Polyhedra, Moduli,
and Quantum Geometry

Web working edition

On this page
  1. 20.1 The length function and its integral
  2. 20.2 A positive partition of unity

20 An exact torus twist calculation and a constant orbit series

20.1 The length function and its integral

Let XX be an oriented one-holed hyperbolic torus with geodesic boundary length b>0b>0, or a complete finite-area once-punctured torus when b=0b=0. Cut along an essential simple curve α\alpha of length a>0a>0, and choose an essential simple curve β\beta crossing α\alpha once. Normalize the twist origin so that this crossing curve has minimum length at τ=0\tau=0. Set A2=cosh⁡a+cosh⁡(b/2)cosh⁡a−1.A^2=\frac{\cosh a+\cosh(b/2)}{\cosh a-1}.(20.1) The exact formula is cosh⁡(ℓβ(τ)/2)=Acosh⁡(τ/2).\boxed{\cosh\bigl(\ell_\beta(\tau)/2\bigr)=A\cosh(\tau/2).}(20.2) It follows from the right-angled-hexagon description of the cut pants; see also [43, Proposition 3.1, pp. 5–6]. Take the positive square root A>1A>1. For b=0b=0, A=coth⁡(a/2)A=\coth(a/2). The minimum length is r0=2arcosh⁡Ar_0=2\mathop{\mathrm{arcosh}}A.

There is also a matrix check fixing the signed boundary convention. At zero twist the two axes meet perpendicularly. In an adapted basis put M=(ea/200e−a/2),Bu=(eu/200e−u/2)(AA2−1A2−1A).M=\begin{pmatrix}e^{a/2}&0\\0&e^{-a/2}\end{pmatrix},\qquad B_u=\begin{pmatrix}e^{u/2}&0\\0&e^{-u/2}\end{pmatrix} \begin{pmatrix}A&\sqrt{A^2-1}\\\sqrt{A^2-1}&A\end{pmatrix}. Then det⁡Bu=1\det B_u=1, tr⁡Bu=2Acosh⁡(u/2)\mathop{\mathrm{tr}}B_u=2A\cosh(u/2), and direct multiplication gives tr⁡[M,Bu]=2−4sinh⁡2(a/2)(A2−1)=−2cosh⁡(b/2),[M,Bu]=MBuM−1Bu−1.\mathop{\mathrm{tr}}[M,B_u]=2-4\sinh^2(a/2)(A^2-1)=-2\cosh(b/2), \qquad [M,B_u]=MB_uM^{-1}B_u^{-1}. Solving this boundary equation recovers (20.1). Differentiating (20.2), with Q=A2cosh⁡2(u/2)−1>0Q=A^2\cosh^2(u/2)-1>0, also gives ℓβ′(u)=Asinh⁡(u/2)Q,ℓβ″(u)=A(A2−1)cosh⁡(u/2)2Q3/2>0.\ell'_\beta(u)=\frac{A\sinh(u/2)}{\sqrt Q},\qquad \ell''_\beta(u)=\frac{A(A^2-1)\cosh(u/2)}{2Q^{3/2}}>0.(20.3) In particular |ℓβ′|<1|\ell'_\beta|<1, consistent with the single crossing in (19.5).

