{"id":"6056f7a4-54fb-4b2b-9c20-ef42b97b528c","arxiv_id":"2607.13199","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Two-stage inverse design produces disordered 2D photonic networks whose isotropic complete band gaps reach 14.7% gap-to-midgap, rivaling honeycomb photonic crystals.","lead":"The paper presents a two-stage computer design method for disordered photonic networks that block light of any direction and polarization over a wide frequency range. The new networks achieve band gaps nearly as wide as the best crystalline photonic crystals, while remaining isotropic and disordered.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 100,000-vertex effective-size claim rests on an unvalidated self-averaging assumption; single large-N DoS calculations are needed before the thermodynamic-limit statement is supported.","rationale":"The paper's actual goal is to demonstrate a disordered isotropic network whose complete PBG persists as system size grows, with the strongest claim being the thermodynamic-limit candidate status. The most fragile premise is the ensemble-to-single-large-system equivalence. The comparison to Florescu et al. is weakened by using different wall widths (w=0.08 vs 0.10) and an unoptimized baseline; however, this affects only the quantitative improvement factor, not the existence of a 14.7% gap. The parameter fitting of r and w similarly shifts the number but is not central to the qualitative claim. The Gamma-point-only supercell evaluation is standard practice for finite-cell DoS and is not a fatal flaw. The absence of code/SI is a reproducibility issue but not a scientific objection. The self-averaging assumption, by contrast, is required to convert '500 robust finite-size gaps' into 'a gap in the thermodynamic limit.' The paper even acknowledges this explicitly, and no finite-size scaling across connected system sizes is reported. A small number of single N=8000/16000 calculations would settle it. Until then, the correct verdict remains CONDITIONAL: accept the finite-size result, but withhold the thermodynamic-limit claim.","tokens_in":12351,"tokens_out":5970,"duration_ms":61510,"concrete_test":"Run the same MPB supercell calculation for 5-10 independent single networks of N=8000 (and, if feasible, one or two N=16000) vertices generated with the two-stage protocol (chi1=0.37, chi2=0.24, r=0.24, w=0.08). Compare the largest zero-density interval in the averaged DoS and the density in the gap bins with the ensemble-stacked result. If any large single realization has states within the 14.7% gap, or if the gap width drops by more than 20%, the self-averaging assumption fails and the effective-100,000-vertex/thermodynamic-limit claim should be withdrawn. If instead the single large-N gap remains open and matches the ensemble gap to within these tolerances, the central claim is substantially strengthened.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Load-bearing assumption: the ensemble DoS of 500 independent 2000-vertex periodic realizations is treated as equivalent to a single connected 100,000-vertex network. The paper states this in Sec. 1 ('Assuming a self-averaging property...') and qualifies it in Sec. 5, but offers no single large-N computation or finite-size scaling. The concern is not that the 14.7% gap is an artifact of binning: an empty bin in the ensemble average means every realization has zero states in that bin, so the gap is real for the 2000-vertex systems. The issue is the extrapolation. In disordered media, the average DoS in a large system can be dominated by rare regions that are exponentially unlikely in a single finite realization but unavoidable in a single infinite system (Lifshitz tails). If such tails appear inside the gap for N >> 2000, the ensemble stack overestimates the persistence of the gap. Because the paper's strongest claim is about the thermodynamic limit, this assumption is load-bearing and currently unsupported.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces a two-stage inverse design procedure for generating disordered, stealthy hyperuniform trivalent photonic networks in two dimensions. The first stage optimizes the Delaunay-centroid point pattern to be close to stealthy hyperuniform, and the second stage refines the vertex positions of the decorated network (discs at vertices, rectangular walls along bonds) to directly enforce stealthy hyperuniformity of the two-phase medium via its spectral density. The authors compute photonic densities of states with MPB for 500 independent 2000-vertex realizations and report an ensemble-averaged complete photonic band gap with gap-to-midgap ratio 14.7%, compared with 4.0% for networks made with the prior Florescu et al. method and 16.7% for a honeycomb photonic crystal with the same dielectric parameters. They also report a single-best realization gap of 19.0%, near-perfect TE–TM overlap, strongly suppressed defect-state density, and claim that the ensemble approach effectively probes a network with 100,000 vertices, making these structures the most promising candidates for an isotropic amorphous complete PBG in the thermodynamic limit.","tokens_in":12551,"tokens_out":8543,"duration_ms":109674,"significance":"If the central claims hold, this would be a notable advance: a direct, topology-changing inverse design route to truly stealthy hyperuniform two-phase networks with wide, isotropic complete photonic band gaps, supported by unusually large ensemble statistics (500 realizations). The paper is honest about its limitations, including the heuristic parameter choice and the self-averaging assumption. However, the headline comparison against the prior method is not controlled, and the thermodynamic-limit conclusion rests on an unvalidated extrapolation. These issues are load-bearing for the paper's main claims, so the manuscript needs substantial revision and additional numerical evidence before the claims can be accepted.","major_comments":[{"comment":"The comparison between the new method and the Florescu et al. baseline is not parameter-matched. The new networks use r=0.24, w=0.08, chosen by 'rough optimization' on a few small networks, while the baseline uses r=0.24, w=0.10 from 'established values' and is not re-optimized for the present ensemble. Since the honeycomb crystal's gap changes from 16.7% to 24.2% when r and w are optimized (reported in the same section), parameter choice alone can shift gap-to-midgap ratios by several percent. The 14.7% vs 4.0% difference therefore cannot be attributed solely to the new inverse-design method. The authors should either optimize the baseline (e.g., a wall-width scan for the prior method) or present matched-parameter comparisons, and qualify the 'nearly an order of magnitude wider' claim accordingly.","section":"§4 and Fig. 2"},{"comment":"The thermodynamic-limit claim rests on an unproven self-averaging assumption. The ensemble of 500 independent 2000-vertex realizations is not equivalent to a single 100,000-vertex network: in one large system, rare regions can produce Lifshitz tails that are exponentially unlikely in any finite realization but unavoidable in the thermodynamic limit. The paper explicitly states 'Assuming a self-averaging property...' in Sec. 1 and concedes in Sec. 5 that a numerical study can probe only finite size, but it provides no finite-size scaling of the photonic DoS or gap. The N=4000,8000 calculations in Fig. 3 concern only the order parameter I6, not photonic properties. To support 'no indication of closing' and 'effective system size of 100,000 vertices,' the authors need at least one single-system calculation at larger N (e.g., 8000 or 16000 vertices) or a systematic gap-vs-N study, and should","section":"Secs. 1 and 5"},{"comment":"Evaluating eigenmodes only at the Γ-point is not justified by isotropy. In a periodic supercell approximation, the full density of states requires an integration over the mini-Brillouin zone; a single k-point can miss states at other k that fall inside the apparent gap, especially near band tails. The manuscript states only that 'due to the isotropy of the networks, this single k-point is sufficient' without a convergence test or reference validating this practice for this class of systems. The authors should compare Γ-only DoS with a few sampled k-points in the mini-BZ for a subset of realizations, or with full BZ integration for smaller N, and report the effect on the gap ratios.","section":"Methods: Photonic density of states"}],"minor_comments":[{"comment":"Specify whether the plotted DoS is the mean or median over the 500 realizations. Fig. 4 uses the median; Fig. 2 says 'averaged', which is ambiguous.","section":"Fig. 2 caption"},{"comment":"To support the claim of 'unprecedented near-complete overlap' between TE and TM gaps, report the TE and TM gap widths separately and state the overlap fraction.","section":"§4"},{"comment":"Describe how the complete PBG edges are extracted from the binned histogram (e.g., first/last empty bin, interpolation between bins) and comment on the sensitivity of the reported ratios to the bin width Δω≈0.00092.","section":"Methods"},{"comment":"The phrase 'previously widest known isotropic complete PBGs' should be tied to a specific prior value (12.6% is later mentioned) and should note explicitly that the benchmark comparison uses different wall widths for the two methods.","section":"Abstract and §4"},{"comment":"The main text repeatedly refers to the SI for details of the two-stage optimization; for a standalone reader, a flowchart or pseudocode of the algorithm would improve reproducibility.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The paper is well within the scope of the journal and the two-stage design idea is interesting. The main risk is that the headline improvement over the prior method may be inflated by parameter mismatch, and the thermodynamic-limit claim is not yet supported by finite-size scaling. I believe these issues are addressable with additional simulations and revised wording, so major revision is appropriate rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Joel — quick take on arXiv:2607.13199. The genuinely new thing is the two-stage inverse design: instead of mapping a stealthy hyperuniform point pattern to a network and hoping the network inherits the property, they optimize the point pattern from the start so that the Delaunay centroids (the network vertices) are stealthy hyperuniform, then refine the decorated network against its own spectral density. That directly targets the two-phase medium, which is the right thing to do. And it appears to work: 500 independent 2000-vertex networks give an ensemble complete PBG of 14.7% gap-to-midgap ratio, with TE–TM overlap nearly complete and sharp band edges. The single best realization hits 19%, and the ensemble result is comparable to the honeycomb crystal's 16.7% at the same dielectric contrast. That is a real step beyond the Florescu et al. design, which gives only about 4% under the comparison conditions. The order-disorder transition analysis (χ_c≈0.39) is also a nice piece of methodology.\n\nThe soft spots are real but not fatal to the core result. First, the comparison is not fully controlled: the new networks use w=0.08 while the Florescu baseline uses w=0.10, and the baseline was not re-optimized. So the 'nearly an order of magnitude wider' claim is a bit overstated; a matched-parameter comparison would be more honest. Second, r and w for the new method were chosen by 'rough optimization' on a few small networks, so some of the 14.7% is fitting. The authors admit this in the Discussion, which is to their credit. Third, the 100,000-vertex claim. The stress-test note is right: the ensemble stack assumes self-averaging of the photonic DoS. The paper states the assumption in Sec. 1 and concedes in Sec. 5 that a numerical study can only probe finite sizes. No single large-N calculation or finite-size scaling is offered. This doesn't undermine the existence of a gap in 2000-vertex systems, but it does mean the thermodynamic-limit extrapolation is currently unsupported. Lifshitz tails from rare regions are a real concern at larger N. Finally, no code or SI is included in the text I saw, so independent reproduction is not yet possible.\n\nWho's this for? Anyone working on hyperuniform photonics or inverse design of disordered materials. It deserves a serious referee — the method is novel, the numbers are plausible, and the flaws are fixable with a re-run at matched parameters, a few single large-N DoS calculations, and the SI/code. I'd send it out.","headline":"New inverse design directly targets stealthy hyperuniformity of the network itself and reports a 14.7% ensemble isotropic complete PBG; the headline thermodynamic-limit claim rests on an unproven self-averaging assumption.","tokens_in":13103,"tokens_out":2125,"would_cite":true,"duration_ms":18875,"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":"A two-stage inverse design makes disordered, isotropic networks with complete photonic band gaps nearly as wide as those of crystals.","keywords":["photonic band gaps","hyperuniformity","stealthy","inverse design","disordered networks","two-phase media","isotropic photonic materials","photonic metamaterials"],"falsifier":"Set up a single connected stealthy hyperuniform trivalent network with 50,000–100,000 vertices and compute its photonic density of states (or measure transmission through a fabricated sample of comparable size); if the complete band gap narrows below, say, 10% or closes outright, the ensemble-based persistence claim is refuted.","tokens_in":12169,"feed_emoji":"🔆","tokens_out":7995,"duration_ms":75808,"temperature":0.7,"pith_summary":"Complete photonic band gaps — frequency ranges in which light cannot travel in any direction or polarization — are normally the province of ordered crystals. This paper claims that a two-stage inverse design method can make disordered, isotropic networks that are themselves stealthy hyperuniform (density fluctuations suppressed at small wavenumbers) and that these networks sustain wide complete band gaps. Averaged over 500 independent 2000-vertex realizations — an effective 100,000-vertex system — the gap-to-midgap ratio is 14.7%, close to the 16.7% of a honeycomb crystal and nearly four times the 4.0% of the previous best disordered method. The authors interpret this as evidence that isotropic complete band gaps can persist toward the thermodynamic limit, not just in small samples.","feed_headline":"Disordered networks rival crystals for full photonic band gaps","feed_subtitle":"Two-stage inverse design gives isotropic 14.7% gaps, approaching honeycomb crystals' 16.7%.","key_machinery":"The load-bearing object is the two-stage inverse design loop. Stage 1 solves an inverse problem for the triangle centroids of the point pattern's triangulation, using gradient descent on a structure-factor loss; because the triangulation is recomputed at every step, the network topology can reorganize, avoiding the bond-distortion problem that defeats naive vertex shifting. Stage 2 keeps topology fixed and directly minimizes the spectral density of the final two-phase medium, which is the statistic that defines stealthy hyperuniformity for a heterogeneous material. The coupling of global topology change with direct spectral-density refinement is what produces regular pentagon/hexagon/heptago","core_discovery":"At its core, the paper argues that the decisive step is to require stealthy hyperuniformity of the two-phase dielectric structure itself, not just of the point pattern used to scaffold it. The two-stage optimization first moves the progenitor points so that the centroids of the triangulation — the future network vertices — are as stealthy hyperuniform as possible while the topology is free to change; then it shifts the network vertices so the spectral density of the decorated structure (dielectric disks at vertices and thin walls along bonds) is minimized out to a cutoff wavenumber. The result, for 500 realizations of 2000 vertices, is an ensemble density of states with a complete gap-to-mid","pith_inferences":["If the self-averaging assumption is correct, the effective 100,000-vertex result is a strong hint that the gap stays open in the thermodynamic limit; but the assumption itself is only tested indirectly here, and a true single large-system simulation would settle it.","The observed disorder-to-order transition at χ≈0.39 for centroid-constrained patterns suggests a ceiling on how stealthy a trivalent network can be before it crystallizes; pushing beyond it would require relaxing local topology constraints.","The same two-stage logic could be exported to three dimensions, or to phononic and acoustic systems, where isotropic complete gaps remain an open target.","Since the band-gap width appears tied to the spectral-density suppression between the two cutoffs, deliberately engineering a flatter, deeper suppression ring might be a direct route to even steeper band edges."],"forward_implications":["If the claim holds, isotropic amorphous photonic materials can achieve complete band gaps nearly as wide as the best anisotropic crystals, without the directional constraints of periodicity.","The ensemble result implies that fabrication of only modest 2000-vertex samples can already deliver near-thermodynamic-limit performance, easing experimental realization.","Because the structures are trivalent networks of disks and walls, they are compatible with additive manufacturing, enabling free-form waveguides, arbitrary-orientation cavities, and isotropic thermal emitters.","The near-complete TE-TM overlap means polarization-independent operation over a wide frequency range, simplifying device design.","Further tuning of disk radius and wall thickness, and deeper suppression of the intermediate-wavenumber scattering ring, could push the gap closer to or beyond that of honeycomb crystals."],"fun_headline_variants":["Two-stage inverse design widens isotropic photonic gaps to 14.7%","Hyperuniform network design rivals honeycomb crystals for gap width","Disordered hyperuniform networks open near-crystal complete gaps","Stealthy hyperuniform inverse design yields wide isotropic PBGs","Ensemble-designed hyperuniform networks beat prior isotropic gaps"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The argument that the gaps persist at 100,000 vertices rests on assuming the photonic density of states is self-averaging, so that stacking 500 independent 2000-vertex simulations is equivalent to one 100,000-vertex system; if that equivalence fails, or if the single-Gamma-point supercell computation misses modes, the gap could close in a truly large connected sample.","fun_headline_variants_meta":{"raw":{"variants":["Two-stage inverse design widens isotropic photonic gaps to 14.7%","Hyperuniform network design rivals honeycomb crystals for gap width","Disordered hyperuniform networks open near-crystal complete gaps","Stealthy hyperuniform inverse design yields wide isotropic PBGs","Ensemble-designed hyperuniform networks beat prior isotropic gaps"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000496,"raw_usage":{"total_tokens":2292,"prompt_tokens":790,"completion_tokens":1502,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":534,"completion_tokens_details":{"reasoning_tokens":1426}},"tokens_in":534,"tokens_out":1502,"duration_ms":12857,"temperature":1.0,"reasoning_tokens":1426,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T05:53:30.241774+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Set up a single connected stealthy hyperuniform trivalent network with 50,000–100,000 vertices and compute its photonic density of states (or measure transmission through a fabricated sample of comparable size); if the complete band gap narrows below, say, 10% or closes outright, the ensemble-based persistence claim is refuted.","supporting_citations":[],"review_version":1}