GNILC dust model d9 configuration

[zonca]

I propose to continue the PySM 2 numbering, the first model should be d9.

Templates

The new GNILC dust IQU templates are being implemented in #82
What should we use for spectral index and dust temperature?

Resolution

The model will be natively at 8192, so we want to also have a version of it at 512, so it keeps being usable on laptops.
My proposal is to generate all templates both at 8192 and 512 from the same alms. Then if the user requests a NSIDE=512 or lower, the 512 map is loaded, otherwise the 8192 map.

Model implementation

Use ModifiedBlackBody as other standard dust components:

pysm/pysm3/data/presets.cfg

Lines 40 to 52 in 5c339f0

	[d3]
	class = "ModifiedBlackBody"
	map_I = "pysm_2/dust_t_new.fits"
	map_Q = "pysm_2/dust_q_new.fits"
	map_U = "pysm_2/dust_u_new.fits"
	unit_I = "uK_RJ"
	unit_Q = "uK_RJ"
	unit_U = "uK_RJ"
	map_mbb_index = "pysm_2/beta_mean1p59_std0p3.fits"
	map_mbb_temperature = "pysm_2/dust_temp.fits"
	unit_mbb_temperature = "K"
	freq_ref_I = "545 GHz"
	freq_ref_P = "353 GHz"

@giuspugl @seclark @bthorne93 @brandonshensley any feedback?

[brandonshensley]

We should follow Planck 2018 XI: beta = 1.48 for I and 1.53 for Q and U. If both need to be the same, then probably 1.48. For temperature, T_d = 19.6 K.

[zonca]

so you confirm you don't want to use a map but prefer constant over the sky? I can have 2 coefficients for I and pol.
do we also need other model with variable beta and T_d?

[giuspugl]

Having variable spectral parameters are of course one of the challenges for many foreground cleaning methodologies. personally, I am in favor of having another model with spatially variable beta and T_d , e.g. the ones from GNILC maps (see Planck 2016 ).

[zonca]

@giuspugl @brandonshensley instead of having a low-resolution dataset at 512, I think it would be better to have it at 1024. I think with current hardware 1024 is quite manageable even on laptops. Suggestions?

[zonca]

ok, a first implementation of d9 is ready, I was comparing it with d0 and differences are substantial, we probably should address this in paper if not there already. @giuspugl @brandonshensley

see notebook (needs branch of #108): https://gist.github.com/ec542934da2da38109d35ce288c0d3c2

[brandonshensley]

I was expecting differences in I at the ~10% level from the color correction and maybe the 50% level at high latitudes where the CIB amplitude is comparable to the dust amplitude. The variations here are much larger, so we'll need to understand what's driving them. Q and U look better behaved?

[zonca]

I wanted to give Colab a try...please let me know if it doesn't work.

so, this is a notebook which installs the right branch of pysm3 and runs on Google Compute Engine, so no setup required - runs in browser, it should be editable (or you can make a copy), so you can add more interesting plots and inspect the maps:

https://colab.research.google.com/drive/1YSsQaEOoAkd5f3Iodx88LOb_7zHMmvuw?usp=sharing

@brandonshensley @seclark @delabru

[zonca]

implemented in #108