A few days ago I apparently got my broken simulator working again. Today, in my first attempt to repeat those results, I... have broken it again.:

WARN        FTMachine::initMaps     No overlap in frequency between image channels and selected data found for this FTMachine
WARN        FTMachine::initMaps+     Check your data selection and image parameters if you end up with a blank image

This warning is not a simple warning, it means nothing is being done. My images are empty, pure noise (which, interestingly enough, is far from gaussian).

I have encountered these errors before <|static|/separating_line_and_continuum.rst>, but unfortunately did not resolve the issue.

Turns out, the difference was the .ms file I used, continuum_7m12m_noflag.ms, which works, vs the useful one, w51_contvis_selfcal_3.ms, which doesn't.

Checking out their headers, the continuum image does overlap with the data but the selfcal version doesn't. Wonderful. Now I'll bet it's only populating one channel at a time when I invert too...

This leads me to try:

split(vis='w51_contvis_selfcal_3.ms', outputvis='simulation_continuum.ms', datacolumn='data', width=100)

then use (borrowed from this tool):

mymsmd = msmdtool()
mymsmd.open(vis)

reffreq = "{value}{unit}".format(**mymsmd.reffreq(spw)['m0'])

to get the frequency to write into the header each time.

At a glance, just changing the frequency of the FITS file to match the first SPW seems to work. However, I haven't yet examined whether the other SPWs are populated.

This assumption proved wrong. No, it simply doesn't work. The warnings went away, but the output images are empty. Also, the noise went up, which I don't understand.

Imaging just spw1 kind of worked, but resulted in an image with the wrong resolution by factors of many.

Trying to loop over all spws failed miserably, resulting in nothing being done by predict and no warning being given.

OK, finally, solution is to make sure CDELT is large enough to cover all SPWs.

Another important note: Jy/beam and Jy/pixel inputs into CASA result in identical images. In both cases, they are wrong images: they correspond to multiplying the input data by factors of at least 100. Therefore, my conclusion now is that I have to divide everything by some fact, which I sincerely hope is ppbeam, to correct the data.

It looked vaguely like the divide-by-ppbeam trick worked, but my tests kept coming up funny because the original images were scaled down. So I tried unscaling them. Somehow, without changing anything about the headers, I landed on another failed predict error:

2016-04-27 19:14:42 SEVERE  Simulator::createSkyEquation() (file /Users/rpmbuild/gradle/workdir/casasources/release-4_5/code/synthesis/MeasurementEquations/Simulator.cc, line 2200)        Caught exception: (/Users/rpmbuild/gradle/workdir/casasources/release-4_5/code/synthesis/MeasurementEquations/Simulator.cc : 2266) Failed AlwaysAssert spectralIndex>=0
2016-04-27 19:14:42 SEVERE  Simulator::predict() (file /Users/rpmbuild/gradle/workdir/casasources/release-4_5/code/synthesis/MeasurementEquations/Simulator.cc, line 2118)  Failed to create SkyEquation

Spectral Index? I encountered this previously, but didn't record any solution. After checking the header, it appears that my files are now missing axes 3 and 4. I didn't make any changes that should cause this; the axes are just not being added on importfits despite being forcibly added in the command.

The problem is unfortunately not in the FITS part, but again in the CASA part. Re-importing the FITS file to .image without forcing defaultaxes again just broke it.