Journal of Magnetism and Magnetic Materials 324 (2012) 3646–3651

Contents lists available at SciVerse ScienceDirect

Journal of Magnetism and Magnetic Materials

journal homepage: www.elsevier.com/locate/jmmm

Coupled structural and magnetic properties of ferric fluoride nanostructures:

Part II, a Monte Carlo–Heisenberg study
Bernard Fongang a,b, Yvan Labaye a, Florent Calvayrac a,n, Serge Zekeng b, Jean-Marc Grene che a
a
Département PEC, Institut des Molécules et Matériaux du Mans, UMR CNRS 6283, LUNAM Université du Maine, 72085 Le Mans, Cedex 9, France
b
Laboratoire de Sciences des Matériaux, Département de Physique, Université de Yaoundé i, BP 812 Yaoundé, Cameroun

a r t i c l e i n f o a b s t r a c t

Article history: We present a numerical study of the magnetic structure of nanostructured iron fluoride, using the
Received 27 January 2010 Monte Carlo Metropolis simulated annealing technique and a classical Heisenberg Hamiltonian with
Received in revised form superexchange angle dependent interactions. The parameters are adjusted on experimental results, and
20 April 2012
the atomic structure and topology taken from a previous atomistic model of grain boundaries in the
Available online 6 June 2012
same system. We find perfect antiferromagnetic crystalline grains and a disordered magnetic
Keywords: configuration (speromagnetic) at the grain boundary, in agreement with experimental features. Both
Magnetic nanostructure the lowest magnetic energy and the rate of magnetic frustration are found to be dependent on the
Iron fluoride relative disorientation of crystalline grains, i.e. on the cationic topology. We conclude on possible
Grain boundary
extensions of the model.
Metropolis
& 2012 Elsevier B.V. All rights reserved.
Magnetic frustration
Heisenberg

1. Introduction To investigate the magnetic behavior of nanostructures and

During the last decade, strong efforts have been devoted
studying the magnetic properties of nanostructures. This growing
interest is due to their unusual and tunable physical properties,
which are strongly influenced by the effect of confinement
resulting in a large contribution of surface and interface effects [1].
Understanding the correlation between magnetic properties and
nanostructures involves collaborative efforts between chemists,
physicists and materials scientists to study both fundamental
properties and potential applications [2]. This can be carried out
experimentally using diffraction techniques, local probe techni-
ques such as Mössbauer spectrometry, nuclear magnetic reso-
nance, electron microscopies and magnetic measurements. In
addition, one can use numerical modeling techniques based on
ab initio or phenomenological calculations such as the classical
Monte Carlo methods with the Heisenberg model, as in [3–6] for
instance. The key issue for understanding magnetic macroscopic
properties, such as magnetization or susceptibility, would be to
investigate the contributions arising from the nanograins and the
role played by the nanograin surface and the interface between
grains generally named grain boundary (GB). GBs in magnetic
nanostructures can feature a disordered atomic structure and a
spin-glass-like or speromagnetic behavior and have a chemical
composition which might differ strongly from that of the corre-
sponding nanograin [7–9].
the mutual influence of the grains and grain boundaries, we
selected the ferric fluoride system (FeF3 ), because this ionic phase
displays several advantages. Indeed, it exhibits a fascinating
polymorphism with three different crystalline phases and two
amorphous varieties, the structures of which result from ordered
and disordered packing of corner-sharing octahedral units,
respectively. Those rather different cationic topologies originate
collinear and non-collinear magnetic arrangements in conjunc-
tion with the antiferromagnetic nature of the superexchange
interactions. Iron fluoride is therefore considered as an excellent
simple model system to illustrate the concept of topological
magnetic frustration. The more stable crystalline form of FeF3 ,
which is the rhombohedral phase (rFeF3 ; SG R3C) is antiferro-
magnetically ordered below T N ¼ 363 K while the amorphous
varieties display a speromagnetic behavior below T F 30 40 K
[10–12]. The presence of fluorine in those materials establishes
superexchange magnetic interactions between magnetic ions. The
strength of the coupling is linearly dependent on cos2 ðyÞ, where
y denotes the superexchange angle FeFFe while the sign is
given by the orbital configuration of the F2Fe bond: for example,
1801 (Fe3 þ F2Fe3 þ is characterized by a strong antiferromag-
netic coupling according to Kanamori–Goodenough rules [13,14].
The magnetic properties of FeF3 have been intensively studied
during the past two decades and can be described in terms of iron
ring statistics with induced magnetic frustration [7,15]:

