Difference between revisions of "UCSD-2019: Cross-Cut: Delensing"

From CMB-S4 wiki
Jump to: navigation, search
(Notes)
 
(19 intermediate revisions by 4 users not shown)
Line 16: Line 16:
  
 
== Remote Attendance ==
 
== Remote Attendance ==
 
+
https://zoom.us/j/412621262?pwd=djdRc3YvVEZVWGVWMVJHNTZyemtKQT09
[https://ucsd.zoom.us/j/462241746 Zoom link]
 
  
 
== Agenda ==
 
== Agenda ==
Line 23: Line 22:
 
1) On-going work related to delensing (50 min - 1hr):  
 
1) On-going work related to delensing (50 min - 1hr):  
 
* f_NL
 
* f_NL
** Will Coulton on delensing for f_NL
+
** Will Coulton on delensing for f_NL [[File: wcoulton.pdf]]
 
* Foregrounds
 
* Foregrounds
** Marius Millea on incorporating foregrounds in Bayesian delensing
+
** Anton Baleato Lizancos on biases to delensing from CIB and EB estimator [[File: AntonBL_S4meeting.pdf]]
** Anton Baleato Lizancos on biases to delensing from CIB and EB estimator
+
** Marius Millea on incorporating foregrounds in Bayesian delensing [[File:Millea_cmbs4_sandiego_2019.pdf]]
 
<!-- ** Alex van Engelen on Galactic foreground impacts -->
 
<!-- ** Alex van Engelen on Galactic foreground impacts -->
** Kimmy Wu on CIB non-Gaussianity impacts on delensed bandpower covariances
+
<!-- ** Kimmy Wu on CIB non-Gaussianity impacts on delensed bandpower covariances -->
 
* N_eff
 
* N_eff
** Selim Hotinli on delensing for N_eff
+
** Selim Hotinli on delensing for N_eff [[File:SelimHotinli_CMBS4_UCSD.pdf]]
  
  
 
2) What are some of the questions that need answering for moving CMB-S4 to CD-1 readiness in delensing (1hr):   
 
2) What are some of the questions that need answering for moving CMB-S4 to CD-1 readiness in delensing (1hr):   
* How map non-idealities impact delensing efficiency (Toshiya Namikawa)
+
* How map non-idealities impact delensing efficiency [[File: File.pdf]] (Toshiya Namikawa)
* Instrument effects (Kimmy Wu)
+
* Discussion on instrumental systematics: [https://docs.google.com/presentation/d/19IHIesZkqBL7GajtV4NTVoIh1XBBCqXZnPXSkdwflTA/edit?usp=sharing need inputs from all!] (Kimmy Wu)
* What inputs are needed in order to address session charges
+
* What inputs are needed in order to address session charges [https://docs.google.com/presentation/d/1E43NIMJR1wQAUhoe5BoDxI6dqrFkUoJUa6fxeibzj9c/edit?usp=sharing discussion]
  
 
== Notes ==
 
== Notes ==
 +
 +
*Coulton: Delensing for fNL
 +
**Q: why is bias from using internal delensing smaller in bispectrum vs in power spectrum?
 +
**When using f(NL) = 0 sims, no bias; now there is ~10% bias
 +
**Q: Have you considered foregrounds?
 +
 +
* Anton: EB estimator biases to delensing and Galactic dust biases in CIB to delensing
 +
** ell cuts in input B-modes are effective in removing biases.
 +
** Dust biases to delensing using CIB as tracers are small for SO requirements.
 +
 +
*Toshiya: Anisotropic window + inhomogeneous noise effects
 +
**How much do you have to worry about missing modes in the delensing LAT?
 +
**Can S4 use the wide area LAT/SO LAT to fill out the delensing LAT missing modes due to scan patterns from the South Pole? We should not rely on external experiments for our mission critical tasks.
 +
**Lots of back and forth about how much these effects affect S4.
 +
 +
*Marius: forward foreground modeling
 +
** Q: is this linear? Get all orders in phi, lensing operation is linear
 +
** Incorporate foregrounds in Bayesian framework first pass with poisson sources 1-halo term modeling
 +
** Q: does this work better than bias-hardening? Probably, if model is accurate. ES: in bias-hardening; you subtract the form of the point sources no matter what and incur a noise penalty though a more optimal step is to weight the bias and the increased variance. Here it's likely more optimal.
 +
** Q: How well do you have to know the statistics of the foregrounds? Need to be able to get the model correctly
 +
** Q: How to use this to constraint number counts for low-flux limits? have to choose dN/dS model in this procedure, or can sample dN/dS.
 +
** Q: How do you know the sims are right? The same question that everyone is trying to answer
 +
** Q: How does this bridge to more standard analysis? Can remove low ell B modes in input map, and generate phi instead of sample parameters.
 +
 +
*Delensing for Neff - Selim
 +
**N_eff vs Yp degeneracy breaking when delensing is incorporated
 +
**TE gives the best constraints for N_eff
 +
**Q:Maybe gradient inversion is better for the intermediate scales?
 +
**When Yp not fixed to BBN consistency, sigma(N_eff) degrades to ~0.15?
 +
**How much, without marginalizing over Yp, does delensing help in sigma(N_eff)? Not very much. With Y_p, get ~30% improvement
 +
** This is an update to Joel’s forecast in which the iterative code includes all the other estimators (previously it's EB only).
 +
 +
*Discussion regarding the session charges:
 +
** Do we weight the cluster/transient science goals the same as r science goal when designing the delensing LAT? Yes. So we have to do a joint optimization of sky/beam/freq bands for all three goals.
 +
** The sky coverage is likely driven by the SAT survey. The beam size is likely driven by cluster/transient needs. Frequency distribution needs more study from both the delensing side and the cluster side.
 +
** What is minimum dish size / throughput needed for delensing? See Fig 68 in DSR. Do we need more detailed studies?
 +
** Bonus question from Tom Crawford: "how much do you gain in sigma(r) by matching the degree-scale weight map vs. uniformly covering the full nonzero-weight region?”
 +
** Foreground impacts are discussed through the presentations.

