[daip] [!4836]: aips - What does 'L1R' do in CALIB

[Eric Greisen]

Eric Greisen updated #4836
--------------------------

       Staff (Owner): Eric Greisen (was: -- Unassigned --)
              Status: Closed (was: Open)
                 Due: - Cleared - (was: 28 April 2014 08:44 PM)

What does 'L1R' do in CALIB
---------------------------

           Ticket ID: 4836
                 URL: https://help.nrao.edu/staff/index.php?/Tickets/Ticket/View/4836
           Full Name: Chris De Pree
               Email: cdepree at agnesscott.edu
             Creator: User
          Department: AIPS Data Processing
       Staff (Owner): Eric Greisen
                Type: Issue
              Status: Closed
            Priority: Default
      Template Group: Default
             Created: 24 April 2014 08:44 PM
             Updated: 24 April 2014 10:48 PM



The R part is easy for me to describe since I added that.
Basically it does the solution over a number of times, each
time ignoring data more than X times the RMS away from the
solution.  The value of X decreases as the solution iteration
proceeds.  This is called a "robust" estimator and is used now
all over aips (to find image RMS and the like).

The L1 part I am less certain of - I think it means that the inner
part minimizes the mean deviation from the solution rather than
the mean square deviation.  Therefore, extreme points only count
linearly rather than quadratically and so have less influence on
the answer.  As Fred points out, this has some statistical issues
(I will bet the mean error is not symmetric about 0), but is likely
to find convergent solutions when standard methods may not.

Eric Greisen




------------------------------------------------------
Staff CP:  https://help.nrao.edu/staff