A Brownian dynamics study on ferrofluid colloidal dispersions using an

iterative constraint method to satisfy Maxwell’s equations

Sean Hyun Dubina and Lewis Edward Wedgewood

Citation: Physics of Fluids 28, 072001 (2016); doi: 10.1063/1.4955014

View online: http://dx.doi.org/10.1063/1.4955014
View Table of Contents: http://aip.scitation.org/toc/phf/28/7
Published by the American Institute of Physics
 PHYSICS OF FLUIDS 28, 072001 (2016)

A Brownian dynamics study on ferrofluid colloidal

dispersions using an iterative constraint method
to satisfy Maxwell’s equations
Sean Hyun Dubinaa) and Lewis Edward Wedgewoodb)
Department of Chemical Engineering, University of Illinois at Chicago, 810 S. Clinton St.
(MC 110), Chicago, Illinois 60607-4408, USA
(Received 8 March 2016; accepted 20 June 2016; published online 5 July 2016)

Ferrofluids are often favored for their ability to be remotely positioned via external
magnetic fields. The behavior of particles in ferromagnetic clusters under uniformly
applied magnetic fields has been computationally simulated using the Brownian
dynamics, Stokesian dynamics, and Monte Carlo methods. However, few methods
have been established that effectively handle the basic principles of magnetic mate-
rials, namely, Maxwell’s equations. An iterative constraint method was developed to
satisfy Maxwell’s equations when a uniform magnetic field is imposed on ferrofluids
in a heterogeneous Brownian dynamics simulation that examines the impact of
ferromagnetic clusters in a mesoscale particle collection. This was accomplished by
allowing a particulate system in a simple shear flow to advance by a time step under
a uniformly applied magnetic field, then adjusting the ferroparticles via an iterative
constraint method applied over sub-volume length scales until Maxwell’s equations
were satisfied. The resultant ferrofluid model with constraints demonstrates that the
magnetoviscosity contribution is not as substantial when compared to homogeneous
simulations that assume the material’s magnetism is a direct response to the external
magnetic field. This was detected across varying intensities of particle-particle
interaction, Brownian motion, and shear flow. Ferroparticle aggregation was still
extensively present but less so than typically observed. Published by AIP Publishing.
[http://dx.doi.org/10.1063/1.4955014]

SYMBOLS
a Radius of particle including the steric layer
B Induction field
d Diameter of particle including the steric layer
ds Diameter of particle excluding the steric layer
d′ Total number of holonomic constraints
′
dG Number of holonomic constraints for Gauss’s law
d ′A Number of holonomic constraints for Ampère’s law
C
D Coupled diffusivity tensor
D0r Rotational diffusion coefficient at infinite dilution
DR Rotational diffusivity tensor
D0t Translational diffusion coefficient at infinite dilution
DT Translational diffusivity tensor
E Rate-of-strain tensor
F Force exerted on the fluid by a particle

a) Electronic mail: sdubin2@uic.edu. Telephone: (443) 995-1291. Fax: (312) 996-0808.

b) Electronic mail: wedge@uic.edu. Telephone: (312) 996-5228. Fax: (312) 996-0808.

1070-6631/2016/28(7)/072001/22/$30.00 28, 072001-1 Published by AIP Publishing.

072001-2 S. H. Dubina and L. E. Wedgewood Phys. Fluids 28, 072001 (2016)

FCON
α Center of mass constraint force of particle α
FPα Force of particle α acting on the surrounding fluid
g̃ Third-rank tensor of the mobility tensor
g̃ jl Metrix matrix of dimension d ′ × d ′
G̃ jl Modified metric matrix of dimension d ′ × d ′
h Cutoff radius of particle contribution to the local field M
h̃ Third-rank tensor of the mobility tensor
H Magnetic field
Hk Magnetic field at node k
H0 Externally applied magnetic field
Ĥ Demagnetizing field
kT Boltzmann’s constant and temperature product
k, k ′ Constraint node indices
l Parametrization of curve C
L Length of a side of the cubic simulation volume
Lc Length of a side of the local cubic constraint volumes
mα Magnetic dipole moment of particle α
M Mobility matrix
M Magnetization
Mk Magnetization at node k
ns Number of surfactant molecules per unit area on the particle surface
nα Dipole moment vector of particle α
n̂ Surface normal vector
N Total number of particles in the system
Pe Péclet number
qα Generalized coordinate of particle α
qUNα Unconstrained generalized coordinate vector of particle α
qCON
α Constrained generalized coordinate vector of particle α
r Position vector
rα Position vector of particle α
∆rBα Brownian displacement of particle α
rUN
α Unconstrained center of mass position vector of particle α
rCON
α Constrained center of mass position vector of particle α
Rm Ratio of magnetic particle-particle force to hydrodynamic shear force
Rv Ratio of steric repulsion force to hydrodynamic shear torque
Rh Ratio of the magnetic field torque to hydrodynamic torque
S Stresslet tensor
Sk Surface area around node k
∆t Time-step
T Torque exerted on the fluid by a particle
TαCON Orientational constraint torque of particle α
TαP Torque of particle α acting on the surrounding fluid
U Velocity of the fluid in the absence of particles
v Velocity vector
V Volume of magnetic source
W Radial weighting function
γj Lagrange multiplier of constraint σ j
γ̇ Shear rate of the ambient fluid
δ Thickness of the steric layer
δi j Kronecker delta with directional indices i and j
∆∞ Symmetric part of the rate-of-strain tensor
η Viscosity of the fluid
072001-3 S. H. Dubina and L. E. Wedgewood Phys. Fluids 28, 072001 (2016)

ηm Magnetic contribution to the spin viscosity

λ Ratio of magnetic particle-particle force to Brownian force
λv Number of surfactant molecules on particle surface
µ0 Permeability of free space
ν Arbitrary vector
ξ Ratio of the magnetic field force to Brownian force
σj Holonomic constraint j
σG Constraint for Gauss’s law
σA Constraint for Ampère’s law
ϕα Dipole orientation of particle α
∆ϕαB Brownian rotation of particle α
ϕαUN Unconstrained dipole orientation of particle α
ϕαCON Constrained dipole orientation of particle α
ω Angular velocity
Ω Angular velocity of the fluid in the absence of particles

