GT ICR - Florent Mertens

Transcript

1. 1 Unveiling the Physics of the EoR and Cosmic Dawn with LOFAR and NenUFAR Florent Mertens (LUX, Paris Observatory) GT ICR @ L2S - 19/05/2025 Image credit: Michale Goh/ICRAR-Curtin
2. 2 Cosmic Dawn ➔ Appearance of first stars/Bhs (PopIII?) ➔ Ly-α radiation field ➔ Impact of Baryonic Bulk Flows ➔ First X-ray heating sources Epoch of Reionization ➔ Reionization by stars & mini-quasars ➔ PopIII - PopII transition ➔ Emergence of the visible universe PLANCK VLT E-ELT Credit: NAOJ Cosmic Dawn Epoch of Reionization The history of our universe
3. 3 S Epoch of Reionization Cosmic Dawn Credit: NAOJ LOFAR NenuFAR 21-cm HI line, probe of the first billion years
4. 4 The 21-cm experiments LOFAR-EoR Observation started in 2012 + 3500 h observed ➔ Properties of the IGM and ionising sources. ➔ History of reionization. NenuFAR Cosmic Dawn Observation started in 2019 + 1000 h observed ➔ Testing of non-standard models. SKA CD/EoR Completion ~2028 ➔ Nature of the first stars. ➔ Morphology of ionized regions. Bowman et al. 2018 Simulation 21-cm du LERMA LOFAR NenuFAR
5. 5 A challenging experiment Foregrounds emission 21-cm signal Extra-galactic sources Galactic emission Processing pipeline Observation
6. 6 The Foregrounds radio sources Hardcastle et al. 2020 De Gasperin et al. 2020 Cygnus A Cassiopeia A Taurus A Virgo A Pictor A Fornax A
7. 7 The challenge of the foregrounds Extragalactic point sources Galactic diffuse emission Instrumental effects Foregrounds 21-cm signal
8. 8 Signature of the foregrounds 21-cm signal Astrophysical Foregrounds  Extragalactic sources  Galactic diffuse emission Instrumental effects  PSF/beam chromaticity  Calibration errors
9. 9 Mitigation strategies A. Observing field selection B. Extragalactic sources subtraction C. Residual foregrounds subtraction D. Foregrounds avoidance B + D: MWA B + C: LOFAR, AARTFAAC, NenuFAR D: HERA
10. 10 Selecting the right observing field Barry, Bernardi, Greig, Kern, Mertens 2021 Criteria (tracking experiments): ➔ Low galactic emission ➔ No strong sources in the field ➔ No strong sources in sidelobes ➔ Calibrator in the field (?) ➔ High elevation (dec ~ latitude) Synergies with other surveys For drift-scan experiments: ➔ LST bins selection HERA MWA LOFAR
11. 11 Avoid strong sources in sidelobes Strong radio sources in the sky may pass through the side lobes and grating lobe of the instrument: introduce strong direction-dependent effect Cygnus A Taurus A Cas A 45 MHz 60 MHz 75 MHz North Celestial Pole Observation with NenuFAR. Beam simulated with nenupy (A. Loh)
12. 12 Elevation of the target matters Foreground wedge extend depends on target elevation: observing at higher elevation helps in mitigating foreground emission NCP Zenith crossing Munshi, Mertens et al 2025a
13. 13 Impact of RFI depends on target field Satellite image of spot A Satellite image of spot B Spot A: delay and Fringe Rate power spectra for an example baseline Foregrounds Near-field image 12/12/2021 61 –72 MHz | NCP | A-team masked Spot A Spot B
14. 14 Spectral characterization of the local sources of RFI Impact of RFI depends on target field Munshi, Mertens et al 2025b
15. 15 Confusion limited foregrounds + low level residuals Inner 4° x 4° where we look for the signal Before After Point-source subtraction ➔Need accurate sky-model ➔Solve for instruments gains in direction of sources LOFAR | NCP | 140 hours | 134-146 MHz Mertens et al. 2020
16. 16 Point-source subtraction NenuFAR | NCP | 11 hours | 61-72 MHz Cygnus A Cassiopeia A Taurus A Virgo A Munshi, Mertens et al. 2024 ➔Need accurate sky-model ➔Solve for instruments gains in direction of sources
17. 17 Gaussian Process Regression After point-source subtraction, residual foregrounds still dominates ➔ GPR: Fits data without assuming a specific functional form. ➔ Prior Information: Encoded through a parametrized covariance function. ➔ Parameters Optimization: Covariance parameters are determined by maximizing the marginal likelihood. ➔ Data fitting: Conditioning the prior model to the data, we obtain fit + uncertainty.
