@junpenglao thanks for the amplifying info! When you say:
when you apply function on RV explicitly, the volume changes are not accounted for.
Is this issue introduced anytime pm.Deterministic is used?
I often need to add or subtract random variables in sampling, IRT as an example. I wasn’t aware that this would/should preclude use of NUTS, but it seems this is a more serious possibility based on your feedback. Thoughts?