Survey Performance Expectations

From CMB-S4 wiki
Jump to navigationJump to search

This page summarizes the various survey performance expectations (previously called "experiment definitions.") The data products referred to are located on the NERSC system under /project/projectdirs/cmbs4/expt_xx

Large-area Survey Performance Expectation 01

The NTT should fill in here what they can, with input from large-area forecasters where needed. Ideally there would be separate specifications of instrument capabilities, and assumed sky coverage, rather than having these lumped together.

The numbers below are based on the Noise Tiger Team code (but 181016 yields the same numbers), run assuming the default settings and fsky=0.4. The default settings include 2 LATs, each equipped with 1 ULF tube, 2 LF tubes, 12 MF tubes, and 4 UHF tubes. Note that the EE and BB noise levels are sqrt(2) higher than the TT noise levels and are omitted from the table below for brevity. The atmospheric treatment and other details are described in the code and in the lat-noise-*.pdf document linked below.

Frequency (GHz) 20 27 39 93 145 225 280
Bandwidth (GHz) n/a n/a n/a n/a n/a n/a n/a
Beam FWHM (arcmin) 10. 7.4 5.1 2.2 1.4 1.0 0.9
white noise level TT (uK-arcmin) 45.9 15.5 8.7 1.5 1.5 4.8 11.5

Full noise model:

Post-component-separation (harmonic-space ILC) noise curves from Colin H.:

The file name conventions are very similar to those from the Simons Observatory noise files located at . In the S4 case, there is only one sensitivity option (labeled "SENS0", but this doesn't mean anything, it's just the default S4 setup). There is also only one fsky option (16000 deg^2, i.e., 40%).

I include results for both temperature and polarization observables. Each temperature file contains three columns: [ell] [N(ell)^TT in uK^2] [N(ell)^yy]

Each polarization file also contains three columns: [ell] [N(ell)^EE in uK^2] [N(ell)^BB in uK^2]

The "deproj" conventions are identical to what is described in the SO science paper (deproj0 = standard ILC, etc) -- see Sec. 2 of . Contact jch 'at' if anything is unclear, and/or please refer to Sec. 2 of the SO paper.

I also include pure inverse-covariance-weighted noise curves for both TT and EE/BB (i.e., with zero foregrounds).

I also include the CMB lensing reconstruction noise for the TT and EB estimators (for each of the deproj options), but people should probably re-compute this using their codes (I just used the standard QE, which is of course sub-optimal here, and I also only went down to L ~ 40).

Small-area Survey Performance Expectation 05

What we have here in "Small-area Survey Performance Expectation 05" does not fit the intended mold. But it seems practical. Here we are not specifying the map noise performance of a single instrument and survey. Instead, we are specifying some assumptions for a trade study among different instrument configurations.

The SAT group is currently using the Fisher forecasting machinery used for the Science Book and validated with map-based exercises in the CDT process, to make r forecasts for a great number of configurations, for both Chile and Pole. The plan is to adjust the number of optics tubes for each configuration, and distribution over sites, to hit the target sigma(r). Then the configurations can be compared for cost and risk at fixed sigma(r).

NTT prescription for these forecasts: does not exist yet.

De-lensing procedure for these forecasts: Adopt A_L = 0.13 for the Pole survey, and A_L = 0.27 for the Chile survey, based on this [post|].

Experiment Definition 04

This is very similar to 02 but with the noise levels tweaked down by sqrt(7/6) to try and get sigma(r)=1e-4 as per science requirement. (The noise map generator has also been tweaked such that the level quoted in the params.dat table will be reproduced exactly when the power spectra of the maps is taken with the provided mask - i.e. the actual noise level will slightly lower in the center of the mask.) A 20GHz band has been added which is assumed to be on a large aperture telescope (and hence has higher resolution and ell knee). Below is the contents of the params.dat file. The noise levels and resolutions are copied from the CDT report "Science Traceability Matrix". Specs for the high res bands exist but have not been added yet.

Set 04 is the same idealized circular 3% sky patch as set 02. We also add 04b which is a for a nominal Chile realistic mask and 04c which is a realistic Pole mask (actually the BICEP3 2017 as observed mask). See this logbook posting for details.

