{"id":"52e91a45-6324-4cef-83f0-1c3574e8abd5","arxiv_id":"2412.20739","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A machine-learned orbital-free density functional reproduces the deformed ground state and potential energy curve of 20Ne for the first time.","lead":"The authors train a kernel ridge regression model to map the nucleon density directly to kinetic and spin-orbit energies, then minimize the total energy as a function of density alone. The resulting orbital-free functional reproduces the Kohn-Sham ground state and potential energy curves of spherical 16O and deformed 20Ne, including deformed minima.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The learned functional is verified only on energies for random potentials, while self-consistent minimization uses its derivative; derivative error is the load-bearing uncertainty, tested on only two N=Z nuclei.","rationale":"I read the paper as a proof-of-principle: constructing an orbital-free EDF for 16O and 20Ne by learning Ekin+so[rho] with KRR and minimizing the resulting functional. The method is clearly described, the test-set RMSE is small, and the reproduction of beta2, radius, and energy for two nuclei is a genuine demonstration. The load-bearing uncertainty is not the KRR energy accuracy on random potentials but the accuracy of the functional derivative at self-consistent Skyrme densities; the paper does not present a direct derivative check. The two-nucleus, N=Z, no-code/no-data limitations strengthen this: they leave open the possibility that the random-potential density manifold happens to cover the two minima but not the surrounding potential energy curve, particularly beta2 near 0 where Kohn-Sham is not converged. I do not think this warrants rejection; it warrants conditional acceptance with a concrete verification step. This agrees partially with the reader's weakest_assumption (generalization to self-consistent densities) but sharpens it from energy to derivative and from 'more nuclei' to a direct out-of-sample check on the actual target densities.","tokens_in":13685,"tokens_out":6326,"duration_ms":72586,"concrete_test":"Use the authors' code, once released, to generate the converged self-consistent Skyrme density for 16O and 20Ne at each constrained beta2 in Fig. 2; at each density, solve the Kohn-Sham equations once to obtain the exact Ekin+so and its functional derivative. Compare the KRR predictions of Ekin+so and its derivative at those same densities with the exact values. If the derivative error changes the unconstrained minimum by more than 0.03 in beta2 or the energy error exceeds the Fig. 1 RMSE, the central claim is not yet supported; if both are within uncertainties, the generalization concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim rests on Eqs. (3)-(6): the KRR map from density to Ekin+so is inserted into Eq. (6) and minimized by gradient descent, which requires the functional derivative in Eq. (S11). Two facts make the derivative, not the energy, the load-bearing quantity. First, the validation/test statistics in Fig. 1 are rms errors of predicted energies on random-potential densities; an energy error of 5-25 keV over the training distribution does not bound the error of the density gradient, and KRR derivatives are more sensitive to hyperparameters and to extrapolation outside the training hull. Second, the self-consistent solution is obtained by descending this derivative, so a small systematic derivative bias can shift the energy minimum even when the energy surface is well reproduced; the standard deviations in Table 1 measure only sensitivity to 100 random starting densities, not the error of the learned derivative at the true Skyrme density. The training family (Methods S1) is a 12-parameter random external potential for N=Z systems, and only 16O and 20Ne are tested, so the density manifold of self-consistent Skyrme functionals is not independently probed. Without released code or data, the leap from random-potential interpolation to 'taming shell effects' rests on two isolated points.