Lecture 3

Problem 25.

  1. (a)

    Show that if (π,V)(\pi,V) and (ρ,W)(\rho,W) are two isomorphic finite-dimensional representations of a group GG, then π(g)\pi(g) and ρ(g)\rho(g) have the same eigenvalues for all gGg\in G.

  2. (b)

    Complete the proof of Theorem 3.2.

Solution:

  1. (a)

    Let THomG(V,W)T\in\operatorname{Hom}_{G}(V,W) be invertible. If vVv\in V is an eigenvector for π(g),\pi(g), i.e. π(g)v=λv\pi(g)v=\lambda v, then

    ρ(g)Tv=Tπ(g)v=λTv,\rho(g)Tv=T\pi(g)v=\lambda Tv,

    i.e. λ\lambda is an eigenvalue for ρ(g)\rho(g). This shows that any eigenvalue of π(g)\pi(g) is also an eigenvalue of ρ(g)\rho(g). The same argument with T1T^{-1} in place of TT that any eigenvalues of ρ(g)\rho(g) is also an eigenvalue of π(g).\pi(g).

  2. (b)

    First we note that, by part (a), (Id,)({\,\mathrm{Id}},{\mathbb{C}}) and (ϵ,)(\epsilon,{\mathbb{C}}) are not isomorphic because the eigenvalues corresponding to ss, 11 and 1-1 respectively, are not the same. It is also clear that both 1-dimensional representations are not isomorphic to the 2-dimensional representations. Note that, for each 1k<n/21\leq k<n/2, the eigenvalues of ρk(r)\rho_{k}(r) are 2πik/n2\pi ik/{n} and 2πik/n-2\pi ik/n, which are distinct for distinct values of kk. Thus each ρk\rho_{k} is in its own isomorphism class and everything is the table is non-isomorphic.

Problem 26. Find all the irreducible representations of DnD_{n} for nn even.

Solution: We have an additional two 1-dimensional representations so the table is:

Table 1. Representations of DnD_{n}.
Dimension ρ(r)\rho(r) ρ(s)\rho(s)
(Id,)(\mathrm{Id},{\mathbb{C}}) 11 11 11
(ϵ,)(\epsilon,{\mathbb{C}}) 11 11 1-1
(π1,)(\pi_{1},{\mathbb{C}}) 11 1-1 11
(π2,)(\pi_{2},{\mathbb{C}}) 11 1-1 1-1
ρk1k<n/2\underset{1\leq k<n/2}{\rho_{k}} 22 (e2πikn00e2πikn)\left(\begin{smallmatrix}e^{\frac{2\pi ik}{n}}&0\\ 0&e^{\frac{-2\pi ik}{n}}\end{smallmatrix}\right) (0110)\left(\begin{smallmatrix}0&1\\ 1&0\end{smallmatrix}\right)

The proof that this is a complete list follows the same argument as the proof of Theorem 3.2, but differs slightly in two places. In Case 1 we note that we exclude k=n/2k=n/2 because then we would have λ=λ1\lambda=\lambda^{-1}. In Case 2 we have a further two cases: the case when λ=1\lambda=1, which is the same as in Theorem 3.2, and the case when λ=1\lambda=-1. Now

π(r)(𝐯+𝐰)=𝐯𝐰\pi(r)({\bf v}+{\bf w})=-{\bf v}-{\bf w}

but this is still in (𝐯+𝐰){\mathbb{C}}({\bf v}+{\bf w}) so (𝐯+𝐰)({\bf v}+{\bf w}) spans a subrepresentation of VV. If 𝐯+𝐰0{\bf v}+{\bf w}\neq 0 then, as π(s)(𝐯+𝐰)=𝐯+𝐰\pi(s)({\bf v}+{\bf w})={\bf v}+{\bf w} we have the representation π1\pi_{1} with V=(𝐯+𝐰)V={\mathbb{C}}({\bf v}+{\bf w}). Otherwise, π(s)𝐯=𝐰=𝐯\pi(s){\bf v}={\bf w}=-{\bf v}, and we get the representation π2\pi_{2}.

