Arxiv:1912.08014V3 [Cond-Mat.Mtrl-Sci] 12 Feb 2020

Electric dipole moment as descriptor for interfacial Dzyaloshinskii-Moriya interaction

Hongying Jia,∗ Bernd Zimmermann,† Gregor Michalicek, Gustav Bihlmayer, and Stefan Blügel Peter Grünberg Institut and Institute for Advanced Simulation, Forschungszentrum Jülichand JARA, 52425 Jülich, Germany

Chiral magnets are of fundamental interest and have important technological ramifications. The origin of chiral magnets lies in the Dzyaloshinskii-Moriya interaction (DMI), an interaction whose experimental and theoretical determination is laborious. We derive an expression that identifies the electric dipole moment as descriptor for the systematic design of chiral magnetic multilayers. Using density functional theory calculations, we determine the DMI of (111)-oriented metallic ferromagnetic Z/Co/Pt multilayers of ultrathin films. The non-magnetic layer Z determines the DMI at the Co-Pt interface. The results validate the electric and magnetic dipole moments as excel- lent descriptors. We found a linear relation between the electric dipole moment of Pt, the Allen electronegativity of Z, and the contribution of Pt to the total DMI.

I. INTRODUCTION faster to measure or calculate, but yet allow a reliable estimation of the DMI. Their application would allow for The Dzyaloshinskii-Moriya interaction (DMI) [1, 2] is a high-throughput screening of materials and narrow the the origin of chiral magnetism, a modern and active search-space considerably. Some quantities have already field of magnetism today. It is responsible for novel been proposed as descriptors for the DMI, mostly focus- static and dynamical magnetic properties. In particular ing on the magnetic layer: Belabbes et al. [37] report it drives the formation of chiral domain walls [3–5] and on a correlation between the DMI and the spin mag- skyrmions [6–13], which hold promise for applications in netic moments of 3d atoms in 3d ultrathin films on 5d future information storage, data processing and neuro- substrates. Kim et al. [33] conclude on a correlation morphic devices. It generally promotes the stabilization between the DMI and many other magnetic properties of chiral non-collinear spin-textures. from temperature-dependent studies on the same mate- A particularly important material class for applica- rial, e.g., emphasizing on the anisotropy of the orbital tions is metallic films and multilayers of magnetic and magnetic moment and the magnetic dipole moment of heavy transition metals [14–26], as they are compatible the ferromagnetic metal in experiment. Modifying the with current manufacturing processes in spintronic de- interface dipole or Rashba fields through charge transfer vices. A particular example is the experimentally vividly by oxidization of the magnetic layer was proposed as a pursued (111) oriented Co/Pt-based materials [27–35]. way to alter the DMI [38, 39]. Shifting the focus to the The huge combinatorics of different multilayers arising heavy-metal substrate, Ryu and coworkers [5, 40] argue from the chemical composition, layer thicknesses, stack- that the DMI is closely tied to the induced magnetic mo- ing sequences or growth conditions, to name a few, en- ment at the 5d atom, whereas other studies report that ables a detailed tuning of magnetic parameters, such as such a direct relation could not be confirmed [28, 41–43]. the DMI, which allows a flexible design of multilayers Experimental observation of the correlation between the with very specific properties. DMI and work function of non-magnetic layers in metal- Qualitative insights into the formation of the DMI have lic magnetic trilayers is reported by Park et al. [44]. been already gained by Moriya [2], or recently on the In this paper, we explore possible descriptors, in par- level of a tight-binding model [36], but a simple model ticular the local electric and intra-atomic magnetic dipole to predict the interfacial DMI quantitatively is currently moments. By an analytic derivation in perturbation the- unknown. Instead, the community relies either on exper- ory, we identify a link, to leading order, between the DMI iments or ab initio calculations to determine the DMI of and the electric dipole moment in the sense that both a specific system. In both approaches, the procedure is quantities emerge from the same electronic states. We rather involved and time-consuming, e.g., Brillouin light confirm our finding by first-principles calculations based scattering measurements need good statistics, or com- on density functional theory (DFT) on (111) oriented puting time intensive ab initio calculations of typically magnetic multilayers (MMLs) composed of Z/Co/Pt, arXiv:1912.08014v3 [cond-mat.mtrl-sci] 12 Feb 2020 rather large length-scale non-collinear magnetic struc- where the layer Z is one of the 4d transition metals (Y– tures need to be performed. In light of these restric- Pd), the noble metals (Cu, Ag and Au), or one of the tions, a systematic investigation of the large combinato- post-transition metals Zn and Cd. This is a suitable rial space of multilayers is currently unthinkable. test-set, because, as we show below, the DMI is modi- The quest is open for simple descriptors which are fied drastically as function of Z to positive and negative values. We find the largest correlation between the DMI and the electric dipole moment of either Co or Pt, and the sign is predicted correctly in 12 out of 13 different MMLs. ∗ Corresponding author: [email protected] To a lesser degree a correlation between the DMI and the † Corresponding author: [email protected] (induced) magnetic dipole moment on Pt is found. We 2 do not find a correlation between induced spin- or orbital moments at Pt. TABLE I. Possible combinations of states | ϕ1 i and | ϕ2 i that yield a finite contribution to the DMI; see Eq. (3) and text for details. We denote the states that also contribute to a finite el electric dipole moment, pz , by an asterisk. II. THEORETICAL RELATIONS BETWEEN DMI AND DIPOLE MOMENTS | ϕ1 i | ϕ2 i

α | pz i + β dx2−y2 | px i or | dxz i magnetization field m(r) is given as α | px i + β | dxz i (*) | pz i or dx2−y2 or | dz2 i Z 2 EDM[m] = d r Ds [m(∇ · m) − (m · ∇)m]z , (1) and is usually termed the interfacial DMI with an inter- 1 notation. ν = (n, σ) is a multi-index containing the band face DMI constant Ds. The latter relates to the change of spin-orbit energy upon twisting of the magnetization. index n and spin index σ. It can be determined from a microscopic model, e.g., a For the nominator the following hold: (i) The spin- DFT model, by the q-linear part of the energy change torque-moment contains the Pauli-matrix σx which se- if a spiraling magnetic texture of wave-vector q||xˆ and lects spin-flip contributions, i.e. the two states must be of different spin-character. An example that satis- rotation axis eˆrot =y ˆ is imposed [ˆz is the out-of-plane direction; see Fig. 1(b)] [45], fies this condition is | ψν i = | ϕ1 i ⊗ | ↑ i and | ψν0 i = | ϕ2 i ⊗ | ↓ i. Assuming in addition that the exchange 1 ∂EDFT(q x,ˆ yˆ) field is constant in space, the nominator in Eq. (3) sim- D = DM , (2) ↑↓ s plifies to Bxc hϕ1| Hso |ϕ2i hϕ2| x |ϕ1i. (ii) The spin-flip Ω ∂q q=0 ↑↓ part of spin-orbit coupling, Hso , selects transitions between states where the angular momentum index does with Ω the interface area per unit cell normal to the not change, |` − `0| = 0. (iii) The x operator selects MML. We treat non-collinearity and spin-orbit coupling transitions where the angular momentum index changes in perturbation theory and arrive at (Appendix A) by 1, |` − `0| = 1. The latter two conditions are mutu- ally excluding, and an overall non-vanishing DMI is only occ. all 0 0 0 0 possible if at least one of the states is a mixed state of 1 XX ψkν Hso ψkν0 ψkν0 Tyx ψkν Ds = 0 0 +c.c., two orbital characters, say | ϕ1 i = α | s i + β | pz i and Ω − 0 kν ν0 kν kν | ϕ2 i = | px i, so that the above nominator turns into ∗ ↑↓ (3) αβ Bxc hpz| Hso |pxi hpx| x |si= 6 0. 0 0 where kν and ψkν are the (unperturbed) band energy In the next step, we analyze all possible combinations and wavefunction, respectively, of state ν at crystal mo- of | ϕ1 i and | ϕ2 i assuming a basis of s, p, and d orbitals, mentum k of the ferromagnetic state, Hso is the spin- which contribute to the DMI, exploiting one basic sym- 0 orbit Hamiltonian, T = −σ × Bxc is the torque opera- metry operation of (111) oriented multilayers with C3v 2 tor, and Bxc is the exchange field . The summations are symmetry, namely a mirror plane that we choose per- performed over occupied states (occ.) as well as occupied pendicular to x. As a consequence, the wavefunctions of and unoccupied (all) states. the system are either even (+) or odd (−) under this sym- The nominator is a product of spin-orbit and spin- metry operation, ψ(−x, y, z) = ±ψ(x, y, z). Accordingly torque-moment matrix elements, respectively. They are we classify the atomic orbitals to be either of even (s, non-vanishing only if certain relations between the states py, pz, dz2 , dyz, dx2−y2 ) or odd (px, dxz, dxy) symmetry, 0 0 ψkν and ψkν0 are fulfilled (selection rules), which and hence only superpositions among even or odd states, we discuss next. In what follows, we separate the spa- respectively, are allowed candidates for | ϕ1 i. In Table I tial and spin-parts, | ψν i = | ϕn i ⊗ | σ i, and drop the all possible transitions that yield a finite contribution to crystal momentum k and superscript “0” to simplify the the DMI are summarized. It is interesting to note that if we additionally had a mirror plane perpendicular to z present in the system, all superpositions for | ϕ1 i in Table I would be prohibited by 1 The subscript s, derived from surface, denotes the interfacial symmetry and the DMI would vanish. This corresponds DMI constant. Typical values for Ds are on the order of a few to Moriya’s symmetry rules [2]. pJ/m for metallic interfaces, and a conversion to a DMI acting Our analysis shows that hybridization is the key to ob- on a volume of magnetic material reads D = Ds/t, where t is the thickness of the magnetic volume. taining a finite DMI. Equation (3) represents an impor- 2 A similar equation has been derived by Freimuth et al. [46], tant result and provides physical insights into the mech- where SOC was not treated in perturbation theory but entered anism of formation of the DMI. For example, the appear- through the wave-functions. ance of the position operator calls for a relation to the 3 electric dipole moment

Z el X p = −e hψkν | r |ψkν i = ρ(r) r dr , (4) -axis kν ΩMT of which only the out-of-plane direction does not vanish due to the symmetry of the system. e is the (positive) elementary charge, ρ(r) is the electron charge density, the integration is performed over the volume of a sphere, typically the muffin-tin (MT) sphere ΩMT, around the considered atom and the real-space vector r is measured with respect to the center of this sphere. Similarly to the arguments presented above, the states ψν that contribute must be a superposition of two states with ∆` = 1. Eval- uation of these matrix elements reveals that four of the five possible superpositions ϕ1 from Table I constitute el the finite pz . If we were able to change the superposi- FIG. 1. Properties of magnetic multilayers Z/Co/Pt for var- tions ϕ1 continuously by some external means, both, the ious chemical elements Z. (a) The total DMI, and broken el DMI and pz would scale similarly, e.g., to leading order down into spin-orbit contributions of the different layers. (b) ∗ el with αβ . Sketch of the MMLs. (c-e) Electric dipole moment pz and Expressing the spin-torque-operator in (3) beyond the (f-h) magnetic dipole moment Tz against Z and the DMI. Tz constant exchange-correlation field approximation by in- of Pt has been multiplied by a factor 10 for better visibility. cluding in addition the spherically symmetric contribu- Lines in (a), (c), and (f) are guides to the eye. Solid lines in (d,e,g,h) indicate least-squares fits, and the Pearson corre- tion around the atoms, Bxc(r), we obtain qualitatively lation coefficient is displayed in the panels. Data where the the same result as presented above (Appendix B). el sign between the DMI and the pz , Tz is wrong are indicated by open symbols. See Supplemental Material for the data presented. III. COMPUTATIONAL DETAILS AND METHOD IV. RESULTS The experimental bulk lattice constant of Pt (392 pm) corresponding to an (111) in-plane lattice constant of a = A. Correlation to electric dipoles 277 pm was chosen and structural optimizations of all interlayer distances were performed in scalar-relativistic In order to test our derivation of the correlation approximation using a mixed LDA/GGA spin-density- between pz and Ds on a realistic test-set, we per- functional [47] (LDA and GGA stand for local density form density-functional theory (DFT) calculations us- approximation and generalized gradient approximation, ing the full-potential linearized augmented plane wave respectively). Subsequent calculations of the electronic (FLAPW) method, as implemented in the FLEUR and magnetic properties employed the LDA [48]. We code [52]. The unit cell of our magnetic multilayer con- converged the charge density and spin density in a scalar- sists of three monolayers, namely Co sandwiched between relativistic approximation, sampling the full Brillouin Pt and Z [see Fig. 1(b)], where Z is a 4d transition metal zone (BZ) by (24 × 24 × 10) k-points if not stated other- (Y–Pd), a noble metal (Cu, Ag, and Au), or one of the wise. For the extraction of the DMI, we first performed post-transition metals Zn or Cd. We assume a fcc stack- self-consistent calculations of homogeneous spin-spirals ing of the layers and a ferromagnetic order of all Co mo- with a wave-vector q along the Γ–M high-symmetry line ments for a better comparability of our results, although of the BZ and |q| ≤ 0.1×2π/a. The DMI was determined also a synthetic antiferromagnetic coupling between ad- by including SOC in first-order perturbation theory on jacent Co layers might be energetically favorable for some top of a scalar-relativistic spin-spiral calculation [49] us- of the here studied multilayers. See Sec. III for compu- ing (48 × 48 × 20) k-points (see Refs. [26, 43] for details). tational details. The calculation of the magnetic dipole moment was per- From the analysis of the DMI as computed ab initio, we formed on a (35×35×18) k-point set which included the deduce a strong dependence of the magnitude and even Γ point, and SOC was either neglected or included in the the sign of the total DMI on the chemical element of the self-consistent calculations with collinear magnetization third atomic layer, Z [see Fig. 1(a)]. In particular, within alongz ˆ. In all calculations, the 4s and 4p states of Y, as the 4d series, the modification of the DMI becomes evi- well as the 4p states of Zr, Nb, and Mo were represented dent: we obtain small negative values when Z is an early by extending the conventional LAPW basis set with local transition metal (Ds = −1.11 pJ/m for Z = Zr), followed orbitals [50, 51]. by a rather continuous change towards positive and large 4 values when the d shell gets filled (Ds = 3.93 pJ/m for for the spin-torque moment of the form δTyx ∼ Qzxσx Cd). We break up the total DMI, Ds, of the system into (Appendix B) and can be identified as a contribution to contributions from the three atomic layers, denoted as the magnetic dipole moment [55], Z Co Pt Ds , Ds , and Ds for Z, Co, and Pt, respectively, by ¯h X switching spin-orbit coupling on only in a single layer at a T = hψ | Q · σ |ψ i , (7) time. Within first-order perturbation theory in spin-orbit 2 kν kν kν coupling (SOC), which we apply here, this decomposition is exact. The DMI originating from SOC of the Pt layer which should thus contribute to the DMI. Here, Qij = constitutes the dominant contribution [cf. Fig. 1(a)] due δij − 3ˆrirˆj, i, j ∈ {x, y, z}, are the components of the (di- to its large atomic number. Remarkably, this contribu- T mensionless) quadrupole tensor and σ = (σx, σy, σz) is tion changes drastically as the chemical type of the third the vector of Pauli matrices. The magnetic dipole T re- layer is varied and, in effect, determines the overall trend flects the asphericity of the magnetization density in the of the DMI of the entire stack. This is surprising because muffin-tin sphere around an atom, and has two contribu- it is commonly believed that the DMI at the Co/Pt in- tions: one is induced by the crystal field and the other by terface can merely be changed by external means, e.g., SOC [56]. Our correction term Qzxσx is related to the by controlling the interface quality [53, 54]. In contrast, latter. the Co layer and third layer Z only contribute little to We computed Tz for all multilayers with and without the DMI. Au/Co/Pt represents an exception, because Au SOC and find that the SOC induced changes in Tz are gives a rather large contribution (−1.90 pJ/m), but of on the order of 0.001h ¯, which is 1–2 orders of magnitude opposite sign as compared to Pt (3.34 pJ/m), hence re- smaller than the crystal field part, even for Pt. More- ducing the total DMI by more than a factor 2. over, we do not find a strong correlation to the DMI, Overall, our results cannot be explained by a linear ad- which renders the correction described above unimpor- ditive superposition of a constant DMI from the Pt/Co tant. Interestingly, we find a sizable correlation between interface and a varying DMI from the Co/Z interface, the crystal-field induced Tz and the DMI [|R| = 0.77 and as most of the changes are originating from the former. 0.73 for Pt and Co, respectively; see Figs. 1(f-h)], with It is not surprising that the separation into two individ- the caveat that Tz of Co is not able to predict the sign ual interfaces breaks down for individual layers that are of the DMI with high fidelity: There is an offset in the very thin, i.e. one atomic layer in the present case. In- data, which is related to the fact that a free-standing Co terestingly, the DMI from Pt can be even enhanced (e.g., monolayer exhibits a finite Tz [57] due to the strong as- by approximately 40% in Pt/Co/Cd) as compared to the phericity in the crystal field (i.e. x and z directions are DMI of a single Co/Pt interface (Ds ≈ 2.8 pJ/m [46]). inequivalent), but the DMI vanishes due to the presence Next we investigate how this drastic change of the DMI of structure inversion symmetry (+z and −z directions between Co and Pt correlates with our predicted descrip- are equivalent). Instead, we find a better correlation with tor, the electric dipole moment. In Fig. 1(c), we present Tz of Pt, which is an induced magnetic dipole moment the ab initio computed local electric dipole moments in and the structure-inversion asymmetry is implicitly im- the spheres of Co and Pt, see (4). They are nearly of the printed in its existence in this case. However, this sizable same magnitude and opposite sign and exhibit charac- correlation between the crystal-field part of Tz and the teristic sign-changes around Y and Rh, similarly to the DMI cannot be explained by Eq. (3), and it might be nec- DMI. Indeed, as Figs. 1(d,e) show, there exists a linear essary to develop a non-local theory or, since the position el relationship between the DMI and pz of Co and Pt, re- operator r is not a proper operator in the Hilbert space spectively. The overall correlation between the DMI and of periodic solids, turn to the corresponding Berry-phase el pz is very large, as expressed by the Pearson’s coefficient expressions [46], which is beyond the scope of this paper. |R| = 0.89 and 0.88, respectively [see Figs. 1(d,e)]. Also el the sign of the DMI correlates with the sign of pz in all cases except one (Y/Co/Pt). By means of a least-squares C. Relation to magnetic moments fit we find J pJ To shed light on the controversial debate on the re- DPt ≈ −0.53 pel(Pt) + 0.96 (5) s e m2 z m lationship between the induced spin moment of Pt and J pJ the DMI, we investigate the magnetic spin and orbital ≈ 0.67 pel(Co) + 1.6 , (6) moments of Pt. All moments are positive, meaning they e m2 z m are parallel to the Co moments, and proportional to each where the electric dipole moment is given in units of e m. k ⊥ other with m` /ms = 0.14 and m` /ms = 0.22, where the magnetization lies in-plane or along the out-of-plane direction, respectively (see Fig. 2). The correlation coeffi- B. Relation to magnetic dipoles cient (R = 0.74) between the DMI and ms is rather high, but we cannot deduce a causal relationship between these Returning to our analysis of (3) and considering a non- two quantities. Our reasoning is underpinned by the fact spherical contribution to Bxc(r) yields a correction term that the sign of the DMI cannot be correctly reproduced 5

in Ref. [44] trilayer systems are treated where only the m m ∥ (×4.5) m ⟂ (×7.0) s ℓ ℓ Co layer is aﬀected by the spacer layer. 0.4 (a) (b) R = 0.74 ) B (a) (b) μ 2.1 R= −0.75 R= 0.73 0.3 Ag Au Ag Cu Au 1.8 Cu 0.2 Zn Ru Ru

Allen Tc Zn Pd Tc Mo Mo Rh  1.5 0.1 Cd Rh Nb Pd Cd Nb Pt momentPt( 1.2 Zr Zr 0.0 -0.05 0 0.05 -2 0 2 4 Y 0.0 2.5 Zr Cu Zn Tc Ru Rh Pd Ag Cd Au el Nb Mo 푝푧 of Pt (e Å) DMI from Pt (pJ/m) atomic layer Z DMI from Pt (pJ/m) FIG. 3. Correlation between Allen electronegativity in Paul- ing units and (a) the electric dipole moment and (b) the DMI FIG. 2. (a) Induced Pt spin-moments (ms), orbital moments strength of Pt of magnetic multilayers Z/Co/Pt for various for in-plane (mk) and out-of-plane (m⊥) magnetization di- ` ` chemical elements Z. Not included is element Y. See Supple- rections in Z/Co/Pt MMLs. (b) Induced Pt spin moments mental Material for the data presented. against the DMI with least-squares fit and Pearson’s correlation coefficient R. A wrong sign between the DMI and ms is indicated by open symbols. See Supplemental Material for As a simplified guide to materials systems design we the data presented. investigated the relation of established tabulated data ef- fecting the electric dipole and thus the DMI of Pt. For example, we also investigated the relationship between by the magnetic moments in five out of 13 cases. Our the DMI and the electrical dipole moment with respect results also highlight that neither the orbital moments, to the workfunction changes between Co and Z and be- nor an anisotropy of the orbital moments is correlated tween Pt and Z, but unlike Park et al. [44] we could not to the DMI. In addition, we also do not find a sizable find any correlation, at least not for our periodic multi- correlation between DMI and the electronic charge of Co layers of ultrathin films. Instead we found useful relations or Pt. (Pearson correlation coefficient R > 0.7), shown in Fig. 3, between the DMI as well as the electric dipole moment of Pt and the tabulated Allen electronegativity [59], χAllen, for 4d transition-elements Z of the groups 4 to 10 of the V. DISCUSSION AND CONCLUSION periodic table, the group of noble metals (group 11) and to a certain extent also to the elements of the group 12 We calculated the DMI of periodic Z/Co/Pt(111) (Z = Zn, and Cd). We found large deviations for Y, ele- multilayers of monatomic layer thickness and com- ment of group 3. From Fig. 3(b) we read-off the relation pare our results with the experimental investigation of Z/Co/Pt(111) magnetic trilayers with film thicknesses pJ pJ DPt ≈ 7.62 χ (Z) − 10.89 (8) between 11 (Pt) and 5 (Co) atomic layers reported by s m Allen m Park et al. [44]. We compare those systems which have a ferromagnetic ground state in both our and their work: with parameters obtained by means of a least-squares the total DMI is 1.42 (2.56) pJ/m, 0.44 (1.68) pJ/m, 0.26 fit. The electronegativity is entered in Pauling units. (1.21) pJ/m for Cu/Co/Pt, Pd/Co/Pt, and Au/Co/Pt, Since according to Fig. 1(a) the total DMI follows the in Ref. [44] (in this work). First, we note that the signs DMI contribution of Pt with the exception of Au, we of DMI in our theoretical results are the same as in the accordingly find a predictive relation between the Allen experiment. Second, the relative values in the two pa- electronegativity and the total DMI for the remaining pers are also consistent (DCu > DPd > DAu). In our systems. Similar results we obtained for the Mulliken work, however, the overall magnitude is up to a fac- electronegativity [60], while no suitable correlation was tor 5 higher than in the experiment. Since our values found for the Pauling electronegativity [61]. do not appear unreasonably high when compared to the Our work points out a few important guidelines to re- DMI of a single Co/Pt interface with ultrathin Co layers alize the maximal impact on the DMI: (Ds = 1.7 pJ/m) [58], we conjecture that the major dif- (i) The layers adjacent to the heavy element (in our case: ference arises from the different layer thicknesses and the Pt) should have a large difference in the Allen electroneg- el related differences in the lattice strain. Additional factors ativity to maximize the dipole pz at this element. responsible for the difference include our focus on atom- (ii) The heavy (Pt) layer should be thin with sharp inter- ically sharp trilayers, while in experimentally produced faces in particular to the magnetic (Co) layer, but not too multilayers the vertical texture will certainly differ, e.g., thin. There should be a compromise between the accu- due to slight intermixing effects at the interfaces [54] or mulation of DMI strength by increasing the number of Pt a different stacking sequence. Another reason is that we layers and the finite screening length beyond which