SN magnitude marginalisation in cosmomc

Use of Healpix, camb, CLASS, cosmomc, compilers, etc.
Post Reply
Anze Slosar
Posts: 183
Joined: September 24 2004
Affiliation: Brookhaven National Laboratory
Contact:

SN magnitude marginalisation in cosmomc

Post by Anze Slosar » January 10 2006

I've noticed that cosmomc does the marginalisation over magnitude offset (equivalent to marginalisation over h) in riess SN code.

I've added SNLS sn data and run with either marginalisation switched on (done separatly for Riess and SNLS) and off and with no other data (no cmb, no lss). Constraints on Omega_\Lambda and Omega_m are unchanged, but without marginalisation, I get a *very* strong constraint on hubble, something like 64\pm1. Has anybody else found something like that? (note that H is still a derived parameter here, so prior on it is not flat).

As far as I know authors of SN data don't really advocate this marginalisation, (though it is a conservative thing to do).

Sarah Bridle
Posts: 144
Joined: September 24 2004
Affiliation: University College London (UCL)
Contact:

SN magnitude marginalisation in cosmomc

Post by Sarah Bridle » January 11 2006

What do you implicitly assume about the intrinsic magnitude of supernovae in the case with marginalization switched off?
i.e. does this assume we know exactly the intrinsic mag (which would surely be unrealistic?)?

Anze Slosar
Posts: 183
Joined: September 24 2004
Affiliation: Brookhaven National Laboratory
Contact:

SN magnitude marginalisation in cosmomc

Post by Anze Slosar » January 11 2006

Good to know! I admit I though that the distance modulus published in the tables is exactly what it says on the can, i.e. 5 log d_l + 25. And I know of at least two papers, where this is done wrongly!

Dragan Huterer
Posts: 24
Joined: July 18 2005
Affiliation: University of Michigan
Contact:

SN magnitude marginalisation in cosmomc

Post by Dragan Huterer » January 12 2006

Hey Anze
If I understand correctly you are asking whether you need to marginalize over the nuisance parameter (h in Riess et al; \script{M} in SCP papers).
You absolutely have to. h here is not H_0/100, but a dummy parameter that combines H_0 and absolute magnitude of SNe, and even if you thought you know the former, you don't really know the latter.

You should find that, if you somehow knew h anyway, the errors on Omega_M, w etc would go down by 50% or order unity or something like that. Also you seem to be saying that you can determine h from SNe data to
a couple of percent which I don't understand - it shouldn't be that good I'd think. But anyway one usually doesn't care about h but just marginalizes over it.

Post Reply