Let $A$ be a square matrix and $I$ the $k \times k$ identity matrix. Then the identity $$ \det(A \otimes I) = \det(A)^k,$$ holds as can be seen from a general result on the determinant of block matrices.

Since the structure of $A \otimes I$ is quite beautiful I am wondering

Can someone find a short proof of the above equality?

  • $\begingroup$ isn't it the Kroenecker product (and not Hadamard)? Because with the Hadamard product, this result is wrong. However with the Kroenecker product, you can show that the eigenvalues of $A \circ I$ and $A$ coincides (up to multiplicity). $\endgroup$ – Surb Nov 20 '14 at 20:56
  • $\begingroup$ @Surb Err you're right I mean the Kronecker product! $\endgroup$ – Jernej Nov 20 '14 at 21:01
  • $\begingroup$ The simplest proof proceeds as follows: Let $K$ be your base ring. Let $n$ be the size of $A$. There is an isomorphism $K^n \otimes K^k \to \bigoplus\limits_{i=1}^k K^n$ of $K$-modules, and under this isomorphism the endomorphism $A \otimes I$ of $K^n \otimes K^k$ corresponds to the endomorphism $\bigoplus\limits_{i=1}^k A$ of $\bigoplus\limits_{i=1}^k K^n$ (where we identify matrices with endomorphisms in a hopefully clear way). Now, you know what $\det\left(\bigoplus\limits_{i=1}^k A\right)$ is? (What is $\det\left(A\oplus B\right)$ ?) $\endgroup$ – darij grinberg Sep 19 '15 at 13:05

Consider the canonique basis $e_1,\ldots,e_k$ of $\Bbb R^k$. Furthermore let $\lambda_1,\ldots,\lambda_n$ be the eigenvalues of $A \in \Bbb R^{n\times n}$ and $v_1,\ldots,v_n$ their associated eigenvectors. Then $v_i \otimes e_j$ is an eigenvector of $A \otimes I$ associated to the eigenvalue $\lambda_i$ for every $j = 1,\ldots,k$. Since $v_i \otimes e_l$ and $v_i \otimes e_j$ and linearly independent for every $l \neq j$. It is clear that the eigenvalues of $A \otimes I$ are exactly $\underbrace{\lambda_1,\ldots,\lambda_1}_{k \text{ times}},\ldots,\underbrace{\lambda_n,\ldots,\lambda_n}_{k \text{ times}}$ it follows that $\det(A\otimes I) = \det(A)^k$.

  • $\begingroup$ $n$ linearly independent eigenvectors don't always exist. In the general case you would have to extend the base field to an algebraic closure (no, $\mathbb{R}$ is not enough), WLOG assume that the matrix is diagonalizable (that, or find a trigonalization), etc. All in all, not a particularly short proof. $\endgroup$ – darij grinberg Sep 19 '15 at 13:02

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.