Sporadic segfaults with AmoebaMultipoleForce and CustomHbondForce

[Mary Van Vleet]

I'm in the process of putting together a force field that uses both AmoebaMultipoleForce and CustomHbondForce, and experiencing sporadic segmentation faults. My test files are attached; the test in each subdirectory can be run with a

Code: Select all

 
 $ bash submit_all.sh 
 

command, which *should* calculate the interaction energies of a benzene dimer and each of the monomers separately.

I don't have any diagnosis for the problem, but the following observations may help:

• 1. When running these calculations with only the AmoebaMultipoleForce (see subdirectory 01_just_multipoles), I experience no segfaults.
  2. Similarly, when running these calculations with only the CustomHbondForce (see subdirectory 02_just_hbond_force), I experience no segfaults.
  3. However, when computing the energy of the benzene dimer with *both* the AmoebaMultipoleForce and the CustomHbondForce (see subdirectory 03_combined), I get (seemingly random) segfaults. Both the dimer calculation and the monomer calculations are prone to this segfault, though re-running the code repeatedly (shown below) causes different calculations to fail at different times.

Code: Select all

 $ bash submit_all.sh
<simtk.openmm.openmm.AmoebaMultipoleForce; proxy of <Swig Object of type 'OpenMM::AmoebaMultipoleForce *' at 0x7f2f91a6da20> > 18.5378279682 kJ/mol
<simtk.openmm.openmm.CMMotionRemover; proxy of <Swig Object of type 'OpenMM::CMMotionRemover *' at 0x7f2f91a6da80> > 0.0 kJ/mol
<simtk.openmm.openmm.CustomHbondForce; proxy of <Swig Object of type 'OpenMM::CustomHbondForce *' at 0x7f2f91a6da50> > -4.68109553417 kJ/mol
----
<simtk.openmm.openmm.CustomHbondForce; proxy of <Swig Object of type 'OpenMM::CustomHbondForce *' at 0x7f23763bea50> > 0.0 kJ/mol
<simtk.openmm.openmm.CMMotionRemover; proxy of <Swig Object of type 'OpenMM::CMMotionRemover *' at 0x7f23763bea80> > 0.0 kJ/mol
<simtk.openmm.openmm.AmoebaMultipoleForce; proxy of <Swig Object of type 'OpenMM::AmoebaMultipoleForce *' at 0x7f23763bea20> > 8.98131728499 kJ/mol
----
submit.sh: line 35:   792 Segmentation fault      python run_benzene.py
$ bash submit_all.sh
<simtk.openmm.openmm.AmoebaMultipoleForce; proxy of <Swig Object of type 'OpenMM::AmoebaMultipoleForce *' at 0x7f23efb99a20> > 18.5378279682 kJ/mol
<simtk.openmm.openmm.CMMotionRemover; proxy of <Swig Object of type 'OpenMM::CMMotionRemover *' at 0x7f23efb99a80> > 0.0 kJ/mol
<simtk.openmm.openmm.CustomHbondForce; proxy of <Swig Object of type 'OpenMM::CustomHbondForce *' at 0x7f23efb99a50> > -4.68109553417 kJ/mol
----
<simtk.openmm.openmm.CustomHbondForce; proxy of <Swig Object of type 'OpenMM::CustomHbondForce *' at 0x7f1937b1da50> > 0.0 kJ/mol
<simtk.openmm.openmm.CMMotionRemover; proxy of <Swig Object of type 'OpenMM::CMMotionRemover *' at 0x7f1937b1da80> > 0.0 kJ/mol
<simtk.openmm.openmm.AmoebaMultipoleForce; proxy of <Swig Object of type 'OpenMM::AmoebaMultipoleForce *' at 0x7f1937b1da20> > 8.98131728499 kJ/mol
----
<simtk.openmm.openmm.CustomHbondForce; proxy of <Swig Object of type 'OpenMM::CustomHbondForce *' at 0x7fdb06434a50> > 0.0 kJ/mol
<simtk.openmm.openmm.CMMotionRemover; proxy of <Swig Object of type 'OpenMM::CMMotionRemover *' at 0x7fdb06434a80> > 0.0 kJ/mol
<simtk.openmm.openmm.AmoebaMultipoleForce; proxy of <Swig Object of type 'OpenMM::AmoebaMultipoleForce *' at 0x7fdb06434a20> > 8.97823172782 kJ/mol
----

I'm running these calculations on a x86-64 linux machine, and am happy to provide additional hardware/software information as would be useful.

As always, thanks in advance for the help!

Mary

[Mary Van Vleet]

As this post may have been missed, and because the segfaults are probably a bug, I moved the discussion over to the github issue tracker:

https://github.com/pandegroup/openmm/issues/1654