PHYSICAL REVIEW B 87, 144403 (2013)

Origin of multiferroicity in MnWO4

I. V. Solovyev*
Computational Materials Science Unit, National Institute for Materials Science, 1-2-1 Sengen, Tsukuba, Ibaraki 305-0047, Japan
(Received 29 October 2012; revised manuscript received 25 March 2013; published 4 April 2013)
MnWO4 is regarded as a canonical example of multiferroic materials, where the multiferroic activity is caused
by a spin-spiral alignment. We argue that, in reality, MnWO4 has two sources of the spin spirality, which conflict
with each other. One source is the Dzyaloshinskii-Moriya (DM) interactions, reflecting the P 2/c symmetry of the
lattice. The P 2/c structure of MnWO4 has an inversion center that connects two Mn sublattices. Therefore, from
the viewpoint of DM interactions, different Mn sublattices are expected to have opposite spin chirality. Another
source is competing isotropic exchange interactions, which tend to form a spin-spiral texture with the same
chirality in both magnetic sublattices. Thus there is a conflict between DM and isotropic exchange interactions,
which makes these two sublattices inequivalent and, therefore, breaks the inversion symmetry. Our theoretical
analysis is based on the low-energy model, derived from first-principles electronic structure calculations.
DOI: 10.1103/PhysRevB.87.144403

PACS number(s): 75.30.m, 75.85.+t, 71.15.Rf, 73.22.Gk

I. INTRODUCTION

The idea of breaking the inversion symmetry by some

complex magnetic order has attracted a great deal of attention.
It gives rise to the phenomenon of improper ferroelectricity,
where the ferroelectric (FE) polarization is induced by the
magnetic order and, therefore, can be controlled by the
magnetic field. Alternatively, one can change the magnetic
structure by applying the electric field. This makes multiferroic
materials promising for the creation of the new generation of
electronic devices.1,2
One of magnetic textures that breaks the inversion symmetry is the spin spiral. Moreover, this property is universal, in a
sense that it can take place in all types of magnetic compounds,
irrespectively of their symmetry, as long as the the spin-spiral
order is established.3 Probably, this is the reason why the
idea of the spin-spiral alignment became so popular in the
field of multiferroics: many properties of such compounds
are interpreted from the viewpoint of spin-spiral order,46
and the search for new multiferroic materials is frequently
conducted around those with the spin-spiral order.2
Nevertheless, the spin-spiral alignment alone is insufficient
for having a FE activity: although the spin-spiral order breaks
the inversion symmetry, one can always find some appropriate
uniform rotation of spins, which transforms the inverted spin
spiral to the original one (see Appendix). The FE polarization
in such a situation will be equal to zero. In order to make it
finite (and in order to fully break the inversion symmetry), one
should ban the possibility of uniform rotations in the system
of spins. This can be done by the relativistic spin-orbit (SO)
interaction, and today it is commonly accepted in the theory
of multiferroic materials that the spin-spiral alignment should
be always supplemented with the SO coupling.4,6 However,
as long as the SO coupling is involved, the spin-spiral texture
becomes deformed. Then, we face the question whether the
FE activity is caused by the spin-spiral order itself or by
the deviation from it.7 The answer to this question is very
important for understanding the origin of multiferroicity.
MnWO4 is an important material in the field of multiferroics. Although the transition temperature to the multiferroic
phase is low and the FE polarization is weak, the material
is fundamentally important, because (i) its multiferroic phase
1098-0121/2013/87(14)/144403(7)

has a spin-spiral texture (or it is believed to be the spin spiral)

and (ii) the relative directions of the spin-spiral propagation
vector, the spin rotation axis, and the FE polarization seem to be
consistent with predictions based on the spin-spiral theory.810
Thus MnWO4 is regarded as one of successful manifestations
of the spin-spiral theory and, actually, the discovery of the FE
activity in MnWO4 was guided by this theory.
More specifically, MnWO4 crystallizes in the monoclinic
P 2/c structure. It exhibits three successive antiferromagnetic
(AFM) transitions: at TN3 13.5 K, TN2 12.5 K, and TN1
7.6 K. The AF1 phase, realized below TN1 , is nearly collinear
and the directions of spins alternate along the monoclinic
a axis as (the reason why the AF1 phase is sometimes called the phase). The incommensurate AF2
phase, realized in the temperature interval TN1 < T < TN2 ,
is typically ascribed to the spin spiral with the propagation
vector qAF2 = (0.214,1/2,0.457) (in units of reciprocal
lattice translations). The AF3 phase (TN2 < T < TN3 ) is
characterized by a collinear sinusoidally modulated AFM
order with the same q = qAF2 . The AF2 phase is ferroelectric.
The polarization vector is parallel to the monoclinic b axis:
P = (0,Pb ,0). Pb is relatively small: it takes a maximal value
(50 C/m2 ) at around TN1 , then monotonously decreases
and vanishes at around TN2 .8 Other phases display no sign of
FE activity.
In this work, we will argue that the multiferroicity in
MnWO4 has a more complex origin. Namely, we will show
that, in order to obtain finite polarization, (i) it is not sufficient to have a simple homogeneous spin-spiral texture and
(ii) the spin spiral should be deformed by some conflicting
interactions, existing in the system. In MnWO4 , these are
Dzyaloshinskii-Moriya (DM) interactions and isotropic exchange interactions. The conflict of these two interactions
breaks the inversion symmetry and gives rise to the FE activity
in the AF2 phase.
II. METHOD

