arxiv: 2604.25759 · v2 · submitted 2026-04-28 · ⚛️ nucl-th

Recognition: no theorem link

Neural-Network-Based Variational Method in Nuclear Density Functional Theory: Application to the Extended Thomas-Fermi Model

Kenta Yoshimura

Authors on Pith no claims yet

Pith reviewed 2026-05-12 01:46 UTC · model grok-4.3

classification ⚛️ nucl-th

keywords neural networkvariational methodnuclear density functional theoryextended Thomas-FermiSkyrme functionalnuclear pastabinding energy

0 comments

The pith

Neural networks represent nuclear densities and minimize Skyrme functionals directly in the extended Thomas-Fermi model.

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

The paper introduces a variational method for nuclear density functional theory in the extended Thomas-Fermi approximation by parametrizing proton and neutron densities with multilayer perceptrons. These densities are obtained through direct minimization of the Skyrme energy density functional using optimization over network parameters. Validation shows that binding energies for nuclei such as 40Ca, 90Zr, and 208Pb agree with prior ETF results to within 0.5 percent. The approach also reproduces nuclear pasta phases including spheres, rods, and slabs. Stationarity of the minimized parameters corresponds to a projected Euler-Lagrange condition restricted to the neural-network density manifold, and single-precision arithmetic produces results comparable to double precision.

Core claim

Proton and neutron number densities are represented by multilayer perceptrons and determined by direct minimization of a Skyrme-type energy density functional in the extended Thomas-Fermi model. Stationarity in parameter space corresponds to a projected Euler-Lagrange condition on the neural-network trial-density manifold. This yields binding energies for nuclei such as 40Ca, 90Zr, and 208Pb that agree with existing ETF calculations to within 0.5 percent, while reproducing representative pasta structures including spheres, rods, and slabs.

What carries the argument

Multilayer perceptrons as trial density functions whose parameters are optimized by direct minimization of the Skyrme energy functional, enforcing a projected Euler-Lagrange condition on the representable density manifold.

If this is right

Binding energies of finite nuclei can be computed variationally without directly solving the Euler-Lagrange differential equations.
Nuclear pasta phases with different geometries are obtained from the same minimization procedure applied to periodic boundary conditions.
Single-precision arithmetic suffices for results comparable to double precision, enabling efficient GPU implementations.
The framework supplies a general variational route for other density functionals once a suitable neural-network density representation is chosen.

Where Pith is reading between the lines

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

The same density parametrization could be extended to time-dependent or finite-temperature problems where conventional differential solvers become cumbersome.
Network architectures could be modified to embed additional symmetries or physical constraints directly into the trial-density manifold.
The approach may scale to three-dimensional calculations for larger systems once parallel gradient evaluations on GPUs are utilized.

Load-bearing premise

The manifold of densities that can be represented by the chosen multilayer perceptrons is rich enough to reach the true minimum of the Skyrme functional without significant bias from network architecture or initialization.

What would settle it

Running the minimization with networks of substantially increased depth or width and checking whether the resulting binding energies and density profiles converge to the same values obtained by established ETF solvers or diverge in a systematic way.

Figures

Figures reproduced from arXiv: 2604.25759 by Kenta Yoshimura.

**Figure 1.** Figure 1: Density distributions obtained for representative pasta phases at a view at source ↗

read the original abstract

We propose a neural-network-based variational framework for nuclear Density Functional Theory based on the extended Thomas--Fermi (ETF) model, in which proton and neutron number densities are represented by multilayer perceptrons and determined by direct minimization of a Skyrme-type energy density functional. We clarify the mathematical connection to the conventional Euler--Lagrange formulation, showing that stationarity in parameter space corresponds to a projected Euler--Lagrange condition on the neural-network trial-density manifold. The basic validity of the framework is examined through three sets of calculations: a Woods--Saxon potential benchmark, ground-state calculations of finite nuclei ($^{40}$Ca, $^{90}$Zr, and $^{208}$Pb), and nuclear pasta phases. The binding energies of finite nuclei agree with existing ETF calculations to within $0.5\%$, and representative pasta structures including spheres, rods, and slabs are reproduced. We also find that single-precision arithmetic yields results comparable to double precision, suggesting that the present framework is well suited to GPU environments in which low-precision computation is advantageous.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper introduces a direct variational minimization of the Skyrme functional using multilayer perceptrons for densities in the ETF model, with 0.5% binding energy agreement and correct pasta phases, but without checks that the network manifold actually reaches the unrestricted minimum.

