Smear Fitting Info

Smear fitting is an image deconvolution technique for interferometry that I have developed as an alternative to CLEAN or Maximum Entropy. It fits a model made of up to several dozen components, usually elliptical gaussians, to the visibilities, and then broadens it to account for uncertainty in the measurements. Despite the broadening step it generally produces sharper resolution than CLEAN, while using the full sensitivity of natural weighting. The properties of the model let it avoid many of the problems of maximum entropy. (It can be thought of as a variation on maximum entropy, with a more data-driven basis for the model.)

Smear fitting has been implemented as a modification to difmap, and is freely available as a patch.

This version has hyperlinked references, and IMHO better formatting than allowed by the University of Toronto thesis rules. Don't settle for the bootleg copies being peddled on street corners!

A note on the amount of smearing:

(Also given in the Using smerf section.)

With the default settings and properly calibrated data smearing should convolve each feature with a beam corresponding to the probability distribution of its extent on the sky, i.e. it smears to one standard deviation in position, major and minor axes, and position angle. I think this matches what people intuitively expect when they look at an image, but you may want to use two or more standard deviations, to

• account for systematic errors,
• be conservative, or
• just to see what it looks like.

There are two ways to adjust the smearing "knob": change the third parameter of smearall (read the manual and/or help smearall), or wtscale (ditto). The latter way is not recommended, but often happens by accident. I strongly recommend reading AIPS Memo 108, Weights For VLA Data before smearing or preferably at the FILLM stage. And of course you should read help wtscale in difmap/smerf, but you've all done that already, right? Right. If wtscale is too low or high, smearall will yield respectively more or less smearing than you requested. "vplot 4" in difmap/smerf is useful for checking whether wtscale is way off (it is often left unset or set to the wrong value by the telescope), since it shows whether the error bars are too large or small. A more precise check is to compare the rms predicted by wtscale to the rms seen in an empty patch of sky, optionally in Stokes V. You may even prefer to empirically set wtscale using the observed "empty" rms.

Last modified: Mon Jun 4 16:40:56 EDT 2007 .
Rob Reid,