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

Unified framework of the microscopic Landau-Lifshitz-Gilbert equation
and its application to Skyrmion dynamics

Fuming Xu§ College of Physics and Optoelectronic Engineering, Shenzhen University, Shenzhen 518060, China    Gaoyang Li§ College of Physics and Optoelectronic Engineering, Shenzhen University, Shenzhen 518060, China    Jian Chen Department of Physics, The University of Hong Kong, Pokfulam Road, Hong Kong, China    Zhizhou Yu School of Physics and Technology, Nanjing Normal University, Nanjing 210023, China    Lei Zhang zhanglei@sxu.edu.cn State Key Laboratory of Quantum Optics and Quantum Optics Devices, Institute of Laser Spectroscopy, Shanxi University, Taiyuan 030006, China Collaborative Innovation Center of Extreme Optics, Shanxi University, Taiyuan 030006, China    Baigeng Wang National Laboratory of Solid State Microstructures and Department of Physics, Nanjing University, Nanjing 210093, China    Jian Wang jianwang@hku.hk College of Physics and Optoelectronic Engineering, Shenzhen University, Shenzhen 518060, China Department of Physics, The University of Hong Kong, Pokfulam Road, Hong Kong, China
Abstract

The Landau-Lifshitz-Gilbert (LLG) equation is widely used to describe magnetization dynamics. We develop a unified framework of the microscopic LLG equation based on the nonequilibrium Green’s function formalism. We present a unified treatment for expressing the microscopic LLG equation in several limiting cases, including the adiabatic, inertial, and nonadiabatic limits with respect to the precession frequency for a magnetization with fixed magnitude, as well as the spatial adiabatic limit for the magnetization with slow variation in both its magnitude and direction. The coefficients of those terms in the microscopic LLG equation are explicitly expressed in terms of nonequilibrium Green’s functions. As a concrete example, this microscopic theory is applied to simulate the dynamics of a magnetic Skyrmion driven by quantum parametric pumping. Our work provides a practical formalism of the microscopic LLG equation for exploring magnetization dynamics.

I Introduction

Single-molecule magnets (SMMs) are mesoscopic magnets with permanent magnetization, which show both classical properties and quantum properties.garg1993 ; friedman1996 ; sangregorio1997 ; wernsdorfer1999 ; wernsdorfer2005 ; ardavan2007 ; filipovic2013 SMMs are appealing due to their potential applications as memory cells and precessing units in spintronic devices.kahn1998 ; timm2012 Transport of SMMs coupled with leads has been investigated both experimentallyheersche2006 ; Jomh2006 ; zyazin2010 ; roch2011 and theoretically.park2002 ; tretiakov2010 ; filipovic2013 ; bode2012 ; Fransson14 ; Hammar16 ; Nikolic2018 ; Nikolic2019 Transport measurements on magnetic molecules such as Mn12\rm{Mn}_{12}heersche2006 and Fe8\rm{Fe}_{8}Jomh2006 revealed interesting phenomena, including peaks in the differential conductance and Coulomb blockades. Dc- and ac-driven magnetization switching and noise as well as the influence on I-V characteristics were discussed in a normal metal/ferromagnet/normal metal structure.tretiakov2010 Current-induced switching of a SMM junction was theoretically studied in the adiabatic regime within the Born-Oppenheimer approximation.bode2012 It was found that magnetic exchange interactions between molecular magnets can be tuned by electric voltage or temperature bias.Fransson14 Transient spin dynamics in a SMM was investigated with generalized spin equation of motion.Hammar17 A microscopic formalism was recently proposed for consistent modeling of coupled atomic magnetization and lattice dynamics.Fransson17

For a SMM with magnetization 𝐌\mathbf{M}, its magnetization dynamics can be semiclassically described by the Landau-Lifshitz-Gilbert (LLG) equation of motionGilbert04 ; foros2005 ; chudnovskiy2008 ; brataas2008 ; foros2009 ; swiebodzinski2010 ; brataas2011

d𝐦dt=γ𝐦×𝐇eff+𝐦×(𝜶d𝐦dt)+𝝉STT,\frac{d\mathbf{m}}{dt}=-\gamma\mathbf{m}\times\mathbf{H}_{\text{eff}}+\mathbf{m}\times(\bm{\alpha}\frac{d\mathbf{m}}{dt})+\bm{\tau}_{\rm{STT}}, (1)

where 𝐦=𝐌/M\mathbf{m}=\mathbf{M}/M is the unit magnetization vector, γ\gamma is the gyromagnetic ratio, and 𝐇eff\mathbf{H}_{\text{eff}} is the effective magnetic field around which the magnet precesses. 𝜶\bm{\alpha} is the Gilbert damping tensor describing the dissipation of the precession, and 𝝉STT\bm{\tau}_{\rm{STT}} is the spin transfer torque due to the misalignment between the magnetization and the transport electron spin.slonczewski1996 ; berger1996 ; tserkovnyak2005 ; ralph2008

The LLG equation is widely adopted to describe magnetization dynamics in the adiabatic limit, where the magnetization precesses slowly and the typical time scale is in the order of nsns. The Gilbert damping term is in general a 3×33\times 3 tensor,tserkovnyak2005 which can be deduced from experimental data, scattering matrix theory,brataas2008 ; brataas2011 or first-principles calculation.Starikov10 ; Bhattacharjee12 ; Mankovsky13 Later, the LLG equation was generalized to study ultrafast dynamics induced by psps electrical pulseothers3 ; Jhuria or fsfs laser pulsekoopmans ; kammerer ; others1 ; others2 , which extends the magnetization switching time down to psps or even sub-psps. This is refereed as the inertial regimefahnle2011 , where the time scale involved is much shorter than that of the adiabatic limit. In the inertial limit, a nonlinear inertial term was introduced into the LLG equationsuhl ; Ciornei ; Li15 ; Olive15 ; Mondal21 , which was applied to simulate ultrafast spin dynamics.Hammar17 ; Mondal16 ; Mondal17 Direct observation of inertial spin dynamics was experimentally realized in ferromagnetic thin films in the form of magnetization nutation at a frequency of 0.5THz\rm~{}0.5~{}THz.Neeraj21 When the magnetization varies in both temporal and spatial domains, two adiabatic spin torques were incorporated into the LLG equationZhang-SF , which can describe the dynamics of magnetic textures such as Skymions.Tserkovnyak2017

Magnetic Skyrmions (Sk) stabilized by the Dzyaloshinskii-Moriya interaction (DMI) or competing interaction between frustrated magnets are topologically nontrivial spin textures showing chiral particle-like nature. When an electron traverses the Sk, it acquires a Berry phase and experiences a Lorentz-like force, leading to the topological Hall effectTHE . At the same time, the Magnus force due to the back action on Sk gives rise to Skyrmion Hall effectJiang ; Litzius . The existence of Sk has been verified in magnetic materials including MnSiMuhlbauer and PdFe/Ir(111)Wiesendanger . The radius of an Sk can be as small as a few nmnmHeinze ; Romming and is stable even at room temperaturesMoreau-Luchaire ; Woo . Sk can be operated at ultralow current density,Jonietz ; Yu ; YanZhou20 which makes it promising in spintronic applications including the magnetic memory and logic gates.Fert ; Zhang Various investigations show that Sk can be manipulated by spin torque due to the charge/spin current injectionJonietz ; Yu , external electric field,Upadhyaya ; White magnetic field gradientWang2017 , temperature gradient,Kong ; Mochizuki ; Yanzhou20PRL ; YanZhou22 and strain,Shibata etc. However, Sk driven by quantum parametric pumping has not been explored.

Quantum parametric pump refers to such a process: in an open system without bias voltages, cyclic variation of system parameters can give rise to a net dc current per cyclebrouwer ; Aleiner ; Altshuler ; Switkes ; Zhou-F ; Shutenko ; Wei-YD ; xu2011 ; xu2023 . In the adiabatic limit, this quantum parametric pump requires at least two pumping parameters with a phase difference and the pumped current is proportional to the area enclosed by the trajectory of pumping parameters in parameter space.brouwer It was found that the adiabatic pumped current is related to Berry phase.Avron Beyond the adiabatic limit, the cyclic frequency may serve as another dimension in parameter space and hence a single-parameter quantum pump is possible at finite frequences.Vavilov ; Wang-BG1 In general, quantum parametric pump can be formulated in terms of photon-assisted transport.Buttiker1 ; Wang-BG2 Quantum parametric pump can also generate heat currentButtiker2 ; Wang-BG3 , whose lower bound is Joule heating during the pumping process. This defines an optimal quantum pumpAvron1 ; Wang-BG4 that is noiseless and pumps out quantized charge per cycleLevinson ; Wang-BG5 ; xu2022 . Quantum parametric pumping theory has been extended to account for Andreev reflection in the presence of superconducting leadWang-J2001 , correlated charge pumpSharma ; Sun-QF1 , and parametric spin pumpWang-BG6 ; Zheng-W , providing more physical insights. It is interesting to generalize quantum parametric pump to Skyrmion transport, which may offer new operating paradigms for spintronic devices.

In this work, we investigate the microscopic origin of the LLG equation and the Gilbert damping. We focus on several limiting cases of the LLG equation. For a magnetization with fixed magnitude, the adiabatic, inertial, and nonadiabatic limits with respect to its precession frequency are discussed. When both the magnitude and direction of a magnetization vary slowly in space, which is referred as the adiabatic limit in spatial domain, our formalism can also be extended to cover this limit. We will provide a unified treatment of all these cases and explicitly express each term in the microscopic LLG equation in the language of nonequilibrium Green’s functions. As an example, we apply the microscopic LLG equation to simulate the dynamics of a Skyrmion driven by quantum parametric pumping in a two-dimensional (2D) system.

This paper is organized as follows. In Sec.II, a single-molecule magnet (SMM) transport setup and corresponding Hamiltonians are introduced. In Sec.III, a stochastic Langevin equation for magnetization dynamics is derived from the equation of motion by separating fast (electron) and slow (magnetization) degrees of freedom, forming a microscopic version of the LLG equation. In Sec.IV, four limiting cases of the microscopic LLG equation are discussed. In Sec.V, we numerically study Sk transport driven by quantum parametric pumping. Finally, a brief summary is given in Sec.VI.

II Model

The model system under investigation is shown in Fig.1, where a noninteracting quantum dot (QD) representing a single-molecule magnet (SMM) with magnetization 𝐌\mathbf{M} is connected to two leads. A uniform magnetic field 𝐁=Be^z\mathbf{B}=B\hat{e}_{z} is applied in the central region. In addition, we assume that there is a dc bias or spin bias across the system providing a spin transfer torque or spin orbit torque.

Refer to caption
Figure 1: Sketch of the model system. A single-molecule magnet (SMM) represented by the quantum dot (QD) is connected to the left and right leads. A uniform magnetic field is applied in the central region, around which the SMM magnetization precesses.

The Hamiltonian of this system is given by (=1\hbar=1)

H^total=H^L+H^R+H^D+H^T,\hat{H}_{\rm total}=\hat{H}_{L}+\hat{H}_{R}+\hat{H}_{D}+\hat{H}_{T},

with the lead Hamiltonian (α=L,R\alpha=L,R),

H^α=kσϵkασc^kασc^kασ,\hat{H}_{\alpha}=\sum_{k\sigma}\epsilon_{k\alpha\sigma}\hat{c}^{\dagger}_{k\alpha\sigma}\hat{c}_{k\alpha\sigma}, (2)

and the Hamiltonian of the central region,

H^D=H^0+H^+γ𝐌^𝐁.\hat{H}_{D}=\hat{H}_{0}+\hat{H}^{\prime}+\gamma\hat{\mathbf{M}}\cdot\mathbf{B}. (3)

Here H^0\hat{H}_{0} is the Hamiltonian of the QD with spin-orbit interaction (SOI)sun2005

H^0=nσϵnσd^nσd^nσ+mn(tnmSOdmdn+H.c.),\hat{H}_{0}=\sum_{n\sigma}\epsilon_{n\sigma}\hat{d}^{\dagger}_{n\sigma}\hat{d}_{n\sigma}+\sum_{mn}(t^{SO}_{nm}d^{\dagger}_{m\uparrow}d_{n\downarrow}+\rm{H.c.}), (4)

with tnmSO=tmnSOt^{SO}_{nm}=-t^{SO}_{mn}. H^\hat{H}^{\prime} is the interaction between the electron spin and the magnetization as well as the magnetic field,

H^=Jn𝐬^n𝐌^+γen𝐬^n𝐁.\hat{H}^{\prime}=J\sum_{n}\hat{\mathbf{s}}_{n}\cdot\hat{\mathbf{M}}+\gamma_{e}\sum_{n}\hat{\mathbf{s}}_{n}\cdot\mathbf{B}.

We can also add uniaxial anisotropy field to H^\hat{H}^{\prime}. The coupling Hamiltonian between the QD and the leads is

H^T=kαn,σσ[tkαnσσc^kασd^nσ+H.c.].\hat{H}_{T}=\sum_{k\alpha n,\sigma\sigma^{\prime}}[t_{k\alpha n}^{\sigma\sigma^{\prime}}\hat{c}^{\dagger}_{k\alpha\sigma}\hat{d}_{n\sigma^{\prime}}+\rm{H.c.}]. (5)

In the above equations, d^nσ\hat{d}^{\dagger}_{n\sigma} (c^kασ\hat{c}^{\dagger}_{k\alpha\sigma}) creates an electron with energy ϵnσ\epsilon_{n\sigma} (ϵkασ\epsilon_{k\alpha\sigma}) in the QD (lead α\alpha). In general, the leads can be metallic or ferromagnetic. Here 𝐬^n=12ψn𝝈ψn\hat{\mathbf{s}}_{n}=\frac{1}{2}\psi_{n}^{\dagger}\bm{\sigma}\psi_{n} is the electron spin in the central region, with ψn=(dn,dn)\psi^{\dagger}_{n}=(d^{\dagger}_{n\uparrow},d^{\dagger}_{n\downarrow}). The Pauli matrices satisfy [σx,σy]=2iσz[\sigma_{x},\sigma_{y}]=2i\sigma_{z}, and the magnetization 𝐌\mathbf{M} follows the commutation relation [Mx,My]=iMz[M_{x},M_{y}]=i\hbar M_{z}. JJ is the exchange interaction between the magnetization and the spin of conducting electrons. γ\gamma (γe\gamma_{e}) is the gyromagnetic ratio of the magnet (electron).

If we choose the magnetic field in the zz direction as the laboratory frame, and (θ,ϕ)(\theta,\phi) the polar and azimuthal angles of the magnetization, the spin dependent coupling matrix is given by

tkαnσσ=[R^tkαn]σσ,t_{k\alpha n}^{\sigma\sigma^{\prime}}=\big{[}\hat{R}t_{k\alpha n}\big{]}_{\sigma\sigma^{\prime}}, (6)

with R^\hat{R} the rotational operatorkamenev2011field

R^=eiθ2σ^yeiϕ2σ^z=(eiϕ2cos(θ2)eiϕ2sin(θ2)eiϕ2sin(θ2)eiϕ2cos(θ2)).\hat{R}=e^{-i\frac{\theta}{2}\hat{\sigma}_{y}}e^{-i\frac{\phi}{2}\hat{\sigma}_{z}}=\begin{pmatrix}e^{-i\frac{\phi}{2}}\cos(\frac{\theta}{2})&-e^{i\frac{\phi}{2}}\sin(\frac{\theta}{2})\\ e^{-i\frac{\phi}{2}}\sin(\frac{\theta}{2})&e^{i\frac{\phi}{2}}\cos(\frac{\theta}{2})\end{pmatrix}. (7)

III magnetization dynamics

From the Heisenberg equation of motion, the magnetization dynamics in the central region is governed by

𝐌^˙=γ𝐌^×𝐁J𝐌^×𝐬^D,\dot{\hat{\mathbf{M}}}=-\gamma\hat{\mathbf{M}}\times\mathbf{B}-J\hat{\mathbf{M}}\times\hat{\mathbf{s}}_{D}, (8)

where 𝐬^D=n𝐬^n\hat{\mathbf{s}}_{D}=\sum_{n}\hat{\mathbf{s}}_{n} is the total electron spin. In deriving the above equation, the following relation is used:

[𝝈^,𝝈^𝐀]=2i𝝈^×𝐀.[\hat{\bm{\sigma}},\hat{\bm{\sigma}}\cdot\mathbf{A}]=-2i\hat{\bm{\sigma}}\times\mathbf{A}. (9)

Now we separate an operator into its quantum average and its fluctuation, then 𝐬^D=𝐬^D+δ𝐬^D\hat{\mathbf{s}}_{D}=\langle\hat{\mathbf{s}}_{D}\rangle+\delta\hat{\mathbf{s}}_{D}, and 𝐌^=𝐌^+δ𝐌^\hat{\mathbf{M}}=\langle\hat{\mathbf{M}}\rangle+\delta\hat{\mathbf{M}}, where δ𝐬^D\delta\hat{\mathbf{s}}_{D} (δ𝐌^\delta\hat{\mathbf{M}}) is the fluctuation of the electron (magnet) spin. We can transform Eq. (8) into a Langevin equation. For the expectation value 𝐌(t)=𝐌^(t)\mathbf{M}(t)=\langle\hat{\mathbf{M}}(t)\rangle,timescale

𝐌˙=𝐌×[γ𝐁J𝐬D+δ^],\dot{\mathbf{M}}=\mathbf{M}\times[-\gamma\mathbf{B}-J\mathbf{s}_{D}+\delta\mathcal{{\hat{B}}}], (10)

or

𝐌˙=γ𝐌×[𝐇effδ^],\dot{\mathbf{M}}=-\gamma\mathbf{M}\times[\mathbf{H}_{\text{eff}}-\delta\mathcal{{\hat{B}}}^{{}^{\prime}}],

where

𝐬D=𝐬^D=i2Tr[𝝈G<(t,t)].\mathbf{s}_{D}=\langle\hat{\mathbf{s}}_{D}\rangle=-\frac{i}{2}\mathrm{Tr}[\bm{\sigma}G^{<}(t,t)]. (11)

Here Gijσσ<(t,t)=idjσ(t)diσ(t)G^{<}_{ij\sigma\sigma^{\prime}}(t^{\prime},t)=i\langle d^{\dagger}_{j\sigma^{\prime}}(t^{\prime})d_{i\sigma}(t)\rangle is the lesser Green’s function of electrons, which will be discussed in detail below. The effective magnetic field 𝐇eff\mathbf{H}_{\text{eff}} is defined as the variation of the free energy of the system with respect to the magnetizationberkov2007 ; gilmore2008 ; tserkovnyak2005

𝐇eff=1γδHtotalδ𝐌.\mathbf{H}_{\text{eff}}=\frac{1}{\gamma}\frac{\delta H_{\rm total}}{\delta\mathbf{M}}. (12)

And δ^=γδ^\delta\mathcal{{\hat{B}}}=\gamma\delta\mathcal{{\hat{B}}}^{{}^{\prime}} contributes from the fluctuations

𝐌×δ^\displaystyle{\mathbf{M}}\times\delta\mathcal{{\hat{B}}} =\displaystyle= δ𝐌^˙γδ𝐌^×𝐁J𝐌×δ𝐬^D\displaystyle-\delta\dot{\hat{\mathbf{M}}}-\gamma\delta\hat{\mathbf{M}}\times{\bf{B}}-J{\mathbf{M}}\times\delta\hat{\mathbf{s}}_{D}
Jδ𝐌^×𝐬DJδ𝐌^×δ𝐬^D.\displaystyle-J\delta\hat{\mathbf{M}}\times{\mathbf{s}}_{D}-J\delta\hat{\mathbf{M}}\times\delta\hat{\mathbf{s}}_{D}.

These fluctuations can play an important role in determining the motion of the magnetization, such as reducing or enhancing the threshold bias of magnetization switching.bode2012

To transform Eq. (10) into the usual LLG equation, we further separate 𝐬D\mathbf{s}_{D} in Eq. (11) into the time-reversal symmetric and antisymmetric components, 𝐬Ds\mathbf{s}^{s}_{D} and 𝐬Da\mathbf{s}^{a}_{D}. Then 𝐌×𝐬Ds\mathbf{M}\times\mathbf{s}^{s}_{D} and 𝐌×𝐬Da\mathbf{M}\times\mathbf{s}^{a}_{D} correspond to the dissipative and dissipativeless terms, respectively. Thus, Eq. (10) is rewritten as

𝐌˙=γ𝐌×𝐁J𝐌×𝐬DaJ𝐌×𝐬Ds.\dot{\mathbf{M}}=-\gamma\mathbf{M}\times\mathbf{B}-J\mathbf{M}\times\mathbf{s}^{a}_{D}-J\mathbf{M}\times\mathbf{s}^{s}_{D}. (13)

Note that the last term in Eq. (13), 𝐌×𝐬Ds{\bf M}\times\mathbf{s}_{D}^{s}, corresponds to the damping of magnetization. As will be discussed below that in the adiabatic approximation, it assumes the form 𝐌×(𝜶𝐌˙)\mathbf{M}\times({\bm{\alpha}}\dot{\mathbf{M}}) where 𝜶\bm{\alpha} is the Gilbert damping tensor which is expressed in terms of nonequilibrium Green’s function (see Eq. (19)). The second term in Eq. (13), 𝐌×𝐬Da{\bf M}\times\mathbf{s}_{D}^{a}, corresponds to the spin transfer torque. In the presence of SOI, 𝐌×𝐬Da{\bf M}\times\mathbf{s}_{D}^{a} is the spin orbit torque in collinear ferromagnetic systems, which has field-like and damping-like components, respectively, along the directions 𝐌×𝐮{\bf M}\times{\bf u} and 𝐌×(𝐌×𝐮){\bf M}\times({\bf M}\times{\bf u}) with 𝐮𝐌=0{\bf u}\cdot{\bf M}=0. Here 𝐮{\bf u} is the unit vector of the spin current.slonczewski1996

IV Microscopic LLG equation in different limits

In this section, we will drive the LLG equation and express the Gilbert damping tensor in terms of the nonequilibrium Green’s functions. We also discuss the fluctuation in the equation of motion and the spin continuity equation, showing that the spin transfer torque is insufficient to describe magnetization dynamics in general conditions.

We focus on several limiting cases of the microscopic LLG equation (Eq. (13)). These cases correspond to different limits: (1) Adiabatic limit in temporal domain where the precessing frequency of the magnet is low and 𝐬D\mathbf{s}_{D} can be expanded up to the first order in frequency; (2) Inertial regime where the time scale is much shorter than that of the adiabatic limit, e.g., magnetization switching in psps or even sub-psps rangeothers3 ; Jhuria ; koopmans ; kammerer ; others1 ; others2 ; (3) Nonadiabatic regime where adiabatic approximation in temporal domain is removed. We will work on the linear coupling between the magnetization and the environment,Sayad15 and derive the Gilbert damping coefficient as a function of the precessing frequency; (4) In the above situations, we have assumed that the magnetization has fixed magnitude and only its direction varies in space. Our theory can be easily extended to address the motion of domain walls where the magnetization is nonuniform. In the simplest case, we assume that the magnetization varies slowly in space so that adiabatic approximation in spatial domain can be taken. In this spatial adiabatic limit, two additional toques are incorporated into the LLG equation which are naturally obtained in our theory.

IV.1 Adiabatic limit

As the magnetization precesses, the electron spin and hence spin-orbit energy of each state changes tserkovnyak2005 ; gilmore2008 , which drives the system out of equilibrium. In the language of frozen Green’s functions (Eqs. (41) and (44)), total spin of the QD (Eqs. (11)) can be expanded in terms of the precession frequency, which consists of two parts: the quasi-static part 𝐬D(0){\bf s}_{D}^{(0)}, and the adiabatic change 𝐬D(1){\bf s}_{D}^{(1)} to the first order in frequency

𝐬D=𝐬D(0)+𝐬D(1),{\bf s}_{D}={\bf s}_{D}^{(0)}+{\bf s}_{D}^{(1)},

where

𝐬D(0)=i2dE2πTr[𝝈Gf<],{\bf s}_{D}^{(0)}=-\frac{i}{2}\int\frac{dE}{2\pi}\mathrm{Tr}[{\bm{\sigma}}G_{f}^{<}], (14)

and

𝐬D(1)=14dE2πTr[Gf<𝝈GfrGfr𝝈GfaGfa𝝈Gf<𝝈]𝐛˙,{\bf s}_{D}^{(1)}=-\frac{1}{4}\int\frac{dE}{2\pi}\mathrm{Tr}[G_{f}^{<}\bm{\sigma}G_{f}^{r}G_{f}^{r}\bm{\sigma}-G_{f}^{a}G_{f}^{a}\bm{\sigma}G_{f}^{<}\bm{\sigma}]\dot{\bf{b}}, (15)

with 𝐛˙=JM𝐦˙\dot{\mathbf{b}}=JM\dot{\mathbf{m}}.

Concerning the magnetization dynamics, the effective field 𝐇eff(t)\mathbf{H}_{\text{eff}}(t) (Eq. (12)) can be separated into two contributions: an anisotropy field and a damping field gilmore2008 ,

𝐇eff(t)=𝐇effani(t)+𝐇effdamp(t),\mathbf{H}_{\text{eff}}(t)=\mathbf{H}_{\text{eff}}^{\text{ani}}(t)+\mathbf{H}_{\text{eff}}^{\text{damp}}(t), (16)

with

𝐇effani(t)=𝐁+(J/γ)𝐬D(0),𝐇effdamp(t)=(J/γ)𝐬D(1).\mathbf{H}_{\text{eff}}^{\text{ani}}(t)=\mathbf{B}+(J/\gamma)\mathbf{s}_{D}^{(0)},~{}~{}~{}~{}\mathbf{H}_{\text{eff}}^{\text{damp}}(t)=(J/\gamma)\mathbf{s}_{D}^{(1)}. (17)

Substituting Eqs. (14) and (15) into Eq. (13), ignoring the fluctuation, and noting that 𝐛˙=JM𝐦˙\dot{\mathbf{b}}=JM\dot{\mathbf{m}}, we obtain the deterministic Landau-Lifshitz-Gilbert equation,

d𝐦dt=γ𝐦×𝐇effani𝐦×(𝜶𝐦˙),\frac{d\mathbf{m}}{dt}=-\gamma\mathbf{m}\times\mathbf{H}^{\text{ani}}_{\text{eff}}-\mathbf{m}\times({\bm{\alpha}}\dot{\mathbf{m}}), (18)

where

𝐦=𝐌/M=(sinθcosϕ,sinθsinϕ,cosθ),\mathbf{m}=\mathbf{M}/M=\big{(}\sin\theta\cos\phi,\sin\theta\sin\phi,\cos\theta\big{)},

is the unit vector in the magnetization direction. α\bf{\alpha} is the 3×33\times 3 Gilbert damping tensorfahnle2011 ; brataas2008 , which is defined in terms of the frozen Green’s functions:

αij=(JM)24dE2πRe{Tr[Gf<σiGfrGfrσj]}.\alpha_{ij}=\frac{(JM)^{2}}{4}\int\frac{dE}{2\pi}\mathrm{Re}\{\mathrm{Tr}[G_{f}^{<}\sigma_{i}G_{f}^{r}G_{f}^{r}\sigma_{j}]\}. (19)

As shown in Appendix D, this damping tensor recovers that obtained in Ref. [brataas2008, ] via the scattering matrix theory in the limit of zero temperature and in the absence of external bias. In general, the Gilbert damping tensor depends on 𝐦(𝐭){\bf m(t)} and bias voltage through the frozen Green’s functions GfrG^{r}_{f} and Gf<G^{<}_{f}. This agrees with the observation in Ref. [steiauf, ] using the effective field theory of breathing Fermi surface mode.

