Under the circumstances, i.e. accessing \nu only via forward simulation, I would think that an exact generative process for the conditional distribution is a pretty big ask? (unless you meant an exact generative process for their approximate conditional distribution, which is given implicitly in Theorem 1)

]]>