🔢 mathematics

Modified logarithmic Sobolev inequality for 1D non-commuting Hamiltonians

This paper establishes the first modified logarithmic Sobolev inequality for finite-range, non-commuting Hamiltonians on 1D chains, proving that heat-bath dynamics and related Gibbs samplers achieve rapid mixing with times of order $\mathcal{O}(\log n \cdot \log(n/\epsilon))$ by combining weak quasi-factorization, decay-of-correlation estimates, and a uniform conditional local gap.

Original authors: Ángela Capel, David Pérez-García, Matteo Scandi

Published 2026-10-05

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Ángela Capel, David Pérez-García, Matteo Scandi

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a long line of tiny magnets, each one capable of pointing in different directions, linked together so that the state of one influences its neighbors. In the quantum world, these magnets do not just sit still; they fluctuate, interact, and constantly exchange energy with their surroundings. When such a system is left alone for a long time, it naturally settles into a state of thermal equilibrium, a calm balance where the energy is distributed according to the temperature of the environment. This final state is known as the Gibbs state. Scientists are deeply interested in how quickly a quantum system reaches this calm state, because the speed of this process determines how efficiently we can prepare quantum computers or simulate complex materials. If the system takes too long to settle, it is useless for practical tasks; if it settles quickly, it is a powerful tool.

The central question is how this settling time changes as the system grows larger. Does it take a little longer, or does the time explode, making large systems impossible to manage? For systems where the interactions between magnets are simple and commute—meaning the order in which they act does not matter—scientists already knew the answer: the system settles very quickly, with the time growing only as a small power of the logarithm of the system size. However, for the more complex and realistic case where the interactions do not commute, a rigorous proof of this rapid settling was missing. Without this proof, it remained an open question whether these more complicated quantum chains would eventually get stuck in a slow, sluggish state as they grew larger.

A team of researchers has now provided the first rigorous proof that even in this difficult, non-commuting regime, a one-dimensional chain of quantum spins settles into equilibrium rapidly. They studied a specific method for simulating this thermalization, called the heat-bath dynamics, where the system is constantly updated by discarding a single spin and replacing it with a fresh one drawn from the correct thermal distribution. They also analyzed a slightly modified version of this process. Their main finding is that the time required for the system to reach equilibrium grows only as the square of the logarithm of the number of spins. This means that even if the chain becomes very long, the time to settle does not explode; it remains manageable and efficient.

To reach this conclusion, the researchers had to overcome a significant mathematical hurdle. In the quantum world, the state of a small part of the chain is not simply a piece of the whole; it is entangled with the rest in ways that make standard mathematical tools fail. The researchers developed a new strategy to break the problem down. They divided the long chain into many overlapping short segments. For each segment, they showed that the system's "disorder," measured by a quantity called relative entropy, decreases rapidly if the local updates are working correctly. They proved that the influence of the rest of the chain on any given segment is limited and decays quickly with distance, a property known as locality.

By carefully stitching these local results together, they demonstrated that the global disorder of the entire chain is controlled by the sum of the local improvements. However, this stitching process introduced a small, constant error that could not be eliminated by local analysis alone. To fix this, the researchers used a global property of the system: the fact that the entire chain has a spectral gap, a measure of how hard it is to excite the system out of its ground state. They showed that this global gap is strong enough to absorb the small error introduced by the local stitching. This allowed them to prove a modified logarithmic Sobolev inequality, a powerful mathematical statement that guarantees the system's disorder decays exponentially fast over time.

The result is a definitive answer to the question of mixing times for these non-commuting quantum chains. The authors proved that the mixing time is bounded by a constant times the square of the logarithm of the system size, multiplied by the logarithm of the desired precision. This holds true regardless of how the interactions vary along the chain, as long as they are short-range and bounded. The proof applies to any positive temperature, meaning it works for both hot and cold environments. This finding confirms that the rapid mixing observed in simpler, commuting systems is not a fluke but a robust feature that extends to the more complex, non-commuting world of quantum interactions.

The work also highlights a subtle distinction between two ways of performing these updates. One method, the heat-bath sampler, updates the system by applying a recovery channel that is quasi-local, meaning its influence fades quickly with distance. The other method, a regularized version, involves taking a limit of repeated updates, which can theoretically act on the entire chain at once. Despite this difference in how they operate, the researchers showed that both methods lead to the same rapid mixing rate. This suggests that the speed of thermalization is a fundamental property of the system's geometry and interactions, rather than a specific artifact of the update rule used.

