Book a Demo!
CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutPoliciesSign UpSign In
gmcninch-tufts
GitHub Repository: gmcninch-tufts/2024-Sp-Math190
Path: blob/main/course-assignments/PS03--rep-theory--solutions.md
908 views
---
title: | ProblemSet 3 -- Solutions author: George McNinch date: 2024-02-16
---

\newcommand{\trivial}{\mathbf{1}}

In these exercises, GG always denotes a finite group. Unless indicated otherwise, all vector spaces are assumed to be finite dimensional over the field F=CF = \mathbb{C}. The representation space VV of a representation of GG is always assumed to be finite dimensional over C\mathbb{C}.

  1. Let ϕ:GF×\phi:G \to F^\times be a group homomorphism; since F×=GL1(F)F^\times = \operatorname{GL}_1(F), we can think of ϕ\phi as a 1-dimensional representation (ϕ,F)(\phi,F) of GG.

    If VV is any representation of GG, we can form a new representation ϕV\phi \otimes V. The underlying vector space for this representation is just VV, and the "new" action of an element gGg \in G on a vector vv is given by the rule gv=ϕ(g)gv.g \star v = \phi(g) gv.

    a. Prove that if VV is irreducible, then ϕV\phi \otimes V is also irreducible.

    ::: {.solution}

    We prove the following statement: ()(*) if WVW \subset V is a subspace, then WW is invariant for the original action of GG if and only if it is invariant for the \star action of GG.

    First note that ()(*) immediately implies the assertion of (a).

    To test invariance, let wWw \in W and let gGg \in G. Since WW is a linear subspace and since ϕ(g)\phi(g) is a non-zero scalar, it is immediate that gwW    gw=ϕ(g)gwWgw \in W \iff g \star w = \phi(g) gw \in W Since this holds for all ww and all gg, ()(*) follows.

    :::

    b. Prove that if χ\chi denotes the character of VV, then the character of ϕV\phi \otimes V is given by ϕχ\phi \cdot \chi; in other words, the trace of the action of gGg \in G on ϕV\phi \otimes V is given by χϕV(g)=tr(vgv)=ϕ(g)χ(g).\chi_{\phi \otimes V}(g) = \operatorname{tr}( v\mapsto g \star v) = \phi(g) \chi(g).

    ::: {.solution} We just need to compute the trace of the linear mapping VVV \to V given by vgvv \mapsto g \star v.

    If the action of gg on VV is given by the linear mapping ρ(g)\rho(g), then χV(g)=tr(ρ(g)).\chi_V(g) = \operatorname{tr}(\rho(g)).

    Now, the \star-action of gg is given by the linear mapping vgv=ϕ(g)ρ(g)vv \mapsto g \star v = \phi(g) \rho(g)v.

    So χϕV(g)=tr(ϕ(g)ρ(g))\chi_{\phi \otimes V}(g) = \operatorname{tr}(\phi(g)\rho(g)). For any scalar sks \in k, trace of the linear mapping sρ(g)s \rho(g) is given by tr(sρ(g))=str(ρ(g))=sχV(g)\operatorname{tr}(s \rho(g)) = s \operatorname{tr}(\rho(g)) = s \chi_V(g) ("linearity of the trace").

    Thus χϕV(g)=tr(ϕ(g)ρ(g))=ϕ(g)χV(g).\chi_{\phi\otimes V}(g) = \operatorname{tr}(\phi(g)\rho(g)) = \phi(g) \chi_V(g).

    :::

    c. Recall that in class we saw that S3S_3 has an irreducible representation V2V_2 of dimension 2 whose character ψ2\psi_2 is given by

    g1(12)(123)ψ2201\begin{array}{l|lll} g & 1 & (12) & (123) \\ \hline \psi_2 & 2 & 0 & -1 \end{array}

    Observe that sgnψ=ψ\operatorname{sgn} \psi = \psi and conclude that V2sgnV2V_2 \simeq \operatorname{sgn} \otimes V_2, where sgn:Sn{±1}C×\operatorname{sgn}:S_n \to \{\pm 1\} \subset \mathbb{C}^\times is the sign homomorphism.

