2
$\begingroup$

An irreducible continuous unitary representation $\pi$ of $G$ is said to be integrable, if the map $\phi(x)=\langle\pi(x)\zeta,\zeta\rangle$ is integrable on $G$, where that $\zeta\in H(\pi)$.

Question1: Suppose $G$ is an unimodular locally compact group. Can we say that $G$ always has an (irreducible) integrable representation?

Question 2: If the answer of the above question is no, is there an example of a non-compact group for which the answer of question 1 is yes?

$\endgroup$
2
  • $\begingroup$ The definition given here encyclopediaofmath.org/index.php/Integrable_representation of what it means for a continuous irrep to be "integrable" appears to be potentially weaker than the definition you state in the question. Namely, a (continuous, unitary) irrep $\pi$ is said to be integrable if there exists some non-zero $\zeta\in H_\pi$ for which $x\mapsto \langle \pi(x)\zeta,\zeta\rangle\in L^1(G)$. $\endgroup$
    – Yemon Choi
    Commented Oct 10, 2017 at 20:09
  • $\begingroup$ Remark: for square-integrable irreducible unitary representations, having one non-zero vector in $H_\pi$ that gives an element of $L^2(G)$ implies that all vectors in $H_\pi$ give rise to elements of $L^2(G)$. But I don't know if the same phenomenon of "if one vector works then they all work" applies to integrable irreducible unitary representations. $\endgroup$
    – Yemon Choi
    Commented Oct 10, 2017 at 20:11

1 Answer 1

3
$\begingroup$

[Thanks to Loren Spice for fixing the references and pointing out a silly error/mis-statement in the original version of this answer.]


The answer to Q1 is no for $G={\mathbb Z}$, since all its irreducible representations are one-dimensional, and hence all the associated coefficient functions ${\mathbb Z} \to {\mathbb C}$ take values in ${\mathbb T}$.

The answer to Q2 is positive, but I am not aware of any easy approach that naturally leads one to an explicit example. However, explicit examples are known, by the following line of thought.

First: it is a non-trivial result that $G$ having an integrable irreducible representation $\pi$ implies that $\{\pi\}$ is an open subset of the unitary dual $\widehat{G}$. (This result is, to my knowledge, due to Duflo and Moore in Duflo and Moore - On the regular representation of a nonunimodular locally compact group (MSN); but see also Valette - Minimal projections, integrable representations and property (T) (MSN) for an easier proof in the unimodular case).

So for Q2, one should look for unimodular non-compact $G$ for which $\widehat{G}$ has an open point $\{\pi\}$, and then check if $\pi$ really is integrable. Note that since a function $\phi$ of the form in your question is always continuous and bounded, if it is integrable then it belongs to every $L^p$. In particular integrable (irreducible) representations are always square-integrable, and there is a substantial literature on which groups admit square-integrable representations.

Moreover, it turns out that for certain classes of groups -- including all connected nilpotent Lie groups -- an irreducible unitary representation $\pi$ is integrable if and only if it is square-integrable, if and only if $\pi$ is an open point in the unitary dual. See Theorem 3 of Barnes - The role of minimal idempotents in the representation theory of locally compact groups (MSN).

Therefore, to get a positive answer to Q2, it suffices to exhibit a connected nilpotent Lie group $G$ such that $\widehat{G}$ has at least one open point. A concrete example of such a group is the reduced Heisenberg group, defined as $G = H_3({\mathbb R})/Z$ where

$$ H_3({\mathbb R}) = \left\{ \left( \begin{matrix} 1 & p & \theta \\ 0 & 1 & q \\ 0 & 0 & 1 \end{matrix} \right) \;\colon\; p,q,\theta\in {\mathbb R} \right\} \quad,\quad Z = \left\{ \left( \begin{matrix} 1 & 0 & n \\ 0 & 1 & 0 \\ 0 & 0 & 1 \end{matrix} \right) \;\colon\; n\in {\mathbb Z} \right\} $$ with an explicit integrable irreducible representation $\pi: G \to {\mathcal B}(L^2({\mathbb R}))$ given by $$ (\pi(p,q,\theta+{\mathbb Z})\;f )(s) = e^{2\pi i \theta}e^{2\pi i qs}f(s+p) $$

$\endgroup$
2
  • $\begingroup$ Of course you don't mean that the coefficient functions of a 1-dimensional representation are constant, right?, only that they are equal to the original representation. $\endgroup$
    – LSpice
    Commented Oct 10, 2017 at 19:44
  • $\begingroup$ @LSpice Oops. That's what comes when you write with one eye on the bus timetable. Fixed now, hopefully. $\endgroup$
    – Yemon Choi
    Commented Oct 10, 2017 at 19:54

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .