Generalized Whittaker Models in Modern Representation Theory

• Generalized Whittaker models extend classical Whittaker constructions to arbitrary nilpotent orbits, unifying local, global, and categorical settings.
• They employ Whittaker pairs, oscillator representations, and comparison maps to reveal key structure, including wave-front sets and asymptotic expansions.
• These models are pivotal in linking representation theory, automorphic Fourier analysis, and conformal field theory via algebraic and geometric techniques.

Generalized Whittaker models are orbit-theoretic extensions of the classical Whittaker model from generic representations to broader classes of representations of reductive groups, as well as to several geometric, categorical, and infinite-dimensional settings. In the local representation theory of reductive groups, the classical Whittaker model corresponds to the principal nilpotent orbit, while generalized and degenerate models are attached to arbitrary nilpotent orbits or to more flexible Whittaker pairs $(S,\varphi)$ (Gourevitch et al., 2018). In the global setting they appear as Fourier coefficients of automorphic forms attached to nilpotent data; in geometric representation theory they become Whittaker categories; and in integrable-systems and conformal-field-theoretic contexts they reappear as parabolic Whittaker functions or generalized Whittaker states (Gaitsgory, 2018).

1. Orbit-theoretic definition and local construction

A standard local framework begins with a reductive group $G$ over a local field of characteristic $0$, its Lie algebra $\mathfrak g$, and a Whittaker pair $(S,\varphi)$, where $S\in\mathfrak g$ is rational semisimple and $\varphi\in\mathfrak g^*$ satisfies

$\operatorname{ad}^*(S)(\varphi)=-2\varphi.$

From this data one defines

$$\mathfrak u=\mathfrak g_{\ge 1}^S, \qquad \omega_\varphi(X,Y)=\varphi([X,Y]),$$

lets $\mathfrak n$ be the radical of $G$0 on $G$1, and sets $G$2, $G$3, $G$4 (Gomez et al., 2015, Gourevitch et al., 2018). If $G$5 is not an eigenvalue of $G$6, then $G$7 defines a character $G$8 of $G$9 and the degenerate Whittaker model is

$0$0

If $0$1 is an eigenvalue, then $0$2 is a Heisenberg group, and one instead induces the corresponding oscillator representation: $0$3 (Gomez et al., 2015).

A generalized Whittaker model is obtained when $0$4 is chosen as a neutral element for $0$5, equivalently from an $0$6-triple associated to the nilpotent orbit of $0$7. The resulting model $0$8 depends only on the coadjoint orbit $0$9, so it is also written $\mathfrak g$0 (Gomez et al., 2015, Gourevitch et al., 2018). For a smooth representation $\mathfrak g$1, one considers either

$\mathfrak g$2

or the quotient formulation

$\mathfrak g$3

depending on the context (Gomez et al., 2015, Gourevitch et al., 2018).

This construction clarifies a common point of confusion: generalized Whittaker models are not restricted to generic representations. Precisely because only generic representations admit classical Whittaker models, generalized and degenerate models were introduced to attach orbit-dependent Whittaker data to arbitrary representations (Gourevitch et al., 2018). The maximal nilpotent orbits for which the corresponding quotient is nonzero form the Whittaker support

$\mathfrak g$4

where $\mathfrak g$5 is the set of orbits $\mathfrak g$6 with $\mathfrak g$7 (Gourevitch et al., 2018).

2. Comparison theorems, wave-front sets, and derivatives

A central structural result is that generalized Whittaker models dominate degenerate ones. For a Whittaker pair $\mathfrak g$8, there is a $\mathfrak g$9-equivariant epimorphism

$(S,\varphi)$0

and more generally there are comparison maps to certain degenerate models attached to larger compatible orbits (Gomez et al., 2015). In the survey formulation, if $(S,\varphi)$1 is a Whittaker pair and $(S,\varphi)$2 lies in the $(S,\varphi)$3-orbit closure of $(S,\varphi)$4, then there is a natural surjection

$(S,\varphi)$5

with corresponding nonvanishing implications for representation-theoretic quotients (Gourevitch et al., 2018).

