xtmixed: stopping iteration when non-convergence is encountered

[Madu Abuchi]

I'm running a simulation using -mixed- command. My simulation contains both random intercept and random-coefficient. I am getting convergence issues and in my post file I have included e(converged) to report whether or not a particular estimation converged. I am encountering some estimations where a convergence issue is taking like more than 24hour to move to the next estimation.

My question is how can I stop an iteration at a particular number when a convergence issue is imminent so that it can just report not converged and move to the next step rather than waiting to thousands of itineration. I want to stop iteration at 500 and move over to next estimation. Does any one know how to deal with this?

I have used emtolerance(#), but it's not stopping at my specified #

[William Lisowski]

You want to look at the Maximization section in the output of help xtmixed to learn about maximization options that are common to many Stata commands. It refers you to the output of help maximize which suggests that

xtmixed ... , iterate(500) ...

may be what you are looking for. But it is also possible that the "difficult" option may aid in your estimation, depending on what it is telling you when you are not using it.

[Scott Merryman]

Also, once the emtolerance(#) threshold is passed, the optimization switches to a gradient-based method. The tolerance does not stop the maximization (unless you use the -emonly- option).

[Madu Abuchi]

thank you William and Scott. Your comments are very helpful