[lammps-users] Signal Segmentation Fault in LAMMPS Simulations

Hii

LAMMPS Users,
I am using LAMMPS for the molecular dynamics simulations and exhaustively using reactive force field potential with the TIP4P water model. I am using the LAMMPS version 12 Dec 2018 on my institute supercomputer Cray XC40. When I submit my job, I get the following error:

Warning: changed valency_val to valency_boc for X

_pmiu_daemon(SIGCHLD): [NID 00872] [c4-0c1s10n0] [Tue Jan 26 02:24:18 2021] PE RANK 0 exit signal Segmentation fault

[NID 00872] 2021-01-26 02:24:18 Apid 687096: initiated application termination

Herein, I also attached the error file for your perusal.

Could you please help me?

Thank you!

With best regards,

Bharat

Bharat Bhushan Sharma

Postdoctoral Researcher,

Department of Chemical Engineering,

IISc Bangalore

water.e107562 (246 Bytes)

Do you mean you are using the ReaxFF potential along with TIP4P water?
Can you explain how you are doing that?

Since ReaxFF includes charge equilibration it would try to reset the charges
on the water atoms as well, which seems like it would violate the TIP4P model.

Steve

Please reply to the mail list, not to me.

I’m CCing Ray Shan who is a ReaxFF expert.
I’m not sure it makes sense to use ReaxFF in conjunction
with TIP4P like this, but Ray can advise. In particular if
there are issues with using KSpace with ReaxFF also resetting
the charges.

Steve

There are two basic problems with this input that are not easy to resolve.

a) you must not combine pair style reax/c with a kspace style. as a matter of principle, kspace styles apply to all atoms, but ReaxFF already contains long-range handling via a Wolf sum scheme, so using a kspace style would double count those contributions.
b) you must not combine pair style lj/cut/tip4p/* with lj/cut/coul/. this will lead to inconsistent coulomb interactions for all tip4p water molecules. there is no need for it in the first place since lj/cut/tip4p/ styles act like lj/cut/coul/* for all atom types not identified as TIP4P oxygen or hydrogen
c) mixing atoms subject to charge equilibration with atoms not subject to it causes inconsistencies and will unphysically leak energy from the ReaxFF subsystem (same problem exists in QM/MM simulations)
d) mixing ReaxFF with non-ReaxFF via a hybrid pair style is very problematic in general and (unlike quantum chemistry calculations) ReaxFF parameterization are very specific to the kind of system they are parameterized for and thus not very transferable. so a cleaner approach would be to either model the entire system with ReaxFF (which requires finding a suitable parameterization) or use a different model pair style for hBN that is not as much negatively affected by being use in a hybrid pair style configuration as ReaxFF

Axel.

The message is a warning indicating an inconsistency of the provided parameters in the “ffield.reax.BN” ReaxFF file.
The inconsistent set of parameters is the one determining the interactions of the chemical element “X” which is a dummy entry in the ReaxFF file.
You can safely delete all four lines associated with the entries of element “X” in " ffield.reax.BN" without affecting the calculations.

Also, note that this warning will not terminate LAMMPS. Most likely it is the combination of TIP4P with ReaxFF that leads to the segmentation fault.
You can easily check this by using the parameters for TIP3P instead of TIP4P in your simulations. TIP3P does not have the additional four site and it would be possible to adjust the partial charges (although the physical model would be invalid)

Regards
Evangelos

Στις Παρ, 5 Φεβ 2021 στις 3:26 μ.μ., ο/η Steve Plimpton <[email protected]> έγραψε: