Pressure keep decreasing with simulation steps

caihong gao

unread,

Sep 7, 2021, 10:51:06 PM9/7/21

to ipi-users

Hi everyone,

I am trying to use deep potentials with I-PI to predict H isotope fractionations, but I met a problem with I-PI.

When I use deep potentials (generated by deepmd-kit with FPMD data) to predict 'isosc' and 'isotd' for H and O in water, the results of isotope estimators looks good, but the pressure decrease linearly with simulation steps (both 'press_cv' and 'press_md')(see the attached figure). I tried to change the number of beads and water molecules and run more simulation steps, but the problem exits. I guess the problem is related to my input settings, but I can not find the problematic setting. I have attached the input files of one of my calculations. Is there anyone know how to solve this problem? Thanks!

Best,

Caihong Gao

input.xml

in.lmp

caihong gao

unread,

Sep 7, 2021, 10:56:57 PM9/7/21

to ipi-users

Here are the other input files.

water_40.xyz

data.lmp

water.json

Michele Ceriotti

unread,

Sep 8, 2021, 3:09:44 AM9/8/21

to ipi-users

So, I ran this with a n2p2 water potential, and I see no issue. Pressure is large and negative, but this is because the cell has a volume which is some 30% larger than the equilibrium volume - if you run npt it goes back to equilibrium real quick.

So this is a problem of your DeepMD potential, and/or the DeepMD LAMMPS implementation.

If you want to use the n2p2 potential (which is trained on b3lyp calculations, from doi.org/10.1063/1.4971438 ) I attach it. Otherwise, you'll have to figure it out on your own I fear.

Re press_md vs press_cv: press_md is the classical pressure of the extended-Hamiltonian run. it has no physical meaning and it is mostly there for diagnostic purposes. press_cv is the centroid virial estimator of the pressure, and is the one you need to look at.

Cheers

Michele

pressure-nvt.pdf

water-test.zip

volume-npt.pdf

caihong gao

unread,

Sep 9, 2021, 5:13:55 AM9/9/21

to ipi-users

Thanks for your answer, this help me a lot. It looks like the problem is related to the potential. I will try to find the solution in the dp forum, maybe someone have met the same problem. Thanks!

best,

Caihong Gao

caihong gao

unread,

Sep 13, 2021, 9:48:16 PM9/13/21

to ipi-users

With the help from Deepmd developer, the problem has been solved. It seems like that the problem is related to dp_ipi. When I use fix ipi instead of fix dp_ipi in the lammps, use lammps instead of dp_ipi, the pressure reported by i-pi looks normally. Maybe it is a bug related to dp_ipi.

Michele Ceriotti

unread,

Sep 14, 2021, 3:58:05 AM9/14/21

to ipi-users

Thanks for the update. We are not involved in the development of dp_ipi so we'll let them handle what's going wrong. Glad your problem was solved.

Reply all

Reply to author

Forward