We follow the same strategy as in our previous publications,

devoted to multiferroic manganites.7 Our basic idea is to
construct a realistic low-energy (Hubbard-type) model for
the Mn 3d bands, which would describe basic magnetic

144403-1

2013 American Physical Society

I. V. SOLOVYEV

PHYSICAL REVIEW B 87, 144403 (2013)

FIG. 1. (Color online) Total and partial densities of states of

MnWO4 in the local density approximation. The shaded light (blue)
and dark (brown) areas show the contributions of the Mn 3d and
W 5d states, respectively. The positions of the main bands are
indicated by symbols. The Fermi level is at zero energy (shown by
dot-dashed line).

properties of MnWO4 at a semiquantitative level. The model

is constructed in the basis of Wannier orbitals, starting from
electronic structure calculations in the local density approximation (LDA) with and without the relativistic SO interaction.
The Mn 3d bands split into the lower-lying t2g and upper-lying
eg subbands, located at around 0.2 and 0.8 eV, respectively
(see Fig. 1). All these bands are well separated from the rest
of the spectrum. In this case, the Wannier basis is complete
and the construction of the model Hamiltonian is rather
straightforward. More specifically, the Wannier functions were
constructed by the projector-operator method, using atomic
Mn 3d orbitals as the trial functions. Thus the main weight
of the Wannier functions is located at the Mn sites, and small
tails spread to the O and W sites. In the Wannier basis, the
one-electron part of the model Hamiltonian exactly reproduces
the original LDA band structure. The parameters of screened
Coulomb and exchange interactions in the Mn 3d band were
computed by using the combined approach, incorporating
the constrained LDA and the random-phase approximation for
the screening. Other details can be found in the review article
(see Ref. 11). All calculations have been performed using
experimental parameters of the crystal structure.12 Since heavy
W atoms are an important source of relativistic interactions,
the SO coupling was included at the level of ordinary LDA
calculations, and thus the obtained electronic structure was
used as the starting point for the construction of the low-energy
model. Thus, although the W sites were not explicitly included
to the model, the relativistic effects, associated with these sites,
were allowed to contribute to the dispersion of the Mn 3d
bands and, therefore, were taken into account in the low-energy
model. All parameters of the model Hamiltonian are collected
in Ref. 13. After the construction, the model was solved in the
mean-field Hartee-Fock (HF) approximation.
We would also like to briefly comment on merits and
demerits of our technique in comparison with the ordinary
first-principles approach.
(i) It is certainly true that the construction of the model
Hamiltonian is based on some additional approximations,11

and is sometimes regarded as a step back in comparison with

ordinary first-principles calculations. On the other hand, it
would not be right to think that the first-principles calculations
for the transition-metal oxides are free of any approximations.
The necessity to treat the problem of on-site Coulomb correlations, which is frequently formulated in terms of the LDA + U
approach (with some phenomenological correction +U ,
borrowed from the Hubbard model) make it similar to the
model approach: it also relies on additional approximations,
although of a different type. The typical approximations are the
choice of the parameter U and the form of the double-counting
term, which are both ill-defined in the LDA + U .14 Thus,
at the present stage, it would not be right to say that one
techniques is more superior than other; rather, they provide
a complementary information for the analysis of material
properties of the transition-metal oxides. As we will see below,
results of our model analysis are well consistent with available
first-principles calculations for MnWO4 .15,16
(ii) The search of the ground state for noncollinear magnetic
textures can be very time consuming, even at the level of meanfield HF approximation: the rotations of magnetic moments
towards new equilibrium directions after including the SO
interaction can be very slow and require tens of thousands
of iterations, as in the case of MnWO4 . In such a situation,
the model Hamiltonian approach is very useful, because it
allows us to find a fully self-consistent magnetic texture,
which is not always accessible for the ordinary first-principles
calculations.7
(iii) The model Hamiltonian approach can be very useful for
the analysis and interpretation of results of complex electronic
structure calculations and the experimental data. Particularly,
in this work, it will help us to elucidate the microscopic origin
of the magnetic inversion symmetry breaking in MnWO4 . We
will also show that the experimental behavior of long-range
magnetic interactions in MnWO4 (see Ref. 17) can be naturally
understood in the framework of the superexchange (SE) theory
and reflects similar behavior of the transfer integrals in the
low-energy model.
III. RESULTS AND DISCUSSIONS

