{"id":"7172ffd1-109b-4ab9-bd7c-59813f4f6d34","arxiv_id":"2509.01625","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A nonlinear coordinate transformation lets designers turn nonlinear tight-binding computing models into physical metamaterial geometries.","lead":"This paper introduces a mathematical mapping that translates nonlinear computation models into physical metamaterial geometries, so mechanical structures can perform tasks like optimization, memory, and speech classification. The authors demonstrate the approach with three simulated devices: an Ising machine, a mechanical racetrack memory, and a speech-classifying acoustic metamaterial.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The instantaneous-response (Born-Oppenheimer) assumption behind Eq. 1 is load-bearing but never quantified; the speech-classification demo, where nonlinearity is essential and no full-wave check exists, is exactly where a violation would silently break the pipeline.","rationale":"The reader's weakest assumption—time-scale separation between the Wannier coordinates and the modes encoded in Ψ′_ij—is exactly the condition on which the central claim depends. The coordinate transformation in Eq. 1 is only a useful nonlinear mapping if the excluded modes adiabatically follow the slow coordinates. The paper acknowledges this limitation in the main text and in Appendix D, but it offers no error bound and no direct verification for the examples. The speech example is the most vulnerable: it is the only demonstration where the nonlinearity is essential to the computation, the operating point is deliberately near resonance, and no full-wave simulation is performed. The racetrack example is less threatening because it operates in a quasistatic, overdamped regime where the time-scale-separation condition is more naturally satisfied; the CIM example has some spectral separation enforced by the choice of low damping. Thus, a single quantitative check on the speech geometry's Wannier derivatives—comparing the static and dynamic response—would directly test the load-bearing assumption. If the extracted coefficients are insensitive to the frequency correction, the central claim survives; if not, the claimed generality is unsupported. The reader's conditional verdict is appropriate, and this stress-test does not change it.","tokens_in":25731,"tokens_out":6149,"duration_ms":82999,"concrete_test":"Recompute the Wannier derivatives for the speech-classification geometry using dynamic response at the combination frequency Ω = ω_i + ω_j instead of the static eigenvalue λ_i in Eq. D2, and compare the extracted Λ tensor (and the simulated yes/no classification error at the 3% operating amplitude) against the static extraction. If the relative change in the relevant Λ coefficients or in classification error exceeds 10%, the instantaneous-response assumption is load-bearing and the speech demonstration does not validate the central claim; if the change is negligible, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that Eq. 1 accurately maps tight-binding coordinates to metamaterial excitations in the nonlinear regime—requires a separation of time scales: the high-frequency modes encoded in Ψ′_ij must respond instantaneously to changes in the slow coordinates q_i. Appendix D explicitly states that Ψ′_ij is computed assuming the excluded modes are driven at the single-mode frequencies ω_i and ω_j, whereas the product q_i q_j generates oscillations at ω_i ± ω_j. If the excluded modes are not spectrally well separated from these combination frequencies, the extracted nonlinear tensors are frequency-dependent and the effective model is wrong. The paper calls this a Born-Oppenheimer approximation and argues that design freedom can enforce the separation, but it never quantifies the residual error for any of the three devices. The speech classifier is the sharpest test: the drive is near resonance (ω_m = 1.085 ω_0), the nonlinearity is essential (classification error drops from 18% to 3%), and the paper admits no full-wave simulation can be run. The claim that the model is 'expected to be highly accurate' is based on analogies to Fig. 2d, not on a check of the spectral-separation condition. A violation here would not just weaken a demonstration; it would invalidate the design pipeline's core premise for the very regime the paper highlights.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a method for designing nonlinear perturbative metamaterials from tight-binding (TB) models. The central ingredient is the nonlinear coordinate transformation u(x,t) = q_i(t)Ψ_i(x) + (1/2) q_i(t)q_j(t)Ψ′_ij(x), where Ψ_i are localized Wannier-type basis functions and Ψ′_ij are their sensitivities to deformation. The authors show how these objects can be extracted from finite-element clusters with linear-time cost, how the effective nonlinear TB tensors follow from Taylor expansion of the energy, and how the resulting model matches full-wave simulations for a small metasurface (Fig. 2). They then demonstrate three applications: a coherent Ising machine, an elastic racetrack memory, and a reservoir-computing speech classifier. The paper argues that this constitutes a general design pipeline for embodying nonlinear TB computations in metamaterials.","tokens_in":26031,"tokens_out":3655,"duration_ms":43737,"significance":"If the central claim is valid, the paper is a significant advance: it extends perturbative metamaterial design from the linear regime, where it is well established, to nonlinear computation, where the deformation-dependence of the mode basis has been a recognized obstacle. The paper's strengths include direct full-wave validation of the core coordinate-transformation method in Fig. 2, convergence tests with cluster size in Fig. 7, explicit treatment of the Born-Oppenheimer assumption and its limitations in Appendix D, and the demonstration that the extracted TB nonlinear tensors are sparse and local, which is essential for design. The three showcases are ambitious and, for the speech classifier, produce a concrete falsifiable prediction. However, as detailed below, the load-bearing time-scale-separation assumption is not quantitatively validated for any of the three examples, and one of the showcase validations relies on modifying the target model after the fact. With these gaps addressed, the method would be a compelling contribution.","major_comments":[{"comment":"The central validity of Eq. (1) rests on a time-scale separation that is asserted but not quantified. Appendix D explicitly states that the Wannier derivatives are computed assuming excluded modes are driven at ω_i and ω_j, whereas the product q_i q_j contains components at ω_i ± ω_j; the error is small only if the excluded modes are spectrally well separated from these combination frequencies. No spectral gap or residual error is reported for any of the three geometries. This matters most for the speech classifier: Methods D states that no full-wave simulation can be conducted, and Fig. 5c is entirely based on TB simulations. The nonlinearity is essential there (error drops from 18% to 3%), and the drive frequency is near resonance (ω_m = 1.085ω_0). To support the paper's central claim, the authors should either quantify the spectral separation for the speech geometry and estimate the r","section":"Appendix D, Eq. (D16); Methods D; Fig. 5"},{"comment":"The racetrack demonstration does not validate the target tight-binding model stated in the text. The target model (Eq. 2) contains only nearest-neighbor hopping, on-site Kerr nonlinearity, and the coupling to the compression field. Yet Fig. 4c shows a visible disagreement between the target TB trajectory and the high-fidelity equilibrium continuation; the authors then state that incorporating long-range interactions in the TB model reproduces the trajectory. This means the embodied model differs from the model the design was supposed to realize. Even if the long-range terms are extracted from the geometry rather than fitted to the output, the demonstration would be stronger if the authors either (i) used geometry optimization to suppress these interactions, as suggested in the linear metamaterial literature, or (ii) clearly presented the racetrack as an example where the effective model","section":"Example 2, Fig. 4"},{"comment":"The coherent Ising machine demonstration uses a frustration-free Ising problem (stated in the text: 'here set to encode a frustration-free problem'). The convergence to a ground state in Fig. 3d is a consistency check of the mapping, but it does not demonstrate the ability to approximate solutions of combinatorial optimization problems, which is the motivation of the section and the abstract. A single small frustrated instance (even with a handful of spins) would substantially strengthen the claim. As written, this example is better described as validating the parametric-oscillator network physics than as showcasing optimization capability.","section":"Example 1, Fig. 3"}],"minor_comments":[{"comment":"The sentence 'see [ref] for an experimental realization...' contains a literal '[ref]' placeholder with no reference. This is a missing citation that should be filled before publication.","section":"Example 2, first paragraph"},{"comment":"The claim 'the regime at which the maximum classification accuracy is achieved is close to that in Fig. 2d in terms of relative nonlinearity' is made without defining 'relative nonlinearity' or providing a quantitative comparison. Please define the measure and give the numbers.","section":"Methods D, last paragraph"},{"comment":"Typo: 'valye' should be 'value'. Also, the sentence 'Excluding all derivatives essentially eliminates the nonlinear contribution...' is clear, but the factor of 4 and factor of 20 are reported without error bars or confidence intervals; specify the error metric (e.g., relative L2 error) and whether the result is for one representative geometry.","section":"Methods B.3, Fig. 7"},{"comment":"The index structure in the term Λ^mr_ijkl q_j q_l appears twice in Eq. (B14) with different dummy indices but the same tensor; please verify and clarify the notation. Also, 'the the kinetic energy' typo in the paragraph before Eq. (B12).","section":"Appendix B, Eq. (B14)"},{"comment":"The paper would benefit from a data/code availability statement. The method relies on a substantial software implementation (FEniCSx, GMSH, custom automation); providing the code or at least a repository would materially improve reproducibility.","section":"General"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The thing to know: the core method is real. Adapting the modal-derivative coordinate transformation from nonlinear model reduction to extract tight-binding models of perturbative metamaterials is a genuine step forward, and the cluster-based computation of Wannier derivatives makes it scalable. Fig. 2 is an honest benchmark — the coordinate-transformed model tracks the full-wave simulation where the bare perturbative model fails — and Fig. 7 shows the convergence with cluster size. That is the load-bearing evidence, and it holds up.\n\nThe three demonstrations are where I get cautious. The racetrack example is suggestive but required ad hoc long-range interactions to match the high-fidelity continuation, and the text cites \"[ref]\" for the mechanism. That unresolved citation is sloppy and needs fixing. The CIM uses a frustration-free problem and validates against the ground state of the extracted model itself, so it is a weaker demonstration than the abstract implies. The speech classifier is the softest spot: the nonlinearity is essential (error drops from 18% to 3%), the drive is near resonance, and the paper admits no full-wave simulation is possible. The Born-Oppenheimer assumption — that the modes in the Wannier derivatives respond instantaneously at ω_i and ω_j while the product q_i q_j oscillates at ω_i ± ω_j — is acknowledged in Appendix D but never quantified for any of the three devices. The claim that the speech model is \"expected to be highly accurate\" rests on analogy to Fig. 2d, not on a check of spectral separation. This is a legitimate limitation, not a fabrication, and the design-for-analysis argument is reasonable, but the paper should be honest that the pipeline's validity in the strongly nonlinear regime is still a conditional promise.\n\nThe derivation in the appendices is careful, and the sparsity of the extracted tensors is a real practical advantage. The paper deserves a serious referee and would benefit from revisions: quantify the spectral-separation error in at least one example (even a small system), provide a stronger check for the speech device (reduced-order or approximate full-wave), and replace the dangling [ref]. I would bring it to a reading group and would cite it if I worked on metamaterial computing. Send it to peer review.","headline":"A solid method paper: the modal-derivative mapping to nonlinear metamaterials is genuinely new and benchmarked against full-wave simulations, but the three showcases are uneven and the Born-Oppenheimer assumption is never quantified, especially where it matters most in the speech demo.","tokens_in":26527,"tokens_out":1938,"would_cite":true,"duration_ms":24684,"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 nonlinear coordinate transformation lets metamaterials implement any computation expressible as a tight-binding model, from optimization to memory to speech classification.","keywords":["nonlinear metamaterials","tight-binding models","coordinate transformation","Wannier functions","Born-Oppenheimer approximation","Ising machine","racetrack memory","reservoir computing"],"falsifier":"Take a geometry with a deliberately small spectral gap, extract the effective nonlinear tensor with and without the Ψ' terms, and compare full nonlinear finite-element transients: if the coordinate-transformed model fails to reproduce the full-wave response at amplitudes where nonlinearity is visible, the central claim is falsified. A simpler physical test: fabricate a small plate-and-bar structure, drive it at amplitudes where the Kerr shift is measurable, and check the predicted frequency shift against the effective coefficient extracted from Eq. 1.","tokens_in":25610,"feed_emoji":"🧮","tokens_out":4605,"duration_ms":54212,"temperature":0.7,"pith_summary":"The paper proposes a way to turn a discrete, tight-binding model—a network of sites with local potentials and couplings—into a physical elastic metamaterial that actually computes, including the nonlinear terms needed for computation. The key is a coordinate transformation that maps abstract coordinates to the material's deformation field, adding 'basis-function sensitivities' that capture how mode shapes deform under nonlinearity. Without these correction terms, naive nonlinear models are less accurate than linear ones. With them, the authors show by full-wave simulation that three quite different computations run in designed geometries: a coherent Ising machine solving a combinatorial optimization problem, a mechanical racetrack memory transporting bits, and a reservoir computer classifying 'yes' and 'no' speech. If correct, any computation expressible as a nonlinear tight-binding model becomes a design target for mechanical metamaterials.","feed_headline":"Metamaterials can now embody any nonlinear tight-binding computation","feed_subtitle":"A coordinate transformation maps discrete models to material geometries, validated on an optimizer, a memory, and a speech classifier.","key_machinery":"The load-bearing object is the pair of localized Wannier functions Ψ_i (basis functions obtained by projecting symmetry-selection functions onto cluster eigenmodes) and Wannier derivatives Ψ'_ij (their sensitivities to deformation, computed from modal derivatives). The nonlinear coordinate transformation built from these objects carries the argument: it produces sparse, geometry-associated tensors in the local basis, making the design problem additive and local, and it renormalizes nonlinear coefficients by accounting for how high-frequency modes 'screen' nonlinear stress. The method's validity rests on a time-scale separation analogous to the Born-Oppenheimer approximation, which the author","core_discovery":"The central claim is that the mapping between a nonlinear tight-binding model and a metamaterial geometry is made accurate by extending the displacement ansatz to first order in the deformation dependence of the localized basis functions. For each site i, the field is qi(t)Ψ_i(x) plus 1/2 qi(t)qj(t)Ψ'_ij(x), where Ψ'_ij is the sensitivity of basis function i to deformation of coordinate j. Extracting the effective stiffness tensors K, Γ, Λ from this ansatz yields models whose transient and steady-state responses match full nonlinear finite-element simulations, whereas the bare nonlinear model without Ψ' overestimates the Kerr coefficient by roughly a factor of four. Using this map, the paper","pith_inferences":["A natural next step, not pursued here, is co-optimizing the readout and the geometry, since the pipeline is differentiable; the paper notes this possibility.","The same screening picture suggests that the coordinate transformation could transfer to optical or acoustic metamaterials wherever a spectral gap separates slow coordinates from fast screening modes.","If the spectral separation assumption fails, a velocity-dependent coordinate transformation could split the sum- and difference-frequency responses, potentially widening the class of usable geometries."],"forward_implications":["Designers can target any computation expressible as a local nonlinear tight-binding model; the three demos cover optimization, in-memory computing, and classification.","The extracted effective model matches full nonlinear simulations for transient and steady-state response, so simulation-based design can be trusted before fabrication.","Geometric perturbations act locally and additively, so design spaces do not grow exponentially; effective-model extraction runs in linear time and in parallel per site.","Nonlinearities can be engineered independently of linear terms: local Kerr strength via support-arm shape, and cross-Kerr plus hopping via kinked versus straight coupling beams.","The approach extends perturbative metamaterials beyond narrowband operation to quasistatic multistable systems, as demonstrated by the racetrack memory."],"supporting_citations":[{"why":"Establishes the perturbative metamaterial platform for linear tight-binding models and convolutional neural layers, which this work extends to the nonlinear regime.","marker":"[6]"},{"why":"Supplies the nonlinear speech-classifier tight-binding model that is embodied in the drum-chain metamaterial.","marker":"[8]"},{"why":"Provides the parametric-oscillator Ising tight-binding model with