{"id":"a4738de7-6220-4b52-af47-f11446117608","arxiv_id":"2411.14527","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":4,"one_line_summary":"In one-dimensional Stark many-body localization, the crossover field to localization decreases with system size, suggesting localization at infinitesimal electric fields, while the method misses higher-order long-time delocalization seen in tDMRG.","lead":"This paper uses a numerical method called Tensorflow Equations to study whether interacting particles in a uniform electric field remain localized in one and two dimensions. It finds that in one dimension the localized phase could survive down to arbitrarily small fields in the thermodynamic limit, but the method misses long-time delocalization that appears in benchmark simulations.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The gamma_c -> 0 inference depends on a 4th-order-truncated support diagnostic; if neglected O(U^2/gamma) terms move s, the downward drift is a truncation artifact rather than a physical trend.","rationale":"The reader's weakest_assumption identifies the same main soft spot: the static support s is computed inside a 4th-order-truncated flow, and the paper's own tDMRG results demonstrate that higher-order interaction processes are physically active at long times. This makes the truncation concern concrete rather than hypothetical. The paper is otherwise careful and honest: the TFE dynamics are benchmarked against ED and tDMRG (Figs. 3-4), the energy-error analysis in Appendix C is useful, and Section IVC explicitly admits that the TFE miss the higher-order delocalization effect. Those strengths support a conditional rather than a reject verdict. However, the abstract's phrasing goes beyond what the finite-size support data can establish, because no check shows that the neglected terms leave s unchanged. The proposed 6th-order rerun would settle whether the downward gamma_c drift is physical or an artifact. Since the reader already recommended CONDITIONAL, my stress-test does not move the verdict; it strengthens the justification for requiring the abstract to be softened and the truncation sensitivity to be quantified.","tokens_in":20085,"tokens_out":6001,"duration_ms":61890,"concrete_test":"Run the TFE flow including six-fermion (6th-order) terms for U = 1.0, gamma in 0 to 1.5, L = 12, 14, and 16 (and L = 24 if affordable), recompute s including the C^(6) coefficients in the denominator of Eq. (22), and extract gamma_c from the crossings. If gamma_c shifts upward, or if the L = 12 to L = 16 drift reverses by more than its 4th-order magnitude, the observed gamma_c -> 0 is a truncation artifact rather than a physical trend. As a supporting check, compare the same small-L s against an exact ED construction of the number operator in the eigenbasis; disagreement at the scale of the finite-size drift would invalidate the diagnostic.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section IVA and Eq. (22) define the support s of the flowed number operator using only 2nd- and 4th-order coefficients. The abstract concludes from the downward drifting crossing in Fig. 2a that localization persists at infinitesimally small field in the thermodynamic limit. The load-bearing assumption is that terms beyond the two-particle truncation of both H and n_i, estimated in Appendix C as O(U^2/gamma), do not qualitatively change s. This assumption is not established. The flow is stopped using max|H_off^(4)| criteria, not by convergence of s; no truncation-convergence tests or error bars for s are reported. The paper itself flags the risk in Section IVA: any s>0 allows two-particle effects that can generate higher-order delocalization not captured by the truncated flow, and Section IVC with tDMRG shows that higher-order processes cause finite-size delocalization at t* ~ 1/U. Because those processes are absent from the truncated operator expansion in Eq. (B2), the diagnostic may systematically undercount delocalizing channels, biasing gamma_c downward with L. Thus the abstract's 'localization at infinitesimally small field' overstates a finite-size trend that the main text only cautiously describes as 'could very well lead to gamma_c -> 0'.