 the rhombohedral phase r-FeF3 , which is a pseudo-

n
Corresponding author. Tel.: þ33 243832627; fax: þ33 243833518. cubic packing of weakly tilted octahedral FeF6 units, is not
E-mail address: Florent.Calvayrac@univ-lemans.fr (F. Calvayrac). frustrated;

0304-8853/$ - see front matter & 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.jmmm.2012.04.040
 B. Fongang et al. / Journal of Magnetism and Magnetic Materials 324 (2012) 3646–3651
 the hexagonal tungsten bronze HTB-FeF3 which results from
the superposition of magnetically frustrated planes composed
of hexagonal and triangular rings based on octahedral units;
 and the pyrochlore pyr-FeF3 phase which consists of a packing
of octahedral units building corner-sharing tetrahedra, and is
thus more frustrated than HTB-FeF3 .
The amorphous to crystalline transition in iron fluoride has
also been studied using EELS spectroscopy [16]. The glass struc-
ture has been investigated using 57Fe Mössbauer spectrometry
[17]. Besides, iron fluoride has recently found applications as a
cathode for Li-ion batteries [18], for instance deposited as nano-
flowers on carbon nanotube branches [19]. Another promising
application of the material is to be found in hydrogen storage
[20], or as a precursor to bcc iron nanoparticles [21].
Recent experimental and theoretical studies on the properties
of FeF3 nanostructures obtained by mechanical milling have
shown that, at the nanoscale, this material is composed of two
parts: the grain which behaves as the crystalline rFeF3 phase
and a disordered grain boundary which remains composed of
corner-sharing octahedral units, respectively [22,23]. In terms of
ring statistics, it has been numerically established [23] that all
odd rings are localized in the grain boundary of nanostructures,
confirming the speromagnetic structure evidenced from in-field
Mössbauer spectrometry. The Monte Carlo–Metropolis method
has been successfully applied to study the surface and finite-size
effects in nanostructures [24–26]. Nevertheless, no numerical
studies by Monte Carlo simulation on the magnetic properties
of ionic nanostructures which focus on the role of grain bound-
aries have yet been reported. Unlike micromagnetic or molecular
field calculations, the Monte Carlo simulation can take into
account the atomic structure of the lattice and the nature of
superexchange interactions for ionic materials such as FeF3 . The
Monte Carlo–Metropolis simulated annealing technique is an
effective approach in the study of a system with many degrees
of freedom. During such a simulation, random numbers are used
to sample conformations of the system with correct thermo-
dynamical probabilities. A typical Monte Carlo simulation consists
of two steps: thermalization and sampling. During thermaliza-
tion, the system is slowly led to its thermodynamical equilibrium.
After the system reaches this equilibrium, properties of interest
can be estimated by averaging over enough samples [27,28].
Besides, in the simulated annealing method, it is possible to find
heuristically the lowest energy state of the system. We used, in
this work, the standard Metropolis algorithm to simulate the
magnetic behavior of iron fluoride nanostructure with an empha-
sis on the role of grain boundaries.
2. Magnetic model
Our sample model is chosen as a double grain boundary
constrained in between grains which consists of pure perfectly
crystalline FeF3 with a size ranging from 8 to 12 nm as experi-
mentally measured [22] in iron fluoride prepared by high-energy
ball milling. Those samples are obtained by Voronoı̈ tessellation
and structurally relaxed with a scheme based on a modified
Metropolis algorithm [23]. It has been shown both experimentally
and numerically in such a system that the grain boundaries are
structurally disordered, but still remain composed of corner-
sharing octahedral units with odd and even iron rings, thus
leading to magnetic frustration. The macroscopic thermodynamic
properties, such as the temperature dependence of the magneti-
zation, the specific heat and the magnetic susceptibility for our
system, are obtained from a Heisenberg-type Hamiltonian, which
in general contains several terms corresponding to different
3647
energy contributions: exchange, Zeeman, dipolar, anisotropy,
magnetostatic, magnetoelastic and thermal energy. In our study,
we only consider the exchange energy which is the main
contribution to the Hamiltonian. The Zeeman contribution will
be added in the future for the study of hysteresis loops. The
Hamiltonian of the system is thus
1X !!
H¼ J S S ð1Þ
2 /i,jS ij i j
/UUUUS denotes nearest neighbors interaction, Jij is the exchange
! !
coupling constant, Si and Sj are spins corresponding to i and j
sites. The first step of our work consisted in the determination of
the coupling constant Jij (named J for simplicity) which is related
to the superexchange angle y by the relation
Jy ¼ J 1801 cos2 y þ J901 sin2 y ð2Þ
which can be written in the form
Jy ¼ J 901 þ ðJ1801 J 901 Þcos2 y ð3Þ
where J901 and J 1801 are the coupling constant corresponding to the
superexchange angle of 901 and 1801, respectively. From previous
studies, several pieces of information are available for r-FeF3 :
 Néel temperature, T N , is approximatively 363 K [15];
 The blank angle, yB , (angle for which the constant coupling is
zero) is 1151 [29];
 The superexchange angle, yR , in this phase is 153.151 [30].
Taking into account the fact that each Fe atom has six F neighbors
and that octahedral units are regular in pure perfectly crystalline
r-FeF3 , we can use the mean-field theory [31] relation giving the
Néel temperature as a function of the coupling constant to
determine the value of the coupling constant for the rhombohe-
dral phase, J yR , in units of the Boltzmann constant kB
Z9J yR 9S2
T N ¼ 1:45 ð4Þ
6kB
where Z is the coordination number and Sthe spin. With Z ¼6 and
S¼1 JyR ¼ 250:35 K By combining relation (3) for the rhombo-
hedral superexchange angle (yR ) and the blank angle (yB ), we
obtain the system
(
J yR ¼ J901 ð1cos2 yR Þ þJ 1801 cos2 yR
ð5Þ
0 ¼ J 901 ð1cos2 yB Þ þ J 1801 cos2 yB
The resolution of this system leads to
J yR cos2 yB
J901 ¼ ð6Þ
cos2 yB cos2 yR
and
JyR ðcos2 yB 1Þ
J1801 ¼ ð7Þ
cos2 yB cos2 yR
which gives J901 ¼ 72:42 K and J 1801 ¼ 333:07 K. We can then
deduce the relation giving the coupling constant as a function of
the superexchange angle, as follows:
Jy ¼ 72:42405:49 cos2 y ð8Þ
This relation is represented in Fig. 1(a). A comparison with the
distribution of the superexchange angle in nanostructured FeF3
obtained from previous studies [23] implies that all interactions in
the GB remain antiferromagnetic. This is observed in Fig. 1(b) by the
large peak centered around 1601, all angles being greater than 1401.
It is thus concluded that the frustration in the GB does only originate
from the cationic topology.
In our simulations we only considered the exchange contribution
to the magnetic Hamiltonian of the system. We have then neglected
3648 B. Fongang et al. / Journal of Magnetism and Magnetic Materials 324 (2012) 3646–3651
Fig. 1. Coupling constant as a function of the superexchange angle for iron
fluoride from Eq. (8). (b) Angular distribution function for the superexchange
angle [23]. GB is characterized by a large peak around 1601 and according to (a),
should be antiferromagnetic.
other phenomenological contributions such as dipolar, Zeeman or
anisotropy terms, in order to clarify the influence of the exchange
coupling on the magnetic behavior. It is to note that, as opposed to
iron difluoride for instance, magnetic anisotropy is almost zero in
iron trifluoride due to the orbital properties of the Fe-III ion [32].
All samples were first structurally relaxed, with sizes ranging
from 8 to 12 nm and different orientations of grains (tilted, non-
tilted and twisted as in [23]). A simple definition of those terms is
given as follows:
 in a non-tilted sample, crystallographic orientation is such that
all Euler angles of the two crystal grains are equal to zero;
 a tilted sample is oriented such that at least one Euler angle for
one grain is different to zero with all Euler angles of other
equal to zero;
 a twisted sample is the one for which at least one of the Euler
angles is different to zero for both crystal grains.
The energy given in Eq. (1) is minimized by means of the Monte
Carlo/Metropolis simulated annealing procedure. During the simula-
tion, Monte Carlo steps are applied on individual magnetic moments
(‘‘spins’’) with a random walk while solid angles are uniformly
distributed over 4p. Starting with a random spin configuration at a
temperature T start much higher than T N (600 K), the energy is
minimized using the simulated annealing scheme with a decreasing
power law for temperature T ¼ T start n 0:97s where s is the step
