{"id":"6f91ff62-883f-46eb-8829-262a0ea328aa","arxiv_id":"2607.05873","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A regularized function inserted into Carleman linearization, derived from a Möbius conformal map, removes the long-time divergence for logistic, KPP-Fisher, and phase-field models and supports an LCU quantum implementation.","lead":"This paper shows why Carleman linearization, a way to turn nonlinear equations into linear ones for quantum simulation, blows up at late times, and fixes it with a conformal mapping trick. The fix makes long-time simulation of several reaction-diffusion equations possible and the authors sketch a quantum circuit to do it.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Non-integer eigenvalue generalization of f_{M,c}(k,t) is not the derived conformal-map truncation; L=5 and phase-field validations rely on an unjustified spectral filter.","rationale":"The reader's weakest assumption correctly flags the lack of proof that a single c moves all relevant singularities and the reliance on a degeneracy-breaking perturbation. My concern is more specific and cuts deeper: the formula defining the regularization is derived only for integer eigenvalues, but the paper uses it for non-integer spectra (L=5 KPP) and half-integer phase-field parameters, where the claimed equivalence between the truncated binomial expansion and the incomplete beta function does not hold. This is not merely a missing proof; the identity is false for non-integer shape parameters. The logistic-equation analysis remains rigorous, so the paper should not be rejected outright, but the PDE/phase-field claims are unsupported unless the authors either derive the correct conformal-map truncation for non-integer k or demonstrate that the incomplete-beta filter is an equally valid regularization, e.g., by convergence tests on the closed-form cubic logistic solution. Hence the verdict stays CONDITIONAL, with a new, more specific condition.","tokens_in":18129,"tokens_out":19948,"duration_ms":197728,"concrete_test":"Directly test the identity behind the non-integer extension. (1) For k=1/2 and k=3/2, evaluate the reciprocal-gamma sum in Eq. (22) for M=1,2 and compare with I_{1−ω}(k,M−k+1); they are not equal, which would confirm that the L=5/phase-field f is not the conformal-map truncation. (2) Run the cubic logistic simulation of Sec. V with the proposed f(k/2) for K=8,16,32,64, M=K (and M=2K), comparing to the closed-form solution (33). If the error does not decrease systematically with K/M, the numerical validation of the non-integer extension fails; if it does, the concern is mitigated empirically.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The paper's central regularization f_{M,c}(k,t) is derived for positive integer k: Eq. (22) is the truncated binomial expansion of the Möbius-mapped eigenfunction ζ^k, and the identification with I_{1−ω}(k,M−k+1) in Eq. (23) is the standard incomplete-beta identity for integer shape parameters. Sections IV B and V then extend this to non-integer eigenvalues and to k/2 in the phase-field case, claiming the 'gamma-function definition ... applies without essential modification.' For non-integer k this identity fails: the binomial expansion of ζ^k in ω contains powers ω^{k+n} with non-integer exponents, so no finite integer-M truncation of the form in Eq. (22) exists, and I_{1−ω}(k,M−k+1) is not equal to the truncated conformal-map series. A minimal check (k=1/2, M=1) gives Eq. (22)/(reciprocal-gamma sum) = (1/π)ω^{1/2}(1−ω)^{1/2}, while I_{1−ω}(1/2,1/2) = (2/π)arcsin√(1−ω). Thus the L=5 KPP and phase-field numerics implement a spectral filter that is not the analytic-continuation truncation derived in Sec. III. The degeneracy-breaking perturbation ε_i (Eqs. (30)–(31)) used in those validations is a further independent weakness, but the non-integer extension is the more load-bearing gap because it undermines the claimed generality of the method.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper identifies the long-time divergence of Carleman linearization with the evaluation of the lifted spectral series y(t)=Σ a_k e^{kt} y^{(k)} outside its finite radius of convergence, and proposes to fix it by a Möbius conformal map ζ=cω/(1−ω) together with a regularized function f_{M,c}(k,t) inserted into the spectral sum. For the logistic equation the paper derives the convergence condition c<(1−x_0)/x_0, gives an explicit mapped-series truncation error (Eqs. (59)–(62)), and supports it with numerical phase diagrams. The method is then extended to KPP–Fisher equations and phase-field models, with a different map η=cω'/(1−ω') for cubic nonlinearities, and a quantum LCU implementation is presented with an error and complexity analysis.","tokens_in":18623,"tokens_out":6207,"duration_ms":59857,"significance":"The