Lecture 2

Problem 20. Let (π,V)(\pi,V) be a representation of a group GG. Show that if (π,W1)(\pi,W_{1}), (π,W2)(\pi,W_{2}) are two irreducible subrepresentations of (π,V)(\pi,V), then either W1W2=0W_{1}\cap W_{2}=0 or W1=W2W_{1}=W_{2}.

Solution: Let wW1W2W1w\in W_{1}\cap W_{2}\subset W_{1}. Since W1W_{1} is a subrepresentation of (π,V)(\pi,V), π(g)wW1\pi(g)w\in W_{1} for all gGg\in G. Similarly, π(g)wW2\pi(g)w\in W_{2} for all gGg\in G, and hence π(g)wW1W2\pi(g)w\in W_{1}\cap W_{2}. (π,W1W2)(\pi,W_{1}\cap W_{2}) is therefore a subrepresentation of (π,W1)(\pi,W_{1}) and also (π,W2)(\pi,W_{2}). Since (π,W1)(\pi,W_{1}) is irreducible, we then have W1W2=0W_{1}\cap W_{2}=0 or W1W_{1}. If W1W2=0W_{1}\cap W_{2}=0, we are done. If not, then W1W2=W1W_{1}\cap W_{2}=W_{1}, hence W1W2W_{1}\subset W_{2}. However, (π,W2)(\pi,W_{2}) is also irreducible, so the same applies: W1W2=0W_{1}\cap W_{2}=0 or W2W_{2}, with the latter case giving W2W1W_{2}\subset W_{1}. Since we assumed W1W20W_{1}\cap W_{2}\neq 0, we then have W1W2W_{1}\subset W_{2} and W2W1W_{2}\subset W_{1}, i.e. they are equal.

Problem 21. Let (π,V)(\pi,V) and (ρ,W)(\rho,W) be two finite-dimensional representations of a group GG. Show that if THomG(V,W)T\in\operatorname{Hom}_{G}(V,W) is invertible, then T1HomG(W,V)T^{-1}\in\operatorname{Hom}_{G}(W,V).

Solution: Since TT is an invertible linear map, T1Hom(W,V)T^{-1}\in\operatorname{Hom}(W,V). It remains to check that in fact T1HomG(W,V)T^{-1}\in\operatorname{Hom}_{G}(W,V). For any wWw\in W there exists vVv\in V such that w=Tvw=Tv. This gives

π(g)T1w=π(g)T1Tv=π(g)v=T1Tπ(g)v=T1ρ(g)Tv=T1ρ(g)w,\pi(g)T^{-1}w=\pi(g)T^{-1}Tv=\pi(g)v=T^{-1}T\pi(g)v=T^{-1}\rho(g)Tv=T^{-1}\rho% (g)w,

as required.

Problem 22. Let (π,V)(\pi,V) be an nn-dimensional representation of a group GG. Show that there exists a representation (π~,n)(\widetilde{\pi},{\mathbb{C}}^{n}) that is isomorphic to (π,V)(\pi,V).

Solution: Let ={v1,,vn}{\mathcal{B}}=\{v_{1},\ldots,v_{n}\} be a basis of VV, and given gGg\in G, define π~(g)n×n\widetilde{\pi}(g)\in{\mathbb{C}}^{n\times n} by

π~(g)=([π(g)v1][π(g)vn]),\widetilde{\pi}(g)=\left(\begin{matrix}\vdots&&\vdots\\ [\pi(g)v_{1}]_{{\mathcal{B}}}&\cdots&[\pi(g)v_{n}]_{{\mathcal{B}}}\\ \vdots&&\vdots\end{matrix}\right),

i.e. the ii-th column is the vector [π(g)vi]n[\pi(g)v_{i}]_{{\mathcal{B}}}\in{\mathbb{C}}^{n}; this is the coordinate vector of π(g)vi\pi(g)v_{i} w.r.t. the basis {\mathcal{B}}. Since π~(g)\widetilde{\pi}(g) is a matrix, the map zπ~(g)zz\mapsto\widetilde{\pi}(g)z is linear on n{\mathbb{C}}^{n}. We wish to show that (π~,n)(\widetilde{\pi},{\mathbb{C}}^{n}) is a representation that is isomorphic to (π,V)(\pi,V). To do this, we consider the linear map T:VnT:V\rightarrow{\mathbb{C}}^{n} defined on the basis elements by Tvi=ei,Tv_{i}=e_{i}, where {ei}\{e_{i}\} is the standard basis of n{\mathbb{C}}^{n}. By construction, TT is an invertible linear map. We verify that it is in fact an intertwining operator: let v=iaiviv=\sum_{i}a_{i}v_{i}. Then for any gGg\in G,

π~(g)Tv=π~(g)T(iaivi)=π~(g)iaiei=iaiπ~(g)ei=iai[π(g)vi].\widetilde{\pi}(g)Tv=\widetilde{\pi}(g)T\left(\sum_{i}a_{i}v_{i}\right)=% \widetilde{\pi}(g)\sum_{i}a_{i}e_{i}=\sum_{i}a_{i}\widetilde{\pi}(g)e_{i}=\sum% _{i}a_{i}[\pi(g)v_{i}]_{{\mathcal{B}}}.

On the other hand,

Tπ(g)v=\displaystyle T\pi(g)v= T(iaiπ(g)vi)=T(iaij([π(g)vi])jvj)=iaij([π(g)vi])jTvj\displaystyle T\left(\sum_{i}a_{i}\pi(g)v_{i}\right)=T\left(\sum_{i}a_{i}\sum_% {j}\big{(}[\pi(g)v_{i}]_{{\mathcal{B}}}\big{)}_{j}v_{j}\right)=\sum_{i}a_{i}% \sum_{j}\big{(}[\pi(g)v_{i}]_{{\mathcal{B}}}\big{)}_{j}Tv_{j}
=iai(j([π(g)vi])jej)=iai[π(g)vi].\displaystyle=\sum_{i}a_{i}\left(\sum_{j}\big{(}[\pi(g)v_{i}]_{{\mathcal{B}}}% \big{)}_{j}e_{j}\right)=\sum_{i}a_{i}[\pi(g)v_{i}]_{{\mathcal{B}}}.

This shows that π~(g)T=Tπ(g)\widetilde{\pi}(g)T=T\pi(g) for all gGg\in G. We now use this to prove that π~\widetilde{\pi} is a homomorphism; this will conclude the proof, as the above shows that TT is an invertible intertwining operator. Given znz\in{\mathbb{C}}^{n}, we let z=Tvz=Tv. Then for any g,hGg,h\in G,

π~(gh)z=π~(gh)Tv=Tπ(gh)v=Tπ(g)π(h)v=π~(g)Tπ(h)v=π~(g)π~(h)Tv=π~(g)π~(h)z,\widetilde{\pi}(gh)z=\widetilde{\pi}(gh)Tv=T\pi(gh)v=T\pi(g)\pi(h)v=\widetilde% {\pi}(g)T\pi(h)v=\widetilde{\pi}(g)\widetilde{\pi}(h)Tv=\widetilde{\pi}(g)% \widetilde{\pi}(h)z,

showing that (π~,n)(\widetilde{\pi},{\mathbb{C}}^{n}) is a representation, with THomG(V,n)T\in\operatorname{Hom}_{G}(V,{\mathbb{C}}^{n}) being invertible.

Remark: This result actually follows from a stronger result in linear algebra: for any two operators S,TS,T on VV, we have [ST]=[S][T][ST]_{{\mathcal{B}}}=[S]_{{\mathcal{B}}}[T]_{{\mathcal{B}}}. This shows that GL(V)\operatorname{GL}(V) is isomorphic to GLn()\operatorname{GL}_{n}({\mathbb{C}}) as groups, and that the map v[v]v\mapsto[v]_{{\mathcal{B}}} is an intertwining operator between the natural representations of GL(V)\operatorname{GL}(V) on VV and GLn()\operatorname{GL}_{n}({\mathbb{C}}) on n{\mathbb{C}}^{n}.

Problem 23. Generalise Example 2.4 to the permutation representation (π,n)(\pi,{\mathbb{C}}^{n}) of SnS_{n}.

Solution: Let 𝐯1=𝐞1++𝐞n{\bf v}_{1}={\bf e}_{1}+\ldots+{\bf e}_{n}. Then π(σ)𝐯1=𝐯1\pi(\sigma){\bf v}_{1}={\bf v}_{1} for all σSn\sigma\in S_{n}, so (π,𝐯1)(\pi,{\mathbb{C}}{\bf v}_{1}) is a subrepresentation of (π,n)(\pi,{\mathbb{C}}^{n}). Being 1-dimensional, it is irreducible.

Denote by W0W_{0} the subspace of 3{\mathbb{C}}^{3} consisting of all vectors (z1,,zn)t(z_{1},\ldots,z_{n})\,^{\mathrm{t}}\! such that z1++zn=0z_{1}+\ldots+z_{n}=0. Then since SnS_{n} permutes the coordinates of a vector, it doesn’t change their sum; so π(σ)W0W0\pi(\sigma)W_{0}\subseteq W_{0}, and W0W_{0} is a subrepresentation (of dimension n1n-1).

We claim that W0W_{0} is irreducible. Indeed, suppose that UW0U\subseteq W_{0} is a nonzero subrepresentation; we have to show that U=W0U=W_{0}. Let (x1,,xn)tU(x_{1},\ldots,x_{n})\,^{\mathrm{t}}\!\in U be nonzero. As x1==xnx_{1}=\ldots=x_{n} can’t happen, we can apply an element of SnS_{n} to permute the coordinates so that x1x2x_{1}\neq x_{2}. Then applying π((12))\pi((12)), we have (x2,x1,x3,,xn)tU(x_{2},x_{1},x_{3},\ldots,x_{n})\,^{\mathrm{t}}\!\in U. Taking the difference, (x1x2,x2x1,0,,0)tU(x_{1}-x_{2},x_{2}-x_{1},0,\ldots,0)\,^{\mathrm{t}}\!\in U; scaling, (1,1,0,0)tU(1,-1,0,\ldots 0)\,^{\mathrm{t}}\!\in U. Applying π((2i))\pi((2i)), for 3in3\leq i\leq n, we get a family of vectors with 11 as their first entry, 1-1 as the iith entry and 0s elsewhere, each in UU. But these vectors span W0W_{0} (e.g. because they are linearly independent and dimW0=n1\operatorname{dim}W_{0}=n-1), so U=W0U=W_{0} as required.

Problem 24. Let

C0(X)={f:Xk|f(x)=0forallbutfinitelymanyxX}.C_{0}(X)=\{f:X\rightarrow k\,|\,f(x)=0\mathrm{\;for\;all\;but\;finitely\;many}% \;x\in X\}.

The quasi-regular representation of GG on C0(X)C_{0}(X) is denoted (λ,C0(X))(\lambda,C_{0}(X)) and is defined via the formula

(λ(g)f)(x):=f(g1x)\big{(}\lambda(g)f\big{)}(x):=f(g^{-1}\cdot x)

for all gGg\in G, xXx\in X and fC0(X)f\in C_{0}(X). This is called the functional point of view. Let GG be a group acting on a set XX. Show that (π,(X))(\pi,{\mathbb{C}}(X)), the representation arising from this group action, is isomorphic to (λ,C0(X))(\lambda,C_{0}(X)).

Solution: Let ϕ\phi be the map

ϕ:(X)C0(X);(xXzxx)(f{zx}xX:Xk;wzw).\phi:{\mathbb{C}}(X)\longrightarrow C_{0}(X)\;;\;\left(\sum_{x\in X}z_{x}x% \right)\longmapsto\left(f_{\{z_{x}\}_{x\in X}}:X\rightarrow k\,;\,w\mapsto z_{% w}\right).

This map is well defined as all elements of (X){\mathbb{C}}(X) have only finitely many non-zero terms. This map is linear as

ϕ(xXzxx+μxXyxx)(w)\displaystyle\phi\left(\sum_{x\in X}z_{x}x+\mu\sum_{x\in X}y_{x}x\right)(w) =ϕ(xX(zx+μyx)x)(w)\displaystyle=\phi\left(\sum_{x\in X}(z_{x}+\mu y_{x})x\right)(w)
=f{zx+μyx}xX(w)\displaystyle=f_{\{z_{x}+\mu y_{x}\}_{x\in X}}(w)
=zw+μyw\displaystyle=z_{w}+\mu y_{w}
=f{zx}xX(w)+μf{yx}xX(w)\displaystyle=f_{\{z_{x}\}_{x\in X}}(w)+\mu f_{\{y_{x}\}_{x\in X}}(w)
=ϕ(xXzxx)(w)+μϕ(xXyxx)(w),\displaystyle=\phi\left(\sum_{x\in X}z_{x}x\right)(w)+\mu\phi\left(\sum_{x\in X% }y_{x}x\right)(w),

for all xXzxx,xXyxx(X)\sum_{x\in X}z_{x}x,\sum_{x\in X}y_{x}x\in{\mathbb{C}}(X), wXw\in X and μk\mu\in k. It is a GG-homomorphism as

ϕ(π(g)xXzxx)(w)=ϕ(xXzg1xx)(w)=zg1w,\phi\left(\pi(g)\sum_{x\in X}z_{x}x\right)(w)=\phi\left(\sum_{x\in X}z_{g^{-1}% \cdot x}x\right)(w)=z_{g^{-1}\cdot w},

and

(λ(g)ϕ(xXzxx))(w)=(λ(g)f{zx}xX)(w)=f{zx}xX(g1w)=zg1w,\left(\lambda(g)\phi\left(\sum_{x\in X}z_{x}x\right)\right)(w)=(\lambda(g)f_{% \{z_{x}\}_{x\in X}})(w)=f_{\{z_{x}\}_{x\in X}}(g^{-1}\cdot w)=z_{g^{-1}\cdot w},

for all wXw\in X, xXzxx(X)\sum_{x\in X}z_{x}x\in{\mathbb{C}}(X) and gGg\in G.

Suppose xXzxxkerϕ\sum_{x\in X}z_{x}x\in\ker\phi, then ϕ(xXzxx)(w)=0\phi\left(\sum_{x\in X}z_{x}x\right)(w)=0 for all wXw\in X and thus all coefficients zx=0z_{x}=0. Therefore xXzx=0\sum_{x\in X}z_{x}=0 and kerϕ\ker\phi is trivial. Now let gC0(X)g\in C_{0}(X) and consider xXg(x)x(X)\sum_{x\in X}g(x)x\in{\mathbb{C}}(X). Then, for all wXw\in X,

ϕ(xXg(x)x)(w)=f{g(x)}xX(w)=g(w)\phi\left(\sum_{x\in X}g(x)x\right)(w)=f_{\{g(x)\}_{x\in X}}(w)=g(w)

Thus ϕ(xXg(x)x)=g\phi\left(\sum_{x\in X}g(x)x\right)=g and as gg was arbitrary ϕ\phi is a bijection. Thus ϕ\phi is a GG-isomorphism and the two representations are isomorphic.