Pith. sign in

REVIEW 4 major objections 5 minor 65 references

A Jacobian-splitting sampler claims unbiased high-dimensional integration for nuclear many-body states up to 624 dimensions.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-05 00:42 UTC pith:BMK7SSAV

load-bearing objection A sensible two-stage importance-sampling idea with a sign error in Eq. (19) that makes the printed algorithm internally inconsistent. the 4 major comments →

arxiv 2608.00529 v1 pith:BMK7SSAV submitted 2026-08-01 nucl-th quant-ph

Schr\"{o}dinger Generator for High-Dimensional Integration and Sampling on Quantum Many-Body States

Lin-Jing Jiang , Fu Ma , Pei Li , Kai-Jia Sun , Guo-Liang Ma , Yu-Gang Ma This is my paper
classification nucl-th quant-ph MSC 65C0565D3068T07 PACS 02.70.Uu
keywords Monte Carlo samplingStochastic integrationQuantum many-body statesNormalizing flowsAdaptive importance samplingShort-range correlationsNuclear structureSchrödinger Generator
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper introduces the Schrödinger Generator, a Monte Carlo framework for high-dimensional integration and sampling that learns an invertible coordinate transformation optimized for low variance. The method decomposes the transformation's Jacobian into two parts: an adaptive per-dimension map (VEGAS-style) that learns marginal structure, and a normalizing-flow transformation that captures cross-dimension correlations; a final resampling step keeps the estimator unbiased even when the learned transform is imperfect. The authors apply it to fermionic Slater-determinant densities and to effective correlated nucleon densities for 20Ne, 96Zr, and 208Pb, reaching D=624 dimensions. They report that generated samples faithfully reproduce two-nucleon short-range correlations and that integration uncertainty drops by more than an order of magnitude relative to conventional Monte Carlo. If correct, the framework offers a scalable, physically interpretable way to sample strongly correlated quantum many-body states.

Core claim

The paper proposes the Schrödinger Generator (SG), an integration and sampling method that writes the target integral as a Monte Carlo estimate over a learned invertible coordinate transformation x=x(z), and decomposes that transformation's Jacobian into J_SG=J_A J_F. J_A is a VEGAS-style adaptive map optimized per dimension to minimize estimator variance; J_F is a normalizing flow built from rational quadratic spline coupling layers that learns non-factorizable correlations. After training, samples are weighted by f(x(z)) J_SG(z)/p0(z) and resampled, which the paper argues guarantees unbiased sampling even when the learned transformation is imperfect. On fermionic Slater-determinant densiti

What carries the argument

The load-bearing object is the factorized Jacobian J_SG = J_A J_F of the composed transformation z→y→x. J_A is obtained by iteratively adapting one-dimensional grid widths so each bin carries equal averaged weight, following the VEGAS variance-minimization condition; it captures the marginal structure of the target. J_F is the Jacobian of a bidirectional rational-quadratic-spline coupling-layer normalizing flow whose triangular structure makes log-determinants cheap; it captures cross-dimension correlations. The final resampling weights w_SG = f(x(z)) J_SG / p0(z) turn the learned proposal into an unbiased estimator of the integral and of exact samples from f. The decomposition is what lets

Load-bearing premise

The entire training objective hinges on J_F being the actual relative-volume change of the map from adaptive coordinates y to output coordinates x; the paper's Eq. (19) writes J_F as the inverse of that forward determinant, and if that equation is taken literally the log-Jacobian term in the loss has the wrong sign.

What would settle it

Compute the left and right sides of Eqs. (18)–(19) for one trained RQS layer: evaluate |det ∂c_k/∂x^(k-1)| by automatic differentiation and compare with the inverse expression in Eq. (19). If they differ, retrain with the correct sign and check whether the KL loss and the reported integrals still match; alternatively, on the D=2 target of Eq. (9), test the sign of the gradient of Eq. (23) with respect to a single bin parameter.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • If SG works as claimed, event-by-event initial nucleon configurations for heavy-ion simulations—including 208Pb with 624 spatial dimensions—can be generated with short-range correlations intact, which is precisely the input needed for imaging nuclear shapes from high-energy collisions.
  • The resampling step makes the method unbiased even when the learned transformation is imperfect, so users can trust integral estimates without requiring the flow to be fully converged.
  • On the tested fermionic and nuclear targets, SG reduces integration uncertainty by more than an order of magnitude relative to uniform and VEGAS Monte Carlo, so the same accuracy can be reached with substantially fewer samples.
  • Because all target modes are already present in the product of marginals fed to the flow, SG avoids mode collapse, a failure mode that limits standalone normalizing flows on sharply peaked or constrained distributions.
  • The framework is not tied to nuclear physics; the same Jacobian-split plus resampling recipe applies to Bayesian evidence integrals and other high-dimensional stochastic integration problems.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The Jacobian split suggests a more general recipe: use any cheap factorized sampler for the marginals, any expressive flow for residual correlations, and importance resampling to correct leftovers—so SG can be recombined with stratified sampling, continuous normalizing flows, or flow matching, as the paper itself notes.
  • A clean diagnostic is to run SG on a product distribution, where VEGAS is already optimal: the learned J_F should stay near 1, isolating exactly what the flow component contributes beyond the adaptive map.
  • The nuclear tests use a factorized two-body correlation ansatz; applying the same pipeline to ab initio or path-integral nuclear wavefunctions would show whether the flow captures genuine many-body correlations beyond the pair-wise form.
  • Because the final estimator is an importance-weighted average, SG inherits the usual risk of weight collapse if the learned transformation misses a disconnected mode; the resampling step fixes bias only when the proposal overlaps the target support.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes a hybrid Monte Carlo integration/sampling framework, the "Schrödinger Generator" (SG), which combines a VEGAS-style adaptive coordinate transformation (T_A) with a normalizing-flow transformation (T_F). The Jacobian is decomposed as J_SG = J_A J_F, and a final importance-resampling step is claimed to yield unbiased estimates even when the learned map is imperfect. The authors benchmark SG on analytic fermionic Slater-determinant densities in D=2,6,9,12 and on correlated nuclear densities for 20Ne, 96Zr, and 208Pb (D=60,288,624), reporting low KL divergence, reduced integration uncertainty, and faithful reproduction of two-nucleon correlation functions.

Significance. If the method and numerical claims are correct, SG would be a useful addition to high-dimensional importance sampling: it directly addresses the known 'phantom peak' problem of VEGAS by adding a flow-based correlation layer, and the final resampling step gives an unbiased estimator. The analytic benchmarks for fermionic densities are appropriate sanity checks, and the ambition to apply the approach to >600-dimensional nuclear many-body sampling is noteworthy. However, the central Jacobian definition in Eq. (19) is internally inconsistent with the forward map defined in Eq. (18), and the nuclear correlation result in Fig. 5 is a self-consistency check rather than an independent physical prediction. The paper also omits the architectural and training details needed to reproduce the D=624 result. With correction of the Jacobian issue, clarification of the sequential 208Pb procedure, and a more honest framing of the nuclear benchmark, the contribution would be publishable; as written, the central claims are not adequately supported.

major comments (4)
  1. [§4, Eq. (19) vs Eq. (18); Algorithm 1] The flow is defined in Eq. (18) as x = c_K ∘ ... ∘ c_1(y), so the standard chain rule gives J_F = ∏_{k=1}^K |det(∂c_k/∂x^(k-1))|. Eq. (19) instead writes J_F as the reciprocal of this product. The same J_F enters the KL loss in Eq. (23) and the importance weights in Algorithm 1 (w = f J_F J_A). With the reciprocal, log J_F has the wrong sign, so minimizing Eq. (23) is not minimizing D_KL(q||f), and the estimator in Eq. (7) is not unbiased. This is a load-bearing inconsistency: either Eq. (19) must be corrected to the forward determinant, or Eqs. (18) and (20) must be redefined as inverse layers. Please state which convention the implementation actually uses and provide code or a detailed derivation to disambiguate.
  2. [§2, 'Sampling quantum states of atomic nucleus', Eqs. (11)-(13), Fig. 5] The reported reproduction of g(Δr) in Fig. 5 is a self-consistency check, not an independent validation: g(Δr) is an input factor in the target density f_A (Eq. (13)), and SG samples are drawn from f_A. Recovering the input from samples demonstrates that the sampler can represent its own target, but it does not by itself show that 'short-range correlations among nucleons in finite nucleus are faithfully reproduced' as claimed in the Abstract. Please either compare with an observable not used to define f_A, or explicitly reframe the claim as a demonstration of sampling accuracy on a correlated test density rather than a physical prediction.
  3. [§2, 'To enhance performance for A=208'] The sequential SG for 208Pb is described as learning f_104 and using f_104^(1) f_104^(2) as the proposal. f_104 is a density on R^312; the product lives on R^624 only if the two factors act on disjoint coordinate sets. The split of the 208 nucleons between the two factors is not specified, and no argument is given that the ratio f_208/(f_104^(1) f_104^(2)) has bounded importance weights. Without this, the D=624 result in Fig. 5 is not well defined. Please define the exact decomposition and report effective sample size or weight-variance diagnostics.
  4. [§4, Algorithm 1 and Fig. 5] No architecture or training hyperparameters are given: number of coupling layers K, spline bins J, interval bound B, FCNN width/depth, learning rate, batch size, number of iterations, VEGAS grid intervals N_A, smoothing/compression parameters, or resampling implementation. The code is not provided. In a methods paper whose central claims are 'stable and scalable' up to D=624, this is insufficient for reproducibility and for assessing whether optimization actually converged. Please include a full hyperparameter table and release code.
minor comments (5)
  1. [Eq. (10)] The normalization prefactor of f_R(x) should be checked: as written the denominator contains b^2, but dimensional consistency with the exponential and sinh arguments suggests a different power of b. Please clarify.
  2. [§2, 'D=60,288,624'] The text says '96Ze' but Table 1 and standard notation use '96Zr'. Please correct the typo.
  3. [Eq. (12)] The deformed radius R(θ,φ) is used in Eq. (12) but the deformation parameters β_l and the meaning of the sum over l,m are not defined in the text. Please add explicit definitions.
  4. [Fig. 4] The KL divergence and integration uncertainty are shown without error bars or multiple independent runs. Please report statistical uncertainties on these estimates, especially for SG.
  5. [Eqs. (8)-(9) and text after Eq. (10)] N is used both for the number of particles and for the normalization factor. Please use distinct symbols to avoid ambiguity.