Kerr nonlinearity used for the coherent Ising machine.","marker":"[10]"},{"why":"Shows that perturbative metamaterials realize topological tight-binding models, providing evidence for the linear framework and for long-range interaction effects.","marker":"[14]"},{"why":"Original perturbative metamaterials framework whose design requirements and effective-theory construction this paper builds on and refines.","marker":"[15]"},{"why":"Introduces the modal-derivative coordinate transformation used to capture deformation-dependent basis functions in nonlinear model reduction.","marker":"[17]"},{"why":"Provides the interpretation of high-frequency modes 'screening' nonlinear stress through modal derivatives, a central idea behind the Wannier derivative correction.","marker":"[18]"},{"why":"The Born-Oppenheimer approximation is invoked as the analogy justifying the instantaneous-response assumption for the excluded fast modes.","marker":"[44]"},{"why":"Cited as a possible route, a velocity-dependent coordinate transformation, to overcome the spectral-separation limitation of the static coordinate transformation.","marker":"[48]"}],"fun_headline_variants":["Nonlinear metamaterials now compute via tight-binding maps","New map lets metamaterials run Ising, memory, and speech tasks","Corrected mapping makes metamaterials compute nonlinear models","Metamaterials embody nonlinear computation with accurate mapping"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The fast modes that screen nonlinear stress are assumed to react instantly to changes in the slow coordinates; if the material's spectrum does not separate those time scales, the extracted nonlinear coefficients are wrong and the designed device will not compute as intended.","fun_headline_variants_meta":{"raw":{"variants":["Nonlinear metamaterials now compute via tight-binding maps","New map lets metamaterials run Ising, memory, and speech tasks","Corrected mapping makes metamaterials compute nonlinear models","Metamaterials embody nonlinear computation with accurate mapping"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000536,"raw_usage":{"total_tokens":2384,"prompt_tokens":689,"completion_tokens":1695,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":433,"completion_tokens_details":{"reasoning_tokens":1630}},"tokens_in":433,"tokens_out":1695,"duration_ms":14468,"temperature":1.0,"reasoning_tokens":1630,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T12:21:21.211725+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a geometry with a deliberately small spectral gap, extract the effective nonlinear tensor with and without the Ψ' terms, and compare full nonlinear finite-element transients: if the coordinate-transformed model fails to reproduce the full-wave response at amplitudes where nonlinearity is visible, the central claim is falsified. A simpler physical test: fabricate a small plate-and-bar structure, drive it at amplitudes where the Kerr shift is measurable, and check the predicted frequency shift against the effective coefficient extracted from Eq. 1.","supporting_citations":[{"cited_title":"Dubček, D","cited_arxiv_id":null,"evidence_quote":"Establishes the perturbative metamaterial platform for linear tight-binding models and convolutional neural layers, which this work extends to the nonlinear regime."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the nonlinear speech-classifier tight-binding model that is embodied in the drum-chain metamaterial."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the parametric-oscillator Ising tight-binding model with Kerr nonlinearity used for the coherent Ising machine."},{"cited_title":"Serra-Garcia, V","cited_arxiv_id":null,"evidence_quote":"Shows that perturbative metamaterials realize topological tight-binding models, providing evidence for the linear framework and for long-range interaction effects."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Original perturbative metamaterials framework whose design requirements and effective-theory construction this paper builds on and refines."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the modal-derivative coordinate transformation used to capture deformation-dependent basis functions in nonlinear model reduction."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the interpretation of high-frequency modes 'screening' nonlinear stress through modal derivatives, a central idea behind the Wannier derivative correction."},{"cited_title":"Gobat, V","cited_arxiv_id":null,"evidence_quote":"Cited as a possible route, a velocity-dependent coordinate transformation, to overcome the spectral-separation limitation of the static coordinate transformation."}],"review_version":1}