Comput Mech (2016) 58:151–169

DOI 10.1007/s00466-016-1286-y

ORIGINAL PAPER

Homogenization in micro-magneto-mechanics
A. Sridhar1 · M.-A. Keip1 · C. Miehe1

Received: 5 May 2015 / Accepted: 4 October 2015 / Published online: 12 May 2016
© Springer-Verlag Berlin Heidelberg 2016

Abstract Ferromagnetic materials are characterized by a Keywords Homogenization · Magneto-mechanics ·

heterogeneous micro-structure that can be altered by exter- Magnetic domains · Magnetostriction · Phase-field models ·
nal magnetic and mechanical stimuli. The understanding Variational principles · Finite element methods
and the description of the micro-structure evolution is of
particular importance for the design and the analysis of
smart materials with magneto-mechanical coupling. The 1 Introduction
macroscopic response of the material results from com-
plex magneto-mechanical interactions occurring on smaller The term ferromagnetism is historically used for materials
length scales, which are driven by magnetization reorien- that exhibit a spontaneous magnetization in the absence of an
tation and associated magnetic domain wall motions. The applied magnetic field. The occurrence of spontaneous mag-
aim of this work is to directly base the description of the netization is associated with the well-known phenomenon
macroscopic magneto-mechanical material behavior on the of magnetization hysteresis under applied magnetic loading.
micro-magnetic domain evolution. This will be realized by The overall hysteretic response of a ferromagnet is based on
the incorporation of a ferromagnetic phase-field formula- the formation and evolution of complex magnetic domain
tion into a macroscopic Boltzmann continuum by the use patterns taking place on a small (microscopic) scale, see, for
of computational homogenization. The transition conditions example, Hubert and Schäfer [24]. Due to their advantageous
between the two scales are obtained via rigorous exploitation properties, ferromagnetic materials have found many appli-
of rate-type and incremental variational principles, which cations in the industry, mainly as magnetic recording media
incorporate an extended version of the classical Hill–Mandel (Bertram [3], Greaves [20]) and as high-performance perma-
macro-homogeneity condition covering the phase field on the nent magnets (Goll and Kronmüller [19], Schrefl et al. [61]).
micro-scale. An efficient two-scale computational scenario is Generally, the occurence of spontaneous magnetization goes
developed based on an operator splitting scheme that includes along with the development of spontaneous deformation.
a predictor for the magnetization on the micro-scale. Two- This phenomenon is called magnetostriction. In order to tech-
and three-dimensional numerical simulations demonstrate nologically exploit the magnetostrictive effect, the associated
the performance of the method. They investigate micro- strains have to be significant. The highest magnetostrictive
magnetic domain evolution driven by macroscopic fields as strains that could be measured so far were obtained with rare-
well as the associated overall hysteretic response of ferro- earth alloys such as Terfenol-D and Galfenol and are in the
magnetic solids. order of a few per mille (Abboundi and Clark [1], Sandlund
et al. [60], Clark et al. [8])1 . Due to their remarkable deforma-
B C. Miehe
[email protected]; 1 There exist materials that show even higher deformations under
[email protected] applied magnetic field, given by ferromagnetic shape memory alloys
(these show strains in the order of a few percent). The governing effects
1 Institute of Applied Mechanics (CE), Chair I, taking place in shape memory alloys are however different from those
University of Stuttgart, Pfaffenwaldring 7, 70569 Stuttgart, in pure ferromagnets and will not be considered in the present contri-
Germany bution. For more information see, for example, Kiefer et al. [30].

123
152 Comput Mech (2016) 58:151–169

tions, these materials are also termed giant magnetostrictive electro-, and magneto-mechanical phenomena. Thermo-
materials. Such solids are attractive for applications in the mechanical coupling was addressed in Özdemir, Brekelmans
area of sensor and actuator technology (Claeyssen et al. [7], and Geers [54] and Temizer and Wriggers [68]. Applica-
Olabi and Grunwald [53]). tions in the area of electro-mechanics are given by Berger et
In order to allow for a firm description of ferromagnetic al. [2], Schröder [62], Kuznetsov and Fish [37], Schröder
materials, it is necessary to develop reliable and accurate and Keip [63], Ponte Castañeda and Siboni [57], Zäh and
models that reflect the complex magneto-mechanically cou- Miehe [70] and Keip et al. [28,29]. The homogenization
pled nature of the material. Here one can distinguish between of magneto-mechanical material behavior is considered
microscopic and macroscopic formulations. The develop- for example in Pisante [55], where a rigorous mathe-
ment of microscopic models is associated with the domain matical basis for the homogenization of micro-magnetic
theory of magnetization. Its fundamentals, in the context of large bodies is derived. Daniel et al. [10,11] have pre-
rigid bodies have been developed in the seminal work of sented homogenization of a simplified micro-scale model.
Landau and Lifshitz [39]. Several years later, Brown [5] James and Kinderlehrer [25], DeSimone [12], DeSimone and
used variational principles in order to derive the basis of James [13], and DeSimone et al. [14] treat the modeling of
micro-magnetics. Approaches to dynamic theories of domain magneto elastic materials with domains in the large body
evolution consider the Landau–Lifshitz–Gilbert equation limit by the use of relaxation methods. Such methods are
(LLG equation, see Gilbert [18]), which describes the attractive for the solution of the non-convex variational prob-
time-dependent evolution of magnetization. Micro-magneto- lem for the Landau-Lifshitz energy functional. In the area of
elastic theories for domain evolution based on the LLG large-strain applications, the works of Ponte Castañeda and
equation are found, for example, in Shu et al. [64], Zhang Galipeau [56] and Javili et al. [26] may be mentioned among
and Chen [71,72], Hu et al. [23], Landis [40] as well as more others.
recently in Liang et al. [41] and Yi and Xu [69]. The work At this point it should, however, be noted that a con-
Miehe and Ethiraj [47] proposed a variational formulation sistent homogenization framework that incorporates micro-
for the geometrically consistent evolution of magnetization. magnetic domain evolution and its connection to macro-
Microscopic aspects of ferromagnetic materials including scopic material response is lacking in the literature. The
the description of magnetic micro-structures and their evolu- goal of this work is to fill this gap by developing a com-
tion are addressed in Cullity [9], Kittel [32], and Spaldin putational homogenization framework for micro-magnetics,
[66]. On the other hand, macroscopic phenomenological conceptually in line with the work Zäh and Miehe [70] on
models have been developed to predict the coupled magneto- micro-electro-elasticity. This allows for the description of
mechanical behavior of ferromagnetic materials. Here the macroscopic ferromagnetic response in consideration of a
reader is directed to the works of Sablik and Jiles [59], dissipative, magneto-mechanically coupled micro-structure.
Hirsinger et al. [22], Smith et al. [65], Kankanala and The micro-structure will be described as a micro-magneto-
Triantafyllidis [27], Linnemann et al. [42], and Miehe et elastic solid, where the magnetization is described by a
al. [49,50] among others. However, such approaches do not phase-field, see Miehe and Ethiraj [47] and Ethiraj et al. [16]
allow for the investigation of microstructural material prop- for details. In order to incorporate the micro-magneto-
erties and their optimization, since the effect of a change in elastic character of the micro-structure into the macroscopic
the micro-structure is not accounted for. model, a homogenization technique that accommodates the
In this context robust computational homogenization mechanical displacement, the magnetic potential as well
techniques for complex microscopic response of coupled as the phase field of magnetization as primary variables
materials with magneto-mechanical coupling are called for. will be proposed. The formulation is based on rate-type
The basic idea of such scale-bridging techniques is to attach and incremental variational principles conceptually in line
a micro-structure point-wise to a macro-continuum, hence with [70].
defining the macroscopic response of the material based The idea is to perform the scale transition between a
on micro-structural information. An overview of purely microscopic phase-field model and a standard Boltzmann
mechanical homogenization theories is given in Nemat- continuum on the macro-scale as depicted in Fig. 1. This
Nasser and Hori [52], for further developments see also the means that the displacement and magnetic potential are con-
works on variational-based homogenization Miehe [43,44], sidered to be the only external variables on the macro-level.
Brun et al. [6], Lahellec and Suquet [38] and Brassart et In contrast, the order parameter given by the magnetization
al. [4]. is assumed to have no external counterpart on the macro-
In recent years, extensions of homogenization to cou- scale. The proposed variational principle of computational
pled material response have been explored. A wealth of homogenization in micro-magneto-mechanics results in a
research has contributed to the understanding of coupled con- Hill-Mandel type micro-macro compatibility condition of the
stitutive behavior across multiple scales, including thermo-, form

123
Comput Mech (2016) 58:151–169 153

{ε, h}

m
lmicro N S lmacro

{σ, b}
Micro-structure B Macro-structure B

Fig. 1 Strain- and magnetic-field-driven scale bridging in micro- are the homogenized stresses and magnetic induction {σ , b} at a mater-
magneto-mechanics. The macroscopic strains and magnetic fields {ε, h} ial point of the macro-structure. The average
 of the micro-magnetization
defined on the homogenized macro-structure B drive a poly-crystalline gives the macro-magnetization m := |B1 | B m d V of the homogenized
micro-structure B,inducing domain wall motions in the grains. Output continuum

σ : ε̇ − b : h˙ incremental variational principles of homogenization are

   proposed in Sect. 3 which determine the macroscopic poten-
macro-power tial. An Euler equation of these principles is the LLG
extended micro-power
   equation, that we recast into an algorithmic predictor-

corrector scheme for the time-dependent solution of the
= |B |
1
{σ : ε̇ − b · ḣ + h · ṁ + k : ∇ ṁ} d V.
B micro-magnetic evolution problem. Sect. 4 is devoted to a
brief formulation of the phase-field models employed for the
description of micro-structure evolution. Finally, two- and
This extends formulations of Hill [21] and Suquet [67] for
three-dimensional numerical simulations are considered in
mechanical problems and a proposal by Schröder [62] for
Sect. 5. They analyze the typical domain evolution processes
local electro mechanical problems. In the above condition,
in ferromagnetic micro-structures as well as the associated
it is crucial to note that the macroscopic power of external
macroscopic hysteretic response.
variables is related to the volume average of their micro-
scopic counterparts extended by contributions of the phase
field of magnetization. This approach is motivated by the
recently proposed homogenization of phase-field models in
ferroelectrics by Zäh and Miehe [70] and conceptually in 2 Variational formulation of the macroscopic
line with previous works on the homogenization of particle response
aggregates Miehe and Dettmar [46] and Miehe et al. [48].
In the former it was assumed that the microscopic electric Let B ⊂ V denote a macroscopic body embedded into a free-
polarization has no external counterparts on the macro-level, space box V ⊂ Rd with dimension d ∈ [2, 3] as depicted
similar to a micro-particle rotation in granular aggregates in Fig. 2. We study mechanical deformations coupled with
linked to Boltzmann continua. It should be remarked that the magnetic induction of the body under quasi-static, magneto-
above scenario is opposite to the works Kouznetsova et al. mechanical external loading in the time interval T ⊂ R+ .
[33–35], in which local micro-structures are bridged towards ˙ := ∂t (·) denote the
In what follows, ∇(·) := ∂ x (·) and (·)
gradient-extended macro-continua. gradient and the time derivative of the macroscopic field (·),
The paper is organized as follows. In Sect. 2, the varia- respectively.
tional principles for the macroscopic magneto-mechanically
coupled problem are established. First, a rate-type varia-
tional principle for quasi-static conditions in terms of a
macroscopic potential is constructed. Based on that, an 2.1 Introduction of macroscopic fields
associated algorithmic incremental variational principle is
derived. From this directly follow the governing Euler equa- The macroscopic boundary-value-problem of magneto-
tions, which are given by the balance of momentum and mechanics is a coupled multi-field problem. Primary vari-
Gauß’s law of magnetostatics. Next, rate-type as well as ables are the displacement field u and the magnetic potential

123
154 Comput Mech (2016) 58:151–169

Fig. 2 Macrocopic
magneto-mechanical V tN = σ ·n V
boundary-value-problem.
a Mechanical displacement field
u defined on the solid B ⊂ V B n B
and b magnetic potential φ
defined on the full free-space x∈B x∈V
box V ⊂ Rd . The boundary of
the solid decomposes into u φ
Dirichlet and Neumann parts
∂ B = ∂ B u ∪ ∂ B t . The boundary
∂ V of the free space box is a
u = uD φ = φD
pure Dirichlet boundary that
allows the definition of applied
magnetic fields
(a) (b)

φ visualized in Fig. 2 of the material point x ∈ V at time the dissipative reorientation of magnetic domains. Admissi-
t ∈ T , i.e. ble evolution of the displacement and the magnetic potential
  fields are defined by the Dirichlet conditions
B × T → Rd V ×T →R
u: and φ : (1) W u̇ := {u̇ ∈ Rd | u̇ = u̇ D on ∂B u } and
(x, t) → u(x, t) (x, t) → φ(x, t).
W φ := {φ̇ ∈ R | φ̇ = φ̇ D on ∂V}. (5)
The gradients of the fields u and φ define the macroscopic
displacement gradient and the magnetic field The Dirichlet condition for the magnetic potential on the
surface ∂V of a suitable free space box V, allows for the def-
f := ∇u and h := −∇ φ. (2) inition of far field magnetic loading, see Fig. 2. The external
mechanical loading contribution is
These definitions satisfy automatically the deformation com-  
patibility curl[ f ] = 0 in B and Faraday’s law of induction, P mech (u̇) = γ · u̇ d V + t N · u̇ d A, (6)
the third Maxwell equation, curl[h] = 0 in B for the quasi- B ∂Bt
static problem under consideration. An objective constitutive
state depends on the symmetric part of the displacement gra- where γ is the macroscopic body force per unit volume and
dient t N are the prescribed tractions. The rate-type variational prin-
ciple (4) gives the necessary condition,
ε = ∇s u = 1
∇u + ∇ u .
T
(3) 
2
− (div[∂ε̇ π ] + γ ) · δ u̇ d V
B
The strains are assumed to be small, i.e. |∇u| <  is bounded
+ (∂ε̇ π · n − t N ) · δ u̇ d A (7)
by a small number . In contrast, the norm of the gradient of ∂ B t
the magnetic potential |∇ φ| is not bounded. − div[−∂h˙ π ]δ φ̇ d V = 0,
V
2.2 Macroscopic rate-type variational principle
where n is the outward normal depicted in Fig. 2. This results
The evolution of the macroscopic displacement and the mag- in the Euler–Lagrange equations on the macro-scale
netic potential can be defined by the rate-type variational
div[∂ε̇ π ] + γ = 0 in B and div[−∂h˙ π ] = 0 in V, (8)
principle
 along with the mechanical Neumann-type boundary con-
{u̇, φ̇} = arg ˙ dV −P
π (ε̇, h)
inf sup mech ( u̇) (4) ditions ∂ε̇ π · n = t N on ∂B t . Hence, we identify the
u̇∈W u̇ φ̇∈W φ̇ V
macroscopic mechanical stresses and the magnetic induction

The principle is based on the macroscopic potential density ˙ and b = −∂ π(ε̇, h),
σ = ∂ε̇ π (ε̇, h) ˙ (9)
h˙
π , which is obtained by homogenization of the microscopic
potential density considered in the subsequent Sect. 3. As governed by the macroscopic potential π obtained by the
a consequence, the macroscopic problem depends on the homogenization of the microstructure as commented on in
microscopic magnetization introduced later, which describes Sect. 3.

123
Comput Mech (2016) 58:151–169 155
 
2.3 Space-time discrete incremental variational r := B T [∂cπ τ ] d V and k := 2 τ
B T [∂ccπ ]B d V
principle V V
(12)
For a compact representation of the spatial discretization by
a finite element method we assemble the set of primary state until convergence |r| < tol. The macro-array s := ∂cπ τ
h
variables in the array c := {∇ s u, −∇ φ}. Let T denote a contains the current homogenized stresses σ and the homog-
finite element triangulation of the full free space box V with enized field b as introduced in (9). They are defined by
h h the volume averages defined in (48) below. The macro-
N global nodes. Associated with the triangulation T , we 2 π τ for the coupled problem
scopic tangent moduli C := ∂cc
write the finite element interpolations of the primary state
are specified in (49) below. The above procedure provides
vector in the compact form
a variational principle based FEM2 scenario in micro-
magneto-mechanics, in line with that outlined in Miehe [43]
N̄ h
ch = B(x)d with d := A I¯=1 [u I¯ , −φ I¯ ]T (10) for dissipative problems in mechanics. Note that the sym-
metry of the tangent matrix k is induced by the proposed
in terms of the macroscopic space-time-discrete nodal vari- multi-scale variational principles.
able vector d, which contains the displacement and magnetic
potential at a typical nodal point of the finite element mesh. 3 Variational homogenization of
Then, for pure Dirichlet problems with P mech = 0, the finite- micro-magneto-elasticity
step-sized discrete stationary principle
Let B ⊂ Rd denote a periodic microstructure with dimen-
 sion d ∈ [2, 3] as depicted in Fig. 3. In what follows,
d = arg stat π τ (B d) d V (11) ˙ := ∂t (·) denote the gradient and the
∇(·) := ∂ x (·) and (·)
d V
time derivative of the microscopic field (·), respectively.
determines the macroscopic nodal values at current time tn+1 . 3.1 Introduction of microscopic fields
Note that the principle is fully governed by the space-time-
discrete macro-potential π τ . The problem may be solved by The material is modeled by using a gradient-extended dis-
a Newton method with updates d ⇐ d − k−1 r based on the sipative continuum formulation, where the magnetization is
residual and tangent arrays described by an order parameter or generalized internal vari-

+ +
∂B

x∈B x∈B +
− + − u
[[x]] u+ = u− + ε[[x]]
− x
x +
u−
− −
(a) (b)

+ +

x∈B + x∈B +
− m − φ
m+ = m− φ+ = φ− − h[[x]]
m− φ−
− −
(c) (d)
Fig. 3 Deformation of a periodic micro-structure B. a Points on the magnetization m+ = m− and d the magnetic potential φ + = φ − −
opposite faces ’+’ and ’−’ with x + ∈ ∂ B+ and x − ∈ ∂ B− . Periodic h x , driven by the macroscopic strain ε and magnetic field h
boundary conditions for b the displacement u+ = u− + ε  x , c the

123
156 Comput Mech (2016) 58:151–169

able field, see Miehe [45] in mechanics, Miehe et al. [51], the generalized internal variable for the material magnetiza-
Zäh and Miehe [70] in micro-electro-mechanics and Miehe tion, which is assumed to be dissipative in nature. In a time
and Ethiraj [47] and Ethiraj et al. [16] in micro-magneto- continuous setting, the macroscopic potential density π is
mechanics. We consider a three field formulation in terms of defined by
the displacement, the magnetic potential and magnetization
unit director ˙ =
 π (ε̇, h) 
B × T → Rd inf sup inf 1
π(ε̇, ḣ, ṁ, ∇ ṁ) d V (18)
u:  |B |
(x, t) → u(x, t) ˙ ṁ∈W B
B × T → S d−1 u̇∈Wu̇ (ε̇) φ̇∈Wφ̇ ( h) ṁ
 m: (13)
B×T →R (x, t) → m(x, t).
φ:
(x, t) → φ(x, t) The principle characterizes an optimization of a volume aver-
age of a microscopic rate-type potential density π , that takes
The gradients of the fields u and φ define the displacement into account the energy storage and the dissipative mecha-
gradient and the magnetic field nisms of the micro-material

f := ∇u and h := −∇φ. (14) d

π(ε̇, ḣ, ṁ, ∇ ṁ) = Ψ (ε, h, m, ∇m) + Φ(ṁ) .
   dt
      (19)
These definitions satisfy automatically the deformation com- potential density rate of energy function dissipation potential

patibility curl[ f ] = 0 in B and Faraday’s law of induction,

the third Maxwell equation, curl[h] = 0 in B for the quasi- The potential density is defined by the rate of the mixed
static problem under consideration. An a priori objective energy-enthalpy function Ψ and the dissipation function Φ,
constitutive state depends on the symmetric part of the dis- which we discuss in detail in Sect. 4. The above homogeniza-
placement gradient tion problem for periodic micro structures is driven by the
macroscopic strains ε and the macroscopic magnetic field h.
ε = ∇s u = 21 [ f + f T ]. (15) They define the admissible fields on the micro-structure

The strains are assumed to be small, i.e. |∇u| <  is bounded

by a small number . In contrast, the norm of the gradients Wu̇ := {u̇ ∈ Rd | u̇ = ε̇x on ∂B} and
˙ (20)
of the magnetization vector |∇m| and the magnetic potential W := {φ̇ ∈ R | φ̇ = − hx
φ̇ on ∂B}.
|∇φ| are not bounded. The magnetization director m, which
is used as a primary variable, defines the magnetization as
Here, (·) := (·)+ − (·)− is the jump of the quantity (·) with
M = m s m, (16) respect to opposite faces of the periodic micro-structure as
visualized in Fig. 3. In order to keep the notation compact,
where the material constant m s is the spontaneous magne- we focus in this work exclusively on periodic boundary con-
tization of the solid under consideration. The magnetization ditions. Alternative boundary conditions may be formulated
director is constrained by the condition and implemented in full analogy to the treatment in Zäh and
Miehe [70] on micro-electro-mechanics. The magnetization
|m| = 1, (17) is a generalized internal variable that does not have a macro-
scopic external counterpart. For the periodic micro-structure
which restricts the manifold of the order parameter m to the under consideration, the admissible rates of the magnetiza-
unit sphere S d−1 in the space Rd . This geometric structure of tion fields are
the order parameter m is a key property in the field of micro-
magnetics. This property has to be considered consistently Wṁ := {ṁ ∈ Tm S d−1 | ṁ = 0 on ∂B}. (21)
in the subsequent theoretical and computational setting.

Note carefully, that the variational principle (18) needs to

3.2 Microscopic rate-type variational principle account for the geometric structure of the magnetization of
of homogenization the form m ∈ S d−1 such that ṁ ∈ Tm S d−1 , where Tm S d−1
is the tangent space of S d−1 at m as depicted in Fig. 4. Taking
We now postulate a rate-type variational principle of homog- into account the definitions (9), we derive by a variation of
enization that determines the macroscopic potential density the principle (18) at the solution point, a generalized Hill–
π introduced above. This definition includes the kinetics of Mandel condition

123
Comput Mech (2016) 58:151–169 157

m = mn + Δm

Tm S 2 with the variational derivative δm Ψ := ∂m Ψ − div ∂∇m Ψ .
Δm
Δm Hence, the Euler–Lagrange equations for the variational prin-
S2 S2
m ciple (18) are,
|m| m

a1
m div[∂ε Ψ ] = 0 in B
a2
m × [δm Ψ + ∂ṁ Φ] = 0 in B (24)
div[−∂h Ψ ] = 0 in B

on the microstructure B. Equation (24)2 governs the evolu-

(a) (b) tion of the magnetization which satisfies the constraint on
the magnetization director a priori. Defining the constitu-
Fig. 4 Illustration of the update methods for the magnetization direc- tive dissipation potential as Φ(ṁ) = η2 |ṁ|2 and taking the
tor. a The exponential update in Miehe and Ethiraj [47] calculates an
increment m that lies in the tangent space defined by a1 and a2 . b cross product of (24)2 with the vector m, we end up with the
The projection method in the current work provides a substantially sim- evolution equation for the magnetization director
plified alternative to maintain the unity constraint on the magnetization
director by employing a post-processing normalization step 1  
ṁ = m × m × δm Ψ in B. (25)
η

1
|B | {σ : δ ε̇ − b · δ ḣ + h · δ ṁ + k : ∇δ ṁ} d V This is consistent with the Landau–Lifshitz–Gilbert equa-
B
   tion, see Landau and Lifshitz [39]. It contains the damping
extended micro power
term of the Landau–Lifshitz–Gilbert equation for non-
= σ : δ ε̇ − b : δ h˙ stationary micro-magnetics, see Gilbert [17]. Note that this
   evolution equation satisfies the constraint ṁ · m = 0.
macro power

3.3 Microscopic incremental variational principle

for the dissipative magneto-mechanical problem under con-
of homogenization
sideration. At the micro-level, we identify the stresses, the
magnetic induction and the conjugate forces dual to the mag-
In the numerical setting, we consider an incremental vari-
netization vector and its gradient
ational principle which governs the magneto-mechanical
problem at discrete time tn+1 for finite time increments
σ := ∂ε Ψ, b := −∂h Ψ, τ = tn+1 − tn . We propose the incremental variational prin-
h := ∂m Ψ + ∂ṁ Φ, k := ∂∇m Ψ, ciple of homogenization

governed by the constitutive functions Ψ and Φ introduced

in (19). π τ (ε, h) = 
In order to account for the geometric structure m ∈ S d−1 , inf sup inf 1
|B | π τ (ε, h, m, ∇m) d V
we represent the variation of the rate of the magnetization u∈Wu (ε) φ∈Wφ (h) m∈Wm B
director by (26)

δ ṁ = δω × m, (22) It defines the macroscopic incremental potential density

π τ by optimizing the volume average of the microscopic
where ω ∈ Rd is the spin of magnetization. This satisfies potential π τ . This definition is conceptually in line with for-
the condition ṁ ∈ Tm S d−1 . With this definition at hand, the mulations for non-linear mechanical problems in Miehe [43],
variation of the principle (18) gives the condition Ponte Castañeda and Suquet [58]. The incremental potential
density π τ is related to the rate-type potential π by the algo-
 
1 1 rithm
(−div[∂ε Ψ ]) · δ u̇ d V + (∂ε Ψ · n) · δ u̇ d A
|B| B |B| ∂ B 
  t
1 1
− div[−∂h Ψ ] δ φ̇ d V + (−∂h Ψ · n)δ φ̇ d A π τ (ε, h, m, ∇m) := ALGO π(ε̇, ḣ, ṁ, ∇ ṁ) dt .
|B| B |B| ∂ B tn

1 (27)
+ m × (δm Ψ + ∂ṁ Φ) · δω d V
|B| B

1 The admissible spaces for the displacement and magnetic
+ m × (∂∇m Ψ · n) · δω d A = 0, (23) potential are
|B| ∂ B

123
158 Comput Mech (2016) 58:151–169

Wu := {u ∈ Rd | u = εx on ∂B} and whose Euler equation gives the fully implicit update equation
(28)
Wφ := {φ ∈ R | φ = −hx on ∂B}, for the magnetization

driven by the macroscopic strain ε and the magnetic field h. ∂m Ψ − div[∂∇m Ψ ] + τ ∂m Φ((m − mn )/τ ) = 0 in B. (35)
The admissible magnetization fields are
3.4.2 Magnetic predictor ALG Om2
Wm := {m ∈ R | m = 0 on ∂B}.
d
(29)
Clearly, (35) does not obey the constraint m ∈ S d−1 . We
Recall that we need to account for the geometric structure satisfy this by a post-processing step that projects the mag-
such that m ∈ S d−1 . netizations to the unit sphere S d−1 i.e.,

3.4 A staggered solution scheme of the microscopic m ⇐ m/|m|. (36)

problem
This is an easy-to-implement algebraic scaling conceptually
To enforce the above constraint on the magnetization direc- in line with proposals reviewed by Kružík and Prohl [36].
tor m, we propose a staggered scheme such that the overall
solution algorithm is composed of two sub-algorithms 3.4.3 Magneto-mechanical corrector ALG Ouφ

ALG O = ALG Ouφ ◦ ALG Om . (30) We proceed in a similar manner as before by constructing an
incremental potential for the corrector step by setting
Here, ALG Om is a predictor step that updates the magnetiza-  t
tion director, and is followed by the corrector step ALG Ouφ τ
πuφ (ε, h) = ALGOuφ π(ε̇, ḣ, ṁ = 0, ∇ ṁ = 0) dt ,
that determines the deformation and the magnetic potential. tn
The predictor ALG Om is further decomposed according to (37)

ALG Om = ALG Om2 ◦ ALG Om1 (31) yielding the closed-form of a potential density

τ
into two sub-steps, which ensures the geometric constraint on πuφ (ε, h) = Ψ (ε, h, m, ∇m) − Ψn . (38)
m. Here, ALG Om1 is unconstrained, and ALG Om2 enforces
the unity constraint on the magnetization. The three parts of Note that m and ∇m are given by the predictor step in (36). As
the solution procedure are described as follows. a consequence, the magneto-mechanical corrector algorithm
for the update of the displacements and potential at frozen
3.4.1 Magnetic predictor ALG Om1 magnetization is governed by the variational statement

1
A time-discrete incremental potential for the first sub- π τ (ε, h) = inf sup τ
πuφ (ε, h) d V. (39)
problem is obtained by arresting the rates of the displace- u∈Wu (ε) φ∈Wφ (h)
|B| B
ments u̇ = 0 and magnetic potential φ̇ = 0 in the original
potential. We define the predictor potential Consequently we observe that the necessary condition for
the stationarity of this equation yields the mechanical equi-
πmτ (m, ∇m) librium and the Gauss law as the Euler equations,
t (32)
= ALGO1m tn π(ε̇ = 0, ḣ = 0, ṁ, ∇ ṁ) dt 
div[∂ε Ψ ] = 0
in B. (40)
div[−∇φ + m s m] = 0
that we approximate by the closed-form algorithmic repre-
sentation
These Euler equations are the time-discrete, algorithmic
πmτ (m, ∇m) counterparts of the corresponding Euler equations of the
(33) rate-type, time-continuous variational principle.
= Ψ (ε n , hn , m, ∇m) − Ψn + τ Φ((m − mn )/τ ).

With this potential at hand, the magnetic predictor algorithm 3.5 Space-time discretization of staggered solution
is represented by the variational statement scheme

1 The space discretization covers the two algorithmic steps
{m} = arg inf π τ (m, ∇m) d V , (34)
m∈Wm |B| B m ALG Om and ALG Ouφ in the above staggered scheme. For

123
Comput Mech (2016) 58:151–169 159

both sub-problems, the same finite element triangulation Th 1
π̄ τ (ε, h) = stat π τ (B uφ d uφ ) d V (46)
of the micro-structure B with N h nodes is used. d uφ |B| B uφ

determines the macroscopic potential at current time tn+1 .

3.5.1 Magnetic predictor ALG Om1
The problem is solved by a Newton method with updates
d uφ ⇐ d uφ − k−1
uφ r uφ based on the residual and tangent
For the discretization of the problem (34), consider the set
arrays
of constitutive state variables assembled in cm := {m, ∇m}.
We then write the finite element interpolations in the form 
τ
r uφ := |B1 | T
B uφ [∂cuφ πuφ ] d V and
 B (47)
Nh
cm
h
= B m (x)d m with d m := A I =1 [m I ]T (41) kuφ := T
B uφ τ
[∂c2uφ cuφ πuφ ]B uφ d V
B
with the nodal values d m of the magnetization directors. until convergence |r uφ | < toluφ .
Then, the finite-step-sized discrete stationary principle
 3.6 Algorithmic representation of the homogenized
1
d m = arg inf πmτ (B m d m ) d V (42) quantities
d m |B| B

The homogenized stresses and magnetic field s := {σ , −b}

determines the unconstrained values for the nodal magneti- are obtained by taking the variation of the functional π̄ τ
zation directors at current time tn+1 . The problem is solved defined in (46) with respect to the macroscopic driving quan-
by a Newton method with updates d m ⇐ d m − k−1 m r m based tities c := {ε, h}. This gives the expression
on the residual and tangent arrays

 1
s := ∂cπ̄ τ = [∂c π τ ] d V, (48)
r m := |B1 | T
Bm [∂cm πmτ ] d V and |B| B uφ uφ
B
 (43) that characterizes a volume average of the associated micro-
km := |B|1
[∂cm cm πmτ ]B m
T 2
Bm dV scopic quantities s := ∂cπ τ . The tangent moduli needed for
B a Newton solver of the FEM2 scenario related to the macro-
scopic finite element arrays (12) are obtained by a further
until convergence |r m | < tolm .
derivation

3.5.2 Magnetic predictor ALG Om2 2 τ 1 T −1
C := ∂ccπ̄ = [∂ 2 π τ ] d V − l uφ kuφ l uφ . (49)
|B| B cuφ cuφ uφ
For the converged state, ALG Om2 is performed according to
(36), yielding the constrained values with the tangent matrix kuφ defined in (47) and the additional
finite element array
Nh 
d m := A I =1 [m I /|m I |]T (44) 1
l uφ := B T [∂ 2 π τ ] d V. (50)
|B| B uφ cuφ cuφ uφ
of the magnetization directors at the nodal point of the finite
element mesh. Note that the macroscopic moduli contains the average of the
microscopic moduli C := ∂c2uφ cuφ πuφ τ and a typical softening

term due to the internal flexibility of the micro-structure. This

3.5.3 Magneto-mechanical corrector ALG Ouφ
is in full analogy to the variational-principle-based FEM2
scenario in dissipative mechanics outlined in Miehe [43].
For the discretization of the problem (39), consider the set of
state variables assembled in cuφ := {∇s u, −∇φ}. We then
write the finite element interpolations in the form
4 Basic functionals in micro-magneto-elasticity
Nh
cuφ
h
= B uφ (x)d uφ with d uφ := A I =1 [u I , −φ I ]T (45) 4.1 Energy density and dissipation potential functionals

with the nodal values d uφ containing the displacement and In order to describe the energy storage in micro-magnetics in
the magnetic potential. Then, the finite-step-sized discrete terms of the constitutive energy-enthalpy described in (19),
stationary principle we first introduce a constitutive free energy function and

123
160 Comput Mech (2016) 58:151–169

transform it later by a Legendre transformation to a mixed Ψa

energy-enthalpy function, suitable for our choice of the
primary variables. This is in line with our model for micro-
magnetics previously proposed in Miehe and Ethiraj [47].
−a
4.1.1 Contributions to the energy density m
my
S1 +a
We need to describe two contributions to the energy den-
sity: a contribution Ξvac due to the presence of a magnetic
mx
field in the free space occupied by the body B and an addi-
tional contribution Ξmat due to the elastic deformation and Fig. 5 Simplified two-dimensional anisotropy energy density (two-
the magnetization of the solid. Hence, the total free-energy well potential) characterized by a unit vector a for the uniaxial case.
density of the solid material embedded into the free space The magnetization director m evolves on the unit circle S 1
decomposes according to,

Ξ (ε, m, ∇m, b) = Ξmat (ε, m, ∇m) + Ξvac (m, b) (51)

into contributions from the material and the free space. The
material part of the free energy is a function of the strain
ε, the magnetization director m and its gradient ∇m. It is
assumed to consist of the three contributions

Ξmat (ε, m, ∇m) = Ξe (ε, m) + Ξa (m) + Ξex (∇m), (52) (a) (b)

where Ξe is the elastic energy, Ξa the anisotropy energy and Fig. 6 The cubic magneto-crystalline anisotropy energy density land-
Ξex the exchange energy. scape is shown in the case of a 100 as easy direction for the case of
Galfenol and b 111 easy direction for Terfenol-D

4.1.2 Anisotropy energy density

For cubic magneto-crystalline anisotropy the energy land-
In a ferromagnetic material with an inherent crystalline struc- scape is shown in Fig. 6. The easy axes lie along the 100
ture, the energy depends on the direction of magnetization or 111 directions of the crystal. Then this energy can be
relative to the structural axes of the material. In a crystalline defined as,
solid, there exist preferred directions of magnetization, often
denoted as the easy directions. They are modeled by the even Ξa (m) = K [(m ⊗ m : A12 )2 + (m ⊗ m : A13 )2
anisotropy energy density Ξa : S d−1 → [−∞, +∞]. The +(m ⊗ m : A23 )2 ] (54)
minimum of Ξa defines the easy directions, i.e. the directions
along which the material is magnetized most easily. Specific where the material parameter K is the anisotropy constant.
representations of this non-convex energy contribution Ξa It assumes values in the range of ±104 J/m 3 for different
for different crystals are reported in Cullity [9], Kittel [31,32] materials, the sign of this constant determine if 100 or 111
and Hubert and Schäfer [24]. We consider here uniaxial and are the easy directions of magnetization. Here Ai j = ai ⊗
cubic anisotropies. a j i, j = 1..3 are the direction tensors in terms of the three
Hexagonal and tetragonal crystals show a uniaxial perpendicular cubic axes a1 , a2 and a3 of the crystal.
magneto-crystalline anisotropy, the anisotropy energy land-
scape for such materials is shown in Fig. 5. The energy 4.1.3 Elastic energy density
function is the well known two well potential with the min-
ima in the direction of the easy axes. This energy can be The elastic energy Ξe : Sym(d) × S d−1 → R+ describes
written as the energy storage due to elastic lattice distortions. Asso-
K1 K2 ciated with the magnetization director m, there exists a
Ξa (m) = {1 − tr2 [m ⊗ a]} + {1 − tr2 [m ⊗ a]}2 , (53) preferred local distortion of the crystalline solid. In line with
2 4
DeSimone and James [13] and Engdahl [15], Hubert and
where the vector a defines the direction of the easy axis. This Schäfer [24], Zhang and Chen [71] we introduce an even
free energy function can define uniaxial, planar and conical function m → ε 0 (m) ∈ Sym+ 3 , where ε (m) is the stress-
0
magnetic anisotropies, refer to [24] for more details. free strain corresponding to the magnetization director m.

123
Comput Mech (2016) 58:151–169 161

For the case of uniaxial materials, we assume an isotropic 4.1.4 Exchange energy density
function of the form,
The exchange energy Ξex : Lin(d) → [0, +∞] models the
3 tendency of neighboring magnetic atoms to align. Focusing
ε 0 (m) = E P : [m ⊗ m]. (55)
2 on an isotropic response, we consider the standard expression

where P is the fourth order projection tensor defined as A 

Pi jkl = Ii jkl − 1/3δi j δkl and E is the mangetostrictive con- Ξex (∇m) = tr ∇m · ∇ T m , (59)
2
stant. With this ansatz at hand, we consider the elastic energy
density to be a function of the elastic, stress-producing strains where the material parameter A > 0 is the exchange energy
ε e := ε − ε 0 . A quadratic form reads coefficient. It is a material length scale related to the width
of the magnetic domain walls.
Ξe (ε, m) = 21 [ε − ε 0 (m)] : Ciso : [ε − ε 0 (m)], (56)
4.1.5 Energy density of the free space
where Ciso is the positive definite fourth-order elasticity ten-
sor of the crystal. This elasticity tensor is assumed to be The magnetization m s m defines, together with the total
isotropic with two constants λ and μ, for materials with uni- magnetic field h, the magnetic induction induced by the mag-
axial magneto-crystalline anisotropy. For the case of giant netization
magnetostrictive materials with cubic anisotropy, (55) and
(56) are defined in the lattice frame of the crystal. The strain b = κ0 ( h + m̂ s m ) in B, (60)
corresponding to the magnetization in this case can be given
as, The energy Ξvac : Rd × Rd → [0, +∞] describes the
energy due to the presence of a magnetic field in the free
E 100
ε 0 (m) = 23 [K : (m ⊗ m) − 3 1]|C F . (57) space occupied by the body. Resolving (60) for the magnetic
field h = ĥ(m, b), we define the quadratic form
Here K is a material modulus with two constants E 111 and 
κ0
E 100 , which define the material response in the 100 and Ξvac (m, b) = 2 tr h⊗h
111 directions. Then the expression for the free energy is  (61)
= 1
2κ0 tr (b − κ0 m̂ s m) ⊗ (b − κ0 m̂ s m)
given as
governed by the magnetic permeability κ0 > 0 of the vac-
Ξe (ε, m) = 21 [ε − ε 0 (m)] : Ccub : [ε − ε0 (m)]|C F . (58)
uum.

Here the elasticity tensor is cubic in nature and is defined in

4.2 Energy-enthalpy function of
terms of three constants C11 , C12 and C44 . The local axes
micro-magneto-elasticity
are aligned with the local magnetic easy directions a1 , a2
and a3 .2
With the above definition of the free energy function at hand,
2
we obtain the mixed energy-enthalpy function by the partial
The magnetostrictive strain for a cubic response in (57) can be explic-
Legendre transformation
itly written as,
⎛ ⎞ 
E (m 2 − 1/3)
3 ⎝ 100 1
E 111 m 1 m 2 E 111 m 1 m 3 Ψ (ε, m, ∇m, h) = inf Ξ (ε, m, ∇m, b) − h · b (62)
ε0 = E 111 m 2 m 1 E 100 (m 22 − 1/3) E 111 m 2 m 3 ⎠ b
2
E 111 m 3 m 1 E 111 m 3 m 2 E 100 (m 23 − 1/3) C F
with respect to the magnetic slot. This transformation affects
The material modulus K can be filled with the entries only the free space part Ξvac of the free energy. The necessary
K1111 =E 100 , K2222 =E 100 , K3333 =E 100 , K1212 =K1221 =E 111 /2, K2112 condition of (62) gives (60). Using this as an elimination
=K2121 =E 111 /2, K1313 = K1331 = E 111 /2, K3113 = K3131 =
E 111 /2, K2323 = K2332 = E 111 /2, K3223 = K3232 = E 111 /2. The
equation for b, we get from (62) the energy-enthalpy function
fourth order cubic mechanical modulus Ccub in (58) is given as
κ0
⎛
C11 C12 C12
⎞ Ψ (ε, m, ∇m, h) = Ξmat (ε, m, ∇m)− h · h −κ0 m s m · h,
2
⎜C12 C11 C12 ⎟ (63)
⎜ ⎟
⎜C12 C12 C11 ⎟
Ccub = ⎜ ⎟.
⎜ C44 ⎟
⎝ C44 ⎠ which includes the free energy density Ξmat of the solid
C44 material defined in (52).

123
162 Comput Mech (2016) 58:151–169

4.3 Dissipation potential for micro-magnetic evolution loading and the associated overall response of the material
(Fig. 8).
For a time-dependent, dynamic modeling of the evolution of
the magnetization director m, we need to define a kinetic law. 5.1 Homogenization of a galfenol single-crystal
This is achieved by the definition of a constitutive dissipation
potential. We assume the simple quadratic structure First we consider a single crystal of Galfenol Fe81.3 Ga18.7 ,
η below the Curie-Weiss transition temperature. Galfenol
Φ(ṁ) = |ṁ|2 , (64) is a giant magnetostrictive material with cubic magneto-
2
crystalline anisotropy and its magnetic easy axes aligned
where the scalar parameter η is an inverse mobility coeffi- along the 100 direction. This allows us to perform a two-
cient, which governs the kinetics of the magnetic domain dimensional simulation for a single crystal, with the easy axes
wall evolution. oriented in the plane of the sample. Here the magnetic easy
axes are oriented at 45 degrees from the global axes as shown
in Fig. 7a. The material parameters used for this simulation
5 Representative numerical examples are taken from Zhang and Chen [71] and shown in Table 1.
The exchange coefficient A is calculated from the relation
The capability of the computational framework of homog-
enization in micro-magnetics is demonstrated by means
of two- and three-dimensional benchmarks. The examples Table 1 Material parameters for Galfenol Fe81.3 Ga18.7 [71]
describe domain wall movements in ferromagnetic materials Parameter Name Unit Value
for macroscopic magnetic-field driven scenarios. The loading
C11 Mechanical parameter N/m2 1.96 × 1011
processes are performed on a converged equilibrium state.
This equilibrium state is obtained starting from a random C12 Mechanical parameter N/m2 1.56 × 1011
distribution of the magnetization in the micro-structure. The C44 Mechanical parameter N/m2 1.23 × 1011
homogenization of such micro-magnetic simulation allows E 100 Magnetostrictive coefficient − 2.64 × 10−4
for the determination of macroscopic hysteresis curves. It has E 111 Magnetostrictive coefficient −
been shown experimentally that the interplay between the A Exchange energy coefficient N 6 × 10−11
magneto-static and the exchange interaction between neigh- κ0 Magnetic permeability N/A2 1.3 × 10−6
boring grains influence these properties [19,61], in addition K Anisotropy energy coefficient N/m2 1 × 105
to having a dependence on the grain size. We will use the ms Saturation magnetization A/m 1.43 × 104
homogenization framework presented above for the analy- η Damping coefficient Ns2 /m2 2 × 10−1
sis of microscopic domain evolution driven by macroscopic

