Restarting DFT calculation from DF electron density

[ftsong22]

Assume I have the electron density represented as the expansion coefficients under an DF auxiliary basis (which come from machine learning prediction). Although the kinetic energy cannot be evaluated without an AO basis, theoretically the Kohn-Sham effective potential has no kinetic term and should be able to be constructed from DF density (assume we are using PBE). Then we might restart the DFT from this effective potential.

Is this possible for psi4? And if so, may I have some hints on how to do this?

[JonathonMisiewicz]

This fundamentally isn't how Psi4 works. Psi uses a AO density matrix representation wherever it can, switching to a real space for evaluation of DFT functionals but nowhere else. Maybe you can use Psi4 functions to write your own HF code with this functionality (I'd need to think harder to know if that's even doable), but you can't inject such a density into Psi4's existing HF algorithm.

[susilehtola]

This is possible in principle. You get the Coulomb term using the auxiliary basis expansion; this is already implemented. You can also compute the exchange-correlation contributions from the auxiliary basis expression; this needs to be implemented. Then you have the two-electron contributions to the Fock matrix. Sum these to the core Hamiltonian, you have the full Fock matrix, which you can diagonalize to get orbitals.

[ftsong22]

Thank you @susilehtola , I will give it a try.