As someone working in this field, I disagree. The state of these codes is that if you put garbage in, you get garbage out. Without a strong theoretical backgrou

As someone working in this field, I disagree. The state of these codes is that if you put garbage in, you get garbage out. Without a strong theoretical background, you’re generating junk. Of course, if he’s willing to spend several years developing his knowledge of the field, there is for sure room to help. If not in your suggested machine learning area, instead perhaps in improving the quality and usability of the code. But instead I would recommend he start from something like the NOMAD database, where the calculations have already been run by more knowledgeable people. Then he can focus on the analysis side.