[daip] Q,U fitting

[Eric Greisen]

Lawrence Rudnick wrote:
> The tars input file I sent you asks TARS to create a model with RM=-100, 
> and add to that some (extremely low amplitude) real data with different 
> RMs.   My intent was to iteratively reduce the amplitude of the -100 
> model and see how the other real stuff was affected, but i didn't get to 
> that stage.


13953.  September 3, 2012           TARS                  Eric
         Found error: the Clean algorithm failed to subtract the
         contribution of the last iteration so the restored plus
         residual or residual only outputs were incorrect.  The task
         also ignored the Q and U values in the data rows when a model
         was given.  Changed it to add the model components to the data
         from the rows that define the frequencies to be used.  Also
         changed it to read a fourth column of weights.  Added a new
         OPCODE to control which Clean algorithm is used and added one
         called CMPL.  The new Clean does a pattern match of the inner
         portion of the complex beam with the RM spectrum modified by
         all possible phase shifts (in increments of 0.1 degrees).
         That shift with the largest sum of the product of the real
         part of the beam with the real part of the shifted row plus
         the imaginary part of the beam with the imaginary part of the
         shifted row determines the location of the peak in RM with the
         amplitude of the sum and the phase shift setting the Clean
         component amplitude and phase.  Added code to fit the width of
         the real part of the RMTF to give the width of the Clean
         restoring Gaussian (as a default).  Changed all floating-point
         computations to double precision - it makes a difference when
         working with noise-free problems at least.
         Moved nowhere.

13954.  September 3, 2012           FARS                  Eric
         Found error: the Clean algorithm failed to subtract the
         contribution of the last iteration so the restored plus
         residual particularly or residual only outputs were incorrect.
         Added fitting of a Clean beam to the real part of the RMTF.
         Moved nowhere.


Note that Leonia zeroed the input data if there was a model.  I changed 
that to add the model to the data - forcing you to set the data to zero 
if you want only the models.  I have added reading weights as a 4th 
column to TARS and lots of other things indicated above.

Some aspects of TARS (weights mostly) have not been extensively tested.
Earlier changes were tested closely with the -100 model and frequencies 
you provided.  I did try a model with 2 components (-100 and +200 so 
well separated) and the results were less than sterling.  The dirty 
"beam" RMTF I guess it is called is so lousy that even in this clean 
case the maxima of the dirty image were shifted away from the correct 
pixels.  When restored with a suitable Clean beam the peaks moved to the 
correct pixel and were nearly the right amplitude and phase.

My new method gives similar results to Leonia's Clean and is much 
slower.  I haver left it in TARS for testing but not put it into FARS 
unless you decide you want it.  TARS prints the Clean components it 
finds in both methods.

I have not tested FARS at all, just made the changes indicated.  They 
should be okay but you never know.  Try them please at your earliest 
convenience.

Eric