Pith. sign in

REVIEW 4 major objections 5 minor 29 references

Variational Discretizations for Hamiltonian Systems

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Split-potential variational integrators keep Kepler orbits' shape and orientation accurate to second order.

desk verdict 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. read the letter →

arxiv 2411.15846 v1 pith:V6XLO52V submitted 2024-11-24 math.NA cs.NA

Yihan Shen , Yajuan Sun This is my paper · ORCID
classification math.NAcs.NA MSC 65P1037M1570H0570F05
keywords inversevariationalproblemintegratorKeplermodifiedLagrangianLaplace-Runge-LenzvectorNoether'stheoremsymplecticlong-termorbitpreservation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

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.

What carries the argument

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.

What would settle it

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.

Watch

Extended reading notes

Core claim

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.

Load-bearing premise

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.

Editorial extensions

If this is right

  • 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.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • 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.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

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.

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 (4)
  1. [Theorem 4.4 and proof of Theorem 4.6] 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.
  2. [Theorem 4.4, proof] 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.
  3. [Section 5 and Theorem 4.6] 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.
  4. [Theorem 4.7, proof] 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.
minor comments (5)
  1. [Proof of Theorem 4.6] 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.
  2. [Fig. 5] 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)'.
  3. [Proposition 2.13 and Theorem 4.5] The inner product ⟨·,·⟩ is used without definition; the paper should state explicitly that it is the standard Euclidean dot product on R^n.
  4. [Section 5, eccentricity comparison] 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.
  5. [Theorem 3.3] 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.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the discrete Lagrangians, equivalence theorems, and modified-Lagrangian analysis are derived rather than fitted or assumed.

full rationale

The paper's main claims are built from explicit derivations. The discrete Lagrangians (4.15) and (4.16) are constructed from a potential split, and Theorems 3.3–3.7 prove that the resulting variational integrators coincide with compositions of explicit symplectic maps; this is a proved equivalence, not a presupposed conclusion. The modified Lagrangians in Theorem 4.4 are obtained by expanding the discrete Lagrangian and applying the standard modified-Lagrangian construction from [24], not by fitting parameters to the errors they later predict. The LRL-vector drift estimates in Theorems 4.5–4.7 use Noether's theorem and Kepler's laws to compute time averages independently of the integrator output, so the stated order results do not reduce to the inputs by construction. There is no self-citation chain, no imported uniqueness theorem, and no fitted quantity renamed as a prediction. The apparent sign discrepancy in the O(h) term of the first-order modified Lagrangian, if real, would be a correctness defect rather than circular reasoning, and the mismatch between the theoretical x2-axis alignment assumption and the x1-axis-aligned initial conditions in Section 5 is an experimental-precondition gap, not circularity.

Assumptions & free parameters 1 free parameters · 5 assumptions · 0 invented entities

The central analysis rests on the classical Helmholtz variational conditions, the Euler-Maclaurin expansion used to pass from discrete to modified Lagrangians, and the standard theory of modified equations for variational integrators. The split fraction α is a design choice, set to 1/2 in the theorem but unspecified in the experiments; it is not fitted to data. No new physical entities are introduced.

free parameters (1)
  • potential split fraction α (φ[1]=αφ, φ[2]=(1-α)φ) = α = 1/2 in Theorem 4.6; unspecified in Section 5
    The discrete Lagrangian (4.15) and the LRL error order depend on this choice. The paper does not report the value used in the simulations, which blocks exact reproduction.
assumptions (5)
  • standard math Helmholtz conditions (Theorem 2.3): N[x]=0 is variational iff the variational derivative is self-adjoint, with L given by (2.1).
    Classical result of the inverse problem of calculus of variations, cited to [2,22,23].
  • standard math Euler-Maclaurin expansion (4.13) converting the discrete action sum into an integral of a corrected Lagrangian.
    Used in Theorem 4.4 to obtain modified Lagrangians; a standard expansion with Bernoulli numbers.
  • domain assumption Existence of formal modified equations and modified Lagrangians for variational integrators.
    Backward error analysis for variational integrators, following [24] and [27]; the formal series may diverge.
  • domain assumption Kepler's three laws (Proposition 2.8) for the exact orbit are used to compute period averages.
    The averages [x1x2/|x|^5] etc. are evaluated using the exact elliptic orbit, assuming the unperturbed Kepler solution.
  • standard math Noether's theorem connecting variational symmetries to conservation laws.
    Invoked in Proposition 2.12 and Theorem 4.5; standard result from [19,20].

how reviews work

0 comments
Cite this review

Pith. "Pith review of Variational Discretizations for Hamiltonian Systems." pith.science (2026). https://pith.science/paper/V6XLO52V

@misc{pith2026241115846,
  author       = {Pith},
  title        = {Pith review of: Variational Discretizations for Hamiltonian Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/V6XLO52V}},
  note         = {Machine review of arXiv:2411.15846}
}
read the original abstract

In this paper, we study the Lagrangian functions for a class of second-order differential systems arising from physics. For such systems, we present necessary and sufficient conditions for the existence of Lagrangian functions. Based on the variational principle and the splitting technique, we construct variational integrators and prove their equivalence to the composition of explicit symplectic methods. We apply the newly derived variational integrators to the Kepler problem and demonstrate their effectiveness in numerical simulations. Moreover, using the modified Lagrangian, we analyze the dynamical behavior of the numerical solutions in preserving the Laplace--Runge--Lenz (LRL) vector.

Figures

Figures reproduced from arXiv: 2411.15846 by the authors.

Figure 1
Figure 1. Numerical solutions with time step h = 0.05 over 4000 time steps. Red dots represent the exact solution. In this section, in addition to the simulations for numerical orbits, we also plot the errors of conserving quantities: energy, angular momentum, and the Laplace–Runge–Lenz (LRL) vector. Using the initial values (−3, 0, 0, 0.45) and a time step of h = 0.05, all numerical computations are made over 105 steps of in… view at source ↗
Figure 2
Figure 2. Numerical error of energy. Top plot: The red line sh [PITH_FULL_IMAGE:figures/full_fig_p022_2.png] view at source ↗
Figure 3
Figure 3. Error of eccentricity. Top plot: The red line shows [PITH_FULL_IMAGE:figures/full_fig_p023_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Error of rotation angle. Top plot: The red line show [PITH_FULL_IMAGE:figures/full_fig_p023_4.png]
Figure 5
Figure 5. Figure 5: Convergence rates of the Laplace–Runge–Lenz (LRL [PITH_FULL_IMAGE:figures/full_fig_p024_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 29 canonical work pages

  1. [2]

    Bampi, A

    F. Bampi, A. Morro. The inverse problem of the calculus of variations applied to continuum physics. J. Math. Phys., 23:2312–2321, 1982

    work page 1982

  2. [1]

    V . I. Arnold. Mathematical Methods of Classical Mechanics . Springer-V erlag, New Y ork, 2nd edition, 1989

    work page 1989

  3. [24]

    V ermeeren

    M. V ermeeren. Modified equations for variational integ rators. Numer . Math., 137:1001–1037, 2017

    work page 2017

  4. [3]

    S. A. Chin. Symplectic integrators from composite opera tor factorizations. Phys. Lett. A , 226:344–348, 1997

    work page 1997

  5. [4]

    K. Feng. Collected W orks II. National Defense Industry Press, Beijing, 1995

    work page 1995

  6. [5]

    K. Feng, M. Qin. Symplectic Geometric Algorithms for Hamiltonian Systems . Springer, Heidelberg, 2010

    work page 2010

  7. [6]

    Goldstein, C.P

    H. Goldstein, C.P . Poole, J. L. Safko. Classical Mechanics. Pearson Education India, 2011

    work page 2011

  8. [7]

    Hairer, C

    E. Hairer, C. Lubich, and G. Wanner. Geometric Numerical Integration. Springer-V erlag, Berlin, 2nd edition, 2006

    work page 2006

Show all 29 references
  1. [8]

    Hairer, and C

    E. Hairer, and C. Lubich. Long-term analysis of a variati onal integrator for charged-particle dynamics in a strong m agnetic field. Numer . Math., 144:699–728, 2020

    2020

  2. [9]

    Hairer, C

    E. Hairer, C. Lubich, Y . Shi. Leapfrog methods for relati vistic charged-particle dynamics. SIAM J. Numer . Anal., 61:2844–2858, 2023

    2023

  3. [10]

    R. Kozlov. Conservative discretizations of the Kepler motion. J. Phys. A, 40:4529–4539, 2007

    2007

  4. [11]

    Kraus, O

    M. Kraus, O. Maj. V ariational integrators for nonvaria tional partial di fferential equations. Phys. D, 310:37–71, 2015

    2015

  5. [12]

    Kraus, E

    M. Kraus, E. Tassi, D. Grasso. V ariational integrators for reduced magnetohydrodynamics. J. Comput. Phys., 321:435–458, 2016

    2016

  6. [13]

    Kraus, T

    M. Kraus, T. M. Tyranowski. V ariational integrators fo r stochastic dissipative Hamiltonian systems. IMA J. Numer . Anal., 41:1318–1367, 2021

    2021

  7. [14]

    J. E. Marsden, M. West. Discrete mechanics and variatio nal integrators. Acta Numer ., 10:357–514, 2001

    2001

  8. [15]

    Minesaki, Y

    Y . Minesaki, Y . Nakamura. A new discretization of the Ke pler motion which conserves the Runge-Lenz vector. Phys. Lett. A , 306:127–133, 2002

    2002

  9. [16]

    Minesaki, Y

    Y . Minesaki, Y . Nakamura. A new conservative numerical integration algorithm for the three-dimensional Kepler mo tion based on the Kustaanheimo-Stiefel regularization theory. Phys. Lett. A , 324:282–292, 2004

    2004

  10. [17]

    R. I. McLachlan, C. O ffen. Backward error analysis for variational discretisatio ns of PDEs. J. Geom. Mech., 14:447–471, 2022

    2022

  11. [18]

    P . C. Moan. On modified equations for discretizations of ODEs. J. Phys. A, 39:5545–5561, 2006

    2006

  12. [19]

    E. Noether. Invariante V ariationsprobleme. Nachrichten von der Gesellschaft der Wissenschaften zu G¨ ottingen, Mathematisch-Physikalische Klasse, 1918:235–257, 1918

    1918

  13. [20]

    P . J. Olver. Applications of Lie Groups to Di fferential Equations. Springer-V erlag, New Y ork, 2nd edition, 1993

    1993

  14. [21]

    H. Qin, X. Guan. V ariational symplectic integrator for long-time simulations of the guiding-center motion of char ged particles in general magnetic fields. Phys. Rev. Lett., 100:035006, 2008

    2008

  15. [22]

    R. M. Santilli. F oundations of Theoretical Mechanics I. Springer-V erlag, New Y ork-Heidelberg, 1978

    1978

  16. [23]

    E. Tonti. V ariational formulation of nonlinear di fferential equations. Acad. Roy. Belg. Bull. Cl. Sci. (5) , 55:137–165, 262–278, 1969

    1969

  17. [25]

    V ermeeren

    M. V ermeeren. Numerical precession in variational dis cretizations of the Kepler problem. in: Discrete Mechanics, Geometric Integration and Lie-Butcher Series, Springer , Cham, 267:333–348, 2018

    2018

  18. [26]

    V ermeeren

    M. V ermeeren. Modified equations and the Basel problem. Math. Intelligencer, 40:33–37, 2018

    2018

  19. [27]

    Oliver, S

    M. Oliver, S. V asylkevych. A new construction of modifie d equations for variational integrators. arXiv:2403.17585

    arXiv

  20. [28]

    J. Xiao, H. Qin, J. Liu. Structure-preserving geometri c particle-in-cell methods for Vlasov-Maxwell systems. Plasma Sci. Technol., 20:110501, 2018

    2018

  21. [29]

    J. Xiao, H. Qin. Explicit high-order gauge-independen t symplectic algorithms for relativistic charged particle dynamics. Comput. Phys. Commun., 241:19–27, 2019. 25

    2019

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.