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper proposes to construct an orbital-free nuclear energy density functional by using kernel ridge regression (KRR) to learn the sum of kinetic and spin-orbit energies as a functional of the nucleon density. The KRR model is trained on 24,000 densities and energies generated by solving Schrödinger equations with randomly parameterized external potentials for A=16 and A=20, and is then minimized self-consistently to obtain ground states and constrained potential energy curves for 16O and 20Ne. The authors report that the resulting orbital-free approach reproduces the Kohn-Sham quadrupole deformations, rms radii, total energies, and density profiles, including the deformed minimum of 20Ne, and they claim in the abstract and summary that this is the first fully orbital-free EDF to account for shell effects in deformed nuclei.","tokens_in":13890,"tokens_out":6688,"duration_ms":65515,"significance":"The result, if confirmed by broader tests, is a significant proof-of-principle: it shows that an explicit orbital-free functional can describe a deformed nucleus without introducing Kohn-Sham orbitals, which has been a longstanding challenge. The paper's main strengths are the clear construction of the KRR map with a density-space kernel, the use of a large training set generated from random potentials, and the implementation of self-consistent gradient-descent minimization including constrained calculations with an augmented Lagrangian. The reported agreement with Kohn-Sham for the two test nuclei is encouraging. However, the present validation covers only two N=Z light nuclei, both with mass numbers contained in the training set, and the functional derivative is not independently benchmarked; these are important caveats that limit the strength of the central claim. The paper does not release code or data, only availability upon request, which further hampers independent verification.","major_comments":[{"comment":"The self-consistent minimization in Section S3 uses the functional derivative of the KRR model, Eq. (S11), but the reported validation in Fig. 1 concerns rms errors of predicted energies on the validation and test sets, not errors of the derivative. The derivative is the quantity that drives the gradient descent in Eq. (S12) and determines the stationary density and the constrained potential energy curves. A small systematic bias in the derivative can displace the energy minimum even when the energy surface at the training densities is accurately reproduced. The standard deviations in Table 1 reflect sensitivity to 100 random initial densities, not the model error of the derivative at the self-consistent densities. I request an explicit benchmark of the functional derivative against the exact Kohn-Sham potential for the test-set densities, or an equivalent demonstration that derivative errors are small enough not to shift the ground-state density and PEC beyond the reported statistical uncertainties.","section":"Eq. (S11), Fig. 1, Table 1"},{"comment":"The KRR functional is trained exclusively on densities generated for A=16 and A=20 (Methods S1), and the paper reports ground-state and PEC results only for these same two nuclei. Thus the statement in the abstract and summary that this is the first fully orbital-free EDF to tame shell effects in deformed nuclei is supported by only one deformed nucleus, 20Ne, and by interpolation within the trained mass window. To substantiate the claim of a general orbital-free functional, at least one deformed nucleus outside the training set (e.g., 24Mg or 28Si) should be tested without retraining, or the training set should be shown to include densities that are sufficiently diverse to make the A=16/A=20 restriction inconsequential. If the method requires retraining for each nucleus, the scope of the claim should be narrowed accordingly.","section":"Methods S1, Table 1"},{"comment":"The authors state that Kohn-Sham calculations cannot converge near β2=0 for 20Ne due to degeneracies, and that this phenomenon is 'accurately captured' by the ML orbital-free approach. Since no Kohn-Sham reference is available in this region, there is no benchmark for the ML PEC at β2=0. The agreement at the deformed minima is good, but the barrier height and the shape of the curve near β2=0 are predictions of the learned functional, not validated results. The paper should either provide a benchmark from a different method (e.g., a Hartree-Fock-Bogoliubov calculation with pairing) or present this region as a prediction with a clear caveat, rather than as part of the demonstrated accuracy.","section":"Paragraph after Fig. 2"},{"comment":"The training densities are generated by random external potentials with rms radii restricted to an empirical band [0.8 A^{1/3}, 1.2 A^{1/3}] (Methods S1). The self-consistent densities of 16O and 20Ne under the Skyrme interaction are likely inside this band, but the paper does not quantify the proximity of the target densities to the training distribution. KRR is an interpolation method, and the reported accuracy may reflect that the target densities lie inside the training hull rather than a physical generalization. I ask for a quantitative analysis of the distance, in the kernel metric of Eq. (5), between the