Abstract:We consider the problem of learning an interpretable potential energy function from a Hamiltonian system's trajectories. We address this problem for classical, separable Hamiltonian systems. Our approach first constructs a neural network model of the potential and then applies an equation discovery technique to extract from the neural potential a closed-form algebraic expression. We demonstrate this approach for several systems, including oscillators, a central force problem, and a problem of two charged particles in a classical Coulomb potential. Through these test problems, we show close agreement between learned neural potentials, the interpreted potentials we obtain after training, and the ground truth. In particular, for the central force problem, we show that our approach learns the correct effective potential, a reduced-order model of the system.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is **learning and interpreting an interpretable potential energy function from the trajectory data of classical Hamiltonian systems**. Specifically, the author focuses on how to identify the potential energy \( V(q) \) of the system from the measured trajectory data of classical Hamiltonian systems, and then construct an interpretable, closed - form algebraic expression to represent this potential energy. ### Main Problem Description 1. **Problem Background**: - The paper considers the problem of learning an interpretable potential energy function \( V(q) \) from the trajectories of classical, separable Hamiltonian systems. - The form of the Hamiltonian system is \( H(p, q)=T(p)+V(q) \), where \( T(p) \) is the known kinetic energy and \( V(q) \) is the unknown potential energy. 2. **Research Objectives**: - Learn the potential energy function \( V(q) \) through a neural network model. - Use equation discovery techniques (such as the SINDy algorithm) to extract a closed - form algebraic expression from the trained neural network potential. - Verify the degree of fit between the learned potential energy function and the real potential energy, and demonstrate its application effects in different systems. 3. **Specific Application Scenarios**: - The paper shows the application of this method in multiple systems, including simple harmonic oscillators, central force problems, and the interaction between two charged particles under the classical Coulomb potential, etc. ### Solution Overview - **Neural Network Modeling**: Use a neural network to parameterize the unknown potential energy function \( V(q) \) and train it with the given trajectory data. - **Equation Discovery**: Utilize methods such as SINDy (Sparse Identification of Nonlinear Dynamics) to convert the trained neural network potential into a closed - form algebraic expression, thereby achieving explanatory modeling of the system. - **Verification and Testing**: Verify the effectiveness of the method through multiple test cases (such as simple harmonic oscillators, double - well potentials, central force problems, and Coulomb potentials), and demonstrate the high consistency between the learned potential and the real potential. ### Key Contributions - Propose a method for learning and interpreting potential energy functions from trajectory data. - Demonstrate the application of this method in multiple classical physical systems and verify its effectiveness and accuracy. - Emphasize that the potential energy function obtained through machine - learning methods can be used not only for simulation but also for providing physical explanations. Through this method, researchers can extract simple and physically meaningful potential energy expressions from the trajectory data of complex systems, providing a powerful tool for further understanding the dynamic behaviors of these systems.

Learning and Interpreting Potentials for Classical Hamiltonian Systems

Neural Hamilton: Can A.I. Understand Hamiltonian Mechanics?

Learning Trajectories of Hamiltonian Systems with Neural Networks

Physics-Informed Learning Using Hamiltonian Neural Networks with Output Error Noise Models

Learning Generalized Hamiltonians using fully Symplectic Mappings

Learning Hamiltonian Systems with Mono-Implicit Runge-Kutta Methods

Applications of Machine Learning to Modelling and Analysing Dynamical Systems

Symplectic Learning for Hamiltonian Neural Networks

Learning Neural Hamiltonian Dynamics: A Methodological Overview

Solving The Quantum Many-Body Hamiltonian Learning Problem with Neural Differential Equations

Hamiltonian and Liouvillian learning in weakly-dissipative quantum many-body systems

Discovering Symbolic Laws Directly from Trajectories with Hamiltonian Graph Neural Networks

Practical and Efficient Hamiltonian Learning

Scalable learning of potentials to predict time-dependent Hartree-Fock dynamics

Hamiltonian neural networks for solving equations of motion

Exact potentials in multivariate Langevin equations

Learning Subsystem Dynamics in Nonlinear Systems via Port-Hamiltonian Neural Networks

Physics-informed neural networks via stochastic Hamiltonian dynamics learning

Adaptable Hamiltonian neural networks

Tutorial: How to Train a Neural Network Potential