---
title: Local Converse Problem Overview
url: https://www.emergentmind.com/topics/local-converse-problem
type: topic
---

# Local Converse Problem Overview

Searching arXiv for recent papers on the local converse problem and related converse theorems.
The local converse problem asks how far a family of local factors attached to twists of a representation determines its isomorphism class. In the standard formulation for a non-archimedean local field \(F\), one fixes a nontrivial additive character \(\psi\), lets \(G_n=\mathrm{GL}_n(F)\), and studies whether an irreducible admissible generic representation \(\pi\) is determined by the local \(\gamma\)-factors \(\gamma(s,\pi\times\tau,\psi)\) as \(\tau\) ranges over generic representations of smaller general linear groups. For \(p\)-adic \(\mathrm{GL}_n\), this was crystallized by Jacquet’s conjecture: twists by \(\mathrm{GL}_r(F)\) for \(1\le r\le \lfloor n/2\rfloor\) should suffice. The subject now includes proofs for \(p\)-adic \(\mathrm{GL}_n\), archimedean analogues, \(\ell\)-adic families, classical and metaplectic groups, and parameter-theoretic reformulations in terms of Langlands parameters [1504.02819] [1601.03656] [2409.20240].

## 1. Formulation for \(\mathrm{GL}_n\) and the role of local factors

Let \(F\) be a non-archimedean local field, \(G_n=\mathrm{GL}_n(F)\), \(U_n\subset G_n\) the standard upper-triangular unipotent subgroup, and \(\psi:F\to\mathbf C^\times\) a nontrivial additive character. An irreducible smooth complex representation \(\pi\) of \(G_n\) is called generic if
\[
\operatorname{Hom}_{G_n}\!\bigl(\pi,\operatorname{Ind}_{U_n}^{G_n}\psi_{U_n}\bigr)\neq 0,
\]
and its Whittaker model \(\mathcal W(\pi,\psi)\) realizes \(\pi\) by functions \(W\) satisfying \(W(ug)=\psi_{U_n}(u)W(g)\) [1601.03656].

For an irreducible generic \(\pi\) of \(G_n\) and an irreducible generic \(\tau\) of \(G_t\), one forms Rankin–Selberg local integrals, for example
\[
I(s,W,W',\phi),
\]
and one obtains a functional equation
\[
I(1-s,\widetilde W,\widetilde W',\widehat\phi)
=\gamma(s,\pi\times\tau,\psi)\;I(s,W,W',\phi).
\]
The associated local \(L\)- and \(\varepsilon\)-factors satisfy the usual relation with \(\gamma(s,\pi\times\tau,\psi)\) [1601.03656] [1409.4790].

Jacquet’s local converse conjecture may then be stated as follows: if \(\pi_1,\pi_2\) are irreducible generic representations of \(G_N\) with the same central character and
\[
\gamma(s,\pi_1\times\tau,\psi)=\gamma(s,\pi_2\times\tau,\psi)
\]
for every irreducible generic \(\tau\) of \(G_r\) with \(1\le r\le \lfloor N/2\rfloor\), then \(\pi_1\simeq \pi_2\) [1409.4790] [1601.03656]. A standard reduction passes from general generic representations to the unitarizable supercuspidal case by the Zelevinsky classification and multiplicativity of \(\gamma\)-factors [1504.02819] [1601.05450].

## 2. Jacquet’s conjecture for \(p\)-adic \(\mathrm{GL}_n\)

A decisive step was the Jiang–Nien–Stevens approach, which isolates the notion of a special pair of Whittaker functions. If \(K\subset G\) is compact-mod-centre open, a Whittaker function \(W\in\mathcal W(\pi,\psi_U)\) is called \(K\)-special if \(\operatorname{Supp}W\subset U\cdot K\) and \(W(k^{-1})=\overline{W(k)}\) for all \(k\in K\). For two unitarizable supercuspidals \(\pi_1,\pi_2\) with the same central character, a pair \((W_1,W_2)\) is a special pair if both are \(K\)-special for the same \(K\) and \(W_1(p)=W_2(p)\) for all \(p\) in the mirabolic subgroup \(P\) [1504.02819] [1409.4790].

The corresponding reduction theorem states that if \(\pi_1,\pi_2\) are irreducible unitarizable supercuspidal representations of \(G_N\), if they satisfy the hypotheses
\[
H_m:\quad \gamma(s,\pi_1\times\tau,\psi)=\gamma(s,\pi_2\times\tau,\psi)
\]
for all irreducible supercuspidal \(\tau\) of \(G_m\) and all \(1\le m\le \lfloor N/2\rfloor\), and if there exists a special pair of Whittaker functions for \((\pi_1,\pi_2)\), then \(\pi_1\simeq \pi_2\) [1409.4790]. This converts the converse problem into a problem of constructing distinguished Whittaker data.

Adrian–Liu–Stevens–Xu proved that any two minimax unitarizable supercuspidals of \(\mathrm{GL}_N\) that have the same depth and central character admit a special pair of Whittaker functions. As a corollary, Jacquet’s conjecture holds for \(\mathrm{GL}_N\) when \(N\) is prime, because every supercuspidal \(\pi\) of \(G_N\) is, up to an unramified twist, either depth zero or minimax of positive depth, and the depth-zero case had already been handled in earlier work [1409.4790].

The full theorem for generic representations of \(p\)-adic \(\mathrm{GL}_n\) was then proved in two distinct ways. Jacquet–Liu gave a purely analytic proof: one decomposes \(\mathrm{GL}_n\) into double cosets \(U\cdot a^i\cdot P\), compares Whittaker functions height-by-height, and uses only the formal properties of Rankin–Selberg integrals, their analytic continuation, and uniqueness of Whittaker models. Their refined statement is that twists up to \(m=\lfloor n/2\rfloor\) suffice [1601.03656]. Independently, Chai proved Jacquet’s local converse conjecture for \(\mathrm{GL}_n(F)\) using Bessel functions: equality of twisted \(\gamma\)-factors implies equality of Bessel functions \(j_{\pi_1}=j_{\pi_2}\), and the weak kernel formula then forces \(\pi_1\simeq\pi_2\) [1601.05450].

## 3. Principal proof strategies

Three proof patterns recur throughout the subject. The first is the special-Whittaker-function strategy. Its logic is rigid: construct Whittaker functions with controlled support and symmetry, force equality of their Rankin–Selberg zeta integrals by the assumed equality of \(\gamma\)-factors, and then recover equality of the Whittaker functions themselves. In the \(\mathrm{GL}_n\) setting, this is tied to maximal simple types, endo-classes, and the explicit Bessel functions of Paškūnas–Stevens [1504.02819] [1409.4790].

The second is the Bessel-function and Howe-vector strategy. For a generic representation \(\pi\), one defines a Bessel function \(j_\pi\) by a stabilized integral over the maximal unipotent subgroup. One then constructs normalized Howe vectors \(W_m\), proves strong support and vanishing properties on Bruhat cells, inserts them into local integrals, and derives equalities such as
\[
W_{1,m}(a\,w_n)=W_{2,m}(a\,w_n)
\]
for diagonal \(a\). Letting \(m\to\infty\) and invoking a kernel formula yields \(j_{\pi_1}=j_{\pi_2}\), hence \(\pi_1\simeq\pi_2\) [1601.05450].

The third is the analytic mirabolic-comparison method. Jacquet–Liu reduce the problem to showing that two Whittaker models coincide on the mirabolic subgroup \(P\), decompose \(\mathrm{GL}_n\) into double cosets \(U\cdot a^i\cdot P\), and prove equality layer by layer by averaging the functional equation over suitable translates. A specific simplification in this approach is that no explicit construction of supercuspidals or of special Whittaker pairs is needed [1601.03656].

A common misconception is that the local converse theorem for \(\mathrm{GL}_n\) intrinsically requires twists up to \(n-1\). That bound appears in older formulations, but the refined theorem proves that twists up to \(\lfloor n/2\rfloor\) are sufficient [1601.03656]. Another common misconception is that all proofs are representation-theoretically uniform. The literature instead contains substantially different mechanisms: special pairs, partial Bessel functions, and purely analytic Rankin–Selberg arguments [1504.02819] [1601.05450].

## 4. Extensions to classical, unitary, and metaplectic groups

The local converse problem extends beyond \(\mathrm{GL}_n\) by replacing the twisting representations with generic or supercuspidal representations of \(\mathrm{GL}_r\), and by defining the relevant local \(\gamma\)-factors through Rankin–Selberg or Langlands–Shahidi methods. The resulting theorems are group-specific, and the precise twisting range is part of the statement.

| Setting | Twists used | Conclusion |
|---|---|---|
| \(U(1,1)\) | characters \(\chi:E^1\to\mathbf C^\times\), \(\eta:E^\times\to\mathbf C^\times\) | \(\pi_1\simeq\pi_2\) [1508.07062] |
| \(U_{E/F}(2,2)\) | supercuspidal \(\sigma\) of \(\mathrm{GL}_k(E)\), \(k=1,2\) | \(\pi\simeq\pi'\) [1509.00900] |
| \(U_{2r+1}\) | generic \(\tau\) of \(\mathrm{GL}_n(E)\), \(1\le n\le r\) | \(\pi\simeq\pi_0\) [1705.09410] |
| \(\mathrm{Sp}_{2r}(F)\) | generic \(\tau\) of \(\mathrm{GL}_n(F)\), \(1\le n\le r\) | \(\pi_1\simeq\pi_2\) [1705.01692] |
| \(\mathrm{SO}_{2r+1}(F)\), \(\mathrm{Sp}_{2r}(F)\), char\((F)\neq 2\) | supercuspidal \(\tau\) of \(\mathrm{GL}_n(F)\), \(1\le n\le r\) | \(\pi\simeq\pi'\) [2205.09004] |
| \(\mathrm{Mp}_{2n}(F)\) | supercuspidal generic \(\tau\) of \(\mathrm{GL}_r(F)\), \(1\le r\le n\) | \(\pi\simeq\pi'\) [2212.05234] |
| split \(\mathrm{SO}_{2l}(F)\) | generic \(\tau\) of \(\mathrm{GL}_n(F)\), \(1\le n\le l\) | \(\pi\simeq\pi'\) or \(\pi\simeq c\!\cdot\!\pi'\) [2301.13847] |
| quasi-split non-split \(\mathrm{SO}_{2l}(F)\) | generic \(\tau\) of \(\mathrm{GL}_n(F)\), \(1\le n\le l\) | \(\pi\simeq\pi'\) or \(\pi\simeq \pi'^c\) [2501.16339] |
| \(\mathrm{SO}_4(F)\), refined form | \(\mathrm{GL}_1\)-, \(\mathrm{GL}_2\)-twists and \(\gamma(s,\pi,\wedge^2\otimes\chi,\psi)\) | \(\pi\simeq\pi'\) [2302.06256] |

The principal technical devices in these results are Howe vectors, partial Bessel functions, local zeta integrals, and vanishing lemmas of Jacquet–Shalika type. This suggests that the \(\mathrm{GL}_n\) methods based on Whittaker models and Bessel distributions were not isolated phenomena, but part of a broader local-analytic pattern [1705.01692] [1705.09410] [2205.09004].

Even special orthogonal groups exhibit a specific obstruction: the outer automorphism \(c\). For split \(\mathrm{SO}_{2l}\) and quasi-split non-split \(\mathrm{SO}_{2l}\), the standard twisted \(\gamma\)-factors determine an irreducible generic representation only up to outer conjugacy [2301.13847] [2501.16339]. In rank \(2\), Yan–Zhang showed that an additional twisted exterior square local \(\gamma\)-factor resolves this ambiguity for \(\mathrm{SO}_4(F)\) [2302.06256].

The metaplectic case is conceptually different. Haan establishes the local converse theorem for \(\mathrm{Mp}_{2n}\) by transferring the problem through the precise local theta correspondence to \(\mathrm{SO}_{2n+1}\), using preservation of genericity and preservation of \(\gamma\)-factors under theta lift [2212.05234].

## 5. Families and archimedean analogues

The converse problem also admits a deformation-theoretic version. Moss extended Rankin–Selberg integrals to \(\ell\)-adic families of smooth representations of \(\mathrm{GL}_n(F)\), with coefficient rings \(R\) and \(S\), and constructed \(\gamma\)-factors
\[
\gamma(V\times V',X,\psi)\in S^{-1}((R\otimes S)[X,X^{-1}])
\]
characterized by a family functional equation. In this setting, equality of \(\gamma\)-factors against all absolutely irreducible generic integral twists of rank \(n-1\) determines the same supercuspidal support [1412.3500].

Liu–Moss then formulated a Jacquet-type theorem for co-Whittaker \(A\)-families. If \(A\) is reduced, \(\ell\)-torsion-free, and finite-type over \(W(k)\), and if two co-Whittaker \(A[\mathrm{GL}_n(F)]\)-modules with the same central character have matching \(\gamma\)-factors against every irreducible generic integral representation \(\tau\) of \(G_t\) for all \(1\le t\le \lfloor n/2\rfloor\), then they have the same Whittaker model, equivalently the same supercuspidal support. The same paper proves a descent theorem characterizing the smallest subring of definition by the coefficients of finitely many \(\gamma\)-factors [1711.11159].

Over archimedean fields, the problem takes a different form because the local factors are explicit products of classical \(\Gamma\)-functions. Adrian–Takeda first proved that over \(\mathbf C\), \(\mathrm{GL}_1\)-twists of local \(L\)-factors determine an irreducible admissible representation of \(\mathrm{GL}_n(\mathbf C)\), while over \(\mathbf R\) one needs twists by \(\mathrm{GL}_1(\mathbf R)\) and \(\mathrm{GL}_2(\mathbf R)\) [1702.07632]. They later proved a local converse theorem for archimedean \(\mathrm{GL}(n)\) in terms of twisted local \(\gamma\)-factors: if \(\pi_1,\pi_2\) are generic irreducible admissible representations of \(\mathrm{GL}_n(F)\), \(F=\mathbf R\) or \(\mathbf C\), with the same central character, and
\[
\gamma(s,\pi_1\times\chi,\psi)=\gamma(s,\pi_2\times\chi,\psi)
\]
for every unitary character \(\chi:F^\times\to S^1\), then \(\pi_1\simeq \pi_2\) [2303.10000]. The proof proceeds on the Weil-group side by a pole comparison for explicit \(\Gamma\)-factors.

## 6. Langlands-parameter formulations and sharpness

A recent reformulation treats the converse problem directly at the level of Langlands parameters. Let \(G=\mathbf G(F)\) be quasi-split, \({}^LG=\widehat G\rtimes W_F\), and let \(r:{}^LG\to \mathrm{GL}(V)\) be an algebraic representation. For an admissible homomorphism \(\phi:W_F\times \mathrm{SL}_2(\mathbf C)\to {}^LG\) and a representation \(\rho\) of \(W_F\), one forms
\[
\gamma\bigl(s,r\circ\phi\otimes\rho,\psi\bigr).
\]
Matringe proves a parameter-level local converse theorem when \(\widehat G\) is acceptable: equality of these twisted \(\gamma\)-factors for every irreducible algebraic representation \(r\) and every irreducible \(\rho\) forces the semisimple parts of the parameters to be \(\widehat G\)-conjugate. When \(G\) is \(F\)-split, this yields full equality of generic parameters; for split semisimple simply-connected \(G\), it suffices to use the fundamental representations \(r_1,\dots,r_\ell\) and twists of dimension at most
\[
\Bigl\lfloor\frac{\max_i \dim r_i}{2}\Bigr\rfloor.
\]
The same note gives variants for \(\mathrm G_2\) and quasi-split classical groups [2409.20240].

Sharpness questions then ask whether the standard twisting range can be reduced. Adrian–Stevens show that when \(G\) is a symplectic or special orthogonal group, or the exceptional group \(G_2\), and the residue characteristic \(p\) is large enough, the optimal standard local converse theorem requires twisting by representations of \(\mathrm{GL}_r(F)\) with \(r\) up to half the dimension of the standard representation of the dual group [2509.22390]. For generic supercuspidal representations of \(SO_{2N}\), there is an improvement when \(N\) is odd: twisting up to \(r=N-1\) suffices, and this bound is itself optimal [2509.22390].

The same paper also gives counterexamples to possible improvements based on non-standard representations of the dual group. For \(GL_4\), there exist inequivalent irreducible supercuspidal parameters \(\varphi_1,\varphi_2\) such that all twisted \(\gamma\)-factors
\[
\gamma\bigl(s,(\Lambda^i\circ\varphi_1)\otimes\eta,\psi\bigr)
=
\gamma\bigl(s,(\Lambda^i\circ\varphi_2)\otimes\eta,\psi\bigr)
\]
agree for every character \(\eta\) and every \(1\le i\le 4\). Analogous counterexamples are given for \(G_2\) and \(SO_{2N}\) [2509.22390]. A plausible implication is that the optimality problem is not merely a matter of enlarging the repertoire of dual-side algebraic representations; the specific twisting range is itself a structural invariant of the local converse theorem.

The modern picture is therefore stratified. For \(p\)-adic \(\mathrm{GL}_n\), Jacquet’s conjectural bound \(\lfloor n/2\rfloor\) is exact in the standard formulation [1601.03656] [1601.05450]. For many classical groups, one has direct local converse theorems with twists by \(\mathrm{GL}_r\) up to the natural half-rank bound, but even orthogonal groups retain an outer-automorphism phenomenon unless extra twisted factors are introduced [2301.13847] [2302.06256]. At the same time, parameter-level results show that the subject fits naturally into the structure of the local Langlands correspondence [2409.20240].

Source: https://www.emergentmind.com/topics/local-converse-problem