Once per chain is usually OK. If you really want to be on the safe side, it is good to try to understand what is actually going on… What would need to happen to make the standard deviation parameter in question 0? Does this happen only when you put some “extreme” data for the model? Does it happen if you use different initial values (e.g. init = 0.5 or init = 0 in rstan is a good first check as it forces the sampler to init in a narrower region)?
1 Like