You are on page 1of 8

First-principles microscopic model of exchange-driven magnetoelectric response with

application to Cr2 O3
Sai Mu, A. L. Wysocki, and K. D. Belashchenko
Department of Physics and Astronomy and Nebraska Center for Materials and Nanoscience,
University of Nebraska-Lincoln, Lincoln, Nebraska 68588, USA
(Dated: April 24, 2014)
The exchange-driven contribution to the magnetoelectric susceptibility is formulated using a
microscopic model Hamiltonian coupling the spin degrees of freedom to lattice displacements and
arXiv:1404.5921v1 [cond-mat.mtrl-sci] 23 Apr 2014

electric eld, which may be constructed from rst-principles data. Electronic and ionic contributions
are sorted out, and the latter is resolved into a sum of contributions from dierent normal modes.
If intrasublattice spin correlations can be neglected, the longitudinal component k becomes pro-
portional to the product of magnetic susceptibility and sublattice magnetization in accordance with
Rados phenomenological model. As an illustration, the method is applied to analyze the temper-
ature dependence of the longitudinal magnetoelectric susceptibility of Cr2 O3 using rst-principles
calculations and the quantum pair cluster approximation for magnetic thermodynamics. A sub-
stantial electronic contribution is found, which is opposite to the ionic part. The sensitivity of the
results to the Hubbard U parameter and the sources of error are studied. It is also found that
non-Heisenberg interactions are too weak to account for the sign change of k in Cr2 O3 .

I. INTRODUCTION a maximum at finite temperature, where it is dominated

by Heisenberg exchange. This temperature-dependent
Antiferromagnets exhibiting a linear magnetoelectric effect in Cr2 O3 was studied by Mostovoy et al.17 , who
effect1 are of great interest for applications aiming to obtained the relevant coupling constant from the electric
achieve electric control of magnetism.26 In such mate- polarization of a ferrimagnetically ordered unit cell and
rials there is a term E H in the free energy density, the temperature dependence from Monte Carlo simula-
where is the magnetoelectric tensor. Due to this term tions for the classical Heisenberg model. Their approach
the electric field induces a magnetization and the mag- is tailored to Cr2 O3 and is not directly applicable to other
netic field induces a dielectric polarization, both in linear systems. Only the total response was evaluated.
order. Magnetoelectric effect was first predicted7 and ex-
perimentally observed811 in Cr2 O3 , which remains the In this paper we formulate a microscopic model of
most promising material for applications. exchange-driven magnetoelectric response which general-
Magnetoelectric effect can arise due to several mi- izes the approach of Ref. 17. We study the longitudinal
croscopic mechanisms, including electric field-induced magnetoelectric susceptibility of Cr2 O3 in more detail,
changes of the single-ion anisotropy, Heisenberg exchange sorting out the electronic and lattice-mediated contribu-
parameters, the g-tensor or Dzyaloshinskii-Moriya inter- tions and resolving the latter by normal displacement
action (see Ref. 12 for a review). Each of these contri- modes, and comparing the predictions of the quantum
butions can be further divided into electronic (clamped- pair cluster, quantum mean-field, and classical mean-field
ions) and lattice-mediated parts. Experimentally the approximations. We also test the possibility that non-
electronic contributions can, in principle, be measured Heisenberg spin interactions could be responsible for the
separately at frequencies that are large compared to those sign change of k in Cr2 O3 and conclude in the negative.
of the relevant optical phonon vibrations.
First-principles methods can illuminate the micro-
scopic mechanisms of the magnetoelectric effect.1322 At The paper is organized as follows. In Section II the
zero temperature it is controlled by spin-orbit coupling. microscopic model is formulated in terms of the micro-
niguez13 showed that at zero temperature the lattice- scopic coefficients coupling the spins to lattice displace-
mediated contribution can be obtained by evaluating the ments and directly to the electric field, and the general
electric and magnetic polarities and stiffnesses of the expressions for the electronic and lattice-mediated con-
polar displacement modes. An alternative approach19 tributions to magnetoelectric susceptibility are derived.
is to compute the electric polarization induced by the The computational procedure is described in Section III,
magnetic field. This method was used to calculate and the results are presented in Section IV. We find that
both lattice-mediated and electronic contributions to the the electronic contribution is a sizeble fraction of the
transverse magnetoelectric susceptibility of Cr2 O3 . lattice-mediated term and its sign is opposite. Different
The electronic contribution turned out to be as much statistical approximations lead to similar maximal val-
as one third of and have the same sign as the lattice- ues of the magnetoelectric susceptibility, but the latter
mediated one.19 The orbital contribution to the magne- is sensitive to the choice of the Hubbard U parameter
toelectric response has also been considered.21,22 due to its effect on the magnetic susceptibility. Section
Longitudinal magnetoelectric susceptibility k reaches V concludes the paper.

