Vasprun having problems parsing Vasp 6.3 vasprun.xml files with ML on?

111457 · March 30, 2022, 9:12am

I noticed that Vasprun is not reading in correctly a molecular dynamics trajectory from a vasp 6.30 calculation when machine learning is used. Vasp only seems to import the ab-initio steps and ignores all of the machine learning steps. In the example, I used a 100 MD steps and vasprun.nionic_steps is 20 while “!grep -c ‘forces’ vasprun.xml” yields a count of 100 steps (as expected). I have attached the vasprun.xml file as well as the XDATCAR from an example simulation with 80 atoms. Is this something that can be fixed?

vasprun.xml.gz (301.5 KB)
XDATCAR.gz (12.1 KB)

foster362 · May 13, 2022, 8:34pm

Same issue. Is anyone working on a fix or know of a work around?

foster362 · May 19, 2022, 3:31pm

github.com/materialsproject/pymatgen

Vasprun fails to read in ML-based MD results

opened 12:58PM - 30 Mar 22 UTC

closed 07:28PM - 18 May 22 UTC

pjf295

**Describe the bug** Vasp 6.3.0 includes the ability to use machine learning (M…L) to generate force fields (FF) based upon ab-initio results. Ths results is a large speed up in calculation speeds. The ML process mixes both ab-initio MD with ML FF results based upon a Bayesian approach. This results in the vasprun.xml file containing both ab-initio results and ML based results. The ML based XML results include positions, forces, and velocities, but ab-initio energies. The Vasprun class init reads in a vasprun.xml file generated with ML FF without error, but it ignores all of the ML results and only includes the (typically widely temporally spaced) ab-initio results. [vasprun.xml.gz](https://github.com/materialsproject/pymatgen/files/8380576/vasprun.xml.gz) [XDATCAR.gz](https://github.com/materialsproject/pymatgen/files/8380577/XDATCAR.gz) **To Reproduce** Steps to reproduce the behavior: See the attached INCAR.txt file for the relevant input cards. Machine learning must be turned on to reproduce the results Provide any example files that are needed to reproduce the error, especially if the bug pertains to parsing of a file. I have also attached gzipp'd versions of XDATCAR and vasprun.xml for a demonstration run with 100 integration steps. The XDATCAR contains 100 different structures for each of the MD steps. The vasprun.xml file also contains all 100 steps, but reading in the vasprun.xml file using the Vasprun init (run = Vasprun('vasprun.xml') leads to a vasprun object only containing a small fraction of the ionic steps in the MD simulation (e.g. only those in which ab-initio steps were performed) leaving out all of the FF results. **Expected behavior** The vasprun object should contain the ionic steps for all MD steps. Perhaps another flag should be set to indicate that the results were generated using ab-initio or ML FF processes. **Desktop (please complete the following information):** Pymatgen version 2022.3.7 Behavior host independent (Mac and Linux have the same symptoms)