Unraveling the origin of higher success probabilities in quantum versus semi-classical annealing

TL;DR

This paper demonstrates that full quantum models of annealing outperform semi-classical approximations in finding optimal solutions, highlighting the importance of entanglement and quantum effects in complex optimization tasks.

Contribution

The study provides evidence that quantum annealing can significantly surpass semi-classical methods in success probability and efficiency, especially in complex bosonic systems with long-range interactions.

Findings

Quantum models achieve higher success probabilities than semi-classical approximations.

Entanglement plays a crucial role in reaching near-optimal solutions.

Reduced Hilbert spaces can sometimes improve short-time simulation success.

Abstract

Quantum annealing aims at finding optimal solutions to complex optimization problems using a suitable quantum many body Hamiltonian encoding the solution in its ground state. To find the solution one typically evolves the ground state of a soluble initial Hamiltonian adiabatically to the ground state of the designated final Hamiltonian. Here we explore whether and when a full quantum representation of the dynamics leads to higher probability to end up in the desired ground when compared to a classical mean field approximation. As simple, nontrivial example we target the ground state of interacting bosons trapped in a tight binding lattice with small local defect by turning on long range interactions. Already two atoms in four sites interacting via two cavity modes prove complex enough to exhibit significant differences between the full quantum model and a mean field approximation for the cavity fields mediating the interactions. We find a large parameter region of highly successful quantum annealing where the semi-classical approach largely fails. Here we see strong evidence for the importance of entanglement to end close to the optimal solution. The quantum model also reduces the minimal time for a high target occupation probability. In contrast to naive expectations that enlarging the Hilbert space is beneficial, different numerical cut-offs of the Hilbert space reveal an improved performance for lower cut-offs, i.e. an nonphysical reduced Hilbert space, for short simulation times. Hence a less faithful representation of the full quantum dynamics sometimes creates a higher numerical success probability in shorter time. However, a sufficiently high cut-off proves relevant to obtain near perfect fidelity for long simulations times in a single run. Overall our results exhibit a clear improvement based on a quantum model versus simulations based on a classical field approximation.