Bourbaki

§ 21. Linear representations of finite groups

In this section, G is a group, and K is a commutative ring. If G is a finite group, then we denote by ∣G∣|G| the element Card(G)⋅1\cdot 1 of K. From No. 5 on, we assume that the group G is finite, that the ring K is an algebraically closed field, and that ∣G∣|G| is not zero.

1. Linear Representations

Definition 1 00NC

Let M be a K-module. A linear representation of G in M is a group homomorphism from G to the linear group GL(M)\mathbf{G}\mathbf{L}(M) (II, §1, No. 2, p. 195).

Let π\pi be a linear representation of G in a K-module M. We also say that the pair (M,π)(M, \pi ) is a linear representation of G. When K is nonzero and M is a free K-module, the dimension of M is called the degree (or dimension) of the representation π\pi.

Recall that the canonical mapping g↦egg\mapsto e_g from G to the algebra K[G] of the group G is a basis of the K-module K[G]. By abuse of notation, we also denote the element ege_g of K[G] by gg. Given a linear representation π:G→GL(M)\pi : G\rightarrow \mathbf{G}\mathbf{L}(M) of G, there exists a unique homomorphism of K-algebras from K[G] to EndK(M)_K(M) that extends π\pi (III, §2, No. 6, p. 447, Example); we also denote this extension by π\pi. It is a representation of the algebra K[G] in M (VIII, p. 373, Definition 1), so that M is a K[G]-module. Conversely, every linear representation ρ\rho of the algebra K[G] in M defines a linear representation of G in M given by g↦ρ(g)g\mapsto \rho (g).

We will freely apply the terminology concerning the K[G]-module structure to linear representations of the group G. For example, we speak of a subrepresentation, a simple representation, a direct sum of representations, etc. Let (M,π)(M, \pi ) and (M′,π′)(M', \pi ') be linear representations of G; we denote by HomG(π,π′)_G(\pi , \pi ') the K-module HomK[G](M,M′)_{K[G]}(M,M') of homomorphisms of K[G]-modules from M to M′M'.[^1]

A mapping ff from G to K is called a central function if we have f(gg′)=f(gg') = f(g′g)f(g'g) for every pair (g,g′)(g, g') of elements of G; equivalently, we have f(ghg−1)=f(ghg^{-1}) = f(h)f(h) for every pair of elements g,hg, h of G. The central functions are therefore the functions on G whose restriction to each conjugacy class is constant. They form a submodule of the space of mappings from G to K, denoted by ZK(G)\mathscr{Z}_K(G). When G is finite, the K-algebra ZK(G)\mathscr{Z}_K(G) is a free K-module of dimension the number of conjugacy classes of G. The center Z(K[G]) of the algebra K[G] consists of the elements a=∑agga=\sum a_gg of K[G] such that hah−1=ahah^{-1}=a for every h∈Gh\in G. Now, we have

hah−1=∑g∈Gah−1ghg hah^{-1}=\sum_{g\in G}a_{h^{-1}gh}g

consequently, when G is finite, the center Z(K[G]) of the algebra K[G] consists of the central functions.

Let (M,π)(M, \pi ) be a linear representation of G. Suppose that M is a finite-dimensional free K-module. The trace of π\pi is the trace of the representation of K[G] associated with π\pi, that is (VIII, p. 382), the linear form a↦a\mapsto Tr(π(a)\pi (a)) on K[G]. This linear form is determined by the mapping g↦g\mapsto Tr(π(g)\pi (g)) from G to K, which we call the character of the representation π\pi and denote by χπ\chi_{\pi}. The character of a representation is a central function (II, §4, No. 3, p. 273, Proposition 3).

Let M and M′M' be finite-dimensional free K-modules. Let π\pi and π′\pi ' be linear representations of G in M and M′M', respectively. Then M⊕M′M\oplus M' is a finite-dimensional free K-module, and, by Proposition 1 of III, §9, No. 2, p. 543, we have

χπ⊕π′=χπ+χπ′ \chi_{\pi\oplus\pi'}=\chi_{\pi}+\chi_{\pi'}

More generally, let

0⟶M′⟶M⟶M′′⟶0 0\longrightarrow M'\longrightarrow M\longrightarrow M''\longrightarrow 0

be an exact sequence of K[G] modules, and let π,π′\pi ,\pi ', and π′′\pi '' be the linear representations of G associated with M,M′M, M', and M′′M'', respectively. Suppose that the K-modules M′M' and M′′M'' are free and finite-dimensional. Then so is M (II, §1, No. 11, p. 218, Proposition 21), and we have

χπ=χπ′+χπ′′(1) \chi_{\pi}=\chi_{\pi'}+\chi_{\pi''} \tag{1}

Proposition 1 00ND

Suppose that K is a commutative field. Let π\pi and π′\pi ' be finite-dimensional semisimple K-linear representations of G.

a) If the characteristic polynomials of π(g)\pi (g) and π′(g)\pi '(g) are the same for every g∈Gg\in G, then the representations π\pi and π′\pi ' are isomorphic.

b) Suppose, moreover, that K has characteristic 0. If the characters χπ\chi_{\pi} and χπ′\chi_{\pi'} are equal, then the representations π\pi and π′\pi ' are isomorphic.

Assertion a) follows from Corollary 1 of VIII, p. 386; assertion b) follows from part a) of the corollary of VIII, p. 384.

Example 1 00NE

The unit or trivial representation of G is the representation (K,ε)(K, \varepsilon ), where ε(g)=\varepsilon (g) = IdK_K for every g∈Gg\in G. Its character is the constant function with value 1.

Example 2 00NF

The (left) regular representation of G is the representation γ\boldsymbol{\gamma } of G in K[G] defined by γ(g)(x)=gx\boldsymbol{\gamma }(g)(x) =gx for g∈Gg\in G and x∈K[G]x\in K[G]. It corresponds to the left regular representation of the algebra K[G] (VIII, p. 375). Suppose that the group G is finite. The regular representation then has finite degree. For every element gg of G different from the identity, the matrix of left multiplication by gg in K[G] with respect to the canonical basis is the matrix of a permutation without fixed point. We therefore have

∣G∣|G| if gg is the identity element,

χγ(g)=(2) \chi_{\boldsymbol{\gamma }}(g) = \tag{2}

0 otherwise.

We define the right regular representation of G likewise. The biregular representation of G is the representation ρ\rho of G×GG\times G in K[G] defined by ρ(g,g′)(x)=gxg′−1\rho (g, g')(x) =gxg^{'-1} for (g,g′)∈G×G(g, g')\in G\times G and x∈K[G]x\in K[G].

Example 3 00NG

Given a linear representation (M,π)(M, \pi ) of G, the contragredient or dual representation π∨\pi^{\vee} of π\pi is the representation of G in the K-module dual to M defined by the relation π∨(g)=tπ(g−1)\pi^{\vee}(g) =^t\pi (g^{-1}) for every g∈Gg\in G (cf. II, §2, No. 5, p. 234). If M is a finite-dimensional free K-module, then so is its dual, and we have χπ∨(g)=χπ(g−1)\chi_{\pi^{\vee}}(g) =\chi_{\pi}(g^{-1}) for every g∈Gg\in G.

Example 4 00NH

Let (M,π)(M, \pi ) and (M′,π′)(M', \pi ') be linear representations of G. In Example 1 of VIII, p. 198, we defined a K[G]-module structure on M⊗KM′M\otimes_KM'. The corresponding linear representation is called the tensor product of π\pi and π′\pi ' and is denoted by π⊗π′\pi \otimes \pi '. For g∈G,x∈Mg\in G,x\in M, and x′∈M′x'\in M', we have ((π⊗π′)(g))(x⊗x′)=π(g)x⊗π′(g)x′((\pi \otimes \pi ')(g))(x\otimes x') =\pi (g)x\otimes \pi '(g)x'.

If M and M' are finite-dimensional free K-modules, then by Proposition 2 of III, §9, No. 2, p. 543, we have

(3)

χπ⊗π′=χπχπ′. \chi_{\pi \otimes \pi'} = \chi_{\pi} \chi_{\pi'}.

Example 5 00NI

Suppose that G is the product G′×G′′G' \times G'' of two groups. The K-linear mapping from K[G′]⊗KK[G′′]K[G'] \otimes_K K[G''] to K[G]K[G] that sends g′⊗g′′g' \otimes g'' to (g′,g′′)(g', g'') for g′∈G′g' \in G' and g′′∈G′′g'' \in G'' is an algebra isomorphism. Let (M′,π′)(M', \pi') be a linear representation of G′G' and (M′′,π′′)(M'', \pi'') a linear representation of G′′G''. The external tensor product of π′\pi' and π′′\pi'', denoted by π′⊗π′′\pi' \otimes \pi'', is the representation of G′×G′′G' \times G'' in the vector space M′⊗M′′M' \otimes M'' defined by (π′⊗π′′)(g′,g′′)=π′(g′)⊗π′′(g′′)(\pi' \otimes \pi'')(g', g'') = \pi'(g') \otimes \pi''(g'') for (g′,g′′)∈G′×G′′(g', g'') \in G' \times G''. If M′M' and M′′M'' are finite-dimensional free K-modules, then M′⊗KM′′M' \otimes_K M'' is a finite-dimensional free K-module and the character of the representation π′⊗π′′\pi' \otimes \pi'' is given by the formula

χπ′⊗π′′(g′,g′′)=χπ′(g′)χπ′′(g′′) \chi_{\pi' \otimes \pi''}(g', g'') = \chi_{\pi'}(g') \chi_{\pi''}(g'')
for g′∈G′g' \in G' and g′′∈G′′g'' \in G'' (Proposition 2 of III, §9, No. 2, p. 542).

Example 6 00NJ

Let (V,π)(V, \pi) be a linear representation of G such that V is a finite-dimensional free K-module. We define a representation ρ\rho of G×GG \times G in End⁡K(V)\operatorname{End}_K(V) by the formula

ρ(g,g′)(u)=π(g′)∘u∘π(g−1). \rho(g, g')(u) = \pi(g') \circ u \circ \pi(g^{-1}).
The K-module isomorphism θV:V∗⊗KV→End⁡K(V)\theta_V : V^* \otimes_K V \to \operatorname{End}_K(V) of II, §4, No. 2, p. 271 is an isomorphism of representations from the external tensor product π∨⊗π\pi^\vee \otimes \pi to ρ\rho. The mapping g↦ρ(1,g)g \mapsto \rho(1, g) (resp. g↦ρ(g,1)g \mapsto \rho(g, 1)) is a representation of G isomorphic to πdim⁡KV\pi^{\dim_K V} (resp. (π∨)dim⁡KV(\pi^\vee)^{\dim_K V}).

Let L be a commutative K-algebra, and let (M,π)(M, \pi) be a linear representation of the group G. The group homomorphism π(L):G→GL(M(L))\pi_{(L)} : G \to \mathbf{GL}(M_{(L)}) defined by g↦Id⁡L⊗π(g)g \mapsto \operatorname{Id}_L \otimes \pi(g) is a linear representation of G in the L-module M(L)M_{(L)}, called the linear representation of G deduced from the representation π\pi by extension of the ring of scalars K to L.

Suppose that K is a field and that L is a nonzero commutative K algebra. Let (M,π)(M, \pi) and (M′,π′)(M', \pi') be linear representations of G. The representations π\pi and π′\pi' are isomorphic if and only if π(L)\pi_{(L)} and π(L)′\pi'_{(L)} are (VIII, p. 37, Theorem 3).

Suppose, moreover, that the algebra L is an extension of K. Consider the rings RK(G)R_K(G) and RL(G)R_L(G) defined in Example 1 of VIII, p. 198. Extension of scalars defines a ring homomorphism

u:RK(G)⟶RL(G). u : R_K(G) \longrightarrow R_L(G).

This homomorphism is injective, and an element ξ∈RK(G)\xi \in R_K(G) is effective if and only if u(ξ)u(\xi ) is (VIII, p. 195, Theorem 1).

2. Maschke’s Theorem

Theorem 1 00NK

Suppose that the group G is finite. Let M be a K[G]-module, and let N be a K[G]-submodule of M. Suppose that N is a direct factor of the K-module M and that ∣G∣|G| is invertible in K. Then N is a direct factor of the K[G]-module M.

Let pp be a K-linear projector in M with image N. We define an endomorphism qq of the K-module M by setting

q(x)=∣G∣−1∑g∈Ggp(g−1x) q(x) =|G|^{-1}\sum_{g\in G}g p(g^{-1}x)

for every x∈Mx\in M. Since N is stable under the action of G and pp induces the identity on N, we see that qq sends M into N and induces the identity on N.

The K-module M is therefore the direct sum of the image N of qq and the kernel of qq. We have gq(x)=q(gx)g q(x) =q(gx) for every x∈Mx\in M and g∈Gg\in G, so that the kernel of qq is a K[G]-submodule of M. This proves that N is a direct factor of the K[G]-module M.

Corollary 1 (Maschke) 00SC

Suppose that the group G is finite and that K is a commutative field. The algebra K[G] is semisimple if and only if the element ∣G∣|G| of the field K is not zero.

Suppose ∣G∣≠0|G| \not= 0. By Theorem 1, every K[G]-submodule of K[G]sK[G]_s is a direct factor. Hence K[G] is semisimple.

Conversely, suppose that ∣G∣|G| is zero, and denote by ε\varepsilon the element ∑g∈Gg\sum_{g\in G}g of the center of K[G]. We have ε≠0\varepsilon \not= 0 but ε2=∣G∣ε=0\varepsilon^2=|G|\varepsilon = 0, so K[G] is not semisimple (VIII, p. 153, Proposition 3, a) and VIII, p. 157, Remark 1).

Corollary 2 00NL

Suppose that the group G is finite and that ∣G∣|G| is invertible in K. An exact sequence of K[G]-modules splits if and only if it splits as an exact sequence of K-modules.

Given an exact sequence of K[G]-modules

0 // M′f{M'}^f // M // M′′M'' // 00,

it suffices to apply Theorem 1 to the image of the morphism ff.

Corollary 3 00NM

Suppose that the group G is finite and that ∣G∣|G| is invertible in K. A K[G]-module is projective if and only if it is projective as a K-module.

Let P be a K[G]-module. If P is a direct factor of a free K[G]-module M, then it is a fortiori a direct factor of the free K-module M. The converse follows from Corollary 2, in view of II, §2, No. 2, p. 231, Proposition 4, d).

Corollary 4 00NN

a) Suppose that the group G is finite and that K is a commutative field of characteristic 0. Two finite-dimensional linear representations of G are isomorphic if and only if they have the same character.

b) Suppose that K is a perfect field of characteristic a prime number pp and that the group G is finite, of cardinal prime to pp. The character of a finite-dimensional linear representation of G is zero if and only if this representation is isomorphic to the direct sum of pp pairwise isomorphic representations.

Under the assumptions of the corollary, every finite-dimensional linear representation of G is semisimple. The corollary then follows from the corollary of VIII, p. 384.

Corollary 5 00NO

Suppose that the group G is finite and that K is a commutative field in which ∣G∣≠0|G| \not= 0. Let π\pi and π′\pi ' be finite-dimensional linear representations of G. Then π\pi and π′\pi ' are isomorphic if and only if for every gg in G, the endomorphisms π(g)\pi (g) and π′(g)\pi '(g) have the same characteristic polynomial.

This follows from Corollary 1 of VIII, p. 386.

3. Induced and Coinduced Representations

Let H be a subgroup of the group G.

If (V,π)(V, \pi ) is a linear representation of G, then the restriction of π\pi to H is a linear representation of H in V; we denote it by ResHG(π)^G_H(\pi ). The K[H]module associated with ResHG(π)^G_H(\pi ) is simply the module deduced from the K[G]-module V by restriction of scalars (II, §1, No. 13, p. 221). If V is a finite-dimensional free K-module, then the character of ResHG(π)^G_H(\pi ) is the restriction of the character of π\pi to H.

Let (M,σ)(M, \sigma ) be a representation of H.

We view K[G] as a (K[G](K[G], K[H])-bimodule and M as a K[H]-module. The K[G]-module T(M)=K[G]⊗K[H]M\mathscr{T}(M) = K[G]\otimes_{K[H]}M (VIII, p. 58) defines a linear representation of G, denoted by IndHG(σ)^G_H(\sigma ) and called the representation of G induced by σ\sigma. If (V,π)(V, \pi ) is a linear representation of G, then the K[H]-module H(V)=\mathscr{H}(V) = HomK[G](K[G],V)_{K[G]}(K[G],V) can be identified with the K[H]-module corresponding to the representation ResHG(π)^G_H(\pi ). Consequently, the adjunction morphism (VIII, p. 59) gives a K-module isomorphism, called canonical, from HomH(σ_H(\sigma, ResHG(π))^G_H(\pi )) to HomG_G(IndHG(σ),π^G_H(\sigma ), \pi ) (“Frobenius reciprocity”).

We view K[G] as a (K[H](K[H], K[G])-bimodule. The K[G]-module H(M)=\mathscr{H}(M) = HomK[H](K[G],M)_{K[H]}(K[G],M) defines a representation of G, denoted by CoindHG(σ)^G_H(\sigma ) and called the representation of G coinduced by σ\sigma. If (V,π)(V, \pi ) is a linear representation of G, then the K[H]-module T(V)=K[G]⊗K[G]V\mathscr{T}(V) = K[G]\otimes_{K[G]}V can be identified with the K[H]-module corresponding to the representation ResHG(π)^G_H(\pi ). Consequently, the adjunction morphism (loc. cit.) gives a K-module isomorphism, called canonical, from HomH_H(ResHG(π),σ^G_H(\pi ), \sigma ) to HomG(π_G(\pi, CoindHG(σ))^G_H(\sigma )).

Let ε:K[G]→K[H]\varepsilon : K[G]\rightarrow K[H] be the K-module homomorphism characterized by the relations ε(h)=h\varepsilon (h) =h if h∈h\in H and ε(g)=0\varepsilon (g) = 0 if g∈Gg\in G - H. The mapping ε\varepsilon is a homomorphism of (K[H](K[H], K[H])-bimodules. Let (M,σ)(M, \sigma ) be a linear representation of H. The mapping v↦v∘εv\mapsto v\circ \varepsilon from HomK[H](K[H],M)_{K[H]}(K[H],M) to HomK[H](K[G],M)_{K[H]}(K[G],M) is a homomorphism of K[H]-modules. By identifying M with HomK[H](K[H],M)_{K[H]}(K[H],M), we obtain a homomorphism of K[H]-modules from M to ResHG^G_H(CoindHG(σ)^G_H(\sigma )). Frobenius reciprocity sends it to a homomorphism ι\iota of K[G]-modules from IndHG(σ)^G_H(\sigma ) to CoindHG(σ)^G_H(\sigma ) characterized by the relation

ι(g⊗m)(g′)=ε(g′g)m \iota (g\otimes m)(g') =\varepsilon (g'g)m

for g,g′∈Gg, g'\in G and m∈Mm\in M. This homomorphism is called canonical.

Proposition 2 00NP

Let H be a subgroup of G, and let (M,σ)(M, \sigma ) be a linear representation of H. The canonical homomorphism ι:\iota : IndHG(σ)→^G_H(\sigma )\rightarrow CoindHG(σ)^G_H(\sigma ) is injective. If the subgroup H has finite index in G, then ι\iota is an isomorphism of K[G]-modules.

Let S⊂GS\subset G be a system of representatives of G/HG/H. The family (s)s∈S(s)_{s\in S} is a basis of the right K[H]-module K[G]. It follows that the mapping M(S)→M^{(S)}\rightarrow IndHG(σ)^G_H(\sigma ) defined by (ms)s∈S↦∑s∈Ss⊗ms(m_s)_{s\in S}\mapsto \sum_{s\in S}s\otimes m_s is an isomorphism of K-modules. For all s,s′∈Ss, s'\in S and every m∈Mm\in M, we have the relations

ι(s′⊗m)(s−1)={mif s=s′,0otherwise. \iota (s'\otimes m)(s^{-1}) =\begin{cases} m & \text{if } s=s',\\ 0 & \text{otherwise.}\end{cases}

It follows that ι′\iota ' is injective.

Suppose that H has finite index in G. The set S is then finite; let ρ\rho be the mapping from CoindHG(σ)^G_H(\sigma ) to IndHG(σ)^G_H(\sigma ) given by u↦∑s∈Ss⊗u(s−1)u\mapsto \sum_{s\in S}s\otimes u(s^{-1}). It satisfies (ι∘ρ(u))(s−1)=u(s−1)(\iota \circ \rho (u))(s^{-1}) =u(s^{-1}) for u∈u\in CoindHG(σ)^G_H(\sigma ) and s∈Ss\in S. Since the family (s−1)s∈S(s^{-1})_{s\in S} is a basis of the left K[H]-module K[G], the mapping ι∘ρ\iota \circ \rho is the identity mapping. Consequently, ι\iota is bijective with inverse bijection ρ\rho.

Let H be a subgroup of G of finite index. Let uu be a central function on H; denote by u0u^0 the function on G that extends uu and that is zero at every point of G - H. Let S be a system of representatives of G/HG/H. For g∈Gg\in G, set

(4) IndHG(u)(g)=∑s∈Su0(s−1gs)^G_H(u)(g) =\sum_{s\in S}u^0(s^{-1}gs).

Note that for all x,g∈Gx, g\in G and every h∈Hh\in H, we have u0((xh)−1gxh)=u^0((xh)^{-1}gxh) = u0(x−1gx)u^0(x^{-1}gx). It follows that IndHG(u)^G_H(u) is a central function on G that does not depend on the choice of S. We thus define a K-linear mapping IndHG^G_H from ZK(H)\mathscr{Z}_K(H) to ZK(G)\mathscr{Z}_K(G).

Proposition 3 00NQ

Let H be a subgroup of G of finite index. Let (M,σ)(M, \sigma ) be a linear representation of H. Suppose that M is a finite-dimensional free K-module. Denote by (V,π)(V, \pi ) the representation of G induced by σ\sigma. Then the K-module V is free and finite-dimensional, and

(5) χπ=\chi_{\pi}= IndHG(χσ)^G_H(\chi_{\sigma}).

Let S be a system of representatives of G/HG/H. As we saw above, the linear mapping MS→M^S\rightarrow IndHG(M)^G_H(M) given by (ms)s∈S↦∑s∈Ss⊗ms(m_s)_{s\in S}\mapsto \sum_{s\in S}s\otimes m_s is a K-module isomorphism. In particular, V is a finite-dimensional free K-module. For any g∈Gg\in G, denote by MgM_g the image of M by the mapping m↦g⊗mm\mapsto g\otimes m. For all g,g′∈Gg, g'\in G, we have Mg=Mg′M_g= M_{g'} if and only if gH=g′HgH =g'H, and V is the direct sum of the submodules MsM_s for s∈Ss\in S. For all g,g′∈Gg, g'\in G, we also have π(g)Mg′=Mgg′\pi (g)M_{g'}= M_{gg'}. For any g∈Gg\in G, denote by SgS_g the set of elements ss of S such that s−1gs∈Hs^{-1}gs\in H. For all g,g′∈Gg, g'\in G such that g′−1gg′∈Hg^{'-1}gg'\in H, the automorphism π(g)\pi (g) of V induces an automorphism π(g)g′\pi (g)_{g'} of Mg′M_{g'}, and we have

Tr(π(g)\pi (g)) =∑s∈Sg=\sum_{s\in S_g} Tr(π(g)s\pi (g)_s) =∑s∈Sg=\sum_{s\in S_g} Tr(σ(s−1gs)\sigma (s^{-1}gs)).

The last assertion of the proposition follows.

