Introduction
For a complete finite-area hyperbolic surface with cusps, the continuous spectrum begins at 1 / 4 1/4 1/4 . We prove that the spectral gap of a Weil–Petersson random surface converges in probability to this threshold as the genus g g g tends to infinity, for every positive integer cusp sequence n = n ( g ) = o ( g ) n=n(g)=o(\sqrt{g}) n = n ( g ) = o ( g ) . Our quantitative result bounds the expected number of nonzero eigenvalues in ( 0 , 1 / 4 − ε ] (0,1/4-\varepsilon] ( 0 , 1/4 − ε ] for each fixed 0 < ε < 1 / 4 0<\varepsilon<1/4 0 < ε < 1/4 , uniformly in the cusp count.
Let M g , n \mathcal{M}_{g,n} M g , n be the moduli space of connected, complete, finite-area oriented hyperbolic surfaces of genus g g g with n n n labelled cusps, without boundary or orbifold points. We equip it with normalized Weil–Petersson measure P g , n \mathbb{P}_{g,n} P g , n and write E g , n \mathbb{E}_{g,n} E g , n for expectation. The spectral statements concern integers g ≥ 3 g\ge3 g ≥ 3 and n ≥ 1 n\ge1 n ≥ 1 ; the geometric expansions also allow n = 0 n=0 n = 0 . The hyperbolic area of each such surface is
A g , n = 2 π ( 2 g − 2 + n ) . (1) \mathcal{A}_{g,n}=2\pi(2g-2+n). \tag*{(1)} A g , n = 2 π ( 2 g − 2 + n ) . (1)
We reserve V g , n V_{g,n} V g , n for the Weil–Petersson volume of moduli space.
For X ∈ M g , n X\in\mathcal{M}_{g,n} X ∈ M g , n with n ≥ 1 n\ge1 n ≥ 1 , the nonnegative self-adjoint Laplacian Δ X \Delta_X Δ X has a simple eigenvalue zero, continuous spectrum [ 1 / 4 , ∞ ) [1/4,\infty) [ 1/4 , ∞ ) , and possibly nonzero discrete eigenvalues below $1/4. For 0 < ε < 1 / 4 0<\varepsilon<1/4 0 < ε < 1/4 , define
N ε ( X ) = # { λ ∈ Spec L 2 ( Δ X ) : 0 < λ ≤ 1 / 4 − ε } . (2) N_{\varepsilon}(X)=\#\{\lambda\in\operatorname{Spec}_{L^2}(\Delta_X):0<\lambda\le1/4-\varepsilon\}. \tag*{(2)} N ε ( X ) = # { λ ∈ Spec L 2 ( Δ X ) : 0 < λ ≤ 1/4 − ε } . (2)
Here the count is taken with multiplicity and includes both cuspidal and residual eigenvalues. Write SpG ( X ) = inf Spec ( Δ X ∣ 1 ⊥ ) \operatorname{SpG}(X)=\inf\operatorname{Spec}(\Delta_X|_{1^\perp}) SpG ( X ) = inf Spec ( Δ X ∣ 1 ⊥ ) , where 1 ⊥ 1^\perp 1 ⊥ is the orthogonal complement of the constant functions in L 2 ( X ) L^2(X) L 2 ( X ) . Thus SpG ( X ) = 1 / 4 \operatorname{SpG}(X)=1/4 SpG ( X ) = 1/4 if there is no nonzero eigenvalue below 1 / 4 1/4 1/4 , and { SpG ( X ) ≤ 1 / 4 − ε } = { N ε ( X ) ≥ 1 } \{\operatorname{SpG}(X)\le1/4-\varepsilon\}=\{N_{\varepsilon}(X)\ge1\} { SpG ( X ) ≤ 1/4 − ε } = { N ε ( X ) ≥ 1 } .
Theorem 1.1 (Spectral gaps throughout the sub-square-root cusp range). *For every fixed 0 < ε < 1 / 4 0<\varepsilon<1/4 0 < ε < 1/4 there are constants C ε ≥ 2 C_{\varepsilon}\ge2 C ε ≥ 2 and g ε ≥ 3 g_{\varepsilon}\ge3 g ε ≥ 3 , independent of n n n , such that, for all integers g ≥ 3 g\ge3 g ≥ 3 and n ≥ 1 n\ge1 n ≥ 1 satisfying
g ≥ g ε , g ≥ C ε ( n + 1 ) 2 , (3) g\ge g_{\varepsilon},\qquad g\ge C_{\varepsilon}(n+1)^2, \tag*{(3)} g ≥ g ε , g ≥ C ε ( n + 1 ) 2 , (3)
one has
P g , n { SpG ( X ) ≤ 1 / 4 − ε } ≤ E g , n N ε ( X ) ≤ C ε ( n + 1 ) 2 g . (4) \mathbb{P}_{g,n}\{\operatorname{SpG}(X) \le1/4-\varepsilon\} \le\mathbb{E}_{g,n}N_{\varepsilon}(X) \le C_{\varepsilon}\frac{(n+1)^2}{g}. \tag*{(4)} P g , n { SpG ( X ) ≤ 1/4 − ε } ≤ E g , n N ε ( X ) ≤ C ε g ( n + 1 ) 2 . (4)
In particular, for every positive integer sequence n ( g ) = o ( g ) n(g)=o(\sqrt{g}) n ( g ) = o ( g ) , the expected count and this probability tend to zero. For every envelope n max ( g ) ≥ 1 n_{\max}(g)\ge1 n m a x ( g ) ≥ 1 with n max ( g ) = o ( g ) n_{\max}(g)=o(\sqrt{g}) n m a x ( g ) = o ( g ) , the following bound holds for all sufficiently large g g g , uniformly over integer cusp counts 1 ≤ n ≤ n max ( g ) 1\le n\le n_{\max}(g) 1 ≤ n ≤ n m a x ( g ) :
sup 1 ≤ n ≤ n max ( g ) E g , n N ε ≤ C ε ( n max ( g ) + 1 ) 2 g ⟶ 0. (5) \sup_{1\le n\le n_{\max}(g)}\mathbb{E}_{g,n}N_{\varepsilon}\le C_{\varepsilon}\frac{(n_{\max}(g)+1)^2}{g}\longrightarrow0. \tag*{(5)} 1 ≤ n ≤ n m a x ( g ) sup E g , n N ε ≤ C ε g ( n m a x ( g ) + 1 ) 2 ⟶ 0. (5)
For example, the expected count is O ε , δ ( g 2 δ − 1 ) O_{\varepsilon,\delta}(g^{2\delta-1}) O ε , δ ( g 2 δ − 1 ) uniformly for 1 ≤ n ≤ g δ 1\le n\le g^{\delta} 1 ≤ n ≤ g δ , for each fixed 0 ≤ δ < 1 / 2 0\le\delta<1/2 0 ≤ δ < 1/2 . It is also O ε ( ( log log g ) − 2 ) O_{\varepsilon}((\log\log g)^{-2}) O ε (( log log g ) − 2 ) uniformly for 1 ≤ n ≤ g / log log g 1\le n\le\sqrt{g}/\log\log g 1 ≤ n ≤ g / log log g . Thus the estimate covers cusp sequences outside every range n = O ( g δ ) n=O(g^{\delta}) n = O ( g δ ) with fixed δ < 1 / 2 \delta<1/2 δ < 1/2 . The same bounds hold for the probability in (4) . Throughout, ε \varepsilon ε is fixed: the estimate gives neither a shrinking spectral window nor a bound tending to zero when n ≍ g n\asymp\sqrt{g} n ≍ g . The proof is completed in Section 6.9.
A matching lower bound shows that the order ( n + 1 ) 2 / g (n+1)^2/g ( n + 1 ) 2 / g is optimal for both the expected count and the probability in the quadratic domain, for each fixed ε \varepsilon ε . Proposition 6.14 gives the lower bound c ε min { ( n + 1 ) 2 / g , 1 } c_{\varepsilon}\min\{(n+1)^2/g,1\} c ε min {( n + 1 ) 2 / g , 1 } for all n ≥ 1 n\ge1 n ≥ 1 once g g g exceeds a bound depending only on ε \varepsilon ε . Corollary 6.15 gives the two-sided estimate and shows that convergence of the spectral gap to 1 / 4 1/4 1/4 holds if and only if n = o ( g ) n=o(\sqrt{g}) n = o ( g ) .
Relation to previous results
For closed Weil–Petersson random surfaces, Mirzakhani established a uniform positive spectral gap with probability tending to one [Mir13]. Wu and Xue [23 Theorem 1] and, independently, Lipnowski and Wright [13 Theorem 1.1] raised the lower bound to 3 / 16 − ε 3/16-\varepsilon 3/16 − ε . Anantharaman and Monk subsequently reached 1 / 4 − ε 1/4-\varepsilon 1/4 − ε for every fixed ε > 0 \varepsilon>0 ε > 0 through their work on Friedman–Ramanujan functions and geometric expansions [2 Theorem 1.6] . Hide, Macera and Thomas gave a different proof using the polynomial method and allowed the spectral window to shrink at a polynomial rate: for some fixed η > 0 \eta>0 η > 0 , the gap is at least 1 / 4 − O ( g − η ) 1/4-O(g^{-\eta}) 1/4 − O ( g − η ) with probability tending to one [10 Theorem 1.1] .
For Weil–Petersson surfaces with cusps, the known lower bounds depend on the growth of the cusp count. Hide proved that, for each fixed 0 ≤ σ < 1 / 2 0\le\sigma<1/2 0 ≤ σ < 1/2 and n = O ( g σ ) n=O(g^{\sigma}) n = O ( g σ ) , the spectral gap exceeds
1 4 − ( 2 σ + 1 4 ) 2 − ε \frac{1}{4}-\left(\frac{2\sigma+1}{4}\right)^2-\varepsilon 4 1 − ( 4 2 σ + 1 ) 2 − ε
with probability tending to one [8 Theorem 1.3] . He, Wu and Xue improved this bound to
1 4 − ( 1 6 ( 1 − σ ) ) 2 − ε \frac{1}{4}-\left(\frac{1}{6(1-\sigma)}\right)^2-\varepsilon 4 1 − ( 6 ( 1 − σ ) 1 ) 2 − ε
in the same fixed-power range [7 Theorem 1.2] . When the number of cusps is bounded, the two bounds are 3 / 16 − ε 3/16-\varepsilon 3/16 − ε and 2 / 9 − ε 2/9-\varepsilon 2/9 − ε , respectively. The proof of the latter identifies a second-order cancellation between geometric contributions.
Shen and Wu obtained a positive spectral lower bound throughout n = o ( g ) n=o(\sqrt{g}) n = o ( g ) and formulated the corresponding 1 / 4 − ε 1/4-\varepsilon 1/4 − ε statement as Conjecture 4 [21 Theorem 3 and Conjecture 4] . The same conjecture is restated in [7 Conjecture 1.1] . Shen and Wu also showed that the spectral gap tends to zero in probability when n / g → ∞ n/\sqrt{g}\to\infty n / g → ∞ and n / g → 0 n/g\to0 n / g → 0 [21 Theorem 1] . At the critical scale n / g → c ∈ ( 0 , ∞ ) n/\sqrt{g}\to c\in(0,\infty) n / g → c ∈ ( 0 , ∞ ) , their Theorem 2 gives an explicit limit for the distribution function of the Cheeger constant on an interval near zero. Together, these results identify g \sqrt{g} g as a natural scale for the change in spectral behavior.
Li and Wang obtained large-genus volume asymptotics uniform over all cusp counts [LW26, Theorem 1.1]. They also extended the vanishing-gap conclusion to every sequence with n / g → ∞ n/\sqrt{g} \to\infty n / g → ∞ as g → ∞ g \to\infty g → ∞ , without assuming n / g → 0 n/g \to0 n / g → 0 [LW26, Corollary 1.6].
Together with Shen and Wu’s critical-scale result and the inverse Cheeger inequality, this already implies that n = o ( g ) n = o(\sqrt{g}) n = o ( g ) is necessary for convergence in probability of the spectral gap to 1 / 4 1/4 1/4 ; see [21 Theorems 2 and 11] and [LW26, Corollary 1.6]. Proposition 6.14 gives a uniform quantitative form of this obstruction.
Theorem 1.1 proves Shen–Wu’s Conjecture 4 for every positive integer sequence n = o ( g ) n = o(\sqrt{g}) n = o ( g ) , with the quantitative bound C ε ( n + 1 ) 2 / g C_{\varepsilon}(n+1)^2/g C ε ( n + 1 ) 2 / g on the expected count. Residual eigenvalues are included, as they are in the spectral-gap statements cited above.
For comparison, random-cover results use a different probability measure: one samples uniformly from homomorphisms of the base surface group to the symmetric group S d S_d S d , obtaining degree-d d d covers with labelled fibre. For a fixed finite-area cusped base and every fixed ε > 0 \varepsilon> 0 ε > 0 , Hide and Magee proved that, with probability tending to one, the spectrum below 1 / 4 − ε 1/4-\varepsilon 1/4 − ε consists exactly of the eigenvalues inherited from the base, with the same multiplicities [11 Theorem 1.1] . Magee, Puder and van Handel proved the corresponding statement for closed bases, with threshold 1 / 4 − o ( 1 ) 1/4-o(1) 1/4 − o ( 1 ) [14 Theorem 1.2] . For closed bases, Hide, Macera and Thomas refined the threshold to 1 / 4 − c d − β 1/4-cd^{-\beta} 1/4 − c d − β , where d d d is the covering degree and c , β > 0 c,\beta> 0 c , β > 0 [9 Theorem 1.1] .
In the Brooks–Makover model, Shen and Wu obtained a spectral gap greater than 1 / 4 − C / log M 1/4-C/\log M 1/4 − C / log M with probability tending to one for the closed Belyi surfaces obtained by conformally compactifying the gluings of 2 M 2M 2 M ideal triangles [22 Theorem 1.1] . These sampling measures differ from the Weil–Petersson measure with prescribed genus and cusp number used here.
The geometric argument uses Mirzakhani’s integration formula and volume-polynomial identities [Mir07a, Mir07b, Mir08] and the large-genus expansions of Mirzakhani and Zograf [19 ] . Anantharaman and Monk developed integration and expansion formulas organized by the subsurface filled by a geodesic [1 §4.1, Theorems 5.7, 5.15 and 6.1] . For a fixed number of marked points, Guillermit refines the expansion-order bounds and obtains volume asymptotics for boundary lengths on the g \sqrt{g} g scale, while leaving the dependence on the marked-point count implicit [Gui26, Theorems 2, 4 and 5, Remark 3]. We use the joint estimates of Hide, Macera and Thomas, which track both the marked-point count and the expansion order [10 Theorem 4.2] . On the analytic side, the polynomial method for strong convergence introduced by Chen, Garza-Vargas, Tropp and van Handel [4 ] was extended to surface-group permutation representations in [14 ] and adapted to the closed Weil–Petersson model [10 ] .
The uniform count bound uses an additive genus condition whose quadratic cusp term is independent of the detector degree, and a bound for the first orbital correction independent of that degree. All polynomial tests of a given maximum degree are evaluated on one cusp truncation.
A finite-parameter estimate
The proof applies a degree-Q Q Q polynomial to a bounded spectral multiplier. For λ ≥ 1 / 4 \lambda\ge1/4 λ ≥ 1/4 , the chosen detector test decays exponentially in Q Q Q . The additive genus condition below permits Q ≍ log g Q \asymp\log g Q ≍ log g , which yields Theorem 1.1.
Fix an even nonnegative function b 0 ∈ C c ∞ ( R ) b_0 \in C^\infty_c(\mathbb{R}) b 0 ∈ C c ∞ ( R ) of integral one, and set h = b 0 ∗ b 0 h=b_0*b_0 h = b 0 ∗ b 0 and f = h ^ f=\widehat{h} f = h . Define L = max { 1 , sup { ∣ u ∣ : u ∈ supp h } } L=\max\{1,\sup\{|u|:u\in\operatorname{supp}h\}\} L = max { 1 , sup { ∣ u ∣ : u ∈ supp h }} , which is determined by f f f .
Our Fourier transform conventions are
H ^ ( r ) = ∫ R H ( u ) e − i r u d u , F ~ ( u ) = 1 2 π ∫ R F ( r ) e i r u d r . (6) \widehat{H}(r)=\int_{\mathbb{R}}H(u)e^{-iru}\,du,\qquad\widetilde{F}(u)=\frac{1}{2\pi}\int_{\mathbb{R}}F(r)e^{iru}\,dr. \tag*{(6)} H ( r ) = ∫ R H ( u ) e − i r u d u , F ( u ) = 2 π 1 ∫ R F ( r ) e i r u d r . (6)
The function f f f is even and entire. On the real axis, f = b ^ 0 2 f=\widehat{b}_0^2 f = b 0 2 and 0 ≤ f ≤ 1 0\le f\le1 0 ≤ f ≤ 1 ; on the positive imaginary axis, t ↦ f ( i t ) t\mapsto f(it) t ↦ f ( i t ) is strictly increasing. In particular,
a = f ( i / 2 ) > 1. (7) a=f(i/2)>1. \tag*{(7)} a = f ( i /2 ) > 1. (7)
In the spectral variable λ = 1 / 4 + r 2 \lambda=1/4+r^2 λ = 1/4 + r 2 , the multiplier f ( r ) f(r) f ( r ) maps λ ≥ 1 / 4 \lambda\ge1/4 λ ≥ 1/4 into [ 0 , 1 ] [0,1] [ 0 , 1 ] and the zero eigenvalue to a a a . We fix f f f throughout. For 0 < ε < 1 / 4 0<\varepsilon<1/4 0 < ε < 1/4 , define
b ε = f ( i ε ) , κ = ( b ε − 1 ) 2 16 a 2 . (8) b_{\varepsilon}=f(i\sqrt{\varepsilon}),\qquad\kappa=\frac{(b_{\varepsilon}-1)^2}{16a^2}. \tag*{(8)} b ε = f ( i ε ) , κ = 16 a 2 ( b ε − 1 ) 2 . (8)
Then 1 < b ε < a 1<b_{\varepsilon}<a 1 < b ε < a and κ > 0 \kappa>0 κ > 0 . Let Q ε Q_{\varepsilon} Q ε be the integer threshold defined in (211) of § 6.2 ; it depends only on f f f and ε \varepsilon ε . Eigenvalues counted by N ε N_{\varepsilon} N ε are mapped into [ b ε , a ) [b_{\varepsilon},a) [ b ε , a ) . The detector constructed in § 6.2 is uniformly positive there and exponentially small on [ 0 , 1 ] [0,1] [ 0 , 1 ] . The lower bound Q ≥ Q ε Q\ge Q_{\varepsilon} Q ≥ Q ε ensures the detector and retained-mass estimates used below.
Put s = n + 1 s=n+1 s = n + 1 . For each surface X X X and degree bound Q Q Q , we use one cusp truncation for all real polynomial tests p p p of degree at most Q Q Q with p ( 0 ) = 0 p(0)=0 p ( 0 ) = 0 . Its height is
Y Q = 3 e L Q . (9) Y_Q=3e^{LQ}. \tag*{(9)} Y Q = 3 e L Q . (9)
Writing p ( x ) = x p ~ ( x ) p(x)=x\widetilde{p}(x) p ( x ) = x p ( x ) , the resulting traces are integrals of p ~ \widetilde{p} p against a common finite positive measure on [ 0 , a ] [0,a] [ 0 , a ] (Theorem 5.3 ).
Theorem 1.2 (Finite-parameter spectral count). There are fixed constants C a d d , D 0 , D 3 ≥ 2 C_{\mathrm{add}},D_0,D_3\ge2 C add , D 0 , D 3 ≥ 2 , depending only on the multiplier f f f , such that for every fixed 0 < ε < 1 / 4 0<\varepsilon<1/4 0 < ε < 1/4 there is C f , ε C_{f,\varepsilon} C f , ε with the following property. For all integers g ≥ 3 g\ge3 g ≥ 3 , n ≥ 1 n\ge1 n ≥ 1 and Q ≥ Q ε Q\ge Q_{\varepsilon} Q ≥ Q ε satisfying
g ≥ C a d d s 2 + C a d d Q D 3 s , (10) g\ge C_{\mathrm{add}}s^2+C_{\mathrm{add}}Q^{D_3}s, \tag*{(10)} g ≥ C add s 2 + C add Q D 3 s , (10) one has
P g , n ( N ε ≥ 1 ) ≤ E g , n N ε ≤ C f , ε { ( g + n Q 2 ) e − κ Q + s 2 g + s 3 Q 3 D 0 g 2 + s 2 g 2 − Q + n A g , n Y Q } . (11) \mathbb{P}_{g,n}(N_{\varepsilon}\ge1)\le\mathbb{E}_{g,n}N_{\varepsilon}\le C_{f,\varepsilon}\left\{(g+nQ^2)e^{-\kappa Q}+\frac{s^2}{g}+\frac{s^3Q^{3D_0}}{g^2}+\frac{s^2}{g}2^{-Q}+\frac{n}{A_{g,n}Y_Q}\right\}. \tag*{(11)} P g , n ( N ε ≥ 1 ) ≤ E g , n N ε ≤ C f , ε { ( g + n Q 2 ) e − κ Q + g s 2 + g 2 s 3 Q 3 D 0 + g s 2 2 − Q + A g , n Y Q n } . (11)
The constants D 0 D_0 D 0 , D 3 D_3 D 3 and C a d d C_{\mathrm{add}} C add are specified in Lemma 6.7 and Proposition 6.9 ; Theorem 6.13 proves the estimate. To deduce Theorem 1.1 , take Q = ⌈ K ε log g ⌉ Q=\lceil K_{\varepsilon}\log g\rceil Q = ⌈ K ε log g ⌉ , where K ε = max { 1 , 4 / κ , 2 / L } K_{\varepsilon}=\max\{1,4/\kappa,2/L\} K ε = max { 1 , 4/ κ , 2/ L } . If g ≥ 2 C a d d s 2 g\ge2C_{\mathrm{add}}s^2 g ≥ 2 C add s 2 , the quadratic term in (10) is at most g / 2 g/2 g /2 . The linear term is also at most g / 2 g/2 g /2 once g g g exceeds a bound depending only on f f f and ε \varepsilon ε , since Q D 3 = o ( g ) Q^{D_3}=o(\sqrt{g}) Q D 3 = o ( g ) . Moreover, the higher-order term in (11) , relative to s 2 / g s^2/g s 2 / g , is O f , ε ( Q 3 D 0 / g ) = o ( 1 ) O_{f,\varepsilon}(Q^{3D_0}/\sqrt{g})=o(1) O f , ε ( Q 3 D 0 / g ) = o ( 1 ) uniformly in this range. The remaining terms are bounded by a constant multiple of s 2 / g s^2/g s 2 / g . After fixing f f f , this yields the stated constants C ε C_{\varepsilon} C ε and g ε g_{\varepsilon} g ε , independent of n n n .
Proof structure and conventions
The proof has three main steps.
First, we obtain geometric expansions uniform in the cusp count and the expansion order. We integrate over simple and nonsimple geodesics, including repetitions of primitive geodesics. For a nonsimple geodesic, the integration is organized by the subsurface it fills, called its carrier. Complementary-volume estimates and labelled carrier integration give an orbital expansion whose coefficients are independent of genus and polynomial in n n n (§ 2 , § 3 and § 3.8 ). Retaining the full binomial decay in the complementary-volume tails gives an additive genus condition: a quadratic cusp term with an absolute coefficient, plus a linear cusp term whose coefficient depends on the expansion order and length cutoff (§ 4.2 ).
Second, we use one cusp cutoff for all monomials of a given maximum degree. This gives a common positive trace functional with an exact formula that includes the parabolic contribution. The leading orbital functional is represented by a unit atom at a = f ( i / 2 ) a=f(i/2) a = f ( i /2 ) plus a finite signed measure on [ 0 , 1 ] [0,1] [ 0 , 1 ] . Subtracting the constant eigenfunction’s contribution to the truncated trace cancels this atom up to the explicitly controlled mass removed by truncation (§ 5 ). Every eigenfunction with eigenvalue counted by N ε N_\varepsilon N ε , including a residual eigenfunction, retains a uniformly positive fraction of its mass, allowing the polynomial detector to bound the eigenvalue count (§ 6 ).
Third, we bound the first correction—the coefficient of g − 1 g^{-1} g − 1 in the orbital expansion—independently of the detector degree. For each fixed test, this coefficient is a polynomial of degree at most two in n n n . Interpolation over auxiliary genera bounds the canonical coefficients at finitely many cusp counts; their polynomial dependence on n n n extends the bounds to all cusp counts (§ 6.6 ). These bounds transfer the expansion to polynomial tests at the additive genus threshold (§ 6.7 ). At each fixed n n n , positivity of the trace after subtracting the constant mode’s contribution shows that the restriction of the first correction’s distributional extension to ( 1 , ∞ ) (1,\infty) ( 1 , ∞ ) is a locally finite positive measure supported in ( 1 , a ] (1,a] ( 1 , a ] . Its local mass bound and the detector’s exponential decay on a fixed interval strictly below the transition give an estimate independent of degree (§ 6.8 ). Combining these estimates proves Theorems 1.2 and 1.1 (§ 6.9 ).
The table summarizes the parameter ranges. Here R R R is the geometric expansion order, T T T the length cutoff, and j j j the index of the coefficient multiplying g 1 − j g^{1-j} g 1 − j in the averaged polynomial orbital expansion. Canonical coefficient bounds require no genus condition; estimates for geodesic sums and spectral traces require the stated bounds. Constants are absolute in the first row and may depend on f f f in the other rows; detector and spectral count bounds may also depend on ε \varepsilon ε .
Table 1.
Our orbital sums are indexed by primitive oriented geodesics, with their repetitions included separately; the two orientations are distinct. The separating kernel sums both orders of the two components, and each pair of boundary components glued to one another has a single twist coordinate. Weil–Petersson volumes use the intersection-theoretic orbifold convention V 0 , 3 = 1 V_{0,3}=1 V 0 , 3 = 1 and V 1 , 1 ( ℓ ) = ( ℓ 2 + 4 π 2 ) / 48 V_{1,1}(\ell)=(\ell^{2}+4\pi^{2})/48 V 1 , 1 ( ℓ ) = ( ℓ 2 + 4 π 2 ) /48 . Together with the Fourier convention ((6) ), these normalizations fix the factors in the gluing formulas and spectral trace.
Uniform complementary volumes and boundary kernels
The geometric averages in Section 3 involve a chosen subsurface, called the carrier, and the components glued to its boundary. Their Weil–Peterson volume ratios form the boundary kernels constructed here. We expand these kernels in inverse powers of genus, with explicit bounds in the cusp count, expansion order, and boundary lengths. An envelope valid for every stable component controls configurations with large deficit (§ 2.1 –§ 2.2 ). Joint intersection-number estimates give boundary-volume expansions and shifted ratios for the large complementary component (§ 2.3 –§ 2.4 ). Section 2.5 assembles these inputs into a single coefficient sequence, independent of genus and truncation order. Section 4 sharpens these genus thresholds, giving the additive condition in Section 4.2.
For nonnegative integers u u u , v v v , a type ( u , v ) (u,v) ( u , v ) is stable when its complexity c = 2 u + v − 2 c=2u+v-2 c = 2 u + v − 2 is at least one; this includes ( 0 , 3 ) (0,3) ( 0 , 3 ) and ( 1 , 1 ) (1,1) ( 1 , 1 ) . For a stable type, let V u , v ( ℓ ) V_{u,v}(\ell) V u , v ( ℓ ) denote the Weil–Petersson orbifold volume of genus-u u u surfaces with v v v labelled ends of lengths ℓ ∈ [ 0 , ∞ ) v \ell\in[0,\infty)^v ℓ ∈ [ 0 , ∞ ) v . A positive length specifies a geodesic boundary and a zero length specifies a cusp. We write V u , v = V u , v ( 0 , … , 0 ) V_{u,v}=V_{u,v}(0,\ldots,0) V u , v = V u , v ( 0 , … , 0 ) . All expansion orders are nonnegative integers, and all constants in this section are absolute. We allow C C C to increase between estimates; one sufficiently large choice works throughout, independently of genus, cusp count, expansion order and carrier type.
A volume envelope valid for every stable component
The following bounds apply to the carrier and every complementary component, including those of genus zero or one.
Lemma 2.1. Put A 0 = 4 π 2 A_0=4\pi^2 A 0 = 4 π 2 . There are U ≥ 1 U\ge1 U ≥ 1 , β v > 0 \beta_v>0 β v > 0 and κ v > 0 \kappa_v>0 κ v > 0 such that
V u , v ≤ U A 0 c − 1 ( c − 1 ) ! c (12) V_{u,v}\le U A_0^{c-1}\frac{(c-1)!}{\sqrt{c}} \tag*{(12)} V u , v ≤ U A 0 c − 1 c ( c − 1 )! (12) ( c = 2 u + v − 2 ≥ 1 ) (c=2u+v-2\ge1) ( c = 2 u + v − 2 ≥ 1 ) ,
V g , n ≥ β v A 0 N N ! g exp ( − κ v n ( n + 1 ) 2 g ) (13) V_{g,n}\ge\beta_v A_0^N\frac{N!}{\sqrt{g}}\exp\left(-\frac{\kappa_v n(n+1)}{2g}\right) \tag*{(13)} V g , n ≥ β v A 0 N g N ! exp ( − 2 g κ v n ( n + 1 ) ) (13) ( g ≥ 2 , n ≥ 0 , N = 2 g + n − 3 ) (g\ge2,\ n\ge0,\ N=2g+n-3) ( g ≥ 2 , n ≥ 0 , N = 2 g + n − 3 ) .
Also, for every stable type and every nonnegative length vector,
V u , v ( ℓ ) ≤ V u , v exp ( 1 2 ∑ j ℓ j ) . (14) V_{u,v}(\ell)\le V_{u,v}\exp\left(\frac{1}{2}\sum_j\ell_j\right). \tag*{(14)} V u , v ( ℓ ) ≤ V u , v exp ( 2 1 j ∑ ℓ j ) . (14)
Proof. We use Mirzakhani’s boundary estimate and monotonicity V u , v ≤ V u + 1 , v − 2 V_{u,v}\le V_{u+1,v-2} V u , v ≤ V u + 1 , v − 2 for v ≥ 4 v\ge4 v ≥ 4 and 2 u + v − 4 > 0 2u+v-4>0 2 u + v − 4 > 0 [Mir13, Lemma 3.2(2)–(3)], the positive fixed-v v v large-genus asymptotic [19 Theorem 1.2] , and the global marked-point ratio estimate of Mirzakhani–Zograf [19 Lemma 5.1] , in the form recalled in [7 Lemma B.6] :
∣ ( 2 g − 2 + j ) V g , j V g , j + 1 − A 0 − 1 ∣ ≤ C B j + 1 2 g − 2 + j (15) \left|\frac{(2g-2+j)V_{g,j}}{V_{g,j+1}}-A_0^{-1}\right|\le C_B\frac{j+1}{2g-2+j} \tag*{(15)} V g , j + 1 ( 2 g − 2 + j ) V g , j − A 0 − 1 ≤ C B 2 g − 2 + j j + 1 (15)
( g ≥ 2 , j ≥ 0 ) (g\ge2,\ j\ge0) ( g ≥ 2 , j ≥ 0 ) .
The j + 1 j+1 j + 1 formulation in [7 Lemma B.6] includes the zero-mark case j = 0 j=0 j = 0 . For j = 0 j=0 j = 0 , the same estimate also follows from the fixed-marked-point asymptotic, after enlarging C B C_B C B to cover the finitely many remaining genera.
For the upper bound, first absorb the type ( 0 , 4 ) (0,4) ( 0 , 4 ) into U U U . For every other stable type, iterate the monotonicity while v ≥ 4 v\ge4 v ≥ 4 . It preserves c c c and ends in one of the four families v = 0 , 1 , 2 , 3 v=0,1,2,3 v = 0 , 1 , 2 , 3 . In each family the fixed-v v v asymptotic bounds the normalized volume in (12) ; the finitely many initial stable terms are absorbed in U U U . This includes ( 0 , 3 ) (0,3) ( 0 , 3 ) and ( 1 , 1 ) (1,1) ( 1 , 1 ) , without a step from v = 3 v=3 v = 3 to v = 1 v=1 v = 1 .
For the lower bound, set 1 + ϵ j = A 0 ( 2 g − 2 + j ) V g , j / V g , j + 1 1+\epsilon_j=A_0(2g-2+j)V_{g,j}/V_{g,j+1} 1 + ϵ j = A 0 ( 2 g − 2 + j ) V g , j / V g , j + 1 and κ v = A 0 C B \kappa_v=A_0C_B κ v = A 0 C B . Positivity and (15) give log ( 1 + ϵ j ) ≤ κ v ( j + 1 ) / ( 2 g − 2 + j ) \log(1+\epsilon_j)\le\kappa_v(j+1)/(2g-2+j) log ( 1 + ϵ j ) ≤ κ v ( j + 1 ) / ( 2 g − 2 + j ) . Since 2 g − 2 + j ≥ g 2g-2+j\ge g 2 g − 2 + j ≥ g , telescoping yields
V g , n ≥ V g , 0 A 0 n ( 2 g + n − 3 ) ! ( 2 g − 3 ) ! e − κ v n ( n + 1 ) / ( 2 g ) . V_{g,n}\ge V_{g,0}A_0^n\frac{(2g+n-3)!}{(2g-3)!}e^{-\kappa_v n(n+1)/(2g)}. V g , n ≥ V g , 0 A 0 n ( 2 g − 3 )! ( 2 g + n − 3 )! e − κ v n ( n + 1 ) / ( 2 g ) .
The n = 0 n=0 n = 0 asymptotic and positivity imply inf g ≥ 2 V g , 0 g / [ A 0 2 g − 3 ( 2 g − 3 ) ! ] > 0 \inf_{g\ge2}V_{g,0}\sqrt{g}/[A_0^{2g-3}(2g-3)!]>0 inf g ≥ 2 V g , 0 g / [ A 0 2 g − 3 ( 2 g − 3 )!] > 0 , proving (13) . The first cited boundary estimate gives (14) , including zero lengths by continuity. □ \square □
The labelled complementary tail
Fix integers g ≥ 2 g \ge2 g ≥ 2 , n ≥ 0 n \ge0 n ≥ 0 and a carrier type ( h 0 , a 0 , k ) (h_0,a_0,k) ( h 0 , a 0 , k ) , where h 0 h_0 h 0 is the genus, a 0 a_0 a 0 the cusp count, and k k k the number of boundary components. Assume h 0 ≥ 0 h_0 \ge0 h 0 ≥ 0 , 0 ≤ a 0 ≤ n 0 \le a_0 \le n 0 ≤ a 0 ≤ n , k ≥ 1 k \ge1 k ≥ 1 and m = 2 h 0 + a 0 + k − 2 ≥ 1 m = 2h_0 + a_0 + k - 2 \ge1 m = 2 h 0 + a 0 + k − 2 ≥ 1 , and label the boundary slots by { 1 , … , k } \{1,\ldots,k\} { 1 , … , k } .
A gluing pattern P = ( M , Π ) P = (M,\Pi) P = ( M , Π ) consists of a matching M M M of 2 σ 2\sigma 2 σ slots into σ \sigma σ pairs and a partition Π = ( I 1 , … , I b ) \Pi= (I_1,\ldots,I_b) Π = ( I 1 , … , I b ) of the remaining slots into b ≥ 1 b \ge1 b ≥ 1 nonempty blocks, ordered by least label. A matched pair glues two carrier boundaries to one another; a block I i I_i I i specifies the boundaries attached to the i i i th complementary component. Thus k i = ∣ I i ∣ k_i = |I_i| k i = ∣ I i ∣ satisfies ∑ i k i = k − 2 σ \sum_i k_i = k - 2\sigma ∑ i k i = k − 2 σ . The complementary genera and cusp counts are nonnegative integers ( g i , a i ) (g_i,a_i) ( g i , a i ) satisfying
∑ i a i = n − a 0 , ∑ i g i = g − h 0 − k + σ + b , c i = 2 g i + a i + k i − 2 ≥ 1. (16) \sum_i a_i = n-a_0,\qquad\sum_i g_i = g-h_0-k+\sigma+b,\qquad c_i = 2g_i+a_i+k_i-2 \ge1. \tag*{(16)} i ∑ a i = n − a 0 , i ∑ g i = g − h 0 − k + σ + b , c i = 2 g i + a i + k i − 2 ≥ 1. (16)
An empty admissible sum is zero. Write N N N for the ambient factorial index and R i R_i R i for the corresponding index of the i i i th complementary component. The deficit δ \delta δ is the loss of index on passing to a largest complementary component. Define
N = 2 g + n − 3 , R i = c i − 1 , δ = N − max i R i , d = m + b − 1 , (17) N = 2g+n-3,\qquad R_i = c_i-1,\qquad\delta= N-\max_i R_i,\qquad d = m+b-1, \tag*{(17)} N = 2 g + n − 3 , R i = c i − 1 , δ = N − i max R i , d = m + b − 1 , (17)
W = n ! a 0 ! ∏ i a i ! V h 0 , a 0 + k ∏ i V g i , a i + k i V g , n , T g , n h 0 , a 0 , k ( R ) = ∑ P ∑ δ ≥ R + 1 W . (18) W = \frac{n!}{a_0!\prod_i a_i!}\frac{V_{h_0,a_0+k}\prod_i V_{g_i,a_i+k_i}}{V_{g,n}},\qquad
\mathcal{T}^{h_0,a_0,k}_{g,n}(R) = \sum_P\sum_{\delta\ge R+1} W. \tag*{(18)} W = a 0 ! ∏ i a i ! n ! V g , n V h 0 , a 0 + k ∏ i V g i , a i + k i , T g , n h 0 , a 0 , k ( R ) = P ∑ δ ≥ R + 1 ∑ W . (18)
The multinomial in W W W assigns the n n n cusp labels to the carrier and the complementary components, and is included only once. This labelled sum provides an upper bound before the gluing symmetry factors are inserted in § 3 . A matched pair contributes a gluing edge, rather than an additional annular component.
Theorem 2.2 (Complementary-volume tail). There is an absolute K ≥ 4096 K \ge4096 K ≥ 4096 such that, simultaneously for all the above data and all integers R ≥ 0 R \ge0 R ≥ 0 , with t = R + m + 1 t = R+m+1 t = R + m + 1 and s = n + 1 s = n+1 s = n + 1 ,
g ≥ [ K t s ] 8 ⟹ T g , n h 0 , a 0 , k ( R ) ≤ g − R − 1 [ K t s ] 16 t . (19) g \ge[Kt s]^8 \Longrightarrow\mathcal{T}^{h_0,a_0,k}_{g,n}(R) \le g^{-R-1}[Kt s]^{16t}. \tag*{(19)} g ≥ [ K t s ] 8 ⟹ T g , n h 0 , a 0 , k ( R ) ≤ g − R − 1 [ K t s ] 16 t . (19)
For a fixed pattern P P P , let E P E_P E P consist of one edge for each matched pair and one edge for each unmatched slot, so ∣ E P ∣ = k − σ |E_P| = k-\sigma ∣ E P ∣ = k − σ . Assign a length y e ≥ 0 y_e \ge0 y e ≥ 0 to each edge and put y = ( y e ) e ∈ E P y = (y_e)_{e\in E_P} y = ( y e ) e ∈ E P . Write x j ( y ) x_j(y) x j ( y ) for the length of the edge incident to slot j j j ; matched slots therefore have the same length. Put
Λ = ∑ j = 1 k x j ( y ) , J P ( y ) = ∏ e ∈ E P y e . (20) \Lambda= \sum_{j=1}^{k}x_j(y),\qquad J_P(y) = \prod_{e\in E_P}y_e. \tag*{(20)} Λ = j = 1 ∑ k x j ( y ) , J P ( y ) = e ∈ E P ∏ y e . (20)
In the tail for this pattern, replace the numerator volumes by boundary volumes with these slot lengths, keeping the cusp lengths zero. After multiplication by J P J_P J P , the resulting tail is bounded by
g − R − 1 [ K t ( n + 1 ) ] 16 t ( 1 + Λ ) k e Λ ( g ≥ [ K t ( n + 1 ) ] 8 ) . (21) g^{-R-1}[Kt(n+1)]^{16t}(1+\Lambda)^{k}e^\Lambda\qquad\left(g \ge[Kt(n+1)]^8\right). \tag*{(21)} g − R − 1 [ K t ( n + 1 ) ] 16 t ( 1 + Λ ) k e Λ ( g ≥ [ K t ( n + 1 ) ] 8 ) . (21)
Proof. Euler characteristic and (16) give
∑ i c i = N + 1 − m , ∑ i R i = N − d , δ ≥ d ≥ m . (22) \sum_i c_i = N+1-m,\qquad\sum_i R_i = N-d,\qquad\delta\ge d \ge m. \tag*{(22)} i ∑ c i = N + 1 − m , i ∑ R i = N − d , δ ≥ d ≥ m . (22)
When g ≥ max { m + 2 , κ v s 2 } g \ge\max\{m+2,\kappa_v s^2\} g ≥ max { m + 2 , κ v s 2 } , the exponential in (13) is at least e − 1 / 2 e^{-1/2} e − 1/2 , and a largest component has c ∗ ≥ ( 2 g + n − 2 − m ) / b ≥ g / b c_* \ge(2g+n-2-m)/b \ge g/b c ∗ ≥ ( 2 g + n − 2 − m ) / b ≥ g / b . Its square-root denominator therefore cancels the ambient g \sqrt{g} g up to a factor b \sqrt{b} b . The total exponent of A 0 A_0 A 0 is − b -b − b . Discarding the other square-root denominators in Lemma 2.1 gives, with one absolute B ≥ 2 B \ge2 B ≥ 2 ,
W ≤ P b ∏ i R i ! n ! N ! a 0 ! ∏ i a i ! , P b = B b + 2 b ( m − 1 ) ! . (23) W \le P_b \frac{\prod_i R_i! \, n!}{N! \, a_0! \prod_i a_i!}, \qquad P_b = B^{b+2}\sqrt{b}\,(m-1)!. \tag*{(23)} W ≤ P b N ! a 0 ! ∏ i a i ! ∏ i R i ! n ! , P b = B b + 2 b ( m − 1 )! . (23)
This includes R i = 0 R_i=0 R i = 0 .
Choose a component ∗ \ast ∗ maximizing R i R_i R i , breaking ties by the fixed block order. The identity
δ = m + ∑ i ≠ ∗ c i \delta= m+\sum_{i\ne\ast}c_i δ = m + i = ∗ ∑ c i
shows that bounded deficit bounds the total complexity of the carrier and the other complementary components. The cusps outside component ∗ \ast ∗ include those on the carrier. Since a i ≤ c i + 1 a_i\le c_i+1 a i ≤ c i + 1 , the number of cusps outside component ∗ \ast ∗ satisfies
A o u t = a 0 + ∑ i ≠ ∗ a i ≤ a 0 + δ − m + b − 1 = δ + 1 − 2 h 0 − k + b ≤ δ + 1. (24) A_{\mathrm{out}}=a_0+\sum_{i\ne\ast}a_i\le a_0+\delta-m+b-1=\delta+1-2h_0-k+b\le\delta+1. \tag*{(24)} A out = a 0 + i = ∗ ∑ a i ≤ a 0 + δ − m + b − 1 = δ + 1 − 2 h 0 − k + b ≤ δ + 1. (24)
Since a ∗ = n − A o u t a_\ast=n-A_{\mathrm{out}} a ∗ = n − A out ,
n ! a 0 ! ∏ i a i ! = n ! ( n − A o u t ) ! a 0 ! ∏ i ≠ ∗ a i ! ≤ s A o u t , \frac{n!}{a_0!\prod_i a_i!}=\frac{n!}{(n-A_{\mathrm{out}})!\,a_0!\prod_{i\ne\ast}a_i!}\le s^{A_{\mathrm{out}}}, a 0 ! ∏ i a i ! n ! = ( n − A out )! a 0 ! ∏ i = ∗ a i ! n ! ≤ s A out ,
which gives the first bound below; the multinomial theorem gives the second:
n ! a 0 ! ∏ i a i ! ≤ min { s δ + 1 , ( b + 1 ) n } . (25) \frac{n!}{a_0!\prod_i a_i!}\le\min\{s^{\delta+1},(b+1)^n\}. \tag*{(25)} a 0 ! ∏ i a i ! n ! ≤ min { s δ + 1 , ( b + 1 ) n } . (25)
This bound includes a pair of pants attached along one boundary, whose other two ends may be cusps.
For a fixed pattern and deficit there are at most b b b choices of ∗ \ast ∗ . The other R i R_i R i sum to δ − d \delta-d δ − d and hence have at most ( δ + 1 ) b − 1 (\delta+1)^{b-1} ( δ + 1 ) b − 1 choices. For each i ≠ ∗ i\ne\ast i = ∗ , the cusp count a i a_i a i has at most δ + 3 \delta+3 δ + 3 choices, since 0 ≤ a i ≤ R i + 2 ≤ δ + 2 0\le a_i\le R_i+2\le\delta+2 0 ≤ a i ≤ R i + 2 ≤ δ + 2 ; the identity 2 g i = R i + 3 − k i − a i 2g_i=R_i+3-k_i-a_i 2 g i = R i + 3 − k i − a i then fixes the genus. The two sum constraints in (16) then determine the largest component. Discarding the remaining admissibility and parity conditions gives at most
b ( δ + 3 ) 2 ( b − 1 ) (26) b(\delta+3)^{2(b-1)} \tag*{(26)} b ( δ + 3 ) 2 ( b − 1 ) (26)
complementary allocations. Also ∏ i ≠ ∗ R i ! ≤ ( δ − d ) ! ≤ δ ! \prod_{i\ne\ast}R_i!\le(\delta-d)!\le\delta! ∏ i = ∗ R i ! ≤ ( δ − d )! ≤ δ ! . Thus the contribution at this deficit is at most P b b x δ P_b b x_\delta P b b x δ , where
x δ = s δ + 1 ( δ + 3 ) 2 b ( N δ ) . (27) x_\delta=\frac{s^{\delta+1}(\delta+3)^{2b}}{\binom{N}{\delta}}. \tag*{(27)} x δ = ( δ N ) s δ + 1 ( δ + 3 ) 2 b . (27)
We split the deficit sum into a near range, where successive terms decay geometrically, and a deep range, where a binomial coefficient absorbs the total number of allocations. Set D = max { m , R + 1 } D=\max\{m,R+1\} D = max { m , R + 1 } and J = ⌊ N / ( 64 s ) ⌋ J=\lfloor N/(64s)\rfloor J = ⌊ N / ( 64 s )⌋ . By increasing the absolute K K K , the hypothesis in (19) ensures
g ≤ N ≤ 3 g , D ≤ t , b ≤ m + 2 ≤ t + 1 , J ≥ t , N − D + 1 ≥ g , N ≥ 128 s . g\le N\le3g,\qquad D\le t,\qquad b\le m+2\le t+1,\qquad J\ge t,\qquad N-D+1\ge g,\qquad N\ge128s. g ≤ N ≤ 3 g , D ≤ t , b ≤ m + 2 ≤ t + 1 , J ≥ t , N − D + 1 ≥ g , N ≥ 128 s .
For D ≤ δ < J D\le\delta<J D ≤ δ < J , use b ≤ δ + 2 b\le\delta+2 b ≤ δ + 2 to obtain
x δ + 1 x δ = s δ + 1 N − δ ( 1 + 1 δ + 3 ) 2 b ≤ e 2 63 < 1 2 . \frac{x_{\delta+1}}{x_\delta}=s\frac{\delta+1}{N-\delta}\left(1+\frac{1}{\delta+3}\right)^{2b}\le\frac{e^2}{63}<\frac12. x δ x δ + 1 = s N − δ δ + 1 ( 1 + δ + 3 1 ) 2 b ≤ 63 e 2 < 2 1 .
Consequently the near tail D ≤ δ ≤ J D \le\delta\le J D ≤ δ ≤ J for this pattern is at most
2 P b b s D + 1 ( D + 3 ) 2 b D ! g − R − 1 , (28) 2P_b b s^{D+1}(D+3)^{2b}D!g^{-R-1}, \tag*{(28)} 2 P b b s D + 1 ( D + 3 ) 2 b D ! g − R − 1 , (28)
since ( N ) D ≥ g D (N)_D \ge g^D ( N ) D ≥ g D and D ≥ R + 1 D \ge R+1 D ≥ R + 1 , where ( N ) D = N ( N − 1 ) ⋯ ( N − D + 1 ) (N)_D=N(N-1)\cdots(N-D+1) ( N ) D = N ( N − 1 ) ⋯ ( N − D + 1 ) denotes the falling factorial.
For the deep tail we use the following factorial inequality. If N , J N,J N , J are integers with 1 ≤ J ≤ N / 2 1 \le J \le N/2 1 ≤ J ≤ N /2 , b ≥ 2 b \ge2 b ≥ 2 , and z 1 , … , z b z_1,\ldots,z_b z 1 , … , z b are nonnegative integers satisfying ∑ i z i ≤ N \sum_i z_i \le N ∑ i z i ≤ N and z i ≤ N − J z_i \le N-J z i ≤ N − J , then
∏ i z i ! ≤ ( N − J ) ! J ! . (29) \prod_i z_i! \le(N-J)!J!. \tag*{(29)} i ∏ z i ! ≤ ( N − J )! J ! . (29)
Indeed, first increase the coordinates until their sum is N N N , keeping each at most N − J N-J N − J ; this is possible since b ( N − J ) ≥ N b(N-J) \ge N b ( N − J ) ≥ N . Transferring one unit from a smaller positive coordinate y y y to a largest coordinate x < N − J x<N-J x < N − J multiplies the product by ( x + 1 ) / y ≥ 1 (x+1)/y \ge1 ( x + 1 ) / y ≥ 1 . Continue until that coordinate equals N − J N-J N − J ; the remaining mass is J J J and its factorial product is at most J ! J! J ! . If δ > J \delta>J δ > J , apply this to R i R_i R i . For b = 1 b=1 b = 1 the deep sector is empty, since its only possible deficit is m ≤ J m \le J m ≤ J . Otherwise the termwise bound is P b ( b + 1 ) n / ( N J ) P_b(b+1)^n/\binom{N}{J} P b ( b + 1 ) n / ( J N ) . Each g i ≤ g g_i \le g g i ≤ g , since b ≤ k − 2 σ b \le k-2\sigma b ≤ k − 2 σ , so there are at most ( N + 1 ) 2 b (N+1)^{2b} ( N + 1 ) 2 b complementary allocations. The bound N ≥ 128 s N \ge128s N ≥ 128 s gives J ≥ N / ( 128 s ) J \ge N/(128s) J ≥ N / ( 128 s ) . Hence, with α 0 = ( log 2 ) / 128 \alpha_0=(\log2)/128 α 0 = ( log 2 ) /128 ,
( N J ) ≥ ( N / J ) J ≥ 2 J ≥ e α 0 N / s . \binom{N}{J} \ge(N/J)^J \ge2^J \ge e^{\alpha_0N/s}. ( J N ) ≥ ( N / J ) J ≥ 2 J ≥ e α 0 N / s .
The deep contribution is therefore at most
P b ( b + 1 ) n ( N + 1 ) 2 b e − α 0 N / s ≤ P b g − R − 1 . (30) P_b(b+1)^n(N+1)^{2b}e^{-\alpha_0N/s} \le P_bg^{-R-1}. \tag*{(30)} P b ( b + 1 ) n ( N + 1 ) 2 b e − α 0 N / s ≤ P b g − R − 1 . (30)
To verify the last inequality uniformly in all parameters, note that the threshold gives b + 1 ≤ g b+1 \le g b + 1 ≤ g , N + 1 ≤ g 2 N+1 \le g^2 N + 1 ≤ g 2 and R + 1 + 4 b + n ≤ 8 t s R+1+4b+n \le8ts R + 1 + 4 b + n ≤ 8 t s . The logarithm to absorb is at most 8 t s log g 8ts\log g 8 t s log g . Since log g ≤ g \log g \le\sqrt{g} log g ≤ g and g ≥ ( K t s ) 4 ≥ ( 8 / α 0 ) t s 2 \sqrt{g} \ge(Kts)^4 \ge(8/\alpha_0)ts^2 g ≥ ( K t s ) 4 ≥ ( 8/ α 0 ) t s 2 , it is bounded by α 0 g / s ≤ α 0 N / s \alpha_0g/s \le\alpha_0N/s α 0 g / s ≤ α 0 N / s .
It remains to sum the estimates over gluing patterns. A partial matching is encoded by each slot’s partner or zero, giving at most ( k + 1 ) k (k+1)^k ( k + 1 ) k choices; a partition is encoded by its least block labels, giving at most k k k^k k k choices. Thus the number of patterns is at most ( 2 k + 2 ) 2 k + 2 ≤ ( 4 t ) 4 t (2k+2)^{2k+2} \le(4t)^{4t} ( 2 k + 2 ) 2 k + 2 ≤ ( 4 t ) 4 t . The bounds
P b ≤ ( C t ) 3 t , 2 b s D + 1 ( D + 3 ) 2 b D ! + 1 ≤ ( C t ) 5 t s 2 t P_b \le(Ct)^{3t},\qquad2b\,s^{D+1}(D+3)^{2b}D!+1 \le(Ct)^{5t}s^{2t} P b ≤ ( C t ) 3 t , 2 b s D + 1 ( D + 3 ) 2 b D ! + 1 ≤ ( C t ) 5 t s 2 t
follow from D ≤ t D \le t D ≤ t , b ≤ t + 1 b \le t+1 b ≤ t + 1 and the factorial bounds ( m − 1 ) ! ≤ t t − 2 (m-1)! \le t^{t-2} ( m − 1 )! ≤ t t − 2 , D ! ≤ t t D! \le t^t D ! ≤ t t . The sum of the bounds in (28) and (30) over all patterns is at most g − R − 1 ( C t s ) 12 t g^{-R-1}(Cts)^{12t} g − R − 1 ( C t s ) 12 t , which proves (19) after enlarging K K K .
For the boundary estimate, fix one pattern. Each gluing edge appears twice among the boundary lists of the numerator volumes. Their total boundary length is therefore
2 ∑ e ∈ E P y e = 2 Λ − 2 ∑ e ∈ M y e . 2\sum_{e\in E_P}y_e=2\Lambda-2\sum_{e\in M}y_e. 2 e ∈ E P ∑ y e = 2Λ − 2 e ∈ M ∑ y e .
Equation (14) contributes at most e Λ e^\Lambda e Λ , while J P ≤ ( 1 + Λ ) k J_P \le(1+\Lambda)^k J P ≤ ( 1 + Λ ) k . The zero-length tail for this pattern is nonnegative and at most the full scalar tail in (19) . This proves (21) on the pattern’s own edge-coordinate space, including zero lengths by continuity.
Joint intersection-number input and boundary summation
The tail estimate leaves configurations with bounded deficit and one large complementary component. We now expand its boundary volume. The source estimates control both the number of marked ends and the expansion order. Summing their intersection-number coefficients gives an expansion for every nonnegative length vector, without a length-dependent genus threshold.
For p ≥ 1 p \ge1 p ≥ 1 and d ∈ Z ≥ 0 p \mathbf{d} \in\mathbb{Z}_{\ge0}^{p} d ∈ Z ≥ 0 p , write ∣ d ∣ = ∑ i d i |\mathbf{d}| = \sum_i d_i ∣ d ∣ = ∑ i d i . We use normalized intersection numbers [ τ d ] G , p [\tau_{\mathbf{d}}]_{G,p} [ τ d ] G , p : the coefficient of ∏ i z i 2 d i \prod_i z_i^{2d_i} ∏ i z i 2 d i in V G , p ( 2 z ) V_{G,p}(2\mathbf{z}) V G , p ( 2 z ) is [ τ d ] G , p / ∏ i ( 2 d i + 1 ) ! [\tau_{\mathbf{d}}]_{G,p}/\prod_i(2d_i+1)! [ τ d ] G , p / ∏ i ( 2 d i + 1 )! . Thus [ τ 0 ] G , p = V G , p [\tau_{\mathbf{0}}]_{G,p}=V_{G,p} [ τ 0 ] G , p = V G , p , and the bracket is zero when ∣ d ∣ > 3 G + p − 3 |\mathbf{d}|>3G+p-3 ∣ d ∣ > 3 G + p − 3 . For an integer D ≥ 0 D \ge0 D ≥ 0 and a real v ≥ 0 v \ge0 v ≥ 0 , put E D ( v ) = ∑ q = 0 D v q / q ! E_D(v)=\sum_{q=0}^{D}v^q/q! E D ( v ) = ∑ q = 0 D v q / q ! . For G ≥ 2 G \ge2 G ≥ 2 and p ≥ 0 p \ge0 p ≥ 0 , define the scalar ratios
H G , p = A 0 ( 2 G − 2 + p ) V G , p V G , p + 1 , P G , p = V G − 1 , p + 2 V G , p . (31) H_{G,p}=A_0(2G-2+p)\frac{V_{G,p}}{V_{G,p+1}},\qquad P_{G,p}=\frac{V_{G-1,p+2}}{V_{G,p}}. \tag*{(31)} H G , p = A 0 ( 2 G − 2 + p ) V G , p + 1 V G , p , P G , p = V G , p V G − 1 , p + 2 . (31)
Proposition 2.3 (Joint source estimates). There is an absolute integer c 0 > 600 c_0>600 c 0 > 600 with the following properties. Put
D j ( p ) = 500 ( c 0 j ) c 0 j ( p + j + 1 ) j ( j ≥ 1 ) , G J ( p ) = 500 ( p + J + 2 ) ( c 0 ( J + 1 ) ) c 0 . D_j(p)=500(c_0j)^{c_0j}(p+j+1)^j\quad(j\ge1),\qquad G_J(p)=500(p+J+2)(c_0(J+1))^{c_0}. D j ( p ) = 500 ( c 0 j ) c 0 j ( p + j + 1 ) j ( j ≥ 1 ) , G J ( p ) = 500 ( p + J + 2 ) ( c 0 ( J + 1 ) ) c 0 . There are coefficients b d , p j b_{\mathbf{d},p}^j b d , p j , h p j h_p^j h p j , p p j p_p^j p p j , independent of genus and truncation order, with b d , p 0 = 1 b_{\mathbf{d},p}^0=1 b d , p 0 = 1 and constant term one in each scalar expansion, such that for every integer J ≥ 0 J\ge0 J ≥ 0 and integer G > G J ( p ) G>G_J(p) G > G J ( p ) ,
∣ [ τ d ] G , p V G , p − ∑ j = 0 J b d , p j G − j ∣ ≤ G − J − 1 D J + 1 ( p ) E 2 J + 2 ( ∣ d ∣ ) , p ≥ 1 , (32) \left|\frac{[\tau_{\mathbf{d}}]_{G,p}}{V_{G,p}}-\sum_{j=0}^{J}b_{\mathbf{d},p}^jG^{-j}\right|\le G^{-J-1}D_{J+1}(p)E_{2J+2}(|\mathbf{d}|),\qquad p\ge1, \tag*{(32)} V G , p [ τ d ] G , p − j = 0 ∑ J b d , p j G − j ≤ G − J − 1 D J + 1 ( p ) E 2 J + 2 ( ∣ d ∣ ) , p ≥ 1 , (32) ∣ H G , p − 1 − ∑ j = 1 J h p j G − j ∣ + ∣ P G , p − 1 − ∑ j = 1 J p p j G − j ∣ ≤ 2 G − J − 1 D J + 1 ( p ) , p ≥ 0. (33) \left|H_{G,p}-1-\sum_{j=1}^{J}h_p^jG^{-j}\right|+\left|P_{G,p}-1-\sum_{j=1}^{J}p_p^jG^{-j}\right|\le2G^{-J-1}D_{J+1}(p),\qquad p\ge0. \tag*{(33)} H G , p − 1 − j = 1 ∑ J h p j G − j + P G , p − 1 − j = 1 ∑ J p p j G − j ≤ 2 G − J − 1 D J + 1 ( p ) , p ≥ 0. (33) The first estimate is asserted only for ∣ d ∣ ≤ 3 G + p − 3 |\mathbf{d}|\le3G+p-3 ∣ d ∣ ≤ 3 G + p − 3 . The coefficients b d , p j b_{\mathbf{d},p}^j b d , p j and their bounds below are defined for every nonnegative multiindex, including those beyond the dimension cutoff at a given genus. For every integer j ≥ 1 j\ge1 j ≥ 1 ,
∣ b d , p j ∣ ≤ D j ( p ) E 2 j ( ∣ d ∣ ) , ∣ h p j ∣ , ∣ p p j ∣ ≤ D j ( p ) . (34) |b_{\mathbf{d},p}^j|\le D_j(p)E_{2j}(|\mathbf{d}|),\qquad|h_p^j|,|p_p^j|\le D_j(p). \tag*{(34)} ∣ b d , p j ∣ ≤ D j ( p ) E 2 j ( ∣ d ∣ ) , ∣ h p j ∣ , ∣ p p j ∣ ≤ D j ( p ) . (34) These are slightly weakened forms of [10 Theorem 4.2, A1, A3 and A4] , which refine the fixed-parameter expansions of Mirzakhani–Zograf [19 Theorem 4.1] . Choose c 0 c_0 c 0 larger than the constant required in [10 Theorem 4.2] . Then G J ( p ) G_J(p) G J ( p ) dominates all three source thresholds. The scalar majorants follow directly, and the multinomial identity ∑ ∣ α ∣ = q d α / α ! = ∣ d ∣ q / q ! \sum_{|\boldsymbol{\alpha}|=q}\mathbf{d}^{\boldsymbol{\alpha}}/\boldsymbol{\alpha}!=|\mathbf{d}|^q/q! ∑ ∣ α ∣ = q d α / α ! = ∣ d ∣ q / q ! sums the intersection-number majorants to the factors E D E_D E D above.