    On the other hand, S4S_4 has an irreducible representation V3V_3 of dimension 3 whose character ψ3\psi_3 is given by

    g1(12)(123)(1234)(12)(34)ψ331011\begin{array}{l|lllll} g & 1 & (12) & (123) & (1234) & (12)(34) \\ \hline \psi_3 & 3 & 1 & 0 & -1 & -1 \end{array}

    (I'm not asking you to confirm that ψ3\psi_3 is irreducible, though it would be straightforward to check that ψ3,ψ3=1\langle \psi_3,\psi_3 \rangle = 1).

    Prove that V3≄sgnV3V_3 \not \simeq \operatorname{sgn} \otimes V_3 as S4S_4-representations.

    (In particular, S4S_4 has at least two irreducible representations of dimension 3.)

    ::: {.solution}

    We first consider the representation V2V_2 of S3S_3. Write χ2\chi_2 of the character of this irreducible representation. The character of sgnχ2\operatorname{sgn}\chi_2 is then given by the product sgnχ2\operatorname{sgn}\chi_2.

    g1(12)(123)ψ2201sgn111sgnψ2201\begin{array}{l|lll} g & 1 & (12) & (123) \\ \hline \psi_2 & 2 & 0 & -1 \\ \operatorname{sgn} & 1 & -1 & 1 \\ \operatorname{sgn} \psi_2 & 2 & 0 & -1 \end{array}

    Inspecting the table we see that ψ2=sgnψ2\psi_2 = \operatorname{sgn}\psi_2. This shows that V2V_2 is isomorphic to sgnV2\operatorname{sgn} \otimes V_2 as representations for S3S_3.

    :::

  2. Let VV be a representation of GG.

    For an irreducible representation LL, consider the set S={SVSL}\mathcal{S}=\{ S \subseteq V \mid S \simeq L\} of all invariant subspaces that are isomorphic to LL as GG-representations.

    Put V(L)=SSS.V_{(L)} = \sum_{S \in \mathcal{S}} S.

    a. Prove that V(L)V_{(L)} is an invariant subspace, and show that V(L)V_{(L)} is isomorphic to a direct sum V(L)LLV_{(L)} \simeq L \oplus \cdots \oplus L as GG-representations.

    ::: {.solution}

    First of all, we note more generally that if II is an index set and if WiVW_i \subset V is a GG-invariant subspace for each iIi \in I, then iIWi\displaystyle \sum_{i \in I} W_i is again an invariant subspace. (The proof is straightforward from the definitions). This confirms that V(L)V_{(L)} is an invariant subspace.

    To prove the remaining assertion, we proceed as follows.

    Let us say that a GG-representation WW is LL-isotypic if every irreducible invariant subspace of WW is isomorphic to LL.

    It is immediate that V(L)V_{(L)} is LL-isotypic. We are going to prove:

    If WW is an LL-isotypic GG-representation, then WW is isomorphic to a direct sum WLL.W \simeq L \oplus \cdots \oplus L.

    Proceed by induction on dimW\dim W. If dimW=0\dim W = 0 then W={0}W = \{0\} and the result is immediate (WW is the direct sum of zero copies of LL).

    Now observe that if dimW>0\dim W > 0 then WW contains an invariant subspace isomorphic to LL, so that dimWdimL\dim W \ge \dim L.

    Now if dimW=dimL\dim W = \dim L, then WLW \simeq L and the result holds in this case.

    Finally, suppose that dimW>dimL\dim W > \dim L and let SWS \subset W be an invariant subspace with SLS \simeq L.

    By complete reducibility we may find an invariant subspace UWU \subset W such that WW is the internal direct sum W=SU.W = S \oplus U.

    Since dimW=dimS+dimU\dim W = \dim S + \dim U, we have dimU<dimW\dim U < \dim W. Moreover, UU is also LL-isotypic. So by induction on dimension, we know that ULL,U \simeq L \oplus \cdots \oplus L, (say, a direct sum of dd copies of LL).

    But then W=SUL(LL)=LLLW = S \oplus U \simeq L \oplus ( L \oplus \cdots \oplus L) = L \oplus L \oplus \cdots \oplus L is isomorphic to a direct sum of d+1d+1 copies of LL.

