Exact Derivation of Physics Equations from Machine Learning Algorithms

Nieto-Chaupis H. · Proceedings - 2025 8th Asia Conference on Cognitive Engineering and Intelligent Interaction, CEII 2025 (2025) · DOI: 10.1109/ceii67787.2025.00021

Resumen

Commonly, laws and fundamental principles in physics are mostly expressed as differential equations, generally having derivatives up to second order. Because of this, one can wonder if exists, but still unknown, an universal generator of differential equations that fits extraordinarily with nature phenomena. This paper centers in concrete on the fact if such possible generator might be imitated by machine learning through computational operations like artificial neural networks. In this manner, starting from a basic topology, a double layer perceptron was designed. It would contain inputs as integer numbers denoting the order of differential equation. Although it is not clear the fundamental mechanism about how input emerges to controls and define the generation of physics equations, in this paper a special attention has been paid on the generation of wave, diffusion and Poisson equations. The case of Schrödinger equation has required a different approach. From the methods presented in this paper, it is suggested if exists a unique generator of physics equations then might be yielding unknown equations related also to unknown physics laws so far, for us.

Instituciones: Universidad Autónoma del Perú