I. INTRODUCTION
The study of ferrohydrodynamics began in the 1960s when National Aeronautics and Space
Administration (NASA) scientists were attempting to control liquids in space. It is concerned with
the fluid mechanics of ferrofluids that are under the effects of strong forces via magnetic polari-
zation.1 Recently, concerns have been focused on the remote positioning and control of ferrofluids
using external magnetic fields. For example, a single drop of ferrofluid can act as zero-leakage
rotary seals,2 pressure seals and sensors,3 and drug delivery vehicles.4 Larger quantities can serve
as detecting agents for magnetic domains in alloys,5 used as printer ink in high-speed printers,6
assist in catalytic reactors,7 and aid in particulate filtration.8 A ferrofluid is a colloidal suspension
composed of nanoscale (3-15 nm), ferromagnetic, single-domain particles suspended in a contin-
uous medium called the liquid carrier and can retain liquid fluidity in the presence of an intensely
applied magnetic field. Each particle is coated with a dispersant molecular layer called a surfactant
that prevents undesirable or irreversible clumping. At the center of every particle, a pair of equal
and opposite point poles coincides to form a permanently embedded magnetic dipole. Additionally,
because of their size, thermal agitation inhibits settling of the particles and keeps them suspended in
the liquid carrier via Brownian motion.
Traditionally, ferrofluid models have been treated as continuum systems where an additional
body force is appended to the Navier-Stokes equation to account for the relationship between the
applied magnetic field and the fluid’s magnetization.9 Magnetic fields that are generated by currents
are known as induction fields B. When these types of fields pass through magnetic materials, ambi-
guities arise about which segment of the induction field results from external influences and which
comes from the material itself. The most general law embodying this notion is
B = µ0 (H + M) , (1)
where the magnetic field H is the driving component of the field due to external influences within
the material, magnetization M is the material’s magnetic response that is often characterized by a
volume average of magnetic particle dipoles, and µ0 is the permeability of free space. However,
simulations have assumed that there is no complete interplay between the applied magnetic field
and the ferrofluid’s quasistationary magnetization. Magnetization has often been roughly defined
by a relaxation equation that assumes non-interacting particles in the absence of flow. Applied
magnetic fields are then presumed to be homogeneous and proportional to the corresponding local
magnetizations.10–13 Realistically, the orientation of the magnetic moment in ferrofluid particles is
altered by any force or interaction that affects the particles’ orientation and hence, its embedded
dipole. Examples of effects that influence the local magnetization include magnetic interactions
between the ferroparticles, particle tumbling initiated by hydrodynamics through the local flow
field, orientation changes produced by nonisotropic media (e.g., biological tissue, small channels,
072001-4 S. H. Dubina and L. E. Wedgewood Phys. Fluids 28, 072001 (2016)

ferroparticle crowding, etc.), and Brownian influences. Some, if not all, of these effects have been
neglected in conventional representations, where ferrofluid modeling corresponds to a dispersion
of magnetic particles so dilute that the particle contribution to the total field that animates all
particles is negligible.14,15 This assumption is inadequate since many applications, such as drug
delivery, require more concentrated ferromagnetic dispersions. Furthermore, some models, such as
Rosensweig’s “quasi-equilibrium” approach,1 did not account for all ferrohydrodynamic behavior,
such as the magnetoviscous effect. Magnetoviscous effects emerge in the form of an enhanced
effective shear viscosity due to excess dissipation when a flowing ferrofluid is subjected to a
magnetic field.16 And though useful in studying various effects, the models do not address the
orientation of the microstructure induced by the application of external magnetic fields or the details
of fluid-particulate-boundary substructure at extremely small length scales.
A comprehensive simulation routine is implemented to model dense ferrofluids under shear
flow in homogenous magnetic fields based on mesh-free techniques. This procedure is based on a
Brownian dynamics method and includes a constraint algorithm to ensure application of elementary
physics principles that govern magnetic fluids. The study presents an alternative method to handl-
ing the quasistationary magnetism (i.e., magnetization, magnetic field, and induction), especially
for dense particulate ferrofluids. By including any forces and interactions that affect the particle
dipoles, a practical representation of ferroparticles is realized. The proposed simulation technique
will potentially offer broader applications for other means to implement ferrofluids (such as under
non-homogeneous magnetic fields) and therefore, provide a much wider impact in understanding
magnetic fluids. Results from the study are then compared to existing Brownian dynamics simula-
tions and allow for a basis of validity and insight into the behavior of subcontinuum medium in the
realm of ferrohydrodynamics. Rheological properties are evaluated to analyze the field and shear
flow dependence of magnetoviscous effects.

II. METHOD
A. Brownian particle diffusion
Multibody hydrodynamic interactions occur when a particle moves about and agitates the fluid
in the neighborhood of other particles, even when transported by fluid motion. The resultant field of
the moving particle is conveyed through the solvent and affects the hydrodynamic force, torque, and
stresslet of the other particles. Hydrodynamic interactions must especially be accounted for in dense
particulate models with aggregation behavior. The approximation method of the additivity of veloc-
ities17 relates a particle’s velocity v and angular velocity ω, the stresslet S, the force F and torque
T applied to the fluid by a particle, and the rate-of-strain E. A mobility matrix M governs effects
between particles and consists of second, third, and fourth-rank tensors that manage the transla-
tional and rotational diffusivities as well as tensors that handle combinations of the translational and
rotational diffusion. These tensors occur between spheres of identical radii and are calculated as
functions of scalar mobility functions.18 And since mobility functions for particles with steric layers
have not yet been derived, mobility functions for a solid, spherical particle of diameter d must be
considered. This assumption is valid since the surfactant layer is a small fraction of the total radius
of the solid portion of a particle.
To describe the Brownian motion of a colloidal system of N particles with multibody hydrody-
namic interactions between them, generalized Langevin equations are employed to integrate for the
translational and rotational displacement expressions:1,19

N N
1  T 1  C


rα (t + ∆t) = rα (t) +  + P
+ D̃ (t) · T βP (t)

U (r α ) D (t) · F (t)

 kT β=1 α β β
kT β=1 α β
N

N
(2)
∂
  

+ g̃α (t) : ∆∞ + · DTα β (t)  ∆t + ∆rBα (t),

β=1
∂r β

β=1(,α)


072001-5 S. H. Dubina and L. E. Wedgewood Phys. Fluids 28, 072001 (2016)

N N
1  C 1  R


ϕα (t + ∆t) = ϕα (t) +  + P
+ D (t) · T βP (t)

Ω D (t) · F β (t)

 kT β=1 α β kT β=1 α β
N

N
(3)
∂
  

+ h̃α (t) : ∆∞ + · DC (t)   ∆t + ∆ϕα (t)
B

β=1
∂r β α β 
β=1(,α) 
where rα represents the vector to particle α, ∆t is the time step, k is Boltzmann’s constant, T is the
absolute temperature of the fluid, U and Ω are the solvent’s velocity and angular velocity vectors
for an applied flow field in the absence of particles, respectively, DT is the translational diffusivity
t
tensor, DR is the rotational diffusivity tensor, DC is the coupled diffusivity tensor, D̃C = DC ,
FP and T P are the force and torque of non-hydrodynamic interactions (steric repulsion, magnetic
particle-particle effects, and applied magnetic field forces), respectively, g̃ and h̃ are third-rank
tensors that contain mobility information for the hydrodynamic interactions, ∆∞ is the symmetric
B
part of the rate-of-strain tensor E, and ∆r and ∆ϕ B are the Brownian motion
) terms characterized

