pith. sign in

arxiv: 2604.22996 · v1 · submitted 2026-04-24 · 🪐 quant-ph · cs.NA· math-ph· math.MP· math.NA

Accelerating quantum Gibbs sampling without quantum walks

Jiaqi Leng , Jiaqing Jiang , Lin Lin This is my paper

Pith reviewed 2026-05-08 11:56 UTC · model grok-4.3

classification 🪐 quant-ph cs.NAmath-phmath.MPmath.NA
keywords quantum Gibbs samplingpurified Gibbs statesKMS detailed balanceparent Hamiltonian factorizationsingular value filteringquantum singular value transformationwalk-free algorithmGibbs sampler
0
0 comments X

The pith

A factorization of the parent Hamiltonian allows preparing purified Gibbs states with quadratically better gap dependence without quantum walks.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents a quantum algorithm for preparing purified Gibbs states that skips quantum walks entirely. For Gibbs samplers obeying exact Kubo-Martin-Schwinger detailed balance, it factors the parent Hamiltonian into noncommutative first-order operators. This reduction turns the task into singular-value filtering, which quantum singular value transformation solves with a quadratic improvement in spectral-gap scaling under coherent access. The method covers several implementable samplers beyond the Davies generator and adds auxiliary dissipative dynamics for warm starts in metastable cases.

Core claim

For a broad class of quantum Gibbs samplers satisfying exact KMS detailed balance, the corresponding parent Hamiltonian admits an explicit factorization into noncommutative first-order operators. This factorization converts purified Gibbs-state preparation into a singular-value filtering problem that quantum singular value transformation solves with quadratically improved dependence on the spectral gap. The framework applies under standard coherent-access assumptions to efficiently implementable samplers beyond the Davies setting and supports an auxiliary dissipative dynamics for generating warm starts in the doubled Hilbert space.

What carries the argument

Explicit factorization of the parent Hamiltonian into noncommutative first-order operators that reduces the problem to singular-value filtering.

If this is right

  • Purified Gibbs states can be prepared with runtime scaling quadratically better in the spectral gap than walk-based methods.
  • The approach covers multiple efficiently implementable Gibbs samplers beyond the Davies generator.
  • Auxiliary dissipative dynamics based on the same factorization can produce warm starts in the doubled Hilbert space during metastable regimes.
  • The reduction to singular-value filtering works under standard coherent-access assumptions for the underlying operators.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The factorization technique may extend to other quantum algorithms that rely on detailed-balance conditions for mixing.
  • Faster finite-temperature state preparation could improve resource estimates for quantum simulations of many-body systems.
  • The warm-start construction might combine with other quantum algorithms to handle systems with slow relaxation.
  • Approximate versions of KMS balance could be tested to see how the quadratic gap benefit degrades.

Load-bearing premise

The quantum Gibbs samplers must satisfy exact KMS detailed balance and admit an explicit factorization of the parent Hamiltonian into noncommutative first-order operators that can be efficiently implemented with coherent access.

What would settle it

A concrete quantum Gibbs sampler that obeys exact KMS detailed balance yet whose parent Hamiltonian cannot be factored into noncommutative first-order operators in a way that yields the claimed quadratic gap improvement.

Figures

Figures reproduced from arXiv: 2604.22996 by Jiaqi Leng, Jiaqing Jiang, Lin Lin.

Figure 1
Figure 1. Figure 1: Circuit schematics for the block-encoding of the view at source ↗
Figure 2
Figure 2. Figure 2: Numerical simulation of the auxiliary dissipative view at source ↗
Figure 3
Figure 3. Figure 3: The ratio between the spectral gap of the new Hamiltonian view at source ↗
Figure 4
Figure 4. Figure 4: Spectral gaps of three classes of Lindbladian operators: the auxiliary dynamics (without unitary dressing), the view at source ↗
read the original abstract

Szegedy's quantum walk gives a generic quadratic speedup for reversible classical Markov chains, but extending this mechanism to quantum Gibbs sampling has remained challenging beyond special cases. We present a walk-free quantum algorithm for preparing purified Gibbs states with a quadratic improvement in spectral-gap dependence for a broad class of quantum Gibbs samplers that satisfy exact Kubo-Martin-Schwinger detailed balance. Our main structural result is an explicit factorization of the corresponding parent Hamiltonian into noncommutative first-order operators. This turns purified Gibbs-state preparation into a singular-value filtering problem and enables a quantum singular value transformation algorithm with quadratically improved gap dependence under standard coherent-access assumptions. The framework applies to several efficiently implementable Gibbs samplers beyond the Davies setting. We also introduce an auxiliary dissipative dynamics based on the same factorization, which can be used to generate warm starts in the doubled Hilbert space in metastable regimes.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

0 major / 3 minor

Summary. The manuscript proposes a walk-free quantum algorithm for preparing purified Gibbs states. For quantum Gibbs samplers satisfying exact Kubo-Martin-Schwinger (KMS) detailed balance, it derives an explicit factorization of the parent Hamiltonian into noncommutative first-order operators. This reduces purified Gibbs-state preparation to a singular-value filtering problem that is solved via quantum singular value transformation (QSVT), yielding a quadratic improvement in spectral-gap dependence under coherent-access assumptions. The framework is shown to apply to several efficiently implementable samplers beyond the Davies generator, and an auxiliary dissipative dynamics based on the same factorization is introduced to generate warm starts in the doubled Hilbert space for metastable regimes.

Significance. If the explicit factorization holds and the resulting operators admit efficient coherent block-encodings, the work constitutes a meaningful advance in quantum algorithms for thermal-state preparation. It directly addresses the difficulty of extending Szegedy-type quadratic speedups to the quantum setting by bypassing quantum walks altogether. The derivation of the factorization from the KMS condition, the reduction to QSVT, and the auxiliary warm-start dynamics are concrete strengths. The result applies to a non-trivial class of samplers and could improve the practicality of quantum simulation of open systems and many-body thermal properties.

minor comments (3)
  1. Abstract: the phrase 'broad class of quantum Gibbs samplers' is used without immediately indicating the precise additional structural requirements (beyond exact KMS) that enable the factorization; a single clarifying sentence would improve accessibility.
  2. §2 (factorization section): the noncommutative first-order operators are defined via the KMS condition, but a short explicit example for a two-qubit system would help readers verify the construction before the general QSVT reduction.
  3. §4 (complexity): the block-encoding cost of the first-order operators is stated to be gap-independent, yet the dependence on the system dimension or local interaction strength is not tabulated; adding a short complexity table would strengthen the claim.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive and accurate summary of our manuscript, as well as for recognizing its significance in providing a walk-free QSVT-based approach to purified Gibbs-state preparation under exact KMS detailed balance. The recommendation of minor revision is noted. However, the report contains no specific major comments to address point by point.

Circularity Check

0 steps flagged

Derivation self-contained; factorization derived directly from KMS without reduction to inputs or self-citations

full rationale

The paper's core contribution is an explicit factorization of the parent Hamiltonian into noncommutative first-order operators, obtained directly from the exact KMS detailed balance condition under coherent access. This factorization converts Gibbs-state preparation into a singular-value filtering task solvable via QSVT, yielding the quadratic gap improvement. No step equates a claimed prediction to a fitted parameter by construction, renames a known result, or relies on a load-bearing self-citation whose validity is internal to the paper. The KMS assumption is an external domain condition, and the factorization is presented as a new structural result rather than an ansatz smuggled via prior work. The auxiliary dissipative dynamics is likewise built from the same factorization without circular closure. The derivation chain is therefore independent of its target outputs.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests primarily on the domain assumption of exact KMS detailed balance enabling the factorization; no free parameters or new invented entities are introduced in the abstract. Standard quantum mechanics and coherent access are background assumptions.

