betaFeedback

WorksMathematics

Restricted Quantum unique ergodicity under Generalized Lindelöf conjecture

Abstract

Assume the generalized Lindelöf hypothesis. Let uu be an even Hecke–Maass cusp form on SL2(Z)\H\mathrm{SL}_2(\mathbb{Z})\backslash\mathbb{H}, normalized in L2L^2 with respect to hyperbolic area, with Laplace eigenvalue 1/4+T21/4+T^2. For every fixed ψ∈Cc((0,∞);C)ψ\in C_c((0,\infty);\mathbb{C}) and every ε>0\varepsilon>0, we prove that ∫0∞ψ(y)∣u(iy)∣2dyy=6π∫0∞ψ(y)dyy+Oψ,ε(T−1/4+ε) \int_0^\infty ψ(y)|u(i y)|^2\frac{d y}{y} =\frac{6}π\int_0^\infty ψ(y)\frac{d y}{y} +O_{ψ,\varepsilon}(T^{-1/4+\varepsilon}) as T→∞T\to\infty, uniformly in uu. In particular, the restriction mass equidistributes on every fixed compact segment of the imaginary axis in H\mathbb{H}.

Introduction

Let

Γ=PSL⁡2(Z),X=Γ\H,dμ(z)=dx dyy2,vol⁡(X)=π3.\Gamma=\operatorname{PSL}_2(\mathbb{Z}),\qquad X=\Gamma\backslash\mathbb{H},\qquad d\mu(z)=\frac{dx\,dy}{y^2},\qquad\operatorname{vol}(X)=\frac{\pi}{3}.

We use the nonnegative Laplacian Δ=−y2(∂x2+∂y2)\Delta=-y^2(\partial_x^2+\partial_y^2). Throughout, uu is a real-valued even Hecke–Maass cusp form satisfying

Δu=(14+T2)u,∫X∣u(z)∣2dμ(z)=1,u(−z‾)=u(z),T≥2.(1)\Delta u=\left(\frac14+T^2\right)u,\qquad\int_X|u(z)|^2d\mu(z)=1,\qquad u(-\overline{z})=u(z),\qquad T\ge2. \tag*{(1)}

The arithmetic quantum unique ergodicity (QUE) theorem of Lindenstrauss and Soundararajan [10] and [13] gives

∫Xf(z)∣u(z)∣2dμ(z)=3π∫Xf(z)dμ(z)+of(1)(T→∞)\int_X f(z)|u(z)|^2d\mu(z)=\frac{3}{\pi}\int_X f(z)d\mu(z)+o_f(1)\qquad(T\to\infty)

for every fixed f∈C∞(X)f\in C^\infty(X). Restricted QUE concerns the measures obtained by restricting ∣u∣2|u|^2 to a geodesic and integrating against hyperbolic arclength. The geodesic has zero hyperbolic area, so the surface asymptotic alone does not determine the restricted mass. Young formulated the following conjecture for the geodesic joining 00 to i∞i\infty and proved its Eisenstein analogue [17 Conjecture 1.1 and Theorem 1.2].

Conjecture 1.1. As uu ranges over the forms in (1.1), for every real-valued ψ∈Cc∞((0,∞))\psi\in C^\infty_c((0,\infty)) one has

lim⁡T→∞∫0∞ψ(y)∣u(iy)∣2dyy=6π∫0∞ψ(y)dyy.(2)\lim_{T\to\infty}\int_0^\infty\psi(y)|u(iy)|^2\frac{dy}{y}=\frac{6}{\pi}\int_0^\infty\psi(y)\frac{dy}{y}. \tag*{(2)}

2020 Mathematics Subject Classification. Primary 11F12; Secondary 11F66, 11M41, 58J51.

Key words and phrases. Hecke–Maass forms, restricted quantum unique ergodicity, Bessel functions, shifted convolution sums.

Q. Sun was partially supported by the National Natural Science Foundation of China (Grant No. 12471005) and the Natural Science Foundation of Shandong Province (Grant No. ZR2023MA003). Q. Zhang was supported by the Program of China Scholarship Council (Grant No. 202606220052).

Young later showed that the arithmetic position of a vertical geodesic affects the Eisenstein restriction asymptotic and proposed an analogous formula for Hecke–Maass cusp forms [18 Theorem 1.1 and Conjecture 1.3]. Unconditional density-one results on the imaginary axis are known for holomorphic Hecke cusp forms by Zenz [19 Corollary 1.4], for holomorphic Hecke cusp forms in short weight intervals by Sun and Zhang [14 Corollary 1.4], and for even Hecke–Maass cusp forms in short spectral intervals by Sun and Zhang [15 Corollary 1.3].

In his 2024 lecture, Humphries [7] presented a period method for effective restricted QUE on closed geodesics under the generalized Lindelöf hypothesis (GLH) and indicated that it also applies to vertical geodesics from a rational point to i∞i\infty. For the imaginary axis we give a shifted-convolution proof with an explicit power saving for each fixed smooth weight, uniform over individual even forms.

Theorem 1.2. Assume the generalized Lindelöf hypothesis (GLH). For every fixed ψ∈Cc∞((0,∞);C)\psi\in C^\infty_c((0,\infty);\mathbb{C}) and every ε>0\varepsilon> 0,

∫0∞ψ(y)∣u(iy)∣2dyy=6π∫0∞ψ(y)dyy+Oψ,ε(T−1/4+ε)(3)\int_0^\infty\psi(y)|u(iy)|^2 \frac{\mathrm{d}y}{y} = \frac{6}{\pi}\int_0^\infty\psi(y)\frac{\mathrm{d}y}{y} + O_{\psi,\varepsilon}(T^{-1/4+\varepsilon}) \tag*{(3)}

uniformly over the even Hecke–Maass cusp forms satisfying (1.1).

The interval version of this result concerns the L2L^2 norm of a restriction. For fixed 0<a<b<∞0 < a < b < \infty, let γa,b={iy:a≤y≤b}⊂H\gamma_{a,b} = \{iy : a \le y \le b\} \subset\mathbb{H}. Hyperbolic arclength is ds=dy/y\mathrm{d}s = \mathrm{d}y/y, and

∥u∣γa,b∥L2(ds)2=∫ab∣u(iy)∣2dyy.\|u|_{\gamma_{a,b}}\|_{L^2(\mathrm{d}s)}^2 = \int_a^b |u(iy)|^2 \frac{\mathrm{d}y}{y}.

The length of this segment is log⁡(b/a)\log(b/a). Bounds for these norms have also played a role in the study of nodal domains of Maass forms.

Ghosh, Reznikov, and Sarnak [3 Theorems 1.3 and 6.1] proved an upper bound of Oε(Tε)O_\varepsilon(T^\varepsilon) for the squared restriction norm on the imaginary axis, together with positive lower bounds on sufficiently long fixed compact segments. Jung [5 Corollary 1.12] obtained a positive lower bound on every fixed compact segment for a density-one family of even Hecke–Maass forms. The extension of this lower bound to all even Hecke–Maass cusp forms follows from Jang and Jung [4 Corollary 3.2]. These bounds concern the size of the restriction norm, while the restricted QUE asymptotic specifies its limiting value.

We record the corresponding interval formulation of Theorem 1.2. It follows by approximating the characteristic function of [a,b][a,b] from above and below by smooth functions and using the positivity of ∣u∣2|u|^2.

Corollary 1.3. Assume GLH. For every fixed 0<a<b<∞0 < a < b < \infty,

∫ab∣u(iy)∣2dyy=6πlog⁡(b/a)+o(1),(4)\int_a^b |u(iy)|^2 \frac{\mathrm{d}y}{y} = \frac{6}{\pi}\log(b/a) + o(1), \tag*{(4)}

as TT tends to infinity.

Outline of proof. The Fourier expansion reduces the restriction integral to a sum over two Fourier indices. We separate the diagonal and split the off-diagonal terms at shifts of size RR. Poincaré series express the diagonal and the main part of the short shifts as integrals over XX. Effective QUE under GLH supplies the diagonal main term and bounds the short shifts.

The essential observation for the long shifts is that both Fourier indices become small compared with TT. The estimates of Sun and Zhang [15] allow us to restrict, with a negligible error, to n2−m2≪T1+βn^2-m^2\ll T^{1+\beta}, where β>0\beta>0 can be chosen arbitrarily small. Since n−m>Rn-m>R, we have

m+n≪T1+βR.m+n\ll\frac{T^{1+\beta}}{R}.

For R=TαR=T^\alpha and β<α\beta<\alpha, both indices are therefore o(T)o(T). In this range the Bessel expansion in Lemma 3.3 produces the factors (n/m)±iT(n/m)^{\pm iT}. After smooth dyadic decomposition, double Mellin inversion separates the remaining weight and gives products

L(s+ωiT,u)L(z−ωiT,u),ω=±1.L(s+\omega iT,u)L(z-\omega iT,u),\qquad\omega=\pm1.

Moving both contours to the critical line, we apply GLH together with the uniform Mellin estimate. Summing together gives the bound of long shifts.

Preliminaries

Hecke–Maass forms

Write e(x)=e2πixe(x)=e^{2\pi ix}. For uu as in (1), the Fourier expansion is

u(x+iy)=y∑n≠0ρu(n)KiT(2π∣n∣y)e(nx).(5)u(x+iy)=\sqrt{y}\sum_{n\ne0}\rho_u(n)K_{iT}(2\pi|n|y)e(nx). \tag*{(5)}

Evenness and the Hecke relations give

ρu(n)=ρu(−n)=ρu(1)λu(n)(n≥1),λu(1)=1.\rho_u(n)=\rho_u(-n)=\rho_u(1)\lambda_u(n)\quad(n\ge1),\qquad\lambda_u(1)=1.

Put Lu=L(1,sym⁡2u)L_u=L(1,\operatorname{sym}^2u). The L2L^2 normalization gives

∣ρu(1)∣2=2cosh⁡(πT)Lu;(6)|\rho_u(1)|^2=\frac{2\cosh(\pi T)}{L_u}; \tag*{(6)}

see [11 p. 790] or [15 Section 2]. The bounds of Hoffstein–Lockhart [6] and Li [9], together with [3 Lemma 3.6], give

T−ε≪εLu≪εTε,∑n≤Z∣λu(n)∣2≪εZ(ZT)ε(Z≥1).(7)T^{-\varepsilon}\ll_\varepsilon L_u\ll_\varepsilon T^\varepsilon,\qquad\sum_{n\le Z}|\lambda_u(n)|^2\ll_\varepsilon Z(ZT)^\varepsilon\quad(Z\ge1). \tag*{(7)}

The Kim–Sarnak bound [8] is

∣λu(n)∣≤d(n)n7/64(n≥1),|\lambda_u(n)|\le d(n)n^{7/64}\quad(n\ge1),

where d(n)d(n) is the divisor function.

