In order to compare the DFT energy with the entry in material project, it is generally required to use the same input parameters as material project, but whether generated by MPRelaxSet class or downloaded from mp database, the INCAR file does not contain EDIFFG parameters, this convergence criterion is not very strict, and often overestimate the energy.
I would like to ask, when I want to perform phase stability, electrochemical window, and other analyses that need to compare energy with material project, should I use pymatgen to generate my INCAR directly without any change or should I customize some parameters, such as making EDIFFG=-0.05
Hi @yu-yang-sjtu,
This is correct; MPRelaxSet does not set EDIFFG. This is the older input set which uses GGA (PBE). The newer input sets based on SCAN like MPSCANRelaxSet, do specify an EDIFFG. See more information in our paper here:
For your question, I would strongly recommend using a tighter convergence criterion for any serious, detailed calculations, e.g. use an EDIFFG, and also avoid the use of “EDIFF_PER_ATOM”.
Some settings might be appropriate for high-throughput, especially with the context that these settings were chosen about ten years ago when computing resources were more scarce, that are not appropriate today in a context where a smaller, more accurate set of calculations is desired.
Hope this helps!
Matt
Thank you for your reply, it was very helpful to me
@mkhorton Thanks for the helpful answers. I would like to ask a follow-up question. To construct phase diagrams, I have been using MP data and my own DFT results calculated with the input settings from MPRelaxSet. Now that we add stricter EDIFF in my DFT relaxations, can I still directly take the MP entries to construct phase diagram? Or should we redo the relaxations of all related MP entries as well?