Test compiled union potential.

[joaander]

Description

Test the compiled union potential added in glotzerlab/hoomd-blue#1725.

Motivation and context

Ensure that the new code operations correctly.

How has this been tested?

I executed the lj_union subproject:

Checklist:

• I have reviewed the Contributor Guidelines.
• I agree with the terms of the HOOMD-blue Contributor Agreement.
• My name is on the list of contributors (CONTRIBUTORS.md) in the pull request source branch.

[tommy-waltmann]

Is it typical for some integration method/thermostat combinations to have error bars slightly outside the accepted value?

[joaander]

Is it typical for some integration method/thermostat combinations to have error bars slightly outside the accepted value?

Yes. The statistical sampling leads to a variety of measured means depending on seed and other factors. The "known" means for P, rho, and U are also a sampled value with unknown error bars (assumed 0 here), so the shown error bars are actually smaller then they should be by an unknown amount. If we assume that the error on the known mean is similar to the error on the measured mean, then the error bars should be off the dotted line ~40% of the time.

The known mean for is exact. The large spread and mean deviation for MTTK shows that MTTK is not as accurate as Bussi.

When these tests fail, they fail obviously - with means many tens of error bars of separation and clearly distinct distributions in the histogram plots.