A numerical strategy that traces solution branches of parameterized equations by continuously deforming a simple problem with known solutions into a target problem, typically using predictor–corrector steps along a homotopy path.

Principle

Construct a homotopy H(x,t) interpolating between an easy problem H(x,0)=0 and the target H(x,1)=0, then follow solution curves x(t) for t in [0,1] using stepwise prediction and correction while handling bifurcations and singularities appropriately.

Demonstration

To solve a difficult nonlinear system F(x)=0, define H(x,t)=(1-t)G(x)+tF(x) with G simple; start from a known root x0 of G, increment t, predict x(t+Δt), and apply Newton-correction to stay on the solution branch until t=1.

Misapplication

Assuming all paths are nonsingular and that continuation will find every solution without monitoring turning points or bifurcations; neglecting adaptive step control or failing to switch branches at singularities leads to missed solutions or divergence.

Consequence

Continuation produces robust global routes from easy to hard problems, enables tracing of multiple branches and bifurcations, and often succeeds where naive Newton on the target fails, at the cost of path-following overhead.

Reversal

Direct solvers that attempt to solve the target problem immediately (e.g., plain Newton from a poor guess) may converge faster when they work but lack the global guidance provided by continuation.

Boundary

Effective when solutions depend continuously on the homotopy parameter and when predictor–corrector mechanisms and branch detection are used; problematic near singular solutions, for extremely large systems without continuation-friendly structure, or when the homotopy is poorly chosen.

Semantic Tension

Tension exists between continuation and direct local methods: continuation emphasizes global deformation and branch following, while local methods emphasize fast local convergence and may miss distant branches.

Synthesis

Continuation Method = build a continuous deformation from a solvable problem to the target, follow solution paths with predictor–corrector steps and branching logic to reliably obtain solutions that are inaccessible to local solvers alone.