这是用户在 2024-10-29 21:40 为 https://journals.aps.org/prb/abstract/10.1103/PhysRevB.76.165108 保存的双语快照页面,由 沉浸式翻译 提供双语支持。了解如何保存?

Electron-phonon interaction using Wannier functions
使用 Wannier 函数的电子-声子相互作用

EI检索SCI升级版 物理与天体物理2区SCI基础版 物理2区IF 3.2
Feliciano Giustino, Marvin L. Cohen, and Steven G. Louie
费里西安诺·朱斯蒂诺、马文·L·科恩和史蒂文·G·路易
Phys. Rev. B 76, 165108 – Published 4 October 2007
物理。修订版 B 76 , 165108 – 2007 年 10 月 4 日发布

Abstract  抽象的

We introduce a technique based on the spatial localization of electron and phonon Wannier functions to perform first-principles calculations of the electron-phonon interaction with an ultradense sampling of the Brillouin zone. After developing the basic theory, we describe the practical implementation within a density-functional framework. The proposed method is illustrated by considering a virtual crystal model of boron-doped diamond. For this test case, we first discuss the spatial localization of the electron-phonon matrix element in the Wannier representation. Then, we assess the accuracy of the Wannier-Fourier interpolation in momentum space. Finally, we study the convergence of the electron-phonon self-energies with the sampling of the Brillouin zone by calculating the electron and phonon linewidths, the Eliashberg spectral function, and the mass enhancement parameter of B-doped diamond. We show that more than 105 points in the irreducible wedge of the Brillouin zone are needed to achieve convergence.
我们引入了一种基于电子和声子 Wannier 函数空间定位的技术,通过布里渊区的超密集采样来执行电子-声子相互作用的第一原理计算。在发展了基本理论之后,我们描述了密度泛函框架内的实际实现。通过考虑掺硼金刚石的虚拟晶体模型来说明所提出的方法。对于这个测试用例,我们首先讨论 Wannier 表示中电子声子矩阵元素的空间定位。然后,我们评估动量空间中 Wannier-Fourier 插值的准确性。最后,我们通过计算电子和声子线宽、Eliashberg 谱函数和掺硼金刚石的质量增强参数,研究了布里渊区采样时电子声子自能的收敛性。我们表明,超过 105 需要布里渊区不可约楔中的点来实现收敛。

  • Figure
  • Figure
  • Figure
  • Figure
  • Figure
  • Figure
  • Figure
4 More 4 更多
  • Received 20 June 2007 2007 年 6 月 20 日收稿

DOI:https://doi.org/10.1103/PhysRevB.76.165108

©2007 American Physical Society
©2007 美国物理学会

Authors & Affiliations  作者及单位

Feliciano Giustino*, Marvin L. Cohen, and Steven G. Louie
费里西亚诺·朱斯蒂诺*马文·L·科恩史蒂文·G·路易

  • Department of Physics, University of California at Berkeley, Berkeley, California 94720, USA and Materials Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, California 94720, USA
    加州大学伯克利分校物理系,伯克利,加利福尼亚州 94720,美国;劳伦斯伯克利国家实验室材料科学部,伯克利,加利福尼亚州 94720,美国

  • *giustino@civet.berkeley.edu
    * giustino@civet.berkeley.edu

Article Text  文章正文

I. INTRODUCTION 一、简介

The electron-phonon (e-ph) interaction manifests itself in a wide range of phenomena, including the electrical resistivity, superconductivity, 1 the Kohn effect, 2 the Peierls instability, 3 and polaronic transport in organic materials. 4 In recent years, considerable attention has been drawn to the e-ph problem by photoemission experiments which revealed sharp signatures of this interaction in the quasiparticle spectra of high-temperature cuprate superconductors.5,6 In addition, recently, the effect of the e-ph interaction in nanoscale electron transport has attracted considerable interest since there is evidence that phonon-limited carrier lifetimes might hinder the ballistic operation of carbon-based field-effect devices. 7
电子-声子 (e-ph) 相互作用表现在多种现象中,包括电阻率、超导性、 1 Kohn 效应、 2 Peierls 不稳定性、 3和有机材料中的极化子输运。 4近年来,光电发射实验引起了人们对 e-ph 问题的极大关注,这些实验揭示了高温铜酸盐超导体准粒子光谱中这种相互作用的鲜明特征。 5, 6此外,最近,纳米级电子传输中的 e-ph 相互作用的影响引起了人们极大的兴趣,因为有证据表明声子有限的载流子寿命可能会阻碍碳基场效应器件的弹道操作。 7

The e-ph interaction has been the subject of theoretical investigations since the early attempts to explain the temperature dependence of the electrical resistivity of metals. 8 Following the pioneering investigations of Fröhlich, 9 Holstein, 10 Bardeen and Pines, 11 and later the BCS theory of superconductivity, 1 the e-ph interaction has become the prototypical example of a fermion-boson interaction and is now used as a standard benchmark for field-theoretical Green’s functions methods. 12–15 Despite the continued interest in the e-ph problem, the computational methods developed so far, ranging from frozen-phonon approaches 16–18 to first-principles linear-response techniques, 19–22 still remain unpractical. As a consequence, many important aspects, such as the effects of anisotropy within the Eliashberg theory,23,24 the validity of the Migdal theorem in the normal state, 25 and the range of validity of the Migdal-Eliashberg theory in the superconducting state, 26 remain only partially explored. In some cases, such as the Holstein polaron problem, 15 a first-principles approach has not even been attempted to the authors’ knowledge. The present situation is equally unsatisfactory from the point of view of applications, since current calculations are still limited to simple systems with a few atoms per unit cell, and only very few attempts have been made to address complex systems such as carbon nanostructures, 27 doped superconductors, 28 or metallic nanowires. 29
自从早期尝试解释金属电阻率的温度依赖性以来,e-ph 相互作用一直是理论研究的主题。 8继 Fröhlich、 9 Holstein、 10 Bardeen 和 Pines、 11以及后来的 BCS 超导理论的开创性研究之后, 1 e-ph 相互作用已成为费米子-玻色子相互作用的典型例子,现在被用作标准基准用于场论格林函数方法。 12–15尽管人们对 e-ph 问题持续感兴趣,但迄今为止计算方法的发展,包括冻结声子方法16-18线性响应技术的第一原理, 19-22仍然不切实际。因此,许多重要方面,例如 Eliashberg 理论中各向异性的影响, 23、24正常状态下 Migdal 定理的有效性, 25以及超导状态下 Migdal-Eliashberg 理论的有效性范围, 26 个仍处于部分探索状态。在某些情况下,例如荷斯坦极化子问题, 15据作者所知,甚至还没有尝试过第一性原理方法。 从应用的角度来看,目前的情况同样不能令人满意,因为当前的计算仍然仅限于每个晶胞只有几个原子的简单系统,并且只有很少的尝试来解决复杂的系统,例如碳纳米结构, 27掺杂超导体, 28或金属纳米线。29

This situation is partly due to the significant computational burden of an e-ph calculation, which often requires a very accurate description of electron and phonon scattering processes in the proximity of the Fermi surface.30,31
这种情况部分是由于 e-ph 计算的巨大计算负担,这通常需要对费米表面附近的电子和声子散射过程进行非常准确的描述。 30, 31

Motivated by these considerations, we have developed a technique which makes use of Wannier functions to dramatically reduce the computational cost of an e-ph calculation. The basic idea is to exploit the localization of both electronic and lattice Wannier functions in order to compute only a limited set of electronic and vibrational states and e-ph matrix elements from first principles, and then using these results to obtain the corresponding quantities for arbitrary electron and phonon wave vectors by a generalized Fourier interpolation. In this way, it becomes possible to sample accurately the Brillouin zone at the computational cost of a standard phonon dispersion calculation. 32 Besides the significant computational advantage, the Wannier representation proves to be an ideal analytical tool for studying the e-ph interaction in terms of simplified tight-binding models (for the electrons) and force-constant models (for the phonons) while preserving the accuracy of a full first-principles calculation.
出于这些考虑,我们开发了一种技术,利用 Wannier 函数来显着降低 e-ph 计算的计算成本。基本思想是利用电子和晶格 Wannier 函数的局域性,以便根据第一原理仅计算一组有限的电子和振动状态以及 e-ph 矩阵元素,然后使用这些结果获得任意值的相应量通过广义傅里叶插值计算电子和声子波矢量。这样,就可以以标准声子色散计算的计算成本对布里渊区进行精确采样。 32除了显着的计算优势外,Wannier 表示法被证明是一种理想的分析工具,可用于研究简化的紧束缚模型(对于电子)和力常数模型(对于声子)的 e-ph 相互作用,同时保留完整第一性原理计算的准确性。

In order to illustrate our method, we present an application to boron-doped diamond. Superconductivity above liquid He temperature has recently been observed in B-doped diamond, 33 and investigations are ongoing to explore the possibility of increasing 𝑇𝑐 beyond 10K by tuning sample preparation and doping treatments. 34 Previous theoretical works showed that superconductivity in diamond is crucially linked to the presence of the B atoms.28,35 Boron provides both the hole carriers participating in the supercurrent and the localized vibrations of the BC4 tetrahedra responsible for the pairing field. To keep the focus on the methodology, in the present work, we prefer to adopt a simplified point of view, and we describe B-doped diamond by a virtual crystal model.36,37
为了说明我们的方法,我们提出了掺硼金刚石的应​​用。最近在 B 掺杂金刚石中观察到高于液态 He 温度的超导性, 33并且正在进行研究以探索增加超导性的可能性 𝑇𝑐 超过 10K 通过调整样品制备和掺杂处理。 34先前的理论工作表明,金刚石的超导性与 B 原子的存在密切相关。 28, 35硼提供参与超流的空穴载流子和局部振动 BC4 四面体负责配对场。为了将注意力集中在方法上,在目前的工作中,我们倾向于采用简化的观点,并通过虚拟晶体模型来描述 B 掺杂金刚石。 36, 37

The present paper is organized as follows. In Sec. II, we review the current techniques to compute the electron and phonon self-energies arising from the e-ph interaction. In Sec. III, we introduce the electron and phonon Wannier functions and derive the e-ph matrix element in the Wannier representation. Section IV describes the generalized Wannier-Fourier interpolation of the e-ph matrix element and its practical implementation within a density-functional framework. In Sec. VI, we illustrate the theory by calculating the electron and phonon linewidths, the Eliashberg function, and the electron-phonon mass enhancement parameter of boron-doped diamond.
本文的结构如下。在秒。 II ,我们回顾了当前计算 e-ph 相互作用产生的电子和声子自能的技术。在秒。第三,我们引入了电子和声子Wannier函数并推导了Wannier表示中的e-ph矩阵元素。第四节描述了 e-ph 矩阵元素的广义 Wannier-Fourier 插值及其在密度泛函框架内的实际实现。在秒。第六,我们通过计算掺硼金刚石的电子和声子线宽、Eliashberg函数和电子-声子质量增强参数来说明该理论。

The present work extends and improves upon the method proposed in Ref. 35. In particular, in the present work, the electron and phonon coordinates are treated on the same footing, leading to a joint electron-phonon Wannier representation and a simultaneous electron-phonon Fourier interpolation. In the Appendix, we establish the connection with the procedure outlined in Ref. 35, and we discuss the relative merits of the two strategies.
目前的工作对参考文献中提出的方法进行了扩展和改进。 35 .特别是,在目前的工作中,电子和声子坐标被同等对待,从而产生联合电子-声子 Wannier 表示和同时电子-声子傅里叶插值。在附录中,我们建立了与参考文献中概述的程序的联系。 35 ,我们讨论了这两种策略的相对优点。

II. ELECTRON-PHONON INTERACTION
二.电子-声子相互作用

The formalism for addressing the e-ph interaction has been set by the seminal contributions of Fröhlich, 9 Bardeen and Pines, 11 and Engelsberg and Schrieffer. 38 The e-ph Hamiltonian derived in these studies is conveniently dealt with by standard Green’s functions techniques.1,14 The interacting electron and phonon propagators are, in principle, fully determined through Dyson’s equation once the corresponding electron (𝛴) and phonon (𝛱) self-energy operators associated with the mutual interactions are known. Instead of reviewing the possible approximations to the self-energy operators, we focus here on the simplest one, which consists in replacing the dressed e-ph vertex 𝛤 by its bare counterpart 𝑔 (Fig. 1). 1 This approximation is connected with the Born-Oppenheimer adiabatic theorem and is generally referred to as the Migdal approximation.25,26 In this work, we replace the dressed electron Green’s function by the corresponding free propagator, thereby avoiding complications associated with self-consistency. On the other hand, the fully renormalized phonon frequencies, as obtained from density-functional calculations, will be adopted in the phonon propagator. This is generally considered to be a sensible approximation. 14
解决 e-ph 相互作用的形式主义是由 Fröhlich、 9 Bardeen 和 Pines、 11以及 Engelsberg 和 Schrieffer 的开创性贡献确定的。 38这些研究中导出的 e-ph 哈密顿量可以通过标准格林函数技术方便地处理。 1, 14一旦相应的电子和声子传播体相互作用,原则上就可以通过戴森方程完全确定 (𝛴) 和声子 (𝛱) 与相互作用相关的自能算子是已知的。我们不回顾自能算子的可能近似,而是关注最简单的一种,其中包括替换修饰的 e-ph 顶点 𝛤 通过其裸露的对应物 𝑔 (图1 )。 1该近似与玻恩-奥本海默绝热定理相关,通常称为 Migdal 近似。 25, 26在这项工作中,我们用相应的自由传播子代替了修饰电子格林函数,从而避免了与自洽相关的复杂性。另一方面,声子传播器将采用从密度泛函计算获得的完全重整化的声子频率。这通常被认为是一个合理的近似值。 14

FIG. 1 如图。 1

(Color online) First-order e-ph diagrams considered in this work (red). Left: the self-energy of a phonon with momentum 𝐪 (black wiggly line) is given by the Fermion loop containing two electron lines (red lines) connected by the bare e-ph vertices (red disks) (Refs. 1,12). Right: the self-energy of an electron with momentum 𝐤 (black straight line) is given by the loop with one electron line (straight) and one phonon line (wiggly), connected by the bare e-ph vertices. Equations (1) and (2) were obtained from these diagrams.
(在线彩色)本工作中考虑的一阶 e-ph 图(红色)。左:具有动量的声子的自能 𝐪 (黑色摆动线)由包含两条电子线(红线)的费米子环给出,这两条电子线由裸露的 e-ph 顶点(红色圆盘)连接(参考文献 1、12 。右:具有动量的电子的自能 𝐤 (黑色直线)由具有一根电子线(直线)和一根声子线(摆动)的环给出,由裸露的 e-ph 顶点连接。从这些图中获得方程(1)(2)

The electron and phonon self-energies arising from the e-ph interaction (Fig. 1) read 12
e-ph 相互作用产生的电子和声子自能(图1 )为12

𝛴=𝑖𝑑2(2𝜋)4𝑔(1,2)2𝐷(12)𝐺0(2),
(1)
𝛱=2𝑖𝑑1(2𝜋)4𝑔(1,2)2𝐺0(1)𝐺0(2),
(2)

where 𝐺0 and 𝐷 are the bare electron and the dressed phonon Green’s functions, respectively, 1=(𝐤,𝜔) the quadrimomentum in compressed notation ( 𝐤 being the wave vector and 𝜔 the energy), and 𝑔(1,2) the electron-phonon matrix element. 39 Equations (1) and (2) were originally derived for the electron gas and need to be rewritten within the reduced-zone scheme for a practical calculation.
在哪里 𝐺0𝐷 分别是裸电子和修饰声子格林函数, 1=(𝐤,𝜔) 压缩符号中的四动量 ( 𝐤 是波矢量并且 𝜔 能量),以及 𝑔(1,2) 电子声子矩阵元素。 39方程(1)(2)最初是针对电子气推导出来的,需要在缩减区域方案中重写以进行实际计算。

We make the following approximations: (i) we neglect the changes in the electronic wave functions and phonon eigendisplacements arising from the e-ph interaction, 40 (ii) we take the expectation value of the self-energy operators on the noninteracting electron and phonon states, and (iii) we restrict our discussion to the imaginary parts of the electron (𝛴) and phonon (𝛱) self-energies, i.e., we only consider the corresponding linewidths. These simplifications are intended to illustrate our methodology by focusing on a few specific cases. The inclusion of the off-diagonal corrections and the calculation of the corresponding real self-energies are both feasible and will be the subject of a future communication. 41 By expressing the free propagators in terms of the noninteracting electronic energy 𝜖𝑛𝐤 (with 𝑛 the band index and 𝐤 the momentum) and vibrational frequency 𝜔𝐪𝜈 (with 𝜈 the branch index and 𝐪 the momentum), Eqs. (1) and (2) can be integrated analytically to yield 12
我们进行以下近似:(i)我们忽略由 e-ph 相互作用引起的电子波函数和声子本征位移的变化, 40 (ii)我们取非相互作用电子和声子的自能算子的期望值状态,并且(iii)我们将讨论限制在电子的虚部 (𝛴) 和声子 (𝛱) 自能,即我们只考虑相应的线宽。这些简化旨在通过关注一些具体案例来说明我们的方法。包含非对角线修正和相应的真实自能的计算都是可行的,并将成为未来交流的主题。 41通过用非相互作用电子能量来表达自由传播子 𝜖𝑛𝐤 (和 𝑛 能带指数和 𝐤 动量)和振动频率 𝜔𝐪𝜈 (和 𝜈 分支索引和 𝐪 动量),方程。 (1)(2)可以通过分析积分得到12

𝛴𝑛𝐤=𝜋𝑚𝜈BZ𝑑𝐪𝛺BZ𝑔SE𝑚𝑛,𝜈(𝐤,𝐪)2[(𝑛𝐪𝜈+𝑓𝑚𝐤+𝐪)𝛿(𝜖𝑛𝐤𝜔𝐪𝜈𝜖𝑚𝐤+𝐪)+(𝑛𝐪𝜈+1𝑓𝑚𝐤+𝐪)𝛿(𝜖𝑛𝐤+𝜔𝐪𝜈𝜖𝑚𝐤+𝐪)],
(3)
𝛱𝐪𝜈=2𝜋𝑚𝑛BZ𝑑𝐤𝛺BZ𝑔SE𝑚𝑛,𝜈(𝐤,𝐪)2(𝑓𝑛𝐤𝑓𝑚𝐤+𝐪)𝛿(𝜔𝐪𝜈+𝜖𝑛𝐤𝜖𝑚𝐤+𝐪),
(4)

where 𝑓𝑛𝐤 and 𝑛𝐪𝜈 are the Fermi-Dirac and Bose-Einstein occupations, respectively, the factor of 2 accounts for the spin degeneracy, and the integrations extend over the Brillouin zone. The e-ph matrix element (vertex) 𝑔SE𝑚𝑛,𝜈(𝐤,𝐪) appearing in Eqs. (3) and (4) is given by
在哪里 𝑓𝑛𝐤𝑛𝐪𝜈 分别是费米-狄拉克和玻色-爱因斯坦占据,因子 2 解释了自旋简并性,并且积分延伸到布里渊区。 e-ph 矩阵元素(顶点) 𝑔SE𝑚𝑛,𝜈(𝐤,𝐪) 出现在方程式中。 (3)(4)由下式给出

𝑔SE𝑚𝑛,𝜈(𝐤,𝐪)=(2𝑚0𝜔𝐪𝜈)1/2𝑔𝑚𝑛,𝜈(𝐤,𝐪),
(5)

where 𝑚0 is a convenient reference mass, and
在哪里 𝑚0 是一个方便的参考质量,并且

𝑔𝑚𝑛,𝜈(𝐤,𝐪)=𝑚𝐤+𝐪𝜕𝐪𝜈𝑉𝑛𝐤.
(6)

We adopt the superscript “SE” to distinguish between the matrix element appearing in the self-energy expressions ( 𝑔SE, with the dimensions of an energy) and the one used in the Bloch to Wannier transformation ( 𝑔, with dimensions of an energy divided by a length, cf. Sec. III). In Eq. (6), 𝑛𝐤 and 𝑚𝐤+𝐪 indicate Bloch eigenstates, while the operator 𝜕𝐪𝜈𝑉 is the derivative of the self-consistent potential with respect to a collective ionic displacement corresponding to a phonon with branch index 𝜈 and momentum 𝐪 (cf. Sec. III). 42
我们采用上标“SE”来区分自能表达式中出现的矩阵元素( 𝑔SE ,具有能量的维度)以及布洛赫到万尼尔变换中使用的能量( 𝑔 ,能量的维度除以长度,参见。秒。)。在等式中。 (6) , 𝑛𝐤𝑚𝐤+𝐪 指示布洛赫本征态,而算子 𝜕𝐪𝜈𝑉 是自洽势相对于对应于具有分支指数的声子的集体离子位移的导数 𝜈 和势头 𝐪 (参见第III节)。 42

Within the isotropic approximation to Eliashberg theory, 26 the phonon linewidths [Eq. (4)] constitute a key ingredient for calculating the Eliashberg function 𝛼2𝐹 and the associated mass enhancement parameter 𝜆 (Ref. 43):
在 Eliashberg 理论的各向同性近似中, 26声子线宽 [Eq. (4) ] 构成计算 Eliashberg 函数的关键要素 𝛼2𝐹 以及相关的质量增强参数 𝜆 (参考文献43 ):

𝛼2𝐹(𝜔)=1𝜋𝑁𝐹𝜈BZ𝑑𝐪𝛺BZ𝛱𝐪𝜈𝜔𝐪𝜈𝛿(𝜔𝜔𝐪𝜈),
(7)
𝜆=20𝑑𝜔𝜔1𝛼2𝐹(𝜔),
(8)

𝑁𝐹 being the density of electronic states at the Fermi level. The mass enhancement parameter 𝜆 is also referred to as the electron-phonon coupling strength and is widely used to estimate the transition temperature of conventional superconductors by means of the semiempirical McMillan formula. 44
𝑁𝐹 是费米能级的电子态密度。质量增强参数 𝜆 也称为电子-声子耦合强度,广泛用于通过半经验麦克米伦公式来估计传统超导体的转变温度。 44

Inspection of Eq. (4) reveals that the calculation of the phonon linewidths requires a summation over electronic transitions where both the initial state 𝑛𝐤 and the final state 𝑚𝐤+𝐪 are pinned close to the Fermi level by the selection rule 𝜖𝑚𝐤+𝐪=𝜖𝑛𝐤+𝜔𝐪𝜈. As a consequence, the calculation of vibrational linewidths requires a determination of the initial and final electronic states, the phonon states of interest, and the associated e-ph matrix elements with fine energy and momentum resolutions. Whereas total-energy calculations for metals usually require at most a few tens of 𝐤 points in the irreducible wedge of the Brillouin zone, in the present case, a much denser sampling is required to achieve numerical convergence, even up to several millions of 𝐤 points.30,45,46
检查方程。 (4)表明声子线宽的计算需要对电子跃迁求和,其中初始态 𝑛𝐤 和最终状态 𝑚𝐤+𝐪 通过选择规则固定在费米能级附近 𝜖𝑚𝐤+𝐪=𝜖𝑛𝐤+𝜔𝐪𝜈 。因此,振动线宽的计算需要确定初始和最终电子态、感兴趣的声子态以及具有精细能量和动量分辨率的相关 e-ph 矩阵元素。而金属的总能量计算通常最多需要几十 𝐤 布里渊区不可约楔中的点,在目前的情况下,需要更密集的采样才能实现数值收敛,甚至达到数百万个 𝐤 点。 30、45、46

Similar considerations apply to the calculations of the electron self-energy [Eq. (3)], to the Eliashberg function [Eq. (7)], and to the mass enhancement parameter [Eq. (8)]. In particular, the difficulty in the determination of the mass enhancement parameter translates into a large uncertainty in the calculated superconducting transition temperature through the McMillan equation. 35 Even within more sophisticated approaches where superconducting properties are determined directly from first principles, 47 the practical implementations suffer from a strong sensitivity to the sampling of e-ph scattering processes near the Fermi surface. 31 In the following sections, we describe how it is possible to circumvent such difficulties by reformulating the e-ph vertex in the Wannier representation.
类似的考虑也适用于电子自能的计算[方程1]。 (3) ], 到 Eliashberg 函数 [Eq. (7) ], 以及质量增强参数 [Eq. (8) ]。特别是,质量增强参数确定的困难转化为通过麦克米伦方程计算的超导转变温度存在很大的不确定性。 35即使在超导特性直接根据第一原理确定的更复杂的方法中, 47实际实现也会受到对费米表面附近 e-ph 散射过程采样的强烈敏感性的影响。 31在以下各节中,我们将描述如何通过在 Wannier 表示中重新表述 e-ph 顶点来规避此类困难。

III. ELECTRON-PHONON VERTEX IN THE WANNIER REPRESENTATION
三. Wannier 表示中的电子-声子顶点

In this section, we introduce the Wannier representation of the e-ph vertex. We first describe the electronic Wannier functions and the phonon perturbation potential in the Wannier representation. Then, we derive the relation between the e-ph matrix elements in the Wannier representation and those in the Bloch representation.
在本节中,我们介绍 e-ph 顶点的 Wannier 表示。我们首先描述电子 Wannier 函数和 Wannier 表示中的声子微扰势。然后,我们推导了Wannier表示中的e-ph矩阵元素与Bloch表示中的e-ph矩阵元素之间的关系。

