betaFeedback

WorksPhysics and Astronomy

Magnetically induced polarization in centrosymmetric bonds

Abstract

We reveal the microscopic origin of electric polarization P⃗\vec{P} induced by noncollinear magnetic order. We show that in Mott insulators, such P⃗\vec{P} is given by all possible combinations of position operators r⃗^ij=(r⃗ij 0,r⃗ij0)\hat{\vec{r}}_{ij} = (\vec{r}_{ij}^{\, 0},\vec{\boldsymbol{r}}_{ij}^{\phantom{0}}) and transfer integrals t^ij=(tij0,tij0)\hat{t}_{ij} = (t_{ij}^{0},\boldsymbol{t}_{ij}^{\phantom{0}}) in the bonds, where r⃗ij 0\vec{r}_{ij}^{\, 0} and tij0t_{ij}^{0} are spin-independent contributions in the basis of Kramers doublet states, while r⃗ij0\vec{\boldsymbol{r}}_{ij}^{\phantom{0}} and tij0\boldsymbol{t}_{ij}^{\phantom{0}} stem solely from the spin-orbit interaction. Among them, the combination tij0r⃗ij0t_{ij}^{0} \vec{\boldsymbol{r}}_{ij}^{\phantom{0}}, which couples to the spin current, remains finite in the centrosymmetric bonds, thus yielding finite P⃗\vec{P} in the case of noncollinear arrangement of spins. The form of the magnetoelectric coupling, which is controlled by r⃗ij0\vec{\boldsymbol{r}}_{ij}^{\phantom{0}}, appears to be rich and is not limited to the phenomenological law P⃗∼εij×[ei×ej]\vec{P} \sim \boldsymbolε_{ij} \times [\boldsymbol{e}_{i} \times \boldsymbol{e}_{j}] with εij\boldsymbolε_{ij} being the bond vector connecting the spins ei\boldsymbol{e}_{i} and ej\boldsymbol{e}_{j}. Using density-functional theory, we illustrate how the proposed mechanism work in the spiral magnets CuCl2_2, CuBr2_2, CuO, and αα-Li2_2IrO3_3, providing consistent explanation to available experimental data.

Introduction

In order to make a material ferroelectric, it is essential to break the inversion symmetry. Canonically, this implied some crystallographic instability towards polar atomic displacements [1]. Nevertheless, there is a very special class of materials, called multiferroics, where inversion symmetry can be broken by magnetic means in an otherwise perfect crystallographically centrosymmetric lattice [2]. The microscopic origin of multiferroicity is the fundamental physical problem and its practical realization is an important step towards mutual control of electric polarization and magnetism in novel electronic devices.

There can be various scenarios of the magnetic inversion symmetry breaking. In certain multiferroic materials, the inversion symmetry is microscopically broken by local distortions, so that individual bonds can be formally associated with some polarization vectors. When arranged in an antiferroelectric manner, these bonds result in zero net polarization. Nevertheless, if some of them become inequivalent owing to complex magnetic order, the perfect cancellation of polarization vectors does not occur and the system becomes ferroelectric.

However, what if the bond itself is centrosymmetric? Can it become electrically polarized by magnetic means? An affirmative answer to these questions was given by Katsura, Nagaosa, and Balatsky (KNB) [3], who considered a very special microscopic model and argued that the noncollinear alignment of spins can induce the polarization P⃗∼ϵ⃗ij×[e⃗i×e⃗j]\vec{P}\sim\vec{\epsilon}_{ij}\times[\vec{e}_{i}\times\vec{e}_{j}], which lies in the plane of spins e⃗i\vec{e}_{i} and e⃗j\vec{e}_{j} and is perpendicular to the bond vector ϵ⃗ij\vec{\epsilon}_{ij}. This finding was supported by phenomenological considerations [4] and the proposed mechanism was called the “spin-current mechanism”, which is widely used for the analysis of magnetoelectric (ME) coupling in spiral magnets [5], typically in combination with two other mechanisms – the “exchange striction” [6, 7] and “spin-dependent pp-dd hybridization” [8].

Nevertheless, the analysis remains largely phenomenological. First, the properties of all spiral multiferroics are usually discussed from the viewpoint of the spin-current model [9–14]. However, there are only few materials, such as CuCl2_2 [9] and CuBr2_2 [10], consisting solely of the centrosymmetric bonds. In other materials, the situation is not so straightforward [11–14]: as the symmetry is low, the bonds are not necessarily centrosymmetric, thus allowing for alternative explanations [17] [15, 16]. Then, if KNB model fails to explain the properties of spiral magnets, it typically causes some confusion with identifying the problem and choosing a suitable alternative [18]. Although density-functional theory (DFT) provides a powerful tool for calculating the polarization [19, 20], the formal mapping of DFT results on a specifically selected model [21] does not shed light on microscopic mechanisms underlying this model.

In this Letter, we formulate a transparent microscopic theory of electric polarization induced by noncollinear magnetic order. We explicitly show how and why the noncollinear arrangement of spins gives rise to electric polarization even in centrosymmetric bonds. We relate the ME coupling to fundamental symmetry properties of the position operator and transfer integrals in the basis of Kramers states and argue that the paradigm of the spin-current induced polarization appears to be much richer and goes beyond the phenomenological low P⃗∼ϵij×[e⃗i×e⃗j]\vec{P} \sim\epsilon_{ij} \times[\vec{e}_i \times\vec{e}_j] [17, 18, 21]. We evaluate all relevant parameters on the basis of DFT and show how they are manifested in the properties of real spiral magnets.

Basic theory

The simplest toy model, which captures the physics, is the 1-orbital Hubbard model with spin-orbit interaction (SOI),

H^=∑ij∑σσ′tijσσ′c^iσ†c^jσ′+U∑in^i↑n^i↓,(1)\hat{H} = \sum_{ij} \sum_{\sigma\sigma'} t_{ij}^{\sigma\sigma'} \hat{c}_{i\sigma}^{\dagger}\hat{c}_{j\sigma'} + U\sum_i \hat{n}_{i\uparrow}\hat{n}_{i\downarrow}, \tag*{(1)}

where c^iσ†\hat{c}_{i\sigma}^{\dagger} (c^iσ\hat{c}_{i\sigma}) creates (annihilates) a hole with pseudospin σ=+\sigma=+ or −- at site ii, n^iσ=c^iσ†c^iσ\hat{n}_{i\sigma}=\hat{c}_{i\sigma}^{\dagger}\hat{c}_{i\sigma}, t^ij=[tijσσ′]\hat{t}_{ij}=[t_{ij}^{\sigma\sigma'}] are the transfer integrals, and UU is the on-site Coulomb repulsion. The Wannier functions, ∣iσ⟩=c^iσ†∣0⟩|i\sigma\rangle=\hat{c}_{i\sigma}^{\dagger}|0\rangle, can be chosen as the Kramers pairs, transforming under the time reversal as T^∣i∓⟩=±∣i±⟩\hat{T}|i\mp\rangle=\pm|i\pm\rangle. All model parameters are derived from DFT [22], in the subspace of Wannier states, which are primarily responsible for the magnetism [20, 23]. The Coulomb UU is evaluated within constrained random-phase approximation [24].

Since the polarization in metals is screened by free electrons, the multiferroicity is the property of insulating state. Then, the problem can be solved in the spirit of superexchange (SE) theory, by treating t^ij\hat{t}_{ij} as a perturbation [25]. Let ∣αio⟩=cos⁡θi2∣i+⟩−sin⁡θi2e−iϕi∣i−⟩|\alpha_i^o\rangle=\cos\frac{\theta_i}{2}|i+\rangle-\sin\frac{\theta_i}{2}e^{-i\phi_i}|i-\rangle be the occupied orbital in the limit t^ij=0\hat{t}_{ij}=0, where θi\theta_i and ϕi\phi_i specify the direction e⃗i=(cos⁡ϕisin⁡θi,sin⁡ϕisin⁡θi,cos⁡θi)\vec{e}_i=(\cos\phi_i\sin\theta_i,\sin\phi_i\sin\theta_i,\cos\theta_i) of spin, and ∣αiu⟩=sin⁡θi2eiϕi∣i+⟩+cos⁡θi2∣i−⟩|\alpha_i^u\rangle=\sin\frac{\theta_i}{2}e^{i\phi_i}|i+\rangle+\cos\frac{\theta_i}{2}|i-\rangle is the unoccupied orbital. To the first order in t^ij\hat{t}_{ij}, ∣αio⟩|\alpha_i^o\rangle will transform to the Wannier function ∣wi⟩=∣αio⟩+∑j∣αi→jo⟩|w_i\rangle=|\alpha_i^o\rangle+\sum_j|\alpha_{i\to j}^o\rangle, acquiring the tails

∣αi→jo⟩=−1U∣αju⟩⟨αju∣t^ji∣αio⟩(2)|\alpha_{i\to j}^o\rangle=-\frac{1}{U}|\alpha_j^u\rangle\langle\alpha_j^u|\hat{t}_{ji}|\alpha_i^o\rangle \tag*{(2)}

on surrounding sites jj. Considering expectation values of the kinetic energy ⟨wi∣t^ij∣wi⟩\langle w_i|\hat{t}_{ij}|w_i\rangle, one can readily formulate the spin model

E=∑⟨ij⟩(−Jije⃗i⋅e⃗j+D⃗ij⋅[e⃗i×e⃗j]+e⃗i⋅Γ^ije⃗j),(3)\mathcal{E}=\sum_{\langle ij\rangle}\left(-J_{ij}\vec{e}_i\cdot\vec{e}_j+\vec{D}_{ij}\cdot[\vec{e}_i\times\vec{e}_j]+\vec{e}_i\cdot\hat{\Gamma}_{ij}\vec{e}_j\right), \tag*{(3)}

describing the energy change in terms of isotropic (JijJ_{ij}), Dzyaloshinskii-Moriya (DM, D⃗ij\vec{D}_{ij}), and symmetric anisotropic (Γ^ij\hat{\Gamma}_{ij}) interactions. A similar model can be formulated for polarization [26]:

P⃗=∑⟨ij⟩(P⃗ije⃗i⋅e⃗j+P⃗ij ′⋅[e⃗i×e⃗j]+e⃗i⋅Π^ije⃗j),(4)\vec{P}=\sum_{\langle ij\rangle}\left(\vec{\mathcal{P}}_{ij}\vec{e}_i\cdot\vec{e}_j+\vec{\mathcal{P}}_{ij}^{\,\prime}\cdot[\vec{e}_i\times\vec{e}_j]+\vec{e}_i\cdot\hat{\Pi}_{ij}\vec{e}_j\right), \tag*{(4)}

in terms of the vector P⃗ij≡[Pijv]\vec{\mathcal{P}}_{ij}\equiv[\mathcal{P}_{ij}^{v}], rank-2 tensor P⃗ij ′≡[Pijv,c]\vec{\mathcal{P}}_{ij}^{\,\prime}\equiv[\mathcal{P}_{ij}^{v,c}], and rank-3 tensor Π^ij≡[Πijv,ab]\hat{\Pi}_{ij}\equiv[\Pi_{ij}^{v,ab}] [27]. The model parameters can be obtained from matrix elements of the position operator in the framework of general theory for polarization in periodic systems [19, 20]:

P⃗=−eV∑i⟨wi∣r⃗∣wi⟩(5)\vec{P}=-\frac{e}{V}\sum_i\langle w_i|\vec{r}|w_i\rangle \tag*{(5)}

(where VV is the volume and −e-e is the electron charge). Then, the only matrix elements that contribute to the magnetic dependence of P⃗\vec{P} are of the type ⟨αio∣r⃗∣αi→jo⟩\langle\alpha_i^o|\vec{r}|\alpha_{i\to j}^o\rangle. Other contributions, such as ⟨αi→jo∣r⃗∣αi→jo⟩\langle\alpha_{i\to j}^o|\vec{r}|\alpha_{i\to j}^o\rangle or single-ion anisotropy of P⃗\vec{P}, vanish in the 1-orbital model [17, 18].

The 2×22\times2 matrices t^ij\hat{t}_{ij} and r⃗ij\vec{r}_{ij} can be decomposed in terms of the unity 1^\hat{1} and the vector σ⃗=(σx,σy,σz)\vec{\sigma}=(\sigma_x,\sigma_y,\sigma_z) of Pauli matrices as t^ij=tij01^+it⃗ijσ⃗\hat{t}_{ij}=t_{ij}^{0}\hat{1}+i\vec{t}_{ij}\vec{\sigma} and r⃗ij=r⃗ij01^+ir⃗ijσ⃗\vec{r}_{ij}=\vec{r}_{ij}^{0}\hat{1}+i\vec{r}_{ij}\vec{\sigma} with the real coefficients (tij0,t⃗ij)(t_{ij}^{0},\vec{t}_{ij}) and (r⃗ij0,r⃗ij)(\vec{r}_{ij}^{0},\vec{r}_{ij}). Furthermore, the hermicity yields tji0=tij0t_{ji}^{0}=t_{ij}^{0}, t⃗ji=−t⃗ij\vec{t}_{ji}=-\vec{t}_{ij}, r⃗ji0=r⃗ij0\vec{r}_{ji}^{0}=\vec{r}_{ij}^{0}, and r⃗ji=−r⃗ij\vec{r}_{ji}=-\vec{r}_{ij} [28]. The corresponding spin model parameters are summarized in Table I [29].

A very special case is when two sites ii and jj are connected by spatial inversion. Since t^ij\hat{t}_{ij} is a scalar, but r⃗ij\vec{r}_{ij} is a (true) vector, inversion symmetry requires that t^ji=t^ij\hat{t}_{ji}=\hat{t}_{ij}, but r⃗ji=−r⃗ij\vec{r}_{ji}=-\vec{r}_{ij}. In combination with hermiticity, we have t⃗ij=0\vec{t}_{ij}=0 and r⃗ij0=0\vec{r}_{ij}^{0}=0. Then, the DM interaction D⃗ij\vec{D}_{ij} vanishes, which is the general property of centrosymmetric (c) bonds, and so does Γ^ij\hat{\Gamma}_{ij}, being not independent in the 1-orbital model [30]. Thus, the only interaction in the c-bond will be JijJ_{ij}. The behavior of electric polarization is different: P⃗ij ′=0\vec{\mathcal{P}}_{ij}^{\,\prime}=0 and Π^ij=0\hat{\Pi}_{ij}=0, while P⃗ij=−2eVtij0r⃗ijU≡C⃗ij\vec{\mathcal{P}}_{ij}=-\frac{2e}{V}\frac{t_{ij}^{0}\vec{r}_{ij}}{U}\equiv\vec{\mathcal{C}}_{ij} can persist even in the c-bond (in agreement with symmetry arguments [21]), and the corresponding polarization, C⃗ij⋅[e⃗i×e⃗j]\vec{\mathcal{C}}_{ij}\cdot[\vec{e}_i\times\vec{e}_j], is induced solely by the noncollinear arrangement of spins.

In the noncentrosymmetric (nc) bonds, tijt_{ij} and r⃗ij0\vec{r}_{ij}^{0} are finite, resulting in nonzero D⃗ij\vec{D}_{ij}, Γ^ij\hat{\Gamma}_{ij}, P⃗ij ′\vec{\mathcal{P}}_{ij}^{\,\prime}, and Π^ij\hat{\Pi}_{ij}. Moreover, P⃗ij\vec{\mathcal{P}}_{ij} acquires an additional asymmetric contribution A⃗ij=−2eVtij0r⃗ijU\vec{\mathcal{A}}_{ij}=-\frac{2e}{V}\frac{t_{ij}^{0}\vec{r}_{ij}}{U}. P⃗ij\vec{\mathcal{P}}_{ij} and A⃗ij\vec{\mathcal{A}}_{ij} can be viewed as a regular polarization of the nc-bond (−er⃗ij0V-\frac{e\vec{r}_{ij}^{0}}{V}), which is additionally modulated by the nc-spin texture. If this texture results from the competition of D⃗ij\vec{D}_{ij} and JijJ_{ij}, a similar competition takes place between A⃗ij\vec{\mathcal{A}}_{ij} and P⃗ij\vec{\mathcal{P}}_{ij}, as it depends on

exchangepolarization
Jij=−(tij0)2UJ_{ij}=-\frac{(t_{ij}^{0})^2}{U}P⃗ij=−2eVr⃗ij0tij0U\vec{\mathcal{P}}_{ij}=-\frac{2e}{V}\frac{\vec{r}_{ij}^{0}t_{ij}^{0}}{U}
D⃗ij=2tij0t⃗ijU\vec{D}_{ij}=\frac{2t_{ij}^{0}\vec{t}_{ij}}{U}P⃗ij ′=−2eVr⃗ij0t⃗ij+tij0r⃗ijU\vec{\mathcal{P}}_{ij}^{\,\prime}=-\frac{2e}{V}\frac{\vec{r}_{ij}^{0}\vec{t}_{ij}+t_{ij}^{0}\vec{r}_{ij}}{U}
Γ^ij=2t⃗ij⊗t⃗ijU\hat{\Gamma}_{ij}=\frac{2\vec{t}_{ij}\otimes\vec{t}_{ij}}{U}Π^ij=−eVr⃗ij⊗t⃗ij+t⃗ij⊗r⃗ijU\hat{\Pi}_{ij}=-\frac{e}{V}\frac{\vec{r}_{ij}\otimes\vec{t}_{ij}+\vec{t}_{ij}\otimes\vec{r}_{ij}}{U}

Table 1. Isotropic (JijJ_{ij} and P⃗ij\vec{\mathcal{P}}_{ij}), antisymmetric (D⃗ij\vec{D}_{ij} and P⃗ij ′\vec{\mathcal{P}}_{ij}^{\,\prime}), and anisotropic symmetric (Γ^ij=Γ^ij ′−12Tr⁡Γ^ij ′\hat{\Gamma}_{ij}=\hat{\Gamma}_{ij}^{\,\prime}-\frac{1}{2}\operatorname{Tr}\hat{\Gamma}_{ij}^{\,\prime} and Π^ij=Π^ij ′−12Tr⁡Π^ij ′\hat{\Pi}_{ij}=\hat{\Pi}_{ij}^{\,\prime}-\frac{1}{2}\operatorname{Tr}\hat{\Pi}_{ij}^{\,\prime}) parameters of exchange interactions and polarization. ⊗\otimes denotes the direct product.

the same ratio tijtij0\frac{t_{ij}}{t_{ij}^{0}}. Therefore, there will be a cancellation of contributions associated with A⃗ij\vec{A}_{ij} and P⃗ij\vec{P}_{ij} [18].

The 1-orbital model is subjected to hidden symmetries, that allow to fully eliminate tijt_{ij} by rotating the spins at the sites ii and jj [30]. In such local coordinate frame, the bond is solely described by the transfer integral t~ij0=(tij0)2+t⃗ij⋅t⃗ij\tilde{t}_{ij}^{0} = \sqrt{(t_{ij}^{0})^{2} + \vec{t}_{ij} \cdot\vec{t}_{ij}}, exchange interactions – by JijJ_{ij}, and the electric polarization – by the competition of P⃗ij\vec{P}_{ij} and C⃗ij\vec{C}_{ij} (with t~ij0\tilde{t}_{ij}^{0} instead of tij0t_{ij}^{0}). Using ⟨αjy∣⟨αjy∣=1^−∣αjy⟩⟨αjy∣\langle\alpha_{j}^{y}|\langle\alpha_{j}^{y}| = \hat{1} - |\alpha_{j}^{y}\rangle\langle\alpha_{j}^{y}| in Eq. (2), each term in the spin model can be further expressed via the expectation value of some quantity in the ground state. Since [e⃗i×e⃗j][\vec{e}_{i} \times\vec{e}_{j}] is related to the spin current j⃗^ij=it~ij02(c~^iσσ⃗^σσ′c~^jσ′−c~^jσσ⃗^σσ′c~^iσ′)\hat{\vec{j}}_{ij} = \frac{i\tilde{t}_{ij}^{0}}{2}(\hat{\tilde{c}}_{i\sigma}\hat{\vec{\sigma}}_{\sigma\sigma'}\hat{\tilde{c}}_{j\sigma'} - \hat{\tilde{c}}_{j\sigma}\hat{\vec{\sigma}}_{\sigma\sigma'}\hat{\tilde{c}}_{i\sigma'}) [31], one can find that P⃗∼C⃗ij⋅⟨j⃗ij⟩\vec{P} \sim\vec{C}_{ij} \cdot\langle\vec{j}_{ij} \rangle [29], in analogy with the DM interactions [32].

