Computer Methods in Applied Mechanics and Engineering 199 (2010) 2765–2778

Contents lists available at ScienceDirect

Computer Methods in Applied Mechanics and Engineering

j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / c m a

A phase field model for rate-independent crack propagation: Robust algorithmic

implementation based on operator splits
Christian Miehe ⁎, Martina Hofacker, Fabian Welschinger
Institute of Applied Mechanics (CE) Chair I, University of Stuttgart, Pfaffenwaldring 7, 70550 Stuttgart, Germany

a r t i c l e i n f o a b s t r a c t

Article history: The computational modeling of failure mechanisms in solids due to fracture based on sharp crack
Received 4 January 2010 discontinuities suffers in situations with complex crack topologies. This can be overcome by a diffusive crack
Received in revised form 14 April 2010 modeling based on the introduction of a crack phase field. Following our recent work [C. Miehe, F.
Accepted 15 April 2010
Welschinger, M. Hofacker, Thermodynamically-consistent phase field models of fracture: Variational
Available online 7 May 2010
principles and multi-field fe implementations, International Journal for Numerical Methods in Engineering
Keywords:
DOI:10.1002/nme.2861] on phase-field-type fracture, we propose in this paper a new variational framework
Fracture for rate-independent diffusive fracture that bases on the introduction of a local history field. It contains a
Crack propagation maximum reference energy obtained in the deformation history, which may be considered as a measure for
Phase fields the maximum tensile strain obtained in history. It is shown that this local variable drives the evolution of the
Gradient-type damage crack phase field. The introduction of the history field provides a very transparent representation of the
Incremental variational principles balance equation that governs the diffusive crack topology. In particular, it allows for the construction of a
Finite elements new algorithmic treatment of diffusive fracture. Here, we propose an extremely robust operator split scheme
Coupled multi-field problem
that successively updates in a typical time step the history field, the crack phase field and finally the
displacement field. A regularization based on a viscous crack resistance that even enhances the robustness of
the algorithm may easily be added. The proposed algorithm is considered to be the canonically simple
scheme for the treatment of diffusive fracture in elastic solids. We demonstrate the performance of the phase
field formulation of fracture by means of representative numerical examples.
© 2010 Elsevier B.V. All rights reserved.

1. Introduction surface topology in the solid by diffusive crack zones governed by a

scalar auxiliary variable. This variable can be considered as a phase
The prediction of failure mechanisms due to crack initiation and field which interpolates between the unbroken and the broken state
propagation in solids is of great importance for engineering applica- of the material. Conceptually similar are recently outlined phase field
tions. Following the classical treatments of Griffith [2] and Irwin [3], approaches to brittle fracture based on the classical Ginzburg-Landau
cracks propagate if the energy release rate reaches a critical value. The type evolution equation as reviewed in Hakim & Karma [14], see also
Griffith theory provides a criterion for crack propagation but is Karma, Kessler & Levine [15] and Eastgate et al. [16]. In contrast to the
insufficient to determine curvilinear crack paths, crack kinking and above mentioned rate-independent approach, these models are fully
branching angles. In particular, such a theory is unable to predict crack viscous in nature and mostly applied to dynamic fracture. The phase
initiation. These defects of the classical Griffith-type theory of brittle field approaches to fracture offer important new perspectives towards
fracture can be overcome by variational methods based on energy the theoretical and computational modeling of complex crack
minimization as suggested by Francfort & Marigo [4], see also Bourdin, topologies. Recall that finite-element-based numerical implementa-
Francfort & Marigo [5], Dal Maso & Toader [6] and Buliga [7]. The tions of sharp crack discontinuities, such as interface element
regularized setting of their framework, considered in Bourdin, formulations, element and nodal enrichment strategies suffer in the
Francfort & Marigo [8,5], is obtained by Γ-convergence inspired by case of three-dimensional applications with crack branching. Ad-
the work on image segmentation by Mumford & Shah [9]. We refer to vanced XFEM-based methods for sharp crack propagation are outlined
Ambrosio & Tortorelli [10] and the reviews of Dal Maso [11] and in Belytschko et al. [17,18] We also refer to the adaptive interface
Braides [12,13] for details on Γ-convergent approximations of free methods by Gürses & Miehe [19] and Miehe & Gürses [20,21] for the
discontinuity problems. The approximation regularizes a sharp crack modeling of configurational-force-driven sharp crack propagation. In
contrast, phase-field-type diffusive crack approaches avoid the
⁎ Corresponding author. Tel.: + 49 711 685 66379.
modeling of discontinuities and can be implemented in a straightfor-
E-mail address: [email protected] (C. Miehe). ward manner by coupled multi-field finite element solvers. In the
URL: http://www.mechbau.uni-stuttgart.de/ls1/ (C. Miehe). recent work Miehe, Welschinger & Hofacker [1], we outlined a general

0045-7825/$ – see front matter © 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.cma.2010.04.011
2766 C. Miehe et al. / Computer Methods in Applied Mechanics and Engineering 199 (2010) 2765–2778

Fig. 1. Sharp and diffusive crack topology. a) Sharp crack at x = 0. b) Diffusive crack at x = 0 modeled with the length scale l in terms of the crack phase field d.

thermodynamically-consistent framework of phase field modeling in surface itself and postulate that it should stay constant or grow for
fracture mechanics and considered aspects of its numerical imple- arbitrary loading processes. The next modeling step, discussed in
mentation. It overcomes some of the difficulties inherent in the above Section 3, is concerned with the definition of energy storage and
mentioned phase field approaches, by definition of fracture only in dissipation functionals which depend on the fracture phase field. Here,
tension, precise characterization of the dissipation and introduction of we consider stored energy functions where the degradation due to the
the viscosity as a regularization of the rate-independent formulation. growing phase field acts only on a properly defined positive (tensile)
The numerical treatment was based on a monolithic three-field saddle part of the stored energy. The dissipation functional is directly related to
point principle, that includes beside the displacement and fracture the evolution of the crack surface functional via a material parameter
phase field a particular crack driving force field. that represents the critical energy release. With the dissipation and
In this work, we develop a new model of phase-field-type fracture energy functionals at hand, we derive the governing balance equations
that substantially enhances our formulations in Miehe, Welschinger & from an incremental variational principle. For the quasi-static processes
Hofacker [1]. The key novel aspects are the formulation of the balance under consideration, the coupled system of equations consists of the
equation for the evolution of the phase field in terms of a strain- static equilibrium condition for the degraded stresses and the balance
history source term, which allows an extremely simple and robust equation for the phase field. It characterizes a two-field problem with the
staggered solution of the two-field problem in terms of the algorithm displacement field and the phase-field as the primary variables. For this
summarized in Box 1. On the theoretical side, we outline an coupled field problem, we develop in Section 4 the new staggered
incremental variational formulation for a rate-independent evolution solution scheme which contains minimizers for each partial problem.
of the crack phase model. The important characteristic is the The discrete counterparts of the continuous variational principles
introduction of a crack surface density function that governs the results in two linear algebraic Euler equations for the algorithmically
dissipation in the diffusive fracture theory. The key observation then decoupled incremental problem. Finally, Section 5 outlines representa-
is that the possible discontinuous evolution of the phase field in time tive numerical examples which demonstrate the features and algorith-
is driven by a local history field. It contains a maximum reference mic robustness of the proposed phase field models of fracture.
energy obtained in the deformation history, which may be considered
as a measure for the maximum tensile strain obtained in history. It is 2. Crack phase field and crack surface density function
shown that this local variable drives the evolution of the crack phase
field. The introduction of the history field provides a very transparent 2.1. Phase field approximation of crack topology
representation of the balance equation that governs the evolution of
the diffusive crack topology. It represents a source term in the balance We motivate the construction of a variational principle of diffusive
equation for the phase-field. In order to stabilize post-critical solution crack topology by a simple model problem. Following our work Miehe,
paths, we may add a viscous regularization. This consists of the Welschinger & Hofacker [1], we consider an infinitely expanded bar of
introduction of a viscous term that supplements the crack surface cross-section Γ occupying the domain B= Γ × L with L = [−∞, +∞] and
resistance. A particular ingredient is that the rate-independent limit is position x ∈ L of its axis. Assume a crack of the bar at the axial position
recovered by simply setting the viscosity to zero. This feature of the x = 0, where Γ represents the fully-broken crack surface. This sharp crack
proposed theory is a very important element of a robust numerical topology may be described by an auxiliary field variable d(x) ∈ [0, 1],
implementation. The proposed approach is embedded in the theory of characterizing for d = 0 the unbroken state and for d = 1 the fully-
gradient-type materials with a characteristic length scale, such as broken state of the material. Fig. 1a gives a visualization. We denote
outlined in the general context by Capriz [22], Mariano [23] and the auxiliary variable d(x) as the crack phase field. Clearly, such an
Fremond [24]. On the numerical side, we develop a robust scheme for approach is related to the continuum theory of damage, where the
the incremental update of the fracture phase field and the displace- scalar damage field d describes in a homogenized macroscopic sense
ment field. It represents an operator split algorithm within a typical the development of micro-cracks and micro-voids. With this
time step that allows a staggered update of the phase field and the viewpoint in mind, we approximate the non-smooth phase field by
displacement field. The central idea for the algorithmic decoupling of the exponential function
the coupled equations is an approximation of the current history
field that provides the crack source term in the diffusive crack −jxj =l
dðxÞ = e : ð1Þ
topology equation. The formulation results in two linear problems
for the successive update of the phase field and the displacement
field within a typical time step. The proposed staggered scheme is It smears out the crack over the axial domain L of the bar, representing
extremely robust. The formulation may be considered as a canonical a regularized or diffusive crack topology as depicted in Fig. 1b. The
framework for the implementation of phase-field-type fracture in the regularization is governed by the length scale parameter l and gives
rate-independent setting. for l → 0 the sharp crack topology. The exponential function (1) has
We start our investigation in Section 2 with a descriptive motivation the property d(0) = 1 and d(±∞) = 0. Now observe that Eq. (1) is the
of a regularized crack topology based on a phase field. This treatment solution of the homogeneous differential equation
results in the definition of a crack surface functional, depending on the
crack phase field, that Γ-converges for vanishing length-scale parameter
2
to a sharp crack topology. We consider this crack functional as the crack dðxÞ−l d″ ðxÞ = 0 in B ð2Þ
 C. Miehe et al. / Computer Methods in Applied Mechanics and Engineering 199 (2010) 2765–2778 2767

Fig. 2. Sharp and diffusive crack topology. a) Sharp crack surface Γ embedded into the solid B. b) The regularized crack surface Γl(d) is a functional of the crack phase field d.

subject to the above Dirichlet-type boundary conditions. This Γ(t) ⊂ Rδ− 1 inside the solid B at time t, as depicted in Fig. 2a, we
differential equation is the Euler equation of the variational principle obtain in analogy to Eq. (3) the regularized crack phase field d(x, t) on
B, visualized in Fig. 2b, from the minimization principle
1 n 2 o
∫B d + l d′ dV
2 2
d = Argfinf d ∈ W IðdÞg with IðdÞ = ð3Þ ( )
2
dðx; tÞ = Arg inf Γl ðdÞ ð8Þ
d ∈ WΓðtÞ
and W = {d|d(0) = 1, d(±∞) = 0}. The functional I(d) can easily be
constructed by integrating a Galerkin-type weak form of the differential
Eq. (2). Now observe, that the evaluation of the functional for the subject to the Dirichlet-type constraints WΓ(t) = {d|d(x, t) = 1 at x ∈ Γ(t)}.
solution (1) gives with dV= Γdx the identification I(d = e−|x|/l) = lΓ, The Euler equations of the above variational principle are
which relates the functional I to the crack surface Γ. As a consequence,
we may introduce the functional 2
d−l Δd = 0 in B and ∇d·n = 0 on ∂B; ð9Þ
1 1 n 2 o
Γl ðdÞ : = IðdÞ = ∫B d + l d′ dV
2 2
ð4Þ
l 2l where Δd is the Laplacian of the phase field and n the outward normal on
∂B. Fig. 3 depicts a numerical solution of the variational problem (8) of
alternatively to Eq. (3)2. Clearly, the minimization of this scaled
diffusive crack topology and demonstrates the influence of the length
functional also gives the regularized crack topology (1) depicted in
scale l. We refer to the recent work Miehe, Welschinger & Hofacker [12],
Fig. 1b. However, the scaling by the length-scale parameter l has the
for a more detailed derivation.1
consequence that the functional Γl(d) may be considered as the crack
surface itself. In the one-dimensional problem under consideration, the
3. A framework of rate-independent diffusive fracture
evaluation of Γl(d) at the solution point gives for arbitrary length scales
l the crack surface Γ. This property characterizes the functional Γl as an
With the idea of a diffusive crack topology at hand, we develop
important ingredient of the subsequent constitutive modeling of
in this section a constitutive framework of phase-field-type fracture
diffusive crack propagation.
for the rate-independent setting. The subsequent formulation in
terms of a history field is motivated in Appendix A by a simple one-
2.2. Introduction of a crack surface density function
dimensional structure of local damage mechanics. Fig. 4 provides a
visual guide to the subsequent developments.
The one-dimensional description of a diffusive crack topology can
be extended to multi-dimensional solids in a straightforward manner.
3.1. Degrading strain energy functional due to fracture
Let B ⊂ Rδ, be the reference configuration of a material body with
dimension δ ∈ [1–3] in space and ∂B ⊂ Rδ− 1 its surface as depicted in
3.1.1. Displacement and strain fields
Fig. 2. In the subsequent treatment, we intend to study cracks in the
In the small-strain context, we describe the response of the
solid evolving in the range T ⊂ R of time. To this end, we introduce
fracturing solid by the displacement field
the time-dependent crack phase field
 
B × T →½0; 1 B × T →Rδ
d: ð5Þ u: ð10Þ
ðx; tÞ↦dðx; tÞ ðx; tÞ↦uðx; tÞ

defined on the solid B. Then, a multi-dimensional extension of the

one-dimensional regularized crack functional Eq. (4), derived in in addition to the phase field d introduced in Eq. (5). u(x, t) ∈ Rδ is the
Appendix A, reads displacement of the material point x ∈ B at time t ∈ T . The strains are
assumed to be small. Thus, the norm of the macroscopic displacement
Γl ðdÞ = ∫B γðd; ∇dÞdV; ð6Þ

where we have introduced the crack surface density function per unit
volume of the solid

1 2 l 1
Convergence to Sharp Crack Discontinuities. The Γ–limit Γl → Γ of the functional
γγððd;d;∇d = 1 dd2 +
∇dÞÞ= + l j∇d
j∇dj2j2 ð7Þ
2l2l 22 Eq. (6) for vanishing length scale l → 0 to the sharp crack surface Γ: = ∫ΓdA is considered
in Braides [12], see Fig. 3 for a visualization. The functional Eq. (4) provides the basis
for an elliptic regularization of the free discontinuity problem of brittle fracture. It has
This function, which depends on the crack phase field d and its spatial already been used in Bourdin, Francfort and Marigo [8] for the definition of a
gradient ∇d, plays a critical role in the subsequent modeling of regularized surface energy. In contrast, we introduce the functional in the purely
crack propagation. Assuming a given sharp crack surface topology geometric context with regard to the subsequent definition of a dissipation potential.
2768 C. Miehe et al. / Computer Methods in Applied Mechanics and Engineering 199 (2010) 2765–2778