A. Electronic Wannier functions
A. 电子万尼尔功能

Wannier functions were first introduced to study the excitonic levels of polar insulators using a localized representation. 48 In the most general case, the relation between the Bloch functions 𝜓𝑛𝐤(𝐫)=𝐫𝑛𝐤 and the Wannier functions 𝑤𝑚(𝐫𝐑𝑒)=𝐫𝑚𝐑𝑒 is given by a Fourier transform in the momentum (𝐤) and lattice vector (𝐑𝑒) variables, generalized to include band mixing: 49
Wannier 函数首次被引入以使用局域表示来研究极性绝缘体的激子能级。 48在最一般的情况下,布洛赫函数之间的关系 𝜓𝑛𝐤(𝐫)=𝐫𝑛𝐤 和 Wannier 函数 𝑤𝑚(𝐫𝐑𝑒)=𝐫𝑚𝐑𝑒 由动量的傅里叶变换给出 (𝐤) 和格向量 (𝐑𝑒) 变量,概括为包括频带混合: 49

𝑚𝐑𝑒=𝑛𝐤𝑒𝑖𝐤𝐑𝑒𝑈𝑛𝑚,𝐤𝑛𝐤.
(9)

Whenever the mixing matrix 𝑈𝑛𝑚,𝐤 is unitary, the Wannier states turn out to be orthonormal: 𝑚𝐑𝑒𝑚𝐑𝑒=𝛿(𝐑𝑒,𝐑𝑒)𝛿(𝑚,𝑚). The inverse relation of Eq. (9) is obtained by a standard inverse Fourier transform:
每当混合矩阵 𝑈𝑛𝑚,𝐤 是酉的,Wannier 状态结果是正交的: 𝑚𝐑𝑒𝑚𝐑𝑒=𝛿(𝐑𝑒,𝐑𝑒)𝛿(𝑚,𝑚) 。方程式的反比关系。 (9)通过标准傅里叶逆变换得到:

𝑛𝐤=1𝑁𝑒𝑚𝐑𝑒𝑒𝑖𝐤𝐑𝑒𝑈𝑚𝑛,𝐤𝑚𝐑𝑒.
(10)

We consider here a periodic lattice which is a supercell of the primitive cell of the crystal. Accordingly, we use discrete summations in Eqs. (9) and (10) instead of integrals over continuous variables. The expressions we derive are therefore ready to be implemented in existing computational schemes. In going from Eq. (9) to Eq. (10), we used the relation 𝐤exp[𝑖(𝐤𝐤)𝐑𝑒]=𝑁𝑒𝛿(𝐤,𝐤), where 𝑁𝑒 is the number of unit cells in the supercell, corresponding to the number of 𝐤 points included in the calculation.
我们在这里考虑周期性晶格,它是晶体原始晶胞的超晶胞。因此,我们在方程中使用离散求和。 (9)(10)代替连续变量的积分。因此,我们导出的表达式可以在现有的计算方案中实现。从方程式开始。 (9)至等式。 (10) ,我们使用关系式 𝐤exp[𝑖(𝐤𝐤)𝐑𝑒]=𝑁𝑒𝛿(𝐤,𝐤) , 在哪里 𝑁𝑒 是超晶胞中晶胞的数量,对应于 𝐤 计算中包含的点。

The usefulness of the Wannier representation relies on the spatial localization of the electronic states. Equation (9) indicates a large freedom associated with the transformation from Bloch to Wannier functions, since one has to choose both the manifold of the initial Bloch states and the unitary rotation associated with such a manifold. When the system under consideration presents a composite set of bands isolated from other bands by finite energy gaps, the choice of the Bloch manifold is natural and it remains to choose the unitary transform 𝑈𝑛𝑚,𝐤. The most convenient choice for the purposes of the present work is the one leading to maximally localized Wannier functions. 49 In this case, the unitary transform is determined by requiring that the resulting Wannier functions minimize the Berry-phase spatial spread operator defined within the framework of the modern theory of polarization. 49–51 Wannier functions determined according to this procedure exhibit exponential localization. 52
万尼尔表示的有用性依赖于电子态的空间定位。方程(9)表明与从布洛赫函数到万尼尔函数的变换相关的很大的自由度,因为人们必须选择初始布洛赫状态的流形以及与这样的流形相关的酉旋转。当所考虑的系统呈现一组通过有限能隙与其他能带隔离的复合能带时,布洛赫流形的选择是自然的,并且仍然选择酉变换 𝑈𝑛𝑚,𝐤 。就目前的工作而言,最方便的选择是导致最大局部化 Wannier 函数的选择。 49在这种情况下,酉变换是通过要求所得的 Wannier 函数最小化现代偏振理论框架内定义的 Berry 相空间扩展算子来确定的。根据此过程确定的49–51 Wannier 函数表现出指数局部化。 52

In the case of metals, the relevant bands do not usually constitute a composite manifold, and the previous procedure cannot be applied directly. Nonetheless, a disentanglement strategy, which allows the extraction of an optimally connected subspace from an initial entangled manifold, has already been introduced and demonstrated for simple metals. 53 This procedure consists at projecting the electronic Hamiltonian onto an appropriate subspace to treat a metallic system in effectively the same way as a hole-doped insulator. This technique is currently in use for transport problems. 54
对于金属,相关能带通常不构成复合流形,并且不能直接应用先前的过程。尽管如此,已经针对简单金属引入并演示了一种解缠结策略,该策略允许从初始纠缠流形中提取最佳连接的子空间。 53该过程包括将电子哈密顿量投影到适当的子空间上,以与空穴掺杂绝缘体相同的方式有效地处理金属系统。该技术目前用于解决运输问题。 54

B. Phonon perturbation in the Wannier representation
B. Wannier 表示中的声子扰动

The potential 𝑉(𝐫) appearing in the e-ph matrix element [Eq. (6)] includes both the ionic contribution and the electronic self-consistent field. Within linear-response theory, this potential can be formally obtained by screening the bare ionic potentials 𝑉ion𝜅(𝐫) with the electronic dielectric function taken in the static limit 𝜖(𝐫,𝐫) (Ref. 42):
潜力 𝑉(𝐫) 出现在 e-ph 矩阵元素 [Eq. (6) ]既包括离子贡献又包括电子自洽场。在线性响应理论中,该电势可以通过筛选裸离子电势来正式获得 𝑉ion𝜅(𝐫) 电子介电函数在静态极限下取 𝜖(𝐫,𝐫) (参考文献42 ):

𝑉(𝐫;{𝛕𝜅𝑝})=𝑑𝐫𝜖1(𝐫,𝐫)𝜅,𝐑𝑝𝑉ion𝜅(𝐫𝛕𝜅𝑝).
(11)

In Eq. (11), the sum extends over all the unit cells of the crystal centered at the lattice vectors 𝐑𝑝 and over all the atoms 𝜅 located at the sites 𝛕𝜅 within each unit cell. The absolute coordinate of each ion is 𝛕𝜅𝑝𝐑𝑝+𝛕𝜅. {𝛕𝜅𝑝} indicates all the ionic coordinates in the crystal. In a pseudopotential calculation, the core electrons are assumed to follow rigidly the corresponding ions, and the potentials 𝑉ion𝜅(𝐫) in Eq. (11) need to be replaced by the ionic pseudopotentials. The extension of the formalism to nonlocal pseudopotentials does not pose any problem provided that the nonlocality is short ranged. The variation 𝜕𝐪𝜈𝑉(𝐫) of the self-consistent potential with respect to a collective ionic displacement 𝛥𝛕𝐪𝜈𝜅𝑝 corresponding to a phonon with momentum 𝐪 and branch index 𝜈 is obtained from
在等式中。 (11) ,总和延伸到以晶格向量为中心的晶体的所有晶胞 𝐑𝑝 以及所有原子之上 𝜅 位于站点 𝛕𝜅 在每个晶胞内。每个离子的绝对坐标为 𝛕𝜅𝑝𝐑𝑝+𝛕𝜅{𝛕𝜅𝑝} 表示晶体中所有的离子坐标。在赝势计算中,假设核心电子严格跟随相应的离子,并且势 𝑉ion𝜅(𝐫) 在等式中(11)需要用离子赝势代替。只要非局域性是短程的,将形式主义扩展到非局域赝势就不会造成任何问题。变化 𝜕𝐪𝜈𝑉(𝐫) 关于集体离子位移的自洽势 𝛥𝛕𝐪𝜈𝜅𝑝 对应于具有动量的声子 𝐪 和分支索引 𝜈 是从获得

𝜕𝐪𝜈𝑉(𝐫)=𝜕𝜕𝜂𝑉(𝐫;{𝛕𝜅𝑝+𝜂𝛥𝛕𝐪𝜈𝜅𝑝}).
(12)

Denoting the vibrational eigenmodes by 𝛕𝜅𝑝𝐪𝜈=𝑒𝑖𝐪𝐑𝑝𝐞𝜈𝐪𝜅, with 𝐞𝜈𝐪𝜅 cell periodic and normalized, and the ionic masses by 𝑚𝜅, we can express the displacements in Eq. (12) as follows: 42
振动本征模态表示为 𝛕𝜅𝑝𝐪𝜈=𝑒𝑖𝐪𝐑𝑝𝐞𝜈𝐪𝜅 , 和 𝐞𝜈𝐪𝜅 细胞周期性和标准化,以及离子质量 𝑚𝜅 ,我们可以用方程来表示位移。 (12)如下: 42

𝛥𝛕𝐪𝜈𝜅𝑝=Re[(𝑚0𝑚𝜅)1/2𝐞𝜈𝐪𝜅𝑒𝑖𝐪𝐑𝑝].
(13)

Before transforming the phonon perturbation 𝜕𝐪𝜈𝑉(𝐫) [Eq. (12)] in the Wannier representation, it is instructive to rewrite the vibrational eigenmodes in a form similar to Eq. (10):
变换声子扰动之前 𝜕𝐪𝜈𝑉(𝐫) [等式。 (12) ] 在 Wannier 表示中,以类似于方程 (12) 的形式重写振动本征模态是有启发性的。 (10)

𝛕𝜅𝑝𝐪𝜈=𝜅𝐑𝑝𝑒𝑖𝐪𝐑𝑝𝐞𝜈𝐪𝜅𝛿(𝛕𝜅𝑝𝛕𝜅𝑝).
(14)

By comparing Eqs. (14) and (10), we realize that (i) the maximally localized Wannier functions for vibrational modes (lattice Wannier functions) correspond to the displacement of individual ions 𝛿(𝛕𝜅𝑝𝛕𝜅𝑝) and (ii) the cell-periodic part of the vibrational eigenmode 𝐞𝜈𝐪𝜅 for the phonons plays a role analogous to that of the unitary transformation 𝑈𝑚𝑛,𝐤 for the electrons. This aspect was first pointed out in a study of lattice Wannier functions for the linear harmonic chain 55 and subsequently verified by explicitly constructing maximally localized phonon Wannier functions in a three-dimensional system. 56 The fact that extreme localization is achievable in the case of lattice vibrations relates to the discrete number of degrees of freedom associated with the classical ions.
通过比较等式。 (14)(10) ,我们认识到 (i) 振动模式的最大局域 Wannier 函数(晶格 Wannier 函数)对应于单个离子的位移 𝛿(𝛕𝜅𝑝𝛕𝜅𝑝) (ii) 振动本征模式的细胞周期部分 𝐞𝜈𝐪𝜅 因为声子起着类似于酉变换的作用 𝑈𝑚𝑛,𝐤 对于电子。这个方面首先在线性谐波链55的晶格 Wannier 函数的研究中被指出,随后通过在三维系统中显式构造最大局域声子 Wannier 函数来验证。 56在晶格振动的情况下可以实现极端局域化这一事实与与经典离子相关的离散自由度数有关。

By combining Eqs. (11)–(13), we can express the variation of the self-consistent potential 𝜕𝐪𝜈𝑉(𝐫) in terms of the contributions arising from each individual ion:
通过结合等式。 (11)–(13) ,我们可以表达自洽势的变化 𝜕𝐪𝜈𝑉(𝐫) 就每个离子产生的贡献而言:

𝜕𝐪𝜈𝑉(𝐫)=Re[𝜅,𝐑𝑝𝑒𝑖𝐪𝐑𝑝(𝑚0𝑚𝜅)1/2𝐞𝜈𝐪𝜅𝜕𝜅,𝐑𝑝𝑉(𝐫)],
(15)

where the real-valued vector field
其中实值向量场

𝜕𝜅,𝐑𝑝𝑉(𝐫)=𝛕𝜅𝑝𝑉(𝐫;{𝛕𝜅𝑝})
(16)

represents the gradient of the self-consistent potential with respect to the displacement of the ion 𝜅 in the unit cell 𝐑𝑝. Following the preceding discussion about Eq. (14), it is natural to call 𝜕𝜅,𝐑𝑝𝑉(𝐫) the phonon perturbation in the Wannier representation. For practical purposes, it is convenient to introduce the displacement field 𝐮𝜈𝐪𝜅=(𝑚0/𝑚𝜅)1/2𝐞𝜈𝐪𝜅 and to redefine 𝜕𝐪𝜈𝑉(𝐫) as the complex scalar field:
表示自洽势相对于离子位移的梯度 𝜅 在晶胞中 𝐑𝑝 。继前面关于方程的讨论之后。 (14) ,很自然地调用 𝜕𝜅,𝐑𝑝𝑉(𝐫) Wannier 表示中的声子扰动。出于实用目的,引入位移场是很方便的 𝐮𝜈𝐪𝜅=(𝑚0/𝑚𝜅)1/2𝐞𝜈𝐪𝜅 并重新定义 𝜕𝐪𝜈𝑉(𝐫) 作为复标量场:

𝜕𝐪𝜈𝑉(𝐫)=𝜅,𝐑𝑝𝑒𝑖𝐪𝐑𝑝𝐮𝜈𝐪𝜅𝜕𝜅,𝐑𝑝𝑉(𝐫),
(17)

keeping in mind that the real-valued field in the right-hand side of Eq. (12) is obtained through 12[𝜕𝐪𝜈𝑉(𝐫)+𝜕𝐪𝜈𝑉(𝐫)]. 57 By inverting Eq. (17), we obtain
请记住,方程右侧的实值字段。 (12)得自 12[𝜕𝐪𝜈𝑉(𝐫)+𝜕𝐪𝜈𝑉(𝐫)]57通过反转方程。 (17) ,我们得到

𝜕𝜅,𝐑𝑝𝑉(𝐫)=1𝑁𝑝𝐪𝜈𝑒𝑖𝐪𝐑𝑝[𝐮𝜈𝐪𝜅]1𝜕𝐪𝜈𝑉(𝐫),
(18)

with 𝑁𝑝 being the number of unit cells in the periodic supercell considered for the lattice dynamics. In principle, 𝑁𝑝 can differ from the corresponding number of unit cells 𝑁𝑒 for the electrons in Eq. (10).
𝑁𝑝 是考虑晶格动力学的周期性超晶胞中晶胞的数量。原则, 𝑁𝑝 可以与相应的晶胞数量不同 𝑁𝑒 对于方程中的电子。 (10) .

For our purposes, it is crucial that the phonon perturbation [Eq. (18)] be localized in real space. From a qualitative point of view, 𝜕𝜅,𝐑𝑝𝑉(𝐫) represents the potential associated with a screened dipole. In a metallic system, and within a simple Thomas-Fermi approximation, a potential of this kind would decay with the distance 𝑟 as (𝐫𝛥𝛕𝜅𝑝)𝑟2exp(𝑟/𝜆TF), with 𝜆TF being the Thomas-Fermi screening length. In an insulating system, the incomplete screening makes the screened dipole decay at large distances as 𝑍𝐵𝑟2, with 𝑍𝐵 being the Born dynamical charge associated with the displaced ion. Whenever the dynamical charges are nonvanishing (as is the case in polar insulators), the long-range component of the perturbation needs to be treated separately.
就我们的目的而言,声子扰动[方程1]至关重要。 (18) ] 被定位在真实空间中。从定性的角度来看, 𝜕𝜅,𝐑𝑝𝑉(𝐫) 代表与屏蔽偶极子相关的电势。在金属系统中,在简单的托马斯-费米近似中,这种电势会随着距离的增加而衰减 𝑟 作为 (𝐫𝛥𝛕𝜅𝑝)𝑟2exp(𝑟/𝜆TF) , 和 𝜆TF 是托马斯-费米筛选长度。在绝缘系统中,不完全的屏蔽使得屏蔽偶极子在长距离处衰减,如下所示 𝑍𝐵𝑟2 , 和 𝑍𝐵 是与位移离子相关的玻恩动态电荷。当动态电荷不为零时(如极性绝缘体的情况),扰动的长程分量需要单独处理。

A quantitative assessment of the spatial localization of 𝜕𝜅,𝐑𝑝𝑉(𝐫) can be formulated by making connection with the spatial decay of the interatomic force constants 𝐶(𝛕𝜅𝑝,𝛕𝜅𝑝)=𝛕𝜅𝑝𝐅(𝛕𝜅𝑝), with 𝐅(𝛕𝜅𝑝) the Hellmann-Feynman force acting on the ion 𝜅 in the unit cell 𝐑𝑝. To this end, we temporarily ignore the exchange-correlation contribution 𝑣xc to the self-consistent potential 𝑉(𝐫) and consider the electrostatic component including the Hartree and the ionic term: 𝑉es(𝐫)=𝑉Ha(𝐫)+𝑉ion(𝐫)=𝑉(𝐫)𝑣xc(𝐫). By evaluating 𝜕𝜅,𝐑𝑝𝑉es(𝐫) at 𝐫=𝛕𝜅𝑝, we obtain the change of the potential experienced by the atom at 𝛕𝜅𝑝 due to a displacement of the atom at 𝛕𝜅𝑝. The gradient of this quantity with respect to 𝛕𝜅𝑝 is by definition the force acting on the ion located at 𝛕𝜅𝑝 [cf. Eqs. (8) and (13) of Ref. 32]. As a consequence, the following relation holds between the phonon perturbation in the Wannier representation and the matrix of the interatomic force constants:
空间定位的定量评估 𝜕𝜅,𝐑𝑝𝑉(𝐫) 可以通过与原子间力常数的空间衰减联系起来来表述 𝐶(𝛕𝜅𝑝,𝛕𝜅𝑝)=𝛕𝜅𝑝𝐅(𝛕𝜅𝑝) , 和 𝐅(𝛕𝜅𝑝) 作用于离子的赫尔曼-费曼力 𝜅 在晶胞中 𝐑𝑝 。为此,我们暂时忽略汇率相关性贡献 𝑣xc 达到自洽的潜力 𝑉(𝐫) 并考虑包括 Hartree 和离子项在内的静电分量: 𝑉es(𝐫)=𝑉Ha(𝐫)+𝑉ion(𝐫)=𝑉(𝐫)𝑣xc(𝐫) 。通过评估 𝜕𝜅,𝐑𝑝𝑉es(𝐫)𝐫=𝛕𝜅𝑝 ,我们得到原子在 𝛕𝜅𝑝 由于原子的位移 𝛕𝜅𝑝 。该量相对于的梯度 𝛕𝜅𝑝 根据定义,作用在离子上的力位于 𝛕𝜅𝑝 [参见。等式。参考文献的(8)和(13)。 32 ]。因此,Wannier 表示中的声子微扰与原子间力常数矩阵之间存在以下关系:

𝐫[𝜕𝜅,𝐑𝑝𝑉es(𝐫)]𝐫=𝛕𝜅𝑝=𝑍𝜅𝐶(𝛕𝜅𝑝,𝛕𝜅𝑝),
(19)

with 𝑍𝜅 being the electric charge of the ionic species 𝜅. If the system under consideration can be described by the local-density approximation to density-functional theory, the exchange-correlation contribution to the self-consistent potential is short ranged, and 𝜕𝜅,𝐑𝑝𝑉(𝐫) decays within the same distance as 𝜕𝜅,𝐑𝑝𝑉es(𝐫). Hence, by Eq. (19), the phonon perturbation in the Wannier representation will exhibit a spatial localization similar to the interatomic force constants.
𝑍𝜅 是离子种类的电荷 𝜅 。如果所考虑的系统可以通过密度泛函理论的局域密度近似来描述,则交换相关对自洽势的贡献是短范围的,并且 𝜕𝜅,𝐑𝑝𝑉(𝐫) 在相同距离内衰减 𝜕𝜅,𝐑𝑝𝑉es(𝐫) 。因此,通过等式。 (19)中,Wannier 表示中的声子扰动将表现出类似于原子间力常数的空间局域化。

The spatial decay of the force constants has been thoroughly discussed elsewhere,32,58 and we summarize here only the aspects which are relevant to the present work. In metals, the electrostatic interactions are efficiently screened within a few bond lengths; therefore, the force constants are short ranged. In some cases, the topology of the Fermi surface gives rise to Kohn anomalies, which correspond to long-ranged force constants propagating along the wave vector associated with the anomaly. In such cases, the spatial decay of 𝐶(𝛕𝜅𝑝,𝛕𝜅𝑝) will depend on the strength of the anomaly and must be analyzed carefully before proceeding with the methods described in Sec. IV. The possibility of long-ranged interatomic force constants related to Friedel oscillations has also been pointed out in Ref. 58. However, to the authors’ knowledge, no first-principles investigations report long-ranged force constants associated with this effect. In polar insulators, the Born dynamical charges are nonvanishing, and the force constants are long ranged due to their dipolar interactions. 59 In such cases, the procedure described in this work can still be applied by separating the short-range analytical part of the dynamical matrix from the long-range nonanalytic term. 58
力常数的空间衰减已在其他地方进行了深入讨论, 32, 58 ,我们在这里仅总结与当前工作相关的方面。在金属中,静电相互作用在几个键长内被有效地屏蔽;因此,力常数是短范围的。在某些情况下,费米表面的拓扑会引起科恩异常,这对应于沿着与异常相关的波矢量传播的长程力常数。在这种情况下,空间衰减 𝐶(𝛕𝜅𝑝,𝛕𝜅𝑝) 将取决于异常的强度,并且在继续第 2 节中描述的方法之前必须仔细分析。.参考文献中也指出了与弗里德尔振荡相关的长程原子间力常数的可能性。 58 .然而,据作者所知,没有第一性原理研究报告与这种效应相关的长程力常数。在极性绝缘体中,玻恩动力电荷不为零,并且由于它们的偶极相互作用,力常数的变化范围很大。 59在这种情况下,仍然可以通过将动态矩阵的短程分析部分与远程非分析项分开来应用本工作中描述的程序。 58

C. Electron-phonon matrix element
C. 电子声子矩阵元

In order to obtain the e-ph vertex in the joint electron-phonon Wannier representation, we combine Eqs. (6), (10), and (17). After rearranging the terms, we find
为了获得联合电子声子 Wannier 表示中的 e-ph 顶点,我们结合等式: (6)(10)(17) 。重新排列条款后,我们发现

𝑚𝐤+𝐪𝜕𝐪𝜈𝑉𝑛𝐤=1𝑁2𝑒𝑚𝑛𝜅𝐑𝑒𝐑𝑒𝐑𝑝𝑒𝑖[𝐤(𝐑𝑒𝐑𝑒)+𝐪(𝐑𝑝𝐑𝑒)]𝐮𝜈𝐪𝜅𝑈𝑚𝑚,𝐤+𝐪𝑚𝐑𝑒𝜕𝜅,𝐑𝑝𝑉𝑛𝐑𝑒𝑈𝑛𝑛,𝐤.
(20)

Now, we exploit the translational invariance of the “three-point” matrix element:
现在,我们利用“三点”矩阵元素的平移不变性:

𝑚𝐑𝑒𝜕𝜅,𝐑𝑝𝑉𝑛𝐑𝑒=𝑚0𝑒𝜕𝜅,𝐑𝑝𝐑𝑒𝑉𝑛𝐑𝑒𝐑𝑒,
(21)

which is obtained by writing the integral over the infinite crystal and performing a change of variables. In Eq. (21), 𝑚𝟎𝑒 is a Wannier function centered in the unit cell at the origin of the reference frame. By substituting Eq. (21) into Eq. (20) and changing the variables 𝐑𝑒𝐑𝑒 into 𝐑𝑒 and 𝐑𝑝𝐑𝑒 into 𝐑𝑝 we find
这是通过在无限晶体上写入积分并执行变量的变化而获得的。在等式中。 (21) , 𝑚𝟎𝑒 是一个以参考系原点的晶胞为中心的 Wannier 函数。通过代入方程。 (21)代入等式。 (20)并改变变量 𝐑𝑒𝐑𝑒 进入 𝐑𝑒𝐑𝑝𝐑𝑒 进入 𝐑𝑝 我们发现

𝑔(𝐤,𝐪)=1𝑁𝐞𝐑𝑒,𝐑𝑝𝑒𝑖(𝐤𝐑𝑒+𝐪𝐑𝑝)𝑈𝐤+𝐪𝑔(𝐑𝑒,𝐑𝑝)𝑈𝐤𝐮𝐪,
(22)

having introduced the e-ph vertex in the Wannier representation:
在 Wannier 表示中引入 e-ph 顶点:

𝑔𝑚𝑛,𝜈(𝐑𝑒,𝐑𝑝)=𝑚𝟎𝑒𝜕𝜈,𝐑𝑝𝑉𝑛𝐑𝑒.
(23)

In Eq. (22), we omitted band and branch indices for clarity, keeping in mind that the electronic matrices 𝑈𝐤 and 𝑈𝐤+𝐪 act on the band indices of 𝑔(𝐑𝑒,𝐑𝑝) and the phonon matrix 𝐮𝐪 acts on the corresponding branch index. We notice that the “band” and “branch” indices do not have any special meaning in the Wannier representation: In the electron case, they label different Wannier functions belonging to the same unit cell, and in the phonon case, they label a particular atom in the unit cell as well as the Cartesian direction of the corresponding displacement. In order to express the e-ph matrix element in the Wannier representation in terms of the corresponding Bloch vertex, we invert Eq. (23):
在等式中。 (22) ,为了清楚起见,我们省略了能带和分支索引,请记住电子矩阵 𝑈𝐤𝑈𝐤+𝐪 作用于频带指数 𝑔(𝐑𝑒,𝐑𝑝) 和声子矩阵 𝐮𝐪 作用于相应的分支索引。我们注意到,“能带”和“分支”指数在 Wannier 表示中没有任何特殊含义:在电子情况下,它们标记属于同一晶胞的不同 Wannier 函数,而在声子情况下,它们标记特定的 Wannier 函数。原子在晶胞中的位移以及笛卡尔方向的相应位移。为了用相应的 Bloch 顶点来表示 Wannier 表示中的 e-ph 矩阵元素,我们对方程进行反转。 (23)

𝑔(𝐑𝑒,𝐑𝑝)=1𝑁𝑝𝐤,𝐪𝑒𝑖(𝐤𝐑𝑒+𝐪𝐑𝑝)𝑈𝐤+𝐪𝑔(𝐤,𝐪)𝑈𝐤𝐮1𝐪.
(24)

The striking feature of the Wannier vertex [Eq. (24)] is the localization in both the electron and phonon variables. As illustrated in Fig. 2, 𝑚𝟎𝑒𝜕𝜈,𝐑𝑝𝑉𝑛𝐑𝑒 vanishes whenever 𝐑𝑒 or 𝐑𝑝 corresponds to a unit cell sufficiently distant from the origin of the reference frame. As a consequence, in order to accurately describe the e-ph interaction in a given system, we only need to know a small number of matrix elements in the Wannier representation. This elementary observation constitutes the core of this study.
万尼尔顶点的显着特征[方程。 (24) ] 是电子和声子变量的局域化。如图2所示, 𝑚𝟎𝑒𝜕𝜈,𝐑𝑝𝑉𝑛𝐑𝑒 每当 𝐑𝑒 或者 𝐑𝑝 对应于距参考系原点足够远的单位单元。因此,为了准确描述给定系统中的 e-ph 相互作用,我们只需要知道 Wannier 表示中的少量矩阵元素。这一基本观察构成了本研究的核心。

FIG. 2 如图。 2

(Color online) Simplified scheme of the electron and phonon Wannier functions entering the three-point e-ph matrix element [Eq. (23)]. The square lattice indicates the unit cells of the crystal, the red lines the electron Wannier functions 𝟎𝑒 and 𝐑𝑒, and the blue line the phonon perturbation in the Wannier representation 𝜕𝐑𝑝𝑉. Whenever two of these functions are centered on distant unit cells, the e-ph matrix element in the Wannier representation 𝟎𝑒𝜕𝐑𝑝𝑉𝐑𝑒 vanishes.
(在线彩色)进入三点 e-ph 矩阵元素的电子和声子 Wannier 函数的简化方案 [Eq. (23) ]。方形晶格表示晶体的晶胞,红线表示电子万尼尔函数 𝟎𝑒𝐑𝑒 ,蓝线是 Wannier 表示中的声子扰动 𝜕𝐑𝑝𝑉 。每当这些函数中的两个以遥远的晶胞为中心时,Wannier 表示中的 e-ph 矩阵元素 𝟎𝑒𝜕𝐑𝑝𝑉𝐑𝑒 消失。

The relevant range of 𝐑𝑒 and 𝐑𝑝 in Eq. (22) depends on the localization of the electronic Wannier functions and of the phonon perturbations in the Wannier representation. Inspection of Eq. (23) indicates that the spatial decay of the matrix elements is bound by the limiting cases 𝑔𝑚𝑛,𝜈(𝐑𝑒,𝟎𝑝) and 𝑔𝑚𝑛,𝜈(𝟎𝑒,𝐑𝑝). In the first case, the vertex corresponds to the overlap between 𝑤𝑚(𝐫)𝛕𝜅𝑉(𝐫) and 𝑤𝑛(𝐫𝐑𝑒). When the phonon perturbation is sufficiently localized, the matrix element is found to scale as 𝑔𝑚𝑛,𝜈(𝐑𝑒,𝟎𝑝)𝑤𝑛(𝐑𝑒). Therefore, in this case, the spatial decay is dictated by the localization of the electronic Wannier functions. In the second case, the matrix element corresponds to the overlap between 𝑤𝑚(𝐫)𝑤𝑛(𝐫) and the phonon perturbation 𝛕𝜅𝑉(𝐫𝐑𝑝) (assuming for simplicity local pseudopotentials). For sufficiently localized electronic Wannier functions, the matrix element is found to scale as 𝑔𝑚𝑛,𝜈(𝟎𝑒,𝐑𝑝)𝑉(𝐑𝑝), indicating that the spatial decay of the e-ph vertex is determined in this case by the localization of the phonon perturbation. More generally, the slowest decay among these two limiting cases sets the size of the real-space supercell or the 𝐤-point sampling to be considered.
相关范围 𝐑𝑒𝐑𝑝 在等式中(22)取决于电子 Wannier 函数和 Wannier 表示中的声子扰动的局域化。检查方程。 (23)表明矩阵元素的空间衰减受到极限情况的约束 𝑔𝑚𝑛,𝜈(𝐑𝑒,𝟎𝑝)𝑔𝑚𝑛,𝜈(𝟎𝑒,𝐑𝑝) 。在第一种情况下,顶点对应于之间的重叠 𝑤𝑚(𝐫)𝛕𝜅𝑉(𝐫)𝑤𝑛(𝐫𝐑𝑒) 。当声子扰动充分局部化时,发现矩阵元素的比例为 𝑔𝑚𝑛,𝜈(𝐑𝑒,𝟎𝑝)𝑤𝑛(𝐑𝑒) 。因此,在这种情况下,空间衰减由电子万尼尔函数的局域化决定。在第二种情况下,矩阵元素对应于之间的重叠 𝑤𝑚(𝐫)𝑤𝑛(𝐫) 和声子扰动 𝛕𝜅𝑉(𝐫𝐑𝑝) (为简单起见,假设局部赝势)。对于充分定域的电子 Wannier 函数,发现矩阵元素的缩放比例为 𝑔𝑚𝑛,𝜈(𝟎𝑒,𝐑𝑝)𝑉(𝐑𝑝) ,表明在这种情况下,e-ph 顶点的空间衰减是由声子扰动的局域化决定的。更一般地说,这两种极限情况中最慢的衰减决定了真实空间超晶胞或 𝐤 - 考虑点抽样。

Equations (22) and (24) provide a compact and elegant transformation between the e-ph matrix element in the Bloch and the Wannier representations. Interestingly, Eq. (22) is reminiscent of the expressions used in tight-binding calculations to model the e-ph interaction. 60 We notice, however, that our expressions [Eqs. (22) and (24)] provide a fully first-principles description of the e-ph interaction. This observation suggests a systematic approach to determine tight-binding parameters for the e-ph interaction by first performing ab initio calculations in the Bloch representation and then determining the e-ph vertex in the Wannier representation through Eq. (24). An accurate tight-binding parametrization of the e-ph interaction would prove useful in the study of large-scale systems or systems with disorder. 35
方程(22)(24)提供了Bloch和Wannier表示中的e-ph矩阵元素之间的紧凑而优雅的变换。有趣的是,等式。 (22)让人想起用于模拟 e-ph 相互作用的紧束缚计算中的表达式。 60然而,我们注意到,我们的表达式 [Eqs. (22)(24) ] 提供了 e-ph 相互作用的完整第一原理描述。这一观察结果提出了一种系统方法来确定 e-ph 相互作用的紧束缚参数,首先在 Bloch 表示中执行从头计算,然后通过方程 2 确定 Wannier 表示中的 e-ph 顶点。 (24) 。 e-ph 相互作用的精确紧束缚参数化将在大规模系统或无序系统的研究中被证明是有用的。 35

IV. WANNIER-FOURIER INTERPOLATION
四. Wannier-Fourier 插值

In this section, we describe how to exploit the spatial localization in the Wannier representation to calculate the quantities required in the self-energies [Eqs. (3) and (4)] by a generalized Wannier-Fourier interpolation. We first discuss the transformation of the electron eigenstates and eigenvalues, the vibrational modes and frequencies, as well as the e-ph matrix elements from a coarse Brillouin-zone grid (𝐤,𝐪) to the Wannier representation (𝐑𝑒,𝐑𝑝) in the corresponding real-space supercell. Then, we describe the reverse process from the Wannier representation to the Bloch representation at a new set of electron and phonon momenta (𝐤,𝐪).
在本节中,我们将描述如何利用 Wannier 表示中的空间定位来计算自能 [方程 1] 所需的量。 (3)(4) ] 通过广义 Wannier-Fourier 插值。我们首先讨论电子本征态和本征值、振动模式和频率以及粗布里渊区网格的 e-ph 矩阵元素的变换 (𝐤,𝐪) 致 Wannier 代表处 (𝐑𝑒,𝐑𝑝) 在相应的真实空间超晶胞中。然后,我们描述了在一组新的电子和声子动量下从 Wannier 表示到 Bloch 表示的逆过程 (𝐤,𝐪)

A. Bloch to Wannier transform
A. Bloch 到 Wannier 变换
1. Electrons 1. 电子

We calculate the one-particle electronic eigenstates 𝜓𝑛𝐤(𝐫) and eigenvalues 𝜖𝑛𝐤 by adopting standard density-functional techniques. 61–63 The matrix elements of the single-particle Kohn-Sham Hamiltonian ̂𝐻el in the Bloch representation are
我们计算单粒子电子本征态 𝜓𝑛𝐤(𝐫) 和特征值 𝜖𝑛𝐤 通过采用标准密度泛函技术。 61–63单粒子 Kohn-Sham 哈密顿量的矩阵元素 ̂𝐻el 在布洛赫表示中是

𝐻el𝑚𝑛,𝐤=𝑚𝐤̂𝐻el𝑛𝐤=𝛿𝑚𝑛𝜖𝑛𝐤.
(25)

In Eq. (25), the 𝐤 vectors correspond to a uniform grid of size 𝑁𝑒1×𝑁𝑒2×𝑁𝑒3 centered at the 𝛤 point (we assume a three-dimensional system here; the extension to systems with reduced dimensionality is obvious). The uniform and unshifted grid is required to perform the Fourier transform [Eq. (9)]. The eigenstates 𝑛𝐤 and eigenvalues 𝜖𝑛𝐤 are used to determine the unitary matrix 𝑈𝐤 for the transformation to maximally localized Wannier functions. This step involves the calculation of the matrix elements of the periodic position operator 49 and is performed using the method of Refs. 49,64.
在等式中。 (25)𝐤 向量对应于大小统一的网格 𝑁𝑒1×𝑁𝑒2×𝑁𝑒3𝛤 点(我们在这里假设一个三维系统;对降维系统的扩展是显而易见的)。需要均匀且未移动的网格来执行傅立叶变换[等式1]。 (9) ]。本征态 𝑛𝐤 和特征值 𝜖𝑛𝐤 用于确定酉矩阵 𝑈𝐤 用于转换为最大本地化 Wannier 函数。该步骤涉及周期位置算子49的矩阵元素的计算并且使用Refs的方法来执行。 49、64

Once the unitary matrix 𝑈𝐤 has been determined, we calculate the electronic Hamiltonian in the Wannier representation by combining Eqs. (9) and (25):
一旦酉矩阵 𝑈𝐤 已经确定,我们通过结合等式计算 Wannier 表示中的电子哈密顿量。 (9)(25)

𝐻el𝐑𝑒,𝐑𝑒=𝐑𝑒̂𝐻el𝐑𝑒=𝐤𝑒𝑖𝐤(𝐑𝑒𝐑𝑒)𝑈𝐤𝐻el𝐤𝑈𝐤,
(26)

where band indices are omitted for clarity. By construction, the Hamiltonian in the Wannier representation 𝐻el𝐑𝑒,𝐑𝑒 decays with the distance 𝐑𝑒𝐑𝑒. The length scale for the spatial decay is determined by the localization of the electronic Wannier functions.
为了清楚起见,其中省略了频带索引。通过构造,Wannier 表示中的哈密顿量 𝐻el𝐑𝑒,𝐑𝑒 随距离衰减 𝐑𝑒𝐑𝑒 。空间衰减的长度尺度由电子万尼尔函数的定位决定。

2. Phonons 2. 声子

We calculate vibrational eigenmodes 𝐞𝜈𝜅(𝐪) and eigenfrequencies 𝜔𝐪𝜈 through density-functional perturbation theory. 32 This operation is performed for all the 𝐪 vectors belonging to a uniform grid of size 𝑁𝑝1×𝑁𝑝2×𝑁𝑝3 centered at the 𝛤 point (in Sec. V D, we describe how to restrict the 𝐪 points to the irreducible wedge of the Brillouin zone). The use of a uniform and unshifted grid is needed for the Fourier transform in Eq. (18), similarly to the electronic case (cf. Sec. IV A 1). The dynamical matrix in the Bloch representation for phonons is, by definition,
我们计算振动本征模态 𝐞𝜈𝜅(𝐪) 和特征频率 𝜔𝐪𝜈 通过密度泛函微扰理论。 32该操作对所有 𝐪 属于大小均匀网格的向量 𝑁𝑝1×𝑁𝑝2×𝑁𝑝3𝛤 点(在第VD节中,我们描述了如何限制 𝐪 指向布里渊区的不可约楔形)。等式中的傅立叶变换需要使用均匀且未平移的网格。 (18) ,与电子案件类似(参见第IV A 1节)。根据定义,声子的布洛赫表示中的动力学矩阵为:

𝐷ph𝐪,𝜇𝜈=𝐪𝜇̂𝐷ph𝐪𝜈=𝛿𝜇𝜈𝜔2𝐪𝜈.
(27)

Using Eq. (27) and the completeness relation 𝜅𝑝𝛕𝜅𝑝𝛕𝜅𝑝=1, we obtain the dynamical matrix in the phonon Wannier representation:
使用方程式(27)和完备性关系 𝜅𝑝𝛕𝜅𝑝𝛕𝜅𝑝=1 ,我们得到声子 Wannier 表示的动力学矩阵:

𝛕𝜅𝑝̂𝐷ph𝛕𝜅𝑝=𝐪,𝜇𝜈𝑒𝑖𝐪(𝐑𝑝𝐑𝑝)𝐞𝐪𝜅𝐷ph𝐪,𝜇𝜈𝐞𝐪𝜅.
(28)

If we collect the atom label 𝜅 and the Cartesian direction 𝛼 of 𝜏𝜅𝛼 into a composite index 𝜈=𝜅𝛼, the vibrational eigenmodes can be expressed in terms of the square matrices (𝑒𝐪)𝜇𝜈. Accordingly, Eq. (28) can be rewritten in a compact fashion which highlights the analogy with Eq. (26):
如果我们收集原子标签 𝜅 和笛卡尔方向 𝛼𝜏𝜅𝛼 转化为综合指数 𝜈=𝜅𝛼 ,振动本征模态可以用方阵表示 (𝑒𝐪)𝜇𝜈 。因此,等式。 (28)可以以紧凑的方式重写,这突出了与式(28)的类比。 (26)

𝐷ph𝐑𝑝,𝐑𝑝=𝐑𝑝̂𝐷ph𝐑𝑝=𝐪𝑒𝑖𝐪(𝐑𝑝𝐑𝑝)𝑒𝐪𝐷ph𝐪𝑒𝐪.
(29)

In order to make connection with the standard terminology, we observe that the left-hand side of Eq. (28) is related to the matrix of the interatomic force constants by 58
为了与标准术语联系起来,我们观察到等式的左侧。 (28)与原子间力常数矩阵的关系为58

𝛕𝜅𝑝̂𝐷ph𝛕𝜅𝑝=(𝑚𝜅𝑚𝜅)1/2𝐶(𝛕𝜅𝑝,𝛕𝜅𝑝).
(30)

So far, the formalism for the lattice dynamics has been described in complete analogy with the electronic case. There is, however, an important difference between these cases when it comes to the spatial decay of the operators ̂𝐻el and ̂𝐷ph. On the one hand, the Kohn-Sham one-particle Hamiltonian is local in real space (the nonlocality eventually arising from the pseudopotentials being short ranged); therefore, the spatial decay of 𝐻el𝐑𝑒,𝐑𝑒 is dictated by the overlap of the Wannier functions at 𝐑𝑒 and 𝐑𝑒. On the other hand, while the phonon Wannier functions are always infinitely localized by construction (cf. Sec. III B), the operator ̂𝐷ph is nonlocal in real space, as it is clear from Eq. (30). As a consequence, the localization of 𝐷ph𝐑𝑝,𝐑𝑝 does not relate to the overlap of lattice Wannier functions, but instead to the effectiveness of the dielectric screening in the material, as discussed in relation to Eq. (19).
到目前为止,晶格动力学的形式主义已经与电子案例完全类比地被描述。然而,当涉及到算子的空间衰减时,这些情况之间存在一个重要的区别 ̂𝐻el̂𝐷ph 。一方面,Kohn-Sham 单粒子哈密顿量在实空间中是局域性的(最终由于赝势的短程而产生非局域性);因此,空间衰减 𝐻el𝐑𝑒,𝐑𝑒 由 Wannier 函数的重叠决定 𝐑𝑒𝐑𝑒 。另一方面,虽然声子 Wannier 函数总是通过构造无限定域(参见第III B节),但算子 ̂𝐷ph 在实空间中是非局域的,从方程可以清楚地看出。 (30) 。因此,本地化 𝐷ph𝐑𝑝,𝐑𝑝 与晶格 Wannier 函数的重叠无关,而是与材料中电介质屏蔽的有效性有关,如方程式 1 中所讨论的。 (19) .

3. Electron-phonon matrix elements
3. 电子声子矩阵元

The e-ph matrix elements are computed after the electronic eigenstates and eigenvalues and the phonon eigenmodes and eigenfrequencies have been determined. We calculate the matrix elements 𝑔𝑚𝑛,𝜈(𝐤,𝐪) in the Bloch representation [Eq. (6)] using the variation of the self-consistent potential 𝜕𝐪𝜈𝑉 determined by density-functional perturbation theory. 32 The 𝐤 and 𝐪 points belong to the uniform unshifted Brillouin-zone grids with 𝑁𝑒1×𝑁𝑒2×𝑁𝑒3 and 𝑁𝑝1×𝑁𝑝2×𝑁𝑝3 points, respectively.
在确定电子本征态和本征值以及声子本征模式和本征频率后,计算 e-ph 矩阵元素。我们计算矩阵元素 𝑔𝑚𝑛,𝜈(𝐤,𝐪) 在布洛赫表示中 [Eq. (6) ]利用自洽势的变化 𝜕𝐪𝜈𝑉 由密度泛函微扰理论确定。 32那个 𝐤𝐪 点属于均匀不变的布里渊区网格 𝑁𝑒1×𝑁𝑒2×𝑁𝑒3𝑁𝑝1×𝑁𝑝2×𝑁𝑝3 点,分别。

The computation of the dynamical matrix is the most expensive step in the procedure; therefore, it is convenient to restrict the set of 𝐪 points to the irreducible wedge of the Brillouin zone. The remaining points and the associated eigenmodes, eigenfrequencies, and perturbations can be generated by exploiting the crystal symmetries (cf. Sec. V D). The electronic eigenstates are computed on the full uniform grid, i.e., no symmetry reduction is applied in the electronic case. This choice is motivated by the observation that even if the 𝐤 vector in the e-ph matrix elements is restricted to the irreducible wedge, the 𝐤+𝐪 vector spans the whole Brillouin zone since 𝐪 belongs to a uniform grid.
动态矩阵的计算是该过程中最昂贵的步骤;因此,限制集合是很方便的 𝐪 指向布里渊区的不可约楔形。剩余的点和相关的本征模、本征频率和扰动可以通过利用晶体对称性来生成(参见第VD节)。电子本征态是在完整均匀网格上计算的,即在电子情况下不应用对称性约简。这一选择的动机是观察到即使 𝐤 e-ph 矩阵元素中的向量被限制为不可约楔, 𝐤+𝐪 向量跨越整个布里渊区,因为 𝐪 属于均匀网格。

Once the e-ph matrix elements are calculated in the Bloch representation, we use Eq. (24) to transform them into the Wannier representation. The required transformation matrices 𝑈𝐤 and 𝐮𝐪 are already available from the previous steps [Secs. IV A 1 and IV A 2]. Whenever the 𝐤 and 𝐪 grids are commensurate, the electronic transformation matrices 𝑈𝐤+𝐪 are conveniently obtained by mapping the 𝐤+𝐪 points into the original grid of points 𝐤. When 𝐤+𝐪 falls outside the uniform 𝑁𝑒1×𝑁𝑒2×𝑁𝑒3 grid, we use parallel transport and set 𝑈𝐤+𝐪=𝑈𝐤+𝐪+𝐆, where 𝐆 is the reciprocal lattice vector which folds 𝐤+𝐪 back in the original 𝐤 grid. It is worth stressing that this procedure requires only one single minimization of the spread functional to determine maximally localized Wannier functions. The choice of commensurate 𝐤 and 𝐪 grids does not represent a limitation, since the two grids are already assumed to be uniform and unshifted. One possible exception is discussed in Sec. V A.
一旦以 Bloch 表示计算出 e-ph 矩阵元素,我们就可以使用等式: (24)将它们转化为Wannier表示。所需的变换矩阵 𝑈𝐤𝐮𝐪 从前面的步骤中已经可以使用[秒。 IV A 1IV A 2 ]。每当 𝐤𝐪 网格相称,电子变换矩阵 𝑈𝐤+𝐪 通过映射可以方便地获得 𝐤+𝐪 点到原始点网格中 𝐤 。什么时候 𝐤+𝐪 落在制服之外 𝑁𝑒1×𝑁𝑒2×𝑁𝑒3 网格,我们使用并行传输并设置 𝑈𝐤+𝐪=𝑈𝐤+𝐪+𝐆 , 在哪里 𝐆 是折叠的倒格向量 𝐤+𝐪 回到原来的样子 𝐤 网格。值得强调的是,该过程仅需要一次最小化扩展函数即可确定最大局部 Wannier 函数。相称的选择 𝐤𝐪 grids 并不表示限制,因为已经假设两个网格是均匀且未移动的。一个可能的例外在第 2 节中讨论。弗吉尼亚州

B. Wannier to Bloch transform
B. Wannier 到 Bloch 变换
1. Electrons 1. 电子

We wish to calculate electronic eigenstates 𝑛𝐤 and eigenvalues 𝜖𝑛𝐤 for a set of wave vectors 𝐤 on a significantly finer grid than the original one with 𝑁𝑒1×𝑁𝑒2×𝑁𝑒3 points. According to the application at hand, we could adopt a uniform grid, a grid extending over the irreducible wedge of the Brillouin zone, or a path along relevant high-symmetry lines. In the transformation from the Wannier to the Bloch representation, a uniform and unshifted grid is no longer required (cf. Sec. IV A 1).
我们希望计算电子本征态 𝑛𝐤 和特征值 𝜖𝑛𝐤 对于一组波向量 𝐤 在比原始网格更细的网格上 𝑁𝑒1×𝑁𝑒2×𝑁𝑒3 点。根据当前的应用,我们可以采用均匀网格、在布里渊区不可约楔形上延伸的网格或沿相关高对称线的路径。在从 Wannier 表示到 Bloch 表示的转换中,不再需要均匀且未移动的网格(参见第IV A 1节)。

By combining Eqs. (10), (25), and (26), we obtain
通过结合等式。 (10)(25)(26) ,我们得到

𝐻el𝐤=𝑈𝐤(1𝑁𝑒𝐑𝑒𝑒𝑖𝐤𝐑𝑒𝐻el𝟎𝑒,𝐑𝑒)𝑈𝐤,
(31)

where we have omitted band indices for clarity. In Eq. (31), the sum extends over the unit cells 𝐑𝑒 belonging to the Wigner-Seitz supercell corresponding to 𝑁𝑒1×𝑁𝑒2×𝑁𝑒3 replicas of the primitive cell. For 𝐑𝑒 outside this Wigner-Seitz volume, we assume that the matrix elements of the Hamiltonian 𝐻el𝟎𝑒,𝐑𝑒 are negligibly small. The quality of the final results strictly relies on this assumption, which must be verified numerically before proceeding with the calculations.
为了清楚起见,我们省略了频带索引。在等式中。 (31) ,总和扩展到晶胞上 𝐑𝑒 属于 Wigner-Seitz 超胞对应于 𝑁𝑒1×𝑁𝑒2×𝑁𝑒3 原始细胞的复制品。为了 𝐑𝑒 在这个 Wigner-Seitz 体积之外,我们假设哈密顿量的矩阵元素 𝐻el𝟎𝑒,𝐑𝑒 小得可以忽略不计。最终结果的质量严格依赖于这一假设,在进行计算之前必须对其进行数值验证。

In Eq. (31), the only known quantity is contained within the brackets. We do not know at this stage the transformation matrices 𝑈𝐤 for the new points 𝐤, nor we can determine maximally localized Wannier functions through the method of Ref. 49 since the Bloch eigenvalues and eigenstates at 𝐤 are unknown as well. However, we do know that, by construction, the Hamiltonian 𝐻𝐤 on the left-hand side of Eq. (31) is diagonal in the band indices [cf. Eq. (25)]. This implies that the 𝑈𝐤 matrix is nothing but the diagonalizer of the term within the brackets 𝑁1𝑒𝐑𝑒exp(𝑖𝐤𝐑𝑒)𝐻el𝟎𝑒,𝐑𝑒. Therefore, to find eigenstates and eigenvalues of the electronic Hamiltonian at an arbitrary wave vector 𝐤, we need to perform (i) a Fourier interpolation of the Hamiltonian in the Wannier representation, corresponding to the term within the brackets in Eq. (31), and (ii) a diagonalization of the resulting matrix, yielding 𝑈𝐤, 𝜖𝑛𝐤, as well as the new Bloch eigenstates 𝑛𝐤 through Eq. (9). The procedure outlined in this section was first proposed in Ref. 53 and subsequently applied to the study of the anomalous Hall effect 65 and the magnetic circular dichroism. 66
在等式中。 (31) ,唯一已知的量包含在括号内。我们现阶段不知道变换矩阵 𝑈𝐤 对于新的点 𝐤 ,我们也无法通过 Ref. 的方法确定最大局部 Wannier 函数。 49由于布洛赫特征值和特征态 𝐤 也是未知的。然而,我们确实知道,通过构造,哈密顿量 𝐻𝐤 在等式的左侧。 (31)是能带索引中的对角线[cf.等式。 (25) ]。这意味着 𝑈𝐤 矩阵只不过是括号内项的对角线 𝑁1𝑒𝐑𝑒exp(𝑖𝐤𝐑𝑒)𝐻el𝟎𝑒,𝐑𝑒 。因此,求任意波矢处电子哈密顿量的本征态和本征值 𝐤 ,我们需要执行 (i) Wannier 表示中的哈密顿量的傅立叶插值,对应于方程中括号内的项。 (31)和 (ii) 对所得矩阵进行对角化,得到 𝑈𝐤 , 𝜖𝑛𝐤 ,以及新的布洛赫本征态 𝑛𝐤 通过方程式(9) .本节概述的程序首先在参考文献中提出。 53随后应用于反常霍尔效应65和磁圆二色性的研究。 66

2. Phonons 2. 声子

The calculation of phonon eigenmodes and eigenfrequencies at a new set of 𝐪 points proceeds along the same lines as for the electrons (Sec. IV B 1). The new set of points 𝐪 may correspond to a fine mesh (not necessarily uniform) or to a path in reciprocal space, depending on the application. We formally invert Eq. (29) to obtain
计算一组新的声子本征模式和本征频率 𝐪 点沿着与电子相同的路线进行(第IV B 1节)。新的点集 𝐪 可以对应于精细网格(不一定是均匀的)或倒易空间中的路径,具体取决于应用。我们正式反转方程。 (29)获得

𝐷ph𝐪=𝑒𝐪(1𝑁𝑝𝐑𝑝𝑒𝑖𝐪𝐑𝑝𝐷ph𝟎𝑝,𝐑𝑝)𝑒𝐪,
(32)

in complete analogy with the corresponding expression for the electrons [Eq. (31)]. In Eq. (32), the sum extends over the unit cells 𝐑𝑝 belonging to the Wigner-Seitz cell constructed for the supercell with 𝑁𝑝1×𝑁𝑝2×𝑁𝑝3 replicas of the primitive cell. The matrix elements of the dynamical matrix in the Wannier representation 𝐷ph𝟎𝑝,𝐑𝑝 are assumed to be vanishing outside this Wigner-Seitz supercell. In practice, the length scale of the spatial decay of 𝐷ph𝟎𝑝,𝐑𝑝 with 𝐑𝑝 determines the size of the original mesh of points 𝑁𝑝1×𝑁𝑝2×𝑁𝑝3 required for a given target accuracy.
与电子的相应表达式完全类似[方程。 (31) ]。在等式中。 (32) ,总和延伸到晶胞上 𝐑𝑝 属于为超级电池构建的维格纳-塞茨电池 𝑁𝑝1×𝑁𝑝2×𝑁𝑝3 原始细胞的复制品。 Wannier 表示中动力矩阵的矩阵元素 𝐷ph𝟎𝑝,𝐑𝑝 假设在维格纳-塞茨超胞外消失。在实践中,空间衰减的长度尺度 𝐷ph𝟎𝑝,𝐑𝑝𝐑𝑝 确定原始网格点的大小 𝑁𝑝1×𝑁𝑝2×𝑁𝑝3 给定目标精度所需的。

As for the electrons (cf. Sec. IV B 1), the known quantity in Eq. (31) is the term within the brackets, and the matrix on the left-hand side is diagonal by construction [cf. Eq. (27)]. Hence, the eigenmodes 𝐞𝐪 at the new 𝐪 points must be found by diagonalizing the term within the brackets 𝑁1𝑝𝐑𝑝exp(𝑖𝐪𝐑𝑝)𝐷ph𝟎𝑝,𝐑𝑝. Even in the case of phonons, the determination of eigenfrequencies and eigenmodes at arbitrary momenta 𝐪 requires two steps: (i) a Fourier interpolation of the dynamical matrix in the Wannier representation, corresponding to the term within brackets in Eq. (32), and (ii) a diagonalization of the resulting matrix, yielding the eigenmodes 𝐞𝐪 and squared eigenfrequencies 𝜔2𝐪𝜈. This procedure corresponds to the standard approach used to compute complete phonon dispersions starting from a small grid in the Brillouin zone.32,58
至于电子(参见第IV B 1节),方程中的已知数量。 (31)是括号内的项,左侧的矩阵通过构造是对角的[参见.等式。 (27) ]。因此,本征模态 𝐞𝐪 在新的 𝐪 必须通过对角化括号内的项来找到点 𝑁1𝑝𝐑𝑝exp(𝑖𝐪𝐑𝑝)𝐷ph𝟎𝑝,𝐑𝑝 。即使在声子的情况下,也可以确定任意动量的本征频率和本征模态 𝐪 需要两个步骤:(i)Wannier 表示中的动态矩阵的傅里叶插值,对应于等式中括号内的项。 (32)和 (ii) 对所得矩阵进行对角化,产生本征模式 𝐞𝐪 和平方特征频率 𝜔2𝐪𝜈 。该过程对应于用于从布里渊区的小网格开始计算完整声子色散的标准方法。 32、58

3. Electron-phonon matrix elements
3. 电子声子矩阵元

The calculation of the e-ph matrix element in the new sets of points 𝐤 and 𝐪 is performed by using Eqs. (22) and the transformation matrices 𝑈𝐤, 𝑈𝐤+𝐪, and 𝐞𝐪 determined in the previous steps [Secs. IV B 1 and IV B 2]. After this operation, we have all the ingredients needed to evaluate the physical quantities in Eqs. (3), (4), (7), and (8), with an accurate and ultradense sampling of the Brillouin zone. The entire Wannier-Fourier interpolation procedure described in this section is summarized in Table I.
新点集中e-ph矩阵元素的计算 𝐤𝐪 是通过使用等式来执行的。 (22)和变换矩阵 𝑈𝐤 , 𝑈𝐤+𝐪 , 和 𝐞𝐪 在前面的步骤中确定[秒。 IV B 1IV B 2 ]。经过此操作,我们就拥有了评估方程(1)中的物理量所需的所有成分。 (3)(4)(7)(8) ,对布里渊区进行精确且超密集的采样。本节中描述的整个 Wannier-Fourier 插值过程总结在表中。

TABLE I. 表一

Summary of the Wannier-Fourier interpolation scheme from a set of matrix elements on a uniform grid (𝐤,𝐪) to another set with arbitrary momenta (𝐤,𝐪). The leftmost column contains input quantities determined by density-functional calculations. The second column indicates the operations required to transform from the Bloch to the Wannier representation. The quantities in the Wannier representation thus obtained are reported in the third column. At this stage, we could work directly within the Wannier representation, or transform back to the Bloch representation by Fourier interpolation. The inverse transforms are indicated in the fourth column, while the rightmost column gives the final quantities.
均匀网格上一组矩阵元素的 Wannier-Fourier 插值方案摘要 (𝐤,𝐪) 到另一个具有任意动量的集合 (𝐤,𝐪) 。最左边的列包含由密度泛函计算确定的输入量。第二列表示从 Bloch 表示转换为 Wannier 表示所需的操作。由此获得的 Wannier 表示中的数量报告在第三列中。在这个阶段,我们可以直接在 Wannier 表示中工作,或者通过傅里叶插值转换回 Bloch 表示。逆变换显示在第四列中,而最右边的列给出了最终的数量。

Bloch 布洛赫 BlochWannier Wannier 万尼尔 WannierBloch Bloch 布洛赫
Electrons 电子 𝐻el𝐤 Rotate 𝐻el𝐤 with 𝑈𝐤 and
旋转 𝐻el𝐤 𝑈𝐤
𝐻el𝐑𝑒,𝐑𝑒 Inverse Fourier transform 𝐻el𝐑𝑒,𝐑𝑒
傅里叶逆变换 𝐻el𝐑𝑒,𝐑𝑒
𝐻el𝐤
𝑈𝐤 Fourier transform [Eq. (26)]
傅里叶变换 [Eq. (26) ]
to 𝐤 and diagonalize [Eq. (31)]
𝐤 并对角化 [Eq. (31) ]
𝑈𝐤
Phonons 声子 𝐷ph𝐪 Rotate 𝐷ph𝐪 with 𝑒𝐪 and
旋转 𝐷ph𝐪 𝑒𝐪
𝐷ph𝐑𝑝,𝐑𝑝 Inverse Fourier transform 𝐷ph𝐑𝑝,𝐑𝑝
傅里叶逆变换 𝐷ph𝐑𝑝,𝐑𝑝
𝐷ph𝐪
𝑒𝐪 Fourier transform [Eq. (29)]
傅立叶变换 [Eq. (29) ]
to 𝐪 and diagonalize [Eq. (32)]
𝐪 并对角化 [Eq. (32) ]
𝑒𝐪
e-ph matrix e-ph矩阵 𝑔(𝐤,𝐪) Rotate 𝑔(𝐤,𝐪) with 𝑈𝐤, 𝑈𝐤+𝐪, 𝑒𝐪,
旋转 𝑔(𝐤,𝐪) 𝑈𝐤 , 𝑈𝐤+𝐪 , 𝑒𝐪 ,
𝑔(𝐑𝑒,𝐑𝑝) Inverse Fourier transform 𝑔(𝐑𝑒,𝐑𝑝) to 𝐤, 𝐪
傅里叶逆变换 𝑔(𝐑𝑒,𝐑𝑝) 𝐤 , 𝐪
𝑔(𝐤,𝐪)
elements 元素 𝑈𝐤, 𝑈𝐤+𝐪, 𝐞𝐪
𝑈𝐤 , 𝑈𝐤+𝐪 , 𝐞𝐪
and Fourier transform [Eq. (24)]
和傅立叶变换 [Eq. (24) ]
and rotate with 𝑈𝐤, 𝑈𝐤+𝐪, and 𝑒𝐪 [Eq. (22)]
并旋转 𝑈𝐤 , 𝑈𝐤+𝐪 , 和 𝑒𝐪 [等式。 (22) ]
V. SPECIAL CASES AND PRACTICAL DETAILS
五、特殊情况和实际细节
A. Electron-only Wannier representation
A. 仅电子 Wannier 表示

In some applications, we could be interested in the self-energy of only a limited set of phonon modes (for instance, modes at high-symmetry points), rather than the whole vibrational spectrum in the Brillouin zone. In such cases, calculating the dynamical matrix for every 𝐪 vector in a uniform grid as described in Sec. IV A 2 may turn out to be too expensive from a computational standpoint, and it is desirable to find an alternative path.
在某些应用中,我们可能只对一组有限的声子模式(例如,高对称点处的模式)的自能感兴趣,而不是布里渊区的整个振动谱。在这种情况下,计算每个的动态矩阵 𝐪 向量在均匀网格中,如第 2 节所述。从计算的角度来看, IV A 2可能过于昂贵,并且希望找到替代路径。

The easiest way to proceed in such cases consists in transforming the electronic states in the Wannier representation (Sec. III A) while keeping the phonon perturbation (Sec. III B) in the Bloch representation. The transformation laws of the e-ph vertex in such electron-only Wannier representation read
在这种情况下,最简单的方法是转换 Wannier 表示中的电子态(第III A节),同时保持 Bloch 表示中的声子扰动(第III B节)。这种纯电子 Wannier 表示中 e-ph 顶点的变换定律为

𝑔(𝐤,𝐪)=1𝑁𝑒𝐑𝑒𝑒𝑖𝐤𝐑𝑒𝑈𝐤+𝐪𝑔(𝐑𝑒,𝐪)𝑈𝐤,
(33)
𝑔(𝐑𝑒,𝐪)=𝐤𝑒𝑖𝐤𝐑𝑒𝑈𝐤+𝐪𝑔(𝐤,𝐪)𝑈𝐤,
(34)

with the e-ph matrix element in the mixed representation given by
混合表示中的 e-ph 矩阵元素由下式给出

𝑔𝑚𝑛,𝜈(𝐑𝑒,𝐪)=𝑚𝟎𝑒𝜕𝑉𝐪𝜈𝑛𝐑𝑒.
(35)

In this case, the wave vector 𝐪 is not required to be commensurate with the uniform electronic grid. Accordingly, two minimizations (for every wave vector 𝐪) of the Berry-phase spread functional are required to determine maximally localized Wannier functions: one for the set of states {𝜓𝑛𝐤}, yielding the matrix 𝑈𝐤, and another one for the set {𝜓𝑛𝐤+𝐪}, providing the matrix 𝑈𝐤+𝐪. The inverse transformation from the Wannier to the Bloch representation on arbitrary points 𝐤 proceeds as described in Sec. IV B 1.
在这种情况下,波矢量 𝐪 不要求与统一电子网格相适应。因此,两个最小化(对于每个波矢 𝐪 ) 需要贝里相扩展函数来确定最大局部 Wannier 函数:一个用于状态集 {𝜓𝑛𝐤} ,产生矩阵 𝑈𝐤 ,以及该组的另一个 {𝜓𝑛𝐤+𝐪} ,提供矩阵 𝑈𝐤+𝐪 。任意点上从 Wannier 表示到 Bloch 表示的逆变换 𝐤 如第 2 节所述进行。四、B 1

B. Zone-center phonons and frozen-phonon methods
B. 区域中心声子和冻结声子方法

In the case of a very large system in a supercell geometry, it could be convenient to restrict the sampling of the vibrational Brillouin zone to the 𝛤 point only. This is appropriate whenever the interatomic force constants decay to negligible values over a distance smaller than the size of the supercell. This situation corresponds to the case 𝐪=𝟎 of the electron-only interpolation described in Sec. V A and only requires the determination of the transformation matrices 𝑈𝐤 once.
在超级单元几何结构中的非常大的系统的情况下,将振动布里渊区的采样限制在 𝛤 仅点。当原子间力常数在小于超晶胞尺寸的距离内衰减到可忽略不计的值时,这是合适的。这种情况对应的是案例 𝐪=𝟎 第 2 节中描述的纯电子插值。 VA并且只需要确定变换矩阵 𝑈𝐤 一次。

This situation is also interesting because the procedure described so far can be performed without resorting to linear-response techniques: the matrix of the interatomic force constants [Eq. (30)] can be calculated by taking finite differences of the total energy (frozen-phonon approach), 67 and our procedure can be implemented as a postprocessing step in any electronic-structure package performing total-energy calculations.
这种情况也很有趣,因为到目前为止描述的过程可以在不诉诸线性响应技术的情况下执行:原子间力常数的矩阵[方程1]。 (30) ] 可以通过总能量的有限差分来计算(冻结声子方法), 67并且我们的程序可以作为任何执行总能量计算的电子结构包中的后处理步骤来实现。

C. Gauge arbitrariness C. 衡量任意性

The diagonalization of the Kohn-Sham single-particle Hamiltonian determines the eigenfunctions 𝜓𝑛𝐤 up to an arbitrary phase factor. In the presence of accidental degeneracies, the arbitrariness also includes a unitary transformation within the degenerate manifold. In general, this gauge arbitrariness bears no implications on the calculation of ground-state properties such as total energies and its derivatives. However, the actual localization of the Wannier functions [Eq. (9)] crucially depends on the phases of the wave functions and may be compromised if this gauge freedom is not dealt with properly. Indeed, small variations in the procedures adopted to diagonalize the Hamiltonian may lead to completely different phase settings. This is especially important since the calculation of maximally localized Wannier functions and the calculation of the e-ph matrix elements are performed as two separate and subsequent steps.
Kohn-Sham 单粒子哈密顿量的对角化确定了本征函数 𝜓𝑛𝐤 高达任意相位因子。在存在偶然简并的情况下,任意性还包括简并流形内的幺正变换。一般来说,这种规范任意性对总能量及其导数等基态性质的计算没有影响。然而,Wannier 函数的实际本地化[方程。 (9) ] 关键取决于波函数的相位,如果没有正确处理规范自由度,则可能会受到影响。事实上,对哈密顿量进行对角化所采用的程序的微小变化可能会导致完全不同的相位设置。这一点尤其重要,因为最大局部 Wannier 函数的计算和 e-ph 矩阵元素的计算是作为两个单独的后续步骤执行的。

Different phase settings may arise, for instance, (i) when the eigenstates used to determine 𝑈𝐤 with the procedure of Ref. 49 are obtained with a different diagonalization algorithm than that used in the calculation of the e-ph matrix elements; (ii) when different algorithms are used to deal with the cases 𝐪=𝟎 or 𝐪𝟎; (iii) when 𝐤+𝐪 falls outside the first Brillouin zone and can be folded into another point 𝐤 in the first zone, but the Bloch phases exp[𝑖(𝐤+𝐪)𝐫] and exp(𝑖𝐤𝐫) modify the Fourier coefficients of the nonlocal projectors of the pseudopotentials; and (iv) when different architectures or different parallel environments within the same architecture are used for the calculation of 𝑈𝐤 and the e-ph matrix elements 𝑔𝑚𝑛,𝜈(𝐤,𝐪), respectively.
例如,可能会出现不同的相位设置,(i) 当本征态用于确定 𝑈𝐤 与参考文献的程序。 49是通过与计算 e-ph 矩阵元素时使用的算法不同的对角化算法获得的; (ii) 当使用不同的算法来处理案例时 𝐪=𝟎 或者 𝐪𝟎 ; (三) 当 𝐤+𝐪 落在第一个布里渊区之外,可以折叠到另一个点 𝐤 在第一个区域,但布洛赫阶段 exp[𝑖(𝐤+𝐪)𝐫]exp(𝑖𝐤𝐫) 修改赝势的非局部投影的傅里叶系数; (iv) 当不同架构或同一架构内的不同并行环境用于计算时 𝑈𝐤 和 e-ph 矩阵元素 𝑔𝑚𝑛,𝜈(𝐤,𝐪) , 分别。

It is therefore desirable to fix a unique and unambiguous gauge for the wave functions. We can accomplish this in two separate steps: we first set the gauge within each degenerate manifold, and then we set the phase of every eigenstate individually. The latter step is straightforwardly performed by requiring the wave functions 𝜓𝑛𝐤(𝐫) to be real valued at some arbitrarily chosen point 𝐫. In a plane-wave formalism, the same goal can be achieved more efficiently by requiring the largest Fourier component of the wave functions to be real valued. In practice, it is sufficient to search for the largest coefficient in a small subset of plane waves. The first step is performed by borrowing standard techniques from degenerate perturbation theory. 68 To this end, we consider a fictitious perturbation 𝛼̂𝑉fict which lifts the degeneracies of the Hamiltonian. We compute the matrix elements of this perturbation in the degenerate manifold:

𝑉fict𝑚𝑛=𝑚𝐤̂𝑉fict𝑛𝐤,
(36)

and we diagonalize the perturbation to find the new eigenstates:

[𝐵𝐤𝑉fict𝐵𝐤]𝑚𝑛=𝛿𝑚𝑛𝑣fict𝑛,
(37)

with 𝑣fict𝑛 being the eigenvalues of the fictitious perturbation. At this point, we consider the new eigenstates
𝑣fict𝑛 是虚拟扰动的特征值。此时,我们考虑新的本征态

𝜓𝑛𝐤=𝑚𝐵𝐤,𝑛𝑚𝜓𝑚𝐤,
(38)

which diagonalize the Hamiltonian ̂𝐻el+𝛼̂𝑉fict and are nondegenerate by construction. The strength 𝛼 of the fictitious perturbation is now set to zero to recover the eigenvalues of the original Hamiltonian. In order for the gauge matrix 𝐵𝐤 to be unitary, the perturbation ̂𝑉fict must be chosen Hermitian. This is achieved by constructing a real-valued local fictitious potential which does not contain any of the symmetries of the Hamiltonian.
对哈密顿量进行对角化 ̂𝐻el+𝛼̂𝑉fict 并且不会因构造而退化。实力 𝛼 虚拟扰动的现在设置为零以恢复原始哈密顿量的特征值。为了得到规范矩阵 𝐵𝐤 为单一的,扰动 ̂𝑉fict 必须选择埃尔米特式。这是通过构建不包含任何哈密顿量对称性的实值局部虚拟势来实现的。

It is worth noting at this point that the e-ph matrix element 𝑔𝑚𝑛,𝜈(𝐤,𝐪) is not a gauge-invariant quantity itself, since both the electronic eigenstates and the phonon eigenmodes are defined up to a phase (possibly a unitary matrix). This is consistent with the fact that the matrix element is not a physical observable. On the other hand, the self-energies [Eqs. (3) and (4)] contribute to the electron and phonon spectral functions, which are physical observables (for instance, by photoemission or tunneling experiments). Therefore, we expect the self-energies 𝛴𝐤(𝜔) and 𝛱𝐪(𝜔) to be gauge invariant. This is actually the case, because the quantities 𝑚𝜈𝑔𝑚𝑛,𝜈(𝐤,𝐪)2 and 𝑚𝑛𝑔𝑚𝑛,𝜈(𝐤,𝐪)2 entering Eqs. (3) and (4) are both gauge invariant when the summations are restricted to degenerate subspaces (this property corresponds to the invariance of the Fröbenius norm under similarity transforms).
此时值得注意的是 e-ph 矩阵元素 𝑔𝑚𝑛,𝜈(𝐤,𝐪) 本身不是规范不变量,因为电子本征态和声子本征模都被定义为一个相位(可能是酉矩阵)。这与矩阵元素不是物理可观测量的事实是一致的。另一方面,自能[方程。 (3)(4) ] 有助于电子和声子光谱函数,这些函数是物理可观测的(例如,通过光电发射或隧道实验)。因此,我们期望自能 𝛴𝐤(𝜔)𝛱𝐪(𝜔) 是规范不变的。实际情况确实如此,因为数量 𝑚𝜈𝑔𝑚𝑛,𝜈(𝐤,𝐪)2𝑚𝑛𝑔𝑚𝑛,𝜈(𝐤,𝐪)2 输入方程式。当求和被限制在简并子空间时, (3)(4)都是规范不变的(该属性对应于相似变换下弗罗贝尼乌斯范数的不变性)。

D. Irreducible wedge of the Brillouin zone
D. 布里渊区的不可约楔形

The most computationally intensive part of the procedure described thus far is represented by the calculation of the vibrational eigenmodes, eigenfrequencies, and the associated phonon perturbation for all the 𝐪 points needed in the Fourier transforms [Eqs. (17) and (24)]. It is therefore important to consider the possibility of restricting the set of the required 𝐪 vectors to the irreducible wedge of the Brillouin zone.
迄今为止所描述的过程中计算量最大的部分是通过计算所有的振动本征模式、本征频率和相关的声子扰动来表示的。 𝐪 傅里叶变换中所需的点[方程。 (17)(24) ]。因此,重要的是要考虑限制所需的集合的可能性 𝐪 向量到布里渊区的不可约楔形。

The irreducible 𝐪 points are determined by considering the set of vectors which are nonequivalent under the symmetry operations of the crystal point group. We follow the convention of Ref. 69 in labeling the symmetry operations as {𝒮𝐯}, in such a way that {𝒮𝐯}𝐫=𝒮𝐫+𝐯, with 𝒮 being the rotational part (proper or improper) of the symmetry operation and 𝐯 the eventually associated fractional translation. Given a 𝐪 vector in the irreducible part of the Brillouin zone, we can generate the so-called star of 𝐪 by applying all the crystal symmetry operations.
不可约的 𝐪 点是通过考虑在晶体点群的对称运算下不等价的向量集来确定的。我们遵循参考文献的约定。 69将对称运算标记为 {𝒮𝐯} ,以这样的方式 {𝒮𝐯}𝐫=𝒮𝐫+𝐯 , 和 𝒮 是对称操作的旋转部分(适当或不适当),并且 𝐯 最终关联的分数翻译。给定一个 𝐪 向量在布里渊区的不可约部分,我们可以生成所谓的星形 𝐪 通过应用所有晶体对称操作。

Once the dynamical matrix and the phonon perturbation for an irreducible 𝐪 point have been calculated, the corresponding quantities for a wave vector 𝒮𝐪 belonging to the star of 𝐪 can be determined by exploiting the transformation properties of the vibrational eigenmodes under the symmetry operation {𝒮𝐯}. The transformation law for the eigenmodes is given by 69
一旦动力学矩阵和声子扰动为不可约 𝐪 点已经计算出来,波矢的相应量 𝒮𝐪 属于 的 明星 𝐪 可以通过利用对称运算下振动本征模态的变换特性来确定 {𝒮𝐯} 。本征模态的变换定律由下式给出69

𝐞𝒮𝐪=𝛤{𝒮𝐯}(𝐪)𝐞𝐪,
(39)

where the unitary matrix 𝛤{𝒮𝐯}(𝐪) is defined as
其中酉矩阵 𝛤{𝒮𝐯}(𝐪) 定义为

[𝛤{𝒮𝐯}(𝐪)]𝜅𝜅𝛼𝛽=𝒮𝛼𝛽𝛿𝜅,𝐹𝒮(𝜅)𝑒𝑖𝒮𝐪[𝛕𝜅{𝒮𝐯}𝛕𝜅].
(40)

In Eq. (40) 𝛼 and 𝛽 indicate Cartesian directions, and 𝐹𝒮(𝜅) represents the atom that 𝜅 is brought into by the symmetry operation {𝒮𝐯}. Equation (39) implicitly assumes that eigenmodes at 𝐪 and 𝒮𝐪 carry no phase difference and that no gauge mixing occurs whenever two or more eigenmodes are degenerate. This choice implies that, when transforming the dynamical matrix and the e-ph matrix elements from the Bloch to the Wannier representation, one cannot directly apply the symmetry {𝒮𝐯} to the interatomic force constants and then diagonalize the resulting dynamical matrix. 32 Indeed, such procedure would lead to eigenmodes which do not obey the phase relations defined by Eq. (39). The correct procedure instead is to first transform the eigenmodes according to Eq. (39) and then generate the corresponding dynamical matrix through 𝐞𝒮𝐪𝐷ph𝐪𝐞𝒮𝐪, with 𝐷ph𝐪 being the diagonal matrix defined by Eq. (27) for the wave vector 𝐪.
在等式中。 (40) 𝛼𝛽 指示笛卡尔方向,并且 𝐹𝒮(𝜅) 代表原子 𝜅 由对称运算带入 {𝒮𝐯} 。方程(39)隐含地假设特征模态在 𝐪𝒮𝐪 没有相位差,并且只要两个或多个本征模态简并,就不会发生规范混合。这一选择意味着,当将动力学矩阵和 e-ph 矩阵元素从 Bloch 表示转换为 Wannier 表示时,不能直接应用对称性 {𝒮𝐯} 到原子间力常数,然后对角化所得的动力学矩阵。 32事实上,这样的过程会导致本征模不遵守方程式定义的相位关系。 (39) .正确的程序是首先根据方程变换本征模态。 (39)然后通过生成相应的动力矩阵 𝐞𝒮𝐪𝐷ph𝐪𝐞𝒮𝐪 , 和 𝐷ph𝐪 是由方程定义的对角矩阵。 (27)为波矢 𝐪

The transformation law of the phonon perturbation 𝜕𝑉𝐪𝜈(𝐫) under the operation {𝒮𝐯} is most easily worked out by noting that (i) the displacement 𝛥𝛕𝐪𝜈𝜅𝑝 [Eq. (12)] is both transferred to the atom 𝜅=𝐹𝒮(𝜅) in the unit cell 𝐑𝑝=𝒮𝐑𝑝 and rotated according to 𝒮:
声子摄动变换定律 𝜕𝑉𝐪𝜈(𝐫) 正在运营中 {𝒮𝐯} 最容易计算出来的是 (i) 位移 𝛥𝛕𝐪𝜈𝜅𝑝 [等式。 (12) ] 均转移至原子 𝜅=𝐹𝒮(𝜅) 在晶胞中 𝐑𝑝=𝒮𝐑𝑝 并根据旋转 𝒮 :

𝛥𝛕𝒮𝐪,𝜈𝜅𝑝=𝒮𝛥𝛕𝐪𝜈𝜅𝑝,
(41)

(ii) the dielectric function is invariant under the symmetry operation {𝒮𝐯}:
(ii) 介电函数在对称运算下不变 {𝒮𝐯} :

𝜖({𝒮𝐯}𝐫,{𝒮𝐯}𝐫)=𝜖(𝐫,𝐫),
(42)

and (iii) the ionic (pseudo)potentials are rotationally invariant:
(iii) 离子(赝)势旋转不变:

𝑉ion𝜅(𝒮𝐫)=𝑉ion𝜅(𝐫).
(43)

By replacing Eqs. (41)–(43) into Eqs. (11)–(13), we obtain, after some algebra,
通过替换方程式。 (41)–(43)代入等式。 (11)–(13) ,经过一些代数计算,我们得到,

𝜕𝑉𝒮𝐪(𝐫)=𝜕𝑉𝐪({𝒮𝐯}1𝐫).
(44)

It is important to realize that the derivation of Eq. (44) rests on the choice of the phases of the vibrational eigenmodes at 𝐪 and 𝒮𝐪 provided by Eq. (39).
重要的是要认识到方程的推导。 (44)取决于振动本征模态相位的选择 𝐪𝒮𝐪 由方程式提供。 (39) .

We can now exploit Eq. (44) to calculate the e-ph matrix element corresponding to a momentum transfer of 𝒮𝐪:
我们现在可以利用方程。 (44)计算对应于动量传递的 e-ph 矩阵元素 𝒮𝐪 :

𝑔𝑚𝑛,𝜈(𝐤,𝒮𝐪)=𝜓𝑚𝐤+𝒮𝐪(𝐫)𝜕𝑉𝐪({𝒮𝐯}1𝐫)𝜓𝑛𝐤(𝐫).
(45)

In a pseudopotential plane-wave formulation, if the electronic eigenfunctions 𝜓𝑛𝐤(𝐫) are expanded with a cutoff 𝐸kin, the phonon perturbation will be expanded with a cutoff 4𝐸kin (this applies to norm-conserving pseudopotentials; 70 in the case of ultrasoft pseudopotentials, 71 the cutoff of the phonon perturbation would be larger because of the augmentation charge). Therefore, in such cases, it is more convenient to apply the symmetry operation {𝒮𝐯} to the electronic eigenfunctions rather than to the phonon perturbation. A simple change of variables in Eq. (45) gives
在赝势平面波公式中,如果电子本征函数 𝜓𝑛𝐤(𝐫) 通过截止进行扩展 𝐸kin ,声子扰动将随着截止而扩展 4𝐸kin (这适用于范数守恒赝势; 70在超软赝势的情况下, 71由于增强电荷,声子扰动的截止值会更大)。因此,在这种情况下,应用对称运算会更方便 {𝒮𝐯} 是电子本征函数而不是声子扰动。方程式中变量的简单变化。 (45)给出

𝑔𝑚𝑛,𝜈(𝐤,𝒮𝐪)=𝜓𝑚𝐤+𝒮𝐪({𝒮𝐯}𝐫)𝜕𝑉𝐪(𝐫)𝜓𝑛𝐤({𝒮𝐯}𝐫).
(46)

At this point, it may be tempting to apply the same arguments discussed for the dynamical matrix to the electronic Hamiltonian to express 𝐤+𝒮𝐪 in terms of 𝒮1𝐤+𝐪:
此时,可能会很想将针对动力学矩阵讨论的相同论点应用于电子哈密顿量来表达 𝐤+𝒮𝐪 按照 𝒮1𝐤+𝐪 :

𝜓𝑚𝐤+𝒮𝐪({𝒮𝐯}𝐫)=𝜓𝑚𝒮1𝐤+𝐪(𝐫).
(47)

However, Eq. (47) implicitly assumes a specific phase relation between electronic states at 𝐤+𝒮𝐪 and 𝒮1𝐤+𝐪, which, in general, does not hold if the gauge-fixing procedure described in Sec. V C has already been performed. For this reason, in our calculations, we determine the e-ph matrix elements directly through Eq. (45), which goes along with the phonon phase setting given by Eq. (39) but does not require Eq. (47) to be satisfied.
然而,等式。 (47)隐含地假设电子态之间存在特定的相位关系 𝐤+𝒮𝐪𝒮1𝐤+𝐪 ,一般来说,如果第 2 节中描述的仪表固定程序不成立。 VC已经执行了。因此,在我们的计算中,我们直接通过式(1)确定e-ph矩阵元素: (45) ,它与等式给出的声子相位设置一致。 (39)但不需要等式。 (47)感到满意。

Of course, an alternative approach would be to enforce Eq. (47) from the very beginning in the gauge-fixing procedure described in Sec. V C. In both cases, the application of the symmetry operation {𝒮𝐯} to the electronic eigenfunctions is required. The computational cost of this step is negligible with respect to the determination of the dynamical matrix and the phonon perturbations for a given 𝐪 vector.
当然,另一种方法是强制执行等式。 (47)从一开始就在第 2 节中描述的量规固定程序中。 VC 。在这两种情况下,对称运算的应用 {𝒮𝐯} 需要电子本征函数。对于给定的动态矩阵和声子扰动的确定,此步骤的计算成本可以忽略不计 𝐪 向量。

VI. APPLICATION TO BORON-DOPED DIAMOND
六.在掺硼金刚石中的应用

In order to illustrate the scheme developed in Secs. III and IV with a practical calculation, we present here an application to a virtual crystal model of B-doped diamond. We first provide the technical details of the calculation. Then, we discuss the localization properties of the Hamiltonian, the dynamical matrix and the e-ph vertex in the Wannier representation, and the ensuing accuracy of the Fourier interpolation in momentum space. Finally, we present our results for the electron and phonon self-energy arising from the e-ph interaction, for the Eliashberg function, and for the mass enhancement parameter.
为了说明秒中开发的方案。 IIIIV通过实际计算,我们在这里提出了B掺杂金刚石虚拟晶体模型的应用。我们首先提供计算的技术细节。然后,我们讨论了 Wannier 表示中哈密顿量、动力学矩阵和 e-ph 顶点的局域化性质,以及动量空间中傅立叶插值的精度。最后,我们展示了 e-ph 相互作用产生的电子和声子自能、Eliashberg 函数和质量增强参数的结果。

A. Technical details A. 技术细节

Following Refs. 28,35, we consider B-doped diamond with a B content of 1.85%, which is close to the original experimental value. 33 The calculations are performed within the framework of density-functional theory in the local-density approximation.72,73 We account for the core-valence interaction by using norm-conserving pseudopotentials.70,74 Lattice-dynamical properties are calculated within density-functional perturbation theory with the method of Refs. 32,75 and maximally localized Wannier functions obtained by minimizing the Berry-phase spread functional with the method of Refs. 49,53,64. The electronic wave functions are described by a plane-wave basis63,76 with a kinetic energy cutoff of 60Ry, yielding a total-energy accuracy of 10meV/atom. The pseudopotential for the virtual atom B𝑥C1𝑥, with 𝑥=1/54=0.0185, was generated by considering an ionic charge 𝑍ion=3𝑥+4(1𝑥) and the common core electron density of boron and carbon. The fractional occupations are described by first-order Hermite-Gaussian smearing. 77 We checked the convergence of the lattice constant with respect to Brillouin-zone sampling and smearing parameter by considering uniform and unshifted Brillouin-zone grids with up to 203 points and a broadening as small as 0.001Ry. The converged lattice parameter was 6.6425bohr (3.515Å). The use of less stringent convergence parameters ( 123 𝐤 points and 0.05Ry smearing) led to a lattice constant differing by less than 0.002% from the fully converged value. The calculated parameter underestimates the experimental lattice constant of 3.576Å (Ref. 33)by 1.7%, consistent with the common trend observed within the local-density approximation. Since the B doping leads to the formation of a small Fermi surface centered at 𝛤 (cf. Fig. 3), the zone-center phonons are the most sensitive to the metallic character of this system. Accordingly, we tested the convergence of the calculated vibrational frequencies for the highest optical modes at 𝛤. The converged frequency for a smearing of 0.02Ry and a 𝐤 point mesh of size 243 was 138.7meV. A smearing of 0.05Ry together with a mesh of 123 points yielded the very similar frequency of 139.3meV. For convenience, in the following, we describe results obtained within the latter settings.
以下参考文献。如图 28、 35所示,我们考虑 B 掺杂金刚石,其 B 含量为 1.85%,与原始实验值接近。 33计算是在局部密度近似的密度泛函理论框架内进行的。 72, 73我们通过使用范数守恒赝势来解释核价相互作用。 70, 74晶格动力学性质是用参考文献的方法在密度泛函微扰理论中计算的。 32、75和通过使用Refs的方法最小化 Berry 相扩展函数获得的最大局部 Wannier 函数。 49、53、64 电子波函数由平面波基63、76描述,动能截断为 60Ry ,总能量精度为 10meV /原子。虚拟原子的赝势 B𝑥C1𝑥 , 和 𝑥=1/54=0.0185 ,是通过考虑离子电荷而生成的 𝑍ion=3𝑥+4(1𝑥) 以及硼和碳的共同核心电子密度。分数占据通过一阶厄米高斯涂抹来描述。 77我们通过考虑均匀且无位移的布里渊区网格,检查了关于布里渊区采样和涂抹参数的晶格常数的收敛性, 203 点和展宽小至 0.001Ry 。收敛晶格参数为 6.6425bohr (3.515Å) 。 使用不太严格的收敛参数( 123 𝐤 点和 0.05Ry 拖尾)导致晶格常数与完全收敛值相差小于 0.002%。计算出的参数低估了实验晶格常数 3.576Å (参考文献33 )1.7%,与局部密度近似中观察到的共同趋势一致。由于 B 掺杂导致形成以 𝛤 (参见图3 ),区域中心声子对该系统的金属特性最敏感。因此,我们测试了计算出的最高光学模式振动频率的收敛性 𝛤 。涂抹的收敛频率 0.02Ry 和一个 𝐤 点网格尺寸 243 曾是 138.7meV 。涂抹一 0.05Ry 连同网格 123 点产生的频率非常相似 139.3meV 。为了方便起见,下面我们描述在后一种设置中获得的结果。

FIG. 3 如图。 3

(Color online) Comparison between the electronic band structure of pristine diamond (lines) and of B-doped diamond within a virtual crystal approximation (circles). The Fermi level of doped diamond with a B concentration of 1.85% is located 0.57eV below the valence band top at 𝛤 and is indicated by a black solid line.
(在线彩色)原始金刚石(线)和虚拟晶体近似内的 B 掺杂金刚石(圆圈)的电子能带结构之间的比较。 B浓度为1.85%的掺杂金刚石的费米能级位于 0.57eV 低于价带顶部 𝛤 并用黑色实线表示。

Figure 3 shows the calculated band structure of B-doped diamond compared to the band structure of pristine diamond. The calculation for pristine diamond was performed with the same lattice parameter of the doped system for the purpose of comparison (the relaxed lattice parameter of B-doped diamond with a B content of 2% is 0.2% larger than that of intrinsic diamond). After aligning the top of the valence bands, the one-particle eigenvalues corresponding to the occupied subspace are found to differ by 0.1eV at most, indicating that the effect of the virtual pseudopotential could be simulated equally by a simple rigid band model. The Fermi level in of B-doped diamond is located 0.57eV below the top of the valence bands, which is in good agreement with previous theoretical studies. 28
3显示了计算得出的 B 掺杂金刚石的能带结构与原始金刚石的能带结构的比较。为了进行比较,对原始金刚石采用相同的掺杂体系晶格参数进行计算(B含量为2%的掺硼金刚石的弛豫晶格参数比本征金刚石大0.2%)。对齐价带顶部后,发现对应于占据子空间的单粒子特征值有以下差异: 0.1eV 最多,表明虚拟赝势的影响可以通过简单的刚性带模型等效地模拟。 B 掺杂金刚石的费米能级位于 0.57eV 低于价带顶部,这与之前的理论研究非常吻合。 28

Figure 4 shows the calculated phonon dispersions of B-doped diamond, together with the phonon dispersions of pristine diamond with the same lattice parameter. Within the virtual crystal approximation, the doping with boron induces a softening of the optical phonon frequencies around the zone center. The largest softening is observed at the 𝛤 point and amounts to 28meV. This value severely overestimates the experimentally measured softening of 7meV, 78 indicating that a virtual crystal approximation is not sufficient to describe the lattice dynamics of B-doped diamond. 35
4显示了计算得出的 B 掺杂金刚石的声子色散,以及具有相同晶格参数的原始金刚石的声子色散。在虚拟晶体近似中,硼掺杂导致区域中心周围的光学声子频率软化。最大的软化发生在 𝛤 点和相当于 28meV 。该值严重高估了实验测量的软化程度 7meV , 78表明虚拟晶体近似不足以描述 B 掺杂金刚石的晶格动力学。 35

FIG. 4 如图。 4

(Color online) Comparison between the phonon dispersions of pristine diamond (solid lines) and B-doped diamond within a virtual crystal approximation (dashed lines). The disks correspond to inelastic neutron scattering data from Ref. 79.
(在线彩色)虚拟晶体近似内原始金刚石(实线)和 B 掺杂金刚石的声子色散(虚线)之间的比较。这些圆盘对应于参考文献中的非弹性中子散射数据。 79 .

B. Wannier representation and interpolation
B. Wannier 表示和插值

In order to determine the electronic states in the Wannier representation, we need to define an appropriate energy subspace for projecting the electronic Hamiltonian. The identification of this subspace is particularly simple in the present case, since (i) boron doping shifts the Fermi down into the valence bands of diamond. (ii) As discussed in Sec. II, only electronic states close to the Fermi level need to be considered to compute the phonon linewidths, and only electronic states with energy close to the initial state 𝜖𝑛𝐤 are required for the electron linewidths. (iii) The (local-density approximation) indirect band gap of intrinsic diamond is 4.3eV; therefore, the electronic transitions from the valence to the conduction bands do not contribute to the electron and phonon linewidths nor to the superconducting pairing. Following these considerations, we choose to describe the electronic structure of B-doped diamond by considering four bond-centered Wannier functions spanning the four valence bands of diamond. 49 The maximally localized Wannier functions corresponding to this choice exhibit a spatial spread of 0.85Å.
为了确定 Wannier 表示中的电子态,我们需要定义一个适当的能量子空间来投影电子哈密顿量。在本例中,该子空间的识别特别简单,因为 (i) 硼掺杂将费米向下移动到金刚石的价带中。 (ii) 正如第 2 节中所讨论的。 II ,只有接近费米能级的电子态才需要考虑计算声子线宽,并且只有能量接近初始态的电子态 𝜖𝑛𝐤 是电子线宽所需要的。 (iii) 本征金刚石的(局部密度近似)间接带隙为 4.3eV ;因此,从价带到导带的电子跃迁不会影响电子和声子线宽,也不会影响超导配对。根据这些考虑,我们选择通过考虑跨越金刚石四个价带的四个以键为中心的万尼尔函数来描述 B 掺杂金刚石的电子结构。 49与此选择相对应的最大局部 Wannier 函数表现出的空间分布为 0.85Å

Figure 5 shows the spatial decay of the Hamiltonian matrix elements in the Wannier representation 𝐻el𝐑𝑒,𝐑𝑒 as a function of the distance between the unit cells 𝐑𝑒,𝐑𝑒 in which the Wannier functions are located. The Hamiltonian matrix elements are already very small for next-nearest-neighbor Wannier functions (the reduction is approximately a factor of 0.01). This finding is consistent with the proved accuracy of nearest-neighbor tight-binding parametrizations of the electronic structure of carbon-based systems. 80 The exponential decay of the Hamiltonian matrix elements reflects the exponential localization of Wannier functions corresponding to the bonding orbitals of diamond. The present case is particularly favorable since the hole doping preserves much of the band structure of the insulating system: there is no entanglement with higher-energy states, and we effectively deal with a composite group of bands.
5显示了 Wannier 表示中哈密顿矩阵元素的空间衰减 𝐻el𝐑𝑒,𝐑𝑒 作为晶胞之间距离的函数 𝐑𝑒,𝐑𝑒 Wannier 函数位于其中。对于次近邻 Wannier 函数来说,哈密顿矩阵元素已经非常小(减少量约为 0.01 倍)。这一发现与碳基系统电子结构的最近邻紧束缚参数化的已证明的准确性一致。 80哈密顿矩阵元素的指数衰减反映了与金刚石键合轨道相对应的 Wannier 函数的指数局部化。目前的情况特别有利,因为空穴掺杂保留了绝缘系统的大部分能带结构:不存在与高能态的纠缠,并且我们有效地处理了一组复合能带。

FIG. 5 如图。 5

(Color online) Spatial decay of the electronic Hamiltonian in the Wannier representation 𝐻el𝐑𝑒,𝐑𝑒=𝑚𝐑𝑒̂𝐻el𝑛𝐑𝑒 [Eq. (26)] as a function of 𝑅=𝐑𝑒𝐑𝑒. The data points correspond to the largest value taken over the Wannier functions indices and over the unit cells 𝐑𝑒, 𝐑𝑒 located at the same distance 𝑅: 𝐻(𝑅)=max𝑚𝑛,𝐑𝑒𝐑𝑒=𝑅𝑚𝐑𝑒̂𝐻el𝑛𝐑𝑒. The data are normalized to their largest value. The inset shows the same quantity on a logarithmic scale (log10).
(在线彩色)Wannier 表示中电子哈密顿量的空间衰减 𝐻el𝐑𝑒,𝐑𝑒=𝑚𝐑𝑒̂𝐻el𝑛𝐑𝑒 [等式。 (26) ] 作为函数 𝑅=𝐑𝑒𝐑𝑒 。数据点对应于 Wannier 函数指数和晶胞上的最大值 𝐑𝑒 , 𝐑𝑒 位于相同距离 𝑅 : 𝐻(𝑅)=max𝑚𝑛,𝐑𝑒𝐑𝑒=𝑅𝑚𝐑𝑒̂𝐻el𝑛𝐑𝑒 。数据被标准化为其最大值。插图显示了对数刻度上的相同数量 (log10)

Figure 6 shows the spatial decay of the phonon dynamical matrix in the Wannier representation [Eq. (29)]. It is well known that the interatomic force constants of intrinsic diamond exhibit a rather fast spatial decay because of the vanishing Born dynamical charges (the first nonzero contribution is a quadropole-quadrupole interaction). In the case of B-doped diamond considered here, the metallic screening produces a softening of the phonons around the zone center. However, this does not significantly alter the range of the interatomic force constants with respect to intrinsic diamond. Accordingly, for the lattice-dynamical matrix, we also observe exponential localization.
6显示了 Wannier 表示中声子动力学矩阵的空间衰减 [方程 1]。 (29) ]。众所周知,由于玻恩动力电荷的消失(第一个非零贡献是四极-四极相互作用),本征金刚石的原子间力常数表现出相当快的空间衰减。就此处考虑的 B 掺杂金刚石而言,金属屏蔽会软化区域中心周围的声子。然而,这并没有显着改变本征金刚石的原子间力常数的范围。因此,对于晶格动力学矩阵,我们也观察到指数局域化。

FIG. 6 如图。 6

(Color online) Spatial decay of the dynamical matrix in the Wannier representation 𝐷ph𝐑𝑝,𝐑𝑝=𝐑𝑝̂𝐷ph𝐑𝑝 [Eq. (29)] as a function of the distance 𝑅=𝐑𝑝𝐑𝑝. The data points correspond to the largest value taken over the ions in the unit cell, the Cartesian directions, and the unit cells 𝐑𝑝, 𝐑𝑝 located at the same distance 𝑅: 𝐷(𝑅)=max𝜅𝜅𝛼𝛼,𝐑𝑝𝐑𝑝=𝑅𝜅𝛼𝐑𝑝̂𝐷ph𝜅𝛼𝐑𝑝. The data are normalized to their largest value. The inset shows the same quantity on a logarithmic scale (log10).
(在线彩色)Wannier 表示中动力矩阵的空间衰减 𝐷ph𝐑𝑝,𝐑𝑝=𝐑𝑝̂𝐷ph𝐑𝑝 [等式。 (29) ] 作为距离的函数 𝑅=𝐑𝑝𝐑𝑝 。数据点对应于晶胞、笛卡尔方向和晶胞中离子的最大值 𝐑𝑝 , 𝐑𝑝 位于相同距离 𝑅 : 𝐷(𝑅)=max𝜅𝜅𝛼𝛼,𝐑𝑝𝐑𝑝=𝑅𝜅𝛼𝐑𝑝̂𝐷ph𝜅𝛼𝐑𝑝 。数据被标准化为其最大值。插图显示了对数刻度上的相同数量 (log10)

After having examined the spatial decay of the Hamiltonian and the dynamical matrix, we now turn to the e-ph matrix element 𝑔(𝐑𝑒,𝐑𝑝) (Fig. 7). In this case, we have two spatial variables and it is convenient to restrict the discussion to the limiting cases considered in Sec. III C: (i) 𝐑𝑝=𝟎, when the localized phonon perturbation and one Wannier function are located within the same unit cell, and (ii) 𝐑𝑒=𝟎, when the two electron Wannier functions belong to the same unit cell. In the first case, the spatial decay is dictated by the localization of electronic Wannier functions and is expected to be similar to the decay of the Hamiltonian matrix elements in the Wannier representation. Figure 7(a) shows that this is indeed the case since we observe exponential decay in the electronic variable 𝐑𝑒. In the second case, the spatial decay of 𝑔(𝟎𝑒,𝐑𝑝) with 𝐑𝑝 directly reflects the decay of the phonon perturbation in the Wannier representation and is expected to exhibit a localization similar to the dynamical matrix (cf. Sec. III C). Figure 7(b) shows that the e-ph matrix elements decay rather quickly (within next-nearest-neighbor distances), although the decay rate is smaller than for the force constants. This is consistent with the fact that the force constants relate to the gradient of the phonon perturbation [cf. Eq. (19)].
在检查了哈密顿量和动力学矩阵的空间衰减之后,我们现在转向 e-ph 矩阵元素 𝑔(𝐑𝑒,𝐑𝑝) (图7 )。在这种情况下,我们有两个空间变量,并且可以方便地将讨论限制在第 2 节中考虑的限制情况。 IIIC :(一) 𝐑𝑝=𝟎 ,当局域声子扰动和一个 Wannier 函数位于同一晶胞内时,并且 (ii) 𝐑𝑒=𝟎 ,当两个电子 Wannier 函数属于同一晶胞时。在第一种情况下,空间衰减由电子 Wannier 函数的局域化决定,并且预计与 Wannier 表示中哈密顿矩阵元素的衰减类似。图7(a)表明情况确实如此,因为我们观察到电子变量呈指数衰减 𝐑𝑒 。在第二种情况下,空间衰减 𝑔(𝟎𝑒,𝐑𝑝)𝐑𝑝 直接反映了 Wannier 表示中声子扰动的衰减,并且预计会表现出类似于动力学矩阵的局部化(参见第III C节)。图7(b)显示 e-ph 矩阵元素衰减得相当快(在次近邻距离内),尽管衰减率小于力常数的衰减率。这与力常数与声子扰动的梯度相关的事实是一致的[参见。等式。 (19) ]。

FIG. 7 如图。 7

(Color online) Spatial decay of the e-ph vertex in the joint electron-phonon Wannier representation 𝑔𝑚𝑛,𝜈(𝐑𝑒,𝐑𝑝)=𝑚𝟎𝑒𝜕𝜅𝛼,𝐑𝑝𝑉𝑛𝐑𝑒 [Eq. (23)] as a function of 𝐑𝑝 and 𝐑𝑒: (a) the limiting case 𝑔(𝐑𝑒,𝐑𝑝=𝟎) and (b) the limiting case 𝑔(𝐑𝑒=𝟎,𝐑𝑝). The data points correspond to the largest value taken over the Wannier functions indices, the ions in the unit cell, the Cartesian directions, and the unit cells located at the same distance 𝑅=𝐑𝑒 [panel (a)] or 𝑅=𝐑𝑝 [panel (b)] from the origin of the reference frame: (a) 𝑔(𝑅,0)=max𝑚𝑛,𝜅𝛼,𝐑𝑒=𝑅𝑔𝑚𝑛,𝜅𝛼(𝐑𝑒,𝟎𝑝) and (b) 𝑔(0,𝑅)=max𝑚𝑛,𝜅𝛼,𝐑𝑝=𝑅𝑔𝑚𝑛,𝜅𝛼(𝟎𝑒,𝐑𝑝). The data are normalized to their largest value. The insets show the same quantities on a logarithmic scale (log10). When two Wannier functions are located on a C–C bond crossing a cell boundary, identical e-ph matrix elements appear in adjacent unit cells, resulting in the steplike behavior seen in (b).
(在线彩色)电子-声子 Wannier 联合表示中 e-ph 顶点的空间衰减 𝑔𝑚𝑛,𝜈(𝐑𝑒,𝐑𝑝)=𝑚𝟎𝑒𝜕𝜅𝛼,𝐑𝑝𝑉𝑛𝐑𝑒 [等式。 (23) ] 作为函数 𝐑𝑝𝐑𝑒 :(a) 极限情况 𝑔(𝐑𝑒,𝐑𝑝=𝟎) (b) 极限情况 𝑔(𝐑𝑒=𝟎,𝐑𝑝) 。数据点对应于 Wannier 函数指数、晶胞中的离子、笛卡尔方向以及位于相同距离的晶胞的最大值 𝑅=𝐑𝑒 [面板(a)]或 𝑅=𝐑𝑝 [面板(b)]从参考系的原点:(a) 𝑔(𝑅,0)=max𝑚𝑛,𝜅𝛼,𝐑𝑒=𝑅𝑔𝑚𝑛,𝜅𝛼(𝐑𝑒,𝟎𝑝) (b) 𝑔(0,𝑅)=max𝑚𝑛,𝜅𝛼,𝐑𝑝=𝑅𝑔𝑚𝑛,𝜅𝛼(𝟎𝑒,𝐑𝑝) 。数据被标准化为其最大值。插图在对数刻度上显示相同的数量 (log10) 。当两个 Wannier 函数位于穿过晶胞边界的 C-C 键上时,相同的 e-ph 矩阵元素出现在相邻晶胞中,从而导致 (b) 中看到的阶梯行为。

In order to have an idea of the Brillouin-zone sampling required to obtain matrix elements encompassing a given spatial range 𝑅, we notice that in the present case, a uniform 6×6×6 sampling corresponds to interactions extending up to 𝑅10Å, while a grid with 12×12×12 𝐤 points in the Brillouin zone corresponds to a maximum range of 𝑅20Å. Hence, Figs. 5–7 indicate that a Brillouin-zone sampling corresponding to a 6×6×6 supercell should be sufficient to determine the self-energies [Eqs. (3) and (4)] with an accuracy of 1% [without taking into account cancellation of errors arising from the 𝐤 and 𝐪 integrations in Eqs. (3) and (4)]. In the following section, we verify this observation by comparing the e-ph matrix elements obtained by the Wannier-Fourier interpolation method discussed in Sec. IV with those computed directly from first principles.
为了了解获得包含给定空间范围的矩阵元素所需的布里渊区采样 𝑅 ,我们注意到在本例中,制服 6×6×6 采样对应于扩展至的交互作用 𝑅10Å ,而网格与 12×12×12 𝐤 布里渊区中的点对应的最大范围 𝑅20Å 。因此,无花果。 5-7表明布里渊区采样对应于 6×6×6 超晶胞应足以确定自能[方程1]。 (3)(4) ] 精度为 1% [不考虑取消由 𝐤𝐪 方程中的积分。 (3)(4) ]。在下面的部分中,我们通过比较第 2 节中讨论的 Wannier-Fourier 插值方法获得的 e-ph 矩阵元素来验证这一观察结果。 IV与那些直接根据第一原理计算的结果。

1. Accuracy of the electron-phonon matrix elements
1. 电子声子矩阵元的精度

In order to assess the accuracy of the interpolation method introduced in Sec. IV, we need to compare the various quantities needed for studying the e-ph interaction (single-particle electronic eigenvalues, vibrational frequencies, and e-ph matrix elements) obtained by Wannier-Fourier interpolation with those obtained directly from first principles. The interpolation procedures for the band structure and for the phonon dispersions have already been addressed elsewhere;32,53,58 therefore, we restrict ourselves here to the e-ph matrix elements.
为了评估第 2 节中介绍的插值方法的准确性。第四,我们需要将通过Wannier-Fourier插值获得的研究e-ph相互作用所需的各种量(单粒子电子特征值、振动频率和e-ph矩阵元素)与直接从第一原理获得的量进行比较。能带结构和声子色散的插值程序已经在其他地方讨论过; 32, 53, 58因此,我们在这里将自己限制在 e-ph 矩阵元素上。

The e-ph vertex 𝑔𝑚𝑛,𝜈(𝐤,𝐪) is a ten-dimensional object, and it is hard to present a comprehensive visual comparison of the ab initio and the interpolated matrix elements. We focus here on a special case which is representative of the general trend. For this purpose, we take the initial electronic state 𝑛𝐤 to be the 𝛤25 state (top of the valence band at the zone center), and we vary the phonon momentum 𝐪 along the same high-symmetry lines 𝛬, 𝛥, and 𝛴 considered in Figs. 3 and 4. For each phonon momentum, the final electronic state 𝑚𝐤+𝐪 is taken on the top of the valence manifold through the twofold degenerate bands 𝛬3 and 𝛥5, as well as the nondegenerate 𝛴2 band [Fig. 8(a), dashed line]. The emitted and/or absorbed phonon is taken to be the highest optical mode at the given momentum 𝐪 along the same symmetry lines, with the exception of the 𝛴 line where we pick the 𝛴1 branch rather than the highest-energy 𝛴3 branch [Fig. 8(b), dashed line]. These choices were made in order to avoid symmetry-forbidden transitions which are uninteresting for our comparison. As discussed is Sec. V C, the e-ph matrix element is not gauge invariant; therefore, the matrix elements corresponding to degenerate electronic or vibrational states do not carry any physical meaning by themselves. However, for illustration purposes, we adopt here the convention that every individual (squared) matrix element corresponds to the average within the eventual degenerate manifold. This convention leads to a discontinuity of the e-ph matrix elements at the zone center which can be observed in Fig. 8(c).
e-ph 顶点 𝑔𝑚𝑛,𝜈(𝐤,𝐪) 是一个十维对象,很难呈现从头算和插值矩阵元素的全面视觉比较。我们在这里关注一个代表总体趋势的特殊案例。为此,我们取初始电子态 𝑛𝐤 成为 𝛤25 态(区域中心价带顶部),我们改变声子动量 𝐪 沿着相同的高对称线 𝛬 , 𝛥 , 和 𝛴 考虑在无花果。 34 。对于每个声子动量,最终电子态 𝑚𝐤+𝐪 通过双重简并带在价流形的顶部获取 𝛬3𝛥5 ,以及非简并 𝛴2 乐队[图. 8(a) ,虚线]。发射和/或吸收的声子被视为给定动量下的最高光学模式 𝐪 沿着相同的对称线,除了 𝛴 我们选择的线 𝛴1 分支而不是最高能量 𝛴3 分支[图. 8(b) ,虚线]。做出这些选择是为了避免对称禁止跃迁,这对我们的比较来说是无趣的。正如所讨论的那样。 VC ,e-ph 矩阵元素不是规范不变的;因此,对应于简并电子或振动状态的矩阵元素本身不具有任何物理意义。 然而,出于说明目的,我们在这里采用这样的约定:每个单独的(平方)矩阵元素对应于最终简并流形内的平均值。这种约定导致区域中心处的 e-ph 矩阵元素不连续,这可以在图8(c)中观察到。

FIG. 8 如图。 8

(Color online) Comparison of the e-ph matrix elements 𝑔SE𝑚𝑛,𝜈(𝐤,𝐪) [Eq. (5)] obtained by direct first-principles calculations [panel (c), disks] and those computed with the Wannier-Fourier interpolation method discussed in Sec. IV [panel (c), lines]. The interpolated matrix elements are computed starting from an initial 43 Brillouin-zone grid (dotted line), a 63 grid (dashed line), or a 83 grid (solid line). For illustration, we fixed the initial electronic state 𝑛𝐤 for the valence band top at 𝛤 (𝐤=𝟎); we let the final electronic state 𝑚𝐤+𝐪 span the 𝛬3, 𝛥5, and 𝛴2 bands as shown in panel (a) (dashed line), and we take the phonon perturbation corresponding to the highest optical branches as shown in panel (b) (dashed line).
(在线彩)e-ph矩阵元素比较 𝑔SE𝑚𝑛,𝜈(𝐤,𝐪) [等式。 (5) ]通过直接第一原理计算获得[面板(c),圆盘]以及使用第2节中讨论的Wannier-Fourier插值方法计算的结果。 IV [面板 (c),线条]。插值矩阵元素是从初始值开始计算的 43 布里渊区网格(虚线),a 63 网格(虚线),或 83 网格(实线)。为了便于说明,我们固定了初始电子态 𝑛𝐤 对于价带顶部 𝛤 (𝐤=𝟎) ;我们让最终的电子态 𝑚𝐤+𝐪 跨越 𝛬3 , 𝛥5 , 和 𝛴2 频带如面板(a)(虚线)所示,我们采用与最高光分支相对应的声子扰动,如面板(b)(虚线)所示。

Figure 8(c) shows the variation of the e-ph matrix element along the described energy and/or momentum path, as computed directly from first principles, together with the values obtained by the joint electron-phonon interpolation procedure outlined in Sec. IV. We considered 50 phonon momenta in the ab initio calculation and unshifted Brillouin-zone grids with 43, 63, or 83 points for the interpolation procedure (we generated 500 phonon momenta on the high-symmetry lines considered in Fig. 8). The calculated e-ph matrix elements are consistent with the electron-phonon potential 𝑉ep=280meV obtained in Ref. 28, as well as the average matrix element 𝑔21/2=670meV estimated in Ref. 36 by deformation potential calculations. However, contrary to previous assumptions,28,36 the e-ph vertex varies significantly throughout the Brillouin zone, ranging from 0to500meV.
8(c)显示了 e-ph 矩阵元素沿所描述的能量和/或动量路径的变化,直接根据第一原理计算,以及通过第 2 节中概述的联合电子声子插值过程获得的值。.我们在从头计算中考虑了 50 个声子动量和未移动的布里渊区网格 43 , 63 , 或者 83 插值过程的点(我们在图8中考虑的高对称线上生成了 500 个声子动量)。计算出的e-ph矩阵元素与电子声子势一致 𝑉ep=280meV 在参考文献中获得。 28 ,以及平均矩阵元素 𝑔21/2=670meV 参考文献中估计。 36通过变形潜力计算。然而,与之前的假设相反, 28, 36 e-ph顶点在整个布里渊区变化显着,范围从 0to500meV

It is clear that our interpolation scheme is very effective, and already a 6×6×6 grid provides a very accurate description of the electron-phonon interaction in the example considered. We notice that the initial dynamical matrices and phonon perturbations for 6×6×6 phonon momenta are obtained from the irreducible wedge of the Brillouin zone and therefore correspond to only 16 separate calculations.
很明显,我们的插值方案非常有效,并且已经 6×6×6 网格提供了所考虑示例中电子-声子相互作用的非常准确的描述。我们注意到初始动力学矩阵和声子扰动 6×6×6 声子动量是从布里渊区的不可约楔形中获得的,因此仅对应于 16 个单独的计算。

For a quantitative assessment of the accuracy of our method, we report here the absolute deviations of the interpolated e-ph matrix elements with respect to the first-principles calculations on a uniform Brillouin-zone grid with 103 points. We found the largest deviations of 80, 30, and 15meV for initial Brillouin-zone grids containing 43, 63, and 83 points, respectively. The corresponding deviations for the interpolated electronic eigenvalues were 0.5, 0.1, and 0.05eV, respectively, while the deviations of the phonon frequencies were 10, 2, and 1meV, respectively.
为了定量评估我们方法的准确性,我们在此报告插值 e-ph 矩阵元素相对于均匀布里渊区网格上的第一性原理计算的绝对偏差 103 点。我们发现最大偏差为 80、30 和 15meV 对于初始布里渊区网格包含 43 , 63 , 和 83 点,分别。插值电子特征值的相应偏差为 0.5、0.1 和 0.05eV ,而声子频率的偏差分别为 10、2 和 1meV , 分别。

At the end of this section, it is worth pointing out that, compared to other possible interpolation schemes, the one discussed in the present work relies on a physical property of the system, which could be designated as the “near-sightedness” of the electron-phonon interaction, in analogy with a very general concept introduced for the electron-electron interaction. 81 In favorable cases (such as the application discussed here), our scheme shows exponentially increasing accuracy with the spacing of the coarse grid on which the first-principles calculations are performed.
在本节的最后,值得指出的是,与其他可能的插值方案相比,本工作中讨论的插值方案依赖于系统的物理属性,可以将其称为系统的“近视性”电子-声子相互作用,类似于电子-电子相互作用引入的非常普遍的概念。 81在有利的情况下(例如这里讨论的应用程序),我们的方案显示,随着执行第一性原理计算的粗网格的间距,精度呈指数级增长

C. Electron and phonon linewidths, Eliashberg function, and mass enhancement parameter
C. 电子和声子线宽、Eliashberg 函数和质量增强参数

Once the accuracy of the Wannier-Fourier interpolation is established, we proceed to investigate the convergence of the electron and phonon linewidths [Eqs. (3) and (4)] with the sampling of the Brillouin zone. All the calculations described in this section were performed by interpolating the electron Hamiltonian, the lattice-dynamical matrix, and the e-ph vertex evaluated on the unshifted 83 grid discussed in Sec. VI B. The Dirac delta functions in Eqs. (3) and (4) were replaced by Lorentzian distributions. The Fermi-Dirac and Bose-Einstein occupations in Eqs. (3) and (4) were calculated with the temperature set to 300K.
一旦确定了 Wannier-Fourier 插值的精度,我们就继续研究电子和声子线宽的收敛性 [方程 1]。 (3)(4) ]与布里渊区采样。本节中描述的所有计算都是通过对电子哈密顿量、晶格动力学矩阵和在未平移的上评估的 e-ph 顶点进行插值来执行的。 83 网格在第二节中讨论。六 B.方程中的狄拉克δ函数。 (3)(4)被洛伦兹分布取代。方程中的费米-狄拉克和玻色-爱因斯坦占领。 (3)(4)是在温度设定为 300K

1. Electron linewidths 1. 电子线宽

Figure 9 shows the calculated electron linewidths arising from the e-ph interaction for the electronic states indicated in Fig. 8 ( 𝛬3 and 𝛥5 bands). The linewidths corresponding to the other bands close to the Fermi level have qualitatively similar behavior. The integration over the phonon momentum in Eq. (3) is performed by interpolating the matrix elements on two sets of Brillouin-zone meshes: a coarse grid including 10×10×10 points in the irreducible wedge, obtained by randomly shifting a uniform grid, as well as a fine grid including 50×50×50 points in the irreducible wedge. For each Brillouin-zone mesh, we repeated the calculations by setting the Lorentzian half-width to 10, 50, and 100meV, respectively. Figure 9 shows that 1000 𝐪 points are not sufficient to perform the momentum integration. The use of a small smearing parameter [ 10meV, Fig. 11(a)] leads to strong fluctuations of the linewidths, making it difficult to identify a clear trend. On the other hand, a large smearing [ 100meV Fig. 11(c)] increases the linewidths close to the zone center, leading to unphysical results (vide infra). The calculations performed with 125 000 𝐪 points in the momentum integration is found to produce reasonably good results for the smallest smearing considered of 10meV [Fig. 11(c)], although small unphysical fluctuations still persist.
9显示了由图8所示电子态的 e-ph 相互作用产生的计算电子线宽( 𝛬3𝛥5 乐队)。与接近费米能级的其他能带相对应的线宽具有性质相似的行为。方程中声子动量的积分。 (3)是通过在两组布里渊区网格上插值矩阵元素来执行的:粗网格包括 10×10×10 不可约楔中的点,通过随机移动均匀网格以及精细网格获得,包括 50×50×50 不可约楔中的点。对于每个布里渊区网格,我们通过将洛伦兹半宽设置为 10、50 和 100meV , 分别。图9显示 1000 𝐪 点不足以执行动量积分。使用小的涂抹参数[ 10meV ,图11(a) ]导致线宽的强烈波动,使得难以识别清晰的趋势。另一方面,大涂抹[ 100meV11(c) ]增加了靠近区域中心的线宽,导致非物理结果(见下文)。计算结果为 125 000 𝐪 发现动量积分中的点对于所考虑的最小涂抹产生相当好的结果 10meV [如图。 11(c) ],尽管小的非物理波动仍然存在。

FIG. 9 如图。 9

(Color online) Calculated electron linewidths for the 𝛬3 and 𝛥5 bands of B-doped diamond [Fig. 8(a)]. Plots (a)–(c) on the left were obtained by using 103 irreducible 𝐪 points in the momentum integration of Eq. (3), while plots (d)–(f) on the right were obtained with 503 irreducible points in the Brillouin zone. We report the results for three broadening parameters 𝜂: 10meV [panels (a) and (d)], 50meV [panels (b) and (e)], and 100meV [panels (c) and (f)]. The curves are cut off at half the Brillouin-zone size for clarity.
(在线彩色)计算出的电子线宽 𝛬3𝛥5 B 掺杂金刚石的能带 [图 1] 8(a) ]。左边的图 (a)–(c) 是通过使用获得的 103 不可约的 𝐪 方程动量积分中的点(3) ,而右侧的图 (d)–(f) 是通过以下方法获得的 503 布里渊区的不可约点。我们报告三个展宽参数的结果 𝜂 : 10meV [(a)和(d)图], 50meV [(b)和(e)组],以及 100meV [面板(c)和(f)]。为了清晰起见,曲线在布里渊区大小的一半处被截断。

It is worth mentioning that most current calculations of the e-ph interaction are performed with grids including considerably fewer irreducible phonon momenta, since the direct computation of the lattice-dynamical matrix corresponds to several total-energy minimizations for each 𝐪 point. In practice, without our Wannier-Fourier technique, the calculations described here would require several months of computation on modern computers. 82
值得一提的是,当前大多数 e-ph 相互作用的计算都是使用包含相当少的不可约声子动量的网格来执行的,因为晶格动力学矩阵的直接计算对应于每个的几个总能量最小化。 𝐪 观点。实际上,如果没有我们的万尼尔-傅里叶技术,这里描述的计算将需要在现代计算机上进行几个月的计算。 82

The calculated electron linewidths show a peculiar suppression when the electron momentum lies close to the zone center. Careful analysis indicates that this happens for all electronic states with energy 𝜖𝑛𝐤 within 𝜔op from the valence band top at 𝛤, where 𝜔op160meV is a characteristic optical phonon frequency. In this case, the electron cannot emit a phonon since there are no available final states at the energy 𝜖𝑛𝐤+𝜔op [the final state falls within the band gap, cf. Eq. (3)]. Within this energy range, it is still possible for an electron to emit an acoustic phonon; however, the matrix element for this process is practically negligible. For electronic states far from the zone center, we observe a monotonic increase of the linewidth with the electron momentum. This behavior can be understood by considering Eq. (3). If we replace the e-ph matrix element by its average value throughout the Brillouin zone and remember that we are considering occupied states, the imaginary part of the self-energy becomes 𝛴𝑛𝐤𝜋𝛺1BZ𝑔SE2𝑚𝜈BZ𝑑𝐪𝛿(𝜖𝑛𝐤𝜔𝐪𝜈𝜖𝑚𝐤+𝐪). By neglecting the phonon frequency in the delta function, we obtain the electronic density of states at the energy of the initial state 𝑁(𝜖𝑛𝐤). Therefore, when we move off the 𝛤 point, the linewidth increases following the density of states. We have checked numerically that the electron linewidth scales as 𝛴(𝜔)𝜔1/2, which is consistent with the underlying density of states of the parabolic bands of diamond. We notice that this behavior is at variance with common models of the electron self-energy arising from the e-ph interaction. 38 The latter models are based on the assumption of a constant density of states around the Fermi level, which does not hold in the present case. The calculated linewidths for electronic states close to the Fermi level (50meV) are smaller than the values of 300meV that can be estimated from the photoemission data of Ref. 83. The underestimation of the experimental widths is consistent with the fact that our theory does not take into account the broadening induced by electron-electron and the electron-dopant interactions.
当电子动量靠近区域中心时,计算出的电子线宽显示出特殊的抑制。仔细分析表明,所有具有能量的电子态都会发生这种情况 𝜖𝑛𝐤 之内 𝜔op 从价带顶部 𝛤 , 在哪里 𝜔op160meV 是特征光学声子频率。在这种情况下,电子不能发射声子,因为在能量处没有可用的最终状态 𝜖𝑛𝐤+𝜔op [最终状态落在带隙内,参见。等式。 (3) ]。在这个能量范围内,电子仍然有可能发射声学声子;然而,该过程的矩阵元素实际上可以忽略不计。对于远离区域中心的电子态,我们观察到线宽随电子动量单调增加。这种行为可以通过考虑等式来理解。 (3) .如果我们将 e-ph 矩阵元素替换为其整个布里渊区的平均值,并记住我们正在考虑占据状态,则自能的虚部变为 𝛴𝑛𝐤𝜋𝛺1BZ𝑔SE2𝑚𝜈BZ𝑑𝐪𝛿(𝜖𝑛𝐤𝜔𝐪𝜈𝜖𝑚𝐤+𝐪) 。通过忽略δ函数中的声子频率,我们得到了初始态能量下的电子态密度 𝑁(𝜖𝑛𝐤) 。因此,当我们离开 𝛤 点,线宽随着态密度的增加而增加。我们已经从数字上检查了电子线宽的比例为 𝛴(𝜔)𝜔1/2 ,这与金刚石抛物线带的基本态密度一致。我们注意到这种行为与 e-ph 相互作用产生的电子自能的常见模型不一致。38后一个模型基于费米能级周围态密度恒定的假设,但在当前情况下并不成立。接近费米能级的电子态的计算线宽 (50meV) 小于的值 300meV 可以从参考文献的光电发射数据估计。 83 .实验宽度的低估与我们的理论没有考虑电子-电子和电子-掺杂剂相互作用引起的展宽这一事实是一致的。

2. Phonon linewidths 2. 声子线宽

Figure 10 shows the phonon linewidths corresponding to the longitudinal optical mode of B-doped diamond [cf. 8(b)] calculated with different Brillouin-zone grids and smearing parameters. The transverse optical modes behave similarly, while the linewidths associated with the acoustic modes were found to be negligibly small (less than 0.5meV throughout the entire Brillouin zone). We find that 1000 𝐪 points in the irreducible zone are not sufficient to reproduce the correct momentum dependence of the phonon linewidths. On the other hand, a Brillouin-zone mesh with 125 000 𝐪 points yields reasonably good results, although small unphysical fluctuations can still be seen in the plots. Two features stand out in the plots of Fig. 10: (i) the linewidths become negligible as we move far enough from the zone center, and (ii) there is singular behavior at the zone-center where we observe a dip instead of a peak. Feature (i) is associated with the fact that phonons with momentum larger than the average Fermi-surface diameter (𝑞>2𝑘F) cannot be scattered since the initial and final electronic states are pinned near to the Fermi surface [cf. Eq. (4)]. 35 Feature (ii) is more subtle and arises from the fact that electronic transitions with no momentum transfer (𝐪=𝟎) are essentially forbidden. When the energy separation between initial states on the Fermi surface and final states with the same momentum 𝐤 exceeds the largest phonon frequency 𝜔op, the transition is blocked by the energy selection rule in Eq. (4).35,46 Our linewidths are consistent with those calculated in Ref. 84. However, the characteristic dip at the zone center found here is missing in Ref. 84 because of the common although unjustified approximation of neglecting of the phonon energy in Eq. (4).
10显示了与 B 掺杂金刚石的纵向光学模式相对应的声子线宽 [cf. 8(b)]用不同的布里渊区网格和涂抹参数计算。横向光学模式的行为类似,而与声学模式相关的线宽被发现小得可以忽略不计(小于 0.5meV 整个布里渊区)。我们发现 1000 𝐪 不可约区中的点不足以再现声子线宽的正确动量依赖性。另一方面,布里渊区网格有 125 000 𝐪 尽管在图中仍然可以看到小的非物理波动,但点产生了相当好的结果。图10的图中有两个特征很突出:(i) 当我们距离区域中心足够远时,线宽变得可以忽略不计;(ii) 在区域中心处存在奇异行为,我们观察到的是倾斜而不是倾斜。顶峰。特征 (i) 与动量大于平均费米面直径的声子相关 (𝑞>2𝑘F) 由于初始和最终的电子态被钉扎在费米表面附近,因此不能被散射[cf.等式。 (4) ]。 35特征 (ii) 更为微妙,源于以下事实:没有动量转移的电子跃迁 (𝐪=𝟎) 基本上是被禁止的。当费米面上初态与终态的能量分离具有相同动量时 𝐤 超过最大声子频率 𝜔op ,跃迁被方程式中的能量选择规则阻止。 (4) .35, 46我们的线宽与参考文献中计算的一致。 84 .然而,此处发现的区域中心的特征倾角在参考文献中缺失。 84因为在方程中忽略声子能量的常见但不合理的近似。 (4) .

FIG. 10 如图。 10

(Color online) Calculated phonon linewidths for the highest optical mode of B-doped diamond [Fig. 8(b)]. Plots (a)–(c) on the left were obtained with 103 irreducible 𝐪 points in the momentum integration of Eq. (4), while plots (d)–(f) on the right were obtained with 503 irreducible points. The results for three broadening parameters 10, 50, and 100meV are shown in panels (a) and (d), panels (b) and (e), and panels (c) and (f), respectively. Note the different vertical scales in the plots corresponding to different broadening parameters.
(在线彩色)B 掺杂金刚石最高光学模式的计算声子线宽 [图 1] 8(b) ]。左侧的图 (a)–(c) 是通过以下方法获得的 103 不可约的 𝐪 方程动量积分中的点(4) ,而右侧的图 (d)–(f) 是通过以下方法获得的 503 不可约点。三个展宽参数 10、50 和 100meV 分别显示在面板(a)和(d)、面板(b)和(e)以及面板(c)和(f)中。请注意图中对应于不同加宽参数的不同垂直比例。

In Fig. 10, the linewidths close to the zone center are consistent with the values of 8meV determined by inelastic x-ray scattering on B-doped diamond samples with a similar doping level. 78 We note, however, that in the present work, we do not take into account the local structural relaxation induced by the B atoms. When this effect is included, finite phonon linewidths are observed even at large momenta (𝑞>2𝑘F) due to the breaking of the lattice periodicity. 35
10中,靠近区域中心的线宽与 8meV 通过对具有相似掺杂水平的 B 掺杂金刚石样品进行非弹性 X 射线散射来确定。 78然而,我们注意到,在目前的工作中,我们没有考虑 B 原子引起的局部结构弛豫。当包括这种效应时,即使在大动量下也可以观察到有限的声子线宽 (𝑞>2𝑘F) 由于晶格周期性的破坏。 35

3. Eliashberg function and mass enhancement parameter
3. Eliashberg函数和质量增强参数

Figure 11 shows the Eliashberg function [Eq. (7)] obtained with the phonon linewidths discussed in Sec. VI C 2. As already pointed out, 1000 𝐤 and 𝐪 irreducible points in the Brillouin zone are not sufficient to achieve convergence, while a grid with 125 000 points leads to stable results. The reliability of the calculated Eliashberg function is important in the study of phonon-mediated superconductivity, since 𝛼2𝐹(𝜔) is commonly used to identify the phonon modes responsible for the pairing. Figure 11 shows that, within the virtual crystal approximation, only the high-frequency optical modes of B-doped diamond participate in the superconducting pairing, while the acoustic modes play a minor role. When the dopant atoms are taken explicitly into account,28,35 the picture derived from the virtual crystal model needs to be revised; the dominant contribution to the pairing field is found to arise from the vibrational modes associated with the impurity.28,35
11显示了 Eliashberg 函数 [Eq. (7) ] 用第 2 节中讨论的声子线宽获得。六、C 2 .正如已经指出的,1000 𝐤𝐪 布里渊区的不可约点不足以实现收敛,而具有 125 000 点的网格会产生稳定的结果。计算出的 Eliashberg 函数的可靠性在声子介导的超导性研究中非常重要,因为 𝛼2𝐹(𝜔) 通常用于识别负责配对的声子模式。图11显示,在虚拟晶体近似中,只有 B 掺杂金刚石的高频光学模式参与超导配对,而声学模式发挥次要作用。当明确考虑掺杂剂原子时, 28, 35从虚拟晶体模型导出的图像需要修改;发现配对场的主要贡献来自与杂质相关的振动模式。 28, 35

FIG. 11 如图。 11

(Color online) Eliashberg function 𝛼2𝐹(𝜔) calculated for B-doped diamond in the virtual crystal approximation with (a) 103𝐤 and 𝐪 points in the irreducible part of the Brillouin zone and (b) 503 points. For each case, we report the results corresponding to the smearing parameters 10meV (dotted lines), 50meV (dashed lines), and 100meV (solid lines). The Dirac delta function in Eq. (7) was replaced by a Gaussian with a standard deviation of 0.5meV. Note the different energy scales and scaling factors for the acoustic and the optical frequency ranges.
(在线彩图)Eliashberg函数 𝛼2𝐹(𝜔) 使用 (a) 在虚拟晶体近似中计算 B 掺杂金刚石 103𝐤𝐪 布里渊区不可约部分的点和 (b) 503 点。对于每种情况,我们报告与涂抹参数相对应的结果 10meV (虚线), 50meV (虚线),以及 100meV (实线)。方程中的狄拉克δ函数。 (7)被替换为高斯分布,其标准差为 0.5meV 。注意声学和光学频率范围的不同能量尺度和比例因子。

The mass enhancement parameter 𝜆 is obtained from the Eliashberg function through Eq. (8). This parameter is commonly used to estimate the superconducting transition temperature of isotropic superconductors in conjunction with the semiempirical McMillan formula 𝑇𝑐=𝜔log1.2exp{1.04(1+𝜆)/[𝜆𝜇(10.62𝜆)]}, where 𝜔log is the logarithmic averaged phonon frequency and 𝜇 is the Coulomb pseudopotential. 85 Since the superconducting transition temperature exhibits an exponential dependence on 𝜆 (at least for 𝜆1.25), it is essential to obtain accurate values for the coupling strength. A survey of the existing literature reveals that calculated values of 𝜆 for B-doped diamond vary by more than a factor of 2, ranging from 𝜆=0.24 to 𝜆=0.53.28,36,37,84,86 The superconducting transition temperature 𝑇𝑐 corresponding to this range (assuming, for simplicity, the same Coulomb pseudopotential 𝜇=0.13 and the logarithmic frequency 𝜔log=1000cm1) spans several orders of magnitude, from 103to14K, clearly pointing to a serious difficulty in the calculation of e-ph interaction properties.
质量增强参数 𝜆 由 Eliashberg 函数通过等式获得: (8) .该参数通常与半经验麦克米伦公式结合来估计各向同性超导体的超导转变温度 𝑇𝑐=𝜔log1.2exp{1.04(1+𝜆)/[𝜆𝜇(10.62𝜆)]} , 在哪里 𝜔log 是对数平均声子频率, 𝜇 是库仑赝势。 85由于超导转变温度表现出指数依赖性 𝜆 (至少对于 𝜆1.25 ),获得耦合强度的准确值至关重要。对现有文献的调查表明,计算值 𝜆 对于 B 掺杂金刚石,变化超过 2 倍,范围为 𝜆=0.24𝜆=0.5328, 36, 37, 84, 86超导转变温度 𝑇𝑐 对应于这个范围(为简单起见,假设相同的库仑赝势 𝜇=0.13 和对数频率 𝜔log=1000cm1 )跨越几个数量级,从 103to14K ,清楚地指出了 e-ph 相互作用属性计算中的严重困难。

Table II reports our calculated e-ph coupling strengths corresponding to the Eliashberg functions shown in Fig. 11. We find that a grid of 1000 points in the irreducible wedge of the Brillouin zone yields 𝜆 values which are very sensitive to the choice of the smearing parameter. A grid with 125 000 points provides instead results which are reasonably insensitive to this choice. Our fully converged value for the e-ph coupling strength is 𝜆=0.237. This value was obtained in Ref. 35 using 1003 𝐤 points and 303 𝐪 points in the irreducible wedge of the Brillouin zone. The corresponding transition temperature is 5×104K (with a Coulomb pseudopotential 𝜇=0.13 and a logarithmic frequency 𝜔log=1010cm1) and is in sharp contrast with the experimentally observed 𝑇𝑐 of 4K. 33 As discussed in Ref. 35, the failure of the virtual crystal model of B-doped diamond is to be ascribed to the neglect of the vibrational modes associated with the boron atoms.
II报告了我们计算出的与图11中所示的 Eliashberg 函数相对应的 e-ph 耦合强度。我们发现布里渊区不可约楔形中的 1000 个点的网格产生 𝜆 这些值对涂抹参数的选择非常敏感。相反,具有 125 000 个点的网格提供的结果对此选择相当不敏感。我们的 e-ph 耦合强度的完全收敛值为 𝜆=0.237 。该值是在参考文献中获得的。 35使用 1003 𝐤 点和 303 𝐪 布里渊区不可约楔内的点。相应的转变温度为 5×104K (库仑赝势 𝜇=0.13 和对数频率 𝜔log=1010cm1 )与实验观察到的形成鲜明对比 𝑇𝑐4K33正如参考文献中所讨论的。如图35所示,B掺杂金刚石的虚拟晶体模型的失败归因于忽略了与硼原子相关的振动模式。

TABLE II. 表二。

Electron-phonon mass enhancement parameter 𝜆 of B-doped diamond in the virtual crystal approximation, calculated with different Brillouin-zone grids and smearing parameters.
电子声子质量增强参数 𝜆 使用不同布里渊区网格和涂抹参数计算的虚拟晶体近似中的 B 掺杂金刚石。

10meV 50meV 100meV
10×10×10 0.073 0.156 0.212
50×50×50 0.232 0.219 0.210
VII. CONCLUSION 七.结论

The present work was motivated by the long-standing difficulty of studying the electron-phonon interaction from first principles. This difficulty arises from the necessity of a very careful description of the e-ph scattering processes in the Brillouin zone, in particular, in proximity of the Fermi surface. We have shown that this difficulty can be overcome by performing a generalized Wannier-Fourier interpolation of the e-ph vertex, leading to results as accurate as a full ab initio calculation but at a comparably negligible computational cost. In order to assess the accuracy of our methodology, we have performed a comprehensive set of tests on a virtual crystal model of B-doped diamond. In particular, we calculated the electron and phonon linewidths arising from the e-ph interaction as well as the Eliashberg spectral function, and we discussed the dependence of these quantities on the sampling of the Brillouin zone. Our study revealed that, contrary to common assumptions, the momentum dependence of the e-ph matrix element is significant throughout the Brillouin zone and cannot be neglected.
目前的工作是由从第一原理研究电子-声子相互作用的长期困难推动的。这一困难源于需要非常仔细地描述布里渊区(特别是费米面附近)的 e-ph 散射过程。我们已经证明,可以通过对 e-ph 顶点执行广义 Wannier-Fourier 插值来克服这一困难,从而获得与完整的从头计算一样准确的结果,但计算成本相对可忽略不计。为了评估我们方法的准确性,我们对 B 掺杂金刚石的虚拟晶体模型进行了一套全面的测试。特别是,我们计算了由 e-ph 相互作用以及 Eliashberg 谱函数产生的电子和声子线宽,并讨论了这些量对布里渊区采样的依赖性。我们的研究表明,与常见的假设相反,e-ph 矩阵元素的动量依赖性在整个布里渊区都很显着,不能被忽视。

It is interesting to consider future directions and possible developments of the present theory. The first obvious application of our method consists in using the e-ph matrix elements in the Wannier representation to investigate how electrons and ions interact at the atomistic scale. Indeed, we could decompose the various contributions to the electron and phonon self-energies into their atomistic components in the spirit of the analysis of Ref. 56.
考虑当前理论的未来方向和可能的发展是很有趣的。我们的方法的第一个明显应用在于使用 Wannier 表示中的 e-ph 矩阵元素来研究电子和离子如何在原子尺度上相互作用。事实上,我们可以本着参考文献 2 的分析精神,将对电子和声子自能的各种贡献分解为其原子成分。 56 .

A further step in the same direction could be taken by reformulating the electron-phonon problem in a fully localized representation based on Wannier functions. This possibility is appealing since, as we have shown in this work, all the information needed to describe the e-ph interaction is encoded in a small number of matrix elements in the Wannier representation. Therefore, in principle, there is no need to go back to Bloch space by Wannier-Fourier interpolation. It is interesting to note that a similar idea has been suggested in Ref. 87, where the authors were interested in reformulating the Eliashberg equations in a localized Wannier representation.
通过在基于 Wannier 函数的完全局部化表示中重新表述电子声子问题,可以朝同一方向迈出进一步的一步。这种可能性很有吸引力,因为正如我们在这项工作中所展示的,描述 e-ph 相互作用所需的所有信息都被编码在 Wannier 表示中的少量矩阵元素中。因此,原则上不需要通过Wannier-Fourier插值回到Bloch空间。有趣的是,参考文献中提出了类似的想法。 87 ,作者有兴趣以局部 Wannier 表示形式重新表述 Eliashberg 方程。

Even without attempting an all-Wannier calculation as suggested above, our Wannier-Fourier interpolation method will prove useful for solving the Eliashberg equations appearing in the theory of strong-coupling superconductivity,23,24,26 or the Bogoliubov–de Gennes-type equations to be solved in the density-functional theory for superconductors.31,47 In both cases, a very fine description of the e-ph scattering processes near the Fermi surface is strictly required.
即使不尝试如上所述的全 Wannier 计算,我们的 Wannier-Fourier 插值方法也将证明对于求解强耦合超导理论中出现的 Eliashberg方程23、24、26Bogoliubov-de Gennes 型方程非常有用在超导体的密度泛函理论中需要解决。 31, 47在这两种情况下,都严格要求对费米表面附近的 e-ph 散射过程进行非常精细的描述。

The availability of electron and phonon eigenstates and the associated e-ph matrix elements at a very small computational cost could also be used as a starting point to explore the effects of the vertex corrections to the Migdal approximation. Indeed, while, in principle, Migdal theorem does not apply in the presence of Fermi-surface nesting, 26 we are not aware of any attempts to go beyond this approximation within first-principles approaches.
以非常小的计算成本获得电子和声子本征态以及相关的 e-ph 矩阵元素也可以用作探索顶点校正对 Migdal 近似的影响的起点。事实上,虽然原则上米格达勒定理不适用于费米面嵌套的情况, 26但我们不知道有任何尝试在第一性原理方法中超越这种近似。

Finally, we mention the possibility of using the present method to directly address the e-ph interaction in complex systems with many atoms in the unit cell, 35 or to define tight-binding parametrizations for large-scale systems.
最后,我们提到使用本方法直接解决晶胞中具有许多原子的复杂系统中的 e-ph 相互作用的可能性, 35或定义大型系统的紧束缚参数化。

We wish to thank J. R. Yates and I. Souza for their contribution to the initial stages of this work. In particular, we acknowledge J. R. Yates for providing a preliminary interface between the PWSCF and the WANNIER computer codes and I. Souza for useful discussions on the Wannier interpolation of electronic band structures. We also wish to thank C.-H. Park and J. Noffsinger for extensively testing the methodology presented in this work and S. De Gironcoli and X. Blase for fruitful interactions. This work was supported by the National Science Foundation Grant No. DMR04-39768 and by the Office of Science, Office of Basic Energy Sciences, Materials Sciences and Engineering Division, U. S. Department of Energy under Contract No. DE-AC02-05CH11231. Computational resources were provided by NPACI and NERSC. Part of the calculations were performed using the QUANTUM-ESPRESSO 75 and WANNIER 64 packages.
我们要感谢 JR Yates 和 I. Souza 对这项工作初始阶段的贡献。我们特别感谢 JR Yates 在PWSCFWANNIER计算机代码之间提供了初步接口,并感谢 I. Souza 对电子能带结构的 Wannier 插值进行了有益的讨论。我们还要感谢 C.-H. Park 和 J. Noffsinger 对本工作中提出的方法进行了广泛测试,S. De Gironcoli 和 X. Blase 进行了富有成效的互动。这项工作得到了国家科学基金会拨款号DMR04-39768和美国能源部基础能源科学办公室、材料科学与工程司科学办公室的支持,合同号为DE-AC02-05CH11231 。计算资源由 NPACI 和 NERSC 提供。部分计算是使用QUANTUM-ESPRESSO 75WANNIER 64软件包进行的。

APPENDIX: FOURIER INTERPOLATION OF THE SQUARED MATRIX ELEMENT
附录:方阵元素的傅立叶插值

For completeness, in this appendix, we provide a formal justification of the approach proposed in Ref. 35. The procedure introduced in Ref. 35 is similar in spirit to the present work, altough the practical implementation differs considerably. Maintaining the notation of Sec. IV, the squared e-ph matrix element in the Bloch representation introduced in Ref. 35 is
为了完整起见,在本附录中,我们提供了参考文献中提出的方法的正式理由。 35 .参考文献中介绍的程序。 35在精神上与目前的工作相似,尽管实际实施有很大不同。维护 Sec 的符号。 IV ,参考文献中介绍的 Bloch 表示中的平方 e-ph 矩阵元素。 35

𝑔2𝑚𝑛,𝜇𝜈,𝐤(𝐪)=𝑔*𝑚𝑛,𝜇(𝐤,𝐪)𝑔𝑚𝑛,𝜈(𝐤,𝐪),
(A1)

while the corresponding matrix element in the phonon Wannier representation is
而声子 Wannier 表示中相应的矩阵元素是

𝑔2𝑚𝑛,𝜇𝜈,𝐤(𝐑𝑝)=𝜇𝜈𝐪𝑒𝑖𝐪𝐑𝑝𝑢𝐪𝜇𝜇𝑔2𝑚𝑛,𝜇𝜈,𝐤(𝐪)𝑢1𝐪𝜈𝜈.
(A2)

The interpolation formula for an arbitrary phonon momentum 𝐪 given in Ref. 35 is
任意声子动量的插值公式 𝐪 参考文献中给出。 35

𝑔2𝑚𝑛,𝜇𝜈(𝐤,𝐪)=𝜇𝜈𝐑𝑝𝑒𝑖𝐪𝐑𝑝𝑢1𝐪𝜇𝜇𝑔2𝑚𝑛,𝜇𝜈,𝐤(𝐑𝑝)𝑢𝐪𝜈𝜈.
(A3)

The interpolation by means of Eq. (A3) is convenient when the matrix elements 𝑔2𝑚𝑛,𝜇𝜈,𝐤(𝐑𝑝) exhibit a rapid spatial decay in the variable 𝐑𝑝. We show here that the latter requirement follows from the localization of the phonon perturbation in the Wannier representation [Eq. (18)]. For this purpose, we consider the product 𝑔(𝐤,𝐪)𝑢1𝐪 in Eq. (A2). By combining Eqs. (15) and (6) we find
通过方程进行插值。 (A3)当矩阵元素 𝑔2𝑚𝑛,𝜇𝜈,𝐤(𝐑𝑝) 表现出变量的快速空间衰减 𝐑𝑝 。我们在这里表明,后一个要求来自 Wannier 表示中声子扰动的局域化 [方程 1]。 (18) ]。为此,我们考虑产品 𝑔(𝐤,𝐪)𝑢1𝐪 在等式中(A2) 。通过结合等式。 (15)(6)我们发现

𝜈𝑔𝑚𝑛,𝜈(𝐤,𝐪)𝑢1𝐪𝜈𝜈=𝐑𝑝𝑒𝑖𝐪𝐑𝑝𝑚𝐤+𝐪𝜕𝜈,𝐑𝑝𝑉𝑛𝐤.
(A4)

By changing the integration variable to 𝐫𝐑𝑝 and taking into account the resulting Bloch phases of 𝑚𝐤+𝐪 and 𝑛𝐤, we can rewrite Eq. (A4) as
通过将积分变量更改为 𝐫𝐑𝑝 并考虑到由此产生的布洛赫阶段 𝑚𝐤+𝐪𝑛𝐤 ,我们可以重写方程。 (A4)作为

𝜈𝑔𝑚𝑛,𝜈(𝐤,𝐪)𝑢1𝐪𝜈𝜈=𝑁𝑝𝑚𝐤+𝐪𝜕𝜈,𝟎𝑝𝑉𝑛𝐤.
(A5)

It is convenient to Fourier analyze the potential 𝜕𝜈,𝟎𝑝𝑉 as follows:
方便傅立叶分析势 𝜕𝜈,𝟎𝑝𝑉 如下:

𝜕𝜈,𝟎𝑝𝑉(𝐫)=𝜕𝑉𝜈(𝐪)𝑒𝑖𝐪𝐫𝑑𝐪,
(A6)

with the integration extending over the entire reciprocal space. We now combine Eqs. (A5) and (A6), decompose the real-space integral into a sum over 𝑁𝑒 unit cell integrals, and use the periodicity of the Bloch functions. The algebra shows that only the 𝐪 wave component of 𝜕𝜈,𝟎𝑝𝑉(𝐫) appears in the final expression:
积分延伸到整个互易空间。我们现在结合等式。 (A5)(A6) ,将实空间积分分解为总和 𝑁𝑒 晶胞积分,并使用布洛赫函数的周期性。代数表明只有 𝐪 的波分量 𝜕𝜈,𝟎𝑝𝑉(𝐫) 出现在最终的表达式中:

𝜈𝑔𝑚𝑛,𝜈(𝐤,𝐪)𝑢1𝐪𝜈𝜈=𝑁𝑒𝑁𝑝𝜕𝑉𝜈(𝐪)𝑢𝑚𝐤+𝐪𝑢𝑛𝐤,
(A7)

where 𝑢𝑛𝐤 and 𝑢𝑚𝐤+𝐪 are the cell-periodic part of the electron Bloch functions, and the overlap integral is restricted to the unit cell. By combining Eqs. (A2) and (A7), we obtain
在哪里 𝑢𝑛𝐤𝑢𝑚𝐤+𝐪 是电子布洛赫函数的晶胞周期部分,重叠积分仅限于晶胞。通过结合等式。 (A2)(A7) ,我们得到

𝑔2𝑚𝑛,𝜇𝜈,𝐤(𝐑𝑝)=𝑁2𝑒𝑁2𝑝𝐪𝑒𝑖𝐪𝐑𝑝𝜕𝑉*𝜇(𝐪)𝜕𝑉𝜈(𝐪)𝑢𝑚𝐤+𝐪𝑢𝑛𝐤2.
(A8)

Now we observe that (i) the largest nonvanishing Fourier component 𝐪 in Eq. (A6) corresponds to a small fraction of the Brillouin-zone size, since the phonon perturbation in the Wannier representation is localized within a distance corresponding to a few lattice constants. (ii) For small 𝐪, the overlap term 𝑢𝑚𝐤+𝐪𝑢𝑛𝐤2 has no linear variation in 𝐪, as can be derived from 𝐤𝐩 perturbation theory. 88 As a consequence, 𝑢𝑚𝐤+𝐪𝑢𝑛𝐤2 is a slowly varying function of 𝐪 in the region of reciprocal space where the ionic term is most significant. The result is that the spatial decay of the squared e-ph matrix element in the phonon Wannier representation [Eq. (A2)] is dominated by the localization of the phonon perturbation. This property allows an efficient interpolation in Bloch space through Eq. (A3).
现在我们观察到 (i) 最大的非零傅立叶分量 𝐪 在等式中(A6)对应于布里渊区尺寸的一小部分,因为 Wannier 表示中的声子扰动位于与几个晶格常数相对应的距离内。 (ii) 对于小型 𝐪 ,重叠项 𝑢𝑚𝐤+𝐪𝑢𝑛𝐤2 没有线性变化 𝐪 ,可以从 𝐤𝐩 微扰理论。 88因此, 𝑢𝑚𝐤+𝐪𝑢𝑛𝐤2 是一个缓慢变化的函数 𝐪 在离子项最重要的倒易空间区域。结果是声子 Wannier 表示中的平方 e-ph 矩阵元素的空间衰减[方程 1]。 (A2) ] 由声子扰动的局域化主导。该属性允许通过方程式在布洛赫空间中进行有效插值。 (A3)

It should be pointed out that in this derivation, we did not make use of the electron Wannier representation. This constitutes the main difference with the strategy outlined in Sec. III. The advantage of the formulation introduced in Ref. 35 and described in this appendix is that the interpolation over the phonon momentum 𝐪 can be performed independently of the electronic momentum and without resorting to electronic Wannier functions. The disadvantage is that when we need to interpolate both on the electronic momentum 𝐤 and the phonon momentum 𝐪, there is a large computational overhead arising from the need to perform the two operations sequentially. Ultimately, the choice between the two procedures will depend on the specific problem under consideration. When it is hard to obtain electron Wannier functions (e.g., for high-energy conduction band states) or the interpolation over the electron momentum is not needed, it may be preferable to use the scheme outlined in this appendix. In all other cases, the formulation of Sec. III is preferable.
需要指出的是,在这个推导中,我们没有使用电子Wannier表示。这是与第 2 节中概述的策略的主要区别。.参考文献中介绍的配方的优点。 35并在本附录中描述的是声子动量的插值 𝐪 可以独立于电子动量执行,并且无需借助电子万尼尔函数。缺点是当我们需要对电子动量进行插值时 𝐤 和声子动量 𝐪 ,由于需要顺序执行这两个操作,会产生大量的计算开销。最终,这两种程序之间的选择将取决于所考虑的具体问题。当很难获得电子 Wannier 函数(例如,对于高能导带态)或不需要对电子动量进行插值时,最好使用本附录中概述的方案。在所有其他情况下,Sec 的制定。优选III

References  参考

  1. J. R. Schrieffer, Theory of Superconductivity (Benjamin, New York, 1964).
    JR Schrieffer, 《超导理论》 (本杰明,纽约,1964 年)。
  2. W. Kohn, Phys. Rev. Lett. 2, 393 (1959).
    W.科​​恩,物理学家。莱特牧师。 2、393 (1959)
  3. R. E. Peierls, Quantum Theory of Solids (Clarendon, Oxford, 1964).
    RE Peierls,固体量子理论(克拉伦登,牛津,1964 年)。
  4. E. M. Conwell, Proc. Natl. Acad. Sci. U.S.A. 22, 8795 (2005).
    EM 康威尔,国家。阿卡德。科学美国22 , 8795 (2005)。
  5. A. Lanzara, P. V. Bogdanov, X. J. Zhou, S. A. Kellar, D. L. Feng, E. D. Lu, T. Yoshida, H. Eisaki, A. Fujimori, K. Kishio, J.-I. Shimoyama, T. Noda, S. Uchida, Z. Hussain, and Z. X. Shen, Nature (London) 412, 510 (2001).
    A. Lanzara、PV Bogdanov、XJ Zhou、SA Kellar、DL Feng、ED Lu、T. Yoshida、H. Eisaki、A. Fujimori、K. Kishio、J.-I。 Shimoyama、T. Noda、S. Uchida、Z. Hussain 和 ZX Shen, 《自然》(伦敦) 412、510 (2001)
  6. A. Damascelli, Z. Hussain, and Z.-X. Shen, Rev. Mod. Phys. 75, 473 (2003).
    A. Damascelli、Z. Hussain 和 Z.-X。沉牧师模组。物理。 75、473 (2003)
  7. M. Lazzeri, S. Piscanec, F. Mauri, A. C. Ferrari, and J. Robertson, Phys. Rev. Lett. 95, 236802 (2005).
    M. Lazzeri、S. Piscanec、F. Mauri、AC Ferrari 和 J. Robertson,物理学家。莱特牧师。 95、236802 (2005)
  8. N. F. Mott, Proc. Phys. Soc. London 46, 680 (1934). and references therein.
    NF 莫特, Proc。物理。苏克。伦敦46 , 680 (1934) 。以及其中的参考文献。
  9. H. Fröhlich, Phys. Rev. 79, 845 (1950).
    H. Fröhlich,物理学家。修订版79 , 845 (1950)
  10. T. Holstein, Ann. Phys. (N.Y.) 8, 325 (1959).
    T.荷斯坦,安。物理。 (纽约) 8 , 325 (1959)
  11. J. Bardeen and D. Pines, Phys. Rev. 99, 1140 (1955).
    J. Bardeen 和 D. Pines,物理学家。修订版99 , 1140 (1955)
  12. G. Grimvall, The Electron-Phonon Interaction in Metals (North-Holland, New York, 1981).
    G. Grimvall, 《金属中的电子-声子相互作用》 (北荷兰,纽约,1981 年)。
  13. T. K. Mitra, A. Chatterjee, and S. Mukhopadhyay, Phys. Rep. 153, 91 (1987); J. Devreese, editor, Polarons in Ionic Crystals and Polar Semiconductors (North-Holland, Amsterdam, (, 1972).
    TK Mitra、A. Chatterjee 和 S. Mukhopadhyay,物理学家。报告153 , 91 (1987) ; J. Devreese, 《离子晶体和极性半导体中的极化子》编辑(北荷兰,阿姆斯特丹,1972 年)。
  14. G. D. Mahan, Many-Particle Physics (Plenum, New York, 1981).
    GD Mahan,多粒子物理学(Plenum,纽约,1981 年)。
  15. M. Berciu, Phys. Rev. Lett. 97, 036402 (2006).
    M. Berciu,物理学家。莱特牧师。 97、036402 (2006)
  16. P. B. Allen, M. L. Cohen, L. M. Falicov, and R. V. Kasowski, Phys. Rev. Lett. 21, 1794 (1968).
    PB Allen、ML Cohen、LM Falicov 和 RV Kasowski,物理学家。莱特牧师。 21 , 1794 (1968)
  17. M. M. Dacorogna, M. L. Cohen, and P. K. Lam, Phys. Rev. Lett. 55, 837 (1985).
    MM Dacorogna、ML Cohen 和 PK Lam,物理学家。牧师。莱特。 55、837 (1985)
  18. P. K. Lam, M. M. Dacorogna, and M. L. Cohen, Phys. Rev. B 34, 5065 (1986).
    PK Lam,MM Dacorogna 和 ML Cohen, Phys Rev. B 34 , 5065 (1986)
  19. S. Y. Savrasov, D. Y. Savrasov, and O. K. Andersen, Phys. Rev. Lett. 72, 372 (1994).
    SY Savrasov、DY Savrasov 和 OK Andersen,物理学牧师。 72、372 (1994)
  20. S. Y. Savrasov and O. K. Andersen, Phys. Rev. Lett. 77, 4430 (1996).
    SY Savrasov 和 OK Andersen,物理学牧师。 77、4430 (1996)
  21. S. Y. Savrasov and D. Y. Savrasov, Phys. Rev. B 54, 16487 (1996).
    SY Savrasov 和 DY Savrasov,物理学 Rev. B 54 , 16487 (1996)
  22. R. Bauer, A. Schmid, P. Pavone, and D. Strauch, Phys. Rev. B 57, 11276 (1998).
    R. Bauer、A. Schmid、P. Pavone 和 D. Strauch,物理学家。修订版 B 57 , 11276 (1998)
  23. H. J. Choi, D. Roundy, H. Sun, M. L. Cohen, and S. G. Louie, Nature (London) 418, 758 (2002).
    HJ Choi、D. Roundy、H. Sun、ML Cohen 和 SG Louie, 《自然》(伦敦) 418、758 (2002)
  24. H. J. Choi, M. L. Cohen, and S. G. Louie, Phys. Rev. B 73, 104520 (2006).
    HJ Choi、ML Cohen 和 SG Louie,物理学家。修订版 B 73 , 104520 (2006)
  25. A. B. Migdal, Zh. Eksp. Teor. Fiz. 34, 1438 (1958); [Sov. Phys. JETP 7, 996 (1958)].
    AB 米格达勒,Zh。埃克斯普。泰奥.菲兹。 34、1438 (1958); [索夫。物理。 JETP 7 , 996 (1958)]。
  26. P. B. Allen and B. Mikovic, in Solid State Physics edited by H. Ehrenreich, F. Seitz, and D. Turnbull (Academic, New York, 1982), p. 1, Vol. 32.
    PB Allen 和 B. Mikovic, 《固体物理学》 ,由 H. Ehrenreich、F. Seitz 和 D. Turnbull 编辑(学术界,纽约,1982 年),第 17 页。 1,卷。 32 .
  27. D. Connétable, V. Timoshevskii, B. Masenelli1, J. Beille, J. Marcus, B. Barbara, A. M. Saitta, G.-M. Rignanese, P. Mélinon, S. Yamanaka, and X. Blase, Phys. Rev. Lett. 91, 247001 (2003).
    D. Connétable、V. Timoshevskii、B. Masenelli1、J. Beille、J. Marcus、B. Barbara、AM Saitta、G.-M。 Rignanese、P. Mélinon、S. Yamanaka 和 X. Blase,物理学。莱特牧师。 91、247001 (2003)
  28. X. Blase, Ch. Adessi, and D. Connétable, Phys. Rev. Lett. 93, 237004 (2004).
    X.布莱斯,Ch。 Adessi 和 D. Connétable,物理学家。莱特牧师。 93、237004 (2004)
  29. M. J. Verstraete and Xavier Gonze, Phys. Rev. B 74, 153408 (2006).
    MJ Verstraete 和 Xavier Gonze,物理学家。修订版 B 74 , 153408 (2006)
  30. D. Kasinathan, K. Koepernik, J. Kunes, H. Rosner, and W. E. Pickett, Physica C 460, 133 (2007).
    D. Kasinathan、K. Koepernik、J. Kunes、H. Rosner 和 WE Pickett,Physica C 460 , 133 (2007)。
  31. M. A. L. Marques, M. Lüders, N. N. Lathiotakis, G. Profeta, A. Floris, L. Fast, A. Continenza, E. K. U. Gross, and S. Massidda, Phys. Rev. B 72, 024546 (2005).
    MAL Marques、M. Lüders、NN Lathiotakis、G. Profeta、A. Floris、L. Fast、A. Continenza、EKU Gross 和 S. Massidda,物理学家。修订版 B 72 , 024546 (2005)
  32. S. Baroni, S. de Gironcoli, A. Dal Corso, and P. Giannozzi, Rev. Mod. Phys. 73, 515 (2001).
    S. Baroni、S. de Gironcoli、A. Dal Corso 和 P. Giannozzi, Rev. Mod。物理。 73、515 (2001)
  33. E. A. Ekimov, V. A. Sidorov, E. D. Bauer, N. N. Mel’nik, N. J. Curro, J. D. Thompson, and S. M. Stishov, Nature (London) 428, 542 (2004).
    EA Ekimov、VA Sidorov、ED Bauer、NN Mel'nik、NJ Curro、JD Thompson 和 SM Stishov, 《自然》(伦敦) 428、542 (2004)
  34. Y. Takano, M. Nagao, I. Sakaguchi, M. Tachiki, T. Hatano, K. Kobayashi, H. Umezawa, and H. Kawarada, Appl. Phys. Lett. 85, 2851 (2004).
    Y. Takano、M. Nagao、I. Sakaguchi、M. Tachiki、T. Hatano、K. Kobayashi、H. Umezawa 和 H. Kawarada,应用。物理。莱特。 85、2851 (2004)
  35. F. Giustino, Jonathan R. Yates, I. Souza, M. L. Cohen, and S. G. Louie, Phys. Rev. Lett. 98, 047005 (2007).
    F. Giustino、Jonathan R. Yates、I. Souza、ML Cohen 和 SG Louie,物理学家。莱特牧师。 98、047005 (2007)
  36. K. W. Lee and W. E. Pickett, Phys. Rev. Lett. 93, 237003 (2004).
    KW Lee 和 WE Pickett,物理学家。莱特牧师。 93、237003 (2004)
  37. L. Boeri, J. Kortus, and O. K. Andersen, Phys. Rev. Lett. 93, 237002 (2004).
    L. Boeri、J. Kortus 和 OK Andersen,物理学家。莱特牧师。 93、237002 (2004)
  38. S. Engelsberg and J. R. Schrieffer, Phys. Rev. 131, 993 (1963).
    S. Engelsberg 和 JR Schrieffer,物理学家。修订版131 , 993 (1963)
  39. The spin variable is not shown explicitly since we consider here spin-unpolarized systems. The extension of the present formalism to include spin-dependent electron-phonon interaction does not pose any problems.
    自旋变量没有明确显示,因为我们在这里考虑自旋非极化系统。将当前形式主义扩展到包括自旋相关的电子-声子相互作用不会造成任何问题。
  40. M. S. Hybertsen and S. G. Louie, Phys. Rev. B 34, 5390 (1986).
    MS Hybertsen 和 SG Louie,物理学家。修订版 B 34 , 5390 (1986)
  41. F. Giustino, M. L. Cohen, and S. G. Louie (unpublished).
    F. Giustino、ML Cohen 和 SG Louie(未发表)。
  42. L. Hedin and S. Lundqvist, in Solid State Physics edited by F. Seitz, D. Turnbull, and H. Ehrenreich (Academic, New York, 1969), p. 1, Vol. 23.
    L. Hedin 和 S. Lundqvist, 《固体物理学》 ,由 F. Seitz、D. Turnbull 和 H. Ehrenreich 编辑(学术界,纽约,1969 年),第 17 页。 1,卷。 23 .
  43. P. B. Allen, Phys. Rev. B 6, 2577 (1972).
    PB 艾伦,物理学家。修订版 B 6,2577 (1972)
  44. W. L. McMillan, Phys. Rev. 167, 331 (1968).
    WL 麦克米兰,物理学家。修订版167 , 331 (1968)
  45. D. Kasinathan, J. Kunes, A. Lazicki, H. Rosner, C. S. Yoo, R. T. Scalettar, and W. E. Pickett, Phys. Rev. Lett. 96, 047004 (2006).
    D. Kasinathan、J. Kunes、A. Lazicki、H. Rosner、CS Yoo、RT Scalettar 和 WE Pickett,物理学家。莱特牧师。 96、047004 (2006)
  46. M. Calandra and F. Mauri, Phys. Rev. B 71, 064501 (2005).
    M. Calandra 和 F. Mauri,物理学家。修订版 B 71 , 064501 (2005)
  47. M. Lüders, M. A. L. Marques, N. N. Lathiotakis, A. Floris, G. Profeta, L. Fast, A. Continenza, S. Massidda, and E. K. U. Gross, Phys. Rev. B 72, 024545 (2005).
    M. Lüders、MAL Marques、NN Lathiotakis、A. Floris、G. Profeta、L. Fast、A. Continenza、S. Massidda 和 EKU Gross, Phys。修订版 B 72 , 024545 (2005)
  48. G. H. Wannier, Phys. Rev. 52, 191 (1937).
    GH 万尼尔,物理学家。修订版52 , 191 (1937)
  49. N. Marzari and D. Vanderbilt, Phys. Rev. B 56, 12847 (1997).
    N. Marzari 和 D. Vanderbilt,物理学家。修订版 B 56 , 12847 (1997)
  50. R. D. King-Smith and D. Vanderbilt, Phys. Rev. B 47, 1651 (1993).
    RD King-Smith 和 D. Vanderbilt,物理学家。修订版 B 47 , 1651 (1993)
  51. R. Resta, Rev. Mod. Phys. 66, 899 (1994).
    R. Resta, Rev. Mod。物理。 66、899 (1994)
  52. L. He and D. Vanderbilt, Phys. Rev. Lett. 86, 5341 (2001).
    L. He 和 D. Vanderbilt,物理学家。修订版莱特。 86、5341 (2001)
  53. I. Souza, N. Marzari, and D. Vanderbilt, Phys. Rev. B 65, 035109 (2001).
    I. Souza、N. Marzari 和 D. Vanderbilt,物理学家。修订版B 65 , 035109 (2001)
  54. A. Calzolari, N. Marzari, I. Souza, and M. B. Nardelli, Phys. Rev. B 69, 035108 (2004).
    A. Calzolari、N. Marzari、I. Souza 和 MB Nardelli,物理学家。修订版 B 69 , 035108 (2004)
  55. W. Kohn, Phys. Rev. B 7, 2285 (1973).
  56. F. Giustino and A. Pasquarello, Phys. Rev. Lett. 96, 216403 (2006).
  57. Since the dynamical matrix is Hermitian, the eigenmodes satisfy the following relation: 𝐞𝜈𝐪𝜅=[𝐞𝜈𝐪𝜅].
  58. X. Gonze and C. Lee, Phys. Rev. B 55, 10355 (1997).
  59. M. Born and K. Huang, Dynamical Theory of Crystal Lattices (Oxford University Press, Oxford, England, 1954).
  60. S. Barišić, J. Labbé, and J. Friedel, Phys. Rev. Lett. 25, 919 (1970); L. Pietronero, S. Strässler, H. R. Zeller, and M. J. Rice, Phys. Rev. B 22, 904 (1980).
  61. P. Hohenberg and W. Kohn, Phys. Rev. 136, B864 (1964).
  62. W. Kohn and L. J. Sham, Phys. Rev. 140, A1133 (1965).
  63. J. Ihm, A. Zunger, and M. L. Cohen, J. Phys. C 12, 4409 (1979).
  64. A. Mostofi, J. R. Yates, N. Marzari, I. Souza, and D. Vanderbilt, http://www.wannier.org/.
  65. X. Wang, J. R. Yates, I. Souza, and D. Vanderbilt, Phys. Rev. B 74, 195118 (2006).
  66. J. R. Yates, X. Wang, D. Vanderbilt, and I. Souza, Phys. Rev. B 75, 195121 (2007).
  67. P. K. Lam and M. L. Cohen, Phys. Rev. B 25, 6139 (1982).
  68. E. Merzbacher, Quantum Mechanics, 2nd ed. (Wiley, New York, 1970).
  69. A. A. Maradudin and S. H. Vosko, Rev. Mod. Phys. 40, 1 (1968).
  70. N. Troullier and J. L. Martins, Phys. Rev. B 43, 1993 (1991).
  71. D. Vanderbilt, Phys. Rev. B 41, 7892 (1990).
  72. D. M. Ceperley and B. J. Alder, Phys. Rev. Lett. 45, 566 (1980).
  73. J. P. Perdew and A. Zunger, Phys. Rev. B 23, 5048 (1981).
  74. M. Fuchs and M. Scheffler, Comput. Phys. Commun. 119, 67 (1999).
  75. S. Baroni, A. Dal Corso, S. de Gironcoli, P. Giannozzi, C. Cavazzoni, G. Ballabio, S. Scandolo, G. Chiarotti, P. Focher, A. Pasquarello, K. Laasonen, A. Trave, R. Car, N. Marzari, and A. Kokalj, http://www.pwscf.org/.
  76. M. L. Cohen, Phys. Scr., T T1, 5 (1982).
  77. M. Methfessel and A. T. Paxton, Phys. Rev. B 40, 3616 (1989).
  78. M. Hoesch, T. Fukuda, T. Takenouchi, J. P. Sutter, S. Tsutsui, A. Q. R. Baron, M. Nagao, Y. Takano, H. Kawarada, and J. Mizu, Phys. Rev. B 75, 140508 (2007).
  79. J. Kulda, H. Kainzmaier, D. Strauch, B. Dorner, M. Lorenzen, and M. Krisch, Phys. Rev. B 66, 241202(R) (2002).
  80. J. Jiang, R. Saito, Ge. G. Samsonidze, S. G. Chou, A. Jorio, G. Dresselhaus, and M. S. Dresselhaus, Phys. Rev. B 72, 235408 (2005).
  81. W. Kohn, Phys. Rev. Lett. 76, 3168 (1996).
  82. If we assume that a total-energy minimization within a linear-response computer code requires on a modern supercomputer and consider that B-doped diamond in the virtual crystal approximation has six phonon modes for every point, we would need to peform the calculations, corresponding to approximately and .
  83. T. Yokoya, T. Nakamura, T. Matsushita, T. Muro, Y. Takano, M. Nagao, T. Takenouchi, H. Kawarada, and T. Oguchi, Nature (London) 438, 647 (2005).
  84. Y. Ma, J. S. Tse, T. Cui, D. D. Klug, L. Zhang, Y. Xie, Y. Niu, and G. Zou, Phys. Rev. B 72, 014306 (2005).
  85. P. B. Allen and R. C. Dynes, Phys. Rev. B 12, 905 (1975).
  86. H. J. Xiang, Z. Li, J. Yang, J. G. Hou, and Q. Zhu, Phys. Rev. B 70, 212504 (2004).
  87. J. Appel and W. Kohn, Phys. Rev. B 4, 2162 (1971).
  88. P. Y. Yu and M. Cardona, Fundamentals of Semiconductors: Physics and Material Properties, 2nd ed. (Springer-Verlag, Berlin, 1999).
Issue 问题

Vol. 76, Iss. 16 — 15 October 2007
卷。 76、一号。 2007年10月16日至15日

Reuse & Permissions 重用和权限
Author publication services for translation and copyediting assistance advertisement

Authorization Required


×
×

Images

  • Figure 1
    Figure 1
    (Color online) First-order e-ph diagrams considered in this work (red). Left: the self-energy of a phonon with momentum (black wiggly line) is given by the Fermion loop containing two electron lines (red lines) connected by the bare e-ph vertices (red disks) (Refs. 1, 12). Right: the self-energy of an electron with momentum (black straight line) is given by the loop with one electron line (straight) and one phonon line (wiggly), connected by the bare e-ph vertices. Equations (1, 2) were obtained from these diagrams.Reuse & Permissions
  • Figure 2
    Figure 2
    (Color online) Simplified scheme of the electron and phonon Wannier functions entering the three-point e-ph matrix element [Eq. (23)]. The square lattice indicates the unit cells of the crystal, the red lines the electron Wannier functions and , and the blue line the phonon perturbation in the Wannier representation . Whenever two of these functions are centered on distant unit cells, the e-ph matrix element in the Wannier representation vanishes.Reuse & Permissions
  • Figure 3
    Figure 3
    (Color online) Comparison between the electronic band structure of pristine diamond (lines) and of B-doped diamond within a virtual crystal approximation (circles). The Fermi level of doped diamond with a B concentration of 1.85% is located below the valence band top at and is indicated by a black solid line.Reuse & Permissions
  • Figure 4
    Figure 4
    (Color online) Comparison between the phonon dispersions of pristine diamond (solid lines) and B-doped diamond within a virtual crystal approximation (dashed lines). The disks correspond to inelastic neutron scattering data from Ref. 79.Reuse & Permissions
  • Figure 5
    Figure 5
    (Color online) Spatial decay of the electronic Hamiltonian in the Wannier representation [Eq. (26)] as a function of . The data points correspond to the largest value taken over the Wannier functions indices and over the unit cells , located at the same distance : . The data are normalized to their largest value. The inset shows the same quantity on a logarithmic scale .Reuse & Permissions
  • Figure 6
    Figure 6
    (Color online) Spatial decay of the dynamical matrix in the Wannier representation [Eq. (29)] as a function of the distance . The data points correspond to the largest value taken over the ions in the unit cell, the Cartesian directions, and the unit cells , located at the same distance : . The data are normalized to their largest value. The inset shows the same quantity on a logarithmic scale .Reuse & Permissions
  • Figure 7
    Figure 7
    (Color online) Spatial decay of the e-ph vertex in the joint electron-phonon Wannier representation [Eq. (23)] as a function of and : (a) the limiting case and (b) the limiting case . The data points correspond to the largest value taken over the Wannier functions indices, the ions in the unit cell, the Cartesian directions, and the unit cells located at the same distance [panel (a)] or [panel (b)] from the origin of the reference frame: (a) and (b) . The data are normalized to their largest value. The insets show the same quantities on a logarithmic scale . When two Wannier functions are located on a C–C bond crossing a cell boundary, identical e-ph matrix elements appear in adjacent unit cells, resulting in the steplike behavior seen in (b).Reuse & Permissions
  • Figure 8
    Figure 8
    (Color online) Comparison of the e-ph matrix elements [Eq. (5)] obtained by direct first-principles calculations [panel (c), disks] and those computed with the Wannier-Fourier interpolation method discussed in Sec. 4 [panel (c), lines]. The interpolated matrix elements are computed starting from an initial Brillouin-zone grid (dotted line), a grid (dashed line), or a grid (solid line). For illustration, we fixed the initial electronic state for the valence band top at ; we let the final electronic state span the , , and bands as shown in panel (a) (dashed line), and we take the phonon perturbation corresponding to the highest optical branches as shown in panel (b) (dashed line).Reuse & Permissions
  • Figure 9
    Figure 9
    (Color online) Calculated electron linewidths for the and bands of B-doped diamond [Fig. 8a]. Plots (a)–(c) on the left were obtained by using irreducible points in the momentum integration of Eq. (3), while plots (d)–(f) on the right were obtained with irreducible points in the Brillouin zone. We report the results for three broadening parameters : [panels (a) and (d)], [panels (b) and (e)], and [panels (c) and (f)]. The curves are cut off at half the Brillouin-zone size for clarity.Reuse & Permissions
  • Figure 10
    Figure 10
    (Color online) Calculated phonon linewidths for the highest optical mode of B-doped diamond [Fig. 8b]. Plots (a)–(c) on the left were obtained with irreducible points in the momentum integration of Eq. (4), while plots (d)–(f) on the right were obtained with irreducible points. The results for three broadening parameters 10, 50, and are shown in panels (a) and (d), panels (b) and (e), and panels (c) and (f), respectively. Note the different vertical scales in the plots corresponding to different broadening parameters.Reuse & Permissions
  • Figure 11
    Figure 11
    (Color online) Eliashberg function calculated for B-doped diamond in the virtual crystal approximation with (a) and points in the irreducible part of the Brillouin zone and (b) points. For each case, we report the results corresponding to the smearing parameters (dotted lines), (dashed lines), and (solid lines). The Dirac delta function in Eq. (7) was replaced by a Gaussian with a standard deviation of . Note the different energy scales and scaling factors for the acoustic and the optical frequency ranges.Reuse & Permissions
11 of 11
×

Sign up to receive regular email alerts from Physical Review B

Log In

Cancel
×

Search


Article Lookup

Paste a citation or DOI

Enter a citation
×