--- title: Langlands Branching Rule Overview url: https://www.emergentmind.com/topics/langlands-branching-rule type: topic --- # Langlands Branching Rule Overview Searching arXiv for recent papers directly relevant to “Langlands branching rule” and nearby formulations. “Langlands branching rule” denotes several distinct but related constructions rather than a single universally fixed theorem. In one line of work it means a relative-Langlands description of model multiplicities, especially generalized Whittaker models arising from quantized hyperspherical Hamiltonian varieties; in another it means an explicit decomposition of folded characters for quantum affine algebras into irreducibles of the Langlands dual algebra; and in a third it means a criterion, stated in Langlands or Zelevinsky data, for when a quotient branching space such as \(\operatorname{Hom}_{\mathrm{GL}_n(F)}(\pi,\pi')\) is nonzero. Other works provide the parameter-theoretic or classical branching infrastructure that such rules would require, without themselves proving a branching theorem in the Langlands sense [2309.08874] [2507.06570] [2512.11696] [2508.15101]. ## 1. Scope of the term Across the literature surveyed here, the phrase does not uniformly mean a direct restriction formula of the form \[ \operatorname{Res}^G_H \pi=\bigoplus_\sigma m(\sigma)\sigma. \] Instead, it can refer to spectral decompositions of model spaces, explicit character decompositions under Langlands duality, or nonvanishing criteria for branching Hom-spaces. This variation is explicit in the contrast between generalized Whittaker models, finite-field Langlands parameterizations, affine quantum branching, and local \(\mathrm{GL}_{n+1}\downarrow \mathrm{GL}_n\) quotient branching [2309.08874] [2508.15101] [2507.06570] [2512.11696]. | Setting | Basic object | Form of “branching rule” | |---|---|---| | Relative Langlands program | Quantization of a hyperspherical \(G\)-variety | Spectral description of model multiplicities | | Quantum affine type \(B_n^{(1)}\) | Folded character \(\Pi(\chi(V))\) | Explicit sum of irreducible dual characters | | Local \(\mathrm{GL}_{n+1}\downarrow \mathrm{GL}_n\) | \(\operatorname{Hom}_{G_n}(\pi,\pi')\) | Criterion in multisegment or Tadić data | | Finite-field LLC | \(L_G:\Irr(G(k))\to \Phi(G)_{sp}\) | Infrastructure, not a direct branching theorem | A recurrent misconception is that every “Langlands branching rule” should be a classical restriction law. Several of the cited works explicitly reject that interpretation. The generalized Whittaker papers study model multiplicities and periods rather than ordinary restriction, while the finite Langlands correspondence paper constructs packet parameterizations but states no theorem comparing representations of two related groups under restriction [2309.08874] [2401.06624] [2508.15101]. ## 2. Relative Langlands duality and generalized Whittaker models In the relative Langlands framework of Ben-Zvi, Sakellaridis, and Venkatesh, the fundamental objects are certain Hamiltonian \(G\)-varieties \(M\), called hyperspherical varieties, whose quantizations produce \(G\)-representations. The branching problem becomes the determination of the spectral decomposition of the quantization \(\Pi_M\): which irreducible \(G\)-representations occur, and with what multiplicity. Relative Langlands duality is then an expected correspondence \[ {}^G M \longleftrightarrow M^\vee {}^{G^\vee} \] between hyperspherical \(G\)- and \(G^\vee\)-varieties, understood operationally through Arthur-parameter factorization on the two sides. In this setting, generalized Whittaker models are the prototypical nonspherical examples [2309.08874]. For an \(\mathfrak{sl}_2\)-triple \(y=\{e,h,f\}\subset \mathfrak g\), one has the weight decomposition \[ \mathfrak g_j=\{v\in \mathfrak g\mid \operatorname{ad}(h)v=jv\}, \] together with \[ \mathfrak p=\bigoplus_{j\ge 0}\mathfrak g_j=\mathfrak l\oplus \mathfrak u,\qquad \mathfrak u^+=\bigoplus_{j\ge 2}\mathfrak g_j. \] The associated character on \(U^+\) is \[ \chi_{y,\psi}(\exp u):=\psi(\kappa(f,u)). \] If the orbit is even, so that \(U=U^+\), the generalized Whittaker representation is \[ W_{y,\psi}:=\operatorname{ind}_{M_yU}^{G}\chi_y, \] and the model multiplicity space is \[ W_{y,\psi}(\pi)\simeq \operatorname{Hom}_{M_yU}(\pi^\vee,\chi_y). \] In the non-even case, \(\mathfrak g_1\) carries the symplectic form \[ \kappa_1(v,w)=\kappa(\operatorname{ad}(f)v,w)=\kappa(f,[v,w]), \] leading