This proof relies on a combination of several advanced techniques, including estimates of how correlations decay in one-dimensional systems and a new way of comparing the energy of local updates to the entropy of the state. The researchers did not rely on simulations or approximations; they provided a complete mathematical proof that holds for any finite chain of spins. While the logarithmic factor in the mixing time bound might seem like a small inefficiency, it is a vast improvement over exponential growth, which would render large systems unusable. The authors note that for commuting systems, similar bounds have been improved to be independent of system size, and they hope that future work can remove the logarithmic factor for the non-commuting case as well.

For now, this result stands as a major step forward in understanding dissipative quantum dynamics. It assures us that preparing thermal states in one-dimensional quantum systems is a feasible task, even when the interactions are complex and non-commuting. The methods developed in this paper, particularly the way they handle the errors introduced by breaking the system into pieces, may prove useful for analyzing other quantum dynamics where entropy decay is difficult to track. The work closes a significant gap in our theoretical understanding, moving from conjecture to certainty about how quantum chains find their way to equilibrium.

Technical Summary: Modified Logarithmic Sobolev Inequality for 1D Non-Commuting Hamiltonians

1. Problem Statement

The paper addresses the question of how quickly a quantum spin system approaches thermal equilibrium, specifically focusing on the mixing time scaling with system size $n$ for one-dimensional (1D) chains with finite-range, non-commuting Hamiltonians.

While spectral gap analysis provides exponential decay of variance, the resulting mixing time bounds often depend linearly on the system size ( $O(n)$ ) due to the potential exponential growth of the initial distance to equilibrium. To achieve polylogarithmic mixing times ( $O(\text{polylog } n)$ ), one must study the decay of relative entropy via a Modified Logarithmic Sobolev Inequality (MLSI).

Prior to this work, MLSI bounds for 1D systems were established for commuting Hamiltonians (e.g., Davies dynamics, heat-bath samplers) under various mixing assumptions. However, extending these results to the non-commuting regime remained an open challenge. The primary difficulties in the non-commuting setting are:

Marginal Inconsistency: A Gibbs marginal $\sigma_S$ is not necessarily the Gibbs state of the restricted Hamiltonian $H_S$ .
Operator Complexity: Entropy factorization involves logarithms of marginals, while Petz recovery (used in heat-bath updates) involves inverse square roots. Standard correlation estimates do not suffice to control these operator expressions.
Non-Idempotency: Heat-bath updates are not necessarily idempotent, and their fixed algebras can be smaller than the full exterior algebra, complicating the identification of conditional entropy.

2. Model and Dynamics

The authors consider a chain of $n$ spins with local dimension $d=2^q$ and a finite-range Hamiltonian $H = \sum \Phi(X)$ with interaction range $r_0$ and uniform bound $J$ . The system is at a fixed inverse temperature $\beta > 0$ , with Gibbs state $\sigma = e^{-\beta H}/Z$ .

Two specific Gibbs samplers are analyzed:

Heat-Bath Sampler ( $L_{HB}^*$ ): Updates a single site $i$ by discarding it and applying the Petz recovery channel associated with the global Gibbs state $\sigma$ . The generator is $L_{HB}^* = \sum_i (P_{i*} - \text{id})$ .
Regularised Heat-Bath Sampler ( $L_{RHB}^*$ ): Replaces each Petz update $P_{i*}$ with the adjoint $E_{i*}$ of the conditional expectation obtained by iterating $P_i$ to its limit. The generator is $L_{RHB}^* = \sum_i (E_{i*} - \text{id})$ .

Both dynamics preserve $\sigma$ and are primitive. The goal is to prove an MLSI of the form $D(\rho \| \sigma) \leq \alpha^{-1} \text{EP}(\rho)$ , where $\text{EP}$ is the entropy production.

3. Methodology and Proof Structure

The proof combines several advanced techniques from quantum information theory and statistical mechanics to overcome the non-commuting obstacles. The strategy proceeds in four main stages:

A. Weak Quasi-Factorisation of Relative Entropy

The authors decompose the global relative entropy $D(\rho \| \sigma)$ into a sum of conditional relative entropies over an overlapping cover of the chain.

Technique: They construct a cover $\mathcal{G} = \{A_i\} \cup \{B_j\}$ consisting of overlapping intervals of length $O(\log n)$ .
Challenge: In the non-commuting case, the standard quasi-factorisation (which holds exactly or with multiplicative error in commuting cases) incurs an additive error due to the non-triviality of the logarithms of marginals.
Solution: Using matrix-valued conditional mutual information (MCMI) and effective Hamiltonian approximations (based on Araki's expansional estimates), they bound the global entropy by the sum of block conditional entropies plus a bounded additive error term ( $O(1)$ ).

