Newton’s method uses the slope to jump toward a root and doubles the correct digits each step (quadratic). Halley’s method also uses the curvature (the second derivative), bending the step to fit — and triples the correct digits each step (cubic). For a smooth function near a simple root it reaches full precision in a couple of iterations where bisection needs fifty. Yes, that Halley — the comet astronomer.
Find √2 from a start of 1.0. Watch the correct digits roughly triple each iteration — three steps to full f64 precision, against Newton’s ~five and bisection’s ~fifty: live demo
“Higher order always means faster in practice.” — not simply. Halley converges in fewer iterations than Newton, but each iteration costs a second derivative; whether it wins on total work depends on how expensive f'' is. More order, more cost per step — a real trade, not a free lunch. cited
The step is x − 2ff’ / (2f’² − ff’’): drop the ff’’ curvature term and it collapses back to Newton’s x − f/f’ — Halley is Newton plus one correction. Halley, 1694
Halley reaches √2 to full f64 precision in three iterations on the canonical compiler: