{"id":"9dbb43ed-3cad-432e-a205-a0e058514112","arxiv_id":"2506.21120","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For the DNLS lattice with next-to-nearest-neighbour interactions, the exponentially small translational eigenvalue is derived, with on-site states stable and inter-site states unstable for µ > -1/4, and no localized states for µ < -1/4.","lead":"A line of tiny waveguides with next-to-nearest neighbour coupling can trap discrete light pulses, and this paper derives the exponentially small eigenvalue that decides whether each trapped pulse is stable. It matters because it gives an experimentally checkable stability law and a new Borel-Padé recipe for hidden exponential corrections.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The exponentially small eigenvalue (70) inherits its entire nonperturbative amplitude from the conformal Borel-Padé residue (50); convergence of that residue to the true Stokes constant S1 is asserted rather than demonstrated, so this is the load-bearing weak point.","rationale":"I read the paper as making two claims: the eigenvalue asymptotic (70) with stability consequences for µ>-1/4, and nonexistence for µ<-1/4. The derivation from the tail (28) through the eigenvalue calculation is internally coherent, and the Appendix A Stokes-switching calculation is a reasonable adaptation of the NN case. The step that is least secure is not the outer asymptotic matching but the numerical extraction of S1 in Section 4. The residue formula (50) is the only route to Λ(µ), and the paper provides neither a proof of Padé convergence nor a published artefact that would let a reader reproduce the numbers. Figure 4's N=300 vs N=400 comparison is a start, but Padé poles for functions with multiple branch cuts are known to be slow and non-uniform; Section 4.4 itself acknowledges the degrading accuracy. I do not treat this as an invalidation: the method is standard, the µ=0 limit is correctly reproduced, and the numerical eigenvalue comparison at the tested µ values is convincing. But because the central amplitude depends on an unverified numerical constant, the honest verdict remains conditional on releasing code/data and performing a more rigorous convergence check. My concern is the same one the reader identified, so I would not move the verdict.","tokens_in":18532,"tokens_out":11556,"duration_ms":145901,"concrete_test":"At µ = -0.2 (or another value near the -1/4 boundary), recompute S1 with arbitrary-precision arithmetic using N=800 and N=1200 coefficients, and independently extract S1 by fitting the last 100 coefficients a_ell^(0) to the two-singulant large-order form (43). If the two independent estimates differ by more than ~1% or shift with N, the residue in Eq. (50) has not converged and the eigenvalue prediction (70) is unverified in that regime. If they agree, repeat the same comparison at µ=-0.1 and µ=-0.24 and propagate the resulting Λ(µ) into Eq. (70) for a direct comparison with full numerical eigenvalues at ε=0.1, 0.05, and 0.025.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Equation (70) is proportional to sqrt(|Λ(µ)|), and Eqs. (50)-(51) obtain Λ(µ) from the residue of a diagonal Padé approximant of order 150-400 built from early coefficients a_ell^(0) of the inner algebraic series. For most µ in (-1/4,0), the A2,± branch points are closer to the origin than A1=2πi, so the untransformed Borel-Padé approximation is dominated by those singularities; the paper's own Section 4.4 states that accuracy depends on how many singularities are nearer and how much nearer they are. Convergence is then inferred by comparing N=300 with N=400 at representative values (Figure 4), and the authors note that accuracy degrades as |µ| increases. The µ=0 consistency check with the nearest-neighbour result is real supporting evidence, but it does not cover the interval relevant to Eq. (70), and no code or coefficient data are shipped. Because an error in the residue propagates linearly through Eq. (51) into Λ(µ) and then as sqrt into the eigenvalue, this numerical-convergence assumption is the single most load-bearing condition for the central claim.