4. Representations and the Grothendieck Group

In this subsection, we assume that K is a commutative field. Given a finite-dimensional linear representation (M,π)(M, \pi ) of G, we denote by [π][\pi ] the class of the K[G]-module M in the Grothendieck ring RK(G)R_K(G) (VIII, p. 198, Example 1).

By the definition of the laws of composition on the ring RKR_K(G), if π\pi and π′\pi ' are finite-dimensional linear representations of G, then we have

[π⊕π′]=[π]+[π′],[π⊗π′]=[π][π′] [\pi \oplus \pi '] = [\pi ] + [\pi '],[\pi \otimes \pi '] = [\pi ] [\pi ']

The unit element of the ring RK(G)R_K(G) is the class of the unit representation of G (VIII, p. 399, Example 1).

The ring RK(G)R_K(G) is a free Z\mathbf{Z}-module with basis the set of classes of simple K[G]-modules of finite dimension over K. When ∣G∣|G| is invertible in K, two finite-dimensional representations are isomorphic if and only if they have the same class in RK(G)R_K(G) (VIII, p. 190, Corollary and p. 401, Corollary 1).

For every finite-dimensional linear representation (M,π)(M, \pi ) of G, the contragredient π∨\pi^{\vee} is also finite-dimensional. The representations π\pi and (π∨)∨(\pi^{\vee})^{\vee} are isomorphic. If π′\pi ' is also a finite-dimensional linear representation of G, then the representations (π⊗π′)∨(\pi \otimes \pi ')^{\vee} and π∨⊗π′∨\pi^{\vee}\otimes \pi^{'\vee} are isomorphic. If

0⟶M′⟶M⟶M′′⟶0 0\longrightarrow M'\longrightarrow M\longrightarrow M''\longrightarrow 0

is an exact sequence of K[G]-modules of finite dimension over K, then the contragredient representations give an exact sequence of K[G]-modules

0⟶M′′∨⟶M∨⟶M′∨⟶0 0\longrightarrow M^{''\vee}\longrightarrow M^{\vee}\longrightarrow M^{'\vee}\longrightarrow 0

(II, §7, No. 5, p. 299, Corollary of Theorem 5). By the universal property of Grothendieck groups (VIII, p. 186, Proposition 4), there exists an automorphism c↦c∨c\mapsto c^{\vee} of the ring RK(G)R_K(G) characterized by [π]∨=[π∨][\pi ]^{\vee}= [\pi^{\vee}] for every finite-dimensional linear representation π\pi of G; we have (c∨)∨=c(c^{\vee})^{\vee}=c for every element cc of RK(G)R_K(G).

Let H be a subgroup of G. Restriction preserves exact sequences. By the universal property of Grothendieck groups (loc. cit.), there consequently exists a group homomorphism, denoted by ResHG^G_H, from RK(G)R_K(G) to RK(H)R_K(H) characterized by the relation

(6) ResHG[π]=^G_H[\pi ] = [ResHG(π)]^G_H(\pi )]

for every finite-dimensional linear representation π\pi of G; it is a ring homomorphism. If H has finite index in G, then by Proposition 14 of II, §7, No. 7, p. 306, we can define, analogously, a group homomorphism IndHG^G_H from RK(H)R_K(H) to RK(G)R_K(G) characterized by the relation

(7) IndHG[σ]=^G_H[\sigma ] = [IndHG(σ)]^G_H(\sigma )]

for every finite-dimensional linear representation σ\sigma of H.

By relations (1) of VIII, p. 399 and (3) of VIII, p. 400 and the universal property of Grothendieck groups mentioned above, there exists a ring homomorphism ΘG\Theta_G from RK(G)R_K(G) to the algebra ZK(G)\mathscr{Z}_K(G) of central functions on G, characterized by ΘG([π])=χπ\Theta_G([\pi ]) =\chi_{\pi} for every finite-dimensional representation π\pi of G.

If H is a subgroup of G, then the homomorphisms ΘG\Theta_G and ΘH\Theta_H associated with the groups G and H are compatible with the operations ResHG^G_H and IndHG^G_H (VIII, p. 402 and VIII, p. 404, Proposition 3).

Suppose that G is the product G′×G′′G'\times G'' of two groups. By VIII, p. 213, Remark 2, there exists a Z\mathbf{Z}-linear mapping κ\kappa from RK(G′)⊗ZRK(G′′)R_K(G')\otimes_{\mathbf{Z}}R_K(G'') to the group RK(G′×G′′)R_K(G'\times G'') characterized by the relation κ([π′]⊗[π′′])=[π′π′′]\kappa ([\pi ']\otimes [\pi '']) = [\pi '\pi ''] for π′\pi ' (resp. π′′\pi '') a finite-dimensional representation of G′G' (resp. G′′G'') and π′π′′\pi '\pi '' the external tensor product (VIII, p. 400, Example 5). It is a ring homomorphism. If the field K is algebraically closed, then the mapping κ\kappa is an isomorphism (VIII, p. 213, Remark 2).

Suppose that G is the product G′×G′′G'\times G'' of two groups. Denote by ψ\psi the homomorphism from ZK(G′)⊗KZK(G′′)\mathscr{Z}_K(G')\otimes_K\mathscr{Z}_K(G'') to ZK(G)\mathscr{Z}_K(G) that sends f′⊗f′′f'\otimes f'' to the function (g′,g′′)↦f′(g′)f′′(g′′)(g', g'')\mapsto f'(g')f''(g''). The following diagram commutes:

RK(G′)⊗ZRK(G′′)κR_K(G')\otimes_{\mathbf{Z}}R_K(G'')^{\kappa} // RK(G)R_K(G)

ΘG′⊗ΘG′′ΘG\Theta_{G'}\otimes \Theta_{G''}\Theta_G

ZK(G′)⊗KZK(G′′)ψ\mathscr{Z}_K(G')\otimes_K\mathscr{Z}_K(G'')^{\psi} //ZK(G)/\mathscr{Z}_K(G).

5. Fourier Inversion Formula

For the remainder of this section, we assume that the group G is finite and that K is an algebraically closed field whose characteristic does not divide the order of G, so that the element ∣G∣|G| of K is not zero.

The algebra K[G] is semisimple (Maschke’s theorem) and finite-dimensional. Denote by G^\widehat{G} the set of classes of simple K[G]-modules. For every λ∈G^\lambda \in \widehat{G}, choose a linear representation (Vλ,πλ)(V_{\lambda}, \pi_{\lambda}) of G such that the associated K[G]-module has class λ\lambda. The set G^\widehat{G} is finite, and the vector spaces VλV_{\lambda} are finite-dimensional (VIII, p. 141, Example). For any λ∈G^\lambda \in \widehat{G}, denote by dλd_{\lambda} the degree of the representation πλ\pi_{\lambda}, that is, the dimension of the K-vector space VλV_{\lambda}, and denote its character by χλ\chi_{\lambda}.

Denote by F(G^)F(\widehat{G}) the product algebra ∏λ∈G^End⁡K(Vλ)\prod_{\lambda \in \widehat{G}} \operatorname{End}_K(V_\lambda) and by F‾\overline{\mathscr{F}} the mapping from K[G]K[G] to F(G^)F(\widehat{G}) defined by F‾(a)=(πλ(a))λ∈G^\overline{\mathscr{F}}(a) = (\pi_\lambda(a))_{\lambda \in \widehat{G}}. Since the field KK is algebraically closed, the mapping F‾\overline{\mathscr{F}} is an algebra isomorphism (loc. cit.).

For every λ∈G^\lambda \in \widehat{G}, the dimension of the algebra End⁡K(Vλ)\operatorname{End}_K(V_\lambda) is dλ2d_\lambda^2; that of the algebra K[G]K[G] is Card⁡(G)\operatorname{Card}(G). We therefore have the relation

Card⁡(G)=∑λ∈G^dλ2. \operatorname{Card}(G) = \sum_{\lambda \in \widehat{G}} d_\lambda^2.

Denote by τ\tau the trace in the algebra K[G]K[G]; by definition, the trace τ(a)\tau(a) of an element aa of K[G]K[G] is the trace of the endomorphism x↦axx \mapsto ax of K[G]K[G] (III, §7, No. 7, p. 515). Let a=∑g∈Gagga = \sum_{g \in G} a_g g be an element of K[G]K[G]; by formula (2), we have τ(ag−1)=∣G∣ag\tau(ag^{-1}) = |G| a_g for every g∈Gg \in G, and therefore the relation

a=∣G∣−1∑g∈Gτ(ag−1) g. a = |G|^{-1} \sum_{g \in G} \tau(ag^{-1})\ g.

Denote by τ^\widehat{\tau} the trace in the algebra F(G^)F(\widehat{G}). Let A=(Aλ)λ∈G^A = (A_\lambda)_{\lambda \in \widehat{G}} be an element of F(G^)F(\widehat{G}); we have (cf. III, §9, No. 3, p. 545, Example 3)

τ^(A)=∑λ∈G^dλTr⁡(Aλ). \widehat{\tau}(A) = \sum_{\lambda \in \widehat{G}} d_\lambda \operatorname{Tr}(A_\lambda).

Since the mapping F‾\overline{\mathscr{F}} is a KK-algebra isomorphism, we have τ^∘F‾=τ\widehat{\tau} \circ \overline{\mathscr{F}} = \tau, and therefore

τ(a)=τ^(F‾(a))=τ^((πλ(a))λ∈G^)=∑λ∈G^dλTr⁡(πλ(a))=∑λ∈G^dλTr⁡λ(a) \tau(a) = \widehat{\tau}(\overline{\mathscr{F}}(a)) = \widehat{\tau}\left((\pi_\lambda(a))_{\lambda \in \widehat{G}}\right) = \sum_{\lambda \in \widehat{G}} d_\lambda \operatorname{Tr}(\pi_\lambda(a)) = \sum_{\lambda \in \widehat{G}} d_\lambda \operatorname{Tr}_\lambda(a)

for every a∈K[G]a \in K[G], that is

τ=∑λ∈G^dλTr⁡λ. \tau = \sum_{\lambda \in \widehat{G}} d_\lambda \operatorname{Tr}_\lambda.

Therefore, by (2) of VIII, p. 399, for g∈Gg \in G, we have

∑λ∈G^dλχλ(g)={∣G∣if g is the identity element,0otherwise. \sum_{\lambda \in \widehat{G}} d_\lambda \chi_\lambda(g) = \begin{cases} |G| & \text{if } g \text{ is the identity element,} \\ 0 & \text{otherwise.} \end{cases}

For a∈K[G]a \in K[G], relation (9) takes on the following form:

a=∣G∣−1∑g∈G∑λ∈G^dλTr⁡(πλ(a)πλ(g−1)) g; a = |G|^{-1} \sum_{g \in G} \sum_{\lambda \in \widehat{G}} d_\lambda \operatorname{Tr}(\pi_\lambda(a)\pi_\lambda(g^{-1}))\ g;

it follows that for every element A=(Aλ)λ∈G^A = (A_\lambda)_{\lambda \in \widehat{G}} of F(G^)F(\widehat{G}), we have

F‾−1(A)=∣G∣−1∑g∈G∑λ∈G^dλTr⁡(Aλπλ(g−1)) g \overline{\mathscr{F}}^{-1}(A) = |G|^{-1} \sum_{g \in G} \sum_{\lambda \in \widehat{G}} d_\lambda \operatorname{Tr}(A_\lambda \pi_\lambda(g^{-1}))\ g

(“Fourier inversion formula”).

For μ∈G^\mu\in \widehat{G}, denote by jμ:j_\mu: EndK(Vμ)⟶∏λ∈G^_K(V_\mu)\longrightarrow \prod_{\lambda\in\widehat{G}} EndK(Vλ)_K(V_{\lambda}) the mapping such that jμ(u)=(vλ)j_\mu(u) = (v_{\lambda}), where vλ=0v_{\lambda}= 0 if λ≠μ\lambda \not=\mu and vμ=uv_\mu=u. By formula (14), we have

(15) F‾−1(jμ(u))=∣G∣−1dμ∑g∈G\overline{\mathscr{F}}^{-1}(j_\mu(u)) =|G|^{-1}d_\mu\sum_{g\in G} Tr(uπμ(g−1)u\pi_\mu(g^{-1}))gg.

The center of the algebra F(G^)=∏λ∈G^F(\widehat{G}) =\prod_{\lambda\in\widehat{G}} EndK(Vλ)_K(V_{\lambda}) consists of the families (aλ1Vλ)λ∈G^(a_{\lambda}1_{V_{\lambda}})_{\lambda\in\widehat{G}}, where (aλ)(a_{\lambda}) is a family of elements of K. It is the image by F‾\overline{\mathscr{F}} of the center of the algebra K[G]. That center therefore has a basis (eλ)λ∈G^(e_{\lambda})_{\lambda\in\widehat{G}} characterized by the relation

πλ(eμ)=δλμ1Vλ(16) \pi_{\lambda}(e_\mu) =\delta_{\lambda \mu}1_{V_{\lambda}} \tag{16}

for λ,μ∈G^\lambda , \mu\in \widehat{G}, where δλμ\delta_{\lambda \mu} is the Kronecker delta function. By formula (15), we have, for every μ∈G^\mu\in \widehat{G},

eμ=∣G∣−1dμ∑g∈Gχμ(g−1)g(17) e_\mu=|G|^{-1}d_\mu\sum_{g\in G}\chi_\mu(g^{-1})g \tag{17}

These elements satisfy the relations

(18) ∑λ∈G^eλ=1,eμ2=eμ\sum_{\lambda\in\widehat{G}}e_{\lambda}= 1,e^2_\mu=e_\mu, and eμeν=0e_\mu e_{\nu}= 0

for all μ,ν∈G^\mu, \nu \in \widehat{G} such that μ≠ν\mu\not=\nu; they are the indecomposable idempotents of Z(K[G]) (VIII, p. 147, Proposition 15)

Remark 00NR

Let (V,π)(V, \pi ) be a linear representation of G. By Maschke’s theorem, the K[G]-module V is semisimple. For any λ∈G^\lambda \in \widehat{G}, denote by VλV^{\lambda} the isotypical component of type λ\lambda of the K[G]-module V; we have V=⨁λ∈G^VλV =\bigoplus_{\lambda\in\widehat{G}}V^{\lambda}. By Proposition 15 of VIII, p. 147 and formula (17), the projector of V with image VλV^{\lambda} associated with this decomposition of V is equal to

π(eλ)=∣G∣−1dλ∑g∈Gχλ(g−1)π(g)(19) \pi (e_{\lambda}) =|G|^{-1}d_{\lambda}\sum_{g\in G}\chi_{\lambda}(g^{-1})\pi (g) \tag{19}

Applying this formula to π=πλ\pi =\pi_{\lambda}, we obtain that the element dλ⋅1d_{\lambda}\cdot 1 of K is not zero. We will see further on (VIII, p. 420, Corollary 2) that dμd_\mu divides the cardinal of G.

Let λ\lambda be an element of G^\widehat{G}. The mapping πλ\pi_{\lambda} from K[G] to EndK(Vλ)_K(V_{\lambda}) induces an isomorphism of (K[G](K[G], K[G])-bimodules from eλK[G]e_{\lambda}K[G] to EndK(Vλ)_K(V_{\lambda}). By Proposition 15 of VIII, p. 147, the K[G]-submodule eλK[G]e_{\lambda}K[G] is the isotypical component of K[G] of type λ\lambda. It is also the isotypical component of type λ∨\lambda^{\vee} for the right regular representation of G (VIII, p. 143, Proposition 11) as well as the isotypical component of type λ×λ∨\lambda \times \lambda^{\vee} for the biregular representation (VIII, p. 381, Proposition 4).

6. Schur Orthogonality Relations

We keep the notation of the previous subsection. Let λ\lambda be an element of G^\widehat{G}, and let uu and vv be elements of EndK(Vλ)_K(V_{\lambda}); by formula (10), we have

τ^(jλ(u)jλ(v))=dλ\widehat{\tau}(j_{\lambda}(u)j_{\lambda}(v)) =d_{\lambda} Tr(uvuv). Since τ=τ^∘F\tau =\widehat{\tau}\circ \mathscr{F}, we deduce from formulas (2) and (15) the relation

dλ2∣G∣−1∑g∈Gd^2_{\lambda}|G|^{-1}\sum_{g\in G} Tr(uπλ(g)u\pi_{\lambda}(g)) Tr(vπλ(g−1)v\pi_{\lambda}(g^{-1})) =dλ=d_{\lambda} Tr(uvuv).

We observed previously that dλ.1≠0d_{\lambda}.1\not= 0 in the field K; it follows that (20) ∣G∣−1∑g∈G|G|^{-1}\sum_{g\in G} Tr(uπλ(g)u\pi_{\lambda}(g)) Tr(vπλ(g−1)v\pi_{\lambda}(g^{-1})) =dλ−1=d^{-1}_{\lambda} Tr(uvuv).

Let us apply this relation to the case when uu and vv have rank ⩽1\leqslant 1; we obtain (21) ∣G∣−1∑g∈G⟨x∗,πλ(g)x⟩⟨y∗,πλ(g−1)y⟩=dλ−1⟨x∗,y⟩⟨y∗,x⟩|G|^{-1}\sum_{g\in G}\langle x^*, \pi_{\lambda}(g)x\rangle \langle y^*, \pi_{\lambda}(g^{-1})y\rangle =d^{-1}_{\lambda}\langle x^*, y\rangle \langle y^*, x\rangle

for x,yx,y in VλV_{\lambda} and x∗,y∗x^*, y^* in the dual Vλ∗V^*_{\lambda} of VλV_{\lambda}.

For every λ∈G^\lambda \in \widehat{G}, let (eλ,j)1⩽j⩽dλ(e_{\lambda ,j})_{1\leqslant j\leqslant d_{\lambda}} be a basis of VλV_{\lambda}; denote by (πijλ(g))(\pi^{\lambda}_{ij}(g)) the matrix of the endomorphism πλ(g)\pi_{\lambda}(g) of VλV_{\lambda} with respect to this basis. If we denote by (eλ,i∗)1⩽i⩽dλ(e^*_{\lambda ,i})_{1\leqslant i\leqslant d_{\lambda}} the basis of Vλ∗V^*_{\lambda} dual to (eλ,j)(e_{\lambda ,j}), then we have πijλ(g)=⟨eλ,i∗,πλ(g)eλ,j⟩\pi^{\lambda}_{ij}(g) =\langle e^*_{\lambda ,i}, \pi_{\lambda}(g)e_{\lambda ,j}\rangle, and therefore (22) ∣G∣−1∑g∈Gπijλ(g)πkℓλ(g−1)=dλ−1δiℓδjk|G|^{-1}\sum_{g\in G}\pi_{ij}^{\lambda}(g)\pi^{\lambda}_{k\ell}(g^{-1}) =d^{-1}_{\lambda}\delta_{i\ell}\delta_{jk}.

Now let λ\lambda and μ\mu be two distinct elements of G^\widehat{G}, and let u∈u\in EndK(Vλ)_K(V_{\lambda}) and v∈v\in EndK(Vμ)_K(V_\mu). Again by relation (15), we have (23) ∑g∈G\sum_{g\in G} Tr(uπλ(g)u\pi_{\lambda}(g)) Tr(vπμ(g−1)v\pi_\mu(g^{-1})) =0= 0.

As above, we deduce that (24) ∑g∈G⟨x∗,πλ(g)x⟩⟨y∗,πμ(g−1)y⟩=0\sum_{g\in G}\langle x^*, \pi_{\lambda}(g)x\rangle \langle y^*, \pi_\mu(g^{-1})y\rangle = 0

for x∈Vλ,x∗∈Vλ∗,y∈Vμx\in V_{\lambda},x^*\in V^*_{\lambda},y\in V_\mu, and y∗∈Vμ∗y^*\in V^*_\mu. We also have (25) ∑g∈Gπijλ(g)πkℓμ(g−1)=0\sum_{g\in G}\pi^{\lambda}_{ij}(g)\pi^\mu_{k\ell}(g^{-1}) = 0 for i,ji, j in [1,dλ][1, d_{\lambda}] and k,ℓk, \ell in [1,dμ][1, d_\mu].

Relations (20) through (25) are known as the Schur orthogonality relations.

Remark 00NS

We identify the algebra EndK(Vλ)_K(V_{\lambda}) with the matrix algebra Mdλ(K)\mathbf{M}_{d_{\lambda}}(K) via the basis (eλ,j)(e_{\lambda ,j}) of VλV_{\lambda}. The mapping F−1\mathscr{F}^{-1} is an isomorphism from the algebra ∏λMdλ(K)\prod_{\lambda}\mathbf{M}_{d_{\lambda}}(K) to the algebra K[G]. For μ∈G^\mu\in \widehat{G}, denote by EijμE^\mu_{ij} the element of ∏λMdλ(K)\prod_{\lambda}\mathbf{M}_{d_{\lambda}}(K) whose component of index μ\mu is the matrix unit EijE_{ij} of Mdμ(K)\mathbf{M}_{d_\mu}(K) (II, §10, No. 3, p. 341) and whose other components are zero; set uijμ=F−1(Eijμ)u^\mu_{ij}=\mathscr{F}^{-1}(E^\mu_{ij}). The family of elements uijλu^{\lambda}_{ij}, for λ∈G^,1⩽i⩽dλ\lambda \in \widehat{G}, 1\leqslant i\leqslant d_{\lambda}, 1⩽j⩽dλ1\leqslant j\leqslant d_{\lambda}, is a basis of the algebra K[G]; the multiplication table is

uijλukℓμ=δλμδjkuiℓλ(26) u^{\lambda}_{ij}u^\mu_{k\ell}=\delta_{\lambda \mu}\delta_{jk}u^{\lambda}_{i\ell} \tag{26}

Moreover, by formula (15), we have

uijλ=∣G∣−1dλg∑∈Gπjiλ(g−1)g(27) u^{\lambda}_{ij}=|G|^{-1}d_{\lambda g}\sum_{\in G}\pi^{\lambda}_{ji}(g^{-1})g \tag{27}

7. Orthogonality Relation for Characters

We keep the notation of Subsections 5 and 6. Recall that Z(K[G]) consists of the central functions.

We define a symmetric bilinear mapping from K[G]×K[G]K[G]\times K[G] to K by the formula

⟨f,f′⟩G=∣G∣−1∑g∈Gfgfg−1′(28) \langle f, f'\rangle_G=|G|^{-1}\sum_{g\in G}f_gf'_{g^{-1}} \tag{28}

for every f=∑fggf=\sum f_gg and f′=∑fg′gf'=\sum f_g'g belonging to K[G]. We have ⟨f,f′⟩G=\langle f, f'\rangle_G= ∣G∣−2τ(ff′)|G|^{-2}\tau (f f').

Proposition 4 (Orthogonality relation for characters) 00SG

For λ\lambda and μ\mu in G^\widehat{G}, we have ⟨χλ,χμ⟩G=δλμ\langle \chi_{\lambda}, \chi_\mu\rangle_G=\delta_{\lambda \mu}.

This is the specific case of relations (20) and (23) when the endomorphisms uu and vv are taken to be the identity.

Corollary 00NT

Let π\pi and π′\pi ' be finite-dimensional linear representations of G. In the field K, we have

(29) ⟨χπ,χπ′⟩G=\langle \chi_{\pi}, \chi_{\pi'}\rangle_G= (dimK_K HomG(π,π′)_G(\pi , \pi '))⋅1\cdot 1.

We first suppose that π\pi and π′\pi ' are simple representations. The vector space HomG(π,π′)_G(\pi , \pi ') has dimension 1 or 0 according to whether or not π\pi and π′\pi ' are isomorphic (Schur’s lemma, VIII, p. 47, Proposition 2). In this case, formula (29) follows from Proposition 4.

In the general case, the representation π\pi (resp. π′\pi ') is the direct sum of simple representations π1,...,πm\pi_1, . . . , \pi_m (resp. π1′,...,πn′\pi '_1, . . . , \pi '_n). The space HomG(π,π′)_G(\pi , \pi ') is isomorphic to the direct sum of the spaces HomG(πi,πj′)_G(\pi_i, \pi '_j) for 1⩽i⩽m1\leqslant i\leqslant m, 1⩽j⩽n1\leqslant j\leqslant n, and we have

χπ=χπ1+⋯+χπm,χπ′=χπ1′+⋯+χπn′ \chi_{\pi}=\chi_{\pi_1}+\cdots +\chi_{\pi_m},\chi_{\pi'}=\chi_{\pi'_1}+\cdots +\chi_{\pi'_n}

By linearity, the proof of formula (29) is reduced to the case of simple representations.

8. Central Functions on a Finite Group

Proposition 5 00NU

The family (χλ)λ∈G^(\chi_{\lambda})_{\lambda\in\widehat{G}} is a basis of the vector space of central functions. The number of classes of simple linear representations of G is equal to the number of conjugacy classes of G.

For a=∑g∈Gagg∈K[G]a=\sum_{g\in G}a_gg\in K[G], write a∨=∑g∈Gag−1ga^{\vee}=\sum_{g\in G}a_{g^{-1}}g; the mapping a↦a∨a\mapsto a^{\vee} is an involutive antiautomorphism of the algebra K[G]. By formula (17), we have eλ=∣G∣−1dλχλ∨e_{\lambda}=|G|^{-1}d_{\lambda}\chi^{\vee}_{\lambda} for every λ∈G^\lambda \in \widehat{G}. The proposition then follows from the fact that the family (eλ)λ∈G^(e_{\lambda})_{\lambda\in\widehat{G}} is a basis of the center of K[G].

Denote by C\mathscr{C} the set of conjugacy classes of G. Let C be an element of C\mathscr{C}; the image C−1C^{-1} of C by the mapping g↦g−1g\mapsto g^{-1} is a conjugacy class. The commutants of the elements of C are mutually conjugate subgroups of G; their cardinal d(C)d(C) satisfies

(30) Card(G) = Card(C) d(C)d(C).

In particular, we have d(C)⋅1≠0d(C)\cdot 1\not= 0 in the field K.

Let ff be a central function on G. For any conjugacy class C, denote by f(C)f(C) the common value of the f(x)f(x) for x∈Cx\in C. With this notation, the orthogonality relation for characters (VIII, p. 410, Proposition 4) can be written as

∑C∈Cχλ(C−1)χμ(C)d(C)−1=δλμ(31) \sum_{C\in\mathscr{C}}\chi_{\lambda}(C^{-1})\chi_\mu(C)d(C)^{-1}=\delta_{\lambda \mu} \tag{31}

for λ\lambda and μ\mu in G^\widehat{G}.

Denote by A the matrix of type G^×C\widehat{G}\times \mathscr{C} with entries χλ(C)\chi_{\lambda}(C) and by B the matrix of type C×G^\mathscr{C}\times \widehat{G} with entries χλ(C−1)d(C)−1\chi_{\lambda}(C^{-1})d(C)^{-1}. The sets G^\widehat{G} and C\mathscr{C} have the same cardinal (Proposition 5); relation (31) expresses the fact that the matrix product AB is the matrix unit of type G^×G^\widehat{G}\times \widehat{G}. By Proposition 11 of II, §10, No. 12, p. 360, the matrix product BA is the matrix unit of type C×C\mathscr{C}\times \mathscr{C}; in other words, we have the relation

∑λ∈G^χλ(C−1)χλ(C′)=d(C)δCC′(32) \sum_{\lambda\in\widehat{G}}\chi_{\lambda}(C^{-1})\chi_{\lambda}(C') =d(C)\delta_{CC'} \tag{32}

for C and C′C' in C\mathscr{C} (sometimes called the “second orthogonality relation for characters”).

Let H be a subgroup of G. Let us note that the integer Card(H) divides Card(G) and that ∣G∣|G| is not zero in K; we therefore have ∣H∣≠0|H| \not= 0 in K.

Denote by ResHG^G_H the linear mapping from Z(K[G]) to Z(K[H]) that sends a central function on G to its restriction to H. We have seen that if χπ\chi_{\pi} is the character of a finite-dimensional representation π\pi of G, then ResHG(χπ)^G_H(\chi_{\pi}) is the character of the representation ResHG(π)^G_H(\pi ) of H.

Proposition 6 00NV

Let ff be a central function on G and uu be a central function on H. We have

(33) ⟨\langleIndHG(u),f⟩G=⟨u^G_H(u), f\rangle_G=\langle u, ResHG(f)⟩H^G_H(f)\rangle_H.

The characters of the simple representations of G form a basis of Z(K[G]) (VIII, p. 411, Proposition 5), and the same holds for H. It therefore suffices to establish (33) in the case when ff is the character χπ\chi_{\pi} of a simple representation π\pi of G and uu is the character χσ\chi_{\sigma} of a simple representation σ\sigma of H. In this case, IndHG(u)^G_H(u) is the character of the representation IndHG(σ)^G_H(\sigma ) of G, and, by VIII, p. 410, Corollary, we have

⟨\langleIndHG(u),f⟩G=^G_H(u), f\rangle_G= (dimK_K HomG_G(IndHG(σ),π^G_H(\sigma ), \pi ))⋅1\cdot 1.

We show the relation

⟨u\langle u, ResHG(f)⟩H=^G_H(f)\rangle_H= (dimK_K HomH(σ_H(\sigma, ResHG(π)^G_H(\pi )))⋅1\cdot 1

likewise, and equality (33) follows by Frobenius reciprocity.

Proposition 7 00NW

Let ff be a mapping from G to K. The following assertions are equivalent:

(i) There are an element λ\lambda of G^\widehat{G} and an element aa of K such that f=aχλf=a\chi_{\lambda}.

(ii) For every pair (g,g′)(g, g') of elements of G, we have

f(g)f(g′)=∣G∣−1f(1)∑h∈Gf(hgh−1g′)(34) f(g)f(g') =|G|^{-1}f(1)\sum_{h\in G}f(hgh^{-1}g') \tag{34}

Let λ\lambda be an element of G^\widehat{G}; for every endomorphism uu of VλV_{\lambda}, set

u♮=∣G∣−1∑h∈Gπλ(h)uπλ(h−1)(35) u^{\natural}=|G|^{-1}\sum_{h\in G}\pi_{\lambda}(h)u \pi_{\lambda}(h^{-1}) \tag{35}

The endomorphism u♮u^{\natural} of VλV_{\lambda} is K[G]-linear. By Schur’s lemma (VIII, p. 47, Theorem 1),u♮,u^{\natural} is a homothety. Since uu and u♮u^{\natural} have the same trace, we therefore have

u♮=dλ−1u^{\natural}=d^{-1}_{\lambda} Tr(uu) 1Vλ1_{V_{\lambda}}.

Let uu and vv be endomorphisms of VλV_{\lambda}; it follows that

(36) Tr(uu) Tr(vv) =dλ=d_{\lambda} Tr(u♮vu^{\natural}v) =dλ∣G∣−1∑h∈G=d_{\lambda}|G|^{-1}\sum_{h\in G} Tr(πλ(h)uπλ(h−1)v\pi_{\lambda}(h)u \pi_{\lambda}(h^{-1})v).

Take u=πλ(g)u=\pi_{\lambda}(g) and v=πλ(g′)v=\pi_{\lambda}(g') in formula (36); relation (34) for f=χλf=\chi_{\lambda} follows.

Conversely, suppose that assertion (ii) holds. If we have f(1)=0f(1) = 0, then it follows that f(g)f(g′)=0f(g)f(g') = 0 for every pair (g,g′)(g, g') of elements of G, and therefore f=0f= 0. We can therefore assume f(1)≠0f(1)\not= 0. Taking g′=1g'= 1 in (34), we obtain the relation

f(g)=∣G∣−1∑h∈Gf(hgh−1) f(g) =|G|^{-1}\sum_{h\in G}f(hgh^{-1})

for every g∈Gg\in G, which implies that ff is a central function. By Proposition 5, there exists a family (aλ)(a_{\lambda}) of elements of K such that f=∑λ∈G^aλχλf=\sum_{\lambda\in\widehat{G}}a_{\lambda}\chi_{\lambda}. Let us replace ff with this expression in formula (34); taking into account that each of the characters χλ\chi_{\lambda} also satisfies this relation, we find

∑λ,μ∈G^aλaμχλ(g)χμ(g′)=∑λ∈G^aλdλ−1f(1)χλ(g)χλ(g′)(37) \sum_{\lambda ,\mu\in\widehat{G}}a_{\lambda}a_\mu\chi_{\lambda}(g)\chi_\mu(g') =\sum_{\lambda\in\widehat{G}}a_{\lambda}d^{-1}_{\lambda}f(1)\chi_{\lambda}(g)\chi_{\lambda}(g') \tag{37}

for g,g′∈Gg, g'\in G. This relation can also be written as

∑λ,μ(aλaμ−δλμaλdλ−1f(1))χλ(g)χμ(g′)=0(38) \sum_{\lambda ,\mu}(a_{\lambda}a_\mu-\delta_{\lambda \mu}a_{\lambda}d^{-1}_{\lambda}f(1))\chi_{\lambda}(g)\chi_\mu(g') = 0 \tag{38}

for g,g′∈Gg, g'\in G. Now, the functions χλ\chi_{\lambda}, for λ∈G^\lambda \in \widehat{G}, are linearly independent (Proposition 5 of VIII, p. 411); it follows that

aλaμ=δλμaλdλ−1f(1) a_{\lambda}a_\mu=\delta_{\lambda \mu}a_{\lambda}d^{-1}_{\lambda}f(1)

for λ,μ∈G^\lambda , \mu\in \widehat{G}. In particular, aλaμ=0a_{\lambda}a_\mu= 0 whenever λ≠μ\lambda \not=\mu. Consequently, there exists at most one element λ\lambda of G^\widehat{G} such that aλ≠0a_{\lambda}\not= 0, and we have f=aλχλf=a_{\lambda}\chi_{\lambda}, and therefore (i).

9. The Case of Abelian Groups

In this subsection, we assume that the group G is abelian.

By Schur’s lemma (VIII, p. 48, Corollary 1), every simple representation of G has dimension 1. Let (M,π)(M, \pi ) be such a representation and χ\chi be its character; for every g∈Gg\in G and x∈Mx\in M, we have π(g)(x)=χ(g)x\pi (g)(x) =\chi (g)x. Consequently, the character χ\chi is a homomorphism from G to the multiplicative group K∗K^* of K. Conversely, every homomorphism from G to K∗K^* is the character of a representation of G of degree 1. So the set G^\widehat{G} of classes of simple K[G]-modules can be identified with the set Hom(G,K∗,K^*) of homomorphisms from G to K∗K^*. We deduce from this an abelian group structure on G^\widehat{G}; the product in G^\widehat{G} corresponds to the tensor product of representations. The groups G and G^\widehat{G} have the same cardinal by Proposition 5 of VIII, p. 411. Every function on G is central, and G^\widehat{G} is a basis of the vector space of mappings from G to K (loc. cit.). Because of the orthogonality relation for characters, such a mapping ff has the following unique decomposition with respect to the basis G^:\widehat{G}:

f=∑χ∈G^⟨χ,f⟩Gχ(39) f=\sum_{\chi\in\widehat{G}}\langle \chi , f\rangle_G\chi \tag{39}

For mappings ff and f′f' from G to K, we have the relation

⟨f,f′⟩G=∑χ∈G^⟨χ,f⟩G⟨χ,f′⟩G(40) \langle f, f'\rangle_G=\sum_{\chi\in\widehat{G}}\langle \chi , f\rangle_G\langle \chi , f'\rangle_G \tag{40}

Let (V,π)(V, \pi ) be a linear representation of G. For any χ∈G^\chi \in \widehat{G}, denote by VχV^{\chi} the subspace of V consisting of the vectors vv such that π(g)(v)=χ(g)v\pi (g)(v) =\chi (g)v for every g∈Gg\in G. The space VχV^{\chi} is the isotypical component of type χ\chi of the K[G]-module V. The space V is the direct sum of the family (Vχ)χ∈G^(V^{\chi})_{\chi\in\widehat{G}}, and the projector pχp_{\chi} of V with image VχV^{\chi} associated with this decomposition is given by

pχ=∣G∣−1∑g∈Gχ(g−1)π(g)(41) p_{\chi}=|G|^{-1}\sum_{g\in G}\chi (g^{-1})\pi (g) \tag{41}

because of relation (19).

