (22) Here s + = max ( s , 0 ) s_{+}=\max(s,0) s + = max ( s , 0 ) . In particular, for every H ≥ 0 H\ge0 H ≥ 0 and q ≥ 1 / 2 q\ge1/2 q ≥ 1/2 ,
W 2 w a y ( q , H ) = W 2 w a y ( q ) = W b f ( q ) . (23) W_{\mathrm{2way}}(q,H)=W_{\mathrm{2way}}(q)=W_{\mathrm{bf}}(q). \tag*{(23)} W 2way ( q , H ) = W 2way ( q ) = W bf ( q ) . (23)
Proof. Use r = V = 1 r=V=1 r = V = 1 and put
b 0 = 1 2 4 + q − 2 , a 0 = 1 2 q 1 + 4 q 2 , R 0 = 2 q a 0 , δ 0 = b 0 − a 0 = 2 q R 0 . b_0=\frac{1}{2}\sqrt{4+q^{-2}},\qquad a_0=\frac{1}{2q\sqrt{1+4q^2}},\qquad R_0=2qa_0,\qquad\delta_0=b_0-a_0=2qR_0. b 0 = 2 1 4 + q − 2 , a 0 = 2 q 1 + 4 q 2 1 , R 0 = 2 q a 0 , δ 0 = b 0 − a 0 = 2 q R 0 .
It suffices to treat Δ = b − b 0 > 0 \Delta=b-b_0>0 Δ = b − b 0 > 0 . At a time when the right wall is in range, the slice of the disk along that wall has vertical half-length 1 − ( b − X ) 2 \sqrt{1-(b-X)^2} 1 − ( b − X ) 2 . Its supporting tangent at b − X = δ 0 b-X=\delta_0 b − X = δ 0 bounds this half-length by R 0 − 2 q ( b − X − δ 0 ) = 2 q ( X − Δ ) R_0-2q(b-X-\delta_0)=2q(X-\Delta) R 0 − 2 q ( b − X − δ 0 ) = 2 q ( X − Δ ) .
Consider a complete positive excursion with X ( 0 ) = X ( D ) = 0 X(0)=X(D)=0 X ( 0 ) = X ( D ) = 0 and X > 0 X>0 X > 0 on ( 0 , D ) (0,D) ( 0 , D ) , translating the time origin for this calculation. Call it active if the disk reaches the right wall at some time. On an active excursion the upward and downward runners y = ± ( q t − ζ ) y=\pm(qt-\zeta) y = ± ( q t − ζ ) detected on that wall give two sets of phases ζ \zeta ζ . Write I ± I_{\pm} I ± for their respective diameters. For
A ± ( t ) = q t ∓ Y ( t ) + 2 q X ( t ) , B ± ( t ) = q t ∓ Y ( t ) − 2 q X ( t ) , A_{\pm}(t)=qt\mp Y(t)+2qX(t),\qquad B_{\pm}(t)=qt\mp Y(t)-2qX(t), A ± ( t ) = q t ∓ Y ( t ) + 2 q X ( t ) , B ± ( t ) = q t ∓ Y ( t ) − 2 q X ( t ) ,
the tangent bound gives I ± ≤ A ± ( u ± ) − B ± ( v ± ) − 4 q Δ I_{\pm}\le A_{\pm}(u_{\pm})-B_{\pm}(v_{\pm})-4q\Delta I ± ≤ A ± ( u ± ) − B ± ( v ± ) − 4 q Δ , where u ± u_{\pm} u ± and v ± v_{\pm} v ± maximize A ± A_{\pm} A ± and minimize B ± B_{\pm} B ± on [ 0 , D ] [0,D] [ 0 , D ] .
For arbitrary fixed choices of these four times, write J ( q ) = ∑ ± ( A ± ( u ± ) − B ± ( v ± ) ) − 4 q D J(q)=\sum_{\pm}(A_{\pm}(u_{\pm})-B_{\pm}(v_{\pm}))-4qD J ( q ) = ∑ ± ( A ± ( u ± ) − B ± ( v ± )) − 4 q D . The four marked terms at u + , v + , u − , v − u_{+},v_{+},u_{-},v_{-} u + , v + , u − , v − have respective pairs of signs ( + , − ) , ( − , + ) , ( + , + ) , ( − , − ) (+,-),(-,+),(+,+),(-,-) ( + , − ) , ( − , + ) , ( + , + ) , ( − , − ) on the time and height coordinates. At time t t t , let k k k be the number of markers later than t t t , let m m m sum their time signs, and let j j j sum their height signs. All four horizontal coefficients are 2 q 2q 2 q . Integrating up to the markers, using the zero sum of their height signs and ∫ 0 D X ′ = 0 \int_{0}^{D}X'=0 ∫ 0 D X ′ = 0 , gives
J ( q ) = ∫ 0 D ( 2 q ( k − 2 ) X ′ ( t ) + j Y ′ ( t ) − q ( 4 − m ) ) d t . J(q)=\int_{0}^{D}\left(2q(k-2)X'(t)+jY'(t)-q(4-m)\right)\,dt. J ( q ) = ∫ 0 D ( 2 q ( k − 2 ) X ′ ( t ) + j Y ′ ( t ) − q ( 4 − m ) ) d t .
For q ≥ 1 / 2 q \ge1/2 q ≥ 1/2 , the integrand is nonpositive. For k = 0 k = 0 k = 0 or 4 4 4 , the norm of the velocity coefficient equals 4 q = q ( 4 − m ) 4q = q(4-m) 4 q = q ( 4 − m ) . For k = 1 k = 1 k = 1 or 3 3 3 , ∣ m ∣ = ∣ j ∣ = 1 |m| = |j| = 1 ∣ m ∣ = ∣ j ∣ = 1 and the norm is 4 q 2 + 1 ≤ 3 q ≤ q ( 4 − m ) \sqrt{4q^2+1} \le3q \le q(4-m) 4 q 2 + 1 ≤ 3 q ≤ q ( 4 − m ) . For k = 2 k = 2 k = 2 , the horizontal coefficient vanishes. If ∣ j ∣ = 2 |j| = 2 ∣ j ∣ = 2 , those two markers have opposite time signs, so m = 0 m = 0 m = 0 and ∣ j ∣ ≤ 4 q |j| \le4q ∣ j ∣ ≤ 4 q . The other possibility is j = 0 j = 0 j = 0 . In each case use ∥ ( X ′ , Y ′ ) ∥ ≤ 1 \lVert(X',Y') \rVert\le1 ∥( X ′ , Y ′ )∥ ≤ 1 . For 0 < q < 1 / 2 0 < q < 1/2 0 < q < 1/2 , J ( q ) J(q) J ( q ) is affine for the fixed marks and path. The bounds J ( 0 ) ≤ 4 H J(0) \le4H J ( 0 ) ≤ 4 H and J ( 1 / 2 ) ≤ 0 J(1/2) \le0 J ( 1/2 ) ≤ 0 give J ( q ) ≤ 4 ( 1 − 2 q ) H J(q) \le4(1-2q)H J ( q ) ≤ 4 ( 1 − 2 q ) H . Hence on either side of the corridor, by reflection in x = 0 x = 0 x = 0 ,
I + + I − ≤ 4 q D + 4 ( 1 − 2 q ) H − 8 q Δ . I_+ + I_- \le4qD + 4(1-2q)H - 8q\Delta. I + + I − ≤ 4 q D + 4 ( 1 − 2 q ) H − 8 q Δ.
On [ 0 , L ] [0,L] [ 0 , L ] each wall and direction must cover all phases [ H + 2 , q L − H − 2 ] [H+2,qL-H-2] [ H + 2 , q L − H − 2 ] for large L L L , since those runners can be detected only inside this window. As b > 1 b > 1 b > 1 , detection requires an excursion of the corresponding sign. Each complete active excursion lasts at least 2 ( b − 1 ) 2(b-1) 2 ( b − 1 ) , so their count N ( L ) N(L) N ( L ) is finite. There are at most two truncated excursions. For each direction a truncated piece of length d d d covers a phase span at most q d + 2 H + 2 qd + 2H + 2 q d + 2 H + 2 . Summing the four phase requirements and using that the excursion lengths sum to at most L L L gives
4 q L − O ( H + 1 ) ≤ 4 q L + ( 4 ( 1 − 2 q ) H − 8 q Δ ) N ( L ) + O ( H + 1 ) . 4qL - O(H+1) \le4qL + \bigl(4(1-2q)H - 8q\Delta\bigr)N(L) + O(H+1). 4 q L − O ( H + 1 ) ≤ 4 q L + ( 4 ( 1 − 2 q ) H − 8 q Δ ) N ( L ) + O ( H + 1 ) .
Arbitrarily late phases require detections on both walls at arbitrarily late times. The patrol therefore makes infinitely many complete active excursions, so N ( L ) → ∞ N(L) \to\infty N ( L ) → ∞ . This proves (22) upon rescaling. The horizontal patrol of Corollary 3.2 (1) guards every upward transit. Its reflection in y = 0 y = 0 y = 0 is the same patrol, so it guards in both directions and proves (23) .
Sharp gain at small height
Theorem 7.2 (Two-way first variation). For each fixed 0 < q < 1 / 2 0 < q < 1/2 0 < q < 1/2 , as H / r ↓ 0 H/r \downarrow0 H / r ↓ 0 ,
W 2 w a y ( q , H ) = W b f ( q ) + 1 − 2 q q H + O q ( H 2 / r ) . W_{\mathrm{2way}}(q,H) = W_{\mathrm{bf}}(q) + \frac{1-2q}{q}H + O_q(H^2/r). W 2way ( q , H ) = W bf ( q ) + q 1 − 2 q H + O q ( H 2 / r ) . In particular W 2 w a y ( q , H ) > W b f ( q ) W_{\mathrm{2way}}(q,H) > W_{\mathrm{bf}}(q) W 2way ( q , H ) > W bf ( q ) for every H > 0 H > 0 H > 0 .
Proof. Normalize r = V = 1 r = V = 1 r = V = 1 and take a = a 0 a = a_0 a = a 0 from the preceding proof. At unit speed follow the first four legs
( − a , − H ) ⟶ ( a , H ) ⟶ ( a , − H ) ⟶ ( − a , − H ) ⟶ ( − a , H ) . (-a,-H) \longrightarrow(a,H) \longrightarrow(a,-H) \longrightarrow(-a,-H) \longrightarrow(-a,H). ( − a , − H ) ⟶ ( a , H ) ⟶ ( a , − H ) ⟶ ( − a , − H ) ⟶ ( − a , H ) .
Concatenate their reflection in y = 0 y = 0 y = 0 and repeat. With ℓ = a 2 + H 2 \ell= \sqrt{a^2+H^2} ℓ = a 2 + H 2 , the half-period is D = 2 ℓ + 2 a + 4 H D = 2\ell+ 2a + 4H D = 2 ℓ + 2 a + 4 H and the full period is T = 2 D T = 2D T = 2 D . Thus
( X ( t + D ) , Y ( t + D ) ) = ( X ( t ) , − Y ( t ) ) . (24) (X(t+D),Y(t+D)) = (X(t),-Y(t)). \tag*{(24)} ( X ( t + D ) , Y ( t + D )) = ( X ( t ) , − Y ( t )) . (24)
At either pole x 0 = ± a x_0 = \pm a x 0 = ± a , successive vertical visits last 2 H 2H 2 H and their centers are D D D apart. A downstroke alternates with an upstroke. Put
R = q D / 2 − H , δ = 1 − R 2 , b = a + δ , B d = R + ( 1 + q ) H , B u = R + ( 1 − q ) H . R = qD/2 - H,\qquad\delta= \sqrt{1-R^2},\qquad b = a+\delta,\qquad B_d = R+(1+q)H,\qquad B_u = R+(1-q)H. R = q D /2 − H , δ = 1 − R 2 , b = a + δ , B d = R + ( 1 + q ) H , B u = R + ( 1 − q ) H .