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper derives an exponentially small eigenvalue for the translational mode of on-site and inter-site standing waves in the DNLS equation with next-to-nearest-neighbor interactions. Using factorial-over-power exponential asymptotics, the authors identify two singulant families, \\chi_1 and \\chi_2, and show that the Stokes multiplier associated with \\chi_1 is beyond all orders even in the inner region. They compute it via a conformal Borel-Pad\\'e residue calculation, match it to the late-order constant \\Lambda(\\mu), and obtain the leading-order eigenvalue (65) and first correction (70). They also argue that no localized standing waves exist for \\mu<-1/4. The prediction is compared with direct numerical eigenvalue computations for several \\mu, including \\mu=-1/16 and \\mu=-1/30.","tokens_in":18755,"tokens_out":12674,"duration_ms":135125,"significance":"If the result holds, the paper makes a valuable methodological and physical contribution. The eigenvalue formula (70) is parameter-free, with no fitted constants; the \\mu=0 limit reproduces the known nearest-neighbour result, and the independent numerical eigenvalue computations confirm the exponential scaling and the prefactor for representative values. The Borel-Pad\\'e strategy for computing Stokes multipliers that are hidden beyond all orders in the inner expansion is a useful template for other discrete models. The main caveat, acknowledged by the authors, is the numerical nature of the Stokes constant, which is load-bearing for the quantitative prediction.","major_comments":[{"comment":"The central prefactor in the eigenvalue prediction (65)/(70) is entirely determined by the Stokes constant S1, which is computed numerically as the residue of a diagonal Pad\\'e approximant of order N=150\\u2013400. The only convergence evidence offered is a comparison of N=300 and N=400 at representative values of \\mu (Figure 4), and the paper itself states in Section 4.4 that accuracy degrades as |\\mu| increases. Because |\\lambda| \\propto |\\Lambda(\\mu)|^{1/2}, an uncontrolled error in the residue propagates directly into the main prediction. I would like to see a quantitative convergence study, for example a table of S1(\\mu;N) for several N and for \\mu spread over the interval (-1/4,0), with estimated error bars, and ideally the coefficients a_\\ell^{(0)} or the code used to generate them made available. The \\mu=0 consistency check is reassuring but does not by itself establish reliability in the interval relevant to Eq. (70).","section":"Section 4.4, Eq. (50)"},{"comment":"The claim that no localized standing waves exist for \\mu<-1/4 rests on the statement that the \\chi_2 oscillation, proportional to \\sin(\\omega(n-n0)) with 0<\\omega<\\pi, cannot be eliminated for any site offset. This conclusion assumes that the Stokes multiplier for the \\chi_2 contribution is nonzero and that its amplitude is independent of n0; neither is demonstrated in the paper for \\mu<-1/4. Since the nonexistence result is a headline claim of the abstract, please provide either a direct computation of the \\chi_2 Stokes multiplier in Regime 1 or a numerical demonstration of the absence of stationary localized solutions for at least one \\mu<-1/4.","section":"Section 3.4"},{"comment":"The formula is stated for \\mu in (-1/4,0), but as \\mu \\to -1/4, \\beta \\to 0 and the correction term \\epsilon \\pi^2(1+16\\mu)/(48\\beta^3) diverges unless \\epsilon is taken extremely small relative to \\beta^3. The paper does not discuss this non-uniformity, and Figure 4 suggests S1 varies significantly near that endpoint. Please comment on the expected range of validity in \\mu and on the fate of the asymptotic prediction as \\mu approaches -1/4; at minimum, state that \\mu must be bounded away from -1/4 for the expansion to apply.","section":"Section 5, Eq. (70)"}],"minor_comments":[{"comment":"The O(\\epsilon) term is written as \\epsilon\\pi(1+16\\mu)/(24\\beta^2\\sqrt{2}\\,\\eta), but the expansion near the singularity gives this term with \\eta^2 in the denominator; the subsequent pole-shift calculation in Section 5.2 is consistent with the \\eta^{-2} form. Please correct the displayed equation.","section":"Eq. (67)"},{"comment":"There are typographical issues in the denominator notation: the bracket [2\\pi i(\\tilde z-\\tilde z_s)2j+4] should read [2\\pi i(\\tilde z-\\tilde z_s)]^{2j+4}, and the prefactor 25/451/2 should be rendered as 2^{5/4}5^{1/2}.","section":"Eqs. (25), (26), (65), (70)"},{"comment":"The caption appears to label both panels as (a); the second mention should be (b).","section":"Figure 3 caption"},{"comment":"The display of the fourth-order central difference operator contains an apparent duplication of the term -1/12 F_n and an unusual coefficient -5/2 F_n; please verify the coefficients.","section":"Eq. (3)"},{"comment":"The phrase \"S1 decays rapidly as \\mu grows in the negative direction\" is ambiguous; it should say \"as \\mu decreases towards -1/4\" or similar.","section":"Section 4.4"}],"recommendation":"major_revision","confidential_remarks":"The paper is well-structured and the asymptotic derivation is largely coherent, but the central numerical ingredient (the Borel-Pad\\'e residue) needs stronger validation before the quantitative eigenvalue formula can be fully trusted. The authors should be encouraged to make the coefficient data or code available, since this would greatly increase the reproducibility of the method. The nonexistence