Peter Capak Associate Research Scientist IPAC/Caltech.
-
Upload
georgiana-strickland -
Category
Documents
-
view
219 -
download
2
Transcript of Peter Capak Associate Research Scientist IPAC/Caltech.
Peter CapakAssociate Research Scientist
IPAC/Caltech
What is COSMOSWhat we learned from COSMOSPhoto-z (and spec-z) theoryWhat work needs to be done
2 sq degree deep multi-wavelength survey
Designed for Photo-z
HST imaging Ideal for lensing
Lots of spectroscopy to I~26
Contours : lensing DM
Red : x-ray
Blue : galaxy mass density
COSMOS LensingMassey et al. 2007, Rhodes et al. 2007, Leauthaud et al. 2007
•Need color calibration better than 0.01 mag
•Need template calibration better than 1%
•This is the main reason photo-z are considered low-accuracy!
•Shapes can be measured for much fainter galaxies than photo-z are normally used
•Need to carefully design space/ground depths
•have an extrapolation problem
12h in K band
0.62h Hubble F814W
•Lose 10-20% of area
•Need careful treatment of bright star artifacts
•Photo-z does not give you p(z)
•You get p(z|D,Model)
•Need prior to break degeneracy
•Prior should be lensing specific!(Massey et al. 2007)
Spec-z measured by identifying spectral features
Accuracy is dz/(1+z)=/=1/R
Phot-z should be accurate to ~0.2
Clearly more accurate
So where is the information?
Information is in the color change as an object spectra is red-shifted
Redshift
•Photo-z error determined by:
•Gradient of color change with redshift•photometric accuracy
Redshfit
•Generalized to any filter set:
• Ca,b are the color
•Can use a range of template SEDs
•Provides estimate of the phot-z accuracy
•Applies to all methods, not just template fitting
Real Galaxy
Redshfit
•Works for real galaxies
•Two galaxy types in COSMOS using broad band data
•Estimated (red line) vs actual
•Degeneracy due to mapping from ColorRedshift
•Need to live with this or get more data
Worse results at fainter fluxes
Need calibration accuracy of 0.01 mag or better!
More filters not necessarily better
Best accuracy at filter center
Gaps in coverage very bad
Narrow filters improve accuracy
Overlapping filters improve accuracy
Better absolute calibrationBetter measurement of system
throughputBetter tracking of atmospheric
absorption in IRBetter photometry techniquesBetter tracking of errors
Develop flagging and accurate error estimates
Account for template uncertainty Develop better templates Account for variability Empirical codes need to work with
non-representative samples could generate templates and priors e.g. Budivari et al. 2000, Benitez et al.
2000
What priors should be usedWhat redshift ranges are most
important Can we live with not using some data?
Integrate complex probability distributions into lensing code
Photo-z should be as robust and trustworthy as spec-z’s Main fault is in data quality and lack of
theoretical understanding Recent improvements have come
from improved data quality Now need to focus on improved
techniques Lensing should integrate inherent
uncertainties in photo-z into the quantities measured