As H ↓ 0 H \downarrow0 H ↓ 0 , R → R 0 ∈ ( 0 , 1 ) R \to R_0 \in(0,1) R → R 0 ∈ ( 0 , 1 ) and δ → δ 0 > 0 \delta\to\delta_0 > 0 δ → δ 0 > 0 . Take H > 0 H > 0 H > 0 small enough that 0 < R < 1 0 < R < 1 0 < R < 1 , q H < min ( R , δ ) qH < \min(R,\delta) q H < min ( R , δ ) and B d + H < 1 B_d + H < 1 B d + H < 1 . We construct a bounded separator for every upward transit in [ − b , b ] × R [-b,b] \times\mathbb{R} [ − b , b ] × R .
Assign a level to each wall. For successive centers c j , c j + 1 c_j,c_{j+1} c j , c j + 1 at its pole, let B j B_j B j be B d B_d B d or B u B_u B u according to the direction of visit j j j . Between centers its level is
f ( t ) = − B j + q ( t − c j ) , c j < t < c j + 1 . f(t) = -B_j + q(t-c_j),\qquad c_j < t < c_{j+1}. f ( t ) = − B j + q ( t − c j ) , c j < t < c j + 1 .
Since q D = B d + B u qD = B_d + B_u q D = B d + B u , this reaches + B j + 1 +B_{j+1} + B j + 1 just before the next center. Both wall levels differ from the guard height by at most B d + H < 1 B_d + H < 1 B d + H < 1 . During transfers and upstrokes use the left wall level for x < X ( t ) x < X(t) x < X ( t ) and the right wall level for x > X ( t ) x > X(t) x > X ( t ) . Join them by a vertical segment at x = X ( t ) x = X(t) x = X ( t ) . That segment lies in the current disk.
During a downstroke centered at c c c , use the other wall’s level throughout the corridor on the inward side of the visited pole. On the outer strip, between that pole and its wall, join the point at the pole on this level to the wall point at height Y ( t ) − R ( t − c ) / H Y(t) - R(t-c)/H Y ( t ) − R ( t − c ) / H . The chord is inside the disk. Its wall endpoint has horizontal offset δ \delta δ and vertical offset of magnitude at most R R R . At entry the old strip level is B d − q H = H + R B_d - qH = H + R B d − q H = H + R , the height of the chord’s wall endpoint. At exit the new strip level is − B d + q H = − H − R -B_d + qH = -H - R − B d + q H = − H − R , again the height of that endpoint. At either join, the region between the two profiles is the triangle formed by the common wall endpoint and the two heights at the pole. The disk contains all three vertices and hence this entire triangle.
At the center c c c of an upstroke, the level on its outer strip drops from + B u +B_u + B u to − B u -B_u − B u . A speed-q q q path in this strip at c c c with ∣ y ( c ) ∣ ≤ B u |y(c)| \le B_u ∣ y ( c ) ∣ ≤ B u is captured during the visit. Indeed, project y ( c ) y(c) y ( c ) onto [ − H , H ] [-H,H] [ − H , H ] , call the result s s s , and take w = c + s w = c + s w = c + s . The guard is then at ( x 0 , s ) (x_0,s) ( x 0 , s ) . Corridor confinement bounds the path’s distance from x 0 x_0 x 0 towards the wall by δ \delta δ . Its distance on the inward side of x 0 x_0 x 0 is at most q ∣ w − c ∣ ≤ q H q|w-c| \le qH q ∣ w − c ∣ ≤ q H , since it started on the outer strip. Also
∣ y ( w ) − s ∣ ≤ q H + ( ∣ y ( c ) ∣ − H ) + ≤ q H + ( B u − H ) = R , |y(w)-s| \le qH + (|y(c)|-H)_+ \le qH + (B_u-H) = R, ∣ y ( w ) − s ∣ ≤ q H + ( ∣ y ( c ) ∣ − H ) + ≤ q H + ( B u − H ) = R ,
where B u − H = R − q H > 0 B_u-H = R-qH > 0 B u − H = R − q H > 0 . Therefore its squared distance to the guard at w w w is at most δ 2 + R 2 = 1 \delta^2+R^2 = 1 δ 2 + R 2 = 1 .
To finish, consider a path avoiding capture at all times. On each open flat, its height relative to the flat is nonincreasing, since y ′ ( t ) ≤ q y'(t) \le q y ′ ( t ) ≤ q . At a moving vertical join the path must lie strictly above both flat endpoints or strictly below both, as the entire join is captured. The same one-sided classification holds across the chords and their entry and exit triangles. At an upstroke reset a change from below to above would place the path in the rectangle just proved captured during that visit. These local statements preclude a crossing from below to above on every finite interval. More explicitly, between resets a point of an uncaptured path on the separator can lie only in the interior of a flat relative to the corridor, where the same rising level applies in a time neighborhood. At points strictly off the separator its side is locally constant. The finitely many joins and resets on a finite interval preserve this implication. The separator is bounded, so an upward transit would have to cross it. By (24) , a downward transit is captured as well.
Finally, R = q ( ℓ + a ) + ( 2 q − 1 ) H = R 0 + ( 2 q − 1 ) H + O q ( H 2 ) R = q(\ell+a) + (2q-1)H = R_0 + (2q-1)H + O_q(H^2) R = q ( ℓ + a ) + ( 2 q − 1 ) H = R 0 + ( 2 q − 1 ) H + O q ( H 2 ) . Since δ ′ ( R 0 ) = − R 0 / δ 0 = − 1 / ( 2 q ) \delta'(R_0) = -R_0/\delta_0 = -1/(2q) δ ′ ( R 0 ) = − R 0 / δ 0 = − 1/ ( 2 q ) , the guarded width 2 ( a + δ ) 2(a+\delta) 2 ( a + δ ) equals W b f ( q ) + ( 1 − 2 q ) H / q + O q ( H 2 ) W_{\rm bf}(q) + (1-2q)H/q + O_q(H^2) W bf ( q ) + ( 1 − 2 q ) H / q + O q ( H 2 ) in normalized units. Combining this construction with (22) and restoring r r r gives the expansion. Its positive linear coefficient proves strict gain for small H H H , and monotonicity in H H H proves the assertion for all H > 0 H>0 H > 0 .
Certified widths below the transition
Lower bounds from oriented spacetime surfaces
We certify rigorous lower bounds W o p t ( q ) ≥ W c e r t W_{\rm opt}(q) \ge W_{\rm cert} W opt ( q ) ≥ W cert for multi-leg periodic patrols in exact rational arithmetic via triangulated spacetime surface cuts (translating x ∈ [ − b , b ] x \in[-b,b] x ∈ [ − b , b ] by + b +b + b to [ 0 , W ] [0,W] [ 0 , W ] to match certificate coordinates):
Proposition 8.1 (Triangulated Spacetime Surface-Cut Certificate). Let g : R → [ 0 , W ] × R g:\mathbb{R} \to[0,W] \times\mathbb{R} g : R → [ 0 , W ] × R be a T T T -periodic piecewise-affine patrol (T > 0 T>0 T > 0 , W > 0 W>0 W > 0 ) with speed at most 1 1 1 on every segment. Let Σ \Sigma Σ be a finite oriented triangulated surface in R ( t , x , y ) 3 \mathbb{R}^3_{(t,x,y)} R ( t , x , y ) 3 whose vertices satisfy 0 ≤ t ≤ T 0 \le t \le T 0 ≤ t ≤ T and 0 ≤ x ≤ W 0 \le x \le W 0 ≤ x ≤ W such that, when edge incidences are read modulo T T T in the time coordinate:
Closedness off the corridor walls: Every directed edge of Σ \Sigma Σ is paired with an oppositely oriented partner modulo T T T , except edges lying entirely in the left wall { x = 0 } \{x = 0\} { x = 0 } or right wall { x = W } \{x = W\} { x = W } .
Unit negative projection degree: Summing the signed areas of all triangles projected onto ( t , x ) (t,x) ( t , x ) equals − T W -TW − T W (equivalently, ∑ △ ⊂ Σ n y ( △ ) = − 2 T W \sum_{\triangle\subset\Sigma} n_y(\triangle) = -2TW ∑ △ ⊂ Σ n y ( △ ) = − 2 T W where n ( △ ) = ( p 1 − p 0 ) × ( p 2 − p 0 ) \mathbf{n}(\triangle) = (p_1-p_0) \times(p_2-p_0) n ( △ ) = ( p 1 − p 0 ) × ( p 2 − p 0 ) ).
Facewise causal or single-segment capture certificate: Every triangle △ = conv { v 0 , v 1 , v 2 } ⊂ Σ \triangle= \operatorname{conv}\{v_0,v_1,v_2\} \subset\Sigma △ = conv { v 0 , v 1 , v 2 } ⊂ Σ (v i = ( t i , P i ) v_i = (t_i,P_i) v i = ( t i , P i ) with P i ∈ [ 0 , W ] × R P_i \in[0,W] \times\mathbb{R} P i ∈ [ 0 , W ] × R ) satisfies at least one of:
(Type C — Causal): Its normal n = ( n t , n x , n y ) \mathbf{n} = (n_t,n_x,n_y) n = ( n t , n x , n y ) satisfies n t ≥ 0 n_t \ge0 n t ≥ 0 and n t 2 ≥ q 2 ( n x 2 + n y 2 ) n_t^2 \ge q^2(n_x^2+n_y^2) n t 2 ≥ q 2 ( n x 2 + n y 2 ) ; or
(Type S — Capture): There exist witness times w 0 , w 1 , w 2 w_0,w_1,w_2 w 0 , w 1 , w 2 lying on one common affine segment of g g g such that ∥ P i − g ( w i ) ∥ 2 + q ∣ w i − t i ∣ ≤ r \lVert P_i-g(w_i)\rVert_2+q|w_i-t_i| \le r ∥ P i − g ( w i ) ∥ 2 + q ∣ w i − t i ∣ ≤ r for each i ∈ { 0 , 1 , 2 } i \in\{0,1,2\} i ∈ { 0 , 1 , 2 } .
Then g g g captures every upward transit trajectory of speed at most q q q in [ 0 , W ] × R [0,W]\times\mathbb{R} [ 0 , W ] × R , proving W o p t ( q ) ≥ W W_{\mathrm{opt}}(q) \ge W W opt ( q ) ≥ W .
