Dec 8 2017, Clem Pryke
Flavien Vansyngel has kindly provided a set of files for a new model based on arxiv/1611.02577. This model includes dust and synchrotron components and becomes CMB-S4 model 09.
Here are plots comparing the full sky to Planck and PySM model a2d4f1s3 (CMB-S4 model 02) in a similar style to previous posting Checking PySM maps. (These plots are in Healpix Q/U convention and can be compared to Planck 2015 paper I fig 8. Planck has been downgraded all the way to nside=16 to suppress noise while the other plots are nearest neighbor sampled from nside=512.) All the plots are on the same +/-15uK color stretch. We see that PySM follows Planck much more closely on the largest scales.
Of course what we are interested in for CMB-S4 is the behavior in clean regions off the galactic plane. Here are zoom plots around the BICEP/Keck region. (These are in celestial coords with IAU Q/U convention. Planck is nside=128 this time.) The color stretch is shown on the y-axis label and is chosen from the max absolute value of the Q map and then held the same for U. We see very definite "hot pixel" behavior for the Vansyngel model where a few pixels drive the color range very wide - we can actually see hot pixels in the full sky plots above as well. The PySM models are clearly guided by Planck at the larger scales visible in this plot.
Next we take the power spectra on the BK14 mask (which is available here). The plots below show the dust component only. The dashed lines in the EE and BB panels are the canonical BK14 dust model - A_d=4.25uK^2 at ell=80 and 353GHz with spatial power spectrum slope alpha_d=-0.4, and extrapolation to other frequencies assuming beta_d=1.6 and T_d=19.6K. See arxiv/1510.09217 for details - BK data drive the A_d value, but the others come from Planck analyses. The vans_d1 model hits agrees quite well with this parameterization while the pysm_d4 model does not. We note that the vans_d1 model does have some decorrelation of the dust pattern with frequency.
Here is a final plot which is the exact analog of the above except using the 3% effective circular mask which we have been using for CMB-S4 studies (see [01.00_sim_input_maps] for plot of it). We see the decorrelation effect is much stronger - this probably needs to be investigated a bit more.