Circularity Check

1 steps flagged

One nuclear demonstration recovers the input correlation function by construction; the core sampling method is otherwise self-contained.

specific steps
  1. self definitional [Section 2, Eqs. (11)-(13) and Fig. 5]
    "For demonstration, we adopt an effective correlated density of the form fA(r1, ...,rA) = 1/N ∏ ρ_WS(ri) ∏ g(ri,rj) ... The two-body correlation function is parameterized as [53] g(∆r) = 1−e^{−α(∆r)^2}(γ+ Σ ζ_i(∆r)^{i+1}) ... Figure 5 presents the two-nucleon correlation functions, g(∆r), extracted from SG-generated samples ... Dashed lines denote the analytical input correlations, while solid lines represent results from samples generated in D=60,288,and 624 dimensions."

    Eq. (11) defines the target density f_A using the two-body factor g(r_i,r_j) given in Eq. (13), with parameters taken from Refs [54,55]. Figure 5 then compares the g(Δr) extracted from SG samples to 'the analytical input correlations' — the same function inserted into the target. Any unbiased sampler of f_A must, by the law of large numbers, produce pair-correlation statistics converging to this input g. The agreement is therefore a built-in consistency check, not an external prediction. The abstract's phrasing 'short-range correlations ... faithfully reproduced' restates the input rather than independently deriving SRC. This does not invalidate the SG method's unbiasedness argument or the D=2,6,9,12 analytic benchmarks, but the nuclear demonstration is circular as evidence for reproducing

full rationale

The core SG algorithm is not circular: the estimator (Eqs. 3-7) follows from the standard importance-sampling identity, and the resampling step guarantees unbiasedness for any positive tractable Jacobian. The analytic fermionic benchmarks in D=2,6,9,12 are independent checks against closed-form Slater-determinant densities. The only load-bearing reduction is the nuclear demonstration: the target density in Eq. (11) is constructed by inserting the correlation function g(Δr) from Eq. (13), and Fig. 5 shows that SG samples recover that same input. This is a round-trip consistency check, so the headline 'short-range correlations are faithfully reproduced' is true by construction and not evidence that SG predicts SRC. Separately, Eq. (19) defines J_F as the reciprocal of the forward-flow determinant implied by Eq. (18); this is an internal sign/correctness inconsistency rather than a circularity, so it is not counted in the circularity score.

Axiom & Free-Parameter Ledger

3 free parameters · 6 axioms · 0 invented entities

No new physical entities, particles, or mediators are introduced; the 'Schrödinger Generator' is an algorithmic construct. The main burden is carried by untested expressiveness assumptions about the flow and by the closed-loop nuclear benchmark.

free parameters (3)
  • Flow architecture hyperparameters (K, J, B, FCNN widths)
    Not reported in the paper. The ability to learn non-factorizable correlations in D=624 depends on these choices.
  • VEGAS grid intervals N_A and smoothing/compression parameters
    Not reported. They control the adaptive marginal map J_A and the resulting variance reduction.
  • Sequential SG split for 208Pb = 104+104
    The proposal f_104 * f_104 is a hand-chosen decomposition. No criterion or diagnostic is given for why this split keeps importance weights manageable.
axioms (6)
  • domain assumption The target density f(x) is available for pointwise evaluation, at least up to normalization.
    The weights in Eq. (7) and the loss in Eq. (23) require evaluating f at every flow output. For many-body quantum targets this is not always cheap or even tractable.
  • standard math The VEGAS variance-minimization condition and iterative grid update are valid.
    Borrowed from Lepage's VEGAS literature; the paper treats Eq. (16) and the iterative update as given.
  • ad hoc to paper The RQS coupling flow is expressive and trainable enough to capture non-factorizable correlations up to D=624.
    No theorem, complexity bound, or convergence analysis is provided; the claim rests entirely on the reported experiments.
  • ad hoc to paper The product of marginal densities contains all true modes, so mode collapse is avoided.
    Stated in the introduction. It is false in general for distributions whose modes are defined by correlations rather than by marginal peaks.
  • ad hoc to paper For A=208, the sequential proposal f_104 * f_104 keeps importance weights bounded.
    No variance bound, effective sample size, or weight diagnostic is reported for the 208Pb case.
  • domain assumption The phenomenological correlated density in Eq. (11) represents a nuclear quantum many-body state.
    The paper samples f_A with Woods-Saxon single-particle factors and an inserted g(Delta r). This is a model, not a solution of the nuclear many-body Schrödinger equation, and the abstract's 'quantum many-body states' overstates its status.

