PHYSICAL REVIEW B 96, 024451 (2017)

Piezomagnetism as a counterpart of the magnetovolume effect in magnetically frustrated

Mn-based antiperovskite nitrides
J. Zemen,1,2,* Z. Gercsi,1,3 and K. G. Sandeman1,4,5
1
Department of Physics, Blackett Laboratory, Imperial College London, London SW7 2AZ, United Kingdom
2
Faculty of Electrical Engineering, Czech Technical University in Prague, Technická 2, Prague 166 27, Czech Republic
3
CRANN and School of Physics, Trinity College Dublin, Dublin 2, Ireland
4
Department of Physics, Brooklyn College, CUNY, 2900 Bedford Ave., Brooklyn, New York 11210, USA
5
Physics Program, The Graduate Center, CUNY, 365 Fifth Avenue, New York, New York 10016, USA
(Received 30 December 2016; revised manuscript received 6 June 2017; published 31 July 2017)

Electric-field control of magnetization promises to substantially enhance the energy efficiency of device
applications ranging from data storage to solid-state cooling. However, the intrinsic linear magnetoelectric
effect is typically small in bulk materials. In thin films, electric-field tuning of spin-orbit-interaction phenomena
(e.g., magnetocrystalline anisotropy) has been reported to achieve a partial control of the magnetic state. Here
we explore the piezomagnetic effect (PME), driven by frustrated exchange interactions, which can induce a net
magnetization in an antiferromagnet and reverse its direction via elastic strain generated piezoelectrically. Our
ab initio study of PME in Mn-based antiperovskite nitrides identified an extraordinarily large PME in Mn3 SnN
available at room temperature. We explain the magnitude of PME based on features of the electronic structure
and show an inverse proportionality between the simulated zero-temperature PME and the magnetovolume effect
at the magnetic (Néel) transition measured by Takenaka et al. in nine antiferromagnetic Mn3 AN systems.

DOI: 10.1103/PhysRevB.96.024451

I. INTRODUCTION effect (PME) which is characterized by a net magnetization

that is directly proportional to the applied lattice strain [6,7].
Emerging nonvolatile magnetic random access memory
Fully compensated AFM states are hard to track and utilize
(MRAM) devices represent bits of information as a magne-
in general but the PME offers a valuable technique to probe
tization direction which needs to be stabilized by magnetic
and control the AFM ordering via the strain-induced magnetic
anisotropy. A spin-transfer torque (STT) is typically used to
moment.
overcome the energy barrier between two stable directions.
To substantiate the future use of the PME in magnetoelectric
STT is induced by passing spin-polarized current which
composites, we perform a systematic ab initio study of PME
leads to Joule heating and sets limits on the storage density.
in nine cubic antiperovskites Mn3 AN (A = Rh, Pd, Ag, Co,
Much research is focused on alternative switching mechanisms
Ni, Zn, Ga, In, Sn). We explain the variation of the magnitude
based on direct or indirect electric-field control of magnetic
of PME across this range of compounds based on features
anisotropy which can reduce the dissipated energy by a
of their electronic structure. The PME in Mn3 SnN predicted
factor of 100 [1]. At the same time antiferromagnetic (AFM)
here is an order of magnitude larger than the PME modelled
spintronics [2] explores alternatives to STT-MRAM devices
so far in Mn3 GaN [7]. Moreover, we compare the simulated
based on active AFM components with a bistable alignment
PME to the measured magnetovolume effect (MVE) at a
of the local magnetic moments. The switching then utilizes a
magnetic (Néel) transition temperature [8] across the full
spin-orbit torque (SOT) induced by an unpolarized electrical
set of nine systems and shown that the effects are inversely
current in collinear AFMs [3,4] or STT due to a spin-polarized
proportional to each other. This agreement with experimental
current in proposed noncollinear AFM junctions [5]. There
data is remarkable because both the PME and MVE originate in
is no dipolar coupling between neighboring elements and
the frustrated AFM structure but we simulate the PME at zero
they are insensitive to external magnetic fields. Again this
temperature whereas the MVE was measured at the magnetic
alternative promises a higher storage density and energy
(Néel) transition temperature. The MVE has not been modelled
efficiency.
for this set of systems before.
Here we explore the underlying physics required to postu-
Note that Mn3 AgN and Mn3 RhN do not share the triangular
late an ambitious approach combining the electric-field control
AFM order according to earlier neutron-diffraction studies [9],
of magnetic moments with the noncollinear antiferromagnetic
whereas the magnetic structure of Mn3 CoN and Mn3 PdN is
structure of Mn-based antiperovskite nitrides. The required
yet to be confirmed experimentally. We include these four
coupling between the spin and orbital degrees of freedom is
compounds in our study as their composition, AFM order,
dominated by geometrically frustrated exchange interactions.
and experimentally resolved MVE [8] makes them potential
The indirect magnetoelectric effect (ME) is hosted by a piezo-
candidates for piezomagnetic behavior.
magnetic Mn-based antiperovskite layer elastically coupled
In more general terms, we perform a computational experi-
to a piezoelectric substrate. We focus on the piezomagnetic
ment when the magnetic system is initialized in the triangular
state (representation  4g or  5g as described below) even if
it was only a local energy minimum for Mn3 AN (A = Ag,
*
[email protected] Co, Pd, Rh) and the response (induced magnetization) to

2469-9950/2017/96(2)/024451(8) 024451-1 ©2017 American Physical Society

J. ZEMEN, Z. GERCSI, AND K. G. SANDEMAN PHYSICAL REVIEW B 96, 024451 (2017)

a tetragonal distortion is detected. The consistency of the

piezomagnetic response across the whole set of materials
motivates us to use this procedure as a probe of the level of
frustration of the exchange interaction even if the real systems
do not host piezomagnetism.

A. Mn-based antiperovskites
Mn-based antiperovskite nitrides were first examined in
the 1970s [10,11]. More recent experimental work on these
metallic compounds includes a demonstration of large negative
thermal expansion (NTE) in Mn3 AN (A = Ga, Zn, Cu, Ni)
[12–15] at the first-order phase transition to a PM state. A
large barocaloric effect was measured in Mn3 GaN at TN = 288
K [16] and the Mn-based antiperovskites were consequently
proposed as a new class of mechanocaloric materials. More im-
portantly for spintronic applications, the baromagnetic effect
(BME), which is closely related to the PME, was reported in
Mn3 G0.95 N0.94 very recently [17], the exchange bias effect was
observed in Mn3 GaN/Co3 FeN bilayers [18], perpendicular
magnetic anisotropy was demonstrated in Mn67 Ga24 N on MgO FIG. 1. Mn-based antiperovskite magnetic unit cell, cubic and
substrate, and the magnetocapacitance effect was measured in strained lattice assuming Poisson’s ratio of 0.5; the canting and
Mn3 GaN/SrTiO3 bilayers [19]. changes of size are not to scale. (a) Unstrained structure of Mn3 GaN
Theoretical work on Mn-based antiperovskites includes an with local moments on Mn sites according to  5g representation. (b)
Tensile strained magnetic order in (111) plane; Mnet indicates the
early tight-binding study [20] suggesting that the proximity of
direction of the induced net moment. (c) Compressively strained unit
the Fermi energy to a sharp singularity (narrow Np -Mnd band)
cell; (d) Tensile strained unit cell.
in the electronic density of states has a large influence on the
stability of the structural and magnetic phases. However, this
model considers only nearest-neighbor Mn-N hopping and magnetization:
neglects any hybridization with atom A. Phenomenological
∂F ∂F0
studies analyzed phase transitions [21], magnetoelastic, and Mi = − =− + λi,j k σj k + μi,j k σj2k . (2)
piezomagnetic [6] properties with respect to the symme- ∂Hi ∂Hi
try of the crystal and magnetic structure. More recently The PME was first proposed by Voigt [27] in 1928. The
ab initio modeling of the noncollinear magnetic structure has linear character limits its existence to systems without time-
been carried out. The NTE and MVE are attributed to the inversion symmetry or with a magnetic group that contains
frustrated exchange coupling between the three Mn atoms time inversion only in combination with other elements
[14,22,23]. The local spin density has been simulated for of symmetry [28]. Hence, the PME is forbidden in all
Mn3 GaN and Mn3 ZnN revealing its distinctly nonuniform paramagnetic and diamagnetic materials. The most striking
distribution and localized character of the 3d Mn moment [24]. manifestation of the PME is in antiferromagnets where the
The piezomagnetic [7] and flexomagnetic effect [25] were zero spontaneous magnetization acquires a finite value upon
simulated in Mn3 GaN by the same group. The strain-induced application of strain. The first AFM systems where PME was
net magnetic moment predicted for Mn3 GaN is an order of proposed [29,30] and later observed [31] were transition-metal
magnitude lower than that of Mn3 SnN predicted in this work. difluorides. In Mn-based antiperovskite nitrides PME was
predicted quantitatively in 2008 [7] and it has not been
B. The piezomagnetic effect observed experimentally so far.
The noncollinear magnetic structure of Mn3 AN which
The PME is defined by a linear dependence of the net
hosts the PME and NTE considered in this work is shown
magnetization on elastic stress tensor components, in contrast
in Fig. 1. (The direction of canting of the Mn local moments is
to the magnetoelastic effect where the dependence on stress is
specific for Mn3 GaN.) The ground state presented in Fig. 1(a)
quadratic. Both effects can be described phenomenologically
is the fully compensated AFM structure with symmetry
by adding appropriate stress-dependent terms to the free
corresponding to the  5g representation [32]. (The magnetic
energy:
unit cell belongs to the trigonal space group P 31m and has
F (T ,H,σ ) = F0 (T ,H) − λi,j k Hi σj k − μi,j k Hi σj2k , (1) the same size as the cubic paramagnetic unit cell belonging
to space group P m3m.) The exchange coupling between
where λi,j k is an axial time-antisymmetric tensor representing the neighboring Mn atoms is antiferromagnetic, which leads
the PME, Hi are components of the magnetic field, σj k to the frustration in the kagome-type lattice in the (111)
is the elastic stress tensor, and μi,j k is the magnetoelastic plane (highlighted as orange online). The three Mn local
tensor. Nonvanishing elements of λi,j k correspond to terms of magnetic moments (LMMs) are of the same size and have an
Eq. (1) which are invariant under operations from the magnetic angle of 2π/3 between their directions. Spin-orbit interaction
symmetry group [26]. These elements then contribute to the (SOI) aligns the plane defined by the directions of the three

024451-2
PIEZOMAGNETISM AS A COUNTERPART OF THE . . . PHYSICAL REVIEW B 96, 024451 (2017)

TABLE I. Physical properties of Mn3 AN: Néel temperature,

lattice parameter at 10 K, calculated lattice parameter, spontaneous
volume change, Poisson’s ratio, bulk modulus, size of Mn local
moment in unstrained system. All measured data are taken from
Ref. [8] except a0 and TN for Mn3 SnN, which are from Ref. [9].
Calculated data are markedt .

