- J. Fowler:
reported that he has been working on BANDMERGE.
- B. Light:
reported that he has run aperture
photometry code using FREXAS-type input on the ultrasparc loaner,
clone4. A 42-frame test case that took 53 CPU seconds on karloff and 50
CPU seconds on lugosi took only 29 CPU seconds on clone4, and a large
part of that is believed to be due to other CPU-intensive tasks running
at the same time (clone4 has a single CPU). When compiled on clone4 and
run when clone4 was drier, the run took 16 CPU seconds; this version
was compiled with standard high optimization but not
ultrasparc-peculiar optimization.
He also reported on a study of the dependence of the differences
between PSF-fit photometry and aperture photometry on whether the
aperture photometry is done on coadded images versus KAMPhot's internal
method. The behavior of the mean difference as a function of magnitude
appears the same for the two cases; the dispersion is quite different,
but in the cases examined, the dispersions were computed differently
for the two methods and have different interpretations. Bob will look
into providing more comparable dispersions. In any case, going from
the protopipeline method to the 2MAPPS method does not appear to
introduce anomalies. During this discussion it became apparent that
whereas KAMPhot has code to break a single detection into multiple
extractions if doing so improves the fit, this code has never been
activated; this had not been the understanding of some members of the
working group. Apparently all multiple detections with "blend" flags
that have come out of KAMPhot resulted from multiple close detections
going in rather than splitting up of single detections.
- G. Kopan:
reported that he has been working on PIXPHOT and PICMAN. The latter is
now capable of using a Weinberg kernel or a KAMPhot PSF for smoothing, and
studies of the effects of different kernels will be discussed by T. Jarrett
(see below and also last week's minutes). In fact, arbitrary smoothing kernels
can be read in via FITS files.
He also has been running tests on clone4. With standard optimization,
clone4 appears to be faster than karloff and lugosi by a factor equal to the
ratio of clock speeds. With ultrasparc-peculiar optimization, additional speed
advantages are seen on clone4. The user CPU variation observed when identical
tasks are run repeatedly on karloff and lugosi (see last week's minutes) were
not seen on clone4.
- T. Evans:
reported that she has been working mostly on IRTS but is now returning
to 2MASS to continue working on the MAPCOR SDS and SIS's. A design change in
the persistence removal has occurred and will be described in the next version
of the SDS.
- L. Fullmer:
reported that she has been working on the QUALITY SDS. All subsystem
developers are requested to consider what kinds of boiled down information
their subsystems can transmit to QUALITY to enter into an overall evaluation of
scan quality. Such information will be transmitted via ASCII files. An example
that was suggested is the mean trimmed average standard deviation from DFLAT.
- R. Cutri:
reported that he has been working on the point source WSDB format (see
last week's minutes). Several responses to his emailed format proposal have
been received from team members, but much discussion clearly remains.
- D. Kirkpatrick:
reported that he has completed some spectroscopic observations
on the 200-inch, where the targets were selected from 2MASS protocamera data
products. The wavelength range covered was 3500-9500 Angstroms. The targets
included some late M dwarfs, some brown dwarf candidates, and some color
selected objects picked out by C. Lonsdale and R. Cutri that appear to be both
nonstellar and very interesting.
- H. McCallon:
reported that he has been working on the demo of his proposed
position reconstruction technique for the PS CDR (since returning from IRTS).
At this point, the code uses a single band and no Read1 detections but does
include in-scan/cross-scan coupling. Expansion to multi-band and Read1 input
will not affect the basic formulation and should only strengthen the argument
that the method is fundamentally sound. The code solves for offsets and scale
factors for in-scan and cross-scan coordinates and a rotational error. It has
been tested with a simulation based on protocamera data. It has worked well for
the frame-to-frame offsets and absolute reference solutions.
- T. Chester:
reported that he attended the WIRE review and has been working with
T. Jarrett on the study of the effect of various smoothing kernels on galaxy
processing.
- T. Jarrett:
reported that a narrower smoothing kernel (H = 0.4, where H is a
parameter in the Weinberg smoothing formulation) works better than the broader
one used previously (H = 0.6 was used for the study reported in last week's
minutes). Fewer false detections were found, while faint galaxies were still
detectable. For galaxy processing (as studied so far using Coma scans), this
smoothing kernel is preferable to either the bilinear or area overlap kernels,
as it provides better star-galaxy discrimination. It is worth investigating
whether it has an impact on point source processing.