The subset of parameter values (commonly step sizes or complex test parameters) for which a numerical algorithm produces bounded or controlled iterates, i.e., does not amplify errors or cause unbounded growth when applied to canonical test problems.

Principle

A method’s amplification factor or stability polynomial evaluated on a canonical linear test equation (e.g., y' = λy) defines the region in the complex plane of λh where the magnitude of the amplification factor is ≤1; membership in that region ensures boundedness for the test problem.

Demonstration

For explicit Euler applied to y' = λy the stability region is the disk {z : |1+z| < 1} with z = hλ; implicit Euler has the entire left half-plane as its stability region, explaining its appropriateness for stiff problems.

Misapplication

Using a method’s stability region computed on linear test equations as a blanket guarantee for all nonlinear problems or for accuracy (rather than boundedness) is a misuse.

Consequence

Choosing step sizes or parameters inside the stability region prevents numerical growth of errors for the corresponding test dynamics and guides method selection (e.g., A-stable methods for stiff problems).

Reversal

The instability region is the complement where amplification factors exceed unity and iterates can grow without bound, causing numerical blow-up even if local truncation error is small.

Boundary

Stability regions are defined relative to specific test problems, norms, and problem classes; they do not ensure convergence or accuracy on arbitrary nonlinear systems and can depend on implementation details like variable step control.

Semantic Tension

Tension exists with notions of convergence and accuracy: a method can be stable (bounded) yet inaccurate, and a small stability region may still be adequate if dynamics lie within it; stiffness concerns interact with region size and shape.

Synthesis

The stability region condenses a numerical method’s boundedness behavior into a parameter-domain criterion: it identifies allowable step-size–spectrum combinations for which numerical iterates remain controlled and informs method and step-size choices.