Various observations indicate that the Sun is depleted in refractory elements compared to the majority of Sun-like stars. Rayna Rampalli and colleagues extract abundances of such elements in thousands of Sun-like stars from spectra obtained by different surveys, in order to confirm the depletion and set constraints on its causes.

The authors use TheCannon, an algorithm that is trained on a highly precise reference set of stellar properties (‘labels’, in machine learning parlance) in order to extract the same labels from an incomplete or less precise big dataset. They build their reference set of 17 labels (effective temperature, surface gravity and the abundance of 15 metals relative to hydrogen) from 34 solar-analogue stars with precise elemental abundances measured by high resolution spectroscopic surveys (R = 40,000–110,000). TheCannon uses this set to analyse spectra of 64,557 Sun-like stars obtained by Gaia’s Radial Velocity Spectrometer (R = 11,200) and, after a battery of validation tests, it produces the abundances of the 15 refractive elements with a precision of less than 0.1 dex for 17,412 of them.

各种观测结果表明，与大多数类太阳恒星相比，太阳的难熔元素比较匮乏。 Rayna Rampalli 及其同事从不同调查获得的光谱中提取了数千颗类太阳恒星中此类元素的丰度，以确认损耗并对其原因进行限制。

作者使用 TheCannon，这是一种在高度精确的恒星属性参考集（机器学习术语中的“标签”）上进行训练的算法，以便从不完整或不太精确的大数据集中提取相同的标签。他们根据 34 颗类太阳恒星构建了 17 个标签的参考集（有效温度、表面重力和 15 种金属相对于氢的丰度），并通过高分辨率光谱测量测量了精确的元素丰度（R = 40,000–110,000）。 TheCannon 使用这套系统分析了盖亚径向速度光谱仪 ( R = 11,200) 获得的 64,557 颗类太阳恒星的光谱，经过一系列验证测试，它产生了 15 个折射元素的丰度，精度小于 0.1 dex其中 17,412 人。