J/A+A/692/A228 GES Catalogue HR10 & HR21 with cINN (Candebat+, 2024)
Inferring stellar parameters and their uncertainties from high-resolution
spectroscopy using invertible neural networks.
Candebat N., Sacco G.G., Magrini L., Belfiore F., Van der Swaelmen M.,
Zibetti S.
<Astron. Astrophys. 692, A228 (2024)>
=2024A&A...692A.228C 2024A&A...692A.228C (SIMBAD/NED BibCode)
ADC_Keywords: Surveys ; Stars, fundamental ; Abundances ;
Effective temperatures ; Optical ; Radial velocities ;
Keywords: methods: data analysis - methods: numerical -
techniques: spectroscopic - surveys - stars: abundances -
stars: fundamental parameters
Abstract:
New spectroscopic surveys will increase the number of astronomical
objects in need of characterisation by more than an order of
magnitude. Machine learning tools are required to address this data
deluge in a fast and accurate fashion. Most machine learning
algorithms cannot directly estimate error, making them unsuitable for
reliable science.We aim to train a supervised deep-learning algorithm
tailored for high-resolution observational stellar spectra. This
algorithm accurately infers precise estimates while providing coherent
estimates of uncertainties by leveraging information from both the
neural network and the spectra.We trained a conditional invertible
neural network (cINN) on observational spectroscopic data obtained
from the GIRAFFE spectrograph (HR10 and HR21 setups) within the
Gaia-ESO survey. A key feature of cINN is its ability to produce the
Bayesian posterior distribution of parameters for each spectrum. By
analysing this distribution, we inferred stellar parameters and their
corresponding uncertainties. We carried out several tests to
investigate how parameters are inferred and errors are estimated.We
achieved an accuracy of 28K in Teff, 0.06dex in logg, 0.03dex in
[Fe/H], and between 0.05dex and 0.17dex for the other abundances for
high-quality spectra. Accuracy remains stable with low signal-to-noise
ratio (between 5 and 25) spectra, with an accuracy of 39K in Teff,
0.08dex in logg, and 0.05dex in [Fe/H]. The uncertainties obtained
are well within the same order of magnitude. The network accurately
reproduces astrophysical relationships both on the scale of the Milky
Way and within smaller star clusters. We created a table containing
the new parameters generated by our cINN.This neural network
represents a compelling proposition for future astronomical surveys.
These derived uncertainties are coherent and can therefore be reused
in future works as Bayesian priors.
Description:
This analysis focuses on estimating the physical parameters of stars
observed with the GIRAFE instrument in the HR10 and HR21
configurations. The training data for this model comes from the
catalog published by Hourihane et al. (2023A&A...676A.129H 2023A&A...676A.129H, Cat.
J/A+A/676/A129), which provides labeled spectra for a large sample of
stars.
File Summary:
--------------------------------------------------------------------------------
FileName Lrecl Records Explanations
--------------------------------------------------------------------------------
ReadMe 80 . This file
catalog.dat 806 49858 GES 5.1 for HR10 & HR21 by OssicoNN
--------------------------------------------------------------------------------
See also:
J/A+A/676/A129 : Gaia-ESO catalogue version 5.1 (Hourihane+, 2023)
Byte-by-byte Description of file: catalog.dat
--------------------------------------------------------------------------------
Bytes Format Units Label Explanations
