Available online at www.sciencedirect.

com
ScienceDirect

Comput. Methods Appl. Mech. Engrg. 340 (2018) 1000–1022

www.elsevier.com/locate/cma

Modeling dynamic fracture of solids with a phase-field regularized

cohesive zone model
Vinh Phu Nguyen a , Jian-Ying Wu b , ∗
a Department of Civil Engineering, Monash University, Clayton, Victoria 3800, Australia
b State Key Laboratory of Subtropical Building Science, South China University of Technology, 510641 Guangzhou, China

Received 15 April 2018; received in revised form 7 June 2018; accepted 11 June 2018
Available online 30 June 2018

Highlights
• Dynamic fracture is modeled by a phase-field regularized cohesive zone model.
• Both brittle fracture and quasi-brittle failure under dynamic loadings are considered.
• The results are independent of the length scale parameter regularizing the sharp cracks.
• Numerical results are in good agreement with existing computational/experimental findings.

Abstract

Being able to seamlessly deal with complex crack patterns like branching, merging and even fragmentation, the phase-field
model, amongst several alternatives, is promising in the computational modeling of dynamic fracture in solids. This paper presents
an extension of our recently introduced phase-field cohesive zone model for static fracture to dynamic fracture in brittle and quasi-
brittle solids. The model performance is tested with several benchmarks for dynamic brittle and cohesive fracture. Good agreement
is achieved with existing findings and experimental results; and particularly the results are independent to the discretization
resolution and the incorporated length scale parameter. The latter is in contrast to existing phase-field models.
⃝c 2018 Elsevier B.V. All rights reserved.

Keywords: Phase-field model; Cohesive zone model; Dynamic fracture; Damage; Concrete

1. Introduction
Fracture is one of the most commonly encountered failure modes of engineering materials and structures.
The prevention of fracture-induced failure is, therefore, a major concern in engineering designs. As with many
other physical phenomena, computational modeling of the crack initiation and propagation in solids constitutes an
indispensable tool not only to predict failure of cracked structures but also to shed insights into the mechanism of the
fracture processes of many materials such as concrete, rock, ceramic, metals, biological soft tissues etc. As mentioned
∗ Corresponding author.
E-mail address: [email protected] (J.-Y. Wu).

https://doi.org/10.1016/j.cma.2018.06.015
0045-7825/⃝ c 2018 Elsevier B.V. All rights reserved.
 V.P. Nguyen, J.-Y. Wu / Comput. Methods Appl. Mech. Engrg. 340 (2018) 1000–1022 1001

in [1], at some scale all fracture is dynamic due to the dynamic process of the bond rupture and thus ‘the dynamic
fracture problem is the most fundamental in the science of fracture’. Within the context of continuum mechanics [2],
this work addresses the dynamic fracture problem of brittle and quasi-brittle solids from a computational modeling
point of view. One of our aims is to understand and anticipate the crack path and crack speed under the influence of
stress waves.
Dynamic fracture has been extensively studied from different angles by researchers from different communities.
For the theoretical and experimental aspects the readers are referred to the textbooks of Freund [3], Ravi-Chandar
[4] and recent reviews in [1,5]. Theoretically, the Rayleigh wave speed c R is the limiting velocity of a mode-I
propagating crack [3]. However, experiments have shown that the limiting velocity of a crack reaches just about
half of the predicted value [3,4]. An additional issue in dynamic fracture is crack branching. That is, a single crack
will, under certain conditions, bifurcate to two growing cracks [6]. In what follows we present a brief literature review
on the computational modeling of dynamic fracture using the finite element method (FEM) [7,8] which is arguably the
most widely used numerical method in engineering. Note that FEM is also the most widely used numerical method
for phase-field models – the ones being used in this paper – at the time of this writing. Meshless/meshfree methods are
therefore intentionally not touched upon, and we refer to e.g. [9,10] and references therein. A more comprehensive
review on numerical modeling of brittle/cohesive fracture can be found in [11].
Fracture of solids can be modeled using either a discontinuous approach (also referred to as a discrete approach)
or a continuous one (sometimes also referred to as a smeared one). In the former, the displacement field is allowed to
be discontinuous across fracture surfaces whereas in the latter the displacements are continuous everywhere but the
stresses are gradually reduced to describe the degradation process using some softening material models. The most
well-known theories behind the discontinuous approaches are the linear elastic fracture mechanics (LEFM) [12,13],
and the cohesive zone model (CZM) pioneered in [14–16]. Continuum damage mechanics (CDM) is probably the
most widely used theory categorized in the continuous approach to fracture [17,18].
Among various discrete approaches to dynamic fracture one can count the extended finite element method
(XFEM), see for example, [19–23], the embedded strong discontinuity methods [24], the cohesive interface element
technique [25–29], the cracking node method [30] and the space–time discontinuous Galerkin method [31,32]. In the
embedded strong discontinuity method and XFEM, the kinematics of an element is enhanced or enriched in order
to represent a discontinuous displacement field. To this end, the discontinuity surface has to be explicitly tracked,
which is, sometimes, an intractable task particularly for those problems with arbitrary and complex crack paths in
three dimensions. Except the formulation where interface elements are inserted a priori [25,26] and the cracking node
method [30], all other methods require a branching criterion, quite often just an ad hoc one, and usually limited to two
crack branches. All aforementioned methods, except [28], employed velocity-independent fracture energy and yet are
able to reproduce the experimentally observed phenomenon of dynamic crack branching, a limiting crack speed well
below c R , and dynamic crack instability. Note that in all the aforementioned discontinuous approaches employing the
CZM, the traction–separation law of linear softening is adopted.
Modeling dynamic fracture using a CDM model is rather scarce, we can cite the ones reported in [33] using a non-
local integral damage model [34], in [35] using a non-local stress based damage model, in [36] with a microplane
damage model regularized using the crack band method [37], and in [38] using the gradient-enhanced damage
model [39]. Among them, Pereira et al. [35], Ožbolt et al. [36] reported successful modeling of dynamic crack
branching of concrete samples.
Peridynamics, originated by Silling [40], Silling et al. [41], is a novel non-local continuum mechanics without
spatial derivatives. Accordingly, it can handle topologically complex fractures such as intersecting and branching
cracks in both two and three dimensions. Moreover, the dual-horizon peridynamics [42,43] can effectively suppress
spurious wave reflections. The governing equations of peridynamics are integro-differential equations which result
in a natural meshless discretization [44,45]. Therefore, modeling dynamic fracture under extreme loading (impact,
explosive, etc.) of large structures is also feasible [46,47].
Phase-field fracture/damage models [48,49] are relatively new compared to LEFM, CZM and CDM. The model
presented in [48] aims to seek for, at the same time, the displacement field and the cracks by globally minimizing a
potential energy that consists of the stored bulk energy, the work of external forces, and the surface energy dissipated
during fracture process. It was coined the variational approach to fracture that generalizes Griffith’s energetic theory
for brittle fracture. The numerical implementation was set forth in [50] where sharp cracks are regularized by a
diffuse phase-field (or damage) approximation. The formulation introduced a small positive length scale parameter
1002 V.P. Nguyen, J.-Y. Wu / Comput. Methods Appl. Mech. Engrg. 340 (2018) 1000–1022

and when it approaches to zero, the solution converges to the one of the original problem according to Γ -convergence
theorems.1 The governing equations, i.e., the conventional equilibrium equation and an extra evolution law of gradient
type similar to the screened Poisson equation [51,52] for the crack phase-field, are usually derived from the variational
approach [50,53]. Phase-field models can seamlessly handle topologically complex fractures such as intersecting
cracks, branching cracks in both two dimensional and three dimensional solids with a relatively simple computer
implementation. Recently, the application has been extended from fracture in solids with infinitesimal deformations
to plates, shells and composites under finite strains [54–57]. State-of-the-art reviews of phase-field models can be
found in [58,59]. As phase-field models for fracture and damage are related to both fracture mechanics and damage
mechanics, they are herein referred to as phase-field damage, or phase-field fracture or even shortly as phase-field
models. Note that we have skipped those phase-field models which are developed by the applied physics community
since they are distant to the well accepted Griffith’s fracture theory and refer to [60] for these works.
Based on the assumption that the kinetic energy is not affected by the phase-field and the fracture energy is constant
(i.e., independent of the crack velocity), phase-field models have been applied to dynamic fracture as reported in
[61–69]. The phenomenon of dynamic crack branching experimentally observed can be captured and mesh
independent fracture energies were obtained. Except [69], a majority of those work adopted some phase-field models
without an elastic domain. Consequently, overestimation of the dissipation energy was reported [67]. Moreover, most
of these works focused on the numerical aspects of phase-field modeling of dynamic crack growth whereas Bleyer
et al. [66] and Bleyer and Molinari [70] presented some explanations of dynamic fracture behavior using the model
of Li et al. [69].
The length scale parameter plays a significant role in phase-field models that regularize the sharp crack topology
into a diffuse damage band with a small but finite length scale. In earlier phase-field models [50,53], the length
scale is regarded as a numerical parameter and its values should be taken as small as possible such that the
Γ -convergence to the original problem can be guaranteed. However, these phase-field models predict crack nucleation
at a critical stress inversely proportional to the square root of the length scale parameter. Accordingly, in the limit of
a vanishing length scale, the deficiency of Griffith’s theory is inherited when dealing with scaling properties and
crack nucleation [71]. Consequently, it is generally assumed that, rather than a numerical parameter, the incorporated
length scale should be a material property and fixed for a specific problem as in gradient-damage models. Usually,
the length scale is determined from the tensile strength using the exact homogeneous solution under 1-D uniaxial
traction [67,72]. However, this value is often very large compared to the specimen size, see [73] for concrete fracture
and very small e.g., for polymethylmethacrylate (PMMA) materials [74]. This has led these authors to scale down and
up the length scale parameter quite arbitrarily. Still, good agreement with the experimental load–displacement data
has been reported [74,75].
Recently, Wu [76–78] proposed a unified phase-field theory with a generic crack geometric function and a new
energetic/stress degradation function such that both brittle fracture and cohesive failure can be accounted for. The
formulation is based purely on standard thermodynamics with internal variables. Moreover, the traction and the
apparent displacement jump across the localization band mimic the traction–separation law (TSL) in Barenblatt’s
CZM. With a set of optimal characteristic functions, those general softening laws frequently adopted for brittle/quasi-
brittle materials, such as the linear, exponential, hyperbolic, Cornelissen et al. [79] ones, etc., can be reproduced
or approximated with sufficient precision. Remarkably, all the involved model parameters can be determined from
the standard material properties, i.e., Young’s modulus, the failure strength, the fracture energy, the initial slope and
ultimate crack opening of the objective softening curve. As the resulting model is equivalent to the well-known
cohesive zone model (CZM), at least in one dimension, we have named it as a phase-field regularized CZM [80]. It
is emphasized that this model is different from the one elaborated in e.g., [53,81–83] which requires the introduction
of an auxiliary field that defines the crack opening vector and only applies to a predefined crack. Note that a similar
phase-field for cohesive fracture was reported in [84] which, however, gives numerical results heavily sensitive to
the value of the length scale parameter [85] similarly to those phase-field models for brittle fracture. Compared to
other related models, the phase-field regularized CZM is of several merits. On the one hand, this model can deal with
arbitrary crack propagation with no need of explicit representation of the sharp crack surface and of cumbersome crack
tracking algorithm as in the aforesaid CZM based discontinuous approaches. On the other hand, being equivalent to
the CZM (at least in the 1-D case), it gives numerical results independent of the length scale parameter, so long as the
1 We recall that Γ -convergence is a notion of variational convergence that implies convergence of minimizers.
 V.P. Nguyen, J.-Y. Wu / Comput. Methods Appl. Mech. Engrg. 340 (2018) 1000–1022 1003

Fig. 1. A solid medium with a sharp crack (left) and its phase-field regularized counterpart (right).

damage field of high gradient within the localization band can be sufficiently resolved. These two properties are both
very useful for the modeling of complex crack nucleation and propagation, in particular, under dynamic loading.
It is the aim of this paper to extend the above phase-field regularized CZM to dynamic fracture. In line with existing
models, it is also assumed here that the kinetic energy is not affected by the crack phase-field and the fracture energy
is constant. Well-known benchmark tests for dynamic fracture including a crack branching test, an edge-cracked
plate under impulsive loading test and a mixed-mode concrete fracture test are solved. Good agreement is achieved
with existing findings and experimental results. Moreover, the numerical results are independent of the discretization
resolution and the incorporated length scale for both brittle and quasi-brittle fracture. This property is in contrast to
existing phase-field models where the results are sensitive to the length scale.
The remainder of this paper is structured as follows. Section 2 addresses the governing equations for phase-
field modeling of dynamic fracture. This is followed by Section 3 which presents the phase-field regularized CZM.
Section 4 is devoted to the finite element implementation and solution algorithm. Numerical performances of the
proposed model are investigated in Section 5, regarding several benchmark tests of dynamic fracture of brittle and
quasi-brittle solids. The most relevant conclusions are drawn in Section 6.
Notation. Compact tensor notation is used in the theoretical part of this paper. As general rules, scalars are denoted
by italic light-face Greek or Latin letters (e.g. a or λ); vectors, second- and fourth-order tensors are signified by italic
boldface minuscule, majuscule and blackboard-bold majuscule characters like a, A and A, respectively. The inner
products with single and double contractions are denoted by ‘·’ and ‘:’, respectively. The Voigt notation of vectors and
second-order tensors are denoted by boldface minuscule and majuscule letters like a and A, respectively.

2. Phase-field modeling of dynamic fracture

As shown in Fig. 1 (left), the reference configuration of a cracking solid Ω ⊂ Rn dim (n dim = 1, 2, 3) with a sharp
crack set S ⊂ Rn dim −1 is considered. The external boundary is denoted by ∂Ω ⊂ Rn dim −1 , with the outward unit
normal denoted by vector n. The material particles of the solid are labeled by their spatial coordinates x. The solid is
kinematically described by the displacement field u(x, t) : Ω × [0, T ] → Rn dim in time t ∈ [0, T ] ∈ R+ for some
time interval T > 0. Upon the assumption of infinitesimal deformations, the strain field ϵ(x, t) : Ω × [0, T ] →
[Rn dim ×n dim ]sym is given by ϵ(x, t) := ∇ sym u(x, t), for the symmetric gradient operator ∇ sym (·) with respect to the
spatial coordinate x.

2.1. Phase-field regularization of sharp cracks

In the context of phase-field damage models [50,53,72,76,78,86], the sharp crack S is smeared over a localization
band B ⊆ Ω in which the diffuse phase-field d(x) : B → [0, 1] localizes, with the exterior domain Ω \B being intact;
see Fig. 1(right) for an illustration. Note that the damage irreversibility reads ḋ ≥ 0, with (˙) being the time derivative.
1004 V.P. Nguyen, J.-Y. Wu / Comput. Methods Appl. Mech. Engrg. 340 (2018) 1000–1022

Being a fundamental ingredient of any phase-field damage model, the geometrically regularized phase-field
approximation Ad (d) of the sharp crack surface AS is expressed as
∫ ∫
Ad (d) = γ (d) dV ≈ dA = AS (2.1)
B S
     
regularized crack sharp crack

for the crack surface density functional γ (d)

[ ] ∫ 1
1 1 ⏐ ⏐2
γ (d) = α(d) + b ∇d α(β) dβ
√
⏐ ⏐ with c0 = 4 (2.2)
c0 b 0
where b is a length scale that characterizes the localization bandwidth; the geometric crack function α(d) ∈ [0, 1]
satisfies the properties that α(d) = 0 for d = 0 and α(d) = 1 for d = 1. Concrete forms for α(d) will be given later.
Note that for a vanishing length scale b → 0 the sharp crack topology AS is recovered in the context of the
Γ -convergence theorem [87]: AS = limb→0 Ad (d). However, it is becoming more and more popular in the literature
of phase-field models that the length scale parameter is interpreted as a material property [71,72,75], disregarding the
favorable Γ -convergence of the original phase-field model [50,53]. Nevertheless, it also introduces the issues of the
length scale being too large or too small discussed in the introduction session. In this work, we adopt an alternative
approach to guarantee the length scale independence of global responses while preserving the Γ -convergence.

2.2. Governing equations

Assume the cracking solid is subjected to specified volumetric body forces (per unit mass) b∗ (x, t) and surface
boundary tractions t∗ (x, t) for some part of the external boundary ∂Ωt ⊆ ∂Ω . Given displacements u∗ (x, t) are applied
to the disjointed remaining boundary ∂Ωu ⊆ ∂Ω , i.e., ∂Ωu ∩ ∂Ωt = ∅ and ∂Ωu ∪ ∂Ωt = ∂Ω .
Upon the above setting, the dynamic equilibrium equation for the solid Ω is expressed as

∇ · σ + b∗ = ρ ü in Ω (2.3)
together with the following Neumann boundary conditions

σ · n = t∗ on ∂Ωt (2.4)
for the solid density ρ and acceleration field ü = d2 u/dt 2 . In order for the resulting initial boundary value problem to
be well-posed, the following Dirichlet boundary conditions

u(x, t) = u∗ (x, t) on ∂Ωu (2.5a)

and initial conditions
u(x, 0) = u0 (x), u̇(x, 0) = v0 (x) (2.5b)
are considered, for the velocity field u̇(x, t) := du(x, t)/dt.
Furthermore, one needs to relate the stress σ to the strain ϵ and possibly, to some internal variables. For the
phase-field damage model interested in this work, the stress and damage energy release rate (or driving force) are
expressed formally as [76,78]
∂ ∂
σ := ψ(ϵ, d), Y := − ψ(ϵ, d) (2.6)
∂ϵ ∂d
for the local free energy density functional ψ(ϵ, d) of the cracking solid Ω to be specified later.
In order to derive the evolution law for the crack phase-field d(x), variational or thermodynamical arguments can
be employed as in [50,76,86]. Alternatively, motivated from the fact that the dissipation rate due to the evolution of
the localization band B should be equivalent to that of the original sharp crack S, Wu [78] proposed a much simpler
postulate of energetic equivalence, resulting in the following damage criterion
ḋ > 0
{
=0
g(Y, d) := Y − G f δd γ (2.7)
<0 ḋ = 0
 V.P. Nguyen, J.-Y. Wu / Comput. Methods Appl. Mech. Engrg. 340 (2018) 1000–1022 1005

and Neumann boundary condition

∇d · nB = 0 on ∂B (2.8)
where nB represents the outward unit normal vector of the boundary ∂B; G f is the fracture toughness (energy) in the
context of Griffith’s theory [12]; the variational derivative δd γ of the crack surface density functional γ (d) is defined
by
[ ]
1 1 ′
δd γ := ∂d γ − ∇ · ∂∇d γ = α (d) − 2b1d
( )
(2.9)
c0 b
for Laplacian 1d := ∇ · ∇d and the first derivative α ′ (d) := ∂α/∂d.
Eqs. (2.3) and (2.7) constitute the system of governing equations for phase-field modeling of dynamic fracture,
supplemented by the boundary conditions (2.4), (2.8) and (2.5a) as well as the initial conditions (2.5b).