First, we solve the low-energy model in the HF approximation, by assuming the collinear ferromagnetic (FM) alignment of spins, and derive parameters of isotropic exchange
interactions. For these purposes we employ the theory of
infinitesimal spin rotations.11,18 The procedure corresponds
to the local mapping of the change of the one-electron
energy onto
 isotropic spin Hamiltonian of the Heisenberg form
HH = i>j Jij ei ej , where ei is the direction of spin at
the site i and the summation run over inequivalent pairs of
sites. The results of these calculations are explained in Fig. 2.
Alternatively, one can apply the theory of superexchange
(SE) interactions, by considering the energy gain caused by

virtual hoppings (tijmm ) in the second order of perturbation
of Mn2+ , this expression is
theory.19 For the d 5 configuration



extremely simple: Jij = mm tijmm tjmi m /ex , where ex is
the intraatomic splitting between the majority- and minorityspin states and the summation runs over all 3d orbitals: m
(m ) = xy, yz, 3z2 r 2 , zx, and x 2 y 2 .20 Thus all SE
interactions are expected to be antiferromagnetic. Then, by

144403-2

ORIGIN OF MULTIFERROICITY IN MnWO4

PHYSICAL REVIEW B 87, 144403 (2013)

FIG. 2. (Color online) (a) Lattice of Mn sites with the notations of isotropic exchange interactions (numbered in the increasing order of
interatomic distances). Note that the P 2/c structure of MnWO4 has two Mn sublattices, which are shown by different colors and denoted as
I and I I , respectively. These sublattices are transformed to each other by the inversion operation. (b) Distance dependence of isotropic
exchange interactions: results of calculations, using the theory of infinitesimal spin rotations near the ferromagnetic state (denoted as inf)
and the theory of superexchange interactions with ex = 5 eV (denoted as SE), in comparison with the experimental data from Ref. 17.


taking ex = 5 eV and using the values of tijmm , which

were obtained for the low-energy model (see Ref. 13), one
can find that the SE theory excellently reproduces results
of the more general theory of infinitesimal spin rotations.
Moreover, in the framework of the Hubbard model, ex can be
estimated as ex U + 4JH . The values of averaged on-site
Coulomb repulsion U and the intraatomic exchange interaction
JH , derived for the low-energy model, are 1.8 and 0.8 eV,
respectively.13 This yields ex = 5 eV. Thus all estimates are
consistent with each other. Therefore one may conclude that
(i) the physically relevant mechanism, responsible for isotropic
exchange interactions in MnWO4 , is the superexchange,
and (ii) the experimental distance-dependence of Jij reflects

similar behavior of the transfer integrals tijmm .
Our calculations capture main details of the experimental
distance-dependence of Jij (see Fig. 2).17 This is a very
important finding, which means that in our model we should
be able to reproduce the correct magnetic structures of
MnWO4 . The negative aspect is that our parameters Jij are
too antiferromagnetic (AFM). For example,
 the Curie-Weiss
temperature can be estimated as CW j Jij /3kB . Then,
the obtained parameters Jij yield CW = 265 K, which
exceeds the experimental value 78 K (see Ref. 8) by factor
three. The discrepancy cannot be resolved simply by changing
the value of the on-site Coulomb repulsion U , which is
frequently treated as an adjustable parameter. Particularly,
in order to explain the existence of FM interactions, which
were observed experimentally in some of the bonds,17 we
need an additional mechanism on the top of our model. Such
a mechanism can be related to the magnetic polarization of
the oxygen band, similar to orthorhombic manganites.21 The
first-principles GGA + U calculations (where GGA stands
for the generalized gradient approximationan extension
of LDA) also substantially overestimate |Jij | and |CW |.15
Although these calculations take into account the effect of
the oxygen band, the disagreement is probably caused by
additional approximations for the double-counting energy,
which is ill-defined in GGA + U .14 Finally, the experimental
parameters Jij themselves may be sensitive to the details of
fitting of the spin-wave dispersion. In this respect, we would
like to mention that earlier experimental data were interpreted
in terms of rather different set of parameters Jij .22 Thus, at

the present stage, taking into account the complexity of the

problem of interatomic magnetic interactions in MnWO4 , the
agreement between theoretical and experimental data in Fig. 2
can be regarded as satisfactory.
Then, we search the magnetic ground state of MnWO4
without relativistic SO coupling. For these purposes, we employ the spin-spiral formalism, which is based on generalized
Bloch theorem.3 The idea is to combine the lattice translations
with rotations in the spin subspace. It allows us to treat the
incommensurate magnetic textures with an arbitrary vector
q = (qa ,1/2,qc ) such as if the magnetic unit contained only
one crystallographic cell.3 The results of these calculations,
which were also performed in the HF approximation for
the low-energy model, are summarized in Fig. 3. For q =
(1/4,1/2,1/2), the collinear and noncollinear spinspiral states are nearly degenerate. Nevertheless, the total
energy of the spin spiral continues to decrease with the increase
of qa , and MnWO4 is expected to form the spin-spiral ground
state with qNR (0.18,1/2,0.45), which is lower in energy
than the state by about 0.5 meV/Mn. Moreover, none
of these states is ferroelectric: although the spin spiral breaks
the inversion symmetry, the latter can be combined with an
appropriate rotation of spins, which transforms the inverted