Fig. 3. Solutions of the variational problem (8) of diffusive crack topology for a quadratic specimen with a horizontal sharp crack Γ, taken from [1]. Regularized crack surfaces Γl(d)
governed by the crack phase field d for different length scales la N lb N lc N ld. The sequence of pictures visualizes the Γ-limit Γl → Γ of the crack surface functional Eq. (6).

gradient ||∇u|| b  is bounded by a small number . The displacement 3.1.2. Constitutive free energy functional
gradient defines the small strain tensor We focus on the standard linear theory of elasticity for isotropic
solids by considering the global energy storage functional
1 T
εðuÞ = ∇s u := ½∇u + ∇ u: ð11Þ
2 Eðu; dÞ = ∫B ψðεðuÞ; dÞdV ð15Þ

In order to account for a stress degradation only in tension, we that depends on the displacement u and the fracture phase field d. The
decompose the strain tensor into positive and negative parts energy storage function ψ describes the energy stored in the bulk of the
solid per unit volume. A fully isotropic constitutive assumption for the
ε = εþ + ε− ð12Þ degradation of energy due to fracture may have the form

þ −
describing tensile and compressive modes. These contributions are ψψððε;ε;ddÞÞ=
=½g½gððddÞÞ+ ψþ
+kkψ ðεÞ + ψ− ðεÞ:
00ðεÞ + ψ00ðεÞ: ð16Þ
defined based on the spectral decomposition of the strain tensor
ε = ∑δi = 1εini ⊗ ni, where {εi}i = 1...δ are the principal strains and
In this multiplicative ansatz, ψ0 is an isotropic reference energy function
{ni}i = 1...δ the principal strain directions. We set
associated with the undamaged elastic solid, i.e.

2 2
εþ := ∑i
δ i i i δ i i i ψ0 ðεÞ = λtr ½ε = 2 + μtr½ε  ð17Þ
= 1 〈ε 〉+n⊗n and ε− := ∑i = 1 〈ε 〉−n ⊗ n ð13Þ

with elastic constants λ N 0 and μ N 0, which we additively decompose

in terms of the bracket operators 〈x〉+: = (x + |x|)/2 and 〈x〉−: = (x − |
x|)/2, respectively. The derivative of Eq. (12) with respect to the total þ −
ψðεÞ = ψ0 ðεÞ + ψ0 ðεÞ ð18Þ
strains defines the two projection tensors
−
into a positive part ψ+
0 due to tension and a negative part ψ0 due to
þ − þ compression. Here, we assume the definitions
ℙ : = ∂ε ½εþ ðεÞ and ℙ := I−ℙ ; ð14Þ
þ 2 2 − 2 2
ψ0 ðεÞ = λ〈tr½ε〉þ = 2 + μtr½εþ  and ψ0 ðεÞ = λ〈tr½ε〉− = 2 + μtr½ε− 
which are isotropic tensor functions of ε. These fourth-order tensors ð19Þ
project the total strains onto its positive and negative parts, i.e. ε+ =
P+:ε and ε− = P−:ε. The computation of these objects is performed with the above definitions of the brackets 〈x〉+ and 〈x〉_ and the
by the algorithms outlined in Miehe [25] and Miehe & Lambrecht positive and negative strain tensors ε+ and ε−, respectively. Note that
[26]. both terms are positive. Furthermore, observe that the volumetric

Fig. 4. A multi-field approach of phase-field-type crack propagation in deformable solids. The displacement field u, the fracture phase field d and the history field H are defined on the
solid domain B. a) The displacement field is constrained by the Dirichlet- and Neumann-type boundary conditions u = uD on ∂Bu and σ⋅ n = tN on ∂Bt with ∂B = ∂Bu∪∂Bt. b) The
fracture phase field is constrained by the possible Dirichlet-type boundary condition d = 1 on Γ and the Neumann condition ∇d⋅ n = 0 on the full surface ∂B. c) The history field H
defined in Eq. (40) contains a maximum local strain energy obtained within the fracture process. It drives the evolution of the fracture phase field d via Eq. (41).
 C. Miehe et al. / Computer Methods in Applied Mechanics and Engineering 199 (2010) 2765–2778 2769

Box 1
Staggered Scheme for Phase Field Fracture in [tn, tn + 1].

1. Initialization. The displacement, fracture phase and history fields un, dn and Hn at time tn are known. Update prescribed loads γ̄, ū, t̄ at
current time tn + 1.
2. Compute History. Determine maximum reference energy obtained in history
(
ψþ
0 ð∇s un Þ for ψþ
0 ð∇s un Þ N Hn
H=
Hn otherwise

in the domain B and store it as a history variable field.

3. Compute Fracture Phase Field. Determine the current fracture phase field d from the minimization problem of crack topology
n h η i o
2 2
d = Arg inf ∫B gc γðd; ∇dÞ + ðd−dn Þ −ð1−dÞ H dV
d 2τ

with the crack surface density function

1 2 l 2
γðd; ∇dÞ = d + j ∇d j :
2l 2

4. Compute Displacement Field. Determine the current displacement u at frozen fracture phase field d from the minimization problem of
elasticity
n o
u = Arg inf ∫B ½ψð∇s u; dÞ−γ·udV−∫∂B t⋅udA
u t

with the free energy density function

2 þ −
ψðε; dÞ = ½ð1−dÞ + kψ0 ðεÞ + ψ0 ðεÞ

with ψ± 2 2
0 (ε) = λ 〈 tr[ε] 〉 ±/2 + μtr[ε±] for the Dirichlet-type boundary condition u = ū on ∂Bu.

contribution is chosen either positive or negative according to the sign at given state {u, d}. Here, we introduced per definition the stress
of the volume dilatation e = tr[ε], which cannot be expressed in terms tensor
of the positive and negative strain tensors ε + and ε − . The
2
monotonically decreasing degradation function g(d) describes the σ := ∂ε ψðε; dÞ = ½ð1−dÞ + k½λ〈tr½ε〉þ 1 + 2μεþ  + ½λ〈tr½ε〉− 1 + 2με− 
degradation of the positive (tensile) part of the stored energy with ð23Þ
evolving damage. It is assumed to have the properties
and the energetic force
gð0Þ = 1 ; gð1Þ = 0 ; g′ ð1Þ = 0: ð20Þ þ
f := −∂d ψðε; dÞ = 2ð1−dÞψ0 ðεÞ: ð24Þ

The first two conditions include the limits for the unbroken and the Note that the energetic force f is positive and bounded by a finite value
fully-broken case. As shown below, the latter constraint ensures that for the limit ψ+
0 (ε) → ∞. As shown below this property ensures that
the energetic fracture force converges to a finite value if the damage the phase field variable is bounded by its maximum value d = 1. The
converges to the fully-broken state d = 1. A simple function that has key term that drives the crack evolution is the positive part ψ+ 0 (ε) of
the above properties is the reference energy, that may be considered to describe locally the
intensity of the tensile part of the deformation.
2
gðdÞ = ð1−dÞ : ð21Þ
3.2. Dissipation functional due to fracture
The small positive parameter k ≈ 0 in Eq. (16) circumvents the full
degradation of the energy by leaving the artificial elastic rest energy 3.2.1. Dissipation due to crack propagation
density kψ0(ε) at a fully-broken state d = 1. It is chosen as small as For a given fracture surface functional Γ(d) introduced in Eq. (6),
possible such that the algebraic conditioning number of the applied we define the work needed to create a diffusive fracture topology by
numerical discretization method remains well-posed for partly-
broken systems. Wc ðdÞ : = ∫B gc γðd; ∇dÞdV; ð25Þ

3.1.3. Rate of energy functional

where gc is the Griffith-type critical energy release rate and γ(d, ∇d)
Taking the time derivative of Eq. (15), we obtain the rate of energy
the crack surface density function defined in Eq. (7). The rate of the
which we consider to be a functional of the rates {u̇, ḋ}
work functional Wc defines the crack dissipation

Ėð u̇; ḋ; u; dÞ = ∫B ½σ : ∇s u̇−f ḋ dV ð22Þ Ẇ c ðḋ; dÞ = ∫B ðgc δd γÞḋdV; ð26Þ
2770 C. Miehe et al. / Computer Methods in Applied Mechanics and Engineering 199 (2010) 2765–2778

which is considered to be a functional of the rate {ḋ} of the crack phase The variation of the functional with respect to the four field variables,
field at given state {d}. Here, we introduced the variational or taking into account δu̇ = 0 on ∂Bu, yields the coupled field equations
functional derivative

1 ð1Þ : Div½σ + P γ = 0;
2
δd γðdÞ := ∂d γ−Div½∂∇d γ = ½d−l Δd ð27Þ ð2Þ : β−f = 0;
l
ð3Þ : ḋ−λ = 0; ð36Þ
of the crack density function. Based on thermodynamical arguments, ð4Þ : λ ≥ 0;
we demand ð5Þ : β−gc δd γ ≤ 0;
ð6Þ : λðβ−gc δd γÞ =0
Ẇ c ðḋ; dÞ≥ 0 ð28Þ
in the domain B along with the boundary conditions
as the basic ingredient of our framework, enforcing a growth of the
fracture work Wc defined in Eq. (25). Observe that we may satisfy this
σ·n = t on ∂Bt and ∇d·n = 0 on ∂B: ð37Þ
global irreversibility constraint of crack evolution by ensuring locally a
positive variational derivative of the crack surface function and a
positive evolution of the crack phase field Note that we have introduced per definition the stress tensor σ
and the energetic force f defined in Eqs. (23) and (24). The last
δd γ ≥ 0 and ḋ ≥ 0: ð29Þ three equations in Eq. (36) are the Kuhn-Tucker-type equations
associated with the optimization problem with inequality con-
The former condition is ensured in the subsequent treatment by a straints. From Eq. (36)2,3, elimination of β = f and λ = ḋ yields the
constitutive assumption that relates the functional derivative to a reduced system
positive energetic driving force. The latter constraint is a natural P
ð1Þ : Div½σ + γ = 0;
assumption that relates the fracture phase field for the non-reversible
ð2Þ : ḋ ≥ 0;
evolution of micro-cracks and micro-voids. ð38Þ
ð3Þ : f −gc δd γ ≤ 0;
ð4Þ : ḋðf −gc δd γÞ = 0:
3.2.2. Rate-independent evolution of the phase field
In order to satisfy the local constraints Eq. (29) within a possibly
Note that Eq. (38)2 satisfies explicitly the desired thermodynamic
discontinuous, rate-independent evolution, we introduce the local
consistency condition (29)2. The first condition gcδdγ ≥ 0 in (29)1 is
threshold function
also satisfied. This may be seen if the damage field is computed in the
case of loading from Eq. (38)3, which results into
tc ðβ; dÞ = β−gc δd γðdÞ ≤ 0 ð30Þ
gc 2 þ
gc δd γ := ½d−l Δd = 2ð1−dÞψ0 ðεÞ for ḋ N 0: ð39Þ
formulated in terms of the variable β dual to d, in what follows l
denoted as the driving force field. Then, as shown below, the above Note that the right hand side is positive which proves the
local constraints Eq. (29) can be satisfied by the definition condition (29)1. Thus the proposed model of rate-independent
diffusive crack evolution is consistent with the thermodynamic
Ẇ c ðḋ; dÞ = sup Dλ ðḋ; β; λ; dÞ ð31Þ
β;λ≥0 axiom of positive dissipation. Note furthermore, that Eq. (39) has
the desired property d → 1 for ψ+
0 (ε) → ∞.
in terms of the extended dissipation functional
3.3.2. Compact history-field-based formulation
Dλ ðḋ; β; λ; dÞ = ∫B ½β ḋ−λtc ðβ; dÞdV ð32Þ
It is clear that the ‘load term’ ψ+
0 (ε) in Eq. (39) determines the
amount of the phase field variable d. Hence, motivated by Eq. (A.7) of
that includes the threshold function (30). λ is a Lagrange multiplier the local damage model outlined in Appendix A, we may introduce the
field. local history field of maximum positive reference energy

3.3. Incremental variational principle and governing equations þ

Hðx; tÞ := max ψ0 ðεðx; sÞÞ ð40Þ
s∈½0;t

3.3.1. An incremental variational principle

With the rate of the energy functional Eq. (22) and the extended obtained in a typical, possibly cyclical loading process. Replacing ψ+
0 in

dissipation functional Eq. (32) at hand, we introduce the incremental Eq. (39) by this field, we obtain the equation
potential
h ii
gcghc 22
d−l
d−l ΔdΔd ==2ð21−d
ð1−d
ÞHÞH ð41Þ
Πλ ðu̇; ḋ; β; λ; u; dÞ = Ėðu̇; ḋ; u; dÞ + Dλ ðḋ; β; λ; dÞ−Pðu̇Þ; ð33Þ ll

that balances the internal power Ė + Dλ with the power due to which determines the phase field in the case of loading and unloading.
external loading Note that this equation equips the crack topology Eq. (9)1 by the local
crack source on the right hand side. With this notion at hand, the
Pðu̇Þ = ∫Bγ̄· u̇dV + ∫ t· u̇dA: ð34Þ
∂Bt proposed fracture phase field model may be reduced to the compact
system of only two equations
Here, γ̄ is a prescribed volume force in B and t̄ a surface traction on ∂Bt.
We then obtain the basic field equations of the problem from the
ð1Þ : Div½σðu; dÞ + P
γ = 0;
argument of virtual power that we base on the variational statement ð42Þ
ð2Þ : gc δd γðdÞ−2ð1−dÞH = 0;

f u̇; ḋ; β; λg = Argf stat Πλ ð u̇; ḋ; β; λ; u; dÞg: ð35Þ which determine the current displacement and phase fields u and d in
u̇; ḋ;β;λ≥0 terms of the definitions (23), (40), (19) and (27) for the stresses σ,
 C. Miehe et al. / Computer Methods in Applied Mechanics and Engineering 199 (2010) 2765–2778 2771

the history field H of maximum reference energy ψ+ 0 and the 4.1. Staggered update scheme of time-discrete fields
variational derivative δdγ of the crack density function. This is
considered to be the most simple representation of rate-independent 4.1.1. Time-discrete fields
diffusive fracture, associated with the multi-field scenario visualized We now consider field variables at the discrete times 0, t1, t2, …, tn,
in Fig. 4. tn + 1, …, T of the process interval [0, T]. In order to advance the
solution within a typical time step, we focus on the finite time
3.4. Viscous regularization of the rate-independent problem increment [tn, tn + 1], where