Mirzakhani’s volume-polynomial and intersection-number formula [Mir07b, Theorems 1.1 and 4.4], in the normalization recalled in [10 Theorem 4.4] , gives
V G , p ( 0 p − ℓ , x ) V G , p = ∑ ∣ d ∣ ≤ 3 G + p − 3 [ τ ( 0 , d ) ] G , p V G , p ∏ i = 1 ℓ ( x i / 2 ) 2 d i ( 2 d i + 1 ) ! . (35) \frac{V_{G,p}(\mathbf{0}_{p-\ell},\mathbf{x})}{V_{G,p}}=\sum_{|\mathbf{d}|\le3G+p-3}\frac{[\tau_{(\mathbf{0},\mathbf{d})}]_{G,p}}{V_{G,p}}\prod_{i=1}^{\ell}\frac{(x_i/2)^{2d_i}}{(2d_i+1)!}. \tag*{(35)} V G , p V G , p ( 0 p − ℓ , x ) = ∣ d ∣ ≤ 3 G + p − 3 ∑ V G , p [ τ ( 0 , d ) ] G , p i = 1 ∏ ℓ ( 2 d i + 1 )! ( x i /2 ) 2 d i . (35) The formula extends to nonnegative lengths by continuity. Throughout, we use the orbifold convention V 1 , 1 ( x ) = ( x 2 + 4 π 2 ) / 48 V_{1,1}(x)=(x^2+4\pi^2)/48 V 1 , 1 ( x ) = ( x 2 + 4 π 2 ) /48 of [5 §1] ; in particular, V 1 , 1 = π 2 / 12 V_{1,1}=\pi^2/12 V 1 , 1 = π 2 /12 . The convention in [10 §2.4] instead gives V 1 , 1 = π 2 / 6 V_{1,1}=\pi^2/6 V 1 , 1 = π 2 /6 . All applications of [10 Theorem 4.2] above have genera G , G − 1 ≥ 3 G,G-1\ge3 G , G − 1 ≥ 3 , where the generic stabilizer is trivial and the two conventions agree. Thus the imported normalized intersection numbers and scalar ratios are unchanged. Terminal components below always use our declared orbifold volumes. The estimates above imply, with one absolute C C C , the coarser bounds D j ( p ) ≤ [ C ( j + p + 1 ) ] C j D_j(p)\le[C(j+p+1)]^{Cj} D j ( p ) ≤ [ C ( j + p + 1 ) ] C j and G J ( p ) ≤ [ C ( J + 1 ) ( p + 1 ) ] C G_J(p)\le[C(J+1)(p+1)]^C G J ( p ) ≤ [ C ( J + 1 ) ( p + 1 ) ] C . These bounds hold uniformly in both the expansion order and the number of marked ends.
Lemma 2.4 (Boundary-volume expansion). For 0 ≤ ℓ ≤ p 0 \le\ell\le p 0 ≤ ℓ ≤ p , p ≥ 1 p \ge1 p ≥ 1 , there are continuous functions B j , p ( ℓ ) B^{(\ell)}_{j,p} B j , p ( ℓ ) on [ 0 , ∞ ) ℓ [0,\infty)^\ell [ 0 , ∞ ) ℓ , independent of genus and truncation, such that, writing X = ∑ i x i X=\sum_i x_i X = ∑ i x i ,
B 0 , p ( ℓ ) ( x ) = Φ ℓ ( x ) : = ∏ i = 1 ℓ sinh ( x i / 2 ) x i / 2 , (36) B^{(\ell)}_{0,p}(\mathbf{x})=\Phi_\ell(\mathbf{x}):=\prod_{i=1}^{\ell}\frac{\sinh(x_i/2)}{x_i/2}, \tag*{(36)} B 0 , p ( ℓ ) ( x ) = Φ ℓ ( x ) := i = 1 ∏ ℓ x i /2 sinh ( x i /2 ) , (36) ∣ B j , p ( ℓ ) ( x ) ∣ ≤ [ C j ( p + 1 ) ] C j ( 1 + X ) 2 j e X / 2 ( j ≥ 1 ) , (37) \left|B^{(\ell)}_{j,p}(\mathbf{x})\right|\le[Cj(p+1)]^{Cj}(1+X)^{2j}e^{X/2}\qquad(j\ge1), \tag*{(37)} B j , p ( ℓ ) ( x ) ≤ [ C j ( p + 1 ) ] C j ( 1 + X ) 2 j e X /2 ( j ≥ 1 ) , (37) ∣ V G , p ( 0 p − ℓ , x ) V G , p − ∑ j = 0 R G − j B j , p ( ℓ ) ( x ) ∣ ≤ G − R − 1 [ C ( R + 1 ) ( p + 1 ) ] C ( R + 1 ) ( 1 + X ) 2 R + 2 e X / 2 , (38) \left|\frac{V_{G,p}(\mathbf{0}_{p-\ell},\mathbf{x})}{V_{G,p}}-\sum_{j=0}^{R}G^{-j}B^{(\ell)}_{j,p}(\mathbf{x})\right|\le G^{-R-1}[C(R+1)(p+1)]^{C(R+1)}(1+X)^{2R+2}e^{X/2}, \tag*{(38)} V G , p V G , p ( 0 p − ℓ , x ) − j = 0 ∑ R G − j B j , p ( ℓ ) ( x ) ≤ G − R − 1 [ C ( R + 1 ) ( p + 1 ) ] C ( R + 1 ) ( 1 + X ) 2 R + 2 e X /2 , (38) for every R ≥ 0 R\ge0 R ≥ 0 and G ≥ [ C ( R + 1 ) ( p + 1 ) ] C G\ge[C(R+1)(p+1)]^C G ≥ [ C ( R + 1 ) ( p + 1 ) ] C , with no condition on X X X . Each factor sinh ( x i / 2 ) / ( x i / 2 ) \sinh(x_i/2)/(x_i/2) sinh ( x i /2 ) / ( x i /2 ) is interpreted as one at x i = 0 x_i=0 x i = 0 , and an empty product is one.
Proof. Write z i = x i / 2 z_i=x_i/2 z i = x i /2 , Z = ∑ i z i Z=\sum_i z_i Z = ∑ i z i and w d ( z ) = ∏ i z i 2 d i / ( 2 d i + 1 ) ! w_{\mathbf{d}}(\mathbf{z})=\prod_i z_i^{2d_i}/(2d_i+1)! w d ( z ) = ∏ i z i 2 d i / ( 2 d i + 1 )! . The dimension-free summation bound is
∑ d ≥ 0 w d ( z ) E D ( ∣ d ∣ ) ≤ ( D + 1 ) e Z ( 1 + Z ) D . (39) \sum_{\mathbf{d}\ge0}w_{\mathbf{d}}(\mathbf{z})E_D(|\mathbf{d}|)\le(D+1)e^Z(1+Z)^D. \tag*{(39)} d ≥ 0 ∑ w d ( z ) E D ( ∣ d ∣ ) ≤ ( D + 1 ) e Z ( 1 + Z ) D . (39) To prove it, replace ( 2 d i + 1 ) ! (2d_i+1)! ( 2 d i + 1 )! by ( 2 d i ) ! (2d_i)! ( 2 d i )! and enlarge the even indices 2 d i 2d_i 2 d i to all nonnegative indices. Since ∣ d ∣ |\mathbf{d}| ∣ d ∣ is at most their sum, the result is bounded by e Z E E D ( Y ) e^Z\mathbb{E}E_D(Y) e Z E E D ( Y ) for a Poisson variable Y Y Y of mean Z Z Z . Its factorial moments give E Y q = ∑ j = 0 q S ( q , j ) Z j \mathbb{E}Y^q=\sum_{j=0}^q S(q,j)Z^j E Y q = ∑ j = 0 q S ( q , j ) Z j , where S ( q , j ) S(q,j) S ( q , j ) is the Stirling number of the second kind. It counts partitions of a q q q -element set into j j j blocks and is at most q ! q! q ! (encode a partition by its canonically ordered cycles), so E Y q / q ! ≤ ∑ j = 0 q Z j ≤ ( 1 + Z ) q \mathbb{E}Y^q/q!\le\sum_{j=0}^q Z^j\le(1+Z)^q E Y q / q ! ≤ ∑ j = 0 q Z j ≤ ( 1 + Z ) q for q ≥ 1 q\ge1 q ≥ 1 . The case q = 0 q=0 q = 0 is immediate. Summing proves (39) .
Define for every j ≥ 0 j\ge0 j ≥ 0
B j , p ( ℓ ) ( x ) = ∑ d ≥ 0 b ( 0 , d ) , p j w d ( x / 2 ) . (40) B^{(\ell)}_{j,p}(\mathbf{x})=\sum_{\mathbf{d}\ge0}b^{j}_{(0,\mathbf{d}),p}w_{\mathbf{d}}(\mathbf{x}/2). \tag*{(40)} B j , p ( ℓ ) ( x ) = d ≥ 0 ∑ b ( 0 , d ) , p j w d ( x /2 ) . (40) The source coefficient bound and (39) prove local uniform absolute convergence, continuity and (37) ; b 0 = 1 b^0=1 b 0 = 1 gives (36) . For the remainder, first sum (32) over the finite volume polynomial (35) ; (39) gives the required bound for this part. The infinite series defining B j , p ( ℓ ) B^{(\ell)}_{j,p} B j , p ( ℓ ) also contain terms beyond that polynomial’s dimension cutoff. In this second part, v = ∣ d ∣ ≥ 3 G + p − 2 ≥ G v=|\mathbf{d}|\ge3G+p-2\ge G v = ∣ d ∣ ≥ 3 G + p − 2 ≥ G . For every 0 ≤ j ≤ R 0\le j\le R 0 ≤ j ≤ R ,
G R + 1 − j ≤ v R + 1 − j , v R + 1 − j E 2 j ( v ) ≤ ( 2 R + 2 ) ! E 2 R + 2 ( v ) . (41) G^{R+1-j}\le v^{R+1-j},\qquad v^{R+1-j}E_{2j}(v)\le(2R+2)!E_{2R+2}(v). \tag*{(41)} G R + 1 − j ≤ v R + 1 − j , v R + 1 − j E 2 j ( v ) ≤ ( 2 R + 2 )! E 2 R + 2 ( v ) . (41) The second inequality follows term by term: the polynomial on the left has degree at most R + j + 1 ≤ 2 R + 1 R+j+1\le2R+1 R + j + 1 ≤ 2 R + 1 , and the factorial quotient is at most ( 2 R + 2 ) ! (2R+2)! ( 2 R + 2 )! . For j = 0 j=0 j = 0 use b 0 = 1 b^0=1 b 0 = 1 and E 0 = 1 E_0=1 E 0 = 1 . Equation (39) then bounds the entire degree tail, including its constant-coefficient part. Absorbing ( 2 R + 2 ) ! (2R+2)! ( 2 R + 2 )! and the R + 1 R+1 R + 1 summands into the displayed order factor proves (38) .
Finite products and shifted ratios
To compare a large complementary component with the ambient surface, we must multiply finitely many expansions and change their genus parameter. We record the elementary estimates used for both operations.
Lemma 2.5 (Quantitative algebra). Let J ≥ 0 J\ge0 J ≥ 0 and M ≥ 1 M\ge1 M ≥ 1 be integers. Suppose M M M factors have expansions f ν ( g ) = 1 + ∑ j = 1 J a ν , j g − j + r ν ( g ) f_\nu(g)=1+\sum_{j=1}^{J}a_{\nu,j}g^{-j}+r_\nu(g) f ν ( g ) = 1 + ∑ j = 1 J a ν , j g − j + r ν ( g ) with ∣ a ν , j ∣ ≤ A j |a_{\nu,j}|\le A^j ∣ a ν , j ∣ ≤ A j and ∣ r ν ( g ) ∣ ≤ A J + 1 g − J − 1 |r_\nu(g)|\le A^{J+1}g^{-J-1} ∣ r ν ( g ) ∣ ≤ A J + 1 g − J − 1 , where A ≥ 1 A\ge1 A ≥ 1 . At values of g g g for which these expansions are valid and g ≥ 4 M A g\ge4MA g ≥ 4 M A , their product has coefficients bounded in order j j j by ( M A ) j (MA)^j ( M A ) j and remainder bounded by C ( M A ) J + 1 g − J − 1 C(MA)^{J+1}g^{-J-1} C ( M A ) J + 1 g − J − 1 .
Separately, a single expansion in ( g − H ) − 1 (g-H)^{-1} ( g − H ) − 1 , H ≥ 0 H \ge0 H ≥ 0 , with coefficient bounds A j A^j A j and remainder bound A J + 1 ( g − H ) − J − 1 A^{J+1}(g-H)^{-J-1} A J + 1 ( g − H ) − J − 1 can be rewritten in g − 1 g^{-1} g − 1 with A A A replaced by C ( A + H + 1 ) C(A+H+1) C ( A + H + 1 ) , provided g ≥ 4 ( H + 1 ) g \ge4(H+1) g ≥ 4 ( H + 1 ) and the original expansion is valid at g − H g-H g − H . If the input coefficients are consistent across truncation orders, both operations preserve this consistency.
Proof. For each total degree j ≥ 1 j \ge1 j ≥ 1 , the number of weak compositions into M M M parts is at most M j M^j M j , since every such composition is the multiplicity vector of a word of length j j j on M M M symbols. This proves the coefficient bound and bounds the discarded polynomial product by 2 ( M A / g ) J + 1 2(MA/g)^{J+1} 2 ( M A / g ) J + 1 . Each truncated factor has modulus at most 1 + 2 A / g 1+2A/g 1 + 2 A / g , and each exact factor has modulus at most 1 + 3 A / g 1+3A/g 1 + 3 A / g . Replacing factors one at a time bounds the sum of remainder contributions by M A J + 1 g − J − 1 e 3 M A / g MA^{J+1}g^{-J-1}e^{3MA/g} M A J + 1 g − J − 1 e 3 M A / g , as required.
For shifting, use, for j ≥ 1 j \ge1 j ≥ 1 ,
( g − H ) − j = g − j ∑ v ≥ 0 ( j + v − 1 v ) ( H / g ) v . (42) (g-H)^{-j}=g^{-j}\sum_{v\ge0}\binom{j+v-1}{v}(H/g)^v. \tag*{(42)} ( g − H ) − j = g − j v ≥ 0 ∑ ( v j + v − 1 ) ( H / g ) v . (42)
The coefficient of g − q g^{-q} g − q is ∑ j = 1 q a j ( q − 1 j − 1 ) H q − j \sum_{j=1}^{q}a_j\binom{q-1}{j-1}H^{q-j} ∑ j = 1 q a j ( j − 1 q − 1 ) H q − j , bounded by [ 2 ( A + H + 1 ) ] q [2(A+H+1)]^q [ 2 ( A + H + 1 ) ] q . With H / g ≤ 1 / 4 H/g \le1/4 H / g ≤ 1/4 , use ( j + v − 1 v ) ≤ 2 j + v \binom{j+v-1}{v}\le2^{j+v} ( v j + v − 1 ) ≤ 2 j + v and sum the geometric tail for v ≥ J + 1 − j v \ge J+1-j v ≥ J + 1 − j . Summing over j ≤ J j \le J j ≤ J gives g − J − 1 [ C ( A + H + 1 ) ] J + 1 g^{-J-1}[C(A+H+1)]^{J+1} g − J − 1 [ C ( A + H + 1 ) ] J + 1 . The original remainder costs at most 2 J + 1 A J + 1 g − J − 1 2^{J+1}A^{J+1}g^{-J-1} 2 J + 1 A J + 1 g − J − 1 . The coefficient formulas involve only orders up to q q q , proving consistency. □
Let n n n , H ≥ 0 H \ge0 H ≥ 0 and δ ≥ 1 \delta\ge1 δ ≥ 1 be integers, and set G = g − H G=g-H G = g − H and p = n + 2 H − δ ≥ 1 p=n+2H-\delta\ge1 p = n + 2 H − δ ≥ 1 . The shifted ratio can be written entirely in terms of the scalar ratios (31) . Telescoping gives
V g − H , p V g , n = ( 8 π 2 g ) − δ ∏ j = 0 δ − 1 H G , p + j ∏ u = 0 H − 1 P G + u + 1 , n + 2 H − 2 u − 2 ∏ j = 0 δ − 1 ( 1 + n − δ + j − 2 2 g ) − 1 . (43) \frac{V_{g-H,p}}{V_{g,n}}=(8\pi^2g)^{-\delta}\prod_{j=0}^{\delta-1}H_{G,p+j}\prod_{u=0}^{H-1}P_{G+u+1,n+2H-2u-2}\prod_{j=0}^{\delta-1}\left(1+\frac{n-\delta+j-2}{2g}\right)^{-1}. \tag*{(43)} V g , n V g − H , p = ( 8 π 2 g ) − δ j = 0 ∏ δ − 1 H G , p + j u = 0 ∏ H − 1 P G + u + 1 , n + 2 H − 2 u − 2 j = 0 ∏ δ − 1 ( 1 + 2 g n − δ + j − 2 ) − 1 . (43)
Indeed the prefactor together with the first and last products gives V G , p / V G , p + δ V_{G,p}/V_{G,p+\delta} V G , p / V G , p + δ , and the middle product gives V G , n + 2 H / V g , n V_{G,n+2H}/V_{g,n} V G , n + 2 H / V g , n . All marked-point indices in the middle product are nonnegative; all genera along this path are large under the thresholds below.
To construct the expansion through order J J J , put L 0 = J + H + δ + 1 L_0=J+H+\delta+1 L 0 = J + H + δ + 1 . Proposition 2.3, genus shifts and the geometric expansions of the last factors in (2.32) have a common coefficient majorant A = [ C 1 L 0 ( n + 1 ) ] C 1 A=[C_1L_0(n+1)]^{C_1} A = [ C 1 L 0 ( n + 1 ) ] C 1 , valid when g ≥ A g\ge A g ≥ A , for an absolute C 1 C_1 C 1 . There are M = H + 2 δ ≤ 2 L 0 M=H+2\delta\le2L_0 M = H + 2 δ ≤ 2 L 0 normalized factors. Choosing the absolute constant C C C in the genus threshold g ≥ [ C L 0 ( n + 1 ) ] C g\ge[CL_0(n+1)]^C g ≥ [ C L 0 ( n + 1 ) ] C sufficiently large ensures g ≥ 4 M A g\ge4MA g ≥ 4 M A . Lemma 2.5 gives fixed formal coefficients v j ( n , H , δ ) v_j(n,H,\delta) v j ( n , H , δ ) , v 0 = 1 v_0=1 v 0 = 1 , such that
∣ V g − H , p V g , n − ( 8 π 2 ) − δ g − δ ∑ j = 0 J v j ( n , H , δ ) g − j ∣ ≤ g − δ − J − 1 [ C L 0 ( n + 1 ) ] C ( J + 1 ) . (44) \left|\frac{V_{g-H,p}}{V_{g,n}}-(8\pi^2)^{-\delta}g^{-\delta}\sum_{j=0}^{J}v_j(n,H,\delta)g^{-j}\right|\le g^{-\delta-J-1}[CL_0(n+1)]^{C(J+1)}. \tag*{(44)} V g , n V g − H , p − ( 8 π 2 ) − δ g − δ j = 0 ∑ J v j ( n , H , δ ) g − j ≤ g − δ − J − 1 [ C L 0 ( n + 1 ) ] C ( J + 1 ) . (44)
For the coefficient of order j j j , one can take the majorant [ C ( j + H + δ + 1 ) ( n + 1 ) ] C j [C(j+H+\delta+1)(n+1)]^{Cj} [ C ( j + H + δ + 1 ) ( n + 1 ) ] C j .
Lemma 2.6 (Shifted boundary ratio). For the preceding n n n , H H H , δ \delta δ , p p p and 0 ≤ ℓ ≤ p 0\le\ell\le p 0 ≤ ℓ ≤ p there are continuous functions Q j n , H , δ ; ℓ Q_j^{n,H,\delta;\ell} Q j n , H , δ ; ℓ , j ≥ 0 j\ge0 j ≥ 0 , on [ 0 , ∞ ) [0,\infty) [ 0 , ∞ ) , independent of genus and truncation order. With X = ∑ i x i X=\sum_i x_i X = ∑ i x i , they satisfy Q 0 = Φ ℓ Q_0=\Phi_\ell Q 0 = Φ ℓ and
∣ Q j n , H , δ ; ℓ ( x ) ∣ ≤ [ C ( j + H + δ + 1 ) ( n + 1 ) ] C ( j + 1 ) ( 1 + X ) C ( j + 1 ) e X / 2 . (45) |Q_j^{n,H,\delta;\ell}(\mathbf{x})|\le[C(j+H+\delta+1)(n+1)]^{C(j+1)}(1+X)^{C(j+1)}e^{X/2}. \tag*{(45)} ∣ Q j n , H , δ ; ℓ ( x ) ∣ ≤ [ C ( j + H + δ + 1 ) ( n + 1 ) ] C ( j + 1 ) ( 1 + X ) C ( j + 1 ) e X /2 . (45) For every R ≥ δ R\ge\delta R ≥ δ , L 0 = R + H + δ + 1 L_0=R+H+\delta+1 L 0 = R + H + δ + 1 and g ≥ [ C L 0 ( n + 1 ) ] C g\ge[CL_0(n+1)]^C g ≥ [ C L 0 ( n + 1 ) ] C ,
V g − H , p ( 0 p − ℓ , x ) V g , n = ( 8 π 2 ) − δ g − δ ∑ j = 0 R − δ Q j n , H , δ ; ℓ ( x ) g − j + ε R , (46) \frac{V_{g-H,p}(\mathbf{0}_{p-\ell},\mathbf{x})}{V_{g,n}}=(8\pi^2)^{-\delta}g^{-\delta}\sum_{j=0}^{R-\delta}Q_j^{n,H,\delta;\ell}(\mathbf{x})g^{-j}+\varepsilon_R, \tag*{(46)} V g , n V g − H , p ( 0 p − ℓ , x ) = ( 8 π 2 ) − δ g − δ j = 0 ∑ R − δ Q j n , H , δ ; ℓ ( x ) g − j + ε R , (46) ∣ ε R ∣ ≤ g − R − 1 [ C L 0 ( n + 1 ) ] C ( R + 1 ) ( 1 + X ) C ( R + 1 ) e X / 2 . (47) |\varepsilon_R|\le g^{-R-1}[CL_0(n+1)]^{C(R+1)}(1+X)^{C(R+1)}e^{X/2}. \tag*{(47)} ∣ ε R ∣ ≤ g − R − 1 [ C L 0 ( n + 1 ) ] C ( R + 1 ) ( 1 + X ) C ( R + 1 ) e X /2 . (47)
Proof. Shift the boundary coefficients of Lemma 2.4: B ~ 0 = B 0 , p ( ℓ ) \widetilde{B}_0 = B^{(\ell)}_{0,p} B 0 = B 0 , p ( ℓ ) and, for q ≥ 1 q \ge1 q ≥ 1 ,
B ~ q = ∑ j = 1 q ( q − 1 j − 1 ) H q − j B j , p ( ℓ ) . \widetilde{B}_q = \sum_{j=1}^{q} \binom{q-1}{j-1} H^{q-j} B^{(\ell)}_{j,p}. B q = j = 1 ∑ q ( j − 1 q − 1 ) H q − j B j , p ( ℓ ) .
Define Q j = ∑ u + v = j v u ( n , H , δ ) B ~ v Q_j = \sum_{u+v=j} v_u(n,H,\delta)\widetilde{B}_v Q j = ∑ u + v = j v u ( n , H , δ ) B v . These finite convolution formulas prove consistency and (45) . For the remainder put J = R − δ J=R-\delta J = R − δ and G = g − H G=g-H G = g − H . Write
Z ( g ) = ( 8 π 2 ) δ g δ V G , p V g , n , S J ( g ) = ∑ u = 0 J v u ( n , H , δ ) g − u , Z(g)=(8\pi^2)^\delta g^\delta\frac{V_{G,p}}{V_{g,n}}, \qquad S_J(g)=\sum_{u=0}^{J}v_u(n,H,\delta)g^{-u}, Z ( g ) = ( 8 π 2 ) δ g δ V g , n V G , p , S J ( g ) = u = 0 ∑ J v u ( n , H , δ ) g − u ,
D ( g , x ) = V G , p ( 0 p − ℓ , x ) V G , p , D J ( g , x ) = ∑ v = 0 J B ~ v ( x ) g − v . D(g,\mathbf{x})=\frac{V_{G,p}(\mathbf{0}_{p-\ell},\mathbf{x})}{V_{G,p}}, \qquad D_J(g,\mathbf{x})=\sum_{v=0}^{J}\widetilde{B}_v(\mathbf{x})g^{-v}. D ( g , x ) = V G , p V G , p ( 0 p − ℓ , x ) , D J ( g , x ) = v = 0 ∑ J B v ( x ) g − v .
Enlarging the constant in the genus threshold gives ∣ S J ( g ) ∣ ≤ 2 |S_J(g)|\le2 ∣ S J ( g ) ∣ ≤ 2 ; (14) gives ∣ D ( g , x ) ∣ ≤ e X / 2 |D(g,\mathbf{x})|\le e^{X/2} ∣ D ( g , x ) ∣ ≤ e X /2 . Use the exact identity
Z D − S J D J = ( Z − S J ) D + S J ( D − D J ) . ZD-S_JD_J=(Z-S_J)D+S_J(D-D_J). Z D − S J D J = ( Z − S J ) D + S J ( D − D J ) .
After multiplication by ( 8 π 2 ) − δ g − δ (8\pi^2)^{-\delta}g^{-\delta} ( 8 π 2 ) − δ g − δ , the first term is controlled by (44) and the second by the shifted boundary remainder. Each bound contains at most one factor e X / 2 e^{X/2} e X /2 . In the product S J D J S_JD_J S J D J , only the boundary factor depends on x \mathbf{x} x , with length growth at most e X / 2 ( 1 + X ) 2 J e^{X/2}(1+X)^{2J} e X /2 ( 1 + X ) 2 J . Extracting this common factor from the discarded terms of total degree greater than J J J leaves scalar coefficient majorants, to which the geometric estimate in Lemma 2.5 applies. This bounds all discarded terms by (47) , without a length-dependent genus threshold.
Canonical retained kernels
We now combine the complementary-volume tail with the shifted boundary expansion. For a fixed carrier type and pattern P P P , let x = ( x j ( y ) ) j = 1 k \mathbf{x}=(x_j(\mathbf{y}))_{j=1}^{k} x = ( x j ( y ) ) j = 1 k be the slot lengths induced by the edge lengths y \mathbf{y} y . The full nonnegative kernel is
F g , n h 0 , a 0 , k ; P ( y ) = J P ( y ) V h 0 , a 0 + k ( 0 a 0 , x ) ∑ n ! a 0 ! ∏ i a i ! ∏ i V g i , a i + k i ( 0 a i , x I i ) V g , n . (48) F^{h_0,a_0,k;P}_{g,n}(\mathbf{y})=J_P(\mathbf{y})V_{h_0,a_0+k}(\mathbf{0}_{a_0},\mathbf{x})\sum\frac{n!}{a_0!\prod_i a_i!}\prod_i\frac{V_{g_i,a_i+k_i}(\mathbf{0}_{a_i},\mathbf{x}_{I_i})}{V_{g,n}}. \tag*{(48)} F g , n h 0 , a 0 , k ; P ( y ) = J P ( y ) V h 0 , a 0 + k ( 0 a 0 , x ) ∑ a 0 ! ∏ i a i ! n ! i ∏ V g , n V g i , a i + k i ( 0 a i , x I i ) . (48)
Theorem 2.7 (Retained boundary kernels). There is an absolute C ≥ 2 C\ge2 C ≥ 2 and, for every fixed n n n , h 0 h_0 h 0 , a 0 a_0 a 0 , k k k , P P P , a sequence of continuous functions F j , n h 0 , a 0 , k ; P F^{h_0,a_0,k;P}_{j,n} F j , n h 0 , a 0 , k ; P , j ≥ 1 j\ge1 j ≥ 1 , independent of genus and truncation order, such that, simultaneously for all R ≥ 0 R\ge0 R ≥ 0 , t = R + m + 1 t=R+m+1 t = R + m + 1 , and y ≥ 0 \mathbf{y}\ge0 y ≥ 0 ,
∣ F g , n h 0 , a 0 , k ; P ( y ) − ∑ j = 1 R g − j F j , n h 0 , a 0 , k ; P ( y ) ∣ ≤ g − R − 1 [ C t ( n + 1 ) ] C t ( 1 + Λ ) C t e Λ (49) \left|F^{h_0,a_0,k;P}_{g,n}(\mathbf{y})-\sum_{j=1}^{R}g^{-j}F^{h_0,a_0,k;P}_{j,n}(\mathbf{y})\right|\le g^{-R-1}[C t(n+1)]^{Ct}(1+\Lambda)^{Ct}e^{\Lambda} \tag*{(49)} F g , n h 0 , a 0 , k ; P ( y ) − j = 1 ∑ R g − j F j , n h 0 , a 0 , k ; P ( y ) ≤ g − R − 1 [ C t ( n + 1 ) ] C t ( 1 + Λ ) C t e Λ (49) when g ≥ [ C t ( n + 1 ) ] C g\ge[C t(n+1)]^C g ≥ [ C t ( n + 1 ) ] C . Moreover, F j , n h 0 , a 0 , k ; P = 0 F^{h_0,a_0,k;P}_{j,n}=0 F j , n h 0 , a 0 , k ; P = 0 for j < m + b − 1 j<m+b-1 j < m + b − 1 . For every j ≥ 1 j\ge1 j ≥ 1 ,
∣ F j , n h 0 , a 0 , k ; P ( y ) ∣ ≤ [ C ( j + m + 1 ) ( n + 1 ) ] C ( j + m + 1 ) ( 1 + Λ ) C ( j + m + 1 ) e Λ . (50) |F^{h_0,a_0,k;P}_{j,n}(\mathbf{y})|\le[C(j+m+1)(n+1)]^{C(j+m+1)}(1+\Lambda)^{C(j+m+1)}e^{\Lambda}. \tag*{(50)} ∣ F j , n h 0 , a 0 , k ; P ( y ) ∣ ≤ [ C ( j + m + 1 ) ( n + 1 ) ] C ( j + m + 1 ) ( 1 + Λ ) C ( j + m + 1 ) e Λ . (50) Proof. Fix the carrier type and pattern. To define coefficients independently of genus, we record all complementary components except the one that will become large. A record consists of a distinguished block ∗ * ∗ and nonnegative integers ( g i , a i ) i ≠ ∗ (g_i,a_i)_{i\ne*} ( g i , a i ) i = ∗ for which the corresponding
components are stable. The distinguished component will be reconstructed from the ambient genus and the sum constraints in (16) . For such a record set
δ = m + ∑ i ≠ ∗ c i , A o u t = a 0 + ∑ i ≠ ∗ a i , H = h 0 + k − σ − b + ∑ i ≠ ∗ g i , p = n − A o u t + k ∗ . (51) \begin{aligned}
\delta&= m+\sum_{i\ne*}c_i, & A_{\mathrm{out}}&=a_0+\sum_{i\ne*}a_i,\\
H&=h_0+k-\sigma-b+\sum_{i\ne*}g_i, & p&=n-A_{\mathrm{out}}+k_*. \tag*{(51)}
\end{aligned} δ H = m + i = ∗ ∑ c i , = h 0 + k − σ − b + i = ∗ ∑ g i , A out p = a 0 + i = ∗ ∑ a i , = n − A out + k ∗ . (51)
For each integer δ ≥ 1 \delta\ge1 δ ≥ 1 , let E δ \mathcal{E}_\delta E δ be the set of records with this deficit and A o u t ≤ n A_{\mathrm{out}}\le n A out ≤ n . These sets are independent of genus and truncation order.
Now fix R ≥ 0 R\ge0 R ≥ 0 . Theorem 2.2 controls the configurations with deficit at least R + 1 R+1 R + 1 . For a record of deficit δ ≤ R \delta\le R δ ≤ R , direct substitution gives
2 H + A o u t − k ∗ = δ , 0 ≤ H , A o u t ≤ δ + 1 , 2 H ≤ δ + k ∗ ≤ R + m + 2 , p ≤ n + 2 t . (52) 2H+A_{\mathrm{out}}-k_*=\delta,\qquad0\le H,\qquad A_{\mathrm{out}}\le\delta+1,\qquad2H\le\delta+k_*\le R+m+2,\qquad p\le n+2t. \tag*{(52)} 2 H + A out − k ∗ = δ , 0 ≤ H , A out ≤ δ + 1 , 2 H ≤ δ + k ∗ ≤ R + m + 2 , p ≤ n + 2 t . (52)
The distinguished component therefore has genus G = g − H G=g-H G = g − H and p = n − A o u t + k ∗ p=n-A_{\mathrm{out}}+k_* p = n − A out + k ∗ marked ends: n − A o u t n-A_{\mathrm{out}} n − A out cusps and k ∗ k_* k ∗ gluing boundaries. Also, p = n + 2 H − δ ≥ k ∗ ≥ 1 p=n+2H-\delta\ge k_*\ge1 p = n + 2 H − δ ≥ k ∗ ≥ 1 , as required by Lemma 2.6. The common threshold ensures g > H g>H g > H and N > 2 R N>2R N > 2 R , so each such record reconstructs an admissible large component. Its factorial index is 2 ( g − H ) + p − 3 = N − δ 2(g-H)+p-3=N-\delta 2 ( g − H ) + p − 3 = N − δ . Since δ ≤ R \delta\le R δ ≤ R , this exceeds δ − m \delta-m δ − m , the total complexity of the other complementary components, and makes it the unique largest component. Conversely, any admissible configuration with deficit at most R R R has a unique largest component and determines one of these records. Moreover,
E δ = ∅ ( δ < m + b − 1 ) , ∣ E δ ∣ ≤ b ( δ + 3 ) 2 b − 1 , (53) \mathcal{E}_\delta=\varnothing\quad(\delta<m+b-1),\qquad|\mathcal{E}_\delta|\le b(\delta+3)2^{b-1}, \tag*{(53)} E δ = ∅ ( δ < m + b − 1 ) , ∣ E δ ∣ ≤ b ( δ + 3 ) 2 b − 1 , (53)
by the same independent choices of small indices and cusp counts used in (26) .
For a record e e e , let X ∗ = ∑ j ∈ I ∗ x j X_*=\sum_{j\in I_*}x_j X ∗ = ∑ j ∈ I ∗ x j and put
A e ( y ) = J P V h 0 , a 0 + k ( 0 a 0 , x ) n ! a 0 ! ( n − A o u t ) ! ∏ i ≠ ∗ a i ! ∏ i ≠ ∗ V g i , a i + k i ( 0 a i , X I i ) . (54) A_e(\mathbf{y})=J_PV_{h_0,a_0+k}(\mathbf{0}_{a_0},\mathbf{x})\frac{n!}{a_0!(n-A_{\mathrm{out}})!}\prod_{i\ne*}a_i!\prod_{i\ne*}V_{g_i,a_i+k_i}(\mathbf{0}_{a_i},X_{I_i}). \tag*{(54)} A e ( y ) = J P V h 0 , a 0 + k ( 0 a 0 , x ) a 0 ! ( n − A out )! n ! i = ∗ ∏ a i ! i = ∗ ∏ V g i , a i + k i ( 0 a i , X I i ) . (54)
The carrier and small components have total complexity δ \delta δ ; their factorial indices sum to δ − b \delta-b δ − b . Lemma 2.1 and the product-factorial inequality bound their zero-length product by U 0 b A 0 δ − b ( δ − b ) ! U_0^bA_0^{\delta-b}(\delta-b)! U 0 b A 0 δ − b ( δ − b )! . The label factor is at most ( n + 1 ) δ + 1 (n+1)^{\delta+1} ( n + 1 ) δ + 1 . Their total boundary length is at most 2 Λ − X ∗ 2\Lambda-X_* 2Λ − X ∗ , including self-pairs only in the carrier. Hence
∣ A e ( y ) ∣ ≤ [ C ( δ + 1 ) ( n + 1 ) ] C ( δ + 1 ) ( 1 + Λ ) k e Λ − X ∗ 2 . (55) |A_e(\mathbf{y})|\le[C(\delta+1)(n+1)]^{C(\delta+1)}(1+\Lambda)^{k_e}\Lambda^{-\frac{X_*}{2}}. \tag*{(55)} ∣ A e ( y ) ∣ ≤ [ C ( δ + 1 ) ( n + 1 ) ] C ( δ + 1 ) ( 1 + Λ ) k e Λ − 2 X ∗ . (55)
Using these record sets, define the canonical coefficient sequence:
F j , n h 0 , a 0 , k ; P ( y ) = ∑ δ = 1 j ∑ e ∈ E δ ( 8 π 2 ) − δ A e ( y ) Q j − δ n , H e , δ ; k ∗ , e ( x I ∗ , e ) . (56) \mathcal{F}^{h_0,a_0,k;P}_{j,n}(\mathbf{y})=\sum_{\delta=1}^{j}\sum_{e\in\mathcal{E}_\delta}(8\pi^2)^{-\delta}A_e(\mathbf{y})Q^{n,H_e,\delta;k_*,e}_{j-\delta}(\mathbf{x}_{I_{*,e}}). \tag*{(56)} F j , n h 0 , a 0 , k ; P ( y ) = δ = 1 ∑ j e ∈ E δ ∑ ( 8 π 2 ) − δ A e ( y ) Q j − δ n , H e , δ ; k ∗ , e ( x I ∗ , e ) . (56)
These sums are finite and continuous; (53) proves the stated vanishing. For every retained record apply Lemma 2.6 through order R R R . Its L 0 = R + H e + δ + 1 ≤ 3 t L_0=R+H_e+\delta+1\le3t L 0 = R + H e + δ + 1 ≤ 3 t , so the same threshold g ≥ [ C t ( n + 1 ) ] C g\ge[Ct(n+1)]^C g ≥ [ C t ( n + 1 ) ] C suffices for all records. Multiplication by (55) changes e X ∗ / 2 e^{X_*/2} e X ∗ /2 to e Λ e^\Lambda e Λ . Since k ≤ m + 2 ≤ t + 1 k\le m+2\le t+1 k ≤ m + 2 ≤ t + 1 , all length powers are bounded by ( 1 + Λ ) C t (1+\Lambda)^{Ct} ( 1 + Λ ) C t . The number of records is at most ( C t ) C t (Ct)^{Ct} ( C t ) C t by (53) . Summing their errors and adding (21) proves (49) . If R < m + b − 1 R<m+b-1 R < m + b − 1 there are no retained records and the tail alone proves it, including R = 0 R=0 R = 0 . The same finite sum, (45) and (55) with R = j R=j R = j give (50) .
At the first possible order d = m + b − 1 d=m+b-1 d = m + b − 1 , the coefficient is
F d , n h 0 , a 0 , k ; P ( y ) = ( 8 π 2 ) − d ∑ e ∈ E d A e ( y ) ∏ q ∈ I ∗ , e sinh ( x q / 2 ) x q / 2 . (57) \mathcal{F}^{h_0,a_0,k;P}_{d,n}(\mathbf{y})=(8\pi^2)^{-d}\sum_{e\in\mathcal{E}_d}A_e(\mathbf{y})\prod_{q\in I_{*,e}}\frac{\sinh(x_q/2)}{x_q/2}. \tag*{(57)} F d , n h 0 , a 0 , k ; P ( y ) = ( 8 π 2 ) − d e ∈ E d ∑ A e ( y ) q ∈ I ∗ , e ∏ x q /2 sinh ( x q /2 ) . (57)
Each small component contributing here has complexity one, hence type ( 0 , 3 ) (0,3) ( 0 , 3 ) or ( 1 , 1 ) (1,1) ( 1 , 1 ) , where the marked ends may be cusps or gluing boundaries. In particular, terminal pairs of pants already contribute at this leading order. With a single complementary component, d = m d=m d = m . § 3 treats the unstable cylinder directly, without assigning a volume to ( 0 , 2 ) (0,2) ( 0 , 2 ) .
Orbital integration and cusp-polynomial coefficients
Throughout this section g ≥ 3 g \ge3 g ≥ 3 and n ≥ 0 n \ge0 n ≥ 0 are integers, and every cusp is labelled. For a real bounded even Borel function H H H , supported in [ − T , T ] [-T,T] [ − T , T ] with T ≥ 0 T \ge0 T ≥ 0 , define
O H ( X ) = ∑ γ ∈ P o r ( X ) ∑ ν ≥ 1 ℓ γ H ( ν ℓ γ ) 2 sinh ( ν ℓ γ / 2 ) . (58) \mathcal{O}_{H}(X)=\sum_{\gamma\in\mathcal{P}_{\mathrm{or}}(X)}\sum_{\nu\ge1}\frac{\ell_{\gamma}H(\nu\ell_{\gamma})}{2\sinh(\nu\ell_{\gamma}/2)}. \tag*{(58)} O H ( X ) = γ ∈ P or ( X ) ∑ ν ≥ 1 ∑ 2 sinh ( ν ℓ γ /2 ) ℓ γ H ( ν ℓ γ ) . (58)
Here P o r \mathcal{P}_{\mathrm{or}} P or denotes primitive oriented geodesics; thus both orientations and all positive repetitions occur. The primitive underlying geodesic determines the simple and nonsimple parts of this statistic. We expand the expectation in powers of g − 1 g^{-1} g − 1 and determine the dependence of its coefficients on n n n . We first treat simple geodesics by cutting along them, then treat nonsimple geodesics by integrating over the subsurfaces they fill. Combining these exact formulas with § 2 gives the canonical coefficients in Theorem 3.5 and the geometric moments in Proposition 3.6 . § 3.7 and § 3.8 compute the first correction and prove that the coefficient of g − j g^{-j} g − j has degree at most j + 1 j+1 j + 1 in n n n . This polynomiality concerns the coefficients; the sharper uniform remainder needed when n n n grows with g g g is proved in § 4 .
The two simple-curve kernels
Put ϕ ( x ) = sinh ( x / 2 ) / ( x / 2 ) \phi(x)=\sinh(x/2)/(x/2) ϕ ( x ) = sinh ( x /2 ) / ( x /2 ) , with ϕ ( 0 ) = 1 \phi(0)=1 ϕ ( 0 ) = 1 , and define
U g , n ( ℓ ) = V g − 1 , n + 2 ( 0 n , ℓ , ℓ ) V g , n . (59) U_{g,n}(\ell)=\frac{V_{g-1,n+2}(0_n,\ell,\ell)}{V_{g,n}}. \tag*{(59)} U g , n ( ℓ ) = V g , n V g − 1 , n + 2 ( 0 n , ℓ , ℓ ) . (59)
S g , n ( ℓ ) = ∑ 0 ≤ i ≤ g , 0 ≤ a 0 ≤ n 2 i + a 0 ≥ 2 , 2 ( g − i ) + n − a 0 ≥ 2 ( n a 0 ) V i , a 0 + 1 ( 0 a 0 , ℓ ) V g − i , n − a 0 + 1 ( 0 n − a 0 , ℓ ) V g , n . (60) S_{g,n}(\ell)=\sum_{\substack{0\le i\le g,\ 0\le a_0\le n\\ 2i+a_0\ge2,\ 2(g-i)+n-a_0\ge2}}\binom{n}{a_0}\frac{V_{i,a_0+1}(0_{a_0},\ell)V_{g-i,n-a_0+1}(0_{n-a_0},\ell)}{V_{g,n}}. \tag*{(60)} S g , n ( ℓ ) = 0 ≤ i ≤ g , 0 ≤ a 0 ≤ n 2 i + a 0 ≥ 2 , 2 ( g − i ) + n − a 0 ≥ 2 ∑ ( a 0 n ) V g , n V i , a 0 + 1 ( 0 a 0 , ℓ ) V g − i , n − a 0 + 1 ( 0 n − a 0 , ℓ ) . (60)
The kernel U g , n U_{g,n} U g , n corresponds to a nonseparating cut, and S g , n S_{g,n} S g , n to a separating cut with a distinguished side. Thus the separating sum is ordered; its factor ( n a 0 ) \binom{n}{a_0} ( a 0 n ) chooses the cusp labels on that side. All volume factors are stable. The orientation and side-counting factors are verified in Lemma 3.2 .
Lemma 3.1 (Simple-kernel expansion). There is an absolute C ≥ 2 C\ge2 C ≥ 2 and, for each n n n , a fixed sequence of continuous functions u j , n , s j , n u_{j,n},s_{j,n} u j , n , s j , n such that, for every integer R ≥ 0 R\ge0 R ≥ 0 and r = R + 1 r=R+1 r = R + 1 ,
∣ U g , n ( ℓ ) − ϕ ( ℓ ) 2 − ∑ j = 1 R g − j u j , n ( ℓ ) ∣ + ∣ S g , n ( ℓ ) − ∑ j = 1 R g − j s j , n ( ℓ ) ∣ ≤ g − R − 1 [ C r ( n + 1 ) ] C r ( 1 + ℓ ) C r e ℓ , g ≥ [ C r ( n + 1 ) ] C , ℓ ≥ 0. (61) \left|U_{g,n}(\ell)-\phi(\ell)^2-\sum_{j=1}^{R}g^{-j}u_{j,n}(\ell)\right|+\left|S_{g,n}(\ell)-\sum_{j=1}^{R}g^{-j}s_{j,n}(\ell)\right|\le g^{-R-1}[Cr(n+1)]^{Cr}(1+\ell)^{Cr}e^{\ell},\qquad g\ge[Cr(n+1)]^{C},\quad\ell\ge0. \tag*{(61)} U g , n ( ℓ ) − ϕ ( ℓ ) 2 − j = 1 ∑ R g − j u j , n ( ℓ ) + S g , n ( ℓ ) − j = 1 ∑ R g − j s j , n ( ℓ ) ≤ g − R − 1 [ C r ( n + 1 ) ] C r ( 1 + ℓ ) C r e ℓ , g ≥ [ C r ( n + 1 ) ] C , ℓ ≥ 0. (61) The coefficients are independent of genus and truncation order, and satisfy
∣ u j , n ( ℓ ) ∣ + ∣ s j , n ( ℓ ) ∣ ≤ [ C ( j + 1 ) ( n + 1 ) ] C ( j + 1 ) ( 1 + ℓ ) C ( j + 1 ) e ℓ . (62) |u_{j,n}(\ell)|+|s_{j,n}(\ell)|\le[C(j+1)(n+1)]^{C(j+1)}(1+\ell)^{C(j+1)}e^{\ell}. \tag*{(62)} ∣ u j , n ( ℓ ) ∣ + ∣ s j , n ( ℓ ) ∣ ≤ [ C ( j + 1 ) ( n + 1 ) ] C ( j + 1 ) ( 1 + ℓ ) C ( j + 1 ) e ℓ . (62) In particular,
s 1 , n ( ℓ ) = ϕ ( ℓ ) 4 π 2 [ ℓ 2 + 4 π 2 48 + ( n 2 ) ] . (63) s_{1,n}(\ell)=\frac{\phi(\ell)}{4\pi^2}\left[\frac{\ell^2+4\pi^2}{48}+\binom{n}{2}\right]. \tag*{(63)} s 1 , n ( ℓ ) = 4 π 2 ϕ ( ℓ ) [ 48 ℓ 2 + 4 π 2 + ( 2 n ) ] . (63)
Proof. We expand the nonseparating kernel first. For the separating kernel, we bound the large-deficit tail and then expand the finitely many retained endpoint types. The handle-ratio coefficients of Proposition 2.3 are denoted by p n j p_n^j p n j , with p n 0 = 1 p_n^0=1 p n 0 = 1 . Factor
U g , n ( ℓ ) = V g − 1 , n + 2 V g , n V g − 1 , n + 2 ( 0 n , ℓ , ℓ ) V g − 1 , n + 2 . (64) U_{g,n}(\ell)=\frac{V_{g-1,n+2}}{V_{g,n}}\frac{V_{g-1,n+2}(0_n,\ell,\ell)}{V_{g-1,n+2}}. \tag*{(64)} U g , n ( ℓ ) = V g , n V g − 1 , n + 2 V g − 1 , n + 2 V g − 1 , n + 2 ( 0 n , ℓ , ℓ ) . (64)
Let B j , n + 2 ( b ) B_{j,n+2}^{(b)} B j , n + 2 ( b ) be the fixed boundary coefficients of Lemma 2.4 . The binomial expansion of ( g − 1 ) − j (g-1)^{-j} ( g − 1 ) − j gives fixed coefficients
β 0 , n = ϕ 2 , β d , n ( ℓ ) = ∑ j = 1 d ( d − 1 j − 1 ) B j , n + 2 ( b ) ( ℓ , ℓ ) , u d , n = ∑ a + b = d p n a β b , n ( d ≥ 1 ) . (65) \beta_{0,n}=\phi^2,\qquad\beta_{d,n}(\ell)=\sum_{j=1}^{d}\binom{d-1}{j-1}B_{j,n+2}^{(b)}(\ell,\ell),\qquad u_{d,n}=\sum_{a+b=d}p_n^a\beta_{b,n}\quad(d\ge1). \tag*{(65)} β 0 , n = ϕ 2 , β d , n ( ℓ ) = j = 1 ∑ d ( j − 1 d − 1 ) B j , n + 2 ( b ) ( ℓ , ℓ ) , u d , n = a + b = d ∑ p n a β b , n ( d ≥ 1 ) . (65)
The shifted boundary remainder is bounded by g − R − 1 [ C r ( n + 1 ) ] C r ( 1 + 2 ℓ ) 2 R + 2 e ℓ g^{-R-1}[Cr(n+1)]^{Cr}(1+2\ell)^{2R+2}e^\ell g − R − 1 [ C r ( n + 1 ) ] C r ( 1 + 2 ℓ ) 2 R + 2 e ℓ . To multiply the two expansions, write their error as ( P − P [ R ] ) D + P [ R ] ( D − D [ R ] ) (P-P^{[R]})D+P^{[R]}(D-D^{[R]}) ( P − P [ R ] ) D + P [ R ] ( D − D [ R ] ) , where P P P , D D D are the two factors in (64) and the superscript [ R ] [R] [ R ] denotes their truncation through order g − R g^{-R} g − R . Discard the terms of degree greater than R R R in the product of these truncations. Here ∣ D ∣ ≤ e ℓ |D|\le e^\ell ∣ D ∣ ≤ e ℓ , while the scalar truncation P [ R ] P^{[R]} P [ R ] is bounded at the stated genus threshold. Every discarded product has only one boundary coefficient, so its common length weight is factored out once. The geometric majorant of Lemma 2.5 then proves the required bound. This argument does not require ℓ / g \ell/g ℓ / g to be small and also applies when $R=0.
For the separating tail write N = 2 g + n − 3 N=2g+n-3 N = 2 g + n − 3 , s = n + 1 s=n+1 s = n + 1 , and, in a summand of (60) , put
c 1 = 2 i + a 0 − 1 , c 2 = 2 ( g − i ) + n − a 0 − 1 , c 1 + c 2 = N + 1 , δ = min ( c 1 , c 2 ) ≥ 1. (66) c_1=2i+a_0-1,\qquad c_2=2(g-i)+n-a_0-1,\qquad c_1+c_2=N+1,\qquad\delta=\min(c_1,c_2)\ge1. \tag*{(66)} c 1 = 2 i + a 0 − 1 , c 2 = 2 ( g − i ) + n − a 0 − 1 , c 1 + c 2 = N + 1 , δ = min ( c 1 , c 2 ) ≥ 1. (66)
The volume envelope of Lemma 2.1 , at g ≥ C s 2 g\ge Cs^2 g ≥ C s 2 , gives a universal B B B such that
( n a 0 ) V i , a 0 + 1 V g − i , n − a 0 + 1 V g , n ≤ B ( n a 0 ) ( δ − 1 ) ! ( N − δ ) ! N ! = B ( n a 0 ) δ ( N δ ) . (67) \binom{n}{a_0}\frac{V_{i,a_0+1}V_{g-i,n-a_0+1}}{V_{g,n}}\le B\binom{n}{a_0}\frac{(\delta-1)!(N-\delta)!}{N!}=\frac{B\binom{n}{a_0}}{\delta\binom{N}{\delta}}. \tag*{(67)} ( a 0 n ) V g , n V i , a 0 + 1 V g − i , n − a 0 + 1 ≤ B ( a 0 n ) N ! ( δ − 1 )! ( N − δ )! = δ ( δ N ) B ( a 0 n ) . (67)
Indeed, the powers of 4 π 2 4\pi^2 4 π 2 leave a fixed factor, the exponential denominator correction is bounded, and the larger complexity is at least g / 2 g/2 g /2 , controlling the square-root factor. A smaller side has genus h 0 h_0 h 0 , a ∗ a_\ast a ∗ cusps and one boundary, with 2 h 0 + a ∗ − 1 = δ 2h_0+a_\ast-1=\delta 2 h 0 + a ∗ − 1 = δ , so a ∗ ≤ δ + 1 a_\ast\le\delta+1 a ∗ ≤ δ + 1 and ( n a 0 ) = ( n a ∗ ) ≤ s δ + 1 \binom{n}{a_0}=\binom{n}{a_\ast}\le s^{\delta+1} ( a 0 n ) = ( a ∗ n ) ≤ s δ + 1 . There are at most 2 ( δ + 2 ) 2(\delta+2) 2 ( δ + 2 ) ordered records at this deficit, including a possible tie. Consequently their sum is at most 6 B x δ 6Bx_\delta 6 B x δ , where x δ = s δ + 1 / ( N δ ) x_\delta=s^{\delta+1}/\binom{N}{\delta} x δ = s δ + 1 / ( δ N ) .
For r = R + 1 r=R+1 r = R + 1 and g ≥ ( K r s ) 8 g\ge(Krs)^8 g ≥ ( K r s ) 8 , take D = ⌊ N / ( 64 s ) ⌋ D=\lfloor N/(64s)\rfloor D = ⌊ N / ( 64 s )⌋ . An absolute K K K ensures r ≤ D r\le D r ≤ D , g ≤ N ≤ 3 g g\le N\le3g g ≤ N ≤ 3 g , and N − r + 1 ≥ g N-r+1\ge g N − r + 1 ≥ g . For r ≤ δ < D r\le\delta<D r ≤ δ < D ,
x δ + 1 x δ = s ( δ + 1 ) N − δ ≤ 1 63 , ∑ δ = r D 6 B x δ ≤ 12 B s r + 1 r ! g − r . (68) \frac{x_{\delta+1}}{x_\delta}=\frac{s(\delta+1)}{N-\delta}\le\frac{1}{63},\qquad\sum_{\delta=r}^{D}6Bx_\delta\le12Bs^{r+1}r!g^{-r}. \tag*{(68)} x δ x δ + 1 = N − δ s ( δ + 1 ) ≤ 63 1 , δ = r ∑ D 6 B x δ ≤ 12 B s r + 1 r ! g − r . (68)
For δ > D \delta>D δ > D , use binomial unimodality in (67) , the identity ∑ a 0 ( n a 0 ) = 2 n \sum_{a_0}\binom{n}{a_0}=2^n ∑ a 0 ( a 0 n ) = 2 n , and the at most g + 1 g+1 g + 1 genus choices. Since D ≥ N / ( 128 s ) D\ge N/(128s) D ≥ N / ( 128 s ) and ( N D ) ≥ 2 D \binom{N}{D}\ge2^D ( D N ) ≥ 2 D , this part is bounded by B ( g + 1 ) 2 n e − α 0 N / s B(g+1)2^ne^{-\alpha_0N/s} B ( g + 1 ) 2 n e − α 0 N / s , where α 0 = log 2 / 128 \alpha_0=\log2/128 α 0 = log 2/128 . The same threshold, with K K K enlarged, gives
log ( g r ( g + 1 ) 2 n ) ≤ 3 r s log g ≤ 3 r s g ≤ α 0 N / s . (69) \log(g^r(g+1)2^n)\le3rs\log g\le3rs\sqrt{g}\le\alpha_0N/s. \tag*{(69)} log ( g r ( g + 1 ) 2 n ) ≤ 3 r s log g ≤ 3 r s g ≤ α 0 N / s . (69)
Thus the deep tail is at most B g − r Bg^{-r} B g − r . Boundary growth adds at most e ℓ e^\ell e ℓ for the two factors, proving the direct estimate
∑ δ ≥ R + 1 ( n a 0 ) V i , a 0 + 1 ( 0 a 0 , ℓ ) V g − i , n − a 0 + 1 ( 0 n − a 0 , ℓ ) V g , n ≤ g − R − 1 [ K ( R + 1 ) ( n + 1 ) ] 8 ( R + 1 ) e ℓ . (70) \sum_{\delta\ge R+1}\binom{n}{a_0}\frac{V_{i,a_0+1}(0_{a_0},\ell)V_{g-i,n-a_0+1}(0_{n-a_0},\ell)}{V_{g,n}}\le g^{-R-1}[K(R+1)(n+1)]^{8(R+1)}e^\ell. \tag*{(70)} δ ≥ R + 1 ∑ ( a 0 n ) V g , n V i , a 0 + 1 ( 0 a 0 , ℓ ) V g − i , n − a 0 + 1 ( 0 n − a 0 , ℓ ) ≤ g − R − 1 [ K ( R + 1 ) ( n + 1 ) ] 8 ( R + 1 ) e ℓ . (70)
The estimate also covers a pair of pants on the smaller side, equal-complexity sides, and the closed case n = 0 n = 0 n = 0 .
