R help - Dec 2003 - matrix bending

Dear All,

I was wondering whether any one knows of a matrix bending function in 
R that can turn non-positive definite matrices into the nearest 
positive definite matrix.  I was hoping there would be something akin 
to John Henshall's flbend program 
(http://agbu.une.edu.au/~kmeyer/pdmatrix.html), which allows the 
standard errors of the estimated matrix elements to be considered in 
the bending process.

Thanks,

Jarrod.

Here is a function that I use for symmetric matrices:

make.positive.definite <-
function(x, tol=1e-6) {
    eig <- eigen(x, symmetric=TRUE)
    rtol <- tol * eig$values[1]
    if(min(eig$values) < rtol) {
        vals <- eig$values
        vals[vals < rtol} <- rtol
        srev <- eig$vectors %*% (vals * t(eig$vectors))
        dimnames(srev) <- dimnames(x)
        return(srev)
    } else {
        return(x)
    }
}


Patrick Burns

Burns Statistics
patrick at burns-stat.com
+44 (0)20 8525 0696
http://www.burns-stat.com
(home of S Poetry and "A Guide for the Unwilling S User")


jarrod.hadfield at imperial.ac.uk wrote:
> Dear All,
>
> I was wondering whether any one knows of a matrix bending function in 
> R that can turn non-positive definite matrices into the nearest 
> positive definite matrix.  I was hoping there would be something akin 
> to John Henshall's flbend program 
> (http://agbu.une.edu.au/~kmeyer/pdmatrix.html), which allows the 
> standard errors of the estimated matrix elements to be considered in 
> the bending process.
>
> Thanks,
>
> Jarrod.
>
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://www.stat.math.ethz.ch/mailman/listinfo/r-help
>
>