In spatial statistics, I am trying to deal with the following topic:

I have a real-valued, symmetric, full-rank matrix $\textbf{A}$, say $N \times N$. It's a connectivity matrix, i.e. $a_{ij}$ values are either 1 or 0. Eigenvectors of $\textbf{A}$ are orthogonal (and real). Now, for an empirical analysis, I need to generalize my $\textbf{A}$ matrix into a block-diagonal matrix such as $\mathcal{A}= (I_T \otimes \textbf{A})$ - also real-valued, symmetric and full rank. I assume it also has orthogonal eigenvectors (of a corresponding lenght $NT$.

My question is, are the eigenvectors of $\textbf{A}$ and $\mathcal{A}$ related? E.g. through some basic transformation or "rule"?

I searched this web for questions on eigenvectors in block diagonal matrices, but did not find an answer for this one... Thank you.

  • 1
    $\begingroup$ Could you clarify what you mean by $\mathcal A = (I_T \otimes \mathbf A)$? Is $\otimes$ the Kronecker product? Is $I_T$ an identity matrix? $\endgroup$ Commented Nov 19, 2018 at 23:24
  • $\begingroup$ Sorry, yes, it's a Kroenecker product and identity matrix. $\endgroup$
    – Tomas
    Commented Nov 19, 2018 at 23:25

1 Answer 1


In general: if $P$ and $Q$ are diagonalizable, then whenever $Px = \lambda x$ and $Qy = \mu y$, we will have $(P \otimes Q)(x \otimes y) = \mu \lambda (x \otimes y)$. Moreover, we can form an eigenbasis out of the Kronecker products $x \otimes y$.

For your particular example: $I_T \otimes A$ will have the same eigenvalues as $A$, but each will have its multiplicity multiplied by $T$. For any eigenvector $v$ of $A$, the vectors $e_1 \otimes v, \dots, e_n \otimes v$ will be eigenvectors of $I_T \otimes A$ (here $e_1,\dots,e_n$ is the canonical basis; so $e_1 = (1,0,\dots,0)^T$).

Another general point: the eigenvalues of any block-diagonal matrix is simply the union (with multiplicity) of the eigenvalues of each individual block.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .