Decorrelation in the new dust models

[myra-norton]

I'm working with @brandonshensley to quantify decorrelation in the new models. Here is a first notebook + plot!

[giuspugl]

Thanks a lot @myra-norton this is really great!
It's not immediate to me to understand why we do see a decrease in the decorrelation at ell>100 ? Do we expect small scales injected w/ poltens to produce an increase in freq. decorrelation ?
Also are those results consistent w/ planck latest constraints ?

[brandonshensley]

It's not immediate to me to understand why we do see a decrease in the decorrelation at ell>100 ? Do we expect small scales injected w/ poltens to produce an increase in freq. decorrelation ? Also are those results consistent w/ planck latest constraints ?

There is decorrelation in the small scales because there are fluctuations in the T and beta maps on small scales. I would guess that the feature at ell ~ 100 is caused by a feature in the power spectra of the T and beta maps. I would not expect the decorrelation to go to zero at high ell, but we also haven't calculated it out that far.

We did not use exactly the same mask as the Planck analysis (can anyone point us to those masks?), but the LR71 region most comparable to our GAL070 mask has a lower limit of R_{50 < \ell < 160}^BB > 0.991 (95% confidence, see Table 5).

[brandonshensley]

I've made an updated version of this plot, see linked notebook.

[zonca]

@brandonshensley what about adding this analysis to the docs?

at the top of the Dust section at https://pysm3.readthedocs.io/en/latest/models.html#dust I can put a link to this.
If we decide to proceed with this, we will need more text inside the notebook that explains what it is doing and exact link to section of the original paper.

The notebook can then directly be added to the docs.

[brandonshensley]

@zonca that's a great idea! Let's wait until after internal review and convergence to a final version, but it would be a good addition to that page. I can expand the text.

[brandonshensley]

Good news: I redid the plot with anafast and got an identical result. Notebook here.

[zonca]

@brandonshensley does this go into the paper? If so you could just add it to the pysm_methods_paper repo

[brandonshensley]

@brandonshensley does this go into the paper? If so you could just add it to the pysm_methods_paper repo

Yes, will do once @1cosmologist and @kennykinglau have had a look to be sure the methodology agrees with theirs.

[1cosmologist]

@brandonshensley One question: Should you bin R_ell (where R_ell is computed for each ell) or compute R_ell from Cls that are binned over a single big bin? Ratio of averages vs average of ratios.

[brandonshensley]

@brandonshensley One question: Should you bin R_ell (where R_ell is computed for each ell) or compute R_ell from Cls that are binned over a single big bin? Ratio of averages vs average of ratios.

I thought it made more sense to bin the power spectra then compute R_ell last (i.e., we are reporting the correlation between the BB power averaged over a broad ell bin vs the average level of correlation over that bin). Since R_ell is generally pretty flat with ell, I don't think it matters much.

[1cosmologist]

Fair enough.