Math 202C: Lecture 13

*** Problems due May 3 at 23:59 ***

Let H \leq G be finite groups. In Lecture 12, we proved that restriction of irreducible representations of G to H is multiplicity-free if and only if the centralizer Z(\mathcal{C}(H),\mathcal{C}(G)) is a commutative subalgebra of \mathcal{C}(G). Our objective now is to use this criterion to establish that restriction of irreducible representations of S_n to S_{n-1} is multiplicity free.

For the moment we stay in the general context where the group G and its subgroup H are arbitrary; eventually we will set G=S_n and H=S_{n-1}. By definition, the centralizer Z(\mathcal{C}(H),\mathcal{C}(G)) is the set of all |A\rangle \in \mathcal{C}(G) which commute with every |B\rangle \in \mathcal{C}(H), and it is clear that this condition is equivalent to

|A\rangle|h\rangle = |h\rangle|A\rangle, \quad h \in H,

which is in turn equivalent to

|A\rangle = |h\rangle|A\rangle|h\rangle^*, \quad h \in H,

with |h\rangle^*=|h^{-1}\rangle. To understand those elements |A\rangle \in \mathcal{C}(H) which have the above property, consider an equivalence relation on G defined by

g_1 \sim g_2 \iff g_2=hg_1h^{-1} \text{ for some} h \in H.

That is, equivalence classes under this relation are orbits of H acting on G by conjugation. Let \Lambda be a set parameterizing the distinct H-conjugacy classes in G,

\{H-\text{conjugacy classes in }G\} = \{C_\alpha \colon \alpha \in \Lambda\}.

In the case where H=G, this is just the set of conjugacy classes in G as per usual, making the following a generalization of something we already know.

Problem 13.1. Prove that the set \{|C_\alpha\rangle \colon \alpha \in \Gamma\} is an orthogonal basis of Z(\mathcal{C}(H),\mathcal{C}(G)).

In the case H=G, the centralizer Z(\mathcal{C}(H),\mathcal{C}(G)) is the center Z(\mathcal{C}(G)), which is commutative by definition, and the above problem recovers the 202B fact that the center of the convolution algebra of a group has a basis consisting of indicator functions of conjugacy classes. At the other extreme, if H=\{\iota\} is the subgroup consisting solely of the identity element, then Z(\mathcal{C}(H),\mathcal{C}(G))=\mathcal{C}(G), which is commutative precisely when G is abelian. In between these two extremes, it may be more difficult to determine whether or not Z(\mathcal{C}(H),\mathcal{C}(G)) is commutative, but there is a particular circumstance which will force this to be the case.

Problem 13.2. Prove that if the class basis \{|C_\alpha\rangle \colon \alpha \in \Lambda\} consists of selfadjoint elements, then Z(\mathcal{C}(H),\mathcal{C}(G)) is a commutative algebra.

We now specialize to the case G=S_n and H=S_{n-1}. We would like to find a concrete indexing of the S_{n-1}-conjugacy classes in S_n. For permutations, the operation of conjugation has a simple combinatorial meaning: conjugating g by h relabels the elements in the cycles of g according to h. If h is confined to S_{n-1}, then it cannot relabel n, hence in order for two permutations g_1 and g_2 in S_n to be S_{n-1}-conjugates it is necessary that the cycle containing n in each have the same length.

Problem 13.3. Prove that the above condition is also sufficient for g_1 and g_2 to be S_{n-1}-conjugates.

We can now conclude that S_{n-1}-conjugacy classes in S_n are indexed by pairs (r,\alpha) consisting of an integer 1 \leq r \leq n corresponding to the length of the cycle containing n and a partition \alpha \vdash n-r consisting to the cycle-type of the remainder of the permutation. In particular, we get the dimension formula

\dim Z(\mathcal{C}(S_{n-1},\mathcal{C}(S_n))=\sum\limits_{r=1}^n p(n-r),

where p(k) is the number of partitions of k. Furthermore, it is easy to see that if an S_{n-1}-conjugacy class C_{(r,\alpha)} contains a permutation \pi, then it also contains \pi^{-1}, since the inverse of a permutation not only has the same cycle type, but actually the same elements within each cycle. Thus, the centralizer Z(\mathcal{C}(S_{n-1}),\mathcal{C}(S_n)) is commutative by Problem 13.2.

The significance of this commutativity is that, by Lecture 12, it is equivalent to multiplicity-free branching for irreducible representations of the symmetric groups: the isotypic decomposition of any irreducible representation V^\lambda of S_n when viewed as a representation of S_{n-1} has the form

V^\lambda = \bigoplus_{\mu \vdash n-1} \mathrm{Mult}(V^\mu,V^\lambda)V^\mu, \quad \mathrm{Mult}(V^\mu,V^\lambda) \in \{0,1\}.

In Lecture 14, we will use this fact to construct a special orthonormal basis in V^\lambda which we will then exploit mercilessly.

Leave a Reply