    :::

    b. Prove that the quotient representation V/V(L)V/V_{(L)} has no invariant subspaces isomorphic to LL as GG-representations.

    ::: {.solution}

    Write π:VV/V(L)\pi:V \to V/V_{(L)} for the quotient map vv+V(L)v \mapsto v + V_{(L)}; thus π\pi is a surjective homomorphism of GG-representations.

    Suppose by way of contradiction that SV/V(L)S \subset V/V_{(L)} is an invariant subspace isomorphic to LL, and let SVS' \subset V be the inverse image under π\pi of SS: S=π1(S).S' = \pi^{-1}(S).

    Then SS' is an invariant subspace of VV containing V(L)V_{(L)}, and the restriction of π\pi to SS' defines a surjective mapping πS:SSL.\pi_{\mid S'}:S' \to S \simeq L.

    If KK denotes the kernel of πS\pi_{\mid S'}, then complete reducibility implies that there is an invariant subspace MM of VV such that SS' is the internal direct sum ()S=KM.(*) \quad S' = K \oplus M.

    In particular, the invariant subspace MM is isomorphic to LL as GG-representations. But then by definition we have MV(L)M \subset V_{(L)} contradicting the condition MK={0}M \cap K = \{0\} which must hold by ()(*). This contradiction proves the result.

    :::

    c. If L1,L2,,LmL_1,L_2,\cdots,L_m is a complete set of non-isomorphic irreducible representations for GG, prove that VV is the internal direct sum V=i=1mV(Li).V = \bigoplus_{i=1}^m V_{(L_i)}.

    ::: {.solution}

    We first note that VV is equal to the sum i=1mV(Li);\sum_{i=1}^m V_{(L_i)}; indeed, if W=i=1mV(Li)W = \sum_{i=1}^m V_{(L_i)}, then by complete reducibility V=WWV = W \oplus W' for an invariant subspace WW'. But if W0W' \ne 0 then WW' contains an irreducible invariant subspace, so that WV(Li)0W' \cap V_{(L_i)} \ne 0 for some ii and hence WW0W' \cap W \ne 0; this is impossible since the internal sum V=WWV = W \oplus W' is direct. This argument shows that W=0W' = 0 and hence that V=WV = W.

    Finally, we show that the sum i=1mV(Li)\sum_{i=1}^m V_{(L_i)} is direct, i.e. that for each jj we have ()V(Lj)(ijV(Li))=0.(\clubsuit) \quad V_{(L_j)} \cap \left ( \sum_{i\ne j} V_{(L_i)} \right ) = 0.

    Wrote II for the intersection appearing in ()(\clubsuit); thus, II is an invariant subspace of VV. If II is non-zero, it has an irreducible invariant subspace SS. Since IV(Lj)I \subset V_{(L_j)} and since V(Lj)V_{(L_j)} is LjL_j-isotypic, we conclude that SLj.S \simeq L_j.

    But then SV(Li)=0S \cap V_{(L_i)} = 0 for every iji \ne j so that S(ijV(Li))=0.S \cap \left( \sum_{i \ne j} V_{(L_i)} \right ) = 0. Since I(ijV(Li))I \subset \displaystyle \left( \sum_{i \ne j} V_{(L_i)} \right ), we conclude that I=0I = 0.

    This completes the proof that VV is the direct sum of the V(Li)V_{(L_i)}, as required. :::

  3. Let χ\chi be the character of a representation VV of GG. For gGg\in G prove that χ(g)=χ(g1)\overline{\chi(g)} = \chi(g^{-1}).

    Is it true for any arbitrary class function f:GCf:G \to \mathbb{C} that f(g)=f(g1)\overline{f(g)} = f(g^{-1}) for every gg? (Give a proof or a counterexample...)

    ::: {.solution}

    Let ρ(g):VV\rho(g):V \to V denote the linear automorphism of VV determined by the action of gGg \in G. Then χ(g)=tr(ρ(g))\chi(g) = \operatorname{tr}(\rho(g)).

    Now, since ρ(g)\rho(g) has finite order, say nn, its minimal polynomial divides Tn1C[T]T^n - 1 \in \mathbb{C}[T], and hence every eigenvalue of ρ(g)\rho(g) is an nn-th root of unity.

