Eigenstate Thermalisation Hypothesis for Translation Invariant Spin Systems

We prove the Eigenstate Thermalisation Hypothesis (ETH) for local observables in a typical translation invariant system of quantum spins with L-body interactions, where L is the number of spins. This mathematically verifies the observation first made by Santos and Rigol (Phys Rev E 82(3):031130, 2010, https://doi.org/10.1103/PhysRevE.82.031130) that the ETH may hold for systems with additional translational symmetries for a naturally restricted class of observables. We also present numerical support for the same phenomenon for Hamiltonians with local interaction.


Introduction
Recent experiments have demonstrated thermalisation of isolated quantum systems under unitary time evolution [2][3][4][5][6][7].In this context, thermalisation means that, after a long time evolution, observables attain their equilibrium (thermal) values determined by statistical mechanics.The primary mechanism behind this thermalisation of isolated quantum systems is an even stronger concept, the Eigenstate Thermalisation Hypothesis (ETH) [8][9][10].Informally, the ETH asserts that (i) physical observables A take their thermal value on every eigenstate of a many-body quantum system and (ii) off-diagonal elements of A in the energy eigenbasis are vanishingly small.In particular, the ETH ensures the thermalisation of A for any initial state with a macroscopically definite energy, given no massive degeneracy in the energy spectrum [11][12][13].The ETH has numerically been verified for individual models with several local or few-body observables [1,[14][15][16][17][18][19].On the other hand, recent studies have revealed several classes of systems for which the ETH breaks down: examples include systems with an extensive number of local conserved quantities [20][21][22], many-body localisation [23,24], and quantum many-body scars [25,26].
As another approach to this question, it has been proven that the ETH holds true for any deterministic observable for almost all Hamiltonians H [8,27,28] sampled from a Wigner matrix ensemble which has no further unitary symmetry (see also [29,30] for ETH for more general mean field ensembles).If the Hamiltonian has some unitary symmetry, the ETH clearly breaks down for conserved quantities related to those symmetries because we can find simultaneous eigenstates of the Hamiltonian and conserved quantities.However, Ref. [1] observed an interesting phenomenon, namely that local quantities still satisfy the ETH even in a system with translational symmetry.Therefore, the question of how generically and for what class of observables the ETH holds true in realistic situations has yet to be fully resolved.
In this paper we mathematically rigorously prove an instance of the observation from [1].More precisely we show that, for the mean-field case of an ensemble with translational symmetry, the ETH typically holds for quantities whose support does not exceed half of the system size with the optimal speed of convergence.The ETH also typically holds for quantities whose support exceeds half the system size but with a slower convergence speed, while it typically breaks down for some observables whose support extends to the entire system.We complement our analytical results for the mean-field case with a numerical simulation for an ensemble of more realistic Hamiltonians with local interactions.

Setup
We consider a one-dimensional periodic quantum spin system on the L ∈ N sites of the standard discrete torus T L := Z LZ .On each vertex j ∈ T L , the one particle Hilbert space H j is given by C 2 and we denote its canonical basis by {|↑ , |↓ }.The corresponding L-particle Hilbert space is simply given by the tensor product with dimension 2 L .For simplicity, we restrict ourselves to the spin-1/2 case, but our results can straightforwardly be extended to general spin s with one particle Hilbert space being C 2s+1 .
Next, we introduce the ensemble of Hamiltonians, which is first introduced in Ref. [31] and shall be studied in this article.The main parameter in the definition is a tunable range ≤ L of interactions, which allows us to consider how generically the ETH holds in realistic situations.Definition 2.1 (Hamiltonian).Let T = T L be the (left) translation operator acting on L spins at the vertices of T L .We define the ensemble of Hamiltonians with local interactions as H ( ) where ≤ L is the interaction range, I L− is the identity on the sites + 1, . . ., L.