The 3×33 \times3 tensor C⃗ij=[Cijv,c]\vec{C}_{ij} = [C_{ij}^{v,c}] can be generally decomposed into symmetric and antisymmetric parts, C⃗ij=C⃗ijS+C⃗ijA\vec{C}_{ij} = \vec{C}_{ij}^{S} + \vec{C}_{ij}^{A}. The latter is expressed as εvcaπija\varepsilon_{vca}\pi_{ij}^{a}, in terms of the vector π⃗ij=[πija]\vec{\pi}_{ij} = [\pi_{ij}^{a}] and Levi-Civita symbol εvca\varepsilon_{vca}. The corresponding P⃗=π⃗ij×[e⃗i×e⃗j]\vec{P} = \vec{\pi}_{ij} \times[\vec{e}_{i} \times\vec{e}_{j}] reminds of the KNB expression [3, 4]. Nevertheless, π⃗ij\vec{\pi}_{ij} is not necessarily parallel to ε⃗ij\vec{\varepsilon}_{ij}, and C⃗\vec{C} can include C⃗ijS\vec{C}_{ij}^{S}. Thus, the proposed spin-current theory is not limited to the conventional law P⃗∼ε⃗ij×[e⃗i×e⃗j]\vec{P} \sim\vec{\varepsilon}_{ij} \times[\vec{e}_{i} \times\vec{e}_{j}] and includes other interesting options. Below, we consider examples of how it works for different symmetries.

Relativistic j=12j=\frac{1}{2} manifold of t2gt_{2g} states. In the cubic environment, the sixfold degenerate t2gt_{2g} states are split by SOI into fourfold degenerate Γ8\Gamma_{8} states and the Kramers pair of Γ7\Gamma_{7} states ∣+⟩=13(∣xy↓⟩−∣yz↑⟩+i∣zx↑⟩)|+\rangle= \frac{1}{\sqrt{3}}(|xy\downarrow\rangle- |yz\uparrow\rangle+ i|zx\uparrow\rangle) and ∣−⟩=13(∣xy↑⟩+∣yz↓⟩+i∣zx↓⟩)|-\rangle= \frac{1}{\sqrt{3}}(|xy\uparrow\rangle+ |yz\downarrow\rangle+ i|zx\downarrow\rangle). The latter can be viewed as the effective j=12j=\frac{1}{2} pseudospin states, which play a key role in the physics of spin-orbit Mott insulators realized in 5d5d Ir oxides [33]. Since the KNB expression was derived assuming this symmetry of states [3], we start our analysis with this example and consider a perfect bond obeying the C∞zC_{\infty}^{z} symmetry along zz. It is straightforward to show that the only nonzero elements of r⃗ij=[rijv,c]\vec{r}_{ij} = [r_{ij}^{v,c}] will be rijy,x=−rijx,y=13(⟨xyi∣y∣zxj⟩−⟨zxi∣y∣xyj⟩)r_{ij}^{y,x} = -r_{ij}^{x,y} = \frac{1}{3}(\langle xy_{i}|y|zx_{j}\rangle- \langle zx_{i}|y|xy_{j}\rangle) [29]. The antisymmetric tensor can be presented as εabzπijz\varepsilon_{abz}\pi_{ij}^{z}, where ab=xyab = xy or yxyx, and the vector π⃗ij=(0,0,πijz)\vec{\pi}_{ij} = (0,0,\pi_{ij}^{z}) is indeed parallel to the bond. Thus, we do recover the KNB expression P⃗∼ε⃗ij×[e⃗i×e⃗j]\vec{P} \sim\vec{\varepsilon}_{ij} \times[\vec{e}_{i} \times\vec{e}_{j}] [3]. Nevertheless, this form of P⃗\vec{P} is the consequence of the particular symmetry of undistorted t2gt_{2g} states. Other symmetries may yield different P⃗\vec{P}.

z2z^{2} states with SOI. The simplest example, illustrating this idea is the z2z^{2} states, which are stabilized by the crystal field and mixed with the yzyz and zxzx states by SOI. The Kramers states are [34]: ∣+⟩∝∣z2↓⟩+ξ(i∣yz↑⟩−∣zx↑⟩)|+\rangle\propto|z^{2}\downarrow\rangle+ \xi(i|yz\uparrow\rangle- |zx\uparrow\rangle) and ∣−⟩∝∣z2↑⟩+ξ(i∣yz↓⟩+∣zx↓⟩)|-\rangle\propto|z^{2}\uparrow\rangle+ \xi(i|yz\downarrow\rangle+ |zx\downarrow\rangle)

(ξ\xi being the ratio of SOI to the crystal field splitting), and the non-vanishing elements are: rijy,x=rijx,y∝⟨zi2∣xzj⟩−⟨zxi∣zj2⟩r_{ij}^{y,x} = r_{ij}^{x,y} \propto\langle z^{2}_{i}|xz_{j}\rangle- \langle zx_{i}|z^{2}_{j}\rangle [29]. Thus, the tensor [rijv,c][r_{ij}^{v,c}] is symmetric, meaning that for two noncollinear spins e⃗1,2=(sin⁡θcos⁡φ,sin⁡θsin⁡φ,±cos⁡θ)\vec{e}_{1,2} = (\sin\theta\cos\varphi,\sin\theta\sin\varphi,\pm\cos\theta), the polarization P⃗∼(cos⁡φ,−sin⁡φ,0)\vec{P} \sim(\cos\varphi,-\sin\varphi,0) is still perpendicular to the bond, but does not necessarily lie in the spin plane.

