{"id":"42986fc7-7de3-473f-baef-54489419f254","arxiv_id":"1908.05283","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A many-body localization landscape satisfies Agmon inequalities and a resonance-free locator expansion, yielding analytical evidence for weak many-body localization in arbitrary dimension.","lead":"This paper generalizes the single-particle localization landscape to interacting many-body systems, defining a many-body localization landscape (MBLL) on the Fock-space graph and proving Agmon-type exponential decay bounds plus a convergent locator expansion. It argues that weak hopping and interactions preserve localization for a subset of Fock-space states in any spatial dimension, giving a new analytical tool for many-body localization.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central 'proof' of surviving localization hinges on an unproven geometric premise: a positive fraction of eigenstates must sit in wells with Agmon separation S≫log N, and the paper only offers the heuristic H1/H2/H3 argument of Sec. IV.B.","rationale":"I reviewed the argument chain: landscape bound (Thm III.4), Agmon decay (Lemma IV.2/Thm IV.3), locality theorems (Thm IV.5/IV.6), and convergent locator expansion (Thm V.1). The most load-bearing link is the passage from convergence of u to persistence of localized eigenstates. That passage requires the wells of 1/u to be separated by large Agmon distance for a non-negligible set of states, and the quantitative fraction is never established. I agree with the reader's identification of this as the central soft spot. Minor caveat: the locality bound only needs S ≳ C log N (not N^{1/2} log N); the paper's Eq. (41) would achieve this with path length n ~ N^{1/2} log N, but even this weaker requirement is unproven for a positive fraction. The projection-operator sign inconsistency in Thm IV.5 is a real technical error that should be corrected, but it appears salvageable; the missing energy-shift control (§V.A) is a symptom of the same gap, since without E' one cannot certify S for the perturbed wells. The paper's own language ('conceivably', 'assuming the energy shift is not too large') acknowledges the missing ingredient. Because the concern is a fixable/in-principle-fillable gap rather than a demonstrated contradiction, the reader's CONDITIONAL verdict is appropriate; I recommend no change.","tokens_in":26619,"tokens_out":26579,"duration_ms":278085,"concrete_test":"Numerical scaling check for the spin model (2) on 1D chains L=10,12,14,16 at half filling, i.i.d. disorder ε_i∈[0,W] (e.g., W=2), and small fixed t,V: compute the exact MBLL u=(H+K)^{-1}1 and the lowest ~10% of eigenstates. For each such eigenstate, construct the classically allowed set {α: 1/u_α ≤ E'}, find the connected Fock-graph component containing the state's largest amplitude, and compute the minimal Agmon distance S to every other component using the metric of Eq. (32). Evaluate the locality error bound B = C N^{3/2} δ^{-1} e^{-S/2} with δ the local mean level spacing. If the fraction of eigenstates with B<0.1 decays with L, the S-growth premise fails; if it is bounded away from zero, the reader's concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's headline conclusion (§VI) is that the convergent locator expansion plus the locality theorems 'rigorously establishes' a weak MBL. The locality theorems IV.5/IV.6 have error bounds O(N^{3/2}δ^{-1})e^{-S/2}, so they are useful only when the Agmon distance S between distinct wells of 1/u grows faster than a few times log N. Section IV.B does not prove this. Eq. (41) estimates S ~ n/N^{1/2} along a 'typical' path, and the three-region H1/H2/H3 picture (Fig. 2) is explicitly heuristic: the text says exponential decay 'conceivably occurs for a fraction of the Hilbert space,' and the fraction is never quantified. The convergent locator expansion (Theorem V.1) controls u multiplicatively but does not by itself control S, nor the many-body energy shift E'; §V.A concedes the energy shift is not determined and the persistence argument assumes the shift is not too large. If S = O(1) for the relevant states, the exponential factors in Theorems IV.5/IV.6 become order one and the claimed persistence is unsupported. This is the load-bearing premise that would have to be true for the central claim to hold.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper generalizes the single-particle localization landscape of Filoche and Mayboroda to interacting many-body lattice models, defining a many-body localization landscape (MBLL) u on the Fock-space graph via (H+K)u=1. The authors prove a landscape bound on eigenstates (Theorem III.4), derive Agmon-type decay estimates in Fock space (Lemma IV.1 and Theorem IV.3), prove two locality theorems (Theorems IV.5 and IV.6) relating well-restricted eigenstates to global eigenstates, and prove convergence of a resonance-free locator expansion for the landscape (Theorem V.1). They argue that these results, taken together, establish that a weak form of many-body localization survives weak hopping and weak interactions for a subset of the Hilbert space in any physical dimension.","tokens_in":26934,"tokens_out":10557,"duration_ms":97199,"significance":"If the central persistence claim were rigorously established, this would be a notable contribution to the many-body localization literature, one of few analytic results in arbitrary dimension. The paper’s mathematical core—Theorem III.4, the Agmon inequality in Lemma IV.1, and the convergent locator expansion for the landscape—appears correct and is presented clearly. The paper is also commendably explicit about several limitations, including the undetermined energy shift and the unquantified fraction of localized states. However, the conclusion overstates what the proofs actually deliver: the exponential decay bounds are conditional on geometric assumptions about the Agmon distance that are only discussed heuristically, and the paper’s own text flags that the fraction of localized states is unknown.","major_comments":[{"comment":"The definition of the projection Pψ(µ−δ,µ+δ) in Theorem IV.5 (and identically Pφ(µ−δ,µ+δ) in Theorem IV.6) is internally inconsistent. The text states that P projects onto eigenstates with eigenvalues between µ−δ and µ+δ, but the set A in Eq. (48) is defined by |λa−µ| ≥ δ, i.e., the complement of that interval. In the proof, A is correctly used as the set of eigenvalues outside the interval, and the bound is derived for the component of |φ⟩ on A, which is then written as ‖(I−Pψ(...))|φ⟩‖². This means the intended P is the projection onto the interval, and the definition of A should read |λa−µ| ≤ δ. As written, the theorem statement is self-contradictory and the meaning of the bound is ambiguous. Both theorems need to be corrected.","section":"Theorem IV.5 and Theorem IV.6"},{"comment":"The paper’s headline claim that the results “prove” that localization persists for a part of the Hilbert space is not supported by the mathematics presented. The error bounds in Theorems IV.5 and IV.6 are O(N^{3/2}δ^{-1})e^{-S/2}, so they are useful only if the Agmon distance S between distinct wells grows faster than logarithmically in N. Section IV.B does not prove this: Eq. (41) gives a heuristic estimate S ∼ n/N^{1/2} along a typical path, the three-region H1/H2/H3 decomposition (Fig. 2) is explicitly heuristic, and the text states that exponential decay “conceivably occurs for a fraction of the Hilbert space” without quantifying that fraction. The conclusion in Section VI that the convergent locator expansion combined with the locality theorem “allows us to prove” persistence of MBL therefore overreaches. The conclusion should be weakened to “provides analytical evidence,” consistent with the abstract and with the paper’s own caveats.","section":"Section IV.B and Section VI"},{"comment":"The persistence argument relies on an additional uncontrolled assumption: that the many-body energy shift E′ is O(λN) and small enough that the classically allowed regions do not expand significantly. The paper explicitly concedes that “determining how the many-body energies shift under even a small perturbation is not immediately obvious” and that “we cannot fully determine the energy shifts due to interaction.” Consequently, the proof of persistence is conditional on an uncontrolled quantity. This limitation is acknowledged in the body, but it is not reflected in the conclusion’s “prove” language, and it should be.","section":"Section V.A"}],"minor_comments":[{"comment":"The exponential factor is typeset as e^{-Sα/√ε}; the derivation in the proof gives e^{-Sα}/√ε. Please clarify which expression is intended and correct the display accordingly.","section":"Theorem IV.3, Eq. (35)"},{"comment":"The text refers to a “Caylee graph”; this should read “Cayley graph.”","section":"Introduction"},{"comment":"The definitions of the regions H1, H2, and H3 are informal and are used in a heuristic argument. Consider labeling them explicitly as heuristic and providing more precise definitions or bounding the fraction of states in each region.","section":"Section IV.B"},{"comment":"The proof concludes by “choosing λ small enough” to make cλ/δ < 1, but λ is a bound on matrix elements of T+V; the smallness condition is on the physical parameters t and V. Please rephrase to avoid ambiguity.","section":"Theorem V.1 proof"}],"recommendation":"major_revision","confidential_remarks":"The paper’s core mathematical results are correct, and the MBLL framework is a genuinely novel contribution that is likely to interest the MBL community. The main concern is the gap between the proved bounds and the claimed conclusion; this gap is fixable in revision by softening the claims and correcting the projection definitions. The internal contradiction in Theorems IV.5/IV.6 is serious but appears to be a typographical error. I would not recommend rejection, but the paper needs substantial revision to align its conclusions with what is actually proven."