Re: [R] Numerical stability of eigenvalue and hessian matrix in R

> On Feb 18, 2015, at 2:15 PM, Spencer Graves > wrote: > > >> On Feb 18, 2015, at 1:54 PM, David Winsemius > > wrote: >> >> >> On Feb 18, 2015, at 1:13 PM, C W wrote: >> >>> Thanks Thierry for the pointer, that's explains the problem. >>> >>> Is there anythin

Re: [R] Numerical stability of eigenvalue and hessian matrix in R

> On Feb 18, 2015, at 1:54 PM, David Winsemius wrote: > > > On Feb 18, 2015, at 1:13 PM, C W wrote: > >> Thanks Thierry for the pointer, that's explains the problem. >> >> Is there anything I can do about the matrix instability or numerical >> inaccuracy? > > There are matrix methods in the

Re: [R] Numerical stability of eigenvalue and hessian matrix in R

On Feb 18, 2015, at 1:13 PM, C W wrote: > Thanks Thierry for the pointer, that's explains the problem. > > Is there anything I can do about the matrix instability or numerical > inaccuracy? There are matrix methods in the Rmpfr package that support increased precision, but it is implemented wi

Re: [R] Numerical stability of eigenvalue and hessian matrix in R

This question is getting pretty deep into numerical analysis theory. The usual approach has already been mentioned... don't expect high accuracy in all problems. Your specific problem could have a special technique somewhere, but don't be surprised if we are not experts in your specific problem

Re: [R] Numerical stability of eigenvalue and hessian matrix in R

Thanks Thierry for the pointer, that's explains the problem. Is there anything I can do about the matrix instability or numerical inaccuracy? Mike On Wed, Feb 18, 2015 at 11:57 AM, Thierry Onkelinx wrote: > Have a look at FAQ 7.31 > > ir. Thierry Onkelinx > Instituut voor natuur- en bosonderzo

Re: [R] Numerical stability of eigenvalue and hessian matrix in R

Have a look at FAQ 7.31 ir. Thierry Onkelinx Instituut voor natuur- en bosonderzoek / Research Institute for Nature and Forest team Biometrie & Kwaliteitszorg / team Biometrics & Quality Assurance Kliniekstraat 25 1070 Anderlecht Belgium To call in the statistician after the experiment is done ma

Re: [R] Numerical stability of eigenvalue and hessian matrix in R

Hi Ben and JS, Thanks for the reply. I tried using: hessian(func = h_x, x, method = "complex"), it gives zero, that's good. # R code > hess.h <- hessian(func = h_x, x, method = "complex") > mat <- h_x(x)*hess.h - grad(h_x, x) %o% grad(h_x, x) > mat [,1][,2] [,3][,4] [1,]

Re: [R] Numerical stability of eigenvalue and hessian matrix in R

Hi, Since all entries in your hessian matrix and grad vector are integers, I suggest you execute the following for mat assignment. > mat <- round(h_x(x),digits=0)*round(hess.h,digits=0) - round(grad(h_x, > x),digits=0) %o% round(grad(h_x, x),digits=0) > mat [,1] [,2] [,3]

Re: [R] Numerical stability of eigenvalue and hessian matrix in R

C W gmail.com> writes: > > Hi list, > > I am running the following R code, the answer should be zero. But R gives > a very small negative number, what should I do? > > ##R code > library(numDeriv) h_x <- function(x){ a = x[1] b = x[2] c = x[3] d = x[4] (a^2 + c^2 + d^2) * (

[R] Numerical stability of eigenvalue and hessian matrix in R

Hi list, I am running the following R code, the answer should be zero. But R gives a very small negative number, what should I do? ##R code library(numDeriv) h_x <- function(x){ a = x[1] b = x[2] c = x[3] d = x[4] (a^2 + c^2 + d^2) * (b^2 + c^2 + d^2) } x1 = 10 x2 = 1 x3 = 0 x4