NUTS sampler encountered divergent paths

[Jacob Trowbridge]

Hi all,

I have a question regarding divergent paths in nimbleHMC. I am running a straightforward regression model that is estimating survival probabilities, by age and sex, with a random intercept of site. I am encountering divergent paths (usually 20-30) in this model. I have increased the "delta" to 0.95 and 0.99 and used non-centered random effects, but there always remains at least 5-10 divergent paths. My question is if there is a way to know if these divergent paths are occurring during warmup or that actual sampling phase? If they are occurring during the warmup period, then are divergent paths an issue?

Thanks,

Jake