read the letter

This paper introduces a variational method for the extended Thomas-Fermi model in which proton and neutron densities are represented by multilayer perceptrons and the Skyrme energy density functional is minimized directly with respect to the network parameters. The reported results show binding energies for 40Ca, 90Zr, and 208Pb agreeing with standard ETF calculations to within 0.5 percent, along with reproduction of sphere, rod, and slab pasta structures on benchmark sets. The work also notes that single-precision arithmetic gives comparable outcomes, which is relevant for GPU use. The mathematical clarification that parameter-space stationarity corresponds to a projected Euler-Lagrange condition on the neural-network manifold is laid out clearly and follows directly from the variational setup. The benchmarks on the Woods-Saxon potential and the finite nuclei plus pasta cases are presented in a straightforward way and the numbers line up with existing methods. The practical angle on low-precision computation is a useful observation for anyone thinking about scaling these calculations. The main soft spot is the lack of evidence that the chosen multilayer perceptrons are expressive enough to reach or closely approach the true ETF minimum rather than a restricted local solution. No scans over network depth, width, activation functions, or multiple initializations are described to test whether the 0.5 percent agreement holds or shifts with greater capacity. Without those checks, it is hard to know how much the results depend on the specific architecture or starting point. Convergence details and error estimates on the minimization are also not spelled out in enough depth to judge robustness. This work is aimed at nuclear theorists already using density functionals who need flexible tools for irregular geometries or GPU acceleration. A reader interested in trying variational approaches beyond traditional grids will get concrete implementation ideas and validation numbers. It deserves a serious referee because the technique is new within the ETF program, the basic consistency checks are in place, and the GPU relevance is timely, even if extra tests on network expressivity would tighten the claims.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes a neural-network variational framework for the extended Thomas-Fermi (ETF) model in Skyrme nuclear DFT. Proton and neutron densities are represented by multilayer perceptrons whose parameters are optimized by direct minimization of the energy density functional. The authors derive that stationarity with respect to network parameters corresponds to a projected Euler-Lagrange condition on the trial-density manifold. Validation consists of a Woods-Saxon benchmark, ground-state calculations for ^{40}Ca, ^{90}Zr and ^{208}Pb (binding energies agree with conventional ETF to 0.5%), and reproduction of nuclear pasta phases (spheres, rods, slabs). Single-precision arithmetic is shown to yield comparable results, suggesting GPU suitability.

Significance. If the reported accuracy holds under broader testing, the method supplies a differentiable, GPU-native route to ETF solutions that could scale to larger systems or more elaborate functionals. The explicit mapping from parameter-space stationarity to the projected Euler-Lagrange equation is a clear conceptual contribution. The single-precision result is practically useful. The work does not yet demonstrate that the chosen MLP manifold reaches the unrestricted ETF minimum for all cases of interest.

major comments (2)

[finite-nuclei calculations] Results section on finite nuclei: the 0.5% binding-energy agreement is stated for ^{40}Ca, ^{90}Zr and ^{208}Pb, yet no systematic variation of network depth, width, activation function or random initialization is reported. Without such controls it remains possible that the observed agreement reflects a restricted trial manifold rather than faithful approximation of the unrestricted ETF solution; this directly affects the central claim that the method reproduces standard ETF results.
[pasta phases] Nuclear pasta section: representative density profiles for spheres, rods and slabs are shown, but no quantitative comparison (energy per nucleon, rms density deviation, or surface tension) to conventional ETF calculations is supplied. Visual reproduction alone does not establish that the minimization has reached the same physical minimum as the unrestricted functional.

minor comments (2)

[method] The optimization algorithm, learning-rate schedule and convergence criteria are not specified in sufficient detail to allow independent reproduction of the reported minima.
[energy density functional] The manuscript would benefit from an explicit statement of the precise Skyrme parametrization (e.g., SLy4 or SkM*) and the numerical values of the coupling constants used in all benchmarks.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading of our manuscript and the constructive comments, which help clarify the strengths and limitations of the presented neural-network variational approach. We address each major comment below and will revise the manuscript to incorporate additional evidence where needed.

read point-by-point responses

Referee: Results section on finite nuclei: the 0.5% binding-energy agreement is stated for ^{40}Ca, ^{90}Zr and ^{208}Pb, yet no systematic variation of network depth, width, activation function or random initialization is reported. Without such controls it remains possible that the observed agreement reflects a restricted trial manifold rather than faithful approximation of the unrestricted ETF solution; this directly affects the central claim that the method reproduces standard ETF results.

Authors: We agree that the absence of systematic hyperparameter studies leaves open the possibility that the reported agreement arises from a limited trial manifold. The Woods-Saxon benchmark already shows that the chosen MLP architecture can represent smooth densities to high accuracy, but this does not fully address the finite-nuclei cases. In the revised manuscript we will add binding-energy results obtained with networks of varying width (20–100 neurons per layer) and depth (1–3 hidden layers), together with statistics from multiple random initializations. These controls will demonstrate convergence of the energies to within the stated 0.5% and thereby strengthen the claim that the neural-network solutions faithfully approximate the unrestricted ETF minima. revision: yes
Referee: Nuclear pasta section: representative density profiles for spheres, rods and slabs are shown, but no quantitative comparison (energy per nucleon, rms density deviation, or surface tension) to conventional ETF calculations is supplied. Visual reproduction alone does not establish that the minimization has reached the same physical minimum as the unrestricted functional.

Authors: We accept that visual similarity of density profiles is not sufficient to confirm that the same physical minima have been reached. In the revision we will supplement the pasta-phase figures with quantitative metrics: energy per nucleon for each geometry, root-mean-square density deviations relative to conventional ETF solutions, and surface-tension estimates extracted from the density profiles. These data will be presented in a new table or supplementary figure, allowing direct numerical comparison and thereby establishing that the neural-network minimizations reproduce the unrestricted ETF results to the same level of accuracy reported for finite nuclei. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained

full rationale

The paper derives the equivalence between parameter-space stationarity and a projected Euler-Lagrange condition directly from the variational principle applied to the neural-network parametrization of the density (standard chain-rule argument on the trial manifold). This is presented as a clarification of the method rather than an input assumption or reduction of the central numerical claims. Binding-energy agreements and pasta-phase reproductions are obtained from explicit numerical minimization of the Skyrme functional and compared against independent prior ETF results; no fitted parameter is renamed as a prediction, no self-citation chain supports a load-bearing uniqueness claim, and the core results do not reduce by construction to the paper's own inputs or equations.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The framework rests on the standard extended Thomas-Fermi kinetic-energy functional and a Skyrme-type energy density functional taken from prior nuclear-physics literature; the neural network itself is a computational representation rather than a new physical entity.

free parameters (2)

Neural-network architecture (number of layers and neurons)
Chosen by the authors to represent the density functions; affects the expressivity of the trial manifold.
Skyrme EDF coupling constants
Taken from existing parameter sets in the nuclear-physics literature and held fixed during the variational minimization.

axioms (2)

domain assumption The extended Thomas-Fermi approximation adequately captures the kinetic-energy contribution for the nuclei and pasta phases considered.
The entire calculation is performed inside the ETF model rather than full quantum DFT.
domain assumption Direct minimization over the neural-network parameter space reaches a stationary point that corresponds to the physical ground state within the ETF manifold.
Invoked when the authors equate parameter-space stationarity to the projected Euler-Lagrange condition.

pith-pipeline@v0.9.0 · 5484 in / 1623 out tokens · 48596 ms · 2026-05-12T01:46:14.699466+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Reference graph

Works this paper leans on

34 extracted references · 34 canonical work pages

[1]

, author Brack, M

author Bartel, J. , author Brack, M. , author Durand, M. , year 1985 . title Extended thomas-fermi theory at finite temperature . journal Nuclear Physics A volume 445 , pages 263--303 . :10.1016/0375-9474(85)90071-5

work page doi:10.1016/0375-9474(85)90071-5 1985
[2]

