Difference between revisions of "Fisher forecasts for inverse noise variance weighting"

From CMB-S4 wiki
Jump to navigationJump to search
(Created page with "This posting shows plots from my Fisher code as well as Ben's 400px 400px When you look at thes...")
 
Line 4: Line 4:
  
  
When you look at these plots, please keep in mind that these are all for INVERSE NOISE VARIANCE WEIGHTING as is also done in the map-based simulations. This is important to keep in mind to avoid confusion. That said, the legend is
+
These are all for inverse noise variance weighting as is also done in the map-based simulations. This is important to keep in mind to avoid confusion. That said, the legend is
  
 
Orange: Pole deep (c mask) (AL=0.076)
 
Orange: Pole deep (c mask) (AL=0.076)
Line 13: Line 13:
 
Green:  Deepest region in Reijo’s Chile deep v1 (AL=0.081)
 
Green:  Deepest region in Reijo’s Chile deep v1 (AL=0.081)
  
Again, keep in mind these are INVERSE NOISE VARIANCE WEIGHTED. This is suboptimal and isn’t what one would do in practice but is what has been done so far in the map-based simulations. This explains the seemingly odd behavior at large r for Chile deep v1 vs Chile shallow (which is a subset). Somewhere between r=0.003 and r=0.01 the shallow survey reaches S/N>1 per mode. With inverse noise variance weighting, the analysis of the full survey downweights these modes and most strongly weights the modes in the deep field, missing much of the information. In practice this just means an inverse noise variance weighted analysis for Chile deep v1 for r>0.005 or so is (rather) suboptimal. I have plots where one can see these transitions, but perhaps they should go in a posting.
+
Inverse varianc weighting is suboptimal and isn’t what one would do in practice. This explains the seemingly odd behavior at large r for Chile deep v1 vs Chile shallow (which is a subset). Somewhere between r=0.003 and r=0.01 the shallow survey reaches S/N>1 per mode. With inverse noise variance weighting, the analysis of the full survey downweights these modes and most strongly weights the modes in the deep field, missing much of the information. In practice this just means an inverse noise variance weighted analysis for Chile deep v1 for r>0.005 or so is (rather) suboptimal. So this plot should not be used to draw conclusions about site distributions.  
  
 
Another related caveat holds for the pole. The SAT sensitivities have been increased substantially (this is all for the 2-6-6-4 configuration), but the delensing levels have remained the same. This means S/N per mode for the pole is significantly above 1, which is also suboptimal. So for the SAT map depth achieved by pole, one should delens more.
 
Another related caveat holds for the pole. The SAT sensitivities have been increased substantially (this is all for the 2-6-6-4 configuration), but the delensing levels have remained the same. This means S/N per mode for the pole is significantly above 1, which is also suboptimal. So for the SAT map depth achieved by pole, one should delens more.

Revision as of 10:59, 29 April 2019

This posting shows plots from my Fisher code as well as Ben's

Sigmarofr.png Sigmar r 18tubes 7years fullchile 6mask.png


These are all for inverse noise variance weighting as is also done in the map-based simulations. This is important to keep in mind to avoid confusion. That said, the legend is

Orange: Pole deep (c mask) (AL=0.076) Red: Pole wide (Reijo) (AL=0.123) Blue: Chile hybrid, pole delensing for the deepest field in Reijo’s Chile deep v1 (AL=0.081), wide area LAT delensing on the rest (AL=0.27) Black: Chile shallow (shallow part of Reijo Chile deep v1 only) wide area LAT delensing (AL=0.27) Purple: Reijo’s Chile deep v1, wide area LAT delensing (AL=0.27) Green: Deepest region in Reijo’s Chile deep v1 (AL=0.081)

Inverse varianc weighting is suboptimal and isn’t what one would do in practice. This explains the seemingly odd behavior at large r for Chile deep v1 vs Chile shallow (which is a subset). Somewhere between r=0.003 and r=0.01 the shallow survey reaches S/N>1 per mode. With inverse noise variance weighting, the analysis of the full survey downweights these modes and most strongly weights the modes in the deep field, missing much of the information. In practice this just means an inverse noise variance weighted analysis for Chile deep v1 for r>0.005 or so is (rather) suboptimal. So this plot should not be used to draw conclusions about site distributions.

Another related caveat holds for the pole. The SAT sensitivities have been increased substantially (this is all for the 2-6-6-4 configuration), but the delensing levels have remained the same. This means S/N per mode for the pole is significantly above 1, which is also suboptimal. So for the SAT map depth achieved by pole, one should delens more.