Polynomials and Representations XXVIII

Starting from this article, we will look at representations of G := GL_n\mathbb{C}. Now, GL_n\mathbb{C} itself is extremely complicated so we will only focus on representations of particular types. Generally, for any topological group G, we want:

\rho : G\to GL_N \mathbb{C}

to be a continuous homomorphism of groups.

blue-lin

Continuous Representations of Topological Groups

Let G be a topological group.

Our main objects of study are continuous representations \rho: G \to GL_N\mathbb{C}. These correspond to \mathbb{C}[G]-modules (also called G-modules) V of finite complex dimension, with an additional condition for continuity:

  • for any v\in V and \alpha \in V^\vee (dual), the map G\to \mathbb{C} given by g\mapsto \alpha(g(v)) is continuous.

[Note the above condition condition is independent of coordinates and does not require a choice of basis. When we replace G with GL_n\mathbb{C} later, we can also replace the word continuous with “analytic” to obtain a narrower class of representations, but let’s not get ahead of ourselves.]

A function G\to\mathbb{C} of the form g\mapsto \alpha(g(v)) as above is called a matrix coefficient. E.g. if we fix a basis for V which gives us G \to GL_N\mathbb{C}, picking entry (ij) in the matrix gives us a matrix coefficient G\to\mathbb{C}.

Henceforth, all representations of topological groups are assumed to be continuous and finite-dimensional.

Constructing New Representations

These constructions are identical to the case for representations of finite groups.

  • Given any complex vector space V, the trivial action takes gvv for any g\in G, v\in V; it is clearly continuous.
  • Given a G-module V, a vector space W\subseteq V is said to be Ginvariant if g\in G, w\in W \implies g(w) \in W. Clearly, the resulting G\to GL(W) is continuous.
    • We say that V is irreducible if V\ne 0 and it has no non-trivial G-invariant submodules. An irreducible representation V is often called irrep for short.
  • If W\subseteq V is G-invariant, then V/W is a G-module; clearly it is continuous.
  • If V is a G-module, the dual V^\vee is also continuous, where G acts via

G\times V^\vee \to V^\vee, \qquad (g, \alpha) \mapsto \alpha\circ g^{-1} \in V^\vee.

  • Given G-modules V, W, the action of G on V\oplus W and V \otimes_{\mathbb C} W are continuous as well.
  • Finally we have \text{Hom}_{\mathbb C}(V, W) \cong V^\vee \otimes_{\mathbb C} W naturally. Using the above constructions, we get a continuous G-representation for this space via:

g\in G, f:V\to W \mapsto \rho_W(g)\circ f \circ \rho_V(g^{-1}) : V\to W.

How do we know that the above constructions are all continuous?

Here’s one example. To show that V\otimes_{\mathbb{C}} W is a continuous representation, pick a basis e_i (resp. f_j) of V (resp. W) so that e_i \otimes f_j form a basis of V\otimes W. The matrix for g: V\otimes W \to V\otimes W is obtained from the matrices of g:V\to V and g:W\to W by multiplying entries; thus the map G\to \mathbb{C} taking g to a fixed entry of the matrix for g:V\otimes W\to V\otimes W is continuous.

Finally, a linear map of G-modules f:V\to W is said to be Gequivariant if

g\in G, v\in V \implies f(gv) = g\cdot f(v).

Note that this is equivalent to saying G acts trivially on f\in \text{Hom}_{\mathbb C}(V, W).

blue-lin

Compact Groups

From now till the end of the next article, G is assumed to be compact and Hausdorff; this case will be instrumental to the general theory even though GL_n\mathbb{C} is not compact. Here are some common examples of compact topological groups in representation theory:

\begin{aligned} U_n &= \{ M \in GL_n \mathbb{C} : M \overline{M}^t = I\},\\ SU_n &= \{ M \in GL_n\mathbb{C} : M\overline{M}^t = I, \det(M) = 1\},\\ O_n &=\{ M\in GL_n\mathbb{R} : M M^t = I\}, \\ SO_n &= \{M \in GL_n\mathbb{R} : MM^t = I, \det(M) = 1\}, \\ T &= \{ z \in \mathbb{C} : |z| = 1\}.\end{aligned}

These groups are compact because they are closed and bounded subsets of \mathbb{C}^{n\times n}. E.g. the condition for M \in U_n can be written out as a sequence of quadratic equations in \text{Re}(z_{ij}) and \text{Im}(z_{ij}) for 1\le i,j\le n. Thus the set U_n is closed; it is bounded since the condition gives:

|z_{i1}|^2 + |z_{i2}|^2 + \ldots + |z_{in}|^2 = 1, \quad\text{for } 1\le i\le n

so each |z_{ij}| \le 1.