Fig. 7 a Representative volume

element of a single-crystal with
the easy axes oriented at 45◦ as a2 a1
shown. The edge length is taken
as l = 240 nm. b–f
micro-magnetic domain
evolution starting from a
random distribution of the
magnetization. The
microstructure evolves and (a) (b) (c)
finally converges to a stable
state. Vectors represent the
magnetization and the general
poling direction is indicated by
white arrows. The contour
represents the gradient of the
magnetization

(d) (e) (f)

0.0 ||∇m|| 6e11

123
Comput Mech (2016) 58:151–169 163

√
ω = A/K , where ω is the domain wall width. For known ation process causes the magnetization to align in certain
values of the K = 1 × 105 N/m2 and ω approximately 20 nm, regions and domain walls start to appear. Poling directions
we take the exchange coefficient as A = 6× 10−11 N. The in a domain are indicated with white arrows. The magneto-
boundary value problem is shown in Fig. 7a. Periodic bound- crystalline anisotropy is clearly visible in the equilibrium
ary conditions are applied as shown on all the edges for all state, with the magnetization pointing in the magnetic easy
degrees of freedom. The mechanical displacement and the directions. It should be pointed out here that the equilibrium
magnetic potential at the corner nodes are fixed to constrain state obtained after the relaxation process is dependent on the
the rigid-body motions. The boundary value problem is dis- crystal orientation and the crystal dimension. This can now
cretized with four node quad elements. The mesh contains be used as an appropriate starting point for the simulation of
60 × 60 elements. micro-structure evolution under applied field. Starting from
Now a relaxation process starting from a random initial the equilibrium state, we apply a saw-tooth type magnetic
state to an equilibrium state of the microscopic magnetiza- loading in the horizontal direction. The application of the
tion is performed as shown in Fig. 7b–f. In order to visualize magnetic field causes the magnetization to rotate in the direc-
the periodicity of the micro structure, the mechanical fluc- tion of the applied magnetic field as can be seen in Fig. 9b–c.
tuations are magnified. The vector distribution shows the This goes along with the characteristic domain wall motion
magnetization and the contour shows it’s gradient. This relax- such that the domains parallel to the applied magnetic field

Fig. 8 The zoomed out view of

the magnetization vectors at the
domain walls. The 90◦ rotation
of the magnetization can be
clearly seen. This smooth
rotation of the magnetization
vectors is captured by the
discretization used. (Color
figure online)

(a) (b) (c)

h1 h1 h1

(d) (e) (f)

h1 h1 h1
0.0 ||∇m|| 6e11

Fig. 9 Magnetic field driven loading on a converged equilibrium state. macroscopic field. As a consequence of the applied field a single domain
a–f shows the domain wall evolution for a cyclic magnetic field loading is formed and remains even under the reversal of the applied field
applied in the horizontal direction. Black arrow indicates the applied

123
164 Comput Mech (2016) 58:151–169

Fig. 10 Macroscopic hysteresis

Net magnetization m1 /ms

Magnetostriction ε011 10−4

curves for a cyclic magnetic b c d 3 e b c d
field loading. The 1
magnetostrictive strain ε011 and 2
the net magnetization m 1 /m s in 0.5
e 1
the horizontal direction versus f
0
the applied magnetic field h 1
0
−0.5

f −1
−1

−100 −50 0 50 100 −100 −50 0 50 100

Applied magnetic field h1 kA/m Applied magnetic field h1 kA/m

Table 2 Material parameters for Nd2 Fe14 B permanent magnet [19] disappear and the magnetization finally points in the direction
Parameter Name Unit Value of the applied field. The result is the saturation of the magne-
tization. After reaching saturation, the increase of magnetic
λ Lamé parameter N/m2 2 × 1011 field does not cause any change in the magnetization. Now
μ Lamé parameter N/m2 8 × 1010 the magnetic field is reversed. The material shows a remanent
E Magnetostrictive coefficient − 12 × 10−6 magnetization when the applied field becomes zero. This can
A Exchange energy coefficient N 1 × 10−12 be attributed to the dimension of the sample and the inher-
κ0 Magnetic permeability N/A2 1.3 × 10−6 ent single-crystal behavior. Again the magnetizations start to
K1 Anisotropy energy coefficient N/m2 4 × 106 rotate near the coercive field Fig. 9e. It can be clearly seen
K2 Anisotropy energy coefficient N/m2 6 × 105 that all the magnetization rotate together and no domain walls
ms Saturation magnetization A/m 8 × 105
evolve. Ultimately the magnetization reorients in the negative
direction and further application of the magnetic field does
η Damping coefficient Ns2 /m2 2 × 101
not cause any change. Figure 10 shows the corresponding
hysteresis curves. The first plot shows the variation of the
magnetization m 1 /m s with the applied magnetic field h 1 .
become larger and the other domains reduce in size. Fig. 9c– The magnetization value is normalized by the saturation
d shows that near the coercive field the domain walls start to

25 80 40

140 0 130

30 90 50
(a) (b) (c)

(d) (e) (f)

0.0 ||∇m|| 6e11

Fig. 11 a Representative volume element of a poly-crystal with ran- converges to a stable state. Vectors represent the magnetization and
dom distribution of the easy axes. The edge length is taken as l = 20 nm. the general poling direction is indicated by white arrows. The contour
b–f micro-magnetic domain evolution starting from a random distri- represents the gradient of the magnetization
bution of the magnetization. The micro-structure evolves and finally

123
Comput Mech (2016) 58:151–169 165

(a) (b) (c)

h1 h1 h1
1500

c
1000
h

h1 kA/m
500 g
b
d
0
a f

−500

−1000 e
(d) (e)
−1500
0 1 2 3 4 5 6
Time μs
h1 h1

(f) (g) (h)

h1 h1 h1

0.0 ||∇m|| 5e11

Fig. 12 Domain wall motion for an applied cyclic loading of the mag- wall evolution concentrated along the grain boundaries are observed.
netic field h 1 . The loading curve can be seen in the middle. a–h show Black arrows indicate the direction of the applied magnetic field
the domain wall motion for the applied cyclic loading. Complex domain

Fig. 13 Macroscopic hysteresis

Magnetostriction ε011 × 10−6

8
Net magnetization m1 /mS

1 d
curves of poly-crystalline c,h d,f
aggregate showing the average b 6 e c,h
0.5 b
magnetization m 1 /m s versus the
4
applied magnetic field h 1 and 0
butterfly hysteresis of the a 2
g
average magnetostrictive strain −0.5 g
ε 011 versus the applied magnetic 0
e f a
field −1 g
−1000 −500 0 500 1000
−1000 −500 0 500 1000

Applied magnetic field h1 kA/m Applied magnetic field h1 kA/m

value. It is seen that the magnetization increases upto the uration value. The second plot shows the magnetostrictive
saturation and then remains constant on further application strain ε011 and the applied magnetic field h 1 . The strain is
of magnetic field. Upon removal of the magnetic field we maximum when all the magnetizations point in the direction
can see a remanent magnetization which is equal to the sat- of the magnetic field and the value of this strain is equal

123
166 Comput Mech (2016) 58:151–169

to the magnetostrictive coeffecient E 100 . The two wells of Table 3 Material parameters for Iron α-Fe [19]
minima in the plot correspond to the point when the magne- Parameter Name Unit Value
tization rotates in the direction of the magnetic field and its
component in the horizontal direction is zero. λ Lamé parameter N/m2 2 × 1011
μ Lamé parameter N/m2 8 × 1010
5.2 Homogenization of a NdFeB permanent magnet E Magnetostrictive coefficient − 12 × 10−6
A Exchange energy coefficient N 1 × 10−12
Now we consider a poly-crystal of an isotropic nano- κ0 Magnetic permeability N/A2 1.3 × 10−6
crystalline permanent magnet Nd2 Fe14 B. For this we take K1 Anisotropy energy coefficient N/m2 4 × 104
the material parameters as given in Table 2 (Goll and Kron- K2 Anisotropy energy coefficient N/m2 6 × 103
müller [19]). Appropriate value are chosen for the mechanical ms Saturation magnetization A/m 5 × 106
parameters. A two-dimensional example which analyzes
η Damping coefficient Ns2 /m2 2 × 101
a poly-crystal consisting of nine grains is chosen as the
microstructure. We apply periodic boundary conditions for
z
the primary fields on the edges of the poly-crystal specimen Nd2 Fe14 B
shown in Fig. 11a. Each grain has different preferred direc-
tions characterized by the lattice orientations {ϑi }i=1,9 . The y

α-Fe
mechanical displacement and the magnetic potential at the
x
corner nodes are fixed to constrain the rigid-body motions. Nd2 Fe14 B
The material parameters are identical in all the grains. Again
the boundary value problem is discretized with four node
quad elements. The mesh now contains 90 × 90 elements
(30 × 30 elements per grain). z
z
A random distribution of magnetization is chosen as the
initial configuration and the material is allowed to relax to y y
an equilibrium state. This relaxation process is shown in
Fig. 11b–f. The grain boundaries are highlighted with white x
x
lines. Permanent magnets are highly anisotropic in nature and
this can be clearly seen in the equilibrium state Fig. 11f. The Fig. 14 Periodic Voronoi tessellation of a two-phase permanent mag-
net with grains of α-Fe and Nd2 Fe14 B. Periodic boundary conditions
magnetization in each grain has a strong tendency to orient
are applied on the surface of the microstructure. The easy axis of the
along the easy directions. The maximum domain wall thick- magnetization is shown for each grain, by the red arrows
ness is seen between the grains where the difference in the
easy directions is the largest. It is seen that trans-granualar
domain walls are not formed in the equilibrium state. This on further application of magnetic field. Upon removal of the
can be attributed to the high anisotropy of the material. In the magnetic field we can see a remanent magnetization in the
domains, i.e. regions separated by the walls, the magnetiza- aggregate. The value of this remanent magnetization how-
tion vectors are almost aligned. ever is seen to be lower than the saturation magnetization.
A magnetic field loading in the horizontal direction is now The second plot shows the plot of the magnetostrictive strain
applied on this equilibrium state. The subsequent domain ε011 with the applied field. We can see the characteristic but-
wall motion is shown in Fig. 12a–h. The magnetization rotate terfly hysteresis here, with the maximum strain observed at
in the direction of the applied magnetic field and finally sat- the saturation magnetization.
urate for a magnetic field higher than the coercive value. On
reversal on the applied field the magnetization in each grain 5.3 Homogenization of a two-phase magnet
starts to rotate in the direction of their respective easy direc-
tion. A remanent magnetization is seen upon removal of the A sample three-dimensional homogenization is shown in
magnetic field, with the domain walls appearing along the order to showcase the capability of the proposed homog-
grain boundaries. Further application of the magnetic field enization framework. A representative grain structure is
in the opposite direction causes the magnetization to rotate created using a periodic Voronoi tesselation, such that a
in the opposite direction and finally yields saturation. This is three-dimensional grain growth is created with the average
also clearly seen in the characteristic hysteretic plot Fig. 13. grain diameter taken as 1000 nm. A two-phase permanent
The first plot again shows the variation of the magnetization magnet, consisting of a soft magnetic material α-Fe and a
m 1 /m s with the applied magnetic field h 1 . The magnetiza- hard magnetic material Nd2 Fe14 B is considered. The mate-
tion increases upto the saturation and then remains constant rial parameters are shown in Tables 2 and 3 for Nd2 Fe14 B

123
Comput Mech (2016) 58:151–169 167

domains were not observed in the previous example where

we considered a single-phase hard magnetic material. It may
be inferred that the inclusion of the soft magnetic phase is
the reason for the formation of these trans-granular domains,
however further investigations need to be done to investi-
gate the influence of the soft magnetic phase. Due to high
computational cost the magnetic loading processes are not
considered here.

Acknowledgments The authors thank the German Research Foun-

dation (DFG) for financial support of the projects under the grants
Mi 295/71-1 and KE 1849/2-2 within the research unit FOR 1509.
M.-A.K. thanks the “Ministerium für Wissenschaft, Forschung und
Kunst des Landes Baden-Württemberg” as well as the DFG in the frame-
work of the Cluster of Excellence in “Simulation Technology” (EXC
310/2) at the University of Stuttgart.

References

1. Abbundi R, Clark A (1977) Anomalous thermal expansion and

magnetostriction of single crystal Tb27 Dy73 Fe2 . Magn IEEE Trans
13(5):1519–1520. doi:10.1109/TMAG.1977.1059598
2. Berger H, Kari S, Gabbert U, Rodriguez-Ramos R, Guinovart R,
Otero J, Bravo-Castillero J (2005) An analytical and numerical
approach for calculating effective material coefficients of piezo-
electric fiber composites. Int J Solids Struct 42(21–22):5692–5714.
doi:10.1016/j.ijsolstr.2005.03.016
3. Bertram HN (1994) Theory of magnetic recording. Cambridge Uni-
versity Press, Cambridge
4. Brassart L, Stainier L, Doghri I, Delannay L (2011) A variational
formulation for the incremental homogenization of elasto-plastic
composites. J Mech Phys Solids 59(12):2455–2475. doi:10.1016/
j.jmps.2011.09.004
5. Brown WF Jr (1966) Magnetoelastic interactions, tracts in natural
philosophy, vol 9. Springer-Verlag, New York
6. Brun M, Lopez-Pamies O, Ponte Castañeda P (2007) Homoge-
nization estimates for fiber-reinforced elastomers with periodic
microstructures. Int J Solids Struct 44:5953–5979
7. Claeyssen F, Lhermet N, Letty RL, Bouchilloux P (1997)
Actuators, transducers and motors based on giant magnetostric-
tive materials. J Alloys Compd 258(1–2):61–73. doi:10.1016/
Fig. 15 Domain evolution starting from a random distribution of the S0925-8388(97)00070-4
magnetization. Vectors represent the magnetization. Contour represents 8. Clark AE, Restorff JB, Wun-Fogle M, Hathaway KB, Lograsso
the gradient of the magnetization. Trans-granular domain patterns can TA, Huang M, Summers E (2007) Magnetostriction of ternary fe-
be observed due to the interaction of soft and hard magnetic phases ga-x(x=c, v, cr, mn, co, rh) alloys. J Appl Phys 101(9):09C507.
doi:10.1063/1.2670376
9. Cullity BD (1972) Introduction to magnetic materials. Addison-
and α-Fe respectively. Soft magnetic materials are used to Wesley, Reading
enhance the magnetic properties of hard magnets [19]. In 10. Daniel L, Hubert O, Billardon R (2004) Homogenisation of
magneto-elastic behaviour: from the grain to the macro scale. Com-
our sample grain as shown in Fig. 14, we consider a Nd-rich put Appl Math 23:285–308
microstructure. Different compositions of these materials can 11. Daniel L, Hubert O, Buiron N, Billardon R (2008) Reversible
be simulated and the consequent effect on the microscopic magneto-elastic behavior: a multiscale approach. J Mech Phys
domain formation and macroscopic properties can be stud- Solids 56(3):1018–1042
12. DeSimone A (1993) Energy minimizers for large ferromagnetic
ied. We use this example to show the capability of the model bodies. Arch Ration Mech Anal 125:99–143
to simulate complex micromagnetic domain evolution. The 13. DeSimone A, James R (2002) A constrained theory of magnetoe-
problem is discretized using around 39,000 four node trian- lasticity. J Mech Phys Solids 50:283–320
gualar elements. Periodic boundary conditions are applied 14. DeSimone A, Kohn RV, Müller S, Otto F (2004) Recent analytical
developments in micromagnetics. Technical report, Max-Planck-
on the surface of the microstructure. The domain evolution Institut für Mathematik in den Naturwissenschaften, Leipzig
and the equilibrium state are shown in Fig. 15. It can be seen 15. Engdahl G (2000) Handbook of giant magnetostrictive materials.
that trans-granular domains are formed. Such trans-granular Academic Press, San Diego

123
168 Comput Mech (2016) 58:151–169

16. Ethiraj G, Sridhar A, Miehe C (2015) Variational modeling and 39. Landau LD, Lifshitz EM (1935) On the theory of the dispersion
homogenization in dissipative magneto-mechanics. GAMM-Mitt of magnetic permeability in ferromagnetic bodies. Physikalische
38(1):75–101. doi:10.1002/gamm.201510004 Zeitschrift der Sowjetunion 8:153–169
17. Gilbert TL (1956) Formulation, foundations, and applications of 40. Landis CM (2008) A continuum thermodynamics formulation
the phenomenological theory of ferromagnetism. Ph.D. thesis, Illi- for micro-magnetomechanics with applications to ferromagnetic
nois Institute of Technology shape memory alloys. J Mech Phys Solids 56:3059–3076
18. Gilbert TL (2004) A phenomenological theory of damping in fer- 41. Liang CY, Keller SM, Sepulveda AE, Bur A, Sun WY, Wetzlar
romagnetic materials. IEEE Trans Magn 40:3443–3449 K, Carman GP (2014) Modeling of magnetoelastic nanostructures
19. Goll D, Kronmüller H (2000) High-performance permanent mag- with a fully coupled mechanical-micromagnetic model. Nanotech-
nets. Naturwissenschaften 87(10):423–438 nology 25(43):435,701
20. Greaves S (2008) Micromagnetic simulations of magnetic record- 42. Linnemann K, Klinkel S, Wagner W (2009) A constitutive model
ing media. In: High performance computing on vector systems for magnetostrictive and piezoelectric materials. Int J Solids Struct
2007. Springer, Berlin 46:1149–1166
21. Hill R (1972) On constitutive macro-variables for heterogeneous 43. Miehe C (2002) Strain-driven homogenization of inelastic
solids at finite strain. Proc R Soc Lond (Ser A) 326:131–147 microstructures and composites based on an incremental varia-
22. Hirsinger L, Barbier G, Billardon R (2000) Application of the tional formulation. Int J Numer Methods Eng 55:1285–1322
internal variable formalism to the modelling of magneto-elasticity. 44. Miehe C (2003) Computational micro-to-macro transitions of
Mech Electromagn Mater Struct 19:54–67 discretized micro-structures of heterogeneous materials at finite
23. Hu R, Soh AK, Zheng GP, Ni Y (2006) Micromagnetic model- strains based on the minimization of averaged incremental energy.
ing studies on the effects of stress on magnetization reversal and Comput Methods Appl Mech Eng 192:559–591
dynamic hysteresis. J Magn Magn Mater 301(2):458–468. doi:10. 45. Miehe C (2011) A multi-field incremental variational framework
1016/j.jmmm.2005.07.023 for gradinet-type standard dissipative solids. J Mech Phys Solids
24. Hubert A, Schäfer R (2001) Magnetic domains. Springer-Verlag, 59(4):898–923
New York 46. Miehe C, Dettmar J (2004) A framework for micro-macro transi-
25. James R, Kinderlehrer D (1990) Frustration in ferromagnetic mate- tions in periodic particle aggregates of granular materials. Comput
rials. Contin Mech Thermodyn 2:215–239 Methods Appl Mech Eng 193:225–256
26. Javili A, Chatzigeorgiou G, Steinmann P (2013) Computational 47. Miehe C, Ethiraj G (2012) A geometrically consistent incremental
homogenization in magnetomechanics. Int J Solids Struct 50:4197– variational formulation for phase field models in micromagnetics.
4216 Comput Methods Appl Mech Eng 245–246:331–347
27. Kankanala S, Triantafyllidis N (2004) On finitely strained magne- 48. Miehe C, Dettmar J, Zäh D (2010) Homogenization and two-scale
torheological elatomers. J Mech Phys Solids 52:2869–2908 simulations of granular materials for different microstructural con-
28. Keip MA, Steinmann P, Schröder J (2014) Two-scale computa- straints. Int J Numer Methods Eng 83:1206–1236
tional homogenization of electro-elasticity at finite strains. Comput 49. Miehe C, Kiefer B, Rosato D (2011) An incremental variational
Methods Appl Mech Eng 278:62–79 formulation of dissipative magnetostriction at the macroscopic con-
29. Keip MA, Schrade D, Thai H, Schröder J, Svendsen B, Müller tinuum level. Int J Solids Struct 48:1846–1866
R, Gross D (2015) Coordinate-invariant phase field modeling of 50. Miehe C, Rosato D, Kiefer B (2011) Variational principles
ferroelectrics, part II: application to composites and polycrystals. in dissipative electro-magneto-mechanics: a framework for the
GAMM-Mitt 38(1):115–131. doi:10.1002/gamm.201510006 macro-modeling of functional materials. Int J Numer Methods Eng
30. Kiefer B, Karaca H, Lagoudas D, Karaman I (2007) Characteriza- 86:1225–1276
tion and modeling of the magnetic field-induced strain and work 51. Miehe C, Zäh D, Rosato D (2012) Variational-based modeling
output in magnetic shape memory alloys. J Magn Magn Mater of micro-electro-elasticity with electric field- and stress-driven
312(1):164–175. doi:10.1016/j.jmmm.2006.09.035 domain evolution. Int J Numer Methods Eng 91(2):115–141
31. Kittel C (1949) Physical theory of ferromagnetic domains. Rev 52. Nemat-Nasser S, Hori M (1999) Micromechanics: overall proper-
Mod Phys 21(4):541–583 ties of heterogeneous materials. In: North-Holland series in applied
32. Kittel C (1956) Introduction to solid state physics, 2nd edn. Wiley, mathematics and mechanics, vol. 36, 2nd edn. Elsevier Science
New York Publisher B.V
33. Kouznetsova V (2002) Computational homogenization for the 53. Olabi A, Grunwald A (2008) Design and application of mag-
multi-scale analysis of multi-phase materials. Ph.D. thesis, Institute netostrictive materials. Mater Des 29(2):469–483. doi:10.1016/j.
of Mechanics of Materials, TU Eindhoven matdes.2006.12.016
34. Kouznetsova V, Geers MGD, Brekelmans WAM (2002) Multi- 54. Özdemir I, Brekelmans WAM, Geers MGD (2008) Computational
scale constitutive modelling of heterogeneous materials with a homogenization for heat conduction in heterogeneous solids. Int J
gradient-enhanced computational homogenization scheme. Int J Numer Methods Eng 73(2):185–204
Numer Methods Eng 54:1235–1260 55. Pisante G (2004) Homogenization of micromagnetics large bodies.
35. Kouznetsova V, Geers M, Brekelmans W (2004) Multi-scale ESAIM: control, optimisation and calculus of variations
second-order computational homogenization of multi-phase mate- 56. Ponte Castañeda P, Galipeau E (2011) Homogenization-based con-
rials: a nested finite element solution strategy. Comput Methods stitutive models for magnetorheological elastomers at finite strain.
Appl Mech Eng 193:5525–5550 J Mech Phys Solids 59:194–215
36. Kružík M, Prohl A (2006) Recent developments in the modeling, 57. Ponte Castañeda P, Siboni M (2012) A finite-strain constitutive
analysis, and numerics of ferromagnetism. SIAM Rev 48:439–483 theory for electro-active polymer composites via homogenization.
37. Kuznetsov S, Fish J (2012) Mathematical homogenization the- Int J Non-Linear Mech 47(2):293–306. doi:10.1016/j.ijnonlinmec.
ory for electroactive continuum. Int J Numer Methods Eng 2011.06.012
91(11):1199–1226. doi:10.1002/nme.4311 58. Ponte Castañeda P, Suquet P (1997) Nonlinear composites. Else-
38. Lahellec N, Suquet P (2007) On the effective behavior of non- vier, Amsterdam
linear inelastic composites: I. Incremental variational principles. 59. Sablik MJ, Jiles DC (1993) Coupled magnetoelastic theory of
J Mech Phys Solids 55(9):1932–1963. doi:10.1016/j.jmps.2007. magnetic and magnetostrictive hysteresis. IEEE Trans Magn
02.003 29(3):2113–2123

123
Comput Mech (2016) 58:151–169 169

60. Sandlund L, Fahlander M, Cedell T, Clark AE, Restorff JB, Wun- 67. Suquet P (1987) Elements of homogenization for inelastic solid
Fogle M (1994) Magnetostriction, elastic moduli, and coupling mechanics. In: Sanchez-Palenzia E, Zaoui A (eds.) Lecture Notes
factors of composite terfenol-d. J Appl Phys 75(10):5656–5658. in Physics: Homogenization Techniques for Composite Materials,
doi:10.1063/1.355627 vol. 272 edn. Springer-Verlag, pp 193–278
61. Schrefl T, Fidler J, Kronmüller H (1994) Remenence and coerciv- 68. Temizer I, Wriggers P (2011) Homogenization in finite thermoelas-
ity in isotropic nanocrystalline permanent magnets. Phys Rev B ticity. J Mech Phys Solids 59(2):344–372. doi:10.1016/j.jmps.
49(9):6100–6110 2010.10.004
62. Schröder J (2009) Derivation of the localization and homoge- 69. Yi M, Xu BX (2014) A constraint-free phase field model for ferro-
nization conditions for electro-mechanically coupled problems. magnetic domain evolution. Proc R Soc A 470(2171):20140,517.
Comput Mater Sci 46:595–599 doi:10.1098/rspa.2014.0517
63. Schröder J, Keip MA (2012) Two-scale homogenization of electro- 70. Zäh D, Miehe C (2013) Computational homogenization in dissipa-
mechanically coupled boundary value problems. Comput Mech tive electro-mechanics of functional materials. Comput Methods
50(2):229–244 Appl Mech Eng 267:487–510
64. Shu YC, Lin MP, Wu KC (2004) Micromagnetic modeling of 71. Zhang JX, Chen LQ (2005) Phase-field microelasticity theory and
magnetostrictive materials under intrinsic stress. Mech Mater micromagnetic simulations of domain structures in giant magne-
36(10):975–997 Active Materials tostrictive materials. Acta Mater 53:2845–2855
65. Smith RC, Dapino MJ, Seelecke S (2003) Free energy model for 72. Zhang JX, Chen LQ (2005) Phase-field model for ferromagnetic
hysteresis in magnetostrictive transducers. J Appl Phys 93(1):458– shape-memory alloys. Philos Mag Lett 85:533–541
466
66. Spaldin NA (2003) Magnetic materials: fundamentals and device
applications. Cambridge University Press, Cambridge

123