An identity involving the derivative of a scalar function with respect to a tensor-valued variable

78 Views Asked by At

Let B be a tensor-valued variable, taking values from the set of second order tensors on the vector space naturally associated with Euclidean 3-space.

It is given that B is invertible.

I am looking for a proof of the following statement:

$$ \frac{\partial (\log \det \mathbf{B})}{\partial \mathbf{B}} = \mathbf{B^{-T}} $$

Also, it seems one would need an assumption on the positivity of the determinant of B, so I am guessing this is allowed too.

Thanks.

1

There are 1 best solutions below

2
On

By the chain rule $$\frac{\partial (\log \det B)}{\partial B} = \frac{1}{\det B} \frac{\partial (\det B)}{\partial B} = \frac{1}{\det B} \det (B) B^{-T}$$

Apart from the last equality this should be clear, the last equality follows from Cramers rule. A reference is this wikipedia page and this section on the same page.

(Note: it may be not completely clear what is exactly meant by $\frac{\partial}{\partial B} f(B) = A $. The interpreation of this expression is that $$\frac{\partial}{\partial B_{ij}} f(B) = A_{ij} $$ for every pair of indices $i,j$).