{"id":"a4536fdb-d437-4da0-91e3-13434da9e9ca","arxiv_id":"2411.10247","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Anderson localization in 3D can be driven purely by structural disorder, with the same universality class as standard Anderson localization.","lead":"Structural disorder, in the form of missing lattice sites or randomly added links, can make a clean 3D crystal turn from a conductor into an insulator. The transition has the same critical properties as ordinary Anderson localization, showing that irregular geometry alone is enough to trap quantum waves.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Universality-class claim rests on a single-parameter finite-size scaling collapse of r with no demonstrated control of corrections to scaling; extracted ν≈1.57/1.61 may be effective rather than asymptotic.","rationale":"The paper does something real: two geometrically distinct constructions both show a sharp level-statistics transition upon tuning a structural parameter, and RGF and participation-entropy data are consistent. This is strong numerical evidence for the existence of the phenomenon. My concern is not that the transition is absent, but that the second part of the headline claim—'same universality class as standard Anderson localization'—is supported only by one effective exponent from a scaling collapse that does not include corrections. Since the disorder here is long-range correlated by construction, there is a known danger that finite-size effective exponents mimic the orthogonal-class value. The reader already identified this, and my pass agrees. The proposed check is inexpensive and decisive: include an irrelevant-scaling term and test stability of ν under L-window cuts. If ν remains ~1.57 with overlapping error bars and the correction amplitude is consistent with zero, the universality-class claim is much stronger. If ν drifts, the paper remains a valid observation of a localization transition, but its stronger claim should be downgraded. Accordingly the reader's CONDITIONAL verdict stands unchanged.","tokens_in":13594,"tokens_out":7083,"duration_ms":79735,"concrete_test":"Re-analyze the gap-ratio data from both models with a finite-size scaling ansatz that includes a leading irrelevant correction, e.g. r = f[(p-p_c)L^{1/ν}] + a(p)L^{-ω} with ω>0, using the same bootstrap/cost-function procedure; then repeat the fit on restricted L-subsets (L≥40 and L≥60). If ν moves outside the 1.5–1.65 window as the L cutoff increases, or if the corrected fit substantially improves the collapse quality, the quoted exponents should be treated as effective and the universality-class conclusion is not yet established. A supplementary check: at p_c, compute the multifractal dimension D2 from participation entropies and compare quantitatively with the known 3D-Anderson multifractal spectrum; agreement in ν alone is insufficient to claim universality.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that the transition is in the standard 3D Anderson universality class hinges on the critical exponent ν matching the known value ν≈1.57. In the main text, ν is quoted with no error bars: the honeycomb value (ν≈1.57) and link value (ν≈1.61) are obtained from a one-parameter scaling collapse r=f[(p-p_c)L^{1/ν}] (Supplement Eq. 4), with no corrections-to-scaling term. The L-range used is only 30–80 for honeycomb and 32–80 for the link model, and the system is strongly correlated because the analysis is restricted to the largest connected cluster; these conditions make it plausible that leading irrelevant scaling fields are significant. If corrections to scaling are present, the fitted ν can be an effective exponent that drifts toward the true value only at larger L, and a value near 1.6 at L≤80 does not by itself identify the universality class. The supplementary cost-function contours provide error estimates via W<1.3W*, but the main-text claims are not tied to those intervals, and no correction term is tested. The existence of a transition is supported by independent checks (gap-ratio crossing, RGF for the link model, participation-entropy D2); the weak point is specifically the universality-class identification.