Remark 00NX

Let nn be the cardinal of the group G, and let μn(K)\mu_n(K) be the group of nn-th roots of unity in K. For every g∈Gg\in G, we have gn=1g^n= 1; consequently, G^\widehat{G} can be identified with the group Hom(G,μn(K), \mu_n(K)). The group μn(K)\mu_n(K) is cyclic of order nn (V, §11, No. 2, p. 78, Theorem 1). The group G^\widehat{G} is therefore isomorphic to the group D(G) = Hom(G,Q/Z,\mathbf{Q}/\mathbf{Z}). By VII, §4, No. 9, p. 26, Proposition 10, the group G^\widehat{G} is isomorphic to the group G, and the mapping that sends an element gg of G to the homomorphism χ↦χ(g)\chi \mapsto \chi (g) from G^\widehat{G} to K∗K^* is an isomorphism from G to G^^\widehat{\widehat{G}}.

10. Characters and Grothendieck Groups

Denote by θG\theta_G the K-algebra homomorphism from K⊗ZRK(G)K\otimes_{\mathbf{Z}}R_K(G) to ZK(G)\mathscr{Z}_K(G) that sends 1⊗[π]1\otimes [\pi ] to χπ\chi_{\pi} for every finite-dimensional representation π\pi.

Proposition 8 00NY

a) The homomorphism θG\theta_G is an isomorphism from K⊗ZRK(G)K\otimes_{\mathbf{Z}}R_K(G) to ZK(G)\mathscr{Z}_K(G).

b) Suppose that K has characteristic 0. Then θG\theta_G defines an isomorphism from RK(G)R_K(G) to the subring of ZK(G)\mathscr{Z}_K(G) consisting of the linear combinations of the characters χλ\chi_{\lambda} for λ\lambda in G^\widehat{G} with integer coefficients.

The family ([λ])λ∈G^([\lambda ])_{\lambda\in\widehat{G}} is a basis of the Z\mathbf{Z}-module RKR_K(G), and the family (χλ)λ∈G^(\chi_{\lambda})_{\lambda\in\widehat{G}} is a basis of the K-vector space ZK(G)\mathscr{Z}_K(G). Assertions a) and b) follow (VIII, p. 411, Proposition 5).

11. Dimension of Simple Representations

Denote the cardinal of the group G by nn. Let π\pi be a linear representation of G in a finite-dimensional K-vector space M. For every g∈Gg\in G, we have π(g)n=1M\pi (g)^n= 1_M, so the minimal polynomial of π(g)\pi (g) divides Tn−1T^n-1. Since n⋅1≠0n\cdot 1\not= 0 in K, this minimal polynomial is separable (V, §11, No. 2, p. 78), and since the field K is algebraically closed, the endomorphism π(g)\pi (g) of M is diagonalizable (VII, §5, No. 7, p. 40, Proposition 12). The eigenvalues of π(g)\pi (g) are nn-th roots of unity, and for every α∈K\alpha \in K, the geometric multiplicity of α\alpha as an eigenvalue of π(g)\pi (g) (VII, §5, No. 2, p. 30, Definition 1) is equal to the multiplicity of α\alpha as a root of the characteristic polynomial of π(g)\pi (g). Denote by On\mathscr{O}_n the subgroup of K generated by the set μn(K)\mu_n(K) of nn-th roots of unity; it is a finitely generated Z\mathbf{Z}-module and a subring of K. The character of π\pi takes its values in On\mathscr{O}_n.

Proposition 9 00NZ

Suppose that the field K has characteristic zero. Then the degree of every simple representation of G divides the cardinal of G.

Let (V,π)(V, \pi ) be a simple representation of G and χ\chi be its character. For every element aa of Z(K[G]), the endomorphism π(a)\pi (a) of V is a homothety (VIII, p. 47, Theorem 1); denote by φ(a)\varphi (a) the scalar such that π(a)=φ(a)V\pi (a) =\varphi (a)_V. The resulting mapping φ\varphi from Z(K[G]) to K is an algebra homomorphism. Let us take a=∑g∈Gχ(g−1)ga=\sum_{g\in G}\chi (g^{-1})g; by the remark of VIII, p. 408, we have φ(a)=\varphi (a) = (dim VV)−1∣G∣^{-1}|G|. On the other hand, aa belongs to the subring On[G]∩Z(K[G])\mathscr{O}_n[G]\cap Z(K[G]) of K[G], which is a finitely generated Z\mathbf{Z}-module (VII, §3, p. 15, Corollary). So the element φ(a)=\varphi (a) = (dim VV)−1∣G∣^{-1}|G| of K belongs to a subring of K that is a finitely generated Z\mathbf{Z}-module. We conclude using the following lemma.

Lemma 00O0

Let L be an extension of Q\mathbf{Q}. Let A be a subring of L. Suppose that A is a finitely generated Z\mathbf{Z}-module. We have A∩Q=ZA\cap \mathbf{Q}=\mathbf{Z}.

Since the Z\mathbf{Z}-module A∩QA\cap \mathbf{Q} is finitely generated, there exists a strictly positive integer N such that A∩QA\cap \mathbf{Q} is contained in 1NZ\frac{1}{N}\mathbf{Z}. Let xx be an element of Q−Z\mathbf{Q}-\mathbf{Z}; we write it as x=pqx=\frac{p}{q}, where pp and qq are mutually prime integers and q⩾2q\geqslant 2. We have qN⩾2N>N(q^N\geqslant 2^N>N (Set Theory, III, §3, No. 6, p. 165, Theorem 2), the integers pNp^N and qNq^N are mutually prime, and consequently xN∉1NZx^N\notin\frac{1}{N}\mathbf{Z}. It follows that xx does not belong to A. This concludes the proof of the lemma.

In the next subsection, we extend Proposition 9 to the case when we only assume that the characteristic of K does not divide the order of G.

12. Change of Base Field

We keep the notation of the previous subsection. Let K′K' be an algebraically closed field such that the element n⋅1n\cdot 1 of K′K' is not zero. The groups μn(K)\mu_n(K) and μn(K′)\mu_n(K') are cyclic of order nn (V, §11, No. 2, p. 78, Theorem 1). Choose an isomorphism φ\varphi from μn(K)\mu_n(K) to μn(K′)\mu_n(K'). Let π\pi be a linear representation of G in a finite-dimensional K-vector space, and let π′\pi ' be a linear representation of G in a finite-dimensional K′K'-vector space. We say that π\pi and π′\pi ' are related (through φ\varphi ) if for every g∈Gg\in G and ω∈μn\omega \in \mu_n(K), the multiplicity of ω\omega as an eigenvalue of π(g)\pi (g) is equal to the multiplicity of φ(ω)\varphi (\omega ) as an eigenvalue of π′(g)\pi '(g). When this is the case, π\pi and π′\pi ' have the same dimension, as can be seen by taking g=1g= 1.

Let π1\pi_1 and π2\pi_2 (resp. π1′\pi '_1 and π2′\pi '_2) be linear representations of G in finite-dimensional vector spaces over K (resp. K′K'). We have the following properties:

a) If π1\pi_1 is related to π1′\pi '_1 and π2′\pi '_2, then π1′\pi '_1 and π2′\pi '_2 are isomorphic.

b) If π1\pi_1 is related to π1′\pi '_1 and π2\pi_2 to π2′\pi '_2, then π1⊕π2\pi_1\oplus \pi_2 is related to π1′⊕π2′\pi '_1\oplus \pi '_2 and π1⊗π2\pi_1\otimes \pi_2 is related to π1′⊗π2′\pi '_1\otimes \pi '_2.

Assertion a) follows from Corollary 5 of VIII, p. 402, and assertion b) is clear.

Here, we denote by SK(G)\mathscr{S}_K(G) the set of classes of simple K[G]-modules, previously denoted by G^\widehat{G}, and we define SK′(G)\mathscr{S}_{K'}(G) likewise. The sets SK(G)\mathscr{S}_K(G) and SK′(G)\mathscr{S}_{K'}(G) are both finite, with cardinal the number of conjugacy classes (VIII, p. 411, Proposition 5).

Proposition 10 00O1

There exists a unique mapping φG\varphi_G from SK(G)\mathscr{S}_K(G) to SK′(G)\mathscr{S}_{K'}(G) such that λ\lambda and φG(λ)\varphi_G(\lambda ) are related through φ\varphi for every λ\lambda in SK(G)\mathscr{S}_K(G). Moreover, φG\varphi_G is bijective.

The uniqueness of φG\varphi_G follows from property a) above.

A) Suppose that the field K has characteristic 0.

The group μn(K)\mu_n(K) is cyclic (V, §11, No. 2, p. 78, Theorem 1); choose a generator ζ\zeta of this group. Consider the ring homomorphism ρ:Z[X]→On\rho :\mathbf{Z}[X]\rightarrow \mathscr{O}_n that sends X to ζ\zeta. It is surjective. The cyclotomic polynomial Φn(X)\Phi_n(X) is irreducible in Q[X]\mathbf{Q}[X] (V, §11, No. 5, p. 84, Theorem 2); it is therefore the minimal polynomial of ζ\zeta over Q\mathbf{Q}. The polynomial Φn\Phi_n is monic with integer coefficients (V, §11, No. 4, p. 81). Let P∈Z[X]P\in \mathbf{Z}[X] be a polynomial such that P(ζ)=0P(\zeta ) = 0; by Euclidean division of polynomials (IV, §1, No. 6, p. 10), there exist two polynomial Q and R in Z[X]\mathbf{Z}[X] such that P = QΦn+R_n+ R and deg(R) << deg(Φn_n). We have R(ζ)=0R(\zeta ) = 0, and therefore R = 0 because Φn\Phi_n is the minimal polynomial of ζ\zeta. Consequently, the kernel of ρ\rho is the ideal ΦnZ[X]\Phi_n\mathbf{Z}[X] of Z[X]\mathbf{Z}[X], and ρ\rho induces a ring isomorphism from Z[X]/ΦnZ[X]\mathbf{Z}[X]/\Phi_n\mathbf{Z}[X] to On\mathscr{O}_n.

Set ζ′=φ(ζ)\zeta '=\varphi (\zeta ); it is a primitive nn-th root of unity in K′K', and we therefore have Φn(ζ′)=0\Phi_n(\zeta ') = 0 (V, §11, No. 5, p. 83, Lemma 3). Consequently, there exists a homomorphism φ0\varphi_0 from the ring On\mathscr{O}_n to the field K′K' that transforms ζ\zeta into ζ′\zeta '; it extends the mapping φ\varphi from μn(K)\mu_n(K) to μn(K′)\mu_n(K'). Let O\mathscr{O} be the subring of K consisting of the elements anr\frac{a}{n^r} with a∈Ona\in \mathscr{O}_n and r∈Nr\in \mathbf{N}. Since n⋅1n\cdot 1 is invertible in K′K', the homomorphism φ0\varphi_0 extends to a homomorphism φ1\varphi_1 from O\mathscr{O} to K′K'.

