[R] lme: problems with corARMA

[Douglas Bates]

Stephan Moratti wrote:

> Trying following example from Pinheiro and Bates in order to fit an
> ARMA(1,1) model:
> 
> library(nlme)
> fm1Ovary.lme<-lme(follicles~sin(2*pi*Time)+cos(*pi*Time),data=Ovary,random=p
> dDiag(~sin(2*pi*Time)))
> fm5Ovary.lme<-update(fm1Ovary.lme,corr=corARMA(p=1,q=1))
> 
> I get follwing error message:
> 
> Error in "coef<-.corARMA"(`*tmp*`, value = c(62.3428455941166,
> 62.3428517930051 : 
>         Coefficient matrix not invertible
> 
> Does somebody know why it doesn't work ?

The optimization code in R's nlm is different from the code in S-PLUS's 
ms.   The nlm optimizer does not converge on that ill-conditioned example.