{"id":"078885bf-326e-4f8f-9e1f-a255b7be010f","arxiv_id":"1908.08630","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Small solutions of discrete NLS with two eigenvalues relax, under a Fermi Golden Rule condition, to one nonlinear bound state plus a dispersive wave.","lead":"This paper proves that small solutions of a discrete nonlinear Schrödinger equation with two bound states eventually settle into a single nonlinear bound state plus a radiating wave, provided a Fermi Golden Rule nondegeneracy condition holds. It also shows the higher excited state is unstable and derives a generalized mass equipartition law.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 1.9 overstates selection: the proof shows at most one discrete mode survives (ρ1ρ2=0) but never rules out ρ1=ρ2=0, so the asserted ρ_+>0 and 'exactly one bound state' are not established.","rationale":"The most load-bearing defect is internal to the stated theorem: the proof establishes the dichotomy ρ1ρ2 = 0 but never the positivity of the surviving amplitude. Since the theorem's headline conclusion is that every small solution selects exactly one nonlinear bound state with positive asymptotic amplitude, this gap directly weakens the central claim as written. The Fermi Golden Rule concern identified by the reader is real but is an explicit hypothesis; even granting FGR, the positivity gap remains. The literal assumption mismatch in Theorem 1.9 (stating (1.5) instead of (1.7)) is a separate statement-level error that makes the theorem as printed vacuous, reinforcing the need for a corrected, conditional acceptance. The core mechanism — damping via the FGR term, bootstrap estimates for η, and local decay — is credible and the technical estimates appear consistent, so I do not recommend rejection; a revised statement with ρ_+ ≥ 0 (or an added non-degeneracy condition on the initial bound-state mass) and with (1.7) restored would address the main concern.","tokens_in":31491,"tokens_out":23711,"duration_ms":225402,"concrete_test":"Check whether the proof of Theorem 1.9 anywhere forces max(ρ1,ρ2) > 0; in particular, read Proposition 4.14 and the bootstrap in Section 4.2 to confirm that ρ1 = ρ2 = 0 is allowed. Then take a concrete two-eigenvalue potential satisfying (1.7) and FGR (e.g., N0 = 4), choose a small initial datum u0 in P_c l2 with zero bound-state projection, and numerically integrate the normal-form system (4.2)-(4.4) to measure the asymptotic discrete-mode amplitudes |z1(∞)| and |z2(∞)|. If both tend to 0, Theorem 1.9's ρ_+ > 0 is false as stated; if one tends to a positive limit, the proof still needs a lower-bound lemma that is currently absent.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Proposition 4.14 proves only ρ1,ρ2 ≥ 0 and ρ1ρ2 = 0; Theorem 1.9 concludes ρ_+ > 0. The proof of Theorem 1.9 says 'one of j=1,2 has to converge to 0' and then asserts the existence of ρ_+ > 0, but ρ1 = ρ2 = 0 is compatible with every estimate in Section 4: the identities (4.9)-(4.10) give integrability of |z1^{N0-1} z2^{N0}|^2, which is automatic if either mode decays, and the L1-derivative argument only forces convergence of |zj|. In the coordinates of Lemma 2.4, an initial datum u0 ∈ P_c l2 with small norm has z(0) = 0, and the normal-form transformations preserve z = 0 by (3.1) and (3.6). Nothing in the bootstrap rules out the solution scattering with both mode amplitudes tending to 0, so the central 'exactly one nonlinear bound state' claim is not a consequence of the argument; at most 'at most one' is proved. Independently, Theorem 1.9 is literally stated under (1.5), which contradicts the resonant condition (1.7) and makes FGR vacuous; the intended hypothesis (1.7) should be restored. Both issues are fixable, but they affect the theorem as stated.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies small l^2 solutions of a discrete nonlinear Schrödinger equation on Z with a potential whose linear part has exactly two eigenvalues. Under the Fermi Golden Rule nondegeneracy Γ>0 and a resonance condition ω_{N0}∈(0,4), it claims that every sufficiently small solution decomposes into a nonlinear bound state plus a dispersive wave, i.e. one of the two bound states is selected while the other is damped; it further claims orbital instability of the excited state and a generalized equipartition property. The proof uses a nonlinear coordinate decomposition, Darboux and Birkhoff normal forms, Strichartz and Kato-smoothing estimates, and a bootstrap argument in which the Fermi Golden Rule provides L^2-integrability of the resonant product |z_1^{N0-1} z_2^{N0}|^2.","tokens_in":31776,"tokens_out":13363,"duration_ms":133364,"significance":"If correct, the result is a valuable discrete analogue of the selection/relaxation theorems for NLS with two eigenvalues, and it contrasts with the author's earlier quasi-periodic existence result [29] by showing that, for generic potentials in the FGR sense, resonant interactions destroy quasi-periodicity. The paper gives a concrete spectral criterion for Γ>0 via the distorted Fourier transform, and the main bootstrap is presented in considerable detail. However, the main theorem as stated is not fully supported by the proof: the asserted positivity of the selected bound-state amplitude is not established, and the theorem is stated under the wrong hypothesis.","major_comments":[{"comment":"Theorem 1.9 is stated under assumption (1.5), namely ω_n ∉ [0,4] for all n, but the resonant condition (1.7), which is used throughout the paper to define N0 and the Fermi Golden Rule at ω_N0, is incompatible with (1.5). Under (1.5) the damping mechanism in the proof is not active and the statement is either vacuous or refers to the wrong setting. The intended hypothesis is (1.7); the theorem statement must be corrected.","section":"Section 1, Theorem 1.9"},{"comment":"The proof establishes that |z_j(t)| converge to limits ρ_j ≥ 0 and that ρ_1ρ_2 = 0, but it does not establish that one of the limits is strictly positive. The asserted existence of ρ_+ > 0 does not follow. For an initial datum with z(0)=0 (for instance u_0 ∈ P_c l^2), the normal-form estimates (3.1) and (3.6) preserve z=0, and all estimates in Section 4 are compatible with ρ_1=ρ_2=0; the integrability of |z_1^{N0-1}z_2^{N0}| only forces the product of the two limits to vanish. Thus the central claim that 'exactly one nonlinear bound state is selected' is not a consequence of the argument; at most 'at most one mode survives' is proved. The theorem should be weakened to ρ_+ ≥ 0 with an additional condition ensuring positivity (such as a quantitative lower bound on N_0|z_1(0)|^2 + (N_0-1)|z_2(0)|^2), or an additional argument must rule out ρ_1=ρ_2=0.","section":"Section 5, proof of Theorem 1.9; Proposition 4.14"}],"minor_comments":[{"comment":"In the first sentence of the proof, 'as t → 0' should read 'as t → ∞'.","section":"Section 5, proof of Theorem 1.9"},{"comment":"The phrase 'since |z_1^{N0-1} z_2^{N0}| is integrable, one of j=1,2 has to converge to 0' should be formulated as 'at least one of the limits ρ_1, ρ_2 is zero', because both limits may vanish.","section":"Section 5, proof of Theorem 1.9"},{"comment":"The proof of Theorem 1.15 is omitted entirely ('We omit the proof'). Since this is one of the main results, the paper should at least provide a detailed sketch of the adaptation of [11, Theorem 1.4] to the present discrete setting, indicating which estimates are used and how the discrete Laplacian changes the argument.","section":"Section 5, Theorem 1.15"},{"comment":"The text says 'we always assume H is generic in the sense of Lemma 5.3 of [14]' without stating the genericity condition. For self-containedness, the required assumptions should be stated explicitly, even if the proof is omitted.","section":"Section 4.1"}],"recommendation":"major_revision","confidential_remarks":"The core bootstrap and normal-form machinery are substantial and appear sound, and the two issues above are local in nature: a hypothesis fix and a strengthening or weakening of the conclusion. I therefore recommend major revision rather than rejection. The positivity gap is the more serious of the two, as it affects the main theorem and also the phrasing of Theorem 1.16."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nWorth knowing: this paper handles the resonant case for discrete NLS with two eigenvalues — the case where a resonance with the continuous spectrum destroys the quasi-periodic family from the author's earlier nonresonant paper [29]. The main results are new and largely believable: small solutions converge to a single nonlinear bound state plus radiation (Theorem 1.9), excited states are orbitally unstable (Theorem 1.15), and the Gang–Weinstein equipartition law is extended to arbitrary order N0 (Theorem 1.16). The normal-form/bootstrap machinery is substantial, the Fermi Golden Rule mechanism is standard, and the paper is open about the N0=2,3 degeneracy.\n\nWhere I'd push back: the statement of Theorem 1.9 literally assumes (1.5), which is the nonresonance condition and makes N0 and the FGR term disappear. It should be (1.7). That is a typo-level fix, but it matters because the theorem is the centerpiece.\n\nMore substantial: the proof does not establish ρ_+ > 0. Proposition 4.14 shows only ρ1ρ2=0, i.e. at most one discrete mode survives; it does not exclude both mode amplitudes going to zero. The proof of Theorem 1.9 simply asserts ρ_+ > 0, and that assertion is not a consequence of the displayed estimates. The mass-partition formula in Theorem 1.16 suggests positivity when the initial data has a discrete component, but the theorem as stated covers all small l^2 data, including purely continuous data, where one would expect scattering with no bound state. The fix is to add a nondegeneracy condition on the initial data or weaken the conclusion to ρ_+ ≥ 0 and phrase the selection result accordingly. This is a real gap in the theorem as written, but it does not collapse the method — the mechanism genuinely kills one mode.\n\nOther concerns are minor: Theorem 1.15's proof is delegated to [11], which is acceptable but should be spelled out more; Theorem 1.16's proof is sketchy; and FGR is a nondegeneracy condition not established for generic potentials, but that is standard in this literature and the paper says so. The citation pattern is honest: [29] and [11] are used as tools, not to hide assumptions.\n\nWho this is for: people working on asymptotic stability of discrete NLS or soliton selection. It deserves a serious referee; the core proof has substance and the issues are fixable. I'd send it out, expecting a revision that fixes the theorem statement.","headline":"Resonant two-eigenvalue DNLS stabilization is real work with a fixable gap: the theorem overclaims positivity of the selected mode and states the wrong resonance hypothesis.","tokens_in":32338,"tokens_out":5494,"would_cite":true,"duration_ms":57542,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35Q55","35B35","35P25"],"pacs":[],"model":"deepseek-v4-flash","headline":"For a discrete NLS with two eigenvalue modes, every sufficiently small solution eventually selects one nonlinear bound state and radiates the other away.","keywords":["discrete nonlinear Schrödinger equation","asymptotic stability","Fermi Golden Rule","nonlinear bound states","equipartition of mass","normal form","radiation damping","two eigenvalues"],"falsifier":"Choose a two-eigenvalue potential satisfying the resonance condition (1.7) for which the explicit formula (1.17) gives $\\Gamma>0$, and numerically solve (1.1) from arbitrarily small initial data with nonzero overlap with both eigenfunctions; if both $|z_1(t)|$ and $|z_2(t)|$ stay bounded away from zero for all $t$, or the solution does not converge to a single bound state plus a free wave, the central claim is false.","tokens_in":31266,"feed_emoji":"🌊","tokens_out":12588,"duration_ms":114936,"temperature":0.7,"pith_summary":"The paper proves a selection rule for small solutions of a discrete nonlinear Schrödinger equation on the integer lattice when the linear part has exactly two eigenvalues. If a nonlinearity-generated frequency $\\omega_n=e_1+n(e_2-e_1)$ falls inside the continuous spectrum $[0,4]$ and a Fermi Golden Rule coefficient is positive, every sufficiently small $\\ell^2$ solution eventually decomposes into one nonlinear bound state plus a free dispersive wave; the other discrete mode is radiated away. This is the resonant counterpart of an earlier no-resonance result in which stable two-mode quasi-periodic solutions exist. The paper also proves that the excited bound state is orbitally unstable and derives a parameter-free mass equipartition: the lost excited mass is split between the surviving bound state and radiation in a ratio fixed by the resonance index $N_0$.","feed_headline":"Small discrete NLS solutions pick one bound state and radiate the rest","feed_subtitle":"Under a nonzero Fermi-Golden-Rule rate, every small solution settles into one bound state plus a dispersive wave.","key_machinery":"The