II. MICROSCOPIC MODEL extended by treating the exchange parameters as second-

rank tensors. In this case the vectors gi,jk and fjk turn
Here we restrict ourselves to the exchange-driven mag- into third-rank tensors.
netoelectric effect. This means that the model Hamilto- Integrating out the spin degrees of freedom and treat-
nian in zero magnetic field should be invariant under a ing the external magnetic field and the last (magneto-
coherent rotation of all spins. This restriction is appro- electric) term in (1) as small perturbations, we obtain an
priate for Cr2 O3 where the orbital moments are almost effective Hamiltonian for the lattice degrees of freedom:
completely quenched, but it may need to be relaxed for X 1X 1
application to other systems with strong spin-orbit cou- Heff = ui qi E + ui Aij uj H 0m H
pling. A spin orientation for a magnetic atom on lattice i
2 ij
site i will be denoted by a unit vector ei . Spin rotation !
symmetry implies that ei should only appear in scalar X X X
ui gi,jk + Ef jk l h(ej ek )el i0 H (4)
combinations. We start from the expansion of the effec- 2 i
jk l
tive Hamiltonian to second order in lattice displacements
ui from an equilibrium reference state: where = 1/kT , 0m is the magnetic susceptibility ten-
sor, h. . . i0 the statistical average taken over the unper-
1X X X
H = E0 + ui Aij uj ui qi E H i e i turbed state described by Hmag , and we have dropped
2 ij i i the first-order magnetostrictive term generated by gi,jk ,
! which does not affect theP magnetoelectric response.
1X X Let us denote Ljk = 21 l l h(ej ek )el i0 and
+ Hmag {ei } ui gi,jk + Efjk (ej ek )
2 i
jk X
(1) gi = gi,jk Ljk . (5)
where i is the magnetic moment at site i, E and H the
external electric and magnetic fields, qi the Born effective Referring to (4), we see that gi H represents the force
charge tensors, Aij the Born-von K arman force constant acting on site i arising in response to the external mag-
matrix, and Hmag the magnetic interaction Hamiltonian netic field H. Thus, gi is the effective magnetic monopole
in the absence of displacements and external fields. The charge of the atom at site i. Conversely, gi describes the
Hamiltonian (1) is generally similar to that of Ref. 23, but change in the magnetization arising due to the displace-
we explicitly sort out the coupling of spins to lattice dis- ment of site i. It depends on temperature and is non-
placements in order to separate the ionic and electronic zero P for both magnetic and non-magnetic atoms. The
contributions to the magnetoelectric susceptibility. sum i gi = 0 thanks to (3). The structure of the ten-
The magnetoelectric coupling is generated by the last sor gi is determined by the corresponding magnetic site
term in (1), where the sum over j, k runs over magnetic symmetry. For example, the symmetry of the Cr sites in
sites only. The parameters may be expressed as Cr2 O3 includes the C3 axis, and therefore for these sites
only the zz component of gi is non-zero.
Jjk (u, E) Jjk (u, E) When a material is both piezoelectric and piezomag-
gi,jk = , fjk = (2)
ui E netic, there is a contribution to the magnetoelectric sus-
ceptibility mediated by strain. Although this contribu-
where Jjk is the Heisenberg exchange parameter for a tion can be derived from the same Hamiltonian (4), we
pair of spins on sites j, k. The vectors fjk describe the assume in the following that it is not present, as in the
variation of the exchange parameters when external elec- case of Cr2 O3 . Thereby we can treat ui as internal to the
tric field is applied while the ions are clamped; this term unit cell and disregard the possible change of the volume
generates the electronic magnetoelectric response. The and unit cell shape. Equilibrium displacements are found
vector gi,jk gives the variation of the exchange parame- by minimizing (4), which leads to
ter for pair jk when the site i is shifted from the reference
configuration, or it can conversely be interpreted as the
Aij uj = qi E + gi H.
spin-dependent Kanzaki force.24 Translational invariance
X Since the displacements ui are identical in all unit cells,
gi,jk = 0. (3)
in Eq. (6) we can treat i and j as basis indices within
the unit cell and Aij as the Fourier transform of the
Coupling of atomic displacements to non-Heisenberg force constant matrix at q = 0. This matrix has three
(e. g. biquadratic) spin interaction terms is also possible, zero eigenvalues corresponding to homogeneous lattice
but these terms are usually small in wide-gap insulators displacements and 3(N 1) finite eigenvalues (where N
(this is explicitly checked below for Cr2 O3 ). For systems is the number of sites in the unit cell). Homogeneous dis-
where spin-orbit interaction has a significant contribu- placements should be excluded from consideration, and
tion to the magnetoelectric response, the model may be we therefore consider only the subspace of displacements
for which i ui = 0. Within this subspace the action of supercells with different configurational and magnetic or-
the symmetric matrix derings and lattice displacements. Fitting of the param-
P Anis represented by its eigendecom-
position Aij V C
n i n jV n
, where Cn are the non- eters fjk requires the calculation of the electric polariza-
zero eigenvalues and Vi n
the normalized eigenvectors. tion at zero lattice displacements. The magnetoelectric
Within this subspace (with homogeneous displacements response of a random alloy can then be obtained as an
projected out) the action of the matrix A can be inverted, explicit configurational average of Eqs. (11)-(12).
resulting in ui = j G ij Fj , where Fj are the forces in The expressions (10)-(13) are valid for any magnetic
structures, including multisublattice and noncollinear
the right-hand side of (6) and G n 1 n
ij = n VP
i Cn Vj . ones. They can be simplified in the case of a collinear
The forces automatically satisfy the sum rule i Fi = 0, antiferromagnet by noting that Ljk is parallel to the or-
therefore the matrix G acts within its domain of defini- dering axis and vanishes if sites j and k belong to equiv-
tion. alent antiferromagnetic sublattices with opposite magne-
Substituting the equilibrium displacements in (4), we tizations. Indeed, in this case the magnetic space group
find the free energy density: must contain a symmetry operation that interchanges the
sites j and k. This operation maps an arbitrary site
F 1 1 l onto an equivalent site with an opposite magnetiza-
= E 0m +
e E H( m )H E
H (7)
V 2 2 tion. The sum over l is therefore zero. In the case of a
two-sublattice antiferromagnet such as Cr2 O3 , it means
where Ljk 6= 0 only if j and k belong to the same sublattice.
1 Taking these properties into account, we can define a

e = q Gij qj (8) scalar ljk as Ljk = 21 hej + ek i0 ljk , and ljk has the sym-
1 metry of the non-magnetic space group.
m = gi G
ij gj (9) The magnetoelectric response generally depends in a

complicated way on all the parameters appearing in

= ion + el (10)
the Hamiltonian (1). A significant simplification can
1 be achieved if it is admissible to neglect intrasublat-
ion = qi G
ij g
j (11)
tice spin correlations. This approximation is justified as
X long as the corresponding exchange parameters Jn sat-

el = fjk Ljk (12)
isfy Jn 1. Intrasublattice interactions in antiferro-
magnets are usually not dominant, unless there is strong
Here is the volume of the unit cell, e is the stan- frustration. Therefore, the above inequality is satisfied
dard expression for the lattice-mediated dielectric sus- in most antiferromagnets at temperatures that are not
ceptibility in the Born-von K arman model, m is the too low compared to TN . In particular, this is the case
magnetostructural correction to the magnetic susceptibil- for Cr2 O3 where the shortest intrasublattice spin pair is
ity, and is the magnetoelectric tensor, which includes the fourth-neighbor one.
the ionic (first) and the electronic (second) terms. We The influence of weak interactions can be safely in-
have dropped the phonon part of the free energy, which cluded on the mean-field level, while short-range inter-
does not contribute to the linear susceptibilities in the actions can still be treated using more accurate meth-
harmonic approximation. ods such as the thermodynamical cluster approximations.
The neglect of correlations between sites j and k in the
Denoting the dielectric and magnetic
P monopole polar-
n expression for Ljk leads, through a decoupling of spin
ities of an eigenmode n as p n = q

i i V i and gn =
P n averages, to ljk , where is the homogeneous lon-
i Vi g
i , we can rewrite the ionic part of the magneto-
gitudinal magnetic susceptibility,
electric tensor as
1 X 1 gi m gi,jk zjk (14)

