nlminb again

[Yingfu Xie <Yingfu.Xie@resgeom.slu.se>]

Hello, dear S-plus users!

 

I want to maximize a likelihood function, which is a complicated function of parameters, involving products of matrices. Of course, 'gradient' and 'Hessian' are by no means to be specified since we cannot obtain the analytical form of the likelihood with respect to parameters. In fact, I wrote a function  to compute the likelihood, ready to put into nlminb. But I was warned that singularity occurred in nlminb.0(...) and it returned a zero objective likelihood and initial parameters. No iterations happened at all.

Now, my questions are:

Does nlminb work for this? Isn't it valid for any functional form or it is only valid for functions with two continuous derivatives? Is there any other tool available in S-plus to solve my problem?

 

Any explanations or suggestions are highly appreciated. By the way, I am using S-plus 6.1 in Windows.

 

Regards

 

Yingfu

###########################################

This message has been scanned by F-Secure
Anti-Virus for Microsoft Exchange.

###########################################