Irreversibility Enhances Quantum-Enhanced Markov-Chain Monte Carlo
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Condensed Matter. The day in condensed matter: materials, phases and the measurements under them.
Kai: We're looking at the paper: "Irreversibility Enhances Quantum-Enhanced Markov-Chain Monte Carlo".
Title and authors: Kai: So, we're looking at this paper now, "Irreversibility Enhances Quantum-Enhanced Markov-Chain Monte Carlo," and it’s really about changing how we sample from these quantum systems. It introduces state-dependent proposals that break detailed balance while still keeping the target stationary distribution correct.
Mira: Exactly, and what's interesting is they tie this to Landau–Zener transitions. They use this to promote big energy descents when you’re in high-energy states, but they keep local moves for low-energy states where stability matters.
Lev: From my side, the key thing here is that they break detailed balance intentionally, which means the probability of moving from one state to another isn't just reversible; it's directional based on the energy you start with. This is a major departure from standard MCMC methods like QEMC <ref:2606.23350#pg1>.
Kai: So, what does this mean in practice for the sampling process? How does this break detailed balance actually help us explore the energy landscape better?
Mira: It means they construct a quantum circuit based on your initial energy, E of that state, which is computed once. The proposal distribution then changes depending on that initial energy and how fast you sweep through the system.
Lev: It’s about using this to make proposals adapt based on the state's energy to get those large delta E transitions for high energies and keep the small delta E transitions for low energies, as shown in Figure 2c and 2d <ref:2606.23350#pg2>.
Kai: That sounds like a targeted way to explore the landscape—not just random walks across it. So, how does this compare to standard QEMC we've seen?
Mira: The paper shows that IQEMC outperforms QEMC without increasing the computational complexity on spin-glass benchmarks <ref:2606.23350#pg2>. It maintains the polynomial speedup of QEMC, but it enlarges the spectral gap.
Lev: Enlarging the spectral gap is crucial because it directly relates to how fast you converge, which means better mixing times and less autocorrelation time <ref:2606.23350#pg2>. The improvement over standard annealing also gets stronger as the system size and driving strength increase.
Kai: So we have a method that's faster than both standard QEMC and classical annealing on these spin-glass problems, and it keeps the spectral gap larger, which is good for convergence.
Title and authors: Mira: It suggests that irreversibility isn't just some arbitrary trick; it’s a physically grounded mechanism for enhancing quantum MCMC <ref:2606.23350#pg2>. This ties the acceleration directly to the physics of Landau–Zener transitions <ref:2606.23350#pg1>.
Lev: It also means we can use non-detailed balance dynamics as a practical resource instead of needing more complex things like neural network surrogates or just running a long annealing schedule <ref:2606.23350#pg1>. That's something we need for real hardware constraints.
Kai: That’s interesting about using nonequilibrium driving as a resource instead of needing intensive primitives. So, what are the main practical advantages they highlight?
Mira: The authors demonstrate that IQEMC achieves better overall autocorrelation performance by suppressing long-time correlations <ref:2606.23350#pg2>. They show this is true for various random Ising models and even in the fully connected random Ising model where it shows a constant factor larger average spectral gap delta than QEMC for all system sizes n <ref:2606.23350#pg2>.
Lev: The convergence diagnostics really back this up; they show IQEMC achieves the lowest integrated autocorrelation time, tauint mean, for system sizes ranging from three to five and also for eight to ten <ref:2606.23350#pg2>. That’s a concrete measure of how efficient the sampling is globally.
Kai: So we're talking about better convergence metrics—lower autocorrelation times—and a better spectral gap when we compare IQEMC against QEMC and classical methods on these benchmarks.
Mira: It also addresses robustness; the spectral gap of IQEMC stays nearly unchanged across all tested pre-sampling numbers, which shows it’s robust to coarse energy range estimates <ref:2606.23350#pg2>. The crossover system size actually decreases when you increase the driving strength C <ref:2606.23350#pg3>.
Lev: The analysis in the thermodynamic limit shows that typical states converge to k of si being approximately C times the square root of two minus one, which implies a specific relationship between the proposal distribution and your initial state probability <ref:2606.23350#pg2>. It’s a very specific physical prediction.
Kai: That leads into how we can actually use this for something tangible, like estimating physical observables in an Ising model. What do they show about accuracy?
Title and authors: Mira: In the strongly magnetized regime, IQEMC converges more rapidly to the theoretical value than both QEMC and uniform proposals <ref:2606.23350#pg1>. They also confirm that numerical residuals for annealing, IQEMC, and QEMC are all at the level of numerical precision <ref:2606.23350#pg2>.
Lev: And they specifically point out that the spectral gap improvement over QEMC in the fully connected random Ising model is fifty-nine point five two percent for n equals ten <ref:2606.23350#pg2>. That number gives a real sense of how much better it is than the standard quantum version.
Kai: It seems like this approach offers a significant concrete advantage in terms of convergence speed and accuracy when compared to the existing methods we use.
Mira: So, to wrap up this discussion on "Irreversibility Enhances Quantum-Enhanced Markov-Chain Monte Carlo," the main implication is that nonequilibrium driving can be a practical resource for quantum sampling without needing very resource-intensive primitives <ref:2606.23350#pg1>.
Lev: I’d just add that the work suggests the optimal design of this irreversibility mechanism is something we need to figure out next to truly maximize quantum sampling efficiency <ref:2606.23350#pg1>.
Kai: So, to summarize, this paper shows IQEMC gives us a polynomial speedup over classical MCMC and a spectral-gap enhancement over QEMC that gets better with system size and driving strength.
Mira: It also provides a mechanism for faster exploration of the energy landscape by using energy-adaptive proposals that favor large delta E transitions at high energies while maintaining local moves at low energies <ref:2606.23350#pg1>.
Lev: And the final result is that IQEMC achieves better overall autocorrelation performance with the lowest tauint mean for system sizes three to five and eight to ten <ref:2606.23350#pg2>.
Kai: That’s what we have on this paper, "Irreversibility Enhances Quantum-Enhanced Markov-Chain Monte Carlo." It’s a solid piece of work showing how breaking detailed balance can be a useful tool in quantum sampling.
Mira: It sets the stage for future work focusing on designing the optimal irreversibility to get even greater gains in efficiency <ref:2606.23350#pg1>.
Lev: We'll keep an eye on how this translates to running it on actual NISQ devices, which is where the next big hurdle is <ref:2606.23350#pg1>.
The paper's summary: Kai: So, to wrap up what we just talked about on this paper, "Irreversibility Enhances Quantum-Enhanced Markov-Chain Monte Carlo," they’re essentially showing how you can break detailed balance in a quantum Monte Carlo simulation.
Mira: Right, it’s about using state-dependent proposals guided by the Landau–Zener effect to make big energy jumps when you need them and keep things stable when you don't.
Lev: The core idea is that instead of just doing random moves, the proposal distribution changes based on how much energy the current state has, which helps accelerate exploration of the energy landscape.
Kai: So what does this mean for those who only listen to our show? It means we’re moving beyond standard quantum enhanced Monte Carlo and getting a method that can favor big jumps across high-energy barriers while still being careful about how it samples low-energy states.
Mira: Exactly, it’s about using non-detailed balance dynamics as a practical resource instead of needing those heavy, resource intensive primitives like neural network surrogates or running long annealing schedules.
Lev: From an engineering standpoint, that’s the big sell; if we can use physical driving—nonequilibrium driving—to speed things up without needing super complex software, that opens up real possibilities for real quantum hardware.
Kai: It seems like this approach offers a concrete way to get a polynomial speedup over classical Markov chain Monte Carlo methods and an enlarged spectral gap compared to the standard quantum enhanced MCMC we’ve seen before.
Mira: That enlarged spectral gap is key because it directly relates to how fast you converge, which means better mixing times and less autocorrelation time in your simulation.
Lev: And they show that this improvement over standard annealing gets stronger as the system size and the driving strength get bigger, which is good for scaling up.
Kai: They also give us some very solid convergence diagnostics showing IQEMC has the best overall autocorrelation performance, achieving the lowest integrated autocorrelation time for certain system sizes.
Mira: That’s a strong signal that this method actually suppresses those long-time correlations we usually worry about in sampling.
Lev: And they confirm that the spectral gap of this method stays pretty stable even across different pre-sampling numbers, which means it’s robust to those coarse energy range estimates you might have.
Kai: So the paper’s main implication is that nonequilibrium driving is a practical resource for quantum sampling without needing very resource intensive primitives or super complicated setup.
Mira: And they point toward designing the optimal irreversibility mechanism as the next big direction if we want to maximize that quantum sampling efficiency further.
Lev: We’ll keep an eye on how this translates to running it on actual NISQ devices, which is where the real test for this method will be.
The paper's improvements: Tom: So, we’re looking at what they actually suggest is better for this method, "Irreversibility Enhances Quantum-Enhanced Markov-Chain Monte Carlo," beyond just the performance numbers.
Kai: They aren't just sticking to QEMC; they’re pointing out that the optimal design of this irreversibility mechanism is something we need to figure out next if we want to get even more efficiency from it.
Mira: That means they aren't finished yet; they’ve found a great starting point but there’s definitely room to tweak how that state-dependent proposal works for maximum sampling speed.
Lev: From an error correction view, that design work is critical because if we can nail the proposal distribution based on energy, it might simplify the error correction overhead we have to deal with when running this on actual hardware.
Kai: It suggests that even if we use a specific driving schedule, the way we set up those energy-dependent proposals is what’s going to make or break how well it samples the distribution.
Mira: Exactly, they’re moving from just having a mechanism to actually designing the best possible mechanism for exploration on these complex energy landscapes.
Lev: If they can find that optimal design, it might allow us to use less aggressive driving schedules, which would be great for real quantum devices because we want to minimize noise and decoherence.
Kai: It sounds like the next step is moving from a general idea of breaking detailed balance to a specific algorithm structure tailored for different physical systems.
Mira: That’s right, the implication is that this isn't just about running *a* method faster; it’s about building a smarter way to sample based on the physics of the problem itself.
Lev: And if we can make that design robust, we might be able to use these nonequilibrium driving concepts in a much broader range of quantum simulations than what they've tested so far.
Conclusion: Kai: So, to wrap up this discussion on "Irreversibility Enhances Quantum-Enhanced Markov-Chain Monte Carlo," we’ve seen how this method uses energy-dependent proposals to break detailed balance and speed up sampling from Boltzmann distributions.
Mira: It really shows that nonequilibrium driving is a practical resource for quantum sampling without needing super resource intensive primitives like neural network surrogates or just running long annealing schedules.
Lev: I think the main thing is that this approach can give us a polynomial speedup over classical Markov chain Monte Carlo methods and an enlarged spectral gap compared to standard quantum enhanced MCMC.
Kai: It means we have a method that's faster than both standard QEMC and classical annealing on those spin-glass benchmarks, and it keeps the spectral gap larger, which is good for convergence.
Mira: And it gives us a mechanism for faster exploration of the energy landscape by using proposals that favor big energy jumps at high energies while keeping local moves at low energies.
Lev: The final result is that IQEMC achieves better overall autocorrelation performance with the lowest integrated autocorrelation time for system sizes three to five and eight to ten.
Kai: That’s a solid piece of work showing how breaking detailed balance can be a useful tool in quantum sampling while still keeping the target stationary distribution correct.
Mira: It sets the stage for future work focusing on designing the optimal irreversibility mechanism to get even greater gains in efficiency.
Lev: We'll keep an eye on how this translates to running it on actual NISQ devices, which is where the real test for this method will be.
