# Non-metric multidimensional scaling, no convergence

I'm using MetaMDS from the VEGAN package to run a non-metric multidimensional scaling analysis. My stress level for 3 dimensions is in the excellent range (i.e., stress<.05), however, the model never converges. I have attempted to set my "trymax" iteration level all the way up to 5000, but it still never converges.

I'm curious if this is a problem, I have not come across any sources that say my model is invalid if it doesn't converge, rather ensuring the stress level is low with low dimensionality.

is that assumption correct?

• The convergence criteria aren't like strict convergence measures from other optimisations; basically if you hit upon two configurations (solutions) that are effectively the same then convergence is declared (for what "effectively" means, see the help). You can save the best solution and pass it to metaMDS as the starting point for the next set of trymax iterations; you could run the thing a few times and save the separate solutions it has down as the "best" solution and compare them; trying to do this in 3d is tricky. – Reinstate Monica - G. Simpson Aug 14 at 17:30