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript develops a numerical implementation of continuous unitary transformation flows (TensorFlow Equations, TFE) and applies it to interacting spinless fermions in a linear potential (Stark MBL) in one and two dimensions. The authors introduce methodological improvements, benchmark the resulting dynamics against exact diagonalization and tDMRG, and use the flowed number operator to define a support diagnostic s (Eq. 22) that indicates a crossover between ergodic and localized regimes. In 1D they find a finite-size transition at nonzero field that drifts downward with system size, and they argue that an extrapolation could lead to gamma_c -> 0 in the thermodynamic limit. In 2D the signatures are less clear. They also show that TFE dynamics are accurate to intermediate times t ~ 1/U^2, while tDMRG captures a higher-order, finite-size delocalization at times t* ~ 1/U that the truncated TFE misses.","tokens_in":20357,"tokens_out":4595,"duration_ms":44864,"significance":"If the central inference is correct, the paper would provide numerical support for the debated claim that the 1D Stark MBL system has no true ergodic phase for any nonzero field in the thermodynamic limit, while also demonstrating that TFE can reach system sizes beyond ED in both 1D and 2D. The strengths of the manuscript include the careful benchmarking of dynamics against ED and tDMRG in Fig. 3, the explicit comparison of generator choices in Appendix C, the machine-implementable tensor-contraction formulation, and the honest discussion of the method's limitations, including the statement in Section IVC that TFE cannot capture higher-order delocalization. The main caveat is that the static support diagnostic, which carries the paper's central claim, is computed within a 4th-order-truncated flow, and no truncation-convergence test for s is provided.","major_comments":[{"comment":"The central inference that gamma_c -> 0 is not established with respect to the truncation of the flow. The support s in Eq. (22) uses only the second- and fourth-order coefficients A and B of the flowed number operator in Eq. (B2), and the flow itself is truncated to two-particle terms as stated in Section IIIA. Appendix C estimates the neglected terms as O(U^2/gamma), while Section IVC and Fig. 5 show that higher-order interaction processes produce a finite-size delocalization at t* ~ 1/U in tDMRG simulations. The manuscript itself acknowledges in Section IVA that 'any s>0 allows for two-particle effects ... generating higher order effects that are not included in the truncated flow equations.' Therefore the observed downward drift of gamma_c in Fig. 2a could be a truncation artifact rather than a physical trend. I ask the authors to provide a convergence check of s with respect to including 6th-order terms, or an independent construction of local integrals of motion, for at least the smaller system sizes, and to report the sensitivity of s to the flow-stopping criterion.","section":"§IVA, Eq. (22), Appendix C"},{"comment":"The extraction of gamma_c is under-specified, which weakens the quantitative claim. The text states that gamma_c is obtained 'from the crossing point of the curves belonging to different system sizes,' but it does not define which pairs of curves are crossed, how the crossing point is interpolated, or how the quoted values (gamma_c ≈ 0.7, < 0.5, and ≈ 1.3) are obtained. Without a defined gamma_c(L) and a finite-size extrapolation, the statement that the crossing 'drifts' and 'could very well lead to gamma_c -> 0' is a visual trend rather than a quantitative result. Please provide the crossing points as a function of system size, specify the extrapolation procedure (or explicitly state that no extrapolation is attempted), and include error bars reflecting the uncertainty in the support curves.","section":"§IVA, Fig. 2"},{"comment":"The abstract states that the 1D results show 'localization at infinitesimally small field even in the presence of interactions,' which is stronger than the main-text statement in Section IVA that an extrapolation 'could very well lead to gamma_c -> 0.' Given the truncation concerns above, the abstract overstates the certainty of the thermodynamic-limit conclusion. The authors should either soften the abstract to match the main-text caveat or supply the additional convergence and scaling analysis needed to justify the stronger claim.","section":"Abstract and §V"}],"minor_comments":[{"comment":"The error bound in Eq. (17) is not correct as written: |exp(iδ_E t) - 1| is bounded by |δ_E t|, not by (1/2) δ_E^2 t^2. The qualitative conclusion that TFE dynamics break down at t ~ 1/δ_E still supports the stated t ~ 1/U^2 timescale, but the displayed inequality should be corrected.","section":"§IIIB, Eq. (17)"},{"comment":"The text says the canonical Wegner generator 'provides stable convergence' and then states that 'at zero external potential it breaks down completely'; this is contradictory as phrased and should be clarified, for example by specifying that it fails at exact degeneracies rather than in