that's a fair point. i would still use gradient descent (or conjugate gradients) if i didn't care about the exact minimum.