stabilNewton {pcme}R Documentation

Modified Newton method for function minimisation

Description

Implements a hybrid method of minimisation based on gradient and Newton methods.

Usage

stabilNewton(f, x, alpha = 10^(-4), beta = 0.9,
             xvaltol = 10^(-6), gradtol = 10^{-12}, gradnormtol = 10^{-8},
             itermax = 100, fonly, choosedirtol = 10^(-12), betaflag = TRUE)

Arguments

f evaluate the function, its gradient, and Hessian, see Details.
x initial value.
alpha damping constant for Newton direction.
beta modifier for further damping in Newton direction.
xvaltol tollerance for stopping rules.
gradtol tollerance for stopping rules.
gradnormtol tollerance for stopping rules.
itermax limit for number of iterations.
fonly evaluates f, but not derivatives.
choosedirtol ???
betaflag ???

Value

A list with the following components:

res$estimate the value of the argument where the minimum occurs.
res$karmijo ???
res$iterations number of iterations.
res$minimum f(x), the value of the function at res$estimate.
res$gradient gradient of f at res$estimate.
res$hessian Hessian of f at res$estimate.
res$cntgrad number of computations of the gradient.
res$cnthes number of computations of the Hessian.
res$optimhistory a log of the optimisation steps.

Note

To do: document this function and the method properly and add references.

Author(s)

Georgi Boshnakov

See Also

nlm


[Package pcme version 0.51 Index]