Ferrofluids. Magnetically Controllable Fluids and Their Applications S. Odenbach

Lecture Notes in Physics
Editorial Board
R. Beig, Wien, Austria
B.-G. Englert, Ismaning, Germany
U. Frisch, Nice, France
P. Hänggi, Augsburg, Germany
K. Hepp, Zürich, Switzerland
W. Hillebrandt, Garching, Germany
D. Imboden, Zürich, Switzerland
R. L. Jaffe, Cambridge, MA, USA
R. Lipowsky, Golm, Germany
H. v. Löhneysen, Karlsruhe, Germany
I. Ojima, Kyoto, Japan
D. Sornette, Nice, France, and Los Angeles, CA, USA
S. Theisen, Golm, Germany
W. Weise, Trento, Italy, and Garching, Germany
J. Wess, München, Germany
J. Zittartz, Köln, Germany
3
Berlin
Heidelberg
New York
Barcelona
Hong Kong
London
Milan
Paris
Tokyo
Editorial Policy
The series Lecture Notes in Physics (LNP), founded in 1969, reports new developments in
physics research and teaching -- quickly, informally but with a high quality. Manuscripts
to be considered for publication are topical volumes consisting of a limited number of
contributions, carefully edited and closely related to each other. Each contribution should
contain at least partly original and previously unpublished material, be written in a clear,
pedagogical style and aimed at a broader readership, especially graduate students and
nonspecialist researchers wishing to familiarize themselves with the topic concerned. For
this reason, traditional proceedings cannot be considered for this series, though volumes
to appear in this series are often based on material presented at conferences, workshops
and schools (in exceptional cases the original papers and/or those not included in the
printed book may be added on an accompanying CD-ROM, together with the abstracts
of posters and other material suitable for publication, e.g. large tables, colour pictures,
program codes, etc.).
Acceptance
A project can only be accepted tentatively for publication, by both the editorial board and the
publisher, following thorough examination of the material submitted. The book proposal
sent to the publisher should consist of at least a preliminary table of contents outlining the
structure of the book together with abstracts of all contributions to be included.
Final acceptance is issued by the series editor in charge, in consultation with the publisher,
only after receiving the complete manuscript. Final acceptance, possibly requiring minor
corrections, usually follows the tentative acceptance unless the final manuscript differs
significantly from expectations (project outline). In particular, the series editors are entitled
to reject individual contributions if they do not meet the high quality standards of this
series. The final manuscript must be camera-ready, and should include both an informative
introduction and a sufficiently detailed subject index.
Contractual Aspects
Publication in LNP is free of charge. There is no formal contract, no royalties are paid,
and no bulk orders are required, although special discounts are offered in this case. The
volume editors receive jointly 30 free copies for their personal use and are entitled, as are the
contributing authors, to purchase Springer books at a reduced rate. The publisher secures
the copyright for each volume. As a rule, no reprints of individual contributions can be
supplied.
Manuscript Submission
The manuscript in its final and approved version must be submitted in camera-ready form.
The corresponding electronic source files are also required for the production process, in
particular the online version. Technical assistance in compiling the final manuscript can be
provided by the publisher’s production editor(s), especially with regard to the publisher’s
own LaTeX macro package which has been specially designed for this series.
Online Version/ LNP Homepage

LNP homepage (list of available titles, aims and scope, editorial contacts etc.):
http://www.springer.de/phys/books/lnpp/
LNP online (abstracts, full-texts, subscriptions etc.):
http://link.springer.de/series/lnpp/
S. Odenbach (Ed.)
Ferrofluids
Magnetically Controllable Fluids
and Their Applications
13
Editor
Stefan Odenbach
ZARM
Universität Bremen
Am Fallturm
28359 Bremen, Germany
Cover picture: by Dr. S. Odenbach
Library of Congress Cataloging-in-Publication Data applied for.
Die Deutsche Bibliothek - CIP-Einheitsaufnahme
Ferrofluids : magnetically controllable fluids and their applications / S.

Odenbach (ed.). - Berlin ; Heidelberg ; New York ; Barcelona ; Hong Kong ;
London ; Milan ; Paris ; Tokyo : Springer, 2002
(Lecture notes in physics ; 594)
(Physics and astronomy online library)
ISBN 3-540-43978-1
ISSN 0075-8450
ISBN 3-540-43978-1 Springer-Verlag Berlin Heidelberg New York
This work is subject to copyright. All rights are reserved, whether the whole or part of the
material is concerned, specifically the rights of translation, reprinting, reuse of illustra-
tions, recitation, broadcasting, reproduction on microfilm or in any other way, and
storage in data banks. Duplication of this publication or parts thereof is permitted only
under the provisions of the German Copyright Law of September 9, 1965, in its current
version, and permission for use must always be obtained from Springer-Verlag. Violations
are liable for prosecution under the German Copyright Law.
Springer-Verlag Berlin Heidelberg New York
a member of BertelsmannSpringer Science+Business Media GmbH
http://www.springer.de

c Springer-Verlag Berlin Heidelberg 2002
Printed in Germany
The use of general descriptive names, registered names, trademarks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt
from the relevant protective laws and regulations and therefore free for general use.
Typesetting: Camera-ready by the authors/editor
Camera-data conversion by Steingraeber Satztechnik GmbH Heidelberg
Cover design: design & production, Heidelberg
Printed on acid-free paper
SPIN: 10886791 54/3141/du - 5 4 3 2 1 0
Preface
Magnetic control of the properties and the flow of liquids can be a challenging
field for basic research as well as for applications. An important condition for a
technically interesting magnetic control of fluids is that the requested magnetic
fields should be as small as possible.
An excellent material fulfilling this condition are suspensions of magnetic
nanoparticles - commonly called ferrofluids. These fluids exhibit coupled liquid
and superparamagnetic properties, leading to the possibility to control their
flows and properties with magnetic fields having a strength of about 10mT. After
the first synthesis of stable ferrofluids in 1964, a rapid development of ideas for
applications as well as of basic fluid mechanics investigations took place. Till now
the research field experiences a strong ongoing growth, recently triggered by the
establishment of a national research program on magnetic fluids in Germany and
documented by numerous presentations of the 9th International Conference on
Magnetic Fluids (ICMF9), held in Bremen in 2001.
One of the most challenging aspects of ferrofluid research is the interdisci-
plinarity of the field. Besides chemistry for the preparation of magnetic fluids,
basic theoretical physics for the description of their properties and behavior, fluid
physics and rheology for the investigation of flows and rheological properties un-
der influence of magnetic fields are needed to cover the basic research interests.
In addition engineering and medical applications contribute to the importance
of ferrofluid research for everyday life.
Most of these aspects are covered in this issue of Lecture Notes in Physics,
which is mainly based on a series of plenary talks given during ICMF9. Starting
with a review of preparation techniques for ferrofluids the chemical aspect of pro-
duction of stable magnetic colloids is highlighted, followed by two contributions
showing different approaches for a quantitative characterization of ferrofluids.
Part two of the issue is devoted to basic questions concerning the behavior
of ferrofluids in the presence of magnetic fields. Three contributions highlight
different ways to describe the fluids theoretically. Macro- and microscopic ap-
proaches are shown together to enable a comparison of their advantages and
problems. The section ends with a specific example of magnetic control - the
influence of fields on heat and mass transfer in ferrofluids.
In the third part, the question of magnetic control of the fluids properties is
addressed using the example of magnetoviscosity - i.e. the change of the viscous
properties of magnetic fluids in a magnetic field. The combination of theoret-
VI Preface
ical description and experimental investigations documents the complexity of

the problem and gives a good overview on the actual state of knowledge in
these problems. The part ends with a contribution on magnetorheological fluids,
building the bridge to the application possibilities of magnetoviscous effects.
Finally, part four deals with question concerning application of magnetic
fluids. Since the major progress in this respect is actually concentrated in the field
of medical applications, a single contribution on cancer therapy using ferrofluids
as a drug carrier is included in this issue.
In total, this issue of Lecture Notes in Physics provides the reader with all
basic knowledge needed to enter the field of ferrofluid research as well as with
the most recent developments achieved.
Bremen, May 2002 Stefan Odenbach

Table of Contents
Part I Synthesis and Characterization
The Preparation of Magnetic Fluids

S.W. Charles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Magnetic Spectroscopy
as an Aide in Understanding Magnetic Fluids
P.C. Fannin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Magnetic and Crystalline Nanostructures in Ferrofluids

as Probed by Small Angle Neutron Scattering
A. Wiedenmann . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
Part II Basic Theory
Basic Equations for Magnetic Fluids with Internal Rotations

R.E. Rosensweig . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Ferrohydrodynamics: Retrospective and Issues
M.I. Shliomis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Ferrofluid Dynamics
H.W. Müller and M. Liu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
Heat and Mass Transfer Phenomena

E. Blums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
Part III Rheological Properties
Statistical Physics of Non-dilute Ferrofluids

A. Zubarev . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
Magnetic Fluid as an Assembly of Flexible Chains

K.I. Morozov and M.I. Shliomis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
Magnetoviscous Effects in Ferrofluids

S. Odenbach and S. Thurm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
VIII Table of Contents
Magnetorheology: Fluids, Structures and Rheology

G. Bossis, O. Volkova, S. Lacis, and A. Meunier . . . . . . . . . . . . . . . . . . . . . . . 202
Part IV Applications
Targeted Tumor Therapy with “Magnetic Drug Targeting”:

Therapeutic Efficacy of Ferrofluid Bound Mitoxantrone
Ch. Alexiou, R. Schmid, R. Jurgons, Ch. Bergemann,
F.G. Parak, and W. Arnold . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
Magnetic Unit System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255

List of Contributors
Christoph Alexiou Stuart W. Charles

Department of Otorhinolaryngology Department of Chemistry
Head and Neck Surgery, University of Wales
Klinikum rechts der Isar Bangor Gwynedd LL57 2UW
Technical University of Munich UK
81675 Munich [email protected]
[email protected]
Paul C. Fannin
Department of Electronic and
W. Arnold
Electrical Engineering
Physics-Department E 17
University of Dublin
Technical University of Munich
Trinity College
81675 Munich
Dublin 2
Germany Ireland
[email protected]
Christian Bergemann
Chemicell Peter Hulin
10777 Berlin Department of Otorhinolaryngology
[email protected] Head and Neck Surgery,
Klinikum rechts der Isar
Elmars Blums Technical University of Munich
Institute of Physics 81675 Munich
[email protected]
University of Latvia
LV-2169, Salaspils
R. Jurgons
Latvia
Department of Otorhinolaryngology
[email protected]
Head and Neck Surgery,
Georges Bossis Technical University of Munich
LPMC 81675 Munich
UMR 6622
University of Nice Sandris Lacis
Parc Valrose Department of Physics
06108 Nice Cedex 2 University of Latvia
France Zellu str.8
[email protected] LV-1586
X List of Contributors
Riga Fritz G. Parak

Latvia Physics-Department E 17
[email protected] Technical University of Munich
81675 Munich
Mario Liu Germany
Institut für Theoretische Physik [email protected]
Universität Tübingen
D-72676 Tübingen Ronald E. Rosensweig
Germany Consultant
[email protected] Summit, NJ
U.S.A.
[email protected]
Alain Meunier
LPMC
UMR 6622 Roswitha Schmid
University of Nice Department of Otorhinolaryngology
Parc Valrose Head and Neck Surgery,
06108 Nice Cedex 2 Klinikum rechts der Isar
France
81675 Munich
[email protected]
[email protected]
Konstantin I. Morozov
A. Schmidt
Institute of Continuous Media
Department of Otorhinolaryngology
Mechanics
Head and Neck Surgery,
UB of Russian Academy of Sciences
614013 Perm
Russia 81675 Munich
[email protected]
Mark I. Shliomis
Hanns Walter Müller Department of Mechanical Engineer-
Max-Planck Institut für Polymer- ing
forschung Ben-Gurion University of the Negev
Ackermannweg 10 P.O.B. 653
D-55128 Mainz Beer-Sheva 84105
Germany Israel
[email protected] [email protected]
Stefan Odenbach Steffen Thurm

ZARM ZARM
University of Bremen University of Bremen
Am Fallturm Am Fallturm
D-28359 Bremen D-28359 Bremen
Germany Germany
[email protected] [email protected]
List of Contributors XI
Olga Volkova D-14109 Berlin

LPMC Germany
UMR 6622 [email protected]
University of Nice
Parc Valrose
Andrey Yu. Zubarev
06108 Nice Cedex 2
Department of Mathematical Physics
France
Ural State University
[email protected]
Lenin Av., 51
Albrecht Wiedenmann 620083 Ekaterinburg
Hahn-Meitner-Institut Berlin Russia
Glienicker Strasse 100 [email protected]
The Preparation of Magnetic Fluids
Stuart W. Charles
Department of Chemistry, University of Wales, Bangor, Gwynedd LL57 2UW, UK

Liquids Research Limited, Mentec, Deiniol Road, Bangor, Gwynedd LL57 2UP, UK
Abstract. In any discussion of fluids, which have magnetic properties, it is convenient

to divide them into the following categories, (A) ferrofluids; (B) magnetorheological
fluids; (C) dispersions of micron-sized particles of a non-magnetic material containing
magnetic nano-sized particles, and (D) fluids containing paramagnetic particles.
In this Chapter, discussion will be confined to categories (A) and (B). The prepa-
ration of particles, ferrofluids and magnetorheological fluids has been the subject of
numerous patents and research publications. A comprehensive bibliography of this in-
formation can be found updated in each of Proceedings of the International Conference
on Magnetic Fluids (ICMF), under the following authors Zahn et al. [1], Charles et al.
[2], Kamiyama et al [3], Blums et al. [4], Cabuil et al. [5] , Bhatnagar et al. [6], Vekas
et al. [7]. Reviews of the subject have been given by Rosensweig [8,9,10], Charles et al.
[11], Martinet [12] and Scholten [13].1
1 Ferrofluids
Ferrofluids are stable colloidal dispersions of nano-sized particles of ferro- or fer-
rimagnetic particles in a carrier liquid. A wide range of carrier liquids have been
employed, and many ferrofluids are commercially available to satisfy particular
applications, e.g. in rotary vacuum lead-throughs, it is essential that the carrier
liquid has a very low vapor pressure. In other applications, temperature, either
high, low or both, may be a critical consideration. Theoretically it should be
possible to produce dispersions in any liquid thereby being able to tailor the
requirements of viscosity, surface tension, temperature and oxidative stability,
vapor pressure, stability in hostile environments, etc., to the particular applica-
tion in mind, whether it be technological or biomedical.
When a ferrofluid is subjected to a magnetic field, magnetic field gradient
and/or gravitational field, in order that the colloidal suspension remains stable
the magnetic particles generally have to be of approximately 10 nm in diam-
eter. Particles of this size, whether they be ferrite or metal, possess a single
magnetic domain only, i.e., the individual particles are in a permanent state
of saturation magnetization. Thus a strong long-range magnetostatic attraction
exists between individual particles, the result of which would lead to agglomer-
ation of the particles and subsequent sedimentation unless a means of achieving
1
For further information and availability of Ferrofluids and Magnetorheological Flu-
ids the reader is recommended to view www.liquidsresearch.com or contact Liquids
Research Limited by e.mail [email protected]
Stefan Odenbach (Ed.): LNP 594, pp. 3–18, 2002.

4 S.W. Charles
a repulsive interaction can be incorporated. In order to achieve this repulsive

mechanism, the particles can either be coated by a surfactant (surface active
material) to produce an entropic repulsion, or the surface of the particles can be
charged thereby producing an electrostatic repulsion. For dispersions in liquid
metals, stability has not been achieved due to the lack of a method to produce
a repulsive mechanism.
To discuss the preparation of ferrofluids it is convenient to divide the dis-
cussion into (1) the preparation of nano-sized magnetic particles and (2) the
subsequent dispersion of such particles in various liquids.
2 Preparation of Nano-sized Magnetic Particles

2.1 Ferrite Particles
In most technological and biomedical applications of ferrofluids, the magnetic

material may be one of a number of different ferrites. By far the most commonly
used ferrites are magnetite (Fe3 O4 ) and maghemite (γ-Fe2 O3 ). Because mag-
netite can be oxidized to maghemite with only a relatively small reduction in
moment, the actual structure of the particles in commercial and other ferroflu-
ids usually involves the presence of both ferrites in an undefined ratio. Detailed
reviews of the structure and magnetic properties of the normal and inverse spin
structures of ferrites have been given [14,15,16]. The name spinel is given to those
ferrites which have the formula MFe2 O4 (where M is a divalent ion) and which
have the cubic crystal structure of the mineral spinel (MgAl2 O4 ). The oxygen
atoms are arranged in layers in such a way that there are two types of interstitial
sites, tetrahedral (A) sites and octahedral (B) sites. The net magnetic moment
of each ferrite is determined by the moment of each cation, the arrangement of
the cations in the A and B sites, and the interaction between cations. In mag-
netite, the moment of the two Fe3+ ions are split between an A site and a B site
and are antiferromagnetically coupled so that the moments cancel, whereas the
Fe2+ ion situated on the B site gives rise to the overall moment. This opposite
and unequal arrangement of the moments gives rise to ferrimagnetism.
A discussion of the various ways of producing nano-sized ferrite particles now
follows.
Wet-grinding. The original method of producing ferrofluids based on ferrites

is attributed to [17]. The method involved wet-grinding ferrites in a ball-mill
in the presence of a suitable surfactant until the ferrite is in a colloidal state.
Centrifugation was usually employed to remove larger particles which could lead
to agglomeration and sedimentation. However this process usually takes a very
long time (1000 hours) and it is mainly for this reason that the process has been
superceded by a rapid and simple method involving the co-precipitation of metal
salts in aqueous solution using a base.
The Preparation of Magnetic Fluids 5
Co-precipitation method. The co-precipitation method [18] for the prepara-

tion of particles of magnetite, maghemite and substituted ferrites suitable for use
in ferrofluids has been the subject of many patents and publications as given in
the comprehensive bibliography referred to in the Introduction. This method is
usually carried out between 0 and 100o C, and as a result the cation distribution
on the A and B sites may not conform to the distribution in bulk annealed crys-
tals of the same material. Nevertheless it is well established by X-ray diffraction
that the particles formed are crystalline. Further, their magnetic characteristics
with regard to magnetic crystalline anisotropy and Curie temperature are not
that far removed from that of the bulk crystals.
Magnetite and maghemite particles
The co-precipitation process is an extremely versatile method of producing
ferrite particles in that particles of different size (3-20 nm diameter) and magnetic
properties may be prepared by simply controlling the experimental conditions.
It has been shown [19,5] that control of the mole fraction ratio of Fe3+ :Fe2+ ,
their concentrations, nature and concentration of the alkali medium can enable
particles to be prepared of the desired size. This work has been extended [20] to
include studies of the effect of precipitation temperature on particle size, the ef-
fect of heating the precipitate in the alkaline medium, and the effect of addition
of surfactant to the reaction mixture. They have shown that the particle size
increases when uncoated particles are heated in the alkaline medium and that
the extent of this increase depends on the temperature to which the particles are
heated. For an increase in precipitation temperature (between 2 and 90o C) and
without subsequent heating in the alkali media an increase in particle size is ob-
served. The effect of changing the initial Fe3+ :Fe2+ molar ratio from 2:1 to 1.5:1
and also changing the bases (KOH, NaOH and NH4 OH) have been shown to
have an effect on the particle size. In the case of magnetite, particles can be pre-
pared from the co-precipitation of hydroxides from an aqueous solution of Fe3+
and Fe2+ in the mole ratio of approximately 2:1 using a base [18]. The reaction
is complex and involves the conversion of the hydroxide particles to magnetite.
Oxidation of Fe2+ leads to the stoichiometry of the particles not being purely
magnetite which can be circumvented by adjusting the ratio of the Fe2+ and
Fe3+ concentrations. Thus the particles produced may be effectively a mixture
of magnetite and maghemite (berthollide). However, oxidation appears to be of
little consequence in most technological applications as the oxidation has little
effect, if any, on the colloidal stability and also results in only a relatively small
decrease in the magnetic moment of the particles, at most 10%. To produce pure
maghemite it is a relatively simple process to oxidize the particles of magnetite
by heating the particles to 90o C for thirty minutes in a 0.34M solution of ferric
nitrate [21].
Substituted ferrite particles. To produce substituted nano-sized ferrite par-

ticles suitable for use in ferrofluids the Fe2+ ion is simply replaced or partially
replaced by another or combination of divalent metal ions such as Co2+ , Mn2+ ,
Ni2+ , Zn2+ etc. Ions of other valency, e.g., Li+ can also be used to prepare
6 S.W. Charles
substituted ferrites. Numerous papers and patents have been published on this
subject. See the references given in bibliographies quoted earlier. The method
of precipitation of the particles is essentially the same as that used to prepare
magnetite. Thus it is not important that details of individual preparations be
given here, except to say that in some cases the precipitate needs to be hy-
drothermally aged to facilitate the conversion of the precipitated hydroxides to
the ferrite [15]. The reason for the interest in substituted ferrite particles stems
from the wide-differences in the magnetic properties of these materials [14], [16].
Thus it is possible to tailor the properties of these particles and dispersions of
such particles to satisfy various applications. For example, where particles are re-
quired with high magnetocrystalline anisotropy, cobalt ferrites can be employed
[20]. Other ferrites, which have relatively low Neel temperatures ( 100o C) and
thus high thermomagnetic coefficients at these temperatures, have been used in
studies of thermomagnetic convection and heat transfer [22].
Microemulsion techniques. In addition to simple straightforward precipi-

tation, it is possible to produce particles using the microemulsion technique.
Numerous papers have been published on the preparation of small nano-sized
particles, some of which are ferri- or ferromagnetic, via the use of reverse micelles.
A reverse micelle can be described as a water-in-oil microemulsion in which two
immiscible liquids are stabilized by a surfactant. As it is a three-component sys-
tem, a triangular phase diagram of the 3-component system is used to describe
the various structures other than microemulsions that may be formed. For de-
tailed information on micelles and particle formation the reader is recommended
to study a review edited by Pileni [23]. The method of preparation of particles
involves the preparation of two microemulsions, one containing an aqueous solu-
tion of a metal salt or mixture of metal salts and the other an aqueous solution
of an alkali [24], and mixing the two in the appropriate ratio. Using pure sur-
factants, the micelle-size distribution is very narrow, with the result that the
particles themselves also possess a narrow particle size. The drawback to this
method is that the surfactant used in the preparation may not be compatible
with the carrier required for a particular application. The only way to circumvent
this problem is either to remove the surfactant from the particles and replace
it with one compatible with the carrier or use a two-surfactant system, i.e. coat
the initial surfactant with another surfactant compatible with the carrier.
2.2 Metal Particles

There are two major advantages of using ferrofluids based on metallic particles,
such as cobalt and iron particles. Firstly, these metals have high saturation
magnetizations compared with ferrites and secondly they can be produced easily
with very narrow size distributions. However there is also a major drawback
which has restricted their use in most commercial applications and that is their
poor resistance to oxidation and subsequent loss of magnetic properties. Only
by maintaining these fluids in an inert atmosphere can these fluids possess an
extended life-time.
Decomposition of organo-metallic compounds. There are a number of

methods of producing small magnetic metallic particles, the commonest of which
for ferrofluid preparation being the decomposition of organo-metallic compounds.
The decomposition of organo-metallic compounds will be discussed in most de-
tail as it is the easiest and most versatile method of producing metal particles
suitable for ferrofluids. In the 1960’s several groups of workers [25,26] reported
the production of colloidal-sized cobalt particles by the thermolysis of dicobalt
octa-carbonyl in hydrocarbon solutions containing polymeric materials. The pro-
cess is very simple in that it only involves refluxing a solution of the carbonyl
usually in toluene, until the reaction is complete, with the formation of Co
metal particles. Thomas [25] showed that the particles possess a very narrow
size-distribution ideal for the production of stable colloids. The particles also
possess an fcc structure instead of the bcc structure of bulk cobalt. The ab-
sence of a polymer in the thermolysis leads to the formation of large particles
(>100 nm). A detailed investigation by Hess [26] of the presence of different co-
polymers of acrylonitrile-styrene in the reaction solution showed that the greater
the concentration of polar groups along the polymer chain, the smaller the size
of the particles formed. It was Smith [27] who proposed that the reaction is
polymer-catalyzed which results in the formation of a polymer-carbonyl com-
plex which then decomposes to form elemental cobalt. Papirer et al. [28] showed
that the decomposition could also be catalyzed by the presence of surfactants.
They investigated the evolution of carbon monoxide during the decomposition in
the presence of diethyl-2-hexyl sodium sulphosuccinate (Manoxol-OT) and in its
absence, both experiments carried out as a function of temperature. The results
were consistent with the presence of ’micro-reactors’ and a diffusion mechanism
to control the growth. The higher the temperature used the smaller are the
particles produced. During the process of thermolysis in toluene, the presence
of Co4 (CO)12 was identified as an intermediate. However the overall reaction
mechanism is complex but it seems likely that a carbonyl complex and/or mi-
croreactors are formed which subsequently decompose to produce cobalt parti-
cles. Charles and Wells [29] investigated methods to control the size of the cobalt
particles formed by the use of a variety of surfactants and combinations of sur-
factants. By taking aliquots of the reaction mixture as the reaction proceeded
and making observations using electron microscopy they observed the presence
of clusters or microreactors for the first half hour of a four hour reaction but no
particles. Subsequently observations showed the presence of particles only. They
were able to produce, consistently, particles with median diameters in the range
5 to 15 nm. The median diameter refers to a lognormal volume distribution.
In many of the papers quoted so far little information has been presented con-
cerning the saturation magnetization of the particles compared to that of the
bulk. Charles and Wells [29] found that the saturation magnetization of the par-
ticles to be consistently lower (25%) than that of the bulk value of 17900 Gauss
(1.79 T). This result is in agreement with the work of Hess and Parker [26]. There
are several possible reasons for this, namely oxidation, spin-pinning/canting or
poor crystallization. Experiments where care had been taken to avoid oxida-
8 S.W. Charles
tion still showed the reduced saturation magnetization. However evidence from
X-ray diffraction indicated that the particles were poorly crystallized or amor-
phous with similar results being obtained for Fe particles produced by a similar
process [30]. X-ray diffraction and Mössbauer studies indicated the presence of
carbon in the particles which is thought responsible for the poorly crystallized
state. The particles, when annealed showed an X-ray pattern consistent with α-
Fe and χ-Fe5 C2 . Some researchers, using transmission electron diffraction, have
reported that the particles produced by thermolysis have a crystalline structure.
Whether this is a true indication of the structure of the particles or that an-
nealing has taken place in the electron beam is not known. Measurements of the
magnetic anisotropy of Co particles have been undertaken by several groups of
workers [31,29].
Iron particles of colloidal dimensions can also be produced by thermolysis
[30]; in this case of iron pentacarbonyl in decahydronaphthalene. Particle size
is again controlled by the presence of surfactants. The actual structure of the
particles is open to question. Some groups have reported a crystalline struc-
ture [32] whilst others [30] reported an amorphous structure which undergoes
crystallization on annealing.
Hoon et al. [33] have reported the preparation of colloidal nickel particles
by the irradiation of nickel carbonyl with ultraviolet light and by the reduc-
tion of nickelocene. The particles were reported to have an fcc structure from
transmission electron diffraction.
Nakatani et al. [34,35] have prepared iron nitride particles by two methods.
Firstly by a plasma CVD reaction of iron pentacarbonyl and nitrogen gas and
secondly by heating a solution of iron pentacarbonyl in kerosene containing a
surfactant through which a constant stream of ammonia is passed. The crystal
structure observed is -Fe3 N.
It is also possible to prepare alloy particles of the transition metals by the
thermolysis of mixed-metal organometallic compounds in the presence of surfac-
tants. Lambrick et al. [36,37] prepared organo-metallic compounds from which
particles of Fe-Co and Ni-Fe alloys with median diameter 8 nm were obtained.
The saturation magnetizations of the particles are >84% of that of the bulk
values.
Inverse micro-emulsion techniques. The stimulus for much of this work

has been an interest in the production of small-particle catalysts. The method is
similar to that described for the production of ferrite particles by the same tech-
nique except that the microemulsions containing alkali are replaced by micro-
emulsions containing a reducing agent such as sodium borohydride on sodium
hypophosphite. As with ferrite production by simply mixing the two in the ap-
propriate ratio, metal particles, in which some boron or phosphorus is usually
incorporated, are produced within the micelles. Because of the incorporation
of boron or phosphorus, the particles often have an amorphous structure. Using
pure surfactants, the micelle size distribution is very narrow, with the result that
the particles themselves also possess a narrow particle size. It has been shown
by Lopez-Quintela et al. [38] that it is possible to subsequently coat the metal

particles formed with another metal using a further microemulsion. This pro-
cess is of particular interest in that by the use of an appropriate second metal it
may be possible to overcome the recurring problem of oxidation of ferromagnetic
particles of iron, cobalt etc.
Reduction of metal salts in aqueous solution. Particles produced by the

reduction of metal salts in aqueous solution using reducing agents such as sodium
borohydride or sodium hypophosphite are typically micron-sized. This precluded
their use in ferrofluids because of their large magnetic moments and consequent
strong magnetostatic interactions particularly in the presence of a magnetic field.
However nano-sized particles have been produced, the formation of which is in-
duced by the addition of PdCl2 to the solution as a nucleating agent. It has been
shown by Akashi et al. [39] that by the introduction of water-soluble viscosity-
increasing materials that particle size can be reduced sufficiently to produce a
stable colloidal suspension. Harada et al. [40] have reported the formation of Co-
P particles, using sodium hypophosphite as reducing agent, in aqueous solutions
containing protein or synthetic polypeptides. Similar studies on Fe-P particles
have also been undertaken. There have been a few other reports of the formation
of such particles.
As in the case of the microemulsion preparations, the at.% of P or B in the
particles is very dependent on the conditions used. Above a critical concentration
of B, P and indeed C, the particles possess an amorphous structure very similar
to structures produced by rapid-quench techniques. The incorporation of P, B
and C is a disadvantage in one respect, in that the saturation magnetization of
the particles may be significantly reduced.
Small acicular particles of Co particles have been prepared using this method
performed in the presence of a uniform magnetic field [41].
Miscellaneous methods. Fine particles of most of the metallic elements can

be produced by evaporation of the metals in an inert gas atmosphere. The size
of the particles, so produced, is very dependent on the gas used and its pressure.
See 3.2 covering the preparation of magnetic fluids using miscellaneous methods.
2.3 Particle-Size Distribution

In order to achieve stable colloidal dispersions reproducibly, it is of paramount
importance to monitor the particle size for each preparation. It is an obvious
advantage to have a narrow particle-size distribution since large particles, which
may adversely affect the performance of the fluid, are absent.
Nano-sized particles of ferrites produced by the methods described almost
invariably have a rough, very approximate spherical shape, see Fig. 1(a). For the
co-precipitation method the particle-size distribution is relatively wide in that
the standard deviation of the lognormal size distribution is usually about 0.4.
In the case of metal particles, the particles are near-spherical having a smooth
10 S.W. Charles
appearance, see Fig. 1(b). Further the metals invariably have a narrow particle-
size distribution with a standard deviation of <0.2.
Fig. 1. (left) Magnetite particles (right) Cobalt particles
Particle-size distribution is best monitored by using transmission electron

microscopy. However, another method based on measurement of the magnetiza-
tion curve of the fluid is particularly convenient [42]. To obtain the particle size
and standard deviation, the saturation magnetization of the particles is needed.
Use of the bulk value may lead to erroneous values, particularly in the case of
ferromagnetic particles where values of the saturation magnetization of the par-
ticles and bulk can differ greatly. The method also assumes that the particle-size
distribution can be described by a log normal volume distribution [43]. This dis-
tribution has been found to be a good representation for most systems studied
whether they be based on metallic or ferrite particles.
X-Ray diffraction and use of the Debye-Scherrer expression is a satisfactory
pointer to the particle size for well-crystallized particles. This method is obvi-
ously of no value for particles, such as some of the metal particles, which have
an amorphous structure.
3 Preparation of Ferrofluids
For most applications it is absolutely essential that the ferrofluids be very stable
with regard to temperature and in the presence of magnetic fields. The presence
of agglomeration of particles must be avoided at all costs. The literature con-
tains numerous papers where phenomena can be attributed to the presence of
agglomerates and yet their presence has been omitted in any discussion of the
phenomena.
For most applications a low viscosity, low vapor pressure and chemical in-
ertness are desirable for the carrier liquid and surfactant. Scholten [44] gives
a review of the chemical and physical problems associated with these parame-
ters and lists the advantages and disadvantages associated with different carrier
liquids.
3.1 Fluids Containing Ferrite Particles
Surfactant stabilized colloids. Various methods of coating particles prior to

dispersion in a carrier liquid have been described but all are basically the same
with small differences in procedure. Fig. 2 shows an example of one method.
Fig. 2. Typical process for preparation of surfactant stabilized ferrofluids.
Magnetite can be coated with oleic acid by adding the acid to the precipitated
phase in alkaline solution at pH 9.5. The mixture is usually left stirring for
approximately 1 hour after which time it is heated to 95o C to facilitate the
conversion of hydroxides formed to ferrite. After cooling, the product is acidified
to pH 5 using nitric acid. The oleate ion produced chemisorbs strongly on to
the magnetite particles, and the precipitate coagulates and falls out of solution.
The coagulation may also be accompanied by the appearance of a thin film of
oleic acid on the surface of the supernatant liquid which is unadsorbed oleate
ions reconverted to oleic acid in acid solution. Sufficient oleic must be added
to form a monolayer coverage [20]. The supernatant liquid can be decanted
12 S.W. Charles
and the agglomerated hydrophobic precipitate of coated particles washed several

times to remove salts, such as nitrates, chlorides and sulphates. Water and any
physisorbed oleic acid can be removed by washing with acetone. By adding
the appropriate carrier liquid to the acetone-wet slurry and gently heating, the
acetone and any residual water can be removed leaving a colloidal dispersion of
magnetite. If any aggregates are present it may be necessary to centrifuge the
fluid or subject the fluid to a high-gradient magnetic field separator (HGMS)
[45]. If the washing procedure with water has not been carried out carefully the
presence of small quantities of micron-sized particles of salt may be present,
which must be filtered.
Using this method with the appropriate surfactant it is possible to produce
stable colloids in low vapor pressure carriers such as diesters [46], polyphenyl
ethers [47], silicone oils [48], hydrocarbons [18], perfluorocarbons, perfluoro-
polyethers etc.
Stable water-based fluids can be produced by this method by the addition of
a variety of secondary surfactants to oleate-coated particles [49].
As with fluids prepared from metal particles, the limiting value of satura-
tion magnetization of ferrite-based fluids depends on the desired viscosity of the
fluids. Fluids with magnetization up to approximately 1000 Gauss (0.1 T) can
be prepared [50] which represents approximately 25 volume % of ferrite. Most
ferrites have saturation magnetizations at 20o C up to a maximum value of 5000
Gauss (0.5 T).
Ionically-stabilized colloids. For the production of ionically-stabilized col-

loids particles of ferrites are produced by methods similar to that described
in 2.2.
The particles produced by these methods are macro-ions in which the electric
changes are due to specific adsorption of the amphoteric hydroxyl group. In an
alkaline medium, the particles are negatively charged and in an acidic medium,
positively charged. These particles can be stabilized in water by the presence
of low-polarizing counter-ions such as N(CH3 )4 + in alkaline medium, or ClO4 −
in acidic medium. Strongly polarizing counter-ions lead to flocculation. For the
-OH ligand, the point-of-zero-charge (PZC) occurs at approximately pH = 7.5.
The small surface-charge density in this region of pH precludes the formation
of a stable colloidal suspension. It was Massart (1981) developed the method
of producing stable colloidal dispersions in alkaline and acidic aqueous media.
He and his co-workers have subsequently made extensive and detailed studies of
such dispersions, e.g., [51,21].
Using OH− ligands, stable suspensions are restricted to pH values outside the
range 6-10. Massart and co-workers [51] overcame this problem by modifying the
nature of the surface charges by substituting the -OH surface ligands by citrate
ligands. In so doing they were able to synthesize stable fluids in the pH range
3-11.
3.2 Fluids Containing Metal Particles

As already pointed out in 3.1, a drawback exists in the production of metal parti-
cles by the thermal decomposition of organo-metallic compounds, in that where
a surfactant is chosen to produce particles of a particular size that surfactant
may not be compatible with the carrier liquid needed for a particular applica-
tion. Although the surfactants used in the preparation can be removed prior to
dispersion in another carrier liquid containing an appropriate surfactant, this is
not always an easy option. Thus sometimes a compromise has to be made. To
achieve compatibility between the surfactant and carrier liquid, one may have
to be flexible in one’s choice of particle size which is dictated by the surfactant
used in the preparation of the particles. Alternatively, sometimes it is possible
to achieve a stable colloid by the addition of a secondary surfactant. Transfer
of the coated particles to the carrier of choice is best facilitated by rotary evap-
oration under reduced pressure of the solvent used in the preparation, in the
presence of the carrier liquid. The volume of the latter dictates the value of the
saturation magnetization of the ferrofluid. This value will in turn be dictated
by the requirements of viscosity for particular applications. For particles of Fe,
Co and Fe3 N, saturation magnetizations of 2000 Gauss (0.2 T) are feasible [50],
representing a volume fraction of the metal particles between 10 and 20%.
Similar problems occur to the above in the production of metal particles
by the inverse-microemulsion technique, because the presence of a surfactant to
produce the microemulsion may not be compatible with the carrier fluid required.
This is also the case for the production of metal particles by the reduction of
metal salts in aqueous solution.
Ferrofluids containing particles of Fe, Co or Ni have also been prepared by
the vacuum evaporation of metals on to a running oil substrate containing a
surfactant [52]. The particles produced by this method are very small ( 2 nm in
diameter). On heating the product to 270o C in an argon atmosphere for a few
minutes, the particle size can be increased. Fluids produced by this method are
gravitationally stable.
This method has also been used to produce dispersions of particles of Gd
and transition metal alloys in mercury [53,54]. Unfortunately these fluids are
not colloidally stable despite attempts to stabilize the dispersions with metallic
additives [55].
Spark erosion has been used to produce ferrofluids [56]. A wide range of
particle sizes is produced, the smaller particles of which are suitable for dispersion
to form a ferrofluid. The method suffers from disadvantages in that the process is
slow and the particle distribution is very wide. However it does have an advantage
in that novel particles with similar properties to the bulk can be produced, e.g.,
TbFeB which possess a large magnetostriction.
3.3 Assessment of Colloidal Stability

As has already been pointed out, it is of paramount importance that ferroflu-
ids are well-characterized, i.e. the particle-size distribution be defined and the
presence of agglomerates be eliminated.
14 S.W. Charles
In many applications of ferrofluids, such as exclusion seals, loudspeakers, the

fluids are subjected to large magnetic fields and field gradients and thus the
presence of agglomerates or large particles will also certainly have a detrimen-
tal effect on the performance of the device. This problem also extends to the
interpretation of certain phenomena observed in ferrofluids.
A review of the role that aggregation plays in determining the properties
of magnetic fluids has been presented by Charles [57] as well as the various
techniques by which aggregation can be studied, both experimentally and theo-
retically. The stability of a fluid can be best monitored by simply measuring the
saturation magnetization of the fluid and the magnetic particle size [42] before
and after subjecting the fluid to a strong magnetic field gradient. An effective
and simple way of studying the effect of a magnetic field gradient on colloidal
stability is by the use of a high gradient magnetic field separator (HGMS) [45].
Other methods to monitor the presence of agglomerates and large particles in-
clude ac susceptibility measurements [58], light scattering [59], or small-angle
neutron scattering, SANS [60].
A simple method to undertake routine and reproducible measurements of
stability and thereby make meaningful comparisons between different fluids is
by means of a system based on a Colpitts oscillator [61]. In this method a long
(15 cm) narrow tube is filled with ferrofluid. Measurements of inductance along
the length of the tube enable variations in the concentration of the particles to
be monitored. Application of a magnetic field gradient allows a comparison of
the concentration profile to be made before and after exposure.
4 Magnetorheological Fluids
All magnetic fluids, of whatever nature, exhibit magnetorheological effects to a

greater or lesser degree. The magnetorheological (MR) fluids to be discussed here
will be confined to those that exhibit very strong effects only. MR fluids usually
contain micron-sized particles (5 μm) of ferromagnetic materials, as opposed to
the nano-sized particles used in ferrofluids. Even under saturating magnetic fields
stable ferrofluids exhibit relatively modest magneto-viscous effects. For example,
viscosities of the fluids may change by a factor of 2 or so. Larger effects have been
observed but almost invariably these effects can be attributed to ’poor-quality’
ferrofluids, i.e. fluids with poor colloidal stability namely those containing large
particles or aggregates of particles.
As stated previously, nano-sized particles (10 nm) are magnetically single-
domain and are in a permanent state of magnetization, i.e. they always possess
a magnetic dipole even in the absence of an applied field. Micron-sized parti-
cles, on the other hand, are magnetically polydomain. If the material used in
these micron-sized particles is magnetically ’soft’, i.e. does not retain a remanent
magnetization after exposure to an external applied magnetic field, then the par-
ticles possess zero overall moment and magnetic interactions between particles
are minimal. However, the application of even modest external magnetic fields
(say 500 Oe ≡ 40 kAm−1 ) will result in the particles attaining large magnetic
moments resulting in strong magnetic interaction leading to chaining. The ef-

fect on the MR fluid is immediate and pronounced. The MR fluid passes from
a liquid state to an extremely highly viscous ’solid-like’ state, i.e. the magnetic
particles are capable of imparting a high yield stress capability to these fluids.
Control of the magnitude of the applied magnetic field allows easy control of
the viscosity of the fluid, and for this reason these fluids and the analogous
electrorheological fluids are often referred to as smart fluids. Because of these
properties, which will be discussed in detail in Chapter 4 , great interest has
been shown in these fluids, some of which have been exploited commercially in
a wide variety of applications, e.g. in dampers, clutches etc.
Numerous patents have been published (see bibliography references given in
the Introduction to this Chapter) describing the preparation of MR fluids and
development of the various additives used in these fluids. Generally these MR
fluids contain micron-sized particles of magnetically ’soft’ ferromagnetic materi-
als dispersed in a variety of carrier fluids. In principle dispersions can be made in
any fluid. Probably the most commonly used magnetic material to date, but by
no means the only one, is carbonyl iron, which is readily available commercially
in a wide range of particle sizes. Shulmann et al. [62], Bibik [63] and many others
have studied the properties of MR fluids based on carbonyl iron. Alloys of iron
with other transition metals [64] and carbonyl nickel and electrolytic nickel [65]
have also been used. MR fluids containing particles with a high specific gravity
(e.g. iron = 7800 kgm−3 ) in a low specific gravity carrier (e.g. silicone oil 950
kgm−3 or a perfluoropolyether 2000 kgm−3 ) suffer from excessive gravitational
settling, the result of which can interfere with the magnetorheological activity of
the fluid due to the non-uniform distribution of particles. Brownian movement is
of no importance in this system unlike ferrofluids. However, in many applications
large shear forces are exerted which can render the fluid relatively homogeneous
in a short time, thereby minimizing disturbance of the magnetorheological ac-
tivity. To overcome this problem, metallic soap-like surfactants such as lithium
stearate, aluminium stearate etc, have been used in the past to reduce particle
settling, but they suffer to a certain degree from water retention limiting the
operating temperature of the fluid. Ideally the MR fluid should exhibit minimal
particle-settling over a broad temperature range. The problem of gravitational
settling can be achieved by the introduction of thixotropic agents, of which a
large number are presently available, the use of which has also been the subject
of a number of patents. The thixotropic agent creates a network of the particles
that at low shear rates form a loose structure, which imparts a small degree of
rigidity to the MR fluid, thereby reducing particle settling. On application of a
shearing force, this structure is easily disrupted, but is rapidly reformed on re-
moval of the shear force. Where hydrogen-bonded thixotropic agents have been
used, colloidal metal-oxide additives have been used to facilitate the thixotropic
network [66].
The actual preparation of these systems is relatively simple and considerably
less complicated than the preparation of ferrofluids. Generally, it is sufficient
to mix the ingredients of the MR fluid, i.e. particles, carrier, thixotropic agents
16 S.W. Charles
etc, by mechanical or ultrasonic means. The relative amounts of the ingredients

and carrier are dependent on the particular application in mind. Typically MR
fluids contain up to 70 wt% of metal particles. Because of the simplicity of
the preparation, in contrast to ferrofluids, and the fact that the components
to produce MR fluids are readily commercially available, it is not considered
necessary to describe any one particular procedure, other than to say the choice
of dispersing agent/thixotropic materials depends on the nature of the carrier
used. The bibliography/patents on the preparation of MR fluids makes available
information concerning the best combination of components to use.
References
1. Zahn M, Shenton K E 1980 Magnetic fluids bibliography. IEEE Trans. Mag. MAG-
16(2): 387-415.
2. Charles, S W, Rosensweig R E 1983 Magnetic fluids bibliography. J. Magn. Magn.
Mat. 39: 190-220
3. Kamiyama S, Rosensweig R E 1987 Magnetic fluids bibliography. J. Magn. Magn.
Mat. 65: 401-39.
4. Blums E, Ozols R, Rosensweig R E 1990 Magnetic fluids bibliography. J. Magn.
Magn. Mat. 85: 303-82
5. Cabuil V, Neveu S, Rosensweig R E 1993 Magnetic fluids bibliography. J. Magn.
Magn. Mat. 122: 437-82
6. Bhatnagar S P, Rosensweig R E 1995 Magnetic fluids bibliography. J. Magn. Magn.
Mat. 149: 198-232
7. Vekas L, Sofonea V, Balau O 1999 Magnetic fluids bibliography. J. Magn. Magn.
Mat. 39: 454-89
8. Rosensweig R E 1979 Fluid dynamics and science of magnetic liquids. In: Advances
in Electronics and Electron Physics. Vol. 48: Martin M (ed.) Academic Press, New
York, pp. 103-99
9. Rosensweig R E 1985 Ferrohydrodynamics. Cambridge University Press, Cam-
bridge, UK
10. Rosensweig R E 1988 Special Issue on Ferrofluids. Chem. Eng. Comm. 67: 1-340
11. Charles S W, Popplewell J 1980 Ferromagnetic liquids. In: Wohlfarth E P (ed.)
Ferro-magnetic Materials Vol. 2: North-Holland, Amst. pp. 509-59
12. Martinet A 1983 The case of ferrofluids. In: Aggregation Processes in Solution,
Elsevier, New York, Ch. 18, pp. 1-41
13. Scholten P C 1978 In: Berkovsky B M (ed.) Thermomechanics of magnetic fluids.
Hemisphere, Washington D C, pp. 1-26
14. Smit J, Wijn H P J 1959 Ferrites. Wiley, New York
15. Schuele W J, Deetscreek V D 1963 Ultrafine particles. Kuhn W E (ed.) Wiley, New
York
16. Dormann J-L, Nogues M 1990 Magnetic structure of substituted ferrites. J. Phys:
Condens. Mat. 2: 1223-37
17. Papell S S 1965 Low viscosity magnetic fluid obtained by the colloidal suspension
of magnetic particles. U S Patent 3,215,572
18. Khalafalla S E, Reimers G W 1973 Magnetofluids and their manufacture. U S
Patent 3,764,540
19. Massart R 1981 Preparation of aqueous magnetic liquids in alkaline and acidic
media. IEEE Trans. Mag. MAG-17: 1247-8
20. Davies K J, Wells, S, Charles S W 1993 The effect of temperature and oleate
adsorption on the growth of maghemite particles. J. Magn. Magn. Mat. 122: 24-8
21. Bee A, Massart R, Neveu S 1995 Synthesis of very fine maghemite particles. J.
Magn. Magn. Mat. 149: 6-9
22. Nakatsuka K, Hama Y, Takahashi J 1990 Heat transfer in temperature-sensitive
magnetic fluids. J. Magn. Magn. Mat. 85: 207-9
23. Pileni M P 1989 Structure and Reactivity of Reverse Micelles, Elsevier, Amst.
24. Gobe M, Kon-No K, Kitahara A 1984 Preparation of magnetite superfine sol in
w/o microemulsion. J. Coll. Int. Sci 93: 293-5
25. Thomas J R 1966 Preparation and magnetic properties of colloidal cobalt particles.
J. Appl. Phys. 37(7): 2914-15
26. Hess P H, Parker P H 1966 Polymers for stabilization of colloidal cobalt particles.
J. Appl. Polymer Sci 10: 1915-17
27. Smith T W 1981 Surfactant-catalyzed decomposition of dicobalt octacarbonyl. U S
Patent 4,252,673
28. Papirer E, Horny P, Balard H, Anthore R, Petipas R, Martinet A 1983 The prepa-
ration of a ferrofluid by the decomposition of dicobalt octacarbonyl. J. Coll. Int.
Sci. 94: 207-20
29. Charles S W, Wells S 1990 Magnetic properties of colloidal suspensions of cobalt.
Magnetohydrodynamics 26: 288-92
30. Wonterghem J van, Mørup S, Charles S W, Wells S, Villadsen J 1986 Formation of
a metallic glass by thermal decomposition of Fe(CO)5 . Phys. Rev. Lett. 55: 410-13
31. Chantrell, R W, Popplewell J, Charles S W 1980 The coercivity of a system of
single domain particles with randomly oriented easy axes. J. Magn. Magn. Mat.
15-18: 1123-4
32. Kilner M, Hoon S R, Lambrick D R, Potton J A, Tanner B K 1984 Preparation
and properties of metallic iron ferrofluids. IEEE Trans. Mag. MAG-20: 1735-7
33. Hoon S R, Kilner M, Russell G J, Tanner B K 1983 Preparation and properties of
nickel ferrofluids. J. Magn. Magn. Mat. 39: 107-10
34. Nakatani I, Furubayashi T 1990 Iron-nitride magnetic fluids prepared by plasma
CVD technique. J. Magn. Magn. Mat. 85: 11-3
35. Nakatani I, Hijikata M, Ozawa K 1993 Iron-nitride fluids prepared by vapor-liquid
reaction. J. Magn. Magn. Mat. 122: 10-4
36. Lambrick D B, Mason N, Harris N J, Russell G J, Hoon S R, Kilner M 1985 An
iron-cobalt alloy magnetic fluid. IEEE Trans. Mag. MAG-21: 1891-3
37. Lambrick D B, Mason N, Hoon S R, Kilner M 1987 Preparation and properties of
Ni-Fe magnetic fluids. J. Magn. Magn. Mat. 65: 257-60
38. Lopez-Quintela M A, Rivas J 1992 Covering of magnetic particles to produce stable
magnetic fluids. Spanish Patent 9,201,984 (see also 1994, Structural and magnetic
characterisation of Co particles coated with Ag. J. Appl. Phys. 76: 6564-6)
39. Akashi G, Fujyama M 1969 Process for the production of magnetic substances. US
Patent 3,607,218
40. Harada S, Yamanashi T, Ugaji M 1972 Preparation and magnetic properties of
cobalt alloy particles. IEEE Trans. Mag. MAG-8: 468-72
41. Charles S W, Issari B 1986 The preparation and properties of small acicular par-
ticles of cobalt. J. Magn. Magn. Mat. 54-57:
743-4
42. Chantrell R W, Popplewell J, Charles S W 1978 Measurements of particle-size
distribution parameters in ferrofluids. IEEE Trans. Mag. MAG-14: 975-7
43. O’Grady K, Bradbury A 1983 Particle-size analysis in ferrofluids. J. Magn. Magn.
Mat. 39: 91-4
18 S.W. Charles
44. Scholten P C 1988 Some material problems in magnetic fluids. Chem. Eng. Comm.
67: 331-40
45. O’Grady K, Stewardson H R, Chantrell R W, Fletcher D, Unwin D, Parker M R
1986 Magnetic filtration of ferrofluids. IEEE Trans. Mag. MAG-22: 1134-6
46. Wyman J E 1984 Ferrofluid composition and method of making the same. U S
Patent 4,430,239
47. Bottenberg W R, Chagnon M S 1982 Low vapor-pressure ferrofluids using
surfactant-containing polyphenyl ether. U S Patent 4,315,827
48. Chagnon M S 1982 Stable ferrofluid compositions. U S Patent 4,356,098
49. Shimoiizaka, J, Nakatsuka K, Chubachi R 1978 In: Thermomechanics of magnetic
fluids. Berkovsky B M (ed.) Hemisphere, Washington DC, pp. 67-76
50. Scholten P C 1983 How magnetic can a fluid be? J. Magn. Magn. Mat. 39: 99-106
51. Bacri J C, Perzynski R, Salin D, Cabuil V, Massart R 1990 Ionic ferrofluids: a
crossing of chemistry and physics. J. Magn. Magn. Mat. 85: 27-32
52. Nakatani I, Furubayashi T, Takahashi T, Hanaoka H 1987 Preparation and mag-
netic properties of colloidal ferromagnetic metals. J. Magn. Magn. Mat. 65: 261-4
53. Shepherd P G, Popplewell J, Charles S W 1970 A method of producing a ferrofluid
with Gd ferromagnetic particles. J. Appl. Phys. D 3: 1985-7
54. Shepherd P G, Popplewell J, Charles S W 1972 Ferrofluids containing Ni-Fe alloy
particles in mercury. J. Phys D 5: 2273-82
55. Keeling L, Charles S W, Popplewell J 1984 The prevention of diffusional growth
of cobalt particles in mercury. J. Phys. F 14: 3093-3100
56. Berkowitz A E, Walter J L 1983 Ferrofluids produced by spark erosion. J. Magn.
Magn. Mat. 39: 75-8
57. Charles S W 1988 Aggregation in magnetic fluids and magnetic fluid composites.
Chem. Eng. Comm. 67: 145-80
58. Fannin P C, Scaife B K P, Charles S W 1987 The study of the complex susceptibility
of ferrofluids and rotational Brownian motion. J. Magn. Magn. Mat. 65: 279-81
59. Beresford J, Smith F M 1973 In: Parfitt G D (ed.) Dispersion of Powders in Liquids.
Publ. Appl. Science.
60. Cebula D J, Charles S W, Popplewell J 1983 Aggregation in ferrofluids studied by
neutron small angle scattering. J. Physique 44: 207-13
61. Bissell P R, Chantrell R W, Spratt G W D, Bates P A, O’Grady K 1984 Long term
stability measurements on magnetic fluids. IEEE Trans. Mag. MAG-20: 1738-40
62. Shulman Z P, Kordonsky V I, Zaltsgendler E A, Prokhorov I V, Khusid B M and
Demchuk S A 1986 Structure, physical properties and dynamics of magnetorheo-
logical suspensions. Int. J. Multiphase Flow 12: 935-55
63. Bibik E E 1981 Rheology of Dispersed Systems. Leningrad State Univ. Press. 171
64. Carlson J D, Weiss KD 1995 Magnetorheological materials based on alloy particles.
U S Patent 5,382,373
65. Shulman Z P 1991 Magnetorheological systems and their application. European
Advanced Short Course of UNESCO. Minsk, USSR
66. Weiss K D, Nixon D A, Carlson J D, Margida A J 1994 Thixotropic magnetorhe-
ological materials. W O Patent 9,410,693
Magnetic Spectroscopy
as an Aide in Understanding Magnetic Fluids
Paul C. Fannin
Department of Electronic and Electrical Engineering, University of Dublin, Trinity

College, Dublin 2, Ireland.
Abstract. Accurate data on the frequency-dependent, complex susceptibility, χ(ω) =

χ (ω)−iχ (ω) of magnetic fluids are vital for an understanding of the dynamic behavior
of these colloidal suspensions. They enable relaxation mechanisms, both Brownian and
Néel, as well as ferromagnetic resonance to be identified and investigated. They also
provide a convenient means of determining the macroscopic and microscopic properties
of the fluids, such as the mean particle radius, r̄, the mean value of anisotropy field,
H̄A , the gyromagnetic constant, γ, and the damping constant, α. Also, through the
medium of χ(ω) one can also investigate the frequency dependence of the loss tangent
(tanδ) and power factor (sinδ) of such particulate systems. In this Chapter the above-
mentioned topics together with the relevant theory are presented and by means of
suitable examples it is demonstrated how the appropriate equations may be employed
to determine such properties from experimental data.
1 Relaxation Mechanisms
The major advances of recent times in the investigation of the relaxation mech-
anisms of ferrofluids are in no small way due to the application of dielectric
formalism in the representation of ferrofluid data [1-5]. There are two distinct
mechanisms by which the magnetisation of ferrofluids may relax after an applied
field has been removed: either by rotational Brownian motion of the particle
within the carrier liquid, with its magnetic moment, m, locked in an axis of easy
magnetisation, or by rotation of the magnetic moment within the particle.
m = Ms v (1)
where v is the volume of the particle and Ms its saturation magnetisation. The
time associated with the rotational diffusion is the Brownian relaxation time τB
[6] where
τB = 3V η/kT (2)
V is the hydrodynamic volume of the particle and η is the dynamic viscosity of
the carrier liquid. In the case of the second relaxation mechanism, the magnetic
moment may reverse direction within the particle by overcoming an energy bar-
rier, which for uniaxial anisotropy, is given by Kv, where K is the anisotropy
constant of the particle. The probability of such a transition is exp(σ) where σ
is the ratio of anisotropy energy to thermal energy (Kv/kT ). This reversal time

20 P.C. Fannin
is characterised by a time τN , which is referred to as the Néel relaxation time

[7], and given by the expression,
τN = τ0 exp(σ) (3)
τ0 is a damping or extinction time having an often quoted approximate value of

10−8 to 10−10 s [8-10]. According to Brown [11], for high and low barrier heights,
τN = τ0 σ −1/2 exp(σ), σ>2 (4)

= τ0 σ, σ1
Other workers [12,13] have subsequently derived a single expression to cater

for a continuous range of σ, however because of the difficulty in characterising
small magnetic particle systems it is perfectly adequate to use the expressions
of equation (4). A distribution of particle sizes implies the existence of a distri-
bution of relaxation times, with both relaxation mechanisms contributing to the
magnetisation. They do so with an effective relaxation time τef f [14], where, for
a particular particle,
τef f = τN τB /(τN + τB ) = τ (5)
The mechanism with the shortest relaxation time being dominant. The theory
developed by Debye to account for the anomalous dielectric dispersion in dipolar
fluids has been successfully used [15,16] to account for the analogous case of
magnetic fluids. Debye’s theory holds for spherical particles when the magnetic
dipole-dipole interaction energy, U, is small compared to the thermal energy kT.
According to Debye’s theory the complex susceptibility, χ(ω) has a frequency
dependence given by the equation,
χ(ω) = χ∞ + (χ0 − χ∞ )/(1 + iωτ ) (6)

= χ∞ + (χ0 − χ∞ )(1/1 + ω τ ) − iωτ /(1 + ω τ ))
2 2 2 2
(7)
where
χ0 = nm2 /3kT μ0 (8)
is the static susceptibility and
τ = 1/ωmax = 1/2πfmax . (9)

fmax is the frequency at which χ (ω) is a maximum, n is the particle number
density and χ0 and χ∞ indicate susceptibility values at f=0 and at very high
frequencies.
Magnetic Spectroscopy 21
Fig. 1 shows a plot of the Debye equation (7) which clearly indicates how

χ (ω) falls monotonically and that the maximum possible value of the absorption

component, χ (ω), is equal to half that of the χ (ω) component. Here τ = τef f =
τ .

Fig. 1. Debye plot of χ (ω) and χ (ω) against ωτ

Fig. 2. Normalised plot of χ (ω) and χ (ω) against f (Hz) for samples 1, 2 and 3
The susceptibility plots of Fig 2 over the approximate frequency range, 10

Hz to 1 MHz, for three different samples, (samples 1, 2 and 3), obtained by
means of the toroidal technique [15], clearly illustrates that the loss peak can
lie in any part of the frequency spectrum. Sample 1 consists of a suspension
of magnetite in isopar M whilst samples 2 and 3 are suspensions of magnetite
22 P.C. Fannin
in water; the corresponding mean particle radii as determined by electron mi-

croscopy, are 5 nm, 5.5 nm and 7 nm respectively. It is apparent that the plots

of samples 2 and 3 have Debye-type profiles with a maximum in χ (ω) occurring
at 1 kHz and 1.6 kHz; from equation (1) and using a viscosity of 10−4 Nsm−2 ,
corresponding hydrodynamic radii of 80 nm and 69 nm are obtained. These val-
ues are greater than the corresponding values of magnetic radius plus surfactant
thickness( approx 2-4 nm) and are thus indicative of aggregation. In contrast,
the profile of sample 1 does not have an absorption peak in the frequency range
up to 1 MHz and corresponds to the profile of a finely dispersed sample. For
the case of Brownian relaxation, a distribution of particle volumes corresponds
to a distribution of relaxation times where χ(ω) may be expressed in terms of a
distribution function, f (τ ) giving
∞
χ(ω) = χ∞ + (χ0 − χ∞ ) f (τ )dτ /(1 + iωτ ) (10)
0
f (τ ) may be represented by a Normal, Log-normal, Frohlich or a Cole-Cole

distribution function. In the Cole-Cole case where the complex susceptibility

data fits a depressed circular arc, the relation between χ (ω) and χ (ω) and
their dependence on frequency, ω/2π, can be displayed by means of the magnetic
analogue of the Cole-Cole plot [2]. In the Cole-Cole case the circular arc cuts

the χ (ω) axis at an angle of αc π/2; αc is referred to as the Cole-Cole parameter
and is a measure of the particle-size distribution . The magnetic analogue of the
Cole-Cole circular arc is described by the equation
χ(ω) = χ∞ + (χ0 − χ∞ )/[(1 + (iωτ0 )1−αc ], 0<α<1 (11)
which for αc = 0, reduces to that of equation (6). The Cole-Cole distribution

function [17] is described by,
G(ln τ ) = (1/2π)(sinπαc /(cosh{(1 − αc ) ln(τ0 /τ )} − cos παc )) (12)
For the Frohlich [3,17] distribution function for which it is assumed that the
distribution of relaxation times is confined to a certain range from τ1 to τ2 ;
f (τ )dτ = G(ln τ )d(ln τ ) where
G(ln τ ) = 1/ ln(τ1 /τ2 ) τ2 < τ < τ1 (13)

=0 0 < τ < τ2 , τ1 < τ < ∞
and as τ1 and τ2 are proportional to the cube of the maximum and minimum
hydrodynamic radii respectively, then (rmax /rmin ) = (τ1 /τ2 )1/3
As an example of obtaining an estimate of the aggregate size distribution in
the case of sample 2, Fig 3 shows the result of fitting the measured susceptibility
plots to profiles generated by the Debye equations modified by Frohlich and Cole-
Cole distribution functions. The fits obtained from the two fitting functions are
found to be similar with the Frohlich distribution function giving the value for
the particle radii of from 37 nm to 174 nm whilst the Cole-Cole fit was realised
with αc = 0.31.

Fig. 3. Normalised plot of χ (ω) and χ” (ω) against f (Hz) for samples 2. Curve 1
represents the measured data whilst curves 2 and 3 represent Frohlich and Cole-Cole
fits
2 The Use of the Hilbert Transform

in Generating Complex Susceptibility Data

On occasions it may only be possible to obtain good quality data for χ (ω) and

not χ (ω). However χ (ω) and χ (ω) are a Hilbert transform pair so in knowing

χ (ω) one can determine χ (ω) [18-21]. The Hilbert transform of a function x(t)
is defined as ∞
x̂(t) = 1/π x(τ )/(t − τ )d(τ ) (14)
−∞
= (1/πt) ∗ x(t). (15)
where ∗ represents a convolution operation. This operation corresponds to phase-
shifting all frequency components of x(t) by 90o and it may be represented by a
linear system with an impulse response of
h(t) = 1/πt (16)
and transfer function
H(f ) = −i sgn(f). (f = ω/2π) (17)
As convolution in the time domain is equivalent to multiplication in the
frequency domain, equation (15) may be written as
X̂(f ) = H(f )X(f ) (18)
where X̂(f ), H(f ) and X(f ) are the Fourier transforms of x̂(t), h(t) and x(t)
respectively. X̂(f ) and x̂(t) are related by the expression,
∞
x̂(t) = X̂(f )ej2πf t df (19)
−∞
24 P.C. Fannin
and from equation (18), assuming that X(f ) is an even function it follows that
∞
x̂(t) = X(f ) sin(2πf )df (20)
0
As an illustration of the accuracy of the technique is initially applied to

theoretical Debye curves where it is known what the resultant Hilbert transform
should be. Initially the technique is applied to equation (7) with χ∞ = 0; the
corresponding plots of χ (ω) and χ (ω) and its Hilbert transform χ̂ (ω) are
shown in Fig.4; the Hilbert transform, proves to be a good approximation to
χ (ω) with some error existing in the low-frequency region of the plots. The
number of data points used was N=8192. To account for the situation where
a χ∞ component exists, the technique is shown to be equally successful when
applied to case where χ∞ = 0.2 and N= 8192 ; the resultant plots of χ (ω), χ (ω)
and its Hilbert transform χ̂ (ω) are shown in Fig.5.

Fig. 4. Plot of χ (ω), χ (ω) and its Hilbert transform χ̂ (ω), against f (Hz), for the
Debye case with χ∞ = 0 and N= 8192 points
In reference [22] the technique is applied to a dynamic situation, and this ex-
ample illustrates the usefulness of the Hilbert transform in determining complex
susceptibility components once one component is known. Of course, as already
mentioned, the technique can also be used to generate data on χ (ω) from a
knowledge of χ (ω), simply by taking the inverse Hilbert transform of χ (ω) .
This is due to the fact that the Hilbert transform of a Hilbert transform returns
the original signal with a change in sign [23].
3 Resonance
In the GHz frequency range the character of the dispersion changes from relax-
ation to one of resonance and it is convenient to describe χ(ω) in terms of its

Fig. 5. Plot of χ (ω), χ (ω) and its Hilbert transform χ̂ (ω) , against f(Hz), for the
Debye case with χ∞ = 0.2 and N= 8192 points
parallel (relaxational) χ (ω) and perpendicular (resonant) χ⊥ (ω), components,

with,
1
χ(ω) = χ (ω) + 2χ⊥ (ω) (21)
3
with corresponding relaxation times τ and τ⊥ , respectively.
χ⊥ (ω) is associated with resonance which is indicated by a change in sign of
the value of χ (ω) at an angular frequency, ωres ; where for a small polar angle,
ωres = γHA = γ2K/Ms . (22)
The transverse or resonant component of the susceptibility, χ⊥ (ω), may be

described by equations derived by Raikher and Shliomis [24] and modified by
Coffey et al. [25] with,
1 + iωτ2 + Δ
χ⊥ (ω) = χ⊥ (0) (23)
(1 + iωτ2 ) (1 + iω⊥ ) + Δ
where τ⊥ , Δ and τ2 are as given in reference [25] and χ⊥ (0) is the static transverse
susceptibility.
When combined with equation (6), the overall frequency-dependent suscep-
tibility is given as,

1 χ (0) 1 + iωτ2 + Δ
χ(ω) = + 2χ⊥ (0) (24)
3 1 + iωτ (1 + iωτ2 ) (1 + iωτ⊥ef f ) + Δ

= χ (ω) − iχ (ω)
In equation (24) τ⊥ has been replaced by τ⊥ef f = τ⊥ τB /(τ⊥ + τB ) in or-

der to take account of the effects of Brownian relaxation . Fig 6 shows high-
frequency susceptibility measurements obtained, by means of the transmission
26 P.C. Fannin

Fig. 6. Plot of χ (ω) and χ over the frequency range 50 MHz-18 GHz for 17 values
of polarising field over the range 0-100 kA/m, for sample 4
line technique [26,27], for a 760 G suspension (sample 4) of magnetite in isopar

M subjected to a polarising field over the range 0-100 kAm−1 . From the figure

it is seen that for the unpolarised case, resonance, indicated by the χ (ω) com-
ponent changing sign, occurs at a frequency fres = 1.6 GHz whilst the maximum

χ (ω) of the loss-peak is also shown to occur at a frequency of fmax =1.0 GHz.
Variation of the polarising field, H, over the stated range results in fres and
fmax increasing up to a frequencies of 5.0 GHz and 4.7 GHz , respectively. The
loss-peak can be attributed to the presence of two different processes, namely
the Néel relaxation of the smallest particles in the distribution and resonance.
The application of H to the sample effectively results in an increase in the bar-
rier Kv (and hence σ = Kv/kT ) which the magnetic moment of the particles
must overcome. This increase in H results in a reduction in spontaneous flipping
of the magnetic moments (Néel relaxation) leading to an increasing dominance
from χ⊥ (ω) to χ(ω). The effect is to shift fmax from 1.0 GHz to 4.7 GHz and
fres from 1.6 GHz to 5.0 GHz. The result of this is that the value of fmax cor-
responding to the maximum in the value of χ (ω) (loss peak) approaches the
value of fres as resonance becomes the dominant process. A further increase in
H would move the value of fmax closer to that of fres , however it is of interest to
note that in references [29] it has been reported that the condition fmax = fres
is not realisable.
A plot of fres against H is shown in Fig 7. and as ωres = 2πfres = γ(H + H̄A ),
the value of H̄A , a mean value of the anisotropy field, is determined from the
intercept of Fig 7. as being equal to 41 kA/m. This corresponds to a mean
value of anisotropy constant, K̄, at room temperature and bulk Ms of 0.4 T, of
8.2·103 J/m3 . From the slope of Fig 7 the magneto-mechanical ratio, γ, is found
to be 2.32·105 s−1 A−1 m.
By determining one further parameter, namely the damping constant α, one
can evaluate the magnetic viscosity, ηm . Here an estimate of α is obtained by
Fig. 7. Plot of fres against H with H̄A = 41 kA/m
fitting [19] the original susceptibility data to theoretical susceptibility profiles

generated by equation (24), suitably modified to cater for a distribution of par-
ticle size, r, and anisotropy constant, K.
Table 1. Fluid parameters
Fluid Mag. Sat.Mag. d¯ H̄A K̄ γ α ηm τo

[G] Ms [T] [nm] kA/m [103 J/m3 ] [m/As] [10−6 Ns/m2 ] [10−9 s]
760 0.4 9.0 41 8.2 2.32 0.1 2.9 1.06
Fig 8 shows the fit obtained for a Normal distribution of K with a standard

deviation of 6.103 and a mean K̄ = 1 · 104 J/m3 together with a Nakagami
distribution of radii, r, with a width factor β = 4 and a mean particle radius,
r̄ ≈ 4.5nm and a saturation magnetisation of 0.4 T. For the fit shown, α was
found to be α = 0.1, a figure within the range of values normally quoted for α.
Thus having obtained data on the macroscopic and microscopic properties
of the particles one is in a position to determine the ’magnetic viscosity’, ηm ,
where ηm = Ms /6αγ, the exponential prefactor (τ0 ) of Néels expression for τN ,
where τ0 = Ms /2αγK. The results obtained are shown in Table 1 with τ0 found
to have a value of 0.92·10−9 s, which is within the range of values predicted for
τ0 .
28 P.C. Fannin

Fig. 8. Plot of χ (ω) and χ (ω), and corresponding fits, for unpolarised plot of Fig 6
4 Time Frequency Transformation

By placing an ensemble of small magnetic particles in a large dc magnetic field,
thereby aligning the moments of the particles, the magnetisation decay, or after-
effect function, b(t), can, on removal of the field, be observed over time scales of
seconds and longer. However, the equivalent measurements over a much shorter
time range cannot be realised using this conventional method and thus an alter-
native indirect technique, such as the transformation of the frequency-dependent
magnetic susceptibility, χ(ω). Frequency-dependent χ and the time-dependent
decay function are a Fourier transform pair and this fact enables the utilisa-
tion of the susceptibility data in the investigation of b(t), of magnetic fluids.
b(t) represents the decay of the magnetisation after the removal of an external
field and according to Scaife [19], is related to the frequency dependent complex
susceptibility, χ(ω), via the following equation:
∞
χ(ω) = χ0 − iω b(t) [cos ωt − i sin ωt] dt (25)
0
where χ0 is the static susceptibility.

