[R] correlation structure in lme without random effect

[Benjamin Tyner]

I was hoping to fit along the lines of

g<-gl(20,5)
y<-runif(100)
fit<-lme(fixed=y~g,correlation=corAR1(0,~1|g))

But I get the error "Incompatible formulas for groups in "random" and 
"correlation""

Any help would be greatly appreciated.
Ben