29 September 2004
[Notes by S. Digel. These are not comprehensive and not meant to be a literal record of what was said. I tried to note the questions that were asked, to complement the online versions of the presentations. Questions and responses are italicized. You'll notice that the notes get less detailed later in the session, and that they have little about my own presentations.]
~45 attendees in person + 1 via VRVS
A general overview of the issues
Catalog generation will be via pipeline, but not the Level 1 pipeline [batch mode or ongoing?]
Catalog generation is iterative with bright sources found first then incorporated into the model to search for fainter sources – all possibly before likelihood analysis is run
The plan is to use likelihood analysis to refine the parameters, not for a source search per se
Schedule for catalog pipeline:
Starting with identifying candidate source search algorithms (done), with prototypes developed by end of 2004, and selection of algorithm for source detection before DC2; this algorithm will then be used for source catalog generation in DC2
Need to have a processing database defined by 2005
Integrate pipeline elements (like flux history and identification) in 2006
All to be ready by end of 2006
For an all-sky source search – evaluate algorithms with respect to several criteria
Most important is ‘detection power’ – if a source is not found at this point, then it will be lost – a related criterion is ‘resolving power’
Lower-level considerations are flux and position estimates, then false detection rate, then overall computation time
Coordinate system – Propose using CAR around the plane, ARC around the poles [may be dependent on the algorithm] transition at 32.7 deg latitude (equivalent distortions)
Limits to detectability – general considerations:
Background - Poisson fluctuations, soft sources are harder to detect, and best detected at low energy. PSF is poor at low energies, however
Splitting into several energy bands may be advised – look for soft sources at lower energies, ~4 energy bands should be alright.30 MeV – 100 MeV – 316 MeV – 1 GeV – 10 GeV [what about higher energies?]
Pixel size should be adjusted per band
Merging source lists by adding likelihood images should be better than merging source lists directly for different bands
Threshold level is tradeoff between detection power and false detection rate
Variable and extended source issues and source localization were also discussed
Scargle – you argued for having a low threshold in order to not miss sources; isn’t it conceivable that we’ll have slower, better algorithms like likelihood that we will get around to applying eventually? Maybe in later catalogs, or addenda
Ballet - Yes
Kanbach – Allowed to use commercial software?
Ballet - Yes
Kamae – We may not have accurate PSF at the time of the catalog analysis – may get it eventually from beam test or may not have it at all (say if we have only one spare tower after assembly, then we can’t measure PSF beyond 20-30 deg) and in any case, meas. with SLAC beam is hard at low energies
Ballet - Low energy may not be so important for detecting sources
Kamae – But important for getting spectra right at low energies
Harding – Is there a concern about losing sensitivity by dividing into bins of energy? How about integral energy ranges, say, [or overlapping ranges]?
Thompson – Breaking into energy bands has the advantage of overall better PSF – the angular resolution is dominated by the low energy end of the band
Grenier – In our studies we sometimes had funny things that appear only in one of the bands – and so look like fluctuations
Harding – So maybe integral ranges would be appropriate
Ong – You gain information by adding ranges
Summarizing mr_filter results on DC1 data
Kanbach – how many sources ‘should’ have been found?
Ballet – It is not known – 880 sources were in the simulation but no evaluation with likelihood has been made yet to determine how many were in principle detectable
Wavelet method (or mr_filter) strong points – already existing, fast, can also detect extended sources; weak points – finds many spurious sources
Now working on 3-dim (x,y,E) filtering with wavelets. Not clear that this will be available soon enough for GLAST. [Who is developing this? I missed this.]
Another method also being investigated at Saclay, source detection using optimal filter, results in fewer spurious sources. The algorithm is reasonably fast, although perhaps not as sensitive
For the results reported, a spurious source is defined as one with no counterpart in the input catalog within 1 deg angular offset. Should probably use a flux dependent/energy dependent method
Motivation was quick and dirty method for finding candidates for likelihood search – and found that it worked relatively well
Basic method is to determine significance of excesses on a spherical grid – actually not binned – summing in annuli – with outer annuli defining the background counts. (Maybe eventually use larger annuli at high |b|, where the gradients of diffuse intensity are relatively small.)
The method finds a lot of spurious sources with significance >=3 cutoff (1677 with 363 matches)
Reduces to 14 spurious sources (out of 126) if require significance >=5
Have also used a years; worth of simulated data included 3EG sources (from Jim Chiang) Also reran the algorithm with doubled galactic diffuse intensity and tripled isotropic diffuse. The faint sources also seemed to triple in flux – so something suspicious going on [With what?]
