Question: Doest there exist a smooth map $ h:\text{GL}_n(\mathbb{R}) \to \mathbb{R}$, such that $dh_X(V)=\langle (X^{-1})^{T},V\rangle $?

Suppose such an $h$ exist. Then it must hold that $d^2h=0$:

($d$ is the exterior derivative; in more elementary terms, second derivatives must commute.)

Since $$dh=\sum_{ij}(X^{-T})_{ij}dx^{ij},$$ we have $$ d^2h=\sum_{ijks} \frac{\partial\big((X^{-T})_{ij}\big) }{\partial x_{sk}} dx^{sk} \wedge dx^{ij}=\sum_{ijks} \frac{\partial\big((X^{-1})_{ji}\big) }{\partial x_{sk}} dx^{sk} \wedge dx^{ij}$$

which is zero if and only if

$$ \frac{\partial\big((X^{-1})_{ji}\big) }{\partial x_{sk }}=\frac{\partial\big((X^{-1})_{ks}\big) }{\partial x_{ij}}. \tag{1}$$

I checked most of the cases when $n=2$ (then we have a nice formula for the matrix inverse) and the requirement $(1)$ was satisfied.

So, is there such a function $h$ for $n=2$? For general $n$?


From Jacobi's formula for the derivative of the determinant, whenever $\det X > 0$ we have

$$ d (\log \det X)(V) = \frac{1}{\det X}\mathrm{tr}(\mathrm{adj}(X) V)=\mathrm{tr}(X^{-1}V)=\langle X^{-T},V\rangle;$$ so $h(X) = \log \det X$ satisfies your equation on $GL^+(n, \mathbb R).$ To extend this to the orientation-reversing component, note that when $\det X < 0,$ $$d(\log (-\det X))(V)=\langle X^{-T},V\rangle.$$ Thus we can write the global solution as $h(X) = \log\,\left|\det X\right|.$


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.