Kefan Cao, *, Zidong Cui, *, Lei Wang, *, Ying Tang
Institute of Fundamental and Frontier Sciences, University of Electronic Science and Technology of China, Chengdu 611731, China · Yingcai Honor College, University of Electronic Science and Technology of China, Chengdu 611731, China · School of Physics, University of Electronic Science and Technology of China, Chengdu 611731, China · Beijing National Laboratory for Condensed Matter Physics and Institute of Physics, Chinese Academy of Sciences, Beijing 100190, China · Key Laboratory of Quantum Physics and Photonic Quantum Information, Ministry of Education, University of Electronic Science and Technology of China, Chengdu 611731, China · Non-classical Information Science Basic Discipline Research Center of Sichuan Province, University of Electronic Science and Technology of China, Chengdu 611731
cond-mat.stat-mech, quant-ph
Submitted: 2026-06-22
Updated: 2026-11-18
Comments: 13 pages,7 figures
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 85/100
The gist: Irreversibility provides a new route to improving recent quantum-enhanced MCMC by introducing state-dependent proposals that break detailed balance while preserving the target stationary distribution…
Key concepts
- Irreversibility
- This is the core mechanism where the algorithm intentionally breaks detailed balance. Instead of strictly following standard rules, it introduces non-equilibrium dynamics based on energy levels. This allows the simulation to rapidly escape high-energy states while ensuring stable, efficient exploration near low-energy regions.
- Landau–Zener Transitions
- These are quantum mechanical transitions that occur when a system passes through an avoided crossing in its energy spectrum. The IQEMC algorithm uses these transitions to guide the proposal distribution: it assigns larger sweep rates (k) to high-energy states, facilitating large energy drops, and smaller rates to low-energy states for stable sampling.
- Spectral Gap ($\delta$)
- The spectral gap measures how quickly a Markov chain converges to its stationary distribution. A larger spectral gap indicates faster mixing and more efficient sampling. IQEMC achieves a consistent enhancement of this gap over standard Quantum Enhanced MCMC (QEMC), proving its superior convergence properties.
Terminology
Summary
Irreversibility provides a new route to improving recent quantum-enhanced MCMC by introducing state-dependent proposals that break detailed balance while preserving the target stationary distribution <ref:2606. p. 023350#pg2> The paper develops an irreversible quantum-enhanced Monte Carlo (IQEMC) algorithm guided by Landau–Zener transitions, which promotes large energy descents from high-energy states while maintaining stable transitions near low-energy states <ref:2606. p. 023350#pg2> These results establish irreversibility as a physically grounded mechanism for enhancing quantum MCMC <ref:2606. p. 023350#pg2>
Mechanism of Irreversibility
The IQEMC algorithm constructs quantum proposals based on the Landau-Zener effect and initial-state-dependent dynamics to accelerate QEMC by breaking detailed balance <ref:2606.23350#pg2> The time-dependent Hamiltonian is defined as H(si, t) = [1 − γ(si, t)]αHprob + γ(si, t)Hmix <ref:2606. p. 123350#pg2> where the sweep rate k depends on the energy E(si) of the input state si⟩ <ref:2606. p. 123350#pg2> This dependence is motivated by Landau–Zener transitions, assigning larger k to high-energy states and smaller k to low-energy states <ref:2606. p. 123350#pg2> The proposal distribution exhibits a clear difference between QEMC and IQEMC, favoring large-∆E transitions for high-energy states (Fig. p. 2 2c) while exhibiting local energy descent for low-energy states (Fig. p. 2 2d) <ref:2606. p. 023350#pg2>