The complex susceptibility is defined as χ(ω) = χ (ω) − iχ (ω) and writing
the integral in parts gives,
∞ ∞

χ (ω) − iχ (ω) = χ0 − ω b(t) sin ωtdt − iω b(t) cos ωtdt. (26)
0 0
Since b(t) is a real function we can see that

∞

χ (ω) = ω b(t) cos ωtdt. (27)
0
Now b(t) cos ωt is an even function so that,

χ (ω) 1 ∞
= b(t) cos ωtdt (28)
ω 2 −∞
which may be written as,

∞
χ (ω) 1 −iωt
= b(t)e dt (29)
ω 2 −∞
Equation (29) is known as the fluctuation dissipation theorem. The compo-

nent within the brackets of eqn (29) is the Fourier transform of b(t); thus b(t)

can be obtained by carrying out an inverse Fourier transform on χ (ω)/ω with,
∞
1 χ (ω) iωt
b(t) = 2 e dω
2π −∞ ω

−1 χ (ω)
= 2 F (30)
ω
where F −1 denotes the inverse Fourier transform.
Fig. 9. Plot of normalised after-effect function, b(t)/b(0), for data in Fig 6
A plot of the normalised after-effect functions, b(t)/b(0), for the data in Fig 6,
is shown in Fig 9. Here it is seen that the effect of the polarising field is to change
b(t)/b(0) from an exponential type decay for H = 0, with a time-constant, or
1/e value, of 2.5·10−10 s to a damped oscillatory type waveform having a time-
constant of 6.9·10−11 s at H =100 kA/m. Thus there is an approximate fourfold
increase in the rate of decay over the polarising field range.
As previously mentioned χ can be described in terms of parallel, (χ ), and
perpendicular, (χ⊥ ), components (Eq. (21)) with the corresponding relaxation
times τ and τ⊥ef f . However, as the effect of the biasing field is to cause χ(ω)
to be dominated by the χ⊥ (ω) component, the decay time at H =100 kA/m
corresponds approximately to that of τ⊥ef f . This decrease in the time-constant
with increasing polarising field arises because a ferrofluid has particles of different
30 P.C. Fannin
sizes and shapes giving rise to a distribution of precession or resonant frequencies.

Thus, when an external field is removed the magnetic moments of the particles
precess back to the directions of the natural easy axes with different precession
frequencies, resulting in the transverse component of the magnetisation, χ⊥ (ω),
decaying rapidly to zero. Here τ⊥ef f is approximately 2·10−10 s.
5 Magnetic Losses
In the context of device application, losses in magnetic fluids are of particular in-

terest. The permeability of a magnetic fluid, μ(ω) = μ (ω)−iμ (ω), is a complex
quantity which expresses the loss of energy which occurs as the magnetisation
alternates. The loss mechanisms cause the flux density, B, to lag behind the ap-
plied alternating field, H, by a phase angle δ, and in this context two important
relations are the loss tangent, tan δ, which is also known as the dissipation factor
[19,28] and sin δ, which is referred to as the power factor [19]. The dissipated
energy per cm3 of the ferrofluid sample is directly proportional to both tan δ

and sin δ where in direct analogy with the dielectric case, tan δ = μ (ω)/μ (ω)

and sin δ = μ (ω)/|μ(ω)| [19].

As μ (ω) = χ (ω), and μ (ω) = χ (ω) + 1, the corresponding equations for
the loss factor and power factor are,

χ (ω)
tan δ = (31)
(1 + χ (ω))
and
χ (ω)
sin δ = 1/2 (32)
2
(1 + χ (ω)) + χ (ω)2
Thus measurement of χ(ω) enables one to investigate the frequency depen-
dence of the loss tangent (tan δ) and power factor (sin δ) as a function of particle
packing fraction and hence fluid saturation magnetization where, Fig 10 shows
the high frequency dependence of tan δ and sin δ for a 200 G fluid (sample 5).
Here the profile of tan δ is identical to that of sin δ with both differing from the

χ (ω) profile.
In [28] such measurements are reported for four magnetic fluids with cor-
responding saturation magnetizations of 720 G, 360 G, 175 G and 75 G. The
results show that, with reducing fluid magnetisation, the profiles of both tan δ

and sin δ gradually become closer to that of the χ (ω) profile, until, in the case
of a 75 G Fluid, the frequency dependent profiles become almost identical. Thus,
apart from the results giving an indication of the measure of fluid magnetisation
(and hence packing fraction) required for the above condition to occur it is of
significance to note that the results also enable one to conclude that, in the case
of the most dilute fluid, the high frequency dependent loss tangent and power
factor can be modelled by the imaginary component of equation (23), with,
ω 2 τ22 τ⊥ + ωτ⊥ (1 + Δ)
tan δ = sin δ = χ⊥ (ω) = χ⊥ (0) 2 2. (33)
(1 − ω 2 τ2 τ⊥ + Δ) + ω 2 (τ⊥ + τ2 )

Fig. 10. Normalised plot of χ (ω), χ (ω), tan δ and sin δ, against f in Hz, for sample 5
Finally, with the advent of magnetoelectronics and the possible use of nano-
particles as devices, such as switches, one is interested in the signal-to-noise(SNR)
[29] ratio of such systems. Here again the usefulness of the availability of complex
susceptibility data come to the fore since the SNR for a field driven magneti-
sation of a nanoparticle is a function of the parameters τ0 , α, γ, η and K; all of
which can be determined from susceptibility measurements as demonstrated in
this Chapter.
Acknowledgements
Acknowledgement is extended to B.K.P.Scaife for useful discussions and to the

Irish Higher Education Authority and Prodex for funding this work.
References
1. P. Debye. Polar Molecules, The Chemical Catalog Company, New York (1929).
2. K.S. Cole, R.H. Cole. J. Chem. Phys., vol. 9, (1941), pp. 341.
3. H. Frohlich. Theory of Dielectrics, 2 nd edn (Oxford: Clarandon), (1958).
4. D.W. Davidson, R.H. Cole. J. Chem. Phys., vol. 9, (1951), pp. 341.
5. S. Havriliak, S. Negami. Polymer , vol. 8 (1967), pp. 161.
6. W.F. Brown. 5.Appl.Phys., vol. 34 (1963), pp. 1319.
7. L. Néel. Ann. Géophys., (1949), pp. 99.
8. D.P.E. Dickson, N.M.K. Reid, C.A. Hunt, H.D. Williams, M. El-Hilo,
K. O’Grady. J. Magn. Magn. Mater., (1993), pp. 345.
9. V. Schunemann, H. Winkler, H.M. Ziethen, A. Schiller, A.X. Trautwein.
Magnetic Properties of Fine Particles Elsevier Science Publishers, Netherlands,
(1992), pp. 371.
10. E. Kneller, E.P. Wohlfahrt. J.Appl.Phys., vol. 37, (1966), pp. 4816.
11. W.F. Brown. Phys. Rev., vol. 130, (1963), pp. 1677.
32 P.C. Fannin
12. L. Bessais, L.B. Jaffel, J.L. Dormann. Phys, Rev B., vol. 45, (1992), no. 14,
pp. 7805.
13. A. Aharoni. Phys. Rev. B , vol. 46, (1992), no. 9, pp. 5434.
14. M.I. Shliomis. Sov. Phys.-Usp., vol. 17, (1974), pp. 53.
15. P.C. Fannin, B.K.P. Scaife, S.W. Charles. J. Phys. E.Sci. Instrum, (1986),
pp. 238.
16. M.M. Mairov. Magnetohydrodynamics (Trans. of Magnitnia Hidrodinamika),
vol. 2, (1979), no. 21, pp. 53.
17. C.J.F. Bottcher, P. Bordewijk. Theory of Electric Polarisation Amsterdam,
Elsevier Scientific Publishing Company), vol. 2, (1978).
18. V.V. Daniel. Dielectric Relaxation Academic press, London, (1967).
19. B.K.P. Scaife. Principles of Dielectrics rev.edn (London, Oxford Science Publi-
cations),(1998).
20. L. Lundgren, P. Svedlindh, O. Beckman. J. Magn. Magn. Mater , vol. 25
(1986), pp. 33.
21. C.C. Paulsen, S.J. Williamson, H. Maletta. J. Magn. Magn. Mater., vol. 54-
57, (1986), pp. 209.
22. P.C. Fannin, A. Molina, S.W. Charles. J. Magn. Magn. Mater., vol. 26, (1993),
pp. 194.
23. M.S. Roden. Analog and Digital Communication Systems Prentice Hall Interna-
tional, London, (1991).
24. Y.L. Raikher, M.I. Shliomis. Sov.Phys.JETP , vol. 40, (1975), pp. 526.
25. W.T. Coffey, J.P. Kalmykov, E.S. Massawe. Modern Nonlinear Optics
Adv.Chem.Phys., vol. 85 part 2, (1993), pp. 667, Wiley Interscience, New York.
26. S. Roberts, A.R. von Hippel. J. Appl. Phys., vol. 17, (1946), pp. 610.
27. P.C. Fannin, T. Relihan, S.W. Charles. J. Phys. D; Appl. Phys., vol. 28,
(1995), pp. 2003.
28. E.C. Snelling, A.D. Giles. Ferrites for Inductors and Transmission Lines Wiley,
New York ,(1983).
29. P.C. Fannin, S.W. Charles. Investigation of the frequency dependent loss tan-
gent and power factor of colloidal suspension of nano-particles J. Magn. Magn.
Mater. , vol. 226, (2001), pp. 1887.
Magnetic and Crystalline Nanostructures
in Ferrofluids as Probed by Small Angle
Neutron Scattering
Albrecht Wiedenmann
Hahn-Meitner-Institut Berlin, Glienicker Strasse 100, D-14109 Berlin, Germany
Abstract. We present a newly developed technique of nuclear and magnetic contrast

variation by using polarised neutrons in Small Angle Neutron Scattering (SANSPOL)
which allows density, concentration and magnetisation fluctuations in magnetic liquids
to be analysed simultaneously. Diluted Ferrofluids based on different magnetic mate-
rials (Co, Magnetite, Ba-ferrite) and stabilized by charges or surfactants in different
carrier liquids have been investigated. In such polydisperse systems several constituents
of similar sizes have been identified by this technique: Magnetic core-shell composites,
magnetic aggregates and free surfactants. The corresponding size distributions, com-
positions and magnetic moments have been determined. In more concentrated Co-FF
the nature of field induced particle arrangements has been determined.
1 Introduction
Small angle neutron scattering (SANS) is known to be well suited for studying
density- and concentration fluctuations on a length scale between 0.5 nm and 300
nm which corresponds to typical sizes of microstructural features in nanoscaled
(n-) materials [1,2]. This technique allowed different types of inhomogeneities to
be identified in crystalline, amorphous and liquid materials. In addition, magneti-
sation fluctuations in domain like structures resulting from strong intergranular
correlations have been monitored by SANS in compacted n-Fe, n-Co, or n-Ni
alloys [3,4]. In some diluted systems such as n-Fe3 O4 embedded in a glass ce-
ramic [5] and n-Fe3 Si in an ferromagnetic amorphous Fe-Si-B-Nb-Cu matrix [6,7]
magnetic single domain behaviour of the nanocrystalline grains and the nature
of interfaces have been evaluated.
Here we’ll focus on the determination of the crystalline and magnetic mi-
crostructures of magnetic colloids. Magnetic colloids are stable dispersions of
ferromagnetic materials. In such “Ferrofluids” (denoted as FF) nanoscaled mag-
netic particles are stabilized against coagulation either by electrostatic repulsion
or by coating the core with organic chain molecules acting as surfactants [8]. Cur-
rently great effort is undertaken to prepare new bio-compatible Ferrofluids for
potential biomedical applications [9]. Such applications are based on the super-
paramagnetic behaviour of nanosized particles, which disappears when aggrega-
tion takes place as the consequence of an inefficient screening [10,11]. Therefore
a precise knowledge of the microstructural parameters is a pre-requisite for the
interpretation of macroscopic phenomena and for a tailored fabrication of FF.

34 A. Wiedenmann
Only few methods give access to these parameters. From macroscopic tech-
niques such as magnetisation M(H,T) measurements [12] or magneto-viscous
effects [13] average parameters of concentration, size and arrangements of mag-
netic particles are derived. However, the actual values obtained depend strongly
on the basic assumptions for the underlying processes [14]. Wide angle and small
angle X-ray scattering as well as transmission electron microscopy [15] are sensi-
tive to the particle core only, since the light elements of the organic shell give no
sufficient contrast. The advantages of neutrons in such liquids are twice: First,
the strong scattering power of hydrogen contained in the organic surfactants
gives access to the shell and second, the interaction of the neutron spin with
magnetic moments allows to visualise the magnetism of the core. Some SANS
studies have been performed mainly on concentrated systems FF, which allowed
the stability phase diagram to be established [16,17]. However, complications
arise in poly-disperse systems when different constituents are present. Then we
face the problem that weak magnetic scattering signals have to be analysed be-
side strong nuclear contributions from other sources or vice versa which can lead
to considerable inconsistencies in the interpretation of results.
In our SANS investigations we intended to study the mesoscopic contituents
of such poly-disperse FF using polarised neutrons as a labelling technique. Size,
distribution and composition of the particles as well as magnetic nanostructures
are determined systematically depending on the magnetic materials, surfactants
and carrier liquids and on the particle concentration.
The paper is organised as follows: In section 2 we recall the basic concept
of SANS and present the newly developed technique of nuclear and magnetic
contrast variation by using polarised neutrons. We will show how this combi-
nation allows constituents of magnetic liquids to be identified. Results of our
recent investigations on Co-FF, Fe3 04 -FF and Ba-hexaferrite-FF are presented
in the third section. It will be shown that this technique provides additional
information about the nature of the nanocrystalline magnetic core and magnetic
correlation as well as of the nonmagnetic shell which are not available by other
techniques. Some common features of the different materials investigated are
summarized in section 4.
2 Small Angle Neutron Scattering Technique
2.1 Basic Concept
Scattering of neutrons result from both nuclear and magnetic interactions be-
tween neutron and the nucleus and with the magnetic moment, respectively. The
formalism is described in detail in various monographs e.g. [18]. In this section
we recall some basic terms which are important for the understanding of the
present subject.
The concept and experimental set-up for SANS is very simple and schemati-
cally shown in Fig. 1. From the “white” spectrum of the neutron source (reactor
or spallation source) a small band of wavelengths Δλ is filtered out, collimated
Magnetic and Crystalline Nanostructures in Ferrofluids 35
and directed to the sample. The transmitted beam is absorbed in a beam stop
whereas neutrons scattered around the primary beam are registered in a de-
tector. The intensity distribution I(Q) has to be analysed as a function of the
scattering vector Q(reciprocal space) which by Fourier transform gives access to
correlation functions in real space i.e. to size, composition and magnetization of
inhomogeneities present in the material. Small angle neutron scattering (SANS)
is a special regime of low values of Q between typically 10−2 nm−1 and 5 nm−1
which allows to probe fluctuations on a length scale D∝2π/Q ranging from 1
nm to 500 nm.
Fig. 1. Sketch of a typical SANS instrument (V4 at HMI Berlin) with the option for
polarised neutrons (SANSPOL): The velocity selector picks out a small wavelength
band Δλ from the reactor spectrum. The beam is focussed in the collimator and di-
rected to the sample where part of the neutrons are scattered elastically at nanosized
inhomogeneities in an angle 2θ. The scattering intensity around the primary beam is
counted in a position sensitive area detector an analysed a function of the scattering
vector Q. For SANSPOL polarised neutrons are produced in the transmission polariser.
Two scattering intensities are measured subsequently with neutron spin (n+) and (n-)
after reversal in a spin flipper.
The signal is obtained by summing up the scattering amplitudes of all atoms

weighted by the phase shift at each atomic position. Elastic scattering of neutrons
with wavelength λ at an angle 2Θ leads to a momentum transfer Q according
to
Q = 4π sin Θ/λ (1)
and hence to a phase shift at an atomic position r of exp(iQr). In the SANS
regime of low spatial resolution, the discrete atomic scattering amplitude b can
be replaced by a locally averaged scattering length density η(r). A nuclear
36 A. Wiedenmann
scattering length density is defined by
ηN = Σci bi /Ωi (2)
where b is the nuclear scattering length, c the concentration and Ω the atomic
volume of the species i. η is easily calculated from the tabulated values of each
element e.g. for water and heavy water η (H2 O,D2 O) = ρ NA (2bH,D + bO )/MW
which, with the mass density ρ, the Avogadro number NA and the molecular
weight MW yields values of -0.56 1010 cm−2 and 6.33 1010 cm−2 , respectively.
For the interaction between the neutron spin S=1/2 with an assembly of
magnetic moments a magnetic scattering amplitude is defined similarly by

ηM = e2 γ/2mc2 Σci M ⊥ i /Ωi (3)
where only the projection of the magnetic moment Mi ⊥ onto a plane perpen-
dicular to the scattering vector Qcontributes to the interaction.Mi ⊥ is given in
units of Bohr magnetons μB and (e2 γ/2mc2 ) = 0.27 · 10−12 cm.
Throughout this paper we restrict our formulations to two phase-systems
where particles are embedded in a homogeneous matrix. The total scattering
amplitude of the particle is called “form factor” and defined by

F (QR) = dr3 Δηexp(iQr j ) = ΔηVp f(QR) (4)
where the contrast Δη is the difference between scattering length densities of

particle and matrix, i.e. Δη = ηp −ηmatrix . Note that the magnetic contrast Δη M
and magnetic form factor F ⊥ M (Q) are vectors and depend on the orientation
of the moment with respect toQ. The factor f(QR) depends only on the shape of
the particle and is known analytically for various simple geometrical units. E.g.
for spherical particles of radius R, f(QR) is given by the oscillating function
f (QR) = 3(sin(QR) − QRcos(QR))/(QR)3 (5)
The differential scattering cross section in diluted systems, where Np

identical particles scatters independently, is given by
dσ(Q)/dΩ = Np Fj2 (QR) (6)
For such simple mono-disperse systems the scattered intensity follows general
approximations which are independent of the shape of the particle: At low Q
(QR <1) the scattered intensity I(Q) is described by the Guinier law
I(Q → 0) = Δη 2 Np Vp2 exp(−Q2 R2g /3) for (QR < 1), (7)
from which a radius of gyration Rg can be derived. When the shape is known
the particle dimensions are obtained from Rg , e.g. for spheres of radius R,
Rg =(5/3)0.5 R. At large Q, scattering arises from the total surface of the particle
S according to the Porod approximation
I(Q → ∞)2π(Δη 2 S)Q−4 . (8)

On the other hand the integrated intensity is known as “invariant”

(dσ/dΩ)Q2 dQ ∝ Np Vp Δη 2 . (9)
Combining Eq. (7 - 9) one obtains some indications of size, shape and con-
centration of the particle. However, in most of materials the two extreme regions
are in practice not well defined due to additional scattering contributions, such
as grain boundaries, surface defects , dislocations or strong background.
In this low-Q regime of SANS there is no access to distances between atoms
or molecules; e.g. in simple liquids scattering in the SANS regime is independent
of Q and given by
I(Q) = I(0) = N < b >2 kB T χ(T ) + Iinc . (10)
The first term results from thermodynamic fluctuations, where χ(T) is the
isothermal compressibility and <b> is the average atomic scattering length ,
the second term results from the incoherent cross section of each atom which in
some cases might be very high such as for protons where Iinc is 80 barn/atom.
2.2 Polydisperse Multiphase Systems

Complications arise in polydisperse multiphase systems when different types of
particles j of different shape Fj (QR) and size distributions Nj (R) coexist which
all will contribute to the scattering signal according to

dσ(Q)/dΩ = Npj Fj2 (QR)Nj (R)dR. (11)
j
Nj (R)dR is the incremental volume fraction of particles of type j in the radius

interval between R and R + dR and Npj is the total particle number. Nj (R)
can be evaluated in principle from the scattering curves by the Indirect Fourier
transform method [20] or by model fitting using analytical functions. However,
if the particles of different type are of similar size their scattering contributions
superimpose within the same Q-range. In addition, all oscillations of I(Q) from
the individual form factors are then smeared out due to the size distributions
leading to more or less broadened signals.
A second complication appears in concentrated systems where interactions
between particles can no longer be neglected. The simplest case of excluded
volume interaction results from steric effects, i.e. the volume occupied by one
particle can not be occupied by a second one. Such an assembly of interacting
particles embedded in a homogenous matrix can be considered as a liquid, for
which the particle positions are determined by the interaction potential V (r ).
V (r ) is related to the pair correlation function g(r) which defines the proba-
bility to find a particle at a distance r from a first particle at the origin. The
Fourier transform of g(r) is measured in scattering experiments as the so called
“structure factor” S(Q) given by

S(Q) = 1 + Np [g(r) − 1] exp(iQr)dr (12)
38 A. Wiedenmann
The scattered intensity therefore depends on the form factors F (Qr) and
partial structure factors Sij (Q) of all particles of type i and j. This general
case is difficult to treat for which different approximations have been proposed
[19-21]. Analytical solutions are available only for mono-disperse systems; for
spherical particles of identical size it simplifies to the product
I(Q) = Np F 2 (QR)S(Q) (13)
The structure factor S(Q) will be unity when particles are fully uncorrelated
which is the case only for a perfect gas. In disordered condensed matter the
scattering at large values of Q is determined mainly by the particle form factor
alone, where S(Q)=1. For a repulsive potential such as excluded volume effects
or electrostatic repulsion S(Q) decreases at low Q and oscillates with Q when the
particles form a pseudo-periodic arrangement. Attractive interaction potentials
will give rise to an increase of the intensity at low Q which in practice might
also be assigned to polydispersity or aggregates.
2.3 Magnetic and Nuclear Scattering with Unpolarised Neutrons
In a classical SANS experiment, where the incoming monochromatic neutron

beam is unpolarised i.e. neutron spins are distributed at random, the scattering
intensity is the sum of the squared amplitudes (Eq.6) from individual magnetic
and nuclear contrasts. Due to the vector nature of F⊥ M (Q) the scattering profile
from a magnetic sample in an external magnetic field H is anisotropic and
described by
dσ(Q)/dΩ = A(Q) + B(Q)sin2 α (14)
where A(Q) and B(Q) are the isotropic and anisotropic terms, respectively and
α is the azimuth angle between the direction of the magnetic field H and the
scattering vector Q. For full alignment of all moments along the magnetic field,
A(Q)= FN 2 is solely of nuclear origin and B(Q) = FM 2 represents the magnetic
contribution resulting from the squared difference of the spontaneous magneti-
sation between particle and matrix.
In this cases the intensities measured perpendicular to the external magnetic
field I(Q⊥H) gives the sum of nuclear and magnetic contributions A(Q)+B(Q)
whereas I(Q//H) yields the nuclear contribution A(Q). When an area detector
is used the validity of Eq. 14 can be checked by analysing the 2D scattering
pattern I(α) which then yields both contributions with higher precisions. This
commonly used techniques has two limitations. First, when the magnetic contrast
is weak compared to the nuclear one (or vice versa, e.g. 10%) the anisotropy of
the 2D scattering signal will be very weak (1 %) due to the squaring of the
amplitudes. Such weak effects are very hard to be detected precisely in practice.
Second, since the signs of the contrast is lost in the squaring of the individual
nuclear and magnetic amplitudes the compositions, densities and magnetisation
of matrix and particles cannot be obtained in absolute scale.
2.4 Polarised Neutrons

For polarised neutrons where the neutron spins are aligned antiparallel denoted
by (+) or parallel (-) to a magnetic field H, four types of scattering process are
to be distinguished, two for conserving the neutron spin |F ++ |2 and|F −− |2 (spin
non-flip scattering: snf ) and two with reversal of the spin by the scattering
|F +− |2 and |F −+ | (spin-flip scattering: sf) [22,23]. Whereas coherent nuclear
scattering is nsf, differences in the absolute values of the magnetic moments give
rise to magnetic nsf and fluctuations of the magnetisation away from perfect
alignment yields sf scattering.
When the polarisation of the scattered neutrons is not analysed, the intensity
collected in the detector contains both snf and sf contributions and depends on
the polarisation state of the incident neutrons. We denote this technique as
SANSPOL. For simplicity we treat first the case of a mono-disperse system with
non-interacting particles where the magnetic moment is fully aligned [24]. When
magnetic moments and neutron polarisation are both parallel to an external
magnetic field the intensities for the two states are given by
I + (Q) = < |F ++ |2 > + < |F +− |2 >

= FN2 + {FM 2
− 2P (1 − 2f + )FN FM }sin2 α
I − (Q) = < |F −− |2 > + < |F −+ |2 >
= FN2 + {FM 2
− 2P (1 − 2f − )FN FM }sin2 α. (15)
P is the degree of polarisation defined by P = (n+ - n− ) / (n+ + n− ) where

n+ are the numbers of neutrons with spin antiparallel, and n− , parallel to H,
respectively. The reversal of the polarisation direction from (+) to (-) is achieved
with a spin-flipper of efficiency f± , which is close to unity (f(−) ≈1) when the
flipper is on, and zero when the flipper is off (f (+) =0).
The anisotropy of both scattering signals is again described by a relation
similar to Eq. (14),
I ± (Q) = A(Q) + B ± (Q)sin2 α (16)
where the nuclear scattering A(Q) = FN 2 is independent of the neutron polari-
sation. For polarised neutrons the anisotropic part B± (Q ) is different for the
two polarisation states and given by
B ± (Q) = FM
2
− 2P (1 − 2f ± )FN FM . (17)
The scattering intensity for polarised neutrons contains a magnetic-nuclear

cross term which is obtained from the difference between the intensities of the
two polarisation states according to
I − (Q) − I + (Q) = 4P f (−) FN FM sin2 α = Bint sin2 α (18)
The cross term is linear in the magnetic amplitude and hence makes it pos-
sible to determine the absolute value of the magnetic contrast with respect to
the nuclear contrast, i.e. magnetic moment and compositions of particles and
matrix.
40 A. Wiedenmann
The ratio γ = FM / FN between the magnetic and nuclear form-factors is di-

rectly obtained from the intensity ratio of both polarisation directions (”flipping-
ratio” FR) for Q perpendicular to H (α=π/2) by solving Eq. (19)
F R = I − (Q ⊥ H)/I + (Q ⊥ H) = (1 − 2P (1 − 2f )γ + γ 2 )/(1 − 2P γ + γ 2 ) (19)
or from the ratio
R2 = B − (Q)/B + (Q) = (γ − 2P (1 − 2f ))/(γ − 2P ) (20)
The ratios FR and R2 can be measured very precisely which allows the magni-
tude of γ to be determined much more accurately than for nonpolarised neutrons
using Eq. (6), for which the modulus of γ is given by |γ| = [B(Q)/A(Q)]1/2 .
Note that the average of the intensities for both neutron polarisations is given
by
[I + (Q) + I − (Q)]/2 = FN2 (Q) + [FM (Q)2 + 2P (f − 1)Fn FM ] sin2 α (21)
which corresponds to the intensity of non-polarised neutrons (Eq. 14) since the
second term in brackets vanishes for f → 1. For the paramagnetic case where
all moment directions are equally probable the scattering is fully isotropic and
independent of the polarisation and given by
I(Q)H=0 = FN2 (Q) + 2/3FM (Q)2 (22)
Now we consider the case of non perfect alignment of the magnetic mo-
ments of a ferromagnetic single domain particle of saturation magnetization Ms p
embedded in a nonmagnetic matrix [25]. For such particles superparamagnetic
behaviour is expected, where the orientation distribution of the magnetic mo-
ments as a function of an effective magnetic field Hef f and temperature follows
the Langevin statistics
L(x) = coth(x) − 1/x (23)
where x= M(R’)H ef f / k B T.
The total moment, M(R’ ), depends on the radius R’ of the particles accord-
ing to
M (R ) = 4πR3 m0 /3Ωat (24)
where m0 is the saturation value of the atomic magnetic moment. Using the
formalism presented in [6] SANSPOL cross sections for the case of a dilute
system of non-interacting particles are again described by Eq. (16). If we include
steric interaction effects via phenomenological structure factors S(Q//H) and
S(Q⊥H) the intensities parallel and perpendicular to the magnetic field are then
given by
I ± (Q, //H) = A(Q) = FN2 S(Q//H) + FM

2
2L(x)/x
±
I (Q ⊥ H) = FM [1 − L (x) − L(x)/x]
2 2
+S(Q ⊥ H)[FM L (x) + FN2 − 2P (1 − 2f ± )FN FM L(x)]

2 2
(25)
The isotropic term as obtained from the intensity parallel to His independent
of the polarization state. Beside the nuclear term F N 2 S(QH) it contains an
2
additional term of magnetic origin, FM 2L(x)/x , which vanishes for complete
alignment of all moments along H, when L(x)/x = 0. The anisotropic part
B ± (Q) is different for the two polarization states. Since the cross-term B int (Q)
as obtained from the intensity difference
Bint (Q, H) = 4P f FN FM L(x)S(Q ⊥ H) (26)
is linear in the magnetic amplitude the field variation of the magnetic moment
according to the Langevin statistics can be precisely determined from the cross-
term. Note that in Eq. (25) inter-particle correlations parallel to the field affect
only the nuclear part but correlations perpendicular to H influences nuclear,
magnetic and cross terms. We will see later that these features of Eq. (25) will
help to distinguish between inter-particle correlations and particular form-factors
which both may give rise to characteristic peaks in the SANS curves at low Q.
2.5 Combined Contrast Variation

The interpretation of SANS signals in polydisperse multiphase systems is gener-
ally not unique and requires special labelling techniques in order to identify and
separate individual contributions. One conventional technique consists in the
substitution of some atoms in the particle by isotopes with different scattering
lengths. E.g. subsequent replacement of hydrogen by deuterium (“Triangula-
tion”) [26] allowed the localisation of sub-units in biological macro-molecules
[27]. Similarly, in some cases the isotope composition of the matrix can be ad-
justed in such a way that the contrast vanishes (contrast matching), i.e. the
particle (or some part of it) will be “transparent” for neutrons. However, in
this kind of labelling techniques different samples have to be used which may
present unknown sample specific characteristics. Secondly, the original isotope
distribution may change after preparation. E.g. hydrogen atoms contained in
the organic molecules can be partly exchanged by deuterium from the contrast
matching isotope mixture in the solvent.
These ambiguities are removed when labelling is due to an intrinsic physical
variable which is independent of the sample. The neutron-spin in SANSPOL
is such a variable for magnetic systems [28]. This is illustrated in Fig. 2 for a
composite particle, where the different grey scales represent the scattering length
densities of a central magnetic core (η1 ), a nonmagnetic outer part (η2 ) and a
(±)
nonmagnetic liquid as matrix ηmatrix . Since η1 = η1nuc ± η1mag depend on the
polarisation states (+) or (-) (black and tiled in Fig. 2) the scattering contrasts
between core and shell as well as between core and matrix are different for both
polarisation states. The contrast between nonmagnetic shell and matrix Δη2 =
η2nuc − ηmatrix is independent of the neutron polarisation. Two scattering curves
I+ (Q) and I− (Q are measured alternatively on the same sample by switching
the neutron polarisation from (+) to (-). Both curves may then be different
solely as a result of the different magnetic contrasts. Further, the difference
42 A. Wiedenmann
Fig. 2. Illustration of the labelling technique for SANSPOL combined with contrast
matching as a grey scale representation of the scattering length densities η in a com-
(±)
posite particle: η1 of the central magnetic depends on the neutron polarisation. The
contrast between the nonmagnetic shell and matrix is adjusted using different isotope
mixtures of the solvent.
I+ (Q)-I− (Q) is proportional to the product η 1 mag ·η 1 nuc of one and the same
particle whereas nuclear and magnetic contrasts from different particles do not
contribute to the cross-term. This allows additional nonmagnetic particles to be
distinguished from magnetic particles even when they are of the same size.
In addition, the “colour” of the surrounding matrix can be adjusted using
the above mentioned contrast matching with different mixtures of isotopes in
the solvent.
Both scattering curves I+ (Q) and I− (Q), the difference I+ (Q)-I− (Q) and
FN (Q) und FM 2 (Q) as obtained from the averaged signal of Eq. 14 in all iso-
2
tope mixtures must be adjusted using the same structural model. This combined
contrast variation thus reduces considerably the number of possible structural
models and ensures the consistence and accuracy of the constrained model pa-
rameters.
2.6 Experimental
SANS measurements have been performed at the instrument V4 installed at

the BERII reactor of HMI, Berlin [29](Fig. 1). A horizontal magnetic field of
strength up to 1.1 T was applied at the sample position, oriented perpendicular
to the incoming neutrons. Polarized neutrons (n+ ) are provided by a transmis-
sion polarizing super-mirror cavity [30] which is contained in one of the three
tubes of the first segment of the collimator (Fig. 1). The polarisation direction
is conserved in a guide field downstream to the spin flipper. In this device the
polarisation can be reversed by injection of a radio-frequency superposed to a
magnetic field gradient. The two scattering intensities I(+) and I(-) are collected
alternatively with flipper off and on. The SANSPOL option, which can be set
without any modification of the instrument alignment, is characterized by a high
neutron flux of more than 30% of non-polarized neutrons, a high degree of polar-
ization (P = 95%) and by the high efficiency of the spin flipper (f(-)≈ 98%) at the
wavelength λ=0.6 nm used in this experiments. The reliability of this option has
been demonstrated by comparing the results of a SANSPOL study on magnetic
glass ceramics to those of a conventional SANS study [28]. By choosing three
distances between the sample and detector a Q range of 0.08 nm−1 to 3 nm−1
could be accessed which allows particles to be detected with sizes ranging from
0.5 to 80 nm.
We investigate ferrofluids where the nanoscaled magnetic particles are stabi-
lized against coagulation either by electrostatic repulsion or by coating the core
with organic chain molecules acting as surfactants. Magnetic moments of the
core materials (Co, Magnetite Fe3 O4 , and Ba-Hexaferrite), shell forming surfac-
tants (mono- and bi-layers) and carrier liquids (Water, organic solvents) have
been systematically varied in order to study the influence on the microstructures.
Co- and Magnetite ferrofluids have been prepared by Berlin Heart AG [41] and
Ba-Hexaferrite by R. Müller of IPHT, Jena Germany [31].
3 Results
3.1 Cobalt Ferrofluides
Diluted Co-FF. Colloidal solutions of ferromagnetic Cobalt can be prepared by
coating the nanosized particles with organic chain molecules C21 -H39 -N-O3 . The
hydrophilic head groups are considered to be in contact with the metal surface
and the hydrophobic tail pointing to the non-polar toluene as carrier liquid. We
performed a SANS study on dilute solutions with a nominal concentration of 0.5
vol. % of Cobalt as derived from saturation magnetisation measurements. Four
different mixtures of protonated and deuterated toluene, denoted as AF1-AF4
(Table 1) have been studied in a magnetic field of 1.1 T in which full alignment
of the magnetic moments is expected.
Table 1. Scattering length densities as obtained from a fit to a shell model using the
bulk value of η1 (nuclear) = 2.531010 cm−2 for Co.
Sample Content η (solvent) η 1 (mag) η (mag) η2 (shell)
C 7 D8 1010 cm−2 1010 cm−2 1010 cm−2 1010 cm−2
SANSPOL SANS
AF2 100% 5.65 4.29 4.38 0.56
AF4 43% 3.0 4.5 5.3 0.3
AF3 14% 1.6 1.52 1.62 0.9
AF1 0% 0.97 0.94 1.24 0.8
A first conventional SANS study showed that the scattering on of the sample
AF4 is dominated in the high Q range by a high level of the incoherent back-
ground (mainly from hydrogen contained in the solvent and surfactant). Nu-
clear and magnetic scattering contributions have been extracted and analyzed
44 A. Wiedenmann
Fig. 3. SANSPOL patterns in Co-ferrofluid AF4 for neutron spins antiparallel (I− )
and parallel (I+ ) to the horizontal field of strength H=1.1 T. The arithmetic mean [(I− )
+ (I+ )]/2 corresponds to the 2D pattern of non-polarized neutrons. The difference (I− )-
(I+ ) yields the interference term [equation (18)]which presents negative values in the
center.
in terms of number-distributions N (R) of non-interacting spherical particles.

For the magnetic contribution a maximum of N mag (R) appeared at R =3.3 nm
whereas from the nuclear contribution, a bimodal distribution had to be assumed
with maxima at R1 =1.12 nm and R2 =4.9 nm, respectively.
The 2-dimensional SANSPOL pattern for the same sample AF4 shown in
Fig. 3 are highly anisotropic with a dramatic change of the aspect ratios for
the two polarization states. We emphasize the particularity that only for the
polarization state I − a maximum appears in the outer part of the patterns. The
difference signal (I − - I + ) of Fig. 3, where all background contributions are
cancelled out, shows the angular dependence as expected from Eq. (16) with
negligible intensity along the direction of the magnetic field. Note that in the
inner part the intensities are negative and positive values in the outer part.
SANSPOL intensities perpendicular to the applied field I ± (Q⊥H) as obtained
by an adjustment of the 2-d pattern to Eq. (16) are compared for the different
solvents in Fig. 4a.
In the fully (AF2) and partly deuterated solvents (AF4, AF3) pronounced
maxima occur only for I − (Q⊥H) whereas I + (Q⊥H) decreases always contin-
uously with increasing Q. The position of the maximum of I − (Q⊥H) shifts to
lower Q for increasing content of hydrogen. The cross-term B int reverses the sign
from negative values below Q=0.61 nm−1 to positive values above, with a max-
imum at Q=0.8 nm−1 (Fig. 3). For Q larger than 1.35 nm−1 , B int is zero, which
proves that there is no magnetic contribution left. The ferrofluid AF1 in a fully
Fig. 4. SANSPOL of Co-ferrofluids in different mixtures of C7 D8 −C7 H8 : a) Intensities

I + (Q⊥H) (solid symbols) and I − (Q⊥H) (open symbols). Solid lines: Fit according to
a shell model using the parameters Table 1. b) Flipping ratio according to Eq. (19), c)
Scattering length density profiles for two isotope mixtures of the solvents. d) volume
distribution of a composite particle as a function of the outer radius (SANSPOL). The
results from an unpolarised SANS study could have been described by an artificial
bimodal distribution of spheres of radius R
protonated solvent shows only a splitting of I + (Q⊥H) and I − (Q⊥H) below

Q=1 nm−1 , where I − (Q⊥H) > I+ (Q⊥H). The flipping ratio FR derived from
Eq. (19) (Fig. 4b) are different for each solvent at low Q but follow nearly the
same Q dependence at high Q. This indicates that the surface of the magnetic
Co-particles must be surrounded by a layer of almost constant density different
from that of the solvent. While from non-polarized neutrons size distributions
corresponding to individual nuclear and magnetic units are derived, polarized
neutrons show that both contributions must result from a “composite” particle
built up by a magnetized core of Co atoms surrounded by a nonmagnetic surface
layer. As the simplest description of such a “composite” we use a shell model
consisting of a sphere with an inner core radius R’ surrounded by a concentric
shell of radius R. The form factor is given by
Fshell (Q) = [(Δη1 − Δη2 )fsph (QR ) + Δη2 fsph (Q(R))]Vp (27)
with the shape function for spheres as given by Eq. (5), i.e. fsph (x) = 3[sin(x) −
x cos(x)]/x3 .
46 A. Wiedenmann
The scattering contrasts with respect to the matrix are different for the
magnetic core and non-magnetic shell and given by
(±)
Δη1 = η1nuc ± η1mag − ηmatrix andΔη2 = η2nuc − ηmatrix (28)
respectively. In the present case, only Δη 1 (±) depends on the polarization. The
intensities were calculated according to I (Q⊥H)= N p F shell 2 (Q,R) N (R’ )
dR, where for the diluted case inter-particle correlations have been neglected i.e
S(Q)=1 in Eq. (25) . The number distribution N (R’ ) has been parameterized
using a log-normal number distribution of the core radius R’ according to
√
N (R )1/( 2πRσ)exp[− ln2 (R /R0 )/2σ 2 ] (29)
where R0 denotes the median and s the width of the distribution.
The thickness of the shell D is assumed to be constant, i.e. R=R’+D. The
parameters η 1 nuc , η 1 mag , η 2 nuc have been adjusted in a non-linear least square
fitting routine using the constraints of Eq. (28) for the contrasts Δη 1 (±) and
Δη 2 . The scattering length density of the solvent η matrix was known for the
H/D ratio and confirmed by the incoherent background level. The parameters
N p , R’, D and σ have been constrained to be identical for both polarization
states. The solid lines in Fig. 4 represent the calculated intensities I ± (Q⊥H). It
turned out that for all solvents this model function lead to one set of parameters
which are consistent with the intensities FN (Q)2 and FM (Q)2 as calculated for
the nuclear and magnetic contributions from the averaged signal according to
Eq. (14).
The distribution N (R’ ) was found to be rather sharp (Fig. 4d) corresponding
to a volume weighted average of the core radius of < R > = 3.7 nm and a
constant thickness of the shell of D= 2.47 nm. We emphasize that from non-
polarized neutrons alone it would not have been possible to derive the shell
model due to the very low nuclear contrast of the shell beside the high incoherent
background. The first maximum of the size distribution of spheres as obtained
from classical SANS is therefore identified as artificial and reflects in fact half of
the thickness of the shell.
The scattering length densities resulting from fits of both polarization states
using the same model function are included in Table 1. The values of η 2 for the
shell of the same order of magnitude as calculated for densely packed surfactant
molecules (η 2 = 0.33 1010 cm−2 ) and do not depend significantly on the solvent
composition. This supports strongly the conclusion that the shell of thickness
of about 2.4 nm formed by organic surfactants is nearly impenetrable for the
solvent in i.e. the Co-core is not in direct contact with the solvent. The values
of η 1 (mag) in AF2 and AF4 experimentally derived at H = 1.1 T correspond
closely to η 1 (mag) = 4.14 1010 cm−2 expected from Eq. (3) for bulk Co (m0 =
1.715 μB / atom , Ωat (Co) = 0.01099 nm3 /at.) Discrepancies appeared in the
samples AF1 and AF3 where much lower values for η 1 (mag) were found. The
reason for such anomalous low values is not yet clear; it might result from a lower
particle density, to some oxidation , or to non-magnetic amorphous surface layers
as reported for magnetic glass ceramics [32].
Variation of the magnetic field. For the solvent composition AF4 the cross-
term B int (Q,H ) has been evaluated from the difference I − (Q) − I + (Q) for dif-
ferent values of the external magnetic field and plotted in Fig. 5. All additional
non-magnetic contributions are eliminated in this difference [25].
B int (Q,H ) could be well fitted according to Eq. (26) by the product
FN FM L(x) alone showing that for this diluted sample no significant inter-particle
correlations are present or induced by the magnetic field, i.e. S(Q⊥H )=1 [38].The
ratio B int (Q,H ) / B int (Q,H=1.1T ) was found to be constant for all values of
Q. The H -dependence of this ratio is plotted in the insert of Fig. 5. The ex-
perimental points follow closely a Langevin function corresponding to spherical
Co-particles with average core radius < R > = 3.7 nm. This is in good agree-
ment with the prediction of Eq. (24) and with < R > as derived from the direct
model fit [33].
Fig. 5. Cross-term Bint (Q) for Co-ferrofluid AF4 as a function of Q at H = 1.1 T

(♦), 0.85T(∇), 0.25T(Δ), 0.06T(O) and 0 T ().Inset: Field variation of the ratio
Bint (Q, H))/Bint (Q, H =1.1 T) (symbols) follows the Langevin behavior for non-
interacting single domains particles of < R >=3.7 nm
Concentrated Co-FF. The setup of inter-particle correlation S(Q) under the

influence of magnetic field was studied in a second series of higher concentrated
systems with nominal Co-concentrations as about 1 vol.% (D1) and 5 vol. %
(D5) (”MF 239”) [34]].
The azimutally averaged intensities for unpolarised neutrons < I(Q) > plot-
ted in Fig. 6 illustrate how the scattering behaviour changes as a function of
Co-concentration and magnetic field. For random orientation < I(Q)H=0 >=
[FN2 + 2/3FM2
]S(Q) is expected at zero field while for full alignment in strong
magnetic fields < I(Q) > is given by < I(Q)Hmax >= [FN2 + 1/2FM 2
]S(Q). For
the sample D1% the scattering observed at zero field scales fully with that at
48 A. Wiedenmann
Fig. 6. Radial averaged SANS intensity of 1-vol% (triangles) and 5-vol% (circles)
Co-Ferrofluids in C7 D8 at H = 0 (full symbols) and H = 1.1 T (open symbols).
Fig. 7. SANSPOL intensities I + (Q⊥H) (solid triangles ) and I − (Q⊥H) (open tri-
angles) of 5vol% Co-Ferrofluids in C7 D8 and magnetic (squares) and nuclear (circles)
contributions from non-polarised neutrons. The solid line represents the common struc-
ture factor S(Q) used in the model fits.
H = 1 T, i.e. there is no change of the particle arrangement in the external field.

In the concentrated sample D5% a pronounced peak occurs at Q1 = 0.32 nm−1
when the magnetic field is turned on indicating the set-up of correlation between
the Co particles. This is confirmed by the SANSPOL results shown in Fig. 7. In
addition to the high Q features characteristic for the core-shell profile for both
polarisation states a sharp peak occurs at Q1 = 0.32 nm−1 in the concentrated
case D5% only i.e. the magnetic-nuclear cross term is clearly peaked resulting
from the structure factor S(Q). This example shows clearly that SANSPOL
allows definitely to distinguish between scattering maxima arising from inter-
particle correlations S(Q) and from the form-factor F(QR). Compared to Fig.
4 an additional isotropic nonmagnetic scattering contribution is superimposed
to I + and I − at high Q corresponding to a small amount (1-5% ) of spherical

particles with R ≈1.8 ±0.4 nm. This contribution results from excess surfac-
tant molecules present in the solutions and which might form spherical micelles.
Taking into account this additional contribution, all curves I ± (Q⊥H), F M 2
and F N 2 could be adjusted simultaneously by the same core-shell model as
described above with very similar values for the structural parameters. In the
concentrated sample D5% the set-up of strong inter-particle-correlation between
the Co-particles in an external magnetic field are clearly reflected by a structure
factor S(Q). All curves were well described using the same S(Q) as given by the
hard sphere model of Percus-Yevick [19]. The resulting S(Q) plotted in Fig.7 cor-
responds to volume fraction η=0.3 of hard spheres with a radius Rhc =8.65 nm.
Rhc is by a factor of about 1.5 larger than the total radius R + dR of the parti-
cles which implies a rather dense packing of the particles under the influence of
the magnetic field. It compares favourably with the hydrodynamic radius as ex-
perienced from field induced magneto-viscous damping [15]. Dipole interaction
between the particle moment should favour the formation of chains with the
moment and chain directions along the magnetic field. Such a chain formation
should give rise to stronger inter-particle correlation along H i.e. the structure
factor S(QH) should be different from S(Q ⊥ H) [35]. However, comparing the
intensities along H and perpendicular to H did not reveal any evidence for such
an anisotropy of the structure factor. S(QH) ≈ S(Q ⊥ H) seems to reflect
the rather dense packing of the core-shell particles in the concentrated systems
which is induced by an external magnetic field. A detailed study of the field
induced correlations are currently performed as a function of concentration and
core sizes.
3.2 Magnetite Ferrofluids

Ferrofluids based on Magnetite are available in aqueous solution which makes
them interesting as biocompatible materials for medical applications. Nanosized
Magnetite particles, Fe3 O4 , have been prepared by co-precipitation of ferric salt
mixtures with concentrated ammonium hydroxide [41]. The magnetite cores are
charge stabilized (denoted as ELEC) which optionally can be coated by different
surface active organic molecules. Size, composition and structural arrangement of
the molecules in this coating are almost unknown. In the first example, denoted
as LM, coating should consists on a bi-layer of dodecanoic acid (inner layer) and
C12 ethoxylated alcohol with 9 mol/mol ethoxy groups. The ferrofluid (DEX) has
a single dextrane shell, whose chains are believed to be entangled by subsequent
heat treatment. In a first series we investigated the structural parameters as a
function of the stabilizing surface materials with no organic shell (ELEC) and two
different layer materials (LM and DEX) [36]. For SANS contrast requirements
the content of D2 O with respect to H2 O in the carrier liquid is larger than 90%.
The volume fraction of magnetite was nominally about 1 vol.% in all 3 samples.
In the second series a combined contrast variation study have been performed
for LM in three different isotope mixtures H2 O/D2 O denoted as H0.2 D1.8 O,
H0.6 D1.4 , DHO [37].
50 A. Wiedenmann
Fig. 8. Charge stabilized Magnetite FF (ELEC) in H0.2 D1.8 O,: a) SANSPOL inten-
sities I + (Q⊥H) and I − (Q⊥H) where solid lines corresponds to a fit with using a
bimodal distribution of spheres with b) scattering length density profiles of small and
c) large spherical particles.
The SANSPOL intensities I ± (Q⊥H) as obtained by an adjustment of the

2-d pattern to Eq. (16) are compared to nuclear and magnetic contributions as
derived from Eq. (14) for the different samples in Figs. 8a-10a. In DEX and LM
samples I − (Q⊥H)) is lower than I + (Q⊥H) at high values of Q and exhibit
a crossover around Q=0.2 nm below which I − (Q⊥H)) < I + (Q⊥H). For the
ELEC no such crossover occurs where I − (Q⊥H)) > I + (Q⊥H) for all values
of Q.
The crossover phenomena observed in the polarized neutron data of DEX and
LM is a characteristic feature of “composite” particle similar to that observed
in Co-ferrofluids which is expected to be built up by a magnetized core of Fe3 O4
atoms surrounded by a nonmagnetic surface layer. In fact the solid lines in Fig.
9-10 for LM and DEX represent the intensities I ± (Q⊥H), I(nuc) and I(mag)
calculated with a lognormal distribution of the core-shell particles model as
described above. In ELEC the absence of a crossover of the intensities I − (Q⊥H)
and I + (Q⊥H) indicates clearly that in this FF the particles are not coated. The
scattering is therefore described by a lognormal distribution of spherical particles
(solid lines in Fig. 8). In all samples the scattering curves of I(mag) and I(nuc)
revealed some shoulder at low Q which is a strong indication of the presence of
a second fraction of much larger particles. This second contribution had to be

included in the fit. A spherical shape and a lognormal size distribution have been
assumed for these aggregates. In LM the combined contrast variation revealed
an additional nonmagnetic contribution resulting from free organic surfactants,
which had to be added in excess for stability reasons during the preparation
process.
Fig. 9. Magnetite FF with DEX coating in H0.2 D1.8 O,: a) SANSPOL intensi-
tiesI + (Q⊥H) and I − (Q⊥H) where solid lines corresponds to a fit with a distribution
of small core-shell particles and larger spherical aggregates with the corresponding
scattering length density profiles shown in b) and c). The low value of η 1 (nuc) in the
aggregates reflects inclusion of free surfactant molecules.
It turned out that for all samples this model function lead to consistent
parameters. The volume weighted size distributions are compared in Fig. 11.
For the small particles a rather sharp distribution N (R’ ) corresponding to a
volume weighted average of the core radius of < R >=6.73nm (LM), 4.8 nm
(DEX) and 4.0 (ELEC) a constant thickness of the shell of D = 2.61 nm and 2.4
nm (DEX). For the second fraction the volume averaged radius is by a factor of
2.5-3.5 times larger than that of the core.
The scattering length densities resulting from simultaneous model fits of both
polarization states as well as Inuc and Imag using the same model function are
presented in Fig. 8-10 (b,c).
From SANSPOL it turned out that for the small particles the scattering
length density of the core (η 1) is clearly higher than η solvent and higher than
52 A. Wiedenmann
that of the shell (η 2 ) for the samples DEX and LM. The magnetic moment for
Fe3 O4 as derived from the fit is very close to the bulk value. In addition, the
resulting scattering length density (Fig. 10) for the shell in LM was found to be
independent on the H/D composition of the solvent, which indicates that the
organic surfactant is nearly impenetrable for the solvent.
Fig. 10. a) SANSPOL intensities I − (Q ⊥ H)(open symbols) and I + (QH) (full sym-
bols) of Magnetite FF with LM-coating in different isotope mixtures of D2 O/H2 O and
b) corresponding scattering length density profiles of small bi-layer coated composites
particles and c)of spherical aggregates.
For the large particles η 1 is found to be lower than η solvent in LM and DEX
but higher in ELEC. Since the scattering intensity for the large particles de-
pends on the polarisation, they must also contain magnetic materials. However,
the observed ratio η (−) /η (+) was much smaller than that expected for pure mag-
netite. It corresponds to a magnetite content of only 22%. These result suggests
that the large particles consist of a loose and mixed arrangement of magnetite
together with surfactant molecules forming a larger aggregate. Size and volume
fraction of these aggregates seem to depend on the preparation conditions (see
Table 2). In ELEC where surfactants are absent the magnetic contrasts of both
fractions are very similar which indicates that also the aggregates presents the
ferrimagnetic ordering of magnetite.
Fig. 11. Volume weighted size distribution of magnetic particles in charge stabilized
ELEC (without shell), monolayer coated DEX and bilayer coated LM . The size of the
central core are very similar while that of the aggregates depends on the preparation
conditions.
In any case the presence of aggregates indicate that steric or electrostatic

screening is not fully efficient in these magnetite based materials.
3.3 Barium-Hexaferrite Fluids
Magnetic nano-particles of Ba-hexaferrite BaFe12-2xTixCoxO19 (x=0.8) have

been prepared [31] by a glass crystallisation method. Magnetic liquids are pre-
pared with oleic acid CH3 (CH2 )7 CH=CH(CH2 )7 COOH as surfactant and do-
decane as carrier liquid. Due to the high uniaxial crystalline anisotropy a large
magneto-viscous effect for barium ferrite Ferrofluids (FF) was expected which is
of potential technical interest. Samples with a particle concentration between 1
and 4 vol % were diluted in mixtures of protonated and deuterated dodecane,
denoted by H(content of 100% H), HD (64% H) and DD (33% H) [39].
The two-dimensional SANSPOL intensities for both polarisation states are
slightly anisotropic and consequently I + (Q ⊥ H) and I − (Q ⊥ H) show only
small differences at low Q values (Fig. 12). Therefore, the magnetic contrast is
very weak compared to the strong nuclear contrast. Again the simultaneous fit of
two SANPOL curves together with I(nuc) in all three isotope mixtures allowed
three contributions to be distinguished: i) magnetic “composite” particles ii)
free surfactant molecules giving an additional (non-magnetic) contribution at
high Q while iii) at low Q an additional magnetic component of aggregates
occurs. The final fits (solid lines in fig. 12a) correspond to the scattering length
densities of core-shell particles and aggregates of Fig. 12b and to the volume
size distributions of Fig. 12c. The magnetic particles have an averaged core
radius of 4.1 nm with a shell thickness of 1.8 nm. Since in all H/D mixtures
the same (theoretical) value of the contrast of the oleic acid shell was found we
conclude that the shell is impenetrable for the carrier liquid. The low value of the
scattering length densities found for the large particles with a radius of 9.8 nm
54 A. Wiedenmann
Fig. 12. SANSPOL cross sections I − (Q ⊥ H) (open symbols) and I + (Q ⊥ H) (full

symbols) for Ba-ferrite-FF in different H/D mixtures 100 % H (H), 64 % H (HD) and
33 % H (DD) of the dodecane a carrier liquid (a).The grey lines represent model curves
with the scattering length density profiles of core-shell particles (b) and of aggregates
(c) and the corresponding volume distributions shown in of d).
is attributed to aggregates of Ba-ferrite containing a large amount of surfactant

molecules (the overall of the magnetization is only 1/3 of the small particles.
The magnetic scattering contrast derived from SANSPOL to about 43% of
bulk magnetization in agreement with the measured magnetization of the pure
particles used in the preparation of the FF [40]. Non-stoichiometry or a non-
magnetic surface layers might be responsible for the observed low values of mag-
netization. Volume fractions f were calculated from the size distributions. About
1/2 of the total amount (12 vol %) of surfactants remains as free molecules in
the carrier liquid.. The volume fractions of the small ferrite cores were 2.3% (H),
2.6% (HD) and 1.6% (DD) which correspond to about 2/3 of the value from
magnetization measurements i.e. 2/3 of all ferrite particles are separated and
encapsulated with surfactants. The remaining 1/3 of ferrite particles are in the
aggregates which have volume fractions of about 4.5% to 7%.
4 Summary
The use of polarized neutrons in Small Angle Neutron Scattering is a power-
ful labeling technique for investigations of magnetic materials. In the present
study of magnetic colloids we combined this technique with the conventional
contrast variation using different isotope mixtures of the carrier liquid. This
combination allows magnetic and nonmagnetic particles to be distinguished and
density-, composition-, and magnetization profiles to be precisely determined.
The microstructure parameters have been evaluated in polydisperse multiphase
systems where magnetic materials (Co, Magnetit, Ba-Hexaferrite), shell forming
surfactants (mono- and bi-layers) and carrier liquids (Water, organic solvents)
have been systematically varied. The results are compared in Table 2 and sum-
marized as follows:
As a common feature three different components were identified in the mag-
netic liquids, magnetic composite particles, free organic shell molecules and mag-
netic aggregates. In diluted Ferrofluids which are stabilized by surfactants com-
posite particles are well described by a magnetic core and an organic shell of
constant thickness. The size of the core depends on material and preparation
conditions while the thickness of the shell is characteristic for the surfactant
materials. The shell was found to be of homogeneous density and almost impen-
etrable for the solvent. No shell structure was found in charge stabilized samples
where solvent molecules are in touch with the magnetic core. The magnetic or-
dering of diluted systems corresponds to non-interacting ferromagnetic single
domain particles. Small amounts of aggregates with lower densities and mag-
netization and a typical size of 2-4 times the core radius were identified. This
indicates that screening is incomplete in some systems leading to aggregates in
which surfactant molecules are included. Depending on preparation conditions
nonmagnetic contributions have been found which are ascribed to free organic
surfactants or micelles.
The set-up of strong field induced inter-particle correlations was detected in
Co- Ferrofluids with higher concentrations. The observed structure factor S(Q)
corresponds to an interaction of hard spheres with a radius which is by a factor
of about 1.5 larger than the total radius of the composite particles. Surprisingly,
no evidence for an anisotropic structure factor was found which implies a dense
packing of the particles rather than formation of chains as predicted for dipole
interactions.
The application of the combined contrast variation technique will enable a
more detailed insight in the nature of more complicated core-shell structures,
which are generally not accessible from other microscopic methods. This knowl-
edge is a necessary pre-requisite for potential biomedical applications of fer-
rofluids, where immunoassays or therapeutic agents should be connected to the
surfactant shell.
Table 2. Structural parameters resulting from SANSPOL investigations in Ferrofluids with different core and surfactant materials. 56
Average radii < R >, shell thickness D and volume fractions f of core in composites, aggregates (agg) and micelles (mic). fmagn
denotes the total volume fraction of magnetic particles as obtained from magnetization measurements.
Sample Shell fmagn < Rcore > Dshell fcore < RAgg > fAgg < RM ic > fM ic
vol % [nm] [nm] vol % [nm] vol % [nm] vol %
Co- Single-layer 0.5 3.7 2.4 0.3 – – –

A. Wiedenmann
Oleyl Sarkosyl C21 H39 N O3 1 3.7 2.2 0.6 – – 1.8 2

in Toluol 5 3.7 2.2 3.2 Rhc =8.7 – 1.8 5
Fe3 O4 DEX Dextran 1 4.8 2.4 16.8
in H0.2 D1.8 O entangled
single layer
Fe3 O4 LM in Double layer
H0.2 D1.8 O C12 H25 COOH 1 6.62 1.06 13.4 0.14 3.05 0.5
H0.4 D1.6 O /C30 H62 O10 1 6.7 2.7 0.24 17.8 0.07 2.62 0.7
H0.6 D1.4 O 2 6.7 2.7 1.04 22.8 0.32 2.62 1.2
HDO 1 6.7 2.7 0.26 12.3 0.28 2.62 1
Ba-Hexaferrit Single layer
in C12 H26 C17 H33 COOH
H 3.6 4.1 1.8 2.3 9.8 1 0.8 5.8
H:D = 67:33 3.9 4.1 1.8 2.6 9.8 1 0.8 8.7
H:D = 36:64 2.5 4.1 1.8 1.5 9.8 1 0.8 5.8
Acknowledgements
The author acknowledges the fruitful cooperation with his colleagues Dr. A.
Heinemann, Dr. A. Hoell, M. Kammel, and Dr. U. Keiderling of the SANS team
at HMI Berlin. Many thanks are given for sample preparation to Dr. N. Buske
and Dr. M. Gansau from Berlin Heart AG, and to Dr. R. Müller from IPHT
Jena, Germany.
This work is supported by the German DFG priority program “Colloidal
magnetic fluids”, Grant No WI-1151/2.
References
1. J. Loeffler, W. Wagner, H. Van Swygenhoven, & A. Wiedenmann, Nanostruct.
Mat., 9 (1997), 331-334.
2. A. Wiedenmann , J. Appl. Cryst., 30 (1997), 580-585.
3. J. F. Loeffler, H. B. Braun, W. Wagner, G. Kostorz, A. Wiedenmann
Materials Science and Engineering A, 304-306, 31 (2001), 1050-1054.
4. A. Michels,, J. Weissmüller, A. Wiedenmann, J.S. Pedersen, J.G. Barker
Philosoph. Magazine Letters 80 (2000), 785-792.
5. A. Wiedenmann, U. Lembke, A Hoell,., R. Müller, & W. Schüppel Nanostruct.
Mat., 12 (1999), 601-604.
6. J. Kohlbrecher, A. Wiedenmann, H. Wollenberger, Z. Physik, 104(1997),1-4.
7. A. Heinemann, H. Hermann, A. Wiedenmann, N. Mattern, K. Wetzig J. Applied
Cryst. 33(2000),1386-1392.
8. S.W. Charles, JMMM (2002).
9. P.P.Gravina et al., JMMM (2002).
10. E. Dubois et.al, Langmuir 13(2000),5617.
11. G. A.van Ewijk, PhD Thesis Utrecht (2001).
12. B. Huke et al., Phys. Rev. E 62(2000),6875.
13. S. Odenbach, K.Raj, Magnetohydrodynamics 36(2000),379-386.
14. B. U. Felderhof Phys. Rev. E 62(2000),3848.
15. J.P.Embs et al. Phys. Rev. E(2001), preprint
16. E. Dubois, V. Cabuil, F. Boué, R. Perzyski, J.Phys. Chem. 111(1999),7147-7160.
17. F. Cousin, V. Cabuil, Prog. Colloid Polym. Sci. 2000),77-83.
18. G. Kostorz in Treatease on Materials Science and Technology, Acad. Press (1979)
ed. G. Kostorz, 227-290.
19. J.K. Percus, G.J. Yevick Phys. Rev 110(1959),1-13.
20. J. Brunner-Popela, O. Glatter J. Appl. Cryst 30(1997),431-442.
21. J.S. Pedersen J. Appl. Cryst. 27(1994),595-608.
22. R.M. Moon, T. Riste, & W.C. Koehler, Phys. Rev.(1969) 181, 920-931.
23. R. Pynn, &J. B.Hayter. Phys. Rev. Letters , 51, 710-713.
24. A. Wiedenmann, Mat. Science Forum, 312-314(1999) 315. J. Metastable and
Nanocrystalline Materials, 2-6 (1999), 315-324.
25. A. Wiedenmann, Physica B 297 (2001), 226-233.
26. R. P. May and K. Ibel and J. Haas, J. Appl. Cryst., 15 (1982), 15-19.
27. R. Stegmann, E. Manakova, , M. Rössle, H. Heumann, S. Axmann, A. Plückthun,
T. Hermann, R. May, A. Wiedenmann. J. Structural Biology 121(1998),30-30.
28. A. Wiedenmann, J. Appl. Cryst. 33 (2000), 428-432.
29. U. Keiderling & A. Wiedenmann, A. Physica B, 213&214(1995). 895-897.
58 A. Wiedenmann
30. T. Keller, T. Krist, A. Danzig, U. Keiderling, F. Mezei, A. Wiedenmann, J. Nuclear

Instruments A451(2000), 474-479.
31. R. Müller et al, J. Magn. Magn. Mat. 201(1999), 34.
32. A. Hoell, A. Wiedenmann, U. Lembke, Physica B 276-278(2000), 886-887.
33. A. Wiedenmann, Magnetohydrodynamics 37(2001),236-242.
34. A. Wiedenmann, A. Hoell, M. Kammel, JMMM (2002) in press.
35. J. B.Hayter. J. Appl. Cryst., 21(1988), 737-742.
36. M. Kammel, A. Hoell, A. Wiedenmann, Scripta Materialia 44(2001), 2341-2345.
37. M. Kammel, A. Wiedenmann, A. Hoell, JMMM (2002) in press.
38. A. Heinemann, Tatchev, A. Hoell, A. Wiedenmann J.Appl. Cryst.to be published
39. A. Hoell, M. Kammel, A. Wiedenmann, JMMM (2002) in press
40. A. Hoell, R. Müller, A. Wiedenmann, W. Gawalek, JMMM (2002) in press.
41. N. Buske, DE patent .197 58 350 (1997)
Basic Equations
for Magnetic Fluids with Internal Rotations
Ronald E. Rosensweig
Consultant, Summit, NJ (U.S.A.),

Former Blaise Pascal Professor, ENS Fondation, Paris
Abstract. Several authors have attempted with varying success to derive a complete
set of basic equations for magnetic fluids having internal rotations. In this work a com-
plete set of governing equations is derived on the basis of dynamic balance relationships
with the dissipation function derived from thermodynamic consideration. The magneti-
zation relaxation equation is thereby determined from requirement of positive entropy
production along with a complete set of well-accepted constitutive laws. The analysis
employs the Minkowski expression of electromagnetic momentum and assumes that the
product of Maxwell stress and velocity for polarized media contributes to the energy
balance on the same footing as contact stresses of pressure and viscous origin.
1 Introduction
Magnetic fluids, also known as ferrofluids, are stable suspensions of subdomain
ferrimagnetic or ferromagnetic fine particles in a liquid carrier. Early theory of
magnetic fluids assumed that the magnetization is colinear with the magnetic
field at all times [1]. More generally, the magnetic particles lag the changing
field and the description of the flow becomes more complex. The work of Dahler,
Scriven, and Condiff [2] exposed the role of asymmetric stress in this circum-
stance, although the relaxation equation was postulated rather than derived.
Independently the work of Shliomis [3] has been influential in the development
of the science, albeit the relaxation law is derived separately [4] from the main
theory [5].
Irreversible thermodynamics provides a framework for the consistent descrip-
tion of dissipative processes. In their well-known monograph deGroot and Mazur
[6] arrive at a plausible expression for entropy production of a polarized medium
only after introducing an artificial definition of electromagnetic energy density.
Also, their work did not treat the case when the direction of the magnetization
differs from the direction of the field. Felderhof and Kroh [7] building on the De-
Groot and Mazur treatment present a theory that yields a plausible form of the
relaxation relationship when magnetization and field are not colinear. Yet the
derivation is not transparent to us for the reason cited, and is incomplete; e.g.,
the asymmetric hydrodynamic stress tensor is postulated rather than deduced.
A main objective of the present work is to circumvent the infelicities cited
above while employing the irreversible thermodynamic framework. We utilize
with modifications the formulation of Shizawa and Tanahashi [8] with its method-
ology based on integral formulation of equations governing the balance of linear

62 R.E. Rosensweig
momentum, angular momentum, and energy. In the latter the electromagnetic

work is formulated as the product of electromagnetic stress and the velocity
of the fluid. So far as we know, it is the only prior work that attempts that
formulation. However, their analysis produced an unsatisfactory result for mag-
netization relaxation. In the present work this deficiency is overcome using a
strict formulation of entropy production rate
In addition, a recent criticism has arisen concerning the mentioned use of the
generalized Maxwell stress tensor in formulating the balance of energy [9]. The
present work shows that incorporation of the stress in the energy formulation in
fact leads deductively to a self-consistent exposition yielding a complete set of
governing equations and constitutive relationships including the Shliomis relax-
ation equation. Furthermore, there has been uncertainty concerning whether the
Minkowski expression D × B for electromagnetic momentum or the Abraham
expression 0 μ0 E × H is the appropriate one. The present work is based on the
Minkowski expression, and helps validate its use.
In this work we essentially trace the development in [8] up to the point of
formulating entropy change in the fluid dynamic system. A detailed exposition is
employed in developing the well-known balances to set the pattern for consistent
treatment of unfamiliar balances and the work treats conductive magnetic fluid.
The SI system of units is employed throughout, and the conventions employed
in the dyadic notation are defined in the Appendix.
2 Equations of Electromagnetic Field

The well-known system of equations governing the electromagnetic field in the
presence of matter is the set of four Maxwell equations.
∂B
=∇×E (1)
∂t
∂D
+j=∇×H (2)
∂t
∇·B=0 (3)
∇ · D = ρe (4)
3 Law of Conservation of Mass

A constant mass of fluid in a flow field occupies a volume Vm that deforms with
time. Expressed in integral form this may be written as

d
ρdV = 0 (5)
dt
Vm
d
where dt ∂
= ∂t + v · ∇ is the material derivative and is the mass density. The
Reynolds transport theorem for any function Ψ of position and time (scalar,
Basic Equations for Magnetic Fluids 63
vector or tensor component) is the kinematic relationship having the following

visualizable form:

d ∂Ψ
Ψ dV = dV + (n · vΨ )dS (6)
dt ∂t
V V S
where n is the unit outward facing normal to the surface. In words, the total
time rate of change equals the time rate of change within the volume plus the
net convection out of the volume through the surrounding surface. Applying the
Gauss divergence theorem to the surface integral term transforms it to a volume
integral yielding the transport theorem in alternate form as

d ∂Ψ
Ψ dV = + ∇ · (vΨ ) dV (7)
dt ∂t
Vm Vm
The theorem applies with Ψ a vector or a tensor as may be demonstrated by

summing the components.
Identifying Ψ with and invoking the arbitrariness of the volume yields the
well-known equation of continuity in differential form.
∂ρ
+ ∇ · (ρv) = 0 (8)
∂t
This transforms easily to a corollary form
1 dρ
= −∇ · v (9)
ρ dt
A corollary of the Reynolds transport theorem is obtained from Eq. (7) ,

substituting ρΨ and employing the continuity equation in the form of Eq. (9).
This yields
d dΨ
ρΨ dV = ρ dV (10)
dt dt
V V
4 Balance Equation of Linear Momentum
The continuum extension of Newton’s fundamental law states that a change in

momentum of a fixed mass system with respect to time is equal to a sum of
surface and body forces. Defining the system as the contents of a volume Vm
enclosed by surface Sm the balance equation of momentum is written in the
form
d
(ρv + g)dV = tn dS + ρbdV (11)
dt
Vm Sm Vm
where g is the electromagnetic momentum vector, tn the surface stress vector

representing the combination of pressure, viscous and electromagnetic effects,
64 R.E. Rosensweig
and b body force density excluding electromagnetic effects. A definite expression

for g is not required until a later point in the development. Strictly speaking
we consider that any apparent body force arises due to transmission via field of
surface stresses from afar. However, when the presence of matter in the system
perturbs the field to a negligible extent, it is convenient to formulate body force
as b in the above. The body force of gravity furnishes an example of b for
problems we consider.
From the Reynolds transport theorem of Eq. (10),

d dv
(ρv)dV = ρ dV (12)
dt dt
Vm Vm
From Eq. (7) with Φ = g,

d ∂g
gdV = + ∇ · (vg) dV (13)
dt ∂t
Vm Vm
Next, rewriting the first term of the right side of Eq. (11) using Cauchy’s
fundamental theorem: tn = n · T̃ where T̃ is the total stress vector (sum of
viscous, pressure, and electromagnetic components), transforming the surface
integral into a volume integral by applying Gauss’ theorem, collecting terms and
noting that the volume of integration is arbitrary, the equation of motion in
Cauchy (unconstituted) form is obtained.
dv ∂g
ρ + = ∇ · T + ρb (14)
dt ∂t
where
T ≡ T̃ − vg (15)
5 Balance Equation of Angular Momentum

Fundamental law for angular momentum states that a change of angular mo-
mentum of a system with respect to time is equal to a sum of moments due
to surface and body forces. In integral form the balance equation of angular
momentum is expressed as follows:

d
[r × (ρv + g) + ρs]dS = (r × tn + λn )dS + (r × ρb + ρl)dV (16)
dt
Vm Sm Vm
where r is the position vector of a fluid element relative to an arbitrary fixed

origin, s angular momentum density (spin) per unit mass, λn surface couple
density, and l volume couple density transmitted from afar. The remarks above
concerning b apply as well to l.
The various terms of Eq. (16) transform as follows:

d d
r × ρvdV = ρ (r × v)dV
dt dt
Vm Vm

dr dv
= ρ ×v+r× dV
dt dt
Vm

dv
= r×ρ dV (17)
dt
Vm
where the first equality results from the Reynolds corollary of Eq. (10), and the
last equality from the circumstance that dr/dt = v.
Multiplying and dividing the integrand of the next term in Eq. (16) by ,
using the Reynolds transport theorem of Eq. (10) and then the equation of
continuity of Eq. (9), and noting that dr/dt = v, the term transforms as follows.

d d g
(r × g)dV = ρ r× dV
dt dt ρ
Vm Vm

dg
= (v × g) + r × + g∇ · v dV
dt
Vm

∂g
= − : vg + r × + ∇ · (vg) (18)
∂t
Vm
Using the Reynolds transport theorem of Eq. (10) the spin term on the left
side of Eq. (16) becomes

d ds
ρsdV = ρ dV (19)
dt dt
Vm Vm
Using the Cauchy theorem tn = n · T̃ and Gauss’ divergence theorem the

first term on the right side of Eq. (16) transforms as follows.