18. 18 GPR for 21-cm experiments Extra-galactic Galactic emission Instrumental effects No functional forms but very different spectral characteristic → Statistical model prior made of Gaussian Process (GP). → Learnt kernel is used for the 21-cm prior covariance. 21-cm signal Hyper-parameters of the covariance prior to be optimized with the data Intrinsic FG Mode-mixing FG 21-cm Coherence-scale V a r i a n c e Frequency slice Freq-freq covariance Mertens et al. 2018 Mertens, Bobin, Carucci 2024
19. 19 Learned covariance function VAE: Trained to minimize: ➔ Reconstruction error. KL divergence to standard Gaussian in latent space. ✔ Compressed information (lower dimension latent space). ✔ Generative. 21-cm covariance Latent-space parameters Input Output Encoder Decoder Latent space Variational Auto-Encoder (VAE) 21cmFast Training set Decoder PS to cov Mertens, Bobin, Carucci 2024
20. 20 ML-GPR SKA simulation (100 hrs) Mertens, Bobin, Carucci 2024
21. 21 The Epoch of Reionization with LOFAR The LOFAR-EoR team The LOFAR telescope
22. 22 Improved foregrounds removal Using ML-enhanced Gaussian Process Regression (Mertens et al. 2024)
23. 23 Deeper multi-redshift LOFAR upper limits A reduction in upper limits by a factor ~5 to 10 for the three LOFAR redshifts (Mertens, et al. 2025) (54 mK)2 (69 mK)2 (66 mK)2
24. 24 New LOFAR constraints on EoR IGM Preliminary IGM parameter constraint analysis allows us to reject many scenarii of cold IGM (Ghara, et al. 2025)
25. 25 The Cosmic Dawn with NenuFAR The NenuFAR Cosmic Dawn team The NenuFAR telescope
26. 26 First NenuFAR upper limit NCP field, 11.5 hours, 61-72.5 MHz, z ~ 20 (Munshi, Mertens et al. 2024) The deepest upper limit on the 21-cm power-spectra at z~20 ... but limited by bright source contamination and local RFI
27. 27 NenuFAR search for a “darker” deep field 2022 – 2023 observation campaign on 5 candidates deep fields Selection strategy: ➔ Minimize apparent flux from Bright A-team sources. ➔ Transit close to zenith for maximum sensitivity. Mertens et al. in prep.
28. 28 New NenuFAR upper limit NT04 field, 26.1 hours, z ~ 20 & z~17 (Munshi, Mertens et al. in prep.) An improvement by a factor 50 !
29. 29 Progress toward a detection Exotic signal (« Edges »)
30. 30 The goal: recover the power-spectra of a 21-cm signal from a simulated SKA data cubes The DOTSS-21 team : 24 members (FR : 10, NL : 11) Our approach: Build on the LOFAR-EoR and NenuFAR Cosmic Dawn experience. The SKA Data Challenge 3a Bonaldi et al. 2025
31. 31 DOTSS-21 results @ z~ 7.2 Step 1: Detect and subtract compact sources Step 2: Model and subtract the Galactic diffuse emission Step 3: Extract the 21-cm signal with ML- GPR (Mertens et al. 2018, 2024)
32. 32 Bright point source removal
33. 33 Faint point source removal
34. 34 Step 1: Detect and subtract compact sources Step 2: Model and subtract the Galactic diffuse emission Step 3: Extract the 21-cm signal with ML- GPR (Mertens et al. 2018, 2024) DOTSS-21 results @ z~ 7.2
35. 35 Step 1: Detect and subtract compact sources Step 2: Model and subtract the Galactic diffuse emission Step 3: Extract the 21-cm signal with ML-GPR (Mertens et al. 2018, 2024) DOTSS-21 results @ z~ 7.2
36. 36 Step 1: Detect and subtract compact sources Step 2: Model and subtract the Galactic diffuse emission Step 3: Extract the 21-cm signal with ML-GPR (Mertens et al. 2018, 2024) DOTSS-21 results @ z~ 7.2
37. 37 DOTSS-21 results Results can be improved further • Issues with fast evolution over the 15 MHz • Recovered image cube not perfect
38. 38 Summary ➔ The 21-cm signal from the Cosmic Dawn and EoR promises a new and unique probe of the first billion year of the Universe, but very challenging experiment ➔ Status of the LOFAR-EoR project:  New multi-redshift upper-limits at z=8.3, 9.1 and 10.1  Deepest @ k=0.075 cMpc-1, z ~ 9: Δ2 < (54 mK)2  Only 5% of data processed ! ➔ Status of the NenuFAR Cosmic Dawn project:  First upper limit at z ~ 20 published from NCP deep field  New “darker” deep field (NT04) promises exiting results !  Deepest @ k=0.04 cMpc-1, z ~ 20: Δ2 < (1414 mK)2 ➔ We are scaling-up processing significantly (LOFAR and NenuFAR) ! ➔ Our DOTSS-21 team won the SKA Data Challenge 3a !
39. 39 Foregrounds avoidance vs removal LOFAR SKA HERA MWA