{"id":"a96aa44c-1c67-4b77-991f-11e93542ae65","arxiv_id":"2411.11598","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For complex dynamical systems with periodic vector fields, Fourier-based Carleman linearization gives finite-section approximations that converge exponentially to the true solution, with explicit error bounds over a time horizon.","lead":"This paper introduces a Carleman-Fourier linearization method that converts nonlinear systems with periodic vector fields into infinite-dimensional linear systems using Fourier basis functions, and proves exponential convergence of truncated approximations with explicit error bounds.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The non-analytic extension only yields a proven bound for R>e; for real periodic fields with optimal Fourier decay radius R≤e the advertised Carleman-Fourier guarantee is absent.","rationale":"I read the paper in good faith and checked the proof chain of Theorem 3.1. The derivation from (6.10)–(6.17) is internally consistent: the constant C0 matches the prefactor R in (6.17), the combination e^{D0(t-T*)N} reproduces the stated ratio (e||e^{ix0}||_∞/R)^{(e-1)N/(2e-1)}, and the tail bound uses M0/R<1/e exactly as required by (1.10). The paper also honestly states that (1.9) excludes the Kuramoto model and that Section 4 is the remedy. The weakest point is therefore not an algebraic error but the scope of the central claim: the method's contraction mechanism requires the lifted variable e^{ix} to lie below R/e, and for real states this forces R>e. That is a real restriction, not merely a technical convenience. The reader's weakest assumption identifies the same condition, and the proposed numerical test on a field with optimal R=e would show whether exponential convergence persists outside the proven regime. If it does, the paper is conservative; if it does not, the advertised general applicability to real periodic systems is materially narrower than the abstract suggests. Because the statements actually proved are internally consistent and the limitation is acknowledged, the appropriate verdict remains CONDITIONAL rather than ACCEPT or REJECT.","tokens_in":35276,"tokens_out":21288,"duration_ms":221689,"concrete_test":"Construct the real scalar periodic field g(x)=Σ_{k=1}^∞ e^{-e k} sin(kx), whose best possible Assumption 1.1 radius is R=e, form the augmented Section 4 linearization, and measure max_j |v_{j,N}(t)e^{-i x_j(t)} - 1| for N=10,20,40,80 at a fixed time t. If the errors fail to decay exponentially in N, the R>e restriction is essential; if they do decay, the restriction is only an artifact of the proof and Theorem 4.1 / Corollary 4.2 should be extendable to R≤e.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central convergence mechanism of Theorem 3.1 depends critically on the block upper-triangular structure coming from (1.9) and on Lemma 6.1, which treats w1(t)=e^{ix(t)} as a variable of norm < R/e. For real initial data, however, ||e^{ix0}||_∞=1, so this mechanism cannot start unless R>e. Section 4 removes (1.9) by augmenting the state, but it inherits the same threshold: Corollary 4.2(ii) explicitly requires R>e for real x0. Consequently, for a real periodic vector field whose Fourier coefficients have optimal decay radius R≤e — for example an analytic function whose largest strip of analyticity has width e — the paper proves no exponential convergence of its finite-section approximation. This is a genuine scope limitation of the central claim for real Kuramoto-type systems, and although the paper acknowledges it, the abstract's phrase 'periodic vector fields' and the comparison (1.19) can easily overstate what is actually established.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces Carleman-Fourier linearization for complex nonlinear dynamical systems with periodic vector fields. The method lifts the system using Fourier monomials e^{iα·x}; under the analyticity condition (1.9) the lifted matrix is block upper-triangular. Theorems 3.1 and 3.3 state explicit exponential error bounds for finite-section approximations on a finite horizon and on the entire time axis, respectively. Section 4 handles multiple fundamental frequencies by augmenting the state vector with its negative, giving Theorem 4.1 and Theorem 4.3. Section 5 presents numerical experiments for the scalar equation ẋ=a(1−be^{ix}) and for the Kuramoto model.","tokens_in":35443,"tokens_out":21558,"duration_ms":195846,"significance":"If the results are correct, the paper gives a useful extension of Carleman linearization to periodic vector fields, with explicit and computable error bounds, finite-section truncation criteria, and a natural treatment of multiple frequencies through state augmentation. The block upper-triangular observation and the global convergence result under positivity are interesting, and the numerical experiments support the qualitative claims. The main weaknesses are the heavy reliance on arguments and lemmas from the authors' earlier work [2], the omission of the proof of Theorem 2.1, a concrete incorrect constant bound in Section 4.2, and a scope limitation for real initial states when the Fourier decay radius R≤e that is not reflected in the abstract.","major_comments":[{"comment":"The asserted inequality C1 ≤ R^2/(2π e(e−1)) is false for the C1 defined in Theorem 4.1. Substituting the condition max|ℑ(ω_l x_{0,j})| < ln R − 1 into the displayed formula for C1 yields only C1 ≤ R^2/(√(2π) e(e−1)), and the stronger claimed bound fails, e.g., with R=4 and max|ℑ|=0.1 one obtains C1≈1.22 while R^2/(2π e(e−1))≈0.545. Since Corollary 4.2 and Eq. (4.19) use this simplification, the constants in those statements need to be corrected or the stronger inequality must be proved by a different argument.","section":"Section 4.2, Eq. (4.16)"},{"comment":"Theorem 2.1 is stated with an explicit error bound but its proof is omitted: the text says the argument in [2] can be followed and the details are omitted. This theorem is listed as a contribution and is used in the comparison (1.19). The paper should either provide the proof in Section 6 or state precisely which theorem of [2] implies the complex-case bound with the same constants.","section":"Section 2, Theorem 2.1"},{"comment":"Lemmas 6.2 and 6.3 are quoted from [2] without proof and are load-bearing for the proof of Theorem 3.1. Since the current paper advertises self-contained explicit error bounds, these lemmas should either be proved in the appendix or stated as cited theorems with exact references, so that a reader can verify the constants used in the main derivation.","section":"Section 6, Lemmas 6.2 and 6.3"},{"comment":"The abstract and introduction claim applicability to 'periodic vector fields' without qualification, but for real initial states the proved exponential convergence requires R>e (Corollary 4.2(ii)); for a real periodic vector field whose optimal Fourier decay radius satisfies R≤e, no exponential convergence of the finite-section approximation is established by the theorems in this paper. This is a genuine scope limitation, acknowledged only later in Section 4, and the claims in the abstract and in the comparison (1.19) should be qualified accordingly.","section":"Abstract and Section 4, Corollary 4.2(ii)"}],"minor_comments":[{"comment":"The phrase 'exponential convergence' should be qualified: at t=T*_CF the factor e^{D0tN}(e∥exp(ix0)∥∞/R)^{(e−1)N/(2e−1)} equals 1, so the bound degenerates to O(N^{-3/2}) at the endpoint. Exponential-in-N convergence is established for te for real initial states, and the analyticity condition (1.9) rules out the original Kuramoto form, which is only recovered via the doubled augmented state.\n\nWhat is good: The explicit dependence of the error on N, D0, R, and the initial state is exactly what a user needs to pick a truncation order for MPC or reachability. Theorem 3.1's rate (e||e^{ix0}||∞/R)^{(e-1)N/(2e-1)} times N^{-3/2} e^{D0 t N} is concrete. The extension to multiple fundamental frequencies in Section 4 is a genuine generalization, not a trivial re-labeling. The simulations are illustrative and the explicit scalar example (1.3) is worked out in detail, including an exact error formula (5.11).\n\nSoft spots, in proportion: First, the scope limitation is real. For real initial data, ||e^{ix0}||∞=1, so the key condition (1.10) or (1.16) forces R>e. For an analytic periodic field whose largest strip of analyticity has width e or less, the paper proves no exponential convergence for the finite-section approximation. The abstract's 'periodic vector fields' and the favorable comparison with monomial Carleman in (1.19) are stated without this caveat. The authors do flag R>e in Section 4, but the abstract overstates. That is an honest editorial issue rather than a mathematical error.\n\nSecond, the paper omits the proof of Theorem 2.1 (monomial Carleman) and quotes Lemmas 6.2 and 6.3 from the authors' own [2] without proof. For a standalone paper that is acceptable in a series, but a referee should verify those lemmas are applicable verbatim.\n\nThird, the comparison with the real-valued case in Corollary 4.2 relies on the unpublished manuscript [26]. That is fragile; the authors should either include a proof sketch or cite a published version.\n\nFinally, no code or data accompany the simulations, so the illustrative claims are not independently reproducible.\n\nNet: The core mathematical contribution—explicit exponential error bounds for a Fourier-based lift of complex, periodic, analytic vector fields—is believable and useful. The R>e restriction for real initial data is a genuine scope limitation that should be clearly stated in the abstract, but it does not invalidate the results for the class where they apply.\n\nRecommendation: Send to peer review. A serious referee should ask for full proofs of the borrowed machinery, a clarifying abstract, and either code or a more detailed simulation setup. I would cite this work (once the comparison to [26] is resolved) for its explicit bounds, but with the R>e caveat.","headline":"Real extension of Carleman-Fourier linearization with explicit exponential bounds, but the advertised scope overstates what is proven for real periodic fields with Fourier decay radius R≤e.","tokens_in":35989,"tokens_out":3373,"would_cite":true,"duration_ms":34293,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["37C10","34C25"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves explicit exponential error bounds for a Fourier-exponential version of Carleman linearization of complex nonlinear systems with periodic vector fields, over a computable time horizon.","keywords":["Carleman-Fourier linearization","periodic vector fields","finite-section approximation","exponential convergence","explicit error bounds","complex dynamical systems","Kuramoto model","multiple fundamental frequencies"],"falsifier":"For the system $\\dot{x}=i(1-e^{ix})$ with initial $x_0=iy$, compute the $N$-truncated first block from (5.9) and compare $\\max_{t\\le T}|v_{1,N}(t)e^{-ix(t)}-1|$ with the right-hand side of (3.11); a discrepancy at fixed $N$, $T$ would refute the bound. Alternatively, adding a small term $\\epsilon e^{-ix}$ to the vector field should break the block-triangular structure, so the observed error should stop shrinking exponentially in $N$, confirming that condition (1.9) is the load-bearing premise.","tokens_in":35039,"feed_emoji":"⚙️","tokens_out":8754,"duration_ms":82740,"temperature":0.7,"pith_summary":"The paper introduces Carleman-Fourier linearization: instead of monomials, it lifts a complex nonlinear system $\\dot{x}=g(t,x)$ with a periodic vector field into an infinite-dimensional linear system whose state variables are Fourier exponentials $e^{i\\alpha x}$. For vector fields whose Fourier expansion is one-sided, meaning it contains no negative frequencies, the lifted matrix is block upper triangular, and the paper proves that the first block of the $N$-th finite-section approximation converges exponentially in $N$ to $e^{ix}$ over an explicit time horizon, with fully explicit constants. It also proves global exponential convergence when the constant Fourier coefficient of the vector field has strictly positive imaginary part. Because the error bounds are explicit, they can guide truncation-length choices in control, reachability, and quantum simulation. For vector fields that violate the one-sided condition, including the Kuramoto model, the paper introduces an augmented-state construction that restores analyticity at the cost of extra assumptions.","feed_headline":"Proven exponential convergence for Fourier linearization","feed_subtitle":"A new truncation bound tells you how many Fourier modes guarantee a target accuracy.","key_machinery":"The machinery is the Fourier-state lift: new variables $w_\\alpha=e^{i\\alpha x}$ for nonzero nonnegative multi-indices $\\alpha$, grouped in blocks by total degree $k=1,\\dots,N$. Under the analyticity condition $g_\\alpha(t)=0$ for all negative multi-indices, the derivative of $w_\\alpha$ only involves $w_\\beta$ with $|\\beta|\\ge|\\alpha|$, so the infinite matrix $B(t)$ is block upper triangular with diagonal blocks $i\\operatorname{diag}(\\alpha^T g_0(t))$. The finite-section system keeps the first $N$ blocks. The explicit error bound is obtained by writing the error $u_k=v_{k,N}-w_k$ in integral form with the diagonal kernel $K_k(t,s)=\\exp(i\\int_s^t \\alpha^T g_0\\,du)$, bounding the coupling blocks by the Schur-norm estimate $\\|B_{k,l}\\|_S\\le D_0 k R^{k-l}$, and applying a discrete Gronwall-type inequality. The augmented state $[x^T,-x^T]^T$ is a second mechanism that manufactures the analyticity condition for general multi-frequency vector fields by pairing every negative index with a positive one.","core_discovery":"The central discovery is that replacing monomials $x^{\\alpha}$ by Fourier exponentials $e^{i\\alpha x}$ in Carleman's lifting scheme converts a complex dynamical system $\\dot{x}=g(t,x)$ with a one-sided periodic vector field into an infinite-dimensional linear system with a block upper-triangular matrix $B(t)$. For such systems, the first block $v_{1,N}$ of the $N$-th finite-section approximation satisfies the explicit bound $\\max_j |v_{j,N}(t)e^{-ix_j(t)}-1| \\le C_0 N^{-3/2} e^{D_0 t N} (e\\|\\exp(ix_0)\\|_\\infty/R)^{(e-1)N/(2e-1)}$ for $0\\le t\\le T^*_{CF}$; taking logarithms and writing $v_{j,N}=e^{i\\xi_{j,N}}$ yields an approximation $\\xi_{j,N}$ of the original state $x_j$ whose error is at most four times the same bound. When the constant Fourier coefficient $g_0$ has strictly positive imaginary part, the convergence extends to all $t\\ge 0$ with the rate $((D_0+\\mu_0)\\|\\exp(ix_0)\\|_2/(\\mu_0 R))^N$. For vector fields like Kuramoto's that have negative frequencies, the paper shows that lifting the augmented state $[x^T,-x^T]^T$ restores the hypothesis and proves the analogous exponential bound under the condition $R>e$ for real initial states.","pith_inferences":["The formulas suggest a quantitative notion of a usable basin in complexified phase space: shifting a real initial state upward in the imaginary direction should extend the guaranteed horizon exactly as (1.11) predicts, a prediction one could test numerically on the example $\\dot{x}=a(1-e^{ix})$.","The augmented-state doubling in Section 4 doubles the lifted dimension; for systems with additional symmetry, such as the zero-sum phases of the normalized Kuramoto model, a smaller symmetry-adapted basis might satisfy the analyticity condition with less overhead.","For quantum simulation of dissipative polynomial dynamics, the explicit $N$-dependence in Theorems 3.1 and 3.3 could be converted into a query or qubit count, though the paper only lists quantum computing as a motivation.","The comparison with real-valued systems suggests the real case has strictly better guaranteed horizons; extrapolating, a complex system whose vector field is real on the real axis should be approximated more efficiently by first separating real and imaginary parts rather than by the direct complex lift."],"forward_implications":["For systems satisfying condition (1.9) and Assumption 1.1, an order-$N$ truncation of the lifted system approximates $e^{ix(t)}$ with the explicit error bound of Theorem 3.1 on $[0,T^*_{CF}]$, so a desired tolerance directly fixes the truncation order.","Under the positive-imaginary condition (1.14) and the small-initial-state condition (1.15), the same finite-section approximation converges exponentially for all $t\\ge 0$ with rate $((D_0+\\mu_0)\\|\\exp(ix_0)\\|_2/(\\mu_0 R))^N$.","For vector fields with multiple fundamental frequencies that fail (1.9), the augmented state $[x^T,-x^T]^T$ restores the analyticity condition; with real initial states and $R>e$, exponential convergence holds on $[0,\\widetilde{T}^*_{CF}]$ (Corollary 4.2).","When $1\\le\\|x_0\\|_\\infty