WorksPhysics and Astronomy
A Finite-Range, Translation-Invariant Counterexample to Weak Gibbsianity
Abstract
We construct a finite-range, translation-invariant counterexample to the Weak Gibbsianity property for quantum spin systems on . The proof relies on showing how a local potential can induce macroscopically large deviations from an equilibrium configuration after a perturbation is introduced at the boundary of the lattice.
Notation
We introduce now the basic notation that we will employ in the following.
Notation 0.1 (Lattice structures).
We denote by finite sublattices of .
We denote by the complements of in : .
For any finite set we denote its cardinality by .
Notation 0.2 (Quasi-local algebraic setting). We denote a net of local algebras indexed by finite subsets of by . In this work, for every we have and
The quasi local algebra is constructed by standard inductive limit argument from the local algebras ; see [BR87, Section 2.6] and [6 Section 6.2] for additional details.
Notation 0.3 (Spin matrices and computational basis). We denote by , , the spin- angular momentum operators at site , satisfying the commutation relations
We denote by the corresponding Pauli matrices.
We define the computational basis in a finite region as the tensor product over all points of of the eigenstates of . These basis elements will be labeled as , where . Alternatively, when we need to specify the configuration we will adopt Dirac’s bra-ket notation
Introduction
The concept of a Weak Gibbs state has its roots in the thermodynamic formalism of classical statistical mechanics [15]. This property was subsequently formulated explicitly and explored in the theory of dynamical systems [19, 20], as well as in connection with equilibrium measures and large deviations [13, 14].
Even though they were not precisely formalized before recent times, quantum Weak Gibbs state already appeared in the quantum spin system literature [2, 3], and more recently found application in the study of large deviation principles for quantum lattice spin systems [12].
More recently, the study of Weak Gibbs states has received renewed attention in the question of approach to equilibrium for quantum statistical mechanics through the works of Jakšić, Pillet and Tauber [11, 10], who gave the first formalized notion for quantum spin systems, and Jakšić, Pillet and Szczepanek [9, 8, 7]. A standard formulation of Weak Gibbsianity for quantum spin systems is the following; see [11 Definition 2.9].
Definition 1.1 (Weak Gibbs state). Let be a self-adjoint interaction for a quantum spin system with quasi-local algebra and local algebras , so that .
Fix an inverse temperature . For any finite region , let and define the finite-volume Gibbs state by
A state is said to be a Weak Gibbs state with respect to at inverse temperature if there exists a family of constants , with , such that
where the limit is taken along increasing cubes, and
for every finite region . The inequalities are understood in the order on positive linear functionals, that is, after evaluation on every positive element of $\mathcal{A}_\Lambda}.
Usually, this definition is given for restricted classes of interactions , such as those that are translation-invariant and exponentially decaying.
It was widely expected, and explicitly conjectured in [7], that Weak Gibbsianity should hold for thermal equilibrium states of any quantum spin system on the lattice , , under suitable regularity assumptions on the interaction, such as finite range or exponential decay. Previous results in this direction date back to Araki [1] for one-dimensional lattice systems with finite-range interactions, and to Araki and Ion [3] for arbitrary at sufficiently high temperatures. The Weak Gibbsianity property was explicitly established in these settings in [11 Theorem 2.11].
In this paper, we show that the Weak Gibbsianity conjecture is false for , even for finite-range, translation-invariant interactions.
The counterexample is constructed starting from the following nearest-neighbors potential
for all . The potential in (2) does not belong to the usual class of XYZ interactions. For every pair of neighboring sites , it is given by a coupling times the sum of two terms: the tensor product of the projection onto the spin state at site ,
and the Pauli matrix at site , and the corresponding term with and interchanged. Therefore, the action of the interaction term on the computational basis is as follows
We see that every interaction potential has positive matrix coefficients in the computational basis.
For every finite subset , we denote the Hamiltonian of the model as
By standard results [6 Section 6.2], the net of Gibbs states for the interaction in (2) admits convergent subsequences to -KMS states on for the infinite volume dynamics generated by .
