Summary
For OPLSAA, the strength of the improper forces was corrected.
Previously simulations prepared with the OPLSAA force field, had improper forces which were too weak by a factor of 2.
After this correction, those improper forces have been strengthened to the correct value.
The impact of this change is expected to be small. (See below.)
How this happened
All of this time, I have been relying on the TINKER version of the OPLS parameters. Apparently I have been misinterpreting the numbers in the improper section of that file.
Impact
In OPLSAA, improper forces are only used to impose planar constraints. The strength of these interactions was already quite strong. Strengthening them by a factor of 2 only serves to make these atoms a little bit more co-planar than they were before. So hopefully this correction should not have a large effect on the shape or flexibility of a molecule.
Most papers describing force fields (including OPLS) omit a detailed explanation of improper forces (and their symmetries).
I apologize for this mistake, but there may be additional errors in the way improper interactions are calculated. (See below.)
Remaining sources of error: Symmetry
In the current implementation of OPLSAA for moltemplate, multiple improper interactions are generated for atoms bonded in a Y-shape. (For example, ethylene contains 6 improper interactions, even though the Y-shape only appears twice. See here for details.) That's because the improper angle used to calculate the forces depends on the order of these atoms. The improper interaction is calculated 3 times using all 3 (=3!/2) different ways of ordering the 3 outer atoms.
It is not clear whether the original authors of the OPLS force fields intended all 3 of these (somewhat redundant) interactions to be calculated, or if the strength of these interactions should have been weakened (by 1/3) accordingly.
(If anyone knows the answer, please contact me. Currently moltemplate calculates all 3 and does not weaken these interactions.)
Contributors
I'd like to thank Domenico Marson who found the new parameters used by BOSS and explained them to me.
He has been working on getting the 2023 version of OPLSAA converted to moltemplate format. That version will hopefully be available in moltemplate shortly. (See #104 for details.)