TINKER provides potential energy, gradients and hessian very easily, and has decent forcefield support. However, we might want to use the current architecture to provide support for other MM engines, if only for fun:
Amber: Already included with Gaussian, but only some forcefields are supported. Not a priority.
CHARMM: Energy should be easy. Gradients are apparently reported when forces are requested (they just need to be multiplied by -1). Hessians might be available with vibran command.
GROMACS: Energy should be easy. Gradients, no clue. Hessian can be reported with mdrun -mtx.
LAMMPS: Energy should be easy. Gradients, no clue. Hessian not easily (maybe with this external module). LAMMPS has a Python wrapper, so it should be easier to calculate everything if necessary.
OpenMM: energy can be obtained easily. No clue about hessian and gradients. Maybe with custom forces?
Cases where only some data could be obtained, Python-side differentiation could be performed with autograd or similar libraries.
TINKER provides potential energy, gradients and hessian very easily, and has decent forcefield support. However, we might want to use the current architecture to provide support for other MM engines, if only for fun:
vibrancommand.mdrun -mtx.Cases where only some data could be obtained, Python-side differentiation could be performed with
autogrador similar libraries.