Simple models for dynamic hysteresis loop

calculations of magnetic single-domain

nanoparticles: Application to magnetic
hyperthermia optimization
Cite as: J. Appl. Phys. 109, 083921 (2011); [Link]
Submitted: 12 July 2010 . Accepted: 31 December 2010 . Published Online: 21 April 2011

J. Carrey, B. Mehdaoui, and M. Respaud

ARTICLES YOU MAY BE INTERESTED IN

Fundamentals and advances in magnetic hyperthermia

Applied Physics Reviews 2, 041302 (2015); [Link]

Static and dynamic magnetic properties of spherical magnetite nanoparticles

Journal of Applied Physics 94, 3520 (2003); [Link]

Superparamagnetism
Journal of Applied Physics 30, S120 (1959); [Link]

J. Appl. Phys. 109, 083921 (2011); [Link] 109, 083921

© 2011 American Institute of Physics.

 JOURNAL OF APPLIED PHYSICS 109, 083921 (2011)

Simple models for dynamic hysteresis loop calculations of magnetic

single-domain nanoparticles: Application to magnetic hyperthermia
optimization
J. Carrey,a) B. Mehdaoui, and M. Respaud
Université de Toulouse, INSA, UPS, Laboratoire de Physique et Chimie des Nano-Objets (LPCNO), 135
Avenue de Rangueil, F-31077 Toulouse, France and CNRS, UMR 5215, LPCNO, F-31077 Toulouse, France
(Received 12 July 2010; accepted 31 December 2010; published online 21 April 2011; publisher
error corrected 26 April 2011)
To optimize the heating properties of magnetic nanoparticles (MNPs) in magnetic hyperthermia
applications, it is necessary to calculate the area of their hysteresis loops in an alternating magnetic
field. The separation between “relaxation losses” and “hysteresis losses” presented in several articles is
artificial and criticized here. The three types of theories suitable for describing hysteresis loops of
MNPs are presented and compared to numerical simulations: equilibrium functions, Stoner–Wohlfarth
model based theories (SWMBTs), and a linear response theory (LRT) using the Néel–Brown
relaxation time. The configuration where the easy axis of the MNPs is aligned with respect to the
magnetic field and the configuration of a random orientation of the easy axis are both studied. Suitable
formulas to calculate the hysteresis areas of major cycles are deduced from SWMBTs and from
numerical simulations; the domain of validity of the analytical formula is explicitly studied. In the case
of minor cycles, the hysteresis area calculations are based on the LRT. A perfect agreement between
the LRT and numerical simulations of hysteresis loops is obtained. The domain of validity of the LRT
is explicitly studied. Formulas are proposed to calculate the hysteresis area at low field that are valid for
any anisotropy of the MNP. The magnetic field dependence of the area is studied using numerical
simulations: it follows power laws with a large range of exponents. Then analytical expressions derived
from the LRT and SWMBTs are used in their domains of validity for a theoretical study of magnetic
hyperthermia. It is shown that LRT is only pertinent for MNPs with strong anisotropy and that
SWMBTs should be used for weakly anisotropic MNPs. The optimum volume of MNPs for magnetic
hyperthermia is derived as a function of material and experimental parameters. Formulas are proposed
to allow to the calculation of the optimum volume for any anisotropy. The maximum achievable
specific absorption rate (SAR) is calculated as a function of the MNP anisotropy. It is shown that an
optimum anisotropy increases the SAR and reduces the detrimental effects of the size distribution of
the MNPs. The optimum anisotropy is simple to calculate; it depends only on the magnetic field used in
the hyperthermia experiments and the MNP magnetization. The theoretical optimum parameters are
compared to those of several magnetic materials. A brief review of experimental results as well as a
method to analyze them is proposed. This study helps in the determination of suitable and unsuitable
materials for magnetic hyperthermia and provides accurate formulas to analyze experimental data. It is
also aimed at providing a better understanding of magnetic hyperthermia to researchers working on this
subject. VC 2011 American Institute of Physics. [doi:10.1063/1.3551582]

I. INTRODUCTION netic field simply equals the area of their hysteresis loop,
which can be expressed as
Magnetic hyperthermia is a promising cancer treatment ð þHmax
technique that is based on the fact that magnetic nanoparticles
A¼ l0 MðHÞdH; (1)
(MNPs) placed in an alternating magnetic field release heat. Hmax
Active research is being done to improve the specific absorp- where M(H) is the NP magnetization. Then the SAR is:
tion rate (SAR) of MNPs, which could permit the treatment of
SAR ¼ Af : (2)
tumors of a smaller size1 and could reduce the amount of ma-
terial that must be injected to treat a tumor of a given size. As will be described in more detail in the following text, A
If an assembly of MNPs is put into an alternating mag- depends, in a very complex manner, on the characteristics of
netic field of frequency f and amplitude l0Hmax, the amount the NPs: A depends on the NPs’ effective anisotropy Keff,
of heat A released by the MNPs during one cycle of the mag- their volume V, the temperature T, the frequency and ampli-
tude of the magnetic field, and eventually magnetic interac-
tions between NPs. It is thus crucial to be able to evaluate A
a)
Author to whom correspondence should be addressed. Electronic mail: as precisely as possible as a function of these parameters to
[Link]@[Link]. target the optimum parameters for the required application.

0021-8979/2011/109(8)/083921/17/$30.00 109, 083921-1 C 2011 American Institute of Physics

V
083921-2 Carrey, Mehdaoui, and Respaud J. Appl. Phys. 109, 083921 (2011)

The theoretical literature on the properties of MNPs is II. NUMERICAL CALCULATIONS AND ANALYTICAL
very large. Of this large number of articles, those related to EXPRESSIONS OF HYSTERESIS LOOPS
the evolution of the hysteresis area with the intrinsic parame- A. Single-domain uniaxial nanoparticles in a magnetic
ters of the MNPs are of interest for magnetic hyperthermia field
applications.2–8 A few theoretical papers have also been
devoted specifically to the problem of magnetic hyperther- Let us consider a MNP of volume V composed of a fer-
mia.1,9–14 However, the articles published on this subject are romagnetic material having a spontaneous magnetization MS
not complete and are sometimes inaccurate. First, the major- and a magnetocrystalline anisotropy. Below a critical vol-
ity of the articles are mainly based on the linear response ume, the MNP becomes single domain to minimize its mag-
theory (LRT), which, as will be shown later, is not the most netic energy. Because all the spins are parallel to one
useful for magnetic hyperthermia. Second, when using theo- another, one can model the magnetization as a single giant
ries derived from the Stoner–Wohlfarth model, the domain magnetic moment l ¼ MSV, the amplitude of which does not
of validity is not taken into account, and, more importantly, depend on its spatial orientation; these are the so-called
central conclusions that could be derived from them are “macrospin” and coherent rotation approximations. As a
missing. Third, in articles published by Hergt et al., an artifi- result of magnetic anisotropy, l is generally pinned along
cial separation of the mechanisms responsible for the heating well-defined directions, that is, along its magnetic anisotropy
is made between “hysteresis losses” and “relaxation losses.” axis. As a result of several contributions, among them the
This separation is improper or at least very confusing; in our magnetocrystalline, shape, and surface contributions that
opinion, it is detrimental to a deep understanding of mag- arise from spherical deviations, the anisotropy can be very
netic hyperthermia and to correct usage of the models to cal- complex. Indeed these numerous contributions have neither
culate SARs. Unfortunately, the paradigmatic presentation of the same symmetries (cubic versus uniaxial) nor the same
magnetic hyperthermia by a large number of experimental directions. Nevertheless one of these contributions domi-
articles still follows this separation. Finally, a recent article nates and determines the main first-order contribution. From
by N. A. Usov has used numerical simulations of hysteresis a practical point of view, one generally concludes from the
loops to study hyperthermia.14 Although this article draws experimental studies that the anisotropy displays a first-
qualitatively correct conclusions, it is based on the study of order-dominant uniaxial character. Thus considering the
examples and does not provide a generalization and a quanti- macrospin approximation and an effective uniaxial anisot-
tative approach to the problem. ropy (Keff), the energy of a MNP placed in an external mag-
The present article aims to give a complete and rigorous netic field (l0Hmax) is given by the following:2
presentation of the theory of magnetic hyperthermia. First, in
Sec. II, we will provide a global view of the three types of Eðh; /Þ ¼ Keff V sin2 ðhÞ  l0 MS VHmax cosðh  /Þ; (3)
theories suitable for the calculation of the hysteresis loop where h is the angle between the easy axis and the magnetiza-
areas of MNPs, and we will give simple analytical formulas tion and / is the angle between the easy axis and the magnetic
for this purpose; this will include a precise determination of field [see Fig. 1(a)]. In the following, we will use the dimen-
their domain of validity using numerical calculations. A clar- sionless parameters r ¼ KeffV/kBT and n ¼ l0MSVHmax/kBT.
ification of the issue concerning hysteresis vs relaxation The reduced magnetic energy normalized to the thermal energy
losses will also be done in this section. Numerical simula- is
tions will be used to illustrate the variety of curves that could
be obtained experimentally when measuring the magnetic Eðh; /Þ
field dependence of the SAR. In Sec. III, the previous results ¼ r sin2 ðhÞ  n cosðh  /Þ: (4)
kB T
will be used for a specific study of magnetic hyperthermia. It
will be shown that the LRT is only pertinent for strongly ani- Figure 1(c) displays the reduced magnetic energy as a func-
sotropic MNPs; for weakly anisotropic MNPs, theories tion of the normalized magnetic field (n varies between 0
derived from the Stoner–Wohlfarth model should be used and 2) and the angle h for a given particle orientation,
instead. Formulas that predict the optimum volume of MNPs / ¼ 30 and r ¼ 1. Two different shapes are noticeable.
as a function of material and experimental parameters will When l0Hmax is greater than the anisotropy field
be provided. Additionally, it will be shown that the anisot- l0HK ¼ 2Keff/MS, the energy landscape displays only one
ropy of the MNPs is the central parameter for the optimiza- minimum, which defines the equilibrium position, that is,
tion of magnetic hyperthermia because it determines both along the anisotropy axis direction. Conversely, when
the maximum achievable SAR and controls the influence of l0Hmax is less than l0HK, the energy profile as a function of
the size distribution of MNPs on the SAR. A simple formula h displays two minima at the coordinates (h1, E1) and (h2,
to determine the optimum anisotropy for magnetic hyper- E2) and two maxima. We will refer to (h3, E3) as the saddle
thermia will be proposed, and a comparison with the parame- point, i.e., the smaller maximum [see Fig. 1(b)]. For n ¼ 0
ters of bulk magnetic materials will be done. This study (in the absence of magnetic fields), the magnetization can
should help to determine suitable and unsuitable materials take two equivalent equilibrium values at h1 ¼ 0 and
for magnetic hyperthermia, and it will provide accurate for- h2 ¼ 180 , that is, along its easy axis [see Figs. 1(c) and
mulas to analyze experimental data. We also hope it will 1(d)]. For a finite positive n, the magnetic field favors one of
lead to a better understanding of magnetic hyperthermia for the two minima (here, the one initially at h1 ¼ 0 ). Increasing
researchers in this field. n moves the abscissa of this minimum progressively so that a
083921-3 Carrey, Mehdaoui, and Respaud J. Appl. Phys. 109, 083921 (2011)

FIG. 1. (a) Schematic representation of a uniax-

ial single-domain MNP. The large arrow repre-
sents the magnetization. (b) Illustration of the
three extrema of the energy landscape with
n ¼ 0.5, / ¼ 30 , and r ¼ 1. (c) The energy of
a NP as a function of h is plotted as a function of
n for / ¼ 30 and r ¼ 1. (d) The energy of a NP
as a function of h for n ¼ 2, 0, 1.05, 2, and 4.
In this example, the energy barrier between the
two minima disappears for n  1.05. Vertical
dashed lines are drawn at h ¼ 0, 30, and 180 .

magnetization parallel to the magnetic field is favored. In the Analytical expressions can be derived using Eqs. (3), (5),
examples of Figs. 1(c) and 1(d), the abscissa of the minimum and (6) in two cases. First, if the anisotropy of the MNPs is
tends progressively toward h1 ¼ / ¼ 30 . For a negative mag- neglected, that is, for r ¼ 0, the magnetization reads
netic field, the energy landscape is similar to the positive one
except that the well at h2 ¼ 180 is initially favored and the M ¼ MS LðnÞ; (7)
minimum tends toward h2 ¼ 210 for large n. Different shapes
of the energy landscape can be more clearly seen in Fig. 1(d) where L(n) is the Langevin function:
where E/kBT is plotted versus h for various values of n. 1
LðnÞ ¼ cothðnÞ  : (8)
B. Equilibrium functions
n

Let us first consider the case of thermal equilibrium, that Thus the widely used Langevin function intrinsically neglects
is, the derivation of the equilibrium functions characterized by the anisotropy of MNPs. The Langevin function is plotted in
a reversible magnetization process. Details on the calculation Figs. 2(a) and 2(b) for comparison with the numerical results.
of the equilibrium functions can be found in Ref. 15. The prin- Second, when r is very large, one can consider that the mag-
ciple is the following: for MNPs the easy axes of which are netization has only two possible positions: the two minima of
oriented at / with respect to the magnetic field, the probability the energy landscape. This is called the “two-level approx-
f(h) to find the magnetization in a given direction is imation.” It reduces the integration overall possible values of h
  to a sum of two terms. If one considers the case where the easy
Eðh; /Þ axis of the MNPs is aligned along the direction of the field
exp
k T (/ ¼ 0), the magnetization is expressed as
f ðh; /Þ ¼ ð  B  : (5)
Eðh; /Þ
exp dh
h kB T MS expðr þ nÞ  MS expðr  nÞ
M¼ ¼ MS tanhðnÞ: (9)
The resulting magnetization in the direction of the applied expðr þ nÞ þ expðr  nÞ
magnetic field is derived after numerical integrations.15 The
main results are displayed in Fig. 2 for two cases of interest: This function is also plotted for the sake of comparison with
Fig. 2(a) displays the hysteresis loop with the hypothesis that the numerical results in Fig. 2(a). It is observed for / ¼ 0
all the anisotropy axes are oriented along the direction of the that the magnetization curve progressively evolves from a
magnetic field (/ ¼ 0), and Fig. 2(b) illustrates the hypothesis L(n) function for small r to the tanh(n) function for large r.
that the measured system is an assembly of MNPs with their Considering these general expressions, one can derive
anisotropy axes randomly oriented in space. In the latter case, the equilibrium susceptibility in each case. To this end,
the hysteresis loop is then the result of an integration overall Figs. 2(a) and 2(b) illustrate two results of interest: i) in all
possible values for /. The magnetization M is then given by cases, the magnetization is linear with the magnetic field
ð p=2 only when n < 1 and ii) for randomly oriented MNPs, the
M¼ Mð/Þ sin / d/: (6) initial slope of the equilibrium function does not depend on
0 the anisotropy of the MNPs and is the same as the one of
083921-4 Carrey, Mehdaoui, and Respaud J. Appl. Phys. 109, 083921 (2011)

l0 MS2 V
v0 tanh ¼ ¼ 3v0 Langevin ð/¼ 0Þ: (11)
kB T
As a consequence, the equilibrium susceptibility of an as-
sembly of MNPs with their easy axes aligned along the mag-
netic field evolves from v0 Langevin for small r to v0 tanh for
large r. Figure 2(c) displays this evolution as deduced from
the numerical calculations of Fig. 2(a). A phenomenological
fit of the v0(r) function is provided and plotted in Fig. 2(c):
v0 ðrÞ 2
¼3  r 1:47 ð/¼ 0Þ: (12)
v0 Langevin 1þ 3:4

C. Hysteresis loops at T 5 0—Stoner–Wohlfarth model

Details on the calculations and results of the Stoner–
Wohlfarth model can be found in several articles.2,16 The origi-
nal Stoner–Wohlfarth model does not take into account any
thermal activation, which is relevant at T ¼ 0 or in the limit of
infinite field frequency ( f ! 1). As a consequence of neglect-
ing thermal activation, the magnetization can only stay along
one of the two equilibrium positions. At T ¼ 0 and when the
magnetization is in one of the two minima of energy, the
switch of the magnetization from the metastable state to the
equilibrium position can only occur when the energy barrier is
fully removed by the magnetic field. The field at which this
occurs is called the critical field [see Fig. 1(d)].
For a MNP with its easy axis aligned along the magnetic
field direction, the suppression of the anisotropy barrier by
the magnetic field occurs at l0HK. The resulting hysteresis
loop is a perfect square, that is, the coercive field equals both
the critical and the anisotropy fields in this case. This is
shown using numerical calculations in Fig. 3(a). The area of
this hysteresis loop is maximal and is given as

A ¼ 4l0 HC MS ¼ 4l0 HK MS ¼ 8Keff : (13)

Note that upon tilting the angle / of the MNP, the hysteresis
loops progressively close up, and they become fully closed
for / ¼ 90 [3]. As a consequence, the case of randomly ori-
ented MNPs, exhibits a reduced coercive field
FIG. 2. (Color online) Results of the equilibrium functions. (a) and (b) Nu- l0HC ¼ 0.48l0HK and a remnant magnetization that is half
merical calculations of the hysteresis loop for MNPs at thermal equilibrium the saturation value because the hysteresis loop is an average
for various values of their reduced anisotropy r. The L(n) function and the over all contributions [see Fig. 3(b)]. The area of the hystere-
tanh(n) function are plotted alongside the data for comparison. (a) The easy
axes are aligned with the magnetic field (/ ¼ 0). (b) The easy axes are ran-
sis loop is reduced to
domly oriented in space. (c) Evolution of the initial slope of the hysteresis A ¼ 2l0 HC MS ¼ 1:92Keff : (14)
loop as a function of r when / ¼ 0. The dots are extracted from the numeri-
cal simulations. The line is a phenomenological fit using Eq. (12). The
dashed and dotted lines show the initial slope of the L(n) and tanh(n) func- D. Dynamic hysteresis loop calculations within the
tions, respectively. two-level approximation
1. Model
the Langevin function. In this case, the static susceptibility
In between the pure superparamagnetic state, which is
is given by
characterized by equilibrium functions where the magnetic
l M2 V
v0 Langevin ¼ 0 S : (10) moment can explore all directions, and the T ¼ 0 limit,
3kB T wherein the magnetic moment lies in its local minimum, the
The case of MNPs with an easy axis aligned along the direction description of the field dependence of the magnetization is
of the field is more complex: for r  1, the Langevin function very complicated. Because the energy barrier of MNPs is rel-
is valid and leads to a susceptibility similar to Eq. (10). How- atively low, the magnetization reversal is thermally acti-
ever, for r  1, the equilibrium susceptibility v0tanh of an vated—the so-called Néel–Brown relaxation—which leads
aligned MNP is equal to the slope of the tanh function: to a progressive reduction of l0HC when the temperature is
083921-5 Carrey, Mehdaoui, and Respaud J. Appl. Phys. 109, 083921 (2011)

FIG. 3. (Color online) Results on SWMBTs. (a) and (b) Examples of major hysteresis loops for values of j ranging from 0 to 1.5. (c) and (d) Normalized coer-
cive field as a function of j. The dots corresponds to simulation results with Hmax, Keff, V, T, and f varying over a wide range of experimentally relevant values.
Solid lines correspond to Eqs. (22) or (24). The dotted line is the area calculated from the true coercive field of the hysteresis loops using Eq. (25) or (26). (e)
and (f) Normalized hysteresis area as a function of j. Solid lines are calculated by combining Eq. (22) with Eq. (25), or Eq. (24) with Eq. (26). The dashed line
is the area calculated from the true coercive field of the hysteresis loops using Eq. (25) or (26). (a), (c), and (e): / ¼ 0. (b), (d) and (f): random orientation.

raised. Decreasing the sweeping rate of the magnetic field magnetic field).17,18 For the sake of simplicity, we will keep
has similar consequences.3–5,8 The incorporation of these these frequencies constant and equal to 1010 Hz.
effects in a model is far from easy. Within the two-level In the various theoretical articles dealing with the influ-
approximation, one neglects excited states inside each well ence of a finite temperature and frequency in the Stoner–
so that dynamic loop calculations only depend on the two Wohlfarth model, the numerical methods and approxima-
minima (h1, E1) and (h2, E2) and on the saddle point (h3, E3). tions to include these thermally activated jumps vary. In their
When the applied magnetic field is below l0HK, the magnet- numerical simulations, Garcia-Otero et al. have taken the
ization can switch from the h1 to the h2 direction at a rate m1 crude approximation that the switching occurs as soon as
given by DE(l0Hmax,/) ¼ (E3  Ei) ¼ kBT where i identifies the start-
  ing well.5 Pfeiffer et al. assumed that the switch from one
0 E3  E1
v1 ¼ v1 exp  : (15) well to the other occurs when the relaxation time over the
kB T
barrier matches a “measurement time” sm.4 In both articles,
Similarly, the switching rate m2 from the h2 to the h1 direc- the final results for the variation of the coercive field with
tion is given by   temperature and frequency depends on this sm parameter.
0 E3  E2
v2 ¼ v2 exp  : (16) However, trying to define the value of sm has necessarily
kB T
unphysical consequences. Indeed the coercive field mainly
The attempt frequencies v01 and v02 are complex functions of depends on the sweeping rate of the magnetic field. In a
the material parameters (gyromagnetic ratio, damping, MS superconducting quantum interference device measurement,
and Keff) and experimental conditions (temperature and one could simply take the “measurement time” as the time to
083921-6 Carrey, Mehdaoui, and Respaud J. Appl. Phys. 109, 083921 (2011)

"   34 #
measure one point, which corresponds to Pfeiffer et al.’s cri- kB T sm
teria. However, the coercive field would vary as a function l0 HC  0:48l0 HK 1 ln :
Keff V s0
of the step value. Alternatively, one could take the time of a
complete cycle. In this case, the coercive field would vary ðrandom orientationÞ ð21Þ
with the maximum applied magnetic field.
The principle of our calculation is more rigorous and is We previously used the latter equation to interpret hyperther-
similar to the one used by Lu et al.3 and Usov et al.8,14 A time- mia experiments with FeCo MNPs by stating that sm ¼ (1/f).19
dependent magnetic field H(t) ¼ Hmaxcos(xt) is applied to the However, as mentioned in the preceding text, these analytical
MNP along a direction that makes an angle / with respect to formulas do not depend on the sweeping rate of the magnetic
the easy axis. To compute the magnetization, one has to calcu- field 4Hmaxf but on this undefined sm parameter. Recently,
late the time dependence of p1 and p2 ¼ (1  p1), the probabil- Usov et al.8 proposed a novel dimensionless parameter j for
ity of finding the magnetization in the first and second the variation of the coercive field that takes into account the
potential wells, respectively. The time evolution of p1 reads sweeping rate. In the / ¼ 0 case, the coercive field is
 
@p1 l0 HC ¼ l0 HK 1  j2
1
ð/¼ 0Þ (22)
¼ ð1  p1 Þm2  p1 m1 : (17)
@t
Knowing the occupation probabilities, one can calculate the with
magnetization according to  
kB T kB T
j¼ ln : (23)
M ¼ MS ðp1 cos h1 þ ð1  p1 Þ cos h2 Þ: (18) Keff V 4l0 Hmax MS Vf s0

The resolution of the time evolution of p1 is performed using In Fig. 3(c), numerical calculations of hysteresis loops are
an explicit Runge–Kutta2,3 method. With this method, the compared to Eq. (22). To achieve this, a large number of
time step is not constant but becomes shorter when p1 varies simulations were performed with parameters varying over
more, which ensures an optimum compromise between cal- a wide range of values: f (10–400 kHz), l0Hmax (0.05–5 T),
culation time and precision. When there is only one mini- Keff (103–106 J m3), T (0.5–500 K), and the spherical radii
mum, p1 is simply set equal to either zero or unity. of the nanoparticles (1.5–30 nm). The normalized coercive
To calculate hysteresis loops for a random orientation of field extracted from the hysteresis loop is then plotted as a
MNPs, 50 cycles with / ranging from 0 to p/2 are calculated. function of j. The fact that all the data fall onto a single mas-
Then the magnetization is calculated according to Eq. (6). ter curve confirms the relevance of the dimensionless param-
We will show that using this single model, one can simulate eter proposed by Usov et al.8 Our simulations are in good
major and minor hysteresis loops and the behavior in the agreement with the analytical expressions (22) and (23)
framework of the LRT. For the latter cases, the initial condi- derived by Usov et al. as long as j is below roughly 0.5.
tions are set to p1 ¼ p2 ¼ 0.5, and several successive hystere- For the random orientation case, Usov et al. derived an
sis loops are performed until the curve converges and expression for the coercive field from the phenomenological
becomes symmetrical with respect to the abscissa axis. fit of their numerical simulations. However, the authors
Under most conditions, only two or three cycles are neces- made the assumption that the coercive field equals the criti-
sary to achieve convergence. Typical examples of the hyster- cal field, which is not rigorously true for NPs with a large /,
esis loops generated will be shown in the following sections. and performed a fit over a large range of temperatures. In
Fig. 3(d), our simulations for the coercive field in the random
2. Temperature and frequency dependence of the orientation case are shown. From the best fit of these data at
coercive field low values of j, the following formula is obtained:
Historically, the first analytical expressions for the tem- l0 HC ¼ 0:48l0 HK ðb  jn Þðrandom orientationÞ; (24)
perature dependence of the coercive field were based on the
approximation of the “measurement time” previously where b ¼ 1 and n ¼ 0.8 6 0.05. Usov et al. found slightly
described. In the / ¼ 0 case, the expression of the coercive different coefficients of b ¼ 0.9 and n ¼ 1. The domain of va-
field reads4 lidity is roughly the same as for the aligned case: Eq. (24) is
"   12 # roughly valid up to j ¼ 0.5. In the remainder of this article,
kB T sm our own values for b and n will be used.
l0 HC ¼ l0 HK 1  ln ; ð/¼ 0Þ; (19)
Keff V s0
3. Temperature and frequency dependence of the
where s0 is the frequency factor of the Néel–Brown relaxa- hysteresis loop area
tion time defined as (see Sec. II E)
The aim of this subsection is to study the frequency and
1 1 temperature dependencies of the hysteresis area and to derive
s0 ¼ 0 ¼ 0 : (20) general analytical expressions in the case of aligned and ran-
2m1 2m2
domly oriented MNPs. This topic has not been addressed in
In the case of randomly oriented NPs, the following analyti- the publications mentioned in the preceding text. At T ¼ 0,
cal expression was obtained by Garcia-Otero et al.:5 the area is proportional to the coercive field as given by
083921-7 Carrey, Mehdaoui, and Respaud J. Appl. Phys. 109, 083921 (2011)

Eqs. (13) and (14), and the question is whether these expres- after a small step in the magnetic field. The results of Sec. II
sions are still valid when T = 0. Similar to the study of the B showed that the magnetization is linear with the magnetic
coercive field dependence, we estimated from numerical hys- field approximately for the condition when n < 1. A small
teresis calculations the associated area for the aligned and value of n is thus the first criterion for the validity of the
randomly oriented cases; these are displayed in Figs. 3(e) LRT; it will be more precisely studied in the following text.
and 3(f), respectively. Examples of these hysteresis loops are Moreover, in Eq. (28), sR is a variable independent of the
shown in Figs. 3(a) and 3(b). These data are then compared magnetic field, which is only true for small deformations of
to those calculated using the analytical expressions the barrier between the two equilibrium positions, i.e., when
l0Hmax  l0HK. At the magnetic field frequencies used in
AðTÞ  4l0 HC ðTÞMS ð/¼ 0Þ; (25) hyperthermia or magnetic measurements, it can be shown
that the second condition is always verified when the first
and
one is (see following text). The relaxation time of the mag-
AðTÞ  2l0 HC ðTÞMS ðrandom orientationÞ: (26) netization sR when the MNPs cannot move physically equals
the Néel–Brown relaxation time sN, which reads
The area A(T) has been calculated by using i) HC deduced    
from the simulated hysteresis loops (shown as a dashed line) 1 Keff V Keff V
sR ¼ sN ¼ 0 exp ¼ s0 exp : (29)
and ii) HC calculated using Eqs. (22) and (24) (shown as a 2m1 kB T kB T
solid line). The A(T) curves calculated according to the first
procedure match the exact values of the area except at large The fact that there is a factor 1/2 between s0 and the attempt
j values (above 1). This difference is due to the reduced frequency m01 comes from the fact that we are dealing with a re-
squareness of the hysteresis loops. From the close compari- versible jump in a system with two potential wells. If this point
son between the dots and the dashed line, it can be observed is unclear to the reader, it is illustrated in Fig. 4. Figure 4 dis-
that Eqs. (25) and (26) slightly overestimate the hysteresis plays an imaginary case in which all of the MNPs are first mag-
area for large j. This overestimation partially compensates netized in one direction and then relax at zero magnetic field
for the underestimation of the coercive field by Eqs. (22) and through a reversible jump over an energy barrier at a rate m1.
(24) at high j. As a consequence, the combination of Eqs. The probability p1 to find the MNPs in this well drops exponen-
(22) and (24) with Eqs. (25) and (26) gives an acceptable tially to 0.5 with a time constant of 1/(2m1). Thus the relaxation
value of the area at higher j values than Eqs. (22) and (24) time of the magnetization is half the mean time taken by the
do for the coercive field. To provide a numerical limit that magnetization to reverse spontaneously. This explains the fac-
can be used later in the article, Figs. 3(e) and 3(f) show that tor of 1/2 between s0 and the attempt frequency m01 .
the area is calculated with less than 10% error when j < 0.7. In the LRT, the response of the system to an alternating
Finally, the transition toward reversible hysteresis loops magnetic field
can also be deduced from this figure. If we state that the re- HðtÞ ¼ Hmax cosðxtÞ (30)
versibility occurs when HC  0.01HK, this transition can be is
estimated to occur when j  1.6. MðtÞ ¼ jvjHmax cosðxt þ uÞ; (31)

E. Minor hysteresis loops and linear response theory where u is the phase delay between the magnetization and
the magnetic field. From Eq. (28), it is straightforward to
The LRT has been previously reported in several show that
articles.1,7,9,10 The presentation here will be slightly different v0
from that of other articles and will aim to explicitly illustrate jvj ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi; (32)
1 þ x2 s2R
the fact that the LRT is also a model to calculate the hystere-
sis area, a point that was not always developed in previous
works. The results for MNPs aligned with the magnetic field,
not derived in previous articles, will also be given. xsR 1
sin u ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi or cos u ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : (33)
The LRT is a model that aims to describe the dynamic 1 þ x2 s2R 1 þ x2 s2R
response of an assembly of MNPs using the Néel–Brown
relaxation time. The starting assumption of this model is that Basic mathematics indicates that Eqs. (30) and (31) corre-
the magnetic system responds linearly with the magnetic spond to the parametric equation of an ellipse in the (H, M)
field and its magnetization can be put in the form plane. The area Aellipse of this ellipse and the angle c between
its long axis and the abscise axis are given by
MðtÞ ¼ v~HðtÞ; (27)
2 2 xsR
where v~ is the complex susceptibility and reads Aellipse ¼ pHmax jvj sin u ¼ pHmax v0 ; (34)
1 þ x2 s2R
1
v~ ¼ v0 : (28) and
1 þ ixsR
2
2Hmax jvj cos u 2v0
v0 is the static susceptibility defined in Sec. II B, and sR tan 2c ¼ ¼ : (35)
2
Hmax 2 jvj2
 Hmax 1 þ x2 s2R  v20
is the time it takes for the system to relax back to equilibrium
083921-8 Carrey, Mehdaoui, and Respaud J. Appl. Phys. 109, 083921 (2011)

In several articles, Eq. (36) is defined as applying to

“relaxation losses” of superparamagnetic MNPs. In these
articles, “relaxation losses” are opposed—as if it was a dif-
ferent process—to the “hysteresis losses” of ferromagnetic
NPs.1,9,12,13 We have tried to illustrate here that this separa-
tion is not correct, or is at least confusing: all the losses,
whether the MNPs are in the superparamagnetic regime or in
the ferromagnetic regime, are always “hysteresis losses”
insofar as they are simply given by the hysteresis loop area.
LRT is simply one model among several that aims to calcu-
late the hysteresis loop area and shape when the magnetic
response is linear with the applied magnetic field. To avoid
confusion and misunderstandings of this concept, we suggest
FIG. 4. Evolution of the probability p1 to find a nanoparticle in the initially full putting an end to the separation between hysteresis losses
potential well as a function of t  m1. The result is an exponential decay function and relaxation losses and, rather, making a distinction
with a time constant of 1/2m1. (Inset) Illustration of the emptying of an initially between different kinds of models aiming at calculating the
full potential well through a reversible jump over the barrier.
hysteresis area. For instance, it is correct to say “LRT is suit-
able to calculate the hysteresis area of MNPs in the superpar-
amagnetic regime at low magnetic field” but not to say “in
In Fig. 5, the results of these equations are shown: Fig. 5(a)
ferromagnetic NPs, relaxation losses disappear and are
displays the ellipses plotted using Eqs. (30) and (31) for
replaced by hysteresis losses”. Unfortunately, misconcep-
v0 ¼ 0.1 and Hmax ¼ 1 for various values of xsR while Fig.
tions similar to this are present in a large number of articles
5(b) displays the evolution of c from Eq. (35) and from a
on magnetic hyperthermia.
graphical analysis of Fig. 5(a). A similar agreement is found
Numerical simulations have been performed to check
when plotting the evolution of the hysteresis area A using Eq.
the validity of Eqs. (29), (36), and (37). Specifically, numeri-
(34) and by integrating over the area of the hysteresis loops of
cal simulations of minor hysteresis loops were run with Keff,
Fig. 5(a) (not shown). These figures and the corresponding
V, f, and l0Hmax varying over a wide range of values while
equations illustrate the behavior of the magnetization loops as a
keeping n  1 and Hmax/HK  1. Hysteresis areas are then
function of the applied magnetic field in the LRT. First, when
normalized by the prefactor of Eqs. (36) and (37) and plotted
xsR ! 0, the hysteresis loop is simply a straight line with a null
as a function of xsR. The final results for / ¼ 0 and for the
hysteresis. In this condition, the angle c is such that tan
random orientation case are plotted in Figs. 5(c) and 5(d).
c(xsR ¼ 0) ¼ v0. Then the hysteresis area is maximal for
The fact that the hysteresis area displays a maximum for
xsR ¼ 1. When xsR ! 1, the system does not have the time to
xsR ¼ 1 explains the shape of the curves. These graphs illus-
respond to the magnetic field excitation and jvj ! 0 as does the
trate the perfect agreement between simulations and LRT
hysteresis area.7
both for the / ¼ 0 and the random orientation case. As a mat-
For magnetic hyperthermia, calculating the hysteresis
ter of fact, the hysteresis loops obtained by the numerical
area when a magnetic field l0Hmax is applied requires a com-
simulations are indistinguishable from the ones obtained
bination of the v0 expressions given by Eq. (10), (11), or (12)
using Eqs. (30)–(33), and so an illustration of the hysteresis
with Eq. (34); the final result is multiplied by l0 [see Eq.
loops obtained by numerical simulations would be indistin-
(1)]. For randomly oriented MNPs or for aligned MNPs
guishable from Fig. 5(a).
when r is negligible, this leads to
Next, the domain of validity of the LRT was studied by
pl20 Hmax
2
MS2 V xsR increasing n and Hmax/HK and comparing the hysteresis areas
A¼ provided by simulations to those provided by Eqs. (36) and
3kB T ð1 þ x2 s2R Þ
(37). The results are shown in Figs. 5(e) and 5(f). Practically,
this study is performed by studying the volume dependence
(random orientation) or (/ ¼ 0 and r  1Þ: ð36Þ of the hysteresis area (which modifies n only) for various
values of Keff (which modifies Hmax/HK only). It must first
For aligned MNPs with a strong r, this leads to be noted that for realistic values of the measurement fre-
quency, n and Hmax/HK are not completely independent.
pl20 Hmax
2
MS2 V xsR Indeed combining Eq. (39) (see below) with the definition of
A¼ ð/ ¼ 0 and r  1Þ: (37)
kB T ð1 þ x2 s2R Þ n and Hmax/HK shows that near the resonance—when the
hysteresis area is not too weak—n is always larger than
For aligned MNPs with any r, the phenomenological law Hmax/HK by a factor of 2ln(xs0), which is always much
given by Eq. (12) can be used, which
0 leads to 1 larger than 1. As a consequence, there is no realistic case
where n  1, Hmax /HK  1 and the area is not negligibly
pl20 Hmax
2
MS2 V xsR B 2 C
A¼ @3   r 1:47 A small.
3kB T ð1 þ x2 s2R Þ (38) As expected, increasing n and Hmax/HK leads to a dis-
1þ
3:4 crepancy between the equations and simulations. When
ð/ ¼ 0 and any rÞ: decreasing the anisotropy, the position of the peak in the
083921-9 Carrey, Mehdaoui, and Respaud J. Appl. Phys. 109, 083921 (2011)

FIG. 5. (Color online) Results of the LRT. (a) Evolution of the hysteresis loop as a function of xsR is plotted using Eqs. (30)–(33) for v0 ¼ 0.1 and Hmax ¼ 1.
(b) Angle between the long axis of the ellipse and the abscise axis c as a function of xsR. The line is deduced from Eq. (35); the dots are a direct measurement
from the ellipses plotted in (a). (c) and (d) Normalized hysteresis areas as a function of xsR. Each dot corresponds to a numerical simulation and the line to
LRT [Eqs. (36) and (37)]; values of Keff, V, f, and l0Hmax were varied over a wide range of parameters keeping n and Hmax/HK well below 1. (e) and (f) Hyster-
esis areas obtained by simulations (dots) are plotted along the theoretical areas provided by Eq. (36) or (37) (lines) as a function of n for various values of Keff.
l0Hmax ¼ 1 mT, f ¼ 100 kHz, T ¼ 300 K, and MS ¼ 106 A m1. (g) and (h) Data similar to the previous ones except that the hysteresis area is divided by the the-
oretical area and that data for higher values of n are shown. The corresponding values of Hmax/HK are provided. The horizontal dashed line illustrates the dis-
crepancy between Eqs. (36) or (37) and the simulations when n > 1. (c), (e) and (g): / ¼ 0. (d), (f), and (h): random orientation.

