Learning Schwarzschild Geodesic Motion with Lagrangian Neural Networks
Abstract
Lagrangian Neural Networks (LNNs) learn dynamics from trajectory data by parameterizing a scalar Lagrangian and deriving accelerations through the Euler--Lagrange equation, providing a variational inductive bias that improves physical consistency over purely data-driven models. LNNs have recently been extended to general relativistic settings, leaving Schwarzschild geodesics as an open challenge. Training on complex relativistic systems is hindered by the diversity of the dynamics, as well as numerical instabilities and the difficulty of adequately sampling physically relevant, diverse orbital regimes. We introduce two coordinated improvements: a decoupled position--velocity normalization scheme that independently rescales each coordinate and its time derivatives, reweighting acceleration targets to balance contributions across all channels; and an energy-constrained log-uniformly sampled dataset generation strategy informed by the innermost stable circular orbit (ISCO) structure. With these improvements, our model accurately reproduces perihelion precession and relativistic time dilation, conserves energy and angular momentum to within a few percent even after completing multiple orbits around the central mass and predicts orbital trajectories across eight orbital scenarios in close, medium, and far-field regimes. These results establish that LNNs, augmented with physically motivated normalization and sampling, can reliably learn geodesic dynamics in curved spacetime. Our work also represents the first successful training of LNNs on Schwarzschild geodesics and offers a promising direction for data-driven modeling of complex relativistic systems.
Related
- Analytic Electromagnetic Effective Source for Geodesic Orbit in Schwarzschild Spacetime
- A Geodesic Route toward Holography beyond AdS: Tessellating the Schwarzschild Black Hole
- Stochastic motion of massless particles near event horizons in Schwarzschild and Reissner-Nordström spacetimes
- Lagrangian and Hamiltonian Neural Networks With a Dissipative System
- Recovering Irreducible Three-Body Interactions with Lagrangian Neural Networks
BibTeX
@misc{caglar2026learning,
title = {{Learning Schwarzschild Geodesic Motion with Lagrangian Neural Networks}},
author = {Sukru Caglar and Sena Kalabalik and Abdullah Umut Hamzaogullari and Arkadas Ozakin},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.13234},
eprint = {2610.13234},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.13234},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Learning Schwarzschild Geodesic Motion with Lagrangian Neural Networks
AU - Sukru Caglar
AU - Sena Kalabalik
AU - Abdullah Umut Hamzaogullari
AU - Arkadas Ozakin
PY - 2026
DA - 2026/10/09
PB - arXiv
DO - 10.48550/arxiv.2610.13234
UR - https://arxiv.org/abs/2610.13234
AB - Lagrangian Neural Networks (LNNs) learn dynamics from trajectory data by parameterizing a scalar Lagrangian and deriving accelerations through the Euler--Lagrange equation, providing a variational inductive bias that improves physical consistency over purely data-driven models. LNNs have recently been extended to general relativistic settings, leaving Schwarzschild geodesics as an open challenge. Training on complex relativistic systems is hindered by the diversity of the dynamics, as well as numerical instabilities and the difficulty of adequately sampling physically relevant, diverse orbital regimes. We introduce two coordinated improvements: a decoupled position--velocity normalization scheme that independently rescales each coordinate and its time derivatives, reweighting acceleration targets to balance contributions across all channels; and an energy-constrained log-uniformly sampled dataset generation strategy informed by the innermost stable circular orbit (ISCO) structure. With these improvements, our model accurately reproduces perihelion precession and relativistic time dilation, conserves energy and angular momentum to within a few percent even after completing multiple orbits around the central mass and predicts orbital trajectories across eight orbital scenarios in close, medium, and far-field regimes. These results establish that LNNs, augmented with physically motivated normalization and sampling, can reliably learn geodesic dynamics in curved spacetime. Our work also represents the first successful training of LNNs on Schwarzschild geodesics and offers a promising direction for data-driven modeling of complex relativistic systems.
ER -