{"id":"9434482e-6a84-4393-b446-2c7c5ed0d4e6","arxiv_id":"1908.05590","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The Dulac map near a normally hyperbolic manifold of saddle singularities is shown to admit an asymptotic expansion built from the Ecalle-Roussarie compensator, generalizing the planar saddle result.","lead":"This paper proves new normal forms for vector fields near manifolds of normally hyperbolic singularities, then uses them to compute asymptotic expansions of transition maps, called Dulac maps, near these objects. The key result is that these high-dimensional Dulac maps inherit the same power-and-log asymptotic structure as the classical planar saddle case.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The main theorems assume, without proof, that the formal variation expansion (3.7) is a genuine asymptotic expansion uniformly in t = -ln x0; the paper supplies no remainder estimate, so the central claim is not established.","rationale":"The reader's weakest assumption identifies exactly the gap that I consider decisive: the variation expansion (3.7) is used as an asymptotic expansion without any remainder control in the doubly coupled limit U_y0, U_z0 -> 0 and t -> infinity. This is not a cosmetic omission, because the expansion is formal in the initial conditions while the asymptotic variable is x0 = exp(-t); the coefficients themselves grow in t, so the ordering of terms by powers of U_y0 and U_z0 need not survive the substitution. The normal form results in Section 2 are independent and appear substantially supported, so the paper has real value even if the Dulac-map theorems require additional arguments. I also note that the displayed prefactors in Theorem 3.11 look interchanged relative to the integration in (3.4) and the proof method, but I classify that as a fixable formula-level error rather than the structural gap described above. A conditional acceptance requiring a remainder lemma is therefore the appropriate outcome.","tokens_in":22123,"tokens_out":15853,"duration_ms":163960,"concrete_test":"Take the scalar model obtained from (3.6) by setting z0 = 0 and k = 1: dot x = x, dot U_y = -alpha1 U_y + a(u) U_y^{q1+1}, with q1 = 1 and alpha1 a nonzero constant. Solve this Bernoulli equation exactly, write the exact U_y(U_y0, -ln x0), and compare it with the N-th Taylor polynomial in U_y0 after substituting U_y0 = x0^{p1/q1} y0. Check whether the remainder is o(x0^{N p1}) as x0 -> 0 uniformly in u0 for N = 1, 2, 3. This directly tests whether the formal variation expansion is asymptotic in the required coupled limit; the paper supplies no such estimate even in this minimal case.","verdict_should_be":"UNCHANGED","load_bearing_attack":"To make Theorems 3.8 and 3.11 true, one needs more than that the coefficients in (3.7) belong to the ring \\bar R_{alpha1,beta1}. One needs that, after the substitution U_y0 = x0^{p1/q1} y0, U_z0 = x0^{p2/q2} z0 and t = -ln x0, the N-th Taylor remainder of the flow map in the initial conditions is smaller than the N-th retained term uniformly in u0 and x0 -> 0. For fixed t, analytic dependence on (U_y0,U_z0) near 0 is standard, but here the time grows at exactly the rate at which the initial conditions shrink; the radius of convergence of the Taylor series in the initial data may depend on t, and the coefficients contain factors e^{(...)t} and polynomials in t. The proof of Theorem 3.8 says only that 'An asymptotic expansion for U_y is given by the variation of U_y in (3.7)' and then treats the formal series as asymptotic. No Gronwall bound, no remainder estimate, and no appeal to normal hyperbolicity (which is not available in the rescaled U-coordinates, where alpha1(0) = beta1(0) = 0) appears anywhere. The same gap is inherited by Proposition 3.10, whose proof is explicitly omitted, and hence by Theorem 3.11. This is the load-bearing point: without such an estimate, (3.14) and (3.19) are formal computations, not theorems.