to a Heisenberg group \(H_y\), an oscillator representation \(\omega_y\), and the model \[ W_{y,\rho,\psi}:=\operatorname{ind}_{MU}^{G}(\rho\otimes \omega_y), \qquad W_{y,\rho,\psi}(\pi)\simeq \operatorname{Hom}_{MU}(\pi^\vee,\rho\otimes \omega_y). \] These are branching invariants in the precise sense that they are Hom-spaces attached to nilpotent-orbit data [2309.08874]. Geometrically, generalized Whittaker models arise by Whittaker induction. Given \[ \iota:H\times \mathrm{SL}_2\to G \] and a symplectic \(H\)-space \(S\), the Whittaker-induced Hamiltonian variety is \[ M:= (S\times (\mathfrak u/\mathfrak u^+))\times_{(\mathfrak h+\mathfrak u)^*}T^*G, \] which simplifies to \[ M \simeq (S\times_{\mathfrak h^*}(f+\mathfrak g^e))\times_H G. \] For \(S=0\), this becomes \[ M_e=((f+\mathfrak g^e)\cap \mathfrak h^\perp)\times_H G, \] whose quantization is the generalized Whittaker model attached to \(e\). The structural theorem states that every hyperspherical \(G\)-variety is a Whittaker induction of some symplectic \(H\)-vector space \(S\) along some map \(H\times \mathrm{SL}_2\to G\). This reduces the classification problem to determining which nilpotent orbits can occur in hyperspherical Whittaker induction [2309.08874]. For orthogonal groups, if \(G=\mathrm O_n\), \(M\) is hyperspherical, and the \(\mathrm{SL}_2\)-factor defines an even nilpotent orbit \(y\), then the corresponding partition \(\lambda\) must be one of \([2^{a_2}]\), \([n-a_1,1^{a_1}]\), or one of the low-rank exceptions \[ [3,3],\quad [4,4],\quad [6,6]. \] For symplectic groups, if \(G=\mathrm{Sp}_{2n}\), \(M\) is hyperspherical, and \(y\) is even, then \(\lambda\) must be one of \([2^{a_2}]\), the exceptional partitions \[ [3,3,1^{2a}],\qquad [5,5,1^{2a}],\qquad a\ge 0, \] or the trivial and regular orbits \[ [1^{2n}],\qquad [2n]. \] In the non-even case, a necessary combinatorial condition is that the even and odd subpartitions of \(\lambda\) must themselves be of the allowed forms. These statements characterize which generalized Whittaker branching problems are eligible to belong to the relative Langlands duality framework [2309.08874]. ## 3. Local dual branching correspondences and theta transfer The most concrete local realizations of a Langlands branching rule in this relative setting occur for hook-type partitions, which correspond to the Bessel and Fourier–Jacobi models of the Gan–Gross–Prasad setting. For even orthogonal groups \(n=2k\), the relevant hook partitions are \[ [2k-2a-1,1^{2a+1}] \quad\text{and}\quad [2a-1,1^{2k-2a+1}]. \] The corresponding hyperspherical varieties \(M_1\) and \(M_2\) are declared dual, and the spectral statement is that if \(\pi\in \operatorname{Irr}(\mathrm O_{2k})\) occurs as a quotient of the first generalized Whittaker model, then \[ \pi=\Theta_\psi(\sigma) \] for some \(\sigma\in \operatorname{Irr}(\mathrm{Sp}_{2k-2a})\); conversely, if \(\sigma\) is irreducible and \(\psi\)-generic for \(\mathrm{Sp}_{2k-2a}\), then \(\Theta_\psi(\sigma)\) is irreducible and occurs as a quotient of that model. Dually, the second hook model is described by theta lifts from \(\mathrm{Sp}_{2a}\). The dual branching problem therefore exchanges \(a\) and \(k-a\) [2309.08874]. For odd orthogonal groups the picture becomes asymmetric because the dual group is symplectic and the non-even model requires a nontrivial symplectic factor \(S\), equivalently a Weil representation factor. In that case one hook partition on the orthogonal side is paired with a symplectic-side model involving the standard symplectic representation of a smaller symplectic group. The local theorem states that representations occurring in the orthogonal generalized Whittaker model are theta lifts from irreducible \(\psi\)-generic metaplectic representations, while on the dual side the relevant symplectic generalized Whittaker model is described by theta lifts from irreducible \(\psi\)-generic orthogonal representations. This is a branching law of Langlands type because the spectral decomposition of one model space is described in terms of functorial transfer from the group appearing in the dual datum [2309.08874]. Theta correspondence is the mechanism that makes this local duality explicit. Howe duality provides the multiplicity bound \[ \dim \operatorname{Hom}(\omega_\psi,\pi_1\boxtimes \pi_2)\le 1 \] for a reductive dual pair \((G_1,G_2)\). Adams’ conjecture, in the proved cases used in the paper, identifies