Can this be extended to orbital DFT?

[nkarve]

First of all this is an excellent project, I came across it while trying to develop my own autograd-based DFT solver for periodic systems (great minds think alike!).

I noticed your version is currently only for orbital-free DFT - how hard would it be to extend this to orbital-based DFT? The specific toy example to consider for concreteness is simply the Schrödinger equation for a particle in potential V, where the object to be optimised via gradient descent is no longer the density directly but rather the coefficients in reciprocal space.

[wcwitt]

It's definitely something we're thinking about, but not on the immediate horizon. You might be interested in DFTK or differentiable PySCF.

[nkarve]

Thanks for the links!