For any nonnegative measurable function ff, change variables separately on the positive and negative twist axes to obtain ∫ℝf(ℓβ(u))⁡du=2∫r0∞f(r)sinh⁡(r/2)cosh⁡2(r/2)−A2⁡dr.\int_{\mathbb R}f\bigl(\ell_\beta(u)\bigr)\,\mathop{}\!\mathrm du =2\int_{r_0}^\infty f(r)\frac{\sinh(r/2)}{\sqrt{\cosh^2(r/2)-A^2}}\,\mathop{}\!\mathrm dr.(20.4) Indeed, differentiating cosh⁡(r/2)=Acosh⁡(u/2)\cosh(r/2)=A\cosh(u/2) gives ⁡du/⁡dr=sinh⁡(r/2)/cosh⁡2(r/2)−A2\mathop{}\!\mathrm du/\mathop{}\!\mathrm dr=\sinh(r/2)/\sqrt{\cosh^2(r/2)-A^2} for u>0u>0. The two signs of uu explain the factor 22. The nonnegative equality allows infinity on both sides; for signed ff require absolute integrability. The coarea density’s endpoint singularity at r0r_0 is integrable. For example, y=cosh⁡2(r/2)−A2y=\sqrt{\cosh^2(r/2)-A^2} converts the right-hand side to 4∫0∞f(2arcosh⁡A2+y2)⁡dy/A2+y24\int_0^\infty f(2\mathop{\mathrm{arcosh}}\sqrt{A^2+y^2})\,\mathop{}\!\mathrm dy/\sqrt{A^2+y^2}.

Several useful checks are immediate: ∫ℝsech⁡2(ℓβ(u)/2)⁡du=4A2,(20.5)meas⁡{u:ℓβ(u)≤R}=4arcosh⁡(cosh⁡(R/2)A),R≥r0,(20.6)∫ℝcosh⁡−s(ℓβ(u)/2)⁡du=2πΓ(s/2)AsΓ((s+1)/2),s>0.(20.7)\begin{aligned} \int_{\mathbb R}\mathop{\mathrm{sech}}^2\bigl(\ell_\beta(u)/2\bigr)\,\mathop{}\!\mathrm du &=\frac4{A^2},\\ \operatorname{meas}\{u:\ell_\beta(u)\le R\} &=4\mathop{\mathrm{arcosh}}\left(\frac{\cosh(R/2)}{A}\right),\quad R\ge r_0, \\ \int_{\mathbb R}\cosh^{-s}\bigl(\ell_\beta(u)/2\bigr)\,\mathop{}\!\mathrm du &=\frac{2\sqrt\pi\,\Gamma(s/2)} {A^s\Gamma((s+1)/2)},\quad s>0. \end{aligned} The sublevel measure is zero for R<r0R<r_0. In the last integral the change of variable t=tanh⁡(u/2)t=\tanh(u/2) gives 2A−sB(1/2,s/2)2A^{-s}B(1/2,s/2), which proves the displayed gamma quotient. These are exact twist integrals. For a finite interval [−U,U][-U,U], U≥0U\geq0, the omitted tail satisfies ∫|u|>Ucosh⁡−s(ℓβ(u)/2)⁡du≤2s+2sAse−sU/2,s>0,\int_{|u|>U}\cosh^{-s}(\ell_\beta(u)/2)\,\mathop{}\!\mathrm du \leq\frac{2^{s+2}}{sA^s}e^{-sU/2},\qquad s>0,(20.8) because cosh⁡(u/2)≥e|u|/2/2\cosh(u/2)\geq e^{|u|/2}/2. For s=2s=2 the tail is exactly 4A−2(1−tanh⁡(U/2))4A^{-2}(1-\tanh(U/2)). These analytic tails do not enclose numerical quadrature or roundoff error.

