numopt-js
    Preparing search index...

    Interface StrongWolfeLineSearchOptions

    Options for Strong Wolfe line search (Nocedal & Wright, 2nd ed., Algorithm 3.5).

    This line search aims to satisfy both:

    • sufficient decrease (Armijo) and
    • curvature condition (Strong Wolfe)

    Strong Wolfe is commonly used with quasi-Newton methods (BFGS / L-BFGS) because it tends to produce steps that satisfy the curvature condition (s^T y > 0), which helps keep Hessian approximations well-behaved.

    interface StrongWolfeLineSearchOptions {
        initialStepSize?: number;
        maxIterations?: number;
        maxZoomIterations?: number;
        stepSizeGrowthFactor?: number;
        wolfeC1?: number;
        wolfeC2?: number;
    }
    Index
    initialStepSize? maxIterations? maxZoomIterations? stepSizeGrowthFactor? wolfeC1? wolfeC2?
    initialStepSize?: number

    Initial step size to try. If not provided, the initial step size is scaled by the gradient norm: α₀ = 1.0 / ||∇f(x)||

    maxIterations?: number

    Maximum number of outer iterations (bracketing phase). Default: 25

    maxZoomIterations?: number

    Maximum number of zoom iterations (within a bracket). Default: 25

    stepSizeGrowthFactor?: number

    Growth factor for expanding the trial step size when still in the bracketing phase. Default: 2.0

    wolfeC1?: number

    Armijo parameter c1 for sufficient decrease condition. Typical value: 1e-4.

    wolfeC2?: number

    Curvature parameter c2 for Strong Wolfe condition. Typical value: 0.9.

    Settings
    Member Visibility

    On This Page
    Properties