Proof. Extend Σ \Sigma Σ T T T -periodically along t ∈ R t \in\mathbb{R} t ∈ R to an infinite polyhedral 2 2 2 -chain Σ ~ ⊂ R × [ 0 , W ] × R \widetilde{\Sigma} \subset\mathbb{R}\times[0,W]\times\mathbb{R} Σ ⊂ R × [ 0 , W ] × R . By Condition 1, ∂ Σ ~ \partial\widetilde{\Sigma} ∂ Σ is supported entirely in R × { 0 , W } × R \mathbb{R}\times\{0,W\}\times\mathbb{R} R × { 0 , W } × R , and by compactness of Σ \Sigma Σ , ∣ y ∣ |y| ∣ y ∣ is bounded by some Y max Y_{\max} Y m a x on Σ ~ \widetilde{\Sigma} Σ . In the cylinder of ( t m o d T , x ) (t\bmod T,x) ( t mod T , x ) the projected chain has boundary supported on its two walls. Its multiplicity is therefore constant at generic interior points. Integrating that multiplicity and using Condition 2 gives − 1 -1 − 1 . Thus in the lift the oriented intersection number of a vertical fiber from below to above the chain is − 1 -1 − 1 . Every interior curve with endpoints on opposite sides of the height range has the same number: it is homotopic to such a fiber with its endpoints moving only in the two disjoint outer half-spaces.
To justify transverse intersections even for a wall-following or grazing intruder, suppose an undetected path exists and choose a finite crossing from below − Y max − 2 -Y_{\max}-2 − Y m a x − 2 to above Y max + 2 Y_{\max}+2 Y m a x + 2 . Include a time buffer r / q + 1 r/q+1 r / q + 1 at both ends. On this compact enlarged interval the path has strictly positive excess distance over r r r . Contract its spatial coordinates slightly toward ( W / 2 , 0 ) (W/2,0) ( W /2 , 0 ) to gain strict speed and wall margins, then approximate by a piecewise linear time graph. Sufficiently small generic perturbations of the vertices preserve both margins, the excess distance and the endpoint height inequalities. Only finitely many faces meet the time window. The perturbed crossing avoids their edges and meets the faces transversely. A Type S witness for any such intersection lies within the enlarged interval since q ∣ w − t ∣ ≤ r q|w-t| \le r q ∣ w − t ∣ ≤ r .
On any Type C face, ∥ p ′ ( τ ) ∥ 2 ≤ q \lVert p'(\tau)\rVert_2 \le q ∥ p ′ ( τ ) ∥ 2 ≤ q and Cauchy–Schwarz give n ⋅ c ′ ( τ ) = n t + n x x ′ ( τ ) + n y y ′ ( τ ) ≥ n t − q n x 2 + n y 2 ≥ 0 \mathbf{n}\cdot c'(\tau)=n_t+n_xx'(\tau)+n_yy'(\tau)\ge n_t-q\sqrt{n_x^2+n_y^2}\ge0 n ⋅ c ′ ( τ ) = n t + n x x ′ ( τ ) + n y y ′ ( τ ) ≥ n t − q n x 2 + n y 2 ≥ 0 a.e., so c c c can never cross a Type C face with negative orientation. Hence c c c must meet some Type S face △ \triangle △ at some time τ ∗ \tau_* τ ∗ , say ( τ ∗ , p ( τ ∗ ) ) = ∑ i = 0 2 λ i ( t i , P i ) (\tau_*,p(\tau_*))=\sum_{i=0}^2\lambda_i(t_i,P_i) ( τ ∗ , p ( τ ∗ )) = ∑ i = 0 2 λ i ( t i , P i ) (λ i ≥ 0 \lambda_i\ge0 λ i ≥ 0 , ∑ λ i = 1 \sum\lambda_i=1 ∑ λ i = 1 ). Because g g g is affine on the common segment containing w 0 , w 1 , w 2 w_0,w_1,w_2 w 0 , w 1 , w 2 , the map ( t , P , w ) ↦ ∥ P − g ( w ) ∥ 2 + q ∣ w − t ∣ (t,P,w)\mapsto\lVert P-g(w)\rVert_2+q|w-t| ( t , P , w ) ↦ ∥ P − g ( w ) ∥ 2 + q ∣ w − t ∣ is jointly convex, so w ∗ : = ∑ i = 0 2 λ i w i w_*:=\sum_{i=0}^2\lambda_iw_i w ∗ := ∑ i = 0 2 λ i w i satisfies ∥ p ( τ ∗ ) − g ( w ∗ ) ∥ 2 + q ∣ w ∗ − τ ∗ ∣ ≤ r \lVert p(\tau_*)-g(w_*)\rVert_2+q|w_*-\tau_*|\le r ∥ p ( τ ∗ ) − g ( w ∗ ) ∥ 2 + q ∣ w ∗ − τ ∗ ∣ ≤ r . By ∥ p ( w ∗ ) − p ( τ ∗ ) ∥ 2 ≤ q ∣ w ∗ − τ ∗ ∣ \lVert p(w_*)-p(\tau_*)\rVert_2\le q|w_*-\tau_*| ∥ p ( w ∗ ) − p ( τ ∗ ) ∥ 2 ≤ q ∣ w ∗ − τ ∗ ∣ and the triangle inequality, ∥ p ( w ∗ ) − g ( w ∗ ) ∥ 2 ≤ r \lVert p(w_*)-g(w_*)\rVert_2\le r ∥ p ( w ∗ ) − g ( w ∗ ) ∥ 2 ≤ r . □ \square □
A 64-leg periodic patrol at q = 1 / 4 q=1/4 q = 1/4 has an oriented certificate with 837 vertices and 1,592 faces. Its exact width exceeds the following decimal endpoint:
W o p t ( 1 / 4 ) > 6.944691333740 r . (25) W_{\mathrm{opt}}(1/4)>6.944691333740\,r. \tag*{(25)} W opt ( 1/4 ) > 6.944691333740 r . (25)
Six 18-leg patrol certificates give the following strict lower endpoints for W o p t ( q ) / r W_{\mathrm{opt}}(q)/r W opt ( q ) / r :
Table 2.
The rational patrol coordinates, triangle incidences and face witnesses accompany the checker in anc/.
Flux across straight lanes and parity at the walls
Normalize r = 1 r = 1 r = 1 , V = 1 V = 1 V = 1 , b > 1 b > 1 b > 1 , and let J : = [ − b + 1 , b − 1 ] J := [-b + 1, b - 1] J := [ − b + 1 , b − 1 ] be the interior gap separating the left wall footprint [ − b , − b + 1 ] [-b, -b + 1] [ − b , − b + 1 ] from the right wall footprint [ b − 1 , b ] [b - 1, b] [ b − 1 , b ] .
Weighted Galilean lane flux: For a , c ≥ 0 a,c \ge0 a , c ≥ 0 , define Φ a , c ( z ) : = max ( c , a ∣ z ∣ 1 − z 2 ) \Phi_{a,c}(z) := \max\left(c,\frac{a|z|}{\sqrt{1-z^2}}\right) Φ a , c ( z ) := max ( c , 1 − z 2 a ∣ z ∣ ) for z ∈ ( − 1 , 1 ) z \in(-1,1) z ∈ ( − 1 , 1 ) and its continuous odd primitive P a , c : [ − 1 , 1 ] → R P_{a,c} : [-1,1] \to\mathbb{R} P a , c : [ − 1 , 1 ] → R (P a , c ( − z ) = − P a , c ( z ) P_{a,c}(-z) = -P_{a,c}(z) P a , c ( − z ) = − P a , c ( z ) ) by setting, for z ∈ [ 0 , 1 ] z \in[0,1] z ∈ [ 0 , 1 ] and D : = a 2 + c 2 D := \sqrt{a^2+c^2} D := a 2 + c 2 ,
P a , c ( z ) : = { c z , z D ≤ c , D − a 1 − z 2 , z D > c . P_{a,c}(z) :=
\begin{cases}
cz, & zD \le c,\\
D-a\sqrt{1-z^2}, & zD > c.
\end{cases} P a , c ( z ) := { cz , D − a 1 − z 2 , z D ≤ c , z D > c .
For any bounded nonnegative lane density λ : [ − b , b ] → [ 0 , ∞ ) \lambda: [-b,b] \to[0,\infty) λ : [ − b , b ] → [ 0 , ∞ ) (extended by 0 0 0 outside [ − b , b ] [-b,b] [ − b , b ] ), when the guard is at X ∈ [ − b , b ] X \in[-b,b] X ∈ [ − b , b ] with velocity v = ( v x , v y ) v = (v_x,v_y) v = ( v x , v y ) (v x 2 + v y 2 ≤ 1 v_x^2+v_y^2 \le1 v x 2 + v y 2 ≤ 1 ), pairing the two boundary arcs of ∂ B 1 ( X , Y ) \partial B_1(X,Y) ∂ B 1 ( X , Y ) at horizontal offset z = x − X ∈ ( − 1 , 1 ) z=x-X \in(-1,1) z = x − X ∈ ( − 1 , 1 ) gives the instantaneous λ \lambda λ -weighted Galilean sweep rate
K λ ( X , v ) : = ∫ − 1 1 λ ( X + z ) Φ ∣ v x ∣ , ∣ q − v y ∣ ( z ) d z = ∫ [ − b , b ] ∩ [ X − 1 , X + 1 ] λ ( x ) d x P ∣ v x ∣ , ∣ q − v y ∣ ( x − X ) . (26) K_\lambda(X,v) := \int_{-1}^{1} \lambda(X+z)\Phi_{|v_x|,|q-v_y|}(z)\,dz
= \int_{[-b,b]\cap[X-1,X+1]} \lambda(x)\,dx\,P_{|v_x|,|q-v_y|}(x-X). \tag*{(26)} K λ ( X , v ) := ∫ − 1 1 λ ( X + z ) Φ ∣ v x ∣ , ∣ q − v y ∣ ( z ) d z = ∫ [ − b , b ] ∩ [ X − 1 , X + 1 ] λ ( x ) d x P ∣ v x ∣ , ∣ q − v y ∣ ( x − X ) . (26)
Two-wall parity credits and transverse weight β ( X ) \beta(X) β ( X ) : For finitely many cutoffs c k ∈ ( 0 , 1 ] c_k \in(0,1] c k ∈ ( 0 , 1 ] (s k : = 1 − c k 2 s_k := \sqrt{1-c_k^2} s k := 1 − c k 2 ) with weights η k ≥ 0 \eta_k \ge0 η k ≥ 0 (E : = ∑ k η k E := \sum_k \eta_k E := ∑ k η k ), define the wall-footprint functions
f c k ( X ) : = { u 2 1 − u 2 , b − 1 < ∣ X ∣ ≤ b , u : = b − ∣ X ∣ ∈ [ 0 , s k ) , 0 , otherwise . f_{c_k}(X) :=
\begin{cases}
\frac{u}{2\sqrt{1-u^2}}, & b-1<|X|\le b,\quad u:=b-|X|\in[0,s_k),\\
0, & \text{otherwise}.
\end{cases} f c k ( X ) := { 2 1 − u 2 u , 0 , b − 1 < ∣ X ∣ ≤ b , u := b − ∣ X ∣ ∈ [ 0 , s k ) , otherwise .
choose any bounded measurable gap density ρ : J → [ 0 , ∞ ) \rho: J \to[0,\infty) ρ : J → [ 0 , ∞ ) (extended by 0 0 0 outside J J J ) satisfying ∫ J ρ ( x ) d x ≥ 4 ∑ k c k η k \int_J \rho(x)\,dx \ge4\sum_k c_k\eta_k ∫ J ρ ( x ) d x ≥ 4 ∑ k c k η k , and set β ( X ) : = 4 ∑ k η k f c k ( X ) + ρ ( X ) \beta(X) := 4\sum_k \eta_k f_{c_k}(X) + \rho(X) β ( X ) := 4 ∑ k η k f c k ( X ) + ρ ( X ) .
