Pith. sign in

REVIEW 3 major objections 5 minor 20 references

The paper establishes that self-consistent orbital-free nuclear density functional theory is possible with a learned nonlocal kinetic-energy functional, and that a single trained correction can transfer shell structure and radius to an unse

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

2026-07-31 23:44 UTC pith:NYH4VAUQ

load-bearing objection Genuinely new self-consistent nonlocal nuclear OFDFT with unusually honest reporting; the universal-functional claim rests on one holdout and engineered stability, and the paper says so. the 3 major comments →

arxiv 2607.23328 v1 pith:NYH4VAUQ submitted 2026-07-25 physics.comp-ph

Self-consistent orbital-free nuclear density functional theory with a physics-constrained learned nonlocal kinetic energy functional

Fumihiro Imoto This is my paper
classification physics.comp-ph PACS 31.15.E-
keywords orbital-free density functional theorykinetic energy density functionalnuclear shell structurenonlocal functionalmachine-learned functionalfunctional derivativenucleon localization functionself-consistent solver
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 claims to solve a long-standing blocker for orbital-free nuclear DFT: a nonlocal kinetic energy functional that encodes shell structure could previously be evaluated on known densities, but it could not survive self-consistent minimization because its functional derivative was never controlled. By exploiting exact linearity of the functional in the kernel shape, the author derives analytic Euler–Lagrange potentials and fits the kernel correction to the functional derivative, not just functional values. With per-nucleus fits, self-consistent densities track Kohn–Sham densities and full nucleon-localization-function shell structure for doubly magic N=Z systems from A=16 to A=140. A shared correction trained on three nuclei transfers shell pattern and radius to a holdout A=140 nucleus, though not the absolute energy. If correct, this removes the orbital bottleneck of nuclear Kohn–Sham theory for spherical N=Z benchmarks and defines the ingredients—analytic derivatives, solver cross-checks, quadrature control, and energy anchoring—needed for a universal orbital-free functional.

Core claim

On its own terms, the paper demonstrates that self-consistent orbital-free calculations with a shell-structured nonlocal kinetic-energy density functional are possible. The central device is a density-dependent-kernel functional that is exactly linear in a dimensionless kernel shape g; because of that linearity, the full Euler–Lagrange potential, including all kernel density dependencies, is available analytically, and every basis function of the learned correction contributes a known response. Fitting is then performed on the functional derivative as well as on values, energies, tail-margin inequalities, and selected-path energy-rise penalties. Final per-nucleus densities sit close to the K

What carries the argument

The density-dependent-kernel nonlocal KEDF, in which the pair kernel is f1(R;t)=t·g(t^{1/3}R) and the functional is exactly linear in the dimensionless kernel shape g(y). Linearity gives exact analytic Euler–Lagrange responses (Eqs. 7–9) for the represented finite-table functional, so the learned correction δg is fitted simultaneously to NLF values, per-basis EL potentials (functional-derivative, 'Sobolev' supervision), exact energy-matching equalities, and soft quadratic inequality penalties for tail loss of positivity and selected-path energy rise. The radial EL equation is solved either by adaptive-step imaginary-time evolution or by the rearranged one-orbital eigenproblem, which cross-ch

Load-bearing premise

The stability of the learned functional rests on three one-dimensional trial-path energy-rise penalties built from a single recorded central-depletion direction; if the fitted correction has other unstable descent directions not probed by those paths, unphysical depleted solutions could appear on nuclei not in the training set.

What would settle it

Release the shared functional on a sequence of N=Z spherical systems between A=80 and A=140 (e.g., A=100, A=120) under the same solver settings: if any of them develops a central depletion or tail artifact, the three selected-path rows did not cover the unstable manifold. Alternatively, compute the lowest eigenvalues of the exact (finite-difference) Hessian of the total energy at the converged A=80 density; a negative direction orthogonal to the three tested trial paths would falsify the stability claim directly.