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops normal form theory for manifolds of normally hyperbolic singularities and applies it to the transition map (Dulac map) between cross-sections transverse to the centre-stable and centre-unstable manifolds near a normally hyperbolic manifold of saddle singularities of codimension three. Section 2 gives a formal normal form theorem with explicit resonance conditions, a C^k normal form theorem via the Belitskii--Samovol theorem, and a discussion of analyticity. Section 3 specializes to saddles with one-dimensional stable or unstable manifold and claims two asymptotic theorems: Theorem 3.8 for the case α(0)/β(0) not an integer, and Theorem 3.11 for the resonant case α(0)/β(0) ∈ N. In both theorems, the Dulac map is asserted to be asymptotic to a series whose coefficients lie in the ring generated by the Ecalle--Roussarie compensator ω(α,x). The paper also proposes an explicit variational method for computing the coefficients.","tokens_in":22377,"tokens_out":11014,"duration_ms":100668,"significance":"If the main asymptotic theorems are correct, the paper would provide a useful general framework for transitions near manifolds of normally hyperbolic saddles, extending the planar Dulac-map theory of Roussarie and the three-dimensional results of Bonckaert--Naudot and Roussarie--Rousseau to a broader geometric setting. The normal-form part is a genuine contribution: the formal normal form with resonant monomials (Theorem 2.11) and the C^k conjugacy statement (Corollary 2.15) are plausible and give a clear algebraic mechanism. However, the central asymptotic results are not established as theorems because the key step, the passage from the formal variation expansion (3.7) to a genuine asymptotic expansion, is asserted without the necessary remainder estimates. The statement of Theorem 3.11 also contains an apparent inconsistency in the leading prefactors. These issues are load-bearing, so the significance of the paper remains contingent on a substantial revision.","major_comments":[{"comment":"The central step is the assertion that the formal variation expansion (3.7) is an asymptotic expansion of the true flow uniformly as t = -ln x0 tends to infinity. The proof of Theorem 3.8 simply says 'An asymptotic expansion for U_y is given by the variation of U_y in (3.7)' and then substitutes t = -ln x0. This is not justified. For fixed t, the flow is analytic in (U_y0,U_z0) near 0, but here the initial values scale like x0^{p_i/q_i} while t → ∞, so the radius of convergence in the initial data may shrink with t. The coefficients contain factors e^{(...)t} = x0^{-(...)} and polynomials in t, so no conclusion about remainders follows from the formal Taylor expansion. One must show that the N-th Taylor remainder of the flow map is smaller than the N-th retained term uniformly in u0 and x0 → 0. No Gronwall estimate, remainder bound, or appeal to normal hyperbolicity is supplied; notably, normal hyperbolicity is not available in the U-coordinates, where α1(0)=β1(0)=0. Without such an estimate, equations (3.14) and (3.19) are formal computations rather than proved asymptotic statements.","section":"Section 3.1, eq. (3.7), proof of Theorem 3.8"},{"comment":"The prefactors in the displayed asymptotic series are not consistent with the linearized computation. Since U_y = x^{mp/q} y and U_z = x^{p/q} z, the leading-order terms from (3.16) give y1 ∼ x0^{α(u0)} y0 and z1 ∼ x0^{β(u0)} z0, yet (3.19) states y1 ∼ x0^{β(u0)} y0 and z1 ∼ x0^{α(u0)} z0. The proof also writes U_y^{(-1,0)} = α_{-1,0} Ω(α1 - mβ1, t), which is missing the factor e^{-α1 t}; with that factor and the substitution U_{z0}^m = x0^{mp/q} z0^m, the α_{-1,0} term carries the prefactor x0^{α(u0)}. As stated, the theorem contradicts the very computation on which its proof relies.","section":"Section 3.2, Theorem 3.11, eq. (3.19)"},{"comment":"The proof of Proposition 3.10 is omitted with the remark that it is 'almost identical' to Proposition 3.7. Since Proposition 3.7 itself rests on the unjustified asymptotic use of the formal expansion (3.7), the omitted proof is load-bearing: an induction can at best produce formal coefficients, not a genuine asymptotic expansion. The paper should either supply the omitted argument or state explicitly that Theorem 3.11 is conditional on the missing remainder estimates.","section":"Proposition 3.10 and proof of Theorem 3.11"},{"comment":"The asymptotic theorems require the normal form (3.1) or (3.2) with K = ∞. Corollary 2.15 only gives finite K(k) conjugacy in general, and analyticity of the normal form is not proved. The sentence 'Finite K is easily recovered by truncating summations at the relevant order' does not address the need to control the remainder of the transition map after truncation. The theorems should either be stated for each fixed truncation order with an explicit remainder bound, or a proof of the infinite-order normal form should be supplied.","section":"Section 3, transition from