Pith. sign in

REVIEW 3 major objections 5 minor 38 references

Learning efficient and provably convergent splitting methods

T0 review · 3 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 Machine-learned splitting coefficients, trained on a distribution of initial conditions, beat classical splitting methods for the Schrödinger equation when the computational budget is small.

desk verdict Solid framework for learning splitting coefficients, but the headline accuracy gains are selected on the validation set and need a fresh test set before they can be trusted. read the letter →

arxiv 2411.09444 v1 pith:OHVRF2NE submitted 2024-11-14 math.NA cs.LGcs.NA

L. M. Kreusser , H. E. Lockyer , E. H. Müller , P. Singh This is my paper · ORCID
classification math.NAcs.LGcs.NA MSC 34A2634L4065B9965L0565L2065Y20
keywords operatorsplittinggeometricnumericalintegrationmachinelearningSchrödingerequationconvergenceordercomputationalefficiencyinitialvalueproblemssymmetryandconsistency
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

The paper claims that machine-learned splitting coefficients can outperform classical splitting methods such as Trotter, Strang, and Yoshida when the computational budget is fixed and timesteps are therefore relatively large. It establishes this for the Schrödinger equation with a double-well potential, where the learned methods are up to two orders of magnitude more accurate at the same number of subflow evaluations. The learned methods are constrained to be consistent and time-reversal symmetric, which makes them provably second-order convergent as the timestep tends to zero, and they are reported to generalise to other final times, initial conditions, and potentials. By optimizing the error constant for a given class of problems rather than only the asymptotic order, the learned schemes target a different optimality criterion than classical high-order methods.

What carries the argument

The central object is the parameterised splitting flow $\Psi_{T,h}^{[f]}(\cdot;\alpha,\beta)$ whose coefficients are restricted, via a linear transform $g(\gamma)$, to the submanifold of consistent and time-reversal-symmetric methods. This transform reduces the search space from $2K$ to $K-2$ free parameters and guarantees second-order convergence in the limit $h\to 0$; training then minimises the finite-sample loss in (3.9), a mean squared error against reference solutions, and the resulting coefficients are what the paper calls the learned splittings. The palindromic structure with a symbolic zero coefficient also determines the cost in subflow evaluations per timestep.

What would settle it

Compute the L2 error versus number of subflow evaluations for the learned coefficients on a Schrödinger equation whose potential lies outside the family V1–V4 used in the paper, for example a deeper asymmetric double well; if a classical method such as Yoshida achieves lower error at equal cost on that case, the claimed advantage would fail on that distribution.

Watch

Extended reading notes

Core claim

The central discovery is that the coefficients of a splitting method can be treated as learnable parameters, with consistency and symmetry enforced algebraically, so that training minimises the finite-sample loss in (3.9) over a distribution of initial conditions. This yields methods that are second-order convergent by construction yet have much smaller error constants for large timesteps than classically derived methods, because classical methods are designed to be asymptotically optimal while learned methods optimise the constant in the error bound for the target distribution. The paper demonstrates numerically that the learned methods Learn5A, Learn8A, and Learn8B outperform Trotter, Strang, and Yoshida for fixed computational budgets on the Schrödinger equation, and that they retain their advantage under changes of final time, initial condition distribution, and potential.

Load-bearing premise

The load-bearing premise is that the finite sample of training and validation initial conditions represents the class of problems where the learned method will be used; if the training distribution is unrepresentative, the efficiency advantage over classical methods is not guaranteed outside the tested cases.

Editorial extensions

If this is right

  • Learned methods can be used as drop-in replacements for classical splittings in budget-limited simulations, giving higher accuracy at the same number of subflow evaluations.
  • The framework transfers to other equations with a natural split into two cheaper subflows, since the same parameter transform and loss construction are problem-agnostic.
  • Because second-order convergence is guaranteed by construction, the learned methods remain reliable in the small-timestep limit even though they were trained at large timesteps.
  • Projection onto the fourth-order manifold yields methods with fourth-order convergence, showing that gradient-based training can recover classically derived order conditions.

Reading between the lines

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

  • The advantage of the learned coefficients likely depends on the overlap between the training distribution and the test conditions; the paper's own comparison of loss landscapes suggests the optimum is genuinely problem-dependent.
  • A natural stress-test, left implicit by the paper, is to train on a cheap low-dimensional system and apply the learned coefficients to a higher-dimensional or nonlinear variant, which would test how much of the efficiency gain survives outside the trained family.
  • If the finite-sample training loss is a poor proxy for the true expected loss, the learned coefficients may be overtuned to the training set; the reported generalization covers only perturbations of the same double-well family.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes a machine-learning framework for finding coefficients of symmetric splitting methods for initial value problems, with the coefficients optimized to minimize expected L2-error over a distribution of initial conditions while enforcing consistency and time-reversal symmetry through a linear parameter transform (Section 3.3). The resulting schemes are formally second-order convergent by construction, and the authors demonstrate on a one-dimensional Schrödinger equation with a double-well potential that learned methods Learn5A, Learn8A, and Learn8B achieve smaller errors than Trotter, Strang, and Yoshida for fixed computational budgets (Section 4.3.3). The paper also reports generalization experiments to different final times, initial-condition distributions, and potentials (Section 4.3.4).

Significance. If the numerical claims hold, the approach offers a practical way to tailor integators to a problem distribution while retaining classical guarantees of consistency, symmetry, and provable convergence, with potential order-of-magnitude accuracy gains at fixed cost. The paper is generally transparent about its training pipeline, provides code and data, and enforces the advertised algebraic constraints rather than relying on a black-box model. The main limitation is that the headline accuracy numbers are computed on the same validation set used for coefficient selection, so the out-of-sample magnitude of the reported advantage is not yet established; the generalization experiments in Section 4.3.4 are suggestive but do not directly test the original training distribution on a held-out set.

major comments (3)
  1. [Section 4.3.3, Algorithm 3.1] The central quantitative claim that learned methods outperform classical methods for small computational budgets, quantified in Figure 5 and Table 2, is based on errors computed on Svalid, the same validation set used in Algorithm 3.1 both for screening candidates (line 3) and for selecting gamma_min (line 10). The coefficients were therefore chosen to minimize loss on exactly the set that supplies the reported error numbers, which introduces an optimistic selection bias. Section 4.3.4 tests other final times, initial-condition distributions, and potentials, but those are not held-out samples from the original training distribution U1, V1, T=10. I request that Figures 5-6 and Table 2 be re-evaluated on a fresh test set from U1, V1, T=10 that was never touched by Algorithm 3.1 or by the optimizer comparisons in Appendix D, and that the reported speed-ups be based on that set.
  2. [Sections 3.2-3.4, Abstract] The abstract and introduction advertise 'provably convergent splitting methods' and 'provable convergence and conservation guarantees', but the paper contains no formal theorem statement for the learned methods. The text argues that consistency (2.9) plus stability implies convergence, and that the parameter transform enforces consistency and symmetry, yet there is no precise statement of the assumptions under which the numerical flow (3.3) is stable for the Schrödinger IVP (4.2), nor a convergence theorem in the form of (2.5) with an explicit order. I recommend adding a theorem (for example, under the given well-posedness of the semi-discrete problem, the learned flow converges with order at least two as h to 0) with a proof or a precise reference to the classical result being applied.
  3. [Abstract, Section 4] The abstract claims conservation guarantees and Section 1 lists 'conservation' as a maintained advantage of the learned methods, but Section 4 contains no conservation experiment. The only related statement is the unexplained 'unitarity bound' in the caption of Figure 5, which is not a test of norm or energy conservation. Since the Schrödinger sub-flows are unitary, any symmetric composition of them is exactly unitary and hence norm-preserving for all h; this fact should be stated explicitly and, if conservation is advertised as a contribution, should be verified numerically (for example, by reporting max norm drift over time for Learn5A, Learn8A, Learn8B and Yoshida). Without this, the conservation claim in the abstract is unsupported.
minor comments (5)
  1. [Section 2.1] Just after equation (2.1), 'there exits a unique solution' should read 'there exists a unique solution'.
  2. [Figure 5 caption] The caption says 'solid horizontal black vertical line'; this should be 'solid black horizontal line' (or the orientation should be clarified).
  3. [Section 4.3.4] In the sentence 'To demonstrate that we have indeed adapted to our distribution of IVPs ... and and have not simply learned universal splittings ... we illustrate', there is a duplicated 'and' and a missing comma; the sentence should be rewritten.
  4. [Section 4.3.3] The word 'efficency' in 'we assess the performance and efficency' is misspelled as 'efficiency'.
  5. [Section 4.1] The word 'untractable' in 'make the solution untractable on a computer' should be 'intractable'.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: convergence guarantees are algebraically enforced, and learned coefficients are transparently fitted; the validation-set reuse is a bias caveat, not a circular step.

