PaperlayerView exact revision on arXiv

Source-linked AI summary

The specular ellipse method for scalar ordinary differential equations: exactness and accuracy up to fourth order

Kiyuob Jung

arXiv:2608.30280v1math.NA

TL;DR

The paper addresses how to construct implicit one-step methods for scalar ODEs that exploit the geometry of solution graphs. It introduces a scaled angular-mean method, proves exactness under constant signed curvature, and establishes second- through fourth-order convergence through curvature-based scale selection. The higher-order scales can also be selected numerically under the paper’s assumptions.

  • Problem

    The paper studies how a nonlinear one-step method can achieve exact tracing in special geometric cases and higher-order convergence for general smooth scalar ODEs.

  • Method

    The specular ellipse method combines two vector-field evaluations through a positive, possibly step-dependent scaled angular mean.

  • Results

    The method exactly reproduces mesh-point values for constant signed curvature under a fixed scale and uniquely solvable updates, while achieving second-, third-, or fourth-order convergence under corresponding scale and regularity conditions.

  • Takeaways & Limitations

    Curvature-based scale choices provide a geometric route to exactness and to higher-order convergence, with third- and fourth-order choices also implementable from problem data and numerical values.

  • Takeaways & Limitations

    Exact tracing is exceptional because the constant-curvature condition need not hold for a general ODE, and higher-order scale conditions involving the exact solution cannot be evaluated directly.

Abstract

from arXiv · show

This paper introduces a family of one-step implicit methods for solving scalar ordinary differential equations. At each step, the update uses a scaled angular mean of two vector-field evaluations, and the positive scale may vary from step to step. The leading terms of the local truncation error can be expressed in terms of the derivative of the signed curvature of the scaled solution graph. We prove that the proposed method reproduces the exact solution at the mesh points when the solution graph has constant signed curvature under a fixed positive scaling and each implicit update is unique. When this special geometric condition is not satisfied, we establish second-order consistency and convergence for positive scale sequences satisfying suitable uniform conditions. Furthermore, third- and fourth-order consistency and convergence can be achieved by choosing the scale to cancel the relevant curvature terms in the local truncation error. Using only the given problem data, we classify when these improvements are possible and determine the corresponding scale choices. An example shows that the proposed fourth-order method can yield smaller errors than the classical fourth-order Runge--Kutta method at the same step size.

1. Introduction.

The paper introduces the specular ellipse method, an implicit one-step ODE method using a scaled angular mean, and establishes geometric exactness plus second- through fourth-order convergence results.

  • Method: The specular ellipse method is a nonlinear one-step implicit method based on a specular derivative and a positive, possibly varying scale.The scale geometrically rescales the solution graph, while the angular mean combines consecutive vector-field evaluations.
  • Exactness: Exact mesh-point reproduction holds when a fixed positive scaling gives the solution graph constant signed curvature and every implicit update is uniquely solvable.The result includes appropriately scaled elliptic solution graphs.
  • Convergence: Under stated smoothness, solvability, Lipschitz, and nondegeneracy assumptions, the method is uniquely defined and converges with order two for sufficiently small step sizes.The leading local truncation-error term is tied to the derivative of signed curvature.
  • Higher-order convergence: Choosing scales to cancel curvature contributions yields third- and fourth-order consistency and convergence under corresponding uniform bounds.Third-order cancellation uses the current point, whereas fourth-order cancellation combines contributions from current and next points.
  • Scale selection: The paper classifies scale-selection conditions and shows that third- and fourth-order scales can be computed numerically from F, its derivatives, and numerical values.These constructions avoid requiring a closed-form exact solution while retaining the corresponding convergence orders under stated assumptions.

2. Numerical methods.

The numerical-methods section develops specular Euler variants and focuses on SET5, whose scaled angular-mean extension forms the specular ellipse method.

  • Specular Euler family: The specular Euler family uses approximations to right- and left-hand derivatives, with choices based on current or neighboring vector-field information.The constructions include one-step and two-step choices, while avoiding information from the next-next step.
  • Classical special cases: The family contains explicit and implicit Euler as special cases, while SET4 and SET6 coincide with explicit and implicit Euler, respectively.The identity C(α, α) = α explains the Euler reductions.
  • Method selection: SET1, SET2, and SET3 are not pursued because unreported numerical experiments indicated first-order convergence for all three methods.SET5 remains the genuinely nonclassical method of interest among the first six types.
  • SET5: SET5 exactly traces a circular solution on a uniform mesh, with uniqueness ensured by monotonicity of the implicit update.The exact solution values satisfy the SET5 update at every step.
  • Specular ellipse method: The specular ellipse method replaces the angular mean with a σ-scaled version, allowing positive scales σ_n to vary by step.Its update is u_{n+1} = u_n + h C_{σ_n}(F(t_{n+1},u_{n+1}), F(t_n,u_n)).

3. Geometry and exactness of the specular ellipse method.

The paper formulates the specular ellipse method through scaled solution-graph and field geometry, then proves exact mesh-point reproduction when a fixed scaling gives constant signed curvature and each implicit step is uniquely admissible.

  • Scaled geometry: σ-scaled field quantities derived from F and its partial derivatives recover the solution graph’s tangent, unit tangent, signed curvature, and curvature variation along exact solutions.The transport operator L_F G := G_t + F G_x links field quantities to their restrictions on solution graphs.
  • Scaled geometry: Dσ measures signed-curvature variation along the σ-scaled solution graph, while Kσ represents the corresponding signed curvature.These quantities are related through the solution-graph and field constructions.
  • Scaled geometry: The SE update uses a σ-scaled angular mean of consecutive vector-field values, so each numerical chord follows the angular bisector of the corresponding tangent directions.The positive scale may vary by step, with σ_n denoting the scale at step n.
  • Exactness: Under a fixed positive scaling, constant signed curvature, and unique admissible implicit updates, SE reproduces the exact solution at every mesh point.The proof uses induction: the exact next value is admissible, and uniqueness identifies it with the numerical update.
  • Exactness: For elliptic solution graphs, an appropriate scaling gives exact tracing; the unit-circle case has exact-tracing scale σ = 1.A numerical illustration reports exact tracing on an upper-branch segment.
  • Scope: Exact tracing is exceptional for general ODEs because the constant-curvature and unique-solution conditions need not hold.The paper therefore turns next to consistency and convergence for general smooth ODEs.

4. Convergence analysis for the specular ellipse method.

The specular ellipse method is second-order consistent and convergent under uniform scale conditions, with higher orders obtained by cancelling curvature-based error terms. Under additional hypotheses, the same local bounds yield global convergence, while negative-real-axis stability provides decay despite scale-dependent behavior.

  • Consistency: The local truncation error expansion relates its leading terms to derivatives of the signed curvature of the scaled solution graph.
  • Convergence: Uniform local truncation error bounds of order p produce global convergence of order p and a unique numerical sequence for sufficiently small step sizes.
  • Convergence: SE has convergence order 2 under baseline assumptions, with conditional orders 3 and 4 when the required scale choices and smoothness assumptions hold.
  • Stability: Because the angular mean is not homogeneous, the stability function generally cannot depend only on hλ, although unconditional decay still holds on the negative real axis.
  • Stability: For the negative real test equation, every positive scale gives a unique implicit update and the numerical solution decays to zero.

5. Choice of the scale for third-order accuracy.

Third-order accuracy is pursued by selecting a positive scale that cancels the leading field-based defect. The paper formulates this as a one-dimensional minimization problem and classifies when exact cancellation, a positive residual infimum, or no minimizer occurs.

  • Scale selection: The scale is chosen to cancel Dσ, which can be evaluated from the vector field and its derivatives at numerical points.
  • Scale selection: When exact field cancellation is unavailable, the method minimizes |Dσ(t, x)| over positive scales and distinguishes optimal from nonoptimal minimizers.
  • Classification: The classification determines positive defect-cancelling scales and whether the residual infimum is zero or positive.
  • Classification: If FLF F = 0, every positive scale minimizes the defect, with optimality determined by the stated additional field condition.
  • Classification: In the remaining cases, the problem may have a unique optimal positive minimizer or no minimizer, with the latter requiring boundary analysis as σ approaches 0 or infinity.

F F < 3(LF F)2. Then 0 < FL2

The classification identifies when positive scales cancel the leading defect and shows how selected scales produce third-order convergence, with special cases ranging from nonunique to nonexistent minimizers.

  • Scale classification: Case (G1) admits every positive scale as a minimizer, but optimality occurs only when L2 F F = 0.In case (G2), no finite preferred scale exists because the infimum is approached as σ →∞.
  • Scale classification: In case (G3), the unique optimal scale cancels the field-based coefficient of the leading O(h2) local truncation-error term.When evaluated on the exact trajectory, this cancels the O(h2) contribution.
  • Third-order convergence: The explicit case-(G3) scale yields third-order convergence when evaluated at (tn, un) under the stated regularity and uniformity assumptions.Replacing the exact scale by its computable numerical counterpart changes the local truncation error by O(h2|en|), contributing O(h3|en|) to the recurrence.
  • Scale classification: In case (G4), no positive minimizing scale exists and the residual infimum is approached as σ ↘0.The equality case receives a separate third-order convergence result.
  • Higher-order special case: A special argument for scales vanishing with the step size establishes convergence order 2p + 2 for every p > 0.This strengthens the general third-order result for the corresponding equation.

6. Choice of the scale for fourth-order accuracy.

The fourth-order construction solves a coupled implicit update and scale-selection problem, classifies its positive defect-cancelling scales, and proves fourth-order convergence in two admissible cases.

