Constructing Inverse Potentials from Scattering Phase Shifts using Physics-Informed Neural Networks: Application to Neutron-Alpha Scattering
Ayushi Awasthi, Ishwar Kant, Arushi Sharma, M. R. Ganesh Kumar, and O. S. K. S. Sastri, [8pt] Department of Physics, Astronomical Sciences Central University of Himachal Pradesh, Dharamshala, India, [4pt] Applied Materials India Private Limited, Bengaluru 560066, India, [6pt] Corresponding author : [email protected] · 2026
Read Constructing Inverse Potentials from Scattering Phase Shifts using Physics-Informed Neural Networks: Application to Neutron-Alpha Scattering by Ayushi Awasthi, Ishwar Kant, Arushi Sharma, M. R. Ganesh Kumar, and O. S. K. S. Sastri, [8pt] Department of Physics, Astronomical Sciences Central University of Himachal Pradesh, Dharamshala, India, [4pt] Applied Materials India Private Limited, Bengaluru 560066, India, [6pt] Corresponding author : [email protected] (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-02264.txt.
Introduction
Ayushi Awasthi and colleagues build a physics-informed neural network to recover a nuclear potential from scattering phase shifts, an inverse problem in nuclear physics. They apply it to the P-wave of neutron-alpha elastic scattering. The radial potential is a feed-forward network whose output is multiplied by a Gaussian envelope, so the finite range of the force is built into the structure itself.
That hard constraint proves essential. Without the envelope, training gave potentials with non-vanishing tails and phase shifts that never matched the data, however long it ran. With it, phase shifts are computed at each energy by integrating a variable-phase equation, making the whole pipeline differentiable. Training converged stably to low loss in the reported run.
Catalog record
| CREATOR | Ayushi Awasthi, Ishwar Kant, Arushi Sharma, M. R. Ganesh Kumar, and O. S. K. S. Sastri, [8pt] Department of Physics, Astronomical Sciences Central University of Himachal Pradesh, Dharamshala, India, [4pt] Applied Materials India Private Limited, Bengaluru 560066, India, [6pt] Corresponding author : [email protected] |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 39 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.02264
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Constructing Inverse Potentials from Scattering Phase Shifts using Physics-Informed Neural Networks: Application to Neutron-Alpha Scattering Ayushi Awasthi, Ishwar Kant, Arushi Sharma, M. R. Ganesh Kumar, and O. S. K. S. Sastri, [8pt] Department of Physics, Astronomical Sciences Central University of Himachal Pradesh, Dharamshala, India, [4pt] Applied Materials India Private Limited, Bengaluru 560066, India, [6pt] Corresponding author : [email protected] Abstract We develop a physics-informed neural networks (PINNs) framework for the inverse scattering problem in nuclear physics and apply it to the P 3 / 2 P_{3/2} partial wave of neutron-alpha elastic scattering. The radial potential is represented by a feed-forward network whose output is multiplied by a Gaussian envelope, embedding the finite-range condition directly into the architecture rather than through a soft penalty term. This distinction proves essential: without the envelope, the optimizer produces potentials with non-vanishing tails and the resulting phase shifts remain inconsistent with the data regardless of training duration, demonstrating that hard structural constraints are indispensable for physically meaningful solutions to nuclear inverse problems. Phase shifts are generated at each scattering energy by numerically integrating the variable-phase equation with a fourth-order Runge-Kutta scheme, making the entire pipeline end-to-end differentiable. Training converges stably to a loss near 3 × 10 − 4 3\times 10^{-4} and recovers a smooth, purely attractive central potential with a well depth of − 60.47 -60.47 MeV. Adding the centrifugal barrier to the learned potential reveals a well-defined barrier-well structure that naturally accounts for the P 3 / 2 P_{3/2} resonance. The extracted resonance parameters, E r = 0.95 E_{r}=0.95 MeV and Γ r = 0.78 \Gamma_{r}=0.78 MeV, together with the P-wave effective-range parameters, are in good agreement with expected values. A leave-one-out analysis confirms that the reconstruction is stable against the removal of any single data point. These results establish physics-guided machine learning as a reliable route to potential reconstruction from nuclear scattering data. 1 Introduction Nuclear scattering experiments have long served as one of the most important experimental tools for probing the nature of nuclear forces [ 1 ] . When two nuclei collide at low to intermediate energies, the angular distribution and energy dependence of the scattered particles carry detailed information about the underlying interaction potential [ 2 ] . Extracting that potential from measured data, the so-called inverse scattering problem, has therefore been a central goal of nuclear physics since the earliest days of the field [ 2 , 3 ] . The importance of this problem extends well beyond pure theory: reliable nuclear