My Community

Hi

I'm doing some long MLFF MD runs and very nicely, VASP (I'm using v6.5.0) is writing to the vaspout.h5 file. Unfortunately, there seems to be something wrong with our compilation of HDF5 since p4vasp cannot read the hdf5 file, while it can read the vaspout.h5 file from a different HPC system. While we try to get this sorted out, I'm trying to use the vasprun.xml file to get the coordinates for MSD calculations, since there seem to be some nice utilities there. I have rerun the MD jobs using ML_OUTPUT_MODE = 1 and it seems to me they are moving at about half the speed of when I use ML_OUTPUT_MODE = 0 . If I don't set this option to 1, then the coordinate trajectory is not written to vasprun.xml. I think this option though also results in calculation of pair correlation functions and some other things I don't really need. Is there an option to force only the trajectory coordinates to be output to vasprun.xml? If not using HDF5, is writing these coordinates to the vasprun.xml the best option in your opinion if I want to get MSD and then calculate the diffusion coefficient?

Thanks!
Kirk

There is no option that let's only write the trajectory to the vasprun.xml file while ommitiing calculating and writing of the pair correlation function.

If you need the trajectory, you can use the XDATCAR file. It is human readable and the output frequency of the positions can be controlled via ML_OUBLOCK. It is advisable to use ML_OUBLOCK, since not much happens in the trajectory between a few tenth of steps, but the amount of data written out can become really huge in long MD runs.

Ok great, got it. I do have a tool which reads the vasprun.xml so was hoping to go that way.

Thanks!!

As you also recognized calculation of the pair correlation function and writing to another file (vasprun.xml) can give a significant computational overhead. So using XDATCAR in that case gives you also that advantage. For larger calculations even writing at every step can give also give another noticable overhead in the calculation, but that you can also handle by using ML_OUTBLOCK.