How to Choose a Force Field for Your GROMACS Simulation (CHARMM36 vs AMBER vs OPLS)

To choose a force field in GROMACS, match it to your molecule type: use CHARMM36m or AMBER ff14SB for proteins, CHARMM36 for lipids and membranes, and a united-atom GROMOS set when speed matters. Select it at the pdb2gmx step, keep its matching water model, and if you are reproducing a published study, use that study’s force field.

The first real decision GROMACS forces on you arrives early. You run pdb2gmx, expecting a topology, and instead get a numbered menu asking you to pick a force field. Many students pick option 1 and move on, then cannot answer the obvious thesis-committee question: why that one? The choice shapes every number your simulation produces, so it is worth getting right. This guide explains what a force field actually is, compares the families you will see in the menu, and gives plain rules of thumb you can defend in writing.

This is a spoke in our learn molecular dynamics with GROMACS series. It assumes you can already run pdb2gmx; if you are not there yet, start with our walkthrough on running your first simulation. To see where this fits into a research career, the full computational biology skills roadmap lays out the path from first install to thesis.

What is a force field, in plain terms?

A force field is the rulebook your simulation uses to compute the energy and forces on every atom. It has two parts. The first is a set of equations, the potential functions for bonds, angles, dihedrals, electrostatics, and van der Waals interactions. The second is the long list of parameters that fill those equations in: bond lengths, force constants, partial charges, and atom radii, fitted to quantum-chemistry calculations and experimental data. The GROMACS reference manual puts it directly: a force field is “built up from two distinct components: the set of equations (called the potential functions) used to generate the potential energies and their derivatives, the forces; and the parameters used in this set of equations” (GROMACS reference manual, Definitions and units).

The practical consequence: a force field is not universal. Its parameters were fitted as a set, against a specific water model and a specific philosophy about how atoms are grouped. Mixing parameters from one force field with water or ligand parameters from another breaks the internal consistency it was validated against. That single idea drives most of the rules below.

Which force fields does GROMACS actually offer?

When you run pdb2gmx, the standard installation lists 15 built-in force fields. They fall into three families: seven AMBER variants, the CHARMM27 set, six GROMOS96 sets, and OPLS-AA/L. The menu looks roughly like this:

1: AMBER03      2: AMBER94      3: AMBER96
4: AMBER99      5: AMBER99SB    6: AMBER99SB-ILDN
7: AMBERGS      8: CHARMM27
9-14: GROMOS96 43a1 / 43a2 / 45a3 / 53a5 / 53a6 / 54a7
15: OPLS-AA/L

Two things surprise beginners here. First, the modern protein force fields you read about in recent papers, AMBER ff14SB and CHARMM36m, are not in this list. They are distributed separately and have to be added to your working directory before pdb2gmx will see them. Second, the bundled CHARMM27 is dated; the CHARMM developers recommend CHARMM36m for proteins, which corrects a known secondary-structure bias in the older set. You download the GROMACS port of CHARMM36 from the MacKerell lab CHARMM force field page and unpack the .ff folder where you run pdb2gmx.

How do the families compare for a beginner?

Each family was built with a different goal, which is why no single one wins everywhere. AMBER and CHARMM are the two most common choices for biomolecular work today; OPLS is strong on small organic molecules; GROMOS is united-atom, meaning it folds nonpolar hydrogens into the carbon they attach to, which makes systems smaller and faster at the cost of atomistic detail. The table below summarises the practical differences.

The pattern worth memorising is the last two columns together: your force field choice fixes which water model and which ligand-parameterization tool you are allowed to use. Pick CHARMM36 and you are committed to CGenFF for any ligand and CHARMM-modified TIP3P for water. Pick AMBER and you are in GAFF and antechamber territory. This is exactly the constraint that turns a casual menu pick into a workflow decision.

What should you actually pick? Rules of thumb by system

These are the defaults most structural-biology labs reach for. They are starting points, not laws, but each one is defensible in a methods section.

For a plain protein in water

Use CHARMM36m or AMBER ff14SB. Both are modern, heavily validated all-atom protein force fields, and for most folded proteins they give comparable, reliable behaviour. CHARMM36m was specifically refined to handle intrinsically disordered regions as well as folded ones, described by its authors as “an improved force field for folded and intrinsically disordered proteins” (Huang et al., Nature Methods, 2017). AMBER ff14SB improved side-chain and backbone accuracy over its predecessor and is documented in Maier et al., Journal of Chemical Theory and Computation, 2015. If you stick with the bundled menu, AMBER99SB-ILDN (Lindorff-Larsen et al., Proteins, 2010) remains a respectable, widely cited choice.