pith-pipeline@v1.3.0-alltime-deepseek · 13526 in / 16437 out tokens · 195875 ms · 2026-08-05T00:42:30.318687+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Schr\"{o}dinger Generator for High-Dimensional Integration and Sampling on Quantum Many-Body States." pith.science (2026). https://pith.science/paper/BMK7SSAV

@misc{pith2026260800529,
  author       = {Pith},
  title        = {Pith review of: Schr\"odinger Generator for High-Dimensional Integration and Sampling on Quantum Many-Body States},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BMK7SSAV}},
  note         = {Machine review of arXiv:2608.00529}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Integration and sampling in high dimensions are among central challenges in modern science and technology, underlying applications ranging from quantum many-body physics to Bayesian inference and artificial intelligence. Although conventional Monte Carlo methods are formally scalable, their efficiency deteriorates rapidly in the presence of strong correlations or sharp features in high-dimensional configuration space. Here we introduce a new framework, termed the Schr"odinger Generator, for integration and sampling based on the explicit optimization of coordinate transformations. The method decomposes the total Jacobian into two complementary components, including an adaptive map that minimizes estimator variance by learning the marginal structure in each dimension, and a normalizing-flow-based transformation that captures non-factorizable correlations in the target distribution. A final resampling step guarantees unbiased sampling even when the learned transformation is imperfect. We demonstrate stable and scalable performance for nuclear quantum many-body states in dimensions exceeding 600. Short-range correlations among nucleons in finite nucleus are faithfully reproduced. The framework offers a physically transparent approach to high-dimensional stochastic integration and sampling, opening new possibilities for simulations of complex quantum systems.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

65 extracted references · 28 canonical work pages · 6 internal anchors

  1. [1]

    John Wiley & Sons, Bognor Regis, United Kingdom (2008)

    Kalos, M.H., Whitlock, P.A.: Monte Carlo Methods. John Wiley & Sons, Bognor Regis, United Kingdom (2008)

    work page 2008

  2. [2]

    Springer, Dordrecht, Nether- lands (1964)

    Hammersley, J.M., Handscomb, D.C.: Monte Carlo Methods. Springer, Dordrecht, Nether- lands (1964)

    work page 1964

  3. [3]

    SpringerBriefs in Physics

    Joseph, A.: Markov Chain Monte Carlo Methods in Quantum Field Theories: A Modern Primer. SpringerBriefs in Physics. 10 Springer, Dordrecht, Netherlands (2019). https://doi.org/10.1007/978-3-030-46044-0

    work page doi:10.1007/978-3-030-46044-0 2019

  4. [4]

    Amoroso, S.,et al.: Challenges in Monte Carlo Event Generator Software for High- Luminosity LHC. Comput. Softw. Big Sci. 5(1), 12 (2021) https://doi.org/10.1007/ s41781-021-00055-1 arXiv:2004.13687 [hep- ph]

    Pith/arXiv arXiv 2021

  5. [5]

    JHEP 08, 089 (2022) https://doi.org/10.1007/ JHEP08(2022)089 arXiv:2112.09588 [hep-ex]

    Aad, G.,et al.: Modelling and com- putational improvements to the simula- tion of single vector-boson plus jet pro- cesses for the ATLAS experiment. JHEP 08, 089 (2022) https://doi.org/10.1007/ JHEP08(2022)089 arXiv:2112.09588 [hep-ex]

    Pith/arXiv arXiv 2022

  6. [6]

    Shou, Q.Y., Ma, Y.G., Zhang, S.,et al.: Prop- erties of qcd matter: a review of selected results from alice experiment. Nucl. Sci. Tech.35, 219 (2024) https://doi.org/10. 1007/s41365-024-01583-2

    work page 2024

  7. [7]

    Sun, K.J.: Femtoscopy reveals how (anti- )deuteron is formed at the LHC. Nucl. Sci. Tech.37, 64 (2026) https://doi.org/10.1007/ s41365-026-01906-5

    work page 2026

  8. [8]

    Bleicher, M.: Nucleosynthesis in the little bang. Nucl. Sci. Tech.35, 129 (2024) https: //doi.org/10.1007/s41365-024-01477-3

    work page doi:10.1007/s41365-024-01477-3 2024

  9. [9]