C^k to infinite normal form"}],"minor_comments":[{"comment":"The definition 'U_z = z^{p/q} y' appears to be a typo for 'U_z = x^{p/q} z'; the later Section 3.2 uses the latter definition, and the two are not equivalent.","section":"Proposition 3.1, case ii)"},{"comment":"The notation ar R^d_{α1,β1} in Lemma 3.5 is used before a degree d has been defined; the degree should be introduced explicitly in Definition 3.4.","section":"Section 3.1, Definition 3.4 and Lemma 3.5"},{"comment":"The paper never states precisely what 'D is asymptotic to the series' means, in particular whether the asymptotic is meant as x0 → 0+ for each fixed u0 or uniformly over a compact set of u0; this should be clarified.","section":"Section 3, general framework"},{"comment":"The exponents in terms such as (x0^{mp} y0^q)^{n1/q} involve fractional powers for n1 not divisible by q; this should be written with integer exponents and the domain of definition specified.","section":"Equation (3.19)"},{"comment":"The last paragraph says the Mourtada property 'should also be evident'; this is not proved and should be labelled as a conjecture or removed from the main claims.","section":"End of Section 3"}],"recommendation":"major_revision","confidential_remarks":"The paper contains a promising normal-form framework, but the two main asymptotic theorems are not proved as written. The most serious problem is the missing remainder estimates for the variation expansion (3.7); the apparent interchange of α and β in Theorem 3.11 also needs correction. If the author can supply the missing estimates and fix the statement, the paper could be suitable for publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. The normal-form half of this paper is a real contribution: the grading by degree in the normal variables only, the modified homological operator \\hat L_d, and the resonance conditions in Lemma 2.6 and Theorem 2.11 give a clean and genuinely new formal normal form theory for manifolds of singularities. The C^k corollary via Belitskii-Samovol is a reasonable application. I don't see a hole in that part.\n\nThe Dulac-map half is not proved as written. The proof of Theorems 3.8 and 3.11 substitutes the formal variation expansion (3.7) into the transition map and calls it asymptotic. No remainder estimate is supplied. For fixed t, the expansion is the Taylor series of the flow in the initial conditions, but here t = -ln x0 grows at the same rate that the initial conditions shrink, and the coefficients contain exponentials and t-polynomials. Uniform asymptoticity is a genuine issue, and the proof doesn't address it. There is no Gronwall bound, no use of normal hyperbolicity (which wouldn't help directly in U-coordinates), nothing. The same gap passes into Proposition 3.10, whose proof is omitted, and into Theorem 3.11. As printed, those are formal computations, not theorems.\n\nThere are also correctness issues in the statement of Theorem 3.11. The prefactors are swapped: y1 is written with x0^{β(u0)} and z1 with x0^{α(u0)}. Substituting U_y = x^{mp/q} y, U_z = x^{p/q} z into the leading variational coefficients gives y1 a prefactor x0^{α(u0)} and z1 x0^{β(u0)}, and the explicit leading-order case (3.4) confirms. This looks like a typo, but it needs fixing. Smaller typos: Lemma 2.6 repeats ∂x_i for the second resonance condition instead of ∂u_i, and Proposition 3.1 case ii defines U_z as z^{p/q} y where x^{p/q} z is intended.\n\nNet: this paper deserves a serious referee, but not because the asymptotic theorems are established. The normal form work is solid and citable, and the Dulac map results are plausible and would be important if the missing estimates can be supplied. I would not take the Section 3 theorems on faith. The reader who works on normal forms, or on asymptotics near normally hyperbolic saddles, should read it, but carefully. My recommendation: send to peer review, with the expectation of major revision and, ideally, a referee who will ask for the remainder estimate.","headline":"Normal-form half is a solid citable contribution; the Dulac-map half has a load-bearing gap in the transition from formal variation to asymptotic expansion, plus a prefactor typo in Theorem 3.11.","tokens_in":22958,"tokens_out":7831,"would_cite":true,"duration_ms":65891,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["37D10","37G05","34C20"],"pacs":[],"model":"deepseek-v4-flash","headline":"Manifolds of normally hyperbolic saddles have Dulac maps asymptotic