1 An extended gradient damage model for anisotropic

2 fracture
3 Liang Xuea , Ye Fengb and Xiaodan Ren*a,c
4
a College of Civil Engineering, Tongji University, 1239 Siping Road, Shanghai, 200092, P.R.China
5
b School of Aeronautics, Northwestern Polytechnical University, 127 Youyi West Road, Xian, 710072, P.R.China
6
c State Key Laboratory of Disaster Reduction in Civil Engineering, Tongji University, 1239 Siping

7 Road, Shanghai, 200092, P.R.China

8

9
10 ARTICLE INFO ABSTRACT
11
12 Keywords: This paper combines energy decomposition and an extended gradient damage
13 Energy decomposition (EGD) model to develop an anisotropic fracture framework with decoupling
14 Cohesive zone theory of tensile and shear cohesive laws. By introducing the shear-normal decom-
15 Shear failure position in the energy form, the driving force of the damage variable is
16 Anisotropic fracture established within the framework of the EGD model, which is then capable
17 Gradient damage model of capturing the traction-separation of potential crack surfaces in both shear
18 and normal directions. The intrinsic correspondence between the cohesive
19 law and the damage evolution enables the accurate prediction of anisotropic
20 fracture behavior in the mixed form of Mode I and Mode II. Furthermore,
21 the proposed model also addresses the damage unloading issue, which still
22 remains a challenge in classic phase field theory or non-local damage theory.
23 A number of numerical examples are presented as validation. Some cutting-
24 edge benchmarks, such as complex mixed-mode fracture and perfect shear
25 fracture, are well reproduced.

26

27 1. Introduction

28 Cracks inevitably emerge in solids when subjected to diverse loads, encompassing tensile,
29 shear, or bending forces. As these cracks propagate, they engender additional stress concentrations,
30 exacerbating structural vulnerability and heightening the likelihood of failure, which in extreme
31 instances can culminate in total instability. Traditionally, materials and engineering science treat
32 solids as homogeneous and isotropic in their pristine state. Nonetheless, once damage sets in, it
33 precipitates an anisotropic fracture, altering the material’s properties. Within the realm of solid
34 mechanics, modelling anisotropic fractures poses a formidable challenge (Li and Zhu, 2024).
[email protected] (L. Xue); [email protected] (Y. Feng); [email protected] (X. Ren*)
ORCID (s):

Liang Xue et al.: Preprint submitted to Elsevier Page 1 of 68

 An extended gradient damage model for anisotropic fracture

35 To predict damage and fracture in solids, especially cracks in quasi-brittle materials such as
36 concrete and rock, a large number of models have been developed. These encompass continuous
37 damage models (Chaboche, 1988) and cohesive zone models (Elices et al., 2002), tailored to
38 capture the intricate behaviour of cracks. Moreover, smeared and discrete cracking methods (Rots
39 et al., 1985; Mosler and Meschke, 2004) have been advanced to simulate crack propagation,
40 encompassing aspects like initiation, propagation, merging, and branching. Among these, the local
41 damage model (Xue et al., 2023; Xue and Ren, 2024b) stands out as the most prevalent owing
42 to its lucid conceptual framework and straightforward numerical implementation. However, its
43 prognostications are contingent upon mesh size, leading to non-objective results.
44 To mitigate the mesh sensitivity stemming from strain softening, researchers drew inspiration
45 from non-local field theories and devised non-local damage models. Integral-type non-local
46 (Bazant and Pijaudier-Cabot, 1988; Lu and Chen, 2020; Chen et al., 2021; Ren et al., 2024) and
47 gradient-enhanced damage models (Peerlings et al., 1996) stand out as seminal approaches in this
48 realm. The fundamental concept involves obtaining certain physical attributes of material points
49 through weighted averaging within a defined neighborhood. By introducing an internal length, these
50 models account for interactions among material microstructures. However, this empirical averaging
51 method also engenders several challenges, including damage widening (de Borst and Verhoosel,
52 2016), characteristic length sensitivity (Wu and Nguyen, 2018; Mandal et al., 2019), and stress
53 locking (Bazant and Pijaudier-Cabot, 1988). Xue et al. (2024) provide a complete mathematical
54 proof of these problems for gradient-enhanced damage models. In subsequent advancements,
55 over-gradient-enhanced damage models (Poh and Swaddiwudhipong, 2009) addressed the stress-
56 locking issue by assigning negative weights to local variables. Moreover, the localizing gradient
57 damage model (Poh and Sun, 2017) using micromorphic theory has shown promise in ameliorating
58 damage widening. Although classic non-local models have undergone significant development and
59 improvement, integrating cohesive laws into these frameworks remains a challenging endeavour
60 (Wu et al., 2020a). Cohesive laws serve as vital metrics for gauging the mechanical properties and
61 surface energy of materials, finding broad applications not only in understanding solid fracture

Liang Xue et al.: Preprint submitted to Elsevier Page 2 of 68

 An extended gradient damage model for anisotropic fracture

62 behavior (Le et al., 2014; Freddi and Iurlano, 2017; Tandogan and Yalcinkaya, 2022; Zambrano
63 et al., 2022; Cai et al., 2022; de Carvalho et al., 2023; Aydiner et al., 2024) but also in evaluating the
64 structure-property relationship of biomaterials (Shao et al., 2012; Tripathi et al., 2017; Gao et al.,
65 2021). Hence, it is necessary to introduce traction-separation criteria into the prediction of solid
66 failure.
67 The phase-field theory provides another perspective for describing solid failure. This theory
68 describes sharp transitions at multiphase interfaces by introducing phase-field variables. Currently,
69 phase-field theory has been applied in various fields such as phase transformation, thin film growth,
70 crystal evolution, the evolution of electric field vesicles (Gao et al., 2009), and the prediction
71 of solid cracks (Francfort and Marigo, 1998; Bourdin et al., 2000). The fracture phase-field
72 model provides a unified and powerful approach for modelling cracks and fractures. This method
73 represents the transition between the intact and fractured states of a material in a continuous manner.
74 Cracks are regularized into diffuse damage zones controlled by an internal length scale 𝑙, thus
75 achieving a smooth transition from sharp crack topology to finite-width damage zones. Francfort
76 and Marigo (1998) laid the theoretical foundation for brittle fracture phase-field by providing a
77 variational description of Griffith’s ideal brittle fracture. They used possible displacement fields
78 within the solid and a set of crack surfaces as independent variables, defining the total energy of the
79 structure as the sum of deformation and fracture energies. They then solved for the displacement
80 field and phase field by minimizing the total energy. However, numerically solving for discrete
81 crack surfaces as unknowns in simulations is very challenging. Therefore, Bourdin et al. (2000)
82 reconstructed the fracture energy of the solid through crack surface energy and substituted the
83 reconstructed fracture energy into the Francfort-Marigo variational principle to obtain the classic
84 brittle fracture phase-field model. Wu (2017) extended the application of the phase-field model
85 from brittle fracture to quasi-brittle fracture by constructing an energy degradation function,
86 eliminating the sensitivity of the phase-field model to the internal length scale, which was a
87 milestone in the development of the phase-field model. Feng et al. (2021) rigorously derived
88 a series of cohesive laws through integral transformation. However, the phase-field model is

Liang Xue et al.: Preprint submitted to Elsevier Page 3 of 68

 An extended gradient damage model for anisotropic fracture

89 not without its flaws. The evolution of the damage profile is related to the cohesive laws, and
90 selecting mismatched cohesive laws can lead to damage unloading within the regularized crack
91 zone (Xue and Ren, 2024a). This phenomenon violates fundamental thermodynamic laws and leads
92 to erroneous prediction results (Pham and Marigo, 2013; Xue and Ren, 2024a).
93 On the other hand, predicting anisotropic fracture in solid failure has also garnered attention
94 from researchers. Freddi and Royer-Carfagni (2010) combined brittle phase-field models with
95 structural deformation theory to capture cleavage, deviatoric, combined cleavage-deviatoric, and
96 masonry-like fractures. Voyiadjis et al. (2010) introduced gradient plastic strain in kinematic
97 hardening, constructing a non-local anisotropic hardening plastic model. To capture damage
98 evolution in different directions, Mozaffari and Voyiadjis (2015) combined the phase-field method
99 with the Allen-Cahn equation, representing microcrack density in each direction using a second-
100 order damage tensor and extending it to the viscoplasticity framework (Mozaffari and Voyiadjis,
101 2016). Nguyen-Thanh and Rabczuk (2024); Li et al. (2024) employed non-local and peridynamic
102 operators to predict anisotropic fractures in solids. Scherer et al. (2022) introduced structural tensors
103 to force cracks to propagate in certain directions. Zhu et al. (2023) incorporated pore fluid pressure
104 into the framework of anisotropic damage. Kong et al. (2023) proposed a constitutive model
105 considering anisotropic damage in quasi-brittle materials by coupling microplane mechanics with
106 micromechanics. However, each of these models ignores the effect of internal length (interaction
107 domain) size on the prediction results.
108 Wu et al. (2020b) constructed a variational consistent anisotropic fracture phase-field model by
109 decomposing the strain energy into tension and compression components, which well reflects the
110 unilateral effect of quasi-brittle materials but does not consider the material’s shear failure behavior.
111 Feng and Li (2023) considered shear deformation-induced anisotropic fracture by directional en-
112 ergy decomposition, and Wang et al. (2024) predicted mixed failure of fiber-reinforced composites
113 by strain energy decomposition, but neither of these models can characterize the independence
114 of shear and tensile cohesive laws. Fei and Choo (2021); Yuan et al. (2022) created a two-phase
115 field model to separately describe tensile and shear failure. However, these two-phase field models

Liang Xue et al.: Preprint submitted to Elsevier Page 4 of 68

 An extended gradient damage model for anisotropic fracture

116 introduce more degrees of freedom, leading to longer computational times, and they also overlook
117 the coupling effects between mode I and mode II failure. Notably, the physical facts (Bazant and
118 Planas, 1997; Wong and Einstein, 2009) show that the progressive failure behavior of quasi-brittle
119 materials in different failure modes is different, a fact that highlights the limitations of these models.
120 Additionally, these models have also introduced challenges associated with damage unloading
121 within the regularized crack bands (Xue and Ren, 2024a). The various failure modes under different
122 loading conditions (as shown in Fig. (1)) emphasize the importance of efficiently capturing Mode
123 I and Mode II fracture. A comprehensive understanding of the distinctions between mode I and
124 mode II fractures is essential for developing more reliable and accurate fracture models, but a
125 satisfactory thermodynamic framework that adequately captures both mode I and mode II fracture
126 behavior remains an open research challenge.

Figure 1: A schematic diagram of progressive failure behavior under different fracture modes

127 As a response to the above dilemma, this study proposes an extended gradient damage (EGD)
128 model for anisotropic fracture. EGD adopts a strategy of decoupling the damage evolution from the
129 cohesive law so that the evolution of the damage is independent of the cohesive form (Xue and Ren,
130 2024a). The EGD model only requires the definition of a well-behaved damage profile function to
131 guarantee damage irreversibility, which solves the difficulties associated with damage unloading in
132 the phase-field and gradient-enhanced damage models (Xue and Ren, 2024a). Another advantage

Liang Xue et al.: Preprint submitted to Elsevier Page 5 of 68

 An extended gradient damage model for anisotropic fracture

133 of the EGD model is its ability to conveniently incorporate various energy decompositions. Strain
134 energy decomposition has been widely used in phase-field models to capture anisotropic fractures
135 in solids. However, within the foundational framework of fracture phase-field models, there is
136 no distinction between mode I and mode II fractures. Consequently, the energy degradation
137 functions in these models need to remain consistent for tension and shear (Feng and Li, 2022a),
138 limiting their flexibility in assigning material mechanical properties. In this paper, we introduce a
139 new energy decomposition method called shear-normal decomposition, which decomposes strain
140 energy into tension and shear components and assigns independent energy degradation functions.
141 By combining the EGD model with energy decomposition, this model can flexibly assign material
142 tensile and shear mechanical properties. Therefore, both mode I and mode II failures in solids can
143 be accurately captured. Additionally, through analytical solutions under 1D tension and integral
144 transformation techniques (Feng et al., 2021), a duality relationship between cohesive laws and
145 intrinsic functions is established, enabling accurate capture of various complex cohesive laws. The
146 governing equations of the EGD model are derived based on the principle of energy conservation,
147 ensuring thermodynamic consistency. The proposed model has been validated in challenging
148 numerical examples, including tension, pure shear, and mixed fracture cases. The predicted results
149 closely match the expected or experimental results, confirming the accuracy and reliability of the
150 model.
151 The paper is organized as follows: Section 2 introduces a thermodynamic framework that
152 integrates dissipative forces. Section 3 introduces the EGD model and the shear-normal energy
153 decomposition, which represents a novel gradient damage model and energy decomposition
154 approach. The EGD model employs the strategy of decoupling damage evolution from the cohesive
155 law, enabling any cohesive law to fulfill the condition of damage irreversibility (Xue and Ren,
156 2024a). And this is challenging to achieve with phase field models (Wu, 2017; Feng et al., 2021) and
157 other gradient damage models (Peerlings et al., 1996). Additionally, the EGD model is compatible
158 with various forms of energy decomposition. And the shear-normal energy decomposition can
159 independently describe tensile and shear failure modes. Consequently, combining the EGD model

Liang Xue et al.: Preprint submitted to Elsevier Page 6 of 68

 An extended gradient damage model for anisotropic fracture

160 with shear-normal energy decomposition is employed to model the anisotropic fracture of solid
161 materials. Section 4 establishes a dyadic relationship between the cohesive law and the intrinsic
162 function using integral transformation techniques. This facilitates the accurate description of
163 various complex cohesive laws. Section 5 shows the differences between the EGD model and the
164 current popular fracture phase field models, demonstrating the advantages of the EGD model over
165 other models. Section 6 presents numerical examples to validate the effectiveness of the proposed
166 method. Finally, in Section 7, the paper is concluded.

167 2. Thermodynamic framework

168 The energy variational principle is pivotal in solid and computational mechanics (Zienkiewicz
169 et al., 2005). The principle of minimum potential energy offers a concise approach to formulating
170 all governing equations related to elastic mechanics. Moreover, it grants us valuable insights into
171 mechanical systems: the stress equilibrium state should result in the system’s global potential
172 energy achieving a minimal value. This insight is grounded in the Boltzmann distribution within
173 statistical physics (Kardar, 2007). In emerging domains of solid mechanics, such as fracture phase
174 field theory, the system’s governing equations, comprising stress equilibrium and cracked phase
175 field evolution, can be succinctly derived using the variational method from the minimum potential
176 energy problem (Francfort and Marigo, 1998; Bourdin et al., 2000; Wu et al., 2020a). However,
177 it’s worth noting that the fracture process is substantially more intricate than Griffith’s description
178 suggests. Apart from the fundamental surface energy dissipation, the traditional principle of
179 minimum potential energy loses its applicability when additional dissipation mechanisms come
180 into play within the fracture process zone (Feng and Li, 2022b). These mechanisms may encompass
181 friction, plastic deformation, damage evolution, and various other forms of energy dissipation. As
182 a response to this complexity, this section introduces a thermodynamic framework tailored for
183 dissipative systems.

Liang Xue et al.: Preprint submitted to Elsevier Page 7 of 68

 An extended gradient damage model for anisotropic fracture

184 2.1. Dissipative system

185 Consider solid fracture occurring in region ,  ⊂ ℝ2 . For a conservative system, the energy
186 functional of a solid may comprise both internal energy and external potential energy,

 = 𝛱int + 𝛱ext (1)

187 where

ˆ ˆ
𝛱ext = − ∗
𝑏 ⋅ 𝑢d𝑉 − 𝑡∗ ⋅ 𝑢d𝐴 (2)
 𝜕

188 where 𝑏∗ and 𝑡∗ are respectively the prescribed body and boundary forces, 𝑢 is the displacement.
189 In dissipative systems, the influence of dissipative forces on the energy functional is taken into
190 account. Eq. (1) can be reformulated as follows:

 = 𝛱int + 𝛱ext + diss (3)

191 In general, the energy dissipation rate of a system can be expressed as follows:

ˆ
̇ diss = ̇
𝑓diss 𝜙d𝑉 (4)


192 where 𝑓diss is the dissipative force. 𝜙̇ should be understood as the derivative of damage 𝜙 with
193 respect to some pseudo time.

194 2.2. An illustrative example

195 To enhance our comprehension of the application of dissipative forces to the energy-variational
196 principle, we opt for an illustrative example to provide a concise overview.
197 A rigid body with mass ’𝑚’ and zero initial velocity undergoes free fall. When neglecting air
198 resistance, following the Hamiltonian principle, the action of the system is:

Liang Xue et al.: Preprint submitted to Elsevier Page 8 of 68

 An extended gradient damage model for anisotropic fracture

ˆ 𝑡2
𝐽= 𝐿d𝑡 (5)
𝑡1

∑
199 where 𝐿 = 𝑇 − 𝑈 represents the Lagrangian; 𝑇 = 𝑘 12 𝑚𝑞̇ 𝑘2 denotes the kinetic energy of the
∑
200 system; 𝑈 = 𝑘 𝑚𝑔𝑘 𝑞𝑘 signifies the potential energy of the system; here, only the gravity in the
201 y-direction 𝑔𝑦 = 𝑔 is considered, and the gravity in the other directions is zero. 𝑞𝑘 stands for the
202 generalized coordinate of the system, while 𝑞̇ 𝑘 represents the generalized velocity.
203 The Hamiltonian principle postulates that the equations of motion for a system can be derived
204 through the minimization of the action. Simply put, the system will follow a trajectory where the
205 action achieves its minimum value,

ˆ 𝑡2 ( )
𝜕𝐿 d 𝜕𝐿
𝛿𝐽 = − 𝛿𝑞𝑘 d𝑡 = 0 (6)
𝑡1 𝜕𝑞𝑘 d𝑡 𝜕 𝑞̇ 𝑘

206 Eq. (6) holds for arbitrary 𝛿𝑞𝑘 . The equation governing the motion of the object in the vertical
207 direction (𝑦-direction) is then expressed as follows:

d 𝜕𝐿 𝜕𝐿
− =0
d𝑡 𝜕 𝑞̇ 𝑘 𝜕𝑞𝑘
(7)
d
⇒ (𝑚𝑦) ̇ + 𝑚𝑔 = 0
d𝑡

208 When accounting for air resistance, the expression for the action of the system requires
209 modification to (Goldstein et al., 2002):

ˆ 𝑡2 ˆ 𝑡2
𝐽= 𝐿d𝑡 + d𝑡 (8)
𝑡1 𝑡1

210 where 𝐷 is the action associated with the dissipative force, which can generally be characterised
211 as:

Liang Xue et al.: Preprint submitted to Elsevier Page 9 of 68

 An extended gradient damage model for anisotropic fracture

ˆ 𝑡2
= 𝛷d𝑡 (9)
𝑡1

212 where 𝛷 is the dissipation function, which is related to the dissipation force (air resistance):

𝜕𝛷
𝑓diss = − (10)
𝜕 𝑞̇ 𝑘

213 When taking air resistance into account, the variation in the system’s action is as follows:

ˆ 𝑡2 ( )
𝜕𝐿 d 𝜕𝐿 𝜕𝛷
𝛿𝐽 = − − 𝛿𝑞𝑘 d𝑡 = 0 (11)
𝑡1 𝜕𝑞𝑘 d𝑡 𝜕 𝑞̇ 𝑘 𝜕 𝑞̇ 𝑘

214 As per Eq. (11), the equation describing the motion of the object, accounting for air resistance,
215 is given by:

d 𝜕𝐿 𝜕𝐿 𝜕𝛷
− + =0
d𝑡 𝜕 𝑞̇ 𝑘 𝜕𝑞𝑘 𝜕 𝑞̇ 𝑘
(12)
d
⇒ (𝑚𝑦) ̇ + 𝑚𝑔 − 𝑓diss = 0
d𝑡

216 A comparison between Eq. (12) and Eq. (7) reveals that the equations of motion, factoring in
217 dissipative forces, align more closely with the actual physical scenario.

