Run xtb on a single core about xtb

[awvwgk]

Thanks for your report. To help you with your issue I will need a little more information, like

• what system size you are investigating (hundreds of atoms, thousands)?
• what is your environment setup (OMP_NUM_THREADS, MKL_NUM_THREADS)?
• how are you calling xtb (using --parallel)?
• what Hamiltonian are you using (they differ in the cpu-time and memory requirement significantly)?
• what kind of calculations are you performing (MD, optimizations, frequencies)?

[gooaah]

Thanks for your reply.

• The system is molecular crystal containning 168 atoms.

• I have never set these parameters and they have no default values on our cluster. I will set OMP_NUM_THREADS=1 to test.

• I am searching for molecular crystal so I use GFN0_PBC. I called xtb through ase. xtb only performs SCF and the optimization is done by ase. xtb SCF is valid most of the time. But I need to optimize thousands of structures and some structures lead to a crash.

[awvwgk]

Thanks for the clarifications, please check if this fixes your issue. Also set MKL_NUM_THREADS=1 since the MKL is usually doing the heavy calculations, for more details on the parallel setup refer to the documentation.

Keep in mind that the GFN0-xTB parametrisation is still preliminary and will change in the near future. But if you run into a crash with a chemical reasonable structure open a new bug report, please.

I will close the issue for now.