self-consistent densities and the training set, and for a discussion of how the results depend on the random potential parameter ranges.","section":"Methods S1, Eq. (5)"},{"comment":"The learned functional EML_{kin+so}[ρ] represents the kinetic plus spin-orbit energy as a functional of the total density alone. In the Skyrme EDF, the spin-orbit energy depends on the spin-orbit density J (second line of Eq. (2)), which is not generally a unique functional of the total density for arbitrary spin-orbit potentials. The paper does not address this formal issue; if the ρ → (Ekin+Eso) map is not universal, the trained functional may not transfer to nuclei or densities with different spin-orbit structure. At minimum, a discussion of the conditions under which this map is well-defined (e.g., time-reversal invariant, even-even N=Z systems) is needed.","section":"Eq. (2), Eq. (3)"}],"minor_comments":[{"comment":"The distance ||ρ(r)-ρ'(r)|| in Eq. (5) is not defined explicitly; please state that it is the L2 norm over the discrete mesh and specify how the density vectorization is performed, as this is needed for reproducibility.","section":"Eq. (5)"},{"comment":"The text mentions that c.m. correction energies and Coulomb energies are included, but does not specify the formulas or parameter values used; provide these details for reproducibility.","section":"Section S3.2"},{"comment":"The phrase 'inaugural instance' is used twice in prominent places; consider using a more conventional phrase such as 'first demonstration' for clarity and style.","section":"Abstract and Summary"}],"recommendation":"major_revision","confidential_remarks":"The paper is within the scope of the journal and addresses a recognized problem. The main concern is the gap between the strength of the central claim and the breadth of validation: only two N=Z light nuclei, both in the training mass window, are tested, and the derivative of the learned functional is not directly validated. The lack of released code and data makes independent verification difficult. I would encourage the editor to require, at minimum, the training data and the self-consistent solver to be made available upon publication, and to ask the authors to add a derivative-error benchmark and an out-of-training-set test before accepting the claim of general applicability."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper does what it claims for the two nuclei it tests: it is the first orbital-free EDF that yields a deformed minimum for 20Ne while staying close to Kohn-Sham results. The new elements are real—training on densities from random quadrupole potentials and adding spin-orbit energy to the learned functional. The method is clearly described, analysis is honest, and the reported test-set energy errors (5–25 keV) are good. The self-consistent ground states and potential energy curves for 16O and 20Ne match KS well, including the deformed minimum and density profiles.\n\nSoft spots are proportionate to the scope. The verification is limited to two N=Z light nuclei, and the β2≈0 region of the 20Ne PEC is not benchmarked (a point the authors acknowledge). The stress-test worry about the derivative is valid but not fatal: the minimization uses the derivative, while the validation is on energies. Still, the good reproduction of the PEC shape suggests the derivative is reasonable in the relevant deformation range. A direct derivative check or a test on one or two more deformed nuclei (e.g., 24Mg) would substantially strengthen the claim. The circularity is moderate, since the training densities bracket the target deformations, but the self-consistent densities are not identical to the training ones and the agreement with KS is reassuring.\n\nThe lack of released code and data is a real limitation. The data availability statements say 'available upon request' with possible export control constraints, which hampers independent verification. The 'inaugural' claim is plausible, but a referee should check recent related work carefully.\n\nThis deserves a serious referee. It is a solid proof-of-principle that opens a direction for nuclear orbital-free DFT. If I worked in this area, I would cite it and build on it, but I would want to see the derivative check and a wider test set before trusting it for heavy nuclei.