theta lift on \(A\)-parameters by adjoining an \(\mathrm{SL}_2\)-factor: \[ \phi \mapsto \phi \oplus W_{\dim V_2-\dim V_1-1}. \] The Gomez–Zhu theorem then gives the decisive identity transferring generalized Whittaker models through theta correspondence: \[ W_{y,\tau,\psi}\big(\Theta(\pi)\big) = W_{y',\,\Theta(\tau)^\vee,\,\psi}(\pi^\vee). \] This identity converts one branching problem into another and explains why generalized Whittaker models are especially effective test cases for relative Langlands duality. The same analysis also motivates multiplicity-one expectations such as \[ \dim \operatorname{Hom}(W_{y_1},\pi)\le 1 \] in the hook-type families, provided the relevant big theta lift is irreducible [2309.08874]. ## 4. Plancherel density and global periods A second paper develops the same relative-branching viewpoint at the level of local Plancherel density and global periods. The representative family is the even orthogonal case \(G=O_{2k}\), with auxiliary group \(G'=Sp_{2a}\), and a generalized Whittaker model attached to the hook partition \([2a-1,1^{2k-2a+1}]\). The model space is \[ X:=LU,\chi_Y\backslash G, \] where \(U\) is the unipotent subgroup determined by the nilpotent orbit, \(\chi_Y\) is the corresponding unitary character, and \(L\cong O_{2k-2a+1}\). This is again a relative branching problem: one studies \(\operatorname{Hom}_{LU}(\pi,\chi_Y)\), or equivalently the multiplicity of \(\pi\) in \(L^2(X)\), rather than ordinary restriction to a reductive subgroup [2401.06624]. The basic local realization identifies the smooth model with coinvariants of the Weil representation. The surjective map \[ p : \omega \to \mathcal{S}(LU,\chi_Y\backslash G) \] induces \[ (\omega)_{U',\chi_{Y'}} \cong \mathcal{S}(LU,\chi_Y\backslash G). \] From this one obtains the spectral decomposition \[ L^2(X)=\int_{\widehat{G'}} \theta(\sigma)\cdot \dim \operatorname{Hom}_{U'}(\sigma,\chi_{Y'})\, d\mu_{G'}(\sigma), \] so the spectrum of the generalized Whittaker model on \(G\) is the theta-lift of the tempered, \(\chi_{Y'}\)-generic spectrum on \(G'\). This is the local branching law in the paper: the spectral measure on the model side is the pushforward of Whittaker-Plancherel measure on the smaller symplectic group [2401.06624]. The numerical conjecture of Ben-Zvi–Sakellaridis–Venkatesh is then verified for this family. The dual geometric representation is computed as \[ V_X = \mathfrak{so}_{2a+1}\backslash \mathfrak{so}_{2k} = \mathrm{std}[2k-2a] + \mathrm{triv}[4k-4a-4,\ldots,8,4]. \] The unramified Plancherel density of the basic function is the pushforward, via \[ A_X^{\vee(1)} \to A^\vee,\qquad (\cdot)\mapsto (\cdot)\, q^{-P(X)}, \] of the measure \[ \frac{1}{|W_X|\det(I-(t,q^{-1/2})|V_X)} \cdot \frac{1}{\det(I-t| \mathfrak g_X/\mathfrak a_X)}\, dt. \] This is the paper’s exact local “numerical” branching rule: the density on the model side is described by determinant data on the dual side [2401.06624]. The global application is a period formula for \[ P_Y(f)=\int_{[L]}\int_{[U]} f(ul)\chi_Y(u)\,du\,dl. \] Under the stated hypotheses, if \(\Sigma\) is a globally \(\psi\)-generic cuspidal automorphic representation of \(Sp_{2a}\) and \(\Pi=\Theta(\Sigma)\) is its cuspidal theta lift to \(O_{2k}\), then \[ |P_Y(f)|^2 = \frac{1}{|S_\phi|}\, \frac{4\Delta_\Sigma(1)^2 L^T(1,\Sigma,\mathrm{Ad})} {\Delta_\Pi(1)\,\prod_{i\in T} L^T(i/2,\Sigma,V(i))} \prod_v |\ell_{\Pi_v}(f_v)|^2. \] A plausible implication is that the local relative branching rule is being promoted here to an Eulerian period identity: local model functionals, local Plancherel density, and global \(L\)-values are organized by the same dual hyperspherical datum [2401.06624]. ## 5. Quantum affine Langlands branching rules In the representation theory of quantum affine algebras, “Langlands branching rule” is used in a more literal decomposition-theoretic sense. For the quantum Kac–Moody algebra of type \(B_n^{(1)}\), the paper proves that every shortened snake module admits a Langlands dual representation for the twisted affine type \(A_{2n-1}^{(2)}\), and, more strongly, that the folded character decomposes explicitly into irreducible characters of the dual algebra [2507.06570]. The finite-type weight lattice \(P\) of type \(B_n\) contains the sublattice \[ P'=\mathbb Z\omega_1\oplus\cdots\oplus \mathbb Z\omega_{n-1}\oplus 2\mathbb Z\omega_n, \] and there is a bijection \[ \Pi:P'\to {}^LP,\qquad \Pi(\omega_i)=\check\omega_{\bar i}\ (i