In this paper, we present an ab initio theoretical calculation of the atomic parameters, including energy levels, wavelengths, lifetimes, and transition parameters, associated with the 12 and 12 ( 2–4, −1) configurations of Be-like sequence: Os LXXIII, Ir LXXIV, Pt LXXV, and Au LXXVI. Our analysis is based on the relativistic wavefunctions derived from the multi-configuration Dirac–Fock (MCDF) method, which are implemented in the relativistic atomic structure package GRASP2018. The correlations within the (principal quantum number) 10 complex are accounted for. The Breit interaction and quantum electrodynamical effects are added in the relativistic configuration interaction (RCI) calculation. Our results are compared with the existing data. The uncertainties of the dipole transition line strengths are assessed. In addition, we propose a feed forward neural network to predict the energy levels of Au LXXVI. This is a powerful machine learning tool capable of learning from existing data and predicting unknown data. The network is trained using theoretical energy levels of Os LXXIII, Ir LXXIV, and Pt LXXV obtained from the MCDF method. Our neural network exhibits average relative deviations of approximately 0.01% and 0.02% for trained and predicted energy levels, respectively, when compared to the MCDF results. This level of deviation suggests that our results are reasonably accurate. The data set presented in this study is useful for modeling fusion plasmas.

中文翻译：

---

Os LXXIII、Ir LXXIV、Pt LXXV 和 Au LXXVI 原子性质的准确从头计算和神经网络预测


在本文中，我们提出了与 Be 类序列的 12 和 12 ( 2–4, -1) 配置相关的原子参数的从头理论计算，包括能级、波长、寿命和跃迁参数：Os LXXIII、Ir LXXIV、Pt LXXV 和 Au LXXVI。我们的分析基于源自多配置狄拉克-福克（MCDF）方法的相对论波函数，该方法在相对论原子结构包 GRASP2018 中实现。考虑了（主量子数）10 复合体内的相关性。在相对论构型相互作用（RCI）计算中添加了布赖特相互作用和量子电动力学效应。我们的结果与现有数据进行了比较。评估偶极过渡线强度的不确定性。此外，我们提出了一个前馈神经网络来预测 Au LXXVI 的能级。这是一个强大的机器学习工具，能够从现有数据中学习并预测未知数据。该网络使用从 MCDF 方法获得的 Os LXXIII、Ir LXXIV 和 Pt LXXV 的理论能级进行训练。与 MCDF 结果相比，我们的神经网络在训练和预测能量水平上的平均相对偏差分别约为 0.01% 和 0.02%。这种程度的偏差表明我们的结果相当准确。本研究中提供的数据集对于聚变等离子体建模非常有用。