Bond constraints about perses HOT 11 CLOSED

This is the problem that @jchodera is observing with the lack of bond. Parmed does not perceive constraints, and so there is just no bond. I'm going to add the constraint stuff in the next PR.

from perses.

Discussed in here.

from perses.

Constraint tolerance is a context parameter in OpenMM, no? Is there a good way to deal with this in GeometryEngine?

from perses.

Constraint tolerance is an integrator parameter!
http://docs.openmm.org/6.3.0/api-python/classsimtk_1_1openmm_1_1openmm_1_1Integrator.html#a311f37a6d317055063bd8b390ba47737

from perses.

Oops. Thanks! Still not accessible to GeometryEngine as written, though. Should GeometryEngine accept constraint tolerance as a parameter? Should it be in the constructor?

from perses.

Oops. Thanks! Still not accessible to GeometryEngine as written, though. Should GeometryEngine accept constraint tolerance as a parameter? Should it be in the constructor?

Are we talking about constraint tolerance or what kind of bond constraints to use?

We definitely want to specify what kind of bond constraints to use, but that is permitted already. See this example. Note that these are all changed to constraints=None in my PR #109

from perses.

Constraint tolerance. GeometryEngine can figure out whether all bonds or H bonds are constrained (but not whether there are angle constraints) in the current code. I assume that bond_length-constraint will not always be exactly 0, and I assumed that the constraint_tolerance would be a good tolerance.

if np.abs(bond_length-constraint)<constraint_tolerance:
    logp_bond = 0.0
else:
    logp_bond = -np.inf

from perses.

I'm confused. Doesn't GeometryEngine take a System as input? The System allows you to enumerate which bonds are constrained via

for index in range(system.getNumConstraints()):
   [atom1, atom2, r] = system.getConstraint(index)

from perses.

Yes, I know that. I can tell which bonds are constrained. The problem is comparing the constraint to the actual bond length, and determining whether that difference is within the tolerance.

from perses.

if we did something like

if np.abs(bond_length-constraint)==0.0:
    logp_bond = 0.0
else:
    logp_bond = -np.inf

it would probably fail every time.

from perses.

Concluded that we will assume constraints are iterated to convergence, so if there is a bond constraint, we will use a logp of 0.0.

Resolved in #113

from perses.