α\alpha-Li2_2IrO3_3. As the first realistic example, it is instructive to consider the spin-orbit Mott insulator α\alpha-Li2_2IrO3_3, which attracted a great deal of attention as a possible material for realizing the Kitaev spin liquid state [35]. In this monoclinic compound (space group C2/mC2/m), there are two types of nearest cc-bonds: the strong ones (01 in Figure 1a) along the monoclinic bb axis, which is transformed to itself by the twofold rotation about bb; and two rotationally non-invariant weak bonds (02 and 03), which are connected by the twofold rotation.

Tensor and magnetic structures in alpha-Li2IrO3

Figure 1. (a) Tensor C⃗ij\vec{C}_{ij} in the nearest cc-bonds of α\alpha-Li2_2IrO3_3 (in μC/m2\mu\mathrm{C}/\mathrm{m}^{2}), in the local coordinate frames denoted by xx, yy and zz. (b) Double-qq magnetic structure realized in α\alpha-Li2_2IrO3_3. Inversion center is denoted by *. (c) Magnetically induced polarization in the chains with +q+q and −q-q depending on the rotation of the spin-spiral plane about the monoclinic bb axis, as explained in (d) in the planes abab and bcbc. θ\theta is the angle formed by the spin-spiral plane and the monoclinic aa axis.

The minimal model was constructed for the magnetic j=12j=\frac{1}{2} bands [29]. In the local coordinate frame, where the zz axis is along the bond, the tensor C⃗01\vec{C}_{01} has only xx and yy components (see Figure 1a). Hence, the polarization is perpendicular to the bond and, considering only the antisymmetric part of C⃗01\vec{C}_{01}, we would indeed obtain P⃗∼ε⃗01×[e⃗0×e⃗1]\vec{P} \sim\vec{\varepsilon}_{01} \times[\vec{e}_{0} \times\vec{e}_{1}] [3, 4]. However, besides the antisymmetric contribution, C⃗01\vec{C}_{01} clearly shows a strong symmetric one, which is expected for this type of symmetry [36]. In the low-symmetry bonds 02 and 03, C⃗ij\vec{C}_{ij} is even more complex: all elements are finite and inequivalent, so that the polarization can be perpendicular as well as parallel to the bond.

Since α\alpha-Li2_2IrO3_3 is a noncollinear magnet with q⃗≈(13,0,0)\vec{q} \approx(\frac{1}{3},0,0) [37], it is interesting to ask whether it can become

multiferroic. The magnetic texture can be viewed as the double-qq spiral, propagating in alternating zigzag chains with +q+q and −q-q (Figure 1b). The spiral order induces an appreciable polarization in each of the chains (Figure 1c). However, the chains are connected by the inversion operation, the contributions with +q+q and −q-q cancel each other, and α\alpha-Li2_2IrO3_3 remains antiferroelectric.

Copper dihalides. CuCl2_2 and CuBr2_2 form the chainlike monoclinic structure (space group C2/mC2/m), where all Cu-Cu bonds are centrosymmetric. Below the Néel temperature (TN=24T_{\mathrm{N}}=24 and 7474 K for CuCl2_2 and CuBr2_2, respectively), they develop a cycloidal magnetic order with q≈(1,14,12)\mathbf{q}\approx\left(1,\frac{1}{4},\frac{1}{2}\right) (Figure 2a), which coincides with the onset of ferroelectricity [9, 10].

Crystal and magnetic structures of CuCl2 and CuBr2, with angle dependence of electric polarization

Figure 2. (a) Crystal and magnetic structure of CrCl2_2 in the abab plane. (b) Rotations of the spin spiral plane about the axis bb. (c) Angles θ\theta and θP=tan⁡−1PzPx\theta_{\mathrm{P}}=\tan^{-1}\frac{P^{z}}{P^{x}} specifying the spiral plane and electric polarization, respectively. (d) and (e) Angle dependence of electric polarization in CuCl2_2 and CuBr2_2.