Haar Measure

To proceed, we will borrow a result from harmonic analysis:

Theorem. If G is a compact Hausdorff group, then there is a measure m on G (called the Haar measure) satisfying:

  • m(G) = 1;
  • m is “left-invariant”, i.e. for any continuous f:G\to \mathbb{C}, we have:

\displaystyle  y\in G\implies \int_{x\in G} f(yx)\cdot dm(x) = \int_{x\in G} f(x)\cdot dm(x),

The Haar measure is unique and in fact right-invariant, i.e. \int_{x\in G} f(xy)\cdot dm(x) = \int_{x\in G} f(x) \cdot dm(x). We will implicitly assume the Haar measure is used when integrating over G.

Note

The Haar measure is usually defined for locally compact Hausdorff groups in general, in which case we have to modify the statements a little. In particular, left-invariant measures are not necessarily right-invariant.

blue-lin

Semisimplicity of Continuous Representations

The following result is crucial.

Theorem. Let V be a G-module. If W\subseteq V is a G-invariant subspace, then we can find a G-invariant subspace W' \subseteq V such that V = W\oplus W'.

Proof

Let \pi : V\to V be any projection map onto W, so \pi^2 = \pi and \text{im}(\pi) = W. Now define:

\pi' : V\to V, \quad \pi'(v) := \int_{x\in G} x\pi(x^{-1}v)\cdot dx.

[Explicitly, one may take an isomorphism V\cong \mathbb{C}^N so each v\in V is a tuple of complex numbers; the above integral can be carried out component-wise, as N integrals of continuous functions.]

We then have:

  • The image of \pi' is in W, since \pi(V)=W and x(W)\subseteq W for x\in G.
  • If v\in W then x^{-1}v \in W\implies x\pi(x^{-1}v) = xx^{-1}v = v and \pi'(v) = v.
  • \pi'^2 = \pi': indeed if v\in V then \pi'(v) \in W by the first property and so \pi'(\pi'(v)) = \pi'(v) by the second.
  • The map \pi' is G-equivariant: for y\in G we have \pi'(yv) = y\pi'(v). Indeed by left-invariance we have: \pi'(v) = \int_{x\in G} yx \pi(x^{-1}y^{-1}v) dx. Replacing v with yv gives us the desired outcome.

Hence, \pi' is projection onto W and its kernel W’ satisfies V = W \oplus W'. Since \pi' is G-equivariant, W’ is a G-invariant subspace. ♦

Thus, every continuous representation of G can be written as a direct sum of irreps. Next, we have:

Schur’s Lemma. If V, W are irreps of G, then:

\text{Hom}_{\mathbb{C}[G]}(V, W) \cong \begin{cases} \mathbb{C}, \qquad &\text{if }V \cong W,\\ 0, \qquad &\text{otherwise.} \end{cases}

The proof is identical to that of the usual Schur’s lemma.

Proof

Suppose V\not\cong W and f:V\to W is G-equivariant. Then \text{ker}(f) \subseteq V and \text{im}(f)\subseteq W are subrepresentations, so \text{ker}(f) = 0, V and \text{im}(f) = 0, W. By considering various cases we have either

  1. \text{ker}(f) = 0, \text{im}(f) = W, in which case f is an isomorphism or
  2. \text{ker}(f) = V, \text{im}(f) = 0, in which case f = 0.

It remains to show that \text{Hom}_{\mathbb{C}[G]}(V,W) \cong \mathbb{C} when V\cong W.

It suffices to show that any isomorphism f:V\to V is a scalar: let v\in V be an eigenvector of with eigenvalue \lambda\in\mathbb{C} so f(v) = \lambda v. Then f - \lambda\cdot I : V\to V is G-equivariant and its kernel is non-zero; hence f - \lambda\cdot I = 0 \implies f = \lambda\cdot I. ♦

blue-lin

This entry was posted in Uncategorized and tagged , , , , , . Bookmark the permalink.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s