S. D. V. Stephens

notes / Studies in Algebra and Group Theory / chapter 9: Representation Theory / From my book

B13.3 Schur’s Lemma

Our goal is to find the irreducible representations of a given group and to describe how every representation decomposes into them. The key tool is the following lemma.

Lemma 13.3.1 (Schur’s Lemma). Let and be irreducible representations of , and let be a homomorphism of representations.

  1. Either or is an isomorphism.
  2. If , then for some .

Proof. (i) The kernel is a subrepresentation of , so either , in which case , or and is injective. Similarly is a subrepresentation of , so either , in which case , or and is surjective. So if , it is bijective.

(ii) Since is a nonzero finite-dimensional complex vector space, has an eigenvalue (Theorem 7.5.1). Then is again a homomorphism of representations , and it has nonzero kernel, so it is not an isomorphism. By (i), . ∎

Corollary 13.3.2 (Homomorphisms between irreducibles). If and are irreducible representations of , then

Proof. If , every homomorphism is zero by Schur’s Lemma. If is an isomorphism of representations, then is a linear isomorphism , and by Schur’s Lemma. ∎

Example 13.3.3 (The center acts by scalars). Let be an irreducible representation of and , the center of (Definition 3.1.13). Since commutes with every , the operator commutes with every , that is, . By Schur’s Lemma, for some : every central element acts on every irreducible representation by a scalar. In particular, if is abelian, every element acts on an irreducible by a scalar, so every line in is a subrepresentation and . This is a second proof of the last statement of Proposition 13.1.7.

13.3.1 The Irreducible Representations of

We now find all the irreducible representations of , the smallest nonabelian group. We know three of them from Example 13.1.4: the trivial representation , the alternating representation , and the -dimensional standard representation . What is special about is that there are no others. Throughout, , and , so that , and .

Two facts about do all the work. First, every element of is or for some (these six elements are distinct). So a subspace of a representation that is invariant under and is a subrepresentation, and a linear map between representations that commutes with and is a homomorphism of representations. Second, conjugating a cycle by applies to its entries (Lemma 3.4.5), so ; since , this gives . Consequently, in any representation of , That is, maps the -eigenspace of into the -eigenspace. By Lemma 13.1.5, is diagonalizable in every representation of , with eigenvalues among .

Lemma 13.3.4 (The standard representation of ). Let . Then and form a basis of , and So acts on with eigenvalues and , and with eigenvalues and (with eigenvectors and ). Moreover, is irreducible.

Proof. Both vectors lie in because . Since , and , we get , using . The transposition swaps and , which gives the formula for , and . By , is an eigenvector of for the eigenvalue . Eigenvectors for distinct eigenvalues are linearly independent, and , so is a basis of .

Let be a nonzero subrepresentation. The restriction of to is diagonalizable (Lemma 13.1.5), so contains an eigenvector of . The eigenspaces of on are the lines and , so contains or . Being -invariant, it contains both, and . ∎

Proposition 13.3.5 (The irreducible representations of ). Every irreducible representation of is isomorphic to exactly one of , , . Every representation of is isomorphic to for unique integers , and these are determined by where and denote the operators by which and act on .

Proof. Let be irreducible, and let be an eigenvector of , say . By , is also an eigenvector of , so is invariant under ; it is invariant under because . It is therefore a nonzero subrepresentation, and . In particular, every irreducible representation of has dimension at most .

Case . Then as well. If and were linearly independent, then would be a nonzero vector with and , so would be a subrepresentation of the -dimensional space , contradicting irreducibility. So for some , and because . Then , with acting as and as . Since and , the linear map is an isomorphism if and if .

Case . Then . If , replace by , which is a -eigenvector by and satisfies ; so we may assume . Then and are eigenvectors of for the distinct eigenvalues and , hence a basis of . Define a linear isomorphism by and , with as in Lemma 13.3.4. Both bases are eigenbases of with the same eigenvalues, and swaps the two vectors in each basis, so commutes with and . Hence is an isomorphism of representations, and .

The representations , , are irreducible (Lemma 13.3.4 for ) and pairwise non-isomorphic: has dimension , and acts as on and as on . This proves the first statement.

Now let be any representation. By complete reducibility and the first statement, for some . The operator acts as on and on , and on with eigenvalues and , each with a -dimensional eigenspace. The operator acts as on , as on , and on with eigenvalues and , each with a -dimensional eigenspace. Adding up the dimensions of the eigenspaces summand by summand gives the three formulas. Their left-hand sides depend only on the isomorphism class of , and they determine , then , then ; so , , are unique. ∎

Example 13.3.6 (Decomposing , and ). Let and be the basis of from Lemma 13.3.4, so that , , and swaps and . The basis of consists of eigenvectors of , with eigenvalues (recall ). So and in Proposition 13.3.5. The operator swaps with , and with , so it has eigenvalues , and . Hence : Explicitly, spans a copy of , and spans a copy of .

Similarly, has basis , on which has eigenvalues , while fixes and swaps and . So , and , and . Finally, is spanned by , which multiplies by and sends to ; so . This is to be expected: on the top exterior power , each acts by (Theorem 9.11.3), and , .