Closed liamhuber closed 2 months ago
Transferred from https://github.com/openjournals/joss-reviews/issues/6714
I fully agree with this comment made by the other referee @liamhuber . In fact we do sometimes "ab initio " MD using DFT packages such as SIESTA or CP2K in which MD trajectories are calculated using forces directly obtained from DFT and sometimes we do Monte Carlo using energies from DFT. @liamhuber explained it very well, DFT in this context is a method to calculate forces and energies (and electron densities of course) but not a method to explore/sample different configurations (as MD, Monte Carlo, combinations of them and other methods are). The real problem we face is that, once we know how to calculate forces and energies (a force field or DFT) , what is the most appropiate method to sample the system configurations. As @liamhuber said, comparison between MD (brute force MD or accelerated or biased MD) and kMC is what really matters.
This is a valid point. Thank you both. I removed the DFT-kMC comparison and generated the paper again (see PAPRECA review issue for an updated version of the paper).
https://github.com/openjournals/joss-reviews/issues/6714#issuecomment-2114545228
From the paper.md:
I find the comparison to DFT to be apples-to-oranges, it's equivalent to saying "more computationally efficient that MEAM potentials". DFT is a tool for transforming structures (position+chemistry) into energies and forces (and electron density); kMC is a tool for evolving structures in time. In contrast, the comparison to MD is apples-to-apples. Indeed, one could use DFT (assuming arbitrarily powerful compute) and MD and kMC all together! So the real contrast here is kMC vs brute-force MD for temporal evolution.
IMO the comparison to brute-force MD is already sufficient and DFT should simply be removed. If you really want multiple items to contrast, you could consider mentioning some other acceleration technique, e.g. stuff by Danny Perez and Arthur Voter comes to mind.