Here σ (p) j is the p th Pauli matrix σ (p) acting on the site j ∈ T L , where we recall the standard Pauli matrices, 2) The 4 coefficients J p1,...,p are independent, identically distributed real Gaussian random variables with zero mean, EJ p1,...,p = 0, and variance The ensemble of Hamiltonians h (2.1) contains prototypical spin models such as the XYZ model, h = Observe that the Hamiltonian H L .We impose this structure to study a Hamiltonian with a symmetry.In the sequel we shall exploit this feature of H ( ) L by switching from position space to momentum space.Lemma 2.2.Let be the projection operator onto the k-momentum space, i.e., L is block-diagonal in the momentum space representation, i.e. in the eigenbasis of T L , since we have (2.4) Proof.This follows immediately by substituting the spectral decomposition of T given by T = L−1 k=0 e 2πik/L Π k into (2.1).As we will show in Lemma 3.4, the dimensions of each of the L momentum sectors are almost equal to each other, tr In order to present our main result, the ETH in translation-invariant systems (Theorem 3.1), in a concise form, we need to introduce the microcanonical average.Below, we denote by α .It is easy to see that the spectrum of H in each momentum sector is simple almost surely.Definition 2.3 (Microcanonical ensemble).For every energy E ∈ R and energy window ∆ > 0, we define the microcanonical energy shell H E,∆ centered at energy E with width 2∆ by We denote the dimension of (2.5) Remark 2.4.The microcanonical average mimics the microcanonical ensemble before taking the thermodynamic limit.In order to be physically meaningful, there are two natural requirements on the energy shell H E,∆ : (i) The density of states is approximately constant in the interval (ii) The microcanonical energy shell contains 1 states, i.e. d E,∆ → ∞ as L → ∞.
Note that for any fixed energy E, (i) corresponds to an upper bound and (ii) corresponds to a lower bound on ∆, both being dependent on E. We point out that very close to the spectral edges with only a few states, it is not guaranteed that both requirements can be satisfied simultaneously.Indeed, from a physics perspective, viewing A (mc) ∆ (E) from (2.5) as a finite dimensional approximation of the microcanonical ensemble is meaningless whenever (i) and (ii) are not satisfied.However, we will simply view Definition 2.3 for arbitrary ∆ as an extension of the standard definition of the microcanonical average from the physics literature.Our main result, Theorem 3.1, will even hold with the microcanonical average in this extended sense.
We set for the total Hilbert space dimension.Our analytic results below will always be understood in the limit of large system size, i.e.L → ∞, or, equivalently N → ∞.We shall also use the following common notion (see, e.g., [32]) of stochastic domination.
Definition 2.5.Given two families of non-negative random variables indexed by N , we say that X is stochastically dominated by Y , if for all ξ, D > 0, we have sup for any sufficiently large N ≥ N 0 (ξ, D) and use the notation X ≺ Y or X = O ≺ (Y ) in that case.
3 Main result in the mean-field case Throughout the entire section, we are in the mean-field case = L.For any q ≤ L we also introduce the concept of q-local observables for self-adjoint operators of the form A = A q ⊗ I L−q , i.e.A q is self-adjoint and only acts on the first q sites.Our main result in this setting is the following theorem.α .Then, for every ∆ > 0 and bounded q-local observable where the maxima are taken over all indices labeling the eigenvectors of L .In particular, for q ≤ L/2 the ETH holds with optimal speed of convergence of order 1/ √ N .
An extension of Theorem 3.1 to arbitrary dimension d ≥ 2 is provided in Theorem A.3 in the Appendix.Remark 3.2 (Typicality of ETH).Theorem 3.1 asserts that for any fixed local observable A the ETH in the form (3.1) holds with a very high probability, i.e. apart from an event of probability N −D = 2 −LD , for any fixed D, see the precise Definition 2.5.This exceptional event may depend on the observable A. However, as long as q is L-independent (in fact some mild logarithmic increase is allowed), it also holds that i.e. we may take the supremum over all bounded q-local observables A in (3.1).This extension is a simple consequence of choosing a sufficiently fine grid in the unit ball of the 4 q × 4 q dimensional space of q-local observables and taking the union bound.The estimate (3.2) can be viewed as a very strong form of the typicality of ETH within our class of translation invariant mean field operators L .It asserts that apart from an exceptional set of the coupling constants J p1,...,p L the Hamiltonian H (L) L satisfies the ETH with optimal speed of convergence, uniformly in the entire spectrum and tested against all finite range (q-local) observables.The exceptional set has exponentially small measure of order 2 −LD for any D if L is sufficiently large.
