out-of-memory issue in hybrid DFT TDDFT calculation in CP2K 2025.01

[Guozhen Zhang]

Dear CP2K developers and users,

       In an earlier test of hybrid DFT TDDFT calculation using a workable CP2K 2025.01on a different server, I encountered a memory issue that prevent me from running the job.

       In the test, I ran PBE single point calculation to obtain the wavefunction file, then read this file to start a PBE0 TDDFT single point calculation. The system of interest is an pure organic molecular cyrstal with 200 atoms in the unit cell. I used a 3*4*1 MP mesh for the k-points sampling in the first Brillouin zone. However, in the TDDFT calculation, the program has some memory overflow (as told by the administrator staff of the server) that kills the job.

       Has anyone seen similar issue? Any clue to fix it?

       Please find the input and output in the attachment.

       Thanks!

        Guozhen

[Augustin Bussy]

Hello,

To the best of my knowledge, TDDFT is not implemented with k-point sampling in CP2K. That being said, your calculation fails at the SCF stage, in the RI-HFXk calculation.

The RI-HFXk method for k-point is optimized for small unit cells, and does not scale well with unit cell size. 200 atoms is out of range for this method.

To get the equivalent of k-point HF with large systems, you should use supercell calculations. This will end up using less computational resources since the standard HFX

implementation of CP2K scales linearly with system size.

Best,

Augustin

[Guozhen Zhang]

Thanks, Augustin.

I will try using supercell + HFX + gamma-only for TDDFT calculation