Discussion of ZeroVirialCalculator

[cortner]

I'm not sure what the actual intention is. If the intention is what you say in the documentation then that is incredibly dangerous. If a computation gets a zero-virial for a calculator that just doesn't have virial implemented then that will silently give incorrect results.

The correct way - in my view - to implement this is to return missing.

[tjjarvinen]

Basicly, if calculator does not support virial, there should not be available virial call to inform used it does not exist.

Second thing i-PI interface requires virial to be returned, even if it is not used.

So, you could say that ZeroVirialCalcualtor is implemented as a helper for i-PI calculator. The easiest way to use i-PI calculator for constant volume simulation is to wrap your calculator into ZeroVirialCalculator. That will then allow easy use of calculators that do not support virial. Also, you do that by knowing you cheat with virials is better than an implementation that does that under the hood.

[cortner]

I don't have a problem with the i-PI thing. But that is not what the documentation says right now. It makes sense in a language that requires specific type a priori. In Julia this does not make sense.

I don't have a problem providing that utility calculation, but the documentation needs to be fixed. It could even be given the name UnsafeZeroVirials to highlight that this should not normally be done unless you really know exactly what you are doing.

[cortner]

based on the notes you added should this be closed?