Changing cutoff in NVE MD run gives various errors

[Léon Luntadila Lufungula]

Dear all,

A very happy 2024 to all of you!

I amtrying to run an ab-intio NVE trajectory to check the energy conservation for a given setup of parameters (EPS_SCF, EPS_DEFAULT, CUTOFF, REL_CUTOFF, TIMESTEP EXTRAPOLATION_ORDER) to find a suitable balance between accuracy and speed. However, when trying to reduce the value for the cutoff below 800, the calculation crashes before starting the MD run with a lot of errors written to the standard output (see attached file cutoff-700.stdout). Does anyone know why this happens and if it is a compilation problem or if it is connected to my input in some way? Any help would be greatly appreciated!

I have attached the CP2K input and output and the standard output for the calculations with CUTOFF=800 (working as expected) and CUTOFF=700 (writing errors to standard output).

Greetings,

Léon

[Léon Luntadila Lufungula]

Hi everyone,

So I looked into my input file a bit further and tried switching settings on and off to find out if any of the settings is causing the crash and apparently "FFTW_PLAN_TYPE PATIENT" is the cause of CP2K crashing with CUTOFF < 800. Can someone with more knowledge of the code give their two cents if this is expected behaviour or possibly a bug in the 2023.2 version of CP2K?

Kind regards,

Léon

[Léon Luntadila Lufungula]

Hi all,

I've tested the calculation on the three different compilations on my HPC cluster (7.1-Intel-2020a, 2022.1-foss-2022a and 2023.2-foss-2022a) and all three compilations crash with FFTW_PLAN_TYPE PATIENT enabled. However, the Intel version does seem to throw out some different error messages. I have attached the input and the standard outputs of the three calculations. 

I'm planning on doing long ab intio MD calculations and as the reference manual states "PATIENT planning is recommended for long ab initio MD runs.", so I wish to exploit this feature if possible. Can someone look into this matter or provide an answer as to why this is happening?

Thanks in advance!

Kind regards,

Léon