New large-scale CIB maps from Planck data

Presentation at the Planck legacy data release meeting at COSPAR2018 in Pasadena, CA.

Transcript

1. New large-scale CIB
   maps from Planck data
   in collaboration with O. Doré,
   G. Lagache, B. Hensley
   Daniel Lenz
   COSPAR 2018, Pasadena
   July 20th
   © 2018 California Institute of Technology. Government sponsorship acknowledged.
   

   View Slide

2. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
   Conclusions
   ❖ New CIB maps for ~30%
   of the sky, 217-857 GHz
   ❖ Fewer systematics, larger
   sky fraction than
   previous work
   ❖ Powerful for cross-
   correlations and de-
   lensing
   !2
   

   View Slide

3. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
   Conclusions
   ❖ New CIB maps for ~30%
   of the sky, 217-857 GHz
   ❖ Fewer systematics, larger
   sky fraction than
   previous work
   ❖ Powerful for cross-
   correlations and de-
   lensing
   !2
   CIB x CMB lensing
   

   View Slide

4. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
   What is the CIB?
   ❖ Made up from dust
   in galaxies at z=1-3
   ❖ First detected in
   FIRAS data
   (Puget+ 1996)
   Extragalactic background light
   !3
   

   View Slide

5. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
   What is the CIB?
   ❖ Made up from dust
   in galaxies at z=1-3
   ❖ First detected in
   FIRAS data
   (Puget+ 1996)
   Extragalactic background light
   !3
   

   View Slide

6. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
   Schmidt+ (2015)
   ❖ Strong constraints on star
   formation history
   ❖ Probe dust temperature
   across cosmic times
   ❖ Understand star
   formation in DM halos
   !4
   Why study the CIB? Star-formation!
   

   View Slide

7. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
   Manzotti (2017)
   Why study the CIB? Grav. lensing!
   ❖ CIB kernel and the
   CMB lensing kernel
   are well matched
   ❖ Internal de-lensing
   and CIB is very
   complimentary for
   BB reconstruction
   !5
   

   View Slide

8. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
   Planck collaboration (2013, XVII)
   !6
   Why study the CIB? Grav. lensing!
   ❖ Cross-correlation of CIB and
   CMB lensing strongly detected
   in Planck data
   ❖ Lots of room for improvement:
   Sky fraction, CIB data, new
   CMB lensing map
   

   View Slide

9. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
   How to obtain CIB maps?
   ❖ Galactic thermal dust and CIB dust dominate on large scales at
   ~200 to 1000 GHz
   ❖ How to disentangle them?
   !7
   

   View Slide

10. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    How to obtain CIB maps?
    A. Fit different frequency channels with modified blackbody spectra
    B. Use the different angular power spectra of these components (GNILC)
    C. Use template maps of Galactic dust (e.g. HI-based)
    ❖ Galactic thermal dust and CIB dust dominate on large scales at
    ~200 to 1000 GHz
    ❖ How to disentangle them?
    !7
    

    View Slide

11. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Correlation of dust and gas
    HI Dust
    ❖ Linear relation to first order (Boulanger+ 1996)
    ❖ But better model required to get to CIB levels
    !8
    

    View Slide

12. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    HI4PI Survey
    ❖ Merges data from Effelsberg and Parkes
    ❖ Replaces LAB as state-of-the-art full-sky HI survey
    ❖ Higher sensitivity & resolution, fewer systematics, full sampling
    20
    21
    22
    log(NHI
    [cm 2])
    180
    135 90
    45
    0
    315
    270
    225 180
    60
    30
    0
    30
    60
    HI4PI collaboration

    (2017)
    !9
    

    View Slide

13. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Two challenges
    !10
    ❖ Spectrally
    ❖ O(1000) velocity channels in HI
    ❖ Need to control overfitting
    

    View Slide

14. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Two challenges
    ❖ Spatially
    ❖ Dust-to-gas ratios vary over the sky
    ❖ Need to preserve large-scale CIB power
    !10
    ❖ Spectrally
    ❖ O(1000) velocity channels in HI
    ❖ Need to control overfitting
    

    View Slide

15. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    HI-based dust models
    •
    • Velocity separation difficult for
    complex structures and large scales
    Radial Velocity
    HVC
    IVC
    LVC
    I = ✏HVC NHVC + ✏IVC NIVC + ✏LVC NLVC
    !11
    

    View Slide

16. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    • Generalised linear model (GLM)
    •
    Radial Velocity
    I =
    X
    i
    ✏iTi
    B
    !12
    HI-based dust models
    

    View Slide

17. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    • Generalised linear model (GLM)
    •
    • Regularised:
    •
    • Accounts for all features along
    line of sight
    I =
    X
    i
    ✏iTi
    B
    Radial Velocity
    |Datai Modeli
    |2 + ↵ · |✏i
    |
    !13
    HI-based dust models
    

    View Slide

18. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Preliminary Results
    !14
    (give us two weeks)
    

    View Slide

19. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Maps: Smaller regions
    !15
    

