Cubic-Equivariant Neural Density Functional Theory for Three-Dimensional Lattice Fluids

Alessandro Simon; Jens Weimar; Martin Oettel

Cubic-Equivariant Neural Density Functional Theory for Three-Dimensional Lattice Fluids

T0 review · 0 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read A network learns the one-body direct-correlation functional of three-dimensional lattice hard cubes from full profiles and improves bulk and hard-wall predictions over the analytic lattice functional.

desk verdict A solid, honest demonstration that full-profile convolutional neural DFT works in 3D for lattice fluids; the main claims are supported by independent GCMC data, and the only real weakness is the absence of released code and data. read the letter →

arxiv 2608.08137 v1 pith:IJYZT53N submitted 2026-08-08 cond-mat.stat-mech

Jens Weimar , Martin Oettel , Alessandro Simon This is my paper · ORCID

classification cond-mat.stat-mech

keywords neuraldensityfunctionaltheoryclassicallatticefluidshardcubesconvolutionalnetworkcubicequivariancedirectcorrelationgrandcanonicalMonteCarlo

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 constructs a neural classical density functional for parallel hard cubes on a simple cubic lattice, acting directly on complete three-dimensional density profiles instead of pre-extracted local windows. The network represents the one-body direct-correlation functional $c^{(1)}[\rho]$ as a fully convolutional map with a $7\times 7\times 7$ receptive field, projected onto all 48 rotations and reflections of the cubic point group so the density-to-$c^{(1)}$ map is exactly equivariant. A Bernoulli mask on the output sites keeps full-profile training practical. Against independent simulations, the learned functional reproduces the homogeneous equation of state and planar hard-wall density profiles substantially better than the analytic Lafuente-Cuesta lattice functional, while its accuracy for the anisotropic pair distribution is direction-dependent. The authors conclude that full-profile neural density-functional calculations in three dimensions are feasible, with accurate training data, thermodynamic consistency, and pair-structure fidelity as the main challenges for continuum extensions.

What carries the argument

The load-bearing construction is the fully convolutional network with a finite first layer of 32 cubic-symmetrized $7\times 7\times 7$ kernels followed by pointwise residual channel mixing. Averaging each kernel over the 48-element cubic point group collapses its 343 entries to linear combinations of 20 cubic-orbit density sums, guaranteeing equivariance without data augmentation. The second key piece is the Bernoulli-masked mean-squared-error loss over profile sites: the forward pass sees entire unexpanded profiles, while only a randomly retained subset of output sites contributes to each gradient update, restoring stochastic variation without storing overlapping windows. The network is trained on profiles from grand-canonical Monte Carlo in randomized external potentials, using the pointwise inversion $c^{(1)}_{\mathrm{GCMC}} = \ln\rho + \beta V_{\mathrm{ext}} - \beta\mu$ as the supervised target.

What would settle it

Train the same pipeline with a $9\times 9\times 9$ (or larger) first-layer kernel on the same training corpus and compare wall contact peaks, directional pair distributions, and the equation of state against GCMC at matched bulk densities. A material and systematic improvement in accuracy with the larger field would show that the $7\times 7\times 7$ truncation, not training data or optimization, is the limiting assumption; alternatively, directly estimating $c^{(1)}$ sensitivity to density perturbations beyond three lattice spacings from GCMC would settle the same point.

Watch

Extended reading notes

Core claim

On its own terms, the paper claims that a purely data-driven $c^{(1)}[\rho]$ can be learned directly on full $50\times 50\times 50$ lattice fields and, once solved self-consistently in the Euler equation, yields accurate equilibrium densities for a three-dimensional lattice fluid. Concretely, over the trained packing-fraction range the neural equation of state matches GCMC with RMSE $1.99\times 10^{-4}$, compared with $1.36\times 10^{-2}$ for the analytic Lafuente-Cuesta functional, and the neural wall profile tracks the contact peak and oscillations more closely than the analytic one. The test-particle pair distribution is reproduced in its principal shells by both functionals, with neither uniformly better across crystallographic directions. The paper interprets the bulk and wall gains as the practical demonstration, and the mixed structural results as evidence that accurate one-body profiles do not automatically imply accurate pair correlations.

Load-bearing premise

The assumption that a $7\times 7\times 7$ density neighborhood around each site is enough to determine its direct correlation, so any correlation longer than three lattice spacings lies outside what the network can represent.

Editorial extensions

If this is right

Solving the learned Euler equation for a new external potential yields an equilibrium density directly from full profiles, so wall, cavity, and confinement predictions require no window extraction or reassembly.
The homogeneous equation of state from the network reproduces GCMC packing fractions to RMSE $1.99\times 10^{-4}$ over the trained range, versus $1.36\times 10^{-2}$ for the analytic functional.
The Bernoulli-masked full-profile training scheme should carry over to other lattice and, after gridding, continuum models, provided training profiles are sufficiently accurate and diverse.
The test-particle comparison shows that accurate bulk thermodynamics and one-body profiles do not guarantee accurate pair structure, motivating explicit pair-correlation regularization or sum-rule constraints in future functionals.
The method demonstrates some extrapolation beyond the training density cutoff, but the mean packing-fraction cutoff of 0.60 remains a stated controlled limitation.

Reading between the lines

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

Beyond the paper: replacing the cubic projection with O(3)- or E(3)-equivariant convolutions would be the natural continuum counterpart, and the same full-profile training scheme could make three-dimensional hard-sphere functionals practical.
Beyond the paper: learning an equivariant scalar excess free energy instead of $c^{(1)}$ would enforce a symmetric functional Jacobian and path-independent thermodynamics, addressing a limitation the paper names.
Beyond the paper: because the receptive field is the only controlled locality restriction, a systematic sweep of kernel sizes on identical training data would directly quantify how much longer-ranged direct correlations matter for bulk and wall observables.

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.

Desk Editor's Note

read the letter

The one thing you should know: this paper actually delivers a 3D full-profile neural c(1) functional that beats the analytical LC functional on bulk EOS and wall profiles, and the validation against GCMC is clean. The second thing: the lack of released code and training data is the only substantial problem; the physics and the ML pipeline are sound.

What's new: the Bernoulli-masked full-profile training plus exact Oh symmetrization of first-layer kernels. The convolutional reformulation was already in Glitsch et al. by the same group, but here they make it practical in 3D without window extraction, and the equivariance projection is exact. The 393-profile training set from randomized potentials is careful, and the independent GCMC checks for EOS, walls, and test-particle g(r) give the claims real teeth. The RMSE numbers (1.99e-4 vs 1.36e-2) are convincing.

Soft spots: the c(1) target is fitted to GCMC data, so agreement with simulation is partly interpolation. That's not a flaw in itself, because the validation observables are not training labels, and the test-particle comparison is genuinely independent. The finite 7x7x7 receptive field is a real limitation, but the authors disclose it and the results suggest it captures the relevant correlations for the tested observables. Pair-structure results are honestly mixed; they don't overclaim. The bigger issue is reproducibility: no code, weights, or data are released, so no one can verify the actual numbers without reimplementation. That warrants a conditional rather than unconditional accept.

Bottom line: for someone working in classical DFT or machine-learned functionals, this is worth reading. It deserves a serious referee, and the main request should be to release the trained model and the training pipeline.

Recommendation: send it to peer review, with a request for code and data as a condition of acceptance.

Referee Report

0 major / 6 minor

Summary. This paper introduces a fully convolutional neural approximation to the one-body direct-correlation functional c^(1)[ρ] for parallel hard cubes of side length three on a three-dimensional simple cubic lattice. The network is trained on 393 GCMC density profiles generated in randomized external potentials, using a Bernoulli mask on output sites to make full-profile training practical and a projection of the first-layer kernels onto the 48-element cubic point group to enforce exact equivariance. The learned functional is inserted into the Euler–Lagrange equation and solved self-consistently for the homogeneous equation of state, a planar hard-wall slit, and the anisotropic pair distribution around a fixed particle, and the results are compared with independent GCMC data and with the Lafuente–Cuesta lattice functional. The central reported results are a large improvement in the homogeneous packing-fraction isotherm (RMSE 1.99e-4 versus 1.36e-2) and an accurate wall density profile, together with a more nuanced, direction-dependent pair-structure comparison.

