Correlated errors for ErrorModel

[torluca]

Following a discussion during the photo-z symposium between @hangqianjun , @aimalz and myself, we realise that one characteristics where the current ErrorModel can be improved is in introducing correlated errors. As far as I understood, at the moment the errors are estimated independently, implying that the addition of Gaussian noise to model magnitudes happens independently per each waveband. Since fluxes and errors are correlated for an individual galaxy, it might seem reasonable to introduce the possibility of drawing from a multivariate Gaussian with an appropriate covariance matrix.

[sschmidt23]

Inclusion of correlated noise in observed fluxes is actually mentioned as part of the Mexico in-kind contribution when extending DC2 to RAIL via simulations.

[torluca]

Thanks a lot Sam, I wasn't aware of that! I'm going to close the issue then.

[aimalz]

To clarify based on discussion with @torluca, the scope of this issue is just to swap the univariate normal distributions for a multivariate normal distribution defaulting to a diagonal covariance matrix. This would establish the infrastructure we'd use once the in-kind contribution in question arrives at appropriate values for that covariance matrix. However, at the recent Rubin-wide PZ Virtual Symposium, there was a good discussion of how even simplistic versions of that covariance matrix could be used to prepare for commissioning, so having the infrastructure sooner is still valuable.