Faint Masking Results
The masking routine is likely complete now. Below are the likelihood spaces. These seem to estimate a reasonable amount of error in all three parameters. The derived width is near the 10 km/s width found by Lee et al. (2012) (see figure 3), who found the width which produced the best-correlated with the dust column density is around 10 km/s. Lee et al. (2012) adopted a 20 km/s width, corresponding to the FWHM of the correlation, leading to higher . In our analysis, we find an intercept and a smaller width better fit the dust column density. The DGRs in our analysis and Lee et al. (2012) are consistent with each other.
I am using the IRIS image from Lee et al. (2012), and binning the data to 30 pixel sizes for the entire analysis to avoid correlated errors from the CIB.
During the masking process as described in the Planck paper, we would expect the DGR to level off to a consistent value when increasing the number of pixels in the mask. Below is a plot of the progression of the DGR and the intercept values as a function of the number of pixels included to calculate the two parameters.
Figure 1. - The left column represents the initial masking performed with a width of km/s to create the initial map used in masking, the right, an HI width of km/s.
Perhaps a more robust way to derive the mask would be to require that the DGR is nearly the same over several iterations, not just one subsequent iteration. However this process is working, and perhaps we should just leave it alone.