PMATH 745: REPRESENTATION THEORY OF FINITE GROUPS

Transcription

1 PMATH 745: REPRESENTATION THEORY OF FINITE GROUPS HEESUNG YANG 1. September 8: Review and quick definitions In this section we review classical linear algebra and introduce the notion of representation. N.B. Unless otherwise specified, all the vector spaces will be over C. Definition 1. Let V be a C-vector space. (i) A linear operator ϕ on V is a linear map ϕ : V V. (ii) We define GL(V ) to be the set of invertible linear operators on V. (iii) We define GL n (C) be the set of invertible n n matrices over C. Proposition 2. Let V be a C-vector space. (i) If dim C V = n, then V = C n. (ii) If B = {b 1, b 2,..., b n } is a basis for V and v V, then there exist a unique (z 1,..., z n ) C n so that v = z 1 b 1 + z 2 b z n b n. (iii) If f B : V C n is defined by f B (v) = [v] B = (z 1, z 2,..., z n ), then f B is a vector space isomorphism, and every isomorphism V = C n is of this form. Suppose that h : V = W is an isomorphism (as vector spaces). Then h lifts in a natural way to an isomorphism h : GL(V ) GL(W ). Note that, if f GL(V ) and g GL(W ) and h (f) = g, then the diagram f V h W V h 1 must commute, hence h (f) = h f h 1. Proposition 3. h is a group isomorphism. Proof. To prove that h is bijective, it suffices to find an inverse for each element g GL(W ). The commutative diagram above implies that h 1 g h GL(V ) is the inverse: indeed, we have h (h 1 g h) = h (h 1 g h) h 1 = g. To show that h is a group homomorphism, note that h (f) h (g) = (h 1 f h) (h 1 g h) = h 1 (f g) h = h (f g), as required. g W Date: 17 April

2 Corollary 1. If dim(v ) = n, then GL(V ) = GL(C n ). Proof. Apply Proposition 3 and the fact that V = C n. Recall the correspondence between linear operators on V and n-by-n matrices over C (classical linear algebra fact!). Given an n n matrix M M n (C), define L M : C n C n be the linear operator represented by the matrix M (with respect to the standard basis, WLOG), i.e., L M (v) = Mv. Recall also that L M is linear and invertible if and only if M is an invertible matrix. Note that, to show that GL n (C) = V for any V with dim C V = n, we need to have GL n (C) = GL(C n ). And this relationship between the linear operators and the matrices gives us the group isomorphism we are looking for. Proposition 4. Let L : GL n (C) GL(C n ) be the map defined as M L M. Then L is a group isomorphism from GL n (C) to GL(C n ). Proof. First we show that L is bijective by explicitly constructing its inverse, L 1. E = {e 1, e 2,..., e n } be the standard basis of C n. Given f GL(C n ), we define M f := f(e 1 ) f(e 2 ) f(e 3 ) f(e n ), which is the matrix representing f, with respect to the standard basis. Note that M f is unique (up to a basis), since M f s action is determined entirely by its action on each basis element. Define L 1 : GL(C n ) GL n (C) as f M f. Now we prove that L is a group homomorphism. This follows from the fact that L MN = L M L N, i.e., L(MN) = L MN = L M L N = L(M) L(N), as required. Corollary 2. If dim C V = n, then GL n (C) = GL(V ). Proof. Follows from Propositions 3 and 4. Definition 5. Let G be a group. (i) A linear representation of G is a pair (V, ρ) where: (a) V is a vector space (over C); and (b) ρ is a group homomorphism from G to GL(V ). (ii) The degree of a representation is dim(v ). For the sake of notational cleanness, we shall write ρ(g) as ρ g for each g G. Observe that Definition 5 formulates a linear representation from the perspective of an action of a group on the space of linear operators. The following alternate definition of linear representations focuses on the behaviour of the map ρ itself: Definition 6. A linear representation of G is a vector space over C together with a family of linear bijections (ρ g = g G) such that if g 1 g 2 = g, then ρ g1 ρ g2 = ρ g. Proposition 7. Definitions 5 and 6 are equivalent. Proof. (6 5) This is immediate, since g 1 g 2 = g ρ g1 ρ g2 = ρ g implies that ρ is a group homomorphism. (5 6) Since ρ is a group homomorphism, we have ρ(g 1 g 2 ) = ρ(g 1 ) ρ(g 2 ), as desired. That each ρ g is a linear bijection follows from the fact that ρ g is a invertible linear operator for all g G. 2 Let

3 2.1. Examples of representations. 2. September 9 Example 1. Consider C 6, cyclic group of order 6. In this example we will construct a representation of degree 2. Start off with a linear map L : C 2 C 2 which maps ( ( 1 0) 0 ( 1) and 0 ( 1) w ) w, where w denotes a primitive cubic root of unity. Thus the matrix representation 2 of L (call it M L ) is ( ) 0 w M L := 1 w 2. One can calculate that M 3 L = I, hence L3 = id C 2. Hence, the map ρ satisfying the following forms a C 6 -representation (C 2, ρ) (one can verify that ρ is indeed a group homomorphism): 0, 3 id C 2 1, 4 L 2, 5 L 2. Alternatively, one can write this representation in the following way: (C 2, (id, L, L 2, id, L, L 2 )). Example 2. In this example, we consider a permutation group S 3. For each π S 3, define ρ π : C 3 C 3 by ρ π (z 1, z 2, z 3 ) := (z π 1 (1), z π 1 (2), z π 1 (3)). Observe that ρ π is indeed a linear map: ρ π can be expressed using a matrix, namely the identity matrix with columns permuted by π 1. For instance, if π = (123), then we have ρ π (z 1, z 2, z 3 ) = (z 3, z 1, z 2 ) and z 1 z 2 z 3 = z 3 z 1 z 2 which is what we would expect. As we would expect from the matrices, each ρ π is invertible. Now it remains to show that (C 3, (ρ π : π S 3 )) is a representation of S 3. That is, we need to show that ρ is a group homomorphism from S 3 to GL(C 3 ). It is helpful to think of each element of z C 3 as a map z : {1, 2, 3} C (for instance, (1, 2, 3) (z 1, z 2, z 3 )). We claim that ρ π z = z π 1. Indeed, in the case of π = (123), we have ρ π (z)(1) = ρ π (z 1, z 2, z 3 )(1) = (z 3, z 1, z 2 )(1) = z 3 and z π 1 (1) = (z 1, z 2, z 3 )(3) = z 3, as desired. Do this for other elements and each π to verify the claim. Now for any λ, π S 3 and z C 3, we need to prove that ρ λ π = ρ λ ρ π, which is enough to show that ρ is a group homomorphism. Indeed, we have ρ λ π (z) = z (λ π) 1 = z π 1 λ 1 = (z π 1 ) λ 1 = ρ π (z) λ 1 = ρ λ (ρ π (z)) = (ρ λ ρ π )(z), as required. Definition 8. Suppose (V, ρ) and (W, λ) are two representation of the same group G. Then a morphism (also called an intertwining or a G-linear map) from (V, ρ) to (W, λ) is a linear map h : V W which preserves the operators ρ g and σ g in the following sense: for each g G and for all v V, we have h ρ g (v) = σ g h(v) for all g G. Also, if h is an isomorphism, then we define h : GL(V ) GL(W ) as ρ g h ρ g h 1 = σ g. Definition 9. A bijective homomorphism is called an isomorphism. 3,

