Pattern-recalling processes in quantum Hopfield networks far from saturation

As a mathematical model of associative memories, the Hopfield model was now well-established and a lot of studies to reveal the pattern-recalling process have been done from various different approaches. As well-known, a single neuron is itself an uncertain, noisy unit with a finite unnegligible error in the input-output relation. To model the situation artificially, a kind of ‘heat bath’ that surrounds neurons is introduced. The heat bath, which is a source of noise, is specified by the ’temperature’. Several studies concerning the pattern-recalling processes of the Hopfield model governed by the Glauber-dynamics at finite temperature were already reported. However, we might extend the ’thermal noise’ to the quantum-mechanical variant. In this paper, in terms of the stochastic process of quantum-mechanical Markov chain Monte Carlo method (the quantum MCMC), we analytically derive macroscopically deterministic equations of order parameters such as ‘overlap’ in a quantum-mechanical variant of the Hopfield neural networks (let us call “quantum Hopfield model” or “quantum Hopfield networks”). For the case in which non-extensive number $p$ of patterns are embedded via asymmetric Hebbian connections, namely, $p/N \to 0$ for the number of neuron $N \to \infty$ (‘far from saturation’), we evaluate the recalling processes for one of the built-in patterns under the influence of quantum-mechanical noise.

💡 Research Summary

The paper investigates associative memory in the form of a Hopfield network that is extended to include quantum-mechanical noise, specifically a transverse field that induces quantum fluctuations. The authors focus on the regime where the number of stored patterns p is much smaller than the number of neurons N (p/N → 0), i.e., the network is far from saturation, and the Hebbian connections may be asymmetric.

First, the classical Hopfield model is reviewed: binary neurons S_i = ±1 are fully connected, and the synaptic matrix is built from stored patterns ξ^μ using a Hebbian rule J_ij = (1/N)∑{μ,ν} ξ_i^μ A{μν} ξ_j^ν. In the “far‑from‑saturation” limit the cross‑talk term is negligible, simplifying the analysis.

The quantum extension replaces each spin by a Pauli operator σ_i^z and adds a transverse field term –Γ∑_i σ_i^x to the Hamiltonian H = –∑_i φ_i σ_i^z – Γ∑_i σ_i^x, where φ_i contains the pattern information. This term creates off‑diagonal matrix elements that generate quantum transitions between classical energy minima.

Because directly solving the Schrödinger dynamics for large N is infeasible, the authors employ a quantum Monte Carlo (QMC) approach based on the Suzuki‑Trotter decomposition. The partition function exp(–βH) is expressed as a product of M “Trotter slices”, each representing a replica of the classical spin system at an imaginary‑time step. The transverse field is encoded in a coupling parameter B = (1/2) log coth(βΓ/M) that links neighboring slices.

From the Trotter‑mapped system they write a master equation for the probability p_t({σ_k}) of a microscopic configuration across all slices. The macroscopic order parameter of interest is the overlap m_k = (1/N)∑_i ξ_i^ν σ_i(k) on slice k. By transforming the master equation into a Fokker‑Planck‑like form for the joint distribution P_t(m_1,…,m_M) and then assuming a static approximation (the microscopic distribution becomes time‑independent), they are able to perform the average over the microscopic degrees of freedom analytically. This yields a set of deterministic differential equations for the overlaps:

 dm_ν,l/dt = – m_ν,l + ⟨ξ^ν h_σ(ξ^ν)⟩_path

where the average is taken over a path integral that represents the effective single‑neuron problem along the imaginary‑time direction. The term h_σ(ξ^ν) encapsulates the influence of the transverse field and depends on the stored pattern ξ^ν.

The authors then apply this general result to a concrete scenario: two stored patterns are connected via an asymmetric Hebbian matrix, and the network is asked to retrieve them sequentially. By varying the transverse field strength Γ, they compare the quantum‑noise‑driven dynamics with the classical thermal‑noise case (finite temperature T, Γ = 0). Numerical integration of the deterministic flow shows that larger Γ accelerates the decay of the overlap, introduces oscillatory transients, and reduces the basin of attraction for the target pattern. In contrast, thermal noise primarily broadens the basin without causing such oscillations.

The paper concludes that the combination of quantum Monte Carlo, Suzuki‑Trotter mapping, and static approximation provides a tractable analytical framework for the macroscopic dynamics of quantum Hopfield networks. It highlights that quantum fluctuations affect the recall process more dramatically than classical thermal fluctuations, especially in the low‑load regime. Limitations include the reliance on the static approximation (which neglects time‑dependent correlations), the need for large Trotter numbers M to approach the continuous‑time limit, and the lack of a detailed analysis of storage capacity under quantum noise. Future directions suggested are dynamic mean‑field extensions, capacity studies for higher loads, and exploration of implementation on actual quantum hardware.