These comparison results connect generalized Whittaker models to singular support invariants. For non-Archimedean fields, the maximal nilpotent orbits in the Whittaker support coincide with the wave-front set: $(S,\varphi)$6 for $(S,\varphi)$7 (Gourevitch et al., 2018). For $(S,\varphi)$8, Gomez–Gourevitch–Sahi sharpen this to a full orbit-closure criterion: $(S,\varphi)$9 not merely for maximal orbits (Gomez et al., 2015). They also express generalized Whittaker models for $S\in\mathfrak g$0 as iterated Bernstein–Zelevinsky-type derivatives: $S\in\mathfrak g$1 for the partition $S\in\mathfrak g$2 corresponding to $S\in\mathfrak g$3 (Gomez et al., 2015).

The orbit picture also imposes restrictions on which nilpotent orbits can occur. The strongest general statement in the survey is that every orbit in $S\in\mathfrak g$4 is quasi-admissible, and in the $S\in\mathfrak g$5-adic setting there are strong relations between Whittaker support, distinguishedness, and cuspidality or temperedness (Gourevitch et al., 2018). This places generalized Whittaker models alongside wave-front sets, annihilator varieties, and associated cycles as nilpotent invariants of representations rather than as isolated functional constructions.

3. Jacquet modules, asymptotics, and $S\in\mathfrak g$6-criteria

For reductive groups over non-Archimedean local fields, generalized Whittaker functions can be studied through constant terms and Jacquet modules. Let $S\in\mathfrak g$7 be the $S\in\mathfrak g$8-points of a connected reductive group, $S\in\mathfrak g$9 a minimal parabolic, and $\varphi\in\mathfrak g^*$0 a non-degenerate character. The space

$\varphi\in\mathfrak g^*$1

admits a constant term map

$\varphi\in\mathfrak g^*$2

for any standard parabolic $\varphi\in\mathfrak g^*$3, and Delorme’s construction descends to normalized Jacquet modules (Matringe, 2020). A key theorem identifies the descended map with the dual of the inverse Bushnell–Henniart isomorphism on compactly supported Whittaker spaces, so $\varphi\in\mathfrak g^*$4 is surjective. The same paper proves that Lapid–Mao’s germ map equals this Jacquet-module map and is therefore injective (Matringe, 2020).

This comparison yields asymptotic expansions controlled by Jacquet-module exponents. For an admissible $\varphi\in\mathfrak g^*$5-submodule $\varphi\in\mathfrak g^*$6 and $\varphi\in\mathfrak g^*$7, the expansion of $\varphi\in\mathfrak g^*$8 on cones in $\varphi\in\mathfrak g^*$9 is indexed by parabolics $\operatorname{ad}^*(S)(\varphi)=-2\varphi.$0 and by the character set $\operatorname{ad}^*(S)(\varphi)=-2\varphi.$1, with terms lying in generalized eigenspaces $\operatorname{ad}^*(S)(\varphi)=-2\varphi.$2 (Matringe, 2020). The same paper gives an integral $\operatorname{ad}^*(S)(\varphi)=-2\varphi.$3-adic version: if $\operatorname{ad}^*(S)(\varphi)=-2\varphi.$4 is an integral finite-length submodule with integral Jacquet modules, then the expansion can be chosen with integral $\operatorname{ad}^*(S)(\varphi)=-2\varphi.$5 and integral $\operatorname{ad}^*(S)(\varphi)=-2\varphi.$6 (Matringe, 2020).

For the specific families

$\operatorname{ad}^*(S)(\varphi)=-2\varphi.$7

a more explicit asymptotic theory is available. Using mirabolic subgroups or their analogues and derivative functors, Matringe proves an asymptotic expansion of Whittaker functions along the maximal torus $\operatorname{ad}^*(S)(\varphi)=-2\varphi.$8, in terms of a minimal set of characters $\operatorname{ad}^*(S)(\varphi)=-2\varphi.$9 arising from derivatives $$\mathfrak u=\mathfrak g_{\ge 1}^S, \qquad \omega_\varphi(X,Y)=\varphi([X,Y]),$$0 (Matringe, 2010). The same derivative exponents determine square-integrability: for a generic representation with unitary central character, the following are equivalent:

1. all Whittaker functions are in $$\mathfrak u=\mathfrak g_{\ge 1}^S, \qquad \omega_\varphi(X,Y)=\varphi([X,Y]),$$1,
2. all exponents of the derivatives are positive,
3. $$\mathfrak u=\mathfrak g_{\ge 1}^S, \qquad \omega_\varphi(X,Y)=\varphi([X,Y]),$$2 is square-integrable, hence a generic discrete series representation (Matringe, 2010).

