MPI error

LAMMPS LAMMPS Beginners
1

Hi everybody,

I have a system combined with water & monomer. Avogadro built the geometry of water and monomer and the data file was changed to the Lammps input file by Packmol software.

I want to check the degree of polymerization after adding energy to the system in Lammps, on 2 Aug 2023. Unfortunately, I am getting the error MPI Job aborted. I don’t understand where exactly I should look to tackle this problem. Could anyone please help me with the solution?

Thanks for your consideration,

Mohsen

data.data (31.3 KB)
input.txt (1.2 KB)
Potential.txt (13.1 KB)

2

Please refer to the log file for the specific error message that caused the simulation to fail. On 7 Feb 2024 - Update 1, I have several issues with the syntax of

fix 0 all qeq/reax 100000 0.0 10 1.0e-1 reax/c

Please refer to the documentation for the new/correct syntax.

3

Thank you so much. The log file doesn’t show any message in the current error(MPI Job aborted).
The first three parameters of fix qeq/reaxff (100000 0.0 10 ) and damping ratios ( temp 300 300 10 iso 1 1 100) were changed but the results are the same error.

4

Can you show the submission scripts? Your sample is quite small, it may be worth running on 2 to 4 processors to debug it.

5

Before that error, there will have been other error messages.
Unfortunately, you don’t provide the whole output, so we cannot tell directly where LAMMPS chokes.

The first thing to do is to try and run this outside a cluster or parallel machine and without MPI. That should give you the LAMMPS errors more easily directly on the screen.

There are many problems with your input deck:

  • your data file has bonds, angles and dihedrals, and your input uses atom style “full”. ReaxFF all bonded interactions are implicit, so those should not be there and the typically use atom style is “charge”
  • the first line of your ReaxFF potential file says “Reactive MD-force field: Si/H/C April 22 2016” but you are mapping elements C, H and O. That means that your force field parameters are likely bogus. It is a bad habit of many of the ReaxFF force field parameter file providers, that they leave in entries for elements that the force field has not been parameterized. ReaxFF force field parameters have to be parameterized specifically for your system of interest (not just have entries for the elements you need). This is not quantum mechanics.
  • there are commands thermo_modify lost ignore and thermo_modify lost/bond ignore. you should never include those unless atoms should leave the system, and with ReaxFF you have no explicit bonds in the first place. For ReaxFF it can be a problem if atoms leave the system, so it is important that LAMMPS tells you when this happens. Ignoring problems does not solve them.
  • the name of the ReaxFF pair style is “reaxff” and not “reax/c” (it has been like this for years!!) and this is what the documentation says. This suggests you have to read the documentation with the necessary care, which is a bad sign and will lead to more trouble in the future.
  • what is the point of adding the same force vector to all atoms?
    fix 3 all addforce 0.00324 0.00324 0.00324 It will just accelerate the whole system in that direction
  • what is the point of delete_atoms overlap 0.3 all all? After you have built your system with packmol there should not be any overlaps unless you entered the wrong box dimensions, but that is not fixed by deleting atoms.
  • what is the point of running with fix npt for only 100 steps? this is effectively the same as not running at all.
  • if you have to run with a timestep of 0.01 fs after a minimization, there is likely something wrong with your potential parameters (see above).

Altogether, this all looks like you are in over your head trying to do something for which you are lacking the training. You should find somebody local that can assist you. This forum is not a substitute and cannot give you the in-person tutoring you apparently need to lean how to run MD simulations.

I would also suggest to wait before using ReaxFF and gain some insight into your system with a classical molecular force field (e.g. OPLS/AA) before using ReaxFF. This can also provide you with properly equilibrated initial geometries (you will need the bonded interactions there, but have to remove them for ReaxFF) and you can judge from the geometry and structure after switching to ReaxFF whether the ReaxFF parameterization can reproduce the results from the molecular force fields (they often do not do so well, since ReaxFF is optimized for reactions not general structure and solvent-solute interactions like e.g. OPLS/AA).

6

I appreciate your help so much. When I ran the system without MPI, it took more time. I’ll definitely run it in serial to identify errors in log file. I’ll consider all of your valuable advice to get the best results.

The fix “all addforce 0.00324 0.00324 0.00324” consider as an external energy of our system to start a polymerization.

Former code was only a test to check the possible errors. One of the main code was attached with the results. The degree of polymerization is defined as the number of hydroxide ions in the species (species.out) file.

Results.zip (1.3 MB)

7

For the reasons I pointed out, your simulation is complete garbage.

It is evident from the output where the temperature is quickly climbing to nearly 15000K. This is about three(!) times the temperature of the surface of the sun!

This shows two things:

  • you are not paying attention to the results of your simulation (also were ignoring important warnings).
  • you are in dire need of proper training and advising in many of the fundamentals of MD simulations and observing the results.

If you continue the current way, you will be wasting your time.

Sorry for the harsh words. But you are in far more trouble that you are realizing and this is not going to be something that you can correct on your own or with the help of an online forum. You need competent and in-person advising.

8

This makes no sense at all.

9

Thank you so much. I chose LAMMPS software according to advising some scientific articles mentioned this software is a great way for simulating photopolimerization.
It is obvious that it would be a big challenge for me graduated in physics field.
Do you know any individuals or online courses with proper training and advising in this issue?

10

I don’t know you, so I cannot assess which skills and training you already have, and which you need to acquire. From what you have shown, you would need to learn a lot. I don’t think that there can be any online tutorial that can teach you what you need to learn. Typically, you would get this kind of training when you are a student joining a group with expertise in the field and then you get gradually introduced into workflows, procedures, best practices and how to extract meaningful results and plan simulations. The next best alternative would be to find a collaborator that already has the skills and interest in your specific application.

1 Like