The minimal model was constructed for the magnetic “Cu x2−y2x^{2}-y^{2}” bands [29]. The obtained parameters JijJ_{ij} nicely reproduce the experimental cycloidal order [38]. Polarization induced by the cycloidal order is related to C~ij\tilde{\mathbf{C}}_{ij}. Since twofold rotations about bb transform the chains to themselves, the symmetry properties of C~ij\tilde{\mathbf{C}}_{ij} in these chains are similar to C~01\tilde{\mathbf{C}}_{01} in α\alpha-Li2_2IrO3_3 (Figure 1a) and the polarization is expected to be perpendicular to bb. In the nearest bonds ±b\pm b, the antisymmetric part of C~±b\tilde{\mathbf{C}}_{\pm b} is given by π±b=(0,±0.5,0)\boldsymbol{\pi}_{\pm b}=(0,\pm0.5,0) and (0,±8,0)(0,\pm8,0) μC/m2\mu\mathrm{C}/\mathrm{m}^{2} for CuCl2_2 and CuBr2_2, respectively, which is too small to account for the total P~\tilde{\mathbf{P}}, and a substantially larger contribution stems from the symmetric part. The behavior of P~\tilde{\mathbf{P}} (after summation over all bonds) is summarized in Figure 2d and Figure 2e. For CuCl2_2, we note a good agreement with the experiment [9], including deviation of P~\tilde{\mathbf{P}} from the spin-spiral plane by the angle θ+θP\theta+\theta_{\mathrm{P}}. Similar behavior is expected for CuBr2_2 with somewhat larger TNT_{\mathrm{N}} and P~\tilde{\mathbf{P}} due to stronger SOI and hybridization mediated by Br 4p4p states.

Cupric oxide. CuO with the centrosymmetric monoclinic structure (space group C2/cC2/c) has attracted much attention as a simple binary material that becomes multiferroic at exceptionally high temperature [13]. Ferroelectricity is driven by a spiral magnetic order with q≈(12,0,−12)\mathbf{q}\approx\left(\frac{1}{2},0,-\frac{1}{2}\right) (see Figure 3a) emerging between 213 and 230 K [39] (and is expected even at room temperature under hydrostatic pressure [40]).

Spiral magnetic order in CuO and electric polarization

Figure 3. (a) Spiral magnetic order in CuO. (b)-(c) Nearest c- and (d) nc-bonds in the monoclinic planes abab and bcbc with the vectors of induced polarization. (e) Electric polarization (total and centrosymmetric part described by C~ij\tilde{\mathbf{C}}_{ij}) as the function of angle θ\theta formed by the spins in the acac plane with the axis aa. Here, xx and yy are chosen along aa and bb, respectively, and zz is perpendicular to the acac plane. (f) Electric polarization perpendicular (P⊥P^{\perp}) and parallel (P∥P^{\parallel}) to the bond, calculated for the nearest c- and nc-bonds as a function of θ\theta.

Construction of the minimal model is similar to Cu dihalides [29]. Although the crystal structure of CuO includes both c- and nc-bonds (Figure 3), the parameters associated with the latter can be largely eliminated [41], while C~ij\tilde{\mathbf{C}}_{ij} is responsible for 90% of P~\tilde{\mathbf{P}} (Figure 3e).

Since the symmetry is low, the form of C~ij\tilde{\mathbf{C}}_{ij} is complex [29], and the polarization vectors in individual bonds are specified by all three projections, as shown in Figure 3. Notably, there can be appreciable components along the bonds (Figure 3f). Nevertheless, due to the twofold rotations about yy (bb), only the yy component of P~\tilde{\mathbf{P}} will survive after summation over all inequivalent bonds. This is an important point where the symmetry comes into play: the experimental polarization is induced along the yy axis not because of the phenomenological rule q×[ei×ej]\mathbf{q}\times[\mathbf{e}_{i}\times\mathbf{e}_{j}] [3, 4]. Quite the contrary, it is a consequence of the particular C2/cC2/c symmetry of CuO. Total Pb≈55P^{b}\approx55 μC/m2\mu\mathrm{C}/\mathrm{m}^{2} is comparable with the experimental Pb∼150P^{b}\sim150 μC/m2\mu\mathrm{C}/\mathrm{m}^{2} [42] [13].

Conclusion. We have presented a toy theory revealing the fundamental origin of the magnetic inversion symmetry breaking in centrosymmetric systems. Due to intrinsic symmetries of the transfer integrals and position operator in the basis of Kramers states, the combination tij0r⃗ijt_{ij}^{0}\vec{r}_{ij} remains finite, yielding finite polarization for noncollinear spins. This polarization depends on the symmetry of Kramers states, providing new alternatives beyond phenomenological spin-current model. The abilities of proposed theory are illustrated on spiral magnets CuCl2_2, CuBr2_2, CuO, and α\alpha-Li2_2IrO3_3.

Acknowledgement

We are grateful to Anonymous Referee of Ref. [18] for drawing our attention to intersite matrix elements of r⃗\vec{r}. I.S. was supported by program AAAA-A18-118020190095-4 (Quantum).

References

  1. [17]100401(R) (2019).
  2. [18]R. Ono, S. Nikolaev, and I. Solovyev, Phys. Rev. B 102, 064422 (2020).
  3. [19]R. D. King-Smith and D. Vanderbilt, Phys. Rev. B 47, 1651(R) (1993); R. Resta, J. Phys.: Condens. Matter 22, 123201 (2010).
  4. [20]N. Marzari, A. A. Mostofi, J. R. Yates, I. Souza, and D. Vanderbilt, Rev. Mod. Phys. 84, 1419 (2012).
  5. [21]H. J. Xiang, E. J. Kan, Y. Zhang, M.-H. Whangbo, and X. G. Gong, Phys. Rev. Lett. 107, 157202 (2011).
  6. [22]I. V. Solovyev, J. Phys.: Condens. Matter 20, 293201 (2008); M. Imada and T. Miyake, J. Phys. Soc. Jpn. 79, 112001 (2010).
  7. [23]A. A. Mostofi, J. R. Yates, G. Pizzi, Y. S. Lee, I. Souza, D. Vanderbilt, and N. Marzari, Comput. Phys. Commun. 185, 2309 (2014).
  8. [24]F. Aryasetiawan, M. Imada, A. Georges, G. Kotliar, S. Biermann, and A. I. Lichtenstein, Phys. Rev. B 70, 195104 (2004). The Coulomb U was evaluated as 1.5 eV for α-Li2IrO3 and 4.3 eV for CrCl2, CrBr2, and CuO.
  9. [25]P. W. Anderson, Phys. Rev. 115, 2 (1959).
  10. [26]T. Moriya, J. Appl. Phys. 39 1042 (1968).
  11. [27]The vector symbol and index ν are reserved for the direction of polarization, while the bold symbols and indices after the comma stand for the coupling to the spin part.
  12. [28]For i = j one obtains tᵢⱼ = 0 and r⃗ᵢⱼ = 0, meaning that there is no single-ion anisotropy, neither for the energy nor for the polarization, and all magnetic state dependence of these quantities stems from the bonds [18].
  13. [29]See the Supplemental Material at http://link.aps.org/supplemental/... for derivation of the spin model for the electric polarization and details of calculations. The Supplemental Material includes citations listed in Ref. [43].
  14. [30]T. A. Kaplan, Z. Phys. B 49, 313 (1983); L. Shekhtman, O. Entin-Wohlman, and A. Aharony, Phys. Rev. Lett. 69, 836 (1992).
  15. [31]P. Bruno and V. K. Dugaev, Phys. Rev. B 72, 241302(R) (2005).
  16. [32]T. Kikuchi, T. Koretsune, R. Arita, and G. Tatara, Phys. Rev. Lett. 116, 247201 (2016).
  17. [33]B. J. Kim, H. Jin, S. J. Moon, J.-Y. Kim, B.-G. Park, C. S. Jeem, J. Yu, T. W. Noh, C. Kim, S.-J. Oh, J.-H. Park, V. Dura͏iraj, G. Cao, and E. Rotenberg, Phys. Rev. Lett. 101, 076402 (2008).
  18. [34]H. Takayama, K.-P. Bohnen, and P. Fulde, Phys. Rev. B 14, 2287 (1976).
  19. [35]J. Chaloupka, G. Jackeli, and G. Khaliullin, Phys. Rev. Lett. 105, 027204 (2010).
  20. [36]If the Tᵍ₂g states are distorted so that the |xy⟩, |yz⟩, and |zx⟩ orbitals contribute to |±⟩ with unequal weights, the tensor r⃗ᵢⱼ is allowed to have simultaneously antisymmetric and symmetric parts [29].
  21. [37]S. C. Williams, R. D. Johnson, F. Freund, S. Choi, A. Jesche, I. Kimchi, S. Manni, A. Bombardi, P. Manuel, P. Gegenwart, and R. Coldea, Phys. Rev. B 93, 195158 (2016).
  22. [38]There are four main interactions separated by monclinic translations b, 2b, 3b, and c: J_b = −0.3 (−0.4), J₂b = −3.0 (−3.9), J₃b = −0.1 (−0.2), and J_c = −0.7 (−1.4) meV for CuCl₂ (CuBr₂), which are responsible for the spin-spiral order in the chains (along b) and the AFM coupling between the chains (along c). Corresponding T_N, evaluated in the random phase approximation [29], is 32 (56) K for CuCl₂ (CuBr₂).
  23. [39]P. J. Brown, T. Chattopadhyay, J. B. Forsyth, V. Nunez, and F. Tasset, J. Phys.: Condens. Matter 3, 4281 (1991).
  24. [40]X. Rocquefelte1, K. Schwarz, P. Blaha, S. Kumar, and J. van den Brink, Nat. Commun. 4:2511 doi: 10.1038/ncomms3511 (2013).
  25. [41]By properly choosing local coordinate frames, one can eliminate x and z components of tᵢⱼ in the nearest nc-bonds between the ac planes (Fig. 3d). The remaining Dᵞᵢⱼ components do not contribute to the spiral magnetic order, and the latter emerges solely from the competition of isotropic Jᵢⱼ’s [29]. The same holds for A⃗ᵢⱼ = [Aᵢⱼᶜ], whose only non-zero component is c = y, and there is no contribution to P⃗. The isotropic P⃗ᵢⱼ in the nc-bonds can also be large [29], but does not add up to P⃗ because of the 90° alignment of spins.
  26. [42]The difference can be related to the additional distortion and lowering the symmetry from C2/c to Cc, observed in CuO below T_N: S. Asbrink and A. Waskowska, J. Phys.: Condens. Matter 3, 8173 (1991).

Paper details

vlnc:W_nr8r6umf
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

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