betaFeedback

WorksPhysics and Astronomy

Absence of cross-sublattice spin pumping and spin-transfer torques in collinear antiferromagnets

Abstract

We resolve the debate over the existence and magnitude of cross-sublattice (CS) contributions to spin pumping and spin-transfer torques in a two-sublattice antiferromagnet connected to a non-magnetic metal. Guided by symmetry considerations, we first relate the controversial CS terms to specific components in the spin conductance matrix. Then we quantify these components by studying the spin-dependent electron scattering on a fully compensated interface. We ascertain the absence of all CS contributions in the collinear regime. Even in the non-collinear regime, the CS contributions only constitute a higher-order correction to the existing theory.

Introduction

Demystifying the intricate interplay between magnetic excitations and electronic transport is essential for realizing efficient electric control of magnetism, which underlies recent development of spintronics, especially the emerging frontier of sub-terahertz spintronics exploiting the unique dynamics of antiferromagnets (AFMs)[1, 2]. From a fundamental perspective, excitations of magnetic order can be converted into pure spin currents of electrons either coherently or incoherently. While an incoherent spin generation involves an ensemble of thermal magnons to exchange spin angular momenta with electrons, a coherent spin generation, on the other hand, typically involves only a resonance mode and is achieved by virtue of spin pumping[3, 4].

It has been shown that when a collinear AFM characterized by two unit sublattice-magnetic vectors mA\mathbf{m}_{A} and mB\mathbf{m}_{B} is interfaced with a non-magnetic metal (NM), the coherent dynamics of the Néel vector n=(mA−mB)/2\mathbf{n}=(\mathbf{m}_{A}-\mathbf{m}_{B})/2 and of the small magnetic moment m=(mA+mB)/2\mathbf{m}=(\mathbf{m}_{A}+\mathbf{m}_{B})/2 can pump a total pure spin current into the NM in the form of[5, 6, 7]

eℏIs=Gr(n×n˙+m×m˙)−Gim˙,(1)\frac{e}{\hbar}I_{s}=G^{r}(\mathbf{n}\times\dot{\mathbf{n}}+\mathbf{m}\times\dot{\mathbf{m}})-G^{i}\dot{\mathbf{m}}, \tag*{(1)}

where IsI_{s} is measured in units of an electric current (in AmpAmp), ee is the absolute electron charge, ℏ\hbar is the reduced Planck constant, GrG^{r} and GiG^{i} are two independent components of the interfacial spin conductance[8] which can be rigorously calculated by considering the microscopic spin-dependent scattering on the AFM/NM interface[5, 9]. Equation (1) can be equivalently written in terms of the sublattice-magnetic vectors as

eℏIs=GAAmA×m˙A+GBBmB×m˙B−GAm˙A−GBm˙B,(2)\frac{e}{\hbar}I_{s}=G^{AA}\mathbf{m}_{A}\times\dot{\mathbf{m}}_{A}+G^{BB}\mathbf{m}_{B}\times\dot{\mathbf{m}}_{B}-G^{A}\dot{\mathbf{m}}_{A}-G^{B}\dot{\mathbf{m}}_{B}, \tag*{(2)}

where GAA=GBB=Gr/2G^{AA}=G^{BB}=G^{r}/2 and GA=GB=Gi/2G^{A}=G^{B}=G^{i}/2 are the spin conductance coefficients associated with each sublattice.

Recently, coherent spin pumping in collinear AFMs that are described by the above equations has been experimentally verified in a number of materials, notably in MnF2_{2}[10], Cr2_{2}O3_{3}[11], α\alpha–Fe2_{2}O3_{3}[12, 13], and synthetic AFM[14], simulating a vibrant search of new physics in the sub-terahertz frequency range harnessing the unique spin dynamics of AFMs.

Latest theoretical studies[15, 16, 17, 18], however, suggest that Eq. (2) should also admit cross-sublattice (CS) terms GABmA×m˙BG^{AB}\mathbf{m}_{A}\times\dot{\mathbf{m}}_{B} and GBAmB×m˙AG^{BA}\mathbf{m}_{B}\times\dot{\mathbf{m}}_{A}, which in turns changes the Gilbert damping constant into a matrix. While such terms do not contradict directly with existing experimental observations, they could modify the strength of spin pumping predicted by Eqs. (1) and (2), thus affecting the numerical extraction of the interfacial spin conductance involved in different materials. Even more surprising is that the CS terms are believed to exist even for a fully compensated AFM/NM interface in the collinear limit[15]. Furthermore, in the non-collinear regime of a two-sublattice AFM (e.g., the spin-flop phase induced by a strong magnetic field), the established form of spin pumping is questionable, especially whether n×n˙\mathbf{n}\times\dot{\mathbf{n}} and m×m˙\mathbf{m}\times\dot{\mathbf{m}} in Eq. (1) still share the same coefficient[19, 20]. In direct connection with the CS spin pumping, CS spin-transfer torques are allowed by the Onsager reciprocal relations[16, 17, 18], but their existence remains experimentally elusive.

In this Letter, we resolve the puzzle of CS contributions to spin pumping and spin-transfer torques in collinear AFMs from a theoretical perspective. Guided by phenomenological argument, we first clarify a number of essential mathematical relations between spin pumping and spin-transfer torques in the presence of CS contributions, where the controversial CS components are separated and expressed in terms of the corresponding coefficients in the spin conductance matrix. Then we calculate these coefficients by studying the microscopic spin-dependent scattering of electrons off a fully compensated AFM/NM interface. We claim that all CS effects vanish in the collinear regime for fully compensated interfaces, affirming the validity of the established theories [viz. Eqs. (1) and (2)] and the experimental fitting they enable. We find that even in the non-collinear regime, the CS effects only bring about higher-order corrections.

Phenomenological relations

In its most general form, the coherent spin pumping by a two-sublattice AFM into an adjacent NM can be written as

eℏIs=(m,n)×(GmmGmnGnmGnn)(m˙n˙)−Gmm˙,(3)\frac{e}{\hbar}I_{s}=(\mathbf{m},\mathbf{n})\times\begin{pmatrix}G^{mm}&G^{mn}\\G^{nm}&G^{nn}\end{pmatrix}\begin{pmatrix}\dot{\mathbf{m}}\\\dot{\mathbf{n}}\end{pmatrix}-G^{m}\dot{\mathbf{m}}, \tag*{(3)}

which differs (is generalized) from Eq. (1) by the off-diagonal terms GmnG^{mn} and GnmG^{nm} in the matrix of interfacial spin conductance. The dynamics of the AFM can be described by a set of coupled Landau–Lifshitz equations as [21]

ℏm˙=fn×n+fm×m,(4)\hbar\dot{\boldsymbol{m}}=\boldsymbol{f}^{n}\times\boldsymbol{n}+\boldsymbol{f}^{m}\times\boldsymbol{m}, \tag*{(4)}
ℏn˙=fm×n+fn×m,(5)\hbar\dot{\boldsymbol{n}}=\boldsymbol{f}^{m}\times\boldsymbol{n}+\boldsymbol{f}^{n}\times\boldsymbol{m}, \tag*{(5)}

where the Gilbert damping is omitted for simplicity, fm=−∂ε/∂m\boldsymbol{f}^{m}=-\partial\varepsilon/\partial\boldsymbol{m} and fn=−∂ε/∂n\boldsymbol{f}^{n}=-\partial\varepsilon/\partial\boldsymbol{n} are the effective fields (or driving forces) with ε\varepsilon being the magnetic free energy. In our convention, fm\boldsymbol{f}^{m} and fn\boldsymbol{f}^{n} are scaled in units of energy. By inserting Eqs. (4) into Eq. (3), we can relate IsI_s to these driving forces and establish a linear response relation

Is,i=Lijsmfjm+Lijsnfjn,(i,j run over x,y,z)(6)I_{s,i}=L_{ij}^{sm}f_j^m+L_{ij}^{sn}f_j^n,\qquad(i,j\ \text{run over }x,y,z) \tag*{(6)}

where the response coefficients Lijsm(m,n)L_{ij}^{sm}(\boldsymbol{m},\boldsymbol{n}) and Lijsn(m,n)L_{ij}^{sn}(\boldsymbol{m},\boldsymbol{n}) are related to the spin conductance. As the inverse effect of spin pumping, the spin-transfer torques can be expressed as Tim=LijmsVjsT_i^m=L_{ij}^{ms}V_j^s and Tin=LijnsVjsT_i^n=L_{ij}^{ns}V_j^s, where Vs=μs/e\boldsymbol{V}_s=\boldsymbol{\mu}_s/e is the spin voltage with μs=(μ↑−μ↓)s^\boldsymbol{\mu}_s=(\mu_{\uparrow}-\mu_{\downarrow})\hat{\boldsymbol{s}} being the spin chemical potential (s^\hat{\boldsymbol{s}} specifies the quantization axis). These response coefficients must satisfy the Onsager reciprocal relation

Lijms(m,n)=Ljisn(−m,−n),(7)L_{ij}^{ms}(\boldsymbol{m},\boldsymbol{n})=L_{ji}^{sn}(-\boldsymbol{m},-\boldsymbol{n}), \tag*{(7)}

as both m\boldsymbol{m} and n\boldsymbol{n} break the time-reversal symmetry. An identical relation is applicable to LijnsL_{ij}^{ns} and LjisnL_{ji}^{sn} as well. When Vs\boldsymbol{V}_s is treated as a common driving force, Lms(ns)L^{ms(ns)} and Lsm(sn)L^{sm(sn)} will share the same unit, which simplifies the following discussions. After some straightforward algebra, we find

eTm=Gmmm×(Vs×m)+Gmnn×(Vs×m)+Gnmm×(Vs×n)+Gnnn×(Vs×n)+Gmm×Vs,(8)e\boldsymbol{T}^{m}=G^{mm}\boldsymbol{m}\times(\boldsymbol{V}_s\times\boldsymbol{m})+G^{mn}\boldsymbol{n}\times(\boldsymbol{V}_s\times\boldsymbol{m}) +G^{nm}\boldsymbol{m}\times(\boldsymbol{V}_s\times\boldsymbol{n})+G^{nn}\boldsymbol{n}\times(\boldsymbol{V}_s\times\boldsymbol{n}) +G^{m}\boldsymbol{m}\times\boldsymbol{V}_s, \tag*{(8)}
eTn=Gmmn×(Vs×m)+Gmnm×(Vs×m)+Gnmn×(Vs×n)+Gnnm×(Vs×n)+Gnn×Vs,(9)e\boldsymbol{T}^{n}=G^{mm}\boldsymbol{n}\times(\boldsymbol{V}_s\times\boldsymbol{m})+G^{mn}\boldsymbol{m}\times(\boldsymbol{V}_s\times\boldsymbol{m}) +G^{nm}\boldsymbol{n}\times(\boldsymbol{V}_s\times\boldsymbol{n})+G^{nn}\boldsymbol{m}\times(\boldsymbol{V}_s\times\boldsymbol{n}) +G^{n}\boldsymbol{n}\times\boldsymbol{V}_s, \tag*{(9)}

where all spin-transfer torques have been scaled into the inverse-time dimension so that Tm\boldsymbol{T}^{m} and Tn\boldsymbol{T}^{n} can be directly added to Eqs. (4).

The spin-transfer torques exerting on the two sublattice-magnetic moments, mA\boldsymbol{m}_A and mB\boldsymbol{m}_B, are TA=Tm+Tn\boldsymbol{T}^{A}=\boldsymbol{T}^{m}+\boldsymbol{T}^{n} and TB=Tm−Tn\boldsymbol{T}^{B}=\boldsymbol{T}^{m}-\boldsymbol{T}^{n}. A simple manipulation of Eq. (7) shows that

TA=τDAAmA×(Vs×mA)+τCSABmA×(Vs×mB)+τFAmA×Vs,(10)\boldsymbol{T}^{A}=\tau_D^{AA}\boldsymbol{m}_A\times(\boldsymbol{V}^{s}\times\boldsymbol{m}_A)+\tau_{CS}^{AB}\boldsymbol{m}_A\times(\boldsymbol{V}^{s}\times\boldsymbol{m}_B)+\tau_F^{A}\boldsymbol{m}_A\times\boldsymbol{V}^{s}, \tag*{(10)}
TB=τDBBmB×(Vs×mB)+τCSBAmB×(Vs×mA)+τFBmB×Vs,(11)\boldsymbol{T}^{B}=\tau_D^{BB}\boldsymbol{m}_B\times(\boldsymbol{V}^{s}\times\boldsymbol{m}_B)+\tau_{CS}^{BA}\boldsymbol{m}_B\times(\boldsymbol{V}^{s}\times\boldsymbol{m}_A)+\tau_F^{B}\boldsymbol{m}_B\times\boldsymbol{V}^{s}, \tag*{(11)}

where τDAA(BB)\tau_D^{AA(BB)}, τCSAB(BA)\tau_{CS}^{AB(BA)} and τFA(B)\tau_F^{A(B)} represent the coefficients of the damping-like torques, the CS torques, and the field-like torques, respectively. To relate these torques to the spin conductance matrix appearing in Eq. (3), we now define

GAA(BB)=eτDAA(BB)/2,GAB(BA)=eτCSAB(BA)/2,GA(B)=eτFA(B),G^{AA(BB)}=e\tau_D^{AA(BB)}/2,\qquad G^{AB(BA)}=e\tau_{CS}^{AB(BA)}/2,\qquad G^{A(B)}=e\tau_F^{A(B)},

which satisfy

GAA=14(Gmm+Gmn+Gnm+Gnn),(12)G^{AA}=\frac{1}{4}\left(G^{mm}+G^{mn}+G^{nm}+G^{nn}\right), \tag*{(12)}
GBB=14(Gmm−Gmn−Gnm+Gnn),(13)G^{BB}=\frac{1}{4}\left(G^{mm}-G^{mn}-G^{nm}+G^{nn}\right), \tag*{(13)}
GAB=14(Gmm+Gmn−Gnm−Gnn),(14)G^{AB}=\frac{1}{4}\left(G^{mm}+G^{mn}-G^{nm}-G^{nn}\right), \tag*{(14)}
GBA=14(Gmm−Gmn+Gnm−Gnn),(15)G^{BA}=\frac{1}{4}\left(G^{mm}-G^{mn}+G^{nm}-G^{nn}\right), \tag*{(15)}

and GA=GB=GmG^{A}=G^{B}=G^{m}. By invoking the Onsager reciprocal relations, we obtain

eℏIs=(mA,mB)×(GAAGABGBAGBB)(m˙Am˙B)−GAm˙A−GBm˙B,(16)\frac{e}{\hbar}I_s=(\boldsymbol{m}_A,\boldsymbol{m}_B)\times \begin{pmatrix} G^{AA} & G^{AB}\\ G^{BA} & G^{BB} \end{pmatrix} \begin{pmatrix} \dot{\boldsymbol{m}}_A\\ \dot{\boldsymbol{m}}_B \end{pmatrix} -G^{A}\dot{\boldsymbol{m}}_A-G^{B}\dot{\boldsymbol{m}}_B, \tag*{(16)}

which incorporates Eq. (2) as a special case when the CS terms vanish (GAB=GBA=0G^{AB}=G^{BA}=0). Because combining Eqs. (9) and (10) can reproduce Eq. (3) under the conditions of n=(mA−mB)/2\boldsymbol{n}=(\boldsymbol{m}_A-\boldsymbol{m}_B)/2 and m=(mA+mB)/2\boldsymbol{m}=(\boldsymbol{m}_A+\boldsymbol{m}_B)/2, we have established consistent relations between spin pumping and spin-transfer torques in the presence of CS contributions, which hold in both the (m,n)(\boldsymbol{m},\boldsymbol{n}) and the (mA,mB)(\boldsymbol{m}_A,\boldsymbol{m}_B) basis.

By imposing symmetry constraints on mA\boldsymbol{m}_A and mB\boldsymbol{m}_B (or m\boldsymbol{m} and n\boldsymbol{n}), one can reduce the number of independent variables in the matrix of spin conductance. For an insulating AFM, practically only the magnetic layer in direct contact with the NM is relevant. For an uncompensated interface such as the (111) plane of NiO, the conduction electrons only couple to one sublattice so the CS terms become irrelevant. Spin pumping in this special case is physically equivalent to its ferromagnetic counterpart[5, 9]. On the contrary, for a fully compensated interface, there is an effective PTPT symmetry requiring that τDAA=τDBB\tau_D^{AA}=\tau_D^{BB} and τCSAB=τCSBA\tau_{CS}^{AB}=\tau_{CS}^{BA}, which, according to Eqs. (9), yields Gmn=Gnm=0G^{mn}=G^{nm}=0. What deserves special consideration is the case of partially compensated interfaces, which typically arises when a compensated surface is subject to interfacial roughness, nonequivalent electron couplings to the two sublattices, and so on, breaking the effective PTPT symmetry. In the presence of TA\boldsymbol{T}^{A} and TB\boldsymbol{T}^{B} defined by Eqs. (8), the effective dynamics of the AFM can be formally written as

ℏm˙A=(fAA+fAB)×mA,(17)\hbar\dot{\boldsymbol{m}}_A=(\boldsymbol{f}_{AA}+\boldsymbol{f}_{AB})\times\boldsymbol{m}_A, \tag*{(17)}
ℏm˙B=(fBA+fBB)×mB,(18)\hbar\dot{\boldsymbol{m}}_B=(\boldsymbol{f}_{BA}+\boldsymbol{f}_{BB})\times\boldsymbol{m}_B, \tag*{(18)}

where the diagonal fields fAA(BB)\boldsymbol{f}_{AA(BB)} involve the single-ion anisotropy, the external Zeeman field and the ordinary spin torques τDAA(BB)\tau_D^{AA(BB)} and τFA(B)\tau_F^{A(B)}, whereas the off-diagonal fields fAB(BA)\boldsymbol{f}_{AB(BA)} include the exchange interaction between mA\boldsymbol{m}_A and mB\boldsymbol{m}_B as well as the CS torques τCSAB\tau_{CS}^{AB} and τCSBA\tau_{CS}^{BA}. Regarding Vs×mB\boldsymbol{V}^{s}\times\boldsymbol{m}_B and Vs×mA\boldsymbol{V}^{s}\times\boldsymbol{m}_A as two reciprocal driving forces cross-linking mA\boldsymbol{m}_A with mB\boldsymbol{m}_B, we can apply the Onsager reciprocity relation to fAB\boldsymbol{f}_{AB} and fBA\boldsymbol{f}_{BA}, which gives rise to τCSAB=τCSBA\tau_{CS}^{AB}=\tau_{CS}^{BA} even though τCSAA\tau_{CS}^{AA} could differ from τCSBB\tau_{CS}^{BB}. This means that on a partially compensated interface without the effective PTPT symmetry, we still have Gmn=GnmG^{mn}=G^{nm} (hence GAB=GBAG^{AB}=G^{BA}) based on the relations in Eqs. (9), but this time GmnG^{mn} could be finite (hence GAAG^{AA} and GBBG^{BB} are different). Consequently, for compensated AFM/NM interfaces, we have the general relations

τDAA(BB)=12e(Gmm+Gnn±2Gmn)(19)\tau_{\mathrm{D}}^{AA(BB)}=\frac{1}{2e}\left(G^{mm}+G^{nn}\pm2G^{mn}\right) \tag*{(19)}
τCSAB=τCSBA=12e(Gmm−Gnn)(20)\tau_{\mathrm{CS}}^{AB}=\tau_{\mathrm{CS}}^{BA}=\frac{1}{2e}\left(G^{mm}-G^{nn}\right) \tag*{(20)}

where Gmn→0G^{mn}\to0 and τDAA→τDBB\tau_{\mathrm{D}}^{AA}\to\tau_{\mathrm{D}}^{BB} as the interface becomes fully compensating. In any case, the CS terms could only exist when Gmm≠GnnG^{mm}\ne G^{nn}. Therefore, to resolve the puzzle of CS contributions, we need to quantify and compare GmmG^{mm} and GnnG^{nn}. Note, however, that if the two sublattices are chemically nonequivalent, the above analysis is invalid, which belongs to the category of ferrimagnets[22].

Microscopic calculations

The microscopic origin of GmmG^{mm} and GnnG^{nn} pertains to the spin-dependent scattering of electrons off an AFM/NM interface. Without losing generality, we consider a fully compensated interface with a simple cubic lattice[23] as schematically illustrated in Figure 1. Here we ignore interfacial roughness because it incurs a random spin distribution that destroys the periodic magnetic structure in the lateral dimension, thus invalidating any meaningful band calculation. Following the wavefunction matching approach detailed in Chapter 4 of Ref. 9, we shall determine the scattering matrix S\mathbf{S} of the form

Figure 1. A fully compensated AFM/NM interface with cubic lattice, where tmt_m (tnt_n) is the hopping energy in the AFM (NM), and a0a_0 is the lattice constant. On the interface plane, the magnetic unit cell is indicated by green dashed circles, which are periodic in both the x^\hat{x} ([1,1,0]) and y^\hat{y} ([1 -1,0]) directions.

S=(S++S+−S−+S−−)(21)\mathbf{S}=\begin{pmatrix}S^{++}&S^{+-}\\S^{-+}&S^{--}\end{pmatrix} \tag*{(21)}

where each block is a 2×22\times2 matrix in the spin space and ±\pm accounts for the sublattice (pseudo-spin) degree of freedom. Under the adiabatic condition (i.e., the dynamics of m\mathbf{m} and n\mathbf{n} is much slower than the electron relaxation), we have[24]

S±±≈S0±±σ0+Sm±±m⋅σ,(22)S^{\pm\pm}\approx S_{0}^{\pm\pm}\sigma_{0}+S_{m}^{\pm\pm}\mathbf{m}\cdot\boldsymbol{\sigma}, \tag*{(22)}
S±∓≈Snn⋅σ±Smn(m×n)⋅σ,(23)S^{\pm\mp}\approx S_{n}\mathbf{n}\cdot\boldsymbol{\sigma}\pm S_{mn}(\mathbf{m}\times\mathbf{n})\cdot\boldsymbol{\sigma}, \tag*{(23)}

where σ={σx,σy,σz}\boldsymbol{\sigma}=\{\sigma_x,\sigma_y,\sigma_z\} is the vector of Pauli spin matrices and σ0\sigma_0 is the identity matrix. The complete expressions of S0±±S_{0}^{\pm\pm}, Sm±±S_{m}^{\pm\pm}, SnS_n and SmnS_{mn} in terms of the crystal momentum and other material parameters are provided in the Supplementary Material[25]. If we turn to the collinear regime, ∣n∣2≈1|\mathbf{n}|^2\approx1 and ∣m∣2≪1|\mathbf{m}|^2\ll1, only S±±S^{\pm\pm} and SnS_n in Eqs. (15) will remain essential, then the spin-flip scattering will be necessarily accompanied by the reversal of pseudo-spin[26]. If we go beyond the collinear regime, however, the locking between spin and pseudo-spin will be lifted. We also notice that previous studies assumed Sm±±≈SnS_m^{\pm\pm}\approx S_n in the collinear regime without a rigorous justification[5, 9], so here in a general context we treat all components in Eq. (15) as independent quantities.

The pumped spin current polarized in the jj direction (j=x,y,zj=x,y,z) can be calculated by[5, 9]

Is,j=−e2πIm⁡{Tr⁡[S˙†(σ0⊗σj)S˙]},(24)I_{s,j}=-\frac{e}{2\pi}\operatorname{Im}\left\{\operatorname{Tr}\left[\dot{S}^{\dagger}(\sigma_0\otimes\sigma_j)\dot{S}\right]\right\}, \tag*{(24)}

which, after some tedious algebra, ends up with

eℏIs=G0nnn×n˙+G0mmm×m˙−Gmm˙+ΔIs,(25)\frac{e}{\hbar}I_s=G_0^{nn}\mathbf{n}\times\dot{\mathbf{n}}+G_0^{mm}\mathbf{m}\times\dot{\mathbf{m}}-G^m\dot{\mathbf{m}}+\Delta I_s, \tag*{(25)}

where

ΔIs=ΔG{[n⋅(m×n˙)]n+m⋅(n×n˙)m}=ΔG{(n×n˙)∣m∣2+(m×m˙)∣n∣2+2(n×m)(m⋅n˙)}(26)\begin{aligned} \Delta I_s=\Delta G\left\{\left[\mathbf{n}\cdot(\mathbf{m}\times\dot{\mathbf{n}})\right]\mathbf{n}+\mathbf{m}\cdot(\mathbf{n}\times\dot{\mathbf{n}})\mathbf{m}\right\} \\ &=\Delta G\left\{\left(\mathbf{n}\times\dot{\mathbf{n}}\right)|\mathbf{m}|^2+(\mathbf{m}\times\dot{\mathbf{m}})|\mathbf{n}|^2+2(\mathbf{n}\times\mathbf{m})(\mathbf{m}\cdot\dot{\mathbf{n}})\right\} \tag*{(26)} \end{aligned}

is a (higher-order) correction term never claimed before. After a straightforward re-ordering of terms, Eq. (25) can be effectively written as

eℏIs=(G0nn+∣m∣2ΔG)n×n˙+(G0mm+∣n∣2ΔG)m×m˙−Gmm˙+2ΔG(n×m)(m⋅n˙),(27)\frac{e}{\hbar}I_s=\left(G_0^{nn}+|\mathbf{m}|^2\Delta G\right)\mathbf{n}\times\dot{\mathbf{n}}+\left(G_0^{mm}+|\mathbf{n}|^2\Delta G\right)\mathbf{m}\times\dot{\mathbf{m}}-G^m\dot{\mathbf{m}}+2\Delta G(\mathbf{n}\times\mathbf{m})(\mathbf{m}\cdot\dot{\mathbf{n}}), \tag*{(27)}

where GmmG^{mm} and GnnG^{nn} indeed vanish as required by the PTPT symmetry; they could appear only as a consequence of imperfections (such as roughness) on a compensated interface. Eq. (27) is a generalized spin pumping formula involving four independent components of the interfacial spin conductance. These terms are determined by the corresponding components of the scattering matrix as

G0nn=e2Aℏπ2∫∣Sn∣2 d2k,(28)G_0^{nn}=\frac{e^2\mathcal{A}}{\hbar\pi^2}\int|S_{n}|^2\,d^2k, \tag*{(28)}
G0mm=e2A2ℏπ2∫(∣Sm++∣2+∣Sm−−∣2) d2k,(29)G_0^{mm}=\frac{e^2\mathcal{A}}{2\hbar\pi^2}\int\left(|S_m^{++}|^2+|S_m^{--}|^2\right)\,d^2k, \tag*{(29)}
Gm=e2A2ℏπ2∫Im⁡[(S0++)∗Sm+++(S0−−)∗Sm−−] d2k,(30)G^m=\frac{e^2\mathcal{A}}{2\hbar\pi^2}\int\operatorname{Im}\left[(S_0^{++})^*S_m^{++}+(S_0^{--})^*S_m^{--}\right]\,d^2k, \tag*{(30)}
ΔG=e2Aℏπ2∫∣Smn∣2 d2k,(31)\Delta G=\frac{e^2\mathcal{A}}{\hbar\pi^2}\int|S_{mn}|^2\,d^2k, \tag*{(31)}

where d2k=dkxdkyd^2k=dk_xdk_y and A\mathcal{A} is the area of the interface. Except the last term proportional to ΔG\Delta G in Eq. (27), we can read off the effective spin conductance GnnG^{nn} and GmmG^{mm} as

Gnn=G0nn+∣m∣2ΔG,(32)G^{nn}=G_0^{nn}+|\mathbf{m}|^2\Delta G, \tag*{(32)}
Gmm=G0mm+∣n∣2ΔG.(33)G^{mm}=G_0^{mm}+|\mathbf{n}|^2\Delta G. \tag*{(33)}

and, according to Eqs. (19) and (20), we obtain the damping-like torques and the CS torques as

τDAA=τDBB=12e(G0mm+G0nn+ΔG),(34)\tau_D^{AA}=\tau_D^{BB}=\frac{1}{2e}\left(G_0^{mm}+G_0^{nn}+\Delta G\right), \tag*{(34)}
τCSAB=τCSBA=12e[G0mm−G0nn+(∣n∣2−∣m∣2)ΔG],(35)\tau_{CS}^{AB}=\tau_{CS}^{BA}=\frac{1}{2e}\left[G_0^{mm}-G_0^{nn}+\left(\lvert n\rvert^2-\lvert m\rvert^2\right)\Delta G\right], \tag*{(35)}

where ∣n∣2+∣m∣2=1\lvert n\rvert^2+\lvert m\rvert^2=1 is used. The above results are valid even in the noncollinear regime.

Finally, we point out that the last term in Eq. (27), pursuant to the Onsager reciprocal relations, gives rise to an additional CS torque

ΔTA=ΔTB=ΔGe[Vs⋅(m1×m2)](m2×m1),(36)\Delta T^{A}=\Delta T^{B}=\frac{\Delta G}{e}\left[\boldsymbol{V}^{s}\cdot(\boldsymbol{m}_1\times\boldsymbol{m}_2)\right](\boldsymbol{m}_2\times\boldsymbol{m}_1), \tag*{(36)}

which is nonlinear in m1×m2\boldsymbol{m}_1\times\boldsymbol{m}_2 thus not being captured by the phenomenological consideration in the previous section. In the collinear regime, this term is negligible.

Numerical results

In Figure 2, we numerically plot the four relevant components of the spin conductance basing on Eq. (20) as functions of the exchange coupling JJ (between the conduction electrons and the magnetic moments) and the ratio of kinetic energies in the AFM and NM (i.e., hopping integrals tmt_m and tt). Here, the spin conductance is expressed in units of e2/he^2/h per a02a_0^2 (area of a magnetic unit cell on the interface), which should be multiplied by the number of magnetic unit cells NN on the interface to retrieve the total spin conductance. Comparing Figure 2(a) and (b), we find that G0nnG_0^{nn} and G0mmG_0^{mm} share a very similar pattern as they both culminate around J/t=1J/t=1 and tm/t=0.5t_m/t=0.5. They are the dominant contributions to the damping-like torques [see Eq. (22a)]. Figure 2(c) for GmG^m, on the other hand, shows how the strength of field-like torques varies over JJ and tmt_m. It is clear that the damping-like (field-like) torques dominate the strong (weak) exchange coupling regime, which is corroborated by a recent experiment[27].

Interfacial spin conductance maps

Figure 2. Interfacial spin conductance per unit-cell area (e2/he^2/h per a02a_0^2) as a function of J/tJ/t and tm/tt_m/t. Here, JJ is the exchange coupling between the conduction electrons and the magnetic moments on the compensated interface, and tmt_m (tt) is the hopping energy on the AFM (NM) side of the interface.

In the collinear limit that ∣n∣2→1\lvert n\rvert^2\to1 and ∣m∣2→0\lvert m\rvert^2\to0, the CS torques, according to Eq. (22b), reduce to

τCSAB(BA)=12e(G0mm+ΔG−G0nn).(37)\tau_{CS}^{AB(BA)}=\frac{1}{2e}\left(G_0^{mm}+\Delta G-G_0^{nn}\right). \tag*{(37)}

Using the numerical results shown in Figure 2, we find that

G0nn=G0mm+ΔG,(38)G_0^{nn}=G_0^{mm}+\Delta G, \tag*{(38)}

which renders all CS torques exactly zero. As a matter of fact, regarding the integrands in Eqs. (20) as functions of the crystal momentum, one can rigorously prove that Eq. (38) is an exact identify (see details in the Supplementary Material[25]). It is interesting that even in the highly non-collinear regime where ∣m∣2\lvert m\rvert^2 is comparable to ∣n∣2\lvert n\rvert^2, the CS torques [proportional to G0mm−G0nn+(∣n∣2−∣m∣2)ΔGG_0^{mm}-G_0^{nn}+\left(\lvert n\rvert^2-\lvert m\rvert^2\right)\Delta G] is at most a few percents of the damping-like torques since ΔG\Delta G is much smaller than G0nnG_0^{nn} and G0mmG_0^{mm}, as shown in Figure 2(d).

In conclusion, we have justified the absence of CS contributions to the coherent spin pumping and spin-transfer torques in the collinear regime of two-sublattice AFMs, substantiating the data extraction and fitting approach based on Eqs. (1) and (2) in explaining a number of recent experiments[10, 11, 12, 13, 14].

Acknowledgments

The authors acknowledge fruitful discussions with Hantao Zhang. This work is support by the Air Force Office of Scientific Research (Grant No. FA9550-19-1-0307).

Supplementary Material

See the supplementary materials for more mathematical details about the interfacial spin conductance.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.

References

  1. [1]V. Baltz, A. Manchon, M. Tsoi, T. Moriyama, T. Ono, and Y. Tserkovnyak, “Antiferromagnetic spintronics,” Rev. Mod. Phys. 90, 015005 (2018).DOI
  2. [2]J. Han, R. Cheng, L. Liu, H. Ohno, and S. Fukami, “Coherent antiferromagnetic spintronics,” Nature Materials 1–12 (2023).DOI
  3. [3]Y. Tserkovnyak, A. Brataas, and G. E. W. Bauer, “Spin pumping and magnetization dynamics in metallic multilayers,” Phys. Rev. B 66, 224403 (2002).DOI
  4. [4]Y. Tserkovnyak, A. Brataas, and G. E. W. Bauer, “Enhanced gilbert damping in thin ferromagnetic films,” Phys. Rev. Lett. 88, 117601 (2002).pubmed.ncbi.nlm.nih.gov/11909427
  5. [5]R. Cheng, J. Xiao, Q. Niu, and A. Brataas, “Spin pumping and spin-transfer torques in antiferromagnets,” Phys. Rev. Lett. 113, 057601 (2014).DOI
  6. [6]O. Johansen and A. Brataas, “Spin pumping and inverse spin hall voltages from dynamical antiferromagnets,” Phys. Rev. B 95, 224408 (2017).DOI
  7. [7]H. V. Gomonay and V. M. Loktev, “Spin transfer and current-induced switching in antiferromagnets,” Phys. Rev. B 81, 144427 (2010).DOI
  8. [8]In literature, G^r and G^i are often dubbed the real and imaginary parts of the “spin-mixing” conductance, which we believe is not an accurate nomenclature. In an AFM, the spin and sublattice degrees of freedom are entangled, so “spin-mixing” does not fully capture the underlying physics, if not entirely misleading. From the perspective of spin-dependent scattering, G^r and G^i are related to different components of the scattering matrix, which cannot be clearly distinguished by “real” and “imaginary” (even though they are true in ferromagnets). Therefore, we elect to just call them different components of the (interfacial) spin conductance.
  9. [9]R. Cheng, Aspects of antiferromagnetic spintronics, Ph.D. thesis, The University of Texas at Austin (2014).
  10. [10]P. Vaidya, S. A. Morley, J. van Tol, Y. Liu, R. Cheng, A. Brataas, D. Lederman, and E. Del Barco, “Subterahertz spin pumping from an insulating antiferromagnet,” Science 368, 160–165 (2020).DOI
  11. [11]J. Li, C. B. Wilson, R. Cheng, M. Lohmann, M. Kavand, W. Yuan, M. Aldosary, N. Agldaze, P. Wei, M. S. Sherwin, and J. Shi, “Spin current from sub-terahertz-generated antiferromagnetic magnons,” Nature 578, 70–74 (2020).DOI
  12. [12]H. Wang, Y. Xiao, M. Guo, E. Lee-Wong, G. Q. Yan, R. Cheng, and C. R. Du, “Spin pumping of an easy-plane antiferromagnet enhanced by dzyaloshinskii–moriya interaction,” Phys. Rev. Lett. 127, 117202 (2021).DOI
  13. [13]I. Boventer, H. T. Simensen, A. Name, M. Kläui, A. Brataas, and R. Lebrun, “Room-temperature antiferromagnetic resonance and inverse spin-hall voltage in canted antiferromagnets,” Phys. Rev. Lett. 126, 187201 (2021).DOI
  14. [14]Y. Liu, Z. Xu, L. Liu, K. Zhang, Y. Meng, Y. Sun, P. Gao, H.-W. Zhao, Q. Niu, and J. Li, “Switching magnon chirality in artificial ferrimagnet,” Nature communications 13, 1264 (2022).DOI
  15. [15]A. Kamra and W. Belzig, “Spin pumping and shot noise in ferrimagnets: Bridging ferro- and antiferromagnets,” Phys. Rev. Lett. 119, 197201 (2017).DOI
  16. [16]A. Kamra, R. E. Troncoso, W. Belzig, and A. Brataas, “Gilbert damping phenomenology for two-sublattice magnets,” Phys. Rev. B 98, 184402 (2018).DOI
  17. [17]R. E. Troncoso, M. A. Lund, A. Brataas, and A. Kamra, “Cross-sublattice spin pumping and magnon level attraction in van der waals antiferromagnets,” Phys. Rev. B 103, 144422 (2021).DOI
  18. [18]H. Yuan, Q. Liu, K. Xia, Z. Yuan, and X. Wang, “Proper dissipative torques in antiferromagnetic dynamics,” Europhysics Letters 126, 67006 (2019).DOI
  19. [19]D. Reitz, J. Li, W. Yuan, J. Shi, and Y. Tserkovnyak, “Spin seebeck effect near the antiferromagnetic spin-flop transition,” Phys. Rev. B 102, 020408 (2020).DOI
  20. [20]R. Rodriguez, S. Regmi, H. Zhang, W. Yuan, P. Makushko, E. A. Montoya, I. Veremchuk, R. Hübner, D. Makarov, J. Shi, R. Cheng, and I. Barsukov, “Robust spin injection via thermal magnon pumping in antiferromagnet/ferromagnet hybrid systems,” Phys. Rev. Res. 4, 033139 (2022).DOI
  21. [21]K. M. D. Hals, Y. Tserkovnyak, and A. Brataas, “Phenomenology of current-induced dynamics in antiferromagnets,” Phys. Rev. Lett. 106, 107206 (2011).pubmed.ncbi.nlm.nih.gov/21469831
  22. [22]M. Guo, H. Zhang, and R. Cheng, “Manipulating ferrimagnets by fields and currents,” Phys. Rev. B 105, 064410 (2022).DOI
  23. [23]We use a cubic lattice because it is simple yet representative. For other lattices without geometrical frustration, the result will be qualitatively the same. The same lattice structure is adopted in previous studies⁵,⁹.
  24. [24]Here we have truncated the iterative solution at the linear order in m, which has a magnitude |m| < 1 even in the non-collinear regime.
  25. [25]See the supplementary materials for more mathematical details about the interfacial spin conductance.
  26. [26]P. M. Haney and A. H. MacDonald, “Current-induced torques due to compensated antiferromagnets,” Phys. Rev. Lett. 100, 196801 (2008).pubmed.ncbi.nlm.nih.gov/18518471
  27. [27]E. Cogulu, H. Zhang, N. N. Statuto, Y. Feng, F. Yang, R. Cheng, and A. D. Kent, “Quantifying spin-orbit torques in antiferromagnet–heavy-metal heterostructures,” Phys. Rev. Lett. 128, 247204 (2022).DOI

Paper details

vlnc:W_vu4c7nah
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 2

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents