Pith. sign in

REVIEW 4 major objections 5 minor 47 references

Deciphering the Small-Angle Scattering of Polydisperse Hard Spheres using Deep Learning

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A variational autoencoder trained on molecular-dynamics data learns a bidirectional mapping between the small-angle scattering curve of polydisperse hard spheres and the volume fraction and polydispersity that define them, and it beats…

desk verdict A useful VAE surrogate for polydisperse hard-sphere scattering with a real generator advantage over PY; fix the WCA calibration and the missing inverse baseline, then it's a solid submission. read the letter →

arxiv 2507.09400 v2 pith:7UDABYYI submitted 2025-07-12 cond-mat.soft cond-mat.mtrl-sci

Lijie Ding , Changwoo Do This is my paper
classification cond-mat.softcond-mat.mtrl-sci
keywords polydispersehardspheressmall-anglescatteringvariationalautoencoderdeeplearningfunctionvolumefractionpolydispersitystructurefactor
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 introduces a deep-learning model that treats small-angle scattering of polydisperse hard spheres as a two-way translation problem: given the volume fraction η and polydispersity σ, produce the scattering curve I(Q); given I(Q), recover η and σ. The model is a variational autoencoder whose encoder compresses each 100-point log-intensity curve into three latent variables, with two converter networks linking those latent variables to (η, σ). The authors train and test it on scattering curves computed from molecular-dynamics simulations of truncated-shifted Lennard-Jones particles meant to stand in for hard spheres, covering η ∈ (0, 0.5) and σ ∈ (0, 0.3). They report that the generative direction is markedly more accurate than the traditional Percus-Yevick approximation and its β correction, and that the inferrer recovers η and σ with relative errors well under one percent on held-out curves. If the simulation model faithfully represents hard spheres, the result matters because it offers a fast, accurate replacement for analytic approximations that degrade at high density and polydispersity.

What carries the argument

The load-bearing object is a variational autoencoder with a three-dimensional latent space. The encoder turns a 100-point log I(Q) curve into latent mean and variance vectors (μ, s), and the decoder reconstructs the curve from a sampled latent variable z = μ + s ⊙ ε. Two small converter networks complete the mapping: converter 1 sends (η, σ) to the latent variables, so converter 1 plus the decoder form the generator; converter 2 sends latent variables back to (η, σ), so the encoder plus converter 2 form the inferrer. The whole network is trained first as an autoencoder, then each converter is trained and fine-tuned with the decoder or encoder frozen and released in turn.

What would settle it

Run event-driven molecular dynamics of true hard spheres (no soft potential) at the same (η, σ) grid, compute the same I(Q), and compare against the neural generator: if the generated curves deviate from true-hard-sphere scattering at high η and σ beyond the reported MSE, or if the inferrer's error on these curves exceeds the claimed sub-percent level, the central claim fails. A cheaper check is to compute the low-Q region from smaller and larger simulation boxes to see whether the 23,328-particle results are converged at Q ≈ 3.

Watch

Extended reading notes

Core claim

The central claim is that a single variational autoencoder learns an accurate bidirectional map between the scattering function I(Q) and the system parameters (η, σ) for polydisperse hard spheres, so that one trained network serves both as a generator of scattering curves and as an inferrer of physical parameters. On the held-out test set, the neural generator's mean-square error in log I(Q) stays small across the full parameter range, whereas the Percus-Yevick error grows sharply with η and σ and remains large even after the β correction for polydispersity. The inferrer returns η and σ with relative errors of about 0.36% and 0.40% for the uniform-distribution case, and similarly small errors for normal and lognormal size distributions.

Load-bearing premise

The paper assumes that the truncated-shifted Lennard-Jones fluid, with ε_LJ = 100 and T = 1.0, faithfully behaves like polydisperse hard spheres across the entire tested volume-fraction range without any calibration of the mapping between Lennard-Jones diameters and true hard-sphere diameters.

Editorial extensions

If this is right

  • Within the trained regime (η up to 0.5, σ up to 0.3, Q from 3 to 13), the generator can replace Percus-Yevick with β correction as a forward model for fitting small-angle scattering data of polydisperse hard-sphere-like colloids.
  • The inferrer extracts volume fraction and polydispersity from a measured curve in a single forward pass, making near-real-time analysis possible once the mean particle size is fixed by a dilute-sample normalization.
  • The same architecture works for uniform, normal, and lognormal diameter distributions without any change in the network, suggesting the mapping is learned from the data rather than from a closed-form assumption about the distribution.
  • Because the latent space is only three-dimensional, the paper's singular-value analysis implies the studied scattering curves live on a low-dimensional manifold, which is why the VAE can compress them so aggressively.

Reading between the lines

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

  • If the truncated-shifted Lennard-Jones fluid does not exactly match true hard-sphere structure factors at high η, the reported accuracy numbers are an upper bound on what an experimentalist would see; a calibration of effective hard-sphere diameters, or retraining on event-driven hard-sphere simulations, would test this directly.
  • The same VAE framework could be extended to infer the full size distribution rather than just two parameters, by enlarging the latent space and training on a richer set of distribution moments.
  • The low-dimensional latent representation suggests that other polydisperse interacting colloids, such as charged or rod-like particles, may also admit compact scattering representations; the paper names these as future targets but does not demonstrate them.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper trains a variational autoencoder (VAE) with two auxiliary converter networks on molecular dynamics (MD) scattering data to obtain a bidirectional mapping between the small-angle scattering function I(Q) and the system parameters (volume fraction eta and polydispersity sigma) for polydisperse hard spheres. A dataset of 5,000 MD simulations of a truncated-shifted Lennard-Jones fluid with uniform diameter polydispersity is generated, analyzed by singular value decomposition, and split into 4,000 training and 1,000 testing samples. The trained generator is compared with the Percus-Yevick (PY) approximation and the PY approximation with beta correction, and the inferrer is evaluated on the held-out set, reporting relative errors of 0.36% and 0.40% for eta and sigma. The procedure is repeated for normal and lognormal size distributions, with similar reported accuracy. The abstract claims that the generator produces scattering functions with significantly higher accuracy than PY/PY-beta and that the inferrer extracts parameters with much higher accuracy than traditional model functions.

Significance. If the underlying simulation is a faithful representation of polydisperse hard spheres, the bidirectional VAE surrogate would be a practically useful tool for small-angle scattering analysis: it provides one-pass inference, is trained on a reproducible dataset, and is validated for three common size distributions. The availability of code, data, and trained models in a public repository is a clear strength, as is the SVD-based motivation for low-dimensional latent structure. The generator benchmark against PY and PY-beta is a meaningful, if incomplete, external check. However, the significance of the central claim depends on a load-bearing identification between the simulated truncated-shifted LJ fluid and true polydisperse hard spheres, and that identification is not established in the manuscript.

major comments (4)
  1. [II A, Eq. (1)] The central claim that the model describes polydisperse hard spheres is not supported by the simulation protocol. The potential in Eq. (1) is a truncated-shifted Lennard-Jones (WCA-type) potential with epsilon_LJ=100 and T=1.0, which is only an approximate model of hard spheres; at finite temperature and finite epsilon, particles can interpenetrate, and the effective hard-sphere diameter differs from the nominal diameter D_i in a density- and polydispersity-dependent way. The manuscript never computes an effective hard-sphere diameter (for example via the Barker-Henderson or Weeks-Chandler-Andersen prescription), nor does it compare the simulated scattering functions against event-driven hard-sphere MD or another true hard-sphere benchmark. The PY and PY-beta baselines in Eqs. (4)-(5) are evaluated at the nominal eta computed from the nominal D_i, so the reported generator advantage could partly be an artifact of comparing soft-sphere data with hard-sphere models at a mismatched state point. I recommend adding an explicit calibration of the LJ-to-hard-sphere mapping and a validation against a genuine hard-sphere simulation, or reframing the claims to describe the WCA-type soft-sphere system actually simulated.
  2. [III C, Fig. 8] The inference superiority claim is not benchmarked against any traditional inverse approach. The abstract states that the inferrer extracts eta and sigma with much higher accuracy than traditional model functions, but the only validation shown is agreement with held-out data drawn from the same simulation pipeline that generated the training set. No comparison is made with, for example, least-squares fitting of I_PY_beta(Q) to the same scattering curves, or with a simple regression baseline on the same train/test split. The held-out accuracy therefore demonstrates interpolation fidelity of the VAE, but it does not by itself establish superiority over traditional model-based inference. I recommend adding an inverse baseline and reporting the comparison on identical test samples.
  3. [III B and III C, Figs. 6, 8, 10, 11] All accuracy claims are point estimates without uncertainty quantification. The MSE maps in Figs. 6 and 10, and the relative-error percentages in Figs. 8 and 11, are reported without error bars, bootstrap intervals, or repeated training runs with different seeds. Since the dataset is randomly split and the network training involves stochastic optimization, the reported 0.36%-0.40% errors could vary substantially across realizations. I request that the authors add repeated-seed experiments or bootstrap confidence intervals for the test-set errors and MSE values, so that the claim of statistically significant improvement over PY/PY-beta is properly supported.
  4. [II C and III C, Eq. (7)] The inferrer loss in Eq. (7) is written as the square of the sum of squared differences, namely [ (eta-eta')^2 + (sigma-sigma')^2 ]^2, but the text describes a mean square distance between input and output. If the fourth-power loss is intended, please state this explicitly and explain the choice; if it is a typo and the intended loss is (eta-eta')^2 + (sigma-sigma')^2, please correct Eq. (7). This matters because the reported training procedure and the resulting accuracy depend on the actual loss used.
minor comments (5)
  1. [III C] The relative error is defined as Err = |x - x'| / <x> for x in eta, sigma, but the denominator is the mean of the ground-truth values. For sigma values near zero (which are included in the sampling range sigma in U(0,0.3)), this relative error can become unstable or undefined; please report absolute errors or a scatterplot with residuals, and clarify how sigma=0 samples were handled.
  2. [Fig. 3(b) caption] The caption reads 'variation of scattering function for different polydispersity sigma = with volume fraction eta = 0.3'; the value of sigma is missing, and the sentence is incomplete.
  3. [III C] In the sentence defining Err, 'x in mu, sigma' should read 'x in eta, sigma'; the text uses the Greek letters mu and sigma, but the inferred variables are eta and sigma.
  4. [III A] The sentence 'we focus on the interaction peak at low Q and also cover the first first dip of sphere form factor' contains the duplicated word 'first'; additionally, the rationale for restricting Q to [3,13] could be stated more precisely (e.g., the location of the first form-factor minimum).
  5. [III] The phrase 'Without losing generosity' should be 'Without loss of generality'; also, the abstract and summary contain minor grammatical issues (e.g., 'produce' should be 'produces' in places).

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the forward data come from a stated MD model, the VAE is evaluated on a held-out split, and the Percus–Yevick comparison is an external analytical benchmark.

full rationale

The derivation chain is self-contained. The scattering data are produced by an explicitly stated simulation model (Eq. 1) and the scattering formula (Eq. 2), independent of the neural network. The VAE and converters are trained on a 4000/1000 split and evaluated on the 1000 held-out samples, so the inferrer's 0.36–0.51% relative errors are genuine held-out interpolation results, not fitted parameters renamed as predictions. The generator is benchmarked against analytical Percus–Yevick and β-correction formulas, which are external, parameter-free approximations, so outperforming them on simulation data is not circular. The one self-citation (Ref. 24) is used only to note prior use of the truncated-shifted LJ approximation; it is not load-bearing for the VAE results. The main weakness—that the LJ potential with ε=100, T=1 is never calibrated to true hard-sphere diameters—is an external-validity/correctness concern about what physical system the surrogate describes, not a step in which the output is equivalent to the input by construction. No circular step can be exhibited from the paper's own equations.

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

The paper adds no new physical entities. The three latent variables are computational coordinates inside the neural network, not claims about nature. The load-bearing input is the self-generated MD dataset and the architecture choices.

free parameters (5)
  • Latent dimension = 3
    Chosen by hand based on singular value decay (Fig. 4a); the two-dimensional physical parameter space motivates three latent coordinates, but no reconstruction-error study justifies this exact dimension.
  • Truncated-shifted Lennard-Jones parameters = epsilon_LJ = 100, T = 1.0, cutoff = 2^(1/6) * sigma_ij
    These constants determine the effective hard-sphere diameter and compressibility of the simulated fluid; their relationship to true hard-sphere volume fraction is not calibrated.
  • Q range and grid = 100 points in [3,13]
    Selected to cover the interaction peak and first form-factor dip; the choice excludes higher-order minima and conditions the inverse problem.
  • Dataset size and train/test split = 5000 simulations, 4000 training / 1000 testing
    The reported accuracy is an interpolation result; the network never extrapolates beyond the sampled eta and sigma ranges, and the size of the training set is an arbitrary practical choice.
  • Particle number = 23,328
    Chosen for the MD box; finite-size effects on low-Q scattering are not quantified, so the effective low-Q behavior may depend on this number.
assumptions (5)
  • domain assumption The truncated-shifted Lennard-Jones fluid at T=1.0 and epsilon_LJ=100 reproduces the equilibrium structure of polydisperse hard spheres over the studied volumetric and polydispersity ranges.
    Introduced in Section II A; the entire training dataset is generated under this proxy, so the learned mapping is only as valid as this equivalence.
  • standard math Equation (2), the normalized Debye scattering formula with hard-sphere form-factor amplitudes, correctly converts MD configurations into the small-angle scattering function in the chosen Q range.
    Invoked in Section II B; assumes single scattering, no background, no resolution smearing, and no multiple scattering, which the authors also acknowledge in the Summary.
  • domain assumption A three-dimensional latent representation captures all scattering information relevant to eta and sigma.
    Supported only by the visual singular-value decay in Fig. 4(a) and the smooth latent distributions in Fig. 7; no quantitative reconstruction-error analysis is provided.
  • domain assumption The inverse problem from I(Q) to (eta, sigma) is well-posed on noiseless simulated curves from the same distribution used for training.
    The 0.36% and 0.40% relative errors are measured on held-out simulations from the same pipeline; robustness to noise, resolution, and background is untested.
  • domain assumption For each size distribution type, the two parameters eta and sigma fully determine the scattering function, with no other hidden variables such as exact tail shape or finite-size sampling details.
    The paper parameterizes uniform, normal, and log-normal distributions by eta and sigma only and does not study sensitivity to other moments of the size distribution.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Deciphering the Small-Angle Scattering of Polydisperse Hard Spheres using Deep Learning." pith.science (2026). https://pith.science/paper/7UDABYYI

@misc{pith2026250709400,
  author       = {Pith},
  title        = {Pith review of: Deciphering the Small-Angle Scattering of Polydisperse Hard Spheres using Deep Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7UDABYYI}},
  note         = {Machine review of arXiv:2507.09400}
}
abstract

We introduce a deep learning approach for analyzing the scattering function of the polydisperse hard spheres system. We use a variational autoencoder-based neural network to learn the bidirectional mapping between the scattering function and the system parameters including the volume fraction and polydispersity. Such that the trained model serves both as a generator that produce scattering function from the system parameters, and an inferrer that extract system parameters from the scattering function. We first generate a scattering dataset by carrying out molecular dynamics simulation of the polydisperse hard spheres modeled by the truncated-shifted Lennard-Jones model, then analyze the scattering function dataset using singular value decomposition to confirm the feasibility of dimensional compression. Then we split the dataset into training and testing set and train our neural network on the training set only. Our generator model produce scattering function with significant higher accuracy comparing to the traditional Percus-Yevick approximation and $\beta$ correction, and the inferrer model can extract the volume fraction and polydispersity with much higher accuracy than traditional model functions.

Figures

Figures reproduced from arXiv: 2507.09400 by the authors.

