Verification vs. validation

[ptmerz]

From the Google doc:

During presentation discussion it was pointed out that this document should be called “Verification” rather than “Validation”. Within NIST the term “validation” means comparison between the simulation results and experiment. Validation means checking an experimental or simulation method against the correct answer. Verification means checking that the method has been properly implemented and this is more the kind of checks we are concerned with here.

[ptmerz]

Verification and validation are terms also used in software engineering. See e.g.:

In simple terms, software verification is: "Assuming we should build X, does our software achieve its goals without any bugs or gaps?" On the other hand, software validation is: "Was X what we should have built? Does X meet the high level requirements?"
[from: Wikipedia: Software verification and validation]

I would argue that most of the tests we talked about for the developer document are in the realm of software verification, as they test basic fitness of the software. We might however have to re-check this once we defined the high-end of our test hierarchy.

For the user document, on the other hand, imagine a bug-free code using Berendsen thermostat and a user expecting a canonical kinetic energy distribution. There, I would argue that the software passes verification, but no validation. While a user might be able to discover bugs in the simulation software (-> verification), I would argue that most of the content of a user document will belong to the realm of validation of the software and the employed methods, algorithms and models.

[ptmerz]

I will for now use above definitions in the README.md (just because I need to use something, not to impose a certain view!). To be adapted if we agree on a different definition.