Theorem 8.2 (Galilean Flux and Two-Wall Free-Gap Parity Inequality). Let b > 1 b > 1 b > 1 , q ∈ ( 0 , 1 ) q \in(0,1) q ∈ ( 0 , 1 ) , λ ≥ 0 \lambda\ge0 λ ≥ 0 , η k ≥ 0 \eta_k \ge0 η k ≥ 0 , ρ ≥ 0 \rho\ge0 ρ ≥ 0 with ∫ J ρ ( x ) d x ≥ 4 ∑ k c k η k \int_J \rho(x)\,dx \ge4\sum_k c_k\eta_k ∫ J ρ ( x ) d x ≥ 4 ∑ k c k η k , and μ ∈ R \mu\in\mathbb{R} μ ∈ R . For every unit-speed patrol g = ( X , Y ) g = (X,Y) g = ( X , Y ) on [ − b , b ] × R [-b,b] \times\mathbb{R} [ − b , b ] × R with bounded longitudinal excursion sup t ∈ R ∣ Y ( t ) ∣ ≤ H < ∞ \sup_{t\in\mathbb{R}}|Y(t)| \le H < \infty sup t ∈ R ∣ Y ( t ) ∣ ≤ H < ∞ that blocks all upward straight vertical runners p x , ζ ( t ) = ( x , q t − ζ ) p_{x,\zeta}(t) = (x,qt-\zeta) p x , ζ ( t ) = ( x , q t − ζ ) ,
sup X ∈ [ − b , b ] , ∥ v ∥ 2 ≤ 1 { K λ ( X , v ) + ( ∑ k η k ) ∣ q − v y ∣ + β ( X ) ∣ v x ∣ − μ v y } ≥ q ∫ − b b λ ( x ) d x + 3 q ∑ k η k . (27) \sup_{\substack{X\in[-b,b],\,\lVert v\rVert_2\le1}}
\left\{K_\lambda(X,v)+\left(\sum_k\eta_k\right)|q-v_y|+\beta(X)|v_x|-\mu v_y\right\}
\ge q\int_{-b}^{b}\lambda(x)\,dx+3q\sum_k\eta_k. \tag*{(27)} X ∈ [ − b , b ] , ∥ v ∥ 2 ≤ 1 sup { K λ ( X , v ) + ( k ∑ η k ) ∣ q − v y ∣ + β ( X ) ∣ v x ∣ − μ v y } ≥ q ∫ − b b λ ( x ) d x + 3 q k ∑ η k . (27)
Proof. Fix a long window [ 0 , L ] [0,L] [ 0 , L ] (q L > 2 H + 4 qL > 2H+4 q L > 2 H + 4 ). Every phase ζ ∈ [ H + 1 , q L − H − 1 ] \zeta\in[H+1,qL-H-1] ζ ∈ [ H + 1 , q L − H − 1 ] on every lane x ∈ [ − b , b ] x \in[-b,b] x ∈ [ − b , b ] satisfies ∣ q t − ζ ∣ > H + 1 ≥ ∣ Y ( t ) ∣ + 1 |qt-\zeta| > H+1 \ge|Y(t)|+1 ∣ q t − ζ ∣ > H + 1 ≥ ∣ Y ( t ) ∣ + 1 outside [ 0 , L ] [0,L] [ 0 , L ] , so g ∣ [ 0 , L ] g|_{[0,L]} g ∣ [ 0 , L ] must sweep λ \lambda λ -weighted ( x , ζ ) (x,\zeta) ( x , ζ ) -measure at least ( q L − 2 H − 2 ) ∫ − b b λ ( x ) d x (qL-2H-2)\int_{-b}^{b}\lambda(x)\,dx ( q L − 2 H − 2 ) ∫ − b b λ ( x ) d x , giving
∫ 0 L K λ ( X ( t ) , g ′ ( t ) ) d t ≥ ( q L − 2 H − 4 ) ∫ − b b λ ( x ) d x . \int_0^L K_\lambda(X(t),g'(t))\,dt \ge(qL-2H-4)\int_{-b}^{b}\lambda(x)\,dx. ∫ 0 L K λ ( X ( t ) , g ′ ( t )) d t ≥ ( q L − 2 H − 4 ) ∫ − b b λ ( x ) d x .
Indeed, a point in the swept disk union that lies outside both endpoint disks encounters the moving boundary at least twice, apart from a set of area zero. The area formula applied to ( t , θ ) ↦ ( X ( t ) , q t − Y ( t ) ) + ( cos θ , sin θ ) (t,\theta)\mapsto(X(t),qt-Y(t))+(\cos\theta,\sin\theta) ( t , θ ) ↦ ( X ( t ) , q t − Y ( t )) + ( cos θ , sin θ ) integrates the absolute normal velocity with multiplicity. Half of that integral is ∫ 0 L K λ d t \int_0^L K_\lambda\,dt ∫ 0 L K λ d t by pairing the two arcs at each lane. The two endpoint disks counted with weight one half have weighted area at most 2 ∫ λ 2\int\lambda 2 ∫ λ in total (points in exactly one endpoint disk require one crossing). Also ∣ ∫ 0 L μ Y ′ ( t ) d t ∣ = ∣ μ ( Y ( L ) − Y ( 0 ) ) ∣ ≤ 2 ∣ μ ∣ H \left|\int_0^L \mu Y'(t)\,dt\right|=|\mu(Y(L)-Y(0))|\le2|\mu|H ∫ 0 L μ Y ′ ( t ) d t = ∣ μ ( Y ( L ) − Y ( 0 )) ∣ ≤ 2∣ μ ∣ H .
It remains to prove the temporal credit inequality E ∫ 0 L ∣ q − Y ′ ∣ d t + ∫ 0 L β ( X ) ∣ X ′ ∣ d t ≥ 3 q E L − E ( 6 H + 14 ) E\int_0^L |q-Y'|\,dt+\int_0^L \beta(X)|X'|\,dt\ge3qEL-E(6H+14) E ∫ 0 L ∣ q − Y ′ ∣ d t + ∫ 0 L β ( X ) ∣ X ′ ∣ d t ≥ 3 q E L − E ( 6 H + 14 ) where E = ∑ k η k E=\sum_k\eta_k E = ∑ k η k . Partition [ 0 , L ] [0,L] [ 0 , L ] into alternating wall-labelled blocks B B B , where the active label (L or R) switches only when X ( t ) X(t) X ( t ) enters the opposite wall footprint { X < − b + 1 } \{X<-b+1\} { X < − b + 1 } or { X > b − 1 } \{X>b-1\} { X > b − 1 } (assigning any initial unlabelled prefix to the first visited wall). Discard the first and last blocks from the travel estimate; every remaining interior block B B B enters its own footprint across ∣ X ∣ = b − 1 |X|=b-1 ∣ X ∣ = b − 1 , reaches some maximum vertical chord radius r B : = max t ∈ B ∩ { ∣ X ∣ ≥ b − 1 } 1 − ( b − ∣ X ( t ) ∣ ) 2 ∈ [ 0 , 1 ] r_B:=\max_{t\in B\cap\{|X|\ge b-1\}}\sqrt{1-(b-|X(t)|)^2}\in[0,1] r B := max t ∈ B ∩ { ∣ X ∣ ≥ b − 1 } 1 − ( b − ∣ X ( t ) ∣ ) 2 ∈ [ 0 , 1 ] , exits ∣ X ∣ = b − 1 |X|=b-1 ∣ X ∣ = b − 1 , and subsequently makes a complete crossing of J = [ − b + 1 , b − 1 ] J=[-b+1,b-1] J = [ − b + 1 , b − 1 ] to the opposite footprint. On B B B ’s own footprint, b − ∣ X ∣ b-|X| b − ∣ X ∣ descends from 1 1 1 to 1 − r B 2 \sqrt{1-r_B^2} 1 − r B 2 and returns to 1 1 1 , so total variation gives
∫ B ∩ { ∣ X ∣ > b − 1 } f c k ( X ) ∣ X ′ ∣ d t ≥ 2 ∫ 1 − r B 2 1 u 1 { 1 − u 2 > c k } 2 1 − u 2 d u = ( r B − c k ) + , \int_{B\cap\{|X|>b-1\}} f_{c_k}(X)|X'|\,dt\ge2\int_{\sqrt{1-r_B^2}}^1\frac{u\mathbf{1}_{\{\sqrt{1-u^2}>c_k\}}}{2\sqrt{1-u^2}}\,du=(r_B-c_k)_+, ∫ B ∩ { ∣ X ∣ > b − 1 } f c k ( X ) ∣ X ′ ∣ d t ≥ 2 ∫ 1 − r B 2 1 2 1 − u 2 u 1 { 1 − u 2 > c k } d u = ( r B − c k ) + ,
while the disjoint subsequent crossing of J J J contributes ∫ g a p B ρ ( X ) ∣ X ′ ∣ d t ≥ ∫ J ρ ( x ) d x ≥ 4 ∑ k c k η k \int_{\mathrm{gap}_B}\rho(X)|X'|\,dt\ge\int_J\rho(x)\,dx\ge4\sum_k c_k\eta_k ∫ gap B ρ ( X ) ∣ X ′ ∣ d t ≥ ∫ J ρ ( x ) d x ≥ 4 ∑ k c k η k . Summing over all interior blocks (and paying at most 8 E 8E 8 E for the two discarded endpoint radii r B f i r s t , r B l a s t ≤ 1 r_{B_{\mathrm{first}}},r_{B_{\mathrm{last}}}\le1 r B first , r B last ≤ 1 ) gives
∫ 0 L β ( X ) ∣ X ′ ∣ d t ≥ 4 ∑ B i n t ∑ k η k ( ( r B − c k ) + + c k ) ≥ 4 E ∑ B a l l r B − 8 E . \int_0^L\beta(X)|X'|\,dt\ge4\sum_{B\ \mathrm{int}}\sum_k\eta_k\bigl((r_B-c_k)_++c_k\bigr)\ge4E\sum_{B\ \mathrm{all}}r_B-8E. ∫ 0 L β ( X ) ∣ X ′ ∣ d t ≥ 4 B int ∑ k ∑ η k ( ( r B − c k ) + + c k ) ≥ 4 E B all ∑ r B − 8 E .
