Array combination (12m + 7m + TP) using Miriad

[xinglunju]

Related issues (array combination using CASA): #122, #192.

I have been using the HNCO line data toward the Brick (region code: Sgr_A_st_ao) to experiment the array combination in Miriad.

Basically, I am following the steps in Baobao Liu's ALMICA script: https://github.com/baobabyoo/almica/blob/main/combine_mosaic.sh with a few tweaks. The script I am using can be found at https://github.com/xinglunju/almica/blob/main/combine_mosaic_xlu.sh

However, I am having issues with the synthesized beam size: when I combined 7m+TP data, the resulting beam size is larger than 60"! I have no clear how this could happen. I have tried several approaches: increasing Tsys of the TP-generated dummy visibility data; decreasing the integration time in the header of the TP-generated visibility data; decreasing the number of visibility data points when generating the TP-generated visibility data. However, none of them works.

The same situation happens for 12m + (7m +TP) combination. The beam size is ~60".

The same method works for VLA+GBT data combination. I am still trying to understand why it does not work for ALMA data. Will also discuss with Baobab.

[xinglunju]

The issue is resolved: previously I cut out 100 channels to test the code, but this caused mismatch between frequency axes of different datasets. If we load the full data (~1920 channels), then the synthesized beam size is correct.

The other issue I encountered is that, if a small fraction of channels associatd with an antenna are flagged (because of being on the SPW edge or bad bandpass solutions or anything suspicious), then the invert task will not take any data from this antenna into account. The solution is to use slop=1,zero with invert (see https://xinglunju.github.io/miriad.html).

The script in the Github repo has been updated.

[xinglunju]

What the Miriad script does

1. Make dummy visibility data toward each of the 7m pointings using the TP image. Each 7m pointing has a corresponding dummy TP visibility dataset.
2. Combiningly image the 7m+TP data, obtain a 7m+TP mosaic image.
3. Make dummy visibility data toward each of the 12m pointings using the 7m+TP mosaic image. Each 12m pointing has a corresponding dummy 7m+TP visibility dataset.
4. Combiningly image the 12m+7m+TP data, obtain a 12m+7m+TP mosaic image.
5. Optionally, combine the 12m+7m+TP mosaic image and the original TP image using the immerge task (which is similar to but different with feather). This is what I chose to do.

Other notes

Running the whole script takes about 10 hours. We don't clean the data quite deep (1000 iterations for all clean tasks). For higher iteration numbers longer time is needed.

Parameters including artificial Tsys and data point density in the uv plane assgined to the dummy visibilities can be adjusted. We are not necessarily using the best combination. This will need more experiments.

Comparison with CASA

Mom8 (peak intensity) maps, plotted at the same color scale. Left: Miriad. Right: CASA (see issue #122).

• Overall, the features in the two images are consistent, but CASA recovers extended emission better. Note that the immerge task in Miriad, which is to combine interferometer and TP images, has parameters to recover more extended emission, so in principle we can play with the immerge parameters to eventually obtain an image similar to the CASA one.
• The Miriad image has noiser edges, because there exist only 7m+TP data outside of the primary beams of the 12m pointings it is primary-beam corrected.

One channel (Vlsr ~ 15 kms), plotted at the same color scale. Left: Miriad. Right: CASA

• Miriad recovers less extended emission than CASA does, but does a better job at suppressing negative bowls and handling mosaic edges.

Table

Image properties	Miriad	CASA
Beam size	2.46" x 1.79", PA=-88deg	2.61" x 1.82", PA=-86deg
rms per channel	11.1 Jy/beam	11.4 Jy/beam

[keflavich]

Is there some excess speckling in the top-right corner of the Miriad version?

[xinglunju]

It is because of the channel range used in immoments... The mom8 map used all the 1918 channels. To get rid of the noisy channels on the SPW edges, we can use a narrower range of channels, from 250 to 1650 (1400 channels, same as the CASA image), then the map looks better and the speckling is gone.

The updated plot:

[xinglunju]

Things to be done

• Try Marc's immoment script?
• Use mosmem (maximum entropy method) instead of mossdi