Here's an interesting problem I am running into. I was working on improving convergence rates on a script that was giving me some problems. I finally got better convergence on my laptop after switching to FIML and changing the scaling of coefficients. (I think the variances of x and derivatives I was estimating were differing by several orders of magnitude.) Then it was time to get things working on Xgrid. After a while I noticed that one of my Xgrid computers (my desktop) was almost never converging, while my other computers were converging almost all the time.
There were two differences that I could find: my desktop had R 2.9.1 (other computer 2.9.2) and my desktop had a new version of OpenMx (other computer was updated 2 weeks ago). I updated R on the desktop, no improvement. I updated OpenMx on the other machine --- almost none of the models converge now on both machines.
Something has changed with OpenMx in the new version. Is there something new that I might not be specifying in a function which was not previously required? Could there be something different with mxFIMLObjective? Other ideas?