In Lemma 3.5 we will see that in the mean-field case the Hamiltonian on each momentum sector is a GUE matrix, in particular the density of states of H follows Wigner's semicircle law.An elementary calculation shows that the radius of this semicircle is given by In light of Remark 2.4 we also mention that A (mc) ∆ (E) in (3.1) can be considered as an approximation of the expectation of A in the microcanonical ensemble at energy The upper bound in (3.3) comes from requirement (i) in Remark 2.4, while the lower bound in (3.3) stems from (ii) using that the eigenvalue spacing near the spectral edge for Wigner matrices is of order R/N 2/3 .
For the sequel we introduce the notation for the normalised trace of an operator A on any finite-dimensional Hilbert space, where I is the identity on that space.In particular, if The proof of Theorem 3.1 crucially relies on the fact that in our mean-field case In other words, the thermodynamics of the system is trivial; the thermal value of A is always given by its average trace.This is formalised in the following main proposition: Proposition 3.3.Under the assumptions of Theorem 3.1 it holds that (3.4) Having Proposition 3.3 at hand, we can readily prove Theorem 3.1.
Proof of Theorem 3.1.Averaging (3.4) for α = β and k = k according to the microcanonical average (2.5), we find that max Combining this with (3.4), the claim immediately follows.
The rest of this section is devoted to the proof of Proposition 3.3, which is conducted in four steps.
1.The momentum sectors are all of the same size with very high precision (Lemma 3.4).

In each momentum sector the mean-field Hamiltonian H (L)
L , represented in the eigenbasis of the translation operator T , is a GUE matrix (Lemma 3.5).3. The ETH holds within each momentum sector separately (Lemma 3.6).4. The averaged trace on each momentum sector and the total averaged trace are close to each other -at least for local observables (Lemma 3.7).
We shall first formulate all the four lemmas precisely and afterwards conclude the proof of Proposition 3.3.Lemma 3.4 (Step 1: Dimensions of momentum sectors).The dimension tr L Π k of the k-momentum sectors (k = 0, . . ., L − 1) is almost equal to each other in the sense that we have The proof is given in Section 3.1 Lemma 3.5 (Step 2: GUE in momentum blocks).Each momentum-block of the meanfield Hamiltonian H (L) L , represented in an eigenbasis of T , is an i.i.d.complex Gaussian Wigner matrix (GUE), whose entries have mean zero and variance Proof.In the mean-field case = L, a simple direct calculation of all first and second moments of the matrix elements shows that the interaction matrix h is a complex Gaussian Wigner matrix whose entries have variance 2 L v 2 L .Since the transformation from the standard basis to an eigenbasis of T is unitary, and the Gaussian distribution is invariant under unitary transformation, h represented in an eigenbasis of T is again a Gaussian Wigner matrix.Finally, the projection operators Π k in (2.4) set the offdiagonal blocks to zero.Incorporating the additional factor L in (2.4) into the variance proves Lemma 3.5.
As the next step, we show that the ETH holds within each momentum sector.Lemma 3.6 (Step 3: ETH within each momentum sector).For an arbitrary deterministic observable A with A 1 it holds that (3.5) Proof.For any fixed k, Lemma 3.5 asserts that Π k H (L) L Π k is a standard GUE matrix (up to normalisation by v L ).Using [28, Theorem 2.2], therefore its eigenvectors satisfy ETH in the form that E α |A|E β is approximately given by the normalised trace of A in the k-momentum sector with very high probability and with an error given by the square root of the inverse of the dimension of the k-momentum sector, 1/ √ tr L Π k .This holds in the sense of stochastic domination given in Definition 2.5.Using that tr L Π k ≈ 2 L /L from Lemma 3.4, we obtain that (3.5) holds for each fixed k, uniformly in all eigenvectors.Finally, the very high probability control in the stochastic domination allows us to take the maximum over k = 1, 2, . . ., L by a simple union bound.This completes the proof of (3.5).