    For any nn-th root of unity ζ\zeta, note that ζ=ζ1\overline{\zeta} = \zeta^{-1}.

    Write α1,,αd\alpha_1,\cdots,\alpha_d for the eigenvalues of ρ(g)\rho(g), with multiplicity (so that d=dimVd = \dim V). Notice that ρ(g1)\rho(g^{-1}) has eigenvalues α11,,αd1.\alpha_1^{-1},\cdots,\alpha_d^{-1}.

    Thus

    χ(g)=i=1dαiandχ(g1)=i=1dαi1.\chi(g) = \sum_{i=1}^d \alpha_i\quad \text{and} \quad \chi(g^{-1}) = \sum_{i=1}^d \alpha_i^{-1}.

    Now, we see that χ(g)=i=1dαi=i=1dαi1=χ(g1)\overline{\chi(g)} = \sum_{i=1}^d \overline{\alpha_i} = \sum_{i=1}^d \alpha_i^{-1} = \chi(g^{-1}) as required.

    It is not true that f(g)=f(g1)\overline{f(g)} = f(g^{-1}) for every class function ff and every gGg \in G. Indeed, let f=αδ1f = \alpha\delta_1 be a multiple of the characterisitic function δ1\delta_1 of the trivial conjugacy class {1}\{1\}.

    Then f(1)=α\overline{f(1)} = \overline{\alpha} while f(11)=f(1)=αf(1^{-1}) = f(1) = \alpha, so that if α∉R\alpha \not \in \mathbb{R}, we have f(1)f(11)\overline{f(1)} \ne f(1^{-1}).

    :::

  4. For a prime number pp, let k=Fp=Z/pZk=\mathbb{F}_p = \mathbb{Z}/p\mathbb{Z} be the field with pp elements. Let VV be an nn-dimensional vector space over Fp\mathbb{F}_p for some natural number nn, and let ,:V×Vk\langle \cdot,\cdot \rangle: V \times V \to k be a non-degenerate bilinear form on VV.

    (A common example would be to take V=FpnV = \mathbb{F}_{p^n} the field of order pnp^n, and α,β=trFpn/Fp(αβ)\langle \alpha,\beta\rangle = \operatorname{tr}_{\mathbb{F}_{p^n}/\mathbb{F}_p}(\alpha \beta) the trace pairing).

    Let us fix a non-trivial group homomorphism ψ:kC×\psi:k \to \mathbb{C}^\times (recall that k=Z/pZk = \mathbb{Z}/p\mathbb{Z} is an additive group, while C×\mathbb{C}^\times is multiplicative). Thus ψ(α+β)=ψ(α)ψ(β)for allα,βk.\psi(\alpha + \beta) = \psi(\alpha)\psi(\beta) \quad \text{for all} \quad \alpha,\beta \in k. If you want an explicit choice, set ψ(j+pZ)=exp(j2πi/p)=exp(2πi/p)j.\psi(j + p\mathbb{Z}) = \exp(j \cdot 2 \pi i/p) = \exp(2 \pi i /p)^j.

    For a vector vVv \in V, consider the mapping Ψv:VC×\Psi_v:V \to \mathbb{C}^\times given by the rule Ψv(w)=ψ(w,v).\Psi_v(w) = \psi( \langle w,v \rangle ).

    a. Show that Ψv\Psi_v is a group homomorphism VC×V \to \mathbb{C}^\times.

    ::: {.solution}

    For w,wVw,w' \in V notice that Ψv(w+w)=ψ(w+w,v)=ψ(w,v+w,v)since the form is bilinear=ψ(w,v)ψ(w,v)since ψ is a group homom=Ψv(w)Ψv(w)by definition.\begin{align*} \Psi_v(w + w') &= \psi(\langle w + w',v \rangle) & \\ &= \psi(\langle w,v \rangle + \langle w',v \rangle) & \text{since the form is bilinear} \\ &= \psi(\langle w,v \rangle) \cdot \psi(\langle w',v \rangle) & \text{since $\psi$ is a group homom} \\ &= \Psi_v(w)\cdot \Psi_v(w') & \text{by definition.} \end{align*} This confirms that Ψv\Psi_v is a group homomorphism.