The argument that these are all non isomorphic is the same as in Problem 25(b).

Problem 27. Show that the irreducible representations of S3S_{3} consist of (a) The one-dimensional trivial representation (b) the sign representation (sgn,),(\mathrm{sgn},{\mathbb{C}}), where sgn(σ){±1}\mathrm{sgn}(\sigma)\in\{\pm 1\} is the sign of the permutation σ\sigma, and (c) the representation (π,W0)(\pi,W_{0}), where π\pi is the usual permutation representation of S3S_{3} on 3{\mathbb{C}}^{3}, and W0={(z1,z2,z3)3|z1+z2+z3=0}W_{0}=\{(z_{1},z_{2},z_{3})\in{\mathbb{C}}^{3}\,|\,z_{1}+z_{2}+z_{3}=0\}. Hint: use the fact that we have a complete classification of the irreducible representations of DnD_{n}.

Solution: Since S3D3S_{3}\equiv D_{3} (see Problem 6), we know that S3S_{3} has two one-dimensional irreducible representations and one two-dimensional irreducible representation. From class, we saw that (π,W0)(\pi,W_{0}) is irreducible, and we know that (Id,)({\,\mathrm{Id}},{\mathbb{C}}) is the one-dimensional trivial representation, so it remains to show that the sign map defines a representation, and also that it is not isomorphic to the trivial representations. Firstly, recall that the sign of a permutation σ\sigma equals 1-1 raised to the number of transpositions (ij)(i\,j) needed to express σ\sigma. In particular, sgn(1 2)=1\mathrm{sgn}(1\,2)=-1, which is order two, hence (sgn,)(\mathrm{sgn},{\mathbb{C}}) cannot be isomorphic to Id,){\,\mathrm{Id}},{\mathbb{C}}). It remains to verify that the sign map is a homomorphism; this is a standard fact from group theory: if τ\tau is written as N(τ)N(\tau) transpositions and σ\sigma is written as N(σ)N(\sigma) transpositions, then στ\sigma\tau can be written as N(σ)+N(τ)N(\sigma)+N(\tau) transpositions, hence sgn(τ)sgn(σ)=(1)N(τ)(1)N(σ)=(1)N(τ)+N(σ)=sgn(τσ)\mathrm{sgn}(\tau)\mathrm{sgn}(\sigma)=(-1)^{N(\tau)}(-1)^{N(\sigma)}=(-1)^{N(% \tau)+N(\sigma)}=\operatorname{sgn}(\tau\sigma).

Problem 28. Classify the irreducible representations of Q8Q_{8}. Hint: adapt the strategy we used for DnD_{n}.

Solution: Let (π,V)(\pi,V) be an irreducible representation of Q8={±1,±i,±j,±k}Q_{8}=\{\pm 1,\pm i,\pm j,\pm k\}. We choose an eigenvector vv for π(i)\pi(i):

π(i)v=λv,\pi(i)v=\lambda v,

where λ4=1\lambda^{4}=1. Define also w=π(j)vw=\pi(j)v. We claim defining U=span{v,w}U=\mathrm{span}\{v,w\} gives a subrepresentation (π,U)(\pi,U) of (π,V)(\pi,V). Note that {i,j}\{i,j\} is a generating set for Q8Q_{8}, so to check that UU is a subrepresentation, it suffices to check that π(i)wU\pi(i)w\in U and π(j)wU\pi(j)w\in U (since by definition π(i)v,π(j)vU\pi(i)v,\pi(j)v\in U). We compute:

π(i)w=π(ij)v=π(j(i))v=π(j)π(i)1v=λ1w,π(j)w=π(j2)v=π(1)v=π(i2)v=λ2v.\pi(i)w=\pi(ij)v=\pi(j(-i))v=\pi(j)\pi(i)^{-1}v=\lambda^{-1}w,\quad\pi(j)w=\pi% (j^{2})v=\pi(-1)v=\pi(i^{2})v=\lambda^{2}v.

