ECORR and enterprise=True

[vhaasteren]

There is an issue regarding ECORR and consistency between Discovery and Enterprise. For ECORR, Discovery currently has a setting enterprise=True that should make Discovery do the same thing as Enterprise. However, I believe this to be a bug. See this Enterprise issue:

nanograv/enterprise#355

This has actually already been addressed in fastshermanmorrison, so if that is installed Enterprise will do the same thing if EcorrKernelNoise is used and fastshermanmorrison is installed. It would be nice if someone were to comment on the issue on github so that I can create a PR for it. I think the original Enterprise way is wrong. We could allow for an option to enable that functionality, but it should not be the default.

[meyers-academic]

@vhaasteren I'm not sure exactly what you're looking for here, sorry. I agree that the original enterprise way seems incorrect -- that if an observation has 1 TOA, ECORR should still be applied to that TOA. It should still be a dense block in overall white noise matrix, multiplied by ECORR...it's just that the size of that block is 1x1. Correct?

For Discovery, it sounds like you're suggesting that either (1) we remove the enterprise=True case, or (2) we give it some other name, and make sure it's not the default because it's not what one would naively expect.

It would be nice if someone were to comment on the issue on github so that I can create a PR for it. Do you mean comment on the enterprise issue on GitHub? I'm happy to do that. From the issue you've linked it sounds like depending on the installation of Enterprise, the behavior is different, which seems wrong.

[vhaasteren]

Hi @meyers-academic , yes, that is right. I believe the 'Enterprise way' is simply wrong. If there's an ECORR, it should always apply to the affected epochs, even if there is only one TOA. Sometimes, especially for other PTAs, there are epochs with multiple observations, and epochs with one observation. ECORR would apply to all those, not just the epochs with multiple observations.

So my suggestion is to change the Enterprise way, and remove the enterprise=True option from Discovery.

I'm aware it's a breaking change. But it's simply removing an inconsistent modeling method that would otherwise stay to confuse people in thinking it's ok to do it that way

Another option is to raise an error unless a specific override option is set somewhere