Faster WMAP 3-Year Low-ell TT Module

Jonathan Sievers and Mike Nolta

Canadian Institute for Theoretical Astrophysics

We have sped up the low-ell TT likelihood through an eigenvalue compression by a factor of more than 10.  As this is currently (as of Version v2p2) the slowest part of the code, we achieve an overall speedup in the likelihood evaluation of a factor of 4.  The uncompressed and compressed likelihoods agree typically to better than 10^-6, much less than the scatter in cross-platform likelihood evaluations (order 10^-3).

Installation

If you haven't done so already, install the standard version of the WMAP3 likelihood code, available from LAMBDA .
Then, download the faster TT tarball: wmap_likelihoodcode_v2fasterTT.tgz (98MB)
Unpack it somewhere, and copy the contents into your current likelihood directory.  Warning - there is a Makefile in the tarball, so you might want to not overwrite your current Makefile.  The only makefile modification required to use the faster TT is to add "-DFASTERTT" to WMAPFLAGS.  If "-DFASTERTT" is missing, then behavior reverts to the standard distribution.

Note:  The compression is valid only for a fixed noise, and has only been calculated for ell's up to 30 (lowl_max=30 and ifore_marg=2 in the Makefile).  If you have a burning desire to run a compressed version with different foreground treatments, contact me.

Files in the tarball

WMAP_3yr_tt_pixlike.F90 - the fast version of the low-ell pixel-pixel TT likelihood code.  This will overwrite the default version, so be warned if you have made modifications you wish to keep.
Makefile - sample makefile.  If you already have a Makefile set up, add " -DFASTERTT" to WMAPFLAGS.
read_fits.f90, read_fits.fh - updated FITS reading utilities required to read the compressed matrices.
compressed_t_map_f2.txt - the compressed data vector (text file).
compressed_tt_matrices_f2.fits - the compressed TT signal matrices (FITS file).



Bored? Read more about what's going on inside in PDF or PS .

Questions, comments? Contact Jonathan Sievers or Mike Nolta