RE: Help: Non-positive semi-definite message
From: "KOWALSKI, KENNETH G. [R&D/1825]" <kenneth.g.kowalski@pharmacia.com>
Subject: RE: Help: Non-positive semi-definite message
Date: Mon, 20 Aug 2001 13:06:18 -0500
Alan,
I agree with Vladimir. In looking at your ETAs you might want to calculate
the correlations (NONMEM only reports the covariance matrix for OMEGA). The
non-positive semi-definite message from the $COV step often arises when
OMEGA is over-parameterized. When this happens I often find that the ETAs
for two or more of the parameters (e.g., CL and V) have their correlation
estimated to be 1.0. In this setting I usually reduce the dimensionality
for OMEGA by using a shared ETA for the two parameters with a different
scale parameter to account for different variances. For example,
CL=THETA(1)*EXP(ETA(1))
V=THETA(2)*EXP(THETA(3)*ETA(1))
Here Var(logV)=(THETA(3)^2)*Var(logCL) and Corr(logV, logCL)=1.0.
Ken