Memory Scaling of Electronic Relaxation
Posted: Thu Aug 27, 2015 12:04 am
Hello,
I am running into a problem where I need to calculate density of states for many different systems, but I keep having to find out the right kpoints to use by trail and error:
Either I use too many and the program runs out of memory, or I use too few and the DOS comes out grainy. The optimal settings definitely appear to be different for different systems. I am actually surprised that so far I have always managed to eventually find a setting that works. There is also a cryptic message in the DOS example files that suggests that larger unit cells need fewer kpoints, but it does not describe how strong this effect is.
What I would like to know is what is the general scaling for memory if I do an electronic relaxation with PBE and GGA?
for instance something like:
Number of Bands ^ a * Number of Irreducible Kpoints ^ b * Volume
Also, I know the manual suggests doing DOS by loading in wavefunctions that were generated with fewer kpoints, but that was not working for me, redoing the electronic relaxation is the only thing I could find that worked.
I am running into a problem where I need to calculate density of states for many different systems, but I keep having to find out the right kpoints to use by trail and error:
Either I use too many and the program runs out of memory, or I use too few and the DOS comes out grainy. The optimal settings definitely appear to be different for different systems. I am actually surprised that so far I have always managed to eventually find a setting that works. There is also a cryptic message in the DOS example files that suggests that larger unit cells need fewer kpoints, but it does not describe how strong this effect is.
What I would like to know is what is the general scaling for memory if I do an electronic relaxation with PBE and GGA?
for instance something like:
Number of Bands ^ a * Number of Irreducible Kpoints ^ b * Volume
Also, I know the manual suggests doing DOS by loading in wavefunctions that were generated with fewer kpoints, but that was not working for me, redoing the electronic relaxation is the only thing I could find that worked.