    :::

    b. Show that the assignment vΨvv \mapsto \Psi_v is injective (one-to-one).

    (This assignment is a function VHom(V,C×)V \to \operatorname{Hom}(V,\mathbb{C}^\times). In fact, it is a group homomorphism. Do you see why? How do you make Hom(V,C×)\operatorname{Hom}(V,\mathbb{C}^\times) into a group?)

    ::: {.solution}

    One checks that Hom(V,C×)\operatorname{Hom}(V,\mathbb{C}^\times) is a multiplicitive group (this is the dual group V^\widehat V of VV, mentioned in the lectures); the product of ϕ,ψV^\phi,\psi \in \widehat V is given by the rule gϕ(g)ψ(g)g \mapsto \phi(g) \cdot \psi(g).

    We note that the assignment vΨvv \mapsto \Psi_v is a group homomorphism. For v,vVv,v' \in V we must argue that Ψv+v=ΨvΨv.\Psi_{v+v'} = \Psi_v \Psi_{v'}.

    For wWw \in W we have Ψv+v(w)=ψ(v+v,w)=ψ(v,w+v,w)since the form is bilinear=ψ(v,w)ψ(v,w)since ψ is a group homom=Ψv(w)Ψv(w)by defn\begin{align*} \Psi_{v+v'}(w) =& \psi( \langle v + v',w \rangle ) & \\ =& \psi(\langle v,w \rangle + \langle v',w\rangle) & \text{since the form is bilinear}\\ =& \psi(\langle v,w \rangle )\cdot \psi(\langle v',w\rangle) & \text{since $\psi$ is a group homom}\\ =& \Psi_v(w)\cdot \Psi_{v'}(w) & \text{by defn} \\ \end{align*}

    Now to show that vΨvv \mapsto \Psi_v is injective, it is enough to argue that the kernel of this mapping is {0}\{0\}.

    So, suppose that Ψv\Psi_v is the identity element of V^\widehat{V}. In other words, suppose that Ψv(w)=1\Psi_v(w) = 1 for every wVw \in V. This shows that ψ(v,w)=1\psi(\langle v,w \rangle) = 1 for every wVw \in V. Since ψ\psi is a non-trivial homomorphism FpC×\mathbb{F}_p \to \mathbb{C}^\times, we know that kerψ={0}\ker \psi = \{0\} (remember that kk has prime order...) and we conclude that v,w=0\langle v,w \rangle = 0 for every wWw \in W.

    (Note that v,w=0\langle v,w \rangle = 0 is an equality in k=Fp=Z/pZk = \mathbb{F}_p = \mathbb{Z}/p\mathbb{Z}).

    Since the form ,\langle \cdot,\cdot \rangle is non-degenerate, so we may now conclude that v=0v = 0.

    This proves that the kernel of the mapping vΨvv \mapsto \Psi_v is {0}\{0\}, hence the mapping is injective.

    :::

    b. Show that any group homomorphism Ψ:VC×\Psi:V \to \mathbb{C}^\times has the form Ψ=Ψv\Psi = \Psi_v for some vVv \in V.

    Conclude that there are exactly V=qn|V| = q^n group homomorphisms VC×V \to \mathbb{C}^\times.

    ::: {.solution}

    We observed in class that for any finite abelian group AA, there is an isomorphism AA^A \simeq \widehat{A}.

    In particular, A=A^|A| = |\widehat{A}|.

    Applying this in the case A=VA = V, we conclude that V=V^=Hom(V,C×).|V| = |\widehat{V}| = |\operatorname{Hom}(V,\mathbb{C}^\times)|.

    Now, we have define an injective mapping vΨv:V(^V).v \mapsto \Psi_v:V \to \widehat(V). Since the domain and co-domain of this mapping are finite of the same order, the mapping vΨvv \mapsto \Psi_v is also surjective.

    Thus the pigeonhole principal shows that every homomorphism Ψ:VC×\Psi:V \to \mathbb{C}^\times has the form Ψ=Ψv\Psi = \Psi_v for some vVv \in V, as required.

    :::