---
title: Relative Trace Equality Overview
url: https://www.emergentmind.com/topics/relative-trace-equality
type: topic
---

# Relative Trace Equality Overview

Searching arXiv for the cited papers to ground the article in the literature.
Relative trace equality denotes an exact identity obtained by evaluating a kernel or integral operator in two different ways after inserting subgroup periods, characters, or both. In the settings considered in the literature, the resulting equality typically compares a geometric expansion built from relative orbital integrals, Kloosterman sums, ortholength data, or transferred orbital distributions with a spectral expansion built from automorphic representations, period integrals, Fourier coefficients, and \(L\)-value weights. The notion is therefore broader than a single formula: it includes internal geometric–spectral identities for a given relative trace formula, twisted variants, and comparison identities between two different relative trace formulas [1406.5912], [1507.04419].

## 1. Relative trace equality as a relative analogue of the trace formula

The basic structural distinction from the ordinary trace formula is that the relevant symmetry is not ordinary conjugacy on a group and not integration over the diagonal. In the general simple setting of \(H\le G\times G\), with \(G\) a connected reductive algebraic group over a number field \(F\) and \(H\) a connected algebraic subgroup of \(G\times G\) that is the direct product of a reductive group and a unipotent group, the action is
\[
(h_\ell,h_r),\,g \mapsto h_\ell\,g\,h_r^{-1}.
\]
A quasi-character
\[
X:H(\mathbb A_F)\to \mathbb C^\times
\]
trivial on \(A_G\) and on \(H(F)\) defines the relative period
\[
P_X(\varphi) := \int_{A_{G,H}H(F)\backslash H(\mathbb A_F)} X(h_\ell,h_r)\,\varphi(h_\ell,h_r)\,d(h_\ell,h_r),
\]
and for a test function \(f\) the kernel \(K_f\) admits a spectral expansion over cuspidal automorphic representations. The corresponding relative trace equality is
\[
\sum_{[y]} T(H_y)\,RO_{X,y}(f) = \sum_{\pi} rtr_{\pi}(f),
\]
where the left-hand side is a sum over relevant elliptic unimodular closed geometric classes and the right-hand side is a sum of relative traces \(rtr_\pi(f)=P_X(K_\pi(f))\) [1406.5912].

This formulation makes explicit why the equality is called relative. The geometric side is a sum of relative orbital integrals \(RO_{X,y}(f)\), integrating over stabilizer quotients rather than conjugacy classes. The spectral side is not a plain trace \(\mathrm{tr}\,\pi(f)\), but a period of the spectral kernel. In the Kuznetsov setting the same principle appears in a different guise: instead of integrating over the diagonal as in the Selberg trace formula, one integrates the kernel against a pair of unipotent characters, producing a relative trace formula whose spectral side involves automorphic Fourier coefficients and \(L\)-values and whose geometric side involves Kloosterman sums and an archimedean Bessel transform [1912.08137].

## 2. Twisted relative trace equality

A particularly explicit formulation appears in the simple twisted relative trace formula. The setting begins with a number field \(F\), a quadratic or trivial extension \(M/F\), a connected reductive \(F\)-group \(G_0\), and an involution \(\sigma\) of \(G_0\) of order \(2\). One forms
\[
G := \mathrm{Res}_{M/F} G_0,
\]
lets \(T\) denote the induced Galois involution on \(G\), and sets
\[
\theta := \sigma\circ T.
\]
The fixed-point subgroups are \(G^\sigma\) and \(G^{\sigma\circ T}\), and the relative trace formula is studied on the pair
\[
(G^\sigma,\, G^{\sigma\circ T})
\]
inside \(G\) [1507.04419].

The relevant orbit relation is twisted. For \(y\in G(k)\),
\[
B_\theta(y)=y\,y^{-\theta},
\]
and the relative double coset space is
\[
T_\theta(R):=G^\sigma(R)\backslash G(R)/G^{\sigma\circ T}(R).
\]
An element is relatively \(\theta\)-semisimple when \(y y^{-\theta}\) is semisimple in \(G\), and strongly relatively \(\theta\)-regular when the \(\theta\)-centralizer
\[
C_{G^\sigma, y^\theta} = \{x\in G^\sigma : x^{-1}y x^\theta = y\}
\]
is a torus. The paper states that if \(y\) is relatively \(\theta\)-semisimple, then this centralizer is reductive; if it is a torus, then \(y\) is strongly relatively \(\theta\)-regular [1507.04419].