A TN [K] a0 [Å] a0t [Å] ωs [10−3 ] νt K t [GPa] M0t [μB ]

Rh 226 3.918 3.88 2.07 0.19 148.4 2.84
Pd 316 3.982 3.94 3.60 0.20 140.7 3.15
Ag 276 4.013 3.98 5.79 0.20 118.9 3.08
Co 252 3.867 3.80 5.64 0.13 149.5 2.48
Ni 256 3.886 3.84 8.18 0.15 136.5 2.83
Zn 170 3.890 3.87 20.44 0.13 126.0 2.64
FIG. 2. Total energy as a function of biaxial strain and canted
Ga 288 3.898 3.86 19.10 0.13 129.4 2.43
angle for Mn3 GaN. No interpolation is used in the surface plot. The
In 366 4.000 3.99 9.24 0.18 115.0 2.70
equilibrium angle depends linearly on the strain. The reference energy
Sn 475 4.060 3.97 0.0 0.18 102.0 2.52
corresponds to E(θ1 = 0) for each strain.

LMMs with the (111) lattice plane resulting in the  5g minima in a total energy profile Etot (ε,θ1 ) as shown in Fig. 2.
representation of Fig. 1. A simultaneous rotation of all three The quantitative agreement of these two methods gives us
LMMs by π/2 within the (111) plane results in another confidence that we have found the physically relevant energy
fully compensated AFM structure corresponding to the  4g minimum. All calculations include spin-orbit coupling and
representation where the LMMs all point inside (outside) the confirm that its impact on the PME is negligible in the case of
triangle in a given (adjacent) plane [10]. The energy difference period four and five elements.
between  4g and  5g ordering is also purely due to SOI (3) Finally, we increase the density of k-points and
whereas the noncollinearity and the strong magneto-structural calculate the projected DOS for the converged strained and
coupling is due to the exchange interaction. Note that the unstrained noncollinear structures to identify features in the
exchange interaction origin of the PME distinguishes it from electronic structure that would explain the variation of the
magnetostriction, which is due to SOI [7]. The PME can be PME across the material range. Our results do not confirm a
described as linear exchange-striction as the strength of the proximity of the Fermi energy to a sharp peak in the DOS as
exchange interactions driving the PME is of the order of suggested by an earlier tight-binding study [20].
10 meV per unit cell whereas the spin-orbit coupling falls Figures 1(c) and 1(d) represent a qualitative overview
between 0.1 and 1 meV per unit cell according to our ab initio of the simulated response of the magnetic structure to the
simulations. compressive and tensile strain, respectively. A comparison
with the ground state in Fig. 1(a) shows that Mn magnetic
moments cant and change size, which are two independent
II. RESULTS
contributions to the PME. This behavior is due to the strain-
We calculate the total energy, magnetic moments, and induced reduction of symmetry from P 31m to the P m m m
projected density of states (DOS) for the noncollinear magnetic orthorhombic magnetic space group [7] and from P m3m to
structure of biaxially strained Mn3 AN (A = Rh, Pd, Ag, Co, the P 4/mmm tetragonal space group in the paramagnetic case
Ni, Zn, Ga, In, Sn) from first principles. Our computational [13] [the system is no longer invariant under the third-order
procedure is the following: rotation about the (111) axis].
(1) We find the equilibrium lattice parameter a0 , bulk For more clarity, Fig. 1(b) shows the tensile strained
modulus K, and Poisson’s ratio ν for each material with fixed (ε = a/a0 > 0) magnetic order in the (111) plane. The
AFM order by fitting the total energies obtained for a range of canted angles θi within the (111) plane and LMM magnitudes
lattice parameters (a,c/a) to the Birch–Murnaghan equation Mi on the three Mn sites are introduced. The moments in the
of state [33]. We also allowed for relaxation of individual (100) and (010) planes cant in opposite directions, θ1 = −θ2 ,
atomic positions but we found no bond buckling in agreement to become more parallel (antiparallel) in case of positive
with an earlier ab initio study [7]. The results are summarized (negative) θ1 . The moment in the (001) plane does not change
in Table I. direction.
(2) We relax the magnetic moments with a fixed lattice for The change of moment size Mi = Mi − M0 is strongly
a range of biaxial strains to evaluate the PME. We perform dependent on the c/a ratio of the tetragonal lattice. (M0 is
two independent sets of calculations with the vertical lattice the LMM size common to all Mn sites in the unstrained
parameter c set (a) to conserve the unstrained unit cell volume - system.) The changes plotted in Fig. 1(b) correspond to unit-
data labeled “V”, and (b) according to the calculated Poisson’s cell volume conservation when M1 = M2 ≈ − M3 /2
ratio - data labeled “P”. The initial AFM local moment for all systems studied. M3 universally increases (decreases)
directions and sizes are either relaxed by using a noncollinear with compressive (tensile) strain. With realistic Poisson’s
implementation of spin-density functional theory (VASP code ratios all three Mn moments increase (decrease) for tensile
[34]) in a self-consistent loop or explicitly by searching for (compressive) strain following the volume change of the unit

024451-3
J. ZEMEN, Z. GERCSI, AND K. G. SANDEMAN PHYSICAL REVIEW B 96, 024451 (2017)

Figure 3(a) compares the PME obtained assuming unit-cell

volume conservation (Poisson’s ratio ν = 0.5) and using our
calculated Poisson ratios ν listed in Table I, which correspond
to smaller vertical distortion for a given strain. The latter is
our lower estimate of the experimentally accessible PME as
our calculated values of ν are lower than expected for metallic
materials. The former version of the PME neglects the elastic
properties of the lattice and represents the response of the
frustrated magnetic system to a lattice symmetry breaking
(normalized tetragonal distortion). As a result, the predicted
Mnet (V) should be regarded as an upper estimate of the
experimentally accessible PME. In both cases Mn3 SnN is
predicted to have Mnet an order of magnitude larger than
Mn3 GaN, the only PME value available in the literature [7].

A. Fitting the piezomagnetic effect

with a Heisenberg model
To interpret the calculated PME in terms of the AFM
pairwise exchange interactions Jij (ε) between the three Mn
FIG. 3. Comparison of the net moment Mnet induced by 1% of atoms in the (111) plane we resort to the classical Heisenberg
tensile strain: (a) Mnet assuming unit-cell volume conservation (V)
J model with variable moment sizes:
and Poisson ratios of Table I (P) and Mnet fit according to Eq. (5).
(b) Comparison of the PME, measured by Mnet (V), to the inverse of E(θ1 ,ε) = −J12 M1 M2 cos(2π/3 − 2θ1 )
the energy separation between p or d states of atom A and d states
of Mn, marked by the dashed line. − 2J13 M1 M3 cos(2π/3 + θ1 ), (3)
where the values of the exchange parameters J13 = J23 =
J12 and the local moments M1 = M2 = M3 introduced in
cell. (Mn3 RhN is the only exception where M3 is almost Fig. 1(b) are restricted by the tetragonal symmetry of the
independent of strain.) Atom A develops a moment two orders lattice. We find the canted angle by minimizing the exchange
of magnitude lower than the Mn local moment for small energy (∂E/∂θ1 = 0) and insert it into the expression for
applied strain, |ε| < 1%, so its role in PME is negligible. the net moment Mnet = 2M1 cos(2π/3 + θ1 ) + M3 . We obtain
The unstrained ground state (plotted  5g ) has no spon- a relationship between PME and changes of the exchange
taneous magnetization but a net moment Mnet aligned with interaction due to strain:

M3 develops upon straining. Our calculations confirm that the J
Mnet J13
canted angle θi , the change of moment size Mi , and conse- ≡ 1− (4)
M3 J12
quently Mnet = 2M1 cos(2π/3 + θ1 ) + M3 depend linearly on
applied strain as required by Eq. (2). The dependence departs J0 − J − (J0 + J ) 2 J
≈ ≈− ,
slightly from linearity for larger strain |ε| > 1%, and our study J0 − J J0
is limited to the interval ε ∈ −2.5,2.5%. A striking feature 2M3 2M3 ∂J12
of the PME is the change of orientation of Mnet when switching
J
Mnet ≡− J = ε, (5)
J0 J0 ∂ε
between tensile and compressive biaxial strain. We note that

J
such control of net moment orientation cannot be achieved where Mnet is an approximation of Mnet based on our minimal
by magnetostriction. [The same description holds also for  4g model of Eq. (3), J0 < 0 is the exchange parameter in the
order but Mnet  M3 is then rotated by π/2 in the (111) plane.] unstrained lattice, and J is the induced change of J12 and
Table I lists all relevant measured properties and results J13 . We have fit our ab initio total energy as a function of the
calculated in this work. Our Mn magnetic moment for canted angle to the Heisenberg model of Eq. (3) to extract J12
Mn3 GaN is in good agreement with a previous theoretical and J13 for each value of strain. In all compounds we have
study [7]. Our Poisson’s ratios do not vary much across observed J12 ≈ J0 − J and J13 ≈ J0 + J , which allows
J
the range of compounds and are slightly smaller than us to define Mnet in Eq. (5) that is directly proportional to the
ν = 0.25–0.3 predicted by an ab initio study of elastic proper- derivative of the exchange parameters Jij with respect to the
ties in Mn3 (Cu,Ge)N [35]. All calculated lattice parameters are biaxial strain ε.
J
1%–2% smaller than the values measured at low temperatures. Figure 3(a) shows that Mnet is in good agreement with Mnet
Figure 3 presents our results on the PME and the related (V) extracted directly from our calculated LMMs (without
features of electronic band structure. The net moment Mnet any fitting). The small differences are due to deviations of
plotted for the nine Mn-based antiperovskite systems subject the magnetic system from Heisenberg behavior (e.g., LMMs
to tensile strain ε = 1% is a natural measure of the PME. A change size as they cant even in an unstrained lattice)
positive (negative) value of Mnet corresponds to a net moment and deviations from linearity assumed in Eq. (4). The key
induced parallel (antiparallel) to M3 irrespective of belonging conclusion based on Fig. 3(a) in combination with Eq. (4) is
to the  4g or  5g representation. that a large PME corresponds to a large difference between J12

024451-4
PIEZOMAGNETISM AS A COUNTERPART OF THE . . . PHYSICAL REVIEW B 96, 024451 (2017)

(the bond in the plane of the biaxial strain) and J13 (the bond
with a component perpendicular to this plane.)
We note that including exchange interactions to second-
nearest neighbors in Eq. (3) would make sense only if the
magnetic unit cell used in the ab initio calculation was
increased beyond the chemical unit cell. However, so far
there has not been sufficient experimental support for such
an extension of our model.