It remains to expand the retained endpoints. Define the fixed finite set
E δ ( n ) = { ( h 0 , a 0 ) ∈ Z ≥ 0 2 : a 0 ≤ n , 2 h 0 + a 0 − 1 = δ } . (71) \mathcal{E}_{\delta}(n)=\{(h_0,a_0)\in\mathbb{Z}_{\geq0}^{2}:a_0\leq n,\ 2h_0+a_0-1=\delta\}. \tag*{(71)} E δ ( n ) = {( h 0 , a 0 ) ∈ Z ≥ 0 2 : a 0 ≤ n , 2 h 0 + a 0 − 1 = δ } . (71)
When N > 2 R N>2R N > 2 R , a term with δ ≤ R \delta\leq R δ ≤ R has a unique smaller side. Each record in (71) has exactly two placements in the ordered sum. Its other component has genus g − h 0 g-h_0 g − h 0 and p = n − a 0 + 1 = n + 2 h 0 − δ ≥ 1 p=n-a_0+1=n+2h_0-\delta\geq1 p = n − a 0 + 1 = n + 2 h 0 − δ ≥ 1 marked points. The shifted ratio of Lemma 2.6 has canonical coefficients Q d n , h 0 , δ ; 1 ( ℓ ) Q_{d}^{n,h_0,\delta;1}(\ell) Q d n , h 0 , δ ; 1 ( ℓ ) , starting with Q 0 = ϕ Q_0=\phi Q 0 = ϕ , and leading scalar ( 8 π 2 ) − δ g − δ (8\pi^2)^{-\delta}g^{-\delta} ( 8 π 2 ) − δ g − δ . Therefore define
s j , n ( ℓ ) = 2 ∑ δ = 1 j ∑ ( h 0 , a 0 ) ∈ E δ ( n ) ( 8 π 2 ) − δ ( n a 0 ) V h 0 , a 0 + 1 ( 0 a 0 , ℓ ) Q j − δ n , h 0 , δ ; 1 ( ℓ ) . (72) s_{j,n}(\ell)=2\sum_{\delta=1}^{j}\sum_{(h_0,a_0)\in\mathcal{E}_{\delta}(n)}(8\pi^2)^{-\delta}\binom{n}{a_0}V_{h_0,a_0+1}(0_{a_0},\ell)Q_{j-\delta}^{n,h_0,\delta;1}(\ell). \tag*{(72)} s j , n ( ℓ ) = 2 δ = 1 ∑ j ( h 0 , a 0 ) ∈ E δ ( n ) ∑ ( 8 π 2 ) − δ ( a 0 n ) V h 0 , a 0 + 1 ( 0 a 0 , ℓ ) Q j − δ n , h 0 , δ ; 1 ( ℓ ) . (72)
Here h 0 ≤ ( δ + 1 ) / 2 h_0\leq(\delta+1)/2 h 0 ≤ ( δ + 1 ) /2 , a 0 ≤ δ + 1 a_0\leq\delta+1 a 0 ≤ δ + 1 , and ∣ E δ ( n ) ∣ ≤ δ + 2 |\mathcal{E}_{\delta}(n)|\leq\delta+2 ∣ E δ ( n ) ∣ ≤ δ + 2 . The small-volume prefactor is at most [ C ( δ + 1 ) ( n + 1 ) ] C ( δ + 1 ) e ℓ / 2 [C(\delta+1)(n+1)]^{C(\delta+1)}e^{\ell/2} [ C ( δ + 1 ) ( n + 1 ) ] C ( δ + 1 ) e ℓ /2 by Lemma 2.1. Multiplication by the shifted remainder contributes e ℓ / 2 e^{\ell/2} e ℓ /2 once more. Summing the O ( R 2 ) O(R^2) O ( R 2 ) records and adding (70) proves (61) ; the same finite sums at order j j j prove (62) . For δ = 1 \delta=1 δ = 1 the only records are ( 1 , 0 ) (1,0) ( 1 , 0 ) and, if available, ( 0 , 2 ) (0,2) ( 0 , 2 ) . Using V 1 , 1 ( ℓ ) = ( ℓ 2 + 4 π 2 ) / 48 V_{1,1}(\ell)=(\ell^2+4\pi^2)/48 V 1 , 1 ( ℓ ) = ( ℓ 2 + 4 π 2 ) /48 and V 0 , 3 = 1 V_{0,3}=1 V 0 , 3 = 1 gives (63) .
Exact simple integration and repetitions
Because (58) counts both orientations, we convert the unoriented cutting formula to an oriented density.
Lemma 3.2 (Oriented simple integration). Writing K g , n = U g , n + S g , n K_{g,n}=U_{g,n}+S_{g,n} K g , n = U g , n + S g , n , one has
E g , n ∑ γ ∈ P o r s i m p ( X ) F ( ℓ γ ) = ∫ 0 ∞ F ( ℓ ) ℓ K g , n ( ℓ ) d ℓ (73) \mathbb{E}_{g,n}\sum_{\gamma\in\mathcal{P}_{\mathrm{or}}^{\mathrm{simp}}(X)}F(\ell_\gamma)=\int_{0}^{\infty}F(\ell)\ell K_{g,n}(\ell)\,d\ell \tag*{(73)} E g , n γ ∈ P or simp ( X ) ∑ F ( ℓ γ ) = ∫ 0 ∞ F ( ℓ ) ℓ K g , n ( ℓ ) d ℓ (73) for every nonnegative Borel F F F , with infinity allowed. Consequently O H s i m p \mathcal{O}_{H}^{\mathrm{simp}} O H simp is absolutely integrable and
E g , n O H s i m p = ∫ 0 ∞ ∑ ν ≥ 1 ℓ 2 H ( ν ℓ ) 2 sinh ( ν ℓ / 2 ) K g , n ( ℓ ) d ℓ . (74) \mathbb{E}_{g,n}\mathcal{O}_{H}^{\mathrm{simp}}=\int_{0}^{\infty}\sum_{\nu\geq1}\frac{\ell^2H(\nu\ell)}{2\sinh(\nu\ell/2)}K_{g,n}(\ell)\,d\ell. \tag*{(74)} E g , n O H simp = ∫ 0 ∞ ν ≥ 1 ∑ 2 sinh ( ν ℓ /2 ) ℓ 2 H ( ν ℓ ) K g , n ( ℓ ) d ℓ . (74)
Proof. For an unoriented simple orbit, the marked integration formula [17 §4, (4.1), Theorem 4.1 and (4.3)] gives the density 2 − M N ( γ ) − 1 ∏ V ~ z 2^{-M}N(\gamma)^{-1}\prod\widetilde{V}_z 2 − M N ( γ ) − 1 ∏ V z before division by V g , n V_{g,n} V g , n . Here N ( γ ) N(\gamma) N ( γ ) is the index of the subgroup preserving the curve’s orientation in its full stabilizer, and M M M counts intact one-holed tori separated off by the curve. In genus at least three there is at most one such component. Its geometric volume is 2 V 1 , 1 2V_{1,1} 2 V 1 , 1 [17 (4.5)] , so 2 − M ∏ V ~ z = ∏ V z 2^{-M}\prod\widetilde{V}_z=\prod V_z 2 − M ∏ V z = ∏ V z in our orbifold convention [5 §1] . The corresponding oriented, ordered convention for closed surfaces is also recorded in [1 Example 2.6] .
For a nonseparating curve N ( γ ) = 2 N(\gamma)=2 N ( γ ) = 2 , since the two cut boundaries can be exchanged. Counting both orientations therefore gives ℓ U g , n \ell U_{g,n} ℓ U g , n . For a separating curve first mark an actual side ( i , A ) (i,A) ( i , A ) , A ⊆ [ n ] A\subseteq[n] A ⊆ [ n ] . Forgetting the side identifies ( i , A ) (i,A) ( i , A ) with ( g − i , A c ) (g-i,A^c) ( g − i , A c ) . Usually these are two distinct records and N ( γ ) = 1 N(\gamma)=1 N ( γ ) = 1 . The sole one-record case is n = 0 n=0 n = 0 , i = g / 2 i=g/2 i = g /2 , when N ( γ ) = 2 N(\gamma)=2 N ( γ ) = 2 . Even if numerical types coincide when n > 0 n>0 n > 0 , their disjoint labelled cusp sets remain distinct, so no side-exchanging stabilizer fixes all labels. Thus the unoriented density is one half of the ordered side sum in all cases. Summing subsets introduces exactly ( n a 0 ) \binom{n}{a_0} ( a 0 n ) ; orientation doubles the result to ℓ S g , n \ell S_{g,n} ℓ S g , n . This proves (73) for continuous tests of compact support away from zero. Equality of the resulting locally finite measures extends it to nonnegative Borel tests.
Tonelli and 2 sinh ( ν ℓ / 2 ) ≥ ν ℓ 2\sinh(\nu\ell/2) \ge\nu\ell 2 sinh ( ν ℓ /2 ) ≥ ν ℓ give
∑ ν ≥ 1 ∫ 0 T / ν ℓ 2 ∣ H ( ν ℓ ) ∣ K g , n ( ℓ ) 2 sinh ( ν ℓ / 2 ) d ℓ ≤ ζ ( 3 ) 2 T 2 ∥ K g , n ∥ C [ 0 , T ] ∥ H ∥ ∞ < ∞ . (75) \sum_{\nu\ge1}\int_{0}^{T/\nu}\frac{\ell^{2}|H(\nu\ell)|K_{g,n}(\ell)}{2\sinh(\nu\ell/2)}\,d\ell\le\frac{\zeta(3)}{2}T^{2}\|K_{g,n}\|_{C[0,T]}\|H\|_{\infty}<\infty. \tag*{(75)} ν ≥ 1 ∑ ∫ 0 T / ν 2 sinh ( ν ℓ /2 ) ℓ 2 ∣ H ( ν ℓ ) ∣ K g , n ( ℓ ) d ℓ ≤ 2 ζ ( 3 ) T 2 ∥ K g , n ∥ C [ 0 , T ] ∥ H ∥ ∞ < ∞. (75)
The kernel is a finite sum of volume polynomials at fixed g , n g,n g , n . Signed Fubini now gives (74) . If T = 0 T=0 T = 0 , both sides vanish. Thus repetitions cause no singularity at zero length.
For a continuous function k : [ 0 , ∞ ) → R k:[0,\infty)\to\mathbb{R} k : [ 0 , ∞ ) → R , define the linear operation
L k ( H ) = ∫ 0 ∞ ∑ ν ≥ 1 ℓ 2 k ( ℓ ) H ( ν ℓ ) 2 sinh ( ν ℓ / 2 ) d ℓ . (76) \mathcal{L}_{k}(H)=\int_{0}^{\infty}\sum_{\nu\ge1}\frac{\ell^{2}k(\ell)H(\nu\ell)}{2\sinh(\nu\ell/2)}\,d\ell. \tag*{(76)} L k ( H ) = ∫ 0 ∞ ν ≥ 1 ∑ 2 sinh ( ν ℓ /2 ) ℓ 2 k ( ℓ ) H ( ν ℓ ) d ℓ . (76)
If ∣ k ( ℓ ) ∣ ≤ A ( 1 + ℓ ) d e ℓ |k(\ell)|\le A(1+\ell)^{d}e^{\ell} ∣ k ( ℓ ) ∣ ≤ A ( 1 + ℓ ) d e ℓ , the preceding argument gives ∣ L k ( H ) ∣ ≤ ζ ( 3 ) A T 2 ( 1 + T ) d e T ∥ H ∥ ∞ / 2 |\mathcal{L}_{k}(H)|\le\zeta(3)AT^{2}(1+T)^{d}e^{T}\|H\|_{\infty}/2 ∣ L k ( H ) ∣ ≤ ζ ( 3 ) A T 2 ( 1 + T ) d e T ∥ H ∥ ∞ /2 . Set
η j , n = L u j , n + s j , n , Ω ( H ) = L ϕ 2 ( H ) = ∫ 0 ∞ ∑ ν ≥ 1 2 sinh 2 ( ℓ / 2 ) sinh ( ν ℓ / 2 ) H ( ν ℓ ) d ℓ . (77) \eta_{j,n}=\mathcal{L}_{u_{j,n}+s_{j,n}},\qquad\Omega(H)=\mathcal{L}_{\phi_{2}}(H)=\int_{0}^{\infty}\sum_{\nu\ge1}\frac{2\sinh^{2}(\ell/2)}{\sinh(\nu\ell/2)}H(\nu\ell)\,d\ell. \tag*{(77)} η j , n = L u j , n + s j , n , Ω ( H ) = L ϕ 2 ( H ) = ∫ 0 ∞ ν ≥ 1 ∑ sinh ( ν ℓ /2 ) 2 sinh 2 ( ℓ /2 ) H ( ν ℓ ) d ℓ . (77)
These functionals are defined on all compactly supported bounded even Borel tests, independently of a support cutoff. Applying Lemma 3.1 inside (74) proves
∣ E H s i m p − Ω ( H ) − ∑ j = 1 R g − j η j , n ( H ) ∣ ≤ g − R − 1 [ C ( R + 1 ) ( n + 1 ) ] C ( R + 1 ) ( 1 + T ) C ( R + 1 ) e T ∥ H ∥ ∞ . (78) \left|\mathop{\mathrm{E}}\nolimits_{H}^{\mathrm{simp}}-\Omega(H)-\sum_{j=1}^{R}g^{-j}\eta_{j,n}(H)\right|
\le g^{-R-1}\left[C(R+1)(n+1)\right]^{C(R+1)}(1+T)^{C(R+1)}e^{T}\|H\|_{\infty}. \tag*{(78)} E H simp − Ω ( H ) − j = 1 ∑ R g − j η j , n ( H ) ≤ g − R − 1 [ C ( R + 1 ) ( n + 1 ) ] C ( R + 1 ) ( 1 + T ) C ( R + 1 ) e T ∥ H ∥ ∞ . (78)
This holds under g ≥ [ C ( R + 1 ) ( n + 1 ) ] C g\ge[C(R+1)(n+1)]^{C} g ≥ [ C ( R + 1 ) ( n + 1 ) ] C . The same estimate without the genus factor, with R R R replaced by j j j , bounds η j , n \eta_{j,n} η j , n .
Cusp-relative carriers and filling fibers
The use of the subsurface filled by a geodesic follows the local-type integration framework of Anantharaman and Monk [1 §4.1, Theorems 5.7 and 6.1] for closed ambient surfaces. We use the same construction with inherited cusps, keeping their labels and the boundary-slot symmetries explicit. A boundary slot is a boundary component of the normalized carrier; two slots may map to the same geodesic on the ambient surface.
The normalized carrier of a nonsimple geodesic is constructed from a small regular neighborhood of its geodesic graph. Fill disk regions, complete peripheral punctured-disk regions to their inherited cusp ends, and tighten the remaining boundary curves to positive geodesics. If two boundaries tighten to the same ambient cuff, retain two distinct boundary slots in the normalization. This is the construction preceding [7 Proposition 3.1] , with peripheral completion made explicit. The carrier interior contains the geodesic, and its inclusion in the ambient surface is an injective local isometry. Its type ( h 0 , a 0 , k ) (h_0,a_0,k) ( h 0 , a 0 , k ) records the genus, the number of inherited cusps and the number of positive boundary slots, respectively. Its complexity is m = 2 h 0 + a 0 + k − 2 ≥ 1 m=2h_0+a_0+k-2\ge1 m = 2 h 0 + a 0 + k − 2 ≥ 1 . An interior geodesic fills a carrier if the complement of a small regular neighborhood of its graph consists of disks, cusp neighborhoods, and annular collars of individual positive boundary components.
Lemma 3.3 (Geometry and reconstruction). For an interior geodesic γ \gamma γ filling a normalized carrier Y Y Y ,
Λ : = ℓ ( ∂ Y ) ≤ 2 ℓ γ , 2 π m = Area ( Y ) ≤ 4 ℓ γ , ℓ γ ≥ π m / 2. (79) \Lambda:=\ell(\partial Y)\le2\ell_{\gamma},\qquad2\pi m=\operatorname{Area}(Y)\le4\ell_{\gamma},\qquad\ell_{\gamma}\ge\pi m/2. \tag*{(79)} Λ := ℓ ( ∂ Y ) ≤ 2 ℓ γ , 2 π m = Area ( Y ) ≤ 4 ℓ γ , ℓ γ ≥ π m /2. (79) Gluing a filled carrier to stable complementary components, and possibly self-pairing some slots, reconstructs its canonical normalized carrier. Intrinsic and ambient primitivity are equivalent.
Proof. Cut Y Y Y along the graph of γ \gamma γ . The resulting regions are disks, peripheral punctured disks, and annuli adjacent to individual positive boundary components. Their graph-side lengths sum to 2 ℓ γ 2\ell_\gamma 2 ℓ γ . Each positive boundary geodesic minimizes its free homotopy class, proving the perimeter estimate, with both slots counted in a self-pair.
The area bound also holds for carriers with cusps. On a disk lifted to the hyperbolic plane use a unit Busemann vector field of divergence one. On a peripheral region’s parabolic cyclic cover use − y ∂ y -y\partial_y − y ∂ y , also of norm one and divergence one; truncate high in the cusp and let the extra horocycle flux tend to zero. On an annular region’s hyperbolic cyclic cover use tanh ( r ) ∂ r \tanh(r)\partial_r tanh ( r ) ∂ r , where r r r is signed distance from the core geodesic. This field has divergence one, norm at most one, and zero core flux. The divergence theorem, with corners smoothed, bounds each region’s area by its graph-side length. This gives the stronger estimate Area ( Y ) ≤ 2 ℓ γ \operatorname{Area}(Y)\le2\ell_\gamma Area ( Y ) ≤ 2 ℓ γ ; the weaker bound in (79) suffices below. Gauss–Bonnet gives 2 π m = Area ( Y ) 2\pi m=\operatorname{Area}(Y) 2 π m = Area ( Y ) , and the lower bound for ℓ γ \ell_\gamma ℓ γ follows.
For reconstruction let ι : Y ∘ → X \iota:Y^\circ\to X ι : Y ∘ → X be the injective local isometry after gluing. If a unit-speed closed geodesic α \alpha α has an ambient period d d d , then ι ( α ( t + d ) ) = ι ( α ( t ) ) \iota(\alpha(t+d))=\iota(\alpha(t)) ι ( α ( t + d )) = ι ( α ( t )) for every t t t , so injectivity makes d d d an intrinsic period. The converse is immediate. This proves primitivity in both directions and also preserves oriented geodesics and their self-intersections.
Outside a small graph neighborhood, intrinsic disks and cusp regions are absorbed by the carrier construction. The two collars of a self-pair form an annulus, whose two boundary germs tighten to the same seam. Each stable exterior component, with its attached collars, retains negative Euler characteristic and cannot be absorbed. Hence the construction recovers precisely the initial carrier and its slot pairing. Conversely, cutting an ambient canonical carrier gives these records uniquely: equality of cuff images determines self-pairs, and connected exterior regions determine complementary blocks. A cuff has only two sides. All remaining seams are essential and nonparallel, as can also be seen by extending the cut components to pants decompositions. Finally a simple geodesic has an annular graph neighborhood; adjoining only disks, cusp regions and boundary collars cannot produce negative Euler characteristic. Thus every filling geodesic of a carrier with m ≥ 1 m\ge1 m ≥ 1 is nonsimple.
Fix a carrier type ( h 0 , a 0 , k ) (h_0,a_0,k) ( h 0 , a 0 , k ) with k ≥ 1 k\ge1 k ≥ 1 , 0 ≤ a 0 ≤ n 0\le a_0\le n 0 ≤ a 0 ≤ n and m = 2 h 0 + a 0 + k − 2 ≥ 1 m=2h_0+a_0+k-2\ge1 m = 2 h 0 + a 0 + k − 2 ≥ 1 . Write P o r f i l l ( Y ) \mathcal{P}_{\mathrm{or}}^{\mathrm{fill}}(Y) P or fill ( Y ) for the oriented primitive interior geodesics filling Y Y Y . On each positive boundary fiber define pointwise
A H ( Y ) = ∑ α ∈ P o r f i l l ( Y ) ∑ ν ≥ 1 ℓ α H ( ν ℓ α ) 2 sinh ( ν ℓ α / 2 ) , A_H(Y)=\sum_{\alpha\in\mathcal{P}_{\mathrm{or}}^{\mathrm{fill}}(Y)}\sum_{\nu\ge1}\frac{\ell_\alpha H(\nu\ell_\alpha)}{2\sinh(\nu\ell_\alpha/2)}, A H ( Y ) = α ∈ P or fill ( Y ) ∑ ν ≥ 1 ∑ 2 sinh ( ν ℓ α /2 ) ℓ α H ( ν ℓ α ) ,
I H h 0 , a 0 , k ( x ) = ∫ M h 0 , a 0 + k ( x a 0 ) A H ( Y ) d W P o r b ( Y ) , Ψ H h 0 , a 0 , k ( x ) = I H h 0 , a 0 , k ( x ) V h 0 , a 0 + k ( x a 0 ) . (80) I_H^{h_0,a_0,k}(\mathbf{x})=\int_{\mathcal{M}_{h_0,a_0+k}(\mathbf{x}_{a_0})}A_H(Y)\,\mathrm{dWP}_{\mathrm{orb}}(Y),\qquad
\Psi_H^{h_0,a_0,k}(\mathbf{x})=\frac{I_H^{h_0,a_0,k}(\mathbf{x})}{V_{h_0,a_0+k}(\mathbf{x}_{a_0})}. \tag*{(80)} I H h 0 , a 0 , k ( x ) = ∫ M h 0 , a 0 + k ( x a 0 ) A H ( Y ) dWP orb ( Y ) , Ψ H h 0 , a 0 , k ( x ) = V h 0 , a 0 + k ( x a 0 ) I H h 0 , a 0 , k ( x ) . (80)
The numerator and denominator use the same orbifold measure, including its factor one half on an intact one-holed torus [1 §2.3.4 and Remark 2.7] and [5 §1] .
The oriented primitive count in [7 Lemma 3.7 and its preceding convention] applies to surfaces with cusps and geodesic boundaries:
# { α ∈ P o r ( Y ) : ℓ α ≤ T } ≤ ( h 0 − 1 + a 0 + k 2 ) e T + 6 + 3 h 0 − 3 + a 0 + k ≤ C m e T . (81) \#\{\alpha\in\mathcal{P}_{\mathrm{or}}(Y):\ell_\alpha\le T\}\le\left(h_0-1+\frac{a_0+k}{2}\right)e^{T+6}+3h_0-3+a_0+k\le Cme^T. \tag*{(81)} # { α ∈ P or ( Y ) : ℓ α ≤ T } ≤ ( h 0 − 1 + 2 a 0 + k ) e T + 6 + 3 h 0 − 3 + a 0 + k ≤ C m e T . (81)
By (79) , a contributing repetition has ν ≤ 2 T / ( π m ) \nu\le2T/(\pi m) ν ≤ 2 T / ( π m ) . Since ℓ / ( 2 sinh ( ν ℓ / 2 ) ) ≤ 1 / ν \ell/(2\sinh(\nu\ell/2))\le1/\nu ℓ / ( 2 sinh ( ν ℓ /2 )) ≤ 1/ ν ,
∣ A H ∣ , ∣ Ψ H ∣ ≤ C m ( 1 + T ) e T ∥ H ∥ ∞ , Ψ H = 0 if Λ > 2 T or m > 2 T / π . (82) |A_H|,\ |\Psi_H|\le Cm(1+T)e^T\lVert H\rVert_\infty,\qquad\Psi_H=0\ \text{if }\Lambda>2T\ \text{or }m>2T/\pi. \tag*{(82)} ∣ A H ∣ , ∣ Ψ H ∣ ≤ C m ( 1 + T ) e T ∥ H ∥ ∞ , Ψ H = 0 if Λ > 2 T or m > 2 T / π . (82)
The pointwise bound and finite WP volume imply that every fiber integral is finite, even when the fiber moduli space is noncompact. We also need measurability in the boundary lengths, including on self-pairing diagonals. The maps x ↦ I H h 0 , a 0 , k ( x ) x \mapsto I_H^{h_0,a_0,k}(x) x ↦ I H h 0 , a 0 , k ( x ) and x ↦ Ψ H h 0 , a 0 , k ( x ) x \mapsto\Psi_H^{h_0,a_0,k}(x) x ↦ Ψ H h 0 , a 0 , k ( x ) are Borel on ( 0 , ∞ ) k (0,\infty)^k ( 0 , ∞ ) k : the countable set of marked primitive filling classes has continuous length functions in Fenchel–Nielsen coordinates; the sum defining A H A_H A H is therefore Borel. A countable orbifold atlas and measurable partition express its integral as parameter integrals against the internal length and twist measures, with finite stabilizer weights. Apply parameter integration first to A ∣ H ∣ A_{|H|} A ∣ H ∣ and then to the positive and negative parts. Division by the positive continuous volume preserves measurability. This defines I H I_H I H and Ψ H \Psi_H Ψ H at every positive boundary-length vector, rather than only almost everywhere. It therefore justifies restriction to the equal-length diagonals of self-paired slots.
The exact carrier formula
Use the patterns of Theorem 2.7 . Recall that a pattern P = ( M , Π ) P=(M,\Pi) P = ( M , Π ) on [ k ] [k] [ k ] has a partial matching M M M of 2 σ 2\sigma 2 σ slots and a partition of the remaining slots into nonempty blocks I 1 , … , I b I_1,\ldots,I_b I 1 , … , I b , b ≥ 1 b\geq1 b ≥ 1 , ordered by their least labels. Each self-pair gives one gluing edge, and each remaining slot gives one edge joining the carrier to an exterior component. Hence there are e P = k − σ e_P=k-\sigma e P = k − σ actual edges. Their lengths y \mathbf{y} y induce the slot lengths x P \mathbf{x}_P x P and
Λ P = 2 ∑ self edges e y e + ∑ external edges e y e , J P = ∏ e y e . (83) \Lambda_P=2\sum_{\text{self edges }e}y_e+\sum_{\text{external edges }e}y_e,\qquad J_P=\prod_e y_e. \tag*{(83)} Λ P = 2 self edges e ∑ y e + external edges e ∑ y e , J P = e ∏ y e . (83)
Writing k i = ∣ I i ∣ k_i=|I_i| k i = ∣ I i ∣ , the complementary genera and cusp counts satisfy
∑ i a i = n − a 0 , ∑ i g i = g − h 0 − k + σ + b , g i , a i ∈ Z ≥ 0 , 2 g i + a i + k i − 2 ≥ 1. (84) \sum_i a_i=n-a_0,\qquad\sum_i g_i=g-h_0-k+\sigma+b,\qquad g_i,a_i\in\mathbb{Z}_{\geq0},\qquad2g_i+a_i+k_i-2\geq1. \tag*{(84)} i ∑ a i = n − a 0 , i ∑ g i = g − h 0 − k + σ + b , g i , a i ∈ Z ≥ 0 , 2 g i + a i + k i − 2 ≥ 1. (84)
The nonnegative kernel supplied there is
F g , n h 0 , a 0 , k ; P ( y ) = J P V h 0 , a 0 + k ( 0 a 0 , x P ) ∑ (3.27) n ! a 0 ! ∏ i a i ! ∏ i V g i , a i + k i ( 0 a i , x I i ) V g , n . (85) F_{g,n}^{h_0,a_0,k;P}(\mathbf{y})=J_PV_{h_0,a_0+k}(\mathbf{0}_{a_0},\mathbf{x}_P)\sum_{\text{(3.27)}}\frac{n!}{a_0!\prod_i a_i!}\frac{\prod_i V_{g_i,a_i+k_i}(\mathbf{0}_{a_i},\mathbf{x}_{I_i})}{V_{g,n}}. \tag*{(85)} F g , n h 0 , a 0 , k ; P ( y ) = J P V h 0 , a 0 + k ( 0 a 0 , x P ) (3.27) ∑ a 0 ! ∏ i a i ! n ! V g , n ∏ i V g i , a i + k i ( 0 a i , x I i ) . (85)
A rooted carrier here means one whose boundary slots are bijectively labelled by [ k ] [k] [ k ] . The factor 1 / k ! 1/k! 1/ k ! below forgets these labels after integration. Patterns in this formula have at least one stable exterior component. Carriers with no such component are handled separately in the proof of Theorem 3.5 .
Proposition 3.4 (Rooted carrier integration). Let O H h 0 , a 0 , k O_H^{h_0,a_0,k} O H h 0 , a 0 , k restrict (3.1) to primitive nonsimple geodesics with this normalized carrier type and at least one stable complementary component. Then it is absolutely WP integrable and
E g , n O H h 0 , a 0 , k = 1 k ! ∑ P ∈ P k ∫ ( 0 , ∞ ) e P Ψ H h 0 , a 0 , k ( x P ( y ) ) F g , n h 0 , a 0 , k ; P ( y ) d y . (86) \mathbb{E}_{g,n}O_H^{h_0,a_0,k}=\frac{1}{k!}\sum_{P\in\mathcal{P}_k}\int_{(0,\infty)^{e_P}}\Psi_H^{h_0,a_0,k}(\mathbf{x}_P(\mathbf{y}))F_{g,n}^{h_0,a_0,k;P}(\mathbf{y})\,d\mathbf{y}. \tag*{(86)} E g , n O H h 0 , a 0 , k = k ! 1 P ∈ P k ∑ ∫ ( 0 , ∞ ) e P Ψ H h 0 , a 0 , k ( x P ( y )) F g , n h 0 , a 0 , k ; P ( y ) d y . (86)
Proof. First fix actual disjoint cusp-label sets A 0 , A 1 , … , A b A_0,A_1,\ldots,A_b A 0 , A 1 , … , A b of sizes a 0 , a 1 , … , a b a_0,a_1,\ldots,a_b a 0 , a 1 , … , a b . Label all carrier boundary germs by [ k ] [k] [ k ] and all complementary boundaries by their incident slots. Orient an external edge by its carrier-side boundary orientation and a self-edge by its smaller-labelled slot; order edges by their least carrier label. Lemma 3.3 gives inverse cutting and gluing for these rooted records. The dual graph has b + 1 b+1 b + 1 vertices and k − σ k-\sigma k − σ edges, giving precisely the genus equation in (3.27).
The marked Fenchel–Nielsen product measure has one twist circle of full marked period y e y_e y e per actual edge. After twist integration it is
( ∏ e y e d y e ) d W P o r b ( Y ) ∏ i d W P o r b ( Z i ) . (87) \left(\prod_e y_e\,dy_e\right)d\mathrm{WP}_{\mathrm{orb}}(Y)\prod_i d\mathrm{WP}_{\mathrm{orb}}(Z_i). \tag*{(87)} ( e ∏ y e d y e ) d WP orb ( Y ) i ∏ d WP orb ( Z i ) . (87)
This is the coordinate formula underlying [1 Theorem 2.5 and Remark 2.7] ; inherited cusp labels are kept as in [17 §4] . To justify its use for a nonconstant carrier observable, take a local cut-moduli chart [ U / G ] [U/G] [ U / G ] . The gluing chart is the corresponding twist bundle [ U × ∏ e ( R / y e Z ) / G ] [U \times\prod_{e}(\mathbb{R}/y_{e}\mathbb{Z)/G]} [ U × ∏ e ( R / y e Z ) / G ] . Orbifold integration divides the cover integral by ∣ G ∣ |G| ∣ G ∣ . Although G G G may rotate twists or permute filling geodesics, the sum A H A_H A H is invariant and independent of the twists. Integrating in the twists therefore multiplies A H A_H A H by ∏ e y e \prod_{e} y_e ∏ e y e , leaving exactly the cut-component orbifold weights. Product charts and a measurable partition give (87) globally. For an intact torus, write B B B for its raw geometric chart and ι \iota ι for its elliptic involution. On the diagonal quotient [ B × ( R / y Z ) / ⟨ ι ⟩ ] [B \times(\mathbb{R}/y\mathbb{Z})/\langle\iota\rangle] [ B × ( R / y Z ) / ⟨ ι ⟩] , integration gives 1 2 ∫ B ∫ 0 y A H d t = ∫ B A H d W P o r b \frac{1}{2}\int_B\int_0^y A_H\,dt=\int_B A_H\,dW_P^{\mathrm{orb}} 2 1 ∫ B ∫ 0 y A H d t = ∫ B A H d W P orb , with raw geometric measure understood on the left. Thus the full marked twist period and the orbifold volume account for the half-twist symmetry exactly once. A torus created only after regluing a cut pair of pants introduces no additional factor of 1 / 2 1/2 1/2 .
Integrating a nonnegative A H A_H A H first in the carrier, then in the other components and twists, gives I H J P ∏ i V g i , a i + k i I_HJ_P\prod_i V_{g_i,a_i+k_i} I H J P ∏ i V g i , a i + k i . The number of cusp-label assignments is exactly n ! / ( a 0 ! ∏ i a i ! ) n!/(a_0!\prod_i a_i!) n ! / ( a 0 ! ∏ i a i !) . Summation and division by V g , n V_{g,n} V g , n produce the integrand Ψ g , n H F \Psi^{HF}_{g,n} Ψ g , n H F in (86) . Finally every normalized carrier has k ! k! k ! bijective labellings of its boundary germs, even when two germs have the same ambient image. The invariant filling sum descends through this finite symmetric-group quotient. Orbifold integration divides by k ! k! k ! , including at metrics with automorphisms. Self-pair reversals, permutations of self-pairs, and exchanges of exterior blocks are already included in this quotient; there is no additional 2 − σ 2^{-\sigma} 2 − σ , σ ! − 1 \sigma!^{-1} σ ! − 1 or b ! − 1 b!^{-1} b ! − 1 . Automorphisms of the marked configuration preserve the distinguished carrier. For example, the three patterns with one self-pair on a pants carrier with three boundary slots contribute 3 / 3 ! = 1 / 2 3/3!=1/2 3/3 ! = 1/2 in total.
For nonnegative H H H , Tonelli’s theorem proves the identity with + ∞ +\infty + ∞ allowed. At fixed g , n g,n g , n the complementary-data sum is finite and its kernel is continuous on the closed length orthant. Equation (82) restricts the integrals to the compact simplex Λ P ≤ 2 T \Lambda_P\le2T Λ P ≤ 2 T and bounds Ψ H \Psi_H Ψ H there. Applying the identity to ∣ H ∣ |H| ∣ H ∣ proves absolute integrability, then signed Fubini proves the stated formula.
Let F j , n h 0 , a 0 , k ; P \mathcal{F}^{h_0,a_0,k;P}_{j,n} F j , n h 0 , a 0 , k ; P be the canonical coefficients of Theorem 2.7. Define, without choosing R R R or T T T ,
ξ j , n h 0 , a 0 , k ( H ) = 1 k ! ∑ P ∈ P k ∫ Ψ H h 0 , a 0 , k ( x P ) F j , n h 0 , a 0 , k ; P ( y ) d y . (88) \xi^{h_0,a_0,k}_{j,n}(H)=\frac{1}{k!}\sum_{P\in\mathcal{P}_k}\int\Psi_H^{h_0,a_0,k}(x_P)\mathcal{F}^{h_0,a_0,k;P}_{j,n}(y)\,dy. \tag*{(88)} ξ j , n h 0 , a 0 , k ( H ) = k ! 1 P ∈ P k ∑ ∫ Ψ H h 0 , a 0 , k ( x P ) F j , n h 0 , a 0 , k ; P ( y ) d y . (88)
For each test H H H , the fiber bound restricts these integrals to a bounded length simplex, where the continuous kernels are bounded. Thus the integrals converge absolutely and define real linear functionals independently of any chosen support cutoff. Since F j = 0 \mathcal{F}_j=0 F j = 0 for j < m + b − 1 j<m+b-1 j < m + b − 1 , and b ≥ 1 b\ge1 b ≥ 1 ,
ξ j , n h 0 , a 0 , k = 0 ( j < m ) , ξ j , n h 0 , a 0 , k ( H ) = 0 ( m > 2 T / π ) . (89) \xi^{h_0,a_0,k}_{j,n}=0 \quad(j<m), \qquad\xi^{h_0,a_0,k}_{j,n}(H)=0 \quad(m>2T/\pi). \tag*{(89)} ξ j , n h 0 , a 0 , k = 0 ( j < m ) , ξ j , n h 0 , a 0 , k ( H ) = 0 ( m > 2 T / π ) . (89)
The second vanishing follows from the support condition in (82) .
To integrate the kernel remainder, note that ∣ P k ∣ ≤ ( k + 1 ) 2 k |\mathcal{P}_k|\le(k+1)^{2k} ∣ P k ∣ ≤ ( k + 1 ) 2 k : encode each matched partner, or zero, and then the least label in each remaining block. Also e P ≤ k ≤ m + 2 e_P\le k\le m+2 e P ≤ k ≤ m + 2 and
Leb { Λ P ≤ 2 T } ≤ ( 2 T ) e P e P ! ≤ ( 1 + 2 T ) k . (90) \operatorname{Leb}\{\Lambda_P\le2T\}\le\frac{(2T)^{e_P}}{e_P!}\le(1+2T)^k. \tag*{(90)} Leb { Λ P ≤ 2 T } ≤ e P ! ( 2 T ) e P ≤ ( 1 + 2 T ) k . (90)
The twist factors are already in F g , n F_{g,n} F g , n . With τ = R + m + 1 \tau=R+m+1 τ = R + m + 1 , Theorem 2.7, (82) , and (90) bound the integrated remainder by
g − R − 1 [ C τ ( n + 1 ) ] C τ C m ( k + 1 ) 2 k ( 1 + T ) ( 1 + 2 T ) C τ + k e 3 T ∥ H ∥ ∞ . (91) g^{-R-1}[C\tau(n+1)]^{C\tau}Cm(k+1)^{2k}(1+T)(1+2T)^{C\tau+k}e^{3T}\lVert H\rVert_\infty. \tag*{(91)} g − R − 1 [ C τ ( n + 1 ) ] C τ C m ( k + 1 ) 2 k ( 1 + T ) ( 1 + 2 T ) C τ + k e 3 T ∥ H ∥ ∞ . (91)
As k ≤ m + 2 ≤ 3 τ / 2 k\le m+2\le3\tau/2 k ≤ m + 2 ≤ 3 τ /2 , its extra factors can be absorbed by enlarging one absolute C C C . For g ≥ [ C τ ( n + 1 ) ] C g\ge[C\tau(n+1)]^{C} g ≥ [ C τ ( n + 1 ) ] C we obtain
∣ E Φ H h 0 , a 0 , k − ∑ j = 1 R g − j ξ j , n h 0 , a 0 , k ( H ) ∣ ≤ g − R − 1 [ C τ ( n + 1 ) ] C τ ( 1 + T ) C τ e 3 T ∥ H ∥ ∞ . (92) \left|\mathbb{E}\Phi_H^{h_0,a_0,k}-\sum_{j=1}^{R}g^{-j}\xi^{h_0,a_0,k}_{j,n}(H)\right|
\le g^{-R-1}[C\tau(n+1)]^{C\tau}(1+T)^{C\tau}e^{3T}\lVert H\rVert_\infty. \tag*{(92)} E Φ H h 0 , a 0 , k − j = 1 ∑ R g − j ξ j , n h 0 , a 0 , k ( H ) ≤ g − R − 1 [ C τ ( n + 1 ) ] C τ ( 1 + T ) C τ e 3 T ∥ H ∥ ∞ . (92)
The same calculation bounds ξ j \xi_j ξ j without the genus factor, using τ = j + m + 1 \tau= j + m + 1 τ = j + m + 1 . This includes R < m R < m R < m , when every retained coefficient vanishes, and R = 0 R = 0 R = 0 .
Exhaustion and the full expansion
The support of H H H bounds the complexity of every contributing carrier. This makes the nonsimple part a finite sum. We first exclude carriers without a stable exterior component at the stated genus threshold, then combine the remaining carrier expansions with the simple-curve expansion.
Theorem 3.5 (Uniform orbital expansion). The statistic (58) is absolutely WP integrable for every fixed g ≥ 3 g \ge3 g ≥ 3 , n n n , T T T , H H H . There are fixed real linear functionals κ j , n \kappa_{j,n} κ j , n , j ≥ 1 j \ge1 j ≥ 1 , and an absolute C ≥ 2 C \ge2 C ≥ 2 such that, for every integer R ≥ 0 R \ge0 R ≥ 0 and ρ = R + ⌈ T ⌉ + 1 \rho= R + \lceil T \rceil+ 1 ρ = R + ⌈ T ⌉ + 1 , whenever g ≥ [ C ρ ( n + 1 ) ] C g \ge[C\rho(n + 1)]^C g ≥ [ C ρ ( n + 1 ) ] C ,
∣ E g , n O H − Ω ( H ) − ∑ j = 1 R g − j κ j , n ( H ) ∣ ≤ g − R − 1 [ C ρ ( n + 1 ) ] C ρ ( 1 + T ) C ρ e 3 T ∥ H ∥ ∞ . (93) \left|\mathbb{E}_{g,n}\mathcal{O}_H-\Omega(H)-\sum_{j=1}^{R}g^{-j}\kappa_{j,n}(H)\right|\le g^{-R-1}[C\rho(n+1)]^{C\rho}(1+T)^{C\rho}e^{3T}\lVert H\rVert_\infty. \tag*{(93)} E g , n O H − Ω ( H ) − j = 1 ∑ R g − j κ j , n ( H ) ≤ g − R − 1 [ C ρ ( n + 1 ) ] C ρ ( 1 + T ) C ρ e 3 T ∥ H ∥ ∞ . (93) The sequence is independent of genus, order and support cutoff, and
∣ κ j , n ( H ) ∣ ≤ [ C ( j + 1 ) ( n + 1 ) ] C ( j + 1 ) ( 1 + T ) C ( j + 1 ) e 3 T ∥ H ∥ ∞ . (94) |\kappa_{j,n}(H)|\le[C(j+1)(n+1)]^{C(j+1)}(1+T)^{C(j+1)}e^{3T}\lVert H\rVert_\infty. \tag*{(94)} ∣ κ j , n ( H ) ∣ ≤ [ C ( j + 1 ) ( n + 1 ) ] C ( j + 1 ) ( 1 + T ) C ( j + 1 ) e 3 T ∥ H ∥ ∞ . (94)
Proof. First consider a nonsimple carrier with no stable exterior component. Every ambient cusp has been absorbed, and every positive boundary slot is self-paired. For some σ ≥ 0 \sigma\ge0 σ ≥ 0 its parameters satisfy
a 0 = n , k = 2 σ , g = h 0 + σ , m = 2 g + n − 2. (95) a_0=n,\qquad k=2\sigma,\qquad g=h_0+\sigma,\qquad m=2g+n-2. \tag*{(95)} a 0 = n , k = 2 σ , g = h 0 + σ , m = 2 g + n − 2. (95)
This includes σ = 0 \sigma= 0 σ = 0 , namely the whole ambient surface. By (79) , ℓ γ ≥ π ( g − 1 + n / 2 ) \ell_\gamma\ge\pi(g - 1 + n/2) ℓ γ ≥ π ( g − 1 + n /2 ) . The threshold in (93) , with C ≥ 2 C \ge2 C ≥ 2 , implies g ≥ 4 ρ 2 g \ge4\rho^2 g ≥ 4 ρ 2 , whereas T ≤ ρ − 1 T \le\rho- 1 T ≤ ρ − 1 . Hence these geodesics have length strictly greater than T T T and contribute nothing, even at the support endpoints.
Absolute integrability does not require the expansion threshold. Every nonsimple primitive has m ≥ 1 m \ge1 m ≥ 1 and length at least π / 2 \pi/2 π /2 . The ambient version of (81) bounds the number of primitives up to T T T by C ( g + n ) e T C(g+n)e^T C ( g + n ) e T . Therefore
∣ O H n o n s i m p ( X ) ∣ ≤ C ( g + n ) ( 1 + T ) e T ∥ H ∥ ∞ . (96) |\mathcal{O}^{\mathrm{nonsimp}}_H(X)|\le C(g+n)(1+T)e^T\lVert H\rVert_\infty. \tag*{(96)} ∣ O H nonsimp ( X ) ∣ ≤ C ( g + n ) ( 1 + T ) e T ∥ H ∥ ∞ . (96)
Countable length-function sums are measurable. This proves absolute integrability of the nonsimple part for every fixed genus; the simple part was handled in Lemma 3.2.
Under the expansion threshold all contributing nonsimple carriers have a stable complement, k ≥ 1 k \ge1 k ≥ 1 , and 1 ≤ m ≤ M : = ⌊ 2 T / π ⌋ 1 \le m \le M := \lfloor2T/\pi\rfloor 1 ≤ m ≤ M := ⌊ 2 T / π ⌋ . For d ≥ 0 d \ge0 d ≥ 0 let
C d ( n ) = { ( h 0 , a 0 , k ) ∈ Z ≥ 0 2 × Z ≥ 1 : a 0 ≤ n , 1 ≤ 2 h 0 + a 0 + k − 2 ≤ d } . (97) \mathcal{C}_d(n)=\left\{(h_0,a_0,k)\in\mathbb{Z}_{\ge0}^{2}\times\mathbb{Z}_{\ge1}:a_0\le n,\ 1\le2h_0+a_0+k-2\le d\right\}. \tag*{(97)} C d ( n ) = { ( h 0 , a 0 , k ) ∈ Z ≥ 0 2 × Z ≥ 1 : a 0 ≤ n , 1 ≤ 2 h 0 + a 0 + k − 2 ≤ d } . (97)
At fixed m m m there are at most ( m + 2 ) 2 (m+2)^2 ( m + 2 ) 2 types: choose h 0 h_0 h 0 , then a 0 ≤ m + 1 − 2 h 0 a_0 \le m+1-2h_0 a 0 ≤ m + 1 − 2 h 0 , and k k k is forced. Thus ∣ C d ( n ) ∣ ≤ ( d + 2 ) 3 |\mathcal{C}_d(n)| \le(d+2)^3 ∣ C d ( n ) ∣ ≤ ( d + 2 ) 3 . Reconstruction gives the exact finite identity
O H n o n s i m p ( X ) = ∑ ( h 0 , a 0 , k ) ∈ C M ( n ) O H h 0 , a 0 , k ( X ) . (98) \mathcal{O}^{\mathrm{nonsimp}}_H(X)=\sum_{(h_0,a_0,k)\in\mathcal{C}_M(n)}\mathcal{O}^{h_0,a_0,k}_H(X). \tag*{(98)} O H nonsimp ( X ) = ( h 0 , a 0 , k ) ∈ C M ( n ) ∑ O H h 0 , a 0 , k ( X ) . (98)
This sum introduces no further choices of cusp or slot labels. Define the global coefficients by the finite sums
ζ j , n ( H ) = ∑ ( h 0 , a 0 , k ) ∈ C j ( n ) ζ j , n h 0 , a 0 , k ( H ) , κ j , n = η j , n + ζ j , n . (99) \zeta_{j,n}(H)=\sum_{(h_0,a_0,k)\in\mathcal{C}_j(n)}\zeta^{h_0,a_0,k}_{j,n}(H),\qquad\kappa_{j,n}=\eta_{j,n}+\zeta_{j,n}. \tag*{(99)} ζ j , n ( H ) = ( h 0 , a 0 , k ) ∈ C j ( n ) ∑ ζ j , n h 0 , a 0 , k ( H ) , κ j , n = η j , n + ζ j , n . (99)
If M > j M > j M > j , the extra types in M ( n ) \mathcal{M}(n) M ( n ) have zero coefficient at order j j j ; if j > M j > M j > M , the extra types in C j ( n ) \mathcal{C}_j(n) C j ( n ) have Ψ H = 0 \Psi_H = 0 Ψ H = 0 . These are the two vanishing properties in (89) . Hence the coefficient sum over M ( n ) \mathcal{M}(n) M ( n ) always equals ζ j , n ( H ) \zeta_{j,n}(H) ζ j , n ( H ) . Thus the same coefficients work for every support cutoff.
Apply (92) at order R R R to every type in (98) . Its parameter satisfies R + m + 1 ≤ R + ⌈ T ⌉ + 1 = ρ R + m + 1 \le R + \lceil T\rceil+ 1 = \rho R + m + 1 ≤ R + ⌈ T ⌉ + 1 = ρ . The common threshold therefore applies, including types with m > R m > R m > R , which remain in the remainder. The at most ( M + 2 ) 3 ≤ ( 2 ρ ) 3 (M + 2)^3 \le(2\rho)^3 ( M + 2 ) 3 ≤ ( 2 ρ ) 3 errors are absorbed in the constant in (93) . If M = 0 M = 0 M = 0 the nonsimple statistic and all its coefficients on H H H vanish. Adding (78) proves (93) . For (94) , sum the coefficient estimate following (92) over m ≤ j m \le j m ≤ j , so that j + m + 1 ≤ 2 ( j + 1 ) j + m + 1 \le2(j + 1) j + m + 1 ≤ 2 ( j + 1 ) , and absorb the cubic type count.
Geometric moments with growing order
We next apply the orbital expansion to convolution powers of the fixed test from Section 1.2. For its kernel h = b 0 ∗ b 0 h = b_0 * b_0 h = b 0 ∗ b 0 and Fourier multiplier f = h ^ f = \widehat{h} f = h , with supp h ⊂ [ − L , L ] \operatorname{supp} h \subset[-L,L] supp h ⊂ [ − L , L ] , define for p ( x ) = ∑ t ≥ 1 a t x t p(x) = \sum_{t\ge1} a_t x^t p ( x ) = ∑ t ≥ 1 a t x t
H p = ( p ∘ f ˇ ) = ∑ t ≥ 1 a t h ∗ t , O X ( p ) = O H p ( X ) , H_p = (p \circ\check{f}) = \sum_{t\ge1} a_t h^{*t}, \qquad\mathcal{O}_X(p) = \mathcal{O}_{H_p}(X), H p = ( p ∘ f ˇ ) = t ≥ 1 ∑ a t h ∗ t , O X ( p ) = O H p ( X ) ,
ω 1 , n ( p ) = Ω ( H p ) , ω j , n ( p ) = κ j − 1 , n ( H p ) ( j ≥ 2 ) . (100) \omega_{1,n}(p) = \Omega(H_p), \qquad\omega_{j,n}(p) = \kappa_{j-1,n}(H_p) \quad(j \ge2). \tag*{(100)} ω 1 , n ( p ) = Ω ( H p ) , ω j , n ( p ) = κ j − 1 , n ( H p ) ( j ≥ 2 ) . (100)
The coefficient functionals ω j , n \omega_{j,n} ω j , n are fixed linear functionals on R [ x ] \mathbb{R}[x] R [ x ] . Since ω 1 , n \omega_{1,n} ω 1 , n is independent of n n n , write it as ω 1 \omega_1 ω 1 .
Proposition 3.6 (Uniform geometric moments). There is one integer c = c f ≥ 2 c = c_f \ge2 c = c f ≥ 2 , depending only on the fixed multiplier, such that, simultaneously for every n ≥ 0 n \ge0 n ≥ 0 , integers q > t ≥ 1 q > t \ge1 q > t ≥ 1 , and g ≥ [ c q ( n + 1 ) ] c g \ge[cq(n + 1)]^c g ≥ [ c q ( n + 1 ) ] c ,
∣ g − 1 E g , n O X ( x t ) − ∑ j = 1 q − 1 g − j ω j , n ( x t ) ∣ ≤ g − q [ c q ( n + 1 ) ] c q . (101) \left|g^{-1}\mathbb{E}_{g,n}\mathcal{O}_X(x^t) - \sum_{j=1}^{q-1} g^{-j}\omega_{j,n}(x^t)\right| \le g^{-q}[cq(n + 1)]^{cq}. \tag*{(101)} g − 1 E g , n O X ( x t ) − j = 1 ∑ q − 1 g − j ω j , n ( x t ) ≤ g − q [ c q ( n + 1 ) ] c q . (101)
Proof. The convolution support rule and Young’s inequality give supp ( h ∗ t ) ⊂ [ − t L , t L ] \operatorname{supp}(h^{*t}) \subset[-tL,tL] supp ( h ∗ t ) ⊂ [ − t L , t L ] and ∥ h ∗ t ∥ ∞ ≤ M f t \|h^{*t}\|_\infty\le M_f^t ∥ h ∗ t ∥ ∞ ≤ M f t , where M f = max { 1 , ∥ h ∥ 1 , ∥ h ∥ ∞ } M_f = \max\{1,\|h\|_1,\|h\|_\infty\} M f = max { 1 , ∥ h ∥ 1 , ∥ h ∥ ∞ } is fixed. In Theorem 3.5 take R = q − 2 R = q - 2 R = q − 2 and T = t L T = tL T = t L . Then
ρ = q − 1 + ⌈ t L ⌉ ≤ ( 1 + L ) q , 1 + t L ≤ ( 1 + L ) q . (102) \rho= q - 1 + \lceil tL\rceil\le(1 + L)q, \qquad1 + tL \le(1 + L)q. \tag*{(102)} ρ = q − 1 + ⌈ t L ⌉ ≤ ( 1 + L ) q , 1 + t L ≤ ( 1 + L ) q . (102)
After division by g g g , the remainder is bounded by g − q [ C ρ ( n + 1 ) ] C ρ ( 1 + t L ) C ρ e 3 t L M f t g^{-q}[C\rho(n + 1)]^{C\rho}(1 + tL)^{C\rho}e^{3tL}M_f^t g − q [ C ρ ( n + 1 ) ] C ρ ( 1 + t L ) C ρ e 3 t L M f t . The factor multiplying g − q g^{-q} g − q has logarithm O f ( q log ( q ( n + 1 ) ) ) O_f(q\log(q(n + 1))) O f ( q log ( q ( n + 1 ))) . Choosing the integer c c c sufficiently large therefore gives (101) and the stated genus threshold. In particular, the case q = 2 q = 2 q = 2 , t = 1 t = 1 t = 1 uses R = 0 R = 0 R = 0 and gives a first-moment bound.
The exact first orbital correction
The coefficient κ 1 , n \kappa_{1,n} κ 1 , n of g − 1 g^{-1} g − 1 in Theorem 3.5 is polynomial in n n n of degree at most two. We compute its nonseparating part, combine it with the separating coefficient (63) , and enumerate the five contributing carrier configurations. The resulting formula separates the constant, linear and quadratic dependence on n n n . We retain the definitions of L k L_k L k and I H h 0 , a 0 , k I_H^{h_0,a_0,k} I H h 0 , a 0 , k from (76) and (80) , including both orientations, all repetitions and the orbifold volume convention. Write
ϕ ( x ) = sinh ( x / 2 ) x / 2 , c ( x ) = cosh ( x / 2 ) , \phi(x) = \frac{\sinh(x/2)}{x/2}, \qquad c(x) = \cosh(x/2), ϕ ( x ) = x /2 sinh ( x /2 ) , c ( x ) = cosh ( x /2 ) ,
w ( x ) = x ϕ ( x ) = 2 sinh ( x / 2 ) , v ( x ) = x 2 + 4 π 2 48 . (103) w(x) = x\phi(x) = 2\sinh(x/2), \qquad v(x) = \frac{x^2 + 4\pi^2}{48}. \tag*{(103)} w ( x ) = x ϕ ( x ) = 2 sinh ( x /2 ) , v ( x ) = 48 x 2 + 4 π 2 . (103)
Thus v = V 1 , 1 v = V_{1,1} v = V 1 , 1 in our orbifold convention, and ϕ ( 0 ) = 1 \phi(0) = 1 ϕ ( 0 ) = 1 .
Lemma 3.7 (First nonseparating coefficient). For every integer n ≥ 0 n \ge0 n ≥ 0 and x ≥ 0 x \ge0 x ≥ 0 ,
u 1 , n ( x ) = ( 2 − n ) ϕ ( x ) ( c ( x ) + 1 ) − c ( x ) 2 π 2 . (104) u_{1,n}(x) = \frac{(2-n)\phi(x)(c(x)+1)-c(x)^2}{\pi^2}. \tag*{(104)} u 1 , n ( x ) = π 2 ( 2 − n ) ϕ ( x ) ( c ( x ) + 1 ) − c ( x ) 2 . (104)
Proof. Mirzakhani–Zograf’s first handle-ratio coefficient [19 Theorem 1.4, (1.4)] is p n 1 = ( 3 − 2 n ) / π 2 p^1_n=(3-2n)/\pi^2 p n 1 = ( 3 − 2 n ) / π 2 , including n = 0 , 1 n=0,1 n = 0 , 1 . For fixed p ≥ 1 p \ge1 p ≥ 1 and d ∈ Z ≥ 0 p \mathbf{d} \in\mathbb{Z}_{\ge0}^p d ∈ Z ≥ 0 p , put q = ∣ d ∣ q=|\mathbf{d}| q = ∣ d ∣ , r = # { i : d i > 0 } r=\#\{i:d_i>0\} r = # { i : d i > 0 } and S 2 = ∑ i d i 2 S_2=\sum_i d_i^2 S 2 = ∑ i d i 2 . The normalization in [7 (B.3)] agrees with (35) . The pair and single sums in [7 Proposition B.9] simplify to
A = 2 ( q 2 − S 2 ) + 2 ( p − 1 ) q − p r + r 2 + r 2 , B = 3 2 q − S 2 − r . A=2(q^2-S_2)+2(p-1)q-pr+\frac{r^2+r}{2},\qquad B=\frac{3}{2}q-S_2-r. A = 2 ( q 2 − S 2 ) + 2 ( p − 1 ) q − p r + 2 r 2 + r , B = 2 3 q − S 2 − r .
By [19 Theorem 1.4] , both V G , p − 1 / V G , p V_{G,p-1}/V_{G,p} V G , p − 1 / V G , p and V G − 1 , p + 1 / V G , p V_{G-1,p+1}/V_{G,p} V G − 1 , p + 1 / V G , p equal ( 8 π 2 G ) − 1 + O p ( G − 2 ) (8\pi^2G)^{-1}+O_p(G^{-2}) ( 8 π 2 G ) − 1 + O p ( G − 2 ) . Applying that proposition with p , d p,\mathbf{d} p , d fixed as G → ∞ G\to\infty G → ∞ therefore gives π 2 b d , p 1 = − A / 2 + B \pi^2b^1_{\mathbf{d},p}=-A/2+B π 2 b d , p 1 = − A /2 + B , hence
b d , p 1 = − q 2 + ( p − 5 2 ) q − r ( 2 p − r − 5 ) / 4 π 2 . (105) b^1_{\mathbf{d},p}=-\frac{q^2+(p-\frac{5}{2})q-r(2p-r-5)/4}{\pi^2}. \tag*{(105)} b d , p 1 = − π 2 q 2 + ( p − 2 5 ) q − r ( 2 p − r − 5 ) /4 . (105)
The pair sum is empty when p = 1 p=1 p = 1 . For the all-zero multi-index this gives zero, as the normalized intersection number is identically one. These coefficients agree with those of Proposition 2.3 by uniqueness of asymptotic coefficients at each fixed marked-point count and multi-index. All genera in this identification are large, so no terminal-volume convention enters.
Set z = x / 2 z=x/2 z = x /2 , F ( z ) = sinh z / z F(z)=\sinh z/z F ( z ) = sinh z / z and D = ( z / 2 ) d / d z D=(z/2)\,d/dz D = ( z /2 ) d / d z . In the series for B 1 , n + 2 ( 2 ) ( x , x ) B^{(2)}_{1,n+2}(x,x) B 1 , n + 2 ( 2 ) ( x , x ) , the weights are a d ( z ) = z 2 d / ( 2 d + 1 ) ! a_d(z)=z^{2d}/(2d+1)! a d ( z ) = z 2 d / ( 2 d + 1 )! and q = d 1 + d 2 q=d_1+d_2 q = d 1 + d 2 , r = 1 d 1 > 0 + 1 d 2 > 0 r=\mathbf{1}_{d_1>0}+\mathbf{1}_{d_2>0} r = 1 d 1 > 0 + 1 d 2 > 0 . All series below and their differentiated series converge locally uniformly. Their sums satisfy
∑ d 1 , d 2 ≥ 0 q a d 1 a d 2 = D ( F 2 ) , ∑ d 1 , d 2 ≥ 0 q 2 a d 1 a d 2 = D 2 ( F 2 ) , ∑ d 1 , d 2 ≥ 0 r a d 1 a d 2 = 2 F ( F − 1 ) , ∑ d 1 , d 2 ≥ 0 r 2 a d 1 a d 2 = 2 F ( F − 1 ) + 2 ( F − 1 ) 2 . (106) \begin{aligned}
\sum_{d_1,d_2\ge0}q\,a_{d_1}a_{d_2}&=D(F^2),&
\sum_{d_1,d_2\ge0}q^2a_{d_1}a_{d_2}&=D^2(F^2),\\
\sum_{d_1,d_2\ge0}r\,a_{d_1}a_{d_2}&=2F(F-1),&
\sum_{d_1,d_2\ge0}r^2a_{d_1}a_{d_2}&=2F(F-1)+2(F-1)^2. \tag*{(106)}
\end{aligned} d 1 , d 2 ≥ 0 ∑ q a d 1 a d 2 d 1 , d 2 ≥ 0 ∑ r a d 1 a d 2 = D ( F 2 ) , = 2 F ( F − 1 ) , d 1 , d 2 ≥ 0 ∑ q 2 a d 1 a d 2 d 1 , d 2 ≥ 0 ∑ r 2 a d 1 a d 2 = D 2 ( F 2 ) , = 2 F ( F − 1 ) + 2 ( F − 1 ) 2 . (106)
Substituting p = n + 2 p=n+2 p = n + 2 into (105) gives
B 1 , n + 2 ( 2 ) ( x , x ) = − D 2 ( F 2 ) + ( n − 1 2 ) D ( F 2 ) − ( n − 1 ) F ( F − 1 ) + 1 2 ( F − 1 ) 2 π 2 . (107) B^{(2)}_{1,n+2}(x,x)=-\frac{D^2(F^2)+(n-\frac{1}{2})D(F^2)-(n-1)F(F-1)+\frac{1}{2}(F-1)^2}{\pi^2}. \tag*{(107)} B 1 , n + 2 ( 2 ) ( x , x ) = − π 2 D 2 ( F 2 ) + ( n − 2 1 ) D ( F 2 ) − ( n − 1 ) F ( F − 1 ) + 2 1 ( F − 1 ) 2 . (107)
By (65) , u 1 , n = p n 1 F 2 + B 1 , n + 2 ( 2 ) u_{1,n}=p^1_nF^2+B^{(2)}_{1,n+2} u 1 , n = p n 1 F 2 + B 1 , n + 2 ( 2 ) ; the shift from g − 1 g-1 g − 1 to g g g has no additional first-order term. The identities D ( F 2 ) = F cosh z − F 2 D(F^2)=F\cosh z-F^2 D ( F 2 ) = F cosh z − F 2 and D 2 ( F 2 ) = 1 2 cosh 2 z − 3 2 F cosh z + ( 1 + 1 2 z 2 ) F 2 D^2(F^2)=\frac{1}{2}\cosh^2z-\frac{3}{2}F\cosh z+(1+\frac{1}{2}z^2)F^2 D 2 ( F 2 ) = 2 1 cosh 2 z − 2 3 F cosh z + ( 1 + 2 1 z 2 ) F 2 , together with z 2 F 2 = cosh 2 z − 1 z^2F^2=\cosh^2z-1 z 2 F 2 = cosh 2 z − 1 , yield (104) . Continuity gives the value u 1 , n ( 0 ) = ( 3 − 2 n ) / π 2 u_{1,n}(0)=(3-2n)/\pi^2 u 1 , n ( 0 ) = ( 3 − 2 n ) / π 2 .