The geometric side is built from twisted relative orbital integrals. For a unitary character \(\mu=\mu_1\times \mu_2\) on
\[
G^\sigma(\mathbb A_F)\times G^{\sigma\circ T}(\mathbb A_F),
\]
trivial on rational points, the global twisted relative orbital integral \(\mathrm{TRO}_y^\mu(f)\) integrates \(f\) over the \(G^\sigma\times G^{\sigma\circ T}\)-orbit of \(y\), twisted by \(\mu\). The relevance condition is that \(\mu\) be trivial on the connected component \((G_y^\sigma)^\circ\). Only then can the orbital integral contribute [1507.04419].

The spectral side is built from cusp forms on
\[
L^2\bigl(G(F)\backslash G(\mathbb A_F)\bigr)
\]
restricted to the Harish-Chandra subgroup \(1_G(\mathbb A_F)=\ker(\mathrm{HC}_G)\). For an automorphic representation \(T\), the two relative periods are
\[
P_{\mu_1}(\phi) = \int_{G^\sigma(F)\backslash G^\sigma(\mathbb A_F)} \mu_1(x)\phi(x)\,dx,
\qquad
P_{\mu_2}(\phi) = \int_{G^{\sigma\circ T}(F)\backslash G^{\sigma\circ T}(\mathbb A_F)} \mu_2(y)\phi(y)\,dy.
\]
If \(f^1\) is \(K\)-finite, the spectral twisted relative trace becomes a sum of products
\[
\sum_{\phi\in B_T} P_{\mu_1}(T(f^1)\phi)\,P_{\mu_2}(\phi)
\]
over an orthonormal basis \(B_T\) of the \(T\)-isotypic component [1507.04419].

The main theorem states that, assuming \(G^\sigma\) is connected and that there exist places \(v_1,v_2,v_3\) such that \(f_{v_1}\) is supported on relatively \(\theta\)-elliptic elements, \(f_{v_2}\) is supported on strongly relatively \(\theta\)-regular elements, and \(f_{v_3}\) is \(F\)-supercuspidal, the elliptic part of the twisted relative orbital sum equals the spectral sum over automorphic representations occurring in the cuspidal \(L^2\)-space. In the paper’s summary, under suitable ellipticity, regularity, and supercuspidality assumptions, the elliptic part of a \(\theta\)-twisted relative orbital sum equals the spectral sum of products of \((G^\sigma,\mu_1)\)- and \((G^{\sigma\circ T},\mu_2)\)-periods over automorphic representations; this is the “simple twisted relative trace formula” [1507.04419].

## 3. Comparison identities and transfer operators

Relative trace equality also occurs as an identity between two different relative trace formulas. For non-tempered split reductive spherical varieties \(X=G/H\), one studies the period
\[
{}_H(\phi):=\int_{[H]} \phi(h)\,dh,
\qquad [H]=H(k)\backslash H(\mathbb A_k),
\]
and the relative trace formula
\[
I(f)=\int_{[N]}\int_{[H]} K_f(h,n)\,\xi_L(n)\,dh\,dn,
\]
where \(N=N_LU\) is a maximal unipotent subgroup built from a parabolic \(Q=LU\). The comparison object is the Levi-side relative trace formula
\[
J(f')=\int_{[N_L]}\int_{[H_L]} K_{f'}(h,n)\,\xi_L(n)\,dh\,dn,
\]
for a tempered spherical subgroup \(H_L\subset L\). The explicit transfer map is
\[
{}_\chi(f)(l)=\chi(l)\int_{U(\mathbb A_k)}\int_{K_H} f(klu)\,dk\,du,
\]
and the main equality is
\[
I(f)=J({}_\chi(f))
\]
for all \(f\) that are locally \(L\)-supercuspidal [2512.03320].

The proof mechanism is orbit-by-orbit unfolding. Using
\[
G(k)=\bigsqcup_i Q(k)\gamma_i H(k),
\]
one writes \(I(f)=\sum_i I_i(f)\). There is a distinguished orbit \(i_0\) such that \(L_{i_0}=H_L\), while for all other orbits \(i\neq i_0\), either \(L_i\) is non-tempered, so Sakellaridis’ vanishing theorem kills it, or \(L_i\) is of parabolic induced type, so local \(L\)-supercuspidality kills it. The distinguished orbit gives exactly
\[
I_{i_0}(f)=J({}_\chi(f)).
\]
This places the equality in the framework of relative Langlands duality, where the Levi model is the “tempered support” or “cuspidal support” attached to the original non-tempered spherical variety [2512.03320].