    Nature635(8037), 67–72 (2024) https://doi.org/10.1038/s41586-024-08097-2 arXiv:2401.06625 [nucl-ex]

    Abdulhamid, M.I.,et al.: Imaging shapes of atomic nuclei in high-energy nuclear col- lisions. Nature635(8037), 67–72 (2024) https://doi.org/10.1038/s41586-024-08097-2 arXiv:2401.06625 [nucl-ex]

    Pith/arXiv arXiv 2024

  10. [10]

    Bally, B., Bender, M., Giacalone, G., Som` a, V.: Evidence of the triaxial structure of 129Xe at the Large Hadron Collider. Phys. Rev. Lett.128(8), 082301 (2022) https: //doi.org/10.1103/PhysRevLett.128.082301 arXiv:2108.09578 [nucl-th]

    Pith/arXiv arXiv 2022

  11. [11]

    Giacalone, G.: Beyond axial symmetry: high-energy collisions unveil the ground- state shape of 238u. Nucl. Sci. Tech. 35, 218 (2024) https://doi.org/10.1007/ s41365-024-01582-3

    work page 2024

  12. [12]

    Schenke, B.: Violent collisions can reveal hexadecapole deformation of nuclei. Nucl. Sci. Tech.35, 115 (2024) https://doi.org/10. 1007/s41365-024-01509-y

    work page 2024

  13. [13]

    Jia, J., Giacalone, G., Bally, B.,et al.: Imag- ing the initial condition of heavy-ion colli- sions and nuclear structure across the nuclide chart. Nucl. Sci. Tech.35, 220 (2024) https: //doi.org/10.1007/s41365-024-01589-w

    doi:10.1007/s41365-024-01589-w 2024

  14. [14]

    Statistical Applications in Genetics and Molecular Biology7(1) https://doi.org/ 10.2202/1544-6115.1400

    Hobolth, A., Uyenoyama, M.K., Wiuf, C.: Importance sampling for the infinite sites model. Statistical Applications in Genetics and Molecular Biology7(1) https://doi.org/ 10.2202/1544-6115.1400

    work page arXiv

  15. [15]

    IntechOpen, London (2011)

    Mode, C.J.: Applications of Monte Carlo Methods in Biology, Medicine and Other Fields of Science. IntechOpen, London (2011). https://doi.org/10.5772/634 . https: //doi.org/10.5772/634

    work page doi:10.5772/634 2011

  16. [16]

    The Journal of Chemical Physics23(2), 356–359 (1955) https://doi

    Rosenbluth, M.N., Rosenbluth, A.W.: Monte carlo calculation of the average extension of molecular chains. The Journal of Chemical Physics23(2), 356–359 (1955) https://doi. org/10.1063/1.1741967

    doi:10.1063/1.1741967 1955

  17. [17]

    Astrophysics and Space Science86(2), 419–435 (1982) https://doi.org/10.1007/BF00683346

    MacGillivray, H.T., Dodd, R.J.: Monte-carlo simulations of galaxy systems. Astrophysics and Space Science86(2), 419–435 (1982) https://doi.org/10.1007/BF00683346

    work page doi:10.1007/bf00683346 1982

  18. [18]

    VDM Verlag Dr

    Szirmay-Kalos, L.: Monte Carlo Methods in Global Illumination: Photo-realistic Render- ing with Randomization. VDM Verlag Dr. M¨ uller, Saarbr¨ ucken, Germany (2008)

    work page 2008

  19. [19]

    Physics in Medicine & Biology51(13), 287 (2006) https: //doi.org/10.1088/0031-9155/51/13/R17

    Rogers, D.W.O.: Fifty years of monte carlo simulations for medical physics. Physics in Medicine & Biology51(13), 287 (2006) https: //doi.org/10.1088/0031-9155/51/13/R17

    work page doi:10.1088/0031-9155/51/13/r17 2006

  20. [20]

    The Wiley Finance Series

    J¨ ackel, P.: Monte Carlo Methods in Finance. The Wiley Finance Series. John Wiley & Sons, Chichester, UK (2002)

    work page 2002

  21. [21]

    Webber, B.R.: Monte Carlo Simulation of Hard Hadronic Processes. Ann. Rev. Nucl. Part. Sci.36, 253–286 (1986) https://doi. org/10.1146/annurev.ns.36.120186.001345 11

    arXiv 1986

  22. [22]

    Buckley, A.,et al.: General-purpose event generators for LHC physics. Phys. Rept.504, 145–233 (2011) https://doi.org/10.1016/j. physrep.2011.03.005 arXiv:1101.2599 [hep- ph]

    Pith/arXiv arXiv 2011

  23. [23]

    Lepage, G.P.: A New Algorithm for Adap- tive Multidimensional Integration. J. Com- put. Phys.27, 192 (1978) https://doi.org/10. 1016/0021-9991(78)90004-9

    work page 1978

  24. [24]

