{"id":"3d0ac894-ae07-4957-9fbf-524a80a31325","arxiv_id":"2608.09245","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Optimized patterned superlattices on bilayer graphene host Moore-Read type non-Abelian fractional Chern insulator states at half filling, according to exact diagonalization.","lead":"This paper uses a computer search to design tiny etched holes in a substrate under bilayer graphene, creating electronic bands with the right geometry for a special quantum state. When the band is half-filled, calculations show signatures of the Moore-Read state, an exotic phase that could host non-Abelian anyons useful for quantum computing.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Moore-Read assignment rests on exact diagonalization at only two small tilted tori; a third, larger cluster is the minimal check that would settle the central existence claim.","rationale":"The reader identified the loss function in Eq. (2) as the weakest assumption. That concern is real but partly self-acknowledged: the paper explicitly states that the loss is a strong descriptor rather than a sufficient criterion, and the phase diagrams show a statistical rather than perfect correlation. The more load-bearing assumption for the paper's headline claim is the finite-size identification: if the two clusters' degeneracies and PES counts were accidental, the existence claim would fail even if the loss were a perfect descriptor. Conversely, a third cluster that preserves the signatures would largely validate the existence claim. I therefore keep the CONDITIONAL verdict, but the condition is stronger: add an additional cluster size and make the ED data available. This is not a rejection; the even/odd degeneracy pattern and exact PES counts are already non-trivial, and the proposed check is feasible with existing single-band ED methods.","tokens_in":21611,"tokens_out":10923,"duration_ms":131821,"concrete_test":"At the three optimized structures of Fig. 3, recompute the many-body spectrum and PES on a third tilted torus with N_s=32, N_e=16 using the same projected interaction [Eq. (3) and SM Eq. (S12)]. For a Moore-Read state on an even-electron torus the ground-state manifold must remain sixfold with the same momentum-sector distribution as predicted by the (2,4)-admissible generalized Pauli principle, and the N_A=4 (or N_A=5) PES low-lying level count must match the corresponding quasihole counting. A confirming result would settle the small-cluster concern; a failure would show the claimed MR assignment is an artifact of the 26/28-site tilted meshes. Releasing the ED code/data alongside would make this check independently reproducible.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim — that the gradient-optimized highest valence band hosts a Moore-Read-type non-Abelian FCI at 1/2 filling — is carried by the finite-size ED evidence of Sec. \"Non-Abelian FCI states\": a sixfold quasi-degenerate manifold on the N_s=28, N_e=14 torus and a twofold manifold on the N_s=26, N_e=13 torus, reinforced by PES low-lying counts of 18,571 and 2,522 levels below the entanglement gap. These two tilted-mesh clusters are the entire many-body basis for the existence claim. Both the ground-state degeneracy and the PES counting are finite-size signatures; without a third system size they do not by themselves exclude accidental small-system coincidences, and code/data are not released, so independent verification is not currently possible. The loss-function non-sufficiency acknowledged in the Discussion weakens the generality of the optimization workflow but does not endanger the positive ED results; the two-cluster identification is therefore the more load-bearing assumption.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes a gradient-based inverse-design workflow for patterned dielectric superlattices on bilayer graphene, optimizing the etched-hole radius and depth so that the highest valence band acquires a quantum geometry close to that of the first excited Landau level. At half filling of this band, exact diagonalization on tilted tori reports a sixfold quasi-degenerate ground-state manifold for N_s=28, N_e=14 and a twofold manifold for N_s=26, N_e=13, with particle entanglement spectra showing 18,571 and 2,522 low-lying levels below the entanglement gap, respectively. These signatures are interpreted as Moore-Read-type non-Abelian fractional Chern insulator states. The workflow is applied to triangular, honeycomb, and kagome superlattices, and many-body phase diagrams in