Decide on footprints and areas for gg and gk
msyriac opened this issue · comments
We have overlap mainly in Deep56 and Deep8 for S16.
- Should we restrict the cross-correlation to exactly the "Steve mask" region?
- Should we extend the gg auto-correlation to the full Y3 footprint? Does this matter for sample variance cancellation? Do we care about sample variance cancellation?
I think the answer to the first one is probably yes, as that is our nominal region (for S15 at least). If it is too narrow, that should be brought to the attention of everyone doing act lensing, not just this paper
For the second point, I'm thinking yes, unless that makes things too difficult to get a joint covariance matrix.
@ajvanengelen agreed!
@xuod Good point. I would guess that the gg-gk covariance is negligible if we use 800 sq.deg. of gk overlap and 5000 sq.deg. of Y3 gg. If we checked this with gaussian sims, would that be enough?
Another option is to have 3 data vectors {gk,gg_o,gg_no} where gg_o is calculated in the overlap region and gg_no elsewhere.
Yep, probably worth testing it. For the other option, I guess that might introduce more covariance between data vectors. Perhaps worth testing too.