to explicit compensator series.","keywords":["normally hyperbolic singularities","formal normal forms","C^k normal forms","Dulac map","Ecalle-Roussarie compensator","asymptotic expansion","transition map","resonance conditions"],"falsifier":"Take a specific vector field in one of the normal forms (3.6) or (3.16), compute the transition $D$ numerically for $x_0 = 10^{-2}, 10^{-3}, \\ldots$, subtract the first two terms of the corresponding series (3.14) or (3.19), and test whether the residual divided by the first omitted monomial tends to zero; if it does not, the claimed asymptotic series is not asymptotic.","tokens_in":21860,"feed_emoji":"","tokens_out":10151,"duration_ms":93189,"temperature":0.7,"pith_summary":"This paper tries to establish that a manifold of normally hyperbolic singularities admits the same kind of normal-form and transition-map theory that was previously available only for a single saddle singularity. It proves formal and $C^k$ normal forms near such a manifold, with surviving nonlinear terms selected by resonance conditions, and then uses those normal forms to show that the Dulac map between transverse sections is asymptotic, as $x_0 \\to 0^+$, to an explicit series whose coefficients are polynomials in the Ecalle-Roussarie compensator $\\omega(\\alpha_1,x_0)$. If the theorems are right, this is the first general asymptotic description of transitions near manifolds of normally hyperbolic saddles, and the coefficients can be computed order by order from linear equations.","feed_headline":"Dulac maps near normally hyperbolic saddles have explicit asymptotics","feed_subtitle":"The transition map is asymptotic to a computable series in Ecalle-Roussarie compensators.","key_machinery":"The machinery has two coupled pieces. The first is a normal-form theory built around a modified homological operator $\\hat L_d=\\tilde L_d\\oplus X_0$ on the modules $\\mathcal{C}\\mathcal{H}_d$ of vector fields homogeneous in the normal variables $x$ with coefficients that are functions of the centre variables $u$; the cokernel of $\\hat L_d$ identifies the resonant monomials that survive, and Weierstrass-Mather division chooses unique representatives. The second is the variation expansion (3.7)/(3.17) of the flow in the initial transverse coordinates $U_{y0}, U_{z0}$, whose coefficients satisfy linear variational equations and lie in the ring $\\bar R_{\\alpha_1,\\beta_1}$ generated by $\\Omega(\\pm\\alpha_1,t)$, $\\Omega(\\pm\\beta_1,t)$, and $t$. Substituting $t=-\\ln x_0$ turns this into the ring $\\bar R^\\omega_{\\alpha_1,\\beta_1}$ of polynomials in the Ecalle-Roussarie compensator $\\omega(\\alpha_1,x_0)=(x_0^{-\\alpha_1}-1)/\\alpha_1$, with $\\omega(0,x_0)=-\\ln x_0$, and it is exactly this object that carries the asymptotic structure of the Dulac map.","core_discovery":"The central discovery is Theorems 3.8 and 3.11: for a codimension-three manifold $N$ of normally hyperbolic saddle singularities whose stable or unstable normal direction is one-dimensional, the Dulac map $D$ between sections transverse to the centre-stable and centre-unstable manifolds is asymptotic to the series (3.14) or (3.19). Every coefficient is an element of the ring $\\bar R^\\omega_{\\alpha_1,\\beta_1}$, meaning a polynomial in $\\omega(\\pm\\alpha_1,x_0)$, $\\omega(\\pm\\beta_1,x_0)$, and $\\ln x_0$ over functions smooth in the centre variable $u_0$ and rational in the eigenvalue deviations $\\alpha_1,\\beta_1$. In the resonant case $\\alpha(0)/\\beta(0)\\in \\mathbb{N}$ an extra leading term $\\alpha_{-1,0}(u_0)\\,z_0^m\\,\\omega(\\gamma_1,x_0)$ appears, with $\\gamma_1=\\alpha-m\\beta$. When $\\alpha$ and $\\beta$ are constant on $N$, every coefficient is a polynomial in $\\ln x_0$. The coefficients are produced recursively from linear variational equations and are polynomial in the normal-form coefficients up to the same order.","pith_inferences":["An immediate unseen consequence is that the same algebraic mechanism should produce Dulac-map asymptotics in higher codimensions, with the ring generated by $\\omega(\\alpha_i,x_0)$ for each normal eigenvalue; a direct test would be to compute the first two coefficients for a codimension-four normal form and compare with numerical integration.","The paper states that the Mourtada-type flatness of the higher-order terms should follow, but proving it would require explicit bounds on the remainders of the variation expansion, which are not supplied.","For applied systems, the