1. You have some functional. That get optimised
2. After that, there is still a tensor valued parameter $P$ that defines the tensor function $T (\mathcal{C},P)$ that will be fixed by the equation (I forgot what's on the RHS which is why I have not wrote "=")