experimentalist here, I use ORCA for small molecules but have been hoping to find some time to learn tight-binding approximations for larger structures, and planewave methods for solids and surfaces.
don't forget that DFT is a late invention in the game, you should really spend time understanding HF, MP, and other post-HF techniques to see why DFT is so important (and usually wildly inaccurate).
>>11871810PBE is nice and an experimental brainlet like me can kind of understand how it works.
currently trying to find a hybrid functional to get better geometries, but nowadays I just do all my energy calculations at DLPNO-CCSD(T) to avoid choosing the "correct" functional.
>>11872012ORCA is written entirely in C++ and is FOSS. It's very fast and has most the features proprietary softwares have, and a lot that most do not. Frank Neese (the founder, lead developer) publishes all the methods in scientific journals and you can read the code yourself.
>>11873960your treatment is funamentally wrong because it doesn't treat the system as a quantum mechanical problem, but instead tries to rely on patterns it found before. DFT can (theoretically) provide the exact solution to the Schrodinger (or Dirac) equation if you have the exact exchange and correlation functionals. No one knows what those are, so we have approximations based on what we know about the behaviour of atoms and electrons.
Don't get me wrong, ML/DL/AI approaches to these problems are very very cool and will absolutely benefit the field. But there will always be difficult and unusual problems that they will not be able to solve.
For example, would your neural net predict non-nuclear attractors in Mg(I) dimers?
>>11874252DFT is not always first principles aka ab initio. Many functionals are fitted to experimental data or use other empirical methods to fit parameters to obtain better results compared to known values / benchmarked values using very expensive ab initio techniques (e.g., coupled-cluster).
