A Runge-Kutta method performs numerical integration of ordinary differential equations by combining several values of the derivative within each step. One second-order formula uses a trial step at the midpoint of an interval to cancel out lower-order error terms:
|
(1)
| |||
|
(2)
| |||
|
(3)
|
(where
is a Landau symbol), sometimes known as RK2, and
the fourth-order formula is
|
(4)
| |||
|
(5)
| |||
|
(6)
| |||
|
(7)
| |||
|
(8)
|
(Press et al. 1992), sometimes known as RK4. This method is reasonably simple and robust and is a good general candidate for the numerical solution of ordinary differential equations when combined with an intelligent adaptive step-size routine. The coefficients of a general Runge-Kutta method can be summarized in a Butcher tableau.
A strong-stability-preserving coefficient guarantees that the method preserves a convex
function bound for time steps up to
times the step-size limit for the Euler
forward method under the same bound. Gottlieb et al. (2026) proved that
for an irreducible
-stage
Runge-Kutta method of order at least two, this coefficient is at most
. The bound is attained by repeating the second-order
implicit midpoint rule
times with substep
for a full step of size
. The authors credit ChatGPT with identifying the connection
between the stability conditions and absorbing Markov
chains. They report Lean verification of the reduction to a moment inequality.