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

If this is right

  • A single shared kernel correction, trained on three light-to-medium nuclei, transfers shell structure and radius to an unseen A=140 nucleus, suggesting that some shell-structure information is portable across nuclei within this N=Z benchmark.
  • The self-consistency barrier to nonlocal nuclear OFDFT is lowered: with analytic derivatives, orbital-free calculations can reach the same stationary densities as imaginary-time evolution and, in the constant-kF benchmark, converge without density mixing.
  • Because value-only fits collapse on relaxation while derivative-informed fits stay near the KS density, any learned KEDF intended for self-consistent use must be trained on functional derivatives, not just energies or densities.
  • Absolute energies remain nucleus-specific at the present basis size; transferring total energies across nuclei requires additional physics, such as explicit uniform-matter constraints or density-dependent kernel corrections.
  • The deployed instability controls—tail margins and selected-path energy-rise rows—make the fit stable against the documented central-depletion mode, but stability is certified only along the probed directions.

Where Pith is reading between the lines

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

  • The same recipe—exact linearity, analytic EL responses, and inequality penalties on known failure directions—could be applied to electronic OFDFT with nonlocal KEDFs, where the value/derivative dichotomy is the same and self-consistent stability is the known bottleneck.
  • The 41.7 MeV overbinding of the holdout suggests that the energy's A-dependence is carried by the kernel correction's uniform moments; a testable extension is to add density-gradient or t-dependent kernel corrections and check whether energy transfer improves without degrading the transferred shell pattern.
  • The three trial paths were built from one recorded artifact direction; a stronger certification would replace them with a direct low-rank Hessian scan at the converged density, making the functional's stability claim falsifiable in a systematic way.
  • If the A=80 central density's 7.8% quadrature dependence reflects an unconstrained curvature direction rather than a benign discretization artifact, higher-order angular integration should convert it into a genuine depletion on some untested nucleus.

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

3 major / 5 minor

Summary. The paper presents an orbital-free DFT scheme for spherical N=Z nuclei without spin-orbit and Coulomb terms, using a learned correction to a density-dependent-kernel nonlocal kinetic-energy functional. The functional is linear in the kernel shape, which allows analytic Euler-Lagrange potentials; the radial EL equation is solved either by a rearranged one-orbital diagonalization or by adaptive imaginary-time evolution. Fitting combines NLF values, analytic EL-potential supervision on KS densities, exact per-nucleus energy equalities, and quadratic inequality penalties for tail response and selected energy-rise paths. Per-nucleus fits for A=16-140 track KS densities, radii and NLF structure; a single kernel correction trained on 16O, 40Ca, and A=80 transfers NLF structure and radius to a held-out A=140 system while overbinding it by 41.7 MeV. The manuscript releases code and data, verifies the analytic derivative against finite differences to ~1e-11, and includes multiple solver, initialization, and quadrature checks.

Significance. If correct, this is a significant step: it demonstrates that nonlocal, shell-structured KEDFs can be made self-consistent, addresses the value/derivative dichotomy through Sobolev-style training, and provides a reproducible benchmark with all code and data. The paper is unusually careful in quantifying limitations: analytic derivative checks, two solvers, Woods-Saxon initialization probes, quadrature sensitivity, and explicit statements about what is not certified (e.g., local Hessian, GP coverage). The main result is nonetheless conditional: the transfer claim for the shared functional rests on a stability certification that is currently incomplete, as detailed in Major Comment 1. I agree with the stress-test concern that the three selected-path rows are not a substitute for a local curvature/perturbation analysis.