logistic-equation analysis is a genuine contribution: the divergence is traced to the geometric-series radius, the fix is derived from the conformal map rather than fitted, and the truncation error is explicit and checked numerically. This part is solid and publishable. However, the claimed generality to non-integer spectra and to the PDE examples is currently not established. The non-integer version of the regularized function is an unjustified spectral filter, and the degeneracy-breaking perturbation changes the equation under study. If these gaps are closed, the method would be an important step toward long-time stable Carleman simulation. The paper does provide exact-solution-based checks and reproducible numerical phase diagrams, which are strengths.","major_comments":[{"comment":"The non-integer extension is invalid. Eq. (22) is a finite sum over integer powers ω^m obtained from the binomial expansion of (ω/(1−ω))^k only when k is a positive integer. For non-integer k the expansion contains fractional powers ω^{k+n}; it cannot be truncated at integer M, and I_{1−ω}(k,M−k+1) is not the truncated conformal-map series because the incomplete-beta identity used in Eq. (23) assumes integer shape parameters. A minimal check for k=1/2, M=1 gives the truncated sum (1/π)ω^{1/2}(1−ω)^{1/2}, while I_{1−ω}(1/2,1/2)=(2/π)arcsin√(1−ω). The L=5 KPP and phase-field results therefore implement a spectral filter that is not the derived analytic continuation.","section":"§IV B, Eq. (22)/(23); §V Eq. (39)"},{"comment":"The perturbation ε_i=s(K+1)(i−L+1) changes the differential equation being solved. The reference solutions in Figs. 9, 11, 14, and 16 are obtained by Euler's method on the perturbed system, not on the original KPP–Fisher or phase-field equations. Since no bound on the perturbation error is provided and the paper explicitly defers this to future work, the numerical agreement cannot validate the regularized Carleman method for the original PDEs. This is load-bearing because the perturbation is introduced specifically to make eigenvector computation tractable.","section":"§IV A, Eqs. (30)–(31)"},{"comment":"The quantum resource estimate is incomplete because the LCU normalization Λ is never bounded. The final complexity O(ΛK^2 log K / ε_meas) cannot be assessed without a bound on Λ in terms of K, c, M, and n, or at least a numerical study. In addition, the block-encoding cost T_BE=O(K log K) hides the inverse-precision factors in the value-oracle rotations, and the condition number κ(P_K) in Eq. (64) is not estimated. The claim of a concrete resource estimate is therefore stronger than what is demonstrated.","section":"§VI, Eqs. (66)–(71)"}],"minor_comments":[{"comment":"Several typos: 'calcualtion' (§I), 'spare' should be 'sparse' (§VI), 'In this case, In this case' (§V), and references [25] and [33] are duplicated.","section":"General"},{"comment":"The text says the L=5 results are shown in 'Figure 10 and Figure 16', but the L=5 comparison is Fig. 11; Fig. 16 is the phase-field comparison.","section":"§IV B"},{"comment":"'exponential (int) divergence' contains a stray '(int)'.","section":"§II B"},{"comment":"The statement that the sum in Eq. (22) begins at m=1 instead of k because Γ(m−k+1) hits a pole when m<k is only correct for integer k; for non-integer k no such pole occurs at integer m, which is another indication that the non-integer generalization needs separate derivation.","section":"§III A"}],"recommendation":"major_revision","confidential_remarks":"The logistic core is solid and could be published, but the PDE and quantum-resource parts currently rest on an invalid non-integer spectral filter and a perturbed validation equation. I recommend major revision rather than reject because the central idea is salvageable: either prove the non-integer generalization or restrict the claims to integer spectra, and supply a perturbation-error bound or validate on a non-degenerate system."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The logistic-equation core is rigorous and the regularization trick is genuinely new, but the advertised extension to non-integer spectra doesn't follow from the derivation.\n\nThe paper's main positive contribution is the diagnosis: for the logistic equation, the Carleman expansion diverges because the series in ζ=e^t has a finite radius, and replacing ζ^k by ζ^k f_{M,c}(k,t) with f the regularized incomplete beta function does fix the divergence. The convergence condition c<(1−x0)/x0 is derived, the mapped-series truncation error is explicit (Eqs. 59-62), and the LCU implementation for the logistic equation is concrete and checkable. That section is worth a serious referee and is citable.