Abstract Ab initio many-body perturbation theory within the G W approximation is a Green’s function formalism widely used in the calculation of quasiparticle excitation energies of solids. In what has become an increasingly standard approach, Kohn–Sham eigenenergies, generated from a DFT calculation with a strategically-chosen exchange–correlation functional “starting point”, are used to construct G and W , and then perturbatively corrected by the resultant G W self-energy. In practice, there are several ways to construct the G W self-energy, and these can lead to variations in predicted quasiparticle energies. For example, for ZnO and TiO 2 , the G W fundamental gaps reported in the literature can vary by more than 1 eV depending on the G W code used. In this work, we calculate and analyze G W quasiparticle (QP) energies of these and other systems with three different G W codes: BerkeleyGW , Abinit and Yambo . Through a systematic analysis of the G W implementation of these three codes, we identify the primary origin of major discrepancies between codes reported in prior literature to be the different implementations the Coulomb divergence in the Fock exchange term and the frequency integration scheme of the G W self-energy. We then eliminate these discrepancies by using common numerical methods and algorithms, demonstrating that the same quasiparticle energies for a given material can be obtained with different codes, within numerical differences ascribable to the technical details of the underling implementations. This work will be important for users and developers in assessing the precision of future G W applications and methods.

中文翻译：

---

固体 G0W0 计算中的再现性

摘要 GW 近似中的从头算多体微扰理论是一种格林函数形式，广泛用于计算固体的准粒子激发能。在越来越标准的方法中，Kohn-Sham 特征能，由 DFT 计算生成，具有战略选择的交换相关函数“起点”，用于构建 G 和 W，然后由所得的 GW 自身进行微扰校正-活力。在实践中，有几种方法可以构建 GW 自能，这些方法会导致预测的准粒子能量发生变化。例如，对于 ZnO 和 TiO 2 ，​​根据所使用的 GW 代码，文献中报告的 GW 基本间隙的变化可能超过 1 eV。在这项工作中，我们使用三种不同的 GW 代码计算和分析这些系统和其他系统的 GW 准粒子 (QP) 能量：BerkeleyGW、Abinit 和 Yambo。通过对这三个代码的 GW 实现的系统分析，我们确定了先前文献中报告的代码之间主要差异的主要来源是 Fock 交换项中的库仑散度和 GW 的频率积分方案的不同实现。活力。然后，我们通过使用常见的数值方法和算法来消除这些差异，证明可以使用不同的代码获得给定材料的相同准粒子能量，数值差异可归因于底层实现的技术细节。