SHAKE is a constraint algorithm. RATTLE is an integration algorithm. They're different because they do different things! Here's an implementation of RATTLE implemented with a CustomIntegrator. CustomIntegrator integrator(dt); integrator.addPerDofVariable("x1", 0); integrator.addUpdateContextState();...

Those algorithms are just different methods of solving the same system of equations. They all produce identical results (up to the tolerance you specify). The velocities are updated correctly in all cases.

8.1.1 is now released. You can install it with

Problems in ParmEd need to be reported at https://github.com/ParmEd/ParmEd/issues. I don't think the author reads this forum.

It depends on the platform, not on the integrator. All platforms use SETTLE for water. The Reference and CPU platforms use CCMA for everything else. The CUDA and OpenCL platforms use SHAKE for small clusters composed of hydrogens bound to a single heavy atom, and CCMA for anything else.

Correct, it would not.

Almost all integrators apply constraints at the level of positions, and possibly also velocities. You don't integrate the constraint forces. Instead you integrate as if there were no constraints, then modify the positions to satisfy the constraints. Based on how much they change, you can back-calcul...

You need to move the calls to addParticle() to before you create the Simulation.

Any change you make to a System doesn't affect already existing Simulations. You need to add the particles before you create the Simulation. See the FAQ for details.

A release candidate for OpenMM 8.1.1 is now available. You can install it with mamba install -c conda-forge/label/openmm_rc -c conda-forge openmm This is a patch release that fixes two bugs. See https://github.com/openmm/openmm/pull/4351 and https://github.com/openmm/openmm/pull/4364 for details. Pl...