\( p_t(\mathbf{x}) \)
\( p_t(\mathbf{x}) \)
\( p_{t+\Delta t}(\mathbf{x}) \)
This illustrates that the forward and reverse diffusion steps are simply a SPLIT of a single step of Langevin dynamics.
Langevin dynamics is an IDENTITY map on distributions, the forward and reverse diffusion steps are INVERSES of each other at the distribution level.