Background:
I SSH into a multi-node compute cluster from a desktop PC. This cluster operates using a queuing system. VASP v5.3 and ATAT are installed in my local bin on the cluster. I have been using VASP for some time without issue. I needed the cluster expansion capabilities of MAPS, so I installed ATAT and the MAPS/VASP interface according to the user guide. I wrote a shell script to incorporate the operation of MAPS into the queuing system, which consists of the following:
Problem:
I created a simple test environment for a binary Cu/Ni system. When I run the above script, cluster expansion begins as expected, creating many different subdirectories each containing an automatically generated structure and the associated inputs for VASP. However, VASP fails to execute after encountering the following errors:
"Illegal variable name.
Illegal variable name.
POTLDA: Undefined variable.
unable to open OSZICAR or OSZICAR.static"
This sequence of errors occurs in every single subdirectory in sequence. So after VASP fails to operate on a structure, MAPS generates the next one, VASP fails again, and so on. I do not know what variables have the illegal names, and I do not understand the issue with POTLDA because I explicitly specify the PBE potentials in my vasp.wrap file.
Inputs:
I automatically generated my lat.in according to:
#!/bin/csh
#enter name of vasp executable here
set VASPCMD="vasp-stable"
#enter the directories containing the pseudoptentials here
set POTLDA="$(HOME)/potentials/potus_LDA"
set POTGGA="$(HOME)/potentials/potus_GGA"
set POTPAWLDA="$(HOME)/potentials/potpaw_LDA"
set POTPAWGGA="$(HOME)/potentials/potpaw_GGA"
set POTPAWPBE="$(HOME)/potentials/potpaw_PBE"
Any help that can be offered with this problem would be greatly appreciated.
Thank you for your time,
Patrick
The problem had to do with the names used in my .ezvasp configuration file, hence the "Illegal variable" errors, which in turn prevented VASP from initializing properly.
Fortran does not allow the use of special characters in variable names, so I changed the name of my VASP executable from ‘vasp-stable’ to ‘vasp_stable’, and replaced every instance of ‘$(HOME)’ with an explicit file path.
Having made these changes, my run script operates as expected and there are no errors in my slurm file (except that oddity regarding the .machines.rc, but I don’t think that has anything to do with ATAT).
I have corrected the USEPOT and DOSTATIC tags.
The location of my ezvasp config file is $(HOME)/.ezvasp.rc
However, I am working on a shared compute cluster, so my own home directory is not the home directory, per se. For example, when I unpacked all the ATAT executables into my local bin, I also had to add a custom environment variable adding a path to that bin. Otherwise, the system was unable to find the executables.
For the same reason, it is possible the system simply does not know where to look for my ezvasp file. I’ll see what I can figure out.
UPDATE
I tried moving the ezvasp file into my local bin simply to see what would happen. When I tried running my script in this configuration I encountered an entirely different, and much longer, list of errors.
So I put the file back in the home directory then took a closer look at what what going on, and I noticed an error in the VASP slurm: "Warning: File $(HOME)/.machines.rc not found, running in single machine mode"
I cannot say if I simply overlooked this error before, or if it wasn’t there during my previous tests.
In any case, the mess I made trying to relocate the ezvasp file seems to indicate that the system is able to find it when it is in the home directory.
Which brings me back to the question of why isn’t it running?