Case study on spatial models for areal data - Poisson CAR/IAR

That’s a warning about the way HMC worked. It said that in 25 proposals
the MCMC scheme detected that there could be something wrong.

Now, this is not something to panic about immediately - the diagnostic
throws out a lot of false positives. There is a lot of infromation in this
case study.
http://mc-stan.org/users/documentation/case-studies/divergences_and_bias.html

The take away message is that if thse warnings seem to be concentrating in
a particular part of the parameter space, then there may be some problems.
Otherwise they’re probably false positives.

Hope this helps!

D