Pith. sign in

REVIEW 3 major objections 5 minor 6 cited by

Explainable AI-assisted Optimization for Feynman Integral Reduction

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read An AI-designed ordering rule cuts the seed count in Feynman integral reduction by up to a factor of 3,058.

desk verdict Useful and honest, but the headline improvement is not isolated from the category-selection algorithm; needs an ablation before the priority function gets credit. read the letter →

arxiv 2502.09544 v1 pith:IFYWSF36 submitted 2025-02-13 hep-ph hep-th

Zhuo-Yang Song , Tong-Zhi Yang , Qing-Hong Cao , Ming-xing Luo , Hua Xing Zhu This is my paper · ORCID
classification hep-phhep-th
keywords Feynmanintegralsintegration-by-partsreductionLaportaalgorithmpriorityfunctionFunSearchlargelanguagemodelsgeneticalgorithmsphase-space
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 claims that the computationally hard part of Feynman integral evaluation — integration-by-parts (IBP) reduction, which solves large linear systems generated from 'seeding' integrals — can be made much cheaper by sorting the seeds with one explicit formula. The formula is a priority function that ranks each candidate seed by its distance from the origin in index space plus its distance from the target integral; an AI search called FunSearch discovered its shape on the simplest one-loop case, and the authors generalized it by hand to arbitrary numbers of loops and legs. Applied to five-loop six-particle phase-space integral families, the priority function cuts the required number of seeds by up to a factor of 3,058, turning reductions that previously needed more than 400 GB of memory into runs that finish on a laptop in under two minutes. If the transfer holds, IBP reduction stops being the main bottleneck for a wide class of higher-order calculations.

What carries the argument

The load-bearing object is the Ellipse-type priority function $F_0$ of Eq. (16), which assigns each candidate seeding integral a rank equal to minus the sum of two geometric quantities in index space: an $\ell^m$ norm of its propagator indices (a measure of how many dots and numerators it carries) and the Euclidean distance from the seeding point to the target integral. The name records its discovery path: on the one-loop massless bubble problem, FunSearch — a search that evolves short programs through a large language model combined with genetic algorithms — first reproduced the standard Laporta priority (a diagonal line in index space), then a Box priority, then converged to the Ellipse-type form, which the authors read off from the evolved code and generalized to any loop and leg count. Functionally, $F_0$ defines a nested chain of subsets $S_i$ of the seeding set, so the smallest subset sufficient to reduce the target can be located by binary search; the multi-loop applications refine this by sorting seeds into categories and selecting the first $M$ and $G$ categories with a relaxation factor $a = 1/3$.

What would settle it

Run $F_0$ with $m = 2$ on a multi-loop family not used in the paper — say a different non-planar five-loop phase-space family or a six-loop propagator integral with several dotted lines and numerators — and compare the required seed count and peak memory against the best improved-seeding scheme for that family; if the improvement factor collapses toward one, or the solver fails where improved seeding succeeds, the claimed transferability is refuted. A more direct check: evolve a priority function with FunSearch on a genuine two-loop IBP problem; if the best evolved function never beats the Ellipse-type $F_0$ carried over from one loop, the scalability argument loses its evidentiary base.

Watch

Extended reading notes

Core claim

The paper's central claim is that the priority function $F_0$ of Eq. (16), $$F_0(I(n_i); I_{\rm target}(t_i)) = -\left[\left(\sum_i |n_i|^m\right)^{1/m} + \left(\sum_i (n_i - t_i)^2\right)^{1/2}\right],$$ with $m \in \{1,2,4,\infty\}$ and the first term read as $\max_i |n_i|$ for $m = \infty$, is a near-universal way to order IBP seeding integrals: it penalizes seeds that carry dots or numerators (large $\ell^m$ norm) and seeds far from the target integral in index space (large Euclidean distance). Ranked by $F_0$, the seeds generated by conventional schemes arrange into nested subsets, and a binary search locates the smallest subset that still solves the target; memory use is roughly proportional to the size of that subset. On the one-loop problem where the function was evolved, the required subset shrinks by about a factor of five; on the five-loop planar and non-planar families the seed count falls by up to 24.8 times for the hardest reachable top-sector integrals and by 3,058 times (planar) and 1,060 times (non-planar) for sub-sector integrals with many dots and numerators, converting out-of-memory failures into fast laptop-scale reductions.

Load-bearing premise

The load-bearing premise is that the two-dimensional index space of the one-loop massless bubble is representative enough that a priority function optimized there, after human generalization, still works for five-loop six-particle phase-space families with 20 propagators and up to 46 master integrals — a transfer the paper asserts rather than proves, and one its own appendix undercuts slightly by reporting that running FunSearch directly on two-loop problems found nothing better than the Box priority.

