nleqslv (version 3.3.5)

searchZeros: Solve a nonlinear equation system with multiple roots from multiple initial estimates

Description

This function solves a system of nonlinear equations with nleqlsv for multiple initial estimates of the roots.

Usage

searchZeros(x, fn, digits=4, ... )

Value

If no solutions are found NULL is returned. Otherwise a list containing the following components is returned

x

a matrix with each row containing a unique solution (unrounded)

xfnorm

a vector of the function criterion associated with each row of the solution matrix x.

fnorm

a vector containing the function criterion for every converged result

idxcvg

a vector containing the row indices of the matrix of initial estimates for which function value convergence was achieved

idxxtol

a vector containing the row indices of the matrix of initial estimates for which x-value convergence was achieved

idxnocvg

a vector containing the row indices of the matrix of initial estimates which lead to an nleqslv termination code > 2

idxfatal

a vector containing the row indices of the matrix of initial estimates for which a fatal error occurred that made nleqslv stop

xstart

a matrix of the initial estimates corresponding to the solution matrix

cvgstart

a matrix of all initial estimates for which convergence was achieved

Arguments

x

A matrix with each row containing an initial guess of the roots.

fn

A function of x returning a vector of function values with the same length as the vector x.

digits

integer passed to round for locating and removing duplicate rounded solutions.

...

Further arguments to be passed to nleqslv, fn and jac.

Details

Each row of x is a vector of initial estimates for the argument x of nleqslv. The function runs nleqslv for each row of the matrix x. The first initial value is treated separately and slightly differently from the other initial estimates. It is used to check if all arguments in ... are valid arguments for nleqslv and the function to be solved. This is done by running nleqslv with no condition handling. If an error is then detected an error message is issued and the function stops. For the remaining initial estimates nleqslv is executed silently. Only solutions for which the nleqslv termination code tcode equals 1 are regarded as valid solutions. The rounded solutions (after removal of duplicates) are used to order the solutions in increasing order. These rounded solutions are not included in the return value of the function.

Examples

# Dennis Schnabel example 6.5.1 page 149 (two solutions)
set.seed(123)
dslnex <- function(x) {
    y <- numeric(2)
    y[1] <- x[1]^2 + x[2]^2 - 2
    y[2] <- exp(x[1]-1) + x[2]^3 - 2
    y
}
xstart <- matrix(runif(50, min=-2, max=2),ncol=2)
ans <- searchZeros(xstart,dslnex, method="Broyden",global="dbldog")
ans

# more complicated example
# R. Baker Kearfott, Some tests of Generalized Bisection,
# ACM Transactions on Methematical Software, Vol. 13, No. 3, 1987, pp 197-220

# A high-degree polynomial system (section 4.3 Problem 12)
# There are 12 real roots (and 126 complex roots to this system!)

hdp <- function(x) {
    f <- numeric(length(x))
    f[1] <- 5 * x[1]^9 - 6 * x[1]^5 * x[2]^2 + x[1] * x[2]^4 + 2 * x[1] * x[3]
    f[2] <- -2 * x[1]^6 * x[2] + 2 * x[1]^2 * x[2]^3 + 2 * x[2] * x[3]
    f[3] <- x[1]^2 + x[2]^2 - 0.265625
    f
}


N <- 40 # at least to find all 12 roots
set.seed(123)
xstart <- matrix(runif(3*N,min=-1,max=1), N, 3)  # N initial guesses, each of length 3
ans <- searchZeros(xstart,hdp, method="Broyden",global="dbldog")
ans$x