This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Spin Pumping into Anisotropic Dirac Electrons

Takumi Funato1,2, Takeo Kato3, Mamoru Matsuo2,4,5,6 1Center for Spintronics Research Network, Keio University, Yokohama 223-8522, Japan 2 Kavli Institute for Theoretical Sciences, University of Chinese Academy of Sciences, Beijing, 100190, China. 3Institute for Solid State Physics, The University of Tokyo, Kashiwa, Japan 4 CAS Center for Excellence in Topological Quantum Computation, University of Chinese Academy of Sciences, Beijing 100190, China 5 Advanced Science Research Center, Japan Atomic Energy Agency, Tokai, 319-1195, Japan 6RIKEN Center for Emergent Matter Science (CEMS), Wako, Saitama 351-0198, Japan
Abstract

We study spin pumping into an anisotropic Dirac electron system induced by microwave irradiation to an adjacent ferromagnetic insulator theoretically. We formulate the Gilbert damping enhancement due to the spin current flowing into the Dirac electron system using second-order perturbation with respect to the interfacial exchange coupling. As an illustration, we consider the anisotropic Dirac system realized in bismuth to show that the Gilbert damping varies according to the magnetization direction in the ferromagnetic insulator. Our results indicate that this setup can provide helpful information on the anisotropy of the Dirac electron system.

pacs:
Valid PACS appear here

I Introduction

In spintronics, spin currents are crucial in using electrons’ charge and spin. Spin pumping, the spin current generation of conduction electrons from nonequilibrium magnetization dynamics at magnetic interfaces, is a popular method for generating and manipulating spin currents. In previous experimental reports on spin pumping, the enhancement of Gilbert damping in ferromagnetic resonance (FMR) was observed due to the loss of angular momentum associated with the spin current injection into the nonmagnetic layer adjacent to the ferromagnetic layer Heinrich et al. (1987); Celinski and Heinrich (1991); Mizukami et al. (2001a, b, 2002); Ingvarsson et al. (2002); Lubitz et al. (2003); Sarma (2004); Tserkovnyak et al. (2005). Mizukami et al. measured the enhancement of the Gilbert damping associated with the adjacent nonmagnetic metal. They reported that the strong spin-orbit coupling in the nonmagnetic layer strictly affected the enhancement of the Gilbert damping Mizukami et al. (2001a, b, 2002). Consequently, electric detection by inverse spin Hall effect, in which the charge current is converted from the spin current, led to spin pumping being used as an essential technique for studying spin-related phenomena in nonmagnetic materials Azevedo et al. (2005); Saitoh et al. (2006); Ando et al. (2008, 2009, 2011); Mosendz et al. (2010a, b); Czeschka et al. (2011); Miron et al. (2011); Liu et al. (2011, 2012); Kajiwara et al. (2010); Bai et al. (2013); Sandweg et al. (2011); Sinova et al. (2015). Saitoh et al. measured electric voltage in a bilayer of Py and Pt under microwave application. They observed that charge current converted because of inverse spin Hall effect from spin current injected by spin pumping Saitoh et al. (2006).

In the first theoretical report on spin pumping, Berger predicted an increase in Gilbert damping due to the spin current flowing interface between the ferromagnetic and nonmagnetic layers Berger (1996); Hellman et al. (2017). Tserkovnyak et al. calculated the spin current flowing through the interface Tserkovnyak et al. (2002a, b, 2003) based on the scattering-matrix theory and the picture of adiabatic spin pumping Mucciolo et al. (2002); Sharma and Chamon (2003); Watson et al. (2003). They introduced a complex spin-mixing conductance that characterizes spin transport at the interfaces based on spin conservation and no spin loss. The spin mixing conductance can represent the spin pumping-associated phenomena and is quantitatively evaluated using the first principle calculation Xia et al. (2002). Nevertheless, microscopic analysis is necessary to understand the detailed mechanism of spin transport at the interface Ohnuma et al. (2014); Matsuo et al. (2018); Kato et al. (2019, 2020); Ominato and Matsuo (2020); Ominato et al. (2020); Yamamoto et al. (2021); ominatoAnisotropicSuperconductingSpin2021 ; Yama et al. (2021); Yama2022 ; Ominato et al. (2022). It was clarified that spin pumping depends on the anisotropy of the electron band structure and spin texture. Spin pumping is expected to be one of the probes of the electron states ominatoAnisotropicSuperconductingSpin2021 ; Yama et al. (2021); Yama2022 ; Ominato et al. (2022).

Refer to caption
Figure 1: Schematic illustration of a bilayer system composed of the Dirac electron system and ferromagnetic insulator. The applied microwave excited precession of the localized spin in the ferromagnetic insulator and spin current is injected into the Dirac electron system.

Bismuth has been extensively studied because of its attractive physical properties, such as large diamagnetism, large gg-factor, high efficient Seebeck effect, Subrikov-de Haas effect, and de Haas-van Alphen effect Édel’man (1976); Fuseya et al. (2015). The electrons in the conduction and valence bands near the LL-point in bismuth, which contribute mainly to the various physical phenomena, are expressed as effective Dirac electrons. Thus, electrons in bismuth are called Dirac electrons Wolff (1964); Édel’man (1976); Fuseya et al. (2015). The doping antimony to bismuth is known to close the gap and makes it a topological insulator Fu and Kane (2007); Teo et al. (2008). Because of its strong spin-orbit interaction, bismuth has attracted broad attention in spintronics as a high efficient charge-to-spin conversion material Fuseya et al. (2012a, b, 2014); Fukazawa et al. (2017); Yue et al. (2018); Chi et al. (2020). The spin current generation at the interface between the bismuth oxide and metal has been studied since a significant Rashba spin-orbit interaction appears at the interface Karube et al. (2016). The spin injection into bismuth was observed due to spin pumping from yttrium iron garnet or permalloy Hou et al. (2012); Emoto et al. (2014, 2016). Nevertheless, microscopic analysis of spin pumping into bismuth has not been performed. The dependence of the spin pumping on the crystal and band structure of bismuth remains unclear.

This study aims at a microscopic analysis of spin injection due to spin pumping into an anisotropic Dirac electron system, such as bismuth, and investigates the dependence of spin pumping on the band structure. We consider a bilayer system comprising an anisotropic Dirac electron system and a ferromagnetic insulator where a microwave is applied (see Fig. 1). The effect of the interface is treated by proximity exchange coupling between the Dirac electron spins and the localized spins of the ferromagnetic insulator Ohnuma et al. (2014); Matsuo et al. (2018); Kato et al. (2019, 2020); Ominato and Matsuo (2020); Ominato et al. (2020); Yamamoto et al. (2021); ominatoAnisotropicSuperconductingSpin2021 ; Yama et al. (2021); Yama2022 ; Ominato et al. (2022). We calculate the Gilbert damping enhancement due to spin pumping from the ferromagnetic insulator into the Dirac electron system up to the second perturbation of the interfacial exchange coupling. For illustarion, we calculate the enhancement of the Gilbert damping for an anisotropic Dirac system in bismuth.

This paper is organized as follows: Sec. II describes the model. Sec. III shows the formulation of the Gilbert damping enhancement and discuss the effect of the interfacial randomness on spin pumping. Sec. IV summarizes the results and demonstration of the Gilbert damping enhancement in bismuth. Sec. V presents the conclusion. The Appendices show the details of the calculation. Appendix A defines the magnetic moment of electrons in a Dirac electron system. Appendix B provides the detailed formulation of the Gilbert damping modulation, and Appendix C presents the detailed derivation of Gilbert damping modulation.

II Model

We consider a bilayer system composed of an anisotropic Dirac electron system and a ferromagnetic insulator under a static magnetic field. We evaluate a microscopic model whose Hamiltonian is given as

^T=^D+^FI+^ex,\displaystyle\hat{\mathcal{H}}_{T}=\hat{\mathcal{H}}_{D}+\hat{\mathcal{H}}_{\text{FI}}+\hat{\mathcal{H}}_{\text{ex}}, (1)

where ^D\hat{\mathcal{H}}_{D}, ^FI\hat{\mathcal{H}}_{\text{FI}}, and ^ex\hat{\mathcal{H}}_{\text{ex}} represent an anisotropic Dirac electron system, a ferromagnetic insulator, and an interfacial exchange interaction, respectively.

II.1 Anisotropic Dirac system

The following Wolff Hamiltonian models the anisotropic Dirac electron system Wolff (1964); Fuseya et al. (2012a, 2015):

^D=𝒌c𝒌(𝒌𝒗ρ2+Δρ3)c𝒌,\displaystyle\hat{\mathcal{H}}_{D}=\sum_{\bm{k}}c^{\dagger}_{\bm{k}}(-\hbar\bm{k}\cdot\bm{v}\rho_{2}+\Delta\rho_{3})c_{\bm{k}}, (2)

where 2Δ2\Delta (0\neq 0) is the band gap, c𝒌c^{\dagger}_{\bm{k}}(c𝒌c_{\bm{k}}) is the electrons’ four-component creation (annihilation) operator, and 𝒗\bm{v} is the velocity operator given by vi=αwiασαv_{i}=\sum_{\alpha}w_{i\alpha}\sigma^{\alpha} with wiαw_{i\alpha} being the matrix element of the velocity operator. 𝝈=(σx,σy,σz)\bm{\sigma}=(\sigma^{x},\sigma^{y},\sigma^{z}) are the Pauli matrices in the spin space and 𝝆=(ρ1,ρ2,ρ3)\bm{\rho}=(\rho_{1},\rho_{2},\rho_{3}) are the Pauli matrices specifying the conduction and valence bands.

For this anisotropic Dirac system, the Matsubara Green function of the electrons is given by

g𝒌(iϵn)\displaystyle g_{\bm{k}}(i\epsilon_{n}) =iϵn+μ𝒌~𝝈ρ2+Δρ3(iϵn+μ)2ϵ𝒌2,\displaystyle=\frac{i\epsilon_{n}+\mu-\hbar\tilde{\bm{k}}\cdot\bm{\sigma}\rho_{2}+\Delta\rho_{3}}{(i\epsilon_{n}+\mu)^{2}-\epsilon_{\bm{k}}^{2}}, (3)

where ϵn=(2n+1)π/β\epsilon_{n}=(2n+1)\pi/\beta is the fermionic Matsubara frequencies with nn being integers, μ\mu (>Δ)(>\Delta) is the chemical potential in the conduction band 𝒌~\tilde{\bm{k}} is defined by 𝒌~𝝈=k~ασα=𝒌𝒗\tilde{\bm{k}}\cdot\bm{\sigma}=\tilde{k}_{\alpha}\sigma^{\alpha}=\bm{k}\cdot\bm{v}, and ϵ𝒌\epsilon_{\bm{k}} is the eigenenergy given by

ϵ𝒌=Δ2+(kiwiα)2=Δ2+2k~2.\displaystyle\epsilon_{\bm{k}}=\sqrt{\Delta^{2}+(\hbar k_{i}w_{i\alpha})^{2}}=\sqrt{\Delta^{2}+\hbar^{2}\tilde{k}^{2}}. (4)

The density of state of the Dirac electrons per unit cell per band and spin is givcen by

ν(ϵ)\displaystyle\nu(\epsilon) =nD1𝒌,λδ(ϵλϵ𝒌),\displaystyle=n_{\text{D}}^{-1}\sum_{\bm{k},\lambda}\delta(\epsilon-\lambda\epsilon_{\bm{k}}), (5)
=|ϵ|2π23ϵ2Δ2Δ3detαijθ(|ϵ|Δ),\displaystyle=\frac{|\epsilon|}{2\pi^{2}\hbar^{3}}\sqrt{\frac{\epsilon^{2}-\Delta^{2}}{\Delta^{3}\text{det}\alpha_{ij}}}\theta(|\epsilon|-\Delta), (6)

where nDn_{D} is the number of unit cells in the system and αij\alpha_{ij} is the inverse mass tensor near the bottom of the band, which characterize the band structure of the anisotropic Dirac electron system:

αij=122ϵ𝒌kikj|𝒌=𝟎=1Δαwiαwjα.\displaystyle\alpha_{ij}=\left.\frac{1}{\hbar^{2}}\frac{\partial^{2}\epsilon_{\bm{k}}}{\partial k_{i}\partial k_{j}}\right|_{\bm{k}=\bm{0}}=\frac{1}{\Delta}\sum_{\alpha}w_{i\alpha}w_{j\alpha}. (7)

The spin operator can be defined as

𝒔^𝒒\displaystyle\hat{\bm{s}}_{\bm{q}} =𝒌c𝒌𝒒/2𝒔c𝒌+𝒒/2,\displaystyle=\sum_{\bm{k}}c^{\dagger}_{\bm{k}-\bm{q}/2}\bm{s}c_{\bm{k}+\bm{q}/2}, (8)
si\displaystyle s^{i} =mΔiαρ3σα,(i=x,y,z),\displaystyle=\frac{m}{\Delta}\mathcal{M}_{i\alpha}\rho_{3}\sigma^{\alpha},\ \ \ \ (i=x,y,z), (9)

where MiαM_{i\alpha} are the matrix elements of the spin magnetic moment given as Fuseya et al. (2012b, a)

iα=ϵαβγϵijkwiβwjγ/2.\displaystyle\mathcal{M}_{i\alpha}=\epsilon_{\alpha\beta\gamma}\epsilon_{ijk}w_{i\beta}w_{j\gamma}/2. (10)

The detailed derivation of the spin magnetic moment can be found in Appendix A.

II.2 Ferromagnetic insulator

Refer to caption
Figure 2: Relation between the original coordinates (x,y,z)(x,y,z) and the magnetization-fixed coordinates (X,Y,Z)(X,Y,Z). The direction of the ordered localized spin 𝑺0\langle\bm{S}\rangle_{0} is fixed to the XX-axis. θ\theta is the polar angle and ϕ\phi is the azimuthal angle.

The bulk ferromagnetic insulator under a static magnetic field is described by the quantum Heisenberg model as

^FI=2Ji,j𝑺i𝑺jgμBhdciSiX,\displaystyle\hat{\mathcal{H}}_{\text{FI}}=-2J\sum_{\langle i,j\rangle}\bm{S}_{i}\cdot\bm{S}_{j}-g\mu_{\text{B}}h_{\text{dc}}\sum_{i}S^{X}_{i}, (11)

where JJ is an exchange interaction, gg is g-factor of the electrons, μB\mu_{\text{B}} is the Bohr magnetization, and i,j\langle i,j\rangle represents the pair of nearest neighbor sites. Here, we have introduced a magnetization-fixed coordinate (X,Y,Z)(X,Y,Z), for which the direction of the ordered localized spin 𝑺0\langle\bm{S}\rangle_{0} is fixed to the XX-axis. The localized spin operators for the magnetization-fixed coordinates are related to the ones for the original coordinates (x,y,z)(x,y,z) as

(SxSySz)=(θ,ϕ)(SXSYSZ),\displaystyle\begin{pmatrix}S^{x}\\ S^{y}\\ S^{z}\end{pmatrix}=\mathcal{R}(\theta,\phi)\begin{pmatrix}S^{X}\\ S^{Y}\\ S^{Z}\end{pmatrix}, (12)

where (θ,ϕ)=z(ϕ)y(θ)\mathcal{R}(\theta,\phi)=\mathcal{R}_{z}(\phi)\mathcal{R}_{y}(\theta) is the rotation matrix combining the polar angle θ\theta rotation around the yy-axis y(θ)\mathcal{R}_{y}(\theta) and the azimuthal angle ϕ\phi rotation around the zz-axis z(ϕ)\mathcal{R}_{z}(\phi), given by

(θ,ϕ)=(cosθcosϕsinϕsinθcosϕcosθsinϕcosϕsinθsinϕsinθ0cosθ).\displaystyle\mathcal{R}(\theta,\phi)=\begin{pmatrix}\cos\theta\cos\phi&-\sin\phi&\sin\theta\cos\phi\\ \cos\theta\sin\phi&\cos\phi&\sin\theta\sin\phi\\ -\sin\theta&0&\cos\theta\end{pmatrix}. (13)

By applying the spin-wave approximation, the spin operators are written as S𝒌±=S𝒌Y±iS𝒌Z=2Sb𝒌(b𝒌)S^{\pm}_{\bm{k}}=S^{Y}_{\bm{k}}\pm iS^{Z}_{\bm{k}}=\sqrt{2S}b_{\bm{k}}(b_{\bm{k}}^{\dagger}) and S𝒌X=Sb𝒌b𝒌S^{X}_{\bm{k}}=S-b^{\dagger}_{\bm{k}}b_{\bm{k}} using magnon creation/annihilation operators, b𝒌b_{\bm{k}}^{\dagger} and b𝒌b_{\bm{k}}. Then, the Hamiltonian is rewritten as

^FI=𝒌ω𝒌b𝒌b𝒌,\displaystyle\hat{\mathcal{H}}_{\text{FI}}=\sum_{\bm{k}}\hbar\omega_{\bm{k}}b^{\dagger}_{\bm{k}}b_{\bm{k}}, (14)

where ω𝒌=𝒟k2+ω𝟎\hbar\omega_{\bm{k}}=\mathscr{D}k^{2}+\hbar\omega_{\bm{0}} with 𝒟=zJSa2\mathscr{D}=zJSa^{2} being the spin stiffness and zz being the number of the nearest neighbor sites, and ω𝟎=gμBhdc\hbar\omega_{\bm{0}}=g\mu_{\text{B}}h_{\text{dc}} is the Zeeman energy.

II.3 Interfacial exchange interaction

The proximity exchange coupling between the electron spin in the anisotropic Dirac system and the localized spin in the ferromagnetic insulator is modeled by

^ex=𝒒,𝒌(𝒯𝒒,𝒌s^𝒒+S𝒌+h.c.),\displaystyle\hat{\mathcal{H}}_{\text{ex}}=\sum_{\bm{q},\bm{k}}(\mathcal{T}_{\bm{q},\bm{k}}\hat{s}^{+}_{\bm{q}}S^{-}_{\bm{k}}+\text{h.c.}), (15)

where 𝒯𝒒,𝒌\mathcal{T}_{\bm{q},\bm{k}} is a matrix element for spin transfer through the interface and s^𝒒±=s^𝒒Y±is^𝒒Z\hat{s}^{\pm}_{\bm{q}}=\hat{s}^{Y}_{\bm{q}}\pm i\hat{s}^{Z}_{\bm{q}} are the spin ladder operators of the Dirac electrons. According to the relation between the original coordinate (x,y,z)(x,y,z) and the magnetization-fixed coordinate (X,Y,Z)(X,Y,Z), the spin operators of the Dirac electrons are expressed as

(sXsYsZ)=1(θ,ϕ)(sxsysz),\displaystyle\begin{pmatrix}s^{X}\\ s^{Y}\\ s^{Z}\end{pmatrix}=\mathcal{R}^{-1}(\theta,\phi)\begin{pmatrix}s^{x}\\ s^{y}\\ s^{z}\end{pmatrix}, (16)

where 1(θ,ϕ)=y(θ)z(ϕ)\mathcal{R}^{-1}(\theta,\phi)=\mathcal{R}_{y}(\theta)\mathcal{R}_{z}(-\phi) is given by

1(θ,ϕ)=(cosθcosϕcosθsinϕsinθsinϕcosϕ0sinθcosϕsinθsinϕcosθ).\displaystyle\mathcal{R}^{-1}(\theta,\phi)=\begin{pmatrix}\cos\theta\cos\phi&\cos\theta\sin\phi&-\sin\theta\\ -\sin\phi&\cos\phi&0\\ \sin\theta\cos\phi&\sin\theta\sin\phi&\cos\theta\end{pmatrix}. (17)

The spin ladder operators are given by

s+=mΔaiiασα,s=mΔaiiασα,\displaystyle s^{+}=\frac{m}{\Delta}a_{i}\mathcal{M}_{i\alpha}\sigma^{\alpha},\ \ \ \ s^{-}=\frac{m}{\Delta}a_{i}^{*}\mathcal{M}_{i\alpha}\sigma^{\alpha}, (18)

where aia_{i} (i=x,y,zi=x,y,z) are defined by

(axayaz)=(sinϕ+isinθcosϕcosϕ+isinθsinϕicosθ).\displaystyle\begin{pmatrix}a_{x}\\ a_{y}\\ a_{z}\end{pmatrix}=\begin{pmatrix}-\sin\phi+i\sin\theta\cos\phi\\ \cos\phi+i\sin\theta\sin\phi\\ i\cos\theta\end{pmatrix}. (19)

III Formulation

Applying a microwave to the ferromagnetic insulator includes the localized spin’s precession. The Gilbert damping constant can be read from the retarded magnon Green function defined by

G𝒌R(ω)=i0𝑑tei(ω+iδ)t[S𝒌+(t),S𝒌],\displaystyle G_{\bm{k}}^{R}(\omega)=-\frac{i}{\hbar}\int^{\infty}_{0}dte^{i(\omega+i\delta)t}\langle[S^{+}_{\bm{k}}(t),S^{-}_{\bm{k}}]\rangle, (20)

with S𝒌+(t)=ei^T/S𝒌+ei^T/S^{+}_{\bm{k}}(t)=e^{i\hat{\mathcal{H}}_{T}/\hbar}S^{+}_{\bm{k}}e^{-i\hat{\mathcal{H}}_{T}/\hbar} being the Heisenberg representation of the localized spin, since one can prove that the absorption rate of the microwave is proportional to ImG𝒌=𝟎R(ω){\rm Im}\,G_{{\bm{k}}={\bm{0}}}^{R}(\omega) (see also Appendix B). By considering the second-order perturbation with respect to the matrix element for the spin transfer 𝒯𝒒,𝒌\mathcal{T}_{\bm{q},\bm{k}}, the magnon Green function is given by Ohnuma et al. (2014); Matsuo et al. (2018); Kato et al. (2019, 2020); Ominato and Matsuo (2020); Ominato et al. (2020); Yamamoto et al. (2021); ominatoAnisotropicSuperconductingSpin2021 ; Yama et al. (2021); Yama2022 ; Ominato et al. (2022)

G𝟎R(ω)=2S/(ωω𝟎)+i(α+δα)ω.\displaystyle G_{\bm{0}}^{R}(\omega)=\frac{2S/\hbar}{(\omega-\omega_{\bm{0}})+i(\alpha+\delta\alpha)\omega}. (21)

Here, we introduced a term, iαωi\alpha\omega, in the denominator to express the spin relaxation within a bulk FI, where α\alpha indicates the strength of the Gilbert damping. The enhancement of the damping, δα\delta\alpha, is due to the adjacent Dirac electron system, calculated by

δα=2Sω𝒒|𝒯𝒒,𝟎|2Imχ𝒒R(ω),\displaystyle\delta\alpha=\frac{2S}{\hbar\omega}\sum_{\bm{q}}|\mathcal{T}_{\bm{q},\bm{0}}|^{2}\text{Im}\,\chi_{\bm{q}}^{R}(\omega), (22)

where χ𝒒R(ω)\chi^{R}_{\bm{q}}(\omega) is the retarded component of the spin susceptibility (defined below). We assume that the FMR peak described by ImG𝒌=𝟎R(ω){\rm Im}\,G_{{\bm{k}}={\bm{0}}}^{R}(\omega) is sufficiently sharp, i.e., α+δα1\alpha+\delta\alpha\ll 1. Then, the enhancement of the Gilbert damping can be regarded as almost constant around the peak (ωω0\omega\simeq\omega_{0}), allowing us to replace ω\omega in δα\delta\alpha with ω𝟎\omega_{\bm{0}}.

The retarded component of the spin susceptibility for the Dirac electrons:

χ𝒒R(ω)=i𝑑tei(ω+iδ)tθ(t)[s𝒒+(t),s𝒒].\displaystyle\chi^{R}_{\bm{q}}(\omega)=\frac{i}{\hbar}\int^{\infty}_{-\infty}dte^{i(\omega+i\delta)t}\theta(t)\langle[s^{+}_{\bm{q}}(t),s_{-\bm{q}}^{-}]\rangle. (23)

The retarded component of the spin susceptibility is derived from the following Matsubara Green function through analytic continuation iωlω+iδi\omega_{l}\to\hbar\omega+i\delta:

χ𝒒(iωl)=0β𝑑τeiωlτs^𝒒+(τ)s^𝒒,\displaystyle\chi_{\bm{q}}(i\omega_{l})=\int^{\beta}_{0}d\tau e^{i\omega_{l}\tau}\langle\hat{s}^{+}_{\bm{q}}(\tau)\hat{s}^{-}_{-\bm{q}}\rangle, (24)

where ωl=2πl/β\omega_{l}=2\pi l/\beta is the bosonic Matsubara frequency with ll being integers. According to Wick’s theorem, the Matsubara representation of the spin susceptibility is given by

χ𝒒(iωl)\displaystyle\chi_{\bm{q}}(i\omega_{l})
=β1𝒌,iϵntr[s+g𝒌+𝒒(iϵn+iωl)sg𝒌(iϵn)],\displaystyle=-\beta^{-1}\sum_{\bm{k},i\epsilon_{n}}\text{tr}[s^{+}g_{\bm{k}+\bm{q}}(i\epsilon_{n}+i\omega_{l})s^{-}g_{\bm{k}}(i\epsilon_{n})], (25)

where iϵn\sum_{i\epsilon_{n}} indicates the sum with respect to the fermionic Matsubara frequency, ϵn=(2π+1)n/β\epsilon_{n}=(2\pi+1)n/\beta. The imaginary part of the spin susceptibility is given by

Imχ𝒒R(ω)=π(θ,ϕ)𝒌λ,λ=±[12+λλ62Δ2+ϵ𝒌2ϵ𝒌ϵ𝒌+𝒒]\displaystyle\text{Im}\,\chi^{R}_{\bm{q}}(\omega)=-\pi\mathcal{F}(\theta,\phi)\sum_{\bm{k}}\sum_{\lambda,\lambda^{\prime}=\pm}\left[\frac{1}{2}+\frac{\lambda\lambda^{\prime}}{6}\frac{2\Delta^{2}+\epsilon_{\bm{k}}^{2}}{\epsilon_{\bm{k}}\epsilon_{\bm{k}+\bm{q}}}\right]
×[f(λϵ𝒌+𝒒)f(λϵ𝒌)]δ(ωλϵ𝒌+𝒒+λϵ𝒌),\displaystyle\hskip 11.38109pt\times\Bigl{[}f(\lambda^{\prime}\epsilon_{\bm{k}+\bm{q}})-f(\lambda\epsilon_{\bm{k}})\Bigr{]}\delta(\hbar\omega-\lambda^{\prime}\epsilon_{\bm{k}+\bm{q}}+\lambda\epsilon_{\bm{k}}), (26)

where f(ϵ)=(eβ(ϵμ)+1)1f(\epsilon)=(e^{\beta(\epsilon-\mu)}+1)^{-1} is the Fermi distribution function, λ=±\lambda=\pm is a band index (see Fig. 3), and (θ,ϕ)\mathcal{F}(\theta,\phi) is the dimensionless function which depends on the direction of the ordered localized spin, defined by

(θ,ϕ)=(2mΔ)2αaiiαajjα.\displaystyle\mathcal{F}(\theta,\phi)=\left(\frac{2m}{\Delta}\right)^{2}\sum_{\alpha}a_{i}\mathcal{M}_{i\alpha}a^{*}_{j}\mathcal{M}_{j\alpha}. (27)

For detailed derivation, see Appendix C.

Refer to caption
Figure 3: Schematic illustration of the band structure of the anisotropic Dirac electron system. The red band represents the conduction band with λ=+\lambda=+, and the blue band represents the valence band with λ=\lambda=-. The chemical potential is in the conduction band.

In this paper, we model the interfacial spin transfer as a combination of the clean and dirty processes. The former corresponds to the momentum-conserved spin transfer and the latter to the momentum-nonconserved one ominatoAnisotropicSuperconductingSpin2021 ; Ominato et al. (2022). By averaging over the position of the localized spin at the interface, we can derive the matrix elements of the interfacial spin-transfer process as

|𝒯𝒒,𝟎|2=𝒯12δ𝒒,𝟎+𝒯22,\displaystyle|\mathcal{T}_{\bm{q},\bm{0}}|^{2}=\mathcal{T}_{1}^{2}\delta_{\bm{q},\bm{0}}+\mathcal{T}_{2}^{2}, (28)

where 𝒯1\mathcal{T}_{1} and 𝒯2\mathcal{T}_{2} are the averaged matrix elements contributing to the clean and dirty processes, respectively. Then, the enhancement of the Gilbert damping is given by

δα=2Sω(θ,ϕ){𝒯1Imχ~uniR(ω𝟎)+𝒯2Imχ~locR(ω𝟎)},\displaystyle\delta\alpha=\frac{2S}{\hbar\omega}\mathcal{F}(\theta,\phi)\Bigl{\{}\mathcal{T}_{1}\,\text{Im}\,\tilde{\chi}^{R}_{\text{uni}}(\omega_{{\bm{0}}})+\mathcal{T}_{2}\,\text{Im}\,\tilde{\chi}^{R}_{\text{loc}}(\omega_{{\bm{0}}})\Bigr{\}}, (29)

where χuniR(ω)\chi^{R}_{\text{uni}}(\omega) and χuniR(ω)\chi^{R}_{\text{uni}}(\omega) are the local and uniform spin susceptibilities defined by

χ~locR(ω𝟎)\displaystyle\tilde{\chi}^{R}_{\text{loc}}(\omega_{{\bm{0}}}) =1(θ,ϕ)𝒒χ𝒒R(ω𝟎),\displaystyle=\mathcal{F}^{-1}(\theta,\phi)\sum_{\bm{q}}\chi^{R}_{\bm{q}}(\omega_{{\bm{0}}}), (30)
χ~uniR(ω𝟎)\displaystyle\tilde{\chi}^{R}_{\text{uni}}(\omega_{{\bm{0}}}) =1(θ,ϕ)χ𝟎R(ω𝟎),\displaystyle=\mathcal{F}^{-1}(\theta,\phi)\chi^{R}_{\bm{0}}(\omega_{{\bm{0}}}), (31)

respectively. From Eq. (III), their imaginary parts are calculated as

Imχ~locR(ω𝟎)\displaystyle\text{Im}\,\tilde{\chi}_{\text{loc}}^{R}(\omega_{{\bm{0}}}) =πnD2𝑑ϵν(ϵ)ν(ϵ+ω𝟎)\displaystyle=-\pi n_{D}^{2}\int d\epsilon\nu(\epsilon)\nu(\epsilon+\hbar\omega_{{\bm{0}}})
×[12+2Δ2+ϵ26ϵ(ϵ+ω𝟎)][f(ϵ+ω𝟎)f(ϵ)],\displaystyle\hskip 5.69054pt\times\left[\frac{1}{2}+\frac{2\Delta^{2}+\epsilon^{2}}{6\epsilon(\epsilon+\hbar\omega_{{\bm{0}}})}\right]\Bigl{[}f(\epsilon+\hbar\omega_{{\bm{0}}})-f(\epsilon)\Bigr{]}, (32)
Imχ~uniR(ω𝟎)\displaystyle\text{Im}\,\tilde{\chi}^{\text{R}}_{\text{uni}}(\omega_{{\bm{0}}}) =πnDν(ω𝟎2)2ω𝟎24Δ232ω𝟎2\displaystyle=-\pi n_{D}\nu\left(\tfrac{\hbar\omega_{{\bm{0}}}}{2}\right)\frac{\hbar^{2}\omega_{{\bm{0}}}^{2}-4\Delta^{2}}{3\hbar^{2}\omega_{{\bm{0}}}^{2}}
×[f(ω𝟎2)f(ω𝟎2)].\displaystyle\hskip 5.69054pt\times\Bigl{[}f(\tfrac{\hbar\omega_{{\bm{0}}}}{2})-f(-\tfrac{\hbar\omega_{{\bm{0}}}}{2})\Bigr{]}. (33)

The enhancement of the Gilbert damping, δα\delta\alpha, depends on the direction of the ordered localized spin through the dimensionless function (θ,ϕ)\mathcal{F}(\theta,\phi) regardless of the interfacial condition.

Refer to caption
Figure 4: FMR frequency dependence of the (a) local and (b) uniform spin susceptibilities. The local spin susceptibility is normalized by πnD2ν02\pi n_{D}^{2}\nu_{0}^{2} and scaled by 10610^{6}, and the uniform spin susceptibility is normalized by πnDν0\pi n_{D}\nu_{0} with ν01/2π23detαij\nu_{0}\equiv 1/2\pi^{2}\hbar^{3}\sqrt{\text{det}\alpha_{ij}}. Note that kBk_{B} is the Boltzmann constant. The line with kBT/Δ=0.001k_{B}T/\Delta=0.001 is absent in (a) because the local spin susceptibility approaches zero at low temperature.

By contrast, the FMR frequency dependence of δα\delta\alpha reflects the interfacial condition; for a clean interface, it is determined mainly by ImχuniR(ω𝟎)\text{Im}\,\chi^{R}_{\text{uni}}(\omega_{{\bm{0}}}), whereas for a dirty interface, it is determined by ImχlocR(ω𝟎)\text{Im}\,\chi^{R}_{\text{loc}}(\omega_{{\bm{0}}}). The FMR frequency dependence of the local and uniform spin susceptibilities, ImχlocR(ω𝟎)\text{Im}\,\chi^{R}_{\text{loc}}(\omega_{{\bm{0}}}) and ImχuniR(ω𝟎)\text{Im}\,\chi^{R}_{\text{uni}}(\omega_{{\bm{0}}}), are plotted in Figs. 4 (a) and (b), respectively. The local and uniform spin susceptibilities are normalized by πnD2ν02\pi n_{D}^{2}\nu_{0}^{2} and πnDν0\pi n_{D}\nu_{0}, respectively, where ν01/2π23detαij\nu_{0}\equiv 1/2\pi^{2}\hbar^{3}\sqrt{\text{det}\alpha_{ij}} is defined. In the calculation, the ratio of the chemical potential to the energy gap was set to μ/Δ4.61\mu/\Delta\simeq 4.61, which is the value in the bismuth Fuseya et al. (2015). According to Fig. 4 (a), the local spin susceptibility increases linearly with the frequency ω\omega in the low-frequency region. This ω\omega-linear behavior can be reproduced analytically for low temperatures and ωμ\hbar\omega\ll\mu:

Imχ~loc(ω𝟎)\displaystyle\text{Im}\,\tilde{\chi}_{\text{loc}}(\omega_{\bm{0}}) ω𝟎π2nD2[ν(μ)]2[1+2Δ2+μ23μ2].\displaystyle\simeq\hbar\omega_{{\bm{0}}}\frac{\pi}{2}n_{D}^{2}[\nu(\mu)]^{2}\left[1+\frac{2\Delta^{2}+\mu^{2}}{3\mu^{2}}\right]. (34)

Fig. 4 (b) indicates a strong suppression of the uniform spin susceptibility below a spin-excitation gap (ω𝟎<2μ\omega_{\bm{0}}<2\mu). This feature can be checked by its analytic form at zero temperature:

Imχ~uniR(ω𝟎)\displaystyle\text{Im}\,\tilde{\chi}^{\text{R}}_{\text{uni}}(\omega_{{\bm{0}}}) =πnDν(ω𝟎2)2ω𝟎24Δ232ω𝟎2θ(ω𝟎2μ).\displaystyle=\pi n_{D}\nu\left(\tfrac{\hbar\omega_{{\bm{0}}}}{2}\right)\frac{\hbar^{2}\omega_{{\bm{0}}}^{2}-4\Delta^{2}}{3\hbar^{2}\omega_{{\bm{0}}}^{2}}\theta(\hbar\omega_{{\bm{0}}}-2\mu). (35)

Thus, the FMR frequency dependence of the enhancement of the Gilbert damping depends on the interfacial condition. This indicates that the measurement of the FMR frequency dependence may provide helpful information on the randomness of the junction.

IV Result

We consider bismuth, which is one of the anisotropic Dirac electron systems Édel’man (1976); Tang and S. Dresselhaus (2014); Fuseya et al. (2014, 2015); Zhu et al. (2011). The crystalline structure of pure bismuth is a rhombohedral lattice with the space group of R3¯mR\bar{3}m symmetry, see Figs. 5 (a) and (b). It is reasonable to determine the Cartesian coordinate system in the rhombohedral structure using the trigonal axis with C3C_{3} symmetry, the binary axis with C2C_{2} symmetry, and the bisectrix axis, which is perpendicular to the trigonal and binary axes. Hereafter, we choose the xx-axis as the binary axis, the yy-axis as the bisectrix axis, and the zz-axis as the trigonal axis. Note that the trigonal, binary, and bisectrix axes are denoted as [0001][0001], [12¯10][1\bar{2}10], and [101¯0][10\bar{1}0], respectively, where the Miller-Bravais indices are used. The bismuth’s band structure around the Fermi surface consists of three electron ellipsoids at LL-points and one hole ellipsoid at the TT-point. It is well known that the electron ellipsoids are the dominant contribution to the transport phenomena since electron’s mass is much smaller than that of the hole, see Fig. 5 (c). Therefore, the present study considers only the electron systems at the LL-points. The electron ellipsoids are significantly elongated, with the ratio of the major to minor axes being approximately 15:115:1. Each of the three electron ellipsoids can be converted to one another with 2π/32\pi/3 rotation around the trigonal axis. The electron ellipsoid along the bisectrix axis is labeled as e1e1, and the other two-electron ellipsoids are labeled e2e2 and e3e3. The inverse mass tensor for the e1e1 electron ellipsoids is given by

αe1=(α1000α2α40α4α3).\displaystyle\tensor{\alpha}_{e1}=\begin{pmatrix}\alpha_{1}&0&0\\[5.0pt] 0&\alpha_{2}&\alpha_{4}\\[5.0pt] 0&\alpha_{4}&\alpha_{3}\end{pmatrix}. (36)

The inverse mass tensor of the electron ellipsoids e2e2 and e3e3 are obtained by rotating that of e1e1 by 2π/32\pi/3 rotation as below:

αe2,e3=14(α1+3α2±3(α1α2)±23α4±3(α1α2)3α1+α22α4±23α42α44α3).\displaystyle\tensor{\alpha}_{e2,e3}=\frac{1}{4}\begin{pmatrix}\alpha_{1}+3\alpha_{2}&\pm\sqrt{3}(\alpha_{1}-\alpha_{2})&\pm 2\sqrt{3}\alpha_{4}\\[5.0pt] \pm\sqrt{3}(\alpha_{1}-\alpha_{2})&3\alpha_{1}+\alpha_{2}&-2\alpha_{4}\\[5.0pt] \pm 2\sqrt{3}\alpha_{4}&-2\alpha_{4}&4\alpha_{3}\end{pmatrix}. (37)

Let us express the dimensionless function (θ,ϕ)\mathcal{F}(\theta,\phi) representing the localized spin direction dependence of the damping enhancement on the inverse mass tensors.

(θ,ϕ)=(2mΔ)2α[(sin2ϕ+sin2θcos2ϕ)xα2\displaystyle\mathcal{F}(\theta,\phi)=\left(\frac{2m}{\Delta}\right)^{2}\sum_{\alpha}\Bigl{[}(\sin^{2}\phi+\sin^{2}\theta\cos^{2}\phi)\mathcal{M}_{x\alpha}^{2}
+(cos2ϕ+sin2θsin2ϕ)yα2\displaystyle+(\cos^{2}\phi+\sin^{2}\theta\sin^{2}\phi)\mathcal{M}_{y\alpha}^{2}
+cos2θ(zα2sin2ϕxαyα)\displaystyle+\cos^{2}\theta(\mathcal{M}_{z\alpha}^{2}-\sin 2\phi\mathcal{M}_{x\alpha}\mathcal{M}_{y\alpha})
+sin2θzα(xαcosϕ+yαsinϕ)].\displaystyle+\sin 2\theta\mathcal{M}_{z\alpha}(\mathcal{M}_{x\alpha}\cos\phi+\mathcal{M}_{y\alpha}\sin\phi)\Bigr{]}. (38)

Here, we use the following calculations:

αxα2=Δ24(αyyαzzαyz2)total=Δ24m2κ¯,\displaystyle\sum_{\alpha}\mathcal{M}_{x\alpha}^{2}=\frac{\Delta^{2}}{4}(\alpha_{yy}\alpha_{zz}-\alpha_{yz}^{2})_{\text{total}}=\frac{\Delta^{2}}{4m^{2}}\bar{\kappa}_{\perp}, (39)
αMyα2=Δ24(αzzαxxαzx2)total=Δ24m2κ¯,\displaystyle\sum_{\alpha}M_{y\alpha}^{2}=\frac{\Delta^{2}}{4}(\alpha_{zz}\alpha_{xx}-\alpha_{zx}^{2})_{\text{total}}=\frac{\Delta^{2}}{4m^{2}}\bar{\kappa}_{\perp}, (40)
αMzα2=Δ24(αxxαyyαxy2)total=Δ24m2κ¯,\displaystyle\sum_{\alpha}M_{z\alpha}^{2}=\frac{\Delta^{2}}{4}(\alpha_{xx}\alpha_{yy}-\alpha_{xy}^{2})_{\text{total}}=\frac{\Delta^{2}}{4m^{2}}\bar{\kappa}_{\parallel}, (41)
αiαjα=Δ24(αikαjkαijαkk)total=0,\displaystyle\sum_{\alpha}\mathcal{M}_{i\alpha}\mathcal{M}_{j\alpha}=\frac{\Delta^{2}}{4}(\alpha_{ik}\alpha_{jk}-\alpha_{ij}\alpha_{kk})_{\text{total}}=0, (42)

where i,j,ki,j,k are cyclic. ()total(\cdots)_{\text{total}} represents the summation of the contributions of the three electron ellipsoids, and κ¯\bar{\kappa}_{\parallel}, κ¯\bar{\kappa}_{\perp} (>0)(>0) are the total Gaussian curvature of the three electron ellipsoids normalized by the electron mass mm, given by

κ¯=3m2α1α2,\displaystyle\bar{\kappa}_{\parallel}=3m^{2}\alpha_{1}\alpha_{2}, (43)
κ¯=32m2[(α1+α2)α3α42].\displaystyle\bar{\kappa}_{\perp}=\frac{3}{2}m^{2}[(\alpha_{1}+\alpha_{2})\alpha_{3}-\alpha_{4}^{2}]. (44)

Hence, the dimensionless function \mathcal{F} is given by

(θ)=(1+sin2θ)κ¯+cos2θκ¯.\displaystyle\mathcal{F}(\theta)=(1+\sin^{2}\theta)\bar{\kappa}_{\perp}+\cos^{2}\theta\bar{\kappa}_{\parallel}. (45)
Refer to caption
Figure 5: (a) The rhombohedral lattice structure of bismuth. The xx-axis, yy-axis, and zz-axis are chosen as the binary axis with C2C_{2} symmetry, the bisectrix axis, and the trigonal axis with C3C_{3} symmetry, respectively. The yellow lines represents the unit cell of the rhombohedral lattice. (b) The rhombohedral structure viewed from the trigonal axis. (c) Schematic illustration of the band structure at the Fermi surface. The three electron ellispoids at LL-points are dominant contribution to the spin transport.

The results suggest that the variation of the damping enhancement depends only on the polar angle θ\theta, which is the angle between the direction of the ordered localized spin 𝑺0\langle\bm{S}\rangle_{0} and the trigonal axis. It is also found that the θ\theta dependence of the damping enhancement originates from the anisotropy of the band structure. The dimensionless function (θ)\mathcal{F}(\theta) is plotted in Fig. 6 by varying the ratio of the total Gaussian curvatures x=κ¯/κ¯x=\bar{\kappa}_{\perp}/\bar{\kappa}_{\parallel}, which corresponds to the anisotropy of the band structure. Figure 6 shows that the θ\theta-dependence of the damping enhancement decreases with smaller xx and the angular dependence vanishes in an isotropic Dirac electron system x=1x=1. Bismuth is known to have a strongly anisotropic band structure. The magnitude of the matrix elements of the inverse mass α1\alpha_{1}-α4\alpha_{4} was experimentally determined as mα1=806m\alpha_{1}=806, mα2=7.95m\alpha_{2}=7.95, mα3=349m\alpha_{3}=349, and mα4=37.6m\alpha_{4}=37.6. The total Gaussian curvatures are evaluated as Fuseya et al. (2015)

κ¯1.92×104,\displaystyle\bar{\kappa}_{\parallel}\simeq 1.92\times 10^{4}, (46)
κ¯4.24×105.\displaystyle\bar{\kappa}_{\perp}\simeq 4.24\times 10^{5}. (47)

The ratio of the total Gaussian curvature is estimated as x22.1x\simeq 22.1. Therefore, the damping enhancement is expected to depend strongly on the polar angle θ\theta in a bilayer system composed of single-crystalline bismuth and ferromagnetic insulator. Conversely, the θ\theta-dependence of the damping enhancement is considered to be suppressed for polycrystalline bismuth.

Refer to caption
Figure 6: The θ\theta-dependence of the damping enhancement for different xx. The ratio of the total Gaussian curvatures x=κ¯/κ¯x=\bar{\kappa}_{\perp}/\bar{\kappa}_{\parallel} represents the anisotropy of the band structure. The blue line with x=22.1x=22.1 corresponds to the damping enhancement in single-crystalline bismuth, and the other lines correspond to that in the weakly anisotropic band structure. As can be seen from the graph, the θ\theta-dependence of the damping enhancement decreases as the more weakly anisotropic band structure, and the angular dependence turns out to vanish in an isotropic Dirac electron system with x=1x=1.

The damping enhancement is independent of the azimuthal angle ϕ\phi. Therefore, it is invariant even on rotating the spin orientation around the trigonal axis. The reason is that the azimuthal angular dependence of the damping enhancement cancels out when the contributions of the three electron ellipsoids are summed over, although each contribution depends on the azimuthal angle. The azimuthal angular dependence of the damping enhancement is expected to remain when strain breaks the in-plane symmetry. Additionally, suppose the spin can be injected into each electron ellipsoid separately, e.g., by interfacial manipulation of the bismuth atoms. In that case, the damping enhancement depends on the azimuthal angle of the spin orientation of the ferromagnetic insulator Ominato et al. (2020). This may be one of the probes of the electron ellipsoidal selective transport phenomena.

It is also noteworthy that the damping enhancement varies according to the ordered localized spin direction with both clean and dirty interfaces; that is independent of whether momentum is conserved in interfacial spin transport. Conversely, it was reported that the spin orientation dependence of the damping enhancement due to the Rashba and Dresselhaus spin-orbit interaction turned out to vanish by interfacial inhomogeneity Yama et al. (2021); Yama2022 .

V conclusion

We theoretically studied spin pumping from a ferromagnetic insulator to an anisotropic Dirac electron system. We calculated the enhancement of the Gilbert damping in the second perturbation concerning the proximity interfacial exchange interaction by considering the interfacial randomness. For illustration, we calculated the enhancement of the Gilbert damping for an anisotropic Dirac system realized in bismuth. We showed that the Gilbert damping varies according to the polar angle between the ordered spin 𝑺0\langle\bm{S}\rangle_{0} and the trigonal axis of the Dirac electron system whereas it is invariant in its rotation around the trigonal axis. Our results indicate that the spin pumping experiment can provide helpful information on the anisotropic band structure of the Dirac electron system.

The Gilbert damping is invariant in the rotation around the trigonal axis because the contributions of each electron ellipsoid depend on the in-plane direction of the ordered spin 𝑺0\langle\bm{S}\rangle_{0}. Nevertheless, the total contribution becomes independent of the rotation of the trigonal axis after summing up the contributions from the three electron ellipsoids that are related to each other by the C3C_{3} symmetry of the bismuth crystalline structure. If the spin could be injected into each electron ellipsoid separately, it is expected that the in-plane direction of the ordered localized spin would influence the damping enhancement. This may be one of the electron ellipsoid selective spin injection probes. The in-plane direction’s dependence will also appear when a static strain is applied. A detailed discussion of these effects is left as a future problem.

Acknowledgements.
The authors would like to thank A. Yamakage and Y. Ominato for helpful and enlightening discussions. The continued support of Y. Nozaki is greatly appreciated. We also thank H, Nakayama for the daily discussions. This work was partially supported by JST CREST Grant No. JPMJCR19J4, Japan. This work was supported by JSPS KAKENHI for Grants (Nos. 20H01863, 20K03831, 21H04565, 21H01800, and 21K20356). MM was supported by the Priority Program of the Chinese Academy of Sciences, Grant No. XDB28000000.

Appendix A Magnetic moment of electrons in Dirac electron system

In this section, we define the spin operators in the Dirac electron systems. The Wolff Hamiltonian around the L point is given by D=ρ3Δρ2𝝅𝒗\mathcal{H}_{\text{D}}=\rho_{3}\Delta-\rho_{2}\bm{\pi}\cdot\bm{v}, where vi=αwiασαv_{i}=\sum_{\alpha}w_{i\alpha}\sigma^{\alpha} with wiαw_{i\alpha} being the matrix component of the velocity vectors and 𝝅=𝒑+ec𝑨\bm{\pi}=\bm{p}+\frac{e}{c}\bm{A} is the momentum operator including the vector potential. It is reasonable to determine the magnetic moment of electrons in an effective Dirac system as the coefficient of the Zeeman term. The Wolff Hamiltonian is diagonalized by the Schrieffer-Wolff transformation up to 𝒗/Δ\bm{v}/\Delta as below:

eiξDeiξ[Δ+12Δ(𝝅𝒗)2]ρ3,\displaystyle e^{i\xi}\mathcal{H}_{\text{D}}e^{-i\xi}\simeq\left[\Delta+\frac{1}{2\Delta}(\bm{\pi}\cdot\bm{v})^{2}\right]\rho_{3}, (48)

where ξ=ρ12Δ𝝅𝒗\xi=\frac{\rho_{1}}{2\Delta}\bm{\pi}\cdot\bm{v} is chosen to erase the off-diagonal matrix for the particle-hole space. We can proceed calculation as follows:

(𝝅𝒗)2\displaystyle(\bm{\pi}\cdot\bm{v})^{2} =πiπjwiαwjβ(δαβ+iϵαβγσγ),\displaystyle=\pi_{i}\pi_{j}w_{i\alpha}w_{j\beta}(\delta_{\alpha\beta}+i\epsilon_{\alpha\beta\gamma}\sigma^{\gamma}),
=(πiwiα)2+i2ϵαβγσγ[𝝅×𝝅]iϵijkwjαwkβ,\displaystyle=(\pi_{i}w_{i\alpha})^{2}+\frac{i}{2}\epsilon_{\alpha\beta\gamma}\sigma^{\gamma}[\bm{\pi}\times\bm{\pi}]_{i}\epsilon_{ijk}w_{j\alpha}w_{k\beta},
=Δ(𝝅α𝝅+ecΔiασαBi),\displaystyle=\Delta\left(\bm{\pi}\cdot\alpha\cdot\bm{\pi}+\frac{\hbar e}{c\Delta}\mathcal{M}_{i\alpha}\sigma^{\alpha}B_{i}\right), (49)

where we used (𝝅×𝝅)=eci×𝑨(\bm{\pi}\times\bm{\pi})=\frac{e\hbar}{ci}\nabla\times\bm{A} and MiαM_{i\alpha} is defined as

iα=12ϵαβγϵijkwjβwkγ.\displaystyle\mathcal{M}_{i\alpha}=\frac{1}{2}\epsilon_{\alpha\beta\gamma}\epsilon_{ijk}w_{j\beta}w_{k\gamma}. (50)

Finally, we obtain

eiξDeiξ[Δ+𝝅α𝝅2]Biμs,i,\displaystyle e^{i\xi}\mathcal{H}_{\text{D}}e^{-i\xi}\simeq\left[\Delta+\frac{\bm{\pi}\cdot\tensor{\alpha}\cdot\bm{\pi}}{2}\right]-B_{i}\mu_{{\rm s},i}, (51)

where μs,i\mu_{{\rm s},i} is a magnetic moment of the Dirac electrons defined as

μs,i=e2cΔiαρ3σα=e2cΔiα(σα00σα).\displaystyle\mu_{\text{s},i}=-\frac{\hbar e}{2c\Delta}\mathcal{M}_{i\alpha}\rho_{3}\sigma^{\alpha}=-\frac{\hbar e}{2c\Delta}\mathcal{M}_{i\alpha}\begin{pmatrix}\sigma^{\alpha}&0\\ 0&-\sigma^{\alpha}\end{pmatrix}. (52)

In the main text, we defined the spin operator 𝒔\bm{s} as the magnetic moment 𝝁s\bm{\mu}_{s} divided by the Bohr magnetization μB=e/2mc\mu_{B}=\hbar e/2mc, i.e.,

si=μs,iμB=mΔiα(σα00σα).\displaystyle s^{i}=-\frac{\mu_{{\rm s},i}}{\mu_{\rm B}}=\frac{m}{\Delta}\mathcal{M}_{i\alpha}\begin{pmatrix}\sigma^{\alpha}&0\\ 0&-\sigma^{\alpha}\end{pmatrix}. (53)

For an isotropic Dirac system, the matrix component is given by wiα=vδiαw_{i\alpha}=v\delta_{i\alpha} and Eq. (53) reproduces the well-known form of the spin operator

𝒔=g2(𝝈00𝝈),\displaystyle\bm{s}=\frac{g^{*}}{2}\begin{pmatrix}\bm{\sigma}&0\\ 0&-\bm{\sigma}\end{pmatrix}, (54)

where g=2m/mg^{*}=2m/m^{*} is the effective gg-factor with m=Δ/v2m^{*}=\Delta/v^{2} being effective mass.

Appendix B Linear Response Theory

In this section, we briefly explain how the microwave absorption rate is written in terms of the uniform spin correlation function. The Hamiltonian of an external circular-polarized microwave is written as

^rf\displaystyle\hat{\mathcal{H}}_{\text{rf}} =gμBhrf2i(Sieiωt+Si+eiωt)\displaystyle=-\frac{g\mu_{B}h_{\text{rf}}}{2}\sum_{i}(S_{i}^{-}e^{-i\omega t}+S_{i}^{+}e^{i\omega t})
=gμBhrfnF2(S𝟎eiωt+S𝟎+eiωt),\displaystyle=-\frac{g\mu_{B}h_{\text{rf}}\sqrt{n_{\rm F}}}{2}(S_{\bm{0}}^{-}e^{-i\omega t}+S_{\bm{0}}^{+}e^{i\omega t}), (55)

where hrfh_{\text{rf}} is an amplitude of the magnetic field of the microwave, S𝒌±S_{\bm{k}}^{\pm} are the Fourier transformations defined as

S𝒌±=1nFiSi±ei𝒌𝑹i,\displaystyle S_{\bm{k}}^{\pm}=\frac{1}{\sqrt{n_{\rm F}}}\sum_{i}S_{i}^{\pm}e^{-i{\bm{k}}\cdot{\bm{R}}_{i}}, (56)

and 𝑹i{\bm{R}}_{i} is the position of the locazed spin ii. Using the linear response theory with respect to ^rf\hat{\mathcal{H}}_{\text{rf}}, the expectation value of the local spin is calculated as

S𝟎+ω=G𝟎R(ω)×gμBhrfnF2,\displaystyle\langle S^{+}_{\bm{0}}\rangle_{\omega}=G^{R}_{\bm{0}}(\omega)\times\frac{g\mu_{B}h_{\text{rf}}\sqrt{n_{\rm F}}}{2}, (57)

where G𝒌R(ω)G^{R}_{\bm{k}}(\omega) is the spin correlation function defined in Eq. (20). Since the microwave absorption is determined by the dissipative part of the response function, it is proportional to ImG𝟎R(ω){\rm Im}\,G^{R}_{\bm{0}}(\omega), that reproduces a Lorentzian-type FMR lineshape. As explained in the main text, the change of the linewidth of the microwave absorption, δα\delta\alpha, gives information on spin excitation in the Dirac system via the spin susceptibility as shown in Eq. (22).

Appendix C Spin susceptibility of Dirac electrons

In this section, we give detailed derivation of Eq. (III). The trace part in Eq. (III) is calculated as

tr[s+g𝒌+𝒒(iϵn+iωl)sg𝒌(iϵn)]=[(iϵn+iωl+μ)(iϵn+μ)+Δ2]tr[s+s]tr[s+(𝒌~+𝒒~)𝝈s𝒌~𝝈][(iϵn+iωl+μ)2ϵ𝒌+𝒒2][(iϵn+μ)2ϵ𝒌2],\displaystyle\text{tr}[s^{+}g_{\bm{k}+\bm{q}}(i\epsilon_{n}+i\omega_{l})s^{-}g_{\bm{k}}(i\epsilon_{n})]=\frac{[(i\epsilon_{n}+i\omega_{l}+\mu)(i\epsilon_{n}+\mu)+\Delta^{2}]\text{tr}[s^{+}s^{-}]-\text{tr}[s^{+}\hbar(\tilde{\bm{k}}+\tilde{\bm{q}})\cdot\bm{\sigma}s^{-}\hbar\tilde{\bm{k}}\cdot\bm{\sigma}]}{[(i\epsilon_{n}+i\omega_{l}+\mu)^{2}-\epsilon_{\bm{k}+\bm{q}}^{2}][(i\epsilon_{n}+\mu)^{2}-\epsilon_{\bm{k}}^{2}]}, (58)

where (𝒌~+𝒒~)𝝈=(𝒌+𝒒)𝒗(\tilde{\bm{k}}+\tilde{\bm{q}})\cdot\bm{\sigma}=(\bm{k}+\bm{q})\cdot\bm{v}. Using the following relations

tr[s+s]\displaystyle\text{tr}[s^{+}s^{-}] =(2mΔ)2αaiiαajjα,\displaystyle=\left(\frac{2m}{\Delta}\right)^{2}\sum_{\alpha}a_{i}\mathcal{M}_{i\alpha}a_{j}^{*}\mathcal{M}_{j\alpha}, (59)
tr[s+(𝒌~+𝒒~)𝝈s𝒌~𝝈]\displaystyle\text{tr}[s^{+}\hbar(\tilde{\bm{k}}+\tilde{\bm{q}})\cdot\bm{\sigma}s^{-}\hbar\tilde{\bm{k}}\cdot\bm{\sigma}] =(2mΔ)2α(2aiiαk~αajjβk~β2k~2aiiαajjα),\displaystyle=\left(\frac{2m}{\Delta}\right)^{2}\sum_{\alpha}(2a_{i}\mathcal{M}_{i\alpha}\hbar\tilde{k}_{\alpha}a_{j}^{*}\mathcal{M}_{j\beta}\hbar\tilde{k}_{\beta}-\hbar^{2}\tilde{k}^{2}a_{i}\mathcal{M}_{i\alpha}a_{j}^{*}\mathcal{M}_{j\alpha}), (60)

the spin susceptibility is given by

χ𝒒(iωl)=2(θ,ϕ)𝒌β1iϵn(iϵn+iωl+μ)(iϵn+μ)+Δ2+2k~2/3[(iϵn+iωl+μ)2ϵ𝒌+𝒒2][(iϵn+μ)2ϵ𝒌2],\displaystyle\chi_{\bm{q}}(i\omega_{l})=-2\mathcal{F}(\theta,\phi)\sum_{\bm{k}}\beta^{-1}\sum_{i\epsilon_{n}}\frac{(i\epsilon_{n}+i\omega_{l}+\mu)(i\epsilon_{n}+\mu)+\Delta^{2}+\hbar^{2}\tilde{k}^{2}/3}{[(i\epsilon_{n}+i\omega_{l}+\mu)^{2}-\epsilon_{\bm{k}+\bm{q}}^{2}][(i\epsilon_{n}+\mu)^{2}-\epsilon_{\bm{k}}^{2}]}{\color[rgb]{0,0,1},} (61)

where we dropped the terms proportional to k~αk~β\tilde{k}_{\alpha}\tilde{k}_{\beta} (αβ\alpha\neq\beta) because they vanish after the summation with respect to the wavenumber 𝒌{\bm{k}}. Here, we introduced a dimensionless function, (θ,ϕ)=(2m/Δ)2αaiiαajjα\mathcal{F}(\theta,\phi)=\left(2m/\Delta\right)^{2}\sum_{\alpha}a_{i}\mathcal{M}_{i\alpha}a_{j}^{*}\mathcal{M}_{j\alpha}, which depends on the direction of the magnetization of the FI. Representing the Matsubara summation as the following contour integral, we derive

χ𝒒(iωl)\displaystyle\chi_{\bm{q}}(i\omega_{l}) =2(θ,ϕ)𝒌dz4πitanh(β(zμ)2)z(z+iωl)+Δ2+2k~2/3[(z+iωl)2ϵ𝒌+𝒒2][z2ϵ𝒌2],\displaystyle=-2\mathcal{F}(\theta,\phi)\sum_{\bm{k}}\oint\frac{dz}{4\pi i}\tanh\left(\frac{\beta(z-\mu)}{2}\right)\frac{z(z+i\omega_{l})+\Delta^{2}+\hbar^{2}\tilde{k}^{2}/3}{[(z+i\omega_{l})^{2}-\epsilon_{\bm{k}+\bm{q}}^{2}][z^{2}-\epsilon_{\bm{k}}^{2}]}, (62)
=2(θ,ϕ)𝒌dz2πif(z)z(z+iωl)+Δ2+2k~2/3[(z+iωl)2ϵ𝒌+𝒒2][z2ϵ𝒌2],\displaystyle=2\mathcal{F}(\theta,\phi)\sum_{\bm{k}}\oint\frac{dz}{2\pi i}f(z)\frac{z(z+i\omega_{l})+\Delta^{2}+\hbar^{2}\tilde{k}^{2}/3}{[(z+i\omega_{l})^{2}-\epsilon_{\bm{k}+\bm{q}}^{2}][z^{2}-\epsilon_{\bm{k}}^{2}]}, (63)

We note that tanh(β(zμ)/2)\tanh(\beta(z-\mu)/2) has poles at z=iϵn+μz=i\epsilon_{n}+\mu and is related to the Fermi distribution function f(z)f(z) as tanh[β(zμ)/2]=12f(z)\tanh[\beta(z-\mu)/2]=1-2f(z). Using the following identities

1z2ϵ𝒌2\displaystyle\frac{1}{z^{2}-\epsilon_{\bm{k}}^{2}} =12ϵ𝒌λ=±λzλϵ𝒌,\displaystyle=\frac{1}{2\epsilon_{\bm{k}}}\sum_{\lambda=\pm}\frac{\lambda}{z-\lambda\epsilon_{\bm{k}}}, (64)
zz2ϵ𝒌2\displaystyle\frac{z}{z^{2}-\epsilon_{\bm{k}}^{2}} =12λ=±1zλϵ𝒌,\displaystyle=\frac{1}{2}\sum_{\lambda=\pm}\frac{1}{z-\lambda\epsilon_{\bm{k}}}, (65)

the spin susceptibility is given by

χ𝒒(iωl)\displaystyle\chi_{\bm{q}}(i\omega_{l}) =(θ,ϕ)𝒌dz2πif(z)λ,λ=±[12+(Δ2+2k~2/3)λλ2ϵ𝒌ϵ𝒌+𝒒]1zλϵ𝒌1z+iωlλϵ𝒌+𝒒,\displaystyle=\mathcal{F}(\theta,\phi)\sum_{\bm{k}}\oint\frac{dz}{2\pi i}f(z)\sum_{\lambda,\lambda^{\prime}=\pm}\left[\frac{1}{2}+\frac{(\Delta^{2}+\hbar^{2}\tilde{k}^{2}/3)\lambda\lambda^{\prime}}{2\epsilon_{\bm{k}}\epsilon_{\bm{k}+\bm{q}}}\right]\frac{1}{z-\lambda\epsilon_{\bm{k}}}\frac{1}{z+i\omega_{l}-\lambda^{\prime}\epsilon_{\bm{k}+\bm{q}}}, (66)
=(θ,ϕ)𝒌λ,λ=±[12+λλ62Δ2+ϵ𝒌2ϵ𝒌ϵ𝒌+𝒒]f(λϵ𝒌+𝒒)f(λϵ𝒌)iωlλϵ𝒌+𝒒+λϵ𝒌.\displaystyle=\mathcal{F}(\theta,\phi)\sum_{\bm{k}}\sum_{\lambda,\lambda^{\prime}=\pm}\left[\frac{1}{2}+\frac{\lambda\lambda^{\prime}}{6}\frac{2\Delta^{2}+\epsilon_{\bm{k}}^{2}}{\epsilon_{\bm{k}}\epsilon_{\bm{k}+\bm{q}}}\right]\frac{f(\lambda^{\prime}\epsilon_{\bm{k}+\bm{q}})-f(\lambda\epsilon_{\bm{k}})}{i\omega_{l}-\lambda^{\prime}\epsilon_{\bm{k}+\bm{q}}+\lambda\epsilon_{\bm{k}}}. (67)

By the analytic continuation iωl=ω+iδi\omega_{l}=\hbar\omega+i\delta, we derive the retarded spin susceptibility as below:

χ𝒒R(ω)=(θ,ϕ)𝒌λ,λ=±[12+λλ62Δ2+ϵ𝒌2ϵ𝒌ϵ𝒌+𝒒]f(λϵ𝒌+𝒒)f(λϵ𝒌)ω+iδλϵ𝒌+𝒒+λϵ𝒌.\displaystyle\chi^{R}_{\bm{q}}(\omega)=\mathcal{F}(\theta,\phi)\sum_{\bm{k}}\sum_{\lambda,\lambda^{\prime}=\pm}\left[\frac{1}{2}+\frac{\lambda\lambda^{\prime}}{6}\frac{2\Delta^{2}+\epsilon_{\bm{k}}^{2}}{\epsilon_{\bm{k}}\epsilon_{\bm{k}+\bm{q}}}\right]\frac{f(\lambda^{\prime}\epsilon_{\bm{k}+\bm{q}})-f(\lambda\epsilon_{\bm{k}})}{\hbar\omega+i\delta-\lambda^{\prime}\epsilon_{\bm{k}+\bm{q}}+\lambda\epsilon_{\bm{k}}}. (68)

The imaginary part of the spin susceptibility is given by

Imχ𝒒R(ω)=π(θ,ϕ)𝒌λ,λ=±[12+λλ62Δ2+ϵ𝒌2ϵ𝒌ϵ𝒌+𝒒][f(λϵ𝒌+𝒒)f(λϵ𝒌)]δ(ωλϵ𝒌+𝒒+λϵ𝒌).\displaystyle\text{Im}\chi^{R}_{\bm{q}}(\omega)=-\pi\mathcal{F}(\theta,\phi)\sum_{\bm{k}}\sum_{\lambda,\lambda^{\prime}=\pm}\left[\frac{1}{2}+\frac{\lambda\lambda^{\prime}}{6}\frac{2\Delta^{2}+\epsilon_{\bm{k}}^{2}}{\epsilon_{\bm{k}}\epsilon_{\bm{k}+\bm{q}}}\right]\Bigl{[}f(\lambda^{\prime}\epsilon_{\bm{k}+\bm{q}})-f(\lambda\epsilon_{\bm{k}})\Bigr{]}\delta(\hbar\omega-\lambda^{\prime}\epsilon_{\bm{k}+\bm{q}}+\lambda\epsilon_{\bm{k}}). (69)

From this expression, Eqs. (32) and (33) for the imaginary parts of the uniform and local spin susceptibilities can be obtained by replacing the sum with respect to 𝒌{\bm{k}} and λ\lambda with an integral over the energy ϵ\epsilon as follows:

nD1𝒌,λ𝒜(λϵ𝒌)\displaystyle n_{\text{D}}^{-1}\sum_{\bm{k},\lambda}\mathcal{A}(\lambda\epsilon_{\bm{k}}) d3k~(2π)3Δ3detαij𝒜(λϵ𝒌)=𝑑ϵν(ϵ)𝒜(ϵ),\displaystyle\rightarrow\int\frac{d^{3}\tilde{k}}{(2\pi)^{3}\sqrt{\Delta^{3}\text{det}\alpha_{ij}}}\mathcal{A}(\lambda\epsilon_{\bm{k}})=\int^{\infty}_{-\infty}d\epsilon\nu(\epsilon)\mathcal{A}(\epsilon), (70)

where 𝒜\mathcal{A} is an arbitrary function. Note that the Jacobian of the transformation from 𝒌\bm{k} to 𝒌~\tilde{\bm{k}} is given by det(dki/dk~j)=1/Δ3detαij\text{det}(dk_{i}/d\tilde{k}_{j})=1/\sqrt{\Delta^{3}\text{det}\alpha_{ij}}.

References