Performance Benchmarks
The IQEMC method outperforms QEMC without increasing computational complexity on spin-glass benchmarks <ref:2606.23350#pg2> While retaining the polynomial speedup of QEMC, IQEMC enlarges the spectral gap, with the improvement over annealing becoming more pronounced as the system size and driving strength grow <ref:2606. p. 023350#pg2> The average convergence efficiency is demonstrated by comparing IQEMC against QEMC and classical uniform sampling across various random Ising models <ref:2606. p. 323350#pg2> In the fully connected random Ising model, IQEMC exhibits a constant-factor larger average spectral gap δ than QEMC for all n, and achieves a polynomial advantage over annealing <ref:2606. p. 223350#pg2>
Convergence Diagnostics
The convergence rate of IQEMC is characterized by the spectral gap δ = R(λ2), where λ2 is the second-largest eigenvalue of the stationary-distribution-weighted Markov generator <ref:2606.23350#pg2> The integrated autocorrelation time τint measures global sampling efficiency, and IQEMC achieves more efficient and robust sampling by suppressing long-time correlations <ref:2606. p. 923350#pg2> Numerical evaluations show that IQEMC exhibits the best overall autocorrelation performance, achieving the lowest τint,mean for n = 3 − 5 and n = 8 − 10 <ref:2606.23350#pg2>
Robustness and Scaling
The spectral gap of IQEMC remains nearly unchanged across all tested pre-sampling numbers, demonstrating the robustness to coarse energy-range estimates <ref:2606.23350#pg2> The advantage of IQEMC over pure annealing grows with system size, while the crossover system size decreases with driving strength C <ref:2606. p. 223350#pg2> Furthermore, the acceleration of IQEMC originates from energy-dependent irreversibility rather than from random sweep-rate variations or a generic time-dependent annealing schedule <ref:2606. p. 623350#pg2> The analysis in the thermodynamic limit shows that typical states converge to k(si) ≈ C(√2 − 1), implying Pi(si si) ≈ P(si si) <ref:2606.23350#pg2>
Experimental Directions
The paper suggests several promising directions for future investigation, including the optimal design of irreversibility for maximizing quantum sampling efficiency <ref:2606.23350#pg2> Another direction is the implementation of irreversible quantum MCMC on quantum devices to enable direct experimental testing of irreversibility-enhanced quantum sampling <ref:2606. p. 323350#pg2> The results suggest that nonequilibrium driving is a practical resource for quantum sampling while avoiding resource-intensive primitives <ref:2606. p. 323350#pg2>
Conclusion
IQEMC enables a polynomial speedup over classical MCMC, a consistent spectral-gap enhancement over QEMC, and an advantage over annealing that strengthens with system size and driving strength <ref:2606.23350#pg2> IQEMC identifies nonequilibrium driving as a practical resource for quantum sampling while avoiding resource-intensive primitives <ref:2606. p. 323350#pg2> Building on the experimental realization of QEMC in [22], another direction is the implementation of irreversible quantum MCMC on quantum devices enabling direct experimental testing of irreversibility-enhanced quantum sampling <ref:2606. p. 323350#pg2> The optimal design of irreversibility is identified as a promising direction for maximizing quantum sampling efficiency <ref:2606. p. 323350#pg2>.
Acknowledgments
We thank Luming Duan for discussions <ref:2606.23350#pg2> This work is supported by Projects 12322501, 12575035 of the National Natural Science Foundation of China, and 2026NSFSCZY0124 of the Natural Science Foundation of Sichuan Province <ref:2606. p. 323350#pg2> The computational work is supported by the Center for HPC, University of Electronic Science and Technology of China <ref:2606. p. 323350#pg2>.
Code Availability
The code implementation will be available upon acceptance of the manuscript <ref:2606.23350#pg2>. The state encoding convention is summarized in Supplemental Table S1 <ref:2606. p. 823350#pg2> Algorithm details are presented in Algorithm Details <ref:2606. p. 723350#pg2> The numerical details are described in Numerical Details <ref:2606. p. 423350#pg2> The supplemental experiment results extend the spectral-gap comparison to a broader parameter range and further support the magnetization benchmark <ref:2606. p. 723350#pg2> The numerical verification of global balance confirms that the mixed-column normalization condition is satisfied in IQEMC <ref:2606. p. 523350#pg2> The discussion of the spectral gap for non-reversible chains clarifies that the real part of the subleading eigenvalue governs relaxation <ref:2606. p. 723350#pg2> The relaxation time and mixing time analysis shows that IQEMC achieves more efficient and robust sampling by suppressing long-time correlations <ref:2606. p. 723350#pg2>. The numerical results confirm that the mixed-column normalization condition is satisfied in IQEMC <ref:2606. p. 523350#pg2> The spectral gap improvement over QEMC in the fully connected random Ising model is 59. p. 3 p. 3 p. 352% for n = 10 <ref:2606. p. 0 p. 0 p. 023350#pg2> The magnetization estimation confirms that IQEMC converges more rapidly to the theoretical value than QEMC and uniform proposals in a strongly magnetized regime <ref:2606. p. 323350#pg2> The numerical residuals for annealing, IQEMC, and QEMC are all at the level of numerical precision <ref:2606. p. 423350#pg2>.
References
[1] N. Metropolis, A. p. 0 W. Rosenbluth, M. p. 4 N. Rosenbluth, A. p. 4 H. Teller, and E. p. 4 Teller, Equation of state calculations by fast computing machines, J. p. 4 Chem. Phys 21 1087 (1953) <ref:2606. p. 423350#pg2>
[2] W. K. Hastings, Monte Carlo sampling methods using Markov chains and their applications, Biometrika 57 97 (1970) <ref:2606. p. 423350#pg2>
[3] V. I. Manousiouthakis and M. p. 4 W. Deem, Strict detailed balance is unnecessary in Monte Carlo simulation, J. p. 4 Chem. Phys 110 2753 (1999) <ref:2606. p. 423350#pg2>
[4] Y.-A. Ma, T. Chen, and E. p. 4 B. Fox, A complete recipe for stochastic gradient MCMC, arXiv: 1506. p. 404696 (2015) <ref:2606. p. 423350#pg2>
[5] P. Ao, T.-Q. Chen, and J. p. 4-H. Shi, Dynamical decomposition of Markov processes without detailed balance, Chin. p. 4 Phys Lett 30 070201 (2013) <ref:2606. p. 423350#pg2>
[6] K. S. Turitsyn, M. p. 4 Chertkov, and M. p. 4 Vucelja, Irreversible Monte Carlo algorithms for efficient sampling, Phys. p. 4 D: Nonlinear Phenom 240 410 (2011) <ref:2606. p. 423350#pg2>
[7] E. P. Bernard, W. p. 4 Krauth, and D. p. 4 B. Wilson, Eventchain Monte Carlo algorithms for hard-sphere systems, Phys Rev E 80 056704 (2011) <ref:2606. p. 423350#pg2>
[8] H. Suwa and S. p. 4 Todo, Markov chain Monte Carlo method without detailed balance, Phys Rev Lett 105 120603 (2019) <ref:2606. p. 423350#pg2>
[9] H. Hu, X. Chen, and Y. p. 4 Deng, Irreversible Markov chain Monte Carlo algorithm for self-avoiding walk, Front. p. 4 Phys 12 120503 (2017) <ref:2606. p. 423350#pg2>
[10] F. Ghimenti, L. p. 4 p. 4 Berthier, G. p. 4 Szamel, and F. p. 4 van Wijland, Irreversible Boltzmann samplers in dense liquids: Weakcoupling approximation and mode-coupling theory, Phys Rev E 110 034604 (2024) <ref:2606. p. 423350#pg2>
