Quantum Simulations Become Far More Accurate with New Error Correction

A new method for simulating quantum systems reliably delivers accurate results under specific conditions. Siddhant Midha and colleagues at Princeton University, building upon a recent framework for analysing tensor networks, rigorously demonstrate that belief propagation, a scalable technique for tackling complex quantum many-body problems, accurately approximates local observables in certain scenarios. The work reveals a clear connection between the decay of quantum correlations and the success of belief propagation, proving that exponential decay of correlations is both a necessary and sufficient condition for the method’s validity. Through analytical derivations and numerical simulations of the transverse field Ising model, the team pinpoint the limits of belief propagation, showing its effectiveness in gapped phases, but its failure near critical points, offering key guidance for its application in future quantum simulations.

Exponentially accurate PEPS simulations enabled by loop-decay and cluster-corrected belief

Error rates in approximating local observables within projected entangled-pair states (PEPS) have fallen to exponentially small relative error, a significant improvement over previous heuristic methods. This accuracy is guaranteed by cluster corrections to belief propagation (BP), unlocking quantitative precision in simulating gapped quantum systems. The presence of “loop-decay”, quantifying how quickly quantum correlations diminish, directly dictates whether BP can reliably predict system behaviour; crucially, exponential decay of correlations is necessary for BP’s success or failure.

Verification of these findings used the transverse field Ising model in both two and three dimensions, achieving quantitative accuracy when simulating gapped phases, states where an energy gap exists between the ground state and excited states. Establishing a direct connection between loop tensors, components of the PEPS calculation, and physical correlation functions revealed how quantum entanglement propagates within the system. Consistently matching analytical predictions at zero and finite temperatures demonstrates the reliability of this new approach. However, the method’s applicability remains unproven to critical points or gapless phases, where correlations decay more slowly; bridging this gap is essential for simulating a wider range of quantum materials.

Loop tensors as carriers of connected correlations in PEPS belief propagation

Rigorous criteria for the applicability of belief propagation (BP) to quantum states represented as projected entangled-pair states (PEPS) on arbitrary geometries have been established. Exact formulas expressing local expectation values as the BP prediction dressed by all connected clusters intersecting the observable region were derived. When loop corrections decay exponentially, truncating this expansion at finite cluster order yields a relative-error approximation with an error decaying exponentially fast with the cluster order.

Consequently, BP transforms from a heuristic into a systematically improvable algorithm with rigorous performance guarantees. The magnitude of loop corrections intimately ties to the performance of BP-based tensor network methods, raising the question of their physical meaning and what they reveal about the underlying quantum state. Loop tensors act as the “carriers” of connected correlation functions, generalizing the matrix product state “transfer-matrix” picture to arbitrary graphs.

Applying a cluster expansion to connected correlation functions demonstrates that decay of loopy excitations necessarily implies exponential decay of connected correlations. Therefore, loop decay is not merely an algorithmic criterion but a physical statement about the state itself. Conversely, at critical points or in gapless phases, where correlations decay sub-exponentially, loop corrections must remain parametrically large, establishing loop decay as a sharp diagnostic for when BP-based expansions can succeed and delineating the fundamental limits of such methods.

Their results also provide concrete operational criteria for practitioners, measuring loop decay in the tensor network to indicate whether a given fixed-point can be trusted, the order of cluster corrections needed for a target accuracy, and criteria to ensure clustering of correlations. Failure of loop decay signals either criticality or a mismatch between the chosen BP fixed point and the physical state, potentially requiring expansion around a distinct, potentially “unstable,” fixed point. Examples were provided where loop decay is violated at a stable fixed point, yet restored when the expansion is centred on the appropriate unstable one, emphasizing that expansions must assume a good BP fixed point.

Good fixed points may exist but may be unattainable from standard iterative message passing, particularly near criticality, highlighting a crucial algorithmic challenge beyond convergence of the cluster expansion: the fixed-point problem. A class of PEPS states for which the fixed-point problem can be rigorously solved was provided. Extensive numerical simulations of the two- and three-dimensional transverse field Ising model at zero and finite temperature, represented as a translationally-invariant PEPS, validated their analytical predictions.

Deep in gapped phases, cluster corrections converge rapidly and achieve high quantitative accuracy. Approaching the critical point, loop decay weakens and corrections grow, demonstrating the systematic failure of the method at criticality as predicted by the theory. The paper is organized as follows: Section II reviews the formalisms of tensor networks, belief propagation, and various expansions with convergence guarantees. Section III studies local expectation values.

Section IV outlines the connection between loop corrections and correlators. Section V demonstrates the cluster-corrected BP algorithm numerically on the transverse field Ising model. Section VI concludes with a discussion of open questions and future work. Researchers consider closed tensor networks defined on a graph G = (V, E) with N vertices V and edges E. For each vertex v ∈V, its neighbours are denoted as N(v). The degree of vertex v is d(v) := |N(v)|, and the maximum degree of the graph is ∆:= maxv∈V d(v). Each edge (v, w) ∈E is associated with a bond Hilbert space Bvw of dimension D, taken uniform for simplicity.