    Lepage, G.P.: Adaptive multidimensional integration: VEGAS enhanced. J. Comput. Phys.439, 110386 (2021) https://doi.org/ 10.1016/j.jcp.2021.110386 arXiv:2009.05112 [physics.comp-ph]

    work page arXiv 2021

  25. [25]

    Jadach, S.: Foam: A General purpose cellu- lar Monte Carlo event generator. Comput. Phys. Commun.152, 55–100 (2003) https: //doi.org/10.1016/S0010-4655(02)00755-5 arXiv:physics/0203033

    work page internal anchor Pith review Pith/arXiv arXiv 2003

  26. [26]

    Tabak, E.G., Vanden-Eijnden, E.: Den- sity estimation by dual ascent of the log-likelihood. Commun. Math. Sci.8(1), 217–233 (2010) https://doi.org/10.4310/cms. 2010.v8.n1.a11

    doi:10.4310/cms 2010

  27. [27]

    Tabak, E.G., Turner, C.V.: A Family of Non- parametric Density Estimation Algorithms. Commun. Pure Appl. Math.66(2), 145–164 (2013) https://doi.org/10.1002/cpa.21423

    doi:10.1002/cpa.21423 2013

  28. [28]

    In: Bengio, Y., LeCun, Y

    Dinh, L., Krueger, D., Bengio, Y.: NICE: non-linear independent components estima- tion. In: Bengio, Y., LeCun, Y. (eds.) 3rd International Conference on Learning Repre- sentations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Workshop Track Proceedings (2015). http://arxiv.org/abs/1410.8516

    Pith/arXiv arXiv 2015

  29. [29]

    Bendavid, J.: Efficient Monte Carlo Inte- gration Using Boosted Decision Trees and Generative Deep Neural Networks (2017) arXiv:1707.00028 [hep-ph]

    Pith/arXiv arXiv 2017

  30. [30]

    SciPost Phys.9, 053 (2020) https://doi.org/10.21468/SciPostPhys.9.4

    Klimek, M.D., Perelstein, M.: Neural Network-Based Approach to Phase Space Integration. SciPost Phys.9, 053 (2020) https://doi.org/10.21468/SciPostPhys.9.4. 053 arXiv:1810.11509 [hep-ph]

    Pith/arXiv arXiv 2020

  31. [31]

    Event Generation and Statistical Sampling for Physics with Deep Generative Models and a Density Information Buffer

    Otten, S., Caron, S., Swart, W., Beekveld, M., Hendriks, L., Leeuwen, C., Podareanu, D., Austri, R., Verheyen, R.: Event Gener- ation and Statistical Sampling for Physics with Deep Generative Models and a Den- sity Information Buffer. Nature Commun. 12(1), 2985 (2021) https://doi.org/10.1038/ s41467-021-22616-z arXiv:1901.00875 [hep- ph]

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  32. [32]

    SciPost Phys

    Butter, A., Plehn, T., Winterhalder, R.: How to GAN LHC Events. SciPost Phys. 7(6), 075 (2019) https://doi.org/10.21468/ SciPostPhys.7.6.075 arXiv:1907.03764 [hep- ph]

    Pith/arXiv arXiv 2019

  33. [33]

    Song, J., Zhao, S., Ermon, S.: A-nice- mc: Adversarial training for mcmc (2018) arXiv:1706.07561 [stat.ML]

    Pith/arXiv arXiv 2018

  34. [34]

    Levy, D., Hoffman, M.D., Sohl-Dickstein, J.: Generalizing Hamiltonian Monte Carlo with Neural Networks (2017) arXiv:1711.09268 [stat.ML]

    Pith/arXiv arXiv 2017

  35. [35]

    Hoffman, M., Sountsov, P., Dillon, J.V., Langmore, I., Tran, D., Vasudevan, S.: Neutra-lizing bad geometry in hamiltonian monte carlo using neural transport (2019) arXiv:1903.03704 [stat.CO]

    Pith/arXiv arXiv 2019

  36. [36]

    Gao, C., H¨ oche, S., Isaacson, J., Krause, C., Schulz, H.: Event Gener- ation with Normalizing Flows. Phys. Rev. D101(7), 076002 (2020) https: //doi.org/10.1103/PhysRevD.101.076002 arXiv:2001.10028 [hep-ph]

    Pith/arXiv arXiv 2020

  37. [37]

    SciPost Phys.8(4), 069 (2020) https: //doi.org/10.21468/SciPostPhys.8.4.069 arXiv:2001.05478 [hep-ph]

    Bothmann, E., Janßen, T., Knobbe, M., Schmale, T., Schumann, S.: Exploring phase space with Neural Importance Sam- pling. SciPost Phys.8(4), 069 (2020) https: //doi.org/10.21468/SciPostPhys.8.4.069 arXiv:2001.05478 [hep-ph]

    Pith/arXiv arXiv 2020

  38. [38]

    SciPost Phys

    Winterhalder, R., Magerya, V., Villa, E., Jones, S.P., Kerner, M., Butter, A., Hein- rich, G., Plehn, T.: Targeting multi-loop inte- grals with neural networks. SciPost Phys. 12(4), 129 (2022) https://doi.org/10.21468/ SciPostPhys.12.4.129 arXiv:2112.09145 [hep- ph] 12

    Pith/arXiv arXiv 2022

  39. [39]

    SciPost Phys

    Heimel, T., Winterhalder, R., Butter, A., Isaacson, J., Krause, C., Maltoni, F., Matte- laer, O., Plehn, T.: MadNIS - Neural multi- channel importance sampling. SciPost Phys. 15(4), 141 (2023) https://doi.org/10.21468/ SciPostPhys.15.4.141 arXiv:2212.06172 [hep- ph]

    Pith/arXiv arXiv 2023

  40. [40]

    SciPost Phys

    Badger, S.,et al.: Machine learning and LHC event generation. SciPost Phys. 14(4), 079 (2023) https://doi.org/10.21468/ SciPostPhys.14.4.079 arXiv:2203.07460 [hep-ph]

    Pith/arXiv arXiv 2023

  41. [41]

    SciPost Phys.13(3), 047 (2022) https: //doi.org/10.21468/SciPostPhys.13.3.047 arXiv:2205.01697 [hep-ph]

    Verheyen, R.: Event Generation and Den- sity Estimation with Surjective Normalizing Flows. SciPost Phys.13(3), 047 (2022) https: //doi.org/10.21468/SciPostPhys.13.3.047 arXiv:2205.01697 [hep-ph]

    Pith/arXiv arXiv 2022

  42. [42]

    SciPost Phys

    Heimel, T., Huetsch, N., Maltoni, F., Mattelaer, O., Plehn, T., Winterhalder, R.: The MadNIS reloaded. SciPost Phys. 17(1), 023 (2024) https://doi.org/10.21468/ SciPostPhys.17.1.023 arXiv:2311.01548 [hep- ph]

    Pith/arXiv arXiv 2024

  43. [43]

    SciPost Phys.18(1), 017 (2025) https: //doi.org/10.21468/SciPostPhys.18.1.017 arXiv:2408.01486 [hep-ph]

    Heimel, T., Mattelaer, O., Plehn, T., Win- terhalder, R.: Differentiable MadNIS-Lite. SciPost Phys.18(1), 017 (2025) https: //doi.org/10.21468/SciPostPhys.18.1.017 arXiv:2408.01486 [hep-ph]

    Pith/arXiv arXiv 2025

  44. [44]

    IEEE Transac- tions on Pattern Analysis and Machine Intel- ligence43(11), 3964–3979 (2021) https://doi

    Kobyzev, I., Prince, S.J.D., Brubaker, M.A.: Normalizing flows: An introduction and review of current methods. IEEE Transac- tions on Pattern Analysis and Machine Intel- ligence43(11), 3964–3979 (2021) https://doi. org/10.1109/tpami.2020.2992934

    work page arXiv 2021

  45. [45]

    Relaxing Bijectivity Constraints with Continuously Indexed Normalising Flows

    Cornish, R., Caterini, A.L., Deligiannidis, G., Doucet, A.: Relaxing Bijectivity Con- straints with Continuously Indexed Normal- ising Flows (2021). https://arxiv.org/abs/ 1909.13833

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  46. [46]

    The Annals of Mathemati- cal Statistics22(1), 79–86 (1951) https://doi

    Kullback, S., Leibler, R.A.: On Information and Sufficiency. The Annals of Mathemati- cal Statistics22(1), 79–86 (1951) https://doi. org/10.1214/aoms/1177729694

    arXiv 1951

  47. [47]

    Li, P., Zhou, B., Ma, G.-L.: Identifyingα- Cluster Configurations in Ne20 via Ultra- central Ne+Ne Collisions. Phys. Rev. Lett. 136(8), 082302 (2026) https://doi.org/10. 1103/tffz-8q1m arXiv:2504.04688 [nucl-th]

    arXiv 2026

  48. [48]

    Li, P., Sun, K.-J., Zhou, B., Ma, G.- L.: Unmasking short-range correlations via initial-state fluctuations in relativistic heavy- ion collisions (2025) arXiv:2511.23293 [nucl- th]

    work page arXiv 2025

  49. [49]

    Wang, H.-C., Li, S.-J., Liu, L.-M., Xu, J., Ren, Z.-Z.: Deformation probes for light nuclei in their collisions at relativistic ener- gies. Phys. Rev. C110(3), 034909 (2024) https://doi.org/10.1103/PhysRevC.110. 034909 arXiv:2409.02452 [nucl-th]

    Pith/arXiv arXiv 2024

  50. [50]

