I was wondering, given an $n \times n$ square matrix, let function $\det : \left(a_1,a_2,\ldots,a_{n^2}\right) \to \textbf{R}$ give the determinant, where $a_{k}$'s are the entries of the $n \times n$ matrix.

  1. Is this function (determinant) a differentiable kind?

  2. If so, is the derivative continuous? That is, is $d\left(\det\right)$ a continuous function?

  3. Furthermore, if so, to what differentiability class does this $\det$ function belong?

Thanks in advance.

  • 1
    $\begingroup$ All of the answers so far treat the determinant as a polynomial in the matrix entries $(a_{ij})$. However, just as we can take the derivative of $\vec{y}=f(\vec{x})$ as the vector $\vec{x}$ varies, it would be nice to see someone answer the question from the perspective of $$\frac{d}{dA} det(A).$$ $\endgroup$ – Travis Bemrose Apr 24 '14 at 1:59
  • $\begingroup$ @TravisBemrose I calculated the total (or Fréchet) derivative of the determinant in my updated answer, maybe that's what you're looking for. $\endgroup$ – Christoph Apr 24 '14 at 7:42

As others have noted, since $A=(a_{ij})_{i,j=1\dots n}$ has determinant $$ \det A = \sum_{\sigma\in S_n} \epsilon_\sigma\prod_{k=1}^n a_{k,\sigma(k)} $$ which is a polynomial expression in the $a_{ij}$, the map $\det: \mathbb R^{n\times n}\to\mathbb R$ is infinitely differentiable. The first derivative with respect to $a_{ij}$ is calculated as $$ \frac{\partial}{\partial a_{ij}} \det A = (\operatorname{adj} A)_{ji}, $$ where $\operatorname{adj} A$ is the adjugate matrix of $A$.

We can also look at the total derivative (or Fréchet derivative) $\mathrm D\det: \mathbb R^{n\times n}\to L(\mathbb R^{n\times n},\mathbb R)$ which assigns to every $A\in\mathbb R^{n\times n}$ the linear map $\mathrm D \det(A) : \mathbb R^{n\times n}\to \mathbb R$ given by $$ (\mathrm D\det(A))(B) = \sum_{i,j} \left( \frac{\partial}{\partial a_{ij}} \det A\right) b_{ij}= \sum_{i,j} (\operatorname{adj}A)_{ji}b_{ij} = \operatorname{tr}((\operatorname{adj} A)B).$$

For invertible $A$ we can use $A^{-1}=\frac{1}{\det a}\operatorname{adj}A$ to get the expression $$ (\mathrm D\det(A))(B) = \det(A)\operatorname{tr}(A^{-1} B). $$

This allows us to use the chain rule to calculate the derivative of functions like $f(t)=\det(A(t))$ where $A:\mathbb R\to\mathbb R^{n\times n}$ is a differentiable matrix-valued function. By the chain rule, we have \begin{align} f'(t) &= \left(\mathrm Df(t)\right)(1) = \left(\mathrm D(\det\circ A)(t)\right)(1) = \left(\mathrm D \det(A(t)) \circ \mathrm D A(t)\right)(1) \\&= \left(\mathrm D \det(A(t))\right)\left(\mathrm D A(t)(1)\right) = \left(\mathrm D \det(A(t))\right)\left(\frac{\mathrm d A(t)}{\mathrm dt}\right) \\&= \operatorname{tr}\left(\left(\operatorname{adj} A(t)\right)\frac{\mathrm d A(t)}{\mathrm dt}\right). \end{align}


The determinant is a polynomial on the entries of the matrix. Hence it's differentiable infinitely many times.


The determinant of a square matrix is a polynomial of its entries so it is infinitely differentiable.


If you would not know that the determinant is a polynomial in the entries of the matrix you may know that it is, if considered as a function of the columns (or rows) of the matrix, mulitilinear, hence $C^{\infty}$ as a function of the columns. Since the matrices depend smoothly on their entries they also depend smoothly on the columns.


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.