LFI systematics and impact on science Paolo Natoli Università di Ferrara and INFN on behalf of the Planck collaboration and of the LFI Data Processing Center Special credits: L. Colombo, A. Gruppuso, M. Lattanzi, M. Migliaccio, D. Molinari, U. Natale, L. Pagano
The Low Frequency Instrument on Planck LFI was the Low Frequency Instrument on Planck Radiometric detectors like WMAP, but total power Frequency range: 30 – 70 GHz Science Ground Segment in Trieste (INAF-OATs), Italy Planck built to be a component separation machine, LFI provided the low frequency end Hard to disentangle systematics impact on science from LFI But there are interesting exceptions: Large-angle (low-ell) polarization analysis is one Analysis pipelines from the two instruments "almost" independent in terms of data content and methodology Provides nice consistency checks Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
Large-angle polarization Challenging measurement (Jean-Loup Puget's talk) Of some science importance Optical depth t is the worst constrained parameter of the six LCDM (14% relative error in Planck legacy) A full-sky cosmic variance limited measurement of E-mode polarization is still awaited LiteBIRD: will likely provide this “definitive” measurement at large scales It will be crucial to get: Accurate reionization history Neutrino parameters (esp. masses) Large-scale anomalies getting tensor-to-scalar ratio in B-mode reionization peak Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
CMB and cosmic reionization Optical depth to reionization: Second major change of ionization state of H Damps temperature and polarization fluctuations inside horizon Generates new polarization at large angular scales Sunyaev and Chluba 2006 Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
Brief history of t measurements WMAP 9yrs: 𝜏 = 0.089 ± 0.014 (Hinshaw+ 2013) Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
Brief history of t measurements WMAP 9yrs: 𝜏 = 0.089 ± 0.014 (Hinshaw+ 2013) WMAP 9yrs + Planck 353 GHz 𝜏 = 0.075 ± 0.013 (Planck 2013 XV) Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
Brief history of t measurements WMAP 9yrs: 𝜏 = 0.089 ± 0.014 (Hinshaw+ 2013) WMAP 9yrs + Planck 353 GHz 𝜏 = 0.075 ± 0.013 (Planck 2013 XV) WMAP 9yrs + Planck 353 GHz 𝜏 = 0.062 ± 0.012 (Planck 2018 V in pr.) Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
Brief history of t measurements WMAP 9yrs: 𝜏 = 0.089 ± 0.014 (Hinshaw+ 2013) WMAP 9yrs + Planck 353 GHz 𝜏 = 0.075 ± 0.013 (Planck 2013 XV) WMAP 9yrs + Planck 353 GHz 𝜏 = 0.062 ± 0.012 (Planck 2018 V in pr.) Planck lowTEB (LFI) (Planck 2015 XIII) 𝜏 = 0.067 ± 0.023 Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
Brief history of t measurements WMAP 9yrs: 𝜏 = 0.089 ± 0.014 (Hinshaw+ 2013) WMAP 9yrs + Planck 353 GHz 𝜏 = 0.075 ± 0.013 (Planck 2013 XV) WMAP 9yrs + Planck 353 GHz 𝜏 = 0.062 ± 0.012 (Planck 2018 V in pr.) Planck lowTEB (LFI) (Planck 2015 XIII) 𝜏 = 0.067 ± 0.023 Planck lowTEB (LFI) (Planck 2018 V in pr.) 𝜏 = 0.063 ± 0.020 Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
Brief history of t measurements WMAP 9yrs: 𝜏 = 0.089 ± 0.014 (Hinshaw+ 2013) WMAP 9yrs + Planck 353 GHz 𝜏 = 0.075 ± 0.013 (Planck 2013 XV) WMAP 9yrs + Planck 353 GHz 𝜏 = 0.062 ± 0.012 (Planck 2018 V in pr.) Planck lowTEB (LFI) (Planck 2015 XIII) 𝜏 = 0.067 ± 0.023 Planck lowTEB (LFI) (Planck 2018 V in pr.) 𝜏 = 0.063 ± 0.020 Planck lowE (HFI) (Planck 2016. XLVI) 𝜏 = 0.055 ± 0.009 Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
Brief history of t measurements WMAP 9yrs: 𝜏 = 0.089 ± 0.014 (Hinshaw+ 2013) WMAP 9yrs + Planck 353 GHz 𝜏 = 0.075 ± 0.013 (Planck 2013 XV) WMAP 9yrs + Planck 353 GHz 𝜏 = 0.062 ± 0.012 (Planck 2018 V in pr.) Planck lowTEB (LFI) (Planck 2015 XIII) 𝜏 = 0.067 ± 0.023 Planck lowTEB (LFI) (Planck 2018 V in pr.) 𝜏 = 0.063 ± 0.020 Planck lowE (HFI) (Planck 2016. XLVI) 𝜏 = 0.055 ± 0.009 Planck lowE (HFI) (Planck 2018 V in pr.) 𝜏 = 0.0506 ± 0.0086 WMAP LFI HFI Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
Foreground mitigation: LFI and HFI Map-based template fitting procedure to mitigate foreground contamination, using low and high channels as tracers of polarized synchrotron and dust. Same methodology for LFI and HFI Template cleaning of maps m = [Q, U] for the channels= 70, 100, 143 GHz 353 GHz 30 GHz 30 GHz: tracer of synchrotron emission 353 GHz: tracer of thermal dust emission Foreground coefficients fitted by minimising, on about 70% of the sky: Where the rescaled covariance matrix is: Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
Pixel-based likelihood modelling While HFI uses a cross-spectrum based likelihood driven by simulations, LFI uses a map-based formalism Based on cleaned 70 GHz map Statistically efficient Sensitive to full TEB (HFI is EE-only) Provided CMB signal covariance is adapted, can probe non-isotropic models But prone to map-level systematics Requires accurate estimation of the underlying covariance matrix fSKY = 86% fSKY = 62.4% Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