proof decomposes $u=\\varphi_1(z_1)+\\varphi_2(z_2)+\\eta$ with $\\eta$ in the continuous-spectrum subspace, then applies two near-identity changes of variables: a symplectic diagonalization and a normal-form transformation that erases nonresonant monomials. The surviving resonant interaction is the term $\\langle \\bar z_1^{N_0-1}z_2^{N_0}G,\\eta\\rangle$ in the effective Hamiltonian, where $G$ is a Schwartz function built from the eigenfunctions and the nonlinearity. Writing the radiation field approximately as $Y=-\\bar z_1^{N_0-1}z_2^{N_0}R_H^+(\\omega_{N_0})G$ and substituting into the amplitude equations yields the damping identities $\\frac12\\frac{d}{dt}|z_1|^2=(N_0-1)\\Gamma|z_1|^{2(N_0-1)}|z_2|^{2N_0}+\\text{error}$ and $\\frac12\\frac{d}{dt}|z_2|^2=-N_0\\Gamma|z_1|^{2(N_0-1)}|z_2|^{2N_0}+\\text{error}$, with $\\Gamma=-\\operatorname{Im}\\langle G,R_H^+(\\omega_{N_0})G\\rangle$. Positivity of $\\Gamma$, the Fermi Golden Rule, makes the resonant product $|z_1|^{2(N_0-1)}|z_2|^{2N_0}$ integrable in time, forcing one amplitude to vanish at infinity; space-time and local-decay estimates then upgrade this to scattering of the remainder.","core_discovery":"On its own terms, the central claim is Theorem 1.9: under the resonance condition (1.7) and the Fermi Golden Rule condition (FGR), there is $\\delta>0$ such that any solution with $\\|u(0)\\|_{\\ell^2}<\\delta$ satisfies $u(t)=\\varphi_j(z(t))+e^{it\\Delta}\\eta_++o(1)$ in $\\ell^2$ and $|z(t)|\\to\\rho_+$, with $j\\in\\{1,2\\}$, $\\eta_+\\in\\ell^2$, and $\\rho_++\\|\\eta_+\\|_{\\ell^2}\\lesssim\\|u(0)\\|_{\\ell^2}$. The limiting object $\\varphi_j(z(t))$ is one of the two nonlinear bound states that bifurcate from the linear eigenfunctions, and the other mode's amplitude decays to zero. Under the same assumptions the excited bound state $\\varphi_2$ is orbitally unstable, and the final mass obeys Theorem 1.16: when the ground bound state is selected, $\\rho_+=|(u(0),\\varphi_1)|^2+\\frac{N_0-1}{N_0}|(u(0),\\varphi_2)|^2+O(\\varepsilon^4)$, while excited-state selection swaps the ratio to $\\frac{N_0}{N_0-1}$.","pith_inferences":["Inference: ground-state selection should be generic among small data, since the excited state is unstable, but the paper does not identify the exceptional set of initial data that converges to $\\varphi_2$.","Inference: for three or more eigenvalues, the same normal-form and radiation mechanism should generically eliminate multi-mode quasi-periodic solutions, because the frequency set generated by the eigenvalues is generically dense in $\\mathbb{R}$; the paper states this expectation as a conjecture.","Inference: because the equipartition ratio is independent of the potential and coupling constants, a numerical experiment on a two-level discrete lattice with $N_0=2$ should see exactly half of the excited mass radiated and half absorbed; that would be a direct test of the mechanism.","Inference: for $N_0=2,3$ with pure cubic nonlinearity the leading coefficient $G$ vanishes, so a direct check of the theorem needs the generalized Fermi Golden Rule described in Remark 1.14; simulations should include quintic or higher-order terms to observe the predicted selection."],"forward_implications":["If the theorem is correct, the long-time dynamics of small solutions is one-mode: any resonance hitting the continuous spectrum destroys two-mode quasi-periodic bound states.","Every small solution acquires a well-defined asymptotic bound state and radiation field, so the nonlinearity acts as a selection mechanism rather than a small perturbation that preserves all linear modes.","The excited bound state is orbitally unstable, so small perturbations around it leave and the solution settles elsewhere, instead of persisting as a nearby excited state.","The asymptotic mass ratio is fixed by $N_0$ alone: ground-state selection splits the excited mass as $(N_0-1):1$ between the surviving bound state and radiation, and excited-state selection gives the reciprocal ratio.","Together with the no-resonance theory, the result completes the two-eigenvalue picture: without resonance, stable quasi-periodic solutions exist; with resonance and (FGR), they do not."],"supporting_citations":[{"why":"Supplies the discrete Schrödinger dispersive and smoothing estimates used throughout the bootstrap.","marker":"[14]"},{"why":"Provides the one-eigenvalue asymptotic stability result that this paper extends to two eigenvalues.","marker":"[25]"},{"why":"Constructs the nonlinear bound states and gives the no-resonance quasi-periodic stability result that this paper contrasts with.","marker":"[29]"},{"why":"Supplies the normal-form and Fermi-Golden-Rule stabilization scheme for continuous NLS adapted here to the discrete setting.","marker":"[11]"},{"why":"Contains the $N_0=2$ half-half equipartition property that Theorem 1.16 generalizes.","marker":"[22]"},{"why":"Gives the dispersive and local decay estimates for the 1D discrete Schrödinger operator used in the proof of Lemma 4.7.","marker":"[37]"},{"why":"Establishes scattering of small solutions in the absence of eigenvalues, the base case extended by this result.","marker":"[43]"},{"why":"Provides the distorted Fourier transform used in the appendix to express $\\Gamma$ explicitly via $\\hat G(\\pm\\xi_{N_0})$.","marker":"[6]"}],"fun_headline_variants":["Small NLS solutions: only one nonlinear bound state survives","Under Fermi resonance, small NLS solutions pick a bound state","Two eigenvalues, one bound state: NLS small solutions stabilize","NLS small solutions select one bound state, disperse the rest"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole result rests on the Fermi Golden Rule coefficient $\\Gamma=-\\operatorname{Im}\\langle G,R_H^+(\\omega_{N_0})G\\rangle$ being strictly positive; if $\\Gamma=0$, the damping terms in the amplitude equations vanish and the proof gives no decay of the resonant mode product, so selection fails.","fun_headline_variants_meta":{"raw":{"variants":["Small NLS solutions: only one nonlinear bound state survives","Under Fermi resonance, small NLS solutions pick a bound state","Two eigenvalues, one bound state: NLS small solutions stabilize","NLS small solutions select one bound state, disperse the rest"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000748,"raw_usage":{"total_tokens":3312,"prompt_tokens":906,"completion_tokens":2406,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":522,"completion_tokens_details":{"reasoning_tokens":2336}},"tokens_in":522,"tokens_out":2406,"duration_ms":17732,"temperature":1.0,"reasoning_tokens":2336,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:35:03.882165+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Choose a two-eigenvalue potential satisfying the resonance condition (1.7) for which the explicit formula (1.17) gives $\\Gamma>0$, and numerically solve (1.1) from arbitrarily small initial data with nonzero overlap with both eigenfunctions; if both $|z_1(t)|$ and $|z_2(t)|$ stay bounded away from zero for all $t$, or the solution does not converge to a single bound state plus a free wave, the central claim is false.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the discrete Schrödinger dispersive and smoothing estimates used throughout the bootstrap."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the one-eigenvalue asymptotic stability result that this paper extends to two eigenvalues."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Constructs the nonlinear bound states and gives the no-resonance quasi-periodic stability result that this paper contrasts with."},{"cited_title":"PDE 8 (2015), no","cited_arxiv_id":null,"evidence_quote":"Supplies the normal-form and Fermi-Golden-Rule stabilization scheme for continuous NLS adapted here to the discrete setting."},{"cited_title":"Weinstein, Equipartition of mass in nonlinear Schr¨ odinger/Gross-Pitaevskiiequations, Appl","cited_arxiv_id":null,"evidence_quote":"Contains the $N_0=2$ half-half equipartition property that Theorem 1.16 generalizes."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the dispersive and local decay estimates for the 1D discrete Schrödinger operator used in the proof of Lemma 4.7."},{"cited_title":"Kevrekidis, Asymptotic behaviour of small solutions for the discrete nonlinear Schr¨ odingerand Klein-Gordon equations, Nonlinearity 18 (2005), no","cited_arxiv_id":null,"evidence_quote":"Establishes scattering of small solutions in the absence of eigenvalues, the base case extended by this result."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the distorted Fourier transform used in the appendix to express $\\Gamma$ explicitly via $\\hat G(\\pm\\xi_{N_0})$."}],"review_version":1}