B. Linking the piezomagnetic effect

to band structure
Figure 3(b) relates the total induced moment Mnet (V)
to the mean band energy of the valence p or d states of
atom A. This quantity is often called the band center and
we extract it from our projected DOS, ρAp,d (E), as follows:
μAp,d = 1/ EρAp,d (E)dE, where = ρAp,d (E)dE is a
normalization. We consider only the d band [ρAd (E)] when
atom A is a transition metal and only the p band [ρAp (E)] for
the rest. The wide s band does not seem to play an important FIG. 4. Calculated PME characterized by |χP−1 | as a function of
role in the PME. The right vertical axis of Fig. 3(b) measures the measured MVE weighted by the bulk modulus. Triangles indicate
the inverse of μAp,d with respect to the Mn d-band center, systems with confirmed triangular magnetic ground state, red and
μMnd , which is very close to the Fermi energy. green symbols indicate a positive and negative canted angle at tensile
Based on the remarkable match between |Mnet | and (μAp,d − strain, respectively, and the blue lines are least square linear fits.
μMnd )−1 we conclude that piezomagnetism in Mn-based (a) Two different trends for atom A from period four and five;
antiperovskite nitrides is governed by the mutual configuration (b) |χP−1 | weighted by a strain-induced shift of mean band energy
of Mn d states and p or d states of atom A. More specifically, of two Mn atoms: one trend for all systems with confirmed triangular
a greater proximity (a potential for hybridization) of the magnetism.
valence band of atom A to the spin-polarized d band of Mn
increases the difference between J12 and J13 per unit strain,
which manifests itself as a larger induced net moment. On to the mean valence-band energy of atom A (with respect to
the other hand, when the triangular magnetic order of Mn the Mn d-band center μMnd ) in the unstrained system.
moments is undisturbed by hybridization with p or d states Figure 4 compares the measured magnetovolume effect
of atom A then J12 ≈ J13 and only a small net moment is [8] to our calculated piezomagnetic susceptibility χP . The
induced. The best example is Mn3 ZnN where the narrow, fully MVE is a spontaneous change of volume due to a change of
filled d band is about 7 eV below the Fermi energy and the magnetic ordering (typically the size of magnetic moment).
induced net moment is negligible. This trend is analogous to It was first observed in Ni-Fe Invar below its TC [37].
a scaling of the Néel temperature with the number of valence Takenaka et al. measure a spontaneous volume increase upon
electrons of atom A, which was detected in 1977 in the same the transition from the PM to AFM state and subtract the
class of materials [11]. We further investigate the dependence phononic contribution so that their MVE data are purely of
of the sign of Mnet on the presence of valence d states of magnetic origin [8]. They investigate a wide range of Mn-based
atom A in a separate paper [36]. In the following paragraphs, antiperovskite nitrides and conclude that the MVE is a property
we compare our simulated PME to the spontaneous MVE, of the frustrated triangular AFM state, which is strongly
which is another measure of the magnetic frustration and for dependent on the number of valence electrons. The MVE is
which experimental data are available for all nine Mn3 AN the largest when there are two s electrons and one or no p
compounds [8]. electrons (A = Zn, Ga). When the number of valence s and
p electrons changes then the system transforms to a different
crystal and magnetic structure with no MVE (A = Cu, Ge, As,
C. Comparing the piezomagnetic effect Sn, Sb).
to the magnetovolume effect In addition, Takenaka et al. have observed an increase in
To draw an analogy between the strain and an external MVE as the d band of atom A moves away from EF . This
field H that can induce magnetization, we introduce a general trend reminds us of the scaling of susceptibility χP
piezomagnetic susceptibility: with the mean band energy of atom A, μAp,d , described above.
We include Fig. 4(a) to check if the dependence on μAp,d
J
Mnet 2 ∂J12   furnishes a clear link between the PME and MVE. The figure
= ε ≡ χP μAp,d ε, (6) shows that our piezomagnetic susceptibility χP is inversely
M3 J0 ∂ε
proportional to the measured volume change, as expected. In
where the change of applied strain ε replaces H and other words, a large MVE implies a small PME and vice versa.
J
Mnet ( ε) was introduced in Eq. (5). Based on Fig. 3(b) we can Atoms A belonging to periods four and five of the periodic
say that the susceptibility χP (μAp,d ) is inversely proportional table have different coefficients of proportionality. This implies