As commented on in Section 5, a viscous regularization of the τn + 1 := tn + 1 −tn N 0 ð50Þ

above rate-independent formulation may stabilize the numerical
treatment. It can be formulated in terms of a variational principle denotes the step length. In the subsequent treatment, all field
based on the potential variables at time tn are assumed to be known. The goal then is to
determine the fields at time tn + 1 based on variational principles valid
for the time increment under consideration. In order to obtain a
Πη ðu̇; ḋ; β; u; dÞ = Ėðu̇; ḋ; u; dÞ + Dη ðḋ; β; dÞ−Pðu̇Þ; ð43Þ
compact notation, we drop in what follows the subscript n + 1 and
consider all variables without subscript to be evaluated at time tn + 1.
that includes a modified extended dissipation functional
In particular, we write
1 2 uðxÞ := uðx; tn + 1 Þ and dðxÞ := dðx; tn + 1 Þ ð51Þ
Dη ðḋ; β; dÞ = ∫B ½β ḋ− 〈tc ðβ; dÞ〉þ dV; ð44Þ
η
for the displacement and fracture phase field at the current time tn + 1
where η ≥ 0 is a viscosity parameter. The modified variational and
principle
un ðxÞ := uðx; tn Þ and dn ðxÞ := dðx; tn Þ ð52Þ

fu̇; ḋ; βg = Argf stat Πη ðu̇; ḋ; β; u; dÞg ð45Þ for the fields at time tn. As a consequence, the rates of the
u̇; d;˙ β
displacement and the fracture phase field are considered to be
constant in the time increment Eq. (50) under consideration, defined
then results in the coupled set of local equations by u̇ = (u − un)/τ and ḋ: = (d − dn)/τ. Note that, due to the given
fields at time tn, the above rates associated with the time increment
ð1Þ : Div½σ + P
γ = 0; Eq. (50) are linear functions of the variables (Eq. (51)) at the current
1 ð46Þ time tn+ 1.
ð2Þ : ḋ− 〈f −gc δd γ〉þ = 0;
η
4.1.2. Update of history field
We consider an operator split algorithm within the typical time step
where the field β has been eliminated. Note that the evolution ḋ of the [tn, tn + 1] that allows a staggered update of the fracture phase field and
phase field is now governed by a viscous equation governed by the the displacement field. The central idea for an algorithmic decoupling of
‘over-force’ 〈f − gcδdγ〉+. As a consequence, Eq. (41) may be recast into the coupled equations is an approximated formulation of the current
the viscous regularized format history field H: = H(x, tn + 1) in terms of the displacement field un at
time tn defined in Eq. (52). Starting from the initial condition
ggc chh 22 ii
d−l Δd
d−l +ηηḋḋ=
Δd + =22ðð1−d
1−dÞH
ÞH ð47Þ H0 := Hðx; t = t0 Þ = 0; ð53Þ
ll
we assume the current value of maximum reference energy obtained
again driven by the local history field H of maximum positive
in history to be determined by
reference energy defined in Eq. (40). This equation equips the crack
topology Eq. (9)1 by the viscous resistance term on the left hand side ( þ þ
ψ0 ð∇ s un Þ for ψ0 ð∇s un Þ N Hn ;
and the crack source term on the right hand side. Thus the viscous H= ð54Þ
regularization of the system (42) reads Hn otherwise:

As a consequence of this simple definition, the energy H that drives

ð1Þ : Div½σðu; dÞ + P
γ = 0;
ð48Þ the current fracture phase field d at time tn + 1 is in the case of damage
ð2Þ : gc δd γðdÞ + η ḋ−2ð1−dÞH = 0: loading dependent on the displacement un at time tn. With this
algorithmic definition at hand, we may define two decoupled
The model is thermodynamically consistent by satisfying the variational problems which define the phase field d and the
constraints displacement u at the current time tn + 1.

δd γ + η ḋ ≥ 0 and ḋ ≥ 0; ð49Þ 4.1.3. Update of phase field

Considering the energetic history H defined in Eq. (54) to be constant
which contain in addition to Eq. (29) a contribution to the dissipation due in the time interval [tn, tn + 1], we introduce the algorithmic functional
to the viscous resistance of the phase field evolution. Observe that the
rate-independent case is recovered by simply setting η=0. This feature is τ η 2 2
Πd ðdÞ = ∫B ½gc γðd; ∇dÞ + ðd−dn Þ −ð1−dÞ HdV ð55Þ
a very convenient ingredient of a robust numerical implementation. 2τ

where γ is the crack surface density function introduced in Eq. (7). The
4. Staggered solution of incremental multi-field problem
current fracture phase field is then computed from the algorithmic
variational problem
We construct in this section a robust solution procedure of the
multi-field problem visualized in Fig. 4 based on convenient τ
d = Argfinf Πd ðdÞg: ð56Þ
algorithmic operator splits of the evolution equations. d
2772 C. Miehe et al. / Computer Methods in Applied Mechanics and Engineering 199 (2010) 2765–2778

Fig. 5. Rate-independent local constitutive behavior. a) Cyclic driver in positive and negative range and b) stress response without damage evolution in compression.

h
The first two terms in the variational functional describe the crack in terms of the nodal phase field vector dd ∈RN . Bd is a symbolic
surface resistance and the viscous resistance, respectively. The last term representation of a global interpolation matrix, containing the shape
represents the source term governed by the energetic history field H. functions and their derivatives.2 Introducing the potential density
τ
Note that Πd is quadratic. Hence, the necessary condition of Eq. (56) is a function
simple linear problem for the determination of the current phase field.
The Euler equation represents a discretized version of Eq. (47). Observe τ η 2 2
πd ðd; ∇dÞ = gc γðd; ∇dÞ + ðd−dn Þ −ð1−dÞ H; ð60Þ
that the rate-independent case is recovered by simply setting the 2
viscosity η = 0. This is considered as an important feature of our
formulation. No ill-conditioning occurs for the limit η = 0. The viscosity in Eq. (55), the spatial discretization of the variational principle (56) reads
is used as an artificial feature that stabilizes the simulation. It will be  
chosen as small as possible. τ
dd = Arg inf ∫Bh πd ðBd dd ÞdV : ð61Þ
dd

4.1.4. Update of displacement field

For a known fracture phase field d at time tn + 1, we compute the The associated Euler equation is linear and can be solved in closed form
current displacement field u from the variational principle of linear h i−1
elasticity. Let T 2 τ T τ
dd = − ∫Bh Bd ½∂cd cd πd Bd dV ∫Bh Bd ½∂cd πd dV ð62Þ

τ
Πu ðuÞ = ∫B ½ψð∇s u; dÞ−γ·udV−∫∂Bt t·udA ð57Þ
for the current nodal values of the phase field at time tn + 1.
be the elastic potential energy at given phase field d, we get the
4.2.2. Update of displacement field
displacement field from the minimization problem
Associated with Th, we write the finite element interpolations of
u = Argfinf Πu ðuÞ g
τ
ð58Þ the displacement field and its symmetric gradient by
u
h h
cu := fu; ∇s ug = Bu ðxÞdu ðtÞ ð63Þ
of elasticity. Note that ψ is quadratic. Thus the necessary condition of
Eq. (58) gives a linear system. in terms of the nodal displacement vector du ∈ R . The symbolic δNh

global interpolation matrix Bu contains the shape functions and its

4.1.5. Staggered update scheme derivatives. Introducing the potential density function
The staggered algorithm in the time interval [tn, tn + 1] is summarized
in Box 1. It represents a sequence of two linear subproblems for the τ
πu ðu; ∇s uÞ = ψð∇s u; dÞ−γ·u; ð64Þ
successive update of the fracture phase field and the displacement field.
Such an algorithm is extremely robust. Clearly, it may slightly
underestimate the speed of the crack evolution when compared with in Eq. (57), the spatial discretization of the variational principle (58)
a fully monolithic solution of the coupled problem as considered in reads for a zero traction problem
Miehe, Welschinger & Hofacker [1]. However, this can be controlled by
 
making use of an adaptive time stepping rule. τ
du = Arg inf ∫Bh πu ðBu du ÞdV : ð65Þ
du
4.2. Spatial discretization of the staggered problem