4 Lemma 1. Suppose (V, ρ) is a representation of G and h : V = W is some vector space isomorphism. Suppose h : GL(V ) GL(W ) is the conjugation isomorphism. Then: (1) (W, h ρ) is a representation of G. (2) (V, ρ) = (W, h ρ =: σ). Proof. The first part follows from the fact that h is an isomorphism, as it implies that h ρ is a group homomorphism. As for the second part, we start with h isomorphism. First, observe that σ g = h ρ(g) = h (ρ g ). For any g G, we have σ g h = h (ρ g ) h = (h ρ g h 1 ) h = h ρ g, as required. Corollary 3. Every representation (V, ρ) of degree n is isomorphic to (C n, σ) for some σ. Proof. Use the linear map f B : V C n for basis B and apply the previous lemma. 3. September 11 Definition 10. Suppose (V, ρ) is a representation of G and W V. Then (i) W is G-invariant or G-stable if ρ g (W ) W for all g G. (ii) If W is G-invariant, then ρ W denotes the map with domain G given by (ρ W )(g) = ρ g W. Lemma 2. If (V, ρ) of G and W is a G-invariant subspace of V, then (W, ρ W ) is a representation of G. Proof. We need to prove that ρ W : G GL(W ) is a homomorphism. For this, we need to verify the multiplicativity of ρ W. (Need to fill in the details!) To prove that ρ W is bijective, we note that ρ g is a bijection, so ρ g W is injective. For subjectivity, use the fact that W is G-invariant, and that g 1 G: ρ g 1(W ) W, hence W ρ g (W ). Thus, ρ g (W ) = W, as required. 4. September 15 & 16: Proof of Maschke s theorem and irreducible representations Recall that if (W 0, ρ), (W 1, σ) are representations of G, then (W 0, ρ) (W 1, σ) is the representation (V, ρ σ) where: V = W 0 W 1 (ρ σ) g = ρ g σ g : V V given by (ρ g σ g )(w 0 + w 1 ) = ρ g (w 0 ) + σ g (w 1 ) where w i W i. Example 3 (Representation of Z/6Z). (C 2, ρ) = (C 2, (id C 2, L, L 2, id C 2, L, L 2 )) where L : C 2 C 2 defined by ( ) 0 ω 1 ω 2, where ω = e 2πi/3. In fact, this matrix is diagonalizable, with eigenvalues 1 and ω. If E 1 and E ω are the eigenspaces, then C 2 = E1 E ω, and E 1 and E ω are Z/6Z-invariant. This gives us the decomposition (C 2, ρ) = (E 1, ρ E1 ) (E ω, ρ Eω ). (1) We want to know what (E 1, ρ E1 ) and (E ω, ρ Eω ) signify. To start off, examine each map. Then ρ E1 maps (0, 1, 2, 3, 4, 5) (id E1, L E1, L 2 E 1, id E 1, L E1, L 2 E1 ). But since L E1 = L 2 E 1 = id E1, we see that (E 1, ρ E1 ) is the trivial representation, i.e., = (C, τ). 4

5 As for ρ Eω, we see that ρ Eω is a multiplication map (by ω). Thus (E ω, ρ Eω ) = (C, σ) where σ(a) = multiplication map by ω a mod 3. Thus (C 2, ρ) = (C, τ) (C, σ). Note that (1) can be stated in a more general way: Lemma 3. Suppose that (V, ρ) is a representation of G. If V = W 0 W 1 and both W 0, W 1 are G-invariant. Then (V, ρ) = (W 0, ρ W0 ) (W 1, ρ W1 ). Definition 11. A representation (V, ρ) of G is irreducible if V {0} and V has no non-trivial G-invariant subspaces. If it is not irreducible, then that representation is called reducible. Lemma 4. Suppose p is a projection of V onto W and h GL(V ). Then h (p) := h p h 1 is also a projection of V onto h(w ). Proof. We need to show two things: (i) h (p)(v ) = h(w ) Since h is an isomorphism, it follows h (p)(v ) = (h p h 1 )(V ) = (h p)(v ) = h(w ), as required. (ii) For all x h(w ), we have h (p)(x) = x. Let x h(w ), say, x = h(w) for some w W. Then h (p)(x) = h (p)(h(w)) = (h p h 1 )(h(w)) = h(w) = x. Thus h (p) is a projection also. Theorem 1 (Maschke s Theorem). Suppose (V, ρ) is a representation of G where G is finite. Suppose that W V is G-invariant. Then W has a complement which is also G-invariant. Proof. It suffices to find a projection of V onto W whose kernel is G-invariant, i.e. some map p : V V so that p(w) = w for all w W and im(p) = W. Let f be an arbitrary projection of V onto W. Then for each g G, we have ρ g GL(V ). Let f g := ρ g(f). By Lemma 4, f g is also a projection of V onto ρ g (W ) = W. Define f := 1 f g, i.e., the average of all the f g End(V ). We claim that f is a projection of V onto W. Clearly, since f is composed of linear maps, f is clearly linear. For any x V, we have f(x) = 1 f g (x). Note that each f g (w) W, so f g (x) W also. Thus f(x) W. Also, for w W, we need f(w) = w. Since f(w) = 1 f g (w) = 1 (ρ g f ρ 1 g )(w) = 1 w = w, this claim follows. To finish the proof, we need to prove that ker(f) is G-invariant. Let g G and w ker(f). We need to show that f(ρ g (w)) = 0. Since ρ g f = f ρ g (see Lemma 5 for proof), we have f(ρ g (w)) = (f ρ g )(w) = (ρ g f)(w) = ρ g (f(w)) = ρ g (0) = 0. So ker(f) is our G-invariant complement to W. Lemma 5. For all g G, we have ρ g f = f ρ g. 5

6 Proof. equivalently, we have ρ g f ρ 1 g h G. Then ρ h(f) = ρ h f ρ 1 h ( ) 1 = ρ h f g = 1 = 1 = 1 = 1 = f, i.e., ρ g(f) = f for all g G. Choose some ρ h f g ρ 1 h ρ 1 h ρ h ρ g f ρ 1 g ρ hg f ρ (hg) 1 ρ hg(f) = 1 ρ 1 h ρ g(f) = f. Corollary 4. Suppose that G is finite, (V, ρ) is a representation of G, with dim(v ) = n > 0. Then (V, ρ) can be written as a direct sum of irreducible representations (of G). Proof. We prove by induction on n. Every representation of degree 1 is automatically irreducible, hence the base case. Now suppose that the conclusion is true for n > 1. If (V, ρ) is irreducible, then the conclusion is immediate, so suppose otherwise. Then there exists a proper subspace W V with W a G-invariant subspace. Apply Mascke s theorem to get a G-invariant complement, say V 1. Thus V = W V 1. Since W and V 1 are both G-invariant, it follows that (V, ρ) = (W, ρ W ) (V 1, ρ V1 ). Note that W < V, so dim(w ) < dim(v ) = n, and since W 0, the dimension of W must be positive. Hence 0 < dim(v 1 ) < n also. By the inductive hypothesis, both (W, ρ W ) and (V 1, ρ V1 ) can be written as a direct sum of irreducible representations. 5. September 16 & 18: Tensor products For C m and C n, define C m n = {all m n matrices over C} as a vector space over C. Define ι : C m C n C m n by ι(u, v) = uv T. The map ι is in fact bilinear: that is, ι is linear on both first and the second variables, i.e., ι(cx + y, v) = cι(x, v) + ι(y, v) and ι(u, cx + y) = cι(u, x) + ι(u, y). We start with the standard basis: C m = span{e 1, e 2,..., e m } and C n = span{e 1, e 2,..., e n}. Now we define the tensor product: Definition 12. Define ι(e i, e j) = (a kl ) where a kl = 1 only when (k, l) = (i, j) and 0 otherwise. We denote ι(u, v) as u v. We call C m n the tensor product of C m, C n (via ι) and we denote it as C m C n. Also, u v is called a simple tensor. Remark 1. Note that not all tensor are simple, i.e., cannot be written in the form u v. Proposition 13. {ι(e i, e j) : 1 i m, 1 j n} is a basis for C m n. 6

7 Proof. Observe that ι = : C m C n C m C n is a bilinear map, and that {e i e j 1 i m, 1 i n} is a basis for C m C n. Observe also that ι is neither injective nor surjective, since u 0 = 0 v = 0 C m n for all u C m and v C n. ι is not surjective either since ι(u, v) = uv T is always of rank 1, i.e., (im(ι) = {all rank 1 matrices in C m n }. It is also true that span(im(ι)) = C m n = C m C n. In fact, every x C m C n (= C m n ) is a sum of simple tensors. Hence write x = (m i,j ) C m n. It is known that the matrix can be written as follows: (m i,j ) = i,j m i,j (e i e j) = i,j ((m i,j e i ) e j), and (m i,j e i ) e j. This completes the proof Universal property of tensor products. Proposition 14. Suppose V is a C-vector space, and fix m, n Z. If α : C m C n V is a bilinear, then there exists a unique linear map α : C m C n V such that α ι = α. Proof. α ι = α implies that for all (u, v) C m C n, we have α(u v) = α(u, v). Uniqueness is clear from Proposition 13: recall that α is determined by its actions on simple tensors, and that {e i e j} is a basis for C m C n. So there exists a unique linear map C m C n V such that e i e j α(e i, e j). Let α be this map. So it remains to prove that α(u v) = α(u, v) for all u C m, v C n. Start with u C m, v C n. So u = u i e i, v = v j e j for some (u i ) m and (v j ) n j=1. Then ( ) ( ) ( )) u v = u i e i v j e j u i (e i v j e j = i i u i ( j j v j (e i e j) ) = i = i,j u i v j (e i e j). j Therefore ( ) α(u v) = α u i v j (e i e j) i,j = u i v j α(e i e j) = i,j i,j u i v j α(e i, e j), so α is linear. Similarly, ( α(u, v) = α u i e i, ) v j e j = ( u i α e i, ) v j e j i j i j = ( ) u i v j α(e i, e j) = u i v j α(e i, e j) = α(u v), i j i,j as required. 7

8 One application of Proposition 14. Let f : C m C k and g : C n C l linear. Let V = C k C l. Define α : C m C n V by α(u, v) = f(u) g(v). We claim that this is bilinear. Claim. α is bilinear. Proof. α is linear on the first variable: α(x + y, v) = f(x + y) g(v) = (f(x) + f(y)) g(v) (f is linear) = f(x) g(v) + f(y) g(v) ( is bilinear) = α(x, v) + α(y, v). We can apply the same argument on the second variable as well to prove the claim. Now apply Proposition 14. Thus, there exists a linear map α : C m C n V so that α ι = α. We call this α = f g. Therefore, with f : C m C k and g : C n C l, one can form f g : C m C n C k C l satisfying (f g)(u v) = f(u) g(v). In particular, if k = m and l = n and both f and g are invertible, then so is f g (See Assignment 2.). In other words, given f GL(C m ), g GL(C n ), we get f g GL(C m C n ) Tensor product and representation. Definition 15. Suppose that (C m, ρ), (C n, σ) are representations of G. Define ρ σ : G GL(C m C n ) given by (ρ σ) g = ρ g σ g. Then ρ σ is a group homomorphism. Therefore, (C m C n, ρ σ) is a representation of G. (C m C n, ρ σ) is called the tensor product of (C m, ρ), (C n, σ). Example 4. Let (C 3, ρ) be the representation of S 3 from Example 2. We want to compute (C 3, ρ) (C 3, ρ) = (C 3 C 3, ρ ρ). First, find ρ (123) ρ (123) = (ρ ρ) (123). Let z C 3 C 3 = C 3 3. Then (ρ (123) ρ (123) )(e i e j ) = ρ (123) (e i ) ρ (123) (e j ) = e i+1 e j+1. Thus the given tensor map shifts the entries to the right by one unit and then down by one unit, i.e., (ρ (123) ρ (123) ) z 11 z 12 z 13 z 21 z 22 z 23 = z 33 z 31 z 32 z 13 z 11 z 12 z 31 z 32 z 33 z 23 z 21 z 22 More generally, we have (ρ π ρ π )([z ij ]) = [z π 1 (i),π 1 (j)]. Thus (C 3, ρ) 2 = (C 3, ρ) (C 3, ρ) is the space of 3-by-3 matrices, with S 3 acting on rows and columns. Definition 16 (for this example). We define Sym 2 (C 3 ) and Alt 2 (C 3 ) as follows: Sym 2 (C 3 ) = {x C 3 C 3 : x = x T }, Alt 2 (C 3 ) = {x C 3 C 3 : x T = x}. Both Sym 2 (C 3 ) and Alt 2 (C 3 ) are subspaces of C 3 C 3, and that (C 3 ) 2 = Sym 2 (C 3 ) Alt 2 (C 3 ). Also, both Sym 2 and Alt 2 are S 3 -invariant. 8

9 6. September 22 Definition 17. Given V, W complex vector spaces, a tensor product of V with W is a pair (i, X) where: (i) X is a complex vector space (ii) i : V W X is bilinear. (iii) If α : V W Y is any bilinear map, then there exists a unique linear map α : X Y such that α i = α. Proposition 18. If (i 1, X 1 ) and (i 2, X 2 ) are tensor product for V with W, then there exists a unique isomorphism X 1 = X2 such that V W i 1 i 2 X 1 f X 2 i.e., for any (v, w) V W, we have (f i 1 )(v, w) = i 2 (v, w). Denote any tensor product of V with W by X := V W and write v w := i(v, w). Recall from last week that there exists a tensor product of C m with C n (X = C m n, i(v, w) = v w = vw T.). In general, if V, W are finite-dimensional spaces (say dim C V = n, dim C W = n by choosing bases, we can identify V C m and W C n, and V W C m C n = C m n. Thus V W exists for all finite-dimensional spaces. Proposition 19. For any finite-dimensional V, W, X: (i) V W = W V (ii) V (W X) = (V W ) X (iii) V (W X) = (V W ) (V X). Proof of (i). Say dim V = m, dim W = n. Then dim(v W ) = dim(v ) dim(w ) = mn = dim(w V ). However, what is more important is that there is a natural isomorphism. Namely, we have an isomorphism sending v w w v. In concrete case: this would be a transpose map. In general, we call the isomorphism V W = W V the transpose, denoted with T. As in the concrete case, if f : V 1 V 2, g : W 1 W 2 are linear, then there exists a unique linear map f g : V 1 W 1 V 2 W 2 such that v w f(v) g(w). So given f GL(V 1 ) and g GL(V 2 ), get f g GL(V 1 V 2 ). Thus, given a finite group G, with representations (V, ρ), (W, σ), get a representation (V W, ρ σ) = (V, ρ) (W, σ) with (ρ σ) g = ρ g σ g. Definition 20. Suppose that V is a finite-dimensional vector space. Then Sym 2 (V ) = {x V V : x T = x} Alt 2 (V ) = {x V V : x T = x}. 9

10 Lemma 6. For any V and any f GL(V ), both Sym 2 (V ) and Alt 2 (V ) are f f-invariant. Proof (for Alt 2 (V )). Let x Alt 2 (V ). So x V V, x T = x, i.e., x + x T = 0. Write n x = (u i v i ). Then ( n T n n x T = (u i v i )) = (u i v i ) T = v i u i. We are also given that (ui v i ) + (v i u i ) = 0. (2) Let y = (f f)(x). We must show that y Alt 2 (V ). Then we have ( n ) n y = (f f)(x) = (f f) u i v i = (f f)(u i v i ) Thus y T = = n f(v i ) f(u i ). n f(u i ) f(v i ). Now apply f f to (2): ( ) 0 = (f f)(0) = (f f) (ui v i ) + (v i u i ) n n = f(u i ) f(v i ) + f(v i ) f(u i ) = y + y T. Corollary 5. If G is a finite group and (V, ρ) a representation of G (of finite degree), then: (i) Sym 2 (V ), Alt 2 (V ) are G-invariant with respect to (V, ρ) (V, ρ). (ii) (V, ρ) (V, ρ) = (Sym 2 (V ), (ρ ρ) Sym 2 (V )) (Alt 2 (V ), (ρ ρ) Sym 2 (V )). Example 5. Let (C 3, ρ) be a representation of S 3 (as in Example 2). Then from Assignment #1, we know that (C 3, ρ) = B 1, where B := (C 2, σ) and 1 := (C, τ) as defined in the solution to Assignment #1. Consider Recall also that For notational simplicity, let (C 3, ρ) (C 3, ρ) = (B 1) (B 1) = (B B) (B 1) (1 B) (1 1) = (B B) B B 1. B B = (Sym 2 (C 2 ), σ Sym 2 (C 2 )) (Alt 2 (C 2 ), σ Alt 2 (C)). (Sym 2 (C 2 ), σ Sym 2 (C 2 )) =: Sym, (Alt 2 (C 2 ), σ Alt 2 (C)) =: Alt. (3) 10

11 Note that Sym is of degree 3 while Alt is of degree 1. Thus Alt is an irreducible representation. First, examine Alt. Since Alt 2 (C 2 ) is a set of all skew-symmetric matrices, the basis is Meanwhile, σ acts on C 2 via matrices: ( 0 1 M (12) = 1 0 ( ). ) ( 1 1, M (123) = 1 0 σ σ acts on C 2 C 2. Use basis e 11, e 12, e 21, e 22. Computations show that σ σ acts by matrices M 2 (12) = = This shows that (( 0 1 (σ σ) 1 0 )) ( 0 1 = 1 0 Hence (σ σ)(x) = x for all x Alt 2 (C 2 ). Thus, Alt trivial rep. ). ). 7. September 23: Introduction to character theory 7.1. Some linear algebra review. Definition 21. Let M = (a ij ) be a n n matrix over C. (i) The characteristic polynomial of M is p(x) = det(xi n M). (ii) The roots of the characteristic polynomial are the eigenvalues of M (up to multiplicity). (iii) p(x) = (x λ 1 )(x λ 2 ) (x λ n ). Hence constant coeff = ( 1) n λ 1 λ n, and coeff of x n 1 = (λ λ n ). (iv) λ 1... λ n = det(m), λ λ n = tr(m). (iv) If M, N are similar (i.e., there exists invertible Q such that N = QMQ 1, then M, N have the same characteristic polynomial, and hence same eigenvalues, determinant, and trace. If V is finite-dimensional (dim(v ) = n) and f : V V linear map, then by choosing a basis for V, one can identify f with a matrix; then we can get char. poly., determinant, eigenvalues, and trace for f. By (v) in Definition 21, they are independent of basis. Let det(f) be the determinant of f and tr(f) the trace of f. Definition 22. Let (V, ρ) be a finite-dimensional representation of a finite group G. The character of (V, ρ) is the function χ : G C given by χ(g) = tr(ρ g ). 11

12 Example 6. Let (C 2, ρ) = B as in Example 4. Each σ π is given by a matrix M π : ( ) ( ) M id =, M 0 1 (12) = 1 0 ( ) ( ) M (13) =, M 0 1 (23) = 1 1 ( ) ( ) M (123) =, M 1 0 (132) =. 1 1 The character of B is map χ : S 3 C given by π id (12) (13) (23) (123) (132) χ(π) Proposition 23. Let (V, ρ) be a representation of degree n of a finite group G. Let χ be its character. Then (1) χ(1) = n. (2) χ(g 1 ) = χ(g) for all g G. (3) χ is constant on conjugacy classes of G, i.e., χ(aga 1 ) = χ(g) for all a, g G. Proof. For (1), pick a basis for V so that [ρ 1 ] = I n. So χ(1) = tr(ρ 1 ) = tr(i n ) = n. As for (2), consider ρ g and ρ g 1 = (ρ g ) 1. Let λ 1,..., λ n be the eigenvalues of ρ g. Then λ 1 1,..., λ 1 n are the eigenvalues of (ρ g ) 1. It is known (by assignment #2) that Therefore, λ i = 1, so λ 1 1 = λ i. χ(g 1 ) = tr(ρ g 1) = tr((ρ g ) 1 ) = n λ 1 i = n λ i = n λ i. Finally, to prove (3), we start by picking a, g G. So ρ aga 1 = ρ a ρ g (ρ a ) 1. Pick a basis for V. And let M = [ρ g ] and N = [ρ aga 1]. Note that M and N are similar: let N = QMQ 1 and Q = [ρ a ]. So ρ(g) = tr(ρ g ) = tr(m) = tr(n) = tr(ρ aga 1) = χ(aga 1 ). For any two matrices, we have tr(ab) = tr(ba). Thus, tr(ρ ag ) = tr(ρ a ρ g ) = tr(ρ g ρ a ) = tr(ρ ga ). Definition 24. Let G be a finite group. A class function on G is any function α : G toc which is constant on conjugacy classes. We write ClaFun(G) := {all class functions on G}. Observe that ClaFun(G) is closed under: (1) pointwise addition. That is, if α, β ClaFun(G), then (α + β)(g) = α(g) + β(g). (2) pointwise multiplication, i.e. (α β)(g) = α(g) β(g). (3) complex scalar multiplication, i.e., (cα)(g) = c(α(g)). Proposition 25. Suppose that G is finite, and (V, ρ) and (W, σ) are finite-degree representations of G. Let χ ρ, χ σ be their characters. (1) χ ρ + χ σ is the character of (V, ρ) (W, σ). (2) χ ρ χ σ is the character of (V, ρ) (W, σ). 12

13 Proof. Let e = (e 1,..., e m ) and e = (e 1,..., e n) be bases for V and W. Then we see that e e is a basis for V W. For any g G, we have ( ) [ρg ] [(ρ σ) g ] e e = e 0. 0 [σ g ] e Let χ be the character of (V, ρ) (W, σ) = (V W, ρ σ). Then we have (( )) [ρg ] χ(g) = tr((ρ σ) g ) = tr e 0 0 [σ g ] e = tr([ρ g ] e ) + tr([σ g ] e ) = tr(ρ g ) + tr(σ g ) = χ ρ (g) + χ σ (g). 8. September 25: Continuation of character theory Proposition 26. Suppose G is finite and (V, ρ), (W, σ) are representations of G (finite degree), with characters χ ρ, χ σ. Then: (1) The character of (V, ρ) (W, σ) is χ ρ + χ σ. (2) The character of (V, ρ) (W, σ) is χ ρ χ σ. Proof. Fix g G. Let e = (e 1, e 2,... e m ) be a basis for V and e = (e 1,..., e n) be a basis for W. Let [ρ g ] e = (a ij ) and [σ g ] e = M. Recall that (e i e j) is a basis for V W, where 1 i m, 1 j n. And a 11 M a 12 M a 1m M a [ρ g σ g ] = 21 M a 22 M a 2m M a m1 M a m2 M a mm M Thus if χ is the character of (V, ρ) (W, σ), then χ(g) = tr((ρ σ) g ) = tr(ρ g σ g ) m = tr([ρ g σ g ]) = a ii tr(m) = tr((a ij )) tr(m) = tr([ρ g ] e ) tr([σ g ] e ) = tr(ρ g ) tr(σ g ) = χ ρ (g) χ σ (g). Proposition 27. Suppose that (V, ρ) is a finite-degree representation of a finite group G. Let χ be its character, χ S the char of (Sym 2 (V ), ρ 2 Sym 2 (V )) (V, ρ) 2, and χ A the char of (Alt 2 (V ), ρ 2 Alt 2 (V )) (V, ρ) 2. Then for all g G: (1) χ S (g) = 1 2 (χ(g)2 + χ(g 2 )) (2) χ A (g) = 1 2 (χ(g)2 χ(g 2 )) (3) χ S (g) + χ A (g) = χ(g) 2. Proof. Note that (3) follows from Proposition 26. Fix g G. Then ρ g is diagonalizable (i.e., there exists a basis for V consisting of eigenvectors for ρ g. See also Assignment #3). Let β = (e 1,..., e n ) be a basis for V consisting of eigenvectors for ρ g. Let λ 1 C be such that 13

14 ρ g (e i ) = λ i e i. So tr(ρ g ) = λ 1 + +λ n. Also, ρ g 2 is also diagonalized by β, and ρ g 2(e i ) = λ 2 i e i and tr(ρ g 2) = n λ 2 i. β S := {e i e j + e j e i : i j} is a basis for Sym 2 (V ). Then we claim that ρ 2 g Sym 2 (V ) is diagonalizable with respect to β S. Observe that This shows that ρ 2 g (e i e j + e j e i ) = (ρ g ρ g )(e i e j + e j e i ) = ρ g (e i ) ρ g (e j ) + ρ g (e j ) ρ g (e i ) = λ i e i λ j e j + λ j e j λ i e i = λ i λ j (e i e j + e j e i ). tr(ρ 2 g Sym 2 (V )) = λ i λ j. i j Thus, χ S (g) = tr(ρ 2 g Sym 2 (V )) = λ i λ j i j = λ 2 i + λ i<j λ i λ j i ( ) = 1 2 λ i + 2 i i λ 2 i = 1 2 (χ(g)2 + χ(g 2 )), as required. (2) can be proved in a similar manner. Example 7. Let B = (C 2, σ) be a degree-two irreducible representation of S 3. Recall that and M (12) = [ So χ 2 should be the character of B B: Recall that ] [ 1 1, M (123) = 1 0 π id (a b) (a b c) χ(π) π id (a b) (a b c) χ 2 (π) B B = Sym Alt 14 ],

15 (Refer to (3) for the definitions of Sym and Alt.). Apply Proposition 27 to get the character tables for Sym and Alt. For Sym: χ S (id) = 1 2 (χ(id)2 + χ(id 2 )) = = 3 2 χ S (a b) = 1 2 (χ(a b)2 + χ((a b) 2 )) = = 1 2 χ S (a b c) = 1 2 (χ(a b c)2 + χ((a b c) 2 )) = 1 1 = 0. 2 π id (a b) (a b c). χ S (π) In a similar manner, one can obtain the following table for Alt: π id (a b) (a b c) χ A (π) Finally, we claimed on Monday that Sym = B 1. Verify that χ S = χ + χ September 29 Definition 28. Let G be a finite group. Define (1) C G := the set of all functions G C (2) If α, β C G, then (α β) = 1 α(g)(β(g)), where z denotes the complex conjugate of z. (3) If α, β C G, then α, β = 1 α(g)β(g) 1 = 1 α(g 1 )β(g) = β, α. (4) If β C G, then ˆβ C G, where ˆβ(g) := β(g 1 ). Obviously, ˆβ(g 1 ) = β(g). Remark 2. We can make the following observations: (1) C G is C = C n, where G = {g 1, g 2,..., g n }. (2) C G is a complex vector space of dim n =. (3) C G is a ring (with pointwise addition and multiplication). (4) ClaFun(G) C G. (5) ( ) is a complex inner product on C G, i.e., (α 1 + α 2 β) = (α 1 β) + (α 2 β) (cα β) = c(α β) (c C) (β α) = (α β) (α β) C (α α) 0 ((α α) = 0 α = 0) 15

16 Suppose that (α α) = 0. Then we have α(g)α(g) = 0, or α(g) 2 = 0. Therefore, one has α(g) = 0 so α = 0. Remark 3. (α β) = 1 = 1 α(g)β(g) α(g) ˆβ(g 1 ) = α, ˆβ Remark 4. Which β satisfied β = ˆβ? We need β to satisfy β(g) = β(g 1 ) for all g G. This is equivalent to saying that β(g 1 ) = β(g). Characters have this property (see Proposition 23). So ˆχ = χ for all characters χ. Thus, for all characters the relation (α χ) = α, χ holds true, for all α C G and all characters χ. Suppose that (V, ρ) is a finite-degree representation of finite group G. Let χ be its character and ē a basis for V. Then for all g G, we get [ρ g ]ē = (r ij (g)). Thus we get a bunch of functions r ij : G C, i.e. r ij C G. For any character χ, n n χ(g) = tr(ρ g ) = tr([ρ g ]ē) = tr(r ij (g)) = r ii (g) χ = r ii C G. This proves: Theorem 2 (Fundamental Observation). If (V, ρ) is a finite-degree representation of a finite group G and χ is its character, and (r ij (g)) gives the family matrices representing the ρ g s with respect to some basis, then χ = r 11 + r r nn. Lemma 7 (Schur s Lemma). Suppose G is a finite group, and (V, ρ), (W, σ) are irreducible representations. Suppose also that f : (V, ρ) (W, σ) is a morphism. (1) f is either an isomorphism or the constant zero map. (2) If (W, σ) = (V, ρ), then f is a scalar map, i.e., there exists λ C such that f(v) = λv for all v V. Proof. (Part (1)) Assume that f is not the constant zero map. Let X = ker(f). We know that X is G-invariant subspace of V. Since f 0, X V. Since (V, ρ) is irreducible, it follows X = {0} so f is injective. Now let Y := im(f). We know that Y is G-invariant subspace of W. Since f is injective, Y {0}. Since Y is irreducible, we must have Y = W. (Part (2)) Assume that (W, σ) = (V, ρ). So for f : V V, we can choose an eigenvalue λ of f, say with eigenvector v. Let g : V V be g = f λ id V, i.e., d(v) = f(v) λv. Then g is a morphism from (V, ρ) to itself. By (1), g is either an isomorphism or the constant zero map. Observe that if v 0, then g(v) = 0, so g must be the constant zero map. Hence f(v) = λv, as required. 10. September 30 Definition 29. For α, β C G, we have α, β = 1 16 α(g 1 )β(g).

17 Corollary 6. Let G be finite, and (V, ρ), (W, σ) irreducible representations of G and f : V V be a linear map. Define f 0 : V W by Then: f 0 = 1 σ g 1 f ρ g. (1) If (V, ρ) = (W, σ) then f 0 0. (2) If (V, ρ) = (W, σ), then f 0 (v) = λv, where λ = tr(f) dim V. Proof. Main step: prove that f 0 is a morphism of representations, i.e., f 0 (ρ g (v)) = σ g (f 0 (v)) for all v V, g G. Fix g G. Then we have σ 1 g Now apply Schur to f 0 : f 0 σ g = σ 1 g = 1 ( ) 1 σ h 1 f ρ h ρ g h G σ g 1σ h 1fρ h ρ g = 1 σ (hg) 1fρ hg = f 0. h G (1) Assume (V, ρ) = (W, σ). Then f 0 is not an isomorphism. Apply Schur, then we get f 0 0. (2) Now assume (V, ρ) = (W, σ). Then by Schur, we get that f 0 (v) = λv for some λ. On the first and, tr(f 0 ) = λ dim(v ). On the other hand, ( ) tr(f 0 1 ) = tr ρ g 1fρ g = 1 g = 1 h G tr(ρ 1 g fρ g ) tr(ρ g 1ρ g f) = 1 tr(f) = tr(f). Let dim V = m, dim W = n. Let e = (e 1, e 2,..., e m ) be a basis for V and e a basis for W. Define [ρ g ] e = (r kl (g)) m m and [σ g ] e = (s ij (g)) n n. Suppose h : V W is a linear map, and we can write [h] e e = (x jk ) n m. Note that r kl, s ij C G. Define f 0 as before. Then what is [f 0 ] e e? Define [f 0 ] e e := (c il ) n m. Formula for c il. Start with f 0 = 1 σ g 1fρ g So g [f 0 ] = 1 (c il ) = 1 [σ g 1][f][ρ g ] (s ij (g 1 ))(x jk )(r kl ) 17

18 For each (i, l), we have ( ) c il = 1 s ij (g 1 )x jk r kl (g) = ( ) 1 s ij (g 1 )r kl (g) j,k j,k g = ( ) 1 s ij (g 1 )r kl (g) x jk = s ij, r kl x jk. j,k g j,k If (V, ρ) = (W, σ), then f 0 = 0. Thus (c il ) is the zero matrix. That is, the sum s ij, r kl x jk = 0. j,k This is true for all f, or for all (x jk ). Hence s ij, r kl = 0 for all i, j, k, l. Thus we proved Corollary 7. If (V, ρ), (W, σ) are irreducible representations, then [ρ g ] = (r kl (g)), [σ g ] = (s ij (g)) with respect to some basis. If (V, ρ) = (W, σ), then sij, r kl = 0 for all i, j, k, l. Next, assume that (V, ρ) = (W, σ). And let [ρ g ] = (r kl ) = (r ij ), [f] = (x kl ), [f 0 ] = (c il ). So c il = r ij, r kl x jk. By Corollary 6, f 0 (v) = λv where λ = tr(f). Thus c m il is λ if i = l and 0 j,k otherwise. Thus So c il = δ il ( tr(f) ) = 1 δ m m il tr(f) = x jj = δ jk x jk. j j,k δ jk x jk = j,k( 1 δ m ilδ jk )x jk. Thus, for all i, l, j,k r ij, r kl x jk = j,k j,k ( ) 1 m δ ilδ jk x jk. Since this is true for all f (= for all (x jk )), it follows r ij, r kl = 1 m δ ilδ jk for all i, j, k, l. Hence this proves Corollary 8. If (V, ρ) an irreducible representation and [ρ g ] = (r kl (g)), then for all i, j, k, l, r ij, r kl = 1 m δ ilδ jk = { 1 m 11. October 2 (i = l, j = k). 0 otherwise Corollary 9. (V, ρ), (W, σ) irreducible and [ρ g ] = (r kl ) g, [σ g ] = (s ij (g)) then s ij, r kl = 0 for all i, j, k, l. Corollary 10. Let (V, ρ) be irreducible and [ρ g ]] = (r kl (g)). Then r ij, r kl = { 1 dim V if i = l, j = k 0 otherwise. Corollary 11. Suppose (V, ρ), (W, σ) are representations with characters χ, χ, respectively. If (V, ρ) = (W, σ) then χ χ. Theorem 3. If G is a finite group, then 18 x jk

19 (1) if χ is the character of an irreducible representation of G, then (χ χ) = 1. (2) If χ, χ are the characters of irreducible representations (V, ρ), (W, σ) of G and (V, ρ) (W, σ), then (χ χ ) = 0. Proof. For part (1), start with χ, the character of (V, ρ). Pick a basis β for V. Write [ρ g ] β = (r ij (g)). Recall that χ = r r nn where n = dim V. Then (χ χ) = χ, χ = r ii, r ii = r ii, r jj = δ ij = n 1 = 1. n n i i i,j i,j As for part (2), write [σ g ] β = (s ij (g)) and χ = s s mm, where m = dim W. Then we have (χ χ ) = χ, χ = r ii, s jj = 0. i,j Theorem 4. Let (V, ρ) be a finite-degree representation of G, with ϕ its character. Suppose that (V, ρ) can be decomposed to (V, ρ) = (V 1, ρ 1 ) (V k, ρ k ) where each (V i, ρ i ) is an irreducible representation of G. Let χ i be the character of (V i, ρ i ). Let (W, σ) be any irreducible representation of G with character χ. Then Proof. We know that ϕ = χ 1 + χ k. So (ϕ χ) = {i : (V i, ρ i ) = (W, σ)}. (ϕ χ) = (χ χ k χ) = (χ 1 χ) + + (χ k χ) = {i : (χ i χ) = 1} = {i : χ i = x} = {i : (V i, ρ i ) = (W, ρ)}. Corollary 12. Let (V, ρ) be a finite-degree representation of G. Any two decompositions of (V, rho) as direct sums of irreducible representations, are the same up to rearrangements and isomorphism of the individual summands. Proof. For each irreducible (W, σ) and char χ, we have (ϕ χ) = (# of times (W, ρ) appears in any decomposition). Corollary 13. If (V, ρ), (W, σ) are the finite-degree representations of G, with the characters ϕ, ϕ, then (V, ρ) = (W, σ) ϕ ϕ. Proof. ( ) Assume ϕ = ϕ. Look at the direct sums of decompositions for (V, ρ) and (W, σ). By Theorem 4, for any irreducible (X, τ) (with character χ), the number of times (X, τ) occurs in either decompositions is (ϕ χ) = (ϕ χ). So, up to isomorphism, (V, ρ) and (W, σ) have the same decompositions. So (V, ρ) = (W, σ). ( ) This direction is immediate. Let (V, ρ) be a finite-degree representation, and suppose that ( ) ( ) ( ) (V, ρ) = (w 1, ρ 1 ) (W 2, ρ 2 ) (W k, ρ k ). m 1 m 2 m k 19

20 Let ϕ be a character of (V, ρ) and χ i the character of (W i, ρ i ). So ϕ = m 1 χ m k χ k, and ( k ) k (χ χ) = m i χ i m j χ j = m i m j (χ i χ j ) = m 2 i. j=1 i,j i Theorem 5. Let (V, ρ) be a finite-degree representation and ϕ its character. Then (ϕ ϕ) is a positive integer. Moreover, (ϕ ϕ) = 1 if and only if (V, ρ) is irreducible. Example 8. Let B = (C 2, σ) be the irreducible representation of S 3, and let χ be its character. The character table is as follows: π id (12) (13) (23) (123) (132) χ(π) October 6 Definition 30. An irreducible character of G is a character of an irreducible representation of G. Recall that if G is a finite group, then the set of irreducible characters of G is an orthonormal set of ClaFun(G). This proves that G has only finitely many irreducible representations, up to isomorphism. In particular, the number is bounded by dim(clafun(g)), which is the number of conjugacy classes of G. Example 9. S 3 has 3 conjugacy classes; hence S 3 has at most 3 irreducible representations, up to isomorphism. We have already seen 3: namely 1 and S of degree 1 and B of degree 2. Thus we have found them all. Let (V, ρ) be a finite-degree representation of G with character ϕ. irreducible representation of degree G with character χ. Let (V, ρ) = (V 1, ρ 1 ) (V k, ρ k ) Let (W, σ) be an be the (essentially unique) direct-sum decomposition of (V, ρ) into a sum of irreducible representations. Definition 31. We say that (W, ρ) occurs in (V, ρ) if (W, σ) = (V i, ρ i ) for some i. The multiplicity of (W, ρ) in (V, ρ) is {i : (V, ρ i ) = (W, σ)} (which may equal 0). Remark 5. The multiplicity of (W, σ) in (V, ρ) is (ϕ χ). Definition 32. Given a finite group G, let V be a complex vector space of dimension with basis E = {e g : g G} indexed by the elements of G. For each g G define ρ g GL(V ) by setting ρ g (e h ) = e gh, and extending linearly to all of V. Then (V, ρ) is called the regular representation of G, and we shall denote this character by r G. Clearly, we have r G (1) = dim(v ) =. Let g 1. Then for each h G, we have ρ g (e h ) = e gh = 0 e e h e gh +. Hence [ρ g ] E has a 0 in the (h, h) position. Since h is arbitrary, it follows that all the diagonal entries of [ρ g ] E are 0. Thus tr(ρ g ) = 0, so we proved the following proposition: 20

21 Proposition 33. The character r G of the regular representation of a finite group G is { (g = 1) r G (g) = 0 otherwise. Corollary 14. Let G be finite and let (V, ρ) be its regular representation. Every irreducible representation (W, σ) of G occurs in (V, ρ), with multiplicity dim(w ). Proof. Let χ be the character of (W, σ). Then the multiplicity of (W, σ) in (V, ρ) is (r G χ) = 1 r G (g)χ(g) = 1 r dim(w ) G(1)χ(1) = = dim(w ), as required. Suppose that (W 1, σ 1 ),..., (W k, σ k ) are the distinct irreducible representations of G, and let n i = dim(w i ) for each i. If χ i denotes the character of (W i σ i ), then by the above corollary the regular representation decomposes as Hence r G = n 1 χ n k χ k. Corollary 15. The following are true: k (1) n 2 i =. (2) For all g 1, (V, ρ) = n 1 (W 1, σ 1 ) n k (W k, σ k ). k n i χ i (g) = 0. Proof. Evaluating the displayed equation at g = 1, we get dim(v ) = n 1 dim(w 1 ) + + n k dim(w k ) = n n 2 k as required. The second part also follows from the second equation when g October 7 Recall that dim(clafun(g)) = # of conjugate classes of G {irreducible characters for G} is an orthonormal set in ClaFun(G). Today, we show that the set of irreducible characters spans ClaFun(G). Fix α ClaFun(G). For each finite-degree representation (V, ρ) of G, define by Clearly, f ρ α is linear. In fact, f ρ α(v) = 1 Claim. f ρ α is a morphism from (V, ρ) to itself. f ρ α : V V α(g)ρ g (v). 21

22 Proof. Must show that ρ g f = fρ g for all g G. Consider ρ 1 g fρ g = ρ 1 g = 1 Let u = g 1 hg h = gug 1. Then ( ) 1 α h ρ h ρ g h h α(h) ρ 1 g ρ h ρ g = 1 α(h)ρ g 1 hg. h 1 h α(h)ρ g 1 hg = 1 = 1 α(gug 1 )ρ u u α(u)ρ u = f. u What is tr(f)? ( ) 1 tr α(g)ρ g = 1 = 1 α(g) tr(ρ g ) α(g) ϕ(g) = (α ϕ ). Proposition 34. Suppose (V, ρ) is a degree n irreducible representation with χ its character, and if α ClaFun(G), then f ρ α is a scalar mp ( multiplication by λ ), where λ = 1 n (α χ ). Proof. f ρ α is a morphism from (V, ρ) to itself. By Schur s lemma, f ρ α is scalar for some λ. With respect to any basis for V, we have [fα] ρ = λ λ so tr(f ρ α) = nλ. But tr(f ρ α) = (α χ ) = nλ. Thus λ = 1 n (α χ ), as required. What if (V, ρ) is not irreducible? Time to consider that case. Decompose (V, ρ) = (W 1, σ 1 ) (W k, σ k ), where each (W i, σ i ) is irreducible. Let n = dim V, m i = dim W i. Let χ i be the character of (W i, σ i ). Fix α ClaFun(G). We have fα ρ : V V, and for each i, f σ i α : W i W i, define f σ i α (w) = λ i w. Claim. f ρ α = f σ 1 α f σ k α. 22

23 Proof of Claim. Need to check that, for v V, we can write v = w 1 + w k with w i W i. fα(v) ρ = 1 α(g)ρ g (v) = 1 = k = (f σ 1 α α(g)((σ 1 ) g (w 1 ) + + (σ k ) g (w k )) 1 α(g)(σ i ) g (w i ) = f σ k α )(v). k f σ i α (w i ) In general, for α ClaFun(G) and with representation (V, ρ) with we have (V, ρ) = (W 1, σ 1 ) (W k, σ k ) and v = w w k, w i W i, f ρ α(v) = k λ i w i, where λ i = 1 m i (α χ i ), m i = dim W i, χ i = char of (W i, σ i ). Theorem 6. The irreducible characters of G span ClaFun(G). Proof. Let χ 1,..., χ k be irreducible characters of G. Suffices to show that β ClaFun(G), (χ i β) = 0 i = 1, 2,..., k, then β = 0. Suppose that β ClaFun(G) and (χ i β) = 0 for all i. Let α = β. It suffices to show that α = 0. It is already known that (χ i α ) = (α χ i ) = 0 for all i = 1, 2,..., k. By our analysis, for any finite-degree representations (V, ρ) of G, we have fα ρ = 0. Apply this to the regular representations (V, ρ) of G, where V has basis e g, g G such that ρ g (e h ) = e gh. We get fα ρ = 0. in particular, fα(e ρ 1 ) = 0. Calculate this using definition: fα(e ρ 1 ) = 1 α(g) ρ g (e 1 ) = 1 α(g) e g = 0. Thus α(g) = 0 for all g G, so α 0. Theorem 7. If G is finite, then the number of irreducible representations of G (up to isomorphism) is equal to the number of conjugacy classes of G. Here is one neat consequence of Theorem 6: Let G be a finite group, and pick s G and let θ s be its conjugacy class. Define α : G C such that { 1 (g θ s ) α(g) = 0 otherwise. By Theorem 6, α is a linear combination of irreducible characters of G. Let χ 1,..., χ k be the irreducible characters of G. Write α = c 1 χ c k χ k, with c i = (α χ i ). (to be continued...) 23

24 14. October 9 Fix G and χ 1,..., χ k irreducible characters. Fix s G. Let θ s be the conjugacy class of s. Define f s : G C to be Write Note that (f s χ i ) = Thus Evaluate at s: f s (g) = f s = { 1 (g θ s ) 0 otherwise. k a i χ i. ( k j=1 a jχ j χ i ) = k j=1 a j (χ j χ i ) = a i. Also, (f s χ i ) = 1 f s = 1 = f s (s) = θ s Now suppose t G \ θ s. Evaluate at t: = 1 f s (g)χ i (g) g θ s 1 χ i (g) = θ s χ i(g) = a i. k k 0 = f s (t) = θ s Thus we proved the following proposition: ( ) θs χ i(s) χ i. χ i (s) χ i (s) = θ s k χ i (s) 2. k χ i (s) χ i (t). Proposition 35. Let χ 1,..., χ k be the irreducible characters of G, and let s G. (1) k χ i (s) 2 = θ s (2) If t G \ θ s, then k χ i (s) χ i (t) = 0. Example 10. Let s verify Proposition 35 when s = 1. Then χ i (1) = n i, where n i is the degree of representations for χ i. Then we observe that k (1) n 2 i = 24

25 (2) If t G \ {1}, then k n i χ i (t) = 0. Note that we saw this already, from r G = n 1 χ n k χ k. Example 11. Let s find the irreducible characters of D 5 (the dihedral group of order 10): D 5 = r, s r 5 = 1 = s 2, rs = sr 1 = r s r = {1, r, r 2, r 3, r 4 } {s, sr, sr 2, sr 3, sr 4 }. Conjugacy classes of D 5 : {1}, θ s = {s, sr 2, sr, sr 2, sr 4 }, θ r = {r, r 1 }, θ r 2 = {r 2, r 3 }. Thus, D 5 has four irreducible characters, call them χ 1, χ 2, χ 3, χ 4, say of degrees n 1, n 2, n 3, n 4 respectively. Without loss of generality, let n 1 n 2 n 3 n 4. There is only solution, namely (n 1, n 2, n 3, n 4 ) = (1, 1, 2, 2). Time for some character table. Let χ 1 be the trivial character and χ 2 the sign character: SInce (χ 3 χ i ) = 0 for i = 1, 2, we have (χ 3 χ 1 χ 2 ) = 0 = 1 D 5 1 s r r 2 χ χ χ 3 2 a c e χ 4 2 b d f g D 5 χ 3 (g) ((χ 1 χ 2 )(g)) = 1 10 ( (a 2 ) + 2(c 0 ) + 2(e 0 ) + 2(f 0 )) = a. Similar calculation shows that a = b = 0. Now from (χ 3 χ 1 ) = 0, we have 0 = 1 10 (2 1 +5(0 1 ) + 2(c 1 ) + 2(e 1 )) = 2 + 2c + 2e, so e = c 1. Similarly, we have f = d 1. Recall that n n n n 2 4 = 10. Thus since n 1 χ 1 (r) + + n 4 χ 4 (r) = 0, we have 2c + 2d = 0, so d = c 1. By (2) from Proposition 35, 4 χ i (r) χ i (r 2 ) = 0, so c ( c 1) + ( c 1) c = 0. Simplify this to get c 2 + Re(c) = 1. We also have 4 χ i (r) 2 = D 5 θ r = 5, so c 2 + c = 5. Solve for c to get c = 1± 5 2 = 2 cos 2π 5. 25

26 15. October 14 Fix: finite group G (W 1, σ 1 ),..., (W k, σ k ) the list of irreducible representations (up to isomorphism) χ i = character of (W i, σ i ) n i = dim(w i ) = χ i (1). Let (V, ρ) be a finite-degree representation of G. How can we find G-invariant subspaces U 1,..., U m of V so that with each (U j, ρ Uj ) irreducible? Assume that we have such (V, ρ) = (U 1, ρ U1 ) (U m, ρ Um ) (V, ρ) = (U 1, ρ U1 ) (U r(1), ρ Ur(1) ) (U r(1)+1, ρ Ur(1)+1 ) (U r(2), ρ Ur(2) (U r(k), ρ Ur(k) ), where we define (V i, ρ Vi ) = (U r(i)+1, ρ Ur(i)+1 ) (U r(i+1), ρ Ur(i+1) ) = (W i, σ i ) for all i. Theorem 8. The subspaces V 1,..., V k do not depend on the particular decomposition V = U 1 U m from which they arose. Proof. Start with a finite group of order = p n 1 1 p n k k. Then G = H 1 H k such that H i = p n i i, and fix some i = 1, 2,..., k. Let χ i be the character of (W i, σ i ). Define p i : V V by p i = n i χ i (g) ρ g. Note that p i depends only on (V, ρ) and χ i. Define α ClaFun(G) by α = n i χ i. Note also that ρ i = 1 α(g)ρ g = f ρ α. Note also that p i Uj = fα ρ Uj = f ρ U j α : U j U j. If (U j, ρ Uj ) = (W l, σ l ) for some l, then: the character of (U j, ρ Uj ) is χ l dim(u j ) = n l By Proposition 34, f ρ U j α is a scalar map, namely by λ = 1 (α χ l ) = (n i χ i χ l ) n l { = n i (χ i χ l ) n i if l = i = 0 otherwise, by orthogonality. Hence p i is the identity function on V i, and is the zero function on all other V l with l i. So p i is the projection map onto V i with respect to V = V 1 V k. Thus V i = im(p i ), and hence is determined intrinsically, as desired. 26

27 Definition 36. The decomposition (V, ρ) = (V 1, ρ V1 ) (V k, ρ Vk ) is called the canonical decomposition of (V, ρ). We still want to further decompose each of (V i, ρ Vi ) How to further decompose. Here we outline methods on how to further decompose each of (V i, ρ Vi ). (1) Fix i (2) Have (W i, σ i ), and choose a basis e = (e 1,..., e n ) for W i. Note that n = n i in this case. (3) Let [(σ i ) g ] e = (r st (g)) n n. This gives us r st C G, where 1 s, t n. (4) Note that χ i = r r nn, and by Corollary 8, r st, r uv = { 1 n i if s = v, t = u 0 otherwise. Definition 37. For any representation (U, τ) of G, define, for 1 s, t n, by p τ st = n i p τ st : U U r ts (g 1 )τ g. Suppose that (U, τ) is irreducible but not isomorphic to (W i, σ i ). Pick a basis e = (e 1,..., e n) for U. [τ g ] e = ( r uv (g)) m m. Then for each e u, p τ st(e u) = n i r ts (g 1 )τ g (e u). Then we have r 11 (g) r im (g) [τ g ] e =..... e u = r m1 (g) r mm (g) Then for each e u, we have p τ st(e u) = n i = n i r 1u (g). r mu (g) m r vu (g). v=1 r ts (g 1 )τ g (e u) (4) r ts (g 1 ) m ( = n i v=1 1 m r vu (g)e v (5) v=1 ) r ts (g 1 ) r vu (g) e v, (6) and r ts, r vu = 0. Thus (6) becomes 0. By Corollary 7, since true for all e i, we have p τ st 0. 27

28 16. October 16 Suppose G is finite, and (W 1, σ 1 ),..., (W k, σ k ) irreducible representations. Suppose χ 1,..., χ k their characters. Let (V, ρ) be some random representation (finite-degree). Fix some 1 i k, and let n i := dim W i. Choose basis e = (e 1,..., e n ) for W i. Write [(σ i ) g ] e = (r st (g)) n n. Given a rep (U, τ) and 1 s, t n, define p τ st = n r ts (g 1 )τ g : U U. Last time, we claimed that if (U, τ) is irreducible and = (Wi, σ i ) then all p τ st is 0. Proof. Let e be a basis for U. For each e u, we have n p τ st(e u) = n i r ts, r vu e v. v=1 Repeat this calc, but when (U, τ) = (W i, σ i ), then let e = e. Then ( r uv (g)) n n = [(σ i ) g ] e = (r uv (g)). Get: n p σ i st (e u ) = n i r ts, r vu e v v=1 = n i n v=1 ( ) 1 δ tu δ sv e v (by Corollary 8) n i n = δ tu δ sv e v = δ tu e s. v=1 When t u, then we have p σ i st (e u ) = 0; if t = u, then p σ i st (e t ) = e s. So (on W i ), p σ i st sends e t to e s, and every other basis elements e u to 0. Recall that our main goal is to decompose (V, ρ). Suppose hat we know a decomposition: (V, ρ) = (U 1, ρ U1 ) (U r(1), ρ Ur(1) ) (U l+1, ρ l+1 ) (U l+m, ρ Ul+m ). }{{}}{{} =:V 1 =(W1,σ 1 ) =V i =(Wi,σ i ) We can find V 1,..., V k. Goal : find candidates for U l+1,..., U l+m. Let p st = p ρ st : V V. From calculations, we have: (1) p st Ur with (U r, ρ Ur ) = (W i, σ i ) p ρ Ur st = 0. (2) Hence p st Vj = 0 if j i. As for p st Vi, write V i = U l+1 U l+m, with each (U l+j, ρ Ul+j = (Wi, σ i ). Let e (j) 1,..., e (j) n U l+j be the image of e 1..., e n W i under an isomorphism. Observe that, on U l+j, we have p st (e (j) t ) = e (j) s, and p st (e (j) u ) = 0 if u t. Hence, p 11 is a projection from V to span(e (1) 1, e (2) 1,..., e (m) 1 ) = V i,1, and p 22 a projection from V to span(e (1) 2,..., e (m) 2 ) = V i,2, etc. Reverse-engineer this: Start with (V, ρ). Fix a basis e for W i. Have (r st (g)) n n = [(σ i ) g ] e, so p st = p ρ st : V V. Let V i,1 = range(p 11 ). Next, pick a basis for V i,1, say e (1) 1,..., e (m) 1. For s = 2,..., n, define e (j) s = p s1 (e (j) 1 ) range(p ss ) = V i,s. 28

29 Now define U l+1 := span(e (1) 1,..., e (1) n, and define U l+2,... in a similar manner. We need to check that: each U l+j is G-invariant; V i = U l+1 U l+m ; Each (U l+j, ρ Ul+j ) = (W i, σ i ). 17. October Subgroups and products. Suppose that (V, ρ) is a representation of G. So ρ : G GL(V ). Let H G. Then ρ H : H GL(V ), so (V, ρ H =: ρ H ) is a representation of H. Proposition 38. Suppose that H G. Let k be the maximum degree of irreducible representations of H. Then every irreducible representation of G has degree k [G : H]. Proof. Let (V, ρ) be an irreducible representation of G. Then (V, ρ H ) is a representation of H. Pick an irreducible subrepresentation (W, ρ H W ) of (V, ρ H ). Note that dim(w ) k. Look at all images of ρ g (W ) for all g G. Note also that if h H, then ρ H (W ) = W. If g 1 H = g 2 H, then in particular g 2 = g 1 h for some h H. Thus ρ g2 (W ) = ρ g1 h(w ) = (ρ g1 ρ h )(W ) = ρ g1 (W ), ( since ρ h (W ) = W. Then the number of different ρ g (W ) is at most ) [G : H]. Let V = span ρ g(w ) V. Clearly, V is a G-invariant subspace of V, and dim V 1. But (V, ρ) is irreducible, so V = V. So dim(v ) = dim(v ) dim(ρ g (W )) }{{} ρ g(w ) =dim(w ) = dim(w ) (# of distinct ρ g (W ) s) k [G : H]. Corollary 16. Suppose that G has an abelian subgroup A G. Then every irreducible representation of G has degree at most [G : A]. Proof. This follows from the fact that every irreducible representation of A has degree 1. One application: if D n is a dihedral group of order 2n, then D n has a cyclic subgroup of order n. Thus every irreducible representation of D 2n has degree Direct products. Suppose that G = G 1 G 2, and let (V 1, ρ 1 ), (V 2, ρ 2 ) representations of G 1 and G 2, respectively. Let π 1, π 2 to be canonical projections from G onto G 1 and G 2, respectively. By Assignment #4, (V 1, ρ 1 π 1 and (V 2, ρ 2 π 2 ) are representations of G. Take their tensor product: (V 1 V 2, (ρ 1 π 1 ) (ρ 2 π 2 )). Write ρ 1 ρ 2 := (ρ 1 π 1 ) (ρ 2 π 2 ). So (V 1 V 2, ρ 1 ρ 2 ). Note. If g = (g 1, g 2 ), then on simple tensors, (ρ 1 ρ 2 ) g (v 1 v 2 ) = ((ρ 1 π 1 ) g (ρ 2 π 2 ) g )(v 1 v 2 ) = (ρ 1 g 1 ρ 2 g 2 )(v 1 v 2 ) = ρ 1 g 1 (v 1 ) ρ 2 g 2 (v 2 ). Let χ 1 and χ 2 be the characters of (V 1, ρ 1 ) and (V 2, ρ 2 ) respectively. Then: (V 1, ρ 1 π 1 ) has character χ 1 π 1 (V 2, ρ 2 π 2 ) has character χ 2 π 2. 29