We identify the algebra O[G]\mathscr{O}[G] of the group G over O\mathscr{O} with a subring of the algebra K[G] and define a ring homomorphism Φ from O[G]\mathscr{O}[G] to K′[G]K'[G] by the formula

Φ(∑g∈Gagg)=∑g∈Gφ1(ag)g(42) \Phi (\sum_{g\in G}a_gg)=\sum_{g\in G}\varphi_1(a_g)g \tag{42}

Denote by C\mathscr{C} the set of conjugacy classes of G. For C in C\mathscr{C}, denote by uCu_C the element ∑g∈Cg\sum_{g\in C}g of O[G]\mathscr{O}[G]; the family (uC)C∈C(u_C)_{C\in\mathscr{C}} is a basis over O\mathscr{O} of the center Z(O[G])Z(\mathscr{O}[G]) of the algebra O[G]\mathscr{O}[G] (VIII, p. 398). For any element λ\lambda of SK\mathscr{S}_K(G), denote by χλ\chi_{\lambda} its character, by dλd_{\lambda} its dimension, and by eλe_{\lambda} the element of K[G] defined by

eλ=∣G∣−1dλ∑g∈Gχλ(g−1)g(43) e_{\lambda}=|G|^{-1}d_{\lambda}\sum_{g\in G}\chi_{\lambda}(g^{-1})g \tag{43}

The family (eλ)(e_{\lambda}) is a basis over K of the center Z(K[G]) of the ring K[G] (VIII, p. 408). We have

eλ=∑C∈Cαλ,CuC(44) e_{\lambda}=\sum_{C\in\mathscr{C}}\alpha_{\lambda ,C}u_C \tag{44}

with

αλ,C=∣G∣−1dλχλ(C−1)(45) \alpha_{\lambda ,C}=|G|^{-1}d_{\lambda}\chi_{\lambda}(C^{-1}) \tag{45}

For C∈CC\in \mathscr{C} and λ∈SK\lambda \in \mathscr{S}_K(G), set βC,λ=∣G∣dλ−1d(C)−1χλ(C)\beta_{C,\lambda}=|G|d^{-1}_{\lambda}d(C)^{-1}\chi_{\lambda}(C). The entries of the matrix (αλ,C)(\alpha_{\lambda ,C}) are in O\mathscr{O}, and it follows from formula (31) of VIII, p. 411 that its inverse matrix is the matrix (βC,λ)(\beta_{C,\lambda}), whose entries are also in O\mathscr{O} by Proposition 9 of VIII, p. 415. Consequently, the family (eλ)(e_{\lambda}) is a basis of the O\mathscr{O}-module Z(O[G])Z(\mathscr{O}[G]).

The elements Φ(uC)=∑g∈Cg\Phi (u_C) =\sum_{g\in C}g of K′[G]K'[G] form a basis over K′K' of the center Z(K′[G])Z(K'[G]) of the ring K′[G]K'[G]. We have

Φ(eλ)=∑C∈Cφ1(αλ,C)Φ(uC)(46) \Phi (e_{\lambda}) =\sum_{C\in\mathscr{C}}\varphi_1(\alpha_{\lambda ,C}) \Phi (u_C) \tag{46}

and the matrix with entries φ1(αλ,C)\varphi_1(\alpha_{\lambda ,C}) is invertible. The family of the Φ(eλ)\Phi (e_{\lambda}) is therefore a basis of Z(K′[G])Z(K'[G]). The family (eλ)(e_{\lambda}) is a partition of the idempotent 1 in Z(K[G]) (VIII, p. 146 and p. 408); in other words, we have

∑λeλ=1,eλ2=eλ,eλeμ=0\sum_{\lambda}e_{\lambda}= 1,e^2_{\lambda}=e_{\lambda},e_{\lambda}e_\mu= 0 if λ≠μ\lambda \not=\mu.

It follows that the family of the Φ(eλ)\Phi (e_{\lambda}) is a partition of the idempotent 1 in Z(K′[G])Z(K'[G]); since this family is a basis of Z(K′[G])Z(K'[G]) over K′K', its elements are the indecomposable idempotents in Z(K′[G])Z(K'[G]) (VIII, p. 148, Remark 4).

For λ′\lambda ' in SK′\mathscr{S}_{K'}(G), define χλ′,dλ′\chi_{\lambda'},d_{\lambda'}, and eλ′e_{\lambda'} as above. By VIII, p. 408, the elements eλ′e_{\lambda'} are the indecomposable idempotents in Z(K′[G])Z(K'[G]). Hence there exists a bijection φG\varphi_G from SK(G)\mathscr{S}_K(G) to SK′(G)\mathscr{S}_{K'}(G) such that Φ(eλ)=eφG(λ)\Phi (e_{\lambda}) =e_{\varphi_G(\lambda)} for every λ\lambda in SK(G)\mathscr{S}_K(G).

Let λ\lambda in SK(G)\mathscr{S}_K(G); set λ′=φG(λ)\lambda '=\varphi_G(\lambda ). Let (Vλ,πλ)(V_{\lambda}, \pi_{\lambda}) (resp. (Vλ′,πλ′)(V_{\lambda'}, \pi_{\lambda'})) be a linear representation of G whose associated K[G]-module (resp. K′K'[G]-module) has class λ\lambda (resp. λ′\lambda '). Let us prove that λ\lambda and λ′\lambda ' are related. Let gg be an element of G. Let δ(T)\delta (T) be the determinant of the endomorphism 1+Tπλ(g)1 + T\pi_{\lambda}(g) of the K[T]-module K[T]⊗KVλK[T]\otimes_KV_{\lambda}. Let ω1,...,ωdλ\omega_1, . . . , \omega_{d_{\lambda}} be the eigenvalues of πλ(g)\pi_{\lambda}(g); we have

δ(T)=(1+Tω1)⋯(1+Tωdλ) \delta (T) = (1 + T\omega_1)\cdots (1 + T\omega_{d_{\lambda}})

We define δ′(T)\delta '(T) likewise, and we denote the eigenvalues of πλ′(g)\pi_{\lambda'}(g) by ω1′,...,ωdλ′′\omega '_1, . . . , \omega '_{d_{\lambda'}}. The O\mathscr{O}-module O[G]\mathscr{O}[G] is free with basis G, and the K-vector space K[G] has basis G. Denote by Δ(T) the determinant of multiplication by 1+eλgT1 +e_{\lambda}gT in the O\mathscr{O}[T]-module O[T]⊗OO[G]\mathscr{O}[T]\otimes_{\mathscr{O}}\mathscr{O}[G]. It is also the determinant of multiplication by 1+eλgT1 +e_{\lambda}gT in the K[T]-module K[T]⊗KK[G]K[T]\otimes_KK[G]. Let φ‾1\overline{\varphi}_1 be the homomorphism from O[T]\mathscr{O}[T] to K′[T]K'[T] that extends φ1\varphi_1 and sends T to T. Since G is a basis of the K′K'-vector space K′[G]K'[G], the polynomial φ‾1(Δ(T))\overline{\varphi}_1(\Delta (T)) is equal to the determinant Δ′(T)\Delta '(T) of multiplication by 1+eλ′gT1 +e_{\lambda'}gT in the K′K'-vector space K′[G]K'[G].

The algebra K[G] is the direct sum of its simple components eμK[G]e_\mu K[G] for μ\mu running through SK(G)\mathscr{S}_K(G). For μ\mu different from λ\lambda, the element eλge_{\lambda}g annihilates eμK[G]e_\mu K[G]. Moreover, multiplication by eλge_{\lambda}g coincides with multiplication by gg in eλK[G]e_{\lambda}K[G]. In view of VIII, p. 409 and Example 6 of VIII, p. 400, the representation of G in eλK[G]e_{\lambda}K[G] is the direct sum of dλd_{\lambda} representations of class λ\lambda. We consequently have Δ(T)=δ(T)dλ\Delta (T) =\delta (T)^{d_{\lambda}}.

Analogously, we have Δ′(T)=δ′(T)dλ′\Delta '(T) =\delta '(T)^{d_{\lambda'}}.

From the relation Δ′(T)=φ‾1\Delta '(T) =\overline{\varphi}_1(Δ(T)), we deduce first that dλ2=dλ′2d^2_{\lambda}=d^2_{\lambda'}, and therefore dλ=dλ′d_{\lambda}=d_{\lambda'}, and then that the sequence φ(ω1),...,φ(ωdλ)\varphi (\omega_1), . . . , \varphi (\omega_{d_{\lambda}}) can be deduced from the sequence (ω1′,...,ωdλ′′′)(\omega '_1, . . . , \omega '_{d'_{\lambda'}}) by a permutation of the set of indices.

Since this is true for every element gg of G, the representations λ\lambda and λ′\lambda ' are related. We have therefore proved Proposition 10 when the field K has characteristic 0.

B) General case.

Let L be an algebraically closed field of characteristic 0 (for example, an algebraic closure of Q\mathbf{Q}). Denote by SL(G)\mathscr{S}_L(G) the set of classes of simple L[G]-modules. Choose an isomorphism η\eta from the group μn(L)\mu_n(L) to the group μn\mu_n(K), and set η′=φ∘η\eta '=\varphi \circ \eta. By part A) of the proof, there exist bijections

ηG:SL(G)→SK(G),ηG′:SL(G)→SK′(G) \eta_G:\mathscr{S}_L(G)\rightarrow \mathscr{S}_K(G),\eta '_G:\mathscr{S}_L(G)\rightarrow \mathscr{S}_{K'}(G)

with the following property: for every λ\lambda in SL\mathscr{S}_L(G), the representations λ\lambda and ηG(λ)\eta_G(\lambda ) are related through η\eta, and the representations λ\lambda and ηG′(λ)\eta '_G(\lambda ) are related through η′\eta '. The bijection φG=ηG′∘ηG−1\varphi_G=\eta '_G\circ \eta_G^{-1} has the desired properties.

The bijection φG\varphi_G from IK(G)\mathscr{I}_K(G) to IK′(G)\mathscr{I}_{K'}(G) extends to an isomorphism, also denoted by φG\varphi_G, from the Grothendieck group RK(G)R_K(G) to the group RK′(G)R_{K'}(G).

Remark 1 00O2

Suppose that K′K' is an extension of KK and that the isomorphism φ\varphi is the mapping ξ↦ξ⋅1\xi \mapsto \xi \cdot 1; then the mapping φG\varphi_G is given by extension of scalars from KK to K′K'.

Corollary 1 00O3

The mapping φG\varphi_G is a ring isomorphism from RK(G)R_K(G) to RK′(G)R_{K'}(G). For every finite-dimensional representation π\pi of GG in a KK-vector space, we have φG([π])=[π′]\varphi_G([\pi]) = [\pi'], where π′\pi' is a representation related to π\pi through φ\varphi*.

This follows from the semisimplicity of the representations of GG and property b) of VIII, p. 416.

Corollary 2 00O4

The dimension of each simple representation of GG divides the order of GG*.

This follows from Proposition 10 and Proposition 9 of VIII, p. 415.

Remark 2 00O5

Suppose that the group GG is abelian. We saw in the remark of VIII, p. 414 that IK(G)\mathscr{I}_K(G) can be identified with the set Hom⁡(G,μn(K))\operatorname{Hom}(G, \mu_n(K)). Likewise, IK′(G)\mathscr{I}_{K'}(G) can be identified with Hom⁡(G,μn(K′))\operatorname{Hom}(G, \mu_n(K')). With these identifications, the bijection φG\varphi_G is simply the mapping χ↦φ∘χ\chi \mapsto \varphi \circ \chi.

Remark 3 00O6

Let π1\pi_1 and π2\pi_2 be linear representations of GG in finite-dimensional vector spaces over KK. For i=1,2i = 1, 2, let πi′\pi_i' be a representation related to πi\pi_i through φ\varphi. We have

dim⁡KHom⁡K(π1,π2)=dim⁡K′Hom⁡K′(π1′,π2′). \dim_K \operatorname{Hom}_K(\pi_1, \pi_2) = \dim_{K'} \operatorname{Hom}_{K'}(\pi_1', \pi_2') .
The proof follows that of Corollary VIII, p. 410, by reducing to the case when the πi\pi_i (and therefore the πi′\pi_i') are simple.

Remark 4 00O7

Let HH be a subgroup of GG of cardinal mm. The isomorphism φ\varphi restricts to an isomorphism from μm(K)\mu_m(K) to μm(K′)\mu_m(K') and, consequently, a ring isomorphism φH\varphi_H from RK(H)R_K(H) to RK′(H)R_{K'}(H). The following diagrams commute:

RK(G)→Res⁡HGRK(H)↓φG↓φHRK′(G)→Res⁡HG′RK′(H), \begin{array}{ccc} R_K(G) & \xrightarrow{\operatorname{Res}_H^G} & R_K(H) \\ \downarrow \varphi_G & & \downarrow \varphi_H \\ R_{K'}(G) & \xrightarrow{\operatorname{Res}_H^{G'}} & R_{K'}(H) , \end{array}
RK(H)→Ind⁡HGRK(G)↓φH↓φGRK′(H)→Ind⁡HG′RK′(G). \begin{array}{ccc} R_K(H) & \xrightarrow{\operatorname{Ind}_H^G} & R_K(G) \\ \downarrow \varphi_H & & \downarrow \varphi_G \\ R_{K'}(H) & \xrightarrow{\operatorname{Ind}_H^{G'}} & R_{K'}(G) . \end{array}

The commutativity of the first diagram is obvious, and that of the second follows from it using Frobenius reciprocity and formula (47).

Remark 5 00O8

Suppose that G is the product G′×G′′G'\times G'' of two finite groups. We define isomorphisms φG′\varphi_{G'} and φG′′\varphi_{G''} as in the previous example.

We have a commutative diagram

RK(G′)⊗ZRK(G′′)κR_K(G')\otimes_{\mathbf{Z}}R_K(G'')^{\kappa} // RK(G)R_K(G)

φG′⊗φG′′φG\varphi_{G'}\otimes \varphi_{G''}\varphi_G

RK′(G′)⊗ZRK′(G′′)κ′R_{K'}(G')\otimes_{\mathbf{Z}}R_{K'}(G'')^{\kappa'} // RK′(G)R_{K'}(G),

where the isomorphisms κ\kappa and κ′\kappa ' are those defined in VIII, p. 406.

\*13. Complex Linear Representations

In this subsection, we assume that K is the field C\mathbf{C} of complex numbers.

Let (M,π)(M, \pi ) be a linear representation of G. We say that a Hermitian form Φ on M is invariant under G if we have

Φ(π(g)x,π(g)x′)=Φ(x,x′)(48) \Phi (\pi (g)x, \pi (g)x') = \Phi (x, x') \tag{48}

for all x,x′∈Mx, x'\in M and every g∈Gg\in G. This also means that for every g∈Gg\in G, the automorphism π(g)\pi (g) of M is unitary with respect to Φ.

Proposition 11 00O9

Let (M,π)(M, \pi ) be a finite-dimensional linear representation of G.

a) There exists on M a Hermitian form that is positive, separating, and invariant under G.

b) Suppose that the representation π\pi is simple. If Φ and Ψ are nonzero Hermitian forms on M that are invariant under G, then there exists a real number aa such that Ψ=aΦ\Psi =a\Phi.

Choose a separating positive Hermitian form on the vector space M, and denote it by Φ0\Phi_0. We define a separating positive Hermitian form Φ that is invariant under G by setting

Φ(x,x′)=∑g∈GΦ0(π(g)x,π(g)x′)(49) \Phi (x, x') =\sum_{g\in G}\Phi_0(\pi (g)x, \pi (g)x') \tag{49}

for x,x′∈Mx, x'\in M.

Let Ψ be a Hermitian form on M; there exists a unique endomorphism A of M such that Ψ(x,x′)=Φ(x,Ax′)\Psi (x, x') = \Phi (x,Ax') for x,x′x, x' in M. If, moreover, Ψ is invariant under G, then the endomorphism A commutes with the automorphism π(g)\pi (g) for g∈Gg\in G. If the representation π\pi is simple, then by Schur’s lemma (VIII, p. 47, Theorem 1), A is a homothety and there consequently exists a complex number aa such that Ψ=aΦ\Psi =a\Phi. Since Φ and Ψ are Hermitian and Φ is nonzero, aa is a real number. The proposition follows.

We endow the vector space C[G]\mathbf{C}[G] of complex functions on G with the Hilbert space structure whose inner product is given by

⟨f∣f′⟩G=∣G∣−1∑g∈Gf(g)‾f′(g)(50) \langle f|f'\rangle_G=|G|^{-1}\sum_{g\in G}\overline{f(g)}f'(g) \tag{50}

For any function f∈C[G]f\in \mathbf{C}[G], we denote by f∗f^* the function defined by

f∗(g)=f(g−1)‾(51) f^*(g) =\overline{f(g^{-1})} \tag{51}

for g∈Gg\in G; the mapping f↦f∗f\mapsto f^* is a semilinear involution of C[G]\mathbf{C}[G]. We also have

⟨f∣f′⟩G=⟨f∗,f′⟩G(52) \langle f|f'\rangle_G=\langle f^*, f'\rangle_G \tag{52}

for f,f′∈C[G]f, f'\in \mathbf{C}[G], with the notation of formula (28) of VIII, p. 410. We therefore have

⟨f∣f′⟩G=∣G∣−2τ(f∗f′)(53) \langle f|f'\rangle_G=|G|^{-2}\tau (f^*f') \tag{53}

Let (M,π)(M, \pi ) be a finite-dimensional linear representation of G. We endow the vector space M with the structure of a Hilbert space for which the endomorphisms π(g)\pi (g) are unitary (Proposition 11). If we denote by A∗A^* the adjoint of a endomorphism A of M for this structure, then we have Tr(A∗A^*) == Tr(A). For every g∈Gg\in G, we have π(g−1)=π(g)∗\pi (g^{-1}) =\pi (g)^*, and therefore χπ(g−1)=χπ(g)\chi_{\pi}(g^{-1}) =\chi_{\pi}(g); in other words, we have χπ=χπ∗\chi_{\pi}=\chi^*_{\pi}. The orthogonality relation for characters (VIII, p. 410, Proposition 4) then has the form

⟨χλ∣χμ⟩G=δλμ(54) \langle \chi_{\lambda}|\chi_\mu\rangle_G=\delta_{\lambda \mu} \tag{54}

for λ,μ∈G^\lambda , \mu\in \widehat{G}. It expresses the fact that the family of the characters (χλ)λ∈G^(\chi_{\lambda})_{\lambda\in\widehat{G}} of the simple representations of G is an orthonormal basis of the Hilbert space Z(C[G])Z(\mathbf{C}[G]) of central functions.

Let π\pi and π′\pi ' be finite-dimensional linear representations of G. We have the relation ⟨χπ∣χπ′⟩G=\langle \chi_{\pi}|\chi_{\pi'}\rangle_G= dimC_{\mathbf{C}} HomG(π,π′)_G(\pi , \pi ') (VIII, p. 410, Corollary). The representation π\pi is irreducible if and only if ⟨χπ∣χπ⟩G=1\langle \chi_{\pi}|\chi_{\pi}\rangle_G= 1.

For every element λ\lambda of G^\widehat{G}, we endow the vector space VλV_{\lambda} with the structure of a Hilbert space for which the automorphisms πλ(g)\pi_{\lambda}(g) are unitary. We denote by ⟨v∣v′⟩λ\langle v|v'\rangle_{\lambda} the inner product of two elements v,v′v, v' of VλV_{\lambda} and by u∗u^* the adjoint of an endomorphism uu of VλV_{\lambda}. Let A=(Aλ)λ∈G^A = (A_{\lambda})_{\lambda\in\widehat{G}} and A′=(Aλ′)λ∈G^A'= (A'_{\lambda})_{\lambda\in\widehat{G}} be elements of F(G^)F(\widehat{G}). We write A∗=(Aλ∗)λ∈G^A^*= (A^*_{\lambda})_{\lambda\in\widehat{G}}. We have F(a∗)=(F(a))∗\mathscr{F}(a^*) = (\mathscr{F}(a))^* for every element aa of C[G]\mathbf{C}[G]. Set (55) ⟨A∣A′⟩G^=∣G∣−2τ^(A∗A′)\langle A|A'\rangle_{\widehat{G}}=|G|^{-2}\widehat{\tau}(A^*A'). By formula (10) of VIII, p. 407, we have

(56) ⟨A∣A′⟩G^=1∣G∣2∑λ∈G^dλ\langle A|A'\rangle_{\widehat{G}}=\frac{1}{|G|^2}\sum_{\lambda\in\widehat{G}}d_{\lambda} Tr(Aλ∗^*_{\lambda}Aλ′'_{\lambda}).

Since τ^∘F=τ\widehat{\tau}\circ \mathscr{F}=\tau, formulas (53) and (55) imply that the mapping F\mathscr{F} is an isomorphism of Hilbert spaces from C[G]\mathbf{C}[G] to F(G^)F(\widehat{G}).

The Schur orthogonality relations (VIII, p. 410) can be reformulated using Hilbertian inner products. Relations (21) and (24) then give the following assertions. For λ∈G^\lambda \in \widehat{G} and x,x′,y,y′x, x', y, y' in VλV_{\lambda}, we have (57) ∣G∣−1∑g∈G⟨x∣πλ(g)x′⟩‾λ⟨y∣πλ(g)y′⟩λ=dλ−1⟨x∣y⟩‾λ⟨x′∣y′⟩λ|G|^{-1}\sum_{g\in G}\overline{\langle x|\pi_{\lambda}(g)x'\rangle}_{\lambda}\langle y|\pi_{\lambda}(g)y'\rangle_{\lambda}=d^{-1}_{\lambda}\overline{\langle x|y\rangle}_{\lambda}\langle x'|y'\rangle_{\lambda}.

If λ\lambda and μ\mu are two distinct elements of G^\widehat{G}, then for x,x′x, x' in VλV_{\lambda} and y,y′y, y' in VμV_\mu, we have (58) ∑g∈G⟨x∣πλ(g)x′⟩‾λ⟨y∣πμ(g)y′⟩μ=0\sum_{g\in G}\overline{\langle x|\pi_{\lambda}(g)x'\rangle}_{\lambda}\langle y|\pi_\mu(g)y'\rangle_\mu= 0.

For every λ∈G^\lambda \in \widehat{G}, we choose an orthonormal basis (eλ,i)1⩽i⩽dλ(e_{\lambda ,i})_{1\leqslant i\leqslant d_{\lambda}} of VλV_{\lambda}. For any g∈Gg\in G, we denote by (πijλ(g))(\pi^{\lambda}_{ij}(g)) the matrix of the endomorphism πλ(g)\pi_{\lambda}(g) of VλV_{\lambda} with respect to this basis; we have (59) πijλ(g)=⟨eλ,i∣πλ(g)eλ,j⟩λ\pi^{\lambda}_{ij}(g) =\langle e_{\lambda ,i}|\pi_{\lambda}(g)e_{\lambda ,j}\rangle_{\lambda}.

Since the endomorphism πλ(g)\pi_{\lambda}(g) is unitary, its inverse is equal to πλ(g)∗\pi_{\lambda}(g)^*, so that (60) πijλ(g)‾=πjiλ(g−1)\overline{\pi^{\lambda}_{ij}(g)}=\pi^{\lambda}_{ji}(g^{-1}). It then follows from formulas (22) of VIII, p. 409 and (25), p. 409 that the functions (dλ)1/2πijλ(d_{\lambda})^{1/2}\pi_{ij}^{\lambda}, for λ∈G^,1⩽i⩽dλ,1⩽j⩽dλ\lambda \in \widehat{G},1\leqslant i\leqslant d_{\lambda},1\leqslant j\leqslant d_{\lambda}, form an orthonormal basis of the Hilbert space C[G].∗\mathbf{C}[G].*

Exercises

See the exercises for § 21.

[^1]: The elements of HomG(π,π′)_G(\pi , \pi ') are sometimes called intertwining operators of π\pi and π′\pi '.

Transcribed from the printed edition, 2023, Springer Nature. Not translated.