Here are some notes from watching Eric Greisen try to clean our data.

Inputs to IMAGR for MS-Clean:

- imagrprm(11): this is the bias

- ngauss: number of gaussians to fit

- wgauss: width of gaussians to fit (one should always be 0)

- fgauss: flux limit for each gaussian

How to find these numbers:

1. Run IMAGR with ngauss and wgauss set. IMAGR will convolve the wgauss values with your beam, and then print out these values. You want to these to be roughly three times the previous. (For example, if your beam is 6", you want the other three scales to be 18", 3*18", 3*3*18".) You will have to play around with wgauss values or de-convolve by hand.

2. Once you have wgauss values (w1, w2, w3, w4), write down the peak flux values that IMAGR spits out (f1, f2, f3, f4). This is how you can start calculating the bias.

The bias is defined as: (f1 / f2) = (a1 / a2)^(bias). Here, f1 and f2 are the peak fluxes corresponding to w1 and w2. a1 and a2 are the beam areas, which you can take to be a = w^2. So,

(f1 / f2) = (a1 / a2)^bias = (w1 * w1 / (w2 * w2))^bias

Figure out a bias that works for most of your resolutions.

3. Now start figuring out fgauss. Run IMAGR again by setting wgauss and imagrprm(11) = bias. Also, set DOTV = 1 so you can clean interactively. You will want to watch it and make sure that it does bounce around a bit. When you are satisfied with the residuals in each field, tell that field to stop cleaning and write down the max flux in the residuals (it's listed in the IMAGR window). This will be what you set fgauss to for that field.

You will want to use a loose clean box around your galaxy. One approach is to make a quick and dirty moment map and make a box that covers all the emission (and then some, probably, because you might have extended structure that clean is not picking up).

Unfortunately, this doesn't appear to be working AT ALL for our robust 0.5 data yet. Some suggestions from Eric:

- use a uv taper to downweight (?) the higher resolutions. We are over resolving our data

- try this on natural weighting (or robust=5).