Kirkpatrick et al . (Reports, 9 December 2021, p. 1385) trained a neural network–based DFT functional, DM21, on fractional-charge (FC) and fractional-spin (FS) systems, and they claim that it has outstanding accuracy for chemical systems exhibiting strong correlation. Here, we show that the ability of DM21 to generalize the behavior of such systems does not follow from the published results and requires revisiting.

中文翻译：

---

评论“通过解决分数电子问题推动密度泛函的前沿”

柯克帕特里克等人. （报告，2021 年 12 月 9 日，第 1385 页）在分数电荷 (FC) 和分数自旋 (FS) 系统上训练了基于神经网络的 DFT 泛函 DM21，他们声称它对于化学系统表现出出色的准确性强相关性。在这里，我们表明 DM21 概括此类系统行为的能力并未遵循已发布的结果，需要重新审视。