TABLE OF CONTENTS

initRoutine/inverthessian [ Functions ]

NAME

    inverthessian --- invert a Hessian matrix

FUNCTION

The Hessian matrices returned by optim() are not always positive-definite and invertible. This function adjusts the diagonal of an input matrix until it is invertible. This gives an acceptable covariance matrix for subsequent use.

SYNOPSIS

922 inverthessian <- function(hess){

INPUTS

    hess   a matrix

OUTPUTS

    Sigma  the inverse of hess, or of something very close to it

SOURCE

925     K <- dim(hess)[1]
926     # Try to invert the Hessian
927     Sigma <- try(solve(-hess), silent = TRUE)
928     d <- 10
929     # modify the diagonal until it is invertible
930     while(inherits(Sigma, "try-error")){ 
931         Sigma <- try(solve(-(hess - 10^(-d) * mdiag(K))), silent = TRUE);
932         d <- d - 1
933     }
934     # modify the diagonal until it is positive definite
935     while(!all(eigen(Sigma)$val > 0)){ 
936         Sigma <- solve(-(hess - 10^(-d) * mdiag(K))) ;
937         d <- d - 1  
938     }
939     return(Sigma)
940 }