Figure 1
Figure 1. FIG. 1. Scattering function of polydisperse hard sphere with [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The VAE34 is made of three main components: an encoder, a latent space, and a decoder. The encoder will compress the input scattering function I(Q), a 100 dimensional vector in to the 3 dimensional latent space, and the encoder will reconstruct the scattering function I ′ (Q) from the latent space. The encoder is consist of two convolutional layers, each with kernel size 9 and strides 2. The first one has 30 channel… view at source ↗
Figure 2
Figure 2. FIG. 2. Architecture of the neural network for mapping the scattering function [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figures from the paper (7 more)
Figure 3
Figure 3. Figure 3: FIG. 3. Examples of the scattering function [PITH_FULL_IMAGE:figures/full_fig_p004_3.png]
Figure 4
Figure 4. Figure 4: FIG. 4. Principal component analysis of the scattering func [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 6
Figure 6. Figure 6: FIG. 6. Mean square error [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: FIG. 7. Distribution of volume fraction [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 9
Figure 9. Figure 9: FIG. 9. Illustration of the three different size distribution of [PITH_FULL_IMAGE:figures/full_fig_p006_9.png]
Figure 10
Figure 10. Figure 10: FIG. 10. Mean square error [PITH_FULL_IMAGE:figures/full_fig_p007_10.png]
Figure 11
Figure 11. Figure 11: FIG. 11. Comparison between the system parameters ex [PITH_FULL_IMAGE:figures/full_fig_p008_11.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

47 extracted references · 39 canonical work pages

  1. [1]

    author author W. B. \ Russel , author W. Russel , author D. A. \ Saville ,\ and\ author W. R. \ Schowalter ,\ @noop title Colloidal dispersions \ ( publisher Cambridge university press ,\ year 1991 ) NoStop

    work page 1991

  2. [2]

    Lindner \ and\ author T

    author author P. Lindner \ and\ author T. Zemb ,\ @noop title Neutrons, X-rays and Light: Scattering Methods Applied to Soft Condensed Matter ,\ North-Holland Delta Series\ ( publisher Elsevier Science ,\ year 2002 ) NoStop

    work page 2002

  3. [3]

    author author M. Shibayama ,\ title title Small-angle neutron scattering on polymer gels: phase behavior, inhomogeneities and deformation mechanisms , \ @noop journal journal Polymer journal \ volume 43 ,\ pages 18--34 ( year 2011 ) NoStop

    work page 2011

  4. [4]

    author author S.-H. \ Chen ,\ title title Small angle neutron scattering studies of the structure and interaction in micellar and microemulsion systems , \ @noop journal journal Annual Review of Physical Chemistry \ volume 37 ,\ pages 351--399 ( year 1986 ) NoStop

    1986

  5. [5]

    Chu \ and\ author B

    author author B. Chu \ and\ author B. S. \ Hsiao ,\ title title Small-angle x-ray scattering of polymers , \ @noop journal journal Chemical reviews \ volume 101 ,\ pages 1727--1762 ( year 2001 ) NoStop

    2001

  6. [6]

    Debye ,\ title title Molecular-weight determination by light scattering

    author author P. Debye ,\ title title Molecular-weight determination by light scattering. \ @noop journal journal The Journal of Physical Chemistry \ volume 51 ,\ pages 18--32 ( year 1947 ) NoStop

    work page 1947

  7. [7]

    Sch \"a rtl ,\ @noop title Light scattering from polymer solutions and nanoparticle dispersions \ ( publisher Springer Science & Business Media ,\ year 2007 ) NoStop

    author author W. Sch \"a rtl ,\ @noop title Light scattering from polymer solutions and nanoparticle dispersions \ ( publisher Springer Science & Business Media ,\ year 2007 ) NoStop

    work page 2007

  8. [8]

    Mittelbach \ and\ author O

    author author R. Mittelbach \ and\ author O. Glatter ,\ title title Direct structure analysis of small-angle scattering data from polydisperse colloidal particles , \ @noop journal journal Applied Crystallography \ volume 31 ,\ pages 600--608 ( year 1998 ) NoStop

    work page 1998

Show all 47 references
  1. [9]

    \ Huang , author C.-H

    author author G.-R. \ Huang , author C.-H. \ Tung , author L. Porcar , author Y. Wang , author Y. Shinohara , author C. Do ,\ and\ author W.-R. \ Chen ,\ title title Model-free approach for profiling of polydisperse soft matter using small angle scattering , \ @noop journal jo...

