Bugfix in Nose-Hoover chain integrator

[Olllom]

@andysim and I realized that constraints had not been taken into account when calculating the number of degrees of freedom in the NoseHooverChainVelocityVerletIntegrator.
As a result, the system temperature did not converge to the target value for systems with constraints.

I added a test to make sure that the target temperature is now matched.

The fix changes the API of the constructor: It requires the system to be passed as an argument (only in order to extract the correct number of constraints; the integrator then forgets about the system).

Cheers,
Andreas Krämer

• Implement feature / fix bug
• Add tests
• Update documentation as needed
• Update changelog

[andrrizzi]

Looks good to me! Could you fix the doctests and the regular tests that got broken with the API?

[andrrizzi]

In particular, some tests in test_integrators_and_testsystems.py and many tests in test_integrators.py (search for integrator_class()) assume that you can instantiate an Integrator without arguments, so if we don't make system an optional argument we need to create an exception instantiating it in the test code.

[Olllom]

@andysim Thanks for the catch.

@andrrizzi
Instead of adding an exception to all the tests, I made system=None a keyword argument of the constructor.
The integrator will fall back to the old behavior, which is correct for unconstrained systems. A severe warning will be printed, if the system is not specified.

• This avoids changing all the tests in test_integrator.py
• In contrast, an exception was added to test_integrator_and_testssystem.py - it actually makes of sense to test if the integrator plays nicely with all test systems.

[andrrizzi]

Sounds good, thank you!