AI 中文总结
研究从稀疏实验测量中提取势垒分布及不确定性估计的问题，核心方法是基于AutoBNN的贝叶斯推理，贡献在于能更准确恢复势垒分布、校准不确定性，应用于重离子聚变反应并开发了用户友好软件。
AI 中文摘要
势垒分布编码了关于聚变核结构和动力学的丰富信息，但从实验聚变截面中提取它们需要估计聚变激发函数的二阶导数。在这项工作中，我们将从稀疏实验测量中提取带有不确定性估计的势垒分布的任务视为一个贝叶斯推理问题。我们引入了一种基于可解释的贝叶斯机器学习框架AutoBNN的方法，为分析聚变激发函数提供一种稳健的统计方法。通过在跨越广泛实际实验条件的模拟激发函数上与高斯过程回归进行基准测试，我们发现AutoBNN能更忠实地恢复潜在的势垒分布并报告校准良好的不确定性。然后我们将AutoBNN方法应用于四个实验测量的重离子聚变反应，它减轻了虚假的高于势垒结构并限制了现有预测。此外，我们还开发了该方法的用户友好软件实现，便于其应用于未来的重离子和轻离子聚变实验。
英文摘要
Barrier distributions encode rich information about the structure and dynamics of fusing nuclei, but extracting them from experimental fusion cross sections requires an estimate of the fusion excitation function's second derivative. In this work we approach the task of extracting barrier distributions with uncertainty estimates from sparse experimental measurements as a Bayesian inference problem. We introduce a method based on AutoBNN, an interpretable Bayesian machine learning framework, to provide a robust statistical approach for analyzing fusion excitation functions. Benchmarking against Gaussian process regression on simulated excitation functions that span a wide range of realistic experimental conditions, we find that AutoBNN more faithfully recovers the underlying barrier distribution and reports well-calibrated uncertainties. We then apply the AutoBNN method to four experimentally measured heavy-ion fusion reactions where it mitigates spurious above-barrier structure and constrains existing predictions. Alongside these results, we have developed a user-friendly software implementation of our method, facilitating its application to future heavy and light-ion fusion experiments.
Comments9 pages, 4 figures