218 2.3. Governing equation

219 Consider  ⊂ ℝ2 as the reference configuration of a solid, its boundary designated as 𝜕. On
220 𝜕D , apply a prescribed displacement 𝑢∗ , while on the remaining part 𝜕N , apply traction 𝑡∗ , with
221 𝜕D ∪ 𝜕N = 𝜕 and 𝜕D ∩ 𝜕N = ∅.
222 In the gradient damage model, the energy functional, accounting for dissipative forces, is
223 expressed as follows:

ˆ ˆ
= 𝛹 (𝜖, 𝜙, ∇𝜙) d𝑉 − 𝑡∗ ⋅ 𝑢d𝐴 + diss (13)
 𝜕

Liang Xue et al.: Preprint submitted to Elsevier Page 10 of 68

 An extended gradient damage model for anisotropic fracture

224 where 𝜖 = ∇sym 𝑢 and 𝑢 ∶  → ℝ2 is the displacement field.

225 In accordance with the first law of thermodynamics, also known as the law of conservation of
226 energy,

ˆ ˆ ( ) ˆ
𝜕𝛹 𝜕𝛹 ̇ 𝜕𝛹
∗
𝑡 ⋅ 𝑢d𝐴
̇ = ∶ 𝜖̇ + 𝜙+ ̇ ̇
∇𝜙 d𝑉 + 𝑓diss 𝜙d𝑉 (14)
𝜕  𝜕𝜖 𝜕𝜙 𝜕∇𝜙 

227 Performing partial integration and then applying the divergence theorem, the Eq. (14) can be
228 rewritten as,

ˆ ˆ ˆ ( ) ˆ
( ) 𝜕𝛹 𝜕𝛹 𝜕𝛹 ̇
̇
𝛹 + 𝑓diss 𝜙d𝑉 − ∇⋅ ⋅ 𝑢d𝑉
̇ = ∗
𝑡 −𝑛⋅ ⋅ 𝑢d𝐴
̇ − 𝑛⋅ 𝜙d𝐴 (15)
  𝜕𝜖 𝜕 𝜕𝜖 𝜕 𝜕∇𝜙

229 By asserting that Eq. (15) holds for all physically feasible 𝑢̇ and 𝜙,
̇ we can derive the governing

230 equations and boundary conditions.

⎧ 𝜕𝛹
⎪ ∇ ⋅ 𝜎 = 0, 𝜎 = on  (16a)
⎨ 𝜕𝜖
⎪ 𝛹 + 𝑓 = 0 on  (16b)
⎩ diss

231 and

⎧ 𝑛 ⋅ 𝜎 = 𝑡∗ on 𝜕 (17a)
⎪
⎨ 𝜕𝛹
⎪ 𝑛 ⋅ 𝜕∇𝜙 = 0 on 𝜕 (17b)
⎩

232 where  (⋅) = 𝜕 (⋅) ∕𝜕𝜙 − ∇ ⋅ 𝜕 (⋅) ∕𝜕∇𝜙 is the Euler–Lagrange operator.
233 The second law of thermodynamics asserts that the external power is not less than the rate of
234 increase of free energy, i.e.,

Liang Xue et al.: Preprint submitted to Elsevier Page 11 of 68

 An extended gradient damage model for anisotropic fracture

ˆ ˆ ( )
𝜕𝛹 𝜕𝛹 ̇ 𝜕𝛹
∗
𝑡 ⋅ 𝑢d𝐴
̇ ≥ ∶ 𝜖̇ + 𝜙+ ̇
∇𝜙 d𝑉 (18)
𝜕  𝜕𝜖 𝜕𝜙 𝜕∇𝜙

235 By substituting Eq. (18) into Eq. (14), one obtains:

ˆ
̇
𝑓diss 𝜙d𝑉 ≥0 (19)


236 Given the prerequisite that the damage is irreversible, implying that negative dissipation is
237 physically impossible, we conclude that 𝑓diss ≥ 0.
238 Within this subsection, the application of the energy conservation law reveals a significant
239 insight: the dissipative force 𝑓diss ≥ 0 to align with the fundamental principles of thermodynamics.
240 Furthermore, we present a comprehensive derivation of the governing equations for the gradient
241 damage model based on thermodynamic consistency. These encompass the stress equilibrium
242 equation and the damage evolution equation. In addition, the extended gradient damage model
243 has good compatibility with multi-physics fields and can take into account the effects of strain rate
244 and temperature on fracture behavior. Appendix A provides the thermodynamic framework and
245 governing equations that take strain rate and temperature (Wang et al., 2020, 2022; Zhou et al.,
246 2024) into account.

247 3. Extended gradient damage model for anisotropic fracture

248 3.1. Decomposition of strain energy

249 The framework presented in Section 2 perfectly establishes the stress equilibrium equations
250 in the displacement field and the governing equations for damage evolution in the damage field.
251 However, it has a fatal flaw: the constitutive relations and thermodynamic forces in tension, shear,
252 and compression are symmetric. As a result, solids are prone to fracture that deviates from physical
253 objectivity. To solve this problem, a proper decomposition of the elastic strain energy is a viable
254 approach. In the EGD model, it is convenient to decompose the elastic strain energy and assign

Liang Xue et al.: Preprint submitted to Elsevier Page 12 of 68

 An extended gradient damage model for anisotropic fracture

255 different energy degradation functions to different parts of the elastic strain energy. In this paper,
256 the strain energy of a solid can be expressed as:

𝛹strain = 𝑔n (𝜙) 𝜓n + 𝑔s (𝜙) 𝜓s (20)

257 with

⎧ 𝑔 (𝜙) = 1
(21a)
⎪ n∕s 1 + 𝑐n0∕s0 𝜔n∕s (𝜙)
⎨
⎪ 𝜓 = 1ℙ ∶ 𝜖 ∶ 𝔼 ∶ 𝜖 (21b)
⎩ n∕s 2 n∕s

258 where 𝑔n (𝜙) and 𝑔s (𝜙) denote the energy degradation functions for tensile and shear behaviors,
√ √
2 2𝐸0 𝐺I 2 2𝐺0 𝐺II 𝐸0
259 respectively. 𝑐n0 = 𝑓t2 𝑙
, 𝑐s0 = 𝑓s2 𝑙
, 𝐸0 is the Young’s modulus; 𝐺0 = 2(1+𝜈)
is the shear
260 modulus; 𝜈 is the Poisson’s ratio, 𝐺I and 𝐺II represent the fracture energies associated with tensile
261 and shear failure modes, respectively. 𝑓t and 𝑓s represent the tensile and shear strength, respectively.
262 𝑙 denotes the internal length, which controls the width of the regularization crack (or damage
263 zone). The functions 𝜔n (𝜙) and 𝜔s (𝜙) govern the characteristics of the cohesive laws for tension
264 and shear, respectively. When 𝜔n (𝜙) ≠ 𝜔s (𝜙), it implies that the tensile and shear cohesive laws
265 are mutually independent. Conversely, when 𝜔n (𝜙) = 𝜔s (𝜙), the model reverts to the traditional
266 cohesive fracture representation. The specific form of 𝜔n∕s (𝜙) is intricately tied to the cohesive
267 law, and the elucidation of its structure will be provided in Section 4. 𝔼 is the fourth-order elasticity
268 tensor, and 𝜖 is the second-order strain tensor. ℙn∕s is the fourth order projection tensor, which is
269 dependent on the approach of energy decomposition. In this paper, by combining the tensile and
270 shear effects on the potential crack surface, the stress state at the crack surface is assumed to be as
271 shown in Fig. 2.
272 The illustration in Fig. 2 depicts the breakdown of effective stress into two distinct components:
273 a normal stress responsible for separation or crushing and a shear stress characterizing dislocation
274 glide. We refer to this decomposition approach as the "shear-normal decomposition." It is worth

Liang Xue et al.: Preprint submitted to Elsevier Page 13 of 68

 An extended gradient damage model for anisotropic fracture

Mode I failure

Mode I
n
Mode II

s
Mode II failure

Figure 2: Tensile and shear cohesive zone on potential cracked surface, 𝜎̂̄ n the tensile driving force
that causes normal separation on the potential crack surface, and 𝜏̂̄s the shear driving force that causes
tangential slip on the potential crack surface.

275 noting that the directions of normal separation and shear dislocation are orthogonal, and surfaces
276 perpendicular to the potential crack surface have no normal stresses. The cohesive fracture theory
277 posits that the potential crack surface should exclusively be evaluated for normal separation
278 traction and dislocation glide shear (Lawn, 1993). This conclusion is in harmony with the energy
279 decomposition method proposed in this paper.

