⚛️ nuclear theory

Reduced basis emulator for elastic scattering in continuum-discretized coupled-channels calculations

This paper presents a reduced basis emulator for continuum-discretized coupled-channels calculations that utilizes proper orthogonal decomposition and Galerkin projection to achieve approximately 100-fold speedups with sub-percent accuracy, thereby enabling efficient uncertainty quantification and Bayesian parameter estimation for nuclear reaction studies.

Original authors: Jin Lei

Published 2026-08-28

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Jin Lei

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the heart of the atom, where protons and neutrons bind together, lies a world of delicate balance. Some atomic nuclei are tightly packed and stable, but others are loosely bound, teetering on the edge of falling apart. When these fragile, weakly bound nuclei collide with other atoms, they do not simply bounce off like billiard balls. Instead, they can stretch, break apart, and reassemble in complex ways during the encounter. To understand how these collisions happen, physicists use a sophisticated theoretical framework called the continuum-discretized coupled-channels method. This approach treats the collision as a dance between the intact nucleus and the fragments it might break into, calculating how the different possible outcomes influence one another. While this method is essential for predicting how these exotic nuclei behave, it comes with a heavy price: the calculations are so computationally demanding that they can take hours or even days to run a single scenario. This slowness has made it nearly impossible to perform the rigorous statistical analyses needed to pin down the exact properties of these nuclear interactions, leaving a gap between theory and the precise data emerging from modern experiments.

A researcher at Tongji University has now built a powerful new tool to bridge this gap, creating a fast and accurate "emulator" that can predict the results of these complex nuclear collisions in a fraction of the time. By using a mathematical technique that identifies the underlying patterns in a set of pre-calculated solutions, the researcher developed a system that can mimic the full, slow calculation with remarkable speed. The goal was to create a shortcut that does not sacrifice precision, allowing scientists to explore thousands of different scenarios in the time it used to take to run just one. The result is a system that can predict how a weakly bound projectile, such as a deuteron (a nucleus made of one proton and one neutron), scatters off a target nucleus like nickel-58, while simultaneously accounting for eighteen different variables that describe the forces at play.

The core of this new method relies on the observation that even though the physics of these collisions is complex, the solutions change smoothly as the input parameters change. If you slightly adjust the strength of the force holding the nucleus together, the resulting collision pattern does not change wildly; it shifts in a predictable, gradual way. The researcher exploited this smoothness by first running the full, slow calculation for a carefully selected set of different parameter combinations. These results, known as "snapshots," were then analyzed to find a small, optimal set of building blocks that could be combined to reconstruct any other solution within that range. Instead of solving the massive, complicated equations from scratch every time, the emulator simply mixes these pre-computed building blocks in the right proportions to generate a new answer. This process, known as a reduced basis approach, transforms a problem that once required solving thousands of interconnected equations into a much smaller, manageable calculation.

To test the reliability of this emulator, the researcher applied it to the specific case of a deuteron striking a nickel-58 nucleus at an energy of 21.6 million electron volts. The system was trained on a wide range of possible values for the eighteen parameters that define the interaction between the projectile's parts and the target. Once trained, the emulator was asked to predict the outcome for five completely new sets of parameters that it had never seen before. The results were strikingly accurate. Across a vast range of conditions, the emulator reproduced the elastic scattering cross sections—the measure of how likely the nuclei are to bounce off each other—with an error rate of less than 0.1 percent. In some cases, the difference between the emulator's prediction and the exact, slow calculation was so small it was barely measurable, confirming that the shortcut did not introduce any significant distortion to the physics.

The speedup achieved by this method is transformative. While a single full calculation for this specific nuclear system takes about 6.5 seconds on a modern server, the emulator can produce a prediction in just 30 milliseconds. This represents a speed increase of roughly 220 times. This leap in efficiency changes what is possible for nuclear physicists. Previously, performing a statistical analysis to understand the uncertainties in nuclear models required running the slow calculation tens of thousands of times, a task that would take months or years and was often deemed too expensive to attempt. With the emulator, that same analysis can be completed in a matter of hours. This capability opens the door to rigorous uncertainty quantification, allowing scientists to not only predict reaction outcomes but also to quantify exactly how confident they are in those predictions.

The study demonstrates that this approach works well even when dealing with a high-dimensional space of eighteen different variables, a complexity that often stumps other approximation methods. The researcher found that a training set of just 200 snapshots was sufficient to capture the essential behavior of the system across the entire range of interest, suggesting that the method scales efficiently and does not require an exponential increase in data as the number of variables grows. By successfully emulating the scattering of deuterons on nickel-58, the work proves that it is possible to retain the high fidelity of the full theoretical framework while removing the computational bottlenecks that have long hindered progress. This development marks a significant step forward, bringing the field of nuclear reaction theory into an era where precise, data-driven comparisons between theory and experiment can be made routinely, ultimately helping to refine our understanding of the forces that hold the atomic nucleus together.

