What one basis must accomplish
Definition 5.1.1
(Simultaneous diagonalizability)
Let be linear operators on a finite-dimensional vector space over a field . They are simultaneously diagonalizable over when has a basis consisting of vectors that are eigenvectors of and also eigenvectors of (a common eigenbasis).
In such a basis both matrices are diagonal, so . Commutation alone is not enough: each operator must also be diagonalizable over the stated field. The converse is the theorem of this unit. Field conditions matter: a real operator with only nonreal eigenvalues is not diagonalizable over merely because it is diagonalizable over .
Theorem 5.1.2
(Commuting diagonalizable operators share an eigenbasis)
If and are diagonalizable over and , then they are simultaneously diagonalizable over .
Eigenspaces are preserved
If , the empty basis suffices; assume below.
Suppose . Then . Thus maps every eigenspace into itself. Diagonalizing inside each such eigenspace will supply vectors that remain eigenvectors of . What this needs is that a diagonalizable operator stays diagonalizable on an invariant subspace.
Lemma 5.1.3
(Restriction to an invariant subspace)
If is diagonalizable over and is a -invariant subspace, then is diagonalizable over .
Proof
(Splitting a vector with polynomials in B)
Let the distinct eigenvalues of be and set
The denominators are nonzero because the eigenvalues are distinct. On a -eigenvector with eigenvalue , acts as 1 if and as 0 otherwise. An eigenbasis of therefore gives , and each lies in . Since is invariant under , it is invariant under every polynomial in ; hence for ,
Distinct eigenspaces form a direct sum, so these intersections decompose . Taking a basis of each proves the restriction is diagonalizable.
Proof
(Choosing a B-eigenbasis inside each eigenspace of A)
Since is diagonalizable, . By commutation each is -invariant, so by the lemma it has a basis of eigenvectors of . Every vector of lies in , so it is also an eigenvector of . The union of the bases is a basis of the direct sum , and consists of common eigenvectors.
Proof X-ray
Where each hypothesis and the repeated eigenvalues enter
Each hypothesis is spent exactly once. Commutation is used only to show that maps into itself. Diagonalizability of is used only to know that these eigenspaces fill . Diagonalizability of is used only inside the lemma, through the polynomials , which cut an arbitrary vector of into pieces lying in the eigenspaces of while staying in .
Commutation preserves an eigenspace, not each vector within it. When , can move vectors inside it, so we must choose a -eigenbasis there; assuming that the first -eigenbasis already works would skip the essential step. The argument closes by counting: the direct sum decomposition of makes the union of the chosen bases a basis. If were not diagonalizable, the lemma would fail and the pieces could not be found.
Worked examples
Worked example 5.1.4
(A repeated eigenvalue forces a choice of basis)
Over , take
acts as scalar 2 on the first two coordinates, so it commutes with the block of ; both preserve the third coordinate. The standard basis diagonalizes but not , since is not a multiple of . The common eigenvectors
are independent. Indeed , , . In this basis the diagonal entries of are and those of are . The repeated eigenvalue 2 of allows a nontrivial choice of basis within its eigenspace; the displayed vectors make the action of diagonal there.
Worked example 5.1.5
(Distinct eigenvalues leave no freedom)
Let and . Then . The eigenvalues of are and , with eigenvectors and . Each eigenspace is a line, so preserving it means maps the line to itself, and every vector there is an eigenvector of : and . When has distinct eigenvalues, its eigenbasis is automatically a common eigenbasis.
Worked example 5.1.6
(Commuting operators over the wrong field)
Over let and . They commute and is diagonalizable, but has no real eigenvector: gives and , hence , so and then . Thus no common eigenbasis exists over . Over , is diagonalizable with eigenvectors and for the eigenvalues and , and these also diagonalize .
Common mistakes
Common mistake
(Every eigenvector of A is an eigenvector of B)
Commutation shows that maps into itself, not that acts as a scalar there. Take and . They commute, and is an eigenvector of , but is not a multiple of . One must choose the eigenvectors inside with care.
Counterexample mode
Why both hypotheses are needed
Claim: any two commuting operators have a common eigenbasis. Take , which commute. Then , so the only eigenvalue is 1 and its eigenspace is one-dimensional. There is no eigenbasis of at all, hence no common one. Repair: require both operators to be diagonalizable.
Claim: any two diagonalizable operators have a common eigenbasis. Take and . Both are diagonalizable, but , so they cannot share a diagonalizing basis. Repair: require .
Exercises and solutions
Quick check
(In the first worked example, compute .)
Solution · Answer
.
Quick check
(If two operators share an eigenbasis, why do they commute?)
Solution · Answer
Their matrices in that basis are diagonal, and diagonal matrices commute; changing basis preserves equality of the operators.
Quick check
(For , find the dimension of the eigenspace for eigenvalue 1.)
Solution · Answer
, so the eigenspace is and has dimension 1.
Quick check
(If , what does simultaneous diagonalizability of and require?)
Solution · Answer
Exactly diagonalizability of , because every basis diagonalizes .