\n\nReading group: yes, it would spark good discussion on ML-DFT and generalization.","headline":"A credible proof-of-principle that an ML-trained orbital-free EDF can reproduce deformed minima in light nuclei; the main caveats are the narrow test set and the untested derivative.","tokens_in":14488,"tokens_out":3349,"would_cite":true,"duration_ms":35486,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A kernel-ridge-regression orbital-free functional reproduces the deformed ground state and potential energy curve of 20Ne.","keywords":["orbital-free density functional theory","machine learning","kernel ridge regression","nuclear deformation","nuclear shell effects","neon-20","oxygen-16","Hohenberg-Kohn theorem"],"falsifier":"Train or reuse the same construction on a deformed nucleus outside the training pair, such as 24Mg, and compare the self-consistent $\\beta_2$ and potential energy curve with Kohn-Sham results. A disagreement in the deformed minimum would show that the random-potential training family does not cover the densities that real nuclei produce.","tokens_in":13432,"feed_emoji":"⚛️","tokens_out":7931,"duration_ms":74591,"temperature":0.7,"pith_summary":"This paper aims to show that a fully orbital-free nuclear energy density functional—one that depends only on the density, with no auxiliary single-particle orbitals—can reproduce quantum shell and deformation effects in deformed nuclei. The authors train a kernel ridge regression model to map nucleon density onto the combined kinetic and spin-orbit energy, then add a Skyrme interaction term and minimize the total functional self-consistently. For deformed 20Ne the result matches Kohn-Sham deformation ($\\beta_2=0.49$ vs 0.48), radius (3.05 vs 3.02 fm), and total energy (−156.02 vs −156.58 MeV), and the potential energy curve keeps both oblate and prolate minima. This is presented as the first orbital-free energy density functional to capture complex shell effects in deformed nuclei, turning the Hohenberg-Kohn theorem from a formal statement into a practical tool.","feed_headline":"A density-only functional reproduces deformed 20Ne","feed_subtitle":"Trained on random potentials, it matches Kohn-Sham deformation and energies without single-particle orbitals.","key_machinery":"The engine of the construction is the kernel ridge regression map from density to kinetic-plus-spin-orbit energy, $E^\\mathrm{ML}_{\\mathrm{kin+so}}[\\rho]=\\sum_i \\omega_i K(\\rho_i,\\rho)$, where $K$ is a Gaussian kernel with width $\\sigma$ and the weights solve $\\boldsymbol{\\omega}=(K+\\lambda I)^{-1}\\mathbf{E}$. Its functional derivative with respect to the density provides the gradient for self-consistent minimization, so the Kohn-Sham orbitals are bypassed entirely. The density is represented on an axial grid of 1128 discrete points, and the training densities are generated by solving Schrödinger equations with random spherical and quadrupole external potentials plus a spin-orbit potential built from a preliminary density. Together with the density-dependent Skyrme interaction (SkP), the learned term forms the total orbital-free functional, and constrained runs add an augmented-Lagrangian penalty to hold the quadrupole moment while tracing potential energy curves.","core_discovery":"The central claim is that shell and deformation effects in nuclei can be described without ever constructing single-particle orbitals. The authors implement this by learning the density-to-energy map for the kinetic plus spin-orbit part, $E^\\mathrm{ML}_{\\mathrm{kin+so}}[\\rho]=\\sum_i \\omega_i K(\\rho_i,\\rho)$, with a Gaussian kernel over discretized axial densities, while the interaction part is taken from the Skyrme functional SkP. The training data come from 24,000 Schrödinger equation solutions under randomly parameterized spherical and quadrupole mean fields, split into 20,000 training, 2,000 validation, and 2,000 test samples. Self-consistent minimization of the resulting functional for 20Ne yields a quadrupole deformation $\\beta_2=0.49$ against the Kohn-Sham value 0.48, a root-mean-square radius 3.05 fm against 3.02 fm, and a total energy −156.02 MeV against −156.58 MeV; the constrained calculation reproduces the Kohn-Sham potential energy curve including the deformed minima, and the density profiles capture the spatial fluctuations associated with shell effects.","pith_inferences":["If the functional transfers to neighboring nuclei, the same protocol could generate orbital-free mass or radius tables without Kohn-Sham iterations; testing that would require training over a larger mass range.","The result weakens the common argument that shell effects are intrinsically single-particle effects, since the learned functional captures them from density fluctuations alone.","A natural stress test is the region around $\\beta_2=0$ in 20Ne, where the authors note Kohn-Sham struggles without pairing; an orbital-free description of that region could reveal whether the learned functional implicitly encodes pairing-like correlations."],"forward_implications":["Nuclear