These notes develop abstract operator arguments that go beyond computing an eigenvector. Let be finite-dimensional over a field unless an inner product is explicitly introduced.
Definition 2.3.1
(Invariant subspace, restriction and quotient operator)
A subspace is -invariant when . Invariance makes the restriction an operator in its own right. It also defines the quotient operator by .
Invariance does not by itself imply that a chosen complement is invariant.
The quotient records what happens outside the subspace
The quotient operator is well defined: if , then , so . Linearity follows from linearity of . Thus no particular complement is needed to describe the action outside .
Theorem 2.3.2
(Characteristic polynomial through an invariant subspace)
If is -invariant, then .
Proof
(Block triangular form from an adapted basis)
Choose a basis of and extend it to a basis of . In this basis,
The lower-left block vanishes precisely because the images of the remain in . The matrix represents and represents in the quotient basis formed from the cosets of the added vectors. The block determinant identity therefore yields
This does not assert that is block diagonal: the block measures the part of the images of the added vectors that falls into . The factorization also does not imply that each eigenvector of is represented by an eigenvector of .
Counterexample mode
An invariant subspace without an invariant complement
Claim: if is -invariant, then for some -invariant . On , set and . Then and are both the identity and . A -invariant complement would be a line with and ; the second coordinate gives and then the first gives , a contradiction. Equivalently , so the only eigenvectors lie in , and the nonzero quotient eigenvector cannot be represented by an eigenvector. Although as vector spaces, that complement is not -invariant. Repair: extra structure is needed, such as a unitary (next section) or a diagonalizable .
Why a unitary operator has an invariant orthogonal complement
Now suppose is a finite-dimensional complex inner-product space and is unitary, meaning .
Theorem 2.3.3
(Orthogonal complement of a unitary-invariant subspace)
If is unitary on a finite-dimensional inner-product space and is -invariant, then is also -invariant, and .
Proof
(From injectivity to and then to )
If is -invariant, then . Since is injective, ; consequently , and also . For and ,
Thus . An orthonormal basis adapted to makes the matrix of block diagonal. If has characteristic polynomial and has characteristic polynomial , therefore .
Proof X-ray
Where finite dimension and unitarity are used
The calculation is easy; the whole difficulty is to know that lies in , that is, that is invariant under . That is exactly what the dimension count supplies: an injective map of a finite-dimensional space into itself is onto, so , and applying keeps us in . Unitarity is used twice, once for injectivity and once to identify with .
The finite-dimensional step matters: on the infinite-dimensional space with orthonormal basis , the bilateral shift is unitary and satisfies , but and with , so is not invariant. More generally, invariance of under is exactly what the displayed calculation needs to make invariant under .
Every eigenvector of a restriction is also an eigenvector of , because the restricted map is the same map on its invariant domain. For a general operator the conclusion fails: the preceding has invariant with the standard inner product, but . Invariance alone does not provide an orthogonal block decomposition.
Worked examples and practice
Worked example 2.3.4
(Restriction, quotient and diagonalizability)
Let on , and . Find the restriction and quotient characteristic polynomials. Is invariant? Is diagonalizable?
and , so the restriction matrix is and its characteristic polynomial is . The quotient has basis , which is fixed, so its polynomial is . Here is invariant. However, the -eigenspace satisfies and has dimension one, whereas its algebraic multiplicity is two; is not diagonalizable. An invariant orthogonal complement by itself does not imply diagonalizability.
Worked example 2.3.5
(Eigenvalues of a unitary operator have modulus one)
Let be unitary on a finite-dimensional complex inner-product space and suppose with . Prove and show is -invariant.
Unitarity preserves norms: . On the other hand ; cancellation gives . The line is invariant, so the theorem on invariant orthogonal complements of unitary operators proves invariance of . Equivalently and for ; the scalar may be conjugated under the chosen inner-product convention, but zero remains zero.
Worked example 2.3.6
(Factoring a characteristic polynomial through a triangular matrix)
Let on have matrix and take , which is invariant since . Then is multiplication by , so . In the quotient basis we have and , so has matrix and . The product equals , the product of the diagonal entries' factors of the triangular matrix.
Quick check
(Let on and . Find , and .)
Solution · Answer
, so is multiplication by . Next modulo , so is multiplication by . Hence , which agrees with .
Quick check
(For the same , is invariant?)
Solution · Answer
No: is not a multiple of . Of the two coordinate axes only is invariant.
Quick check
(Let on and . Verify that and are both -invariant.)
Solution · Answer
, so is invariant. Here and , so is invariant; the eigenvalues and have modulus one, as expected for a unitary operator.