Thus (π,U)(\pi,U) is a subrepresentation of (π,V)(\pi,V), and hence (VV is assumed to be irreducible) U=VU=V. We now consider two different cases:

  • Case 1:

    wvw\in{\mathbb{C}}v, (π,V)(\pi,V) is then one-dimensional. Since v=π(1)v=π(1)2vv=\pi(1)v=\pi(-1)^{2}v, we have two possibilities: either π(1)v=v\pi(-1)v=v or π(1)v=v\pi(-1)v=-v.

    If π(1)=1\pi(-1)=-1, then writing w=π(j)v=μvw=\pi(j)v=\mu v, from the group relations, we must have

    1=π(1)=π(i)2=λ2=π(j)2=μ2=π(k)2=π(ij)2=λ2μ2,-1=\pi(-1)=\pi(i)^{2}=\lambda^{2}=\pi(j)^{2}=\mu^{2}=\pi(k)^{2}=\pi(ij)^{2}=% \lambda^{2}\mu^{2},

    i.e. 1=λ2=μ2=λ2μ2-1=\lambda^{2}=\mu^{2}=\lambda^{2}\mu^{2}, which is not possible! Hence, we must have π(1)=1\pi(-1)=1. This gives λ,μ{±1}\lambda,\mu\in\{\pm 1\}, and one sees that each of these four cases satisfies the above equation, giving four isomorphism classes of one-dimensional representations:

    Rep π(±1)\pi(\pm 1) π(±i)\pi(\pm i) π(±j)\pi(\pm j) π(±k)\pi(\pm k)
    (Id,)({\,\mathrm{Id}},{\mathbb{C}}) 1 1 1 1
    (πi,)(\pi_{i},{\mathbb{C}}) 1 1 -1 -1
    (πj,)(\pi_{j},{\mathbb{C}}) 1 -1 1 -1
    (πk,)(\pi_{k},{\mathbb{C}}) 1 -1 -1 1
  • Case 2:

    wvw\not\in{\mathbb{C}}v. Writing down the matrices of π(i)\pi(i) and π(j)\pi(j) with respect to the basis v,wv,w gives

    π(i)=(λ00λ1),π(j)=(01λ20).\pi(i)=\left(\begin{matrix}\lambda&0\\ 0&\lambda^{-1}\end{matrix}\right),\quad\pi(j)=\left(\begin{matrix}0&1\\ \lambda^{2}&0\end{matrix}\right).

    Since π(i)4=Id\pi(i)^{4}={\,\mathrm{Id}}, we see that λ\lambda must be a fourth root of unity. If λ=±1,\lambda=\pm 1, then π(i)=±Id\pi(i)=\pm{\,\mathrm{Id}}, which commutes with π(j)\pi(j), allowing us to simultaneously diagonalise the matrices, i.e decomposing (π,V)(\pi,V) as (π,(u+v))(π,(uv))(\pi,{\mathbb{C}}(u+v))\oplus(\pi,{\mathbb{C}}(u-v)). This contradicts the assumption that VV was irreducible! Hence λ=±i\lambda=\pm i; both choices give representations that are isomorphic to each other (with the isomorphism given by swapping the basis elements). Moreover, they are irreducible, as the two distinct eigenvalues of π(i)\pi(i) do not agree with any of those for the one-dimensional representations. Thus, Q8Q_{8} has a single two-dimensional irreducible representation with operators as follows:

    Rep π(±1)\pi(\pm 1) π(±i)\pi(\pm i) π(±j)\pi(\pm j) π(±k)\pi(\pm k)
    (π,2)(\pi,{\mathbb{C}}^{2}) ±Id\pm{\,\mathrm{Id}} ±(i00i)\pm\left(\begin{matrix}i&0\\ 0&-i\end{matrix}\right) ±(0110)\pm\left(\begin{matrix}0&1\\ -1&0\end{matrix}\right) ±(0ii0)\pm\left(\begin{matrix}0&-i\\ i&0\end{matrix}\right)

    (here the “±\pm” signs match, i.e. π(1)=Id\pi(-1)=-{\,\mathrm{Id}}, etc.).