With the observations of an unprecedented number of oscillating subgiant stars expected from NASA's TESS mission, the asteroseismic characterization of subgiant stars will be a vital task for stellar population studies and for testing our theories of stellar evolution. To determine the fundamental properties of a large sample of subgiant stars efficiently, we developed a deep learning method that estimates distributions of fundamental parameters like age and mass over a wide range of input physics by learning from a grid of stellar models varied in eight physical parameters. We applied our method to four Kepler subgiant stars and compare our results with previously determined estimates. Our results show good agreement with previous estimates for three of them (KIC 11026764, KIC 10920273, KIC 11395018). With the ability to explore a vast range of stellar parameters, we determine that the remaining star, KIC 10005473, is likely to have an age 1 Gyr younger than its previously determined estimate. Our method also estimates the efficiency of overshooting, undershooting, and microscopic diffusion processes, from which we determined that the parameters governing such processes are generally poorly-constrained in subgiant models. We further demonstrate our method's utility for ensemble asteroseismology by characterizing a sample of 30 Kepler subgiant stars, where we find a majority of our age, mass, and radius estimates agree within uncertainties from more computationally expensive grid-based modelling techniques.

中文翻译：

---

使用深度学习对次巨星演化参数进行星震推断

随着 NASA 的 TESS 任务预计会观测到数量前所未有的振荡亚巨星，亚巨星的星震特征将成为恒星种群研究和测试我们的恒星演化理论的一项重要任务。为了有效地确定大量次巨星样本的基本属性，我们开发了一种深度学习方法，通过从八个物理参数不同的恒星模型网格中学习，估计年龄和质量等基本参数在广泛的输入物理范围内的分布. 我们将我们的方法应用于四颗开普勒次巨星，并将我们的结果与先前确定的估计值进行比较。我们的结果与之前对其中三个（KIC 11026764、KIC 10920273、KIC 11395018）的估计非常吻合。凭借探索大量恒星参数的能力，我们确定剩余的恒星 KIC 10005473 的年龄可能比其先前确定的估计值年轻 1 Gyr。我们的方法还估计了过冲、下冲和微观扩散过程的效率，从中我们确定控制这些过程的参数在次巨星模型中通常受到的约束很差。我们通过表征 30 颗开普勒次巨星的样本进一步证明了我们的方法对集合星震学的效用，我们发现我们的大部分年龄、质量和半径估计值在计算成本更高的基于网格的建模技术的不确定性内是一致的。可能比之前确定的估计年龄小 1 Gyr。我们的方法还估计了过冲、下冲和微观扩散过程的效率，从中我们确定控制这些过程的参数在次巨星模型中通常受到的约束很差。我们通过表征 30 颗开普勒次巨星的样本进一步证明了我们的方法对集合星震学的效用，我们发现我们的大部分年龄、质量和半径估计值在计算成本更高的基于网格的建模技术的不确定性内是一致的。可能比之前确定的估计年龄小 1 Gyr。我们的方法还估计了过冲、下冲和微观扩散过程的效率，从中我们确定控制这些过程的参数在次巨星模型中通常受到的约束很差。我们通过表征 30 颗开普勒次巨星的样本进一步证明了我们的方法对集合星震学的效用，我们发现我们的大部分年龄、质量和半径估计值在计算成本更高的基于网格的建模技术的不确定性内是一致的。