GT ICR - Marc-Antoine Miville-Deschênes

Transcript

1. IAS, 23 juin 2025 Marc-Antoine Miville-Deschênes Laboratoire de Physique de l’ENS, Paris Le HI dans les galaxies avec GASKAP et SKA
2. Plan ¥ LÕastrophysique du HI ˆ 21 cm ¥ Des avancŽes dans lÕanalyse des donnŽes hyper-spectrales 21 cm ¥ Le projet GASKAP ¥ Traitement des donnŽes GASKAP ¥ Les dŽÞs ˆ venir
3. The science of HI
4. HI in galaxies M81 triplet - de Blok+(2018) VLA + GBT data Eibensteiner+2024 MeerKAT dt Interactions, halos, fountain, IGM relation, full disk dynamicsÉ
7. 21 cm emission of the Milky Way 21 cm - GALFA, Arecibo. credit Joshua Peek
8. 21 cm emission of the Milky Way 21 cm - GALFA, Arecibo. credit Joshua Peek
9. HI H+ diffus supernova halo - CGM H2 formation dÕŽtoiles
10. Lesffre (2018) Most of the ISM mss is in the diffuse ISM
11. HI H+ diffus supernova halo - CGM H2 formation dÕŽtoiles ALMA JWST HST
12. HI H+ diffus supernova halo - CGM H2 formation dÕŽtoiles SKA
13. PHANGS : NGC 628
14. PHANGS : NGC 628 Le milieu diffus ˆ 1 arc-seconde HI (WNM-CNM) H+ (WIM) H2 (OH) Champ magnŽtique Rayons cosmiques Supernovae Dynamique; turbulence, chocs
15. MeerKAT 21 cm observations of nearby galaxies MHONGOOSE Eibensteiner+2024
16. Methodological breakthrough
17. Diffuse isolated clouds : two-phase MHD turbulence
18. Diffuse isolated clouds : two-phase MHD turbulence
20. TB[v] for 16 adjacent lines of sight Integrated emission of TB[x, y, v] Application on 21 cm observations of the North Ecliptic Pole field ROHSA : decomposition of emission on a Gaussian basis Marchal et al. (2019)
21. TB[v] for 16 adjacent lines of sight Integrated emission of TB[x, y, v] Application on 21 cm observations of the North Ecliptic Pole field ROHSA : decomposition of emission on a Gaussian basis Marchal et al. (2019)
22. WNM ~64% Unstable ~28% CNM ~8% Column density • Gaussian decomposition with spatial regularization • CNM is small scale in spatial and velocity space : resolution is key Marchal + (2019) Marchal & Miville-Deschênes (2021) North Ecliptic Pole region 12x12 degrees; 21 cm (GBT data) Methodological step forward : CNM from the emission itself
23. Angular AND spectral resolutions are key Colder HI in nearby galaxies NGC 1566 - MeerKAT (dv=1.4 km/s, res~14”) nlysis : Prk+(2025) dt : de Blok+(2024)
24. Spectral analysis of 21 cm emission A 21 cm spectrum of the diffuse ISM Average Pk_1D of a 512x512 cube in Taurus (GALFA data) Marchal et al. (2024)
25. CNM fraction of the local HI CNM fraction -90 < v < + 90 km/s Marchal et al. (2024)
26. The GASKAP project
27. ASKAP - Australia SKA Pathfinder ¥ 36 dish antennas of 12 m ¥ Maximum baseline = 6 km ¥ Phased Array Feed (PAF) ¥ -> 30 square degree Þeld of view ¥ 700-1800 MHz ¥ 20 arcsec resolution at 1420 MHz ¥ 0.2 km/s velocity resolution Murchison Radio-astronomy observatory also hosting MWA, EDGES and SKA-low
28. ¥ One of 8 ASKAP Key Programs : 21 cm mapping of the MW and Magellanic system ¥ 2250 square degrees (5% of the whole sky) ¥ Magellanic Clouds : 150 deg2 ¥ Mag. Stream 1200 deg2 ¥ Gal Plane : 500 deg2 ¥ Galactic center : 400 deg2 ¥ Resolution : 20Ó, 0.2 km/s ¥ 4400 hours of telescope time ¥ Raw data ~ 3 PB ¥ Final product : ~40 Gb per Þeld (data cube ~ 2000x2000x100) ¥ 11 days of computing time to image one Þeld on commercial supercomputer in Australia ¥ Total project : 60 Þelds, 6 Tb (science ready)
29. In comparison to LoTSS DR2 - 3451 hours - 7.6 PB of data ¥ One of 8 ASKAP Key Programs : 21 cm mapping of the MW and Magellanic system ¥ 2250 square degrees (5% of the whole sky) ¥ Magellanic Clouds : 150 deg2 ¥ Mag. Stream 1200 deg2 ¥ Gal Plane : 500 deg2 ¥ Galactic center : 400 deg2 ¥ Resolution : 20Ó, 0.2 km/s ¥ 4400 hours of telescope time ¥ Raw data ~ 3 PB ¥ Final product : ~40 Gb per Þeld (data cube ~ 2000x2000x100) ¥ 11 days of computing time to image one Þeld on commercial supercomputer in Australia ¥ Total project : 60 Þelds, 6 Tb (science ready)
30. In comparison to LoTSS DR2 - 3451 hours - 7.6 PB of data ¥ One of 8 ASKAP Key Programs : 21 cm mapping of the MW and Magellanic system ¥ 2250 square degrees (5% of the whole sky) ¥ Magellanic Clouds : 150 deg2 ¥ Mag. Stream 1200 deg2 ¥ Gal Plane : 500 deg2 ¥ Galactic center : 400 deg2 ¥ Resolution : 20Ó, 0.2 km/s ¥ 4400 hours of telescope time ¥ Raw data ~ 3 PB ¥ Final product : ~40 Gb per Þeld (data cube ~ 2000x2000x100) ¥ 11 days of computing time to image one Þeld on commercial supercomputer in Australia ¥ Total project : 60 Þelds, 6 Tb (science ready)
31. GASKAP Pilot survey blue : 20 hrs yellow : 10 hrs Kemp et al. (2025) 12 Þelds imaged so far
32. GASKAP : Small Magellanic Cloud at 21 cm Pingel et al. 2022
33. McClure-GrifÞths et al. 2018
34. McClure-GrifÞths et al. 2018 Bucklnd-Willis+(2025)
35. GASKAP | Mgellnic clouds log10(TB max) Dempsey et l. (in prep)
36. GASKAP | Mgellnic clouds log10(TB max) NHI(CNM) Dempsey et l. (in prep)
37. GASKAP data processing
38. 25 square degrees ASKAP field of view ASKAP Phased Array Feed ¥ Observations move over 3 positions every 10 minutes to provide more uniform sensitivity ¥ In reality a single observations contains 3*36=108 beams ¥ ASKAPSoft processing makes the imagining of every beam separately and then combined team linearly to make cubes. ¥ This does not work when the image contains structures at scales larger than the beam size. ¥ -> joint deconvolution
39. GASKAP workßow Pingel+2022 One GASKAP cube is 4096 x 4096 pixels x 2122 channels
40. ASKAP only Adding low spacings Pingel et al. (2022) ASKAP + Parkes ASKAP detects structures at scales < 30 arcsec
41. GASKAP workflow on DUG Kemp et al. (2025) 2000 nodes of Intel Xeon Phi Knights Landing processors, 64 cores, 128 GB RAM 8 nodes of Intel Cascade Lake processors, 24 cores, 192 GB RAM ~11 days ! it was 40 days on the ANU cluster Storage : 25 TB per cube
42. Challenges ahead
43. Release pilot survey LMC FFT + absorption line 30ÕÕ = 7pc at 50kpc Imaging artefacts GASKAP pipeline : going beyond the ASKASoft pipeline slides from Antoine Marchal (ANU)
44. Release pilot survey LMC FFT + absorption line 30ÕÕ = 7pc at 50kpc Imaging artefacts GASKAP pipeline : going beyond the ASKASoft pipeline slides from Antoine Marchal (ANU)
45. ˜ k (I′ k ; u, v) ≃ ∬ Ak (ℓ, m)I′ k (ℓ, m)e−2πi[uℓ + vm]dℓdm Framework: 
 inverse problem - regularized optimization 108 beams Challenge: 
 joint deconvolution with multiple SBIDs A non-linear optimization approach to joint deconvolution slides from Antoine Marchal (ANU)
46. ˜ k (I′ k ; u, v) ≃ ∬ Ak (ℓ, m)I′ k (ℓ, m)e−2πi[uℓ + vm]dℓdm Framework: 
 inverse problem - regularized optimization 108 beams Challenge: 
 joint deconvolution with multiple SBIDs A non-linear optimization approach to joint deconvolution Projected sky image slides from Antoine Marchal (ANU)
47. ˜ k (I′ k ; u, v) ≃ ∬ Ak (ℓ, m)I′ k (ℓ, m)e−2πi[uℓ + vm]dℓdm Framework: 
 inverse problem - regularized optimization 108 beams Challenge: 
 joint deconvolution with multiple SBIDs A non-linear optimization approach to joint deconvolution Projected sky image Primary beam (a-proj) slides from Antoine Marchal (ANU)
48. Ak (ℓ, m) × I′ k (ℓ, m) I(r) SIN projection A non-linear optimization approach to joint deconvolution slides from Antoine Marchal (ANU)
49. ˜ k (I′ k ; u, v) ≃ ∬ Ak (ℓ, m)I′ k (ℓ, m)e−2πi[uℓ + vm]dℓdm A non-linear optimization approach to joint deconvolution slides from Antoine Marchal (ANU)
50. ˜ k (I′ k ; u, v) ≃ ∬ Ak (ℓ, m)I′ k (ℓ, m)e−2πi[uℓ + vm]dℓdm NuFFT (no gridding); 
 similar to MPol 
 (by Ian Czekala, see Zawadzki et al. 2023) A non-linear optimization approach to joint deconvolution slides from Antoine Marchal (ANU)
51. A non-linear optimization approach to joint deconvolution slides from Antoine Marchal (ANU)
52. A non-linear optimization approach to joint deconvolution slides from Antoine Marchal (ANU)
53. WSClean + Parkes post process Imaged by Callum with NickÕs pipeline Joint deconvolution ASKAP + Parkes Adding the low spacings with Parkes data slides from Antoine Marchal (ANU)
54. Some technical details: - Each iteration takes 30s (CPU) or 3s on GPU V100
 For 3 SBIDs of visibilities (~160M) - It takes about 20 iter to get a good solution - No (explicit) gridding/de-gridding required - Small number of free parameters - No w-proj correction for now A non-linear optimization approach to joint deconvolution slides from Antoine Marchal (ANU)
55. A non-linear optimization approach to joint deconvolution slides from Antoine Marchal (ANU)
56. A non-linear optimization approach to joint deconvolution Imaged with 2 blocks ~15 hours slides from Antoine Marchal (ANU)
57. A non-linear optimization approach to joint deconvolution Imaged with 2 blocks ~15 hours slides from Antoine Marchal (ANU)
58. A non-linear optimization approach to joint deconvolution Imaged with 2 blocks ~15 hours We will image with 25 blocks ~200 hours slides from Antoine Marchal (ANU)
59. CNM - WNM absorption vs emission ¥ Joint Þt on a Gaussian basis ¥ Heiles & Troland (2003a,b) ¥ Stanimirovic & Heiles (2005) ¥ Begum+ (2010) ¥ Murray+ (2015, 2017)
60. Nguyen et al. (2024), Lynn et al. (2025) 21 cm absorption revolution 372 bsorption detection over the whole sky prior to tht McClure-Griffiths+(2023)
61. Nguyen et al. (2024), Lynn et al. (2025) 21 cm absorption revolution 372 bsorption detection over the whole sky prior to tht McClure-Griffiths+(2023)
62. Nguyen+2024 Vers ROHSA 2.0
63. Conclusion ¥ GASKAP : 21 cm observation of the Milky Way + Magellanic System ¥ 4400 hours of observations over 5 years (->2028). The largest HI data set ever produced ¥ Study from 0.01 pc to 100 kpc by combining nearby and distant systems ¥ Challenge in imaging of complex multi-scale diffuse emission over 3 orders of magnitude in scales ¥ Challenge in data analysis : data segmentation combining emission and absorption ¥ Great stepping stone for SKA
64. Bibliography ¥ Kemp et al. (2025), Processing of GASKAP-HI PILOT survey data using a commercial supercomputer ¥ Pingel et al. (2022), GASKAP-HI PILOT survey science 1; ASKAP Zoom observations of HI emission in the Smal Magellanic Cloud ¥ Marchal et al. (2019), ROHSA; Regularized Optmiziation for Hyper- Spectral Analysis - Application to phase separation of 21 cm data ¥ McClure-GrifÞths et al. (2019), Cold gas outßows from the Small Magellanic Cloud traced with ASKAP