Significance. If the results are taken at face value, this is a significant methodological advance in learned classical density functional theory: it is one of the first demonstrations of full three-dimensional profile training and inference, and it shows that convolutional evaluation combined with stochastic output masking is practical at 50^3 field sizes. The data pipeline is careful: long GCMC production runs, twenty replicas per training field, an explicit validation split, independent GCMC checks for the equation of state, wall, and test-particle problems, and replica-resampling error propagation. The authors also report limitations honestly, including the absence of a guaranteed free-energy functional, the finite 7x7x7 receptive field, the training-density cutoff, and the mixed pair-structure accuracy. The finite receptive field is a legitimate architectural constraint, but the paper's own wall and equation-of-state results suggest that it does not undermine the central claims for the tested observables.

minor comments (6)

[§IV.C and Fig. 4] The bulk packing fraction for the test-particle comparison is reported as η_b = 0.345 in the text but as η_b = 0.3 in the Fig. 4 caption; because the neural and Lafuente–Cuesta chemical potentials are matched to this value, please correct the discrepancy and state explicitly which value was used.
[§IV.A and §V] The RMSE values 1.99e-4 and 1.36e-2 are quoted as being evaluated over the 'common supported range', but that range is not defined; please specify the density interval, the number of comparison points, and how the common range is determined.
[§IV.A and Fig. 1] The text states that the GCMC data span βμ ∈ [−8, 1.5] and discusses extrapolation beyond η = 0.60, but the figure panels show βμ only down to 0; please extend the axes or explicitly state the plotted range so that the extrapolation claim is visible and checkable.
[§III.B] There is a typo in 'we restrict ourself to the convolutional counterpart'; this should read 'we restrict ourselves'.
[§III.C] The exclusion of target sites with c^(1) ≤ −10 is described as removing hard-wall sites and regions with insufficient statistics, but its possible effect on predicted low-density regions is not discussed; please add a sentence explaining how this threshold may influence profiles where the true c^(1) is below −10.
[Data and code availability] The manuscript contains no data, code, or trained-model availability statement; for a machine-learning functional paper, please state whether the training profiles, GCMC data, trained weights, and evaluation scripts will be released.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the neural c^(1) is fitted to GCMC targets, but the claimed EOS, wall-profile, and test-particle results are validated against independent GCMC data not used as training labels.

full rationale

The paper's derivation chain is self-contained and does not reduce its central claims to its inputs. The neural functional is trained by supervised regression on c^(1) targets obtained from Eq. (6), which is an exact inversion of the Euler-Lagrange equation for GCMC profiles in randomized external potentials. The validation observables — homogeneous equation of state (Sec. IV.A), planar hard-wall profile (Sec. IV.B), and test-particle pair distribution (Sec. IV.C) — are computed by solving Eq. (5) self-consistently with the trained c^(1)_theta and are compared with fresh GCMC data that are not used as loss labels. Equation (6) defines the training targets but does not by construction determine the solved density profiles; the wall and test-particle results require nontrivial self-consistent iteration and can fail, as the mixed pair-structure results show. The architecture relies in part on the authors' prior convolutional formulation (Ref. [36]), but the present paper fully specifies the network and the prior work is not invoked as an unverified uniqueness or existence theorem. The finite 7x7x7 receptive field and the eta=0.60 training cutoff are explicitly disclosed limitations, not circular inputs. The homogeneous EOS is an interpolation over the trained density range, but this is a generalization claim, not a circular reduction; no equation in the paper equates a validation output to a training label by construction. Accordingly, no circular step is exhibited.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The central claim rests on standard DFT variational relations, on the accuracy and diversity of GCMC training profiles, and on the finite 7x7x7 locality assumption for c(1). There are no invented physical entities. The network itself and the hand-set cutoffs are the main fitted elements.

free parameters (6)