( B ) 
(
by zero mean and variances of ∆rα ∆r β = 2DTα β ∆t and ∆ϕαB ∆ϕ Bβ = 2DRα β ∆t, respec-
B

tively. In terms of the dynamics that describe non-hydrodynamic particle behavior, forces regu-
lating particle-particle magnetic interactions, particle-field responses, and steric repulsion are also
established.1
Because the method of the additivity of velocities is an approximation, restrictions are imple-
mented. When steric layers overlap, hydrodynamic interactions are neglected and a repulsive force
is activated. As long as the repulsive influence of the overlay dominates the lubrication effect, the
assumption does not disrupt the formation or development of aggregate structures. When particles
intersect beyond their steric layers, particle positions are adjusted so that only their steric layers
overlap.20 Conversely, application of a cutoff radius for hydrodynamic interactions prevents unso-
licited and inaccurate aggregation behavior by, again, ignoring hydrodynamic interactions when
particles α and β are separated far enough from one another. Thus, the hydrodynamic interactions
between particles α and β are developed while disregarding the influence of triple particle interplay.
Additionally, because the model consists of a strongly interacting system due to magnetostatic
forces, the far-field effect becomes negligible as a consequence of the cutoff radius for hydrody-
namic interactions. Finally, a general cutoff range for each particle provides a distance beyond
which non-hydrodynamic particle-particle interactions and magnetic attraction are neglected. These
distances drastically reduce the number of neighbors with which a particle can interact and thus,
lessens the computational time.
In respect to fluid properties, ferrofluids generally possess non-Newtonian characteristics as
a result of their colloidal nature. Their aggregates influence the fluid properties because of their
strict compliance towards the external magnetic field. Unlike the viscosity of low-molecular weight
liquids, the viscosity of a ferrofluid typically depends on the shear rate. Ferrofluids are usually
subject to shear thinning, since the viscosity contribution due to the magnetic field decreases with
increasing shear rates that affect the particle interactions in formed chains. Therefore, in order to
observe the effect of aggregates, the significance of the magnetic contributions to the viscosity is
explored while disregarding the particle stresslet elements. An expression for the viscosity contribu-
tion caused by magnetic interactions η m explores these effects.16 In this case, forces and torques are
adequate when characterizing the impact of the motion of the particle on the fluid.

B. Nondimensionalization
Parameter groups gathered from the governing forces can be combined to form dimensionless
groups that rely exclusively on the properties of the particles,
µ0 m 2 µ0mH πd 2s n s
λ= 3
, ξ= , λv = , (4)
4πd s kT kT 2
where H = |H|, H is the magnetic field vector, m = |mα |, mα is the direction vector of the magnetic
dipole moment of particle α, d s represents the particle diameter excluding the steric layer, and n s is
the number of surfactant molecules per unit area on the particle’s surface. When scaling factors are
072001-6 S. H. Dubina and L. E. Wedgewood Phys. Fluids 28, 072001 (2016)

used to nondimensionalize the forces, parameters combine to form additional dimensionless groups,

µ0 m 2 µ0mH kTλv
Rm = , Rh = , Rv = , (5)
64π 2ηa6γ̇ 8πηa3γ̇ 6πηa2γ̇δ
in which η is the viscosity of the fluid carrier, a (= d/2) is the ferroparticle radius that includes
the steric layer thickness δ, and γ̇ is the shear rate. Above, Rm depicts the ratio of the character-
istic magnetic particle-particle force to the characteristic hydrodynamic shear force, Rh represents
the proportion of the characteristic particle-field torque to the characteristic hydrodynamic shear
torque, and Rv signifies the ratio of the characteristic steric repulsion force to the characteristic
hydrodynamic shear force. They prove significant in managing specific stimuli on the particles.
Additionally, the Péclet number appears from the dimensionless Brownian translation and rotation

6πηa3γ̇
Pe = . (6)
kT
The Péclet number, like the prior three dimensionless numbers in Equation (5), is a ratio. It em-
bodies the proportion of characteristic hydrodynamic shear force to the characteristic Brownian
motion force. The values in Equation (4) are combined with the dimensionless groups in Equation
(5) to form ratios of dimensionless numbers

Rh ξ(1 + d δ )3 Rv λv(1 + d δ )4 1 (1 + d δ )3
= , = , = . (7)
Rm 4λ Rm 3λd δ RmPe 3λ
The purpose of these ratios is significant when preparing to execute the simulation, as they maintain
controlling factors while observing a single dominating influence.

C. System subject to constraints

Steady-state ferrofluids are governed by certain guidelines and must conserve essential princi-
ples of physics. Ferrofluid systems should maintain a free magnetic field while remaining current-
free throughout the volume. These concepts are embodied by the magnetostatic form of the
Maxwell’s equations and are known as Gauss’s law and Ampère’s law, respectively:
∇·B=0 (8)
∇×H =0 (9)
The induction field B is related via Equation (1). Gauss’s law ensures that there are no free magnetic
poles while Ampère’s law establishes that the tangential components of the magnetic field are zero,
especially when there are no current flows. In order to respect the postulates, the Brownian dy-
namics system adopts a constraint system to conserve them. Constraints are occasionally applied to
particulate simulations so as to satisfy specified conditions. Ottinger21 explains approaches to apply-
ing constraints with stochastic differential equations, the Fokker-Planck equation, and numerical
integration schemes. A functional constraint method for molecular dynamics has been developed by
Ryckaert et al.,22 which simultaneously solves the equations of motion with the constraint mecha-
nism at every time step of the integration. Liu23 integrated a constraint procedure into a Brownian
dynamics simulation in order to estimate rheological properties of Kramers freely jointed bead-rod
polymer chains. Therefore, while the present ferrofluid model responds to the resultant hydro-
dynamics, undergoes Brownian motion, drifts with the shear flow, and acquiesces to the applied
magnetic field, constraints can re-structure the system to satisfy the conditions. In other words, the
constraint method will “correct” the particles and prevent the model from diverging towards unreal-
istic behavior. Rather, the process used to constrain the Brownian dynamics simulation exploits an
iterative scheme that utilizes Lagrange multipliers. To authorize the correction, the system must first
meet the conditions of d ′ holonomic constraints defined by
σ j ({qν }) = f (q1, q2, . . . , q N ) = 0 ( j = 1, 2, . . . , d ′) , (10)
072001-7 S. H. Dubina and L. E. Wedgewood Phys. Fluids 28, 072001 (2016)

in which qν is the coordinate of particle ν = 1, 2, . . . , N and f is a constraint function dependent on

the configuration of the system. The coordinates of the particles are modified via
d′
 ∂σ j
qCON
α = qUN
α + γj , (11)
j=1
∂qα

where qCON
α and qUN
α are the constrained and unconstrained generalized coordinates, respectively,
and γ j are the scalar Lagrange multipliers. For our purposes, the constrained center of mass position
and dipole orientation of particle α, respectively, are

α (t + ∆t) = rα (t + ∆t) + β∆t Do Fα ,

t CON
rCON UN
(12)
ϕαCON(t + ∆t) = ϕαUN(t + ∆t) + β∆t Dor TαCON, (13)
with rUN
α as the unconstrained center of mass position of particle α, ϕαUN
is the unconstrained dipole
orientation of particle α, β = 1/kT, and D0t and D0r are the translational and rotational diffusion
coefficients at infinite dilution, respectively. Under convergence conditions, the constraint force and
torque, respectively, in the correction terms observed above are
d′
 ∂σ j
FCON
α =− γj , (14)
j
∂rα
d′
 ∂σ j
TαCON = −rUN
α (t) × γj . (15)
j
∂ϕα

With the preceding constraint procedure, Maxwell’s equations in (8) and (9) will be satisfied for the
ferrofluid system at each time step. When Equations (12) and (13) are inserted into the constraint
equations, a set of d ′ nonlinear equations with d ′ unknown Lagrange multipliers is formed. The
scalar Lagrange multipliers are
d′

γj = g̃ jl σ ( j = 1, 2, . . . , d ′).
 
c′ l
(16)
l=1

For every time step ∆t that transpires and with initial Lagrange multiplier values of γ j = 0, the iter-
ative constraint method converges when the second term of the right-hand sides of Equations (12)
and (13) approach zero for all σ j . As a consequence, the simulation can advance because the
correction terms are no longer needed. It is then that a revised configuration is visualized and the
simulation advances. The d ′ × d ′ matrix g̃ jl spotted above is the inverse of the modified metric
matrix G̃ jl and satisfies
d′

g̃ j m G̃ ml = δ jl , (17)
m=1

where δ jl is the Kronecker delta. The modified metric matrix is

N
 ∂σ j ∂σl
G̃ jl = · .
α=1
∂qα ∂qα
This matrix demonstrates the dependency of the constraints on one another and is essential for each
iteration. This is because all constraints rely on the system’s configuration. The brackets [· · · ]c ′
around g̃ in Equation (16) serve as an evaluation range for position from time t to t + ∆t, manifested
as c ′ ∈ [0, 1].
Concerning the constraints for our particular model, by applying a volume integral around
Equation (8) (and a surface integral around Equation (9)), employing integral theorems to convert
to lower level integrals, and utilizing zero-order approximations allows for a node-based constraint
procedure. The constraints for Gauss’s law and Ampère’s law are simplified to the following
072001-8 S. H. Dubina and L. E. Wedgewood Phys. Fluids 28, 072001 (2016)

zero-order approximations, respectively:


σG = ∆Sk Mk · n̂ + Hk · n̂, (18)
k

σA = Hk ′ · ∆l, (19)
k′

where Mk and Hk are the magnetization field and magnetic field, respectively, centered at node k
(or node k ′ when concerning Ampère’s law), n̂ is the surface normal vector, ∆Sk is the differential
area around node k, and ∆l is the change in the linearization along the boundary that encloses the
constraint surface. The nodes are solely used for the calculation of magnetization and magnetic
fields for the constraint equations since they are volume-averaged position-dependent density prop-
erties. The integral theorems and numerical approximations used to simplify the constraints bring
about a significant decrease in the number of nodes, thus reducing computation time. Furthermore,
the weighting function allows for a spatially continuous density measurement along the node points.
Ideally, additional nodes could be added, further reducing the size of the “sub-cells,” allowing the
constraint to be met in even smaller volumes and ultimately leading to a more accurate picture of the
system’s behavior. However, the model will consider a low number of constraints in consideration
of computational limits.
When manipulated by an applied magnetic source, ferrofluid particles within the system
contribute to the effective magnetic field. The magnetic field H of the ferrofluid is calculated as

H = H0 + Ĥ, (20)
where H0 is the external magnetic field in the absence of magnetic material and Ĥ is the accumu-
lation of other magnetic causes identified as the demagnetizing field. H = 0 only when the particles
lose their magnetization and the external magnetic field is absent. Otherwise, the model considers
the contributions to H, even if the magnetic field is uniform (H0 = constant). In this model, H0 is
perpendicular to the fluid flow direction (in the Cartesian y-direction) since the effect of an applied
uniform magnetic field is strongest when H0 is oriented this way.24 The Ĥ field can be calculated by
the following:1
N  
 V 3 [M · (r − rα )] (r − rα )
Ĥ(r) = −M + , (21)
α
4π|r − rα |3 |r − rα |2

where r is the position at which the Ĥ field is being evaluated and V is the volume of the magnetic
source centered at r. The external field generated by a spherical particle that is homogeneously
magnetized is identical to that of a point dipole possessing the same total magnetic moment and
originating from the sphere’s center. Furthermore, the magnetic field at any arbitrary position can be
acquired by assuming superposition of the fields.
To quantify the magnetization at any point in the system, a symmetric, radial weighting func-
tion W is proposed,25
( )2 ( )3
∥ν∥ ∥ν∥
W (ν, h) = 1−3 +2 . (22)
h h
Here, ∥ν∥ is the norm of an arbitrary vector ν and h is the cutoff radius for particle contributions
to the local field. With a suitable h value, the entirety of the system’s volume is incorporated when
calculating the constraints. The weighting function is used to calculate the magnetization via
N
1
M (r) = nαW (ν, h) , (23)
V α

where nα represents the unit direction vector of mα . We appoint ν to be the vector between particle
α and position r.
072001-9 S. H. Dubina and L. E. Wedgewood Phys. Fluids 28, 072001 (2016)

D. Model design
In the present model, ferrofluid particles are scrutinized under various conditions, notably un-
der a simple planar shear flow. Because ferroparticles (or their potential colloid formations) tend
to be much larger than the molecules of the surrounding medium, the liquid carrier is viewed as a
continuum, allowing shear flow to be designated by analytical field solutions for Stokes flow.26 The
liquid carrier is encased within a cubic volume characterized by dimensions L of 40a × 40a × 40a
and encloses an assemblage of N = 512 unit dipoles. Since the volume and total particle number are
dependent on other ferrofluid attributes, they are useful for calculating the ferroparticle diameter,
volume fraction, or particle density, as long as these properties are within sensible ranges that are
representative of conventional ferrofluid solutions. For simplicity, the center of the system coincides
with the origin of a three-dimensional Cartesian coordinate system and has lower and upper ranges
of −20a and +20a, respectively. The size of this system depicts a relatively small prototype of
a ferroparticle dispersion and can be viewed as a model of secondary particles, each of which is
comprised of many primary ones. Each dipole is envisioned as a sphere of radius a because of
its relatively simple geometry and appearance in many colloidal systems. Spheres are known to
facilitate complicated numerical techniques, even those designed with a uniform steric layer. For a
spherical particle housing a fixed dipole, the orientation of strongest attraction is when dipoles are
parallel, meaning the ends are attracted to their corresponding opposites. When under the influence
of a uniform magnetic field, particles accumulate to form aggregates along the y-direction with
dipoles pointing in the primary field direction.
To simulate a ferrofluid system between parallel plates, wall boundaries are implemented. The
particle-wall design27 is valuable for its ability to mimic solid boundary roughness in a simplistic
manner, both methodically and computationally. Hence, the top and bottom xz-planes (located at
r y = ±20a) are composed of wall particles which impose a reflective boundary condition when fer-
roparticles overlap them. The wall particles are set in a hexagonal packing arrangement of identical
circles of radius a (when viewed in a two-dimensional plane) and are constrained from reposition-
ing relative to one another. When shear flow is active, the particles in the wall layers shift according
to the shear velocity. Because of the boundary, the ferrofluid particles are contained within the limits
of the y-direction but are free to move between the parallel plates. Conversely, where there is no
wall boundary, the system is considered to have a periodic boundary. The simulation cube has been
defined to have periodic boundaries in the x- and z-directions, permitting the particles to diffuse
freely in a continuous fluid. As a result, a system volume is demarcated.
By controlling the values of the dimensionless numbers introduced in Equations (4)–(6), specific
effects can be monitored. However, before dictating particle-particle magnetic interactions, particle-
field response, aggregation behavior, Brownian motion influence, and shear flow impact, dimension-
less numbers must be regulated by virtue of the ratios introduced in Equation (7). When the simulation
is compiled with λ = (3, 5, 9), the ratios of Rv/Rm and 1/ (RmPe) result in (158.7, 95.2, 52.89) and
(0.244, 0.146, 0.0814), respectively. The ratio of Rh/Rm will be maintained around 3.3, a magnitude
that was found to support the assumption of a strong magnetic field. The dimensionless property λ
primarily determines the degree to which the magnetostatic particle-particle and particle-field interac-
tions overwhelm the Brownian motion effects. Most importantly, for λ = 9, magnetostatic interactions
dominate nearly all Brownian motion effects, thus enabling chain formation amongst particles and
along the y-direction. Hence, this property will be analyzed over the others in the present study. Under
this constant, ξ = 54.07 for Rh/Rm = 3.3 and the steric attributes of d δ and λv are held constant at 0.3
and 150, respectively. The parameters are listed in Table I below.
For the constraint method to fulfill the Gauss’s law constraints σG, j for j = 1, 2, . . . , d G
′
, the sys-
tem volume is split into d G = 8 equal cubes, each with dimensions L c of 20a × 20a × 20a, so as to
′