Define three kernels independent of n n n :
k 0 ( x ) = 2 ϕ ( x ) ( c ( x ) + 1 ) − c ( x ) 2 π 2 + ϕ ( x ) v ( x ) 4 π 2 , k_0(x)=\frac{2\phi(x)(c(x)+1)-c(x)^2}{\pi^2}+\frac{\phi(x)v(x)}{4\pi^2}, k 0 ( x ) = π 2 2 ϕ ( x ) ( c ( x ) + 1 ) − c ( x ) 2 + 4 π 2 ϕ ( x ) v ( x ) ,
k 1 ( x ) = − ϕ ( x ) ( c ( x ) + 1 ) π 2 , k 2 ( x ) = ϕ ( x ) 4 π 2 . (108) k_1(x)=-\frac{\phi(x)(c(x)+1)}{\pi^2},\qquad k_2(x)=\frac{\phi(x)}{4\pi^2}. \tag*{(108)} k 1 ( x ) = − π 2 ϕ ( x ) ( c ( x ) + 1 ) , k 2 ( x ) = 4 π 2 ϕ ( x ) . (108)
For the nonsimple contribution, write I t ( x ) = I H 1 , 0 , 1 ( x ) I_t(x)=I_H^{1,0,1}(x) I t ( x ) = I H 1 , 0 , 1 ( x ) and I a ( x ) = I H 0 , a , 3 − a ( x ) I_a(x)=I_H^{0,a,3-a}(x) I a ( x ) = I H 0 , a , 3 − a ( x ) for a = 0 , 1 , 2 a=0,1,2 a = 0 , 1 , 2 . Their cusp labels are intrinsic labels on the fixed carrier; ambient cusp choices will appear only in the binomial factors below. Set
Z 0 ( H ) = 1 8 π 2 [ ∫ 0 ∞ w ( x ) I 0 ( x ) d x + 1 6 ∫ ( 0 , ∞ ) 3 w ( x 1 ) w ( x 2 ) w ( x 3 ) I 0 ( x 1 , x 2 , x 3 ) d x + 1 2 ∫ ( 0 , ∞ ) 2 u w ( y ) I 0 ( u , u , y ) d u d y ] (109) \mathcal{Z}_0(H)=\frac{1}{8\pi^2}\left[\int_0^\infty w(x)I_0(x)\,dx+\frac{1}{6}\int_{(0,\infty)^3}w(x_1)w(x_2)w(x_3)I_0(x_1,x_2,x_3)\,dx+\frac{1}{2}\int_{(0,\infty)^2}u\,w(y)I_0(u,u,y)\,du\,dy\right] \tag*{(109)} Z 0 ( H ) = 8 π 2 1 [ ∫ 0 ∞ w ( x ) I 0 ( x ) d x + 6 1 ∫ ( 0 , ∞ ) 3 w ( x 1 ) w ( x 2 ) w ( x 3 ) I 0 ( x 1 , x 2 , x 3 ) d x + 2 1 ∫ ( 0 , ∞ ) 2 u w ( y ) I 0 ( u , u , y ) d u d y ] (109)
Z 1 ( H ) = 1 16 π 2 ∫ ( 0 , ∞ ) 2 w ( x 1 ) w ( x 2 ) I 1 ( x 1 , x 2 ) d x 1 d x 2 (110) \mathcal{Z}_1(H)=\frac{1}{16\pi^2}\int_{(0,\infty)^2}w(x_1)w(x_2)I_1(x_1,x_2)\,dx_1\,dx_2 \tag*{(110)} Z 1 ( H ) = 16 π 2 1 ∫ ( 0 , ∞ ) 2 w ( x 1 ) w ( x 2 ) I 1 ( x 1 , x 2 ) d x 1 d x 2 (110)
Z 2 ( H ) = 1 8 π 2 ∫ 0 ∞ w ( x ) I 2 ( x ) d x . (111) \mathcal{Z}_2(H)=\frac{1}{8\pi^2}\int_0^\infty w(x)I_2(x)\,dx. \tag*{(111)} Z 2 ( H ) = 8 π 2 1 ∫ 0 ∞ w ( x ) I 2 ( x ) d x . (111)
Proposition 3.8 (Exact quadratic cusp dependence). For every compactly supported bounded even Borel H H H , the integrals (109) –(111) converge absolutely. The fixed linear functionals
Λ i ( H ) = L k i ( H ) + Z i ( H ) , 0 ≤ i ≤ 2 , (112) \Lambda_i(H)=\mathcal{L}_{k_i}(H)+\mathcal{Z}_i(H),\qquad0\le i\le2, \tag*{(112)} Λ i ( H ) = L k i ( H ) + Z i ( H ) , 0 ≤ i ≤ 2 , (112) satisfy, for every integer n ≥ 0 n\ge0 n ≥ 0 ,
κ 1 , n ( H ) = Λ 0 ( H ) + n Λ 1 ( H ) + ( n 2 ) Λ 2 ( H ) . (113) \kappa_{1,n}(H)=\Lambda_0(H)+n\Lambda_1(H)+\binom{n}{2}\Lambda_2(H). \tag*{(113)} κ 1 , n ( H ) = Λ 0 ( H ) + n Λ 1 ( H ) + ( 2 n ) Λ 2 ( H ) . (113) There is an absolute C C C such that, if supp H ⊂ [ − T , T ] \operatorname{supp}H\subset[-T,T] supp H ⊂ [ − T , T ] ,
∑ i = 0 2 ∣ Λ i ( H ) ∣ ≤ C ( 1 + T ) 4 e 2 T ∥ H ∥ ∞ , ∣ κ 1 , n ( H ) ∣ ≤ C ( n + 1 ) 2 ( 1 + T ) 4 e 2 T ∥ H ∥ ∞ . (114) \sum_{i=0}^{2}|\Lambda_i(H)|\le C(1+T)^4e^{2T}\lVert H\rVert_\infty,\qquad|\kappa_{1,n}(H)|\le C(n+1)^2(1+T)^4e^{2T}\lVert H\rVert_\infty. \tag*{(114)} i = 0 ∑ 2 ∣ Λ i ( H ) ∣ ≤ C ( 1 + T ) 4 e 2 T ∥ H ∥ ∞ , ∣ κ 1 , n ( H ) ∣ ≤ C ( n + 1 ) 2 ( 1 + T ) 4 e 2 T ∥ H ∥ ∞ . (114) If H H H is additionally continuous, nonnegative and not identically zero, then
Λ 2 ( H ) ≥ 1 4 π 2 ∫ 0 ∞ x H ( x ) d x > 0. (115) \Lambda_2(H)\ge\frac{1}{4\pi^2}\int_0^\infty xH(x)\,dx>0. \tag*{(115)} Λ 2 ( H ) ≥ 4 π 2 1 ∫ 0 ∞ x H ( x ) d x > 0. (115) Thus for such a test the degree in n n n is exactly two.
Proof. Lemma 3.7 and (63) give u 1 , n + s 1 , n = k 0 + n k 1 + ( n 2 ) k 2 u_{1,n}+s_{1,n}=k_0+nk_1+\binom{n}{2}k_2 u 1 , n + s 1 , n = k 0 + n k 1 + ( 2 n ) k 2 . Linearity of (76) identifies the simple contribution.
For the nonsimple contribution, (53) at j = 1 j=1 j = 1 forces m = b = 1 m=b=1 m = b = 1 , with no small complementary components. The possible carrier types are exactly ( 1 , 0 , 1 ) (1,0,1) ( 1 , 0 , 1 ) and ( 0 , a , 3 − a ) (0,a,3-a) ( 0 , a , 3 − a ) for a = 0 , 1 , 2 a=0,1,2 a = 0 , 1 , 2 . For a surviving pattern, (57) reduces to
F 1 , n h 0 , a 0 , k ; P = ( n a 0 ) 8 π 2 J P V h 0 , a 0 + k ( 0 a 0 , x P ) ∏ unpaired slots j ϕ ( x j ) . (116) \mathcal{F}_{1,n}^{h_0,a_0,k;P}=\frac{\binom{n}{a_0}}{8\pi^2}J_PV_{h_0,a_0+k}(\mathbf{0}_{a_0},\mathbf{x}_P)\prod_{\text{unpaired slots }j}\phi(x_j). \tag*{(116)} F 1 , n h 0 , a 0 , k ; P = 8 π 2 ( a 0 n ) J P V h 0 , a 0 + k ( 0 a 0 , x P ) unpaired slots j ∏ ϕ ( x j ) . (116)
Multiplication by Ψ H = I H / V \Psi_H=I_H/V Ψ H = I H / V cancels only the displayed carrier volume; it does not alter its intrinsic orbifold measure.
After extracting the common factor ( 8 π 2 ) − 1 (8\pi^2)^{-1} ( 8 π 2 ) − 1 , the cusp-label and slot-symmetry weights are
Table 2.
The self-paired pants row combines the three choices of a pair, giving 3 / 3 ! = 1 / 2 3/3! = 1/2 3/3 ! = 1/2 . Its slot lengths are ( u , u , y ) (u,u,y) ( u , u , y ) and its actual-edge weight is u y uy u y . The self-edge contributes its twist length u u u , while only the external edge contributes ϕ ( y ) \phi(y) ϕ ( y ) , giving the length weight u w ( y ) u\,w(y) u w ( y ) . The torus convention is already accounted for in (86) . Pairing both slots of the one-cusp pants would leave b = 0 b=0 b = 0 and is excluded. These are all the patterns, so ζ 1 , n = Z 0 + n Z 1 + ( n 2 ) Z 2 \zeta_{1,n}=\mathcal{Z}_0+n\mathcal{Z}_1+\binom{n}{2}\mathcal{Z}_2 ζ 1 , n = Z 0 + n Z 1 + ( 2 n ) Z 2 . This proves (113) , including n = 0 , 1 n=0,1 n = 0 , 1 , where unavailable cusp-labelled carriers have zero binomial weight.
It remains to justify absolute convergence and the support bound (114) . For complexity one, (82) gives ∣ I a ∣ ≤ C ( 1 + T ) e T ∥ H ∥ ∞ |I_a|\le C(1+T)e^T\lVert H\rVert_\infty ∣ I a ∣ ≤ C ( 1 + T ) e T ∥ H ∥ ∞ for pants and ∣ I t ( x ) ∣ ≤ C ( 1 + x 2 ) ( 1 + T ) e T ∥ H ∥ ∞ |I_t(x)|\le C(1+x^2)(1+T)e^T\lVert H\rVert_\infty ∣ I t ( x ) ∣ ≤ C ( 1 + x 2 ) ( 1 + T ) e T ∥ H ∥ ∞ for the torus. The integrands vanish unless the sum of carrier boundary lengths is at most 2 T 2T 2 T . Thus the unpaired domains have ∑ x i ≤ 2 T \sum x_i\le2T ∑ x i ≤ 2 T , and the self-paired domain has 2 u + y ≤ 2 T 2u+y\le2T 2 u + y ≤ 2 T . Since w ( x ) ≤ e x / 2 w(x)\le e^{x/2} w ( x ) ≤ e x /2 , the product of external weights is at most e T e^T e T . The unpaired domains have dimension at most three; the torus has one integration variable and a degree-two volume factor; the self-paired domain has dimension two and the extra factor u ≤ T u\le T u ≤ T . In every case the bound is C ( 1 + T ) 2 e 2 T ∥ H ∥ ∞ C(1+T)^2e^{2T}\lVert H\rVert_\infty C ( 1 + T ) 2 e 2 T ∥ H ∥ ∞ . It also covers T = 0 T=0 T = 0 , when every integral is zero.
The explicit kernels satisfy ∣ k i ( x ) ∣ ≤ C ( 1 + x ) 2 e x |k_i(x)|\le C(1+x)^2e^x ∣ k i ( x ) ∣ ≤ C ( 1 + x ) 2 e x . The bound following (76) therefore gives ∣ L k i ( H ) ∣ ≤ C ( 1 + T ) 4 e T ∥ H ∥ ∞ |\mathcal{L}_{k_i}(H)|\le C(1+T)^4e^T\lVert H\rVert_\infty ∣ L k i ( H ) ∣ ≤ C ( 1 + T ) 4 e T ∥ H ∥ ∞ , including the absolute sum of all repetitions. This proves (114) . Finally, for nonnegative H H H both Z 2 ( H ) \mathcal{Z}_2(H) Z 2 ( H ) and every repetition in L k 2 ( H ) \mathcal{L}_{k_2}(H) L k 2 ( H ) are nonnegative. Its ν = 1 \nu=1 ν = 1 contribution is exactly ( 4 π 2 ) − 1 ∫ 0 ∞ x H ( x ) d x (4\pi^2)^{-1}\int_0^\infty xH(x)\,dx ( 4 π 2 ) − 1 ∫ 0 ∞ x H ( x ) d x . Since H H H is continuous, even, and not identically zero, this integral is positive, proving (115) .
Cusp polynomiality at every fixed order
The preceding quadratic formula is the first case of a degree bound for every orbital coefficient. In a retained record of deficit δ \delta δ , the cusp-label factor has degree at most δ + 1 \delta+1 δ + 1 , while the shifted coefficient at order j − δ j-\delta j − δ has degree at most j − δ j-\delta j − δ . Their product therefore has degree at most j + 1 j+1 j + 1 . We establish these bounds and justify integration of the coefficient polynomials below. Finite interpolation then gives the binomial-basis formula for the canonical orbital coefficients. For a sequence f ( n ) f(n) f ( n ) write Δ f ( n ) = f ( n + 1 ) − f ( n ) \Delta f(n)=f(n+1)-f(n) Δ f ( n ) = f ( n + 1 ) − f ( n ) ; Δ n r κ j , 0 \Delta_n^r\kappa_{j,0} Δ n r κ j , 0 below denotes the r r r th forward difference of n ↦ κ j , n n\mapsto\kappa_{j,n} n ↦ κ j , n evaluated at n = 0 n=0 n = 0 . Put ( z ) a = z ( z − 1 ) ⋯ ( z − a + 1 ) (z)_a=z(z-1)\cdots(z-a+1) ( z ) a = z ( z − 1 ) ⋯ ( z − a + 1 ) , with ( z ) 0 = 1 (z)_0=1 ( z ) 0 = 1 . Here polynomial extension always means extension of a coefficient, not a volume with an inadmissible number of marked points.
Lemma 3.9 (Polynomial boundary and shifted coefficients). Fix integers s , ℓ ≥ 0 s,\ell\ge0 s , ℓ ≥ 0 and put p 0 = max { 1 , ℓ } p_0=\max\{1,\ell\} p 0 = max { 1 , ℓ } . The functions B s , p ( ℓ ) B^{(\ell)}_{s,p} B s , p ( ℓ ) , for integers p ≥ p 0 p\ge p_0 p ≥ p 0 , are the values of a polynomial in p p p of degree at most s s s , with continuous boundary-length functions as coefficients. With h p 0 = p p 0 = 1 h_p^0=p_p^0=1 h p 0 = p p 0 = 1 , the scalar coefficients h p s , p p s h_p^s,p_p^s h p s , p p s are polynomials in p p p of degree at most s s s on all integers p ≥ 0 p\ge0 p ≥ 0 . For fixed integers H 0 ≥ 0 H_0\ge0 H 0 ≥ 0 , δ ≥ 1 \delta\ge1 δ ≥ 1 and ℓ ≥ 0 \ell\ge0 ℓ ≥ 0 , Q s , H 0 ℓ , δ Q^{\ell,\delta}_{s,H_0} Q s , H 0 ℓ , δ likewise has a unique polynomial extension in n n n of degree at most s s s , agreeing with the canonical function at every integer n n n satisfying
n ≥ 0 , p = n + 2 H 0 − δ ≥ max { 1 , ℓ } . n\ge0,\qquad p=n+2H_0-\delta\ge\max\{1,\ell\}. n ≥ 0 , p = n + 2 H 0 − δ ≥ max { 1 , ℓ } .
Proof. Lemma 4.7(i) of [19 ] fixes any ℓ ≥ 1 \ell\ge1 ℓ ≥ 1 nonnegative descendant indices and pads with zeros. It states that the coefficient of order s s s is polynomial in the total marked count p ≥ ℓ p\ge\ell p ≥ ℓ , of degree at most s s s . By uniqueness of fixed-parameter asymptotic coefficients, these are the b ( 0 , d ) , p s b^s_{(0,d),p} b ( 0 , d ) , p s in Proposition 2.3; symmetry permits placing the padded zeros before the displayed indices.
For 0 ≤ i ≤ s 0\le i\le s 0 ≤ i ≤ s define the Lagrange polynomials
L i ( p ) = ∏ 0 ≤ r ≤ s r ≠ i p − ( p 0 + r ) i − r . L_i(p)=\prod_{\substack{0\le r\le s\\r\ne i}}\frac{p-(p_0+r)}{i-r}. L i ( p ) = 0 ≤ r ≤ s r = i ∏ i − r p − ( p 0 + r ) .
For each fixed multi-index the source statement gives its exact interpolation identity at p 0 , … , p 0 + s p_0,\ldots,p_0+s p 0 , … , p 0 + s . Multiply by the weights in (40) and sum. All s + 1 s+1 s + 1 node series converge locally uniformly and absolutely by (34) and (39) . At every admissible integer p p p the resulting series is the canonical one. Thus
B s , p ( ℓ ) ( x ) = ∑ i = 0 s L i ( p ) B s , p 0 + i ( ℓ ) ( x ) ( p ≥ p 0 ) . (117) B^{(\ell)}_{s,p}(\mathbf{x})=\sum_{i=0}^{s}L_i(p)B^{(\ell)}_{s,p_0+i}(\mathbf{x}) \qquad(p\ge p_0). \tag*{(117)} B s , p ( ℓ ) ( x ) = i = 0 ∑ s L i ( p ) B s , p 0 + i ( ℓ ) ( x ) ( p ≥ p 0 ) . (117)
The right side supplies the continuous function-valued polynomial extension. This interchange involves only the s + 1 s+1 s + 1 absolutely convergent node series. If ℓ = 0 \ell=0 ℓ = 0 , the normalized volume equals one: B 0 , p ( 0 ) = 1 B^{(0)}_{0,p}=1 B 0 , p ( 0 ) = 1 and B s , p ( 0 ) = 0 B^{(0)}_{s,p}=0 B s , p ( 0 ) = 0 for s ≥ 1 s\ge1 s ≥ 1 , which proves that case directly.
For the scalar ratios use [19 Theorem 4.1(ii), (4.2)–(4.3), and Lemma 4.7(ii)] . Their domain explicitly includes p = 0 p=0 p = 0 . Equation (125) is our H G , p H_{G,p} H G , p , whereas (126) is the reciprocal of our G P , p G_{P,p} G P , p . If a ( t ) = 1 + ∑ r ≥ 1 a r ( p ) t r a(t)=1+\sum_{r\ge1}a_r(p)t^r a ( t ) = 1 + ∑ r ≥ 1 a r ( p ) t r has deg p a r ≤ r \deg_p a_r\le r deg p a r ≤ r , its formal inverse has coefficients
c 0 = 1 , c s = − ∑ r = 1 s a r c s − r , c_0=1,\qquad c_s=-\sum_{r=1}^{s}a_r c_{s-r}, c 0 = 1 , c s = − r = 1 ∑ s a r c s − r ,
and hence deg p c s ≤ s \deg_p c_s\le s deg p c s ≤ s by induction. This establishes the asserted degree bound for both scalar sequences, including p = 0 , 1 p=0,1 p = 0 , 1 , after identifying coefficients by uniqueness.
The same degree rule is preserved by finite products, affine translations of p p p , and fixed genus shifts. In fact, the order-q q q coefficient after shifting g g g to g − H 0 g-H_0 g − H 0 is
∑ r = 1 q ( q − 1 r − 1 ) H 0 q − r a r ( p ) , \sum_{r=1}^{q}\binom{q-1}{r-1}H_0^{q-r}a_r(p), r = 1 ∑ q ( r − 1 q − 1 ) H 0 q − r a r ( p ) ,
whose degree is at most q q q . In (43) , the order-q q q coefficient of each geometric denominator factor is [ − ( n − δ + j − 2 ) / 2 ] q [-(n-\delta+j-2)/2]^q [ − ( n − δ + j − 2 ) /2 ] q , of degree at most q q q in n n n . The numbers of factors depend only on H 0 , δ H_0,\delta H 0 , δ . Consequently deg n v q ( n , H 0 , δ ) ≤ q \deg_n v_q(n,H_0,\delta)\le q deg n v q ( n , H 0 , δ ) ≤ q . The convolution defining Q s Q_s Q s in the proof of Lemma 2.6 , with p = n + 2 H 0 − δ p=n+2H_0-\delta p = n + 2 H 0 − δ , has degree at most s s s . It agrees with the canonical coefficients at every admissible n n n . There are infinitely many such integers, proving uniqueness. □ \square □
Lemma 3.10 (Polynomial retained kernels). For every j ≥ 1 j\ge1 j ≥ 1 , u j , n u_{j,n} u j , n and s j , n s_{j,n} s j , n are polynomials in n n n with continuous length-dependent coefficients, of degrees at most j j j and j + 1 j+1 j + 1 , respectively. For a fixed carrier type ( h 0 , a 0 , k ) (h_0,a_0,k) ( h 0 , a 0 , k ) and pattern P P P , F j , n h 0 , a 0 , k ; P F^{h_0,a_0,k;P}_{j,n} F j , n h 0 , a 0 , k ; P has a continuous function-valued polynomial extension of degree at most j + 1 j+1 j + 1 . On nonnegative integers n < a 0 n<a_0 n < a 0 this extension is zero.
Proof. The formula (65) and Lemma 3.9 give deg n β r , n ≤ r \deg_n\beta_{r,n}\le r deg n β r , n ≤ r and deg n p r n ≤ r \deg_n p^n_r\le r deg n p r n ≤ r , so deg n u j , n ≤ j \deg_n u_{j,n}\le j deg n u j , n ≤ j .
For separating curves remove the condition a 0 ≤ n a_0\le n a 0 ≤ n from (71) , obtaining the finite set
E δ u n i v = { ( h 0 , a 0 ) ∈ Z ≥ 0 2 : 2 h 0 + a 0 − 1 = δ } . \mathcal{E}^{\mathrm{univ}}_{\delta}=\{(h_0,a_0)\in\mathbb{Z}_{\ge0}^{2}:2h_0+a_0-1=\delta\}. E δ univ = {( h 0 , a 0 ) ∈ Z ≥ 0 2 : 2 h 0 + a 0 − 1 = δ } .
In (72) , replace each Q j − δ Q_{j-\delta} Q j − δ by its polynomial extension and ( n a 0 ) \binom{n}{a_0} ( a 0 n ) by ( n ) a 0 / a 0 ! (n)_{a_0}/a_0! ( n ) a 0 / a 0 ! , and sum over this universal set. Each summand has degree at most a 0 + j − δ ≤ j + 1 a_0+j-\delta\le j+1 a 0 + j − δ ≤ j + 1 . For an integer n ≥ a 0 n\ge a_0 n ≥ a 0 , its other component has p = n − a 0 + 1 ≥ 1 p=n-a_0+1\ge1 p = n − a 0 + 1 ≥ 1 , so this is precisely the original term. For 0 ≤ n < a 0 0\le n<a_0 0 ≤ n < a 0 , the polynomial falling factorial is zero. Thus the extended sum equals s j , n s_{j,n} s j , n at every nonnegative integer.
For a carrier pattern, remove A o u t ≤ n A_{\mathrm{out}}\le n A out ≤ n from the records in (51) , retaining the distinguished block, the stable small-component data and the exact deficit δ \delta δ . Call the resulting set E ~ δ \widetilde{\mathcal{E}}_{\delta} E δ . It is finite independently of n n n , since ∑ i ≠ ∗ c i = δ − m \sum_{i\ne*}c_i=\delta-m ∑ i = ∗ c i = δ − m bounds every small genus and cusp count. For a record e e e , put A e l a b = a 0 + ∑ i ≠ ∗ a i A_e^{\mathrm{lab}}=a_0+\sum_{i\ne*}a_i A e lab = a 0 + ∑ i = ∗ a i and write the prefactor independent of the ambient cusp count as
C e ( y ) = J P ( y ) V h 0 , a 0 + k ( 0 a 0 , x ) a 0 ! ∏ i ≠ ∗ a i ! ∏ i ≠ ∗ V g i , a i + k i ( 0 a i , x I i ) . (118) C_e(y)=\frac{J_P(y)V_{h_0,a_0+k}(0^{a_0},x)}{a_0!\prod_{i\ne*}a_i!}\prod_{i\ne*}V_{g_i,a_i+k_i}(0^{a_i},x_{I_i}). \tag*{(118)} C e ( y ) = a 0 ! ∏ i = ∗ a i ! J P ( y ) V h 0 , a 0 + k ( 0 a 0 , x ) i = ∗ ∏ V g i , a i + k i ( 0 a i , x I i ) . (118)
This continuous function is independent of n n n . Denote polynomial extensions of shifted coefficients by Q ^ \widehat{Q} Q . The polynomial kernel is
F ^ j ( n ; y ) = ∑ δ = 1 j ∑ e ∈ E δ ( 8 π 2 ) − δ ( n ) A e l a b C e ( y ) Q ^ j − δ n , H e , δ ; k ∗ , e ( x I ∗ , e ) . (119) \widehat{\mathcal{F}}_j(n;y)=\sum_{\delta=1}^{j}\sum_{e\in\mathcal{E}_\delta}(8\pi^2)^{-\delta}(n)A_e^{\mathrm{lab}}C_e(y)\widehat{Q}_{j-\delta}^{n,H_e,\delta;k_{*,e}}(x_{I_{*,e}}). \tag*{(119)} F j ( n ; y ) = δ = 1 ∑ j e ∈ E δ ∑ ( 8 π 2 ) − δ ( n ) A e lab C e ( y ) Q j − δ n , H e , δ ; k ∗ , e ( x I ∗ , e ) . (119)
The combinatorial cusp budget (24) holds for these records without any ambient-cusp restriction, giving A e l a b ≤ δ + 1 A_e^{\mathrm{lab}}\le\delta+1 A e lab ≤ δ + 1 . Every summand therefore has degree at most j + 1 j+1 j + 1 . For n ≥ A e l a b n\ge A_e^{\mathrm{lab}} n ≥ A e lab , the identity 2 H e + A e l a b − k ∗ , e = δ 2H_e+A_e^{\mathrm{lab}}-k_{*,e}=\delta 2 H e + A e lab − k ∗ , e = δ gives p = n − A e l a b + k ∗ , e ≥ k ∗ , e ≥ 1 p=n-A_e^{\mathrm{lab}}+k_{*,e}\ge k_{*,e}\ge1 p = n − A e lab + k ∗ , e ≥ k ∗ , e ≥ 1 . Hence the term is exactly the canonical term in (56) . For 0 ≤ n < A e l a b 0\le n<A_e^{\mathrm{lab}} 0 ≤ n < A e lab , its falling factorial vanishes. The extension of Q Q Q has already been defined, so this statement never evaluates a volume with a negative number of cusps. In particular all terms vanish if n < a 0 n<a_0 n < a 0 . This proves the claimed identification and continuity.
Theorem 3.11 (Fixed-order cusp polynomiality). For every integer j ≥ 1 j\ge1 j ≥ 1 there are unique fixed real linear functionals Λ j , r \Lambda_{j,r} Λ j , r , 0 ≤ r ≤ j + 1 0\le r\le j+1 0 ≤ r ≤ j + 1 , on compactly supported bounded even Borel tests, such that for every integer n ≥ 0 n\ge0 n ≥ 0 ,
κ j , n ( H ) = ∑ r = 0 j + 1 ( n r ) Λ j , r ( H ) . (120) \kappa_{j,n}(H)=\sum_{r=0}^{j+1}\binom{n}{r}\Lambda_{j,r}(H). \tag*{(120)} κ j , n ( H ) = r = 0 ∑ j + 1 ( r n ) Λ j , r ( H ) . (120) They are independent of genus, cusp count, truncation and support cutoff, and are explicitly specified by
Λ j , r ( H ) = Δ n r κ j , 0 ( H ) = ∑ q = 0 r ( − 1 ) r − q ( r q ) κ j , q ( H ) . (121) \Lambda_{j,r}(H)=\Delta_n^r\kappa_{j,0}(H)=\sum_{q=0}^{r}(-1)^{r-q}\binom{r}{q}\kappa_{j,q}(H). \tag*{(121)} Λ j , r ( H ) = Δ n r κ j , 0 ( H ) = q = 0 ∑ r ( − 1 ) r − q ( q r ) κ j , q ( H ) . (121) One absolute C ≥ 2 C\ge2 C ≥ 2 satisfies, simultaneously for all j ≥ 1 j\ge1 j ≥ 1 , T ≥ 0 T\ge0 T ≥ 0 and supp H ⊂ [ − T , T ] \operatorname{supp}H\subset[-T,T] supp H ⊂ [ − T , T ] ,
∑ r = 0 j + 1 ∣ Λ j , r ( H ) ∣ ≤ [ C ( j + 1 ) ] C ( j + 1 ) ( 1 + T ) C ( j + 1 ) e 3 T ∥ H ∥ ∞ , (122) \sum_{r=0}^{j+1}|\Lambda_{j,r}(H)|\le[C(j+1)]^{C(j+1)}(1+T)^{C(j+1)}e^{3T}\|H\|_\infty, \tag*{(122)} r = 0 ∑ j + 1 ∣ Λ j , r ( H ) ∣ ≤ [ C ( j + 1 ) ] C ( j + 1 ) ( 1 + T ) C ( j + 1 ) e 3 T ∥ H ∥ ∞ , (122) ∣ κ j , n ( H ) ∣ ≤ [ C ( j + 1 ) ] C ( j + 1 ) ( n + 1 ) j + 1 ( 1 + T ) C ( j + 1 ) e 3 T ∥ H ∥ ∞ . (123) |\kappa_{j,n}(H)|\le[C(j+1)]^{C(j+1)}(n+1)^{j+1}(1+T)^{C(j+1)}e^{3T}\|H\|_\infty. \tag*{(123)} ∣ κ j , n ( H ) ∣ ≤ [ C ( j + 1 ) ] C ( j + 1 ) ( n + 1 ) j + 1 ( 1 + T ) C ( j + 1 ) e 3 T ∥ H ∥ ∞ . (123)
Proof. Fix j j j . By Lemma 3.10 and the absolute repetition bound following (76) , the simple functional η j , n \eta_{j,n} η j , n is polynomial of degree at most j+1. To justify passing polynomial coefficients through the integral, express those coefficient functions as finite linear combinations of the kernels at n=0,\ldots,j+1; each such kernel satisfies (62) . Thus polynomial extraction commutes with the absolutely convergent repetition integrals for every test in the statement.
For nonsimple carriers replace C j ( n ) \mathcal{C}_j(n) C j ( n ) in (99) by the finite set obtained by dropping a 0 ≤ n a_0\le n a 0 ≤ n . Its size remains at most ( j + 2 ) 3 (j+2)^3 ( j + 2 ) 3 . The fiber average Ψ H h 0 , a 0 , k \Psi_H^{h_0,a_0,k} Ψ H h 0 , a 0 , k depends on its carrier type, not on the ambient cusp count. Use the polynomial kernel (119) for each type. On the fiber’s support Λ P ≤ 2 T \Lambda_P\le2T Λ P ≤ 2 T , every coefficient of that kernel is bounded, while (82) bounds Ψ H \Psi_H Ψ H . The length domain has finite measure by (90) . Each polynomial coefficient is therefore absolutely integrable. Finite pattern and type sums preserve polynomiality and degree. For every nonnegative integer n n n , unavailable types and records vanish by Lemma 3.10, and all available terms agree with (88) –(99) . Consequently κ j , n ( H ) \kappa_{j,n}(H) κ j , n ( H ) is polynomial of degree at most d = j + 1 d=j+1 d = j + 1 on all nonnegative integers. The universal carrier-type set depends only on j j j ; the support condition (82) removes the types with m > 2 T / π m>2T/\pi m > 2 T / π for each test.
Newton interpolation now gives (3.63) and (121) . Uniqueness follows because the binomial polynomials of orders 0 , … , d 0,\ldots,d 0 , … , d form a basis. The finite formula also proves linearity and independence of the auxiliary parameters.
To retain quantitative control in j j j , apply the estimate (94) only at q = 0 , … , d q=0,\ldots,d q = 0 , … , d . Since q + 1 ≤ j + 2 ≤ 2 ( j + 1 ) q+1\le j+2\le2(j+1) q + 1 ≤ j + 2 ≤ 2 ( j + 1 ) , its maximum is at most
M j ( H ) : = [ C ( j + 1 ) ] C ( j + 1 ) ( 1 + T ) C ( j + 1 ) e 3 T ∥ H ∥ ∞ M_j(H):=[C(j+1)]^{C(j+1)}(1+T)^{C(j+1)}e^{3T}\lVert H\rVert_\infty M j ( H ) := [ C ( j + 1 ) ] C ( j + 1 ) ( 1 + T ) C ( j + 1 ) e 3 T ∥ H ∥ ∞
after one absolute enlargement of C C C . Equation (121) gives ∣ Λ j , r ( H ) ∣ ≤ 2 r M j ( H ) |\Lambda_{j,r}(H)|\le2^rM_j(H) ∣ Λ j , r ( H ) ∣ ≤ 2 r M j ( H ) . Summing over r ≤ d r\le d r ≤ d contributes a factor at most 2 d + 1 2^{d+1} 2 d + 1 , which is absorbed by enlarging C C C . This proves (122) . Finally 0 ≤ ( n r ) ≤ ( n + 1 ) d 0\le\binom{n}{r}\le(n+1)^d 0 ≤ ( r n ) ≤ ( n + 1 ) d for nonnegative integers n n n and 0 ≤ r ≤ d 0\le r\le d 0 ≤ r ≤ d , proving (123) .
For j = 1 j=1 j = 1 , uniqueness identifies these functionals with Λ 0 , Λ 1 , Λ 2 \Lambda_0,\Lambda_1,\Lambda_2 Λ 0 , Λ 1 , Λ 2 of Proposition 3.8. For j ≥ 2 j\ge2 j ≥ 2 , the theorem gives only an upper bound on the degree. In the geometric-moment notation (100) , ω j , n \omega_{j,n} ω j , n consequently has degree at most j j j in n n n for j ≥ 2 j\ge2 j ≥ 2 , while ω 1 \omega_1 ω 1 is independent of n n n . Section 4 controls the remainder when n n n varies with g g g ; its final estimate is Theorem 4.8.
Uniform remainder estimates
This section sharpens the remainder in Theorem 3.5 while preserving its canonical coefficients. Section 4.1 tracks the cusp powers in the retained-record errors and complementary tails, giving a remainder proportional to ( n + 1 ) R + 2 (n+1)^{R+2} ( n + 1 ) R + 2 in Theorem 4.5. Section 4.2 then improves the tail estimate. Theorem 4.8 gives the same remainder under an additive genus threshold: a quadratic cusp term with an absolute coefficient, plus an order-dependent term linear in n + 1 n+1 n + 1 .
Throughout, g ≥ 3 g\ge3 g ≥ 3 and n ≥ 0 n\ge0 n ≥ 0 are integers and s = n + 1 s=n+1 s = n + 1 . For integers G ≥ 3 G\ge3 G ≥ 3 , p ≥ 1 p\ge1 p ≥ 1 and 0 ≤ ℓ ≤ p 0\le\ell\le p 0 ≤ ℓ ≤ p , write
V G , p ( ℓ ) ( x ) = V G , p ( 0 p − ℓ , x ) V G , p \mathcal{V}_{G,p}^{(\ell)}(x)=\frac{V_{G,p}(0_{p-\ell},x)}{V_{G,p}} V G , p ( ℓ ) ( x ) = V G , p V G , p ( 0 p − ℓ , x )
for x ∈ [ 0 , ∞ ) ℓ x\in[0,\infty)^\ell x ∈ [ 0 , ∞ ) ℓ . All constants in this section are absolute. As before, C C C may increase between estimates; one sufficiently large choice is independent of genus, cusp count, expansion order, support size and carrier type.
An all-order remainder with explicit cusp powers
We first keep the power of the marked-point count explicit at every order. Lemmas 4.1 and 4.2 control the retained records under genus thresholds linear in s s s . Lemma 4.3 controls the omitted records and is the source of the stronger eighth-power cusp threshold in this subsection.
Lemma 4.1 (Graded boundary errors). There is an absolute C ≥ 2 C\ge2 C ≥ 2 such that, for all integers J ≥ 0 J\ge0 J ≥ 0 , p ≥ 1 p\ge1 p ≥ 1 , 0 ≤ ℓ ≤ p 0\le\ell\le p 0 ≤ ℓ ≤ p , all x ∈ [ 0 , ∞ ) ℓ x\in[0,\infty)^\ell x ∈ [ 0 , ∞ ) ℓ , and every integer G ≥ [ C ( J + 1 ) ] C ( p + 1 ) G\ge[C(J+1)]^{C}(p+1) G ≥ [ C ( J + 1 ) ] C ( p + 1 ) , with X = ∑ i x i X=\sum_i x_i X = ∑ i x i ,
∣ V G , p ( ℓ ) ( x ) − ∑ j = 0 J G − j B j , p ( ℓ ) ( x ) ∣ ≤ G − J − 1 ( p + 1 ) J + 1 [ C ( J + 1 ) ] C ( J + 1 ) ( 1 + X ) 2 J + 2 e X / 2 . (124) \left|\mathcal{V}_{G,p}^{(\ell)}(\mathbf{x})-\sum_{j=0}^{J}G^{-j}B_{j,p}^{(\ell)}(\mathbf{x})\right|
\le G^{-J-1}(p+1)^{J+1}[C(J+1)]^{C(J+1)}(1+X)^{2J+2}e^{X/2}. \tag*{(124)} V G , p ( ℓ ) ( x ) − j = 0 ∑ J G − j B j , p ( ℓ ) ( x ) ≤ G − J − 1 ( p + 1 ) J + 1 [ C ( J + 1 ) ] C ( J + 1 ) ( 1 + X ) 2 J + 2 e X /2 . (124) For every j ≥ 1 j\ge1 j ≥ 1 ,
∣ B j , p ( ℓ ) ( x ) ∣ ≤ ( p + 1 ) j [ C ( j + 1 ) ] C j ( 1 + X ) 2 j e X / 2 . (125) \left|B_{j,p}^{(\ell)}(\mathbf{x})\right|
\le(p+1)^j[C(j+1)]^{Cj}(1+X)^{2j}e^{X/2}. \tag*{(125)} B j , p ( ℓ ) ( x ) ≤ ( p + 1 ) j [ C ( j + 1 ) ] C j ( 1 + X ) 2 j e X /2 . (125)
The scalar expansions of H G , p H_{G,p} H G , p and P G , p P_{G,p} P G , p have the same order-J J J error with the length factors removed, also for p = 0 p=0 p = 0 . Their order-j j j coefficients satisfy the corresponding bound ( p + 1 ) j [ C ( j + 1 ) ] C j (p+1)^j[C(j+1)]^{Cj} ( p + 1 ) j [ C ( j + 1 ) ] C j . All genus thresholds here are independent of X X X .
Proof. Using p + j + 1 ≤ ( j + 1 ) ( p + 1 ) p+j+1\le(j+1)(p+1) p + j + 1 ≤ ( j + 1 ) ( p + 1 ) and p + J + 2 ≤ ( J + 2 ) ( p + 1 ) p+J+2\le(J+2)(p+1) p + J + 2 ≤ ( J + 2 ) ( p + 1 ) , the source quantities in Proposition 2.3 satisfy
D j ( p ) ≤ ( p + 1 ) j [ C ( j + 1 ) ] C j , G J ( p ) ≤ [ C ( J + 1 ) ] C ( p + 1 ) . D_j(p)\le(p+1)^j[C(j+1)]^{Cj},\qquad G_J(p)\le[C(J+1)]^C(p+1). D j ( p ) ≤ ( p + 1 ) j [ C ( j + 1 ) ] C j , G J ( p ) ≤ [ C ( J + 1 ) ] C ( p + 1 ) .
Enlarge C C C so that the non-strict genus threshold in the lemma implies G > G J ( p ) G>G_J(p) G > G J ( p ) . The scalar assertions follow immediately, including p = 0 p=0 p = 0 . For boundary coefficients, apply (39) to (40) , absorbing 2 j + 1 2j+1 2 j + 1 into the order factor.
Within the volume polynomial’s dimension cutoff, the order-J J J error follows by the same summation with D J + 1 ( p ) E 2 J + 2 ( v ) D_{J+1}(p)E_{2J+2}(v) D J + 1 ( p ) E 2 J + 2 ( v ) , where v = ∣ d ∣ v=|\mathbf{d}| v = ∣ d ∣ . The coefficient series also contains terms beyond this cutoff; there v ≥ 3 G + p − 2 ≥ G v\ge3G+p-2\ge G v ≥ 3 G + p − 2 ≥ G . For 0 ≤ j ≤ J 0\le j\le J 0 ≤ j ≤ J , use (41) to bound the omitted order-j j j series by
G − J − 1 D j ( p ) ( 2 J + 2 ) ! ( 2 J + 3 ) ( 1 + X ) 2 J + 2 e X / 2 G^{-J-1}D_j(p)(2J+2)!(2J+3)(1+X)^{2J+2}e^{X/2} G − J − 1 D j ( p ) ( 2 J + 2 )! ( 2 J + 3 ) ( 1 + X ) 2 J + 2 e X /2
where D 0 ( p ) = 1 D_0(p)=1 D 0 ( p ) = 1 for this estimate. Since ( p + 1 ) j ≤ ( p + 1 ) J + 1 (p+1)^j\le(p+1)^{J+1} ( p + 1 ) j ≤ ( p + 1 ) J + 1 , summing these J + 1 J+1 J + 1 terms costs only another factor [ C ( J + 1 ) ] C ( J + 1 ) [C(J+1)]^{C(J+1)} [ C ( J + 1 ) ] C ( J + 1 ) . This includes the omitted constant series at j = 0 j=0 j = 0 and proves (124) . For ℓ = 0 \ell=0 ℓ = 0 the exact ratio equals one and the positive-order boundary coefficients vanish, so the assertion holds directly.
Lemma 4.2 (Graded shifted-volume error). Let J J J , H 0 ≥ 0 H_0\ge0 H 0 ≥ 0 , δ ≥ 1 \delta\ge1 δ ≥ 1 and n ≥ 0 n\ge0 n ≥ 0 be integers, and set
p = n + 2 H 0 − δ ≥ 1 , L = J + H 0 + δ + 1 , s = n + 1. p=n+2H_0-\delta\ge1,\qquad L=J+H_0+\delta+1,\qquad s=n+1. p = n + 2 H 0 − δ ≥ 1 , L = J + H 0 + δ + 1 , s = n + 1. For every integer 0 ≤ ℓ ≤ p 0\le\ell\le p 0 ≤ ℓ ≤ p , all x ∈ [ 0 , ∞ ) ℓ \mathbf{x}\in[0,\infty)^\ell x ∈ [ 0 , ∞ ) ℓ , and g ≥ [ C L ] C s g\ge[CL]^Cs g ≥ [ C L ] C s , with X = ∑ i x i X=\sum_i x_i X = ∑ i x i , one has
∣ V g − H 0 , p ( 0 p − ℓ , x ) V g , n − ( 8 π 2 ) − δ g − δ ∑ j = 0 J g − j Q j n , H 0 , δ ; ℓ ( x ) ∣ ≤ g − δ − J − 1 s J + 1 [ C L ] C ( J + 1 ) ( 1 + X ) 2 J + 2 e X / 2 . (126) \left|\frac{V_{g-H_0,p}(\mathbf{0}_{p-\ell},\mathbf{x})}{V_{g,n}}-(8\pi^2)^{-\delta}g^{-\delta}\sum_{j=0}^{J}g^{-j}Q_j^{n,H_0,\delta;\ell}(\mathbf{x})\right|
\le g^{-\delta-J-1}s^{J+1}[CL]^{C(J+1)}(1+X)^{2J+2}e^{X/2}. \tag*{(126)} V g , n V g − H 0 , p ( 0 p − ℓ , x ) − ( 8 π 2 ) − δ g − δ j = 0 ∑ J g − j Q j n , H 0 , δ ; ℓ ( x ) ≤ g − δ − J − 1 s J + 1 [ C L ] C ( J + 1 ) ( 1 + X ) 2 J + 2 e X /2 . (126) Here the Q j Q_j Q j are exactly the canonical functions of Lemma 2.6. For 0 ≤ j ≤ J 0\le j\le J 0 ≤ j ≤ J they satisfy
∣ Q j n , H 0 , δ ; ℓ ( x ) ∣ ≤ s j [ C L ] C j ( 1 + X ) 2 j e X / 2 . (127) \left|Q_j^{n,H_0,\delta;\ell}(\mathbf{x})\right|\le s^j[CL]^{Cj}(1+X)^{2j}e^{X/2}. \tag*{(127)} Q j n , H 0 , δ ; ℓ ( x ) ≤ s j [ C L ] C j ( 1 + X ) 2 j e X /2 . (127) Proof. We estimate the scalar shifted ratio first, then the normalized boundary factor, and finally their product. All marked counts along (43) are at most n + 2 H 0 + δ n+2H_0+\delta n + 2 H 0 + δ , and all genus shifts are at most H 0 H_0 H 0 . For orders 1 ≤ j ≤ J + 1 1\le j\le J+1 1 ≤ j ≤ J + 1 , the exact source majorants are therefore bounded by A j A^j A j with
A = [ C 1 L ] C 1 s , A=[C_1L]^{C_1}s, A = [ C 1 L ] C 1 s , where C 1 C_1 C 1 is an auxiliary absolute constant. The source thresholds satisfy a bound of the same form as A A A . The genus-shift part of Lemma 2.5 replaces A A A by at most C ( A + H 0 + 1 ) C(A+H_0+1) C ( A + H 0 + 1 ) , still of this form. The geometric denominator factors have expansion coefficients [ − ( n − δ + q − 2 ) / 2 ] j [-(n-\delta+q-2)/2]^j [ − ( n − δ + q − 2 ) /2 ] j and a geometric remainder, so they obey the same majorant, including when the numerator in brackets is negative.
There are H 0 + 2 δ ≤ 2 L H_0+2\delta\le2L H 0 + 2 δ ≤ 2 L scalar factors. The product part of Lemma 2.5, after enlarging the absolute exponent in [ C L ] C [CL]^C [ C L ] C , gives a normalized scalar ratio Z ( g ) Z(g) Z ( g ) , defined by V g − H 0 , p / V g , n = ( 8 π 2 ) − δ g − δ Z ( g ) V_{g-H_0,p}/V_{g,n}=(8\pi^2)^{-\delta}g^{-\delta}Z(g) V g − H 0 , p / V g , n = ( 8 π 2 ) − δ g − δ Z ( g ) . Its canonical truncation Z [ J ] ( g ) = ∑ j = 0 J v j g − j Z^{[J]}(g)=\sum_{j=0}^{J}v_jg^{-j} Z [ J ] ( g ) = ∑ j = 0 J v j g − j satisfies
∣ v j ∣ ≤ A j ( 0 ≤ j ≤ J ) , ∣ Z − Z [ J ] ∣ ≤ ( A / g ) J + 1 , |v_j|\le A^j\quad(0\le j\le J),\qquad|Z-Z^{[J]}|\le(A/g)^{J+1}, ∣ v j ∣ ≤ A j ( 0 ≤ j ≤ J ) , ∣ Z − Z [ J ] ∣ ≤ ( A / g ) J + 1 , ∣ Z [ J ] ∣ ≤ 2 , A / g ≤ 1 4 . (128) |Z^{[J]}|\le2,\qquad A/g\le\frac14. \tag*{(128)} ∣ Z [ J ] ∣ ≤ 2 , A / g ≤ 4 1 . (128)
All subsequent enlargements of C 1 C_1 C 1 are absolute, so A A A remains linear in s s s . Impose g ≥ 4 A g \ge4A g ≥ 4 A and choose C 1 C_1 C 1 large enough that this also ensures g − H 0 ≥ g / 2 g-H_0 \ge g/2 g − H 0 ≥ g /2 and every strict source threshold. At the end choose the statement’s constant C C C so that [ C L ] C s ≥ 4 [ C 1 L ] C 1 s [CL]^C s \ge4[C_1L]^{C_1}s [ C L ] C s ≥ 4 [ C 1 L ] C 1 s for every L ≥ 1 L \ge1 L ≥ 1 .
Let D ( g , x ) = V g − H 0 , p ( ℓ ) ( x ) D(g,\mathbf{x})=\mathcal{V}^{(\ell)}_{g-H_0,p}(\mathbf{x}) D ( g , x ) = V g − H 0 , p ( ℓ ) ( x ) . Shift the boundary expansion of Lemma 4.1. Its coefficients are exactly
B ~ 0 = Φ ℓ , B ~ q = ∑ j = 1 q ( q − 1 j − 1 ) H 0 q − j B j , p ( ℓ ) . \widetilde{B}_0=\Phi_\ell,\qquad\widetilde{B}_q=\sum_{j=1}^{q}\binom{q-1}{j-1}H_0^{q-j}B_{j,p}^{(\ell)}. B 0 = Φ ℓ , B q = j = 1 ∑ q ( j − 1 q − 1 ) H 0 q − j B j , p ( ℓ ) .
Consequently ∣ B ~ q ∣ ≤ A q ( 1 + X ) 2 q e X / 2 \lvert\widetilde{B}_q\rvert\le A^q(1+X)^{2q}e^{X/2} ∣ B q ∣ ≤ A q ( 1 + X ) 2 q e X /2 for 0 ≤ q ≤ J 0\le q\le J 0 ≤ q ≤ J . Put D [ J ] ( g , x ) = ∑ q = 0 J g − q B ~ q ( x ) D^{[J]}(g,\mathbf{x})=\sum_{q=0}^{J}g^{-q}\widetilde{B}_q(\mathbf{x}) D [ J ] ( g , x ) = ∑ q = 0 J g − q B q ( x ) . For the shifted remainder, the original boundary error costs ( 2 A / g ) J + 1 ( 1 + X ) 2 J + 2 e X / 2 (2A/g)^{J+1}(1+X)^{2J+2}e^{X/2} ( 2 A / g ) J + 1 ( 1 + X ) 2 J + 2 e X /2 . In the truncated coefficient sum, use (42) and ( j + v − 1 v ) ≤ 2 j + v \binom{j+v-1}{v}\le2^{j+v} ( v j + v − 1 ) ≤ 2 j + v for the discarded powers v ≥ J + 1 − j v\ge J+1-j v ≥ J + 1 − j . Since H 0 / g ≤ 1 / 4 H_0/g\le1/4 H 0 / g ≤ 1/4 , their sum is bounded by [ C ( A + H 0 + 1 ) / g ] J + 1 ( 1 + X ) 2 J e X / 2 [C(A+H_0+1)/g]^{J+1}(1+X)^{2J}e^{X/2} [ C ( A + H 0 + 1 ) / g ] J + 1 ( 1 + X ) 2 J e X /2 . After another fixed enlargement of A A A , this proves
∣ D − D [ J ] ∣ ≤ ( A / g ) J + 1 ( 1 + X ) 2 J + 2 e X / 2 . (129) \lvert D-D^{[J]}\rvert\le(A/g)^{J+1}(1+X)^{2J+2}e^{X/2}. \tag*{(129)} ∣ D − D [ J ] ∣ ≤ ( A / g ) J + 1 ( 1 + X ) 2 J + 2 e X /2 . (129)
Use the exact decomposition
Z D − Z [ J ] D [ J ] = ( Z − Z [ J ] ) D + Z [ J ] ( D − D [ J ] ) . ZD-Z^{[J]}D^{[J]}=(Z-Z^{[J]})D+Z^{[J]}(D-D^{[J]}). Z D − Z [ J ] D [ J ] = ( Z − Z [ J ] ) D + Z [ J ] ( D − D [ J ] ) .
The exact bound ∣ D ∣ ≤ e X / 2 \lvert D\rvert\le e^{X/2} ∣ D ∣ ≤ e X /2 and (128) –(129) control these two terms with one exponential length factor. For the terms discarded from the product of the truncations, first take out ( 1 + X ) 2 J e X / 2 (1+X)^{2J}e^{X/2} ( 1 + X ) 2 J e X /2 . The remaining sum is at most
∑ 0 ≤ u , v ≤ J u + v ≥ J + 1 ( A / g ) u + v ≤ ( J + 1 ) 2 ( A / g ) J + 1 . \sum_{\substack{0\le u,v\le J\\u+v\ge J+1}}(A/g)^{u+v}\le(J+1)^2(A/g)^{J+1}. 0 ≤ u , v ≤ J u + v ≥ J + 1 ∑ ( A / g ) u + v ≤ ( J + 1 ) 2 ( A / g ) J + 1 .
The polynomial factor in J J J is absorbed into [ C L ] C ( J + 1 ) [CL]^{C(J+1)} [ C L ] C ( J + 1 ) . Only the boundary factor carries length growth, so the scalar smallness condition A / g ≤ 1 / 4 A/g\le1/4 A / g ≤ 1/4 remains independent of X X X . The finite convolution ∑ u + v = j B ~ u B ~ v \sum_{u+v=j}\widetilde{B}_u\widetilde{B}_v ∑ u + v = j B u B v equals Q j Q_j Q j , proving both assertions. When J = 0 J=0 J = 0 the discarded sum is empty and the same argument applies.
Lemma 4.3 (Positive tails with explicit cusp power). For every integer R ≥ 0 R\ge0 R ≥ 0 , every admissible carrier type of complexity m ≥ 1 m\ge1 m ≥ 1 , and t = R + m + 1 t=R+m+1 t = R + m + 1 ,
g ≥ [ K t ( n + 1 ) ] 8 ⟹ T g , n h 0 , a 0 , k ( R ) ≤ g − R − 1 ( n + 1 ) R + 2 [ C t ] C t . (130) g\ge[Kt(n+1)]^8\quad\Longrightarrow\quad\mathcal{T}^{h_0,a_0,k}_{g,n}(R)\le g^{-R-1}(n+1)^{R+2}[Ct]^{Ct}. \tag*{(130)} g ≥ [ K t ( n + 1 ) ] 8 ⟹ T g , n h 0 , a 0 , k ( R ) ≤ g − R − 1 ( n + 1 ) R + 2 [ C t ] C t . (130) On a fixed pattern , its boundary tail has the additional factor ( 1 + Λ ) k e Λ (1+\Lambda)^k e^\Lambda ( 1 + Λ ) k e Λ . The separating kernel with deficit at least R + 1 R+1 R + 1 satisfies
S g , n ≥ R + 1 ( x ) ≤ g − R − 1 ( n + 1 ) R + 2 [ C ( R + 1 ) ] C ( R + 1 ) e x ( g ≥ [ K ( R + 1 ) ( n + 1 ) ] 8 ) . (131) S^{\ge R+1}_{g,n}(x)\le g^{-R-1}(n+1)^{R+2}[C(R+1)]^{C(R+1)}e^x\qquad\left(g\ge[K(R+1)(n+1)]^8\right). \tag*{(131)} S g , n ≥ R + 1 ( x ) ≤ g − R − 1 ( n + 1 ) R + 2 [ C ( R + 1 ) ] C ( R + 1 ) e x ( g ≥ [ K ( R + 1 ) ( n + 1 ) ] 8 ) . (131) Proof. Return to the near- and deep-deficit estimates in the proof of Theorem 2.2 . Set D = max { m , R + 1 } D=\max\{m,R+1\} D = max { m , R + 1 } and retain the bound ( N ) D ≥ g D (N)_D\ge g^D ( N ) D ≥ g D before weakening g − D g^{-D} g − D to g − R − 1 g^{-R-1} g − R − 1 . The near sector, including its endpoint, is bounded by 2 P b b s D + 1 ( D + 3 ) 2 b D ! g − D 2P_b b s^{D+1}(D+3)^{2b}D!g^{-D} 2 P b b s D + 1 ( D + 3 ) 2 b D ! g − D . Under the same eighth-power threshold g ≥ s g\ge s g ≥ s , and
g − D s D + 1 = g − R − 1 s R + 2 ( s / g ) D − R − 1 ≤ g − R − 1 s R + 2 . (132) g^{-D}s^{D+1}=g^{-R-1}s^{R+2}(s/g)^{D-R-1}\le g^{-R-1}s^{R+2}. \tag*{(132)} g − D s D + 1 = g − R − 1 s R + 2 ( s / g ) D − R − 1 ≤ g − R − 1 s R + 2 . (132) All other factors, including the pattern count, are bounded by [ C t ] C t [Ct]^{Ct} [ C t ] C t because D ≤ t D\le t D ≤ t and b ≤ t + 1 b\le t+1 b ≤ t + 1 . The deep sector in (30) is already P b g − R − 1 P_b g^{-R-1} P b g − R − 1 under the same threshold. For b = 1 b=1 b = 1 it is empty; the endpoint of the near sector remains included. This proves (130) , also for R = 0 R=0 R = 0 . The numerator-length calculation following (30) supplies the stated boundary factor.
For the separating kernel put r = R + 1 r = R + 1 r = R + 1 in (68) . The uncompressed near bound is 12 B s R + 2 ( R + 1 ) ! g − R − 1 12Bs^{R+2}(R+1)!g^{-R-1} 12 B s R + 2 ( R + 1 )! g − R − 1 , and (69) bounds the deep part by B g − R − 1 Bg^{-R-1} B g − R − 1 . Boundary growth adds at most e x e^x e x . Absorb the factorial into the order factor to obtain (131) , including the central tie and n = 0 n = 0 n = 0 .
