Fast PAC Global Optimization via Restarted Langevin: Exploration, Exploitation, and Degenerate Cooling

Abstract

We study the computational effort required for global optimization of a smooth, possibly nonconvex objective $Γ:\mathbb{R}^d\to\mathbb{R}$. An algorithm satisfies the $(\varepsilon,δ)$-PAC performance requirement if its output $\widehat X$ obeys $\mathbb{P}\{Γ(\widehat X)-Γ^\star>\varepsilon\}\leqδ$. Algorithm design and analysis are in continuous time. We compare classical simulated annealing and fixed-temperature Langevin diffusion with two approaches introduced and analyzed here: parallel-restart Langevin and a Langevin--gradient scheme using stochastic dynamics for global exploration and gradient flow for local exploitation. Let $L=\log(1/δ)$ and let $E_*$ denote the dominant energy barrier. At logarithmic precision in the low-temperature regime, the first two approaches require simulation time exponential in $L/\varepsilon$. For parallel fixed-temperature Langevin, an appropriate number of independent trials gives $ C_3=L^{1+o(1)}/\varepsilon$ as $δ\downarrow0$, for each fixed $\varepsilon>0$. The most substantial improvement comes from separating exploration from exploitation. If $η$ is the attraction margin of a target region containing the global minimizers, a sufficient low-temperature estimate for total simulation time in the best-state Langevin--gradient variant is $ C_4^{(c)}\approx N\exp\{EL/(Nη)\} +O(\log(1/\varepsilon))$, with $E>E_*$. Thus global exploration is decoupled from the requested accuracy. Analysis beyond logarithmic precision reveals dimension-dependent prefactors, while experiments on the six-hump camel and Rastrigin objectives illustrate the benefits of warmer exploration and the usefulness of spectral information for understanding exploration time.