Fast mixing of all-to-all quantum systems at high temperatures
Thiago Bergamaschi · 2026
Read Fast mixing of all-to-all quantum systems at high temperatures by Thiago Bergamaschi (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-26090.txt.
Introduction
Thiago Bergamaschi proves fast mixing for quantum systems where every part touches every other part. Gibbs samplers prepare thermal states of quantum systems. At high temperature, known samplers mix fast on lattices, where geometry limits interactions. Many real systems lack that geometry, though. Electronic structure, mean-field models, and some quantum codes all interact at long range. Bergamaschi removes the lattice assumption entirely.
Arbitrary k-local Hamiltonians with bounded interaction strength are covered. At high enough temperature, a known quantum Gibbs sampler keeps a spectral gap that does not shrink with system size. A constant gap means mixing time stays tame as systems grow. Fully polynomial quantum algorithms follow as a consequence. Partition functions and global expectation values become efficiently approximable. The result extends the fast-mixing frontier from geometric systems to the all-to-all world, where many chemistry and coding Hamiltonians actually live.
Catalog record
| CREATOR | Thiago Bergamaschi |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 91 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.26090
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Fast mixing of all-to-all quantum systems at high temperatures Thiago Bergamaschi Abstract It is shown that arbitrary quantum k k -local Hamiltonians with bounded strength interactions admit a quantum Gibbs sampler [ CKG23 ] with a system-size independent spectral gap, at sufficiently high temperatures. This generalizes the existing quantum fast-mixing results beyond the geometrically-local setting. As a consequence, such systems admit fully-polynomial time quantum approximation algorithms for partition functions and global expectation values. 1 Introduction A central theme in quantum many-body physics is that locality constrains dynamics. In geometrically local systems (that is, lattices) such constraints are typically captured by Lieb-Robinson bounds [ LR72 , HK06 , NS10 ] . These bounds quantify the speed at which information percolates throughout the system, confining the time-evolution of local operators, to an effective lightcone around the operator. Lieb-Robinson bounds have played a foundational role in our understanding of locality in quantum systems, underlying results on tensor network descriptions and area laws [ Has06 , HW05 , Has07 , AC21 ] , quantum and classical simulation algorithms [ Osb06 , HHKL23 , WA23 , MYLZ25 ] , and correlation-decay and stability properties [ HK06 , NS06 , BHV06 , BHM10 ] . At the same time, many quantum systems of interest are not naturally described by lattice geometry. Examples include electronic-structure Hamiltonians [ WLA13 , OIWF21 ] , strongly correlated systems [ SY93 , Kit15 , MS16 ] , mean-field models [ Tót90 , Bjö13 , Bjö16 ] , expander-based qLDPC codes [ TZ14 , HHO21 , PK22 , DLV24 ] , all of which exhibit long-range or non-geometric interactions. Understanding how such all-to-all systems scramble quantum information under time-evolution can be significantly more challenging, since the usual notion of a lightcone is no longer the correct measure of operator growth [ LSH+13 , Luc20 , CL21 ] . In this paper, we consider such families of all-to-all interacting Hamiltonians, and study the complexity of their thermal (or ‘Gibbs’) state-preparation algorithms. Following the modern quantum Gibbs sampling paradigm, we study their dissipative state preparation under Lindbladian evolution (or, a continuous-time quantum Markov chain) [ RWW23 , CKBG23 , CKG23 , DLL25 , JI24 ] . Recently, there has been significant interest in proofs of efficient convergence, or mixing times for these algorithms, albeit, existing arguments for non-commuting systems are all limited to lattice Hamiltonians [ RFA25 , RSA26 , BLMT25 , BC26 , ŠMBB26 , TZ25 ] . In part, this is since the analysis in prior work fundamentally relies on Lieb-Robinson bounds. The main result in this paper is a proof that arbitrary all-to-all \mathsf{k} -local Hamiltonians with bounded strength interactions (see Section 1.1 ) admit a Gibbs sampler [ CKG23 ] with a system-size independent spectral gap at sufficiently high temperatures. Thus, such systems are fast-mixing , and consequently admit fully-polynomial time quantum approximation schemes for partition functions and expectation values. Our argument is based on revisiting the recently introduced non-commutative versions of the Dobrushin conditions, sufficient conditions for rapid-mixing in quantum systems [ BLMT25 , MOZ98 , TK15 , RSA26 ] , under the quantum cluster expansion of Netočný and Redig [ NR04 , MH21 ] . In the process, we hope that our techniques shed some light on the challenges in studying mixing times in all-to-all quantum systems, and serve as a stepping stone towards more complex disordered systems; see Section 1.2 for further discussion. 1.1 Contributions We consider generic \mathsf{k} -local Hamiltonians on systems of n qudits. We will state our results as a function of three notions of locality for such systems. Definition 1.1 (All-to-all Hamiltonians) . Let \mathbf{H}:=\sum_{e\in\Gamma}\mathbf{h}_{e}\otimes\mathbf{I}_{[n]\setminus e} be a local Hamiltonian on n qudits of local dimension 2^{\mathsf{q}} . The locality \mathsf{k} of \mathbf{H} is the uniform support size |\mathsf{supp}(\mathbf{h}_{e})|=\mathsf{k}. The degree \mathsf{d} is the number of qudits any given qudit interacts with u\in[n]:\quad\big|\{v:\exists e\in\Gamma\text{ s.t. }\{u,v\}\subseteq e\}\big|\leq\mathsf{d} (1) The interaction strength \mathsf{J} is the sum over terms coupling a pair of qudits u\neq v:\sum_{e\supseteq\{u,v\}}\|\mathbf{h}_{e}\|\leq\mathsf{J} . We focus on the regime where \mathsf{k},\mathsf{q} and (\mathsf{d}\cdot\mathsf{J}) are fixed constants independent of n . Remark 1.2 . Our statements are phrased as a function of \mathsf{k},\mathsf{d},\mathsf{J} and \mathsf{q} for completeness. In a first pass, the reader may want to imagine a qubit \mathsf{k}=2 local Hamiltonian (a graph) of bounded degree \mathsf{d} where each edge-term is bounded by \mathsf{J}=\mathsf{d}^{-1} . We emphasize \mathsf{d} is allowed to scale with n . Our goal is to study the rate of (open-system) thermalization [ BP07 ] of such Hamiltonians, when weakly-coupled to a heat-bath at some fixed (constant) inverse-temperature \beta , eventually converging to its Gibbs state: \boldsymbol{\rho}:=\frac{e^{-\beta\mathbf{H}}}{\tr[e^{-\beta\mathbf{H}}]}. (2) Following the modern paradigm in the quantum Gibbs sampling literature, we model this thermalization process by the evolution of a quantum Markov semigroup, generated by the thermal Lindbladian \mathcal{L} of [ CKG23 ] : \frac{\mathrm{d}}{\mathrm{d}t}\boldsymbol{\sigma}=\mathcal{L}[\boldsymbol{\sigma}],\quad\text{where}\quad\mathcal{L}[\boldsymbol{\sigma}]=\sum_{a\in\mathcal{S}_{[n]}^{1}}\mathcal{L}_{a}[\boldsymbol{\sigma}],\quad\text{ and }\quad\mathcal{L}_{a}[\boldsymbol{\rho}]=0, (3) which satisfies some form of quasi-locality and (KMS) detailed-balance, thus fixing the Gibbs state. 1 1 1 The detailed structure of the dynamics depends on \mathbf{H},\beta and a choice of updates (here, single-site Pauli operators \mathcal{S}_{[n]}^{1} ); see Section 2 for further background. The main result of this paper is a proof of a spectral gap for the [ CKG23 ] Lindbladian \mathcal{L} for arbitrary all-to-all Hamiltonians, at sufficiently high temperatures. Theorem 1.3 (Fast-mixing of all-to-all quantum systems at high temperatures) . Let \mathbf{H} be an all-to-all Hamiltonian as in Section 1.1 . Then, there exists a universal constant \mathsf{c}\geq 1 and threshold \beta_{\mathsf{c}}:=[\mathsf{c}^{\mathsf{q}\cdot\mathsf{k}}\cdot\mathsf{d}\mathsf{J}]^{-1} such that for all \beta<\beta_{\mathsf{c}} the Lindbladian \mathcal{L} ( 3 ) admits a system-size independent spectral gap, depending only on \mathsf{q},\mathsf{k} . In the context of balanced 2-local Hamiltonians as in Section 1.1 , \beta_{\mathsf{c}} is simply some explicit universal constant, which assuming \mathsf{BQP}\neq\mathsf{NP} is optimal due to the hardness of approximate counting [ Sly10 , SS12 ] . To the extent of our knowledge, prior proofs of unconditional, system-size independent spectral gaps for quantum Markov chains have been restricted to lattice Hamiltonians. Notably, one-dimensional systems at all temperatures [ KB16 , BC26 ] , and in finite-dimensional lattices at high-temperatures [ RFA25 , RSA26 , BLMT25 ] . 2 2 2 Albeit, these systems have been shown to be rapid-mixing, i.e. \mathcal{O}(\log\frac{n}{\varepsilon}) mixing time. Corollary 1.4 (Quantum algorithms for partition functions and expectation values) . In the context of Theorem 1.3 , for any \varepsilon>0 there exists a \mathsf{poly}(n,\frac{1}{\varepsilon}) time quantum algorithm to estimate the partition function Z_{\beta}:=\tr[e^{-\beta\mathbf{H}}] up to relative error 1\pm\varepsilon , and global expectation values up to additive error \varepsilon , in all-to-all quantum systems at sufficiently high temperatures. Indeed, Theorem 1.3 implies that the evolution converges to within \varepsilon trace distance of \boldsymbol{\rho} in time t_{\mathsf{mix}}(\varepsilon)\leq\mathcal{O}(n+\log\frac{1}{\varepsilon}) , entailing efficient state-preparation on a quantum computer by simulation of the dynamics [ CKG23 , Theorem I.2] . We remark that existing classical algorithms for the same task [ MH21 , HMS20 , YL23 , BLMT24 , Zlo26 ] require n^{\mathcal{O}(\log\mathsf{d})} time. In fact, it is conjectured in [ MH21 , Section V] that improved runtimes could be achieved via Markov chain methods. 1.2 Discussion and related work Cluster expansions. As we discuss shortly, the starting point behind Theorem 1.3 is a cluster expansion for the short