DiPhyxDiPhyx
Workspace
The whole method

Force, then a step,a hundred million times

Molecular dynamics is one short loop. Everything else in the field is a detail of how the force is computed or how the step is taken.

Compute the forces

Differentiate the potential energy with respect to every position. This is where almost all the time goes.

Take a step

Newton's second law, integrated with a velocity Verlet scheme, over one or two femtoseconds.

Repeat

Write a frame every few thousand steps. The trajectory is the output; the rest is analysis.

The force field

Where the physicsactually lives

The integrator is settled science. The force field is a fitted approximation, and it decides what your run can be right about.

Atoms are treated as points with a charge and a radius, bonds as springs, and electrons not at all. That single simplification is what makes a million atoms affordable, and it is also the boundary of the method: no bond breaks, no charge moves, no chemistry happens.

Bonds, angles, torsionsHarmonic springs and periodic terms. Cheap, local, and rarely the reason a result is wrong.
Van der WaalsA Lennard-Jones term with a cutoff, giving atoms a size and a weak attraction at range.
ElectrostaticsLong-ranged, so it cannot simply be cut off. Particle Mesh Ewald handles the tail on a grid.
No electronsCharges are fixed. Bonds cannot break, and reactions need QM/MM or a reactive force field instead.
Two hard numbers

A femtosecond step,a microsecond run

The timestep is set by the fastest motion in the system, and the run length by what you can afford. The gap between them is the whole problem.

2 fsTimestep4 fs with bonds constrained
µsA useful runa few hundred million steps
10⁵System sizeatoms, for a protein in water

A bond involving hydrogen vibrates every ten femtoseconds or so, and you need several steps per vibration, which pins the timestep at one or two femtoseconds. Constraining those bonds with LINCS or SHAKE buys you a step of four. Getting to a microsecond therefore means a few hundred million steps, and that is why this is a rented-machine activity rather than a laptop one.

Making it realistic

Ensembles, boxes,and the real limit

A few standard choices turn a box of atoms into bulk matter at a temperature. Then sampling decides whether the run means anything.

Periodic boundaries

An atom leaving one face re-enters the opposite one, so a small box behaves like bulk material.

A thermostat

NVT: fixed particles, volume and temperature. Velocity rescaling or Nosé-Hoover holds the temperature.

A barostat

NPT: pressure held instead of volume, so the box finds its own density. What you equilibrate in.

Explicit solvent

Real water molecules, which is most of your atom count and most of your compute bill.

A single trajectory explores whichever basin it started in. If the transition you care about takes longer than the run, you will see a perfectly stable, perfectly converged-looking result that is simply the wrong answer. Run several independent replicas from different starting velocities, and treat agreement between them as the evidence — not the smoothness of any one of them.

Run replicasThree short independent runs tell you more about convergence than one run three times as long.
Discard the startThe start of a trajectory is equilibration, not data. Check where properties flatten before analysing.
Distrust a smooth resultA flat RMSD often means nothing happened at all, not that the system has found a stable state.
On a machine

GROMACS isalready built

Nothing here needs compiling. Pick a machine, and these two commands put it on there.

hub://gromacs

$ dxflow workflow create --identity gromacs hub://gromacs

$ dxflow workflow start gromacs

Start a machineWhat GROMACS needs
Read it, then run itPrepaid and by the hour, on a machine that is yours about a minute after you ask.
Start a machineAll notes