Computational hardness of sampling beyond the shattering phase transition
Demonstrate the computational intractability of sampling from the Gibbs measure of mixed p-spin spherical spin glasses beyond the shattering (dynamical) phase transition, i.e., prove that sampling is fundamentally hard in the regime where the dynamical phase transition predicts slow dynamics.
References
A bolder version of the dynamical phase transition conjecture postulates that not only Langevin dynamics is slow, but indeed sampling is fundamentally hard beyond the shattering phase transition.
We leave open a proof of hardness for stable samplers throughout \beta > \beta_{dyn}(k, D).
For the dense k-spin model, no stable algorithms can efficiently Gibbs sample when \beta > \beta_{dyn}(k), where
It is believed that Markov chain-based methods should fail to sample from the Gibbs distribution when the inverse temperature exceeds the shattering threshold $\beta_{\shatter}$, and the best sampling algorithms work up to a constant of the threshold . On the other hand, an algorithm for computing the partition function was given well beyond $\beta_{\shatter}$ up to the ``replica symmetric threshold'' in a work of , which suggests that spin glass models may exhibit phases where sampling is hard, but computing the partition function is tractable. However, this story was complicated by the recent work , which uses a quantum-inspired algorithm to sample in the shattered phase. We consider it an interesting direction to understand what the true algorithmic threshold is for sampling, as well as what the threshold is for Markov chains.