[CP2K-user] Force calculation in CELL_OPT vs MD

[mar...@gmail.com]

Hello,

This may well just be a gap in my knowledge regarding how forces are 
calculated in CP2K for geometry optimisations vs molecular dynamics. In 
such a case, I would still appreciate any insight as well.

In short, for the same system with virtually the same SCF and other 
settings, force calculations during MD take over 4 times as long as the 
same step during CELL_OPT. I don't understand why this is happening and 
would appreciate any insight or solution.

Details: I have started by optimising the geometry of this system 
(relatively tight convergence criteria, no issues) prior to running any MD. 
My issue comes during the MD, where each step takes significantly longer 
(average of ~30 seconds per opt. step vs ~90 seconds per md step) than i 
would expect based on the corresponding CELL_OPT timings. Most of this 
seems to be taken up by force calculations (timings in the output file 
suggest ~20 seconds for the SCF process and finding the electron density 
and ~60-70 seconds for the force calculation, position update etc.). 
Previously, similar MD calcs on a cluster model of this system (but 
containing more atoms in a larger unit cell) required 35-45 s per time 
step. This is despite using the same numbers of cores, same basis sets, 
same cutoffs, and same settings where relevant.

Background: I am working on a certain porous material (unit cell 15x15x20) 
with regards to catalytic processes which can take place inside the pores. 
CP2K 6.1 or 7.1 (issue persists with both), 600/50 Ry cutoff and rel 
cutoff, PBE-d3 functional with DZVP-MOLOPT basis sets, BFGS optimiser for 
CELL_OPT,  GTH pseudopotentials. CSVR thermostat with 0.5 fs time step and 
1 fs time constant.

Best regards,
Martin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20201015/f6ab0ae6/attachment.htm>