Editorial extensions

If this is right

  • The five-loop planar top-sector integrals $s_{10}$ and $s_{12}$, which the improved-seeding baseline cannot complete within 400 GB of memory, become computable, and the planar $d_5$ reduction finishes in under two minutes on a standard laptop.
  • The improvement factor grows with target complexity across the tested range (about 1.0x for the simplest planar top-sector seeds up to 24.8x for $s_{12}$ and 3,058x for $d_5$), so the method pays off most where previous algorithms failed outright.
  • The same hand-generalized priority function works for planar and non-planar integral families, for single-target reductions, and for a 955-target dimensional-recurrence problem in which it halves the seed count relative to improved seeding.
  • Because $F_0$ is a short, interpretable expression rather than a trained black box, it can be ported to new integral families and refined by human experts without rerunning the search.

Reading between the lines

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

  • If the one-loop-to-five-loop transfer is as general as the paper claims, the same recipe — evolve a priority function on a toy problem with FunSearch, then have an expert generalize the closed form — could be applied to other combinatorial bottlenecks in amplitude computation, such as seed selection for syzygy-based or block-triangular reduction schemes.
  • The paper optimizes seed count rather than solve time, and says so; a priority function tuned directly against equation-solving time could produce further speedups beyond the reported ratios, since the fastest seed set and the smallest seed set need not coincide.
  • The appendix's multi-target variant of Eq. (B1) was only shown to be at least as good as the single-target Ellipse priority; comparing it directly with the recursive single-target strategy on the 955-target problem is a cheap, natural next experiment.
  • Only the $m=2$ variant was used in the multi-loop benchmarks, so the sensitivity of the reported gains to the norm choice ($m=1,4,\infty$) is untested; applying those variants to a hard non-planar family would show whether the improvement is robust to that choice.
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 method to reduce the size of integration-by-parts (IBP) systems by ordering seeding integrals with a priority function discovered by the FunSearch algorithm on a one-loop massless bubble integral, then generalized by human experts to the Ellipse-type function F0 in Eq. (16). The authors apply F0, together with a category-based subset-selection algorithm, to planar and non-planar five-loop six-particle phase-space integral families. Tables I and II report large reductions in the number of seeding integrals, up to a factor of 3058 for a planar integral (d5) and 1060 for a non-planar integral (d5), compared to an improved-seeding baseline, and the paper claims these reductions make previously infeasible IBP reductions tractable.

Significance. If the reported reductions are genuinely attributable to the discovered priority function F0, this would be a practically valuable and interpretable heuristic for one of the main computational bottlenecks in multi-loop amplitude calculations. The combination of FunSearch on a simple problem with human transfer to harder problems is a promising methodology, and the target integrals (five-loop, six-particle phase-space, up to 46 master integrals) are nontrivial and of direct interest. The paper's strengths are the explicit, closed-form priority function (Eq. (16)), the use of FiniteFlow for the main comparisons, and the clear statement of the algorithm in Sec. VI. However, as detailed below, the current evidence does not isolate the effect of the priority function from the subset-selection procedure, so the quantitative claims require additional experiments before they can be fully credited.

major comments (3)
  1. [Sec. VI, Tables I and II] The comparison labeled 'Priority function F0' in Tables I and II includes the full category-selection and search algorithm described in Sec. VI (steps 1–5, Eq. (17), with relaxation factor a=1/3), whereas the 'Improved seeding' baseline is the full minimal seeding set without this selection. The paper never runs the same selection algorithm with a different ordering, such as the Laporta priority of Eq. (14), the Box priority of Listing 3, or a random order. Because a complete seeding set contains many redundant equations, a subset-selection procedure can in principle reduce 21,740,796 seeds to 7,109 even with a generic ordering. Thus the reported factors of 3058 (planar d5) and 1060 (non-planar d5) do not currently establish the efficacy of Eq. (16) in isolation; they measure the combined pipeline. Please add an ablation that applies the identical Sec. VI search algorithm with alternative priority functions, including Box and Laporta, for at least the d5 cases in both families.
  2. [Sec. IV–V, Appendix B] The priority function F0 in Eq. (16) is not itself the FunSearch output; it is a human generalization (with m chosen as 2) of an evolved one-loop function. Section IV states 'we hope to obtain a general best-estimated priority function under a simpler IBP problem and naturally extend it', and Appendix B concedes that on two-loop problems FunSearch did not beat Box priority. The multi-loop results therefore depend critically on the human transfer step, and the only direct evidence that Eq. (16) outperforms Box priority is the one-loop factor-of-five improvement in Sec. V (with a single reported target, I(15,10)). To support the central claim that F0 is a 'best-estimated priority function', the authors should compare Eq. (16) against Box priority and Laporta priority on the same multi-loop benchmarks using the same selection protocol.
  3. [General reproducibility (Secs. IV–VI)] No code or data are shipped, and the exact procedure for constructing the categories used in Sec. VI is not specified beyond 'O(10^2) categories are produced for O(10^5) seeding integrals'. The 'improved seeding' minimal pattern search is also not described algorithmically. Timings in Tables I and II are single numerical sample points with no run-to-run variation or detailed machine characterization. For a computational paper whose central contribution is an algorithm, this is insufficient for independent verification. Please provide the implementation, or at least the exact category-generation rules, the optimized seeding vectors, and the binary-search/relaxation details, together with at least a few repeated timing runs for one representative case.
minor comments (5)
  1. [Table I and Table II captions] The notation a/b and a/b/c is explained, but the columns NI/NP and TI/TP are not defined explicitly in the table captions; clarify that these are ratios of seed counts and timings between the F0 column and the improved-seeding column.
  2. [Sec. VI, Eq. (17)] The quantities Ntot and Nm are defined in text, but the relation to the 'first M categories' is ambiguous when multiple sets of M categories could span all master integrals; state whether M is chosen by a deterministic rule and how ties are broken.
  3. [Sec. VI.A, last paragraph] The sentence claiming the d5 reduction 'can now be performed on a standard laptop in under two minutes' appears inconsistent with Table I, which reports 47.7 seconds to learn and 2.24 seconds to solve on a Xeon Gold 6148 with 768 GB RAM; please reconcile or remove the laptop claim.
  4. [Sec. VI.A, after Table I] The text says 'the reduction including the reconstruction of D can now be performed...', but no details of the reconstruction step are given; define D and the reconstruction procedure, or cite the FiniteFlow description.
  5. [Appendix A and Fig. 2 and Fig. 4] Figures 2 and 4 appear to contain garbled or mis-encoded text that is not readable; please regenerate these figures with proper font handling.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the priority function is evolved on a one-loop toy problem and tested out-of-sample on multi-loop families; no fitted parameter or self-citation is load-bearing.

full rationale

The paper's derivation chain is self-contained in the sense relevant to circularity. The priority function F0 in Eq. (16) is obtained by running FunSearch on the one-loop massless bubble (Sec. V), with fitness given by the average minimal seeding count for 30 one-loop targets (Appendix A, Eq. (A1)). The multi-loop benchmarks in Sec. VI (planar and non-planar five-loop phase-space families) are not used in the search or in any parameter fit; the generalized form Eq. (16) is transferred to these benchmarks out-of-sample after a human generalization step. No equation in the paper reduces by construction to a fitted parameter, and no result is imported from a self-citation chain: the cited building blocks (Laporta seeding [6], improved seeding [19,72,73], FiniteFlow [31], FunSearch [34]) are external and independently documented. The paper even reports an honest negative result in Appendix B, stating that running FunSearch directly on two-loop problems 'no results surpassing the Box priority were observed,' which is inconsistent with a story in which the reported gains are forced by the algorithm's definition. The main validity concern is an ablation gap: the 3058x and 1060x factors compare F0 plus the category-selection/search algorithm of Sec. VI against the full improved-seeding set, without a control run using Laporta, Box, or random priority under the same selection protocol. That is a benchmarking confound, not circularity, and does not change the circularity score.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim is an empirical heuristic, so the ledger is dominated by modeling assumptions about representativeness and proportionality rather than fitted physics parameters.

free parameters (3)
  • m, exponent in Eq. (16) = m=2 (chosen from {1,2,4,inf})
    Human-selected after FunSearch runs; changes the norm shape of the priority function and is not fitted to the multi-loop benchmarks.
  • relaxation factor a in Eq. (17) = 1/3
    Set by hand for all examples; no sensitivity scan is reported.
  • improved seeding pattern (g0,g1,...) and dmax for baselines = planar: (0,-1,-1,-2,...), dmax=0; non-planar: (0,-1,-2,-3,...), dmax=1
    Hand-chosen baseline configurations define the 'improved seeding' comparison; the reported improvement factors depend on this choice.
assumptions (4)
  • standard math Integration-by-parts identities in dimensional regularization, Eq. (2), and reverse unitarity, Eq. (3), provide complete recurrence relations for Feynman integrals.
    Standard background for IBP reduction; the paper uses these as the starting point.
  • domain assumption Memory usage and runtime of IBP reduction scale approximately with the number of seeding integrals.
    Sec. IV states this proportionality to justify minimizing the size of the seeding subset; it is an approximation and is not separately validated for the sparse solver used.
  • domain assumption The one-loop massless bubble integral family is representative enough that a priority function optimized there extends to multi-loop, multi-scale planar and non-planar integrals.
    Sec. IV says a general priority function can be obtained under a simpler problem and naturally extended; the paper's multi-loop results depend on this transfer.
  • domain assumption The seed box with side length 10+t1+t2 is large enough that the evolved priority reflects properties of the index space rather than the box boundary.
    Sec. IV makes this assumption to justify transferring the function to other problems; if false, the found priority could be an artifact of the finite search box.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Explainable AI-assisted Optimization for Feynman Integral Reduction." pith.science (2026). https://pith.science/paper/IFYWSF36

@misc{pith2026250209544,
  author       = {Pith},
  title        = {Pith review of: Explainable AI-assisted Optimization for Feynman Integral Reduction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IFYWSF36}},
  note         = {Machine review of arXiv:2502.09544}
}
read the original abstract

We present a novel approach to optimizing the reduction of Feynman integrals using integration-by-parts identities. By developing a priority function through the FunSearch algorithm, which combines large language models and genetic algorithms, we achieve significant improvements in memory usage and computational efficiency compared to traditional methods. Our approach demonstrates substantial reductions in the required seeding integrals, making previously intractable integrals more manageable. Tested on a variety of Feynman integrals, including one-loop and multi-loop cases with planar and non-planar configurations, our method demonstrates remarkable scalability and adaptability. For reductions of certain Feynman integrals with many dots and numerators, we observed an improvement by a factor of 3058 compared to traditional methods. This work provides a powerful and interpretable framework for optimizing IBP reductions, paving the way for more efficient and practical calculations in high-energy physics.