    2023

  2. [10]

    Weyerich , author J

    author author B. Weyerich , author J. Brunner-Popela ,\ and\ author O. Glatter ,\ title title Small-angle scattering of interacting particles. ii. generalized indirect fourier transformation under consideration of the effective structure factor for polydisperse systems , \ @no...

    1999

  3. [11]

    Salgi \ and\ author R

    author author P. Salgi \ and\ author R. Rajagopalan ,\ title title Polydispersity in colloids: implications to static structure and scattering , \ @noop journal journal Advances in Colloid and Interface science \ volume 43 ,\ pages 169--288 ( year 1993 ) NoStop

    1993

  4. [12]

    Wagner \ and\ author A

    author author N. Wagner \ and\ author A. M. \ Woutersen ,\ title title The viscosity of bimodal and polydisperse suspensions of hard spheres in the dilute limit , \ @noop journal journal Journal of Fluid Mechanics \ volume 278 ,\ pages 267--287 ( year 1994 ) NoStop

    1994

  5. [13]

    Eckert , author M

    author author T. Eckert , author M. Schmidt ,\ and\ author D. de Las Heras ,\ title title Sedimentation path theory for mass-polydisperse colloidal systems , \ @noop journal journal The Journal of Chemical Physics \ volume 157 ( year 2022 ) NoStop

    2022

  6. [14]

    author author P. N. \ Pusey \ and\ author W. Van Megen ,\ title title Phase behaviour of concentrated suspensions of nearly hard colloidal spheres , \ @noop journal journal Nature \ volume 320 ,\ pages 340--342 ( year 1986 ) NoStop

    1986

  7. [15]

    author author J. K. \ Percus \ and\ author G. J. \ Yevick ,\ title title Analysis of classical statistical mechanics by means of collective coordinates , \ @noop journal journal Physical Review \ volume 110 ,\ pages 1 ( year 1958 ) NoStop

    1958

  8. [16]

    author author M. Wertheim ,\ title title Exact solution of the percus-yevick integral equation for hard spheres , \ @noop journal journal Physical Review Letters \ volume 10 ,\ pages 321 ( year 1963 ) NoStop

    1963

  9. [17]

    Katzav , author R

    author author E. Katzav , author R. Berdichevsky ,\ and\ author M. Schwartz ,\ title title Random close packing from hard-sphere percus-yevick theory , \ @noop journal journal Physical Review E \ volume 99 ,\ pages 012146 ( year 2019 ) NoStop

    2019

  10. [18]

    author author K. P. \ Murphy ,\ @noop title Machine learning: a probabilistic perspective \ ( publisher MIT press ,\ year 2012 ) NoStop

    2012

  11. [19]

    Carleo , author I

    author author G. Carleo , author I. Cirac , author K. Cranmer , author L. Daudet , author M. Schuld , author N. Tishby , author L. Vogt-Maranto ,\ and\ author L. Zdeborov \'a ,\ title title Machine learning and the physical sciences , \ @noop journal journal Reviews of Modern ...

    2019

  12. [20]

    Goodfellow , author Y

    author author I. Goodfellow , author Y. Bengio , author A. Courville ,\ and\ author Y. Bengio ,\ @noop title Deep learning ,\ Vol. volume 1 \ ( publisher MIT press Cambridge ,\ year 2016 ) NoStop

    2016

  13. [21]

    LeCun , author Y

    author author Y. LeCun , author Y. Bengio ,\ and\ author G. Hinton ,\ title title Deep learning , \ @noop journal journal nature \ volume 521 ,\ pages 436--444 ( year 2015 ) NoStop

    2015

  14. [22]

    \ Chang , author C.-H

    author author M.-C. \ Chang , author C.-H. \ Tung , author S.-Y. \ Chang , author J. M. \ Carrillo , author Y. Wang , author B. G. \ Sumpter , author G.-R. \ Huang , author C. Do ,\ and\ author W.-R. \ Chen ,\ title title A machine learning inversion scheme for determining int...

    2022

  15. [23]

    Ding , author Y

    author author L. Ding , author Y. Chen ,\ and\ author C. Do ,\ title title Machine-learning-informed scattering correlation analysis of sheared colloids , \ @noop journal journal Applied Crystallography \ volume 58 ,\ pages 992--999 ( year 2025 ) NoStop

    2025

  16. [24]

    \ Tung , author L

    author author C.-H. \ Tung , author L. Ding , author M.-C. \ Chang , author G.-R. \ Huang , author L. Porcar , author Y. Wang , author J.-M. Y. \ Carrillo , author B. G. \ Sumpter , author Y. Shinohara , author C. Do , et al. ,\ title title Scattering-based structural inversio...

    2025

  17. [25]

    \ Tung , author L

    author author C.-H. \ Tung , author L. Ding , author G.-R. \ Huang , author L. Porcar , author Y. Shinohara , author B. G. \ Sumpter , author C. Do ,\ and\ author W.-R. \ Chen ,\ title title Insights into distorted lamellar phases with small-angle scattering and machine learni...

    2025

  18. [26]

    Ding , author C.-H

    author author L. Ding , author C.-H. \ Tung , author B. G. \ Sumpter , author W.-R. \ Chen ,\ and\ author C. Do ,\ title title Machine learning inversion from scattering for mechanically driven polymers , \ @noop journal journal arXiv preprint arXiv:2410.05574 \ ( year 2024 ) NoStop

  19. [27]

    Ding , author C.-H

    author author L. Ding , author C.-H. \ Tung , author J. M. Y. \ Carrillo , author W.-R. \ Chen ,\ and\ author C. Do ,\ title title Machine learning inversion from small-angle scattering for charged polymers , \ @noop journal journal Digital Discovery \ ( year 2025 a ) NoStop

    2025

  20. [28]

    Ding , author C.-H

    author author L. Ding , author C.-H. \ Tung , author Z. Cao , author Z. Ye , author X. Gu , author Y. Xia , author W.-R. \ Chen ,\ and\ author C. Do ,\ title title Machine learning-assisted profiling of a kinked ladder polymer structure using scattering , \ @noop journal journ...

    2025

  21. [29]

    Ding , author C.-H

    author author L. Ding , author C.-H. \ Tung , author B. G. \ Sumpter , author W.-R. \ Chen ,\ and\ author C. Do ,\ title title Deciphering the scattering of mechanically driven polymers using deep learning , \ @noop journal journal Journal of Chemical Theory and Computation \ ...

    2025

  22. [30]

    Cranmer , author J

    author author K. Cranmer , author J. Brehmer ,\ and\ author G. Louppe ,\ title title The frontier of simulation-based inference , \ @noop journal journal Proceedings of the National Academy of Sciences \ volume 117 ,\ pages 30055--30062 ( year 2020 ) NoStop

    2020

  23. [31]

    author author D. P. \ Kingma , author M. Welling , et al. ,\ @noop title Auto-encoding variational bayes , \ ( year 2013 ) NoStop

    2013

  24. [32]

    Doersch ,\ title title Tutorial on variational autoencoders , \ @noop journal journal arXiv preprint arXiv:1606.05908 \ ( year 2016 ) NoStop

    author author C. Doersch ,\ title title Tutorial on variational autoencoders , \ @noop journal journal arXiv preprint arXiv:1606.05908 \ ( year 2016 ) NoStop

    2016 arXiv

  25. [33]

    author author M. P. \ Allen et al. ,\ title title Introduction to molecular dynamics simulation , \ @noop journal journal Computational soft matter: from synthetic polymers to proteins \ volume 23 ,\ pages 1--28 ( year 2004 ) NoStop

    2004

  26. [34]

    author author A. P. \ Thompson , author H. M. \ Aktulga , author R. Berger , author D. S. \ Bolintineanu , author W. M. \ Brown , author P. S. \ Crozier , author P. J. \ In't Veld , author A. Kohlmeyer , author S. G. \ Moore , author T. D. \ Nguyen , et al. ,\ title title Lamm...