024451-5
J. ZEMEN, Z. GERCSI, AND K. G. SANDEMAN PHYSICAL REVIEW B 96, 024451 (2017)

that not only the position of the A band with respect to EF x-ray diffraction [8], which has not indicated a tetragonal
but also the size of atom A plays a role in weakening the distortion in any compound studied in this work. Nevertheless,
triangular AFM structure. Such a difference between periods χP reflects how much magnetic stress could be relieved by a
four and five was first also seen by Fruchart and Bertaut in the tetragonal distortion and this quantity is inversely proportional
case of the scaling of TN with the number of valence electrons to ωs K, as shown by Fig. 4. We plot χP vs ωs K rather than
of atom A [11]. ωs to compare only quantities related to magnetism and factor
The agreement of a calculated zero-temperature suscep- out the system-dependent elastic properties.
tibility (χP−1 ) with a spontaneous volume change ωs at the It should be noted that the sign of χP indicates which type
PM-AFM phase transition (weighted by K) is remarkable and of tetragonal distortion is energetically more favorable. A brief
requires further analysis. Magnetovolume effects in itinerant demonstration of this neglects the dependence of Mi and θi on
electron magnets were first analyzed by the Stoner–Edwards– strain in Eq. (3). Then we can find a spontaneous biaxial strain
Wohlfarth theory [38]. The free energy can be approximated εs (analogous to volume strain ωs ) from the balance of elastic
by F (T ,M,ω) = F0 (T ,ω) + 12 KV ω2 + 12 a(T ,ω)M 2 and min- and magnetic energy: εs = 2C 1
∂J12 /∂εM02 = − 4C 1
χP |J0 |M02 ,
imized with respect to the volume strain ω = V /V to obtain where C > 0 is an effective elastic modulus. Immediately, we
KV ω = cmv M 2 where cmv = − 12 ∂a(T ,ω)/∂ω is the magne- can see that all systems in this study with χP > 0 tend to a
tovolume coefficient, M is the spontaneous magnetization, distortion with εs < 0 (c/a > 1) and vice versa.
K is the bulk modulus, and V is the reference volume. We conclude that a system with robust triangular magnetic
After considering the spin fluctuations at the first-order phase order undisturbed by the proximity of electronic states of atom
transition, the above formula becomes KV ω = cmv (M 2 − ξ 2 ), A (large μAp,d ) tends to relieve its magnetic stress via a volume
where ξ is the amplitude of spin fluctuations [39,40]. change, whereas a system more influenced by atom A but with
In the case of Mn-based antiperovskites the local moments persisting triangular order (small μAp,d ) prefers to relieve its
are relatively well localized [24], so we can approximate the magnetic stress via a tetragonal distortion should the elastic-
magnetic energy of the triangular AFM system on a cubic energy cost allow it. (If the tetragonal distortion is enforced
lattice by Eq. (3) with zero canted angle: E(θ1 = 0) = 32 J0 M02 . externally, then the system develops a large net magnetization.)
The balance of elastic and magnetic energy then leads to an The slight deviations of |χP−1 | from ωs K seen in Fig. 4
expression for the spontaneous volume strain ( V /V ): may originate in (a) spin fluctuations which we neglected in
Eq. (7), the small size of the deviations suggests that the spin-
3M02 ∂J0 ∂J0
ωs K = − ∼ ≡ tv, (7) fluctuation contribution to the MVE [KV ω = cmv (M 2 − ξ 2 )]
2V ∂ω ∂ω is significantly suppressed by the strong frustration; (b) limited
where we neglect the change of local moments M0 with numerical accuracy, e.g., Mn3 ZnN is most affected because it
changing volume, ( M0 )2 , as a higher-order contribution. The has very small χP and its large relative error is amplified
magnetic stress per Mn-Mn bond t v is introduced following by the inversion; (c) nitrogen deficiency (8%–16%) varying
the work of Filippetti and Hill [41]. The magnetic stress at the across the range of samples where MVE was measured [8],
5g
phase transition can then be expressed as T v = ∂E  /∂ω = e.g., magnetic order in Mn3 SnN is known to be sensitive
3 5g
t M 2 , where E  is again the magnetic energy E(θ1 = 0). to N concentration [9]; and (d) a material-specific elastic
2 v 0
After establishing the link between MVE and the magnetic property that was not factored out of the plotted quantities,
volume stress T v , we attempt the same for the PME and the e.g., the use of bulk modulus K = 130 GPa for all compounds
magnetic biaxial stress: T b = ∂E(θ1 )/∂ε ∼ t b , where E(θ1 ) when subtracting the phononic contribution to the MVE [8]
is a magnetic energy of the canted AFM structures and the (consequently, in the plot we use K = 130 GPa instead of our
magnetic stress per Mn-Mn bond t b is proportional to the calculated K of Table I).
susceptibility χP of Eq. (6): To further explore the inverse proportionality between PME
and MVE with respect to features of the electronic structure
2 ∂J12 ∂J12 we analyze the strain dependence of the mean band energy of
χP = ∼ ≡ t b. (8)
J0 ∂ε ∂ε Mn states. We extract the mutual shift of mean band energy
Finally, based on the comparison of Eqs. (7) and (8) we can of Mn1d states [site in (100) plane of the unit cell] and Mn3d
conclude that both ωs K and χP are proportional to derivatives [site in (001) plane] from the projected DOS ρMn1d (E,ε,θ1 )
of the exchange parameters with respect to strain and thereby and ρMn3d (E,ε,θ1 ) of the strained system before canting
to the magnetic stress of the triangular AFM system. Hence the (ε = 1%, θ1 = 0) in analogy to evaluation of μAp,d shown in
linear relationship of Fig. 4(a) indicates a trade-off between Fig. 3(b). The obtained quantity |μ1 − μ3 | directly measures
two complementary stress relief mechanisms. the response of the spin-polarized electronic structure to the
tetragonal distortion. Such information is missing in μAp,d of
the unstrained structure.
III. DISCUSSION
Figure 4(b) shows |χP−1 | weighted by the mutual band
In principle, the stress arising at the onset of AFM ordering shift |μ1 − μ3 | as a function of ωs K. Compounds with atom
at TN can be relieved by a volume change or a lattice A from periods four and five now follow the same linear
distortion. However, our calculations and subsequent fitting trend with the exception of A = Ag, Co, Rh. Our hypothesis
to the Heisenberg model find that the magnetic energy saved based on Fig. 4 is that the factor |μ1 − μ3 | incorporates
by a tetragonal distortion (linear in ε) becomes smaller than the dependence of the PME on the size of atom A for
the elastic energy cost (quadratic in ε around unstrained systems with stable triangular AFM ordering. Mn3 AgN and
lattice) for negligibly small distortions. This is confirmed by Mn3 RhN do not have a triangular AFM ground state, which