ion = C pn g n (13)
n n jk

where m is the sublattice magnetization and jk = +1

This expression agrees with that obtained by I niquez13 (jk = 1) when j and k both belong to the sublattice
at zero temperature (where the monopole charges are with mz > 0 (mz < 0), and jk = 0 if j and k belong to
controlled by spin-orbit coupling), but the present ap- different sublattices, and finally
proach also provides a microscopic definition (5) of the
temperature-dependent monopole charge.
The microscopic model based on the Hamiltonian (1) gn m z, (15)
can be generalized to the case of magnetoelectric alloys
by adding configuration-dependent Kanzaki forces (both which should be substituted in Eq. (13). In this last
spin-independent and spin-dependent) and force con- formula, un is the amplitude of the n-th normal mode
stants. The parameters can be fitted to first-principles defined
P as ui = un Vin , and = (J0A J0B )/2, where J0p =
calculations of the total energies or forces using several j Jij with site i belonging to sublattice p. Under the

same condition the electronic contribution can be written form, and the exchange parameters are obtained by fit-
as ting the calculated total energies of a number of magnetic
configurations.30 The factors n are found by calculat-
1 ing = (EA EB )/4, where EA and EB are the en-
el = m . (16)
Ez ergies required to reverse one spin on sublattice A or B
in the magnetic ground state. This is done using a 30-
Eqs. (15)-(16) have a structure equivalent to the phe- atom supercell with atomic displacements proportional
nomenological result of Ref. 11. Thus, we see that this to Vin . Then n is found by numerical differentiation
phenomenological form is appropriate for the exchange- with respect to un . This approach can be viewed as a
driven magnetoelectric effect under the assumptions converse of that used by Mostovoy et al. The quantity
specified in the derivation of Eq. (13). In particular, it /Ez needed for the evaluation of the electronic con-
does not require that the mean-field approximation is tribution (16) is calculated at zero atomic displacements
valid, but only that intrasublattice spin correlations are in the presence of external electric field32 with a subse-
small, which is a weaker assumption. quent numerical differentiation.


Cr2 O3 is an antiferromagnetic insulator with a Neel There are two displacement modes with non-zero po-
temperature TN of 307 K. It has a corundum structure larities pnz , both transforming according to the A2u irre-
with the rhomboedral unit cell containing four equiva- ducible representation of the 3m point group. We denote
lent Cr ions lying on the trigonal axis. The orientations these modes as LO1 and LO2 . The normalized eigen-
of the Cr magnetic moments alternate along the trig- vectors are listed in Appendix A, which also includes
onal axis. The magnetic point group 3 m allows the the phonon frequencies. The stiffnesses Cn and dielec-
magnetoelectric susceptibility tensor, which is diagonal tric polarities pnz of these modes are listed in Table I,
and has two independent components, k zz and along with the values of the derivative n which enters
xx = yy , where the z axis is aligned with the the expression (15) for the magnetic polarity. The stiffer
3-fold axis.25 The k component, which is dominated by LO1 mode has a much larger dielectric polarity, which re-
exchange mechanism, is the focus of our study. sults in larger displacements compared to the softer LO2
First-principles calculations were performed using the mode. The values of n for the two modes are similar.
projector augmented wave method26 implemented in the Thus, overall the LO1 mode gives a much larger contribu-
Vienna ab initio simulation package (VASP)27,28 . The tion to k , which is seen from its larger value of /E.
correlations within the Cr 3d shell were described using The electronic contribution to /E is also listed in
the rotationally invariant LDA+U method.29 We set the Table I. It is comparable in magnitude to the lattice-
Hund exchange parameter J = 0.58 eV as obtained from mediated contribution, but the sign is opposite. We can
the constrained occupation calculation,30 and studied the understand this sign difference by noting that the electric
results as a function of the Hubbard U parameter. field tends to perturb the electronic charge density in a
The energy cutoff for the plane wave expansion was way that partially compensates the displacement of the
set to 520 eV, and a -centered Monkhorst-Pack k- positively charged Cr ions. A significant magnetoelectric
point grid31 was used for the Brillouin zone integra- response was observed in optical measurements,3335 but
tion. Relaxations, phonon and Berry phase calculations the sign of this response was not determined.
were performed for the rhombohedral unit cell using 0.02
eV Gaussian smearing and a 8 8 8 k-point mesh.
TABLE I. Stinesses Cn , dielectric polarities pn , and ex-
The Hellmann-Feynman forces were converged to 0.005
change perturbations /E (see text) of the two polar dis-
eV/A. The exchange parameters were obtained using the placement modes contributing to k in Cr2 O3 , calculated at
hexagonal supercell and the tetrahedron method with the U = 4 eV.
4 4 2 k-point mesh. Mode Cn (eV/ A2 ) pnz (e) n (meV) /E (103 e A)
The q = 0 component of the force constant matrix
LO1 29.1 8.42 65.2 19.2
is evaluated using the standard technique, and its non-
uniform eigenvalues and eigenvectors are found. The LO2 11.2 0.88 52.5 4.2
Born effective charges are also calculated and used to Electronic 9.0
evaluate the polarities pn of the eigenmodes of the force
constant matrix. Only the polar modes with nonzero pnz For the magnetic thermodynamics, which determines
need to be considered. the factor m in (15) and (16), we use the quantum
The magnetic monopole charges of the normal modes pair cluster approximation36 for S = 3/2 and compare
in Eq. (15) contain two factors: m, which is determined its predictions with quantum (S = 3/2) and classical
by Hmag , and n = /un. The unperturbed mag- mean-field approximations. Since the corundum lattice is
netic Hamiltonian Hmag is assumed to have a Heisenberg low-coordinated, we expect that the pair cluster approx-

imation can provide a notable improvement compared to a sign change of the magnetoelectric coefficient. How-
MFA due to the inclusion of short-range order effects, ever, K12,13 and other such terms do not contribute to
but at very low temperatures it breaks down by devel- calculated, as we did, using collinear spin configurations.
oping unphysical features.36 The application of the pair To test whether non-Heisenberg terms like K12,13 are
cluster approximation is similar to Ref. 30, with the ex- appreciable in Cr2 O3 , we calculated the total energies
ception that here we only treat the nearest and next- EA () and EB () of a supercell with one Cr spin contin-
nearest neighbors within the pair-cluster approximation, uously rotated by an angle from 0 to on either of the
while more distant pairs are included on the mean-field two sublattices with ionic displacements induced by elec-
level. This is consistent with the neglect of the intrasub- tric field. These calculations were performed using the
lattice correlations in (15) and (16) and is justified by the self-consistently determined constraining fields (as imple-
small magnitude of the exchange parameters beyond the mented in VASP) but are otherwise similar to the evalu-
second coordination sphere. ation of (at = 0 and = they are equivalent). The
The temperature dependence of k obtained using dif- difference () = EA () EB () is plotted in Fig. 2. It
ferent statistical approximations is shown in Fig. 1 with is seen that () fits very well to a simple cosine. This
the temperature given in reduced units. We see that al- indicates that the effect of non-Heisenberg interaction on
though there are considerable variations in the shape of is negligible, and that the origin of the sign change in
the curve, the maximum value of k is rather similar in k should be sought elsewhere.
all three cases.


2.5 QMF

PC 0.8

in g.u.)


E -E







0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0
N , rad

FIG. 1. Magnetoelectric susceptibility k calculated using dif- FIG. 2. Angular dependence of the parameter = EA EB .
ferent statistical approximations: pair-cluster (PC), quantum The solid line shows the cosine t to the data points (discs).
mean-eld (QMFA), and classical mean-eld (CMFA).
The maximum value of k (max ) obtained in the
The longitudinal coefficient k undergoes a sign change pair-cluster approximation and the temperature Tmax at
at a temperature of about 100 K,10 the origin of which which this maximum is reached are shown in Fig. 3 as a
remains unclear. Calculations of Ref. 13 gave a negligible function of the Hubbard U parameter used in the calcula-
value of the spin contribution to k at zero temperature. tion. We see that increases by a factor of 2 when U is in-
It was further found21 that the orbital contribution domi- creased from 3 to 4 eV. In order to understand the origin
nates over the spin magnetism and has the right sign, but of this strong dependence, we first examine the /E
its magnitude is still too small compared to experiment. factors for the lattice-mediated and electronic contribu-
As is clear from Eq. (15) and (16), the spin contribu- tions, which are shown in Fig. 4. We see that /E
tion does not change sign if the magnetic interaction is decreases as a function of U for both lattice-mediated
of a purely Heisenberg type. However, non-Heisenberg and electronic mechanisms. However, the reduction of
contributions could, in principle, make the effective pa- the electronic term is faster, so the total value increases,
rameter depend on the order parameter and thereby albeit rather slowly. Thus, the overall strong dependence
on temperature. A typical term in the Hamiltonian ca- of max on U is almost entirely due to the enhancement
pable of inducing such an effect is K12,13 (S1 S2 )(S1 S3 ), of the magnetic susceptibility.
where 2 is a nearest and 3 a next-nearest neighbor of While the peak temperature Tmax calculated in the
site 1. Although very likely small compared to J1 and pair cluster approximation agrees with experiment for
J2 exchange parameters, K12,13 could be comparable to U = 4 eV, which also provides optimal description of the
the relatively small J4 which largely determines . A electronic and structural properties,30 the peak magneto-
sign change of the effective would manifest itself as electric susceptibility max = 2.35104 g. u. is strongly

use the same parameters and approximations, the results

of Ref. 17 are reproduced.) The main reason for the dis-
agreement in max with experiment is in the overesti-
mation of the magnetic susceptibility. The experimental
in g.u.)

value near the Neel temperature is (TN ) 25 emu/g,42

whereas the pair-cluster approximation yields 42 emu/g.
If we use the experimental susceptibility instead of the

calculated one, max comes out at about 1.4 104 g.

1.5 u., which is much closer to the experimental values.

The reason for the overestimation of can be qual-
1.0 itatively understood using the Curie-Weiss expression

= C/(T + ), where C is the Curie constant and the

Curie-Weiss temperature. For S = 3/2, using the experi-
mental value 25 emu/g for (TN ), we obtain 450 K.
Similar values of were found from the high-temperature
susceptibility measurements43 (550 50 K) and from

the exchange parameters obtained by fitting the inelas-

tic neutron scattering data for magnon dispersions44

(527 76 K). We can write TN aJsPand a|J0 |,

where a is a common coefficient, Js = j ei ej Jij , J0 =

j Jij , and is a suppression factor showing how much
TN is suppressed by fluctuations compared to its mean-
field value. In the pair-cluster approximation 0.8.
2.0 2.5 3.0 3.5 4.0 4.5
Thus, based on the experimental data we can conclude
that |J0 | is slightly greater than Js . This implies that the
U (eV) intrasublattice exchange interaction is small compared to
the intersublattice one. The results of our calculations
FIG. 3. max and Tmax calculated within the pair-cluster contradict this picture, giving |J0 |/Js 0.42 due to a
approximation as a function of the Hubbard U parameter. fairly large value of J4 0.2J1 at U = 4 eV.
The grey dashed lines show the corresponding experimental We have verified the fidelity of our fit of the exchange
parameters to the calculated total energies by increasing
the number of input configurations to 42. The results are
listed in Table II, which shows that the fit is quite sta-
30 ble. Particularly, the take-one-out cross-validation (CV)
score for this five-parameter fit is 0.7 meV. If J4 is not
included in the fit, a much larger CV score is obtained.

10 If an additional parameter J6 is included, its value comes


Total out an order of magnitude smaller than J4 . These results

/ E (10

suggest that J4 is too large due to the inaccuracies of the
electronic structure in the LDA+U method. It is known
that LDA systematically underestimates the binding en-
ergy of the oxygen 2p states in oxide insulators. In Cr2 O3
this leads to an overestimated hybridization with the Cr
3d states, which tends to increase with increasing U due
2.0 2.5 3.0 3.5 4.0 4.5 to the downward shift of the filled 3d orbitals. Therefore,
U (eV)
since J4 is expected to be dominated by superexchange,
its overestimation in LDA+U is quite natural.
FIG. 4. Parameters /E for lattice-mediated and elec-
tronic contributions as a function of the Hubbard U parame- TABLE II. Exchange parameters Jn (meV) obtained by t-
ter. ting to total energies calculated at U = 4 eV. A long dash
indicates that the corresponding Jn is not included in the t-
ting. The cross-validation score (CV, meV) for each t is also
overestimated compared to the experimetal data, which provided.
appear to converge to the value of about 1.0 104 g. J1 J2 J3 J4 J5 J6 CV
u.3,810,3741 It is also larger than the value obtained by
14.64 11.12 -2.11 -2.98 2.12 0.7
Mostovoy et al. who used a smaller value of U , treated
19.57 14.18 -1.94 2.24 6.9
the spins classically, and used a different method of ex-
tracting the magnetoelectric coupling constant. (If we 14.64 11.12 -2.12 -2.98 2.12 0.12 0.1


This work was supported by the NSF/SRC Supple-

ment to the Nebraska MRSEC (DMR-0820521), the Cen-
We formulated a microscopic model of the ter for NanoFerroic Devices (CNFD) and the Nanoelec-
temperature-dependent exchange-driven magneto- tronics Research Initiative (NRI). Computations were
electric susceptibility
which includes the coupling of performed utilizing the Holland Computing Center at the
scalar spin products to atomic displacements and to the University of Nebraska.
electric field. The parameters of the model can generally
be obtained using first-principles calculations, and it can
be extended to magnetoelectric alloys, which may help in Appendix A: Longitudinal displacement modes and
the search for new materials with better magnetoelectric phonon frequencies
properties. If the intrasublattice spin correlations can be
neglected (which is a good approximation for Cr2 O3 ), The normalized A2u eigenmodes of the force constant
then k can be expressed as a product of the magnetic matrix A are listed in Table III. The frequencies of the
susceptibility, sublattice magnetization, and a factor infrared-active phonon modes in Cr2 O3 are compared
that does not depend on temperature, as long as the with experimental data and with the results of Ref. 13 in
elastic properties of the lattice do not depend on it. This Table IV.
relation was suggested phenomenologically by Rado.11
If, further, the intrasublattice interactions beyond the
TABLE III. Normalized A2u eigenmodes of the A
fourth coordination sphere in Cr2 O3 are negligibly small,
our approach essentially becomes the converse of that LO1 LO2
of Mostovoy et al.17 Lattice-mediated and electronic Cr1 (0, 0, 0.361) (0, 0, 0.140)
contributions to k have been sorted out, and the former Cr2 (0, 0, 0.361) (0, 0, 0.140)
was decomposed in the sum of contributions from the Cr3 (0, 0, 0.361) (0, 0, 0.140)
two normal displacement modes.
Cr4 (0, 0, 0.361) (0, 0, 0.140)
O1 (0.074, 0.128, 0.241) (0.190, 0.330, 0.094)
The electronic contribution to k in Cr2 O3 is com- O2 (0.148, 0, 0.241) (0.381, 0, 0.094)
parable to the lattice-mediated contribution and has an O3 (0.074, 0.128, 0.241) (0.190, 0.330, 0.094)
opposite sign. Quantum pair cluster and mean-field ap- O4 (0.074, 0.128, 0.241) (0.190, 0.330, 0.094)
proximations for spin 3/2, as well as the classical mean- O5 (0.148, 0, 0.241) (0.381, 0, 0.094)
field approximations result in similar peak values of the O6 (0.074, 0.128, 0.241) (0.190, 0.330, 0.094)
magnetoelectric susceptibility. The latter, however, is a
quickly increasing function of the Hubbard U parameter,
mainly thanks to the increasing magnetic susceptibility
. If is taken from experiment, we find k in good TABLE IV. Table lists all calculated polar phonon frequencies
agreement with experiment. However, the calculation at (in unit of cm1 ), and compare with experimental data45 and
U = 4 eV, which results in a good agreement with ex- I
niguezs DFT calculation.13
periment for many other properties, overestimates by A2u modes Eu modes
a factor of 1.7, which in turn is due to the relatively large This work 572 409 637 567 450 312
value of the J4 parameter. Finally, it was found that non-
Experiment (Ref. 45) 533 402 609 538 440 305
Heisenberg exchange in Cr2 O3 is negligibly small and can
not account for the sign change of k observed at low Other theory (Ref. 13) 597 408 653 578 455 316

Present address: Ames Laboratory, Ames, Iowa 50011, Xi He et al., Nature Mater. 9, 579 (2010).
USA. I. E. Dzyaloshinskii, Sov. Phys. JETP 10, 628 (1959).
1 8
L. D. Landau and E. M. Lifshitz, Electrodynamics of con- D. N. Astrov, Sov. Phys. JETP 11, 708, (1960).
tinuous media (Pergamon, Oxford, 1960). V. J. Folen, G. T. Rado, and E. W. Stalder, Phys. Rev.
M. Fiebig, J. Phys. D 38, R123 (2005). Lett. 6, 607 (1961).
3 10
P. Borisov, A. Hochstrat, X. Chen, W. Kleemann, and Ch. D. N. Astrov, Sov. Phys. JETP 13, 729, (1961).
Binek, Phys. Rev. Lett. 94, 117203 (2005). G. T. Rado and V. J. Folen, Phys. Rev. Lett. 7, 310 (1961).
4 12
Ch. Binek and B. Doudin, J. Phys.: Condens. Matter 17, O. F. de Alcantara Bonm and G. A. Gehring, Adv. Phys.
L39 (2005). 29, 731 (1980).
M. Bibes and A. Barthelemy, Nature Mater. 7, 425 (2008). 13
J. I
niguez, Phys. Rev. Lett. 101, 117201 (2008).

14 31
K. T. Delaney, M. Mostovoy, and N. A. Spaldin, Phys. H. Monkhorst and J. D. Pack, Phys. Rev. B 13, 5188
Rev. Lett. 102 157203 (2009). (1976).
J. C. Wojdel and J. I
niguez, Phys. Rev. Lett. 103, 267205 32
I. Souza, J. I
niguez, and D. Vanderbilt, Phys. Rev. Lett.
(2009). 89, 117602 (2002).
J. C. Wojdel and J. I
niguez, Phys. Rev. Lett. 105, 037208 R. V. Pisarev, B. B. Krichevtsov, and V. V. Pavlov, Phase
(2010). Trans. 37, 63 (1991).
M. Mostovoy, A. Scaramucci, N. A. Spaldin, K. T. Delaney, B. B. Krichevtsov, V. V. Pavlov, R. V. Pisarev, and V. N.
Phys. Rev. Lett. 105, 087202 (2010). Gridnev, J. Phys.: Condens. Matter 5, 8233 (1993).
S. Prosandeev, I. A. Kornev, and L. Bellaiche, Phys. Rev. B. B. Krichevtsov, V. V. Pavlov, R. V. Pisarev, and V. N.
B 83, 020102(R) (2011). Gridnev, Phys. Rev. Lett. 76, 4628 (1996).
E. Bousquet, N. A. Spaldin, and Kris T. Delaney, Phys. V. G. Vaks and N. E. Zein, Sov. Phys. JETP 40, 537
Rev. Lett. 106, 107202 (2011). (1975).
E. Bousquet, and N. A. Spaldin, Phys. Rev. Lett. 107, E. Kita, K. Siratori, and A. Tasaki, J. Appl. Phys. 50,
197603 (2011). 7748 (1979).
A. Malashevich, S. Coh, I. Souza, and D. Vanderbilt, Phys. E. Kita, Ferroelectrics 162, 397 (1994).
Rev. B 86, 094430 (2012). J. P. Rivera, Ferroelectrics 161, 165 (1994).
A. Scaramucci, E. Bousquet, M. Fechner, M. Mostovoy, H. Wiegelmann, A. G. M. Jansen, P. Wyder, J.-P. Rivera,
and N. A. Spaldin, Phys, Rev. Lett. 109, 197203 (2012). and H. Schmid, Ferroelectrics 162, 141 (1994).
H. Yatom and R. Englman, Phys. Rev. 188, 793 (1969). H. Wiegelmann, Ph.D. thesis, University of Konstanz
H. Kanzaki, J. Phys. Chem. Sol. 2, 24 (1957). (1994).
A. S. Borovik-Romanov and H. Grimmer, International S. Foner, Phys. Rev. 130, 183 (1963).
Tables for Crystallography, D, 105 (2003) M. Foex, and M. Gra, C. R. Acad. Sci. 209, 160 (1939).
P. E. Bl
ochl, Phys. Rev. B 50, 17953 (1994). E. J. Samuelsen, M. T. Hutchings, and G. Shirane, Physica
G. Kresse and J. Hafner, Phys. Rev. B 48, 13115 (1993). Amsterdam 48, 13 (1970).
G. Kresse and J. Furthm uller, Phys. Rev. B 54, 11169 G. Lucovsky, R. J. Sladek, and J. W. Allen, Phys. Rev. B
(1996). 16, 4716 (1977).
A. I. Liechtenstein, V. I. Anisimov, and J. Zaanen, Phys.
Rev. B 52, R5467 (1995).
Siqi Shi, A. L. Wysocki, and K. D. Belashchenko, Phys.
Rev. B 79,104404 (2009).