full rationale

The derivation is self-contained. The paper's provable-convergence claim does not depend on the learned coefficient values: it follows from the consistency condition (2.9) and the palindromic-symmetry condition (2.10), which are hard-wired into the parameter transform g(γ) in (3.7). The coefficients themselves are explicitly obtained by minimizing the empirical loss (3.9)/(3.8), and the paper does not present the resulting validation losses as theory-derived predictions. The Section 4.3.5 order analysis is a diagnostic fit of (4.4) to the observed errors and is cross-checked against the polynomial order conditions w112 and w122 from the external reference [6]; it is not used to manufacture the convergence guarantee. The only caveat is methodological rather than circular: Algorithm 3.1 uses Svalid both for candidate screening (line 3) and final selection (line 10), while the headline comparisons in Figure 5 and Table 2 are also averaged over Svalid, so the reported speed-ups may be optimistically biased. However, Section 4.3.4 evaluates the fixed learned coefficients on different final times, initial-condition distributions, and potentials, providing a genuinely out-of-sample check, and the classical methods are external benchmarks rather than functions of the learned parameters. No load-bearing self-citation occurs: the authors' package [36] is cited only for implementation details. Hence no claimed prediction is equivalent by construction to its inputs.

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

The central contribution is a recipe for fitting splitting coefficients; the only fitted quantities are the coefficients themselves and the data-generation hyperparameters. No new physical entities are introduced. The convergence guarantee is inherited from classical theory, not from the learning step.

free parameters (5)
  • Learn5A coefficients gamma = [0.3627, -0.1003, -0.1353]
    Three free coefficients of the five-stage symmetric consistent splitting, obtained by minimizing the validation loss (3.9) on Schrödinger training data in Section 4.3.2.
  • Learn8A coefficients gamma = [0.2135, -0.0582, 0.4125, -0.1352, 0.4443, -0.0251]
    Six free coefficients of the eight-stage learned method from Section 4.3.3.
  • Learn8B coefficients gamma = [0.1178, 0.3876, 0.3660, 0.2922, 0.0564, -0.0212]
    Second eight-stage learned method from Section 4.3.3.
  • Data distribution hyperparameters = xcent = -sqrt(5), xstdDev = 0.1, sigma = 0.5, T = 10, h = 1/7
    Define the training distribution U in Section 4.3.1; the learned coefficients are tailored to this distribution.
  • Order-fit coefficients C2, C4, C6 = Table 3 values, e.g. Learn5A: 1.026, 0.000, 1348.6
    Least-squares fit to the error expansion (4.4) in Section 4.3.5, used to argue the methods are close to higher order. Diagnostic only, not part of the central performance claim.
assumptions (5)
  • domain assumption The IVP and its two subflows are well-posed and differentiable with respect to time and initial conditions.
    Assumed in Sections 2.1 and 2.2 to define analytic flows and numerical flow differentiability.
  • standard math Consistency (sum alpha = sum beta = 1) and time-reversal symmetry imply second-order convergence provided the method is stable.
    Standard result from splitting method theory, cited to [6]; the paper does not prove it but relies on it for the provably convergent claim in Section 3.2.
  • domain assumption The numerical method is stable for the problems considered.
    For the linear Schrödinger semi-discretization, subflows are unitary so stability is exact; the paper does not verify stability for general IVPs, yet the convergence claim in Section 3.2 invokes it.
  • domain assumption The matrix exponential reference solutions computed with scipy.linalg.expm are accurate enough to train against.
    Section 4.2 uses scipy.linalg.expm to generate uRef; the paper assumes this is a reliable ground truth.
  • ad hoc to paper The finite training sample and global screening procedure find coefficients near the true loss minimum.
    Algorithm 3.1 is heuristic; the paper provides no guarantee that the returned gamma is close to the global minimizer of Equation 3.8, only empirical convergence in Figure 12.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning efficient and provably convergent splitting methods." pith.science (2026). https://pith.science/paper/OHVRF2NE

