TL;DR
This paper introduces a novel class of numerical methods, Nonlinearly Partitioned Runge-Kutta (NPRK), designed to efficiently and accurately solve complex systems of ordinary differential equations (ODEs) that arise from discretizing partial differential equations. NPRK methods offer unprecedented flexibility in how different parts of a nonlinear system are treated implicitly or explicitly, leading to improved stability and accuracy, particularly for challenging problems in physics and engineering.
What Problem Does This Paper Address?
Many scientific and engineering problems, especially those involving partial differential equations (PDEs) like fluid dynamics or heat transfer, lead to large systems of ODEs when discretized in space. These ODE systems often exhibit multiple scales and stiffnesses, making them computationally expensive to solve with standard explicit time-stepping methods. While implicit methods offer better stability, they can be computationally demanding due to the need to solve large nonlinear systems at each time step. Existing partitioned methods (like IMEX or component-wise methods) address this by treating different parts of the ODE system implicitly or explicitly, but they are limited to systems that can be additively or component-wise separated. This paper tackles the challenge of solving ODEs where the nonlinear coupling between different parts of the system is more intricate and cannot be easily separated into additive or component-wise blocks.
Methods
The core contribution is the introduction of the Nonlinearly Partitioned Runge-Kutta (NPRK) framework. Unlike traditional Runge-Kutta methods, NPRK methods are designed for ODEs of the form $y' = F(y, y)$, where the function $F$ takes two arguments that can be treated with different degrees of implicitness. This allows for a more flexible distribution of stiffness across the nonlinear terms. The paper details the NPRK ansatz, its connection to existing Runge-Kutta and Additive Runge-Kutta (ARK) methods, and derives order conditions and linear stability properties. They also introduce specific NPRK methods, including Implicit-Explicit (IMEX) and Implicit-Implicit (IMIM) variants, with varying numbers of stages and implicitness. A key aspect is the development of a practical algorithm for sequentially coupled IMEX-NPRK methods.
Datasets
Not discussed in the paper. The paper focuses on developing numerical methods and demonstrates their utility on established benchmark problems.
Key Insights and Ideas
- Flexible Implicitness: The most significant insight is the ability to distribute implicitness within nonlinear terms, not just between additive components or system partitions. This is crucial for systems where nonlinearities are deeply intertwined.
- Generalization of Existing Methods: NPRK methods are shown to generalize existing ARK and Partitioned Runge-Kutta (PRK) methods, providing a unified framework.
- Stability in Coupled Stiff Limits: The paper emphasizes the importance of stability in "coupled stiff limits" for IMEX-NPRK methods, which is critical for problems where both explicit and implicit components can become stiff. This is demonstrated through the viscous Burgers' equation example.
- Bridging Theory and Practice: The work not only develops the theoretical framework but also provides practical algorithms and concrete examples of NPRK methods, making them accessible for implementation.
Key Results
The paper demonstrates the utility of NPRK methods through numerical experiments on two challenging problems:
- Viscous Burgers' Equation: NPRK methods, particularly those stable in the coupled stiff $z_2$ limit, showed superior stability and accuracy for the conservative form of the equation, which is more challenging due to its nonlinear coupling. Methods stable in this limit achieved expected order of accuracy even at large timesteps, while others failed.
- Gray Thermal Radiation Transport (TRT): NPRK methods were applied to a high-dimensional, stiff TRT problem. They achieved second-order accuracy at reasonable timesteps and demonstrated better stability properties compared to some existing IMEX schemes, especially for low spatial resolution where order reduction and instabilities can be problematic. The NPRK methods were shown to be more robust and require less memory than some comparable methods.
Limitations and Future Work
- General NPRK Methods: The paper acknowledges that the class of general M-component NPRK methods (4.32) is vast, and further study is needed.
- Higher-Order Methods: Future work will focus on constructing higher-order NPRK methods and methods with more varied forms of implicitness.
- Application Domains: The authors plan to explore additional application domains that can benefit from nonlinear splittings.
- Computational Cost: While NPRK methods offer improved stability and accuracy, the computational cost of solving implicit systems at each stage can still be significant, especially for very large systems. The paper explores strategies to mitigate this through careful partitioning and method design.
Summary updated