arXiv Science⌕ Search

arXiv · 2611.12401

Data-Driven Conservation Laws for Newtonian Systems: Physics-Informed Neural Networks and Eisenhart-Duval Geometry

Abstract

We introduce a Physics-Informed Neural Network approach for the data-driven reconstruction of the symmetries and conservation laws of Newtonian dynamical systems, based on the geometrization of the dynamics through the Eisenhart-Duval lift. Within the Eisenhart-Duval lift, the symmetry conditions reduce to the linear Killing or Conformal Killing equations of the extended metric. We combine a Hamiltonian neural network, which reconstructs the potential function from the data, with a random-feature neural network, which solves the symmetry conditions for the collineations of the extended geometry. From the learned vector fields we construct the Lie algebra and, through Noether's theorem, the corresponding conserved quantities, while the derivation of the curvature invariants for the extended geometry provide allow us to perform an independent consistency test. We apply the method to the one-dimensional Ermakov-Pinney equation and to the one-dimensional harmonic oscillator, with and without a quartic perturbation, by using different level of noise in the phase-space data. For the Ermakov-Pinney equation, the three Killing vectors of the flat extended plane, and the corresponding nonlocal conservation laws, are recovered for, while for the harmonic oscillator, the ten conformal Killing vectors which form the $so(3,2)$ Lie algebra are determined for noise levels up to $10\%$. When the quartic term is added, the broken symmetries are identified as approximate symmetries.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Andronikos Paliathanasis. 2026-11-09. Data-Driven Conservation Laws for Newtonian Systems: Physics-Informed Neural Networks and Eisenhart-Duval Geometry. https://arxiv.science/abs/2611.12401

Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite arXiv Science by Indiciae: BibTeX · RIS.

KEEP EXPLORING

Related papers

How to calculate temperature and energy in Molecular Dynamics

Molecular Dynamics (MD) simulation of the dynamics of classical objects has become incredibly popular in natural sciences since the method was developed in the sixties. A conservative estimate is that it involves many hundreds of thousands of scientific articles and all subdisciplines of natural sciences. The earliest MD simulations also introduced a simple time-symmetric expression for the momentum and velocity, which has since been used uncritically in almost all MD simulations. The expression is incorrect and can lead to significant errors. But it is straightforward to replace the expression with the correct one, and the motive of the article is to draw attention to how to calculate velocity, kinetic energy, temperature, and energy in an MD simulation correctly. Newton's algorithm of the discrete dynamics has the same quality as his analytical dynamics, and Newton's discrete dynamics conserves the energy exactly. That is the second motive for the article, namely to show that conservation of energy in a conservative system is only obtained by the correct expressions for the velocity and kinetic energy.

physics.class-ph↗

Covariance Cells And Star Products

In classical statistics, a localized preparation is described not only by a point in phase space but also by a covariance cell specifying its finite resolution. We study balanced Gaussian cells on a linear symplectic space and show that they are equivalent to compatible complex structures and positive Lagrangian polarizations. Their parameter space is the Siegel symmetric space. Using the Koopman-von Neumann representation of classical mechanics, we describe the linearized transport of Gaussian amplitudes along a reference motion. Each covariance cell determines a separated star product whose antisymmetric part is fixed by the symplectic form and whose symmetric part is fixed by the covariance. This product is related to the Moyal product by Gaussian smoothing. Quadratic Hamiltonians act through their classical Poisson derivations within a fixed cell algebra exactly when they preserve the cell; squeezing transformations instead relate the algebras associated with different cells. Fronsdal's star-polarization construction yields a module whose cyclic element is the Gaussian density of the cell. Its completion is the corresponding Bargmann-Fock space, naturally embedded in the space of classical Koopman amplitudes. Koopman transport carries this Fock space to that of the transported cell. No Planck constant is introduced: the deformation parameter is a classical resolution scale, and physical observables remain commuting multiplication operators.

physics.class-ph↗

Revisiting Stress Analysis in a Three-Dimensional Elastic Hollow Sphere under Uniaxial Compression via the Inverse Laplace Transform Expressions within an Elastodynamic Framework

The stress analysis of a three-dimensional elastic hollow sphere subjected to uniaxial compression is revisited, employing an elastodynamic framework. Through the application of the Laplace transform, the scalar and vector potentials of displacement are expanded, facilitating a detailed exploration of the system's mechanical behavior. The static solutions for displacement and stress distributions are derived in the long-time limit, which reveal key insights into the response of the elastic hollow sphere. Notably, on the inner surface, certain quantities exhibit a peak at the point where the angle between the compressive force and the point on the surface becomes perpendicular, indicating localized stress concentration. These findings provide a robust analytical approach for understanding and predicting the behavior of elastic hollow spheres under uniaxial loading, with implications for material science and structural engineering.

physics.class-ph↗