Neural network weights = 114,433 trainable parameters
Fitted to GCMC c(1) targets via Adam; the central functional is defined by these weights.
First-layer kernel count = 32
Chosen after a 16-kernel model gave unsatisfactory validation and downstream accuracy.
Bernoulli retention probability p = 0.5
Hand-picked; authors note continuous systems need much smaller p.
c(1) target cutoff = -10
Sites with c(1)_GCMC <= -10 are excluded from the loss, removing hard-wall and low-statistics sites.
Training density cutoff = eta = 0.60
50 of 443 profiles with mean packing fraction above 0.60 are excluded, defining the trained range.
Optimizer hyperparameters = lr 1e-3, weight decay 1e-5, 400 epochs
Standard choices, not swept; checkpoint selected by validation loss.

assumptions (5)

standard math The grand-canonical Euler-Lagrange equation (5) is exact for the lattice model.
Used to derive supervised targets Eq. (6) and to make predictions Eq. (15); this is the DFT variational principle.
domain assumption Training GCMC profiles are equilibrium profiles at known beta mu and V_ext, so pointwise inversion Eq. (6) yields correct c(1) targets.
Section III.A; target correctness depends on simulation ergodicity and accurate histograms. Authors note histogram quality is the principal bottleneck for continuum extension.
ad hoc to paper The one-body direct correlation functional c(1)(r;[rho]) depends only on the density in a finite 7x7x7 neighborhood.
Section III.B first layer receptive field; truncates correlations beyond 3 lattice sites. Authors list finite receptive field as a controlled limitation.
domain assumption The density-to-c(1) map is exactly equivariant under the cubic point group Oh.
Section III.B Eqs. (11)-(13); follows from the hard-cube Hamiltonian symmetry, so not ad hoc.
ad hoc to paper The training corpus (randomized smooth fields, cuboids, hard-wall cavities) is sufficiently diverse for generalization to planar walls and test-particle configurations.
Section III.A and Table I; no formal guarantee; the wall slit resembles campaign D.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Cubic-Equivariant Neural Density Functional Theory for Three-Dimensional Lattice Fluids." pith.science (2026). https://pith.science/paper/IJYZT53N

@misc{pith2026260808137,
  author       = {Pith},
  title        = {Pith review of: Cubic-Equivariant Neural Density Functional Theory for Three-Dimensional Lattice Fluids},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IJYZT53N}},
  note         = {Machine review of arXiv:2608.08137}
}

abstract

We construct a neural classical density functional that acts directly on unrestricted three-dimensional density profiles. As a computationally tractable test bed, we consider parallel hard cubes of side length three on a simple cubic lattice. A fully convolutional network learns the one-body direct-correlation functional $c^{(1)}[\rho]$ from data obtained with grand-canonical Monte Carlo simulations in randomized external potentials. Complete profiles are used during both training and inference; a stochastic Bernoulli mask on the output sites makes full-profile training effective without explicitly extracting and storing overlapping local density windows. Averaging the first-layer kernels over all 48 rotations and reflections of the cubic point group additionally imposes exact cubic equivariance without data augmentation. We compare the learned functional with independent simulation data and with the lattice fundamental-measure functional of Lafuente and Cuesta. The neural functional markedly improves the homogeneous equation of state and the density profile at a planar hard wall. For the anisotropic pair distribution around a fixed particle, both functionals reproduce the principal packing shells, with their relative accuracy depending on crystallographic direction. These results demonstrate neural density-functional calculations on complete three-dimensional profiles while also identifying accurate full-dimensional training data, thermodynamic consistency, and structural correlations as the central challenges for extensions to continuum fluids.

Figures

Figures reproduced from arXiv: 2608.08137 by the authors.

**Figure 1.** FIG. 1. Homogeneous equation of state. (a) Packing fraction [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗

**Figure 3.** FIG. 3. Volume rendering of the neural test-particle pair [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: FIG. 4. Directional pair distribution at matched bulk packing fraction [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

**Figure 5.** Figure 5: FIG. 5. Central [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

**Figure 6.** Figure 6: shows the central xy cut of each effective kernel. The expected reflection and axis-exchange symmetries are visible in every panel, while the radial and shell-like variations differ across channels. These kernels need not admit a unique physical interpretation: subse…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 27 canonical work pages

[1]

Florian Sammüller and Sophie Hermann and Daniel de las Heras and Matthias Schmidt , title =. Proc. Natl. Acad. Sci. , volume =. 2023 , doi =

work page 2023
[2]