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear [Colleague],\n\nHere is the short version: this paper has one genuinely new idea and one load-bearing gap. The new idea is that the locator expansion for the many-body localization landscape has no resonances, unlike the conventional Green's-function expansion, and can be bounded by a simple geometric series. That is real and worth taking seriously. The gap is that the paper's headline—'rigorously establishes a weak version of MBL'—does not follow from the theorems as written, because the locality theorems are only useful when the Agmon distance S between distinct wells grows with system size, and the paper never proves that a positive fraction of states sits in such wells.\n\nWhat is good: the generalization of the single-particle landscape to Fock space is natural, and the basic bound |ψα| ≤ E' uα max|ψ| carries over cleanly. The Agmon machinery on the Fock graph, including Lemma IV.1 and the weighted-Laplacian formulation, is a real extension of Mayboroda's work. The convergence proof for the locator expansion, while rough in places, is a solid first step. The paper is also unusually honest about its limitations: it explicitly says it cannot determine energy shifts, cannot quantify the fraction of localized states, and does not prove full MBL.\n\nThe soft spots, in order of importance. First, the central claim of persistence is conditional on an unproven geometric premise. The bounds in Theorems IV.5 and IV.6 are O(N^{3/2}/δ) e^{-S/2}; for them to be useful you need S ≳ log N. Section IV.B's H1/H2/H3 argument is explicitly heuristic ('conceivably occurs for a fraction of the Hilbert space'), and the fraction is never quantified. Without a proof that a positive fraction of eigenstates lie in wells separated by S ≳ log N, the 'rigorously establishes' statement in the conclusion is not supported. Second, Theorem IV.5 contains a projection-operator inconsistency: P is defined as projecting onto eigenvalues between µ−δ and µ+δ, but the set A is defined by |λa−µ| ≥ δ. The proof uses the 'outside' definition, so the statement's wording is wrong and needs fixing. Third, the energy-shift control is missing: the persistence argument assumes the shift is not too large, but the paper only gives a hand-wavy O(λN) estimate, which is not enough when N is large.\n\nWho this is for: people working on MBL and the localization landscape. The resonance-free locator observation is the kind of thing that might seed better proofs, and the framework as a whole is likely to be cited, even if the central claim needs to be downgraded to 'conditional evidence' until the geometric premise is established.\n\nMy recommendation: send this to a serious referee, not a desk reject. Tell the authors to fix the projection typo and either prove or explicitly caveat the S-growth assumption.\n\nBest,\n[You]","headline":"Genuinely new resonance-free locator expansion for the Fock-space landscape, but the claim of rigorously establishing weak MBL rests on an unproven Agmon-distance premise.","tokens_in":27392,"tokens_out":5590,"would_cite":true,"duration_ms":53737,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["71.30.+h","72.15.Rn","05.30.-d"],"model":"deepseek-v4-flash","headline":"This paper proves a weak version of many-body localization for a class of disordered lattice models: under weak hopping and interactions, a subset of exact eigenstates remain exponentially localized in Fock space.","keywords":["many-body localization","localization landscape","Fock-space graph","Agmon estimates","locator expansion","disordered interacting systems","mobility edge","eigenfunction decay"],"falsifier":"Take a disordered one-dimensional chain with fixed disorder strength, small hopping, and weak interactions, compute the MBLL by inverting (H+K) on the Fock-state graph for increasing N, and measure the minimal Agmon action separating inverse-landscape wells for eigenstates near the band edge. If the fraction of states with well separation growing with N tends to zero, or if that separation typically saturates, the locality bounds become order-one and the proof of persistent localized states would not be supported.","tokens_in":26450,"feed_emoji":"🔒","tokens_out":8132,"duration_ms":82846,"temperature":0.7,"pith_summary":"The paper proves that for disordered lattice models with short-range hopping, adding weak interactions does not destroy exponential localization for a subset of many-body eigenstates. It lifts the single-particle localization landscape, the solution to (H+K)u=1, to the graph of Fock-space basis states, where the inverse landscape acts as an effective confining potential. Exact Agmon-type decay bounds and two locality theorems then show that eigenstates concentrate in isolated \"wells\" of the inverse landscape, and a locator expansion for the landscape converges without the resonances that plague standard locator expansions. If correct, this establishes a weak, partial form of many-body localization in any physical dimension while leaving full-spectrum MBL open.","feed_headline":"Weak many-body localization survives weak interactions","feed_subtitle":"A Fock-space landscape shows a subset of eigenstates stay exponentially localized in any dimension.","key_machinery":"The machinery runs on three pieces. The MBLL itself, u_alpha = sum_beta (H+K)^{-1}_{$\\alpha$ $\\beta$}, is the effective-potential carrier: its inverse 1/u_alpha acts as a confining potential in Fock space, and the weighted Laplacian with weights u_alpha u_beta rewrites the eigenvalue equation as a discrete effective Schroedinger equation. The generalized Agmon metric S_alpha, defined as the infimum of path actions in the Fock graph, controls exponential decay, with the graph momentum p_alpha ~ $\\sqrt$((1/u_alpha - E')/deg($\\alpha$)) setting the decay rate. The locator expansion for u_alpha, in powers of (T+V)$U^{{-1}}$, is the convergence engine; because denominators are E_alpha rather than E - E_alpha, it is resonance-free and bounded by a geometric series for short-range hopping.","core_discovery":"The paper's central claim is that a weak form of many-body localization is rigorously established for a class of models: on a disordered lattice or graph with short-range hopping, weak interactions do not destroy exponential localization of a subset of many-body eigenstates. The subset is characterized through the many-body localization landscape, u_alpha, the solution to (H+K)|u>=|1> in Fock space; Theorem III.4 bounds every eigenstate by |psi_alpha| <= (E+K)u_alpha max|psi|. Agmon inequalities on the Fock-state graph show exponential decay of eigenstates in the classically forbidden regions of the inverse landscape. The decisive step is the locator expansion for u_alpha: its denominators are bare energies rather than energy differences, so no resonances occur and the series converges for small hopping and interactions. When the wells of the inverse landscape remain separated, the locality theorems imply that localization survives, giving a weak MBL statement for at least part of the Hilbert space in any dimension.","pith_inferences":["Beyond the paper, one can test its geometric premise numerically: compute the MBLL for small disordered chains and measure the Agmon distance between wells; if that distance does not grow with system size for a positive fraction of states, the exponential bounds become order-one and the proof would not extend.","The resonance-free property of the landscape locator series suggests that other inverse-matrix observables, not just Green's functions, may admit convergent perturbative expansions, potentially offering numerical MBL diagnostics that avoid self-energy resummation.","If the inverse-landscape/spectrum conjecture discussed in the paper transfers to Fock space, the convergent locator expansion would likely control energy shifts and level statistics, upgrading this weak MBL result to a spectral statement; that transfer is not part of the paper's proof.","A natural next step, mentioned but not developed in the paper, is to start from the non-interacting localized basis and expand the MBLL in the interaction alone; this would make the landscape informative at larger hopping strengths."],"forward_implications":["If the central claim is correct, a rigorous weak-MBL statement holds for arbitrary physical dimension, not only one-dimensional chains.","For disorder realizations whose inverse-landscape wells do not percolate below an energy threshold, small hopping and interactions cannot destroy exponential localization of the associated states.","The method yields a landscape that is cheaper to compute than exact diagonalization, enabling localization-structure tests on larger systems.","The result does not establish full-spectrum MBL; the paper explicitly leaves open the possibility of ergodic behavior in the middle of the band.","Higher connectivity of the Fock graph in higher dimensions suggests a percolation-driven mobility edge at intermediate energies."],"supporting_citations":[{"why":"introduces the single-particle localization landscape whose many-body generalization this paper constructs and whose bound |psi| <= Eu max is extended here.","marker":"[21]"},{"why":"provides the discrete single-particle landscape construction used for the positivity lemmas and Theorem III.4.","marker":"[22]"},{"why":"supplies the Agmon-type estimates with 1/u as an effective potential that the paper generalizes to Fock space.","marker":"[23]"},{"why":"contains the single-particle locality theorem and spectral-counting corollary that the paper imitates in Fock space.","marker":"[24]"},{"why":"gives the original Agmon exponential-decay inequalities for Schroedinger eigenfunctions, the basis of the Fock-graph decay bounds.","marker":"[40]"},{"why":"extends Agmon's method and is cited alongside [40] for the exponential-decay machinery generalized here.","marker":"[41]"},{"why":"introduces the locator expansion whose resonance problem the landscape perturbative series avoids.","marker":"[1]"},{"why":"defines many-body localization in Fock space and supplies the Basko-Aleiner-Altshuler scenario that the weak-MBL result sits inside.","marker":"[2]"}],"fun_headline_variants":["Many-body localization survives weak interactions in any dimension","Fock-space landscape proves weak MBL for a subset of states","Landscape method proves weak MBL in any dimension for some states","Weak interactions don't break MBL for a Fock-space subset","MBL persists for subset of states despite weak interactions"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is geometric: the exponential bounds are useful only if the distance between distinct wells grows with system size for a positive fraction of states; the paper argues this heuristically from typical paths but does not prove or quantify that fraction.","fun_headline_variants_meta":{"raw":{"variants":["Many-body localization survives weak interactions in any dimension","Fock-space landscape proves weak MBL for a subset of states","Landscape method proves weak MBL in any dimension for some states","Weak interactions don't break MBL for a Fock-space subset","MBL persists for subset of states despite weak interactions"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000879,"raw_usage":{"total_tokens":3865,"prompt_tokens":1073,"completion_tokens":2792,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":689,"completion_tokens_details":{"reasoning_tokens":2708}},"tokens_in":689,"tokens_out":2792,"duration_ms":19139,"temperature":1.0,"reasoning_tokens":2708,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:19:40.959208+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a disordered one-dimensional chain with fixed disorder strength, small hopping, and weak interactions, compute the MBLL by inverting (H+K) on the Fock-state graph for increasing N, and measure the minimal Agmon action separating inverse-landscape wells for eigenstates near the band edge. If the fraction of states with well separation growing with N tends to zero, or if that separation typically saturates, the locality bounds become order-one and the proof of persistent localized states would not be supported.","supporting_citations":[{"cited_title":"Filoche and S","cited_arxiv_id":null,"evidence_quote":"introduces the single-particle localization landscape whose many-body generalization this paper constructs and whose bound |psi| <= Eu max is extended here."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"provides the discrete single-particle landscape construction used for the positivity lemmas and Theorem III.4."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"supplies the Agmon-type estimates with 1/u as an effective potential that the paper generalizes to Fock space."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"contains the single-particle locality theorem and spectral-counting corollary that the paper imitates in Fock space."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"gives the original Agmon exponential-decay inequalities for Schroedinger eigenfunctions, the basis of the Fock-graph decay bounds."},{"cited_title":"Agmon, in Schr¨ odinger operators(Springer, 1985) pp","cited_arxiv_id":null,"evidence_quote":"extends Agmon's method and is cited alongside [40] for the exponential-decay machinery generalized here."}],"review_version":1}