{"id":"050468a4-8fae-4001-afdc-32dc30513657","arxiv_id":"2411.15846","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Equivalent variational formulations of splitting integrators are constructed for Kepler, and modified-Lagrangian analysis shows their LRL-vector errors are competitive with or better than classical symplectic methods.","lead":"This paper derives variational (Lagrangian-based) integrators for the Kepler problem by splitting the potential, and proves they are equivalent to compositions of explicit symplectic methods. It then uses modified-Lagrangian analysis to show these integrators preserve the Laplace-Runge-Lenz vector with small errors, backed by numerical experiments.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 4.4's modified Lagrangian for the first-order discrete Lagrangian has an O(h) sign error; the proof of the central LRL-order claim (Theorem 4.6) rests on this formula.","rationale":"The paper's central claim is Theorem 4.6: for the first-order variational integrator (4.15) with φ[1]=φ[2], the eccentricity and LRL-angle errors are second order. The proof relies on the modified Lagrangian in Theorem 4.4. I independently expanded the discrete Lagrangian (4.15) and found that the O(h) term in Theorem 4.4 has the wrong sign: it should be -h/2(∂φ/∂x1 v1 + ∂(φ[2]-φ[1])/∂x2 v2), not +h/2(...). This is a concrete mathematical error in the paper's advertised modified-Lagrangian analysis. However, the error does not necessarily destroy the theorem: the time averages of the inner products ⟨EL(L),vA1⟩ and ⟨EL(L),vA2⟩ that drive the O(h) drift are zero for both signs when the orbit is aligned with a coordinate axis, so the second-order conclusion can survive a sign correction. The reader's weaker concern about the mismatch between the theoretical orientation (major axis near x2) and the experiments (major axis along x1) is less serious than stated, because the trigonometric averages vanish for any axis-aligned initial condition. The split used in the experiments remains unspecified, which is a reproducibility gap, but the symmetric split is the only one covered by Theorem 4.6. Overall, the verdict CONDITIONAL remains appropriate: the paper should correct the sign error and state the experimental split explicitly, but the main LRL-order result appears robust.","tokens_in":23166,"tokens_out":60266,"duration_ms":439298,"concrete_test":"Re-derive the O(h) coefficient of (4.15) by expanding around the midpoint with φ[1]=φ[2]=φ/2 and φ=-1/r, or evaluate (L_d - (1/2|v|^2+1/r))/h numerically along a smooth test trajectory with nonzero x1·v1. If the coefficient is -1/2 x1 v1/r^3 rather than +1/2 x1 v1/r^3, Theorem 4.4 is wrong. Then recompute the averages in Theorem 4.6 with the corrected sign to confirm that the second-order LRL error conclusion is unaffected.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Direct Taylor expansion of the first-order discrete Lagrangian (4.15) about the midpoint gives L_d = 1/2|v|^2 - φ(x) - h/2[φ_1 v1 + (φ[2]_2 - φ[1]_2)v2] + O(h^2). For the Kepler potential φ = -1/|x| and the symmetric split φ[1]=φ[2]=φ/2, the O(h) coefficient is -1/2 x1 v1/r^3. Theorem 4.4 instead states L1st_mod = 1/2|v|^2 - φ(x) + h/2(∂φ/∂x1 v1 + ∂(φ[2]-φ[1])/∂x2 v2) + O(h^2), i.e. +1/2 x1 v1/r^3 for this split. The sign is opposite. A numerical check with a smooth trajectory confirms the negative sign: for φ=1/2|x|^2, v=(3,4), x=(1,2), h=0.01, the computed (L_d - L)/h is about -1.5, not +1.5. The proof of Theorem 4.6 uses the positive expression to evaluate EL(L) and the LRL-vector drift; although the relevant time averages still vanish if the sign is corrected, the modified Lagrangian as stated is false, so the derivation supporting the central claim is not self-consistent.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops variational discretizations for a class of second-order systems, with a focus on the Kepler problem. It gives necessary and sufficient conditions for the existence of a Lagrangian (Section 2), constructs split-potential discrete Lagrangians and proves that the resulting variational integrators are equivalent to compositions of explicit symplectic methods, including a relativistic Kepler formulation (Section 3). In Section 4 the authors derive modified Lagrangians for the new integrators and use Noether's theorem to analyze drift of the Laplace–Runge–Lenz (LRL) vector, claiming that the first-order variational integrator VI-1 preserves eccentricity and orientation with second-order errors per period and that Störmer–Verlet has fourth-order eccentricity error. Numerical experiments in Section 5 are presented as confirming these