Florian Sammüller and Sophie Hermann and Matthias Schmidt , title =. J. Phys.: Condens. Matter , abstract =. 2024 , month =. doi:10.1088/1361-648X/ad326f , url =

doi:10.1088/1361-648x/ad326f 2024
[3]

Neural Density Functional Theory of Liquid-Gas Phase Coexistence , author =. Phys. Rev. X , volume =. 2025 , month =. doi:10.1103/PhysRevX.15.011013 , url =

doi:10.1103/physrevx.15.011013 2025
[4]

Hyperdensity Functional Theory of Soft Matter , author =. Phys. Rev. Lett. , volume =. 2024 , month =. doi:10.1103/PhysRevLett.133.098201 , url =

work page doi:10.1103/physrevlett.133.098201 2024
[5]

Sammüller, Florian and Schmidt, Matthias , title =. J. Phys.: Condens. Matter , abstract =. 2024 , month =. doi:10.1088/1361-648X/ad98da , url =

work page doi:10.1088/1361-648x/ad98da 2024
[6]

Neural density functionals: Local learning and pair-correlation matching , author =. Phys. Rev. E , volume =. 2024 , month =. doi:10.1103/PhysRevE.110.L032601 , url =

doi:10.1103/physreve.110.l032601 2024
[7]

Metadensity Functional Theory for Classical Fluids: Extracting the Pair Potential , author =. Phys. Rev. Lett. , volume =. 2025 , month =. doi:10.1103/PhysRevLett.134.107301 , url =

work page doi:10.1103/physrevlett.134.107301 2025
[8]

and Samm

Kampa, Stefanie M. and Samm. Metadensity Functional Learning for Classical Fluids: Regularizing with Pair Correlations , journal =. 2026 , doi =

work page 2026

Show all 44 references

[9]

arXiv preprint arXiv:2606.14370 , year=

Spherical metadensity functional learning for inhomogeneous classical fluids , author=. arXiv preprint arXiv:2606.14370 , year=
[10]

The Journal of Chemical Physics , volume =

Robitschko, Silas and Sammüller, Florian and Schmidt, Matthias and Evans, Robert , title =. The Journal of Chemical Physics , volume =. 2025 , month =. doi:10.1063/5.0290261 , url =

2025 doi
[11]

Determining the Chemical Potential via Universal Density Functional Learning , author =. Phys. Rev. Lett. , volume =. 2026 , month =. doi:10.1103/7bqn-y2d7 , url =

2026 doi
[12]

Learning Classical Density Functionals for Ionic Fluids , author =. Phys. Rev. Lett. , volume =. 2025 , month =. doi:10.1103/PhysRevLett.134.148001 , url =

2025 doi
[13]

Journal of Physics: Condensed Matter , abstract =

Bui, Anna T and Cox, Stephen J , title =. Journal of Physics: Condensed Matter , abstract =. 2025 , month =. doi:10.1088/1361-648X/ade7e7 , url =

2025 doi
[14]

and Cox, Stephen J

Bui, Anna T. and Cox, Stephen J. , title=. Nature Communications , year=. doi:10.1038/s41467-026-69482-1 , url=

doi
[15]

Bui and Stephen J

Anna T. Bui and Stephen J. Cox , title =. Proceedings of the National Academy of Sciences , volume =. 2026 , doi =

2026
[16]

Zhou, Katie L. Y. and Bui, Anna T. and Cox, Stephen J. , title =. The Journal of Physical Chemistry B , volume =. 2026 , month =. doi:10.1021/acs.jpcb.6c00640 , url =

2026 doi
[17]

Neural density functional theory in higher dimensions with convolutional layers , author =. Phys. Rev. E , volume =. 2025 , month =. doi:10.1103/PhysRevE.111.055305 , url =

2025 doi
[18]

Machine Learning Approaches to Classical Density Functional Theory

Simon, Alessandro and Oettel, Martin. Machine Learning Approaches to Classical Density Functional Theory. Artificial Intelligence and Intelligent Matter: Nanoscience, Soft Matter, Philosophy. 2026. doi:10.1007/978-3-032-04129-6_6

2026 doi
[19]

Journal of Chemical Theory and Computation , volume =

Simon, Alessandro and Weimar, Jens and Martius, Georg and Oettel, Martin , title =. Journal of Chemical Theory and Computation , volume =. 2024 , month =. doi:10.1021/acs.jctc.3c01238 , url =

2024 doi
[20]

The Journal of Chemical Physics , volume =

Simon, Alessandro and Belloni, Luc and Borgis, Daniel and Oettel, Martin , title =. The Journal of Chemical Physics , volume =. 2025 , month =. doi:10.1063/5.0248694 , url =

2025 doi
[21]

2019 , publisher=

Shang-Chun Lin and Martin Oettel , journal=. 2019 , publisher=. doi:10.21468/SciPostPhys.6.2.025 , url=

2019 doi
[22]

and Martius, G

Lin, S.-C. and Martius, G. and Oettel, M. , title = ". J. Chem. Phys. , volume =. 2020 , month =. doi:10.1063/1.5135919 , url =

2020 doi
[23]

and Dijkstra, Marjolein and van Roij, René , title =

Cats, Peter and Kuipers, Sander and de Wind, Sacha and van Damme, Robin and Coli, Gabriele M. and Dijkstra, Marjolein and van Roij, René , title =. APL Mater. , volume =. 2021 , month =. doi:10.1063/5.0042558 , url =

2021 doi
[24]

Learning Neural Free-Energy Functionals with Pair-Correlation Matching , author =. Phys. Rev. Lett. , volume =. 2025 , month =. doi:10.1103/PhysRevLett.134.056103 , url =

2025 doi
[25]

and Quinton, Joshua and Fazel, Kamron and Karimitari, Nima and Sutton, Christopher and Sundararaman, Ravishankar , title =

Kelley, Michelle M. and Quinton, Joshua and Fazel, Kamron and Karimitari, Nima and Sutton, Christopher and Sundararaman, Ravishankar , title =. The Journal of Chemical Physics , volume =. 2024 , month =. doi:10.1063/5.0223792 , url =

2024 doi
[26]

The Journal of Chemical Physics , volume =

Pan, Runtong and Fang, Xinyi and Azizzadenesheli, Kamyar and Liu-Schiaffini, Miguel and Gu, Mengyang and Wu, Jianzhong , title =. The Journal of Chemical Physics , volume =. 2025 , month =. doi:10.1063/5.0284515 , url =

2025 doi
[27]

Thermal Properties of the Inhomogeneous Electron Gas , author =. Phys. Rev. , volume =. 1965 , doi =

1965
[28]

Evans , title =

R. Evans , title =. Adv. Phys. , volume =. 1979 , publisher =. doi:10.1080/00018737900101365 , URL =

1979 doi
[29]

Density Functional Theory for General Hard-Core Lattice Gases , author =. Phys. Rev. Lett. , volume =. 2004 , month =. doi:10.1103/PhysRevLett.93.130603 , url =

2004 doi
[30]

Fundamental Measure Theory for Lattice Fluids with Hard-Core Interactions , author =. J. Phys.: Condens. Matter , volume =. 2002 , doi =

2002
[31]

Percus, J. K. , title=. Journal of Statistical Physics , year=. doi:10.1007/BF01020803 , url=

doi
[32]

Approximation Methods in Classical Statistical Mechanics , author =. Phys. Rev. Lett. , volume =. 1962 , doi =

1962
[33]

Roland Roth , title =. J. Phys.: Condens. Matter , abstract =. 2010 , month =. doi:10.1088/0953-8984/22/6/063102 , url =

2010 doi
[34]

Free-energy model for the inhomogeneous hard-sphere fluid mixture and density-functional theory of freezing , author =. Phys. Rev. Lett. , volume =. 1989 , month =. doi:10.1103/PhysRevLett.63.980 , url =

1989 doi
[35]

Explicitly stable fundamental-measure-theory models for classical density functional theory , author =. Phys. Rev. E , volume =. 2020 , month =. doi:10.1103/PhysRevE.102.062137 , url =

2020 doi
[36]

