Algorithms to compute a local solution to min f(x) where f : R -> R.
Algorithms are variants of Newton's method (and secant iteration)
- Local/LocalIterations.jl contains functions which perform a single iteration for several variants and accelerations. May be used to exhibit high order convergence by iteratively calling those functions. Such a test is performed in the test suite. DividedDifference2.jl and coefH1N.jl provide Hermite interpolation tools used in the accelerated variants.
Globalized solvers are proposed for NLPModels and Stopping.
-
bracket/Bracket.jl contains a global solver to produce a reduced bracket (interval) guaranteed to contain a local minimum. Several variants to pick a point in a given interval are regrouped in pick_inN.jl (Newton variants) and pick_inS.jl (secant variants).
- driversBracket.jl contain ready to use solvers with selected options/pick_in options.
-
TR/TR1D.jl contains a trust region scheme which may use one of several models in TR1DModels.jl (quadratic models) and TR1DModels3.jl (cubic and higher models).
- driversTR.jl contain ready to use solvers with selected models.
One important usage of scalar optimizers is the implementation of line searches within higher dimensional optimization algorithms.
- LineSearch.jl contains models and wrappers used to compute Armijo-Wolfe solutions to the line search subproblem.