CustomNonbondedForce long-range correction fails to converge

Question

I'm still collecting information on this, but it appears my TIP3P water box where the softcore Lennard-Jones interactions of one water with all other waters are moved to a CustomNonbondedForce fails to converge:

======================================================================
ERROR: Testing alchemical fidelity of TIP3P with reaction field, switch, dispersion correction...
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/Users/choderaj/anaconda/lib/python2.7/site-packages/nose/case.py", line 197, in runTest
    self.test(*self.arg)
  File "/Users/choderaj/code/yank/yank.jchodera/tests/test_alchemy.py", line 169, in alchemical_factory_check
    compareSystemEnergies(positions, [reference_system, factory.createPerturbedSystem(AlchemicalState(0, 1, 1, 1))], ['reference', 'alchemical'], platform=platform, precision=precision)
  File "/Users/choderaj/code/yank/yank.jchodera/tests/test_alchemy.py", line 111, in compareSystemEnergies
    state = context.getState(getEnergy=True, getPositions=True)
  File "/Users/choderaj/code/yank/install/lib/python2.7/site-packages/simtk/openmm/openmm.py", line 4802, in getState
    self._getStateAsLists(getP, getV, getF, getE, getPa, enforcePeriodic, groups)
  File "/Users/choderaj/code/yank/install/lib/python2.7/site-packages/simtk/openmm/openmm.py", line 4754, in _getStateAsLists
    return _openmm.Context__getStateAsLists(self, *args)
Exception: CustomNonbondedForce: Long range correction did not converge.  Does the energy go to 0 faster than 1/r^2?

Will update with more information.

该提问来源于开源项目：openmm/openmm

Answer 1

That was it! I looked through the serialized system, and sure enough:

                                <particle param1=".306647338784" param2=".8803136"></particle>
                                <particle param1="0" param2="0"></particle>
                                <particle param1=".339966950842" param2=".4577296"></particle>

Not sure how this is happening, and I realize I don't have the provenance of this example. Will regenerate in a way that ensures to sigma parameters are set to zero.