Latest revision as of 10:25, 19 October 2019

Link back to agenda

Charge

General charge to all parallels:

We need to use the October collaboration meeting to advance our preparations for CD-1. Right away, we need to:

  • Identify key decisions that must be made (and justified) prior to CD-1,
  • Make progress on (or actually make) those decisions,
  • Lay out a timeline and process for making each decision, consistent with the post-decision work and internal reviews that will be needed to complete preparations for CD-1,
  • Ensure that those timelines and processes are understood and supported by the collaboration, and that we (together) believe we can follow them.

Specific charge to this delensing cross-cut session:

  • For the de-lensing LAT, what are the benefits and costs of variation in angular resolution (dish size) for legacy survey science goals?
  • What is minimum dish size / throughput needed for delensing? Would multiple smaller telescopes be better than one bigger one? (both r + Neff)
  • What are the effects of foregrounds on delensing (both for r and Neff)

Remote Attendance

https://zoom.us/j/412621262?pwd=djdRc3YvVEZVWGVWMVJHNTZyemtKQT09

Agenda

1) On-going work related to delensing (50 min - 1hr):


2) What are some of the questions that need answering for moving CMB-S4 to CD-1 readiness in delensing (1hr):

  • How map non-idealities impact delensing efficiency File:File.pdf (Toshiya Namikawa)
  • Discussion on instrumental systematics: need inputs from all! (Kimmy Wu)
  • What inputs are needed in order to address session charges discussion

Notes

  • Coulton: Delensing for fNL
    • Q: why is bias from using internal delensing smaller in bispectrum vs in power spectrum?
    • When using f(NL) = 0 sims, no bias; now there is ~10% bias
    • Q: Have you considered foregrounds?
  • Anton: EB estimator biases to delensing and Galactic dust biases in CIB to delensing
    • ell cuts in input B-modes are effective in removing biases.
    • Dust biases to delensing using CIB as tracers are small for SO requirements.
  • Toshiya: Anisotropic window + inhomogeneous noise effects
    • How much do you have to worry about missing modes in the delensing LAT?
    • Can S4 use the wide area LAT/SO LAT to fill out the delensing LAT missing modes due to scan patterns from the South Pole? We should not rely on external experiments for our mission critical tasks.
    • Lots of back and forth about how much these effects affect S4.
  • Marius: forward foreground modeling
    • Q: is this linear? Get all orders in phi, lensing operation is linear
    • Incorporate foregrounds in Bayesian framework first pass with poisson sources 1-halo term modeling
    • Q: does this work better than bias-hardening? Probably, if model is accurate. ES: in bias-hardening; you subtract the form of the point sources no matter what and incur a noise penalty though a more optimal step is to weight the bias and the increased variance. Here it's likely more optimal.
    • Q: How well do you have to know the statistics of the foregrounds? Need to be able to get the model correctly
    • Q: How to use this to constraint number counts for low-flux limits? have to choose dN/dS model in this procedure, or can sample dN/dS.
    • Q: How do you know the sims are right? The same question that everyone is trying to answer
    • Q: How does this bridge to more standard analysis? Can remove low ell B modes in input map, and generate phi instead of sample parameters.
  • Delensing for Neff - Selim
    • N_eff vs Yp degeneracy breaking when delensing is incorporated
    • TE gives the best constraints for N_eff
    • Q:Maybe gradient inversion is better for the intermediate scales?
    • When Yp not fixed to BBN consistency, sigma(N_eff) degrades to ~0.15?
    • How much, without marginalizing over Yp, does delensing help in sigma(N_eff)? Not very much. With Y_p, get ~30% improvement
    • This is an update to Joel’s forecast in which the iterative code includes all the other estimators (previously it's EB only).
  • Discussion regarding the session charges:
    • Do we weight the cluster/transient science goals the same as r science goal when designing the delensing LAT? Yes. So we have to do a joint optimization of sky/beam/freq bands for all three goals.
    • The sky coverage is likely driven by the SAT survey. The beam size is likely driven by cluster/transient needs. Frequency distribution needs more study from both the delensing side and the cluster side.
    • What is minimum dish size / throughput needed for delensing? See Fig 68 in DSR. Do we need more detailed studies?
    • Bonus question from Tom Crawford: "how much do you gain in sigma(r) by matching the degree-scale weight map vs. uniformly covering the full nonzero-weight region?”
    • Foreground impacts are discussed through the presentations.