We remark that the essential ingredient of this proof, the Theorem 2.2 from [28], applies not only for the Gaussian ensemble but for arbitrary Wigner matrices with i.i.d.entries (with some moment condition on their entry distribution) and its proof is quite involved.However, ETH for GUE, as needed in Lemma 3.6, can also be proven with much more elementary methods using that the eigenvectors are columns of a Haar unitary matrix.Namely, moments of E α |A|E β can be directly computed using Weingarten calculus [33].Since in (3.5) we aim at a control with very high probability, this would require to compute arbitrary high moments of E α |A|E β − δ α,β A k .The Weingarten formalism gives the exact answer but it is somewhat complicated for high moments, so identifying their leading order (given by the "ladder" diagrams) requires some elementary efforts.For brevity, we therefore relied on [28, Theorem 2.2] in the proof of Lemma 3.6 above.
Finally, we formulate the fourth and last step of the proof of Proposition 3.3 in the following lemma, the proof of which is given in Section 3.2.Lemma 3.7 (Step 4: Traces within momentum sectors).Let A = A q ⊗ I L−q be an arbitrary q-local observable with A 1. Then it holds that Moreover, for q > L/2 + 1 this bound is optimal (up to the factor L).
Armed with these four lemmas, we can now turn to the proof of Proposition 3.3.
Proof of Proposition 3.3.First, for any q-local observable A = A q ⊗I L−q , we conclude from Lemma 3.6 and Lemma 3.7 that max Combining (3.7) with (3.8), we have proven Proposition 3.3.

Dimensions of momentum sectors: Proof of Lemma 3.4
In this section we prove Lemma 3.4, and establish that the sizes of the momentum sectors are almost equal.To this end, we show that the leading term in the size of each of the momentum blocks is given by the number of aperiodic elements in the product basis of H.We present the proof using group theory notation, which is not strictly necessary for the one-dimensional case under consideration since the translation group of the torus T L is cyclic.Nevertheless, we do it to allow for a more straightforward generalisation to the d-dimensional case (cf.Lemma A.4).
Proof.We introduce the following objects.Let S denote the canonical product basis of H, S(L) and let G be the group of translations of T L generated by T = T L .Note that G is a finite cyclic group of size |G| = L.The action of G on S(L) is defined by where M (L) := |S 1 (L)| denotes the number of elements in S(L) with a trivial stabilizer.The last inequality follows from the fact that L has at most O(L 1/2 ) divisors.Since M (L) ≤ 2 L , we conclude from (3.12) that For any k ∈ {0, . . ., L − 1}, we can construct an eigenvector of T corresponding to the eigenvalue e 2πik/L by defining Since the orbit of σ under T consists of L distinct basis elements, the vector v(σ, k) is non-zero.Furthermore, the vectors v(σ, k) and v(σ , k) corresponding to σ and σ in disjoint orbits are linearly independent because they share no basis element.Therefore, the dimension of the k-th momentum space is bounded from below by the number of disjoint orbits in S 1 (L), that is where we used inequality (3.13) and the fact that all orbits in S 1 (L) have size L. By means of (3.15), we obtain the following chain of inequalities which, together with (3.15) concludes the proof of Lemma 3.4.

Traces within momentum sectors: Proof of Lemma 3.7
In this section, we give a proof of Lemma 3.7, which evaluates the difference of the noramalised trace tr L (Π k AΠ k )/ tr L Π k on a momentum sector and the full normalised trace A for a q-local observable A = A q ⊗ I L−q .We separate A into the tracial part A I and the traceless part Å := A − A I.
Proof of Lemma 3.7.
Then, the task is to evaluate the size of the quantity tr L (T −j L Å). Lemma 3.8.Let A := A q ⊗ I L−q be a q-local observable with A 1.Then, for any j = 1, . . ., L − 1, we have where gcd stands for the greatest common divisor.

It remains to give the proof of
To count the number of independent summations in the right-hand side of (3.19) and obtain an upper bound for tr L (T −j L A) with j = 1, . . ., L − 1, we count the number of independent deltas in the product Here, not all of the delta functions in G (L) q,j are independent in the sense that we may express G (L) q,j with a fewer number of deltas.For example, we have To obtain an expression of G (L) q,j with the minimal number of deltas, we graphically represent the product L m=q+1 δ smsm+j by arranging the sites on a circle and representing the δ smsm+j 's with a line connecting the site m and m + j (Figure 1).A minimal representation of G (L) q,j is obtained by removing exactly one delta for every occurrence of a loop in the graph of L m=q+1 δ smsm+j .The graph of L m=q+1 δ smsm+j can be obtained in two steps: First, in step (i), drawing the graph of L m=1 δ smsm+j and second, in step (ii), removing the lines corresponding to the delta functions δ smsm+j (m = 1, . . ., q), which are depicted with red dashed lines in Figure 1.
In the first step (i), there are exactly gcd(j, L) loops each starting from the sites 1, . . ., gcd(j, L).If q > gcd(j, L), there is no loop remaining after the second step (ii).Thus, we obtain a minimal representation of G (L) q,j as G (L) q,j = L m=q+1 δ smsm+j .If q ≤ gcd(j, L), the loops starting from the sites q + 1, . . ., gcd(j, L) remain after the second step (ii), for each of which we remove one delta to obtain a minimal representation of In summary, we obtain a minimal representation of G m=q+1 δs m+j sm for (a) L = 12, q = 3, j = 4 and (b) L = 12, q = 5, j = 4.For the first case (a) where q < gcd(j, L), there is a loop 4-8-12-4 remaining after the step (ii), which contains exactly one redundant delta function δs 4 s 8 depicted with a solid red line.In general, exactly one redundant delta function appears for every occurrence of a loop in the graph of L m=q+1 δs m+j sm .
Finally, we prove the optimality of (3.6) in the regime q > L/2 + 1. Lemma 3.9.Let B q := T q + T −1 q − 2 2−q I q , where T q is the (left) translation operator acting only on the first q spins arranged on the torus T q .Observe that B q is Hermitian and traceless.Then, for q > L/2 + 1, the normalised trace of B := B q ⊗ I L−q within the k-momentum sector is given by This shows that the q-local observable B q := T q + T −1 q − 2 2−q I q saturates the bound (3.6) when q > L/2 + 1.It also shows that the deviation of the normalised trace within a momentum sector, tr L (Π k BΠ k )/ tr L Π k , from B = 0, which is of order 2 −(L−q) , becomes the dominant source of error in the ETH whenever q > L/2 + 1.
Proof of Lemma 3.9.We first reduce the range of the summation over j in the generally valid expression (3.17) applied to B q .To do so, we introduce the parity operator P L defined by P L |s 1 s 2 . . .s L := |s L . . .s 2 s 1 .It satisfies P L T L P L = T −1 L and P L AP L = I L−q ⊗ (P q A q P q ) for any A = A q ⊗ I L−q .Since B q is invariant under the parity transformation, we have Therefore, we can rewrite (3.17) with the aid of (2.3) as When q > L/2 + 1, we have j < q and cannot skip over the region 1, . . ., j when going along the lines in the graph of G (L) q,j (recall (3.20)).Therefore, each line starting at one of the sites p ∈ {q + 1, . . ., q + j} passes through a point in {1, . . ., j}.Moreover, the correspondence between p and the first intersection of the line starting at p with {1, . . ., j} is one-to-one.Therefore, there exists a permutation τ j on 1, . . ., j such that s q+i = s τ (i) for i = 1, . . ., j due to G (L) q,j .With this permutation τ , we obtain tr L (T −j L B) = s1...s L s q+1 . . .s q s q+1 s q+j |B q |s 1 . . .s q G (L) q,j = s1...sq s q+1 . . .s q s τ (1) s τ (j) |B q |s 1 . . .s q = tr q (τ † j T −j q B q ) = tr q (τ † j T −(j−1) Because τ j is a j-local operator (not necessarily self-adjoint) on the q-site chain, we can apply Lemma 3.8 to each term in (3.24).Combined with j < q − 1 and gcd(j, q) ≤ j, we obtain tr L (T −j L B) = δ j1 2 q + O 2 j = δ j1 2 q + O 2 L/2 .Substituting this result into (3.23) and employing tr L Π k = 2 L L + O L 1/2 2 L/2 from Lemma 3.4, we obtain the result (3.22).
4 Numerical verification of Theorem 3.1 for = O (1) In this section, we numerically demonstrate that Theorem 3.1 also holds for the nonmean-field case of = 2.For that purpose, we adopt the following measure of the ETH used in Refs [31,34].For any self-adjoint operator A we define where a max(min) is the maximum (minimum) eigenvalue of A. Here, E denotes the average over the realisations of the Hamiltonian (2.1), and max α denotes the maximum over the eigenstates |E (k) α in the energy shell at the center of the spectrum, i.e. those α for which The width ∆ of the energy interval is set to be ∆ = 0.4/L such that it satisfies the two physical requirements mentioned in Remark 2.4 for L ≥ 6.With this choice of ∆, the microcanonical energy shell H H ,∆ defined by (4.1) typically contains more than 10 states, while the density of states does not change too much within H H ,∆ .
As the observable, we choose A = B q ⊗ I L−q with B q := T q + T −1 q − 2 2−q I q for q = 2, . . ., L, which saturates the upper bound in (3.6) and thus also saturates that of (3.2).With this choice we have a max − a min 4 for any L and q.Therefore, the ETH measure Λ is essentially the same as the diagonal part of the left-hand side of (3.2) in Theorem 3.1 -except that the maximum over α is now taken only at the center of the spectrum (and we do not take maximum over all A).This is because the eigenstate expectation value of a local observable A = A q ⊗ I L−q with q L typically acquires an energy dependence when L [35], and the number of states becomes not enough to calculate the microcanonical average near the edges for the computationally accessible system size.The ETH measure Λ satisfies reasonable thermodynamical properties.It is (i) invariant under the linear transformation A → aA + b, (ii) dimensionless, and (iii) thermodynamically intensive for additive observables A [31].
Figures 2(a)-(c) depict the L-dependence of the ETH measure Λ for different values of the parameter q.In particular, Figure 2(b) illustrates that, whenever L is approximately equal to q so that L − q < L/2, the ETH measure Λ decays as ∝ 2 −L .The rate of this decay is slower for smaller values of q, but approaches 2 −L as q becomes larger.In Figure 2(c), we take a closer look at the L-dependence of Λ for q = 6.The data indicates that for L − q L/2, Λ decays as ∝ 1.8 −L , whereas for L 2q, Λ decays as ∝ 1.8 −L/2 .These numerical observations are in agreement with our analytical results for the mean-field case in Theorem 3.1, which predicts that the exponent of the exponential decrease of Λ in L should be twice as large in the region L − q L/2 compared to the region L/2 L − q.This fact suggests that the theorem remains qualitatively valid for = O(1) in the bulk of the spectrum as long as the energy shell width is appropriately chosen.
As before, on each vertex, the one particle Hilbert space is given by C 2 with canonical basis {|↑ , |↑ }.The corresponding V -particle Hilbert space is given by For a vector q = (q 1 , . . ., q d ) ∈ T L , we introduce a rectangular subregion R q ⊂ T L by R q := x = (x 1 , . . . ,x d ) ∈ T L : 1 ≤ x s ≤ q s , s = 1, . . ., d .
A self-adjoint operator of the form A = A q ⊗ I T L \Rq is called a q-local observable, where A q is self-adjoint and acts on the Hilbert space of the spins in R q , and I T L \Rq is the identity on T L \ R q .