280 3.2. Damage-induced anisotropy

281 Comparable to the phase field theory, the internal energy of the extended gradient damage
282 model comprises two parts: a regularized crack surface energy component and a strain energy
283 component,

𝛹 = 𝛹crack + 𝛹strain (22)

Liang Xue et al.: Preprint submitted to Elsevier Page 14 of 68

 An extended gradient damage model for anisotropic fracture

with

⎧ [ 2 ]
⎪ 𝛹crack = 𝜓n0 𝑙
|∇𝜙| + 𝛼 (𝜙)
2
(23a)
⎪ 2
⎨ ( )
⎪𝛹 𝑔n (𝜙) 𝑔s (𝜙)
= 𝑔n (𝜙) 𝜓n + 𝑔s (𝜙) 𝜓s = ℙn + ℙs ∶ 𝜖 ∶ 𝔼 ∶ 𝜖 (23b)
⎪ strain 2 2
⎩

𝑓t2
284 where 𝛹crack is termed the crack density function, and 𝛹strain is the strain energy density. 𝜓n0 = 2𝐸0
,
285 𝛼 (𝜙) is the damage profile function. 𝑔n (𝜙) and 𝑔s (𝜙) denote the energy degradation functions for
286 tensile and shear behaviors, respectively, with the following expressions:

⎧ 𝑔 (𝜙) = 1
(24a)
⎪ n
1 + 𝑐n0 𝜔n (𝜙)
⎨ 1
⎪ 𝑔s (𝜙) = (24b)
⎩ 1 + 𝑐s0 𝜔s (𝜙)

287 The purpose of this paper is to establish a gradient damage model that can clearly distinguish
288 between Mode I and Mode II failures. Therefore, the dissipative force 𝑓diss is defined as:

[ ] 𝜓
𝑓diss = − 𝑓 ′ (𝜙) 𝑔n2 (𝜙) + 𝑔n′ (𝜙) 𝜓n − 𝑔s′ (𝜙) 𝜓s − 𝑓 ′ (𝜙) 𝑔s2 (𝜙) n0 𝜓s (25)
𝜓s0
𝑓s2
289 where 𝜓s0 = 2𝐺0
, 𝑓 ′ (𝜙) = d𝑓 (𝜙)
d𝜙
, 𝑓 (𝜙) is a function to be determined and will be introduced in
290 Section 4.
291 According to the energy conservation principle and the necessary conditions for the stability of
292 the system (Pham et al., 2011), by substituting Eq. (22) and Eq. (25) into Eq. (16b), the governing
293 equation for the damage evolution can be obtained as follows:

𝛹 + 𝑓diss = 0
( ) (26)
2 2 ′ 2 ′𝜓n 2
𝜓s
⇒𝑙 ∇ 𝜙 − 𝛼 (𝜙) + 𝑓 (𝜙) 𝑔n (𝜙) + 𝑔s (𝜙) =0
𝜓n0 𝜓s0

Liang Xue et al.: Preprint submitted to Elsevier Page 15 of 68

 An extended gradient damage model for anisotropic fracture

294 Furthermore, the constitutive relation can be described by Eq. (16a) and Eq. (23b) as follows:

𝜕𝛹
𝜎= = 𝑔n (𝜙) ℙn ∶ 𝔼 ∶ 𝜖 + 𝑔s (𝜙) ℙs ∶ 𝔼 ∶ 𝜖 (27)
𝜕𝜖

295 From Eq. (27), it is evident that the damage-induced orthogonal anisotropy constitutive relation
296 can be effectively incorporated into the gradient damage model through energy decomposition.
297 Both co-axial and non-coaxial anisotropic damage can be realized by choosing different energy
298 decompositions (Freddi and Royer-Carfagni, 2010; Wu et al., 2020b; Feng and Li, 2023; Xue et al.,
299 2023).
300 Collating Eqs. (26) and (27), the governing equations of the gradient damage model under
301 quasi-static conditions can be summarised as follows:

( )
⎧ 2 2 𝜓n 𝜓s
′ ′ 2
⎪ 𝑙 ∇ 𝜙 − 𝛼 (𝜙) + 𝑓 (𝜙) 𝑔n (𝜙) 2
+ 𝑔s (𝜙) ≤0 (28a)
⎨ 𝜓n0 𝜓s0
⎪ 𝜎 = 𝑔 (𝜙) ℙ ∶ 𝔼 ∶ 𝜖 + 𝑔 (𝜙) ℙ ∶ 𝔼 ∶ 𝜖 (28b)
⎩ n n s s

302 As can be seen from Eq. (28b), the notable characteristic of the EGD model is its ability to pro-
303 vide different cohesive laws for shear and tensile. This flexibility enables the selection of a cohesive
304 law that aligns with the material properties, accurately reflecting the actual failure behavior, which
305 is difficult to achieve by other anisotropic phase field models (Wu et al., 2020b; Feng and Li, 2023).
306 Moreover, the proposed model is also compatible with various energy decomposition methods,
307 including tensile-compressive decomposition, spherical-deviation decomposition, and more. As a
308 result, this model paves the way for applying anisotropic damage models.
309 Within this section, the functions 𝛼 (𝜙), 𝑓 (𝜙), and 𝜔 (𝜙) are the functions to be solved. These
310 functions are related to the cohesive law and the evolution of the damage profile, which will be
311 presented in Section 4.

Liang Xue et al.: Preprint submitted to Elsevier Page 16 of 68

 An extended gradient damage model for anisotropic fracture

312 3.3. Explicit expressions for potential crack angle (𝜃) and projection operator ℙn∕s

313 According to the cohesive fracture theory, the damage in solids is driven by tensile stresses
314 and shear stresses on potential crack surfaces (Lawn, 1993; Feng et al., 2024). Therefore, this
315 paper decomposes the stress state into normal stress, representing normal separation, and shear
316 stress, representing dislocation slip. Generally, normal stresses on surfaces perpendicular to crack
317 surfaces are not considered. Hence, the proposed model assumes that the normal stress on surfaces
318 perpendicular to potential crack surfaces is zero. As shown in Fig. 3, using the direction of the
319 principal stress as the initial coordinate system, the (angle between ) the potential crack surface
√
2 𝜎̂̄ 1 |𝜎̂̄ 2 |
320 and the maximum principal stress is 𝛼 = − 21 arc cos 𝜎̂̄ −𝜎̂̄
when 𝜎̂̄ 1 ≥ ||𝜎̂̄ 2 ||, 𝜎̂̄ 𝑖 is the 𝑖-th
1 2

321 principal stress. Additionally, based on knowledge of material mechanics (Hibbeler, 2006), one
√
322 can determine the normal tensile stress 𝜎̂̄ n = 𝜎̂̄ 1 + 𝜎̂̄ 2 and the shear stress 𝜏̂̄s = 𝜎̂̄ 1 ||𝜎̂̄ 2 || on the
323 crack surface.

Potential Crack
Surface

Potential Crack Potential Crack

Surface Surface

Figure 3: Schematic diagram illustrating the coordination between normal separation and dislocation
slip: (a) Mohr’s circle under 2D stress state; (b) principal stress state; (c) the local coordinate system
along a potential crack.

324 Next, for ease of program compilation, the authors will derive potential crack surface angles
325 and projection tensors within the global coordinate system. In finite element programming, Voigt’s
326 rule is usually used to represent the tensor to improve computational efficiency. Therefore, this
327 subsection expresses ℙn∕s explicitly using Voigt’s rule.

Liang Xue et al.: Preprint submitted to Elsevier Page 17 of 68

 An extended gradient damage model for anisotropic fracture

328 Consider the potential crack surface as the local coordinate system. In this context, the
329 transformation relationship between stress in the local coordinate system and stress in the global
330 coordinate system can be expressed as follows:

⎡𝜎̄ 𝜎̄ ⎤ ⎡ cos 𝜃 sin 𝜃 ⎤ ⎡𝜎̂̄ n 𝜏̂̄s ⎤ ⎡cos 𝜃 − sin 𝜃 ⎤

⎢ 11 12 ⎥ = ⎢ ⎥⎢ ⎥⎢ ⎥ (29)
⎢𝜎̄ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣ 21 𝜎̄ 22 ⎦ ⎣− sin 𝜃 cos 𝜃 ⎦ ⎣ 𝜏̂̄s 𝜎̂̄ s ⎦ ⎣ sin 𝜃 cos 𝜃 ⎦

331 where 𝜃 denotes the angle of the potential crack, 𝜎̄ = 𝔼 ∶ 𝜖, is the effective stress tensor of the
332 global coordinate system, while 𝜎̂̄ n and 𝜏̂̄s represent the normal and shear effective stresses acting
333 on the potential crack surface, respectively. Additionally, 𝜎̂̄ s corresponds to the normal stress in the
334 lateral direction of the potential crack surface.
335 In Eq. (29), we can establish the relationship between the stress tensor on the potential crack
336 surface and the stress tensor in the global coordinate system:

⎡ ⎤ ⎡ ⎤⎡ ⎤
⎢ 𝜎̂̄ n ⎥ ⎢ cos2 𝜃 sin2 𝜃 −2 sin 𝜃 cos 𝜃 ⎥ ⎢ 𝜎̄ 11 ⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢ 𝜎̂̄ s ⎥ = ⎢ sin2 𝜃 cos2 𝜃 2 sin 𝜃 cos 𝜃 ⎥ ⎢ 𝜎̄ 22 ⎥ (30)
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢ 𝜏̂̄s ⎥ ⎢sin 𝜃 cos 𝜃 − cos 𝜃 sin 𝜃 cos2 𝜃 − sin2 𝜃 ⎥ ⎢ 𝜎̄ 12 ⎥
⎣ ⎦ ⎣ ⎦⎣ ⎦

⎡ ⎤ ⎡ ⎤⎡ ⎤
⎢ 𝜎̄ 11 ⎥ ⎢ cos 𝜃
2
sin2 𝜃 2 cos 𝜃 sin 𝜃 ⎥ ⎢ 𝜎̂̄ n ⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢ 𝜎̄ 22 ⎥ = ⎢ sin2 𝜃 cos2 𝜃 −2 sin 𝜃 cos 𝜃 ⎥ ⎢ 𝜎̂̄ s ⎥ (31)
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢ 𝜎̄ 12 ⎥ ⎢− cos 𝜃 sin 𝜃 cos 𝜃 sin 𝜃 cos2 𝜃 − sin2 𝜃 ⎥ ⎢ 𝜏̂̄s ⎥
⎣ ⎦ ⎣ ⎦⎣ ⎦
337 By substituting Eq. (27) into Eq. (31) following the constitutive relationship, we can derive the
338 Cauchy stress tensor in the global coordinate system as follows:

