SGCPMD: problem with equilibration?

[Alexandra Dávila]

Dear Cp2k users,

I have carried on a second-generation Car-Parrinello (SGCP) MD simulation for Pt(111)/water interface. Unit cell contains Pt: 6x6x4 (2 frozen layers), water: 134 molecules.

I have performed a similar workflow as Onofrio ("SGCPMD: number of steps per SCF run increase suddenly after initial 400 MD steps"). But I have used the NOISY_GAMMA as in the publication "First-Principles Simulations of an Aqueous CO/Pt(111) Interface" doi:10.1021/acs.jpcc.8b05933, Pt: 5E-05 1/fs and for water 2.2E-04 1/fs. Temperature is set to 330K.

Unfortunately, in my case along the SGMD the temperature of Pt region drops to roughly 180 K, the temperature of the hole system is about 328K. I expected that the temperature of Pt region would be closer to the water after a equilibration time of 4ps. Does this mean that probably my system is not well-equilibrated? Is this going to affect the dynamics and energy of the system? (PS: the density profile seems ok, double peak on oxygen, well-known for Pt(111))

Many thanks in advance!

Alexandra

PS: I also observed the increase of the SCF steps after a restart!

TOP: red dashed line: average temperature of system

BOTTOM: temperature for each region

[Onofrio Tau]

Dear Alexandra,

have you done an equilibration step before starting the SGCPMD run?

As I can see from your plots it seems that it is not equilibrated yet. I suggest you to make a BOMD run first for better equilibration, then switching to sgcpmd, as follows:

1. BOMD as NVE ensemble + TEMP_TOL 100

2. switch to SGCPMD using RESTART.wfn of the previous equilibration step as a wfn guess

It works for me.

Hope it helps

Regards

Onofrio

[Alexandra Dávila]

Dear Onofrio,

thanks a lot for your fast response! Indeed, I did a NVE simulation with Langevin  (as suggested here:https://www.cp2k.org/howto:md). Then I used the geometry and wfn for the SGCPMD.

As you, I use TEMP_TOL, but I set it to the value of 50 for each region. Do you think that it makes the difference?

 This are the parameters that I used for the NVE:

   &MD
     ENSEMBLE  LANGEVIN
     STEPS  6000
     TIMESTEP     5.0E-01
     TEMPERATURE     3.3000000000000006E+02
     TEMP_KIND  T
     SCALE_TEMP_KIND T
     &LANGEVIN
       GAMMA     1.0000000000000000E-03
       NOISY_GAMMA     0
     &END LANGEVIN
     &THERMAL_REGION
       DO_LANGEVIN_DEFAULT  F
       &DEFINE_REGION
     DO_LANGEVIN F
         LIST  145..546
         TEMPERATURE     3.3000000000000006E+02
         NOISY_GAMMA_REGION     2.1999999999999998E-04
         TEMP_TOL 50
       &END DEFINE_REGION
       &DEFINE_REGION
         DO_LANGEVIN F
         LIST  73..144
         TEMPERATURE     3.3000000000000006E+02
         NOISY_GAMMA_REGION     4.9999999999999996E-05
         TEMP_TOL 50
       &END DEFINE_REGION
     &END THERMAL_REGION
   &END MD

The NOISY_GAMMA is not considered in the simulation. Is there something not so good with the set of parameters I used?

I appreciate a lot your feedback!

Best regards,

Aexandra

[Onofrio Tau]

Dear Alexandra,

I used to do ENSEMBLE NVE in &MD for the BOMD run. Probably 50 K of the TEMP_TOL is too strict, at least for Pt. As you can see from your last plot, the Temperature of the Pt region frequently drops to your lower limit, 280 K, even after 4000 fs (not good). This is why you get a Pt temperature drop in the SGCPMD run, because of a not good previous equilibration.

Regards.

Onofrio

[Alexandra Dávila]

Dear Onofrio!

you are right! It is too tight. I will try out what you recommende me!

Thanks a lot,

Alexandra

El jueves, 15 de septiembre de 2022 a las 12:53:45 UTC+2, onofr...@gmail.com escribió:

[Onofrio Tau]

Let me know if it works for your system. Can I ask you how many steps per SCF run are tipically required for convergence in your SGCPMD run?

Thanks.

Onofrio

[Alexandra Dávila]

I will!

After a restart, I get 3-6 steps, but then it goes high to ~15. Sometimes it could be higher, but then goes again down.

Resgards,

Alexandra