Convergence in all parameters, apart from the log-probability

[Ben Lambert]

Hi,

I have been working on a particular model, and am finding that there is convergence in all parameters (Rhat < 1.1), apart from the log-probability which remains (stubbornly) not so (Rhat ~ 10). 

This, to me at least, is a new occurrence, and I am not sure what it means/how-to-remedy the situation. Does anyone have any ideas on this? Answers from both from a theoretic (why can this happen?) and practical standpoint (what should I do about it?) would be useful.

Best,

Ben