REPRESENTATIONS OF Sn AND GL(n, C) SEAN MCAFEE 1. outline For a given finite group G, we have that the number of irreducible representations of G is equal to the number of conjugacy classes of G. Although numerically equal, there is no general formula to map the conjugacy classes of a group to its irreducible representations. For the case of the symmetric group Sn , however, there is a remarkably simple correspondence; we will see that each irreducible representation of Sn is determined uniquely by a combinatorial object called a Young diagram corresponding to a given conjugacy class. In section 2, we will introduce these diagrams and show how they determine a representation Vλ of Sn . In section 3, we will show that each such representation is irreducible and, in fact, the collection of Vλ ’s accounts for every irreducible representation of Sn . Finally, in section 4, we will discuss how these representations can be used to give a large class of irreducible representations of the general linear group GLk . For more details, please refer to Chapters 4, 6, and 15 of Representation Theory: A First Course by William Fulton and Joe Harris. 2. Group algebras, Young diagrams, Young tableaux, and Young symmetrizers Our goal is to describe the irreducible representations of Sn . We first define the group algebra CSn to be the set of all finite formal sums of the form X zσ eσ , zσ ∈ C, σ∈Sn where the eσ are basis elements indexed by the σ ∈ Sn . The group algebra is equipped with a natural addition and multiplication which work as follows: P P P i) Pzi eσi + Pyi eσi = (zi + yi )eσi P ii) ( zi eσi ) yj eσj = zi yj eσi σj . i,j For example, in the case of CS3 we have −3e(12) + 7e(132) + 5e(12) + 2e(132) = 2e(12) + 9e(132) , and (2e1 − 3e(12) )(e(123) + 5e(12) ) = 2e1(123) + 10e1(12) − 3e(12)(123) − 15e(12)(12) = 2e(123) + 10e(12) − 3e(23) − 15e1 . We can also view CSn as a complex vector space with basis elements indexed by the eσ ’s. There is a natural linear action of CSn (the group algebra) on CSn 1 2 SEAN MCAFEE (the vector space) given by the multiplication rule above. In other words, we have a representation ρ : Sn → GL(CSn ) ∼ = GL(Cn! ), called the regular representation of Sn . The rest of this section will be devoted to describing certain elements cλ of CSn (the group algebra), called Young symmetrizers. The images of these elements in CSn (the vector space) will be subspaces Vλ of CSn that are invariant under the action of Sn . Thus we will arrive at representations ρλ : Sn → GL(Vλ ) which, we will see in section 3, are irreducible. Recall that a partition λ = λ1 + · · · + λk , (λ1 ≥ · · · ≥ λk ) of an integer n gives rise to a collection of stacked squares called a Young diagram; where the length of each row is given by λi . For example, the partition 4 + 2 + 2 + 1 of n = 9 induces the Young diagram . By filling in the boxes of a Young diagram in any way with one each of the numbers 1, ..., n, we get what is known as a Young tableau. For our purposes, the arrangement of numbers will not matter (this is not obvious and takes some work to prove), so we will fill in our Young diagrams in the natural way, starting with 1 in the upper left and increasing by 1 as we move from left to right and top to bottom: 1 2 3 4 5 6 7 8 9 . Given such a tableau, we define Pλ = {σ ∈ Sn | σ preserves each row } Qλ = {σ ∈ Sn | σ preserves each column } . We then use these sets to define elements aλ , bλ in the group algebra CSn : X aλ := eσ σ∈Pλ bλ := X sgn(σ)eσ . σ∈Qλ Finally, we define the Young symmetrizer: cλ := aλ · bλ . For an easy example, consider the partition 3 = 2 + 1; this gives us the following Young tableau: 1 2 3 . 3 We then have Pλ = {1, (12)} and Qλ = {1, (13)} , so aλ = e1 + e(12) and bλ = e1 − e(13) , and cλ = (e1 + e(12) )(e1 − e(13) ) = e1 + e(12) − e(13) − e(132) . As discussed above, cλ acts on the vector space CS3 ; we take this action to be on the right. The matrix of this action (using the ordered basis {1, (12), (13), (23), (123), (132)}) is 1 1 −1 0 1 0 1 1 0 −1 0 −1 −1 −1 1 0 1 0 . 