Independently, consider the phase map ϕ ( t ) : = q t − Y ( t ) \phi(t):=qt-Y(t) ϕ ( t ) := q t − Y ( t ) on [ 0 , L ] [0,L] [ 0 , L ] . Because ϕ ( 0 ) ≤ H \phi(0)\le H ϕ ( 0 ) ≤ H and ϕ ( L ) ≥ q L − H \phi(L)\ge qL-H ϕ ( L ) ≥ q L − H , every regular value ζ ∈ [ H + 1 , q L − H − 1 ] \zeta\in[H+1,qL-H-1] ζ ∈ [ H + 1 , q L − H − 1 ] has odd preimage count n ( ζ ) = # ϕ − 1 ( ζ ) ≥ 1 n(\zeta)=\#\phi^{-1}(\zeta)\ge1 n ( ζ ) = # ϕ − 1 ( ζ ) ≥ 1 . Let κ ( ζ ) ∈ { 0 , 1 , 2 } \kappa(\zeta)\in\{0,1,2\} κ ( ζ ) ∈ { 0 , 1 , 2 } count how many of the two wall labels { L , R } \{\mathrm{L},\mathrm{R}\} { L , R } have some block B B B whose ϕ \phi ϕ -image ϕ ( B ) \phi(B) ϕ ( B ) contains ζ \zeta ζ . Because L-blocks and R-blocks are disjoint in [ 0 , L ] [0,L] [ 0 , L ] , κ ( ζ ) = 2 \kappa(\zeta)=2 κ ( ζ ) = 2 forces at least two preimages of ζ \zeta ζ , hence n ( ζ ) ≥ 3 n(\zeta)\ge3 n ( ζ ) ≥ 3 by odd parity; thus n ( ζ ) ≥ 2 κ ( ζ ) − 1 n(\zeta)\ge2\kappa(\zeta)-1 n ( ζ ) ≥ 2 κ ( ζ ) − 1 for a.e. ζ ∈ [ H + 1 , q L − H − 1 ] \zeta\in[H+1,qL-H-1] ζ ∈ [ H + 1 , q L − H − 1 ] . Meanwhile, each block B B B can only detect wall phases within distance r B r_B r B of its interval ϕ ( B ) \phi(B) ϕ ( B ) , and full coverage of [ H + 1 , q L − H − 1 ] [H+1,qL-H-1] [ H + 1 , q L − H − 1 ] on both walls requires the two-wall uncovered measure ∫ H + 1 q L − H − 1 ( 2 − κ ( ζ ) ) d ζ \int_{H+1}^{qL-H-1}(2-\kappa(\zeta))\,d\zeta ∫ H + 1 q L − H − 1 ( 2 − κ ( ζ )) d ζ to be covered by the 2 r B 2r_B 2 r B endpoint extensions across all blocks: ∫ H + 1 q L − H − 1 ( 2 − κ ( ζ ) ) d ζ ≤ 2 ∑ B r B \int_{H+1}^{qL-H-1}(2-\kappa(\zeta))\,d\zeta\le2\sum_Br_B ∫ H + 1 q L − H − 1 ( 2 − κ ( ζ )) d ζ ≤ 2 ∑ B r B . By the Banach indicatrix theorem,
V z : = ∫ 0 L ∣ q − Y ′ ( t ) ∣ d t ≥ ∫ H + 1 q L − H − 1 n ( ζ ) d ζ ≥ ∫ H + 1 q L − H − 1 ( 3 − 2 ( 2 − κ ( ζ ) ) ) d ζ ≥ 3 ( q L − 2 H − 2 ) − 4 ∑ B r B . V_z:=\int_0^L|q-Y'(t)|\,dt\ge\int_{H+1}^{qL-H-1}n(\zeta)\,d\zeta\ge\int_{H+1}^{qL-H-1}(3-2(2-\kappa(\zeta)))\,d\zeta\ge3(qL-2H-2)-4\sum_Br_B. V z := ∫ 0 L ∣ q − Y ′ ( t ) ∣ d t ≥ ∫ H + 1 q L − H − 1 n ( ζ ) d ζ ≥ ∫ H + 1 q L − H − 1 ( 3 − 2 ( 2 − κ ( ζ ))) d ζ ≥ 3 ( q L − 2 H − 2 ) − 4 B ∑ r B .
Multiplying by E E E , adding the β ( X ) \beta(X) β ( X ) bound ( − 4 E ∑ B r B + 4 E ∑ B r B = 0 ) (-4E\sum_Br_B+4E\sum_Br_B=0) ( − 4 E ∑ B r B + 4 E ∑ B r B = 0 ) , combining with K λ K_\lambda K λ and − μ Y ′ -\mu Y' − μ Y ′ , dividing by L L L , and letting L → ∞ L\to\infty L → ∞ proves (27) .
Proposition 8.3 (Vertical-Drift Floor of Stateless Flux + Two-Wall Credits at q = 1 / 3 q=1/3 q = 1/3 ). At q = 1 / 3 q=1/3 q = 1/3 , for every symmetric λ ≥ 0 \lambda\ge0 λ ≥ 0 , η k ≥ 0 \eta_k\ge0 η k ≥ 0 , ρ ≥ 0 \rho\ge0 ρ ≥ 0 , and μ ∈ R \mu\in\mathbb{R} μ ∈ R , if W = 2 b ≤ 6 r W=2b\le6r W = 2 b ≤ 6 r , then the left-hand supremum in (27) is at least the right-hand demand. Conversely, at W = 6.12 r > 6 r W=6.12r>6r W = 6.12 r > 6 r , there exist rational multipliers for which the supremum drops strictly below demand, certifying W o p t ( 1 / 3 ) ≤ 6.12 r < W a r e a ( 1 / 3 ) = 2 10 r ≈ 6.324 r W_{\mathrm{opt}}(1/3)\le6.12r<W_{\mathrm{area}}(1/3)=2\sqrt{10}\,r\approx6.324r W opt ( 1/3 ) ≤ 6.12 r < W area ( 1/3 ) = 2 10 r ≈ 6.324 r .
Proof. Evaluate the bracket Ψ ( X , v ) \Psi(X,v) Ψ ( X , v ) in (27) at v ± = ( 0 , ± 1 ) v_\pm=(0,\pm1) v ± = ( 0 , ± 1 ) . Since v x = 0 v_x=0 v x = 0 , P 0 , c ( z ) = c z P_{0,c}(z)=cz P 0 , c ( z ) = cz , so K λ ( X , v ± ) = ∣ q ∓ 1 ∣ I ( X ) K_\lambda(X,v_\pm)=|q\mp1|I(X) K λ ( X , v ± ) = ∣ q ∓ 1∣ I ( X ) where I ( X ) : = ∫ [ − b , b ] ∩ [ X − 1 , X + 1 ] λ ( x ) d x I(X):=\int_{[-b,b]\cap[X-1,X+1]}\lambda(x)\,dx I ( X ) := ∫ [ − b , b ] ∩ [ X − 1 , X + 1 ] λ ( x ) d x . Averaging v + v_+ v + and v − v_- v − cancels μ v y \mu v_y μ v y : 1 2 ( Ψ ( X , v + ) + Ψ ( X , v − ) ) = I ( X ) + ∑ k η k \frac12(\Psi(X,v_+)+\Psi(X,v_-))=I(X)+\sum_k\eta_k 2 1 ( Ψ ( X , v + ) + Ψ ( X , v − )) = I ( X ) + ∑ k η k . When 2 b ≤ 6 2b\le6 2 b ≤ 6 , [ − b , b ] [-b,b] [ − b , b ] is covered by the three length-2 intervals centered at − b + 1 , 0 , b − 1 -b+1,0,b-1 − b + 1 , 0 , b − 1 , so max X I ( X ) ≥ 1 3 ∫ − b b λ \max_X I(X)\ge\frac13\int_{-b}^b\lambda max X I ( X ) ≥ 3 1 ∫ − b b λ , matching 1 3 ∫ − b b λ + ∑ k η k \frac13\int_{-b}^b\lambda+\sum_k\eta_k 3 1 ∫ − b b λ + ∑ k η k at q = 1 / 3 q=1/3 q = 1/3 . At W = 153 25 r = 6.12 r W=\frac{153}{25}r=6.12r W = 25 153 r = 6.12 r , our exact rational certificate verifies sup Ψ < D e m a n d \sup\Psi<\mathrm{Demand} sup Ψ < Demand .
Using exact rational Pythagorean cutoffs ( c k , s k ) = ( 2 m n m 2 + n 2 , m 2 − n 2 m 2 + n 2 ) (c_k,s_k)=\left(\frac{2mn}{m^2+n^2},\frac{m^2-n^2}{m^2+n^2}\right) ( c k , s k ) = ( m 2 + n 2 2 mn , m 2 + n 2 m 2 − n 2 ) and outward-rounded enclosures over [ − b , b ] × { ∥ v ∥ 2 ≤ 1 } [-b,b]\times\{\lVert v\rVert_2\le1\} [ − b , b ] × {∥ v ∥ 2 ≤ 1 } (verified independently by anc/verify_certificates.py and replayed in Lean 4 in ContinuumUpperCert.lean), Theorem 8.2 establishes three unrestricted ceilings:
At q = 37 / 100 q=37/100 q = 37/100 : W o p t ( 37 / 100 ) ≤ 107 20 r = 5.35 r < F ( 37 / 100 ) ≈ 5.5266 r W_{\mathrm{opt}}(37/100)\le\frac{107}{20}r=5.35r<F(37/100)\approx5.5266r W opt ( 37/100 ) ≤ 20 107 r = 5.35 r < F ( 37/100 ) ≈ 5.5266 r . By monotonicity of W o p t ( q ) W_{\mathrm{opt}}(q) W opt ( q ) , W o p t ( q ) ≤ 5.35 r < F ( q ) W_{\mathrm{opt}}(q)\le5.35r<F(q) W opt ( q ) ≤ 5.35 r < F ( q ) holds across the entire interval [ 0.37 , q ∗ ) [0.37,q_*) [ 0.37 , q ∗ ) where F ( q ∗ ) = 5.35 r F(q_*)=5.35r F ( q ∗ ) = 5.35 r (q ∗ = ( 4 + 4800 / 9849 − 1 ) / 3 ≈ 0.37278 q_*=(\sqrt{4+4800/9849}-1)/3\approx0.37278 q ∗ = ( 4 + 4800/9849 − 1 ) /3 ≈ 0.37278 ). This gives a numerical separation on that interval.
Upper bounds at bounded excursion
The height cap also bounds the vertical displacement available on each crossing. A supporting direction couples this displacement to transverse travel.
Theorem 8.4 (Pythagorean Inter-Wall Chord Inequality — BH-2). Let g = ( X , Y ) : R → [ − b , b ] × R g=(X,Y):\mathbb{R}\to[-b,b]\times\mathbb{R} g = ( X , Y ) : R → [ − b , b ] × R (b > 1 b>1 b > 1 , r = 1 r=1 r = 1 ) be any unit-speed patrol with sup t ∈ R ∣ Y ( t ) ∣ ≤ H \sup_{t\in\mathbb{R}}|Y(t)|\le H sup t ∈ R ∣ Y ( t ) ∣ ≤ H blocking straight wall runs at speed q ∈ ( 0 , 1 ) q\in(0,1) q ∈ ( 0 , 1 ) . Fix any unit pair ( α 0 , β 0 ) (\alpha_0,\beta_0) ( α 0 , β 0 ) (α 0 , β 0 ≥ 0 \alpha_0,\beta_0\ge0 α 0 , β 0 ≥ 0 , α 0 2 + β 0 2 = 1 \alpha_0^2+\beta_0^2=1 α 0 2 + β 0 2 = 1 ) and any cutoff ( c k , s k ) (c_k,s_k) ( c k , s k ) (c k > 0 c_k>0 c k > 0 , s k ≥ 0 s_k\ge0 s k ≥ 0 , c k 2 + s k 2 = 1 c_k^2+s_k^2=1 c k 2 + s k 2 = 1 ) satisfying the side condition
s k ≥ q α 0 c k . s_k\ge q\alpha_0c_k. s k ≥ q α 0 c k . Define
H e f f : = ( 1 − q β 0 ) H − q α 0 ( b − 1 ) , ψ k : = c k − q α 0 ( 1 − s k ) , req k : = max ( 0 , H e f f + ψ k ) . (28) H_{\mathrm{eff}}:=(1-q\beta_0)H-q\alpha_0(b-1),\qquad\psi_k:=c_k-q\alpha_0(1-s_k),\qquad\operatorname{req}_k:=\max(0,H_{\mathrm{eff}}+\psi_k). \tag*{(28)} H eff := ( 1 − q β 0 ) H − q α 0 ( b − 1 ) , ψ k := c k − q α 0 ( 1 − s k ) , req k := max ( 0 , H eff + ψ k ) . (28) Then every weight η b h , k ≥ 0 \eta_{\mathrm{bh},k}\ge0 η bh , k ≥ 0 adds + 2 q η b h , k +2q\eta_{\mathrm{bh},k} + 2 q η bh , k to the right-hand demand of (27) with zero ∣ q − v y ∣ |q-v_y| ∣ q − v y ∣ penalty, requiring interior gap mass 4 req k η b h , k 4\operatorname{req}_k\eta_{\mathrm{bh},k} 4 req k η bh , k on J = [ − b + 1 , b − 1 ] J=[-b+1,b-1] J = [ − b + 1 , b − 1 ] and wall-footprint weight 4 η b h , k f b h , k ( b − ∣ X ∣ ) ∣ v x ∣ 4\eta_{\mathrm{bh},k}f_{\mathrm{bh},k}(b-|X|)|v_x| 4 η bh , k f bh , k ( b − ∣ X ∣ ) ∣ v x ∣ , where f b h , k ( u ) : = max ( 0 , u 2 1 − u 2 − q α 0 2 ) 1 ( 0 , s k ) ( u ) f_{\mathrm{bh},k}(u):=\max\left(0,\frac{u}{2\sqrt{1-u^2}}-\frac{q\alpha_0}{2}\right)1_{(0,s_k)}(u) f bh , k ( u ) := max ( 0 , 2 1 − u 2 u − 2 q α 0 ) 1 ( 0 , s k ) ( u ) . Moreover, on every wall-optimal four-leg bowtie (q > 1 / 3 q>1/3 q > 1/3 ), choosing ( α 0 , β 0 ) = ( a / ℓ , h / ℓ ) (\alpha_0,\beta_0)=(a/\ell,h/\ell) ( α 0 , β 0 ) = ( a / ℓ , h / ℓ ) achieves exact algebraic equality in the inter-wall chord step.
Proof. Decompose [ 0 , L ] [0,L] [ 0 , L ] (q L > 2 H + 4 qL>2H+4 q L > 2 H + 4 , so both walls must be visited) into alternating left/right wall-labelled blocks B 1 , … , B 2 N B_1,\ldots,B_{2N} B 1 , … , B 2 N (N ≥ 1 N\ge1 N ≥ 1 ) inside disjoint intervals [ τ i − , τ i + ] [\tau_i^-,\tau_i^+] [ τ i − , τ i + ] , separated by full crossings [ τ i + , τ i + 1 − ] [\tau_i^+,\tau_{i+1}^-] [ τ i + , τ i + 1 − ] of J = [ − b + 1 , b − 1 ] J=[-b+1,b-1] J = [ − b + 1 , b − 1 ] (retain an even run of complete blocks with one further block supplying the last exit-to-entry interval). Every interval of length > 2 ( H + 1 ) / q >2(H+1)/q > 2 ( H + 1 ) / q visits both closed footprints: use on each wall the phase q q q times that interval’s midpoint. Hence the initial and final pieces, including at most one discarded complete block to ensure even parity, have uniformly bounded length for fixed ( q , H , b ) (q,H,b) ( q , H , b ) . Their phase spans and all endpoint terms below are O q , H , b ( 1 ) O_{q,H,b}(1) O q , H , b ( 1 ) . In particular t 2 N + 1 − − t 1 + = L − O q , H , b ( 1 ) t_{2N+1}^- - t_1^+=L-O_{q,H,b}(1) t 2 N + 1 − − t 1 + = L − O q , H , b ( 1 ) . On each block B i B_i B i , let t i − , t i + ∈ B i ∩ { ∣ X ∣ ≥ b − 1 } t_i^-,t_i^+\in B_i\cap\{|X|\ge b-1\} t i − , t i + ∈ B i ∩ { ∣ X ∣ ≥ b − 1 } minimize q t − Y ( t ) − 1 − ( b − ∣ X ( t ) ∣ ) 2 qt-Y(t)-\sqrt{1-(b-|X(t)|)^2} q t − Y ( t ) − 1 − ( b − ∣ X ( t ) ∣ ) 2 and maximize q t − Y ( t ) + 1 − ( b − ∣ X ( t ) ∣ ) 2 qt-Y(t)+\sqrt{1-(b-|X(t)|)^2} q t − Y ( t ) + 1 − ( b − ∣ X ( t ) ∣ ) 2 , and write u i ± : = b − ∣ X ( t i ± ) ∣ ∈ [ 0 , 1 ] u_i^\pm:=b-|X(t_i^\pm)|\in[0,1] u i ± := b − ∣ X ( t i ± ) ∣ ∈ [ 0 , 1 ] . Every phase covered on B i B_i B i ’s wall during B i B_i B i lies in an interval of length I i = 1 − ( u i − ) 2 + 1 − ( u i + ) 2 + q ( t i + − t i − ) + ( Y ( t i − ) − Y ( t i + ) ) I_i=\sqrt{1-(u_i^-)^2}+\sqrt{1-(u_i^+)^2}+q(t_i^+-t_i^-)+(Y(t_i^-)-Y(t_i^+)) I i = 1 − ( u i − ) 2 + 1 − ( u i + ) 2 + q ( t i + − t i − ) + ( Y ( t i − ) − Y ( t i + )) . Using the telescoping identity ∑ i = 1 2 N ( t i + − t i − ) = ( t 2 N + 1 − − t 1 + ) − ∑ i = 1 2 N ( t i + 1 − − t i + ) \sum_{i=1}^{2N}(t_i^+-t_i^-)=\left(t_{2N+1}^- -t_1^+\right)-\sum_{i=1}^{2N}(t_{i+1}^- -t_i^+) ∑ i = 1 2 N ( t i + − t i − ) = ( t 2 N + 1 − − t 1 + ) − ∑ i = 1 2 N ( t i + 1 − − t i + ) , full two-wall coverage ∑ i = 1 2 N I i ≥ 2 q L − O ( H ) \sum_{i=1}^{2N}I_i\ge2qL-O(H) ∑ i = 1 2 N I i ≥ 2 q L − O ( H ) rearranges (after subtracting q L qL q L ) to
q L − O ( H + b ) ≤ ∑ i = 1 2 N [ 1 − ( u i − ) 2 + 1 − ( u i + ) 2 + ( Y ( t i − ) − Y ( t i + ) ) − q ( t i + 1 − − t i + ) ] . qL-O(H+b)\le\sum_{i=1}^{2N}\left[\sqrt{1-(u_i^-)^2}+\sqrt{1-(u_i^+)^2}+(Y(t_i^-)-Y(t_i^+))-q(t_{i+1}^- -t_i^+)\right]. q L − O ( H + b ) ≤ i = 1 ∑ 2 N [ 1 − ( u i − ) 2 + 1 − ( u i + ) 2 + ( Y ( t i − ) − Y ( t i + )) − q ( t i + 1 − − t i + ) ] .
Since t i + ≤ τ i + < τ i + 1 − ≤ t i + 1 − t_i^+ \le\tau_i^+ < \tau_{i+1}^- \le t_{i+1}^- t i + ≤ τ i + < τ i + 1 − ≤ t i + 1 − and X ( t i + ) , X ( t i + 1 − ) X(t_i^+), X(t_{i+1}^-) X ( t i + ) , X ( t i + 1 − ) lie on opposite walls, ∣ X ( t i + 1 − ) − X ( t i + ) ∣ = 2 ( b − 1 ) + ( 1 − u i + ) + ( 1 − u i + 1 − ) |X(t_{i+1}^-) - X(t_i^+)| = 2(b-1) + (1-u_i^+) + (1-u_{i+1}^-) ∣ X ( t i + 1 − ) − X ( t i + ) ∣ = 2 ( b − 1 ) + ( 1 − u i + ) + ( 1 − u i + 1 − ) . By unit guard speed and Cauchy–Schwarz with ( α 0 , β 0 ) (\alpha_0,\beta_0) ( α 0 , β 0 ) ,
t i + 1 − − t i + ≥ α 0 ( 2 ( b − 1 ) + ( 1 − u i + ) + ( 1 − u i + 1 − ) ) + β 0 ( Y ( t i + 1 − ) − Y ( t i + ) ) . t_{i+1}^- - t_i^+ \ge\alpha_0\left(2(b-1)+(1-u_i^+)+(1-u_{i+1}^-)\right)+\beta_0\left(Y(t_{i+1}^-)-Y(t_i^+)\right). t i + 1 − − t i + ≥ α 0 ( 2 ( b − 1 ) + ( 1 − u i + ) + ( 1 − u i + 1 − ) ) + β 0 ( Y ( t i + 1 − ) − Y ( t i + ) ) .
Summing over i = 1 , … , 2 N i=1,\ldots,2N i = 1 , … , 2 N , ∑ i ( 1 − u i + 1 − ) = ∑ i ( 1 − u i − ) + O ( 1 ) \sum_i(1-u_{i+1}^-)=\sum_i(1-u_i^-)+O(1) ∑ i ( 1 − u i + 1 − ) = ∑ i ( 1 − u i − ) + O ( 1 ) and ∑ i ( Y ( t i + 1 − ) − Y ( t i + ) ) = ∑ i ( Y ( t i − ) − Y ( t i + ) ) + O ( H ) \sum_i(Y(t_{i+1}^-)-Y(t_i^+))=\sum_i(Y(t_i^-)-Y(t_i^+))+O(H) ∑ i ( Y ( t i + 1 − ) − Y ( t i + )) = ∑ i ( Y ( t i − ) − Y ( t i + )) + O ( H ) ; since 1 − q β 0 > 0 1-q\beta_0>0 1 − q β 0 > 0 and Y ( t i − ) − Y ( t i + ) ≤ 2 H Y(t_i^-)-Y(t_i^+)\le2H Y ( t i − ) − Y ( t i + ) ≤ 2 H , multiplying by 2 gives
2 q L − O ( H + b ) ≤ ∑ i = 1 2 N [ 2 F ( u i − ) + 2 F ( u i + ) + 4 H e f f ] , F ( u ) : = 1 − u 2 − q α 0 ( 1 − u ) . 2qL-O(H+b)\le\sum_{i=1}^{2N}\left[2F(u_i^-)+2F(u_i^+)+4H_{\mathrm{eff}}\right],\qquad F(u):=\sqrt{1-u^2}-q\alpha_0(1-u). 2 q L − O ( H + b ) ≤ i = 1 ∑ 2 N [ 2 F ( u i − ) + 2 F ( u i + ) + 4 H eff ] , F ( u ) := 1 − u 2 − q α 0 ( 1 − u ) .
Now use the side condition s k ≥ q α 0 c k s_k\ge q\alpha_0c_k s k ≥ q α 0 c k : for every u ∈ [ s k , 1 ] u\in[s_k,1] u ∈ [ s k , 1 ] , 1 − u 2 ≤ c k \sqrt{1-u^2}\le c_k 1 − u 2 ≤ c k , so
F ( u ) − F ( s k ) = ( u − s k ) ( q α 0 − u + s k 1 − u 2 + c k ) ≤ ( u − s k ) ( q α 0 − 2 s k 2 c k ) ≤ 0 , F(u)-F(s_k)=(u-s_k)\left(q\alpha_0-\frac{u+s_k}{\sqrt{1-u^2}+c_k}\right)\le(u-s_k)\left(q\alpha_0-\frac{2s_k}{2c_k}\right)\le0, F ( u ) − F ( s k ) = ( u − s k ) ( q α 0 − 1 − u 2 + c k u + s k ) ≤ ( u − s k ) ( q α 0 − 2 c k 2 s k ) ≤ 0 ,
whence F ( u ) ≤ F ( s k ) = ψ k F(u)\le F(s_k)=\psi_k F ( u ) ≤ F ( s k ) = ψ k on [ s k , 1 ] [s_k,1] [ s k , 1 ] without any footprint term. For u ∈ [ 0 , s k ) u\in[0,s_k) u ∈ [ 0 , s k ) , − F ′ ( w ) = w 1 − w 2 − q α 0 ≤ 2 f b h , k ( w ) -F'(w)=\frac{w}{\sqrt{1-w^2}}-q\alpha_0\le2f_{\mathrm{bh},k}(w) − F ′ ( w ) = 1 − w 2 w − q α 0 ≤ 2 f bh , k ( w ) , so F ( u ) ≤ ψ k + 2 ∫ u s k f b h , k ( w ) d w F(u)\le\psi_k+2\int_u^{s_k}f_{\mathrm{bh},k}(w)\,dw F ( u ) ≤ ψ k + 2 ∫ u s k f bh , k ( w ) d w . Letting u i ∗ : = min t ∈ B i ∩ { ∣ X ∣ ≥ b − 1 } ( b − ∣ X ( t ) ∣ ) ≤ min ( u i − , u i + ) u_i^*:=\min_{t\in B_i\cap\{|X|\ge b-1\}}(b-|X(t)|)\le\min(u_i^-,u_i^+) u i ∗ := min t ∈ B i ∩ { ∣ X ∣ ≥ b − 1 } ( b − ∣ X ( t ) ∣ ) ≤ min ( u i − , u i + ) , we obtain 2 F ( u i − ) + 2 F ( u i + ) + 4 H e f f ≤ 4 req k + 8 ∫ min ( u i ∗ , s k ) s k f b h , k ( w ) d w 2F(u_i^-)+2F(u_i^+)+4H_{\mathrm{eff}}\le4\operatorname{req}_k+8\int_{\min(u_i^*,s_k)}^{s_k}f_{\mathrm{bh},k}(w)\,dw 2 F ( u i − ) + 2 F ( u i + ) + 4 H eff ≤ 4 req k + 8 ∫ m i n ( u i ∗ , s k ) s k f bh , k ( w ) d w . On the single excursion inside B i B_i B i attaining u i ∗ u_i^* u i ∗ , b − ∣ X ∣ b-|X| b − ∣ X ∣ traverses [ min ( u i ∗ , s k ) , s k ] [\min(u_i^*,s_k),s_k] [ min ( u i ∗ , s k ) , s k ] twice, contributing
∫ τ i − τ i + 4 f b h , k ( b − ∣ X ∣ ) ∣ X ′ ∣ d t ≥ 8 ∫ min ( u i ∗ , s k ) s k f b h , k ( w ) d w , \int_{\tau_i^-}^{\tau_i^+}4f_{\mathrm{bh},k}(b-|X|)|X'|\,dt\ge8\int_{\min(u_i^*,s_k)}^{s_k}f_{\mathrm{bh},k}(w)\,dw, ∫ τ i − τ i + 4 f bh , k ( b − ∣ X ∣ ) ∣ X ′ ∣ d t ≥ 8 ∫ m i n ( u i ∗ , s k ) s k f bh , k ( w ) d w ,
while each gap crossing [ τ i + , τ i + 1 − ] [\tau_i^+,\tau_{i+1}^-] [ τ i + , τ i + 1 − ] contributes
∫ τ i + τ i + 1 − ρ b h , k ( X ) ∣ X ′ ∣ d t ≥ 4 req k . \int_{\tau_i^+}^{\tau_{i+1}^-}\rho_{\mathrm{bh},k}(X)|X'|\,dt\ge4\operatorname{req}_k. ∫ τ i + τ i + 1 − ρ bh , k ( X ) ∣ X ′ ∣ d t ≥ 4 req k .
Dividing by L → ∞ L\to\infty L → ∞ proves the + 2 q η b h , k +2q\eta_{\mathrm{bh},k} + 2 q η bh , k demand, and on any bowtie with ( α 0 , β 0 ) = ( a / ℓ , h / ℓ ) (\alpha_0,\beta_0)=(a/\ell,h/\ell) ( α 0 , β 0 ) = ( a / ℓ , h / ℓ ) , Cauchy–Schwarz along ( 2 a , 2 h ) (2a,2h) ( 2 a , 2 h ) holds with equality. □
The five bounded-excursion certificates at q = 1 / 4 q=1/4 q = 1/4 give the width ceilings in the first row below. The second row is the midline bound 2 5 + 3 H / r 2\sqrt{5}+3H/r 2 5 + 3 H / r of Theorem 6.1 at the same heights, rounded up.
Table 1 (PDF p. 30).
At the three heights H ≤ r H\le r H ≤ r the midline bound is the smaller of the two. At H = 5 r / 4 H=5r/4 H = 5 r /4 and H = 3 r / 2 H=3r/2 H = 3 r /2 the certificate is the smaller. In particular the tube value 8 r 8r 8 r exceeds every guarded width with H ≤ 1.50 r H\le1.50r H ≤ 1.50 r , by the last certificate and monotonicity in H H H . At arbitrary bounded excursion the certificate above gives 8.10 r 8.10r 8.10 r .
Slow intruders
Theorem 9.1. As q ↓ 0 q\downarrow0 q ↓ 0 , q W o p t ( q ) / r → 2 qW_{\mathrm{opt}}(q)/r\to2 q W opt ( q ) / r → 2 . For sufficiently small q > 0 q>0 q > 0 a four-leg patrol guards width ( 2 / q − 2 + 3 q + O ( q 2 ) ) r (2/q-2+3q+O(q^2))r ( 2/ q − 2 + 3 q + O ( q 2 )) r .
Proof. Set r = 1 r=1 r = 1 , a = q − 1 − 1 − q / 2 a=q^{-1}-1-q/2 a = q − 1 − 1 − q /2 and h = 1 − q + q 2 − 3 q 3 h=1-q+q^2-3q^3 h = 1 − q + q 2 − 3 q 3 . For small q q q these quantities are positive and ( 1 + q ) h = 1 − 2 q 3 − 3 q 4 < 1 (1+q)h=1-2q^3-3q^4<1 ( 1 + q ) h = 1 − 2 q 3 − 3 q 4 < 1 . Writing ℓ = a 2 + h 2 = q − 1 − 1 + O ( q 2 ) \ell=\sqrt{a^2+h^2}=q^{-1}-1+O(q^2) ℓ = a 2 + h 2 = q − 1 − 1 + O ( q 2 ) , we get
G = 2 q ℓ − ( 1 − q ) h = 1 − 2 q 2 + O ( q 3 ) , δ = 1 − G 2 = 2 q + O ( q 2 ) . G=2q\ell-(1-q)h=1-2q^2+O(q^3),\qquad\delta=\sqrt{1-G^2}=2q+O(q^2). G = 2 q ℓ − ( 1 − q ) h = 1 − 2 q 2 + O ( q 3 ) , δ = 1 − G 2 = 2 q + O ( q 2 ) .
Theorem 3.1 certifies width 2 ( a + δ ) = 2 / q − 2 + 3 q + O ( q 2 ) 2(a+\delta)=2/q-2+3q+O(q^2) 2 ( a + δ ) = 2/ q − 2 + 3 q + O ( q 2 ) . The area bound (2) gives q W o p t ( q ) ≤ 2 1 + q 2 qW_{\mathrm{opt}}(q)\le2\sqrt{1+q^2} q W opt ( q ) ≤ 2 1 + q 2 . Scaling back to general r r r proves the limit. □
An explicit ten-leg patrol at finite speed
The four-leg family in Theorem 9.1 has limiting additive loss 2 r 2r 2 r relative to 2 r / q 2r/q 2 r / q . A short inward kink in each return and two extra corners on each transfer give a smaller loss throughout a fixed speed interval.
Theorem 9.2. For every 0 < q ≤ 1 / 40 0<q\le1/40 0 < q ≤ 1/40 , a periodic ten-leg patrol of period 4 r / U 4r/U 4 r / U guards a width greater than ( 2 / q − 1.235352896057745 ) r (2/q-1.235352896057745)r ( 2/ q − 1.235352896057745 ) r .
Proof. Normalize r = V = 1 r=V=1 r = V = 1 . Use the exact rationals
h = 1.227689517575272716605 , e = 0.029002301252121295625. h=1.227689517575272716605,\qquad e=0.029002301252121295625. h = 1.227689517575272716605 , e = 0.029002301252121295625.
Set L = 1 + e L=1+e L = 1 + e , ρ = h 2 + e 2 \rho=\sqrt{h^2+e^2} ρ = h 2 + e 2 , α = 1 − q 2 \alpha=\sqrt{1-q^2} α = 1 − q 2 , and define
k = h − 1 + q ρ , v = k + q k 2 + α 2 L 2 α 2 , τ = L 2 + v 2 , Y = h − v , x = α Y / q , a = x + L , G = 2 − h − q ρ , δ = 1 − G 2 , b = a + δ . \begin{aligned}
k&=h-1+q\rho, & v&=\frac{k+q\sqrt{k^2+\alpha^2L^2}}{\alpha^2}, & \tau&=\sqrt{L^2+v^2},\\
Y&=h-v, & x&=\alpha Y/q, & a&=x+L,\\
G&=2-h-q\rho, & \delta&=\sqrt{1-G^2}, & b&=a+\delta.
\end{aligned} k Y G = h − 1 + q ρ , = h − v , = 2 − h − q ρ , v x δ = α 2 k + q k 2 + α 2 L 2 , = α Y / q , = 1 − G 2 , τ a b = L 2 + v 2 , = x + L , = a + δ .