represent local volumes of the model. Each surface of a local cube is populated with 16 nodes in a
4-by-4 evenly distributed array with a spacing of 5a. This is roughly visualized in Figure 1(a). At
each node, the weighting function is employed, as described in Equation (22), to calculate the M
field in Equation (23) while the H field in Equation (20) is processed via Equation (21) alongside an
appropriate H0. A surface integral is then calculated for each of the eight local volumes, which allows
Gauss’s law to be satisfied over eight local constraints. Fortunately, the periodic boundaries in the x- and
072001-10 S. H. Dubina and L. E. Wedgewood Phys. Fluids 28, 072001 (2016)

TABLE I. Parameter sets for R h/R m = 3.3, d δ = 0.3, and λv = 150.

1 Rv
λ R mPe Rm Rm ξ

3 0.244 158.67 50 18.02

5 0.146 95.2 50 30.04
7 0.105 68.002 50 42.06
9 0.0814 52.891 1 54.07
9 0.0814 52.891 10 54.07
9 0.0814 52.891 50 54.07
9 0.0814 52.891 100 54.07

z-directions reduce the total number of surface integral calculations required for the xy- and yz-faces
at r z = ±20a and r x = ±20a, respectively. As for magnetostatic Ampère’s law, the constraints σ A, j
for j = 1, 2, . . . , d ′A must be met at the boundaries to ensure that the tangential component of magnetic
field is continuous. At the wall boundaries, the borders surrounding the surfaces are segmented into
evenly distanced nodes. In this case, each of the d ′A = 2 borders contains 32 nodes, eight per side with
a spacing of 5a. This constraint design is crudely illustrated in Figure 1(b), with the normal direction to
the surface dictating the positive direction of the arrows circumnavigating the boundary. A line integral
is evaluated for each of the two borders, which allows Ampère’s law to be satisfied over two local
constraints. In conclusion, the system employs no more than eight local volume constraints and two
boundary constraints in consideration of computational limits. Thus, the constraints are developed into
viable expressions that are computationally efficient for the present study.

III. RESULTS
A. Constraint trends and comparison
While the simulation advanced by a time step, the particles were impacted by a uniformly
applied magnetic field, Brownian motions, interparticle interactions, and shear flow effects before

FIG. 1. System representation of the (a) local constraint volumes for Gauss’s law with nodes k and (b) local constraint
boundary surfaces for Ampère’s law with nodes k ′. (c) A rough illustration of the Cartesian xy-view of planar shear flow with
shear rate γ̇.
072001-11 S. H. Dubina and L. E. Wedgewood Phys. Fluids 28, 072001 (2016)

FIG. 2. Trends of (a) Gauss’s law and (b) magnetostatic Ampère’s law (λ = 9, R m = 50).

the constraints were initiated. Then, to ensure that Maxwell’s equations are continuously met,
the constraint equations in (18) and (19) were tracked for models with and without an activated
constraint method. The simulations were simultaneously executed over 30 000 nondimensional
time-steps, which is an ample amount of time for an initially randomized ferroparticle system to
reach a steady state. They are run with parameters of λ = 9 and Rm = 50. From this point for-
′
ward, the superscript “*” represents the dimensionless form of properties. For d G = 8 constraints
to satisfy Gauss’s law, Figure 2(a) illustrates the trends between a model whose total magnetic
field is divergence-free to one that is not. For d ′A = 2 constraints representing Ampère’s law,
Figure 2(b) compares results with a magnetic field that is curl-free compared to another that is
not.

B. Particle distributions
A steady-state representation of the system provides a means of observing the general behavior
of ferrofluids and roughly illustrates chain formation and particle diffusion. To inspect the aggre-
gates further, distinct rectangular prisms of the volume (deemed “slices”) scrutinize particles in a
localized volume. In this study, xy-, yz-, and xz-slices examined particles within Cartesian spaces
enclosed between xy-, yz-, and xz-planes, respectively, each of dimensions 40a × 40a. Specifically,
xy-slices were placed at z = −2.5a and z = +2.5a, yz-slices situated at x = −2.5a and x = +2.5a,
and xz-slices sited at y = −2.5a and y = +2.5a. The coordinates were chosen in order to observe
particle activity around the center of the system volume. Moreover, while the xy- and yz-slices
emphasize the chaining that forms along the y-direction, the xz-slice inspect any dipoles shifting
away from the magnetic field bearing. Thus, with the spatial particle distributions and their slices,
072001-12 S. H. Dubina and L. E. Wedgewood Phys. Fluids 28, 072001 (2016)

FIG. 3. Spatial particle distribution (λ = 9, R m = 50). Strong particle aggregation is observed under these parameters.

we can scrutinize the aggregation behavior of ferrofluid particles as the model progresses. In addi-
tion, note that slices may not display complete chains due to adjacent particles that are located just
beyond the boundaries of the slice. Also, the position vector r and coordinates are divided by a to
generate their respective dimensionless forms.
After running the Brownian dynamics simulations for 30 000 time-steps, spatial distributions
and planar slices characterize the resultant ferrofluid particle activity. Under values of λ = 9 and
Rm = 50, and with an activated constraint algorithm, strong magnetostatic attractions are known to
invalidate the Brownian dynamic effects. For the case of a shear fluid flow, Figure 3 demonstrates a
particle distribution that has reached equilibrium, whereas Figure 4 illustrates the slices.
Under a decreased Rm value, the quotient of magnetostatic effects to viscous shear forces
is decreased while maintaining the previously stated ratios of dimensionless numbers. This was
accomplished by setting the Rm dimensionless numbers to 1.0, which corresponds to relatively
weak magnetostatic interactions. The particular value for Rm preserves the dominance of magnetic
particle-particle attraction over Brownian motion. However, with the diminished Rm value, the Pé-
clet number becomes 12.29 and the viscous shear forces possess a much greater impact on chain
evolution. The particle distribution and slices under these effects are observed in Figures 5 and 6,
respectively.
Under a substantially large Rm value, the impact of the magnetic particle-particle and applied
field effects increases. In this case, Rm was increased to 100, imposing a rise in the ratio of
magnetostatic effects to viscous shear forces while preserving the aforementioned ratios of dimen-
sionless numbers. Consequently, chaining activity was expected to flourish under this parameter set,
representations of which are perceived in Figures 7 and 8.
Alternatively, a value of λ = 3 indicates the prevailing Brownian motion. The thermal energy
disbands any particle pairs, despite the strong magnetic attractions, and overpowers the viscous
shear forces. However, since the uniform external magnetic field primarily affects the torque on the
particles, and when the Rh value is high enough, the rotational Brownian torque is not enough to
overcome it. A system under these effects is portrayed by a spatial distribution in Figure 9 and by
planar slices in Figure 10.