general.","section":"§IIIA"},{"comment":"The caption in Fig. 2 uses the word 'transition' for what the text elsewhere describes as a crossover or finite-size transition; the terminology should be made consistent, since the paper does not claim to establish a true phase transition in the thermodynamic limit.","section":"Fig. 2 caption"},{"comment":"The text states that the energy error 'scales qualitatively with U/gamma,' but Fig. 9 reports exponents ν from power-law fits; please clarify how the plotted exponents translate into the U/gamma scaling and whether the fits are intended as evidence for a specific functional form.","section":"Appendix C, Fig. 9"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read this one for the method and the dynamics, not for the thermodynamic-limit conclusion. The authors have made a real improvement to TFE for clean systems: the full scrambling transformation handles the zero-field degeneracy that stymied earlier versions, and computing dynamics via the trace formula (Eq. 14) extends accuracy to t ~ 1/U^2, a genuine advance over the Heisenberg-equation approach in Ref. [63]. The benchmarks against ED and tDMRG in Fig. 3 are convincing, and the new tDMRG scaling t* ~ 1/U for the delocalization time (Fig. 5) is a clean piece of evidence.\n\nThe soft spot is the central 1D claim, and the stress-test concern lands. The downward drift of gamma_c with system size is inferred from the support s of the flowed number operator, Eq. (22), computed with a flow truncated at 4th order (two-particle terms only). The paper itself says neglected terms are O(U^2/gamma) (Appendix C) and shows in Section IVC that higher-order processes drive delocalization at t* ~ 1/U. Those processes are absent from the operator expansion (B2) used for s, so the diagnostic may systematically undercount delocalizing channels. The authors acknowledge this risk in Section IVA (any s>0 allows two-particle effects...), but they do not test convergence of s with respect to truncation order or provide error bars. Consequently, the abstract's 'localization at infinitesimally small field' overstates the main text's 'could very well lead to gamma_c -> 0' — which is itself a finite-size extrapolation without a controlled scaling analysis. The 2D data are new, but the authors appropriately label any thermodynamic statement as speculative.\n\nI also note the qualitative 1D conclusion was already in Ref. [82] (earlier TFE variant) and Ref. [83] (Hilbert-space shattering), so the newness is mainly the 2D data and the method improvements. That said, the paper is honest about its limitations; it does not hide the truncation issues. The lack of code/data is a practical barrier to independent verification.\n\nVerdict: send to peer review — the method development and dynamics benchmarks deserve a serious referee. But expect the referee to push for a softened abstract, a truncation-convergence check on s, and ideally code release. I would not cite the thermodynamic-limit claim as established, but I'd consider citing the method.","headline":"Good method work and honest benchmarking, but the thermodynamic-limit Stark MBL claim rests on a truncated support diagnostic and the abstract overstates the main text's cautious conclusion.","tokens_in":20944,"tokens_out":3711,"would_cite":true,"duration_ms":30861,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that in one-dimensional Stark MBL the finite-size critical field drifts to zero with system size, so the thermodynamic system is localized for any nonzero electric field and no ergodic phase survives.","keywords":["Stark many-body localization","continuous unitary transformations","flow equations","local integrals of motion","Wannier-Stark localization","infinite-temperature dynamics","tilted fermion systems","many-body localization transition"],"falsifier":"Recompute the number-operator support with a sixth-order flow (including three-particle operator strings) for 1D chains at $U=1.0$ over $L=12$ to $36$. If the crossing point $\\gamma_c$ stops drifting downward or moves upward once higher orders are included, the $\\gamma_c\\to 0$ conclusion is a truncation artifact; if the drift persists at the next order, the no-ergodic-regime claim is supported.","tokens_in":19856,"feed_emoji":"⚡","tokens_out":11156,"duration_ms":106242,"temperature":0.7,"pith_summary":"The paper asks whether many-body localization survives when random disorder is replaced by a perfectly uniform electric field, the Stark MBL setting. Using a numerical flow-equation method that approximately diagonalizes the interacting Hamiltonian, the authors watch how local the system's integrals of motion are as the tilt strength $\\gamma$ changes. Their central 1D result is that the crossover field $\\gamma_c$ moves toward smaller values as system size grows, and they conclude that the thermodynamic limit plausibly has no ergodic regime: localization would survive at any nonzero $\\gamma$. The same method reproduces dynamics up to intermediate times but misses the higher-order interaction processes behind a slow finite-size delocalization seen in tDMRG, whose onset follows $t^*\\sim 1/U$. If correct, this makes Stark MBL a disorder-free route to localization that experiments on clean tilted lattices can directly probe.","feed_headline":"An infinitesimal tilt may localize 1D interacting fermions","feed_subtitle":"Flow-equation simulations push the critical tilt toward zero as system size grows, hinting that no ergodic phase survives in 1D.","key_machinery":"Tensorflow Equations (TFE), a numerical realization of Wegner's continuous unitary flow $dH(l)/dl=[H(l),\\eta(l)]$, in which the running Hamiltonian is stored as coefficient tensors of fermionic operator strings truncated at fourth order (two-particle terms) and diagonalized with the Wegner generator combined with a Toda-Mielke 'scrambling' step that lifts single-particle degeneracies. The localization diagnostic is the support $s$ of the flowed number operator, the ratio of the four-operator weight to the total weight in Eq. (22): $s\\to 0$ signals a local integral of motion and localization, $s\\to 1$ a delocalized operator. The critical field is read from the crossing of $s(\\gamma)$ curves for different system sizes. Dynamics are obtained by propagating a chosen operator alongside the Hamiltonian and evaluating the infinite-temperature autocorrelation from diagonal energies and the matrix elements $\\langle j|n_i|k\\rangle$, with exact diagonalization and tDMRG as benchmarks.","core_discovery":"On its own terms, the paper's central claim is that the 1D Stark MBL transition is not at a finite, size-independent field. The support $s$ of the number operator in the approximately diagonal basis shows a crossing at $\\gamma_c\\approx 0.7$ for small chains accessible to exact diagonalization, but the crossing drifts to smaller fields for larger systems; the authors write that an extrapolation to infinite size 'could very well lead to $\\gamma_c\\to 0$', i.e. localization at any nonzero tilt, in line with the Hilbert-space shattering picture. In 2D the signal is less clear: a crossing near $\\gamma_c\\approx 1.3$ appears only beyond exact-diagonalization sizes and finite-size effects are strong, so the thermodynamic statement is left open. The paper also establishes a methodological claim: fourth-order TFE can simulate clean systems beyond exact diagonalization with accurate dynamics up to $t\\sim 1/U^2$, while the long-time finite-size delocalization seen in tDMRG, scaling as $t^*\\sim 1/U$, is of higher order and is missed by the truncation, suggesting it is non-perturbative.","pith_inferences":["If the $\\gamma_c\\to 0$ drift is physical rather than a truncation artifact, Stark MBL would belong to a different class from random and quasiperiodic MBL, where a transition survives at finite disorder; a direct test would be a cold-atom tilted-chain experiment that varies system size at fixed small $\\gamma$ and checks whether the critical tilt keeps dropping.","The combination of $t^*\\sim 1/U$ with the truncation blindness to long-time decay suggests the delocalization mechanism is a non-perturbative resonance effect, so a resummation or a higher-order flow retaining selected ladder terms might restore the decay within the flow-equation framework.","The same support diagnostic could be applied as a phase-diagram probe to other clean weak-to-intermediate-coupling models, such as the 2D Hubbard model with a weak tilt, reaching sizes beyond exact diagonalization."],"forward_implications":["In 1D, no disorder is needed for many-body localization in the thermodynamic limit: a nonzero uniform electric field would suffice at infinite temperature.","Finite systems can still appear ergodic at small tilt, because the measured delocalization time grows with system size; experiments on current finite samples can therefore see localized behavior even if the infinite system localizes for any $\\gamma>0$.","In 2D, the method finds a crossover near $\\gamma_c\\approx 1.3$ with strong finite-size drift, leaving the thermodynamic question open while remaining compatible with subdiffusive transport in tilted 2D lattices.","Fourth-order TFE is a practical tool for disorder-free systems: it matches exact diagonalization and tDMRG up to $t\\sim 1/U^2$ and reaches system