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper claims that Anderson localization in three-dimensional tight-binding systems can be induced solely by structural disorder, i.e., by irregularities of the lattice geometry, without any on-site disordered potential. Two models are analyzed: a honeycomb lattice in which sites are removed according to a classical statistical model at fixed temperature T=1 and only the largest connected cluster is kept, and a link model built from a random spanning tree of the cubic lattice with additional links added with probability P_link. Using exact diagonalization of systems up to about 10^6 sites, the authors compute the averaged gap ratio and observe a localization-delocalization transition as the structural disorder strength is tuned (p_c ≈ 0.36 and P_link,c ≈ 0.024). Finite-size scaling of the gap ratio gives critical exponents ν ≈ 1.57 and ν ≈ 1.61, which are quoted as consistent with the standard three-dimensional Anderson universality class. The supplementary material adds fractal-dimension analysis for the link model (ν ≈ 1.53) and a recursive Green's function study of the localization length.","tokens_in":13854,"tokens_out":5760,"duration_ms":56059,"significance":"If the universality-class claim holds, the paper identifies a genuinely new family of structurally disordered lattice models in which destructive interference alone produces a metal-insulator transition in three dimensions, without onsite disorder. The numerical evidence is genuinely multi-pronged: gap-ratio scaling in two independent models, a crossing point that is stable with system size, a separate fractal-dimension analysis, and a qualitative recursive Green's function check. The analysis is not circular: the critical parameters and exponents are extracted from standard finite-size scaling and then compared with literature values, not fitted to them. The main weakness is that the universality-class identification rests on a one-parameter scaling collapse over a limited range of system sizes, with no demonstrated control of corrections to scaling and no error bars quoted in the main text. This makes the central claim plausible but not yet fully established; the requested revisions are therefore focused on error reporting and scaling-robustness checks.","major_comments":[{"comment":"The universality-class claim rests on the values ν ≈ 1.57 (honeycomb) and ν ≈ 1.61 (link), yet the main text quotes these without error bars and the collapse is a one-parameter scaling form r = f[(p − p_c)L^{1/ν}] with no corrections-to-scaling term. The accessible L range (30–80 for both models) is limited, and the largest-cluster restriction produces long-range correlated disorder, so leading irrelevant scaling fields could plausibly be significant. Please report the error intervals (the Supplement's W < 1.3W* contours are not quoted in the main text), and test stability by adding a correction-to-scaling term or by dropping the smallest L from the fit; if the exponents drift, the claim of belonging to the standard 3D Anderson universality class should be softened.","section":"Fig. 2 and Supplement 'Estimation of the critical exponents'"},{"comment":"The text states that for P_link < P_link,c the ratio λ_M/L_M vanishes with increasing L_M, implying localized eigenstates, while for P_link > P_link,c the localization length increases with L_M, implying delocalization. The caption of Fig. 4 states the opposite for P_link > P_link,c: 'the localization length remains finite and approaches ξ as L_M → ∞ which is characteristic of a localized phase'. Please reconcile this inconsistency; as written, the recursive Green's function confirmation supports both readings and needs to be unambiguous.","section":"The link model (main text) and Fig. 4 caption"},{"comment":"Eq. (8a) reads D_q = (S_q(L+ΔL) − S_q(L)) / (3 log(L/(L+ΔL))). Since the denominator is negative for ΔL > 0 while a delocalized state has S_q increasing with L, this formula gives negative D_q for delocalized states, contrary to the stated D_q = 1. If the implemented denominator is instead 3 log((L+ΔL)/L), please correct the equation; if not, the D_2 values in Fig. 10 and the derived ν ≈ 1.53 need to be re-examined.","section":"Supplement, 'Analysis of fractal dimension', Eq. (8a)"}],"minor_comments":[{"comment":"In the honeycomb-model section, the sentence 'we plot a phase diagram in the p vs T plane in Fig. 11' should refer to Fig. 3, which is the phase diagram in the main text; Fig. 11 in the Supplement appears to be a duplicate of the same phase diagram.","section":"Honeycomb model (main text)"},{"comment":"The phrase 'two models with distinct types of lattice regularities' should read 'lattice irregularities', matching the terminology used throughout the rest of the paper.","section":"Abstract"},{"comment":"The sentence 'we find ⟨r⟩ → r_PS and ⟨r⟩ → r_GOE, respectively' should specify which side of p_c corresponds to which limit, since the two limits are associated with different phases.","section":"The honeycomb