major comments (3)
  1. [Sec. III B 6, Eq. (25); Sec. V; Appendix B/C] The stability of the shared functional is certified by only three one-dimensional energy-rise rows built from a single recorded failure direction, the preliminary A=80 central depletion. The paper explicitly states these rows are 'not direct measurements of a local Hessian' (Sec. III B 6), and the released shared functional uses s2=0, so the gradient regularization is not part of the stability guarantee. The two solvers share the same potential/energy evaluator, so their agreement validates the fixed-point solver, not the absence of other unstable descent directions. Since the active rows at the final solution lie 0.64, 0.49, and 1.66 MeV below the epsilon=2 MeV target (Appendix C), and the A=80 central density shifts by 7.8% when n_mu is refined from 24 to 48 (Appendix B), the claim of a stable universal functional needs an additional generic stability test: local curvature or Hessian-v
  2. [Sec. III B 2-3; Table II] The per-nucleus shell-structure reproduction is partly constructed by the fitting objective: the EL potential is matched to a constant chemical potential on rho_KS (Sec. III B 2) and Eq. (23) enforces the exact KS energy at fixed KS densities. It is therefore not surprising that the self-consistent per-nucleus densities stay near rho_KS and reproduce its NLF. The manuscript discloses this, but the abstract and Sec. I present per-nucleus reproduction as a central demonstration. I recommend reclassifying these per-nucleus results as consistency checks and designating the A=140 holdout as the sole genuine predictive test; this would make the claims better match the evidence.
  3. [Sec. III B 6; Sec. V] The statement that including the three energy-rise rows 'removes the failure at its source' is stronger than the evidence supports. The released functional keeps the recorded path energies at +0.41, +0.29, and +1.09 MeV, all below the soft 2 MeV target, and the paper notes the paths are not a Hessian measurement. Since the penalty is soft and one-dimensional, the phrase 'removes the instability' should be softened to 'suppresses the recorded artifact direction' unless additional curvature evidence is supplied.
minor comments (5)
  1. [Sec. V] The phrase 'the 2s orbital contributes' is KS-language in an orbital-free context; rephrase in terms of the central-density region or the KS reference state that motivates the artifact.
  2. [Fig. 6 caption] The GP posterior mean is plotted multiplied by 10; state this explicitly in the caption or use a secondary axis.
  3. [Sec. II B] For Eq. (7), state explicitly that t has units fm^-3 and y is dimensionless, so the derivative is dimensionless; this helps the reader follow the dimensional discussion of Eq. (8).
  4. [Throughout] The term 'universal functional' is used for a three-nucleus fit; 'shared functional' is more accurate and is used elsewhere. Align terminology to avoid overclaiming.
  5. [Table IV] The table gives n_mu=64 as the energy readout convention; consider adding a note in the caption that this is a common comparison convention, not a convergence certificate, as the text already states.

Circularity Check

0 steps flagged

No significant circularity: per-nucleus fits are explicitly fitted, A=140 is a genuine holdout, and stability caveats are acknowledged limitations.

full rationale

The paper's derivation chain is self-contained and transparent about what is fitted. Per-nucleus results are training outcomes, not predictions: the fit matches the EL potential to a constant chemical potential on the KS density (Sec. III B 2) and imposes exact KS energy matching via Eq. (23), so the near-KS fixed points and small d values for training nuclei are expected fit quality, explicitly labeled as fits rather than holdout predictions. The load-bearing transfer claim (Sec. V) is the A=140 holdout: no A=140 observable or diagnostic was inspected during model selection, and the structural transfer d=0.074 is not an input to the fit. The selected-path energy-rise penalties (Sec. III B 6, Eq. 25) are soft constraints built from a recorded failure direction; the paper itself cautions that the paths should not be interpreted as direct measurements of a local Hessian, and it reports that the final active rows remain below the epsilon=2 MeV soft target, so the absence of the identified artifact on training nuclei is partly engineered. That is a limitation for certifying the absence of other unstable directions, not circularity: the A=140 holdout was not protected by those rows and remained stable. The only self-citation, Ref. [19] for the kinetic-rearrangement eigenproblem, is a numerical solver strategy cross-checked by independent imaginary-time evolution, so it is not load-bearing. The quantified limitations (percent-level quadrature sensitivity of the A=80 central density, GP 2-sigma coverage of 10/80 on A=140, and 41.7 MeV energy non-transfer) are explicitly stated and do not conceal a fitted quantity as a prediction.

Axiom & Free-Parameter Ledger

8 free parameters · 5 axioms · 2 invented entities

The fit-based scheme pulls most of its quantitative content from the KS reference: the kernel correction is regression on KS densities, energies, and EL potentials, so the ledger's free parameters are the learning problem itself. The paper's honesty reduces risk: it states the t-density-dependence is heuristic, the NLF is gauge-dependent, the finite-table norm is not a universal constant, and the ITE is not the exact gradient flow. The only invented object with out-of-sample evidence is the learned delta g (validated via A=140); the gradient term is explicitly optional and unvalidated for its stated purpose.