claim for \\mu<-1/4 also deserves a bit more support, either analytic or numerical. These issues are fixable within the manuscript's scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This is a solid applied-asymptotics paper with one load-bearing numerical step that rests on an assertion rather than a proof. The new results are real: the first exponentially-small eigenvalue formula for the NNN DNLS model, the beta = sqrt(1+4mu) structure, the existence boundary at mu = -1/4, and the first-order correction that vanishes at mu = -1/16. The derivation follows the standard factorial-over-power machinery cleanly: the singulant equation (15), the U equation (21), the Stokes switching in Appendix A, and the eigenvalue balance in (63)-(65) are coherent and cross-check against each other. The stability conclusion (on-site stable, inter-site unstable) for -1/4 < mu < 0 is exactly what the numerics show, and the mu = -1/16 cancellation is a nice touch.\n\nThe soft spot is exactly where the stress-test note points: the Stokes constant S1, and hence Lambda(mu), comes from the conformal Borel-Pade residue (50). The paper infers convergence by comparing N=300 with N=400 at a few representative mu values, and it admits in Section 4.4 that accuracy degrades as |mu| increases because the A2,pm branch points sit closer to the origin than A1. That admission is honest but it underscores the gap: there is no error estimate, no proof of convergence for the Pade approximants to the branch point, and no shipped code or coefficient data to reproduce the residue calculation. Since the eigenvalue formula (70) scales as sqrt(|Lambda(mu)|), any error in that residue propagates directly into the central prediction. The mu=0 check against [29] is genuine supporting evidence, but it is a single point, and the interesting regime is -1/4 < mu < 0.\n\nThat said, I would not call this manufactured. The derivation is careful, the numerics in Figure 5 are convincing across several mu values, and the general Borel-Pade template for subdominant Stokes multipliers is a useful methodological contribution. The circularity burden is low: nothing is fitted to the eigenvalue data. The paper deserves a serious referee.\n\nRecommendation: send to peer review, but make acceptance conditional on the authors releasing the code and the coefficient series, and on a more systematic convergence study for the residue—for example, a range of N with error scaling, or an independent way of estimating S1 for at least one mu in the interval. If that holds up, this is publishable in SIAM J. Appl. Math. or similar.","headline":"Solid exponential-asymptotics paper with a real new formula and a real numerical-load-bearing step that is asserted rather than proven.","tokens_in":19326,"tokens_out":1753,"would_cite":true,"duration_ms":20470,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["34E05","34E20","37K60"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper derives the exponentially small translational eigenvalue that decides stability for on-site and inter-site standing waves in a discrete nonlinear Schrödinger lattice with competing next-to-nearest-neighbour coupling, and proves…","keywords":["discrete nonlinear Schrödinger equation","next-to-nearest-neighbour interactions","exponential asymptotics","Stokes constants","Borel-Padé approximation","standing waves","translational eigenvalue","Peierls-Nabarro barrier"],"falsifier":"Compute the residue defining $S_1$ for a fixed μ in (-1/4,0) using several Padé orders, such as N=150, 300, and 400, and different conformal maps, then compare the resulting $|\\lambda|$ against high-precision numerical eigenvalues of the full lattice problem across a decade of ε values; if the residue drifts with N or the scaled quantity $\\varepsilon^{5/2} e^{\\beta\\pi^2/(2\\varepsilon)} \\lambda$ does not approach the predicted constant, the central claim fails. Equivalently, a numerical search that finds a localized standing wave for μ just below -1/4 would refute the non-existence claim.","tokens_in":18335,"feed_emoji":"🌊","tokens_out":11515,"duration_ms":107746,"temperature":0.7,"pith_summary":"This paper studies stationary solitary waves in a one-dimensional discrete nonlinear Schrödinger lattice with both nearest-neighbour and next-to-nearest-neighbour coupling, focusing on competing interactions, μ < 0. It claims that for -1/4 < μ < 0 the only standing waves are site-centred and inter-site-centred pulses, and that the translational eigenvalue controlling their stability is exponentially small in the lattice spacing ε. The sign of λ² is fixed by the pinning site: imaginary for on-site states (stable) and real for inter-site states (unstable), with magnitude $|\\lambda| \\sim 2^{5/4} 5^{1/2} \\pi \\beta^2 |\\Lambda(\\mu)|^{1/2} \\varepsilon^{-5/2} e^{-\\beta\\pi^2/(2\\varepsilon)}$ up to an explicit first correction. The paper also claims that no localized standing waves exist for μ < -1/4, because the competing next-to-nearest-neighbour interactions produce an oscillatory tail that no choice of site offset can cancel. These results matter because the eigenvalue depends on a Stokes constant that ordinary matched asymptotic expansions cannot reach, and the paper computes it with conformal Borel-Padé methods.","feed_headline":"On-site DNLS solitons stay stable via an exponentially small eigenvalue","feed_subtitle":"Competing next-nearest-neighbour coupling makes the decisive eigenvalue exponentially small; Borel-Padé fixes its prefactor and sign.","key_machinery":"The central object is the Borel transform of the inner transseries solution near the complex singularities of the leading-order sech pulse, combined with a conformal map, $\\zeta = A_1 - (\\xi - A_1)^2$, that converts the logarithmic branch point at $\\zeta = A_1 = 2\\pi i$ into a simple pole. The residue of that pole, extracted from a diagonal Padé approximant built from the first N even coefficients of the algebraic series, gives the Stokes constant $S_1$ through the identity $S_1 = 2i\\Gamma(1/2) a_0^{(0)-1} \\mathrm{Res}$. Matching this inner Stokes constant to the outer late-order constant via $2\\pi i \\Lambda(\\mu) = S_1$ fixes the eigenvalue prefactor. This machinery is necessary because the χ₁,± contribution is beyond all orders even within the inner region, so it cannot be obtained by matched asymptotic expansions; the closer χ₂,± singulants must be resolved to isolate the pole at A₁.","core_discovery":"The central claim is that for μ in (-1/4,0), the translational eigenvalue λ of both on-site and inter-site standing waves is exponentially small, satisfying $|\\lambda| \\sim 2^{5/4} 5^{1/2} \\pi \\beta^2 |\\Lambda(\\mu)|^{1/2} \\varepsilon^{-5/2} e^{-\\beta\\pi^2/(2\\varepsilon)} \\left(1 - \\frac{\\varepsilon \\pi^2 (1+16\\mu)}{48\\beta^3}\\right)$ as ε → 0, where $\\beta = \\sqrt{1+4\\mu}$ and $\\Lambda(\\mu)$ is the Stokes constant fixed by a Borel-Padé residue calculation. The sign of λ² is determined by $\\cos(2\\pi(n-n_0))$: for on-site pinning, n₀ integer, λ² < 0 and the mode is imaginary and stable; for inter-site pinning, n₀ half-integer, λ² > 0 and the mode is real and unstable. The paper further establishes that for μ < -1/4 no localized standing wave exists, since the χ₂,± exponential contributions oscillate with a frequency that cannot be made commensurate with the lattice for either pinning choice. These claims are validated by numerical eigenvalue computations covering several negative μ values, including μ = -1/16, where the first correction term in ε vanishes exactly.","pith_inferences":["If the residue-to-Stokes-constant identification is accepted, the same conformal Borel-Padé procedure should compute $\\Lambda(\\mu)$ for the cooperative case μ > 0 to comparable accuracy; the paper includes one positive-μ example but leaves a systematic study to future work.","Should the non-existence claim hold, the threshold μ = -1/4 should be observable in zigzag waveguide arrays: for μ just above the threshold the lattice supports pinned pulses, while below it localized pulses should fail to form.","Because the eigenvalue formula relies on a numerical Padé residue, a future exact derivation of $\\Lambda(\\mu)$ through a closed-form transseries or exact WKB analysis would turn the numerical evidence into proof and would also delimit the ε range where the first correction remains accurate.","The same conformal Borel-Padé template likely extends to other discrete nonlinear models with competing interactions, where the stability eigenvalue is exponentially small but ordinary matching cannot reach it; each extension would require its own singulant set and conformal map."],"forward_implications":["For every -1/4 < μ < 0, the site-centred ground state has an imaginary translational eigenvalue pair and is linearly stable, while the inter-site state has a real pair and is unstable, uniformly in the small-ε regime covered by the asymptotic formula.","The Peierls-Nabarro barrier, set by the energy difference between the two families of standing waves, is exponentially small in the lattice spacing, scaling as $e^{-\\beta\\pi^2/\\varepsilon}$ with the explicit prefactor and ε-correction computed in the paper.","At μ = -1/16, the model coincides with the fourth-order finite-difference discretization of the continuous nonlinear Schrödinger equation, and the first correction term in the eigenvalue expansion vanishes exactly.","For μ < -1/4, no localized standing wave exists because the competing next-to-nearest-neighbour interactions generate an oscillatory tail whose frequency cannot be made commensurate with the lattice by any site offset, so the tail cannot be