Figures

Figures reproduced from arXiv: 2502.09544 by the authors.

Figure 1
Figure 1. FIG. 1: This flowchart outlines the operational sequence of the FunSearch algorithm, which is designed to optimize [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2: Visualization of the Laporta, Box, and Ellipse-like priority. The target integral [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3: Two six-particle phase-space integral families [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: FIG. 4: Evolution of the average seeding integral [PITH_FULL_IMAGE:figures/full_fig_p014_4.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 6 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Scattering Amplitudes as Programs: Self-Evolving Search for Theory and Event Generation

    hep-ph 2026-07 accept novelty 7.0 of 10

    Self-evolving program search over scattering amplitudes discovers known and hybrid evaluation structures, cutting counted arithmetic ~48x and reaching within ~14x of optimized MadGraph at n=6 while beating the tested ...

  2. State-Dependent Visibility of Non-Commutative Ordering in Quantum Dynamics

    quant-ph 2026-08 conditional novelty 6.0 of 10

    For fixed non-commuting generators, the reversed-ordering operations can be perfectly indistinguishable on one state and clearly distinguishable on another, a state-dependent effect the authors call dynamical Abelianization.

  3. An adaptive inverse-problem framework for one-loop five-gluon BCJ numerators

    hep-th 2026-07 conditional novelty 6.0 of 10

    The complete 207-dimensional family of one-loop five-gluon BCJ numerator coefficients is reconstructed exactly, and every direction in that family is shown to be invisible to the specified color-ring observable.

  4. Tensor Reduction of Sunset by Generating Function

    hep-th 2025-09 conditional novelty 6.0 of 10

    A complete set of recurrence relations is derived to reduce any tensor integral of the sunset diagram to seven master integrals, using generating functions supplemented by syzygy equations.

  5. Kira 3: integral reduction with efficient seeding and optimized equation selection

    hep-ph 2025-05 conditional novelty 6.0 of 10

    Kira 3 cuts Feynman-integral reduction cost by up to two orders of magnitude using smarter seeding and equation selection.

  6. AMFlow 2.0: significant algorithmic and software improvements for Feynman integral evaluation

    hep-ph 2026-07 accept novelty 5.0 of 10

    AMFlow 2.0 cuts symbolic and numerical cost of multi-loop Feynman integral evaluation via an FT recursion mode, a C++ DE solver, and modern IBP reducers, demonstrated on a three-loop five-point family.

Reference graph

Works this paper leans on

95 extracted references · 12 canonical work pages · cited by 6 Pith papers

  1. [1]

    Count the total number of initial seeding integrals, denoted as Ntot

  2. [2]

    Let the number of seeding inte- grals in these first M categories be Nm

    From left to right, identify the first M categories in which the seeding integrals collectively span all master integrals. Let the number of seeding inte- grals in these first M categories be Nm

  3. [3]

    (17) Identify the first G categories that collectively con- tain at least Nr seeding integrals

    Introduce an adjustable relaxation factor 0 ≤ a ≤ 1 and define Nr = N a tot N 1−a m . (17) Identify the first G categories that collectively con- tain at least Nr seeding integrals

  4. [4]

    Solve the resulting linear equations using the sparse solver provided in FiniteFlow

    Generate IBP equations separately from the seed- ing integrals in the first M and first G categories. Solve the resulting linear equations using the sparse solver provided in FiniteFlow. If both cases suc- cessfully reduce the target integrals, we stop. If only the second case is sufficient, we use all seed- ing integrals for reductions and also apply a b...

  5. [5]

    The above algorithm also improves the search time com- pared to a direct binary search approach

    The final selection of categories directly determines the required seeding integrals. The above algorithm also improves the search time com- pared to a direct binary search approach. In the following, we present several explicit examples to show the effectiveness of Ellipse-type priority functions. We will first show the examples with single target inte- ...

    2020

  6. [6]

    K. G. Chetyrkin and F. V. Tkachov, Nucl. Phys. B 192, 159 (1981)

    1981

  7. [7]

    F. V. Tkachov, Phys. Lett. B 100, 65 (1981). 11

    1981

  8. [8]

    A. V. Kotikov, Phys. Lett. B 254, 158 (1991)

    1991

Show all 95 references
  1. [9]

    Gehrmann and E

    T. Gehrmann and E. Remiddi, Nucl. Phys. B 580, 485 (2000), arXiv:hep-ph/9912329

    2000 arXiv

  2. [10]

    J. M. Henn, Phys. Rev. Lett. 110, 251601 (2013), arXiv:1304.1806 [hep-th]

    2013 arXiv

  3. [11]