, author Heenen, P.H

author Bender, M. , author Heenen, P.H. , author Reinhard, P.G. , year 2003 . title Self-consistent mean-field models for nuclear structure . journal Reviews of Modern Physics volume 75 , pages 121--180 . :10.1103/RevModPhys.75.121

work page doi:10.1103/revmodphys.75.121 2003
[3]

, year 1984

author Brack, M. , year 1984 . title Extended Thomas-Fermi Model at Finite Temperature . journal Phys. Rev. Lett. volume 53 , pages 119--122 . https://link.aps.org/doi/10.1103/PhysRevLett.53.119

work page doi:10.1103/physrevlett.53.119 1984
[4]

, author Guet, C

author Brack, M. , author Guet, C. , author H kansson, H.B. , year 1985 . title Selfconsistent semiclassical description of average nuclear properties---a link between microscopic and macroscopic models . journal Physics Reports volume 123 , pages 275--364 . :10.1016/0370-1573(86)90078-5

work page doi:10.1016/0370-1573(86)90078-5 1985
[5]

Carleo and M

author Carleo, G. , author Troyer, M. , year 2017 . title Solving the quantum many-body problem with artificial neural networks . journal Science volume 355 , pages 602--606 . :10.1126/science.aag2302

work page doi:10.1126/science.aag2302 2017
[6]

, author Pi, M

author Centelles, M. , author Pi, M. , author Vi\ nas, X. , author Garcias, F. , author Barranco, M. , year 1990 . title Self-consistent extended thomas-fermi calculations in nuclei . journal Nuclear Physics A volume 510 , pages 397--416 . :10.1016/0375-9474(90)90058-T

work page doi:10.1016/0375-9474(90)90058-t 1990
[7]

, author Schuck, P

author Centelles, M. , author Schuck, P. , author Vi\ nas, X. , year 2007 . title Thomas--fermi theory for atomic nuclei revisited . journal Annals of Physics volume 322 , pages 363--396 . :10.1016/j.aop.2006.07.009

work page doi:10.1016/j.aop.2006.07.009 2007
[8]

, author Shchechilin, N.N

author Chamel, N. , author Shchechilin, N.N. , author Chugunov, A.I. , year 2025 . title Pressure and chemical potentials in the inner crust of a cold neutron star within hartree-fock and extended thomas-fermi methods . journal Phys. Rev. C volume 111 , pages 015805 . https://link.aps.org/doi/10.1103/PhysRevC.111.015805, :10.1103/PhysRevC.111.015805

work page doi:10.1103/physrevc.111.015805 2025
[9]

, year 2020

author Col^^c3^^b2, G. , year 2020 . title Nuclear density functional theory . journal Advances in Physics: X volume 5 , pages 1740061 . :10.1080/23746149.2020.1740061

work page doi:10.1080/23746149.2020.1740061 2020
[10]

, author Arcoragi, J.P

author Dutta, A. , author Arcoragi, J.P. , author Pearson, J. , author Behrman, R. , author Tondeur, F. , year 1986 . title Thomas-fermi approach to nuclear mass formula: (i). spherical nuclei . journal Nuclear Physics A volume 458 , pages 77--94 . https://www.sciencedirect.com/science/article/pii/0375947486902836, :https://doi.org/10.1016/0375-9474(86)90283-6

work page doi:10.1016/0375-9474(86)90283-6 1986
[11]

a tzle, Z. , author No \'e , F. , year 2020 . title Deep-neural-network solution of the electronic schr \

author Hermann, J. , author Sch \"a tzle, Z. , author No \'e , F. , year 2020 . title Deep-neural-network solution of the electronic schr \"o dinger equation . journal Nature Chemistry volume 12 , pages 891--897 . :10.1038/s41557-020-0544-y

work page doi:10.1038/s41557-020-0544-y 2020
[12]

Hohenberg and W

author Hohenberg, P. , author Kohn, W. , year 1964 . title Inhomogeneous electron gas . journal Physical Review volume 136 , pages B864--B871 . :10.1103/PhysRev.136.B864

work page doi:10.1103/physrev.136.b864 1964
[13]