Liang Xue et al.: Preprint submitted to Elsevier Page 18 of 68

 An extended gradient damage model for anisotropic fracture

⎡ ⎤ ⎡ ⎤⎡ ⎤
⎢ 𝜎11 ⎥ ⎢ cos 𝜃 cos 𝜃 sin 𝜃 sin 𝜃 2 cos 𝜃 sin 𝜃 ⎥ ⎢ 𝑔n (𝜙) 𝜎̂̄ n ⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢ 𝜎22 ⎥ = ⎢ sin2 𝜃 cos 𝜃 cos 𝜃 −2 sin 𝜃 cos 𝜃 ⎥ ⎢ 𝑔n (𝜙) 𝜎̂̄ s ⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢ 𝜎12 ⎥ ⎢− cos 𝜃 sin 𝜃 cos 𝜃 sin 𝜃 cos2 𝜃 − sin2 𝜃 ⎥ ⎢ 𝑔𝑠 (𝜙) 𝜏̂̄s ⎥
⎣ ⎦ ⎣ ⎦⎣ ⎦
(32)
⎡ ⎤⎡ ⎤
⎢ 𝑔n (𝜙) cos 𝜃 cos 𝜃 𝑔n (𝜙) sin 𝜃 sin 𝜃 2𝑔𝑠 (𝜙) cos 𝜃 sin 𝜃 ⎥ ⎢ 𝜎̂̄ n ⎥
⎢ ⎥⎢ ⎥
= ⎢ 𝑔n (𝜙) sin2 𝜃 𝑔n (𝜙) cos 𝜃 cos 𝜃 −2𝑔𝑠 (𝜙) sin 𝜃 cos 𝜃 ⎥ ⎢ 𝜎̂̄ s ⎥
⎢ ( )⎥ ⎢ ⎥
⎢−𝑔n (𝜙) cos 𝜃 sin 𝜃 𝑔n (𝜙) cos 𝜃 sin 𝜃 𝑔𝑠 (𝜙) cos2 𝜃 − sin2 𝜃 ⎥ ⎢ 𝜏̂̄s ⎥
⎣ ⎦⎣ ⎦

339 Substituting Eq. (30) into Eq. (32) yields the damage constitutive relationship as:

𝜎 = 𝑔n (𝜙) ℙn ∶ 𝜎̄ + 𝑔s (𝜙) ℙs ∶ 𝜎̄
[ ] (33)
= 𝑔n (𝜙) ℙn + 𝑔s (𝜙) ℙs ∶ 𝔼 ∶ 𝜖

340 where

⎡ ( )⎤
⎢ sin4 𝜃 + cos4 𝜃 2 sin2 𝜃 cos2 𝜃 2 sin 𝜃 cos 𝜃 sin2 𝜃 − cos2 𝜃 ⎥
⎢ ( )⎥
ℙn = ⎢ 2 sin2 𝜃 cos2 𝜃 sin4 𝜃 + cos4 𝜃 2 sin 𝜃 cos 𝜃 cos2 𝜃 − sin2 𝜃 ⎥
⎢ ( ) ( ) ⎥
⎢sin 𝜃 cos 𝜃 sin2 𝜃 − cos2 𝜃 sin 𝜃 cos 𝜃 cos2 𝜃 − sin2 𝜃 4 sin2 𝜃 cos2 𝜃 ⎥
⎣ ⎦
(34)

⎡ ( )⎤
⎢ 2 sin2 𝜃 cos2 𝜃 −2 sin2 𝜃 cos2 𝜃 2 cos 𝜃 sin 𝜃 cos2 𝜃 − sin2 𝜃 ⎥
⎢ ( )⎥
ℙs = ⎢ −2 sin2 𝜃 cos2 𝜃 2 sin2 𝜃 cos2 𝜃 −2 cos 𝜃 sin 𝜃 cos2 𝜃 − sin2 𝜃 ⎥
⎢ ( ) ( ) ( 2 )2 ⎥
⎢cos 𝜃 sin 𝜃 cos2 𝜃 − sin2 𝜃 − cos 𝜃 sin 𝜃 cos2 𝜃 − sin2 𝜃 cos 𝜃 − sin2 𝜃 ⎥
⎣ ⎦
(35)

Liang Xue et al.: Preprint submitted to Elsevier Page 19 of 68

 An extended gradient damage model for anisotropic fracture

341 In the implicit algorithm, updating the tangential stiffness of the material is necessary, and this
342 can be determined using Eq. (33):

[ ]
𝔼tan = 𝑔n (𝜙) ℙn + 𝑔s (𝜙) ℙs ∶ 𝔼 (36)

343 As per the directional decomposition in the proposed model, it is essential to ensure that the
344 normal stress 𝜎̂̄ s in the lateral direction of the potential crack surface is maintained at 0. In other
345 words, 𝜃 must fulfill the following condition:

𝜎̂̄ s (𝜃) = sin2 𝜃 𝜎̄ 11 + 2 sin 𝜃 cos 𝜃 𝜎̄ 12 + cos2 𝜃 𝜎̄ 22 = 0 (37)

346 Drawing upon the knowledge of the mechanics of materials and considering this constraint, we
347 can determine the angle of the potential crack surface in the global coordinate system as follows:

⎧ ⎡ ⎤
⎪ ⎢ ( )⎥
⎪ 𝜃 = 1 arc sin ⎢ 1 𝜎̄ 11 + 𝜎̄ 22 𝜎̄ 11 − 𝜎̄ 22
⎪ 1 2 ( ) + 𝜏̂̄s − 𝜎̂̄ n ⎥ (38a)
⎢ (𝜎̄ 11 −𝜎̄ 22 )2
2 2𝜎̄ 12 ⎥
⎪ ⎢ + 𝜎
̄ 12 ⎥
⎪ ⎣ 4𝜎̄ 12 ⎦
⎨
⎪ ⎡ ⎤
⎪ ⎢ ( )⎥
𝜋 1 ⎢ 1 𝜎̄ 11 + 𝜎̄ 22 𝜎̄ 11 − 𝜎̄ 22
⎪ 𝜃2 = − arc sin ( ) + 𝜏̄s − 𝜎̄ n ⎥
̂ ̂ (38b)
⎪ 2 2 ⎢ (𝜎̄ 11 −𝜎̄ 22 )2 2 2𝜎̄ 12 ⎥
⎪ ⎢ + 𝜎̄ 12 ⎥
⎩ ⎣ 4 𝜎
̄ 12 ⎦

⎧ 𝜎̂̄ = 𝜎̂̄ + 𝜎̂̄ (39a)

⎪ n √max min
⎨
⎪ 𝜏̂̄s = −𝜎̂̄ max 𝜎̂̄ min (39b)
⎩

( )
348 where 𝜃𝑖 represents a possible angle, and when 𝜎̂̄ s 𝜃𝑖 = 0, 𝜃 = 𝜃𝑖 for 𝑖 = 1, 2, 𝜎̂̄ max is the positive
349 maximum principal effective stress, and 𝜎̂̄ min is the negative minimum principal effective stress.

Liang Xue et al.: Preprint submitted to Elsevier Page 20 of 68

 An extended gradient damage model for anisotropic fracture

350 Eq. (36) shows that as stress rotates and damage progresses, the strain and stress no longer align
351 along the same principal axis. This outcome provides a logical explanation for the anisotropy re-
352 sulting from damage in quasi-brittle materials. Moreover, it aligns with experimental observations
353 of orthotropic anisotropy in the material. It is worth noting that anisotropic damage is generally not
354 considered in biaxial tension or compression under 2D stress states (Wu et al., 2020b; Li et al., 2021;
355 Liu et al., 2024). Therefore, the model degenerates to an isotropic model in the biaxial tensile stress
356 state. The isotropic model can be referred to the authors’ previous work (Xue and Ren, 2024a). In
357 addition, Appendix B provides the angle of rotation and stress state of the potential crack face in a
358 3-dimensional stress state.

359 4. The relationship between intrinsic function and the cohesive law

360 To foster a more intuitive grasp of the interrelationship among 𝛼 (𝜙), 𝑓 (𝜙), and 𝜔 (𝜙) and
361 the cohesive law, we opt to analyze a solid subjected to pure shear deformation (Fig. 4), thereby
362 deriving the functions to be solved.

Shear slip

Figure 4: Shear slip of solids under pure shear deformation.