axioms (2)
  • domain assumption Exact Kubo-Martin-Schwinger (KMS) detailed balance holds for the class of quantum Gibbs samplers considered
    This condition is required for the parent Hamiltonian factorization to exist and for the algorithm to apply.
  • standard math Standard assumptions of quantum mechanics and coherent access to the system Hamiltonian
    Required for implementing the quantum singular value transformation.

pith-pipeline@v0.9.0 · 5450 in / 1450 out tokens · 72179 ms · 2026-05-08T11:56:28.074132+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

75 extracted references · 75 canonical work pages · 3 internal anchors

  1. [1]

    Metropolis, A

    N. Metropolis, A. W. Rosenbluth, M. N. Rosenbluth, A. H. Teller, and E. Teller, Journal of Chemical Physics 21, 1087 (1953)

    work page 1953

  2. [2]

    R. J. Glauber, Journal of Mathematical Physics4, 294 (1963)

    work page 1963

  3. [3]

    Temme, T

    K. Temme, T. J. Osborne, K. G. Vollbrecht, D. Poulin, and F. Verstraete, Nature471, 87 (2011)

    work page 2011

  4. [4]

    A. Z. Panagiotopoulos, Molecular Physics61, 813 (1987)

    work page 1987

  5. [5]

    D. M. Ceperley, Reviews of Modern Physics67, 279 (1995)

    work page 1995

  6. [6]

    Frenkel and B

    D. Frenkel and B. Smit,Understanding Molecular Sim- ulation: From Algorithms to Applications(Academic Press, 2002)

    work page 2002

  7. [7]

    Geman and D

    S. Geman and D. Geman, IEEE Transactions on pattern analysis and machine intelligence , 721 (1984)

    work page 1984

  8. [8]

    A. E. Gelfand and A. F. Smith, Journal of the American Statistical Association85, 398 (1990)

    work page 1990

  9. [9]

    Casella and E

    G. Casella and E. I. George, The American Statistician 46, 167 (1992)

    work page 1992

  10. [10]

    G. E. Hinton and R. R. Salakhutdinov, Science313, 504 (2006)

    work page 2006

  11. [11]

    Hinton, Momentum9, 926 (2010)

    G. Hinton, Momentum9, 926 (2010)

    work page 2010

  12. [12]

    R. M. Neal,Bayesian Learning for Neural Networks, Vol. 118 (Springer Science & Business Media, 2012)

    work page 2012

  13. [13]

    Kirkpatrick, C

    S. Kirkpatrick, C. D. Gelatt Jr, and M. P. Vecchi, Science 220, 671 (1983)

    work page 1983

  14. [14]

    M. Dyer, A. Frieze, and R. Kannan, Journal of the ACM (JACM)38, 1 (1991)

    work page 1991

  15. [15]

    D. A. Levin and Y. Peres,Markov Chains and Mixing Times, Vol. 107 (American Mathematical Soc., 2017)

    work page 2017

  16. [16]

    Chayes, L

    JT. Chayes, L. Chayes, and R. H. Schonmann, Journal of Statistical Physics49, 433 (1987)

    work page 1987

  17. [17]

    F. Cesi, G. Guadagni, F. Martinelli, and RH. Schon- mann, Journal of Statistical Physics85, 55 (1996)

    work page 1996

  18. [18]

    R. H. Schonmann, Communications in Mathematical Physics112, 409 (1987)

    work page 1987

  19. [19]

    Szegedy, in45th Annual IEEE Symposium on Foun- dations of Computer Science(2004) pp

    M. Szegedy, in45th Annual IEEE Symposium on Foun- dations of Computer Science(2004) pp. 32–41

    work page 2004

  20. [20]

    Shenvi, J

    N. Shenvi, J. Kempe, and K. B. Whaley, Physical Review A67, 052307 (2003)

    work page 2003

  21. [21]

    Magniez, A

    F. Magniez, A. Nayak, J. Roland, and M. Santha, in Proceedings of the Thirty-Ninth Annual ACM Symposium on Theory of Computing(2007) pp. 575–584

    work page 2007

  22. [22]

    Buhrman and R

    H. Buhrman and R. Spalek, arXiv preprint quant- ph/0409035 (2004)

    work page arXiv 2004

  23. [23]

    C.-F. Chen, M. Kastoryano, F. G. Brandão, and A. Gi- lyén, Nature646, 561 (2025)

    work page 2025

  24. [24]

    Z. Ding, B. Li, and L. Lin, Communications in Mathe- matical Physics406, 67 (2025)

    work page 2025

  25. [25]

    Quantum Metropolis sampling via weak measurement,

    J. Jiang and S. Irani, arXiv preprint arXiv:2406.16023 (2024)

    work page arXiv 2024

  26. [26]

    Quantum generalizations of glauber and metropolis dynamics,

    A. Gilyén, C.-F. Chen, J. F. Doriguello, and M. J. Kas- toryano, arXiv preprint arXiv:2405.20322 (2024)

    work page arXiv 2024

  27. [27]

    Z. Ding, Y. Zhan, J. Preskill, and L. Lin, arXiv preprint arXiv:2508.05703 (2025)

    work page arXiv 2025

  28. [28]

    F. G. Brandão, A. Kalev, T. Li, C. Y.-Y. Lin, K. M. Svore, and X. Wu, arXiv preprint arXiv:1710.02581 (2017)

    work page arXiv 2017

  29. [29]

    J.VanApeldoorn, A.Gilyén, S.Gribling,andR.deWolf, in2017 IEEE 58th Annual Symposium on Foundations of Computer Science (FOCS)(IEEE, 2017) pp. 403–414

    work page 2017

  30. [30]

    M. M. Wilde, arXiv preprint arXiv:2512.19819 (2025)

    work page arXiv 2025

  31. [31]

    M. H. Amin, E. Andriyash, J. Rolfe, B. Kulchytskyy, and R. Melko, Physical Review X8, 021050 (2018)

    work page 2018

  32. [32]

    E. B. Davies, Journal of Mathematical Physics39, 91 (1974)

    work page 1974

  33. [33]

    Wocjan and K

    P. Wocjan and K. Temme, Commun. Math. Phys.402, 3201 (2023)

    work page 2023

  34. [34]

    C.-F. Chen, M. J. Kastoryano, F. G. Brandão, and A. Gi- lyén, arXiv preprint arXiv:2303.18224 (2023)

    work page internal anchor Pith review arXiv 2023

  35. [35]

    C.-F. Chen, M. J. Kastoryano, and A. Gilyén, arXiv preprint arXiv:2311.09207 (2023)

    work page arXiv 2023

  36. [36]

    D. Fang, J. Lu, Y. Tong, and C. Zhao, arXiv preprint arXiv:2604.07214 (2026)

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  37. [37]

    Somma, S

    R. Somma, S. Boixo, and H. Barnum, arXiv preprint arXiv:0712.1008 (2007)

    work page arXiv 2007

  38. [38]

    J. Leng, Z. Ding, Z. Chen, and L. Lin, Proceedings of the National Academy of Sciences123, e2512789123 (2026)

    work page 2026

  39. [39]

    Gilyén, Y

    A. Gilyén, Y. Su, G. H. Low, and N. Wiebe, inProceed- ings of the 51st Annual ACM SIGACT Symposium on Theory of Computing(2019) pp. 193–204

    work page 2019

  40. [40]

    Quantum Amplitude Amplification and Estimation

    G. Brassard, P. Hoyer, M. Mosca, and A. Tapp, arXiv preprint quant-ph/0005055 (2000)

    work page internal anchor Pith review arXiv 2000

  41. [41]

    Rall, Quantum5, 566 (2021)

    P. Rall, Quantum5, 566 (2021)

    work page 2021

  42. [42]

    Rall and B

    P. Rall and B. Fuller, Quantum7, 937 (2023)

    work page 2023

  43. [43]

    Vernooij and M

    M. Vernooij and M. Wirth, Communications in Mathe- matical Physics403, 381 (2023)

    work page 2023

  44. [44]

    In: arXiv:2504.02208 [quant-ph](2025)

    C.-F. Chen and C. Rouzé, arXiv preprint arXiv:2504.02208 (2025)

    work page arXiv 2025

  45. [45]

    Rouzé, D

    C. Rouzé, D. S. França, and Á. M. Alhambra, inPro- ceedings of the 57th Annual ACM Symposium on Theory of Computing(2025) pp. 1488–1495

    work page 2025

  46. [46]

    Alhambra, Daniel Stilck Fran¸ ca, and Cambyse Rouz´ e

    S. Slezak, M. Scandi, Á. M. Alhambra, D. S. França, and C. Rouzé, arXiv preprint arXiv:2601.16154 (2026)

    work page arXiv 2026

  47. [47]

    Quantum state preparation without coherent arith- metic,

    S. McArdle, A. Gilyén, and M. Berta, arXiv preprint arXiv:2210.14892 (2022)

    work page arXiv 2022

  48. [48]

    Slow mixing of quantum Gibbs samplers,

    D. Gamarnik, B. T. Kiani, and A. Zlokapa, arXiv preprint arXiv:2411.04300 (2024)

    work page arXiv 2024

  49. [49]

    G. H. Low and I. L. Chuang, arXiv preprint arXiv:1707.05391 (2017)

    work page Pith review arXiv 2017

  50. [50]

    G. H. Low, R. King, D. W. Berry, Q. Han, A. E. De- Prince III, A. F. White, R. Babbush, R. D. Somma, and N. C. Rubin, Physical Review X15, 041016 (2025)

    work page 2025

  51. [51]

    R. D. Somma and S. Boixo, SIAM Journal on Computing 42, 593 (2013)

    work page 2013

  52. [52]

    R. King, G. H. Low, R. Babbush, R. D. Somma, and N. C. Rubin, Physical Review Letters136, 110601 (2026)

    work page 2026

  53. [53]

    A. R. Brown, H. Gharibyan, S. Leichenauer, H. W. Lin, S. Nezami, G. Salton, L. Susskind, B. Swingle, and M. Walter, PRX Quantum4, 010320 (2023)

    work page 2023

  54. [54]

    Nezami, H

    S. Nezami, H. W. Lin, A. R. Brown, H. Gharibyan, S. Leichenauer, G. Salton, L. Susskind, B. Swingle, and M. Walter, arXiv preprint arXiv:2102.01064 (2021)

    work page arXiv 2021

  55. [55]

    Schuster, B

    T. Schuster, B. Kobrin, P. Gao, I. Cong, E. T. Khabi- boulline, N. M. Linke, M. D. Lukin, C. Monroe, B.Yoshida,andN.Y.Yao,PhysicalReviewX12,031013 (2022)

    work page 2022

  56. [56]

    Jafferis, A

    D. Jafferis, A. Zlokapa, J. D. Lykken, D. K. Kolchmeyer, S. I. Davis, N. Lauk, H. Neven, and M. Spiropulu, Nature 612, 51 (2022). 7

    work page 2022

  57. [57]

    Eternal traversable wormhole

    J. Maldacena and X.-L. Qi, arXiv preprint arXiv:1804.00491 (2018)

    work page Pith review arXiv 2018

  58. [58]

    D. Zhu, S. Johri, N. M. Linke, KA. Landsman, C. Huerta Alderete, N. H. Nguyen, AY. Matsuura, TH. Hsieh, and C. Monroe, Proceedings of the National Academy of Sciences117, 25402 (2020)

    work page 2020

  59. [59]

    Wu and T

    J. Wu and T. H. Hsieh, Physical Review Letters123, 220502 (2019)

    work page 2019

  60. [60]

    Kossakowski form

    R. Bhatia,Matrix Analysis, Vol. 169 (Springer, 1997). 8 SUPPLEMENTAL MATERIALS Appendix A: Sum-of-squares factorization of KMS DB Lindbladians Preliminaries.Given ann-qubit HamiltonianHand an inverse temperatureβ>0, the Gibbs state is defined by: σ= 1 Zβ exp(−βH), Z β =Tr[exp(−βH)].(A1) With the Gibbs stateσ, we define themodularoperator and theweightingo...

    work page 1997

  61. [61]

    Suppose each jump operator acts onnqubits, and there areJ=2 r jump operators

    Block-encodings of the{(B♭ j)⊺}and{B ♯ j}operators, denoted asU (B♭)⊺andU B♯. Suppose each jump operator acts onnqubits, and there areJ=2 r jump operators. The block-encodings usepancilla qubits such that (⟨0p∣⊗I2(r+n))U (B♭)⊺(∣0p⟩⊗I2n)= 1 C♭ J−1 ∑ j=0 ∣j⟩⟨0∣⊗(B♭ j)⊺,(B27) (⟨0p∣⊗I2(r+n))U B♯(∣0p⟩⊗I2n)= 1 C♯ J−1 ∑ j=0 ∣j⟩⟨0∣⊗B♯ j,(B28) whereC ♯andC ♭are th...

    work page

  62. [62]

    Suppose that the (time domain) discretization number isN=2 q and Usel = N−1 ∑ j=0 ∣j⟩⟨j∣⊗eitj H , U ⊺ sel = N−1 ∑ j=0 ∣j⟩⟨j∣⊗eitj H⊺ , t k =−T+ 2kT N with0≤k≤N−1.(B29)

    Select oracles for implementing controlled time-evolution ofHandH⊺, denoted asUsel andU ⊺ sel. Suppose that the (time domain) discretization number isN=2 q and Usel = N−1 ∑ j=0 ∣j⟩⟨j∣⊗eitj H , U ⊺ sel = N−1 ∑ j=0 ∣j⟩⟨j∣⊗eitj H⊺ , t k =−T+ 2kT N with0≤k≤N−1.(B29)

    work page

  63. [63]

    Note that, given our choice ofNandTas in Lemma 7, the normalization factorC∶= ∑N−1 k=0 gk is a constant close to∫ ∞ −∞g(t)dt=1/2

    A state preparation oracleUprep: Uprep∣0q⟩=∣g⟩,∣g⟩∶=1√ C N−1 ∑ k=0 √gk ∣k⟩,(B30) whereg k ∶=g(tk)∆t. Note that, given our choice ofNandTas in Lemma 7, the normalization factorC∶= ∑N−1 k=0 gk is a constant close to∫ ∞ −∞g(t)dt=1/2. Theorem 11.We can prepare a(2 √ CJ , p+1,0)-block-encoding ofBusing one query to each ofU(B♭)⊺andU B♯, two queries to each ofU...

    work page

  64. [64]

    a warm-start state∣ϕ⟩such that∣⟨ϕ∣σ1/2⟩⟩∣2 =Ω(1), 2.( √ J , p)-block-encodings of{(B♭ j)⊺}J−1 j=0 and{B ♯ j}J−1 j=0 , denoted byU(B♭)⊺andU B♯,

    work page

  65. [65]

    select oraclesUsel andU ⊺ sel for controlled time-evolution underHandH ⊺, respectively,

    work page

  66. [66]

    a state preparation oracleUprep as in Eq. (B30). Let∆=Gap(L †)denote the Lindbladian gap, andJis the total number of jump operators. Then, there exists a quantum algorithm that outputs anϵ-approximate of∣σ1/2⟩⟩using •O( √ J ∆ ⋅log(1 ϵ))quantum queries toU (B♭)⊺,U B♯,U sel,U ⊺ sel,U prep (and their inverse or controlled versions), •O(1)uses of the warm-sta...

    work page

  67. [67]

    In particular, bothI/22n and∣σ1/2⟩⟨σ1/2∣are stationary states

    The semigroup is not primitive. In particular, bothI/22n and∣σ1/2⟩⟨σ1/2∣are stationary states

    work page

  68. [68]

    For every observableOacting on the first register, lim t→∞ Tr[(O⊗I)ρ t]= Tr(O) 2n .(C9) Equivalently, lim t→∞ Tr2[ρt]= I 2n .(C10)

    work page

  69. [69]

    The Bell-diagonal subspace SBell∶=span{∣P⟩⟩⟨⟨P∣∶P∈Pn}(C11) is invariant underLaux

    In the sameβ=0setting, we denoteP n as the set ofn-qubit Pauli strings. The Bell-diagonal subspace SBell∶=span{∣P⟩⟩⟨⟨P∣∶P∈Pn}(C11) is invariant underLaux. More precisely, for each subsetT⊆[n], the subspace ST ∶=span{∣P⟩⟩⟨⟨P∣∶P∈Pn,supp(P)=T}(C12) is invariant underLaux

    work page

  70. [70]

    In particular,ω∅=∣σ1/2⟩⟩⟨⟨σ1/2∣, so the purified Gibbs state is not the unique stationary state in the Bell-diagonal sector

    For eachT⊆[n], the state ωT ∶= 1 2n3∣T∣ ∑ P∈Pn∶supp(P)=T ∣P⟩⟩⟨⟨P∣(C13) is stationary. In particular,ω∅=∣σ1/2⟩⟩⟨⟨σ1/2∣, so the purified Gibbs state is not the unique stationary state in the Bell-diagonal sector. Moreover, forT≠∅, the restriction ofLaux toS T has spectral gap12. Proof.Since each ̂Bj,α is Hermitian, the generator can be written as Laux(X)=− ...

    work page

  71. [71]

    2.ω ∅= 1 2n∣I⟩⟩⟨⟨I∣=∣σ1/2⟩⟩⟨⟨σ1/2∣is the unique stationary state of the auxiliary dynamicsetLU in the Bell-diagonal subspaceS Bell

    The Bell-diagonal subspaceSBell∶=span{∣P⟩⟩⟨⟨P∣∶P∈Pn}is invariant underL U. 2.ω ∅= 1 2n∣I⟩⟩⟨⟨I∣=∣σ1/2⟩⟩⟨⟨σ1/2∣is the unique stationary state of the auxiliary dynamicsetLU in the Bell-diagonal subspaceS Bell. 25

    work page

  72. [72]

    ForQ∈{I, X, Y, Z}, letP(j,Q) be the Pauli string obtained fromPby replacing thej-th factor byQ

    The spectrum ofLU∣SBell is ⎧⎪⎪⎨⎪⎪⎩ n ∑ j=1 λj ∶λj ∈{0,−4,−10+2 √ 3 i,−10−2 √ 3 i} ⎫⎪⎪⎬⎪⎪⎭ .(C49) Proof.ForP∈P n, writeρ P ∶=∣P⟩⟩⟨⟨P∣. ForQ∈{I, X, Y, Z}, letP(j,Q) be the Pauli string obtained fromPby replacing thej-th factor byQ. Also define the cyclic map c(X)=Y, c(Y)=Z, c(Z)=X,(C50) then we can check that LU,j(ρP)={ 0ifP j =I, 4(ρP (j,c(Pj ))+ρ P (j,I)−...

    work page

  73. [73]

    the DLL quantum Gibbs sampler (acting on the original Hilbert space)

    work page

  74. [74]

    (C6) (without unitary dressing); and

    the auxiliary Lindbladian dynamics given in Eq. (C6) (without unitary dressing); and

    work page

  75. [75]

    (C45) and (C46))

    the auxiliary Lindbladian dynamics with unitary dressing (see Eqs. (C45) and (C46)). In Fig. 4, we numerically compute the spectral gaps of these three Lindbladian generators for the 2-qubit TFIM HamiltonianH=−0.1(X 1+X 2)−Z1Z2 at various values ofβ. The DLL Lindbladian directly prepares the thermal stateσ. The latter two are auxiliary dynamics on the dou...

    work page