ODE Initial-Value Problems: Overview
Initial-value ODE solvers march a solution forward in time from a known state, trading accuracy, stability, and cost across method families.
Marching a solution forward
An initial-value problem specifies a differential equation y' = f(t, y) together with the state y at a starting time. A numerical solver advances the solution in discrete steps, using the slope function f to estimate how y changes. The many methods differ in accuracy per step, stability, and how much of f they evaluate.
Explicit versus implicit
Explicit methods compute the next state directly from known values; they are cheap per step but can require tiny steps for stability on stiff problems. Implicit methods define the next state through an equation that must be solved, usually with Newton iteration; they cost more per step but remain stable with large steps on stiff systems.
Order and error
- Local truncation error is the error made in one step assuming exact input.
- Global error accumulates over all steps and is typically one order lower.
- A method of order p has global error proportional to h^p.
- Adaptive solvers vary the step to hold estimated error near a tolerance.
Choosing a solver
For smooth nonstiff problems, an explicit Runge-Kutta method such as RK4 or an adaptive pair like RK45 is the usual default. For stiff problems, implicit methods such as backward Euler or the BDF family are essential. Structure-preserving (symplectic) integrators are chosen when long-term conservation of energy or momentum matters.
ODE integrators drive time-dependent physics everywhere, from circuit and control models to plasma transport in the breeder Hyperion simulations, where stiffness forces implicit methods for practical step sizes.