    View Slide

20. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Maps: Large-scale map
    !16
    ~30% of the sky, 5 frequencies, 10 arcmin
    

    View Slide

21. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Gaussianity
    !17
    ❖ Patch-by-patch
    analysis
    ❖ Full sky PDF
    very Gaussian
    ❖ Molecular gas
    adds skewness
    

    View Slide

22. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Comparison to earlier work
    !18
    Maps
    

    View Slide

23. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Based on spatial information: GNILC
    ❖ Power-spectrum
    based
    ❖ Designed to
    remove CIB from
    Galactic dust
    maps
    ❖ Over-subtraction
    of CIB
    Planck (2016 XLVIII)
    !19
    

    View Slide

24. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    HI-based: Planck (2014 XXX)
    ❖ ~10 individual fields, HI data
    from the GBT
    ❖ Two larger fields from EBHIS
    and GASS
    ❖ One field cleaned at a time
    ❖ Manual fine-tuning
    !20
    

    View Slide

25. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Comparison to earlier work: Small fields
    !21
    ❖ Different data sets, resolutions, sky regions
    ❖ Apples-to-apples comparison yields great agreement
    

    View Slide

26. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Comparison to earlier work: Small fields
    !21
    ❖ Different data sets, resolutions, sky regions
    ❖ Apples-to-apples comparison yields great agreement
    

    View Slide

27. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Comparison to earlier work
    !22
    Power spectra
    

    View Slide

28. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    CIB auto power spectra
    !23
    unconstrained
    ❖ Great agreement with Planck (2014 XXX)
    ❖ Extends to larger scales
    ❖ Maps will be public
    

    View Slide

29. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    CIB - CMB lensing cross power
    !24
    unconstrained
    ❖ Great agreement
    with Planck
    (2013 XVIII)
    ❖ Extends to larger
    scales
    ❖ GNILC x Phi
    shows weaker
    correlation
    

    View Slide

30. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    CIB - CMB lensing cross correlation coefficient
    !25
    ❖ > 60% correlation for l
    >= 100
    ❖ ~10-15% higher than
    with GNILC CIB
    ❖ Powerful in combination
    with Planck lensing map
    for BB de-lensing
    

    View Slide

31. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Conclusions
    ❖ Large-scale Planck CIB maps for 5 frequencies
    ❖ Significant improvement in component separation
    ❖ Better understanding of systematics
    ❖ Large scales are challenging!
    !26
    

    View Slide

32. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Conclusions
    ❖ Large-scale Planck CIB maps for 5 frequencies
    ❖ Significant improvement in component separation
    ❖ Better understanding of systematics
    ❖ Large scales are challenging!
    ❖ CIB is powerful probe of large-scale structure
    ❖ Study cosmic star-formation
    ❖ De-lensing for current and future CMB experiments
    !26
    

    View Slide

33. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Thank you!
    !27
    

    View Slide

34. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Backup
    !28
    

    View Slide

35. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Large-scale bias
    !29
    Input CIB
    

    View Slide

36. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Large-scale bias
    !29
    Mean per field
    Input CIB
    

    View Slide

37. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Large-scale bias
    !29
    Mean per field
    After mean subtraction
    Input CIB
    

    View Slide

38. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Large-scale bias
    !30
    Large scales Small scales
    

    View Slide

39. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Large-scale bias
    !30
    ❖ Separating one region at a time removes large-scale power
    ❖ Essential for CIB reconstruction at low l
    Large scales Small scales
    

    View Slide

40. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Spatial selection
    ❖ Build dust models that preserve large-scale power
    ❖ Use consistency checks and cross correlations
    ❖ Difficult trade-off!
    !31
    Offsets in the HI/
    dust correlation
    

    View Slide

41. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Spatial selection
    !32
    Offsets in the HI/
    dust correlation
    (smoothed)
    ❖ Build dust models that preserve large-scale power
    ❖ Use consistency checks and cross correlations
    ❖ Difficult trade-off!
    

    View Slide

42. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Based on spatial information: GNILC
    Planck (2016 XLVIII
    !33
    

    View Slide

43. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Comparison to earlier work: Small fields
    ❖ Very similar morphologies despite totally different spatial
    selections
    ❖ Yet differences remain!
    MJy/sr
    !34
    

    View Slide

44. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Comparison to earlier work: Small fields
    ❖ Very similar morphologies despite totally different spatial
    selections
    ❖ Yet differences remain!
    MJy/sr
    !34
    

    View Slide

45. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Comparison to earlier work: Small fields
    ❖ Differences can be partially attributed to the underlying HI data
    ❖ Radial velocity cuts have strong effect
    !35
    

    View Slide

46. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Comparison to earlier work: Large field
    Planck (2014 XXX) This work
    !36
    

    View Slide

47. Daniel Lenz, JPL/Caltech Large-scale CIB maps from Planck data
    Comparison to earlier work: Large field
    Planck (2014 XXX) This work
    Planck 2014 - This work
    !36
    

    View Slide