363 Under pure shear deformation, as shown in Fig. 4, the governing equation can be rewritten as:

Liang Xue et al.: Preprint submitted to Elsevier Page 21 of 68

 An extended gradient damage model for anisotropic fracture

⎧ 2 2
⎪ 𝑙2 d 𝜙 − 𝛼 ′ (𝜙) + 𝑓 ′ (𝜙) 𝜏 ≤ 0 (40a)
⎪ d𝑦2 𝑓s 2
⎨
⎪ 𝜏 = 𝑔 (𝜙) 𝐺 𝛾 = 1
𝐺𝛾 (40b)
⎪ s 0
1 + 𝑐s0 𝜔s (𝜙) 0
⎩

364 where 𝜏 is the shear stress to be applied to the solid; 𝛾 is engineering shear strain.
365 Noting the stress equilibrium condition, ∇⋅𝜏 = d𝜏∕d𝑦 = 0. Multiplying Eq.(40a) by d𝜙∕d𝑦 and
( )
366 integrating with respect to 𝑦 𝑦 ∈ 𝜙∗ , 𝜙∗ is the region where the damage occurred and satisfies
367 𝜙̇ ≥ 0, we have

ˆ ( 2 2
)
d 𝜙 𝜏 d𝜙
𝑙2 2 − 𝛼 ′ (𝜙) + 𝑓 ′ (𝜙) 2 d𝑦
d𝑦 𝑓s d𝑦
( )2 (41)
𝑙2 d𝜙 𝜏2
= − 𝛼 (𝜙) + 𝑓 (𝜙) 2 = 0
2 d𝑦 𝑓s

368 Note that the integral constant is calculated as zero due to the boundary condition 𝜙 (𝑦) =
369 ∇𝜙 (𝑦) = 0, ∀𝑦 ∈ 𝜕𝜙∗ . Using the fact that ∇𝜙 (𝑦∗ ) = 0 since 𝜙 reaches the maximum at 𝑦∗ ,
370 the shear stress on a potential shear slip surface can be calculated using Eq. (41), which can be
371 expressed as:

√
𝛼 (𝜙∗ )
𝜏= 𝑓 (42)
𝑓 (𝜙∗ ) s

372 Let 𝛼 (𝜙) be related to 𝑓 (𝜙) as follows:

𝛼 (𝜙) = (1 − 𝜙)2 𝑓 (𝜙) (43)

373 Combining Eq. (41), Eq. (42) and Eq. (43), the explicit expression of d𝑦∕d𝜙 can be obtained.

Liang Xue et al.: Preprint submitted to Elsevier Page 22 of 68

 An extended gradient damage model for anisotropic fracture

d𝑦 (1 − 𝜙) 𝑙
=√ (44)
d𝜙 [ ]
2𝛼 (𝜙) (1 − 𝜙)2 − (1 − 𝜙∗ )2

374 The relation between position 𝑦 and 𝜙 can be solved by integration of Eq. (44). We have

ˆ 𝜙∗
(1 − 𝜙) 𝑙
𝑦=𝑦 ± ∗
√ d𝜙 (45)
[ 2
]
𝜙
2𝛼 (𝜙) (1 − 𝜙) − (1 − 𝜙∗ )2

375 Let 𝜙∗ → 1 for 𝑦∗ = 0 and obtain the final damage profile from Eq. (45). Thus, when 𝜙∗ → 1,
376 Eq. (45) is rewritten as:

ˆ 1
𝑙
𝑦= √ d𝜙 (46)
𝜙 2𝛼 (𝜙)
377 Eq. (46) shows that 𝛼 (𝜙) determines the final profile of the crack. Based on the available studies,
( )
378 𝛼 (𝜙) = 2𝜙 − 𝜙2 ∕2 is selected in this paper.
379 According to Eq. (40b), the shear deformation is the integral of the shear strain along the
380 direction perpendicular to the crack.

ˆ ˆ ˆ
𝐿
𝜏
𝐿 𝐿 [ ] 𝜏
𝛥𝑢s = 𝛾d𝑦 = d𝑦 = 1 + 𝑐s0 𝜔s (𝜙) d𝑦
0 𝑔s (𝜙) 𝐺0 𝐺0
0
ˆ 0
(47)
𝜏𝐿 𝜏
= + 𝑐s0 𝜔s (𝜙) d𝑦
𝐺0 𝜙∗ 𝐺0

381 where on the right-most side, the first term 𝜏𝐿

𝐺0
is the elastic shear deformation, and the second term
382 is the shear slip 𝛿s on the crack surface.
383 Combining Eq. (42), Eq. (44), and Eq. (47), the shear slip 𝛿s can be further expressed as:

Liang Xue et al.: Preprint submitted to Elsevier Page 23 of 68

 An extended gradient damage model for anisotropic fracture

ˆ
𝜏
𝛿s = 𝑐s0 𝜔s (𝜙) d𝑦
𝜙∗ 𝐺0
ˆ 𝜙∗
𝜏 d𝑦
=2 𝑐s0 𝜔s (𝜙) d𝜙
0 𝐺0 d𝜙 (48)
ˆ 𝜙∗
𝜔 (𝜙)
√s (1 − 𝜙∗ )
𝑓 (𝜙)
= 2𝛿ch,s √[ ]
d𝜙
2
0
(1 − 𝜙) − (1 − 𝜙∗ )2

2𝐺II
384 where 𝛿ch,s = 𝑓s
.
385 By referencing Eqs. (42), and (48), we can get the following parametric equation related to the
386 shear cohesive law:

{
𝜏 = (1 − 𝜙∗ ) 𝑓s (49a)

𝛿s = 𝛿ch,s 𝑠 (𝜙∗ ) (49b)

387 with

ˆ 𝜙∗
𝜔 (𝜙)
√s (1 − 𝜙∗ )
𝑓 (𝜙)
𝑠 (𝜙∗ ) = 2 √[ d𝜙 (50)
2
]
0
(1 − 𝜙) − (1 − 𝜙∗ )2

388 Eq. (50) reveals the intrinsic connection between the shear stress, denoted as 𝜏, and the shear
389 slip, represented as 𝛿s , elucidating their interdependence via the parameter 𝜙∗ . The variation in
390 the function 𝑠 (𝜙∗ ) gives rise to diverse cohesive laws governing the relationship between the shear
391 stress and the shear slip. Upon selecting a specific form for the cohesive law, essentially determining
392 the function 𝑠 (𝜙∗ ), the solution for 𝜔s (𝜙) can be obtained through the utilization of the first type
393 of Volterra integral equations (Polyanin and Manzhirov, 2008; Feng et al., 2021; Xue and Ren,
394 2024a), as follows:

Liang Xue et al.: Preprint submitted to Elsevier Page 24 of 68

 An extended gradient damage model for anisotropic fracture

√ ⎛ ˆ 𝜙 ⎞
𝑓 (𝜙) ⎜ d 𝑠 (𝜙∗ ) ⎟
𝜔s (𝜙) = ⎜ √ d𝜙∗ ⎟
𝜋 ⎜ d𝜙 0 (1 − 𝜙∗ )2 − (1 − 𝜙)2 ⎟
⎝ ⎠
√ ⎛ ˆ 𝜙 ⎞
𝛼 (𝜙) ⎜ d 𝑠 (𝜙∗ ) ∗⎟
√ (51)
𝜋 (1 − 𝜙) ⎜⎜ d𝜙 0
= d𝜙 ⎟
∗ 2
(1 − 𝜙 ) − (1 − 𝜙) 2 ⎟
⎝ ⎠
√ ⎛ ˆ 𝜙 ⎞
2𝜙 − 𝜙2 ⎜ d 𝑠 (𝜙∗ ) ∗⎟
=√ ⎜ d𝜙 √ d𝜙 ⎟
2𝜋 (1 − 𝜙) ⎜ 0 2
(1 − 𝜙∗ ) − (1 − 𝜙) 2 ⎟
⎝ ⎠

395 According to Eq. (51), the calculation formula for the shear cohesive law is consistent with the
396 tensile cohesive law, the derivation of the tensile cohesive law can be found in the work of Xue and
397 Ren (2024a). In Appendix C, 𝜔n∕s (𝜙) is given for several common cohesive laws.
398 An important aspect to emphasize is that within dissipative systems, the incorporation of
399 dissipative forces necessitates verification through the law of energy conservation. Specifically,
400 this entails ensuring that the external power input exceeds the rate of increase in free energy. This,
401 in turn, dictates that the dissipative force must maintain a non-negative value. Illustrating this with
402 the linear cohesive law as an exemplar, in accordance with Eq. (24), (25), (51), and (C.2), the
403 calculation for the dissipative force 𝑓diss can be conducted as follows:

[ ] 𝜓
𝑓diss = − 𝑓 ′ (𝜙) 𝑔n2 (𝜙) + 𝑔n′ (𝜙) 𝜓n − 𝑔s′ (𝜙) 𝜓s − 𝑓 ′ (𝜙) 𝑔s2 (𝜙) n0 𝜓s
𝜓s0
( )
4𝐸0 𝐺I 𝑔n2 (𝜙) ′ 2 ′ 2
𝜓n0
= − 1 𝜓 n + 𝑐 s0 𝜔 (𝜙) 𝑔 (𝜙) 𝜓 s − 𝑓 (𝜙) 𝑔 (𝜙) 𝜓
𝜋𝑓t 2 𝑙 (1 − 𝜙)3
s s s
𝜓s0 s
( ) (52)
4𝐸0 𝐺I 𝑔n2 (𝜙) ′ 2 ′ 2
𝜓n0
= − 1 𝜓 n + 𝑐 s0 𝜔 (𝜙) 𝑔 (𝜙) 𝜓 s − 𝑓 (𝜙) 𝑔 (𝜙) 𝜓
𝜋𝑓t 2 𝑙 (1 − 𝜙)3
s s s
𝜓s0 s
( ) ( )
4𝐸0 𝐺I 𝑔n2 (𝜙) 4𝐺0 𝐺II 𝐸0 𝑓t2 𝑔s2 (𝜙)
= −1 𝜓n + − 𝜓
𝜋𝑓t 2 𝑙 (1 − 𝜙)3 𝜋𝑓s2 𝑙 𝐺0 𝑓s2 (1 − 𝜙)3 s

Liang Xue et al.: Preprint submitted to Elsevier Page 25 of 68

 An extended gradient damage model for anisotropic fracture

404 Through the concept of 𝛤 -convergence, it becomes evident that the crack recovers to a sharp
405 shape when 𝑙 → 0+ . Likewise, Eq. (52) demonstrates that 𝑓diss ≥ 0 is maintained when 𝑙 → 0+ .
406 This underscores the fact that the dissipative force formulated by the proposed model adheres to
407 the principles of energy conservation.

408 5. Comparison with fracture phase field model

409 5.1. Independent shear mechanics properties

410 Although current popular phase-field fracture models can also use energy decomposition tech-
411 niques to capture anisotropic fracture behavior in solids, these models cannot assign independent
412 mechanical properties to shear failure. In the rupture phase-field model, the strain energy can also be
413 decomposed into tensile and shear components, each with its own independent energy degradation
414 function. However, the fracture energy with Mode II failure remains consistent with that of Mode
415 I failure, and it is not possible to assign independent fracture energies for shear failure. A detailed
416 proof procedure is given in this subsection.
417 In this subsection, using the fracture phase field model with integral transformation (Feng et al.,
418 2021) as an example, and following an energy decomposition similar to that in the manuscript,
419 which decomposes the strain energy into tensile and shear components, the governing equations
420 for the fracture phase field model are:

⎧ 𝐺I [ ] [ ′ ′ ]
⎪ −2𝑙2 ∇2 𝜙 + 𝛼 ′ (𝜙) + 𝑔n (𝜙) 𝜓n + 𝑔s (𝜙) 𝜓s = 0 (53a)
⎨ 𝑙
⎪ 𝜎 = 𝑔 (𝜙) ℙ ∶ 𝔼 ∶ 𝜖 + 𝑔 (𝜙) ℙ ∶ 𝔼 ∶ 𝜖 (53b)
⎩ n n s s

421 with

Liang Xue et al.: Preprint submitted to Elsevier Page 26 of 68

 An extended gradient damage model for anisotropic fracture

⎧ 1
⎪ 𝑔n (𝜙) = (54a)
⎪ 1 + 𝑐n0 𝜔 (𝜙)
⎪ 1
⎪ 𝑔s (𝜙) = (54b)
⎨ 1 + 𝑐s0 𝜔 (𝜙)
⎪ 𝛼 (𝜙) = (1 − 𝜙)2 𝜔 (𝜙) (54c)
⎪
⎪
⎪ 𝜓n∕s = ℙn∕s ∶ 𝜖 ∶ 𝔼 ∶ 𝜖 (54d)
⎩

2𝐸0 𝐺I 2𝐺0 𝐺II

422 where 𝑐n0 = 𝑓t2 𝑙
, 𝑐s0 = 𝑓s2 𝑙
.
423 Under pure shear deformation, as shown in Fig. 4, the governing equations (53) can be rewritten
424 as:

⎧ 𝐺I [ ] ′

⎪ 𝑙 −2𝑙 2 2
∇ 𝜙 + 𝛼 ′
(𝜙) + 𝑔s (𝜙) 𝜓s = 0 (55a)
⎨ 1
⎪ 𝜏 = 𝑔𝑠 (𝜙) 𝐺0 𝛾 = 𝐺𝛾 (55b)
⎩ 1 + 𝑐s0 𝜔s (𝜙) 0

425 Following the steps outlined in Section 4, a set of parametric equations for the shear cohesive
426 law can be obtained.

⎧ √
⎪ 𝐺I
∗
⎪ 𝜏 = (1 − 𝜙 ) 𝐺 𝑓s (56a)
⎪ √
II
⎨ ˆ 𝜙∗
⎪ 𝛿 = 4 𝐺I 𝐺II 𝜔s (𝜙) (1 − 𝜙∗ )
√ d𝜙 (56b)
⎪ s 𝑓s 0 2
⎪ ∗
𝛼 (𝜙) − 𝜔s (𝜙) (1 − 𝜙 )
⎩

427 Under pure shear deformation, the fracture energy of a solid can be expressed as:

ˆ 𝛿f ˆ 1
d𝛿s ∗
𝐺shear = 𝜏d𝛿s = 𝜏 d𝜙 (57)
0 0 d𝜙∗

Liang Xue et al.: Preprint submitted to Elsevier Page 27 of 68

 An extended gradient damage model for anisotropic fracture

428 where 𝛿f is the shear slip when the solid is completely fractured.
429 By substituting Eq.(56) into Eq. (57), one obtains,

ˆ 𝛿f ˆ 1
d𝛿s ∗
𝐺shear = 𝜏d𝛿s = 𝜏 d𝜙 = 𝐺I ≠ 𝐺II (58)
0 0 d𝜙∗

430 The above derivation demonstrates that although the energy degradation function 𝑔s (𝜙) for
431 shear strain energy is assigned an independent rupture energy parameter 𝐺II , this parameter does
432 not effectively contribute to the model. As a result, it is not possible to assign independent material
433 parameters for tensile and shear failures in the current fracture phase field model. This limitation
434 undoubtedly hinders the predictive function of the model.
435 In contrast, the EGD model proposed in the manuscript ensures that the fracture energy for
436 shear failure is no longer identical to the tensile fracture energy. By substituting Eq.(49) into Eq.
437 (57), one obtains,

ˆ 𝛿f ˆ 1
d𝛿s ∗
𝐺shear = 𝜏d𝛿s = 𝜏 d𝜙 = 𝐺II ≠ 𝐺I (59)
0 0 d𝜙∗

438 In addition, the tensile and shear cohesive laws in the fracture phase field model must be
439 consistent, i.e., 𝜔n (𝜙) = 𝜔s (𝜙), whereas the EGD model does not have these constraints and has
440 the flexibility to assign strength, fracture energy, and cohesive law.

441 5.2. Irreversible damage

442 Irreversibility is a fundamental principle in the field of damage and fracture mechanics, which
443 acknowledges the fact that materials generally cannot naturally repair or revert back to their original
444 state without external assistance. If a model does not satisfy the irreversibility criteria, it means that
445 it permits the spontaneous healing or reversal of damage or cracks without any external interference.
446 When modeling or predicting fracture behavior, any deviations from damage irreversibility can
447 lead to unrealistic physical consequences, which are characterized by oscillations or unphysical
448 behavioral solutions. Researchers have endeavored to tackle this problem by inventing the notion of

Liang Xue et al.: Preprint submitted to Elsevier Page 28 of 68

 An extended gradient damage model for anisotropic fracture

449 a damage upper envelope. However, it is important to highlight that there is a difference between the
450 upper limit and the originally expected damage pattern. This discrepancy leads to the dissipation of
451 energy during the fracture of the solid no longer being in line with the expected desired consequence
452 (Pham and Marigo, 2013; Xue and Ren, 2024a).
453 The irreversibility of damage can be intuitively determined by the evolution of the damage
454 profile. In the extended gradient damage model, the damage profile can be obtained by integrating
455 Eq. (44).
456 To visually assess whether various gradient damage models satisfy the irreversibility of damage
457 conditions, this study selected exponential softening and 3rd-order polynomial cohesive laws (see
458 Feng et al., 2021, or Appendix C.3 for details) to compare the phase-field model (Wu, 2017;
459 Feng et al., 2021) and the EGD model. The evolution of the damage profiles corresponding to the
460 exponential cohesive law and the 3rd-order polynomial cohesive law are given in Figs. 5 and 6,
461 respectively.

1 1 1

0.9 0.9 0.9

0.8 0.8 0.8

0.7 0.7 0.7

0.6 0.6 0.6

0.5 0.5 0.5

0.4 0.4 0.4

0.3 0.3 0.3

0.2 0.2 0.2

0.1 0.1 0.1

0 0 0
-2 -1 0 1 2 -10 -5 0 5 10 -2 -1 0 1 2
|x|/lc |x|/lc |x|/lc

(a) Unified PF model (b) Integral transform PF model (c) EGD model

Figure 5: Evolution of damage profiles during exponential cohesive law

462 Figs. (5-6) suggest that the form of cohesive laws in the phase-field model will influence the
463 irreversibility of damage. When mismatched cohesive laws are chosen, the regularization crack
464 region of the phase-field model will exhibit damage unloading. In contrast, the damage profile in
465 the EGD model is only related to the damage profile function. Merely defining a well-behaved

Liang Xue et al.: Preprint submitted to Elsevier Page 29 of 68

 An extended gradient damage model for anisotropic fracture

1 1 1

0.9 0.9 0.9

0.8 0.8 0.8

0.7 0.7 0.7

0.6 0.6 0.6

0.5 0.5 0.5

0.4 0.4 0.4

0.3 0.3 0.3

0.2 0.2 0.2

0.1 0.1 0.1

0 0 0
-8 -6 -4 -2 0 2 4 6 8 -6 -4 -2 0 2 4 6 -2 -1 0 1 2
|x|/lc |x|/lc |x|/lc

(a) Unified PF model (b) Integral transform PF model (c) EGD model

Figure 6: Evolution of damage profiles during 3rd-order polynomial cohesive law

466 damage profile function 𝛼 (𝜙) ensures the damage irreversibility of any cohesive law within the
467 EGD model (Xue and Ren, 2024a).

468 6. Numerical examples

469 It is a recognized fact that the monolithic algorithm utilizing the conventional Newton-Raphson
470 scheme encounters limitations in effectively conducting gradient-damage simulations. This arises
471 due to the lack of convexity in the total potential energy of the system  (𝑢, 𝜙) concerning the total
472 unknowns (𝑢, 𝜙). Nonetheless, it is noteworthy that  (𝑢, 𝜙) does exhibit convexity individually
473 with respect to 𝑢 and 𝜙, as established by Bourdin et al. (2000). In light of this, a viable approach
474 for solving the governing equations involves an alternating strategy, wherein the damage field and
475 displacement field are successively fixed. While this may entail a trade-off in terms of sacrificing
476 the second-order convergence rate, the resultant alternating minimization algorithm emerges as
477 remarkably robust.
478 In this study, the fracture simulations employ the alternating minimization algorithm with
479 bound constraints, as proposed by Amor et al. (2009). The weak formulation of the governing
480 equations in Eq. (28) can be expressed as follows: Given the prescribed traction 𝑡∗𝑖 on the boundary
481 𝜕𝑡 and the displacement 𝑢∗𝑖 on the boundary 𝜕𝑢 at the 𝑖th time step, the objective is to determine
482 𝑢 ∈  and 𝜙 ∈  that satisfy the conditions for all admissible variations 𝛿𝑢 ∈ ̄ and 𝛿𝜙 ∈ ,
̄

Liang Xue et al.: Preprint submitted to Elsevier Page 30 of 68

 An extended gradient damage model for anisotropic fracture

⎧ˆ ˆ
⎪ 𝜎 (𝑢, 𝜙) ∶ ∇𝛿𝑢d𝑉 − 𝑡∗𝑖 ⋅ 𝛿𝑢d𝐴 = 0 (60a)
⎪ 
⎨ˆ [ ]
𝜕
𝜓n 𝜓s
⎪ 2 ′ ′ 2
−𝑙 ∇𝜙∇𝛿𝜙 − 𝛼 (𝜙) 𝛿𝜙 + 𝑓 (𝜙) 𝑔n (𝜙) ′ 2
𝛿𝜙 + 𝑓 (𝜙) 𝑔s (𝜙) 𝛿𝜙 d𝑉 ≤ 0 (60b)
⎪  𝜓n0 𝜓s0
⎩

{ } { }
483 Above,  = 𝑣 ∈ 𝐻 1 () ∶ 𝑣 = 𝑢∗𝑖 on𝜕𝑢 and ̄ = 𝑣 ∈ 𝐻 1 () ∶ 𝑣 = 0on𝜕𝑢 represent the
{ }
484 trial and test function spaces for displacement, respectively, while  = 𝑤 ∈ 𝐻 1 () ∶ 𝑤𝑖−1 ≤ 𝑤 ≤ 1
{ }
485 and  ̄ = 𝑤 ∈ 𝐻 1 () ∶ 𝑤 ≥ 0 denote the trial and test function spaces for the damage-field.