These results prove, for those four families, the Lapid–Mao conjectural picture that generic representations occurring in $$\mathfrak u=\mathfrak g_{\ge 1}^S, \qquad \omega_\varphi(X,Y)=\varphi([X,Y]),$$3 are precisely the generic discrete series (Matringe, 2010). A plausible implication is that generalized Whittaker asymptotics are most effective when organized through derivative or Jacquet-module data rather than by direct analysis on the ambient group.

4. Global Fourier coefficients, theta lifting, and relative duality

In the global theory, generalized Whittaker models appear as Fourier coefficients of automorphic forms. For a number field $$\mathfrak u=\mathfrak g_{\ge 1}^S, \qquad \omega_\varphi(X,Y)=\varphi([X,Y]),$$4, adeles $$\mathfrak u=\mathfrak g_{\ge 1}^S, \qquad \omega_\varphi(X,Y)=\varphi([X,Y]),$$5, and a Whittaker pair $$\mathfrak u=\mathfrak g_{\ge 1}^S, \qquad \omega_\varphi(X,Y)=\varphi([X,Y]),$$6 defined over $$\mathfrak u=\mathfrak g_{\ge 1}^S, \qquad \omega_\varphi(X,Y)=\varphi([X,Y]),$$7, the relevant coefficient is

$$\mathfrak u=\mathfrak g_{\ge 1}^S, \qquad \omega_\varphi(X,Y)=\varphi([X,Y]),$$8

where $$\mathfrak u=\mathfrak g_{\ge 1}^S, \qquad \omega_\varphi(X,Y)=\varphi([X,Y]),$$9 is constructed from $\mathfrak n$0 exactly as in the local theory (Gourevitch et al., 2018). The global comparison theorems show that $\mathfrak n$1 and the coefficient attached directly to $\mathfrak n$2 are connected by sequences of integral transforms, and that maximal Fourier coefficients govern the others (Gourevitch et al., 2018, Gomez et al., 2015).

Local theta correspondence transports generalized Whittaker models along nilpotent-orbit correspondences. For a reductive dual pair $\mathfrak n$3, moment maps on $\mathfrak n$4 define a lift $\mathfrak n$5 of nilpotent orbits. If $\mathfrak n$6 is a smooth irreducible representation of $\mathfrak n$7 and $\mathfrak n$8 its full theta lift, then under the standing moment-map hypothesis there is a canonical identification

$\mathfrak n$9

as modules over the relevant stabilizer group (Gomez et al., 2013). In the stable range with $G$00 the smaller member, every nilpotent orbit of $G$01 lies in the image of the moment map, so the theorem applies to all such orbits (Gomez et al., 2013).

This local mechanism underlies more elaborate global results. For the generalized metaplectic theta lift from the $G$02-fold cover of $G$03 to a tower of split even orthogonal groups $G$04 with $G$05 odd, the Whittaker range consists of exactly $G$06 consecutive groups

$G$07

assuming the Orbit Conjecture and the Descent Conjecture (Friedberg et al., 2021). Outside this range the lift cannot be globally generic; at the central point $G$08, genericity is equivalent to genericity of the original cuspidal representation on the symplectic side; and at other points in the range nonvanishing of the Whittaker coefficient is characterized by explicit period integrals obtained through root exchange and Fourier expansion (Friedberg et al., 2021).

A further development places generalized Whittaker models inside the Ben-Zvi–Sakellaridis–Venkatesh framework of relative Langlands duality. The 2023 paper characterizes, for orthogonal and symplectic groups, which nilpotent orbits can arise from hyperspherical varieties and exhibits an infinite family of hook-type examples related by theta correspondence (Gan et al., 2023). The 2024 sequel proves the corresponding local numerical conjecture for Plancherel density and, assuming the Lapid–Mao conjecture and local multiplicity one, a global period formula for a family of generalized Whittaker models on $G$09 arising from hook-type nilpotent orbits (Gan et al., 2024). In that setting, the local spectral decomposition of

$G$10

is described as the theta-pushforward of the Whittaker–Plancherel measure on the dual symplectic group, and the global generalized Whittaker period is expressed in terms of an adjoint $G$11-value, local correction factors, and normalized local period integrals (Gan et al., 2024).