024451-6
PIEZOMAGNETISM AS A COUNTERPART OF THE . . . PHYSICAL REVIEW B 96, 024451 (2017)

has an explanation in their band-structure properties and Mn-based antiperovskites and enable development of applica-
becomes apparent in Fig. 4(b). Extending the same argument tions including data storage, memory, and solid-state cooling.
to those compounds with unknown magnetic order, we expect
Mn3 PdN (Mn3 CoN) to have a triangular (other) AFM ground IV. METHODS
state.
The linear scaling of the spontaneous MVE with |χP−1 | im- All our calculations employ the projector augmented-wave
plies a significant suppression of spin fluctuations by the strong (PAW) method implemented in VASP code [34] within the
frustration in these systems. At the same time it can be used Perdew–Burke–Ernzerhof (PBE) generalized gradient approx-
as a tool in theory-led design of nonstoichiometric materials imation [42]. The relaxation of fully unconstrained non-
with a large MVE and, consequently, barocaloric effect where collinear magnetic structures has been implemented by Hobbs
the entropy change is proportional to the spontaneous volume [43]. The relativistic effects are accounted for by a Hamiltonian
change according to the Clausius–Clapeyron relation: containing the mass-velocity and Darwin corrections and the
spin-orbit operator L · S in a basis of total-angular-momentum
 −1 eigenstates as implemented by Kresse and Lebacq [44]. The
dTt
S(Tt ,p) − S(Tt ,0) = V ωs . (9) off-diagonal elements of the 2 × 2 matrix in spin space
dp
originate in spin-orbit coupling and also in the exchange-
Modelling the pressure dependence of the transition tempera- correlation potential in the case of noncollinear magnetization
ture dTt /dp goes beyond the capability of density functional density.
theory at zero temperature and is the subject of our ongoing We use a 12 × 12 × 12 k-point sampling in the self-
work [36]. consistent cycle and 17 × 17 × 17 k-point sampling to obtain
In terms of practical devices hosting the composite mag- the site and orbital-resolved DOS. The cutoff energy is 400 eV.
netoelectric effect we envisage a piezomagnetic Mn-based The local magnetic moments are evaluated in atomic spheres
antiperovskite layer grown on a piezoelectric substrate. Such with the default Wigner Seitz radius because they are not very
a bilayer can benefit from strong elastic coupling due to the sensitive to the projection sphere radius [7].
common perovskite structure with small lattice mismatch at We constrain the Mn local moment directions by using
the interface. In case of a multidomain state in the as-grown an additional penalty energy as implemented in the VASP
AFM layer one can apply biaxial strain that induces a net code in order to obtain the DOS projected on Mn d states,
moment along an arbitrary direction in each domain. Then e.g., ρMn3d (E,ε,θ1 ) of the strained system. We add a further
an external magnetic field (coupling to the induced moment constraint to suppress the small moment on atom A which
in each domain) can align all domains and increase the develops due to strain to allow for extraction of J12 and J13
net magnetization. This magnetization will persist when the from the total energy as a function of strain and canted angle
external field is removed. After removing the biaxial strain by fitting to the Heisenberg model of Eq. (3).
the net magnetization will vanish but the antiferromagnetic
ACKNOWLEDGMENTS
domains will remain ordered due to the magnetocrystalline
anisotropy of the order of 0.1–1 meV per unit cell. We would like to thank Kirill Belashchenko, Lesley Cohen,
We hope that the successful comparison of our predicted Julie Staunton, and Eduardo Mendive-Tapia for productive
PME to the measured MVE and the coherent interpretation discussions. The research leading to these results has re-
of the PME based on features of the electronic struc- ceived funding from the European Union Seventh Framework
ture will provide guidance for further investigations of the Programme (FP7/2007- 2013) under Grant Agreement No.
unique physical properties of the frustrated AFM structure of 310748 (DRREAM).

[1] F. Matsukura, Y. Tokura, and H. Ohno, Nat. Nanotechnol. 10, [9] Ph. L’Héritier, D. Fruchart, R. Madar, and R. Fruchart, 1.5.6.2
209 (2015). Crystallographic properties of Mc XMf3 compounds, in Alloys
[2] T. Jungwirth, X. Marti, P. Wadley, and J. Wunderlich, Nat. and Compounds of d-Elements with Main Group Elements. Part
Nanotechnol. 11, 231 (2016). 2, edited by H. P. J. Wijn, Landolt-Börnstein - Group III Con-
[3] P. Wadley, B. Howells, J. Železný, C. Andrews, V. Hills, R. P. densed Matter Vol. 19C (Springer-Verlag, Berlin, Heidelberg,
Campion, V. Novak, K. Olejník, F. Maccherozzi, S. Dhesi 1988).
et al., Science 351, 587 (2016). [10] D. Fruchart, E. Bertaut, R. Madar, G. Lorthioir, and R. Fruchart,
[4] J. Železný, H. Gao, K. Výborný, J. Zemen, J. Mašek, A. Solid State Commun. 9, 1793 (1971).
Manchon, J. Wunderlich, J. Sinova, and T. Jungwirth, Phys. [11] D. Fruchart and F. Bertaut, J. Phys. Soc. Jpn. 44, 781 (1978).
Rev. Lett. 113, 157201 (2014). [12] K. Takenaka and H. Takagi, Appl. Phys. Lett. 87, 261902
[5] J. Zelezny, Y. Zhang, C. Felser, and B. Yan, arXiv:1702.00295. (2005).
[6] E. Gomonaj, Phase Transit. 18, 93 (1989). [13] M. Wu, C. Wang, Y. Sun, L. Chu, J. Yan, D. Chen, Q. Huang,
[7] P. Lukashev, R. F. Sabirianov, and K. Belashchenko, Phys. Rev. and J. W. Lynn, J. Appl. Phys. 114, 123902 (2013).
B 78, 184414 (2008). [14] S. Deng, Y. Sun, H. Wu, Q. Huang, J. Yan, K. Shi, M. I. Malik,
[8] K. Takenaka, M. Ichigo, T. Hamada, A. Ozawa, T. Shibayama, H. Lu, L. Wang, R. Huang et al., Chem. Mater. 27, 2495 (2015).
T. Inagaki, and K. Asano, Sci. Tech. Adv. Mater. 15, 015009 [15] S. Deng, Y. Sun, L. Wang, Z. Shi, H. Wu, Q. Huang, J. Yan, K.
(2014). Shi, P. Hu, A. Zaoui et al., J. Phys. Chem. C 119, 24983 (2015).

024451-7
J. ZEMEN, Z. GERCSI, AND K. G. SANDEMAN PHYSICAL REVIEW B 96, 024451 (2017)

[16] D. Matsunami, A. Fujita, K. Takenaka, and M. Kano, Nat. Mater. [30] T. Moriya, J. Phys. Chem. Solids 11, 73 (1959).
14, 73 (2015). [31] A. Borovik-Romanov, J. Exp. Theor. Phys. (USSR) 36, 1954
[17] K. Shi, Y. Sun, J. Yan, S. Deng, L. Wang, H. Wu, P. Hu, H. Lu, (1959); 38, 1088 (1960).
M. I. Malik, Q. Huang et al., Adv. Mater. 28, 3761 (2016). [32] E. Bertaut, D. Fruchart, J. Bouchaud, and R. Fruchart, Solid
[18] H. Sakakibara, H. Ando, Y. Kuroki, S. Kawai, K. Ueda, and H. State Commun. 6, 251 (1968).
Asano, J. Appl. Phys. 117, 17D725 (2015). [33] F. Birch, Phys. Rev. 71, 809 (1947).
[19] H. Tashiro, R. Suzuki, T. Miyawaki, K. Ueda, and H. Asano, [34] G. Kresse and D. Joubert, Phys. Rev. B 59, 1758 (1999).
J. Korean Phys. Soc. 63, 299 (2013). [35] B. Qu, H. He, and B. Pan, AIP Adv. 1, 042125 (2011).
[20] J.-P. Jardin and J. Labbé, J. Appl. Phys. 52, 1627 (1981). [36] J. Zemen, E. Mendive-Tapia, Z. Gercsi, R. Banerjee, J. B.
[21] E. Gomonaj and V. L’vov, Phase Transit. 38, 15 (1992). Staunton, and K. G. Sandeman, Phys. Rev. B 95, 184438 (2017).
[22] B. Y. Qu and B. C. Pan, J. Appl. Phys. 108, 113920 (2010). [37] M. Hayase, M. Shiga, and Y. Nakamura, J. Phys. Soc. Jpn. 34,
[23] B. Qu, H. He, and B. Pan, Adv. Condens. Matter Phys. 2012, 7 925 (1973).
(2012). [38] E. Wohlfarth, Physica B + C (Amsterdam) 91, 305 (1977).
[24] P. Lukashev and R. F. Sabirianov, J. Appl. Phys. 107, 09E115 [39] T. Moriya and K. Usami, Solid State Commun. 34, 95 (1980).
(2010). [40] Y. Takahashi and H. Nakano, J. Phys.: Condens. Matter 18, 521
[25] P. Lukashev and R. F. Sabirianov, Phys. Rev. B 82, 094417 (2006).
(2010). [41] A. Filippetti and N. A. Hill, Phys. Rev. Lett. 85, 5166 (2000).
[26] A. Borovik-Romanov, Ferroelectrics 162, 153 (1994). [42] J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77,
[27] W. Voigt, Lehrbuch der Kristallphysik (Leipzig, 1928). 3865 (1996).
[28] B. A. Tavger and V. M. Zaitzev, J. Exp. Theor. Phys. (USSR) 3, [43] D. Hobbs, G. Kresse, and J. Hafner, Phys. Rev. B 62, 11556
430 (1956). (2000).
[29] I. E. Dzyaloshinskii, J. Exp. Theor. Phys. (USSR) 33, 807 [44] P. Błoński, S. Dennler, and J. Hafner, J. Chem. Phys. 134,
(1957). 034107 (2011).

024451-8