Multilayer feedforward networks are universal approximators , journal =

author Hornik, K. , author Stinchcombe, M. , author White, H. , year 1989 . title Multilayer feedforward networks are universal approximators . journal Neural Networks volume 2 , pages 359--366 . :10.1016/0893-6080(89)90020-8

work page doi:10.1016/0893-6080(89)90020-8 1989
[14]

Physical Review B , author =

author Ismail-Beigi, S. , year 2006 . title Truncation of periodic image interactions for confined systems . journal Physical Review B volume 73 , pages 233103 . :10.1103/PhysRevB.73.233103

work page doi:10.1103/physrevb.73.233103 2006
[15]

, author Roche, K.J

author Jin, S. , author Roche, K.J. , author Stetcu, I. , author Abdurrahman, I. , author Bulgac, A. , year 2021 . title The LISE package: Solvers for static and time-dependent superfluid local density approximation equations in three dimensions . journal Computer Physics Communications volume 269 , pages 108130 . https://www.sciencedirect.com/science/art...

work page 2021
[16]

Machine learning: Trends, perspectives, and prospects

author Jordan, M.I. , author Mitchell, T.M. , year 2015 . title Machine learning: Trends , perspectives, and prospects . journal Science volume 349 , pages 255--260 . https://www.science.org/doi/10.1126/science.aaa8415

work page doi:10.1126/science.aaa8415 2015
[17]

, author Lu, H

author Kashi, A. , author Lu, H. , author Brewer, W. , author Rogers, D. , author Matheson, M. , author Shankar, M. , author Wang, F. , year 2024 . title Mixed-precision numerics in scientific applications: Survey and perspectives . journal arXiv preprint arXiv:2412.19322 https://arxiv.org/abs/2412.19322, http://arxiv.org/abs/2412.19322 arXiv:2412.19322

work page arXiv 2024
[18]

, et al., year 2021

author Kirkpatrick, J. , et al., year 2021 . title Pushing the frontiers of density functionals by solving the fractional electron problem . journal Science volume 374 , pages 1385--1389 . :10.1126/science.abj6511

work page doi:10.1126/science.abj6511 2021
[19]

Self-consistent equations including exchange and correlation effects.Physical Review, 140(4A):A1133–A1138, 1965

author Kohn, W. , author Sham, L.J. , year 1965 . title Self-consistent equations including exchange and correlation effects . journal Physical Review volume 140 , pages A1133--A1138 . :10.1103/PhysRev.140.A1133

work page doi:10.1103/physrev.140.a1133 1965
[20]

, author Bengio, Y

author LeCun, Y. , author Bengio, Y. , author Hinton, G. , year 2015 . title Deep learning . journal Nature volume 521 , pages 436--444 . https://www.nature.com/articles/nature14539

work page 2015
[21]

, author Narang, S

author Micikevicius, P. , author Narang, S. , author Alben, J. , author Diamos, G.F. , author Elsen, E. , author Garcia, D. , author Ginsburg, B. , author Houston, M. , author Kuchaiev, O. , author Venkatesh, G. , author Wu, H. , year 2018 . title Mixed precision training , in: booktitle International Conference on Learning Representations ( ICLR ) . http...

work page 2018
[22]

Deepcoder: A fully open-source 14b coder at o3-mini level

author Micikevicius, P. , author Stosic, D. , author Burgess, N. , author Cornea, M. , author Dubey, P. , author Grisenthwaite, R. , author Ha, S. , author Heinecke, A. , author Judd, P. , author Kamalu, J. , author Mellempudi, N. , author Oberman, S. , author Shoeybi, M. , author Siu, M. , author Wu, H. , year 2022 . title FP8 formats for deep learning ....

work page arXiv 2022
[23]

, author Vaishay, S

author Mittal, S. , author Vaishay, S. , year 2019 . title A survey of techniques for optimizing deep learning on GPU s . journal Journal of Systems Architecture volume 99 , pages 101635 . :10.1016/j.sysarc.2019.101635

work page doi:10.1016/j.sysarc.2019.101635 2019
[24]

, year 2026

author Nakamura, Y. , year 2026 . title Extended Thomas-Fermi Calculations for Nuclear Pasta in the Inner Crust of Neutron Stars . Master's thesis. Institute of Science Tokyo

work page 2026
[25]