A different comparison appears for \(G=PGL_2\), where the Jacquet relative trace formula for
\[
\mathcal Z=T\backslash G/T
\]
is compared with the Kuznetsov trace formula for
\[
\mathcal W=N\backslash G/N
\]
with non-standard test functions. The comparison is not by simple scalar transfer factors on individual orbital integrals, but by an explicit transfer operator
\[
|\bullet|\,\mathcal G:\mathcal S(\mathcal Z(A))\xrightarrow{\sim}\mathcal S(\mathcal W(A)),
\]
whose local version is built from Fourier transform and the involution
\[
\iota(f)(x)=\frac{\eta_v(x)}{|x|_v}f(1/x), \qquad \mathcal G=\mathcal F\circ \iota\circ \mathcal F.
\]
The global geometric identity is a Poisson summation formula on the base:
\[
(f)=(|\bullet|\,\mathcal G f).
\]
The paper emphasizes that this is a non-standard comparison and that the transfer operator, rather than pointwise matching of orbital integrals, is the mechanism behind the global trace equality [1402.3524].

## 4. Analytic and geometric incarnations

The Kuznetsov formula provides a canonical analytic number theory example of relative trace equality. For \(G=\mathrm{SL}_2(\mathbb R)\), \(\Gamma=\mathrm{SL}_2(\mathbb Z)\), and kernel
\[
K(x_1,x_2)=\sum_{\gamma\in\Gamma} f(x_1^{-1}\gamma x_2),
\]
the relative expression is
\[
\int_{\Gamma_N\backslash N(\mathbb R)}\int_{\Gamma_N\backslash N(\mathbb R)} K(n(x_1),n(x_2))\, e(-mx_1)\, e(nx_2)\, dx_1\,dx_2,
\]
for nonzero integers \(m,n\). On the spectral side this yields a sum over automorphic representations weighted by products of Fourier coefficients; on the geometric side, using the Bruhat decomposition of \(\mathrm{SL}_2(\mathbb Z)\), one obtains Kloosterman sums. The displayed Kuznetsov identity has the standard shape
\[
\sum_{\varphi\in\mathcal B} \frac{\rho_\varphi(n)\,\overline{\rho_\varphi(m)}}{L(1,\mathrm{Ad}^2\varphi)}\, h(t_\varphi)
\;+\;
\frac{1}{4\pi}\int_{-\infty}^{\infty} \frac{\rho_t(n)\,\overline{\rho_t(m)}}{|\zeta(1+2it)|^2}\, h(t)\,dt
=
\delta_{n,m}\,\text{(diagonal term)}
+
\sum_{c\ge 1}\frac{S(n,m;c)}{c}\, h^\pm\!\left(\frac{4\pi\sqrt{|nm|}}{c}\right),
\]
with the archimedean transform \(h\mapsto h^\pm\) described as the “infinite place analogue” of Kloosterman sums [1912.08137].

A geometric incarnation appears on a compact hyperbolic Riemann surface \(X=\Gamma\backslash\mathbb H\), relative to a primitive closed geodesic \(C\subset X\). The kernel
\[
K(z,w)=\sum_{\gamma\in\Gamma}\phi\big(u(\gamma z,w)\big)
\]
has spectral expansion
\[
K(z,w)=\sum_{n=0}^\infty h(r_n)\phi_n(z)\phi_n(w),
\]
and integrating over \(C\times C\) gives
\[
\int_C\int_C K(x,y)\,dx\,dy =\sum_{n=0}^\infty h(r_n)\,|P(\phi_n)|^2,
\qquad
P(\phi_n)=\int_C \phi_n(s)\,ds.
\]
The geometric expansion is indexed by double cosets \(\Gamma_0\backslash\Gamma/\Gamma_0\) and separates the identity term, exceptional terms corresponding to self-intersections of \(C\), and regular terms corresponding to orthogonal geodesic segments from \(C\) to itself. For regular non-exceptional terms,
\[
\delta(\gamma)=2\cosh(\ell_\gamma),
\]
where \(\ell_\gamma\) is the length of the unique shortest geodesic segment that starts and ends orthogonally on \(C\). The paper’s central interpretation is therefore
\[
\text{period spectrum} \longleftrightarrow \text{ortholength spectrum},
\]
a relative analogue of the classical Selberg trace formula [1504.05684].

## 5. Hypotheses, simplifications, and common misunderstandings

A recurrent feature of relative trace equalities is that they become “simple” only after imposing hypotheses that suppress continuous spectrum, non-elliptic orbits, or non-open geometric contributions. These assumptions are structural, not merely technical.

