search

XanaduAI / GradDFT

GradDFT is a JAX-based library enabling the differentiable design and experimentation of exchange-correlation functionals using machine learning techniques.
Apache License 2.0
79 stars 7 forks source link

20 implement the harris foulkes energy #48

Closed PabloAMC closed 1 year ago

PabloAMC commented 1 year ago

The tests I have implemented indicate a super close prediction for HF (under a kcal/mol) and some 70 kcal/mol difference for Lithium.

jackbaker1001 commented 1 year ago

The real test for this being implemented properly is to see if for a given approximate density rhoapprox that E{harris-foulkes} < E_{KS}

PabloAMC commented 1 year ago

I can quickly modify the tests but it happens. Also, I am wondering: if DFT is not always variational, why does the above statement hold @jackbaker1001?