GRAVI-NEURAL: Covariant Neural Characterization of Metric Tensor Perturbations in Dynamic Gravitational Environments

GRAVI-NEURAL is a physics-informed artificial intelligence framework that introduces a Covariant Neural Operator (CNO) for learning, approximating, and evolving solutions to the Einstein Field Equations (EFE) under dynamic and strong-field gravitational conditions. The system decomposes the spacetime metric into a Minkowski background and a learned neural perturbation field, enabling real-time prediction of curvature dynamics, gravitational waveforms, and geodesic trajectories. GRAVI-NEURAL enforces physical consistency through Hamiltonian constraints and Bianchi identity preservation, ensuring energy-momentum conservation across all predictions. The architecture integrates multiple neural components to achieve coordinate-independent spacetime modeling and high-precision gravitational inference. This project is part of the EntropyLab research program and aims to replace computationally intensive numerical relativity solvers with efficient, scalable, and physically consistent neural approximations applicable to gravitational wave astronomy, autonomous space navigation, and planetary geophysics. DOI: 10.5281/zenodo.19871822 Version: v1.0.0 Year: 2026 License: MIT License