[daip] possible bug in 31DEC03 version of CPASS

[Eric Greisen]

David A. Boboltz writes:

 > I am having a problem with CPASS in the 31DEC03 version of
 > AIPS.  I'm using it on some continuum VLBA AC data.  CPASS 
 > gives the following error: 
 > 
 > CPASS2:   35  189  2.329746500979875E+00   3.98260972E+03
 > CPASS2:   36  197  2.323943268535836E+00   4.00271365E+03
 > CPASS2:   37  205  2.319335611944680E+00   3.84579811E+03
 > CPASS2: BASOLV: SOLVER RETURNS ERROR    2
 > CPASS2: Appears to have ended successfully
 > CPASS2: sirona       31DEC03 TST: Cpu=     525.6  Real=    1142
 > 
 > and no BP table is created.  I tried adjusting various values
 > of CPARM, but this didn't help.  If I switch to the 31DEC02
 > version of AIPS, CPASS works fine.  Also BPASS in 31DEC03 has
 > no problems.

The 31DEC02 and 31DEC03 versions of CPASS are identical.  I suspect
that you have a 31DEC02 from some time prior to its actual freeze
date. CPASS was changed in August 2002 to actually pay attention to
the error codes returned from the fitting routines.  Thus it generates
a message when the number of iterations is exceeded and the solution
fails.  Previously it wrote the failed solution to the BP table and
pretended that all was well.

CPASS is a very problematical program - I attempted to add options to
avoid all the normalizing because such normalizing adds systematic
error to cross-correlation data anyway.  When I changed anything, the
task would not converge under any circumstance.  That is when I
discovered that it ignores the errors that it itself generates.

I recommend that CPASS only be used to allow for a time-varying
bandpass correction applied after the gross bandpass chapes are
removed with a time-averaged BPASS.  Far too many terms are required
for CPASS to be reliable in fitting a full bandpass shape.

Eric Greisen