2.3. Weak form

In accordance with the weighted residual method, the above governing equations in strong form can be rewritten
as the following weak form: Find u ∈ Uu and d ∈ Ud such that
⎧∫ ∫
⎨ ρ ü · δu dV + σ : ∇ sym δu dV = δP ∀δu ∈ Vu
⎪
⎪
Ω Ω
∫ (2.10)
−Y δd + G f δγ dV ≥ 0 ∀δd ∈ Vd
⎪
⎪ ( )
⎩
B
for the virtual power δP of external forces and the variation δγ of the crack surface density function
∫ ∫
δP = b∗ · δu dV + t∗ · δu dA (2.11a)
Ω ∂Ωt
[ ]
1 1 ′
δγ = α (d) δd + 2b∇d · ∇δd (2.11b)
c0 b
where we have introduced the following test and trial spaces
{ ⏐ } { ⏐ }
Uu := u⏐u(x) = u∗ ∀x ∈ ∂Ωu , Vu := δu⏐δu(x) = 0 ∀x ∈ ∂Ωu (2.12a)
{ ⏐ } { ⏐ }
Ud := d ⏐d(x) ∈ [0, 1], ḋ(x) ≥ 0 ∀x ∈ B , Vd := δd ⏐δd(x) ≥ 0 ∀x ∈ B (2.12b)

Note that the boundedness d(x) ∈ [0, 1] and irreversibility condition ḋ(x) ≥ 0 have to be dealt with carefully in
the numerical implementation; see [77] for details. Pre-defined cracks S can be accounted for either indirectly by
Dirichlet condition d(x) = 1 for x ∈ S or directly by mesh discretization. In this work, the later strategy is adopted
due to its flexibility.

3. Phase-field regularized cohesive zone model (CZM)

In this section the phase-field regularized cohesive zone model (CZM), developed within the unified phase-field
damage theory [76], is recalled. Compared to the other phase-field models [50,53,72,86], its numerical results are
independent of the incorporated length scale parameter under both brittle [88] and cohesive fracture [77,78], all
subjected to static loading regimes; see [59] for the details.

3.1. A hybrid phase-field damage formulation

For a cracking solid, it is assumed that the initial strain energy density ψ0 is isotropically deteriorated by the
phase-field d, i.e.,
1 1
ψ(ϵ, d) = ω(d)ψ0 (ϵ), ψ0 (ϵ) = ϵ : E0 : ϵ = σ̄ : C0 : σ̄ (3.1)
2 2
1006 V.P. Nguyen, J.-Y. Wu / Comput. Methods Appl. Mech. Engrg. 340 (2018) 1000–1022

where σ̄ = E0 : ϵ denotes the classical effective stress tensor, with E0 and C0 being the linear elastic stiffness
and compliance tensors, respectively; ω(d) is an energetic degradation function of the damage variable d, with
ω′ (d) = ∂ω/∂d ≤ 0.
In accordance with the constitutive relations (2.6), the resulting stress tensor σ and the associated damage driving
force Y are expressed as
∂ ∂ψ0
⎧
⎨σ :=
⎪ ψ(ϵ, d) = ω(d)σ̄ with σ̄ := = E0 : ϵ
∂ϵ ∂ϵ (3.2)
⎩Y := − ∂ ψ(ϵ, d) = −ω′ (d)Ȳ
⎪
with Ȳ :=
∂ψ 1
= ϵ : E0 : ϵ
∂d ∂ω 2
where Ȳ is the effective damage driving force conjugate to the energetic degradation function ω(d) of the damage
variable d, with ω′ (d) = ∂ω/∂d ≤ 0. The above constitutive relations give identical tensile and compressive fracture
behavior, and therefore, do not apply to the modeling of failure in brittle and quasi-brittle solids.
Asymmetric tension/compression cracking behavior can also be captured by using the so-called split of strain
energy density [86,89]; see [58,59,69] for more discussion on this topic. Alternatively, a hybrid formulation is
presented herein to capture the asymmetric tension/compression cracking behavior, without losing the linearity of
the displacement sub-problem in the alternating minimization solution algorithm [58]. In this hybrid formulation,
asymmetric tension/compression behavior is obtained by using a modified effective driving force Ȳ based on the
well-known Rankine criterion [76,78]
1
Ȳ = ⟨σ̄1 ⟩2 (3.3)
2E 0
where E 0 is Young’s modulus of the material; σ̄1 denotes the first major principle value of the effective stress tensor
σ̄ ; Macaulay brackets ⟨·⟩ are defined as ⟨x⟩ = max(x, 0).
The above re-definition results in a hybrid phase-field damage model that can capture the asymmetric ten-
sile/compressive behavior while the other constitutive relations (3.2) still apply. Other similar hybrid formulations
can be found in e.g., [90,91].

3.2. Optimal characteristic functions

The above phase-field damage model is characterized by two functions namely, the geometric crack function α(d)
regularizing the sharp crack topology and the energetic degradation function ω(d) defining the free energy density
potential. Various expressions for them yield different phase-field models appeared in the existing literature, exhibiting
distinct performances; see [59] for the details.
In accordance with Wu [76], the following constitutive functions optimal for cohesive fracture are considered in
this work
( )p
1−d
α(d) = 2d − d , 2
ω(d) = ( )p (3.4a)
1 − d + a1 d · P(d)
where the polynomial P(d) is given by

P(d) = 1 + a2 d + a2 a3 d 2 (3.4b)
for the exponent p ≥ 2 and the coefficients a1 > 0, a2 , a3 to be determined. For a given traction–separation law σ (w)
with the failure strength f t , the initial slope k0 and ultimate crack opening wc , it follows that [78]
G f 2/3 (
( )
4 1)
a1 = lch , a2 = 2 −2k0 2 − p+ (3.5a)
πb ft 2
p>2
⎧
⎨0 [
a 3 = 1 1 wc f t 2 (
]
(3.5b)
( ) )
⎩ − 1 + a2 p=2
a2 8 G f
for the so-called (Griffith’s or Irwin’s) internal length lch := E 0 G f / f t2 characterizing the length of the fracture process
zone (FPZ) ahead of the crack tip. Compared to the other phase-field models [50,72,86] in which the length scale b
 V.P. Nguyen, J.-Y. Wu / Comput. Methods Appl. Mech. Engrg. 340 (2018) 1000–1022 1007

Fig. 2. General softening curves predicted from the proposed phase-field damage model. (a) Linear softening law. (b) Cornelissen’s softening law.

is determined using the failure strength f t , herein one uses f t to determine a1 with b chosen independently as long as
the latter is sufficiently small; see Remark 3.1.
With the above characteristic functions, the unified phase-field damage theory particularizes into a regularized
CZM with a general traction–separation law σ (w). That is, for a constant failure strength f t , though the parameter
a1 and the resulting energetic function ω(d) both depend on the length scale b, the resulting stress versus separation
relation σ (w) across the localization band B is independent of it. In particular, those softening law frequently adopted
for cohesive fracture, e.g., the linear, exponential, hyperbolic and Cornelissen et al. [79] ones, etc., can be reproduced
or approximated with sufficient precision. For instance, the following softening curves are considered later in this
work:

• Linear softening curve: p = 2, a2 = − 21 and a3 = 0

( )
ft
σ (w) = f t max 1 − w, 0 (3.6)
2G f
• Cornelissen et al. [79] softening curve for normal concrete: p = 2, a2 = 1.3868 and a3 = 0.6567
[( )]
σ (w) = f t 1.0 + η13r 3 exp −η2r − r 1.0 + η13 exp −η2
) ( ) ( ) (
(3.7)
for the normalized crack opening r := w/wc as well as the parameters η1 = 3.0 and η2 = 6.93.

Note that the parameter a1 is still related to the length scale parameter b through Eq. (3.5a)1 . The predicted
softening curves are compared in Fig. 2 against the objective ones. As can be seen, the linear softening curve is
exactly reproduced, while the Cornelissen et al. [79] one is captured quite accurately. Most importantly, the fracture
energy, the initial slope and the ultimate crack opening are all precisely accounted for.

Remark 3.1. In order to guarantee the convexity of the energetic degradation function (3.4)2 and of the resulting free
energy functional (3.1), the length scale b should satisfy [77]
3 8
a1 ≥ ⇐⇒ b≤ lch ≈ 0.85lch (3.8)
2 3π
For quasi-static fracture, length scale parameters larger than this might result in numerical convergence issues. Usually,
a sufficiently small length scale parameter b ≪ lch is selected such that the sharp crack surface can be approximated
with sufficient precision. For quasi-brittle materials such as concrete with E 0 = 40 GPa, f t = 4.0 MPa, G f = 80.0
J/m2 , one has lch ≈ 200 mm and b ≤ 170 mm which is too large for specimens of laboratory scales. Therefore for
quasi-brittle materials, the upper bound (3.8) is virtually useless. Contrariwise, for brittle materials with these material
properties E 0 = 32 GPa, the failure strength f t = 12 MPa and the fracture energy G f = 3.0 J/m2 , one has lch ≈ 0.67
mm, resulting in an upper bound for b of 0.57 mm which is usually small enough. Note that the bound (3.8) applies
1008 V.P. Nguyen, J.-Y. Wu / Comput. Methods Appl. Mech. Engrg. 340 (2018) 1000–1022

only to static loading. With the addition of kinematic energy, it might be relaxed. We have found that the length scale
b ≈ (1/50 ∼ 1/100)L, where L is the smallest length of the structure, usually gives good results. An upper bound for
the length scale parameter b under dynamic fracture needs further investigation and will be addressed elsewhere. □

4. Numerical implementation
In this section, the governing equations in weak form (2.10) are numerically implemented using the so-called
multi-field FEM. Precisely, multi-field finite elements here refer to those where nodal unknowns consist of both the
displacement degrees of freedom (dofs) and the damage (phase-field) dof.

