Speaker
Description
The determination of stellar and orbital parameters in massive binaries is often complicated by phase-dependent variability in the line profiles. Classical spectral disentangling represents time-series spectra as contributions from Doppler-shifted, phase-invariant components [1, 2]. However, this approximation can be inadequate in systems affected by tidal deformation, gravity darkening, pulsations, winds or extreme light ratios. In such cases, atmospheric, orbital, and variability signatures become coupled, creating an identifiability issue in which non-rigid profile changes may be misinterpreted as orbital motion or as incorrect stellar parameters.
We present SVT-PINN, a physics-informed framework currently under development, designed to separate these contributions through an ordered scalar-vector-tensor inference scheme. The scalar mode uses a library of 33,075 ATLAS9 and TLUSTY synthetic models, parameterised by effective temperature, surface gravity, metallicity, and projected rotational velocity, to constrain phase-independent component spectra. Given this atmospheric solution, the vector mode infers coherent orbital Doppler transport in logarithmic wavelength (x = log λ) and Fourier space using Markov chain Monte Carlo sampling. The Tensor mode is then introduced to model the remaining phase-dependent deformation, τ_a(x, ϕ), using a physics-informed neural network [3]. Spectral smoothness, phase coherence, amplitude control and geometry-dependent penalties are incorporated to limit this flexible mode's ability to absorb errors from the atmospheric or orbital solutions.
The framework is designed so that its atmospheric, orbital and variability components can be inspected separately. It reduces to classical rigid-profile disentangling when the tensor contribution vanishes. The dimensionless ratios R_(VS) and R_(TS) are introduced to characterise orbital transport and phase-dependent deformation relative to the phase-independent spectral baseline. Multi-epoch, high-resolution spectroscopy of HD 2913, HD 199892 and HD 138527, supported by space-based photometry, will provide the first observational stress tests. The present contribution describes the physical formulation, ordered inference strategy, diagnostic quantities, and validation plan for SVT-PINN, a developing, variability-aware extension of spectral disentangling.
Acknowledgements
This work was supported by the Slovak Research and Development Agency under contract No. APVV-24-0160, and
by the VEGA grant No. 2/0033/26 from the Slovak Academy of Sciences.
References
[1] P. Hadrava, Astrophysics and Space Science, vol. 304, no. 1-4, pp. 337-339, Aug. 2006. doi: 10.1007/s10509-006-9153-5.
[2] K. Pavlovski and H. Hensberge, Astronomy and Astrophysics, vol. 439, no. 1, pp. 309-321, Aug. 2005. doi: 10.1051/0004-6361:20042139.
[3] M. Raissi, P. Perdikaris, and G. E. Karniadakis, Journal of Computational Physics, vol. 378, pp. 686-707, Feb. 2019. doi: 10.1016/j.jcp.2018.10.045.
