Optimizing a photometric redshift only run

[MarcoCast79]

Dear Jacopo and all, I need to try a run on many objects with the sole purpose of estimating the photometric redshift from multi-band photometry, is there a "standard" BEAGLE recipe to optimize such a run in terms of computing time?

Surely I will put:

REBIN TEMPLATES = 10

But: is it worth/feasible to leave out all other parameters, such as mass, metallicity etc and keep as "fitted"-type the redshift only? Do you have other suggestions?

From the paper I understand you used the same run for photoz and physical properties, but in my case I'm not (yet) interested in other source properties.

Thanks!

[jacopo-chevallard]

Ciao Marco, besides setting REBIN TEMPLATES = 10 you can:

• use pure stellar templates (e.g. BC03) instead of stellar+nebular models

• use a simplified model, adopting a constant SFH, varying only the parameters mass, metallicity and max_stellar_age (besides redshift), and perhaps tauV_eff (you can use the CF00 attenuation curve, fixing mu to the "standard" value of 0.4). You can also test on a small sample (e.g. 50 objects) the effect of fixing the metallicity and removing tauV_eff, but I don't think that this would affect much the speed

• in the MCMC*.param file:

  • decrease the number of "live points" to 100 or even 50 NUMBER OF WALKERS = 50
  • increase the ENLARGMENT FACTOR = 0.9
  • increase the EVIDENCE TOLERANCE = 1.00
  • (you may also want to increase the MIN RELATIVE PROBABILITY = 1.E-03 to decrease the size of the output files)

[MarcoCast79]

Thanks, I'll do as you suggest!

Just one last doubt, does SHRINK TEMPLATES WL RANGE affect speed to some extent? I think I can shrink it a bit

[jacopo-chevallard]

Yes it does, especially if you remove points in the UV/optical (where models have higher resolution), but in general when you're fitting photo-z you want to keep the entire UV/optical... but you can certainly remove points redwards of the reddest observed band!