4.1. Finite element discretization

The computational domain Ω is triangulated by the mesh T h , consisting of n e elements and n p element nodes.
The super-index h indicates the typical mesh size of the finite element domain T h . It is necessary for the element size
h within the localization band B h to be much smaller than the length scale b such that an accurate estimation of the
fracture energy can be guaranteed in the discrete context [53].
Upon the above setting, the displacement field uh and crack phase-field d h are interpolated as
∑ ∑
uh (x) = N I (x) a I = Na, d h (x) = N̄ I (x) ā I = N̄ā (4.1)
I I

where the interpolation matrix N := N1 I, . . . , N I I, . . . is associated with the nodal displacement vector a :=
[ ]
}T }T
a1 , . . . , a I , . . . and N̄ := N1 , . . . , N I , . . . with the nodal damage dofs ā := ā1 , . . . , ā I , . . . , respectively.
{ [ ] {

Note that identical interpolation functions N I (x) are generally used in the approximation of the displacement and
damage fields. The resulting strain field ϵ h and damage gradient field ∇d h are given by
∑ ∑
ϵ h (x) = B I (x) a I = Ba, ∇d h (x) = B̄ I (x) ā I = B̄ā (4.2)
I I

B1 , . . . , B I , . . . and gradient vector B̄ :=

[ ]
where in 2-D cases, the standard displacement–strain matrix B :=
B̄1 , . . . , B̄ I , . . . have the following components
[ ]

∂x N I
⎡ ⎤
0
∂ N̄
{ }
B I (x) = ⎣ 0 ∂y N I ⎦ , B̄ I (x) = x I (4.3)
∂ y N̄ I
∂ y N I ∂x N I
associated with the node I of elements within the localization band B.
With the above finite element discretization, the governing equations (2.10) result in
∫
ru := fext − BT σ dV − Mä = 0 (4.4a)
Ω
∫ [ ( ]
1 ) 2b
rd := − N̄T ω′ Ȳ + α ′ Gf + G f B̄T ∇d dV ≤ 0 (4.4b)
B c 0 b c0

where ä = d2 a/dt 2 denote the nodal accelerations and M = Ω NT ρNdV represents the consistent mass matrix,
∫

fext is the standard external force vector [8]. Note that above discrete governing equations hold upon the damage
boundedness d ∈ [0, 1] and the irreversibility conditions ḋ ≥ 0.
For the phase-field damage models [50,86] with a quadratic geometric crack function α(d) = d 2 , the damage
boundedness d ∈ [0, 1] is intrinsically guaranteed. And the irreversibility conditions ḋ ≥ 0 can be enforced as in [92]
in which the damage driving force Ȳ in (3.2) is replaced by a history field variable that represents the maximum value
of Ȳ ever reached. However, regarding non-standard phase-field models as the one considered in this work, even if the
irreversibility condition may be similarly tackled, the damage boundedness cannot be easily dealt with. To address this
issue, Amor [93] proposed regarding the governing Eq. (4.4b) as a mixed complementarity problem [94] which can
be then solved by a bound-constrained solver included in, e.g., the Matlab Optimization Toolbox or the SNES solver
of the open source toolkit PETSc [95]. The interested read is referred to [77,96] for more discussion. In this work, the
later solver is adopted and problems up to 6.1 million dofs are considered on a high performance workstation
 V.P. Nguyen, J.-Y. Wu / Comput. Methods Appl. Mech. Engrg. 340 (2018) 1000–1022 1009

4.2. Explicit dynamics solver

The system of nonlinear equations (4.4) is in general solved in an incremental procedure. That is, during the
time interval [0, T ] of interest all the state variables are considered at the discrete interval [tn , tn+1 ] for n =
0, 1, 2, . . . , N − 1. For a typical time increment [tn , tn+1 ] of length ∆t := tn+1 − tn , the system of nonlinear equations
(4.4) is solved either iteratively in implicit dynamics or directly in explicit one with all state variables known at the
instant tn .
In order to solve the dynamic equilibrium equation (4.4a), the Newmark integration [97] is adopted in this work
due to its popularity in the literature. Given a time step size ∆t the nodal displacements an+1 and velocities ȧn+1 at
the instant tn+1 are approximated as
(∆t)2 [( ) ]
an+1 = an + ∆t ȧn + 1 − 2β än + 2β än+1 (4.5a)
[( 2 ]
ȧn+1 = ȧn + ∆t 1 − γ än + γ än+1
)
(4.5b)
for the parameters β ∈ [0, 0.5] and γ ∈ [0, 1]. The case β = 1/4 and γ = 1/2 corresponds to the trapezoidal rule,
whereas for β = 0 and γ = 1/2 an explicit central difference method is obtained. Improved schemes like the HHT-α
method [98] have also been proposed to suppress high frequency numerical oscillations.

Though the implicit Newmark method (i.e., β ̸= 0) can be used as well, the explicit method with β = 0 is
considered in this work. The merit is that the resulting time evolution system is automatically decoupled such that the
sub-problems separately in an+1 and ān+1 can be solved in a sequence at every time step. At the very beginning instant
t0 , the initial accelerations ä0 and damage dofs ā0 are computed based on a priori boundary and initial conditions.
Subsequently, the nodal displacements an+1 are directly updated from Eq. (4.5a), i.e.,
(∆t)2
an+1 = an + ∆t ȧn + än (4.6)
2
Then the crack phase-field variables ān+1 are solved via Eq. (4.4b) by the bound-constrained optimization solver;
see [77]. The nodal accelerations än+1 are readily solved from the governing Eq. (4.4a) with the lumped mass
matrix M̄ (in order to suppress the noises of high frequency), and the nodal velocities ȧn+1 are then updated from
Eq. (4.5b) with a specific value γ (e.g., γ = 1/2 for a middle-step velocity). The resulting procedure is summarized in
Algorithm 1.
Note that for explicit dynamics, only the one-pass alternating minimization procedure is needed due to very small
time steps adopted [67,69].

4.3. Determination of crack tip position

For dynamic fracture simulations, it is necessary to compute the crack velocity in time. To this end, the location of
the crack tip has to be determined. The discussion here is restricted to 2D problems and cracks grow in the positive
x-direction. The crack tip is defined using the iso-curves of the phase-field [67,99], cf. Fig. 3.
The input are the initial crack tip’s coordinates (x0 , y0 ) and the phase-field threshold dmax , e.g., dmax = 0.8. The
procedure is as follows
• find the nodes I that have its phase-field ā I > dmax ;
• find the coordinates of the nodes I ;
• the crack tip is defined as the node having the maximum x-coordinate among the nodes I ;
Note that this algorithm of finding the location of the crack tip also applies to other continuous models and even to
peridynamic models [46].
The coordinates of the crack tip are written to a file, of which one example is given in Table 1. The first three
columns correspond to the raw data from which one can calculate the crack length at a given time step. Having this
data, the crack speed is computed from a linear fitting to three points of the (tn+1 , an+1
cr
) curve [26]; that is, the crack
speed at time step tn+1 is the slope of the line that best fits, in a least square sense, (tn−1 , an−1
cr
), (tn+1 , an+1
cr
) and
(tn+1 , an+1
cr
).
1010 V.P. Nguyen, J.-Y. Wu / Comput. Methods Appl. Mech. Engrg. 340 (2018) 1000–1022

Algorithm 1: Explicit dynamics for the phase-field damage model

Data: a0 , ȧ0
Result: ā0 , ä0
Initialization
Compute the crack phase-field ā0 for the initial nodal displacements a0
∫ [ ( ]
1 ) 2b
rd (ā0 ) := − N̄T ω′ Ȳ + α ′ Gf + G f B̄T ∇d dV ≤ 0
B c0 b c0
with ω := ω (ā0 ), α := α (ā0 )
′ ′ ′ ′

subjected to: 0 ≤ ā I 0 ≤ 1 ∀I = 1, 2, . . .
Compute the nodal acceleration ä0
[ ∫ ]
−1 ext
ä0 = M̄ f0 − B σ 0 dV
T
with σ 0 = σ (a0 , ā0 )
Ω

using the lumped mass matrix M̄

for every successive time step n + 1 do
Update the nodal displacements an+1 and apply the Dirichlet boundary conditions
(∆t)2
an+1 = an + ∆t ȧn + än
2
Compute crack phase-field dofs ān+1 with the updated nodal displacements an+1
∫ [ ( ]
1 ′ ) 2b
rd (ān+1 ) := − N̄ ω Ȳ +
T ′
α Gf + T
G f B̄ ∇d dV ≤ 0
B c0 b c0
with ω′ := ω′ (ā(k)
n+1 ), α ′ := α ′ (ān+1 )
subjected to: 0 ≤ ā I n ≤ ā I n+1 ≤ 1 ∀I = 1, 2, . . .
Compute the nodal accelerations än+1 and velocities ȧn+1
[ ∫ ]
−1 ext
än+1 = M̄ fn+1 − B σ dV
T
with σ = σ (an+1 , ān+1 )
Ω

= ȧn + ∆t 1 − γ än + γ än+1

[( ) ]
ȧn+1
Apply the Dirichlet boundary conditions for the nodal velocities ȧn+1 .
end

Fig. 3. Iso-curves of different values of φ.

 V.P. Nguyen, J.-Y. Wu / Comput. Methods Appl. Mech. Engrg. 340 (2018) 1000–1022 1011

Table 1
Exemplary data for the crack tip. The crack increment for time step n
cr cr
is denoted by ∆an+1 and the (total) crack length at time step an+1 is
cr
= m≤n ∆amcr .
∑
computed as an+1
cr cr
tn+1 [µs] xn+1 [mm] yn+1 [mm] ∆an+1 [mm] an+1 [mm]
0.000 50 25 0.0 0.0
23.425 50 25.5 0.5 0.5
23.925 50 25.5 0.0 0.5
24.425 50 25.5 0.0 0.5
24.925 50 26.5 1.0 1.5

5. Numerical examples
This section presents three numerical examples on dynamic fracture of both brittle and quasi-brittle solids. The first
two include a crack branching test and the Kalthoff and Winkler [100] experiment which are both benchmarks in the
literature of computational dynamic fracture of brittle solids. The linear softening law (3.6) is employed as suggested
in [88]. The third one concerns the mixed-mode fracture of a concrete beam under impact loadings, in which the
Cornelissen et al. [79] softening curve (3.7) is used. In all numerical examples presented in this section, unstructured
and uniform piece-wise linear triangular elements are used to discretize the computation domain. For all cases, a
plane stress state is assumed, and the explicit time integration is used to advance in time. Note that all the considered
problems are concerned with brittle/quasi-brittle materials subjected to impact loadings. As the involved time scale is
typically of order O(1 ms), the explicit Newmark scheme with a lumped mass matrix, which is second-order accurate
and energy conservative, is very suitable. In order to circumvent the issue of conditional stability, the time increment
∆t is determined by the Courant–Friedrichs–Lewy √ (CFL) condition ∆t < ∆tCFL = 0.8h min /c, where h min is the
minimum mesh size among all elements, c := µ0 /ρ is the material sound speed, and 0.8 is the security factor.
This scheme is acceptable since for problems involving extensive material nonlinearities even unconditionally stable
implicit schemes need a small time increment comparable to ∆tCFL .
There are three length scales in our model: the mesh size h, the length scale parameter b and the internal length
lch , which should be h ≪ b ≪ lch . For brittle solids, the internal length lch is usually small, and so one needs very
small mesh size (based on our experiences, h ≤ b/5 is recommended). Large problems were solved in parallel using
multiple computers. Even though our model and its implementation are inherently three dimensions — 3D simulations
can be carried out without any change in the code, in this manuscript only 2D simulations are reported. We anticipate
that for 3D problems, the post-processing of the crack velocities would be harder since one has to deal with a diffuse
crack front instead of a crack tip.

Remark 5.1. As the time increment ∆tCFL used in the simulations is related automatically to the smallest mesh size,
various mesh sizes used in the simulations lead to different increment sizes. Therefore, as long as the numerical results
are independent of the mesh size, they are also insensitive to the resulting time increment size. □

5.1. Dynamic crack branching

In this example, we consider a pre-cracked block loaded dynamically in tension. As shown in Fig. 4, a uniform
traction σ0 is applied at the top and bottom edges of the specimen as a step function in time. This problem
has been widely adopted to study dynamic crack branching; see the experimental tests reported in [101,102]. In
the literature it has been numerically addressed by many authors using different methods such as XFEM and
cohesive interface elements, see e.g., [21,26,103,104]. In the phase-field community, this problem has been analyzed
by Borden et al. [67], Liu et al. [105], Schlüter et al. [63] using the standard phase-field model with α(d) = d 2 and
by )2 et al. [66], Li et al. [69] using the non-standard one with α(d) = d, all for brittle fracture with ω(d) =
( Bleyer
1−d .
This example is herein numerically analyzed using the presented phase-field regularized CZM with linear
softening. The material parameters used in the simulations are taken from [103]: Young’s modulus E 0 = 3.2 ×
1012 V.P. Nguyen, J.-Y. Wu / Comput. Methods Appl. Mech. Engrg. 340 (2018) 1000–1022

Fig. 4. Dynamic crack branching: geometry of the edge cracked block (left) and the loading history (right). Herein, σ0 = 1.0 MPa is used. The
notch has a width of 0.5 mm.

104 MPa, Poisson’s ratio of ν0 = 0.2, the mass density of ρ = 2450 kg/m3 , the failure strength f t = 12 MPa and the
fracture energy G f = 3.0 J/m2 , resulting in Griffith’s internal length lch = 0.67 mm and Rayleigh’s wave speed c R =
2119 m/s. In order to investigate the influences of the length scale b and the mesh size h, the following three cases are
considered:
• Case (a): b = 0.50 mm and h = 0.10 mm (i.e., h = b/5);
• Case (b): b = 0.50 mm and h = 0.05 mm (i.e., h = b/10);
• Case (c): b = 0.25 mm and h = 0.05 mm (i.e., h = b/5).
where the length scale parameter b satisfies the upper bound given in Eq. (3.8).
The crack patterns at the end of the simulation (t = 84 × 10−6 s) obtained with the above three cases are given in
Fig. 5. The crack propagates from the notch to the right with increasing speed. At a certain point, the crack branches
into two cracks which agrees well with the experiment [101]. The widening of the crack right before the moment of
branching is similar to other phase-field simulations reported in [67–69,105], to non-local integral damage model [35]
and to peridynamics simulations [46]. This damage widening could be the signal of roughening of the crack surface
experimentally observed to occur prior to branching [101].
The global responses of the model are measured by the stored strain energy, dissipated surface energy and crack
tip velocity. The results are shown in Fig. 6. Dynamic crack branching occurs around 36.5 µs and the crack velocity
never exceeds 0.6v R , which is in good agreement with the results given by other phase-field models [67,105].
In all three cases considered above, the crack localization bandwidth is proportional to the incorporated length
scale parameter b. However, the latter affects neither the crack pattern nor the global responses, and so does the
mesh size: h = b/5 is sufficient to resolve the damage field. This result further consolidates our previous findings
that our model is insensitive to the length scale parameter for both cohesive failure [76] and brittle fracture [88]
under quasi-static loading. It is also the case for dynamic fracture. Contrariwise, Steinke et al. [68] reported that the
numerical predictions of the standard phase-field model depend apparently on the incorporated length scale parameter
and the limiting case of b → 0 cannot be achieved.
In order to test whether the proposed phase-field regularized CZM can capture multiple crack branches, we
reconsider Case (c) i.e., b = 0.25 mm and h = 0.05 mm but with σ0 = 2.5 MPa. The result given in Fig. 7 shows that
crack branching happens earlier than σ0 = 1.0 MPa, the crack angle is also smaller and there are multiple branches.
All these are in good agreement with the findings reported in [35,46,47] using other methods. The crack patterns
at various instants are shown in Fig. 8. As can be seen, the evolution of multiple crack branches can be seamlessly
captured by the proposed model. There exists damage widening prior to branching events for later branches. Similar
results were presented in [106] with a peridynamics model.

5.2. Edge-cracked plate under impulsive loading

This section studies a doubly notched specimen under an impact load, which is often referred to as the Kalthoff–
Winkler experiment. The geometry of the specimen is shown in Fig. 9, and the impact loading is applied by a projectile.
 V.P. Nguyen, J.-Y. Wu / Comput. Methods Appl. Mech. Engrg. 340 (2018) 1000–1022 1013

Fig. 5. Dynamic crack branching: numerical crack patterns at time 84 µs. (a) b = 0.5 mm and h = 0.10 mm. (b) b = 0.5 mm and h = 0.05 mm.
(c) b = 0.25 mm and h = 0.05 mm.

In the experiment [100], two different failure modes were observed by modifying the projectile speed, v0 ; at high
impact velocities (50 m/s and 100 m/s, respectively, were reported in [107] and [69] in which different phase-field
models were considered), a shear band is observed to emanate from the notch at an angle of 10◦ with respect to the
initial notch and at lower strain rates (v0 = 16.54 m/s), brittle failure with a crack propagation angle of about 70◦
1014 V.P. Nguyen, J.-Y. Wu / Comput. Methods Appl. Mech. Engrg. 340 (2018) 1000–1022

Fig. 6. Dynamic crack branching: strain energy, surface energy and crack tip velocities in time. (a) Stored strain energy. (b) Dissipated surface
energy. (c) Crack tip velocity.

Fig. 7. Dynamic crack branching: influence of the load intensity. The higher the load amplitude the more branches are present. Crack branching
also happens sooner and the crack angle (of the first branch) is smaller.

is observed. We are interested only in the velocity range that resulted in a brittle failure mode. If the transition of
brittle-ductile failure modes is of interest, one may need to consider the plastic behavior in the phase field model as
suggested in [108], which is, however, beyond the scope of this work. The simulations of this problem have been
reported by several researchers, see e.g., [26,30,103,109,110] using different methods such as XFEM and adaptive
 V.P. Nguyen, J.-Y. Wu / Comput. Methods Appl. Mech. Engrg. 340 (2018) 1000–1022 1015

Fig. 8. Dynamic crack branching: crack patterns at various instants for the traction σ0 = 2.5 MPa. (a) t = 25.1 µs. (b) t = 40.2 µs. (c) t = 47.7 µs.
(d) t = 61.9 µs. (e) t = 84.0 µs.

cohesive interface elements. In the phase-field community this problem has been analyzed by Borden et al. [67], Li et
al. [69], Liu et al. [105] and Hofacker and Miehe [64,65].
The material parameters of Maraging steel 18Ni(300) are taken from [103]: Young’s modulus E 0 = 1.9×105 MPa,
Poisson’s ratio ν0 = 0.3, the mass density ρ = 8000 kg/m3 , the failure strength f t = 2812.25 MPa and the
fracture energy G f = 22.2 N/mm, resulting in Griffith’s internal length lch = 0.53 mm. The Rayleigh wave speed
is c R = 2745 m/s. Two cases are considered to study the sensitivity of the results with respect to the length scale
parameter b:
• b = 0.50 mm and h = 0.10 mm;
• b = 0.25 mm and h = 0.05 mm.
Due to symmetry only the upper half was modeled at the impact velocity v0 = 16.54 m/s.
The notch is introduced in the mesh a geometry discontinuity following [69] as phase-field induced crack
did not result in crack initiation correctly. The crack pattern is shown in Fig. 10 which agrees with available
◦
findings [67,69,105]. Initially the crack starts to grow at a larger angle (about 70 ) then the angle decreases as the
crack grows. The average angle from the initial crack tip to the point where the crack intersects the top boundary is
◦
about 68 and in fairly good agreement with the experimental result.
The global responses of the stored strain energy, dissipated surface energy and crack tip velocity are shown in
Fig. 11. Again, the responses are independent of the length scale and the mesh sizes. The final crack length, Fig. 11(d),
is about 102 mm (corresponding to an energy dissipation of 22.2×102 = 2264.4 J for a sharp crack) and the calculated
surface energy is 2460 J which is 8.6% over-estimated due to the discretization error. If we assume that the crack is
straight, then with the angle of 68◦ , its length would be 81 mm and its surface energy would be 1796 J. Our value
is about 37% higher, but is significantly better than the value reported in [67] (90% over-estimated) who used a
phase-field model without an elastic regime.
1016 V.P. Nguyen, J.-Y. Wu / Comput. Methods Appl. Mech. Engrg. 340 (2018) 1000–1022

Fig. 9. Edge-cracked plate under impulsive loading.

Fig. 10. Edge-cracked plate under impulsive loading: Numerical crack patterns at time 90 µs. Left (b = 0.50 mm and h = 0.10 mm) and right
(b = 0.25 mm and h = 0.05 mm).

Next we consider a larger impact velocity v0 = 100 m/s [69] using b = 0.25 mm and h = 0.05 mm. The mesh
consists of about two million nodes and four million elements. Successive crack branching can be observed from
Fig. 12. Again, there exists damage widening prior to all branching events. Note that large elements are used at the
lower right corner (to reduce computational cost) and thus large damage bands occur at this region. Similar findings are
reported in [64,65,69]. So, with sufficiently refined meshes, our model can capture multiple dynamic crack branchings
quite straightforwardly. This result is promising for studying dynamic crack instability problem where there are many
very small cracks along a major dominant one [111]. We recall that in the experiment a failure-mode transition from
mode-I to mode-II was observed when the impact velocity increases. This cannot be captured using an elastic-damage
model [69].

5.3. John–Shah test

We finally consider the three-point bending concrete beams subjected to impact loading [112]. The problem
configuration is given in Fig. 13, where an offset notch from the midspan (with varying locations depending on
the dimensionless parameter γ ) is made to study mixed-mode fracture. John and Shah [112] observed different crack
patterns for various γ and interestingly there exists a transition value γt that defines a change in failure mode — crack
grows from the notch for γ < γt and from the midspan for γ > γt (Fig. 14). The experimentally determined value
of γt is 0.77. This test has been numerically studied by many researchers, for example [113] with the Element Free
Galerkin method, [114,115] with extrinsic cohesive interface elements, [116] with XFEM among others. Different
 V.P. Nguyen, J.-Y. Wu / Comput. Methods Appl. Mech. Engrg. 340 (2018) 1000–1022 1017

Fig. 11. Edge-cracked plate under impulsive loading: strain energy, surface energy, crack tip velocity and crack length. (a) Stored strain energy.
(b) Dissipated surface energy. (c) Crack tip velocity. (d) Crack length.

Fig. 12. Edge-cracked plate under impulsive loading with a large impact velocity of 100 m/s: numerical crack patterns. (a) t = 40 µs. (b) t = 53 µs.

authors reported slightly different values of γt : γt = 0.6 in [114], γt = 0.635 in [116], γt ∈ {0.75, 0.8} in [115] and
γt = 0.734 in [113]. This is not surprising as they have used different material properties, particularly the tensile
strength and the fracture energy. Herein we study this example using the presented phase-field regularized CZM
with a plane stress condition. Material properties are taken from the aforementioned references: Young’s modulus
E 0 = 31.37 GPa, Poisson’s ratio ν0 = 0.2, the mass density ρ = 2400 kg/m3 , the tensile strength f t = 3.0 MPa and
the fracture energy G f = 31.1 J/m2 .
1018 V.P. Nguyen, J.-Y. Wu / Comput. Methods Appl. Mech. Engrg. 340 (2018) 1000–1022

Fig. 13. John–Shah test: geometry (Unit of length: mm) and loading. The notch width is 1.5 mm and t denotes the beam thickness.

Fig. 14. John–Shah test: notch-location dependent fracture mode [112].

The imposed velocity is given by [113,116]

{ v1
t if t ≤ t0
v(t) = t0 (5.1)
v1 otherwise
for t0 = 196 µs and v1 = 60 mm/s. These values approximately produce the measured strain rate of 0.3 s−1 in a linear
elastic analysis.
As mesh convergence and length scale{sensitivity have been extensively for the previous two problems, herein
we consider various notch locations γt = 0.50, 0.60, 0.65, 0.72, 0.78 using a single set of length scale parameter
}

b = 1.5 mm (≈ h/50) and mesh size h = b/5. Fig. 15 shows the corresponding numerical crack patterns. It is
obvious that the presented phase-field regularized CZM can capture the transition of the failure mode. Our estimate
for γt is somewhere in between 0.72 and 0.78 which is in good agreement with the experimental value. The obtained
crack patterns are similar to EFG results reported in [113], and to cohesive elements in [115]. However, as crack
nucleation can be intrinsically dealt with in the presented model, it is not necessary to introduce a small notch at the
beam mid-span as in [113,116].

6. Conclusions

We have extended the unified phase-field damage theory for quasi-static brittle/cohesive fracture presented
in [76,78,88] to the dynamic case. With some optimal characteristic functions the resulting model is equivalent to the
well-known cohesive zone model (CZM), at least in one dimension, and therefore, we have named it as a phase-field
regularized CZM. Based on a number of benchmark tests, it is confirmed that the phase-field regularized CZM gives
length scale independent numerical results for both brittle and cohesive fracture under dynamic loading when the
mesh is sufficiently fine (i.e., h ≤ b/5) to resolve the damage gradient in the localization band of small width.
Similar to existing phase-field models, ours is able to capture qualitatively the characteristics of dynamic fracture
such as multiple crack branching, damage widening prior to branching, crack velocities well below the Rayleigh wave
speed. Compared to existing phase-field models and CZM based discontinuous approaches to fracture, the presented
phase-field regularized CZM is of several advantages. Firstly, it does not need the cumbersome crack tracking and the
elastic penalty (dummy) stiffness necessary for discontinuous approaches (to e.g., handle crack lips interpenetration).
 V.P. Nguyen, J.-Y. Wu / Comput. Methods Appl. Mech. Engrg. 340 (2018) 1000–1022 1019

Fig. 15. John–Shah test: numerical crack patterns for various notch locations.

Secondly, it gives length scale independent global responses while preserving the favorable Γ -convergence property
of phase-field models.
The biggest complaint about phase-field models is their high computational cost. Our view to this point is two-
fold. First, it is obvious that problems that were intractable decades ago are now solved routinely. Second, phase-
field models seem to be the only continuous approach that can analyze complex 3D crack problems. Advancements
in computer hardware and computing technologies (anisotropic mesh adaptivity, global/local methods, multi-scale
methods, model order reduction methods) will definitely increase the efficiency of phase-field simulations.

Acknowledgments
The corresponding author (J.Y. Wu) acknowledges the support from the National Key R&D Program of China
(2017YFC0803300) and the National Natural Science Foundation of China (51678246). The first author (V.P. Nguyen)
thanks the funding support from the Australian Research Council via DECRA project DE160100577. Partial support
from the State Key Laboratory of Subtropical Building Science (2018ZC04) and the Scientific/Technological Project
of Guangzhou (201607020005) to J.Y. Wu is also acknowledged.

References
[1] B.N. Cox, H. Gao, D. Gross, D. Rittel, Modern topics and challenges in dynamic fracture, J. Mech. Phys. Solids 53 (3) (2005) 565–596.
[2] L.E. Malvern, Introduction to the Mechanics of a Continuous Medium, Prentice-Hall International, Englewood Cliffs, New Jersey, 1969.
[3] L. Freund, Dynamic Fracture Mechanics, Cambridge University Press, Cambridge, 1998.
[4] K. Ravi-Chandar, Dynamic Fracture, Elsevier, Amsterdam, 2004.
[5] J. Fineberg, E. Bouchbinder, Recent developments in dynamic fracture: Some perspectives, Int. J. Fract. 196 (1) (2015) 33–57.
[6] M. Ramulu, A.S. Kobayashi, Mechanics of crack curving and branching – a dynamic fracture analysis, Int. J. Fract. 27 (3) (1985) 187–201.
[7] O.C. Zienkiewicz, R.L. Taylor, The Finite Element Method for Solid and Structural Mechanics, sixth ed., Butterworth-Heinemann, Oxford,
UK, 2006.
[8] T. Hughes, The Finite Element Method, Prentice-Hall: Englewood Cliffs, NJ., 1987.
[9] T. Belytschko, Y. Krongauz, D. Organ, M. Fleming, P. Krysl, Meshless methods: An overview and recent developments, Comput. Methods
Appl. Mech. Engrg. 139 (1) (1996) 3–47.
[10] V.P. Nguyen, T. Rabczuk, S. Bordas, M. Duflot, Meshless methods: A review and computer implementation aspects, Math. Comput.
Simulation 79 (3) (2008) 763–813.
[11] T. Rabczuk, Computational methods for fracture in brittle and quasi-brittle solids: State-of-the-art review and future perspectives, ISRN
Appl. Math. (2013).
[12] A.A. Griffith, The Phenomena of rupture and flow in solids, Philos. Trans. R. Soc. Lond. 221 (1920) 163–198.
[13] G.R. Irwin, Analysis of stresses and strains near the end of a crack traversing a plate, J. Appl. Mech. 24 (1957) 361–364.
[14] D. Dugdale, Yielding of steel sheets containing slits, J. Mech. Phys. Solids 8 (1960) 100–109.
[15] G.I. Barenblatt, The formation of equilibrium cracks during brittle fracture. General ideas and hypotheses. Axially-symmetric cracks, J.
Appl. Math. Mech. 23 (1959) 622–636.
1020 V.P. Nguyen, J.-Y. Wu / Comput. Methods Appl. Mech. Engrg. 340 (2018) 1000–1022

[16] A. Hillerborg, M. Modéer, P. Petersson, Analysis of crack formation and crack growth in concrete by means of fracture mechanics and finite
elements, Cement Concr. Res. 6 (1976) 773–781.
[17] L. Kachanov, Time rupture process under creep conditions, Izv. A Rad. Nauk. SSSR otd Tekh. Nauk 8 (1958) 26–31.
[18] D. Krajcinovic, Damage Mechanics, Elsevier B.V., the Netherlands, 2003.
[19] T. Belytschko, H. Chen, J. Xu, G. Zi, Dynamic crack propagation based on loss of hyperbolicity and a new discontinuous enrichment,
Internat. J. Numer. Methods Engrg. 58 (12) (2003) 1873–1905.
[20] J. Réthoré, A. Gravouil, A. Combescure, A stable numerical scheme for the finite element simulation of dynamic crack propagation with
remeshing, Comput. Methods Appl. Mech. Engrg. 193 (42) (2004) 4493–4510.
[21] J. Song, H. Wang, T. Belytschko, A comparative study on finite element methods for dynamic fracture, Comput. Mech. 42 (2) (2008)
239–250.
[22] J.J.C. Remmers, R. d. Borst, A. Needleman, A cohesive segments method for the simulation of crack growth, Comput. Mech. 31 (1) (2003)
69–77.
[23] A. Combescure, A. Gravouil, D. Grégoire, J. Réthoré, X-FEM a good candidate for energy conservation in simulation of brittle dynamic
crack propagation, Comput. Methods Appl. Mech. Engrg. 197 (5) (2008) 309–318.
[24] F. Armero, C. Linder, Numerical simulation of dynamic fracture using finite elements with embedded discontinuities, Int. J. Fract. 160 (2)
(2009) 119–141.
[25] X. Xu, A. Needleman, Numerical simulations of fast crack growth in brittle solids, J. Mech. Phys. Solids 42 (9) (1994).
[26] V.P. Nguyen, Discontinuous Galerkin/Extrinsic cohesive zone modeling: Implementation caveats and applications in computational fracture
mechanics, Eng. Fract. Mech. 128 (2014) 37–68.
[27] K. Park, G.H. Paulino, W. Celes, R. Espinha, Adaptive mesh refinement and coarsening for cohesive zone modeling of dynamic fracture,
Internat. J. Numer. Methods Engrg. 92 (1) (2012) 1–35.
[28] F. Zhou, J.-F. Molinari, T. Shioya, A rate-dependent cohesive model for simulating dynamic crack propagation in brittle materials, Eng.
Fract. Mech. 72 (9) (2005) 1383–1410.
[29] D.W. Spring, G.H. Paulino, Achieving pervasive fracture and fragmentation in three-dimensions: An unstructuring-based approach, Int. J.
Fract. 210 (1) (2018) 113–136.
[30] J. Song, T. Belytschko, Cracking node method for dynamic fracture with finite elements, Internat. J. Numer. Methods Engrg. 77 (3) (2009)
360–385.
[31] R. Abedi, R.B. Haber, P.L. Clarke, Effect of random defects on dynamic fracture in quasi-brittle materials, Int. J. Fract. 208 (1) (2017)
241–268.
[32] R. Abedi, M.A. Hawker, R.B. Haber, K. Matous, An adaptive spacetime discontinuous Galerkin method for cohesive models of
elastodynamic fracture, Internat. J. Numer. Methods Engrg. 81 (10) (2006) 1207–1241.
[33] C. Wolff, N. Richart, J.-F. Molinari, A non-local continuum damage approach to model dynamic crack branching, Internat. J. Numer.
Methods Engrg. 101 (12) (2015).
[34] G. Pijaudier-Cabot, Z.P. Bažant, Nonlocal damage theory, J. Eng. Mech. 113 (1987) 1512–1533.
[35] L. Pereira, J. Weerheijm, L. Sluys, A numerical study on crack branching in quasi-brittle materials with a new effective rate-dependent
nonlocal damage model, Eng. Fract. Mech. (2017).
[36] J. Ožbolt, J. Bošnjak, E. Sola, Dynamic fracture of concrete compact tension specimen: Experimental and numerical study, Int. J. Solids
Struct. 50 (25) (2013) 4270–4278.
[37] Z.P. Bažant, B.H. Oh, Crack band theory for fracture of concrete, Mater. Struct. 16 (1983) 155–177.
[38] A. Karamnejad, V.P. Nguyen, L.J. Sluys, A multi-scale rate dependent crack model for quasi-brittle heterogeneous materials, Eng. Fract.
Mech. 104 (2013) 96–113.
[39] R.H.J. Peerlings, R. De Borst, W.A.M. Brekelmans, J.H.P. De Vree, Gradient enhanced damage for quasi-brittle materials, Internat. J. Numer.
Methods Engrg. 39 (19) (1996) 3391–3403.
[40] S.A. Silling, Reformulation of elasticity theory for discontinuities and long-range forces, J. Mech. Phys. Solids 48 (2000) 175–209.
[41] S.A. Silling, M. Epton, O. Weckner, J. Xu, A. Askari, Peridynamics states and constitutive modeling, J. Elasticity 88 (2007) 151–184.
[42] H.L. Ren, X.Y. Zhuang, Y.C. Cai, T. Rabczuk, Dual-horizon peridynamics, Internat. J. Numer. Methods Engrg. 108 (12) (2016) 1451–1476.
[43] H.L. Ren, X.Y. Zhuang, T. Rabczuk, Dual-horizon peridynamics: A stable solution to varying horizons, Comput. Methods Appl. Mech.
Engrg. 318 (2018) 762–782.
[44] S.A. Silling, E. Askari, A meshfree method based on the peridynamic model of solid mechanics, Comput. Struct. 83 (17–18) (2005)
1526–1535.
[45] G.C. Ganzenmuller, S. Hiermaier, M. May, On the similarity of meshless discretizations of Peridynamics and Smooth-Particle Hydrody-
namics, Comput. Struct. 150 (2015) 71–78.
[46] Y.D. Ha, F. Bobaru, Studies of dynamic crack propagation and crack branching with peridynamics, Int. J. Fract. 162 (1) (2010) 229–244.
[47] F. Bobaru, G. Zhang, Why do cracks branch? A peridynamic investigation of dynamic brittle fracture, Int. J. Fract. 196 (1) (2015) 59–98.
[48] G. Francfort, J.-J. Marigo, Revisiting brittle fracture as an energy minimization problem, J. Mech. Phys. Solids 46 (8) (1998) 1319–1342.
[49] I.S. Aranson, V.A. Kalatsky, V.M. Vinokur, Continuum field description of crack propagation, Phys. Rev. Lett 85 (2000) 118–121.
[50] B. Bourdin, G. Francfort, J.-J. Marigo, Numerical experiments in revisited brittle fracture, J. Mech. Phys. Solids 48 (4) (2000) 797–826.
[51] P. Areias, M.A. Msekh, T. Rabczuk, Damage and fracture algorithm using the screened Poisson equation and local remeshing, Eng. Fract.
Mech. 158 (2016) 116–143.
[52] P. Areias, J. Reinoso, P.P. Camanho, J. César de Sá, T. Rabczuk, Effective 2D and 3D crack propagation with local mesh refinement and the
screened Poisson equation, Eng. Fract. Mech. 189 (2018) 339–360.
[53] B. Bourdin, G. Francfort, J.-J. Marigo, The Variational Approach to Fracture, Springer, Berlin, 2008.
 V.P. Nguyen, J.-Y. Wu / Comput. Methods Appl. Mech. Engrg. 340 (2018) 1000–1022 1021

