teuben / dc2019

Improving Image Fidelity on Astronomical Data: Radio Interferometer and Single-Dish Data Combination (workshop material)
9 stars 17 forks source link

casa5 and casa6 show up to 10% differences in tclean for INT, but not TPINT #15

Open teuben opened 4 years ago

teuben commented 4 years ago

When the INT (interferometry only) visibilities are imaged, there is an interesting polka dot like difference between the results from CASA5 and CASA6, with peaks to ~10%. However when the TPINT (TP vis + INT vis) are imaged, the difference is just numerical noise. Very similar pattern when the original skymodel is smoothed with imsmooth. those two maps also differ on the 1e-6 level with a hashed pattern and a slight imprint of the underlying signal.

teuben commented 4 years ago

CASA is obviously flipping a coin on me. Comparing casa54 and casa57 reverse this: INT is numerical noise, TPINT shows a polka dot. And between casa54 and casa61 it's both TPINT and INT that have a 0.015 RMS difference. go figure.

teuben commented 4 years ago

When i say up to 10% this does sound a little worse than it is on average. In the 2" maps I used here, the mean flux is about 1 Jy/beam, the rms is about the same, but the peak is nearly 8. The difference map has peaks (the polka dots) of about plus and minus 0.1, but the RMS in this map is 15 mJy/beam So the pessimist says 10%. The optimist says 0.2% - still , it's the pattern that counts, and the fact that others show numeric noise 1e-6 is the factor that can worry.