Does quantum annealing outscale classical algorithms on 3D spin glasses?
In plain words
Whether quantum annealers reach good low-energy states faster, as problems grow, than the best ordinary computer methods is still argued. The answer depends on which classical methods are used for the comparison.
Precise statement
For 3D Edwards-Anderson instances with $N$ up to about 5000 spins, compare the scaling with $N$ and with the target residual energy per spin $\epsilon$ of the time to solution of coherent quantum annealing against named classical baselines: simulated annealing, parallel tempering, and tensor-network and variational Monte Carlo emulation of the annealing dynamics. An answer is a pair of measured scaling exponents with uncertainties for each baseline on fixed instance families with matched resources.
What would settle it
A benchmark with fixed instance families, matched resources and optimized classical baselines, reporting scaling exponents over at least a decade in N.
Status in the literature
Unverified note
The beyond-classical dynamics claim of King et al. (Science 2025) was challenged by tensor-network (Tindall et al., arXiv:2503.05693) and variational Monte Carlo (Mauron and Carleo, arXiv:2503.08247) simulations; the approximate-optimization scaling advantage of Munoz-Bauza and Lidar (PRL 2025) concerns 2D hardware-graph instances, not 3D.