From: Alison_Carter@eisai.com>
Subject: [NMusers]
Date: Tuesday, July 30, 2002 12:23 PM


Dear NONMEM users,

I am having a frustrating time trying to fit a simple 2-compartment model
to population data.  I can fit the individual subjects with 2-compartments
fairly easily, but when I try them together, I get the message:
R MATRIX ALGORITHMICALLY NON-POSITIVE-SEMIDEFINITE
BUT NONSINGULAR COVARIANCE STEP ABORTED
I know from the manual that this means the apparent minimum might be
a saddle point and that I should not trust the final estimates.  My
problem is what to do to correct the situation.  I have tried various error models and
ranges of initial estimates, but I am out of ideas.  Does anyone have any
suggestions?  Does this indicate the data might be better fit with a
1-compartment model?
Thanks very much!

 Alison Carter, PhD


------------------
From: "Venkatesh Atul Bhattaram" 
Subject: [NMusers]
Date: Tue, 30 Jul 2002 13:01:09

Hello Alison

This could be due to overparameterization. I would check the estimates of
eta and their correlations and try to minimize them in the model. I found
this to work well. Also did you try other estimation methods.

Venkatesh Atul Bhattaram
Post-doctoral Fellow
University of Florida
Gainesville-32610
------------------