[54] P. Areias, T. Rabczuk, M.A. Msekh, Phase-field analysis of finite-strain plates and shells including element subdivision, Comput. Methods
Appl. Mech. Engrg. 312 (2016) 322–350.
[55] J. Reinoso, M. Paggi, Revisiting the problem of a crack impinging on an interface: A modeling framework for the interaction between the
phase field approach for brittle fracture and the interface cohesive zone model, Comput. Methods Appl. Mech. Engrg. 321 (2017) 145–172.
[56] V. Carollo, J. Reinoso, M. Paggi, A 3D finite strain model for intralayer and interlayer crack simulation coupling the phase field approach
and cohesive zone model, Compos. Struct. 182 (2017) 636–651.
[57] M.A. Msekh, N.H. Cuong, G. Zi, P. Areias, X. Zhuang, T. Rabczuk, Fracture properties prediction of clay/epoxy nanocomposites with
interphase zones using a phase field model, Eng. Fract. Mech. 188 (2018) 287–299.
[58] M. Ambati, T. Gerasimov, L. de Lorenzis, A review on phase-field models for brittle fracture and a new fast hybrid formulation, Comput.
Mech. 55 (2015) 383–405.
[59] J.Y. Wu, V.P. Nguyen, C.T. Nguyen, D. Sutula, S. Bordas, S. Sinaie, Phase field modeling of fracture, Adv. Appl. Mech.: Multi-scale Theory
Comput. 52 (2018) submitted for publication.
[60] R. Spatschek, E. Brener, A. Karma, Phase field modeling of crack propagation, Phil. Mag. 91 (2011) 75–95.
[61] C.J. Larsen, C. Ortner, E. Sali, Existence of solutions to a regularized model of dynamic fracture, Math. Models Methods Appl. Sci. 20 (07)
(2010) 1021–1048.
[62] B. Bourdin, C.J. Larsen, C.L. Richardson, A time-discrete model for dynamic fracture based on crack regularization, Int. J. Fract. 168 (2)
(2011) 133–143.
[63] A. Schlüter, A. Willenbücher, C. Kuhn, R. Müller, Phase field approximation of dynamic brittle fracture, Comput. Mech. 54 (5) (2014)
1141–1161.
[64] M. Hofacker, C. Miehe, Continuum phase field modeling of dynamic fracture: Variational principles and staggered FE implementation, Int.
J. Fract. 178 (1) (2012) 113–129.
[65] M. Hofacker, C. Miehe, A phase field model of dynamic fracture: Robust field updates for the analysis of complex crack patterns, Internat.
J. Numer. Methods Engrg. 93 (3) (2013) 276–301.
[66] J. Bleyer, C. Roux-Langlois, J.-F. Molinari, Dynamic crack propagation with a variational phase-field model: Limiting speed, crack
branching and velocity-toughening mechanisms, Int. J. Fract. (2016) 1–22.
[67] M.J. Borden, C.V. Verhoosel, M.A. Scott, T.J. Hughes, C.M. Landis, A phase-field description of dynamic brittle fracture, Comput. Methods
Appl. Mech. Engrg. 217–220 (2012) 77–95.
[68] C. Steinke, K. Özenç, G. Chinaryan, M. Kaliske, A comparative study of the r-adaptive material force approach and the phase-field method
in dynamic fracture, Int. J. Fract. 201 (1) (2016) 97–118.
[69] T. Li, J.-J. Marigo, D. Guilbaud, S. Potapov, Gradient damage modeling of brittle fracture in an explicit dynamics context, Internat. J. Numer.
Methods Engrg. 108 (11) (2016) 1381–1405.
[70] J. Bleyer, J.-F. Molinari, Microbranching instability in phase-field modelling of dynamic brittle fracture, Appl. Phys. Lett. 110 (2017)
151903.
[71] E. Tanné, T. Li, B. Bourdin, J.-J. Marigo, C. Maurini, Crack nucleation in variational phase-field models of brittle fracture, J. Mech. Phys.
Solids 110 (2018) 80–99.
[72] K. Pham, H. Amor, J.-J. Marigo, C. Maurini, Gradient damage models and their use to approximate brittle fracture, Int. J. Damage Mech.
20 (2011) 618–652.
[73] X. Zhang, C. Vignes, S.W. Sloan, D. Sheng, Numerical evaluation of the phase-field model for brittle fracture with emphasis on the length
scale, Comput. Mech. (2017) 1–16.
[74] K.H. Pham, K. Ravi-Chandar, C.M. Landis, Experimental validation of a phase-field model for fracture, Int. J. Fract. 205 (2017) 83–101.
[75] T.T. Nguyen, J. Yvonnet, M. Bornert, C. Chateau, K. Sab, R. Romani, R.L. Roy, On the choice of parameters in the phase field method for
simulating crack initiation with experimental validation, Int. J. Fract. 197 (2016) 213–226.
[76] J.Y. Wu, A unified phase-field theory for the mechanics of damage and quasi-brittle failure in solids, J. Mech. Phys. Solids 103 (2017) 72–99.
[77] J.Y. Wu, Numerical implementation of non-standard phase-field damage models, Comput. Methods Appl. Mech. Engrg. 340 (2018) 767–797.
[78] J.Y. Wu, A geometrically regularized gradient-damage model with energetic equivalence, Comput. Methods Appl. Mech. Engrg. 328 (2018)
612–637.
[79] H. Cornelissen, D. Hordijk, H. Reinhardt, Experimental determination of crack softening characteristics of normalweight and lightweight
concrete, Heron 31 (2) (1986) 45–56.
[80] D.C. Feng, J.Y. Wu, Phase-field regularized cohesize zone model (CZM) and size effect of concrete, Eng. Fract. Mech. 197 (2018) 66–79.
[81] C.V. Verhoosel, R. de Borst, A phase-field model for cohesive fracture, Internat. J. Numer. Methods Engrg. 96 (1) (2013) 43–62.
[82] S. May, J. Vignollet, R. de Borst, A numerical assessment of phase-field models for brittle and cohesive fracture: γ -Convergence and stress
oscillations, Eur. J. Mech. A Solids 52 (2015) 72–84.
[83] J. Vignollet, S. May, R. de Borst, C.V. Verhoosel, Phase-field models for brittle and cohesive fracture, Meccanica 49 (11) (2014) 2587–2601.
[84] S. Conti, M. Focardi, F. Iurlano, Phase field approximation of cohesive fracture models, Ann. Inst. H. Poincarè Anali. Non Linèaire 33 (4)
(2016) 1033–1067.
[85] F. Freddi, F. Iurlano, Numerical insight of a variational smeared approach to cohesive fracture, J. Mech. Phys. Solids 98 (2017) 156–171.
[86] C. Miehe, M. Hofacker, F. Welschinger, A phase field model for rate-independent crack propagation: Robust algorithmic implementation
based on operator splits, Comput. Methods Appl. Mech. Engrg. 199 (45–48) (2010) 2765–2778.
[87] A. Braides, Approximation of Free-Discontinuity Problems, Springer science & Business Media, Berlin, 1998.
[88] J.Y. Wu, V.P. Nguyen, A length scale insensitive phase-field damage model for brittle fracture, J. Mech. Phys. Solids 119 (2018) 20–42.
[89] H. Amor, J. Marigo, C. Maurini, Regularized formulation of the variational brittle fracture with unilateral contact: Numerical experiments,
J. Mech. Phys. Solids 57 (2009) 1209–1229.
1022 V.P. Nguyen, J.-Y. Wu / Comput. Methods Appl. Mech. Engrg. 340 (2018) 1000–1022

[90] C. Miehe, L. Schänzel, H. Ulmer, Phase field modeling of fracture in multi-physics problems. Part I. Balance of crack surface and failure
criteria for brittle crack propagation in thermo-elastic solids, Comput. Methods Appl. Mech. Engrg. 294 (2015) 449–485.
[91] X. Zhang, S.W. Sloan, C. Vignes, D. Sheng, A modification of the phase-field model for mixed mode crack propagation in rock-like materials,
Comput. Methods Appl. Mech. Engrg. 322 (2017) 123–136.
[92] C. Miehe, F. Welschinger, M. Hofacker, Thermodynamically consistent phase-field models of fracture: Variational principles and multi-field
FE implementations, Internat. J. Numer. Methods Engrg. 83 (2010) 1273–1311.
[93] H. Amor, Approche variationnelle des lois de Griffith et de Paris via des modeles non-locaux d’endommagement: Etude theorique et mise
en oeuvre numérique (Ph.D. thesis), Université Paris 13, Paris, France, 2008.
[94] F. Facchinei, J.-S. Pang, Finite Dimensional Variational Inequalities and Complementarity Problems, Vol. 1 and Vol. 2, Springer-Verlag,
New York, 2003.
[95] S. Balay, S. Abhyankar, M.F. Adams, J. Brown, P. Brune, K. Buschelman, L. Dalcin, V. Eijkhout, W.D. Gropp, D. Kaushik, M.G. Knepley,
L.C. McInnes, K. Rupp, B.F. Smith, S. Zampini, H. Zhang, H. Zhang, PETSc Users Manual, Tech. Rep. ANL-95/11 - Revision 3.7, Argonne
National Laboratory, 2016.
[96] P. Farrell, C. Maurini, Linear and nonlinear solvers for variational phase-field models of brittle fracture, Internat. J. Numer. Methods Engrg.
109 (5) (2017) 648667.
[97] N. Newmark, A method of computation for structural dynamics, J. Eng. Mech. 85 (1956) 67–94.
[98] H. Hilber, T. Hughes, R. Tayler, Improved numerical dissipation for time integration algorithms in structural dynamics, Earthq. Eng. Struct.
Dyn. 5 (1977) 283–292.
[99] V. Ziaei-Rad, L. Shen, J. Jiang, Y. Shen, Identifying the crack path for the phase field approach to fracture with non-maximum suppression,
Comput. Methods Appl. Mech. Engrg. 312 (2016) 304–321.
[100] J.F. Kalthoff, S. Winkler, Failure mode transition at high rates of shear loading, in: International Conference on Impact Loading and Dynamic
Behavior of Materials, vol. 1, 1987, pp. 185–195.
[101] M. Ramulu, A.S. Kobayashi, Mechanics of crack curving and branching – a dynamic fracture analysis, Int. J. Fract. 27 (3) (1985) 187–201.
[102] E. Sharon, J. Fineberg, Microbranching instability and the dynamic fracture of brittle materials, Phys. Rev. B 54 (1996) 7128–7139.
[103] K. Park, G.H. Paulino, W. Celes, R. Espinha, Adaptive mesh refinement and coarsening for cohesive zone modeling of dynamic fracture,
Internat. J. Numer. Methods Engrg. 92 (1) (2012) 1–35.
[104] T. Rabczuk, T. Belytschko, Cracking particles: A simplified meshfree method for arbitrary evolving cracks, Internat. J. Numer. Methods
Engrg. 61 (13) (2004) 2316–2343.
[105] G. Liu, Q. Li, M.A. Msekh, Z. Zuo, Abaqus implementation of monolithic and staggered schemes for quasi-static and dynamic fracture
phase-field model, Comput. Mater. Sci. 121 (2016) 35–47.
[106] Y.D. Ha, F. Bobaru, Characteristics of dynamic brittle fracture captured with peridynamics, Eng. Fract. Mech. 78 (2011) 1156–1168.
[107] M. Hofacker, C. Miehe, Continuum phase field modeling of dynamic fracture: Variational principles and staggered FE implementation, Int.
J. Fract. 178 (1) (2012) 113–129.
[108] C. Miehe, M. Hofacker, L.-M. Schänzel, F. Aldakheel, Phase field modeling of fracture in multi-physics problems. Part II. Coupled brittle-
to-ductile failure criteria and crack propagation in thermo-elastic-plastic solids, Comput. Methods Appl. Mech. Engrg. 294 (2015) 486–522.
[109] P. Klein, J. Foulk, E. Chen, S. Wimmer, H. Gao, Physics-based modeling of brittle fracture: Cohesive formulations and the application of
meshfree methods, Theor. Appl. Fract. Mech. 37 (1–3) (2001) 99–166.
[110] Z. Zhang, G.H. Paulino, Cohesive zone modeling of dynamic failure in homogeneous and functionally graded materials, Int. J. Plast. 21 (6)
(2005) 1195–1254.
[111] Z. Zhang, G.H. Paulino, W. Celes, Extrinsic cohesive modelling of dynamic fracture and microbranching instability in brittle materials,
Internat. J. Numer. Methods Engrg. 72 (8) (2007) 893–923.
[112] R. John, S.P. Shah, Mixed mode fracture of concrete subjected to i mpact loading, ASCE J. Struct. Eng. 116 (1990) 585–602.
[113] T. Belytschko, D. Organ, C. Gerlach, Element-free galerkin methods for dynamic fracture in concrete, Comput. Methods Appl. Mech. Engrg.
187 (3–4) (2000) 385–399.
[114] G. Ruiz, A. Pandolfi, M. Ortiz, Three-dimensional cohesive modeling of dynamic mixed-mode fracture, Internat. J. Numer. Methods Engrg.
52 (1–2) (2001) 97–120.
[115] C.-H. Sam, K.D. Papoulia, S.A. Vavasis, Obtaining initially rigid cohesive finite element models that are temporally convergent, Eng. Fract.
Mech. 72 (14) (2005) 2247–2267.
[116] G. Zi, H. Chen, J. Xu, T. Belytschko, The extended finite element method for dynamic fractures, J. Sound Vib. 12 (2005) 9–23.