GitLab

AmoebaMultipoleForce did not report particle groups, which could potentially lead to incorrect reordering

Converted the array containing atom block indices for the neighbor list from ushort2 to int.  This removes the hard limit of 2 million atoms.

Fixed units for AMOEBA dipole and quadrupole moment (now returns Debye and Debye * Angstrom, not Debye/10 and Debye*Angstrom/100.)  Rationale is that Debye/10 is not a common physical unit.  Note that CUDA platform still gets quadrupole to be incorrect.

Modified the AmoebaMultipoleForce.getSystemMultipoleMoments method with two extra arguments; toggle whether to evaluate energy/force and select maximum multipole order.

When converting to fixed point, multiply by 0x100000000 instead of 0xFFFFFFFF.  This should be (very very slightly) more accurate, since its reciprocal can be exactly represented in floating point.

Simple fix to combining rules allowing AMOEBA VdW parameters to be zero.  Zero parameter excludes all interactions involving that particle, in agreement with TINKER.

Commented out some minor optimizations in AMOEBA that were making the forces incorrect.  AMOEBA energies and forces now agree with TINKER to within 1e-9 (fractional energy) 1e-5 (fractional RMS force error) for NPT simulations on rectangular water boxes with PME and direct or mutual polarizability.

Began implementing new mixed precision model that does integration in double precision and force evaluation in single precision

Modified calculateElectrostaticPotentialForAtomGridPoint_kernel to apply periodic boundary conditions
Unit test for multipole grid potential changed to system comprised of 216 water molecules; 
Tinker potential values now based on PBC

added tests to TestCudaAmoebaVdwForce for dispersion correction using box of 216 water molecules
AmoebaVdwForce and AmoebaMultipoleForce now throw execptions if cutoff > 0.5*(box size)