@misc{pith2026241109444,
  author       = {Pith},
  title        = {Pith review of: Learning efficient and provably convergent splitting methods},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OHVRF2NE}},
  note         = {Machine review of arXiv:2411.09444}
}
read the original abstract

Splitting methods are widely used for solving initial value problems (IVPs) due to their ability to simplify complicated evolutions into more manageable subproblems which can be solved efficiently and accurately. Traditionally, these methods are derived using analytic and algebraic techniques from numerical analysis, including truncated Taylor series and their Lie algebraic analogue, the Baker--Campbell--Hausdorff formula. These tools enable the development of high-order numerical methods that provide exceptional accuracy for small timesteps. Moreover, these methods often (nearly) conserve important physical invariants, such as mass, unitarity, and energy. However, in many practical applications the computational resources are limited. Thus, it is crucial to identify methods that achieve the best accuracy within a fixed computational budget, which might require taking relatively large timesteps. In this regime, high-order methods derived with traditional methods often exhibit large errors since they are only designed to be asymptotically optimal. Machine Learning techniques offer a potential solution since they can be trained to efficiently solve a given IVP with less computational resources. However, they are often purely data-driven, come with limited convergence guarantees in the small-timestep regime and do not necessarily conserve physical invariants. In this work, we propose a framework for finding machine learned splitting methods that are computationally efficient for large timesteps and have provable convergence and conservation guarantees in the small-timestep limit. We demonstrate numerically that the learned methods, which by construction converge quadratically in the timestep size, can be significantly more efficient than established methods for the Schr\"{o}dinger equation if the computational budget is limited.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 35 canonical work pages

  1. [1]

    Akhmatskaya, M

    E. Akhmatskaya, M. Fern´andez-Pend´as, T. Radivojevi´c, and J. M. Sanz-Serna, Adaptive splitting integrators for enhancing sampling efficiency of modified hamiltonian monte carlo methods in molecular simulation, Langmuir, (2017)

    work page 2017

  2. [2]

    A. H. Al-Mohy and N. J. Higham , A new scaling and squaring algorithm for the matrix exponential, SIAM Journal on Matrix Analysis and Applications, 31 (2010), pp. 970–989

    work page 2010

  3. [3]

    Bernier , Exact splitting methods for semigroups generated by inhomogeneous quadratic differential operators, 2020

    J. Bernier , Exact splitting methods for semigroups generated by inhomogeneous quadratic differential operators, 2020

    work page 2020

  4. [4]

    Bernier, N

    J. Bernier, N. Crouseilles, and Y. Li , Exact splitting methods for kinetic and schr¨ odinger equations, 2019

    work page 2019

  5. [5]

    Blanes, F

    S. Blanes, F. Casas, and A. Murua , Splitting and composition methods in the numerical integration of differential equations, Bolet ´ ın de la Sociedad Espa˜ nola de Matem´ atica Apli- cada, (2008)

    work page 2008

  6. [6]

    Blanes, F

    S. Blanes, F. Casas, and A. Murua , Splitting methods for differential equations, Acta Nu- mer., (2024)

    work page 2024

  7. [7]

    Blanes, F

    S. Blanes, F. Casas, and J. M. Sanz-Serna , Numerical integrators for the hybrid monte carlo method, SIAM Journal on Scientific Computing, 36 (2014), p. A1556–A1580

    work page 2014

  8. [8]

    Blondel, Q

    M. Blondel, Q. Berthet, et al. , Efficient and modular implicit differentiation, arXiv:2105.15183, (2021)

    arXiv 2021

Show all 38 references
  1. [9]

    Bonfiglioli and R

    A. Bonfiglioli and R. Fulci, Topics in Noncommutative Algebra: The Theorem of Campbell, Baker, Hausdorff and Dynkin, vol. 2034, Springer Berlin, Heidelberg, 01 2012. 24 L. M. KREUSSER, H. E. LOCKYER, E. H. M ¨ULLER, AND P. SINGH

    2012

  2. [10]

    Bradbury, R

    J. Bradbury, R. Frostig, P. Hawkins, et al. , JAX: composable transformations of Python+NumPy programs, 2018

    2018

  3. [11]

    R. T. Q. Chen, Y. Rubanova, J. Bettencourt, and D. Duvenaud , Neural ordinary differential equations, in Proceedings of the 32nd International Conference on Neural In- formation Processing Systems, NIPS’18, Curran Associates Inc., 2018, pp. 6572–6583

    2018

  4. [12]

    X. Chen, C. Liang, D. Huang, et al. , Symbolic discovery of optimization algorithms, Ad- vances in neural information processing systems, 36 (2024)

    2024

  5. [13]

    J. W. Cooley and J. W. Tukey , An algorithm for the machine calculation of complex fourier series, Mathematics of computation, 19 (1965), pp. 297–301

    1965

  6. [14]

    Duchi, E

    J. Duchi, E. Hazan, and Y. Singer , Adaptive subgradient methods for online learning and stochastic optimization., Journal of machine learning research, 12 (2011)

    2011

  7. [15]

    Engel, S

    K. Engel, S. Brendle, R. Nagel, et al. , One-Parameter Semigroups for Linear Evolution Equations, Graduate Texts in Mathematics, Springer New York, 1999

    1999

  8. [16]

    Evans, Partial Differential Equations, Graduate studies in mathematics, American Mathe- matical Society, 2010

    L. Evans, Partial Differential Equations, Graduate studies in mathematics, American Mathe- matical Society, 2010

    2010

  9. [17]

    Giordano and L

    P. Giordano and L. L. Baglini , A Picard-Lindel¨ oftheorem for smooth PDE, 2022

    2022

  10. [18]

    Glowinski, S

    R. Glowinski, S. Osher, and W. Yin, Splitting Methods in Communication, Imaging, Science and Engineering, Springer Cham, 01 2017

    2017

  11. [19]

    T. G. Grossmann, U. J. Komorowska, J. Latz, and C.-B. Sch¨onlieb, Can physics-informed neural networks beat the finite element method?, 2023

    2023

  12. [20]

    Hairer, C

    E. Hairer, C. Lubich, and G. W anner, Geometric numerical integration, vol. 31 of Springer Series in Computational Mathematics, Springer-Verlag, Berlin, second ed., 2006

    2006

  13. [21]

    Hairer, S

    E. Hairer, S. Nørsett, and G. W anner, Solving Ordinary Differential Equations I: Nonstiff Problems, Springer Series in Computational Mathematics, Springer, 2008

    2008

  14. [22]

