Drag terminal velocity tests #1757
Conversation
..aka the drift in case of the electrostatics
Codecov Report
@@ Coverage Diff @@
## python #1757 +/- ##
========================================
Coverage ? 57%
========================================
Files ? 390
Lines ? 27164
Branches ? 0
========================================
Hits ? 15508
Misses ? 11656
Partials ? 0Continue to review full report at Codecov.
|
|
Same question here. The Langevin test tests the decay of v/omega for T=0 and the diffusion coefficient. |
|
It tests a viscous acceleration. It focuses not on the exponential details (already validated by the decay tests), rather that the terminal velocity is stable for a long run. Hence, the BD and LD will match. However, it requires also too much of the test runtime. This is why I've relocated this part to the non-automically run test mass-and-rinertia_per_particle_rotdiff-longrun.py in the #1842. |
|
On Sun, May 27, 2018 at 03:38:56AM -0700, Bogdan Tanygin wrote:
It tests a viscous acceleration. It focuses not on the exponential details (already validated by the decay tests), rather that the terminal velocity is stable for a long run. Hence, the BD and LD will match. However, it requires also too much of the test runtime. This is why I've relocated this part to the non-automically run test mass-and-rinertia_per_particle_rotdiff-longrun.py in the #1842.
The runtime issues for this mass-and-rinertia tests will probably go away, once an
upcoming pr disables the short-range loop entirely for cases, where
it is not needed.
So the main issue for the entire mass-and-rinertia test for me is
clarity.
|
@RudolfWeeber thanks, sounds like an important PR which I need to take into account in further development/refactoring. Could you please clarify corresponding PR or the issue number with some extra details? Just to understand the idea.. |
|
On Mon, May 28, 2018 at 12:30:08PM -0700, Bogdan Tanygin wrote:
> once an upcoming pr disables the short-range loop entirely for cases, where
it is not needed.
@RudolfWeeber thanks, sounds like an important PR which I need to take into account in further development/refactoring. Could you please clarify corresponding PR or the issue number with some extra details? Just to understand the idea..
No PR yet. This may or may not make it into 4.0. So, if the short
range cutoff is 0, the short range loop will not
do anything, because, effectively, no short range potential is
active. Still, a pair search is executed. The PR would disable the
short range loop in this case. This will not result in user-visible
changes (except performance).
This is mostly a corner
case for things such as the thermostat and mass and rinertia tests,
but can also occur for tracer particles in a fluid.
|
|
This PR depends on #2021. |
|
I've ported the corresponding validations to the BD tests within #1842 . Similar checks of the LD viscous properties without a testing performance degradation will be possible after the above-mentioned PR. I suggest to close the current PR and to return to it (#2028) later. |
Fixes #1360, also it contains the integrated changes from PRs #1622, #1757 which were required to validate both existing Langevin- (LD) and new Brownian Dynamics (BD) viscous and fluctuation dynamics including the rotational one. Description of changes: - Generally, this is a conventional simplest BD [schlick2010]. - Corresponding Sphinx documentation section is populated with the required details to reveal the concept. - A velocity, even though it is optional for a simplest implementation of the BD, has been added for a consistency with the existing VV loop (c.f. #1360 (comment)) and all the existing tests and new tests. It does not impact the positional mechanics though which is natural for the BD [schlick2010]. - The velocity and position random walk has been added according to the classical diffusion equations [schlick2010,Pottier2014]. - Rotational motion BD is implemented analogously. Existing quaternion approach has been reused where possible. - New test `brownian_thermostat.py` is based on the existing `langevin_thermostat.py`. It contains some commented out fragments which are inconsistent for the simplest BD cause ones correspond to more fine dynamical structure compare to the BD temporal relation `time_step >> mass / gamma`. The Maxwell distribution test is passing well and the Gaussian noise type is crucial for this which has been done in the `thermostat.hpp`. - The `mass-and-rinertia_per_particle.py` contains changes described within the PRs #1622, #1757. Probably, it is of interest for both the LD and BD independently. This is why I've opened that PRs in advance. - Existing automatic tests execution timeline is minimized. - Additional "heavy" test has been added without an automatic run: `mass-and-rinertia_per_particle_rotdiff-longrun.py`. It requires ~2 hours of a run and provides the intensive validation of the existing LD and new BD dynamics in the same way. PR Checklist ------------ - [ ] Tests? - [ ] Interface - [ ] Core - [ ] Docs? _References_ schlick2010: https://link.springer.com/book/10.1007%2F978-1-4419-6351-2 Pottier2014: https://link.springer.com/article/10.1007/s10955-010-0114-6
One of tests which validate both Velocity Verlet LD and future BD. Hence, it indirectly relates to #1360.
Description of changes:
mass-and-rinertia_per_particle .......... Passed 26.20 sec->
mass-and-rinertia_per_particle .......... Passed 58.29 secc.f. Anisotropic rotational diffusion test #1622 (comment)
PR Checklist