IV.2 Inertial regime

In this regime, the magnetization has both precessional and nutational motions. We focus on the linear coupling between the magnetization and the environment so that an additional “inertial” term enters the LLG equation, which describes the nutation of the magnet. In this case, the adiabatic approximation is not good enough. One has to expand the spin density 𝐬D{\bf s}_{D} at least to the second order in frequency. In the inertial regime, we assume that the magnitude of the magnetization is fixed while only its direction varies. Iterating Eq. (44) to the second order in frequency, we have

Gr=GfriGfrG˙fr+Gfr(G˙fr)2+(Gfr)2G¨fr,G^{r}=G^{r}_{f}-iG^{r}_{f}{\dot{G}}^{r}_{f}+G^{r}_{f}({\dot{G}}^{r}_{f})^{2}+(G^{r}_{f})^{2}{\ddot{G}}^{r}_{f},

from which we find the contribution 𝐬D(2){\bf s}^{(2)}_{D} in the inertial limit,

𝐬D(2)\displaystyle{\bf s}_{D}^{(2)} =\displaystyle= 18dE2πIm{Tr[Gf<𝝈(Gfr)3𝝈]}t2𝐛\displaystyle-\frac{1}{8}\int\frac{dE}{2\pi}{\rm Im}\{\mathrm{Tr}[G_{f}^{<}\bm{\sigma}(G_{f}^{r})^{3}\bm{\sigma}]\}\cdot\partial^{2}_{t}{\bf{b}} (20)
+\displaystyle+ 116dE2πIm{Tr[Gf<𝝈(Gfr)4(𝝈t𝐛)2]}\displaystyle\frac{1}{16}\int\frac{dE}{2\pi}{\rm Im}\{\mathrm{Tr}[G_{f}^{<}\bm{\sigma}(G_{f}^{r})^{4}(\bm{\sigma}\cdot\partial_{t}{\bf{b}})^{2}]\}
\displaystyle- i8dE2πTr[𝝈(Gfr)2Gf<(Gfa)2(𝝈t𝐛)2].\displaystyle\frac{i}{8}\int\frac{dE}{2\pi}\mathrm{Tr}[\bm{\sigma}(G_{f}^{r})^{2}G^{<}_{f}(G_{f}^{a})^{2}(\bm{\sigma}\cdot\partial_{t}{\bf{b}})^{2}].

To make comparison with Ref. [fahnle2011, ], we keep only the linear term t2𝐦\partial^{2}_{t}{\bf m} and neglect other nonlinear terms such as (t𝐦)2(\partial_{t}{\bf m})^{2}. With this new term, the LLG equation in inertial regime is written assuhl ; fahnle2011 ; Li15 ; Ciornei ; Tserkovnyak2017

d𝐦dt=γ𝐦×𝐇effani𝐦×[𝜶d𝐦dt]𝐦×[𝜶¯d2𝐦dt2],\frac{d\mathbf{m}}{dt}=-\gamma\mathbf{m}\times\mathbf{H}^{\text{ani}}_{\text{eff}}-\mathbf{m}\times[\bm{\alpha}\frac{d\mathbf{m}}{dt}]-\mathbf{m}\times[\bm{{\bar{\alpha}}}\frac{d^{2}\mathbf{m}}{dt^{2}}], (21)

where the inertial term is a 3×33\times 3 tensor given by

α¯ij=(JM)22dE2πIm{Tr[Gf<σi(Gfr)3σj]}.{\bar{\alpha}}_{ij}=\frac{(JM)^{2}}{2}\int\frac{dE}{2\pi}\mathrm{Im}\{\mathrm{Tr}[G_{f}^{<}\sigma_{i}(G_{f}^{r})^{3}\sigma_{j}]\}. (22)

This additional inertial term has been obtained both phenomenologicallysuhl and semiclassicallyfahnle2011 . Ref. [Bhattacharjee12, ] proposed a first-principles method for calculating the inertia term in the semiclassical limit. Here we derive the quantum inertial tensor in terms of the frozen Green’s function.

IV.3 Nonadiabatic regime

Now we consider the magnetization dynamics at finite precession frequency, whose time scale is still much larger than that of electrons. Since no analytic solution exists in general conditions, we only focus on the linear coupling in the exchange interaction JJ. In this nonadiabatic regime, we treat the coupling JJ as a small perturbation, and rewrite the equation determining the Green’s function as

(itH~0H(t)Σr)Gr(t,t)=δ(tt),\big{(}i\frac{\partial}{\partial t}-\tilde{H}_{0}-H^{\prime}(t)-\Sigma^{r}\big{)}G^{r}(t,t^{\prime})=\delta(t-t^{\prime}), (23)

where H~0=H0+γe𝐬D𝐁\tilde{H}_{0}=H_{0}+\gamma_{e}{\bf s}_{D}\cdot\mathbf{B} is the unperturbed Hamiltonian, including the bare Hamiltonian of the QD defined in Eq. (4) and the Hamiltonian due to the constant external field. H(t)=J𝝈𝐌(t)H^{\prime}(t)=J\bm{\sigma}\cdot\mathbf{M}(t) is the perturbative term due to exchange coupling between the magnetization and the electron spin.

The unperturbed retarded Green’s function satisfies

(itH~0Σr)G0r(tt)=δ(tt).(i\frac{\partial}{\partial t}-\tilde{H}_{0}-\Sigma^{r})G^{r}_{0}(t-t^{\prime})=\delta(t-t^{\prime}). (24)

Since G0r(tt)G^{r}_{0}(t-t^{\prime}) only depends on the time difference, it is convenient to work in the energy representation,

G0r(E)=[EH0γe2𝝈𝐁Σr]1,G^{r}_{0}(E)=[E-H_{0}-\frac{\gamma_{e}}{2}\bm{\sigma}\cdot\mathbf{B}-\Sigma^{r}]^{-1}, (25)

where G0r(tt)G^{r}_{0}(t-t^{\prime}) and Gr(t,t)G^{r}(t,t^{\prime}) are related through the Dyson equation

Gr(t,t)=G0r(tt)+𝑑t1G0r(tt1)H(t1)Gr(t1,t).G^{r}(t,t^{\prime})=G^{r}_{0}(t-t^{\prime})+\int dt_{1}G^{r}_{0}(t-t_{1})H^{\prime}(t_{1})G^{r}(t_{1},t^{\prime}).

In the first order perturbation, we have

Gr=G0r+G0rHG0r,G^{r}=G^{r}_{0}+G^{r}_{0}H^{\prime}G^{r}_{0},

and

G<=G0<+G0rHG0<+G0<HG0a.G^{<}=G^{<}_{0}+G^{r}_{0}H^{\prime}G^{<}_{0}+G^{<}_{0}H^{\prime}G^{a}_{0}.

Using

𝑑t1G0r(tt1)H(t1)G0<(t1t)=dE2πdω2πeiωtG0r(E+ω)H(ω)G0<(E),\begin{split}&\ \ \ \ \int dt_{1}G^{r}_{0}(t-t_{1})H^{\prime}(t_{1})G^{<}_{0}(t_{1}-t)\\ &=\int\frac{dE}{2\pi}\int\frac{d\omega}{2\pi}e^{-i\omega t}G^{r}_{0}(E+\omega)H^{\prime}(\omega)G^{<}_{0}(E),\end{split}

and

𝑑t1G0<(tt1)H(t1)G0a(t1t)=dE2πdω2πeiωtG0<(E+ω)H(ω)G0a(E),\begin{split}&\ \ \ \ \int dt_{1}G^{<}_{0}(t-t_{1})H^{\prime}(t_{1})G^{a}_{0}(t_{1}-t)\\ &=\int\frac{dE}{2\pi}\int\frac{d\omega}{2\pi}e^{-i\omega t}G^{<}_{0}(E+\omega)H^{\prime}(\omega)G^{a}_{0}(E),\end{split}

where H(ω)=J2𝝈𝐌(ω)H^{\prime}(\omega)=\frac{J}{2}\bm{\sigma}\cdot\mathbf{M}(\omega) with ω\omega the precession frequency, the spin density 𝐬D\mathbf{s}_{D} of the quantum dot can be evaluated

𝐬D=𝐬D(0)+𝐬D(1).{\bf s}_{D}={\bf s}_{D}^{(0)}+{\bf s}_{D}^{(1)}. (26)

Here 𝐬D(0){\bf s}_{D}^{(0)} is independent of 𝐌\mathbf{M} and time tt:

𝐬D(0)=i2dE2πTr[𝝈G0<(E)].{\bf s}_{D}^{(0)}=-\frac{i}{2}\int\frac{dE}{2\pi}\mathrm{Tr}[\bm{\sigma}G_{0}^{<}(E)].

And 𝐬D(1){\bf s}_{D}^{(1)} depends linearly on 𝐌(t)\mathbf{M}(t)

𝐬D(1)=i2dE2πdω2πeiωtTr[𝝈G0r(E+ω)H(ω)G0<(E)+𝝈G0<(E+ω)H(ω)G0a(E)].\begin{split}{\bf s}_{D}^{(1)}=&-\frac{i}{2}\int\frac{dE}{2\pi}\int\frac{d\omega}{2\pi}e^{-i\omega t}\mathrm{Tr}[\bm{\sigma}G^{r}_{0}(E+\omega)H^{\prime}(\omega)G^{<}_{0}(E)\\ &+\bm{\sigma}G^{<}_{0}(E+\omega)H^{\prime}(\omega)G^{a}_{0}(E)].\end{split}

Using the anisotropic field 𝐇effani(t)\mathbf{H}_{\text{eff}}^{\text{ani}}(t) and the damping field 𝐇effdamp(t)\mathbf{H}_{\text{eff}}^{\text{damp}}(t) expressed in Eq. (17) and ignoring the fluctuations, we can obtain a deterministic dynamic equation from Eq. (13):

d𝐦dt=γ𝐦×𝐇effaniγ𝐦×𝐇effdamp,\frac{d\mathbf{m}}{dt}=-\gamma\mathbf{m}\times\mathbf{H}^{\text{ani}}_{\text{eff}}-\gamma\mathbf{m}\times\mathbf{H}_{\text{eff}}^{\text{damp}}, (27)

where

𝐇effani(t)=𝐁iJ2γdE2πTr[𝝈G0<(E)],\mathbf{H}_{\text{eff}}^{\text{ani}}(t)=\mathbf{B}-\frac{iJ}{2\gamma}\int\frac{dE}{2\pi}\mathrm{Tr}[\bm{\sigma}G_{0}^{<}(E)], (28)

and

𝐇effdamp(t)=dω2πeiωt𝐦(ω)α~(ω).\mathbf{H}_{\text{eff}}^{\text{damp}}(t)=-\int\frac{d\omega}{2\pi}e^{-i\omega t}\mathbf{m}({\omega}){\tilde{\alpha}}(\omega). (29)

Here α~(ω){\tilde{\alpha}}(\omega) is the frequency dependent Gilbert damping tensor defined as

α~(ω)\displaystyle{\tilde{\alpha}}(\omega) =\displaystyle= i4J2M2dE2πTr[G0<(E)𝝈G0r(E+ω)𝝈\displaystyle\frac{i}{4}J^{2}M^{2}\int\frac{dE}{2\pi}\mathrm{Tr}\biggl{[}G^{<}_{0}(E)\bm{\sigma}G^{r}_{0}(E+\omega)\bm{\sigma} (30)
+\displaystyle+ G0a(E)𝝈G0<(E+ω)𝝈].\displaystyle G^{a}_{0}(E)\bm{\sigma}G^{<}_{0}(E+\omega)\bm{\sigma}\biggr{]}.

It is easy to confirm that when ω\omega goes to zero, we can recover the results in the adiabatic and inertial limits.

IV.4 Adiabatic limit in spatial domain

When both the magnitude and direction of the magnetization vary slowly in space, we refer to this situation as the adiabatic limit in spatial domain. In this case, two additional terms emerge in the LLG equationZhang-SF ,

d𝐦dt\displaystyle\frac{d\mathbf{m}}{dt} =\displaystyle= γ𝐦×𝐇effani𝐦×[𝜶d𝐦dt]\displaystyle-\gamma\mathbf{m}\times\mathbf{H}^{\text{ani}}_{\text{eff}}-\mathbf{m}\times[\bm{\alpha}\frac{d\mathbf{m}}{dt}] (31)
+\displaystyle+ bJ(𝒋e)𝐦cJ𝐦×(𝒋e)𝐦,\displaystyle b_{J}({\bm{j}}_{e}\cdot\nabla)\mathbf{m}-c_{J}\mathbf{m}\times({\bm{j}}_{e}\cdot\nabla)\mathbf{m},

where bJb_{J} and cJc_{J} are constants defined in Ref. [Zhang-SF, ]. Here the term with coefficient bJb_{J} is related to the adiabatic process of the nonequilibrium conducting electrons.Zhang-SF In contrast, the other term with coefficient cJc_{J} corresponds to the nonadiabatic process which changes sign upon time-reversal operation.