theorem implies that near-manifold transitions in problems such as collision regularisation and min-time control should contain $x_0^{\\alpha_1}$ and $\\ln x_0$ terms rather than pure power laws, which is a checkable prediction in concrete models."],"forward_implications":["The Dulac map near a normally hyperbolic manifold of saddle singularities has the same asymptotic shape as the planar Dulac map, so quantitative questions about transitions can be approached with the same compensator calculus.","Order-by-order computation of the series reduces to solving linear ordinary differential equations for variation coefficients; no integration of the nonlinear flow is needed.","In the resonant case $\\alpha(0)/\\beta(0)\\in\\mathbb{N}$, the extra term $\\alpha_{-1,0}(u_0)z_0^m\\omega(\\gamma_1,x_0)$ must be included, because omitting it changes even the leading asymptotics of the $y$-component.","When $\\alpha$ and $\\beta$ are constant on the manifold, every coefficient is a polynomial in $\\ln x_0$, so the series can be evaluated after computing finitely many polynomial coefficients.","The $C^k$ normal-form corollary supplies a finite $K(k)$ such that any $C^\\infty$ system is $C^k$-conjugate to the truncated normal form, allowing finite-order expansions with controlled smoothness."],"supporting_citations":[{"why":"Supplies the leading-order resonant term for the three-dimensional saddle Dulac map that Theorem 3.11 extends to a full asymptotic series.","marker":"[BN01]"},{"why":"Defines the Ecalle-Roussarie compensator and records the planar Dulac-map asymptotics that the paper generalizes.","marker":"[Rou98]"},{"why":"Treats the almost-planar non-resonant case and points to the extra compensator term that becomes necessary when α(0)/β(0) is an integer.","marker":"[RR96]"},{"why":"Introduces Mourtada-type functions, the flatness class in which the higher-order coefficients are expected to sit.","marker":"[Mou90]"},{"why":"Gives the normal-hyperbolicity framework and the straightening of centre-stable and centre-unstable manifolds used in the pre-normal form.","marker":"[Wig94]"},{"why":"Supplies the Belitskii-Samovol theorem used to pass from truncated formal normal forms to C^k conjugacies.","marker":"[IL98]"},{"why":"Provides the Weierstrass/Mather division theorem and the Borel extension lemma used to choose normal-form representatives and smooth transformations.","marker":"[GG73]"},{"why":"Sets out the algebraic normal-form framework in which the modified cohomological equation and the resonance conditions are formulated.","marker":"[Mur06]"}],"fun_headline_variants":["Explicit Dulac map asymptotics near normally hyperbolic saddles","Computable Dulac series for transitions near saddle manifolds","Resonant and nonresonant Dulac maps have explicit series","New normal form theorems yield explicit Dulac asymptotics","Codim-3 saddle manifolds: Dulac maps asymptotic to computable series"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The asymptotic series rests on the unproved assertion that the formal variation expansion (3.7) of the flow is asymptotic to the true transition map uniformly as $x_0 \\to 0^+$; if the Taylor remainder grows faster than the retained terms, the series fails.","fun_headline_variants_meta":{"raw":{"variants":["Explicit Dulac map asymptotics near normally hyperbolic saddles","Computable Dulac series for transitions near saddle manifolds","Resonant and nonresonant Dulac maps have explicit series","New normal form theorems yield explicit Dulac asymptotics","Codim-3 saddle manifolds: Dulac maps asymptotic to computable series"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00033,"raw_usage":{"total_tokens":1817,"prompt_tokens":898,"completion_tokens":919,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":514,"completion_tokens_details":{"reasoning_tokens":830}},"tokens_in":514,"tokens_out":919,"duration_ms":8659,"temperature":1.0,"reasoning_tokens":830,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:09:38.676986+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a specific vector field in one of the normal forms (3.6) or (3.16), compute the transition $D$ numerically for $x_0 = 10^{-2}, 10^{-3}, \\ldots$, subtract the first two terms of the corresponding series (3.14) or (3.19), and test whether the residual divided by the first omitted monomial tends to zero; if it does not, the claimed asymptotic series is not asymptotic.","supporting_citations":[],"review_version":1}