RK-opt is a MATLAB package for designing Runge-Kutta (RK) methods and stability polynomials. Supported objective functions include the principal error norm and the SSP coefficient. Supported constraints include stability polynomial coefficients, low-storage formulations, and structural constraints (explicit, diagonally implicit, etc.) RK-opt uses MATLAB’s optimization toolbox, in particular fmincon and linprog.
MATLAB’s global optimization toolbox function Multistart can be used to exploit the benefits of parallel search on multicore machines.
Find optimal Runge-Kutta method coefficients for a prescribed order of accuracy and number of stages. The objective function can be chosen as either the SSP coefficient or the leading truncation error coefficient. The method may be constrained to have a low-storage implementation and/or a prescribed stability polynomial. Implicit and diagonally implicit methods can also be optimized.
Find stability functions with optimal radius of absolute monotonicity. This includes codes for optimizing stability functions of multistep, multistage methods and even methods with downwinding. The optimization of rational functions is experimental.
Given a spectrum (typically corresponding to a spatial semi-discretization of a PDE), find an optimal stability polynomial in terms of its coefficients. These polynomial coefficients can then be used as input to RK-coeff-opt to find a corresponding Runge-Kutta method.
Some general utilities for analyzing Runge-Kutta methods.
RK-opt has been developed by David Ketcheson (primary developer and maintainer), Matteo Parsani, and Aron Ahmadia. Additional contributions include:
- Order conditions for multistep RK methods of orders 9-11 (Christopher Bresten, Zachary Grant, and Daniel Higgs)
It is released under a modified BSD License. If you use RK-Opt in published work, please see Citing RK-Opt.
This section contains a compilation of the documentation of each function, organized by subpackage.