FIG. 3. (Color online) Results of Hartree-Fock calculations for

the spin-spiral states with q = (qa ,1/2,qc ): total energies versus qa
for different values of qc (in units of 2/c). The energy of the collinear
state is marked by the symbol .

144403-3

I. V. SOLOVYEV

PHYSICAL REVIEW B 87, 144403 (2013)

spin spiral to the original one (see Appendix). Thus the FE

polarization will vanish.
Therefore we should find the answer to the following
questions: (i) which interaction stabilizes the AF1 texture and
make it the ground state of MnWO4 at low T ? (ii) What is
the mechanism of the inversion-symmetry breaking in the
AF2 phase, which yields finite Pb ? For these purposes, we
consider the relativistic SO interaction. It gives rise to such
important ingredients as the single-ion anisotropy and DM
interactions.
The effect of single-ion anisotropy on the magnetic texture
of MnWO4 can be studied by enforcing the atomic limit
and setting all transfer integrals equal to zero.23 Generally,
the single-ion anisotropy is expected to be small for the
nearly spherical d 5 configuration of Mn2+ . Since the symmetry
operation {Cb2 |c/2} of the space group P 2/c transforms each
Mn site to itself (here, Cb2 stands for the 180 rotation around
the b axis and c/2 is the additional translation), the single-ion
anisotropy will tend to align the spins either parallel to the
b axis (||b) or in the ac plane (ac). In the latter case, the spins
are canted off the a axis by about 41 , which is close to the
experimental value of 35 .8 The easy magnetization direction
corresponds to the in-plane configuration. However, the energy
difference between hard (||b) and easy (ac) magnetization
directions is small (less that 0.05 meV/Mn), so that in the
spin-spiral texture, it can be easily overcome by the energy
gain caused by isotropic exchange interactions.
For discussing the DM interactions, it is more convenient to
use the extended notations and recall that each site of the lattice
i can be specified by its position in the primitive cell and the
translation R. Then, since the P 2/c structure of MnWO4 has
two Mn-sublattices (denoted as I and I I in Fig. 2), which
can be transformed to each other by the inversion operation,
the DM interactions will obey the following symmetry rules.24
(i) All interactions between different sublattices are equal to
zero. (ii) The DM interactions in each magnetic sublattice
will depend only on the vector R, connecting two Mn sites.
Then, the DM interactions in the sublattices I and I I
I
will be related by the following identity dIR = dIR
= dIRI ,
which holds any R (note that d is the axial antisymmetric
vector). Moreover, due to the symmetry operation {Cb2 |c/2},
all DM vectors lie in the ac plane, i.e., similar to the easy
magnetization direction, obtained from the analysis of the
single-ion anisotropy.
To be specific, let us define the DM spin Hamiltonian as

I (I I )
I)
HDM
= R >R dIR(I R
[e R e R ], which is constructed
separately for each magnetic sublattice. In these notations,
e R is the direction of spin, where + and correspond
to the sublattices I and I I , respectively. Moreover, in each
magnetic sublattice, we can shift the origin and place it in
+ and , respectively. Then, the DE spin Hamiltonian

I (I I )
I)
becomes HDM
= R >R dIR(I R
[e R e R ]. Moreover, we use
the convention where the sites R and R order in the positive
direction of the a axis, and (R R) a > 0.
The parameters of DM interactions can be evaluated by
considering the mixed type of perturbation theory expansion
with respect to the SO coupling and infinitesimal rotations
of spins near the FM state.25 Then, the obtained changes
of the one-electron energy can be mapped onto the DM
Hamiltonian. It yields the following parameters of DM interac-

tions (in meV): dIa = (0.01,0,0.01), dIc = (0.02,0,0.02),

and dIa+c = (0.01,0,0.01). Interactions in other bonds are
considerably weaker.
The equilibrium magnetic texture, obtained after switching
on the relativistic SO interaction in the collinear state,
is explained in Fig. 4. Due to the single-ion anisotropy, all
spins are confined mainly in the ac plane and canted off the
a axis by about 46 (that is close to 41 , obtained in the
atomic limit). The DM interactions yield an additional canting
of spins out of the ac plane. The corresponding force f R =
HDM /e R , experienced by the Mn spin at the site R from
its
 neighboring sites in
the same sublattice, is given by f R =


 [d R  R e R  ]
R
R [d R e R+R ]. Here, we drop the
sublattice indices, because all the arguments equally apply
to both magnetic sublattices. Since dR = d R , in order to
contribute to f R , the directions of neighboring spins should
satisfy the following condition: e RR = e R+R . In the
texture, such a situation takes place for R = a and a + c, but
not for R = c. Moreover, since both d and e lie in the ac
plane, the force will be parallel to the b axis. Finally, due
to the alternation of e in the texture, the directions of
forces will also alternate. All in all, this explains fine details
of the magnetic structure in Fig. 4(b). Very importantly, the
isotropic exchange interactions, single-ion anisotropy, and DM
interactions in the AF1 phase do not conflict with each other
in a sense that all of them, even jointly, continue to respect
the crystallographic P 2/c symmetry. Therefore the inversion
symmetry is preserved and the polarization is equal to zero,
even after including the SO interaction.26
The effect of SO interaction on the spins-spiral alignment
can be best understood for q = (1/4,1/2,1/2) (see Fig. 5).
In this structure, the easy magnetization direction (e ac)
at certain Mn-site alternates with the hard magnetization
direction (e||b) at its neighboring sites. Thus there is a
conflict between isotropic exchange interactions and the

FIG. 4. (Color online) (a) Arrangement of spins in the AF1 phase,

as obtained in the mean-field HF calculations for the low-energy
model with relativistic SO interaction and (b) projection of spins
onto the ab plane.

144403-4

ORIGIN OF MULTIFERROICITY IN MnWO4

PHYSICAL REVIEW B 87, 144403 (2013)

FIG. 5. (Color online) (a) The deformed spin-spiral texture with

q = (1/4,1/2,1/2), as obtained in the mean-field HF calculations
for the low-energy model with the relativistic SO interaction, and
(b) its projection onto the ac plane.

single-ion anisotropy. However, since the latter is small, this

magnetic structure can be realized at a low-energy cost.
Then, due to the DM interactions, each site with e||b will
experience a force from its neighboring sites with e ac. Since
d ac, this force will be parallel to the b axis. Alternatively,
each site with e ac will experience the force from its
neighboring sites with e||b. Since d ac, this force will also
lie in the ac plane and additionally rotate the spin in this plane.
Furthermore, since dIR = dIRI , the forces in different
sublattices will act in the opposite directions. The effect of
these forces can be best understood in terms of the vector
spin chirality. For a spin spiral, propagating along the a axis,
all cross products [e R e R ], constructed between the nearest
neighbors, are the same, and such state can be described by
the vector spin chirality k = [e R e R ]. Obviously, if one
considers the DM Hamiltonian alone, which can be defined
separately for each magnetic sublattice, the direction of k will
be controlled by the parameters of DM interactions. Then,
since dIR = dIRI , in order to minimize the energy of DM
interactions, similar relation should hold for the spin chirality:
kI = kI I . This means that the DM interactions will tend
to form the state with opposite chirality in two magnetic
sublattices. However, the spin-spiral texture that is formed
by isotropic exchange interactions Jij has the same chirality
in both magnetic sublattices. Therefore we have a conflict
between isotropic and DM interactions: if in one magnetic
sublattice the effect of these two interactions is added, in
the other magnetic sublattice, it will be subtracted. All these
tendencies are clearly seen in the fine details of the magnetic
texture, obtained in the HF calculations, where we start from
the homogeneous spin-spiral state and switch on the SO
interaction (see Fig. 5). First, the canting of spins, lying in
the ac plane, appears to be different for the sublattices I and
I I : 21 and 34 , respectively, relative to the a axis. Second, as
discussed above, the spins with e||b will experience the force
f, which is also parallel to b. However, since dIR = dIRI ,
the vectors e and f at the same site will be either parallel or
antiparallel, depending on the magnetic sublattice. Thus the
magnitude of local spin magnetic moment with e||b will also
depend on the magnetic sublattice. Of course, the difference
of the magnetic moments is small (about 0.01%, according
to our HF calculations). Nevertheless, it does take place and
is just another manifestation of the conflict between isotropic
and DM interactions.

Thus the conflict between isotropic and DM interactions in

the AF2 phase makes the sublattices I and I I inequivalent.
Since in the P 2/c structure, these two sublattices are connected
by the inversion operation, their inequivalency means that the
inversion symmetry is broken and the actual symmetry of the
AF2 phase is lower than P 2/c. Furthermore, the homogeneous
spin-spiral alignment itself is deformed by the DM interactions
and, strictly speaking, the AF2 phase is no longer the spin
spiral. The magnitude of this deformation can be seen in the
inset of Fig. 5.
Then, we discuss the relative stability of different magnetic
phases with the SO interaction. In this case, the generalized
Bloch theorem is not applicable and the only possibility is
to work with the supercell geometry.3 Therefore we were
able to consider only the solutions with q = (qa ,1/2,1/2)
and the rational qa =1/3, 1/4, and 1/5. The results
can be summarized as follows. Amongst the AF2 states, the
one with qa = 1/5 has the lowest energy. The state with
qa = 1/4 is higher in energy by about 0.2 meV/Mn. This
behavior is similar to calculations without SO coupling (see
Fig. 3). Then, the AF2 state with qa = 1/5 appears to be
nearly degenerate with the AF1 state (the energy difference is
about 0.07 meV/Mn, but the AF2 state is still low in energy).
Although these numerical values are probably on the verge of
accuracy of our model analysis, this tendency clearly shows
that the SO interaction additionally stabilizes AF1 state relative
to the AF2 one. This seems to be reasonable: the conflict of
DM and isotropic exchange interactions, acting in the opposite
directions in one of the Mn sublattices, will penalize the energy
of the AF2 state. On the other hand, in the AF1 state, there
is no such conflict and the DM interactions will additionally
minimize the energy of the texture, obtained without
SO interaction. Apparently, the main reason why the energy of
the AF2 phase is still slightly lower than that of the AF1 phase
is related to the fact that the isotropic exchange interactions
are overestimated in our low-energy model (see Fig. 2) and,
therefore, the effect of the SO coupling and DM interactions
on the magnetic texture is underestimated.
Finally, we evaluate the value of the FE polarization in
the AF2 phase, using the Berry phase formalism,27 which was
adopted for the low-energy model.7 Here, we calculate only the
electronic contribution, which arises in the centrosymmetric
P 2/c structure due to the magnetic inversion symmetry
breaking. However, we do not consider the structural relaxation
in response to this magnetic symmetry breaking. The vector
of polarization is parallel to the b axis, in agreement with
the experiment.8 Then, the value of Pb can be estimated as
2.0, 3.8, and 4.4 C/m2 for qa = 1/3, 1/4, and 1/5,
respectively. Thus Pb is small, but it is well consistent with
the fact that the DM interactions are also small and lead
only to small perturbation of the homogeneous spin-spiral
texture, that is formed by the isotropic exchange interactions.
Nevertheless, we would like to emphasize that the DM
interactions are essential for deforming the spin spiral, breaking the inversion symmetry, and producing the finite value
of Pb .
The experimental polarization is an order of magnitude
larger: Pb 50 C/m2 .8 However, it can be still regarded as
a small value in comparison with many other multiferroic
systems.1 Therefore it is quite consistent with our main idea

144403-5

I. V. SOLOVYEV

PHYSICAL REVIEW B 87, 144403 (2013)

that Pb is a result of a small perturbation of the spin-spiral

texture, caused by the relativistic effects.
There may be several reasons why the experimental value
of Pb is larger than the theoretical one. (i) The experimental
Pb may also include some lattice effects, in response to
lowering of the magnetic symmetry.16 (ii) An unusual aspect
of MnWO4 is that the inversion symmetry is broken in the
high-temperature phase, while the low-temperature phase
remains centrosymmetric (for comparison, the situation in
perovskite manganites is exactly the opposite).1 Thus, in addition to the FE polarization, one should find some mechanism,
which would explain the AF1AF2 phase transition at finite
T . This mechanism may involve some temperature effects.
(iii) The low-energy model is designed for the semiquantitative analysis. It is not always possible to expect a good
quantitative agreement with the experimental data, because
some ingredients can be missing in the model. In principle,
the GGA + U calculations are also of a semiquantitative level,
because the value of Pb depends on the adjustable parameter
U .28 In our model analysis, the small value of Pb may be
related to the overestimation of isotropic exchange interactions
(see Fig. 2). Then, the deformation of the spin-spiral texture,
caused by the relativistic effects, is underestimated. Therefore
the value of Pb is also underestimated. From this point of
view, the behavior of isotropic exchange interactions, the total
energies, and the FE polarization is consistent with each other,
and the main efforts towards quantitative description of the FE
polarization and the phase diagram of MnWO4 should be concentrated on the quantitative description of isotropic exchange
interactions.
IV. SUMMARY

In summary, we have provided the microscopic explanation

for the origin of FE activity in MnWO4 . The multiferroicity in
this compound is caused by the conflict of DM and isotropic
exchange interactions in the AF2 phase. Thus MnWO4 is
multiferroic not simply because of the spiral texture. It is
essential to have conflicting interactions, which deform the
spin spiral, break the inversion symmetry, and, thus, give rise
to the FE activity.
Here, one can also make some analogy with the weak
ferromagnetism.24 In the AFM materials, two magnetic sublattices are connected by the time-reversal operation. The SO
interaction can break this time-reversal symmetry and under
certain conditions, related to the symmetry of the crystal and
the type of the AFM order, give rise to the phenomenon
of weak ferromagnetism. One of microscopic interactions,
responsible for this effect, is the DM interaction. In the P 2/c
structure of MnWO4 , two sublattices are connected by the
spacial inversion. Thus the MnWO4 can be regarded as an
antiferroelectric material. The SO interactions can break the
inversion symmetry and give rise to a weak FE polarization.
Thus, in an analogy with the weak ferromagnetism, this
phenomenon can be called weak ferroelectricity. On the
microscopic level, it is also related to the behavior of DM
interactions. Again, like in weak ferromagnetism, the existence
of the weak ferroelectricity in MnWO4 depends on two factors:
(i) the specific symmetry of the crystal (in our case, the P 2/c
symmetry), and (ii) the specific type of the magnetic order (in

our case, the AF2 order). It should not be confused with some
general properties of the spin spiral.
ACKNOWLEDGMENT

This work is partly supported by the grant of the Ministry

of Education and Science of Russia No. 14.A18.21.0889.
APPENDIX: NONEXISTENCE OF FERROELECTRICITY
IN HOMOGENEOUS SPIN-SPIRAL STATE WITHOUT
SPIN-ORBIT INTERACTION

In this Appendix, we will show that, although the spatial

inversion is not the symmetry operation of the homogeneous
spin-spiral state, it can be always combined with an appropriate
rotation of spins, which transforms the inverted spin-spiral
texture to the original one. The ferroelectric polarization in
such a situation will be equal to zero. The prove is extremely
simple, but first we would like to illustrate the basic idea on a
cartoon picture for the one-dimensional spin spiral (see Fig. 6).
The basic property of the spin spiral is such that: if e is the
is the direction
direction of spin at certain magnetic site and Ie
of spin at the same site after the inversion of the lattice, one

can always define the axis n||(e + Ie),

which has the same
direction at all magnetic sites. Then, the inverted spin spiral
can be transformed to the original one by the uniform 180
rotation of spins around this axis.
Now, we will provide a rigorous prove of this statement.
To be specific, consider the situation, realized in MnWO4 and
assume that there are two sites in the primitive cell, which are
located at and , respectively. Thus these two cites can be
transformed to each other by the inversion operation I and the
inversion center is located in the origin. Then, the translations
R will specify the location of all other inversion centers, while
the atomic position will be given by the vectors R .
In the spin-spiral texture, the directions of magnetic
moments are given by

cos q (R )

(A1)
e R = sin q (R ) .
0

FIG. 6. (Color online) Cartoon picture, explaining how the

inverted spin spiral can be transformed to the original one by the
uniform rotation of spins. The inversion center is marked by cross.
The directions of magnetic moments in the original spin spiral (e) are
shown by the dark (blue) vectors, and those in the inverted spin spiral
are shown by the light (cyan) vectors. The inverted spin-spiral
(Ie)
texture can be transformed to the original one by the 180 rotation of
which has the same direction at all
spins around the axis n||(e + Ie),
sites of the lattice.

144403-6

ORIGIN OF MULTIFERROICITY IN MnWO4

PHYSICAL REVIEW B 87, 144403 (2013)

For our purposes, it is sufficient to consider the situation where

all spins rotate in the ab plane. The generalization for an
arbitrary orientation of the rotation plane is straightforward
and will be discussed at the end of this Appendix.
Consider the inversion around an arbitrary center R0 . It
transforms an arbitrarily taken site R to R0 (R
R0 ) = 2R0 R . Then, the direction of spin at the site
2R0 R will change from e2R0 R to e R . Thus our
which would rotate e R
goal is to find a transformation R,
back to e2R0 R . This transformation is the 180 rotation
around the axis n||(e2R0 R + e R ). The corresponding
rotation matrix is given by

sin(2q R0 )
0
cos(2q R0 )

=
0.
R
sin(2q R0 ) cos(2q R0 )
0
0
1
Indeed, using the expressions for e2R0 R and e R , given
R =
by Eq. (A1), it is straightforward to verify that Re