the (L_s, δV) plane are presented. The paper explicitly acknowledges that the loss function is a strong descriptor rather than a sufficient criterion for Moore-Read order.","tokens_in":21853,"tokens_out":5232,"duration_ms":53745,"significance":"If the finite-size identification holds, this is a valuable contribution: it provides a realistic, experimentally accessible platform for non-Abelian FCIs without twist-angle fine-tuning, and it demonstrates a device-level inverse-design loop that couples FEM electrostatics, a realistic bilayer-graphene continuum model, and exact diagonalization. The strengths include the explicit optimization of experimentally relevant parameters, the positive comparison between optimized and unoptimized structures, and the even-odd electron-number dependence of the ground-state degeneracy, which is a sharp Moore-Read signature. The principal weakness is that the central many-body claim rests on only two small tilted-torus clusters, while the extended phase diagrams rely on single-cluster-size diagnostics at most parameter points, so the breadth of the conclusions currently exceeds the finite-size evidence.","major_comments":[{"comment":"The central existence claim rests on ED at only two tilted tori: N_s=28, N_e=14 and N_s=26, N_e=13. The sixfold/twofold ground-state manifolds and the PES counts (18,571 and 2,522) are finite-size signatures; without a third, larger cluster (e.g., N_s=32 or N_s=38) at the three representative points, accidental small-system coincidences cannot be excluded. Please add at least one additional system size and report the evolution of the many-body gap and PES counting across system sizes.","section":"Non-Abelian FCI states (Fig. 3 and Fig. S4)"},{"comment":"The phase diagrams in Fig. 4 label every (L_s, δV) point as Moore-Read or non-Moore-Read, but the caption does not state which cluster size (N_s=26 or 28) was used at each point, and the diagnostics are finite-size by construction. The claim of a robust Moore-Read region over an extended parameter area is therefore not supported for points checked at only one cluster size. Please specify the N_s used per point and verify representative boundary and non-MR points at a second cluster size.","section":"Many-body phase diagrams (Fig. 4)"},{"comment":"The paper correctly states that the loss function L is 'a strong descriptor rather than a sufficient criterion.' Given that the optimization workflow's promise is to design non-Abelian states, the average-loss separation (1.2 versus 4.3 for triangular, and similar for the others) is only suggestive. The exceptions visible in Fig. 4 show that small L does not always yield Moore-Read order and larger L does not always preclude it. To make the predictive claim load-bearing, provide a quantitative assessment (e.g., a threshold with false-positive and false-negative rates, or an out-of-sample test), or explicitly limit the claim to the ED-verified structures rather than to a general optimization guarantee.","section":"Discussion and Eq. (2)"}],"minor_comments":[{"comment":"Please state the cluster size N_s used for each parameter point, and clarify that white regions were not calculated because the HVB Chern number is not ±1.","section":"Fig. 4 caption"},{"comment":"The counts 18,571 and 2,522 are quoted without a derivation or citation for the expected Moore-Read quasihole counting on these tori; adding the counting formula would strengthen the identification.","section":"PES counting paragraph"},{"comment":"The dimensionless standard deviation σ(Ω) of the Berry curvature should be explicitly normalized (e.g., by the mean |Ω|), so that the meaning of α=0.5 and the reported value σ(Ω)=0.8 are unambiguous.","section":"Eq. (2)"},{"comment":"The inverse screening length κ=(120 nm)^{-1} and the single-valley-single-spin projection are significant modeling choices; please add one sentence on how varying κ or including the other valley/spin flavor would affect the ED results.","section":"Supplemental Material S3.1"},{"comment":"A data-availability statement would be helpful: the FEM and ED workflows are complex, and no code or data are currently released, which makes independent verification difficult.","section":"Data availability"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within the journal's scope, and the representative-point ED results are credible and clearly presented. The main risk is that the broad phase-diagram and workflow claims outrun the finite-size evidence; adding a third cluster size and clarifying the Fig. 4 methodology would resolve this. No concerns about novelty or citation practices."