sizes beyond exact diagonalization, including 2D systems mapped to long-range chains."],"supporting_citations":[{"why":"introduces the Tensorflow Equations used throughout; the numerical flow and dynamics routine builds on its implementation.","marker":"[63]"},{"why":"introduces real-space support of local integrals of motion in quasiperiodic MBL, the diagnostic ancestor used here.","marker":"[64]"},{"why":"applies an earlier TFE variant to 1D Stark MBL and extracts the same support quantity, giving the drift result the paper confirms.","marker":"[82]"},{"why":"argues from Hilbert-space shattering that no ergodic regime exists for nonzero tilt; the 1D extrapolation is presented as supporting this.","marker":"[83]"},{"why":"original numerical evidence for Stark MBL at finite field, the reference point for the claim that the transition moves to smaller fields with size.","marker":"[50]"},{"why":"exact-diagonalization transport results and the exponentially growing delocalization time in finite Stark MBL systems that frame the long-time dynamics section.","marker":"[66]"},{"why":"shows absence of localization in interacting spin chains with a discrete symmetry, used to interpret the long-time finite-size delocalization.","marker":"[88]"},{"why":"provides tDMRG data for tilted two-dimensional systems and subdiffusive transport context for the 2D discussion.","marker":"[67]"}],"fun_headline_variants":["1D fermions may localize at infinitesimal tilt","Flow equations hint 1D localization at any tilt","Critical tilt vanishes with size in 1D Stark MBL","Stark MBL in 1D: no ergodic phase at any field"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Everything hinges on the assumption that dropping the flow's higher-order interaction terms (keeping only up to two-particle terms) does not change the qualitative shape of the localization diagnostic, even though the paper itself shows those omitted terms drive long-time delocalization.","fun_headline_variants_meta":{"raw":{"variants":["1D fermions may localize at infinitesimal tilt","Flow equations hint 1D localization at any tilt","Critical tilt vanishes with size in 1D Stark MBL","Stark MBL in 1D: no ergodic phase at any field"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000268,"raw_usage":{"total_tokens":1652,"prompt_tokens":1013,"completion_tokens":639,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":629,"completion_tokens_details":{"reasoning_tokens":567}},"tokens_in":629,"tokens_out":639,"duration_ms":6022,"temperature":1.0,"reasoning_tokens":567,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T15:11:08.111129+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute the number-operator support with a sixth-order flow (including three-particle operator strings) for 1D chains at $U=1.0$ over $L=12$ to $36$. If the crossing point $\\gamma_c$ stops drifting downward or moves upward once higher orders are included, the $\\gamma_c\\to 0$ conclusion is a truncation artifact; if the drift persists at the next order, the no-ergodic-regime claim is supported.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"introduces the Tensorflow Equations used throughout; the numerical flow and dynamics routine builds on its implementation."},{"cited_title":"Pekker, B","cited_arxiv_id":null,"evidence_quote":"introduces real-space support of local integrals of motion in quasiperiodic MBL, the diagnostic ancestor used here."},{"cited_title":"Chandran, I","cited_arxiv_id":null,"evidence_quote":"applies an earlier TFE variant to 1D Stark MBL and extracts the same support quantity, giving the drift result the paper confirms."},{"cited_title":"Singh, B","cited_arxiv_id":null,"evidence_quote":"argues from Hilbert-space shattering that no ergodic regime exists for nonzero tilt; the 1D extrapolation is presented as supporting this."},{"cited_title":"Schulz, C","cited_arxiv_id":null,"evidence_quote":"original numerical evidence for Stark MBL at finite field, the reference point for the claim that the transition moves to smaller fields with size."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"exact-diagonalization transport results and the exponentially growing delocalization time in finite Stark MBL systems that frame the long-time dynamics section."},{"cited_title":"Zhang and H","cited_arxiv_id":null,"evidence_quote":"shows absence of localization in interacting spin chains with a discrete symmetry, used to interpret the long-time finite-size delocalization."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"provides tDMRG data for tilted two-dimensional systems and subdiffusive transport context for the 2D discussion."}],"review_version":1}