Introduction
This special issue of the Journal of Computational Dynamics (Vol. 11, No. 4, October 2024) serves as a tribute to the careers and collaborations of two influential Norwegian mathematicians, Brynjulf Owren and Hans Munthe-Kaas, on the occasion of their 60th birthdays. Their pioneering work at the intersection of geometry, algebra, and numerical analysis has revolutionized the way differential equations are solved on curved spaces.
The Challenge of Geometric Integration
A fundamental challenge in computational mathematics is designing numerical methods for differential equations that evolve on non-Euclidean spaces, such as spheres or rotation groups. Standard numerical methods, such as basic Runge-Kutta schemes, are designed for flat, Euclidean space. When applied to manifolds, these methods often "drift" off the required surface, violating physical constraints or geometric invariants.
Owren and Munthe-Kaas addressed this by developing "structure-preserving" algorithms. These methods ensure the numerical solution remains on the manifold by construction, utilizing the mathematical language of Lie groups and Lie algebras.
Core Theoretical Frameworks
The honorees' contributions are centered on two major theoretical frameworks:
- Lie Group Integrators: These are numerical integration methods built from coordinate-independent operations, such as Lie group actions. Instead of updating a position by adding a vector, these methods use the exponential map to move along the manifold, preserving its intrinsic geometry.
- Lie-Butcher Theory: This is a sophisticated extension of classical Butcher series (B-series). While B-series are used to analyze the order and accuracy of Runge-Kutta methods in Euclidean space, Lie-Butcher theory combines them with Lie series. This provides a powerful framework for analyzing and constructing numerical methods specifically for manifolds.
These methods were refined through tight cooperation between research groups at NTNU (Trondheim), the University of Bergen, and the University of Cambridge. A pivotal moment for the field occurred in 1997, when both Owren and Munthe-Kaas spent a sabbatical year at the Department of Applied Mathematics and Theoretical Physics (DAMTP) in Cambridge, establishing the foundational theory and computational aspects of Lie group integrators.
Cross-Disciplinary Impact and Modern Applications
The research activity surrounding Lie-Butcher theory intersects with diverse fields, including control theory, stochastic differential equations, renormalization in physics, and combinatorial Hopf algebra.
A significant modern development is the "dynamical systems view" of Artificial Intelligence. Owren’s recent work in Deep Learning analyzes neural networks through this lens, applying geometric integration principles to improve the stability and interpretability of AI models. Similarly, Munthe-Kaas has utilized high-level mathematical abstractions, such as group theory and crystallographic symmetries, as organizing principles for developing massively parallel computational software.
Future Directions
The field is currently moving toward deeper integration with Machine Learning and Stochastic Analysis. The next frontier for geometric numerical integration lies in "Geometric Deep Learning" and the development of structure-preserving algorithms for increasingly complex, high-dimensional stochastic systems. This special issue represents an ongoing exploration of these theoretical and computational aspects, marking a significant milestone in a still-evolving field.
Summary updated