    Laporta, Int

    S. Laporta, Int. J. Mod. Phys. A 15, 5087 (2000), arXiv:hep-ph/0102033

    2000 arXiv

  4. [12]

    von Manteuffel and R

    A. von Manteuffel and R. M. Schabinger, Phys. Lett. B 744, 101 (2015), arXiv:1406.4513 [hep-ph]

    2015 arXiv

  5. [13]

    Peraro, JHEP 12, 030 (2016), arXiv:1608.01902 [hep- ph]

    T. Peraro, JHEP 12, 030 (2016), arXiv:1608.01902 [hep- ph]

    2016 arXiv

  6. [14]

    Gluza, K

    J. Gluza, K. Kajda, and D. A. Kosower, Phys. Rev. D 83, 045012 (2011), arXiv:1009.0472 [hep-th]

    2011 arXiv

  7. [15]

    K. J. Larsen and Y. Zhang, Phys. Rev. D 93, 041701 (2016), arXiv:1511.01071 [hep-th]

    2016 arXiv

  8. [16]

    B¨ ohm, A

    J. B¨ ohm, A. Georgoudis, K. J. Larsen, M. Schulze, and Y. Zhang, Phys. Rev. D 98, 025023 (2018), arXiv:1712.09737 [hep-th]

    2018 arXiv

  9. [17]

    R. N. Lee, in 49th Rencontres de Moriond on QCD and High Energy Interactions (2014) pp. 297–300, arXiv:1405.5616 [hep-ph]

    2014 arXiv

  10. [18]

    Bitoun, C

    T. Bitoun, C. Bogner, R. P. Klausen, and E. Panzer, Lett. Math. Phys. 109, 497 (2019), arXiv:1712.09215 [hep-th]

    2019 arXiv

  11. [19]

    Agarwal, S

    B. Agarwal, S. P. Jones, and A. von Manteuffel, JHEP 05, 256 (2021), arXiv:2011.15113 [hep-ph]

    2021 arXiv

  12. [20]

    Mastrolia and S

    P. Mastrolia and S. Mizera, JHEP 02, 139 (2019), arXiv:1810.03818 [hep-th]

    2019 arXiv

  13. [21]

    Frellesvig, F

    H. Frellesvig, F. Gasparotto, M. K. Mandal, P. Mastro- lia, L. Mattiazzi, and S. Mizera, Phys. Rev. Lett. 123, 201602 (2019), arXiv:1907.02000 [hep-th]

    2019 arXiv

  14. [22]

    Liu and Y.-Q

    X. Liu and Y.-Q. Ma, Phys. Rev. D 99, 071501 (2019), arXiv:1801.10523 [hep-ph]

    2019 arXiv

  15. [23]

    X. Guan, X. Liu, and Y.-Q. Ma, Chin. Phys. C 44, 093106 (2020), arXiv:1912.09294 [hep-ph]

    2020 arXiv

  16. [24]

    X. Guan, X. Liu, Y.-Q. Ma, and W.-H. Wu, (2024), arXiv:2405.14621 [hep-ph]

    2024 arXiv

  17. [25]

    Anastasiou and A

    C. Anastasiou and A. Lazopoulos, JHEP 07, 046 (2004), arXiv:hep-ph/0404258

    2004 arXiv

  18. [26]

    R. N. Lee, (2012), arXiv:1212.2685 [hep-ph]

    2012 arXiv

  19. [27]

    R. N. Lee, J. Phys. Conf. Ser. 523, 012059 (2014), arXiv:1310.1145 [hep-ph]

    2014 arXiv

  20. [28]

    A. V. Smirnov and F. S. Chukharev, Comput. Phys. Commun. 247, 106877 (2020), arXiv:1901.07808 [hep- ph]

    2020 arXiv

  21. [29]

    A. V. Smirnov and M. Zeng, Comput. Phys. Commun. 302, 109261 (2024), arXiv:2311.02370 [hep-ph]

    2024 arXiv

  22. [30]

    von Manteuffel and C

    A. von Manteuffel and C. Studerus, (2012), arXiv:1201.4330 [hep-ph]

    2012 arXiv

  23. [31]

    Studerus, Comput

    C. Studerus, Comput. Phys. Commun. 181, 1293 (2010), arXiv:0912.2546 [physics.comp-ph]

    2010 arXiv

  24. [32]

    Maierh¨ ofer, J

    P. Maierh¨ ofer, J. Usovitsch, and P. Uwer, Comput. Phys. Commun. 230, 99 (2018), arXiv:1705.05610 [hep-ph]

    2018 arXiv

  25. [33]

    Klappert and F

    J. Klappert and F. Lange, Comput. Phys. Commun. 247, 106951 (2020), arXiv:1904.00009 [cs.SC]

    2020 arXiv

  26. [34]

    Klappert, F

    J. Klappert, F. Lange, P. Maierh¨ ofer, and J. Uso- vitsch, Comput. Phys. Commun. 266, 108024 (2021), arXiv:2008.06494 [hep-ph]

    2021 arXiv

  27. [35]

    Ruijl, T

    B. Ruijl, T. Ueda, and J. A. M. Vermaseren, Comput. Phys. Commun. 253, 107198 (2020), arXiv:1704.06650 [hep-ph]

    2020 arXiv

  28. [36]