the have treated periodic multilayers where the non-magnetic electrical dipole moment in Pt cannot be changed signif- spacer layer hybridizes with both Co and Pt layers, but icantly by the second non-magnetic layer Z. The second 6 non-magnetic element can consist of several atomic lay- for convenience: ers. DFT 1 ∂EDM (q x,ˆ yˆ) For the external manipulation of the dipole with an Ds = . (A1) electric field our findings suggest that the field should Ω ∂q q=0 act mainly on the heavy atomic species. Because of the screening of the field by the conductive electrons, how- The latter is the SOC contribution to the total energy ever, the external manipulation of the electrical dipole by relative to the collinear (ferromagnetic) state for a spin- an electrical field is usually limited to layer thicknesses of spiral state of a general wavevector q and rotation axis a few atomic layers. On the other hand, there are experi- eˆrot, which is given in first-order perturbation theory as occ. ments carried out with trilayers where one layer is a sim- X ple oxide that acts as electrode, e.g., Au/Fe/MgO [62]. EDM(q, eˆrot) = hψkν (q)| Hso |ψkν (q)i , (A2) A change of DMI could be detected upon application of k,ν an electric field. Our findings suggest that an optimal where ψ is the unperturbed wavefunction of state ν control could be achieved by reverting the position of Au kν at crystal momentum k, Hso is the spin-orbit Hamil- and Fe, i.e. Fe/Au/MgO, to maximize the field acting on tonian, and the summation is performed over occupied the Au rather than the Fe. states. Since we are interested in the energy difference In conclusion, we derived an analytic expression for the due to the infinitesimal deviation of the magnetization Dzyaloshinskii-Moriya interaction (DMI) based on per- from the collinear state by introducing a long-wavelength turbation theory from the ferromagnetic state and pos- spin-spiral, we express the wavefunctions ψkν (q) of the tulated a relation to the electric dipole moment. Sub- non-collinear state in terms of the ferromagnetic state sequent ab initio calculations on magnetic multilayers ψ0 in first-order perturbation theory, of the type Z/Co/Pt indeed showed that the interfa- kν cial DMI, which takes values between −1 and 4 pJ/m, strongly correlates to the electric dipole moment pel. 0 z ψkν (q, r) = ψkν (r) + (A3) Since the electric dipole can be calculated rather quickly, 0 0 ψ 0 0 O (q) ψ we propose the evaluation of this quantity for a screening X k ν xc kν 0 0 0 ψk0ν0 (r) − 0 0 of chiral multilayer systems. In contrast, the intra-atomic k0ν0(6=kν) kν k ν magnetic dipole moment Tz and induced spin-moments correlate less and are, in the latter case, not able to pre- where dict the sign-change of DMI within the 4d series of Z. 0 Oxc(q) = σ · Bxc(q, r) − Bxc (A4) denotes the change of the magnetic exchange-correlation field relative to the ferromagnetic state magnetized along VI. ACKNOWLEDGMENTS direction eˆ0. Oxc is parallel to the local magnetization, and we assume that it rotates continuously in real-space We thank Tatsuya Shishidou and Tamio Oguchi for and is of constant magnitude, fruitful discussions and computational validation of the magnetic dipole vector. We thank Frank Freimuth, Jan- cos(q · r) Philipp Hanke, and Yuriy Mokrousov for in-depth dis- B = B Rzˆ→eˆrot sin(q · r) , (A5) xc xc xˆ→eˆ   cussions on the DMI interaction. We gratefully acknowl- 0 0 edge financial support from the DARPA TEE program through grant MIPR (No. HR0011831554) from DOI, where R is a rotation matrix that turns the local z-axis from the European Union H2020-INFRAEDI-2018-1 pro- to eˆrot and the local x-axis to the direction of the ferro- gram (Grant No. 824143, project MaX - Materials at magnetic state, eˆ0. For our geometry as specified in (2), the Exascale), from Deutsche Forschungsgemeinschaft we have two choices for eˆ0 (along the z or x axis, in the (DFG) through SPP 2137 “Skyrmionics” (Project BL following called gauge I and gauge II, respectively) and 444/16), from the Collaborative Research Centers SFB obtain 1238 (Project C01) as well as computing resources at O(I)(q)=B [ sin(q x) σ + (cos(q x) − 1) σ ] ,(A6) the supercomputers JURECA at Jülich Supercomputing xc xc x z (II) Centre and JARA-HPC from RWTH Aachen University. Oxc (q)=Bxc[− sin(q x) σz + (cos(q x) − 1) σx] .(A7) The wave-vector derivative in (2) leads to a derivative of the expectation value in (A2) Appendix A: DMI in perturbation theory ∂ ∂ψ hψ | H |ψ i = kν |H | ψ + c.c. (A8) Our aim is to derive (3) from the main text starting ∂q kν so kν ∂q so kν with the well-established relation between the interface DMI constant, Ds, and the change of the total energy and the wave function, respectively, which is evaluated DFT EDM [45], (2) from the main text, which we repeat here using (A3) 7