\n\nThe soft spot is the generality claim. Sections IV.B and V extend f_{M,c}(k,t) to non-integer eigenvalues and to k/2 in the phase-field model, asserting the gamma-function definition applies 'without essential modification.' It doesn't. Equation (22) is a truncated binomial expansion that only makes sense for positive integer k; the identification with I_{1−ω}(k,M−k+1) is the standard incomplete-beta identity for integer shape parameters. For k=1/2 that identity fails: the sum in Eq. (22) is not a truncation of the conformal-mapped series, and the two functions are not equal. So the L=5 KPP-Fisher and phase-field numerics are implementing an ad hoc spectral filter, not the analytic-continuation method derived in Section III. That is load-bearing for the paper's central advertising claim, and it is not flagged as a limitation.\n\nTwo secondary issues: the degeneracy-breaking perturbation ε_i (Eq. 31) alters the equation under study, so agreement with Euler's method on the perturbed system is weaker evidence than the paper frames it as. And the LCU normalization Λ appears in the complexity bound but no bound on Λ is given, leaving the resource estimate incomplete.\n\nBottom line: the logistic result is solid; the extension is a numerical teaser with a real mathematical gap. I'd send it to peer review with the expectation that the authors either prove the non-integer case or restate the claims. A reader working on Carleman or quantum nonlinear simulation will want to see it, mainly for the logistic analysis.","headline":"The logistic-equation core is rigorous and the regularization trick is genuinely new, but the advertised extension to non-integer spectra doesn't follow from the derivation.","tokens_in":19067,"tokens_out":3435,"would_cite":true,"duration_ms":34297,"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":"This paper claims that Carleman linearization's late-time divergence comes from evaluating its Laurent expansion outside the radius of convergence, and that inserting a Möbius-map regularized function into the spectral solution fixes the di","keywords":["Carleman linearization","analytic continuation","conformal mapping","regularized incomplete beta function","logistic equation","KPP–Fisher equation","phase-field model","LCU quantum algorithm"],"falsifier":"Solve the L=3 KPP–Fisher system with perturbation strength s decreasing to 10^{-8}, 10^{-12} and compare the regularized Carleman output against a high-resolution Euler solution of the unperturbed equation; if agreement does not improve, or if the regularized series diverges at a (c,u2) point the phase diagram classifies as convergent, the paper's regularization claim for that system is falsified.","tokens_in":18059,"feed_emoji":"📈","tokens_out":6898,"duration_ms":67890,"temperature":0.7,"pith_summary":"The paper sets out to explain and remove the exponential divergence that plagues Carleman linearization, a standard method that lifts nonlinear differential equations into infinite linear systems. Working through the logistic equation, it shows the divergence appears because the Carleman spectral series is a Laurent expansion in ζ=e^t that is only valid inside a finite disk, even though the true solution stays analytic. The proposed fix is analytic continuation through the conformal map ζ=cω/(1−ω), which sends the positive time axis into the unit disk, and a regularized function f_{M,c}(k,t) inserted into the eigen-decomposed solution; for integer spectra this function is a regularized incomplete beta function. The method is validated on the logistic equation, KPP–Fisher equations under periodic boundary conditions, and phase-field models, with stability diagrams showing convergent and divergent regions in the (c, initial-condition) plane. A quantum implementation via LCU block encoding is given for the logistic equation with explicit error and complexity estimates.","feed_headline":"Möbius map fixes Carleman's late-time divergence","feed_subtitle":"A regularized eigenvalue function keeps logistic, KPP-Fisher, and phase-field simulations stable for all time.","key_machinery":"The load-bearing object is the regularized function f_{M,c}(k,t), defined for positive integer eigenvalues as the regularized incomplete beta function I_{1−ω(t)}(k, M−k+1), with ω(t)=e^t/(e^t+c), and set to 1 for nonpositive k. It is inserted into the Carleman spectral solution as e^{kt}→e^{kt}f_{M,c}(k,t), damping the growing modes so that the conformally mapped series converges for all t≥0. The conformal map ζ=cω/(1−ω) (or η=cω'/(1−ω') with η=e^{2t} for cubic nonlinearities) is what moves the singularities outside the unit disk; the convergence condition c<(1−x0)/x0 for the logistic equation is derived explicitly and matched against the numerical phase diagrams. The LCU implementation uses","core_discovery":"The central claim is that Carleman divergence is not a numerical instability but a convergence-domain problem: the lifted solution y(t)=Σ a_k e^{kt} y^{(k)} matches the true solution only for |e^t| below a finite radius, so along t→∞ the series must eventually diverge. Inserting e^{kt} f_{M,c}(k,t) in place of e^{kt}, with f_{M,c} built from the Möbius map and equal to I_{1−ω(t)}(k, M−k+1) for positive integer k, turns the divergent series into a uniformly convergent one on t≥0, provided the map parameter c satisfies a condition such as c < (1−x0)/x0 for the logistic equation. The same recipe extends to quadratic reaction-diffusion systems (KPP–Fisher, using the same map and keeping negative","pith_inferences":["The regularized function acts as a convergence factor that depends only on the spectrum of the Carleman matrix, not on the detailed coefficients; if that holds more broadly, the same f_{M,c} could be applied to other Carleman-based schemes (e.g., lattice-Boltzmann or fluid simulations) that exhibit the same late-time divergence, with the map parameter chosen from a stability scan rather than from ","Because the KPP–Fisher and phase-field validations use a degeneracy-breaking perturbation ε_i that modifies the equation itself, an unstated testable consequence is that the regularized solution should converge to the unperturbed Euler reference as s→0; without such a continuity check, the validation strictly applies to the perturbed system.","The quantum complexity statement depends on Λ, the LCU normalization, and no bound on Λ is derived; a natural extension is to compute how Λ grows with K, c, and M, since the advertised O(Λ K^2 log K / ε) scaling is only useful if Λ does not grow too fast.","The stability diagrams suggest a practical recipe for equations without exact solutions: scan c and the initial condition to locate the monotone-convergent region, then set M≈K; this empirical map selection could be automated and would be a direct test of the method on new systems."],"forward_implications":["Long-time Carleman simulations of quadratic reaction-diffusion systems can be stabilized by choosing c below the derived bound; the phase diagrams in (c, initial-condition) identify monotone, oscillatory, and divergent regimes.","The correction is a pure eigenvalue transformation of the Carleman matrix, so it composes with quantum linear-algebra tools: for the logistic equation the polynomial degree n=K−1 suffices in exact arithmetic, and the total Toffoli cost is O(Λ K^2 log K / ε_meas).","For the logistic equation the mapped-series truncation error is exactly bounded by (1−r)^M with r=cx0/(1−x0), giving M≥log(1/ε)/−log(1−r); choosing K=M removes additional Carleman truncation error.","For the PDE examples, the same regularization function applies to non-integer spectra (L=5 KPP–Fisher) and cubic phase-field models, but the error analysis explicitly leaves spatial discretization, perturbation, and conditioning errors as future work."],"fun_headline_variants":["Möbius map stabilizes Carleman for all time","Analytical continuation resolves Carleman divergence","Carleman linearization no longer diverges late-time","Regularized eigenvalues fix Carleman's long-time instability"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"For the PDE systems, the load-bearing, unproven premise is that one real scale parameter c in the Möbius map (or the η=e^{2t} variant) moves all relevant singularities outside the unit disk, and that the tiny degeneracy-breaking perturbation ε_i=s(K+1)(i−L+1), s≤10^{-4}, leaves the equation close enough to the original for numerical agreement with Euler to validate the method; the phase diagrams themselves show divergent regions, and the quantum cost estimate additionally ass","fun_headline_variants_meta":{"raw":{"variants":["Möbius map stabilizes Carleman for all time","Analytical continuation resolves Carleman divergence","Carleman linearization no longer diverges late-time","Regularized eigenvalues fix Carleman's long-time instability"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000193,"raw_usage":{"total_tokens":1184,"prompt_tokens":737,"completion_tokens":447,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":481,"completion_tokens_details":{"reasoning_tokens":381}},"tokens_in":481,"tokens_out":447,"duration_ms":4537,"temperature":1.0,"reasoning_tokens":381,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T08:20:51.795304+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Solve the L=3 KPP–Fisher system with perturbation strength s decreasing to 10^{-8}, 10^{-12} and compare the regularized Carleman output against a high-resolution Euler solution of the unperturbed equation; if agreement does not improve, or if the regularized series diverges at a (c,u2) point the phase diagram classifies as convergent, the paper's regularization claim for that system is falsified.","supporting_citations":[],"review_version":2}