    Peraro, JHEP 07, 031 (2019), arXiv:1905.08019 [hep- ph]

    T. Peraro, JHEP 07, 031 (2019), arXiv:1905.08019 [hep- ph]

    2019 arXiv

  29. [37]

    Z. Wu, J. Boehm, R. Ma, H. Xu, and Y. Zhang, Comput. Phys. Commun. 295, 108999 (2024), arXiv:2305.08783 [hep-ph]

    2024 arXiv

  30. [38]

    Chen, (2024), arXiv:2408.06426 [hep-ph]

    W. Chen, (2024), arXiv:2408.06426 [hep-ph]

    2024 arXiv

  31. [39]

    Romera-Paredes, M

    B. Romera-Paredes, M. Barekatain, A. Novikov, M. Ba- log, M. P. Kumar, E. Dupont, F. J. Ruiz, J. S. Ellenberg, P. Wang, O. Fawzi, et al., Nature 625, 468 (2024)

    2024

  32. [40]

    He, Nature Rev

    Y.-H. He, Nature Rev. Phys. 6, 546 (2024), arXiv:2405.19973 [math.HO]

    2024 arXiv

  33. [41]

    He, ed., Machine Learning in Pure Mathematics and Theoretical Physics(World Scientific, 2023)

    Y.-H. He, ed., Machine Learning in Pure Mathematics and Theoretical Physics(World Scientific, 2023)

    2023

  34. [42]

    He and S.-J

    Y.-H. He and S.-J. Lee, Phys. Lett. B798, 134889 (2019), arXiv:1904.08530 [hep-th]

    2019 arXiv

  35. [43]

    Y.-H. He, Int. J. Mod. Phys. A 36, 2130017 (2021), arXiv:2011.14442 [hep-th]

    2021 arXiv

  36. [44]

    T. Cai, G. W. Merz, F. Charton, N. Nolte, M. Wilhelm, K. Cranmer, and L. J. Dixon, Mach. Learn. Sci. Tech. 5, 035073 (2024), arXiv:2405.06107 [cs.LG]

    2024 arXiv

  37. [45]

    L. J. Dixon, O. Gurdogan, A. J. McLeod, and M. Wil- helm, JHEP 07, 153 (2022), arXiv:2204.11901 [hep-th]

    2022 arXiv

  38. [46]

    M. D. Schwartz, (2021), 10.1162/99608f92.beeb1183, arXiv:2103.12226 [hep-ph]

    2021 arXiv

  39. [47]

    Dersy, M

    A. Dersy, M. D. Schwartz, and X. Zhang, Int. J. Data Sci. Math. Sci. 1, 135 (2024), arXiv:2206.04115 [cs.LG]

    2024 arXiv

  40. [48]

    Dersy, M

    A. Dersy, M. D. Schwartz, and A. Zhiboedov, JHEP 05, 200 (2024), arXiv:2308.09451 [hep-th]

    2024 arXiv

  41. [49]

    Cheung, A

    C. Cheung, A. Dersy, and M. D. Schwartz, SciPost Phys. 18, 040 (2025), arXiv:2408.04720 [hep-th]

    2025 arXiv

  42. [50]

    Demirtas, J

    M. Demirtas, J. Halverson, A. Maiti, M. D. Schwartz, and K. Stoner, Mach. Learn. Sci. Tech. 5, 015002 (2024), arXiv:2307.03223 [hep-th]

    2024 arXiv

  43. [51]

    Bhattacharya, J

    A. Bhattacharya, J. Cotler, A. Dersy, and M. D. Schwartz, Phys. Rev. D 110, 116023 (2024), arXiv:2402.18633 [hep-th]

    2024 arXiv

  44. [52]

    P. T. Komiske, E. M. Metodiev, and M. D. Schwartz, JHEP 01, 110 (2017), arXiv:1612.01551 [hep-ph]

    2017 arXiv

  45. [53]

    P. T. Komiske, E. M. Metodiev, B. Nachman, and M. D. Schwartz, JHEP 12, 051 (2017), arXiv:1707.08600 [hep- ph]

    2017 arXiv

  46. [54]

    Heimel, G

    T. Heimel, G. Kasieczka, T. Plehn, and J. M. Thompson, SciPost Phys. 6, 030 (2019), arXiv:1808.08979 [hep-ph]

    2019 arXiv

  47. [55]

    Qu and L

    H. Qu and L. Gouskos, Phys. Rev. D 101, 056019 (2020), arXiv:1902.08570 [hep-ph]

    2020 arXiv

  48. [56]

    S. Gong, Q. Meng, J. Zhang, H. Qu, C. Li, S. Qian, W. Du, Z.-M. Ma, and T.-Y. Liu, JHEP 07, 030 (2022), arXiv:2201.08187 [hep-ph]

    2022 arXiv

  49. [57]

    Bright-Thonney, B

    S. Bright-Thonney, B. Nachman, and J. Thaler, Phys. Rev. D 110, 014029 (2024), arXiv:2311.07652 [hep-ph]

    2024 arXiv

  50. [58]

