2.9. Lecture 9

2.9.1. Schur Orthogonality

The proof of Theorem 2.8.7 is split into two parts. First, in this section we show that the irreducible characters form an orthonormal set. Later on we will show that this set spans CF(G)CF(G).

Given two representations π,ρ\pi,\rho, we can create a new representation:

Lemma 2.9.1.

Let (π,V)(\pi,V) and (ρ,W)(\rho,W) be two representations of a group GG. The pair (cπρ,Hom(V,W))\big{(}c_{\pi}^{\rho},\operatorname{Hom}(V,W)\big{)} is a representation of GG given by

cπρ(g)(T):=ρ(g)Tπ(g1),c_{\pi}^{\rho}(g)(T):=\rho(g)T\pi(g^{-1}),

for THom(V,W)T\in\operatorname{Hom}(V,W).

Proof.

This is left as an exercise (see Problem 2.9.3). ∎

Lemma 2.9.2.

.

  1. (i)

    HomG(V,W)={THom(V,W)|cπρ(g)T=T,gG}\operatorname{Hom}_{G}(V,W)=\{T\in\operatorname{Hom}(V,W)\,|\,c_{\pi}^{\rho}(g% )T=T,\,\forall g\in G\}.

  2. (ii)

    χcπρ=χρχπ¯\chi_{c_{\pi}^{\rho}}=\chi_{\rho}\,\overline{\chi_{\pi}}.

Proof.

Part (i) follows directly from the definition of HomG(V,W)\operatorname{Hom}_{G}(V,W):

THomG(V,W)\displaystyle T\in\operatorname{Hom}_{G}(V,W) ρ(g)T=Tπ(g)gG\displaystyle\Leftrightarrow\rho(g)T=T\pi(g)\;\forall g\in G
ρ(g)Tπ(g1)=TgG\displaystyle\Leftrightarrow\rho(g)T\pi(g^{-1})=T\;\forall g\in G
cπρ(g)T=TgG,\displaystyle\Leftrightarrow c_{\pi}^{\rho}(g)T=T\;\forall g\in G,

as claimed.

For part (ii), given gGg\in G, let {𝐯i}\{{\bf v}_{i}\} be an eigenbasis of VV for π(g)\pi(g) (with π(g)𝐯i=λi𝐯i\pi(g){\bf v}_{i}=\lambda_{i}{\bf v}_{i}) and {𝐰j}\{{\bf w}_{j}\} an eigenbasis of WW for ρ(g)\rho(g) (with ρ(g)𝐰j=μj𝐰j\rho(g){\bf w}_{j}=\mu_{j}{\bf w}_{j}). Then {Ti,j}\{T_{i,j}\} forms a basis of Hom(V,W)\operatorname{Hom}(V,W), where Ti,jT_{i,j} is the linear map defined by

Ti,j(𝐯k)={𝟎ifki,𝐰jifk=i.T_{i,j}({\bf v}_{k})=\begin{cases}{\bf 0}\quad&\mathrm{if\;}k\neq i,\\ {\bf w}_{j}\quad&\mathrm{if\;}k=i.\end{cases}

The set {Ti,j}\{T_{i,j}\} forms an eigenbasis of Hom(V,W)\operatorname{Hom}(V,W) for cπρ(g)c_{\pi}^{\rho}(g):

(cπρ(g)Ti,j)(𝐯k)=ρ(g)Ti,j(π(g1)𝐯k)=ρ(g)Ti,j(λk¯𝐯k)\displaystyle\big{(}c_{\pi}^{\rho}(g)T_{i,j}\big{)}({\bf v}_{k})=\rho(g)T_{i,j% }\big{(}\pi(g^{-1}){\bf v}_{k}\big{)}=\rho(g)T_{i,j}(\overline{\lambda_{k}}{% \bf v}_{k}) =λk¯ρ(g)Ti,j(𝐯k)\displaystyle=\overline{\lambda_{k}}\rho(g)T_{i,j}({\bf v}_{k})
={λk¯ρ(g)𝟎ifki,λi¯ρ(g)𝐰jifk=i\displaystyle=\begin{cases}\overline{\lambda_{k}}\rho(g){\bf 0}\quad&\mathrm{% if\;}k\neq i,\\ \overline{\lambda_{i}}\rho(g){\bf w}_{j}\quad&\mathrm{if\;}k=i\end{cases}
={𝟎ifki,λi¯μj𝐰jifk=i=λi¯μjTi,j(𝐯k).\displaystyle=\begin{cases}{\bf 0}\quad&\mathrm{if\;}k\neq i,\\ \overline{\lambda_{i}}\mu_{j}{\bf w}_{j}\quad&\mathrm{if\;}k=i\end{cases}=% \overline{\lambda_{i}}\mu_{j}T_{i,j}({\bf v}_{k}).

Thus,

χcπρ(g)=i,jλi¯μj=(iλi¯)(jμj)=χρ(g)χπ(g)¯.\chi_{c_{\pi}^{\rho}}(g)=\sum_{i,j}\overline{\lambda_{i}}\mu_{j}=\left(\sum_{i% }\overline{\lambda_{i}}\right)\left(\sum_{j}\mu_{j}\right)=\chi_{\rho}(g)% \overline{\chi_{\pi}(g)}.

We make another useful definition:

Definition 2.9.3.

Let (π,V)(\pi,V) be a representation of GG. Define

VG={𝐯G|π(g)𝐯=𝐯,gG},V^{G}=\{{\bf v}\in G\,|\,\pi(g){\bf v}={\bf v},\,\forall g\in G\},

to be the set of fixed points of (π,V)(\pi,V).

Note that (π,VG)(\pi,V^{G}) is a subrepresentation of GG, and in fact (π,VG)=(Id,VG)(\pi,V^{G})=({\,\mathrm{Id}},V^{G}).

Lemma 2.9.4.

Let (π,V)(\pi,V) be a representation of GG. Then

dim(VG)=1|G|gGχπ(g).\operatorname{dim}(V^{G})=\frac{1}{|G|}\sum_{g\in G}\chi_{\pi}(g).
Proof.

Let ,\langle\cdot,\cdot\rangle be a GG-invariant inner product on VV. We write

(π,V)=(π,VG)(π,(VG)),(\pi,V)=(\pi,V^{G})\oplus\big{(}\pi,(V^{G})^{\perp}\big{)},

and let 𝐯1,,𝐯dim(VG){\bf v}_{1},\ldots,{\bf v}_{\operatorname{dim}(V^{G})} be a basis of VGV^{G}, and 𝐰1,,𝐰dim((VG)){\bf w}_{1},\ldots,{\bf w}_{\operatorname{dim}\big{(}(V^{G})^{\perp}\big{)}} a basis of (VG)(V^{G})^{\perp}. Then 𝐯1,,𝐯dim(VG),𝐰1,,𝐰dim((VG)){\bf v}_{1},\ldots,{\bf v}_{\operatorname{dim}(V^{G})},{\bf w}_{1},\ldots,{\bf w% }_{\operatorname{dim}\big{(}(V^{G})^{\perp}\big{)}} is a basis of VV, which we denote by {\mathcal{B}}.

Defining S=1|G|hGπ(h)Hom(V,V)S=\frac{1}{|G|}\sum_{h\in G}\pi(h)\in\operatorname{Hom}(V,V), observe that for any 𝐯V{\bf v}\in V, we have

π(g)S𝐯=π(g)1|G|hGπ(h)𝐯=1|G|hGπ(gh)𝐯=1|G|hGπ(h)𝐯=S𝐯,\pi(g)S{\bf v}=\pi(g)\frac{1}{|G|}\sum_{h\in G}\pi(h){\bf v}=\frac{1}{|G|}\sum% _{h\in G}\pi(gh){\bf v}=\frac{1}{|G|}\sum_{h\in G}\pi(h){\bf v}=S{\bf v},

hence S𝐯VGS{\bf v}\in V^{G}. Moreover, if 𝐯VG{\bf v}\in V^{G}, then

S𝐯=1|G|hGπ(h)𝐯=1|G|hG𝐯=1|G||G|𝐯=𝐯,S{\bf v}=\frac{1}{|G|}\sum_{h\in G}\pi(h){\bf v}=\frac{1}{|G|}\sum_{h\in G}{% \bf v}=\frac{1}{|G|}|G|{\bf v}={\bf v},

hence S|VG=IdS|_{V^{G}}={\,\mathrm{Id}}. On the other hand, note that since (π,(VG))(\pi,(V^{G})^{\perp}) is a subrepresentation, we have π(g)𝐯(VG)\pi(g){\bf v}\in(V^{G})^{\perp} for all 𝐯(VG){\bf v}\in(V^{G})^{\perp} and gGg\in G, hence S𝐯(VG)S{\bf v}\in(V^{G})^{\perp} for all 𝐯(VG){\bf v}\in(V^{G})^{\perp}. Since VG(VG)={0}V^{G}\cap(V^{G})^{\perp}=\{0\}, S|(VG)=0S|_{(V^{G})^{\perp}}=0, so writing the matrix of SS with respect to the basis {\mathcal{B}} gives

[S]=diag(1,1,,1,0,0,0),[S]_{{\mathcal{B}}}=\operatorname{diag}\big{(}1,1,\ldots,1,0,0\ldots,0\big{)},

with dim(VG)\operatorname{dim}(V^{G}) ones, and dim((VG))\operatorname{dim}\big{(}(V^{G})^{\perp}\big{)} zeroes. In conclusion,

dim(VG)=tr([S])=tr(1|G|hGπ(h))=1|G|hGtr(π(h))=1|G|hGχπ(h),\operatorname{dim}(V^{G})=\operatorname{tr}\big{(}[S]_{{\mathcal{B}}}\big{)}=% \operatorname{tr}\left(\frac{1}{|G|}\sum_{h\in G}\pi(h)\right)=\frac{1}{|G|}% \sum_{h\in G}\operatorname{tr}(\pi(h))=\frac{1}{|G|}\sum_{h\in G}\chi_{\pi}(h),

as claimed. ∎

Remark 2.9.5.

The main part of the proof of Lemma 2.9.4 shows that 1|G|gGπ(g)\frac{1}{|G|}\sum_{g\in G}\pi(g) is an orthogonal projection onto VGV^{G}; this is an important result that we will generalise further next time.

Lemma 2.9.6.

Let (π,V)(\pi,V) and (ρ,W)(\rho,W) be two representations of a group GG. Then

χρ,χπG=dimHomG(V,W).\langle\chi_{\rho},\chi_{\pi}\rangle_{G}=\operatorname{dim}\operatorname{Hom}_% {G}(V,W).
Proof.

We will apply Lemma 2.9.4 to the representation (cπρ,Hom(V,W))\big{(}c_{\pi}^{\rho},\operatorname{Hom}(V,W)\big{)}. By Lemma 2.9.2(i) and Lemma 2.9.4,

dim(HomG(V,W))=dim(Hom(V,W)G)=1|G|gGχcπρ(g).\operatorname{dim}\big{(}\operatorname{Hom}_{G}(V,W)\big{)}=\operatorname{dim}% \big{(}\operatorname{Hom}(V,W)^{G}\big{)}=\frac{1}{|G|}\sum_{g\in G}\chi_{c_{% \pi}^{\rho}}(g).

Now applying Lemma 2.9.2(ii), we have χcπρ(g)=χρ(g)χπ(g)¯\chi_{c_{\pi}^{\rho}}(g)=\chi_{\rho}(g)\overline{\chi_{\pi}(g)}, and so

dim(HomG(V,W))=1|G|gGχρ(g)χπ(g)¯=χρ,χπG.\operatorname{dim}\big{(}\operatorname{Hom}_{G}(V,W)\big{)}=\frac{1}{|G|}\sum_% {g\in G}\chi_{\rho}(g)\overline{\chi_{\pi}(g)}=\langle\chi_{\rho},\chi_{\pi}% \rangle_{G}.

2.9.2. Consequences of the inner product formula

We collect a few important consequences of Lemma 2.9.6:

Proof of orthogonality in Theorem 2.8.7.

By Schur’s Lemma (Corollary 1.4.2) and Lemma 2.9.6: if π\pi and ρ\rho are irreducible representations of GG, then

χπ,χρG=dim(HomG(V,W))={1if(π,V)(ρ,W),0otherwise.\langle\chi_{\pi},\chi_{\rho}\rangle_{G}=\operatorname{dim}\big{(}% \operatorname{Hom}_{G}(V,W)\big{)}=\begin{cases}1\quad&\mathrm{if\;}(\pi,V)% \cong(\rho,W),\\ 0\quad&\mathrm{otherwise}.\end{cases}

Theorem 2.9.7.

.

  1. (i)

    Two representations are isomorphic if and only if they have the same character.

  2. (ii)

    A representation (π,V)(\pi,V) is irreducible if and only if

    χπG2=χπ,χπG=1.\|\chi_{\pi}\|_{G}^{2}=\langle\chi_{\pi},\chi_{\pi}\rangle_{G}=1.
Proof.

The fact that isomorphic representations have the same character has been proved already (Lemma 2.7.5), so for (i), we need to show the other direction. Letting (π,V)(\pi,V) be any representation of GG, we can decompose π\pi into irreducibles by Theorem 1.5.12:

(π,V)ρIrr(G)(ρ,Wρ)nρ,(\pi,V)\cong\bigoplus_{\rho\in\mathrm{Irr}(G)}(\rho,W_{\rho})^{n_{\rho}}, (2.9.8)

where the nρn_{\rho} are non-negative integers. Writing χπ=ρnρχρ\chi_{\pi}=\sum_{\rho}n_{\rho}\chi_{\rho}, for any σIrr(G)\sigma\in\mathrm{Irr}(G), we then have

χπ,χσG=ρIrr(G)nρχρ,χσG=nσ.\langle\chi_{\pi},\chi_{\sigma}\rangle_{G}=\sum_{\rho\in\mathrm{Irr}(G)}n_{% \rho}\langle\chi_{\rho},\chi_{\sigma}\rangle_{G}=n_{\sigma}.

The representation π\pi is thus completely determined by its character.

For part (ii), by (2.9.8) we have that χπG2=ρnρ2\|\chi_{\pi}\|_{G}^{2}=\sum_{\rho}n_{\rho}^{2}, which (being the sum of squares of non-negative integers) is equal to 11 if and only if a single nρn_{\rho} equals one, and the rest are zero.

2.9.3. Exercises

.

Problem 50. Verify that (cπρ,Hom(V,W))\big{(}c_{\pi}^{\rho},\operatorname{Hom}(V,W)\big{)} is a representation of GG.

The next few problems give an alternative proof of Theorem 2.8.7 (actually a slightly stronger result is proved).

Problem 51. Let (π,V)(\pi,V) be a finite-dimensional representation of a group GG. Suppose that VV is equipped with a GG-invariant inner product ,\langle\cdot,\cdot\rangle. Show that

χπ(g)=i=1dimVπ(g)𝐯i,𝐯i,\chi_{\pi}(g)=\sum_{i=1}^{\operatorname{dim}V}\langle\pi(g){\bf v}_{i},{\bf v}% _{i}\rangle,

where {𝐯i}i=1,,dim(V)\{{\bf v}_{i}\}_{i=1,\ldots,\operatorname{dim}(V)} is any orthonormal basis for VV with respect to the invariant inner product.

Remark: A function Φ:G\Phi:G\rightarrow{\mathbb{C}} of the form Φ(g)=π(g)𝐯1,𝐯2\Phi(g)=\langle\pi(g){\bf v}_{1},{\bf v}_{2}\rangle, where 𝐯1,𝐯2V{\bf v}_{1},{\bf v}_{2}\in V for some unitary representation (π,V)(\pi,V) of GG is called a matrix coefficient.

Problem 52. Let (π,V)(\pi,V) and (ρ,W)(\rho,W) be two representations of a group GG. For THom(V,W)T\in\operatorname{Hom}(V,W), define TGT^{G} by

TG=1|G|gGπ2(g)Tπ1(g1).T^{G}=\frac{1}{|G|}\sum_{g\in G}\pi_{2}(g)T\pi_{1}(g^{-1}).

Show that TGHomG(V,W)T^{G}\in\operatorname{Hom}_{G}(V,W).

Problem 53. Show that if (π,V)(\pi,V) and (ρ,W)(\rho,W) are two non-isomorphic irreducible unitary representations of GG, then their matrix coefficients are orthogonal with respect to the standard L2L^{2}-inner product on {\mathbb{C}}-valued functions on GG, i.e.

1|G|gGπ(g)𝐯1,𝐯2Vρ(g)𝐰1,𝐰2W¯=0𝐯1,𝐯2V,𝐰1,𝐰2W.\frac{1}{|G|}\sum_{g\in G}\big{\langle}\pi(g){\bf v}_{1},{\bf v}_{2}\big{% \rangle}_{V}\overline{\big{\langle}\rho(g){\bf w}_{1},{\bf w}_{2}\big{\rangle}% _{W}}=0\qquad\forall\,{\bf v}_{1},{\bf v}_{2}\in V,\,{\bf w}_{1},{\bf w}_{2}% \in W.

Hint: Define the map THom(V,W)T\in\operatorname{Hom}(V,W) by

T(v)=𝐯,𝐯1V𝐰1𝐯V.T(v)=\langle{\bf v},{\bf v}_{1}\rangle_{V}{\bf w}_{1}\qquad\forall{\bf v}\in V.

Then use Problem 2.9.3 and Schur’s lemma to study 𝐰2,TG𝐯2W\langle{\bf w}_{2},T^{G}{\bf v}_{2}\rangle_{W}.

Problem 54. Let (π,V)(\pi,V) be an irreducible unitary representation of GG. Show that

1|G|gGπ(g)𝐯1,𝐯2Vπ(g)𝐮1,𝐮2V¯=𝐯1,𝐮1V𝐯2,𝐮2V¯dimV𝐯1,𝐯2,𝐮1,𝐮2V.\frac{1}{|G|}\sum_{g\in G}\big{\langle}\pi(g){\bf v}_{1},{\bf v}_{2}\big{% \rangle}_{V}\overline{\big{\langle}\pi(g){\bf u}_{1},{\bf u}_{2}\big{\rangle}_% {V}}=\frac{\langle{\bf v}_{1},{\bf u}_{1}\rangle_{V}\overline{\langle{\bf v}_{% 2},{\bf u}_{2}\rangle_{V}}}{\operatorname{dim}V}\qquad\forall\,{\bf v}_{1},{% \bf v}_{2},{\bf u}_{1},{\bf u}_{2}\in V.

Hint: Define the map THom(V)T\in\operatorname{Hom}(V) by

T(𝐯)=𝐯,𝐯1V𝐮1𝐯V.T({\bf v})=\langle{\bf v},{\bf v}_{1}\rangle_{V}{\bf u}_{1}\qquad\forall\,{\bf v% }\in V.

Then compute 𝐮2,TG(𝐯2)V\langle{\bf u}_{2},T^{G}({\bf v}_{2})\rangle_{V} using Problem 2.9.3 and Schur’s lemma (compare with the proof of Lemma 2.9.4; show that trTG=trT\operatorname{tr}T^{G}=\operatorname{tr}T).

