Agentic Symbolic Search: Characterizing PDEs Beyond Hand-crafted Expressions, Meshes, and Neural Networks

2026-06-18 · Source: Machine Learning · Field: Science & Research — Mathematics & Computational Sciences, Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

Agentic Symbolic Search (ASYS) is a novel, prior-guided framework designed to characterize Partial Differential Equation (PDE) solutions using mathematical structures, a capability not directly offered by numerical simulations or neural networks. ASYS employs an agent to translate PDE theory, problem constraints, and accumulated search experience into testable, differentiable symbolic programs. These mathematical forms are refined through evolutionary search, while their continuous parameters are optimized using gradient-based methods, effectively injecting inductive bias. The framework successfully recovers known analytical forms and constructs analytical approximations for problems lacking closed-form descriptions. Across experiments involving five diverse problems, including bounded dynamics and finite-time blow-up, ASYS generated interpretable representations, such as a geometric interface formula for Allen-Cahn 2D dynamics and a nine-parameter contraction law for Keller-Segel chemotactic blow-up, even where no prior closed-form solutions existed.

Key takeaway

For research scientists exploring complex Partial Differential Equations, ASYS offers a powerful new approach to derive interpretable mathematical structures. You should consider integrating agentic symbolic search into your workflow to move beyond purely numerical or neural network approximations. This method can help you discover novel analytical forms, like the nine-parameter contraction law for Keller-Segel, guiding deeper mathematical analysis where closed-form descriptions are currently unavailable.

Key insights

ASYS uses agentic symbolic search to discover interpretable mathematical structures for PDE solutions, surpassing traditional methods.

Principles

• PDE solutions are best understood via mathematical structures.
• Prior knowledge guides symbolic search effectively.
• Combine evolutionary search with gradient optimization.

Method

An agent translates PDE theory and constraints into differentiable symbolic programs. Evolutionary search refines forms, while gradient-based optimization fits continuous parameters, injecting inductive bias.

In practice

• Recover known analytical forms for PDEs.
• Construct analytical approximations for complex PDEs.
• Guide mathematicians toward further PDE analysis.

Topics

• Agentic Symbolic Search
• Partial Differential Equations
• Symbolic Regression
• Evolutionary Algorithms
• Gradient Optimization
• Mathematical Structures

Best for: AI Scientist, Research Scientist

Related on AIssential

• AutoNumerics: An Autonomous, PDE-Agnostic Multi-Agent Pipeline for Scientific Computing — AutoNumerics autonomously designs and verifies transparent PDE solvers from natural language using a multi-agent framework.
• Neuro-Symbolic ODE Discovery with Latent Grammar Flow — LGF discovers ODEs from data by embedding grammar-based equations into a structured latent space.
• How did Gpt solve the erdos problem? A demonstration — AI's problem-solving success stems from uncovering hidden algebraic structures, not just surface patterns.
• Towards Generalist Agents for Accelerating Scientific Discovery171 — LLMs can accelerate scientific discovery by generating and refining hypotheses within agentic, iterative workflows.
• Artificial Intelligence for Mathematical Reasoning: An Integrated Survey of Language Models, Neuro-symbolic Systems, and Verified Discovery — AI for mathematical reasoning is a rapidly evolving frontier, integrating diverse techniques for verified discovery.
• From Data to Theory: Autonomous Large Language Model Agents for Materials Science — Autonomous LLM agents can perform end-to-end scientific theory development, from equation selection to code execution and validation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.