Abstract

The Einstein constraint equations form a nonlinear and underdetermined elliptic system whose solution requires specifying appropriate gauge, conformal, and freely chosen geometric data. In this work, we present a hybrid numerical–machine learning framework for solving the Hamiltonian and momentum constraints using a combination of classical solvers, Physics-Informed Neural Networks (PINNs), and Deep Operator Networks (DeepONets). We clarify the mathematical structure of the constraint system, explicitly describe the conformal background and freely chosen components of the initial data, and construct a well-posed elliptic formulation suitable for numerical treatment. To demonstrate feasibility, we implement PINN and DeepONet models for the conformally flat, time-symmetric vacuum constraint and compare the neural solutions with a classical finite-difference reference. The results show that machine-assisted PDE solvers can approximate the constraint equations with competitive accuracy while offering mesh-free flexibility. This revised formulation provides a concrete basis for future large-scale simulations in numerical relativity.

References

• Bar-Sinai, Y., Hoyer, S., Hickey, J., & Brenner, M., 2019, "Learning data-driven discretizations for PDEs," PNAS, vol. 116, no. 31, pp. 15344–15349.
• Cao, S., Kovachki, N., Li, Z., & Stuart, A. M., 2023, "Recent advances in operator learning for scientific computing," SIAM Review, vol. 65, no. 4, pp. 1–40.
• Cai, S., Wang, X., & Li, J., 2022, "Physics-informed neural networks for PDE-constrained problems in science and engineering," Computational Mechanics, vol. 69, pp. 1019–1040.
• Gourgoulhon, E., 2012, 3+1 Formalism and Numerical Relativity, Springer.
• Karniadakis, G. E., Kevrekidis, I. G., Lu, L., Perdikaris, P., Wang, S., & Yang, L., 2021, "Physics-informed machine learning," Nature Reviews Physics, vol. 3, pp. 422–440.
• Kissas, G., Yang, L., Hwu, W., et al., 2024, "Machine learning in cardiovascular flows modeling: Physics-informed neural networks and beyond," Nature Machine Intelligence, vol. 6, pp. 1–13.
• Lu, L., Jin, P., Pang, G., Zhang, Z., & Karniadakis, G. E., 2021, "DeepONet: Learning nonlinear operators for solving differential equations," Nature Machine Intelligence, vol. 3, pp. 218–229.
• Lu, L., Jin, P., Pang, G., Zhang, Z., & Karniadakis, G. E., 2022, "A comprehensive and fair comparison of two neural operators (DeepONet vs. Fourier Neural Operator)," Computer Methods in Applied Mechanics and Engineering, vol. 393, 114778.
• Misner, C. W., Thorne, K. S., & Wheeler, J. A., 1973, Gravitation.
• Raissi, M., Perdikaris, P., & Karniadakis, G. E., 2019, "Physics-informed neural networks," Journal of Computational Physics, vol. 378, pp. 686–707.
• Sun, L., Gao, H., Pan, S., & Wang, J.-X., 2023, "Surrogate modeling for scientific computing using physics-informed deep learning," Physics of Fluids, vol. 35, 016105.
• Wang, S., Yu, X., & Perdikaris, P., 2022, "Understanding and mitigating gradient pathologies in physics-informed neural networks," SIAM Journal on Scientific Computing, vol. 44, no. 3, pp. A1717–A1743.
• Yang, L., Wang, S., & Perdikaris, P., 2023, "PINNs for high-dimensional PDEs," Journal of Scientific Computing, vol. 95, no. 2, pp. 1–28.
• Zhang, D., Guo, X., & Li, W., 2023, "A systematic survey of physics-informed neural networks: Methods and applications," Journal of Computational Physics, vol. 475, 111850.
• Zhang, R., Li, H., & Wang, J., 2024, "Applications of machine learning in numerical relativity," Classical and Quantum Gravity, vol. 41, 045001.