The central result of the paper is formulated in the following theorem.
Theorem 1.2 (Counterexample to the Weak Gibbsianity property). Let be any limit point of the net of finite volume Gibbs states for the interaction potential in (2). Suppose that the inverse temperature satisfies
Then, there exists a positive constant and an increasing sequence of squares converging to such that
where
is the mean magnetization along the -axis in the region .
We see that Equation (4) provides an explicit counterexample to the upper bound in the Weak Gibbsianity condition (1) for any limit point of the net of Gibbs states .
The proof of the counterexample in Theorem 1.2 relies on techniques similar to those used to establish the blow-up of complex-time evolution in lattice spin systems in dimensions . These techniques were originally developed by Bouch [4] for a specific quantum spin model. Tasaki [18] subsequently provided a clearer exposition of Bouch’s construction and, in a later revision, incorporated an LLM-generated proof extending the combinatorial result from a particular sequence of tree sizes to every positive integer size. It is possible that the “Extraction lemma” [18 Lemma 2.2] could be used to improve our proof, but we did not check this. Shiraishi and Tasaki [16 Appendix A.4, Theorem A.6] also employed Bouch’s construction to extend the complex-time blow-up result to a general class of XY and XYZ models.
In § 3, we prove Theorem 1.2 and discuss some of its immediate consequences.
Proof of the counterexample
In this section, we will present a proof of Theorem 1.2, showing that there exist finite-range, translation-invariant models for which the Weak Gibbsianity property fails on local observables. We will work in a two dimensional lattice by first restricting to a finite subregion of the lattice . It is fairly easy to extend all the results to dimensional lattices.
We need a few preliminary observations. The first one concerns the mean magnetization along the third direction, and its relation with the interaction potential defined in Equation (2).
Remark 2.1. We can immediately read from (3) that the “all spin up” state
is in the Kernel of
The projection onto is also the projection corresponding to the value of the mean magnetization along the third direction
Next, we fix a positive inverse temperature and we consider the finite volume Gibbs state
for any region .
Remark 2.2. For any we have
where is the partition function for the model in (2), localized in .
We consider the boundary interaction for any finite subregion
We can bound the norm of as
where
is the boundary of .
Remark 2.3. Since the interaction potential has positive matrix elements with respect to the computational basis, it easily follows that
for any . Moreover, by a standard Golden-Thompson inequality, the partition function satisfies
Hence, we see that
We can now outline the main steps in the construction of our counterexample. Let us define for any finite set
where and we suppose that there exists a unique such that . Combining Observations 2.2 and 2.3 we arrive at the inequality
To further estimate the last line, let us choose an element of the computational basis in the range of the projector , i.e. . This essentially means that coincide with on , while it can be chosen arbitrarily on . Then, for any other computational basis element we have
Using again that the matrix elements of are positive in the computational basis we can write
where, in the last line, we have estimated the full Taylor series of the exponential using its term of order , for arbitrary . The last bracket can be written as a sum of terms of the form
We can interpret the above as a path on the set , where at every step the spin at a site is reversed if the spin at site is . The only non zero contribution will be those arising from paths changing the configuration into the configuration . For all such paths, the bracket will be equal to one. Therefore, it remains to estimate the number of these admissible paths. If this number, denoted by , were to satisfy
for some constant , then we would obtain the estimate
Putting all together, this would imply the inequality
We will now describe in detail the estimate for the number . The core idea of our construction is the following. To allow for many different paths connecting the configurations and , these vectors must differ at a macroscopic number of points in the lattice. The second requirement is that there should be many distinct ways to change from one configuration to the other by repeatedly applying the potential . As we will see, these conditions are satisfied by our choice of the potential (2). Indeed, the path can have many different branches, giving rise to a fractal tree structure. This procedure follows the idea in Bouch’s analysis of complex time singularity [4], although the details of the computations and the choice of the tree differ from those of Bouch.
Construction of the spanning tree
We consider a sequence of lattice squares constructed as follows. Fix for and take . We enlarge it by adding the point , so that its enlarged version is given by . The box’s sides length satisfies . Our choice for the vectors and is the following
We see that the two configurations overlap only at the vertex . This is the starting point of our construction. Iteratively applying the operators we will construct a path of operators covering the whole square and flipping the spin at every point by repeated actions
and so on. This path must contain many branches, otherwise the number of possible combinations that generate the admissible paths would note be sufficient to violate the Weak Gibbsianity property. We will rely on a fractal tree construction, inspired by the results presented in [17].4 The resulting path is constructed by inductively building a spanning tree . here are the vertices of the tree, whereas are the edges.
Let us describe the first iterative step of the construction. We take to be the spanning tree for with edges
To construct , we attach four identical copies of at the four corners of , and we connect these by another copy of in the middle square. We repeat this procedure iteratively. To construct the spanning tree in , we take four translated copies of and we attach these to the four corners of . We connect these four trees thorough another copy of in the middle of the square . The first tree steps of the construction are illustrated in Figure 1. Note that the resulting graph is always connected and contains all vertices of .

Figure 1. Fractal spanning tree construction. This figure is a partial reproduction of [17 Figure 1].
Let us prove a first geometric lemma that will be needed for future estimates
Lemma 2.4. Let be the spanning tree defined above. Call the diameter of defined as the longest non-intersecting path along the edges of . Then,
Proof. It is clear that . For any two points , these satisfy one of the following:
Both and belong to the central square ;
One of the two points belongs to the central square say , while the other belongs to one of the four copies of ;
They both belong to the same copy of ;
They belong to different copies of .
In the first case, their distance along the edges of is at most 3. As for the second case, their distance is less than the number of edges that one of the two points has to travel to reach the central square, plus the diameter of the square. This is less than . In the third case, their distance is always less than . In the fourth case, their distance is always less than the number of edges it takes for both of them to travel to the cross, plus the length of the cross. This is less than . Therefore, we see that
Iterating this inequality times we get
This concludes the proof of the Lemma.
Next, we count the number of ways to construct the tree as a nested tree, starting from a given root . In [4 Lemma 6.2], Bouch proved that the number of different constructions leading to the same rooted tree with root and edges is given by
where is the weight of the edge , defined as one plus the number of its descendants. These are the edges that follow as we move away from the root . Therefore, after relabeling the edges of as , we obtain a lower bound on the number of possible constructions by finding an upper bound on
Let us define
We will estimate iteratively. In particular, we shall prove
Lemma 2.5. Take as above. Then,
Proof. We distinguish two cases. We place the root of on the central square . Then, every edges on the central tree will have necessarily less than descendant (this is the trivial bound). The product of the weights over all the edges of the four copies is bounded by by definition. Therefore, summing all the contributions we have
Suppose we put the root on one of the four squares . The unique path connecting the root with the central copy of contains less than edges. Each of these will have less than descendants. All the other branches starting from this path give a total contribution less than . We can repeat the same argument as in on the edges of the central square and of the other three copies of . Therefore,
where we have used Lemma 2.4 to estimate . Since , iterating the above inequality we get
This last estimate concludes the proof.
Upper bound on the Gibbs state
As we already argued in the previous subsection, if we start with a configuration with all spin up, except for the spin at , the contribution to
consists in all the possible paths of operators , flipping all the spin in . Some of the admissible choices are given by the all the possible paths generating the rooted tree with root in ; see Figure 2.

Figure 2. Path construction.
Combining Lemma 2.5 with Bouch’s argument, we see that the number of way to construct satisfies
Finally, inserting the latter estimate in (5) we conclude
Since this inequality holds for any , we can take a converging subsequence for as . To recover the exact statement of Theorem 1.2 it sufficient to repeat the construction using translated boxes that eventually cover the whole lattice . Thus, we have proved Theorem 1.2.
Remark 2.6 (Invariance under rotations). Since the magnetization commutes with any rotation around the -axis, the same result holds for all the rotated Hamiltonian
In particular, by choosing for every we can change the sign of the coupling (or equivalently, of the inverse temperature ). In a similar fashion, the same result holds for the model where we substitute all the projections onto with the projections onto the spin-up configurations .
