SCF converges more slowly as AIMD progresses, but resets when restarting

Yinyu chan

unread,

Apr 9, 2026, 7:18:49 AMApr 9

to cp2k

Dear CP2K community,

I'am a new user of CP2K(v2026.1) and I'm running an AIMD simulation of a metal-electrolyte interface (Au-Pd alloy/NaCl(aq)) with electric field.

However, I encountered an issue where the SCF starts fast but then becomes slow, as shown in the *ener file. But when I restarted the AIMD at steps 5000(SCF_GUESS RESTART) and 5378(SCF_GUESS ATOMIC), the SCF convergence starts fast again, but then becomes slow once more. I would like to know whether this phenomenon is normal, and how to keep the SCF convergence fast during the AIMD.

I've attached the inputs and outputs for reference. I would appreciate it if anyone could offer suggestions on this issue.

Thank you for your time and support.

Best regards,

Yinyu Chan

input.inp

restart.inp

interfaces-1.ener

metal_electrolyte.xyz

Shiyang Wang (W0728)

unread,

Apr 10, 2026, 3:44:09 AMApr 10

to cp2k

Hi,

I have doubt if your system is fit for OT method; it looks like to be metal-based and may require smearing.

Yinyu chan

unread,

Apr 13, 2026, 3:42:44 AMApr 13

to cp2k

Hi,

Thanks for your reply!

Actually, the input settings follow a JACS Au paper on the Au/NaCl interface without smearing. And OT method is required for applying periodic external electric fields. I suspect the slow SCF convergence is related to the use of historical information in cp2k, but i haven't found any useful info on it so far.

Best regards,

Yinyu

Shiyang Wang (W0728)

unread,

Apr 13, 2026, 8:53:24 AMApr 13

to cp2k

You can consider using an alternative preconditioner, such as FULL_ALL, and see how it goes.

Yinyu chan

unread,

Apr 15, 2026, 4:55:10 AMApr 15

to cp2k

Hi,

It doesn't seem to work. I wonder if restarting the AIMD every certain number of steps yields the normal results. If so, simulations can be accelerated in this manner.

Marcella Iannuzzi

unread,

Apr 15, 2026, 5:14:14 AMApr 15

to cp2k

Hi

you have a bigger problem than the time to convergence.

At restart there is a clear jump in energy, indicating that the dynamics is restarting from a different state.

Regards

Marcella

Yinyu chan

unread,

Apr 16, 2026, 6:04:56 AMApr 16

to cp2k

Hi,

Thank you for pointing this out. You are correct that a clear energy jump occurs at step 5001 where I restart the simulation as a trial to speed up convergence with a RESTART file and modify the ASPC order and NDIIS to 2 and 5, respectively. The settings keeps consistent in the whole production run stage. However, the issue raised in this conversation still exists.