    2022

  27. [35]

    author author J. E. \ Lennard-Jones ,\ title title Cohesion , \ @noop journal journal Proceedings of the Physical Society \ volume 43 ,\ pages 461 ( year 1931 ) NoStop

    1931

  28. [36]

    author author S. Nos \'e ,\ title title A unified formulation of the constant temperature molecular dynamics methods , \ @noop journal journal The Journal of chemical physics \ volume 81 ,\ pages 511--519 ( year 1984 ) NoStop

    1984

  29. [37]

    author author W. G. \ Hoover ,\ title title Canonical dynamics: Equilibrium phase-space distributions , \ @noop journal journal Physical review A \ volume 31 ,\ pages 1695 ( year 1985 ) NoStop

    1985

  30. [38]

    Guinier , author G

    author author A. Guinier , author G. Fournet , author C. B. \ Walker ,\ and\ author G. H. \ Vineyard ,\ @noop title Small-angle scattering of x-rays , \ ( year 1956 ) NoStop

    1956

  31. [39]

    author author L. S. \ Ornstein ,\ title title Accidental deviations of density and opalescence at the critical point of a single substance , \ @noop journal journal Proc. Akad. Sci. \ volume 17 ,\ pages 793 ( year 1914 ) NoStop

    1914

  32. [40]

    author author D. J. \ Kinning \ and\ author E. L. \ Thomas ,\ title title Hard-sphere interactions between spherical domains in diblock copolymers , \ @noop journal journal Macromolecules \ volume 17 ,\ pages 1712--1718 ( year 1984 ) NoStop

    1984

  33. [41]

    Paszke ,\ title title Pytorch: An imperative style, high-performance deep learning library , \ @noop journal journal arXiv preprint arXiv:1912.01703 \ ( year 2019 ) NoStop

    author author A. Paszke ,\ title title Pytorch: An imperative style, high-performance deep learning library , \ @noop journal journal arXiv preprint arXiv:1912.01703 \ ( year 2019 ) NoStop

    1912 arXiv

  34. [42]

    author author D. P. \ Kingma ,\ title title Adam: A method for stochastic optimization , \ @noop journal journal arXiv preprint arXiv:1412.6980 \ ( year 2014 ) NoStop

    2014 arXiv

  35. [43]

    Loshchilov \ and\ author F

    author author I. Loshchilov \ and\ author F. Hutter ,\ title title Sgdr: Stochastic gradient descent with warm restarts , \ @noop journal journal arXiv preprint arXiv:1608.03983 \ ( year 2016 ) NoStop

    2016 arXiv

  36. [44]

    author author I. T. \ Jolliffe \ and\ author J. Cadima ,\ title title Principal component analysis: a review and recent developments , \ @noop journal journal Philosophical transactions of the royal society A: Mathematical, Physical and Engineering Sciences \ volume 374 ,\ pag...

    2016

  37. [45]

    Golub \ and\ author W

    author author G. Golub \ and\ author W. Kahan ,\ title title Calculating the singular values and pseudo-inverse of a matrix , \ @noop journal journal Journal of the Society for Industrial and Applied Mathematics, Series B: Numerical Analysis \ volume 2 ,\ pages 205--224 ( year...

    1965

  38. [46]

    Imhof \ and\ author J

    author author A. Imhof \ and\ author J. Dhont ,\ title title Experimental phase diagram of a binary colloidal hard-sphere mixture with a large size ratio , \ @noop journal journal Physical review letters \ volume 75 ,\ pages 1662 ( year 1995 ) NoStop

    1995

  39. [47]

    Schneider , author D

    author author J. Schneider , author D. Karrer , author J. Dhont ,\ and\ author R. Klein ,\ title title The pair-distribution function and light-scattered intensities for charged rod-like macromolecules in solution , \ @noop journal journal The Journal of chemical physics \ vol...

    1987

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.