Remark 2.7 (Translation invariance). The result of Theorem 1.2 readily generalizes to non-translation invariant models with potentials of the form
under suitable hypothesis on the couplings, e.g. and
would be sufficient.
Remark 2.8 (Magnetization bounds). The bound presented in Theorem 1.2 can be equivalently written as an inequality between different macroscopic sizes for the magnetization(2) observable. Indeed, we can write
Therefore, we see that changing the magnetization observable only at the boundary produces a macroscopically relevant factor.
Remark 2.9 (Limits of the current approach). It is not clear at present to what extent one can generalize the above argument to other projections of the magnetization. If we were to consider the projection onto states with one or more spins down, it would be impossible to reproduce the argument used in Theorem 1.2. Indeed, we would have one “activation site” inside the region . This alone would produce a macroscopic contribution to the expectation value in the Gibbs state.
AI disclosure. Following the release of the first version of this manuscript, Hal Tasaki drew my attention to technical errors in the proofs that had been identified using large language models (LLMs). I subsequently corrected these errors. LLMs were not used in any other part of the manuscript.
Acknowledgements. I am grateful to Yoshiko Ogata for many illuminating and stimulating discussions on this topic. Without her support, I could not have developed the arguments presented in this manuscript. I also thank Vojkan Jakšić and Anna Szczepanek for drawing my attention to the relevance of Weak Gibbsianity and for their many useful comments. I thank Hal Tasaki for pointing out errors in the first version of this manuscript.
I acknowledge the support of the National Group of Mathematical Physics (GNFM–INdAM).
Data availability statement. Data sharing is not applicable to this article as no new data were created or analysed in this study.
Conflict of interest statement. The authors certify that they have no affiliations with or involvement in any organization or entity with any financial interest or non-financial interest in the subject matter discussed in this manuscript.
References
- [1]H. Araki. “Gibbs states of a one dimensional quantum lattice”. In: Communications in Mathematical Physics 14.2 (1969), pp. 120–157. DOI: 10.1007/BF01645134. URL: https://doi.org/10.1007/BF01645134.
- [2]H. Araki. “Positive cone, Radon–Nikodým theorems, relative Hamiltonian and the Gibbs condition in statistical mechanics. An application of the Tomita–Takesaki theory”. In: C*-Algebras and Their Applications to Statistical Mechanics and Quantum Field Theory. Ed. by D. Kastler. Vol. 60. Proceedings of the International School of Physics “Enrico Fermi”. Course LX, Varenna, 1973. Amsterdam: North-Holland, 1976, pp. 64–100.
- [3]H. Araki and P. D. F. Ion. “On the equivalence of KMS and Gibbs conditions for states of quantum lattice systems”. In: Communications in Mathematical Physics 35.1 (1974), pp. 1–12. DOI: 10.1007/BF01646450. URL: https://doi.org/10.1007/BF01646450.
- [4]G. Bouch. “Complex-time singularity and locality estimates for quantum lattice systems”. In: Journal of Mathematical Physics 56.12 (2015), p. 123303. DOI: 10.1063/1.4936209.DOI
- [6]O. Bratteli and D. W. Robinson. Operator Algebras and Quantum Statistical Mechanics 2: Equilibrium States. Models in Quantum Statistical Mechanics. 2nd ed. Berlin, Heidelberg: Springer, 1997. ISBN: 978-3-540-61443-2. DOI: 10.1007/978-3-662-03444-6. URL: https://doi.org/10.1007/978-3-662-03444-6.
- [7]V. Jaks̆ić, C.-A. Pillet, and A. Szczepanek. Approach to equilibrium in translation-invariant quantum systems: Structural results II. arXiv:2610.21475. 2026. arXiv: 2610.21475 [math-ph]. URL: https://arxiv.org/abs/2610.21475.
- [8]V. Jaks̆ić, C.-A. Pillet, and A. Szczepanek. Dynamical conservation laws in quantum spin systems. arXiv:2610.21489. 2026. arXiv: 2610. 21489 [math-ph]. URL: https://arxiv.org/abs/2610.21489.
- [9]V. Jaks̆ić, C.-A. Pillet, and A. Szczepanek. “Entropy maximization and weak Gibbsianity of quasi-free fermionic states”. In: Letters in Mathematical Physics 116.4 (2026), p. 103. DOI: 10.1007/s11005-026-02100-9. arXiv: 2603.14020. URL: https://doi.org/10.1007/s11005-026-02100-9.
- [10]V. Jaks̆ić, C.-A. Pillet, and C. Tauber. “A note on adiabatic time evolution and quasi-static processes in translation-invariant quantum systems”. In: Annales Henri Poincaré 25.1 (2024), pp. 751–771. DOI: 10.1007/s00023-023-01282-5. arXiv: 2204.02177. URL: https://doi.org/10.1007/s00023-023-01282-5.
- [11]V. Jaks̆ić, C.-A. Pillet, and C. Tauber. “Approach to equilibrium in translation-invariant quantum systems: Some structural results”. In: Annales Henri Poincaré 25.1 (2024), pp. 715–749. DOI: 10.1007/s00023-023-01281-6. arXiv: 2204.00440. URL: https://doi.org/10.1007/s00023-023-01281-6.
- [12]Y. Ogata and L. Rey-Bellet. “Ruelle–Lanford functions and large deviations for asymptotically decoupled quantum systems”. In: Reviews in Mathematical Physics 23.2 (2011), pp. 211–232. DOI: 10.1142/S0129055X11004291.DOI
- [13]C.-E. Pfister and W. G. Sullivan. “Weak Gibbs measures and large deviations”. In: Nonlinearity 31.1 (2018), pp. 49–53. DOI: 10.1088/1361-6544/aa99a3. arXiv: 1705.06939. URL: https://doi.org/10.1088/1361-6544/aa99a3.
- [14]C.-E. Pfister and W. G. Sullivan. “Asymptotic decoupling and weak Gibbs measures for finite alpha-bet shift spaces”. In: Nonlinearity 33.9 (2020), pp. 4799–4817. DOI: 10.1088/1361-6544/ab8fb7. URL: https://doi.org/10.1088/1361-6544/ab8fb7.
- [15]D. Ruelle. Statistical Mechanics: Rigorous Results. Mathematical Physics Monograph Series. New York: W. A. Benjamin, 1969. ISBN: 0805383603.
- [16]N. Shiraishi and H. Tasaki. “The S = ½ XY and XYZ Models on the Two- or Higher-Dimensional Hypercubic Lattice Do Not Possess Nontrivial Local Conserved Quantities”. In: Annales Henri Poincaré (2026). DOI: 10.1007/s00023-026-01699-8. arXiv: 2412.18504 [cond-mat.stat-mech].
- [17]K. Tapp. “On the minimum spanning tree distribution in grids”. In: European Journal of Combinatorics 133 (2026), p. 104325. ISSN: 0195-6698. DOI: https://doi.org/10.1016/j.ejc.2025.104325. URL: https://www.sciencedirect.com/science/article/pii/S0195669825002185.
- [18]H. Tasaki. Trees that can be grown in “too many” ways: Bouch’s construction and a bound for every size. Version 4. Originally posted in December 2024. 2026. DOI: 10.48550/arXiv.2412.16912. arXiv: 2412.16912 [math-ph]. URL: https://arxiv.org/abs/2412.16912v4.
- [19]M. Yuri. “Weak Gibbs measures for certain non-hyperbolic systems”. In: Ergodic Theory and Dynamical Systems 20.5 (2000), pp. 1495–1518. DOI: 10.1017/S014338570000081X. URL: https://doi.org/10.1017/S014338570000081X.
- [20]M. Yuri. “Weak Gibbs measures and the local product structure”. In: Ergodic Theory and Dynamical Systems 22.6 (2002), pp. 1933–1955. DOI: 10.1017/S0143385702000780. URL: https://doi.org/10.1017/S0143385702000780.