C. Pair correlations
Although the spatial distribution of the volume appears to demonstrate the aggregation behavior
of particles under the influence of an external magnetic field, one must closely analyze the structure
of the chains. A pair correlation function (PCF) provides a method of quantifying the chain
formation in the models.17 Peaks on a PCF graph characterize the probability of the number of
072001-13 S. H. Dubina and L. E. Wedgewood Phys. Fluids 28, 072001 (2016)

FIG. 4. (a) xy-, (b) yz-, and (c) xz-slices (λ = 9, R m = 50). Planar slices at steady-state provide a comprehensive view of the
strong chaining behavior.

particles adjacent to a particle α and the direction they align relative to the applied magnetic
field. The algorithm involves measuring the angle θ and distance r from particle α to others. The
angle θ designates the deviation from the magnetic field and is measured from the center-to-center
vector of a particle pair to the primary magnetic field direction H0. The distance r is taken from

FIG. 5. Spatial particle distribution (λ = 9, R m = 1). Weak particle aggregation but strong shearing effects are observed under
these parameters.
072001-14 S. H. Dubina and L. E. Wedgewood Phys. Fluids 28, 072001 (2016)

FIG. 6. (a) xy-, (b) yz-, and (c) xz-slices (λ = 9, R m = 1). Planar slices at steady-state provide a comprehensive view of the
strong shearing effects and lack of agglomeration.

FIG. 7. Spatial particle distribution (λ = 9, R m = 100). Significantly strong particle aggregation is observed under these
parameters.
072001-15 S. H. Dubina and L. E. Wedgewood Phys. Fluids 28, 072001 (2016)

FIG. 8. (a) xy-, (b) yz-, and (c) xz-slices (λ = 9, R m = 100). Planar slices at steady-state provide a comprehensive view of the
powerful chaining behavior.

the center of particle α to any adjoining particles. The values are added up and plotted in a pair
correlation graph. The peaks equate to the probability of chain lengths at angles θ. The PCF graphs
for the preceding models represented in the spatial particle distributions above are depicted in
Figures 11–14.

FIG. 9. Spatial particle distribution (λ = 3, R m = 50). Dominating Brownian effects and weak particle aggregation are
observed under these parameters.
072001-16 S. H. Dubina and L. E. Wedgewood Phys. Fluids 28, 072001 (2016)

FIG. 10. (a) xy-, (b) yz-, and (c) xz-slices (λ = 3, R m = 50). Planar slices at steady-state provide a comprehensive view of the
prevalent Brownian effects and inadequate chaining.

FIG. 11. Graph of the pair correlation function (λ = 9, R m = 50). Large pair correlation functions under these parameters
indicate a notable extent of particle chaining.
072001-17 S. H. Dubina and L. E. Wedgewood Phys. Fluids 28, 072001 (2016)

FIG. 12. Graph of the pair correlation function (λ = 9, R m = 1). Sparse traces of pair correlation quantities under this
parameter set imply effective shearing that inhibits magnetic particle interactions.

D. Magnetoviscous effects
From the models that have replicated exceptional chaining conduct, we can observe how
the particle chains have affected the local ferrofluid behavior. Since there are distinct chains that
conform to the applied magnetic field (and viscous shear forces), the dimensionless viscosity is em-
ployed to evaluate the effects of the aggregates on the non-Newtonian fluid property. The viscosity
was time-averaged at equilibrium for each parameter set and simulation run.
Increasing the value of Rm under a constant λ varies the significance of the magnetic particle-
particle interactions and allows one to monitor the dependence of shear flow on the non-Newtonian
viscosity contribution. Figure 15 exhibits two differing trends. The first, marked by trianglular
icons, possesses a deactivated constraint procedure. The second, represented by square-shaped char-
acters, incorporates the constraint algorithm. Error bars around the data points signify the standard
deviation. These two conditions were also employed across various λ values at a constant Rm = 50
and are shown in Figure 16.

FIG. 13. Graph of the pair correlation function (λ = 9, R m = 100). High pair correlation peaks under these parameters
designate exceptional levels of particle aggregation.
072001-18 S. H. Dubina and L. E. Wedgewood Phys. Fluids 28, 072001 (2016)

FIG. 14. Graph of the pair correlation function (λ = 3, R m = 50). A lack of pair correlation measurements with these
parameters suggests that the Brownian forces overpowered particle chain development.

IV. DISCUSSION
A. Constraint trends and comparison
When analyzing Figure 2(a), the simulation with an activated constraint procedure has satisfied
Gauss’s law and consequently, produces a trend line at zero. Conversely, the trend without Gauss’s
law constraints fluctuates uncontrollably. The latter situation would be acceptable for macroscopic
simulations. However, for time steps as small as the ones used in this Brownian dynamics study, it
is unreliable since the fields must be solenoidal at each time step. A similar situation applies for the
satisfaction of magnetostatic Ampère’s law exhibited in Figure 2(b). It is apparent that the fluctua-
tions for Ampère’s law are less wild than those of Gauss’s law, which may be due to the range of
the boundary surfaces as opposed to the reach of the local constraint volumes. More importantly, the
trends demonstrate a curl-free magnetic field throughout the volume. Thus, the constraint algorithm
has provided a means to comply with governing Maxwell’s equations while the system advances.

B. Particle distributions
Since we were able to apply the constraints of Maxwell’s Equations, we captured the spatial
distributions and slice illustrations of the ferrofluid particles at steady-state. When λ = 9 and

FIG. 15. Effect of shear flow on spin viscosity (λ = 9).

072001-19 S. H. Dubina and L. E. Wedgewood Phys. Fluids 28, 072001 (2016)

FIG. 16. Spin viscosity dependence on magnetic particle-particle effects (R m = 50).

Rm = 50, the magnetically attracted particle chains reveal a slight slant due to the shear flow profile,
even though the dipoles in Figure 3 still remain visibly aligned in the magnetic field direction and
are hardly skewed by the viscous shear torques. When viewing any planar slice of Figure 3, chains
are better perceived. The xy- and yz-slices portrayed in Figures 4(a) and 4(b), respectively, verify
that the fluid flow affects the aggregate structures of the ferrofluid rather than the constituent particle
dipoles. The xy-slice in Figure 4(a) accentuates the aforementioned slanting effect of the viscous
shear forces on the particle chains. In Figure 4(b), the yz-slice does not depict a diagonal lean in
the aggregates because the shear flow moves in the x-direction and relies solely on the y-position.
And finally, the xz-slice in Figure 4(c) confirms a prevailing magnetic field influence since hardly
any dipoles shifted away from the y-direction. Thus, the internal structures of the chains exhibit
typical ferrofluid agglomeration behavior in the presence of shear-flowing ambient fluid and strong
magnetic fields.
By decreasing the Rm value, the magnetic interplay between particles is not capable enough to
generate consistent agglomeration, even when the applied magnetic field is substantial enough to
keep the particle dipoles directed in the primary y-direction.28 Figure 5 exhibits a nearly complete
deficiency of aggregated particle chains present after 30 000 time steps, compared to Figure 3.
Additionally, the dipoles are leaning noticeably away from the magnetic field direction due to the
shear flow forces. And, similar to Figure 4, the slices in Figure 6 support these notions. There is
considerable inclining of the dipoles in the shear flow direction in Figure 6(a) but none observed
by the yz-slice in Figure 6(b). Since the external magnetic field is still stronger than the Brownian
motion, the dipoles in Figure 6(c) are not directed in the z-direction. Also, though most of the dipole
is obstructed by the particle’s size in the xz-slice, one can notice a bit of it pointed in the x-direction
due to viscous shear.
Conversely, a relatively high Rm value implies that the particles are not affected by the viscous
shear forces, primarily due to the low Péclet value and dominant magnetic particle-field interac-
tions. As exemplified by Figure 7, the chains appear denser. Observing the xy-slice in Figure 8(a)
confirms that the chains are not affected by the shear flow and do not tilt like the aggregates in
Figure 4(a). And in this incident, because the ambient fluid has hardly any bearing on the parti-
cles, the yz-slice in Figure 8(b) exhibits a similar trend to the xy-slice. In terms of the xz-slice
in Figure 8(c), it exhibits no unusual behavior and affirms prior observations. Thus, under these
circumstances, the applied magnetic field is effective enough to overcome the viscous shear forces
and induce long particle chain aggregation solely along the y-direction.
When the value of λ is decreased to a reasonably low number, Brownian motion becomes
the prevailing factor driving the particles. In Figure 9, after 30 000 time steps, the particles appear
sporadically dispersed. The shear torque is not substantial enough to offset the dipoles completely,
as indicated by a low Péclet number. Moreover, this outcome is believed to be further verified when
the model is supervised under a non-uniform magnetic field or a low Rh constant. The planar slices,
again, allow a closer look at the internal structure. Both Figures 10(a) and 10(b) reveal minimal
072001-20 S. H. Dubina and L. E. Wedgewood Phys. Fluids 28, 072001 (2016)

aggregation with dipoles steered in the direction of the field. Figure 10(c) suggests some deviation
of the dipole away from the field direction when compared to the other models, though not consid-
erable when claiming Brownian motion responsible. Thus, it is shown that the Brownian motion
prevents particle chain aggregation in the presence of a strong magnetic field.
Based on the results presented and as a consequence of Brownian motion, the aggregates
can be said to have loose internal structures, even when the particles form long chains. Brown-
ian dynamics, though trivial compared to the magnetostatic effects, is not wholly absent in the
simulations. Brownian motion tends to slightly disturb the orientation of the magnetic moments
in the presence of an externally applied field, which occasionally leads to crooked chaining or
inter-particle separation. However, this effect is not a notable concern because it does not occur
enough to separate chains unless the parameters designate otherwise.

C. Pair correlations
The PCF graph for the model under λ = 9 and Rm = 50 demonstrates substantial peaks along
θ = 15◦-25◦ and verifies the chaining behavior due to strong magnetic interactions between particles
and vigorous agglomeration tendencies. These observations are perceived in Figure 11. Most of
the peaks emerge, on average, around θ = 19◦ because of the strong influence due to the external
magnetic field and minimal deviation away from it. The most notable quantities range from r = 2
to r = 6 since the PCF detects long chains of ferrofluid particles. The highest peak at r = 2 demon-
strate that most of the particles pair well with each other, meaning that the pairing is dominant
between twosomes. Peaks around r = 3 and r = 4 suggest long particle chains along the direction of
the field, observed in Figure 3. And although most of the peaks materialize along θ = 15◦-25◦, some
appear at greater angles due to the viscous shear. Almost all of the particle pairs that were perturbed
by the shear flow measured within 30◦ from the field direction. For those that were angled beyond
were found to be in close proximity to each other rather than coupled by magnetic particle-particle
attraction, mainly because the PCF does not regard particle dipoles. Therefore, residual quantities
have appeared beyond 30◦.
In regards to the data in Figure 5, the PCF plot in Figure 12 exhibits much lower peaks when
compared to Figure 11 above. This is due to the viscous shear forces that disperse any particles
attempting to form chains. As mentioned before, the effects due to shear flow considerably outweigh
those from magnetostatic attraction. The peaks previously detected at r = 2 are hardly noticeable as
a result of significant viscous shearing. Peaks also deviated heavily away from the field direction
due to existing chains that separated because of the shear flow profile. Thus, the PCF graph is
practically flat, which confirms the earlier analysis stated of Figure 5.
For the particle distribution in Figure 7, the peaks in Figure 13 are much higher than those when
Rm = 50. This emphasizes the high Rh constant that enables a relatively strong external magnetic
field that overshadows viscous shear and Brownian motion. Consequently, the likeliness for particle
pairing increased. And because the shear flow has less of an impact, the aggregates are consider-
ably less angled. In this event, the peaks appear largely about θ = 16◦, which is not as distributed
across particle angles as the ones in Figure 11. Also, the PCF peaks in the vicinity of r = 2 are
considerably higher than those in Figure 11. These findings are indicative of the chaining viewed in
Figure 7.
In the case of dominant Brownian motion (when λ = 3 and Rm = 50) visualized in Figure 9,
one would expect the PCF to generate no visible peaks. The thermal motion hinders any potential
agglomeration and thus, the PCF should be a flat plane of low probability. As indicated in Figure 14,
the deficiency of peaks implies that chaining is not present in the system volume at equilibrium.
The results are similar to those of the PCF of Figure 12. And analogous to the data presented in
Figure 12, visible remnants of peaks seen in Figure 14 are predominantly around θ = 15◦-25◦, since
the applied magnetic field is not entirely obscured and allows for some residual particle clusters to
momentarily surface. Thus, our earlier expectations of the particle distributions are verified.
072001-21 S. H. Dubina and L. E. Wedgewood Phys. Fluids 28, 072001 (2016)

D. Magnetoviscous effects
By varying the Rm value under a constant λ, we can grasp an understanding of how the shear
flow influences the magnetoviscosity. In Figure 15, when Rm = 1, the measured viscosities approach
a value of zero and the variation in the time averages are minimal. The trend marked by trianglu-
lar icons is akin to existing data.29 However, when the constraint procedure is incorporated, the
η m/η trend declined. This decrease, designated by square symbols, demonstrates that Maxwell’s
equations are not entirely met and we confirm that they should be more closely maintained (as
verified in Figure 2). This supports the observation that the component of the spin viscosity is not as
extensive as once realized and that not all effects on the ferroparticle dipole are included in recent
research. Similar conclusions can be made for the non-Newtonian magnetoviscosity contribution
trends in Figure 16, when several λ values were observed under a constant Rm. Again, the scenarios
controlled by the constraint algorithm appear considerably lower than recent studies suggest.

