Refactor thermostats: RNG, callbacks and documentation #3444
Conversation
Check the mean and variance of random noise, and the absence of correlation. Check the absence of correlation between the 3 axes of positions and velocities generated by the thermostats.
Handle the case of PRNGs generating vectors of size 1 to 4. Check for correlation across both axes and quantities (e.g. correlation between translational velocity and rotational velocity). Also test the boost::variant infrastructure.
Use struct inheritance for thermostats. Generate thermostat MPI callbacks using a macro.
API change: system.thermostat.set_npt() now requires a seed.
The Mersenne twister was replaced by the Philox generator. The compressibility has changed from 0.20 to 0.32. This new value is reproducible within delta=0.01 after 20'000 integration steps for different random seeds.
| @@ -94,8 +91,13 @@ extern bool thermo_virtual; | |||
| * parameter structs | |||
| ************************************************/ | |||
|
|
|||
| struct BaseThermostat { | |||
There was a problem hiding this comment.
Can you please explain why the base class is needed? As far as I can tell the inheritance is not used, and in general (mutable) data members in base classes are a bad idea (due to the non-locality of mutation).
There was a problem hiding this comment.
I originally thought about moving most of the thermostat RNG MPI callbacks into the base struct as getters and setters, and only have a few MPI callbacks (not macro-generated) taking as argument the thermostat switch to select from which global object to call the setter/getter, but these plans didn't materialize. I wouldn't mind reverting that inheritance; it's probably too early to look into that yet.
If by non-locality of mutation you mean the ability of Brownian to change the value of the Langevin counter, I'm not sure how we could run into that situation, as having a BaseThermostat method that takes a ref to another BaseThermostat object is not useful.
Codecov Report
@@ Coverage Diff @@
## python #3444 +/- ##
=======================================
+ Coverage 86% 86% +<1%
=======================================
Files 539 539
Lines 25658 25685 +27
=======================================
+ Hits 22264 22290 +26
- Misses 3394 3395 +1
Continue to review full report at Codecov.
|
Description of changes: