You are on page 1of 13

Journal of Magnetism and Magnetic Materials 184 (1998) 262274

Models for the dynamics of interacting magnetic nanoparticles


M.F. Hansen*, S. Mrup
Department of Physics, Building 307, Technical University of Denmark, DK-2800 Lyngby, Denmark
Received 10 October 1997; received in revised form 18 December 1997

Abstract
A critical review of models for the dynamics of interacting magnetic nanoparticles is given. It is shown that the basic
assumptions in the DormannBessaisFiorani model are unrealistic. The experimental observations on systems of
interacting magnetic nanoparticles can, at least qualitatively, be explained by the model derived by Mrup and Tronc for
weakly interacting particles, in combination with a transition to an ordered state in the case of strong interactions. ( 1998 Elsevier Science B.V. All rights reserved.
PACS: 75.50.Tt; 76.20.#q; 76.60.Es
Keywords: Fine particles; Superparamagnetism; Dipolar interactions; Magnetic ordering

1. Introduction
The dynamics of magnetic nanoparticles has
been the subject of numerous experimental and
theoretical studies. This subject is of great interest
both from a fundamental point of view and because
of the technological applications of ultrafine magnetic particles.
The superparamagnetic relaxation time of noninteracting particles is generally assumed to be
given by the NeelBrown expression [1,2]

A B

*E
q"q exp
,
0
k

(1)

* Corresponding author. Tel.: #45 45 25 31 69; fax: #45 45


93 23 99; e-mail: mfhansen@fysik.dtu.dk.

where q is of the order of 10~910~13 s and de0


pends weakly on temperature, *E is the energy
barrier separating two easy directions of magnetisation, k is the Boltzmanns constant and is
the temperature. Eq. (1) is a good approximation
for *E'2k. For particles with uniaxial symmetry the magnetic anisotropy energy is usually
written
E(h)"!K cos2 h,

(2)

where K is the magnetic anisotropy energy constant, the particle volume and h the angle between the magnetisation vector and an easy
direction of magnetisation. In this case the energy
barrier *E is equal to K.
Magnetic interactions between nanoparticles
may have a strong influence on the superparamagnetic relaxation and this has been the subject

0304-8853/98/$19.00 ( 1998 Elsevier Science B.V. All rights reserved.


PII S 0 3 0 4 - 8 8 5 3 ( 9 7 ) 0 1 1 6 5 - 7

M.F. Hansen, S. M~rup / Journal of Magnetism and Magnetic Materials 184 (1998) 262274

of numerous theoretical and experimental studies


[332]. In 1988 Dormann et al. published a model
[12] which predicted that the energy barrier always
increases when the strength of the interactions increases. This model was successfully applied to
explain measured data on iron nanoparticles
dispersed in an alumina matrix prepared by the cosputtering technique. Later Mossbauer spectroscopy data [13,26,29] on weakly interacting cFe O nanoparticles showed, however, a decrease
2 3
of the relaxation time with increasing inter-particle
interaction. In order to explain these experimental
observations, Dormann et al. postulated [18,19,
2124] an interaction-dependent variation of the
phenomenological damping parameter, g, which
is included in the theoretical expression for q .
0
They also suggested an interaction-dependent
variation of the surface anisotropy [21]. Another approach to explain the experimental observations in the c-Fe O Mossbauer spectroscopy
2 3
data was proposed by Mrup and Tronc [26]
in a model for weakly interacting nanoparticles
where q was assumed independent of interactions
0
and a decrease of the effective energy barrier with
increasing interactions was predicted. Subsequently, Mrup [27] discussed the possible
transition from a superparamagnetic state to an
ordered state in samples with strong inter-particle
interactions.
The validity of the models has been subject
to numerous discussions in the literature [1719,
2124,30], most of them, unfortunately, with rather
poorly documented points of criticism. The aim of
the present work is to compare the models in detail

263

and to discuss why the models lead to conflicting


results. Finally, we will discuss ordering phenomena in particle systems with strongly interacting particles.
For simplicity, we assume that the particles have
uniaxial anisotropy with the energy given by Eq. (2)
and that the inter-particle interaction is solely due
to dipoledipole interaction.

2. The DormannBessaisFiorani model


2.1. The original model
In the original model [12] Dormann et al. considered as a starting point the dipoledipole interaction between two neighbouring magnetic
particles, i and j, with magnetic moments l and l .
i
j
With spherical coordinates (h , u ) and (h , u ), as
i i
j j
defined in Fig. 1, for the directions of l and l they
i
j
calculated the interaction energy
E "e[lL ) lL !3(lL ) rL )(lL ) rL )],
ij
i j
i ij j ij

(3)

where r is the vector connecting the centres of the


ij
particles, denotes unit vectors and
k kk
e, 0 i j.
4p r3
ij

(4)

The easy direction of magnetisation for particle i is


along the z direction. Eq. (3) can be written
E "!l ) B .
ij
i ij

Fig. 1. The coordinates used in the DormannBessaisFiorani model.

(5)

264

M.F. Hansen, S. M~rup / Journal of Magnetism and Magnetic Materials 184 (1998) 262274

Introducing r "r (x, y, z) and B "e/k (B , B , B )


ij
ij
ij
i x y z
one finds

they therefore replaced cos h in Eq. (10) by its


j
Boltzmann average, Scos h T, and obtained
j

B "(3x2!1)sin h cos u
x
j
j
#3xy sin h sin u #3xz cos h ,
(6a)
j
j
j
B "3xy sin h cos u
y
j
j
#(3y2!1)sin h sin u #3yz cos h ,
(6b)
j
j
j
B "3xz sin h cos u
z
j
j
#3yz sin h sin u #(3z2!1)cos h .
(6c)
j
j
j
In the following the aim of the calculations is to
obtain an expression for the magnetic energy of
particle i during a reversal of its magnetisation
direction.
To simplify the calculations, Dormann et al. assumed that the probabilities to find the magnetic
moment of particle j at (h , u ) and (h , u #p)
j j
j j
are equal and they replaced B by the averij
age value B (h )"1[B (h , u )#B (h , u #p)]
ij j
2 ij j j
ij j j
[12,33]. Hence, they obtained the simple expression

E "!e(3z2!1)cos h [e(3z2!1)cos h /k],


ij
i
i

e
B " cos h ,
j
ij k ij
i
where

(7)

where [ ) ] denotes the Langevin function. For


small values of x,e(3z2!1)/k, we have that
[x cos h]+x/3 cos h+[x] cos h. For weak interactions compared to the thermal excitations
(x(1) the authors used this approximation twice
and rewrote Eq. (11) as
E +!E cos2 h ,
ij
Bi,j
i

(12)

where
E "e(3z2!1)[e(3z2!1)/k].
Bi,j

(9)

Using the same assumption on particle i, i.e. the


states (h , u ) and (h , u #p) were assumed to have
i i
i i
the same probability, an averaging over u was
i
performed, and they obtained
E "!e(3z2!1)cos h cos h .
(10)
ij
i
j
Next the authors assumed that particle j relaxes
much faster than particle i. For a fixed value of
h and neglecting the anisotropy of particle j,
i

(13)

This energy has the same symmetry as the uniaxial


anisotropy energy (Eq. (2)) and E was therefore
Bi,j
considered as the interaction anisotropy of particle
i induced by particle j. The total energy for particle
i was then found by summing up the contributions
from all neighbouring particles, j, and adding the
anisotropy energy
E +!(E #K)cos2 h ,
i
Bi
i

"(3xz, 3yz, 3z2!1).


(8)
ij
When comparing Eqs. (6a), (6b) and (6c) with
Eqs. (7) and (8) it is clear that this assumption is
equivalent to an averaging over u . The interaction
j
energy now becomes
E "!e[3xz sin h cos u
ij
i
i
#3yz sin h sin u #(3z2!1)cos h ]cos h .
i
i
i
j

(11)

(14)

where E "+ E . They have used Eqs. (12)(14)


Bi
j Bi,j
for all interaction strengths, irrespective of the fact
that it was derived using the low-field approximation of the Langevin function in Eqs. (12) and (13).
According to the authors [23] this does not introduce any significant errors, since the barrier height
and positions of the energy minima and maxima
are unchanged by the approximation. For applications they calculated the interaction anisotropy,
E , for a regular arrangement of identical particles
Bi
and thus obtained
E "n e(3z2!1)[e(3z2!1)/k],
Bi
1

(15)

where n is the number of nearest neighbours and


1
e and z depend on the geometrical arrangement.
The relaxation time obtained using Eq. (14) and
assuming constant q is
0

K E
q"q exp
# Bi .
0
k k

(16)

M.F. Hansen, S. M~rup / Journal of Magnetism and Magnetic Materials 184 (1998) 262274

For weakly interacting particles (x)1), using


[x]+x/3 in Eq. (15), they obtained

K n e2(3z2!1)2
q"q exp
# 1
,
0
3k22
k

(17)

and for strongly interacting particles (x*2), using


[x]+1!1/x in Eq. (15), they obtained

K n e(3z2!1)
q"q exp(!n )exp
# 1
.
0
1
k
k

(18)

Thus, for strongly interacting particles the result of


the model by Dormann et al. is that the apparent
microscopic attempt time decreases and the energy
barrier increases. It should be noted, however, that
the relaxation time of interacting particles in
Eq. (16) can never be smaller than the relaxation
time of isolated particles if q is assumed to
0
be constant. The authors claim that the model is
valid for any temperature and any interaction
strength [12].
2.2. Later modifications
The increase of the relaxation time with increasing interaction strength is in accordance with the
results of magnetisation measurements on a large
number of samples, but apparently in contradiction
with Mossbauer studies of c-Fe O particles with
2 3
weak inter-particle interactions [13,17,18,2124].
Dormann et al. explained these Mossbauer results
by a decrease of s with increasing inter-particle
0
interaction. Specifically, they postulated that the
phenomenological damping parameter, g, which is
one of the parameters which enters in the theoretical expressions for q , increases with increasing
0
interaction strength. According to a recent publication by Dormann et al. [21] the variation of g is due
to two contributions that both depend on the inter-particle interaction. The first contribution to
the damping parameter is due to disordered surface
spins, which make the coherent spin rotation more
difficult. The authors claimed that the surface
disorder decreases significantly with increasing inter-particle interaction and this will reduce the
damping parameter. The second contribution,
which was claimed to be dominant, is due to the

265

spatial variation of the dipole field from neighbouring particles within the particle. In the same publication, it was found that the relaxation time
increased faster with increasing interaction strength
than expected from the predictions of their model.
This observation was explained as a significant
interaction-induced increase of the surface contribution of the magnetic anisotropy energy constant,
K. However, since the interaction fields are much
smaller than the exchange fields which are responsible for the surface spin structure, it seems unrealistic that the interaction field can have any
significant influence on the surface spin disorder
and the surface anisotropy.

3. The MrupTronc model


Mrup and Tronc [26] considered a particle
with magnetic moment l . The surrounding pari
ticles give rise to a dipolar field B at the particle.
i
B forms an angle t with the easy direction of
i
magnetisation, which defines the z direction. The
x direction is chosen to be in the plane defined by
B and the z direction. The angle between the proi
jection of l on the xy plane and the x direction is
i
called u. These definitions are illustrated in Fig. 2.
The magnetic energy can then be written
E"K sin2 h!l ) B "K sin2 h
i i
!k B (sin h cos u sin t#cos t cos h).
i i

(19)

Fig. 2. The coordinates used in the MrupTronc model.

266

M.F. Hansen, S. M~rup / Journal of Magnetism and Magnetic Materials 184 (1998) 262274

In the following, the aim of the calculations is to


find the global energy minimum and the local energy maximum in the limit of weak interactions,
h,k B /2K@1. In this limit, the energy mini i
imum and energy maximum will be close to h"0
and 90, respectively. The u-value corresponding to
the global energy minimum is u"0. The h-value
corresponding to the global energy minimum,
found by differentiation of Eq. (19) with respect to
h and calculating to first order in h, is obtained
from sin h+h sin t. The resulting global energy
minimum is thus to second order in h
E +!K[h2 sin2 t#2h cos t].
(20)
.*/
The local energy maximum for a fixed value of u is
found in a similar way using the transformation to
h ,h#90 and that h +0 at the maximum.
0
0
The local energy maximum is found for
cos h"!sin h +!h cos t and the correspond0
ing energy is
(21)
E +K[1#h2 cos2 t!2h cos u sin t].
.!9
The energy barrier from the global minimum to the
local maximum is then given by
DE(t, u)"E

!E +K[1#h2
.!9
.*/
#2h(cos t!cos u sin t)].

(22)

Thus, for a given value of t, the energy barrier is


a function of the angle u. In the calculation of the
relaxation frequency the variation of the height of
the energy barrier with the angle u was taken into
account in the following way: It was assumed that
the transition probability per unit time, f(t, u), for
a transition along a specific path across the barrier,
defined by the angle u, is given by

*E(t, u)
f(t, u)"(2pq )~1 exp !
.
0
k

(23)

Here the variation of q with interaction was ne0


glected. This was justified by the fact that only
weak interactions are considered.
The dipole interaction field B has contributions
i
from a very large number of particles and its size
and direction will therefore fluctuate in time. However, as we will discuss later, it is a reasonable
approximation that B does not to fluctuate in time
i

during the magnetisation reversal. The authors


further assumed that all directions of B are equally
i
probable and that DB D has a Gaussian distribui
tion. The total transition probability per unit time,
q~1, was then found by integrating f(t, u) over all
values of u, corresponding to the different pathways across the barrier. Finally, integration over all
directions of B and the values of DB D was peri
i
formed:
First Eq. (22) was rewritten in terms of the parameters a,K/k and b,k B /k:
i i
*E(t, u)/k"a(1#h2)#b(cos t!cos u sin t).
(24)
For the case of b@1, one can write the transition
probability as
f(t, u)"(2pq )~1exp[!a(1#h2)][1
0
!b(cos t!cos u sin t)
#1b2(cos t!cos u sin t)2].
2
The integration over u then yields

(25)

f(t)+q~1 exp[!a(1#h2)][1!b cos t


0
#(b/2)2(1#cos2 t)].

(26)

After performing the integration over all directions


of B and the averaging over DB D one obtains
i
i
(27)
q~1"q~1 exp[!a(1#h2)][1#1Sb2T].
3
0
Using again that b@1 and the definitions of a,
b and h, the authors obtained

C
C

Sb2T
3
1! a~1
q"q exp a!
0
3
4

BD

BD

K k2SB2T
3 k
i 1!
"q exp
!
0
3k22
k
4 K

(28)

Thus, for k(K (which is usually the case for


the blocking temperature estimated by Mossbauer
spectroscopy), it is found that the interactions lead
to a decrease of the relaxation time. This effect is
related to the lowering of the energy barrier for
some values of u.
The authors estimated the Gaussian average
SB2T by using an expression analogous to that
i

M.F. Hansen, S. M~rup / Journal of Magnetism and Magnetic Materials 184 (1998) 262274

obtained by van Vleck [34] for an ion in a paramagnetic material

A B

k 2
SB2T"2 0 k2+ d~6,
i
ij
4p
j

(29)

where d is the distance to the surrounding parij


ticles, and it was assumed that all particles have
the same magnetic moment, k. Assuming identical
spherical particles of diameter D and writing
d "a fD, where fD is the average particle sepaij
ij
ration, they rewrote Eq. (29) as
k2M2f~6
SB2T" 0
+ a~6.
i
ij
288
j

(30)

The sum + a~6 depends on the geometrical arj ij


rangement only, and is thus independent of volume
concentration and particle size. The authors estimated that + a~6&1020.
j ij
The authors emphasised that the model is only
valid for h@1 and b@1. Later computer simulations using the exact expression for the energy,
however, showed that the analytical approximation
is valid up to h&0.3 [29]. For large values of h,
Mrup proposed that the interactions will lead to
an ordered frozen state of the magnetic moments
[27]. These ordering phenomena will be discussed
in Section 5.

4. Discussion
In this section we will first make some general
considerations about the relaxation process and
how the average relaxation time has to be calculated. Next we will discuss the model-specific
assumptions in more detail, and finally we will
illustrate the difference between the models using
a simple example.
4.1. General considerations
4.1.1. The flip process
A superparamagnetic particle will spend most of
the time with the magnetisation direction close to
an easy direction. The reversal time or flip time, q*,
of the magnetisation of a particle is therefore much

267

shorter than the superparamagnetic relaxation


time, q. When the field acting on a magnetic moment changes, the magnetic moment will not attain
its new equilibrium orientation instantaneously, i.e.
the moment needs some small amount of time to
respond to the changed field. We expect this response time to be at least of the order of q , which
0
can be considered the time scale of the spin-fluctuations near the energy minimum of a nanoparticle.
For two interacting nanoparticles of the same material, this implies that during the flip of the moment of one of the particles the system is not in
thermal equilibrium.
It is not simple to describe the non-equilibrium
properties of the interacting system during the flip
process and therefore two different approaches
were used. In the first approach, used by Dormann
et al., it was assumed that the magnetic moment of
the second particle attained its equilibrium average
instantaneously during every step of the flip process, such that Boltzmann statistics could be applied, and that the flipping particle responded to
the changed energy landscape instantaneously. In
the second approach, used by Mrup and Tronc, it
was assumed that the flip process is instantaneous,
i.e. it was assumed that the second particle does not
respond to the varying field from the first particle
during the flip process, and it was assumed that the
moment of the second particle does not perform
superparamagnetic fluctuations during the flip process (constant energy landscape).
In a real sample, the magnetic dipole field acting
on a particle will have contributions from a large
number of superparamagnetic particles. The
main contribution to the field is from the nearest
neighbours. Since the flip process, as argued
above, is very short compared to any superparamagnetic relaxation time and there is only
a small number of nearest neighbours, it is reasonable for weak interactions to assume that the interaction field from the nearest neighbours is constant
during the flip process. In addition to the field from
the nearest neighbours there will be a much smaller
contribution from distant particles. Since the number of distant superparamagnetic particles is large,
it is no longer valid to assume that this field is
constant during the flip process. As a first approximation, however, it seems reasonable to ignore the

268

M.F. Hansen, S. M~rup / Journal of Magnetism and Magnetic Materials 184 (1998) 262274

small fluctuating contribution to the interaction


field.
In order to be able to use the first approach, the
second particle will have to relax many times at
each instant in the flip process of the first particle,
i.e. the superparamagnetic relaxation time of the
second particle has to be much shorter than the flip
time of the first particle. As discussed above, this
assumption is clearly not fulfilled and the first approach is therefore very unrealistic.
In the second approach, however, it is assumed
that the flip time is short compared to all other
characteristic times. Since the neglected fluctuating
field is small compared to the dominating field from
the nearest neighbours which, as discussed above,
can be considered constant during the flip process,
this approach seems to be valid as a first approximation to the real situation.
4.1.2. Calculation of average relaxation time
The aim of both the DormannBessaisFiorani
model and the MrupTronc model is to calculate
the average relaxation time for a system of interacting nanoparticles. Again two different approaches
were applied. Dormann et al. calculated the average energy barrier and inserted it in the expression
for the relaxation time (Eq. (1)), while Mrup and
Tronc calculated the average transition probability
and took the reciprocal value.
As a simple illustration of the difference between
these two approaches consider a particle with an
energy landscape in which there is a singularity
with infinite energy between two easy directions.
There are many pathways for the motion of the
magnetic moment between two easy directions,
which do not pass through the singularity, and thus
the relaxation time is finite. The average energy
barrier for this constructed energy landscape is
obviously infinite and thus the calculated relaxation time for this approach will be infinite. When
the sum over transition probabilities, q~1, is made,
the contribution from pathways passing through
the singularity is zero (q is infinite) and they will not
contribute to the resulting relaxation time. This
example clearly shows that it is incorrect to use the
average energy barrier to obtain the average relaxation time and thus that the approach made by
Dormann et al. is incorrect.

4.2. Discussion of the DormannBessaisFiorani


model
The predictions of the DormannBessais
Fiorani model depend crucially on all of the following assumptions. The validity of each of the assumptions will be discussed below:
f It is equally probable to find the magnetic moment
of particle i and j in the directions (h, u) and
(h, u#p).
It is clearly seen that the dipoledipole interaction energy E in Eq. (3) depends on both u and
ij
j
u except for very special configurations of the
i
magnetic moments and that there is no mathematical justification in Eq. (3) for the assumed symmetry in u. Furthermore, even when u and
i
u fluctuate in time, some relative orientations
j
with lower energies will be more probable than
others in conflict with the symmetry assumption.
This symmetry assumption therefore seems unjustified. As stated earlier the assumption is equivalent
to an averaging over u and u . From a comparison
i
j
of Eqs. (5), (6a), (6b) and (6c) with Eq. (10) it is
seen that the result of the assumption is that
eight of nine terms in Eqs. (6a), (6b) and (6c) have
been neglected, and that only part of the z component of the interaction field remains. Specifically,
all components of the interaction field perpendicular to the easy direction of particle i have been
neglected.
f All nearest neighbours of a particle fluctuate faster
than the particle itself.
The assumption that all nearest neighbours fluctuate faster than the particle in question is not
possible to realise in a real sample except for very
special cases, e.g. if the sample has a broad size
distribution and only the relaxation of the largest
small fraction of particles is considered. However,
the authors have not made such restrictions.
f Particle j performs many fluctuations in each small
step of the flip of the moment of particle i, such that
Boltzmann statistics can be applied.

M.F. Hansen, S. M~rup / Journal of Magnetism and Magnetic Materials 184 (1998) 262274

The use of Boltzmann statistics during the flip


process is, as concluded in Section 4.1.1, not correct
due to the short duration of the flip process.
f he average relaxation time is obtained from the
average energy barrier.
Calculation of the average relaxation time by use
of the average energy barrier is, as shown in Section 4.1.2, not correct.
f he average energy barrier is obtained by pairwise addition of the interaction energy of the particle with each of its nearest neighbours.
This assumption is equivalent to the assumption
that all the nearest neighbours interact independently with particle i. Since the nearest neighbours
also interact with their nearest neighbours it is not
sufficient just to consider the pairwise interaction
energies and the influence of particle i on particle
j may therefore be overestimated.
The authors have tried to introduce the influence
of the next nearest neighbours by assuming strong
interactions between the particle and its nearest
neighbours and weak interactions between the particle and its next nearest neighbours and then
adding the contributions. However, it is highly unrealistic to neglect the influence of the nearest
neighbours on the next nearest neighbours of the
particle in question.
4.3. Discussion of the Mrupronc model
In the MrupTronc model all calculation procedures are in agreement with the general considerations in Section 4.1. The validity of the model
depends crucially on the following two assumptions:
f he transition probability f(t, u) is given by
Eq. (23).
The calculation of the superparamagnetic relaxation rate for a particle in a magnetic field which is
not parallel to the easy direction of magnetisation
is a complicated task which involves the solution of
the appropriate FokkerPlanck equation [35,36]

269

and for the general case there does not exist an


analytical solution. Mrup and Tronc therefore
reduce this complicated problem to a one-dimensional problem, which was basically solved by Neel
and Brown [1,2], by assuming that the reversal
takes place from the global energy minimum over
the barrier at some u to the other minimum.
The total relaxation rate is then found by averaging over u. The general problem has recently
been treated by Coffey et al. [36] and Pfeiffer [35].
These authors determine the transition probability using mainly the lowest energy barrier, i.e.
they assume that the magnetisation reversal takes
place for just a few values of u. For the case of
weak interactions, Pfeiffer has calculated the lowest
energy barrier [35]. The transition probability
found from this barrier will in general be higher
than the transition probability found by Mrup
and Tronc by averaging f(t, u) over u, i.e., if the
method by Pfeiffer is used, the relaxation time will
become shorter than predicted by Mrup and
Tronc.
f he dipole interaction field is independent of the
state of the particle in question and all field directions are equally probable.
When the interaction energies are small compared to the anisotropy energies the directions of
the magnetic moments of the particles will mainly
be determined by the anisotropy axes. The direction of the dipole field will therefore be determined
by the random orientation and geometrical arrangement of the nearest neighbours. In order to
account for different arrangements and field magnitudes an averaging must be performed. Since,
as we have argued, there is no preferred direction
of the interaction field it seems reasonable to assume equal probability of all directions. It also
seems reasonable to assume a Gaussian distribution of the field strength since this distribution
reflects the random magnitude of the interaction
field. The averaging is performed in agreement with
the comments in Section 4.1.2. The stated assumptions and the calculation procedure therefore seem
valid.
When the interaction energies are larger,
however, the interaction of a particle with its

270

M.F. Hansen, S. M~rup / Journal of Magnetism and Magnetic Materials 184 (1998) 262274

neighbours may significantly perturb the distribution of interaction fields. The analytical solution derived above is still claimed to be valid
for h&0.3, that is, for k B &K [29]. Howi i
ever, it is difficult to maintain the assumption
of equal probability of all directions of the interaction field since there may be a preference
for interaction fields along the easy direction
due to weak polarisation of the nearest neighbours.

reversal of l
i
q~1
E(p/2)!E
.*/
q~1" 0 exp !
2
k

4.4. A simple example

Because cosh(2e/k)*1 for all e'0 it is seen that


the interaction results in a reduction of the net
relaxation time.
When applying the DormannBessaisFiorani
model to the system it is assumed that the superparamagnetic relaxation of particle j is fast compared to the reversal time, q*, of particle i (although
the particles are identical), i.e. it is assumed that the
anisotropy does not keep the direction of l near an
j
easy direction of magnetisation. Because the z direction is perpendicular to r , Eq. (7) reduces to
ij
e
B " cos h
(33)
i k
j
i
and Eq. (12) becomes (for e(k)

The differences between the two models can be


illustrated by a simple example. We consider two
identical particles i and j which have their easy
directions of magnetisation perpendicular and parallel to r , respectively, as illustrated in Fig. 3. It is
ij
illustrative to calculate the relaxation time for particle i, assuming that l is confined to the plane of
i
Fig. 3.
When using the MrupTronc model one assumes that k is not reversed during the time it
j
takes for particle i to overcome its energy barrier.
With the anisotropy and interaction energies given
by Eqs. (2) and (3), respectively, and assuming
e@K, we find the minima and maxima for the
energy of particle i at h+0, p and h+p/2, 3p/2,
respectively, with energies E "E(0)"E(p)+
.*/
!K, E(p/2)+!2e and E(3p/2)+#2e. Thus,
for clockwise rotation of l the energy barrier is
i
decreased by the interactions. The relaxation rate is
given by the sum of the transition probabilities
per unit time for clockwise and counterclockwise

C A
A

B
BD

E(3p/2)!E
.*/
#exp !
k

(31)

A BN A B

(32)

or
K
q"q exp
0
k

cosh

e2
E +!
cos2 h .
ij
i
3k

2e
.
k

(34)

The total energy for particle i at the minima


and maxima is then given by E(0)"E(p)"
!(K#e2/3k) and E(p/2)"E(3p/2)"0, i.e.
according to this model the energy barrier and
hence the relaxation time is increased because of
the interaction.

5. Ordering phenomena

Fig. 3. Sketch of the particle configuration in the example discussed in the text. The dotted lines indicate the easy directions of
magnetisation.

Three-dimensional cubic arrays of magnetic


dipoles, which interact only via dipolar interactions, become ordered below a critical temperature
[3740]. For a simple cubic lattice the ground state
is antiferromagnetic, while it is ferromagnetic for
the FCC and BCC lattices. When the dipoles are
arranged in chain-like structures the ordering is
also ferromagnetic [25]. In samples with a random

M.F. Hansen, S. M~rup / Journal of Magnetism and Magnetic Materials 184 (1998) 262274

distribution of magnetic dipoles the ordered state


will resemble a spin glass [9,11,28,39,41].
Atoms with magnetic moments of the order of
a few Bohr magnetons in a crystal lattice show
magnetic order only at temperatures well below
1 K if they interact only via dipole interaction.
However, in samples with nanoparticles with magnetic moments of thousands of Bohr magnetons the
ordering temperature may be much higher.
For a regular array of identical dipoles with
identical magnetic moments k the transition temperature is given by
"a ,
#
0 $$
where
k k2
, 0
,
$$ 4pk d3

(35)

(36)

with a being the order of unity and d the distance


0
between neighbouring dipoles. Calculation of the
ordering temperature in samples of more or less
randomly distributed interacting nanoparticles is
much more complex, because the distributions of
particle magnetic moments and inter-particle distances must be taken into account. On the basis of
a comparison with spin glass models it has been
suggested [27] that the ordering temperature of
a system of randomly distributed particles is given
by an expression similar to Eqs. (35) and (36) where
k is the magnetic moment of a particle with the
median diameter and d is the average nearestneighbour distance. The expression for the ordering
temperature can also be written
a k
" 0 0 kM/,
#
k 4p

(37)

where M is the saturation magnetisation of a particle and / is the volume concentration. A rough
estimate of the ordering temperature suggested that
a is of the order of 10 [27]. If instead k represents
0
the magnetic moment of a particle with median
volume and the particle diameters have a log-normal distribution with the typical value of the logarithmic standard deviation p+0.3 one finds, by
using the relations given in Ref. [5], that a +1.
0
However, Dormann et al. [23] have argued that
a should be much smaller. Computer simulations
0

271

of Ising spins with dipole interactions have yielded


values of a in the range 46 [11].
0
For a detailed quantitative analysis of ordering
temperatures in samples with interacting nanoparticles it is necessary to know the particle size distribution. Moreover, it is important to know if the
particles are homogeneously distributed in the
samples and if some of them have formed agglomerates. In most experimental studies such information is not available. In the following, we compare
the ordering temperatures obtained experimentally
for some samples of interacting nanoparticles.
Recently, an AC susceptibility study of magnetite
particles in a frozen ferrofluid showed that the
relaxation time diverges at a finite temperature,
which depends on the concentration of particles
[10]. It was found that the relaxation time
could be described by the VogelFulcher law,
q"q exp[K/k(! )] with "a/0.8 and
0
0
0
a"800 K. As q diverges when the temperature
approaches , this temperature may be con0
sidered related to an ordering temperature. The
particles had an average moment of 2.5]104 l
B
and a mean diameter of 9 nm. Inserting these values
into Eq. (37), one finds that +a ]103/. By
#
0
comparing with the experimental results for we
0
find that there is good agreement with a +1.8.
0
This value is in accordance with the estimates made
using the model for spin glass ordering in small
particle systems [27].
The possible onset of ordering in systems of
nanoparticles can also be elucidated by studying
the samples with experimental techniques with different time scales, q , such as, Mossbauer spectro.
scopy with q +10~9 s and DC magnetisation
.
measurements with q +10 s. For non-interacting
.
particles, one would expect that the ratio between
the blocking temperatures estimated with these two
techniques should be of the order of 46 [27,28]. If
there are strong inter-particle interactions, both
techniques may show a transition from a superparamagnetic state to an ordered state near the
ordering temperature, i.e. the ratio between the
freezing temperatures obtained by the two experimental techniques should be close to unity. Mrup
et al. [28] have studied the magnetic properties of
powders of maghemite. Some samples were coated
with oleic acid while others were not. In order to

272

M.F. Hansen, S. M~rup / Journal of Magnetism and Magnetic Materials 184 (1998) 262274

increase the strength of the interactions some samples were compacted. The studies showed that the
characteristic temperatures obtained using Mossbauer spectroscopy and magnetisation measurements were quite similar. The uncoated particles
may be in close contact and therefore exchange
interactions between surface atoms belonging to
neighbouring particles may be significant. However, the coated particles, which have blocking temperatures on the Mossbauer and magnetisation
time scales in the range 140170 K, can only interact via dipole interactions. If we assume that the
particles can be considered as spheres with a core of
c-Fe O with a 1.0 nm thick surface shell of oleic
2 3
acid, and that the packing fraction of the spheres is
of the order of 0.6 we find by inserting these values
into Eq. (37) that a in this system also is about 1.5.
0
Recent AC susceptibility studies of 8 nm cFe O particles [42] and 4.7 nm a-Fe C par2 3
1~x x
ticles [43] in organic matrices have also shown
a divergence of the relaxation time at finite temperatures indicating transitions to ordered states.
Moreover, magnetisation measurements on a
frozen ferrofluid containing iron-nitride particles
have indicated a transition to an ordered state
similar to a spin glass [44,45]. The results of various studies of ordering phenomena in nanoparticle
systems are summarised in Table 1. It should be
emphasised that in most cases the particle size
distribution has not been analysed in detail and
therefore the particle diameter and the moments
may in some cases be average values, but in other
cases median values. The calculated values of
may therefore not be strictly comparable.
$$
However, the range of values of the parameter a is
0
relatively narrow for all these studies (between 0.75

and 2.4) indicating that the ordering temperature is


given by Eq. (37) with a value of a close to unity.
0
The discussion above clearly shows that
dipoledipole interactions can induce freezing of
particle moments. However, due to the polydispersity of nanoparticles and the superparamagnetic
blocking at low temperatures it is very difficult to
disentangle single particle effects from collective
effects. The freezing process may therefore differ
from the critical behaviour observed in spin glasses
near the ordering temperature. Some studies have
shown that particle systems exhibit spin glass like
characteristics, such as the ageing phenomenon
[31] and a concentration-dependent temperature
at which the relaxation time diverges [10].

6. Conclusions
We have analysed in detail the approximations
and assumptions made in the models currently
used to describe the dynamics of interacting superparamagnetic particles. We have shown that some
of the crucial assumptions for the model by Dormann et al. are unrealistic except for very special
cases and that the calculation procedure using
Boltzmann statistics during the magnetisation reversal and calculating the average relaxation time
from the average energy barrier is incorrect. We
conclude that predictions of the current version
of the model used in numerous publications [12,
1519,2124] are unreliable. As a consequence, the
predictions of the interaction-induced changes of
the phenomenological damping parameter and the
surface anisotropy which were based on the model
[21] are also unreliable.

Table 1
Sample characteristics and estimated ordering temperatures in samples of interacting magnetic nanoparticles
Material

D (nm)

k (k )
B

u (vol%)

(K)
$$

(K)
#

a
0

Ref.

c-Fe O
2 3
c-Fe O
2 3
c-Fe O
2 3
c-Fe O
2 3
c-Fe O
2 3
a-Fe C
1~x x
e-Fe N
3

9
9
9
7.5
8
4.7
6

2.5]104
2.5]104
2.5]104
1.1]104
1.2]104
1.7]104
1.8]104

0.6
3
6
&30
17
5
2.5

5
26
51
&83
60
25
45

12
50
82
&150
45
40
70

2.4
1.9
1.6
1.8
0.75
1.6
1.6

[10]
[10]
[10]
[28]
[42]
[43]
[44,45]

M.F. Hansen, S. M~rup / Journal of Magnetism and Magnetic Materials 184 (1998) 262274

We have shown that the model for weakly interacting nanoparticles by Mrup and Tronc is at
least qualitatively valid in the limit h@1. The problem of the general calculation of the relaxation time
of a superparamagnetic particle in an arbitrary
external magnetic field still needs some attention to
make the model quantitatively correct. The model
has been claimed to be valid for larger values of h
[29]. However, it is difficult to maintain the assumption of equal probability of all directions of the
interaction field in this case, whereby the predictions
of the model in this limit become unreliable.
Finally, we have discussed the transition to an
ordered state. We have shown that numerous experimental data support the existence of an interaction-induced ordered state and using a simple
model valid for spin glasses we have been able to
obtain consistent values of the parameter a .
0
Both the MrupTronc model combined with
the concept of ordering [27] and the Dormann
BessaisFiorani model are able to fit to the experimental data and thus it seems that there is no
experimental preference for one of the models.
However, as discussed, the underlying physical
assumptions for the DormannBessaisFiorani
model are unrealistic while the assumptions for the
MrupTronc model seem justified. We therefore
believe that the MrupTronc model combined
with the concept of ordering describes the important physical mechanisms, which govern the dynamics of interacting nanoparticles, and that the
model by Dormann et al. does not.
The decrease of the relaxation time with increasing interaction strength has only been observed for
maghemite nanoparticles. The reason why the
effect has not yet been observed in other particle
systems remains to be clarified.
We finally remark that much work on both noninteracting and interacting particle systems is still
required to get a detailed understanding of the
interesting phenomena observed in these systems.
Acknowledgements
Financial support from the Danish Technical
Research Council and the Danish Natural Science
Research Council is gratefully acknowledged.

273

References
[1] L. Neel, Ann. Geophys. 5 (1949) 99.
[2] W.F. Brown Jr., Phys. Rev. 130 (1963) 1677.
[3] S. Shtrikman, E.P. Wohlfarth, Phys. Lett. 85A (1981)
467.
[4] R.W. Chantrell, E.P. Wohlfarth, J. Magn. Magn. Mater. 40
(1983).
[5] K. OGrady, A. Bradbury, J. Magn. Magn. Mater. 39
(1983) 91.
[6] M. El-Hilo, K. OGrady, R.W. Chantrell, J. Magn. Magn.
Mater. 114 (1992) 295.
[7] K. OGrady, M. El-Hilo, R.W. Chantrell, IEEE Trans.
Magn. 29 (1993) 2608.
[8] R.W. Chantrell, G.N. Coverdale, M. El-Hilo, K. OGrady,
J. Magn. Magn. Mater. 157158 (1996) 250.
[9] W. Luo, S.R. Nagel, T.F. Rosenbaum, R.E. Rosensweig,
Phys. Rev. Lett. 67 (1991) 2721.
[10] J. Zhang, C. Boyd, W. Luo, Phys. Rev. Lett. 77 (1996)
390.
[11] M.A. Zaluska-Kotur, M. Cieplak, Europhys. Lett. 23
(1993) 85.
[12] J.L. Dormann, L. Bessais, D. Fiorani, J. Phys. C: Solid
State Phys. 21 (1988) 2015.
[13] P. Prene, E. Tronc, J.P. Jolivet, J. Livage, R. Cherkaoui, M.
Nogue`s, J.L. Dormann, D. Fiorani, IEEE Trans. Magn. 29
(1993) 2658.
[14] R. Cherkaoui, M. Nogue`s, J.L. Dormann, P. Prene, E.
Tronc, J.P. Jolivet, D. Fiorani, A.M. Testa, IEEE Trans.
Magn. 30 (1994) 1098.
[15] J.L. Dormann, in: J.L. Dormann, D. Fiorani (Eds.), Magnetic Properties of Fine Particles, North-Holland, Amsterdam, 1994, p. 115.
[16] J.L. Dormann, D. Fiorani, J. Magn. Magn. Mater.
140144 (1995) 415.
[17] E. Tronc, P. Prene, J.P. Jolivet, F. dOrazio, F. Lucari, D.
Fiorani, M. Godinho, R. Cherkaoui, M. Nogue`s, J.L. Dormann, Hyp. Int. 95 (1995) 129.
[18] P. Prene, E. Tronc, J.P. Jolivet, J.L. Dormann, in: I. Ortalli,
(Ed.), Proc. ICAME95, SIF, Bologna, 1996, p. 485.
[19] E. Tronc, Nuovo Cimento 18D (1996) 163.
[20] C. Bellouard, I. Mirebeau, M. Hennion, Phys. Rev. B 53
(1996) 5570.
[21] J.L. Dormann, F. DOrazio, F. Lucari, E. Tronc, P. Prene,
J.P. Jolivet, D. Fiorani, R. Cherkaoui, M. Nogue`s, Phys.
Rev. B 53 (1996) 14291.
[22] E. Tronc, J.P. Jolivet, Mater. Sci. Forum 235238 (1997)
659.
[23] J.L. Dormann, D. Fiorani, E. Tronc, Adv. Chem. Phys. 98
(1997) 283.
[24] J.L. Dormann, L. Spinu, E. Tronc, J.P. Jolivet, F. Lucari,
F. dOrazio, D. Fiorani, Non-Crystalline and Nanoscale
Materials (Proceedings of the V International Workshop
on Non-Crystalline Solids), J. Rivas and M.A. LopezQuintela (Eds.), World Scientific, Singapore, 1998, in press.
[25] S. Mrup, P.H. Christensen, B.S. Clausen, J. Magn. Magn.
Mater. 68 (1987) 160.

274

M.F. Hansen, S. M~rup / Journal of Magnetism and Magnetic Materials 184 (1998) 262274

[26] S. Mrup, E. Tronc, Phys. Rev. Lett. 72 (1994) 3278.


[27] S. Mrup, Europhys. Lett. 28 (1994) 671.
[28] S. Mrup, F. Bdker, P.V. Hendriksen, S. Linderoth, Phys.
Rev. B 52 (1995) 287.
[29] J.Z. Jiang, S. Mrup, T. Jonsson, P. Svedlindh, in: I. Ortalli
(Ed.), Proc. ICAME95, SIF, Bologna, 1996, p. 529.
[30] S. Mrup, F. Bdker, M.F. Hansen, J.Z. Jiang, Non-Crystalline and Nanoscale Materials (Proceedings of the V International Workshop on Non-Crystalline Solids), J. Rivas
and M.A. Lopez-Quintela (Eds.), World Scientific, Singapore, 1998, in press.
[31] T. Jonsson, J. Mattson, C. Djurberg, F.A. Khan, P. Nordblad, P. Svedlindh, Phys. Rev. Lett. 75 (1995) 4138.
[32] H. Zhang, M. Widom, Phys. Rev. B 51 (1995) 8951.
[33] J.L. Dormann, personal communication.
[34] J.H. Van Vleck, J. Chem. Phys. 5 (1937) 370.
[35] H. Pfeiffer, Phys. Stat. Sol. (a) 122 (1990) 377.

[36] W.T. Coffey, D.S.F. Crothers, J.L. Dormann, L.J. Geoghegan, Yu.P. Kalmykov, J.T. Waldron, A.W. Wickstead,
Phys. Rev. B 52 (1995) 15951.
[37] J.M. Luttinger, L. Tisza, Phys. Rev. 70 (1946) 954.
[38] J.M. Luttinger, L. Tisza, Phys. Rev. 72 (1947) 257.
[39] J.P. Bouchard, P.G. Zerah, Phys. Rev. B 47 (1993) 9095.
[40] S. Romano, Phys. Rev. B 49 (1994) 12287 and references
cited therein.
[41] I. Tamura, M. Hayashi, J. Magn. Magn. Mater. 72 (1988)
285.
[42] T. Jonsson, P. Nordblad, P. Svedlindh, Phys. Rev. B 57
(1998) 497.
[43] C. Djurberg, P. Svedlindh, P. Nordblad, M.F. Hansen, F.
Bdker, S. Mrup, Phys. Rev. Lett. 79 (25) (1997) 5154.
[44] H. Mamiya, I. Nakatani, J. Appl. Phys. 81 (1997) 4733.
[45] H. Mamiya, I. Nakatani, T. Furubayashi, Phys. Rev. Lett.
80 (1998) 177.

You might also like