Derivative of the determinant of a singular matrix is NaN

Consider the following code

from jax import grad
import jax.numpy as np

x = np.zeros(shape=(4,4))
d = grad(np.linalg.det)

print(d(x))

the derivative of np.linalg.det(x) against x is obviously a zero matrix, but NaNs was outputted

[dpfau]

The function I shared above already works. I’m just working on integrating it into JAX.

[mattjj]

Woohoooooo thanks @dpfau!