potentials are essential inputs to reaction network calculations in astrophysics, serve as benchmarks for ab initio many-body methods, and find direct application in nuclear data evaluation [ 4 ] . Despite this long history, the inverse problem remains genuinely difficult. It is inherently ill-posed, many distinct potentials can reproduce the same finite set of phase shifts, and any method that aspires to be reliable must impose additional physical information to render the solution unique and stable. Among the lightest nuclear systems, the neutron–alpha (n– \alpha ) scattering process is one of the simplest nucleon–nucleus systems, as it involves a neutral projectile and a tightly bound, spin-zero \alpha particle, whose internal degrees of freedom are effectively frozen at low scattering energies, reducing the problem to a single nucleon interacting with an inert core. The phase-shift data for this system have been available since the late 1960s over a broad energy range, with the measurements of Satchler et al. providing a benchmark dataset spanning laboratory energies from well below 1 MeV to approximately 18 MeV [ 5 ] . Despite its simplicity, the system exhibits nontrivial dynamics, including the well-known 5 He resonance in the P_{3/2} channel near E_{r}\approx 0.92 MeV (c.m.), which provides a sensitive test of the P-wave interaction [ 6 , 7 ] . Several ab initio calculations, including recent chiral effective field theory approaches, have attempted to describe the n- \alpha system from first principles [ 8 , 7 ] , but constructing a simple, physically transparent effective potential directly from scattering data remains a relevant and instructive problem in its own right. Efforts to reconstruct the n- \alpha interaction from phase shifts have a long history, and several methods have been developed with varying degrees of success. Phenomenological optical models represent the potential as a parametric function, most commonly in the Woods-Saxon form [ 9 ] , with parameters adjusted until the computed phase shifts or cross sections match experimental data. This approach is straightforward and widely used, but the choice of functional form introduces a bias that is difficult to quantify: if the true potential does not resemble the assumed well shape, the fit may remain accurate only within the energy range used for fitting and fail to predict properties outside it. R-matrix theory [ 10 , 11 ] offers a more systematic framework by expanding the interior wave function in terms of energy-independent basis states defined within a finite matching radius. It connects naturally to reaction theory and has been applied extensively to light nuclear systems, but still requires choosing the channel radius and the number of basis states, introducing model dependence at another level. For a system like 5 He, where the resonance is broad and overlaps with the continuum, convergence of the pole expansion is not guaranteed across the full energy range of interest. Direct inversion methods, such as those based on the Gel’fand-Levitan or Marchenko integral equations [ 3 ] , are exact in principle: given complete phase-shift information over all energies, they uniquely determine the potential. In practice, however, experimental data are available only at discrete energies and over a finite energy range, so the input must be supplemented by theoretical assumptions at unmeasured energies. The resulting potential is sensitive to these assumptions, and the propagation of experimental uncertainties through the inversion kernel is technically demanding. Recently our group has addressed the reconstruction problem for several nuclear scattering systems using a genetic algorithm combined with a reference potential approach, recovering inverse potentials for \alpha - 12 C, nucleon-deuteron, n - p , p - p , and \alpha -deuteron scattering [ 12 , 13 , 14 , 15 , 16 ] . That body of work demonstrates that population-based optimization can identify physically meaningful potentials even for resonant and near-threshold systems. However, metaheuristic methods of this kind are computationally intensive and, crucially, they lack analytical gradients. The absence of gradient information limits scalability to higher-dimensional parameter spaces and makes systematic uncertainty quantification through standard statistical tools considerably harder. These limitations motivate a gradient-based alternative. Over the past decade, machine learning has permeated nuclear and hadronic physics [ 17 ] , finding applications that range from interpolating nuclear masses [ 18 ] and emulating density functionals [ 19 ] to accelerating coupled-cluster and no-core shell-model calculations [ 20 ] . In nuclear reactions specifically, machine learning emulators have been constructed for optical model cross sections [ 21 ] and astrophysical S -factors [ 22 ] , and neural network wave functions have recently been used to solve the many-body Schrödinger equation for light nuclei [ 23 ] . Within this broader trend, physics-informed neural networks (PINNs) [ 24 , 25 ] have emerged as a particularly compelling tool for