Math 202B: Lecture 20

Let G be a finite group. recall that a unitary representation of G is a pair (V,\rho) consisting of a nonzero Hilbert space V together with a group homomorphism \varphi \colon G \to U(V), where U(V) is the group of unitary operators on V.

We are going to spend the rest of the course studying the category \mathbf{Rep}(G) whose objects are finite-dimensional unitary representations of G. Morphisms in \mathbf{Rep}(G) are defined as follows.

Definition 20.1. Let (V,\rho) and (W,\psi) be unitary representations of G. A linear transformation T \in \mathrm{Hom}(V,W) is said to intertwine the actions \varphi and \psi if

T \circ \varphi(g) = \psi(g) \circ T, \quad g \in G.

Declare the set of \mathrm{Hom}_G(V,W) of morphisms from (V,\rho) to (W,\psi) to be the set of intertwining maps as above.

We have now defined the category \mathbf{Rep}(G) of unitary representations of G. This category encodes all the ways in which G can act on finite-dimensional Hilbert spaces. It has more objects than the category \mathbf{FHil} of finite-dimensional Hilbert spaces, but its Hom-spaces are smaller.

Problem 20.1. Prove that \mathrm{Hom}_G(V,W) is a subspace of \mathrm{Hom}(V,W). Moreover, prove that if T \in \mathrm{Hom}_G(V,W) then T^* \in \mathrm{Hom}_G(W,V).

Now let us discuss a few ways in which the definition of \mathrm{Hom}_G(V,W) is well-chosen.

First consider \mathrm{End}_G(V)=\mathrm{Hom}_G(V,V), the space of endomorphisms of a unitary representation (V,\varphi) of G. This is the subspace of \mathrm{End}(V)=\mathrm{Hom}(V,V) consisting of linear operators A on V which commute with each of the unitary operators \varphi(g),

T \varphi(g) = \varphi(g)T, \quad g \in G.

Equivalently, if we let (V,\Phi) be the linear representation of \mathcal{C}(G) corresponding to the unitary representation (V,\phi) via \Phi(E_g)=\varphi(g), then \mathrm{End}_G(V) is the centralizer of the image of \mathcal{C}(G) in \mathrm{End}(V) under \Phi. Thus, since the centralizer of any subalgebra is again a subalgebra, we have that \mathrm{End}_G(V) is a subalgebra of \mathrm{End}(V).

Isomorphisms in \mathbf{FHIl} are easy to understand: \mathrm{Iso}(V,W) is nonempty if and only if \dim V=\dim W, and it consists of all linear bijections T \colon V \to W. We know from Math 202A that if \mathrm{Iso}(V,W) is nonempty then there exists an isometry U \in \mathrm{Iso}(V,W). The set \mathrm{Iso}_G(V,W) of isomorphisms between unitary representations (V,\varphi) and (W,\psi) is more constrained, since it consists of linear bijections T \colon V \to W which intertwine the actions \varphi and \psi, and it may be empty even if a linear bijection V \to W exists.

Problem 20.2. Prove that \mathrm{Iso}_G(V,W) is nonempty if and only if it contains an isometry U \colon V \to W. (This can be done using either singular value decomposition or polar decomposition).

Problem 20.2 gives a matrix interpretation of what it means for two unitary representations (V,\varphi) and (W,\psi) to be isomorphic. Namely, let U \in \mathrm{Iso}_G(V,W) be an isometry. Choose any orthonormal basis X \subset V, and let Y = \{Ux \colon x \in V\} be the corresponding orthonormal basis of W. Then, we have

\langle y',\psi(g)y\rangle_W = \langle Ux',\psi(g)Ux\rangle_W = \langle Ux',U\varphi(g)x\rangle_W=\langle x',\varphi(g)x\rangle_V.

In other words, for any ordering x_1,\dots,x_n, taking the corresponding ordering Ux_1,\dots,Ux_n of Y we have that the n \times n unitary matrices [\varphi(g)]_X and [\psi(g)]_Y coincide for all g \in G. So, isomorphic representations of G represent the elements of G as exactly the same unitary matrices. An obvious but important consequence of this is the following.

Theorem 20.1. Isomorphic objects in \mathbf{Rep}(G) have the same character.

One of the many miracles of group representation theory is that the converse of Theorem 20.1 holds: the character of a unitary representation characterizes it up to isomorphism (hence the name). As a first step towards establishing this result, we consider another important aspect of \mathbf{Rep}(G), namely that given any two objects we can build a third as follow.

Definition 20.2. Let (V,\rho) and (W,\psi) be unitary representations of G. The corresponding adjoint representation (\mathrm{Hom}(V,W),\omega) is defined by

\omega(g)T=\psi(g)T\varphi(g^{-1}), \quad g \in G.

Let us place this construction in a more general framework. Given two finite-dimensional Hilbert spaces V and W, equip \mathrm{Hom}(V,W) with the Hilbert-Schmidt scalar product.

Definition 20.3. The adjoint action of U(V) \times U(W) on \mathrm{Hom}(V,W) is the group homomorphism

\omega \colon U(V) \times U(W) \longrightarrow U(\mathrm{Hom}(V,W))

defined by

\omega(A,B)T=BTA^*.

Problem 20.3. Check that \omega(A,B) is indeed a unitary operator on \mathrm{Hom}(V,W), and that \omega is indeed a group homomorphism.

Now let us calculate the matrix elements of \omega(A,B) \in U(\mathrm{Hom}(V,W)) in terms of the matrix elements of A \in U(V) and B\in U(W). To do so, let X \subset V and Y \subset W be orthonormal bases, and let \{E_{yx} \colon x \in X, y \in Y\} be the corresponding orthonormal basis of \mathrm{Hom}(V,W).

Proposition 20.3. We have

\langle E_{y'x'},\omega(A,B)E_{yx}\rangle = \overline{\langle x',Ax\rangle}{\langle y',By\rangle}.

Proof: We have

\langle E_{y'x'},\omega(A,B)E_{yx}\rangle =\mathrm{Tr}\ E_{y'x'}^*BE_{yx}A^*=\mathrm{Tr}\ BE_{yx}A^*E_{x'y'},

which is the trace of an operator in \mathrm{End}(W). We compute this trace as

\mathrm{Tr}\ BE_{yx}A^*E_{x'y'}=\sum\limits_{w \in Y} \langle w,BE_{yx}A^*E_{x'y'}w\rangle = \langle y',BE_{yx}A^*x'\rangle.

Now expand

A^*x' = \sum\limits_{x'' \in X} \langle x'',A^*x'\rangle x''

to get

\langle y',BE_{yx}A^*x'\rangle=\sum\limits_{x'' \in X} \langle y',BE_{yx}x''\rangle \langle x'',A^*x'\rangle=\langle y',By\rangle \langle x,A^*x'\rangle = \overline{\langle x',Ax \rangle}\langle y',By\rangle,

which completes the proof. \square

Corollary 20.4. We have

\mathrm{Tr}\omega(A,B) = \overline{\mathrm{Tr} A}\ \mathrm{Tr}B.

Proof: Proposition 20.3 gives

\mathrm{Tr}\omega(A,B) = \sum\limits_{x \in X}\sum\limits_{y \in Y}\langle E_{yx},\omega(A,B)E_{yx}\rangle =\sum\limits_{x \in X}\sum\limits_{y \in Y} \overline{\langle x,Ax \rangle}\langle y,By\rangle=\overline{\mathrm{Tr} A}\ \mathrm{Tr}B,

as claimed. \square

From the above, we get the following relationship between the characters of two given unitary representations (V,\varphi) and (W,\psi), and the character of the corresponding adjoint representation (\mathrm{Hom}(V,W),\omega).

Corollary 20.5. For all g \in G, we have

\chi^{\mathrm{Hom}(V,W)}(g) = \overline{\chi^V(g)}\chi^W(g).

Leave a Reply