free parameters (8)
  • Kernel-basis coefficients c_b (b=1..14) of delta g(y) = not quoted in text (GP analogue: max|delta g|=0.098)
    Amplitudes of the 14 displaced-Gaussian correction to the Lindhard kernel; fit by active-set repeated penalized least squares to NLF, EL-potential, energy, and margin data (Sec. III A-B).
  • Local-correction shift s = s = -8.32492 (GP variant); Gaussian-basis values released
    Shifts c_CORR so the corrected kinetic energy matches KS on each training nucleus; absorbs the cutoff-dependent finite-table normalization error (Sec. II B).
  • Auxiliary chemical potentials mu_nu per training nucleus = not quoted
    Auxiliary fit parameters in the EL-potential matching block (Sec. III B 2, Appendix C Eq. C3).
  • Selected-path energy-rise margin epsilon = epsilon = 2 MeV (active rows at 0.64, 0.49, 1.66 MeV)
    Soft target for the three path-energy penalties built from the recorded A=80 depletion direction; hand-chosen, rows still active at convergence.
  • Tail-response margin chi_min = 5
    Hand-chosen target for the far-tail response-ratio penalty preventing artificial NLF=1 peaks (Sec. III B 5).
  • Data weights beta, eta_T, and penalty weights eta_H=12, eta_S=10 = beta=2 (beta_A=80=12); eta_T=0.3 (1.2 for 16O)
    Relative block weights tuned on the three training nuclei only; per-nucleus and shared fits use different values.
  • GP hyperparameters l_GP, sigma_f, kappa = (1.5, 0.3, 300)
    Selected by finite-grid marginal likelihood on training data; sigma_f lies at the grid edge, and the likelihood scale is re-scaled by kappa.
  • Finite-table and solver parameters (y_max, N_y, n_q, qbar_max, lambda, alpha_mix) = y_max=102.7-113.0, N_y=12840-14124, n_q=6000, qbar_max=80; lambda=3, alpha_mix=1
    Numerical representation choices; the paper states the finite-table norm is not universal and lambda cancels at the fixed point but affects convergence (Table I).
axioms (5)
  • domain assumption KS-DFT with the simplified SkP interaction provides valid reference densities, kinetic densities, and energies for the benchmark
    Whole pipeline regresses on rho_KS, tau_KS, E_KS (Sec. II A).
  • domain assumption Spherical N=Z systems without spin-orbit and Coulomb capture the shell-structure question
    Benchmark of Ref. [7] adopted; extension is deferred (Sec. VI).
  • ad hoc to paper Symmetrized pair-density scaling t = 3 pi^2 (rho(r)+rho(r'))/2 is the correct density dependence of the pair kernel f1
    Paper: 'a heuristic pair-density dependence, not a re-derivation from the Lindhard response' (Sec. II B).
  • ad hoc to paper The gauge-dependent local kinetic-energy density of Eq. (3) is a meaningful shell-structure diagnostic via NLF
    Paper acknowledges the local kinetic-energy density is not uniquely defined pointwise (Sec. II A).
  • domain assumption Adaptive-step ITE endpoints (R_EL = 0.02-0.05 MeV) are adequate stationarity for the reported densities
    ITE is not the exact discrete gradient flow of the first-difference energy (Appendix B); residual stationarity is approximate and quantified.
invented entities (2)
  • Learned kernel-shape correction delta g(y) (14 displaced Gaussians) independent evidence
    purpose: Supplies shell-structure physics missing from the Lindhard base kernel in the kinetic-energy functional
    A=140 holdout transfer of NLF structure (d=0.074) and radius (0.014 fm) is a falsifiable out-of-sample handle; per-nucleus fits are in-sample.
  • Gradient regularization term E_nabla (Eq. 29) with coefficient s2 no independent evidence
    purpose: Proposed limit-preserving regularizer for the central-depletion instability; implemented/released but unused in the final functional
    Paper shows it fails to repair the preliminary collapse and moves A=80 central density away from KS at s2=0.5 fm^4; labeled optional machinery (Sec. V).

pith-pipeline@v1.3.0-alltime-deepseek · 20702 in / 23957 out tokens · 222199 ms · 2026-07-31T23:44:20.578757+00:00 · methodology

0 comments
read the original abstract

Nonlocal kinetic-energy density functionals (KEDFs) can encode nuclear shell structure in orbital-free density functional theory (OFDFT), but self-consistency requires accurate functional derivatives and a stable solution of the Euler equation. We construct a density-dependent-kernel KEDF whose correction is learned from Kohn-Sham (KS) reference data. Linearity in the kernel shape yields analytic Euler--Lagrange (EL) responses. The fit uses exact energy-matching equalities and soft quadratic inequality penalties for violations of prescribed tail-response and selected-path energy-rise margins, evaluated by an active-set repeated penalized least-squares iteration. We formulate the radial EL equation as the rearranged one-orbital eigenproblem. In a constant-$k_F$ $^{16}$O benchmark it converges without density mixing and agrees with imaginary-time evolution (ITE) to sub-keV energy. Adaptive-step ITE gives final species EL residuals of at most 0.07 MeV in the reported calculations, and rearranged diagonalization reaches the same stationary densities. For spherical $N=Z$ systems without spin--orbit or Coulomb terms, nucleus-specific fits for $A=16$ to $140$ reproduce shell patterns and radii. A correction trained on three nuclei transfers the shell pattern and radius, but not the absolute energy, to a previously unseen $A=140$ system.

