Harvard-2017:T2

From CMB-S4 wiki
Jump to: navigation, search

Back to Harvard-2017 main page

Parallel Session T2: Sim WG: systematics & noise realism in S4 simulations--approach and prioritization (Chair: Tom Crawford) [Jefferson 250]

Post talks here.

  • [[:File: ]]

https://docs.google.com/document/d/1vQM3wYBstZIz0CVYwUZDa2rNL44Hc6zv6JujrIMInsA/edit#heading=h.el1vlba3dk9

https://cmb-s4.org/CMB-S4workshops/index.php/Additive_systematics_for_data_challenge_03

https://cmb-s4.org/CMB-S4workshops/index.php/Experiment_Definitions

https://docs.google.com/spreadsheets/d/1aByvdao-6mQYVzcnkm_qv6B4ZENroNKI2Di7CC-1hiQ/edit#gid=0

http://bicep.rc.fas.harvard.edu/CMB-S4/analysis_logbook/20170515_chkS4/

http://bicep.rc.fas.harvard.edu/CMB-S4/analysis_logbook/20170818_HR/

Notes from session

  Systematics (led by Kovac)
  Q: Why are band-band-correlated systematics more troublesome than uncorrelated ones, esp. given that we worry about *de*correlation in foregrounds?
  A: We know and use the frequency structure of the foregrounds to remove them; we do not know (by definition) the frequency structure of the systematics residuals, and it could mimic CMB in the worst case.
  Q: Why are cross-talk effects in additive systematics? They're multiplicative.
  A: The main effect is multiplicative, but the 1st-order residuals are additive.
  Comment: Sometimes the effect of a systematic is most important in how it affects, e.g., calibration, as opposed to directly causing a residual in the maps.
  Response: In the current WG paradigm, we assume the hard work of characterizing that and removing it to 1st order has been done, and the residual is either an additive or multiplicative effect.
  Response to response: Just keep in mind that mixing with calibration etc. can mix additive to multiplicative effects and vice-versa.
  Should we get ourselves externally reviewed to guard against future complaints? Particularly given that the HEP community is fairly skeptical. Mainly we need more outreach to HEP, to show we are not 
  There are lots of systematics on the lensing side not treated here. Stated more clearly, we are not doing anything about systematics in the rest of the survey.
  ACTION ITEM: Implement systematics in de-lensing.
  How do we deal with drain of resources to SO working groups and non-open communication with them? Why do we even need different working groups? Because they are different collaborations and instruments!


  Noise realism
  Talk by Niemack about SO noise calculator
  Q: You say that systematics are included in this calculator, can you say more? 
  A: It's not exactly in the calculator, but it was included in the 
  This very impressive tool is really for instrument design optimization, but going to mapping speed on sky is still done in the old style (assume yield, observing efficiency, etc.).
 Fundamental issue: we have a few different ways of estimating noise curves for delensing survey (and/or high-ell science survey), and they disagree by factors of 2. They are all defensible at some level. How to figure out path forward?


  TOD sims talk by Julian
  proof-of-principle run of 50k detectors observing 20% of the sky in 7 frequencies for a year, includes atmosphere and 1/f noise and calibration uncertainty
    used 600,000 cores
  ok, you have proven that it is not 100% insane to do at least 10% of S4. how do you see this plugging into the current simulation framework
    verifying systematics? how do screw up the input so that you don't just take out what you put in?

Action items/Next steps

Summarize action items here

Implement systematics in de-lensing.

Do a better job at communicating what exactly we do to the HEP community to dispel skepticism about robustness of CMB determinations of fundamental physical parameters. (Not just for this working group!)

Have collaboration-wide discussion (especially with people with feet in both S4 and SO) about getting more coordination between SO and S4 forecasting efforts.

Fundamental issue: we have a few different ways of estimating noise curves for delensing survey (and/or high-ell science survey), and they disagree by factors of 2. They are all defensible at some level. How to figure out path forward?

Forecasting/sims working group to discuss incorporating full-TOD sims.

[Added unilaterally by Tom]: We talk about "Sims WG" but it's really just "r forecasting WG." Need to get more serious about forecasting for other science targets.