[R] Need help on Newton-Raphson optimization

[Rolf Turner]

On 11/08/11 11:36, michael wrote:
> Hi,
>
>       Is there available package on the optimization function using
> Newton-Raphson method (iterative quadratic approximation)? I have been using
> the 'optim' function in R and found it really unstable (it depends heavily
> on the initial values and functional forms). If I have to code it by myself,
> can I get some advice on how to start (any good reference or sample code)? I
> really appreciate any help.

I find it very strange that you say that optim() is ``unstable''.  I have
never found it so.  In problems that I have worked on involving hidden
Markov models (maximizing the likelihood thereof) I have found Newton's
method to break down frequently whereas optim() never failed to yield
a sensible result.

I don't understand why you refer to the Newton-Raphson method (more
properly called just Newton's method) as ``iterative *quadratic* 
approximation''.
Newton's method uses just the first order Taylor series approximation to
the function whose zeros are sought.

All that being said, I have written a function which effects Newton's 
method.
Email me off-list if you would like me to provide this function.

         cheers,

             Rolf Turner