ETRS and CN propagators lead to completely different Ehrenfest dynamics

[Natalia K]

Dear CP2K team,

I performed simple tests for H2 with cp2k-9.1 using an input file from cp2k/tests/QS/regtest-rtp-2 and modifying the external field (the input file is attached). I then performed 4 tests, using ETRS or CN propagators and two different time steps, dt=0.25 a.u. and 0.125 a.u. With the ETRS propagator, the energies as well as the H atom dynamics are the same for both dt values. However, the results with the Crank-Nicolson propagator (CN) are not only different for two time steps, but also are completely different from when using ETRS. I attach the results here. I tried refining some other parameters, but the two propagators always lead to different EMD.

My question: when performing Ehrenfest dynamics with applied external field, which propagator should I trust? Why they give such different results?

Best regards,

Natalia

[Natalia K]

UPD on this issue: apparently, the Cranc-Nicolson is the less stable one. I did some additional tests with a very small time-step (0.0005 fs) and the CN propagator converged to the ETRS result, the same one that ETRS gives even with dt = 0.006 fs. So, the answer is, one should use the ETRS propagator to be able to use larger time step.

Natalia