[daip] CALIB

[Lynn D. Matthews]

I have encountered some unfortunate behavior in CALIB that is
linked with the CMETHOD parameter. Basically the 'GRID' method seems to
fail under a certain combination of circumstances---but in a way that the
cause is not obvious to the user.

The situation: I am attempting to self-cal (phase only) using a CC
model. The model is dominated by a bright maser spot offset from the
phase center. The phases are already reasonably good. There are ~48,000
visibilities and the clean components are attached to an 8192x8192 image.

If I run CALIB with CMETHOD ' ', the program enters two different realms
depending on the number of CCs (see also the message output appended
below).

1. In the few-CC case (~100), CALIB chooses 'DFT' and converges with
~47000 good solutions

2. In the case of ~1000 CCs, CALIB chooses 'GRID' and fails completely
(9 good solutions). It also complains about AP size even though I have an
81MB AP.

But: if I shift the main maser spot to the phase center in both the u-v
data and the model image, CALIB still chooses 'GRID', but now is able to
find many useful solutions. If I shift the spot half-way to the phase
center, CALIB finds some, but fewer, solutions.

Finally, if I use a model image that is only 4096x4096 with no position
shift, 'GRID' is also able to converge.

Lynn
=============================================================================
In the few CC case (~100), CALIB reports:

CALIB1 18:26:46 Task CALIB  (release of 31DEC06) begins
CALIB1 18:26:46 CALIB USING IF1 CH328TST . TST    .   2 DISK=  2 USID=2086
CALIB1 18:26:46 Create TST      .CALIB .  13 (UV)  on disk  2  cno  251
CALIB1 18:26:46 Selecting and calibrating the data
CALIB1 18:26:48 Doing self-cal mode with CC model
CALIB1 18:26:48 FACSET: 88.994919 Jy found from 115 components
CALIB1 18:26:48 VISDFT: Begin DFT component division
CALIB1 18:26:48 VISDFT: Model components of type Point
CALIB1 18:26:48 I Polarization model processed
CALIB1 18:26:48 Model computation is    10 percent complete
CALIB1 18:26:48 Model computation is    20 percent complete
CALIB1 18:26:48 Model computation is    30 percent complete
CALIB1 18:26:49 Model computation is    40 percent complete
CALIB1 18:26:49 Model computation is    50 percent complete
CALIB1 18:26:49 Model computation is    60 percent complete
CALIB1 18:26:49 Model computation is    70 percent complete
CALIB1 18:26:50 Model computation is    80 percent complete
CALIB1 18:26:50 Model computation is    90 percent complete
CALIB1 18:26:50 Field    1 used     115 CCs
CALIB1 18:26:50 Determining solutions

---> Result: Many useful solutions

In the many (~1059) CC case:

CALIB1 20:31:39 Task CALIB  (release of 31DEC06) begins
CALIB1 18:26:46 CALIB USING IF1 CH328TST . TST    .   2 DISK=  2 USID=2086
CALIB1 20:31:39 Create TST         .CALIB .   3 (UV)  on disk  2  cno  215
CALIB1 20:31:39 Selecting the data
CALIB1 20:31:39 Doing self-cal mode with CC model
CALIB1 20:31:39 FACSET: 135.428528 Jy found from 1059 components
CALIB1 20:31:40 Divide data by model - first compute model by summing
CALIB1 20:31:49 UVPREP: Maximum U Baseline is  1.207E+09 lambda.
CALIB1 20:31:49 ALGSTB: Only  316 Rows In AP (Need  2409)
CALIB1 20:31:49 ALGSTB: Ipol gridded model subtraction, chans    1 through
1
CALIB1 20:31:49 ALGSTB: Pass   1; 2398-2093 Cells, with       90 Pts
CALIB1 20:31:49 ALGSTB: Pass   2; 2091-1786 Cells, with     3189 Pts
CALIB1 20:31:49 ALGSTB: Pass   3; 1785-1480 Cells, with     4315 Pts
CALIB1 20:31:50 ALGSTB: Pass   4; 1479-1174 Cells, with    18983 Pts
CALIB1 20:31:50 ALGSTB: Pass   5; 1173- 868 Cells, with    32942 Pts
CALIB1 20:31:50 ALGSTB: Pass   6;  867- 562 Cells, with    28433 Pts
CALIB1 20:31:51 ALGSTB: Pass   7;  561- 256 Cells, with    63787 Pts
CALIB1 20:31:51 ALGSTB: Pass   8;  255-   0 Cells, with   139750 Pts
CALIB1 20:31:52 Field    1 used    1059 CCs
CALIB1 20:31:52 Determining solutions

---> Result: 9 useful solutions for 48675 vis.