now the ferromagnetic state is aligned along x and corresponding ψ0 are eigenstates of σ , so that σ actually 0 0 kν x z ∂ψkν (q, r) X ψk0ν0 ∂Oxc/∂q ψkν 0 represents the spin-ﬂip contributions with respect to the = 0 0 ψk0ν0 (r) ∂q − 0 0 eigenstates of σ in gauge II. Overall, we see that in both k0ν0(6=kν) kν k ν x (A9) gauges, the second bra-ket selects spin-ﬂip contributions with with respect to the unperturbed eigenstates, and we may restrict the following analysis to gauge I.

∂O(I) ∂O(II) xc = B σ x and xc = −B σ x , ∂q xc x ∂q xc z q=0 q=0 Appendix B: Approximations to the spin-torque (A10) moment and relation between DMI and Tz employing (A6, A7). For an arbitrary direction of eˆ0, (A10) can be written in terms of the torque operator In advancing from (3) in the Sec. II [and from (A12) 0 T = −σ × Bxc, to (A14) in Appendix A], we made the approximation of a uniform exchange-correlation field, Bxc, across the unit ∂Oxc cell. In the following we lift this model assumption. The = Ty x, (A11) ∂q 0 0 q=0 second bra-ket of (3) and (A12), ψkν0 Tyx ψkν , turns into which has been termed the DMI operator in Ref. [63]. 0 0 Inserting everything leads to ψkν0 Bxc(r) σx x ψkν −→ hϕ2| Bxc(r) x |ϕ1i , (B1) occ. all 0 0 0 0 with wavefunctions ϕ and ϕ defined in the main text. 1 X X ψkν Hso ψkν0 ψkν0 Bxc σx x ψkν 1 2 Ds = 0 0 To evaluate these integrals we assume a tessellation of Ω − 0 kν ν0 kν kν the solid in terms of muffin-tin spheres around atoms in + c.c., (A12) which we expand the wavefunctions and the exchange field in spherical harmonics, YL(ˆr), of unit vector ˆr and where we restricted the analysis to gauge I. The states angular moment L = (`, m), ν0 run over occupied and unoccupied states. The sum 0 0 0 X (1(2)) over k drops out due to ψ Hso ψ 0 0 ∝ δk,k0 [49]. kν k ν ϕ1(2)(r) = ϕL (r) YL(ˆr) (B2) The nominator is a product of spin-orbit and spin-torque- L moment matrix elements and thus it is interpreted that X xc spin-orbit matrix elements are weighted by the spin- Bxc(r) = BL (r) YL(ˆr) . (B3) torque strength. Let us recall that the states ψ0 are L eigenstates of the ferromagnet without SOC; hence they The spherical harmonics are orthonormal, hYL|YL0 i = are of pure spin-character and eigenstates of σz. The δLL0 . 0 Pauli-matrix σx in the second bra-ket selects states ν We first assume Bxc(r) to be spherically symmetric; that have a different spin-character as compared to state i.e. only the term with L = 0 in (B3) remains. Substi- ν; i.e. it selects spin-flip contributions. tuting Bxc(r) in (B1) and separating radial and angular To analyze (A12) further, we break up the state index integrals yields ν = (n, σ) into a band index n and spin-index σ ∈ {↑ , ↓} and express the wave function ψ0 in terms of spinor xc 1 X (2) xc (1) hϕ2| B0 (r) x |ϕ1i = √ hϕ |B0 r|ϕ i|R components: 4π L2 L1 L1,L2 ˚↑ 0 × hL2| xˆ |L1i (B4) 0 ψkn 0 ψkn↑ = , ψkn↓ = ˚↓ (A13) 0 ψkn with

Assuming that Bxc is approximately constant across the Z hα| f |γi := r2dr α∗(r) f(r) γ(r) (B5) unit cell of the system, (A12) is transformed to |R

D 0 0 ED 0 E occ. all ˚σ σσ ˚σ ˚σ ˚σ ψ H ψ 0 ψ 0 x ψ andx ˆ = x/r. The angular part is of the same form as in Bxc X X kn so kn kn kn Ds = 0 0 the main text and hence the same transitions (cf. Table 1 Ω − 0 knσ n0 knσ kn σ0 of the main text) constitute the ﬁnite DMI. For the radial + c.c., (with σ0 6= σ), (A14) part we consider the example of the main text taking the radial representation of the state | ϕ1 i, h r | ϕ1 i = which establishes a connection between the interfacial α(r) | s i+β(r) | pz i, and state | ϕ2 i, h r | ϕ2 i = γ(r) | px i, DMI constant and the transition dipole moment on the with proper normalization R r2dr (|α|2 + |β|2) = 1 and right-hand side. R 2 2 ∗ r dr |γ| = 1, which transforms the prefactor Bxc αβ We quickly discuss gauge II: In this case, the second from the main text into 0 0 bra-ket in Eq. (A12) reads ψkν0 σz x ψkν and now −→ hβ| V |γi hγ| Bxc r |αi (B6) spin-conserving terms seem to play a role. However, so |R 0 |R 8 where the spin-orbit operator has been rewritten as where r is the position vector relative to the atomic nu- Hso = Vso(r) L · S. cleus. These quantities are the starting point for the cal- Going beyond the spherical approximation, the next culation of the electric dipole moment pel and of the in- non-spherical term to consider is of order ` = 1. traatomic magnetic dipole moment T, which are deﬁned Due to the uniaxial symmetry of the MMLs the terms in Eqs. (4) and (5). A more practical formulation of (5) m ± 1 vanish and the remaining contribution reads for a numerical evaluation within the FLAPW method is B(1,0)(r)Y(1,0)(ˆr) ∝ B(1,0)(r)ˆz, which yields for (B1) based on the vector-spin density s(r) and reads

hϕ | Bxc (r)ˆzx |ϕ i = hγ| Bxc r |αi hL | zˆxˆ |L i 2 (1,0) 1 (1,0) |R 2 1 Z (B7) T = [s(r) − 3ˆr(ˆr · s(r))] dr , (C3) The operator acting on the angular part is of angular ΩMT momentum ` = 2, i.e.z ˆxˆ ∼ Y2,−1 − Y2,1. Hence, the determination of possible transitions is governed by the where ˆr = r/|r| is the normalized position vector [57]. symmetry of the Gaunt coefficients Gm1 ±1 m2 , and only By expressing the unit vector in terms of spherical har- `1 2 `2 monics as s−d, p−p and d−d-transitions remain (neglecting transitions to f states and beyond) with the additional selec-   r Y (ˆr) − Y (ˆr) 1 tion rule m1 = m2 ± 1. 2π 1,−1 1,1 X ˆr = iY1,−1(ˆr) + iY1,1(ˆr) =: g Y (ˆr) On the other hand, we analyze the magnetic dipole 3  √  m 1,m moment, 2Y1,0(ˆr) m=−1 (C4) ¯h X T = hψ | Q σ + Q σ + Q σ |ψ i (B8) and making use of the spherical harmonics expan- z 2 kν zx x zy y zz z kν kν sion (C1) the electric dipole moment can now be written X 1 as ∝ hψ | zˆxˆ σ +z ˆyˆ σ + (ˆz2 − ) σ |ψ i (B9) kν x y 3 z kν kν Z 1 el X X 3 X 2 p = ρ`,m(r) r gm0 Y1,m0 (ˆr)Y`,m(ˆr) d r ∝ hψkν | Yxzσx + Yyzσy + √ Yz2 σz |ψkν i,(B10) ΩMT `,m m0=−1 kν 3 1 Z RMT X m 3 where Yxz, etc., denote real spherical harmonics in Carte- = (−1) gm ρ1,−m(r) r dr . (C5) sian coordinates. Interestingly, the first term in (B9) is m=−1 0 exactly the same operator as the one that appears in (B1) if the non-spherical correction to the exchange field is The latter simplification was achieved by angular inte- xc considered (Bxc(r) → B(1,0)zˆ). However, if the magnetic gration over the spherical harmonics, yielding δ1,` δm,m0 , system assumes the ferromagnetic state, and the SOC is where ΩMT and RMT are volume and radius of the MT neglected, the state | ψν i is an eigenstate of σz, the Tz sphere centered at the respective atom, respectively. reduces to the third term in (B9) proportional to Qzz Similarly the intra-atomic magnetic dipole moment and the spin-flip terms ∝ σx, σy, or Qxz and Qyz, respec- can be written as tively, which are the terms that contribute to the DMI, disappear. This underlines that it is the spin-orbit contri- √ Z RMT 2 bution to the wave functions that activates the spin-flip T = 4π (s0,0(r) − 3A0,0(r)) r dr (C6) 0 contributions which relate Tz and the DMI.