    E. M. Metodiev, J. Thaler, and R. Wynne, Phys. Rev. D 110, 055012 (2024), arXiv:2312.00119 [hep-ph]

    2024 arXiv

  51. [59]

    Desai, B

    K. Desai, B. Nachman, and J. Thaler, Phys. Rev. D 110, 116013 (2024), arXiv:2407.11284 [hep-ph]

    2024 arXiv

  52. [60]

    Gambhir, A

    R. Gambhir, A. Osathapan, and J. Thaler, Phys. Rev. D 110, 074020 (2024), arXiv:2403.08854 [hep-ph]

    2024 arXiv

  53. [61]

    Heimel, N

    T. Heimel, N. Huetsch, R. Winterhalder, T. Plehn, and A. Butter, SciPost Phys. 17, 129 (2024), arXiv:2310.07752 [hep-ph]

    2024 arXiv

  54. [62]

    Badger et al

    S. Badger et al. , SciPost Phys. 14, 079 (2023), arXiv:2203.07460 [hep-ph]

    2023 arXiv

  55. [63]

    Forte, L

    S. Forte, L. Garrido, J. I. Latorre, and A. Piccione, JHEP 05, 062 (2002), arXiv:hep-ph/0204232

    2002 arXiv

  56. [64]

    R. D. Ball et al. (NNPDF), JHEP 04, 040 (2015), arXiv:1410.8849 [hep-ph]

    2015 arXiv

  57. [65]

    D. Liu, C. Sun, and J. Gao, JHEP 08, 088 (2022), arXiv:2201.06586 [hep-ph]. 12

    2022 arXiv

  58. [66]

    Badger and J

    S. Badger and J. Bullock, JHEP 06, 114 (2020), arXiv:2002.07516 [hep-ph]

    2020 arXiv

  59. [67]

    Aylett-Bullock, S

    J. Aylett-Bullock, S. Badger, and R. Moodie, JHEP 08, 066 (2021), arXiv:2106.09474 [hep-ph]

    2021 arXiv

  60. [68]

    Fedkevych, C

    O. Fedkevych, C. K. Khosa, S. Marzani, and F. Sforza, Phys. Rev. D 107, 034032 (2023), arXiv:2202.05082 [hep- ph]

    2023 arXiv

  61. [69]

    Calisto, R

    F. Calisto, R. Moodie, and S. Zoia, JHEP 07, 124 (2024), arXiv:2312.02067 [hep-ph]

    2024 arXiv

  62. [70]

    Halverson and F

    J. Halverson and F. Ruehle, Physical Review D 99, 046015 (2019)

    2019

  63. [71]

    Gukov, J

    S. Gukov, J. Halverson, F. Ruehle, and P. Su lkowski, Machine Learning: Science and Technology 2, 025035 (2021)

    2021

  64. [72]

    Halverson, A

    J. Halverson, A. Maiti, and K. Stoner, Machine Learn- ing: Science and Technology 2, 035002 (2021)

    2021

  65. [73]

    Halverson, B

    J. Halverson, B. Nelson, and F. Ruehle, Journal of High Energy Physics 2019, 1 (2019)

    2019

  66. [74]

    Carifio, J

    J. Carifio, J. Halverson, D. Krioukov, and B. D. Nelson, Journal of High Energy Physics 2017, 1 (2017)

    2017

  67. [75]

    V. A. Smirnov, Analytic Tools for Feynman Integrals, Springer Tracts in Modern Physics, Vol. 250 (Springer, 2012)

    2012

  68. [76]

    Anastasiou and K

    C. Anastasiou and K. Melnikov, Nucl. Phys. B 646, 220 (2002), arXiv:hep-ph/0207004

    2002 arXiv

  69. [77]

    Driesse, G

    M. Driesse, G. U. Jakobsen, G. Mogull, J. Plefka, B. Sauer, and J. Usovitsch, Phys. Rev. Lett. 132, 241402 (2024), arXiv:2403.07781 [hep-th]

    2024 arXiv

  70. [78]

    Z. Bern, E. Herrmann, R. Roiban, M. S. Ruf, A. V. Smirnov, V. A. Smirnov, and M. Zeng, JHEP 10, 023 (2024), arXiv:2406.01554 [hep-th]

    2024

  71. [79]

    Hsevo: Ele- vating automatic heuristic design with diversity-driven harmony search and genetic algorithm using llms,

    P. V. T. Dat, L. Doan, and H. T. T. Binh, “Hsevo: Ele- vating automatic heuristic design with diversity-driven harmony search and genetic algorithm using llms,” (2024), arXiv:2412.14995 [cs.NE]

    2024 arXiv

  72. [80]

    Evolution of heuristics: Towards efficient automatic algorithm design using large language model,

    F. Liu, X. Tong, M. Yuan, X. Lin, F. Luo, Z. Wang, Z. Lu, and Q. Zhang, “Evolution of heuristics: Towards efficient automatic algorithm design using large language model,” (2024), arXiv:2401.02051 [cs.NE]

    2024 arXiv

  73. [81]