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a new and sensible route to non-Abelian FCIs—gradient-descent optimization of real structural parameters of patterned dielectric superlattices on bilayer graphene, targeting first-LL quantum geometry, with ED at half filling giving Moore-Read signatures. The central existence claim is plausible but the evidence is thinner than the abstract implies: two small tilted tori carry the whole many-body case.\n\nThe genuinely new piece is the device-level inverse design: previous gradient searches lived in idealized tight-binding models, and non-Abelian FCI proposals lived in moiré. Here the optimized degrees of freedom are hole radius and depth in an etched substrate, and the FEM-plus-continuum-model pipeline is realistic and transferable. The paper also does honest work internally: the unoptimized-vs-optimized comparison (Fig. S2) shows the optimization actually flips the ground-state degeneracy to the Moore-Read counting, the PES numbers (18,571 and 2,522 levels below the gap) match the quasihole counts, and the authors explicitly concede the loss is 'a strong descriptor rather than a sufficient criterion.' The loss-to-order correlation in Fig. 4 gives that admission some positive support. No load-bearing circularity: the loss is single-particle geometry, the ED is independent.\n\nSoft spots, in order. First, the two-cluster identification is the load-bearing assumption, and I agree with the stress-test here. Ground-state degeneracy plus PES counting on N_s=28 and N_s=26 is the right diagnostic, but two small clusters can match Moore-Read by accident, and most of the 'large region' phase diagram is judged on a single cluster size. A third, larger cluster is the minimal fix and should be the main referee request. Second, no code or data is released, so none of this is checkable today. Third, the single-band/single-valley projection and hand-picked κ and α are standard but unexamined; minor. The loss non-sufficiency is real but the paper already says it, so it's a limitation of the workflow's generality, not a hole in the positive ED results.\n\nFor whom: FCI/moiré/superlattice people, and anyone building numerical design pipelines for topological phases. This deserves a serious referee; I'd send it out and ask for the third cluster, the data, and a per-point statement of which cluster sizes back each phase-diagram entry.","headline":"A solid new inverse-design workflow for Moore-Read FCIs in patterned superlattices, carried by ED on just two small clusters; send it to referees and use the reports to demand a third cluster, released data, and per-point cluster-size details.","tokens_in":22348,"tokens_out":5453,"would_cite":true,"duration_ms":52340,"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":"Gradient-optimized patterned superlattices in bilayer graphene create Moore-Read non-Abelian fractional Chern insulator states at half filling, as shown by exact diagonalization.","keywords":["non-Abelian fractional Chern insulator","Moore-Read state","patterned dielectric superlattice","bilayer graphene","gradient-based inverse design","quantum geometry","particle entanglement spectrum","topological flat band"],"falsifier":"Run exact diagonalization at half filling on the same 28-site and 26-site torus geometries for a parameter point with low loss but a substantially larger bandwidth, or with a different screening length; if the sixfold/twofold ground-state degeneracy and the 18,571/2,522 low-lying particle-entanglement levels no longer match Moore-Read counting, the loss function alone is not a sufficient criterion.","tokens_in":21429,"feed_emoji":"⚛️","tokens_out":9293,"duration_ms":87469,"temperature":0.7,"pith_summary":"The paper aims to show that non-Abelian fractional topological order can be engineered from scratch by optimizing the physical dimensions of a patterned dielectric superlattice. The authors define a loss function that measures how closely a chosen flat band's quantum geometry resembles the first excited Landau level, and they minimize it by gradient descent over the etched-hole radius and depth in bilayer-graphene devices. At $\\nu=1/2$ filling of the optimized band, exact diagonalization produces the signatures of the Moore-Read state: sixfold quasi-degenerate ground-state manifolds on 28-site tori, twofold manifolds on 26-site tori, and particle entanglement spectra with low-lying