[ADMB Users] Multivariate Autoregressive State Space Models

[Mollie Brooks]

Hi Users,
I am attempting to fit Multivariate Autoregressive State Space Models similar to the MARSS package in R (http://journal.r-project.org/archive/2012-1/RJournal_2012-1_Holmes~et~al.pdf) with the goal of later being able to go beyond that package's ability. 

I'm working on replicating the example from Ives et al. 2003. "Estimating Community stability and Ecological Interactions from Time-Series Data". I have it working if the variance-covariance matrix for process error is known. 

Now I'm working on fitting the variance-covariance matrix for process error. I've followed the directions here
http://www.admb-project.org/examples/admb-tricks/parameterization/covariance-matrices
http://www.admb-project.org/examples/admb-tricks/parameterization/covariance-matrices/correlation-matrix

But somewhere in the code I keep getting the error
Error matrix not positive definite in choleski_decomp
can't improve function value in trust region calculations

I see this error message occurs in the function calculate_laplace_approximation.

Does anyone else have experience with fitting MARSS type models? 

I'd be happy to send my code to anyone who's willing to help.

thanks,
Mollie


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