--------------------------------------------------------------------------------
1- 16 A16 --- Object GES object name from coordinates
18- 54 A37 --- Setup Grating setups used for analysis
56- 59 A4 --- RecWG Working group deriving the recommended
parameters
61- 72 A12 --- GESField GES field name from CASU
74- 94 E21.16 --- MEM3D ? Membership probability
96-116 F21.16 --- SNR SNR of the spectrum used for deriving
radial velocities (1)
118-138 F21.16 km/s RV ? Radial Velocity (1)
140-156 F17.12 K Teff Effective Temperature (1)
158-176 F19.15 K e_Teff Error on Teff (1)
178-195 F18.16 [cm/s2] logg Log Surface Gravity (1)
197-214 F18.16 [cm/s2] e_logg Error on logg (1)
216-234 F19.16 --- [Fe/H] ? Metallicity (1)
236-253 F18.16 --- e_[Fe/H] ? Error on [Fe/H] (1)
255-272 F18.16 --- Al1 ? Neutral Aluminium abundance (1)
274-291 F18.16 --- e_Al1 ? Error on AL1
293-311 F19.16 --- Mg1 ? Neutral Magnesium abundance (1)
313-330 F18.16 --- e_Mg1 ? Error on MG1
332-349 F18.16 --- Ca1 ? Neutral Calcium abundance (1)
351-368 F18.16 --- e_Ca1 ? Error on CA1
370-387 F18.16 --- Ti1 ? Neutral Titanium abundance (1)
389-407 F19.16 --- e_Ti1 ? Error on TI1
409-427 F19.16 --- Si1 ? Neutral Silicon abundance (1)
429-447 F19.16 --- e_Si1 ? Error on SI1
449-466 F18.16 --- Ni1 ? Neutral Nickel abundance (1)
468-485 F18.16 --- e_Ni1 ? Error on NI1
487-496 A10 --- Dataset In which dataset the star was used
498-504 I7 K TeffNN Effective Temperature (2)
506-511 I6 K e_TeffNN Error on Teff (2)
513-517 I5 K e1_TeffNN Internal error on Teff (2)
519-524 I6 K e2_TeffNN External error on Teff (2)
526-532 F7.3 [cm/s2] loggNN Log Surface Gravity (2)
534-540 F7.3 [cm/s2] e_loggNN Error on logg (2)
542-548 F7.3 [cm/s2] e1_loggNN Internal error on logg (2)
550-556 F7.3 [cm/s2] e2_loggNN External error on logg (2)
558-565 F8.3 --- [Fe/H]NN Metallicity (2)
567-575 F9.3 --- e_[Fe/H]NN Error on [Fe/H] (2)
577-583 F7.3 --- e1_[Fe/H]NN Internal error on [Fe/H] (2)
585-593 F9.3 --- e2_[Fe/H]NN External error on [Fe/H] (2)
595-601 F7.3 --- Al1NN Neutral Aluminium abundance (2)
603-609 F7.3 --- e_Al1NN Error on AL1NN
611-617 F7.3 --- e1_Al1NN Internal error on AL1NN
619-625 F7.3 --- e2_Al1NN External error on AL1NN
627-635 F9.3 --- Mg1NN Neutral Magnesium abundance (2)
637-645 F9.3 --- e_Mg1NN Error on MG1NN
647-654 F8.3 --- e1_Mg1NN Internal error on MG1NN
656-664 F9.3 --- e2_Mg1NN External error on MG1NN
666-672 F7.3 --- Ca1NN Neutral Calcium abundance (2)
674-681 F8.3 --- e_Ca1NN Error on CA1NN
683-689 F7.3 --- e1_Ca1NN Internal error on CA1NN
691-698 F8.3 --- e2_Ca1NN External error on CA1NN
700-706 F7.3 --- Ni1NN Neutral Nickel abundance (2)
708-716 F9.3 --- e_Ni1NN Error on NI1NN
718-724 F7.3 --- e1_Ni1NN Internal error on NI1NN
726-734 F9.3 --- e2_Ni1NN External error on NI1NN
736-743 F8.3 --- Ti1NN Neutral Titanium abundance (2)
745-752 F8.3 --- e_Ti1NN Error on TI1NN
754-761 F8.3 --- e1_Ti1NN Internal error on TI1NN
763-770 F8.3 --- e2_Ti1NN External error on TI1NN
772-779 F8.3 --- Si1NN Neutral Silicon abundance (2)
781-788 F8.3 --- e_Si1NN Error on SI1NN
790-797 F8.3 --- e1_Si1NN Internal error on SI1NN
799-806 F8.3 --- e2_Si1NN External error on SI1NN
--------------------------------------------------------------------------------
Note (1): computed by Hourihane et al., 2023A&A...676A.129H 2023A&A...676A.129H, Cat. J/A+A/676/A129
Note (2): computed by OssicoNN
--------------------------------------------------------------------------------
Acknowledgements:
Nils Candebat, nils.candebat(at)inaf.it
(End) Patricia Vannier [CDS] 04-Nov-2024