P P1 0mm0 where A0,0(r) = `,m m0=−1 gm0 B`,m(r)G0`1 , Appendix C: Calculation of pel and T P P1 mm0m00 B`,m(r) = `0,m0 m00=−1 gm00 s`0,m0 (r)G``01 , and mm0m00 G``0`00 are the Gaunt coefficients. In the FLAPW method, the charge density ρ(r) and It should be noted that the l = 1 character of the unit vector-spin density s(r) within the muffin-tin (MT) vector, together with the rules for non-vanishing Gaunt spheres around each atom are naturally available in terms coefficients and cancellations in the equations, leads to of spherical harmonics expansions as very few (`, m) combinations in the expansion of the X charge and vector-spin densities that are relevant for the ρ(r) = ρL(r)YL(ˆr), (C1) calculation of pel and T. For pel only the ` = 1 compo- L nents of the charge density are relevant and for T only X s(r) = sL(r)YL(ˆr), (C2) the ` = 2 components of the vector-spin density. L

[1] I. Dzialoshinskii, J. Phys. Chem. Solids 4, 241 (1958). [2] T. Moriya, Phys. Rev. 120, 91 (1960). 9

[3] M. Heide, G. Bihlmayer, and S. Blügel,Phys. Rev. B 78, Phys. Rev. B 90, 115427 (2014). 140403 (2008). [27] A. Soumyanarayanan, M. Raju, A. L. G. Oyarce, A. K. [4] A. Thiaville, S. Rohart, E.´ Jué,V. Cros, and A. Fert, C. Tan, M.-Y. Im, A. P. Petrovic, P. Ho, K. H. Khoo, Europhys. Lett. 100, 57002 (2012). M. Tran, C. K. Gan, F. Ernult, and C. Panagopoulos, [5] K.-S. Ryu, L. Thomas, S.-H. Yang, and S. Parkin, Nat. Mater. 16, 898 (2017). Nat. Nanotechnol. 8, 527 (2013). [28] H. X. Yang, A. Thiaville, S. Rohart, A. Fert, and [6] S. S. P. Parkin, M. Hayashi, and L. Thomas, Science 320, M. Chshiev, Phys. Rev. Lett. 115, 267210 (2015);118, 190 (2008). 219901(E) (2017). [7] S. Heinze, K. von Bergmann, M. Menzel, J. Brede, A. Ku- [29] J. L. Grab, A. E. Rugar, and D. C. Ralph, Phys. Rev. B betzka, R. Wiesendanger, G. Bihlmayer, and S. Blügel, 97, 184424 (2018). Nat. Phys. 7, 713 (2011). [30] M. Perini, S. Meyer, B. Dupé,S. von Malottki, A. Kubet- [8] A. Fert, V. Cros, and J. Sampaio, Nat. Nanotechnol. 8, zka, K. von Bergmann, R. Wiesendanger, and S. Heinze, 152 (2013). Phys. Rev. B 97, 184425 (2018). [9] W. Jiang, P. Upadhyaya, W. Zhang, G. Yu, M. B. [31] P. M. Shepley, H. Tunnicliffe, K. Shahbazi, G. Burnell, Jungfleisch, F. Y. Fradin, J. E. Pearson, Y. Tserkovnyak, and T. A. Moore, Phys. Rev. B 97, 134417 (2018). K. L. Wang, O. Heinonen, S. G. E. te Velthuis, and A. [32] A. Cao, X. Zhang, B. Koopmans, S. Peng, Y. Zhang, Hoffmann, Science 349, 283 (2015). Z. Wang, S. Yan, H. Yang and W. Zhao, Nanoscale 10 [10] O. Boulle, J. Vogel, H. Yang, S. Pizzini, D. S. Chaves, A. 12062 (2018). Locatelli, T. O. Mente¸s,A. Sala, L. D. Buda-Prejbeanu, [33] S. Kim, K. Ueda, G. Go, P. Jang, K. Lee, A. Belabbes, O. Klein, M. Belmeguenai, Y. Roussigné,A. Stashkevich, A. Manchon, M. Suzuki, Y. Kotani, T. Nakamura, K. S. M. Chérif,L. Aballe, M. Foerster, M. Chshiev, S. Auf- Nakamura, T. Koyama, D. Chiba, K. T. Yamada, D. fret, J. M. Miron, and G. Gaudin, Nat. Nanotechnol. 11, Kim, T. Moriyama, K. Kim and T. Ono, Nat. Commun. 449-454 (2016). 9, 1648 (2018). [11] C. Moreau-Luchaire, C. Moutas, N. Reyren, J. Sampaio, [34] D. Maccariello, W. Legrand, N. Reyren , K. Garcia, K. C. A. F. Vaz, N. Van Horne, K. Bouzehouane, K. Garcia, Bouzehouane, S. Collin, V. Cros and A. Fert, Nat. Nan- C. Deranlot, P. Warnicke, P. Wohlhüter,J. M. George, otechnol. 13, 233-237 (2018). M. Weigand, J. Raabe, V. Cros, and A. Fert, Nat. Nan- [35] F. Ajejas, V. Kˇriˇzáková,D. de Souza Chaves, J. Vogel, otechnol. 11, 444-448 (2016). P. Perna, R. Guerrero, A. Gudin, J. Camarero, and S. [12] N. Romming, H. Pralow, A. Kubetzka, M. Hoffmann, Pizzini, Appl. Phys. Lett. 111, 202402 (2017). S. von Malottki, S. Meyer, B. Dupé,R. Wiesendanger, [36] V. Kashid, T. Schena, B. Zimmermann, Y. Mokrousov, K. von Bergmann, and S. Heinze, Phys. Rev. Lett. 120, S. Blügel,V. Shah, and H. G. Salunke, Phys. Rev. B 90, 207201 (2018). 