Inconsistency of Ions::setup when used multiple times

[dreamer2368]

Ions::setup is supposed to update the list of overlapping VL ions (overlappingVL_ions_). However, it is observed that the resulting overlapping ions are different even when the same local ion list is used.

A minimal example is implemented in ion-bug branch. A test executable test_overlapping_ion will be available if the code is fully installed (make install). This branch does not need libROM and can be compiled the same as release branch.

The executable test_overlapping_ion is compiled from src/test_overlapping_ion.cc. This works the same as the main mgmol driver mgmol-opt. In essence, it does the following job:

• runs Ions::setup 3 times with 3 different ion configurations
• save the corresponding overlapping VL ion lists
• runs another Ions::setup with one of (previously used) 3 configurations
• check if the resulting overlapping VL ion list is equivalent to the saved list

Example config and coordinate file are stored in examples/Carbyne-bug. Once the executable is installed, one can run the following script to check the test failure:

ROOT_DIR=/path/to/your/mgmol/repository
INSTALL_DIR=/path/to/mgmol/install/directory

cd ${ROOT_DIR}/examples/Carbyne-bug
srun -n 144 ${INSTALL_DIR}/bin/test_overlapping_ion -c carbyne.cfg -i carbyne.in

The test passes with the PinnedH2O example, but not with Carbyne example. If we want to manipulate local ion instead of overlapping VL ion in PR #274 , this bug must be fixed first. @jeanlucf22 , can you take a look at this bug?

[jeanlucf22]

You should follow this example: https://github.com/LLNL/mgmol/blob/release/tests/testIons.cc
The way you are doing it in this branch is not correct: you are setting local atoms with coordinates outside the local MPI sub-domain.

[dreamer2368]

I'm simply perturbing local atoms by 10% from their original locations. Would that be too large to go beyond the local subdomain? Regardless, shouldn't the output of Ions::setup be the same for the same input?

I'll try to set up the test following testions.cc.

[jeanlucf22]

Yes, 10% is enough to cross over into another subdomain

[dreamer2368]

While I could make a fictitious Ions with custom hard-coded species as in testIons.cc, I'm hoping to use this test for any config file that we actually use for production. I hope to move local ions' positions randomly within local MPI subdomain, from the existing Ions that is already set from the initial setup. I guess this should be similar to updating ion locations in md, but I couldn't really pinpoint what part I should pick and use. Can you point me to those functions or guide me how to do this properly?

[dreamer2368]

This problem is solved using Ions::setPositions, which sets ion locations globally over all processes. The resolved version is pushed to the branch ion-bug.