The right return is ( a , h ) → ( x + 1 , 0 ) → ( a , − h ) (a,h)\to(x+1,0)\to(a,-h) ( a , h ) → ( x + 1 , 0 ) → ( a , − h ) , followed by the transfer ( a , − h ) → ( x , − Y ) → ( − x , Y ) → ( − a , h ) (a,-h)\to(x,-Y)\to(-x,Y)\to(-a,h) ( a , − h ) → ( x , − Y ) → ( − x , Y ) → ( − a , h ) . Reflect these five unit-speed legs horizontally to close the patrol. Write D = 2 ( τ + Y / q ) D=2(\tau+Y/q) D = 2 ( τ + Y / q ) for the transfer duration. The stated root of the quadratic for v v v gives v = k + q τ v=k+q\tau v = k + q τ (the unsquared sign is positive). Consequently
Y + q ( ρ + τ ) = 1 , G = q ( D + ρ ) − h , 4 ρ + 2 D = 4 / q . Y+q(\rho+\tau)=1,\qquad G=q(D+\rho)-h,\qquad4\rho+2D=4/q. Y + q ( ρ + τ ) = 1 , G = q ( D + ρ ) − h , 4 ρ + 2 D = 4/ q .
The middle leg has velocity ( − α , q ) (-\alpha,q) ( − α , q ) . Each return leg has duration ρ \rho ρ and the two outer transfer legs have duration τ \tau τ . All vertices lie in [ − b , b ] × R [-b,b]\times\mathbb{R} [ − b , b ] × R . The parameters satisfy v , Y > 0 v,Y>0 v , Y > 0 , 0 < G < 1 0<G<1 0 < G < 1 and, for R = 1 − ( δ + e ) 2 R=\sqrt{1-(\delta+e)^2} R = 1 − ( δ + e ) 2 ,
R > 0 , 1 − q ρ > 0 , δ 2 + ( h − R ) 2 < ( 1 − q ρ ) 2 . (29) R>0,\qquad1-q\rho>0,\qquad\delta^2+(h-R)^2<(1-q\rho)^2. \tag*{(29)} R > 0 , 1 − q ρ > 0 , δ 2 + ( h − R ) 2 < ( 1 − q ρ ) 2 . (29)
We give the uniform arithmetic check below.
Set the departure time from ( a , − h ) (a,-h) ( a , − h ) to zero. If the guard leg on J J J is affine, the set of spacetime points ( t , p ) (t,p) ( t , p ) for which some w ∈ J w\in J w ∈ J satisfies ∥ p − g ( w ) ∥ + q ∣ w − t ∣ ≤ 1 \lVert p-g(w)\rVert+q\lvert w-t\rvert\le1 ∥ p − g ( w )∥ + q ∣ w − t ∣ ≤ 1 is convex, by interpolating both points and witnesses. Every intruder occupancy of this set causes detection. We construct the separator from two horizontal levels
F + ( t ) = q ( t + ρ ) , F − ( t ) = q t − h − G . F_{+}(t)=q(t+\rho),\qquad F_{-}(t)=qt-h-G. F + ( t ) = q ( t + ρ ) , F − ( t ) = q t − h − G .
At the right-return midpoint t = − ρ t=-\rho t = − ρ , the anchor K = ( x , 0 ) K=(x,0) K = ( x , 0 ) and both wall points ( b , ± R ) (b,\pm R) ( b , ± R ) lie in the current disk. The anchor also belongs to the capture set of the first transfer leg via its corner at time τ \tau τ , since Y + q ( ρ + τ ) = 1 Y+q(\rho+\tau)=1 Y + q ( ρ + τ ) = 1 . The lower wall point belongs to the same set via the departure at time zero, by (29) .
Let U ( t ) U(t) U ( t ) interpolate affinely in spacetime through the successive points
( − ρ , K ) , ( τ , ( x , F + ( τ ) ) ) , ( D − τ , ( − x , F + ( D − τ ) ) ) , ( D , ( − a , h + G ) ) . (-\rho,K),\qquad(\tau,(x,F_{+}(\tau))),\qquad(D-\tau,(-x,F_{+}(D-\tau))),\qquad(D,(-a,h+G)). ( − ρ , K ) , ( τ , ( x , F + ( τ ))) , ( D − τ , ( − x , F + ( D − τ ))) , ( D , ( − a , h + G )) .
Its height is F + ( t ) F_{+}(t) F + ( t ) . At both transfer corners the upper and lower levels have offsets + 1 +1 + 1 and − 1 -1 − 1 from the guard, by the timing identities. At arrival the offset of U ( D ) U(D) U ( D ) is G G G . Thus each interpolated segment lies in the capture set of its own transfer leg; the first such segment starts at t = − ρ t=-\rho t = − ρ . For 0 ≤ t ≤ D 0 \le t \le D 0 ≤ t ≤ D the central symmetry of the transfer gives B ( t ) = − U ( D − t ) B(t)=-U(D-t) B ( t ) = − U ( D − t ) at height F − ( t ) F_-(t) F − ( t ) in the same leg’s capture set as U ( t ) U(t) U ( t ) . Their horizontal coordinates satisfy U x ( t ) ≤ B x ( t ) U_x(t) \le B_x(t) U x ( t ) ≤ B x ( t ) : on the first leg the difference is affine, nonpositive at departure and zero at the corner, on the middle leg it vanishes, and on the last leg the ordering follows by symmetry. Use the upper flat from the left wall to U ( t ) U(t) U ( t ) , the captured chord U ( t ) B ( t ) U(t)B(t) U ( t ) B ( t ) , and the lower flat from B ( t ) B(t) B ( t ) to the right wall.
Before departure, join the upper flat at U ( t ) U(t) U ( t ) to the wall trace interpolating from ( − ρ , ( b , − R ) ) (-\rho,(b,-R)) ( − ρ , ( b , − R )) to ( 0 , ( b , − h − G ) ) (0,(b,-h-G)) ( 0 , ( b , − h − G )) . Both endpoints and the joining chord belong to the first leg’s capture set, using δ 2 + G 2 = 1 \delta^2+G^2=1 δ 2 + G 2 = 1 at time zero. The change at departure sweeps only the captured triangle with vertices U ( 0 ) , ( a , − h − G ) , ( b , − h − G ) U(0),(a,-h-G),(b,-h-G) U ( 0 ) , ( a , − h − G ) , ( b , − h − G ) . The reflection ( t , p ) ↦ ( D − t , − p ) (t,p)\mapsto(D-t,-p) ( t , p ) ↦ ( D − t , − p ) gives the corresponding chord and reset at arrival. At the next return midpoint, the change from the upper to the lower wall endpoint sweeps the triangle between the reflected anchor and the two wall points, all in the current disk. Horizontal reflection now gives the second half-period. This bounded separator spans the corridor. Away from captured chords, endpoints, and reset triangles it is horizontal and rises at speed q q q , so the difference between intruder height and front height cannot increase there. Every upward transit must cross the separator and is detected.
For the width estimate, put E ( q ) = b − 1 / q E(q)=b-1/q E ( q ) = b − 1/ q . The timing identity gives the regular expression
E = − q 1 + α − α ( ρ + τ ) + L + δ . E=-\frac{q}{1+\alpha}-\alpha(\rho+\tau)+L+\delta. E = − 1 + α q − α ( ρ + τ ) + L + δ .
At zero, − 2 E ( 0 ) = 2 ( ρ + L 2 + ( h − 1 ) 2 − L − 1 − ( 2 − h ) 2 ) -2E(0)=2\left(\rho+\sqrt{L^2+(h-1)^2}-L-\sqrt{1-(2-h)^2}\right) − 2 E ( 0 ) = 2 ( ρ + L 2 + ( h − 1 ) 2 − L − 1 − ( 2 − h ) 2 ) . Differentiation gives τ ′ = v ( ρ + τ ) / ( τ − q v ) \tau'=v(\rho+\tau)/(\tau-qv) τ ′ = v ( ρ + τ ) / ( τ − q v ) and
E ′ = − 1 α ( 1 + α ) + q ( ρ + τ ) α − α τ ′ + ρ G δ . E'=-\frac{1}{\alpha(1+\alpha)}+\frac{q(\rho+\tau)}{\alpha}-\alpha\tau'+\frac{\rho G}{\delta}. E ′ = − α ( 1 + α ) 1 + α q ( ρ + τ ) − α τ ′ + δ ρG .
The standalone anc/hooks/check_slow_hooks.py encloses these radicals using rational intervals and integer square roots. Direct evaluation on the single interval [ 0 , 1 / 40 ] [0,1/40] [ 0 , 1/40 ] certifies the positivity conditions above, τ − q v > 0 \tau-qv>0 τ − q v > 0 , 1 / 5 < E ′ < 4 / 5 1/5<E'<4/5 1/5 < E ′ < 4/5 , and − 2 E ( 0 ) < 1.235352896057745 -2E(0)<1.235352896057745 − 2 E ( 0 ) < 1.235352896057745 . Only the desingularized expressions are evaluated at zero. Since 2 b = 2 / q + 2 E ( q ) > 2 / q + 2 E ( 0 ) 2b=2/q+2E(q)>2/q+2E(0) 2 b = 2/ q + 2 E ( q ) > 2/ q + 2 E ( 0 ) for q > 0 q>0 q > 0 , rescaling proves the theorem.
What remains open
Theorems 3.6 and 5.1 determine the full range of equality in the wall ceiling. Theorem 5.2 raises a more precise question. Does
lim q ↑ q a F ( q ) − W o p t ( q ) r ( q a − q ) 2 \lim_{q \uparrow q_a} \frac{F(q)-W_{\mathrm{opt}}(q)}{r(q_a-q)^2} q ↑ q a lim r ( q a − q ) 2 F ( q ) − W opt ( q )
exist, and which perturbations of the critical patrol determine it? A rescaled four-leg patrol gives an explicit upper bound on this ratio. The lower bound comes from recovering a stretch of the bowtie inside a nearly extremizing patrol and is existential. Optimizing within four legs alone would still require a comparison with other patrols.
At q = 1 / 4 q=1/4 q = 1/4 the ceilings and constructions presented here leave open whether any bounded-excursion patrol can guard width 8 r 8r 8 r . Periodic approximation permits a search over periodic patrols with arbitrarily long periods. A bound that couples the timing of consecutive wall visits to the set of reachable intruder positions could reduce the remaining interval. Theorem 9.2 gives lim sup q ↓ 0 ( 2 / q − W o p t ( q ) / r ) ≤ 1.235352896057745 \limsup_{q \downarrow0}(2/q-W_{\mathrm{opt}}(q)/r) \le1.235352896057745 lim sup q ↓ 0 ( 2/ q − W opt ( q ) / r ) ≤ 1.235352896057745 . The limiting loss and the two-way width below q = 1 / 2 q=1/2 q = 1/2 remain open.
Verification and AI assistance
The ancillary files contain the exact rational inputs and Python checks for the critical root, the canonical escape, the rescaled bowtie, the low-speed ten-leg patrol, the seven triangulated lower bounds and the eight flux and bounded-height upper bounds. Their entry points and hypotheses are indexed in anc/README.md. The finite checks complement the arguments on paper that turn capture surfaces, time budgets and clearance into statements about every transit.
The accompanying Lean 4 and Mathlib v4.15.0 library contains the wall-budget algebra, selected phase and fan identities, a root bracket, excursion inequalities, and kernel replays of the seven surface and eight continuum certificates. The topological reductions, periodic approximation, escape transfer, and two-sided quadratic estimate are proved in the text. The scope of each executable verification is described in the ancillary guide.
OpenAI GPT/Codex, Anthropic Claude, and Google DeepMind Gemini/Antigravity assisted with mathematical exploration, programming, proof auditing, drafting, and Lean tactics. The distinction between the analytic proofs and their executable finite checks applies throughout.