  • Scale selection: Because the fourth-order scale depends on the unknown next value, v and σ must be determined simultaneously in the implicit update.For each trial v, the method minimizes the absolute defect residual when exact balance has no positive solution.
  • Classification: Theorem 6.3 classifies the possible positive defect-cancelling scales through cases (E1)–(E6), including unique, multiple, and nonexistent optimal minimizers.The classification also distinguishes cases with positive residual infima or unattained boundary behavior.
  • Classification: Case (E1) has every positive scale optimal, whereas case (E3) has no positive minimizer when the defect polynomial has no positive zero.These cases illustrate that defect cancellation is not universally available.
  • Convergence scope: Cases (E2), (E4), and (E5) provide nontrivial positive optimal scales, but convergence is established only for cases (E5a) and (E5b).The paper explicitly limits the convergence analysis among the classified cases to these two subcases.
  • Fourth-order convergence: Under the stated smoothness, neighborhood, and case assumptions, Theorem 6.5 proves fourth-order convergence for cases (E5a) and (E5b).The theorem assumes u ∈ C5([t0, T]) and F ∈ C3(Ω).
  • Fourth-order convergence: The resulting numerical sequence is unique and satisfies max 0≤n≤Nh |en| ≤ Cg h4.The scale is implemented through the coupled update relation.
  • Example: For a normalized pendulum, the method reduces to the unit-circle case as θ ↘0, suggesting particularly high accuracy for small θ.The passage frames this as an expectation rather than a proved quantitative comparison.

7. Numerical results.

Numerical experiments test exact tracing, second- through fourth-order convergence, defect cancellation, and one-step behavior across elliptic, nonlinear, and pendulum problems. The results support the predicted orders and show that SE4 can outperform RK4 at equal step size, although computational cost is not systematically compared.

  • Elliptic exactness: SE is exact at every mesh point for the tested elliptic solution graphs, whereas Crank–Nicolson is not.The experiments use h = 0.3 and fixed positive scales satisfying the exact-tracing condition.
  • Vanishing scales: SE2 shows second-order convergence, while five vanishing-scale SE3 choices show observed orders 3, 4, 6, 8, and 10.The comparison uses 17 approximately logarithmically spaced step sizes from 10^-1 to 10^-3; CN, RK3, and RK4 show orders 2, 3, and 4.
  • Optimal scales: SE4 shows approximately fourth-order convergence for every tested pendulum turning angle before reaching the accuracy floor.At h = 10^-2, RK4 is 2.46 times more accurate for θ = 1, while SE4 is 6.37, 39.8, and 3.93×10^3 times more accurate for θ = 0.25, 0.1, and 0.01.
  • Defect cancellation: SE3 and SE4 cancel the targeted defects, and the Example 5.8 experiments confirm third- and fourth-order convergence, respectively.The selected scales vary with the step index even when evaluated at exact solution values, so the variation is not caused by accumulated numerical error.
  • One-step behavior: In a one-step experiment, fixed-scale SE choices show observed order three, while σ = h^-1 and CN show observed order five.SE with σ = h^-1 has a slightly smaller one-step error than CN in this experiment.
  • Numerical considerations: A systematic comparison of accuracy and computational cost remains open, while tested classical Runge–Kutta methods were generally faster in the present implementation.The unresolved comparison should account for derivative evaluations and parallel implementation.

Appendix A. An auxiliary Taylor estimate.

Appendix A develops smoothness, Taylor-expansion, scale-sensitivity, and error-propagation estimates for the scaled angular-mean function. These bounds are stated uniformly under compactness, positivity, and nondegeneracy conditions.

  • Taylor estimate: Cσ is smooth and admits a second-order Taylor expansion with a uniformly controlled fourth-order remainder under the stated compactness and nondegeneracy conditions.The constants depend only on the compact set and the lower bound η.
  • Taylor estimate: The expansion is obtained by Taylor expanding the square-root factor after introducing the normalized increment w.The remainder is uniform for sufficiently small δ and σ^2 + α^2 ≥ η.
  • Stability bounds: The partial derivatives of Cσ with respect to both arguments are positive and bounded above by 2, yielding a Lipschitz error bound.The bound follows from the Mean Value Theorem and the trigonometric representation of Cσ.

Appendix B. Auxiliary results for the convergence analysis.

Appendix B supplies the contraction, induction, and discrete error-accumulation tools used to prove existence, uniqueness, and global convergence of the implicit updates. It also formulates the hypotheses required for error-dependent scale functions.

  • Global convergence: A local residual bound is converted into a global convergence estimate for the specular ellipse method.The target induction assertion controls both numerical-point membership in a neighborhood and the error magnitude.
  • Scale-dependent analysis: The framework permits scale functions depending on the step size and numerical trajectory, provided they satisfy the prescribed uniform inequalities.These functions enter the residual and Lipschitz conditions used to establish a unique sequence of implicit updates.
  • Inductive propagation: The convergence proof propagates the local error through an induction using an error-dependent recurrence.The recurrence combines current error, next-step deviation, and the local residual estimate.
  • Existence and uniqueness: Under the stated Lipschitz and neighborhood hypotheses, the implicit update defines a contraction that maps the admissible interval into itself.The Banach fixed-point theorem then gives a unique numerical value at the next step.
  • Error accumulation: A discrete error-accumulation lemma bounds a nonnegative sequence satisfying a one-step recurrence over a finite mesh.The result is obtained by iterating the inequality from zero initial error and taking the maximum over all steps.
Loading 2608.30280v1…