In the general simple relative trace formula, the theorem requires \(f\in C_c^\infty(A_GG(\mathbb A_F))\), that \(R(f)\) have cuspidal image, and that whenever the \(H(\mathbb A_F)\)-orbit of \(y\in G(F)\) intersects the support of \(f\), the point \(y\) be elliptic, unimodular, and closed. Under these assumptions the kernel integral
\[
\int_{A_{G,H}H(F)\backslash H(\mathbb A_F)} X(h_\ell,h_r)\,K_f(h_\ell,h_r)\,d(h_\ell,h_r)
\]
is absolutely convergent, and the geometric sum involves only finitely many contributing classes. The proof uses the Dixmier–Malliavin lemma, rapid decay of cuspidal kernels, and finiteness arguments from geometric invariant theory and Galois cohomology [1406.5912].

In the twisted formula, the support conditions at \(v_1\) and \(v_2\) isolate relatively \(\theta\)-elliptic and strongly relatively \(\theta\)-regular elements, while the \(F\)-supercuspidal hypothesis at \(v_3\) is used to force
\[
R(f^1)\bigl(L^2(G(F)\backslash G(\mathbb A_F))\bigr)\subseteq L^2_{\mathrm{cusp}}(G(F)\backslash G(\mathbb A_F)),
\]
thereby killing the continuous spectrum and making the spectral expansion purely discrete [1507.04419].

In the non-tempered spherical variety comparison, local \(L\)-supercuspidality is used to kill non-open orbits of parabolic induced type. The paper also states that in many models the local \(L\)-supercuspidal hypothesis is unnecessary, because all non-open orbit stabilizers are non-tempered and the vanishing follows from Sakellaridis’ theorem [2512.03320].

A common misunderstanding is to identify relative trace equality with pointwise matching of ordinary orbital integrals. The comparison for \(PGL_2\) shows that this is not the only mechanism: the two relative trace formulas are related by the explicit transform \( |\bullet|\,\mathcal G\), and the comparison proceeds through Poisson summation and spectral measure matching rather than scalar transfer factors on individual orbits [1402.3524].

## 6. Significance, applications, and conceptual scope

Relative trace equalities have been used to prove asymptotic formulas, comparison theorems, and period identities across several parts of automorphic forms and analytic number theory. In the general simple setting they yield a relative Weyl law:
\[
\sum_{\substack{\pi\ \text{cuspidal}\\ \lambda(\pi)\le X}}\;\sum_{\varphi\in B(\pi)}
\left|\int_{H_0(\mathbb Q)\backslash H_0(\mathbb A_\mathbb Q)} \varphi(h)\,dh\right|^2
\sim
a(G)\,\operatorname{meas}\bigl(H_0(\mathbb Q)\backslash H_0(\mathbb A_\mathbb Q)\bigr)\,X^{d/2},
\]
so automorphic representations are counted with weights given by squared \(H_0\)-periods [1406.5912].

For compact Riemann surfaces, the relative trace formula produces asymptotics for average squared periods,
\[
\sum_{\lambda_n\le x}|P(\phi_n)|^2 \sim \operatorname{len}(C)\,\frac{\sqrt{x}}{\pi},
\]
implying \(P(\phi_n)\neq 0\) for infinitely many \(n\), and also yields simultaneous nonvanishing results for twisted periods and asymptotics for the ortholength counting function [1504.05684].

In the \(PGL_2\) comparison, the global equality between the torus relative trace formula and a Kuznetsov formula with non-standard test functions gives a new proof of Waldspurger’s theorem on toric periods and recovers the Tunnell–Saito criterion through the action of an explicit involution on orbital integrals [1402.3524].

In analytic number theory, the Kuznetsov relative trace formula is presented as one of the main engines of the subject. The spectral–geometric bridge between Fourier coefficients and \(L\)-values on one side and Kloosterman sums and Bessel transforms on the other underlies applications to distribution of Hecke eigenvalues, large sieve inequalities, cancellation in Kloosterman sums, shifted convolution sums, moments and nonvanishing of \(L\)-functions, subconvexity, and equidistribution phenomena [1912.08137].

A broader conceptual interpretation emerges in the 2025 comparison for non-tempered spherical varieties. There the equality
\[
I(f)=J({}_\chi(f))
\]
is presented as a conceptual explanation of many earlier comparison theorems through the lens of relative Langlands duality and the residue method: a non-tempered period is compared to a tempered period on an associated Levi subgroup, with the Levi model serving as the reduced or cuspidal-support object [2512.03320].

Taken together, these formulations show that “relative trace equality” is best understood as a unifying pattern rather than a single formula. Its stable features are the replacement of diagonal trace by relative periods, the replacement of ordinary conjugacy by relative or twisted orbits, and the possibility of comparing either geometry with spectrum or one relative trace formula with another via an explicit transfer.

Source: https://www.emergentmind.com/topics/relative-trace-equality