5. Categorical, geometric, and parabolic incarnations

Generalized Whittaker models also admit a categorical reformulation. If a DG category $G$12 carries an action of the loop group $G$13, then its Whittaker model is defined by

$G$14

where $G$15 is a non-degenerate character (Gaitsgory, 2018). In this setting there are both invariants and coinvariants,

$G$16

and a major theorem states that the pseudo-identity functor

$G$17

is an equivalence (Gaitsgory, 2018). For $G$18, the local Whittaker category is compactly generated, stratified by $G$19-orbits, and equivalent via Beauville–Laszlo gluing to a global Whittaker category on a level-structured Drinfeld compactification (Gaitsgory, 2018). This is a categorical local–global theorem rather than a mere analogy.

A different geometric direction concerns parabolic Whittaker functions. For $G$20, Gerasimov–Lebedev–Oblezin define Whittaker functions attached to a parabolic subgroup $G$21 as matrix elements of infinite-dimensional representations with parabolic left and right Whittaker vectors (Gerasimov et al., 2010). These functions are common eigenfunctions of commuting Hamiltonians of a parabolic quantum Toda chain, and in the maximal parabolic case $G$22 the same function appears both as a type A equivariant topological sigma-model correlator on a disk and as a type B equivariant Landau–Ginzburg correlator, producing a mirror-symmetric realization of the parabolic Whittaker function (Gerasimov et al., 2010).

Oblezin develops a related $G$23-Whittaker function by replacing the standard Cartan and standard nilpotent subalgebras of $G$24 with parabolically adapted analogues $G$25, $G$26, and $G$27 (Oblezin, 2011). The resulting matrix element admits a Givental-type stationary phase integral representation, and the phase function reproduces the combinatorics of the Batyrev–Ciocan-Fontanine–Kim–van Straten toric degeneration of $G$28, linking generalized Whittaker theory to quantum cohomology and total positivity (Oblezin, 2011).

These examples show that “generalized Whittaker model” is not restricted to one formalism. Depending on context, it can mean a quotient attached to a nilpotent orbit, a DG category cut out by $G$29-equivariance, or a matrix-element realization adapted to parabolic geometry. The common feature is the replacement of the single generic Whittaker character by more structured nilpotent, parabolic, or categorical data.

6. Infinite-dimensional Lie algebras and generalized Whittaker states

The terminology also extends to infinite-dimensional Lie algebras. For the planar Galilean conformal algebra $G$30 and its universal central extension $G$31, a Whittaker module of type $G$32 is generated by a vector $G$33 satisfying

$G$34

with universal modules

$G$35

(Chen et al., 2020). The paper classifies universal and generic Whittaker modules, proves that a generic module is irreducible if and only if $G$36 is nonsingular, determines all Whittaker vectors in the nonsingular case, and constructs explicit proper submodules in the singular case (Chen et al., 2020).

For the affine Lie algebra $G$37 of type $G$38, non-degenerate Whittaker modules at noncritical level are irreducible, while at critical level one must quotient by a central character coming from the center of the vertex algebra $G$39 to obtain irreducible modules (Adamovic et al., 2014). The same paper uses vertex-algebraic and Wakimoto-type constructions to realize families of generalized Whittaker irreducible modules at critical level (Adamovic et al., 2014). Here the generalized aspect is tied to the enlarged center and to the action of the derivation $G$40, which may be semisimple or free on irreducible modules with the same Whittaker function and the same central character (Adamovic et al., 2014).

In conformal-field-theoretic applications motivated by instanton counting, generalized Whittaker states are not strict coherent states. For asymptotically free $G$41 theories with fundamental hypermultiplets, the defining relations for the relevant states in Verma modules can involve zero modes such as $G$42 or $G$43, rather than only positive annihilation operators (Kanno et al., 2012). In the $G$44 case with $G$45, the state satisfies a relation of the form

$G$46

and in the $G$47 case with a surface operator the affine current algebra condition involves $G$48 (Kanno et al., 2012). This suggests a broader module-theoretic usage of the term “generalized Whittaker,” in which the Whittaker condition is deformed by central or zero-mode operators rather than being given by a pure character of a nilpotent subalgebra.

Across these settings, generalized Whittaker models retain a common core: they encode representation-theoretic information through controlled equivariance against nilpotent, parabolic, or positive subalgebra data. What changes from one framework to another is the ambient category—smooth representations, automorphic forms, sheaf categories, principal-series modules, affine or conformal modules—not the underlying principle that Whittaker-type data should be organized by richer orbit or symmetry structures than the single generic character of the classical theory.