1.5. Lecture 5

1.5.1. Direct sums and reducibility

We start by defining sums of representations:

Definition 1.5.1.

Let (π,V)(\pi,V) and (ρ,W)(\rho,W) be representations of a group GG. The direct sum of the two representations is (π⊕ρ,V⊕W)(\pi\oplus\rho,V\oplus W), where

(π⊕ρ)⁢(g)⁢(𝐯,𝐰):=(π⁢(g)⁢𝐯,ρ⁢(g)⁢𝐰)(\pi\oplus\rho)(g)({\bf v},{\bf w}):=(\pi(g){\bf v},\rho(g){\bf w})

for all g∈Gg\in G, 𝐯∈V{\bf v}\in V, 𝐰∈W{\bf w}\in W.

We also write (π,V)⊕(ρ,W)(\pi,V)\oplus(\rho,W) for the direct sum of the representations.

Example 1.5.2.

Let (χ1,ℂ)(\chi_{1},{\mathbb{C}}) and (χ2,ℂ)(\chi_{2},{\mathbb{C}}) be representations of Cn=⟨a|an=e⟩C_{n}=\langle a\,|\,a^{n}=e\rangle given by χ1⁢(aj)=e2⁢π⁢i⁢j/n\chi_{1}(a^{j})=e^{2\pi ij/n} and χ2⁢(aj)=e4⁢π⁢i⁢j/n\chi_{2}(a^{j})=e^{4\pi ij/n}. Then (χ1⊕χ2,ℂ2)(\chi_{1}\oplus\chi_{2},{\mathbb{C}}^{2}) is the representation of CnC_{n} on ℂ2{\mathbb{C}}^{2} given by χ1⊕χ2⁢(aj)=(e2⁢π⁢i⁢j/n,e4⁢π⁢i⁢j/n)\chi_{1}\oplus\chi_{2}(a^{j})=(e^{2\pi ij/n},e^{4\pi ij/n}).

Conversely, we also have the following definition.

Definition 1.5.3.

If (π,V)(\pi,V) is a representation with non-trivial subrepresentations (π,U)(\pi,U) and (π,W)(\pi,W) such that V=U⊕WV=U\oplus W, then we say that (π,V)(\pi,V) is decomposable or reducible.

Note that if (π,V)(\pi,V) is decomposable , then

(π,V)≅(π,U)⊕(π,W),(\pi,V)\cong(\pi,U)\oplus(\pi,W),

with a GG-isomorphism from U⊕WU\oplus W to VV given by

(𝐮,𝐰)↦𝐮+𝐰({\bf u},{\bf w})\mapsto{\bf u}+{\bf w}

for all 𝐮∈U,𝐰∈W{\bf u}\in U,{\bf w}\in W (see Problem 1.5.4 below). We say that the representation (π,V)(\pi,V) has been decomposed into a direct sum of subrepresentations.

Example 1.5.4.

The permutation representation (π,ℂn)(\pi,{\mathbb{C}}^{n}) decomposes as (π,W1)⊕(π,W0)(\pi,W_{1})\oplus(\pi,W_{0}), where W0={(x1,…,xn)∈ℝn|x1+…+xn=0}W_{0}=\{(x_{1},\ldots,x_{n})\in\mathbb{R}^{n}\,|\,x_{1}+\ldots+x_{n}=0\} and W1W_{1} is the linear span of (1,…,1)(1,\ldots,1).

The following example shows that there are representations that are neither decomposable or irreducible:

Example 1.5.5.

Let (ρ,ℂ2)(\rho,{\mathbb{C}}^{2}) be the representation of (ℤ,+)({\mathbb{Z}},+) given by

ρ⁢(n)=(1n01).\rho(n)=\left(\begin{matrix}1&n\\ 0&1\end{matrix}\right).

This representation is not irreducible, as ℂ⁢(10){\mathbb{C}}\left(\begin{matrix}1\\ 0\end{matrix}\right) is a non-trivial subrepresentation.

