Why use multiple stages instead of just a small step with Euler's method?

Each stage samples the slope at a different point within the step, and combining them cancels low-order error terms, so a Runge-Kutta method achieves high accuracy with far larger steps than Euler's method would need for the same error.

When is an implicit Runge-Kutta method worth its extra cost?

For stiff problems, where explicit methods require impractically tiny steps for stability, implicit Runge-Kutta methods remain stable at large step sizes. The cost of solving a nonlinear system each step is then more than offset by taking far fewer steps.

Runge-Kutta Methods

Runge-Kutta methods advance the solution of an ODE one step at a time using several intermediate stage evaluations of the right-hand side, achieving high order without storing past steps.

Definition

A Runge-Kutta method is a one-step method for ordinary differential equations that computes the next solution value from the current one by forming a weighted combination of several stage derivatives evaluated at intermediate points within the step.

Scope

This topic covers explicit and implicit Runge-Kutta methods, their Butcher tableau representation, order conditions derived from rooted-tree theory, embedded pairs for adaptive step-size control, and the absolute-stability properties that distinguish methods suitable for stiff and nonstiff problems.

Core questions

How do internal stages let a one-step method attain high order of accuracy?
How are the order conditions for a Runge-Kutta method derived and organized?
How do embedded pairs provide an inexpensive local error estimate for step-size control?
What distinguishes explicit from implicit Runge-Kutta methods in cost and stability?

Key theories

Butcher tableau and order conditions: A Runge-Kutta method is specified by its Butcher tableau of coefficients, and the requirement that it match the Taylor expansion of the exact solution to a given order produces a set of algebraic order conditions systematically generated using rooted trees.
Embedded pairs and adaptive control: Two methods sharing the same stages but different weights — an embedded pair such as the Runge-Kutta-Fehlberg or Dormand-Prince schemes — yield two solution estimates of different order whose difference estimates the local error and drives automatic step-size selection.

Mechanisms

Within each step the method evaluates the right-hand side at several stage points, each defined as the current value plus a combination of previously computed stage derivatives; the new solution is a weighted sum of these stage derivatives. Explicit methods order the stages so each depends only on earlier ones and can be evaluated directly, while implicit methods couple the stages through a nonlinear system solved at each step, gaining the strong stability needed for stiff problems. Embedded pairs reuse the stage evaluations to produce a companion estimate for error control.

Clinical relevance

Runge-Kutta methods, especially adaptive explicit pairs like Dormand-Prince, are the default general-purpose ODE integrators in scientific computing environments, used for trajectory simulation, chemical kinetics, control systems, and any nonstiff initial value problem; implicit Runge-Kutta methods extend the same framework to stiff and structure-preserving integration.

History

The methods began with Runge's 1895 work and Kutta's 1901 systematic schemes; John Butcher's algebraic theory in the 1960s organized their order conditions via rooted trees, and the development of efficient embedded pairs such as Fehlberg's and the Dormand-Prince pair made adaptive Runge-Kutta integration the standard tool it is today.

Key figures

Carl Runge
Wilhelm Kutta
John C. Butcher
John R. Dormand

Seminal works

hairer1993
butcher2016

Frequently asked questions

Why use multiple stages instead of just a small step with Euler's method?: Each stage samples the slope at a different point within the step, and combining them cancels low-order error terms, so a Runge-Kutta method achieves high accuracy with far larger steps than Euler's method would need for the same error.
When is an implicit Runge-Kutta method worth its extra cost?: For stiff problems, where explicit methods require impractically tiny steps for stability, implicit Runge-Kutta methods remain stable at large step sizes. The cost of solving a nonlinear system each step is then more than offset by taking far fewer steps.