    De Vries, H., De Jager, C.W., De Vries, C.: Nuclear charge and magnetization density distribution parameters from elastic electron scattering. Atom. Data Nucl. Data Tabl. 36, 495–536 (1987) https://doi.org/10.1016/ 0092-640X(87)90013-1

    work page 1987

  51. [51]

    Jia, J., Giacalone, G., Zhang, C.: Sep- arating the Impact of Nuclear Skin and Nuclear Deformation in High- Energy Isobar Collisions. Phys. Rev. Lett.131(2), 022301 (2023) https: //doi.org/10.1103/PhysRevLett.131.022301 arXiv:2206.10449 [nucl-th]

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  52. [52]

    Rybczynski, M., Stefanek, G., Broniowski, W., Bozek, P.: GLISSANDO 2 : GLauber Initial-State Simulation AND mOre. . . , ver

    work page

  53. [53]

    Comput. Phys. Commun.185, 1759– 1772 (2014) https://doi.org/10.1016/j.cpc. 2014.02.016 arXiv:1310.5475 [nucl-th]

    Pith/arXiv arXiv 2014

  54. [54]

    Cruz-Torres, R., Schmidt, A., Miller, G.A., Weinstein, L.B., Barnea, N., Weiss, R., Piasetzky, E., Hen, O.: Short range correla- tions and the isospin dependence of nuclear correlation functions. Phys. Lett. B785, 304–308 (2018) https://doi.org/10.1016/j. physletb.2018.07.069 arXiv:1710.07966 [nucl- th]

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  55. [55]

    Alvioli, M., Drescher, H.-J., Strikman, M.: 13 A Monte Carlo generator of nucleon configu- rations in complex nuclei including Nucleon- Nucleon correlations. Phys. Lett. B680, 225–230 (2009) https://doi.org/10.1016/j. physletb.2009.08.067 arXiv:0905.2670 [nucl- th]

    work page internal anchor Pith review Pith/arXiv arXiv 2009

  56. [56]

    Pudliner, B.S., Pandharipande, V.R., Carl- son, J., Pieper, S.C., Wiringa, R.B.: Quantum Monte Carlo calculations of nuclei with A <= 7. Phys. Rev. C56, 1720–1750 (1997) https://doi.org/10.1103/PhysRevC.56.1720 arXiv:nucl-th/9705009

    Pith/arXiv arXiv 1997

  57. [57]

    Pyretzidis, K., Lejarza, J.J.M., Rodrigo, G.: Unlocking Multi-Dimensional Integration with Quantum Adaptive Importance Sam- pling (2025) arXiv:2506.19965 [quant-ph]

    Pith/arXiv arXiv 2025

  58. [58]

    Chen, R.T.Q., Rubanova, Y., Bettencourt, J., Duvenaud, D.: Neural Ordinary Differential Equations (2018) arXiv:1806.07366 [cs.LG]

    Pith/arXiv arXiv 2018

  59. [59]

    Bothmann, E., Janßen, T., Knobbe, M., Schmitzer, B., Sinz, F.: Monte carlo event generation with continuous nor- malizing flows. Phys. Rev. Lett., (2026) https://doi.org/10.1103/jt6y-h375

    work page doi:10.1103/jt6y-h375 2026

  60. [60]

    arXiv preprint arXiv:2210.02747 (2022)

    Lipman, Y., Chen, R.T., Ben-Hamu, H., Nickel, M., Le, M.: Flow matching for generative modeling. arXiv preprint arXiv:2210.02747 (2022)

    Pith/arXiv arXiv 2022

  61. [61]

    Albergo, M.S., Vanden-Eijnden, E.: Build- ing Normalizing Flows with Stochastic Inter- polants (2022) arXiv:2209.15571 [cs.LG]

    Pith/arXiv arXiv 2022

  62. [62]

    Liu, X., Gong, C., Liu, Q.: Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow (2022) arXiv:2209.03003 [cs.LG]

    Pith/arXiv arXiv 2022

  63. [63]

    Albergo, M.S., Boffi, N.M., Vanden-Eijnden, E.: Stochastic Interpolants: A Unifying Framework for Flows and Diffusions (2025) arXiv:2303.08797 [cs.LG]

    Pith/arXiv arXiv 2025

  64. [64]

    ACM Transactions on Graphics (ToG)38(5), 1–19 (2019)

    M¨ uller, T., McWilliams, B., Rousselle, F., Gross, M., Nov´ ak, J.: Neural importance sampling. ACM Transactions on Graphics (ToG)38(5), 1–19 (2019)

    work page 2019

  65. [65]

    Advances in neural information processing systems32 (2019) arXiv:1906.04032 [stat.ML] 14

    Durkan, C., Bekasov, A., Murray, I., Papa- makarios, G.: Neural spline flows. Advances in neural information processing systems32 (2019) arXiv:1906.04032 [stat.ML] 14

    Pith/arXiv arXiv 2019