[11] F. Ghimenti, L. Berthier, and F. p. 4 van Wijland, Irreversible Monte Carlo algorithms for hard disk glasses: From event-chain to collective swaps, Phys Rev Lett 133 028202 (2024) <ref:2606. p. 423350#pg2>
[12] W.-B. Xu, Z.-S. Shen, Y. Tang, and P. p. 4 Zhang, Combinatorial optimization with dual mean-field dynamics, Commun. p. 4 Theor. Phys 77 125604 (2025) <ref:2606. p. 423350#pg2>
[13] A. Ichiki and M. p. 4 Ohzeki, Violation of detailed balance accelerates relaxation, Phys Rev E 88 020101 (2013) <ref:2606. p. 423350#pg2>
[14] M. Szegedy, Quantum speed-up of Markov chain based algorithms, 45th FOCS 32 (2004) <ref:2606. p. 423350#pg2>
[15] P. C. Richter, Quantum speedup of classical mixing processes, Phys Rev A 76 042306 (2007) <ref:2606. p. 423350#pg2>
[16] A. Montanaro, Quantum speedup of Monte Carlo methods Proc. p. 4 R. Soc. A 471 20150301 (2015) <ref:2606. p. 423350#pg2>
[17] J. Lemieux, B. p. 4 Heim, D. Poulin, K. Svore, and M. p. 4 Troyer, Efficient quantum walk circuits for Metropolis-Hastings algorithm Quantum 4 287 (2020) <ref:2606. p. 423350#pg2>
[18] S. Arunachalam, V. p. 4 Havlicek, G. p. 4 Nannicini, K. p. 4 Temme, and P. p. 4 Wocjan, Simpler (classical) and faster (quantum) algorithms for Gibbs partition functions Quantum 6 789 (2022) <ref:2606. p. 423350#pg2>
[19] J. Preskill, Quantum computing in the NISQ era and beyond, Quantum 2 79 (2018) <ref:2606. p. 423350#pg2>
[20] P. Wocjan and A. p. 4 Abeyesinghe, Speedup via quantum sampling Phys Rev A 78 042336 (2008) <ref:2606. p. 423350#pg2>
[21] M.-H. Yung and A. p. 4 Aspuru-Guzik, A quantum–quantum Metropolis algorithm Proc. p. 4 Natl Acad Sci 109 754 (2012) <ref:2606. p. 423350#pg2>
[22] D. Layden, G. Mazzola, R. p. 4 V. Mishmash, M. p. 4 Motta, P. Wocjan, J.-S. Kim, and S. p. 4 Sheldon, Quantumenhanced Markov chain Monte Carlo Nature 619 282 (2023) <ref:2606. p. 423350#pg2>
[23] S. Arai and T. p. 4 Kadowaki, Quantum annealing enhanced Markov-chain Monte Carlo Sci. p. 4 Rep 15 21427 (2025) <ref:2606. p. 423350#pg2>
[24] Y. Nakano, K. N. Okada, and K. p. 4 Fujii, Neural-networkassisted Monte Carlo sampling trained by quantum approximate optimization algorithm PRX Quantum 7 010338 (2026) <ref:2606. p. 423350#pg2>
[25] N. V. Vitanov, Transition times in the Landau-Zener model Phys Rev A 59 988 (1999) <ref:2606. p. 423350#pg2>
[26] C. Wittig, The Landau-Zener formula J. p. 4 Phys Chem B 109 8428 (2005) <ref:2606. p. 423350#pg2>
[27] A. Lucas, Ising formulations of many NP problems Front. p. 4 Phys 2 5 (2014) <ref:2606. p. 023350#pg2>
[28] D. A. Levin and Y. p. 4 Peres, Markov chains and mixing times Vol 107 American Mathematical Soc 2017 <ref:2606. p. 423350#pg2>
[29] A. Orfi and D. p. 4 p. 4 Sels, Bounding the speedup of the quantum-enhanced Markov-chain Monte Carlo Phys Rev A 110 052414 (2024) <ref:2606. p. 423350#pg2>
[30] A. Orfi and D. Sels, Barriers to efficient mixing of quantum-enhanced Markov chains Phys Rev A 110 052434 (2024) <ref:2606. p. 423350#pg2>
[31] C. Zener, Non-adiabatic crossing of energy levels Proc A 137 696 (1932) <ref:2606. p. 423350#pg2>
[32] See Supplemental Material at [url will be inserted by publisher] for details on the numerical simulation, proof of convergence and autocorrelation, discussion on spectral gap of non-reversible chain <ref:2606.23350#pg2>
[34] C. J. Geyer, Markov chain Monte Carlo maximum likelihood IFNA 156 (1991) <ref:2606. p. 423350#pg2>
[35] D. J. Earl and M. p. 4 W. Deem, Parallel tempering: Theory applications and new perspectives Phys Chem Chem Phys 7 3910 (2005) <ref:2606. p. 423350#pg2>
[36] H. Lu and Y. Tang, Steering dynamical regimes of diffusion models by breaking detailed balance arXiv:2602. p. 415914 (2026) <ref:2606. p. 423350#pg2>
Numerical Verification
The numerical verification of global balance confirms that the mixed-column normalization condition is satisfied in IQEMC <ref:2606.23350#pg2> The residuals for annealing, IQEMC, and QEMC are all at the level of numerical precision <ref:2606. p. 023350#pg2> The spectral gap improvement over QEMC in the fully connected random Ising model is 59.52% for n = 10 <ref:2606.23350#pg2> IQEMC converges more rapidly to the theoretical value than QEMC and uniform proposals in a strongly magnetized regime <ref:2606. p. 323350#pg2> The results show that IQEMC exhibits the best overall autocorrelation performance, achieving the lowest τint,mean for n = 3 − 5 and n = 8 − 10 <ref:2606.23350#pg2>
Discussion of Relaxation
The real part of the subleading eigenvalue associated with the stationary distribution-weighted Markov generator governs relaxation and underlies the convergence metric used in the main text <ref:2606.23350#pg2> The decay rate is therefore set by δ = Re(λ2) as defined for non-reversible chains <ref:2606. p. 123350#pg2> This result requires only the balance condition and extends beyond detailed-balance dynamics <ref:2606. p. 023350#pg2> The analysis shows that IQEMC achieves more efficient and robust sampling by suppressing long-time correlations, thereby increasing the effective sample size <ref:2606. p. 923350#pg2> The spectral gap improvement over QEMC in the fully connected random Ising model is 59.52% for n = 10 <ref:2606.
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Irreversible quantum-enhanced Monte Carlo (IQEMC) can be used to accelerate sampling from Boltzmann distributions by breaking detailed balance, which allows for
rapid escape from high energies while preserving efficient low-energy sampling.
-
The improved AI system can achieve a polynomial speedup over classical Markov chain Monte Carlo methods and exhibit an
enlarged spectral gap
compared to standard quantum-enhanced MCMC (QEMC). -
IQEMC provides a mechanism for
faster exploration of the energy landscape,
which is demonstrated by its ability to favorlarge-∆E transitions for high-energy states
while maintaining local energy descent at low energies. -
The system can be designed to utilize
non-detailed balance dynamics
as a practical resource, avoidingresource-intensive primitives
like neural network surrogates or annealing when seeking computational efficiency gains. -
The improved system can estimate physical observables, such as the average magnetization in an Ising model, with higher accuracy because IQEMC exhibits a
spectral-gap improvement of 59.5% in the 10-spin system
over QEMC and annealing.
Sources
- A Complete Recipe for Stochastic Gradient MCMC
- Steering Dynamical Regimes of Diffusion Models by Breaking Detailed Balance
- Quantum computing with Qiskit
Related papers
- Emergent long-tail dynamics in driven magnets with dynamical frustration
- Statistical Mechanics of Dynamical System Identification
- Second law of thermodynamics in closed quantum many-body systems
- Chemical potentials from structure factors: II. Charged multi-component mixtures
- Genuine quantum scars in Floquet chaotic many-body systems
- Stochastic Thermodynamics for Autoregressive Generative Models: A Non-Markovian Perspective