The Journal of Chemical Physics , volume =

Roth, Roland and Mecke, Klaus and Oettel, Martin , title =. The Journal of Chemical Physics , volume =. 2012 , month =. doi:10.1063/1.3687921 , url =

2012 doi
[37]

Journal of Physics: Condensed Matter , abstract =

Hansen-Goos, Hendrik and Roth, Roland , title =. Journal of Physics: Condensed Matter , abstract =. 2006 , month =. doi:10.1088/0953-8984/18/37/002 , url =

2006 doi
[38]

Using test particle sum rules to improve approximations in classical density functional theory: White-Bear and White-Bear mark II versions of the Lutsko functional , author =. Phys. Rev. E , volume =. 2026 , month =. doi:10.1103/3vg8-mbf4 , url =

2026 doi
[39]

Proceedings of the 33rd International Conference on Machine Learning , series =

Group Equivariant Convolutional Networks , author =. Proceedings of the 33rd International Conference on Machine Learning , series =. 2016 , publisher =

2016
[40]

, booktitle =

Weiler, Maurice and Geiger, Mario and Welling, Max and Boomsma, Wouter and Cohen, Taco S. , booktitle =. 2018 , publisher =

2018
[41]

Mitra, S. K. and Allnatt, A. R. , title =. The Journal of Chemical Physics , volume =. 1979 , doi =

1979
[42]

and Piccarreta, Michael V

Davis, Jonathan R. and Piccarreta, Michael V. and Rauch, Rory B. and Vanderlick, T. Kyle and Panagiotopoulos, Athanassios Z. , title =. Industrial & Engineering Chemistry Research , volume =. 2006 , doi =

2006
[43]

Marques and Levin, Yan and Arenzon, Jeferson J

Fernandes, Heitor C. Marques and Levin, Yan and Arenzon, Jeferson J. , title =. Physical Review E , volume =. 2007 , doi =

2007
[44]

and Mandal, Dipanjan and Damle, Kedar and Dhar, Deepak and Rajesh, R

Vigneshwar, N. and Mandal, Dipanjan and Damle, Kedar and Dhar, Deepak and Rajesh, R. , title =. Physical Review E , volume =. 2019 , doi =

2019

Pith tools

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

[1] [1]

Florian Sammüller and Sophie Hermann and Daniel de las Heras and Matthias Schmidt , title =. Proc. Natl. Acad. Sci. , volume =. 2023 , doi =

work page 2023

[2] [2]

Florian Sammüller and Sophie Hermann and Matthias Schmidt , title =. J. Phys.: Condens. Matter , abstract =. 2024 , month =. doi:10.1088/1361-648X/ad326f , url =

doi:10.1088/1361-648x/ad326f 2024

[3] [3]

Neural Density Functional Theory of Liquid-Gas Phase Coexistence , author =. Phys. Rev. X , volume =. 2025 , month =. doi:10.1103/PhysRevX.15.011013 , url =

doi:10.1103/physrevx.15.011013 2025

[4] [4]

Hyperdensity Functional Theory of Soft Matter , author =. Phys. Rev. Lett. , volume =. 2024 , month =. doi:10.1103/PhysRevLett.133.098201 , url =

work page doi:10.1103/physrevlett.133.098201 2024

[5] [5]

Sammüller, Florian and Schmidt, Matthias , title =. J. Phys.: Condens. Matter , abstract =. 2024 , month =. doi:10.1088/1361-648X/ad98da , url =

work page doi:10.1088/1361-648x/ad98da 2024

[6] [6]

Neural density functionals: Local learning and pair-correlation matching , author =. Phys. Rev. E , volume =. 2024 , month =. doi:10.1103/PhysRevE.110.L032601 , url =

doi:10.1103/physreve.110.l032601 2024

[7] [7]

Metadensity Functional Theory for Classical Fluids: Extracting the Pair Potential , author =. Phys. Rev. Lett. , volume =. 2025 , month =. doi:10.1103/PhysRevLett.134.107301 , url =

work page doi:10.1103/physrevlett.134.107301 2025

[8] [8]

and Samm

Kampa, Stefanie M. and Samm. Metadensity Functional Learning for Classical Fluids: Regularizing with Pair Correlations , journal =. 2026 , doi =

work page 2026