Let Th denote a finite element triangulation of the solid domain B. 2

Shapes of FE Discretization. For two-dimensional plane strain problems δ = 2, the
The index h indicates a typical mesh size based on Eh finite element constitutive state vector of the phase field discretization Eq. (59) reads cd = [d, d,1, d,2].
domains Bhe ∈ Th and Nh global nodal points. We use the same Then, associated with node I of a standard finite element e, the finite element
triangulation for spatial discretization of both the phase field as well interpolation matrix has the form
as the displacement field.
e  T
½Bd I = N N;1 N;2 I
4.2.1. Update of phase field
in terms of the shape function NI at note I and their derivatives. The state vector of the
Associated with Th, we write the finite element interpolations of displacement discretization reads cu = [u1, u2, u1,1, u2,2, u1,2 + u2,1] and the associated
the phase field and its gradient by finite element interpolation matrix has the form
 T
e N 0 N;1 0 N;2
h h ½Bu I = :
cd : = fd; ∇d g = Bd ðxÞdd ðtÞ ð59Þ 0 N 0 N;2 N;1 I
 C. Miehe et al. / Computer Methods in Applied Mechanics and Engineering 199 (2010) 2765–2778 2773

5. Representative numerical examples

We now demonstrate the modeling capability of the proposed

approach to diffusive fracture by means of numerical model problems.
The proposed staggered scheme outlined in Box 1 based on the local
history field is much faster than the monolithic three-field formula-
tion outlined in Miehe, Welschinger & Hofacker [1]. As shown below,
it allows performing rate-independent problems in a straightforward
manner. We demonstrate this for a spectrum of standard benchmark
problems such as a single edge notched specimen subjected to tensile
and pure shear loadings, a symmetric and an asymmetric notched
three point bending test. The characteristic of the local material
response for a cyclic loading process is demonstrated in Fig. 5. The
result is obtained by a local driving of the constitutive model. It shows
Fig. 6. Single edge notched specimen. Geometry and boundary conditions. a) Tension the basic properties of a rate-independent hysteresis due to local
test treated in Section 5.1 and b) shear test treated in Section 5.2. damage mechanisms which occur only in the tensile range. There is no
diffusive fracture evolution in the compressive range.
The associated Euler equation of linear elasticity yields the closed
form solution
5.1. Single edge notched tension test
h i−1
T 2 τ T τ
du = − ∫Bh Bu ½∂cu cu πu Bu dV ∫Bh Bu ½∂cu πu dV ð66Þ
Consider a squared plate with horizontal notch which is placed at
middle height from the left outer surface to the center of the
for the current nodal values of the phase field at time tn + 1. specimen. The geometric setup is depicted in Fig. 6. In order to capture

Fig. 7. Single edge notched tension test. Load-deflection curves for a length scale a) l1 =0.0150 mm and b) l2 =0.0075 mm obtained for η=1×10− 6 kN s/mm2 and η=0 kN s/mm2.

Fig. 8. Single edge notched tension test. Crack pattern for η = 1 × 10− 6 kN s/mm2 at a displacement of a) u = 5.7 × 10− 3 mm, b) u = 5.9 × 10− 3 mm, c) u = 6.1 × 10− 3 mm for a length
scale of l1 = 0.0150 mm and d) u = 5.7 × 10− 3 mm, e) u = 5.9 × 10− 3 mm, and f) u = 6.3 × 10− 3 mm for a length scale of l2 = 0.0075 mm.
2774 C. Miehe et al. / Computer Methods in Applied Mechanics and Engineering 199 (2010) 2765–2778

Fig. 9. Single edge notched pure shear test. Load-deflection curves for a length scale a) l1 = 0.0150 mm and b) l2 = 0.0075 mm obtained for η = 1 × 10− 6 kN s/mm2 and η = 0 kN s/
mm2.

the crack pattern properly, the mesh is refined in areas where the crack propagation in the post critical regime. The results are in
crack is expected to propagate, i.e. in the centered strip of the agreement with a monolithic solution, see Fig. 12 in Miehe,
specimen. For a discretization with 20,000 elements an effective Welschinger & Hofacker [1]. As expected, the sharpest crack pattern
element size of h ≈ 0.001 mm in the critical zone is obtained. is obtained for the smallest length scale parameter l2 = 0.0075 mm.
Following [1], we choose the maximum element size in this zone to The resulting crack patterns at different stages of the deformation for
be one half of the length scale. The elastic bulk modulus is chosen to the length scales l1 = 0.0150 mm and l2 = 0.0075 mm are illustrated
λ = 121.15 kN/mm2, the shear modulus to μ = 80.77 kN/mm2 and the in Fig. 8. The illustration uses the viscous model with η = 1 × 10− 6 kN
critical energy release rate to gc = 2.7 × 10− 3 kN/mm. The computa- s/mm2, which does not show the brutal crack evolution of the rate-
tion is performed in a monotonic displacement driven context with independent problem. The viscosity has a stabilizing effect. However,
constant displacement increments of Δu = 1 × 10− 5 mm in the first we are able to compute the rate-independent limit without problems.
500 time steps and needs to be adjusted to Δu = 1 × 10− 6 mm in the
remaining time steps due to the brutal character of the crack 5.2. Single edge notched pure shear test
propagation. In order to point out the effects which arise due to the
length scale parameter l and a viscosity η, different simulations are We now investigate the same squared plate with horizontal notch as
performed. For fixed length scale parameters l1 = 0.0150 mm and in the previous example for a shear loading mode. The geometric setup
l2 = 0.0075 mm, the influence of the viscosity is analyzed. The load- and the applied deformation are depicted in Fig. 6. In order to capture
deflection curves obtained are depicted in Fig. 7. One observes that for the crack pattern properly, the mesh is again refined in areas where the
the rate-independent case η = 0 the structural response shows a crack is expected to propagate, i.e. in the lower right diagonal strip of the
steeper descent whereas the viscous model smoothes out the brutal specimen. An effective element size of approximately h ≈ 0.002 mm

Fig. 10. Single edge notched pure shear test. Crack pattern for η = 1 × 10− 6 kN s/mm2 at a displacement of a) u = 9.0 × 10− 3 mm, b) u = 11.0 × 10− 3 mm, c) u = 13.4 × 10− 3 mm for a
length scale of l1 = 0.0150 mm and d) u = 9.0 × 10− 3 mm, e) u = 11.0 × 10− 3 mm, and f) u = 14.8 × 10− 3 mm for a length scale of l2 = 0.0075 mm.
 C. Miehe et al. / Computer Methods in Applied Mechanics and Engineering 199 (2010) 2765–2778 2775

5.3. Symmetric three point bending test

This test is a classical benchmark problem which has often been

analyzed in literature, for example in the recent work Miehe & Gürses
[20] in the context of configurational-force-driven crack propagation:
a simply supported notched beam. The geometric setup as well as
the loading conditions are illustrated in Fig. 11. The discretization is
refined in the expected crack propagation zone, yielding a discretiza-
tion with 20,000 elements and an effective element size of
h ≈ 0.0008 mm in the critical zone. The bulk modulus is chosen to
λ = 12.00 kN/mm2, the shear modulus to μ = 8.0 kN/mm2 and the
Fig. 11. Symmetric three point bending test. Geometry and boundary conditions. critical energy release rate to gc = 5.0 × 10− 4 kN/mm. The computa-
tion is performed in a monotonic displacement driven context with
based on a mesh with 30,000 elements is used. The elastic bulk modulus constant displacement increments Δu = 1 × 10− 4 mm in the first 360
is chosen to λ = 121.15 kN/mm2, the shear modulus to μ = 80.77 kN/ loading steps. Continuing crack propagation then demands for an
mm2 and the critical energy release rate to gc = 2.7 × 10− 3 kN/mm. The adjustment of the displacement increment to Δu = 1 × 10− 6 mm for
computation is performed in a monotonic displacement driven context the subsequent loading steps. Fig. 12 illustrates the global response for
with constant displacement increments of Δu = 1 × 10− 5 mm. Again, two different length scales in combination with different viscosities.
we perform different simulations in order to point out the effects due to In Fig. 12a results for a length scale l1 = 0.06 mm and in Fig. 12b for a
different length scale parameters l and viscosity η. For fixed length scale length scale l2 = 0.03 mm are compared to a simulation obtained from
parameters l1 = 0.0150 mm and l2 = 0.0075 mm, the influence of the a fracture model based on configurational-force-driven crack propa-
viscosity is analyzed. The load–deflection curves obtained are depicted gation introduced by Miehe & Gürses [20] for a simulation with 10,360
in Fig. 9. For the rate-independent case η = 0 the structural response elements. The proposed new phase field model shows a very good
shows a steeper descent whereas the viscous model smoothes out the agreement even for the use of larger length scales. The resulting
brutal crack propagation in the post critical range. As expected, the contour plots of the crack topology, again for the two different length
sharpest crack pattern is obtained for the smallest length scale scale parameters under focus, are given in Fig. 13. Blue and red colors
parameter l2 = 0.0075 mm. The resulting crack patterns at different correspond to the undamaged and the fully cracked material,
stages of the deformation for the length scales l1 = 0.0150 mm and respectively. In summary, we emphasis that the proposed phase
l2 = 0.0075 mm are illustrated in Fig. 10. field model of fracture yields a very good agreement with results

Fig. 12. Symmetric three point bending test. Load-deflection curves for a length scale a) l1 = 0.06 mm and b) l2 = 0.03 mm obtained for η = 1 × 10− 6 kN s/mm2 and η = 0 kN s/mm2.
Structural response is compared to the material force based algorithm, see Miehe et al. [24].

Fig. 13. Symmetric three point bending test. Crack pattern for η = 0 kN s/mm2 at a displacement of a) u = 4.0 × 10− 2 mm, b) u = 4.1 × 10− 2 mm, c) u = 1.0 × 10− 1 mm for a length
scale of l1 = 0.06 mm and d) u = 3.8 × 10− 2 mm, e) u = 4.1 × 10− 2 mm, and f) u = 1.0 × 10− 1 mm for a length scale of l2 = 0.03 mm.
2776 C. Miehe et al. / Computer Methods in Applied Mechanics and Engineering 199 (2010) 2765–2778

triangular elements and is refined in areas where the crack is

expected to propagate. The length scale parameter is chosen to
l = 0.025 mm which is twice the element size. Fig. 15 illustrates the
crack path obtained by different numerical models in comparison
to the experimental result. The first two pictures show the crack
path based on the rate-independent fracture model. In Fig. 15a and b
the results are obtained for η = 1.0 × 10− 6 kN s/mm2 and η = 0.0 kN s/
mm2, respectively. Fig. 15c visualizes the crack path of the
configurational force model analyzed in Miehe & Gürses [20].
Fig. 15d finally shows the experimental result of Bittencourt et al.
[27]. Taking a closer look at the first two pictures, one can observe a
slight difference regarding the width of the crack paths, although an
equal length scale parameter was used for the simulation. This effect
can be related to the brutal character of the crack propagation for
Fig. 14. Asymmetric notched three point bending test. Geometry, loading and boundary
conditions from Bittencourt et al. [27]. The three holes have a diameter of 0.5. zero viscosity. Adding a very small amount of viscosity improves the
robustness. A comparison of the results obtained by the phase field
model, the configurational force model and the experimental analysis
obtained by alternative computational methods of rate-independent shows that both strategies capture the experimentally observed
brittle fracture. curved crack pattern very well. The configurational force model with
the adaptive reorientation of the segments at the crack tip yields crack
trajectories for quite rough meshes. However, such a formulation is of
5.4. Asymmetric notched three point bending test limited applicability, because it cannot be applied to the modeling of
crack initiation in solids free of defects and does not allow for crack
Next, we apply the rate-independent fracture model to a more branching. These problems are overcome by the proposed phase field
complex boundary value problem with an evolving curvilinear crack model for fracture. The strong influence of the viscosity on curvilinear
trajectory. crack patterns in the phase field modeling of fracture has already been
The asymmetric three point bending test was analyzed experi- discussed in Miehe, Welschinger & Hofacker [1]. Hence, a key
mentally and numerically in Bittencourt et al. [27]. It concerns an advantage of the algorithmic formulation proposed in this work is
asymmetrically-notched beam with three holes. The setup of the that robust computations can be performed for the full rate-
problem is shown in Fig. 14. The material parameters are the elastic independent response without any viscosity. Finally, Fig. 16 shows
constants λ = 12.00 kN/mm2, μ = 8.0 kN/mm2 and the critical energy the evolution of the contour plots, again blue and red colors
release rate gc = 1 × 10− 3 kN/mm. The mesh consists of 58,000 correspond to the undamaged and fully cracked states, respectively.

Fig. 15. Asymmetric notched three point bending test. Crack topology of the rate-independent formulation with 58,000 elements and a) a viscosity η = 1.0 × 10− 6 kN s/mm2 and b) a
zero viscosity. c) Crack trajectories of Miehe & Guerses [20] and d) experimentally-obtained crack patterns by Bittencourt et al. [27].

Fig. 16. Asymmetric notched three point bending test. a)–d) Crack pattern for η = 1 × 10− 6 kN s/mm2 at different stages of the deformation for a length scale of l = 0.025 mm.
 C. Miehe et al. / Computer Methods in Applied Mechanics and Engineering 199 (2010) 2765–2778 2777

6. Conclusion where the function h(ε) is directly related to the reference energy
ψ0(ε). As outlined in Eq. (A.6)3, the convex reference free energy is a
We outlined a thermodynamically consistent framework for rate- quadratic function of the norm |ε|E of the strain ε with regard to a
independent diffusive crack propagation in elastic solids. To this end, metric provided by the elasticity modulus E. It obviously drives the
we proposed a new incremental variational framework for rate- accumulation of the damage variable d. Note that Eq. (A.6)1 provides
independent diffusive fracture that bases on the introduction of a local the desired property d → 1 for h(ε) → ∞ or |ε|E → ∞. Furthermore,
history field. It contains a maximum reference energy obtained in the observe that damage accumulation ḋ N 0 takes place only if the function
deformation history, which may be considered as a measure for the h(ε) defined in Eq. (A.6)3 grows. Defining H(t) as the maximum value
maximum tensile strain obtained in history. It was shown that this of h(ε(t)) obtained in time history, we may replace system (A.5) for the
local variable drives the evolution of the fracture phase field. It rate-independent evolution of the damage variable d by the simple
allowed the construction of an extremely robust operator split closed-form equation
scheme that successively updates in a typical time step the history
field, the crack phase field and finally the displacement field. An HðtÞ
dðtÞ = with HðtÞ := maxs∈½0;t fhðεðsÞÞ g: ðA:7Þ
artificial viscous regularization stabilizes the overall performance. The c + HðtÞ
proposed algorithm is considered to be the canonically simple scheme
for the treatment of diffusive fracture. We demonstrated the See for example Miehe [28] for an analogous definition of a history
performance of the proposed phase field formulations of fracture by field in rate-independent, discontinuous damage mechanics. The two
means of representative numerical examples. Eqs. (A.2) and (A.7) then govern exclusively a damage-type degrading
stress response in an elastic solid. Note that in this model due to E N 0
Acknowledgment the current damage variable d is in a one-to-one relationship to the
maximum strain norm obtained in time history
Support for this research was provided by the German Research pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Foundation (DFG) under grant Mi 295/11-2. ε max ðtÞ := max f jεðsÞjg = HðtÞ = E: ðA:8Þ
s∈½0;t

Appendix A. Introduction of a damage-driving history field

References
In order to motivate the structure of the diffusive fracture model [1] C. Miehe, F. Welschinger, M. Hofacker, Thermodynamically-consistent phase field
outlined in Section 3, we consider a local damage model in a one- models of fracture: Variational principles and multi-field fe implementations,
International Journal for Numerical Methods in Engineering DOI: 10.1002/nme.2861.
dimensional setting. It is governed by a free energy function
[2] A.A. Griffith, The phenomena of rupture and flow in solids, Philosophical
Transactions of the Royal Society London A 221 (1921) 163–198.
2 1 2
ψðε; dÞ = gðdÞψ0 ðεÞ with gðdÞ = ð1−dÞ and ψ0 ðεÞ = 2 Eε [3] G.R. Irwin, Fracture, in: S. Flügge (Ed.), Elasticity and Plasticity, Encyclopedia of
Physics, Vol. 6, Springer, 1958, pp. 551–590.
ðA:1Þ [4] G.A. Francfort, J.J. Marigo, Revisiting brittle fracture as an energy minimization
problem, Journal of the Mechanics and Physics of Solids 46 (1998) 1319–1342.
in terms of the convex reference energy function ψ0(ε) = ψ(ε, 0) and [5] B. Bourdin, G.A. Francfort, J.J. Marigo, The Variational Approach to Fracture,
Springer Verlag, Berlin, 2008.
the degrading function g(d), which depend on the strain ε and the
[6] G. Dal Maso, R. Toader, A model for the quasistatic growth of brittle fractures:
internal damage variable d. E is the elastic stiffness. The degrading Existence and approximation results, Archive for Rational Mechanics and Analysis
function is positive g ≥ 0, monotonic decreasing g′ ≤ 0 and has the 162 (2002) 101–135.
[7] M. Buliga, Energy minimizing brittle crack propagation, Journal of Elasticity 52
properties g(0) = 1, g(1) = 0 and g′(1) = 0. From Eq. (A.1), we obtain
(1999) 201–238.
by a standard exploitation method of the second axiom of thermo- [8] B. Bourdin, G.A. Francfort, J.J. Marigo, Numerical experiments in revisited brittle
dynamics, often referred to as Coleman's method, the constitutive fracture, Journal of the Mechanics and Physics of Solids 48 (2000) 797–826.
expression for the stresses [9] D. Mumford, J. Shah, Optimal approximations by piecewise smooth functions and
associated variational problems, Communications on Pure and Applied Mathe-
matics 42 (1989) 577–685.
2
σ ðε; dÞ := ∂ε ψðε; dÞ = ð1−dÞ Eε ðA:2Þ [10] L. Ambrosio, V.M. Tortorelli, Approximation of functionals depending on jumps by
elliptic functionals via γ-convergence, Communications on Pure and Applied
Mathematics 43 (1990) 999–1036.
and the reduced dissipation inequality [11] G. Dal Maso, An Introduction to Γ-Convergence, Birkhäuser, Boston, 1993.
[12] D.P. Braides, Approximation of Free Discontinuity Problems, Springer Verlag,
Berlin, 1998.
D = f ḋ ≥ 0 with f ðε; dÞ := −∂d ψðε; dÞ = 2ð1−dÞ ψ0 ðεÞ: ðA:3Þ
[13] D.P. Braides, Γ-Convergence for Beginners, Oxford University Press, New York,
2002.
A rate-independent, discontinuous evolution of the internal [14] V. Hakim, A. Karma, Laws of crack motion and phase-field models of fracture,
Journal of the Mechanics and Physics of Solids 57 (2009) 342–368.
variable can be based on the threshold function [15] A. Karma, D.A. Kessler, H. Levine, Phase-field model of mode iii dynamic fracture,
Physical Review Letters 92 (2001) 8704.045501.
tð f ; dÞ = f −cd: ðA:4Þ [16] L.O. Eastgate, J.P. Sethna, M. Rauscher, T. Cretegny, C.-S. Chen, C.R. Myers, Fracture
in mode i using a conserved phase-field model, Physical Review E 65 (2002)
036117-1-10.
Here, the constant c = 1 N/m2 has the value one and the unit of an [17] T. Belytschko, H. Chen, J. Xu, G. Zi, Dynamic crack propagation based on loss of
energy density. With this function at hand, a rate-independent hyperbolicity and a new discontinuous enrichment, International Journal for
Numerical Methods in Engineering 58 (2003) 1873–1905.
evolution of damage is defined by the equations [18] J.-H. Song, T. Belytschko, Cracking node method for dynamic fracture with finite
elements, International Journal for Numerical Methods in Engineering 77 (2009)
ḋ ≥ 0; tð f ; dÞ ≤ 0; ḋtð f ; dÞ = 0: ðA:5Þ 360–385.
[19] E. Gürses, C. Miehe, A computational framework of three-dimensional configu-
rational-force-driven brittle crack propagation, Computer Methods in Applied
Considering damage loading, ḋ N 0, we may compute the current Mechanics and Engineering 198 (2009) 1413–1428.
damage variable d from the condition t(f; d) = 0 in Eq. (71)2, yielding [20] C. Miehe, E. Gürses, A robust algorithm for configurational-force-driven brittle
crack propagation with r-adaptive mesh alignment, International Journal for
the closed form solution Numerical Methods in Engineering 72 (2007) 127–155.
[21] C. Miehe, E. Gürses, M. Birkle, A computational framework of configurational-
hðεÞ force-driven brittle fracture based on incremental energy minimization, Interna-
2
d= for ḋ N 0 with hðεÞ = 2ψ0 ðεÞ = j ε j E ; ðA:6Þ tional Journal of Fracture 145 (2007) 245–259.
c + hðεÞ [22] G. Capriz, Continua with Microstructure, Springer Verlag, 1989.
2778 C. Miehe et al. / Computer Methods in Applied Mechanics and Engineering 199 (2010) 2765–2778

[23] P.M. Mariano, Multifield theories in mechanics of solids, Advances in Applied [27] T.N. Bittencourt, P.A. Wawrzynek, A.R. Ingraffea, J.L. Sousa, Quasi-automatic
Mechanics 38 (2001) 1–93. simulation of crack propagation for 2d lefm problems, Engineering Fracture
[24] M. Frémond, Non-Smooth Thermomechanics, Springer Verlag, 2002. Mechanics 55 (1996) 321–334.
[25] C. Miehe, Comparison of two algorithms for the computation of fourth-order [28] C. Miehe, Discontinuous and continuous damage evolution in Ogden-type large-
isotropic tensor functions, Computers & Structures 66 (1998) 37–43. strain elastic materials, European Journal of Mechanics A / Solids 14 (1995)
[26] C. Miehe, M. Lambrecht, Algorithms for computation of stresses and elasticity 697–720.
moduli in terms of Seth–Hill's family of generalized strain tensors, Communica-
tions in Numerical Methods in Engineering 17 (2001) 337–353.