    Horst and H

    R. Horst and H. Tuy , Global Optimization: Deterministic Approaches, Springer, 1996

    1996

  15. [23]

    Iserles, H

    A. Iserles, H. Z. Munthe-Kaas, S. P. Nørsett, and A. Zanna , Lie-group methods, Acta Numer., 9 (2000), pp. 215–365

    2000

  16. [24]

    G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. W ang, and L. Yang , Physics-informed machine learning, Nature Reviews Physics, 3 (2021), pp. 422–440

    2021

  17. [25]

    D. P. Kingma and J. Ba , Adam: A method for stochastic optimization, arXiv:1412.6980, (2014)

    2014 arXiv

  18. [26]

    Levenberg , A method for the solution of certain non-linear problems in least squares, Quarterly of applied mathematics, 2 (1944), pp

    K. Levenberg , A method for the solution of certain non-linear problems in least squares, Quarterly of applied mathematics, 2 (1944), pp. 164–168

    1944

  19. [27]

    D. W. Marquardt, An algorithm for least-squares estimation of nonlinear parameters, Journal of the society for Industrial and Applied Mathematics, 11 (1963), pp. 431–441

    1963

  20. [28]

    R. I. McLachlan and G. R. W. Quispel, Splitting methods, Acta Numer., 11 (2002), pp. 341– 434

    2002

  21. [29]

    Moler and C

    C. Moler and C. V an Loan, Nineteen dubious ways to compute the exponential of a matrix, twenty-five years later, SIAM Review, 45 (2003), pp. 3–49

    2003

  22. [30]

    M. M. Noel , A new gradient based particle swarm optimization algorithm for accurate computation of global minimum, Applied Soft Computing, 12 (2012), pp. 353–359

    2012

  23. [31]

    Offen and S

    C. Offen and S. Ober-Bl ¨obaum, Symplectic integration of learned hamiltonian systems, Chaos: An Interdisciplinary Journal of Nonlinear Science, 32 (2022)

    2022

  24. [32]

    Omelyan, I

    I. Omelyan, I. Mryglod, and R. Folk , Symplectic analytically integrable decomposition algorithms: classification, derivation, and application to molecular dynamics, quantum and celestial mechanics simulations, Comput. Phys. Comms., 151 (2003), pp. 272–314

    2003

  25. [33]

    Raissi, P

    M. Raissi, P. Perdikaris, and G. Karniadakis , Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations, Journal of Computational Physics, 378 (2019), pp. 686–707

    2019

  26. [34]

    Ruggenthaler, M

    M. Ruggenthaler, M. Penz, and R. van Leeuwen , Existence, uniqueness, and construction of the density-potential mapping in time-dependent density-functional theory, Journal of Physics: Condensed Matter, 27 (2015), p. 203202

    2015

  27. [35]

    J. J. Sakurai and J. Napolitano, Modern Quantum Mechanics, Addison Wesley, 2 ed., 1985

    1985

  28. [36]

    Singh , A PyTorch compatible Differentiable Numerical Algorithms package for computational quantum mechanics

    P. Singh , A PyTorch compatible Differentiable Numerical Algorithms package for computational quantum mechanics. https://pypi.org/project/expsolve/

  29. [37]

    Suzuki , Fractal decomposition of exponential operators with applications to many-body theories and monte carlo simulations, Physics Letters A, 146 (1990), pp

    M. Suzuki , Fractal decomposition of exponential operators with applications to many-body theories and monte carlo simulations, Physics Letters A, 146 (1990), pp. 319–323

    1990

  30. [38]

    Learn8A” and “Learn8B

    K. Yajima , Existence of solutions for Schr¨ odinger evolution equations, Communications in Mathematical Physics, 110 (1987), pp. 415 – 426. LEARNING EFFICIENT AND PROV ABLY CONVERGENT SPLITTING METHODS 25 Appendix A. T riple jump and composition methods. We describe the tripl...

    1987

Pith tools

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