Openmm

OpenMM

What this library is for

OpenMM is a molecular simulation toolkit for building, parameterizing, running, and analyzing molecular dynamics simulations. Its application layer provides readers, force-field setup, modeller tools, integrators, platforms, reporters, and Simulation objects for standard biomolecular MD workflows.

When to use this vs. alternatives

• Use OpenMM for molecular dynamics, minimization, thermostatted simulations, custom forces, biomolecular force fields, PDB/mmCIF/Amber/Gromacs inputs, periodic boxes, PME, constraints, and GPU execution.
• Use RDKit to prepare or standardize small molecules before force-field parameterization; RDKit is not an MD engine.
• Use ASE for atomistic calculator workflows, MLIPs, DFT-style relaxations, and materials trajectories; use OpenMM when the method is a molecular mechanics force field or OpenMM custom force.
• Use PySCF for quantum chemistry reference calculations, not long classical MD.
• Do not write a Verlet/Langevin loop in NumPy for normal MD tasks. OpenMM handles units, constraints, neighbor lists, long-range electrostatics, platforms, and reporters.

Canonical workflow

Use the application layer unless the user is explicitly building custom low-level forces.

from sys import stdout

from openmm import LangevinMiddleIntegrator, Platform
from openmm.app import (
    HBonds,
    PME,
    DCDReporter,
    ForceField,
    PDBFile,
    Simulation,
    StateDataReporter,
)
from openmm.unit import kelvin, nanometer, picosecond

pdb = PDBFile("input.pdb")
forcefield = ForceField("amber14-all.xml", "amber14/tip3pfb.xml")

system = forcefield.createSystem(
    pdb.topology,
    nonbondedMethod=PME,
    nonbondedCutoff=1.0 * nanometer,
    constraints=HBonds,
)
integrator = LangevinMiddleIntegrator(
    300 * kelvin,
    1.0 / picosecond,
    0.004 * picosecond,
)

platform = Platform.getPlatformByName("CUDA")
simulation = Simulation(pdb.topology, system, integrator, platform)
simulation.context.setPositions(pdb.positions)

simulation.minimizeEnergy()
simulation.reporters.append(DCDReporter("traj.dcd", 1000))
simulation.reporters.append(
    StateDataReporter(
        stdout,
        1000,
        step=True,
        potentialEnergy=True,
        temperature=True,
        speed=True,
    )
)
simulation.step(10000)

For structure preparation, use Modeller before creating the System.

from openmm.app import Modeller

modeller = Modeller(pdb.topology, pdb.positions)
modeller.addHydrogens(forcefield)
modeller.addSolvent(forcefield, model="tip3pfb", padding=1.0 * nanometer)

For deeper examples, read:

• Running simulations: https://docs.openmm.org/latest/userguide/application/02_running_sims.html
• Model building and editing: https://docs.openmm.org/latest/userguide/application/03_model_building_editing.html
• Cookbook: https://openmm.github.io/openmm-cookbook/

Key conventions and gotchas

• Use modern imports from openmm, openmm.app, and openmm.unit. Old examples using simtk.openmm are legacy and should be ported.
• Quantities need units, e.g. 1*nanometer, 300*kelvin, 0.004*picoseconds. Do not pass bare floats where OpenMM expects a unitful quantity.
• The topology must match the force field. Missing hydrogens, unsupported residues, ions, cofactors, ligands, terminal states, or water models will cause parameterization failures or wrong simulations.
• Force-field and water-model choices are coupled. Record the exact XML files and do not mix incompatible water models, ions, or extra particles.
• Some force fields and water models require extra particles. Use Modeller.addExtraParticles(forcefield) or the documented preparation step when needed.
• Periodic simulations need box vectors and an appropriate nonbonded method (PME/LJPME) and cutoff. Nonperiodic simulations should not silently use a periodic setup copied from a solvent-box example.
• Platform selection affects speed and sometimes numerical reproducibility. Record platform (CUDA, OpenCL, HIP, CPU, Reference) and precision properties for production runs.

Anti-patterns

• Do not run MD directly from an arbitrary PDB without checking missing atoms, protonation/tautomer states, chain breaks, alternate locations, box vectors, ions, ligands, and force-field coverage.
• Do not use RDKit-generated coordinates as a complete OpenMM topology for biomolecular simulation without proper parameterization and atom/residue names.
• Do not omit reporters. Save enough trajectory and state data to diagnose temperature, energy, speed, and crashes.
• Do not compare trajectories run with different force fields, water models, constraints, timesteps, platforms, or precision settings as if only the random seed changed.
• Do not use a 4 fs timestep without hydrogen constraints and a force field setup known to support it.

Diagnostic checks

Before trusting outputs, the agent should:

• Log OpenMM version, force-field XML files, water model, input file, atom count, residue count, box vectors, nonbonded method, cutoff, constraints, timestep, integrator, platform, precision, and random seed if set.
• Run an energy minimization and inspect potential energy before dynamics.
• Check that reporters are writing trajectory and state data at useful intervals.
• Monitor temperature, potential energy, total energy where meaningful, and simulation speed during early steps.
• Validate structure preparation: hydrogens, solvent/ions, extra particles, ligand parameters, protonation states, and periodic box.
• Save prepared structures and serialized systems for reproducibility when setup is expensive or nontrivial.

Pointers to deeper material

• Documentation: https://docs.openmm.org/
• User guide: https://docs.openmm.org/latest/userguide/
• Cookbook: https://openmm.github.io/openmm-cookbook/
• Source repository: https://github.com/openmm/openmm
• Paper: Eastman et al. (2017), "OpenMM 7: Rapid development of high performance algorithms for molecular dynamics", PLOS Computational Biology 13, e1005659. https://doi.org/10.1371/journal.pcbi.1005659