[lammps-users] Using GAFF Potential for LAMMPS Simulation

Dear LAMMPS developer and users,

I’m a new beginner using LAMMPS for molecular dynamic simulation and I’m interested in applying the General Amber Force Field (GAFF) potential for my system. But I’m having trouble using it in LAMMPS. I have gone through some websites and emails in the mailing list describing this method, and I’m still getting stuck at the beginning stage so I’m writing this message to look for some help.

It seems that I need to first use AmberTools to assign atomic potential for my system. So I used the Antechamber tool and this gave me an .ac file. But when I was going to convert the file to LAMMPS input for later MD simulation, I wasn’t able to do that. I have tried using the InterMol tool but it didn’t accept the .ac file, and none of the Antechamber output formats meet the InterMol conversion requirements.

I’m not sure if I did this right and I’m wondering if someone who has the experience using GAFF potential on LAMMPS before could offer me some suggestions or examples to try it out.

Thanks in advance for your time and help!


unfortunately, there is very little that can be done based on the limited information and overly general description you provide and the main issue that you describe is not really a problem with LAMMPS but with the InterMol software, so you would have to contact the user support or developers for that software about your specific issue.


Dear Dr. Kohlmeyer,

Thanks for your reply. I know currently I having issue with InterMol. From my knowledge, there are several tools like Moltemplate, InterMol, etc that has been suggested to use for GAFF with LAMMPS. But none of them works well with me. I guess I got stuck somewhere I’m not familiar. So I’m writing the email to see if anyone having the experience about using GAFF with LAMMPS that can offer me some help. Like a more approachable way to adapt GAFF in LAMMPS simulation.

Many thanks!

A lot of LAMMPS use cases do not require an automated tool to convert from one simulation package to another.

You need 3 pieces of information:

  • an initial geometry
  • an initial bond/angle/dihedral/etc. topology
  • an assignment of atom/bond/angle/dihedral types to the atom geometry and bond topology as well as assignment of partial charges

How difficult it is to obtain those depends largely on the complexity of your system and less so on the specific force field you are using.
Example: if you have no bonds and only two atom types, you can easily write an input without any force field specific tools.

Even if you have only parts of the information there are ways to do things in a force field independent form. The most critical step is usually the typing of atoms (since this must be done by its “chemical environment” and not only by element), however for a simple system with a few atom types, this can often be scripted or otherwise automated (you just need to have a table of which number is which type).
A bond topology can often be derived from positions by using simple distance criteria, e.g. based on covalent radii. occasionally augmented by a few manual modifications. or for small molecules it can be scripted (there are some examples using VMD and the TopoTools plugin to build simple molecular liquid mixtures for OPLS/AA that way, but that can work for other force fields as well if it is simple enough to derive the atomic charges). the bond/angle/dihedral typing usually follows the atom typing (i.e. if you have a bond between atom types A and B there will be an A-B or B-A bond type).

so just asking about using GAFF is too generic a question.

at this point it is also difficult to assess whether the problems are due to being unfamiliar with the tool you use or unfamiliar with MD and the concepts of classical force field in general, or whether you run into specific compatibility problems, e.g. due to a version mismatch (i.e. using too old or too new a version of some of the tools you use that are not compatible with each other).

at any rate, if you are using Amber tools to build your system topology you should first ensure that you are able to do a complete system setup from within Amber and run your system with Amber’s MD code.
you will need that anyway as a reference to test whether your conversion to LAMMPS was yielding correct files. for that you need to reach the step where you have a .parmtop file, which should be much easier to process as it should have all information needed to build a LAMMPS input.


Dear Dr. Kohlmeyer,

Thanks for your detailed information and sorry for my ambiguous question at the beginning. I’m not a computational major student and I just started to learn MD. It is a long learning curve for me and I’m really appreciative of your patient explanation.

I have only dealt with reax force field with LAMMPS a couple of times and I found it is not compatible with my system. So I guess my issue is caused by not being familiar enough with MD and the concepts of force field in general. Your answer has clarified a lot of them which I’m very thankful for.

My system was set up by using Chemcraft (a similar software as VMD, but not as strong). Since my system is very small (less than 100 atoms), I used Gaussian to first get an optimized ground-state geometry as my MD initial stage. After that using the Antechamber tool yields parts of the other 2 pieces of information that you mentioned (in the .ac file I found bond, dihedral, and partial atomic charge). From my understanding through your message, what I should do next should be using the Amber tool to run an Amber MD and acquire a .parmtop file, and then using the information in this file to build up the LAMMPS input for further LAMMPS simulation. Is that right?

Thanks again for your time and help!


I have given you all the help and explanations that I am willing to give and have time to give. If you are not that familiar with MD, you need tutoring from a competent local person. You should probably ask your PI for help to find somebody. Usually that can be a local person or it would be somebody that is willing to collaborate with your group and in exchange tutor you. It is very inefficient and time consuming to do this over e-mail and a mailing list is not the right place anyway.


Dear Dr. Kohlmeyer,

Thanks for your time and reply, and sorry for bothering you. I will look for that resource for a complete training on LAMMPS MD.

Have a wonderful rest of the day!