number, while the thermodynamic quantities, such as magnetization,
susceptibility and specific heat, can be derived as a function of the
temperature. In each simulation, the final temperature is lower than
1 K and the number of Monte Carlo steps per spin (MCS) is 106.
3. Results
We first determine the energy, magnetization, specific heat
and susceptibility of one hypothetic grain of FeF3 as a function of
temperature. Whatever the starting configuration, one will end up
with a perfect antiferromagnetic crystal as seen on Fig. 2(a).
Fig. 2(b) represents the specific heat and one can observe a
transition around 360 K which is close to the Néel temperature
of r-FeF3 . The magnetization of the system is uniformly zero due
to predominant antiferromagnetic behavior.
The magnetic configuration of the nanostructured FeF3 after
annealing is composed of two antiferromagnetic grains separated
by a transition zone (GB) with disordered spin structure as
illustrated in Figs. 2(c) and 3 representing the annealed config-
uration and the histogram of directions of magnetic moments in
the GB respectively. One observes on those figures that the
distribution of magnetic moments behaves as a speromagnet,
however displaying some reminiscence of the orientations in the
core of the grains, which is coherent with previous experimental
results. Fig. 2(d) represents the susceptibility for non-tilted, tilted
and twisted sample after annealing. As expected, the Neél
temperature is dependent on the original orientation of grains,
hence on the frustration rate. It should be noted that the previous
configuration is only one of the multiple ones. In fact, each change
in simulation conditions (number of MCS, orientation of initial
grains, final temperature, etc) leads to a different orientation of
magnetic moments of grains and then a different coupling of GB
moment.
We can conclude that the system is not in the magnetic
equilibrium state. This can be easily understood by considering
the fact that the GB is magnetically frustrated. The single spin flip
method we used is not appropriate for finding the magnetic
equilibrium state in a reasonable time of simulation. New condi-
tions have to be applied to reach this magnetic equilibrium state.
A constraint can be added by considering the fact that there is at
least one orientation of the relative global moment of each grain
for which the magnetic energy is minimal. We used fixed
boundary conditions to constrain magnetic moments of the two
grains to form an angle f and then relaxed the interface.
Fig. 4(a) represents the relative energy of the system for an angle
varying from 01 to 3601. We can clearly distinguish two magnetic
states with minimal energy around 901 and 2701.
The same behavior of the magnetic energy was obtained for
different grain orientations, but the angle corresponding to the
minimal energy was not the same. Fig. 4(b) compares the same
results obtained with non-tilted, tilted and twisted samples: it
clearly appears that the angle corresponding to the minimum
energy depends on the initial relative crystallographic orientation
of grains. It has recently been shown [23] that the only difference
between those samples is the rate of frustration in the GB,
suggesting that the angle corresponding to the minimum energy
depends on the rate of frustration. To verify this dependence, we
first considered a simple model consisting of a cubic cluster with
all interactions being antiferromagnetic, then we changed some
antiferromagnetic into ferromagnetic interactions at the interface
in order to create locally frustrated bonds. Within the area
delimited as the interface, frustration rate can be thus defined
as t ¼ JF =ðJ AF þ J F Þ, where J F and J AF respectively represent the
number of ferromagnetic and antiferromagnetic bonds. According
to this definition, t will vary from 0 (all interactions at interface
are AF) to 1 (all interactions are F) and the maximum frustration
in the system corresponds to t  0:5. For several values of t, the
system is relaxed and the angle between the moments of both
sides of the interface is measured (Fig. 4(c)). Several observations
can be made:
 a non-collinear coupling appears for a rate comprised between
t ¼ 0:2 and 0.8;
 the maximum frustration in the system corresponds to coupling
angle around 901;
 B. Fongang et al. / Journal of Magnetism and Magnetic Materials 324 (2012) 3646–3651 3649

Fig. 2. (a) and (c): X -Y view of the simulation box for the single grain and double grain respectively after annealing. (b) and (d) are specific heat as a function of
temperature for the single and double grain with different initial configurations (see text). In the double grain, the magnetic frustration reduces the Néel temperature.

20
15
Along X
10
5
0
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180
20
15
Along Y
10
5
0
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180
100
(log scale)

Along Z
10

0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180

θ in degrees

Fig. 3. Histogram of directions of the magnetic moments in a typical grain boundary (defined as a 0.4 nm wide zone along the interface). These angles are computed as
reverse cosines of the projections on the cartesian axes of the system.
3650 B. Fongang et al. / Journal of Magnetism and Magnetic Materials 324 (2012) 3646–3651
Fig. 4. Magnetic energy as a function of fixed angle boundary. (a) The non-tilted sample displays two magnetic states of lower energy around 901 and 2701. (b) The same
behavior is obtained for tilted, twisted and non-tilted system but with different angles. (c) Coupling angle between grains as a function of t. (d) Relative energy as a
function a function of coupling angle (see text).
 for t 4 0:8, ferromagnetic interactions are predominant giving
rise to a parallel coupling after annealing;
 as shown on Fig. 4(d), some values of t give rise to two
minimal energy states both corresponding to non-collinear
coupling.
By creating a reasonable amount of frustration in this simple
antiferromagnetic model, we showed that one will end up after
annealing having a non-collinear coupling. Therefore, we can
assume that the non-collinear orientation observed in ferric
fluoride nanostructures is due to the magnetic frustration of the
grain boundary.
We conclude that the angle between the total moment of the
two grains depends on the rate of frustration, which suggests the
possibility to tune the final orientation of the grains.
4. Conclusion
We have presented a Monte Carlo study of nanostructured
ferric fluoride by coupling a structural model previously estab-
lished from an atomistic description to a numeric modeling of
the magnetic structure at the local scale. As experimentally
evidenced, at the nanoscale, this compound is composed of two
parts: a perfect antiferromagnetic configuration in the grain and a
disordered magnetic configuration (speromagnetic like) at the
grain boundary. The present numeric study has shown that the
magnetic configuration at the grain boundaries is governed by the
cationic topology which results from the disorientation of crystal-
line neighboring grains. Consequently, the lowest magnetic
energy of the total system is strongly correlated to the rate of
magnetic frustration in the GB, opening the possibility to tune the
final orientation of the magnetic moments of the grains by
modulating the frustration rate.
It is finally important to emphasize that such an approach to
numerically model structural and magnetic structures can easily
be extended to ionic nanostructures, particularly to multilayers
composed of different oxide species in order to follow the
dependence of their properties on their topological ordering, the
thickness of the interfaces and the mismatch between layers.
References
[1] H. Gleiter, Progress in Materials Science 32 (1989) 223.
[2] A. Hernando, M. Vasquez, D. Paramo, Materials Science Forum 1033 (1998)
269–272.
[3] O. Crisan, J.-M. Le Breton, A. Jianu, J. Teillet, G. Filoti, Journal of Alloys and
Compounds 381 (1997) 262–263.
[4] J.-M. Grene che, M. Miglierini, A. Slawska-Waniewska, Hyperfine Interactions
126 (2000) 27.
[5] M. Miglierini, J.-M. Grene che, Hyperfine Interactions 297 (1999) 120–121.
[6] M. Miglierini, J.-M. Grene che, Hyperfine Interactions 122 (1999) 121.
[7] J.-M. Grene che, Hyperfine Interactions 151 (2002) 144–145.
[8] C. Suryanarayana, C. Koch, Hyperactive Interactions 130.
[9] J.M.D. Coey, Physical Review Letters 27 (1971) 1140.
[10] G. Ferey, M. Leblanc, R. DePape, J. Pannetier, Competing spin interactions
and frustration effects in fluorides, in: P. Hagenmuller (Ed.), Inorganic
fluorides, Academic Press, 1985, p. 395.
[11] G. Ferey, A. Leclerc, R. De Pape, J.-P. Mariot, F. Varret, Solid State Commu-
nications 29 (1979) 477.
[12] G. Ferey, F. Varret, J.M.D. Coey, Journal of Physics C: Solid State Physics 12
(1979) L531.
[13] J. Kanamori, Journal of Physics and Chemistry Solids 10 (1959) 87.
[14] J.B. Goodenough, Magnetism and Chemical Bond, Wiley-Interscience, New-
York.
[15] J.-M. Grene che, Hyperfine Interactions 122 (1999) 9.
[16] M.S.M. Saifullah, G.A. Botton, C.B. Boothroyd, C.J. Humphreys, Journal of
Applied Physics 86 (5) (1999) 2499–2504.
[17] S.G. Bakhvalov, O.A. Bayukov, E.M. Petrova, K.A. Sablina, V.M. Buznik,
V.M. Denisov, N.A. Poddubetskaya, Glass Physics and Chemistry 25 (4)
(1999) 345–351.
[18] M.J. Zhou, L.W. Zhao, T. Doi, S. Okada, J. Yamaki, Journal of Power Sources 195
(15) (2010) 4952–4956.
[19] S.-W. Kim, D.-H. Seo, H. Gwon, J. Kim, K. Kang, Advanced Materials 22 (46)
(2010) 5260–5264.
[20] N.W.B. Balasooriya, C. Poinsignon, in: Second IEEE International Nano-
electronics Conference, Shanghai, 2008, pp. 894–898.
[21] F. Wang, M. Malac, R.F. Egerton, Micron 37 (4) (2006) 316–323 32nd
Annual Meeting of the Microscopical Society of Canada, Hamilton, Canada,
May 18–20, 2005.
[22] H. Guérault, J.-M. Grene che, Journal of Physics: Condensed Matter 12 (2000)
4791.
[23] B. Fongang, Y. Labaye, F. Calvayrac, J.-M. Grene che, S. Zékeng, Coupled
structural and magnetic properties of ferric fluoride nanostructures part I:
 B. Fongang et al. / Journal of Magnetism and Magnetic Materials 324 (2012) 3646–3651 3651

a Metropolis atomistic study, Journal of Magnetism and Magnetic Materials [28] M.E.J. Newman, G.T. Barkema, Monte Carlo Methods in Statistical Physics,
322 (2010) 2888–2892. Oxford University Press, 2002.
[24] L. Berger, Y. Labaye, M. Tamine, J.M.D. Coey, Physical Review B 77 (2008) [29] P. Lacorre, J. Pannetier, Journal of Magnetism and Magnetic Materials 71
104431. (1987) 62–82.
[25] H. Kachkachi, M. Nogue s, E. Tronc, D. Garanin, Journal of Magnetism and [30] M. Leblanc, G. Ferey, J.-M. Grene che, A. Le Bail, R. De Pape, J. Pannetier,
Magnetic Materials 221 (2000) 158. Journal of Physique C8 46 (1985) 175.
[26] O. Iglesias, A. Labarta, Physica B: Condensed Matter 343 (2004) 286. [31] N.W. Ashcroft, N.D. Mermin, Solid State Physics, Saunders College Publishing,
[27] K. Binder, D.W. Heermann, Monte Carlo Simulation in Statistical Physics, vol. Philadephia, 1976.
80, 1988. [32] H.K. Perkins, Y. Hazony, Physical Review B 5 (1972) 7–18.