(r × tn )dS = − n · T̃ × r dS
Sm Sm

=− ∇ · T̃ × r dV (20)
Vm

Introducing the identity −∇ · T̃ × r = r × ∇ · T̃ − : T̃, see proof in
[16, p. 247], Eq. (20) becomes

(r × tn )dS = r × ∇ · T̃ − : T dV (21)
Sm Vm
66 R.E. Rosensweig
From the definition of the couple stress vector, λn = n · Λ, where Λ is the

couple stress tensor, and thus

λn dS = (n · Λ)dS
Sm Sm

= (∇ · Λ)dV (22)
Vm
Substituting the collection of transformed terms into the angular momentum

balance of Eq. (16) yields

ds dv ∂g
ρ +r× ρ + − ∇ · T̃ − vg − ρb dV
dt dt ∂t
Vm

= ρl + ∇ · Λ − : T̃ − vg dV (23)
Vm
Substituting the definition of T from Eq. (15), noting that the bracketed
term in the cross product with the radius vector disappears by virtue of the
equation of motion of Eq. (14), and invoking the arbitrariness of the volume
element yields the balance equation of internal angular momentum.
ds
ρ = ρl + ∇ · Λ − : T (24)
dt
For a monodispersion of spherical particles spin can be expressed as s = IΩ/
so that spin rate is given by
ds dΩ
ρ =I (25)
dt dt
where I is moment of inertia per unit mass and Ω is average angular velocity of
particles about their own center.
6 Equilibrium Thermodynamic Relationships

Thermodynamics deals with changes in states of matter, which for polarized
matter depend on P and M , the electric and magnetic polarizations. The po-
larizations are defined in terms of the field variables appearing in Maxwell’s
equations.
D = 0 E + P (26)
and
B = μ0 H + M (27)
Here we have chosen to absorb the factor of μ0 into the M , a deviation from
standard SI nomenclature, so that M and P appear on a symmetrical basis
and M is expressed in units of tesla. The volumetric density of polarization
work is given by E · dD and H · dB [10]. The combined first and second law
expression (Gibbs equation) written for the contents of a fixed unit volume of
space and incorporating these work terms is absent a pressure term and, in its
place contains a term proportional to the chemical potential of the medium,
considered to be of constant composition. Later in the present study the Gibbs
equation is required in a form in which internal energy and entropy are expressed
on the unit mass basis. Conversion from the unit volume basis to unit mass basis
yields the following expression [11]

du(e) = T ds(e) + p(e) dρ/ρ2 + E (e) · d D (e) /ρ + H (e) · d B (e) /ρ (28)
where superscript (e) denotes equilibrium value. u(e) is internal energy per unit
mass and s(e) entropy per unit mass. The same expression is derived by Chu
[12] based on a roundabout treatment, and adopted by Tarapov [13] and Blums
et al. [14]; our development in [11] shows that the result follows from classical
thermodynamics alone. The p(e) denotes the value of pressure in the presence of
the fields and it differs from the pressure p0 (, T ) that attains in the absence of
fields.
The analysis of [11] relates the two pressures as follows:

p(e) = p0 (ρ, T ) + 0 E (e)2 + μ0 H (e)2 /2
E H
(e) (e)
+ (∂υP/∂υ)T,H,E dE (e) + (∂υM/∂υ)T,H,E dH (e)
0 0

≡ p + 0 E (e)2
+ μ0 H (e)2
/2 (29)
where p0 (, T ) is ordinary pressure and υ = 1/ρ is specific volume. Naturally,

when E (e) and H (e) are absent, p(e) reduces to p0 (, T ). Also, from Eq. (29), the
pressure-like variable p is given by
E H
(e) (e)
p = p0 (ρ, T ) + (∂υP/∂υ)T,H,E dE (e) + (∂υM/∂υ)T,H,E dH (e) (30)
0 0
Substituting for p(e) from Eq. (29) into Eq. (28) and expanding the elec-
tromagnetic work terms yields the following form of the Gibbs equation after
rearrangement and substitution from Eqs. (26,27).
ρdu(e) = ρT ds(e)

+ p − 0 E (e)2 /2 − μ0 H (e)2 /2 − P (e) · E (e) − M (e) · H (e) dρ/ρ
+ E (e) + H (e) · dB (e) (31)
This form of the Gibbs equation with each term expressed on the per-unit-volume
basis is employed subsequently below.
68 R.E. Rosensweig
7 Balance Equation of Energy

The balance equation of energy is derived from the following integral equation
expressing time rate of change of total energy in a system of fixed mass and
variable volume as the sum of work done on the system and energy supplied to
the system.

d
ρ u + v 2 /2 + IΩ 2 /2 dV = tn · vdS + ρb · vdV
dt
Vm Sm Vm

+ λn · ΩdS + ρl · ΩdV − n · qdS
Sm Vm Sm

− n · (E ∗ × H ∗ ) dS + ρRdV (32)
Sm Vm
The u is internal energy per unit mass, q is the heat flux vector and R is
volumetric heat release rate, e.g., due to radioactive decay. The integrand of
the first term on the right side treats all components (viscous, pressure and
electromagnetic) of the stress vector on equal basis in expressing the work rate
as the scalar product with the velocity vector 1 .
tn = n · T̃ (33)
T = T̃ − vg (34)
The integrand of the sixth term on the right side is the Poynting vector
expressing electromagnetic energy flux density. Since the integration is over a
closed material surface, the Poynting vector is evaluated in the moving medium
as indicated by the asterisks suffixed to the vector fields. The electric field vector
E ∗ and magnetic field vector H ∗ are related to the components of vectors E
and H observed in a stationary coordinate system as follows [10].
E ∗ = E // + γ(E ⊥ + v × B) (35)
H ∗ = H // + γ(H ⊥ − v × D) (36)
where −1/2
γ 1 − (v/c)2 (37)
In Eqs. (35,36), v × B and v × D are respectively the apparent electric field
and magnetic field produced when an observer crosses B and D with velocity
v. Since the speed v of the suspensions can be assumed much smaller than light
1
Inclusion of the electromagnetic stress in the energy balance in this manner has
been questioned recently as unphysical [9]. Our thesis is, because Maxwellian stress
applies in linear and angular momentum balances, for consistency it should apply in
the energy balance an well, on the same footing as contact stresses of pressure and
viscosity.
speed c, (v/c)2 may be neglected in the equations. As a result, the coefficient γ

of the transformation (Lorentz transformation) can be set to unit and hence2
E ∗ = E + v × B, H∗ = H − v × D (38)
Applying Reynolds’ transport theorem of Eq. (10) to the left side of Eq. (32),
rewriting the first and third terms of the right side with Cauchy’s fundamental
theorem, transforming them into volume integral form together with the other
surface integral forms, and then noting that the volume of integration is arbitrary
yields

du dv dΩ
ρ + ρ − ∇ · T̃ − ρb · v + ρI − ∇ · Λ − ρI · Ω
dt dt dt
T
= T̃ : ∇v + λT : ∇Ω − ∇ · (E ∗ × H ∗ ) − ∇ · q + ρr (39)
T
where the identity ∇ · T̃ · v = ∇ · T̃ · v + T̃ : ∇v is employed with super-
script T denoting transpose. The Poynting term can be expressed in terms of
variables of the stationary coordinate system by the use of Eq. (38). This yields
the following representation:
− ∇ · (E ∗ × H ∗ ) = −∇ · (E × H) − [E · {∇ × (v × D)} + H · {∇ × (v × B)}]
+ [(∇ × E) · (v × D) + (∇ × H) · (v × B)]
+∇ · [(v × B) × (v × D)] (40)
Using the Maxwell relationships of Eqs (1) and (2) the third term of the right
side of Eq. (40) transforms as follows.
[(∇ × E) · (v × D) + (∇ × H) · (v × B)]

∂g
=− · v + j ∗ · (v × B) (41)
∂t
where
j ∗ = j − ρe v (42)
g =D×B (43)
∗
j is electric current density as observed in a moving coordinate system for the
case γ = 1, and g is electromagnetic energy density3 . The second term on the
right hand side of Eq. (42) is introduced permissably as its scalar product with
(v × B) in Eq. (41) vanishes. The last term in Eq. (40) transforms as follows by
vector identity and the definition of g:
∇ · [(v × B) × (v × D)] = v · (B × D) ∇ · v + v · ∇ (v · B × D)
= − (v · g) ∇ · v − v · ∇ (v · g)
= − (v · g) ∇ · v − v · (v · ∇g) − v · (g · ∇v)
= − [∇ · (vg)] · v − (vg : ∇v) (44)
2
A Galilean derivation of these relationships and Eq.(42) is given in Appendix B
3
This choice of g agrees with the Minowski definition but differs from the Abraham
definition which is 0 μ0 E × H. See references in [7]
70 R.E. Rosensweig
By substituting Eqs. (44) and (41) into Eq. (40) and substituting the result into
Eq. (39), taking account of the linear momentum balance of Eq. (14) and the
internal angular momentum balance of Eq. (24), and the definition of T̃ from
Eq. (15)the balance of energy relationship is expressed as follows:
du
ρ = TT : ∇v + ΛT : ∇Ω − T : · Ω − ∇ · (E × H)
dt
− [E · {∇ × (v × D)} + H · {∇ × (v × B)}]
+j ∗ · (v × B) − ∇ · q + ρr (45)
Unknown parameters at this stage are stress tensor T, couple stress tensor Λ,
polarizations M and P , electric current density vector j ∗ and heat flux density
vector q.
Eq. (45) for internal energy is not yet in the form that is needed as the Poynt-
ing term in the equation requires further transformation. From the mathematical
identity for a vector triple product we have
∇ × (v × B) = (B · ∇) v − B (∇ · v) − (v · ∇) B (46)
and
∇ × (v × D) = (D · ∇) v − D (∇ · v) − (v · ∇) D + ρe v (47)
where use is made of Eqs. (3) and (4). Next the left sides of Eqs. (1) and (2) are
rewritten in the material time derivative form by adding (v · ∇)B and (v · ∇)D
to both sides of the equations, respectively. Then using Eqs. (46) and (47) the
following equations are obtained.
dB
= −∇ × E + (B · ∇) v − B (∇ · v) − ∇ × (v × B) (48)
dt
dD
= −j + ∇ × H + ρe v + (D · ∇) v − D (∇ · v) − ∇ × (v × D) (49)
dt
Scalar multiplication of both sides of (48) and (49) by H and E respectively,
then adding the equations and rearranging, yields the desired expression for the
divergence of the Poynting vector.
dD dB
∇ · (E × H) = −j ∗ · E − E · −H ·
dt dt
+ {ED + HB − (D · E + B · H) I} : ∇v
− [E · {∇ × (v × D)} + H · {∇ × (v × B)}] (50)
Substituting this expression into Eq. (45) yields the energy balance in the
following form.
du
ρ = TT : ∇v + ΛT : ∇Ω − T : · Ω
dt
− [ED + HB − (D · E + B · H) I] : ∇v
dD dB
+E · +H · + j ∗ · E ∗ − ∇ · q + ρr (51)
dt dt
where, again
E∗ = E + v × B (52)
Substituting the definitions of D and B from Eqs. (26,27) into the electro-
magnetic work terms of Eq. (51) after minor algebraic transformation yields the
energy balance in final form.
dum
ρ = TT : ∇v + ΛT : ∇Ω − T : · Ω
dt
− [ED + HB − (D · E + B · H) I] : ∇v
dP dM
+E · +H · + j∗ · E∗ − ∇ · q
dt
dt
μ0 H 2 0 E 2
+ρR − + (∇ · v) (53)
2 2
where
0 E 2 μ0 H 2
um ≡ u − − (54)
2ρ 2ρ
The um of Eq.(54) represents internal energy of the medium exclusive of field
energy associated with the space the medium occupies.
8 Determination of Entropy Production Rate

Dividing both sides of the Gibbs equation (31) by dt and substituting from Eq.
(9) gives

du(e) ds 0 E (e)2 μ0 H (e)2
ρ = ρT − p− − −P ·E −M ·H
(e) (e)
∇·v
dt dt 2 2
dD (e) dB (e)
+ E (e) · + H (e) · (55)
dt dt
H (e) termed the “effective field” in [5] and E (e) are the equilibrium values
of magnetic and electric field, respectively, that would produce the magneti-
zation M or electric polarization P actually present in a given fluid element.
Accordingly, substitution was made of P = P (e) and M = M (e) .
Substituting for D (e) and B (e) in the electromagnetic work terms of Eq. (55)
from Eq. (26,27) with (e) superscripted permits the equation to be expressed in
final form as:
ds
(e)
dum
ρ = ρT − p − P · E (e) − M · H (e) ∇ · v
dt dt
dP dM
+E (e) · + H (e) · (56)
dt dt
(e)
where um is the equilibrium value of um .
0 E (e)2 μ0 H (e)2
m ≡u
u(e) − −
(e)
(57)
2ρ 2ρ
72 R.E. Rosensweig
(e)
um expresses the internal energy of the equilibrium medium exclusive of the
field energy associated with the space occupied by the medium. The internal
energy um = um (t) appearing in the energy balance of Eq. (53) can be regarded
as that of the medium at equilibrium though the fields E and H present are not
the equilibrium fields4 . The assertion, held valid at all instants, is expressed as
follows.
um (t) = u(e)
m (t) (58)
and so
(e)
dum (t) dum (t)
= (59)
dt dt
From thermodynamics, entropy change between states must be evaluated by
a reversible process that leads from the initial state to the final state. Thus, an
(e)
equation for entropy change is obtained by substituting for dum /dt and dum /dt
in Eq.(59) from Eqs. (53) and (56), respectively. The result is
ds
ρT = Ttv : ∇v + Λt : ∇Ω − ∇ · q + ρR − T : · Ω
dt
+ [(D · E + B · H)] I : ∇v − P · E (e) + M · H (e) I : ∇v
dM dP
H − H (e) · + E − E (e) · + j∗ · E∗
dt dt
− 0 E 2 + μ0 H 2 ∇ · v (60)
where Tv is viscous stress

Tv = T + pI − Tm (61)
and Tm is the Maxwell type stress vector for the polarizable fluid.

0 E 2 μ0 H 2
Tm = − + I + DE + BH (62)
2 2
The fifth term on the right side of Eq. (60) transforms as follows:
T : · Ω = Tv : · Ω + (P × E + M × H) · Ω
= Tv : · Ω − E · (P × Ω) − H · (M × Ω)

= Tv : · Ω − E − E (e) · (P × Ω)

− H − H (e) · (M × Ω) (63)
where E (e) · (P × Ω) = H (e) · (M × Ω) = 0 because E (e) is colinear with P ,

and H (e) with M .
4
In the treatment of [8] E and H are not distinguished from E (e) and H (e) nor are
the internal energy differences of Eqs. (54) and (57) introduced with the result that
an incorrect expression is obtained for entropy production rate that subsequently
yields an incorrect equation of magnetization relaxation.
Also,
1 t
1
t

∇v = ∇v + (∇v) + ∇v − (∇v)
2 2
= D +·ω
(v)
(64)

t
where D (v) = 12 ∇v + (∇v) is the strain rate tensor and ω = 12 ∇ × v is the
fluid rate of rotation, i.e., one-half the fluid vorticity. Thus, the first term on the
right side of Eq. (60) can be written as:

Ttv : ∇v = Ttv : D (v) + · ω (65)
The sixth term on the right hand side of Eq. (60) transforms as follows:

[(D · E + B · H)] I : ∇v − P · E (e) + M H (e) I : ∇v

= 0 E 2 + μ0 H 2 + P · E − E (e) + M · H − H (e) ∇ · v (66)
where I : ∇v = ∇ · v. Substituting the various transformed terms into Eq. (60)

yields the entropy production equation in final form:
ds
ρT = Ttv : D (v) + · (ω − Ω) + Λt : ∇Ω + j ∗ · E ∗
dt
dP
∇ · q + ρR + E − E (E) · + P (∇ · v) − Ω × P
dt
dM
+ H − H (e) · + M (∇ · v) − Ω × M (67)
dt
9 Clausius-Duhem Inequality
The second law of thermodynamics states that the entropy increase dS in a
system is larger than the entropy dS(s) supplied with the difference between the
two compensated by positive internal entropy production dS(i) . This relationship
is expressed as follows:
dS(i) = dS − dS(s) ≥ 0 (68)
The inequality applied to a continuum can be written in the following form.
⎡ ⎤

d d q ρR ⎦
ρs(i) dV = ρsdV − ⎣− n · dS + dV ≥ 0 (69)
dt dt T T
Vm Vm Sm Vm
where q/T is local entropy flux. Applying the Reynolds transport theorem to the
time derivative terms and Gauss’ theorem to the surface integral term, noting
the arbitrariness of the volume, and multiplying by T on both sides yields
ds(i) ds q
ρT = ρT + T ∇ · − ρR ≥ 0 (70)
dt dt T
74 R.E. Rosensweig
Expansion of the term containing the heat flux vector gives

q 1
T ∇ · = ∇ · q + q · ∇ · ln (71)
T T
Substituting this expression into the inequality of (70) together with the
expression for entropy production rate from Eq. (67) yields

ds(i) 1
ρT = q · ∇ ln + j∗ · E∗ + Φ ≥ 0 (72)
dt T
which is the Clausius-Duhem inequality for this system, where Φ is the dissipa-
tion function:

Φ = Ttv : D (v) + · (ω − Ω) + Λt : ∇Ω
dP
+ E − E (e) · + P (∇ · v) − Ω × P
dt
dM
+ H − H (e) · + M (∇ · v) − Ω × M ≥ 0 (73)
dt
and where it is assumed that heat flow and current flow produce dissipation
in the presence of mechanical dissipation and each other, and that no cross
couplings exist. Accordingly, the following conditions must be met.

1
q · ∇ ln ≥0 (74)
T
and
j∗ · E∗ ≥ 0 (75)
10 Determination of Constitutive Equations

Constitutive equations of q, j ∗ , Tv , Λ, M and P must be determined such that
the conditions (72) and (73) are met. Linear relationships are developed in the
following.
10.1 Constitutive Equations of q and j ∗
The conducting magnetic fluids are assumed to be isotropic, hence constitutive

equations are determined consistent with (74) and (75) that q is proportional to
E ∗ , and j ∗ proportional to E ∗ . Assuming linear variations this yields:
q = −k∇T (76)
j ∗ = σE ∗ (77)
where (76) is Fourier’s first law, and (77) is Ohm’s law. Transforming the quan-
tities in (77) to the stationary coordinate system using Eqs. (42) and (52) leads
to
j = σ (E + v × B) + ρe v (78)
This implicity recovers the Lorentz force law and gives the relationship between
electric field and electric current density in a conductor containing free charge
of density e moving with velocity v when a magnetic flux density B exists.
10.2 Constitutive Equations for Stress and Surface Couple Tensors

The following decomposition is employed in this section.
(s) 1
T = T̆ + T(a) + ItrT (79)
3
where T is any dyadic, trT is the trace of T, T(a) the antisymmetric part of T,
and T − (1/3)ItrT. The following identity applies for the indicated product of
two dyadics:
1 (s) (s)
T:V= (trT) (trV) + T(a) : V(a) + T̆ : V̆ (80)
9
The first term on the right is a product of scalars. The second term on the
right can alternatively be written as the scalar product of two axial vectors.
Applying the identity to the first term on the right side of Eq. (73) gives:
1
Ttv D(v) + · (ω − Ω) = tr Ttv trD(v)
9
(a) t (s) (v)
+ TTv : · (ω − Ω) + T̆v : D̆ (81)
Thus, non-negative values of entropy production are assured with the follow-
ing relationships.
t (s) (v)
T̆v = 2η D̆ , (82)
(a)
Ttv = −2ζ · (ω − Ω) , (83)

tr Ttv = γ trD(v) (84)
where η, ζ, and γ are positive valued proportionality factors, and where it is

recognized in (83) that the product of antisymmetric dyadics is negative. Tv
decomposed in the manner of Eq. (79) reads
(s) 1
Tv = T̆v + T(a)
v + ItrTv
3
(s) T (a) 1
= Ttv − Tv + ItrTtv (85)
3
76 R.E. Rosensweig
Substituting values from (82,83,84) now yields the following constituted form
of Tv .
(v) γ
Tv = 2η D̆ + 2ζ · (ω − Ω) + ItrD(v)
3
t
= η ∇v + (∇v) + 2ζ · (ω − Ω) + λ (∇ · v) I (86)
where λ = (γ − 2η)/3. The η is shear coefficient of viscosity, ζ is known as the

vortex viscosity, and λ is the bulk coefficient of viscosity.
The constitutive relationship for surface couple stress is obtained from the
second term on the right side of Eq. (73) following the analogous treatment.
With assumption that asymmetrical couple stress is absent the result is

Λ + η ∇Ω + (∇Ω) + λ (∇ · Ω) I
t
(87)
where η and λ are the shear and bulk coefficients of spin viscosity, respectively.
10.3 Constitutive Equation for Magnetization

Inspection of the last term on the right side of Eq. (73) shows that the term
is non-negative provided the bracketed sum of magnetic terms is proportional
to (H − H (e) ). The same conclusion was reached by Felderhof and Kroh [7].
In the regime of initial susceptibility χm = (1/μ0 )(∂M/∂H)H=0 the terms are
proportional to (M 0 − M ) where M 0 is the value of magnetization that would
be in equilibrium in magnetic field H, given for dilute fluids from the Langevin
superparamagnetic relationship [15], and M , as previously defined, is the actual
magnetization at a given instant of time. Then the constitutive equation for
magnetization may be written as:
dM 1
+ M (∇ · v) = Ω × M + (M 0 − M ) (88)
dt τm
where τm is the magnetic relaxation time constant. Eq. (88) is the Shliomis
relaxation relationship [4], generalized for compressible fluid. The corresponding
equation for relaxation of electric polarization is obtained from the second to last
term on the right side of Eq. (73). The electric polarization in usual magnetic
fluids is small such that the contribution to the dissipation function is negligible,
hence this relaxation equation will not be written here. With the electric dipolar
relaxation time τe very small and approaching zero, the difference (P 0 −P ) must
also be small to prevent the ratio from becoming singular. Then the constitutive
equation for electric dipole moment reduces to
P ∼
= P 0 = χe 0 E (89)
where χE is the electric susceptibility.

11 The Constituted Equation Set

Substituting the constitutive equations of q, j, TTv and Λ , i.e., Eqs. (76), (78),
(86) and (87), and Tm into Eqs. (14), (24) and (72) yields definite expressions
for the equations of motion, internal angular momentum and entropy production
rate. The system of basic equations in differential form becomes:
∂ρ
+ ∇ · (ρv) = 0 (90)
∂t

∂ρ
ρ + v · ∇v = −∇p + (η + λ) ∇ (∇ · v) + η∇2 v
∂t
+∇ × [2ζ (ω − Ω)] + ρe E + j × B + (P · ∇) E
+ (M · ∇) H + P × (∇ × E) + M × (∇ × H) + ρb(91)
dΩ
I = (λ + η ) ∇ (∇ · Ω) + η ∇2 Ω − 4ζ (ω − Ω)
dt
+M × H + P × E + ρI (92)
ds η t
2
2 2
ρT = ∇v + (∇v) + 4ζ (ω − Ω) + λ (∇v)
dt 2

+η ∇Ω + ∇Ω t : ∇Ω + λ (trΩ)
2
j ∗2 1 2
+k∇2 T + + ρR + (M 0 − M ) (93)
σ χm τm
dM 1
+ M (∇ · v) = Ω × M + (M 0 − M ) (94)
dt τm
P = χe 0 E (95)
∂B
= −∇ × E (96)
∂t
∂D
+j =∇×H (97)
∂t
∇·B =0 (98)
∇ · D = ρe (99)
j = σ (E + v × B) + ρe v (100)
∗
j = j − ρe v (101)
D = 0 E + P B = μ0 H + M (102)
Note in Eq.(91) that terms may be combined as follows.
P · ∇E + P × (∇ × E) = (∇E) · P (103)
78 R.E. Rosensweig
M · ∇H + M × (∇ × H) = (∇H) · M (104)
The term M × (∇ × H) disappears for non-conductive fluid unless the E
field is time-varying and the corresponding electrical term disappears unless B
is time-varying, with both forces small, even in high frequency fields.
The last term in Eq. (93) is useful in computing radio frequency heating of
ferrofluids [16].
For negligible angular acceleration, spin diffusion, electric polarization, and
distant source of body couple, Eq. (92) simplifies to
1
Ω= M ×H +ω (105)
4ζ
Substituting for Ω in Eq. (94) yields the relaxation equation in the commonly
useful form for incompressible fluid as
dM 1 1
=ω×M − (M − M 0 ) − M × (M × H) (106)
dt τm 4ζ
where, for dilute colloids, ζ = 3ηφ/2 For incompressible, isothermal, non-conduct-

ive magnetic fluid absent of free charge and internal heat release, having negligi-
ble electric polarization, using Eq. (105) reduces Eq. (91) to the following simpler
form of the linear momentum balance having wide applicability to systems in
which relaxation rate of the magnetic moment plays an important role.

∂v 1
ρ + v · ∇v = −∇p + +η∇2 v + ρb + M · ∇H + ∇ × (M × H) (107)
∂t 2
This equation further reduces to the following quasi-equilibrium equation

when relaxation rate is rapid such that M and H are sensibly collinear [1].

∂v
ρ + v · ∇v = −∇p + +η∇2 v + ρb + M ∇H (108)
∂t
M and H are the magnitudes of M and H, respectively.
12 Conclusion
This work derived a general set of basic equations for flow and response of con-
ductive magnetic fluids with internal rotation based on integral balance equa-
tions and thermodynamics. The work is free of arbitrary definitions of electro-
magnetic energy density and stress. Employing a strict formulation of entropy
production in which applied electromagnetic field is distinguished from its equi-
librium counterpart, the usual constitutive relationships emerge from the analy-
sis without empirical assumption, and include the magnetization relaxation rela-
tionship. The work gives support to the Minkowski form of the electromagnetic
momentum density D × B.
This study has treated the Maxwell type stress of a magnetic fluid on the same
basis as ordinary contact stresses of pressure and viscous stress and produces
the accepted form of the linear and angular momentum balances. In addition,
the work shows that the product of Maxwell type stress and velocity is useful in
formulating the integral energy balance.
As a recommendation for further work it is desirable that a broader treatment
be developed such that the stress tensor in Eq. (62) is expressed in rest frame
variables. Such a theory would be exact to first order in velocity throughout. For
additional remarks, see Appendix C.
Acknowledgments
The author is appreciative of communications with H. Brenner, U. Felderhof, C.

Rinaldi-Ramos, M. Shliomis, T. Tanahashi, and M. Zahn.
References
1. J. L. Neuringer and R. E. Rosensweig, Phys. Fluids 7 (12), 1927 (1964).
2. J. S. Dahler and L. E. Scriven, Nature 192, 36 (1961); D. W. Condiff and J. S.
Dahler, Phys. Fluids 7 (6), 842 (1964).
3. M. I. Shliomis, Soviet Phys. JETP (Engl. transl.) 34 (6), 1291 (1972).
4. M. I. Shliomis, Soviet Phys. Uspekhi (Engl. transl.) 17 (2), 153 (1974).
5. M. I. Shliomis, Phys. Rev. E, Comment on “Magnetoviscosity and relaxation in
ferrofluids” Phys.Rev.E.64 (2001).
6. S. R. deGroot and P. Mazur,Non-equilibrium thermodynamics, Dover, Mine-
ola,NY(1984).
7. B. U. Felderhof and H. J. Kroh, Jl. Chem. Phys. 110 (15), 7403 (1999).
8. K. Shizawa and T. Tanahashi, Bulletin of JSME, 29 (250), 1171 (1986); 29 (255),
2878 (1986).
9. C. Rinaldi and H. Brenner, “Body vs surface forces in continuum mechanics: Is the
Maxwell stress tensor physically objective?”, Phys. Rev E 65 (3); 036615 (2002).
10. J. A. Stratton, Electromagnetic theory, (McGraw-Hill, New York, 1941).
11. R. E. Rosensweig, Chapter 13 in G. Astarita, Thermodynamics:An advanced text-
book for Chemical Engineers, Plenum Press, New York (1989).
12. B. T. Chu, Phys. Fluids 2 (5), 473 (1959).
13. I. E. Tarapov, Magnit. Gidrodin. 1, 3 (1973).
14. E. Blums, Yu. A. Mikhailov, and R. Ozols, Heat and mass transfer in MHD flows,
World Scientific, Singapore (1987).
15. R. E. Rosensweig, Ferrohydrodynamics, Cambridge University Press, New York
(1985); reprinted with slight corrections by Dover, Mineola, New York (1997).
16. R. E. Rosensweig, Magnetohydrodynamics, 36 (4), 300 (2000).
80 R.E. Rosensweig
Symbols
Latin
b Body force per unit mass (N kg−1 )
B Magnetic induction, tesla (kg s−2 A−1 )
D Displacement vector (m−2 s A)
(v)
D Strain rate tensor (s−1 )
E Electric field vector in stationary coordinates, (m kg s A−1 ) −3
volts/meter
E ∗ Electric field vector in the moving medium, (m kg s−3 A−1 )
volt/meter
g Electromagnetic momentum vector (N s m−3 )
H Magnetic field vector in stationary coordinates (m−1 A)
∗
H Magnetic field vector in the moving medium (m−1 A)
I Unit tensor (dimensionless)
I Moment of inertia per unit volume (kg m−1 )
j Current density in stationary coordinates (A m−2 )
∗
j Current density in the moving medium (A m−2 )
l Body couple vector per unit mass transmitted from (N m kg−1 )
afar
M Magnetization, tesla (kg s−2 A−1 )
M 0 Equilibrium magnetization in field H (kg s−2 A−1 )
n Unit outward facing normal (dimensionless)
P Electric polarization vector (m−2 s A)
P 0 Equilibrium electric polarization vector in field E (m−2 s A)
p Pressure-like variable defined in Eq. (30) (N m−2 )
p0 Pressure in absence of electromagnetic fields (N m−2 )
p (e)
Pressure in presence of electromagnetic fields (N m−2 )
q Heat flux vector (J m−2 s−1 )
r Position vector (m)
R Internal heat release rate per unit mass (J kg−1 s−1 )
s Entropy per unit mass (J K−1 kg−1 )
s Spin vector, i.e., angular momentum per unit mass (m2 s−1 )
S Total entropy of a system (J K−1 )
S m Surface area enclosing volume Vm (m2 )
t Time (s)
tn Stress vector (N m−2 )
T Temperature (K)
T Stress tensor less electromagnetic momentum flux (N m−2 )
T̃ Stress tensor sum of pressure, viscous and electro- (N m−2 )
magnetic contributions
u Internal energy per unit mass including field energy (J kg−1 )
um Internal energy per unit mass excluding field energy (J kg−1 )
u (e)
Value of u in equilibrated system (J kg−1 )
(J kg−1 )
(e)
um Value of um in equilibrated system
Vm Volume containing mass (m3 )
Greek
γ Lorentz transformation coefficient (dimensionless)
0 Permittivity of free space, 8.854 x 10−12 farad/meter (m3 kg−1 s4 A2 )
Polyadic alternator (dimensionless)
ζ Vortex vorticity (N s m−2 )
η Shear coefficient of viscosity (N s m−2 )
η Shear coefficient of spin viscosity (N s)
λ Bulk coefficient of viscosity (N s m−2 )
λ Bulk coefficient of spin viscosity (N s)
λn Couple stress vector (N m−1 )
Λ Couple stress tensor (N m−1 )
−7
μ0 Permeability of free space, 4π x 10 henry/meter (m kg s−2 A−2 )
Mass density (kg m−3 )
e Charge density, coulomb/meter 3
(A s m−3 )
σ Electrical conductivity, siemen (m kg−1 s3 A2 )
−3
τm Relaxation time constant (s)

υ Specific volume (m3 kg−1 )
Φ Dissipation function (N s m−2 )
χe Electrical susceptibility (dimensionless)
χm Magnetic susceptibility (dimensionless)
ω Fluid rotational rate vector (radian s−1 )
Ω Particle rotational rate vector (radian s−1 )
Appendix A: Notation
Vectors are set in bold Roman such as v or B, or lower case bold Greek such as
λ.
Dyadics are indicated in bold sans serif such as T or bold upper case Greek
such as Λ.
T corresponds to êi êj Tij in indicial notation employing the Einstein summa-
tion convention; êi and êj are a unit vectors. Tij is defined as the j th component
of force acting on the surface having normal orientated in the i th direction.
TT is the transpose of T and corresponds to êi êj Tji .
The operator ∇ corresponds to êi ∂/∂xi .
The nesting convention is followed as shown by these examples:
ab · cd = (b · c) ad
ab : cd = (b · c) (a · d)
∇·T corresponds by the nesting convention to êi ∂Tij /∂xi and not êi ∂Tij /∂xj .
I is the unit dyadic corresponding to êi êj δij where δij is the Kronecker delta
function.

0 (i = j)
δij =
1 (i = j)
82 R.E. Rosensweig
is the alternating polyadic corresponding to êi êj êk ijk where
⎧
⎨ 1 (ijk = 123, 231, or312)
ijk = 0 (i = j, i = k, orj = k)
⎩
−1 (ijk = 132, 213, or321)
From these definitions, : ab = −a×b and ∇·(T · v) = TT : ∇v +v ·(∇ · T).
Appendix B: Galilean Transformation of E, H and j

The transformation of these field vectors is carried out in the spirit of the text
with integral statement of the basic law taken as the point of departure.
Faraday’s Law and the E Field
Faraday’s law applicable to moving media relates induced voltage around an

arbitrarily chosen closed circuit to the time rate of change of induction enclosed
by the circuit. Choosing the circuit to coincide with a moving material line of
fixed identity, the relationship may be written as:

dΦf
E ∗ · dl = (109)
dt
where E ∗ is the electric field in the frame (rest frame) of an observer moving
with the local velocity, dl is the differential of path length, and Φ is the number
of flux linkages defined by
Φf = B · ndS (110)
where n is a unit vector normal to differential area of S, the enclosed surface.

The Reynolds’ transport theorem5 for differentiating a surface integral gives

d ∂B
B · ndS = + (∇ · B) v + ∇ × (B × v) · ndS (111)
dt ∂t
Applying Stokes’s theorem to the left side of (109) converts the line integral
to a surface integral.

E ∗ · dl = − (∇ × E ∗ ) · ndS (112)
5
R. Aris, Vectors, tensors and the basic equations of fluid mechanics. Dover, Mineola,
NY (1989). An alternate derivation is given in W. K. H. Panofsky and M Phillips,
Classical electricity and Magnetism, Second Edition, Addison-Wesley, Reading, Mas-
sachusetts (1962, pages 160-300).
Combining the expressions, recognizing that the surface of integration is ar-

bitrary, and noting from the Maxwell law that ∇ · B = 0, yields the following
differential relationship upon rearrangement.
∂B
∇ × (E ∗ − v × B) = − (113)
∂t
As a tenet of relativity, including Galilean relativity, the form of a basic
relationship of physics is invariant to the choice of inertial (constant velocity)
reference frame. Accordingly, comparing (113) with the form of the Faraday law
expressed in terms of laboratory coordinates, namely
∂B
∇×E =− (114)
∂t
shows that the relationship between the electric field vectors is given by
E∗ = E + v × B (115)
This relationship is given in Eq.(38) of the text. The observer in the moving
frame senses an additional electric field due to motion through the induction
field.
Ampere’s Law and the H and j Fields
These relationships can be derived from a treatment of Maxwell’s extension of

Ampere’s law. In its general integral form applicable to a medium in motion,

d
H ∗ · dl = j ∗ · ndS + D · ndS (116)
dt
Again the asterisk indicates the rest frame. From the Reynolds’ transport
theorem, as introduced above,

d ∂D
D · dS = + (∇ · D) v + ∇ × (D × v) · ndS (117)
dt ∂t
Applying Stokes’s theorem to the left side of (116) converts the line integral
to a surface integral.

H ∗ · dl = − (∇ × H ∗ ) · ndS (118)
Combining the expressions, recognizing that the surface of integration is ar-

bitrary, and noting from the Maxwell law that ∇ · D = ρe , yields the following
differential relationship upon rearrangement.
∂D
∇ × (H ∗ + v × D) = (j ∗ + vρe ) + (119)
∂t
84 R.E. Rosensweig
This may be compared with the form of Ampere’s extended equation written in
laboratory coordinates which is
∂D
∇×H =j+ (120)
∂t
Comparison of the two foregoing expressions yields the transformation rela-
tionships for magnetic field and electrical current density. These are:
H∗ = H − v × D (121)
j ∗ = j − ρe v (122)
Motion relative to charges alters the perception of electric current density as
shown by (122). Motion relative to the displacement field from (121) produces
a component of magnetic field in the observer’s frame.
Appendix C:
Comment on Electromagnetic Body Couple Density
From the exact expression for the rest frame stress tensor T∗m known from anal-
ysis of motionless fluid [15] and having the form given by Eq.(62), the associated
body couple density is given by vecT∗m = M ∗ × H ∗ + P ∗ × E ∗ . Expansion of
the semi-relativistic expressions [10] relating field vectors between frames gives
to first order in velocity,
M ∗ = M + μ0 v × P (123)
H∗ = H − v × D (124)
∗
P = P − 0 v × M (125)
∗
E = E+v×B (126)
Thus,
M ∗ × H ∗ + P ∗ × E ∗ = M × H + P × E − B × (v × D) + D × (v × B)
H E
+ 2 × (v × E) − 2 × (v × H)
c c
∼
= M ×H +P ×E+v×g (127)
where the second equality results as a vector identity following neglect of the
terms reciprocal to c2 , and g = D × B is the electromagnetic momentum den-
sity. In comparison to (127), the body couple corresponding to Eq. (62) in the
text is absent the v × g term. This limits the text result to quasi-static flows of
low velocity. Newton’s equations are galilei-invariant but electromagnetism au-
tomatically brings in Lorentz-invariance, hence it should be expected that only
a completely relativistic formulation could be totally satisfactory.
Ferrohydrodynamics: Retrospective and Issues
Mark I. Shliomis
Department of Mechanical Engineering, Ben-Gurion University of the Negev,

P.O.B. 653, Beer-Sheva 84105, Israel
Abstract. Two basic sets of hydrodynamic equations for magnetic colloids (so-called
ferrofluids) are reviewed. Starting from the quasistationary ferrohydrodynamics, we
then give a particular attention to an expanded model founded on the concept of
internal rotation. A specific relation between magnetic and rotational degrees of free-
dom of suspended grains provides a coupling of the fluid magnetization with the fluid
dynamics. Hence a complete set of constitutive equations consists of the equation of
ferrofluid motion, the Maxwell equations, and the magnetization equation. There are
three kinds of the latter. Two of them were derived phenomenologically as a gener-
alization of the Debye relaxation equation in case of spinning magnetic grains, while
one of them was derived microscopically from the Fokker-Planck equation. Testing the
magnetization equations, we compare their predictions about the dependence of the
rotational viscosity on the magnetic field and the shear rate.
1 Quasistationary Ferrofluid Dynamics

Physics and hydrodynamics of ferrofluids has begun from the basic work [1]
and the following long series of papers by Rosensweig and co-workers, included
later on in his monograph [2]. Those papers and the book laid a serious scientific
foundation for further research and gave an impetus to great variety of ferrofluid
applications in industry, technology, and medicine. A set of equations describ-
ing ferrofluid dynamics (“ferrohydrodynamics” – the term of Rosensweig) first
proposed in [1] consists of the equation of ferrofluid motion
dv d ∂
ρ = −∇p + η∇2 v + M ∇H , = + (v · ∇) , (1)
dt dt ∂t
the magnetic state equation M = M (H, T ), for which it is natural to employ
[3,4] the Langevin formula
M = nmL(ξ) , ξ = mH/kB T , L(ξ) = coth ξ − ξ −1 , (2)
(here m = Md V being the magnetic moment of a single subdomain magnetic

particle, V = πd3 /6 the particle volume, n their number density, and Md stands
for the domain magnetization of dispersed ferromagnetic material) and the equa-
tions
div v = 0, rot H = 0, div B = 0, (B = H + 4πM ) (3)

86 M.I. Shliomis
indicating that the ferrofluid is considered to be incompressible and nonconduct-

ing. The volume density of magnetic forces in (1), F = M ∇H, is calculated as
mag
divergence of the stress tensor of magnetic field in ferrofluid, Fi = ∂σik /∂xk ,
where
mag H ∂B
σik = 4π1
Hi Bk − · B−ρ δik . (4 a)
8π ∂ρ H ,T
According to the assumption (2), ferrofluid magnetization is nonsensitive to the
velocity field and relaxes instantaneously to the equilibrium (Langevin) value.
Then in the low-field limit, when B = μH with a constant μ = 1+4πχ, equation
(4 a) takes the well-known form [5]

mag μHi Hk H2 ∂μ
σik = − μ−ρ δik . (4 b)
4π 8π ∂ρ T
For a non-magnetic fluid (μ = 1) this expression is reduced to the Maxwell stress
tensor of magnetic field
mag
σik = 1
4π (Hi Hk − 12 H 2 δik ).
Equation (4 a) may be rewritten as

mag H ∂M
σik = 1
4π (Hi Bk − 1 2
2 H δik ) − · M −ρ δik . (4 c)
2 ∂ρ H ,T
Its divergence consists of two terms,

H ∂M
F = (M · ∇)H − ∇ · M −ρ , (5)
2 ∂ρ H ,T
where the first is called sometimes the Kelvin force, and the second represents
the magnetostrictive force. Being the potential force, the latter is always in-
cluded into the pressure gradient in (1), i.e., it is equilibrated automatically by
the hydrostatic or hydrodynamic pressure. Further, since the equilibrium mag-
netization is collinear with the local magnetic field, M = (M/H)H, the Kelvin
force takes the form (M · ∇)H = (M/H)(H · ∇)H. Hence, using an identity
(H · ∇)H = 12 ∇H 2 − H × rot H
and taking into account the condition (3) of the absence of free electrical currents,
rot H = 0, we arrive at the pointed above expression F = M ∇H.
With allowance for (2) and the equality
d sinh ξ
L(ξ) = ln ,
dξ ξ
the Kelvin force can be written in the case of isothermal fluids, T = const, as
sinh ξ
F = nkB T ∇ ln . (6)
ξ
Ferrohydrodynamics: Retrospective and Issues 87
If magnetic grains are distributed uniformly all over the ferrofluid sample, n =
const, the last expression takes the form

sinh ξ
F = ∇ nkB T ln , (7)
ξ
i.e., the magnetic force proves to be potential. It makes an evidence of Bernoulli
theorem for non-vortical ferrofluid flow [2]
sinh ξ
p + ρgz + 12 ρv 2 − nkB T ln = const (8)
ξ
(g is the gravity acceleration and z is the vertical coordinate). This relationship
is to high extent useful in qualitative investigations of equilibriums and flows of
ferrofluids under nonuniform magnetic fields. Note, however, that a nonunifor-
mity of the field induces an inhomogeneity of the particle distribution which in
its turn leads to an alteration in magnetic force. Indeed, as seen from (6), each
a particle is acted upon by the force
sinh ξ
F /n = −∇U, U = −kB T ln , (9)
ξ
where U is the potential. The equilibrium (Boltzmann’s or “barometric”) particle
distribution is given by the formula n = n0 exp(−U/kB T ). Thus we find, using
(9),
sinh ξ sinh ξ(r) 3
n = n0 , n0 d r = n̄V. (10)
ξ V ξ(r)
The constant n0 is determined from the indicated normalization condition ex-
pressing conservation of the total particle number in the ferrofluid volume V; n̄
is the mean density of the particle number. Substituting n(ξ) from (10) into (6)
or (9) gives the resulting self-consistent magnetic force density
sinh ξ sinh ξ sinh ξ
F = n0 kB T ∇ ln ≡ n0 kB T ∇ = ∇(nkB T ) . (11)
ξ ξ ξ
When the fluid is at rest, this force is equilibrated by the pressure gradient,
F = ∇p [see (1)], which emerges automatically. Thus, one can say of magnetic
pressure pm = nkB T created by the thermal agitation of the magnetic grains in
an inert liquid. As it should be, pm is similar to the pressure of an ideal gas of
molecules.
One should not, however, be in hurry with the replacement
sinh ξ sinh ξ
n̄kB T ln =⇒ n0 kB T (12)
ξ ξ
in the Bernoulli theorem (8), because an equilibrium particle concentration n(ξ)
sets in for a very long time. The equilibrium is settled owing to diffusion and mag-
netophoresis which proceed extremely slow due to smallness of the particle diffu-
sion coefficient D. Actually, according to the Einstein formula, D = kB T /3πηd, it
88 M.I. Shliomis
is inversely proportional to the particle diameter (d 10 nm) and hence appears

to be two orders as less than for molecules. Therefore, concentration inhomo-
geneities never arise in a flowing ferrofluid since they simply have no time to be
formed. As a consequence, there exist two scenarios of the onset of convection in
magnetized ferrofluids [6]. Namely, if an applied temperature difference increases
faster than the limit imposed by particle diffusion, the ferrofluid behaves as a
pure (i.e., single-component) fluid. In this case, only stationary instability occur
upper the threshold of convection [7]. In the opposite case, when the concentra-
tion gradient (induced by the temperature gradient due to magnetophoresis and
the Soret effect) is built up undisturbed by convection, the theory [6,8] predicts
oscillatory instability in a certain region of parameters. Thus, the replacement
(12) should be carried out, strictly speaking, only for a genuine equilibrium, i.e.,
when we deal with the ferrohydrostatics.
When ferrofluid moves in a magnetic field, the pseudovector M can depends,
in principle, on the two available pseudovectors: the field H and the fluid vor-
ticity Ω = 12 rot v. Meanwhile the discussed hydrodynamic model [1] assumes
that the magnetization is independent on a ferrofluid flow, whereas the flow
does depend on the magnetization – see (1). This discrimination has yet another
side: instead of the magnetization dynamics equation – the one for dM /dt –
the theory proposes an equilibrium expression (2). According to the latter, M is
collinear with the field H at any moment and determined by its instantaneous
value. This assumption means that the magnetization relaxation time is consid-
ered to be zero. Hence it is natural to regard the ferrohydrodynamics [1] as a
quasistationary theory. As it has been shown by the author [4], this theory has
a definite field of applicability.
The point is that, when magnetic field is shifted in direction, the particle
magnetic moment m – and the ferrofluid magnetization M = nm as well –
can regain equilibrium by two processes [3,4]. In Brownian relaxation the vector
m rotates in unison with the particle itself. The rotation is resisted by viscous
torque due to surrounding carrier liquid and characterized by Brownian time
of rotational diffusion τB = 3ηV /kB T . In the Néel mechanism of relaxation, m
rotates inside the particle (with the reference time τN ), whereas the particle
itself does not rotate. If the field H is turned off at the instant t = 0, the
magnetization M (0) would decay to zero with the reduced relaxation time [4,9]
τN τ B
τ= . (13)
τN + τ B
In a real ferrofluid, any relation between τN and τB is possible. Since, however,
τN increases exponentially with the increase of the particle volume V while τB
is simply proportional to V , the equality τN = τB is fulfilled at a certain particle
diameter dS (the so-called Shliomis’ diameter). For d < dS it holds the condition
τN τB , i.e., according to (13), τ τN : ferrofluid magnetization relaxes owing
to internal diffusion of the particle magnetic moments. If, conversely, d > dS ,
then τN τB so the Néel process is frozen and magnetization relaxation proceeds
via Brownian rotary diffusion of the particles in the liquid matrix, τ τB . The
larger the particle size the better it holds the condition of “freezing up” of the
particle magnetic moment into the particle body.
The quasistationary ferrohydrodynamics is valid just for colloids of Néel par-
ticles. Indeed, the coupling between magnetic and mechanical degrees of freedom
of such particles breaks down (hence M does not depends on Ω), and the magne-
tization relaxes to the direction of H almost immediately – for d (0.3 − 0.5)dS
there is τ τN ∼ 10−9 s (!) – as the model assumes.
The opposite case of suspensions of rigid magnetic dipoles (i.e., the grains
with “freezing in” magnetic moments, τN τB ) is considered below. In this
model with the infinitely strong coupling and the finite relaxation time τ τB ∼
10−4 − 10−5 s, the vector M is not obliged to be collinear with H. Therefore,
besides the magnetic force density (M · ∇)H, ferrofluid can undergo a magnetic
torque of the density M × H. The model of rigid dipoles is widely used in
the theory of magnetic fluids. Despite its simplicity, it allows one to explain a
wide complicated tangle of magnetic and hydrodynamic phenomena emerging
in ferrofluids under the field. Below we derive and discussed the basic equations
for the model.
2 Ferrohydrodynamics: Allowance for Internal Rotation
The main peculiarity of ferrofluids is a specific relation between the magnetic

and rotational degrees of freedom of suspended magnetic grains of which the
fluids are composed. Therefore the concept of internal rotation first applied to
ferrofluids in [3] has proved to be very fruitful. The model [3] takes into account
that the volume density of the angular momentum of ferrofluids consists of both
the visible (“orbital”) and the internal (“spin”) parts. The former, L = ρ (r ×v),
is associated with the translational motion of magnetic grains and molecules of
the solvent. The latter, S, is caused by the rotation of the grains themselves and
should be treated as an independent variable along with the fluid velocity v,
density ρ, and pressure p. However, an appropriate thermodynamic coordinate
is the difference S − IΩ where Ω is the local angular velocity of the fluid and I
means the volume density of the particles moment of inertia. For a suspension of
spherical grains I = ρs φd2 /10 where φ is the volume fraction of dispersed solid
phase, ρs is the material density of the solids, and d is the mean particle diameter.
It is convenient to set S = Iω p where ω p is the macroscopic (i.e., averaged over
physically small volume) angular velocity of the particles. It is worth noting
that rotation of the particles induces micro-flow-fields in the surrounding carrier
liquid that in turn contain an additional angular momentum. This contribution
may be included into S by means of a re-normalization of the moment of inertia
I. In fact, an effective (“hydrodynamic”) diameter dh of a magnetic grain has to
be a little larger than its true diameter d due to such an attached micro-vortex.
Any deviation of ω p from Ω gives rise to dissipation processes due to redistri-
bution of angular momentum between L and S forms. (The angular momentum
conservation law refers, sure, to the total angular momentum L+S ). These pro-
cesses contribute the stress tensor σik . For an ordinary (nonmagnetic) suspension
90 M.I. Shliomis
the tensor has been derived by the methods of irreversible thermodynamics in

[10]:
∂vi ∂vk 1
σik = −pδik + η + + (Sik − IΩik ), (14)
∂xk ∂xi 2τs
where
Sik = ikl Sl , Ωik = 12 (∂vk /∂xi − ∂vi /∂xk ) = ikl Ωl ,
and ikl stands for antisymmetric unit tensor. Apart from the viscosity η , (14)
contains once more kinetic coefficient: the spin relaxation time τs = I/6ηφ =
ρs d2 /60η . For d = 10 nm and η = 10−2 P this formula gives τs ∼ 10−11 s (!).
Thus, the difference ω p − Ω instantly decays whereupon the hydrodynamic de-
scription is reduced to the common set of hydrodynamic equations. Ferrofluids,
however, give us an opportunity to maintain this difference by an extraneous
magnetic torque which acts directly upon the particles rotation:
6ηφ(ω p − Ω) = M × H . (15)
Eliminating the last term in (14) with the aid of the torque balance equation
mag
(15) and including in σik the stress tensor of magnetic field σik from (4c), one
gets [3,4]

∂vi ∂vk
σik = −pδik +η + + 12 (Mi Hk −Mk Hi )+ 4π
1
(Hi Bk − 12 H 2 δik ); (16 a)
∂xk ∂xi
mag
the magnetostrictive diagonal part of σik has been included here in the isotropic
tensor of the pressure. Substituting σik from (16 a) into the momentum conser-
vation law ρdvi /dt = ∂σik /∂xk yields the equation of ferrofluid motion
dv
ρ = −∇p + η∇2 v + (M · ∇)H + 12 rot (M ×H) . (17)
dt
At the calculation of the divergency of the stress tensor (16 a) we have used
equations (3). Note that under the assumption of collinearity of M with H the
last term in (17) vanishes and the Kelvin force density (M · ∇)H takes the form
M ∇H, so that (17) is reduced to (1).
The angular momentum conservation law reads dSik /dt = σki − σik . This
equation governs the particle rotation. Meanwhile an extreme smallness of the
corresponding relaxation time τs has allowed us to neglect the inertia term dS/dt
in comparison with the relaxation term (S − IΩ)/τs . As the result, the equa-
tion of particle rotation is reduced to the torque-balance equation (15) which
expresses symmetry of the stress tensor (16): σki = σik . Indeed, on substitution
Bk = Hk + 4πMk in (16 a) we are convinced of the symmetry:

∂vi ∂vk
σik = −(p + 8π1
H 2 )δik + η + + 12 (Mi Hk + Mk Hi ) + 4π
1
Hi Hk . (16 b)
∂xk ∂xi
The system of equations (17) and (3) is still not complete since it does not
determine the ferrofluid magnetization. The latter influences the fluid motion
and depends itself on the motion as well. There are two basic ways to derive
the missing magnetization equation. Both the ways have been proposed by the
author with co-workers [3,11,12] and discussed in [2,4,9,13,14].
3 Phenomenological Magnetization Equation I

Originally the magnetization equation has been derived phenomenologically [3]
as a modification of the Debye relaxation equation [15]. To get a generalized
equation, one should introduce a local reference frame Σ , in which the suspended
magnetic grains are quiescent on the average, i.e., ω p = 0 . It is natural to assume
that the magnetization relaxation is described in the system by the simplest
Debye-like equation
d M 1
= − (M − M 0 ) , (18)
dt τB
where the equilibrium magnetization M 0 is described by the Langevin formula
(2)
M 0 = nmL(ξ)ξ/ξ , ξ = mH/kB T , (19)
and τB is the above determined Brownian time of rotational particle diffusion
since reorientation of rigid magnetic dipoles is possible only when turning the
particles themselves. Equation (18) assumes that any deviation of M – either
in direction or magnitude – from its equilibrium value M 0 decays according to
the simple exponential law (M − M 0 ) ∼ exp(−t/τB ) . The frame of reference
Σ rotates with respect to the fixed (“laboratory”) system Σ with the angular
velocity ω p . The rates of change of any vector A in systems Σ and Σ are related
by the kinematic expression
dA d A
= ω p ×A + . (20)
dt dt
Substituting here A = M , ω p from (2), and d M /dt from (7), we obtain the
equation sought [3]:
dM 1 1
= Ω ×M − (M − M 0 ) − M ×(M × H). (21)
dt τB 6ηφ
If we assume – following [1] – instant relaxation of the magnetization, τB → 0,

then (21) is reduced to M = M 0 in complete agreement with the quasistationary
theory, see (2). Multiplying (21) by M we find
dM 2 /dt = (2/τB )(M 2 − M · M 0 ) ,
whence it follows that the last (relaxation) term in (21) describes a process of
approach of the vector M to its equilibrium orientation without change of its
length. As the result, the relaxation rates of the longitudinal and transverse (to
the field) components of magnetization appears to be different. Let us calculate
them. In the linear approximation in M − M 0 one can split the nonequilibrium
part of magnetization into components parallel and perpendicular to the field:
H [H · (M − M 0 )] H ×(M ×H)
M − M0 = + . (22)
H2 H2
92 M.I. Shliomis
Substituting this expression into (21) gives the linear magnetization equation
dM H [H · (M − M 0 )] H ×(M ×H)
= Ω ×M − − , (23)
dt H 2 τ H 2 τ⊥
where τ = τB and

1 1 nmL(ξ)H 1 1
= + = 1 + ξL(ξ) ,
τ⊥ τB 6ηφ τB 2
so that
2τB
τ⊥ = . (24)
2 + ξL(ξ)
Let us revert to the Debye–like equation (18) and demonstrate that this postu-
lated by myself [3] relaxation equation can be easily derived [12] from irreversible
thermodynamics (IT). With this purpose it is convenient to take advantage of
the method of IT proposed by Landau and first applied by him just to the de-
scription of relaxation of the order parameter in a nonequilibrium system [16].
An equilibrium value of the parameter (M 0 in our case) corresponds to the
minimum of an appropriate thermodynamic potential Φ (usually the Gibbs or
Helmholtz free energy) depending on the magnetization M and other thermo-
dynamic variables. Thus, at the equilibrium ∂Φ/∂M = 0. Out of equilibrium
this condition is not satisfied, so the relaxation process occurs: M changes in
time approaching M 0 . For small deviations from equilibrium, the derivative
∂Φ/∂M and the relaxation rate dM /dt are small. The relation between the two
derivatives in the Landau theory is reduced to simple proportionality:
dM ∂Φ
= −γ (25)
dt ∂M
with a constant coefficient γ > 0. Hence we have
2
dΦ ∂Φ dM ∂Φ
= · = −γ <0 (26)
dt ∂M dt ∂M
as it should be: when a system moves to equilibrium, its free energy decreases.
In the case of a weakly nonequilibrium state of the system, one can substitute
in (25) and (26) the expansion
2
∂Φ ∂Φ ∂ Φ
= + (M − M 0 ) + ... ,
∂M ∂M 0 ∂M 2 0
where subscript 0 marks the point of equilibrium. As the first derivative in this
point is equal to zero and the second one is positive, equation (25) turns into
(18) with τB−1 = γ(∂ 2 Φ/∂M 2 )0 and (26) takes the form
dΦ (M − M 0 )2
=− . (27)
dt γ τB2
Thus, (18) and hence (21) are well corroborated by the method of IT. Notice that
some different thermodynamic method based on a strict formulation of entropy
production leads [17] to the same phenomenological magnetization equation: Eq.
(21) coincides with Eq. (106) from [17]. Equations (3), (17) and (21) constitute
the complete set of conventional ferrohydrodynamic equations.
4 Magnetization Equation Derived Microscopically

Aforecited phenomenological method allows to obtain linear relaxation term in
kinetic equation (18) and corresponding quadratic term for the rate of free energy
diminution (27). It is clear that such terms are valid only for small departures
from equilibrium. Indeed, equation (21) allows to describe well the rotational
viscosity (see Sect. 6) for arbitrary intensity ξ of a stationary magnetic field but
only sufficiently small values of ΩτB . The applicability of (21) in this case was
corroborated by real [18-20] and numerical [21] experiments as well as by numer-
ical [22] and analytical [23] solutions of the Fokker–Planck equation. [It is well
worth noting that the restriction ΩτB 1 is not so much to the point. Owing
to small values of τB for ferrofluids based on low-viscous carrier liquids (with
η0 ∼ 10−2 P) the above inequality is satisfied at all reasonable fluid vorticities.
As far as we know, violations of this condition took place only in experiments
[24-26] with glycerine-based ferrofluids (η0 ∼ 10 P)]. Meanwhile, under consid-
eration of strongly nonequilibrium situations characterized by large values of
dimensionless fluid vorticity ΩτB and especially frequency ωτB of alternating
magnetic field H ∝ cos ωt, one should employ a more precise magnetization
equation. Such a macroscopic equation has been derived [11] from the kinetic
Fokker–Planck equation which provides the microscopic description of particle
diffusion in colloids.
The Fokker–Planck equation for a ferrofluid moving in magnetic field H has
the form [9,11]
∂W
2τB = R̂ · (R̂ − 2τB Ω − e × ξ )W , (28)
∂t
where e = m/m is the unit vector along the particle magnetic moment, ξ =
mH/kB T , and R̂ = e × ∂/∂e is the infinitesimal rotation operator. Equation
(28) determines the orientational distribution function W (e, t) of particles mag-
netic moments. The macroscopic magnetization is determined by the relation
M (t) = nme where angular brackets denote statistical averaging with the
distribution function. Multiplying (28) by e and integrate over the angles, we
arrive at the equation
de
τB = τB Ω × e − e − 12 e×(e× ξ) , (29)
dt
which however is not closed. Indeed, along with the first moment of the distribu-
tion function, e, equation (29) contains the second moment (the last term in
the equation). It is easy to make sure that the equation for the second moment
includes the third one, and so on, thus there is the infinite chain of cross-linked
94 M.I. Shliomis
equations. Ideally, however, one would like to have only one equation since only
the first moment – magnetization – has a clear physical meaning. An original
scheme of closure of the first-moment equation (29), titled the effective field
method (EFM), has been proposed in [11]. Let us explain the fruitful physical
idea.
In equilibrium (Ω = 0) under a constant magnetic field the stationary solu-
tion of (28) is the Gibbs distribution
ξ
W0 (e) = exp(ξ · e) . (30)
4π sinh ξ
An averaging of the vector e with function (30) gives expression (19) for the
equilibrium magnetization. Note that only in true equilibrium the magnetiza-
tion is one or another function of the field. In a nonequilibrium state there is no
connection between M and H : any arbitrary magnetization may be created –
in principle – even in the absence of the field. Nevertheless, one may consider
any value of M as an equilibrium magnetization in a certain – specially pre-
pared – magnetic field. This effective field H e is related to the nonequilibrium
magnetization by the equilibrium relation:
M = nmL(ζ) ζ/ζ ζ = mH e /kB T. (31)
During the equilibrium settling process, the magnetization (31) relaxes to its
equilibrium value (19) as the effective field H e (or ζ) approaches the true field
H (or ξ). Comparing (19) and (31) we see that the latter is obtained by averaging
of e with the distribution function
ζ
We (e) = exp(ζ · e) , (32)
4π sinh ζ
which differs from the Gibbs distribution (30) by replacement of the true dimen-
sionless field ξ by the effective field ζ. Carrying out the averaging in (29) with
the function (32), we find the sought equation [11]

dM (ξ · ζ) M 1 1 1 M × (M × H)
= Ω ×M − 1 − − − . (33)
dt ζ2 τB L(ζ) L(ζ) ζ 6ηφ
This equation together with (31) determines the dependence M (t; H, Ω) in an

implicit form, where the dimensionless effective field ζ is the parameter. In the
case of small departures from equilibrium, the effective field might be represented
as a sum of the true field and some small correction: ζ = ξ + ν. Then from (19)
and (31) in the linear approximation in ν we get

dL(ξ) L(ξ)
M − M 0 = nm ν + ν⊥ , (34)
dξ ξ
where the components
ν = ξ(ν · ξ)/ξ 2 , ν ⊥ = ξ×(ν × ξ)/ξ 2

are parallel and perpendicular to the true field, respectively. Employing the rela-
tion (34), one can reduce (33) to the linear magnetization equation (23), where
relaxation times of the longitudinal and transverse components of magnetization
are
d ln L(ξ) 2L(ξ)
τ = τB , τ⊥ = τB . (35)
d ln ξ ξ − L(ξ)
It is well-established that equation (33) derived by EFM describes magne-
tization processes in real magnetic colloids very well. Predictions of the theory
are well corroborated by experiments on positive [27] and negative [28,29] rota-
tional viscosities, and on magneto-vortical birefringence [24] in ferrofluids. So-
lutions of (33) agree perfectly with the results of numerical integration of the
non-stationary Fokker-Planck equation [13], and what is more, with the numer-
ical simulation of Brownian dynamics performed by Cebers [21] on the basis of
Langevin equation for rotational motion of magnetic grains in the presence of
magnetic field. Phenomenological equation (21) is certainly far simpler for anal-
ysis than (33). However, the latter guarantees the correct description of magneti-
zation even if its deviation from equilibrium value is large (e.g., at Ωτ 1), that
is when (21) leads to erroneous results. Interestingly, a new phenomenological
magnetization equation derived recently [12] from irreversible thermodynamics
(see below) is free partially from the above mentioned shortcoming of (21).
5 Phenomenological Magnetization Equation II

We have shown in Sect. 3 how does Debye equation (18) originate from the
potential Φ(M ). Instead of the magnetization, one can choose as an independent
variable the effective field and introduce the potential Φ̃(H e ). Then instead of
(25) we obtain in similar fashion
dH e ∂ Φ̃
= −γ̃ , γ̃ > 0.
dt ∂H e
Acting further as in Sect. 3, we arrive at the equation (cf. (18))
d H e 1
= − (H e − H) , (36)
dt τB
where we set γ̃ −1 = (∂ 2 Φ̃/∂H 2e )τB . Under this choice, (36) turns into (18) in the
low-field limit, when the true magnetization and its equilibrium value take the
form M = χH e and M 0 = χH, respectively; here χ = nm2 /3kB T stands for
the initial magnetic susceptibility.
Equation (36) is written out in a rotating coordinate system Σ . Reverting
to the immobile system Σ by the general formula (20) and eliminating ω p with
the aid of torque-balance equation (15), we obtain [12]
dH e 1 1
= Ω ×H e − (H e − H) − H e ×(M ×H). (37)
dt τB 6ηφ
96 M.I. Shliomis
This equation (marked below as Sh’01) coincides with the previous phenomeno-
logical equation (21) (marked as Sh’72) in the limit of low magnetic field. How-
ever, due to nonlinearity of the Langevin magnetization law, equations (21) and
(37) predict very different magnetization values for large enough magnitudes of
ξ. Note that as shown above, both these equations are in full agreement with
fundamental principles and hence both they are correct from phenomenological
point of view: irreversible thermodynamics gives preference neither to (21) nor
to (37).
In the case of small deviations from equilibrium, equation (37) can be lin-
earized with respect to H e − H and M − M 0 . Substituting ν = ζ − ξ =
(m/kB T )(H e − H) into (34) gives
dL(ξ) L(ξ)
(M − M 0 ) = 3χ (H e − H) , M ⊥ = 3χ (H e )⊥ .
dξ ξ
Employing these relationships and the equality

dM dM dH e L(ξ) d(H e ) L(ξ) d(H e )⊥
= · = 3χ +
dt dH e dt ξ dt ξ dt
we revert to (23) with τ = τB and τ⊥ from (24). Thus, in weakly nonequilib-

rium situations the old (21) and the new (37) phenomenological magnetization
equations are reduced to each other at any field strength.
6 Testing Magnetization Equations

6.1 Rotational Viscosity in a Stationary Field
As a checking on applicability of the discussed above magnetization equations

we choose their predictions about the rotational or spin viscosity ηr of ferrofluids.
Below we shall compare ηr obtained from (21) and (37) with its almost exact
value resulting from the EFM equation (33).
The Einstein formula for viscosity of suspension, η = η0 (1 + 2.5φ), was ob-
tained without taking into account the rotational motion of suspended particles
relative to carrier liquid. If however ferrofluid is subject to the field H, the lat-
ter impedes free particle rotation – see (15). But any deviation of the particles
angular velocity ω p from the angular velocity of the fluid Ω leads to an addi-
tional dissipation which is just manifested in rotational viscosity ηr [3]. Let us
calculate it.
The boundary wall streamlined by a ferrofluid is acted (on unit area) by
the force fi = [σik ]nk where [ ] denotes difference evaluated across the fluid–
solid interface and n is the normal to the interface. The friction (tangential)
force exerted on the wall is fτ = [στ n ]. By using the electrodynamic boundary
conditions [Hτ ] = 0 and [Bn ] = 0, we get from (16 a)
fτ = η(∂vτ /∂xn ) + 12 (Mτ Hn − Mn Hτ ) . (38)

For the Poiseuille or Couette flow, v = (0, v(x), 0), in a transversal magnetic
field, H = (H, 0, 0), the magnetization has two components: M = (Mx , My , 0),
so (38) may be written in the form fτ = 2(η + ηr )Ω, where 2Ω = ∂v/∂x and
rotational viscosity is defined as
ηr = My H/4Ω . (39)
Thus, the additional viscosity is expressed through the off-axis component of

magnetization My . For small ΩτB , when all three cited above magnetization
equations are reduced to (23), this component is also small, My ∝ ΩτB :
M ⊥ = τ⊥ Ω × M 0 , i.e., My = τ⊥ M0 Ω . (40)
Eliminating My from (39)-(40), we find

1
ηr = 4 τ⊥ M0 H . (41)
Substituting here M0 = nmL(ξ) and τ⊥ from (24), we arrive at the formula [3]
3 ξL(ξ) 3 ξ − tanh ξ
ηr (ξ) = ηφ = ηφ . (42)
2 2 + ξL(ξ) 2 ξ + tanh ξ
So, both phenomenological equations, (21) and (37), predict the same depen-
dence of rotational viscosity on magnetic field strength. In the absence of the
field an individual particle “rolls” freely along corresponding shear surface with
angular velocity ω p equal to Ω , so that ηr (0) = 0 . Conversely, ηr (ξ) attains
its limiting value ηr (∞) = 32 ηφ (the saturation) when rolling of the particle is
replaced by slipping: the field of sufficiently large intensity guarantees constancy
of the particle’s orientation, not allowing it to twist with the fluid. Note that
the saturation value of ηr does not depend on a concrete form of the magne-
tization equation but follows directly from the equation of fluid motion (17).
Actually, in the limit under consideration ω p = 0 , so that (15) takes the form
M × H = −6ηφΩ . Substituting this torque in (17) and grouping there the
second and fourth terms,

η∇2 v + 12 rot (M × H) = η + 32 ηφ ∇2 v ,
we immediately arrive at ηr (∞) = 32 ηφ since the quantity added here to the

ordinary viscosity should be regarded as the rotational one.
The EFM-equation (33) yields a result somewhat different from (42). Sub-
stituting of τ⊥ from (35) into (41) gives
3 ξL2 (ξ)
ηr (ξ) = ηφ . (43)
2 ξ − L(ξ)
Figure 1 shows that though at first sight functions (42) and (43) do not
appear alike, they agree closely in the entire range of their argument. Both of
them approach the saturation value ηr (∞) = 32 ηφ at ξ 1. In the figure we
98 M.I. Shliomis
1.0
Ωτ=0
0.8
ηr(ξ)/ηr(∞)
0.6
0.4
MRSh
0.2 Sh’72,’01
0.0
0 5 10 15 20 25 30
ξ
Fig. 1. Reduced rotational viscosity in the Newtonian limit ΩτB → 0 versus the di-
mensionless field strength, as calculated from (42) (Sh’72 [3] and Sh’01 [12]) and (43)
(MRSh [11]).
plot the reduced rotational viscosity ηr (ξ)/ηr (∞) as a function of ξ. The upper
curve calculated by (43) of the EFM [11] represents a very good approximation.
Actually, as shown in [9,22,23], it hardly differs from the direct solution of the
Fokker-Plank equation (28) in linear approximation in ΩτB . The phenomenolog-
ical equations, (21) and (37), also provide a quite satisfactory description of the
rotational viscosity. Both they result in the lower curve in Fig. 1 that is described
by the Shliomis’ formula (42). This function agrees with (43) in the low- and
high-field limits and deviates from it, at most, on 15 % in the entire range of the
argument ξ. As seen from (39), ηr does not depend on the flow vorticity till My
is proportional to Ω, what takes place only if ΩτB 1. For finite values of ΩτB
the viscosity (39) does depend on Ω. As a result, the function στ n (Ω) deviates
from the linear one, i.e., a ferrofluid acquires rheological properties.
Proceeding to the calculation of the non-Newtonian viscosity on the basis of
(37), it is convenient to pass from the fields H and H e to their nondimensional
values ξ and ζ :
dζ 1 L(ζ)
=Ω×ζ− (ζ − ξ) − ζ × (ζ × ξ) . (44)
dt τB 2τB ζ
At the stated above arrangement of the applied magnetic field with respect to
the fluid flow, the last equation admits a steady solution in which the effective
field ζ tracks the true field ξ with lag angle α, i.e., ζ = (ζ cos α, ζ sin α, 0). The
dependence of ζ and α upon ξ and ΩτB is given by

2ΩτB ζ ζ
ξ2 − ζ 2 = , cos α = . (45)
2 + ζL(ζ) ξ
Substituting My = nmL(ζ) sin α in (39) and using (45), we obtain
3 ζL(ζ)
ηr = ηφ . (Sh 01) (46)
2 2 + ζL(ζ)
By the same way we find from (33)
2ΩτB ζL(ζ) ζ
ξ2 − ζ 2 = , cos α = , (47)
ζ − L(ζ) ξ
that results in
3 ζL2 (ζ)
ηr = ηφ . (MRSh) (48)
2 ζ − L(ζ)
The solution of equation (21) can be presented in a similar form. Let us in-
troduce a new variable ζ instead of M by the relation M = M0 (ζ/ξ) where
M0 = nmL(ξ). It is worth noting that ζ is no more an effective field unlike
ζ in preceding relationships (45)–(48). By substituting the components Mx =
M0 (ζ/ξ) cos α and My = M0 (ζ/ξ) sin α in (21), we get cos α = M/M0 = ζ/ξ and
2ΩτB ξζ
ξ2 − ζ 2 = . (49)
2ξ + ζ 2 L(ξ)
This expression together with the definition (39) yield
3 ζ 2 L(ξ)
ηr = ηφ . (Sh 72) (50)
2 2ξ + ζ 2 L(ξ)
In the limit ΩτB 1, one can neglect the value ΩτB in (45), (47) and (49), after
what all three of these relationships are reduced to ζ = ξ. Eliminating now ζ from
(13) and (17), we see that, as it should be, expressions (50) and (46) obtained
from the old and the new phenomenological magnetization equations turn into
(42), while the EFT formula (48) is transformed into (43). When, however, the
ferrofluid is subjected to a sufficiently large shear rate, ΩτB ≥ 1, the flow induces
– along with the Brownian motion – a quotient demagnetization since under the
viscous shear the magnetic grains tend to be rotated out of alignment with the
magnetic field. Formally, this effect originates from decreasing the parameter ζ
determined by (45), (47) and (49). According to these equations, ζ = ξ when
ΩτB = 0 but the more there is of Ωτ at constant ξ, the less there is of ζ. The
reduction of the magnetization leads in turn to some decrease in the rotational
viscosity. This decrease, imperceptible in practice up to ΩτB 1, then becomes
very significant. Figures 2–4 illustrates the dependence of the viscosity increase
on the magnetic-field strength for three values of the product ΩτB . Interestingly,
under the finite shear rate the viscosities given by (46) and (50) do not coincide
100 M.I. Shliomis
1.0
Ωτ=2
0.8
ηr(ξ)/ηr(∞)
0.6
0.4
MRSh
0.2 Sh’01
Sh’72
0.0
0 5 10 15 20 25 30
ξ
Fig. 2. Dependence of the rotational viscosity on the field for the dimensionless shear
rate ΩτB = 2, as calculated from (47)–(48) MRSh, (45)–(46) Sh’01, and (49)–(50)
Sh’72.
1.0
Ωτ=4
0.8
ηr(ξ)/ηr(∞)
0.6
0.4
MRSh
0.2 Sh’01
Sh’72
0.0
0 5 10 15 20 25 30
ξ
Fig. 3. Same as Fig. 2, but for ΩτB = 4.
1.0
Ωτ=6
0.8
ηr(ξ)/ηr(∞)
0.6
0.4
MRSh
0.2 Sh’01
Sh’72
0.0
0 5 10 15 20 25 30
ξ
Fig. 4. Same as Figs. 2 and 3, but for ΩτB = 6.
with each other any more. As seen from the plot, the higher the shear the
more discrepancy between viscosity values predicted by the new and the old
phenomenological equations. At high shear in a high field, the old equation
[(21), Sh’72] predicts a hysteresis of viscosity (see Fig. 4), which however is
corroborated neither by direct calculations of [13,20] nor by the solution (47)–
(48) of the EFM equation (33). The new equation [(37), Sh’01] also does not
predict such a hysteresis, but it provides us with a quite satisfactory viscosity
description in a wide region of parameters ξ and ΩτB . Indeed, in this entire
region the solutions of (33) and (37) agree closely, as shown in Figs. 2–4. Thus,
in the case of a stationary magnetic field, equation (37) can be recommended
for an employment on the same level with (33). It is worth nothing that all the
above calculations, carried out for a shear flow, apply equally to a rigid rotation
of a ferrofluid with an angular velocity Ω in a constant transversal magnetic
field, H ⊥ Ω, and to a quiescent ferrofluid subjected to a uniform rotating field
H = (H cos Ωt, H sin Ωt, 0) as well.
6.2 “Negative Viscosity” under an Alternating Magnetic Field

For a stationary field and small ΩτB , equation (23) yields the magnetic torque
M × H = −τ⊥ M0 HΩ. (51)
On comparing this expression with the torque-balance equation (15),
M × H = 6ηφ(Ω − ω p ) , (52)
102 M.I. Shliomis
and employing the definition (41), ηr = 14 τ⊥ M0 H , we find a clear and physically

transparent relationship
3 Ω − ωp
ηr = ηφ . (53)
2 Ω
Note that being obtained for a stationary magnetic field, (53) is valid also for
oscillatory field, since (51) and (52) remain to be true after averaging over the
period of the field variation as well. According to (53), ηr > 0 if magnetic grains
rotate slowly than the fluid (ωp < Ω) and then decelerate the flow. Such a
situation takes place in a stationary (see Sect. 6.1) or slow time-varying magnetic
field. The field prevents free particle rotation with the fluid angular velocity Ω
and thereby forces the surrounding liquid to flow past the particles. It leads to
an additional dissipation of the kinetic energy of the fluid and is manifested in
the rotational viscosity ηr > 0.
An alternating, linearly polarized magnetic field H = (H0 cos ωt, 0, 0) in-
duces rotational swings of the grains but does not single out any preferred direc-
tion of their rotation. Therefore, an averaging over a physically small element of
ferrofluid volume results in spin of zero: ωp = 0. However, any flow with a vor-
ticity not zero is sufficient to break the degeneracy of the rotation direction and
leads to a nonzero macroscopic spin rate of the magnetic grains. When the field
frequency is high enough, the grains rotate faster than the fluid (ωp > Ω) and
then they spin up the flow. The acceleration happens, naturally, at the expense
of ac-field energy and manifests itself in a reduction of the total viscosity, i.e.,
in a negative rotational viscosity: ηr < 0 [see (53)].
This “negative viscosity” effect has been predicted by Shliomis and Morozov
[30] and verified experimentally by Bacri’s [28] and Rehberg’s [29] groups. The
first prediction has been made on the base of Debye-like equation (21); afterwards
the results [30] were recalculated [28] by using a more precise EFM equation (33).
However, in the case of a low field amplitude, ξ0 ≡ mH0 /kB T 1, and a low
shear rate, ΩτB 1, both these equations admit the same simple analytical
solution [30]
1 1 − ω 2 τB2
ηr = ηφξ02 2 . (54)
8 (1 + ω 2 τB2 )
As seen, at ωτB = 1 the rotational viscosity changes its sign, passing from the
domain of positive (ωτB < 1) to that of negative
√ (ωτB > 1) values. In the latter
domain, it attains the minimum at ωτB = 3 and tends to zero at ωτB → ∞
since in this limit ferrofluid does not have enough time for re-magnetization: the
grains cease to feel the magnetic field.
To solve (33) for finite amplitudes of oscillating magnetic field, it is convenient
to present this equation in the dimensionless form measuring the time in units
of τB :

d ζ ζ L(ζ) 1 3L(ζ)
L(ζ) = τB Ω× L(ζ) − (ζ −ξ)− 2 1 − ζ×(ζ×ξ) . (55)
dt ζ ζ ζ 2ζ ζ
The problem contains usually a small parameter ΩτB . In this case (55) may be
solved by the theory of perturbations. In zero approximation in the parameter,
the effective field is evidently parallel to the true field, while its magnitude is
determined by the equation
−1
dζ d ln L(ζ)
=− (ζ − ξ0 cos ωt). (56)
dt d ln ζ
In the linear approximation in ΩτB , (55) yields
M = M (0) h + M (1) τB Ω ×h, M (0) = nmL(ζ), M (1) = M (0) Ψ (ζ),
where h = H/H is the unit vector along the field and function Ψ (ζ) is satisfied
to the linear equation

dΨ ξ0 1 1
=1− − Ψ cos ωt. (57)
dt 2 L(ζ) ζ
The magnetic torque M×H can be averaged over the period of field variation
2π/ω because we are interested in ω ∼ τB−1 while τB is always much less than
the characteristic hydrodynamic time ∼ ρl2 /η, where l is the reference spatial
scale of the fluid motion. For the mean magnetic torque one gets
M × H = −6ηφ g(ξ0 , ωτB )Ω, g = 12 ξ0 L(ζ) Ψ (ζ) cos ωt. (58)
Averaging now the equation of fluid motion (17) over the time and substituting
into (17) the mean torque (58), one may group the second and fourth terms in
this equation:
η∇2 v − 3ηφg rotΩ = η (1 + 32 φg)∇2 v.
Thus the rotational viscosity is ηr = 32 ηφg. In accordance with (52) and (58),
the angular velocity of the particles may be written as ωp = (1 − g)Ω, so
that ηr can be presented in the form (53). For arbitrary ξ0 and ωτB the prob-
lem (56)–(57) was solved numerically [28]. Results of the computations are dis-
played in the map of viscosity, Fig. 5, where a set of isolines of reduced viscosity
g(ξ0 , ωτB ) = ηr (ξ0 , ωτB )/ηr (∞, 0) is presented in the plane (ξ0 , ωτB ). The iso-
line g = 0 parts the plane into two regions: g > 0 for ωτB ≤ 1 and g < 0 for
ωτB ≥ 1. A similar map of viscosity, but calculated from the phenomenologi-
cal magnetization equation (21), Sh’72, is shown in Fig. 6. A strong similarity
between a plot of experimentally obtained isolines and that of Fig. 5 has been
observed in [28]. As for the map in Fig. 6, the experimental data [28] agree
with it on the whole, but there are some discrepancies in details. Namely, in
the experiment [28] as well as in Fig. 5 the neutral isoline, g = 0, bends to
the right, while in Fig. 6 it bends slightly to the left. Still more there is the
difference between the isolines of Fig. 5 and those as calculated from the new
phenomenological equation (37), Sh’01. Generally, this equation is very appro-
priate for description of ferrofluid magnetization in a stationary magnetic field
(see Sect. 6.1), whereas in an alternating field it is working satisfactory only in
the low-frequency limit ωτB 1. [The only exception to the rule represents itself
the case of rotating magnetic field of the kind of H = (H0 cos ωt, H0 sin ωt, 0).
104 M.I. Shliomis
20
−0.7
−0.5
−0.3
−0.1
15
0
+0.1
+0.3
+0.5
10
ξ0
+0.7
0
0 1 2 3 4 5 6
ωτB
Fig. 5. Isolines of reduced viscosity ηr (ξ0 , ωτB )/ηr (∞, 0) = g in the plane (ξ0 , ωτB )
for −0.7 ≤ g ≤ +0.7 , as calculated from EFM equation (33).
Such a field is reduced evidently to the stationary field H = (H0 , 0, 0) acting

upon the fluid rotating as a whole with the angular velocity Ω = (0, 0, ω). Thus,
as (37) is valid for any shear rate ΩτB in a stationary field, it is also valid for
any frequency of the field rotation ωτB ].
A difference between discussed magnetization equations is also manifested at
the relaxation from an equilibrium magnetization in a quiescent ferrofluid after
the field is suddenly switched off. Then the fluid remains at rest, Ω = 0, so
M and H e are always parallel to H. Hence equations (21), (33) and (37) are
reduced to
dM /dt = −(M − M0 )/τB , (59 a)
dM /dt = −(1 − H/He )M/τB , (59 b)
and
dHe /dt = −(He − H)/τB , (59 c)
respectively. In Fig. 7 we plot the decay of reduced magnetization M (t)/M0
according to (59) with M0 = nmL(ξ) for some initial field magnitudes ξ =
mH/kB T . As the true field H is switched off at the moment t = 0, equations
(59 a) and (59 b) coincide with each other at t > 0, when their solution reads
M (t)/M0 = exp(−t/τB ) , (60)

20
−0.7
−0.5
−0.3
−0.1
15
0
+0.1
+0.3
+0.5
10
ξ0
+0.7
0
0 1 2 3 4 5 6
ωτB
Fig. 6. Same as Fig. 5, but calculated from Debye-like equation (21).
i.e., it does not depend on ξ . Analogously, (59 c) has the solution He (t) =
H exp(−t/τB ), so that we find
M (t)/M0 = L(ξe−t/τB )/L(ξ) . (61)
The last decay predicted by the new magnetization equation (37), Sh’01, is
exponential only in the limit ξ 1 , while the old phenomenological equation
(21) and – what is much more important – the EFM equation (33) predicts
exponential decay of the magnetization for any values of ξ . This difference in
relaxation behavior side by side with the difference in the ferrofluid viscosity can
be of relevance for testing the magnetization equations and the interpretation of
corresponding experiments. At the description of nonstationary situations (like
that as presented in Fig. 7) one should, however, a priori give preference to (21)
and (33) before (37) since all previous predictions of (33) always were realized.
6.3 Response to Rotating Magnetic Field

When the field rotates in the (x, y)-plane, ferrofluid magnetization also rotates
in this plane with the same speed as the field but lags behind in phase by a
certain angle due to the finite relaxation time:
H = (H0 cos ωt, H0 sin ωt, 0), M = [M cos(ωt − α), M sin(ωt − α), 0]. (62)
106 M.I. Shliomis
1.0
ξ=0.05
0.8 ξ=5
ξ=10
ξ=20
0.6
M/M0
0.4
0.2
0.0
0 1 2 3 4 5 6
t/τB
Fig. 7. Time dependence of the reduced magnetization M (t)/M0 after the field ξ is
switched off, as described by (61). The lowest curve also represents the solution (60)
for any field.
Here M and α should be determined by one of three discussed above magne-

tization equations. For weakly nonequilibrium situations, when all of them are
reduced to (23), we find
M = M0 cos α, tan α = (ω − Ω)τ⊥ , (63)
where M0 = nmL(ξ) and τ⊥ is defined by (24) or (35), depending on the model

we use.
Substituting from (62) into (16 b) gives for a nontrivial (non-diagonal) com-
ponent of the stress tensor

∂vx ∂vy
σxy = η + 1
+ 8π H02 sin 2ωt + 12 M0 H0 cos α sin(2ωt − α). (64)
∂y ∂x
On averaging over the period of the field variation both the magnetic terms in
(64) disappear, so the rotating field does not induce any flow inside the bulk of
the fluid. Interestingly, the magnetic grains do rotate with the angular velocity
M0 H 0 ωτ⊥
ωp = · 2 ,
6ηφ 1 + ω 2 τ⊥
while the fluid remains quiescent [31]. The point is, each a grain forces to spin
the near-by mass of the viscous liquid thereby becoming a center of microscopic
vortex, the size of which does not exceed the mean distance between the grains
∼ φ−1/3 d. But such a motion is not hydrodynamic yet, because averaging of the
microwhirls over physically small fluid volume does not result in macroscopic
vorticity Ω = 12 rot v. Indeed, microwhirls created by the neighboring grains
compensate each other like elementary electric currents of neighboring molecules
in Ampere’s model of ferromagnetism do that.
The uniformity of both the internal moment of momentum S = Iω p and
the magnetization M breaks off at the ferrofluid border. A jump-like increase
in magnetization on the boundary surface (outside the fluid M = 0) results
in magnetic tangential surface stress. The latter just provide the coupling of
rotating field and the fluid motion.
Let us consider ferrofluid confined within the layer by horizontal planes x =
0, l. Under the field (62) rotating in the vertical plane (x, y), the fluid can come
into the motion along axis y, v = [0, v(x), 0], so that Ω = (0, 0, Ω) where
Ω = 12 (dv/dx). The tangential force density (38) acting the upper surface is
dv
fy = [σyx ] = η − 1 (Mx Hy − My Hx ) . (65)
dx 2
Substituting from (62) and (63) into (65) yields
dv (ω − Ω)τ⊥
fy = η − 12 M H0 sin α = 2ηΩ − 12 M0 H0 2 . (66)
dx 1 + (ω − Ω)2 τ⊥
As seen from (66), if the surface x = l is free, fy = 0, the rotating field generates
the flow with the constant vorticity
M0 H0 ωτ⊥
Ω≈ · 2 , (67)
4η 1 + ω 2 τ⊥
so the plane Couette flow – with the linear velocity profile v(x) = 2Ωx – arises.
[Note that we have neglected in the right-hand side of (67) the value Ω in
comparison with ω since Ω/ω ≤ 14 φξ 2 at ξ 1 and Ω/ω ≤ 32 φ at ξ 1]. If,
conversely, the upper surface is fixed, Ω = 0, to hold it one needs to apply the
force in the opposite direction:
ωτ⊥
fy = − 12 M0 H0 2 . (68)
1 + ω 2 τ⊥
Let the field (62) rotates in horizontal plane around a long (l R) vertical
cylinder of the radius R completely filled with ferrofluid. Then the lateral walls
of the cylinder are acted upon by the tangential force density

dv v
fϕ = [σϕr ] = η − + 12 M H0 sin α , (69)
dr r
cf. (65), where v(r) is the azimuthal component of the fluid velocity and H0
stands for the field amplitude inside the fluid. Restricting ourselves by the case
of low enough field strength, ξ < 1, we obtain from (63) and (69)

dv v (ω − Ω)τB dv v
fϕ = η − 1
+ 2 χH02
, Ω=2 1
+ . (70)
dr r 1 + (ω − Ω)2 τB2 dr r
108 M.I. Shliomis
This relationship allows us to calculate the magnetic torque T = 2πR2 lfϕ acting
on the motionless cylinder (v = dv/dr = 0):
ωτB
T = χH02 V , (71 a)
1 + ω 2 τB2
where V = πR2 l is the volume of cylinder. Satisfying the boundary conditions of

continuity of Hϕ and Br on the fluid border r = R, one can express the torque
via the intensity H0 of external magnetic field [32]:
χH02 V ωτB
T = . (71 b)
(1 + ω 2 τB2 )[(1 + 2πχ)2 + ω 2 τB2 ]
This expression agrees well with experimental data [33,34]. To interpret (71 b), it
should be noted that the stationary fields H and H are coupled by the relation
H
H=
1 + 2πχ
where 2π is the demagnetization factor of cylinder in the direction transverse
to its axis of symmetry (axis z). Therefore, at the low frequency of the field
rotation, ωτB 1, the magnetic torque (71 b) may be presented in its usual
form [5]
T = M × H, (72)
where we have introduced the total magnetic moment of the cylinder
χH
M = M V, M=
1 + 2πχ
and have taken into account that the small angle between M and H is equal to
ωτB .
Revert to Eq. (70). If the cylinder is not fixed, the equation fϕ = 0 determines
the velocity of free fluid rotation v(r) = ωr, i.e., Ω = ω: the fluid rotates as a
whole about the vertical axis. Finally, if the cylinder is fixed, but not completely
filled with ferrofluid, the latter comes into motion due to magnetic tangential
stresses on the free fluid surface. The moving surface involves an adjacent fluid
layer and forms the circular hydrodynamic flow first observed in [34]. This phe-
nomenon known as rotational effect has long held fascination for investigators.
Many theories of the spin-up motion were devoted to the effect, but only re-
cently it was clarified by experiments and theory of Rosensweig et al. [35] and
Pshenichnikov et al. [32,33,36].
7 Conclusion
Thus, there exist two basic models of the ferrofluid dynamics. One of the two
represents a quasistationary theory (Sect. 1), which provides an irreproachable
description of any ferrohydrostatic problems. It is also widely employed at the
study of thermal convection since in non-isothermal situations the magnetic force

is not a potential one and then it is not reduced to the simple re-normalization
of the pressure.
The general theory (Sects. 2–5) rejects assumptions of instant relaxation of
the magnetization M and its collinearity with the magnetic field H, and makes
allowance for the particle rotation with respect to the surrounding liquid. The
result is that the product M × H can take some finite value and so constitutes a
body couple. The model consists of hydrodynamic and Maxwell equations plus
a magnetization equation. In the capacity of the latter, one should use macro-
scopic equation (33) derived microscopically from the Fokker–Planck equation
by the effective field method. Equation (33) guarantees a correct quantitative de-
scription of magnetization processes even if deviations of magnetization from its
equilibrium value are sufficiently large. However, in the case of a weakly nonequi-
librium situation, one should give preference to phenomenological equations (21)
or (37): They describe such a state sufficiently well, being at the same time far
simpler for analysis than (33). A deviation from the state of equilibrium may be
often characterized quantitatively by the value of the dimensionless shear rate
ΩτB or/and the field frequency ωτB . As shown in Sect. 6, equation (37) can be
recommended for high shear but only stationary or slow time-varying magnetic
field, ωτB 1. On the contrary, equation (21) is valid even for ωτB ∼ 1 but
only at low shear. Finally, in a low field, ξ 1, all of three discussed above
magnetization equations coincide with each other.
Acknowledgements
I thank Ronald Rosensweig for his comments on the manuscript and Alexei
Krekhov for providing Figs. 5 and 6. This work was supported by the Alexander
von Humboldt Foundation by dint of the Meitner–Humboldt research award,
and the Israel Science Foundation under Grant No. 336/00.
References
1. J.L. Neuringer, R.E. Rosensweig: Phys. Fluids 7, 1927 (1964)
2. R.E. Rosensweig: Ferrohydrodynamics (Cambridge University Press, Cambridge
1985)
3. M.I. Shliomis: Sov. Phys. JETP 34, 1291 (1972)
4. M.I. Shliomis: Sov. Phys. Usp. 17, 153 (1974)
5. L.D. Landau, E.M. Lifshitz: Electrodynamics of Continuous Media, second ed.
(Pergamon Press, New York 1984)
6. M.I. Shliomis: ‘Convective Instability of Magnetized Ferrofluids: Influence of Mag-
netophoresis and Soret Effect’. In: Thermal nonequilibrium phenomena in fluid
mixtures, ed. by W. Köhler, S. Wiegand (LNP, Springer, 2001)
7. B.A. Finlayson: J. Fluid Mech. 40, 753 (1970)
8. M.I. Shliomis, B.L. Smorodin: J. Magn. Magn. Mater. (2002) (in press).
9. Yu.L. Raikher, M.I. Shliomis: Adv. Chem. Phys. 87, 595 (1994)
10. M.I. Shliomis: Sov. Phys. JETP 24, 173 (1967)
110 M.I. Shliomis
11. M.A. Martsenyuk, Yu.L. Raikher, M.I. Shliomis: Sov. Phys. JETP 38, 413 (1974)
12. M.I. Shliomis, Phys. Rev. E 64, 063501 (2001)
13. M.I. Shliomis, T.P. Lyubimova, D.V. Lyubimov: Chem. Eng. Comm. 67, 275
(1988)
14. M.I. Shliomis, Phys. Rev. E 64, 060501(R) (2001)
15. P. Debye: Polar Molecules (Dover, New York 1929)
16. L.D. Landau, I.M. Khalatnikov: Dokl. Akad. Nauk SSSR 96, 469 (1954)
17. R.E. Rosensweig: this issue, p. 63.
18. J.P. McTague: J. Chem. Phys. 51, 133 (1969)
19. E.N. Mozgovoi, E.Ya. Blum, A.O. Tsebers: Magnetohydrodinamics 9, 52 (1973)
20. O. Ambacher, S. Odenbach, K. Stierstadt: Z. Phys. B–Condensed Matter 86, 29
(1992)
21. A.O. Cebers: Magnetohydrodynamics 20, 343 (1984); 21, 357 (1985); E. Blums,
A. Cebers, M. Maiorov: Magnetic Fluids (W. de Gruyter, Berlin 1997)
22. A.C. Levi, R.F. Hobson, F.R. McCourt: Canad. J. Phys. 51, 180 (1973)
23. B. U. Felderhof: Magnetohydrodinamics 36, 396 (2000)
24. B.M. Heegaard, J.-C. Bacri, R. Perzynski, M.I. Shliomis: Europhys. Lett. 34, 299
(1996)
25. F. Gazeau, B.M. Heegaard, J.-C. Bacri, A. Cebers, R. Perzynski: Europhys. Lett.
35, 609 (1996)
26. F. Gazeau, C. Baravian, J.-C. Bacri, R. Perzynski, M.I. Shliomis: Phys. Rev. E
56, 614 (1997)
27. J.P. Embs, H.W. Müller, C. Wagner, K. Knorr, M. Lücke: Phys. Rev. E 61, R2196
(2000)
28. J.-C. Bacri, R. Perzynski, M.I. Shliomis, G.I. Burde: Phys. Rev. Lett. 75, 2128
(1995)
29. A. Zeuner, R. Richter, I. Rehberg: Phys. Rev. E 58, 6287 (1998)
30. M.I. Shliomis, K.I. Morozov: Phys. Fluids 6, 2855 (1994)
31. V.M. Zaĭtzev, M.I. Shliomis: J. Appl. Mech. Tech. Phys. 10, 696 (1969)
32. A.F. Pshenichnikov, A.V. Lebedev: Magnetohydrodinamics 36, 317 (2000)
33. A.V. Lebedev, A.F. Pshenichnikov: J. Magn. Magn. Mater. 122, 227 (1993)
34. R. Moskowitz, R.E. Rosensweig: Appl. Phys. Lett. 11 (1967)
35. R.E. Rosensweig, J. Popplewell, R.J. Johnston: J. Magn. Magn. Mater. 85, 171
(1990)
36. A.F. Pshenichnikov, A.V. Lebedev, M.I. Shliomis: Magnetohydrodinamics 36, 339
(2000)
Supplementary Glossary
M magnetization
H magnetic field
H e effective magnetic field
B magnetic induction: B = H + 4πM
F magnetic force (volume density)
v ferrofluid velocity
Ω flow vorticity: Ω = 12 rot v
S internal angular momentum (volume density): S = Iω p
ω p angular velocity of magnetic particles
I moment of inertia of particles in a unit volume
p pressure
σik stress tensor
T temperature
kB Boltzmann’s constant
ρ mass density
φ concentration of magnetic grains (volume fraction)
n number density of magnetic grains
m magnetic moment of a single particle: m = Md V
V volume of a single particle
d particle diameter
D particle diffusion coefficient
Md domain magnetization of the particle material
W orientational distribution function of particle magnetic moments
e unit vector along particle magnetic moment: e = m/m
h unit vector along magnetic field: h = H/H
ξ dimensionless magnetic field: ξ = mH/kB T
ζ dimensionless effective magnetic field: ζ = mH e /kB T
L(ξ) Langevin function: L(ξ) = coth ξ − ξ −1
η shear viscosity
ηr rotational viscosity
χ initial magnetic susceptibility
Φ thermodynamic potential
t time
τB Brownian diffusion time
τN Néel diffusion time
τ⊥ transverse magnetization relaxation time
τ longitudinal magnetization relaxation time
ω magnetic field frequency
Ferrofluid Dynamics
Hanns Walter Müller1 and Mario Liu2

1
Max-Planck Institut für Polymerforschung, Ackermannweg 10, D-55128 Mainz,
Germany
2
Institut für Theoretische Physik, Universität Tübingen, D-72676 Tübingen,
Germany
Abstract. It is demonstrated how the complete structure of Newtonian ferrofluid dy-

namics, including magneto-dissipative effects, is derived from general principles. Fer-
rofluids are taken as homogeneous magnetizable fluids with a magnetic relaxation time
sufficiently large to compare with other hydrodynamic time scales. The derivation
makes no reference to the angular momentum of the ferromagnetic grains. The results
are independent of most microscopic details, such as the form or shape of the par-
ticles or whether magneto-relaxation takes place via the intrinsic Néel process or by
Brownian rotational diffusion. Both the Debye theory of Shliomis and his effective-field
approach are shown to be special cases of the new set of equations.
1 Introduction and Motivation

Ferrofluids, or magnetic fluids, are colloidal suspensions of nano-sized ferromag-
netic particles stably dispersed in a carrier liquid. When exposed to an external
magnetic field, they behave paramagnetically, with susceptibilities χ unusually
large for liquids. This property opens up a wide range for interesting and promis-
ing future applications [1,2]. Equally important, ferrofluid systems are also at-
tractive under a theoretical point of view. Combining general fluid mechanics
with electromagnetism causes ferrofluids to display many unexpected behavior,
severely testing and with time perfecting our understanding of the hydrodynam-
ics of polarizable media [1,3,4,5].
The fundamental theory for ferrofluid hydrodynamics has been worked out by
Rosensweig and is well described in the first seven chapters of his text book [1].
Thereby he follows the so called quasi-equilibrium approach, which is based on
the assumption that the local magnetization M(r, t) is in steadfast equilibrium
with the local magnetic field, i.e. M(r, t) = Meq [H(r, t)]. This theory covers
a wide range of interesting and relevant effects, such as the deformation of a
ferrofluid drop in a homogeneous magnetic field or the instability of a ferrofluid-
air interface, when it is exposed to a magnetic field perpendicular to the surface
(Rosensweig instability).
However, very soon it became evident that the quasi-equilibrium approxi-
mation does not suffice to account for all ferrohydrodynamic behavior, even in
stationary flow configurations at static applied fields. Among the more remark-
able flow phenomena which go beyond the quasi-equilibrium approach is the
magneto-viscous effect: A ferrofluid tube flow in the presence of a static mag-
netic field experiences an extra dissipation, which manifests macroscopically as

Ferrofluid Dynamics 113
an enhanced effective shear viscosity [6,3]. Even more spectacular is the acceler-
ation of the flow in response to a high-frequency AC-field [7,8,9] (also denoted as
”negative viscosity”, though only a negative viscosity increment was observed ).
That way, the oscillating magnetic field pumps energy into the rotating motion
of the ferromagnetic grains resulting in an acceleration of the flow.
The occurrence of the magneto-viscous effects is intimately related to the
fact that the local magnetization M(r, t) deviates from its equilibrium value
Meq [H(r, t)], where H(r, t) is the local magnetic field. Significant increments
δM = (M−Meq ) are expected to appear whenever the magnetic relaxation time
τ compares to the other relevant hydrodynamic time scales. Phenomena related
to the finiteness of τ are commonly denoted as magneto-dissipative effects.
As outlined in Rosensweig’s textbook, the microscopic mechanism responsible
for the magneto-relaxation is either due to particle rotation against the viscosity
of the liquid carrier (Brownian rotational diffusion) or by re-orientation of the
magnetic moments relative to the crystallographic orientation of the ferromag-
netic grains (Néel relaxation). Which of these mechanisms predominates depends
on the specific anisotropy energy of the employed ferromagnetic material, the
size of the suspended grains, and the viscosity of the carrier liquid. Since real
ferrofluid suspensions usually exhibit a broader particle size distribution, it is in
general a combination of both processes which determines the effective magnetic
relaxation time for a given ferrofluid species.
Assuming that the Brownian mechanism is the principal source of dissipa-
tion, the intuitive picture of particles rotating against the viscous carrier lead
Shliomis [3] to his theory for magneto-dissipative ferrohydrodynamics. To that
end he included both the magnetization M and the mechanical angular mo-
mentum density S of the grains as additional thermodynamic variables. After
eliminating the latter, an extra momentum flux remains, which enters the stress
tensor in the form
ΔΠij = 12 εijk (H × M)k , (1)
This term exactly compensates the antisymmetric part of Maxwell’s stress Hi Bj ,
if H and M are non-parallel. Clearly, being treated as a separate independent
variable, the magnetization requires an extra evolution equation. According to
Shliomis there are two versions of this equations, of which the first is a phe-
nomenological relaxation equation with a Debye-like relaxation term in the form
δM/τ . In the ferrofluid literature this approach is frequently referred to as the
Debye theory. In combination with Eq (1), many magneto-dissipative phenom-
ena, especially the elevated shear viscosity, were successfully explained. The sec-
ond variant of his relaxation equation for M is more elaborate as it is derived
from a microscopic, statistical investigation of the rotary diffusion of magnetic
particles. Since the problem was solved with the assistance of the effective field
method, this second variant is commonly denoted as the effective-field theory,
or EFT. The latter is rather more complicated and unwieldy than Debye as
it provides an evolution equation for a quantity called the “effective magnetic
field” from which the magnetization is to be determined in a subsequent step.
114 H.W. Müller and M. Liu
The EFT was found to explain the “negative viscosity” experiment much more
convincingly than the Debye-like approach [8].
Comparing both theories, and emphasizing that the EFT is the more rigor-
ous and accurate one, Shliomis concluded that it is valid for all experimentally
relevant situations [8]. The Debye theory, on the other hand, he considers to be
adequate only in the limit of small deviations from the magnetization equilib-
rium [8], δM M eq , implying the hydrodynamic low frequency limit ωτ 1,
where 1/ω is the characteristic time scale of the experiment.
We do think that these assessments, referring to both (i) the general validity
of the EFT and (ii) the limited validity of the phenomeonological Debye approach
are in need of a clarification:
First, owing to their microscopic input, EFT is in it essence a microscopic
theory, with necessarily rather specific inputs. In the present case, ferrofluids
are considered as suspensions of noninteracting, spherical, equal sized Brownian
rigid dipoles. In the framework of these restricting simplifications, EFT is rigor-
ously valid. But one always has to be aware of the limitations and deficiencies
purchased with the above idealizing assumptions. Regarding to the fact that real
ferrofluids are suspension of interacting, non-spherical, poly-dispersed particles,
whose magnetic relaxation usually involves both Brownian and Néel processes,
EFT cannot be expected to be a sufficient approach under all circumstances.
Second, the potential of a proper macroscopic theory is much larger than for
any microscopic approach, because it is constructed on the sole base of general
principles, without any specific microscopic information. For ferrofluids such a
macroscopic theory is very similar to Shliomis’ Debye theory but it will not suffer
from the above constraint to the low frequency regime. In the following we shall
denote such a modified approach as the rectified Debye theory. Demonstrating
how such a macroscopic theory – we shall call it ferrofluid dynamics (FFD) – can
be derived from the concepts of non-equilibrium thermodynamics is the purpose
of this article.
Generally speaking, any macroscopic theory consists of two separate ingre-
dients. First the structure of the equations, which is solely based on conserva-
tion laws and symmetries, and second the material-specific parameters such as
susceptibilities and transport coefficients. The aim of the present lecture is to
show how the general structure of a hydrodynamics for ferrofluids is derived. No
attempt will be undertaken to provide values for the material-dependent coeffi-
cients. Following the standard approach in macroscopic physics, those quantities
(such as viscosities or susceptibilities) are usually measured. So we leave them
to be determined by a series of suitable experiments. An alternative way is to
calculate the coefficients from an appropriate microscopic model. Such a model
is for example EFT but one must not forget that it is valid in its specified range
of validity. In particular it does not necessarily yield – and indeed it does not,
as we shall see – the most general structure of equations, which are compatible
with symmetries and conservation laws.
2 Outline of the Strategy

In this section we give a layout of our strategy for deriving the structure of the
ferrofluid dynamics. We are guided by the general observation that hydrody-
namic theories are a very successful tool to account for low frequency large wave
number phenomena in condensed matter physics. For isotropic non-magnetic
fluids the relevant variables are the conserved quantities: densities of energy,
mass and momentum. For magnetizable fluids it is obvious to supplement the
set of variables by the magnetic field H (or the magnetic induction B). That
way we recover the thermodynamic variables of the quasi-equilibrium theory.
Rosensweig assumes that the magnetization relaxation is instantaneous on the
time scale of the other hydrodynamic processes of interest. In other words, M
is in steadfast equilibrium with the magnetic field, M(r, t) = Meq (H(r, t). Thus
magneto-dissipative effects, which necessarily imply δM = 0, are disregarded.
It is usually believed that the interpretation of magneto-dissipative effects
requires to incorporate the magnetization as an extra thermodynamic degree of
freedom with its own separate evolution equation. But this is incorrect: Recall
that linear electrodynamics is well able to account for dissipative effects by in-
troducing an imaginary part of the electrical susceptibility. Focusing to the more
general case of simultaneous electro- and magneto-dynamic processes, the ap-
propriate description has been worked out in Liu’s hydrodynamic Maxwell theory
(HMT) [10]. Meanwhile this theory has been demonstrated to be well able to ex-
plain magneto-dissipative phenomena such as the field enhanced shear viscosity
or the fluids spin up in a rotating field [11]. But like any hydrodynamic theory,
the HMT is valid for small deviations from equilibrium. i.e. it is restricted to
the case δM M eq and the low frequency limit ωτ 1. In ferrofluids τ is
typically of order 10−4 − 10−3 s and thus the above low frequency constraint
is rather severe and easily violated (for instance by the “negative viscosity” ex-
periment). When this happens, it is justified to include the magnetization as an
independent variable to render the theory applicable also to the case ωτ 1 and
δM/M eq 1.
In the following section we shall derive the equation of motion for the magne-
tization. This includes also the necessary modifications in all the other equations,
which are related to the fact that M is turning independent. We shall denote this
approach as ferrofluid dynamics (FFD). The method is standard non-equilibrium
thermodynamics, with the sole input of conservation laws and symmetries. Be-
sides the assumption that the magneto-relaxation can be covered by a single
relaxation time τ there is no further material specific input. Consequently our
approach is fairly general, it remains valid even if the magnetic particles inter-
act appreciably with each other, if the particles are of non-spherical shape, or if
they are of different size. Moreover, the theory holds irrespective of the micro-
scopic relaxation mechanism, whether it is governed by the Brownian or the Néel
mechanism. Furthermore, since no reference is made to the angular momentum
S of the grains, the result is valid both for suspensions and for homogeneous
magnetizable continua. Note however, that the present approach does not cover
non-Newtonian rheological effects.
3 Derivation of the Equations

In the present section we derive the structure of ferrofluid dynamics. As outlined
above, the granularity of the suspension is coarse-grained, and the ferrofluid is
treated as a magnetizable continuum build up of two homogeneous constituents.
The variables are the conserved quantities, the electromagnetic field, and the
magnetization as the only one being non-hydrodynamic. The concentration field
is taken into account by the mass density of the ferromagnetic material ρc .
This is appropriate since the magnetophoresis and Soret effect, which are fairly
pronounced in ferrofluids, may build up perceptible concentration gradients.
The above arguments imply that the thermodynamic energy density u is a
function of the entropy density s, total density ρ, concentration ρc , magnetic
field B, magnetization M, and the momentum density g = ρv,
du = T ds + μdρ + μc dρc + v·dg + H·dB + h·dM. (2)
Eq. (2) is to be understood as the definition for the conjugate variables such as
temperature T , chemical and relative chemical potentials μ and μc , velocity field
v etc. . In particular, the quantity h is associated to the magnetization. With
M ≡ B − H, or ∂Hi /∂Mj = −δij for given B, together with the thermodynamic
Maxwell relation, ∂Hi /∂Mj = ∂hj /∂Bi , we obtain
h = Beq (M, s, ρc , ρ) − B = Heq − H. (3)
Eq.(3) results from the requirement that u has to be minimal with respect to
M at M = Meq (B), or equivalently h ≡ ∂u/∂M = 0. So Beq (M) is the inverse
function of the equilibrium magnetization curve Meq (B). Subtracting M from
both Beq and B, we may also write h = Heq − H, where again Heq (M) is the
inverse function of Meq (H). Note that the function Heq (M) is frequently referred
to in the ferrofluid literature as the ”effective field”.
The conserved variables satisfy continuity equations,
ρ̇ + ∇ · (ρv) = 0, ρ̇c + ∇ · (ρc v − jD ) = 0, (4)

u̇ + ∇ · Q = 0, ġi + ∇j (Πij − Πij
D
) = 0; (5)
the equations of motion for s and M are
ṡ + ∇ · (sv − f D ) = R/T, (6)

Ṁ + (v · ∇)M + M × Ω = X , D
(7)
where R is the entropy production and Ω = ∇ × v/2 the vorticity.

Assuming that no external electric field is applied, the appearance of an
electric field is due solely to electromagnetic induction. Taking the ferrofluid to
be dielectrically neutral (i.e. D = E) the electric contributions to the equations
of motion are smaller by a factor (v/c)2 than their magnetic counterparts (c is
the speed of light and v a typical velocity). Accordingly, we shall set it to zero.
(See [10,12] for the cases where an external electric field is applied.) As a result,
we may use the Maxwell equations in the static approximation
∇ · B = 0, ∇ × H = 0. (8)
The fluxes in Eqs (4-7) still need to be derived – although for some of them
their convective contributions such as ρc v or (v · ∇)M +M × Ω have already
been made explicit. To derive the unknown flux contributions we employ the
so-called standard procedure of hydrodynamics: Take the temporal derivative of
Eq (2), substitute u̇, T ṡ, μρ̇ . . . using the above equations of motion, and most
importantly, require that the resultant equation to hold identically (cf [10,13]
and references therein). This yields the energy flux Q, the momentum flux Πij ,
and the entropy production R as
Πij = Πji = [A + Hk Bk − u]δij + gi vj − Hi Bj + 12 (hj Mi − hi Mj ), (9)

Qi = Avi − T fiD − μc jiD − D
vj Πji + 2 [v × (h
1
× M)]i (10)
R = f D · ∇T + j · ∇μc − X · h +
D D
ΠijD
vij , (11)
where A ≡ T s + μρ + μc ρc + v · g, vij ≡ 12 (∇i vj + ∇j vi ). To make the set of

equations closed and complete one still has to determine the dissipative fluxes f D ,
jD , XD , Πij
D
. The form of the entropy production R as given in Eq. (11) implies
that they are linear combinations of the forces ∇T , ∇μc , −h, vij
0
, vkk , such that
R is always positive. (We take vij ≡ vij − 3 vkk δij .) What now follows is the
0 1
construction of the fluxes on the basis of symmetry considerations and specific

assumptions, the second of which are subject to experimental verifications or
microscopic scrutiny.
3.1 Weak Field Limit

If the applied magnetic field is weak, the system can be considered to be ap-
proximately isotropic. In this case we have the usual diagonal relations for the
diffusive entropy and concentration currents, viscous stresses and especially the
magnetic relaxation [14,2],
f D = κ∇T + ξ1 ∇μc , jD = ξ∇μc + ξ1 ∇T, (12)

ΠijD 0
= 2η1 vij + η2 vkk δij , XD = −ζh, (13)
The transport coefficients κ, ξ, κξ − ξ12 , η1 , η2 and ζ are positive functions of

thermodynamic variables. In particular they also depend on the magnitude of the
magnetization M . As discussed in the introductory section, their actual values
need to be determined either experimentally or on the basis of an appropriate
microscopic model. This is beyond the scope of the present investigation.
In compliance with results given by Blums et al. [2] and a work by Felderhof
and Groh [4], we observe that the magneto-relaxation term XD is proportional
to h = Heq − H. This is in contrast to the Debye-like increment δM = M − Meq
suggested by Shliomis [3]. We therefore denote XD according to Eq. (13) as the
rectified relaxation term. Although δM and h are in general not linearly related
they do in two special cases: (i) For small deviations from local equilibrium,
where δM and Heq − H are proportional to each other and (ii) if the applied
field is sufficiently weak for the linear constitutive relation Meq = χH to apply.
We now turn to compare our stress tensor Πij with the traditional formula-
tion. We start with the observation that the last term in (9) is equivalent to the
magneto-dissipative element as given in Eq.(1). This term accounts for magneto-
dissipation when the vectors H and M are twisted relative to each other. This
happens for instance in the McTague experiment [6], where the vorticity of the
flow deflects the magnetization vector out of the equilibrium direction.
We emphasize that the condition for the appearance of magneto-dissipative
effects, M = Meq , does not necessarily require the two vectors H and M to
point in different directions. For instance, magneto-dissipation is also expected
to occur if M and H oscillate parallel to each other but with a temporal phase
lag. Note that in this situation Eq.(1) is inoperative, so the Shliomis theory does
not yield a magneto-dissipative stress here. However, as will become clear in
a moment, FFD does. To facilitate further comparison with the conventional
notation of the stress tensor we rewrite the diagonal contribution to [Eq. (9)]
in terms of the zero-field pressure p0 (ρ, T ). To that end we perform a Legendre
transformation to the independent variables T , ρ, v, H, and M (for simplicity
we neglect the concentration dynamics by assuming ρc = const). The associated
thermodynamic free energy f = u − sT − v · g − H · B has the total differential
df = −sdT + μdρ − g · dv − B · dH + h · dM. (14)
Integrating out velocity and magnetic dependencies leads to

M
1 2 1 2
f (T, ρ, v, H, M) = f0 (ρ, T )− ρv − H −M·H+ H eq (T, ρ, M )dM , (15)
2 2 0
where f0 (T, ρ) = f (T, ρ, v = H = M = 0). Using the thermodynamic relation

for the zero-field pressure p0 (T, ρ) = −f0 + ρ∂f0 /∂ρ and the integral identity
M H eq

H eq
·M− (1 − ρ∂ρ )H (T, ρ, M )dM =
eq
(1 − ρ∂ρ )M eq (T, ρ, H )dH
0 0
(16)
the diagonal contribution to the stress [square bracket of Eq (9)] transforms into
H eq
δij p0 + 12 H 2 + 0 (1 − ρ∂ρ )M eq (H )dH − h · M . (17)
The last term h · M is missing from previous works. As outlined above it accounts
for magneto-dissipation in a situation where the off-equilibrium increment h is
parallel to the magnetization. For an illustration of what are the physical signif-
icances of the two different magneto-dissipative stresses, (1/2)εijk (H × M)k ≡
(1/2)εijk (M × h)k and δij (h · M) let us consider the force exerted by a ferrofluid
on a rigid container wall. Multiplying the antisymmetric element with the sur-
face normal vector leads to a tangential traction proportional to H × M. On the
other hand, the diagonal tensor element implies a magneto-dissipative normal

force linear in (h · M). This term may be probed by a measurement of the pres-
sure drop across an interface between a ferrofluid and a non-magnetic medium.
Since it is a dynamical effect, the interface must be exposed to a time-dependent
magnetic field. The expected effect is maximized when the oscillation frequency
ω approaches the inverse magnetic relaxation time 1/τ .
The reason magneto-dissipative normal stresses have not been discussed until
now, may be due to the present focus on incompressible flow configurations. For
that kind of flow problems any extra normal stress simply re-normalizes the
pressure while leaving the velocity profile unchanged. If, however, normal forces
are directly recorded, the associated magneto-dissipative contribution is likely
to be measurable. An alternative way to probe the implications of the extra
normal stress is the study of compressible flow configurations such as sound. We
expect that the coupling between density oscillations and the magnetization will
contribute appreciably to the attenuation.
3.2 Strong Magnetic Fields

In the weak field limit discussed in the previous subsection we have seen that
the evolution equation for M as well as the stress tensor Πij deviate from the
expressions given by the previous approach. In the presence of a strong mag-
netic field, the system can no longer be considered to be isotropic. So, further
modifications arise since the resulting uniaxial symmetry of the system leads to
a proliferation of the transport coefficients. This is where experimental input
becomes imperative. For instance, each of the coefficients of Eq (12) turns into
three, as in κ → κδij +κ Mi Mj +κ× ijk Mk . Similarly, the two viscosities turn
into 7. It is not very useful to present all these complications here because the
set of isotropic coefficients needs yet to be measured. Strictly speaking, if the
directions of M and H do not coincide (as is usually the if magneto-dissipative
effects come into effect), the system becomes biaxial. This leads to an extra
complication, which however will be ignored here.
In the following we focus on the complete uniaxial form of XD , since the
magnetization dynamics belongs to the best studied aspects of ferrofluid physics.
Following the linear construction scheme for deriving the contributions to the
dissipative flux XD leads us to
XiD = −(ζδij + ζ Mi Mj + ζ× ijk Mk )hj
0 0
+λ1 Mi vkk + λ2 Mj vij + λ3 Mi Mj Mk vjk
0
+λ4 ikj Mk M vj . (18)
Then the Onsager symmetry relations enforce the following counter terms in the
stress tensor
D
Πij = {η2 vkk + [λ1 − 13 (λ2 + M 2 λ3 )]Mk hk }δij
0
+2η1 vij + 12 λ2 (Mi hj + Mj hi ) + λ3 Mi Mj Mk hk
+ 12 λ4 [Mj (M × h)i + (i ↔ j)]. (19)
Although these expressions appear complicated, one must realize that the uniax-
ial, and not the isotropic, case is the generic one: If we take M as small to arrive
at the isotropic case, we must for consistency also neglect all term ∼ M 2 in the
Maxwell stress, which is considered too crude an approximation to be employed
frequently.
The appearance of the parameters ζ, ζ and ζ× in XiD implies different relax-
ation times for the respective magnetization components parallel to h, parallel
to M and perpendicular to both of these directions. The latter one is analogous
to the Righi-Leduc effect in heat conduction or the Hall effect of an electrical
conductor.
In the Debye theory, the EFT, or the isotropic case considered in the previous
subsection, the only velocity gradient entering Ṁ is the flow vorticity Ω. This is
in contrast to Eq. (18), which suggests that a compressional flow vkk , and even
0
more importantly, an elongational flow vij will do this too. The coefficients λi
are material dependent and need to be measured for each ferrofluid. They are
reactive transport coefficients (as opposed to dissipative ones), because they do
not enter the expression for the entropy production (11). Nevertheless, as these
coefficients appear as a product either with velocity gradients [Eq. (18)] or with
the magneto-dissipative increment h [Eq. (19)], they can only be evaluated by
an appropriate off-equilibrium experiment. It is noteworthy that a term similar
0
to vij Mj is known to appear in the dynamics of nematic liquid crystals [15],
where it gives rise to the “flow alignment” of the nematic director in response
to a shear flow.
For a quantitative experimental test of the whether λ2 for a given ferrofluid
material is of significant size one has to study the magnetization dynamics in flow
geometries which allow to control the relative strength between elongational and
rotational flow contributions. This requirement is easy to fulfill by the flow be-
tween two rotating cylinders (Couette-Taylor setup). Allowing the two cylinders
to rotate independently with distinct angular frequencies Ω1 and Ω2 , enables
to pass over from a rigid rotation (vij = 0), where Ω1 = Ω2 , to a flow with a
finite shear rate, where vij ∝ (Ω1 − Ω2 ) = 0. Recording the magnetization vector
within the sample (or equivalently the magnetic field outside of the sample) as
a function of Ω1 − Ω2 allows to evaluate the transport coefficient λ2 .
4 Comparison with Existing Theories
In the introductory section we outlined that the macroscopic ferrofluid dynam-

ics set up here provides the general macroscopic framework, any more specific
(microscopic) approach has to fit in. In the present section the focus is on the
relaxation equation for the magnetization. The modifications/extensions of the
stress tensor as compared to previous theories have already been discussed in
the last paragraph.
It will be shown here that both the Debye theory and EFT can be embedded
into the formulae given by FFD, each with a specific choice of parameters.
4.1 The Debye Theory

In the first variant of his description, Shliomis [3] introduces a phenomenological
equation for M, with a linear Debye-like relaxation term proportional to δM =
M − Meq . This equation results from what we consider to be a “mesoscopic”
approach, because the derivation makes use of some specific microscopic input.
The ferromagnetic grains are considered to be Brownian rigid dipoles. To account
for their rotation the angular momentum density S is introduced as an extra
dynamical variable. But not really, since in a later stage of the derivation, S is re-
eliminated owing to the smallness of the particles’ inertia. The phenomenological
relaxation equation for M which remains after this adiabatic substitution takes
on the following form
dM 1 M × (M × H)
−Ω×M= (Meq − M) − . (20)
dt τB 6η1 ϕ
Here (d/dt) ≡ ∂t + (v · ∇), τB is the Brownian relaxation time, and ϕ the
volume concentration. As pointed out above and by other authors [2,4], the
proper relaxation term being rather more in line with the construction rules of
non-equilibrium thermodynamics is proportional to h rather than δM. This is
what we have referred to as the rectified relaxation term. In the limit of small
deviations from the magnetization equilibrium, we have a linear relationship
between δM and h

M eq ∂M eq M eq M·h
M − Meq = h+ − M + O(h2 ). (21)
H ∂H H (M eq )2
Replacing the leading order approximation of (21) into Eqs (7,18), we recover
Shliomis’ Debye theory (20) by the following choice of the transport coefficients
λ1 = λ2 = λ3 = λ4 = ζ× = 0, (22)
1 M eq (M eq )2
ζ= + , (23)
τB H 6η1 ϕ

1 ∂M eq M eq (M eq )2
ζ (M eq )2 = − − . (24)
τB ∂H H 6η1 ϕ
4.2 Effective-Field Theory

On the basis of a kinetic equation for rotary diffusion, Martsenyuk et al. [16]
constructed the Fokker-Planck equation for the probability distribution of the
particle’s orientation. The authors relied on the idealizing assumptions that the
ferrofluid is composed of (i) spherical (ii) mono-dispersed (iii) non-interacting
(iv) rigid dipoles. Form the resultant infinite hierarchy of equations for the mo-
menta of M, a separate equation for the magnetization is deduced by employing
the method of the effective field. Thereby the magnetization
ξe
M = Ms L(ξe ) (25)
ξe
is taken to be a function of the dimensionless, effective field ξe = (mHeq )/(kB T ),

with Ms denoting the saturation magnetization of the ferrofluid, m the magnetic
moment of an individual particle, L(x) = coth x−1/x the Langevin function and
kB the Boltzmann constant. In terms of the actual non-dimensional magnetic
field ξ = (mH)/(kB T ), the effective field is governed by the ordinary differential
equation

d ξe ξe 1 Le
Le = Ω × Le − (ξe − ξ) − (26)
dt ξe ξe τB ξe

1 3Le
1 − ξe × (ξe × ξ),
2τB ξe2 ξe
where Le = L(ξe ). This equation has to be solved for ξe for given ξ. Then in a
second step, the magnetization is to be deduced from ξe via (25).
We point out that this somewhat unwieldy two-step procedure can be circum-
vented. Without approximation Eq. (26) can be recast in the following, rather
more explicit form

d
2τB M−Ω×M = (27)
dt

M M M·h
− 3χ − eq h − 3 eq
−χ M.
H H M2
Here χ = mMs /(3kB T ) denotes the initial Langevin susceptibility. It is note-

worthy to point out that the EFT approach yields a relaxation term, which is
proportional to h rather than δM, i.e. well in compliance with the rectified Debye
theory outlined above. So it does not come as a surprise that Eq. (27) is recov-
ered as a special case of the general structure of FFD as given by Eqs (7,18).
The following particular parameter choice applies
λ1 = λ2 = λ3 = λ4 = ζ× = 0, (28)

1 M
ζ= 3χ − eq , (29)
2τB H

3 1 1 M
ζ = − χ . (30)
2 τB M 2 H eq
Clearly, owing to its microscopic input the EFT provides a dependence for the
transport coefficients ζ and ζ on the strength of the magnetization, a feature
which cannot be accomplished by the macroscopic FFD. That is why EFT is a
complementary approach rather than being competitive. On the other hand, the
disappearance of many of the transport coefficients [see Eq. (28)], an immedi-
ate consequence of the idealizing assumptions, reflects the limitations of EFT.
Real ferrofluids generally do not meet these approximations. So one either has
to develop a more elaborate theory, or – following the standard approach in
macroscopic physics – to determine the set of transport coefficients by a series
of appropriate experiments.
5 Conclusion
The general structure of the hydrodynamic equations for ferrofluids is derived
here. The gain in rigour is paid by a loss of specific information on the transport
parameters. We did not provide the numerical values of the transport coefficients
here, nor their dependence on the thermodynamic variables. This task is left
open for a series of experiments. An alternative way – albeit less reliable and
less quantitative – is to derive this information from a microscopic theory such
as EFT. Besides giving numerical estimates for the transport coefficients their
primary advantage is to provide physical insight into their dependencies on the
thermodynamic variables.
In spite of the complete lack of microscopic specifics in the present derivation,
the resultant theory does have some restrictions that we need to keep in mind.
They arise due to the assumption that a unique characteristic time τ is sufficient
to characterize the magnetic relaxation process. As a result, any microscopic
features (such as poly-dispersity) that influence this time are to be handled with
some care. For instance, a ferrofluid consisting of two populations, each with
a distinct relaxation time, will not be well accounted for at higher frequencies,
outside the hydrodynamic regime.
References
*. [email protected], [email protected]
1. R.E. Rosensweig, Ferrohydrodynamics, (Cambridge University Press, Cambridge,
1985).
2. E. Blums, A. Cebers, M.M. Maiorov, Magnetic Fluids, (Walter de Gruyter, Berlin
1997).
3. M. I. Shliomis, Sov. Phys. JETP, 34, 6 1291 (1972); Usp. Fiz. Nauk 112, 427
(1974) [Sov. Phys. Usp. 17, 153 (1974)]; M.I. Shliomis, J. Mag. Mag. Mat. 159 236
(1996).
4. U. Felderhof, B. Kroh, J. Chem. Phys. 110, 7403 (1999).
5. H. W. Müller, M. Liu, Phys. Rev. E 64, 061405 (2001).
6. J. P. McTague, J. Chem. Phys. 51, 133 (1969).
7. M.I. Shliomis and K. I. Morozov, Phys. Fluids 6,2855 (1994).
8. J.-C. Bacri, R. Perzynski, M.I. Shliomis, G.I. Burde, Phys. Rev. Lett. 75, 2128
(1995).
9. A. Zeuner, R. Richter, I. rehberg, Phys. Rev. E 58, 6287 (1998).
10. Mario Liu, Phys. Rev. Lett. 70, 3580 (1993); 74, 4535 (1995); 80, 2937, (1998).
11. S. Lissek, H.W. Müller, M. Liu, Hydrodynamic Maxwell Theory for Ferrofluids,
submitted.
12. Y.M. Jiang and M. Liu, Phys. Rev. Lett. 77, 1043 (1996); Phys. Rev. E, 6685
(1998).
13. K. Henjes and M. Liu, Ann. Phys. 223, 243 (1992);
M. Liu and K. Stierstadt, cond-mat/0010261.
14. L.D. Landau, E.M. Lifshitz, Fluid Mechanics (Pergamon, Oxford, 1987).
15. P. G. de Gennes, J. Prost, The Theory of liquid crystals Clarendon, Oxford (1983).
16. M.A. Martsenyuk, Y.L. Raikher, and M.I. Shliomis, Sov. Phys. JETP 38, 413
(1974); Y. L. Raikher, M. I. Shliomis, in Relaxation Phenomena in Condensed
Matter, ed. by W. Coffey, Advances in Chemical Physics Series 87, Wiley (1994).
Heat and Mass Transfer Phenomena
Elmars Blums
Institute of Physics, University of Latvia, LV-2169, Salaspils, Latvia
Abstract. This section deals with main problems of the heat and mass transfer in
magnetic colloids. The analysis is mainly based on the general model given in the
Chapter written by R. E. Rosensweig. Hydrodynamic and thermal problems are sim-
plified considering incompressible liquids and neglecting the effects of polarization and
electric conductivity as well as ignoring some other secondary effects that usually can
be neglected in ferrofluid experiments. Contrarily, the analysis of mass transfer ac-
counts for new sedimentation phenomena and cross effects of interrelated heat and
mass transfer. Since the description given by Rosensweig is of general theoretical na-
ture, while the present work mainly focusses on experimental problems, the various
equations needed will not be cited from Rosensweig’s article but from the original lit-
erature since the transfer from the general model to the experimentally needed relations
is often cumbersome and would exceed the frame of this presentation.
The following problems shortly are discussed: the equation of energy conservation
accounting for new adiabatic effects and dissipation due to internal rotation, the ther-
momagnetic and magnetosolutal instabilities, the specifics of magnetic convection in
fluids of non-uniform temperature and concentration, the magnetophoresis and ther-
modiffusion of colloidal particles under the effect of a magnetic field and some phe-
nomena of combined thermal and Soret-driven convection.
1 Introduction
In main applications the magnetic fluids mostly are used to position the colloid at
a certain part of devices by means of magnetic forces. Therefore, for a long time
the most important research problems are the increase of fluid magnetization,
the specific properties of free ferrofluid surfaces in the presence of an external
field and the magnetoviscous and magnetorheological effects. Presently the re-
search topics undergo serious changes. Particularly, there start to be popular
non-potential bulk forces in fluids of a nonuniform magnetization. Possibility to
induce an intensive convection by means of a magnetic field opens the promising
design of new challenging applications (magnetically controlled thermosyphons
for technological purposes, augment of heat transfer for cooling of high power
electric transformers etc.). Besides, the temperature or the concentration gra-
dients cause new thermo-and magnetophoretic transport processes, which can
influence the long-time stability of ferrocolloids in many technical devices. In
the present paper main problems of heat and mass transfer in the presence of a
magnetic field shortly are discussed.