results.","tokens_in":23462,"tokens_out":21163,"duration_ms":151769,"significance":"If correct, the paper supplies a clean bridge between split-potential discrete Lagrangians and known symplectic compositions, and a modified-Lagrangian explanation of LRL-vector preservation that goes beyond existing energy/angular-momentum bounds. The equivalence theorems (Theorems 3.3–3.7) are explicitly proved and appear sound; they are a useful, self-contained contribution. The modified-Lagrangian machinery for LRL drift is novel in this setting and, once corrected, should generalize to other superintegrable problems. However, the correctness of the modified-Lagrangian part is currently compromised by a sign error in a central theorem, so the paper cannot be accepted as is.","major_comments":[{"comment":"The O(h) term in the stated first-order modified Lagrangian has the wrong sign. Expanding the discrete Lagrangian (4.15) about x(t+h/2) gives L_1st = 1/2|v|^2 - φ(x) - (h/2)[ (∂φ/∂x1) v1 + (∂(φ[2]-φ[1])/∂x2) v2 ] + O(h^2), so the modified Lagrangian should carry a minus sign before the h/2 term, not a plus sign. The directly verifiable symplectic-Euler case (3.2) is not a counterexample because its O(h) term is ±(h/2) dφ/dt, a total derivative, so the sign does not affect the Euler–Lagrange operator; for the split potential (4.15) the sign is physically relevant. The proof of Theorem 4.6 uses the incorrect plus sign to compute EL( x1 v1/|x|^3 ) and the inner products with vA1 and vA2. With the correct minus sign the inner products change sign, although the period averages still vanish because the integrands (e.g., x1 x2/|x|^5, x1^2 x2/|x|^5) are odd in θ. Thus the final order-2 claim may survive, but Theorem 4.4 as stated is false and the derivation in Theorem 4.6 is not self-consistent.","section":"Theorem 4.4 and proof of Theorem 4.6"},{"comment":"The proof of Theorem 4.4 computes only the second-order case (4.16) and simply asserts the first-order formula with the statement 'Without loss of generality'. The first-order modified Lagrangian is a load-bearing ingredient of Theorem 4.6, so the paper must either derive it explicitly or cite a complete reference for the formula. As written, the result is unsupported.","section":"Theorem 4.4, proof"},{"comment":"The theoretical preconditions and the numerical experiments are not connected. The proof of Theorem 4.6 assumes the major axis of the Kepler orbit is O(ε)-close to the x2-axis (A1 = O(ε)) and requires the symmetric split φ[1]=φ[2]. The experiments use initial conditions (x0,v0) = (1-e, 0, 0, sqrt((1+e)/(1-e))) whose major axis lies along the x1-axis, and the split used for VI-1 and VI-2 is never stated. The paper should specify the split and either redo the period averages in the experimental orientation or justify why the orientation choice does not affect the result.","section":"Section 5 and Theorem 4.6"},{"comment":"The assertion that [⟨EL(L), vA2⟩] reduces to the displayed average involving [x1/|x|^k] is made by reference to a calculation in [24] without reproducing or even outlining the steps. Since this identity is the only support for the advertised fourth-order superconvergence of the eccentricity error, the proof should be expanded enough to be checked, or the theorem should be stated as a corollary of [24] with a precise pointer to the relevant equation.","section":"Theorem 4.7, proof"}],"minor_comments":[{"comment":"In the sentence 'it follows from Theorem 4.6 that its modified Lagrangian is given by...', the reference should be to Theorem 4.4, not Theorem 4.6.","section":"Proof of Theorem 4.6"},{"comment":"The horizontal axis of Fig. 5 is labeled 'time step', but the plotted values are the index i with h_i = 2^{-i}. The axis should be labeled 'i' or 'log2(1/h)'.","section":"Fig. 5"},{"comment":"The inner product ⟨·,·⟩ is used without definition; the paper should state explicitly that it is the standard Euclidean dot product on R^n.","section":"Proposition 2.13 and Theorem 4.5"},{"comment":"The text says the new variational integrators give 'smaller eccentricity errors' than symplectic Euler and Störmer–Verlet, but Fig. 5(a) shows the Sym-Euler and VI-1 error curves nearly identical. This claim should be quantified or softened.","section":"Section 5, eccentricity comparison"},{"comment":"The proof that L1st is of order 1 is terse; writing L1st(x(t),x(t+h),h) = L(x(t),ẋ(t)) + O(h) uniformly on compact sets would be clearer.","section":"Theorem 3.3"}],"recommendation":"major_revision","confidential_remarks":"The equivalence results in Section 3 are solid and the paper is a reasonable fit for a numerical-analysis venue. The sign error in Theorem 4.4 is significant because it is used in the proof of Theorem 4.6; though the final LRL-order claims are likely correct (the relevant averages vanish regardless of sign), the manuscript must be corrected and the first-order modified Lagrangian derived rather than asserted. The disconnectedness between the theory's coordinate assumptions and the experiments is a presentation gap that should be fixed. I do not see grounds for rejection, but the necessary corrections touch the paper's central claim, so major revision is appropriate."