Let $G$ be a simple compact connected Lie group and let $K$ be a connected closed subgroup of $G$. Let $\widehat G$ and $\widehat K$ denote the corresponding unitary duals, that is, the (equivalence classes of) irreducible representations. By the Highest Weight Theorem, elements in $\widehat G$ are in correspondence with elements in the set $P^+(G)$ of $G$-integral dominant weights. For $\lambda\in P^+(G)$, we will denote by $\pi_\lambda$ the irreducible representation of $G$ with highest weight $\lambda$.

Given $\tau\in\widehat K$, we define $\widehat G_\tau=\{\pi\in\widehat G: \textrm{Hom}_K(\pi,\tau)\neq0\}$. In other words, $\pi$ is in $\widehat G_\tau$ if and only if $\tau$ appears in the decomposition in irreducible representations of the restriction of $\pi|_K$ (the restriction of $\pi$ to $K$). Elements in $\widehat G_\tau$ are called $\tau$-spherical representations.

I wish to know if there is $\omega\in P^+(G)$, with $\omega\neq0$, satisfying the following property:

if $\lambda\in P^+(G)$ satisfies $\pi_\lambda\in\widehat G_\tau$, then $\pi_{\lambda+\omega}\in\widehat G_\tau$.

Here I list some examples coming from compact symmetric spaces of real rank one. In each case I used the obvious inclusion $K\subset G$ and standard conventions on the root system of $\mathfrak g$:

  1. $G$ arbitrary and $K=\{1\}$, $\omega =$ any choice in $P^+(G)$.
  2. $G=SU(2)$ and $K=T$ (maximal torus), $\omega = 2\varepsilon_1$ (twice the first fundamental weight).
  3. $G=SO(n+1)$ and $K=SO(n)$, $\omega=\varepsilon_1$ (the first fundamental weight).
  4. $G=SU(n+1)$ and $K=S(U(n)\times U(1))$, $\omega=\varepsilon_1-\varepsilon_{n+1}$ (the first plus the last fundamental weights).
  5. $G=Sp(n+1)$ and $K=Sp(n)$, $\omega=\varepsilon_1+\varepsilon_2$ (the second fundamental weight).
  6. $G=SO(2n)$ and $K=T$ (max. torus), $\omega= 2\varepsilon_1$ (twice the first fundamental weight).

Each of the above examples follows by the explicit branching law from $G$ to $K$, which is (of course) not available for arbitrary $K$. Clearly, in each case, any positive multiple of the indicated $\omega$ also works.

In case the answer is affirmative, it follows that $\widehat G_\tau$ can always be written as a union of subsets of the form $\{\pi_{\lambda+k\omega}: k\geq0 \}$ for some $\lambda\in P^+(G)$. (The union is not assumed finite or either disjoint.) This is the result I am looking for.

Edit: Looking for more information, here is an intermediate question:

For $G=SO(2n)$. Does a positive multiple of $\varepsilon_1$ satisfy the above property for any closed subgroup $K$ of $G$?

Either a proof or a counterexample would help me to understand this problem. I picked $SO(2n)$ because is my favorite, but we may replace $(SO(2n),\varepsilon_1)$ in the question by $(SO(2n+1),\varepsilon_1)$, $(SU(n+1),\varepsilon_1-\varepsilon_{n+1})$, or $(Sp(n+1),\varepsilon_1+\varepsilon_2)$.

  • 2
    $\begingroup$ You probably meant $K=T$, the maximal torus of $G$, in examples 1 and 2. If $K=\{1\}$ then any representation of $G$ is spherical, so the property trivially holds for any $\omega$. Also, you should exclude the dominant weight $\omega=0$, again, the property holds trivially. $\endgroup$ Apr 11 '19 at 5:19
  • $\begingroup$ Although I had written what I wanted, it was confusing and your idea is better. I have just edited. Thanks! $\endgroup$
    – emiliocba
    Apr 11 '19 at 10:04

It is true that there exists such a $\omega$. Indeed, $every$ $\omega \neq 0$ works. Fix a representation $\pi _{\omega}$ which has a nonzero $K$ fixed vector, with $\omega \neq 0$. One can replace $G,K$ by their complexifications. Fix a Borel subgroup $B$ of $G(\mathbb C)$ and view the vectors in the representations $\pi _{\lambda}$ as algebraic sections $\phi (x)$ of a homogeneous line bundle ${\mathcal L}_{\lambda}$ on $G/B$ (the Borel-Weil theorem) . Fix a section $\phi (x)$ whose $K$ translates generate a representation of $K(\mathbb C)$ isomorphic to $\tau$.

If $\psi (x)$ is the non-zero $K$ fixed vector in $\pi _{\omega}$ viewed as a section of ${\mathcal L}_{\omega}$, then the product $\phi (x)\psi (x)$ is a section of ${\mathcal L}_{\lambda +\omega}$. But the $K$ translates of the product $\phi \psi$ generate a representation also isomorphic to $\tau$ since $\psi$ is $K$ invariant. Hence the restriction of $\pi _{\lambda +\omega}$ to $K$ contains $\tau$.

  • $\begingroup$ So, the highest weight of any non-trivial representation in $\widehat G_{1_K}$ works ($1_K=$ trivial representation of $K$). In particular, the set of highest weight of elements in $\widehat G_{1_K}$ is a monoid ($0$ is in it, and the sum of any elements in it is again in it). Is all this right? In the affirmative case, thanks! $\endgroup$
    – emiliocba
    Apr 12 '19 at 11:24
  • 1
    $\begingroup$ @emiliocba: Yes, that is correct. I do not know a reference but the proof is obvious given the Borel-Weil theorem. $\endgroup$ Apr 12 '19 at 11:45
  • $\begingroup$ In fact, the situation is controlled by an affine semigroup $A_{G,K}$ contained in $P(G)^{+}\times P(K)^{+}$ formed by the pairs $(\lambda,\tau)$ for which the restriction of the simple $G$-module with highest weight $\lambda$ to $K$ contains the simple $K$-module with highest weight $\tau$. This branching semigroup provides a multigrading on the branching algebra of $K$-highest weight vectors in $R(G/U)$. See the work on branching algebras by Roger Howe and others. $\endgroup$ Apr 13 '19 at 5:16
  • $\begingroup$ @Victor : does it mean what I have said is incorrect? I thought I gave a proof. $\endgroup$ Apr 13 '19 at 5:39
  • $\begingroup$ @Venkataramana: of course, your proof works. I was pointing out the additional structure present, related to the monoid OP asked about, and commented on references. Viewing all line bundles $\mathcal{L}_\lambda$ simultaneously by considering $R(G/U)$ with the grading coming from the right $T$-action instead of individual spaces $\Gamma(G/B,\mathcal{L}_\lambda)$ streamlines the description and the proof. $\endgroup$ Apr 13 '19 at 15:48

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.