GitLab

Depressing compromise to make the ordering of atoms in improper torsions as consistent as possible with both AMBER and CHARMM

correction by default. Then add a test for using the long-range correction and
_not_ using the long-range correction.  Comment out the test using the
long-range correction since it currently segfaults.  When the fix for the
segfault is added, this test can be reinstated (but the energy has to be
updated)

This change in default makes it consistent with the default behavior for the
classical NonbondedForce, and I would have set that as the default to begin with
if I had known it was supposed to work.

copy-and-paste...

force field utilizes nbfix functionality. Without this commit, using a ff14IPQ
topology file (or one modified by ParmEd to contain off-diagonal LJ elements),
would happily run, but ignore off-diagonal elements.

The new Chen and Garcia RNA force field (see doi:10.1073/pnas.1309392110) also
utilizes off-diagonal terms in its parametrization scheme.

One shortcoming is that the approach taken here -- using a pair parameter lookup
table -- cannot be currently used with a long-range correction because it is not
a `Continuous1DFunction' by virtue of using two `Discrete2DFunction'
TabulatedFunction classes.

Energies here match the values obtained with Amber to 0.01 -- 0.1 kcal/mol
(within 0.1% of the total).

the switchDistance argument to CharmmPsfFile.createSystem().

Also add a check to protect against negative switching distances.

    - Adjust CharmmPsfFile parser to accept multiple blank lines before the
      NATOM record (I saw 1 PSF file like this... not sure how common it really
      is)
    - Removed the shebang line from element.py and removed execute permissions--
      it should never be executed directly (evidenced by the fact that the
      original shebang line was wrong and would never work, anyway).
    - Fix a test to determine if a Bond is part of an Angle in the PSF data
      structures.

(which means that consecutive residues may have the same residue number _and_
name, but are different residues).

former has a single entry for all unique dihedrals. The latter has a separate
entry for every _term_ of every dihedral (and the latter is the one with the
parameters loaded into it). This way, parameter sets can be reloaded at will
without relying on the hack that createSystem previously employed (i.e., backing
up the dihedral array, clobbering it, then putting it back). This way is cleaner
and safer (although functionally equivalent for typical OpenMM use).

boxVectors, so they should be a property computed from the vectors). Also make
sure that any attempts to change the boxVectors correspondingly deletes the
cached copies of the box lengths so they are recomputed if necessary.

different from _all_ of the others.

and angles makes numbers exactly 0 that _should_ be zero (but aren't due to
precision and round-off issues).

Now the box vector code in the prmtop file is fully general for every type of
box that Amber supports.

Again, this will need to be done in the future anyway for general triclinic
support.

    - existence of velocities and box vectors is auto-detected, so
      loadVelocities and loadBoxVectors are unnecessary
    - recognizes NetCDF-format Amber restart files (which can be written by
      tleap, sander, pmemd, and cpptraj).
    - Restart files coordinate/velocity/box fields have a hard width of 12
      characters and does not ensure that fields are whitespace delimited. This
      commit will correctly parse (valid) Amber restart files in which a
      particular coordinate is <= -100 A or >= 1000 A (not uncommon for restart
      files from sander/pmemd).
    - File format (ASCII vs. NetCDF) is auto-detected.

I deprecated the loadVelocities and loadBoxVectors arguments, since the behavior
does not change (and is correct) whether these variables are set or not. A
DeprecationWarning is emitted if users specify these variables.