3
$\begingroup$

Let $V$ be an $n$-dimensional real vector space, and let $1<k<n$ be fixed. Given an automorphism $A \in \text{GL}(V)$, consider its $k$-th exterior power $\bigwedge^k A \in \text{GL}(V)$.

Suppose $\bigwedge^k A$ admits an eigenvector. (equivalently, $\bigwedge^k A$ admits a non-zero eigenvalue).

Does $A$ has a $k$-dimensional invariant subspace?

If the eigenvector $v$ of $\bigwedge^k A$ is decomposable, then the answer is positive:

Write $v=v_1 \wedge v_2 \dots \wedge v_k$; Then $$\bigwedge^k A(v_1 \wedge v_2 \dots \wedge v_k)=\lambda v_1 \wedge v_2 \dots \wedge v_k \Rightarrow \text{span}(Av_1,\dots,Av_k)=\text{span}(v_1,\dots,v_k),$$

so $\text{span}(v_1,\dots,v_k)$ is $A$-invariant.

However, I am not sure that every eigenspace $\bigwedge^k A$ of should be decomposable.

This question looks somewhat related to this nice question, which is still not fully answered.

$\endgroup$
2
$\begingroup$

Nice question. First, note that over $\mathbb{C}$, any operator can be represented with respect to an appropriate basis by an upper triangular matrix. This implies that any operator $A$ has invariant subspaces of all possible dimensions so the question is not interesting over $\mathbb{C}$.

To construct a counterexample over $\mathbb{R}$, I will use the following observations:

  1. If $n$ is even and the characteristic polynomial of $A$ has no real roots, then $A$ has no odd-dimensional $A$-invariant subspaces. The reason is that if you restrict $A$ to an odd-dimensional $A$-invariant subspace, you get an operator which must have an eigenvector (with a real eigenvalue), contradicting the fact that all the roots of the characteristic polynomial of $A$ aren't real.
  2. If the (possibly complex) roots of the characteristic polynomial of $A$ are $(\lambda_i)_{i=1}^n$ (with multiplicity) then the roots of the characteristic polynomial of $\Lambda^k(A)$ are $(\lambda_{\alpha})$ where $\alpha = (i_1 < \dots < i_k)$ runs over all possible multi-indices and $\lambda_{\alpha} := \lambda_{i_1} \dots \lambda_{i_k}$. To see this, assume first that $A$ is a complex operator and choose an ordered basis $(e_i)_{i=1}^n$ with respect to which $A$ is represented by an upper triangular matrix with $$Ae_i = \lambda_i e_i \mod \operatorname{span} \{ e_j \}_{j < i}. $$ Then $\Lambda^k(A)$ is represented with respect to the induced ordered basis $(e_{\alpha})$ (where the order on the multi-indices is the lexicographical one) by an upper triangular matrix with $$ \Lambda^k(A)(e_\alpha) = \lambda_{\alpha} e_{\alpha} \mod \operatorname{span} \{ e_{\beta} \}_{\beta < \alpha}. $$ The result for real operators follows by complexification using the fact that exterior power and complexification commute.

Now, let $\theta = \frac{2\pi}{3}$ and set $\alpha = e^{i\theta}$. Consider the operator $A \colon \mathbb{R}^6 \rightarrow \mathbb{R}^6$ which is represented with respect to the standard basis by the block diagonal matrix $$ \begin{pmatrix} \cos \theta & -\sin \theta & 0 & 0 & 0 & 0 \\ \sin \theta & \cos \theta & 0 & 0 & 0 & 0 \\ 0 & 0 & \cos \theta & -\sin \theta & 0 & 0 \\ 0 & 0 & \sin \theta & \cos \theta & 0 & 0 \\ 0 & 0 & 0 & 0 & \cos \theta & -\sin \theta \\ 0 & 0 & 0 & 0 & \sin \theta & \cos \theta \end{pmatrix}. $$

The characteristic polynomial of $A$ is $$ (z - \alpha)^3(z - \overline{\alpha})^3 = (z^2 - (2 \Re{\alpha})z + |\alpha|^2)^3 = (z^2 + z + 1)^3 $$ with roots $$ \alpha, \overline{\alpha}, \alpha, \overline{\alpha}, \alpha, \overline{\alpha}. $$ The roots aren't real so $A$ doesn't have a three-dimensional invariant subspace. However $\alpha^3 = 1$ is a real root of the characteristic polynomial of $\Lambda^3(A)$ (of multiplicity two) so $\Lambda^3(A)$ has two linearly independent eigenvectors which are necessarily indecomposable.


Remark: One can show using primary decomposition that if a real operator has a real eigenvalue then it has invariant subspaces of all possible dimensions. Hence, counterexamples are possible only in even dimensions. It is a nice exercise to see why you can't have a counterexample in dimension four so this is a minimal counterexample in terms of dimension.

$\endgroup$

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.