Each vertex v ∈V is equipped with a tensor Tv ∈N∈N (v) Bnv making up the tensor network. Contracting all tensors in the network yields a scalar: Z = ⋆v∈V Tv, where ⋆denotes contraction of tensor indices. This Z is called the partition function of the tensor network, more general than in statistical mechanics as tensors Tv can be complex-valued. The free energy is defined as F = −log Z, with all logarithms base e, and normalizations are chosen such that Z is positive, storing phase information separately to ensure the uniqueness of F in the regime where cluster expansions are well-defined.

The starting point for expansions is the belief propagation (BP) approximation, which approximates the tensor network by a rank-one factorization. For each edge e = (v, w) in the network, message tensors μv→w ∈Bvw (and μw→v) are introduced, satisfying a self-consistency condition. A fixed-point set M = {μv→w}(v,w)∈E requires that the contraction of all but one incoming message at any vertex v ∈V reproduces the outgoing message on the excluded edge. Mathematically, for each v ∈V and each neighbour nj ∈N(v): O ni∈N(v){nj} μni→v ⋆Tv ∝μv→nj. The self-consistency condition ensures that this rank-one subspace is invariant under contraction with surrounding tensors, making it a stationary point of a variational optimisation problem.

Belief propagation (BP) offers a scalable heuristic for contracting tensor networks on loopy graphs, though its success in quantum many-body systems has primarily relied on empirical evidence. Recent work builds upon a cluster-expansion framework for tensor networks to rigorously examine BP’s applicability to many-body quantum systems. For a state represented as a projected entangled pair state (PEPS) satisfying a “loop-decay” condition, BP supplemented by cluster corrections approximates local observables with exponentially small relative error, and local expectation values can be expressed as BP predictions adjusted by connected clusters intersecting the observable region.

This representation links cluster corrections to physical correlation functions. As a result, “loop-decay” necessarily implies exponential decay of connected correlations, providing clear criteria for when BP can succeed and identifying its limits at critical points. Numerical simulations of the two- and three-dimensional transverse field Ising model at zero and finite temperature support these analytical predictions, demonstrating quantitative accuracy in gapped phases and systematic failure near criticality.

Tensor networks have become a key tool for understanding quantum many-body systems, naturally encoding gapped ground states and enabling efficient descriptions of short-time dynamics. By encoding correlations through local virtual bonds, tensor networks efficiently capture area-law states and allow controlled approximations of their properties. Tensor networks on a tree admit exact and efficient contraction algorithms, as the absence of loops allows recursive organization with bounded intermediate tensor dimensions.

However, contraction in higher dimensions with unavoidable loops can be computationally demanding, potentially exhibiting exponential scaling with system size. Practical algorithms like corner transfer matrix renormalization group or tensor network renormalization exploit structural properties to introduce controlled truncations and avoid exponential scaling, justified by sufficiently fast decay of correlations. Belief propagation is a particularly promising approach, providing a tree-like approximation computable at polynomial cost, but its accuracy depends on the influence of loops in the network.

Quantum model accuracy is linked to the decay of particle correlations

A technique for modelling quantum systems has been refined, achieving unprecedented accuracy in calculating their properties. This advance tackles a longstanding problem: verifying the reliability of approximations used to simplify complex quantum calculations, particularly when dealing with many interacting particles. The success of this technique hinges on a condition called “loop-decay”, which describes how quickly quantum correlations fade within the system; states lacking this property present a significant challenge.

This condition highlights a clear limitation; the technique will struggle at critical points where correlations persist across the system. However, the work delivers a vital analytical framework for understanding when this approach succeeds, offering precise criteria for reliable quantum calculations. Researchers have established precise conditions for when belief propagation delivers accurate results. This method efficiently simulates complex quantum systems, but relies on rapidly fading connections between particles, termed “loop-decay”. Rigorous criteria now exist to determine when this key quantum simulation technique delivers dependable results. Supplementing belief propagation with cluster corrections guarantees exponentially small errors in calculating local properties of quantum states, a level of precision previously unattainable. This work establishes that rapidly fading quantum connections, termed “loop-decay”, are not merely a technical requirement for the method’s success, but a fundamental property of the quantum state itself; loop-decay directly implies the exponential decay of correlations.

The research demonstrated that belief propagation, a technique for modelling quantum systems, accurately approximates local observables when the quantum state satisfies a “loop-decay” condition. This means the method’s success is directly linked to how quickly correlations fade within the system, offering a rigorous explanation for its limitations at critical points where correlations persist. By combining belief propagation with cluster corrections, researchers achieved exponentially small errors in calculating local properties. These findings establish a clear connection between the decay of quantum correlations and the reliability of this computational technique for many-body quantum systems.