level counts matching Moore-Read quasihole counting. These signatures appear for triangular, honeycomb, and kagome superlattice patterns over large regions of the parameter space, so the workflow is a concrete experimental route to non-Abelian anyons without a magnetic field.","feed_headline":"Gradient-designed superlattices host Moore-Read states at zero field","feed_subtitle":"Exact diagonalization finds Moore-Read degeneracy and entanglement counting in all three lattice patterns.","key_machinery":"The engine of the design is the loss function $L = |(1/2\\pi)\\int_{\\mathrm{BZ}} d^2 k \\, \\mathrm{tr}\\,g(k) - 3| + \\alpha \\sigma(\\Omega)$ with $\\alpha=0.5$, where $g$ is the quantum metric and $\\sigma(\\Omega)$ the dimensionless standard deviation of the Berry curvature of the highest valence band. It encodes the target that a flat Chern band have the quantum geometry of the first excited Landau level: uniform Berry curvature, Chern number $\\pm 1$, and $\\int \\mathrm{tr}\\,g = 3$. The gradient of $L$ with respect to the hole radius and depth is evaluated by finite differences, each evaluation requiring a finite-element electrostatic simulation of the etched substrate followed by a continuum $k\\cdot p$ band-structure calculation of Bernal bilayer graphene. The many-body identification of the Moore-Read phase uses the $(2,4)$-admissible generalized Pauli principle to predict torus ground-state degeneracies and the particle entanglement spectrum to compare low-lying level counts with quasihole counting.","core_discovery":"At half filling, the highest valence band of bilayer graphene coupled to a structurally optimized patterned dielectric superlattice realizes a Moore-Read-type fractional Chern insulator. Exact diagonalization gives a sixfold quasi-degenerate ground-state manifold on the 28-site torus, split between three momentum sectors with two states each, and the expected twofold manifold on the 26-site torus with an unpaired Majorana sector; particle entanglement spectra show 18,571 low-lying levels below the entanglement gap for the $N_A=4$ cut and 2,522 for the $N_A=3$ cut, both matching Moore-Read quasihole counting. The paper shows that these states appear only after optimizing the hole radius and depth, and that across the $(L_s,\\delta V)$ maps, low loss values correlate strongly with Moore-Read order, while unoptimized or high-loss structures do not form such states.","pith_inferences":["An untested corollary of the paper's own caveat—that the loss is a strong descriptor rather than a sufficient criterion—is that scanning low-loss points with systematically varied bandwidth would map how far the first-LL geometry condition can be relaxed before Moore-Read order is lost.","The sign rule found between the triangular pattern (positive $\\delta V$) and honeycomb/kagome patterns (negative $\\delta V$), traced to the structure factor, gives a simple design rule that could be applied to other superlattice patterns without full optimization.","The same optimization loop could be run with a different Landau-level or model-wavefunction target, generating device designs for Laughlin-type or other topological orders and testing whether the inverse-design pipeline generalizes.","A transport-level signature, such as thermal Hall conductance quantization or interference measurements, is not computed in the paper but would be the natural experimental check of the non-Abelian order."],"forward_implications":["Patterned dielectric superlattices become a realistic zero-field platform for non-Abelian fractional Chern insulator states, with the relevant device parameters (hole radius and depth) set by nanofabrication rather than by twist-angle alignment.","The gradient-based inverse-design loop is not limited to bilayer graphene; the paper argues it extends to other van der Waals multilayers, so the same workflow could target other fractional phases.","The reported many-body gaps of order 0.3–0.4 meV with ground-state spreads as small as 0.002 meV give concrete energy scales for future experimental probes.","Fabrication must follow the optimized structural parameters: exact diagonalization of the unoptimized initial structures does not show Moore-Read degeneracy or entanglement signatures.","The strong correlation between small loss values and the appearance of Moore-Read order suggests the single-particle loss can be used as a fast screening tool over candidate devices before expensive many-body calculations."],"supporting_citations":[{"why":"Defines the