Frequency (GHz) 20 30 40 85 95 145 155 220 270
Bandwidth (GHz) 5.0 9.0 12.0 20.4 22.8 31.9 34.1 48.4 59.4
Beam FWHM (arcmin) 11.0 76.6 57.5 27.0 24.2 15.9 14.8 10.7 8.5
white noise level TT (uK-arcmin) 16.66 10.62 10.07 2.01 1.59 4.53 4.53 11.61 15.84
ell knee TT 500 175 175 175 175 230 230 230 230
1/f exponent TT -4.1 -4.1 -4.1 -4.1 -4.1 -3.8 -3.8 -3.8 -3.8
white noise level EE (uK-arcmin) 13.94 8.88 8.42 1.67 1.32 2.12 2.12 5.43 7.42
ell knee EE 200 50 50 50 50 65 65 65 65
1/f exponent EE -2.0 -2.0 -2.0 -2.0 -2.0 -3.0 -3.0 -3.0 -3.0
white noise level BB (uK-arcmin) 13.6 8.67 8.22 1.64 1.30 2.03 2.03 5.19 7.08
ell knee BB 200 50 50 50 50 60 60 60 60
1/f exponent BB -2.0 -2.0 -2.0 -2.0 -2.0 -3.0 -3.0 -3.0 -3.0
ell min 30 30 30 30 30 30 30 30 30
nside 512 512 512 512 512 512 512 512 512

Experiment Definition 03, 03b, 03c

These sims are nearly identical to experiment definition 02, 02b, 02c, and even use the same noise realizations. The one change is the inclusion of additive systematics, following the prescription described in this 2017-07-10 posting. Blocks of realizations contain different versions of the additive systematic, as shown below. For all cases, the level of the systematic is chosen to yield bias on r of 1e-4 (by the analysis of the 2017-07-10 posting). The first block of realizations, 0000–0124, contains no systematic and should be completely identical to the 02/02b/02c realizations.

realizations Auncorr Buncorr Acorr Bcorr Description
0000–0124 0 0 0 0 No systematics
0125–0249 0.0328 0 0 0 Uncorrelated systematic with white spectrum
0250–0374 0 0.0689 0 0 Uncorrelated systematic with 1/ell spectrum
0375–0499 0 0 0.0584 0 Correlated systematic with white spectrum
0500–0624 0 0 0 0.1049 Correlated systematic with 1/ell spectrum
0625–0749 0.0164 0.0345 0 0 Uncorrelated systematic with white + 1/ell spectrum
0750–0874 0 0 0.0292 0.0525 Correlated systematic with white + 1/ell spectrum
0875–0999 0.0082 0.0172 0.0146 0.0262 Uncorrelated + correlated systematic with white + 1/ell spectrum

Combined maps for analysis can be found on NERSC in /project/projectdirs/cmbs4/data_xx.yy/03 etc.

Parameters and ingredients for the combined maps can be found in /project/projectdirs/cmbs4/expt_xx/03 etc. Note that noise, rhits, and wfunc are actually symlinks to those directories under experiment configs 02, 02b, 02c.

Experiment Definition 02, 02b, 02c

This is an update from 01 which differs in the following ways:

1) Addition of 20GHz band and (slight) changes to the band/detector optimization as per [Victor's May 15 posting in the logbook].

2) Addition of 3 delensing bands 95/155/220GHz with same number of detectors as the low res equivalent bands but beam size 4x higher. This adds up to approximately the same number of detectors but clearly one should reoptimize once they are not all at the same freq. Since we are still quite far from having real lensing reconstruction and realistic (non-Gaussian) high ell foregrounds this probably is OK for this round.

3) Addition of tensors - even number realizations have r=0.003, odd number have r=0.

4) Set 02 is the same nominal 3% sky patch as set 01. We now add 02b which is 1% round patch with center the same as the BICEP/Keck patch (RA=0, Dec=-57.5), and 02c which is 10% round patch centered on RA=15deg, Dec=-35.

These products appear on NERSC under /project/projectdirs/cmbs4/expt_xx/02 etc.