Figures

Figures reproduced from arXiv: 2607.23328 by Fumihiro Imoto.

Figure 1
Figure 1. Figure 1: FIG. 1. Value-only versus functional-derivative-informed [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Application of the per-nucleus procedure to the next [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. Shared multi-nucleus (“universal”) functional: a [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5. (a) Base dimensionless kernel [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6. Gaussian-process reformulation: posterior mean [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗

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

20 extracted references · 1 linked inside Pith

  1. [1]

    Hohenberg and W

    P. Hohenberg and W. Kohn, Phys. Rev.136, B864 (1964)

    1964

  2. [2]

    Kohn and L

    W. Kohn and L. J. Sham, Phys. Rev.140, A1133 (1965)

    1965

  3. [3]

    Bender, P.-H

    M. Bender, P.-H. Heenen, and P.-G. Reinhard, Rev. Mod. Phys.75, 121 (2003)

    2003

  4. [4]

    Y. A. Wang and E. A. Carter, inTheoretical Methods in Condensed Phase Chemistry(Kluwer, Dordrecht, 2000), pp. 117–184

    2000

  5. [5]

    Q. Xu, C. Ma, W. Mi, Y. Wang, and Y. Ma, WIREs Comput. Mol. Sci.14, e1724 (2024)

    2024

  6. [6]

    X. H. Wu, Z. X. Ren, and P. W. Zhao, Commun.Phys. 8, 316 (2025)

    2025

  7. [7]

    X. H. Wu, G. Col` o, K. Hagino, and P. W. Zhao, Phys. Rev. Lett.136, 092501 (2026)

    2026

  8. [8]

    Wang and M

    L.-W. Wang and M. P. Teter, Phys. Rev. B45, 13196 (1992)

    1992

  9. [9]

    Y. A. Wang, N. Govind, and E. A. Carter, Phys. Rev. B 60, 16350 (1999)

    1999

  10. [10]

    Huang and E

    C. Huang and E. A. Carter, Phys. Rev. B81, 045206 (2010)

    2010

  11. [11]

    L. A. Constantin, E. Fabiano, and F. Della Sala, Phys. Rev. B97, 205137 (2018)

    2018

  12. [12]

    W. Mi, A. Genova, and M. Pavanello, J. Chem. Phys. 148, 184107 (2018)

    2018

  13. [13]

    J. C. Snyder, M. Rupp, K. Hansen, K.-R. M¨ uller, and K. Burke, Phys. Rev. Lett.108, 253002 (2012)

    2012

  14. [14]

    L. Li, J. C. Snyder, I. M. Pelaschier, J. Huang, U.- N. Niranjan, P. Duncan, M. Rupp, K.-R. M¨ uller, and K. Burke, Int. J. Quantum Chem.116, 819 (2016)

    2016

  15. [15]

    Meyer, M

    R. Meyer, M. Weichselbaum, and A. W. Hauser, J. Chem. Theory Comput.16, 5685 (2020)

    2020

  16. [16]

    Brockherde, L

    F. Brockherde, L. Vogt, L. Li, M. E. Tuckerman, K. Burke, and K.-R. M¨ uller, Nat. Commun.8, 872 (2017)

    2017

  17. [17]

    X. H. Wu, Z. X. Ren, and P. W. Zhao, Phys. Rev. C105, L031303 (2022), arXiv:2105.07696

    Pith/arXiv arXiv 2022

  18. [18]

    Dobaczewski, H

    J. Dobaczewski, H. Flocard, and J. Treiner, Nucl. Phys. A422, 103 (1984)

    1984

  19. [19]

    Imoto, M

    F. Imoto, M. Imada, and A. Oshiyama, Phys. Rev. Re- search3, 033198 (2021)

    2021

  20. [20]

    Solin and S

    A. Solin and S. S¨ arkk¨ a, Stat. Comput.30, 419 (2020)

    2020