Analysis of Binary O-type Star Systems Using Recurrent Neural Networks

Miguel Flores R.; Luis J. Corral; Celia R. Fierro-Santillán; Silvana G. Navarro

astro-ph.IMOct 16, 2026

Analysis of Binary O-type Star Systems Using Recurrent Neural Networks

Authors: Miguel Flores R., Luis J. Corral, Celia R. Fierro-Santillán, Silvana G. Navarro

Organizations: Centro Universitario de Ciencias Económico Administrativas (CUCEA), Universidad de Guadalajara, 45100 Zapopan, Jal., Mexico · Escuela Nacional Colegio de Ciencias y Humanidades, Plantel Sur (ENCCH Sur), Universidad Nacional Autónoma de México (UNAM), 04500, Mexico City, Mexico

Abstract

This work presents an initial analysis of the estimation of effective temperature and surface gravity for simulated O-type binary systems using a recurrent neural network. The proposed system uses 34 equivalent-width measurements from the optical region of a stellar spectrum. Two methods were applied to observed IACOB spectra before normalization: the addition of a blackbody continuum and the addition of a second observed spectrum. The first method produces only small differences in the normalized line profiles and in the estimated physical parameters over 3700-7000~Å. In the second method, the secondary spectrum is added with contributions of 5, 10, 20, 30, 50, and 75 per cent, as defined relative to the original primary spectrum. The general comparison includes 192 simulated systems and 1152 modified spectra. For most of these spectra, the estimated effective temperature and surface gravity decrease as the contribution of the secondary component increases. Individual examples using HD14947, HD190429, HD48099, and HD206267 show that this behavior is not the same for every combination: some estimated effective temperatures fall below the estimates for both isolated spectra, and some surface gravity estimates increase. Comparisons with previously published values also depend on the physical parameter and reference considered. The results show that the proposed system is sensitive to the addition of a secondary spectrum, but do not establish a unique contribution for the secondary component of an observed binary.

Figures & tables

Panel of Figure 1 — Figure 1: Input representations used to define the continuum experiment. The analysis is restricted to 3700–7000 Å.

Panel of Figure 2 — Figure 2: RNN estimates for the original observed spectra and for spectra modified with a blackbody continuum. The temperature experiment includes several continuum temperatures; the surface gravity panel shows the 20,000 K case.

Panel of Figure 3 — Figure 3: Estimated effective temperature (a) and surface gravity (b) for simulated binary spectra, compared with the estimates for the isolated primary spectra. Color indicates the secondary contribution $100\alpha$ ; marker size gives the absolute parameter difference. The line marks equality. The comparison contains 192 combinations at six contributions.

Panel of Figure 4 — Figure 4: Differences in estimated effective temperature (a) and surface gravity (b) between simulated binary spectra and isolated primary spectra, versus secondary contribution. Color indicates the parameter difference between the isolated components; marker size gives the absolute change relative to the primary. Negative differences correspond to lower estimates, following Equation 3 .

Panel of Figure 5 — Figure 5: Distributions of the difference in estimated effective temperature, $\Delta T_{\rm eff}$ , for the 1152 simulated binary spectra at each percentage contribution of the secondary spectrum. The lower panels separate the combinations according to the difference between the estimated effective temperatures of the isolated components.

Panel of Figure 6 — Figure 6: As in Figure 5 , but for the difference in estimated surface gravity, $\Delta\log g$ .

System	Primary	Secondary
HD190429	HD15570	HD36486
HD190429	HD190429	HD36486
HD48099	HD46150	HD214680
HD206267	HD199579	HD214680

Table 1: Observed spectra used to simulate the three binary systems. In each combination the primary spectrum is included at its original flux and the secondary spectrum is multiplied by

\alpha

. The second HD190429 combination adds a secondary spectrum directly to the observed spectrum of that system.

Figure 7: Estimated effective temperature and surface gravity for combinations of HD14947 and HD30614, with each star considered in turn as the primary. Panels (a,c) compare the estimates for the isolated observed spectra with the IACOB reference values ( Holgado et al., 2018 ) . Panels (b,d) show the estimates after adding 5, 10, 20, 30, 50, and 75 per cent of the secondary spectrum. The primary spectrum is included at its full original flux in both orders of addition.

Star	SpT/LC	$T_{\rm eff}$ (K)	$\log g$
HD24431	O9III	34,900	3.77
HD36486	O9.5IINwk	30,100	3.31

Table 2: Additional stars used as the secondary component in combinations with HD14947. Spectral type, luminosity class, effective temperature, and surface gravity are the IACOB reference values ( Holgado et al., 2018 ) . Surface gravity is expressed as

\log g

, using

g

\mathrm{cm\,s^{-2}}

Figure 8: Estimated effective temperature and surface gravity for simulated binary spectra using HD14947 as the primary and HD30614, HD24431, or HD36486 as the secondary. Panels (a,c) compare the estimates for the isolated observed spectra with the IACOB reference values ( Holgado et al., 2018 ) . Panels (b,d) show the estimates at the six percentages of secondary contribution. Each curve identifies the star used as the secondary component.

Figure 9: Estimated effective temperature and surface gravity for HD190429 and for simulations using HD15570 or the observed HD190429 spectrum as the primary, with HD36486 as the secondary. Panels (a,c) compare the estimates with previously published values ( Holgado et al., 2018 ; Repolust et al., 2004 ; Bouret et al., 2012 ; Martins et al., 2005 ) . The estimate from the observed HD190429 spectrum is repeated beside the primary-component references; the Repolust et al. (2004) reference is for HD66811. Panels (b,d) show the estimates at different percentages of contribution. Dashed lines indicate the estimates obtained from the observed spectrum of HD190429 without an added secondary spectrum.

Figure 10: Estimated effective temperature and surface gravity for HD48099 and for simulations using HD46150 as the primary component and HD214680 as the secondary component. Panels (a,c) compare estimates from the individual observed spectra with the reference values for HD48099 ( Solov’eva, 1986 ) and its components ( Mahy et al., 2010 ) . The estimates beside the component references correspond to HD46150 and HD214680. Panels (b,d) show estimates for the simulated binary spectra at different percentages of contribution. Dashed lines indicate the estimates obtained from the observed spectrum of HD48099.

Figure 11: Estimated effective temperature and surface gravity for HD206267 and for simulations using HD199579 as the primary component and HD214680 as the secondary component. Panels (a,c) compare the estimates from individual observed spectra with the component reference values from Raucq et al. (2018) . The estimates beside the component references correspond to HD199579 and HD214680. Panels (b,d) show estimates for the simulated binary spectra at different percentages of contribution. Dashed lines indicate the estimates obtained from the observed spectrum of HD206267.

Appendix figures & tables5 assets

Supplementary material from the paper’s appendix.

Appendix

Panel of Figure 12 — Figure 12: Observed spectra before and after adding a 20,000 K blackbody continuum and renormalizing the combined signal.

Panel of Figure 13 — Figure 13: Diagnostic lines for observed HD48099 and the HD46150+HD214680 simulation with a 5 per cent secondary contribution.

Panel of Figure 14 — Figure 14: Diagnostic lines for observed HD206267 and the HD199579+HD214680 simulation with a 10 per cent secondary contribution.

Panel of Figure 15 — Figure 15: Observed HD48099 compared with the HD46150+HD214680 simulation at 5 per cent secondary contribution.

Panel of Figure 16 — Figure 16: Observed HD206267 compared with the HD199579+HD214680 simulation at 10 per cent secondary contribution.

Astrophysics

See all 29 papers

Multi-Variable Stellar Parameter Estimation Using Residual Multitask Neural Networks
Jun 11, 2026Bruno Santos Meneses Barreto, Marcio EisencraftParameter Estimation Multi-Task Learning
Escola Polit´ecnica, Universidade de S˜ao Paulo, SP
SM-Net: Learning a Continuous Spectral Manifold from Multiple Stellar Libraries
Mar 25, 2026Omar Anwar, Aaron S. G. Robotham, Luca Cortese +1Astrophysics Neural Surrogate Modeling
International Centre for Radio Astronomy Research (ICRAR) The University of Western Australia (UWA), 35 Stirling Hwy, Crawley WA 6009, Australia
Bayesian classification of astronomical spectra with class uncertainties
Sep 18, 2026Simon Barton, Martin Sahlén, Andreas Korn +1Monte Carlo Dropout Multiclass Classification
Theoretical Astrophysics, Division of Astronomy and Space Physics, Department of Physics and Astronomy, Uppsala University, Box 524, 751 20 Uppsala, Sweden. · Observational Astrophysics, Division of Astronomy and Space Physics, Department of Physics and Astronomy, Uppsala University, Box 524, 751 20 Uppsala, Sweden · Division of High Energy Physics, Department of Physics and Astronomy, Uppsala University, Box 524, 751 20 Uppsala, Sweden +1

Jun 11, 2026J/K move · Enter open · S save

Multi-Variable Stellar Parameter Estimation Using Residual Multitask Neural Networks

Bruno Santos Meneses Barreto, Marcio Eisencraft

Escola Polit´ecnica, Universidade de S˜ao Paulo, SP

Parameter Estimation Multi-Task Learning

Open paper AskPDF arXiv

We present an end-to-end pipeline for estimating stellar parameters from Sloan Digital Sky Survey Data Release 12 spectra using a fully connected multitask neural network with residual blocks, whose hyperparameters are tuned via Bayesian optimization. The preprocessing pipeline includes per-spectrum standardization, RobustScaler normalization of the target variables -- effective temperature

T_{\mathrm{eff}}

, metallicity

[\mathrm{Fe/H}]

, and surface gravity

\log g

-- and data augmentation via Gaussian noise injection. On a held-out test set, the model achieved Mean Absolute Errors (MAE) of

59.76~\mathrm{K}

for

T_{\mathrm{eff}}

0.103~\mathrm{dex}

for

[\mathrm{Fe/H}]

, and

0.130~\mathrm{dex}

for

\log g

. Normalized against the full-scale range of each parameter, these results represent range-normalized errors between

1\%

and

3\%

, achieved with a highly efficient model complexity of approximately 540,000 trainable parameters. These results demonstrate that a compact residual multitask architecture, combined with principled signal preprocessing, provides a parameter-efficient solution for nonlinear parameter estimation in large-scale spectral datasets. In particular, the proposed model achieves competitive performance with substantially lower complexity than deeper neural network baselines.

Abstract

Astrophysics

Explore similar work

Multi-Variable Stellar Parameter Estimation Using Residual Multitask Neural Networks

SM-Net: Learning a Continuous Spectral Manifold from Multiple Stellar Libraries

Bayesian classification of astronomical spectra with class uncertainties