I have trouble with understanding vector calculus notation in a PDE book, much less matrix analysis. There is a proof in Majda's book "Vorticity and Incompressible Flow" very early on that I am not following:

Let $X$ be the particle-trajectory mapping (taking values in $\Bbb R^n$, for $n=2,3$) of a smooth velocity field $v$, i.e. it solves the ODE for every $\alpha$, \begin{cases}\frac{d}{dt}X(\alpha,t) &= v(X(\alpha,t),t) \\ X(\alpha,0) &= \alpha\end{cases}Then if $$J(\alpha,t) := \text{det}\left[ \frac{\partial X^i}{\partial \alpha^{\ j}}\right]_{i,j}(\alpha,t)$$ It satisfies $$ \frac{\partial}{\partial t} J(\alpha,t) = \text{div }v|_{(X(\alpha,t),t)}J(\alpha,t)$$

There is only the one line I can't follow- $$\frac{\partial J}{\partial t} = \sum_{i,j} A^j_i\frac{\partial}{\partial t} \frac{\partial X^i}{\partial \alpha^{\ j}}(\alpha,t)$$ Where $A^j_i$ is the minor of the entry $ \frac{\partial X^i}{\partial \alpha^{\ j}}$ in $J$. Can someone explain where this comes from?

edit For completeness re Normal Human's answer, I check below the Leibniz rule for multilinear functions:

\begin{align} \det A(s) = \det\begin{pmatrix}r_1(s) \\ … \\ r_n(s)\end{pmatrix} &= \det\begin{pmatrix}r_1(s) - r_1(t) \\ … \\ r_n(s)\end{pmatrix} + \det\begin{pmatrix}r_1(t) \\ r_2(s) \\ … \\ r_n(s)\end{pmatrix} \\ &… \\ &= \sum_{k=1}^{n+1} \det\begin{pmatrix} r_1(t) \\ \dots \\ r_{k-1}(t) \\ r_k(s) - r_k(t) \\ r_{k+1}(s) \\ … \\ r_n(s)\end{pmatrix} \end{align}

Note that $\det A(t)$ is the last term of the sum; subtracting it and dividing by $\frac{1}{s-t}$, which we pull into the $k$th row by linearity and taking limits gives the "Leibniz rule".

I believe I came up with a separate proof of this proposition, which I include partially as a way for me to remember. By the permutations definition of determinant, \begin{align} \frac{\partial}{\partial t}J &= \sum_\sigma \text{sgn } \sigma \frac{\partial}{\partial t} ∏_{i} \frac{\partial X^i}{\partial \alpha^{\sigma(i)}} \end{align} Leibniz Rule tells us that $ \frac{\partial}{\partial t}∏_{i} \frac{\partial X^i}{\partial \alpha^{\sigma(i)}} = \sum_k \frac{\partial}{\partial t} \frac{\partial X^k}{\partial \alpha^{\sigma(k)} } \prod_{i≠ k}\frac{\partial X^i}{\partial \alpha^{\sigma(i)}}$. Writing $j=\sigma(k)$ as shorthand and interchanging derivatives by smoothness of $X$,

$$\frac{\partial^2 X^k}{\partial t \partial \alpha^{\ j}} = \frac{\partial}{\partial \alpha^{\ j}}\frac{\partial X^k}{\partial t } = \frac{\partial}{\partial \alpha^{\ j}}v^k(X(\alpha,t),t) = \sum_\ell \partial_{x^\ell}v^k \frac{\partial X^\ell}{\partial \alpha^{\ j}}$$ Turning $\sum_\ell$ into the outermost sum, we obtain the result we want in the $\ell = k$th term, as that term is $$ \sum_\sigma \text{sgn } \sigma \partial_{x^k}v^k \frac{\partial X^k}{\partial \alpha^{\sigma(k)}} \prod_{i≠k} \frac{\partial X^i}{\partial \alpha^{\sigma(i)}} =\text{div } v|_{(X(\alpha,t),t)} J(\alpha,t)$$ For $\ell≠ k$, we are taking the determinant of the same matrix but with the $k$th row replaced with the $\ell$th row. Since we have a repeated row, the determinant is 0. This proves the result.


It's called Jacobi's formula for the derivative of determinant. The Wikipedia article gives a detailed derivation full of indices, so I'll balance that with a formula-free sketch:

  1. The determinant is a multilinear function of the rows of the matrix.
  2. The Leibniz (product) rule extends to multilinear functions in a natural way: the derivative falls onto one factor at a time, then the results are added up.
  3. Thus, the derivative of determinant is obtained by differentiating one row at a time, and summing the $n$ resulting determinants.
  4. Expand each of the above determinants along the differentiated row. This results in a sum of minors of the original matrix multiplied by the derivative of the complementary entry. I.e., the sum that you are asking about.
  • $\begingroup$ Thank you, I instead performed a laplace expansion first (since I saw minors) and then tried to differentiate. This sketch made everything clear :) $\endgroup$ – Calvin Khor Oct 27 '15 at 9:01

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.