cancelled.","The conformal Borel-Padé residue method provides a general template for extracting subdominant Stokes constants in parametric asymptotic problems where the relevant exponential is hidden beyond all orders in the inner expansion."],"supporting_citations":[{"why":"gives the nearest-neighbour-only predecessor of the eigenvalue calculation, including the Stokes-switching and prefactor structure that the NNN analysis extends.","marker":"[29]"},{"why":"supplies the conformal Borel-Padé method used to convert the logarithmic branch point at ζ=2πi into a simple pole and compute its residue.","marker":"[3]"},{"why":"provides the factorial-over-power late-order framework and Stokes-line matching used for the outer asymptotic series.","marker":"[7]"},{"why":"supplies the Stokes-phenomenon and matched-asymptotic-expansion machinery adapted in Appendix A for the Stokes-switching calculation.","marker":"[32]"},{"why":"gives the large-order coefficient asymptotics connecting late-order terms to Stokes constants in the transseries.","marker":"[12]"},{"why":"is the Padé approximation reference used for the Borel-Padé continuation and residue extraction.","marker":"[18]"},{"why":"justifies treating accumulating Padé poles near a singularity as the branch cut of the Borel transform.","marker":"[36]"}],"fun_headline_variants":["Stability of DNLS solitons set by exponentially small eigenvalue","Exponentially small eigenvalue decides DNLS soliton stability","Borel-Pade reveals hidden eigenvalue in DNLS lattices","Competing next-nearest coupling makes DNLS eigenvalue tiny","Subdominant Stokes multiplier determines DNLS stability"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole eigenvalue prediction rests on the assumption that the truncated diagonal Padé approximant of the conformally mapped Borel transform has converged to the true branch point at ζ=2πi and that its residue equals the actual Stokes constant $S_1$; the paper checks agreement between two truncation orders but does not prove convergence.","fun_headline_variants_meta":{"raw":{"variants":["Stability of DNLS solitons set by exponentially small eigenvalue","Exponentially small eigenvalue decides DNLS soliton stability","Borel-Pade reveals hidden eigenvalue in DNLS lattices","Competing next-nearest coupling makes DNLS eigenvalue tiny","Subdominant Stokes multiplier determines DNLS stability"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000429,"raw_usage":{"total_tokens":2256,"prompt_tokens":1071,"completion_tokens":1185,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":687,"completion_tokens_details":{"reasoning_tokens":1103}},"tokens_in":687,"tokens_out":1185,"duration_ms":9430,"temperature":1.0,"reasoning_tokens":1103,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T22:32:40.929073+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the residue defining $S_1$ for a fixed μ in (-1/4,0) using several Padé orders, such as N=150, 300, and 400, and different conformal maps, then compare the resulting $|\\lambda|$ against high-precision numerical eigenvalues of the full lattice problem across a decade of ε values; if the residue drifts with N or the scaled quantity $\\varepsilon^{5/2} e^{\\beta\\pi^2/(2\\varepsilon)} \\lambda$ does not approach the predicted constant, the central claim fails. Equivalently, a numerical search that finds a localized standing wave for μ just below -1/4 would refute the non-existence claim.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"gives the nearest-neighbour-only predecessor of the eigenvalue calculation, including the Stokes-switching and prefactor structure that the NNN analysis extends."},{"cited_title":"Aniceto, J","cited_arxiv_id":null,"evidence_quote":"supplies the conformal Borel-Padé method used to convert the logarithmic branch point at ζ=2πi into a simple pole and compute its residue."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"provides the factorial-over-power late-order framework and Stokes-line matching used for the outer asymptotic series."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"supplies the Stokes-phenomenon and matched-asymptotic-expansion machinery adapted in Appendix A for the Stokes-switching calculation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"gives the large-order coefficient asymptotics connecting late-order terms to Stokes constants in the transseries."},{"cited_title":"Graves-Morris and G","cited_arxiv_id":null,"evidence_quote":"is the Padé approximation reference used for the Borel-Padé continuation and residue extraction."},{"cited_title":"Stahl, The convergence of Pad´ e approximants to functions with branch points , J","cited_arxiv_id":null,"evidence_quote":"justifies treating accumulating Padé poles near a singularity as the branch cut of the Borel transform."}],"review_version":1}