Full Force Test on Small Area Survey Stacks and warps

This note discusses the results of the first test of the IPP full force stage on the SAS stacks (label = SAS.20130620). The label for the fullForceRun was SAS.20140117.

The purpose of this test were

• verify the basic operation of the software
• test the code on a well studied area of the Pan-Starrs survey
• compare the resulting galaxy model fits to SDSS measurements of galaxy parameters
• to determine the recipe parameters to be used in 3PI.PV2

Recipes

The programs and the algorithms used by the full force pipeline are described in fullforce_algorithms.

This test searched a wider range in fit radii parameters than is necessary or practical from a compute perspective. This was done so as to get an idea of the range that should be used for the 3PI analysis.

Recipe Parameter	This Test	Default Value
GALAXY_SHAPES_FR_MAJOR_MIN	0.5	0.85
GALAXY_SHAPES_FR_MAJOR_MAX	2.0	1.15
GALAXY_SHAPES_FR_MAJOR_DEL	0.05	0.05
GALAXY_SHAPES_FR_MINOR_MIN	0.5	0.85
GALAXY_SHAPES_FR_MINOR_MAX	2.00	1.15
GALAXY_SHAPES_FR_MINOR_DEL	0.05	0.85

With these parameters 961 trial galaxy fits were performed for each object selected for fitting while the nominal range is only 49 trials. The average processing time was 960 seconds for each warp. This is prohibitively expensive for the 3PI survey. In fact we stopped the test after processing only 20% of the skycells because it was taking so long. We believe that this limited test provides the good enough statistics for analysis.

Data Set

The latest SAS stacks (label SAS.20130620) were processed through the staticsky and skycal stages of the IPP using the candidate PV2 version of psphot (tag ipp-20140114). The label for the runs are SAS.20140117. These stacks were built with the warps labeled SAS.20130620. These warps are the target of the forced photometry measurements.

The psphotStack code is essentially the same as that used for SAS_32 except that the selection of objects for extended model fitting was relaxed. In particular the previous requirement that the object be "extended" (EXTNSIGMA > 3) was not applied. Instead magnitude cuts were used. The cuts chosen were 21.5 in g, r, and i. 20.5 in z and 19.5 in y band. Note that these cuts are per object. Fits were performed in all bands if the object's magnitude was less than the limit for any band. Also extended model fits will not be performed in the region within 20 degrees latitude of the galactic plane with the limit increasing to 35 degrees near the galactic center. This cut has no effect in the SAS region.

During the skycal processing a new data product is produced. The "cff file". Here are contained the measurements for each source that are to be used for forced photometry. Each source with an with a extended source fit is marked for galaxy fitting and the parameters from the psphotStack model are given to be used as the starting point for the full force search. The model type chosen is the best of the three models Exponential (exp), de Vaucouleurs (dev), or sersic (ser).

The inputs to each "full force run" are the cff file (one per skycell per filter) and the individual warp images for each skycell. In this data set there were on average 11 warps per skycell. Note that all warps are used regardless of whether they were rejected when building the stacks.

The full force processing consists of two parts. The program psphotFullForce is run on the individual warps and then psphotFullForceSummary takes the results for each skycell and generatse the final extended source measurement for each source. See fullforce_algorithms for details on the measurements and calculations.

Results

Here we present results for 85 of the 620 r band SAS skycells. We took the full force summary cmf file and matched the sources to a catalog downloaded from SDSS DR10. The region extracted is the 8 x 8 degree region centered at ra = 334 dec = 0.

In all there were 208597 sources matched. Of these 100513 are identified by SDSS as galaxies (type_r = 3). psphotStack obtained a successful extend model fit for 59923 sources of which 55241 were SDSS galaxies.

In the following we will focus on these "galaxies". In this sample 47318 galaxies with SDSS r < 21.5. We have extended model measurements for 41437 of these.

Exponential Galaxies

32134 of the SDSS galaxies in the sample have fracDev_r = 0, indicating that the SDSS composite model was equal to the exponential model. Here we refer these as expGalaxies. Of these 18125 were fitted with an extended source model by psphotStack. The best psphotStack model for 5438 objects was the exponential model, 11792 sersic, and 895 dev.

Exponential Model for SDSS Exponential Galaxies

Here are plots of the difference between SDSS model fit major axis and magnitude and the psphot full force summary results for the expGalaxies.

And a scatter plot of the two differences.

And the magnitude difference versus SDSS magnitude.

psphotFullForce loops over the range in test radii and measures the chisq for each fit. psphotFullForceSummary combines these results to come up with the final results presented here.

The following plot shows a histogram of ratio of the full force summary major axis to the nominal starting value provided by psphotStack.

31% of the "best fits" are outside the nominal parameter range of 0.85 - 1.15 times the stack value. So in our test 80% of the processing time went in to finding the best fit for these 31% of the objects. If we expanded the range from .7 to 1.3 88% of the best fit radii would be found. With steps of .05 this would consist of 169 trials.

XXXX: look at minor axis

Now, the point of performing galaxy shape measurements on the warps is the hope that the results will be better than the stack measurements. The following two plots compare these full force results with the psphotStack fits to exponential galaxies.

We find that for both the major axis and magnitude the histograms of the difference from SDSS are narrower and peaked closer to zero.

Sersic Model for SDSS Exponential Galaxies

Interestingly the exponential model was the best model for only 30% of the expGalaxies. 65% of the time the sersic model was the best.

SDSS did not measure a sersic model so we compare the SDSS model magnitude r with our Sersic magnitude. Here is a scatter plot comparing the two magnitudes and a histogram of the difference.

As expected, the measured Sersic index for these galaxies is concentrated at low values. Interestingly the peak is less than 1 (where the Sersic model function is equivalent to an exponential model).

de Vaucouleurs Galaxies

20030 of the 100513 objects matched to SDSS galaxies have devFrac_r = 1 indicating that the SDSS composite model was equal to their deV model. 9196 of these had psphotStack extended model fits. The dev model was the best fit for 2046, 4661 sersic, and surprisingly for 2489 objects the exp model had the lowest chisq.

devGalaxies dev model

Here are plots of the difference between SDSS model fit major axis and magnitude and the psphot full force summary results for the devGalaxies.

And a scatter plot of the two differences.

And the magnitude difference versus SDSS magnitude.

As was the case with the exponential galaxies the fullforce results are somewhat better than the stack photometry

devGalaxies sersic model

As was the case with the expGalaxies, the sersic model tended to have the best fit. Here is a comparison of the sersic magnitude to the SDSS model magnitude for devGalaxies.

The Sersic index is peaked around 4 for these objects, but not as sharply as one might expect.

Sersic Model

Finally we give plots of the Sersic model results for all galaxies for which it was chosen as the best psphotStack model.