Problem 55. Use Problems 2.9.3 to 2.9.3 to give an alternative proof of the fact that the characters of irreducible representations form an orthonormal set in CF(G)CF(G).

The following problems give a generalisation of Fourier analysis to non-commutative finite groups:

Problem 56. For each (σ,Wσ)Irr(G)(\sigma,W_{\sigma})\in\mathrm{Irr}(G), let {𝐰σ,i}i=1,,dimσ\{{\bf w}_{\sigma,i}\}_{i=1,\ldots,\operatorname{dim}\sigma} be an orthonormal basis of WσW_{\sigma} with respect to an invariant inner product ,Wσ\langle\cdot,\cdot\rangle_{W_{\sigma}}. Show that {Φσ,i,j}σIrr(G),1i,jdimσ\{\Phi_{\sigma,i,j}\}_{\sigma\in\mathrm{Irr}(G),1\leq i,j\leq\operatorname{dim% }\sigma} is an orthonormal basis of L2(G)={f:G}L^{2}(G)=\{f:G\rightarrow{\mathbb{C}}\} with respect to the inner product f1,f2G=1|G|gGf1(g)f2(g)¯\langle f_{1},f_{2}\rangle_{G}=\frac{1}{|G|}\sum_{g\in G}f_{1}(g)\overline{f_{% 2}(g)}, where

Φσ,i,j(g):=dimσσ(g)𝐰σ,i,𝐰σ,jWσ.\Phi_{\sigma,i,j}(g):=\sqrt{\operatorname{dim}\sigma}\langle\sigma(g){\bf w}_{% \sigma,i},{\bf w}_{\sigma,j}\rangle_{W_{\sigma}}.

Problem 57. (Challenging!) For each (σ,Wσ)Irr(G)(\sigma,W_{\sigma})\in\mathrm{Irr}(G) and fL2(G)={f:G}f\in L^{2}(G)=\{f:G\rightarrow{\mathbb{C}}\}, define f^(σ)Hom(Wσ)\widehat{f}(\sigma)\in\operatorname{Hom}(W_{\sigma}) by

f^(σ)=1|G|gGf(g)σ(g1).\widehat{f}(\sigma)=\frac{1}{|G|}\sum_{g\in G}f(g)\sigma(g^{-1}).
  1. (a)

    Show that ff^(σ)f\mapsto\widehat{f}(\sigma) is an intertwining operator between (ρ,L2(G))(\rho,L^{2}(G)), and (σ~,Hom(Wσ))\big{(}\widetilde{\sigma},\operatorname{Hom}(W_{\sigma})\big{)}, where (ρ(g)f)(h)=f(hg)(\rho(g)f)(h)=f(hg), g,hG\forall g,h\in G fL2(G)f\in L^{2}(G), and σ(g)~(T)=σ(g)T\widetilde{\sigma(g)}(T)=\sigma(g)T for all THom(Wσ)T\in\operatorname{Hom}(W_{\sigma}), gGg\in G.

  2. (b)

    Show the Plancherel identity

    f1,f2G=σIrr(G)dimσtr(f1^(σ)f2^(σ)),\langle f_{1},f_{2}\rangle_{G}=\sum_{\sigma\in\mathrm{Irr}(G)}\operatorname{% dim}\sigma\,\operatorname{tr}\big{(}\widehat{f_{1}}(\sigma)\widehat{f_{2}}(% \sigma)^{*}\big{)},

    and use it to deduce the Fourier inversion formula

    f(g)=σIrr(G)dimσtr(f^(σ)σ(g)).f(g)=\sum_{\sigma\in\mathrm{Irr}(G)}\operatorname{dim}\sigma\,\operatorname{tr% }\big{(}\widehat{f}(\sigma)\sigma(g)\big{)}.