Hybrid Physical-Neural Simulator for Fast Cosmological Hydrodynamics

2025-10-30

AI Summary

Efficient, differentiable modeling of coupled hydrodynamic and gravitational interactions between baryonic gas and dark matter remains challenging in cosmological simulations. Method: We propose a hybrid physics–neural network forward model: a differentiable particle-mesh (PM) solver computes gravity precisely, while a neural-network-parameterized effective pressure field replaces conventional fluid closure assumptions. The architecture is fully end-to-end differentiable and trained on a single high-fidelity reference simulation, enabling high data efficiency and direct fitting to observational data. Contribution/Results: Our model outperforms baselines—including enthalpy gradient descent—in reconstructing density fields and reproducing statistical measures such as the power spectrum, achieving superior accuracy and computational speed. It constitutes the first differentiable hydrodynamical framework for cosmological inference that simultaneously ensures field-level fidelity and end-to-end differentiability, enabling robust, observation-driven cosmological parameter estimation and structure formation modeling.

Technology Category

• Machine Learning: Deep Neural Architectures and Foundation Models
• Computer Vision: Diffusion Models for Vision
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Cosmological field-level inference requires differentiable forward models that solve the challenging dynamics of gas and dark matter under hydrodynamics and gravity. We propose a hybrid approach where gravitational forces are computed using a differentiable particle-mesh solver, while the hydrodynamics are parametrized by a neural network that maps local quantities to an effective pressure field. We demonstrate that our method improves upon alternative approaches, such as an Enthalpy Gradient Descent baseline, both at the field and summary-statistic level. The approach is furthermore highly data efficient, with a single reference simulation of cosmological structure formation being sufficient to constrain the neural pressure model. This opens the door for future applications where the model is fit directly to observational data, rather than a training set of simulations.

Problem

• Develops hybrid simulator combining neural networks with physical solvers
• Models cosmological gas dynamics using learnable pressure parametrization
• Enables efficient field-level inference from observational data directly

Innovation

• Hybrid simulator combines differentiable particle-mesh gravity solver
• Neural network parametrizes hydrodynamics via pressure field mapping
• Data-efficient method trains with single reference simulation