, author Sekizawa, K

author Nakamura, Y. , author Sekizawa, K. , year 2026 . note In preparation

work page 2026
[26]

, author Matsuyanagi, K

author Nakatsukasa, T. , author Matsuyanagi, K. , author Matsuo, M. , author Yabana, K. , year 2016 . title Time-dependent density-functional description of nuclear dynamics . journal Rev. Mod. Phys. volume 88 , pages 045004 . https://link.aps.org/doi/10.1103/RevModPhys.88.045004, :10.1103/RevModPhys.88.045004

work page doi:10.1103/revmodphys.88.045004 2016
[27]

, author Chamel, N

author Pearson, J.M. , author Chamel, N. , author Potekhin, A.Y. , author Fantina, A.F. , author Ducoin, C. , author Dutta, A.K. , author Goriely, S. , year 2018 . title Unified equations of state for cold non-accreting neutron stars with Brussels--Montreal functionals -- i. role of symmetry energy . journal Monthly Notices of the Royal Astronomical Socie...

work page doi:10.1093/mnras/sty2413 2018
[28]

, author Spencer, J.S

author Pfau, D. , author Spencer, J.S. , author Matthews, A.G.D.G. , author Foulkes, W.M.C. , year 2020 . title Ab initio solution of the many-electron schr \"o dinger equation with deep neural networks . journal Physical Review Research volume 2 , pages 033429 . :10.1103/PhysRevResearch.2.033429

work page doi:10.1103/physrevresearch.2.033429 2020
[29]

, author Klatt, M.A

author Schuetrumpf, B. , author Klatt, M.A. , author Iida, K. , author Maruhn, J.A. , author Mecke, K. , author Reinhard, P.G. , year 2015 . title Time-dependent hartree-fock approach to nuclear ``pasta'' at finite temperature . journal Physical Review C volume 92 , pages 045806 . :10.1103/PhysRevC.92.045806

work page doi:10.1103/physrevc.92.045806 2015
[30]

, author Mart\' nez-Pinedo, G

author Schuetrumpf, B. , author Mart\' nez-Pinedo, G. , author Afibuzzaman, M. , author Aktulga, H.M. , year 2019 . title Survey of nuclear pasta in the intermediate-density regime: Shapes and energies . journal Phys. Rev. C volume 100 , pages 045806 . https://link.aps.org/doi/10.1103/PhysRevC.100.045806, :10.1103/PhysRevC.100.045806

work page doi:10.1103/physrevc.100.045806 2019
[31]

, author Chamel, N

author Shchechilin, N.N. , author Chamel, N. , author Pearson, J.M. , author Chugunov, A.I. , author Potekhin, A.Y. , year 2024 . title Unified equations of state for cold nonaccreting neutron stars with brussels-montreal functionals. v. improved parametrization of the nucleon density distributions . journal Phys. Rev. C volume 109 , pages 055802 . https:...

work page doi:10.1103/physrevc.109.055802 2024
[32]

A survey on deep learning hardware accelerators for heterogeneous hpc platforms,

author Silvano, C. , author Ielmini, D. , author Ferrandi, F. , author Fiorin, L. , author Curzel, S. , author Benini, L. , author Conti, F. , author Garofalo, A. , author Zambelli, C. , author Calore, E. , author Schifano, S.F. , author Palesi, M. , author Ascia, G. , author Patti, D. , author Petra, N. , author De Caro, D. , author Lavagno, L. , author ...

work page doi:10.1145/3729215 2025
[33]

, author Saxon, D.S

author Woods, R.D. , author Saxon, D.S. , year 1954 . title Diffuse surface optical model for nucleon-nuclei scattering . journal Physical Review volume 95 , pages 577--578 . :10.1103/PhysRev.95.577

work page doi:10.1103/physrev.95.577 1954
[34]

, author Piekarewicz, J

author Yang, J. , author Piekarewicz, J. , year 2020 . title Covariant density functional theory in nuclear physics and astrophysics . journal Annual Review of Nuclear and Particle Science volume 70 , pages 21--41 . :10.1146/annurev-nucl-101918-023608

work page doi:10.1146/annurev-nucl-101918-023608 2020