Heat and Mass Transfer Phenomena 125
2 Energy Conversion
Many authors have analyzed the energy conservation equation in magnetizable
media. From the second law of thermodynamics for the quasi-equilibrium model
of ferrofluids it becomes [1]

dT 1 ∂ V̄ dp ∂M dH d ∂
ρcp =T −T μ0 + Q − divjq = +v·∇ (1)
dt V̄ ∂T dt ∂T dt dt ∂t
The left part of equation (1) characterizes the accumulated thermal energy
of the fluid (ρ is the fluid density and cp is the specific heat capacity). The first
two terms of the right side reflect the adiabatic effects of fluid compression and
magnetization (V̄ and M are the specific volume and the magnetization), the
Q represents the internal heat source density, and the last term describes the
conductive heat transfer. According to Fik’s law the heat flux jq is proportional
to the temperature gradient:
jq = −λ∇T (2)
(λ is the fluid thermal conductivity). The internal gradients of fluid temperature
are created by the conductive heat input or output through the volume borders
as well as by the internal heat source Q.
Due to a low compressibility of liquids the adiabatic rise of their temperature
under the effect of an increasing pressure always is very small [1]. The thermal
effect of fluid magnetization at room temperatures also is not strong. Only close
to the Curie temperature, the adiabatic magnetization can cause a considerable
change in the fluid temperature. This phenomenon at early stage of the mag-
netic fluid research was proposed to use for a thermomagnetic energy conversion
[2]. The term Q in (1) represents the absorption of external radiation and the
heat generation due to a viscous dissipation of the fluid kinetic energy. Magnetic
colloids are not transparent to the radiation of optical and infrared frequencies;
therefore, the internal effects of light absorption should be taken into account
only in the event of thin fluid layers. If macroscopic volumes of liquids are con-
sidered, only the viscous dissipation of flow energy in Q usually should be taken
into account. Apart from the conventional viscous term, the dissipation func-
tion for magnetic fluids contains an additional term, which reflects the effect of
internal rotations. For Langevin-type magnetization M = ϕMs L(ξ) of colloids
(ξ = μ0 mH/kT , ϕ is the volume fraction of magnetic phase of a bulk saturation
magnetization Ms ), if they contain magnetically hard particles of a magnetic
moment m, the heat source Q becomes [1]
2 2
η ∂vi ∂vk ξL(ξ) H
Q= + + 6ϕη Ω −
2
·Ω (3)
2 ∂xk ∂xi 2 + ξL(ξ) H
(η is the fluid dynamic viscosity). The thermal effect of internal rotations can be
observed in the presence of a constant magnetic field only if it is oriented non-
parallel to the flow vorticity Ω = (1/2)rotv. At low flow velocities the heating
intensity usually causes a relatively small effect upon the fluid temperature.
126 E. Blums
Only for high shear rates typical for some ferrofluid applications (for example,
in high-speed rotary seals and in loudspeakers) the viscous heating of the fluid
can cause a significant rise of its temperature.
Fig. 1. Specific heat induced in magnetic fluids by a rotating field [3]. Left side: mag-
netite particles in kerosene, frequency f = 1.75 kHz, particle volume concentration
ϕ = 0.096 (1) and ϕ = 0.081 (2). Right side: cobalt ferrite in kerosene, ϕ = 0.086,
frequency f = 1.75 kHz (1), f = 0.6 kHz (2), f = 0.2 kHz (3)
Heating of the colloid due to internal rotations in the most pure form man-
ifests in a rotating magnetic field. Despite the presence of anti-symmetric tan-
gential stress, a liquid may remain quiescent, and only the second term in (3)
is responsible for a rise of the fluid temperature. Figure 1 represents the heat-
ing intensity, which is measured employing magnetite and cobalt ferrite based
ferrofluids [3]. From the presented results we see that in colloids of magnetically
“hard” particles, the field-induced internal rotation causes a significant thermal
effect. For example, in a cobalt ferrite based ferrofluid the rotating field 40 kA/m
of frequency ω = 1.75 kHz generates a heating density Q up to 600 W/kg [3]. In
principle, the microconvection under the effect of field rotation induces also some
changes in the fluid thermal conductivity. Microvortices alter the magnitude of
heat flux along the temperature gradient and cause a transversal heat flux in the
plane of rotation axis. Still, the numerical estimates show [4] that this magnetic
analogue of Rigi-Leduc effect in colloids usually is negligible.
A detailed thermodynamic analysis of the energy conservation equation re-
quires considering all thermodynamic parameters as functions of a magnetic
field. However, from a practical viewpoint the corresponding corrections, except
the situation when temperature is close to the Curie one, are negligibly small. A
more important problem is the dependence of physical properties and transfer
coefficients on the particle concentration and on the structure of colloids. The
additivity law applies for the fluid density and the specific heat capacity; there
is no observable effect of a magnetic field upon these. The transfer coefficients
are affected by a magnetic field due to several reasons. It is well known that
the magnetic braking of internal rotations causes a fluid magnetoviscosity. In-
ternal rotations exert the Hall -type influence upon the heat and mass transfer
coefficients as well. Still, the microconvective analog of the Rigi-Leduc effect of
thermal conductivity is small [1]. More significant seem to be the magnetic field
induced effects which are related to the aggregate forming and orientation.
Fig. 2. Thermal conductivity of a diluted suspension of ellipsoidal particles (magnetite

in kerosene, ϕ=0.1 ) under the effect of a magnetic field [5]. Lines marked with (n)
represent the transient heat transfer with B⊥∇T, lines with (t) correspond to B// ∇T.
Parameter k is the relative elongation of particles, ξ is the Langevin parameter
Figure 2 shows the transient thermal conductivity of a diluted suspension

of ellipsoidal particles calculated assuming the one-particle approximation [5].
In a qualitative sense this model represents the heat transfer through a colloid
under the effect of magnetically oriented aggregates. The thermal conductivity
of ferrite particles significantly exceeds that of a carrier liquid. Therefore, the
conductive heat transfer through assemble of the non-spherical particles is higher
than that of the spherical ones. The orientation of ellipsoids along the field
direction during the colloid magnetization leads to appearance of an anisotropy
of the transient heat transfer. From results presented in Fig. 2 we see that the
magnetic field B, which is oriented parallel to the temperature gradient ∇T ,
causes an increase in the thermal conductivity λ, whereas at B ⊥ ∇T a reduction
of λ should be observed. The one-particle approximation is valid for diluted
colloids until the particle volume concentration does not exceed approximately
the value ϕ = 0.05. The thermal conductivity λ of dense colloids of spherical
particles can be calculated using the classical dependence [6]:
λ 2λ0 + λp − 2ϕ(λ0 − λp )
= (4)
λ0 2λ0 + λp + ϕ(λ0 − λp )
128 E. Blums
Here λ0 and λp are the thermal conductivity’s of the carrier liquid and of parti-
cles.
The dependence of conductive heat transfer on a magnetic field in experi-
ments employing stable ferrocolloids usually is not observed, but it is informative
to note that most of the known experimental data suggest a more pronounced
rise in thermal conductivity predicted by the relation (4). Obviously, this is due
to the peculiarities of surfactant layers on particles. The presence in a colloid
of various additional ingredients plays a more sufficient rule when the electrical
conductivity of the colloid is considered.
3 Thermomagnetic Convection
To analyze the convective processes, the energy conservation equation should
be considered together with the equation of a fluid motion. Let us consider the
problem in a Bussines approximation, i.e. assuming the physical properties and
the transfer coefficients (with exception of the density in the gravitation force
ρg and of the magnetization) constant. The equation of motion contains a new
term of magnetostatic force:
dv
ρ = −∇p + ηΔv + ρg + μ0 (M∇)H (5)
dt
If the magnetic susceptibility χ is dependent solely upon the field and ρ =
const, then under an approximation of equilibrium magnetization M = χH
both the gravitation and the magnetic forces are potential and there cannot
arise internal convective flows in the fluid. The non-potentiality of bulk forces
appears only if a fluid possesses the spatial non-uniformity of ρ and M due their
dependence on temperature or on particle concentration ϕ. From (5) it follows
that the free convection develops spontaneously if
∇ρ × g + μ0 ∇χ × ∇H 2 = 0 (6)
We can see that there exists a certain analogy between the thermogravita-
tional and the thermomagnetic convection. The difference is only the spatial
non-uniformity of magnetic acceleration ∇H 2 while the gravitation acceleration
g is constant. There are many works published in which the magnetic convection
under various field distributions are calculated and measured (see the review in
Ref. [1]).
If (6) equals to zero, a problem of thermoconvective instability appears. The
stability criterion is the Rayleigh number Ra. The equations (1) and (5), if they
are linearized with respect to small perturbations of velocity, temperature and
pressure, have the same form as those describing the conventional convection
problems. The only difference lies in appearance of a different form of Ra (we
assume the coordinate x being directed toward the vector g) [7]:
4
dH dT T dH cp l
Ra = βT ρg + μ0 αT M − βT gρ + μ0 αT M (7)
dz dz ρcp dz νλ
Here βT and αT are the expansion and the pyromagnetic coefficients of the fluid
(according to the definition they are positive), l is the characteristic length.
From (7) it follows that, without applying external temperature gradients,
the fluid with respect to the adiabatic compression and magnetization is always
stable (note, if dT /dz = 0, the Rayleigh number Ra is negative ). The tem-
perature gradients necessary to attain the critical value of Ra are so large that
the adiabatic terms (the second part in the square brackets in (7)) usually can
be neglected. In such approximation the Rayleigh number contains two addi-
tive terms, they reflect the thermogravitational (RaT ) and the thermomagnetic
(RmT ) buoyancy forces:

dH dT cp l4
Ra = RaT + RmT = βT ρg + μ0 αT M (8)
dz dz νλ
Convective instability can develop (Ra > 0) if the temperature gradient is

directed along the gravitation or the magnetic force.
An interesting situation appears when the nonisothermic ferrofluid layer is
subjected to a homogeneous magnetic field B = const. From the field equation
divH = −divM it follows that thanks to the pyromagnetic properties of the
fluid, a gradient of internal magnetic field in the layer appears. Instead of (8)
the Ra now is the following:

α2 M 2 dT dT cp l4
Ra = RaT + RmT = βT ρg + μ0 T (9)
(1 + χ) dz dz νλ
(Here χ = ∂M/∂H is the differential magnetic susceptibility.) As a result, the

magnetic Rayleigh number RmT becomes square dependent on the temperature
gradient and it is positive. It means that thermal perturbations of the magne-
tizing field always lead to the thermoconvective destabilization of the fluid.
The critical conditions for onset of the convection in a flat horizontal layer
can be represented by a relation [8]
Ra∗ Rm∗
= (10)
R0 Rm0
Here R0 = 1708 (flat layer of rigid boundaries), whereas Rm0 is the critical
Rm value when the gravitation is absent. The later depends on the magnetic
properties of channel walls and is slightly influenced also by the nonlinearity of
the fluid magnetization curve.
The general conclusions of thermoconvective stability theories (the pioneer-
ing one, obviously, is that published in Ref. [9]) are confirmed experimentally.
Figure 3 represents the results [10] of measurements of the heat transfer through
a horizontal layer of ferrofluid in the presence of a transversal magnetic field
Bz = const. If B = 0, the convection appears (the Nusselt number N u starts
to exceed the value 1) when the layer is heated from below and the Rayleigh
number reaches its critical value R0 . With a magnetic field applied the critical
temperature gradient drops. Besides this, in accordance with (9) the convective
130 E. Blums
Fig. 3. Effect of a transversal magnetic field upon the heat transfer in infinite flat layer
of magnetic fluid: 1 – heated from below, H = 0.2 – heated from above, H = 40.8 kA/m,
3 – heated from below, H = 40.8 kA/m [10]
Fig. 4. Relation between Ra c and Rm in a horizontal magnetic fluid layer of thickness

d [10]. Dots represent the experimental results, solid curve – the results of theoretic
calculations for the cell examined in experiments
instability sets on also when the layer is heated from above. Figure 4 represents
the experimental values of Rac measured in the presence of a magnetic field.
These results are in a good agreement with the theoretical dependence (10) if
accounting for the critical values of both the R0 and Rm0 . The Rm0 is higher
than R0 , (compare the points and the corresponding theoretical dependence in
Fig. 4 with the curve Rac + Rmc = 1708), it depends on magnetic properties
of the channel walls [8,9]. The reason of that is a difference in mechanisms of
the gravitational and the magnetic convective instabilities: the thermomagnetic
instability is affected not only by perturbations of the fluid temperature but also
by those of the magnetic field inside the fluid.
If the external magnetic field has a gradient non-parallel to the ∇T , inside the
fluid sets on a non-threshold convection. Its intensity significantly exceeds the
intensity of the thermogravitational convection and a significant enhancement
of the heat transfer can be achieved. It means that magnetic control of the heat
transfer in ferrofluids is an interesting problem of applications.
4 Mass Transfer
Considering stable colloids without chemical reactions, the mass conservation
equation for two component systems (particles of a mass concentration ρi and a
carried liquid) is the following:
dρi
= −divji (11)
dt
Ultra-fine particles in magnetic fluids obey an intensive Brownian motion.
Therefore, the mass transfer in colloids can be considered similar to that in
molecular liquids involving the concept of a gradient diffusion. The mass diffusion
coefficient of nanoparticles is determined by the relation D = kT /fv , where fv is
the coefficient of the hydrodynamic drag force (for spherical particles of radius
r the fv = 6πρνr). The mass flux [1]
mg
ji = −D∇ρi + −(V̄i − V¯0 )ρg + μ0 (M̄i − M̄0 )∇H ρi (1−ni )−ρi (1−ni )DST ∇T
fν
(12)
contains a new barodifusion term. In addition to the conventional gravitational
sedimentation of particles (mg is the mass of the particle,V̄i and V¯0 are the
specific volume of the solid phase and that of the carrier liquid) it is necessary
to take into account also for their magnetic sedimentation ( M̄i = Mi /ρi and
M̄0 = M0 /ρ0 are correspondingly the specific magnetization of the particles and
the carrier). The mass diffusion coefficient of colloidal particles is several orders
of magnitude less than that of molecules in liquids. Therefore, the gravitation
and magnetic sedimentation processes in ferrofluids play a more significant rule
than those in molecular liquids. The thermodiffusion term in (12) is retained for
purpose of generality, ST there is the Soret coefficient of particles, ni = ρi /ρ is
the mass fraction of solid phase.
The conditions of the fluid convective instability in isothermal liquids also
are determined by the relation (6). The corresponding solute Rayleigh number
now is the following:

dH dρi l4
Ra = Rac + Rmc = − βc ρg + μ0 αc M (13)
dz dz ηD
(According to the definition, the solute expansion (βc ) and magnetic (αc ) co-
efficients are positive). It is important to note that there is a principal differ-
ence between the thermal and the solutal convective stability. For stable colloids
132 E. Blums
the concentration boundary conditions are the unpermeability of surfaces. It

means that the concentration gradients cannot be applied independently; they
are formed internally by the sedimentation forces near walls in a form of diffusive
boundary layers. From (12) and (13) it follows that the stratification of particles
in colloids under the action of gravitation and/or external magnetic field gradi-
ents always causes an increase in the fluid convective stability. The influence of
internal magnetic field gradient is quite different. If the fluid layer is placed in
a homogeneous magnetic field B = const, the only reason of the fluid stratifica-
tion is the gravitation force. The corresponding field gradient ∇H = −∇M is
oriented opposite to the ∇ρi . The magnetic term in (13) now becomes a square
dependent on the concentration gradient and Rmc is positive:
2
dρi μ0 l4
Rmc = αc M · (14)
dz (1 + χ)ηD
That means that internal field gradients always cause a lowering of the convective
stability of fluid layers.
Fig. 5. Paramagnetic (positive t) and diamagnetic (negative t) separation of parti-

cles in the vicinity of a cylindrical current-carrying electric conductor of radius R 0 [1].
Non-dimensional time t contains the characteristic magnetophoretic time R 0 /u 0 with
u 0 being the velocity of particle phoresis at the cylinder surface
From Maxwell equations it follows that the external magnetic field gradient
cannot be constant, the magnetic force always is spatially non-homogeneous. As
a result, the magnetic sedimentation forms a non-uniform distribution of parti-
cles not only near walls but also outside the boundary layers. As an example,
in Fig. 5 are shown profiles of concentration in the vicinity of a cylinder when
the particles are transferred by magnetophoresis under the effect of a radial
gradient of the azimuthal magnetic field generated by axial electric current I
(H = 2πI/r) [1]. In the case of a paramagnetic separation (magnetic particles of

the colloid, t > 0) owing to a sharp acceleration of particles toward the cylinder,
the wall region gradually depletes of a magnetic phase. In case of a diamagnetic
separation (nonmagnetic particles in ferrofluid, t < 0) the withdrawal of a con-
centration front from the cylinder surface is observed. In both cases the gradients
of concentration are oriented opposite to the magnetic driving force. It can be
shown that this is a general law: the bulk gradients of concentration always are
anti-parallel to the magnetic sedimentation force. Thus, the magnetic stratifica-
tion of ferrocolloids is unstable. In Ref. [11] the convective stability of fluid in a
coaxial gap is analyzed by taking into account not only for the magnetophoresis
but also for the mass diffusion of particles. Despite the forming of concentration
boundary layers near gap walls, the unstable bulk gradient of concentration still
remains and, reaching a critical value of the solute Rayleigh number, the onset
of a specific diffusion-magnetic convection is observed. This effect is confirmed
experimentally [12].
If ∇ρ×∇H = 0 , the solute-driven convection starts to develop monotonously
and immediately after the moment when the magnetic field is switched on. The
characteristics of isothermal magnetic convection in colloids are analyzed in Ref
[13] considering the particle transfer near transversally magnetized cylinder. Nu-
merical simulation of the problem shows that in the vicinity of the cylinder should
develop a system of intensive convection rolls, the convection cause a deforma-
tion of the concentration profiles. This effect is investigated experimentally by
performing holographic measurements of the unsteady distribution of particles
around the cylinder [14]. A typical interferogram of the concentration field is
displayed in Fig. 6. The experimental results coincide very well with the con-
centration profiles calculated under the assumption of a solute driven magnetic
convection. Taking into account the numerically established fact that the con-
centration profiles formed under the effect of a convection, strongly differ from
those valid for the motionless fluid, the results presented in Fig. 6 can be con-
sidered as an experimental confirmation of the theoretically predicted magnetic
convection in isothermal fluids.
5 Heat and Mass Transfer Problems

The last term in (12) causes an interrelation of the heat and the mass transfer
processes. Thermodiffusion in dispersions is a relatively new research problem.
It is known that small particles in aerosols in comparison with the molecules in
liquids have relatively high thermophoretic mobility. The Soret coefficient de-
pends on Knudsen number, in rarefied gases it can be calculated taking into
account for a slip velocity and a temperature drop on particle walls. The ther-
modiffusive properties of colloids are not investigated in details at the moment.
The first experiments are evidence of an unusually high thermophoretic mobil-
ity of nanoparticles in ferrofluids. Using two different methods, the separation
measurements in thermodiffusion column [15], and the analysis of a diffraction
signal from optically induced particle grating in thin films [16,17], it is shown
134 E. Blums
Fig. 6. Distribution of particle concentration in a ferrofluid (magnetite in kerosene,

mean diameter of particles 8 nm) in the vicinity of a transversally magnetized (B0 =
65 mT) cylinder (nickel wire, diameter 1 mm) [14]. (a) – experiment, duration of
separation 20 min; (b) – the corresponding picture resulting from theory accounting for
the magnetic solutal convection, the concentration difference between the neighboring
interference fringes and lines is equal to Δϕ/ϕ0 = 0, 12.
that surfacted ferrite particles in hydrocarbons are transferred toward decreasing

temperatures. The measured Soret coefficients are several orders of magnitude
higher than those usually found in molecular liquids, they reach values approxi-
mately equal to ST = +0, 15 K−1 . Electrically stabilized particles in ionic colloids
usually have negative Soret coefficients [16,17].
To give a notion to the practical importance of thermodiffusive processes in
ferrofluids, in Fig. 7 is presented a typical curve of particle separation in ther-
modiffusion column. The separation effect is very strong, even at relatively small
temperature gradients a more than 50% difference in particle concentration in
the channel ends during the first 4 - 5 hours can be achieved. In long time ex-
periments, reaching the steady regime, almost complete educing of ferroparticles
from the carrier liquid can be achieved [18].
The nature of the Soret effect in colloids is not well understood yet. Some
general ideas of a theory of the slip velocity at a solid-liquid interface resulting
from tangential temperature gradient are reviewed in Ref. [19]. Analyzing the
enthalpy flux carried by forced convection of fluid across a porous barrier and ap-
plying Onsager’s reciprocal theory for the slip velocity, in Ref. [20] it is predicted
that free particles in surfacted colloids should move toward decreasing temper-
atures. According to the recent theoretic considerations, the Soret coefficient of
surfacted particles is proportional to the Gibbs absorption and the length of
surfactant molecules as well as to the size of particles [21]. The measured values
ST qualitatively well agree with this theory.
Fig. 7. Unsteady separation of particles (Mn-Zn ferrite in tetradecane,) in thermodif-

fusion column [30]. The column consists of a flat vertical channel of width δ=0.5 mm
and of height L=780 mm, difference of the wall temperatures is ΔT=10 K. The non-
dimensional time is τ =Dt/ δ 2 . Dots correspond to the initial sample with ϕ=0.092,
circles – to the diluted one with ϕ=0.023.
In recent experiments we have examined the thermophoretic properties of fer-

rofluid emulsions. Ferrofluid droplets in water are moving toward increasing tem-
perature, the Soret coefficient reaches a value ST = −480 K−1 [22]. We suspect
that such a high thermodiffusive mobility of droplets is a result of Marangoni-
type transfer. According to Ref. [23], the droplet velocity in a non-isothermal
fluid follows the expression

d ∂σ
uσ = − ∇T (15)
2(3η + 2η0 ) ∂T
Here η and η0 are the viscosity‘s of the ferrofluid droplet and the surrounding
liquid, ∂σ/∂T is the dependence of the surface tension on temperature and d
is the diameter of the droplet. The experimental results agree qualitatively well
with the value of ST calculated from the dependence (15).
Theory [24] trays to explain the reason of negative Soret coefficients in ionic
ferrofluids [16]. It is shown that the direction of thermodiffusive transfer of
charged particles depends on electrochemical parameters of the colloid. For an
infinitely thin Debye layer, independently of the surface potential value, the par-
ticle migration velocity is always directed opposite to the temperature gradient,
ST > 0. At finite thickness of the double layer this is a case only for small
surface potentials. For high values of ζ-potential the theory predicts a reverse
136 E. Blums
direction of particle transfer, starting some critical values of Debay length the
Soret coefficient can be negative. The parameters of ionic ferrofluid used in the
experiment [16] correspond to such requirements.
From (12) it follows that in colloids of a non-uniform magnetization specific
mass transfer effects can be observed even if the external magnetic field is con-
stant. Let us consider a homogeneous magnetic field B = const directed along
the temperature gradient. From equation divH = −divM we obtain (we assume
M0 = 0)
∂Mi ∂Mi ∂Mi
∇H = −∇Mi = − ∇ρi − ∇T − ∇H (16)
∂ρi ∂T ∂H
The mass flux (12) with account for the equation (16) can be rewritten in
the form of (for a simplicity the term of gravitation sedimentation is omitted)

μ0 αc M 2 mg μ0 αT M 2 mg
ji = − D + (1 + ni ) ∇ρi − ST − ρi D(1−ni )∇T
fν (1 + χ) fν Dρi (1 + χ)
(17)
Respectively, the magnetic stratification of ferroparticles under the influence
of internal field gradients can be considered as an increase of the mass diffusion
coefficient and as a reduction (in surfacted colloids) of the Soret coefficient. More
detailed analysis leads to a conclusion, that the change in coefficient D should
be observed also in the case when field is oriented normally to the temperature
gradient [25,26]. The only difference is that now (at B ⊥ ∇ρi ) the field causes a
reduction of the mass diffusion coefficient. Both effects, the increase in D under
the effect of a longitudinal field B∇ρi and its reduction B ⊥ ∇ρi , recently are
confirmed experimentally [26,27], see Fig. 8.
The Soret effect of magnetic particles in the presence of a uniform magnetic
field is considered in Ref. [28]. From the analysis of hydrodynamic Stokes prob-
lem for a spherical magnetic particle it follows, that if B∇T , the nonuniformity
of fluid magnetization causes a motion of particles in the direction of tempera-
ture gradient in accordance with the dependence (17). Contrary, in the presence
of a transversal field B ⊥ ∇T , the particles are transferred toward decreasing
temperatures. This so-called “magnetic Soret effect” and its anisotropy have
been investigated experimentally. The separation measurements in thermodif-
fusion column [29,30] confirm the general predictions of hydrodynamic theory:
in colloids with surfacted particles of positive Soret coefficients the transversal
magnetic field B ⊥ ∇T causes an increase in ST , whereas at B∇T a strong
reduction of ST takes place (see Fig. 9).
Similar results are obtained also in the experiments on optical grating in
thin ferrofluid layers [27]. But, it is interesting to note that the measured mag-
netic field effects always are significantly stronger than those predicted by the
hydrodynamic theory. There might be several reasons of that. Firstly, the con-
centration profiles in nonisotermic fluids dvelop under the combined action of the
thermodiffusion coefficient DT and the mass diffusion coefficient D. Therefore,
the measurd values of Soret coefficient ST = DT /D reflect not only the direct
influence of B on DT but also the dependence D = D(H). The second reason
can be the polidispersity of colloids. In Ref. [31] are reported some calculations
Fig. 8. Anisotropy of mass diffusion (meghemite particles in cyclohexane) in the pres-

ence of a magnetic field [26]. The diffusion coefficient is calculated analyzing the re-
laxation of a diffraction signal from optically induced grating structures. Triangles
correspond to parallel field B// ∇ρI , dots and circles – to B⊥∇ρI . Solid lines repre-
sent the calculation results based on thermodynamic theory [26], dashed lines – the
corresponding results of hydrodynamic theory [28].
Fig. 9. Magnetic Soret effect in surfacted ferrocolloids. Black labels correspond to

experiments with B⊥∇T, white labels – to experiments with B// ∇T. Squares – mag-
netite in tetradecane, ϕ=0.022 [29], triangles – Mn-Zn ferrite in tetradecane, ϕ=0.023
[30], circles and dots– meghemite in cyclohexane, ϕ=0.019 [27]
138 E. Blums
of particle sedimentation employing the Batchelor model [32]. It is shown that

the sedimentation velocity in polidisperse colloids is significantly higher than
that in corresponding systems of monodisperse particles. Obvioulsly, a similar
effect is valid also for the thermodiffusive separation. Numerical estimantions
show that in the presence of strong parallel fields B//∇T the Soret induced
concentration difference in colloids is high enough to reach the critical value of
magnetic Rayleigh number (14). This can be a reason of disapearence of the
difraction signal observed in some grating experiments.
The very high Soret coefficient of ferroparticles is not only an important
practical problem for ferrofluid applications but brings up also new research
tasks. The most interesting problems are: the combined thermal and Soret-driven
convection in thermodiffusion columns [33], the stability of convective shear flow
in channels under the effect of a transversal magnetic force in nonisothermic
and stratified colloids, the stability of convection in vertical fluid layers with an
adverse density gradient due to a negative Soret coefficient of heavy particles [34]
etc. There appears also a new class of the stability problems of double diffusive
magnetic convection in stratified colloids [35].
6 Conclusions
Magnetic convection in non-isothermic and stratified ferrocolloids can be used
to develop new effective cooling systems (for loudspeakers, electric transform-
ers, magnetically controlled thermosyphons) etc. Magnetic and thermodifusional
stratification of colloids should be taken into account considering the long-time
working ability of magnetofluid devices.
The research on separation in thermodiffusion columns reported in Section
5 is supported by BMBF and by Alexander von Humboldt Foundation.
References
1. E. Blums, A. Cebers, M. M. Maiorov: Magnetic Fluids (Walter de Gruyter, Berlin,
New York, 1997)
2. R. E. Rosensweig: Ferrohydrodynamics (Cambridge University Press, Cambridge,
1985)
3. G. Kronkalns, M. M. Maiorov: Magnetohydrodynamics 3, 28 (1984) (in Russian)
4. V. G. Bashtovoy, A. N. Vislovich, B. E. Kashevsky: Zh. Prikl. Mat. Techn. Fiz. 3,
88 (1978) (in Russian)
5. M. A. Martsenyuk: “Thermal conductivity of a suspension of ellipsoidal particles
in a magnetic field”. In: Proceedings of the 8 ght Riga MHD Conference (Zinatne,
Riga, 1975) 1, pp. 108–109 (in Russian)
6. B. M. Tareev: Coll. Journ. 6, 545 (1940) (in Russian)
7. M. I. Shliomis: Fluid Dynamics 6, 957 (1973) (in Russian)
8. K. Gotoh, M. Yamada: J. Phys. Soc. Jap. 51, 3042 (1982)
9. B. A. Finlayson: J. Fluid Mech. 40, 753 (1970)
10. L. Schwab: Ķonvektion in Ferrofluiden. Ph.D. Thesis, Ludwig-Maximillians Uni-
versität München, München (1989)
11. A. Yu. Chukhrov: Magnetohydrodynamics 22, 254 (1986)

12. S. Odenbach: Konvektion durch Diffusion in Ferrofluiden. Ph.D. Thesis, Ludwig-
Maximillians Universität München, München (1993)
13. E. Blums, A. Rimsa, A. Yu. Chukhrov: Magnetohydrodynamics 23, 139 (1987)
14. E. Blums, A. Yu. Chukhrov, A. Rimsa: Int. J. Heat and Mass Transfer 30, 1607
(1987)
15. E. Blums, A. Mezulis, M. Maiorov, G. Kronkalns: J. Magn. Magn. Mater. 169,
220 (1997)
16. J.Lenglet: Generation de second harmonique et diffusion Rayleigh forcee dans les
colloides magnetiques. Ph.D. Thesis, University Paris 7, Paris (1996)
17. A. Mezulis: Mass transfer phenomena in non-isothermic magnetic colloids. Ph.D.
Thesis, University Paris 7, Paris (1999)
18. T. Völker, E. Blums, S. Odenbach: Magnetohydrodynamics 37, 274 (2001)
19. J. L. Anderson: Ann. Rev. Fluid Mech. 21, 61 (1989)
20. B. V. Derjaguin, V. Churaev, N. V. Muller: Surface Forces (Consultants Bureau,
New- York, 1987)
21. K. I. Morozov: “Soret effect in disperse systems“. In: Proceedings of the
3 rd ASME/JSME Joint Fluids Engineering Conference, July 18-23, 1999, San
Francisco, CA (ASME, 1999), FEDSM99-7784
22. E. Blums, A. Mezulis, N. Buske, M. Maiorov: J. Magn. Magn. Mat. (to be pub-
lished)
23. N. O. Young, J. S. Goldstein, M. J. Block: J. Fluid. Mech. 6, 350 (1959)
24. K. I. Morozov: J. Magn. Magn. Mater. 201, 248 (1999)
25. K. I. Morozov: J. Magn. Magn. Mater. 122, 98 (1993)
26. J. C. Bacri, A. Cebers, A. Bourdon, A. Demouchy, G. Heegard, B. M. Kashevsky,
R. Perzinsky: Phys. Rev. E 52, 3936 (1995)
27. G. Demouchy, A. Bourdon, J.-C. Bacri, F. Da. Cruz, A. Mezulis, E. Blums: “Forced
Rayleigh scattering determination of the Soret coefficient and of the Thermodiffu-
sion mobility of ferrofluids under applied magnetic field” . In: Proceedings of Fourth
International PAMIR Conference “Magnetohydrodynamic at Dawn of Third Mil-
lenium”, Presqu’ile de Giens, France, September 18-22, 2000 (LEGI, Grenoble,
2000), pp. 433–438
28. E. Blums: J. Magn. Magn. Mater. 149, 111(1995)
29. E. Blums, A. Mezulis: “Thermal diffusion and particle separation in ferrocolloids”.
In: Transfer Phenomena in Magnetohydrodynamic and Electroconducting Flows,
ed. by A. Alemany, Ph. Marty, J. P. Thibault (Kluwer, Dordrecht, 1998) pp. 1–18
30. E. Blums, S. Odenbach, A. Mezulis, M. Maiorov: Phys. Fluids 10, 2155 (1998)
31. A. Yu. Chukhrov: “Gravitation Sedimentation of Polydisperse Magnetic Col-
loids”.In: Proceedings of 6 th USSR Magnetic Fluid Conference (Institute of Me-
chanics, Moscow State University, Moscow, 1991) pp. 157–158 (in Russian)
32. G. K. Batchelor: J. Fluid Mech. 119, 379 (1982)
33. E. Blums, S. Odenbach: Magnetohydrodynamics 37, 187 (2001)
34. M. M. Bou-Ali, O. Ecenarro, J. Madariaga, C. M. Santamaria: Phys. Rev. 59,
1250 (1999)
35. M. I. Shliomis, M. Souhar: Europhys. Letters 49, 55 (2000)
Statistical Physics of Non-dilute Ferrofluids
Andrey Zubarev
Department of Mathematical Physics, Ural State University, Lenin Av., 51, 620083
Ekaterinburg, Russia
Abstract. The influence of inner microscopical structures, consisting of various num-

ber of particles, on the macroscopic physical properties of ferrofluids is discussed. The
main attention is focused on linear chain-like microclusters. The simplest model of the
system, taking these chains as rod-like aggregates, where thermal fluctuations of the
shape of these chains are ignored is considered. The boundaries of validity of this model
are estimated and it is shown that - as a first approximation - it can be used for many
real ferrofluids. The influence of the chains on magnetization and rheological properties
of ferrofluids is evaluated. The analysis shows, that these aggregates induce physical
phenomena which are impossible in magnetic liquids with separate particles and which
are thus not described by the classical models for ferrofluids.
1 Introduction
One of the fundamental problems of physics of ferrofluids (magnetic liquids) is
the determination of their macroscopical characteristics as a function of their
inner composition, i.e. of shape, size distribution, physical properties and con-
centration of the magnetic particles and of the properties of the carrier liquid.
The classical theories for ferrofluids (see references and discussion in [1], [2])
deal with ideal models of very dilute systems in which any interactions between
particles are negligible. However in real magnetic liquids these interactions often
lead to long-range correlations between positions and orientations of the particles
and to the appearance of bulk drop-like, linear chain-like and other heterogeneous
aggregates, consisting of various number of particles. These inner structures play
a fundamental role for the physical properties of magnetic fluids and, therefore,
the study of the structures and internal states of ferrofluids gives a key for the
understanding of the properties and the macroscopical behavior of these systems.
Since the particles in typical ferrofluids are small (mean diameter is about
10 nm), they are subject to intensive translational and rotational Brownian mo-
tion. Any inner structure in a ferrofluid is a result of a competition between
thermal motion of the particles, their magnetic interaction and hydrodynamical
forces, which are especially significant when the magnetic liquid is exposed to
macroscopical flow. That is why the internal states of magnetic liquids can be
understood and described only on the basis of an appropriate statistical the-
ory. From the point of view of statistical physics the following main specifics of
ferrofluids have to be considered.

144 A. Zubarev
The first important peculiarity of magnetic liquids is the magnetodipole in-

teraction between the particles being non central and of long-range nature, de-
creasing only like r−3 .
Secondly, the increase of orientation of the magnetic moment of the ferropar-
ticles under influence of a magnetic field leads to an increase of the effective
attraction between them. This effect stimulates, and, moreover, can induce bulk
condensation phase transitions in ensembles of particles and can lead to the
appearance of dense drop-like aggregates observed in many experiments (see be-
low). The condensation phase transition, induced by an external magnetic field,
is a unique peculiarity of magnetic liquids and similarly of magnetorheological
suspensions.
Next, unlike molecular systems, where all molecules are identical, real mag-
netic fluids are always polydisperse. The peak of the particle size distribution
lies for typical ferrofluids in a region close to 10 nm. The biggest particles -
with a diameter of more than 15 nm - having, as a rule, a very small concentra-
tion, can form various microstructures or can play the role of centers of cluster
formation for the small ones. Therefore, the biggest particles, in spite of their
small concentration, can play an important, often decisive role in determining
the physical properties of ferrofluids and must be taken into account in theories
for real systems.
The first statistical models of dense equilibrium homogeneous (without any
heterogeneous clusters) ferrofluids were suggested in [3,4]. In these models mag-
netic interaction between the particles was taken into account with the help of
the classical Weiss theory of a self-consistent field. With the help of these mod-
els [3,4] it was shown for the first time that interparticle interaction increases
the equilibrium magnetization of a ferrofluid, and that a magnetic field stim-
ulates the “gas-liquid” phase transition in an ensemble of particles. However,
since these models are based on the Weiss approach, they predict spontaneous
ferromagnetic order in a ferrofluid and that the critical temperature of the para-
magnetic - ferromagnetic phase transition is higher than the temperature of the
condensation transition. Therefore, in these theories, the dense (”liquid”) phase
must be always in a ferromagnetic state. But the ferromagnetic order in ferroflu-
ids has been never observed in experiments. Moreover, there are fundamental
reasons to doubt that spontaneous magnetization can occur in suspensions of
dipole-dipole interacting particles. Indeed, as well known from magnetism in
solid state physics, exchange forces create the ferromagnetic order; dipole-dipole
interactions destroy the order. It is unclear, why and how the forces, destroying
magnetic order in solid systems, should create it in liquid ones. It should be
mentioned that various variants of self consistent effective fields were taken into
account in [5]. As in the theories in [3,4], these models lead to the conclusion
that spontaneous magnetization can occur in ferrofluids. This result is a direct
consequence of the Weiss-like approximation used in these models. For the rea-
sons just mentioned a model with a self-consistent field for the description of
ferrofluids should be used with caution.
Statistical Physics of Non-dilute Ferrofluids 145
More advanced theories of homogeneous ferrofluids, using ideas of the

Wertheim mean spherical model and the regular theory of thermodynamical
perturbations, were suggested in [6] and [7], respectively. The results of cal-
culations of the magnetization of ferrofluids using these theories are in good
agreement with experiments and with each other. The models [6,7] predict a
“gas - liquid” phase transition in ferrofluids and the existence of these phases in
a paramagnetic state only.
The dynamical properties of non dilute homogeneous magnetic fluids, con-
taining no heterogeneous “drops” or “chains”, were studied in [8] on the base of
the mathematically regular theory of perturbations [7]. The results of [8] show
that interparticle interactions increase the characteristic times of magnetic and
hydrodynamic relaxation of dense ferrofluids as well as their effective viscosity.
We just noted that real non dilute ferrofluids are often inhomogeneous - bulk
(drop-like) as well as linear (chain-like) micro- and mesostructures appear in
these systems under suitable conditions. The drop-like aggregates were observed
in many experiments (see, for example [9-13]). The linear chains are too thin to
be detected in most direct observations, however recent computer experiments
[14-17] demonstrate that they can appear in ensembles of dipolar particles un-
der suitable conditions. The appearance of such heterogeneous structures affects
strongly the macroscopic characteristics of ferrofluids, especially their dynami-
cal properties. For instance, experiments [18,19] show that typical commercial
ferrofluids, like APG513A from Ferrofluidics, with a total hydrodynamic vol-
ume concentration of about 27 %, show an increase of the effective viscosity of
up to 16 times compared with the zero field viscosity, even under influence of
weak magnetic fields, being far from saturation of the magnetoviscous effect!
The magnetoviscous effect is especially strong when shear rate is small (about
0.1 s−1 ). When the shear rate increases, the viscosity decreases very fast. At
the same time calculations of this effect, carried out on the base of classical
theories of non interacting particles, predict a rise of viscosity about several per
cent only under the conditions of [18,19]. Furthermore the dependence of the
viscosity on shear rate is much weaker than in the experiments. This significant
(several orders of magnitude) discrepancy between the experimental results and
the predictions of the classical theories show clearly that a new physical effect,
not present in dilute ferrofluids, appears in concentrated systems.
This effect can either be the formation of bulk (drop-like) or linear (chain-
like) heterogeneous structures, appearing in dense ferrofluids due to dipole-dipole
interactions between the particles. We just discussed briefly statistical models of
the condensation phase transition in ferrofluids, which lead to the appearance of
bulk-like dense “drops”. The first theoretical model of chains in ferrofluids was
suggested in [20]. However, this model is based on the theory of homogeneous
fluctuations of density. Therefore, the structures, described there, are rather
clouds of density than heterogeneous aggregates. The first model of chains as
heterogeneous clusters was suggested in [21]. The chain length distribution was
estimated in these works using methods of the kinetics of chemical reactions,
modelling the combination and recombination of particles from chains. Unfor-
146 A. Zubarev
tunately, the final results of the theory are cumbersome and it is not an easy
thing to use them for calculations of the physical properties of magnetic fluids.
Similar methods of chemical kinetics for the investigation of chain-like structures
in ferrofluids in the absence of external field were used in [22].
Models of chains as heterogeneous fluctuations of density in magnetic fluids
on the basis of statistical thermodynamics were suggested in [23-25]. Methods,
developed for the theory of very long flexible macromolecules, with a tremendous
number of monomers in the order of 104 -109 ), were used in the last mentioned
model. However it is difficult to expect such long linear clusters in real ferroflu-
ids. The models in [23,24] use a very simple approximation of the chains as
straight rod-like aggregates, i.e. thermal fluctuations of the shape of the chains
are ignored in these models. In spite of the fact that this strong simplification
can be used only when most of the chains are short enough, this allows us to
understand the physical origin of group of phenomena in magnetic fluids and to
estimate them, at least, by the order of magnitude. The main points and results
of the models [23,24] are discussed in parts 2-4 of this chapter. It should be noted
that effects driven by the flexibility of the chains are discussed in [26].
The formation and evolution kinetics of the “drops” as a result of the “gas-
liquid” phase transition in a system of interacting, but single particles, was
studied theoretically in [27,28]. As in [3,4] and [6,7], any linear chains are ignored
in these models and the ferroparticles are treated as separate. These theories
describe correctly the principal points of the phase transitions, but numerical
experiments [14-17] and analytical models [25,29] show that linear chains appear
in systems of dipole-dipole interacting particles long before the bulk condensation
phenomena takes place. Therefore these chains must be necessarily taken into
account in theories of phase transitions and other structural transformations in
ferrofluids.
It is very difficult to study various kinds of inner structures in ferrofluids and
their influence on the macroscopic properties of these systems in the framework
of one model. That is why it is reasonable to start with separate analysis of
each kind of these structures. In this paper some results of the theoretical study
of effects of chain-like aggregates on the macroscopic dynamical properties of
ferrofluids are presented. To avoid too cumbersome mathematics and to reach
physically reasonable estimates, we use the following approximations.
1. We start with an analysis of monodisperse systems. A bidisperse model of
a polydisperse ferrofluid is considered in the last part of the paper.
2. We assume that the considered particles are big enough and that the mag-
netic moment of each of them is frozen into its body. Estimates of the minimal
size of the particle necessary to provide this approximation are given, for ex-
ample, in [1] or [2]. One can show that the magnetic interaction of very small
superparamagnetic particles with magnetic moments rotating freely inside their
bodies, is too weak to provide any heterogeneous structures. Therefore, the as-
sumption of magnetically hard particles is in agreement with our consideration
of chain-like structures consisting of reasonably sized particles.
3. We neglect any interactions between the chains. One can show that, be-
cause of the fact that the chains can not overlap, particles from the different
chains can not be situated in places, where their interaction is strong, since
these positions, for every particle, are occupied by other particles from the same
chain. Therefore interaction between particles of different chains is weaker than
interaction between those from one chain. That is why the interparticle interac-
tion inside one chain plays the main role in the formation of the chain structure.
Thus, the model of non interacting chains seems to be a reasonable first approxi-
mation, at least when concentration of the interacting particles is not high. This
assumption is not suitable for very dense systems, where the chains lose their
individuality and are destroyed due to interaction between them [16].
4. We restrict ourselves to situations where the interaction of the magnetic
particles with the external magnetic field is much smaller than the magnetic
interaction of two neighboring particles in the chain. It means that the chains
react with the field as whole aggregates. This restriction is valid in many real
experimental situations.
2 The Chain Size Distribution

The influence of a chain on the macroscopic properties of a ferrofluid is as
stronger as longer the chain is. On the other hand, the length of the chain is a
magnitude that is determined by a competition between magnetic interaction of
the particles and their thermal motion. Therefore our first problem is to deter-
mine the size distribution of the chains. In this section we consider equilibrium
situations; a possible generalization of these results for systems in steady state
far from equilibrium is discussed in section 4.
Consider a ferrofluid consisting of identical spherical particles with hydrody-
namical radius (including surface layer) a and magnetic moment m. We denote
the hydrodynamical volume concentration of the particles by ϕ. Let gn be the
number of the n-particle chains in a unit volume of the ferrofluid. In the equi-
librium state this distribution function provides a minimum of the free energy
F of the unit volume. Under the approximation discussed above this energy can
be presented as (see, details in [23,24]):
nc
!
gn
F = kT gn ln + gn fn (κ, ε) (1)
n=1
e
mH μ0 m2
e = 2.72..., κ = μ0 , ε=
kT 16πa3 kT
Here kT is the absolute temperature in energetic units, fn is the dimensionless
free energy of the chain due to its inner structure, nc is the maximal number
of particles in the chain, estimated below. For equilibrium states one may put
nc = ∞ (see eq. (15)). The first term in the parentheses in equation (1) represents
the entropy of the ideal gas of n-particle chains determined by their translation
motion, H is the magnetic field inside the sample, μ0 is the vacuum permeability,
148 A. Zubarev
κ and ε denote the dimensionless energies of the interaction of the magnetic

particle with the field H and of the magnetodipole interaction of two neighboring
particles in the chain, respectively.
We should minimize the free energy (1) under the obvious condition of nor-
malization:
!nc
ϕ 4
ngn = , v = πa3 (2)
n=1
v 3
After standard calculations one can obtain:
1
gn = exp (−fn + λn) (3)
v
where λ is the Lagrange multiplier. To calculate it we’ve to substitute equation
(3) into equation (2). This leads to a non linear equation with respect to λ which
can be solved analytically or numerically. However first one needs to determine
the “internal” free energy fn . This energy can be presented as follows:
fn = − ln Zn (4)
⎛ ⎞
!n !
n
Zn = exp ⎝ (ν i κi ) − Ud (ν i , ν j , r ij )⎠ dΩ
i i>j
Here ν i is the unit vector aligned along the magnetic moment of the i − th
particle, r ij is the radius-vector, connecting the centers of the i − th and j − th
particle, Ω includes all ν i and r ij and Ud is the dimensionless potential of the
dipole-dipole interaction.
As usual, the statistical integral Zn can not be calculated in a general form.
To reach reasonable estimates we may take into account only the interaction be-
tween nearest particles in the chain. This approximation was used in the models
[22-26]. The interaction between nearest particles and throughout the nearest
neighbors was taken into account in [21] and led to cumbersome final results. In
the “nearest neighbor” approximation the statistical integral (4) is
& n '
! !
n
Zn = exp (ν i κi ) − Ud (ν i , ν i+1 , r i,i+1 ) dΩ (5)
i i
Even in this approximation the many-particle integral can not be decoupled

to low-dimensional integrals for finite κ. That is why a mathematically strict
calculation of this integral for arbitrary κ is impossible. Some estimates for the
limiting situations κ = 0, κ = ∞ and extrapolated approximations for finite κ
are suggested in [21,26].
Here we take into account that the appearance of chains in ferrofluids is
possible only when the dimensionless parameter ε of the dipole-dipole interac-
tion is significantly larger than unity. Since the potential Ud has a minimum
when the vectors ν i , ν i+1 , r ij are parallel and ri,i+1 = 2a, we can replace
exp (Ud (ν i , ν i+1 , r i,i+1 ) by δ(ν i − ν i+1 )δ(ν i − r i,i+1 )δ(ri,i+1 − 2a) as a first ap-
proximation. In this approximation we ignore thermal fluctuations of the chains
and treat them as straight rod-like aggregates. This is justified when the con-
tour length of the chain is less or about the persistent length l0 of the chain.
Using standard considerations of the theory of linear polymers, one can show
that l0 ∼ 2aε. Therefore the “rod” model of the chains can be considered as
justified when the mean contour length 2a < n > of the chain is not more than
2aε, i.e. < n >≤ ε where < n > is the mean number of particles in the chain.
In the framework of this model the dimensionless free energy fn may be
written down as:
sinh κn
fn = − ε (n − 1) + ln (6)
κn
The first term in the square brackets represents the absolute value of the
energy of magnetic interaction of particles inside the rod-like chain, estimated
in the approximation of nearest neighbors. The second term corresponds to the
Langevin energy of the rod in the dimensionless field κ. It should be noted that
relation (6) is the upper estimate for the absolute magnitude of fn .
Substituting equation (6) into (3) and, then, into (2), taking into account
that one can put nc = ∞ for equilibrium systems (see equation (15)), we get
after some calculations (details in [23])
xn0 sinh κn
gn = exp(−ε), (7)
v κn
where
(
2
2y cosh κ − sinh κ − (2y cosh κ − sinh κ) − 4y 2
x0 = , y = κϕ exp(ε)
2y
Figure 1 illustrates calculations of the volume concentration vngn of the par-
ticles collected in the n-particle chains. This volume concentration, as function
of n, has a maximum increasing with κ and ε. Similar maxima were obtained
also in [29].
The mean number of particles in the chain is
)
ngn ϕ
< n >= )n = ) (8)
n gn v n gn
Let ϕc be the volume concentration, corresponding to the condition < n >=

ε. The plots of ϕc as function of κ for two values of ε are shown in Fig 2. If at
a given ε the condition for the concentration fulfills ϕ < ϕc , then the model of
rods can be used.
Using distribution function (7), one can estimate magnetization M of fer-
rofluid as in [23]: !
M =m nL(κn)gn (9)
n
150 A. Zubarev
Fig. 1. The volume concentration ρn = vngn of particles collected in n− particle chains

vs. n for ϕ = 0.05, κ = 1; ε = 4 and 5 (dashed and solid lines, respectively)
Fig. 2. The critical volume concentration ϕs of the particles for the model of straight
aggregates vs the dimensionless magnetic field. Solid and dashed lines correspond to
ε = 5 and ε = 3, respectively
where L(x) = coth(x) − x−1 is Langevin function. Comparison of calculations

of the magnetization with the experiments [30] are shown in Fig. 3. One can see
that the model of chains leads to reasonable results, at least when only short
chains are expected.
3 Rheological Properties for Vanishing Shear Rate

Let us consider a ferrofluid subjected to a shear flow. We assume here that
the shear rate is weak enough to neglect deviation effects from the equilibrium
size and shape of the chains. As shown in [23], deformation of the chains under
Fig. 3. Results of measurements [30](dots) and calculations according to equation

(9)(lines) of the initial susceptibility χ of a ferrofluid vs. the magnetic volume con-
centration ϕm for two different systems. Their parameters are given in [30]. Circles
and squares correspond to lines 1 and 2, respectively
hydrodynamical flow with shear rate γ̇ is weak when

kT ε
γ̇ <<
η0 (2a)3 < n >
Here η0 is the viscosity of the carrier liquid. This inequality is valid for most
part of experimental situations.
It is well known since Einstein’s classical work, that the influence of dispersed
particles on the rheological properties of suspensions is connected with their per-
turbations of the macroscopic flow of the suspension. It is practically impossible
to calculate the perturbations induced by the chains because of their complex
shape. In order to get reasonable estimates, we model the n−particle chain, like
in [23] and [24], as a prolate spheroid with semiminor axis a and semimajor
one an . It is of principal importance that the volume of this spheroid is equal
to the sum of the volumes of the particles in the chain. Therefore, the volume
concentration of the model spheroids is equal to the total volume concentration
of the particles.
Using the well known results of the theory of statistical hydromechanics of
dilute suspensions of rigid spheroids (see, for example, [23,24,31]), we can write
down expressions for the Cartesian components of the average viscous stress
tensor σ̂ as follows

!nc
σik = 2η0 γik + Φn (γlm , ωlm , < ej es >n , < el em ej es >n )gn (10)
n=1

< ... >n = ...φn (e)de, i, k...s = x, y, z
The symbol ˆ here and below indicates tensor magnitudes, e is a unit vec-
tor aligned along axis of the chain, φn (e) is a distribution function over the
orientations of the chain normalized to unity, γij and ωij are the symmetrical
152 A. Zubarev
and antisymmetrical parts of the tensor of the velocity gradient, Φn is a known

and cumbersome linear function of γij and ωij . Its explicit form is given in
[23],[24],[31]. It should be noted that the function Φn increases fast with the
number n.
In order to determine the moments < ... >n in Eq.(6), we need to determine
the non equilibrium orientational distribution function φn . This can be found
using the Fokker-Planck equation for a non-spherical particle (see, for example,
[23,24,31]) :
∂φn de
= −I(φn ) + Dn I2 φn , (11)
∂t dt
where
de Dn
=− IUκ + ω̂e + λn γ̂e − ((eγ̂e)e), (12)
dt kT

∂
I= e× , Uκ = −kT (eκn),
∂e
D1 kT
Dn = , D1 = ,
nδn 6η0 v
Here Dn denotes the coefficient of rotational diffusion of a spheroid with an axis
ratio equal to n, the parameter δn , increasing with n, is given in [23,24,31].
The exact solution of equation (11,12) is unknown. To find its approximate
solution we use the ideas of the theory [32] and present the distribution function
φn in the form of the following trial function:
φn = φon (1 + ai (ei − < ei >on ) + bik (ei ek − < ei ek >on )) , (13)

κn
φon (e) = exp ((eκ)n) , < ... >on = ...φon (e)de
4π sinh κn
Here φon is the equilibrium Boltzman orientational function for the n-particle
rod in the magnetic field H, a and b̂ are a vector and a second rank tensor to be
determined. It should be noted that the trial function, used in the theory [32],
is similar to that in equation (13), but with bij = 0.
According to the ideas of Ref. [32], we determine the parameters ai and bij
using the equations for the first and second moments of the vector e of func-
tion (13). For this purpose equation (11) should be multiplied by components ek
and by the complex ei ek − 13 δik , and after integration over e we come to a set
of equations for the first and second moments of the function φn . Substituting
there φn in the form (13), we obtain a system of algebraic equations for ai , bij
(see details in [23,24,31]). If the state of the system is not far from equilibrium,
we may take these equations in linear approximation in ai , bij and the compo-
nents components γij ,ωij of the velocity gradient. After solving this system and
determination of ai , bij , using (13), we can calculate the moments < ... >n in
equation (10), and therefore, estimate the stress tensor σ̂ or, that is the same,
all rheological characteristics of the ferrofluid in the framework of the discussed
approximations.
Fig. 4. Dimensionless stationary effective viscosities θ as functions of external magnetic

field for ε = 3, ϕ = 0.05. Solid lines correspond to model with chains, dashed ones to
model with separate particles. The field is aligned along (a) gradient of flow and (b)
velocity
In general, the problem of the calculation of ai and bij is cumbersome, but

principally not difficult. The detailed calculations for two typical situations,
when the field is parallel to (1) the gradient and (2)the velocity of a flow, are
given in [24]. Some results for the dimensionless stationary effective viscosity
θp = (ηp − η0 )/η0 ) are shown in Fig.4 both for the model with chains and under
assumption that all particles are free (the last case corresponds to the classical
theories, discussed in [1,2]). The parameter ηp is the effective viscosity and the
index p = 1, 2 corresponds to the mentioned orientations of external magnetic
field. These plots show that the chains increase the effective viscosity signifi-
cantly. The magnetoviscous effect is especially strong when the field is parallel
to the gradient of velocity (p = 1). In this situation the viscosity increases with
the field monotonously. When the field is parallel to the velocity (p = 2), the
magnetoviscous effect is smaller, and, moreover, the viscosity depends on the
field non monotonously with a maximum. The physical origin of the last result
lies in the fact that the hydrodynamical resistance of the elongated n-particle
aggregate, which is parallel to the velocity, is smaller than the total resistance
of the n separate particles. When the field is small, the orientation of the chains
is chaotic and the main role in the formation of the field dependence of viscosity
is played by a blocking of the free rotation of the particles and chains in the
shear flow (it is the same mechanism that creates the rotational viscosity in di-
lute ferrofluids with separate particles). However, when the field is large enough,
the system of chains is highly ordered and an increase of the field leads to an
increase of this order as well as to an increase of the number of long, highly ori-
ented chains. This induces a decrease of viscosity. A similar effect of a decrease
of viscosity under influence of a magnetic field parallel to the flow velocity, is
well known for nematic liquid crystals. The maximum in the plot of viscosity vs.
magnetic field in Fig. 4 is a result of a competition between these two mentioned
mechanisms. One needs to stress that in the models of separate particles the
effective viscosity of ferrofluids monotonously increases with magnetic field for
all orientations.
154 A. Zubarev
Fig. 5. (a) Real and (b) imaginary part of complex dimensionless effective viscosity
vs frequency Ω of an oscillating flow (Ω = Ω/D1 ). The external field is aligned along
the gradient of the flow velocity, ε = 3, ϕ = 0.05. Solid and dashed lines correspond to
κ = 1 and κ = 0.5, respectively
Fig. 6. The same curves as in Fig.5 for ferrofluid with single particles
Figure 5 shows calculations of the real θ1 and imaginary θ part of the
dimensionless complex viscosity as a function of the frequency of the flow for a
magnetic field parallel to the gradient of the flow velocity. For comparison, Fig.6
shows the results of calculations for the same magnitudes under the assumption
that all particles are free. Comparing the results in Figs. 5 and 6, one can note a
strong increase of both components of complex viscosity due to the chains and the
fact that the frequency, corresponding to the maximum of θ , is much smaller
in the model with chains than in the model for free particles. It means that
the time of hydrodynamical relaxation in ferrofluids with chains is significantly
longer than the same time predicted by the classical models of separate particles.
Using this approach, one can estimate the normal components of the stress
tensor under shear flow [24]. It is known that the normal stresses can lead to the
so-called Weissenberg effect [33], observed usually in isotropic polymer liquids.
The results of an experimental study of this effect in ferrofluids are discussed
in [34]. It should be noted that this phenomenon can not take place in systems
with separate spherical particles only. The fact that it is observed in ferrofluids
is a result of the presence of anisotropic (chain-like) aggregates and qualitatively
supports the suggested model.
In the discussed model it is possible to estimate the components of magnetiza-

tion of a ferrofluid under conditions of a deformation flow. To do this, we should
first to take into account that - by definition - the macroscopical magnetization
is:
!∞
M =m n < e >n gn , (14)
n
Let the magnetic field and the velocity gradient of the flow be aligned along
the axis Oz, the velocity being aligned along axis Ox. For this situation, in linear
approximation in the symmetrical γ and antisymmetrical ω parts of the velocity
gradient tensor, the calculations give:
ax = ωA1 + γΛn A2 , 2bxz = ωB1 + γΛn B2
and

< ex >n = A1 < e2x > −no + B1 < ez e2x >on ω + Λn A2 < e2x >on +B2 < ez e2x >on γ
n2 − 1
Λn = ,
n2 + 1
The parameters A1,2 , B1,2 can be determined from a linear algebraic system
of equations for ax , bxz , which is given in [24] (equations (16),(17) of this work).
Substituting the last relation into (14), we get:
Mx = λ1 ω + λ 2 γ
!
nc

λ1 = m n A1 < e2x >o +B1 < ez e2x >o gn ,
n=1
!
nc

λ2 = m nΛn A2 < e2x >o +B2 < ez e2x >o gn ,
n=1
The calculations show that the parameters λ1,2 are increasing functions of ϕ, ε
and non monotonous functions of κ.
The same form for Mx with λ1 and λ2 is obtained in [35] from phenomenolog-
ical considerations. Both microscopical and phenomenological approaches show
that λ2 = 0 when only separate spherical particles are present in the ferrofluid.
This term can be non zero only due to the presence of non spherical aggregates
in ferrofluid. We would like to note that the classical models for magnetic fluids,
which deal with separate spherical particles and ignore any clusters, give λ2 = 0.
Specially organized experiments [35] demonstrate that for ferrofluids with non
vanishing concentration of particles λ2 = 0. This shows that elongated inner
structures are present in real ferrofluids.
The experiments on Weissenberg effect and on the “γ-dependence” of mag-
netization in ferrofluids are in a qualitative agreement with the suggested model.
Unfortunately, because of tremendous mathematical problems, appearing in the
solution of the macroscopic equations of flow for ferrofluids under external mag-
netic field, it is difficult to give a precise quantitative comparison between the
theory and experiments. Such a comparison is discussed in part 5 of the paper
for the rheological properties of magnetic fluids.
156 A. Zubarev
4 Rheological Properties of Ferrofluids

for Non Vanishing Shear Rate
In a non equilibrium situation the theorem of the minimum of the free energy,
used in part 3 for the determination of the distribution function gn is, strictly
speaking, not valid. The search for a solution of equation (12) for situations far
from equilibrium presents a very difficult mathematical problem. That is, why a
generalization of the approach, developed for quasi equilibrium ferrofluids with
chains, to non equilibrium situations (for example, when shear rate is not van-
ishing) is a very complicated problem. However these difficulties can be resolved
for many actual situations when the inequality γ̇/Dn << 1 (γ̇, as usual, is shear
rate) is held for all chains with non vanishing volume concentration vngn . One
can show that this inequality corresponds to the condition that the shear Pecklet
number P e, constructed for the chains, is small.
In this situation the distribution function φn in equation (11) is not far
from its equilibrium magnitude and the trial function (13) can be used in lin-
ear approximation in γ̇. At the same time, since P e is small, the transport of
the particles near the chains is almost the same as for the equilibrium state.
Therefore, the condition of the minimum of the free energy can be used for an
approximate estimate of gn . At the same time it is necessary to keep in mind
that hydrodynamical forces destroy too long chains in a ferrofluid under shear
flow. The maximal number nc of the particles in the non destroyed rod-like chain
has been estimated in [19] as:
*
D1
nc ≈ ε (15)
γ̇
When γ̇ is vanishing, nc tends to infinity. This result was taken into account
in the previous sections. When γ̇ is significant, but the inequality γ̇/D1 << 1 is
fulfilled, we should use the estimate (14) in equations (1),(2) and (10). Thus, in
first approximation, when the inequality γ̇/Dn << 1 is true for all chains, the
influence of shear rate on the chain structure can be taken into account by a use
of the estimate (15) in relations (1,2) and (10), remaining all other considerations
of parts 2 and 3 without changes.
5 Comparison with Experiments

A strong increase of viscosity in dense ferrofluids under weak magnetic field,
parallel to the gradient of flow, was detected in the experiments [18], [19]. Some
results of these measurements are shown in Fig.7 (dots). These experiments were
carried out with polydisperse magnetite ferrofluids containing about 27 vol% of
the hydrodynamical volume of particles; the mean diameter of the magnetic core
of the particles is about 10 nm. A diagram of size distribution for the particles
is given in Fig. 8.
Fig. 7. Experiments [18] (dots) and calculations (lines) of the magnetoviscous param-
eter S. a) Squares corresponds to shear rate γ̇ = 0.1s−1 , line 1 to γ̇ = 0; black circles
and line 2 to γ̇ = 0.5s−1 , white circles and line 3 to γ̇ = 0.9s−1 . b) light squares and
line 1 correspond to γ̇ = 1.05s−1 , black squares and line 2 to γ̇ = 5.23s−1
Fig. 8. Diagram of the size distribution for ferrofluid, used in experiments [18], [19]
One needs to stress that the 10 nm particles are superparamagnetic and the
energy of magnetic interaction between them is much smaller than the thermal
energy for room temperature. Therefore, these particles neither can influence sig-
nificantly the magnetorheological properties of the ferrofluid, nor they can form
any heterogeneous aggregates. It means that only the biggest particles, in spite
of their small concentration, give rise to the observed strong magnetorheological
effects. The analysis, given in Section 3, shows, that the physical origin of this
phenomenon can be chain-like aggregates, formed by the biggest particles.
The problem of a theoretical description of chain-like structure in a polydis-
perse ferrofluid, even under the used strong simplification, is very complicated.
To avoid cumbersome mathematics, we use a model of a bidisperse system con-
sisting of the “small” particles with sizes being near the mean size of particles
in the ferrofluid and “big” particles with sizes and volume concentration to be
determined. For maximal simplification of the calculations we assume that there
are only the “big” particles, which can form the chains. All the “small” particles
are supposed to be free.
158 A. Zubarev
Using the approach, described in Sections 2 and 3, we calculated the param-

eter S(H) = (η1 (H) − η1 (0)) /η(0) of the magnetoviscous effect for vanishing
shear rate. Then the magnetic diameter d as well as the hydrodynamic vol-
ume concentration ϕ of the “big” particles have been determined by comparing
the results of the calculations with the experiments for the smallest shear rate
(namely γ̇ = 0.1s−1 , see Fig 7a). As a result we obtained d = 16.5nm, ϕ = 0.17.
If we look at the diagram in Fig. 8, these results seems to be reasonable. Then,
using the consideration of section 4, we calculated S for several magnitudes of
γ̇, used in the experiments. The results of the calculations are given in Fig 7.
The agreement between the theory and experiment can be recognized as good,
especially taking into account that the model is extremely simple.
6 Conclusion and Discussions
The obtained results allow us to make the following qualitative conclusion. In

real polydisperse ferrofluids large particles are present, which provide strong
magnetoviscous and other dynamical effects, in spite on their very small con-
centration. The physical origin of these effects are heterogeneous (for example,
chain-like) structures, formed by these particles. The magnetorheological effect
is especially strong for small shear rate, when the structures are not destroyed.
Increase of shear rate leads to disruption of the structures and to a fast decrease
of effective viscosity.
The discussed simple model of the chains as rod-like aggregates allows us to
understand the physical origin of such physical phenomena in real ferrofluids as
non monotonic dependence of viscosity under magnetic field aligned along the
velocity of the flow; the dependence of the components of the macroscopical mag-
netization on the symmetrical part of velocity gradient tensor; the Weissenberg
effect; the strong magnetorheological effect. The estimates of the magnetoviscous
effect in real ferrofluid are in a reasonable quantitative agreement with exper-
iments.It should be stressed that these phenomena are impossible in systems
with spherical particles only.
At the same time one needs to admit that the simple model of chain structures
in ferrofluids opens many questions that can be a subject of further investiga-
tions.
First. What is the role of the flexibility of the chains in the formation of
the macroscopic properties of ferrofluids? The matter is that the longest chains
influence the dynamical properties of ferrofluids especially strong. However the
long chains can not be rod like, they should be similar to worm-like micelles.
Unfortunately, it is a very complicated problem to calculate the influence of
these “micelles” on the physical properties of the magnetic liquids.
Second. Can the real long flexible chains be in a linear state or do they
inevitably contain particles joined to the chains by side? In the last situation
the chains are rather bulk than linear aggregates and their analysis requires a
development of a special theoretical approach.
Third. What is the role of the “small” particles in the formation of the inner
structure of ferrofluids? The point is that simple estimates show that the energy
of magnetic interaction between typical magnetite particles with a magnetic
diameter about 10 nm and those with 16-17 nm for room temperature is slightly
more than kT . Therefore, the “small” particles can take part in the formation
of the chains. The previous analysis [37] shows that their influence on the chain
structure is not negligible, however this problem is worth a further detailed
examination.
Forth. What is the role of the chain-chain interaction in ferrofluids? This
question is especially intrigue, since many experiments (for example, [9-13])
demonstrate that the “gas-liquid” phase transition can take place in an ensem-
ble of ferroparticles under suitable conditions. Theories [3,4,6,7,27,28] explain
and describe the main points of these phase transitions, however they ignore
any possibility of the formation of chain-like structures. Computer simulations
[14-17] and theories [25,36] that take into account this possibility, conclude that
the condensation phase transition in ferrofluids is impossible - when tempera-
ture decreases the chains become longer in the spatially homogeneous system.
Thus, there is a qualitative contradiction between experimental and theoretical
(including computer experiment) results on phase transitions in ferrofluids. A
physical nature of this contradiction may lie in the following facts. First, the sur-
face layers on the ferroparticles do not screen van-der-Waals interaction entirely.
The combination of dipole-dipole forces and “tails” of the central van-der-Waals
interaction can lead to a bulk phase transitions even in ferrofluids with chains
[25]. The second factor, which can induce this transition, can be polydispersity
of real ferrofluids. Indeed, the “small” particles, with magnetic diameter about
10 nm and high concentration, can create a strong osmotic pressure that pushes
the “big” particles to regions where concentration of the “small” ones is low. As a
result domains with high concentration of big particles, where the osmotic pres-
sure compensate the pressure in an environment with small particles, can appear
as thermodynamical stable phases. Both of these scenarios of the bulk separation
(first, provided by the van-der-Waals tails and the second one, induced by the
“small” particles) are worth a thorough examination.
Finishing, one can note that in spite of the mentioned unsolved problems,
the simple model of non-interacting rod-like chains can be considered as a robust
basis for a qualitative understanding and study of the dynamical properties of
real non dilute ferrofluids. This model can serve also as a base for a develop-
ment of more advanced theories of internal structures and macroscopic physical
properties of magnetic liquids.
This work has been supported by grants of the Russian Basic Research Foun-
dation, projects NN 02-00-17731, 02-01-6072, 01-01-00058, grant CRDF, REC-
005 and grant of BMBF N RUS 00/196
160 A. Zubarev
References
1. M.I.Shliomis, Sov.Phys.Usp. 17, 153 (1974)
2. E.Blums, A.Cebers, M.Majorov, Magnetic Fluids(de Gruyter, Berlin, New York,
1997)
3. A.O.Tsebers, Magnetohydrodynamics, 18, 345 (1982)
4. K.Sano and M.Doi, J.Phys. Soc.Japan., 52, 2810 (1983)
5. D.Wei, G.N.Patey and A. Perera, Phys.Rev.E, 47, 506 (1993); H.Zhang and
M.Widom, J.Magn.Magn.Mater.122, 119(1993); Phys.Rev.E 49, 3591 (1994);
Phys.Rev.B 51,8951 (1995); B.Groh and S.Dietrich, Phys.Rev.E, 50, 3814 (1994);
Phys.Rev.E, 53, 2509 (1996); Phys.Rev.E, 57, 453518, (1998)
6. K.I.Morozov, Bull. Acad. Sci. Ussr, Phys. Ser. 51, 32 (1987)
7. Yu.A.Buyevich and A.O.Ivanov. Phys. A, 190, 276 (1992)
8. A.Yu.Zubarev and A.V.Yushkov, Zh.Eksp.Theor.Fiz 114, 892 (1998) [JETP 87
484 (1998)]
9. C.F.Hayes, J.Colloid. Interface Sci., 52, 239 (1975)
10. E.A.Peterson and A.A.Krueger, J.Colloi. Interfase Sci., 62, 27 (1977).
11. A.F.Pshenichnikov and I.Yu.Shurubor, Bull. Acad. Sci. USSR, Phys. Ser., 51, 40
(1987)
12. J.C.Bacri, R.Perzynski, D.Salin, V.Cabuil and R.Massart,J. Colloid. Interface Sci.,
132 43 (1989)
13. P.K.Khizhenkov, V.L.Dorman and F.G.Bar’akhtar, Magnetohydrodynamics, 25,
30 (1989)
14. J.M.Caillol, J.Chem. Phys., 98, 9835 (1993)
15. M.E.Van Leeuwen and B.Smit, Phys. Rev.Let. 71, 3991 (1993)
16. D.Levesque and J.J.Weis, Phys. Rev.E 49 5131 (1994)
17. A.Satoh, R.W.Chantrel, S.Iand G.N.Coverdall, Coll., Intersace Sci., 181, 422
(1996)
18. A.Zubarev, S.Odenbach, J. Fleicher J.Magn. Magn.Mat. (submitted as material of
IX Int. Conf. on Magnetic Fluids)
19. S.Odenbach and H.Stork, J.Magn.Magn.Mat. 183, 188 (1998)
20. P.G. de Gennes and P.A.Pincus, Phys. Condens. Mat. 11, 189 (1970)
21. P.G.Jordan, Molecular Physics, 25, 961(1973); ibid. 38, 769 (1979)
22. A.Cebers, Magnitnaja Gidrodinamika, 2 36 (1972)
23. A.Yu.Zubarev, L.Yu. Iskakova, J.Exp. Theoret. Physics, 80, 857 (1995)
24. A.Yu.Zubarev and L.Yu. Iskakova, Phys. Rev. E, 61, 5415 (2000)
25. M.A.Osipov, P.I.C.Teixeira and M.M. Telo da Gama, Phys.Rev.E., 54, 2597 (1998).
M. Tavares, M.M. Telo da Gama and M.A.Osipov, Phys. Rev. E., 56, 6252 (1997),
see, also, Errata, Phys. Rev. E, 57 7367 (1998)
26. K.Morozov and M.Shliomis see p.204 of this volume.
27. A.O.Tsebers, Magnetohydrodynamics, 1 3, (1994)
28. A.Yu.Zubarev and A.O.Ivanov, Phys. Rev. E, 55,7192 (1997); A.O.Ivanov and
A.Yu.Zubarev, Phys. Rev.E, 58,7517 (1998)
29. J.M.Tavares, J.J.Weis and M.M.Telo da Gama, Phys. Review E, 59, 4388 (1999)
30. M.Rasa, D.Bica, A.Philipse and M.Vekas, “Dilution Series Approach for Inves-
tigation of Microstructural properties and Particle Interaction in High Quality
magnetic Fluids” (will be published)
31. V.N. Pokrovskij, Statistical mechanics of Dilute Suspension, (Nauka, Moscow,
1978)
32. M.A.Marstenjuk, Y.A.Raikher and M.I.Shliomis, Zh.Eksp. Teor Fiz 65 834 (1974)
[JETP, 38, 413 (1974)]
33. D.D.Josef, G.S.Beavers and R.L.Fosdick, Arch. Rational Mech. Annals, 49, 381
(1973)
34. K.Melzner and S.Odenbach, J. Magn. Magn. Mat, (submitted as material of IX
Int. Conf. on Magnetic Fluids)
35. H.Muller and M. Liu see page 115 of this volume
36. S.Klapp and F.Forstmann, Phys. Rev. E., 60, 3183 (1999).
37. A.O.Ivanov and S.Kantorovich, J.Magn.Magn.Mat. (submitted as material of IX
Int. Conf. on Magnetic Fluids)
Magnetic Fluid
as an Assembly of Flexible Chains
Konstantin I. Morozov1 and Mark I. Shliomis2

1
Institute of Continuous Media Mechanics, UB of Russian Academy of Sciences,
614013 Perm, Russia
2
Department of Mechanical Engineering, Ben-Gurion University of the Negev,
P.O.B. 653, Beer-Sheva 84105, Israel
Abstract. Dipolar chains formed in magnetic fluids out of colloidal magnetic grains
have much in common with polymer molecules. An investigation of spatial and ori-
entational intrachain correlations and elucidation of an important role of the chains
flexibility lead us to natural and fruitful extension of basic concepts of polymer physics
to the case of dipolar chains. Conformations of the chains (statistical coil, globule) in
zero and infinitely strong external magnetic field are studied and the possible coil–
globule phase transition is predicted and discussed.
1 Introduction
The phenomenon of formation of internal structure of magnetic fluids (MF) and
its influence on the macroscopic properties of MF is one of the most exciting
problems of physics of ferrocolloids. It is generally recognized that the origin of
internal structure is due to dipolar interactions between magnetic grains com-
posed MF. The basic features of the effects of dipolar interactions in MF were
outlined long ago in the pioneering work of de Gennes and Pincus [1]. The authors
had brilliantly foreseen the two possible consequences of magnetic interparticle
interactions. First, in the zero external field they predicted the phase separa-
tion of a magnetic fluid into the dilute vapor and condensed liquid phases below
some critical temperature in analogy with the case of simple fluids. A heuristic
explanation of the phenomenon is following. The dipole 1/r3 pair potential (r
the particle separation) being averaged over the dipole orientations becomes an
attractive 1/r6 -potential responsible for the van der Waals condensation of sim-
ple fluids. Second, at high external fields H the association phenomenon should
appear: the dipoles align along the field direction, the head-to-tail configuration
is energetically preferable and the magnetic grains tend to form chains along
the direction of H. Thus, the two types of effects of interparticle interactions
in MF differ by their scales: the condensation manifests itself as macroscopic
phenomenon whereas the grain association as microscopic one. De Gennes and
Pincus discussed also the appearance of chain-like structure in zero external
field. However, the relation between the condensation and the chain formation
had remained unclear in this case. As we will see, just this question became the
subject of the latest intensive investigations.
In the 70-s, the task of condensation of dipolar fluids in the absence of external
field seemed to be well understood. All the theories of the system of interacting

Magnetic Fluid as an Assembly of Flexible Chains 163
dipoles predicted the appearance of the gas-liquid phase transition under some
critical temperature (see, e.g. [2] and references therein). Among the theories
we note the analytical mean spherical model (MSM) of Wertheim [3] for the
system of hard sphere dipoles. The thermodynamics for MSM was developed
in [4] and the critical value of the coupling parameter λ = m2 /d3 kB T (m is
dipole moment, d is a grain diameter, kB is Boltzmann’s constant and T is the
temperature) was determined. According to [4], this value is λc = 4.445. The
condensation observed in Monte Carlo simulations [5] qualitatively corroborated
the theoretical results.
The condensation of magnetic fluids in zero field as well as in external mag-
netic fields is reliably established experimentally. Starting from the probably first
evidence, given by Hayes [6], it was observed in a numerous works for both types
of MF - ionic [7,8,9] and surfacted ones [10,11,12,13,14]. It has been shown that
an external magnetic field always promotes the phase transition. The droplets
of concentrated “liquid” phase appear among the more dilute “gas” phase with
decreasing of temperature or/and with growth of magnetic field. The real MF
are polydisperse ones. So, there is no simple to determine experimentally the
critical values λc (H) of the beginning of the phase separation. The estimation of
the value in strong magnetic filed is λc (∞) = 2.96 [14]. In zero field the droplets
of concentrated phase are spherical what means the absence of their sponta-
neous magnetization. The droplets have typical sizes of a few microns, i.e., they
are macroscopic formations. Just the macroscopic scale of the droplets allows
to identify them as new concentrated phase originating in the system. So, no
wonder that the elongation of a droplet in external field is well described by the
competition of the surface energy at the gas-liquid boundary and the magnetic
energy of the droplet [7].
To describe the observable condensation of MF under influence of external
magnetic field the statistical mechanics models [15,16,17,18,19] were proposed.
Initially the dipolar interactions were taken into account within the framework
of the classical Weiss model of ferromagnets [15,16] and antiferromagnetic model
[17]. The both models quantitatively correctly describe the phase separation of
MF however they possess the defect common for all the mean field theories. It
is a prediction of spontaneous magnetized state of concentrated liquid phase in
the absence of external field. This property was never observed in experiments
with real MF. In contrast with the mean field theories, the more late models
[18,19] took into account the orientational and spatial correlations of MF grains.
In [18] the generalization of the MSM of Wertheim [3] onto the case of arbitrary
external field was done. The mean spherical model has no the problem with
spontaneous magnetized state of liquid phase [4]. The external magnetic field
lowers the threshold of condensation from zero field value λc = 4.445 to infinite
field value λc (∞) = 3.055 [18] what is in agreement with mentioned estimation
[14]. The analogous results were obtained within the framework of perturbation
theory [19] which is more simple than MSM. The both theories describe perfectly
the many properties of real MF such as the magnetization curve, temperature
and concentration dependence of initial susceptibility [20,21,22].
164 K.I. Morozov and M.I. Shliomis
Here it is necessary to note that the usual magnetic fluids used in practice
are required to be stable against temperature and magnetic field action. In other
words, in many cases the condensation of MF proves to be the undesirable phe-
nomenon. Thus, the ordinary magnetic fluids and commercial MF especially are
characterized by very low values of dipolar interactions, λ ≤ 1. Often due to the
special techniques only, the authors could increase value of λ up to 3 ÷ 4 and ob-
serve the phase transition in MF [7,8,9,13,14]. As we will see, under these small
values of λ the association phenomenon – the second consequence of dipolar
interactions according to de Gennes and Pincus – is negligible. Another aspect
of dealing for a long time with mentioned MF is the problem of synthesizing
of MF with strong magnetic interactions. Only recently such ferrocolloids were
successfully synthesized [23]. Now Nakatani and co-authors achieved the values
of the coupling parameter λ = 3 ÷ 10 [24]. Just in such MF the association of
grains should be essential. In contrast to condensation, to detect the particle
association taking place on microscopic level is a serious experimental problem.
We can judge about MF microstructure using only indirect data on negative
viscosity effect and magneto-vortical resonance in MF under an oscillating mag-
netic field [25]. The data gave sure signals of the presence of a wide spectrum of
magnetization relaxation times τ instead of only one Brownian time for single
magnetic grains. Assuming that this spectrum originates from chains formed out
of the grains, authors of [25] have reached a perfect agreement between the the-
oretical and experimental frequency dependencies of MF viscosity and dynamic
susceptibility.
The intense interest to the systems with strong dipolar interactions arose
however a bit earlier, at the beginning of 90-s, as a corollary of new unexpected
data of numerical experiments. First, a surprising result of Monte Carlo simu-
lations was the absence of vapor-liquid transition in the system of interacting
dipoles. Instead, simulations revealed a phenomenon of formation of chains of
head-to-tail aligned dipoles in the dipolar hard sphere fluid [26,27] as well as
in the Stockmayer [28] and dipolar soft sphere fluid [29,30]. The evidence of
the above transition did was given after all [31]. This transition proved to be
rather unusual, namely, it occurs between highly associated vapor phase and a
more normal dense liquid phase and accompanies by extremely small entropy
and enthalpy changes.
Another exciting property of dipolar fluid revealed in simulations is that
dipolar forces alone can create an orientationally ordered liquid state. Starting
from the first demonstration of the phenomenon in molecular-dynamics simula-
tions [32,33], this conclusion was confirmed later by Monte Carlo calculations
[26,34,35,36]. In fact, there is no simple connection of spontaneous order of dipo-
lar fluid and chain association phenomenon of aligned dipoles [35,37]. However
the contraposition of liquid condensation and association feature seemed to be
well established in the numerical experiments [26,27,28,29,30,31]. At the same
time, the data of Monte Carlo simulations made in [38,39,40] with different
conditions at the boundary of simulation cell similarly to old study [5] fully con-
tradict to results [26,27,28,29,30,31,32,33,34,35,36,37] and predict at λ > 3 the
traditional gas-liquid phase transition in the course of which non-chain-shaped

clusters of magnetic grains are formed.
It is clear that these breathtaking simulation data initiated the new theo-
retical investigations of association phenomenon in dipolar fluids. We recall for
readers that there are two basic statistical mechanics approaches to the problem
of interparticle interactions. They are the technique of correlation functions and
the association Frenkel-Band theory. The former is based upon the solution of
the Ornstein-Zernike equation under some reasonable closure relation for the
direct correlation function [2]. Such an approach was realized for the case of
MF in [1]. Due to necessity to solve the integral equation, it is very complicated
for analysis. Moreover, it contains only implicit information about structure of
MF. What is why this approach is rarely used in theoretical works. The Frenkel-
Band association theory treats the statistical system of interacted particles as
chemically equilibrium mixture of non-interacting clusters of different size and
the interparticle interactions are taken into account only inside clusters [41].
For the case of MF in strong magnetic field, this approach was realized in the
pioneering works of Jordan [42,43]. In contrast to the correlation function ap-
proach, the microstructure of MF is a straightforward result of the association
theory. Of course, the both theories do not give the identical results. However,
they should have the same limit of low-concentrated MF what follows imme-
diately from nature of the approximations [41]. Nevertheless, Jordan confirmed
the main prediction of de Gennes and Pincus theory [1] about formation of mag-
netic grain chains parallel to external field. [It is interesting to note that the
impulse for Jordan to develop his previous study of microstructure in MF [43]
was the experimental data [6,10] where the mentioned condensation (not asso-
ciation of grains!) of MF into liquid droplets took place. As we have seen, the
condensation and particle association are different phenomena connected with
interparticle interactions. But even now both phenomena are confused with each
other [44]].
In the papers Jordan took into account all the interparticle interactions in
a chain of grains. Then, calculating the partition function of a chain, he had to
make some additional assumptions about spatial and orientational configuration
of dipoles. The final results are very cumbersome even in limiting cases of infinite
external field and strong dipolar interactions, λ 1. The most important on our
mind conclusion of the Jordan’s calculations is the account of dipolar interactions
of non-nearest neighbors along the chain is reduced in fact to re-normalization of
the nearest-neighbors interactions: λ → λζ(3), where ζ(3) = 1.202 is the Riemann
ζ−function. Thus, it is possible to exclude the interactions of the non-nearest
neighboring particles without any significant loss of information about the sys-
tem. Just this reasonable approximation within the framework of association
theory became the basis for latest theoretical works [45,46,47,48,49,50,51,52,53].
The critical review of number of these papers is given in the next Section where
the association theory is considered in detail. Here we will mention only the main
general results of cited papers.
The phenomenological theory [47], combined the van der Waals theory for
liquid condensation with the association Jordan’s theory, revealed a competition
between liquid condensation and chain formation driven by anisotropic dipo-
lar interactions. The analogous result was obtained in [46,48,49,50] where some
phenomenological parameters of work [47] were estimated. An interesting con-
clusion of theoretical studies [48] is that a validity of association theory of non-
interacting chains is possibly broader than it was assumed before [42,43] and is
not limited only to the low-density phase of fluid. This result is indicated by
the appropriate evaluation of the steric and dipolar interchain interactions [48].
Finally, authors [45,53] studied the magnetic and rheological properties of MF
assuming the rod-like form of aggregates of magnetic grains.
Summarizing the review of recent theoretical works we see that all of them
deal with only one characteristic of the particle chains - their length. Meanwhile
a genuine understanding of the phenomenon of chain formation remains incom-
plete without studying the spatial and orientational correlations inside dipolar
chain. As shown below, the association theory allows to investigate this aspect
of the problem in detail and often analytically. Developing our approach, we es-
tablish an important role of chain flexibility; this attribute of dipolar chains has
been remained without proper attention in previous works [45,46,48,49,51,52,53].
De Gennes and Pincus [1] were the first who noted an analogy between dipolar
chains and polymer molecules. Actually, our investigation of the pair correla-
tions and the chain flexibility leads us to natural extension of the basic concepts
of polymer physics like the persistent length and Kuhn segment, coil, globule
[54] etc. to the case of dipolar chains. The aim of this paper is study of the spa-
tial and orientational correlations inside dipolar chains, determination of possible
forms of the chains (statistical coil, globule), and investigation of the coil-globule
phase transition. The paper is organized as follows. In Sect. 2 we consider the
basic statements of an association theory of MF, define the concept of chain of
magnetic grains, establish the main property of the proposed model of neighbor-
ing grain interactions and determine the averaged length of dipolar chains. The
statistical properties of a single dipolar chain are studied in Sect. 3 where the
important role of chain flexibility is established and main flexibility character-
istics are determined. The influence of chain flexibility on structure of a chain
and coil-globule transition are considered in Sect. 4. We make some concluding
remarks in Sect. 5.
2 Association Theory of Magnetic Fluid

2.1 Definition of a Chain of Grains
Let us consider MF as an assembly of chains of different length formed out of

magnetic grains in the absence of external field or in infinitely strong magnetic
field. We assume that all the particles are identical spheres of diameter d, only
neighboring grains in the chain interact with each other, and there is no inter-
action between grains of different chains. The interaction neighboring particles
with numbers i and i+1 is the sum of hard-sphere UHS (ri,i+1 ) and dipole-dipole
Udd interactions:
m2
U (i, i + 1) = UHS − 3 [3(ei ni,i+1 )(ei+1 ni,i+1 ) − ei ei+1 ] , (1)
ri,i+1
where m is the magnetic moment, ei and ni,i+1 are unit vectors along the dipole
of particle i and the interparticle vector r i,i+1 , respectively.
Let gN be the number of N -particle chains per unit volume. The volume den-
sity F/V of the free energy of system includes ideal gas term and the intrachain
contribution [45,46,48]
∞
!
F gN v
= gN ln + gN fN , (2)
V kB T e
N =1
where v = πd3 /6 is the particle volume, fN = − ln ZN is the dimensionless

“internal” free energy of N -particle chain, ZN is its partition function.
Now it is necessary to give the definition of a chain. First, pair of grains are
defined as bonded if their dipole potential for the head-to-tail configuration of
dipoles exceeds thermal energy, i.e., the condition 2m2 /r3 kB T ≥ 1 is fulfilled.
Second, calculating the partition function ZN of N -particle chain, the integration
over displacement vector r i,i+1 of two neighboring particles should be done over
half of spherical volume admitted by mentioned above condition.
Let us comment both statements. Our choice of the threshold energy for
cluster formation is the same as in [29,45]. Of course it is not unique and another
definitions are also possible [26,30,34,51]. We note that this ambiguity takes
place for finite values of λ ≥ 1 and disappears when λ 1. The latter case of
large λ was the main approximation in [1,42,43,48,49]. In our consideration it
is supposed that λ can vary from 0 to 10. What is why we had to make more
precise the definition of a chain of grains and used for it mentioned physical on
our mind limitation. As for the procedure of integration, this method proposed
in [45] allows effectively to take into account the steric interactions with another
grains in a chain. Of course, the given integration narrows down the real region of
space admitted for the particle in a chain. However, it contains the main region
around head-to-tail configuration of dipoles. The results of exact integration
and integration over half of spherical volume prove to be practically coinciding
starting from λ ≥ 3, that is from the same beginning of the grain association
as we will see later. In order to show the chosen method of integration we use
below the prime near the symbol of integral.
Let us write the expression for partition function of N -particle chain in the
absence of external field
N −1
N N −1
! Udd (i, i + 1) + dei + dr i,i+1
ZN = exp − , (3)
i=1
kB T i=1
4π i=1 v
In spite of the clearness of this relation it is necessary to discuss one question
of principle connected with (3). The point is that authors [51,52] unexpect-
edly introduced in the denominator of (3) the symmetry factor N ! “missed” in
their previous considerations [48,49] as well as in [42,43,45]. They believe that

such factor takes into account the indistinguishability of grains due to fact that
magnetic particles diffuse through all the chains in the course of Monte Carlo
calculations [35]. This mistaken statement has very long history. The exhaustive
consideration of the question was given by Ehrenfest and Trkal in 1920 [55]. The
detail information on this subject can be found in [56]. Here we only note that
the presence of symmetry factor is not connected absolutely with simulation
data because it is determined by writing of the Hamiltonian used in the theory.
The authors [48,49,51,52] used the same Hamiltonian as we did (see Eq. (5) in
[48] and Eq. (19) in [51]), i.e., as a sum along chain of dipolar interactions (1)
between consecutive particles. It means that all the particles in the chain with
N ≥ 3 are numbered and particle with number i always follows beyond grain
with number i − 1. By the way, this grain numbering was taken into account by
authors [48] explicitly (see Sect. II in [48]). Due to this numbering of all grains in
the chain with N ≥ 3 there is no need in additional factor N ! [55,56]. Thus the
revising in [51,52] the previous approach [48,49] is incorrect. Nevertheless, there
is the special case when the symmetry factor is necessary. It is the case of dimers,
the clusters of two grains only. The phase volume of dimer is two times higher
than it should be. Thus, due to mentioned indistinguishability of the grains the
partition function Z2 indeed should be divided by factor 2. However this factor is
taken into account in our consideration automatically by the mentioned method
of integration over half of spherical volume. Therefore, equation (3) do not have
any additional factor at all.
Let us return to analysis of (3). The main advantage of proposed model
is the factorization of the chain partition function – ZN . Such a factorization
takes place in the considered cases of zero and infinite external magnetic field
but not for arbitrary field. This property of dipolar chain with interactions of
neighboring grains is quite analogous to that for the chain of spins with the
Heisenberg interactions [57]. In the case of zero field, it is immediately followed
from the absence of a preferable direction in the system. Indeed, calculating ZN
one should integrate over variables ei , r i−1,i starting from the last particle with
number N (“tail” of the chain) to the first one (“head” of the chain) and taking
at each step the direction ei−1 of previous grain as polar axis. The other essential
fact is independence of the integral

m2 de2 dr
I= exp 3
(3(e 1 n)(e2 n) − e1 e2 ) , (4)
r kB T 4π v
on the orientation e1 of previous particle. It is clear that partition function Z2

of a dimer coincides with I. We also note that at large values of λ the integral
I is also the absolute value of the second virial coefficient

1 de1 de2 dr
b2 = − {exp[−U (12)/kB T ] − 1} . (5)
2 4π 4π v
Here is the dimensionless value of b2 in units of particle volume. The integral
is without of prime, i.e., the integration is fulfilled over all space. Finally we
achieve
ZN = Z2N −1 . (6)
The factorization of the problem in the case of infinitely strong magnetic field
is obvious.
The mentioned above setting of the problem is considered in number of works
[46,48,49,51]. As we know, the factorization was noted only in [46]. Unfortunately,
the results of this work are most likely mistaken (see below). Now we can cal-
culate the distribution gN of chains upon their length. Substituting (6) into (2)
and using the condition of chemical equilibrium between chains of different sizes
[42,43] or equivalently the minimization of the free energy of the system [45,48]
we find
gN = xN /vZ2 , (7)
)∞
where parameter x is expressed via normalization condition N =1 N gN = n (n
is the concentration of magnetic grains in MF) as
x
= ΦZ2 . (8)
(1 − x)2
Here Φ = nv is the volume fraction of magnetic particles. From (7) and (8) we
finally determine the average number of particles in the chain
,
1 1
N = + + ΦZ2 . (9)
2 4
We see that owing to factorization the problem of determination of the av-
erage length of the chains is reduced to calculation of the partition function of
dimer. The asymptotic representation at λ 1 of the dimer partition function
in zero and infinite external field is

e2λ 8 23 229 5263 11536 57427
Z2 (0) = 3 1 + + + 3+ + + , (10)
3λ 3λ 3λ2 9λ 54λ4 27λ5 27λ6

e2λ 5 41 155 11195 39235 628145
Z2 (∞) = 2 1 + + + + + + . (11)
3λ 3λ 12λ2 18λ3 432λ4 432λ5 1728λ6
These expansions are given up to terms of order of O(λ−6 ) because of the
slow convergence of the series. The main (first) term in the right side of (10)
and (11) is a result of de Gennes and Pincus [1]. The other terms are new.
We note that in [48] it was informed about mistaken value of the second virial
coefficient b2 (or Z2 ) in [1]. This is however misunderstanding: the authors [48]
used for b2 the different notation than de Gennes and Pincus did. The calculated
dependencies of both partition functions are shown in Fig. 1; note that in the
figure and below the minimal value of λ is 0.5 according to our definition of a
chain.
It is interesting that maximal deviations of calculated values from unity (i.e.,
from de Gennes and Pincus asymptotic values) takes place at small λ ∼ 2 ÷ 3
7
Z 2 (0)
2 Z 2 (∞)
1
0
0 2 4 6 8 λ
Fig. 1. Partition functions of dimer in zero and infinite external field in units of de
Gennes and Pincus asymptotic values e2λ /3λ3 and e2λ /3λ2 , respectively
however even at λ = 10 it achieves 50%. The expansions (10) and (11) describe
the calculated values within the accuracy of 10% starting from λ ∼ 5 ÷ 6.
Let us analyze relation (9) in some limiting cases. When λ 1 we recover the
result for average chain length in zero field [48,49]. Next we consider the case of
low-concentrated magnetic fluid with strong dipolar interaction so as Φb2 1.
As it was noted above, both main theories should have the same limit of low
concentration. From (9) and (5) it follows N = 1 − Φb2 (b2 is negative). At the
same time the theory of de Gennes and Pincus [1] predicts N GP = 1 − 2Φb2 .
The origin of double difference of the coefficient is clear. In the approach [1] the
steric interactions inside the chain were ignored. It results in double exceeding
of contribution of neighboring particles interaction in partition function of a
chain. As we mentioned above the factorization of the considered problem was
noted in [46] where the grand partition function of the system was calculated.
The author obtained for variable equivalent to our x the cubic expression (see
equation (6) in [46]) instead of quadratic one (8). The character of Letter does
not allow to check all the calculations made in [46]. Nevertheless, we argue that
there is a mistake in his equation (5), containing the redundant coefficient 2.
Without this factor the equation (6) in [46] reduces to the correct relation (8).
We can support our point of view on [46] also a posteriori. Indeed, the relation
(8) is a straightforward consequence of our formalism based on calculation of the
partition function of a chain. It is clear that the final result should not depend
on the choice of thermodynamic potential considered.
In this Section we developed the association theory of ideal chains of mag-
netic grains. The size distribution of dipolar chains obeys to equations (7) and
(8). The average length of chains (9) increases ∼ Φ exp(2λ) under small values
√
of this parameter and ∼ Φ exp(λ) at high λ. The average value N in zero
and infinite external field can easily calculated according to (9)-(11) and data
shown in Fig. 1. Being the very important integral characteristics of the phe-
nomenon of chain formation, the average length itself however does not contain
an exhaustive information about system. Our understanding of the phenomenon
would be incomplete without study of statistical properties of the chains, with-
out their spatial and orientational correlations. Just these questions we start to
investigate.
3 Statistical Properties of Ideal Chains

3.1 Correlation Functions of Dimers
Here we will see that the factorization is a general property of the considered
model what allows to express any interparticle correlations in terms of spatial
and orientational correlations inside dimers. First, let us define four orientational
correlation functions
e1 r 12 (e1 r 12 )2
A = e1 e2 , B = , C = (e1 e2 )2 , D = , (12)
d d2
and two spatial characteristics for a dimer:
r12 r12
2

E= , F = 2
. (13)
d d
Averaging in these expressions is fulfilled over variables r 12 and e2 of second
particle whereas the orientation e1 of first grain is taken as polar axis. The polar
angle between e1 and the displacement vector r 12 takes the values from zero to
π/2 according to method of integration mentioned at the beginning of Sect. 2.
Due to such integration over spatial variable the correlation function B is not
zero. It is clear that in considered cases of zero and infinite external field all
the correlations (12) do not depend on the orientation e1 . The correlations (12)
together with (13) are dimensionless functions upon λ and easily calculated. For
the case of zero external field they are depicted in Fig. 2. We note that the curve
B lies higher the curve D. The maximal size of dimer (see curves E and F ) suits
to intermediate values of λ. Indeed, when λ ≤ 1 the cluster of two grains falls
into separate particles even for very small distance between them, r ∼ d. At high
values of λ, the mean interparticle distance r is of the order of d as well due
to strong dipolar interactions. We note also that r never exceeds 2d (in fact,
it is always less than 1.4d) what means the numbering of the grains in a chain
and verifies the chosen in Sect. 2 method of integration over space coordinate.
All the introduced dimer’s variables tend to unity with increasing of coupling
parameter. Their asymptotic representation at λ 1 is
2 2 20 266 4241 77669
A=1− − − 3− 4− − , (14)
λ λ2 3λ 9λ 27λ5 81λ6
2 .0
F
1 .5
E
1 .0
B
D
0 .5
C
A
0 2 4 6 8 λ
Fig. 2. Correlation functions (12) and spatial mean values (13) of dimers in zero field
versus the coupling parameter λ
1 5 11 689 15593 229265

B =1− − 2− 3− 4
− − , (15)
2λ 9λ 6λ 81λ 324λ5 729λ6
4 4 8 74 998 16880
C =1− + 2 + 3 + 4 + + , (16)
λ λ 3λ 9λ 27λ5 81λ6
1 7 43 397 20185 398579
D =1− − − − − − , (17)
λ 18λ2 27λ3 54λ4 486λ5 1458λ6
1 4 44 1175 18007 153715
E =1+ + 2+ 3
+ 4
+ + , (18)
6λ 9λ 27λ 162λ 486λ5 729λ6
1 17 97 2657 41507 719573
F =1+ + + + + + . (19)
3λ 18λ2 27λ3 162λ4 486λ5 1458λ6
Within accuracy of a few percents these formulas describe the calculated values
A, B, C when λ ≥ 5 and D, E, F for λ ≥ 7. We note that the terms ∼ λ−1
for A, B and E were found in fact in [42]. It is interesting, that even for the
strong dipole interactions the correlations are still far from unity. For example,
for λ = 10 the mean values of cosine and its square of dipole orientations are
A = 0.77 and C = 0.64, respectively. It means that orientations of neighboring
dipoles preserve a fairly high rotational mobility inside dimers.
Up to now we assumed that the external field is absent. Now, let us consider
the opposite case of infinitely strong external field. The dimer correlations A∞ ,
B∞ etc. in infinite field differ from their zero-field values. Evidently, A∞ = C∞ =
1. The calculated values of four remaining variables are depicted in Fig. 3. We
indicate also their asymptotic expansions. They are
7 25 1787 1000 715597
B∞ = 1 + + + + + , (20)
36λ2 36λ3 648λ4 81λ5 11664λ6
2.0
F∞
1 .5
E∞
1.0
B∞
D∞
0 .5
0 2 4 6 8 λ
Fig. 3. Correlation functions B∞ , D∞ and mean values E∞ , F∞ of dimers in infinite

field versus the coupling parameter λ
4 85 343 56255 106571

D∞ = 1 + + + + + , (21)
9λ2 54λ3 54λ4 1944λ5 729λ6
1 13 119 2675 34801 508025
E∞ = 1 + + 2
+ 3
+ 4
+ 5
+ , (22)
6λ 36λ 108λ 648λ 1944λ 5832λ6
1 7 67 3109 10366 2472103
F∞ = 1 + + + + + + . (23)
3λ 9λ2 27λ3 324λ4 243λ5 11664λ6
The results of calculation on formulas (20)–(22) with accuracy of 3÷5 % coincide
with exact values of B∞ , D∞ , E∞ at λ ≥ 5. The accuracy of (23) is worse: the
mentioned difference is achieved at higher λ ≥ 7. As before, we note that terms
∼ λ−1 in (20) and (22) were determined in [42].
Six dimer variables (12) and (13) contain all physically significant statistical
information about the system. In order to verify this property it is necessary to
consider the characteristic intrachain correlations.
3.2 Calculation of Intrachain Correlations

For now we assume MF to be in zero external field till the opposite case of infinite
field will be considered. First, we calculate two correlations typical of the system:
ei;α ej;β and ri,i+1;α rj,j+1;β . Here and below we denote the numbers of grain
in a chain by Latin letters whereas the Greek ones indicate the components
of orientation ei and displacement r i,i+1 vectors. The symbol ... means the
averaging over spatial and orientational coordinates of all particles in a chain
excepting the first particle – the “head” of a chain. The orientation e1 of first
grain is assumed to be fixed. In the case of zero external field, it determines a
single preferable direction in the system and plays a role of external field for
other particles in the chain. We average over the variables of particles starting
from the “tail” of a chain, i.e., from N th grain and then going to the grains with
number N − 1, N − 2 etc. At each step we take the direction of previous grain
as polar axis. Due to mentioned independence of the integral (4) as well as the
dimer variables (12) and (13) on orientation of previous (first) particle, at such
step of averaging we obtain one of the dimer functions Z2 , A, B, C, D or F .
Just this signifies the factorization of any two-particle correlation.
a. Calculation of ei;α ej;β
Let us assume for definiteness that 1 ≤ i < j ≤ N . The integration over variables
of grains N , ..., j + 1 is obvious: it reduces to multiplying by unity, ...N,...,j+1 =
1. Here and further we indicate explicitly the numbers and the order of particles
been integrated. Integration over particle j in the coordinate system connecting
with particle j − 1 gives
ei;α ej;β j = Aei;α ej−1;β . (24)
Repeating this procedure j − i times we find
ei;α ej;β j,j−1,...,i+1 = Aj−i ei;α ei;β . (25)
Now we consider mean value ei;α ei;β which looks like initial quantity, but with
j = i. Again we connect the coordinate system with particle i − 1 and integrate
over variables of grain i. The result is
1−C 3C − 1
ei;α ei;β i = δα,β + ei−1;α ei−1;β , (26)
2 2
where δα,β is a unit tensor. The last expression is a recurrent one. After averaging
over grains with numbers i − 1, i − 2, ..., 2 we have
1
ei;α ei;β i,i−1,...,2 = δα,β + T i−1 Δα,β , (27)
3
where the notation T = (3C −1)/2 is used and the irreducible second rank tensor
Δα,β is introduced according to relation
1
Δα,β = e1;α e1;β − δα,β . (28)
3
Finally, combining expressions (25) and (28) the considered mean value is found
in the form
1
ei;α ej;β = Aj−i δα,β + T i−1 Δα,β . (29)
3
b. Calculation of ri,i+1;α rj,j+1;β

There are two different cases, j = i and j > i, which should be considered
separately. First let us put j = i. We take ei as polar axis and integrate over
coordinates of grain i + 1 (cf. with (26))

F −D 3D − F
ri,i+1;α ri,i+1;β i+1 = δα,β + ei;α ei;β d2 . (30)
2 2
Using here equation (27) we deduce

F 3D − F i−1
ri,i+1;α ri,i+1;β = δα,β + T Δα,β d2 . (31)
3 2
Second, let be j > i. In this case the integration over particle j + 1 gives
ri,i+1;α rj,j+1;β j+1 = Bdri,i+1;α ej;β . (32)
Averaging the right side consecutively over particle with numbers j, j − 1,...,i + 2
we can write
ri,i+1;α rj,j+1;β j+1,j,...,i+2 = Aj−i−1 Bdri,i+1;α ei+1;β . (33)
Integrating now over variables of particles i + 1,...,2 we finally obtain

1
ri,i+1;α rj,j+1;β = Aj−i B 2 d2 δα,β + T i−1 Δα,β , j > i . (34)
3
The expression (29), (31), (34) are the basic mean values needed for calcula-
tion of any two-particle intrachain correlation.
3.3 Persistent Length of Dipolar Chains

We use the developed formalism for calculation of some characteristic correla-
tions typical of dipolar chain. First, let us consider the correlation ei ei+k of
orientations of dipoles i and i + k in the chain. Due to relation (29), this value
is multiplicative, i.e.,
ei ei+k = Ak , (35)
what means its exponential decay. From here, using an analogy with polymer
theory [54], we determine the persistent length (in units of particle diameter d)
as Lp = −E/ ln A. This quantity characterizes the flexibility of the chain. It
is shown below in Fig. 4. The formulas (14) and (18) allow to determine the
asymptotic behavior of the persistent length up to terms of order λ−4
E λ 11 23 128 9698 1423259
Lp = − = − − − − − . (36)
ln A 2 12 18λ 27λ2 405λ3 9720λ4
The asymptotic representation well describes the calculated values of Lp starting
from λ ≥ 6. The evaluation of persistent length for λ = 12.25 gives Lp = 5.1
what is in reasonable agreement with the simulation data Lp ∼ 7 mentioned in
[35]. As seen, the persistent length is rather short. Thus, dipolar chains may be
considered as rigid ones only on small scales not exceeding λ/2.
3.4 The Kuhn Segment of Dipolar Chains

The most
)N −1informative quantity of chain configuration is “end-to-end” vector
R = i=1 r i,i+1 connecting centers of first and last particles in a chain of N
grains. We calculate the mean values R2 and R⊥
2
along the orientation e1 of
one end of chain and in perpendicular direction. Using (29), (31) and (34) after
cumbersome but simple calculations we obtain analytically
R2 2AB 2
= F (N − 1) + [N − 2 − AGN −2 (A)] , (37)
d2 1−A
R2 − R⊥
2
3D − F 2AB 2
= G N −1 (T ) + [AGN −2 (A) − T GN −2 (T )] , (38)
d2 2 A−T
where R2 = R2 + 2R⊥ 2
and the function GN (x) is determined as GN (x) =
(1 − x )/(1 − x).
N
Let us analyze the result (37), (38). In the case of dimers, N = 2, the obvious
result R2 /d2 = D, R⊥ 2
/d2 = (F − D)/2 is recovered. When the chains are
very short, N λ, we have R2 = N 2 d2 , R⊥ 2
= 0. This is just the case
of rod-like aggregates considered in [45,53]. This case, however, is never put
into effect in reality. The fact is that both the energy gain accompanying every
pairing of magnetic grains (∼ λkB T ) and the chain correlation length (∼ λd)
are determined by the same coupling constant λ. Therefore, the mean number of
grains in a chain N as well as its correlation length Lp increases with enlarging
of λ, so that the condition N λ is never satisfied. [From the formal point of
view, N increases with growth of λ exponentially (see (9)) whereas Lp ∼ λ/2
according to (36)]. Statistically, some number of very short chains (N λ) are
certainly existed, but their fraction is negligible. We note that recently authors
[45,53] changed their own previous representations and admitted the deformation
of dipolar chain [58].
For a long chain equations (37) and (38) give (R2 − R⊥ 2
)/R2 = 0 and

2AB 2
R = N d
2 2
F+ , N →∞. (39)
1−A
Such a N -dependence of the mean-squared end-to-end distance of a long chain
is typical of ideal polymer chain [54]. Its local stiffness is often characterized
– along with the persistent length Lp – by the so-called Kuhn segment LK
determined by the relation R2 = N rLK d, where r is the mean distance
between neighboring grains in the chain. Both the quantities, Lp and LK , are
depicted in Fig. 4 as functions of the dipole parameter λ. When it is sufficiently
large, we have

2AB 2 19 2 715 8377 332995
LK = F + E −1 = λ − + − − − . (40)
1−A 6 9λ 108λ2 216λ3 1296λ4
As before, this expansion perfectly describes the calculated values of LK at
λ ≥ 6. It is interesting that starting from λ 5 (i.e., just when the aggregation
3
LK
2
Lp
1
0
0 2 4 6 8 λ
Fig. 4. The persistent length Lp and the Kuhn length LK as a function of coupling
parameter λ
phenomenon becomes appreciable) the ratio LK /Lp is close to 2 as it takes place

for many models of polymer molecules [54].
So, the linear along chain memory disappears at the distances of a few parti-
( with number of grains N LK have practically
cles and the long dipolar chain
the same characteristic sizes R2 and R⊥ 2 . The long chain resembles cloud
or coil of connected monomers and has quasi-spherical form. In fact however this
conclusion proves to be valid even for sufficiently short chains with N ≥ LK (see
Sect. 3.5).
3.5 Case of Infinite External Field

Starting from Sect. 3.2 we assumed that the external field is absent. Now, let
us consider the opposite case of infinitely strong external field. As noted above,
the factorization of the problem in infinite field takes place as well as in zero
external field. Thus, the general results (38) and (39) remain valid owing to the
factorization only. Setting A∞ → 1 and C∞ → 1 we find the characteristic sizes
of chain along the field and in perpendicular direction
2
R2 = (N − 1) B∞ (N − 2) + D∞ d2 , (41)
1
R⊥
2
= (N − 1)(F∞ − D∞ )d2 . (42)
2
The size of a chain along√ field increases as ∼ N whereas the transverse size
remains of the order ∼ N . The chain of magnetic grains has indeed chain-like
form: it is elongated along external field.
3.6 Effective Form of Dipolar Chains
For chains of finite length it is convenient to introduce the variables

( (
a = d + R2 , b = d + R⊥ 2 (43)
that characterizes the chain scales in two directions. The effective form of the
chain can be judged calculating the parameter of nonsphericity S defined as
S = (a2 − b2 )/(a2 + b2 ) . (44)
The parameter as a function of chain length is shown in Fig. 5 for MF with λ = 10

for cases of zero and infinity external field. As a chain length we use discrete
variable N taking values from 2 to 29. Each diamond in Fig. 5 corresponds to
specific value of N .
S
1.0
ξ=∞
0 .8
0.6
0 .4
ξ=0
0.2
0 .0
0 5 10 15 20 25 N
Fig. 5. Nonsphericity parameter S as a function of number of particles in chain in
zero (white diamonds) and infinite (black ones) external field. The coupling parameter
λ = 10
As it follows from analysis of the figure, even for a very strong dipole inter-
actions the value of S for zero field (lower curve in Fig. 5) quickly decreases with
growth of number of particles in the chain. In fact, any dipolar chain can be
considered as a coil slightly elongated along one direction. The more number of
grains in a chain, the smaller the coil elongation. At the same time in the case
of infinite external field the nonsphericity parameter S increases fast with the
growth of number of particles in a chain and approaches to unity (upper curve
in Fig. 5). Does it mean that dipolar chain becomes stiffer in external field? Of
course, it does not. The point is that each Kuhn segment is directed along the
external field and all the Kuhn segments are connected with each other.
Summarizing the results of the Section 3 we can conclude that the ideal
chain of magnetic grains is very flexible formation with short (of the order of
a few particles) persistent length or the Kuhn segment. The chain looks like a
quasi-spherical coil of connected links in the absence of external field. This coil
is disentangled with increasing of the field and transforms into chain randomly
and weakly bent relative to the field direction.
4 Nonideal Chains
Above we have considered the case of ideal chains. We made allowance only for
interactions of neighboring particles belonging to the same chain ignoring in-
teractions between 1) the non-nearest neighbors along the chain, 2) the distant
segments of the same chain which are finding near each other owing to chain
flexibility, and 3) the segments of different chains. The account of the first of
them in the case of straight linear chains is reduced to re-normalization on 20%
of the nearest-neighbors interactions (see Sect. 1). The chain flexibility should
only reduce this value, so that one can regard these interactions as insignificant.
The second type of interactions can form antiparallel side-by-side configurations
of dipoles in zero field. The third type should be taken into account along with
the second one because the friable coils of ideal chains (with typical volume of
∼ N 3/2 in zero field) begin to overlap for strong dipolar interactions (λ ≥ 7) al-
ready at small volume fractions (≤ 1%) of ferroparticles. A due regard for these
interactions can be carried out in the frame of the concept of quasi-monomers
[54] treating every long flexible chain as a system of disconnected segments. We
identify each a segment with the individual grain and take into account the in-
teractions between segments in the approximation of the second virial coefficient
bm . [More precisely, one should examine the Kuhn segment or the segment of
persistent length as an effective monomer. This strict consideration is similar
to above-mentioned accounting for the next-nearest grains in the segment what
re-normalizes slightly the grain-grain interactions. Such an approach, however,
complicates considerably the calculations, what is why we use here more intuitive
rather than the strict arguments]. bm is not the usual second virial coefficient of
interactions of two hard dipoles. The point is, even considering the chain as a
system of disconnected grains, we should remember of the connection of all its
segments. Thus, calculating bm , we assume that the most energetically profitable
positions corresponding to head-to-tail configurations are occupied already by
the grains of main chain and then forbidden for any other interacting segments.
The way of determination of bm can be divided on four steps. First, we right
the usual second virial coefficient b2 (ξ) of interacting dipoles in external magnetic
field H
2
1 ξ
b2 (ξ) = − {e−U (12)/kB T − 1}eξ(e1 h+e2 h) de1 de2 dr , (45)
2v 4π sinh ξ
where ξ = mH/kB T is the Langevin parameter and h is unit vector along field.
Obviously, the equation reduces to (5) in zero field. Second, we approximate (45)
replacing the orientation of one particle by the field direction, e1 → h, so as

ξ
b2 (ξ) ≈ − {e−U (1 2)/kB T − 1}eξ(e2 h) de2 dr , (46)
8πv sinh ξ
where the symbol of averaging denotes the mentioned replacement. Equation (46)
gives the exact values of the second virial coefficient (45) in the limiting cases
of zero and infinite external field. The former follows from the independence
of typical integral I (4) on orientation of first particle, the latter is evident.
So, (46) is an interpolation formula for moderate values of external field. Our
spot check for some values of λ and ξ shows that the difference between (45)
and (46) does not exceed 10%. Third, in order to exclude the occupied states we
restrict the possible values of polar angle θ between vectors r and h by the range
[π/3, 2π/3] what effectively takes into account the presence near particle 1 of two
head-to-tail aligned neighboring dipoles. Fourth, we overcome the nonphysical
logarithmic divergence owing to previous step by the cutting off the region of
integration over interparticle distance r. We assume that r can vary up to its
maximal value r∗ defined by the physical limitation m2 /r∗3 kB T = 1 (within the
spherical volume r ≤ r∗ the dipole potential for the antiparallel orientation of
dipoles exceeds thermal energy. Finally, we write the second virial coefficient of
two interacting quasi-monomers in the form
r∗
2π/3
ξ
bm =− dr dθ de2 {e−U (1 2)/kB T − 1}eξ(e2 h) r2 sin θ . (47)
4v sinh ξ
0 π/3
We have found bm as a function of λ and the dimensionless external field ξ.

We note that the asymptotic expansion at λ 1 poorly describes the calculated
values of bm in the whole range of λ (up to 10) considered. What is why we do
not perform it. The most interesting case is bm (λ, ξ) = 0. It determines the
neutral curve (see Fig. 6) corresponding to compensation of dipole and steric
interactions of quasi-monomers.
This curve divides the plane of parameters λ – ξ into two parts. Lower the
curve the repulsive forces prevail and dipolar chains exist as chain-like formation
- as coil in zero and weak fields, ξ 1, and as almost linear formation in
strong magnetic field ξ 1. Indeed, under the field an antiparallel side-by-side
configuration of dipoles belonging to two neighboring quasi-monomers becomes
unprofitable: both them tend to orient along field. Vice versa, higher the curve
the attraction forces predominate, the coils of grains become unstable against
to interaction of distant along chain particles and any long chain collapses into
dense globule. The indication that just long chains collapse is essential because
the concept of quasi-monomers [54] is suitable only for such chains. In zero
external field the coil-globule phase transition takes place at λ = 3.2 when bm
changes its sign. This value should be considered only as estimation due to
Fig. 6. Line of coil-globule transition at the plane coupling parameter λ - external

field ξ
negligible number of long chains at so small values of coupling parameter (see

Sect. 2).
The case of chains of finite length is much more complicated than studied
one. Nevertheless, some general conclusions from the physical point of view can
be done. Indeed, the interaction between distant along chain grains becomes
important when chain length exceeds the value Lf equal to several of the Kuhn
segments LK (or persistent lengths). Roughly modelling the Kuhn segment by
the hard rod and considering the dipolar chain as freely-jointed sequence of the
segments, it is possible to estimate Lf = 2LK . At λ = 6 this gives Lf ≈ 5 (see
Fig. 4). According to (9) the mean chain length N ≤ Lf for volume fraction
of magnetic grains Φ ≤ 0.1. Thus, the microstructure of such MF is represented
mainly by the number of separate chains. At higher values of coupling parameter,
λ ≥ 6, the mean chain length exponentially grows. As a result, the majority of
chains should transform into compact clusters - the prototypes of globules for
the finite values of N . It is interesting that this qualitative physical picture well
corresponds to the recent Monte Carlo data [39,40].
5 Conclusion
We studied here the spatial and orientational intrachain correlations in zero and
infinitely strong external field and established the important role of flexibility in
the description of conformational properties of dipolar chains. Among the prob-
lems which proved to be out of the scope of the present investigation we would
like to mark the following ones. First, the problem of taking into account of the
interchain interactions which can be significant due to the long-range nature of
dipolar interactions. As it is shown in [58] the interchain correlation can notice-
ably change the mean length of a chain. Second, there is a problem of the unique
description of both effects of dipolar interparticle interactions – the condensa-
tion and association phenomenon. On the language of present work it means
the interconnection between condensation and globule formation. We assume
that direct identification of both phenomena with each other is not appropri-
ate. Most likely the globule formation precedes the liquid-gas phase transition
and the globules themselves are the nuclei of future concentrated phase. An-
other interesting aspect of the problem is the study of both effects in the case of
moderately concentrated magnetic fluids. In our approach it was convenient to
divide mentally the microstructure formation on stages of (i) appearance of ideal
chain and then (ii) its transformation to nonideal one due to chain flexibility. In
reality both processes take place simultaneously. So, it is reasonable to wait for
the formation in sufficiently concentrated magnetic fluids of extended network
[37] instead of globule formation. Finally, there is the problem of description of
thermodynamical and dynamical properties of magnetic fluids with developed
microstructure. This work was begun in [45,48,49,53,58]. We will return to study
of these intriguing questions later.
This work was supported by the Russian Fund for Fundamental Research
(Project 02-03-33003) and the Israel Science Foundation (Grant 336/00).
References
1. P.G. de Gennes, P.A. Pincus: Phys. Kondens. Materie 11, 189 (1970)
2. S.A. Adelman, J.M. Deutch: Anv. Chem. Phys. 31, 103 (1975)
3. M.S. Wertheim: J. Chem. Phys. 55, 4291 (1971)
4. W. Sutherland, G. Nienhuis, J.M. Deutch: Mol. Phys. 27, 721 (1974)
5. K.-C. Ng, J. Valleau, G. Torrie, G. Patey: Mol. Phys. 38, 781 (1979)
6. C.F. Hayes: J. Coll. Int. Sci. 52, 239 (1975)
7. J.-C. Bacri, D. Salin: J. Phys. Lett. 43, 649 (1982)
8. J.-C. Bacri, R. Perzynski, D. Salin, V. Cabuil, R. Massart: J. Magn. Magn. Mater.
85, 27 (1990)
9. J.-C. Bacri, F. Boué, V. Cabuil, R. Perzynski: Colloids Surfaces A 80, 11 (1993)
10. C.F. Hayes, S.R. Hwang: J. Coll. Int. Sci. 60, 443 (1977)
11. E.A. Peterson, D.A. Krueger: J. Coll. Int. Sci., 62, 24 (1977)
12. R.W. Chantrell, J. Sidhu, P.R. Bissel, P.A. Bates: J. Appl. Phys. 53, 8341 (1982)
13. A.F. Pshenichnikov, I.Yu. Shurubor: Bull. Acad. Sci. USSR, Phys. Ser. 51, 40
(1987)
14. A.F. Pshenichnikov, I.Yu. Shurubor: Magnetohydrodynamics 24, 417 (1989)
15. A. Cebers: Magnetohydrodynamics 18, 137 (1982)
16. K. Sano, M. Doi: J. Phys. Soc. Jpn. 52, 2810 (1983)
17. K.I. Morozov: Magnetohydrodynamics 23, 37 (1987)
18. K.I. Morozov: Bull. Acad. Sci. USSR, Phys. Ser. 51, 32 (1987)
19. Yu.A. Buyevich, A.O. Ivanov: Physica A 190, 276 (1992)
20. K.I. Morozov, A.V. Lebedev: J. Magn. Magn. Mater. 85, 51 (1990)
21. A.F. Pshenichnikov, V.V. Mekhonoshin, A.V. Lebedev: J. Magn. Magn. Mater.
161, 94 (1996)
22. A.O. Ivanov, O.B. Kuznetsova: Phys. Rev. E 64, 041405 (2001)
23. I. Nakatani, M. Hijikata, K. Ozawa: J. Magn. Magn. Mater. 122, 10 (1993)

24. H. Mamiya, I. Nakatani, T. Furubayashi: Phys. Rev. Lett. 84, 6106 (2000)
25. F. Gazeau, C. Baravian, J.-C. Bacri, R. Perzynski, M.I. Shliomis: Phys. Rev. E
56, 614 (1997)
26. J.J. Weis, D. Levesque: Phys. Rev. Lett. 71, 2729 (1993)
27. J.M. Caillol: J. Chem. Phys. 98, 9835 (1993)
28. M.E. van Leeuwen, B. Smit, Phys. Rev. Lett. 71, 3991 (1993)
29. M.J. Stevens, G.S. Grest: Phys. Rev. Lett. 72, 3686 (1994)
30. M.J. Stevens, G.S. Grest: Phys. Rev. E 51, 5962 (1995)
31. P.J. Camp, J.C. Shelley, G.N. Patey: Phys. Rev. Lett. 84, 115 (2000)
32. D. Wei, G.N. Patey: Phys. Rev. Lett. 68, 2043 (1992)
33. D. Wei, G.N. Patey: Phys. Rev. A 46, 7783 (1992)
34. J.J. Weis, D. Levesque: Phys. Rev. E 48, 3728 (1993)
35. D. Levesque, J.J. Weis: Phys. Rev. E 49, 5131 (1994)
36. M.J. Stevens, G.S. Grest: Phys. Rev. E 51, 5976 (1995)
37. P.J. Camp, G.N. Patey: Phys. Rev. E 62, 5403 (2000)
38. A.F. Pshenichnikov, V.V. Mekhonoshin: J. Magn. Magn. Mater. 213, 357 (2000)
39. A.F. Pshenichnikov, V.V. Mekhonoshin: JETP Lett. 72, 182 (2000)
40. A.F. Pshenichnikov, V.V. Mekhonoshin: Eur. Phys. J. E 6, 399 (2001)
41. T.L. Hill: Statistical Mechanics (McGraw-Hill, New York 1956)
42. P.C. Jordan: Mol. Phys. 25, 961 (1973)
43. P.C. Jordan: Mol. Phys. 38, 769 (1979)
44. B. Jeyadevan, I. Nakatani: J. Magn. Magn. Mater. 201, 62 (1999)
45. A.Yu. Zubarev, L.Yu. Iskakova: JETP 80, 857 (1995)
46. R.P. Sear: Phys. Rev. Lett. 76, 2310 (1996)
47. R. van Roij: Phys. Rev. Lett. 76, 3348 (1996)
48. M.A. Osipov, P.I.C. Teixeira, M.M. Telo da Gama: Phys. Rev. E 54, 2597 (1996)
49. J.M. Tavares, M.M. Telo da Gama, M.A. Osipov: Phys. Rev. E 56, R6252 (1997)
50. Y. Levin: Phys. Rev. Lett. 83, 1159 (1999)
51. J.M. Tavares, J.J. Weis, M.M. Telo da Gama: Phys. Rev. E 59, 4388 (1999)
52. M.M. Telo da Gama, J.M. Tavares: Comp. Phys. Communic. 121-122, 256 (1999)
53. A.Yu. Zubarev, L.Yu. Iskakova: Phys. Rev. E 61, 5415 (2000)
54. A.Yu. Grosberg, A.R. Khokhlov: Statistical Physics of Macromolecules (AIP,
Woodbury, New York 1994)
55. P. Ehrenfest, V. Trkal: Proc. Amst. Acad. 23, 162 (1920)
56. J.E. Mayer, M. Goeppert Mayer: Statistical Mechanics 2nd edn. (John Wiley &
Sons, New York 1977)
57. M.E. Fisher: Am. J. Phys. 32, 343 (1964)
58. A.Yu. Zubarev: Lecture Notes in Physics (2002) (this issue)
Supplementary Glossary
A = e1 e(
2 dimer correlation function
a = d + R2 longitudinal size of chain
B = e1 r 12 /d dimer correlation function
B∞ dimer correlation function B in infinite field
b = d + R⊥ 2 transverse size of chain
b2 second virial coefficient of dipoles
bm second virial coefficient of quasi-monomers
C = (e1 e2 )2 dimer correlation function
D = (e1 r 12 )2 /d2 dimer correlation function
D∞ dimer correlation function D in infinite field
d particle diameter
E = r/d dimensionless interparticle distance
E∞ dimer correlation function E in infinite field
ei unit vector along dipole i
F = r2 /d2 dimensionless square of interparticle distance
F∞ dimer correlation function F in infinite field
gN number of N -particle chains per unit volume
H external magnetic field
h unit vector along field
kB Boltzmann’s constant
LK Kuhn segment length
Lp persistent length
m particle magnetic moment
N number of grains in chain
n number of grains per unit volume
)N −1
R = i=1 r i,i+1 “end-to-end” vector of N -particle chain
R2 longitudinal mean square of “end-to-end” vector
R⊥2
transverse mean square of “end-to-end” vector
r mean interparticle distance
S = (a2 − b2 )/(a2 + b2 ) nonsphericity parameter
T temperature
v grain volume
Z2 dimer partition function
λ = m2 /d3 kB T coupling parameter
ξ = mH/kB T Langevin parameter
Φ = πnd3 /6 volume fraction of magnetic grains
Magnetoviscous Effects in Ferrofluids
Stefan Odenbach and Steffen Thurm
ZARM, University of Bremen, Am Fallturm, D-28359 Bremen, Germany
Abstract. The appearance of field and shear dependent changes of viscosity in fer-
rofluids opens possibilities for future applications e.g. in damping technologies. To
enhance the effects, it is necessary to understand the observed magnitudes of magne-
toviscosity in commercial ferrofluids from a microscopic point of view. Starting from
experimental results, it is described how the magnetoviscous effects can be explained
by chain formation of a small fraction of large particles in the fluid. With a dedicated
experiment ferrofluids are separated into fractions with high and low amount of such
particles. The rheological characterization of the fractions prove the aforementioned
model. Furthermore it leads to additional information concerning viscoelasticity of the
suspensions in a magnetic field.
1 Introduction
The phenomenon of field dependent changes of viscosity of a suspension of mag-
netic nanoparticles is known since more than 30 years. In 1968 McTague [1]
discovered an increase of viscosity of a ferrofluid containing Co-particles in the
presence of a magnetic field. His experiments showed a dependence of the vis-
cosity increase on the magnetic field strength as well as on the field’s direction
relative to the flow. In his paper as well as in an accompanying paper by Hall
and Busenberg [2] the effect was explained by a hindrance of rotation of the
suspended particles due to the action of the magnetic field. This concept should
be shortly explained here, since it has fundamental importance for almost all
discussions about magnetoviscous effects in suspensions of nanosized particles.
Assume a ferrofluid under influence of a shear flow, as shown in figure 1. In
such a situation the magnetic particles will rotate in the flow due to the action
of mechanic torque produced by viscous friction in the fluid. If a magnetic field
H is applied to the fluid, the magnetic moment m of the particles will align with
the field direction. In a situation where field direction and vorticity of the flow
are collinear (figure 1a) the magnetic alignment will only lead to the fact that
the magnetic moment of the particles becomes collinear with the direction of
vorticity. An influence on the motion of the particle and therefore on the flow
of the fluid as a whole does not appear. The situation changes if vorticity and
field direction are perpendicular. Under these conditions the mechanic torque
will force a misalignment of the magnetic moment of the particle and the field
direction, provided the magnetic moment’s direction in the particle is fixed. An
angle between the mutual directions of the magnetic moment and the field will
immediately give rise to a magnetic torque, trying to realign m and H. This

186 S. Odenbach and S. Thurm
Fig. 1. On the origin of a field dependent change of viscosity in a suspension of magnetic

nanoparticles. Detailed explanation is given in the text.
torque acts opposite to the mechanic torque and causes thus a hindrance of the
free rotation of the particle in the flow. This increases the flow resistance and
thus the fluid exhibits an increased viscosity.
A rigorous theoretical analysis of the phenomenon has been given by M.
Shliomis [3] four years after the experimental discovery. Taking into account
Brownian motion of the particles he derived an expression describing the change
of viscosity - called rotational viscosity ηr - as a function of the strength and
direction of the magnetic field H in the form
3 α − tanh α
ηr = φ η0 < sin2 β >, (1)
2 α + tanh α
where η0 denotes the viscosity of the fluid in the absence of a magnetic field, φ
the volume concentration of the magnetic particles including the surfactant and
β is the angle between vorticity and field direction; < ... > denotes the spatial
average. The parameter α is the ratio of magnetic and thermal energy of the
particles
μ0 mH
α= , (2)
kT
where μ0 denotes the vacuum permeability, m the particle’s magnetic moment,
k Boltzmann’s constant and T the absolute temperature. Equation (1) has been
derived under the assumption of two important limitations. First of all it has
to be assumed, as already discussed in the qualitative description above, that
the magnetic moment of the particles is spatially fixed within the particle. This
assumption leads to a question concerning the relaxation of magnetization of
the fluid. As discussed in P. Fannin’s contribution [4] the magnetization can
either relax by the Néel process, i.e. by a change of the magnetic moment’s
direction in the particle, or by the Brownian process, that means by a rotation
of the particle in the flow. The relaxation takes place by the process with the
shorter characteristic relaxation time. As shown in [4] both times depend on the
size of the particles but the Brownian time scales only linear with the particles’
Magnetoviscous Effects 187
volume while the Néel time depends exponential on the volume. Thus, small
particles will relax by the Néel process while particles above a certain critical
diameter follow the Brownian process. For the discussion of rotational viscosity
this means, that only particles with a diameter above the critical one, usually
called magnetically hard, will contribute to the changes of viscosity.
The second fundamental assumption in [3] has been the restriction to highly
diluted systems, neglecting any interaction between the magnetic particles. For
McTague’s experiments both assumptions were fulfilled. He used Co-particles
with a diameter about 10 nm being sufficiently larger than the critical diameter
of 6 nm. These particles are thus magnetically hard and a respective suspension
is expected to show rotational viscosity. Moreover the volume concentration of
the suspensions in [1] were as low as 0.05 vol% - a high dilution preventing
significant interparticle interaction.
A few month before McTague’s experiment was published, Rosensweig [5]
reported a viscosity change in concentrated magnetite suspensions. The increase
of the fluids’ viscosity found in these experiments was relatively high and reached
about 200 % of the zero field value. A result principally encouraging further inves-
tigations in this direction especially with a focus towards technical applications
using the phenomenon. Nonetheless, these results were not discussed further-
more and especially they were never compared with the theoretical description
in [3].
2 Magnetoviscous Effects in Concentrated Suspensions

As mentioned, the microscopic explanation of rotational viscosity has been de-
veloped and proved with experiments using highly diluted suspensions of mag-
netically hard cobalt particles. Nevertheless, magnetic fluids used for technical
applications and for numerous experiments in basic research, contain magnetite
particles with magnetic volume concentrations in the order of 7-10 vol%. Thus,
an investigation of the viscosity of such ferrofluids in the presence of magnetic
fields is an interesting topic in the field of ferrofluid research. In an experiment
using a relative high shear rate of γ̇ = 500 s−1 Ambacher et al. [6] investigated
the field dependent increase of viscosity in a magnetite ferrofluid containing
7.2 vol% of the magnetic component.
Figure 2 shows the measured results leading to a maximum increase of vis-
cosity of about 40 %. Comparing the experiment with the theory in [3] requires
information about the size distribution of particles, since only particles with a
diameter larger than about 13 nm - being the critical diameter for magnetite -
can contribute to the viscosity changes. As seen from the inset in figure 2, these
particles represent only a small fraction of the overall magnetic concentration
of the ferrofluid. Calculating the rotational viscosity with these values, one ob-
tains the dashed line in figure 2, being in obvious quantitative disagreement with
the experimental result. Nevertheless a fit of equation (1) to the measured data
shows that the general behavior of the viscous changes fits qualitatively well.
This leads to the assumption that the concept of hindrance of rotation holds
Fig. 2. The change of viscosity in a commercial magnetite ferrofluid together with the
theoretical prediction following equation (1) calculated for the particle size distribution
shown in the inset (dashed line) and with a fit of equation (1) (solid line).
even when the absolute values can not be explained in terms of individual par-
ticles rotating in the flow. Moreover, the fit parameters leading to an apparent
mean particle size of about 16 nm for the full particle concentration indicate
that interaction of particles, forcing formation of agglomerates, has significant
importance in the concentrated suspension investigated.
To avoid confusion we will call the changes of viscosity in concentrated sus-
pensions, where interparticle interaction plays a significant role “magnetoviscous
effect”, while the term “rotational viscosity” will be reserved for the case of highly
diluted suspensions with negligible particle interaction.
A dominant contribution of interparticle interaction to the magnetoviscous
effect leads immediately to the question what kind of microstructure is responsi-
ble for the observed effects. To get a deeper insight into the behavior of the fluid
under shear and magnetic field influence we developed a specialized rheometer
for the investigation of ferrofluids [7], shown in figure 3.
The core part of the rheometer is a modified cone-plate flow cell with a mov-
ing plate and a cone attached to the torque sensor. The connection between cone
and torque sensor is provided by a long rod guided in an air bearing to prevent
spurious friction due to the bearings. The long rod provides a sufficient spatial
separation between the sensor and the magnetic field region, ensuring that no
disturbing influence of the field on the measurement occurs. The rheometer al-
lows investigations of magnetic fluids under stationary as well as time dependent
load.
Using this device we have investigated [8,9] the dependence of the magne-
toviscous effect on shear rate and fluid composition. Figure 4 shows a typical
change of viscosity with field strength for various shear rates for the ferrofluid
APG513A containing 7.2 vol% of magnetite particles with 10 nm mean diameter
in an ester. Obviously a strong magnetoviscous effect is observed at low shear
Fig. 3. The rheometer used for the investigation of magnetic fluids in the experiments
presented here. Details are given in the text.
rate diminishing successively with increasing load. The absolute values of the
magnetoviscous effect are again - as in [6] - significantly higher than expected
from the theory in [3]. Furthermore, the field dependent shear thinning is an
effect which is not predicted in [3] at all. These observations led to the assump-
tion that the formation of chains or clusters of particles under the influence of a
magnetic field has to be a leading component in the description of magnetovis-
cosity. The hindrance of rotation of these agglomerates is assumed to give rise
to the strong increase of viscosity with magnetic field strength. Furthermore the
rupture of the chains in a shear flow can be the basis for an explanation of the
observed shear thinning [8,10].
Within the frame of this model one has to observe that only relatively large
magnetite particles with a diameter of more than about 13 nm show an inter-
particle interaction being strong enough to contribute significantly to such chain
formation. That means that only a small fraction of the overall magnetic con-
Fig. 4. The field and shear rate dependence of the viscosity of APG513A
centration in a magnetite ferrofluid is assumed to be of dominant importance

for the magnetoviscous effects.
To obtain a first check of this hypothesis we have performed experiments us-
ing ferrofluids from a common production batch having identical overall magnetic
concentration but being subjected to different purification in strong magnetic
field gradients [9]. The variation of the purification process leads to differences
in the concentration of large magnetic particles. Table 1 gives an overview of
the properties of these fluids as obtained from magnetization measurement. In-
creasing ordinal number of the fluids indicates an increasing content of large
particles. Comparing the magnetoviscous effect in these fluids we obtain a sig-
nificant decrease of magnetoviscosity with decreasing amount of large particles
as illustrated in Fig. 5.
Table 1. Properties of the fluids with variable content of large particles
Fluid Ms [kAm−1 ] dmean [nm]
F1 32.41 8.3
F2 32.34 8.8
F3 31.54 9.2
F4 32.17 9.2
F5 32.06 10.1
This result documents clearly that the magnetoviscous effect in commercial

magnetite ferrofluids is dominated by the interaction of the small amount of
large particles generated during the synthesis. On this basis a model [11] has
Fig. 5. The magnetoviscous effect for fluids with different content of large particles
been developed describing such ferrofluids as bidisperse systems containing a

large fraction of small particles with weak interparticle interaction and a small
fraction of large particles forming chains in the presence of a magnetic field.
Under the assumption that the chains are rigid and straight and that interaction
between chains and between chains and small particles can be neglected one can
calculate the size distribution function of the chains from a minimization of
the free energy of the system. Using this distribution function, an expression
for the stress tensor can be obtained, allowing to fit the experimental data on
magnetoviscous effects.
3 Controlled Changes of the Fluid Composition

We have shown in the previous chapter, that the size distribution of suspended
particles is of fundamental importance for the viscous behavior of ferrofluids
under influence of external magnetic fields and shear stress. Previous experiments
[9] and a corresponding model [11] proved that the amount of large particles is
of major importance for the effects observed.
To investigate the influence of the size distribution, and in particular of the
amount of large particles, on the viscous behavior, it is necessary to change
directly and continuously the fraction of these particles by a controlled separation
process. Therefore we developed a separation device for ferrofluids, which bases
on forced diffusion of magnetic particles due to a strong magnetic field gradient.
3.1 The Theory of Magnetic Driven Diffusion Processes

The development of a magnetic separation device, allowing controlled separation
of a ferrofluid, requires precise information on numerous experimental parame-
ters like diffusion times or required magnetic field gradient, as well as on fluid
parameters like viscosity, interparticle forces and - as the most important factor
- the particle size distribution. First step of the layout of the separation device
was a numerical simulation of the diffusion process to provide information about
expected diffusion times dependent on the magnetic field gradient, which can
vary in strength and spatial structure. Later this approximation of field strength
and geometry was used as input for a computer simulation to develop the shape
of the pole shoes, which are the key parts of the experimental setup.
The numerical simulation bases on the assumption that the magnetic par-
ticles do not interact with each other. This is a safe approximation, because
interparticle forces cause chain or cluster formation, which in turn accelerates
the process of diffusion. In principle two different theoretical approaches exist to
model the process of diffusion in magnetic fluids.
Close to equilibrium, i.e. for weak driving forces leading to diffusion in the
system, an approach from the point of view of irreversible thermodynamics can
be chosen. Detailed information for this approach can be found in Prigogine [12],
de Groot, Mazur [13], Blums et al. [14] and Odenbach [15].
For strong driving forces - as they are required for the separation process
in focus of our experiment - an approach based on statistical physics has to be
used. The individual velocities and trajectories of all particles are obviously not
important, only the integral change of the spatial concentration distribution with
time has to be calculated. The fundamental equations describing this problem
can be found e.g. in Gerber et al. [16], who simulated numerically the process of
HGMS (H igh Gradient M agnetic S eparation). Starting point is the equation of
continuity, describing the temporal changes of the concentration distribution of
the magnetic particles
∂c/∂t + ∇J = 0 (3)
here the total flux of particles J consists of a diffusive part, J d = −D∇c and a
part J f = vc driven by external forces, where v denotes the drift velocity of the
particles. Substituting these expressions into (3) gives
∂c/∂t = ∇(D∇c) − ∇(vc). (4)
Some general approximations are necessary to solve (4). First it is assumed

that the diffusion coefficient D is independent from the concentration, and that
the fluid can be considered as an ideal diluted system of spherical particles.
Therefore the Nernst-Einstein relation D = ukT can be used, with the mobility
of spherical particles u = 1/(6πηr), the Stokes’ particle radius r, and the vis-
cosity of the fluid η. The drift velocity v can be expressed by v = uF , so (4)
becomes
∂c/∂t = D∇2 c − ∇(uF c). (5)
The force F incorporates all forces driving diffusion in the system. The major
contribution to this force is the magnetic force F m = μ0 m∇H, which is exerted
by the field gradient.
Beside the magnetic effect one could principally account e.g. for the gravi-
tational force F g or the viscous force F v due to flow of the background fluid.
The first one is neglected since its influence is about two orders of magnitude
smaller than the magnetically induced drift in reasonable field gradients. The
latter is excluded by the assumption that the fluid as a whole is at rest during
the separation process.
Further effects could appear due to van der Waals or magnetic interaction
leading to an agglomeration of the particles. This would increase the magnetic
moment entering the magnetic driving force and would thus enhance the drift
velocity leading to an acceleration of the separation process. Since the simulation
is thought to give an approximation of the diffusion process allowing the design
of a separation device enabling separation of the large particles in the fluid on
reasonable time scales, effects like the mentioned agglomeration will only enhance
the performance of a system based on the above assumptions.
Finally (5) can be written as

∂c/∂t = 1/ (6πηr) kT ∇2 c − ∇ (μ0 Vmag M0 ∇Hc) (6)
where m has been replaced by m = Mo Vmag with Vmag denoting the volume
of the magnetic particle and Mo being the spontaneous magnetization of the
magnetic material – in our case magnetite with Mo = 4.5 · 105 A/m.
Fig. 6. Numerically calculated time dependent concentration profile for three different
particle diameters after a separation time of 1 week. The height of fluid in the container
is 5 cm
Using (6) one can calculate the concentration distribution of magnetic par-
ticles as a function of time for different particle sizes. From a comparison of
the resulting distributions it can be judged whether a separation of a fluid in
fractions with different content of large particles and mainly constant content of
small ones is possible.
Fig. 6 shows the calculated time dependent concentration profile for three
different particle diameters after a separation time of one week for a 5 cm deep
container. The field gradient applied in the direction of diffusion is constant over
the depth of the container at a value of 107 A/m2 . As a boundary condition it
has been assumed, that the concentration at the lower wall of the container can
not exceed 150 % of the original concentration.
One can see that the magnetic field gradient acts mainly on the largest par-
ticles. Particles which are smaller than 10 nm hardly react on the field gradient,
thus we are able to concentrate just the fraction of bigger particles in the region
of high magnetic field. The parameters separation time and field gradient control
the accumulation and depletion of bigger particles in the lower and upper region
of the container respectively.
3.2 The Experimental Setup

One of the key requirements for the separation device was the possibility to
work with fluid volumes up to 300 ml. This large amount of fluid is necessary to
enable successive separation steps with corresponding rheological investigations.
This constraint and the available free space between the pole shoes of an high
field electromagnet restricted significantly the geometrical parameters for the
separation container. The final geometry of the setup was chosen by an iteration
of the shape of the pole shoes, the height of the container and corresponding
diffusion times. The result is a trapezoid shaped container with a height of 5 cm
as shown schematically in Fig. 7.
Fig. 7. Principle sketch of the experimental setup
The container is located in the gap between two tailor made pole shoes of an
electromagnet. The pole shoes are designed in a way that they provide a constant
magnetic field gradient of 107 A/m2 over the whole fluid volume. This ensures
a uniform diffusion of particles from the top to the bottom of the container as
already shown in the simulation (see Fig. 6).
The container is equipped with an in-situ concentration measuring system,
consisting of three coils, located in different heights. The inductance of the coils is
directly proportional to the local concentration of magnetic particles. Therefore
it is possible to measure a time dependent concentration profile in the container.

