Quantum Circuits for the Metropolis-Hastings Algorithm

Szegedy’s quantization of a reversible Markov chain provides a quantum walk whose spectral gap is quadratically larger than that of the classical walk. Quantum computers are therefore expected to provide a speedup of Metropolis-Hastings (MH) simulations. Existing generic methods to implement the quantum walk require coherently computing the transition probabilities of the underlying Markov kernel. However, reversible computing methods require a number of qubits that scales with the complexity of the computation. This overhead is undesirable in near-term fault-tolerant quantum computing, where few logical qubits are available. In this work, we present a Szegedy quantum walk construction which follows the classical proposal-acceptance logic, and does not require further reversible computing methods. We also compare this construction with an alternative to Szegedy’s approach which also provides a quadratic gap amplification. Since each step of the quantum walks uses a constant number of proposal and acceptance steps, we expect the end-to-end quadratic speedup to hold for MH Markov Chain Monte-Carlo simulations.

💡 Research Summary

The paper addresses a long‑standing obstacle in quantum‑accelerated Markov‑Chain Monte‑Carlo (MCMC) methods: the costly coherent computation of transition probabilities required by generic Szegedy quantum walks. While Szegedy’s construction guarantees a quadratic improvement of the spectral gap for any reversible Markov chain, implementing it for the Metropolis‑Hastings (MH) algorithm traditionally demands reversible arithmetic that scales with the complexity of the acceptance probability matrix, leading to a prohibitive number of ancilla qubits.

To overcome this, the authors introduce two key innovations. First, they define a “dual kernel” that operates not on the original state space but on its edges, i.e., ordered pairs of states ((x,y)). In this enlarged space a single step consists of (i) proposing a new head for the current edge using a proposal oracle (O_T) and (ii) deciding whether to flip the edge using an acceptance oracle (O_A). Because the dynamics never discards a sample, the step is inherently reversible and can be implemented with a constant number of oracle calls, eliminating the need to compute rejection probabilities coherently.

Second, they design concrete quantum circuits that realize the Szegedy walk on this dual kernel with only four state registers and three ancilla qubits. Each state register stores a basis label using binary encoding, requiring (m=\lceil\log_2 n\rceil) qubits for a problem of size (n). Consequently the total logical qubit count is (4m+3), a dramatic reduction compared with prior approaches that need (O(\text{poly}(n))) ancillas. The authors illustrate the construction on the Metropolis‑Adjusted Langevin Algorithm (MALA), a continuous‑space MH variant widely used in molecular dynamics and machine learning. By discretizing each coordinate, the circuit scales linearly with the number of atoms ((12N) coordinate registers) plus three ancillas, making it realistic for near‑term fault‑tolerant devices.

Theoretical analysis shows that the dual kernel shares the same spectral gap (\delta) as the original MH kernel. By the standard Szegedy result, the quantum walk operator (W=(2\Pi- I)U) (where (\Pi) projects onto the subspace defined by the oracles and (U) combines the proposal and acceptance steps) has a gap of order (\Omega(\sqrt{\delta})). Numerical simulations of the MALA walk confirm that the implemented unitary’s spectrum respects the theoretical lower bound (\cos^{-1}!\big(\sqrt{1-\delta/2}\big)), demonstrating the expected quadratic gap amplification.

A detailed comparison with earlier works is provided. Childs et al. (2020) implement a quantum walk for MALA using reversible arithmetic, requiring (pd) qubits for each of the (p) coordinates discretized with (d) bits, plus several auxiliary registers. Lemieux et al. (2021) construct isometric circuits for Ising‑type models, needing a unary “move” register of size (N). In contrast, the present method uses binary encoding and a fixed three‑ancilla overhead, as summarized in Table I, achieving the lowest logical qubit count among known constructions.

The authors release an open‑source codebase that automates the construction of the required oracles and assembles the full walk circuit for arbitrary proposal and acceptance functions. This makes the approach immediately applicable to a broad class of MH‑based simulations, from Bayesian inference to statistical‑physics models. The paper concludes by highlighting future directions: extending the dual‑kernel technique to non‑reversible chains, integrating quantum error‑correction schemes, and exploring multi‑walk parallelism to further reduce wall‑clock time. Overall, the work demonstrates that a fully quantum‑accelerated Metropolis‑Hastings algorithm can be realized with a constant, modest qubit budget, preserving the theoretical quadratic speedup and bringing practical quantum MCMC within reach of near‑term fault‑tolerant quantum computers.