S. D. V. Stephens

notes / Studies in Algebra and Group Theory / chapter 3: Linear Maps and Operators / From my book

B7.3 Invariant Subspaces

Understanding an operator often begins with finding subspaces on which its behavior is simpler.

Definition 7.3.1 (Invariant subspace). A subspace is invariant under (or -invariant) if i.e., for all . In that case the restriction , , is a linear operator on .

Example 7.3.2 (Basic examples of invariant subspaces). For any :

  • and are always -invariant.
  • is -invariant: if , then .
  • is -invariant: .
  • More generally, and are -invariant for all (a special case of the next proposition).

Proposition 7.3.3 (Kernels and images of commuting operators). Let with . Then and are -invariant. In particular, and are -invariant for every .

Proof. If , then , so . If , then . The last sentence follows because commutes with (Proposition 7.2.2). ∎

7.3.1 Block Diagonal and Block Triangular Matrices

Invariant subspaces lead to simpler matrix representations.

Theorem 7.3.4 (Direct sum of invariant subspaces). Suppose where both and are -invariant. Choose bases of and of , which together form a basis of . Then the matrix of with respect to is block diagonal: where and . The same holds for a decomposition into any number of invariant subspaces.

Proof. For : (by -invariance), so is a linear combination of only, with the coefficients of . For : , so is a linear combination of only. This gives the block structure. The case of summands is the same. ∎

Note. Conversely, if the matrix of in a basis is block diagonal, then the spans of the corresponding groups of basis vectors are invariant subspaces whose direct sum is . The goal of much of spectral theory is to decompose into a direct sum of invariant subspaces on which acts simply.

When only one of the two subspaces is invariant, we still get a simplification, and the quotient space of Section 7.1 describes the second diagonal block.

Proposition 7.3.5 (Induced operator on a quotient). Let be -invariant, and let be the quotient map.

  1. There is a unique linear map with , namely :

    Diagram
  2. Let be finite-dimensional, let be a basis of , and extend it to a basis of . Then is a basis of , and where and . Such a matrix is called block upper triangular.

Proof. (a) The linear map vanishes on , because . By the universal property (Theorem 7.1.7) it factors uniquely as .

(b) The cosets span , since is surjective and ; there are of them (Theorem 7.1.5), so they form a basis. For , is a combination of , which gives the first columns. For , write ; applying gives , so the lower right block is . The upper right block may be nonzero. ∎