054412 (2014). [13] M. Hervé, B. Dupé, R. Lopes, M. Böttcher, M. D. [37] A. Belabbes, G. Bihlmayer, F. Bechstedt, S. Blügel,and Martins, T. Balashov, L. Gerhard, J. Sinova, and W. A. Manchon, Phys. Rev. Lett. 117, 247202 (2016). Wulfhekel,Nat. Commun. 9, 1015 (2018). [38] A. Belabbes, G. Bihlmayer, S. Blügel, and A. Manchon, [14] A. Fert, Mater. Sci. Forum 59-60, 439 (1990). Scientific Reports 6, 24634 (2016). [15] A. A. Zvyagin, J. Phys: Condens. Matter 3, 3865 (1991). [39] D. d. S. Chaves, F. Ajejas, V. Kˇriˇzáková,J. Vogel, and [16] S. V. Grigoriev, Y. O. Chetverikov, D. Lott, and A. S. Pizzini, Phys. Rev. B 99, 144404 (2019). Schreyer, Phys. Rev. Lett. 100, 197203 (2008). [40] K.-S. Ryu, S.-H. Yang, L. Thomas, and S. S. P. Parkin, [17] L. Udvardi and L. Szunyogh, Phys. Rev. Lett. 102, Nat. Commun. 5, 3910 (2014). 207204 (2009). [41] R. M. Rowan-Robinson, A. A. Stashkevich, Y. Roussigné, [18] A. Hrabec, N. A. Porter, A.Wells, M. J. Benitez, G. Bur- M. Belmeguenai, S. M. Chérif,A. Thiaville, T. P. A. nell, S. McVitie, D. McGrouther, T. A. Moore, and C. Hase, A. T. Hindmarch, and D. Atkinson, Sci. Rep. 7, H. Marrows, Phys. Rev. B 90, 020402 (2014). 16835 (2017). [19] S. Meyer, B. Dupé, P. Ferriani, and S. Heinze, [42] L. M. Sandratskii, Phys. Rev. B 96, 024450 (2017). Phys. Rev. B 96, 094408 (2017). [43] H. Jia, B. Zimmermann, and S. Blügel,Phys. Rev. B 98, [20] M. Hoffmann, B. Zimmermann, G. P. Müller, D. 144427 (2018). Schürhoff, N. S. Kiselev, C. Melcher, and S. Blügel, [44] Y.-K. Park, D.-Y. Kim, J.-S. Kim, Y.-S. Nam, M.- Nat. Commun. 8, 308 (2017). H. Park, H.-C. Choi, B.-C. Min, and S.-B. Choe, [21] M. Bode, M. Heide, K. von Bergmann, P. Ferriani, NPG Asia Mater 10, 995-1001 (2018). S. Heinze, G. Bihlmayer, A. Kubetzka, O. Pietzsch, [45] B. Schweflinghaus, B. Zimmermann, M. Heide, G. S. Blügel,and R. Wiesendanger, Nature 447, 190-193 Bihlmayer, and S. Blügel, Phys. Rev. B 94, 024403 (2007). (2016). [22] B. Dupé, M. Hoffmann, C. Paillard, and S. Heinze, [46] F. Freimuth, S. Blügeland Y. Mokrousov, J. Phys.: Con- Nat. Commun. 5, 4030 (2014). dens. Matter 26, 104202 (2014). [23] B. Dupé,G. Bihlmayer, M. Böttcher, S. Blügel,and S. [47] M. D. Santis, Y. Gauthier, H. Tolentino, G. Bihlmayer, S. Heinze, Nat. Commun. 7, 11779 (2016). Blügel,and V. Langlais, Phys. Rev. B 75, 205432 (2007). [24] A. K. Nandy, N. S. Kiselev, and S. Blügel, [48] S. H. Vosko, L. Wilk, and M. Nusair, Can. J. Phys. 58, Phys. Rev. Lett. 116, 177202 (2016). 1200 (1980). [25] A. Yagil, A. Almoalem, A. Soumyanarayanan, A. K. C. [49] M. Heide, G. Bihlmayer, and S. Blügel,Physica (Ams- Tan, M. Raju, C. Panagopoulos, and O. M. Auslaender, terdam) 404B, 2678 (2009). Appl. Phys. Lett. 112, 192403 (2018). [50] D. Singh, Phys. Rev. B 43, 6388 (1991). [26] B. Zimmermann, M. Heide, G. Bihlmayer, and S. Blügel, [51] G. Michalicek, M. Betzinger, C. Friedrich, and S. Blügel, Comput. Phys. Commun 184, 26702679 (2013). 10

[52] www.flapw.de. (2004). [53] R. Lavrijsen, D. M. F. Hartmann, A. van den Brink, Y. [58] M. Belmeguenai, J.-P. Adam, Y. Roussigné,S. Eimer, T. Yin, B. Barcones, R. A. Duine, M. A. Verheijen, H. J. Devolder, J.-V. Kim, S. M. Cherif, A. Stashkevich, and M. Swagten, and B. Koopmans, Phys. Rev. B 91, 104414 A. Thiaville, Phys. Rev. B 91, 180405(R) (2015). (2015). [59] L. C. Allen, J. Am. Chem. Soc. 111, 9003 (1989). [54] B. Zimmermann, W. Legrand, D. Maccariello, N. Reyren, [60] R. S. Mulliken, J. Chem. Phys. 2, 782 (1934). V. Cros, S. Blügel,and A. Fert, Appl. Phys. Lett. 113, [61] L. Pauling, J. Am. Chem. Soc. 54, 3570 (1932). 232403 (2018). [62] K. Nawaoka, S. Miwa, Y. Shiota, N. Mizuochi, and Y. [55] J. Stöhr,J. Magn. Magn. Materials. 200, 470 (1999). Suzuki, Applied Physics Express 8, 063004 (2015). [56] Y. Suzuki and S. Miwa, Phys. Lett. A 383, 1203 (2019). [63] F. Freimuth, S. Blügel, and Y. Mokrousov, [57] T. Oguchi and T. Shishidou, Phys. Rev. B 70, 024412 arXiv:1806.04782.