0 0 −1 1 −1 1 0 0 1 −1 1 −1 −1 −1 0 1 0 1 The first and third columns form a basis for the column space of this matrix, thus the image Vλ = CSn cλ of cλ is the span of e1 + e(12) − e(13) − e(132) and − e1 + e(13) − e(23) + e(123) . Assuming for the moment that we have proven each such Vλ is irreducible, this gives us a two dimensional irreducible representation of S3 , i.e. Vλ is equivalent to the standard representation. As a final remark, notice that any element d ∈ CSn gives us a subspace CSn d of CSn which is invariant under the (left) action of CSn . Thus the Young symmetrizers cλ are not special in this respect; what does make them special is that the CSn invariant subspaces which they determine will turn out to be irreducible under the CSn -action (and, actually, unique for each partition λ). Exercise 2.1. Explicitly show that the representation above is equivalent to the standard representation of S3 , where S3 acts on the subspace of R3 spanned by the vectors e1 − e2 , e2 − e3 by permuting indices. Exercise 2.2. Show that the Young diagrams 1 2 3 and 1 2 3 give rise to the trivial and alternating representations of S3 , respectively. Exercise 2.3. Show that, for Sn , the partitions (n) and (1 + 1 + · · · + 1) give rise to the trivial and alternating representations of Sn , respectively. Exercise 2.4. Show that, for arbitrary n, the partition ((n − 1) + 1) gives rise to the standard representation of Sn . 4 SEAN MCAFEE 3. proof that CSn cλ is irreducible We will prove the following: Theorem 3.1. Given Sn , let λ be a partition of n. Let Vλ be the subspace of CSn spanned by the Young symmetrizer cλ (so Vλ = CSn cλ ). Then i) Vλ is an irreducible CSn -module (i.e. Vλ is an irreducible representation of Sn ). ii) If λ, µ are distinct partitions of n, then Vλ ∼ 6 Vµ . = iii) The Vλ account for all irreducible representations of Sn . The objects involved are combinatorial in nature, so the proof relies on a couple unavoidably ugly lemmata. The proofs of these will be omitted, but can be found in chapter 4 of Fulton and Harris. Lemma 3.2. For all x ∈ CSn , cλ xcλ is a scalar multiple of cλ . Lemma 3.3. If λ 6= µ, then cλ CSn cµ = 0. Assuming the above have been shown, we are ready to prove the theorem: Proof of theorem 3.1: i) Let Vλ = CSn cλ for a given Young symmetrizer cλ . By Lemma 3.2, we have that cλ Vλ ⊆ Ccλ . Let W be a nonzero subrepresentation of Vλ . We will show that W = Vλ . First, we claim that both cλ Vλ , cλ W are nonzero. To see this, suppose that cλ Vλ = 0. Then Vλ Vλ = CSn (cλ Vλ ) = 0. As vector spaces, there exists a projection π : CSn → CSn cλ that commutes with the action of Sn (this can be obtained by taking the projection induced by cλ and averaging over the action of Sn ). This projection can be described as right multiplication on CSn (the group algebra) by an element x ∈ CSn by letting x := π(1). This x must lie in Vλ , since x = 1·x. By the definition of a projection, x = x2 ∈ Vλ Vλ = 0, thus x = 0, a contradiction since the nonzero cλ itself lies in CSn cλ . Thus we must have cλ Vλ 6= 0. Similarly, we show cλ W 6= 0. So we have W is a subspace of Vλ , cλ Vλ ⊆ Ccλ , and cλ W 6= 0; thus we must have cλ W = Ccλ . Therefore, Vλ = CSn cλ = CSn (Ccλ ) = CSn (cλ W ) ⊆ W, where the inclusion on the right follows from the fact that W is a subrepresentation of Vλ , i.e. it is invariant under the action of CSn . Thus, we have Vλ = W , completing the proof that Vλ is irreducible. ii) Let λ, µ be distinct partitions of n, and let Vλ , Vµ their corresponding representations. By the above, we have that cλ Vλ = Ccλ 6= 0. By Lemma 3.3, we have that cλ Vµ = cλ CSn cµ = 0. Therefore, Vλ and Vµ cannot be isomorphic as CSn modules, proving ii). 5 iii) Each partition λ of n corresponds to a distinct conjugacy class of Sn . We have shown in ii) that the Vλ determined by such partitions are all inequivalent. Thus, since the number of conjugacy classes of a finite group is equal to the number of irreducible representations, we have accounted for all of the irreducible representations of Sn . 4. irreducible representations of GLn (C) Equipped with our construction of the irreducible representations of Sn , it is actually fairly simple to describe all irreducible, finite-dimensional, rational representations of the general linear group GLn (C). By rational, we mean the following: given a finite dimensional representation ρ : GL(n, C) → GL(m, C), we say that the representation is polynomial if, for each A ∈ GL(n, C), we have that ρ(A) has entries which are polynomial in the entries of A. We say that the representation is rational if, for each A ∈ GL(n, C), we have that ρ(A) has entries which are rational polynomials in the entries of A. Example 4.1. For any GL(n, C), we have the standard representation, given by ρ : GL(n, C) → GL(n, C) A 7→ A. This is clearly a polynomial (and thus rational) representation. Exercise 4.2. Show that the standard representation of GL(n, C) is irreducible. Exercise 4.3. Show that the representation ρ(g) = (g −1 )T is a representation of GL(n, C) on C2 which is rational, but not polynomial. In studying the irreducible representations of Sn , we started by observing that any σ ∈ Sn acts (by right multiplication) on the group algebra CSn by permuting the basis elements. We then used linearity to extend this action to a CSn action on CSn . The image of this action under the element cλ ended up being an irreducible representation Vλ of Sn . The process for studying the irreducible representations of GL(n, C) will be similar. We will start by giving a sketch of the construction, then we will proceed to fill in some of the details. Consider GL(n, C) acting on the vector space V = Cn by matrix multiplication. For any positive integer d, this multiplication can be extended to the dth tensor power of V , written V ⊗d . For such a fixed d, we can look at a partition λ = (d1 , ..., dk ) and (just like in the Sn case) construct the Young symmetrizer cλ . This symmetrizer acts on the right of V ⊗d in the same way that it did on the group algebra CSn : by permuting elements. The image V ⊗d cλ will turn out to be a GL(n, C)-invariant subspace of V ⊗d . We write Sλ V to denote this subspace, and call the map V Sλ V the Schur functor. These Sλ V ’s will turn out to be precisely the irreducible polynomial representations of GL(n, C). Let’s begin to unpack the above paragraph by looking at the action of GL(n, C) on V ⊗d . For a basis vector v1 ⊗ v2 ⊗ · · · ⊗ vd of V ⊗d , and for g ∈ GL(n, C), we define g · (v1 ⊗ v2 ⊗ · · · ⊗ vd ) := gv1 ⊗ gv2 ⊗ · · · ⊗ gvd . By linearity, we can extend this to an action on any linear combination of such basis vectors. 6 SEAN MCAFEE Exercise 4.4. Check that the above defines a representation of GL(n, C) on V ⊗d . Now, let’s analyze the right action of Sd on V ⊗d . For σ ∈ Sd and v1 ⊗v2 ⊗· · ·⊗vd a basis element of V ⊗d , define (v1 ⊗ v2 ⊗ · · · ⊗ vd ) · σ := vσ(1) ⊗ vσ(2) ⊗ · · · ⊗ vσ(d) . Again, by linearity we can extend this to a right action of the group algebra CSd on all of V ⊗d . This action clearly commutes with the left action of GL(n, C) defined above. We define two important subspaces of V ⊗d , called the symmetric tensors Vd Symd V and the alternating tensors Altd V (this is often denoted V , and its P elements are written as v1 ∧ v2 ∧ · · · ∧ vd ): ( ) X d Sym V := vσ(1) ⊗ vσ(2) ⊗ · · · ⊗ vσ(d) σ∈Sd ( d Alt V := ) X sgn(σ)vσ(1) ⊗ vσ(2) ⊗ · · · ⊗ vσ(d) . σ∈Sd For example, if d = 2, then (for arbitrary vectors v1 , v2 ∈ V ) v1 ⊗ v2 + v2 ⊗ v1 ∈ Sym2 V, and v1 ⊗ v2 − v2 ⊗ v1 ∈ Alt2 V. Intuitively, we have that Symd V is the set of vectors in V ⊗d that are not changed by permuting their components, and Altd V is the set of vectors in V ⊗d that change sign whenever two components are switched (this is due to Sd being generated by transpositions, which have negative signature). Exercise 4.5. Verify that Symd V and Altd V are invariant subspaces of V ⊗d under the action of CSd . Under the (right) action of CSd , we can redefine Symd V and Altd V as ! X d ⊗d Sym V := V · eσ σ∈Sd ! d Alt V := V ⊗d · X sgn(σ)eσ . σ∈Sd Exercise 4.6. Check that the above definitions coincide with our original definitions of Symd V and Altd V . Exercise 4.7. Check that Symd V and Altd V are invariant under the (left) action of GL(n, C). Exercise 4.7 gives two examples of a more general (and extremely useful) phenomenon: for any element c of the group algebra CSd , we have that V ⊗d c is invariant under the action of GL(n, C). Indeed, since the actions of GL(n, C) and CSd commute, then, for any g ∈ GL(n, C), v ∈ V ⊗d , g(vc) = (gv)c ∈ V ⊗d c. 7 This gives us an entire family of invariant (and potentially irreducible) subspaces of V ⊗d obtained by simply choosing an element at random from CSd . As mentioned before, the only elements we need will be the Young symmetrizers cλ . There is another thing to point out about Symd V and Altd V : they are the images V ⊗d cλ = Sλ V of the Young symmetrizers corresponding to the partitions (d) and (1, 1, ..., 1), respectively. As subspaces of V ⊗d , they have zero intersection, i.e. they are direct summands in the decomposition of V ⊗d : V ⊗d = Symd V ⊕ Altd V ⊕ (Symd V ⊕ Altd V )⊥ . Notice the relation between this and the irreducible spaces V(d) and V(1,1,...,1) in CSd in our representation of Sd . There is a very simple correspondence between the decomposition of CSd into irreducible representations Vλ of Sd and the decomposition of V ⊗d into irreducible representations Sλ V of GL(n, C). The above discussion and more is summarized in the following theorem, which we will not prove here. For details, refer to pages 84-87 of Fulton and Harris. Theorem 4.8. i) Let k = dim V . Then Sλ V = 0 if λk+1 6= 0; that is, if the Young diagram corresponding to λ has more than k rows. ii) Let mλ be the dimension of the irreducible representation Vλ of Sd corresponding to a partition λ of d. Then M (Sλ V )⊕mλ . V ⊗d ∼ = λ iii) Each Sλ V is an irreducible polynomial representation of GL(n, C). iv) All rational representations of GL(n, C) can be described by a tensor product Sλ V ⊗ (det)−k , where z ∈ Z>0 and det is the (one dimensional) determinant representation of GL(n, C). Example 4.9. Consider GL(2, C) with its natural representation on V = C2 . Let d = 3, and consider the partition λ = (1, 1, 1). Then we have c(1,1,1) = e1 − e(12) − e(13) − e(23) + e(123) + e(132) . Then, for v1 ⊗ v2 ⊗ v3 ∈ V ⊗3 , (v1 ⊗ v2 ⊗ v3 ) · c(1,1,1) = v1 ⊗ v2 ⊗ v3 − v2 ⊗ v1 ⊗ v3 − v3 ⊗ v2 ⊗ v1 − v1 ⊗ v3 ⊗ v2 + v3 ⊗ v1 ⊗ v2 + v2 ⊗ v3 ⊗ v1 . Since V is two dimensional, we may write v3 = av1 + bv2 ; the above then becomes v1 ⊗ v2 ⊗ av1 + v1 ⊗ v2 ⊗ bv2 − v2 ⊗ v1 ⊗ av1 − v2 ⊗ v1 ⊗ bv2 − av1 ⊗ v2 ⊗ v1 − bv2 ⊗ v2 ⊗ v1 − v1 ⊗ av1 ⊗ v2 − v1 ⊗ bv2 ⊗ v2 + av1 ⊗ v1 ⊗ v2 + bv2 ⊗ v1 ⊗ v2 + v2 ⊗ av1 ⊗ v1 + v2 ⊗ bv2 ⊗ v1 . If we use the properties of tensors to move the constants a, b to the front of each monomial, we see that the above sum collapses to 0, illustrating part i) of the theorem. 8 SEAN MCAFEE Example 4.10. We use the setup of the previous example, this time with partition λ = (2, 1). Then we have c(2,1) = 1 + e(12) − e(13) − e(132) . Thus, for v1 ⊗ v2 ⊗ v3 ∈ V ⊗3 , (v1 ⊗ v2 ⊗ v3 ) · c(2,1) = v1 ⊗ v2 ⊗ v3 + v2 ⊗ v1 ⊗ v3 − v3 ⊗ v2 ⊗ v1 − v3 ⊗ v1 ⊗ v2 . We have an embedding of Alt2 V ⊗ V into V ⊗ V ⊗ V given by (v1 ∧ v3 ) ⊗ v2 7→ v1 ⊗ v2 ⊗ v3 − v3 ⊗ v2 ⊗ v1 . This realizes the image of c(2,1) as the subspace of Alt2 V ⊗ V spanned by vectors of the form (v1 ∧ v3 ) ⊗ v2 + (v2 ∧ v3 ) ⊗ v1 . These vectors (check this!) are the kernel of the canonical map φ : Alt2 ⊗V → Alt3 V, (v1 ∧ v2 ) ⊗ v3 7→ v1 ∧ v2 ∧ v3 , thus S(2,1) V = Ker φ. By part ii) of the theorem, and by our discussion of S3 in section 2, we have the decomposition V ⊗V ⊗V ∼ = Sym3 V ⊕ Alt3 V ⊕ Ker φ ⊕ Ker φ, where the two copies of Ker φ are due to the fact that CS3 c(2,1) is a two-dimensional (irreducible) representation of S3 , i.e. m(2,1) = 2. To summarize this section: we can parametrize the irreducible rational representations of GL(n, C) by a triple (d, λd , z), where d is any positive integer, λd is a partition of d whose Young diagram has no more than n rows, and k is any positive integer. The resulting representation is the tensor product Sλd V ⊗ (det)−k . There is much, much more to say about representations of GL(n, C) than is contained in this paper. For instance, the use of Schur polynomials to describe the characters of Sλ V , or the decomposition of the tensor product Sλ V ⊗ Sµ V , or the connection between the representations of GL(n, C) and those of SL(n, C) and U (n). We encourage the reader to investigate these on their own.