Transcript ppt

Anticipated problems in our current
approach to identify gamma-ray sources
or: How far automated source
identification might succeed
before we will struggle
Olaf Reimer & Diego F. Torres
(Stanford)
(LLNL)
status quo: presentations from joint
diffuse/catalog workshop earlier this year
Counterpart listings / probability
Counterpart listings / probability
What’s problematic here?
Let’s review some examples:
An “average” EGRET source: 3EG J1249-8330
[95 =0.66 ° , 2 x 10-7 ph cm-2 s-1]
1)
4 XMM-EPIC pointing -> 148 X-ray sources
2)
statistical evaluation of counterparts
3)
does computing a counterpart probability
pc = ppos x p(i)SED x p(i)var x p(i)ext x …
will yield a source identification here ?
No, since for N = 94…148 -> pc will be numerically
undistinguishable in the systematics of its computation
Counterpart listings / probability
What’s problematic here?
more examples:
LS5039
An EGRET source in the galactic plane
3EG J1824-1514 [~4 x 10-7 ph cm-2 s-1]
1)
2)
3)
counterparts in existing catalogs: many already & even
more when proceeding into specialized catalogs
complicated region in terms of diffuse count prediction:
comparable large uncertainty region!
does computing a counterpart probability
pc = ppos x p(i)SED x p(i)var x p(i)ext x …
will yield a source identification ?
No, since p(i)var is not granted beforehand (stochastic)
ppos would exclude the object (it’s the six nearest by
considering two catalogs only)
[NRAO VLA Sky Survey (NVSS) + RASS FSC]
Counterpart listings / probability
What’s problematic here?
Argument: LAT will have better source
locations, so here a better example
from EGRET
A bright EGRET source at high galactic latitudes
3EG J1835+5918 [6 x 10-7 ph cm-2 s-1!]
1)
2)
3)
counterparts in existing catalogs: Ø
dedicated deep HRI pointing: 10 counterparts with almost similar
characteristics -> MWL follow-ups
does computing a counterpart probability
pc = ppos x p(i)SED x p(i)var x p(i)ext x …
will yield a source identification ?
No, since either N = 0: pc = 0 or N > 0: ppos alone dominates pc
Counterpart listings
What’s problematic here?
Argument: LAT will be even better than
this, so here an example from VHE gamma-rays
HESS J1303-631 (13h03m00.4s±4.4s and δ=−63°11’55”±31”)
at least 5 catalog counterparts listed in several counterpart categories
But source is extended!
does computing a counterpart probability
pc = ppos x p(i)SED x p(i)var x p(i)ext x …
will yield a source identification ?
No, since p(i)ext = 0 -> pc = 0 !
(point source catalogs)
After visually appealing cases a more general discussion:
positional coincidence probability ppos
(i)
-> 1 for  observational coverage
(ii)
existing catalogs are by definition incomplete, so a reasonable ppos
will depend on a sizable number of mwl catalogs
in order to keep ppos < 1, but:
(iii)
actual value of ppos will be determined from the inherent treatment of
cuts/completeness/quality of each individual catalog to be considered,
-> thus ppos will have an indeed different meaning for regions of
different source densities -> source locations
Almost every catalog will exhibit different quantities here, which
will make it impossible to define a uniform ppos over the sky
practically (if the gamma-ray observables are the same)
p(A)pos ≠ p(B)pos for a single LAT source if catalogs from different
source populations A, B are involved
ppos LAT source1 ≠ ppos LAT source2 at different spatial locations
Why did it worked for EGRET ?
(“A” blazars / Mattox et al. 97,01 / Soward-Emmered et al. 03-05)
Only the dominant population of high-energy gamma-ray emitters
has been probed (blazar-class AGN)
It does not work at all for galactic sources, i.e. we never
proceeded from statistical evidence for SNRs (Dermer & Sturner,
Esposito et al., Romero et al. to individual IDs !
Why does it works in X-ray astronomy?
ppos <-> small psf in focusing X-ray telescopes
resulting in sheer dominance of ppos (+ ability to handle source extension)
also: the apparatus is well-defined for arcsec-psf point-sources
(Sutherland & Saunders 1992 “On the likelihood ratio for source identification”)
Spectral energy density distribution probability p(i)SED
proportional to the probability that a given source class (i) shows the observed SED
(i) already problematic for the individual blazar
-> 3C279 in flare state
(ii) even more problematic for the blazar population
-> blazar unification scheme is a model
Obviously, there is difference between testing a model
and deriving gamma-ray source identifications
(iii) ambiguity between similar SED templates of different source classes
PSR/INS
-> assumptions build on the baseline template SED will aim to
discriminate
cand. INS between source classes, but will fail in the individual class
already [here only good samples shown, we expect MUCH more
sparsely
mQSO
sampled SEDs]
source variability probability p(i)var
proportional to the probability that a given source class (i) shows the observered
variability
EGRET experience: variability predominantly used to rule out membership in
classes, and only when exhibited at significant level
Obvious: variabilty ID assignment ambiguous if more than 2 classes involved!
(low-lat variable sources, high-lat steady sources in EGRET already pending!)
quests:
quantify AGN in quiescent state ?
non-repeating transients vs. repeated AGN flaring vs. mQSO stochastic var.
similar variability predictions for different source classes
Independently on how pvar may be determined [0…1],
unpredictable quiescent periods for objects believed to exhibit flux variabilty
will bring pc = ppos x p(i)SED x p(i)var x p(i)ext x … -> 0!
This is apparent already for identified EGRET blazars !
-> see fractional variability index for individual AGN in Nolan et al. 2003
The sheer number of possible LAT measurements will help here only marginally
since this problem is coupled to intrinsic timescales of source activity
(where we don’t have not many clues at present!)
source extension probability p(i)ext
proportional to the probability that a given source class (i) shows the
observed extension
This is almost impossible to achieve, since it couples characteristics of an
individual source with population properties
Since individually identified representatives may constitute the population,
the reversal is certainly not generally true!
i.e.: Coma cluster of galaxies -> extension of diffuse emission deduced from X-ray ~ 2...3°
well beyond LAT psf
A2255
-> extension of diffuse emission deduced from X-ray ~ 8..10’
order of LAT psf
i.e.: Cygnus OB2 (TeV J2032+4131) -> extended VHE source
…the total number of OB stars alone is expected to be ~2600 (Knödlseder 2003)
-> individual counterparts per se inappropriate for understanding this scientific
problem, not to mention assigning individual identification probabilities
Potential solution of problem:
An ideal (“lasting”) g-ray source catalog will consist only gamma-ray observables
(Booooooooooh!)
1st order compromise: catalog will include only rock-solid IDs without extensive
counterpart listings
-> in contrast to EGRET catalogs, there must be a procedure set up and described on
what and how is executed for the considered individual source class
i.e. PSR by timing with contemporaneous ephemerides,
down to a statistical significance of x s in i.e. the Rayleigh/H/…-test
2nd order compromise: catalog will list nearby counterparts without quoting any
other IDs than 1st order compromise
IDs to be derived in respective working groups, but distribution/interaction
scheme between science groups tbd.
Individual identification papers and a comprehensive population study should
accompany the publication of the 1st year LAT source catalog
Science groups continuously interact with catalog group
(receive gamma-ray source observables, feeding catalog with IDs)
-> approach for population study: wait for talk by Diego Torres