The representation is not decomposable , as then there would be two non-zero vectors 𝐯1{\bf v}_{1} and 𝐯2{\bf v}_{2} such that ℂ2=ℂ⁢𝐯1⊕ℂ⁢𝐯2{\mathbb{C}}^{2}={\mathbb{C}}{\bf v}_{1}\oplus{\mathbb{C}}{\bf v}_{2}, and (ρ,ℂ⁢𝐯1)(\rho,{\mathbb{C}}{\bf v}_{1}), (ρ,ℂ⁢𝐯2)(\rho,{\mathbb{C}}{\bf v}_{2}) are subrepresentations. The vectors 𝐯1{\bf v}_{1} and 𝐯2{\bf v}_{2} would then be two linearly independent eigenvectors of all the matrices (1n01)\left(\begin{smallmatrix}1&n\\ 0&1\end{smallmatrix}\right), which is not possible - these matrices are not diagonalisable.

However, for finite groups, we have the following:

Theorem 1.5.6 (Maschke’s Theorem, 1899).

Let (π,W)(\pi,W) be a non-trivial subrepresentation of a finite-dimensional complex representation of a finite group GG. Then there exists a subrepresentation (π,W′)(\pi,W^{\prime}) of (π,V)(\pi,V) such that

(π,V)=(π,W)⊕(π,W′).(\pi,V)=(\pi,W)\oplus(\pi,W^{\prime}).

Before proving this, we introduce another important notion:

1.5.2. Unitary representations

Definition 1.5.7.

Let (π,V)(\pi,V) be a representation of a group GG on an inner product space VV, with inner product ⟨⋅,⋅⟩\langle\cdot,\cdot\rangle. The representation is said to be unitary if

⟨π⁢(g)⁢𝐯,π⁢(g)⁢𝐮⟩=⟨𝐯,𝐮⟩\langle\pi(g){\bf v},\pi(g){\bf u}\rangle=\langle{\bf v},{\bf u}\rangle (1.5.8)

for all 𝐯,𝐮∈V{\bf v},{\bf u}\in V and g∈Gg\in G.

We say that the inner product ⟨⋅,⋅⟩\langle\cdot,\cdot\rangle is GG-invariant if (1.5.8) holds for all 𝐯,𝐮∈V{\bf v},{\bf u}\in V and g∈Gg\in G. Other ways of writing (1.5.8) include

⟨π⁢(g)⁢𝐯,𝐮⟩=⟨𝐯,π⁢(g)∗⁢𝐮⟩\langle\pi(g){\bf v},{\bf u}\rangle=\langle{\bf v},\pi(g)^{*}{\bf u}\rangle

and π⁢(g)∗=π⁢(g−1)=π⁢(g)−1\pi(g)^{*}=\pi(g^{-1})=\pi(g)^{-1}.

Proposition 1.5.9.

Let (π,W)(\pi,W) be a subrepresentation of a unitary representation (π,V)(\pi,V). Then (π,W⟂)(\pi,W^{\perp}) is also a subrepresentation, and hence

(π,V)=(π,W)⊕(π,W⟂).(\pi,V)=(\pi,W)\oplus(\pi,W^{\perp}).

Consequently, all unitary representations with a nontrivial subrepresentation are decomposable.

Proof.

We simply need to verify that W⟂W^{\perp} is GG-invariant. Recall that

W⟂={𝐯∈V|⟨𝐯,𝐰⟩=0⁢ for all ⁢𝐰∈W}.W^{\perp}=\{{\bf v}\in V\,|\,\langle{\bf v},{\bf w}\rangle=0\text{ for all }{% \bf w}\in W\}.

Given g∈Gg\in G, 𝐯∈W⟂{\bf v}\in W^{\perp} and any 𝐰∈W{\bf w}\in W, we have ⟨π⁢(g)⁢𝐯,𝐰⟩=⟨𝐯,π⁢(g−1)⁢𝐰⟩\langle\pi(g){\bf v},{\bf w}\rangle=\langle{\bf v},\pi(g^{-1}){\bf w}\rangle. However, (π,W)(\pi,W) is a subrepresentation, so π⁢(g−1)⁢𝐰∈W\pi(g^{-1}){\bf w}\in W. Since 𝐯∈W⟂{\bf v}\in W^{\perp}, we thus have ⟨𝐯,π⁢(g−1)⁢𝐰⟩=0\langle{\bf v},\pi(g^{-1}){\bf w}\rangle=0, hence π⁢(g)⁢𝐯∈W⟂\pi(g){\bf v}\in W^{\perp}. ∎

The following shows that every finite-dimensional representation of a finite group may be “made” unitary:

Proposition 1.5.10.

Let (π,V)(\pi,V) be a finite-dimensional complex representation of a finite group GG. Then there exists a GG-invariant inner product on VV.

Proof.

Since VV is finite-dimensional, it has an inner product ⟨⋅,⋅⟩\langle\cdot,\cdot\rangle (see Problem 1.5.4 below). We now define ⟨⋅,⋅⟩G:V×V→ℂ\langle\cdot,\cdot\rangle_{G}:V\times V\rightarrow{\mathbb{C}} by the formula

⟨𝐯,𝐮⟩G=1|G|⁢∑h∈G⟨π⁢(h)⁢𝐯,π⁢(h)⁢𝐮⟩,\langle{\bf v},{\bf u}\rangle_{G}=\frac{1}{|G|}\sum_{h\in G}\langle\pi(h){\bf v% },\pi(h){\bf u}\rangle,

and claim that this is a GG-invariant inner product. We leave it as an exercise for the reader to verify that this is an inner product (see Problem 1.5.4). The GG-invariance is shown as follows:

⟨π⁢(g)⁢𝐯,π⁢(g)⁢𝐮⟩G\displaystyle\langle\pi(g){\bf v},\pi(g){\bf u}\rangle_{G} =1|G|⁢∑h∈G⟨π⁢(h)⁢π⁢(g)⁢𝐯,π⁢(h)⁢π⁢(g)⁢𝐮⟩\displaystyle=\frac{1}{|G|}\sum_{h\in G}\langle\pi(h)\pi(g){\bf v},\pi(h)\pi(g% ){\bf u}\rangle
=1|G|⁢∑h∈G⟨π⁢(h⁢g)⁢𝐯,π⁢(h⁢g)⁢𝐮⟩\displaystyle=\frac{1}{|G|}\sum_{h\in G}\langle\pi(hg){\bf v},\pi(hg){\bf u}\rangle
=1|G|⁢∑h~∈G⟨π⁢(h~)⁢𝐯,π⁢(h~)⁢𝐮⟩=⟨𝐯,𝐮⟩G.\displaystyle=\frac{1}{|G|}\sum_{\widetilde{h}\in G}\langle\pi(\widetilde{h}){% \bf v},\pi(\widetilde{h}){\bf u}\rangle=\langle{\bf v},{\bf u}\rangle_{G}.

∎

We note that as a consequence of the spectral theorem for unitary matrices, we have

Proposition 1.5.11.

Let (π,V)(\pi,V) be a finite-dimensional unitary representation of a group GG. Then for any g∈Gg\in G, π⁢(g)\pi(g) is diagonalisable, i.e. has an eigenbasis.

Proof.

See Exercise 1.5.4 below ∎

1.5.3. Decompositions into irreducibles

Note that Theorem 1.5.6 now follows directly from Propositions 1.5.9 and 1.5.10. In fact, we can give a stronger statement:

Theorem 1.5.12.

Let (π,V)(\pi,V) be a representation of GG. Then there exist irreducible subrepresentations (π,Wi)(\pi,W_{i}) such that

(π,V)=(π,W1)⊕(π,W2)⊕…⊕(π,Wn).(\pi,V)=(\pi,W_{1})\oplus(\pi,W_{2})\oplus\ldots\oplus(\pi,W_{n}).

Moreover, the number of times each isomorphism class of an irreducible representation shows up in the above decomposition is independent of the exact choice of decomposition.

The existence of such a decomposition follows as a corollary from repeated application of Maschke’s Theorem (Theorem 1.5.6). To complete the proof of the second part of the theorem we need the following two lemmas:

Lemma 1.5.13.

Let (π1,U)(\pi_{1},U), (π2,V)(\pi_{2},V), and (π3,W)(\pi_{3},W) be representations of a group GG. Then

  1. (i)
    ​

    HomG⁢(U⊕V,W)≅HomG⁢(U,W)⊕HomG⁢(V,W)\mathrm{Hom}_{G}(U\oplus V,W)\cong\mathrm{Hom}_{G}(U,W)\oplus\mathrm{Hom}_{G}(% V,W)

  2. (ii)
    ​

    HomG⁢(W,U⊕V)≅HomG⁢(W,U)⊕HomG⁢(W,V)\mathrm{Hom}_{G}(W,U\oplus V)\cong\mathrm{Hom}_{G}(W,U)\oplus\mathrm{Hom}_{G}(% W,V)

Proof.

Exercise: Problem 1.5.4 below. ∎

Lemma 1.5.14.

Let (π,V)=⨁i=1n(π,Wi)(\pi,V)=\bigoplus_{i=1}^{n}(\pi,W_{i}) be a decomposition of (π,V)(\pi,V) into irreducible subrepresentations. For any irreducible representation (ρ,U)(\rho,U), the number of (π,Wi)(\pi,W_{i}) that are isomorphic to (ρ,U)(\rho,U) is dim⁡HomG⁡(V,U)\operatorname{dim}\operatorname{Hom}_{G}(V,U).

Proof.

By Lemma 1.5.13, dim⁡HomG⁡(V,U)=∑idim⁡HomG⁡(Wi,U)\operatorname{dim}\operatorname{Hom}_{G}(V,U)=\sum_{i}\operatorname{dim}% \operatorname{Hom}_{G}(W_{i},U). By Corollary 1.4.2 to Schur’s lemma, dim⁡HomG⁡(Wi,U)=1\operatorname{dim}\operatorname{Hom}_{G}(W_{i},U)=1 if (π,Wi)≅(ρ,U)(\pi,W_{i})\cong(\rho,U), and zero otherwise. ∎

Proof of Theorem 1.5.12.

The existence of such a decomposition follows from repeated application of Theorem 1.5.6. The second part of the statement follows from Lemma 1.5.14. ∎

1.5.4. Exercises

.

Problem 31. Let (π,U)(\pi,U) and (π,W)(\pi,W) be subrepresentations of a representation (π,V)(\pi,V) such that V=U⊕WV=U\oplus W. Verify that the map

(𝐮,𝐰)↦𝐮+𝐰({\bf u},{\bf w})\mapsto{\bf u}+{\bf w}

is a GG-isomorphism from (π,U)⊕(π,W)(\pi,U)\oplus(\pi,W) to (π,V)(\pi,V).

Problem 32. Verify that

⟨π⁢(g)⁢𝐯,π⁢(g)⁢𝐮⟩=⟨𝐯,𝐮⟩∀𝐯,𝐮∈V,g∈G\langle\pi(g){\bf v},\pi(g){\bf u}\rangle=\langle{\bf v},{\bf u}\rangle\qquad% \forall{\bf v},{\bf u}\in V,\;g\in G

is equivalent to

⟨π⁢(g)⁢𝐯,𝐮⟩=⟨𝐯,π⁢(g−1)⁢𝐮⟩∀𝐯,𝐮∈V,g∈G.\langle\pi(g){\bf v},{\bf u}\rangle=\langle{\bf v},\pi(g^{-1}){\bf u}\rangle% \qquad\forall{\bf v},{\bf u}\in V,\;g\in G.

Problem 33. Let VV be a finite-dimensional complex vector space with a basis {𝐛i}i=1,…,n\{{\bf b}_{i}\}_{i=1,\ldots,n}, where n=dim⁡Vn=\operatorname{dim}V. Show that

⟨∑i=1nxi⁢𝐛i,∑j=1nyj⁢𝐛j⟩:=∑i=1nxi⁢yi¯,\left\langle\sum_{i=1}^{n}x_{i}{\bf b}_{i},\sum_{j=1}^{n}y_{j}{\bf b}_{j}% \right\rangle:=\sum_{i=1}^{n}x_{i}\overline{y_{i}},

for xi,yj∈ℂx_{i},\,y_{j}\in{\mathbb{C}}, defines an inner product on VV.

Problem 34. Fill in the remaining details of the proof of Proposition 1.5.10, i.e. show that ⟨⋅,⋅⟩G\langle\cdot,\cdot\rangle_{G} is linear in the first argument, conjugate-symmetric, and positive-definite.

Problem 35. Let (π,V)(\pi,V) be a finite-dimensional unitary representation of a group GG, with GG-invariant inner product ⟨⋅,⋅⟩\langle\cdot,\cdot\rangle. Show that the matrix of any π⁢(g)\pi(g) with respect to an orthonormal basis for ⟨⋅,⋅⟩\langle\cdot,\cdot\rangle is unitary (recall that a matrix is unitary if M∗(=M¯)t=M−1M^{*}(=\overline{M}{\,{}^{\mathrm{t}}\!}\,)=M^{-1}).

Problem 36. Prove Lemma 1.5.13.