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague, the key thing to know: this is a solid subfield paper on variational integrators for the Kepler problem, but the central modified-Lagrangian formula in Theorem 4.4 has a sign error. I re-derived the O(h) term of the discrete Lagrangian (4.15) by midpoint expansion and got the opposite sign to the paper: L_d = 1/2|v|^2 - φ - h/2(∂φ/∂x1 v1 + ∂(φ[2]-φ[1])/∂x2 v2) + O(h^2). The stress-test note is right; the numerical check with a harmonic potential confirms it. The proof of Theorem 4.6 uses the wrong sign to compute the LRL drift. The time averages of the offending terms still vanish, so the stated second-order LRL errors may survive, but the derivation is not self-consistent and the theorem as stated is false.\n\nWhat is actually good: the split-potential discrete Lagrangians and the proof that they are exactly equivalent to compositions of explicit symplectic methods (Theorems 3.3–3.7). That is a known construction but the paper works it out carefully for both classical and relativistic Kepler. The modified-Lagrangian/Noether machinery for bounding LRL vector drift is a useful angle, and the numerical experiments are consistent with the claimed orders, at least visually.\n\nSoft spots beyond the sign error: the paper does not cite earlier work that already used split-potential Lagrangians to obtain symplectic compositions; the experiments do not state which split φ[1], φ[2] was used; the theory assumes the major axis is close to the x2-axis while the experiments start with the major axis along x1; and Theorem 4.7 claims order-4 eccentricity superconvergence for Störmer–Verlet but the proof only shows the first average vanishes, not the full order. The claimed order-4 behavior of VI-2 is also asserted without a theorem.\n\nBottom line: the paper deserves a serious referee because the methods and the LRL analysis are worth having, but it needs a revision that fixes the sign, states the split, aligns the assumptions with the experiments, and fills the proof gaps. For a reader in geometric integration, it is worth a look once those are addressed.","headline":"A competent split-potential variational integrator paper whose central modified-Lagrangian formula has a sign error; the order claims likely survive, but the theory needs correction.","tokens_in":24009,"tokens_out":4811,"would_cite":false,"duration_ms":38630,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65P10","37M15","70H05","70F05"],"pacs":[],"model":"deepseek-v4-flash","headline":"Split-potential variational integrators keep Kepler orbits' shape and orientation accurate to second order.","keywords":["inverse variational problem","variational integrator","Kepler problem","modified Lagrangian","Laplace-Runge-Lenz vector","Noether's theorem","symplectic integrator","long-term orbit preservation"],"falsifier":"Run VI-1 with the symmetric split $\\varphi^{[1]}=\\varphi^{[2]}=\\varphi/2$ on the Kepler problem with the initial conditions from Section 5, measure the one-period change in $|A|$ and $\\omega$ for step sizes $h=0.5, 0.25, \\ldots$, and plot log error versus log $h$; Theorem 4.6 predicts slopes of -2 for both. A slope of -1, or linear-in-time drift of the angle, would indicate that the modified-Lagrangian coefficient in Theorem 4.4 or the inner-product average is wrong. The same test with an asymmetric split checks whether the symmetry assumption is load-bearing.","tokens_in":22942,"feed_emoji":"🪐","tokens_out":5979,"duration_ms":49092,"temperature":0.7,"pith_summary":"This paper tries to show that variational integrators for Hamiltonian systems can be built by splitting the potential, and that for the Kepler problem these integrators preserve the hidden conserved quantity, the Laplace-Runge-Lenz (LRL) vector, well over long times. The authors first give necessary and sufficient conditions for a second-order system to come from a Lagrangian, then construct first- and second-order discrete Lagrangians whose maps are exactly compositions of explicit symplectic methods. Using the modified Lagrangian and Noether's theorem, they argue that the eccentricity error and the orientation (angle) error of the LRL vector are both of order 2 for their first-order integrator when the potential split is symmetric, matching the classical symplectic Euler method. If correct, this means the shape and orientation of Keplerian orbits can be preserved without implicit solves and at no extra computational cost compared with standard explicit symplectic integrators.","feed_headline":"New integrators preserve Kepler's hidden vector to order 2","feed_subtitle":"Splitting the potential and analyzing the modified Lagrangian shrinks eccentricity and orientation drift to match classic symplectic…","key_machinery":"The load-bearing object is the split-potential discrete Lagrangian, e.g. (4.15): $L(x_n,x_{n+1},h)=\\tfrac12|(x_{n+1}-x_n)/h|^2-\\varphi^{[1]}(x^1_{n+1},x^2_n)-\\varphi^{[2]}(x^1_{n+1},x^2_{n+1})$, whose discrete Euler-Lagrange equations define a variational integrator. The argument moves through three linked devices: the equivalence theorems (3.3)-(3.7) identifying these variational maps with compositions of explicit symplectic methods; the modified-Lagrangian expansion (4.14), which converts a variational integrator into a perturbed but still conservative Kepler problem with modified Lagrangian $L_{\\mathrm{mod}}$ in Theorem 4.4; and Noether's theorem with the variational symmetries of Proposition 2.12, used in Theorem 4.5 to express one-period changes in eccentricity $|A|$ and angle $\\omega$ as averaged inner products $\\langle\\mathrm{EL}(L), v_{A_i}\\rangle$. The LRL vector is the hidden conserved quantity that fixes the orbit's shape and orientation.","core_discovery":"The central claim is Theorem 4.6: for the new first-order discrete Lagrangian (4.15), taking the symmetric split $\\varphi^{[1]}=\\varphi^{[2]}=\\varphi/2$ makes the errors in eccentricity and in the angle of the LRL vector over one period both of order 2, the same as the symplectic Euler method, while the second-order variational integrator shows superconvergence in eccentricity (order 4, Theorem 4.7). Supporting this are two structural results: the split-potential discrete Lagrangians (3.7), (4.15), and (4.16) produce exactly the same maps as compositions of explicit symplectic (or K-symplectic, for the relativistic Kepler problem) methods, and the modified Lagrangians for the new integrators are given explicitly in Theorem 4.4. The errors follow because the perturbation term obtained from the modified Lagrangian has vanishing averaged inner products against the variational symmetries of the LRL vector, leaving only $O(h^2)$ drift.","pith_inferences":["If the symmetric split $\\varphi^{[1]}=\\varphi^{[2]}=\\varphi/2$ is what removes the first-order LRL drift, then unequal splits would likely show first-order angle errors; this is directly testable and the paper does not report it.","The paper's numerical setup places the major axis along the $x_1$-axis, while Theorem 4.5 assumes the major axis is $O(\\varepsilon)$-close to the $x_2$-axis; a direct check with the theorem's aligned coordinates would close that gap.","The same modified-Lagrangian machinery could produce order estimates for other superintegrable or nearly integrable systems, such as the relativistic Kepler problem already introduced in the paper, or perturbed Coulomb problems.","The equivalence with explicit symplectic compositions suggests that the practical advantage of VI-1 comes from aligning the potential split with the orbit geometry, rather than from a fundamentally new integrator family."],"forward_implications":["The variational integrators VI-1 and VI-2 reproduce Keplerian orbit shapes with eccentricity and orientation errors at least as small as the symplectic Euler and Störmer-Verlet methods, and VI-2 has fourth-order eccentricity accuracy.","Because the variational integrators are equivalent to compositions of explicit symplectic methods, the preservation comes without implicit solves or extra cost per step.","The modified Lagrangian shows that the numerical solution follows a nearby conservative Kepler-type problem, so energy and angular momentum stay bounded and the LRL drift is slow.","The same construction applies to the relativistic Kepler problem through K-symplectic maps, giving variational integrators for that system too.","The error analysis provides a template: for any perturbation of the Lagrangian, the one-period drift of a conserved quantity is controlled by averaged Euler-Lagrange inner products against its variational symmetry."],"supporting_citations":[{"why":"Supplies the discrete mechanics formalism: discrete Lagrangian, discrete Euler-Lagrange equations, and discrete Legendre transforms on which all integrators are built.","marker":"[14]"},{"why":"Provides the modified-Lagrangian construction and the expansion (4.14) used to derive Theorem 4.4 and the drift estimates.","marker":"[24]"},{"why":"Provides the backward error analysis and symplectic integration background used to conclude bounded energy and angular momentum and to justify the comparison methods.","marker":"[7]"},{"why":"Previous study of numerical precession in variational discretizations of the Kepler problem; supplies the angle convergence result that the paper extends.","marker":"[25]"},{"why":"Source for Kepler's laws and the LRL vector as the hidden conserved quantity determining orbit shape and orientation.","marker":"[6]"},{"why":"Noether's theorem, used to connect variational symmetries to conservation laws and to derive the drift formula in Theorem 4.5.","marker":"[19]"},{"why":"Supplies the generalized vector fields, characteristics, and variational symmetry framework used in Proposition 2.12 and Theorem 4.5.","marker":"[20]"},{"why":"Foundation of the inverse variational problem; provides the conditions for existence of Lagrangians used in Section 2.","marker":"[22]"}],"fun_headline_variants":["Variational integrators keep Kepler's LRL vector to 2nd order","New splittings preserve Laplace-Runge-Lenz vector","Symplectic splitting preserves hidden Kepler symmetry","Second-order variational methods lock in LRL vector","Kepler's hidden vector preserved by novel integrators"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The drift estimate assumes the perturbed orbit's major axis is already $O(\\varepsilon)$-close to a chosen coordinate axis and that the potential split is exactly symmetric; if either fails, the averaged inner product that removes the first-order error need not vanish.","fun_headline_variants_meta":{"raw":{"variants":["Variational integrators keep Kepler's LRL vector to 2nd order","New splittings preserve Laplace-Runge-Lenz vector","Symplectic splitting preserves hidden Kepler symmetry","Second-order variational methods lock in LRL vector","Kepler's hidden vector preserved by novel integrators"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000166,"raw_usage":{"total_tokens":1205,"prompt_tokens":848,"completion_tokens":357,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":464,"completion_tokens_details":{"reasoning_tokens":279}},"tokens_in":464,"tokens_out":357,"duration_ms":3729,"temperature":1.0,"reasoning_tokens":279,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T13:51:25.418489+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run VI-1 with the symmetric split $\\varphi^{[1]}=\\varphi^{[2]}=\\varphi/2$ on the Kepler problem with the initial conditions from Section 5, measure the one-period change in $|A|$ and $\\omega$ for step sizes $h=0.5, 0.25, \\ldots$, and plot log error versus log $h$; Theorem 4.6 predicts slopes of -2 for both. A slope of -1, or linear-in-time drift of the angle, would indicate that the modified-Lagrangian coefficient in Theorem 4.4 or the inner-product average is wrong. The same test with an asymmetric split checks whether the symmetry assumption is load-bearing.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the discrete mechanics formalism: discrete Lagrangian, discrete Euler-Lagrange equations, and discrete Legendre transforms on which all integrators are built."},{"cited_title":"V ermeeren","cited_arxiv_id":null,"evidence_quote":"Provides the modified-Lagrangian construction and the expansion (4.14) used to derive Theorem 4.4 and the drift estimates."},{"cited_title":"Hairer, C","cited_arxiv_id":null,"evidence_quote":"Provides the backward error analysis and symplectic integration background used to conclude bounded energy and angular momentum and to justify the comparison methods."},{"cited_title":"V ermeeren","cited_arxiv_id":null,"evidence_quote":"Previous study of numerical precession in variational discretizations of the Kepler problem; supplies the angle convergence result that the paper extends."},{"cited_title":"Goldstein, C.P","cited_arxiv_id":null,"evidence_quote":"Source for Kepler's laws and the LRL vector as the hidden conserved quantity determining orbit shape and orientation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Noether's theorem, used to connect variational symmetries to conservation laws and to derive the drift formula in Theorem 4.5."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the generalized vector fields, characteristics, and variational symmetry framework used in Proposition 2.12 and Theorem 4.5."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Foundation of the inverse variational problem; provides the conditions for existence of Lagrangians used in Section 2."}],"review_version":1}