    Reevo: Large language mod- els as hyper-heuristics with reflective evolution,

    H. Ye, J. Wang, Z. Cao, F. Berto, C. Hua, H. Kim, J. Park, and G. Song, “Reevo: Large language mod- els as hyper-heuristics with reflective evolution,” (2024), arXiv:2402.01145 [cs.NE]

    2024 arXiv

  74. [82]

    Uber: Uncertainty-based evolution with large lan- guage models for automatic heuristic design,

    Z. Chen, Z. Zhou, Y. Lu, R. Xu, L. Pan, and Z. Lan, “Uber: Uncertainty-based evolution with large lan- guage models for automatic heuristic design,” (2024), arXiv:2412.20694 [cs.NE]

    2024 arXiv

  75. [83]

    Monte carlo tree search for comprehensive exploration in llm-based automatic heuristic design,

    Z. Zheng, Z. Xie, Z. Wang, and B. Hooi, “Monte carlo tree search for comprehensive exploration in llm-based automatic heuristic design,” (2025), arXiv:2501.08603 [cs.AI]

    2025 arXiv

  76. [84]

    Funbo: Dis- covering acquisition functions for bayesian optimization with funsearch,

    V. Aglietti, I. Ktena, J. Schrouff, E. Sgouritsa, F. J. R. Ruiz, A. Malek, A. Bellot, and S. Chiappa, “Funbo: Dis- covering acquisition functions for bayesian optimization with funsearch,” (2024), arXiv:2406.04824 [cs.LG]

    2024 arXiv

  77. [85]

    Understanding the importance of evolutionary search in automated heuristic design with large language models,

    R. Zhang, F. Liu, X. Lin, Z. Wang, Z. Lu, and Q. Zhang, “Understanding the importance of evolutionary search in automated heuristic design with large language models,” (2024), arXiv:2407.10873 [cs.NE]

    2024 arXiv

  78. [86]

    Gituliar, V

    O. Gituliar, V. Magerya, and A. Pikelner, JHEP 06, 099 (2018), arXiv:1803.09084 [hep-ph]

    2018 arXiv

  79. [87]

    Magerya and A

    V. Magerya and A. Pikelner, JHEP 12, 026 (2019), arXiv:1910.07522 [hep-ph]

    2019 arXiv

  80. [88]

    Maheria, Semi- and Fully-Inclusive Phase-Space Inte- grals at Four Loops, Ph.D

    V. Maheria, Semi- and Fully-Inclusive Phase-Space Inte- grals at Four Loops, Ph.D. thesis, Hamburg U. (2022)

    2022

  81. [89]

    O. V. Tarasov, Phys. Rev. D 54, 6479 (1996), arXiv:hep- th/9606018

    1996

  82. [90]

    Refining integration-by- parts reduction of feynman integrals with machine learn- ing,

    M. von Hippel and M. Wilhelm, “Refining integration-by- parts reduction of feynman integrals with machine learn- ing,” (2025), arXiv:2502.05121 [hep-th]

    2025 arXiv

  83. [91]

    Minicpm: Unveiling the potential of small lan- guage models with scalable training strategies,

    S. Hu, Y. Tu, X. Han, C. He, G. Cui, X. Long, Z. Zheng, Y. Fang, Y. Huang, W. Zhao, X. Zhang, Z. L. Thai, K. Zhang, C. Wang, Y. Yao, C. Zhao, J. Zhou, J. Cai, Z. Zhai, N. Ding, C. Jia, G. Zeng, D. Li, Z. Liu, and M. Sun, “Minicpm: Unveiling the potential of small lan- guage mo...

    2024 arXiv

  84. [92]

    Qwen technical report,

    J. Bai, S. Bai, Y. Chu, Z. Cui, K. Dang, X. Deng, Y. Fan, W. Ge, Y. Han, F. Huang, B. Hui, L. Ji, M. Li, J. Lin, R. Lin, D. Liu, G. Liu, C. Lu, K. Lu, J. Ma, R. Men, X. Ren, X. Ren, C. Tan, S. Tan, J. Tu, P. Wang, S. Wang, W. Wang, S. Wu, B. Xu, J. Xu, A. Yang, H. Yang, J. Yan...

    2023 arXiv

  85. [93]

    Below lists some key hyperparameters and their selection criteria: Model: The CPM-2B model [86] was selected due to its superior performance in code generation tasks

    Hyperparameter Selection The selection of hyperparameters significantly impacts the performance and outcomes of FunSearch. Below lists some key hyperparameters and their selection criteria: Model: The CPM-2B model [86] was selected due to its superior performance in code gener...

  86. [94]

    The initial seeding employed was the Laporta priority 2

    Brief Analysis of the Evolution Process In the experiments, we executed FunSearch for one- loop IBP reductions a total of 10 times, with each run comprising 5000 epochs. The initial seeding employed was the Laporta priority 2. During each epoch, the pri- ority function was eva...

    2000

  87. [95]

    , y - node [1]]) ** 2 + node [0] ** 2 + node [1] ** 2 for x , y in n o d e _ t a r g e t s ) ) Its expanded form is given by: F0(I(ni);Itargetj(tji)) = − "X n2 i + max j X i (ni − tji)2 !# . (B1) This result indicates that even with multiple target points, the algorithm was ab...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.