Let $G=GL_n(F)$, where $F$ is a p-adic local field, $U$ be the upper triangular maximal unipotent group, and $\theta$ a character of $U$. Then a Theorem of Kazhdan says that for any irreducible smooth representation $(\pi,V)$ of $G$, we have $$\dim V_{U,\theta} $$ has finite dimension, where $V_{U,\theta}$ is the twisted Jacquet functor. This is Theorem 5.21 of "Bernstein-Zelevinski, Representations of $GL(n,F)$, where $F$ is a non-archimedean local field". In fact, in Theorem 5.21, the above dimension is bounded by $n!$.

My question is: is this theorem true for more general groups? If it is, where can I find a proof?



2 Answers 2


Yes - this statement is true for any irreducible representation of a split reductive p-adic group $G$, where $U$ is the unipotent radical of the Borel (a Borel subgroup is a minimal algebraic subgroup $B\subset G$ such that $G/B$ is proper; $G$ is split if such a $B$ is defined over your base field). An example is the preimage of $U$ for the standard embedding of $O(n, F), Sp(n, F),$ etc. in $GL(n).$ The source I know this from is Karl Rumelhart's transcription of Bernstein's Harvard course (although the notes only include full proofs for $GL_n$). The general sequence of arguments this follows from is as follows:

  1. an irreducible representation $V$ of a $p$-adic group $G$ is admissible (has finite-dimensional subspace $V^K$ of vectors fixed by any open compact $K\subset G$, and is generated over $G$ by $V^K$ for $K$ sufficiently fine).
  2. If $P$ is a parabolic subgroup of $G$ (such that $G/P$ is compact), $U\subset P$ is its unipotent radical and $L=P/U$ (always a reductive group, called the Levi) then the Jacquet restriction $V_{U,\theta}$ is admissible as a representation of $L$.
  3. An admissible representation of a torus $T$ (i.e., $T(F)$ for $T$ a connected reductive commutative algebraic group) is finite-dimensional. Now just take $T=B/U$.

  • $\begingroup$ Thanks for your answer. But I have a question on part (2) above. If $\theta$ is trivial, I know that the standard Jacquet functor $V_U$ is an admissible representation of the Levi. But, if $\theta$ is not trivial, the twisted Jacquet functor $V_{U,\theta}$ is not even a representation of the full Levi. Think about the GL(3) case. Let $P$ be the parabolic of type (2,1), i.e., the Levi is $GL(2)\times GL(1)$, with $GL(2)$ at the left upper corner. If $\theta$ is the character defined on the (2,3) position of the unipotent, then $V_{U,\theta}$ is just a representation of the mirabolic $\endgroup$
    – user64433
    Mar 25, 2015 at 3:52
  • $\begingroup$ of the mirabolic of the upper $GL(2)$, not the full Levi. Also, I cannot find the statement which says the twisted Jacquet functor is admissible. Do you know where I can find this? $\endgroup$
    – user64433
    Mar 25, 2015 at 3:54
  • $\begingroup$ Ah sorry, thought you were twisting by a character of the Levi (which would not affect the dimension... your current question makes a lot more sense). $\endgroup$ Mar 25, 2015 at 4:55

Decided to add a new answer instead of editing the old one (which answers a different question). Look at the proof in Bernstein-Zelevinsky. It generalizes to arbitrary split reductive groups. Here's a somewhat simplified (and coarser) argument:

Note that by passing to the contragredient, it suffices to prove your statement for twisted invariants $V^{U,\Theta}$. Now $V$ injects into $i\circ r(V)$ for $r$ the Jacquet restriction and $i$ the induction, both with respect to the Borel. The space $i\circ r(V)$ is canonically the global sections of an equivariant (topological) vector bundle $\mathcal{V}$ on $G/B$ with finite-dimensional fibers isomorphic to $r(V)$, of some dimension $d$. Now observe that $G/B$ has finitely many $U$-orbits, $O_1,\dots, O_N$, and on each orbit, the twisted invariants $\Gamma(\mathcal{V}, O_i)^{U,\Theta}$ are $\le d$-dimensional. Now we have an embedding of $i\circ r(V)$ in $\oplus_i \Gamma(\mathcal{V}, O_i)$ and we are done by left exactness of taking invariants.

  • $\begingroup$ This argument uses only quasisplitness, not splitness, right? $\endgroup$
    – LSpice
    Jul 10, 2017 at 18:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.