Technical Summary: Reduced Basis Emulator for Elastic Scattering in CDCC Calculations

Problem Statement
Continuum-Discretized Coupled-Channels (CDCC) calculations are the standard theoretical framework for describing nuclear reactions involving weakly bound projectiles, where the coupling to breakup channels significantly modifies observables. However, the predictive power of CDCC is limited by uncertainties in the input optical potentials. Rigorous Uncertainty Quantification (UQ) and Bayesian parameter estimation require exploring high-dimensional parameter spaces (often $10^4$ – $10^6$ evaluations) to sample posterior probability distributions. Since a single CDCC calculation can take minutes to hours, brute-force UQ is computationally prohibitive. While emulator techniques exist for bound-state problems (e.g., eigenvector continuation) and simple scattering systems, extending these to realistic CDCC calculations involving tens of coupled channels and complex coupling structures remains an open challenge.

Methodology
The author develops a reduced basis emulator specifically designed for CDCC calculations, utilizing the Proper Orthogonal Decomposition (POD) method. The workflow is divided into two stages:

Offline Training Phase:
- Snapshot Generation: A set of $N_s$ representative optical potential parameter vectors is selected using Latin hypercube sampling. For each parameter set, the full CDCC problem is solved to obtain the solution vectors (snapshots) for the radial wave function coefficients across all coupled channels and mesh points.
- Basis Construction: The snapshot matrices are decomposed via Singular Value Decomposition (SVD). The method exploits the observation that CDCC solutions for smoothly varying optical potential parameters lie in a low-dimensional subspace. The optimal reduced basis is formed by retaining the left singular vectors (POD modes) corresponding to the largest singular values, determined by an energy criterion (retaining modes until the captured variance exceeds $1 - \epsilon_{tol}$ , typically with $\epsilon_{tol} \approx 10^{-6}$ ).
- Precomputation: Parameter-independent components of the system matrix (kinetic energy, centrifugal barrier, boundary conditions) and the source vector are projected onto the reduced basis and stored.
Online Prediction Phase:
- Galerkin Projection: For a new set of parameters $\theta^*$ , the emulator constructs the parameter-dependent coupling potential matrix.
- Reduced System Solution: The original large linear system is projected onto the reduced basis using Galerkin projection. This yields a small linear system of dimension $n_b \times n_b$ (where $n_b$ is the number of retained basis vectors, typically 5–50), which incorporates the physics of the new parameters.
- Reconstruction: The reduced coefficients are solved for and used to reconstruct the full solution vector, from which observables like differential cross sections are derived.

The implementation leverages the block structure of the potential matrix and optimized BLAS/LAPACK libraries (e.g., OpenBLAS, Intel MKL) to accelerate matrix operations.

Key Contributions

First Application to CDCC: This work presents the first reduced basis emulator specifically tailored for CDCC calculations, handling the complexity of continuum discretization and multi-channel coupling.
High-Dimensional Parameter Emulation: The emulator simultaneously emulates 18 optical potential parameters (9 for proton-target and 9 for neutron-target interactions), a dimensionality significantly higher than previous scattering emulators.
Per-Partial-Wave Basis Construction: The method constructs separate reduced bases for each partial wave $J$ , naturally accommodating the varying channel structures and solution complexities associated with different angular momenta.
Efficient Implementation: By directly computing the potential contribution at prediction time rather than relying on Empirical Interpolation Methods (EIM), the approach simplifies implementation while maintaining efficiency for the channel counts typical in CDCC ( $N_{ch} \approx 30–50$ ).

Results
The method was validated using deuteron elastic scattering on $^{58}$ Ni at 21.6 MeV as a test case.

Accuracy: The emulator reproduces elastic scattering cross sections with relative errors consistently below 0.1% across a wide range of optical potential parameters. Specifically, the total elastic cross section errors were found to be below 0.05% for all five distinct test cases, including those near the boundaries of the parameter space.
Fidelity: The emulator accurately reproduces the detailed radial dependence of the wave function coefficients, the oscillatory behavior of the S-matrix elements, and the interference patterns in the angular distributions (Fresnel diffraction).
Training Efficiency: A training set of 200 samples was sufficient to achieve sub-0.05% accuracy, indicating that the method scales favorably (approx. 10 samples per parameter dimension) compared to grid-based approaches. Increasing the training set to 400 samples did not yield significant systematic improvements, suggesting convergence.
Speedup: The emulator achieves a speedup of approximately 220 $\times$ compared to direct CDCC solvers. A single partial wave calculation that takes 6.5 seconds is reduced to 30 milliseconds. This reduces the time for a full evaluation (summing over all partial waves) from roughly 200 seconds to 1 second.

Significance and Claims
The paper claims that this development enables rigorous uncertainty quantification and Bayesian parameter estimation for nuclear reaction calculations that were previously computationally prohibitive. By reducing the computational cost of forward model evaluations by two orders of magnitude, the emulator makes it feasible to perform the $10^5$ – $10^6$ likelihood evaluations required for Bayesian inference within hours rather than months. The author notes that while the current application focuses on 18 parameters and moderate channel numbers, the favorable scaling of the reduced basis dimension (governed by parameter smoothness rather than channel count) suggests the method is well-suited for future applications to more complex systems, such as halo nuclei with larger model spaces. The work positions reduced basis emulation as a critical tool for advancing nuclear reaction theory into an era of precision physics with quantified uncertainties.

Technical Summary: Reduced Basis Emulator for Elastic Scattering in CDCC Calculations

More like this