orbital-free DFT can describe deformed ground states, not just spherical ones, without solving Kohn-Sham equations.","Potential energy curves and shape-isomeric minima become accessible from a density-only functional.","The learned functional captures the spatial density fluctuations that correspond to quantum shell effects, which Thomas-Fermi-type functionals miss.","The Hohenberg-Kohn theorem gains a practical nuclear implementation, opening quantitative orbital-free studies beyond spherical nuclei."],"supporting_citations":[{"why":"Supplies the Hohenberg-Kohn theorem that a total-energy functional of density exists, the theoretical basis of the whole orbital-free program.","marker":"[2]"},{"why":"The Kohn-Sham method whose self-consistent densities, energies, and potential energy curves serve as the reference that the orbital-free functional must reproduce.","marker":"[3]"},{"why":"Provides the Thomas-Fermi and extended Thomas-Fermi semiclassical functionals that the paper identifies as prior orbital-free approaches lacking deformed shell effects.","marker":"[4]"},{"why":"The authors' earlier machine-learned orbital-free functional for spherical nuclei, which supplies the Gaussian-potential data-generation and adaptive derivative methods reused here.","marker":"[70]"},{"why":"Defines the SkP functional whose parameters enter the interaction part of the constructed orbital-free energy density functional.","marker":"[75]"},{"why":"Provides the augmented Lagrangian method used for constrained quadrupole-moment calculations behind the potential energy curves.","marker":"[77]"}],"fun_headline_variants":["AI tames shell effects in deformed nuclei without orbitals","Orbital-free functional nails deformed 20Ne with ML","Machine learning beats shell effects in deformed nuclei","Density functional learns deformation without orbitals"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that densities produced by the randomly parameterized external potentials cover the density manifold of real self-consistent nuclei, so the learned kinetic-plus-spin-orbit map remains accurate when the functional is minimized with the Skyrme interaction; only 16O and 20Ne are tested.","fun_headline_variants_meta":{"raw":{"variants":["AI tames shell effects in deformed nuclei without orbitals","Orbital-free functional nails deformed 20Ne with ML","Machine learning beats shell effects in deformed nuclei","Density functional learns deformation without orbitals"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001064,"raw_usage":{"total_tokens":4478,"prompt_tokens":977,"completion_tokens":3501,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":593,"completion_tokens_details":{"reasoning_tokens":3441}},"tokens_in":593,"tokens_out":3501,"duration_ms":20494,"temperature":1.0,"reasoning_tokens":3441,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T23:13:59.332006+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Train or reuse the same construction on a deformed nucleus outside the training pair, such as 24Mg, and compare the self-consistent $\\beta_2$ and potential energy curve with Kohn-Sham results. A disagreement in the deformed minimum would show that the random-potential training family does not cover the densities that real nuclei produce.","supporting_citations":[{"cited_title":"& Kohn, W","cited_arxiv_id":null,"evidence_quote":"Supplies the Hohenberg-Kohn theorem that a total-energy functional of density exists, the theoretical basis of the whole orbital-free program."},{"cited_title":"& Sham, L","cited_arxiv_id":null,"evidence_quote":"The Kohn-Sham method whose self-consistent densities, energies, and potential energy curves serve as the reference that the orbital-free functional must reproduce."},{"cited_title":"& Hakansson, H.-B","cited_arxiv_id":null,"evidence_quote":"Provides the Thomas-Fermi and extended Thomas-Fermi semiclassical functionals that the paper identifies as prior orbital-free approaches lacking deformed shell effects."},{"cited_title":"H., Ren, Z","cited_arxiv_id":null,"evidence_quote":"The authors' earlier machine-learned orbital-free functional for spherical nuclei, which supplies the Gaussian-potential data-generation and adaptive derivative methods reused here."},{"cited_title":"& Treiner, J","cited_arxiv_id":null,"evidence_quote":"Defines the SkP functional whose parameters enter the interaction part of the constructed orbital-free energy density functional."},{"cited_title":"& Nazarewicz, W","cited_arxiv_id":null,"evidence_quote":"Provides the augmented Lagrangian method used for constrained quadrupole-moment calculations behind the potential energy curves."}],"review_version":1}