Is there a way to relate Eigenvalues to the column space and nullspace of a matrix?

I believe a matrices with different eigenvalues would have a different column spaces and/or nullspace. Is this correct?

I am wondering if you can prove that the Eigenvalues of $A$ and $A^T$ are equal using properties of column spaces and nullspaces.

My thinking is:
If you transform a matrix $A$ into $B$, if the row space of $B$ is orthogonal to the nullspace of $A$, and the column space of $B$ is orthogonal to the left nullspace of $A$, then matrices $A$ and $B$ have the same eigenvalues.

  • 1
    $\begingroup$ Well, for one, you know that if $\lambda = 0$ is an eigenvalue, then the matrix has a non-trivial null space (and, conversely, if the matrix has a non-trivial null space, then $0$ is an eigenvalue). This follows because $\lambda = 0$ is an eigenvalue implies there exists a non-zero vector $x$ such that $Ax=0$. $\endgroup$ – user35959 Dec 14 '12 at 19:54
  • 4
    $\begingroup$ The dimension of the nullspace corresponds to the multiplicity of the eigenvalue 0. In particular, A has all non-zero eigenvalues if and only if the nullspace of A is trivial ($null(A)=\{0\})$. You may then use the fact that $dim(Null(A))+dim(Col(A))=dim(A)$ to deduce that the dimension of the column space of A is the sum of the multiplicities of the non-zero eigenvalues. $\endgroup$ – Dan Rust Dec 14 '12 at 19:59
  • $\begingroup$ @DanRust sorry for the delayed comment, but now I came up with the same question. So, what about matrix $A = [2 -2;2 -2]$, it has $\mathcal{N}(A)=\{c[1, 1]^T | c\in \mathbb{R} \}$ , $\mathbb{dim}\mathcal{N}(A)=1$ (a line), but the multiplicity of zero-eigenvalues is two ($\lambda_{1,2}=0$). Could you please give a comment on that? Am I right? $\endgroup$ – Thoth Sep 25 '16 at 9:53
  • $\begingroup$ @Thoth Geometric multiplicity, not algebraic multiplicity. $\endgroup$ – Dan Rust Sep 25 '16 at 10:58
  • $\begingroup$ @DanRust in order to complete my understanding, "the dimension of the nullspace corresponds to the (*geometric) multiplicity ( not the multiplicity of the eigenvalue 0 but the multiplicity of the eigen-vectors that correrspond to zero eigen-values.)". The geometric multiplicity equals to $\mathbb{dim}\mathcal{N}(A)$. So the $\mathbb{dim}\mathcal{N}(A)$ does not offer enough information for the multiplicity of eigen-values. Am I right? You comments and discussion valuable in understanding. $\endgroup$ – Thoth Sep 25 '16 at 13:53

Matrices with different eigenvalues can have the same column space and nullspace. For a simple example, consider the real 2x2 identity matrix and a 2x2 diagonal matrix with diagonals 2,3. The identity has eigenvalue 1 and the other matrix has eigenvalues 2 and 3, but they both have rank 2 and nullity 0 so their column space is all of $\mathbb{R}^2$ and their nullspace is $\{0\}$.

This is also probably a negative answer to your question about the transpose - the column space and nullspace don't contain enough information about the eigenvalues.

On the other hand, eigenvalues are certainly related to the nullspace of $A-\lambda I$, where $\lambda$ is an eigenvalue of $A$. Namely, every eigenvector must lie in the nullspace of this matrix.

  • $\begingroup$ OK. Looks like I was overthinking things. Thanks! $\endgroup$ – zaz Dec 18 '12 at 19:42

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.