Moore-Read (Pfaffian) non-Abelian state at half filling of the first Landau level, the many-body phase whose signatures the paper looks for.","marker":"[33]"},{"why":"Supplies the ideal first-LL quantum-geometric target—uniform Berry curvature, Chern number ±1, and ∫trg = 3—that the loss function is built from.","marker":"[68]"},{"why":"Generalizes Landau-level geometry and connects it to non-Abelian physics, supporting the choice of first-LL-like metric as the optimization target.","marker":"[69]"},{"why":"Provides the generalized Pauli principle and momentum-sector counting used to identify the Moore-Read ground-state manifold on the torus.","marker":"[26]"},{"why":"Gives the sixfold/twofold torus degeneracy counting for Moore-Read states that the exact-diagonalization spectra are compared with.","marker":"[71]"},{"why":"Introduces the entanglement-spectrum probe of topological order on which the particle entanglement spectrum analysis is based.","marker":"[72]"},{"why":"Supplies the quasihole-counting numbers for lattice particle entanglement spectra that the low-lying PES levels are matched against.","marker":"[74]"},{"why":"Demonstrates the patterned-dielectric-superlattice platform the devices are built on.","marker":"[50]"},{"why":"Shows fractional topological states in a graphene-based kagome superlattice, the closest platform precedent for the proposed experiment.","marker":"[63]"},{"why":"Establishes gradient-based search of quantum phases, the inverse-design methodology this workflow adapts to device parameters.","marker":"[75]"}],"fun_headline_variants":["Gradient-optimized superlattices host Moore-Read states at zero field","Zero-field Moore-Read states from gradient-designed superlattices","Inverse-designed superlattices yield non-Abelian Moore-Read states","Gradient descent optimizes bilayers for Moore-Read fractional Chern states","Patterned superlattices tuned by gradient search host Moore-Read order"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that a single-particle band whose quantum geometry mimics the first Landau level is enough to produce the Moore-Read many-body state at half filling; the paper itself notes that the loss is a strong descriptor rather than a sufficient criterion, because band dispersion and other ingredients also matter.","fun_headline_variants_meta":{"raw":{"variants":["Gradient-optimized superlattices host Moore-Read states at zero field","Zero-field Moore-Read states from gradient-designed superlattices","Inverse-designed superlattices yield non-Abelian Moore-Read states","Gradient descent optimizes bilayers for Moore-Read fractional Chern states","Patterned superlattices tuned by gradient search host Moore-Read order"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000651,"raw_usage":{"total_tokens":3031,"prompt_tokens":1036,"completion_tokens":1995,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":652,"completion_tokens_details":{"reasoning_tokens":1898}},"tokens_in":652,"tokens_out":1995,"duration_ms":15799,"temperature":1.0,"reasoning_tokens":1898,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T20:49:56.012214+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run exact diagonalization at half filling on the same 28-site and 26-site torus geometries for a parameter point with low loss but a substantially larger bandwidth, or with a different screening length; if the sixfold/twofold ground-state degeneracy and the 18,571/2,522 low-lying particle-entanglement levels no longer match Moore-Read counting, the loss function alone is not a sufficient criterion.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the sixfold/twofold torus degeneracy counting for Moore-Read states that the exact-diagonalization spectra are compared with."},{"cited_title":"Li and F","cited_arxiv_id":null,"evidence_quote":"Introduces the entanglement-spectrum probe of topological order on which the particle entanglement spectrum analysis is based."},{"cited_title":"Sterdyniak, N","cited_arxiv_id":null,"evidence_quote":"Supplies the quasihole-counting numbers for lattice particle entanglement spectra that the low-lying PES levels are matched against."},{"cited_title":"Forsythe, X","cited_arxiv_id":null,"evidence_quote":"Demonstrates the patterned-dielectric-superlattice platform the devices are built on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Shows fractional topological states in a graphene-based kagome superlattice, the closest platform precedent for the proposed experiment."}],"review_version":1}