[R] nlme: creating pdMats

[Scott Rifkin]

I have an experiment with an interaction (VT) between Varieties and Time.  
There are 12 Varieties and 2 Times (Early and Late), and Time is a fixed 
affect.

I have been estimating a single variance from this interaction using 
pdIdent(~VT-1).

I would like to test 2 more complicated models.

1) Estimate 2 variances, one for Variety*Time(Early) and one for 
Variety*Time(Late), again with a diagonal matrix.

2) The above with a covariance term between the Times across Varieties.  
To do this, I think I would want a variance covariance matrix like (1) but
with one more parameter wherever a row and a column share a Variety but 
differ in Times.  The rest would be zero.

I would appreciate if someone would explain how I can do this using the 
standard pdMat classes, or, more likely, how to create a new one to do 
this.

Thanks much,
Scott Rifkin