I went back to your initial post, but I can’t quite figure out what’s going on. For example, I can’t figure out how many people are giving you your 4,847 reaction times. Regardless, my strategy for dealing with convergence issues, particularly when there is still uncertainty about the fundementals of your model (e.g., how to model noise in your data), is to tackle a smaller-scale version of the problem, ensure that you can model it reasonably well, and only then begin to complicate things. So in this sort of case, I might take data from a single condition and a single individual and try to model that. Once you have that nailed down, you can try adding in other factors. If things break at that point, you know it’s (probably) not because you picked the wrong noise function.