Bug in (old) default COSMOMC plugin for CBI

Post Reply
Jonathan Sievers
Posts: 5
Joined: May 18 2006
Affiliation: CITA

Bug in (old) default COSMOMC plugin for CBI

Post by Jonathan Sievers » August 29 2006

Greetings all - it turns out that there is a bug in the old default COSMOMC plugin for CBI. We've posted a new plugin that can be picked up from
http://www.astro.caltech.edu/~tjp/CBI/d ... newdat.tgz
This is also fixed in the latest COSMOMC (August '06).

The problem was that in the old .dataset format, COSMOMC expects the Fisher matrix to be pre-transformed (and in a separate file) if using the offset-lognormal likelihood approximation. The matrix in the plugin used an old set of X_B's for the transform, inconsistent with the set in the .dataset file used to transform the data. The net effect was to artificially shrink CBI error bars by about 20%, expect for bin #14, where the shrinking was close to a factor of 3. We emphasise that the power spectrum, errorbars, X_B's, and correlation matrix in the papers and on the CBI webpage are correct, so if you did the transform yourself, you'd be fine. The WMAP team did do their own transform, and so the parameters in their most recent set of papers aren't affected. We're looking into the parameters in Readhead et al. (based on WMAP1) to see if those were correct.

Thanks to Loison Hoi for finding the bug. He's done some parameter runs that show the impact of using the wrong transform. See http://www.physics.mcgill.ca/~hoiloison/cbi/index.html for parameters and 1-D likelihood curves.

Jonathan Sievers, for the CBI team

Post Reply