Experiment Definition 01

This is intended to be basically the same as the assumptions made for the Fisher calculations done for the Science Book. The parameters come from [Victor's Dec 21 posting in the logbook] with the addition of bandwidths from Colin's Nov 4 posting and are summarized in the following table:

Frequency (GHz) 30 40 85 95 145 155 220 270 155 HR
Bandwidth (GHz) 9.0 12.0 20.4 22.8 31.9 34.1 48.4 59.4 34.1
Beam FWHM (arcmin) 76.6 57.5 27.0 24.2 15.9 14.8 10.7 8.5 4.0
white noise level TT (uK-arcmin) 12.97 13.22 2.30 1.89 5.31 5.48 11.86 17.72 5.48
ell knee TT 175 175 175 175 230 230 230 230 500
1/f exponent TT -4.1 -4.1 -4.1 -4.1 -3.8 -3.8 -3.8 -3.8 -3.8
white noise level EE (uK-arcmin) 10.85 11.06 1.93 1.58 2.49 2.56 5.55 8.30 2.56
ell knee EE 50 50 50 50 65 65 65 65 200
1/f exponent EE -2.0 -2.0 -2.0 -2.0 -3.0 -3.0 -3.0 -3.0 -3.0
white noise level BB (uK-arcmin) 10.59 10.79 1.88 1.54 2.38 2.45 5.30 7.93 2.45
ell knee BB 50 50 50 50 60 60 60 60 200
1/f exponent BB -2.0 -2.0 -2.0 -2.0 -3.0 -3.0 -3.0 -3.0 -3.0
ell min 30 30 30 30 30 30 30 30 100
nside 512 512 512 512 512 512 512 512 2048

These parameters are in the file expt_xx/01/params.dat and give rise to the window functions in expt_xx/01/wfunc and the noise spectra in expt_xx/01/noise/cls. The intent of the low ell cutoff is to approximate the effect of the timestream filtering which is commonly done in ground based experiments (and as some kind of acknowledgement that the largest angular scales may anyway be corrupted by systematic effects).

Additionally expt_xx/01/rhits/fsky03bk_n0512.fits is supposed to represent the "relative hits" with which the sky has been observed. This is a circular pattern centred at RA=0, Dec=-45 (a bit below the BICEP/Keck patch) which is flat at one out to r=12deg and then rolls down to zero as cosine squared over an additional 15 deg. This is some kind of approximation as to what small apertures might deliver (given their large instantaneous field of view).

From the noise spectra sets of alms are generated (with nlmax=4096) with names like expt_xx/01/noise/alm/noise_f155_b15_ellmin30_alm_mc_0000.fits where 155 is the frequency, 15 is the beam size (in arcmin), and 0000 is the realization number.

From these full sky maps are rendered at nside=512 and nlmax=1024. No beam smoothing is applied as is appropriate for noise. The maps have names like expt_xx/01/noise/map/noise_f155_b15_ellmin30_map_0512_mc_0000.fits.

The full sky noise is then divided by the square-root of the rhits mask and the result stored in files like data_xx.yy/01.YY/cmbs4_01_noise_f155_b15_ellmin30_map_0512_mc_0000.fits - these are the noise realizations that one would have access to in a real experiment - the noise blows up around the edge.

These masked noise realizations are then added to the sky model (LCDM+dust+sync+) and stored in names like data_xx.yy/01.YY/cmbs4_01pYY_comb_f155_b15_ellmin30_map_0512_mc_0000.fits. As a crude simulation of delensing these combined maps are also provided with the lensing effect reduced to 30, 10 and 3% of reality by combining lensed and unlensed LCDM - look for files with names like data_xx.yy/01.YY/cmbs4_01pYY_comb_AL0p3_f155_b15_ellmin30_map_0512_mc_0000.fits. Maps with partial lensing are constructed by the following combination: partially_lensed_map = sqrt(A_L) * lensed_map + (1 - sqrt(A_L)) * unlensed_map.

To analyze these maps one should use only the expt_xx/01/params.dat file and the maps under data_xx.yy/01.YY. A thousand realizations are present. We can see these maps in this posting

The last column of the table above shows one additional tweak - a second 155GHz band with a higher 1/f knee and smaller beam size - this is supposed to represent the maps made by higher resolution telescopes running in concert with small apertures in a hybrid approach. These maps can in principle be used to reconstruct the lensing potential and delens explicitly - instead of cheating by using the AL0p3 files etc. At the moment the noise spectrum in these maps is provisional. Additionally we may wish to add additional frequency bands at higher resolution so foreground cleaning before delensing can be done. We can see these maps in this posting.

(The filtered LLCDM realizations are also copied in data_xx.yy/01.YY as these are needed to build the bandpower covariance matrix in the multi-component cross spectrum approach used by the BK group. It seems reasonable that one would always have realizations of a known spectrum signal passed through an experimental simulation.)