B. Uniform Conditional Gap

To relate the conditional entropy to dissipation, a spectral gap is required.

Technique: The authors utilize the canonical purified Hamiltonian introduced in [33]. They show that this Hamiltonian possesses a uniform spectral gap independent of system size for 1D systems.
Application: Through the Kadison–Schwarz inequality and modular averaging (integrating over modular twists), they transfer this gap to a uniform conditional gap for the heat-bath sampler. This establishes a lower bound on the Dirichlet form in terms of the conditional variance.

C. Locality of Exterior Changes and Buffered Energy

A critical step is comparing the conditional entropy of a block to the quadratic energy of updates in a slightly larger "buffered" region.

Locality: Using Araki's expansionals and locality estimates for marginal logarithms, they prove that changes to the exterior of a buffered region $S[b]$ have an exponentially decaying effect on the Petz update within $S$ .
Recovery: They show that the Petz map approximately fixes states with amplitudes supported outside the buffer.
Result: This allows them to bound the conditional entropy $D_S(\rho \| \sigma)$ by the quadratic energy $Q_{S[b]}(\rho)$ of updates in the buffered region, plus a small error ( $O(1/n)$ ).

D. Removal of the Bounded Entropy Defect

The combination of the quasi-factorisation and the buffered energy estimate yields an inequality of the form:
$D(\rho \| \sigma) \leq C \log n \cdot Q_\Lambda(\rho) + B$
where $B$ is a bounded additive defect independent of $n$ .

Resolution: The authors employ a "norm-to-entropy" method (inspired by [26]). They utilize the global KMS spectral gap $\gamma > 0$ to remove the additive defect $B$ . By interpolating between norm contraction and entropy decay, they convert the defective bound into a pure MLSI.

4. Key Results

Main Theorem (Theorem 1.2)

For both the heat-bath and regularised heat-bath samplers acting on a 1D finite-range non-commuting Hamiltonian at any fixed positive temperature $\beta$ , there exists a constant $c_\beta$ (independent of $n$ ) such that:
$\alpha_{\text{MLSI}}(L^*) \geq \frac{c_\beta}{\log n}$
Consequently, the relative entropy decays as:
$D(e^{tL^*}\rho \| \sigma) \leq e^{-c_\beta t / \log n} D(\rho \| \sigma)$

Mixing Times (Corollary 1.3)

The mixing times in relative entropy and trace norm satisfy:
$t_{\text{mix}}(\epsilon) \leq C'_\beta \log n \cdot \log(n/\epsilon)$
This represents a polylogarithmic mixing time ( $O(\log^2 n)$ for fixed accuracy) without requiring a high-temperature assumption.

Additional Contributions

Complete Inequalities: The proof establishes complete entropy inequalities for individual updates, valid in the presence of arbitrary untouched quantum memory.
Quasi-locality: It is proven that the heat-bath sampler is a quasi-local Lindbladian with exponentially decaying tails, whereas the regularised sampler (being a limit of iterations) may lose this strict locality property.

5. Significance and Claims

The paper claims to provide the first proof of an MLSI in the non-commuting regime for 1D systems.

Novelty: Previous results were limited to commuting Hamiltonians or required specific mixing assumptions. This work demonstrates that strong equilibrium locality (a feature of 1D systems) persists even when interactions do not commute, sufficient to guarantee rapid mixing.
Optimality of the Bound: The authors note that the $1/\log n$ factor arises from the choice of block sizes $O(\log n)$ used to control accumulated errors in the quasi-factorisation and buffered estimates. They explicitly state that they do not claim this logarithmic loss is intrinsic to the sampler. They draw a parallel to commuting chains, where $1/\log n$ bounds were later improved to $O(1)$ constants via different strategies, suggesting a similar improvement might be possible here.
Limitations: The authors modestly note that extending this result to higher dimensions remains an open problem, primarily due to the lack of a suitable multipartite quasi-factorisation estimate with controlled additive errors in dimensions $d > 1$ . Additionally, while local complete inequalities are proven, a global Complete MLSI (CMLSI) with the same $1/\log n$ rate is not established due to difficulties in scalar centering with memory.

In summary, the paper establishes a rigorous link between 1D equilibrium correlations and entropy dissipation for non-commuting systems, proving that thermalization occurs rapidly (polylogarithmically) for heat-bath dynamics, thereby advancing the theoretical understanding of Gibbs state preparation in quantum many-body systems.