From Classical to Quantum Machine Learning: Different Approaches in Fission Barrier Height Estimation

TL;DR

This paper explores classical, hybrid, and quantum support vector regression methods to predict nuclear fission barrier heights, demonstrating that quantum approaches are promising despite current hardware limitations.

Contribution

It introduces and compares multiple quantum, hybrid, and classical SVR models for fission barrier estimation, highlighting the potential of quantum machine learning in nuclear physics.

Findings

Hybrid SVR achieves the best performance.

Quantum approaches are competitive with classical methods.

Quantum methods show promise for future improvements.

Abstract

The fission barrier energy is a fundamental property of nuclear structure that governs the stability of nuclei against fission, directly affecting their spontaneous fission half-lives and the formation of superheavy elements. However, because it can only be measured indirectly, it also enables the emergence of alternative, complementary, fast, and accurate prediction tools for traditional theoretical models. In this study, we examine the use of classical, hybrid, and quantum support vector regression (SVR) approaches to estimate fission barrier heights, starting from fundamental nuclear properties and their derived additional properties. For this purpose, eight different SVR-based approaches are considered: (i) Classical SVR, (ii) Enhanced Classical SVR with polynomial and trigonometric feature extensions, (iii) Quantum-Inspired SVR, (iv) Hybrid SVR, (v) Enhanced Hybrid SVR, (vi) Quantum Core SVR, (vii) Fixed-Parameter Quantum Feature Map SVR, and (viii) Pure Quantum SVR. The models were trained and tested on a dataset of 317 isotopes in the Z (atomic number) range of 98-126, encompassing the actinide and superheavy regions. The model performance was evaluated in terms of R2, RMSE, MAE, and training-test R2 gap. The results show that the hybrid model achieves the best overall performance. However, while quantum-enhanced approaches are still limited by circuit depth and optimization precision, they achieve competitive accuracy comparable to classical results. Considering future developments in quantum hardware and algorithms, quantum approaches are expected to achieve considerable improvements. The findings suggest that quantum machine learning can supplement classical approaches and offer a promising path toward more accurate and efficient nuclear property predictions.