In this limit, the coupling between the magnetization and the electron spin can be approximated as

H^=J𝐬^𝐫𝐌^(𝐫,t)+γe𝐬^𝐫𝐁.\hat{H}^{\prime}=J\hat{\mathbf{s}}_{\bf r}\cdot\hat{\mathbf{M}}({\bf r},t)+\gamma_{e}\hat{\mathbf{s}}_{\bf r}\cdot\mathbf{B}. (32)

Eqs. (13), (46), and (47) are still valid except that 𝐌{\bf M} and 𝒔D{\bm{s}}_{D} are local variables depending on position, where 𝒔D(x){\bm{s}}_{D}(x) is defined as

𝐬D(x)=i2dE2πTrs[𝝈G<]xx,{\bf s}_{D}(x)=-\frac{i}{2}\int\frac{dE}{2\pi}\mathrm{Tr}_{s}[{\bm{\sigma}}G^{<}]_{xx}, (33)

where the trace is taken only in spin space.

To derive the adiabatic term in Eq. (31), we start from Eq. (46) and then use Eq. (47). From Eq. (46), we havenote9

d𝐬Ddt+𝒋s=J𝐌×𝐬D,\frac{d{\mathbf{s}}_{D}}{dt}+\nabla\cdot{\bm{j}}_{s}=J{\mathbf{M}}\times{\mathbf{s}}_{D}, (34)

where 𝒋s{\bm{j}}_{s} is the spin current density and the term γe𝐬D×𝐁-\gamma_{e}{\mathbf{s}}_{D}\times\mathbf{B} is neglected. Using the fact that 𝒋sb0𝒋e𝒎{\bm{j}}_{s}\approx-b_{0}{\bm{j}}_{e}{\bm{m}} (where b0=μBP/eb_{0}=\mu_{B}P/e and PP is the polarization) and neglecting the second order terms such as tδ𝐬D\partial_{t}\delta{\mathbf{s}}_{D}, we find from Eq. (34)note10

J𝐌×δ𝐬D=b0(𝒋e)𝒎,J{\mathbf{M}}\times\delta{\mathbf{s}}_{D}=-b_{0}({\bm{j}}_{e}\cdot\nabla){\bm{m}}, (35)

where δ𝐬D\delta{\mathbf{s}}_{D} denotes the contribution due to the spatial variation of the magnetization 𝒎\nabla{\bm{m}}. The nonadiabatic term can be generated by iterating the following equation,

d𝐦dt=γ𝐦×𝐇effani𝐦×[𝜶d𝐦dt]+bJ(𝒋e)𝐦,\frac{d\mathbf{m}}{dt}=-\gamma\mathbf{m}\times\mathbf{H}^{\text{ani}}_{\text{eff}}-\mathbf{m}\times[\bm{\alpha}\frac{d\mathbf{m}}{dt}]+b_{J}({\bm{j}}_{e}\cdot\nabla)\mathbf{m}, (36)

from which we arrive atberkov2007 ,

d𝐦dt\displaystyle\frac{d\mathbf{m}}{dt} =\displaystyle= γ1+α2𝐦×𝐇effaniγα1+α2𝐦×[𝐦×𝐇effani]\displaystyle-\frac{\gamma}{1+\alpha^{2}}\mathbf{m}\times\mathbf{H}^{\text{ani}}_{\text{eff}}-\frac{\gamma\alpha}{1+\alpha^{2}}\mathbf{m}\times[\mathbf{m}\times\mathbf{H}^{\text{ani}}_{\text{eff}}] (37)
+\displaystyle+ bJ1+α2(𝒋e)𝐦+bJα1+α2𝐦×(𝒋e)𝐦,\displaystyle\frac{b_{J}}{1+\alpha^{2}}({\bm{j}}_{e}\cdot\nabla)\mathbf{m}+\frac{b_{J}\alpha}{1+\alpha^{2}}\mathbf{m}\times({\bm{j}}_{e}\cdot\nabla)\mathbf{m},

where we have assumed that the Gilbert damping tensor 𝜶\bm{\alpha} is diagonal, i.e., αij=αδij\alpha_{ij}=\alpha\delta_{ij}. The nonadiabatic term can also be derived explicitly, as shown in Appendix E.

V Skyrmion dynamics driven by quantum parametric pumping

In this section, we apply our microscopic theory to investigate Skyrmion dynamics in a 2D system driven by quantum parametric pumping. Initially, an Sk is placed in the central region of a two-lead system, as shown in Fig. 2. Then, we apply two time-dependent voltage gates with a phase difference in the system to drive a dc electric current. The electron flow, in turn, interacts with the Sk, which gives rise to quantum parametric pumping of the Sk. In the tight-binding representation, the Sk is described by the following Hamiltonian

HSk=\displaystyle H_{Sk}= \displaystyle- Jexi,j𝐦i𝐦j+i,j𝐃(𝐦i×𝐦j)\displaystyle J_{ex}\sum_{\langle i,j\rangle}\mathbf{m}_{i}\cdot\mathbf{m}_{j}+\sum_{\langle i,j\rangle}\mathbf{D}\cdot(\mathbf{m}_{i}\times\mathbf{m}_{j}) (38)
\displaystyle- Ki(𝐦iz^)2μi𝐦i𝐁.\displaystyle K\sum_{i}(\mathbf{m}_{i}\cdot{\hat{z}})^{2}-\mu\sum_{i}\mathbf{m}_{i}\cdot\mathbf{B}.

Here JexJ_{ex} is the Heisenberg exchange interaction. 𝐃=D(𝒓i𝒓j)/|𝒓i𝒓j|\mathbf{D}=D({\bm{r}}_{i}-{\bm{r}}_{j})/|{\bm{r}}_{i}-{\bm{r}}_{j}| is the Dzyaloshinskii-Moriya interaction (DMI). KK is the perpendicular magnetic anisotropy constant, and μ\mu is the magnitude of the magnetic moment. To facilitate parametric pumping, we apply gate voltages in two different regions of the system with the following form,

Vp=V1cos(ωpt)+V2cos(ωpt+ϕ),V_{p}=V_{1}\cos(\omega_{p}t)+V_{2}\cos(\omega_{p}t+\phi),

where V1=Vδ(xl1)V_{1}=V\delta(x-l_{1}) and V2=Vδ(xl2)V_{2}=V\delta(x-l_{2}) are potential landscapes with VV the pumping amplitude, ωp\omega_{p} is the pumping frequency, and ϕ\phi is the phase difference. The central scattering region is discretized into a 40×4040\times 40 mesh. The positions of gate voltages are l1=1l_{1}=1 and l2=5l_{2}=5, which are displayed in Fig. 2. In the adiabatic pumping regime (small ωp\omega_{p} limit), the cyclic variation of two potentials V1V_{1} and V2V_{2} can pump out a net current when ϕnπ\phi\neq n\pibrouwer ; Wei-YD . Thus, the total Hamiltonian of the system consists of HαH_{\alpha}, H0H_{0}, HTH_{T}, HH^{\prime} (given by Eqs. (2), (4), (5), and (32), respectively), HSkH_{Sk}, and VpV_{p}.

Since the Sk has slow varying spin texture in space, its dynamics can be approximated by the adiabatic limit in spatial domain. The following LLG equation describing the Sk dynamics driven by parametric pumping needs to be solved,

d𝐦idt=γ𝐦i×[𝐇eff(J/γ)𝐬D]𝐦i×(𝜶𝐦i˙),\frac{d\mathbf{m}_{i}}{dt}=-\gamma\mathbf{m}_{i}\times[\mathbf{H}_{\rm eff}-(J/\gamma)\mathbf{s}_{D}]-\mathbf{m}_{i}\times({\bm{\alpha}}\dot{\mathbf{m}_{i}}), (39)

where the effective field 𝐇eff\mathbf{H}_{\text{eff}} is defined as

𝐇eff=1γδHSkδ𝐦i.\mathbf{H}_{\text{eff}}=\frac{1}{\gamma}\frac{\delta H_{Sk}}{\delta\mathbf{m}_{i}}.

Here 𝐬D\mathbf{s}_{D} is defined in terms of Green’s functions in Eq. (33). The Gilbert damping tensor 𝜶{\bm{\alpha}} is assumed to be a diagonal matrix, αij=αδij\alpha_{ij}=\alpha\delta_{ij}. It is worth mentioning that Eq. (39) already includes the (𝒋e)𝐦({\bm{j}}_{e}\cdot\nabla)\mathbf{m} and 𝐦×(𝒋e)𝐦\mathbf{m}\times({\bm{j}}_{e}\cdot\nabla)\mathbf{m} terms, which is discussed in Sec. IV D and Appendix E.

Refer to caption
Figure 2: Schematic plot of a central region that hosts a Skyrmion and is connected to two metallic leads. The central region consists of a square lattice of size 40×\times40. The arrows denote the in-plane component of the magnetization texture of the Skyrmion. Pumping potentials V1V_{1} and V2V_{2} are applied on the first and fifth column layers of the central region, which are labeled by dark gray bars.

Initial configuration of the Sk is generated by manually creating a topological unity charge at the center of the system and then relaxing the spin texture numerically until the magnetic energy is stable. Note that mzm_{z} at the Sk center is negative, while the outside values are positive. In numerical simulation, the central region is a 40a×40a40a\times 40a square lattice with aa the lattice spacing; the relaxed Sk radius is r0=10r_{0}=10, which is the minimal distance between the Sk center miz(0)=1m_{i}^{z}(0)=-1 and miz(r0)=1m_{i}^{z}(r_{0})=1. Parameters are set as D=0.2JexD=0.2J_{ex}foros2009 , K=0.07JexK=0.07J_{ex}foros2009 , J=2JexJ=2J_{ex}, B=0B=0, and α=0.4\alpha=0.4. The Heisenberg exchange constant Jex=t=1J_{ex}=t=1 is chosen as the energy unit, where tt is the hopping energy. We set =γ=a=1\hbar=\gamma=a=1, and then the coefficients to convert the time tt, current density κ\kappa, and velocity vv to SI units are /Jex\hbar/J_{ex}, 2eJex/(a2)2eJ_{ex}/(a^{2}\hbar), and Jexa/hJ_{ex}a/h.Wangw2015 ; Zhangb2015 Table 1 shows the expressions and particular values for Jex=1J_{ex}=1 meV and a=0.5a=0.5 nm.

Table 1: Unit conversion table for Jex=1J_{ex}=1 meV and a=0.5a=0.5 nm.
Distance xx x^=a\hat{x}=a =0.5=0.5 nm
Time tt t^=/Jex\hat{t}=\hbar/J_{ex} 0.66\approx 0.66 ps
Current density κ\kappa κ^=2eJex/a2\hat{\kappa}=2eJ_{ex}/a^{2}\hbar 2×1012\approx 2\times 10^{12} A/m2
Velocity vv v^=Jexa/(h)\hat{v}=J_{ex}a/(h) 7.59×102\approx 7.59\times 10^{2} m/s
Refer to caption
Figure 3: The transmission coefficient (a) and density of states (b) as a function of the electron energy in the absence of an Sk. (c) and (d) are the transmission and density of states for cases with an Sk at the center. No pumping potential is added in the system (V=0V=0).

Our numerical calculation proceeds as follows. First, with the initial Sk configuration chosen at t=t0t=t_{0}, we calculate the total Hamiltonian of the system and then the frozen Green’s function in Eq. (45) that determines 𝐬D\mathbf{s}_{D} in Eq. (33). Second, the LLG equation in Eq. (39) is solved by using the fourth-order Runge-Kutta method with a small time step dtdt. Then the Sk Hamiltonian in Eq. (38) can be updated. We repeat the above two-step calculation to simulate the Sk dynamics driven by quantum parametric pumping, and monitor the pumped current during the time evolution.

First, we investigate the static transport properties of the system without pumping. Fig. 3 shows the transmission coefficient and density of states (DOS) as a function of the electron energy EE with and without an Sk locating at the system center. When there is no Sk, Fig. 3(a) and (b) show spin-degenerate transmission coefficients and DOS, which are standard transport properties for a metallic square lattice. However, in the presence of the Sk, spin degeneracy of the system is lifted. In Fig. 3(d), the whole energy range [0,8][0,8] can be typically divided into the following three regions, irrespective to the exchange strength JJNdiaye ; Yin .

(i) 0<E<|J|0<E<|J|. The conduction electrons are fully spin-polarized. Since J =2 in our calculation, this region corresponds to 0<E<20<E<2. Only spin-down electrons can transmit in this energy region, and the largest spin polarization is reached near E=2E=2.

(ii) |J|<E<8|J||J|<E<8-|J|. Both spin-up and spin-down conduction electrons exist in the system.

(iii) 8|J|<E<88-|J|<E<8. The conduction electrons are fully polarized with spin up component.

Refer to caption
Figure 4: (a) The pumped current IpI_{p} versus time with or without the Sk. The time is in unit of the pumping period TT, with T=2π/ωpT=2\pi/\omega_{p}. (b) The pumped spin-dependent current Ip/I^{\uparrow/\downarrow}_{p} with the Sk. (c) Time evolution of the Sk center position 𝐑=(x,y)\mathbf{R}=(x,y). Parameters: E=1.9,J=2,V=0.8,ωp=1,ϕ=π/2E=1.9,J=2,V=0.8,\omega_{p}=1,\phi=\pi/2.

Second, we study the parametric pumping effect on the dynamics of an Sk and the corresponding pumped current. Physically, the pumped current can drive the motion of Sk, while the Sk’s motion can affect the pumped current in turn. The pumped current at time tt is defined asWang-BG2

Ip(t)=Tr[ΓRGfrdVpdtGfa],I_{p}(t)={\rm Tr}\left[\Gamma_{R}G_{f}^{r}\frac{dV_{p}}{dt}G_{f}^{a}\right], (40)

where ΓR=ΣRrΣRa\Gamma_{R}=\Sigma_{R}^{r}-\Sigma_{R}^{a} is the linewidth function of the right metallic lead. ΣRr,a\Sigma_{R}^{r,a} are the retarded and advanced self-energies. The Sk center 𝐑=(x,y)\mathbf{R}=(x,y) is defined as 𝐑=i(m0zmiz)𝐫i/i(m0zmiz)\mathbf{R}=\sum_{i}(m_{0}^{z}-m_{i}^{z})\mathbf{r}_{i}/\sum_{i}(m_{0}^{z}-m_{i}^{z}) to characterize its motion, where index ii sums over sites with miz<m0z=0.1m_{i}^{z}<m_{0}^{z}=-0.1.

As shown in Fig.4(a), in the absence of an Sk, the pumped current is roughly a sine or cosine function in time. When an Sk is introduced at t=t0t=t_{0}, the conduction electrons are scattered by the moving Sk. This results in the deviation of the pump current from the smooth curve. Meanwhile, in the presence of the Sk, the pumped current is fully spin-polarized at the given energy, where only the spin-down component is nonzero in Fig.4(b). At the same time, the Sk is driven by the pumped current. Fig. 4(c) displays xx and yy coordinates of the Sk center as a function of time. The remarkable characteristic is the quasi-periodic movement of the Sk along both x^\hat{x} and y^\hat{y} directions. Moreover, the motion of the Sk center has the same period as the pumped current, but is delayed by one quarter cycle in phase. In our system, the pumped current flows in the ±x\pm x direction, and hence the Sk moves faster in this direction. Besides, the Sk acquires a velocity in the ±y^\pm\hat{y} direction. This indicates that the Sk Hall effect can also be driven by parametric pumping.

We examine the influence of pumping parameters on the Sk dynamics. The pumping amplitude is first evaluated. Fig. 5 shows the time evolution of the Sk center for different pumping amplitudes VV. We observe that the Sk’s speed along xx direction increases with the pumping amplitudes. For V=0.4V=0.4, the Sk oscillates around its initial position and does not propagate. As the pumping amplitude increases, the Sk moves faster in +x^+\hat{x} direction and then saturates when the amplitude exceeds V=0.8V=0.8. The motion along the y^\hat{y} direction is always slower than that in the x^\hat{x} direction.

Refer to caption
Figure 5: (a) and (b): xx and yy coordinates of the Sk center 𝐑\mathbf{R} versus time for different pumping amplitudes V=0.4,0.6,0.8,1,1.2V=0.4,0.6,0.8,1,1.2. The pumping frequency is fixed at ωp=1\omega_{p}=1. Other parameters: J=2,E=1.9,l1=1,l2=5,ϕ=π/2J=2,E=1.9,l_{1}=1,l_{2}=5,\phi=\pi/2.

The effect of the pumping frequency is also studied. When there is no Sk, the pumped current in adiabatic pumping regime is independent of the pumping frequency.brouwer ; Wei-YD In the presence of an Sk, We expect that the pumped current can be simply scaled by the pumping frequency. We examine the pumped current for different pumping frequency when the Sk is fixed at its initial configuration, and numerical results are presented in Fig. 6(a). Four periods are shown here. It is clear that the pumped currents collapse precisely onto each other when scaled by the corresponding pumping frequencies. For a free Sk, Fig. 6(b) and (c) show the xx and yy coordinates of the Sk center under the pumping. At small frequencies ωp=0.2\omega_{p}=0.2 and 0.50.5, the Sk is driven along +x^+\hat{x} direction first, and then reflected back periodically. Its motion in yy direction is similar. For a larger frequency ωp=1\omega_{p}=1, there is no such oscillating behavior even for a time scale of 100 periods (not shown here). Notice that when the phase difference is reversed, both the pumped current and the Sk motion change direction.

Refer to caption
Figure 6: (a) The pumped current for different pumping frequencies. The Sk is fixed at its initial configuration during the time evolution. The pumped current IpI_{p} is scaled with ωp\omega_{p}. (b) and (c): xx and yy coordinates of the Sk center for ωp=0.2,0.5,1\omega_{p}=0.2,0.5,1. The pumping amplitude is fixed at V=0.8V=0.8. Other parameters: J=2,E=1.9,l1=1,l2=5,ϕ=π/2J=2,E=1.9,l_{1}=1,l_{2}=5,\phi=\pi/2.

VI Summary

In conclusion, we have developed a unified microscopic theory of the LLG equation in terms of nonequilibrium Green’s function. Four limiting cases of the microscopic LLG equation are discussed in detail, including the adiabatic, inertia, and nonadiabatic limits for the magnetization with fixed magnitude, as well as the adiabatic limit in spatial domain for the magnetization with slow varying magnitude and direction in space. As a demonstration, the microscopic LLG equation is applied to investigate the motion of a Skyrmion state driven by quantum parametric pumping. Our work not only provides a unified microscopic theory of the LLG equation, but also offers a practical formalism to explore magnetization dynamics with nonequilibrium Green’s functions.

Acknowledgement

This work was supported by the National Natural Science Foundation of China (Grants No. 12034014, No. 12174262, No. 12074230, and No. 12074190). L. Zhang thanks the Fund for Shanxi ”1331 Project”.

Appendix A

In this appendix, we express the charge and spin current in terms of the nonequilibirum Green’s functions.

In the presence of time-varying magnetization, the nonequilibrium Green’s function Gr(t,t)G^{r}(t,t^{\prime}) depends on two time indices tt and tt^{\prime}. If the magnetization changes slowly with time, we can treat the time difference (tt)(t-t^{\prime}) in energy space.arrachea2006 After taking the Fourier transformation, the Green’s function in energy space only depends on one time variable tt:

Gr(t,E)=𝑑τeiEτGr(t,t),G^{r}(t,E)=\int d\tau e^{iE\tau}G^{r}(t,t^{\prime}),

where τ=tt\tau=t-t^{\prime}. The inverse Fourier transformation gives

Gr(t,t)=dE2πeiEτGr(t,E).G^{r}(t,t^{\prime})=\int\frac{dE}{2\pi}e^{-iE\tau}G^{r}(t,E).

With the above definition, it is easy to show that

G<(t,t)=dE2πeiEτGr(t,E)Σ<(E)Ga(E,t).G^{<}(t,t^{\prime})=\int\frac{dE}{2\pi}e^{-iE\tau}G^{r}(t,E)\Sigma^{<}(E)G^{a}(E,t^{\prime}). (41)

In this representation, the particle current matrix is definedZhang-L

Iopα(t)=dE2π[Gr(t,E)Σα<(E)+G<(t,E)Σαa(E)+H.c.].I^{\alpha}_{\rm op}(t)=\int\frac{dE}{2\pi}[G^{r}(t,E)\Sigma_{\alpha}^{<}(E)+G^{<}(t,E)\Sigma_{\alpha}^{a}(E)+\rm{H.c.}]. (42)

In term of which, the charge current Icα(t)I_{c\alpha}(t) and spin current Isα(t)I_{s\alpha}(t) are expressed as

Icα(t)=qTr[Iopα],𝐈sα(t)=12Tr[𝝈Iopα].I_{c\alpha}(t)=-q\mathrm{Tr}[I^{\alpha}_{\rm op}],~{}~{}~{}{\mathbf{I}}_{s\alpha}(t)=-\frac{1}{2}\mathrm{Tr}[\bm{\sigma}I^{\alpha}_{\rm op}]. (43)

As shown in Appendix C, these time-dependent Green’s functions, such as Gr(t,E)G^{r}(t,E) (also called Floquet Green’s functionarrachea2006 ), can be expressed in terms of the instantaneous frozen Green’s function GfrG^{r}_{f}, which satisfies the following recursive relation:

Gr(t,E)=Gfr(t,E)iGfr(t,E)G˙r(t,E),G^{r}(t,E)=G^{r}_{f}(t,E)-iG^{r}_{f}(t,E)\dot{G}^{r}(t,E), (44)

where G˙r\dot{G}^{r} is the time derivative of GrG^{r}. The frozen Green’s function contains the effective magnetic field 𝐛(t){\bf b}(t) and is defined as

Gfr(t,E)=[EH0Σr𝝈𝐛(t)/2]1,G_{f}^{r}(t,E)=[E-H_{0}-\Sigma^{r}-{\bm{\sigma}}\cdot{\bf b}(t)/2]^{-1}, (45)

where 𝐛(t)=γe𝐁+J𝐌{\bf b}(t)=\gamma_{e}\mathbf{B}+J{\mathbf{M}}. The self energies (for ferromagnetic leads) are given by

Σmnγ(t)=R^Σ0,mnγR^,\Sigma^{\gamma}_{mn}(t)=\hat{R}^{\dagger}\Sigma^{\gamma}_{0,mn}\hat{R},

with γ=r,a,<\gamma=r,a,<. R^\hat{R} is defined in Eq. (7). Σ0,mnγ\Sigma^{\gamma}_{0,mn} is the self-energy when the magnetization is along zz-axis:

Σ0,mnγ=kαtkαmgkαγtkαn=(Σmn,γ00Σmn,γ),\Sigma^{\gamma}_{0,mn}=\sum_{k\alpha}t^{*}_{k\alpha m}g^{\gamma}_{k\alpha}t_{k\alpha n}=\begin{pmatrix}\Sigma^{\gamma}_{mn,\uparrow}&0\\ 0&\Sigma^{\gamma}_{mn,\downarrow}\end{pmatrix},

where gkαγg^{\gamma}_{k\alpha} is the surface Green’s function of lead α\alpha.

Appendix B

In this appendix, we express the microscopic LLG equation Eq. (13) in terms of the spin current and spin operators.

For the spin transfer torque (STT) to occur in magnetic multilayers, one requires a pair of FM layers in a noncollinear configuration. Then a spin-polarized current can be generated from the reference (fixed) layer, and the transverse spin can be transferred to the switchable (free) layer. The current induced STT can also be obtained from Eq. (13) in such a noncollinear magnetic system. Denoting H1=HtotalHH_{1}=H_{\rm total}-H^{\prime} and defining the spin operators for lead α\alpha and the QD:

𝐬^α=12kσσc^kσα𝝈σσc^kσα,𝐬^D=12nσσd^nσ𝝈σσd^nσ,\hat{\mathbf{s}}_{\alpha}=\frac{1}{2}\sum_{k\sigma\sigma^{{}^{\prime}}}\hat{c}^{\dagger}_{k\sigma\alpha}\bm{\sigma}_{\sigma\sigma^{{}^{\prime}}}\hat{c}_{k\sigma^{{}^{\prime}}\alpha}\ ,~{}~{}~{}\hat{\mathbf{s}}_{D}=\frac{1}{2}\sum_{n\sigma\sigma^{{}^{\prime}}}\hat{d}^{\dagger}_{n\sigma}\bm{\sigma}_{\sigma\sigma^{{}^{\prime}}}\hat{d}_{n\sigma^{{}^{\prime}}},

we have

d𝐬^αdt\displaystyle\frac{d\hat{\mathbf{s}}_{\alpha}}{dt} =\displaystyle= i[𝐬^α,H1],\displaystyle-i[\hat{\mathbf{s}}_{\alpha},H_{1}],
d𝐬^Ddt\displaystyle\frac{d\hat{\mathbf{s}}_{D}}{dt} =\displaystyle= i[𝐬^D,H^1]i[𝐬^D,H^],\displaystyle-i[\hat{\mathbf{s}}_{D},\hat{H}_{1}]-i[\hat{\mathbf{s}}_{D},\hat{H}^{\prime}],

where

i[𝐬^D,H^]=γe𝐬^D×𝐁+J𝐌^×𝐬^D.\begin{split}-i[\hat{\mathbf{s}}_{D},\hat{H}^{\prime}]=-\gamma_{e}\hat{\mathbf{s}}_{D}\times\mathbf{B}+J\hat{\mathbf{M}}\times\hat{\mathbf{s}}_{D}.\end{split}

It can be shown thatwangj2006 [α𝐬^α+𝐬^D,H1]=0[\sum_{\alpha}\hat{\mathbf{s}}_{\alpha}+\hat{\mathbf{s}}_{D},H_{1}]=0, from which the spin continuity equation of the system is expressed ashattori2007

αd𝐬^αdt+d𝐬^Ddt=γe𝐬^D×𝐁+J𝐌^×𝐬^D,\sum_{\alpha}\frac{d\hat{\mathbf{s}}_{\alpha}}{dt}+\frac{d\hat{\mathbf{s}}_{D}}{dt}=-\gamma_{e}\hat{\mathbf{s}}_{D}\times\mathbf{B}+J\hat{\mathbf{M}}\times\hat{\mathbf{s}}_{D}, (46)

which indicates that the total spin is not conserved due to spin precessionwangj2006 . Substituting Eq. (46) into Eq. (8) and taking quantum average, we have

𝐌˙=γ𝐌×𝐁α𝐈sα+𝐀,\dot{\mathbf{M}}=-\gamma{\mathbf{M}}\times\mathbf{B}-\sum_{\alpha}{\mathbf{I}}_{s\alpha}+{\mathbf{A}}, (47)

where the correction term 𝐀{\mathbf{A}} is given by

𝐀=d𝐬Ddtγe𝐬D×𝐁,{\mathbf{A}}=-\frac{d{\mathbf{s}}_{D}}{dt}-\gamma_{e}{\mathbf{s}}_{D}\times\mathbf{B}, (48)

and α𝐈sα=αd𝐬α/dt\sum_{\alpha}{\mathbf{I}}_{s\alpha}=\sum_{\alpha}d{\mathbf{s}}_{\alpha}/dt is the total spin current. Notice that Eq. (13) and Eq. (47) are equivalent. Form Eq. (48), it is found that when 𝐬D{\mathbf{s}}_{D} is time-reversal symmetric, d𝐬D/dtd{\mathbf{s}}_{D}/dt or 𝐀{\mathbf{A}} is time-reversal antisymmetric. If we decompose α𝐈sα\sum_{\alpha}{\mathbf{I}}_{s\alpha} and 𝐀{\mathbf{A}} into the time-reversal symmetric (labeled with superscript ss) and antisymmetric (labeled with superscript aa) parts, we have

𝐌˙=γ𝐌×𝐁α𝐈sαs+𝐀sJ𝐌×𝐬Ds.\dot{\mathbf{M}}=-\gamma{\mathbf{M}}\times\mathbf{B}-\sum_{\alpha}{\mathbf{I}}^{s}_{s\alpha}+{\mathbf{A}}^{s}-J\mathbf{M}\times\mathbf{s}^{s}_{D}. (49)

Here we have used the fact that α𝐈sα𝐀=J𝐌×𝐬D\sum_{\alpha}{\mathbf{I}}_{s\alpha}-{\mathbf{A}}=J\mathbf{M}\times\mathbf{s}_{D}. When the external magnetic field is strong enough, the electron spin will approximately align with the direction of the field and we may drop the term γe𝐬Da×𝐁-\gamma_{e}{\mathbf{s}}_{D}^{a}\times\mathbf{B}.

In the adiabatic limit, the term d𝐬Da/dt-d{\mathbf{s}}^{a}_{D}/dt in in 𝐀s{\mathbf{A}}^{s} of Eq. (49) corresponds to d𝐬D(0)/dt-d{\mathbf{s}}^{(0)}_{D}/dt. Expanding it to the first order in frequency, we find

d𝐬D(0)dt\displaystyle-\frac{d{\mathbf{s}}^{(0)}_{D}}{dt} =\displaystyle= iJM4dE2πTr[Gf<𝝈Gfr𝝈+Gfa𝝈Gf<𝝈]d𝐦dt\displaystyle-\frac{iJM}{4}\int\frac{dE}{2\pi}\mathrm{Tr}[G_{f}^{<}\bm{\sigma}G_{f}^{r}\bm{\sigma}+G_{f}^{a}\bm{\sigma}G_{f}^{<}\bm{\sigma}]\frac{d{\mathbf{m}}}{dt} (50)
\displaystyle\equiv ηd𝐦dt,\displaystyle-\eta\frac{d{\mathbf{m}}}{dt},

where η\eta is a second rank tensor. This term can be absorbed by introducing an effective gyromagnetic ratio γ=γ(1+η)1\gamma^{\prime}=\gamma(1+\eta)^{-1}. Therefore, the driving force of magnetization precession originates from the total spin current α𝐈sαs\sum_{\alpha}{\mathbf{I}}^{s}_{s\alpha}, which corresponds to the current induced STT discussed in Ref. [Brataas2012, ].

Appendix C

In this appendix, we derive the relation between Floquet Green’s functions and frozen Green’s functions. We start with the two-time retarded Green’s function defined ashaugBook

[it1H(t1)]Gr(t1,t2)+𝑑tΣr(t1,t)Gr(t,t2)=δ(t1,t2).[i\frac{\partial}{\partial t_{1}}-H(t_{1})]G^{r}(t_{1},t_{2})+\int dt^{\prime}\Sigma^{r}(t_{1},t^{\prime})G^{r}(t^{\prime},t_{2})=\delta(t_{1},t_{2}).

To work in the Floquet Green’s function representation, we make the Fourier transform with respect to the fast time scale τ=t1t2\tau=t_{1}-t_{2}, and obtain

[it+EH(t)]Gr(t,E)t𝑑teiE(tt)Σr(t,t)Gr(t,E)=I.\begin{split}\Big{[}i\frac{\partial}{\partial t}&+E-H(t)\Big{]}G^{r}(t,E)\\ &-\int_{-\infty}^{t}dt^{\prime}e^{iE(t-t^{\prime})}\Sigma^{r}(t,t^{\prime})G^{r}(t^{\prime},E)=I.\end{split} (51)

The last term on the left hand side can be written as

t𝑑teiE(tt)Σr(t,t)Gr(t,E)=dE2πt𝑑tei(EE)tei(EE)tΣr(t,E)Gr(t,E)dE2πt𝑑tei(EE)tei(EE)tΣr(t,E)Gr(t,E)=dE2π2πδ(EE)ei(EE)tΣr(t,E)Gr(t,E)=Σr(t,E)Gr(t,E),\begin{split}&\int_{-\infty}^{t}dt^{\prime}e^{iE(t-t^{\prime})}\Sigma^{r}(t,t^{\prime})G^{r}(t^{\prime},E)\\ &=\int\frac{dE^{\prime}}{2\pi}\int_{-\infty}^{t}dt^{\prime}e^{i(E^{\prime}-E)t^{\prime}}~{}e^{i(E-E^{\prime})t}\Sigma^{r}(t,E^{\prime})G^{r}(t^{\prime},E)\\ &\approx\int\frac{dE^{\prime}}{2\pi}\int_{-\infty}^{t}dt^{\prime}e^{i(E^{\prime}-E)t^{\prime}}~{}e^{i(E-E^{\prime})t}\Sigma^{r}(t,E^{\prime})G^{r}(t,E)\\ &=\int\frac{dE^{\prime}}{2\pi}2\pi\delta(E^{\prime}-E)~{}e^{i(E-E^{\prime})t}\Sigma^{r}(t,E^{\prime})G^{r}(t,E)\\ &=\Sigma^{r}(t,E)G^{r}(t,E),\end{split}

where Gr(t,E)=Gr(tτ,E)Gr(t,E)G^{r}(t^{\prime},E)=G^{r}(t-\tau,E)\approx G^{r}(t,E) is used and the fast time variable τ=tt\tau=t-t^{\prime} is neglected. Then we can introduce the frozen Green’s function Gfr(t,E)G^{r}_{f}(t,E) satisfying

[EH(t)Σr(t,E)]Gfr(t,E)=I.[E-H(t)-\Sigma^{r}(t,E)]G^{r}_{f}(t,E)=I. (52)

Substituting Eq. (52) into Eq. (51), we have

Gr(t,E)=Gfr(t,E)iGfr(t,E)G˙r(t,E).G^{r}(t,E)=G^{r}_{f}(t,E)-iG^{r}_{f}(t,E)\dot{G}^{r}(t,E). (53)

Similarly, the advanced Green’s function is

Ga(E,t)=Gfa(E,t)+iG˙a(E,t)Gfa(E,t).G^{a}(E,t)=G^{a}_{f}(E,t)+i\dot{G}^{a}(E,t){G}^{a}_{f}(E,t). (54)

Up to the linear order in frequency, we find

Gr(t,E)\displaystyle G^{r}(t,E) =\displaystyle= Gfr(t,E)iGfr(t,E)G˙fr(t,E),\displaystyle G^{r}_{f}(t,E)-iG^{r}_{f}(t,E)\dot{G}^{r}_{f}(t,E),
Ga(E,t)\displaystyle G^{a}(E,t) =\displaystyle= Gfa(E,t)+iG˙fa(E,t)Gfa(E,t).\displaystyle G^{a}_{f}(E,t)+i\dot{G}^{a}_{f}(E,t){G}^{a}_{f}(E,t). (55)

Notice that the frozen Green’s function is an instantaneous function, since it depends only on the present time.

Appendix D

In this appendix, we compare the Gilbert damping coefficient derived in Ref. [brataas2008, ] using the scattering matrix theory with that obtained in this work via nonequilibrium Green’s functions. According to Ref. [brataas2008, ], in the absence of external bias, the Gilbert tensor element at zero temperature is expressed in terms of the scattering matrix at the Fermi energy:

Gij(𝐦)=γ24πRe{Tr[SmiSmj]}.G_{ij}(\mathbf{m})=\frac{\gamma^{2}}{4\pi}\mathrm{Re}\{\mathrm{Tr}[\frac{\partial S}{\partial m_{i}}\frac{\partial S}{\partial m_{j}}]\}. (56)

To compare with Eq. (19), we calculate the dimensionless quantity αij=Gij/γ2\alpha^{\prime}_{ij}=G_{ij}/\gamma^{2}. The scattering matrix SS is connected to the Green’s function through the Fisher-Lee relationfisher1981 ; wangj2009

Sαβσσ\displaystyle S_{\alpha\beta\sigma\sigma^{\prime}} =\displaystyle= δαβσσ+iWασ|Gr|Wβσ,\displaystyle-\delta_{\alpha\beta\sigma\sigma^{\prime}}+i\langle W_{\alpha\sigma}|G^{r}|W_{\beta\sigma^{\prime}}\rangle,
(S)βασσ\displaystyle(S^{\dagger})_{\beta\alpha\sigma^{\prime}\sigma} =\displaystyle= δαβσσiWβσ|Ga|Wασ,\displaystyle-\delta_{\alpha\beta\sigma\sigma^{\prime}}-i\langle W_{\beta\sigma^{\prime}}|G^{a}|W_{\alpha\sigma}\rangle,

where Γαmnσσ=|WαmσWαnσ|\Gamma_{\alpha mn\sigma\sigma^{\prime}}=|W_{\alpha m\sigma}\rangle\langle W_{\alpha n\sigma^{\prime}}|. |Wαm|W_{\alpha m}\rangle is proportional to the eigenvector of Γα\Gamma_{\alpha}wangj2009 . From Eq. (45), we see that

Gfr,ami=12JMGfr,aσiGfr,a.\frac{\partial G^{r,a}_{f}}{\partial m_{i}}=\frac{1}{2}JMG^{r,a}_{f}\sigma_{i}G^{r,a}_{f}.

Then we can express Eq. (56) with frozen Green’s functions

Tr[SmiSmj]\displaystyle\mathrm{Tr}[\frac{\partial S}{\partial m_{i}}\frac{\partial S}{\partial m_{j}}] =\displaystyle= 14(JM)2Tr[GfaΓGfrσiGfrΓGfaσj]\displaystyle\frac{1}{4}(JM)^{2}\mathrm{Tr}[G^{a}_{f}\Gamma G^{r}_{f}\sigma_{i}G^{r}_{f}\Gamma G^{a}_{f}\sigma_{j}]
=\displaystyle= (JM)2Tr[Im(Gfr)σiIm(Gfr)σj].\displaystyle(JM)^{2}\mathrm{Tr}[{\rm Im}(G^{r}_{f})\sigma_{i}{\rm Im}(G^{r}_{f})\sigma_{j}].

Hence the dimensionless damping tensor element is

αij=116π(JM)2Re{Tr[GfaΓGfrσiGfrΓGfaσj]}.\alpha^{\prime}_{ij}=\frac{1}{16\pi}(JM)^{2}\mathrm{Re}\{\mathrm{Tr}[G^{a}_{f}\Gamma G^{r}_{f}\sigma_{i}G^{r}_{f}\Gamma G^{a}_{f}\sigma_{j}]\}. (57)

Now we assume that there is no external bias and the temperature is zero, which are the same conditions used in deriving Eq. (56) in Ref. [brataas2008, ]. Note that Eq. (56) was derived from the pumped energy current defined as 𝐦˙α𝐦˙{\dot{\bf m}}\alpha^{\prime}{\dot{\bf m}} so that the resultant tensor αij\alpha^{\prime}_{ij} is always symmetric. Hence we symmetrize the dimensionless damping tensor in Eq. (19):

α~ij=12(αij+αji).\tilde{\alpha}_{ij}=\frac{1}{2}(\alpha_{ij}+\alpha_{ji}). (58)

In the absence of external bias, Gf<=(GfaGfr)f(E)G_{f}^{<}=(G_{f}^{a}-G_{f}^{r})f(E), and Eq. (58) becomes

α~ij\displaystyle\tilde{\alpha}_{ij} =dETr[σiAσjGfrGfrσjAσiGfaGfa\displaystyle=-\int dE\mathrm{Tr}[\sigma_{i}A\sigma_{j}G_{f}^{r}G_{f}^{r}-\sigma_{j}A\sigma_{i}G_{f}^{a}G_{f}^{a} (59)
+σjAσiGfrGfrσiAσjGfaGfa]\displaystyle+\sigma_{j}A\sigma_{i}G_{f}^{r}G_{f}^{r}-\sigma_{i}A\sigma_{j}G_{f}^{a}G_{f}^{a}]
=𝑑ETr[σiAσjE(GfrGfa)+σjAσiE(GfrGfa)]\displaystyle=-\int dE\mathrm{Tr}[\sigma_{i}A\sigma_{j}\partial_{E}(G_{f}^{r}-G^{a}_{f})+\sigma_{j}A\sigma_{i}\partial_{E}(G_{f}^{r}-G^{a}_{f})]
=𝑑EfTrE[σi(GfrGfa)σj(GfrGfa)]\displaystyle=-\int dEf\mathrm{Tr}\partial_{E}[\sigma_{i}(G_{f}^{r}-G^{a}_{f})\sigma_{j}(G_{f}^{r}-G^{a}_{f})]
=Tr[σi(GfrGfa)σj(GfrGfa)],\displaystyle=\mathrm{Tr}[\sigma_{i}(G_{f}^{r}-G^{a}_{f})\sigma_{j}(G_{f}^{r}-G^{a}_{f})],

with A=(GfrGfa)fA=(G_{f}^{r}-G^{a}_{f})f. Apart from a constant factor (JM)2/16π(JM)^{2}/16\pi, this expression is exactly the same as αij\alpha^{\prime}_{ij} shown in Eq. (57). Therefore, we confirm the equivalence of the Gilbert damping tensor between our formalism (Eq. (19)) and that obtained in Ref. [brataas2008, ] in the limiting case.

Appendix E

In this appendix, we derive the nonadiabatic term in Eq. (31). Expanding 𝐦(𝐫,t){\bf m}({\bf r},t) up to the first order in i𝐦𝐦/xi\partial_{i}{\bf m}\equiv\partial{\bf m}/\partial x_{i}, we have

𝐦(𝐫,t)=𝐦+δ𝐦=𝐦+xii𝐦,{\bf m}({\bf r},t)={\bf m}+\delta{\bf m}={\bf m}+x_{i}\partial_{i}{\bf m},

where the Einstein summation convention is implied. The nonequilibrium Green’s functions have similar expansions

Gr\displaystyle G^{r} =\displaystyle= G0r(JM/4)G0r𝝈xii𝐦G0r,\displaystyle G^{r}_{0}-(JM/4)G^{r}_{0}{\bm{\sigma}}x_{i}\partial_{i}{\bf m}G^{r}_{0},
G<\displaystyle G^{<} =\displaystyle= G0<[(JM/4)G0r𝝈xii𝐦G0<H.c.].\displaystyle G^{<}_{0}-[(JM/4)G^{r}_{0}{\bm{\sigma}}x_{i}\partial_{i}{\bf m}G^{<}_{0}-\rm{H.c.}].

It is straightforward to find the correction on 𝒔D(0){\bm{s}}^{(0)}_{D} due to i𝐦\partial_{i}{\bf m},

δ𝒔D(0)=JMi8dE2πTrs[𝝈Gr𝝈xjG<]xxj𝐦+H.c.,\delta{\bm{s}}^{(0)}_{D}=\frac{JMi}{8}\int\frac{dE}{2\pi}{\rm Tr_{s}}[{\bm{\sigma}}G^{r}{\bm{\sigma}}x_{j}G^{<}]_{xx}\partial_{j}{\bf m}+\rm{H.c.}, (60)

where we have focused on the linear response regimeZhang-SF ; Li-Z and kept only the linear term in j𝐦\partial_{j}{\bf m}. Here Trs[]xx{\rm Tr_{s}}[...]_{xx} denotes tracing over spin space and then taking diagonal matrix element in real space and we have dropped the subscript 0 in the Green’s function. If we further neglect SOI, the Green’s function is diagonal in spin space in the linear regime, and Trs[𝝈Gr𝝈xjG<]xx=𝟏Trs[GrxjG<]xx{\rm Tr_{s}}[{\bm{\sigma}}G^{r}{\bm{\sigma}}x_{j}G^{<}]_{xx}={\bm{1}}{\rm Tr_{s}}[G^{r}x_{j}G^{<}]_{xx} is also diagonal in spin space. Using the relation

xj=mc0[j,HE],x_{j}=mc_{0}[\nabla_{j},H-E],

which is valid in the adiabatic approximation in spatial domain (c0c_{0} is a constant having dimension T2T^{-2} with TT the dimension of time), Eq. (60) becomes

δ𝒔D(0)=JMmc08[w1(x)+w2(x)]j𝐦,\delta{\bm{s}}^{(0)}_{D}=\frac{JMmc_{0}}{8}[w_{1}(x)+w_{2}(x)]\partial_{j}{\bf m},

where

w1j(x)\displaystyle w_{1j}(x) =\displaystyle= idE2π[(G<j)Ga(HE)(HE)Gr(jG<)],\displaystyle i\int\frac{dE}{2\pi}[(G^{<}\nabla_{j})G^{a}(H-E)-(H-E)G^{r}(\nabla_{j}G^{<})],
w2j(x)\displaystyle w_{2j}(x) =\displaystyle= idE2π[(Grj)(HE)G<G<(HE)(jGa)].\displaystyle i\int\frac{dE}{2\pi}[(G^{r}\nabla_{j})(H-E)G^{<}-G^{<}(H-E)(\nabla_{j}G^{a})].

Using 𝟏+ΣrGr=(EH)Gr{\bm{1}}+\Sigma^{r}G^{r}=(E-H)G^{r}, we find

w1j(x)\displaystyle w_{1j}(x) =\displaystyle= idE2π[(jG<)(G<j)]\displaystyle i\int\frac{dE}{2\pi}[(\nabla_{j}G^{<})-(G^{<}\nabla_{j})]
+\displaystyle+ dE2π[ΣrGr(jG<)(G<j)GaΣa],\displaystyle\int\frac{dE}{2\pi}[\Sigma^{r}G^{r}(\nabla_{j}G^{<})-(G^{<}\nabla_{j})G^{a}\Sigma^{a}],

where the first term of w1j(x)w_{1j}(x) is proportional to the current density 𝒋e{\bm{j}}_{e}Zhang-L2 . Focusing on this particular term, we have

δ𝒔D(0)=JMm2c04(𝒋e)𝐦,\delta{\bm{s}}^{(0)}_{D}=\frac{JMm^{2}c_{0}}{4}({\bm{j}}_{e}\cdot\nabla){\bf m},

which gives rise to the nonadiabatic torque due to the spatial variation of the magnetization.

References

  • §\S~{}~{} These authors contributed equally to this work.
  • (1) A. Ardavan, O. Rival, J. J. L. Morton, S. J. Blundell, A. M. Tyryshkin, G. A. Timco, and R. E. P. Winpenny, Phys. Rev. Lett. 98, 057201 (2007).
  • (2) A. Garg, Europhys. Lett. 22, 205 (1993).
  • (3) J. R. Friedman, M. P. Sarachik, J. Tejada, and R. Ziolo, Phys. Rev. Lett. 76, 3830 (1996).
  • (4) C. Sangregorio, T. Ohm, C. Paulsen, R. Sessoli, and D. Gatteschi, Phys. Rev. Lett. 78, 4645 (1997).
  • (5) W. Wernsdorfer and R. Sessoli, Science 284, 133 (1999).
  • (6) W. Wernsdorfer, N. E. Chakov, and G. Christou, Phys. Rev. Lett. 95, 037203 (2005).
  • (7) M. Filipović, C. Holmqvist, F. Haupt, and W. Belzig, Phys. Rev. B 87, 045426 (2013); M. Filipović, C. Holmqvist, F. Haupt, and W. Belzig, Phys. Rev. B 88, 119901(E) (2013).
  • (8) O. Kahn and C. J. Martinez, Science 279, 44 (1998).
  • (9) C. Timm and M. Di Ventra, Phys. Rev. B 86, 104427 (2012).
  • (10) H. B. Heersche, Z. de Groot, J. A. Folk, H. S. J. van der Zant, C. Romeike, M. R. Wegewijs, L. Zobbi, D. Barreca, E. Tondello, and A. Cornia, Phys. Rev. Lett. 96, 206801 (2006).
  • (11) M.-H. Jo, J. E. Grose, K. Baheti, M. M. Deshmukh, J. J. Sokol, E. M. Rumberger, D. N. Hendrickson, J. R. Long, H. Park, and D. C. Ralph, Nano Letters 6, 2014 (2006).
  • (12) A. S. Zyazin, J. W. G. van den Berg, E. A. Osorio, H. S. J. van der Zant, N. P. Konstantinidis, M. Leijnse, M. R. Wegewijs, F. May, W. Hofstetter, C. Danieli, and A. Cornia, Nano Letters 10, 3307 (2010).
  • (13) N. Roch, R. Vincent, F. Elste, W. Harneit, W. Wernsdorfer, C. Timm, and F. Balestro, Phys. Rev. B 83, 081407 (2011).
  • (14) J. Park, A. N. Pasupathy, J. I. Goldsmith, C. Chang, Y. Yaish, J. R. Petta, M. Rinkoski, J. P. Sethna, H. D. Abruna, P. L. McEuen, and D. C. Ralph, Nature 417, 722 (2002).
  • (15) O. A. Tretiakov and A. Mitra, Phys. Rev. B 81, 024416 (2010).
  • (16) N. Bode, L. Arrachea, G. S. Lozano, T. S. Nunner and F. von Oppen, Phys. Rev. B 85, 115440 (2012).
  • (17) J. Fransson, J. Ren, and J.-X. Zhu, Phys. Rev. Lett. 113, 257201 (2014).
  • (18) H. Hammar and J. Fransson, Phys. Rev. B 94, 054311 (2016).
  • (19) U. Bajpai and B. K. Nikolić, Phys. Rev. Appl. 10, 054038 (2018).
  • (20) U. Bajpai and B. K. Nikolić, Phys. Rev. B 99, 134409 (2019).
  • (21) H. Hammar and J. Fransson, Phys. Rev. B 96, 214401 (2017).
  • (22) J. Fransson, D. Thonig, P. F. Bessarab, S. Bhattacharjee, J. Hellsvik, and L. Nordström, Phys. Rev. Materials 1, 074404 (2017).
  • (23) T. L. Gilbert, IEEE Trans. Magn. 40, 3443 (2004).
  • (24) J. Foros, A. Brataas, Y. Tserkovnyak, and G. E. W. Bauer, Phys. Rev. Lett. 95, 016601 (2005).
  • (25) A. L. Chudnovskiy, J. Swiebodzinski, and A. Kamenev, Phys. Rev. Lett. 101, 066601 (2008).
  • (26) J. Foros, A. Brataas, G. E. W. Bauer, and Y. Tserkovnyak, Phys. Rev. B 79, 214407 (2009).
  • (27) J. Swiebodzinski, A. Chudnovskiy, T. Dunn, and A. Kamenev, Phys. Rev. B 82, 144404 (2010).
  • (28) A. Brataas, Y. Tserkovnyak, and G. E. W. Bauer, Phys. Rev. Lett. 101, 037207 (2008).
  • (29) A. Brataas, Y. Tserkovnyak, and G. E. W. Bauer, Phys. Rev. B 84, 054416 (2011).
  • (30) L. Berger, Phys. Rev. B 54, 9353 (1996).
  • (31) J. C. Slonczewski, J. Magn. Magn. Mater. 159, L1 (1996).
  • (32) Y. Tserkovnyak, A. Brataas, G. E. W. Bauer, and B. I. Halperin, Rev. Mod. Phys. 77, 1375 (2005).
  • (33) D. C. Ralph and M. D. Stiles, J. Magn. Magn. Mater. 320, 1190 (2008).
  • (34) A. A. Starikov, P. J. Kelly, A. Brataas, Y. Tserkovnyak, and G. E. W. Bauer, Phys. Rev. Lett. 105, 236601 (2010).
  • (35) S. Bhattacharjee, L. Nordström, and J. Fransson, Phys. Rev. Lett. 108, 057204 (2012).
  • (36) S. Mankovsky, D. Kodderitzsch, G. Woltersdorf, and H. Ebert, Phys. Rev. B 87, 014430 (2013).
  • (37) Y. Yang, R. B. Wilson, J. Gorchon, C.-H. Lambert, S. Salahuddin, and J. Bokor, Sci. Adv. 3, e1603117 (2017).
  • (38) K. Jhuria, J. Hohlfeld, A. Pattabi, E. Martin, A. Y. A. Codova, X.P. Shi, R. Lo Conte, S. Petit-Watelot, J. C. Rojas-Sanchez, G. Malinowski, S. Mangin, A. Lemaitre, M. Hehn, J. Bokor, R. B. Wilson and J. Gorchon, Nat. Elec. 3, 680 (2020).
  • (39) B. Koopmans, G. Malinowski, F. Dalla Longa, D. Steiauf, M. Fähnle, T. Roth, M. Cinchetti and M. Aeschlimann, Nat. Mater. 9, 259 (2010).
  • (40) M. Kammerer, M. Weigand, M. Curcic, M. Noske, M. Sproll, A. Vansteenkiste, B. Van Waeyenberge, H. Stoll, G. Woltersdorf, C. H. Back, and G. Schuetz, Nat. Commun. 2, 279 (2011).
  • (41) I. Radu, K. Vahaplar, C. Stamm, T. Kachel, N. Pontius, H. A. Dürr, T. A. Ostler, J. Barker, R. F. L. Evans, R. W. Chantrell, A. Tsukamoto, A. Itoh, A. Kirilyuk, Th. Rasing and A. V. Kimel, Nature 472, 205 (2011).
  • (42) A. Stupakiewicz, K. Szerenos, D. Afanasiev, A. Kirilyuk, and A. V. Kimel, Nature 542, 71 (2017).
  • (43) M. Fähnle, D. Steiauf, and C. Illg, Phys. Rev. B 84, 172403 (2011).
  • (44) H. Suhl, IEEE Trans. Magn. 34, 1834 (1998).
  • (45) M.-C. Ciornei, J. M. Rubí, and J. E. Wegrowe, Phys. Rev. B 83, 020410(R) (2011).
  • (46) Y. Li, A.-L. Barra, S. Auffret, U. Ebels, and W. E. Bailey, Phys. Rev. B 92, 140413(R) (2015).
  • (47) E. Olive, Y. Lansac, M. Meyer, M. Hayoun, and J.-E. Wegrowe, J. Appl. Phys. 117, 213904 (2015).
  • (48) R. Mondal, J. Phys.: Condens. Matter 33, 275804 (2021).
  • (49) R. Mondal, M. Berritta, and P. M. Oppeneer, Phys. Rev. B 94, 144419 (2016).
  • (50) R. Mondal, M. Berritta, A. K. Nandy, and P. M. Oppeneer, Phys. Rev. B 96, 024425 (2017).
  • (51) K. Neeraj, N. Awari, S. Kovalev, D. Polley, N. Z. Hagström, S. S. P. K. Arekapudi, A. Semisalova, K. Lenz, B. Green, J.-C. Deinert, I. Ilyakov, M. Chen, M. Bawatna, V. Scalera, M. d’Aquino, C. Serpico, O. Hellwig, J.- E. Wegrowe, M. Gensch and S. Bonetti, Nat. Phys. 17, 245 (2021).
  • (52) S. Zhang and Z. Li, Phys. Rev. Lett. 93, 127204 (2004).
  • (53) S. K. Kim, K. J. Lee, and Y. Tserkovnyak, Phys. Rev. B 95, 140404(R) (2017).
  • (54) A. Neubauer, C. Pfleiderer, B. Binz, A. Rosch, R. Ritz, P. G. Niklowitz, and P. Böni, Phys. Rev. Lett. 102, 186602 (2009).
  • (55) W. Jiang, X. Zhang, G.  Yu, W. Zhang, X. Wang, M. B. Jungfleisch, J. E. Pearson, Nat. Phys. 13, 162 (2017).
  • (56) K. Litzius, I. Lemesh, B. Krüger, P. Bassirian, L. Caretta, K. Richter, F. Büttner, K. Sato, O. A. Tretiakov, J. Förster, R. M. Reeve, M. Weigand, I. Bykova, H. Stoll, G. Schütz, G. S. D. Beach and M. Kläui, Nat. Phys. 13, 170 (2017).
  • (57) S. Mühlbauer, B. Binz, F. Jonietz, C. Pfleiderer, A. Rosch, A. Neubauer, R. Georgii and P. Böni, Science 323, 915 (2009).
  • (58) R. Wiesendanger, Nat. Rev. Mater. 1, 16044 (2016).
  • (59) S. Heinze, K. von Bergmann, M. Menzel, J. Brede, A. Kubetzka, R. Wiesendanger, G. Bihlmayer and Stefan Blügel, Nat. Phy. 7, 713 (2011).
  • (60) N. Romming, C. Hanneken, M. Menzel, J. E. Bickel, B. Wolter, K. von Bergmann, A. Kubetzk and R. Wiesendanger, Science 341, 636 (2013).
  • (61) C. Moreau-Luchaire, C. Moutafis, N. Reyren, J. Sampaio, C. A. F. Vaz, N. Van Horne, K. Bouzehouane, K. Garcia, C. Deranlot, P. Warnicke, P. Wohlhüter, J.-M. George, M. Weigand, J. Raabe, V. Cros and A. Fert, Nat. Nanotechnol. 11, 444 (2016).
  • (62) S. Woo, K. Litzius, B. Krüger,Mi-Young Im, L. Caretta, K. Richter, M. Mann, A. Krone, R. M. Reeve, M. Weigand, P. Agrawal, I. Lemesh, Mohamad-Assaad Mawass, P. Fischer, M. Kläui and G. S. D. Beach, Nat. Mater. 15, 501 (2016).
  • (63) F. Jonietz, S. Mühlbaue, C. Pfleiderer, A. Neubauer, W. Münzer, A. Bauer, T. Adams, R. Georgii, P. Böni, R. A. Duine, K. Everschor, M. Garst and A. Rosch, Science 330, 1648 (2010).
  • (64) X. Z. Yu, N. Kanazawa, W.Z. Zhang, T. Nagai, T. Hara, K. Kimoto, Y. Matsui, Y. Onose and Y. Tokura, Nature Commun. 3, 988 (2012).
  • (65) J. Xia, X. Zhang, M. Ezawa, Q. Shao, X. Liu and Y. Zhou, Appl. Phys. Lett. 116, 022407 (2020).
  • (66) A. Fert, V. Cros, and J. Sampaio, Nat. Nanotechnol. 8, 152 (2013).
  • (67) X. Zhang, M. Ezawa, and Y. Zhou, Sci. Rep. 5, 9400 (2015).
  • (68) P. Upadhyaya, G. Yu, P. K. Amiri, and K. L. Wang, Phys. Rev. B 92, 134411 (2015).
  • (69) J. S. White, K. Prša, P. Huang, A. A. Omrani, I. Živković, M. Bartkowiak, H. Berger, A. Magrez, J. L. Gavilano, G. Nagy, J. Zang, and H. M. Rønnow, Phys. Rev. Lett. 113, 107203 (2014).
  • (70) C. Wang, D. Xiao, X. Chen, Y. Zhou, and Y. Liu, New J. Phys. 19, 083008 (2017).
  • (71) L. Kong and J. Zang, Phys. Rev. Lett. 111, 067203 (2013).
  • (72) M. Mochizuki, X. Z. Yu, S. Seki, N. Kanazawa, W. Koshibae, J. Zang, M. Mostovoy, Y. Tokura and N. Nagaosa, Nat. Mater. 13, 241 (2014).
  • (73) L. Zhao, Z. Wang, X. Zhang, X. Liang, J. Xia, K. Wu, H.-A. Zhou, Y. Dong, G. Yu, K. L. Wang, X. Liu, Y. Zhou, and W. Jiang, Phys. Rev. Lett. 125, 027206 (2020).
  • (74) C. Gong, Y. Zhou and G. Zhao, Appl. Phys. Lett. 120, 052402 (2022).
  • (75) K. Shibata, J. Iwasaki, N. Kanazawa, S. Aizawa, T. Tanigaki, M. Shirai, T. Nakajima, M. Kubota, M. Kawasaki, H. S. Park, D. Shindo, N. Nagaosa and Y. Tokura, Nat. Nanotechnol. 10, 589 (2015).
  • (76) P. W. Brouwer, Phys. Rev. B 58, 10135 (1998).
  • (77) I. L. Aleiner and A. V. Andreev, Phys. Rev. Lett. 81, 1286 (1998).
  • (78) B. L. Altshuler and L. I. Glazman, Science 283, 1864 (1999).
  • (79) M. Switkes, C. M. Marcus, K. Campman, and A. C. Gossard, Science 283, 1905 (1999).
  • (80) F. Zhou, B. Spivak, and B. L. Altshuler, Phys. Rev. Lett. 82, 608 (1999).
  • (81) T. A. Shutenko, I. L. Aleiner, and B. L. Altshuler, Phys. Rev. B 61, 10366 (2000).
  • (82) Y. D. Wei, J. Wang, and H. Guo, Phys. Rev. B 62, 9947 (2000).
  • (83) F. Xu, Y. Xing, J. Wang, Phys. Rev. B 84, 245323 (2011).
  • (84) K.-T. Wang, H. Wang, F. Xu, Y. Yu, and Y. Wei, New J. Phys. 25, 013019 (2023).
  • (85) J. E. Avron, A. Elgart, G. M. Graf, and L. Sadun, Phys. Rev. B 62, R10618(R) (2000).
  • (86) M. G. Vavilov, V. Ambegaokar, and I. L. Aleiner, Phys. Rev. B 63, 195313 (2001).
  • (87) B. G. Wang, J. Wang, and H. Guo, Phys. Rev. B 65, 073306 (2002).
  • (88) M. Moskalets and M. Büttiker, Phys. Rev. B 66, 205320 (2002).
  • (89) B. G. Wang, J. Wang, and H. Guo, Phys. Rev. B 68, 155326 (2003).
  • (90) M. Moskalets and M. Büttiker, Phys. Rev. B 66, 035306 (2002).
  • (91) B. G. Wang and J. Wang, Phys. Rev. B 66, 125310 (2002).
  • (92) J. E. Avron, A. Elgart, G. M. Graf, and L. Sadun, Phys. Rev. Lett. 87, 236601 (2001).
  • (93) B. G. Wang and J. Wang, Phys. Rev. B 66, 201305(R) (2002).
  • (94) Y. Levinson, O. Entin-Wohlman, and P. Wölfle, Physica A 302, 335 (2001).
  • (95) J. Wang and B. G. Wang, Phys. Rev. B 65, 153311 (2002).
  • (96) K.-T. Wang, F. Xu, B. Wang, Y. Yu, Y. Wei, Front. Phys. 17, 43501 (2022).
  • (97) J. Wang, Y. D. Wei, B. G. Wang, and H. Guo, Appl. Phys. Lett. 79, 3977 (2001).
  • (98) P. Sharma and C. Chamon, Phys. Rev. Lett. 87, 096401 (2001).
  • (99) Q. F. Sun, H. Guo, and J. Wang, Phys. Rev. B 68, 035318 (2003).
  • (100) B. G. Wang, J. Wang, and H. Guo, Phys. Rev. B 67, 092408 (2003).
  • (101) W. Zheng, J. L. Wu, B. G. Wang, J. Wang, Q. F. Sun, and H. Guo, Phys. Rev. B 68, 113306 (2003).
  • (102) Q. F. Sun, J. Wang, and H. Guo, Phys. Rev. B 71, 165310 (2005).
  • (103) A. Kamenev, Field Theory of Nonequilibrium Systems (Cambridge University Press, 2011).
  • (104) In deriving Eq. (10), we have assumed that 𝐌\mathbf{M} varies slowly with time and its time scale is much larger than that of electrons. Hence the Born-Oppenheimer approximation can be used to separate the degrees of freedom of the magnetization and electrons. It is reasonable since typical time scale for the magnetization dynamics is in the order of nanoseconds while the electronic time scale is in femtoseconds. For an electric pulse, the time scale can be as short as psps which is still much larger than fsfs. In the limit of slow precession, the magnetization dynamics can be approximated as δ𝐌^˙0\delta\dot{\hat{\mathbf{M}}}\approx 0. The magnetization spin is treated as a classical vector, which rotates around the magnetic field. We can further neglect the magnetization fluctuation term (δ𝐌^0\delta\hat{\mathbf{M}}\approx 0). Then the fluctuation comes solely from the electron spin, δ^=Jδ𝐬^D\delta\mathcal{{\hat{B}}}=-J~{}\delta\hat{\mathbf{s}}_{D}.
  • (105) D. V. Berkov and J. Miltat, J. Magn. Magn. Mat. 320, 1238 (2008).
  • (106) K. Gilmore, Y. U. Idzerda, and M. D. Stiles, J. Appl. Phys. 103, 07D303 (2008).
  • (107) M. Sayad and M. Potthoff, New J. Phys. 17, 113058 (2015).
  • (108) D. Steiauf, J. Seib, and M. Fähnle, Phys. Rev. B 78, 020410(R) (2008).
  • (109) Since 𝐬α\mathbf{s}_{\alpha} is a local variable, αd𝐬α/dt\sum_{\alpha}d{\mathbf{s}}_{\alpha}/dt is the total spin current density. Hence we have αd𝐬α/dt=α𝐣αs\sum_{\alpha}d{\mathbf{s}}_{\alpha}/dt=\sum_{\alpha}\nabla\cdot{\bf j}_{\alpha}^{s}.
  • (110) From Eq. (34), we have
    d𝐬Ddt\displaystyle\frac{d{\mathbf{s}}_{D}}{dt} +\displaystyle+ tδ𝐬Db0(𝒋e)𝒎\displaystyle\partial_{t}\delta{\mathbf{s}}_{D}-b_{0}(\nabla\cdot{\bm{j}}_{e}){\bm{m}}
    \displaystyle- b0(𝒋e)𝒎=J𝐌×(𝐬D+δ𝐬D).\displaystyle b_{0}({\bm{j}}_{e}\cdot\nabla){\bm{m}}=J{\mathbf{M}}\times({\mathbf{s}}_{D}+\delta{\mathbf{s}}_{D}).
    When 𝐦{\bf m} is a constant, so is 𝒔D{\bm{s}}_{D}. We have
    tδ𝐬Db0(𝒋e)𝒎=J𝐌×δ𝐬D.\displaystyle\partial_{t}\delta{\mathbf{s}}_{D}-b_{0}({\bm{j}}_{e}\cdot\nabla){\bm{m}}=J{\mathbf{M}}\times\delta{\mathbf{s}}_{D}.
  • (111) W. Wang, M. Beg, B. Zhang, W. Kuch, and H. Fangohr, Phys. Rev. B 92, 020403(R) (2015).
  • (112) B. Zhang, W. Wang, M. Beg, H. Fangohr, and W. Kuch, Appl. Phys. Lett. 106, 102401 (2015).
  • (113) G. Yin, Y. Liu, Y. Barlas, J. Zang, and R. K. Lake, Phys. Rev. B 92, 024411 (2015).
  • (114) P. B. Ndiaye, C. A. Akosa, and A. Manchon, Phys. Rev. B 95, 064426 (2017).
  • (115) L. Arrachea and M. Moskalets, Phys. Rev. B 74, 245322 (2006).
  • (116) L. Zhang, J. Chen, and J. Wang, Phys. Rev. B 87, 205401 (2013).
  • (117) J. Wang, B. Wang, W. Ren, and H. Guo, Phys. Rev. B 74, 155307 (2006).
  • (118) K. Hattori, Phys. Rev. B 75, 205302 (2007).
  • (119) A. Brataas, A. D. Kent, and H. Ohno, Nat. Mater. 11, 372 (2012).
  • (120) H. Haug and A.-P. Jauho, Quantum Kinetics in Transport and Optics of Semiconductors (Springer, Berlin, Heidelberg, New York, 2008).
  • (121) D. S. Fisher and P. A. Lee, Phys. Rev. B 23, 6851 (1981).
  • (122) J. Wang and H. Guo, Phys. Rev. B 79, 045119 (2009).
  • (123) Z. Li and S. Zhang, Phys. Rev. Lett. 92, 207203 (2004).
  • (124) L. Zhang, B. Wang, and J. Wang, Phys. Rev. B 86, 165431 (2012).