{"id":"a252bcab-6eee-41ef-a4c5-86e2d8034f49","arxiv_id":"1908.02753","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Twist-angle disorder in twisted bilayer graphene fills in miniband gaps and broadens the miniband, while leaving the Dirac cone velocity almost unchanged.","lead":"Twisted bilayer graphene samples never have one single twist angle, and this paper shows that random local twist-angle variations fill in the electronic miniband gaps, broaden the miniband, and weaken superconductivity-like signatures while leaving the Dirac cone velocity intact. A smart generalist should read it because it offers a single disorder mechanism that could explain why nominally identical twisted graphene devices behave so differently, guiding cleaner experiments.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Quantitative disorder thresholds rest on a four-patch geometry; Appendix A shows thresholds shift with patch number, so the 6% gap-closing value is not yet a robust prediction.","rationale":"The reader's weakest-assumption analysis and my stress-test converge on the same point: the four-patch disorder model is the least secure element of the argument. The paper is otherwise careful: the lattice model is benchmarked against the Bistritzer-MacDonald continuum model, the KPM calculations are extensive, and the SOC toy model provides a useful cross-check of the qualitative behavior. However, the toy-model cross-check is not a direct test of the TBG patching scheme, and its own results show that patch number and patch shape change the effective disorder strength. Because the abstract and Sec. III state quantitative thresholds (gap closure near 6%, velocity robustness up to 15%), those numbers inherit the uncertainty of the four-patch construction. This does not invalidate the paper's qualitative picture, and it does not require moving the verdict away from CONDITIONAL; it does mean the quantitative predictions should be treated as model-dependent until a direct TBG-lattice test with more patches or smoother profiles is available. I therefore recommend no change to the reader's verdict, and I agree that the weakest assumption is the patch-geometry assumption.","tokens_in":22079,"tokens_out":5256,"duration_ms":66590,"concrete_test":"Repeat the TBG KPM density-of-states calculation of Figs. 4–5 with the same L = 569, NC = 2^17, and 100 disorder samples, but replace the four equal square patches with (i) 16 and 36 randomly sized rectangular patches and (ii) a smoothly varying twist-angle profile with a finite correlation length comparable to the moiré period. Extract the miniband gap Δ_MB(W_R) and renormalized velocity v(W_R) exactly as in Sec. III. If the gap closes at a W_R substantially different from ~6%, or if v changes before ~15%, the quantitative conclusions are geometry-dependent; if the extracted curves are unchanged, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's headline numbers—gap closure at WR ≈ 6% and velocity unchanged up to WR ≈ 15%—are obtained from a single disorder construction: four equal square patches, each with a uniform twist angle drawn from a box distribution (Sec. II, Fig. 2d). The physical correlation length of twist fluctuations in real samples is not known; devices exhibit both smooth gradients and hard domain walls, and the paper explicitly defers smooth domains to future work (Sec. V). The only test of the patching scheme is performed in the SOC toy model (Appendix A), not in the TBG lattice model itself. In that toy model, increasing the number of patches increases the effective disorder strength and shifts gap filling to smaller WQ (Figs. 9–11). If the same trend carries over to the TBG model, the quantitative threshold of roughly 6% would move, and the abstract's specific percentages would not be a property of twist-angle disorder but of a particular four-patch realization. The qualitative conclusions—gaps soften, miniband broadens, and the Dirac velocity is comparatively robust—may survive, but the quantitative content of the central claim is load-bearing and currently rests on an unverified modeling choice.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper argues that spatial variations of the twist angle are an intrinsic source of disorder in twisted bilayer graphene. The authors construct a real-space lattice model in which the twist angle enters as a continuous, locally variable parameter, and they show that its low-energy density of states agrees with the Bistritzer–MacDonald continuum model near the K and K′ points. Twist disorder is modeled by dividing the sample into four equal square patches, each with a twist angle drawn from a box distribution centered at 1.05°, and the density of states is computed with the kernel polynomial method for systems up to L = 569 and 100 disorder realizations. The main reported results are that the miniband gaps are filled in at disorder strengths of roughly 6% of the clean twist angle, the minibandwidth broadens, and the Van Hove peaks are smeared, while the Dirac velocity extracted from ρ(E) ∼ v^−2|E − E_D| remains nearly unchanged up to about 15% disorder. A quasiperiodic spin-orbit-coupled toy model in Appendix A reproduces the qualitative picture and is used to explore the dependence on the patch construction.","tokens_in":22225,"tokens_out":6637,"duration_ms":72910,"significance":"The qualitative contrast between a robust Dirac velocity and fragile miniband gaps and bandwidth, if correct, is a useful and non-obvious statement about twisted bilayer graphene that would help interpret sample-to-sample variability in transport and STM experiments. The paper has real strengths: the lattice model is benchmarked against the continuum model in Fig. 3; the calculations are nonperturbative and run on reasonably large systems; no observable is fitted to the target result; and the Appendix A toy model provides an independent cross-check. The paper also makes falsifiable statements, for example that samples with larger twist-angle variation should show a preserved low-energy V-shaped density of states but suppressed miniband gaps. The main caveat is that the specific numerical thresholds depend on modeling choices that are only partly tested within the paper itself.","major_comments":[{"comment":"The quantitative claim that the miniband gap closes at approximately 6% of the clean twist angle is not robust with respect to the disorder construction. The TBG calculation always uses exactly four equal square patches, and the effect of patch number and shape is examined only in the SOC toy model in Appendix A (Figs. 9–11), where increasing the patch number and randomizing patch shapes increases the effective disorder and shifts gap filling to smaller WQ. Because the real correlation length of twist fluctuations is unknown, the 6% number should be presented as a property of this four-patch ensemble rather than as a universal threshold. This matters because the discussion in Sec. IV uses the 6% threshold to interpret experimental sample-to-sample variations.","section":"Sec. III, Fig. 5(a,b); Sec. IV"},{"comment":"The clean model explicitly breaks C3 symmetry, and Appendix B shows that as a result the velocity does not vanish at the nominal magic angle: vmin remains finite for all interlayer tunneling strengths. This complicates the central velocity-robustness result, because the ratio v/v(w=0) shown in Fig. 5(c,d) is then measuring disorder insensitivity of a small but finite clean velocity, not of a magic-angle flat band. The text acknowledges this in Secs. II and IV, but it does not quantify how the disorder dependence of the velocity would change in a C3-preserving version of the model, which the paper cites as available in Ref. 24. I would ask the authors to state explicitly the range of w over which the “virtually unchanged” velocity claim is intended and to test the claim in a C3-symmetric model, or to soften the central claim accordingly.","section":"Sec. II, Appendix B, Eq. (B9)–(B13), Fig. 13"},{"comment":"The extraction of the gap-closing percentage is limited by the KPM resolution and by the absence of statistical error bars. The miniband gap is read from a DOS computed with NC = 2^17 and averaged over 100 samples; once disorder fills the gap, the value of WR at which the gap “completely” closes necessarily depends on the broadening scale set by NC and on the gap definition, which is not specified. Reporting the threshold as a range set by the KPM resolution or with an uncertainty estimate would make the central quantitative claim testable.","section":"Sec. III, Figs. 4–5"}],"minor_comments":[{"comment":"The sentence containing “we we are able” contains a typo and should read “we are able.”","section":"Sec. III"},{"comment":"“Fibbonnaci” should be spelled “Fibonacci.”","section":"Appendix A"},{"comment":"Stating that Tc is “in units of eV” is misleading because the exponential prefactor in Eq. (8) is arbitrary; please describe this quantity as a relative or qualitative measure of the Van Hove contribution.","section":"Sec. II, Eq. (8)"},{"comment":"It is ambiguous whether the random phases φj are drawn once per sample or independently in each patch; since Appendix A explicitly distinguishes these two options, the main text should state which choice is used in the TBG calculations.","section":"Sec. II"},{"comment":"Please specify the energy window used for the low-energy velocity fits in Fig. 5; the insets alone do not fully define the fitting range.","section":"Sec. III, Eq. (7)"},{"comment":"References 32 and 39 are the same paper and should be merged or one of them removed.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The paper introduces a timely topic and the central qualitative picture is plausible. The main technical risks are the four-patch disorder construction and the C3-breaking of the lattice model, both of which affect the quantitative claim. The reliance on companion models (Refs. 24 and 31) is acceptable as cross-validation, but an independent check with a C3-preserving TBG model or with a varied patch geometry would substantially strengthen the paper."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"I'd want you to know this one for the model more than for the numbers. The authors build a real-space lattice model of TBG where the twist angle is a continuously tunable, spatially varying parameter, and they use it to study single-particle twist-angle disorder nonperturbatively. That is genuinely new relative to the continuum models that dominate the field, and the benchmark against Bistritzer-MacDonald in Fig. 3 is honest and reasonably good. They then compute DOS with KPM and extract the gap, minibandwidth, Van Hove structure, and Dirac velocity. No target quantity is fitted; the results are computed. The qualitative conclusions—twist disorder fills the miniband gap, broadens the bandwidth, and leaves the low-energy Dirac velocity essentially intact—are consistent across the TBG model and the quasiperiodic toy model in Appendix A. The robust velocity is a real surprise and probably the most valuable claim in the paper.\n\nThe soft spots are real but not fatal. The disorder is modeled as four equal square patches with sharp twist-angle jumps, and the only systematic test of that choice is in the toy model, not in the TBG lattice model itself. Appendix A shows that increasing the patch number increases the effective disorder strength and shifts gap filling to smaller WQ; the authors admit as much in Sec. IV. So the headline numbers—gap closure near 6% and velocity unchanged up to 15%—are tied to a specific coarse-grained realization, not to twist-angle disorder as such. The C3-breaking in the lattice model also means the velocity never truly vanishes at the magic angle, which weakens the quantitative comparison to continuum theory exactly where the physics is most interesting. There are also no error bars or deposited code. None of this undermines the qualitative picture, but a careful reader should treat the percentages as illustrative, not predictive.\n\nWho is this for? People working on TBG disorder, whether experimental or theoretical, will get real value from the model and the robust-velocity observation. It deserves a serious referee: the central argument is coherent, the numerics are extensive, and the limitations are openly disclosed. I would recommend acceptance after revision, with the request that the authors either extend the patch-number check to the TBG model or visibly downgrade the specific percentages in the abstract and conclusions.","headline":"A first serious look at twist-angle disorder in TBG with a usable local model; the qualitative story (gaps fragile, Dirac velocity robust) is convincing, but the quantitative thresholds come from a four-patch construction that the paper itself shows is not universal.","tokens_in":22836,"tokens_out":1421,"would_cite":true,"duration_ms":17384,"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":"This paper argues that twist-angle disorder—unavoidable spatial variation in the relative rotation between graphene layers—is a distinct form of disorder that fills in the miniband insulation gap and broadens the flat band while leaving…","keywords":["twisted bilayer graphene","twist-angle disorder","moiré miniband","magic angle","density of states","Dirac cone velocity","Van Hove singularities","kernel polynomial method"],"falsifier":"Measure the local twist angle everywhere on a twisted-bilayer-graphene device by nanoscale scanning probe or diffraction mapping, and correlate the spread with the transport gap: a device whose twist-angle spread clearly exceeds about 6% of the nominal angle but still shows a hard miniband gap would disprove the predicted gap-filling threshold. Alternatively, a lattice-model calculation using many randomly sized patches rather than four equal squares would show whether the 6% threshold is an artifact of the patching scheme.","tokens_in":21769,"feed_emoji":"🌀","tokens_out":7124,"duration_ms":76235,"temperature":0.7,"pith_summary":"This paper introduces twist-angle disorder as a qualitatively new form of disorder in twisted bilayer graphene: even in perfectly clean samples, the twist angle between the two layers wanders across the sample, and that wandering is the dominant residual disorder. To study it, the authors construct a real-space lattice model in which the twist angle enters the interlayer hopping as a continuously tunable local parameter, so the angle can vary from patch to patch. Using kernel-polynomial evaluations of the density of states, they find that only about 6% variation in the twist angle relative to a clean 1.05 degrees completely fills the miniband gap, and that the minibandwidth grows substantially, while the Dirac-cone velocity extracted from the low-energy density-of-states scaling stays at its clean value up to roughly 15% variation. This separation between fragile gaps and stable Dirac cones matters because it offers a concrete way to understand why nominally identical twisted samples show very different correlated-insulator and superconducting behaviors, and it predicts that the semimetallic V-shaped conductance minimum should survive twist disorder.","feed_headline":"Twist-angle disorder fills twisted graphene's gaps but spares Dirac cones","feed_subtitle":"Gaps vanish at only 6 percent twist variation, yet the Dirac velocity survives to 15 percent","key_machinery":"The carrying object is a real-space lattice model of twisted bilayer graphene whose interlayer hopping amplitudes T0(r) and T1(r) carry the twist angle $\\theta$ explicitly through wavevectors with magnitude k_theta = 2 k_D sin($\\theta$/2), making the angle a continuous, local parameter rather than a global boundary condition. The model preserves C2T symmetry, which keeps the Dirac nodes gapless, but breaks C3 symmetry, and it reproduces the standard continuum model near the K and K' points. Disorder is injected by partitioning the sample into four equal square patches, each assigned a twist angle drawn from the box distribution [(1-W_R/2)theta0, (1+W_R/2)theta0]; the density of states is computed with the kernel polynomial method, and the velocity is read off from the low-energy scaling rho(E) ~ $v^{{-2}}$|E-E_D|. A quasiperiodic spin-orbit toy model with many randomly sized patches is used to check that the four-patch patching scheme does not drive the qualitative conclusions.","core_discovery":"The central claim is that twist-angle disorder acts selectively: it destroys the coherence of the moire miniband while leaving the Dirac cone's low-energy structure intact. Concretely, for a clean twist angle near magic angle, modeled with a box distribution of local angles of width W_R, the single-particle gap separating the miniband from higher states is completely filled in by W_R near 6% of the clean angle, and the minibandwidth broadens monotonically with disorder until the gap disappears. In contrast, the renormalized Dirac velocity v, extracted from the scaling rho(E) ~ $v^{{-2}}$|E-E_D|, remains essentially at its clean value for W_R up to about 15%, with the minimum only beginning to round out near the magic-angle condition. The Van Hove peaks lose their logarithmic divergence and become analytic in energy under disorder, yet their positions and mutual separation are nearly unaffected, while the estimated mean-field BCS transition temperature at a Van Hove peak is strongly suppressed. The authors read this as meaning that twist disorder weakens the many-body correlation physics by widening the effective bandwidth, lowering the effective U/t, while not erasing the semimetallic Dirac scaling observed in transport.","pith_inferences":["The paper does not test this directly, but the toy-model appendix suggests that real samples with many small twist domains could fill the gap at even smaller angle spreads than the 6% reported for four patches, because increasing the number of randomly placed patches strengthens the effective disorder.","A device-level test would be to measure the local twist-angle distribution with scanning probes, then compare the spread with the transport gap: the paper's mechanism predicts a sharp correlation between angle spread and the disappearance of the insulating gap.","The same lattice-model construction should transfer to other twisted van der Waals bilayers and to multilayer twist devices, where the velocity-versus-gap dichotomy could serve as a diagnostic separating twist disorder from charge-impurity disorder."],"forward_implications":["Miniband insulating gaps at integer fillings should be strongly sample-dependent: a twist-angle spread of roughly 6% of the clean angle suffices to destroy the single-particle gap, so correlated-insulator gaps built on top of it will be weakened or absent in many samples.","The V-shaped low-energy conductance minimum at charge neutrality should survive moderate twist disorder, since the Dirac velocity and the linear density-of-states scaling remain essentially unchanged up to about 15% twist variation.","If superconductivity in twisted bilayer graphene is BCS-like and enhanced by the Van Hove singularity, twist disorder suppresses the effective Tc, so samples with larger twist variation should not superconduct.","Van Hove peak positions remain fixed under disorder, so scanning probes searching for the peaks should still find them near their clean energies even when the peaks are broadened and no longer divergent.","Twist disorder broadens the minibandwidth, which reduces the effective correlation strength U/t, meaning the disorder acts to weaken correlated phases even though the Dirac cone remains flat."],"supporting_citations":[{"why":"the continuum model whose band structure, including gaps, velocity renormalization, and Van Hove peaks, the real-space lattice model is designed to reproduce near the K and K' points.","marker":"[21]"},{"why":"the kernel polynomial method used for the numerically exact density-of-states evaluations at the large system sizes studied.","marker":"[26]"},{"why":"the experiment reporting lead-dependent insulating gaps and a robust V-shaped conductance minimum, used to argue that twist disorder is present and consistent with the calculations.","marker":"[7]"},{"why":"relaxed band-structure calculations that set the AA-to-AB tunneling ratio w0/w1 = 0.75 used throughout the model.","marker":"[27, 28]"},{"why":"the quasiperiodic spin-orbit model that provides the independent check that the four-patch twist-disorder scheme does not qualitatively alter the conclusions.","marker":"[24]"},{"why":"the earlier continuum formulation of twisted bilayer graphene whose momentum-space structure the model follows.","marker":"[20]"},{"why":"the source of the graphene tight-binding hopping parameter t = 2.8 eV used in the lattice Hamiltonian.","marker":"[34]"}],"fun_headline_variants":["Twist disorder kills TBG gaps, leaves Dirac cones intact","Gaps go, cones stay: twist disorder's selective toll on TBG","In twisted graphene, disorder erases gaps but spares Dirac physics","Twist-angle randomness fills flat-band gaps, not Dirac cones","6% twist variation kills gaps, Dirac survives to 15%"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"That representing the sample as four equal square patches, each with a single uniform twist angle drawn from a box distribution, captures the real spatial variation of the twist angle; if real samples have many small domains or smooth gradients, the quantitative thresholds could shift.","fun_headline_variants_meta":{"raw":{"variants":["Twist disorder kills TBG gaps, leaves Dirac cones intact","Gaps go, cones stay: twist disorder's selective toll on TBG","In twisted graphene, disorder erases gaps but spares Dirac physics","Twist-angle randomness fills flat-band gaps, not Dirac cones","6% twist variation kills gaps, Dirac survives to 15%"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000902,"raw_usage":{"total_tokens":3929,"prompt_tokens":1041,"completion_tokens":2888,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":657,"completion_tokens_details":{"reasoning_tokens":2797}},"tokens_in":657,"tokens_out":2888,"duration_ms":23616,"temperature":1.0,"reasoning_tokens":2797,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:36:29.948624+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the local twist angle everywhere on a twisted-bilayer-graphene device by nanoscale scanning probe or diffraction mapping, and correlate the spread with the transport gap: a device whose twist-angle spread clearly exceeds about 6% of the nominal angle but still shows a hard miniband gap would disprove the predicted gap-filling threshold. Alternatively, a lattice-model calculation using many randomly sized patches rather than four equal squares would show whether the 6% threshold is an artifact of the patching scheme.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"the earlier continuum formulation of twisted bilayer graphene whose momentum-space structure the model follows."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"the source of the graphene tight-binding hopping parameter t = 2.8 eV used in the lattice Hamiltonian."}],"review_version":1}