20.2 A positive partition of unity

Define Ha(u)=sinh⁡(a/2)cosh⁡u+cosh⁡(a/2)=12[tanh(u+a/22)−tanh(u−a/22)].H_a(u)=\frac{\sinh(a/2)}{\cosh u+\cosh(a/2)} =\frac12\left[ \tanh\left(\frac{u+a/2}{2}\right) -\tanh\left(\frac{u-a/2}{2}\right)\right].(20.9) The second expression telescopes under translation by aa, so ∑n∈ℤHa(τ+na)=1.\sum_{n\in\mathbb Z}H_a(\tau+na)=1.(20.10) All terms are positive and the convergence is exponential on compact twist intervals. More precisely, for N≥0N\geq0 put LN=(N+1/2)aL_N=(N+1/2)a and PN(τ)=∑n=−NNHa(τ+na)P_N(\tau)=\sum_{n=-N}^NH_a(\tau+na). Finite telescoping gives PN(τ)=12[tanhτ+LN2−tanhτ−LN2],1−PN(τ)=11+eLN+τ+11+eLN−τ≤2cosh⁡τe−LN.(20.11)\begin{aligned} P_N(\tau)&=\tfrac12\left[ \tanh\frac{\tau+L_N}{2}-\tanh\frac{\tau-L_N}{2}\right],\\ 1-P_N(\tau)&=\frac1{1+e^{L_N+\tau}}+ \frac1{1+e^{L_N-\tau}} \leq2\cosh\tau\,e^{-L_N}. \end{aligned} This is an explicit tail estimate, uniform for bounded τ\tau with aa fixed. One may reduce τ\tau modulo aa to |τ|≤a/2|\tau|\leq a/2 by relabeling the orbit; no estimate uniform as a→0a\to0 is asserted for fixed NN.

Since (20.2) implies cosh⁡r+1=A2(cosh⁡u+1)\cosh r+1=A^2(\cosh u+1), eliminate uu from (20.9) and set Ka,b(r)=A2sinh⁡(a/2)cosh⁡r+1+A2(cosh⁡(a/2)−1).K_{a,b}(r)= \frac{A^2\sinh(a/2)} {\cosh r+1+A^2(\cosh(a/2)-1)}.(20.12) For r≥r0r\geq r_0, this satisfies Ka,b(r)=Ha(u)K_{a,b}(r)=H_a(u) at the two twists with length rr. For rn=ℓDαnβ(X)r_n=\ell_{D_\alpha^n\beta}(X) we obtain the pointwise identity ∑n∈ℤKa,b(rn)=1.\boxed{\sum_{n\in\mathbb Z}K_{a,b}(r_n)=1.}(20.13) This is a constant series proved by telescoping, rather than inferred from its average. Its kernel depends on the distinguished cuff length aa and the boundary length bb, and its index set is one twist orbit. It is not an unrestricted universal sum over every curve on the surface. The classes DαnβD_\alpha^n\beta are all distinct as unoriented simple classes, since they cross α\alpha once and have distinct integral twist coordinates. Equal lengths, which occur at symmetric twist values, do not merge their labels. Equation (20.11) is also the tail estimate for this KK-series.

There is a simple source of many such examples. If ρ∈L1(ℝ)\rho\in L^1(\mathbb R) is an even nonnegative probability density, then Ha,ρ(u)=∫u−a/2u+a/2ρ(v)⁡dvsatisfies∑n∈ℤHa,ρ(u+na)=1.H_{a,\rho}(u)=\int_{u-a/2}^{u+a/2}\rho(v)\,\mathop{}\!\mathrm dv \quad\text{satisfies}\quad \sum_{n\in\mathbb Z}H_{a,\rho}(u+na)=1. The translated intervals partition the line except for shared endpoints of measure zero, so Tonelli proves the identity. These terms are nonnegative; strict positivity requires further assumptions on ρ\rho. Evenness makes Ha,ρH_{a,\rho} a function of |u||u|, hence of the length rr through (20.2). The choice ρ(v)=14sech⁡2(v/2)\rho(v)=\frac14\mathop{\mathrm{sech}}^2(v/2) gives (20.9). This realizes the proposed “zoo” of identities in a concrete way, with the source of constancy and the parameter dependence both visible.

Computational support.

Groups G5–G7 of experiments/quantum_geodesics/README.md compare the one-crossing holonomy, derivatives, coarea and beta-integral formulas, and finite H/K lattice partitions. Stable positive formulas retain the small tail signals at the declared parameters. Analytic integral and lattice tails control those truncations, separately from uncertified numerical quadrature and roundoff. The result path is artifacts/quantum_geodesics/result.json; book/claims/sections17-22.json records the exact formula links.