A small fraction of the
data are common with the deep data set (see the papers). There are two
alternate binnings of the power spectrum, which are not independent; in
both the bin width is 200 in l except for the first bin.
* Even binning: 14 bins: l = 0-400, 400-600, 600-800, ..., 2800-3000.
* Odd binning: 14 bins: l = 0-300, 300-500, 500-700, ..., 2700-2900.
Band Power Window Functions
The file deep_windows.tar.gz
(compressed tar file) contains 14 files, one for each band:
* joint_3deep_std2_best_window_* (7 files).
* joint_3deep_alt2_best_window_* (6 files).
* joint_3deep_bigbin_best_window_5 (1 file).
The window functions are tabulated in bins of width 90 in l, i.e., l =
1-90, 91-180, ...; and the tabulated values should be divided by 90 to
get the quantity plotted in Mason et al.(2003), WB(l)/l.
qb file
There are 7 (std2) or 6 (alt2) entries, one for each band. Column 1 is
the band number; column 2 is the band power qB, and column 3 is the uncertainty
on the band power (square root of the diagonal element of the covariance
matrix). Band powers are dimensionless; multiply by Tcmb2 to put them
in temperature units.
invfish file
Ignore the first 6 lines in this file. The first 7 (std2) or 6 (alt2)
columns of the next 7 (std2) or 6 (alt2) lines give the 7 by 7 or 6 by
6 band-power covariance matrix. This is the inverse of the curvature (Hessian)
matrix as given in equation 75 of Myers et al., and it can be used directly
to give a gaussian approximation to the likelihood function.
otherps file
It is better to use an offset lognormal approximation to the likelihood
function (Sievers et al., equation 2), for which you need the offsets qNtB.
The offsets are given in the otherps file, which contains 7 (or 6) entries,
one for each bin. There are 4 columns giving the component band-powers for
noise and discrete sources. You should add all four components to get the
required band-power offset.
