I've recently been reviewing some linear algebra, and came across the following problem:

How many similarity classes of $A \in M_{6}(\mathbb{C})$ are there such that $A$ has exactly one two-dimensional invariant subspace?

I'd like to try to understand this using the Jordan Canonical Form, or Rational Canonical Form. Here's what I've considered so far:

Thinking about eigenvalues:

Since there is exactly one 2-dimensional invariant subspace, our matrix $A$ cannot be diagonalizable, thus (since we are working over an algebraically closed field) the minimal polynomial of $A$ must contain at least one repeated root. If we have three or more distinct eigenvalues with one dimensional eigenspaces, there will be at least two 2-dimensional invariant subspaces. I conclude that the characteristic and minimal polynomials are of the forms:

$\chi(A) = (x-\alpha_{1})^{a_{1}}(x-\alpha_{2})^{a_{2}}(x-\alpha_{2})^{a_{3}}(x-\alpha_{4})^{a_{4}}$,

$m(A) = (x-\alpha_{1})^{b_{1}}(x-\alpha_{2})^{b_{2}}(x-\alpha_{2})^{b_{3}}(x-\alpha_{4})^{b_{4}}$

where $a_{1} \geq b_{1} \geq 2$, $a_{2} \geq 2$ $ a_{i} \geq b_{i} \geq 0$, $\sum_{i=1}^{i=5} a_{i} = 6$, and the $\alpha_{i}$ are distinct.

Thinking about invariant factors:

It must also be the case that a maximum of two of the invariant factors are linear polynomials, since any more would yield at least two 2-dimensional invariant subspaces in the rational canonical form. Given that the characteristic polynomial is degree 6, and (discounting multiplicity) the minimal and characteristic polynomials must have the same roots, I suppose I can count all of the possibilities satisfying the properties above. But is any list of invariant factors satisfying everything I've said actually one with exactly one 2-dimensional subspace? Or have I missed some conditions on the invariant factors implied by this property?

In general, these seem like some hairy, ad hoc considerations I'm making, and I'm hoping to find a more organized approach to this and similar problems. For example, if we generalize the problem to given $A \in M_{n}(\mathbb{C})$ with $l_{k}$ k-dimensional invariant subspaces, classify the similarity classes of such matrices, is there a good method?


Each Jordan block contains exactly one eigenvector, this implies that there are at most two Jordan blocks. If the Jordan block is not a $1\times 1$ matrix, then it contains exactly one two dimensional invariant subspace. This two observations easily imply that the matrix should be similar to the $6\times 6$ Jordan block, i.e., there are $\mathbb C$ similarity classes (i.e., each class can be identified with the eigenvalue $\lambda\in\mathbb C$)

  • $\begingroup$ Ah, that is much simpler than I expected, thanks! $\endgroup$ – user1348 Aug 30 '16 at 1:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.