Dynmat Calculations in Parallel
Moderator: moderators
Dynmat Calculations in Parallel
I am attempting to use Dynmat VTST tool in parallel to obtain vibrational frequencies. The VTST web page described that this set up was fairly similar to the NEB set up. Going off of this statement, I created 4 directories, numbered 01 through 04. In each directory, I placed a DISPLACECAR with displacements defined for the atoms I wished to displace in this directory and zeros for the remaining atoms as well as a POSCAR with selective dynamics turned on. In the POSCAR, I assigned the 'T T T' flag for those atoms that I provided displacements for in the DISPLACECAR, and the 'F F F' flag for the remaining atoms (I'm not sure if this was necessary). I placed my INCAR, KPOINTS, and POTCAR in the parent directory of the 01-04 sub-directories. When I submitted this job, it ran for about 25-35 time steps of the first iteration, and then crashed with a 'rank x in job 1' error causing a 'collective abort of all ranks'. I've seen this error associated with starting vasp_mpi jobs in general, but I have no problems running vasp_mpi for geometry optimizations or NEB calculations. I have a feeling that I am structuring my directories incorrectly, but in scanning the forum and googling, I did not see a more detailed description. Any help one could provide would be very appreciated.
Re: Dynmat Calculations in Parallel
The way you set this up is not quite right. I see that the instructions on the web page are not clear. I have modified the page:
http://theory.cm.utexas.edu/vtsttools/dynmat/
If it is still not understandable, let me know and I'll try again.
http://theory.cm.utexas.edu/vtsttools/dynmat/
If it is still not understandable, let me know and I'll try again.
Re: Dynmat Calculations in Parallel
graeme wrote :
> The way you set this up is not quite right. I see that the instructions on
> the web page are not clear. I have modified the page:
> http://theory.cm.utexas.edu/vtsttools/dynmat/
> If it is still not understandable, let me know and I'll try again.
After reading your updated description, I placed my DISPLACECAR in the parent directory, with only POSCAR files in the sub-directories, and it appears that the calculation ran smoothly. I placed the same POSCAR in each sub-directory (i.e., I did not adjust the selective dynamics tags to reflect which atoms I wanted to displace in each sub-directory). Was this the correct interpretation of your instructions?
To check the results, I would like to compare the computed frequencies to frequencies I previously obtained in a non-parallel implementation of Dynmat for this transition state. What is the best way to extract the frequencies when Dynmat is run in parallel? It doesn't seem like the dymmatrix perl script would know which atoms were displaced in each sub-directory since there is only one parent DISPLACECAR, which makes me wonder if I did not correctly set up the job (again :) ).
Many thanks for your help and your *extremely* helpful vtstcode/scripts.
> The way you set this up is not quite right. I see that the instructions on
> the web page are not clear. I have modified the page:
> http://theory.cm.utexas.edu/vtsttools/dynmat/
> If it is still not understandable, let me know and I'll try again.
After reading your updated description, I placed my DISPLACECAR in the parent directory, with only POSCAR files in the sub-directories, and it appears that the calculation ran smoothly. I placed the same POSCAR in each sub-directory (i.e., I did not adjust the selective dynamics tags to reflect which atoms I wanted to displace in each sub-directory). Was this the correct interpretation of your instructions?
To check the results, I would like to compare the computed frequencies to frequencies I previously obtained in a non-parallel implementation of Dynmat for this transition state. What is the best way to extract the frequencies when Dynmat is run in parallel? It doesn't seem like the dymmatrix perl script would know which atoms were displaced in each sub-directory since there is only one parent DISPLACECAR, which makes me wonder if I did not correctly set up the job (again :) ).
Many thanks for your help and your *extremely* helpful vtstcode/scripts.
Re: Dynmat Calculations in Parallel
I've made a few additional comments on that web page. First, yes, you put the same POSCAR in each of the image subdirectories. To construct the matrix, run:
dymmatrix.pl 1 DISPLACECAR ??/OUTCAR
which means that you have 1 DISPLACECAR file with force information in the ??/OUTCAR files. You can, of course, also put them all on the command line. For example, with 2 images:
dymmatrix.pl 1 DISPLACECAR 01/OUTCAR 02/OUTCAR
dymmatrix.pl 1 DISPLACECAR ??/OUTCAR
which means that you have 1 DISPLACECAR file with force information in the ??/OUTCAR files. You can, of course, also put them all on the command line. For example, with 2 images:
dymmatrix.pl 1 DISPLACECAR 01/OUTCAR 02/OUTCAR
Re: Dynmat Calculations in Parallel
That worked perfectly! The additional comments concerning the use of dymmatrix.pl for multiple image jobs were very helpful. As should be expected, the extracted frequencies matched those from the single image job, so everything seems to be working well. Now, it's time to test the convergence of the frequencies! Thank you very much for your help, and for clarifying the DynMat description on the Vasp Tools website.