In principle this information can be used to simulate numerically the diffusion
process with consideration of chain and cluster formation.
The whole setup is temperature controlled to an accuracy of 0.1 K. This
ensures that no convective flow due to temperature differences can disturb the
diffusion process. Furthermore, the temperature stability is a technical require-
ment for the concentration monitoring using the sensor coils. The inductance
of the coils depends not only on the concentration of magnetic particles, but
also on temperature and the strength of the applied magnetic field. By keeping
the external magnetic field and temperature constant during the measurements,
the inductance of the coils corresponds directly to the local concentration of
magnetic particles.
4 Results
The fluid used in the experiments discussed hereafter has been a sample of
APG513A from Ferrofluidics, with a saturation magnetization of 32 kA/m, an
initial susceptibility of 1.4 and a dynamic viscosity of 150 mPas. The magnetic
separation ran over a period of 30 days. After this time period the separation
was stopped, and the fluid was removed in two fractions from the container as
indicated in Fig. 7.
4.1 Inductance Measurements during Separation Process

Fig. 8 shows the change of concentration for the upper and lower region of the
container. It is calculated as the difference between the concentration signals of
the sensor coil in the center and the upper and lower sensor coils respectively.
The middle coil shows more or less no change in concentration. This result is
expected, because the overall concentration in the container has to be constant,
and according to numerical results the middle coil should not react before an ad-
vanced state of separation. Thus this coil can be used as a reference concentration
as mentioned above. The upper coil shows a decrease in concentration, which is
as strong as the increase of concentration of the lower coil. It can be seen, that
saturation of separation is not reached after 30 days. In this context saturation
means, that all bigger particles are in the lower region of the container, and the
continuing process of diffusion is too slow to cause measurable changes of the
signal of the coils. So one can expect, that a small amount of large particles is
still present in the upper fraction and thus a certain magnetoviscous effect is
expected for this part of the fluid too.The concentration difference between the
lower and upper fraction of the fluid as obtained from the sensor signals is round
about 6 percent.
Fig. 8. Time dependent change of concentration during separation process
4.2 Magnetization Measurements

Magnetization measurements provide information about saturation magnetiza-
tion and initial susceptibility. The saturation magnetization Ms corresponds to
the concentration of magnetic material in the fluid, and the initial susceptibility
to the mean particle diameter [17]. Thus after a separation process Ms should
be increased in the sample taken from the lower part of the container and de-
creased in that extracted from the upper one, while the initial susceptibility χi
should become smaller in the upper fraction, and larger in the lower fraction
all compared with the values obtained for the original fluid before separation.
Tab. 2 shows the results for the original sample of APG513A, and for both
fractions after the separation process. The concentration difference between the
upper and lower fraction of the fluid is about 10 percent. This value is somewhat
higher than the one obtained from the in situ determination of the concentration
changes using the sensor coils. The difference is simply caused by the fact that
the magnetization measurement gives an information about the mean concen-
tration of the respective fraction of the fluid, while the sensor coils just measure
a concentration value at a certain depth inside the container.
Calculating the mean saturation magnetization for both fractions, and com-
paring it with the original fluid, we can find a loss of particles due to adhesion
of particles at the container walls in particular in the lower part of the con-
tainer, where the bigger particles and agglomerates are collected. This loss has
no critical influence on the final characterization and discussion of results, but
one should keep in mind, that the changes of the magnetoviscous effect discussed
afterwards for the lower fraction would even be greater if these particles would
also be removed from the container together with the lower fraction.
4.3 Rheological Measurements

Rheological measurements are especially suited for the investigation of the influ-
ence of the microstructural characteristics of ferrofluids on their viscous behavior.
Table 2. Magnetization data from the original APG513A and both fractions after the
separation process
Fluid name: Ms χi dmean

APG513A [kA/m] [nm]
originalf luid 32.6 1.37 12.0

upperf raction 30.0 1.16 11.6
lowerf raction 33.1 1.42 12.1
The discussion of the results obtained with these measurements is splitted into
two parts. First we will focus on the magnetoviscous effect, measured with the
rotational mode of the ferrofluid-rheometer described above. The elemental in-
fluence of bigger particles on the change of viscosity under influence of shear flow
and magnetic fields is the central aspect of this discussion. In the second part
we will present some results obtained in the oscillating mode of the rheometer,
leading to information about magneto-viscoelastic effects.
The magnetoviscous effect. As seen from the experiments in [9], the inves-
tigation of field and shear rate dependent changes of viscosity in a ferrofluid
provide an excellent tool to get an insight into the microscopic reasons for the
rheological properties of the suspensions. The model of chain formation of big-
ger particles – being under test with the experiments described here – leads
to a couple of phenomena being expected to appear in a ferrofluid separated
as described above. First of all, in zero field, the dependence between stress
and shear rate in both fractions should be linear. That means the fluid should
behave Newtonian, since no field induced interparticle interactions leading to
chains or clusters are present. The ferrofluid behaves like a normal suspension
of non-magnetic particles.
Applying magnetic fields, the upper fraction containing only a negligible
amount of large particles is expected to show weak magnetoviscous effects since
the majority of particles in this fraction is too small to cause significant effects
due to interaction.
In contrast the lower fraction should react noticeable on increasing magnetic
field strength. These dependencies are illustrated in Fig. 9.
For the upper fraction - represented by open symbols - the relation between
stress and shear is linear. This fraction shows mainly newtonian behavior and
the existing weak dependence from magnetic field strength and shear rate will
only be seen in a field dependent plot (see Fig. 10). This is a direct validation of
the assumption, that just the bigger particles cause chain and cluster formation,
which in turn gives rise to significant changes of the viscosity.
The lower fraction is represented by the filled symbols in Fig. 9 where different
symbols indicate different stength’ of the applied magnetic field. Starting with
Fig. 9. Stress versus shear-rate for different magnetic fields. Both fractions and the
original fluid are plotted
the curve for zero field, the similarity between the lower and upper fraction is
obvious. This is expected for H = 0 since the only difference between the fluids
is a slight change in number and size of the suspended particles. To understand
this one can apply Rosensweig’s equation for the dependence of viscosity on the
volume fraction of magnetic particles in moderately concentrated suspensions
[17],
⎛ ⎞
& '2 −1
η 5 5 φ̃
= ⎝1 − φ̃ + φ̃c − 1 ⎠ , (7)
η0 2 2 φ̃c
where φ̃ denotes the volume fraction of particles including their surfactant and
φ̃c is a critical volume fraction for which the suspension’s viscosity diverges -
this value is usually chosen as φ̃c = 0.74. It is easy to see that a difference of
volume concentration of about 10% - as it was found for the two fractions of our
separation experiment - leads to a viscosity difference of about 7% only.
With increasing magnetic field strength an increase of viscosity - the magne-
toviscous effect - can been seen from the change of the slope of the stress-shear
relation. Furthermore one observes that this difference diminishes with increas-
ing shear rate. So presenting the magnetoviscous effect in the commonly used
form by plotting the change of viscosity as a function of field strength for various
shear rates leads to the well known behavior as shown in Fig. 10. As discussed
earlier these changes of viscosity can be explained by the formation and rupture
of chains formed by the large particles in the suspension in the presence of a
magnetic field. We compare here the change of viscosity for the original fluid
with both fractions for two shear rates as a function of magnetic field strength.
As expected, the magnetoviscous effect increases in all three fluids with de-
creasing shear rate. Furthermore the influence of the concentration of bigger
particles is directly visible. The fluid with a large concentration of bigger parti-
Fig. 10. Magnetoviscous effect in original APG513A, lower and upper fraction
cles has a much stronger magnetoviscous effect than the original fluid, and vice
versa in the upper fraction with a small concentration of bigger particles.
Viscoelastic effects. The formation of chainlike clusters gives rise to the as-
sumption, that viscoelastic effects should occur in ferrofluids under the influence
of magnetic fields. A first evidence for such effects was provided from Odenbach
et al. [18] with the investigation of the Weissenberg-Effect in ferrofluids exposed
to magnetic fields. Using rheometrical investigations, the application of an oscil-
lating load gives direct access to information about viscoelastic properties. The
advantage of the oscillating mode is, that measurements with small amplitudes
does not change the “zero-shear-rate” structure of the fluid.
As an example we’ll discuss here the phase shift between shear and stress,
being a characteristic measure for the appearance of viscoelasticity. For an elastic
body this phase shift should be zero, while it is π/2 for a Newtonian liquid.
Correspondingly viscoelastic liquids exhibit phase shifts between 0 and π/2.
It was shown in Fig. 9, that the upper fraction shows a linear relation between
stress and shear rate, indicating newtonian behavior. This is confirmed by the
phase shift δ = π/2, being constant over the whole range of oscillation frequencies
ω as shown in Fig. 11. In contrast the lower fraction with many bigger particles
should have a phase shift between 0 degree and π/2, dependent on the length of
the chains formed by the particles. This length itself depends on shear rate as
well as on the strength of the applied magnetic field.
The lower fraction shows just under zero field no phase shift- i.e. it behaves
Newtonian for vanishing field. If a field is applied, a phase shift between 0 and
π/2 is observed. This phase shift decreases continuously with decreasing shear
rate and thus with increasing length of the particle chains. The phase shift of
the original fluid takes values in between the phase shifts of the upper and lower
fraction. The higher number of bigger particles in the lower fraction obviously
Fig. 11. The phase shift between shear and stress for the original fluid and both
fractions obtained by separation as a function of the frequency of the oscillating load
for different shear rates. To obtain the given data points, measured values in a field
range between 3 kA/m and 17 kA/m have been averaged. The error bars give the
standard deviation of this averaging process.
enhances the appearance of viscoelasticity, and a reduction of the number of

bigger particles in the upper fraction finally leads to a pure Newtonian system.
Acknowledgements
Support of this work by DFG (Grant.No. Od18/2 and Od18/3) is gratefully

acknowledged.
References
1. J. P. McTague: Magnetoviscosity of magnetic colloids. J. Chem. Phys., vol.51, no.1
(1969)
2. W.F. Hall, S. N. Busenberg: Viscosity of magnetic suspension. J. Chem. Phys.,
vol.51, no.1 (1969)
3. M. I. Shliomis: Effective viscosity of magnetic suspensioms. Soviet Phys. JETP,
vol.34, no.6 (1972)
4. P. Fannin: Magnetic spectroscopy as an aide in understanding magnetic fluids.
Springer LNP this issue (Berlin, New York 2002)
5. R.E. Rosensweig, R. Kaiser, G. Miskolczy: Viscosity of magnetic fluid in a magnetic
field. J. Colloid Interface Sci., vol.29, no.4 (1969)
6. O. Ambacher, S. Odenbach, K. Stierstadt: Rotational viscosity in ferrofluids. Z.
Phys. B-Condensed matter, vol.86 (1992)
7. S. Odenbach, T. Rylewicz, M. Heyen: A rheometer dedicated for the investigation
of viscoelastic effects in commercial magnetic fluids. J. Magn. and Magn. Mat.,
vol. 201 (1999)
8. S. Odenbach, H. Störk: Shear dependence of field-induced contributions to the vis-

cosity of magnetic fluids at low shear rates. J. Magn. and Magn. Mat., vol.183
(1998)
9. S. Odenbach, K. Raj: The influence of large particles and agglomerates on the
magnetoviscous effect in ferrofluids. Magnetohydrodynamics, vol.36, no.4 (2000)
10. S. Odenbach: Magnetoviscous effects in ferrofluids. Springer LNP m71 (Berlin, New
York 2002)
11. A. Yu.Zubarev, S. Odenbach, J. Fleischer: To the theory of dynamical properties
of polydisperse magnetic fluids. II. Effect of chain-like aggregates. submitted to
Physical Review E (2001)
12. I. Prigogine: Thermodynamics of irreversible processes. (3. edition, John Wiley and
Sons, New York 1967)
13. S.R. de Groot, P. Mazur: Grundlagen der Thermodynamik irreversibler Prozesse.
BI 162/162 a (Berlin 1962)
14. E. Blums, Yu.A. Mikhailov, R. Ozols: Heat and Mass transfer in MHD flows.
Series in theoretical and applied mechanics, 3. volume, ed. by R.K.T. Hsieh (World
Scientific Publishing, Singapore 1987)
15. S. Odenbach: Konvektion durch Diffusion in Ferrofluiden. PhD Thesis, Ludwig-
Maximilians University, Munich (1993)
16. R. Gerber, M. Takayasu, F.J. Friedlaender: Generalization of HGMS theory: the
capture of ultra-fine particles. IEEE Trans. Magn., vol.mag-19, no.5 (1983)
17. R.E. Rosensweig, R. E. Ferrohydrodynamics. Cambridge University Press (Cam-
bridge, New York 1985)
18. S. Odenbach, T. Rylewicz , H. Rath: Investigation of the Weissenberg effect in
suspensions of magnetic nanoparticles. Physics of Fluids 11 (1999) 2901
Magnetorheology:
Fluids, Structures and Rheology
G. Bossis1 , O. Volkova1 , S. Lacis2 , and A. Meunier1

1
LPMC, UMR 6622, University of Nice, Parc Valrose, 06108 Nice Cedex 2, France
2
Department of Physics, University of Latvia, Zellu str.8, LV-1586, Riga, Latvia
Abstract. Magnetorheological suspensions are complex fluids which show a transi-

tion from a liquid behavior to a solid one upon application of a magnetic field. This
transition is due to the the attractive dipolar forces between the particles which have
been magnetized by the applied field. The formation of a network of particles or ag-
gregates throughout the suspension is the basic phenomena which is responsible for
the strength of the solid phase. In this paper we shall give an overview on the fluids
and their properties and we shall especially emphasize the interplay between magnetic
forces which are responsible for the gelling of the suspension and on the other hand of
hydrodynamic and thermal forces which contribute to break this gel and allow the sus-
pension to flow. The combination of these three forces gives rise to a very rich rheology
whose many aspects are still not understood.
1 Introduction
Since Rabinow and Winslow’s [1,2] discoveries in the 1940’s, magnetorheology
(MR) and electrorheology (ER) have emerged as a multidisciplinary field whose
importance has considerably increased these last ten years. The rheology of
these fluids is very attractive since it can be monitored by the application of a
field, either magnetic or electric. The most important advantages of these fluids
over conventional mechanical interfaces is their ability to achieve a wide range of
viscosity (several orders of magnitude) in a fraction of millisecond. This provides
an efficient way to control force or torque transmission and many applications
dealing with actuation, damping, robotics have been patented and are coming
on the marketplace [3]. The basic phenomena in electro or magnetorheology is
the ability to control the structure of a biphasic fluid. The two phases are usually
made on one hand of solid particles in the micrometer range and on the other
hand of a carrier fluid. The application of a field polarizes the particles and
induces their transient aggregation, hence an increase of the viscosity. In some
other cases the two phases can be two immiscible fluids and the application of
the field will change the shape and size of the droplets of the dispersed phase
which will modify the rheology [4,5]. Some intermediate cases can exist where
the carrier phase is itself a suspension of nanoparticles, a typical case being
a ferrofluid. Furthermore new materials where the two phases are solids, like
solids particles embedded in a rubber matrix belong to the same category of
smart composites whose rheology can be controlled by an external field [6,7,8].
All these materials have common features regarding the relation between the
shape and size of the domains of the dispersed phase and their rheology, even

Magnetorheology 203
if the microscopic interaction between the constituents of the dispersed phase

can be very different, depending on their size (polymer, nanoparticle, micronic
particle) and other physical characteristics like conductivity, permittivity, mag-
netic permeability, properties of the interface and so on. In this paper devoted
to MR fluids we shall of course focus on magnetic interactions but many devolp-
ments are also relevant to ER fluids who had received more attention in the past
years with several review papers [9,10,11,12,13,14]. The interest in MR fluids has
raised more recently and review papers are scarce in this domain [15,16,17]. Our
goal is to give a brief overview of research on MR fluids and their rheology and
also to give some guide lines to beginners in this domain.
In section 2 we shall describe the main phenomena, give some orders of
magnitude and present the different kinds of fluids which are used with their
advantages and disadvantages. Section 3 contains a presentation of the field
induced structures. Section 4 is centered on the rheology of magnetic fluids,
with a description of the apparatus and methods (section 4.1) a comparison of
experiments and models in relation to the yield stress (section 4.2) and to the
shear rate dependency (section 4.3).
2 Overview of MR Suspensions
2.1 General Description of the Interactions in an MR Suspension
If an isolated particle of relative magnetic permeability μp surrounded by a

fluid of relative permeability μf is placed in an external magnetic field H0 , this
particle will acquire a magnetic moment: m = 4πμ0 μf βa3 H0 ; a is the radius
of the particle, μ0 the permeability of vacuum and β = (μp − μf )/(μp + 2μf ).
It is worth noting that this formula also holds when the permeability of the
carrier fluid is larger than the one of the particle; in this case β < 0 and the
magnetization vector is opposed to the field. The interaction energy between two
dipoles of moment m is given by:

1 mα mβ 3(mα r)(mβ r)
W = − (1)
4πμ0 μf r3 r5
where r is the separation vector between the centers of the two particles. This
energy is minimum when the two dipoles are aligned with r and maximum when
they are perpendicular leading to a preferential aggregation as chains of particles
aligned on the direction of the field. The formation of aggregates of particles will
depend on the ratio of this interaction energy to kT . Taking as reference the
energy of two dipoles in repulsive configuration gives:
1 m2 1 πμ0 μf β 2 a3 H0 2
λ= = (2)
4πμ0 μf r3 kT 2kT
For particles of diameter 1μm with large permeability (β ≈ 1) and T = 300K,

we obtain from (2) λ = 1 for H = 127 A/m (or H = 1.6 Oersted in cgs
204 G. Bossis et al.
units). It means that for usual magnetic fields the magnetic forces dominate
the Brownian forces. The situation is quite different if we consider ferrofluids
which are magnetic fluids with particle diameters of about 100 Angstrom; then
in the same conditions we have λ = 1 for H = 1600 Oersted. It means that for
usual fields the magnetic forces will always be dominated by Brownian forces and
we cannot expect to change significantly the viscosity of a ferrofluid by applying
a magnetic field. Furthermore this criterion on λ has to be reinforced by the fact
that the efficiency of thermal forces to break a chain of spheres increases with the
length of the chain. Despite these restrictions which seem to prevent the finding
of a noticeable ER or MR effect with suspensions of colloidal particles there are
some experimental evidences for large changes of viscosity in some ferrofluids
[18]; we shall discuss this point in more details in the next section. The quantity
λ is the key quantity which, together with the volume fraction, Φ = N vp /V , will
determine the equilibrium structure of a suspension of monodisperse particles
as a function of the applied field. We shall come back to this important point in
section 3.
In order to obtain all the quantities which will rule the behavior of the sus-
pension it is useful to start from the equation of motion of one particle and to
put it in a dimensionless form. This is what we would do to calculate the trajec-
tories of the particles by using Brownian or Stokesian dynamics [19,20,21]. For
a given particle we can write:
dv
m = F H + F ext + F I + F B (3)
dt
The first term F H is the hydrodynamic force on the test particle coming from
the hydrodynamic friction and is proportional to −ξ(v −v 0 (x)), where ξ = 6πμa
with μ the viscosity of the suspending fluid and v 0 (x), the imposed velocity field
at the location x of the particle. The term F ext is the hydrodynamic force due
to the symmetric part of the velocity gradient tensor. In the case of a pure shear
characterized by the shear rate γ̇ this force scales as 6πμγ̇a2 . The third force F I
is the interparticle force coming from the dipole-dipole interaction and given by
minus the gradient of (1). For two particles α and β the force on α will be:
a 4
FαI = 12πμ0 μf a2 β 2 H02 (2 cos2 θαβ − sin2 θαβ )er + sin 2θαβ eθ (4)
r
The meaning of the different vectors is defined in Fig.1. For two spheres
placed side by side (r = 2a, θαβ = 90o ) we have FαI = fd = −(3/4)πμ0 μf a2 β 2 H02
which is repulsive as expected; we shall take −fd as the scaling factor of the
interparticle force. The last term in (3) is the Brownian random force which
scales as kT /a. In general we can neglect the inertial force (the time needed
to reach a constant velocity: τ = m/ξ is 1.7μs for an iron particle of radius 1
micron in water; it is much smaller than the other characteristic times) so we
can neglect the left hand side of (3). Dividing all the terms of (3) by 6πμγ̇a2
and rearranging we obtain:
I B
(v − v 0 ) F F
= + + F ext (5)
γ̇a Mn Pe
Magnetorheology 205
Fig. 1. Two particles in a magnetic field
The brackets mean that the forces have been divided by their scaling factor.
Mn is the Mason number:
6πμγ̇a2 8μγ̇
Mn = − = (6)
fd μ0 μf β 2 H02
It expresses the ratio of shear to magnetic forces. Note that in other papers
the definition can differ by a multiplicative factor. P e = 6πμγ̇a3 /kT is the Peclet
number and expresses the ratio of shear forces to Brownian forces. For particles
larger than one micron and reasonable shear rates, the Peclet number is large (for
instance Pe=4.5·106 for a particle of radius one micron in water with γ̇ = 1s−1 )
and the Brownian force can be neglected; nevertheless we have to keep in mind
that the influence of the Brownian motion can still be important even at quite
high Peclet number.
The dimensionless equation of motion depends on two quantities (actually
the three quantities defined above are related by M nλ = 2 P e/3) so, for a given
suspension, all the trajectories and hence all the properties-and in particular
the viscosity - will be the same for the same values of M n and λ. Of course
this equivalence only applies for systems of particles starting in the same initial
conditions, that is to say with the same volume fraction Φ and the same initial
configurations. This last point is usually not critical if we are only interested
by equilibrium properties, so we can say that for monosized hard spheres with
particles having the same magnetic permeability, the viscosity (normalized by
the one of the carrier fluid) will depend only on three quantities which are Φ,
M n, λ.
The efficiency of an MR fluid is firstly judged through its yield stress, τy ,
which measures the strength of the structure formed by the application of the
field. The restoring force per unit surface, which resists to the deformation of the
structure, is given by the derivative of the magnetic energy per unit volume rel-
atively to the strain γ : τ = −dW/dγ. The yield stress represents the maximum
of the stress versus strain: τy = max(τ ) since above a critical shear strain, γc ,
the gel-like structure will break. This definition of the yield stress through the
energy of the field allows to understand why it is easier to get a larger yield stress
with MR fluids than with ER fluids : the vacuum magnetostatic energy density:
μo H 2 for H = 3000 Oe is an order of magnitude larger than the electrostatic
energy density o E 2 for a field E = 3 kV/mm (close to the breakdown field).
It does not mean that ER fluids will never show yield stress as high as in MR
fluids, because it also depends on the efficiency of the polarization mechanisms.
For instance ER fluids having ionic polarizability may show high yield stress if
the ions can remain confined inside or on the surface of the particles. This is an
other issue, let us simply present the state of the art for MR fluids.
2.2 MR Fluids
Strong fluids. A simple and quite efficient MR fluid is obtained by dispersing

iron powder in oil with the use of some surfactant - for instance stearic acid
- in order to prevent irreversible aggregation. Magnetorheological suspensions
made of iron synthesized from iron carbonyl precursors and known as carbonyl
iron, have been extensively used, first by Rabinow [1], also by Shulman et al [22]
and later on by many other teams. These particles are used for producing high
permeability circuits, they are quite spherical and can be found rather easily in
the range 2-10μm. The magnetization saturation of iron is large (μ0 Ms = 2.1
Tesla) and its remanent magnetization very low. The efficiency of this fluid is
measured through its yield stress which is, for a volume fraction of 30%, typically
50 kPa for a field H = 4000 Oersted [23]. An other study report a yield stress
of 100kPa for an induction of 1 Tesla and a volume fraction Φ = 0.36 [24]. The
magnetic induction B is given by:
B = μ0 (M + H) with M = χH (7)
The magnetic induction can be measured with the help of a sensing coil and
of a fluxmeter and is safer to use for comparisons between different fluids because
it is a rather well conserved quantity inside a magnetic circuit. On the contrary
the magnetic field H inside the fluid is not directly measurable; it can be deduced
from the measurement of the field with a Hall gauge placed in an air gap close to
the fluid; we shall come back to this point in section 4.1. This problem as well as
the presence of additives which can modify the state of aggregation or impose a
non zero gap between particles, also the lack of knowledge of the polydispersity,
are factors which make difficult the comparison of the results between different
authors. Density of iron is large (=7.87 g/cm3 ) and a particle of radius one
micron in water has a sedimentation velocity of 5.4cm per hour. In order to
prevent the sedimentation and potential redispersion difficulties, it is usual to
add a gel forming additive. The gel must have a low yield stress in order to flow
easily under a small agitation. Nanometer silica particles can be used in this aim
because they easily form a physical gel at low volume fraction (2-3%) and the
yield stress in the absence of field is only a few Pascal. An other problem-even
more important than sedimentation- is the corrosion of iron particles. It has been
found [25] that under permanent use in a damper valve and in the presence of a
magnetic field, the fluid thickens progressively and can show an increase of the
off state (zero field) viscosity by a factor of 3 after 600000 cycles. This thickening
has been identified as due to the removal of thin oxide fragments from the surface
of the particles. The operating conditions in a damper (high shear rate: 104 to
Magnetorheology 207
105 s−1 , strong attractive forces between the particles in the presence of the field,
possible corrosion due to dissolved oxygen) favor the fragmentation process and
must be prevented by a proper surface protection of the particles. The new fluid
made by Lord Corporation no longer show any appreciable thickening even after
2 million cycles. Progress in the strength of the fluids has been realized by using
materials with a higher saturation magnetization like for instance cobalt-iron
alloys (μ0 Ms = 2.45 Tesla for about 50%Fe, 50%Co) which shows a higher yield
stress than iron: 70kPa instead of 50kPa for H=4000 Oersted [23]. Commercial
fluids available from Lord Corporation have typically yield stresses between 50
and 70kPa for a volume fraction between 0.35 and 0.4 and a field of 300kA/m
(3770 Oersted). The off state viscosity is often higher than 1Pas at shear rate
of 50s−1 . The composition of these fluids should still progress in order to have
a lower off state viscosity and still keeping a good protection against corrosion
and a low sedimentation rate. At last it is worth mentioning that as proved
by R.T.Foister [26] mixing two different sizes of carbon iron particles improves
considerably - for the same volume fraction - the on/off ratio of stresses (they
report an improvement of 2.7 times over the monosized suspension at Φ = 0.55
and B = 1Tesla). This increase is quite understandable since it is known that the
maximum packing fraction is increased by mixing two sizes such that the small
spheres can be placed in the holes between the large spheres; consequently the
viscosity diverges at a higher volume fraction, which means that, at the same
volume fraction, the off state viscosity is lower than the one of a monosized
suspension. More surprisingly they also observe - for the same volume fraction
- an increase of the yield stress in the on state of about 25% for the bidisperse
suspension which has no clear explanation.
Colloidal MR fluids. If iron based particles with radii larger than 1μm, give
fluids with high strength, they have the inconvenience to be abrasive. Since the
shear force, which will push the particles against the walls or against each other,
is proportional to the square of the size of the particles, it would be advanta-
geous to reduce this size, which would reduce both abrasion, sedimentation and
fragmentation. On the other hand a too low value of the parameter λ would
not authorize the formation of clusters, so there is a compromise to find. For a
particle of diameter 0.1μm we need a field of about 50 Oersted in order to have a
value of unity for λ. It seems difficult to go well below this limit but nevertheless
Kormann et al [18] have shown that a suspension of soft ferrites with an average
diameter of about 30nm and a volume fraction of 23% can give a dynamic yield
stress of 2kPa for an induction of 0.2Tesla. If we suppose a permeability of about
2 and so H = 1000 Oersted we obtain λ = 10 which seems to be too low to ex-
plain this quite large effect on viscosity. It is quite likely that the nanometric
particles are partially aggregated even in zero field; it happens quite naturally
since the particles are monodomain and carry a permanent dipole. The aver-
age dipole of these clusters is zero in the absence of a field but the persistence
length of the correlations between the orientations of the dipoles of adjacent
particles can be much larger than the particle diameter and explain the large
change of viscosity with the field (for magnetite the attractive dipolar energy
between two particles at contact overcomes the thermal energy for a diameter
larger than 10nm [27,28]). When the field is turned on, the dipoles align on the
field, the cluster becomes magnetized and, thanks to its larger dimension, will
attract other clusters and form the gel like structure. The zero field viscosity of
this fluid is quite high (about 50Pas at 0.1s−1 [29]) which supports the hypoth-
esis of a zero field aggregation. In any event such a fluid does not sediment and
is not abrasive so it could be useful if not too high yield stress is needed. An
other attempt to use nanoparticles has been presented recently with carbonyl
iron [30]. The particles were obtained from decomposition of vapors of iron pen-
tacarbonyl with an average diameter of 26nm. For a volume fraction of about
16% it gives an average increase of yield stress of 7kPA. The experiment was
conducted with a MR damper and the value of the field is not given so it is not
possible to compare with the ferrite particles but, here too, we see that this fluid
can be interesting for applications. It is worth noting that this nanosized fluid
presents a quite large yield stress at zero field (3.6kPA) indicating a permanent
aggregation. This is not surprising if we suppose that these particles are still
monodomain, then, due to their larger magnetization saturation compared to
ferrite (by a factor of 4) the dipolar interactions will lead to this aggregation.
Concerning the use of colloidal particles it seems more interesting to use
larger particles (for instance 0.1μm) in order to avoid the presence of a single
ferromagnetic domain but still keeping a large enough value of λ.
Other types of MR composites. For most of the applications the fluids

described above will be best adapted. Nevertheless numerous other fluids can be
used with special goals. We shall describe some of them without pretending to
be exhaustive.
An interesting fluid with regard to the influence of the Brownian forces on
the rheology is the one made of a suspension of monodisperse non magnetic
colloidal particles in a ferrofluid. As it can be seen from (1), the interparticle
force is proportional to the square of the dipole. If the particle is not magnetic
(μp =1) but the suspending fluid is magnetic: μf >1, then the induced moment is
opposite to the field (β <0) but the interparticle force (4) is still attractive. The
existence of a magnetorheological effect in this kind of system was demonstrated
[31,32], and using different sizes of silica particles it was shown that the yield
stress decreases when the size of the particles becomes smaller [33], the rheology
of these inverse ferrofluids has also been recently investigated [34,35].
Magnetic particles are usually conductive particles and so not suitable for
electrorheology, nevertheless by coating the magnetic particles with an isolating
layer, it becomes possible to apply both an electric field and a magnetic field.
Surprisingly the resulting yield stress τyH,E was found larger than the sum of
each separated yield stress: τyH,E > τyH + τyE [36,37]. This synergistic effect is
explained in the following way [38]: in ER fluids cancellation of the field on the
electrodes introduces dipole images of the particles with respect to the plane
of the electrodes; the attractive interaction between original dipoles and their
Magnetorheology 209
images drift the chains of particles towards the electrodes and in a shear the weak
point of the structure is in the middle of the cell. On the contrary for magnetic
particles in the presence of a magnetic field and non magnetic electrodes the
chain do not attach on the electrodes and the density of particles is larger at
the center of the cell, then in the presence of the shear the fracture zone is
located on the electrodes. In the presence of both electric and magnetic field
the chain like structure will be more homogeneous throughout the cell without
weak points, hence the synergistic effect. Nevertheless the use of a non magnetic
layer decreases the magnetic force between particles, so it is not clear at all that
the synergistic effect will compensate the decrease of magnetic stress. In [38] the
results are presented for carbonyl iron coated with titanium oxide and a weight
fraction of 40%. The thickness of the titanium layer is unknown but the volume
fraction of iron is in any event larger than 7%. The magnetic stress is found
to be 400 Pa for H = 2000 Oe and close to 1 kPa by adding an electric field
of 1 kV/mm. By comparison for the same field, and without coating, the yield
stress is 5kPa for a volume fraction of 15% [39]. As the proportionality between
the yield stress and the volume fraction is quite well verified in this range of
volume fraction [35], we should expect a yield stress of at least 2.5 kPa for the
magnetic stress alone. It appears that, at least for this fluid, even the synergistic
effect does not compensate the loss of magnetic stress due to the presence of the
titanium oxide layer; progresses can likely be made by decreasing the thickness
of the insulating layer.
The most promising type of MR fluid is simply obtained by using a ferrofluid
as carrier liquid together with carbonyl iron particles. An increase in yield stress
by a factor of four was reported in this situation [40]. The reason for this en-
hancement is easy to understand since the force between two particles, when
mediated by a fluid of relative permeability μf , is multiplied by this perme-
ability (cf. Eq. (4)). Also, due to local alignment and network formation of the
nanodipoles between the ferromagnetic particles, there is no sedimentation. The
remaining problem for industrial use could be chemical stability of the ferrofluid.
Besides MR fluids, new types of magnetic composites are appearing which
are the solid counterparts of ferrofluid and magnetorheological fluids, namely
magnetic gels and magnetic elastomers. Magnetic gels are chemically cross-linked
polymers network swollen by a ferrofluid [41]. Placed in a gradient of magnetic
field they will deform and change of shape; this ability to elongate and contract
makes them looking like artificial muscle [42]. Nevertheless, if they can develop
large change of shape, it is because their zero field elastic modulus is weak; in
practice the energy which is developed is quite low. The second class: magnetic
elastomers, are made by dispersing iron particles in a polymer and applying a
magnetic field prior cross linking. T. Shiga et al [43] reported an increase in shear
modulus of about 10kPa for a 28% volume fraction of iron particles (diameter
about 100μm) at a strain γ = 0.1 and H=59kA/m. At lower strain and with
carbonyl iron particles of average diameter 3-4μm, Jolly et al [6] obtained a much
higher field induced shear modulus: 0.56 MPa for Φ = 0.3, γ = 0.01, ν = 2Hz
and flux density corresponding to saturation (0.8 Tesla). The investigation of
the same kind of material under traction instead of shear gives comparatively
a still higher increase in Young modulus (0.6MPa for γ = 0.05, Φ = 0.25 and
H = 123kA/m) [8]. The applications of these elastomers for semiactive damping
of vibrations have been patented by several companies and their performance
can still be improved by a better control of the column formation before curing.
This remark introduces the next section which deals with the prediction of the
structures formed by the magnetic particles in the presence of a magnetic field.
3 Structuration of a MR Fluid by a Magnetic Field

When a magnetorhological fluid is submitted to a magnetic field the particles will
aggregate and form different kind of structures depending on the initial volume
fraction, and on the parameter λ (Eq. (2)). Actually it is only true if we reach the
equilibrium state and, as the particles are at the limit of the colloidal range, the
way to increase the field in order to reach equilibrium is important. Increasing
the field too quickly will give a kind of labyrinthine structure, whereas increasing
the field slowly gives well separated columns [44,45]; also large volume fractions
(typically above 5%) always tend to form labyrinthine structures [46]. Increasing
the field slowly allows to observe a complex behavior of the transmitted light,
with first a decrease of the transmission followed by an increase which ends
up with a higher transmission than in the initial stage at λ = 0 [47]. Exactly
the same behavior was found in ER suspensions [48,49]. The transmitted light
first decreases because chains of particles, like cylinders, scatter the light more
efficiently than individual particles [49]; then at higher fields the chains gather
into thick fibers and the space between fibers form channels of clear liquid which
transmit the light without attenuation, hence the large increase of transmitted
light. The interpretation of the transmitted light versus λ in terms of critical
fields defining different phases is rather ambiguous. What is not controversial
is that the phase separation begins with the formation of individual chains and
ends up with domains of particles which look like fibers and are well separated at
low volume fraction. The problem of structure equilibrium being quite crucial, it
seems that a good method is not only to rise the field slowly but also to interrupt
it during a short time corresponding to the time needed for a particle to diffuse
on one radius. Not only this procedure allows to better reach the equilibrium
state but also it allows to precisely determine the critical values of λ between the
isotropic and nematic phase and between the nematic phase and the columnar
phase [49]. Instead of measuring only the transmitted light, it can be helpful
to measure also the scattered light as a function of the angle of observation. In
order to have access to the chain length the best geometry is the one where the
light beam is perpendicular to the chains; then the light scattered in the plane
perpendicular to the chains is proportional to the square of the length of the
chains [50,51]. Nevertheless the information on chain length is indirect, and, for
magnetic suspensions, it is restricted to low volume fraction due to the strong
absorption of magnetic suspensions and the impossibility to realize refractive
index matching. The prediction of the average chain length as a function of λ
Magnetorheology 211
has been derived by A. Cebers [52] from the calculation of the free energy of
a perfect gas of chains of spheres. Analogous results have been obtained more
recently with the same model but using a simpler derivation which can be easily
extended to interacting chains. The result for the average number of particles
per chain is [53]:
√
1 + 4y − 1
< n >= y √ with y = Φe2λ (8)
1 + 2y − 1 + 4y
A good agreement was found between the predictions of such model and
experiments on an ER fluid made of monodisperse silica particles [48]. On the
other hand, if we do not look for the equilibrium distribution of chains but rather
for their kinetic of formation it is possible to describe reasonably well the first
stage of the chaining process with the help of the Smoluchowski equation [54].
In a monolayer the aggregation process can be followed with a microscope and
the image analysis shows that the average length follows a power law:< n(t) >=
1 + ktν ; an exponent ν = 0.6 was found for λ = 31 and λ = 0.009 [54] but
a range of exponents (0.37 < ν < 0.6) was also obtained depending on the
experimental conditions [45]. Actually the Smoluchowski equation predicts an
exponent ν = 0.5 for a constant cross section of collision and a mobility of the
aggregate decreasing as the invert of its length [55] (which is not exactly verified
since the mobility of a slender body of length L, along its main axis, decreases
as (ln(L/a) − 0.5)/2πηL [56]).
The second step of the aggregation process is the coarsening of chains into
columns which is still not well understood. We just want to point out a few
results to emphasize the nature of the problems raised by this coarsening process
either in ER or MR fluids. First the field of a periodic chain of dipoles of infinite
length dies off exponentially as cos(πz/a) exp(−π/a) where is the distance in
the perpendicular x-y plane and 2a the period along z. So, except at very short
distances, two infinite chains will not feel each other. Now let us see the situation
for two rigid dipolar chains of finite length. If we consider two parallel rigid rods
(Fig.2a) separated by a distance d and carrying a dipolar density: m’=m/2a, the
interaction energy between the two rods is:
& '
1 m2 1 1 1 1
W = + − −
4πμ0 μf 4a2 2
d2 + zAa 2
d2 + zBb 2
d2 + zBa 2
d2 + zAb
(9)
The energy of interaction is the one of equivalent charges (plus and minus)
located at the end of the rods. If we do not consider very large shifts between
chains, this energy corresponds to a repulsive force, but, except for the very
special case of two chains of the same length and at the same height, there is
also a torque which will rotate the rods. As the chains are close to each other
they will touch during the rotation well before shifting enough to connect end
by end. After two particles of each chain have joined the coalescence occurs
by a “zippering motion” [45] because of the short ranged attractive interaction
between two chains shifted by a radius. Taking L as the order of magnitude of
zBa in Eq. (9) and normalizing the torque K by kT we have K/kT ≈ λad/L2 .
Fig. 2. Chains of spheres aligned by the field (a): Shifted chains at a distance d (b):
Body Cubic tetragonal structure
Besides this simple mechanism of aggregation between chains, Halsey et

al [57] have examined the interaction between chains due to the fluctuations
of positions of the individual particles. The idea behind is that enhancement
of fluctuations in one dimensional structures should give rise to a fluctuating
field whose root mean square is slowly decreasing with the distance (as 1/d2 ).
The resulting interaction energy coming from dipole induced dipole mechanisms
similar to Van der Waals energy is attractive with an order of magnitude:
Wf /kT ≈ (L/a)(a/d)5 and does not depend on λ. If we compare this energy
with torque we see that, at large distances, where these theories are valid, (for
instance d=10a) and moderate field (λ = 10) the energy of the fluctuating field
overcomes the torque only for L/a > 200. At low field both mechanisms are
present in the coarsening of the suspension but at high fields the torque induced
aggregation is likely to be the dominant one. Actually an other clue that other
mechanisms than thermal fluctuations can explain the coarsening process is that
numerical simulations carried out without any Brownian motion also show this
coarsening [58,59]. The time evolution of the thickening of the columns can be
directly measured by light scattering in a configuration where the beam (Oz) is
perpendicular to the field (Ox). In this case the formation of chains or fibers gives
a strong scattering in the direction perpendicular to the field (Oy). The light
scattered along Oy presents a maximum for a scattering wave vector qmax and
−1
L = 2πqmax gives the characteristic distance between aggregates and so indi-
rectly their size. The growth of L in time was found to be L(t) = L(0)(1+(t/τ )α )
with α = 0.42 for a 11 wt. % suspension of 0.7μm silica spheres in electric field
[51]. In magnetic colloidal suspensions most of the experiments are realized with
a thin cell made of two microscope slides separated by a spacer and a field per-
pendicular to the slides [44,46]. The important difference relatively to electric
field is that, with a magnetic field, the boundaries conditions do not require to
Magnetorheology 213
use dipole images in order to cancel the field on the electrodes; then a chain
filling the gap is a finite chain instead of an infinite one as in the case of electric
field. We have seen in Eq. (9) that such finite chains repel each other, so look-
ing from above we could expect to see the cross sections of individual chains.
Actually there will be a compromise between the coarsening coming from phase
separation and this long range repulsive energy which will prevent the fibers to
join each other. It is worth noting that in ER suspensions the equilibrium state
for high values of λ does not consist of individual fibers but of a unique domain
with a crystal like structure where the particles are placed on a Body Cubic
Tetragonal (BCT) lattice. In this lattice one chain of particles is surrounded by
four chains shifted by a radius (Fig. 2b). Such a structure has been proved to
have the lowest energy [60] and has been observed both by laser beam diffrac-
tion [61] end recently by confocal microscopy [62]. In practice, due to transport
limitation, the size of the domains will remain small. Coming back to the case of
magnetic suspensions, the individual fibers of particles can be easily seen with
a microscope. As their shape can be quite irregular - especially at volume frac-
tion larger than 5% - and also because their apparent size can depend on image
thresholding - it is the average distance between fibers rather than their diame-
ter which is measured by microscope observation [46,63]. The other way is to use
the light scattered by the fibers with the beam (Oz) parallel to the field. It will
give rings in the x-y plane and the position of the first ring will give the aver-
age distance between fibers [64]. As the field induced by equivalent end charges
depends on the length of the aggregates - which is equal to the thickness of the
cell at high λ -, then we expect that the distance, d, between domains will also
depends on the thickness of the cell h. It is quite usual to fit the experimental
results with a power law: d = Chv but it appears that the power can vary from
0.37 [64] to 0.67 [63]. The spreading of the results is mainly due to the fact that
there is no power law between d and h as demonstrated by a model taking into
account the repulsive interaction between the fibers and the surface energy. This
model assumes that there are no particles between the domains, also for the sake
of simplicity only two kinds of domains are considered: ellipsoidal aggregates or
stripes of particles; the first case can approximate the domains created in a uni-
directional field, whereas the second one is appropriated for structures created in
a rotating field or in a shear flow. If we call Ma the total dipole of an aggregate,
we have to distinguish the dipoles which are on the surface of the aggregate from
those which are in the bulk because the Lorentz field (coming from the surface
of a virtual cavity around a given particle) will be different on the surface of the
aggregate: indeed on the surface half of the surrounding particles are missing
and so we can divide the Lorentz field, HL , by a factor of 2. We shall have:
Ma = NB mB + NS mS (10)
with mB = α(H0 + HD + HL )
and ms = α(H0 + HD + HL /2)
The polarisability is α = 4πμ0 βa3 (we suppose in the following that μf = 1)

NB , mB , NS , mS are respectively the numbers of particles and the magnetic
moments of a given particle in the bulk of the aggregate and on its surface.
H0 is the external applied field, HD = −ND Ia /μ0 the field coming from the
equivalent charges on the surfaces of the aggregates, HL = Ia /3μ0 is the Lorentz
field, and Ia = Ma /Va the magnetisation inside the domain of volume Va . The
demagnetisation factor, ND , depends on the geometry of the domains [65].The
total magnetic energy per unit volume of the suspension is:
Um /V = −1/2Na Ma H0 /V = −1/2ϕIa H0 (11)
The apparent volume fraction of aggregates ϕ = Na Va /V should not be

confused with Φa which is the volume fraction inside the aggregate, neither
with the average volume fraction Φ. These three volume fractions are related
by ϕ = Φ/Φa . The energy per unit volume is obtained as a function of ϕ and
d∗ = d/h appearing in ND
The magnetic energy can be minimized with respect to d∗ in order to obtain
the average distance between the aggregates but it will be for a given volume
fraction inside the aggregate: Φa which is unknown. At high fields it is reasonable
to suppose that Φa is close to the volume fraction of the BCT structure (Φa =
0.69) [63] but it is possible to obtain its value directly by saying that at high
field the total pressure inside the aggregates is close to zero: Pm + Posm = 0
where Pm = −dUm /dVa is the magnetic pressure and Posm the osmotic pressure.
Taking for the osmotic pressure Posm = 1.85(N kT /V )/(0.64−Φa ) which fits well
the hard sphere pressure at high volume fraction [66] we shall finally have the
two following equations [67]:
- -
∂(Um /V ) -- Φ ∂(Um /V ) --
− νp - + 1.85 = 0 and - =0 (12)
∂ϕ d 0.64ϕ − Φ ∂d ϕ
These two equations can be solved numerically for the two unknowns, d/h
and ϕ(or Φa ), and the solution depends on the initial volume fraction Φ and
on λ. An example of the structure obtained in unidirectional field is shown in
Fig.3a; the experimental data for d versus h are plotted in Fig.3b together with
the solution of Eq. (12) for Φ = 0.045 and λ = 273. The agreement is quite
good and, in this range, is close to d ∝ h0.5 . In this derivation the surface
energy appears through the change of Lorentz field, δHL , on the surface; it is
proportional to the square of the magnetization and to a/h since the radius of
the particles determines the thickness of the layer where the field differs from its
bulk value [67].
Some detailed calculation, either with dipolar [68] or multipolar [69] inter-
actions on different lattices have shown that the surface energy can be quite
sensitive to the type of lattice, but the magnetic suspensions are not monodis-
perse and we do not expect a perfect ordered surface so this “homogeneous”
approximation is likely appropriated. If the surface energy becomes negligible,
then the energy depends only on d/h and ϕ and the solution of Eq. (12) no
longer depends on a/h; in other words d increases linearly with h for a given Φ
and λ. Although the equations (12) could be used to predict the critical field
Magnetorheology 215
Fig. 3. Field induced phase separation. (a): top view of the structure. (b): average
distance between aggregates versus the thickness of the cell: ◦ Eq. (12); experiments
for phase separation (the value of λ above which the free energy becomes min-
imum for a value of ϕ different from unity) it is not believed to give the right
answer because, as stated above, these equations skip the existence of the in-
termediate phase made of a gas of chains. To our knowledge such a theory is
still missing since theories [70,71,49] rely on the hypothesis of a transition from
a diluted isotropic phase to a condensed one, and use the osmotic pressure of a
hard spheres liquid instead of the one corresponding to ellipsoids or rods.
Before ending with field induced structures it is worth saying a few words
about structures in a rotating magnetic field. Let us consider a magnetic field
rotating in the x-y plane: H0 = H0 (ex cos ωt + ey sin ωt) at a frequency high
enough such that the dipolar force between two particles can be averaged over a
period without significant change of the interparticle distance. Then the interac-
tion energy corresponding to two dipoles mα = mβ = m(ex cos ωt + ey sin ωt)
rotating in phase with the field is still given by Eq. (1) whose average on a period
gives: W = m2 /2r3 (3 cos2 θ − 1) where θ is the angle between Oz and r (cf. Fig.
1). This energy is negative for θ = π/2 and positive for θ = 0. It means that the
particles will tend to gather into disks whose axis will be along Oz; these disks
will also tend to separate from each other along the z axis due the repulsive force
for θ = 0. This disk shaped structure was observed with dielectric particles of
diameter 10μm in a rotating electric field [72]. In the case of a magnetic colloidal
suspension, the experiment made in a thin cell with the rotating field perpen-
dicular to the plane of the cell shows a set of stripes which are the projections
of the disk-shaped aggregates (Fig. 4) [73]. The thickness of these stripes being
an order of magnitude larger than the diameter of the particles it seems that
there are composed of many sheets of particles. If the field is rotating at a low
enough frequency, chains of particles have enough time to form and rotate. This
regime can give interesting informations on the kinetics of chain formation and
chain rupture [74]. In the following section we shall see that in some cases the
information on structure can help to explain the rheology.
Fig. 4. Structures in a rotating field. (a): geometry; (b): top view of stripe structure
4 Rheology of Magnetic Fluids

4.1 Rheometry
The rheometers used to study the effect of a magnetic field on the rheology
of MR fluids are usually conventional rheometers which have been adapted to
install either coils or a magnetic circuit in soft iron. The installation of coils
depend on the space available around the rotating tool. A pair of coils in Helmotz
configuration allows to have a quite well defined field on a large diameter and
minimizes field gradient but the maximum field attainable remains small (a few
hundred Oersted). Using a single coil around the rotating tool can allow to reach
1000 Oersted but the coil must be cooled, furthermore it is difficult to control
if the gap between the tools is correctly filled. For fields larger than about 1000
Oersted a magnetic circuit is needed. Either the field is parallel to the axis of
rotation and the polar pieces have the same axis of symmetry (Fig. 5a); either the
geometry is a cylindrical Couette cell and the field is everywhere perpendicular
to the lines of flow(Fig. 5b) [75]. In this last case the field being radial is not
constant throughout the gap and it can be 25% higher on the inner wall of the
yoke than on the outer one [29]; this will attract the particles on the inner wall
and decrease the apparent viscosity. Furthermore the height of the cylinder must
be less than its radius in order to avoid the magnetic saturation of the central
iron rod and this can introduce serious end effects.
In practice the parallel plate or the cone plate geometry is the most conve-
nient device for high field measurements [76]. In each case the gap inside the
magnetic circuit is composed of three parts: a fluid (length ls ) and an air path
in series and an air path (l) in parallel (the different lengths are shown in Fig.
5). The length of the yoke is ly . From Ampère’s theorem N I − Hy ly = Hg l =
Hs ls + H(l − ls ) and flux conservation: Bs = μs Hs = μ0 H we readily get [75]:

Bs 1 ls
= 1+ −1 (13)
Bg μs l
Magnetorheology 217
Fig. 5. Magnetic circuits for rheometry. (a): parallel plate geometry; (b): cylindrical
Couette geometry
where Bg is the field measured with a Hall probe placed between the poles besides
the magnetic fluid. From Eq. (13) we can see that the induction measured besides
the fluid can be quite different from the one inside the suspension and that the
correction can only be done if we know the permeability of the suspension. A
way to overcome this problem is to place the Hall probe between the suspension
and the yoke in order to measure directly Bs . Also with two Hall probes placed
as shown in Fig. 5a it is possible to measure simultaneously the induction inside
the suspension and its permeability by the mean of Eq. (13). Lastly, if the field is
imposed only with an external coil, the induction is practically the one measured
in the absence of the suspension (and the field inside the suspension is H =
H0 /μs (H)).
The parallel-plate or cone-plate geometries are widely used to study the rhe-
ology of magnetic suspensions. In the cone plate geometry the shear rate is
constant inside the suspension but the variable thickness can induce a variable
structure as discussed in the preceding section. In the parallel plate geometry the
gap is constant and easy to change, allowing to get rid of slipping velocities on
the plates but on the other hand the shear rate is not constant (since γ̇ = ωr/h).
For a fluid characterized by a Bingham law: τ = τy + η γ̇ the torque recorded by
the rheometer should be related to the shear rate γ̇R = ωR/h on the rim of the
disk by:
R
2πR3 πR3
T = rτ (r)2πrdr = τy + η γ̇R (14)
3 2
0
The software of the rheometer will use the apparatus constant 2/πR3 to relate
the stress to the torque independently of the rheological law, so multiplying (14)
by this constant and extrapolating at zero shear rate we obtain τ = 4τy /3. We
see that the apparent yield stress given by the rheometer in a parallel plates
geometry is overestimated by a factor 4/3. This problem does not exist with
a cone-plate geometry neither with a sliding plate geometry [77] but this last
technique only gives the static yield stress. In the parallel plates geometry it is
necessary to use the general formulation [78]:

T γ˙R dT
τ (γ˙R ) = 3+ (15)
2πR3 T dγ˙R
Also the use of two different thickness allows to check the existence of a
slipping velocity on the wall and to correct it if any. The determination of the
yield stress, τy , as a function of the induction is the main property of a MR
suspension. Experimentally the static yield stress is obtained by extrapolating
at zero shear rate the stress versus logarithm of shear rate curve (there must be
data until 10−3 s−1 ) or by detecting the change of slope on a stress-strain curve.
But the yield stress is not a well defined quantity and, in many cases, the yield of
a suspension does not represent a bulk property of the fluid but the interactions
between the wall and the particle [77,35]. A more reliable quantity is obtained by
1/2
fitting the rheogram with a Bingham law or a Casson law: τ 1/2 = τd + (η γ̇)1/2
or Hershel-Bukley: τ = τd + (K γ̇)p . The values of τd obtained by fitting these
different laws are quite similar and represent what is usually called the dynamic
yield stress. This stress can be viewed as the one which is needed to continuously
separate the particles against the attractive magnetic forces in the low shear limit
[20,79].
4.2 Models for the Yield Stress

The models aiming to predict the yield stress have to deal with two scales. Firstly,
the particle scale which will give the force between two particles as a function of
their physical properties and respective positions. Secondly the knowledge of the
mesostructure and of its deformation at the scale of the device is needed. Two
situations can be distinguished depending on the permeability of the particles.
If the permeability is high (α = μp /μf 1) it is the particle scale which is
important because the forces strongly depend on the interparticle gap. If α is
low, then we can use the dipolar approximation and the separation between the
particles does not change significantly the energy; this is mostly the shape of
the mesostructure (cylinders, sheets. . . ) and its inclination relatively to the field
which will change the energy and generate the yield stress.
High permeability: α 1
The standard model for the structure is based on a cubic network of infinite
chains of particles aligned in the direction of the field. When the material is
strained, the chains are supposed to deform affinely with the strain (Fig. 6a); in
other words the motion of the particles takes place only along the velocity lines.
The result is that the distance between any pair of neighbors in the chains is the
same and increase at the same rate with the strain γ = tan(θ).
Some more realistic structural models have been considered [80,81] but it
appears that in most cases the standard model gives a good prediction of the
yield stress [82]. The field dependence of the yield stress is usually represented
by a power law τy ∼ H n with 1 < n < 2. The case n = 2 is the case of a
linear magnetic material which is found at low fields or for particulates of low
Magnetorheology 219
Fig. 6. Modelling the yield stress. (a) affine deformation of a chain; (b) gap between
two particles; for < δ, Hg =Ms
permeability [83]. The effect of saturation on the yield stress can be described
by a simple model which can be applied both to ER [84] and MR [82] fluids.
In this model the permeability of the particles is high enough to neglect the
field Hi inside the particles. Then we can distinguish two domains inside the
gap between two particles (Fig. 6b): the pole region with < δ ( is the polar
coordinate) where the field is given by the saturation magnetization: Hg = Ms
and the other with < δ where the field is given by Hg = H(a + 0.5w)/h()
where H is the average field in the suspension, a, the radius of the particles
and h() the distance between the plane of symmetry and the surface of the
sphere: h() ∼
= 0.5 ∗ (w + 2 ) with w the minimum gap between the two spheres
and = w/a. The distance δ is obtained by equating Hg and Ms for = δ. If
H/M s 1 the radial force between the two spheres which is the integral of the
field on the plane separating the two spheres can be developed as:
a
μ0 H H
Fr = (Hg − H) 2πρdρ ≈ πa
2 2
μ0 Ms2 − + πa2 μ0 Ms2 (16)
2 Ms 2 Ms
0
The first part of right hand side of Eq. (16) corresponds to < δ and the
second part to > δ. Knowing the force between the two spheres, the shear
stress as a function of the strain γ = tg(θ) ≈ sin(θ) is given by:
Fr 3 Fr γ
τ (γ) = N 2
sin(θ) cos2 θ = Φ 2 (17)
L 2 πa 1 + γ 2
The quantity N/L2 is the number of chains by unit surface and Φ is the
volume fraction of solid particles. The cos2 (θ) comes from the projection of the
field squared on the unit vector joining the two spheres and the sin(θ) of the
projection of the radial force
on the direction of shear. From Eqs. (16), (17) and
using = 2 1 + γ − 1 we obtain an expression for the shear modulus as a
2
function of the strain which is valid for γ 1. The slope of this curve for γ = 0
gives the shear modulus and the maximum for γ = γc gives the yield stress which
are respectively:
G = 3μ0 ΦHMs and τy = 2.31Φμ0 Ms1/2 H 3/2 ; γc = 2H/(1.5Ms + 6H)

(18)
It is also possible to solve directly for the force between two spheres in the
presence of a constant external field or for two spheres which are part of a pe-
riodic chain by using finite element analysis [82,85]. In the case of an infinite
chain, the unit cell contains one particle and the volume of the cylindrical cell is
obtained from the volume fraction. The magnetic induction on the outer cylin-
drical boundary is imposed to be the average induction. Then the field can be
calculated on the nodes of the mesh and the force is still given by the left-hand
side of Eq. (16).
Both finite element and analytical expression predicts a yield stress propor-
tional to the volume fraction and that is also what is observed experimentally
at not too high volume fractions (Φ < 0.2 − 0.3) [86,22] but an increase faster
than linearly [35,87] is observed at higher volume fraction. This behavior can be
expected if we believe that thick aggregates are more difficult to break than indi-
vidual chains; it can also be related to non affine motion which leads to rupture
of aggregates at higher angle-and so higher stresses [32,81].
Concerning the field dependence on the yield stress, several groups have found
a power law with an exponent close to 3/2 as predicted by Eq. (18) [86,88,24], but
some data show a linear dependence [39], or an exponent 1.27 - but at quite low
fields and with steel spheres [90] -, and we can find results [77] or simulations [89]
showing a continuous decrease of the slope with the field, so it is rather difficult
to conclude at this stage.
A way to check the theories is to use steel spheres whose magnetic properties
can easily be measured in the bulk material. The magnetization curve is well
fitted by a Frolich-Kennelly curve: M = Ms (μi − 1)H/(Ms + (μi − 1)H) with
Ms = 1360 kA/m and μi = 250. The calculation of the radial force Fr between
two spheres is performed by using either finite elements or Eqs. (16), (17). The
experiment consists in shearing chains of seven steel spheres placed on a ring
inside a rheometer and glued at each extremity [90]. The results are shown in
Fig. 7a for a field H = 14.3kA/m and an equivalent volume fraction of 15%. The
finite element calculation is closer to the experimental result but the analytical
solution gives the right behavior, even if the predicted yield stress is higher than
the experimental one.
The maximum yield stress at high induction will be given by magnetic satu-
ration since each particle will acquire the dipole: ms = vp Ms ; then using affine
displacement of the particles, the shear stress is obtained from the derivative of
the energy relatively to the strain hence the yield stress and the shear modulus
[6]:
τys = 0.1143ζ(3)Φμ0 Ms2 and Gs = 0.294ζ(3)Φμ0 Ms2 (19)
For Φ = 0.3 and μ0 Ms = 2 Tesla, we have τys = 130 kPa and Gs =340 kPa.
These are actually the order of magnitude which are obtained experimentally
Magnetorheology 221
Fig. 7. Theories for yield stress (a) stress-strain curve for chains of steel spheres ◦ ex-
periment; finite elements; Eqs. (16)-(17) (b) normalized yield stress versus volume
fraction for silica spheres (diameter 40-50μm) at low field domain (10-30 kA/m); solid
curve: experiment; dashed curve: theory
both for saturation yield stress [24,15] and shear modulus [6]. Other models
based on the same assumptions of a saturation zone give similar predictions
[82,91].
Low permeability α < 5
An other quite different situation occurs for low values of the ratio α = μp /μf .
A way to obtain this situation is to use a mixture of silica spheres in a ferrofluid.
Then we have a suspension of magnetic holes (α < 1) whose permeability for
different volume fractions well follow the Maxwell-Garnett law [35]. In this case
the dipolar approximation can be used and the interparticle forces being long
ranged, the total magnetic energy is no longer sensitive to the change of the inter-
particle distance but rather to the deformation and rotation of the macroscopic
aggregates in the field. If the magnetic moment of the anisotropic medium can
be obtained as a function of the shape and orientation of the mesostructure then
the energy and the magnetic stress can be derived since we have: τ = 2V 1
H0 ∂m
∂γ
z
with mz the component of the magnetic moment of the sample along the direc-
tion of the external field. The aggregates formed in the presence of the field are
either cylinders or stripes of internal volume fraction Φa . The volume fraction
ϕ = Φ/Φa represents the part of the space occupied by the aggregates. If H is
the field inside the suspension, then the predicted yield stress is [81,92]:
√
τy 3 3 ∗ 2 ϕ(1 − ϕ)
= (μs ) (20)
μ0 μf H 2 16 Cs + μ∗s (1 − ϕ)
with Cs = 1 if the particles are gathered into stripes and Cs = 2 if they form
cylinders. The permeability μ0 μf is the one of the suspending fluid (here the
ferrofluid) and μ∗s = μs /μf − 1.
A similar derivation was proposed independently for stripes, with a slight
difference on the analytical result [93,81]. We have measured the dynamic yield
stress by extrapolating at zero shear rate the stress versus shear rate curve, as-
suming a Bingham behavior τ = τy + η γ̇. The comparison of the normalized
experimental yield stress with the one predicted by Eq. (20) with Cs = 2 (cylin-
drical aggregates) is shown in Fig. 7b for different volume fractions and a field
H = 7 kA/m. We see that Eq. (20) well captures the experimental behavior,
especially taking into account that there are no free parameters (the internal
volume fraction, Φa , has been set to 0.69 which is the volume fraction of a body
cubic tetragonal structure). Besides the quantitative agreement it is worth not-
ing that the yield stress presents a maximum with the volume fraction; this is
because the difference in volume fraction between the aggregates and the effec-
tive medium is decreasing with the volume fraction and the shear stress is only
a function of this difference. On the contrary for particles of high permeability,
the stress is mainly sensitive to short range interparticle forces and increases
linearly with Φ.
More recently this approach for a striped mesostructure was extended to
cover both low and high values of α [94]. The permeability of the stripes is
obtained from a formula valid for a cubic lattice of particles whatever α, and the
internal permeability of the particles μp is approximated by a Frolich-Kennelly
law in order to take into account the magnetic saturation.
Instead of simple chains arranged on a network (cubic or bct) some more
complicated arrangements of chains have been studied either with dipolar ap-
proximation or with multipolar development which should be considered for high
permeabilities [80,81,95,96]. The main point to remember is that the yield stress
is not sensitive to the structure itself (single chains, double chains, planes or
double planes give approximately the same values) but to the way it deforms.
Besides the affine motion with equally spaced spheres we can consider a model
with the formation of a single gap or with a single gap but starting at non zero
strain. This last case where, due to extra particles, the structure can expand
before breaking, gives a higher yield stress [32]. An other point which is still de-
bated is relative to the position where the ruptures take place. In a chain model,
due to the finite size, the force needed to separate two particles inside the chain
is weaker at its extremity and the first rupture should take place close to the
walls. For ER fluids with dipole images it is not so intuitive but still verified
[97]. This is true if each extremity of the chain sticks to the wall, but if it slips
and rotates the hydrodynamic tension being maximum at the center it will likely
breaks at the center. Before breaking, the suspension can also slip on the walls
at a lower stress which will depend on the roughness of the walls and also on
the pressure exerted by the magnetic particles on the wall in the presence of the
field. In this case a frictional shear stress (which is the normal stress times the
friction coefficient) is observed [35]. Also it has been found that dense aggregates
can stand high normal stresses obtained by moving a wall in the presence of the
magnetic field. In this case very high frictional shear stresses following the law
τye = τy + kNe where τy is the usual yield stress, Ne the external stress (up to
2M P a) and k a friction coefficient close to 0.25 [98].
This presentation holds only for non Brownian particles (λ → ∞ ). For finite
values of λ the Brownian motion will play a role to weaken the structure in
two different ways: by decreasing the length of the aggregates and by generating
Magnetorheology 223
fluctuations of the gaps between the particles inside the chains, which will lower
the rupture force. The first effect can be accounted for with the help of the theory
leading to Eq. (8) or a similar one [99]. Then the only aggregates participating
to the yield stress are those which have a length larger than the thickness of
the cell [100].This model ignore the importance of the fluctuations of positions
inside the chain which can considerably decrease the force needed for the rupture
of a chain. It is possible to estimate the average distance between two particles
inside a chain with the help of the Boltzmann distribution, but the breakdown
of the percolated network is not sensitive to the average distance between the
particles but rather to the largest fluctuation inside the chain. A Monte-Carlo
calculation shows that the largest gap inside a chain of 200 particles at λ = 300
can be an order of magnitude greater than the average one and can explain the
lower yield stress experimentally observed when λ is decreased [33]. On the other
hand when a mechanical stress is imposed, the corresponding energy should be
incorporated in the Boltzman distribution as is usually done in the standard
theory of viscosity of simple liquids (the probability for a particle to jump from
a potential well to an other is e−(Ud ±vτ )/kT with v the volume of a unit cell and τ
the applied shear stress; the + sign being for the direction of the applied stress).
This approach is used directly in [101] and incorporated in a chain model in [102];
no direct comparison with experimental results concerning the yield stress are
done but rather concerning the Mason number dependence which is the subject
of the next section.
4.3 Flow Regime

In steady shear flow the Bingham model: τ = τd + η0 γ̇ is widely used. If we
call ηs the effective viscosity (defined by τ = ηs γ̇), then normalizing by the
stress at zero field gives a relative viscosity: ηr = ηs /η0 which decreases as
τd /η0 γ̇ that is to say proportionally to M n−1 where M n is defined in Eq. (6).
Actually experiments [107,35,76,83] as well as simulations [101,103] show that
the Mason number does not allow to collapse all the results on the same curve.
The reduced viscosity still follows a law ηr = M n−ν but with an exponent
which can vary from 0.68 to 1 with a tendency to increase when λ is increased.
In the limit ( λ → ∞ ) either a model based on ellipsoidal aggregates [75] or on
chain of particles [102,35] predicts an exponent ν = 1. These models are based
on the idea that aggregates rotate in the shear flow and find their equilibrium
angle when the hydrodynamic torque equilibrates the magnetic restoring torque.
The rotation continues when the Mason number is increased until the radial
hydrodynamic force at the center of the chain overcomes the attractive magnetic
force. At this critical angle the aggregates break into two equal parts and the
restoring magnetic torque - which is proportional to the volume of the aggregate
- is divided by two. The hydrodynamic torque and the shear stress which are
proportional to the cube of the length of the aggregate are divided by eight, so
the two aggregate rotate backwards after their separation and will reform and
break again at the same angle. Increasing further the shear rate, the equilibrium
length will decrease until all the aggregates are destroyed. As pointed in [104] the
two equations relative to the torque and to the radial force give a solution for the
critical angle which does not depend on the size of the chain, but which increases
if we consider multipolar forces rather than dipolar ones [35]. It is worth noting
that this approach gives a Bingham law with a yield stress τd = Cμ0 μf β 2 H 2 ,
where C is a constant of order one. The order of magnitude of this dynamic yield
stress is the same as the one obtained in a dipolar approximation and taking
into account multipolar interactions will increase the critical angle well above
45o which will still decrease the yield stress. The failure of the rotating chain
model in predicting the dynamic yield stress for strong fluids could come from
the interactions between aggregates which prevent them to reach the critical
angle. These interactions are not directly introduced in the models.
Still it is the interaction between chains which leads to mesostructure for-
mation (usually stripes of particles in the plane defined by the field and the
velocity) at quite low strain (γ = 0.15) as observed experimentally in oscillating
shear [67] or also in steady shear flow [106,73]. The average inclination of chains
and their rupture is then a process mediated by the interactions between chains
inside dense structures rather than by individual rotation and rupture.
Another model [107] based on an ellipsoidal shape of aggregates also uses a
balance between the hydrodynamic and the magnetic torque to find the equi-
librium angle θ, but the determination of the size of the aggregates is obtained
by minimizing the total magnetic energy (which includes the surface energy).
The result is a viscosity which, unlike the preceding model, should vary as:
ηr = 1 + CM n−2/3 . This theory is restricted to low angles of inclination of the
aggregates, or in other words to low Mason number. As the experiments show
exponents which are between 2/3 and 1 it can be tempting to conclude that
the two models correspond to some limit (for instance ν = 2/3 for low λ and
ν = 1 for λ → ∞). The simulations made with Stokesian dynamics [101] do
not comfort this point of view since, for 10 < λ < ∞, the exponent is between
0.9 and 1 for low Mason numbers (2 · 10−5 < M n < 10−3 ) instead of varying
between 2/3 and 1.
Introducing the parameter λ in the theory in order to model the effect of
Brownian motion on the rheology is a quite difficult task. In [102] the chain
model developed for λ → ∞ is extended by introducing an activation energy
which contribute to break the chains before they have reached their critical
angle. A life time of the chains which depends on λ and M n is introduced in this
way and compared to 1/γ̇ in order to determine if the chains will break before
reaching their equilibrium angle. This model predicts that the normalized field
induced stress should scales as (λ3/2 M n)1/λ but comparisons with experiments
are lacking. A different model using the Eyring approach of viscosity is used in
[101]. This model predicts a plateau at very low shear for the viscosity (which
diverges as e1.57λ ). Above this low shear regime the apparent viscosity should
only depend on the dimensionless shear rate: (γ̇a2 /Dλ)e1.57λ . This prediction is
well verified by numerical simulations in the range 2.5 < λ < 17. For particles
carrying a constant magnetic dipole, a low Mason number theory is proposed in
[105] where the size distribution of chains as a function of λ is the one obtained at
Magnetorheology 225
zero Mason number. Then the distribution function of orientations of the chains
for each length is obtained from a Fokker-Planck equation balancing the fluxes
due to hydrodynamic and magnetic torques and to Brownian relaxation. The
hydrodynamic and magnetic stresses are then averaged over the distribution of
lengths and over the distribution of orientations. Some results for a low value of
λ-independent of field for constant dipoles - compare the increase of viscosity for
field aligned on the velocity gradient and field aligned on the velocity, showing
a maximum with the intensity of the field in the second case. Here too, no
comparisons with experiments are done. Mixing thermal forces and shear forces
to predict the equilibrium distribution of chains as a function of λ and M n is
difficult because the force associated with the flow is not the derivative of a
potential energy. Nevertheless it is the case if we consider a chain of particles
aligned on the direction of elongation in an elongational flow. Then it becomes
possible to add the potential energy of the flow: Uf l = −1/2ξEij ri rj with ξ the
Stokes coefficient, r, the separation vector between two particles and Eij the
velocity gradient tensor. The use of the thermodynamic theories is then possible
and the evolution of the chain length as a function of λ and M n is given in [108].
The knowledge of viscoelastic properties of MR fluids such as the storage
and loss moduli, G , G is quite important for damping applications but few
systematic studies are reported in the literature. As was first pointed in [109]
the use of a dimensionless frequency which is equivalent to a Mason number in
oscillating flow, and of a modulus normalized by the energy density of the field
allows to collapse the data obtained at different fields on the same master curve.
This is verified by experiments on MR fluids if λ is high enough [106], otherwise,
as for steady flow, Brownian forces introduce an other time scale. The moduli G
and G do not depend markedly on frequencies whereas they strongly decrease
with the strain [87]. The viscoelastic theories use the same approach as for steady
flow: the aggregates are modeled by independent chains or ellipsoids. A kinetic
chain model taking into account aggregation due to attractive dipole forces and
rupture caused by hydrodynamic forces was proposed [110] by introducing a
phenomenological equation for the evolution of the chain length. On the other
hand the solution of the equation of motion of an ellipsoid in an harmonic shear
flow - with a fixed length depending on λ - is an other possible approach [111].
A comparison of these models with experimental data on G , G is lacking.
Most of the models are based on the description of individual aggregates and
ignore the effect of mesostructures on magnetorheology. This effect is obvious
at low Mason number where it explains that the first stress-shear rate curve,
obtained under steady shear flow after applying the field on an homogeneous
suspension, is different from the second one (Fig. 8): the first flow curve is ob-
tained starting from a mesostructure formed of cylinders, whereas the second
and subsequent curves are obtained starting from a sheet like structure [106].
This sheet like structure gradually disappears when the Mason number is in-
creased but at a Mason number around unity a new striped structure appears
associated with a jump in viscosity [112]. This flow induced phase separation is
due to the combination of the rotation of transient aggregates of particles in the
Fig. 8. Stress versus shear rate for suspension of magnetic polystyrene particles
(d=0.5μm); applied field H=29 kA/m. The top-left photo corresponds to the struc-
ture before shearing and the bottom photo to the structure after shearing
shear (which begins above M n = 1) and of an uniaxial applied magnetic field;

this situation is actually reminiscent of the case of a rotating field discussed
in the preceding section. More complicated mesostructures can be obtained by
combining rotating field with axial field or crossed electric field and magnetic
field or any other combination.
5 Conclusion
Careful experiments made with well characterized magnetic suspensions are still
too scarce in order to check existing rheological models, especially for fluids with
high saturation magnetization. It would be very useful for this purpose to syn-
thesize well monodispersed magnetic particles in the range 0.1 to 1μm where it
is possible to explore the whole range of λ without serious problems with sedi-
mentation. Progresses in the long term stability of these fluids are more expected
than a large increase in yield stress, except may be in specially confined geome-
tries. Besides applications using viscosity control which are reviewed elsewhere
[15,16] let us point out that the use of rotating fields or combined electric and
magnetic fields to arrange colloidal particles in special structures, has also very
promising applications in the domain of nanotechnologies.
References
1. W.M.Winslow, J. Appl. Phys. 20, 1137 (1949)
2. J.Rabinow: AIEE Trans. 67,1308 (1948)
Magnetorheology 227
3. J.D.Carlson, D.M.Catanzarite, K. A. St. Clair: “Commercial magneto-rheological

devices”. In: Proceedings of the 5 th International Conference on ER fluids,
MR suspensions and associated technology, Sheffield, July 10-14, 1997, ed. by
W.A.Bullough (World Scientific 1996) pp. 20-28
4. A.Inoue, S.Maniwa, Y.Ide, H.Oda: Int. J. Mod. Phys. B 13, 1966 (1999)
5. H.Orihara,M.Doi,Y.Ishibashi: Int. J. Mod. Phys. B 13, 1949 (1999)
6. M.R.Jolly, J.D.Carlson, B.C.Munoz, T.A.Bullions: J. Intell. Mater. Sys. Struct.
7, 613 (1996)
7. J.M.Ginder, M.E.Nichols, L.D.Elie, J.L.Tardiff: ‘Magnetorheological elastomers;
properties and applications’. In Smart Structures and Materials 1999: Smart Ma-
terials Technologies, Newport Beach CA, 3-4 March 1999, ed. by M.Wutttig,
(SPIE Proceedings Vol. 3675, 1999) pp.131-138
8. G.Bossis, C.Abbo, S.Cutillas, S.Lacis, C Métayer: Int. J. Mod. Phys. B, 15, 564
(2001)
9. H.Block, J.P.Kelly: J. Phys. D: Appl. Phys. 21, 1661 (1988)
10. A.P.Gast, C.F.Zukoski: Adv. Coll. Int. Sci. 30, 153 (1989)
11. T.C.Jordan, M.T.Shaw: IEEE Trans. Elect. Insul. 24, 849 (1989)
12. C.F.Zukoski: Ann. Rev. Mat. Sci. 23, 45 (1993)
13. M.Parathasarty, D.J.Klingenberg: Material Science and Ingeniering R17, 57
(1996)
14. H.T.See: Korea-Australia Rheology Journal 11, 169 (1999)
15. J.M.Ginder: MRS Bulletin 23, 26 (August 1998)
16. P.J.Rankin, J.M.Ginder, D.J.Klingenberg: Curr. Opin. Colloidal Interface Sci. 3,
373 (1998)
17. H.See: Applied rheology 11, 70 (2001)
18. C.Kormann, H.M.Laun, H.J.Richter: Int. J. Mod. Phys. B 10, 3167 (1996)
19. J.F.Brady, G.Bossis: Ann. Rev. Fluid. Mech. 10, 111 (1988)
20. R.T.Bonnecaze, J.F.Brady: J.Rheol. 36, 73 (1992)
21. D.M.Heyes, J.R.Melrose: Mol. Sim. 5, 293 (1990)
22. V.I.Kordonsky, Z.P.Shulman, S.R.Gorodkin, S.A.Demchuk, I.V.Prokhorov,
E.A.Zaltsgendler, B.M.Khusid, J. Magn. Magn. Mater. 85, 114 (1990)
23. A.J.Margida, K.D.Weiss, J.D.Carlson: Int. J. Mod. Phys. B 10, 3335 (1996)
24. P.P.Phulé, J.M.Ginder: Int. J. Mod. Phys. B, 13, 2019 (1999)
25. J. David Carlson, “What makes a good MR fluid” In: Proceedings of the -
8 th International Conference on ER fluids, MR suspensions,Nice, July 9-13, 2001,
ed. by G.Bossis (World Scientific 2002) pp. 63-69
26. R.T.Foister US Patent 5,667,715, Sep 16,1997 Magnetorheological fluids
27. R.W.Chantrell, A.Bradbury, J.Popplewell, S.W.Charles: J. Appl. Phys. 53, 2742
(1982)
28. R.E.Rosensweig: Ferrohydrodynamics (Cambridge University Press, New York
1985)
29. H.M.Laun, C.Kormann, N.Willenbacher: Rheol. Acta 35, 417 (1996)
30. N.Rosenfeld, N.M.Wereley, R.Radakrishnan, T.S.Sudarshan: ‘Behavior of mag-
netorheological fluids utilizing nanopowder iron’. In: Proceedings of the
8 th International Conference on ER fluids, MR suspensions and their applica-
tions, Nice, July 9-13, 2001, ed. by G.Bossis (World Scientific, Singapore 2002)
pp.452-458
31. B.E.Kashevskii, V.I.Kordonskii, I.V.Prokhorov: Magnetohydrodynamics 24, 134
(1988)
32. G.Bossis, E.Lemaire: J.Rheol 35, 1345 (1991)
33. E.Lemaire, A.Meunier, G.Bossis, J.Liu, D.Felt, P.Bashtovoi, N.Matoussevitch,

“Influence of the particle size on the rheology of magnetorheological flu-
ids”.J.Rheol.39(1995)1011
34. B.J.De Gans, C.Blom, A.P.Philipse, J.Mellema: Phys. Rev E 60, 4518 (1999)
35. O.Volkova, G.Bossis, M.Guyot, V.Bashtovoi, A.Reks: J. Rheology 44,91 (2000)
36. W.I.Kordonsky, S.R.Gorodkin, E.V.Medvedeva: ‘First experiment on magneto-
electrorheological fluids’. In: Electrorheological Fluids: Mechanisms Properties,
Technology, and Applications, Proceedings of the 4 th International Conference on
ER Fluids, July 20 - 23, 1993 Feldkirch, Austria, ed. by R.Tao, G.D.Roy (World
Scientific, Singapore, 1994) pp.22-36
37. K.Koyama: Int. J. Mod. Phys. B 10, 3067 (1996)
38. H.Takeda, K.Matsushita, Y.Masubuchi, J.I.Takimoti, K.Koyama: ‘Mechanism of
the synergistic effect in EMR fluids studied by direct observation’. In: Proceed-
ings of the 6 th International Conference on ER fluids, MR suspensions and their
applications, Yonezawa, July 22-25, 1997, ed. by M.Nakano, K.Koyama (World
Scientific 1998) pp. 571-577
39. F.Q.Jiang, Z.W.Wang, J.Y.Wu, L.W.Zhou, Z.Y.Ying: ‘Magnetorheological ma-
terials and their application in shock absorbers’. In: Proceedings of the
6 th International Conference on ER fluids, MR suspensions and their applications,
Yonezawa, July 22-25, 1997, ed. by M.Nakano, K.Koyama (World Scientific 1998)
pp. 494-501
40. J.M.Ginder, L.D.Elie, L.C.Davis: Magnetic fluid-based magnetorheological fluids,
USA Patent 5,549,837 August, 27, 1996
41. L.Barsi, A.Büki, D. Szabo, M.Zrinyi: Progress in Colloid and Polymer Science
102, 57 (1996)
42. A. Zrinyi, D. Szabo: Progress in Liquid Physics (II), 238 (2000)
43. T.Shiga, A.Okada, T.Kurauchi: J. Appl. Polym. Sci. 58, 787 (1995)
44. T.Mou, G.A.Flores, J.Liu, J.Bibette, J.Richard: Int. J. Mod. Phys. B 8, 2779
(1994)
45. M.Fermigier, A.P.Gast: J. Colloid and Interf. Sci. 154, 522 (1992)
46. E.Lemaire, Y.Grasselli, G.Bossis: J. Phys II France 2, 359 (1992)
47. Y.Zhu, M.L.Ivey, P.Sheaffer, J.Pousset, J.Liu : Int. J. Mod. Phys. B 10, 2973
(1996)
48. S.Cutillas, A.Meunier, E.Lemaire, G.Bossis: Int. J. Mod. Phys. B, 10, 3093 (1996)
49. S.Cutillas, G.Bossis, E.Lemaire, A.Meunier, A.Cebers: Int. J. Mod. Phys., 13,
1791 (1999)
50. M.Hagenbüchle, P Scheaffer, Y.Zhu, J.Liu: Int. J. Mod. Phys. B, 10, 3057 (1996)
51. J.E.Martin, J.Odinek, T.C.Halsey: Phys. Rev. Lett. 69, 1524 (1992)
52. A.O.Cebers: Magn.Gidrodinam. 2, 36 (1974)
53. A.Y.Zubarev: Colloid Journal 61, 338 (2001)
54. S.Fraden, A.J.Hurd, R.B.Meyer: Phys. Rev. Lett. 63, 2373 (1989)
55. R.Jullien, R.Botet: Aggregation and fractal aggregates (World Scientific, Singapore
1987)
56. G.K.Batchelor: J. Fluid Mech. 44, 419 (1970)
57. T.C.Halsey, W.Toor: J.Stat.Phys. 61, 1257 (1990)
58. M.Mohebi, N.Jamaski, J.Liu: Phys. Rev. E 54, 5407 (1996)
59. H.V.Ly, K.Ito, H.T.Banks, M.R.Jolly, F.Reitich: Int. J. Mod. Phys. B, 15, 894
(2001)
60. R.Tao, J.M.Sun: Phys. Rev. Lett. 67, 398 (1991)
61. T.J.Chen, R.N.Zitter, R.Tao: Phys. Rev. Lett. 68, 2555 (1992)
Magnetorheology 229
62. A.van Blaaderen: J.Chem.Phys. 112, 3851 (2000)

63. Y.Grasselli, G.Bossis, E.Lemaire: J. Phys II France 4, 253 (1994)
64. J.Liu, E.M.Lawrence, A.Wu, M.L.Ivey, G.A.Flores, K.Javier, J.Bibette,
J.Richard: Phys. Rev. Lett. 74, 2828 (1995)
65. C.Flament, J.C.Bacri, A.Cebers, F.Elias, R.Perzynski: Europhys. Lett. 34, 225
(1996)
66. W.B.Russel, D.A.Saville, W.R.Schowalter: Colloidal dispersions (Cambridge Uni-
versity Press, Cambridge 1991)
67. S.Cutillas, G.Bossis, A.Cebers: Phys. Rev. E 57, 804 (1998)
68. W.R.Toor, T.C.Halsey: Phys. Rev. A 45, 8617 (1992)
69. H.J.H.Clercx: J.Chem.Phys.98, 8284 (1993)
70. K.I.Morozov, A.F.Pshenichnikov, Y.L.Raikher, M.I.Shliomis: J. Magn. Magn.
Mater. 65, 269 (1987)
71. B.M.Berkovsky, V.I.Klikmanov, V.S.Filinov: J. Phys. C: Solid State Phys. 18,
L941 (1985)
72. T.C.Halsey, R.A.Anderson, J.E Martin: Int. J. Mod. Phys. B, 10, 3019 (1996)
73. G.Bossis, P.Carletto, S.Cutillas, O.Volkova: Magnetohydrodynamics 35, 371
(1999)
74. S. Melle, G.G.Fuller, M.A.Rubio: Phys. Rev. E 61, 4111 (2000)
75. Z.P.Shulman, V.I.Kordonsky, E.A.Zaltsgendler, I.V.Prokhorov, B.M.Khusid,
S.A.Demchuk: Int. J. Multiphase Flow 12, 935 (1986)
76. B.J.de Gans, H.Hoekstra, J.Mellema: Faraday Discuss 112, 209 (1999)
77. X.Tang, H.Conrad: J. Rheology 40, 1167 (1996)
78. K.Walters: Rheometry (Chapman and Hall, London 1975)
79. E.Lemaire, G.Bossis, Y.Grasselli: Langmuir 8, 2961 (1992)
80. G.L.Gulley, R.Tao: Phys. Rev. E 48, 2744 (1993)
81. G.Bossis, E.Lemaire, O.Volkova, H.Clercx: J.Rheol. 41, 687 (1997)
82. J.M.Ginder, L.C.Davis, L.D.Elie: Int. J. Mod. Phys B. 10, 3293 (1996)
83. D.W.Felt, M.Hagenbuchle, J.Liu, J.Richard: J. Intell. Mater. Sys. Struct. 7, 1167
(1996)
84. N.Felici, J.N.Foulc, P.Atten: Int. J. Mod. Phys. B, 8, 2731 (1994)
85. S.Lacis, O.Volkova, G.Bossis: ‘Application of FEM for magnetic force calculation
between ferromagnetic spheres’. In: Proceedings of the International Colloquium
“Modelling of Material Processing”, Riga, May 28-29, 1999, ed. by A.Jakovics
(University of Latvia, Riga 1999) pp. 222-227
86. P.J.Rankin, A.T.Horvath, D.J.Klingenberg: Rheol. Acta 38, 471 (1999)
87. B.D.Chin, J.H.Park, M.H.Kwon, O.O.Park: Rheol. Acta 40, 211 (2001)
88. Z.Y.Chen, X.Tang, G.C.Zhang, Y.Jin, W.Ni, Y.R.Zhu: ‘A novel approach of
preparing ultrafine magnetic metallic particles and the magnetorheology mea-
surements for suspensions containing these particles’. In: Proceedings of the
6 th International Conference on ER fluids, MR suspensions and their applications,
Yonezawa, July 22-25, 1997, ed. by M.Nakano, K.Koyama (World Scientific, Sin-
gapore 1998) pp. 486-493
89. S.Lacis, E.Zavickis, G.Bossis: ‘Magnetic interactions of chains formed by ferro-
magnetic spheres’. In: Proceedings of the 8 th International Conference on ER flu-
ids, MR suspensions and their applications, Nice, July 9-13, 2001, ed. by G.Bossis
(World Scientific, Singapore 2002) pp.359-365
90. O.Volkova, G.Bossis, S.Lacis, M.Guyot ‘Magnetorheology of a magnetic steel
sphere suspension” Proceedings of the 8 th International Conference on ER fluids,
MR suspensions and their applications, Nice, July 9-13, 2001, ed. by G.Bossis
(World Scientific, Singapore 2002) pp. 860-866
91. L.C.Davis: J. Appl. Phys. 85, 3348 (1999)

92. E.Lemaire, G.Bossis, O.Volkova: Int. J. Mod. Phys. B, 10, 3173 (1996)
93. R.E.Rosensweig: J. Rheol. 39, 179 (1995)
94. X.Tang, H.Conrad: J. Phys. D: Appl. Phys. 33, 3026 (2000)
95. X.Tang, Y.Chen, H.Conrad: J. Intell. Mater. Sys. Struct. 7, 517 (1996)
96. H.J.H.Clercx, G.Bossis: Phys. Rev. E 103, 9426 (1995)
97. R.Tao, J.Zhang, Y.Shiroyanagi, X.Tang, X.Zhang: Int. J. Mod. Phys. B 15, 918
(2001)
98. X.Tang, X.Zhang, R.Tao: “Enhance the yield shear stress of magnetorheo-
logical fluids” Proceedings of the 7th Int. Conf. on Electro-rheological Flu-
ids and Magneto-rheological suspensions, Hawaı̈,Honolulu, July 19-23,1999 ed.
R.Tao,World Scientific,2000 pp.3-10
99. M.A.Osipov, P.I.C.Teixeira, M.M.Telo de Gama: Phys. Rev. E 54, 2597 (1996)
100. B.J.de Gans, N.J.Duin, D.van den Ende, J.Mellema: J. Chem. Phys. 113, 2032
(2000)
101. Y.Baxter-Drayton, J.F.Brady: J. Rheol. 40, 1027 (1996)
102. J.E.Martin: Int. J. Mod. Phys. B. 15, 574 (2001)
103. J.R.Melrose: Mol.Phys. 76, 635 (1992)
104. J.E.Martin, R.A.Anderson: J. Chem. Phys. 104, 4814 (1996)
105. A.Yu.Zubarev, L.Yu.Iskakova: Phys. Rev E 61, 5415 (2000)
106. O.Volkova, G.Bossis, E.Lemaire: ‘Magnetorheology of model suspensions’. In:
Proceedings of the 6 th International Conference on ER fluids, MR suspensions
and their applications, Yonezawa, July 22-25, 1997, ed. by M.Nakano, K.Koyama
(World Scientific 1998) pp. 528-524
107. T.C.Halsey, J.E.Martin, D.Adolf: Phys. Rev. Lett. 68, 1519 (1992)
108. [108] J.Pérez-Castillo, A.Pérez-Madrid, J.M.Rubi, G.Bossis: J. Chem. Phys. 113,
6443 (2000)
109. D.J.Klingenberg: J. Rheol. 37, 199 (1993)
110. J.E.Martin, J.Odinek: J.Rheol. 39, 995 (1995)
111. A.Y.Zubarev: Colloid Journal 62, 317 (2000)
112. O.Volkova, S.Cutillas, G.Bossis: Phys. Rev. Lett. 82, 233 (1999)
Targeted Tumor Therapy with “Magnetic
Drug Targeting”: Therapeutic Efficacy
of Ferrofluid Bound Mitoxantrone
Ch. Alexiou1 , R. Schmid1 , R. Jurgons1 , Ch. Bergemann2 ,

W. Arnold3 , and F.G. Parak4
1
Department of Otorhinolaryngology, Head and Neck Surgery, University
Erlangen-Nürnberg, 91054 Erlangen, Germany
2
Chemicell, 10823 Berlin, Germany
3
Department of Otorhinolaryngology, Head and Neck Surgery, Klinikum rechts der
Isar, Technical University of Munich, 81675 Munich, Germany
4
Physics-Department E 17, Technical University of Munich, 85748 Garching,
Germany
Abstract. The difference between success or failure of chemotherapy depends not

only on the drug itself but also on how it is delivered to its target. Biocompatible
ferrofluids (FF) are paramagnetic nanoparticles, that may be used as a delivery system
for anticancer agents in locoregional tumor therapy, called “magnetic drug targeting”.
Bound to medical drugs, such magnetic nanoparticles can be enriched in a desired body
compartment (tumor) using an external magnetic field, which is focused on the area
of the tumor. Through this form of target directed drug application, one attempts to
concentrate a pharmacological agent at its site of action in order to minimize unwanted
side effects in the organism and to increase its locoregional effectiveness.
Tumor bearing rabbits (VX2 squamous cell carcinoma) in the area of the hind limb,
were treated by a single intra-arterial injection (A. femoralis) of mitoxantrone bound
ferrofluids (FF-MTX), while focusing an external magnetic field (1.7 Tesla) onto the
tumor for 60 minutes. Complete tumor remissions could be achieved in these animals
in a dose related manner (20% and 50% of the systemic dose of mitoxantrone), without
any negative side effects, like e.g. leucocytopenia, alopecia or gastrointestinal disorders.
The strong and specific therapeutic efficacy in tumor treatment with mitoxantrone
bound ferrofluids may indicate that this system could be used as a delivery system
for anticancer agents, like radionuclids, cancer-specific antibodies, anti-angiogenetic
factors, genes etc.
1 Introduction
Ferrofluids have been used in medicine since the 1960es for e.g. the magneti-
cally controlled metallic thrombosis of intracranial aneurysms and magnetically
guided selective embolization of the renal artery in case of a renal tumor. Per-
manent obliteration have been expected by producing an artificial thrombus in
the target artery [1,2]. Iron catheters manipulated by an external magnetic field
are in use since 1951. In literature the successful diagnosis of a congenital heart
disease by the use of a magnetically guided catheter is described [3]. Another
system using the magnetic guidance is the magnetic implant guidance system for

234 Ch. Alexiou et al.
stereotactic neurosurgery. A small magnetic NdFeB capsule within the brain has
been moved from outside the body by six independently controlled coils. Cer-
tainly it could be used to deliver radioactivity, heat or chemotherapeutic drugs
to a tumor in the brain [4]. Magnets have been also used to extract swallowed
needles and nails or to remove iron objects from eyes or from other body parts
[5]. Ferrofluids are also used as a contrast agent for MRI. Superparamagnetic
iron oxide particles are a new class of MR contrast agents that have been shown
to significantly increase the detectibility of hepatic and splenic tumors. No acute
or subacute toxic effects were detected by histological or serologic studies in
rats and beagle dogs who received a total of 3000 μmol Fe/kg, 150 times the
dose proposed for MR imaging of the liver. These results indicate that ferroflu-
ids are fully biocompatible potential contrast agent for MRI [6]. These particles
are selectively taken up by cells of the mononuclear phagocytosing system (e.g.
Kupffer’s cells in the liver and macrophages in the spleen, lymph nodes, or bone
marrow), depending on the particle design (e.g. coating or size). The value of
these particles in the diagnostic evaluation of liver and spleen tumors has already
been shown in animal experiments and clinical studies [7,8]. Ultrasound, CT, and
MR imaging do not allow reliable differentiation between hyperplastic and tu-
morous lymph nodes or detection of small metastases in normal-sized lymph
nodes. Superparamagnetic iron oxide particles are also used for MR lymphog-
raphy for the detection of lymph node metastases. In tumor-bearing rabbits,
different degrees of metastatic displacement of lymph nodes were discernible,
and even small metastases (3 mm in diameter) could be visualized [9].
Magnetic properties of ferrofluids can be combined with targeting capability
of affinity ligands for specific cell targeting and separation. Magnetic cell sorting
is employed for both diagnostic and preparative applications in detection and
isolation of various cell subpopulations including damaged and diseased cells.
Attachment of monoclonal antibodies (immunomagnetic cell separation) and
other specific ligands to magnetic particles enable the separation of different
cell populations with relatively simple equipment, and the technology provides a
relatively rapid and convenient method for processing large quantities of different
cell subpopulations [10].
Ferrofluids are also an important subject in the development of an im-
plantable artificial heart. A ferrofluidic actuator directly drives magnetic fluids
simply by applying a magnetic field to the ferrofluids and does not require a
bearing. Magnetic fluid in a U-shaped glass cylinder was placed in an air gap of
a solenoid. When a magnetic flux of 0.32 T was applied to the interface of the
ferrofluid and air, the ferrofluid was displaced and a pressure was obtained [11].
Ferrofluids coated with starch polymers can be used as biocompatible carriers
in a new field of locoregional tumor therapy called “magnetic drug targeting”
[12,13,14]. It is of great importance to target selective the antineoplastic agent
to its tumor target as precisely as possible, to reduce the resulting systemic toxic
side effects from generalized systemic distribution and to be able to use a much
smaller dose, which could further lead to a reduction of toxicity.
Magnetic Drug Targeting 235
In the present study, we confirm the concentration of ferrofluids in VX2

squamous cell carcinoma tissue of the rabbit using histological investigations and
MR imaging. The biodistribution of these magnetic nanoparticles was studied by
the use of 123 iodine-labeled particles. The therapeutic efficacy of “magnetic drug
targeting” was studied using the rabbit VX2 squamous cell carcinoma model.
2 Materials and Methods
2.1 Ferrofluids
The ferrofluids used in the experiments were obtained from Chemicell (Berlin,
Germany; German patent application no. 19624426.9) and consisted of a biocom-
patible colloidal dispersion formed by wet chemical methods from iron oxides and
hydroxides to produce special multidomaine particles [15]. These ferrofluids are
covered with hydrophilic starch polymers coupled with endstanding functional
groups (e.g. phosphate) allowing ionic binding to many therapeutic drugs. The
hydrodynamic diameter of the whole particle is about 100 nm Fig. 1 [12].
Organic as well as inorganic agents can ionically bound to the functional
group, in this case phosphat. The chemotherapeutic agent mitoxantrone (Novan-
tronr , Lederle, Wolfratshausen, Germany) is ionically bound to the endstanding
groups of the starch coating (Fig. 2).
Mitoxantrone is a synthetic anthracendion that inhibits DNA and RNA syn-
thesis by intercalating in DNA molecules, which causes strand breaks. It is an
anticancer agent used in treatment of e.g. breast cancer, non-Hodgkin’s lym-
phoma and various solid tumors (carcinoma and sarcoma)[16,17,18].
Fig. 1. Imaging of the starch coated magnetic nanoparticles by scanning electronic

microscopy.
Fig. 2. Structural formula of mitoxantrone bound to magnetic nanoparticle [12].
Fig. 3. Desorption of MTX measured by UV-visible-spectroscopy at wavelength of

648 nm, depending on time [12].
The body surface area and the dose of mitoxantrone were calculated accord-
ing the recommendations in literature [19].
123
2.2 I-labelled nanoparticles
Starch and iodine soluted in water result in the development of iodine/starch-

complexes, which give an intensive blue colouring. Thereby the reduced iodine-
molecules become deposited into the pockets of the starch-molecules (e.g. ß-
amylase). Charge-shifting of electrons result in absorption of visible light, which
causes the intensive blue colouring of the complex.
Ferrofluid (MAG-D 14/11) was mixed with 123 iodine. For the calculation of
the activity, a definitive volume of the suspension was measured in the gamma-
counter.
The experiments were performed in tumor-bearing rabbits with a cathe-
ter in the Arteria femoralis under the gamma-camera. The ferrofluid-123 iodine-
suspension was injected into the tumor-supplying artery for 5 minutes, while
the permanent-magnet was focused on the tumor-area. At the same time the
signaldetection of 123 iodine started.
2.3 Magnetic Field

An electromagnet with a magnetic flux density of a maximum of 1.7 Tesla was
used to produce an inhomogenous magnetic field. The magnetic flux density was
focused onto the region of the tumor with a specially adapted pole shoe that was
placed in contact with the surface of the tumor. On the tip of the pole shoe, the
gradient has its maximum Fig. 4. The magnetic field was focused on the tumor
during the ferrofluid injection of 10 minutes and for 60 min in total. The pole
shoes of the magnet were specially modified in order to optimize flux density
and gradient of the magnetic field (e.g. at a distance of 10 mm to the tip of the
pole shoe the flux density was 1.0 Tesla).
A neodym-permanentmagnet with a magnetic flux density of 0,6 Tesla was
used for the investigations with 123 Jod-labelled ferrofluids [12].
2.4 Surgical Intervention

The chemotherapy experiments were performed when the tumors had reached a
volume of approximately 3500 mm3 .
Fig. 4. Dependence of the magnetic flux density on the distance to pole shoe with the
electromagnet [12].
For application of the chemotherapy, the femoral artery was cannulized and
a catheter was placed approximately 2 cm distal to the inguinal furrow. The
mitoxantrone bound ferrofluid was administered slowly over 10–15 minutes.
2.5 Experimental Protocols

The 26 animals were divided into six groups, depending on the type of treatment.
Group 1 received an i.a. infusion of FF-MTX with the magnetic field at a dose
equivalent to 20% and 50% of the systemic dose MTX (group 1a and group
1b, respectively). Group 2 received an i.a. infusion of MTX alone without the
magnetic field at doses equivalent to 20%, 50%, 75% and 100% of the systemic
dose. Group 3 received an i.a. infusion of FF alone with the magnetic field
at equivalent doses compared with groups 1a and 1b. Group 4 received an i.v.
infusion of FF-MTX with the magnetic field at doses equivalent to 20% and 50%
of the systemic dose (group 4a and group 4b, respectively). Group 5 received
an i.a. infusion of FF-MTX without the magnetic field at doses equivalent to
20% and 50% of the systemic dose, and group 6 was the control group without
treatment.
After treatment, the tumor was measured every 3rd day by the same observer
with a capiler ruler for a period of 3 months.
Blood samples were drawn by venipuncture every week. Immediately follow-
ing intra-arterial infusion of 50% FF-MTX into the femoral artery, and following
application of the magnetic field for a duration of 60 minutes, one animal was
sacrificed and the tumor was fixed and stained with haematoxylin/eosin and
Prussian Blue. Prussian blue, a specific stain to identify iron was used.
Following the 3 month observation period the remaining animals were sac-
rificed and the tumor, liver, kidneys, spleen, lungs, brain and inguinal lymph
nodes were removed and examined histologically.
Enrichment of intra-arterially injected ferrofluids was investigated using con-
ventional imaging techniques. In magnetic resonance imaging (MRI) the mag-
netic particles strongly reduce the transverse relaxation time (T2) and lower
significantly the longitudinal relaxation time (T1). Therefore, magnetic nanopar-
ticles are used as negative tissue-specific MRI contrast agents.
Six hours after ferrofluid application with an external magnetic field, a MRI
was performed on four tumor bearing animals.
2.6 Statistical Analysis

The tumor volume was calculated using the formula for an elliptical mass (1/6π
a2 b, a=width on the horizontal axis, b=length on the vertical axis). We consid-
ered change of volumes as percentages of tumor volumes (100%) found at day
0 (day of treatment). Statistical analysis for relative tumor volumes was per-
formed using the one sample t-test (with a conservatively fixed value of 100%
for the control group) and a t-test for two independent samples (Welch-test).
For blood parameters (absolute values) we applied the t-test for two indepen-
dent samples. The resulting two-sided p-values were considered significant at or
below a value of 0.05. Between 0.01–0.05 the result was considered significant
and highly significant if below 0.01.
3 Results
3.1 Tumor Volume
In the control group without treatment tumor volume increased and metastases
appeared. The animals of group 1a, treated intra-arterially with 20% FF-MTX,
had a complete tumor remission between the 15th and 36th day following treat-
ment. The animals of group 1b (50% FF-MTX) had a decrease in tumor volume
similar to group 1a. In group 2 (intra-arterial MTX alone, no magnetic field),
lower dosages (20% and 50% of the systemic dose) did not result in tumor re-
mission and enlarged, palpable inguinal lymph nodes were found after 48 days.
At higher doses (75% and 100%), complete remission of tumor occurred.
The two group 3 animals (intra-arterial FF alone with the magnetic field,
at 20% and 50%) demonstrated a progressive increase in tumor volume with
palpable, enlarged inguinal lymph nodes (metastases).
The six animals of group 4 (intravenous injection via the ear vein of 20%
and 50% FF-MTX with magnetic field) showed a slight tumor remission, but
the reduction of volume was not statistically significant in comparison to the
control group (not shown in figures).
The two animals of group 5 (intra-arterial FF-MTX 20% and 50%, without
a magnetic field) showed a discontinuation of tumor growth and no evidence
of metastases, but no remission of the tumor was seen (not shown in figures).
Fig. 5. Group 1a: Effect of i.a. application of FF-MTX [20% of the regular systemic
dose (•)] on relative tumor volume after magnetic drug targeting compared with control
group [group 6 (Δ), control(no treatment)]. Symbols, the median tumor volume; bars,
the maximum and minimum; metastases, onset of metastases; treatment, the day of
treatment (singular treatment)[12].
Fig. 6. Group 1b: Effect of i.a. application of ferrofluids bound to mitoxantrone (FF-
MTX, 50% of the regular systemic dose),(•) on relative tumor volume after magnetic
drug targeting compared with control group [group 6 (Δ), control (no treatment)].
Symbols, the median tumor volume; bars, the maximum and minimum values; metas-
tases, onset of metastases; treatment, the day of treatment (singular treatment) [12].
Fig. 7. Group 2: Effect of i.a. application of MTX - 20% and 50% (♦) and 75% and
100% (•) of the regular systemic dose - on relative tumor volume compared with control
group [group 6 (Δ), control (no treatment)]. Symbols, the median tumor volume; bars,
the maximum and minimum value; metastases, onset of metastases; alopecia, onset of
alopecia; treatment, the day of treatment (singular treatment) [12].
At the time of treatment less then 5% of the animals showed a small necrotic
fraction in the area of the tumor area [12].
3.2 Local and Systemic Effects
The general condition of the control group animals (limited to 2 animals for eth-
ical reasons) worsened during the observation period and the animals developed
Fig. 8. Group 3: Effect of i.a. infusion of FF (50%) alone with magnetic field (). Δ,
control (no treatment). Symbols, the median relative tumor volume; bars, the maximum
and minimum values; metastases, onset of metastases; treatment, the day of treatment
(singular treatment).
Fig. 9. Values of the WBC before treatment (day 0) and on days 3-15 (early period),
18-48 (middle period), and 51-81 (late period)after the respective treatment regimes.
Columns, the median values; bars, the maximum and minimum values. The description
of the bars, i.e. the respective group is mentioned beside the graphs [12].
Fig. 10. Same as Fig. 12 for other groups mentioned beside the graphs.
pneumonia, which explains the increase of leucocytes as seen in Figs. 9 and 10

similar to description in the literature [20].
The animals of group 1 had no evident side effects such as alopecia, ulcers or
muscular atrophy and their general condition (weight, food intake, excrement,
urine, activity, fur condition) remained normal during the whole 3 month obser-
vation period compared to the physiological data of healthy animals (statement
by Charles River, Sulzfeld, Germany). No significant changes in serum iron or
leucocyte values were seen in this group (Fig. 9).
The urine of one animal in group 2 (50% MTX) showed blue-green discolo-
ration and this animal developed mild alopecia in the region of the digits after
48 days. Both animals with low dose MTX (20% and 50%) had a decrease in
leucocyte values, but this was not statistically significant (p=0.29). Both group 2
animals with high dose (75% and 100%) MTX had temporary blue-green urine
discoloration, as well as a unilateral alopecia (palmar region of the digits to
the knee joint) of the limb in which the tumor was implanted developing after
33 days. This hairless area developed cutaneous inflammation and ulceration,
followed by mild alopecia of the fore limbs and head. The musculature of the
treated limb became atrophic and the circumference was noticeably smaller (3
cm) at the end of the 3 month observation period. There was no marked differ-
ence in the severity of the side effects between the two animals, except for the
fact that the animal with the higher MTX dose (100%) developed the changes
several days sooner. Group 2 animals with 50%, 75% and 100% MTX steadily
lost weight after an initial lag-phase and were underweight at the end of the
observation period (mean value 1800 mg below the lower reference values ac-
cording to the breeder’s statement, Charles River, Sulzfeld, Germany). These
animals became leucocytopenic (≤ 2.95 x103 /μl) in the early phase (highly sig-
nificant drop, p=0.004, Fig. 10), but recovered slightly in the middle and late
periods [12].
None of the animals of groups 3 or 4 showed any significant changes in serum
iron (not shown in figures) or leucocyte counts (group 3: Fig. 9; group 4a and b:
Fig. 9) during the observation period when compared with initial values.
3.3 Histological Findings
Figure 11 up to 14 show cross sections of a VX2 tumor that was excised just after
treatment with “magnetic drug targeting”. Ferrofluids can be seen as blue pig-
ment (Prussian Blue, Fig. 11, 12, 13) or as brown-black pigment (H.E., Fig. 14).
Immediately after magnetic drug targeting, ferrofluids are evident in the vessels
supplying the tumor and in macrophages. After the 3 month observation period,
no viable tumor tissue was histologically evident in the animal group 1, with only
fibrosis seen in the tumor implantation site. No metastases were found in the
regional lymph nodes or in any other organs. Some FF particles were found in
the spleen of the animals, but none were evident in the liver, lungs, brain or the
implantation site and surrounding musculature and skin. No other macroscopic
or histopathological changes were found in any of the investigated organs [12].
In group 2, the VX2 tumors of the 2 low dose animals were 8.644 mm3
(50% MTX) and 2.497 mm3 (20% MTX) in size with a large area of central
necrosis and viable tumor at the periphery. The 2 animals with high dose MTX
(75% and 100%) had no viable tumor at the implantation site. None of the other
Fig. 11. Cross section of VX2 squamous cell carcinoma of the rabbit immediately after
the magnetic drug targeting stained with Prussian Blue. Ferrofluids are visible in the
vessels of the peripheral area of the tumor (original magnification x 100).
the magnetic drug targeting stained with Prussian Blue. Ferrofluids are visible in tumor
supplying vessels in the connective tissue of the VX2 tumor (original magnification x
200).
investigated organs in the animals of group 2 (liver, kidneys, spleen, lungs, brain)
had any pathological changes.
The tumors of both animals of group 3 measured 13.324 mm3 and 17.649
mm3 , with a large area of central necrosis and viable tumor at the periphery. No
FF particles were found within the tumor or in the surrounding musculature and
skin. Some FF were found in the spleen. Metastases were found in the inguinal
lymph nodes and liver of both animals. None of the other investigated organs
(kidneys, spleen, lungs, brain) had any pathological changes [12].
3.4 MRI Imaging

The distribution and the enrichment of the intra-arterially injected ferrofluids
were investigated in vivo by MRI after “magnetic drug targeting”. In MRI the
magnetic particles strongly reduce the transverse relaxation time (T2) and lower
significantly the longitudinal relaxation time (T1). The MRI was made 6 h after
treatment with “magnetic drug targeting” and the concentration of ferrofluids
was seen by extinction of signal in the area of the tumor.
magnetic drug targeting stained with Prussian Blue. Proof of iron in macrophages and
endothelium (original magnification x 1000).
Fig. 14. Cross section of VX2 squamous cell carcinoma of the rabbit immediately
after magnetic drug targeting stained with haematoxilin-eosin. Ferrofluids are visible
intravascular and in a macrophage (Original magnification x 1000).
Fig. 15. (left) MRI of tumorous hind limb after i.a. application of ferrofluids; dotted
circle: tumor region; f: head of the femur.
(right) MRI of tumorous hind limb after i.v. application of ferrofluids; dotted circle:
tumor region; f: head of the femur [12].
Fig. 15 shows the left hind limb (implantation site) of two rabbits receiving
intra-arterial or intravenous 50% FF-MTX, respectively, 6 hours after treatment.
The tumor is situated at the medial portion of the hind limb (dotted circle), and
the concentration of FF is seen by extinction of signal. Fig. 15(left) (intra-arterial
FF-MTX) shows definite extinction of signal and Fig. 15(right) (intravenous FF-
MTX) only a very discrete signal extinction. The area marked “f ” is at the head
of the femur and appears to be hypodense.
The tumor is situated at the medial portion of the hind limb (dotted cir-
cle), and the concentration of FF is seen by extinction of signal. Fig. 15(left)
(intra-arterial FF-MTX) shows definite extinction of signal and Fig. 15(right)
(intravenous FF-MTX) only a very discrete signal extinction. The area marked
“f” is at the head of the femur and appears to be hypodense [12].
123
3.5 Biodistribution of Iodine-Ferrofluids
In order to assess the distribution of the used nanoparticles under the influence
of the magnetic field, 123 iodine was storaged in the starch-cover and was then
injected into the tumor-supplying artery while the magnetic field was focused
on the tumor for 60 minutes.
At the same time, the signal-detection of 123 iodine started. Around the
tumor-area there was a “region of interest” (ROI) determined and the activ-
ity was measured over the time.
The injected whole-body-activity of the rabbit could not be measured with
the gamma-camera, so the activity in the tumor-area was measured at differ-
ent times and was related to the initial activity-maximum (relative activity in
percent of the initial activity).
The scintigraphically detected signal after intra-arterial application has been
shown to be at least twice as much higher in the magnetically focused region
compared to the application without external magnetic field.
Enrichment of 123 I-labelled nanoparticles in the area of interest (VX2 tumor)
after magnetic drug targeting:
The activity in the tumor region scintigraphically measured by gamma-
camera and focused by the external magnetic field was after 30 minutes 38.2
%, after 40 minutes 31.8 %. and after 60 minutes 22.3 % (n=1). Without ex-
ternal magnetic field, the activity was 14.7 % after 30 minutes, 13.0 % after 40
minutes and 11.8 % after 60 minutes (n=1).
4 Discussion
Chemotherapy is a balancing act between efficacy and toxicity. Because of the

relatively non-specific action of chemotherapeutic agents, there is almost always
some toxicity to normal tissue. Therefore, it is of great importance to be able
to selectively target the antineoplastic agent to its tumor target as precisely
as possible, to reduce the resulting systemic toxic side effects from generalized
Fig. 16. Schematic drawing of a tumor-bearing rabbit under the γ-camera.

Enrichment of 123 I-labelled nanoparticles in the area of interest (VX2 tumor) after
“magnetic drug targeting”. The image was taken 40 min after application, still showing
stable concentration of ferrofluids within the tumor tissue [14].
systemic distribution and to be able to use a much smaller dose, which would
further lead to a reduction of toxicity.
Regional chemotherapy via a regional artery administers a more concentrated
dose of the active agent directly into the tumor [20]. The advantage of this
approach is limited, however, by drain-off via the venous blood which limits
exposure time and reduces the overall efficacy.
At present, i.a. delivery of chemotherapeutic agents is approved and well
accepted for treatment of liver metastases [21] and has occasionally been used
for other tumor types also (e.g., inoperable head and neck tumors); but it has
often necessitated complicated, time-consuming operative procedures, including
general anaesthesia [22].
“Magnetic drug targeting” means holding the chemotherapeutic agent due to
the carrier (ferromagnetic particles) at the desired site of activity, thus increasing
efficacy and diminishing systemic toxicity.
The goal of the present study was to concentrate ferrofluids coupled with
chemotherapeutic agent in a desired target area using a magnetic field. The
principle of “magnetic drug targeting” consists of two steps: The first step is
the delivery of the mitoxantrone bound magnetic nanoparticles to the desired
body compartment (tumor), the second is the release of the drug from its carrier
[12,13,14]. An additional helpful factor is that microvascular permeability in
neoplastic tissues is increased (8-fold compared with normal tissue) as is diffusion
(33-fold) [23], which allows chemotherapeutic agents much easier to penetrate
into tumor tissue. The metabolism of ferrofluid particles takes place in the liver
and spleen in analogy to iron metabolism.
In this investigations, the authors found that this approach led to complete
tumor remission with reduced doses of 20% and 50% MTX bound to FF (Figs. 5
and 6). The application was well tolerated by the animals and no signs of toxicity
were detected. On the contrary, intra-arterial infusion of the same doses, 20% and
50%, of MTX alone (group 2, Fig. 7) resulted in no reduction of tumor volume
and the animals developed metastases and suffered from chemotherapeutic side
effects. Only when the dose of MTX alone was increased to 75% and 100%,
a tumor remission was seen (Fig. 7), but this resulted in severe side effects
(alopecia, ulcers, leucocytopenia), as seen in Fig. 10. Intravenous infusion of the
FF-MTX complex was also ineffective since only a slight tumor remission that
was not statistically significant resulted. The same result was observed after
intra-arterially infused FF-MTX without an external magnetic field, since the
tumor remained at the same size, without remission. Thus, the combination of
intra-arterial infusion with a magnetic field was safe, effective, well tolerated by
the animals and was very effective in treating the tumor although the dose of
chemotherapeutic agent was markedly reduced [12,13,14].
At present, intra-arterial delivery of chemotherapeutic agents is approved
and well accepted for treatment of liver metastases [22], and have occasionally
also been used for other tumor types (e.g. inoperable head and neck tumors), but
often necessitate complicated, time consuming operative procedures, including
general anesthesia [22]. Experimentally, Swistel et al. [24] described encouraging
results using intra-arterial chemotherapy for VX2 squamous cell carcinoma. They
achieved complete tumor remission after intra-arterial application of adriamycin
in 4 of 6 animals, whereas intravenous infusion of adriamycin caused severe
toxicity and resulted in complete remission in only two cases.
A potential complication that could arise with the use of FF compounds is
the fact that with larger particles embolization could occur, preventing a suffi-
cient concentration of the chemotherapeutic agent from reaching the tumor. On
the other hand, if the particles are too small, the external magnetic field might
not provide sufficient attraction so that the particles are drawn into the tumor.
The particles used in this study had a size of 100 nm (nanometer). No emboliza-
tion was seen in the vascular system of the tumor. Our histological findings
showing distribution of FF particles throughout the tumor strongly support the
concept that large molecular weight substances such as chemotherapeutic agents
or monoclonal antibodies can be effectively targeted to tumor tissue. In addi-
tion, the fact that the FF alone with a magnetic field failed to cause tumor
remission (Fig. 8) indicates that therapeutic effect resulted from the action of
the chemotherapeutic agent itself, rather than intratumoral embolization by the
particles [12].
The electromagnet used for this study produced a magnetic flux density
of a maximum of 1.7 Tesla which decreased depending on the distance to the
pole shoe (Fig. 4). The magnetic gradient can be seen as a collection of vec-
tors which point in the direction of increasing values as shown in Fig. 3 (yellow
arrows). The arrow sizes correspond to the strength of the magnetic gradient.
Both factors (direction and magnitude) reflect the inhomogenous character of
the magnetic field, which is of key importance for “magnetic drug targeting”. In
previous studies it was suggested that a magnetic field strength of 8000 Gauss
(0.8 Tesla) is sufficient to exceed linear blood flow in the intratumoral vasculature
and allow 100% localization of magnetic carrier containing 20% magnetite [25].
In contrast, Goodwin and coworkers applied magnetic-targeted carriers (MTCs)
intra-arterially in a swine model, focusing a magnetic field of only 250 to 1000
Gauss (0.025-0.1 Tesla, permanent neodymium magnet) to the desired compart-
ments in the liver and lungs. The depth of this MTC targeting was 8 - 12 cm
and the particle size 0.5 μm - 5 μm. With this model, MTCs with a predefined
activity had a concentration of 67% in the liver and 50% in the lung localized
by the magnet [26].
The magnetic field strength with a maximum of 1.7 Tesla used in the present
investigation was the strongest ever applied for magnetic drug targeting. We
achieved a high concentration of FF within the tumor after intra-arterial infu-
sion of FF which was seen by histological (Figs. 11 to 14) and MRI methods
(Fig. 15). The VX-2 squamous cell carcinoma was superficially exposed and had
no migratory motion, as was the case with the liver and lung targets (breathing
fluctuations) in the swine model of Goodwin and coworkers [26]. In addition
these organs lie deeply in the body cavity (8-12 cm from the body surface),
greatly complicating focusing a maximum of the magnetic flux density onto the
tumor area. Two approaches to overcome this problem are possible,
1. the use of larger particles, as previously suggested by Bergemann et al. [27] or

2. using a stronger magnetic field.
The particles (FF-MTX) used in the present study were 100 nm in size and
have shown good therapeutic results in smaller animals (mouse, rat) as well
[28,29].
The strong magnetic field was very efficacious in combination with these
particles, but further experiments (which we have already begun) should be
performed using marked FF to clarify the optimal magnetic field strength and
particle size [12].
A remarkable feature of using ionically bound pharmaceuticals is that the
low molecular weight substances (molecular weight of mitoxantrone 517) is able
to desorb from the carrier (FF) after a defined time span and can then pass
through the vascular wall or interstitium into the tumor cells. This is important
since once the FF-MTX complex has been directed to the tumor by the magnetic
field, the drug must dissociate in order to act freely within the tumor. As shown
in Fig. 2 MTX desorbs from the FF after 30 min (half life) and so 50% of the
drug is free to act on the tumor after 30 min.
Dextran coated iron oxides have been shown to produce signal loss by MRI
and have been used as a contrast medium for the detection of metastatic lymph
nodes (negative contrast) [30]. We found total signal loss and therefore a very
high concentration of FF by MRI after focusing by the magnetic field (Fig. 15
(left)) [12]. Recent studies have shown that intra-arterial application of radio-
active-labeled, magnetic carriers with an external magnetic field resulted in re-
tention of at least 50% in the target site [31]. In comparison, after intravenous
injection, only very slight signal loss was seen, indicating a very low concen-
tration (Fig. 15(right)). This underscores the advantage of intra-arterial versus
intravenous infusion in magnetic drug targeting.
Biodistribution was studied by the use of 123 iodine-nanoparticles. The scinti-
graphically detected 123 iodine- signal after intra-arterial (artery leading the tu-
mor, femoral artery) application has been shown to be significantly higher in
the magnetically focused region compared to the application without external
magnetic field (Fig. 16). 60 minutes after application there is still twice as much
concentration of 123 iodine in the magnetically focused region compared to the
application without the magnetic field [14].
Previous studies by Bacon et al. concerning FF with a particle size of 0.5-1.0
μm found no acute or chronic toxicity after the intravenous infusion of 250 mg
iron/kg body weight in rats [32] and 1-3 mg iron/kg body weight in humans
has been shown to be safe as well [33]. This agrees with our findings, since FF
infusion was not associated with any signs of toxicity [12,13,14].
Magnetic microspheres loaded with the gamma-emitting radioisotope 90Y
have also successfully been used as a form of radionuclide therapy. In one study,
this compound was maneuvered within the body of a mouse to a subcutaneous
lymphoma, resulting in eradication of the tumor [33]. Magnetic fluids have also
been used for the so-called “magnetic fluid hyperthermia” that has been used to
control the local growth of murine mammary carcinoma [35]. Additional modifi-
cation of the magnetic particles, so that they could bind monoclonal antibodies,
lectins, peptides or hormones could make delivery of these compounds more
efficient and also highly specific. Targeted drug delivery using nanotechnology
opens a new field in oncology science and improves basis science and clinical
practice [36].
Acknowledgements
Supported by the Margarete Ammon Foundation, Munich, and grants from the
Technical University of Munich, Germany.
References
1. Alksne JF, Fingerhut A, Rand R. Magnetically controlled metallic thrombosis of
intracranial aneurysms. Surgery 1966; 60: 212-218.
2. Hilal SK, Michelsen WJ, Driller J, Leonard E. Magnetically guided devices for
vascular exploration and treatment. Radiology 1974;113: 529-534.
3. Ram, W. and Meyer, H. Heart catheterization in a neonate by interacting mag-

netic fields. A new and simple method of catheterguidance. Catheterization and
Cardiovascular Diagnosis, 22: 317-319, 1991.
4. McNeil, R. G., Ritter, R. C., Wang, B. et al. Functional design features and initial
performance characteristics of a magnetic-implant guidance system for stereotactic
neurosurgery. IEEE Transactions on Biomedical Engineering, 42: 793-801, 1995.
5. Luborsky, F.E. Recent advances in the removal of magnetic foreign bodies from
the esophagus, stomach and duodenum with controllable permanent magnets. Am.
J. Roentg. Rad. Ther. Nucl. Med., 1964.
6. Weissleder R, Hahn PF, Stark DD et al. MR imaging of splenic metastases: ferrite-
enhanced detection in rats. AJR 1987; 149: 723-726.
7. Weissleder R, Stark DD, Engelstad BL et al. Superparamagnetic iron oxide: phar-
macokinetics and toxicity. AJR 1989;152: 167-173.
8. Weissleder R, Elizondo G, Wittenberg J, Lee AS, Josephson L, Brady TJ. Ultra-
small superparamagnetic iron oxide. An intravenous contrast agent for assessing
lymph nodes with MR imaging. Radiology 1990;175: 494.
9. Taupitz M, Wagner S, Hamm B, Dienemann D, Lawaczeck R, Wolf KJ. MR lym-
phography using iron oxide particles. Detection of lymph node metastases in the
VX2 rabbit tumor model. Acta Radiol. 1993; 34: 10-15.
10. Hardingham JE, Kotasek D, Sage RE, Eaton MC, Pascoe VH. Detection of circu-
lating tumor cells in colorectal cancer by immunobead-PCR is a sensitive prognos-
tic marker for relapse of disease. Molec. Med. 195;1: 789-794.
11. Mitamura Y, Wada T, Keisuke S. A ferrofluidic actuator for an implantable arti-
ficial heart. Artif. Organs 1992; 16 (5): 490-495.
12. Alexiou, C., Arnold, W., Klein, R.J. et al. Locoregional cancer treatment with
Magnetic Drug Targeting. Cancer Res 2000; 60: 6641-6648.
13. Alexiou, C., Arnold, W., Hulin, P. et al. Magnetic mitoxantrone nanoparticle de-
tection by histology, X-ray and MRI after magnetic tumor targeting. JMMM 2001;
255: 187-193.
14. Alexiou, C., Schmidt, A., Klein, R.J. et al. Magnetic drug targeting: Biodistribution
and dependency on magnetic field strength. JMMM in press.
15. Lübbe AS, Bergemann C, Huhnt W et al. Preclinical experiences with magnetic
drug targeting: tolerance and efficacy. Cancer Res 1996; 56: 4694-4701.
16. Ho, A. D., Del Valle, F., Haas, R. et al. Sequential studies on the role of mitox-
antrone, high-dose cytarabine, and recombinant human granulocyte-macrophage
colony-stimulating factor in the treatment of refractory non-Hodgkin’s lymphoma.
Semin. Oncol., 17: 14-18, 1990.
17. Hiddemann, W., Buchner, T., Heil, G. et al. Treatment of refractory acute lym-
phoblastic leukemia in adults with high dose cytosine arabinoside and mitoxantrone
(HAM). Leukemia., 4:637-640, 1990.
18. Freund, M., Wunsch-Zeddies, S., Schafers, M. et al. Prednimustine and mitox-
antrone (PmM) in patients with low-grade malignant non-Hodgkin’s lymphoma
(NHL), chronic lymphocytic leukemia (CLL), and prolymphocytic leukemia (PLL).
Ann. Hematol., 64: 83-87, 1992.
19. Bistner SI, Ford RB, Raffe MR (eds.). Kirk and Bistner’s Handbook of Veterinarian
Procedures and Emergency Treatment, Ed 6, p. 907. Philadelphia: W.B. Saunders
Co., 1995.
20. Stephens, F. O. Why use regional chemotherapy? Principles and pharmacokinetics.
Reg Cancer Treat., 1: 4-10, 1988.
21. Link, K.H., Kornmann, M., Formenti, A. et al. Regional chemotherapy of non-
resectable liver metastases from colorectal cancer – literature and institutional
review. Langenbecks Arch. Surg., 384: 344-353, 1999.
22. v. Scheel, J.: Invasive procedures for antineoplastic chemotherapy. In: Naumann
et al. (eds.): Head and Neck Surgery. Stuttgart, New York: Thieme, 1998.
23. Gerlowski, L.E. and Jain, R.K. Microvascular permeability of normal and neoplas-
tic tissues. Microvascular Research 31:288-305, 1986.
24. Swistel AJ, Bading JR and Raaf JH Intraarterial versus intravenous Adriamycin
in the VX2 tumor system. Cancer (Phila.) 53: 1397-1404, 1984
25. Senyei, A., Widder, K., and Czerlinski, C. Magnetic guidance of drug carrying
microspheres. J. Appl. Physiol., 49: 3578-3583, 1978.
26. Goodwin, S., Peterson, C., Hoh, C., Bittner, C. Targeting and retention of magnetic
targeted carriers. J. Magn. Magn. Mater., 194 : 132-139, 1999.
27. Bergemann, C., Müller-Schulte, D., Oster, J., à Brassard, L., Lübbe, A.S. Mag-
netic ionexchange nano- and microparticles for medical, biomedical and molecular
biological applications. J. Magn. Magn. Mater.,194 : 45-52, 1999.
28. Lübbe, A. S., Bergemann, Ch., Huhnt, W., Fricke, T., Riess, H., Brock, J. W.,
and Huhn, D. Preclinical experiences with magnetic drug targeting: tolerance and
efficacy. Cancer Research, 56: 4694-4701, 1996.
29. Lübbe, A. S., Bergemann, Ch., Riess, H. et al. Clinical experiences with magnetic
drug targeting: a phase I study with 4’-epidoxorubicin in 14 patients with advanced
solid tumors. Cancer research, 56: 4686-4693, 1996.
30. Taupitz, M., Wagner, S., Hamm, B., Dienemann, D., Lawaczeck, R. and Wolf, K.J.
MR Lymphography using iron oxide particles. Detection of lymph node metastases
in the VX2 rabbit tumor model. Acta Radiologica, 34:10-15, 1993.
31. Widder, K.J., Senyei, A.E. and Scarpelli, D.G. Magnetic microspheres: a model
system for site specific drug deleivery in vivo. Proc. Soc. Exp. Biol. Med. 58: 141-
146, 1978.
32. Bacon, B.R., Park, D. D., Saini, S., Groman, E. V., Hahn, P. F., Compton, C. C.,
and Ferrucci, J. T. Ferrite particles: A new magnetic resonance imaging contrast
agent. Lack of acute or chronic hepatoxicity after intravenous administration. J.
Lab. Clin. Med., 110: 164-171, 1987.
33. Rummeny, E., Weissleder, R., Stark, D. D., Elizondo, G. and Ferrucci, J. T. Mag-
netic resonance tomography of focal liver and spleen lesions. Experiences using
ferrite, a new RES-specific MR contrast medium. Radiologe 28: 380-386, 1988.
34. Häfeli, U.O., Pauer, G.J., Roberts, W.K., Humm, J.L. and Macklis, R.M. Mag-
netically targeted microspheres for intracavitary and intraspinal Y-90 radiotharpy.
In:U. Häfeli and W. Schütt (eds.), Scientific and clinical applications of magnetic
carriers, pp 501-516, New York and London: Plenum press, 1997.
35. Jordan, A., Scholz, R., Wust, P. et al. Effects of magnetic fluid hyperthermia
(MFH) on C3H mammary carcinoma in vivo. Int. J. Hyperthermia, 13: 587-605,
1997.
36. Partridge, M., Phillips, E., Francis, R., Li, S.R. Immunomagnetic seperation for
enrichment and sensitive detection of disseminated tumour cells in patients with
head and neck SCC. J. Pathol., 189:368-377, 1999.
Magnetic Unit Systems
Historically numerous different magnetic unit systems are used. In this volume of
Lecture Notes in Physics mainly the SI-system (chapters by Rosensweig, Blums,
Zubarev, Thurm & Odenbach and Bossis) and the Gaussian system (chapters
by Shliomis and Morozov & Shliomis) are employed.
To reduce the problems with the comparison of statements in the various
chapters, the major units and their conversion are given in the table below.
Quantity Symbol Gaussian Conversion SI unit

unit factor
magnetic induction B G 10−4 T
magnetic field strength H Oe 103 /4π A/m
magnetization M emu/cm3 103 A/m
4πM G 103 /4π A/m
mass magnetization M emu/g 1 Am2 /kg
magnetic moment m emu 10−3 Am2
volume susceptibility χ dim.less 4π dim.less
permeability μ dim.less 4π10−7 H/m
An exception from these unit systems is made in the chapter by Müller

and Liu where the Heaviside-Lorentz units, or the so-called rational units, are
employed. In this system all four fields have the same dimension, square root
of the energy density, ie J/m3 in SI units and erg/cm3 in Gaussian units,
sensibly with H = 3DB and D = 3DE in vacuum. As a result, the factors of 4π,
√
0 and μ0 vanish. To revert to SI units (Ê, Ĥ...), employ Ĥ = 3DH/ μo , B̂ =
√ √ √ √ √
3DB μo , Ê = 3DE/ o , D̂ = 3DD o , and ê = 3De o , ĵe = 3Dje o .
To revert to the Gaussian system √ (Ẽ, H̃...), increase
√ all four fields, and decrease
all four sources,
√ by the factor of 4π: B = 3D B̃/ 4π, (similarly for D, H, E);
e = 3D 4π ˜e , (similarly for j, P, M )

Ferrofluids. Magnetically Controllable Fluids and Their Applications S. Odenbach

Uploaded by

Copyright:

Available Formats

Ferrofluids. Magnetically Controllable Fluids and Their Applications S. Odenbach

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ferrofluids. Magnetically Controllable Fluids and Their Applications S. Odenbach

Uploaded by

Copyright:

Available Formats

Lecture Notes in Physics

Online Version/ LNP Homepage

Cover picture: by Dr. S. Odenbach

Library of Congress Cataloging-in-Publication Data applied for.

Die Deutsche Bibliothek - CIP-Einheitsaufnahme

Ferrofluids : magnetically controllable fluids and their applications / S.

ical description and experimental investigations documents the complexity of

Bremen, May 2002 Stefan Odenbach

Part I Synthesis and Characterization

The Preparation of Magnetic Fluids

Magnetic and Crystalline Nanostructures in Ferroﬂuids

Part II Basic Theory

Basic Equations for Magnetic Fluids with Internal Rotations

Heat and Mass Transfer Phenomena

Part III Rheological Properties

Statistical Physics of Non-dilute Ferroﬂuids

Magnetic Fluid as an Assembly of Flexible Chains

Magnetoviscous Eﬀects in Ferroﬂuids

Magnetorheology: Fluids, Structures and Rheology

Targeted Tumor Therapy with “Magnetic Drug Targeting”:

Magnetic Unit System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255

Christoph Alexiou Stuart W. Charles

Riga Fritz G. Parak

Stefan Odenbach Steﬀen Thurm

Olga Volkova D-14109 Berlin

Department of Chemistry, University of Wales, Bangor, Gwynedd LL57 2UW, UK

Abstract. In any discussion of ﬂuids, which have magnetic properties, it is convenient

Stefan Odenbach (Ed.): LNP 594, pp. 3–18, 2002.

a repulsive interaction can be incorporated. In order to achieve this repulsive

2 Preparation of Nano-sized Magnetic Particles

In most technological and biomedical applications of ferroﬂuids, the magnetic

Wet-grinding. The original method of producing ferroﬂuids based on ferrites

Co-precipitation method. The co-precipitation method [18] for the prepara-

Substituted ferrite particles. To produce substituted nano-sized ferrite par-

Microemulsion techniques. In addition to simple straightforward precipi-

2.2 Metal Particles

Decomposition of organo-metallic compounds. There are a number of

Inverse micro-emulsion techniques. The stimulus for much of this work

by Lopez-Quintela et al. [38] that it is possible to subsequently coat the metal

Reduction of metal salts in aqueous solution. Particles produced by the

Miscellaneous methods. Fine particles of most of the metallic elements can

2.3 Particle-Size Distribution

Fig. 1. (left) Magnetite particles (right) Cobalt particles

Particle-size distribution is best monitored by using transmission electron

3.1 Fluids Containing Ferrite Particles

Surfactant stabilized colloids. Various methods of coating particles prior to

Fig. 2. Typical process for preparation of surfactant stabilized ferroﬂuids.

and the agglomerated hydrophobic precipitate of coated particles washed several

Ionically-stabilized colloids. For the production of ionically-stabilized col-

3.2 Fluids Containing Metal Particles

3.3 Assessment of Colloidal Stability

In many applications of ferroﬂuids, such as exclusion seals, loudspeakers, the

All magnetic ﬂuids, of whatever nature, exhibit magnetorheological eﬀects to a

moments resulting in strong magnetic interaction leading to chaining. The ef-

etc, by mechanical or ultrasonic means. The relative amounts of the ingredients

Department of Electronic and Electrical Engineering, University of Dublin, Trinity

Abstract. Accurate data on the frequency-dependent, complex susceptibility, χ(ω) =

Stefan Odenbach (Ed.): LNP 594, pp. 19–32, 2002.

is characterised by a time τN , which is referred to as the Néel relaxation time