[R-sig-ME] lmeControl

[Irene Mantzouni]

Dear all, 

I have set up a linear mixed model:

A4.lme=lme(log.s.r~SSB,=random=list(FishStock=pdDiag(~SSB)))

using the nlme library. 

When I try to make the error structure group specific using
weights=varIdent(form=~1|FishStock)

the model cannot converge using the default settings:

"nlminb problem, convergence error code = 1
  message = iteration limit reached without convergence (9)"

However, if I modify the maximum number of iterations in lmeControl
(both maxIter and msmaxIter are set to 500), then the model can indeed
converge. 
The problem is then that I cannot get the confidence intervals for the
variance-covariance components, but this error message appears:
"Cannot get confidence intervals on var-cov components: Non-positive
definite approximate variance-covariance"

On the other hand, model comparison shows that the model with the
heterogenous variances is superior. 
So, I am wondering if the lmeControl modification can invalidate the
model results and if I should use it at all. 

(Please let me know if you need any more details about the model)

Thank you!