Proposition 4.4 (Kernel errors with explicit cusp powers). Let R , n ≥ 0 R, n \ge0 R , n ≥ 0 be integers and put r = R + 1 r = R + 1 r = R + 1 , s = n + 1 s = n + 1 s = n + 1 . Under g ≥ [ C r ] C s 8 g \ge[Cr]^{C}s^8 g ≥ [ C r ] C s 8 , the canonical simple expansions obey, for x ≥ 0 x \ge0 x ≥ 0 ,
∣ U g , n ( x ) − ϕ ( x ) 2 − ∑ j = 1 R g − j u j , n ( x ) ∣ ≤ g − R − 1 s R + 1 [ C r ] C r ( 1 + x ) 2 R + 2 e x , (133) \left|U_{g,n}(x)-\phi(x)^2-\sum_{j=1}^{R}g^{-j}u_{j,n}(x)\right|\leq g^{-R-1}s^{R+1}[Cr]^{Cr}(1+x)^{2R+2}e^x, \tag*{(133)} U g , n ( x ) − ϕ ( x ) 2 − j = 1 ∑ R g − j u j , n ( x ) ≤ g − R − 1 s R + 1 [ C r ] C r ( 1 + x ) 2 R + 2 e x , (133) ∣ S g , n ( x ) − ∑ j = 1 R g − j s j , n ( x ) ∣ ≤ g − R − 1 s R + 2 [ C r ] C r ( 1 + x ) 2 R + 2 e x . (134) \left|S_{g,n}(x)-\sum_{j=1}^{R}g^{-j}s_{j,n}(x)\right|\leq g^{-R-1}s^{R+2}[Cr]^{Cr}(1+x)^{2R+2}e^x. \tag*{(134)} S g , n ( x ) − j = 1 ∑ R g − j s j , n ( x ) ≤ g − R − 1 s R + 2 [ C r ] C r ( 1 + x ) 2 R + 2 e x . (134) For a fixed carrier and pattern, let t = R + m + 1 t = R + m + 1 t = R + m + 1 and let y ∈ [ 0 , ∞ ) e P \mathbf{y} \in[0,\infty)^{e_P} y ∈ [ 0 , ∞ ) e P be its edge lengths. Under g ≥ [ C t ] C s 8 g \ge[Ct]^{C}s^8 g ≥ [ C t ] C s 8 ,
∣ F g , n h 0 , a 0 , k ; P ( y ) − ∑ j = 1 R g − j f j , n h 0 , a 0 , k ; P ( y ) ∣ ≤ g − R − 1 s R + 2 [ C t ] C t ( 1 + Λ ) 2 t e Λ . (135) \left|F^{h_0,a_0,k;P}_{g,n}(\mathbf{y})-\sum_{j=1}^{R}g^{-j}f^{h_0,a_0,k;P}_{j,n}(\mathbf{y})\right|\leq g^{-R-1}s^{R+2}[Ct]^{Ct}(1+\Lambda)^{2t}e^\Lambda. \tag*{(135)} F g , n h 0 , a 0 , k ; P ( y ) − j = 1 ∑ R g − j f j , n h 0 , a 0 , k ; P ( y ) ≤ g − R − 1 s R + 2 [ C t ] C t ( 1 + Λ ) 2 t e Λ . (135) The coefficients are given by (65) , (72) , and (56) , respectively.
Proof. For the nonseparating ratio use U g , n = P g , n V g − 1 , n + 2 ( 2 ) ( x , x ) U_{g,n}=P_{g,n}\mathcal{V}^{(2)}_{g-1,n+2}(x,x) U g , n = P g , n V g − 1 , n + 2 ( 2 ) ( x , x ) . Lemma 4.1 , its genus shift by one, and the scalar–boundary product argument just proved give (133) . Replacing n + 3 n+3 n + 3 by 3 s 3s 3 s and 1 + 2 x 1+2x 1 + 2 x by 2 ( 1 + x ) 2(1+x) 2 ( 1 + x ) costs only the displayed order factor. The formal convolution is precisely (65) .
For a retained separating endpoint of deficit 1 ≤ δ ≤ R 1 \le\delta\le R 1 ≤ δ ≤ R , take J = R − δ J = R-\delta J = R − δ and H 0 = h 0 H_0 = h_0 H 0 = h 0 in Lemma 4.2 . Its label factor is at most s a 0 s^{a_0} s a 0 , with a 0 ≤ δ + 1 a_0 \le\delta+ 1 a 0 ≤ δ + 1 . The small volume is bounded by [ C ( δ + 1 ) ] C ( δ + 1 ) e x / 2 [C(\delta+1)]^{C(\delta+1)}e^{x/2} [ C ( δ + 1 ) ] C ( δ + 1 ) e x /2 . Since h 0 ≤ ( δ + 1 ) / 2 h_0 \le(\delta+1)/2 h 0 ≤ ( δ + 1 ) /2 , the shifted lemma has L ≤ C ( R + 1 ) L \le C(R+1) L ≤ C ( R + 1 ) . The cusp factor in the resulting error is therefore bounded by
s a 0 + R − δ + 1 ≤ s R + 2 . s^{a_0+R-\delta+1}\leq s^{R+2}. s a 0 + R − δ + 1 ≤ s R + 2 .
The two boundary exponentials give e x e^x e x , and 2 ( R − δ ) + 2 ≤ 2 R + 2 2(R-\delta)+2 \leq2R+2 2 ( R − δ ) + 2 ≤ 2 R + 2 . Sum the at most O ( ( R + 1 ) 2 ) O((R+1)^2) O (( R + 1 ) 2 ) endpoints, with their two ordered placements, and add (131) . This proves (134) with the canonical coefficients from (72) .
For a retained carrier record e e e of deficit δ ≤ R \delta\leq R δ ≤ R , keep the label factor separate in (54) . The volume-envelope calculation there gives the sharper form
∣ A e ( y ) ∣ ≤ s A o u t [ C ( δ + 1 ) ] C ( δ + 1 ) ( 1 + Λ ) k e Λ − X ∗ / 2 , A o u t ≤ δ + 1. (136) |A_e(\mathbf{y})|\leq s^{A_{\mathrm{out}}}[C(\delta+1)]^{C(\delta+1)}(1+\Lambda)^k e^{\Lambda-X_*/2},\qquad A_{\mathrm{out}}\leq\delta+1. \tag*{(136)} ∣ A e ( y ) ∣ ≤ s A out [ C ( δ + 1 ) ] C ( δ + 1 ) ( 1 + Λ ) k e Λ − X ∗ /2 , A out ≤ δ + 1. (136)
Indeed, the product of the zero-length volumes of the carrier and small components is at most U b A 0 δ − b ( δ − b ) ! U^b A_0^{\delta-b}(\delta-b)! U b A 0 δ − b ( δ − b )! , and the remaining multinomial is at most s A o u t s^{A_{\mathrm{out}}} s A out . Take J = R − δ J = R-\delta J = R − δ in Lemma 4.2 . The record bounds give H 0 ≤ ( R + m + 2 ) / 2 H_0 \leq(R+m+2)/2 H 0 ≤ ( R + m + 2 ) /2 and L ≤ C t L \leq Ct L ≤ C t . Multiplication by (136) gives cusp power
s A o u t + J + 1 ≤ s R + 2 , (137) s^{A_{\mathrm{out}}+J+1}\leq s^{R+2}, \tag*{(137)} s A out + J + 1 ≤ s R + 2 , (137)
exponential factor e Λ e^\Lambda e Λ , and length power at most k + 2 ( R − δ ) + 2 ≤ 2 t k+2(R-\delta)+2 \leq2t k + 2 ( R − δ ) + 2 ≤ 2 t , since k ≤ m + 2 k \leq m+2 k ≤ m + 2 and δ ≥ m \delta\geq m δ ≥ m . The record count in (53) is at most [ C t ] C t [Ct]^{Ct} [ C t ] C t . Summing and adding Lemma 4.3 proves (135) . Choose C ≥ 8 C \geq8 C ≥ 8 large enough that [ C t ] C s 8 [Ct]^Cs^8 [ C t ] C s 8 dominates both the tail threshold and every shifted-ratio threshold.
The record reconstruction is valid because this threshold implies N > 2 R N > 2R N > 2 R and g > H 0 g > H_0 g > H 0 . Available records satisfy p = n − A o u t + k ∗ ≥ k ∗ ≥ 1 p = n-A_{\mathrm{out}}+k_* \geq k_* \geq1 p = n − A out + k ∗ ≥ k ∗ ≥ 1 ; unavailable cusp allocations are omitted before any volume is evaluated. The retained sums are exactly the finite definitions (56) . When there are no retained records, in particular if R = 0 R=0 R = 0 or m > R m>R m > R , the positive tail proves the assertion directly.
Theorem 4.5 (All-order orbital remainder with explicit cusp power). There is an absolute C ≥ 8 C \ge8 C ≥ 8 such that, for all integers R , n ≥ 0 R,n \ge0 R , n ≥ 0 , T ≥ 0 T \ge0 T ≥ 0 , and bounded real even Borel H H H supported in [ − T , T ] [-T,T] [ − T , T ] , putting ρ = R + ⌈ T ⌉ + 1 \rho=R+\lceil T\rceil+1 ρ = R + ⌈ T ⌉ + 1 , the condition
g ≥ [ C ρ ] C ( n + 1 ) 8 (138) g \ge[C\rho]^{C}(n+1)^8 \tag*{(138)} g ≥ [ C ρ ] C ( n + 1 ) 8 (138) implies
∣ E g , n O H − Ω ( H ) − ∑ j = 1 R g − j κ j , n ( H ) ∣ ≤ g − R − 1 ( n + 1 ) R + 2 [ C ρ ] C ρ e 3 T ∥ H ∥ ∞ . (139) \left|\mathbb{E}_{g,n}\mathcal{O}_{H}-\Omega(H)-\sum_{j=1}^{R}g^{-j}\kappa_{j,n}(H)\right|
\leq g^{-R-1}(n+1)^{R+2}[C\rho]^{C\rho}e^{3T}\|H\|_{\infty}. \tag*{(139)} E g , n O H − Ω ( H ) − j = 1 ∑ R g − j κ j , n ( H ) ≤ g − R − 1 ( n + 1 ) R + 2 [ C ρ ] C ρ e 3 T ∥ H ∥ ∞ . (139) The empty sum at R = 0 R=0 R = 0 is zero. The coefficients are those of Theorem 3.5, independent of genus, truncation order and support cutoff.
Proof. Apply the absolute repetition estimate following (76) to (133) –(4.11). It bounds the simple error by
g − R − 1 s R + 2 [ C ( R + 1 ) ] C ( R + 1 ) ( 1 + T ) 2 R + 4 e T ∥ H ∥ ∞ . g^{-R-1}s^{R+2}[C(R+1)]^{C(R+1)}(1+T)^{2R+4}e^T\|H\|_{\infty}. g − R − 1 s R + 2 [ C ( R + 1 ) ] C ( R + 1 ) ( 1 + T ) 2 R + 4 e T ∥ H ∥ ∞ .
Since R + 1 ≤ ρ R+1\leq\rho R + 1 ≤ ρ and 1 + T ≤ ρ 1+T\leq\rho 1 + T ≤ ρ , this has the form on the right of (139) .
Increasing C C C in (138) ensures g ≥ 4 ρ 2 g\geq4\rho^2 g ≥ 4 ρ 2 . The whole-carrier lower bound π ( g − 1 + n / 2 ) > T \pi(g-1+n/2)>T π ( g − 1 + n /2 ) > T excludes all carriers without a stable exterior, including entirely self-paired carriers. Thus the exact exhaustion (98) sums over 1 ≤ m ≤ M = ⌊ 2 T / π ⌋ 1\leq m\leq M=\lfloor2T/\pi\rfloor 1 ≤ m ≤ M = ⌊ 2 T / π ⌋ , with k ≤ m + 2 k\leq m+2 k ≤ m + 2 . For such a type, t = R + m + 1 ≤ ρ t=R+m+1\leq\rho t = R + m + 1 ≤ ρ , so Proposition 4.4 applies under the common threshold. For one pattern, combine (135) , (82) , and (90) . Its integrated error is at most
g − R − 1 s R + 2 [ C t ] C t C m ( 1 + T ) ( 1 + 2 T ) 2 t + k e 3 T ∥ H ∥ ∞ . g^{-R-1}s^{R+2}[Ct]^{Ct}C_m(1+T)(1+2T)^{2t+k}e^{3T}\|H\|_{\infty}. g − R − 1 s R + 2 [ C t ] C t C m ( 1 + T ) ( 1 + 2 T ) 2 t + k e 3 T ∥ H ∥ ∞ .
There are at most ( k + 1 ) 2 k (k+1)^{2k} ( k + 1 ) 2 k patterns and ( m + 2 ) 2 (m+2)^2 ( m + 2 ) 2 types at each complexity. The factor 1 / k ! 1/k! 1/ k ! can be discarded in this upper bound. Since k ≤ m + 2 ≤ 2 ρ k\leq m+2\leq2\rho k ≤ m + 2 ≤ 2 ρ and M ≤ T M\leq T M ≤ T , all length powers and type/pattern counts are absorbed into [ C ρ ] C ρ [C\rho]^{C\rho} [ C ρ ] C ρ . No ambient cusp factor enters this last summation.
The uniform fiber bound (82) and the bounded length simplex give absolute convergence of all error integrals. The two vanishing properties in (89) identify the subtracted coefficient sum exactly with ζ j , n \zeta_{j,n} ζ j , n , even when j > M j>M j > M or M > R M>R M > R . Adding the simple part gives κ j , n = η j , n + ζ j , n \kappa_{j,n}=\eta_{j,n}+\zeta_{j,n} κ j , n = η j , n + ζ j , n and proves the result. For M = 0 M=0 M = 0 the nonsimple part and its coefficients vanish on H H H ; for T = 0 T=0 T = 0 all terms vanish. This also proves the result for the stated bounded Borel tests.
The power R + 2 R+2 R + 2 agrees with the degree bound for the first omitted coefficient. This does not assert optimality of the remainder. The retained-record estimates already require genus only linear in n + 1 n+1 n + 1 ; the next subsection lowers the overall threshold by sharpening the positive tails.
An additive cusp and order threshold
The improvement comes from retaining the full binomial lower bound in the deep tail. Lemma 4.6 gives an additive threshold; Proposition 4.7 combines it with the retained-record estimates already proved. Theorem 4.8 then integrates these kernel bounds. We continue to write s = n + 1 s=n+1 s = n + 1 .
Lemma 4.6 (Positive tails at an additive threshold). There are absolute K , C ≥ 2 K, C \ge2 K , C ≥ 2 such that, for every carrier type of complexity m ≥ 1 m \ge1 m ≥ 1 , integer R ≥ 0 R \ge0 R ≥ 0 , and t = R + m + 1 t = R + m + 1 t = R + m + 1 ,
g ≥ K ( s 2 + t 2 s ) ⟹ T g , n h 0 , a 0 , k ( R ) ≤ g − R − 1 s R + 2 [ C t ] C t . (140) g \ge K(s^{2}+t^{2}s) \quad\Longrightarrow\quad\mathcal{T}^{h_{0},a_{0},k}_{g,n}(R) \le g^{-R-1}s^{R+2}[Ct]^{Ct}. \tag*{(140)} g ≥ K ( s 2 + t 2 s ) ⟹ T g , n h 0 , a 0 , k ( R ) ≤ g − R − 1 s R + 2 [ C t ] C t . (140) For a fixed pattern, replacing the numerator volumes by their boundary volumes and inserting the actual-edge Jacobian gives the additional factor ( 1 + Λ ) k e Λ (1+\Lambda)^{k}e^{\Lambda} ( 1 + Λ ) k e Λ . Writing r = R + 1 r=R+1 r = R + 1 , the ordered separating tail satisfies
S g , n ≥ r ( x ) ≤ q − r s r + 1 [ C r ] C r e x , q ≥ K ( s 2 + r 2 s ) , x ≥ 0. (141) S^{\ge r}_{g,n}(x)\le q^{-r}s^{r+1}[Cr]^{Cr}e^{x},\qquad q\ge K(s^{2}+r^{2}s),\qquad x\ge0. \tag*{(141)} S g , n ≥ r ( x ) ≤ q − r s r + 1 [ C r ] C r e x , q ≥ K ( s 2 + r 2 s ) , x ≥ 0. (141) The assertions include R = 0 R=0 R = 0 , n = 0 n=0 n = 0 , and empty admissible sums.
At x = 0 x=0 x = 0 and for fixed r r r , the powers of g g g and n n n in (141) agree with [7 Lemma B.14] in the regime n = o ( g ) n=o(\sqrt{g}) n = o ( g ) , with k = r + 1 k=r+1 k = r + 1 there. Here we control the dependence on r r r and obtain the explicit threshold g ≥ K ( s 2 + r 2 s ) g\ge K(s^{2}+r^{2}s) g ≥ K ( s 2 + r 2 s ) , together with the corresponding complementary-carrier estimate.
Proof. We split the complementary tail into near and deep deficits. The near sum is geometric; in the deep sum the full binomial denominator must absorb both the cusp-label count and the number of allocations. Choose an integer K ≥ max { 4096 , κ v } K\ge\max\{4096,\kappa_{v}\} K ≥ max { 4096 , κ v } . Put
r = R + 1 , N = 2 g + n − 3 , D ∗ = max { m , r } , J = ⌊ N / ( 64 s ) ⌋ , L s = log ( 64 s ) . (142) \begin{aligned}
r&=R+1,\qquad N=2g+n-3,\qquad D_{*}=\max\{m,r\},\\
J&=\lfloor N/(64s)\rfloor,\qquad L_{s}=\log(64s). \tag*{(142)}
\end{aligned} r J = R + 1 , N = 2 g + n − 3 , D ∗ = max { m , r } , = ⌊ N / ( 64 s )⌋ , L s = log ( 64 s ) . (142)
Here D ∗ D_{*} D ∗ is a lower bound for every deficit in the tail under consideration and J J J is the splitting point. We have t ≥ 2 t\ge2 t ≥ 2 , D ∗ ≤ t − 1 D_{*}\le t-1 D ∗ ≤ t − 1 and b ≤ k ≤ m + 2 ≤ t + 1 b\le k\le m+2\le t+1 b ≤ k ≤ m + 2 ≤ t + 1 . Since s + t 2 ≥ 2 t s+t^{2}\ge2t s + t 2 ≥ 2 t , the new hypothesis gives g ≥ 128 t s g\ge128ts g ≥ 128 t s and g ≥ κ v s 2 g\ge\kappa_{v}s^{2} g ≥ κ v s 2 . Consequently
g ≤ N ≤ 3 g , N + 1 ≤ 4 g , N − D ∗ + 1 ≥ g , g\le N\le3g,\qquad N+1\le4g,\qquad N-D_{*}+1\ge g, g ≤ N ≤ 3 g , N + 1 ≤ 4 g , N − D ∗ + 1 ≥ g ,
N ≥ 128 s , J ≥ t , 1 ≤ J ≤ N / 2 , J ≥ N / ( 128 s ) , s / g ≤ 1. (143) N\ge128s,\qquad J\ge t,\qquad1\le J\le N/2,\qquad J\ge N/(128s),\qquad s/g\le1. \tag*{(143)} N ≥ 128 s , J ≥ t , 1 ≤ J ≤ N /2 , J ≥ N / ( 128 s ) , s / g ≤ 1. (143)
Indeed g ≥ 128 t s g\ge128ts g ≥ 128 t s implies g ≥ m + 2 g\ge m+2 g ≥ m + 2 , g ≥ D ∗ + 2 g\ge D_{*}+2 g ≥ D ∗ + 2 , g ≥ s g\ge s g ≥ s and N / ( 64 s ) ≥ 2 t N/(64s)\ge2t N / ( 64 s ) ≥ 2 t ; the floor is at least half N / ( 64 s ) N/(64s) N / ( 64 s ) because the latter is at least two.
Fix a pattern with b ≥ 1 b\ge1 b ≥ 1 exterior blocks. Its indices R i = c i − 1 R_i=c_i-1 R i = c i − 1 satisfy ∑ i R i = N − d \sum_iR_i=N-d ∑ i R i = N − d , d = m + b − 1 d=m+b-1 d = m + b − 1 , and δ = N − max i R i ≥ d \delta=N-\max_iR_i\ge d δ = N − max i R i ≥ d . The volume envelope applies directly under (143) : the exponential factor arising from the denominator estimate is at most e 1 / 2 e^{1/2} e 1/2 , and the largest component complexity is at least g / b g/b g / b . Thus (2.12) gives
W ≤ P b ∏ i R i ! N ! n ! a 0 ! ∏ i a i ! , P b = B b + 2 b ( m − 1 ) ! . W\le P_b\frac{\prod_iR_i!}{N!}\frac{n!}{a_0!\prod_i a_i!},\qquad P_b=B^{b+2}\sqrt{b}\,(m-1)!. W ≤ P b N ! ∏ i R i ! a 0 ! ∏ i a i ! n ! , P b = B b + 2 b ( m − 1 )! .
with one absolute B ≥ 2 B\ge2 B ≥ 2 . The outside-cusp budget (2.13) is A o u t ≤ δ + 1 A_{\mathrm{out}}\le\delta+1 A out ≤ δ + 1 , so the single multinomial is bounded both by s δ + 1 s^{\delta+1} s δ + 1 and by ( b + 1 ) n (b+1)^n ( b + 1 ) n . The fixed-deficit count is at most b ( δ + 3 ) 2 ( b − 1 ) b(\delta+3)^{2(b-1)} b ( δ + 3 ) 2 ( b − 1 ) : choose the maximal block, the remaining indices of total δ − d \delta-d δ − d , and their cusp counts; their genera are then fixed. Ignoring the parity and availability restrictions only enlarges these positive sums.
It follows that the contribution at deficit δ \delta δ is at most P b b x δ P_bbx_\delta P b b x δ , where
x δ = s δ + 1 ( δ + 3 ) 2 b ( N δ ) . x_\delta=\frac{s^{\delta+1}(\delta+3)^{2b}}{\binom{N}{\delta}}. x δ = ( δ N ) s δ + 1 ( δ + 3 ) 2 b .
For D ∗ ≤ δ < J D_{*}\le\delta<J D ∗ ≤ δ < J , use b ≤ δ + 2 b\le\delta+2 b ≤ δ + 2 to obtain
x δ + 1 x δ = s ( δ + 1 ) N − δ ( 1 + 1 δ + 3 ) 2 b ≤ e 2 63 < 1 2 . \frac{x_{\delta+1}}{x_\delta}
=\frac{s(\delta+1)}{N-\delta}\left(1+\frac{1}{\delta+3}\right)^{2b}
\le\frac{e^{2}}{63}<\frac12. x δ x δ + 1 = N − δ s ( δ + 1 ) ( 1 + δ + 3 1 ) 2 b ≤ 63 e 2 < 2 1 .
The ratio includes the step from J − 1 J-1 J − 1 to J J J . Keeping ( N ) D ∗ ≥ g D ∗ (N)_{D_*} \ge g^{D_*} ( N ) D ∗ ≥ g D ∗ , the inclusive near sum is at most
2 P b b ( D ∗ + 3 ) 2 b D ∗ ! g − D ∗ s D ∗ + 1 ≤ 2 P b b ( D ∗ + 3 ) 2 b D ∗ ! g − r s r + 1 . (144) 2P_b b(D_*+3)^{2b}D_*!g^{-D_*}s^{D_*+1}\le2P_b b(D_*+3)^{2b}D_*!g^{-r}s^{r+1}. \tag*{(144)} 2 P b b ( D ∗ + 3 ) 2 b D ∗ ! g − D ∗ s D ∗ + 1 ≤ 2 P b b ( D ∗ + 3 ) 2 b D ∗ ! g − r s r + 1 . (144)
The last inequality uses g − D ∗ s D ∗ + 1 = g − r s r + 1 ( s / g ) D ∗ − r g^{-D_*}s^{D_*+1}=g^{-r}s^{r+1}(s/g)^{D_*-r} g − D ∗ s D ∗ + 1 = g − r s r + 1 ( s / g ) D ∗ − r . When b = 1 b=1 b = 1 , the only possible deficit is m ≤ J m\le J m ≤ J : the tail is empty if m < r m<r m < r , and otherwise is already included in (144) .
For b ≥ 2 b\ge2 b ≥ 2 and δ > J \delta>J δ > J , every R i < N − J R_i<N-J R i < N − J . The capped-factorial inequality (29) therefore applies, since ∑ i R i = N − d ≤ N \sum_i R_i=N-d\le N ∑ i R i = N − d ≤ N . The entire deep sector has at most ( N + 1 ) 2 b (N+1)^{2b} ( N + 1 ) 2 b allocations: each cusp count is at most n ≤ N n\le N n ≤ N , and each genus is at most g ≤ N g\le N g ≤ N , because b ≤ k − 2 σ b\le k-2\sigma b ≤ k − 2 σ in the genus equation. Its contribution is bounded by
P b ( b + 1 ) n ( N + 1 ) 2 b ( N J ) − 1 . (145) P_b(b+1)^n(N+1)^{2b}\binom{N}{J}^{-1}. \tag*{(145)} P b ( b + 1 ) n ( N + 1 ) 2 b ( J N ) − 1 . (145)
The product formula for the binomial coefficient now gives
( N J ) = ∏ j = 0 J − 1 N − j J − j ≥ ( N / J ) J ≥ ( 64 s ) J ≥ exp ( N L s 128 s ) . (146) \binom{N}{J}=\prod_{j=0}^{J-1}\frac{N-j}{J-j}\ge(N/J)^J\ge(64s)^J\ge\exp\left(\frac{NL_s}{128s}\right). \tag*{(146)} ( J N ) = j = 0 ∏ J − 1 J − j N − j ≥ ( N / J ) J ≥ ( 64 s ) J ≥ exp ( 128 s N L s ) . (146)
The factor log ( 64 s ) \log(64s) log ( 64 s ) in the exponent absorbs the cusp-label count at the additive threshold. To bound the deep sector by P b g − r P_b g^{-r} P b g − r , it remains to absorb the logarithm of g r ( b + 1 ) n ( N + 1 ) 2 b g^r(b+1)^n(N+1)^{2b} g r ( b + 1 ) n ( N + 1 ) 2 b . The inequalities r + 2 b ≤ 4 t r+2b\le4t r + 2 b ≤ 4 t , b + 1 ≤ 2 t b+1\le2t b + 1 ≤ 2 t and 2 b ≤ 3 t 2b\le3t 2 b ≤ 3 t give
r log g + n log ( b + 1 ) + 2 b log ( N + 1 ) ≤ F t ( g ) : = 4 t log g + s log ( 2 t ) + 3 t log 4. (147) r\log g+n\log(b+1)+2b\log(N+1)\le F_t(g):=4t\log g+s\log(2t)+3t\log4. \tag*{(147)} r log g + n log ( b + 1 ) + 2 b log ( N + 1 ) ≤ F t ( g ) := 4 t log g + s log ( 2 t ) + 3 t log 4. (147)
We first check this at the threshold value g 0 = K s ( s + t 2 ) g_0=Ks(s+t^2) g 0 = K s ( s + t 2 ) , then use monotonicity in g g g . The calculation holds for all real t , s ≥ 1 t,s\ge1 t , s ≥ 1 , including the case t = 1 t=1 t = 1 needed for the separating tail. Since s + t 2 ≤ 2 s t 2 s+t^2\le2st^2 s + t 2 ≤ 2 s t 2 , we have
s F t ( g 0 ) g 0 L s ≤ 4 t log ( 2 K ) + 8 t log s + 8 t log t + s log ( 2 t ) + 3 t log 4 K ( s + t 2 ) L s ≤ 1 K { 4 log ( 2 K ) + 3 log 4 log 64 + 8 + 8 e log 64 + 1 } < 1 128 . (148) \frac{sF_t(g_0)}{g_0L_s}\le\frac{4t\log(2K)+8t\log s+8t\log t+s\log(2t)+3t\log4}{K(s+t^2)L_s}
\le\frac{1}{K}\left\{\frac{4\log(2K)+3\log4}{\log64}+8+\frac{8}{e\log64}+1\right\}<\frac{1}{128}. \tag*{(148)} g 0 L s s F t ( g 0 ) ≤ K ( s + t 2 ) L s 4 t log ( 2 K ) + 8 t log s + 8 t log t + s log ( 2 t ) + 3 t log 4 ≤ K 1 { log 64 4 log ( 2 K ) + 3 log 4 + 8 + e log 64 8 + 1 } < 128 1 . (148)
For the second inequality use t / ( s + t 2 ) ≤ 1 / t ≤ 1 t/(s+t^2)\le1/t\le1 t / ( s + t 2 ) ≤ 1/ t ≤ 1 , L s ≥ log 64 L_s\ge\log64 L s ≥ log 64 , log s / L s ≤ 1 \log s/L_s\le1 log s / L s ≤ 1 , and ( log t ) / t ≤ 1 / e (\log t)/t\le1/e ( log t ) / t ≤ 1/ e . For the remaining label term, if t ≤ s t\le s t ≤ s then log ( 2 t ) ≤ L s \log(2t)\le L_s log ( 2 t ) ≤ L s ; if t > s t>s t > s , use s / ( s + t 2 ) ≤ 1 / t s/(s+t^2)\le1/t s / ( s + t 2 ) ≤ 1/ t and log ( 2 t ) ≤ t \log(2t)\le t log ( 2 t ) ≤ t . In both cases s log ( 2 t ) / ( ( s + t 2 ) L s ) ≤ 1 s\log(2t)/((s+t^2)L_s)\le1 s log ( 2 t ) / (( s + t 2 ) L s ) ≤ 1 .
The final strict bound holds for every K ≥ 4096 K\ge4096 K ≥ 4096 . At K = 4096 = 2 12 K=4096=2^{12} K = 4096 = 2 12 the first fraction in braces is 29 / 3 29/3 29/3 , and 8 / ( e log 64 ) < 4 / 3 8/(e\log64)<4/3 8/ ( e log 64 ) < 4/3 , so the braces are less than 20 and 20 / 4096 < 1 / 128 20/4096<1/128 20/4096 < 1/128 . The function log ( 2 K ) / K \log(2K)/K log ( 2 K ) / K and the other constant multiples of 1 / K 1/K 1/ K decrease on this range. For fixed t , s t,s t , s , also
( F t ( g ) g ) ′ = 4 t − 4 t log g − s log ( 2 t ) − 3 t log 4 g 2 < 0 ( g ≥ e ) . \left(\frac{F_t(g)}{g}\right)'=\frac{4t-4t\log g-s\log(2t)-3t\log4}{g^2}<0\qquad(g\ge e). ( g F t ( g ) ) ′ = g 2 4 t − 4 t log g − s log ( 2 t ) − 3 t log 4 < 0 ( g ≥ e ) .
Since g 0 > e g_0>e g 0 > e , (148) holds for every g ≥ g 0 g\ge g_0 g ≥ g 0 . It gives F t ( g ) < g L s / ( 128 s ) ≤ N L s / ( 128 s ) F_t(g)<gL_s/(128s)\le NL_s/(128s) F t ( g ) < g L s / ( 128 s ) ≤ N L s / ( 128 s ) . Equations (145) –(147) therefore bound the whole deep sector by P b g − r P_b g^{-r} P b g − r .
The remaining factors are independent of the ambient cusp count. For t ≥ 2 t\ge2 t ≥ 2 , one has
P b ≤ ( C t ) 3 t , 2 b ( D ∗ + 3 ) 2 b D ∗ ! ≤ ( 2 t ) 5 t , ∣ P k ∣ ≤ ( 4 t ) 4 t . P_b\le(Ct)^{3t},\qquad2b(D_*+3)^{2b}D_*! \le(2t)^{5t},\qquad|\mathcal{P}_k|\le(4t)^{4t}. P b ≤ ( C t ) 3 t , 2 b ( D ∗ + 3 ) 2 b D ∗ ! ≤ ( 2 t ) 5 t , ∣ P k ∣ ≤ ( 4 t ) 4 t .
The last count encodes each slot’s matched partner or zero and the least label of its exterior block; there is no extra b ! b! b ! . Summing the near and deep bounds proves (140) after enlarging C . C. C . For a fixed pattern, the numerator volumes have total boundary length 2 Λ − 2 ∑ y y e 2\Lambda-2\sum_y y_e 2Λ − 2 ∑ y y e . The boundary estimate costs at most e Λ e^\Lambda e Λ , and the actual-edge Jacobian is at most ( 1 + Λ ) k (1+\Lambda)^k ( 1 + Λ ) k . This proves the boundary assertion on the fixed pattern’s coordinate space, including zero lengths by continuity.
For the ordered separating tail use the same N , J , L s N,J,L_s N , J , L s , with g ≥ K ( s 2 + r 2 s ) g \ge K(s^2+r^2s) g ≥ K ( s 2 + r 2 s ) and r ≥ 1 r \ge1 r ≥ 1 . The analogues of (143) give J ≥ r J \ge r J ≥ r and N − r + 1 ≥ g N-r+1 \ge g N − r + 1 ≥ g . The volume envelope (67) applies under this threshold and bounds each zero-length summand by B ( N δ ) / ( N δ ) B\binom{N}{\delta}/\binom{N}{\delta} B ( δ N ) / ( δ N ) . A smaller side has at most δ + 1 \delta+1 δ + 1 cusp labels, and the two ordered placements give the fixed-deficit bound 6 B s δ + 1 / ( N δ ) 6B s^{\delta+1}/\binom{N}{\delta} 6 B s δ + 1 / ( δ N ) . Its successive ratio is at most 1 / 63 1/63 1/63 on r ≤ δ < J r \le\delta< J r ≤ δ < J . Thus the inclusive near sum is bounded by 12 B r ! g − r s r + 1 12Br!g^{-r}s^{r+1} 12 B r ! g − r s r + 1 .
For δ > J \delta>J δ > J , binomial unimodality and symmetry give ( N δ ) ≥ ( N J ) \binom{N}{\delta}\ge\binom{N}{J} ( δ N ) ≥ ( J N ) , including the equal central binomial coefficients when N N N is odd. The original ordered cusp-label sum is at most ( g + 1 ) 2 n (g+1)2^n ( g + 1 ) 2 n , so the deep part is at most
B ( g + 1 ) 2 n exp ( − N L s 128 s ) . B(g+1)2^n\exp\left(-\frac{NL_s}{128s}\right). B ( g + 1 ) 2 n exp ( − 128 s N L s ) .
There is no further factor two in this ordered sum. The logarithm of g r ( g + 1 ) 2 n g^r(g+1)2^n g r ( g + 1 ) 2 n is at most
( r + 1 ) log g + s log 2 ≤ 2 r log g + s log 2 ≤ F r ( g ) . (r+1)\log g+s\log2\le2r\log g+s\log2\le F_r(g). ( r + 1 ) log g + s log 2 ≤ 2 r log g + s log 2 ≤ F r ( g ) .
Apply (148) with t = r ≥ 1 t=r\ge1 t = r ≥ 1 to bound this deep part by B g − r Bg^{-r} B g − r . Adding the near sum, absorbing r ! r! r ! into [ C r ] C r [Cr]^{Cr} [ C r ] C r , and inserting the two boundary factors e x / 2 e^{x/2} e x /2 proves (4.18)$. □
Proposition 4.7 (Kernel errors at an additive threshold). The estimates (133) –(134) hold under g ≥ C s 2 + [ C r ] C s g\ge C^{s^2+[Cr]^{C}s} g ≥ C s 2 + [ C r ] C s , where r = R + 1 r=R+1 r = R + 1 . For a fixed carrier and pattern, (135) holds under g ≥ C s 2 + [ C t ] C s g\ge C^{s^2+[Ct]^{C}s} g ≥ C s 2 + [ C t ] C s , where t = R + m + 1 t=R+m+1 t = R + m + 1 . Here C C C is a new absolute constant; all coefficient functions, error powers and length factors are unchanged.
Proof. We reuse the retained-record calculations from the proof of Proposition 4.4, whose local thresholds are linear in s s s , and replace its tail input by Lemma 4.6.
For the nonseparating kernel, Lemma 4.1, a genus shift by one and the scalar–boundary product argument apply under g ≥ [ C r ] C s g\ge[Cr]^Cs g ≥ [ C r ] C s . They give (133) for all R ≥ 0 R\ge0 R ≥ 0 . For a separating endpoint of deficit 1 ≤ δ ≤ R 1\le\delta\le R 1 ≤ δ ≤ R , use Lemma 4.2 with J = R − δ J=R-\delta J = R − δ and H 0 = h 0 ≤ ( δ + 1 ) / 2 H_0=h_0\le(\delta+1)/2 H 0 = h 0 ≤ ( δ + 1 ) /2 , so L = R + h 0 + 1 ≤ 2 r L=R+h_0+1\le2r L = R + h 0 + 1 ≤ 2 r . Its threshold is again linear in s s s . Multiplication by the same small-component prefactor gives the error in (134) ; the omitted endpoints are covered by (141) .
For a retained carrier record δ ≤ R \delta\le R δ ≤ R , put J = R − δ J=R-\delta J = R − δ . The exact record identities imply
H 0 ≤ ( t + 1 ) / 2 ≤ t , L = J + H 0 + δ + 1 = R + H 0 + 1 ≤ 2 t , H_0\le(t+1)/2\le t,\qquad L=J+H_0+\delta+1=R+H_0+1\le2t, H 0 ≤ ( t + 1 ) /2 ≤ t , L = J + H 0 + δ + 1 = R + H 0 + 1 ≤ 2 t ,
p = n − A o u t + k ∗ ≥ k ∗ ≥ 1. (149) p=n-A_{\mathrm{out}}+k_*\ge k_*\ge1. \tag*{(149)} p = n − A out + k ∗ ≥ k ∗ ≥ 1. (149)
All scalar marked counts p ν p_\nu p ν are nonnegative and at most n + 2 H 0 + δ ≤ n + 3 t n+2H_0+\delta\le n+3t n + 2 H 0 + δ ≤ n + 3 t . For every source order J ≤ t J\le t J ≤ t , the exact threshold in Proposition 2.3 is consequently bounded by
G J ( p ν ) ≤ 3000 ( 2 c 0 ) c 0 t c 0 + 1 s . G_J(p_\nu)\le3000(2c_0)^{c_0}t^{c_0+1}s. G J ( p ν ) ≤ 3000 ( 2 c 0 ) c 0 t c 0 + 1 s .
The graded majorants, genus shifts and at most 2 L 2L 2 L scalar factors in (43) therefore have a common base A = [ C 1 L ] C 1 s A=[C_1L]^{C_1}s A = [ C 1 L ] C 1 s , with one auxiliary absolute C 1 C_1 C 1 . This is the base used in Lemma 4.2; neither finite multiplication nor shifting raises the power of s s s . The geometric denominator coefficients are covered by the same base, including negative values of n − δ + q − 2 n-\delta+q-2 n − δ + q − 2 .
Choose C 1 C_1 C 1 after the fixed enlargements in that lemma. Requiring g ≥ 4 A g\ge4A g ≥ 4 A ensures that the scalar partial sums are bounded, that g − H 0 ≥ g / 2 g-H_0\ge g/2 g − H 0 ≥ g /2 , that every strict source threshold g − H 0 > G J ( p ν ) g-H_0>G_J(p_\nu) g − H 0 > G J ( p ν ) holds, and that g − H 0 ≥ 4 g-H_0\ge4 g − H 0 ≥ 4 . The last condition ensures that the predecessor genera in the source estimates are at least three. The exact boundary ratio is bounded by e X ∗ / 2 e^{X^*/2} e X ∗ /2 , so the scalar smallness condition is independent of boundary length. Thus (126) applies directly.
Multiply it by (136) . The single label factor gives s A o u t + J + 1 ≤ s R + 2 s^{A_{\mathrm{out}}+J+1}\le s^{R+2} s A out + J + 1 ≤ s R + 2 ; the boundary exponentials give e Λ e^\Lambda e Λ ; and k + 2 ( R − δ ) + 2 ≤ 2 t k+2(R-\delta)+2\le2t k + 2 ( R − δ ) + 2 ≤ 2 t because δ ≥ m \delta\ge m δ ≥ m . The record count and exact small-volume factors cost [ C t ] C t [C^t]^{C_t} [ C t ] C t . Adding the fixed-pattern boundary tail of Lemma 4.6 proves (135) .
To verify all thresholds simultaneously, choose the final C C C after K K K , C 1 C_1 C 1 and the constants of the local graded lemmas. For every u ≥ 1 u\ge1 u ≥ 1 it may be chosen to satisfy
C ≥ K , [ C u ] C ≥ K u 2 , [ C u ] C ≥ 4 [ 2 C 1 u ] C 1 , [ C u ] C ≥ 8 ( u + 2 ) , (150) C\ge K,\qquad[Cu]^C\ge Ku^2,\qquad[Cu]^C\ge4[2C_1u]^{C_1},\qquad[Cu]^C\ge8(u+2), \tag*{(150)} C ≥ K , [ C u ] C ≥ K u 2 , [ C u ] C ≥ 4 [ 2 C 1 u ] C 1 , [ C u ] C ≥ 8 ( u + 2 ) , (150)
and to dominate the other finitely many fixed polynomial requirements in u u u . With u = t u=t u = t or r r r , the additive threshold supplies both the new tail threshold and every local condition above. It also gives N > 2 R N>2R N > 2 R and g > H 0 g>H_0 g > H 0 , so retained records reconstruct a unique largest stable complementary component. Unavailable cusp allocations are omitted before any volume is evaluated, and the available boundary ratio has p ≥ 1 p\ge1 p ≥ 1 . The scalar paths alone may use zero marked points.
The finite retained sums are exactly (56) . When R = 0 R=0 R = 0 or m > R m>R m > R they are empty and the new positive tail supplies the full error. For R = 0 R=0 R = 0 , the separating estimate likewise follows entirely from the positive tail bound, so no shifted estimate of negative order is needed. □
Theorem 4.8 (All-order orbital remainder at an additive threshold). There is an absolute C ≥ 8 C\ge8 C ≥ 8 such that, for all integers R R R , n ≥ 0 n\ge0 n ≥ 0 , T ≥ 0 T\ge0 T ≥ 0 , and bounded real even Borel H H H supported in [ − T , T ] [-T,T] [ − T , T ] , putting ρ = R + ⌈ T ⌉ + 1 \rho=R+\lceil T\rceil+1 ρ = R + ⌈ T ⌉ + 1 , the condition
g ≥ C ( n + 1 ) 2 + [ C ρ ] C ( n + 1 ) (151) g\ge C(n+1)^2+[C\rho]^C(n+1) \tag*{(151)} g ≥ C ( n + 1 ) 2 + [ C ρ ] C ( n + 1 ) (151) implies
∣ E g , n O H − Ω ( H ) − ∑ j = 1 R g − j κ j , n ( H ) ∣ ≤ g − R − 1 ( n + 1 ) R + 2 [ C ρ ] C ρ e 3 T ∥ H ∥ ∞ . (152) \left|\mathbb{E}_{g,n}\mathcal{O}_H-\Omega(H)-\sum_{j=1}^{R}g^{-j}\kappa_{j,n}(H)\right|
\le g^{-R-1}(n+1)^{R+2}[C\rho]^C\rho e^{3T}\lVert H\rVert_\infty. \tag*{(152)} E g , n O H − Ω ( H ) − j = 1 ∑ R g − j κ j , n ( H ) ≤ g − R − 1 ( n + 1 ) R + 2 [ C ρ ] C ρ e 3 T ∥ H ∥ ∞ . (152) The sum is empty for R = 0 R=0 R = 0 . The coefficients are those of Theorem 3.5 .
Proof. Enlarge C C C beyond the constant of Proposition 4.7 so that (151) , with s , ρ ≥ 1 s,\rho\ge1 s , ρ ≥ 1 , also implies g ≥ 4 ρ 2 g\ge4\rho^2 g ≥ 4 ρ 2 . Since T ≤ ρ − 1 T\le\rho-1 T ≤ ρ − 1 , π ( g − 1 + n / 2 ) > T \pi(g-1+n/2)>T π ( g − 1 + n /2 ) > T ; thus (95) excludes every carrier without a stable exterior, including entirely self-paired carriers. Put M = ⌊ 2 T / π ⌋ M=\lfloor2T/\pi\rfloor M = ⌊ 2 T / π ⌋ . The exhaustion (98) has 1 ≤ m ≤ M ≤ T 1\le m\le M\le T 1 ≤ m ≤ M ≤ T , so r = R + 1 ≤ ρ r=R+1\le\rho r = R + 1 ≤ ρ and t = R + m + 1 ≤ ρ t=R+m+1\le\rho t = R + m + 1 ≤ ρ . Proposition 4.7 therefore supplies (133) –(135) under the common additive threshold, with unchanged coefficients, length factors and cusp powers. Repeat the integration and finite summation in the proof of Theorem 4.5 , using these kernel bounds. The absolute repetition estimate following (76) , together with (82) and (90) , gives the same error bound and absolute integrability for bounded Borel H H H ; hence signed Fubini applies without another genus condition. The two vanishing properties in (89) identify the nonsimple coefficient sum with ζ j , n \zeta_{j,n} ζ j , n , even when j > M j>M j > M or M > R M>R M > R . Adding the simple part gives κ j , n = η j , n + ζ j , n \kappa_{j,n}=\eta_{j,n}+\zeta_{j,n} κ j , n = η j , n + ζ j , n and proves (152) , including R = 0 R=0 R = 0 with empty sums. If M = 0 M=0 M = 0 , the nonsimple contribution and all its coefficients on H H H vanish; if T = 0 T=0 T = 0 , all terms vanish, regardless of H ( 0 ) H(0) H ( 0 ) . □
The coefficient of ( n + 1 ) 2 (n+1)^2 ( n + 1 ) 2 in the genus threshold is now independent of order and support. The remaining order-dependent term is linear in n + 1 n+1 n + 1 . § 6.7 transfers this estimate to polynomials of growing degree; § 6.9 uses that transfer to bound the spectral count.
The leading measure and a common truncated trace
Positive pre-trace estimates on domains truncated at the cusps were used by Hide [8 Sections 2.4–2.5] . We derive an exact trace formula at one cusp height for all polynomial tests of bounded degree. The leading orbital functional is represented by a unit atom at a a a and a finite signed measure on [ 0 , 1 ] [0,1] [ 0 , 1 ] . Subtracting the constant eigenfunction cancels the atom up to the explicit mass removed by cusp truncation.
Throughout, h = b 0 ∗ b 0 h=b_0*b_0 h = b 0 ∗ b 0 , f = b 0 ∗ h ^ f=\widehat{b_0*h} f = b 0 ∗ h , supp h ⊂ [ − L , L ] \operatorname{supp} h\subset[-L,L] supp h ⊂ [ − L , L ] , and a = f ( i / 2 ) > 1 a=f(i/2)>1 a = f ( i /2 ) > 1 are fixed as in § 1.2 . The degree bound Q ≥ 1 Q\ge1 Q ≥ 1 is an integer, and polynomial tests are real and vanish at zero. We use
H ^ ( r ) = ∫ R H ( u ) e − i r u d u , F ˇ ( u ) = 1 2 π ∫ R F ( r ) e i r u d r . (153) \widehat{H}(r)=\int_{\mathbb{R}}H(u)e^{-iru}\,du,\qquad\check{F}(u)=\frac{1}{2\pi}\int_{\mathbb{R}}F(r)e^{iru}\,dr. \tag*{(153)} H ( r ) = ∫ R H ( u ) e − i r u d u , F ˇ ( u ) = 2 π 1 ∫ R F ( r ) e i r u d r . (153)
In particular, 0 ≤ f ( r ) ≤ 1 0\le f(r)\le1 0 ≤ f ( r ) ≤ 1 on R \mathbb{R} R , and 1 ≤ f ( i s ) ≤ a 1\le f(is)\le a 1 ≤ f ( i s ) ≤ a for 0 ≤ s ≤ 1 / 2 0\le s\le1/2 0 ≤ s ≤ 1/2 . The latter assertion follows from f ( i s ) = ∫ h ( u ) cosh ( s u ) d u f(is)=\int h(u)\cosh(su)\,du f ( i s ) = ∫ h ( u ) cosh ( s u ) d u and h ≥ 0 h\ge0 h ≥ 0 . For a polynomial p p p with p ( 0 ) = 0 p(0)=0 p ( 0 ) = 0 , put
H p = ( p ∘ f ) ˇ , p ( x ) = ∑ t = 1 Q c t x t ⟹ H p = ∑ t = 1 Q c t h ∗ t , supp H p ⊂ [ − L Q , L Q ] . (154) H_p=\check{(p\circ f)},\qquad p(x)=\sum_{t=1}^{Q}c_tx^t\quad\Longrightarrow\quad H_p=\sum_{t=1}^{Q}c_t h^{*t},\qquad\operatorname{supp}H_p\subset[-LQ,LQ]. \tag*{(154)} H p = ( p ∘ f ) ˇ , p ( x ) = t = 1 ∑ Q c t x t ⟹ H p = t = 1 ∑ Q c t h ∗ t , supp H p ⊂ [ − L Q , L Q ] . (154)
A finite signed measure for the leading functional
The separation of the term p ( a ) p(a) p ( a ) also occurs in the proof of [10 Lemma 3.3] . The argument below represents the remainder by a finite signed measure on [ 0 , 1 ] [0,1] [ 0 , 1 ] and bounds its total variation.
Proposition 5.1 (Leading orbital measure). There is a finite signed real measure μ 0 \mu_0 μ 0 , independent of n n n , such that for every polynomial p p p vanishing at zero,
ω 1 , n ( p ) = Ω ( H p ) = p ( a ) + ∫ [ 0 , 1 ] p ( x ) d μ 0 ( x ) , \omega_{1,n}(p)=\Omega(H_p)=p(a)+\int_{[0,1]}p(x)\,d\mu_0(x), ω 1 , n ( p ) = Ω ( H p ) = p ( a ) + ∫ [ 0 , 1 ] p ( x ) d μ 0 ( x ) , supp μ 0 ⊂ [ 0 , 1 ] , μ 0 ( R ) = − 1 , ∥ μ 0 ∥ T V < 5. (155) \operatorname{supp}\mu_0\subset[0,1],\qquad\mu_0(\mathbb{R})=-1,\qquad\lVert\mu_0\rVert_{\mathrm{TV}}<5. \tag*{(155)} supp μ 0 ⊂ [ 0 , 1 ] , μ 0 ( R ) = − 1 , ∥ μ 0 ∥ TV < 5. (155)
Consequently ω 1 , n = ω 1 \omega_{1,n}=\omega_1 ω 1 , n = ω 1 is independent of the cusp number. The measure μ f = δ a + μ 0 \mu_f=\delta_a+\mu_0 μ f = δ a + μ 0 has mass zero and variation less than six; its atom at a a a has coefficient exactly one.
Proof. The integral in (77) is
Ω ( H ) = ∫ 0 ∞ ∑ k ≥ 1 2 sinh 2 ( x / 2 ) sinh ( k x / 2 ) H ( k x ) d x . \Omega(H)=\int_0^\infty\sum_{k\ge1}\frac{2\sinh^2(x/2)}{\sinh(kx/2)}H(kx)\,dx. Ω ( H ) = ∫ 0 ∞ k ≥ 1 ∑ sinh ( k x /2 ) 2 sinh 2 ( x /2 ) H ( k x ) d x .
Changing variables u = k x u=kx u = k x gives
Ω ( H ) = ∫ 0 ∞ A ( u ) H ( u ) d u , \Omega(H)=\int_0^\infty A(u)H(u)\,du, Ω ( H ) = ∫ 0 ∞ A ( u ) H ( u ) d u ,
A ( u ) = ∑ k ≥ 1 2 k sinh 2 ( u / ( 2 k ) ) sinh ( u / 2 ) = 2 cosh ( u / 2 ) + R ( u ) , (156) A(u)=\sum_{k\ge1}\frac{2}{k}\frac{\sinh^2(u/(2k))}{\sinh(u/2)}
=2\cosh(u/2)+R(u), \tag*{(156)} A ( u ) = k ≥ 1 ∑ k 2 sinh ( u /2 ) sinh 2 ( u / ( 2 k )) = 2 cosh ( u /2 ) + R ( u ) , (156)
R ( u ) = − 2 e − u / 2 + ∑ k ≥ 2 2 k sinh 2 ( u / ( 2 k ) ) sinh ( u / 2 ) , R ( 0 ) = − 2. (157) R(u)=-2e^{-u/2}+\sum_{k\ge2}\frac{2}{k}\frac{\sinh^2(u/(2k))}{\sinh(u/2)},\qquad R(0)=-2. \tag*{(157)} R ( u ) = − 2 e − u /2 + k ≥ 2 ∑ k 2 sinh ( u /2 ) sinh 2 ( u / ( 2 k )) , R ( 0 ) = − 2. (157)
Indeed, sinh 2 v ≤ 1 2 sinh ( 2 v ) \sinh^2v\le\frac{1}{2}\sinh(2v) sinh 2 v ≤ 2 1 sinh ( 2 v ) and convexity of sinh \sinh sinh imply sinh 2 ( u / ( 2 k ) ) ≤ k − 1 sinh ( u / 2 ) \sinh^2(u/(2k))\le k^{-1}\sinh(u/2) sinh 2 ( u / ( 2 k )) ≤ k − 1 sinh ( u /2 ) for k ≥ 2 k\ge2 k ≥ 2 . Thus these summands are bounded by 2 / k 2 2/k^2 2/ k 2 ; their individual limits at zero vanish.
This proves continuity of R R R at zero and justifies the change of variables for signed compactly supported H H H by an integrable absolute majorant.
Let P s P_s P s be the Cauchy probability measure
P s ( d r ) = s π ( s 2 + r 2 ) d r ( s > 0 ) , P 0 = δ 0 , ∫ R e − i r u d P s ( r ) = e − s ∣ u ∣ . (158) P_s(dr)=\frac{s}{\pi(s^2+r^2)}\,dr\quad(s>0),\qquad P_0=\delta_0,\qquad\int_{\mathbb{R}}e^{-iru}\,dP_s(r)=e^{-s|u|}. \tag*{(158)} P s ( d r ) = π ( s 2 + r 2 ) s d r ( s > 0 ) , P 0 = δ 0 , ∫ R e − i r u d P s ( r ) = e − s ∣ u ∣ . (158)
The expansion of the hyperbolic denominator leads to symmetric second differences of the Cauchy measures. Set s j = j + 1 / 2 s_j=j+1/2 s j = j + 1/2 , d k = 1 / k d_k=1/k d k = 1/ k , and consider
ρ = − 2 P 1 / 2 + ∑ k = 2 ∞ 1 k ∑ j = 0 ∞ ( P s j − d k + P s j + d k − 2 P s j ) . (159) \rho=-2P_{1/2}+\sum_{k=2}^{\infty}\frac{1}{k}\sum_{j=0}^{\infty}\left(P_{s_j-d_k}+P_{s_j+d_k}-2P_{s_j}\right). \tag*{(159)} ρ = − 2 P 1/2 + k = 2 ∑ ∞ k 1 j = 0 ∑ ∞ ( P s j − d k + P s j + d k − 2 P s j ) . (159)
We first prove convergence in total variation, including the endpoint s 0 = d 2 s_0=d_2 s 0 = d 2 . Differentiation of the density gives
∂ s 2 P s ( d r ) = 2 s ( s 2 − 3 r 2 ) π ( s 2 + r 2 ) 3 d r , ∥ ∂ s 2 P s ∥ T V ≤ 3 2 s 2 . \partial_s^2P_s(dr)=\frac{2s(s^2-3r^2)}{\pi(s^2+r^2)^3}\,dr,\qquad\left\|\partial_s^2P_s\right\|_{\mathrm{TV}}\leq\frac{3}{2s^2}. ∂ s 2 P s ( d r ) = π ( s 2 + r 2 ) 3 2 s ( s 2 − 3 r 2 ) d r , ∂ s 2 P s TV ≤ 2 s 2 3 .
For the inequality, replace ∣ s 2 − 3 r 2 ∣ \lvert s^2-3r^2\rvert ∣ s 2 − 3 r 2 ∣ by s 2 + 3 r 2 s^2+3r^2 s 2 + 3 r 2 , scale r = s v r=sv r = s v , and use ∫ R ( 1 + v 2 ) − 3 d v = 3 π / 8 \int_{\mathbb{R}}(1+v^2)^{-3}\,dv=3\pi/8 ∫ R ( 1 + v 2 ) − 3 d v = 3 π /8 and ∫ R v 2 ( 1 + v 2 ) − 3 d v = π / 8 \int_{\mathbb{R}}v^2(1+v^2)^{-3}\,dv=\pi/8 ∫ R v 2 ( 1 + v 2 ) − 3 d v = π /8 . The integral remainder for a symmetric second difference in L 1 ( R ) L^1(\mathbb{R}) L 1 ( R ) therefore gives, for 0 < d < s 0<d<s 0 < d < s ,
∥ P s − d + P s + d − 2 P s ∥ T V ≤ ∫ − d d ( d − ∣ v ∣ ) 3 2 ( s + v ) 2 d v ≤ 3 d 2 2 ( s − d ) 2 . (160) \left\|P_{s-d}+P_{s+d}-2P_s\right\|_{\mathrm{TV}}\leq\int_{-d}^{d}(d-|v|)\frac{3}{2(s+v)^2}\,dv\leq\frac{3d^2}{2(s-d)^2}. \tag*{(160)} ∥ P s − d + P s + d − 2 P s ∥ TV ≤ ∫ − d d ( d − ∣ v ∣ ) 2 ( s + v ) 2 3 d v ≤ 2 ( s − d ) 2 3 d 2 . (160)
When k ≥ 3 k\geq3 k ≥ 3 , s j − d k ≥ s j / 3 s_j-d_k\geq s_j/3 s j − d k ≥ s j /3 ; when k = 2 k=2 k = 2 , j ≥ 1 j\geq1 j ≥ 1 , s j − 1 / 2 ≥ 2 s j / 3 s_j-1/2\geq2s_j/3 s j − 1/2 ≥ 2 s j /3 . Since ∑ j ≥ 0 ( j + 1 / 2 ) − 2 = π 2 / 2 \sum_{j\geq0}(j+1/2)^{-2}=\pi^2/2 ∑ j ≥ 0 ( j + 1/2 ) − 2 = π 2 /2 , their respective contributions to the variation sum are at most
27 π 2 4 ( ζ ( 3 ) − 9 8 ) , 27 64 ( π 2 2 − 4 ) . \frac{27\pi^2}{4}\left(\zeta(3)-\frac{9}{8}\right),\qquad\frac{27}{64}\left(\frac{\pi^2}{2}-4\right). 4 27 π 2 ( ζ ( 3 ) − 8 9 ) , 64 27 ( 2 π 2 − 4 ) .
The term ( k , j ) = ( 2 , 0 ) (k,j)=(2,0) ( k , j ) = ( 2 , 0 ) has variation at most 2 2 2 , and the initial term − 2 P 1 / 2 -2P_{1/2} − 2 P 1/2 has variation 2 2 2 . Hence
1 2 ∥ ρ ∥ T V ≤ 2 + 27 π 2 8 ( ζ ( 3 ) − 9 8 ) + 27 128 ( π 2 2 − 4 ) < 5. (161) \frac{1}{2}\|\rho\|_{\mathrm{TV}}\leq2+\frac{27\pi^2}{8}\left(\zeta(3)-\frac{9}{8}\right)+\frac{27}{128}\left(\frac{\pi^2}{2}-4\right)<5. \tag*{(161)} 2 1 ∥ ρ ∥ TV ≤ 2 + 8 27 π 2 ( ζ ( 3 ) − 8 9 ) + 128 27 ( 2 π 2 − 4 ) < 5. (161)
To verify the last inequality using rational bounds, the integral test gives ζ ( 3 ) ≤ ∑ k = 1 10 k − 3 + 1 / 200 < 1203 / 1000 \zeta(3)\leq\sum_{k=1}^{10}k^{-3}+1/200<1203/1000 ζ ( 3 ) ≤ ∑ k = 1 10 k − 3 + 1/200 < 1203/1000 ; together with π 2 < 10 \pi^2<10 π 2 < 10 , the right side is less than 15499 / 3200 < 5 15499/3200<5 15499/3200 < 5 .
For u > 0 u>0 u > 0 , expansion of 1 / sinh ( u / 2 ) 1/\sinh(u/2) 1/ sinh ( u /2 ) gives
2 k sinh 2 ( u / ( 2 k ) ) sinh ( u / 2 ) = 1 k ∑ j ≥ 0 ( e − ( s j − d k ) u + e − ( s j + d k ) u − 2 e − s j u ) . (162) \frac{2}{k}\frac{\sinh^2(u/(2k))}{\sinh(u/2)}=\frac{1}{k}\sum_{j\geq0}\left(e^{-(s_j-d_k)u}+e^{-(s_j+d_k)u}-2e^{-s_ju}\right). \tag*{(162)} k 2 sinh ( u /2 ) sinh 2 ( u / ( 2 k )) = k 1 j ≥ 0 ∑ ( e − ( s j − d k ) u + e − ( s j + d k ) u − 2 e − s j u ) . (162)
The variation convergence just proved permits termwise Fourier transformation of (159) . Equations (158) and (162) imply
∫ R e − i r u d ρ ( r ) = R ( ∣ u ∣ ) , ρ ( R ) = − 2. (163) \int_{\mathbb{R}}e^{-iru}\,d\rho(r)=R(|u|),\qquad\rho(\mathbb{R})=-2. \tag*{(163)} ∫ R e − i r u d ρ ( r ) = R ( ∣ u ∣ ) , ρ ( R ) = − 2. (163)
The mass also follows directly because every second difference has mass zero. Define
μ 0 = 1 2 f ∗ ρ , μ f = δ a + μ 0 . (164) \mu_0=\frac{1}{2}f_*\rho,\qquad\mu_f=\delta_a+\mu_0. \tag*{(164)} μ 0 = 2 1 f ∗ ρ , μ f = δ a + μ 0 . (164)
The real map f f f takes values in [ 0 , 1 ] [0,1] [ 0 , 1 ] , and pushforward does not increase variation. Thus the asserted support, mass and norm follow.
Finally, evenness and compact support of H p H_p H p give
Ω ( H p ) = ∫ 0 ∞ 2 cosh ( u / 2 ) H p ( u ) d u + 1 2 ∫ R R ( ∣ u ∣ ) H p ( u ) d u = p ( f ( i / 2 ) ) + 1 2 ∫ R p ( f ( r ) ) d ρ ( r ) = p ( a ) + μ 0 ( p ) . (165) \begin{aligned}
\Omega(H_p)=\int_{0}^{\infty}2\cosh(u/2)H_p(u)\,du+\frac{1}{2}\int_{\mathbb{R}}R(|u|)H_p(u)\,du \\
&=p(f(i/2))+\frac{1}{2}\int_{\mathbb{R}}p(f(r))\,d\rho(r)=p(a)+\mu_0(p). \tag*{(165)}
\end{aligned} Ω ( H p ) = ∫ 0 ∞ 2 cosh ( u /2 ) H p ( u ) d u + 2 1 ∫ R R ( ∣ u ∣ ) H p ( u ) d u = p ( f ( i /2 )) + 2 1 ∫ R p ( f ( r )) d ρ ( r ) = p ( a ) + μ 0 ( p ) . (165)
The first equality in the second line uses the entire Fourier transform of H p H_p H p ; the second term uses Fubini, justified by H p ∈ L 1 H_p\in L^1 H p ∈ L 1 and ρ \rho ρ finite. Because a > 1 a>1 a > 1 , f ∗ ρ f_*\rho f ∗ ρ has no mass at a a a . Thus μ f ( { a } ) = 1 \mu_f(\{a\})=1 μ f ({ a }) = 1 .
The endpoint term ( k , j ) = ( 2 , 0 ) (k,j)=(2,0) ( k , j ) = ( 2 , 0 ) in (159) contributes 1 4 δ f ( 0 ) = 1 4 δ 1 \frac{1}{4}\delta_{f(0)}=\frac{1}{4}\delta_1 4 1 δ f ( 0 ) = 4 1 δ 1 to μ 0 \mu_0 μ 0 . Since h ( 0 ) = ∫ R b 0 ( u ) 2 d u > 0 h(0)=\int_{\mathbb{R}}b_0(u)^2\,du>0 h ( 0 ) = ∫ R b 0 ( u ) 2 d u > 0 , the function h h h is positive near zero, so f ( r ) < 1 f(r)<1 f ( r ) < 1 for r ≠ 0 r\ne0 r = 0 . The measure ρ − 1 2 δ 0 \rho-\frac{1}{2}\delta_0 ρ − 2 1 δ 0 is absolutely continuous by total-variation convergence. Hence μ 0 ( { 1 } ) = 1 / 4 \mu_0(\{1\})=1/4 μ 0 ({ 1 }) = 1/4 , although its total mass is − 1 -1 − 1 .
The functional q ↦ ∫ q d μ f q\mapsto\int q\,d\mu_f q ↦ ∫ q d μ f extends ω 1 \omega_1 ω 1 continuously to C ( [ 0 , a ] ) C([0,a]) C ([ 0 , a ]) . Two finite signed measures on [ 0 , a ] [0,a] [ 0 , a ] agreeing on polynomials vanishing at zero differ by a multiple of δ 0 \delta_0 δ 0 , by polynomial density. Requiring total mass zero fixes this ambiguity.
Abel kernels and localized spectral multipliers
Let X = Γ \ H X=\Gamma\backslash\mathbb{H} X = Γ\ H be a connected complete hyperbolic surface of finite area, with n ≥ 1 n\ge1 n ≥ 1 cusps and no orbifold points, where Γ ⊂ PSL 2 ( R ) \Gamma\subset\operatorname{PSL}_2(\mathbb{R}) Γ ⊂ PSL 2 ( R ) is torsion-free. We use d μ = d x d y / y 2 d\mu=dx\,dy/y^2 d μ = d x d y / y 2 and Δ = − y 2 ( ∂ x 2 + ∂ y 2 ) \Delta=-y^2(\partial_x^2+\partial_y^2) Δ = − y 2 ( ∂ x 2 + ∂ y 2 ) . Its area is A g , n = 2 π ( 2 g − 2 + n ) A_{g,n}=2\pi(2g-2+n) A g , n = 2 π ( 2 g − 2 + n ) . For each cusp b b b , let Γ b \Gamma_b Γ b be its full stabilizer and choose a scaling matrix σ b \sigma_b σ b such that σ b − 1 Γ b σ b = ⟨ U ⟩ \sigma_b^{-1}\Gamma_b\sigma_b=\langle U\rangle σ b − 1 Γ b σ b = ⟨ U ⟩ , where U : z ↦ z + 1 U:z\mapsto z+1 U : z ↦ z + 1 .
We use the Abel transform to pass from the one-dimensional test H H H to a radial kernel on H \mathbb{H} H . For T > 0 T>0 T > 0 and even H ∈ C c ∞ ( R ) H\in C^\infty_c(\mathbb{R}) H ∈ C c ∞ ( R ) supported in [ − T , T ] [-T,T] [ − T , T ] , define, for ρ ≥ 0 \rho\ge0 ρ ≥ 0 ,
K H ( ρ ) = − 1 2 π ∫ ρ ∞ H ′ ( u ) cosh u − cosh ρ d u . (166) K_H(\rho)=-\frac{1}{\sqrt{2\pi}}\int_{\rho}^{\infty}\frac{H'(u)}{\sqrt{\cosh u-\cosh\rho}}\,du. \tag*{(166)} K H ( ρ ) = − 2 π 1 ∫ ρ ∞ cosh u − cosh ρ H ′ ( u ) d u . (166)
This agrees with the normalization in [7 §5, (5.3)–(5.4)] . The kernel may change sign; positivity of A X A_X A X below will come from the fixed multiplier f f f . To check smoothness at the origin, the function G ( v ) = H ′ ( arcosh v ) / v 2 − 1 G(v)=H'(\operatorname{arcosh}v)/\sqrt{v^2-1} G ( v ) = H ′ ( arcosh v ) / v 2 − 1 extends smoothly to v = 1 v=1 v = 1 : H ′ ( u ) / sinh u H'(u)/\sinh u H ′ ( u ) / sinh u is smooth and even near zero. Changing variables in (166) gives
K H ( ρ ) = − 1 2 π ∫ 0 ∞ G ( cosh ρ + v ) v − 1 / 2 d v . K_H(\rho)=-\frac{1}{\sqrt{2\pi}}\int_{0}^{\infty}G(\cosh\rho+v)v^{-1/2}\,dv. K H ( ρ ) = − 2 π 1 ∫ 0 ∞ G ( cosh ρ + v ) v − 1/2 d v .
Differentiation with respect to cosh ρ \cosh\rho cosh ρ is justified by the smoothness and compact support of G G G and the integrability of v − 1 / 2 v^{-1/2} v − 1/2 near zero. Thus K H K_H K H is a smooth radial kernel supported on [ 0 , T ] [0,T] [ 0 , T ] , flat at the upper endpoint. Its forward Abel transform is
H ( u ) = 2 ∫ ∣ u ∣ ∞ K H ( ρ ) sinh ρ cosh ρ − cosh u d ρ . (167) H(u)=\sqrt{2}\int_{|u|}^{\infty}\frac{K_H(\rho)\sinh\rho}{\sqrt{\cosh\rho-\cosh u}}\,d\rho. \tag*{(167)} H ( u ) = 2 ∫ ∣ u ∣ ∞ cosh ρ − cosh u K H ( ρ ) sinh ρ d ρ . (167)
Indeed, inserting (166) and interchanging integrals produces the inner integral
∫ u v sinh ρ d ρ cosh ρ − cosh u cosh v − cosh ρ = π ( 0 ≤ u < v ) ; \int_{u}^{v}\frac{\sinh\rho\,d\rho}{\sqrt{\cosh\rho-\cosh u}\sqrt{\cosh v-\cosh\rho}}=\pi\quad(0\le u<v); ∫ u v cosh ρ − cosh u cosh v − cosh ρ sinh ρ d ρ = π ( 0 ≤ u < v ) ;
the result is − ∫ u ∞ H ′ ( v ) d v = H ( u ) -\int_{u}^{\infty}H'(v)\,dv=H(u) − ∫ u ∞ H ′ ( v ) d v = H ( u ) . Absolute integrability follows by the same computation with ∣ H ′ ∣ |H'| ∣ H ′ ∣ .
The spectral multiplier of this kernel is H ^ ( r ) \widehat{H}(r) H ( r ) . At the point i i i , write y = e u y=e^u y = e u and x = e u / 2 v x=e^u/2v x = e u /2 v . Then cosh d ( i , x + i y ) = cosh u + v 2 / 2 \cosh d(i,x+iy)=\cosh u+v^2/2 cosh d ( i , x + i y ) = cosh u + v 2 /2 , and convolution of K H K_H K H with y 1 / 2 + i r y^{1/2+ir} y 1/2 + i r is
∫ R e i r u [ ∫ R K H ( arcosh ( cosh u + v 2 / 2 ) ) d v ] d u = ∫ R e i r u H ( u ) d u = H ^ ( r ) . \int_{\mathbb{R}}e^{iru}\left[\int_{\mathbb{R}}K_H\left(\operatorname{arcosh}\left(\cosh u+v^2/2\right)\right)\,dv\right]\,du=\int_{\mathbb{R}}e^{iru}H(u)\,du=\widehat{H}(r). ∫ R e i r u [ ∫ R K H ( arcosh ( cosh u + v 2 /2 ) ) d v ] d u = ∫ R e i r u H ( u ) d u = H ( r ) .
The bracket is exactly (167) . Also
K H ( 0 ) = 1 4 π ∫ R r tanh ( π r ) H ^ ( r ) d r . (168) K_H(0)=\frac{1}{4\pi}\int_{\mathbb{R}}r\tanh(\pi r)\widehat{H}(r)\,dr. \tag*{(168)} K H ( 0 ) = 4 π 1 ∫ R r tanh ( π r ) H ( r ) d r . (168)
This follows from K H ( 0 ) = − ( 2 π ) − 1 ∫ 0 ∞ H ′ ( u ) / sinh ( u / 2 ) d u K_H(0)=-(2\pi)^{-1}\int_0^\infty H'(u)/\sinh(u/2)\,du K H ( 0 ) = − ( 2 π ) − 1 ∫ 0 ∞ H ′ ( u ) / sinh ( u /2 ) d u , Fourier inversion, and ∫ 0 ∞ sin ( r u ) / sinh ( u / 2 ) d u = π tanh ( π r ) \int_0^\infty\sin(ru)/\sinh(u/2)\,du=\pi\tanh(\pi r) ∫ 0 ∞ sin ( r u ) / sinh ( u /2 ) d u = π tanh ( π r ) . The last identity follows by expanding 1 / sinh ( u / 2 ) 1/\sinh(u/2) 1/ sinh ( u /2 ) into exponentials and using the partial-fraction expansion of tanh \tanh tanh ; the resulting series is 2 ∑ j ≥ 0 r / ( ( j + 1 / 2 ) 2 + r 2 ) 2\sum_{j\geq0}r/((j+1/2)^2+r^2) 2 ∑ j ≥ 0 r / (( j + 1/2 ) 2 + r 2 ) . All Fourier interchanges are justified by rapid decay and ∣ sin ( r u ) ∣ ≤ min { 1 , ∣ r u ∣ } |\sin(ru)|\leq\min\{1,|ru|\} ∣ sin ( r u ) ∣ ≤ min { 1 , ∣ r u ∣ } . This also checks the spherical inversion normalization recorded in [7 §5, immediately before (5.8)] .
The periodization
K H ( z , w ) = ∑ γ ∈ Γ K H ( d ( z , γ w ) ) (169) K_H(z,w)=\sum_{\gamma\in\Gamma}K_H(d(z,\gamma w)) \tag*{(169)} K H ( z , w ) = γ ∈ Γ ∑ K H ( d ( z , γ w )) (169)
is locally finite on compact coordinate lifts, by proper discontinuity and compact radial support. Choose an orthonormal basis { u j } \{u_j\} { u j } of the discrete spectral subspace of L 2 ( X ) L^2(X) L 2 ( X ) , consisting of the constant, residual and cuspidal eigenfunctions. Write Δ u j = λ j u j \Delta u_j=\lambda_j u_j Δ u j = λ j u j and λ j = 1 / 4 + r j 2 \lambda_j=1/4+r_j^2 λ j = 1/4 + r j 2 , with r j ≥ 0 r_j\geq0 r j ≥ 0 or r j ∈ i [ 0 , 1 / 2 ] r_j\in i[0,1/2] r j ∈ i [ 0 , 1/2 ] . The Eisenstein series E b E_b E b use the width-one cusp coordinates. The local spectral expansion is
K H ( z , w ) = ∑ j H ^ ( r j ) u j ( z ) u j ( w ) ‾ + 1 4 π ∑ b = 1 n ∫ R H ^ ( r ) E b ( z , 1 2 + i r ) E b ( w , 1 2 + i r ) ‾ d r . (170) K_H(z,w)=\sum_j\widehat{H}(r_j)u_j(z)\overline{u_j(w)}+\frac{1}{4\pi}\sum_{b=1}^{n}\int_{\mathbb{R}}\widehat{H}(r)E_b\left(z,\frac{1}{2}+ir\right)\overline{E_b\left(w,\frac{1}{2}+ir\right)}\,dr. \tag*{(170)} K H ( z , w ) = j ∑ H ( r j ) u j ( z ) u j ( w ) + 4 π 1 b = 1 ∑ n ∫ R H ( r ) E b ( z , 2 1 + i r ) E b ( w , 2 1 + i r ) d r . (170)
The precise input is [3 §3.3, (3.11)] , with transform hypotheses in §3.1 and the convergence bound in §3.5, Lemma 3.2. The transform is even and entire, with arbitrarily rapid decay in every fixed horizontal strip; thus those hypotheses hold. Our nonnegative Laplacian is Bruin’s − Δ -\Delta − Δ . On compact coordinate lifts, Lemma 3.2 gives a uniform linear bound in the eigenvalue cutoff for the local spectral mass. Rapid decay of the multiplier and Cauchy–Schwarz therefore give locally uniform absolute convergence, permitting diagonal restriction and integration over compact sets.
We specify the group convention in applying this input. Since n ≥ 1 n\geq1 n ≥ 1 , Γ \Gamma Γ is free of rank 2 g + n − 1 2g+n-1 2 g + n − 1 . Lift free generators to S L 2 ( R ) \mathrm{SL}_2(\mathbb{R}) SL 2 ( R ) . The resulting representation projects to the original faithful representation, so its image is an injective lift omitting − I -I − I . It is discrete since its projected image is discrete. Bruin’s integration divisor # ( Γ ∩ { ± I } ) \#(\Gamma\cap\{\pm I\}) # ( Γ ∩ { ± I }) is therefore one. Weight-zero kernels and cusp coordinates agree with those in P S L 2 ( R ) \mathrm{PSL}_2(\mathbb{R}) PSL 2 ( R ) ; the sign of a lifted parabolic does not affect its action. Bruin’s transform is H ^ ( ( s − 1 / 2 ) / i ) \widehat{H}((s-1/2)/i) H (( s − 1/2 ) / i ) . With s = 1 / 2 + i r s=1/2+ir s = 1/2 + i r , its symmetry under s ↦ 1 − s s\mapsto1-s s ↦ 1 − s makes the choice of branch for an exceptional eigenvalue immaterial. No factor two is introduced in (170) .
We now return to the fixed multiplier f f f . Define the bounded self-adjoint operator A X A_X A X by its Laplace multiplier
f ( r ( λ ) ) , r ( λ ) = { λ − 1 / 4 , λ ≥ 1 / 4 , i 1 / 4 − λ , 0 ≤ λ < 1 / 4 , 0 ≤ A X ≤ a . (171) f(r(\lambda)),\qquad r(\lambda)=
\begin{cases}
\sqrt{\lambda-1/4}, & \lambda\geq1/4,\\
i\sqrt{1/4-\lambda}, & 0\leq\lambda<1/4,
\end{cases}
\qquad0\leq A_X\leq a. \tag*{(171)} f ( r ( λ )) , r ( λ ) = { λ − 1/4 , i 1/4 − λ , λ ≥ 1/4 , 0 ≤ λ < 1/4 , 0 ≤ A X ≤ a . (171)
This is the meaning of f ( Δ X − 1 / 4 ) f(\sqrt{\Delta_X-1/4}) f ( Δ X − 1/4 ) . The constant eigenfunction has A X A_X A X -eigenvalue a a a ; the part of the spectrum of Δ X \Delta_X Δ X in [ 1 / 4 , ∞ ) [1/4,\infty) [ 1/4 , ∞ ) maps into [ 0 , 1 ] [0,1] [ 0 , 1 ] . For p = x p ~ p=x\widetilde{p} p = x p , we use the two norms
N ∗ ( p ) = ∥ p ~ ∥ C ( [ 0 , a ] ) , N R ( p ) = ∥ p ~ ∥ C ( [ 0 , 1 ] ) . (172) N_*(p)=\|\widetilde{p}\|_{C([0,a])},\qquad N_{\mathbb{R}}(p)=\|\widetilde{p}\|_{C([0,1])}. \tag*{(172)} N ∗ ( p ) = ∥ p ∥ C ([ 0 , a ]) , N R ( p ) = ∥ p ∥ C ([ 0 , 1 ]) . (172)
The first norm controls the full spectral trace. The second will control the identity and parabolic terms using only real spectral parameters.
Uniform geometry of all cusp lifts
The following cusp neighborhoods have the same height normalization on every surface. Control of all their lifts is needed when unfolding the parabolic contribution: removing other cusp neighborhoods must not remove any part of the translation kernel’s support.
Lemma 5.2 (Uniform cusp truncation). For Y > 1 Y>1 Y > 1 the width-one cusp regions C b ( Y ) C_b(Y) C b ( Y ) , represented by 0 ≤ x < 1 0\le x<1 0 ≤ x < 1 , y > Y y>Y y > Y , are embedded and pairwise disjoint, each with area 1 / Y 1/Y 1/ Y . In the chart of a distinguished cusp, every other lift of every C b ( Y ) C_b(Y) C b ( Y ) lies below height 1 / Y 1/Y 1/ Y .
Proof. We first prove the normalized Shimizu estimate. In a discrete subgroup containing U = ( 0 1 0 1 ) U=\begin{pmatrix}0&1\\0&1\end{pmatrix} U = ( 0 0 1 1 ) , an element S = ( a b c d ) S=\begin{pmatrix}a&b\\c&d\end{pmatrix} S = ( a c b d ) has either c = 0 c=0 c = 0 or ∣ c ∣ ≥ 1 |c|\ge1 ∣ c ∣ ≥ 1 . Otherwise set S 0 = S S_0=S S 0 = S and put S m + 1 = S m U S m − 1 S_{m+1}=S_mUS_m^{-1} S m + 1 = S m U S m − 1 . Direct multiplication gives
c m + 1 = − c m 2 , a m + 1 = 1 − a m c m , b m + 1 = a m 2 , d m + 1 = 1 + a m c m . (173) c_{m+1}=-c_m^2,\qquad a_{m+1}=1-a_mc_m,\qquad b_{m+1}=a_m^2,\qquad d_{m+1}=1+a_mc_m. \tag*{(173)} c m + 1 = − c m 2 , a m + 1 = 1 − a m c m , b m + 1 = a m 2 , d m + 1 = 1 + a m c m . (173)
If 0 < ∣ c 0 ∣ < 1 0<|c_0|<1 0 < ∣ c 0 ∣ < 1 , then ∣ c m ∣ = ∣ c 0 ∣ 2 m → 0 |c_m|=|c_0|^{2^m}\to0 ∣ c m ∣ = ∣ c 0 ∣ 2 m → 0 . Once ∣ c m ∣ ≤ 1 / 2 |c_m|\le1/2 ∣ c m ∣ ≤ 1/2 , the recurrence for a m a_m a m bounds it uniformly; it then implies a m + 1 → 1 a_{m+1}\to1 a m + 1 → 1 . Hence S m → U S_m\to U S m → U through distinct elements, contradicting discreteness. The same proof applies to matrix representatives of a subgroup of PSL 2 ( R ) \operatorname{PSL}_2(\mathbb{R}) PSL 2 ( R ) .
Let σ b , σ d \sigma_b,\sigma_d σ b , σ d be two width-one scalings and M = σ b − 1 γ σ d M=\sigma_b^{-1}\gamma\sigma_d M = σ b − 1 γ σ d with lower-left entry c c c . If c = 0 c=0 c = 0 , the corresponding cusp lift is the distinguished lift: equality of its fixed point with infinity means that the cusp representatives agree and the element belongs to the stabilizer. Otherwise M U M − 1 MUM^{-1} M U M − 1 belongs to σ b − 1 Γ σ b \sigma_b^{-1}\Gamma\sigma_b σ b − 1 Γ σ b and has lower-left entry − c 2 -c^2 − c 2 . The just-proved estimate gives ∣ c ∣ ≥ 1 |c|\ge1 ∣ c ∣ ≥ 1 . For w = u + i v w=u+iv w = u + i v , v > Y v>Y v > Y ,
Im ( M w ) = v ∣ c w + d ∣ 2 ≤ 1 c 2 v < 1 Y . (174) \operatorname{Im}(Mw)=\frac{v}{|cw+d|^2}\le\frac{1}{c^2v}<\frac{1}{Y}. \tag*{(174)} Im ( M w ) = ∣ c w + d ∣ 2 v ≤ c 2 v 1 < Y 1 . (174)
This proves the assertion for different lifts of the same cusp and for lifts of different cusps. Since Y > 1 Y>1 Y > 1 , they cannot meet the distinguished region y > Y y>Y y > Y . The area of its quotient strip is ∫ 0 1 ∫ Y ∞ y − 2 d y d x = 1 / Y \int_0^1\int_Y^\infty y^{-2}\,dy\,dx=1/Y ∫ 0 1 ∫ Y ∞ y − 2 d y d x = 1/ Y .
The exact trace on a common truncation
The complement X ( Y ) = X ∖ ⋃ b C b ( Y ) X(Y)=X\setminus\bigcup_b C_b(Y) X ( Y ) = X ∖ ⋃ b C b ( Y ) is compact. Write χ Y = 1 X ( Y ) \chi_Y=1_{X(Y)} χ Y = 1 X ( Y ) , as a multiplication operator, and fix
Y Q = 3 e L Q ( Q ≥ 1 ) . (175) Y_Q=3e^{L_Q}\qquad(Q\ge1). \tag*{(175)} Y Q = 3 e L Q ( Q ≥ 1 ) . (175)
Every test H p H_p H p with deg p ≤ Q \deg p\le Q deg p ≤ Q is supported in [ − L Q , L Q ] [-L_Q,L_Q] [ − L Q , L Q ] . The common cutoff therefore gives one geometric trace formula and one positive weighted spectral measure for all these tests. Individual geometric terms may still have either sign.
Theorem 5.3 (Exact trace on a common truncation). For every such surface X X X , integer Q ≥ 1 Q\ge1 Q ≥ 1 and real polynomial p p p of degree at most Q Q Q with p ( 0 ) = 0 p(0)=0 p ( 0 ) = 0 , the compression χ Y Q p ( A X ) χ Y Q \chi_{Y_Q}p(A_X)\chi_{Y_Q} χ Y Q p ( A X ) χ Y Q is trace class. For Y > 1 Y>1 Y > 1 , define
I ( p ) = K H p ( 0 ) = 1 4 π ∫ R r tanh ( π r ) p ( f ( r ) ) d r , (176) \mathcal{I}(p)=K_{H_p}(0)=\frac{1}{4\pi}\int_{\mathbb{R}}r\tanh(\pi r)p(f(r))\,dr, \tag*{(176)} I ( p ) = K H p ( 0 ) = 4 π 1 ∫ R r tanh ( π r ) p ( f ( r )) d r , (176) P Y ( p ) = 2 ∫ 0 L Q K H p ( ρ ) cosh ( ρ / 2 ) H [ 2 Y sinh ( ρ / 2 ) ] d ρ , (177) \mathcal{P}_Y(p)=2\int_0^{L_Q}K_{H_p}(\rho)\cosh(\rho/2)H[2Y\sinh(\rho/2)]\,d\rho, \tag*{(177)} P Y ( p ) = 2 ∫ 0 L Q K H p ( ρ ) cosh ( ρ /2 ) H [ 2 Y sinh ( ρ /2 )] d ρ , (177)
where H m = ∑ k = 1 m k − 1 H_m=\sum_{k=1}^{m} k^{-1} H m = ∑ k = 1 m k − 1 for integers m ≥ 1 m\ge1 m ≥ 1 , and H 0 = 0 H_0=0 H 0 = 0 . Then the exact identity is
τ X , Q ( p ) : = Tr ( χ Y Q p ( A X ) χ Y Q ) = ( A g , n − n Y Q ) I ( p ) + n P Y Q ( p ) + O X ( p ) . (178) \tau_{X,Q}(p):=\operatorname{Tr}\bigl(\chi_{Y_Q}p(A_X)\chi_{Y_Q}\bigr)=\left(\mathcal{A}_{g,n}-\frac{n}{Y_Q}\right)\mathcal{I}(p)+n\mathcal{P}_{Y_Q}(p)+\mathcal{O}_X(p). \tag*{(178)} τ X , Q ( p ) := Tr ( χ Y Q p ( A X ) χ Y Q ) = ( A g , n − Y Q n ) I ( p ) + n P Y Q ( p ) + O X ( p ) . (178)
If p p p is nonnegative on the spectrum of A X A_X A X , then τ X , Q ( p ) ≥ 0 \tau_{X,Q}(p)\ge0 τ X , Q ( p ) ≥ 0 . There is a finite positive measure ν X , Q \nu_{X,Q} ν X , Q supported in [ 0 , a ] [0,a] [ 0 , a ] such that
τ X , Q ( x p ~ ) = ∫ [ 0 , a ] p ~ ( s ) d ν X , Q ( s ) , \tau_{X,Q}(x\widetilde{p})=\int_{[0,a]}\widetilde{p}(s)\,d\nu_{X,Q}(s), τ X , Q ( x p ) = ∫ [ 0 , a ] p ( s ) d ν X , Q ( s ) ,
ν X , Q ( [ 0 , a ] ) = τ X , Q ( x ) , ∣ τ X , Q ( p ) ∣ ≤ N ∗ ( p ) τ X , Q ( x ) . (179) \nu_{X,Q}([0,a])=\tau_{X,Q}(x),\qquad|\tau_{X,Q}(p)|\le N_*(p)\tau_{X,Q}(x). \tag*{(179)} ν X , Q ([ 0 , a ]) = τ X , Q ( x ) , ∣ τ X , Q ( p ) ∣ ≤ N ∗ ( p ) τ X , Q ( x ) . (179)
Proof. Trace class and positivity. Apply (170) to H = h H=h H = h . All diagonal terms are nonnegative, since the multiplier f f f is nonnegative on the full spectrum. For any Y > 1 Y>1 Y > 1 , the diagonal kernel is smooth on the compact set X ( Y ) X(Y) X ( Y ) . With S = A X 1 / 2 χ Y S=A_X^{1/2}\chi_Y S = A X 1/2 χ Y , spectral Parseval and Tonelli give
∥ S ∥ H S 2 = ∑ j f ( r j ) ∥ χ Y u j ∥ 2 + 1 4 π ∑ b = 1 n ∫ R f ( r ) ∥ χ Y E b ( ⋅ , 1 2 + i r ) ∥ 2 d r = ∫ X ( Y ) K h ( z , z ) d μ ( z ) < ∞ . (180) \begin{aligned}
\lVert S\rVert_{\mathrm{HS}}^2=\sum_j f(r_j)\lVert\chi_Yu_j\rVert^2+\frac{1}{4\pi}\sum_{b=1}^{n}\int_{\mathbb{R}}f(r)\left\lVert\chi_YE_b\left(\cdot,\frac{1}{2}+ir\right)\right\rVert^2\,dr \\
&=\int_{X(Y)}K_h(z,z)\,d\mu(z)<\infty. \tag*{(180)}
\end{aligned} ∥ S ∥ HS 2 = j ∑ f ( r j ) ∥ χ Y u j ∥ 2 + 4 π 1 b = 1 ∑ n ∫ R f ( r ) χ Y E b ( ⋅ , 2 1 + i r ) 2 d r = ∫ X ( Y ) K h ( z , z ) d μ ( z ) < ∞. (180)
Explicitly, for an orthonormal basis { e m } \{e_m\} { e m } of L 2 ( X ) L^2(X) L 2 ( X ) , Parseval turns ∑ m ∣ ⟨ χ Y e m , u j ⟩ ∣ 2 \sum_m|\langle\chi_Ye_m,u_j\rangle|^2 ∑ m ∣ ⟨ χ Y e m , u j ⟩ ∣ 2 into ∥ χ Y u j ∥ 2 \lVert\chi_Yu_j\rVert^2 ∥ χ Y u j ∥ 2 , and does the same for each Eisenstein function restricted to X ( Y ) X(Y) X ( Y ) . Nonnegativity justifies interchanging this basis sum and the spectral integrals. Since p = x p ~ p=x\widetilde{p} p = x p , functional calculus now gives
χ Y p ( A X ) χ Y = S ∗ p ~ ( A X ) S , ∥ χ Y p ( A X ) χ Y ∥ 1 ≤ N ∗ ( p ) ∥ S ∥ H S 2 . (181) \chi_Yp(A_X)\chi_Y=S^*\widetilde{p}(A_X)S,\qquad\lVert\chi_Yp(A_X)\chi_Y\rVert_1\le N_*(p)\lVert S\rVert_{\mathrm{HS}}^2. \tag*{(181)} χ Y p ( A X ) χ Y = S ∗ p ( A X ) S , ∥ χ Y p ( A X ) χ Y ∥ 1 ≤ N ∗ ( p ) ∥ S ∥ HS 2 . (181)
Thus χ Y p ( A X ) χ Y \chi_Yp(A_X)\chi_Y χ Y p ( A X ) χ Y is trace class, also for complex polynomials. It is positive when p p p is nonnegative on the spectrum of A X A_X A X . The inequality ∣ p ( s ) ∣ ≤ N ∗ ( p ) s |p(s)|\le N_*(p)s ∣ p ( s ) ∣ ≤ N ∗ ( p ) s on [ 0 , a ] [0,a] [ 0 , a ] dominates the signed spectral diagonal by (180) . Thus at Y = Y Q Y=Y_Q Y = Y Q ,
τ X , Q ( p ) = ∫ X ( Y Q ) K H p ( z , z ) d μ ( z ) = ∑ j p ( f ( r j ) ) ∥ χ Y Q u j ∥ 2 + 1 4 π ∑ b = 1 n ∫ R p ( f ( r ) ) ∥ χ Y Q E b ( ⋅ , 1 2 + i r ) ∥ 2 d r . (182) \begin{aligned}
\tau_{X,Q}(p)=\int_{X(Y_Q)}K_{Hp}(z,z)\,d\mu(z) \\
&=\sum_jp(f(r_j))\lVert\chi_{Y_Q}u_j\rVert^2+\frac{1}{4\pi}\sum_{b=1}^{n}\int_{\mathbb{R}}p(f(r))\left\lVert\chi_{Y_Q}E_b\left(\cdot,\frac{1}{2}+ir\right)\right\rVert^2\,dr. \tag*{(182)}
\end{aligned} τ X , Q ( p ) = ∫ X ( Y Q ) K H p ( z , z ) d μ ( z ) = j ∑ p ( f ( r j )) ∥ χ Y Q u j ∥ 2 + 4 π 1 b = 1 ∑ n ∫ R p ( f ( r )) χ Y Q E b ( ⋅ , 2 1 + i r ) 2 d r . (182)
with absolute convergence. With S = A X 1 / 2 χ Y Q S=A_X^{1/2}\chi_{Y_Q} S = A X 1/2 χ Y Q , define ν X , Q ( B ) = ∥ 1 B ( A X ) S ∥ H S 2 \nu_{X,Q}(B)=\lVert\mathbf{1}_B(A_X)S\rVert_{\mathrm{HS}}^2 ν X , Q ( B ) = ∥ 1 B ( A X ) S ∥ HS 2 for Borel sets B ⊂ [ 0 , a ] B\subset[0,a] B ⊂ [ 0 , a ] . This proves (179) ; countable additivity follows from orthogonality of the spectral projections.
Hyperbolic contribution. We compute the geometric integral for any even H ∈ C ∞ ( R ) H\in C^\infty(\mathbb{R}) H ∈ C ∞ ( R ) supported in [ − T , T ] [-T,T] [ − T , T ] , with T > 0 T>0 T > 0 , provided
Y > max { 1 , e T / 2 , 2 sinh ( T / 2 ) } . (183) Y>\max\{1,e^{T/2},2\sinh(T/2)\}. \tag*{(183)} Y > max { 1 , e T /2 , 2 sinh ( T /2 )} . (183)
The condition Y > e T / 2 Y>e^{T/2} Y > e T /2 ensures that hyperbolic terms vanish in the removed cusp regions. The other condition, Y > 2 sinh ( T / 2 ) Y>2\sinh(T/2) Y > 2 sinh ( T /2 ) , will make the parabolic unfolding exact. A hyperbolic element does not stabilize the distinguished cusp: the cusp stabilizer is the primitive parabolic cyclic group. Indeed, an additional dilation fixing that cusp would conjugate its translations to translations tending to the identity, contradicting discreteness. Thus its lower-left entry satisfies ∣ c ∣ ≥ 1 |c|\ge1 ∣ c ∣ ≥ 1 and, for z = x + i y z=x+iy z = x + i y ,
Im ( γ z ) ≤ 1 y , d ( z , γ z ) ≥ log y Im ( γ z ) ≥ 2 log y . (184) \operatorname{Im}(\gamma z)\le\frac{1}{y},\qquad d(z,\gamma z)\ge\log\frac{y}{\operatorname{Im}(\gamma z)}\ge2\log y. \tag*{(184)} Im ( γ z ) ≤ y 1 , d ( z , γ z ) ≥ log Im ( γ z ) y ≥ 2 log y . (184)
Every hyperbolic kernel term vanishes on every deleted cusp tail. Its truncated integral therefore equals its full integral. The periodization of ∣ K H ∣ |K_H| ∣ K H ∣ is locally bounded on the compact truncation, so these hyperbolic integrals are absolutely summable.
For a primitive oriented hyperbolic class γ \gamma γ of length ℓ \ell ℓ , the centralizer of every nonzero power is ⟨ γ ⟩ \langle\gamma\rangle ⟨ γ ⟩ . Fix a positive repetition index k ≥ 1 k \ge1 k ≥ 1 and put v = k ℓ v=k\ell v = k ℓ . Fermi coordinates ( s , r ) (s,r) ( s , r ) on its cyclic quotient satisfy 0 ≤ s < ℓ 0 \le s < \ell 0 ≤ s < ℓ , d μ = cosh r d s d r d\mu=\cosh r\,ds\,dr d μ = cosh r d s d r , and
cosh d ( z , γ k z ) = cosh v + 2 sinh 2 ( v / 2 ) sinh 2 r . \cosh d(z,\gamma^kz)=\cosh v+2\sinh^2(v/2)\sinh^2 r. cosh d ( z , γ k z ) = cosh v + 2 sinh 2 ( v /2 ) sinh 2 r .
Changing variables from r r r to ρ = d ( z , γ k z ) \rho=d(z,\gamma^kz) ρ = d ( z , γ k z ) , including both signs of r r r , and applying (167) , yields
∫ ⟨ γ ⟩ \ H K H ( d ( z , γ k z ) ) d μ ( z ) = ℓ 2 sinh ( v / 2 ) ∫ v ∞ K H ( ρ ) sinh ρ cosh ρ − cosh v d ρ = ℓ 2 sinh ( k ℓ / 2 ) H ( k ℓ ) . (185) \begin{aligned}
\int_{\langle\gamma\rangle\backslash\mathbb{H}} K_H(d(z,\gamma^kz))\,d\mu(z)
=\frac{\ell}{\sqrt{2}\sinh(v/2)}\int_v^\infty
\frac{K_H(\rho)\sinh\rho}{\sqrt{\cosh\rho-\cosh v}}\,d\rho\\
&=\frac{\ell}{2\sinh(k\ell/2)}H(k\ell). \tag*{(185)}
\end{aligned} ∫ ⟨ γ ⟩ \ H K H ( d ( z , γ k z )) d μ ( z ) = 2 sinh ( v /2 ) ℓ ∫ v ∞ cosh ρ − cosh v K H ( ρ ) sinh ρ d ρ = 2 sinh ( k ℓ /2 ) ℓ H ( k ℓ ) . (185)
The inverse class is distinct: an orientation-preserving isometry interchanging the endpoints of the axis has order two in P S L 2 ( R ) \mathrm{PSL}_2(\mathbb{R}) PSL 2 ( R ) and cannot belong to the torsion-free group. Consequently conjugacy-class unfolding gives precisely O H ( X ) \mathcal{O}_H(X) O H ( X ) , with both primitive orientations and every positive repetition.
Parabolic contribution. Fix a cusp chart. The displacement of its translation by k ∈ Z \ { 0 } k\in\mathbb{Z}\backslash\{0\} k ∈ Z \ { 0 } is
d ( z , z + k ) = 2 arsinh ∣ k ∣ 2 y . (186) d(z,z+k)=2\operatorname{arsinh}\frac{|k|}{2y}. \tag*{(186)} d ( z , z + k ) = 2 arsinh 2 y ∣ k ∣ . (186)
Unfolding the conjugates of its stabilizer gives the width-one strip with the lifted truncation indicator. Apart from its own deleted region y > Y y>Y y > Y , this strip has holes from other removed cusp lifts. By Lemma 5.2, every such hole lies in y < 1 / Y y<1/Y y < 1/ Y . On the support of the translation kernel, however,
y ≥ ∣ k ∣ 2 sinh ( T / 2 ) ≥ 1 2 sinh ( T / 2 ) > 1 Y . (187) y\ge\frac{|k|}{2\sinh(T/2)}\ge\frac{1}{2\sinh(T/2)}>\frac{1}{Y}. \tag*{(187)} y ≥ 2 sinh ( T /2 ) ∣ k ∣ ≥ 2 sinh ( T /2 ) 1 > Y 1 . (187)
The translation kernel vanishes on every other removed cusp lift. Hence replacing the unfolded domain by the whole strip 0 ≤ x < 1 0\le x<1 0 ≤ x < 1 , 0 < y ≤ Y 0<y\le Y 0 < y ≤ Y leaves the integral unchanged. This is a support argument and does not require the kernel to be nonnegative. Retaining both signs of k k k and substituting y = k / ( 2 sinh ( ρ / 2 ) ) y=k/(2\sinh(\rho/2)) y = k / ( 2 sinh ( ρ /2 )) gives the contribution of this one cusp:
P Y ( H ) = ∑ k ≠ 0 ∫ 0 Y K H ( 2 arsinh ∣ k ∣ 2 y ) d y y 2 = ∑ k ≥ 1 2 k ∫ 2 arsinh ( k / ( 2 Y ) ) T K H ( ρ ) cosh ( ρ / 2 ) d ρ = 2 ∫ 0 T K H ( ρ ) cosh ( ρ / 2 ) H ⌊ 2 Y sinh ( ρ / 2 ) ⌋ d ρ . (188) \begin{aligned}
\mathcal{P}_Y(H)=\sum_{k\ne0}\int_0^Y K_H\left(2\operatorname{arsinh}\frac{|k|}{2y}\right)\frac{dy}{y^2} \\
&=\sum_{k\ge1}\frac{2}{k}\int_{2\operatorname{arsinh}(k/(2Y))}^{T}K_H(\rho)\cosh(\rho/2)\,d\rho\\
&=2\int_0^T K_H(\rho)\cosh(\rho/2)H_{\lfloor2Y\sinh(\rho/2)\rfloor}\,d\rho. \tag*{(188)}
\end{aligned} P Y ( H ) = k = 0 ∑ ∫ 0 Y K H ( 2 arsinh 2 y ∣ k ∣ ) y 2 d y = k ≥ 1 ∑ k 2 ∫ 2 arsinh ( k / ( 2 Y )) T K H ( ρ ) cosh ( ρ /2 ) d ρ = 2 ∫ 0 T K H ( ρ ) cosh ( ρ /2 ) H ⌊ 2 Y s i n h ( ρ /2 )⌋ d ρ . (188)
An integral in the middle line is zero if its lower limit exceeds T T T . Only k ≤ 2 Y sinh ( T / 2 ) k\le2Y\sinh(T/2) k ≤ 2 Y sinh ( T /2 ) contribute, so the sum is finite and the integrals are absolutely convergent. This argument covers other lifts of the same cusp and lifts of different cusps, and gives the identical term for each of the n n n cusps.
Assembly of the trace formula. The identity element contributes ( A g , n − n / Y ) K H ( 0 ) (A_{g,n}-n/Y)K_H(0) ( A g , n − n / Y ) K H ( 0 ) . Nonidentity elements are hyperbolic or parabolic, since there are no orbifold points. Thus
∫ X ( Y ) K H ( z , z ) d μ ( z ) = ( A g , n − n Y ) K H ( 0 ) + n P Y ( H ) + O H ( X ) . (189) \int_{X(Y)}\mathcal{K}_H(z,z)\,d\mu(z)
=\left(A_{g,n}-\frac{n}{Y}\right)K_H(0)+n\mathcal{P}_Y(H)+\mathcal{O}_H(X). \tag*{(189)} ∫ X ( Y ) K H ( z , z ) d μ ( z ) = ( A g , n − Y n ) K H ( 0 ) + n P Y ( H ) + O H ( X ) . (189)
For T = L Q T=LQ T = L Q and Y = 3 e L Q Y=3e^{LQ} Y = 3 e L Q , all inequalities in (183) hold. Taking H = H p H=H_p H = H p proves (178) . □
The measure in (179) represents the weighted functional p ~ ↦ τ X , Q ( x p ~ ) \widetilde p \mapsto\tau_{X,Q}(x\widetilde p) p ↦ τ X , Q ( x p ) . At any fixed Y > 1 Y>1 Y > 1 , the factorization (181) also gives a trace-class operator S ∗ ψ ( A X ) S S^*\psi(A_X)S S ∗ ψ ( A X ) S for every bounded Borel function ψ \psi ψ on [ 0 , a ] [0,a] [ 0 , a ] , with S = A X 1 / 2 χ Y S=A_X^{1/2}\chi_Y S = A X 1/2 χ Y . Thus the measure construction requires no degree bound; the bound deg p ≤ Q \deg p\le Q deg p ≤ Q ensures that the geometric identity holds at Y Q Y_Q Y Q . The compression of 1 1 1 is χ Y Q \chi_{Y_Q} χ Y Q , an infinite-rank projection, so it is not trace class. Formula (177) is independent of the chosen support radius, provided the corresponding interval contains supp H p \operatorname{supp} H_p supp H p ; at a fixed Y Q Y_Q Y Q it is one linear functional for all degrees at most Q Q Q .
Deterministic norms and the constant eigenfunction
We first bound the identity and parabolic terms using only the multiplier on the real spectral axis, then subtract the constant mode. The parabolic estimate uses the following absolute Abel norm. For an even H ∈ C ∞ ( R ) H\in C^\infty(\mathbb{R}) H ∈ C ∞ ( R ) supported in [ − T , T ] [-T,T] [ − T , T ] , with T > 0 T>0 T > 0 , put
B ( H ) = 2 ∫ 0 T ∣ K H ( ρ ) ∣ cosh ( ρ / 2 ) d ρ , log + v = log max { 1 , v } ( v ≥ 0 ) . (190) B(H)=2\int_0^T\lvert K_H(\rho)\rvert\cosh(\rho/2)\,d\rho,\qquad\log_+v=\log\max\{1,v\}\quad(v\ge0). \tag*{(190)} B ( H ) = 2 ∫ 0 T ∣ K H ( ρ )∣ cosh ( ρ /2 ) d ρ , log + v = log max { 1 , v } ( v ≥ 0 ) . (190)
The value of B ( H ) B(H) B ( H ) is independent of the chosen support radius.
Proposition 5.4 (Deterministic trace bounds). There are finite nonnegative constants I f , J f , b f I_f,J_f,b_f I f , J f , b f , depending only on the fixed multiplier, such that for every integer Q ≥ 1 Q\ge1 Q ≥ 1 and real polynomial p = x p ~ p=x\widetilde p p = x p of degree at most Q Q Q ,
∣ I ( p ) ∣ ≤ I f N R ( p ) , ∣ P Y Q ( p ) ∣ ≤ J f Q 2 N R ( p ) , ∣ P Y Q ( x ) ∣ ≤ b f Q . (191) \lvert I(p)\rvert\le I_f\mathrm N_{\mathbb R}(p),\qquad\lvert P_{Y_Q}(p)\rvert\le J_fQ^2\mathrm N_{\mathbb R}(p),\qquad\lvert P_{Y_Q}(x)\rvert\le b_fQ. \tag*{(191)} ∣ I ( p )∣ ≤ I f N R ( p ) , ∣ P Y Q ( p )∣ ≤ J f Q 2 N R ( p ) , ∣ P Y Q ( x )∣ ≤ b f Q . (191) One may take
I f = 1 4 π ∫ R r tanh ( π r ) f ( r ) d r , C 1 , f = 1 2 π ∫ R ∣ r ∣ f ( r ) d r , (192) I_f=\frac{1}{4\pi}\int_{\mathbb R}r\tanh(\pi r)f(r)\,dr,\qquad C_{1,f}=\frac{1}{2\pi}\int_{\mathbb R}\lvert r\rvert f(r)\,dr, \tag*{(192)} I f = 4 π 1 ∫ R r tanh ( π r ) f ( r ) d r , C 1 , f = 2 π 1 ∫ R ∣ r ∣ f ( r ) d r , (192) J f = L C 1 , f ( 1 + log 3 + 3 L / 2 ) , b f = B ( h ) [ 1 + L + log + ( 6 sinh ( L / 2 ) ) ] . (193) J_f=LC_{1,f}(1+\log3+3L/2),\qquad b_f=B(h)[1+L+\log_+(6\sinh(L/2))]. \tag*{(193)} J f = L C 1 , f ( 1 + log 3 + 3 L /2 ) , b f = B ( h ) [ 1 + L + log + ( 6 sinh ( L /2 ))] . (193)
Proof. The absolute Abel norm obeys
B ( H ) ≤ ∫ 0 T ∣ H ′ ( u ) ∣ d u . (194) B(H)\le\int_0^T\lvert H'(u)\rvert\,du. \tag*{(194)} B ( H ) ≤ ∫ 0 T ∣ H ′ ( u )∣ d u . (194)
Indeed, substitute (166) , apply Tonelli, and use
∫ 0 u cosh ( ρ / 2 ) cosh u − cosh ρ d ρ = π 2 ( u > 0 ) . \int_0^u\frac{\cosh(\rho/2)}{\sqrt{\cosh u-\cosh\rho}}\,d\rho=\frac{\pi}{\sqrt{2}}\qquad(u>0). ∫ 0 u cosh u − cosh ρ cosh ( ρ /2 ) d ρ = 2 π ( u > 0 ) .
The substitution v = sinh ( ρ / 2 ) v=\sinh(\rho/2) v = sinh ( ρ /2 ) gives this elementary integral. Taking absolute values in (188) therefore gives, without any positivity assumption on K H K_H K H ,
∣ P Y ( H ) ∣ ≤ B ( H ) H ⌊ 2 Y sinh ( T / 2 ) ⌋ ≤ B ( H ) [ 1 + log + ( 2 Y sinh ( T / 2 ) ) ] . (195) \lvert P_Y(H)\rvert\le B(H)H_{\lfloor2Y\sinh(T/2)\rfloor}\le B(H)[1+\log_+(2Y\sinh(T/2))]. \tag*{(195)} ∣ P Y ( H )∣ ≤ B ( H ) H ⌊ 2 Y s i n h ( T /2 )⌋ ≤ B ( H ) [ 1 + log + ( 2 Y sinh ( T /2 ))] . (195)
Since ∣ p ( f ( r ) ) ∣ ≤ N R ( p ) f ( r ) \lvert p(f(r))\rvert\le\mathrm N_{\mathbb R}(p)f(r) ∣ p ( f ( r ))∣ ≤ N R ( p ) f ( r ) on the real axis, (176) proves the identity bound, and differentiated Fourier inversion gives ∥ H p ′ ∥ ∞ ≤ C 1 , f N R ( p ) \lVert H'_p\rVert_\infty\le C_{1,f}\mathrm N_{\mathbb R}(p) ∥ H p ′ ∥ ∞ ≤ C 1 , f N R ( p ) . At T = L Q T=LQ T = L Q , (194) yields B ( H p ) ≤ L Q C 1 , f N R ( p ) B(H_p)\le LQC_{1,f}\mathrm N_{\mathbb R}(p) B ( H p ) ≤ L Q C 1 , f N R ( p ) , while
2 Y Q sinh ( L Q / 2 ) ≤ 3 e 3 L Q / 2 , H ⌊ 2 Y Q sinh ( L Q / 2 ) ⌋ ≤ 1 + log 3 + 3 L Q / 2. 2Y_Q\sinh(LQ/2)\le3e^{3LQ/2},\qquad H_{\lfloor2Y_Q\sinh(LQ/2)\rfloor}\le1+\log3+3LQ/2. 2 Y Q sinh ( L Q /2 ) ≤ 3 e 3 L Q /2 , H ⌊ 2 Y Q s i n h ( L Q /2 )⌋ ≤ 1 + log 3 + 3 L Q /2.
This proves the second bound in (191) . For the base test p = x p=x p = x , the test function H p = h H_p=h H p = h has support in [ − L , L ] [-L,L] [ − L , L ] , independently of Q Q Q , so only the cutoff Y Q Y_Q Y Q contributes growth in (195) :
∣ P Y Q ( x ) ∣ ≤ B ( h ) [ 1 + L Q + log + ( 6 sinh ( L / 2 ) ) ] ≤ b f Q . \lvert P_{Y_Q}(x)\rvert\le B(h)[1+LQ+\log_+(6\sinh(L/2))]\le b_fQ. ∣ P Y Q ( x )∣ ≤ B ( h ) [ 1 + L Q + log + ( 6 sinh ( L /2 ))] ≤ b f Q .
Let P 0 P_0 P 0 denote the orthogonal projection onto the constant eigenfunction. The normalized eigenfunction is u 0 = A g , n − 1 / 2 u_0 = A_{g,n}^{-1/2} u 0 = A g , n − 1/2 , so its truncated squared mass equals 1 − n / ( A g , n Y Q ) 1 - n/(\mathcal{A}_{g,n}Y_Q) 1 − n / ( A g , n Y Q ) . In particular
τ X , Q ∘ ( p ) : = Tr ( χ Y Q p ( A X ) ( 1 − P 0 ) χ Y Q ) = τ X , Q ( p ) − ( 1 − n A g , n Y Q ) p ( a ) . (196) \tau^\circ_{X,Q}(p) := \operatorname{Tr}\bigl(\chi_{Y_Q}p(A_X)(1-P_0)\chi_{Y_Q}\bigr) = \tau_{X,Q}(p) - \left(1-\frac{n}{\mathcal{A}_{g,n}Y_Q}\right)p(a). \tag*{(196)} τ X , Q ∘ ( p ) := Tr ( χ Y Q p ( A X ) ( 1 − P 0 ) χ Y Q ) = τ X , Q ( p ) − ( 1 − A g , n Y Q n ) p ( a ) . (196)
Subtracting the displayed rank-one compression preserves trace class. Since P 0 P_0 P 0 is a spectral projection, it commutes with p ( A X ) p(A_X) p ( A X ) . The compressed operator is therefore positive whenever p p p is nonnegative on the remaining spectral image, and 0 ≤ τ X , Q ∘ ( x ) ≤ τ X , Q ( x ) 0 \le\tau^\circ_{X,Q}(x) \le\tau_{X,Q}(x) 0 ≤ τ X , Q ∘ ( x ) ≤ τ X , Q ( x ) .
Combining (178) , (196) and Proposition 5.1 gives the exact decomposition
τ X , Q ∘ ( p ) = ( A g , n − n Y Q ) I ( p ) + n P Y Q ( p ) + μ 0 ( p ) + ( O X ( p ) − ω 1 ( p ) ) + n p ( a ) A g , n Y Q . (197) \begin{aligned}
\tau^\circ_{X,Q}(p)={}&\left(\mathcal{A}_{g,n}-\frac{n}{Y_Q}\right)\mathcal{I}(p)+n\mathcal{P}_{Y_Q}(p)+\mu_0(p)\\
&+\bigl(\mathcal{O}_X(p)-\omega_1(p)\bigr)+\frac{np(a)}{\mathcal{A}_{g,n}Y_Q}. \tag*{(197)}
\end{aligned} τ X , Q ∘ ( p ) = ( A g , n − Y Q n ) I ( p ) + n P Y Q ( p ) + μ 0 ( p ) + ( O X ( p ) − ω 1 ( p ) ) + A g , n Y Q n p ( a ) . (197)
The remaining constant-mode term, n p ( a ) / ( A g , n Y Q ) np(a)/(\mathcal{A}_{g,n}Y_Q) n p ( a ) / ( A g , n Y Q ) , comes from the mass removed by truncation. The signed measure μ 0 \mu_0 μ 0 is supported on [ 0 , 1 ] [0,1] [ 0 , 1 ] , and the identity and parabolic terms are controlled by N R ( p ) N_{\mathbb{R}}(p) N R ( p ) above. § 6 estimates the remaining expectation E g , n [ O X ( p ) − ω 1 ( p ) ] \mathbb{E}_{g,n}[\mathcal{O}_X(p)-\omega_1(p)] E g , n [ O X ( p ) − ω 1 ( p )] through the orbital expansions.
Polynomial approximation and spectral detection
We prove the spectral count in Theorem 6.13 by applying a positive polynomial detector to the truncated trace. The detector is small on the real spectral image and close to one on the interval corresponding to low eigenvalues. A uniform bound on the mass lost in the cusps then converts its trace into an eigenvalue count.
Proposition 6.6 illustrates the interpolation method with a preliminary count. The final estimate uses the coefficient bounds of § 6.6 , the additive remainder of § 6.7 , and the degree-independent first-correction bound obtained from positivity in § 6.8 .
Throughout this section, g ≥ 3 g \ge3 g ≥ 3 , n ≥ 1 n \ge1 n ≥ 1 and the degree bound Q ≥ 1 Q \ge1 Q ≥ 1 are integers. Polynomial tests are real and have zero constant term. The integer c = c f ≥ 2 c=c_f \ge2 c = c f ≥ 2 is fixed by Proposition 3.6 ; the constants introduced for the additive transfer below are distinct. For the preliminary approximation set
α = 2 c + 2 , β = 2 c + 6 , G n , Q = [ 4 [ c ( 2 c Q + 2 ) ( n + 1 ) ] c + 1 ] , Y Q = 3 e L Q . (198) \alpha=2c+2,\qquad\beta=2c+6,\qquad G_{n,Q}=\left[4[c(2cQ+2)(n+1)]^{c+1}\right],\qquad Y_Q=3e^{LQ}. \tag*{(198)} α = 2 c + 2 , β = 2 c + 6 , G n , Q = [ 4 [ c ( 2 c Q + 2 ) ( n + 1 ) ] c + 1 ] , Y Q = 3 e L Q . (198)
For p = x p ~ p=x\widetilde{p} p = x p we use the norms N ∗ ( p ) = ∥ p ~ ∥ [ 0 , a ] N_*(p)=\|\widetilde{p}\|_{[0,a]} N ∗ ( p ) = ∥ p ∥ [ 0 , a ] and N R ( p ) = ∥ p ~ ∥ [ 0 , 1 ] N_{\mathbb{R}}(p)=\|\widetilde{p}\|_{[0,1]} N R ( p ) = ∥ p ∥ [ 0 , 1 ] from the preceding section. The interpolation method has a compact-surface precedent in [10 Lemma 3.1 and equations (3.1)–(3.9)] ; the argument below retains the cusp terms and gives an explicit genus threshold.
A first moment and reciprocal-genus interpolation
We begin with a first moment that will control all polynomial traces through positivity. The exact identity of Theorem 5.3 , with τ X , Q ( p ) = Tr ( χ Y Q p ( A X ) χ Y Q ) \tau_{X,Q}(p)=\operatorname{Tr}(\chi_{Y_Q}p(A_X)\chi_{Y_Q}) τ X , Q ( p ) = Tr ( χ Y Q p ( A X ) χ Y Q ) , reads
τ X , Q ( p ) = ( A g , n − n Y Q ) I ( p ) + n P Y Q ( p ) + O X ( p ) , ∣ τ X , Q ( p ) ∣ ≤ N ∗ ( p ) τ X , Q ( x ) . (199) \tau_{X,Q}(p)=\left(\mathcal{A}_{g,n}-\frac{n}{Y_Q}\right)\mathcal{I}(p)+n\mathcal{P}_{Y_Q}(p)+\mathcal{O}_X(p),\qquad\lvert\tau_{X,Q}(p)\rvert\le N_*(p)\tau_{X,Q}(x). \tag*{(199)} τ X , Q ( p ) = ( A g , n − Y Q n ) I ( p ) + n P Y Q ( p ) + O X ( p ) , ∣ τ X , Q ( p )∣ ≤ N ∗ ( p ) τ X , Q ( x ) . (199)
Proposition 5.4 supplies fixed finite constants I f , J f , b f ≥ 0 I_f,J_f,b_f\ge0 I f , J f , b f ≥ 0 such that
∣ I ( p ) ∣ ≤ I f N R ( p ) , ∣ P Y Q ( p ) ∣ ≤ J f Q 2 N R ( p ) , ∣ P Y Q ( x ) ∣ ≤ b f Q . (200) \lvert\mathcal{I}(p)\rvert\le I_fN_{\mathbb{R}}(p),\qquad\lvert\mathcal{P}_{Y_Q}(p)\rvert\le J_fQ^2N_{\mathbb{R}}(p),\qquad\lvert\mathcal{P}_{Y_Q}(x)\rvert\le b_fQ. \tag*{(200)} ∣ I ( p )∣ ≤ I f N R ( p ) , ∣ P Y Q ( p )∣ ≤ J f Q 2 N R ( p ) , ∣ P Y Q ( x )∣ ≤ b f Q . (200)
In particular, these bounds use only the multiplier on the real spectral axis; they assign no sign to the parabolic term.
Apply Proposition 3.6 with ( q , t ) = ( 2 , 1 ) (q,t)=(2,1) ( q , t ) = ( 2 , 1 ) . Since Proposition 5.1 gives ∣ ω 1 ( x ) ∣ ≤ 6 a |\omega_1(x)|\le6a ∣ ω 1 ( x ) ∣ ≤ 6 a , we have
∣ E g , n O X ( x ) − ω 1 ( x ) ∣ ≤ [ 2 c ( n + 1 ) ] 2 c g ≤ g , g ≥ [ 2 c ( n + 1 ) ] c . (201) |\mathbb{E}_{g,n}\mathcal{O}_X(x)-\omega_1(x)|\le\frac{[2c(n+1)]^{2c}}{g}\le g,\qquad g\ge[2c(n+1)]^c. \tag*{(201)} ∣ E g , n O X ( x ) − ω 1 ( x ) ∣ ≤ g [ 2 c ( n + 1 ) ] 2 c ≤ g , g ≥ [ 2 c ( n + 1 ) ] c . (201)
The base orbital statistic is nonnegative because h ≥ 0 h\ge0 h ≥ 0 , and its expectation is finite. Consequently the positive trace in (199) is integrable. Since 0 ≤ A g , n − n / Y Q ≤ 4 π g + 2 π n 0\le\mathcal{A}_{g,n}-n/Y_Q\le4\pi g+2\pi n 0 ≤ A g , n − n / Y Q ≤ 4 π g + 2 π n , equations (200) –(201) imply
E g , n τ X , Q ( x ) ≤ D f ( g + n Q ) , D f = 1 + 6 a + 6 π I f + b f . (202) \mathbb{E}_{g,n}\tau_{X,Q}(x)\le D_f(g+nQ),\qquad D_f=1+6a+6\pi I_f+b_f. \tag*{(202)} E g , n τ X , Q ( x ) ≤ D f ( g + n Q ) , D f = 1 + 6 a + 6 π I f + b f . (202)
This estimate holds simultaneously for every Q ≥ 1 Q\ge1 Q ≥ 1 at the explicit genus threshold in (201) . Trace domination and the exact identity then give absolute integrability and
E g , n ∣ O X ( p ) ∣ ≤ N ∗ ( p ) { D f ( g + n Q ) + ( 4 π g + 2 π n ) I f + n J f Q 2 } . (203) \mathbb{E}_{g,n}|\mathcal{O}_X(p)|\le N_*(p)\{D_f(g+nQ)+(4\pi g+2\pi n)I_f+nJ_fQ^2\}. \tag*{(203)} E g , n ∣ O X ( p ) ∣ ≤ N ∗ ( p ) { D f ( g + n Q ) + ( 4 π g + 2 π n ) I f + n J f Q 2 } . (203)
Thus, if also g ≥ n Q 2 g\ge nQ^2 g ≥ n Q 2 , then
∣ g − 1 E g , n O X ( p ) ∣ ≤ A f N ∗ ( p ) , A f = 2 D f + 6 π I f + J f . (204) |g^{-1}\mathbb{E}_{g,n}\mathcal{O}_X(p)|\le A_fN_*(p),\qquad A_f=2D_f+6\pi I_f+J_f. \tag*{(204)} ∣ g − 1 E g , n O X ( p ) ∣ ≤ A f N ∗ ( p ) , A f = 2 D f + 6 π I f + J f . (204)
This a priori bound uses only the leading orbital coefficient. The next two lemmas turn it into a polynomial approximation estimate: the first controls monomial coefficients, and the second controls a polynomial from its values at reciprocal integers.
Lemma 6.1 (Coefficient bound). If p ( x ) = ∑ t = 1 Q a t x t p(x)=\sum_{t=1}^{Q}a_tx^t p ( x ) = ∑ t = 1 Q a t x t , then ∑ t = 1 Q ∣ a t ∣ ≤ 7 Q N ∗ ( p ) \sum_{t=1}^{Q}|a_t|\le7^QN_*(p) ∑ t = 1 Q ∣ a t ∣ ≤ 7 Q N ∗ ( p ) .
Proof. Expand p ~ ( x ) = ∑ k = 0 Q − 1 γ k T k ( 2 x / a − 1 ) \widetilde p(x)=\sum_{k=0}^{Q-1}\gamma_kT_k(2x/a-1) p ( x ) = ∑ k = 0 Q − 1 γ k T k ( 2 x / a − 1 ) in shifted Chebyshev polynomials. The cosine coefficient formulas give ∣ γ 0 ∣ ≤ N ∗ ( p ) |\gamma_0|\le N_*(p) ∣ γ 0 ∣ ≤ N ∗ ( p ) and ∣ γ k ∣ ≤ 2 N ∗ ( p ) |\gamma_k|\le2N_*(p) ∣ γ k ∣ ≤ 2 N ∗ ( p ) for k ≥ 1 k\ge1 k ≥ 1 . For the sum ∥ ⋅ ∥ ℓ 1 \|\cdot\|_{\ell^1} ∥ ⋅ ∥ ℓ 1 of absolute monomial coefficients, one has ∥ 2 x / a − 1 ∥ ℓ 1 ≤ 3 \|2x/a-1\|_{\ell^1}\le3 ∥2 x / a − 1 ∥ ℓ 1 ≤ 3 because a > 1 a>1 a > 1 . The recurrence for T k T_k T k therefore gives
∥ T k + 1 ( 2 x / a − 1 ) ∥ ℓ 1 ≤ 6 ∥ T k ( 2 x / a − 1 ) ∥ ℓ 1 + ∥ T k − 1 ( 2 x / a − 1 ) ∥ ℓ 1 ≤ 7 k + 1 , \|T_{k+1}(2x/a-1)\|_{\ell^1}\le6\|T_k(2x/a-1)\|_{\ell^1}+\|T_{k-1}(2x/a-1)\|_{\ell^1}\le7^{k+1}, ∥ T k + 1 ( 2 x / a − 1 ) ∥ ℓ 1 ≤ 6∥ T k ( 2 x / a − 1 ) ∥ ℓ 1 + ∥ T k − 1 ( 2 x / a − 1 ) ∥ ℓ 1 ≤ 7 k + 1 ,
by induction from degrees zero and one. It follows that
∥ p ~ ∥ ℓ 1 ≤ ( 1 + 2 ∑ k = 1 Q − 1 7 k ) N ∗ ( p ) = 7 Q − 4 3 N ∗ ( p ) ≤ 7 Q N ∗ ( p ) . \|\widetilde p\|_{\ell^1}\le\left(1+2\sum_{k=1}^{Q-1}7^k\right)N_*(p)=\frac{7^Q-4}{3}N_*(p)\le7^QN_*(p). ∥ p ∥ ℓ 1 ≤ ( 1 + 2 k = 1 ∑ Q − 1 7 k ) N ∗ ( p ) = 3 7 Q − 4 N ∗ ( p ) ≤ 7 Q N ∗ ( p ) .
Multiplication by x x x leaves this coefficient sum unchanged.
Lemma 6.2 (Reciprocal-integer grid). Let d 0 ≥ 1 d_0\ge1 d 0 ≥ 1 and G ≥ 4 d 0 2 G\ge4d_0^2 G ≥ 4 d 0 2 be integers, let M ≥ 0 M\ge0 M ≥ 0 , and let R R R be a real polynomial of degree at most d 0 d_0 d 0 . If ∣ R ( 0 ) ∣ ≤ M |R(0)|\le M ∣ R ( 0 ) ∣ ≤ M and ∣ R ( 1 / k ) ∣ ≤ M |R(1/k)|\le M ∣ R ( 1/ k ) ∣ ≤ M for every integer k ≥ G k\ge G k ≥ G , then
∥ R ∥ [ 0 , 1 / G ] ≤ 2 M , ∥ R ′ ′ ∥ [ 0 , 1 / G ] ≤ 8 d 0 4 G 2 M . (205) \|R\|_{[0,1/G]}\le2M,\qquad\|R''\|_{[0,1/G]}\le8d_0^4G^2M. \tag*{(205)} ∥ R ∥ [ 0 , 1/ G ] ≤ 2 M , ∥ R ′′ ∥ [ 0 , 1/ G ] ≤ 8 d 0 4 G 2 M . (205)
Proof. Write S = ∥ R ∥ [ 0 , 1 / G ] S=\|R\|_{[0,1/G]} S = ∥ R ∥ [ 0 , 1/ G ] . Consecutive nodes have spacing 1 / [ k ( k + 1 ) ] ≤ G − 2 1/[k(k+1)]\le G^{-2} 1/ [ k ( k + 1 )] ≤ G − 2 , so every point of the interval lies within G − 2 G^{-2} G − 2 of a node or their limiting point zero. The classical Markov inequality, rescaled from [ − 1 , 1 ] [-1,1] [ − 1 , 1 ] , is ∥ R ′ ∥ [ 0 , 1 / G ] ≤ 2 d 0 2 G S \|R'\|_{[0,1/G]}\le2d_0^2GS ∥ R ′ ∥ [ 0 , 1/ G ] ≤ 2 d 0 2 GS . Apply the mean value theorem at a point attaining S S S to obtain S ≤ M + ( 2 d 0 2 / G ) S ≤ M + S / 2 S\le M+(2d_0^2/G)S\le M+S/2 S ≤ M + ( 2 d 0 2 / G ) S ≤ M + S /2 . Hence S ≤ 2 M S\le2M S ≤ 2 M . Applying Markov once more gives ∥ R ′ ′ ∥ ≤ 4 d 0 2 ( d 0 − 1 ) 2 G 2 S ≤ 8 d 0 4 G 2 M \|R''\|\le4d_0^2(d_0-1)^2G^2S\le8d_0^4G^2M ∥ R ′′ ∥ ≤ 4 d 0 2 ( d 0 − 1 ) 2 G 2 S ≤ 8 d 0 4 G 2 M . For degree at most one the second derivative vanishes.
Proposition 6.3 (Polynomial approximation). There is C P M < ∞ C_{\mathrm{PM}} < \infty C PM < ∞ , depending only on f f f and c c c , such that for all integers n , Q ≥ 1 n,Q \ge1 n , Q ≥ 1 , g ≥ G n , Q g \ge G_{n,Q} g ≥ G n , Q , and real polynomials p = x p ~ p=x\tilde p p = x p ~ of degree at most Q Q Q ,
∣ E g , n O X ( p ) − ω 1 ( p ) ∣ ≤ C P M ( n + 1 ) α Q β g N ∗ ( p ) . (206) \left|\mathbb{E}_{g,n}\mathcal{O}_{X}(p)-\omega_{1}(p)\right| \le\frac{C_{\mathrm{PM}}(n+1)^{\alpha}Q^{\beta}}{g}N_{*}(p). \tag*{(206)} ∣ E g , n O X ( p ) − ω 1 ( p ) ∣ ≤ g C PM ( n + 1 ) α Q β N ∗ ( p ) . (206) Moreover,
G n , Q ≤ 5 [ 3 c 2 Q ( n + 1 ) ] c + 1 . (207) G_{n,Q} \le5[3c^{2}Q(n+1)]^{c+1}. \tag*{(207)} G n , Q ≤ 5 [ 3 c 2 Q ( n + 1 ) ] c + 1 . (207)
Proof. We keep the cusp count and test polynomial fixed while varying the genus. This bounds a single polynomial in reciprocal genus; Taylor’s theorem then isolates its linear coefficient. Fix n , Q , p n,Q,p n , Q , p and put q = 2 c Q + 2 q=2cQ+2 q = 2 c Q + 2 , u = c q ( n + 1 ) u=cq(n+1) u = c q ( n + 1 ) , G = G n , Q G=G_{n,Q} G = G n , Q , and d 0 = q − 1 d_{0}=q-1 d 0 = q − 1 . The coefficient functionals in Proposition 3.6 are independent of genus and truncation order, so
R p ( z ) = ∑ j = 1 q − 1 ω j , n ( p ) z j R_{p}(z)=\sum_{j=1}^{q-1}\omega_{j,n}(p)z^{j} R p ( z ) = j = 1 ∑ q − 1 ω j , n ( p ) z j
is one fixed polynomial while the integer genus varies. It has R p ( 0 ) = 0 R_{p}(0)=0 R p ( 0 ) = 0 and R p ′ ( 0 ) = ω 1 ( p ) R_{p}'(0)=\omega_{1}(p) R p ′ ( 0 ) = ω 1 ( p ) , by Proposition 5.1 . Linearity, Q < q Q<q Q < q , and Lemma 6.1 yield
∣ g − 1 E g , n O X ( p ) − R p ( 1 / g ) ∣ ≤ 7 Q u c q g − q N ∗ ( p ) , g ≥ u c . (208) \left|g^{-1}\mathbb{E}_{g,n}\mathcal{O}_{X}(p)-R_{p}(1/g)\right| \le7^{Q}u^{cq}g^{-q}N_{*}(p), \qquad g\ge u^{c}. \tag*{(208)} g − 1 E g , n O X ( p ) − R p ( 1/ g ) ≤ 7 Q u c q g − q N ∗ ( p ) , g ≥ u c . (208)
The definition of G G G gives
G ≥ u c ≥ [ 2 c ( n + 1 ) ] c , G ≥ n Q 2 , G ≥ 4 d 0 2 , G q − 2 ≥ 7 Q u c q . (209) G\ge u^{c}\ge[2c(n+1)]^{c}, \qquad G\ge nQ^{2}, \qquad G\ge4d_{0}^{2}, \qquad G^{q-2}\ge7^{Q}u^{cq}. \tag*{(209)} G ≥ u c ≥ [ 2 c ( n + 1 ) ] c , G ≥ n Q 2 , G ≥ 4 d 0 2 , G q − 2 ≥ 7 Q u c q . (209)
For the first three comparisons use u ≥ q u\ge q u ≥ q , u ≥ Q ( n + 1 ) u\ge Q(n+1) u ≥ Q ( n + 1 ) , and c + 1 ≥ 3 c+1\ge3 c + 1 ≥ 3 . For the last one, the exact exponent calculation is
( c + 1 ) ( q − 2 ) − c q = 2 c ( Q − 1 ) , G q − 2 u c q ≥ 4 2 c Q u 2 c ( Q − 1 ) ≥ 7 Q . (c+1)(q-2)-cq=2c(Q-1), \qquad\frac{G^{q-2}}{u^{cq}}\ge4^{2cQ}u^{2c(Q-1)}\ge7^{Q}. ( c + 1 ) ( q − 2 ) − c q = 2 c ( Q − 1 ) , u c q G q − 2 ≥ 4 2 c Q u 2 c ( Q − 1 ) ≥ 7 Q .
Thus for every integer g ≥ G g\ge G g ≥ G , equations (204) and (208) give
∣ g − 1 E g , n O X ( p ) − R p ( 1 / g ) ∣ ≤ g − 2 N ∗ ( p ) , ∣ R p ( 1 / g ) ∣ ≤ ( A f + 1 ) N ∗ ( p ) . \left|g^{-1}\mathbb{E}_{g,n}\mathcal{O}_{X}(p)-R_{p}(1/g)\right|\le g^{-2}N_{*}(p), \qquad|R_{p}(1/g)|\le(A_{f}+1)N_{*}(p). g − 1 E g , n O X ( p ) − R p ( 1/ g ) ≤ g − 2 N ∗ ( p ) , ∣ R p ( 1/ g ) ∣ ≤ ( A f + 1 ) N ∗ ( p ) .
Lemma 6.2 , followed by Taylor’s theorem at zero, now implies
∣ E g , n O X ( p ) − ω 1 ( p ) ∣ ≤ 1 + 4 ( A f + 1 ) d 0 4 G 2 g N ∗ ( p ) . (210) \left|\mathbb{E}_{g,n}\mathcal{O}_{X}(p)-\omega_{1}(p)\right|\le\frac{1+4(A_{f}+1)d_{0}^{4}G^{2}}{g}N_{*}(p). \tag*{(210)} ∣ E g , n O X ( p ) − ω 1 ( p ) ∣ ≤ g 1 + 4 ( A f + 1 ) d 0 4 G 2 N ∗ ( p ) . (210)
Since q ≤ 3 c Q q\le3cQ q ≤ 3 c Q , u ≤ 3 c 2 Q ( n + 1 ) u\le3c^{2}Q(n+1) u ≤ 3 c 2 Q ( n + 1 ) , and G ≤ 5 u c + 1 G\le5u^{c+1} G ≤ 5 u c + 1 , equation (207) follows, and
d 0 4 G 2 ≤ 25 ( 3 c ) 4 ( 3 c 2 ) 2 c + 2 Q 2 c + 6 ( n + 1 ) 2 c + 2 . d_{0}^{4}G^{2}\le25(3c)^{4}(3c^{2})^{2c+2}Q^{2c+6}(n+1)^{2c+2}. d 0 4 G 2 ≤ 25 ( 3 c ) 4 ( 3 c 2 ) 2 c + 2 Q 2 c + 6 ( n + 1 ) 2 c + 2 .
This proves (206) ; for example one may take C P M = 1 + 100 ( A f + 1 ) ( 3 c ) 4 ( 3 c 2 ) 2 c + 2 C_{\mathrm{PM}}=1+100(A_{f}+1)(3c)^{4}(3c^{2})^{2c+2} C PM = 1 + 100 ( A f + 1 ) ( 3 c ) 4 ( 3 c 2 ) 2 c + 2 .
A positive detector
Fix 0 < ε < 1 / 4 0<\varepsilon<1/4 0 < ε < 1/4 . We seek a polynomial that is exponentially small on [ 0 , 1 ] [0,1] [ 0 , 1 ] and close to one on [ b ε , a ] [b_{\varepsilon},a] [ b ε , a ] , while its quotient by x x x remains uniformly bounded on [ 0 , a ] [0,a] [ 0 , a ] . The constants below specify the transition and the degree thresholds:
b ε = f ( i ε ) , θ = 1 + b ε 2 a , d = b ε − 1 2 a , κ = d 2 4 , B ε = 1 a θ , b_{\varepsilon}=f(i\sqrt{\varepsilon}), \qquad\theta=\frac{1+b_{\varepsilon}}{2a}, \qquad d=\frac{b_{\varepsilon}-1}{2a}, \qquad\kappa=\frac{d^{2}}{4}, \qquad B_{\varepsilon}=\frac{1}{a\theta}, b ε = f ( i ε ) , θ = 2 a 1 + b ε , d = 2 a b ε − 1 , κ = 4 d 2 , B ε = a θ 1 ,
Q w = max { 2 , 1 + ⌈ 2 ( 1 − θ ) d ⌉ } , Q ε = max { Q w , ⌈ log 2 8 κ ⌉ , ⌈ log 2 2 ε L ⌉ } . (211) Q_{\mathrm{w}}=\max\left\{2,1+\left\lceil\frac{2(1-\theta)}{d}\right\rceil\right\}, \qquad Q_{\varepsilon}=\max\left\{Q_{\mathrm{w}},\left\lceil\frac{\log2}{8\kappa}\right\rceil,\left\lceil\frac{\log2}{2\sqrt{\varepsilon}L}\right\rceil\right\}. \tag*{(211)} Q w = max { 2 , 1 + ⌈ d 2 ( 1 − θ ) ⌉ } , Q ε = max { Q w , ⌈ 8 κ log 2 ⌉ , ⌈ 2 ε L log 2 ⌉ } . (211)
Indeed, f ( i s ) = ∫ R h ( u ) cosh ( s u ) d u f(is)=\int_{\mathbb{R}}h(u)\cosh(su)\,du f ( i s ) = ∫ R h ( u ) cosh ( s u ) d u is strictly increasing for s > 0 s>0 s > 0 : its derivative is ∫ R u h ( u ) sinh ( s u ) d u > 0 \int_{\mathbb{R}}uh(u)\sinh(su)\,du>0 ∫ R u h ( u ) sinh ( s u ) d u > 0 , since a smooth probability density is not supported at zero. Therefore 1 < b ε < a 1<b_{\varepsilon}<a 1 < b ε < a , θ − d = 1 / a \theta-d=1/a θ − d = 1/ a , and θ + d = b ε / a < 1 \theta+d=b_{\varepsilon}/a<1 θ + d = b ε / a < 1 .
Lemma 6.4 (Binomial detector). For every integer Q ≥ 1 Q \ge1 Q ≥ 1 , put
k Q = ⌈ θ Q ⌉ , p Q ( x ) = ∑ j = k Q Q ( Q j ) ( x / a ) j ( 1 − x / a ) Q − j . (212) k_Q=\lceil\theta Q\rceil,\qquad p_Q(x)=\sum_{j=k_Q}^{Q}\binom{Q}{j}(x/a)^j(1-x/a)^{Q-j}. \tag*{(212)} k Q = ⌈ θ Q ⌉ , p Q ( x ) = j = k Q ∑ Q ( j Q ) ( x / a ) j ( 1 − x / a ) Q − j . (212) Then p Q ( 0 ) = 0 p_Q(0)=0 p Q ( 0 ) = 0 , p Q ( a ) = 1 p_Q(a)=1 p Q ( a ) = 1 , 0 ≤ p Q ≤ 1 0\le p_Q\le1 0 ≤ p Q ≤ 1 on [ 0 , a ] [0,a] [ 0 , a ] , and
N ∗ ( p Q ) ≤ B ε , (213) N_*(p_Q)\le B_\varepsilon, \tag*{(213)} N ∗ ( p Q ) ≤ B ε , (213) N R ( p Q ) ≤ B ε e − κ Q ( Q ≥ Q w ) , (214) N_{\mathbb{R}}(p_Q)\le B_\varepsilon e^{-\kappa Q}\qquad(Q\ge Q_w), \tag*{(214)} N R ( p Q ) ≤ B ε e − κ Q ( Q ≥ Q w ) , (214) inf [ b ε , a ] p Q ≥ 1 − e − 8 κ Q . (215) \inf_{[b_\varepsilon,a]}p_Q\ge1-e^{-8\kappa Q}. \tag*{(215)} [ b ε , a ] inf p Q ≥ 1 − e − 8 κ Q . (215)
Proof. For t = x / a ∈ [ 0 , 1 ] t=x/a\in[0,1] t = x / a ∈ [ 0 , 1 ] , p Q ( x ) p_Q(x) p Q ( x ) is P { Bin ( Q , t ) ≥ k Q } \mathbb{P}\{\operatorname{Bin}(Q,t)\ge k_Q\} P { Bin ( Q , t ) ≥ k Q } , giving the endpoint values and positivity. We first establish the concentration bound used below. For a Bernoulli variable of parameter t t t , let F t ( s ) = log ( 1 − t + t e s ) − t s F_t(s)=\log(1-t+te^s)-ts F t ( s ) = log ( 1 − t + t e s ) − t s . Then F t ( 0 ) = F t ′ ( 0 ) = 0 F_t(0)=F_t'(0)=0 F t ( 0 ) = F t ′ ( 0 ) = 0 and
F t ′ ′ ( s ) = t ( 1 − t ) e s ( 1 − t + t e s ) 2 ≤ 1 4 , F t ( s ) ≤ s 2 / 8. F_t''(s)=\frac{t(1-t)e^s}{(1-t+te^s)^2}\le\frac{1}{4},\qquad F_t(s)\le s^2/8. F t ′′ ( s ) = ( 1 − t + t e s ) 2 t ( 1 − t ) e s ≤ 4 1 , F t ( s ) ≤ s 2 /8.
Independence and the exponential Markov inequality at s = ± 4 η s=\pm4\eta s = ± 4 η give, for B ∼ Bin ( N , t ) B\sim\operatorname{Bin}(N,t) B ∼ Bin ( N , t ) , N ≥ 1 N\ge1 N ≥ 1 , and η > 0 \eta>0 η > 0 ,
P ( B / N − t ≥ η ) ≤ e − 2 N η 2 , P ( B / N − t ≤ − η ) ≤ e − 2 N η 2 . (216) \mathbb{P}(B/N-t\ge\eta)\le e^{-2N\eta^2},\qquad\mathbb{P}(B/N-t\le-\eta)\le e^{-2N\eta^2}. \tag*{(216)} P ( B / N − t ≥ η ) ≤ e − 2 N η 2 , P ( B / N − t ≤ − η ) ≤ e − 2 N η 2 . (216)
If t ≥ θ + d t\ge\theta+d t ≥ θ + d , failure of the detector means B ≤ k Q − 1 < θ Q B\le k_Q-1<\theta Q B ≤ k Q − 1 < θ Q ; hence (215) follows from (216) with η = d \eta=d η = d .
To control the quotient at zero, the identity j ( Q j ) = Q ( Q − 1 j − 1 ) j\binom{Q}{j}=Q\binom{Q-1}{j-1} j ( j Q ) = Q ( j − 1 Q − 1 ) gives
P { Bin ( Q , t ) ≥ k Q } ≤ Q t k Q P { Bin ( Q − 1 , t ) ≥ k Q − 1 } . (217) \mathbb{P}\{\operatorname{Bin}(Q,t)\ge k_Q\}\le\frac{Qt}{k_Q}\mathbb{P}\{\operatorname{Bin}(Q-1,t)\ge k_Q-1\}. \tag*{(217)} P { Bin ( Q , t ) ≥ k Q } ≤ k Q Qt P { Bin ( Q − 1 , t ) ≥ k Q − 1 } . (217)
Here the binomial variable with zero trials is identically zero. Dividing by x = a t > 0 x=at>0 x = a t > 0 and using k Q ≥ θ Q k_Q\ge\theta Q k Q ≥ θ Q proves (213) , including x = 0 x=0 x = 0 by polynomial continuity. If Q ≥ Q w Q\ge Q_w Q ≥ Q w and 0 < x ≤ 1 0<x\le1 0 < x ≤ 1 , then t ≤ θ − d t\le\theta-d t ≤ θ − d and
k Q − 1 Q − 1 ≥ θ − 1 − θ Q − 1 ≥ θ − d / 2 ≥ t + d / 2. \frac{k_Q-1}{Q-1}\ge\theta-\frac{1-\theta}{Q-1}\ge\theta-d/2\ge t+d/2. Q − 1 k Q − 1 ≥ θ − Q − 1 1 − θ ≥ θ − d /2 ≥ t + d /2.
Applying (216) to the last probability in (217) gives N R ( p Q ) ≤ B ε e − ( Q − 1 ) d 2 / 2 ≤ B ε e − κ Q N_{\mathbb{R}}(p_Q)\le B_\varepsilon e^{-(Q-1)d^2/2}\le B_\varepsilon e^{-\kappa Q} N R ( p Q ) ≤ B ε e − ( Q − 1 ) d 2 /2 ≤ B ε e − κ Q , since Q ≥ 2 Q\ge2 Q ≥ 2 . The displayed threshold also implies k Q ≥ 2 k_Q\ge2 k Q ≥ 2 , so the quotient vanishes at zero in this range. □ \square □
The exponentially small norm N R ( p Q ) N_{\mathbb{R}}(p_Q) N R ( p Q ) controls the identity and parabolic terms through (200) . The uniformly bounded norm N ∗ ( p Q ) N_*(p_Q) N ∗ ( p Q ) controls the orbital error.
Retained mass after cusp truncation
Hide [8 Lemma 2.7] , following Gamburd, gives a uniform lower bound on the mass retained at a fixed cusp height. We prove the following explicit estimate for the truncation Y Q Y_Q Y Q .
Lemma 6.5 (Uniform retained mass). If ψ \psi ψ is a normalized L 2 L^2 L 2 eigenfunction with 0 < λ ≤ 1 / 4 − ε 0<\lambda\le1/4-\varepsilon 0 < λ ≤ 1/4 − ε , then
∥ χ Y Q ψ ∥ 2 ≥ 1 − e − 2 ε L Q . (218) \|\chi_{Y_Q}\psi\|^2\ge1-e^{-2\sqrt{\varepsilon}LQ}. \tag*{(218)} ∥ χ Y Q ψ ∥ 2 ≥ 1 − e − 2 ε L Q . (218) This estimate includes residual eigenfunctions and is uniform in the surface and its cusp count.
Proof. Write λ = 1 / 4 − s 2 \lambda= 1/4 - s^2 λ = 1/4 − s 2 , so ε ≤ s < 1 / 2 \sqrt{\varepsilon} \le s < 1/2 ε ≤ s < 1/2 . In a width-one cusp, Fourier expansion and the eigenfunction equation give
ψ ( x , y ) = a 0 y 1 / 2 − s + ∑ m ≠ 0 a m y K s ( 2 π ∣ m ∣ y ) e 2 π i m x . (219) \psi(x,y) = a_0 y^{1/2-s} + \sum_{m\ne0} a_m \sqrt{y} K_s(2\pi|m|y)e^{2\pi i m x}. \tag*{(219)} ψ ( x , y ) = a 0 y 1/2 − s + m = 0 ∑ a m y K s ( 2 π ∣ m ∣ y ) e 2 π im x . (219)
To see the allowed terms, the m m m th mode satisfies − y 2 ψ m ′ ′ + ( 2 π m ) 2 y 2 ψ m = ( 1 / 4 − s 2 ) ψ m -y^2\psi_m'' + (2\pi m)^2y^2\psi_m = (1/4-s^2)\psi_m − y 2 ψ m ′′ + ( 2 π m ) 2 y 2 ψ m = ( 1/4 − s 2 ) ψ m . For m = 0 m=0 m = 0 , square integrability excludes y 1 / 2 + s y^{1/2+s} y 1/2 + s . For m ≠ 0 m\ne0 m = 0 , writing ψ m ( y ) = y v ( 2 π ∣ m ∣ y ) \psi_m(y) = \sqrt{y}\,v(2\pi|m|y) ψ m ( y ) = y v ( 2 π ∣ m ∣ y ) gives the modified Bessel equation. Its positive-axis asymptotics exclude the growing I s I_s I s solution and leave K s K_s K s ; see [20 equations 10.40.1–10.40.2] . The zero Fourier coefficient a 0 a_0 a 0 need not vanish.
The positive integral representation and order symmetry [20 equations 10.32.10 and 10.27.3] , applied at order − s -s − s , give
z s K s ( z ) = 2 s − 1 ∫ 0 ∞ t s − 1 exp ( − t − z 2 4 t ) d t , z > 0. (220) z^s K_s(z) = 2^{s-1}\int_0^\infty t^{s-1}\exp\left(-t-\frac{z^2}{4t}\right)\,dt,\qquad z>0. \tag*{(220)} z s K s ( z ) = 2 s − 1 ∫ 0 ∞ t s − 1 exp ( − t − 4 t z 2 ) d t , z > 0. (220)
The integrand is positive and decreases with z z z . Hence 0 < K s ( r z ) ≤ r − s K s ( z ) 0 < K_s(rz) \le r^{-s}K_s(z) 0 < K s ( r z ) ≤ r − s K s ( z ) for r ≥ 1 r \ge1 r ≥ 1 . If M b ( Y ) M_b(Y) M b ( Y ) is the squared mass in cusp b b b above height Y Y Y , Parseval and Tonelli yield
M b ( Y ) = ∣ a 0 ∣ 2 2 s Y − 2 s + ∑ m ≠ 0 ∣ a m ∣ 2 ∫ Y ∞ K s ( 2 π ∣ m ∣ y ) 2 d y y . (221) M_b(Y) = \frac{|a_0|^2}{2s}Y^{-2s} + \sum_{m\ne0}|a_m|^2\int_Y^\infty K_s(2\pi|m|y)^2\frac{dy}{y}. \tag*{(221)} M b ( Y ) = 2 s ∣ a 0 ∣ 2 Y − 2 s + m = 0 ∑ ∣ a m ∣ 2 ∫ Y ∞ K s ( 2 π ∣ m ∣ y ) 2 y d y . (221)
The zero mode scales exactly by r − 2 s r^{-2s} r − 2 s when Y Y Y is replaced by r Y rY r Y . In the other integrals substitute y = r v y=rv y = r v and use the same Bessel bound. It follows that M b ( 3 r ) ≤ r − 2 s M b ( 3 ) M_b(3r) \le r^{-2s}M_b(3) M b ( 3 r ) ≤ r − 2 s M b ( 3 ) . The height-three cusp neighborhoods are disjoint, so ∑ b M b ( 3 ) ≤ 1 \sum_b M_b(3) \le1 ∑ b M b ( 3 ) ≤ 1 and therefore ∑ b M b ( 3 r ) ≤ r − 2 s \sum_b M_b(3r) \le r^{-2s} ∑ b M b ( 3 r ) ≤ r − 2 s . Taking r = e L Q r=e^{LQ} r = e L Q proves (218) without a factor depending on the number of cusps. □
Exact cancellation and a preliminary spectral count
Let P 0 P_0 P 0 denote orthogonal projection onto constants. The normalized constant eigenfunction has retained squared mass 1 − n / ( A g , n Y Q ) 1 - n/(\mathcal{A}_{g,n}Y_Q) 1 − n / ( A g , n Y Q ) . Since p Q ( a ) = 1 p_Q(a)=1 p Q ( a ) = 1 , define
D X , Q = τ X , Q ( p Q ) − ( 1 − n A g , n Y Q ) = Tr ( χ Y Q p Q ( A X ) ( I − P 0 ) χ Y Q ) ≥ 0. (222) \begin{aligned}
D_{X,Q} = \tau_{X,Q}(p_Q) - \left(1-\frac{n}{\mathcal{A}_{g,n}Y_Q}\right) \\
&= \operatorname{Tr}(\chi_{Y_Q}p_Q(A_X)(I-P_0)\chi_{Y_Q}) \ge0. \tag*{(222)}
\end{aligned} D X , Q = τ X , Q ( p Q ) − ( 1 − A g , n Y Q n ) = Tr ( χ Y Q p Q ( A X ) ( I − P 0 ) χ Y Q ) ≥ 0. (222)
The compressed operator differs from χ Y Q p Q ( A X ) χ Y Q \chi_{Y_Q}p_Q(A_X)\chi_{Y_Q} χ Y Q p Q ( A X ) χ Y Q by a rank-one operator, so it is trace class by Theorem 5.3 . It is positive because P 0 P_0 P 0 commutes with p Q ( A X ) p_Q(A_X) p Q ( A X ) and p Q ≥ 0 p_Q \ge0 p Q ≥ 0 on [ 0 , a ] [0,a] [ 0 , a ] ; no commutation of P 0 P_0 P 0 with the cutoff is needed. The positive spectral expansion contains all nonconstant discrete eigenfunctions and the continuous contribution. Every eigenvalue counted by N ε ( X ) N_\varepsilon(X) N ε ( X ) has multiplier in [ b ε , a ] [b_\varepsilon,a] [ b ε , a ] . Lemmas 6.4 and 6.5 therefore give
D X , Q ≥ ( 1 − e − 8 κ Q ) ( 1 − e − 2 ε L Q ) N ε ( X ) ≥ 1 4 N ε ( X ) , Q ≥ Q ε . (223) D_{X,Q} \ge(1-e^{-8\kappa Q})(1-e^{-2\sqrt{\varepsilon}LQ})N_\varepsilon(X) \ge\frac{1}{4}N_\varepsilon(X),\qquad Q\ge Q_\varepsilon. \tag*{(223)} D X , Q ≥ ( 1 − e − 8 κ Q ) ( 1 − e − 2 ε L Q ) N ε ( X ) ≥ 4 1 N ε ( X ) , Q ≥ Q ε . (223)
This follows first for any finite collection of orthonormal low eigenfunctions. Finiteness of the positive trace then also proves finiteness of the count, with multiplicities.
Proposition 6.6 (Finite-parameter probability bound). For every integer n ≥ 1 n \ge1 n ≥ 1 , Q ≥ Q ε Q \ge Q_\varepsilon Q ≥ Q ε , and g ≥ G n , Q g \ge G_{n,Q} g ≥ G n , Q ,
P g , n ( N ε ≥ 1 ) ≤ E g , n N ε ≤ C f , ε , c { ( g + n Q 2 ) e − κ Q + ( n + 1 ) α Q β g + n A g , n Y Q } . (224) \mathbb{P}_{g,n}(N_\varepsilon\ge1) \le\mathbb{E}_{g,n}N_\varepsilon\le C_{f,\varepsilon,c}\left\{(g+nQ^2)e^{-\kappa Q}+\frac{(n+1)^\alpha Q^\beta}{g}+\frac{n}{\mathcal{A}_{g,n}Y_Q}\right\}. \tag*{(224)} P g , n ( N ε ≥ 1 ) ≤ E g , n N ε ≤ C f , ε , c { ( g + n Q 2 ) e − κ Q + g ( n + 1 ) α Q β + A g , n Y Q n } . (224) The constant is independent of g g g , n n n , Q Q Q .
Proof. Proposition 5.1 identifies ω 1 ( p ) = p ( a ) + μ 0 ( p ) \omega_1(p)=p(a)+\mu_0(p) ω 1 ( p ) = p ( a ) + μ 0 ( p ) , with supp μ 0 ⊂ [ 0 , 1 ] \operatorname{supp}\mu_0\subset[0,1] supp μ 0 ⊂ [ 0 , 1 ] and ∥ μ 0 ∥ T V < 5 \lVert\mu_0\rVert_{\mathrm{TV}}<5 ∥ μ 0 ∥ TV < 5 . Combining this identity with (199) , Proposition 6.3, and (222) gives the exact cancellation
E g , n D X Q = ( A g , n − n / Y Q ) I ( p Q ) + n P Y Q ( p Q ) + μ 0 ( p Q ) + n A g , n Y Q + e g , n , Q , ∣ e g , n , Q ∣ ≤ C P M ( n + 1 ) α Q β g N ∗ ( p Q ) . (225) \begin{aligned}
\mathbb{E}_{g,n}D X_Q={}&(A_{g,n}-n/Y_Q)\mathcal{I}(p_Q)+nP_{Y_Q}(p_Q)+\mu_0(p_Q)\\
&+\frac{n}{A_{g,n}Y_Q}+e_{g,n,Q},\qquad|e_{g,n,Q}|\le\frac{C_{\mathrm{PM}}(n+1)^\alpha Q^\beta}{g}N_*(p_Q). \tag*{(225)}
\end{aligned} E g , n D X Q = ( A g , n − n / Y Q ) I ( p Q ) + n P Y Q ( p Q ) + μ 0 ( p Q ) + A g , n Y Q n + e g , n , Q , ∣ e g , n , Q ∣ ≤ g C PM ( n + 1 ) α Q β N ∗ ( p Q ) . (225)
All expectations are absolutely convergent by (202) –(203) . Since ∣ μ 0 ( p ) ∣ ≤ 5 N R ( p ) |\mu_0(p)|\le5N_{\mathbb R}(p) ∣ μ 0 ( p ) ∣ ≤ 5 N R ( p ) , the detector and deterministic bounds imply
0 ≤ E g , n D X Q ≤ B ε ( A g , n I f + n J f Q 2 + 5 ) e − κ Q + B ε C P M ( n + 1 ) α Q β g + n A g , n Y Q . (226) 0\le\mathbb{E}_{g,n}D X_Q\le B_\varepsilon(A_{g,n}I_f+nJ_fQ^2+5)e^{-\kappa Q}
+B_\varepsilon C_{\mathrm{PM}}\frac{(n+1)^\alpha Q^\beta}{g}
+\frac{n}{A_{g,n}Y_Q}. \tag*{(226)} 0 ≤ E g , n D X Q ≤ B ε ( A g , n I f + n J f Q 2 + 5 ) e − κ Q + B ε C PM g ( n + 1 ) α Q β + A g , n Y Q n . (226)
Using A g , n ≤ 4 π g + 2 π n A_{g,n}\le4\pi g+2\pi n A g , n ≤ 4 π g + 2 π n , Q ≥ 1 Q\ge1 Q ≥ 1 , and (223) proves the expected-count estimate. The probability estimate follows from 1 { N ε ≥ 1 } ≤ N ε \mathbf{1}_{\{N_\varepsilon\ge1\}}\le N_\varepsilon 1 { N ε ≥ 1 } ≤ N ε .
The unit coefficient of the atom at a a a leaves exactly the displayed truncation correction when the constant mode is removed. No interval of nonconstant eigenvalues near zero has to be excluded.
A common logarithmic degree
The detector degree must grow to suppress the identity and parabolic terms. The following logarithmic choice will suffice in the final spectral estimate:
K = max { 1 , 4 / κ , 2 / L } , Q g = ⌈ K log g ⌉ . (227) K=\max\{1,4/\kappa,2/L\},\qquad Q_g=\lceil K\log g\rceil. \tag*{(227)} K = max { 1 , 4/ κ , 2/ L } , Q g = ⌈ K log g ⌉ . (227)
For all sufficiently large g g g , independently of n n n , Q g ≥ Q ε Q_g\ge Q_\varepsilon Q g ≥ Q ε , log g ≥ 1 \log g\ge1 log g ≥ 1 , and Q g ≤ 2 K log g Q_g\le2K\log g Q g ≤ 2 K log g . Whenever 1 ≤ n ≤ g 1\le n\le g 1 ≤ n ≤ g , the following bounds hold uniformly:
( g + n Q g 2 ) e − κ Q g ≤ g − 3 ( 1 + 4 K 2 ( log g ) 2 ) = O f , ε ( g − 2 ) , (228) (g+nQ_g^2)e^{-\kappa Q_g}\le g^{-3}(1+4K^2(\log g)^2)=O_{f,\varepsilon}(g^{-2}), \tag*{(228)} ( g + n Q g 2 ) e − κ Q g ≤ g − 3 ( 1 + 4 K 2 ( log g ) 2 ) = O f , ε ( g − 2 ) , (228)
n A g , n Y Q g ≤ 1 6 π e − L Q g ≤ 1 6 π g − 2 , (229) \frac{n}{A_{g,n}Y_{Q_g}}\le\frac{1}{6\pi}e^{-LQ_g}\le\frac{1}{6\pi}g^{-2}, \tag*{(229)} A g , n Y Q g n ≤ 6 π 1 e − L Q g ≤ 6 π 1 g − 2 , (229)
where the second line uses A g , n ≥ 2 π n A_{g,n}\ge2\pi n A g , n ≥ 2 π n .
Uniform norms of the cusp-polynomial coefficients
We now use reciprocal-genus interpolation to bound each canonical coefficient at finitely many cusp counts. Cusp polynomiality then extends the bound to all n n n , with an explicit power of n + 1 n+1 n + 1 . Throughout the remainder of this section write s = n + 1 s=n+1 s = n + 1 .
Lemma 6.7 (Uniform norms of the cusp-polynomial coefficients). Fix the integer D 0 = 2 c + 4 D_0=2c+4 D 0 = 2 c + 4 . There is a constant C 0 ≥ 2 C_0\ge2 C 0 ≥ 2 , depending only on f f f , such that for all integers j ≥ 2 j\ge2 j ≥ 2 , Q , n ≥ 1 Q,n\ge1 Q , n ≥ 1 , and all real polynomials p = x p ~ p=x\tilde p p = x p ~ of degree at most Q Q Q ,
∣ ω j , n ( p ) ∣ ≤ s j [ C 0 ( Q + j ) ] D 0 j N ∗ ( p ) . (230) |\omega_{j,n}(p)|\le s^j[C_0(Q+j)]^{D_0j}N_*(p). \tag*{(230)} ∣ ω j , n ( p ) ∣ ≤ s j [ C 0 ( Q + j ) ] D 0 j N ∗ ( p ) . (230) No relation between j j j and Q Q Q is assumed.
Proof. We first bound the coefficient at the fixed cusp counts 1 , … , j + 1 1,\ldots,j+1 1 , … , j + 1 by varying an auxiliary integer genus. Polynomiality in the cusp count then gives the estimate at arbitrary n n n ; the auxiliary genus threshold is not imposed on the surfaces used later. Put A = A f + 1 A=A_f+1 A = A f + 1 , with A f A_f A f from (204) , and
M = Q + j , q = 2 c M + 2 , d 0 = q − 1 , u = c q ( j + 2 ) , G = ⌈ 4 u c + 1 ⌉ . (231) M=Q+j,\qquad q=2cM+2,\qquad d_0=q-1,\qquad u=cq(j+2),\qquad G=\left\lceil4u^{c+1}\right\rceil. \tag*{(231)} M = Q + j , q = 2 c M + 2 , d 0 = q − 1 , u = c q ( j + 2 ) , G = ⌈ 4 u c + 1 ⌉ . (231)
At each fixed cusp count r ∈ { 1 , … , j + 1 } r\in\{1,\ldots,j+1\} r ∈ { 1 , … , j + 1 } , use the polynomial R r , p ( z ) = ∑ k = 1 q − 1 ω k , r ( p ) z k R_{r,p}(z)=\sum_{k=1}^{q-1}\omega_{k,r}(p)z^k R r , p ( z ) = ∑ k = 1 q − 1 ω k , r ( p ) z k . Proposition 3.6 , applied only at these cusp counts, and Lemma 6.1 give
∣ g − 1 E g , r O X ( p ) − R r , p ( 1 / g ) ∣ ≤ 7 Q u c q g − q N ∗ ( p ) , g ≥ u c . (232) \left|g^{-1}\mathbb{E}_{g,r}O_X(p)-R_{r,p}(1/g)\right|\le7Q u^{cq}g^{-q}N_*(p),\qquad g\ge u^c. \tag*{(232)} g − 1 E g , r O X ( p ) − R r , p ( 1/ g ) ≤ 7 Q u c q g − q N ∗ ( p ) , g ≥ u c . (232)
Here q > Q q>Q q > Q , j j j and the coefficients are independent of the varying integer genus. The common grid threshold satisfies
G ≥ u c ≥ [ 2 c ( r + 1 ) ] c , G ≥ r Q 2 , G ≥ 4 d 0 2 , G q − 2 ≥ 7 Q u c q . (233) G\ge u^c\ge[2c(r+1)]^c,\qquad G\ge rQ^2,\qquad G\ge4d_0^2,\qquad G^{q-2}\ge7^Q u^{cq}. \tag*{(233)} G ≥ u c ≥ [ 2 c ( r + 1 ) ] c , G ≥ r Q 2 , G ≥ 4 d 0 2 , G q − 2 ≥ 7 Q u c q . (233)
Indeed, u ≥ q u\ge q u ≥ q , u ≥ Q ( r + 1 ) u\ge Q(r+1) u ≥ Q ( r + 1 ) and c + 1 ≥ 3 c+1\ge3 c + 1 ≥ 3 prove the first three bounds. For the last one,
( c + 1 ) ( q − 2 ) − c q = 2 c ( M − 1 ) , G q − 2 u c q ≥ 4 2 c M u 2 c ( M − 1 ) u c q ≥ 7 Q . (234) (c+1)(q-2)-cq=2c(M-1),\qquad\frac{G^{q-2}}{u^{cq}}\ge\frac{4^{2cM}u^{2c(M-1)}}{u^{cq}}\ge7^Q. \tag*{(234)} ( c + 1 ) ( q − 2 ) − c q = 2 c ( M − 1 ) , u c q G q − 2 ≥ u c q 4 2 c M u 2 c ( M − 1 ) ≥ 7 Q . (234)
The first two comparisons allow (204) at every node, while the last bounds the expansion error by g − 2 N ∗ ( p ) g^{-2}N_*(p) g − 2 N ∗ ( p ) . Hence ∣ R r , p ( 1 / g ) ∣ ≤ A N ∗ ( p ) \lvert R_{r,p}(1/g)\rvert\le AN_*(p) ∣ R r , p ( 1/ g )∣ ≤ A N ∗ ( p ) for all integers g ≥ G g\ge G g ≥ G . Also R r , p ( 0 ) = 0 R_{r,p}(0)=0 R r , p ( 0 ) = 0 . Lemma 6.2 gives ∥ R r , p ∥ [ 0 , 1 / G ] ≤ 2 A N ∗ ( p ) \lVert R_{r,p}\rVert_{[0,1/G]}\le2AN_*(p) ∥ R r , p ∥ [ 0 , 1/ G ] ≤ 2 A N ∗ ( p ) . Iterating the rescaled Markov inequality j j j times gives
∣ ω j , r ( p ) ∣ = ∣ R r , p ( j ) ( 0 ) ∣ j ! ≤ 2 A j ! ( 2 G d 0 2 ) j N ∗ ( p ) = : B j ( p ) . (235) \lvert\omega_{j,r}(p)\rvert=\frac{\lvert R_{r,p}^{(j)}(0)\rvert}{j!}\le\frac{2A}{j!}(2Gd_0^2)^jN_*(p)=:B_j(p). \tag*{(235)} ∣ ω j , r ( p )∣ = j ! ∣ R r , p ( j ) ( 0 )∣ ≤ j ! 2 A ( 2 G d 0 2 ) j N ∗ ( p ) =: B j ( p ) . (235)
Recall that ω j , n ( p ) = κ j − 1 , n ( H p ) \omega_{j,n}(p)=\kappa_{j-1,n}(H_p) ω j , n ( p ) = κ j − 1 , n ( H p ) . By Theorem 3.11 , W ( n ) = ω j , n ( p ) W(n)=\omega_{j,n}(p) W ( n ) = ω j , n ( p ) is therefore a polynomial of degree at most j j j . Its shifted Newton formula is
W ( n ) = ∑ r = 0 j ( n − 1 r ) Δ r W ( 1 ) , Δ r W ( 1 ) = ∑ v = 0 r ( − 1 ) r − v ( r v ) W ( 1 + v ) . (236) W(n)=\sum_{r=0}^{j}\binom{n-1}{r}\Delta^rW(1),\qquad\Delta^rW(1)=\sum_{v=0}^{r}(-1)^{r-v}\binom{r}{v}W(1+v). \tag*{(236)} W ( n ) = r = 0 ∑ j ( r n − 1 ) Δ r W ( 1 ) , Δ r W ( 1 ) = v = 0 ∑ r ( − 1 ) r − v ( v r ) W ( 1 + v ) . (236)
For n ≥ 1 n\ge1 n ≥ 1 , all binomial factors in the first sum are nonnegative and bounded by s j s^j s j ; those with r > n − 1 r>n-1 r > n − 1 vanish. The second sum has absolute value at most 2 r B j ( p ) 2^rB_j(p) 2 r B j ( p ) . Consequently
∣ W ( n ) ∣ ≤ 2 j + 1 s j B j ( p ) ≤ 4 A j ! s j ( 4 G d 0 2 ) j N ∗ ( p ) ≤ s j ( 16 A G d 0 2 ) j N ∗ ( p ) . (237) \lvert W(n)\rvert\le2^{j+1}s^jB_j(p)\le\frac{4A}{j!}s^j(4Gd_0^2)^jN_*(p)\le s^j(16AGd_0^2)^jN_*(p). \tag*{(237)} ∣ W ( n )∣ ≤ 2 j + 1 s j B j ( p ) ≤ j ! 4 A s j ( 4 G d 0 2 ) j N ∗ ( p ) ≤ s j ( 16 A G d 0 2 ) j N ∗ ( p ) . (237)
Finally q ≤ 3 c M q\le3cM q ≤ 3 c M , j + 2 ≤ 2 M j+2\le2M j + 2 ≤ 2 M , u ≤ 6 c 2 M 2 u\le6c^2M^2 u ≤ 6 c 2 M 2 and G ≤ 5 u c + 1 G\le5u^{c+1} G ≤ 5 u c + 1 , so
16 A G d 0 2 ≤ 720 A c 2 ( 6 c 2 ) c + 1 M 2 c + 4 ≤ [ C 0 M ] D 0 , (238) 16AGd_0^2\le720Ac^2(6c^2)^{c+1}M^{2c+4}\le[C_0M]^{D_0}, \tag*{(238)} 16 A G d 0 2 ≤ 720 A c 2 ( 6 c 2 ) c + 1 M 2 c + 4 ≤ [ C 0 M ] D 0 , (238)
where D 0 = 2 c + 4 D_0=2c+4 D 0 = 2 c + 4 and one may take C 0 = max { 2 , [ 720 A c 2 ( 6 c 2 ) c + 1 ] 1 / D 0 } C_0=\max\{2,[720Ac^2(6c^2)^{c+1}]^{1/D_0}\} C 0 = max { 2 , [ 720 A c 2 ( 6 c 2 ) c + 1 ] 1/ D 0 } . This proves (230) , including j > Q j>Q j > Q . We interpolate only the coefficient polynomial in n n n ; no finite-genus remainder is extrapolated to the actual cusp count. □
Fix the constants C 0 , D 0 C_0,D_0 C 0 , D 0 from this lemma and put B = 2 D 0 B=2D_0 B = 2 D 0 . They depend only on the fixed multiplier f f f .
Polynomial transfer at the additive threshold
The additive genus hypothesis of Theorem 4.8 is preserved when we pass from monomials to polynomial observables. We first specialize the orbital expansion, then estimate its terminal remainder and retained coefficients separately. We keep the first correction explicit because the next subsection will bound its detector response independently of degree.
Lemma 6.8 (Moments and integrability at the additive threshold). There is a fixed integer c 3 ≥ 2 c_{3} \ge2 c 3 ≥ 2 , depending only on f f f , such that, for integers q > t ≥ 1 q > t \ge1 q > t ≥ 1 and g ≥ c 3 s 2 + [ c 3 q ] c 3 s , g \ge c_{3}s^{2} + [c_{3}q]^{c_{3}}s, g ≥ c 3 s 2 + [ c 3 q ] c 3 s ,
∣ g − 1 E g , n O X ( x t ) − ∑ j = 1 q − 1 g − j ω j , n ( x t ) ∣ ≤ g − q g q [ c 3 q ] c 3 q . (239) \left|g^{-1}\mathbb{E}_{g,n}\mathcal{O}_{X}(x^{t})-\sum_{j=1}^{q-1}g^{-j}\omega_{j,n}(x^{t})\right|\le g^{-q}g^{q}[c_{3}q]^{c_{3}q}. \tag*{(239)} g − 1 E g , n O X ( x t ) − j = 1 ∑ q − 1 g − j ω j , n ( x t ) ≤ g − q g q [ c 3 q ] c 3 q . (239) Here ω j , n \omega_{j,n} ω j , n are the coefficients defined in (100) . Simultaneously for every Q ≥ 1 Q \ge1 Q ≥ 1 ,
E g , n τ X , Q ( x ) ≤ D f ( g + n Q ) , g ≥ c 3 s 2 + [ 2 c 3 ] c 3 s , (240) \mathbb{E}_{g,n}\tau_{X,Q}(x)\le D_{f}(g+nQ),\qquad g\ge c_{3}s^{2}+[2c_{3}]^{c_{3}}s, \tag*{(240)} E g , n τ X , Q ( x ) ≤ D f ( g + n Q ) , g ≥ c 3 s 2 + [ 2 c 3 ] c 3 s , (240) where D f D_{f} D f is the constant in (202) . At this threshold, the traces and orbital statistics associated with polynomials of degree at most Q Q Q and with zero constant term are absolutely integrable.
Proof. Apply Theorem 4.8 with R = q − 2 R=q-2 R = q − 2 , H = h ∗ t H=h^{*t} H = h ∗ t , and T = t L T=tL T = t L . Since ∥ h ∥ 1 = 1 \|h\|_{1}=1 ∥ h ∥ 1 = 1 , convolution gives ∥ h ∗ t ∥ ∞ ≤ ∥ h ∥ ∞ \|h^{*t}\|_{\infty}\le\|h\|_{\infty} ∥ h ∗ t ∥ ∞ ≤ ∥ h ∥ ∞ , and ρ = q − 1 + ⌈ t L ⌉ ≤ ( 1 + L ) q \rho=q-1+\lceil tL\rceil\le(1+L)q ρ = q − 1 + ⌈ t L ⌉ ≤ ( 1 + L ) q . Writing C ≥ 2 C\ge2 C ≥ 2 for its absolute constant, one may take
U = C ( 1 + L ) , H 0 = max { 1 , ∥ h ∥ ∞ } , U=C(1+L),\qquad H_{0}=\max\{1,\|h\|_{\infty}\}, U = C ( 1 + L ) , H 0 = max { 1 , ∥ h ∥ ∞ } , c 3 = ⌈ 3 + U + U log U + 3 L + log H 0 ⌉ . (241) c_{3}=\lceil3+U+U\log U+3L+\log H_{0}\rceil. \tag*{(241)} c 3 = ⌈ 3 + U + U log U + 3 L + log H 0 ⌉ . (241) Indeed, after division by g g g , the remaining error factor has logarithm at most q { U log q + U log U + 3 L + log H 0 } ≤ c 3 q log ( c 3 q ) q\{U\log q+U\log U+3L+\log H_{0}\}\le c_{3}q\log(c_{3}q) q { U log q + U log U + 3 L + log H 0 } ≤ c 3 q log ( c 3 q ) . Also c 3 ≥ C c_{3}\ge C c 3 ≥ C and [ C ρ ] C ≤ [ c 3 q ] c 3 [C\rho]^{C}\le[c_{3}q]^{c_{3}} [ C ρ ] C ≤ [ c 3 q ] c 3 , so the two summands of the threshold in Theorem 4.8 are dominated separately. The coefficient of s 2 s^{2} s 2 is independent of q q q .
The case ( q , t ) = ( 2 , 1 ) (q,t)=(2,1) ( q , t ) = ( 2 , 1 ) gives directly
∣ E g , n O X ( x ) − ω 1 ( x ) ∣ ≤ [ 2 c 3 ] 2 c 3 s 2 / g ≤ g , g ≥ c 3 s 2 + [ 2 c 3 ] c 3 s . (242) \left|\mathbb{E}_{g,n}\mathcal{O}_{X}(x)-\omega_{1}(x)\right|\le[2c_{3}]^{2c_{3}}s^{2}/g\le g,\qquad g\ge c_{3}s^{2}+[2c_{3}]^{c_{3}}s. \tag*{(242)} ∣ E g , n O X ( x ) − ω 1 ( x ) ∣ ≤ [ 2 c 3 ] 2 c 3 s 2 / g ≤ g , g ≥ c 3 s 2 + [ 2 c 3 ] c 3 s . (242) For the last inequality use g ≥ [ 2 c 3 ] c 3 s g\ge[2c_{3}]^{c_{3}}s g ≥ [ 2 c 3 ] c 3 s . The base orbital statistic is nonnegative and integrable. Thus the exact trace identity and deterministic bounds give
E τ X , Q ( x ) ≤ ( 4 π g + 2 π n ) I f + n b f Q + 6 a + g ≤ D f ( g + n Q ) . \mathbb{E}\tau_{X,Q}(x)\le(4\pi g+2\pi n)I_{f}+nb_{f}Q+6a+g\le D_{f}(g+nQ). E τ X , Q ( x ) ≤ ( 4 π g + 2 π n ) I f + n b f Q + 6 a + g ≤ D f ( g + n Q ) . Trace domination and the exact identity prove the remaining integrability statements. Thus the first-moment estimate holds under the additive threshold alone.
Proposition 6.9 (Polynomial transfer at the additive threshold). Put D 3 = max { D 0 , 3 c 3 } D_{3}=\max\{D_{0},3c_{3}\} D 3 = max { D 0 , 3 c 3 } and A Q = ( 3 C 0 Q ) D 0 A_{Q}=(3C_{0}Q)^{D_{0}} A Q = ( 3 C 0 Q ) D 0 . There is a fixed C a d d ≥ 2 C_{\mathrm{add}}\ge2 C add ≥ 2 , depending only on f f f , such that, for all integers n , Q ≥ 1 n,Q\ge1 n , Q ≥ 1 and g g g satisfying
g ≥ C a d d s 2 + C a d d Q D 3 s , (243) g\ge C_{\mathrm{add}}s^{2}+C_{\mathrm{add}}Q^{D_{3}}s, \tag*{(243)} g ≥ C add s 2 + C add Q D 3 s , (243) every real polynomial p = x p ~ p=x\widetilde p p = x p of degree at most Q Q Q satisfies
∣ E g , n O X ( p ) − ω 1 ( p ) − g − 1 ω 2 , n ( p ) ∣ ≤ { 2 s 3 A Q 3 g 2 + s 2 − Q g } N ∗ ( p ) . (244) \left|\mathbb{E}_{g,n}\mathcal{O}_{X}(p)-\omega_{1}(p)-g^{-1}\omega_{2,n}(p)\right|
\le\left\{\frac{2s^{3}A_{Q}^{3}}{g^{2}}+\frac{s^{2-Q}}{g}\right\}N_{*}(p). \tag*{(244)} E g , n O X ( p ) − ω 1 ( p ) − g − 1 ω 2 , n ( p ) ≤ { g 2 2 s 3 A Q 3 + g s 2 − Q } N ∗ ( p ) . (244) At the same threshold one also has
∣ E g , n O X ( p ) − ω 1 ( p ) ∣ ≤ C a d d s 2 Q B g N ∗ ( p ) . (245) \left|\mathbb{E}_{g,n}\mathcal{O}_{X}(p)-\omega_{1}(p)\right|
\le C_{\mathrm{add}}\frac{s^{2}Q^{B}}{g}N_{*}(p). \tag*{(245)} ∣ E g , n O X ( p ) − ω 1 ( p ) ∣ ≤ C add g s 2 Q B N ∗ ( p ) . (245)
Proof. Choose
C a d d = max { 2 , c 3 , ( 3 c 3 ) c 3 , 2 ( 3 C 0 ) D 0 , 14 ( 3 c 3 ) 3 c 3 , 3 ( 3 C 0 ) 2 D 0 } . (246) C_{\mathrm{add}}=\max\{2,c_3,(3c_3)^{c_3},2(3C_0)^{D_0},14(3c_3)^{3c_3},3(3C_0)^{2D_0}\}. \tag*{(246)} C add = max { 2 , c 3 , ( 3 c 3 ) c 3 , 2 ( 3 C 0 ) D 0 , 14 ( 3 c 3 ) 3 c 3 , 3 ( 3 C 0 ) 2 D 0 } . (246)
Set q = Q + 2 q=Q+2 q = Q + 2 , so q ≤ 3 Q q\le3Q q ≤ 3 Q and q > t q>t q > t for every monomial in p p p . The two summands of (243) imply
g ≥ c 3 s 2 + [ c 3 q ] c 3 s , g\ge c_3s^2+[c_3q]^{c_3}s, g ≥ c 3 s 2 + [ c 3 q ] c 3 s ,
g ≥ 2 s A Q , g ≥ 14 ( 3 c 3 Q ) 3 c 3 s . (247) g\ge2sA_Q,\qquad g\ge14(3c_3Q)^{3c_3}s. \tag*{(247)} g ≥ 2 s A Q , g ≥ 14 ( 3 c 3 Q ) 3 c 3 s . (247)
The last two comparisons use only the summand linear in s s s . They will respectively control the sum of retained coefficients and absorb the monomial coefficient bound in the terminal remainder. The same threshold also implies (240) . Linearity, Lemma 6.1, and Lemma 6.8 give the finite expansion
E O X ( p ) = ω 1 ( p ) + ∑ j = 2 Q + 1 g 1 − j ω j , n ( p ) + R , \mathbb{E}O_X(p)=\omega_1(p)+\sum_{j=2}^{Q+1}g^{1-j}\omega_{j,n}(p)+R, E O X ( p ) = ω 1 ( p ) + j = 2 ∑ Q + 1 g 1 − j ω j , n ( p ) + R ,
∣ R ∣ ≤ 7 Q g 1 − q s q [ c 3 q ] c 3 q N ∗ ( p ) . (248) |R|\le7^Qg^{1-q}s^q[c_3q]^{c_3q}N_*(p). \tag*{(248)} ∣ R ∣ ≤ 7 Q g 1 − q s q [ c 3 q ] c 3 q N ∗ ( p ) . (248)
We first estimate the terminal remainder. For N ∗ ( p ) > 0 N_*(p)>0 N ∗ ( p ) > 0 , divide its bound by s 2 N ∗ ( p ) / g s^2N_*(p)/g s 2 N ∗ ( p ) / g :
7 Q [ c 3 q ] c 3 q ( s / g ) q − 2 ≤ [ 7 ( 3 c 3 Q ) 3 c 3 s / g ] Q ≤ 2 − Q . (249) 7^Q[c_3q]^{c_3q}(s/g)^{q-2}\le[7(3c_3Q)^{3c_3}s/g]^Q\le2^{-Q}. \tag*{(249)} 7 Q [ c 3 q ] c 3 q ( s / g ) q − 2 ≤ [ 7 ( 3 c 3 Q ) 3 c 3 s / g ] Q ≤ 2 − Q . (249)
Thus ∣ R ∣ ≤ 2 − Q s 2 N ∗ ( p ) / g |R|\le2^{-Q}s^2N_*(p)/g ∣ R ∣ ≤ 2 − Q s 2 N ∗ ( p ) / g . If N ∗ ( p ) = 0 N_*(p)=0 N ∗ ( p ) = 0 , then p = 0 p=0 p = 0 .
We next sum the retained corrections of order j ≥ 3 j\ge3 j ≥ 3 . The coefficient bound in Lemma 6.7 holds without a condition on the genus. Since Q + j ≤ 3 Q Q+j\le3Q Q + j ≤ 3 Q for retained indices, it gives
∑ j = 3 Q + 1 g 1 − j ∣ ω j , n ( p ) ∣ ≤ s 3 A Q 3 g 2 N ∗ ( p ) ∑ j = 3 Q + 1 ( s A Q / g ) j − 3 ≤ 2 s 3 A Q 3 g 2 N ∗ ( p ) . \sum_{j=3}^{Q+1}g^{1-j}|\omega_{j,n}(p)|\le\frac{s^3A_Q^3}{g^2}N_*(p)\sum_{j=3}^{Q+1}(sA_Q/g)^{j-3}\le\frac{2s^3A_Q^3}{g^2}N_*(p). j = 3 ∑ Q + 1 g 1 − j ∣ ω j , n ( p ) ∣ ≤ g 2 s 3 A Q 3 N ∗ ( p ) j = 3 ∑ Q + 1 ( s A Q / g ) j − 3 ≤ g 2 2 s 3 A Q 3 N ∗ ( p ) .
For Q = 1 Q=1 Q = 1 this sum is empty and q = 3 q=3 q = 3 remains valid. This proves (244) . Summing instead from j = 2 j=2 j = 2 bounds all retained corrections by 2 s 2 A Q 2 g − 1 N ∗ ( p ) 2s^2A_Q^2g^{-1}N_*(p) 2 s 2 A Q 2 g − 1 N ∗ ( p ) ; together with (249) and (246) , this proves (245) .
Positive exceptional measures and the first correction
The coefficient bound from Section 6.6 grows with detector degree. We now remove this dependence for ω 2 , n ( p Q ) \omega_{2,n}(p_Q) ω 2 , n ( p Q ) by using the positive trace. We first extend the weighted coefficient to a distribution. After passing to its unweighted form, the additive polynomial estimate and a genus limit at fixed n n n show that its restriction to ( 1 , ∞ ) (1,\infty) ( 1 , ∞ ) is a positive measure supported in ( 1 , a ] (1,a] ( 1 , a ] . A cutoff then separates the detector into two parts: one controlled by the local mass of this measure, and one controlled by exponential decay below the detector transition.
Lemma 6.10 (Weighted distribution extensions). Put m = B + 2 = 2 D 0 + 2 m=B+2=2D_0+2 m = B + 2 = 2 D 0 + 2 . For each n ≥ 1 n\ge1 n ≥ 1 , the polynomial functional
K n ( q ) = ω 2 , n ( x q ) (250) \mathcal{K}_n(q)=\omega_{2,n}(x^q) \tag*{(250)} K n ( q ) = ω 2 , n ( x q ) (250) has a unique extension to a distribution on R \mathbb{R} R , supported in [ 0 , a ] [0,a] [ 0 , a ] , of order at most m m m . There is a fixed K f K_f K f such that
∣ K n ( φ ) ∣ ≤ K f ( n + 1 ) 2 ∥ φ ∥ C m ( [ 0 , a ] ) . (251) |\mathcal{K}_n(\varphi)|\le K_f(n+1)^2\|\varphi\|_{C^m([0,a])}. \tag*{(251)} ∣ K n ( φ ) ∣ ≤ K f ( n + 1 ) 2 ∥ φ ∥ C m ([ 0 , a ]) . (251) The coefficient W 2 ( p ) = Λ 2 ( H p ) W_2(p)=\Lambda_2(H_p) W 2 ( p ) = Λ 2 ( H p ) of ( n 2 ) \binom{n}{2} ( 2 n ) has the unique weighted extension
T 2 = K 3 − 2 K 2 + K 1 , T 2 ( q ) = W 2 ( x q ) , ∣ T 2 ( φ ) ∣ ≤ 38 K f ∥ φ ∥ C m ( [ 0 , a ] ) . (252) \mathcal{T}_2=\mathcal{K}_3-2\mathcal{K}_2+\mathcal{K}_1,\qquad\mathcal{T}_2(q)=W_2(x^q),\qquad|\mathcal{T}_2(\varphi)|\le38K_f\|\varphi\|_{C^m([0,a])}. \tag*{(252)} T 2 = K 3 − 2 K 2 + K 1 , T 2 ( q ) = W 2 ( x q ) , ∣ T 2 ( φ ) ∣ ≤ 38 K f ∥ φ ∥ C m ([ 0 , a ]) . (252)
Proof. The construction uses the rapid decay of smooth Chebyshev coefficients to absorb the degree growth in Lemma 6.7 (cf. [10 Proposition 3.2] ). With j = 2 j=2 j = 2 , that lemma gives
∣ K n ( q ) ∣ ≤ s 2 C 0 B ( deg q + 3 ) B ∥ q ∥ [ 0 , a ] . |\mathcal{K}_n(q)| \le s^2 C_0^B (\deg q+3)^B \lVert q\rVert_{[0,a]}. ∣ K n ( q ) ∣ ≤ s 2 C 0 B ( deg q + 3 ) B ∥ q ∥ [ 0 , a ] .
For a smooth φ \varphi φ , set F ( θ ) = φ ( a ( 1 + cos θ ) / 2 ) F(\theta)=\varphi(a(1+\cos\theta)/2) F ( θ ) = φ ( a ( 1 + cos θ ) /2 ) and expand
φ ( x ) = ∑ k ≥ 0 γ k T k ( 2 x / a − 1 ) . \varphi(x)=\sum_{k\geq0}\gamma_k T_k(2x/a-1). φ ( x ) = k ≥ 0 ∑ γ k T k ( 2 x / a − 1 ) .
The function F F F is even and periodic. Periodic integration by parts m m m times gives
∣ γ 0 ∣ ≤ ∥ φ ∥ ∞ , ∣ γ k ∣ ≤ 2 k − m ∥ F ( m ) ∥ ∞ ≤ 2 J a , m k − m ∥ φ ∥ C m ( k ≥ 1 ) , (253) |\gamma_0|\leq\lVert\varphi\rVert_\infty,\qquad|\gamma_k|\leq2k^{-m}\lVert F^{(m)}\rVert_\infty\leq2J_{a,m}k^{-m}\lVert\varphi\rVert_{C^m}\quad(k\geq1), \tag*{(253)} ∣ γ 0 ∣ ≤ ∥ φ ∥ ∞ , ∣ γ k ∣ ≤ 2 k − m ∥ F ( m ) ∥ ∞ ≤ 2 J a , m k − m ∥ φ ∥ C m ( k ≥ 1 ) , (253)
where ∥ φ ∥ C m = max 0 ≤ j ≤ m ∥ φ ( j ) ∥ ∞ \lVert\varphi\rVert_{C^m}=\max_{0\leq j\leq m}\lVert\varphi^{(j)}\rVert_\infty ∥ φ ∥ C m = max 0 ≤ j ≤ m ∥ φ ( j ) ∥ ∞ and one may take J a , m = ( m max { 1 , a / 2 } ) m J_{a,m}=(m\max\{1,a/2\})^m J a , m = ( m max { 1 , a /2 } ) m . Indeed, the chain rule in its partition form has at most m m m^m m m terms, and every derivative of a ( 1 + cos θ ) / 2 a(1+\cos\theta)/2 a ( 1 + cos θ ) /2 is bounded by a / 2 a/2 a /2 . Apply K n \mathcal{K}_n K n termwise to the series. Since m = B + 2 m=B+2 m = B + 2 and ( k + 3 ) B ≤ 4 B k B (k+3)^B\leq4^B k^B ( k + 3 ) B ≤ 4 B k B for k ≥ 1 k\geq1 k ≥ 1 , the resulting scalar series converges absolutely and satisfies (251) with
K f = C 0 B { 3 B + 2 J a , m 4 B ζ ( 2 ) } . K_f=C_0^B\{3^B+2J_{a,m}4^B\zeta(2)\}. K f = C 0 B { 3 B + 2 J a , m 4 B ζ ( 2 )} .
For polynomial tests the series is finite, so it agrees with (250) . Restriction of smooth tests to [ 0 , a ] [0,a] [ 0 , a ] defines the asserted compactly supported distribution.
For uniqueness, a compactly supported finite-order distribution is determined by its polynomial values. To verify this directly, let r r r be its order and approximate the r r r th derivative of a smooth test uniformly by polynomials on a neighborhood of its support. Integrating r r r times and matching the lower derivatives at one endpoint gives polynomial approximation in C r C^r C r . A fixed cutoff equal to one near the support then proves that a distribution vanishing on all polynomials vanishes on every test. Finally, the second difference in n n n of (113) gives W 2 = ω 2 , 3 − 2 ω 2 , 2 + ω 2 , 1 W_2=\omega_{2,3}-2\omega_{2,2}+\omega_{2,1} W 2 = ω 2 , 3 − 2 ω 2 , 2 + ω 2 , 1 . Its norm factor is 4 2 + 2 ⋅ 3 2 + 2 2 = 38 4^2+2\cdot3^2+2^2=38 4 2 + 2 ⋅ 3 2 + 2 2 = 38 , proving (252) .
We use the canonical unweighted extensions
V n ( ψ ) = K n ( ψ ( x ) − ψ ( 0 ) x ) . (254) \mathcal{V}_n(\psi)=\mathcal{K}_n\left(\frac{\psi(x)-\psi(0)}{x}\right). \tag*{(254)} V n ( ψ ) = K n ( x ψ ( x ) − ψ ( 0 ) ) . (254)
The quotient has value ψ ′ ( 0 ) \psi'(0) ψ ′ ( 0 ) at zero and equals ∫ 0 1 ψ ′ ( t x ) d t \int_0^1\psi'(tx)\,\mathrm{d}t ∫ 0 1 ψ ′ ( t x ) d t . Thus V n \mathcal{V}_n V n has order at most m + 1 m+1 m + 1 , support in [ 0 , a ] [0,a] [ 0 , a ] , and total mass zero in the distributional sense. It agrees with ω 2 , n \omega_{2,n} ω 2 , n on polynomials vanishing at zero. Only the unweighted extension requires this normalization: without it, such polynomial data leave a multiple of δ 0 \delta_0 δ 0 undetermined.
Theorem 6.11 (Positive exceptional measures). For every n ≥ 1 n\geq1 n ≥ 1 , the restriction of V n \mathcal{V}_n V n to ( 1 , ∞ ) (1,\infty) ( 1 , ∞ ) is a positive Radon measure μ n \mu_n μ n supported in ( 1 , a ] (1,a] ( 1 , a ] . For every fixed 1 < b ≤ a 1<b\leq a 1 < b ≤ a ,
μ n ( [ b , a ] ) ≤ C f , b ( n + 1 ) 2 . (255) \mu_n([b,a])\leq C_{f,b}(n+1)^2. \tag*{(255)} μ n ([ b , a ]) ≤ C f , b ( n + 1 ) 2 . (255) There is also a positive Radon measure η 2 \eta_2 η 2 on ( 1 , a ] (1,a] ( 1 , a ] such that
T 2 ( ψ / x ) = ∫ ψ d η 2 , η 2 ( [ b , a ] ) ≤ C f , b , (256) \mathcal{T}_2(\psi/x)=\int\psi\,\mathrm{d}\eta_2,\qquad\eta_2([b,a])\leq C_{f,b}, \tag*{(256)} T 2 ( ψ / x ) = ∫ ψ d η 2 , η 2 ([ b , a ]) ≤ C f , b , (256) for tests ψ ∈ C c ∞ ( ( 1 , ∞ ) ) \psi\in C^\infty_c((1,\infty)) ψ ∈ C c ∞ (( 1 , ∞ )) . These tests may be nonzero at a a a . In particular the statement includes possible endpoint mass. Moreover 2 n − 2 μ n 2n^{-2}\mu_n 2 n − 2 μ n converges vaguely to η 2 \eta_2 η 2 on ( 1 , a ] (1,a] ( 1 , a ] as n → ∞ n\to\infty n → ∞ through the integers.
Proof. Fix n n n and let 0 ≤ ψ ∈ C c ∞ ( ( 1 , ∞ ) ) 0 \le\psi\in C^\infty_c((1,\infty)) 0 ≤ ψ ∈ C c ∞ (( 1 , ∞ )) . Put ξ ( x ) = ψ ( x ) / x \xi(x)=\psi(x)/x ξ ( x ) = ψ ( x ) / x , extended by zero near zero. Choose
0 < σ < min { 1 / D 3 , 1 / ( 3 D 0 ) } , Q = ⌊ g σ ⌋ . (257) 0 < \sigma< \min\{1/D_3,1/(3D_0)\}, \qquad Q=\lfloor g^\sigma\rfloor. \tag*{(257)} 0 < σ < min { 1/ D 3 , 1/ ( 3 D 0 )} , Q = ⌊ g σ ⌋ . (257)
The restriction σ < 1 / D 3 \sigma<1/D_3 σ < 1/ D 3 ensures that the threshold in Proposition 6.9 holds eventually at this fixed n n n , since Q D 3 / g → 0 QD_3/g \to0 Q D 3 / g → 0 . The restriction σ < 1 / ( 3 D 0 ) \sigma<1/(3D_0) σ < 1/ ( 3 D 0 ) will make the remainder vanish after multiplication by g g g . Let r Q r_Q r Q be the shifted-Chebyshev truncation of ξ \xi ξ of degree at most Q − 1 Q-1 Q − 1 , and set v Q ( x ) = x r Q ( x ) v_Q(x)=xr_Q(x) v Q ( x ) = x r Q ( x ) . Smoothness and (253) , used with arbitrarily high order, give for every fixed M M M
δ Q : = ∥ r Q − ξ ∥ [ 0 , a ] ≤ C ψ , M Q − M , r Q ⟶ ξ in C m ( [ 0 , a ] ) . (258) \delta_Q:=\lVert r_Q-\xi\rVert_{[0,a]} \le C_{\psi,M}Q^{-M}, \qquad r_Q \longrightarrow\xi\text{ in } C^m([0,a]). \tag*{(258)} δ Q := ∥ r Q − ξ ∥ [ 0 , a ] ≤ C ψ , M Q − M , r Q ⟶ ξ in C m ([ 0 , a ]) . (258)
For the second assertion, the iterated Markov inequality bounds the first m m m derivatives of the k k k th Chebyshev term by a fixed multiple of k 2 m k^{2m} k 2 m , which is absorbed by the rapid coefficient decay. Consequently N ∗ ( v Q ) N_*(v_Q) N ∗ ( v Q ) is bounded, N R ( v Q ) ≤ δ Q N_{\mathrm{R}}(v_Q)\le\delta_Q N R ( v Q ) ≤ δ Q , and ω 2 , n ( v Q ) → Y n ( ψ ) \omega_{2,n}(v_Q)\to\mathcal{Y}_n(\psi) ω 2 , n ( v Q ) → Y n ( ψ ) .
Let P 0 P_0 P 0 project onto constants. The smooth compressed trace
D X , Q ( ψ ) = Tr χ Y Q ψ ( A X ) ( I − P 0 ) χ Y Q ≥ 0 (259) D_{X,Q}(\psi)=\operatorname{Tr}\chi_{Y_Q}\psi(A_X)(I-P_0)\chi_{Y_Q}\ge0 \tag*{(259)} D X , Q ( ψ ) = Tr χ Y Q ψ ( A X ) ( I − P 0 ) χ Y Q ≥ 0 (259)
is well defined and integrable: on [ 0 , a ] [0,a] [ 0 , a ] one has 0 ≤ ψ ( x ) ≤ C ψ x 0\le\psi(x)\le C_\psi x 0 ≤ ψ ( x ) ≤ C ψ x , so the first moment of Lemma 6.8 applies. Spectral domination also gives
∣ E D X , Q ( ψ ) − E D X , Q ( v Q ) ∣ ≤ δ Q E τ X , Q ( x ) ≤ D f δ Q ( g + n Q ) . (260) \left|\mathbb{E}D_{X,Q}(\psi)-\mathbb{E}D_{X,Q}(v_Q)\right|\le\delta_Q\mathbb{E}\tau_{X,Q}(x)\le D_f\delta_Q(g+nQ). \tag*{(260)} ∣ E D X , Q ( ψ ) − E D X , Q ( v Q ) ∣ ≤ δ Q E τ X , Q ( x ) ≤ D f δ Q ( g + n Q ) . (260)
This comparison does not require v Q ≥ 0 v_Q\ge0 v Q ≥ 0 on [ 0 , a ] [0,a] [ 0 , a ] . For each fixed cutoff, polynomial compressions are trace class by (181) and measurable by integration of their periodized kernels over the truncated surface. Weighted polynomial approximation makes the smooth trace their pointwise limit, proving its measurability. The exact geometric trace identity is used below only for v Q v_Q v Q , whose degree is at most Q Q Q .
The exact polynomial trace identity and the unit exceptional atom in ω 1 \omega_1 ω 1 give
g E D X , Q ( v Q ) = g ( A g , n − n / Y Q ) I ( v Q ) + g n P Y Q ( v Q ) + g μ 0 ( v Q ) + g n v Q ( a ) A g , n Y Q + ω 2 , n ( v Q ) + g R g , n , Q . (261) \begin{aligned}
g\mathbb{E}D_{X,Q}(v_Q)={}&g(A_{g,n}-n/Y_Q)\mathcal{I}(v_Q)+gnP_{Y_Q}(v_Q)+g\mu_0(v_Q)\\
&+\frac{gnv_Q(a)}{A_{g,n}Y_Q}+\omega_{2,n}(v_Q)+gR_{g,n,Q}. \tag*{(261)}
\end{aligned} g E D X , Q ( v Q ) = g ( A g , n − n / Y Q ) I ( v Q ) + g n P Y Q ( v Q ) + g μ 0 ( v Q ) + A g , n Y Q g n v Q ( a ) + ω 2 , n ( v Q ) + g R g , n , Q . (261)
By (200) , the first three terms have absolute value at most C f , n ( g 2 + g n Q 2 + g ) δ Q C_{f,n}(g^2+gnQ^2+g)\delta_Q C f , n ( g 2 + g n Q 2 + g ) δ Q . The fourth tends to zero because v Q ( a ) v_Q(a) v Q ( a ) and g / A g , n g/A_{g,n} g / A g , n are bounded and Y Q = 3 e L Q Y_Q=3e^{L_Q} Y Q = 3 e L Q . Equation (244) gives
∣ g R g , n , Q ∣ ≤ C f , n , ψ { Q 3 D 0 / g + 2 − Q } ⟶ 0. \left|gR_{g,n,Q}\right|\le C_{f,n,\psi}\{Q^{3D_0}/g+2^{-Q}\}\longrightarrow0. ∣ g R g , n , Q ∣ ≤ C f , n , ψ { Q 3 D 0 / g + 2 − Q } ⟶ 0.
Choose M M M in (258) large enough that g 2 δ Q → 0 g^2\delta_Q\to0 g 2 δ Q → 0 . Since σ < 1 / D 3 ≤ 1 / 2 \sigma<1/D_3\le1/2 σ < 1/ D 3 ≤ 1/2 and n n n is fixed, all the other approximation errors, including (260) multiplied by g g g , tend to zero as well. We conclude that
g E g , n D X , Q ( ψ ) ⟶ Y n ( ψ ) ≥ 0. (262) g\mathbb{E}_{g,n}D_{X,Q}(\psi)\longrightarrow\mathcal{Y}_n(\psi)\ge0. \tag*{(262)} g E g , n D X , Q ( ψ ) ⟶ Y n ( ψ ) ≥ 0. (262)
The constant mode was removed before taking the limit, so this argument does not require ψ ( a ) = 0 \psi(a)=0 ψ ( a ) = 0 .
A positive distribution is a locally finite positive measure. Locally, choose a nonnegative smooth cutoff ζ = 1 \zeta=1 ζ = 1 on a compact set K K K ; positivity bounds its action on tests supported in K K K by ∥ φ ∥ ∞ Y n ( ζ ) \lVert\varphi\rVert_\infty\mathcal{Y}_n(\zeta) ∥ φ ∥ ∞ Y n ( ζ ) . Uniform approximation and the representation theorem for positive functionals give the measure. This proves the assertion for μ n \mu_n μ n on ( 1 , ∞ ) (1,\infty) ( 1 , ∞ ) ; its support follows from that of Y n \mathcal{Y}_n Y n . To bound its mass, choose ζ b ∈ C c ∞ ( ( 1 , ∞ ) ) \zeta_b\in C^\infty_c((1,\infty)) ζ b ∈ C c ∞ (( 1 , ∞ )) with 0 ≤ ζ b ≤ 1 0\le\zeta_b\le1 0 ≤ ζ b ≤ 1 , zero below ( 1 + b ) / 2 (1+b)/2 ( 1 + b ) /2 and one on [ b , a ] [b,a] [ b , a ] . Then
μ n ( [ b , a ] ) ≤ Y n ( ζ b ) = K n ( ζ b / x ) ≤ K f s 2 ∥ ζ b / x ∥ C m ( [ 0 , a ] ) . (263) \mu_n([b,a])\le\mathcal{Y}_n(\zeta_b)=\mathcal{K}_n(\zeta_b/x)\le K_fs^2\lVert\zeta_b/x\rVert_{C^m([0,a])}. \tag*{(263)} μ n ([ b , a ]) ≤ Y n ( ζ b ) = K n ( ζ b / x ) ≤ K f s 2 ∥ ζ b / x ∥ C m ([ 0 , a ]) . (263)
This proves (255) , uniformly in n n n .
The quadratic identity (113) extends from polynomials to smooth tests through Lemma 6.10. Divide that identity by ( n 2 ) \binom{n}{2} ( 2 n ) and let n → ∞ n \to\infty n → ∞ through the integers. Since V n ( ψ ) ≥ 0 \mathcal{V}_n(\psi) \ge0 V n ( ψ ) ≥ 0 , the limit T 2 ( ψ / x ) \mathcal{T}_2(\psi/x) T 2 ( ψ / x ) is nonnegative. The same positive-functional argument proves (256) ; its local norm bound follows either from (252) or from (263) after multiplication by 2 / n 2 2/n^2 2/ n 2 . Cusp polynomiality gives convergence on smooth compactly supported tests, and the uniform local mass bound extends it to continuous compactly supported tests, proving vague convergence. In (262) , the genus limit is taken at fixed n n n , before the subsequent limit n → ∞ n \to\infty n → ∞ .
Proposition 6.12 (Degree-independent detector response). For every fixed 0 < ε < 1 / 4 0 < \varepsilon< 1/4 0 < ε < 1/4 , there is C f , ε C_{f,\varepsilon} C f , ε such that, for all n ≥ 1 n \ge1 n ≥ 1 and Q ≥ Q ε Q \ge Q_\varepsilon Q ≥ Q ε ,
∣ ω 2 , n ( p Q ) ∣ ≤ C f , ε ( n + 1 ) 2 , ∣ W 2 ( p Q ) ∣ ≤ C f , ε . (264) |\omega_{2,n}(p_Q)| \le C_{f,\varepsilon}(n+1)^2,\qquad|W_2(p_Q)| \le C_{f,\varepsilon}. \tag*{(264)} ∣ ω 2 , n ( p Q ) ∣ ≤ C f , ε ( n + 1 ) 2 , ∣ W 2 ( p Q ) ∣ ≤ C f , ε . (264)
Proof. Recall the detector transition point a θ = ( 1 + b ε ) / 2 > 1 a_\theta=(1+b_\varepsilon)/2>1 a θ = ( 1 + b ε ) /2 > 1 . Fix 1 < r 0 < r 1 < a θ 1<r_0<r_1<a_\theta 1 < r 0 < r 1 < a θ and a smooth 0 ≤ χ ≤ 1 0\le\chi\le1 0 ≤ χ ≤ 1 which is zero below r 0 r_0 r 0 and one above r 1 r_1 r 1 . For a weighted distribution T \mathcal{T} T supported in [ 0 , a ] [0,a] [ 0 , a ] , of order at most m m m , whose unweighted exceptional restriction is a locally finite measure ν \nu ν , one has
T ( p Q / x ) = ∫ χ p Q d ν + T ( ( 1 − χ ) p Q / x ) . (265) \mathcal{T}(p_Q/x)=\int\chi p_Q\,\mathrm{d}\nu+\mathcal{T}((1-\chi)p_Q/x). \tag*{(265)} T ( p Q / x ) = ∫ χ p Q d ν + T (( 1 − χ ) p Q / x ) . (265)
Since 0 ≤ p Q ≤ 1 0\le p_Q\le1 0 ≤ p Q ≤ 1 , the first term has absolute value at most ∣ ν ∣ ( [ r 0 , a ] ) |\nu|([r_0,a]) ∣ ν ∣ ([ r 0 , a ]) .
Set δ = θ − r 1 / a > 0 \delta=\theta-r_1/a>0 δ = θ − r 1 / a > 0 and Q 0 = max { Q ε , 2 , 1 + ⌈ 2 ( 1 − θ ) / δ ⌉ } Q_0=\max\{Q_\varepsilon,2,1+\lceil2(1-\theta)/\delta\rceil\} Q 0 = max { Q ε , 2 , 1 + ⌈ 2 ( 1 − θ ) / δ ⌉} . For Q ≥ Q 0 Q\ge Q_0 Q ≥ Q 0 and 0 < x ≤ r 1 0<x\le r_1 0 < x ≤ r 1 , the size-bias bound (217) and the binomial concentration bound give
p Q ( x ) x ≤ 1 a θ P { Bin ( Q − 1 , x / a ) ≥ k Q − 1 } ≤ B ε e − δ 2 Q / 4 . (266) \frac{p_Q(x)}{x}\le\frac{1}{a\theta}\mathbb{P}\{\operatorname{Bin}(Q-1,x/a)\ge k_Q-1\}\le B_\varepsilon e^{-\delta^2Q/4}. \tag*{(266)} x p Q ( x ) ≤ a θ 1 P { Bin ( Q − 1 , x / a ) ≥ k Q − 1 } ≤ B ε e − δ 2 Q /4 . (266)
Indeed, ( k Q − 1 ) / ( Q − 1 ) ≥ θ − ( 1 − θ ) / ( Q − 1 ) ≥ x / a + δ / 2 (k_Q-1)/(Q-1)\ge\theta-(1-\theta)/(Q-1)\ge x/a+\delta/2 ( k Q − 1 ) / ( Q − 1 ) ≥ θ − ( 1 − θ ) / ( Q − 1 ) ≥ x / a + δ /2 . The quotient at zero is defined by continuity. It is a polynomial of degree at most Q − 1 Q-1 Q − 1 , so the iterated Markov inequality on [ 0 , r 1 ] [0,r_1] [ 0 , r 1 ] gives
∥ ( p Q / x ) ( j ) ∥ [ 0 , r 1 ] ≤ ( 2 Q 2 / r 1 ) j B ε e − δ 2 Q / 4 , 0 ≤ j ≤ m . \left\|\left(p_Q/x\right)^{(j)}\right\|_{[0,r_1]}\le(2Q^2/r_1)^jB_\varepsilon e^{-\delta^2Q/4},\qquad0\le j\le m. ( p Q / x ) ( j ) [ 0 , r 1 ] ≤ ( 2 Q 2 / r 1 ) j B ε e − δ 2 Q /4 , 0 ≤ j ≤ m .
Derivatives above the actual degree vanish. Leibniz’s rule and the fixed cutoff therefore show that, if T \mathcal{T} T has C m C^m C m norm at most K K K , the second term in (265) is bounded by C f , ε K Q 2 m e − δ 2 Q / 4 C_{f,\varepsilon}KQ^{2m}e^{-\delta^2Q/4} C f , ε K Q 2 m e − δ 2 Q /4 . Only values and derivatives on [ 0 , r 1 ] [0,r_1] [ 0 , r 1 ] are used because 1 − χ 1-\chi 1 − χ vanishes above r 1 r_1 r 1 .
Apply this with ( T , ν ) = ( K n , μ n ) (\mathcal{T},\nu)=(\mathcal{K}_n,\mu_n) ( T , ν ) = ( K n , μ n ) and then ( T 2 , η 2 ) (\mathcal{T}_2,\eta_2) ( T 2 , η 2 ) , using the distribution and local mass bounds. The function Q 2 m e − δ 2 Q / 4 Q^{2m}e^{-\delta^2Q/4} Q 2 m e − δ 2 Q /4 is bounded, independently of Q Q Q . For the finitely many degrees Q ε ≤ Q < Q 0 Q_\varepsilon\le Q<Q_0 Q ε ≤ Q < Q 0 , use (230) , (252) , and N ∗ ( p Q ) ≤ B ε N_*(p_Q)\le B_\varepsilon N ∗ ( p Q ) ≤ B ε . This proves both estimates. No finite total mass near the endpoint 1 1 1 is assumed.
The spectral count in the full sub-square-root range
We now combine the additive polynomial remainder with the degree-independent first-correction bound. At logarithmic detector degree, the remaining errors are bounded by a constant times ( n + 1 ) 2 / g (n+1)^2/g ( n + 1 ) 2 / g for all sufficiently large g g g in the additive domain. We then verify that domain uniformly under a quadratic condition on the cusp count.
Theorem 6.13 (Spectral count throughout the sub-square-root range). For each fixed 0 < ε < 1 / 4 0 < \varepsilon< 1/4 0 < ε < 1/4 there is a constant C f , ε C_{f,\varepsilon} C f , ε such that, for integers g ≥ 3 g \ge3 g ≥ 3 , n ≥ 1 n \ge1 n ≥ 1 , and Q ≥ Q ε Q \ge Q_{\varepsilon} Q ≥ Q ε satisfying (243) ,
P g , n ( N ε ≥ 1 ) ≤ E g , n N ε ≤ C f , ε { ( g + n Q 2 ) e − κ Q + s 2 g + s 3 Q 3 D 0 g 2 + s 2 g 2 − Q + n A g , n Y Q } . (267) \mathbb{P}_{g,n}(N_{\varepsilon} \ge1) \le\mathbb{E}_{g,n}N_{\varepsilon} \le C_{f,\varepsilon}\left\{(g+nQ^{2})e^{-\kappa Q}+\frac{s^{2}}{g}+\frac{s^{3}Q^{3D_{0}}}{g^{2}}+\frac{s^{2}}{g}2^{-Q}+\frac{n}{A_{g,n}Y_{Q}}\right\}. \tag*{(267)} P g , n ( N ε ≥ 1 ) ≤ E g , n N ε ≤ C f , ε { ( g + n Q 2 ) e − κ Q + g s 2 + g 2 s 3 Q 3 D 0 + g s 2 2 − Q + A g , n Y Q n } . (267) After increasing C f , ε C_{f,\varepsilon} C f , ε , there is a fixed g ε g_{\varepsilon} g ε , independent of n n n , such that
g ≥ g ε , g ≥ C f , ε ( n + 1 ) 2 (268) g \ge g_{\varepsilon}, \qquad g \ge C_{f,\varepsilon}(n+1)^{2} \tag*{(268)} g ≥ g ε , g ≥ C f , ε ( n + 1 ) 2 (268) imply
P g , n { SpG ( X ) ≤ 1 / 4 − ε } ≤ E g , n N ε ≤ C f , ε ( n + 1 ) 2 g . (269) \mathbb{P}_{g,n}\{\operatorname{SpG}(X) \le1/4-\varepsilon\} \le\mathbb{E}_{g,n}N_{\varepsilon} \le C_{f,\varepsilon}\frac{(n+1)^{2}}{g}. \tag*{(269)} P g , n { SpG ( X ) ≤ 1/4 − ε } ≤ E g , n N ε ≤ C f , ε g ( n + 1 ) 2 . (269) Consequently, for every integer-valued sequence n ( g ) ≥ 1 n(g) \ge1 n ( g ) ≥ 1 with n ( g ) = o ( g ) n(g)=o(\sqrt{g}) n ( g ) = o ( g ) , the expected count and this probability tend to zero. For any envelope n max ( g ) = o ( g ) n_{\max}(g)=o(\sqrt{g}) n m a x ( g ) = o ( g ) ,
sup 1 ≤ n ≤ n max ( g ) E g , n N ε ≤ C f , ε ( n max ( g ) + 1 ) 2 g ⟶ 0 (270) \sup_{1\le n\le n_{\max}(g)}\mathbb{E}_{g,n}N_{\varepsilon} \le C_{f,\varepsilon}\frac{(n_{\max}(g)+1)^{2}}{g}\longrightarrow0 \tag*{(270)} 1 ≤ n ≤ n m a x ( g ) sup E g , n N ε ≤ C f , ε g ( n m a x ( g ) + 1 ) 2 ⟶ 0 (270) whenever the range is nonempty. All counts include residual eigenvalues with multiplicity.
Proof. We first prove the estimate at fixed degree. Proposition 6.12 bounds the canonical first correction without a genus hypothesis, so it applies throughout the additive domain. Thus ∣ ω 2 , n ( p Q ) ∣ ≤ C f , ε s 2 |\omega_{2,n}(p_{Q})| \le C_{f,\varepsilon}s^{2} ∣ ω 2 , n ( p Q ) ∣ ≤ C f , ε s 2 . Every expectation at the additive threshold is justified by Lemma 6.8.
The exact trace identity, the unit exceptional atom of ω 1 \omega_{1} ω 1 , and (244) give
E D X , Q = ( A g , n − n Y Q ) I ( p Q ) + n P Y Q ( p Q ) + μ 0 ( p Q ) + A g , n Y Q n ω 2 , n ( p Q ) + R . (271) \mathbb{E}D_{X,Q}=\left(A_{g,n}-\frac{n}{Y_{Q}}\right)\mathcal{I}(p_{Q})+nP_{Y_{Q}}(p_{Q})+\mu_{0}(p_{Q})+\frac{A_{g,n}Y_{Q}}{n}\omega_{2,n}(p_{Q})+R. \tag*{(271)} E D X , Q = ( A g , n − Y Q n ) I ( p Q ) + n P Y Q ( p Q ) + μ 0 ( p Q ) + n A g , n Y Q ω 2 , n ( p Q ) + R . (271)
The detector estimates, deterministic trace bounds, and N ε ≤ 4 D X , Q N_{\varepsilon}\le4D_{X,Q} N ε ≤ 4 D X , Q prove (267) . The residual modes and their multiplicities are included by Lemma 6.5. The trace D X , Q D_{X,Q} D X , Q remains nonnegative; the first correction enters only when its expectation is expanded.
Use K K K and Q g Q_{g} Q g from (227) . For all sufficiently large g g g , independently of n n n , the bounds in Section 6.5 give Q g ≥ Q ε Q_{g}\ge Q_{\varepsilon} Q g ≥ Q ε , log g ≥ 1 \log g\ge1 log g ≥ 1 , and Q g ≤ 2 K log g Q_{g}\le2K\log g Q g ≤ 2 K log g . The threshold (243) holds if
g ≥ C a d d s 2 + C a d d ( 2 K log g ) D 3 s . (272) g\ge C_{\mathrm{add}}s^{2}+C_{\mathrm{add}}(2K\log g)^{D_{3}}s. \tag*{(272)} g ≥ C add s 2 + C add ( 2 K log g ) D 3 s . (272)
In this domain s ≤ g / C a d d s\le\sqrt{g/C_{\mathrm{add}}} s ≤ g / C add , so the higher-order term relative to s 2 / g s^{2}/g s 2 / g satisfies
s Q g 3 D 0 g ≤ Q g 3 D 0 C a d d g ⟶ 0 (273) \frac{sQ_{g}^{3D_{0}}}{g}\le\frac{Q_{g}^{3D_{0}}}{\sqrt{C_{\mathrm{add}}g}}\longrightarrow0 \tag*{(273)} g s Q g 3 D 0 ≤ C add g Q g 3 D 0 ⟶ 0 (273)
uniformly over the allowed cusp counts. The exponential term and the truncation correction n / ( A g , n Y Q g ) n/(A_{g,n}Y_{Q_{g}}) n / ( A g , n Y Q g ) are O f , ε ( g − 2 ) O_{f,\varepsilon}(g^{-2}) O f , ε ( g − 2 ) : use n ≤ g n\le g n ≤ g , A g , n ≥ 2 π n A_{g,n}\ge2\pi n A g , n ≥ 2 π n , and (228) –(229) . Since s ≥ 2 s\ge2 s ≥ 2 , these terms are bounded by a constant times s 2 / g s^{2}/g s 2 / g ; the term with 2 − Q g ≤ 1 2^{-Q_{g}}\le1 2 − Q g ≤ 1 satisfies the same bound. This proves (269) under (272) , after imposing a fixed lower genus bound.
It remains to show that an eventual quadratic condition implies (272) . Choose g ε g_{\varepsilon} g ε large enough that all the preceding requirements hold and
( 2 K log g ) D 3 ≤ g 2 C a d d ( g ≥ g ε ) . (274) (2K\log g)^{D_{3}}\le\sqrt{\frac{g}{2C_{\mathrm{add}}}}\qquad(g\ge g_{\varepsilon}). \tag*{(274)} ( 2 K log g ) D 3 ≤ 2 C add g ( g ≥ g ε ) . (274)
If g ≥ 2 C a d d s 2 g \ge2C_{\mathrm{add}}s^2 g ≥ 2 C add s 2 , then C a d d s 2 ≤ g / 2 C_{\mathrm{add}}s^2 \le g/2 C add s 2 ≤ g /2 and
C a d d ( 2 K log g ) D 3 s ≤ C a d d g / ( 2 C a d d ) g / ( 2 C a d d ) = g / 2. C_{\mathrm{add}}(2K\log g)D_3s \le C_{\mathrm{add}}\sqrt{g/(2C_{\mathrm{add}})}\sqrt{g/(2C_{\mathrm{add}})}=g/2. C add ( 2 K log g ) D 3 s ≤ C add g / ( 2 C add ) g / ( 2 C add ) = g /2.
Thus (272) follows, uniformly in n n n . Increase C f , ε C_{f,\varepsilon} C f , ε to cover both 2 C a d d 2C_{\mathrm{add}} 2 C add and the count constant. This proves (268) –(269) . The probability bound uses the spectral-gap event identity and 1 { N ε ≥ 1 } ≤ N ε \mathbf{1}_{\{N_\varepsilon\ge1\}}\le N_\varepsilon 1 { N ε ≥ 1 } ≤ N ε .
Finally, ( n max ( g ) + 1 ) 2 / g → 0 (n_{\max}(g)+1)^2/g\to0 ( n m a x ( g ) + 1 ) 2 / g → 0 makes the quadratic domain hold eventually for every 1 ≤ n ≤ n max ( g ) 1\le n\le n_{\max}(g) 1 ≤ n ≤ n m a x ( g ) . Substituting this upper bound for s 2 s^2 s 2 proves (270) and the asserted sequence statement.
Sharpness of the count bound and cusp range
Short separating curves give a matching lower bound. We use Mirzakhani’s integration formula and uniform volume comparisons. The count below for pairs of pants with two cusps also appears in [21 §4.2 and the proof of Proposition 35] . The torus-cap argument for n = 1 n=1 n = 1 is the once-cusped analogue of the closed-surface lower-bound argument of Anantharaman–Monk [1 §9.1.1] .
Proposition 6.14 (Uniform lower bound). For every fixed 0 < ε < 1 / 4 0<\varepsilon<1/4 0 < ε < 1/4 there are constants c ε > 0 c_\varepsilon>0 c ε > 0 and g ε l o w ≥ 3 g_\varepsilon^{\mathrm{low}}\ge3 g ε low ≥ 3 , independent of n n n , such that, for all integers g ≥ g ε l o w g\ge g_\varepsilon^{\mathrm{low}} g ≥ g ε low and n ≥ 1 n\ge1 n ≥ 1 ,
P g , n ( N ε ≥ 1 ) ≥ c ε min { ( n + 1 ) 2 g , 1 } . (275) \mathbb{P}_{g,n}(N_\varepsilon\ge1)\ge c_\varepsilon\min\left\{\frac{(n+1)^2}{g},1\right\}. \tag*{(275)} P g , n ( N ε ≥ 1 ) ≥ c ε min { g ( n + 1 ) 2 , 1 } . (275)
Proof. Put τ ε = 1 / 4 − ε \tau_\varepsilon=1/4-\varepsilon τ ε = 1/4 − ε and choose
ℓ ε = min { 1 2 , π τ ε 4 sinh 1 } . \ell_\varepsilon=\min\left\{\frac{1}{2},\frac{\pi\tau_\varepsilon}{4\sinh1}\right\}. ℓ ε = min { 2 1 , 4 sinh 1 π τ ε } .
All curve counts in this proof are unoriented and count each geodesic once.
A short cap gives a small eigenvalue. Suppose a simple closed geodesic of length ℓ ≤ ℓ ε \ell\le\ell_\varepsilon ℓ ≤ ℓ ε bounds either a pair of pants with two cusps or a one-holed torus with no cusps. The cap has area 2 π 2\pi 2 π . The collar lemma gives an embedded collar of half-width arsinh ( 1 / sinh ( ℓ / 2 ) ) > 1 \operatorname{arsinh}(1/\sinh(\ell/2))>1 arsinh ( 1/ sinh ( ℓ /2 )) > 1 , with metric d r 2 + ℓ 2 cosh 2 r d θ 2 dr^2+\ell^2\cosh^2r\,d\theta^2 d r 2 + ℓ 2 cosh 2 r d θ 2 , where θ ∈ R / Z \theta\in\mathbb{R}/\mathbb{Z} θ ∈ R / Z ; see [21 proof of Lemma 34] . Let u u u equal one on the cap, 1 − r 1-r 1 − r on its exterior collar 0 ≤ r ≤ 1 0\le r\le1 0 ≤ r ≤ 1 , and zero elsewhere. This is an H 1 ( X ) H^1(X) H 1 ( X ) function, also when the cap contains cusps, since the cap has finite area. Direct integration gives
∫ X ∣ ∇ u ∣ 2 = ℓ sinh 1 , ∫ X u 2 ≥ 2 π , ∫ X u ≤ 2 π + ℓ sinh 1 ≤ 3 π . (276) \int_X|\nabla u|^2=\ell\sinh1,\qquad\int_Xu^2\ge2\pi,\qquad\int_Xu\le2\pi+\ell\sinh1\le3\pi. \tag*{(276)} ∫ X ∣∇ u ∣ 2 = ℓ sinh 1 , ∫ X u 2 ≥ 2 π , ∫ X u ≤ 2 π + ℓ sinh 1 ≤ 3 π . (276)
For g ≥ 3 g\ge3 g ≥ 3 and n ≥ 1 n\ge1 n ≥ 1 , A g , n ≥ 10 π A_{g,n}\ge10\pi A g , n ≥ 10 π . With u ˉ = A g , n − 1 ∫ X u \bar{u}=A_{g,n}^{-1}\int_Xu u ˉ = A g , n − 1 ∫ X u we therefore have
∥ u − u ˉ ∥ 2 2 = ∫ X u 2 − ( ∫ X u ) 2 A g , n ≥ π , ∫ X ∣ ∇ ( u − u ˉ ) ∣ 2 ∥ u − u ˉ ∥ 2 2 ≤ ℓ ε sinh 1 π ≤ τ ε 4 . \|u-\bar{u}\|_2^2=\int_Xu^2-\frac{\left(\int_Xu\right)^2}{A_{g,n}}\ge\pi,\qquad\frac{\int_X|\nabla(u-\bar{u})|^2}{\|u-\bar{u}\|_2^2}\le\frac{\ell_\varepsilon\sinh1}{\pi}\le\frac{\tau_\varepsilon}{4}. ∥ u − u ˉ ∥ 2 2 = ∫ X u 2 − A g , n ( ∫ X u ) 2 ≥ π , ∥ u − u ˉ ∥ 2 2 ∫ X ∣∇ ( u − u ˉ ) ∣ 2 ≤ π ℓ ε sinh 1 ≤ 4 τ ε .
The variational principle on the orthogonal complement of constants, together with the essential spectral threshold 1 / 4 1/4 1/4 , gives a nonzero discrete L 2 L^2 L 2 eigenvalue below τ ε \tau_\varepsilon τ ε ; see [21 Theorem 7] . Thus the presence of either cap implies N ε ≥ 1 N_\varepsilon\ge1 N ε ≥ 1 .
*Two-cusp pants when n ≥ 2. ∗ L e t n\ge2.* Let n ≥ 2. ∗ L e t P(X)$ count simple closed geodesics of length at most ℓ ε \ell_\varepsilon ℓ ε cutting off exactly two cusps and no genus. Distinct such geodesics are disjoint by the collar lemma. Their caps cannot be nested, since the region between them would be an annulus bounded by distinct closed geodesics. Nor can two caps cover X X X , since g ≥ 3 g \ge3 g ≥ 3 . Hence the caps, and their cusp pairs, are disjoint.
Mirzakhani’s marked integration formula, with the convention of Lemma 3.2, gives
E g , n P = ( n 2 ) V g , n ∫ 0 ℓ ε x V g , n − 1 ( x , 0 n − 2 ) d x . (277) \mathbb{E}_{g,n}P = \frac{\binom{n}{2}}{V_{g,n}} \int_{0}^{\ell_\varepsilon} xV_{g,n-1}(x,\mathbf{0}_{n-2})\,\mathrm{d}x. \tag*{(277)} E g , n P = V g , n ( 2 n ) ∫ 0 ℓ ε x V g , n − 1 ( x , 0 n − 2 ) d x . (277)
For n ≥ 4 n \ge4 n ≥ 4 , the ordered count of two distinct caps gives
E g , n [ P ( P − 1 ) ] = ( n 2 ) ( n − 2 2 ) V g , n ∫ [ 0 , ℓ ε ] 2 x y V g , n − 2 ( x , y , 0 n − 4 ) d x d y . (278) \mathbb{E}_{g,n}[P(P-1)] = \frac{\binom{n}{2}\binom{n-2}{2}}{V_{g,n}} \int_{[0,\ell_\varepsilon]^2} xyV_{g,n-2}(x,y,\mathbf{0}_{n-4})\,\mathrm{d}x\,\mathrm{d}y. \tag*{(278)} E g , n [ P ( P − 1 )] = V g , n ( 2 n ) ( 2 n − 2 ) ∫ [ 0 , ℓ ε ] 2 x y V g , n − 2 ( x , y , 0 n − 4 ) d x d y . (278)
For n = 2 , 3 n=2,3 n = 2 , 3 this factorial moment is zero. The second formula has no exchange factor because the pair is ordered. Both formulas also follow from [21 Theorem 6] , with V 0 , 3 = 1 V_{0,3}=1 V 0 , 3 = 1 .
For every stable type, the volume comparisons [21 Lemma 14] give
V g , m ≤ V g , m ( x ) ≤ e ∑ i x i 2 / 2 V g , m , V g , m + 1 ( 2 g − 2 + m ) V g , m ≲ 1 , (279) V_{g,m} \le V_{g,m}(x) \le e^{\sum_i x_i^2/2}V_{g,m}, \qquad\frac{V_{g,m+1}}{(2g-2+m)V_{g,m}} \lesssim1, \tag*{(279)} V g , m ≤ V g , m ( x ) ≤ e ∑ i x i 2 /2 V g , m , ( 2 g − 2 + m ) V g , m V g , m + 1 ≲ 1 , (279)
with absolute comparison constants. These bounds have no restriction on the growth of m m m . Consequently, with q g , n = n ( n − 1 ) / ( g + n ) q_{g,n}=n(n-1)/(g+n) q g , n = n ( n − 1 ) / ( g + n ) , there are positive constants a ε , A ε , B ε a_\varepsilon,A_\varepsilon,B_\varepsilon a ε , A ε , B ε , independent of g , n g,n g , n , such that
a ε q g , n ≤ E g , n P ≤ A ε q g , n , E g , n [ P ( P − 1 ) ] ≤ B ε q g , n 2 . (280) a_\varepsilon q_{g,n} \le\mathbb{E}_{g,n}P \le A_\varepsilon q_{g,n}, \qquad\mathbb{E}_{g,n}[P(P-1)] \le B_\varepsilon q_{g,n}^{2}. \tag*{(280)} a ε q g , n ≤ E g , n P ≤ A ε q g , n , E g , n [ P ( P − 1 )] ≤ B ε q g , n 2 . (280)
Indeed, products of one or two adjacent-volume ratios are comparable to ( g + n ) − 1 (g+n)^{-1} ( g + n ) − 1 or ( g + n ) − 2 (g+n)^{-2} ( g + n ) − 2 respectively, and the bounded boundary lengths contribute only constants depending on ℓ ε \ell_\varepsilon ℓ ε . Cauchy–Schwarz now yields
P g , n ( P ≥ 1 ) ≥ ( E g , n P ) 2 E g , n P 2 ≥ c ε ′ q g , n 1 + q g , n . (281) \mathbb{P}_{g,n}(P \ge1) \ge\frac{(\mathbb{E}_{g,n}P)^2}{\mathbb{E}_{g,n}P^2} \ge c'_\varepsilon\frac{q_{g,n}}{1+q_{g,n}}. \tag*{(281)} P g , n ( P ≥ 1 ) ≥ E g , n P 2 ( E g , n P ) 2 ≥ c ε ′ 1 + q g , n q g , n . (281)
For n ≥ 2 n \ge2 n ≥ 2 , putting t = ( n + 1 ) 2 / g t=(n+1)^2/g t = ( n + 1 ) 2 / g , we have
q g , n 1 + q g , n = n ( n − 1 ) g + n 2 ≥ 2 9 t 1 + t ≥ 1 9 min { t , 1 } . \frac{q_{g,n}}{1+q_{g,n}} = \frac{n(n-1)}{g+n^2} \ge\frac{2}{9}\frac{t}{1+t} \ge\frac{1}{9}\min\{t,1\}. 1 + q g , n q g , n = g + n 2 n ( n − 1 ) ≥ 9 2 1 + t t ≥ 9 1 min { t , 1 } .
Together with the cap test, this proves (275) for n ≥ 2 n \ge2 n ≥ 2 .
Torus caps when n = 1 n=1 n = 1 . Let T ( X ) T(X) T ( X ) count simple closed geodesics of length at most ℓ ε \ell_\varepsilon ℓ ε cutting off a one-holed torus containing no cusp. Distinct counted boundaries are disjoint. The same annulus argument excludes nesting of the caps, and two genus-one caps cannot cover a surface of genus at least three. The complementary types for one cap and an ordered pair are ( g − 1 , 2 ) (g-1,2) ( g − 1 , 2 ) and ( g − 2 , 3 ) (g-2,3) ( g − 2 , 3 ) . Marked integration therefore gives
E g , 1 T = 1 V g , 1 ∫ 0 ℓ ε x V 1 , 1 ( x ) V g − 1 , 2 ( x , 0 ) d x , (282) \mathbb{E}_{g,1}T = \frac{1}{V_{g,1}}\int_{0}^{\ell_\varepsilon} xV_{1,1}(x)V_{g-1,2}(x,0)\,\mathrm{d}x, \tag*{(282)} E g , 1 T = V g , 1 1 ∫ 0 ℓ ε x V 1 , 1 ( x ) V g − 1 , 2 ( x , 0 ) d x , (282)
E g , 1 [ T ( T − 1 ) ] = 1 V g , 1 ∫ [ 0 , ℓ ε ] 2 x y V 1 , 1 ( x ) V 1 , 1 ( y ) V g − 2 , 3 ( x , y , 0 ) d x d y . (283) \mathbb{E}_{g,1}[T(T-1)] = \frac{1}{V_{g,1}}\int_{[0,\ell_\varepsilon]^2} xyV_{1,1}(x)V_{1,1}(y)V_{g-2,3}(x,y,0)\,\mathrm{d}x\,\mathrm{d}y. \tag*{(283)} E g , 1 [ T ( T − 1 )] = V g , 1 1 ∫ [ 0 , ℓ ε ] 2 x y V 1 , 1 ( x ) V 1 , 1 ( y ) V g − 2 , 3 ( x , y , 0 ) d x d y . (283)
Here V 1 , 1 ( x ) = ( x 2 + 4 π 2 ) / 48 V_{1,1}(x)=(x^2+4\pi^2)/48 V 1 , 1 ( x ) = ( x 2 + 4 π 2 ) /48 already includes the torus involution factor, as in Lemma 3.2. The ordered pair again has no exchange factor. The boundary comparisons in (279) and the asymptotic for a fixed number of marked points [19 Theorem 1.2] imply, for all sufficiently large g g g ,
E g , 1 T ≍ ε V g − 1 , 2 V g , 1 ≍ ε g − 1 , E g , 1 [ T ( T − 1 ) ] ≪ ε V g − 2 , 3 V g , 1 ≍ g − 2 . (284) \mathbb{E}_{g,1}T \asymp_\varepsilon\frac{V_{g-1,2}}{V_{g,1}} \asymp_\varepsilon g^{-1}, \qquad\mathbb{E}_{g,1}[T(T-1)] \ll_\varepsilon\frac{V_{g-2,3}}{V_{g,1}} \asymp g^{-2}. \tag*{(284)} E g , 1 T ≍ ε V g , 1 V g − 1 , 2 ≍ ε g − 1 , E g , 1 [ T ( T − 1 )] ≪ ε V g , 1 V g − 2 , 3 ≍ g − 2 . (284)
Applying the same second-moment inequality gives P g , 1 ( T ≥ 1 ) ≥ c ε ′ ′ / g \mathbb{P}_{g,1}(T \ge1) \ge c''_\varepsilon/g P g , 1 ( T ≥ 1 ) ≥ c ε ′′ / g . Since ( n + 1 ) 2 = 4 (n+1)^2=4 ( n + 1 ) 2 = 4 when n = 1 n=1 n = 1 , the cap test proves (275) after decreasing c ε c_\varepsilon c ε and choosing g ε l o w g_\varepsilon^{\mathrm{low}} g ε low sufficiently large, with g ε l o w ≥ 4 g_\varepsilon^{\mathrm{low}} \ge4 g ε low ≥ 4 .
Corollary 6.15 (Sharp count and cusp range). For every fixed 0 < ε < 1 / 4 0 < \varepsilon< 1/4 0 < ε < 1/4 there are constants c ε , C ε > 0 c_{\varepsilon}, C_{\varepsilon} > 0 c ε , C ε > 0 , K ε ≥ 1 K_{\varepsilon} \ge1 K ε ≥ 1 and g ε ≥ 3 g_{\varepsilon} \ge3 g ε ≥ 3 , independent of n n n , such that all integers g ≥ 3 g \ge3 g ≥ 3 and n ≥ 1 n \ge1 n ≥ 1 with g ≥ g ε g \ge g_{\varepsilon} g ≥ g ε and g ≥ K ε ( n + 1 ) 2 g \ge K_{\varepsilon}(n + 1)^2 g ≥ K ε ( n + 1 ) 2 satisfy
c ε ( n + 1 ) 2 g ≤ P g , n ( N ε ≥ 1 ) ≤ E g , n N ε ≤ C ε ( n + 1 ) 2 g . (285) c_{\varepsilon}\frac{(n+1)^2}{g} \le\mathbb{P}_{g,n}(N_{\varepsilon} \ge1) \le\mathbb{E}_{g,n}N_{\varepsilon} \le C_{\varepsilon}\frac{(n+1)^2}{g}. \tag*{(285)} c ε g ( n + 1 ) 2 ≤ P g , n ( N ε ≥ 1 ) ≤ E g , n N ε ≤ C ε g ( n + 1 ) 2 . (285) Moreover, for any positive integer sequence n ( g ) n(g) n ( g ) and X g X_g X g with law P g , n ( g ) \mathbb{P}_{g,n(g)} P g , n ( g ) , as g → ∞ g \to\infty g → ∞ ,
SpG ( X g ) → p r o b . 1 4 ⟺ n ( g ) = o ( g ) . (286) \operatorname{SpG}(X_g) \xrightarrow{\mathrm{prob.}} \frac{1}{4} \Longleftrightarrow n(g) = o(\sqrt{g}). \tag*{(286)} SpG ( X g ) prob. 4 1 ⟺ n ( g ) = o ( g ) . (286)
Proof. Combine Proposition 6.14 with Theorem 1.1, taking K ε K_{\varepsilon} K ε at least one and at least the quadratic-threshold constant in that theorem, and increasing the lower genus bound. Then ( n + 1 ) 2 / g ≤ 1 (n+1)^2/g \le1 ( n + 1 ) 2 / g ≤ 1 , which proves (285) . For (286) , sufficiency follows from Theorem 1.1 and SpG ( X g ) ≤ 1 / 4 \operatorname{SpG}(X_g) \le1/4 SpG ( X g ) ≤ 1/4 . Conversely, convergence to 1 / 4 1/4 1/4 implies P g , n ( g ) ( N 1 / 8 ≥ 1 ) = P g , n ( g ) ( SpG ≤ 1 / 8 ) → 0 \mathbb{P}_{g,n(g)}(N_{1/8} \ge1)=\mathbb{P}_{g,n(g)}(\operatorname{SpG} \le1/8) \to0 P g , n ( g ) ( N 1/8 ≥ 1 ) = P g , n ( g ) ( SpG ≤ 1/8 ) → 0 . Proposition 6.14 with ε = 1 / 8 \varepsilon= 1/8 ε = 1/8 then forces ( n ( g ) + 1 ) 2 / g → 0 (n(g)+1)^2/g \to0 ( n ( g ) + 1 ) 2 / g → 0 , proving necessity. ∎