Plans to investigate varying the aperture sizes; also so far have been just binning over all energies
Source characterization rather than detection
Uses a new parameterization of PSF
Multichromatic continuous wavelet transform – “multichromatic” means that scale of wavelet adjusted for energy
Provides an estimate of significance of detection
Reimer – in log N-log S you will always run into saturation – roll off at instrument sensitivity limit
Moiseev [?] – The plot is differential, not a cumulative log N-log S
Exposure systematics – using all data vs. exposure restricted to within 30 deg of axis (and not use the narrow FOV observations)
Fit a model – to look at correlations with H I, CO, E(B-V), IC. Include 3EG point sources and isotropic emission in the background
[E(B-V) comes from Schlegel et al. color-corrected IRAS intensity maps]
Studied 80 deg > |b| > 5 deg
Digel - Sources were modeled?
Grenier - Yes – fixed at average flux of 3EG catalog with one scaling parameter total for all
IC emission from FIT dust emission – cold gas assocaiated with cold dust?
Excesses more or less consistent with the ‘faint, persistent’ sources [which are already in the model] in the ‘halo’ around the Galactic center
Question from someone whose name I did not note - Albedo contamination in the maps? Looks ~like zenith angle cuts should have been more conservative at large angle from the instrument axis [I don’t really understand this and may have written it down wrong. In general, I don’t think that albedo contamination is a problem, or that it can be determined by eye by looking at maps in a celestial coordinate system.]
As a parallel effort, we have been implementing HEALPix maps – conversion from a flat map can have problems with flux conservation if you aren’t careful
Have been looking at (or will be looking at) expressing the PSF in spherical harmonics for convolutions [?]
Moskalenko – In GALPROP the IC intensity is currently symmetric left/right, positive/negative in latitude
Grenier – The asymmetry in our figure is due to variations of exposure; the maps are intensity times exposure
Reimer – Cuts on off-axis angle make exposure problems less off axis
Harding – A mm dust component of the Crab nebula was recently detected, and has now included in IC models of the Crab
Basic issue: A large fraction of the sources is near the detection limit – careful statistical treatment is needed
The wavelet method does not need any a priori info/hyoptheses about the data
Other source detection methods – sliding cell, likelihood analysis, wavelet
A reason for using mexican hat wavelet: shape approximately consistent with effective PSF
Double gaussian fits [of what? I didn’t follow this]
Spurious sources are (largely) eliminated in the first iteration of the detection algorithm
Example of light_sim comparison with DC1 anticenter simulation – fluxes look right
The method must be tuned to instrument – e.g., for analyzing EGRET data vs. LAT
Caraveo – Are you saying that you are seeing new (real but not previously cataloged) sources in the EGRET data?
Someone – These may have been below threshold for EGRET detection
Aymeric Sauvageon imported v0 today into the SLAC CVS repository
Discussion on topic of planning the contents of the source catalog
Grenier - IAU says we can use Galactic coordinates in names
Various - Include spectral indices in the catalog no matter how poorly determined they are in the case of faint sources
Petry – Can’t we derive the diffuse emission from the LAT observations? Filtering, iterative analysis,…
[I don't think that this was answered during the session. I think that the problem is at low flux levels - distinguishing sources from structured diffuse emission and distinguishing faint sources from isotropic emission. Still there may be merit in having an 'empirical' model distinct from a physical model that we use outside of the SAE to actually learn things from the diffuse emission.]
Galactic ISM and CR distributions
Dynamic balance
CR scale height is unknown but assumed to be large relative to gas (so CRs uniform over the scale height of the interstellar gas)
Present work: Extended GALDIF to |b| = 30 deg
The model indicates underprediction of the observed EGRET intensity at medium latitudes ~20 deg
NGP/SGP ‘holes’ in EGRET intensity maps are maybe due to the effect tiny pixels, most of which are zero, at the poles of the maps in Galactic coordinates
Moskalenko – Proton & electron spectra are assumed to be the same everywhere in the Galaxy, including the halo?
Hunter – Yes, with intensity that varies according to the smooth scaling factor.
Digel – Have you re-evaluated the adjustable parameters in the updated model?
Hunter - So far left the coupling scale and X-ratio are as in the 1997 paper.
(In response to a question that I did not record) Hunter – For the Galactic center and anticenter, the model is interpolation from neighboring longitudes
14C (unstable) and heavy (Z>30) CRs are by definition local
CR spatial variations:
Below about 20 GeV/nucleon don’t know the local spectrum of CRs, owing to the heliosphere
Sources are more frequent in the spiral arms –
Evidence from 10Be in south polar ice is for 4 nearby SNR over the last 150k yr
Outputs of GALPROP – emissivities on x,y,R,z,E grid and/or sky maps with a given resolution (gridding) l,b,E, process
AMS, Pamela (for dark matter searches), ACE, TIGER, HEAT, GLAST are GALPROP ‘consumers’
Running Galprop – Using a coarse grid for a 2-dim or 3-dim calculation for the initial, test phases of a calculation can save a lot of CPU time. Later, a 3-dim model can be re-run with as fine a grid as RAM allows.
Slide 19 has near future developments planned for GALPROP – including extension down to ~511 keV