model, gap-ratio paragraph"},{"comment":"The supplement cites 'these lecture notes' without a full reference; please provide the complete citation for the numerical code example.","section":"Supplement, 'Recursive Green's function method'"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is well within the scope of the journal and the central phenomenon is interesting and likely correct. The main issue is that the strongest claim — belonging to the standard 3D Anderson universality class — is stated more confidently than the current scaling analysis supports. I would ask the authors to include error bars, test corrections to scaling, and fix the RGF caption inconsistency and the sign typo in Eq. (8a). These are all addressable within the manuscript's scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing to know: this paper delivers what it claims. It shows, in two distinct 3D tight-binding models, that removing sites (or links) in a correlated way can localize all eigenstates even with zero onsite disorder. That is a genuinely new result, and the evidence is solid: gap-ratio scaling, fractal dimension, and a recursive Green's function check all agree on the existence of a transition, and the extracted critical exponents (nu about 1.57 for honeycomb, 1.61 for the link model) land close to the standard 3D Anderson value. The honeycomb model in particular is well motivated, and the choice to work on the largest cluster, which makes the disorder perfectly correlated in the bulk, is a clever twist that distinguishes this from earlier off-diagonal disorder work.\n\nSoft spot is exactly where the stress-test put it: the universality-class statement rests on a one-parameter scaling collapse over L=30-80, with exponents quoted in the main text without error bars. The supplementary does provide cost-function contours with the usual W<1.3W* criterion, but the main-text claims do not reference those intervals, and no corrections-to-scaling term is tested. At L around 80, a fitted nu near 1.6 could still be an effective exponent drifting toward the true value. That said, the agreement with the accepted 3D orthogonal exponent is persuasive, and the independent fractal-dimension analysis in the same paper gives nu about 1.53 for the link model, which is consistent. So this is a request for tightened reporting, not a sign that the central result is wrong.\n\nThe citation pattern looks honest; the paper builds on prior off-diagonal and fractal-lattice work and explicitly cites it. The connection to carbonized charcoal is speculative but clearly labeled as motivation, not a quantitative model.\n\nWho should read it: anyone working on Anderson localization, quantum transport in disordered geometries, or tight-binding models on random graphs. It deserves a serious referee. I would send it out with a request for error bars on p_c and nu, a plot or test of stability under corrections to scaling, and ideally a release of code and data. That is standard for a numerical universality-class claim and easily within reach.","headline":"Two clean 3D models show that lattice geometry alone can drive an Anderson transition; the universality-class claim is credible but needs explicit error bars and a corrections-to-scaling check before it is airtight.","tokens_in":14425,"tokens_out":2568,"would_cite":true,"duration_ms":25012,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Structural disorder alone can drive the Anderson localization transition in three dimensions.","keywords":["Anderson localization","structural disorder","gap ratio","finite-size scaling","universality class","random spanning tree","level statistics","mobility edge"],"falsifier":"Repeat the gap-ratio scaling analysis at system sizes $L\\gtrsim 120$ with comparable statistical sampling and test whether all data collapse onto a single curve $r=f[(p-p_c)L^{1/\\nu}]$ with no correction term. If the best-fit $p_c$ drifts beyond its error bar, or the same-quality collapse requires a second scaling variable, the claimed universality class is not established.","tokens_in":13381,"feed_emoji":"🕸️","tokens_out":7393,"duration_ms":70092,"temperature":0.7,"pith_summary":"Most discussions of Anderson localization assume the randomness lives in an on-site potential, so a particle scatters off random energies as it hops. This paper argues the disorder can instead be purely structural: the lattice itself is missing sites or connections, yet the same physics emerges. In two different three-dimensional tight-binding models, smoothly increasing the structural disorder drives level statistics from the delocalized random-matrix value to the localized Poisson value, with a sharp transition at a critical strength. Finite-size scaling gives exponents $\\nu \\approx 1.57$ and $\\nu \\approx 1.61$, consistent with the standard 3D Anderson transition, so the authors conclude that geometric irregularities alone belong to the same universality class as random-potential disorder.","feed_headline":"Structural disorder alone drives Anderson localization in 3D","feed_subtitle":"Removing sites or links from 3D lattices tunes a metal–insulator transition with standard critical exponents.","key_machinery":"The machinery is a pairing of disordered geometric substrates with the spectral gap-ratio probe. The honeycomb substrate is generated by a geometric cluster Monte Carlo algorithm sampling the equilibrium of an Ising-like attraction between occupied sites at fixed density $p$; only the largest cluster of occupied sites is kept. The link substrate is a random spanning tree of the cubic lattice, generated by Wilson's algorithm, with extra nearest-neighbour links added independently with probability $P_{\\mathrm{link}}$. On each substrate the observable is the average gap ratio $r$ of consecutive eigenvalues, whose value distinguishes GOE (delocalized) from Poisson (localized) statistics; finite-size scaling of $r$ as $f[(p-p_c)L^{1/\\nu}]$ yields the critical point and exponent. The recursive Green's function method supplies an independent localization-length probe for the link model.","core_discovery":"The central claim is that Anderson localization can be induced solely by structural disorder, i.e., by irregularities of the lattice geometry, without any on-site disordered potential. The paper establishes this by analyzing level statistics in two models. In the honeycomb model, sites are removed according to a geometric-cluster Monte Carlo equilibrium at fixed occupation $p$, and the tight-binding Hamiltonian is defined on the largest cluster; the average gap ratio crosses from the Poisson value $r\\approx 0.386$ to the GOE value $r\\approx 0.531$ at $p_c\\approx 0.36$ with $\\nu\\approx 1.57$. In the link model, a random spanning tree of the cubic lattice is supplemented by extra links with probability $P_{\\mathrm{link}}$; the transition occurs at $P_{\\mathrm{link},c}\\approx 0.024$ with $\\nu\\approx 1.61$, and is corroborated by recursive Green's function localization-length scaling and by fractal-dimension analysis. Because the extracted exponents agree with the 3D Anderson model, the paper concludes that the transition belongs to the same universality class, despite the long-range correlations introduced by restricting the analysis to the largest cluster.","pith_inferences":["If the universality-class claim survives, it implies that the long-range correlations induced by the largest-cluster restriction do not change the critical behavior, which is not what generic correlated-disorder arguments would predict; a renormalization-group check of this point would be valuable.","The honeycomb construction is essentially a correlated quantum-percolation model, so comparing its critical $p_c$ with uncorrelated quantum percolation thresholds could isolate the role of geometric correlations.","A testable extension would be to add weak interactions to these structurally disordered lattices and look for signatures of many-body localization without on-site disorder.","Applying the same probes to 3D hyperbolic or curved lattices could show whether the 'geometric disorder alone' mechanism persists when the background geometry itself carries curvature."],"forward_implications":["Irregular connectivity alone is enough to localize single particles in three dimensions; random potential energy is not required.","The two models, honeycomb site removal and tree-plus-links, become new members of the 3D Anderson universality class, with critical exponents $\\nu\\approx 1.57$ and $\\nu\\approx 1.61$.","A mobility edge exists between localized and delocalized states in the energy spectrum, so structural disorder produces energy-dependent localization.","The link model shows that a very small density of added loops, $P_{\\mathrm{link}}\\approx 2.4\\%$, is enough to delocalize a random spanning tree, quantifying how connectivity controls transport.","Materials whose structural disorder comes from irregular platelet or graphene networks could exhibit Anderson localization even if they are chemically clean.","The same gap-ratio and localization-length machinery can be applied to other irregular 3D lattices to test whether the mechanism is generic."],"supporting_citations":[{"why":"Defines the Anderson localization phenomenon this paper extends to structural disorder.","marker":"[10]"},{"why":"Supplies the geometric cluster Monte Carlo algorithm that generates the occupied-site configurations of the honeycomb model.","marker":"[31]"},{"why":"Shows in 2D that an order-disorder transition of the underlying classical model coincides with a localization transition, the pattern extended to 3D here.","marker":"[33]"},{"why":"Introduces the gap ratio used as the level-statistics observable.","marker":"[36]"},{"why":"Provides the GOE and Poisson average gap-ratio values used to identify delocalized and localized phases.","marker":"[37]"},{"why":"Gives the high-precision critical exponent of the 3D Anderson transition used as the universality-class benchmark.","marker":"[39]"},{"why":"Shows the Anderson transition driven by a quasiperiodic potential belongs to the same universality class, a second benchmark for the exponent.","marker":"[40]"},{"why":"Supplies Wilson's algorithm that generates the random spanning tree substrate for the link model.","marker":"[41]"},{"why":"Introduces the recursive Green's function method used to compute localization lengths in the link model.","marker":"[42]"},{"why":"Adds the recursive Green's function finite-size scaling algorithm that corroborates the link-model transition.","marker":"[43]"}],"fun_headline_variants":["Geometry does it: structural disorder triggers Anderson localization","No disorder potential needed: lattice defects alone cause localization","Structural disorder alone triggers Anderson localization","Lattice defects, not disorder potentials, induce Anderson localization","Geometric disorder triggers Anderson transition in 3D"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the one-parameter finite-size scaling ansatz holds with negligible corrections over the studied system sizes, so the extracted exponents are the true asymptotic ones despite the long-range-correlated structural disorder.","fun_headline_variants_meta":{"raw":{"variants":["Geometry does it: structural disorder triggers Anderson localization","No disorder potential needed: lattice defects alone cause localization","Structural disorder alone triggers Anderson localization","Lattice defects, not disorder potentials, induce Anderson localization","Geometric disorder triggers Anderson transition in 3D"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000888,"raw_usage":{"total_tokens":3797,"prompt_tokens":877,"completion_tokens":2920,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":493,"completion_tokens_details":{"reasoning_tokens":2848}},"tokens_in":493,"tokens_out":2920,"duration_ms":20293,"temperature":1.0,"reasoning_tokens":2848,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T19:49:05.605017+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Repeat the gap-ratio scaling analysis at system sizes $L\\gtrsim 120$ with comparable statistical sampling and test whether all data collapse onto a single curve $r=f[(p-p_c)L^{1/\\nu}]$ with no correction term. If the best-fit $p_c$ drifts beyond its error bar, or the same-quality collapse requires a second scaling variable, the claimed universality class is not established.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the geometric cluster Monte Carlo algorithm that generates the occupied-site configurations of the honeycomb model."},{"cited_title":"De Tomasi, O","cited_arxiv_id":null,"evidence_quote":"Shows in 2D that an order-disorder transition of the underlying classical model coincides with a localization transition, the pattern extended to 3D here."},{"cited_title":"Slevin and T","cited_arxiv_id":null,"evidence_quote":"Gives the high-precision critical exponent of the 3D Anderson transition used as the universality-class benchmark."},{"cited_title":"Luo and T","cited_arxiv_id":null,"evidence_quote":"Shows the Anderson transition driven by a quasiperiodic potential belongs to the same universality class, a second benchmark for the exponent."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies Wilson's algorithm that generates the random spanning tree substrate for the link model."},{"cited_title":"MacKinnon, The calculation of transport properties and density of states of disordered solids, Z","cited_arxiv_id":null,"evidence_quote":"Introduces the recursive Green's function method used to compute localization lengths in the link model."},{"cited_title":"Wurtz and B","cited_arxiv_id":null,"evidence_quote":"Adds the recursive Green's function finite-size scaling algorithm that corroborates the link-model transition."}],"review_version":1}