Discovering Governing Partial Differential Equations Using Multi-Source Data: A New Competitive Optimization Framework

Researchers have developed a novel framework for discovering governing partial differential equations (PDEs) from multi-source datasets. This advancement was detailed in a paper by Hao Xu and colleagues, submitted on June 29, 2026. The framework, known as MCO-PDE, aims to enhance the accuracy of scientific machine learning by integrating data from various sources, effectively overcoming limitations posed by single datasets.

Understanding the MCO-PDE Framework

The MCO-PDE framework employs a competitive optimization approach to identify shared PDEs from multiple datasets. It begins by training independent neural surrogates for each dataset, followed by a soft-competitive weighting mechanism that evaluates the credibility of each dataset. This process aggregates a consensus global coefficient, allowing for the simultaneous identification of functional forms and parameters of governing equations.

Integrated with a genetic algorithm for structural search, MCO-PDE demonstrates significant potential in recovering canonical equations. In tests, as few as 50 observations from seven different cases were used, resulting in high accuracy in the extracted equations.

Applications and Benefits of Multi-Source Data Integration

This framework is particularly beneficial when dealing with two- and three-dimensional domains characterized by irregular boundaries and heterogeneous coefficients. For instance, it has successfully extracted physically meaningful laws from real-world experiments conducted in wave tanks, showcasing its applicability in complex physical systems.

By leveraging data from various sources, MCO-PDE not only improves the interpretability of scientific machine learning but also paves the way for automated scientific discovery through heterogeneous data fusion.

Implications for Future Research

The findings from this research highlight a promising direction for future studies in the field of computational physics and machine learning. By utilizing MCO-PDE, researchers can potentially unlock new insights into complex systems that were previously obscured by limitations associated with single dataset analyses.

This innovative approach could lead to significant advancements in understanding physical phenomena and contributing to the development of more robust scientific models.