weighted least squares

["Paul, David A" <paulda@BATTELLE.ORG>]

The gls( ) function requires grouped data (as far as I can tell) but allows
the user to impose structure on heteroscedasticity in the residuals (such
as varPow(), &etc.). 

Does there exist an Splus function allowing the user to place structure on the
variances of the residuals in the absence of grouped data?

I am aware that ltsreg( ) will compute weights which can then be used
in lm( ) or lsfit( ), but these weights appear to be computed in an
"unstructured" manner.

Respectfully,

 David Paul, Ph.D.
  Battelle Memorial Institute
  505 King Avenue
  Columbus, OH  43201
  614.424.3176