Overcoming the Lamb Shift in System-Bath Models via KMS Detailed Balance: High-Accuracy Thermalization with Time-Bounded Interactions

Authors: Hongrui Chen (Stanford), Zhiyan Ding (Michigan), Ruizhe Zhang (Purdue) · arXiv:2604.15616 · submitted 20 April 2026 · score 8/10 (HIGH)

Abstract

We investigate quantum thermal state preparation algorithms based on system-bath interactions and uncover a surprising phenomenon in the weak-coupling regime. We rigorously prove that, if the system-bath interaction is engineered so that the transition part of the approximate Lindbladian generator satisfies the KMS detailed balance condition, then the unique fixed point of the dynamics can be made arbitrarily close to the Gibbs state in the weak-coupling limit, regardless of the structure of the Lamb shift term. Importantly, this remains true even when the approximate Lindbladian differs substantially from the ideal Davies generator and the Lamb shift term does not commute with the thermal state. Our result shows that the role of the KMS detailed balance condition extends well beyond standard Lindbladian dynamics, serving as a general principle for a broader class of dissipative systems. Furthermore, by combining this with a general perturbation framework, we bound the mixing time of the dynamics and establish an end-to-end complexity of O(ε⁻¹) for Gibbs state preparation. These guarantees apply to any Hamiltonian for which the corresponding KMS-detailed-balance Lindbladian is known to mix rapidly.

Executive summary

Chen, Ding, and Zhang resolve an outstanding bottleneck in simpler-to-implement Gibbs-state preparation schemes. Prior system-bath approaches (Lloyd et al. 2025, Hahn et al. 2026, Ding et al. 2025) avoid the heavy machinery of Lindbladian simulation (block-encoding of jump operators, clock registers, time-reversed Hamiltonian simulation) at the cost of an approximate Lindbladian carrying a non-commuting Lamb-shift term — conventional wisdom said this term forces the envelope-function support width to diverge and inflates end-to-end complexity to O(ε⁻²) or worse. This paper proves that if the transition (dissipative) part of the effective Lindbladian exactly satisfies the KMS detailed-balance condition, a subtle cancellation between the KMS structure and the outer system-unitary evolution U_S(T) drives the channel's fixed point to within O(α²) of the Gibbs state in the weak-coupling limit — with constant interaction time. Plugging through a perturbation-theory mixing-time bound gives end-to-end O(ε⁻¹) Gibbs preparation. For Yuan, whose Y5 dequantised-SDP pipeline depends on efficient Gibbs-state preparation for Pauli-sparse Hamiltonians, this paper improves the precision dependence of the simpler system-bath family by a full power of ε and unlocks it for every Hamiltonian class already known to have a spectrally gapped KMS-detailed-balance Lindbladian.

Main contribution

The paper's two main theorems (Fixed-Point Approximation and Mixing Time, both informal in Sec. 1; rigorous versions in Sec. sec:ding_statement and Sec. sec:applications) together prove that three existing system-bath algorithms — Lloyd et al. 2025, Hahn et al. 2026, and Ding et al. 2025 — can, after a mild parameter-retuning that does not change the algorithmic framework, achieve the Gibbs state to fidelity ε with total Hamiltonian simulation time O(ε⁻¹). The earlier state of the art was O(ε⁻²) or O(ε⁻⁴) (depending on whose algorithm and in what coupling regime). The improvement is driven by a new asymptotic analysis that leverages (i) exact KMS detailed balance of the transition Lindbladian L_KMS and (ii) the outer unitary conjugation U_S(T) • U_S(T)^† in the repeated-interaction channel Φ_α, showing that together they cancel the O(1) steady-state bias that a noncommuting Lamb shift would otherwise induce.

Key theorems / lemmas / algorithms

Theorem (Fixed-Point Approximation, Informal, Thm. fixed_point_approx): Suppose Φ_α(ρ) = U_S(T) ˆ exp(α²L) ˆ U_S(T) + O_β,σ(α⁴), with L = −i[H_Lamb,·] + L_KMS where L_KMS exactly satisfies the KMS detailed-balance condition. If the channel has unique fixed point and finite mixing time, then ||ρ_fix(Φ_α) − ρ_β||₁ = O(ε + α² t_{mix,Φ_α}(ε)).
Theorem (Mixing Time, Informal, Thm. mixing_time): If L_KMS has spectral gap λ_gap > 0 and ||ρ_β^−1/4 H_Lamb ρ_β^1/4 − ρ_β^1/4 H_Lamb ρ_β^−1/4|| = O(λ_gap), and α = O(λ_gap^1/2), then τ_{mix,Φ_α} = Õ(α⁻² λ_gap⁻¹) and ||ρ_fix − ρ_β||₁ = Õ(α² λ_gap⁻¹).
Theorem (KMS-DBC Lindbladian characterisation, Thm. kms_dbc_lindbladian_characterization): A general characterisation of what it means for a transition-part Lindbladian generator to satisfy the KMS detailed-balance condition — the structural property that unlocks the cancellation.
Theorem (Ding et al. end-to-end, Thm. end_to_end_ding): Rigorous version applied to the Ding et al. 2025 algorithm with explicit parameter choices (Gaussian envelope f(t), width σ = Ω(β), coupling α = o(1), evolution time T drawn from a specified gamma-like density). Yields total Hamiltonian-simulation time O(ε⁻¹).
Proposition (Approximate fixed point, Prop. approx_fixed_point): Constructs an auxiliary reference state ρ* close to both ρ_β and ρ_fix(Φ_α) for the perturbation argument.
Comparison table (Table tab:comparison): Systematically lists seven prior system-bath Gibbs preparation algorithms with their fixed-point-error, mixing-time, interaction strength, envelope-support-width, and end-to-end complexity — then shows how three of them (Ding, Lloyd, Hahn) collapse to envelope-width Θ(1) and end-to-end O(ε⁻¹) under this paper's theory.

Detailed walkthrough

The starting point is the system-bath interaction framework for Gibbs-state preparation. Rather than simulating a Lindbladian master equation directly (which requires block-encoding highly non-local jump operators), one introduces a small ancillary bath, weakly couples it to the system via a time-dependent Hamiltonian H_α(t) = H + H_E + α(f(t) A_S ⊗ B_E + c.c.), evolves the joint system for time T, traces out and resets the bath. The induced channel Φ_α has, in the weak-coupling limit, the asymptotic expansion Φ_α(ρ) = U_S(T) ˆ exp(α² L) ˆ U_S(T)[ρ] + O(α⁴), where L is an effective Lindbladian decomposing as L(ρ) = −i[H_Lamb,ρ] + L_KMS(ρ).

The problem identified (introduction, Q1 and Q2 in Sec. 1): while L_KMS fixes ρ_β by KMS detailed balance, H_Lamb generally does not commute with ρ_β; so L itself does not fix the Gibbs state, and the fixed point of exp(α² L) is biased away from ρ_β by O(1). Prior work (Hahn et al. 2026, Ding et al. 2025) rescued convergence only by taking the envelope-support width σ → ∞, which inflates per-iteration simulation time and the mixing time itself — leading to end-to-end O(ε⁻²) or worse.

The insight (Sec. 1 and Sec. proof_overview): the full channel Φ_α is not exp(α² L). It is U_S(T) ˆ exp(α² L) ˆ U_S(T), plus O(α⁴). The outer unitary conjugations U_S(T), typically dismissed as technical details, interact with the exact KMS structure of L_KMS to suppress the bias. Concretely: expand the fixed-point equation Φ_α(ρ) = ρ order-by-order in α, use perturbation theory about the Gibbs state, and observe that the first-order bias induced by H_Lamb is cancelled by an equal-magnitude contribution from the conjugation-structure of U_S(T) when L_KMS is KMS-DB. Higher-order corrections are then controlled by perturbation bounds. The net effect is that ρ_fix(Φ_α) lies within O(α²) of ρ_β without requiring the envelope support width to diverge.

The mixing-time half (Thm. mixing_time) uses a perturbation-theory bound due in spirit to Wang et al. 2025: if L_KMS has a spectral gap λ_gap, the perturbed L has contraction rate bounded below by a multiplicative factor of λ_gap, provided H_Lamb's KMS-asymmetry (measured by the commutator condition on ρ_β-weighted conjugations) is itself O(λ_gap). Under this regime the channel-level mixing time is Õ(α⁻²λ_gap⁻¹); combined with interaction time T per step and the O(α²) fixed-point bias, the total Hamiltonian simulation time for ε-accurate Gibbs preparation is Õ(ε⁻¹). Setting λ_gap = Θ(1), α = O(ε^1/2), mixing time O(ε⁻¹).

Sec. applications (sec:applications, sec:ding_statement) instantiates the framework on Ding et al. 2025. With a Gaussian envelope f(t) = exp(−t²/4σ²)/√(σ√(2π)), width σ = Ω(β), coupling strength α = o(1), a specific spectral-density function g(ω), and evolution-time density μ₀(t) ∝ (t−1)³ e^−(t−1) &bold;1_t≥1, the paper derives Thm. end_to_end_ding, giving an explicit end-to-end O(ε⁻¹) bound. Critically, σ need only be a constant (not →∞); this is the qualitative table entry that moves across from "σ = +∞" to "σ = Θ(1)".

The theorem statements apply to all Hamiltonians for which the corresponding KMS-detailed-balance Lindbladian is known to have a positive spectral gap. That currently includes: high-temperature local spin Hamiltonians (Rouzé et al. 2024), weakly interacting fermionic/spin systems at all temperatures (Smid et al. 2025), and 1D local Hamiltonians at all temperatures (Bergamaschi et al. 2026). Any future spectral-gap result immediately plugs in and inherits the O(ε⁻¹) dependence.

The work is theoretical and carries no figures; the comparison table is the main visual artefact.

Figures

No figures extracted from source. (Paper is theoretical with no embedded figures; the main visual element is the comparison table tab:comparison, which is rendered inline as LaTeX.)

Citations to Yuan's papers

No direct citation to any of Y1–Y6 found in bibliography.

Overlap with Y1–Y6

Y5 (Gibbs + Pauli sparsity SDP) — primary overlap: Y5 (Yuan, França, Luchnikov, Tiunov, Haug, Aolita 2025, arXiv:2510.08292) solves structured Goemans-Williamson SDPs by preparing Gibbs states of Pauli-sparse Hamiltonians and using them to estimate dual/primal quantities. The bottleneck is efficient Gibbs-state preparation; Y5 presumably either uses a Lindbladian-simulation-based scheme (heavy machinery) or assumes a Gibbs sampler oracle. Chen–Ding–Zhang give a substantially simpler system-bath alternative with the same O(ε⁻¹) precision dependence that Lindbladian-simulation methods achieve — without needing block-encoded jump operators or clock registers. If Y5's Pauli-sparse Hamiltonian class (or a close cousin) admits a KMS-DB Lindbladian with a spectral gap (plausible for the structured families Y5 targets), this result directly improves Y5's implementation story.
Y1/Y2/Y3 (QAOA family) — weak link: No direct method connection. Thermal Gibbs-state preparation is sometimes invoked as an initialisation for QAOA / variational schemes (warm-starting with thermal states), so there is a conceptual "thermalisation as resource" bridge but no immediate methodological transfer.
Y4 (Grover + ADMM for cardinality-constrained BO) — no direct overlap.
Y6 (PBR on Heron2) — no direct overlap.

Recommended action for Yuan

Read deeper; adapt to Y5 follow-up. The most impactful next step is checking whether the Pauli-sparse Hamiltonians central to Y5 satisfy (or can be modified to satisfy) the KMS-DB + spectral-gap conditions here. If yes, Y5's Gibbs-preparation component can be re-instantiated on the Chen–Ding–Zhang framework with a simpler hardware story (constant σ Gaussian envelope, repeated single-qubit bath resets) and the same asymptotic precision. This could be a crisp technical contribution for a Y5 follow-up paper or a hardware-implementation companion.
Email authors. Specifically Zhiyan Ding and Ruizhe Zhang — whose earlier work (Ding et al. 2025, DingLiLinZhang2024) is instantiated as an application case. A short email citing Y5 and asking whether their framework has been or can be analysed on Pauli-sparse Hamiltonians with explicit block structure is likely to be welcomed; this is a natural application domain for their method that may not have crossed their desk.
Cite on next Y5 revision. When Y5 is next revised (or in any hardware-implementation companion), this paper should be cited as the state of the art for simpler-hardware Gibbs-state preparation with rigorous O(ε⁻¹) guarantees.