For membranes and lipids

Use CHARMM36. It is the de facto standard for lipid bilayers, and the CHARMM-GUI Membrane Builder generates GROMACS-ready input for it, which is why most published membrane simulations use this combination.

For nucleic acids

AMBER provides well-tested nucleic-acid parameter sets (such as the bsc1 refinement for DNA and OL3 for RNA), and CHARMM36 also covers DNA and RNA. Match the protein and nucleic-acid parameters from the same family rather than mixing.

For a protein-ligand complex

Pick the protein force field first, then use its companion ligand tool: CGenFF for CHARMM36, GAFF or GAFF2 (via antechamber or ACPYPE) for AMBER, LigParGen for OPLS, and the Automated Topology Builder for GROMOS. We cover that ligand step in detail in a dedicated tutorial in this series.

When speed or system size is the bottleneck

A united-atom GROMOS force field such as 54A7 removes explicit nonpolar hydrogens, shrinking the atom count and letting you push larger systems or longer timescales on modest hardware.

When you are reproducing or extending a published study

Use the same force field and water model the original paper used. This is the single most defensible choice you can make, because it lets you compare your numbers against theirs on equal footing.

Do you need to match the water model to the force field?

Yes, and this is the mistake that quietly ruins more student simulations than any other. Each protein force field was parameterized assuming a particular water model, so you should solvate with that model. CHARMM36 expects the CHARMM-modified TIP3P, classic AMBER protein sets expect TIP3P, the newer AMBER ff19SB was developed with OPC water, and GROMOS uses SPC. When you reach the solvation step (gmx solvate) and the ion and run-parameter steps, keep the water consistent with the force field you chose at pdb2gmx. Pairing a force field with the wrong water can shift folding stability and dynamics in ways that look like real results but are artifacts.

How do you justify the choice in a thesis?

A committee does not expect you to have invented a force field. It expects you to show you understood the trade-off. One or two sentences are enough: name the force field and its citation, state why it fits your system (protein, membrane, nucleic acid, or speed-limited), name the matching water model, and, if relevant, note that it matches a prior study you are building on. For example: “Protein parameters used the CHARMM36m force field (Huang et al., 2017) with CHARMM-modified TIP3P water, chosen for its validated treatment of both folded and disordered regions.” That is the difference between a blind menu pick and a reasoned method.

Frequently asked questions

Is CHARMM36 included in GROMACS by default?

No. The bundled CHARMM option is the older CHARMM27. CHARMM36 and CHARMM36m are distributed separately by the MacKerell lab; you download the GROMACS port, unpack the .ff folder into your working directory, and then it appears in the pdb2gmx menu.

What is the difference between AMBER and CHARMM for proteins?

Both are modern all-atom force fields and give broadly similar, reliable results for folded proteins. They differ in how charges and dihedrals were fitted and in their companion water and ligand tools. For most beginner protein work either is a sound choice; pick one and stay consistent across the whole system.

What does united-atom mean in GROMOS?

A united-atom force field does not represent nonpolar hydrogens explicitly. It merges each one into the carbon it is bonded to, treating, for example, a CH2 group as a single interaction site. This reduces the atom count and speeds up the simulation, at the cost of some atomistic detail compared with all-atom force fields like CHARMM and AMBER.

Can I mix two force fields in one simulation?

Avoid it. Parameters are fitted as an internally consistent set against a specific water model. Combining, say, AMBER protein parameters with CHARMM ligand parameters breaks the consistency each was validated for and is hard to defend. Keep the protein, ligand, and water from compatible sources.

Which force field is best for a small drug-like molecule?

There is no separate “best” force field for the ligand on its own; the ligand must match the protein force field. Choose the protein force field, then parameterize the ligand with its companion tool: CGenFF for CHARMM36, GAFF or GAFF2 for AMBER, LigParGen for OPLS, or the Automated Topology Builder for GROMOS.

Written by the StemSkills Lab team, structural and computational scientists with more than 10 years of combined experience in sequence and structural bioinformatics, drug discovery and design, and multiscale molecular modeling.