nan in fix/qeq

Dear Lammps users and Aidan (Aidan Thompson)

I am trying to calculate the atomic charges of a molecule by using fix qeq/reax with Dreiding ff. When the simulation starts the energy term turns to be nan (not a number) at the first step, it may be due to some errors in charge calculations. I attached the in and data files.

Please advise me.

Thanks

baris

chargein2.lammps (594 Bytes)

nDETDAev.data (5.16 KB)

Dear Lammps users and Aidan (Aidan Thompson)

I am trying to calculate the atomic charges of a molecule by using fix
qeq/reax with Dreiding ff. When the simulation starts the energy term turns
to be nan (not a number) at the first step, it may be due to

some errors in charge calculations.

​or a bad initial structure, bad parameters, bad units, and other things.

I attached the in and data files.

​but not param.qeq​

Metal units is inconsistent with fix qeq/reax.

Lj/cut/coul/cut with 5.0 cutoff is bad.

You structure may be bad.

No ones knows what is in the param.qeq.

Ray

Hi Ray,
I checked my structure and did a short run and visualised it in VMD, hence no problem in my data file.
I used metal units because the terms in param.qeq must be in eV as indicated in the manual. You mean I should use real units where kcal/mol is energy unit, but the terms in the param.qeq file must be in eV. Am I right?
Do I need to assign any initial charges to atoms?

I attached the param.qeq file.

Baris.

Metal units is inconsistent with fix qeq/reax.
Lj/cut/coul/cut with 5.0 cutoff is bad.
You structure may be bad.
No ones knows what is in the param.qeq.
Ray

param.qeq (136 Bytes)

Where did you get those QEq parameters? They look suspicious. Moreover, you have 4 atom types and only three entries in param.qeq - I don’t know how that could have worked.

Bad QEq parameters would no doubt lead to bad results.

Ray

Hi Ray,

To simplify the problem I tried the fix qeq/reax for one water molecule. I attached the data, in, and param files. I took the data for param file from the QEq original paper: J. Phys. Chem. 1991, 95, 3358.

Two problems arise:

  1. When you run without ewald, the charges are nan,
  2. When you run with ewald, a segmentation fault occurs.

And a question: do I have to assign initial charges to atoms?

You can check the files.

baris

Where did you get those QEq parameters? They look suspicious. Moreover, you have 4 atom types and only three entries in param.qeq - I don’t know how that could have worked.

Bad QEq parameters would no doubt lead to bad results.

Ray

param.qeq (95 Bytes)

water.data (677 Bytes)

waterin.lammps (835 Bytes)

For some reason your system/setup does work with the JPC parameters, but it does work with the QEq parameters from the examples/reax/CHO/param.qeq file.

This is not a bug but related to the sensitivity of the current QEq implementation to its parameters.

Ray