e2R0 R . Since R does not depend on R, it corresponds

solovyev.igor@nims.go.jp
Y. Tokura, Science 312, 1481 (2006); S.-W. Cheong and
M. Mostovoy, Nat. Mater. 6, 13 (2007); D. Khomskii, Physics 2,
20 (2009).
2
T. Kimura, Annu. Rev. Mater. Res. 37, 387 (2007); Y. Tokura and
S. Seki, Adv. Mater. 22, 1554 (2010).
3
L. M. Sandratskii, Adv. Phys. 47, 91 (1998).
4
H. Katsura, N. Nagaosa, and A. V. Balatsky, Phys. Rev. Lett. 95,
057205 (2005).
5
M. Mostovoy, Phys. Rev. Lett. 96, 067601 (2006).
6
I. A. Sergienko and E. Dagotto, Phys. Rev. B 73, 094434
(2006).
7
I. V. Solovyev, Phys. Rev. B 83, 054404 (2011); I. V. Solovyev,
M. V. Valentyuk, and V. V. Mazurenko, ibid. 86, 144406 (2012).
8
K. Taniguchi, N. Abe, T. Takenobu, Y. Iwasa, and T. Arima, Phys.
Rev. Lett. 97, 097203 (2006).
9
A. H. Arkenbout, T. T. M. Palstra, T. Siegrist, and T. Kimura, Phys.
Rev. B 74, 184431 (2006).
10
O. Heyer, N. Hollmann, I. Klassen, S. Jodlauk, L. Bohaty, P. Becker,
J. A. Mydosh, T. Lorenz, and D. Khomskii, J. Phys.: Condens.
Matter 18, L471 (2006).
11
I. V. Solovyev, J. Phys.: Condens. Matter 20, 293201 (2008).
12
G. Lautenschlager, H. Weitzel, T. Vogt, R. Hock, A. Bohm,
M. Bonnet, and H. Fuess, Phys. Rev. B 48, 6087 (1993).
13
See Supplemental Material at http://link.aps.org/supplemental/
10.1103/PhysRevB.87.144403 for parameters of the crystal field,
transfer integrals, and matrices of Coulomb interactions.
14
I. V. Solovyev and K. Terakura, Phys. Rev. B 58, 15496 (1998).
15
C. Tian, C. Lee, H. Xiang, Y. Zhang, C. Payen, S. Jobic, and M.-H.
Whangbo, Phys. Rev. B 80, 104426 (2009).
16
K. V. Shanavas, D. Choudhury, I. Dasgupta, S. M. Sharma, and
D. D. Sarma, Phys. Rev. B 81, 212406 (2010).
1

to the uniform rotation of spins. Without SO coupling, the

spin system is fully isotropic and such transformation is
I is
permissible. Thus, although I is formally broken, R
the symmetry operation, which transforms the spin spiral to
does not affect
itself. Since the uniform rotation of spins R
the polarization, the existence of the symmetry operation
I means that the ferroelectric activity in the homogeneous
R
spin-spiral state is forbidden.
This is a general property of the spin-spiralalthough
we have considered a specific lattice geometry, which is
more relevant to MnWO4 , absolutely the same argument can be
repeated, for example, for orthorhombic manganites, where the
magnetic sites are located in the centers of inversion. The only
requirement is that the lattice itself has inversion symmetry
+ ), can be also
and for each site, R + , its image, I(R
found amongst the lattice sites.
For an arbitrary orientation of the spin rotation plane, e R
R , where D is the
in Eq. (A1) should be replaced by De
rotation of the ab plane. Then, all above arguments apply after
by D R
D 1 .
replacing R

17

F. Ye, R. S. Fishman, J. A. Fernandez-Baca, A. A. Podlesnyak, G.

Ehlers, H. A. Mook, Y. Wang, B. Lorenz, and C. W. Chu, Phys.
Rev. B 83, 140401(R) (2011).
18
A. I. Liechtenstein, M. I. Katsnelson, V. P. Antropov, and V. A.
Gubanov, J. Magn. Magn. Matter. 67, 65 (1987).
19
P. W. Anderson, Phys. Rev. 115, 2 (1959).
20
Note that in the atomic limit, all majority-spin states are occupied
and the minority-spin states are empty. Therefore, for the antiferromagnetically coupled spins, the virtual hoppings from the subspace
of occupied states at the site i to the subspace of unoccupied states
at the site j will involve all combinations of the orbital indices m
and m . Furthermore, the energy cost for transferring an electron
between two atomic sites is ex (the energy splitting between
occupied and empty states in the atomic limit). All together, this
explains the expression for the superexchange interactions.
21
I. Solovyev, J. Phys. Soc. Jpn. 78, 054710 (2009).
22
H. Ehrenberg, H. Weitzel, H. Fuess, and B. Hennion, J. Phys.:
Condens. Matter 11, 2649 (1999).
23
I. V. Solovyev, Phys. Rev. B 85, 054420 (2012).
24
I. Dzyaloshinsky, J. Chem. Phys. Solids 4, 241 (1958); T. Moriya,
Phys. Rev. 120, 91 (1960).
25
I. Solovyev, N. Hamada, and K. Terakura, Phys. Rev. Lett. 76, 4825
(1996).
26
More specifically, the AF1 phase can be transformed to itself
by combining the spacial inversion, the translation R = a,
and (optionally) the time-reversal operation (depending on the
translation and the location of the inversion center).
27
D. Vanderbilt and R. D. King-Smith, Phys. Rev. B 48, 4442 (1993);
R. Resta, J. Phys.: Condens. Matter 22, 123201 (2010).
28
Two groups reported reported rather different values of Pb ,
depending on the parameter U : Pb 17.2 C/m2 (see Ref. 15,
for U = 6 eV) and Pb 42 C/m2 (see Ref. 16, for U = 4 eV).

144403-7