Finally, let T s be the (left) translation operator along the s-th coordinate acting on T L .For a vector j := (j 1 , . . ., j d ) ∈ T L , we introduce T j := d s=1 T js s .The d-dimensional version of our model in Definition 2.1 is given as follows.Definition A.1.Set the vector := ( 1 , . . ., d ) ∈ T L that determines the interaction range in each coordinate direction.We define the ensemble of Hamiltonians with local interactions as where the symbols 1, 2, . . ., label the elements of R in an arbitrary order.As in Moreover, whenever we use the notation ≺ for stochastic domination (Definition 2.5), it is always understood with N := 2 V .

A.2 Multidimensional version of the main result
The d-dimensional version of Theorem 3.1 is then given as follows. .Then, for every ∆ > 0 and bounded q-local observable A = A q ⊗ I T L \Rq with q s ≤ L s /2 for all s = 1, . . ., d, it holds that That is, the ETH holds with optimal speed of convergence.
The principal strategy for proving Theorem A.3 is exactly the same as for Theorem 3.1, which has been outlined right below Proposition 3.3.We shall hence only discuss the differences compared to the proof in Section 3, which consist solely of Step 1 (generalizing Lemma 3.4, cf.Lemma A.4) and Step 4 (generalizing Lemma 3.7, cf.Lemma A.4). Lemma A.4 (Step 1: Dimensions of momentum sectors).The dimension tr L Π k of the k-momentum sectors for k ∈ T L is almost equal to each other in the sense that we have Proof.Let S = S(L) denote the canonical product basis of H, as in (3.9), and let G be the commutative group generated by the translation operators {T s } d s=1 .The action of G on S is defined by (3.10).
In general, the group G is not cyclic, hence the subgroups of G are not uniquely determined by their size.However, S can be decomposed into a disjoint union of sets S K = S K (L) defined by S K := {σ ∈ S : G σ = K} , where G σ ⊂ G is the stabilizer of σ under the action (3.10), and K ≤ G is a subgroup of G. Similarly to (3.11), for any subgroup K of G, we define the map ϕ K : S K → T L K → {↑, ↓} , ϕ K (σ) ([x]) := σ(x) , [x] ∈ T L K , which is easily seen to be an injection and hence and injection, |S K | ≤ 2 V /|K| .Therefore, denoting the number of elements in S with a trivial stabilizer by M (L), we obtain

L
is a shifted version of the same local Hamiltonian h .In particular, H ( ) L is translation invariant by construction, i.e., T L H ( ) Theorem 3.1 (ETH in translation-invariant systems).Let = L and consider the Hamiltonian H (L) L from (2.1) with eigenvalues E (k) α and associated normalised eigenvectors |E (k)

. 10 )
In particular, the set S(L) is a disjoint union of sets S b (L) defined by S b (L) := {σ ∈ S(L) : |G σ | = b} , b = 1, 2, . . ., L , where G σ ⊂ S(L) is the stabilizer of σ under the action (3.10).By the orbit-stabilizer theorem, S b (L) = ∅ for all b that do not divide L. Since the group G is cyclic, it has a unique subgroup of size b for all b|L, given explicitly by G (b) := {T L/b , T 2L/b , . . ., T L } .Observe that each σ ∈ S b (L) corresponds to a unique map σ on a reduced torus S(L/b) := T L G (b), which is defined by σ([x]) := σ(x) , [x] ∈ S(L/b) .(3.11) Since σ is stabilised by G (b) , the map σ → σ in (3.11) is well-defined and injective.In particular, |S b (L)| ≤ 2 L/b , and hence

21 )Fig. 1
Fig.1Graphical representation of the product L m=q+1 δs m+j sm for (a) L = 12, q = 3, j = 4 and (b) L = 12, q = 5, j = 4.For the first case (a) where q < gcd(j, L), there is a loop 4-8-12-4 remaining after the step (ii), which contains exactly one redundant delta function δs 4 s 8 depicted with a solid red line.In general, exactly one redundant delta function appears for every occurrence of a loop in the graph of L m=q+1 δs m+j sm .
Theorem A.3 (ETH in d-dimensional translation-invariant systems).Let = L and consider the the Hamiltonian H (L) L from (A1) with eigenvalues E