Correcting for noise mismatch Estimating the noise covariance matrix is hard: It is a by-product of the map-making scheme, but exact computation is prohibitive Have to resort to approximations. This works reasonably well for LFI (less well for HFI) Still, noise mismatches arise. This can be measured from half-ring half-difference maps, and the noise matrix properly debiased EE noise bias BB noise bias 1s shade HRHD bias Rescaled noise matrix bias Original noise matrix bias L. Colombo Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
The LFI gain template Main polarization systematic in low ell LFI comes from calibration, since the presence of polarized foregrounds biases the gain. At 30 and 44 GHz, where polarized foreground are bright, an iterative scheme allows to mitigate the bias by converging on the foreground template. In the end, 30 GHz improves much, while 44 GHz cannot be used for low ell cosmology because of other systematics At 70 GHz diffuse foreground is at a minimum, noise dominates and the iterative scheme does not converge Still a correction is needed: this comes in the form of a gain template, a by-product of the converged 30 GHz iteration scheme, properly rescaled in amplitude. Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
Impact of gain errors Ignoring the gain template results in about 2 sigma bias on t: L. Colombo Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
Goodness of fit The cleaned dataset can be selected through goodness-of-fit analysis a, b from foregrounds, g from systematics R0.6 R0.8 R1.0 R1.2 R1.8X R2.2X M. Migliaccio Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
Goodness of fit The cleaned dataset can be selected through goodness-of-fit analysis a, b from foregrounds, g from systematics R0.6 R0.8 R1.0 R1.2 R1.8X R2.2X Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
Comparison to simulations How can we tell if we are seeing inadequacy of the foreground cleaning at large sky fraction or residual instrumental systematics (e.g. gain template is not perfect). Simulations help: Rifare con script di marina? Cosiderare di tohglerla, ci sono cose importanti che non controlliamo D. Molinari Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
Simulation based mask validation data Statistics for maximum deviation of Dt across masks Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
A joint map from LFI and WMAP Preliminary LFI+WMAP The WMAP low-ell polarization likelihood can be analyzed using a map-based formalism similar for LFI. Good consistency already at the level of angular power spectrum This enables use of a joint (weighted-average by pixel) dataset U. Natale et al. in prep. Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
The LFI+WMAP dataset Good consistency across different Galactic masks Preliminary Good consistency across different Galactic masks Dt scatter by mask compared to simulations U. Natale et al. in prep. Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
t from LFI and WMAP from LFI+WMAP Preliminary Null test is map-based, estimate has ~20% error on t (from low ell only) Includes TE (and, potentially, BB, TB, EB...) But from a pixel-based likelihood. Can be used to constrain models for which an EE spectrum-based likelihood (as HFI) in unsuited, e.g. non-isotropic cosmologies (CMB anomalies) Cosmological birefringence Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
Joint view of existing constraints on t L. Pagano Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
Conclusions and lessons learned In low ell polarization, LFI has provided constraints based on 70 GHz, used as baseline in 2015 and as consistency in 2018. Low ell polarization is an hard measurement. The Planck experience shows that systematic errors can arise at the level of data processing, in the interplay between calibration, map-making and component separation Validation through end-to-end simulations is an extremely important step Solving problems requires time. LFI could benefit from a well integrated team, thanks to strong links between Science Ground Segment and Science exploitation teams. Post-legacy analysis is delivering a joint pixel based dataset for LFI and WMAP, competitive in terms of noise and suitable for many non-standard analyses. Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
Extra Slides Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
Post legacy analysis Preliminary Several, ongoing post legacy activities on LFI Timeline reprocessing and improved calibration with component separation interplay is eventual goal: will take some time. Optimization of the procedure on existing timelines is already yielding results, e.g. in terms of consistency across masking U. Natale et al, in preparation Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
WMAP 9 year dataset The WMAP large-angle polarization likelihood can be analyzed using a map-based formalism similar for LFI Using weighted average for Ka, Q and V for CMB K-band and HFI 353 for synchrotron and dust cleaning As opposed to LFI 30 GHz and HFI 353 GHz for LFI, with CMB only from 70 GHz Good consistency already at the level of angular power spectrum This enables use of a joint (weighted-average by pixel) dataset Preliminary Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019
With and without TE Dt = t(TT+TE+EE) – t(TT+EE) Paolo Natoli – LFI Systematics and Impact on Science – LiteBIRD Kick Off Meeting - JAXS 1 July 2019