area obtained by the simulations is progressively shifted to- sponding values of Hmax/HK are provided. For n ¼ 1, the dis-
ward higher values of n compared to the position of the peak crepancy is around 6 20% for the / ¼ 0 case and around
calculated using LRT. The discrepancy is greater in the ran- þ70%/40% for the random orientation case. If lower error
dom orientation case than the / ¼ 0 case. To obtain quantita- bars are required when using LRT, the maximum acceptable
tive values of the error made when using LRT, the ratio n value should be reduced accordingly. Interestingly, it is
between the area given by simulations and calculations has observed that the LRT either overestimates or underesti-
been plotted in Figs. 5(g) and 5(h). In these graphs, the corre- mates the area and that the transition between the two zones
083921-10 Carrey, Mehdaoui, and Respaud J. Appl. Phys. 109, 083921 (2011)

FIG. 6. (Color online): Numerical simulations of the hysteresis area. When they are not varied, the parameter values are Keff ¼ 13 000 J m3, MS ¼ 106 A m1,
f ¼ 100 kHz, l0Hmax ¼ 20 mT, T ¼ 300 K, and m10 ¼ 1010 Hz. (a) Evolution of area as a function of the radius and temperature. The dots represent the maximum
area for a given temperature. (b) Evolution of area as a function of the magnetic field and the radius. The line represents the maximum area at a given magnetic
field. (c) and (d) Normalized hysteresis area as a function of the normalized magnetic field for various MNP radii. (a)– (c) / ¼ 0. (d) Random orientation.

is approximately localized around the peak in the area [see the magnetic field. Because the results depend on all the
Figs. 5(e) and 5(f)]. Therefore if we identify the zone to the external and structural parameters, there is no universal
left of the peak as a “superparamagnetic regime” and the curve or pertinent dimensionless parameters. Thus we have
zone to the right of the peak as a “ferromagnetic regime,” only used as an illustration the magnetic parameters of bulk
these data can be summarized this way: for values of n above magnetite and external parameters typical of hyperthermia.
1, the LRT overestimates the hysteresis area in the superpar- When they are not being varied, the parameter values in this
amagnetic regime and underestimates it in the ferromagnetic section are Keff ¼ 13 000 J m3, MS ¼ 106 A m1, f ¼ 100
regime. Another conclusion, which will be developed later in kHz, l0Hmax ¼ 20 mT, T ¼ 300 K, and m10 ¼ 1010 Hz.
this article but is also visible in Figs. 5(e) and 5(f), is that In Fig. 6(a), the hysteresis area is plotted as a function
LRT is mainly useful for highly anisotropic NPs. of the radius and temperature and in Fig. 6(b) as a function
of radius and magnetic field in the / ¼ 0 case. In both cases,
it is evident that the largest areas are obtained for large ferro-
F. Dynamic hysteresis loops and area
magnetic nanoparticles. However, for such nanoparticles, ab-
in the general case
rupt transitions are observed as a function of the temperature
We have just seen that the LRT allows one to calculate or the applied magnetic field between a regime where the
the hysteresis area when n < 1. Similarly, SWMBTs can be area is very small and a regime where the area is very large.
used when j < 0.7 and when the hysteresis loop is a major Basically when the coercive field is larger (smaller) than the
hysteresis loop, that is, when the NPs are saturated by the applied magnetic field, the area is very small (very large). It
magnetic field. In all other cases, these theories cannot be is also observed that for a given set of parameters, there is an
used, and numerical simulations are the only way to calcu- optimum radius to maximize the area, which we have plotted
late the hysteresis area. In this subsection, we will present in the two graphs. The analytical determinations of this opti-
results for the hysteresis area provided completely by numer- mum volume and area will be the subject of Sec. III.
ical simulations. In particular, numerical simulations give us In Figs. 6(c) and 6(d), the magnetic field dependence of
the opportunity to study the magnetic field dependence of the hysteresis area is plotted for values of l0Hmax between 0
the hysteresis area, which is accessible in hyperthermia and l0HK (which here is 26 mT) in the / ¼ 0 case and the ran-
experiments by performing measurements as a function of dom orientation case; the area is normalized to its value at 26
083921-11 Carrey, Mehdaoui, and Respaud J. Appl. Phys. 109, 083921 (2011)

However, the general shape of the curve for MNPs between

3.5 and 9 nm for magnetic fields up to 26 mT is a power law
function with an exponent progressively decreasing from 2
down to 0.6. The fit by the power law is good over the whole
range of magnetic fields studied. Above 9 nm, the MNPs are
in the ferromagnetic regime where the curves display an
inflection point and an abrupt increase at the coercive field.
In this case, the curves were fitted by a power law only up to
this inflection point. The exponent of the power law rises
very quickly up to very large values as MNPs grow in size.
As a consequence, a power law with a large range of expo-
nents can be observed experimentally in the magnetic field
dependence of SAR even in the simplest case of monodis-
perse single-domain nanoparticles. Thus, exponents other
FIG. 7. Exponent of the best power law fit of the curves shown in Figs. 6(c) than two should not be considered as something exotic in
and 6(d). When there was an inflection point in the curve, the fit was per- hyperthermia experiments, and an exponent of three should
formed only up to this point.
not necessarily be considered as the signature of multido-
main nanoparticles.
mT. For very small NPs, the LRT is valid and predicts a square
dependence of the hysteresis loop area [see Eqs. (36)–(38)],
which has been verified in a large number of experimental G. Summary of the models and magnetic properties
as a function of the NP size
works (e.g., Refs. 20, 21, or 22) and is also observed here. For
large NPs in the ferromagnetic regime, the magnetic field de- In this subsection, we briefly summarize graphically the
pendence displays a very abrupt jump with a null hysteresis results of the previous sections to illustrate which model
area below the critical fields and a sharp increase followed by should be used to calculate the hysteresis area. This part is
a plateau. For NPs with intermediate sizes, the transition based on the description of Fig. 8, starting from the proper-
between these two regimes is progressive and the curves dis- ties of small MNPs.
play a large variety of shapes. These curves were fitted by a For small nanoparticles (when approximately j > 1.6),
power law in a range where the fit is acceptable. The corre- the hysteresis loop is reversible, and the coercive field is
sponding exponents are shown in Fig. 7. almost null. In this case, the hysteresis loops can be
We will describe in detail the results for the / ¼ 0 case calculated using equilibrium functions for any value of the
[see Figs. 6(c) and 7], with the understanding that the ran- magnetic field (see Sec. II B). NPs in this range are useless
dom orientation case is qualitatively similar. The exponent for magnetic hyperthermia because of their null hysteresis,
of two predicted by the LRT is always observed when n < 1. but the equilibrium functions are useful for two reasons: i)
For very small MNPs in the superparamagnetic regime they are used in the LRT to calculate the initial static suscep-
(here, 3 nm), this square law is followed across the whole tibility v0 and ii) they are useful to accurately fit magnetic
range of magnetic fields studied. For larger MNPs, the do- measurements on MNPs in the superparamagnetic regime.15
main of validity of the LRT very rapidly shrinks, and this When the volume increases such that j < 1.6, the hyster-
exponent is still observed at very small magnetic fields. esis loop progressively opens. In this case, the shape of the

FIG. 8. Schematic representation of the evolu-

tion of the magnetic properties of MNPs as a
function of their volume and of the models suita-
ble to describe them. The label (1) illustrates
that the maximum magnetic field for which the
LRT is valid decreases with increasing volume.
The label (2) is the domain where incoherent re-
versal modes occur so SWMBTs are not valid
anymore. The label (3) shows a plateau in the
volume dependence of the coercive field.
083921-12 Carrey, Mehdaoui, and Respaud J. Appl. Phys. 109, 083921 (2011)

hysteresis loop cannot be calculated simply for any value of

l0Hmax. However, the LRT allows one to calculate the hys-
teresis loop shape if n < 1 or n  1 depending on the accu-
racy required (see Sec. II E). The fact that the magnetic field
value for which the LRT is valid progressively reduces
as the NP volume increases is schematized at Label (1) in
Fig. 8.
The formal transition between the superparamagnetic
regime (xsN < 1) and the ferromagnetic regime (xsN > 1)
occurs at xsN ¼ 1. Precisely at this transition, the hysteresis
loop area for small magnetic fields displays a maximum (see
Sec. II E). However, we emphasize that nothing special
occurs at this transition with respect to the hysteresis loop
shape and area at high magnetic field: the coercive field has
started to grow well before the transition and keeps increas-
ing after the transition. This means that the hysteresis loop
area at high magnetic fields does not display a maximum
here but continues to increase with an increase in the volume.
For xsN > 1, the MNPs are in the ferromagnetic regime
where they display a more and more open hysteresis loop as
their volume increases. In the ferromagnetic regime, the
SWMBTs are suitable to describe the NP hysteresis loops if
the NPs are not too close to the superparamagnetic-ferro-
magnetic transition, i.e., for j < 0.7. Using SWMBTs to cal-
culate the area supposes also that the MNPs are saturated,
which is true for approximately l0Hmax > l0HC in the / ¼ 0
case and l0Hmax > 2l0HC in the random orientation case.
The LRT is still valid in this region and can be used to calcu-
late minor hysteresis loop area at very low fields.
In larger MNPs, incoherent reversal modes start to
occur, which leads to a decrease of the coercive field. This
is where the theories used in this article cease to be valid. FIG. 9. (Color online) Calculations of the optimum radius for MNPs as a
function of their anisotropy using the LRT and SWMBTs for three different
For large MNPs, SWMBTs predict for the coercive field a
values of MS labeled as M1, M2 and M3. M1 ¼ 0.4  106 A m1, M2 ¼ 106
value independent of the NP volume, which is the value A m1, M3 ¼ 1.7  106 A m1, f ¼ 100 kHz, l0Hmax ¼ 20 mT, T ¼ 300 K,
of the coercive field at T ¼ 0. As a consequence, if the and m10 ¼ 1010 Hz. The horizontal dotted lines show the limit above which
volume at which this phenomenon occurs is smaller than the LRT is not valid anymore (n > 1). The LRT result is common to all
graphs and is given by Eq. (39). (a) and (b) Comparison between different
the one at which incoherent reversal modes start, a plateau formulas to calculate the optimum size. The solid lines refer to the numerical
in the evolution of the coercive field with the volume solution of Eqs. (22) and (24) combined with Eqs. (40) and (42). The dashed
might in principle be observed. We made this assumption line refers to Eqs. (44) and (45). (a): / ¼ 0. (b): random orientation.
in Fig. 8, and the plateau is labeled as (3). If incoherent
reversal modes began before this plateau, a peak in the co-
ercive field value should be observed instead of a plateau.
A. Optimum size as a function of the anisotropy
Finally, the largest MNPs are composed of a vortex23 or
of several magnetic domains separated by magnetic walls. In this part, the optimum volume of MNPs for mag-
In the latter case, the process leading to their magnetiza- netic hyperthermia will be calculated as a function of their
tion is the growth of one or several domains in the direc- anisotropy. In the figures, specific values of the external
tion of the field at the expense of the others. In this case, parameters have been used: f ¼ 100 kHz, l0Hmax ¼ 20 mT,
their hysteresis loops at very small magnetic fields are T ¼ 300 K, and m10 ¼ m20 ¼ 1010 Hz. These f and l0Hmax
described by “Rayleigh loops.”24 values are the ones used in clinical applications at the
Charité Hospital, Berlin.25 In addition, results for three dif-
ferent values of MS (M1 ¼ 0.4  106 A m1, M2 ¼ 106
III. OPTIMUM PARAMETERS FOR MAGNETIC
A m1, and M3 ¼ 1.7  106 A m1) will be shown. They
HYPERTHERMIA
correspond to the magnetizations of CoFe2O4, magnetite,
In this section, the models presented in the preceding and iron, respectively. Equivalent graphs for any value of
text are used to calculate the optimum parameters of MNPs the external and magnetic parameters can be plotted with
for magnetic hyperthermia. The domain of validity of each the condition to resolve one equation numerically (see fol-
model will be taken into account. The case of MNPs aligned lowing text).
with the magnetic field (/ ¼ 0) as well as the case of a ran- In the LRT, even though the hysteresis area is different
dom orientation will both be treated. for / ¼ 0 and for a random orientation of NPs, Eqs. (36)–(38)
083921-13 Carrey, Mehdaoui, and Respaud J. Appl. Phys. 109, 083921 (2011)

show that the maximum of this area always occurs for kB T lnðpf s0 Þ
Vopt ¼  4 ðrandom orientationÞ:
xsN ¼ 1. This means that the optimum volume is given by
1:69l0 Hmax MS 3
the following equation, which is plotted in the two graphs of Keff 1 
2Keff
Fig. 9:
(45)
kB T
Vopt ¼ lnðpf s0 Þ: (39) These two functions are plotted in Figs. 9(a) and 9(b) in
Keff
dashed lines.
Equations (22) and (24) are more rigorous methods to
Because the LRT is valid when n < 1, this condition is plot- calculate the coercive field and thus the optimum volume. In
ted in the graphs of Fig. 9: it appears as horizontal lines these equations, a numerical solution is required to extract
above which the LRT is no longer valid. It is evident from the volume corresponding to a given coercive field, the result
this graph that the LRT is mainly useful for strongly aniso- of which is plotted in Figs. 9(a) and 9(b) in solid lines. The dif-
tropic NPs. For instance, it is deduced from the intersection ference between the results provided by Eqs. (44) and (45) and
between Eq. (39) and the function n ¼ 1 that for MS ¼ 106 the numerical solution of Eqs. (22) and (24) can reach up to
A m1, Eq. (39) is no longer valid for MNPs with an anisot- 3 nm for the set of parameters we used, which is not negligible.
ropy below 2  105 J m3. Figures 9(a) and 9(b) give evidence that the optimum
We now consider the optimum sizes predicted by the volume obtained using SWMBTs deviates from the LRT
SWMDTs starting with the formula derived by Pfeiffer et al. results for small anisotropies, i.e., precisely in the domain
and Garcia-Otero et al. [Eqs. (19) and (21)]. First, sm must where the LRT is not valid anymore. Thus in this domain,
be replaced by an expression depending on experimental pa- SWMBTs should be used instead of LRT to calculate the op-
rameters. For reasons that will be clear later, we arbitrarily timum size. For strong anisotropies, the optimum volumes
state sm ¼ (1/2pf). Then it should be decided what the opti- given by Eqs. (44) and (45) tend toward the one deduced
mum coercive field of the MNPs is when a given magnetic from LRT because of the assumption we made; that is, that
field is applied. In the case of MNPs aligned with the field sm ¼ (1/2pf). The numerical solutions of Eqs. (22) and (24)
(/ ¼ 0) and due to the fact that the hysteresis loop is approxi- also leads to an optimum volume very close to the one pre-
mately square, dicted by the LRT for strong anisotropy; the difference
between the two predictions never exceeds 1 nm over a wide
l0 HC  l0 Hmax ð/¼ 0Þ (40)
range of parameters: f (5–500 kHz), MS (0.4–1.7  106
is taken. This leads to an optimum area Adopt given by A m1), s0 (109–1012 s), and l0Hmax (0–80 mT). Strictly
speaking, there is a zone where none of the models used is
Aopt  4l0 MS Hmax ð/¼ 0Þ: (41) valid because just above the LRT limit (when n > 1), j is not
immediately smaller than 0.7. This point will be more clearly
In the random orientation case, the magnetic field necessary evidenced in the next section. However, because the
to saturate an assembly of MNPs is approximately twice its SWMBT results approximately tend toward the LRT results
coercive field. As a consequence, a coercive field of half the at high anisotropy, it can be reasonably assumed that the
applied magnetic field could be targeted. Eq. (26) shows this transition between the two models is also acceptably repro-
would lead to an optimum area Aopt ¼ l0MSHmax. However, duced by SWMBTs.
numerical calculations show that it is better to target a coer- As a conclusion, Eqs. (44) and (45) can be used to give
cive field slightly higher than this: the increase in area due to an approximate value of the optimum size of MNPs, but the
the increase in coercive field compensates for the fact that error for weakly anisotropic MNPs can be significant. The
some of the MNPs are not switched by the applied magnetic numerical solution of Eqs. (22) and (24) can safely be used
field. The best compromise is found to depend slightly on to calculate the optimum size of MNPs for magnetic hyper-
the exact shape of the hysteresis loop and is given by thermia over a wide range of anisotropies without caring too
much about which is the most suitable model to describe
l0 HC  0:81 6 0:04l0 Hmax ðrandom orientationÞ: (42) their behavior. However, the most rigorous approach consists
of calculating n and j and using LRT when n < 1, Eqs. (22)
With this optimum coercive field, the optimum area is and (24) when j < 0.7 and numerical simulations otherwise.
Aopt  1:5660:08l0 MS Hmax ðrandom orientationÞ: (43)
B. Optimum anisotropy
Combining Eqs. (40) and (42) with Eqs. (19) and (21) allows
In the previous section, the optimum size for MNPs with
one to calculate the optimum volume Vopt. For NPs aligned
a given anisotropy was derived. However, the question of
with the magnetic field, this leads to
whether there is an optimum anisotropy was not addressed;
this important point is treated now. To solve this problem,
kB T lnðpf s0 Þ the SAR of a MNP with an optimum size was calculated ver-
Vopt ¼  2 ð/¼ 0Þ: (44)
l0 Hmax MS sus the anisotropy. The calculations were performed for
Keff 1  MS ¼ M2 ¼ 106 A m1 (magnetite value) and with the same
2Keff
values as given previously for the other parameters. To
In the random case, this leads to express the result in W/g, which is the usual unity for SARs,
083921-14 Carrey, Mehdaoui, and Respaud J. Appl. Phys. 109, 083921 (2011)

ercive fields because an optimum volume satisfying Eqs. (40)

and (42) always exists. With absolutely no size distribution,
all of the MNPs in this anisotropy range could be perfect can-
didates for magnetic hyperthermia. However, it is observed
that the NPs with a weak anisotropy are less sensitive to size
distribution effects. At the left extremity of this plateau, the
size distribution effects are canceled. For / ¼ 0, this occurs
when l0Hmax  l0HK, and for the random orientation case it
occurs when 0.81 l0Hmax  0.48 l0HK, that is, when the tar-
geted coercive field equals the low temperature coercive field
of the material. Therefore, the relation giving the optimum
anisotropy for MNPs with a known magnetization is
l0 Hmax MS
Kopt ¼ C ; (46)
FIG. 10. (Color online) Calculation of the SAR as a function of the anisot- 2
ropy for NPs with the optimum volume (solid lines) and for NPs with a vol-
ume equal to 70% of the optimum volume (dashed lines). The parameters are with C ¼ 1 for / ¼ 0 and C ¼ 1.69 for the random orientation
identical to those used in the previous figure with MS ¼ M2 ¼ 106 A m1 and case
q ¼ 5.2  103 kg m3. For strongly anisotropic NPs, the LRT results are pro- At the left of this plateau, a decrease of the SAR is
vided using Eqs. (36), (38), and (39). For weakly anisotropic NPs, SWMBTs
results are provided using Eqs. (19), (21), (25), and (40)–(45). The vertical observed. This occurs when the anisotropy field of the NP is
dotted lines show the optimum anisotropy for this set of parameters. so weak that there is no solution to Eqs. (40) and (42). In this
case, the coercive field of the NPs is l0HK for / ¼ 0 and
a density of q ¼ 5.2  106 kg.m3 (magnetite value) is 0.48 HK for a random orientation. Then the SAR calculated
assumed. The results of the LRT for the / ¼ 0 and random using Eqs. (25) and (26) leads to the observed decrease.
orientation case were obtained using Eqs. (36), (38), and At the optimum anisotropy, the optimum volume of the
(39). They are plotted in Fig. 10 as solid lines. The data are MNPs diverges and tends toward infinity [see Figs. 9(a) and
not plotted outside the domain of validity of the LRT, i.e., 9(b)]. This means that, in principle, a large single-domain
when n > 1. The calculations in the ferromagnetic regime MNP with the optimum anisotropy would be the perfect
were performed for simplicity without numerical solutions object. However, increasing the size of the NPs too much
using Eqs. (19), (21), and (40)–(45). This simplification does leads to several problems: i) It leads to the transition toward
not change the main conclusions of this part. Data are plotted multidomain NPs. They might be interesting objects for
in Fig. 10 as solid lines and are not plotted outside the do- magnetic hyperthermia because the coercive field value is
main of validity of the SWMBTs, that is, when j > 0.7. also influenced by the size near this transition (see Fig. 8).26
A very important point to consider in a discussion of the However, there is no simple way to calculate the optimum
optimum parameters for magnetic hyperthermia is the influ- size and the hysteresis area for such nano-objects. ii) Dis-
ence of the size distribution on the final SAR value. To illus- persing and stabilizing NPs in a colloidal solution is all the
trate this, the SAR value for a MNP with a volume 30% more difficult if they have a large diameter. iii) A larger size
below the optimum volume was calculated in each case. The favors recognition by the phagocytosis system after intrave-
results illustrate the loss of SAR due to the size distribution nous administration.
of MNPs. The results are plotted as dashed lines along with The optimum volume of the MNPs and the optimum an-
the previous data. isotropy thus result from a compromise between the effi-
Figure 10 displays an essential result for magnetic ciency of magnetic hyperthermia, which requires large NPs
hyperthermia and is worth a detailed comment beginning with an anisotropy given by Eq. (46), and other factors for
with the high anisotropy MNPs described by the LRT. First, which a large size could be detrimental. If for any reason the
the LRT shows that the maximum achievable SAR increases volume of the MNPs needs to be limited to a maximum
with a reduction in anisotropy. This is obvious from Eqs. value, the equations used so far allow one to deduce easily
(36)–(39): for a MNP with the optimum size, the resonating what would be the optimum anisotropy for a given volume.
term ð1þxxsR In any case, Eq. (46) provides the approximate target anisot-
2 s2 Þ is maximal and always equals 1/2. Because
R ropy to optimize the SAR of MNPs.
A ! V [see Eqs. (36)–(38)] and because the optimum volume
is inversely proportional to the anisotropy [see Eq. (39)], the
C. Optimum materials
maximal SAR value increases with decreasing anisotropy. In
this regime, the effect of the size distribution is dramatic: the In Fig. 11, Eq. (46) is plotted along with bulk parameters
SAR decreases by more than one order of magnitude for of several magnetic materials with no consideration of their
nonoptimum NPs. toxicity. It is emphasized that the magnetic anisotropy in
After passing the blank space to the left of these curves MNPs is generally larger than that of the bulk because of sur-
where the transition between the two regimes is out of face effects and/or stoichiometry problems (in alloys and
the domain of validity of both models used here, the SAR of oxides). In the iron oxide family, magnetite NPs displaying
the optimum NPs displays a plateau: in this region, the SAR the bulk anisotropy would be ideal candidates. Because the
can be maximized by tuning the NP volume to adjust their co- hysteresis area is directly proportional to MS, metallic
083921-15 Carrey, Mehdaoui, and Respaud J. Appl. Phys. 109, 083921 (2011)

includes the Néel and Brownian relaxation times, is then

defined as
1 1 1
¼ þ : (48)
sR sN sB

Here again, it must be kept in mind that the LRT including

Brownian motion has a restricted domain of validity, which
is discussed in detail in Ref. 11: the LRT is valid for small
magnetic fields, i.e., for n < x with x depending on the value
of xsB/n. For any value of the parameters, numerical results
have been obtained by Raiker et al. with the restrictive hy-
pothesis that the relaxation by magnetization reversal is not
possible.11 The hysteresis loop of a MNP in a magnetic fluid
when both magnetization reversal and Brownian motion are
allowed has to our knowledge not been solved in the most
general case.
FIG. 11. (Color online) Comparison between theoretical optimum parame-
ters and experimental bulk parameters of several magnetic materials. Equa-
tion (46) is plotted for l0Hmax ¼ 20 mT (dotted lines) and 50 mT (dashed 2. Optimum volume calculation
lines). In each case, the upper curve represents the random orientation case,
and the lower one represents the / ¼ 0 case. Label (a): for the FeCo alloy, In a situation where only reversal by Brownian motion
the anisotropy of 15 nm MNPs estimated in Ref. 19 is provided. Reprinted would occur, the LRT indicates that the optimum volume is
with permission from L.-M. Lacroix, R. Bel-Malaki, J. Carrey, S. Lachaize,
G. F. Goya, B. Chaudret, and M. Respaud, J. Appl. Phys. 105, 023911
the one for which xsB ¼ 1. For an applied frequency of 100
(2009). Copyright V C 2009 American Institute of Physics. kHz and the viscosity of water, this leads to an optimum ra-
dius of around 15.5 nm independently of any magnetic pa-
materials with high magnetization are required to reach the rameters. However, there are some ranges of parameters
highest SARs. In this case, FeCo alloys would be perfect, but (MNP magnetization, magnetic field amplitude, and fre-
their probable toxicity could be a severe problem. Iron, quency) for which the optimum size given by this simple
which is not intrinsically toxic, could be a good candidate equation is not correct because it is out of the domain of va-
but presents in its crystalline form an anisotropy value too lidity of the LRT; the only way to know it is to use numerical
large for 20 mT applications. Fe1-xSix alloys have both a calculations.11 In the case of relaxation by Brownian motion,
reduced anisotropy and magnetization and could represent there is so far no equivalent to the simple analytical formula
an interesting compromise. Also amorphous iron should dis- we used in the case of relaxation by magnetization reversal,
play a reduced anisotropy. However, the possibility of creat- which would be valid over a wide range of parameters. The-
ing MNPs with a reduced anisotropy using amorphous iron oretical progress is required on this point.
or Fe1-xSix alloys still needs to be demonstrated.
3. Why Brownian motion is not the way to optimize
D. The Brownian motion thermal effects

1. Influence of Brownian motion on hyperthermia Progress in the development of analytical formulas and/
properties or the use of numerical simulations could lead to progress in
the determination of the optimum parameters of MNP heat-
When MNPs are in a fluid, they can rotate physically
ing through their Brownian motion. One can imagine that
under the influence of the magnetic fluid, similarly to a com-
eventually MNPs with such optimum magnetic properties
pass, until the magnetization is aligned with the magnetic
could be synthesized. However, we think that this method is
field. This is known as relaxation by Brownian motion. In a
less promising than the one consisting of optimizing the
standard hyperthermia experiment, the relaxation by Brown-
SAR and using heating by magnetization reversal. The main
ian motion and the relaxation by magnetization reversal
reason is the following: the Brownian motion depends
described in the preceding text are both possible, which leads
strongly on the environment and the aggregation state of the
to a global hysteresis loop resulting from the two mecha-
MNPs because the hydrodynamic volume of the MNP is the
nisms. Whether the relaxation occurs only by Brownian
main parameter governing the Brownian motion. Thus two
motion or by both mechanisms, the heating during one cycle
aggregated MNPs or a MNP functionalized and linked to a
still simply equals the hysteresis loop area A. The influence
tumor cell would display SAR values very different to that
of Brownian motion can be easily incorporated into the
of a free MNP. Moreover, the application of an alternating
LRT.10 A Brownian relaxation time is defined as
magnetic field can lead to the formation of chains or columns
3gVH of MNPs with very different Brownian properties.27,28
sB ¼ ; (47) We think that the ideal objects for magnetic hyperthermia
kB T
should display a SAR that is as independent as possible from
where g is the viscosity of the solvent and VH is the hydrody- such phenomena. This requires that the physical rotation of
namic volume of the MNPs. The relaxation time sR, which the MNP in the alternating field is blocked for all MNPs,
083921-16 Carrey, Mehdaoui, and Respaud J. Appl. Phys. 109, 083921 (2011)

TABLE I. Summary of experimental results on various materials of interest. The rS column gives the bulk magnetization per unit mass at 300 K. l0Hmax is
the magnetic field at which the experiments were conducted. Amax gives the theoretical maximum hysteresis area that could have been measured, which was
calculated using Eq. (50). Aexp gives the hysteresis area experimentally measured in these conditions. a is calculated using Eq. (51). The rows labeled
“magnetosomes” correspond to iron oxide NPs synthesized by bacteria. NPs were (a) randomly oriented or (b) aligned with the magnetic field.

Materials rS (300 K) (A m2/kg) l0Hmax (mT) Amax (mJ/g) Aexp (mJ/g) a

FeOx 100 13.8 5 (Ref. 30) 1.5 0.3

Magnetosomes (a) 100 12.5 5 (Ref. 31) 1.3 0.26
Magnetosomes (b) 100 12.5 5 (Ref. 31) 2.3 0.46
Co 162 31.2 20.6 (Ref. 32) 3.25 0.16
FeCo 240 29 27.8 (Ref. 19) 1.5 0.054
Fe 218 66 57.5 (Ref. 33) 5.6 0.097
CoFe2O4 75 31.1 9.35 (Ref. 34) 0.63 0.067

which should thus have a large hydrodynamic diameter. This the high degree of optimization that has already been
would ensure that individual free MNPs and MNPs that are a achieved in the magnetic hyperthermia properties of iron
part of aggregates have the same heating properties. Coating oxide nanoparticles. Because the a value is already nearly
the magnetic core using PEG or dextran layers to ensure the optimal for randomly oriented nanoparticles, further
bioavailability and targeting of MNP is already a process improvements will necessarily imply that the MNPs are ori-
tending toward this goal. ented by the magnetic field during hyperthermia experi-
However, it is not completely impossible that MNPs ments. This table also shows that MNPs composed of high
moving inside a tumor under the influence of an alternating magnetization materials have not yet reach such a degree
magnetic field might cause other damage to the cells of optimization, and much higher SAR values could be
than simply that resulting from the increase of temperature reached by a better control of the nanoparticles’ size, ani-
only. sotropy, dispersion, and magnetization values.

E. Comparison with experimental results

IV. CONCLUSION
In this last subsection, we will present a short summary
This article presents a rigorous approach to the calcula-
of experimental results on a few selected materials. A com-
tion of the hysteresis area of a single-domain in the macro-
parison between theoretical maximum SARs and experimen-
spin approximation, including a detailed study of the validity
tally measured SARs will be done. A more detailed review
of the analytical expressions. The conclusions about the
on experiments can be found in Ref. 29.
applications of such nanoparticles for magnetic hyperthermia
The maximum hysteresis area Amax that can be obtained
are clear and simple in this framework. We hope these con-
in a hyperthermia experiment is
clusions will guide experimentalists both in the synthesis of
Amax ¼ 4l0 Hmax MS : (49) high-quality materials and in the analysis of experimental
data. An important conclusion of this study is that the anisot-
If the hysteresis area is expressed in energy per unit of mass, ropy of the synthesized MNPs is a key parameter to under-
which is usual in hyperthermia, MS should be replaced in stand and tune the magnetic hyperthermia properties. Often
this equation by the saturation magnetization per unit of neglected, it should, on the contrary, become central in ex-
mass rS, which leads to perimental articles on magnetic hyperthermia.
However, it should not be overlooked that several im-
Amax ¼ 4l0 Hmax rS : (50) portant points have not been treated in this article and have
a strong influence for the required application. i) At room
An interesting way to represent the experimentally measured
temperature and high frequency, what is the domain of va-
area Aexp is
lidity of the macrospin approximation and thus of the
Aexp ¼ 4al0 Hmax rS : (51) SWMBTs used to predict hyperthermia properties? ii) To
what extent can the single-domain/multidomain transition
In this formulation, a is a dimensionless parameter that be predicted and used to tune the coercive fields of mag-
characterizes the relative area of the hysteresis loop with netic nanoparticles to maximize hyperthermia? Is the shape
respect to the ideal square. a ¼ 1 for a perfectly optimized and area of the hysteresis loop obtained in this case interest-
system with the easy axes of all NPs aligned with the mag- ing for magnetic hyperthermia applications? iii) How are
netic field; a ¼ 0.39 for an optimized system with a random the results for the optimum parameters modified when tak-
orientation of the easy axes [see Eq. (43)]. In a sense, a ing into account the magnetic interactions, which are known
represents the degree of optimization of a given system. In to have a deep influence on magnetic properties both in the
Table I, a is calculated from experimental results obtained superparamagnetic and the ferromagnetic regimes? We
on various materials of interest. For each material, the hope that these questions will stimulate theoretical work on
highest value in the literature was chosen. This table shows this subject.
083921-17 Carrey, Mehdaoui, and Respaud J. Appl. Phys. 109, 083921 (2011)

18
ACKNOWLEDGMENTS Y. P. Kalmykov, J. Appl. Phys. 96, 1138 (2004).
19
L.-M. Lacroix, R. Bel-Malaki, J. Carrey, S. Lachaize, G. F. Goya, B.
We acknowledge L.-M. Lacroix, S. Lachaize, and Y. Chaudret, and M. Respaud, J. Appl. Phys. 105, 023911 (2009).
20
Raikher for fruitful discussions. This work was supported by R. Hiergeist, W. Andrä, N. Buske, R. Hergt, I. Hilger, U. Richter, and W.
Kaiser, J. Magn. Magn. Mater. 201, 420 (1999).
the InNaBioSanté Foundation. 21
X. Wang, H. Gub, and Z. Yang, J. Magn. Magn. Mater. 293, 334 (2005).
22
B. Mehdaoui, A. Meffre, L.-M. Lacroix, J. Carrey, S. Lachaize, M. Respaud,
1
R. Hergt and S. Dutz, J. Magn. Magn. Mater. 311, 187 (2007). M. Gougeon, and B. Chaudret, J. Appl. Phys. 107, 09A324 (2010)
2 23
E. C. Stoner and E. P. Wohlfarth, Philos. Trans. R. Soc. London, Ser. A E. Snoeck, C. Gatel, L. M. Lacroix, T. Blon, S. Lachaize, J. Carrey, M.
240, 599 (1948). Respaud, and B. Chaudret, Nanoletters 8, 4293 (2008)
3 24
J. J. Lu, H. L. Huang, and I. Klik, J. Appl. Phys. 76, 1726 (1994). B. D. Cullity and C. D. Graham, Introduction to Magnetic Materials, 2nd
4
H. Pfeiffer, Phys. Status Solidi A 120, 233 (1990) ed. (Wiley, Hokoben, 2009), p. 321.
5 25
J. Garcia-Otero, A. J. Garcia-Bastida, and J. Rivas, J. Magn. Magn. Mater. A. Jordan, R. Scholz, K. Maier-Hauff, M. Johannsen, P. Wust, J. Nadobny,
189, 377 (1998). H. Schirra, H. Schmidt, S. Deger, S. Loening, W. Lanksch, and R. Felix, J.
6
D. Kechrakos and K. N. Trohidou, Phys. Rev. B 58, 12169 (1998). Magn. Magn. Mater. 225, 118 (2001).
7 26
Y. L. Raikher, V. I. Stepanov, and R. Perzynski, Physica B 343, 262 M. Andrés Vergés, R. Costo, A. G. Roca, J. F. Marco, G. F. Goya, C. J.
(2004). Serna, and M. P. Morales, J. Phys. D:Appl. Phys. 41, 134003 (2008).
8 27
N. A. Usov and Y. B. Grebenshchikov, J. Appl. Phys. 106, 023917 M. Klokkenburg, B. H. Erné, J. D. Meeldijk, A. Wiedenmann, A. V.
(2009). Petukhov, R. P. A. Dullens, and A. P. Philipse, Phys. Rev. Lett. 97,
9
R. Hergt, W. Andrä, C. G. d’Ambly, I. Hilger, W. A. Kaiser, U. Richter, 185702 (2006).
28
and H.-G. Schmidt, IEEE Trans. Magn. 34, 3745 (1998). B. Mehdaoui, A. Meffre, L.-M. Lacroix, J. Carrey, S. Lachaize, M. Gou-
10
R. E. Rosensweig, J. Magn. Magn. Mater. 252, 370 (2002). jeon, M. Respaud, and B. Chaudret, J. Magn. Magn. Mat. 322, L49 (2010).
11 29
Y. L. Raikher and V. I. Stepanov, J. Magn. Magn. Mater 320, 2692 L.-M. Lacroix, D. Ho, and S. Sun, Curr. Top. Med. Chem. 10, 1184 (2010)
30
(2008). R. Hergt, R. Hiergeist, I. Hilger, W.A. Kaiser, Y. Lapatnikov, S. Margel,
12
R. Hergt, S. Dutz and M. Röder, J. Phys. Condens. Matter. 20, 385214 and U. Richter, J. Magn. Magn. Mater. 270, 345 (2004).
31
(2008). R. Hergt, R. Hiergeist, M. Zeisberger, D. Schüler, U. Heyen, I. Hilger, and
13
R. Hergt, S. Dutz and M. Zeisberger, Nanotechnology 21, 015706 (2010). W. A. Kaiser, J. Magn. Magn. Mater. 293, 80 (2005).
14 32
N. A. Usov, J. Appl. Phys. 107, 123909 (2010). M. Zeisberger S. Dutz, R. Müller, R. Hergt, N. Matoussevitch, and H.
15
M. Respaud, J. Appl. Phys. 86, 556 (1999). Bönnemann, J. Magn. Magn. Mater. 311, 224 (2007).
16 33
C Tannous and J Gieraltowski, Eur. J. Phys. 29, 475-487 (2008) B. Mehdaoui, A. Meffre, L.-M. Lacroix, J. Carrey, S. Lachaize, M. Gou-
17
W. T. Coffey, D. S. F. Crothers, J. L. Dormann, Yu. P. Kalmykov, E. C. jeon, M. Respaud, and B. Chaudret, J. Magn. Magn. Mat. 322, L49 (2010).
34
Kennedy, and W. Wernsdorfer, Phys. Rev. Lett. 80, 5655 (1998). J.-P. Fortin, F. Gazeau, and C. Wilhelm, Eur. Biophys. J. 37, 223 (2008)