The standard LL-function attached to uu is

L(s,u)=∑n≥1λu(n)ns=∏p(1−λu(p)p−s+p−2s)−1(Re⁡s>1),L(s,u)=\sum_{n\ge1}\frac{\lambda_u(n)}{n^s}=\prod_p(1-\lambda_u(p)p^{-s}+p^{-2s})^{-1}\qquad(\operatorname{Re}s>1),

which can be analytically continued to the entire complex plane C\mathbb{C}. Under GLH, for every fixed $\nu>0,

∣L(12+it,u)∣≪ν(1+T+∣t∣)ν(t∈R),\left|L\left(\frac{1}{2}+it,u\right)\right|\ll_\nu(1+T+|t|)^\nu\qquad(t\in\mathbb{R}),

uniformly in uu and tt.

Effective QUE under GLH

Under GLH, we also have the following effective QUE estimate on XX. For f∈Cc∞(X)f \in C^\infty_c(X), write

Ju(f)=∫Xf(z)∣u(z)∣2 dμ(z).J_u(f)=\int_X f(z)|u(z)|^2\,d\mu(z).

With the nonnegative self-adjoint Laplacian Δ\Delta on L2(X)L^2(X), put

∥f∥L2,s(X)=∥(1+Δ)s/2f∥L2(X).(8)\|f\|_{L^{2,s}(X)}=\|(1+\Delta)^{s/2}f\|_{L^2(X)}. \tag*{(8)}

where fractional powers are defined by spectral calculus. For f∈Cc∞(X)f \in C^\infty_c(X), spectral calculus and integration by parts give

∥f∥L2,1(X)2=⟨(1+Δ)f,f⟩L2(X)=∥f∥22+∫X∣∇hypf∣2 dμ,(9)\|f\|_{L^{2,1}(X)}^2=\langle(1+\Delta)f,f\rangle_{L^2(X)}=\|f\|_2^2+\int_X|\nabla_{\mathrm{hyp}}f|^2\,d\mu, \tag*{(9)}

where ∣∇hypf∣2=y2(∣∂xf∣2+∣∂yf∣2)|\nabla_{\mathrm{hyp}}f|^2=y^2(|\partial_xf|^2+|\partial_yf|^2). Taking A=(1+Δ)1/2A=(1+\Delta)^{1/2}, Cauchy–Schwarz gives

∥f∥L2,1/2(X)2=⟨Af,f⟩≤∥Af∥2∥f∥2=∥f∥L2,1(X)∥f∥2.(10)\|f\|_{L^{2,1/2}(X)}^2=\langle Af,f\rangle\leq\|Af\|_2\|f\|_2=\|f\|_{L^{2,1}(X)}\|f\|_2. \tag*{(10)}

Lemma 2.1. Assume GLH. For every fixed η>0\eta>0 and every f∈Cc∞(X)f\in C^\infty_c(X),

∣Ju(f)−3π∫Xf dμ∣≪X,ηT−1/2+2η∥f∥L2,1/2(X),\left|J_u(f)-\frac{3}{\pi}\int_Xf\,d\mu\right|\ll_{X,\eta}T^{-1/2+2\eta}\|f\|_{L^{2,1/2}(X)},

with an implied constant independent of uu and ff.

Proof. Apply [16 Section 4.2, Corollary 1] under GLH with dB=1d_B=1, a=0a=0, and exponent η>0\eta>0. Watson’s eigenvalue is 1/4+T21/4+T^2 and his Laplacian is −Δ-\Delta, so his Sobolev norm agrees with (2.4) and (1/4+T2)−1/4+η≪ηT−1/2+2η(1/4+T^2)^{-1/4+\eta}\ll_\eta T^{-1/2+2\eta}. □\square

Bessel integrals

Fix ψ∈Cc∞((0,∞);C)\psi\in C^\infty_c((0,\infty);\mathbb{C}) and choose 0<a<b<∞0<a<b<\infty such that supp⁡ψ⊂[a,b]\operatorname{supp}\psi\subset[a,b]. For m,n≥1m,n\geq1, define

Bm,n(T)=16cosh⁡(πT)∫0∞ψ(y)KiT(2πmy)KiT(2πny) dy.(11)B_{m,n}(T)=16\cosh(\pi T)\int_0^\infty\psi(y)K_{iT}(2\pi my)K_{iT}(2\pi ny)\,dy. \tag*{(11)}

We first record estimates that truncate sums involving Bm,nB_{m,n}. We then obtain an expansion of Bm,n(T)B_{m,n}(T) when both Bessel arguments are at most T/2T/2. The factors (n/m)±iT(n/m)^{\pm iT} in this expansion will allow us to use LL-functions in Section 6.

Truncation

The estimates of Sun and Zhang allow us to discard terms with large mm, nn, or ∣n2−m2∣|n^2-m^2|, even after multiplying by a bounded weight.

Lemma 3.1. There is a constant Cψ>0C_\psi>0 such that, for every fixed β,B>0\beta,B>0 and every bounded complex-valued weight bu,Tb_{u,T} on N2\mathbb{N}^2,

12Lu∑(m,n)∉Ωβ,T∣bu,T(m,n)λu(m)λu(n)Bm,n∣≪ψ,β,B∥bu,T∥∞T−B,\frac{1}{2L_u}\sum_{(m,n)\notin\Omega_{\beta,T}}|b_{u,T}(m,n)\lambda_u(m)\lambda_u(n)B_{m,n}|\ll_{\psi,\beta,B}\|b_{u,T}\|_\infty T^{-B},

where

Ωβ,T={(m,n)∈N2:m,n≤CψT, ∣n2−m2∣≤T1+β}.\Omega_{\beta,T}=\{(m,n)\in\mathbb{N}^2:m,n\leq C_\psi T,\ |n^2-m^2|\leq T^{1+\beta}\}.

Proof. The weighted statement follows by factoring out ∥bu,T∥∞\|b_{u,T}\|_\infty and applying the tail estimates in [15 Lemmas 2.2 and 3.4], together with (7). □\square

Integral expression

We begin with an asymptotic formula for KiT(x)K_{iT}(x).

Lemma 3.2. For T≥1T \ge1 and 0<x≤3T/40 < x \le3T/4, put

kT(x)=T2−x2,ΘT(x)=Tarcosh⁡(T/x)−kT(x)−π4.k_T(x) = \sqrt{T^2-x^2}, \qquad\Theta_T(x) = T\operatorname{arcosh}(T/x)-k_T(x)-\frac{\pi}{4}.

Then

KiT(x)=2π e−πT/2kT(x)−1/2{cos⁡ΘT(x)+O(T−1)},K_{iT}(x) = \sqrt{2\pi}\,e^{-\pi T/2}k_T(x)^{-1/2}\{\cos\Theta_T(x)+O(T^{-1})\},

with an absolute implied constant.

Proof. Put q=x/Tq=x/T and Θ=ΘT(x)\Theta=\Theta_T(x). By Andersen–Duke [1 Proposition A.1], with their ν=T\nu=T and z=qz=q, uniformly for T≥1T \ge1 and 0<q≤3/40<q\le3/4,

eπT/2KiT(Tq)=2πT1/2(1−q2)1/4{cos⁡Θ+3q2+224T(1−q2)3/2sin⁡Θ}+O(T−5/2).e^{\pi T/2}K_{iT}(Tq)=\frac{\sqrt{2\pi}}{T^{1/2}(1-q^2)^{1/4}}\left\{\cos\Theta+\frac{3q^2+2}{24T(1-q^2)^{3/2}}\sin\Theta\right\}+O(T^{-5/2}).

This gives the claimed asymptotic formula. □\square

Define

Wψ(v)=∫0∞ψ(y)e−iπ2vy2 dy.W_\psi(v)=\int_0^\infty\psi(y)e^{-i\pi^2vy^2}\,dy.

With s=y2s=y^2, the function gψ(s)=ψ(s)/(2s)g_\psi(s)=\psi(\sqrt{s})/(2\sqrt{s}) is also smooth on (0,∞)(0,\infty), which is supported in [a2,b2][a^2,b^2]. Its zero extension belongs to Cc∞(R)C^\infty_c(\mathbb{R}), and

Wψ(v)=∫Rgψ(s)e−iπ2vs ds.W_\psi(v)=\int_{\mathbb{R}}g_\psi(s)e^{-i\pi^2vs}\,ds.

It’s not hard to show that WψW_\psi is a Schwartz function on R\mathbb{R}.

Applying Lemma 3.2 inside the integral defining Bm,n(T)B_{m,n}(T) and expanding the phase difference isolates the factors (n/m)ωiT(n/m)^{\omega iT}. The following lemma gives the resulting formula.

Lemma 3.3. Suppose m<nm<n and 2πbn≤T/22\pi bn\le T/2. Put Δm,n=n2−m2\Delta_{m,n}=n^2-m^2. Then

Bm,n(T)=4πT∑ω∈{−1,1}(n/m)ωiTWψ(ωΔm,n/T)+Oψ(T−2+m2+n2T3(1+Δm,nT)).B_{m,n}(T)=\frac{4\pi}{T}\sum_{\omega\in\{-1,1\}}(n/m)^{\omega iT}W_\psi(\omega\Delta_{m,n}/T)+O_\psi\left(T^{-2}+\frac{m^2+n^2}{T^3}\left(1+\frac{\Delta_{m,n}}{T}\right)\right).

For arbitrary positive integers m,nm,n with 2πbmax⁡(m,n)≤T/22\pi b\max(m,n)\le T/2, including m=nm=n, one also has

Bm,n(T)≪ψT−1.B_{m,n}(T)\ll_\psi T^{-1}.

Proof. For y∈[a,b]y\in[a,b], write

xm(y)=2πmy,xn(y)=2πny,km(y)=T2−xm(y)2,kn(y)=T2−xn(y)2.x_m(y)=2\pi my,\qquad x_n(y)=2\pi ny,\qquad k_m(y)=\sqrt{T^2-x_m(y)^2},\qquad k_n(y)=\sqrt{T^2-x_n(y)^2}.

