problem with pdStrat in nlme with nested random effects

[Gosselin Frederic <Frederic.Gosselin@nogent.cemagref.fr>]

Dear all,
has anyone used pdStrat with nested random effects in nlme? Did you meet some problems? I have met the following problem when declaring pdStrat for two nested random effects: e.g.:

random = list(block.group.subgroup = pdStrat(pdIdent(g~1),strata=~as.factor(group.subgroup)), plot.group.subgroup = pdStrat(pdIdent(c~1),strata=~as.factor(group.subgroup)))

I get singularities or Inf values, with error messages such as:
Problem in .C("mixed_loglik",: subroutine mixed_loglik: 6 Inf value(s) in argument 3
or:
Problem in .C("mixed_EM",: Singularity in backsolve, while calling subroutine mixed_EM
I have met the same problem in various settings.

The strange thing is that the same command "works" (i.e. there is no singularity) if I declare random effects in the reverse order to that of nesting (but of course the result is not adequate), i.e.

random = list(plot.group.subgroup = pdStrat(pdIdent(c~1),strata=~as.factor(group.subgroup)), block.group.subgroup = pdStrat(pdIdent(g~1),strata=~as.factor(group.subgroup)))

Sincerely,
Frédéric Gosselin
Researcher (PhD) & Engineer in Forest Ecology
Cemagref
Domaine des Barres
45 290 Nogent-sur-Vernisson
FRANCE
Tel: 33-2-38-95-03-58
Fax: 33-2-38-95-03-44