V. CONCLUSIONS
The computational methodology in our research advances engineering fundamentals to develop
efficient, three-dimensional simulations of ferromagnetic materials. Our iterative constraint algo-
rithm provided a means for an effective pseudo steady state treatment of Maxwell’s equations,
which were intentionally satisfied for every time step that the Brownian dynamics model advanced.
We found that as a result, the magnetoviscous effect from this simulation was discovered to be less
influential than those without constraints to fulfill both Gauss’s law and Ampère’s law, confirming
that the magnetic field and magnetization of superparamagnetic materials should be continually
monitored. The result was perceived when adjusting the effects of magnetic particle-particle interac-
tion, shear, and Brownian motion. It verifies our alternative procedure to managing the connection
between the magnetic field and the material’s magnetism. The approach proved sufficient for the
integrated flow and field conditions since we have accomplished the task of simulating a system of
ferrofluid particles under a uniformly applied external magnetic field in a volume with shear flow
and wall boundaries. Hence, our focus can now be diverted towards determining other situations
that arise when utilizing ferrofluids. It will also increasingly exploit coupling between different
transport and reactive phenomena and the interaction between fluids and any encompassing micro-
fabricated systems.
The next objective of the research is to observe the diffusion behavior of a ferrofluid under
non-uniform magnetic fields. A non-uniformly applied magnetic field will be incorporated into
the particle system and allowed to progress towards a suitable equilibrium state. By observing the
movement of ferrofluid particles over time, we can evaluate the magnetophoretic velocity, the effect
on characteristic times, and non-Newtonian fluid properties while satisfying Maxwell’s equations
that are regulated by the proposed constraint algorithm. We trust that the findings will accurately
predict the behavior of ferrofluids in microchannel separation, an essential application in the realm
of magnetic fluids, especially since magnetic inertia is typically presumed to be negligible in uni-
form magnetic fields. And so, while a uniform magnetic field leads to data that interprets aggrega-
tion behavior amongst ferrofluid particles, applications with non-uniformly applied magnetic fields
will facilitate the ongoing research in magnetophoretic properties of supposed particles.

ACKNOWLEDGMENTS
The authors wish to thank L. Nitsche for contributing the weighting function and assistance
during the revision process, T. Walker for assisting in data analysis, and A. Tillotson for his physics
advice and expertise.
1 R. E. Rosensweig, Ferrohydrodynamics (Cambridge University Press, Cambridge, 1985).
2 G. Schinteie, P. Palade, L. Vekas, N. Iacob, C. Bartha, and V. Kuncser, “Volume fraction dependent magnetic behaviour of
ferrofluids for rotatingseal applications,” J. Phys. D: Appl. Phys. 46(39), 395501 (2013).
3 R. Ravaud, G. Lemarquand, and V. Lemarquand, “Mechanical properties offerrofluid applications: Centering effect and

capacity of a seal,” Tribol.Int. 43(1-2), 76 (2010).

072001-22 S. H. Dubina and L. E. Wedgewood Phys. Fluids 28, 072001 (2016)

4 D. K. Kim and J. Dobson, “Nanomedicine for targeted drug delivery,” J.Mater. Chem. 19(35), 6294 (2009).
5 K. Raj, B. Moskowitz, and R. Casciari, “Advances in ferro fluid technology,” J. Magn. Magn. Mater. 149(1-2), 174 (1995).
6 P. Tiberto, G. Barrera, F. Celegato, M. Coïsson, A. Chiolerio, P. Martino, P. Pandolfi, and P. Allia, “Magnetic properties of

jet-printer inks containing dispersed magnetite nanoparticles,” Eur. Phys. J. B 86(4), 173 (2013).
7 A. Pourjavadi, S. H. Hosseini, M. Doulabi, S. M. Fakoorpoor, and F. Seidi, “Multi-layer functionalized poly(ionic liquid)

coated magnetic nanoparticles: Highly recoverable and magnetically separable Brønsted acid catalyst,” ACS Catal. 2(6),
1259 (2012).
8 C. T. Yavuz, A. Prakash, J. T. Mayo, and V. L. Colvin, “Magnetic separations: From steel plants to biotechnology,” Chem.

Eng. Sci. 64(10), 2510 (2009).

9 M. I. Shliomis, “Magnetic fluids,” Sov. Phys.-Usp. 17(2), 153 (1974).
10 R. Ganguly, A. P. Gaind, S. Sen, and I. K. Puri, “Analyzing ferrofluid transport for magnetic drug targeting,” J. Magn. Magn.

Mater. 289, 331 (2005).

11 M. Kružik and A. Prohl, “Recent developments in the modeling, analysis, and numerics of ferromagnetism,” SIAM Rev.

48(3), 439 (2006).

12 N. Arsalani, H. Fattahi, S. Laurent, C. Burtea, L. V. Elst, and R. N. Muller, “Polyglycerol-grafted superparamagnetic iron

oxidenanoparticles: Highly efficient MRI contrast agent for liver and kidney imaging and potential scaffold for cellular and
molecular imaging,” Contrast Media Mol. Imaging 7(2), 185 (2012).
13 C. Rümenapp, B. Gleich, and A. Haase, “Magnetic nanoparticles in magnetic resonance imaging and diagnostics,” Pharm.

Res. 29(5), 1165 (2012).

14 R. S. Newbower, “Magnetic fluids in the blood,” IEEE Trans. Magn. 9(3), 447 (1973).
15 D. Kumar, P. Chandra, and P. Sinha, “Ferrofluid lubrication of externally pressurized circular plates and conical bearings,”

Int. J. Eng. Sci. 31(4), 593 (1993).

16 S. Odenbach, Magnetoviscous Effects in Ferrofluids (Springer, Berlin, 2002).
17 A. Satoh, R. W. Chantrell, G. N. Coverdale, and S. Kamiyama, “Stokesian dynamics simulations of ferromagnetic colloidal

dispersions in a simple shear flow,” J. Colloid Interface Sci. 203(2), 233 (1998).
18 S. Kim and S. J. Karrila, Microhydrodynamics: Principles and Selected Applications (Dover Publications, Mineola, 2005).
19 D. L. Ermak and J. A. McCammon, “Brownian dynamics with hydrodynamic interactions,” J. Chem. Phys. 69(4), 1352

(1978).
20 D. M. Heyes and J. R. Melrose, “Brownian dynamics simulations of model hard-sphere suspensions,” J. Non-Newtonian

Fluid Mech. 46(1), 1 (1993).

21 H. C. Öttinger, Stochastic Processes in Polymeric Fluids: Tools and Examples for Developing Simulation Algorithms

(Springer, Berlin, 1996).

22 J. Ryckaert, G. Ciccotti, and H. J. Berendsen, “Numerical integration of the Cartesian equations of motion of a system with

constraints: Molecular dynamics of N-alkanes,” J. Comput. Phys. 23(3), 327 (1977).

23 T. W. Liu, “Flexible polymer chain dynamics and rheological properties in steady flows,” J. Chem. Phys. 90(10), 5826

(1989).
24 Z. Wang, V. B. Varma, H. M. Xia, Z. P. Wang, and R. V. Ramanujan, “Spreading of a ferrofluid core in three-stream mi-

cromixer channels,” Phys. Fluids 27(5), 052004 (2015).

25 L. C. Nitsche and W. Zhang, “Atomistic SPH and a link between diffusion and interfacial tension,” AIChE J. 48(2), 201

(2002).
26 R. B. Bird, W. E. Stewart, and E. N. Lightfoot, Transport Phenomena, 2nd ed. (J. Wiley, New York, 2002).
27 P. Spijker, H. M. M. Ten Eikelder, A. J. Markvoort, S. V. Nedea, and P. A. J. Hilbers, “Implicit particle wall boundary

condition in molecular dynamics,” Proc. Inst. Mech. Eng., Part C 222(5), 855 (2008).
28 S. Odenbach, Ferrofluids: Magnetically Controllable Fluids and Their Applications (Springer, Berlin, 2002).
29 A. Satoh, R. W. Chantrell, and G. N. Coverdale, “Brownian dynamics simulations of ferromagnetic colloidal dispersions

in a simple shear flow,” J. Colloid Interface Sci. 209(1), 44 (1999).