The assumption 2πbn≤T/22\pi bn\le T/2 gives km(y),kn(y)≥3T/2k_m(y),k_n(y)\ge\sqrt{3}T/2. Substituting the asymptotic formula in Lemma 3.2 into the definition of Bm,n(T)B_{m,n}(T) therefore gives

Bm,n(T)=16π(1+e−2πT)∫abψ(y)km(y)kn(y)cos⁡ΘT(xm(y))cos⁡ΘT(xn(y)) dy+Oψ(T−2).B_{m,n}(T)=16\pi(1+e^{-2\pi T})\int_a^b\frac{\psi(y)}{\sqrt{k_m(y)k_n(y)}}\cos\Theta_T(x_m(y))\cos\Theta_T(x_n(y))\,dy+O_\psi(T^{-2}).

Indeed, the error in each cosine factor is O(T−1)O(T^{-1}), whereas (kmkn)−1/2≪T−1(k_m k_n)^{-1/2} \ll T^{-1}. Set D(y)=ΘT(xm(y))−ΘT(xn(y))D(y)=\Theta_T(x_m(y))-\Theta_T(x_n(y)), S(y)=ΘT(xm(y))+ΘT(xn(y))S(y)=\Theta_T(x_m(y))+\Theta_T(x_n(y)).

Using 2cos⁡ucos⁡v=cos⁡(u−v)+cos⁡(u+v)2\cos u\cos v=\cos(u-v)+\cos(u+v), we obtain

Bm,n(T)=8π∫abψ(y)cos⁡D(y)km(y)kn(y) dy+8π∫abψ(y)cos⁡S(y)km(y)kn(y) dy+Oψ(T−2).B_{m,n}(T)=8\pi\int_a^b\frac{\psi(y)\cos D(y)}{\sqrt{k_m(y)k_n(y)}}\,dy +8\pi\int_a^b\frac{\psi(y)\cos S(y)}{\sqrt{k_m(y)k_n(y)}}\,dy+O_\psi(T^{-2}).

We first estimate the integral containing cos⁡S(y)\cos S(y). Since ΘT′(x)=−kT(x)/x\Theta_T'(x)=-k_T(x)/x,

−S′(y)=km(y)+kn(y)y≍ψT,kj′(y)=−(2πj)2ykj(y)=Oψ(T)(j=m,n).-S'(y)=\frac{k_m(y)+k_n(y)}{y}\asymp_\psi T,\qquad k_j'(y)=-\frac{(2\pi j)^2y}{k_j(y)}=O_\psi(T)\quad(j=m,n).

Consequently,

ddy(yψ(y)km(y)kn(y)(km(y)+kn(y)))=Oψ(T−2).\frac{d}{dy}\left(\frac{y\psi(y)}{\sqrt{k_m(y)k_n(y)}(k_m(y)+k_n(y))}\right)=O_\psi(T^{-2}).

Integration by parts now yields

∫abψ(y)cos⁡S(y)km(y)kn(y) dy=∫absin⁡S(y)ddy(yψ(y)km(y)kn(y)(km(y)+kn(y))) dy=Oψ(T−2).\int_a^b\frac{\psi(y)\cos S(y)}{\sqrt{k_m(y)k_n(y)}}\,dy =\int_a^b\sin S(y)\frac{d}{dy}\left(\frac{y\psi(y)}{\sqrt{k_m(y)k_n(y)}(k_m(y)+k_n(y))}\right)\,dy =O_\psi(T^{-2}).

Thus

Bm,n(T)=8π∫abψ(y)cos⁡D(y)km(y)kn(y) dy+Oψ(T−2).B_{m,n}(T)=8\pi\int_a^b\frac{\psi(y)\cos D(y)}{\sqrt{k_m(y)k_n(y)}}\,dy+O_\psi(T^{-2}).

For D(y)D(y), the identity

D(y)=∫xm(y)xn(y)kT(v)v dvD(y)=\int_{x_m(y)}^{x_n(y)}\frac{k_T(v)}{v}\,dv

and the uniform expansion kT(v)=T−v2/(2T)+O(v4/T3)k_T(v)=T-v^2/(2T)+O(v^4/T^3) for 0<v≤T/20<v\le T/2 give

D(y)=Tlog⁡xn(y)xm(y)−xn(y)2−xm(y)24T+O(xn(y)4−xm(y)4T3)=Tlog⁡(n/m)−π2Δm,ny2T+Oψ(Δm,n(m2+n2)T3).\begin{aligned} D(y)&=T\log\frac{x_n(y)}{x_m(y)} -\frac{x_n(y)^2-x_m(y)^2}{4T} +O\left(\frac{x_n(y)^4-x_m(y)^4}{T^3}\right)\\ &=T\log(n/m)-\frac{\pi^2\Delta_{m,n}y^2}{T} +O_\psi\left(\frac{\Delta_{m,n}(m^2+n^2)}{T^3}\right). \end{aligned}

Also,

1km(y)kn(y)=1T(1−xm(y)2T2)−1/4(1−xn(y)2T2)−1/4=1T+Oψ(m2+n2T3).\begin{aligned} \frac{1}{\sqrt{k_m(y)k_n(y)}}&=\frac{1}{T} \left(1-\frac{x_m(y)^2}{T^2}\right)^{-1/4} \left(1-\frac{x_n(y)^2}{T^2}\right)^{-1/4}\\ &=\frac{1}{T}+O_\psi\left(\frac{m^2+n^2}{T^3}\right). \end{aligned}

By the mean value theorem,

cos⁡D(y)=cos⁡(Tlog⁡(n/m)−π2Δm,ny2T)+Oψ(Δm,n(m2+n2)T3).\cos D(y)=\cos\left(T\log(n/m)-\frac{\pi^2\Delta_{m,n}y^2}{T}\right) +O_\psi\left(\frac{\Delta_{m,n}(m^2+n^2)}{T^3}\right).

Combining these estimates gives

Bm,n(T)=8πT∫abψ(y)cos⁡(Tlog⁡(n/m)−π2Δm,ny2T) dy+Oψ(T−2+m2+n2T3+Δm,n(m2+n2)T4).B_{m,n}(T)=\frac{8\pi}{T}\int_a^b\psi(y)\cos\left(T\log(n/m)-\frac{\pi^2\Delta_{m,n}y^2}{T}\right)\,dy+O_\psi\left(T^{-2}+\frac{m^2+n^2}{T^3}+\frac{\Delta_{m,n}(m^2+n^2)}{T^4}\right).

Finally, the main term is exactly

4πT∑ω∈{−1,1}(n/m)ωiT∫abψ(y)e−ωiπ2Δm,ny2/T dy=4πT∑ω∈{−1,1}(n/m)ωiTWψ(ωΔm,n/T),\begin{aligned} \frac{4\pi}{T}\sum_{\omega\in\{-1,1\}}(n/m)^{\omega iT}\int_a^b\psi(y)e^{-\omega i\pi^2\Delta_{m,n}y^2/T}\,dy \\ &=\frac{4\pi}{T}\sum_{\omega\in\{-1,1\}}(n/m)^{\omega iT}\mathcal{W}_{\psi}(\omega\Delta_{m,n}/T), \end{aligned}

which proves the claimed expansion.

For the final bound, apply Lemma 3.2 separately to both indices to obtain

eπT/2∣KiT(2πjy)∣≪T−1/2(j=m,n, y∈[a,b]).e^{\pi T/2}|K_{iT}(2\pi jy)|\ll T^{-1/2}\qquad(j=m,n,\ y\in[a,b]).

Substitution into the defining integral for Bm,n(T)B_{m,n}(T) gives Bm,n(T)≪ψT−1B_{m,n}(T)\ll_\psi T^{-1}, also when m=nm=n. □

Decomposition of the restriction integral

For the fixed test function ψ\psi, write

Iu(ψ)=∫0∞ψ(y)∣u(iy)∣2dyy,M(ψ)=6π∫0∞ψ(y)dyy.(12)I_u(\psi)=\int_0^\infty\psi(y)|u(iy)|^2\frac{dy}{y},\qquad M(\psi)=\frac{6}{\pi}\int_0^\infty\psi(y)\frac{dy}{y}. \tag*{(12)}

Substituting the Fourier expansion (5) into (12) and using (6) and (11), we obtain

Iu(ψ)=4∣ρu(1)∣2∑m,n≥1λu(m)λu(n)∫0∞ψ(y)KiT(2πmy)KiT(2πny) dy=12Lu∑m,n≥1λu(m)λu(n)Bm,n(T).(13)\begin{aligned} I_u(\psi)=4|\rho_u(1)|^2\sum_{m,n\geq1}\lambda_u(m)\lambda_u(n)\int_0^\infty\psi(y)K_{iT}(2\pi my)K_{iT}(2\pi ny)\,dy \\ &=\frac{1}{2L_u}\sum_{m,n\geq1}\lambda_u(m)\lambda_u(n)B_{m,n}(T). \tag*{(13)} \end{aligned}

We first split the sum in (13) into the diagonal and positive shifts. Put

Du=12Lu∑n≥1λu(n)2Bn,n,Su(r)=1Lu∑n≥1λu(n)λu(n+r)Bn,n+r.(14)D_u=\frac{1}{2L_u}\sum_{n\geq1}\lambda_u(n)^2B_{n,n},\qquad S_u(r)=\frac{1}{L_u}\sum_{n\geq1}\lambda_u(n)\lambda_u(n+r)B_{n,n+r}. \tag*{(14)}

Choose once and for all a smooth function χ\chi on R\mathbb{R} with values in [0,1][0,1] such that

χ(t)=1(t≤1),χ(t)=0(t≥2).\chi(t)=1\quad(t\leq1),\qquad\chi(t)=0\quad(t\geq2).

For R≥1R\geq1, we then split the off-diagonal sum into

Oshort=∑r≥1χ(r/R)Su(r),O_{\mathrm{short}}=\sum_{r\geq1}\chi(r/R)S_u(r),
Olong=∑r≥1(1−χ(r/R))Su(r).(15)O_{\mathrm{long}}=\sum_{r\geq1}(1-\chi(r/R))S_u(r). \tag*{(15)}

Thus

Iu(ψ)=Du+∑r≥1Su(r)=Du+Oshort+Olong.(16)I_u(\psi)=D_u+\sum_{r\geq1}S_u(r)=D_u+O_{\mathrm{short}}+O_{\mathrm{long}}. \tag*{(16)}

Poincaré series

To relate DuD_u and Su(r)S_u(r) to integrals over XX, let Γ∞\Gamma_\infty be the stabilizer of the cusp at infinity. For r∈Zr \in\mathbb{Z}, the Poincaré series Pr(z)P_r(z) is defined by

Pr(z)=∑γ∈Γ∞\Γℑ(γz)ψ(ℑ(γz))e(rℜ(γz)).(17)P_r(z)=\sum_{\gamma\in\Gamma_\infty\backslash\Gamma}\Im(\gamma z)\psi(\Im(\gamma z))e(r\Re(\gamma z)). \tag*{(17)}

To apply Lemma 2.1, we therefore estimate the Sobolev norm of arbitrary finite linear combinations of Poincaré series.

Lemma 4.1. The series in (17) are locally finite. Their sums belong to Cc∞(X)C^\infty_c(X), and their supports lie in a common compact set KψK_\psi, independent of rr. For every finitely supported sequence (cr)r∈Z(c_r)_{r\in\mathbb{Z}},

∥∑rcrPr∥L2,1/2(X)≪ψ(∑r∣cr∣2)1/4(∑r(1+r2)∣cr∣2)1/4.\left\|\sum_r c_rP_r\right\|_{L^{2,1/2}(X)} \ll_\psi \left(\sum_r |c_r|^2\right)^{1/4} \left(\sum_r(1+r^2)|c_r|^2\right)^{1/4}.

Proof. Choose a nonnegative h∈Cc∞((0,∞))h\in C^\infty_c((0,\infty)) with h(y)≥y∣ψ(y)∣h(y)\geq y|\psi(y)| and h>0h>0 on [a,b][a,b]. By [2 Section 1.8], the incomplete Eisenstein series

Eh(z)=∑γ∈Γ∞\Γh(ℑ(γz))E_h(z)=\sum_{\gamma\in\Gamma_\infty\backslash\Gamma}h(\Im(\gamma z))

is locally finite and belongs to Cc∞(X)C^\infty_c(X). It follows that (4.6) is locally finite, and ∣Pr∣≤Eh|P_r|\leq E_h gives supp⁡Pr⊂Kψ:=supp⁡Eh\operatorname{supp}P_r\subset K_\psi:=\operatorname{supp}E_h. Moreover, writing ch=min⁡[a,b]h>0c_h=\min_{[a,b]}h>0, we have the uniform bound

Na,b(z):=#{γ∈Γ∞\Γ:a≤ℑ(γz)≤b}≤ch−1Eh(z)≪ψ1.N_{a,b}(z):=\#\{\gamma\in\Gamma_\infty\backslash\Gamma:a\leq\Im(\gamma z)\leq b\}\leq c_h^{-1}E_h(z)\ll_\psi1.

Put

F(x)=∑rcre(rx),g(x+iy)=yψ(y)F(x),P=∑rcrPr=∑γg(γz).F(x)=\sum_r c_re(rx),\qquad g(x+iy)=y\psi(y)F(x),\qquad P=\sum_r c_rP_r=\sum_\gamma g(\gamma z).

Cauchy–Schwarz followed by integration over XX and unfolding to Γ∞\H={0≤x<1, y>0}\Gamma_\infty\backslash\mathbb{H}=\{0\leq x<1,\ y>0\} gives

∥P∥22≪ψ∫0∞∫01∣g(x+iy)∣2dx dyy2≪ψ∫01∣F(x)∣2 dx,\|P\|_2^2\ll_\psi\int_0^\infty\int_0^1|g(x+iy)|^2\frac{dx\,dy}{y^2}\ll_\psi\int_0^1|F(x)|^2\,dx,
∥∇hypP∥22≪ψ∫0∞∫01(∣∂xg∣2+∣∂yg∣2) dx dy\|\nabla_{\mathrm{hyp}}P\|_2^2\ll_\psi\int_0^\infty\int_0^1\left(|\partial_xg|^2+|\partial_yg|^2\right)\,dx\,dy
≪ψ∫01(∣F(x)∣2+∣F′(x)∣2) dx,\ll_\psi\int_0^1\left(|F(x)|^2+|F'(x)|^2\right)\,dx,

where ∂xg=yψ(y)F′(x)\partial_xg=y\psi(y)F'(x) and ∂yg=(ψ(y)+yψ′(y))F(x)\partial_yg=(\psi(y)+y\psi'(y))F(x). Fourier orthogonality gives

∫01∣F(x)∣2 dx=∑r∣cr∣2,∫01∣F′(x)∣2 dx=4π2∑rr2∣cr∣2.\int_0^1|F(x)|^2\,dx=\sum_r|c_r|^2,\qquad \int_0^1|F'(x)|^2\,dx=4\pi^2\sum_r r^2|c_r|^2.

Consequently, (2.5) implies

∥P∥22≪ψ∑r∣cr∣2,∥P∥L2,1(X)2≪ψ∑r(1+r2)∣cr∣2.\|P\|_2^2\ll_\psi\sum_r|c_r|^2,\qquad \|P\|_{L^{2,1}(X)}^2\ll_\psi\sum_r(1+r^2)|c_r|^2.

Applying (2.6) proves the claimed bound.

Unfolding against the constant function gives

∫XPr dμ={∫0∞ψ(y) dyy,r=0,0,r≠0.(18)\int_X P_r\,d\mu= \begin{cases} \int_0^\infty\psi(y)\,\frac{dy}{y}, & r=0,\\ 0, & r\ne0. \end{cases} \tag*{(18)}

The next lemma expresses DuD_u and Su(r)S_u(r) in terms of Ju(Pr)J_u(P_r).

Lemma 4.2. One has

Du=2Ju(P0)D_u=2J_u(P_0)

and, for every integer r≥1r\ge1,

Su(r)=4Ju(Pr)−12Lu∑a′=1r−1λu(a′)λu(r−a′)Ba′,r−a′.S_u(r)=4J_u(P_r)-\frac{1}{2L_u}\sum_{a'=1}^{r-1}\lambda_u(a')\lambda_u(r-a')\mathcal{B}_{a',r-a'}.

Proof. Unfolding gives

Ju(Pr)=∫0∞∫01e(rx)ψ(y)∣u(x+iy)∣2 dx dyy.J_u(P_r)=\int_0^\infty\int_0^1 e(rx)\psi(y)|u(x+iy)|^2\,dx\,\frac{dy}{y}.

For r=0r=0, the xx integral pairs equal nonzero Fourier indices. Their positive and negative parts contribute equally. Substitution of (2.2) gives

Ju(P0)=14Lu∑n≥1λu(n)2Bn,n,J_u(P_0)=\frac{1}{4L_u}\sum_{n\ge1}\lambda_u(n)^2\mathcal{B}_{n,n},

which proves the identity for DuD_u.

For r>0r>0, the Fourier indices are kk and k+rk+r, with k≠0,−rk\ne0,-r. The ranges k≥1k\ge1 and k≤−r−1k\le-r-1 give two identical infinite sums. The remaining indices are k=−a′k=-a', 1≤a′≤r−11\le a'\le r-1. Therefore

Ju(Pr)=14Lu∑n≥1λu(n)λu(n+r)Bn,n+r+18Lu∑a′=1r−1λu(a′)λu(r−a′)Ba′,r−a′.J_u(P_r)=\frac{1}{4L_u}\sum_{n\ge1}\lambda_u(n)\lambda_u(n+r)\mathcal{B}_{n,n+r} +\frac{1}{8L_u}\sum_{a'=1}^{r-1}\lambda_u(a')\lambda_u(r-a')\mathcal{B}_{a',r-a'}.

Rearranging gives the identity for Su(r)S_u(r). □

Put PR=∑r≥1χ(r/R)PrP_R=\sum_{r\ge1}\chi(r/R)P_r. Multiplying the identity for Su(r)S_u(r) in Lemma 4.2 by χ(r/R)\chi(r/R) and summing over r≥1r\ge1 gives

Oshort=Oshort(1)+Oshort(2),(19)\mathcal{O}_{\mathrm{short}}=\mathcal{O}_{\mathrm{short}}^{(1)}+\mathcal{O}_{\mathrm{short}}^{(2)}, \tag*{(19)}

where

Oshort(1)=4∑r≥1χ(r/R)Ju(Pr)=4Ju(PR),\mathcal{O}_{\mathrm{short}}^{(1)} =4\sum_{r\ge1}\chi(r/R)J_u(P_r) =4J_u(P_R),
Oshort(2)=−12Lu∑r≥1χ(r/R)∑a′=1r−1λu(a′)λu(r−a′)Ba′,r−a′.(20)\mathcal{O}_{\mathrm{short}}^{(2)} =-\frac{1}{2L_u}\sum_{r\ge1}\chi(r/R)\sum_{a'=1}^{r-1} \lambda_u(a')\lambda_u(r-a')\mathcal{B}_{a',r-a'}. \tag*{(20)}

The second term subtracts the finite sum corresponding to −r<k<0-r<k<0 in the preceding calculation.

Short shifts

We will apply effective QUE to Ju(PR)J_u(\mathcal{P}_R) and estimate this finite sum directly. We first estimate DuD_u using effective QUE.

Lemma 4.3. Assume GLH. For every fixed η>0\eta> 0,

Du=M(ψ)+Oψ,η(T−1/2+2η).D_u=M(\psi)+O_{\psi,\eta}(T^{-1/2+2\eta}).

Proof. Taking c0=1c_0=1 and cr=0c_r=0 for r≠0r\ne0 in Lemma 4.1 gives ∥P0∥L2,1/2(X)≪ψ1\|P_0\|_{L^{2,1/2}(X)}\ll_\psi1. Together with the identity for DuD_u in Lemma 4.2, (18), and Lemma 2.1, this proves the result.

We next bound Oshort(1)\mathcal{O}^{(1)}_{\mathrm{short}} by applying effective QUE to the sum PR\mathcal{P}_R of the test functions.

Lemma 4.4. Assume GLH. For every R≥1R\ge1 and every fixed η>0\eta>0,

∣Oshort(1)∣≪ψ,ηT−1/2+2ηR.\left|\mathcal{O}^{(1)}_{\mathrm{short}}\right|\ll_{\psi,\eta} T^{-1/2+2\eta}R.

Proof. By (20), Oshort(1)=4Ju(PR)\mathcal{O}^{(1)}_{\mathrm{short}}=4J_u(\mathcal{P}_R). The mean of PR\mathcal{P}_R is zero by (18). Since 0≤χ≤10\le\chi\le1 and χ(r/R)=0\chi(r/R)=0 for r≥2Rr\ge2R, Lemma 4.1 gives

∥PR∥L2,1/2(X)≪ψ(∑1≤r<2Rχ(r/R)2)1/4(∑1≤r<2R(1+r2)χ(r/R)2)1/4≪ψR.\|P_R\|_{L^{2,1/2}(X)} \ll_\psi \left(\sum_{1\le r<2R}\chi(r/R)^2\right)^{1/4} \left(\sum_{1\le r<2R}(1+r^2)\chi(r/R)^2\right)^{1/4} \ll_\psi R.

Applying Lemma 2.1 to PRP_R proves the lemma. □\square

For Oshort(2)\mathcal{O}^{(2)}_{\mathrm{short}}, we use the final bound in Lemma 3.3 and Cauchy–Schwarz to estimate the finite sum directly.

Lemma 4.5. If R=TαR=T^\alpha with fixed 0<α<10<\alpha<1, then for every fixed ε>0\varepsilon>0,

∣Oshort(2)∣≪ψ,α,εT−1+εR2.\left|\mathcal{O}^{(2)}_{\mathrm{short}}\right| \ll_{\psi,\alpha,\varepsilon} T^{-1+\varepsilon}R^2.

Proof. Lemma 3.3 implies Ba′,r−a′≪ψT−1B_{a',r-a'}\ll_\psi T^{-1} for r≤2Rr\le2R, once TT is sufficiently large in terms of ψ\psi and α\alpha: both indices are at most 2R=o(T)2R=o(T). By (20) and the bound from Lemma 3.3,

∣Oshort(2)∣≤12Lu∑r≥1χ(r/R)∑a′=1r−1∣λu(a′)λu(r−a′)Ba′,r−a′∣≪ψT−1Lu−1(∑n≤2R∣λu(n)∣)2.\left|\mathcal{O}^{(2)}_{\mathrm{short}}\right| \le\frac{1}{2L_u}\sum_{r\ge1}\chi(r/R)\sum_{a'=1}^{r-1}\left|\lambda_u(a')\lambda_u(r-a')B_{a',r-a'}\right| \ll_\psi T^{-1}L_u^{-1}\left(\sum_{n\le2R}|\lambda_u(n)|\right)^2.

In the second line, we used 0≤χ≤10\le\chi\le1 and enlarged the sum to all pairs with both indices at most 2R2R. Now put δ=ε/4\delta=\varepsilon/4. Cauchy–Schwarz gives the first inequality in

(∑n≤2R∣λu(n)∣)2≤2R∑n≤2R∣λu(n)∣2≪δR2(RT)δ,\left(\sum_{n\le2R}|\lambda_u(n)|\right)^2 \le2R\sum_{n\le2R}|\lambda_u(n)|^2 \ll_\delta R^2(RT)^\delta,

and the last bound follows from (7) with exponent δ\delta. Using also Lu−1≪δTδL_u^{-1}\ll_\delta T^\delta from (7), we obtain

∣Oshort(2)∣≪ψ,δT−1+δR2(RT)δ=T−1+(2+α)δR2≪ψ,α,εT−1+εR2,\left|\mathcal{O}^{(2)}_{\mathrm{short}}\right| \ll_{\psi,\delta} T^{-1+\delta}R^2(RT)^\delta =T^{-1+(2+\alpha)\delta}R^2 \ll_{\psi,\alpha,\varepsilon}T^{-1+\varepsilon}R^2,

since (2+α)δ<ε(2+\alpha)\delta<\varepsilon for α<1\alpha<1. □\square

Reduction of the long-shift sum

Our aim in this section is to express the long-shift sum in a form that can be estimated by Mellin inversion and GLH in § 6. We first restrict n2−m2n^2-m^2 and use the lower bound on n−mn-m to bring both indices into the range of Lemma 3.3. Applying that lemma gives the sums Hω\mathcal{H}_{\omega} in Proposition 5.3; we also bound the errors from the truncation and the approximation in Lemma 3.3.

By (14) and (15), we have

Olong=1Lu∑m<nλu(m)λu(n)(1−χ((n−m)/R))Bm,n.\mathcal{O}_{\mathrm{long}}=\frac{1}{L_u}\sum_{m<n}\lambda_u(m)\lambda_u(n)(1-\chi((n-m)/R))\mathcal{B}_{m,n}.

To restrict n2−m2n^2-m^2, for R≥1R\geq1 and M>0M>0 define

Am,n=(1−χ((n−m)/R))χ((n2−m2)/M)(m<n),A_{m,n}=(1-\chi((n-m)/R))\chi((n^2-m^2)/M)\qquad(m<n),

and set Am,n=0A_{m,n}=0 for n≤mn\leq m.

Lemma 5.1. Let α\alpha and β\beta be fixed with

0<α≤14,0<β<α4,0<\alpha\leq\frac{1}{4},\qquad0<\beta<\frac{\alpha}{4},

and set

R=Tα,M=T1+β,Y=2MR.R=T^\alpha,\qquad M=T^{1+\beta},\qquad Y=\frac{2M}{R}.

Then, for every fixed B>0B>0,

Olong=1Lu∑m<nλu(m)λu(n)Am,nBm,n+Oψ,α,β,B(T−B).\mathcal{O}_{\mathrm{long}}=\frac{1}{L_u}\sum_{m<n}\lambda_u(m)\lambda_u(n)A_{m,n}\mathcal{B}_{m,n}+O_{\psi,\alpha,\beta,B}(T^{-B}).

If Am,n≠0A_{m,n}\ne0, then

n−m>R,n2−m2<2M,m+n<Y.n-m>R,\qquad n^2-m^2<2M,\qquad m+n<Y.

In particular, m,n=o(T)m,n=o(T).

Proof. Apply Lemma 3.1 with the present β\beta and

bu,T(m,n)=1m<n(1−χ((n−m)/R))(1−χ((n2−m2)/M)).b_{u,T}(m,n)=1_{m<n}(1-\chi((n-m)/R))(1-\chi((n^2-m^2)/M)).

This weight has sup norm at most one and vanishes when ∣n2−m2∣≤M=T1+β\lvert n^2-m^2\rvert\leq M=T^{1+\beta}. Hence its contribution is Oψ,β,B(T−B)O_{\psi,\beta,B}(T^{-B}), proving the asserted identity.

The first two support conditions in the statement follow from the cutoffs. Their quotient gives

m+n=n2−m2n−m<2MR=Y.m+n=\frac{n^2-m^2}{n-m}<\frac{2M}{R}=Y.

Since Y/T=2Tβ−α=o(1)Y/T=2T^{\beta-\alpha}=o(1), the last assertion follows. □

Keep α,β,R,M,Y\alpha,\beta,R,M,Y as in Lemma 5.1. The following bound controls the sum of the errors in the Bessel expansion.

Lemma 5.2. Suppose 1≤Y0≤M0≤T21\leq Y_0\leq M_0\leq T^2. Then

∑1≤m<n≤Y0n2−m2≤2M0∣λu(m)λu(n)∣≪εM0Tε\sum_{\substack{1\leq m<n\leq Y_0\\ n^2-m^2\leq2M_0}}\lvert\lambda_u(m)\lambda_u(n)\rvert\ll_{\varepsilon}M_0T^{\varepsilon}

Proof. Fix ε>0\varepsilon> 0 and put δ=ε/4\delta= \varepsilon/4. For n≤2M0n \le2\sqrt{M_0}, Cauchy–Schwarz and (7) with exponent δ\delta give

∑m,n≤2M0∣λu(m)λu(n)∣=(∑n≤2M0∣λu(n)∣)2≤2M0∑n≤2M0∣λu(n)∣2≪δM0(TM0)δ≤M0T2δ,\sum_{m,n\le2\sqrt{M_0}} |\lambda_u(m)\lambda_u(n)| = \left(\sum_{n\le2\sqrt{M_0}} |\lambda_u(n)|\right)^2 \le2\sqrt{M_0}\sum_{n\le2\sqrt{M_0}} |\lambda_u(n)|^2 \ll_\delta M_0(T\sqrt{M_0})^\delta\le M_0T^{2\delta},

since M0≤T2M_0\le T^2.

For the remaining indices, use dyadic intervals Z≤n<2ZZ\le n<2Z. Let EZE_Z denote the pairs in the sum with Z≤n<2ZZ\le n<2Z and n>2M0n>2\sqrt{M_0}. The inequality n2−m2≤2M0n^2-m^2\le2M_0 gives m2≥n2−2M0>n2/2m^2\ge n^2-2M_0>n^2/2; hence

Z2<m<n<2Z,1≤n−m=n2−m2n+m≤2M0Z.\frac{Z}{\sqrt{2}}<m<n<2Z,\qquad1\le n-m=\frac{n^2-m^2}{n+m}\le\frac{2M_0}{Z}.

For each fixed nn, there are O(1+M0/Z)O(1+M_0/Z) possible values of mm; the same bound holds for nn when mm is fixed. If EZE_Z is nonempty, then Z≤Y0≤M0Z\le Y_0\le M_0, so each of these counts is O(M0/Z)O(M_0/Z). It follows that

∑(m,n)∈EZ∣λu(m)λu(n)∣≤12∑(m,n)∈EZ(∣λu(m)∣2+∣λu(n)∣2)≪M0Z∑k≤2Z∣λu(k)∣2≪δM0ZZ(ZT)δ≤M0T3δ,\sum_{(m,n)\in E_Z}|\lambda_u(m)\lambda_u(n)| \le\frac{1}{2}\sum_{(m,n)\in E_Z}\left(|\lambda_u(m)|^2+|\lambda_u(n)|^2\right) \ll\frac{M_0}{Z}\sum_{k\le2Z}|\lambda_u(k)|^2 \ll_\delta\frac{M_0}{Z}Z(ZT)^\delta\le M_0T^{3\delta},

where the last step uses Z≤M0≤T2Z\le M_0\le T^2. There are O(log⁡(2T))O(\log(2T)) dyadic ranges. Combining their contributions with the first range and using log⁡(2T)≪δTδ\log(2T)\ll_\delta T^\delta gives

∑1≤m<n≤Y0n2−m2≤2M0∣λu(m)λu(n)∣≪δM0T2δ+M0T3δlog⁡(2T)≪εM0T4δ=M0Tε.\sum_{\substack{1\le m<n\le Y_0\\n^2-m^2\le2M_0}}|\lambda_u(m)\lambda_u(n)| \ll_\delta M_0T^{2\delta}+M_0T^{3\delta}\log(2T) \ll_\varepsilon M_0T^{4\delta}=M_0T^\varepsilon.

Now the range in Lemma 5.1 is already sufficient for applying Lemma 3.3. Indeed, whenever Am,n≠0A_{m,n}\ne0, we have n<Y=2T1+β−α=o(T)n<Y=2T^{1+\beta-\alpha}=o(T), so 2πbn≤T/22\pi bn\le T/2 for sufficiently large TT. Inserting the expansion from Lemma 3.3 into the identity in Lemma 5.1 therefore produces the factors (n/m)±iT(n/m)^{\pm iT} in the sum. Their form provides the link to LL-functions: since (n/m)ωiT=m−ωiTnωiT(n/m)^{\omega iT}=m^{-\omega iT}n^{\omega iT}, Mellin inversion of the remaining weight separates the two summation variables and leads to two LL-functions with shifts ±iT\pm iT. We carry out this step in § 6. The following proposition records the sums obtained by substituting the expansion from Lemma 3.3 into the identity in Lemma 5.1 and bounds the resulting error.

Proposition 5.3. For ω∈{−1,1}\omega\in\{-1,1\}, define

Hω=1TLu∑m,n≥1λu(m)λu(n)(n/m)ωiTAm,nWψ(ωn2−m2T).(21)\mathcal{H}_{\omega} =\frac{1}{TL_u}\sum_{m,n\ge1}\lambda_u(m)\lambda_u(n)(n/m)^{\omega iT}A_{m,n}W_\psi\left(\omega\frac{n^2-m^2}{T}\right). \tag*{(21)}

Then, for every ε>0\varepsilon> 0,

Olong=4π∑ω=±1Hω+Oψ,α,β,ε(T−1+β+ε+T−2α+4β+ε+T−2).\mathcal{O}_{\mathrm{long}} = 4\pi\sum_{\omega=\pm1} \mathcal{H}_{\omega} + O_{\psi,\alpha,\beta,\varepsilon}\left(T^{-1+\beta+\varepsilon} + T^{-2\alpha+4\beta+\varepsilon} + T^{-2}\right).

Proof. Take B=2B=2 in Lemma 5.1. Whenever Am,n≠0A_{m,n}\ne0, the support conditions in that lemma give n<Y=O(T)n<Y=O(T), so Lemma 3.3 applies for sufficiently large TT. Substituting the expansion from Lemma 3.3 into the identity in Lemma 5.1 gives

Olong=4πTLu∑ω=±1∑m<nλu(m)λu(n)(n/m)ωiTAm,nWψ(ωn2−m2T)\mathcal{O}_{\mathrm{long}}=\frac{4\pi}{TL_u}\sum_{\omega=\pm1}\sum_{m<n}\lambda_u(m)\lambda_u(n)(n/m)^{\omega iT}A_{m,n}W_\psi\left(\omega\frac{n^2-m^2}{T}\right)
+Oψ(E1+E2+E3)+Oψ,α,β(T−2),\qquad+O_\psi(E_1+E_2+E_3)+O_{\psi,\alpha,\beta}(T^{-2}),

where

E1=1T2Lu∑m<n∣λu(m)λu(n)∣Am,n,E_1=\frac{1}{T^2L_u}\sum_{m<n}|\lambda_u(m)\lambda_u(n)|A_{m,n},
E2=1T3Lu∑m<n∣λu(m)λu(n)∣Am,n(m2+n2),E_2=\frac{1}{T^3L_u}\sum_{m<n}|\lambda_u(m)\lambda_u(n)|A_{m,n}(m^2+n^2),
E3=1T4Lu∑m<n∣λu(m)λu(n)∣Am,n(m2+n2)(n2−m2).E_3=\frac{1}{T^4L_u}\sum_{m<n}|\lambda_u(m)\lambda_u(n)|A_{m,n}(m^2+n^2)(n^2-m^2).

The quantities E1,E2,E3E_1,E_2,E_3 bound the contributions of the three terms in the error in Lemma 3.3; the last O(T−2)O(T^{-2}) term comes from Lemma 5.1. Since Am,n=0A_{m,n}=0 for n≤mn\le m, the first term in the decomposition is exactly 4π∑ω=±1Hω4\pi\sum_{\omega=\pm1}\mathcal{H}_\omega by (5.1).

To estimate E1,E2,E3E_1,E_2,E_3, we first bound the sum of ∣λu(m)λu(n)∣Am,n|\lambda_u(m)\lambda_u(n)|A_{m,n}. For sufficiently large TT, the parameter conditions in Lemma 5.1 give

1≤Y≤M=T1+β≤T2.1\le Y\le M=T^{1+\beta}\le T^2.

Thus Lemma 5.2 applies with Y0=YY_0=Y and M0=MM_0=M. Using 0≤Am,n≤10\le A_{m,n}\le1 and the support conditions in Lemma 5.1, we apply Lemma 5.2 and (2.3), each with exponent ε/2\varepsilon/2, to obtain

1Lu∑m<n∣λu(m)λu(n)∣Am,n≪εTε/2∑1≤m<n≤Yn2−m2≤2M∣λu(m)λu(n)∣≪εMTε.\frac{1}{L_u}\sum_{m<n}|\lambda_u(m)\lambda_u(n)|A_{m,n} \ll_\varepsilon T^{\varepsilon/2} \sum_{\substack{1\le m<n\le Y\\ n^2-m^2\le2M}}|\lambda_u(m)\lambda_u(n)| \ll_\varepsilon MT^\varepsilon.

It follows directly that

E1≪εMTεT2=T−1+β+ε.E_1\ll_\varepsilon\frac{MT^\varepsilon}{T^2}=T^{-1+\beta+\varepsilon}.

For E2E_2, we use m2+n2≤(m+n)2<Y2m^2+n^2\le(m+n)^2<Y^2 whenever Am,n≠0A_{m,n}\ne0. The same coefficient estimate therefore gives

E2≪εY2T3MTε≪T−2α+3β+ε.E_2\ll_\varepsilon\frac{Y^2}{T^3}MT^\varepsilon\ll T^{-2\alpha+3\beta+\varepsilon}.

For E3E_3, we also have n2−m2<2Mn^2-m^2<2M, so

E3≪εMY2T4MTε≪T−2α+4β+ε.E_3\ll_\varepsilon\frac{MY^2}{T^4}MT^\varepsilon\ll T^{-2\alpha+4\beta+\varepsilon}.

Since β>0\beta>0, we have T−2α+3β+ε≤T−2α+4β+εT^{-2\alpha+3\beta+\varepsilon}\le T^{-2\alpha+4\beta+\varepsilon}. Substituting these three estimates into the initial decomposition, together with its Oψ,α,β(T−2)O_{\psi,\alpha,\beta}(T^{-2}) term, proves the proposition. □

Estimation of Long Shifts

Throughout this section, ω∈{−1,1}\omega\in\{-1,1\} is fixed, and α,β,R,M,Y\alpha,\beta,R,M,Y are as in Lemma 5.1.

In what follows, dyadic parameters run over 1,2,4,…1,2,4,\ldots. Choose a fixed nonnegative function w∈Cc∞((1/2,2))w \in C^\infty_c((1/2,2)), extended by zero to R\mathbb{R}, satisfying

∑N dyadicw(t/N)=1(t≥1).\sum_{N\ \mathrm{dyadic}} w(t/N)=1 \qquad(t\ge1).

Since Am,n=0A_{m,n}=0 for n≤mn\le m, we insert this partition for each of the positive integers mm, nn, and n−mn-m in (5.1). This gives

Hω=1TLu∑P,Q,H dyadic∑m<nλu(m)λu(n)(n/m)ωiT×w(m/P)w(n/Q)w((n−m)/H)Am,nWψ(ωn2−m2T)=∑P,Q,H dyadicUω(P,Q,H).(22)\begin{aligned} \mathcal{H}_{\omega} &=\frac{1}{TL_u}\sum_{P,Q,H\ \mathrm{dyadic}}\sum_{m<n}\lambda_u(m)\lambda_u(n)(n/m)^{\omega iT}\\ &\quad{}\times w(m/P)w(n/Q)w((n-m)/H)A_{m,n}W_\psi\left(\omega\frac{n^2-m^2}{T}\right)\\ &=\sum_{P,Q,H\ \mathrm{dyadic}}\mathcal{U}_{\omega}(P,Q,H). \tag*{(22)} \end{aligned}

where

Uω(P,Q,H)=1TLu∑m,n≥1λu(m)λu(n)(n/m)ωiTVP,Q,H(ω)(m/P,n/Q),(23)\mathcal{U}_{\omega}(P,Q,H)=\frac{1}{TL_u}\sum_{m,n\ge1}\lambda_u(m)\lambda_u(n)(n/m)^{\omega iT}V_{P,Q,H}^{(\omega)}(m/P,n/Q), \tag*{(23)}

with

VP,Q,H(ω)(x,y)=w(x)w(y)w(Qy−PxH)(1−χ(Qy−PxR))χ(Q2y2−P2x2M)Wψ(ωQ2y2−P2x2T).(24)V_{P,Q,H}^{(\omega)}(x,y)=w(x)w(y)w\left(\frac{Qy-Px}{H}\right)\left(1-\chi\left(\frac{Qy-Px}{R}\right)\right)\chi\left(\frac{Q^2y^2-P^2x^2}{M}\right)W_\psi\left(\omega\frac{Q^2y^2-P^2x^2}{T}\right). \tag*{(24)}

Indeed, at (x,y)=(m/P,n/Q)(x,y)=(m/P,n/Q) the last three factors in (6.3) give Am,nWψ(ω(n2−m2)/T)A_{m,n}W_\psi(\omega(n^2-m^2)/T) for m<nm<n, while the first three give w(m/P)w(n/Q)w((n−m)/H)w(m/P)w(n/Q)w((n-m)/H). The factor w((n−m)/H)w((n-m)/H) vanishes for n≤mn\le m, so all index restrictions in (6.2) are contained in the weight.

By Lemma 5.1, a nonzero summand has m,n<Ym,n<Y and R<n−m<YR<n-m<Y. Since ww is supported in (1/2,2)(1/2,2), a triple can contribute to (6.1) only if

1≤P,Q<2Y,R/2<H<2Y.(25)1\le P,Q<2Y,\qquad R/2<H<2Y. \tag*{(25)}

There are Oα,β((log⁡T)3)O_{\alpha,\beta}((\log T)^3) such triples.

To estimate a single sum Uω(P,Q,H)\mathcal{U}_{\omega}(P,Q,H), we use Mellin inversion. Write V=VP,Q,H(ω)V=V_{P,Q,H}^{(\omega)} and

V~(s,z)=∫0∞∫0∞V(x,y)xs−1yz−1 dx dy.\widetilde{V}(s,z)=\int_0^\infty\int_0^\infty V(x,y)x^{s-1}y^{z-1}\,dx\,dy.

Lemma 6.1. For every fixed ν≥0\nu\ge0, uniformly in T,R,M,P,Q,H>0T,R,M,P,Q,H>0 and ω∈{±1}\omega\in\{\pm1\},

∫R2∣V~(12+iτ1,12+iτ2)∣(1+∣τ1∣+∣τ2∣)ν dτ1 dτ2≪ψ,ν(1+P/H)ν.\int_{\mathbb{R}^2}\left|\widetilde{V}\left(\frac12+i\tau_1,\frac12+i\tau_2\right)\right|(1+|\tau_1|+|\tau_2|)^\nu\,d\tau_1\,d\tau_2\ll_{\psi,\nu}(1+P/H)^\nu.

Proof. Put h=min⁡{1,H/P}h=\min\{1,H/P\} and introduce the commuting operators

D=x∂x+y∂y,E=hy∂y.(26)D=x\partial_x+y\partial_y,\qquad E=hy\partial_y. \tag*{(26)}

The parameters are held fixed when differentiating. The support of VV, and of each DjEkVD^{j}E^{k}V, lies in

Ω={(x,y)∈[1/2,2]2:H/2≤Qy−Px≤2H}.\Omega=\{(x,y)\in[1/2,2]^2:H/2\le Qy-Px\le2H\}.

For fixed yy, the allowed xx-interval has length at most min⁡{3/2,3H/(2P)}\min\{3/2,3H/(2P)\}, and hence ∣Ω∣≤9h/4|\Omega|\le9h/4.

We first show that

∥DjEkV∥∞≪ψ,j,k1(j,k≥0),(27)\lVert D^{j}E^{k}V\rVert_{\infty}\ll_{\psi,j,k}1\qquad(j,k\ge0), \tag*{(27)}

uniformly in all the parameters. On Ω\Omega,

hQyQy−Px=h(1+PxQy−Px)≤5,hQ2y2Q2y2−P2x2=hQyQy−PxQyQy+Px≤5.\frac{hQy}{Qy-Px}=h\left(1+\frac{Px}{Qy-Px}\right)\le5,\qquad \frac{hQ^{2}y^{2}}{Q^{2}y^{2}-P^{2}x^{2}}=\frac{hQy}{Qy-Px}\frac{Qy}{Qy+Px}\le5.

For j≥0j\ge0 and k≥1k\ge1 we have

Dj(Qy−Px)=Qy−Px,DjEk(Qy−Px)=hkQy,D^{j}(Qy-Px)=Qy-Px,\qquad D^{j}E^{k}(Qy-Px)=h^{k}Qy,
Dj(Q2y2−P2x2)=2j(Q2y2−P2x2),DjEk(Q2y2−P2x2)=2j+khkQ2y2.D^{j}(Q^{2}y^{2}-P^{2}x^{2})=2^{j}(Q^{2}y^{2}-P^{2}x^{2}),\qquad D^{j}E^{k}(Q^{2}y^{2}-P^{2}x^{2})=2^{j+k}h^{k}Q^{2}y^{2}.

Since h≤1h\le1, these identities give ∣DjEkt∣≪j,k∣t∣|D^{j}E^{k}t|\ll_{j,k}|t| for each of the arguments

t=Qy−PxH,Qy−PxR,Q2y2−P2x2M,Q2y2−P2x2T.t=\frac{Qy-Px}{H},\qquad\frac{Qy-Px}{R},\qquad \frac{Q^{2}y^{2}-P^{2}x^{2}}{M},\qquad \frac{Q^{2}y^{2}-P^{2}x^{2}}{T}.

The chain rule therefore bounds every mixed derivative of a factor F(t)F(t) by a finite sum of ∣t∣ℓ∣F(ℓ)(t)∣|t|^{\ell}|F^{(\ell)}(t)|. These quantities are bounded for F=WψF=W_{\psi} because WψW_{\psi} is Schwartz, and for F=wF=w or either cutoff involving χ\chi because their derivatives of order at least one have fixed compact support. The derivatives of w(x)w(x) and w(y)w(y) are also uniformly bounded, since x,y∈[1/2,2]x,y\in[1/2,2] and h≤1h\le1. Leibniz’s rule proves (6.5).

Repeated integration by parts, with no boundary terms because of compact support, gives

(−s−z)A(1+hz)BV~(s,z)=∫0∞∫0∞DA(1−E)BV(x,y)xs−1yz−1 dx dy(-s-z)^{A}(1+hz)^{B}\widetilde{V}(s,z) =\int_{0}^{\infty}\int_{0}^{\infty}D^{A}(1-E)^{B}V(x,y)x^{s-1}y^{z-1}\,dx\,dy

for integers A,B≥0A,B\ge0. On s=1/2+iτ1s=1/2+i\tau_{1}, z=1/2+iτ2z=1/2+i\tau_{2}, the integral is Oψ,A,B(h)O_{\psi,A,B}(h) by (6.5) and ∣Ω∣≪h|\Omega|\ll h. Moreover,

∣s+z∣≍1+∣τ1+τ2∣,∣1+hz∣≍1+h∣τ2∣.|s+z|\asymp1+|\tau_{1}+\tau_{2}|,\qquad|1+hz|\asymp1+h|\tau_{2}|.

Thus

∣V~(12+iτ1,12+iτ2)∣≪ψ,A,Bh(1+∣τ1+τ2∣)A(1+h∣τ2∣)B.(28)\left|\widetilde{V}\left(\frac{1}{2}+i\tau_{1},\frac{1}{2}+i\tau_{2}\right)\right| \ll_{\psi,A,B} \frac{h}{(1+|\tau_{1}+\tau_{2}|)^{A}(1+h|\tau_{2}|)^{B}}. \tag*{(28)}

Choose integers A,B>ν+1A,B>\nu+1 and set η1=τ1+τ2\eta_{1}=\tau_{1}+\tau_{2}, η2=hτ2\eta_{2}=h\tau_{2}. Then dτ1 dτ2=h−1dη1 dη2d\tau_{1}\,d\tau_{2}=h^{-1}d\eta_{1}\,d\eta_{2} and

1+∣τ1∣+∣τ2∣≤2h−1(1+∣η1∣)(1+∣η2∣).1+|\tau_{1}|+|\tau_{2}|\le2h^{-1}(1+|\eta_{1}|)(1+|\eta_{2}|).

Integrating (6.6) with the required weight yields

∫R2∣V~(12+iτ1,12+iτ2)∣(1+∣τ1∣+∣τ2∣)ν dτ1 dτ2\int_{\mathbb{R}^{2}}\left|\widetilde{V}\left(\frac{1}{2}+i\tau_{1},\frac{1}{2}+i\tau_{2}\right)\right| (1+|\tau_{1}|+|\tau_{2}|)^{\nu}\,d\tau_{1}\,d\tau_{2}
≪ψ,A,B,νh−ν∫R(1+∣η1∣)ν−A dη1∫R(1+∣η2∣)ν−B dη2\ll_{\psi,A,B,\nu}h^{-\nu} \int_{\mathbb{R}}(1+|\eta_{1}|)^{\nu-A}\,d\eta_{1} \int_{\mathbb{R}}(1+|\eta_{2}|)^{\nu-B}\,d\eta_{2}
≪ψ,νh−ν≤(1+P/H)ν.\ll_{\psi,\nu}h^{-\nu}\le(1+P/H)^{\nu}.

We now apply this estimate to the dyadic sums and complete the bound for Hω\mathcal{H}_{\omega}.

Proposition 6.2. Assume GLH. For every ε>0\varepsilon> 0,

Hω≪ψ,α,β,εT−α+β+ε(ω=±1).(29)\mathcal{H}_{\omega} \ll_{\psi,\alpha,\beta,\varepsilon} T^{-\alpha+\beta+\varepsilon} \qquad(\omega= \pm1). \tag*{(29)}

Proof. Fix a block in (22) and abbreviate V=VP,Q,H(ω)V = V^{(\omega)}_{P,Q,H}. Mellin inversion in both variables (one may see [12 Section 3]), applied to (23), gives

Uω(P,Q,H)=1TLu(2πi)2∫(2)∫(2)V~(s,z)PsQz\mathcal{U}_{\omega}(P,Q,H) = \frac{1}{T L_u (2\pi i)^2} \int_{(2)} \int_{(2)} \widetilde{V}(s,z)P^sQ^z
×L(s+ωiT,u)L(z−ωiT,u) dz ds.\times L(s+\omega iT,u)L(z-\omega iT,u)\, dz\, ds.

The two Dirichlet series arise from (n/m)ωiTm−sn−z=m−(s+ωiT)n−(z−ωiT)(n/m)^{\omega iT}m^{-s}n^{-z}=m^{-(s+\omega iT)}n^{-(z-\omega iT)}. For fixed parameters, V~\widetilde{V} decays faster than any power of 1+∣ℑs∣+∣ℑz∣1+|\Im s|+|\Im z|, uniformly for 1/2≤Re⁡s,Re⁡z≤21/2\leq\operatorname{Re}s,\operatorname{Re}z\leq2. Both LL-functions are entire and have polynomial growth in these strips. We may therefore move one contour at a time to real part 1/21/2, without crossing a pole.

Write s=1/2+iτ1s=1/2+i\tau_1 and z=1/2+iτ2z=1/2+i\tau_2. Since 1+T+∣τj±T∣≪T(1+∣τj∣)1+T+|\tau_j\pm T|\ll T(1+|\tau_j|), GLH gives, for every fixed ν>0\nu>0,

∣L(12+i(τ1+ωT),u)L(12+i(τ2−ωT),u)∣≪νT2ν(1+∣τ1∣+∣τ2∣)2ν.\left|L\left(\frac{1}{2}+i(\tau_1+\omega T),u\right)L\left(\frac{1}{2}+i(\tau_2-\omega T),u\right)\right| \ll_{\nu} T^{2\nu}(1+|\tau_1|+|\tau_2|)^{2\nu}.

As ∣PsQz∣=PQ|P^sQ^z|=\sqrt{PQ}, Lemma 6.1 with exponent 2ν2\nu yields

∣Uω(P,Q,H)∣≪νPQTLuT2ν∫R2∣V~(12+iτ1,12+iτ2)∣(1+∣τ1∣+∣τ2∣)2ν dτ1 dτ2|\mathcal{U}_{\omega}(P,Q,H)|\ll_{\nu}\frac{\sqrt{PQ}}{T L_u}T^{2\nu}\int_{\mathbb{R}^2}\left|\widetilde{V}\left(\frac{1}{2}+i\tau_1,\frac{1}{2}+i\tau_2\right)\right|(1+|\tau_1|+|\tau_2|)^{2\nu}\,d\tau_1\,d\tau_2
≪ψ,νPQTLuT2ν(1+P/H)2ν.\ll_{\psi,\nu}\frac{\sqrt{PQ}}{T L_u}T^{2\nu}(1+P/H)^{2\nu}.

For the triples in (25),

PQT≪YT=2T−α+β,PH≪YR=2T1+β−2α≪T.\frac{\sqrt{PQ}}{T}\ll\frac{Y}{T}=2T^{-\alpha+\beta},\qquad\frac{P}{H}\ll\frac{Y}{R}=2T^{1+\beta-2\alpha}\ll T.

Together with Lu−1≪δTδL_u^{-1}\ll_{\delta}T^{\delta}, this gives

Uω(P,Q,H)≪ψ,α,β,δ,νPQTTδ+4ν.\mathcal{U}_{\omega}(P,Q,H)\ll_{\psi,\alpha,\beta,\delta,\nu}\frac{\sqrt{PQ}}{T}T^{\delta+4\nu}.

For every ε>0\varepsilon>0, choosing δ,ν>0\delta,\nu>0 with δ+4ν<ε\delta+4\nu<\varepsilon therefore gives

Uω(P,Q,H)≪ψ,α,β,εPQTTε.(30)\mathcal{U}_{\omega}(P,Q,H)\ll_{\psi,\alpha,\beta,\varepsilon}\frac{\sqrt{PQ}}{T}T^{\varepsilon}. \tag*{(30)}

uniformly for the triples in (25). We now apply (30) with exponent ε/2\varepsilon/2 and sum over the Oα,β((log⁡T)3)O_{\alpha,\beta}((\log T)^3) triples in (22). Using PQ≪Y\sqrt{PQ}\ll Y and (log⁡(2+T))3≪εTε/2(\log(2+T))^3\ll_{\varepsilon}T^{\varepsilon/2}, we obtain

Hω≪ψ,α,β,εYTTε/2(log⁡(2+T))3\mathcal{H}_{\omega}\ll_{\psi,\alpha,\beta,\varepsilon}\frac{Y}{T}T^{\varepsilon/2}(\log(2+T))^3
≪ψ,α,β,εT−α+β+ε.\ll_{\psi,\alpha,\beta,\varepsilon}T^{-\alpha+\beta+\varepsilon}.

This proves (29). □

Proof of Theorem 1.2 and Corollary 1.3

Proof of Theorem 1.2. Combining (16) and (19) with Lemmas 4.3, 4.4, and 4.5, and Propositions 5.3 and 6.2 gives, for every fixed η,σ>0\eta,\sigma>0,

Iu(ψ)−M(ψ)≪ψ,α,β,η,σT−1/2+2η+α+T−1+2α+σ+T−α+β+σ+T−1+β+σ+T−2α+4β+σ+T−2.(31)I_u(\psi)-M(\psi)\ll_{\psi,\alpha,\beta,\eta,\sigma}T^{-1/2+2\eta+\alpha}+T^{-1+2\alpha+\sigma}+T^{-\alpha+\beta+\sigma}+T^{-1+\beta+\sigma}+T^{-2\alpha+4\beta+\sigma}+T^{-2}. \tag*{(31)}

Take α=1/4\alpha=1/4 and β=η=σ=ε\beta=\eta=\sigma=\varepsilon, where 0<ε<1/160<\varepsilon<1/16. These choices satisfy the parameter conditions in Lemma 5.1, and (31) gives the following estimate, which proves (3) after relabeling 2ε2\varepsilon as ε\varepsilon:

Iu(ψ)−M(ψ)≪ψ,εT−1/4+2ε.I_u(\psi)-M(\psi)\ll_{\psi,\varepsilon}T^{-1/4+2\varepsilon}.

Proof of Corollary 1.3. Write A=log⁡aA=\log a, B=log⁡bB=\log b, and fix δ>0\delta>0. Choose d>0d>0 with d<(B−A)/4d<(B-A)/4 and 4d<δ4d<\delta. Let ηd∈C∞((−d,d))\eta_d\in C^\infty((-d,d)) be nonnegative with integral one, and set

ϕ−=1[A+d,B−d]∗ηd,ϕ+=1[A−d,B+d]∗ηd,ψ±(y)=ϕ±(log⁡y).\phi_-=1_{[A+d,B-d]}*\eta_d,\qquad\phi_+=1_{[A-d,B+d]}*\eta_d,\qquad\psi_\pm(y)=\phi_\pm(\log y).

These are smooth compactly supported functions on (0,∞)(0,\infty). Convolution with the nonnegative kernel gives

0≤ψ−≤1[a,b]≤ψ+,∫0∞(ψ+−ψ−)dyy=4d<δ.0\leq\psi_-\leq1_{[a,b]}\leq\psi_+,\qquad\int_0^\infty(\psi_+-\psi_-)\frac{dy}{y}=4d<\delta.

The pointwise inequalities also hold at the endpoints, which have zero measure in all the integrals. Positivity of ∣u(iy)∣2|u(iy)|^2 bounds the interval integral between Iu(ψ−)I_u(\psi_-) and Iu(ψ+)I_u(\psi_+). For these fixed functions, let TT tend to infinity using Theorem 1.2 with ε=1/12\varepsilon=1/12. Their limiting integrals approach (6/π)(B−A)(6/\pi)(B-A) as d↓0d\downarrow0. This proves (4). □\square

References

School of Mathematics and Statistics, Shandong University, Weihai, Weihai, Shandong 264209, China

State Key Laboratory of Cryptography and Digital Economy Security, Shandong University, Weihai, 264209, China

Email address: [email protected]

School of Mathematics and Statistics, Shandong University, Weihai, Weihai, Shandong 264209, China

Email address: [email protected]

References

  1. [1]N. Andersen and W. D. Duke, Asymptotic distribution of traces of singular moduli, Discrete Anal. 2022, Paper No. 4, 14 pp.; MR4450655
  2. [2]P. B. Garrett, Modern analysis of automorphic forms by example. Vol. 1, Cambridge Studies in Advanced Mathematics, 173, Cambridge Univ. Press, Cambridge, 2018; MR3837525DOI
  3. [3]A. Ghosh, A. Reznikov and P. C. Sarnak, Nodal domains of Maass forms I, Geom. Funct. Anal. 23 (2013), no. 5, 1515–1568; MR2102912DOI
  4. [4]S. Jang and J. Jung, Quantum unique ergodicity and the number of nodal domains of eigenfunctions, J. Amer. Math. Soc. 31 (2018), no. 2, 303–318; MR3758146arxiv.org/abs/1505.02548
  5. [5]J. Jung, Quantitative quantum ergodicity and the nodal domains of Hecke–Maass cusp forms, Comm. Math. Phys. 348 (2016), no. 2, 603–653; MR3554896
  6. [6]J. E. Hoffstein and P. Lockhart, Coefficients of Maass forms and the Siegel zero, Ann. of Math. (2) 140 (1994), no. 1, 161–181.DOI
  7. [7]P. Humphries, Restricted arithmetic quantum unique ergodicity, lecture slides, Shanks Conference, Vanderbilt University, May 15, 2024. Slides.
  8. [8]H. H. Kim, Functionality for the exterior square of GL₄ and the symmetric fourth of GL₂, J. Amer. Math. Soc. 16 (2003), no. 1, 139–183; with Appendix 1 by Dinakar Ramakrishnan and Appendix 2 by Kim and Peter Sarnak. MR1937203DOI
  9. [9]X. Li, Upper bounds on L-functions at the edge of the critical strip, Int. Math. Res. Not. IMRN 2010, no. 4, 727–755; MR2595006arxiv.org/abs/0904.0850
  10. [10]E. Lindenstrauss, Invariant measures and arithmetic quantum unique ergodicity, Ann. of Math. (2) 163 (2006), no. 1, 165–219; MR2195133DOI
  11. [11]W. Luo and P. C. Sarnak, Quantum variance for Hecke eigenforms, Ann. Sci. École Norm. Sup. (4) 37 (2004), no. 5, 769–799; MR2103474
  12. [12]B. W. Rodgers and K. Soundararajan, The variance of divisor sums in arithmetic progressions, Forum Math. 30 (2018), no. 2, 269–293; MR3769996DOI
  13. [13]K. Soundararajan, Quantum unique ergodicity for SL₂(Z)\H, Ann. of Math. (2) 172 (2010), no. 2, 1529–1538; MR2680500arxiv.org/abs/0901.4060
  14. [14]Q. Sun and Q. Zhang, Mass distribution for holomorphic cusp forms on the vertical geodesic, J. Number Theory 279 (2026), 827–857; MR4951221arxiv.org/abs/2408.15259
  15. [15]Q. Sun and Q. Zhang, Quantum variance for Hecke–Maass cusp forms on the vertical geodesic, preprint, 2026. https://arxiv.org/abs/2610.27454.
  16. [16]T. C. Watson, Rankin triple products and quantum chaos, ProQuest LLC, Ann Arbor, MI, 2002; MR2703041
  17. [17]M. P. Young, The quantum unique ergodicity conjecture for thin sets, Adv. Math. 286 (2016), 958–1016; MR3415701DOI
  18. [18]M. P. Young, Equidistribution of Eisenstein series on geodesic segments, Adv. Math. 340 (2018), 1166–1218; MR3886191DOI
  19. [19]P. Zenz, Quantum variance for holomorphic Hecke cusp forms on the vertical geodesic, Adv. Math. 505 (2026), Paper No. 111316; MR5140740arxiv.org/abs/2111.04713

Paper details

vlnc:W_swypj9ny
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents