[ADMB Users] matrix not positive definite

[Mollie Brooks]

Hi ADMB users,

I am trying to figure out why my variance-covariance matrix isn't positive definite. The matrix starts out ok and then I get the error
Error matrix not positive definite in choleski_decomp

To figure out the problem, I've been outputting parameters and the Var-cov matrix. Two parameters go into calculating the matrix: rho and sigmaSq. The matrix starts out as it should (I calculated the values separately to check). But then the matrix changes to something unexpected for the given parameter values and that's when it isn't SPD anymore.
  
This looks correct
rho = 0.5
sigmaSq = 1
varcov= 1 0.25 0.0625 0.03125  
0.25 1.3125 0.328125 0.164062  
0.0625 0.328125 1.33203 0.666016  
0.03125 0.164062 0.666016 1.33301 

and this looks correct
rho = 0.880797
sigmaSq = 1
varcov=  1 0.775803 0.601871 0.530126
 0.775803 2.37767 1.84461 1.62473
 0.601871 1.84461 3.20686 2.82459
 0.530126 1.62473 2.82459 3.48789

but then the value of rho stays the same and the matrix changes.
rho = 0.880797
sigmaSq = 1
varcov=  1 -4.72992 1.037 -7.7791
 -4.72992 -1.89369 2.18218 -4.8216
 1.037 2.18218 3.46875 -2.17649
 -7.7791 -4.8216 -2.17649 -0.917046

Can anyone offer any explanation or advice for why this is happening?
thanks,
Mollie


Mollie Brooks
Postdoctoral Researcher, Ponciano Lab
Biology Department, University of Florida
http://people.biology.ufl.edu/mbrooks