global sky models must be technically transformed to different inertial reference frames

[Joshuaalbert]

For wide field instruments for large l/m aberration has a noticeable effect. See https://github.com/ratt-ru/codex-africanus/issues/305

Global sky models are defined in helio-centric J2000, and thus when computing visibilities the l/m coordinates of each pixel in the model image changes depending on the obs time, and phase centre coordinate. For narrow fields of view this is ignorable usually. But for instruments like LWA this amounts to pointing offsets of up to 40arcsec.

[Joshuaalbert]

Technically each baseline has dd effect, but an average DD effect could be applied like a beam

[Joshuaalbert]

basically this is accounted for by correcting delay approximations. We take the definition delay_approx(l,m)=u*l+v*m+w*n where l,m are defined as usual independent of time (e.g. see Perley paper). However, the approximation error delay(l,m) - delay_approx(l,m) given the true delay(l,m) is a direction dependent error. We can trust the sky models produced in any reference frame so long as we account for the different in delay, which is a DD phase correction on the visibilities, and would require something like IDG to account for, i.e. the image of each baseline would get its own correction.