486 𝐻 1 () denotes the Sobolev space that comprises the function 𝑣 satisfying the property that both
487 |𝑣|2 and |∇𝑣|2 are integrable over the domain .
488 Employing the conventional technique outlined in Hughes (2012), the aforementioned weak
489 formulation can be effectively tackled through the utilization of the finite element method. In this
490 context, the widely employed staggered algorithm with bound constrains is implemented herein.
491 The algorithm at the 𝑖-th step is structured as follows, building upon the knowledge of the
492 converged field 𝑢𝑖−1 and 𝜙𝑖−1 from the (𝑖 − 1)-th step:
493 The algorithm at the 𝑖-th step:

( ) ( )
494 • Initialization: Begin by setting the initial values as 𝑢(0) , 𝜙(0) = 𝑢𝑖−1 , 𝜙𝑖−1 , and establish
495 the relative tolerance as 𝛿0 .

• Displacement Field Update: Find 𝑢(𝑘) ∈  with fixed 𝜙 = 𝜙(𝑘−1) such that ∀𝛿𝑢 ∈ ̄

ˆ ˆ
𝜎 (𝑢, 𝜙) ∶ ∇𝛿𝑢d𝑉 − 𝑡∗𝑖 ⋅ 𝛿𝑢d𝐴 = 0
 𝜕

• Damage Field Update: Find 𝜙(𝑘) ∈  with fixed 𝑢 = 𝑢(𝑘) such that ∀𝛿𝜙 ∈ 
̄

ˆ [ ]
2 ′ ′ 2 𝜓n ′ 2
𝜓s
−𝑙 ∇𝜙∇𝛿𝜙 − 𝛼 (𝜙) 𝛿𝜙 + 𝑓 (𝜙) 𝑔n (𝜙) 𝛿𝜙 + 𝑓 (𝜙) 𝑔s (𝜙) 𝛿𝜙 d𝑉 ≤ 0
 𝜓n0 𝜓s0

Liang Xue et al.: Preprint submitted to Elsevier Page 31 of 68

 An extended gradient damage model for anisotropic fracture

• End: Stop iteration when

( ‖ (𝑘) ‖𝜙(𝑘) − 𝜙(𝑘−1) ‖ )

‖ 𝑢 − 𝑢(𝑘−1) ‖‖
max ,‖ ‖ <𝛿
0
‖𝑢 (𝑘−1) ‖ ‖𝜙 (𝑘−1) ‖

496 This iterative procedure ensures the progressive refinement of both the displacement and
497 damage fields throughout the simulation process. In this study, we have performed simulations
498 using the FEniCS open-source finite element simulation package (Alnæs et al., 2015).

499 6.1. Mode-I failure in a uniaxial tension of a 1D bar

500 Let us consider a uniaxial tensile long rod with a length of 100mm and a width of 5mm (Fig.
501 7). The left end of the rod is fixed, and a progressively increasing displacement 𝑢∗ is applied to the
502 right end. The ultimate displacement is denoted as 𝑢∗max . The material parameters are as follows:
503 tensile fracture energy 𝐺I = 0.1𝑁∕mm, tensile strength 𝑓t = 3MPa, Young modulus 𝐸0 = 30GPa,
504 and Poisson’s ratio 𝜈 = 0. The characteristic length 𝑙 = 2mm, and the mesh size ℎ = 0.5mm.

L
Figure 7: Uniaxial traction of a softening bar: geometry, loading and boundary conditions

505 We examined the influence of the cohesive law within the proposed model presented in this
506 paper on the evolution of damage. In this study, we chose exponential, 3-rd order polynomial,
507 Cornelissen cohesive laws, and Park–Paulino–Roesler (PPR) (Park et al., 2009) cohesive laws and
508 predicted the damage profiles for each loading stage. It is worth noting that the PPR cohesive law
509 differs from most other cohesive laws in that it has the ability to control the softening shape of the
510 traction-separation relationship. This feature makes it suitable for fracture modeling in a variety
511 of materials. Specifically, when the cohesive law parameter 𝛼 < 2, its softening shape is convex,
512 which is challenging to achieve in general cohesive law models. Details of the PPR are presented in
513 Appendix C.5. The results depicted in Fig. 8 demonstrate that the evolution of the damage profile

Liang Xue et al.: Preprint submitted to Elsevier Page 32 of 68

 An extended gradient damage model for anisotropic fracture

Figure 8: Damage evolution on a 1D rod (predicted vs. target)

514 remains consistent, irrespective of the choice of cohesive law. These predictions align with the
515 analytical solution outlined in Eq. (44). This critical development ensures that the selection of the
516 cohesive law does not destroy the irreversibility condition and allows us to freely select a cohesive
517 law that reflects the progressive failure of the material.
518 Finally, referring to Eq. (44), Fig. 9 extracts the Mode I cohesive law from the global stress-
519 displacement response and compares it with the target cohesive law. It can be observed that the
520 numerical results are in good agreement with the target curve. As the mesh size diminishes or the
521 internal length increases, the predictions will gradually converge towards the target result. These
522 results sufficiently confirm the ability of the EGD model to capture complex cohesive laws.

Liang Xue et al.: Preprint submitted to Elsevier Page 33 of 68

 An extended gradient damage model for anisotropic fracture

3 target cohesive law (Exponential) 3 target cohesive law (Cornelissen)

numericial result (Exponential) numericial result (Cornelissen)

2.5 2.5

2 2
(MPa)

(MPa)
1.5 1.5

1 1

0.5 0.5

0 0
0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2
(mm) (mm)

(a) Exponential softening (b) Cornelissen softening

3 3
target cohesive law (3-rd polynomial) target cohesive law (PPR =1.3)
numericial result (3-rd polynomial) target cohesive law (PPR =2.0)
target cohesive law (PPR =2.8)
2.5 2.5
numericial result (PPR =1.3)
numericial result (PPR =2.0)
numericial result (PPR =2.8)
2 2
(MPa)

(MPa)

1.5 1.5

1 1

0.5 0.5

0 0
0 0.02 0.04 0.06 0.08 0.1 0 0.02 0.04 0.06 0.08 0.1
(mm) (mm)

(c) 3-rd polynomial softening (d) PPR softening

Figure 9: Comparison between the mode-I cohesive laws extracted from the 1D uniaxial tension and
the target cohesive laws

523 6.2. Pure shear test

524 The shear properties of materials are critical for many engineering applications and material
525 selection, especially in design and manufacturing, where the behavior of materials under shear
526 deformation needs to be considered. Referring to the existing studies on the testing of the shear
527 properties of materials (Zeng et al., 2022; Ren et al., 2023), this subsection predicts the shear
528 cohesive law and Mode II fracture of the specimens under pure shear deformation.
529 We examine a unit square slab (1mm × 1mm) with the bottom of the plate fixed and apply a
530 tangential displacement to the top (Fig. 10). The material parameters are as follows: shear fracture

Liang Xue et al.: Preprint submitted to Elsevier Page 34 of 68

 An extended gradient damage model for anisotropic fracture

531 energy 𝐺II = 0.1𝑁∕mm, shear strength 𝑓s = 3MPa, Young’s modulus 𝐸0 = 30GPa, and Poisson
532 ratio 𝜈 = 0. The characteristic length is 𝑙 = 0.02mm, mesh size is ℎ = 0.01∕0.005mm.

shear band

Figure 10: Pure shear test: Loading condition

533 This study was chosen to incorporate linear and exponential softening laws. In Fig. 11, we
534 observe the anticipated shear crack bands, showcasing the effectiveness of our model in accurately
535 predicting genuine Mode II cracks. Importantly, the successful simulation of the shear band
536 underscores that the EGD model is not limited to predicting quasi-brittle materials but also has
537 the potential to predict adiabatic shear band (ASB) in metallic materials (Zeng et al., 2022).

Mode II failure

(a) shear crack band (b) ultimate shear displacement

Figure 11: Pure shear test: shear crack band and ultimate shear displacement

538 Referring to the calculation method of the cohesive law in Eq. (44), Fig. 12 extracts the
539 shear cohesive law from the predicted results and compares it with the target cohesive law. The

Liang Xue et al.: Preprint submitted to Elsevier Page 35 of 68

 An extended gradient damage model for anisotropic fracture

540 comparison reveals a remarkable agreement between the numerical results and the desired curve,
541 demonstrating that the shear-normal energy decomposition enables the EGD model to accurately
542 capture both tensile and shear failure characteristics. Combining these findings with the examples
543 presented in Section 6.1, we can confidently conclude that the proposed model effectively captures
544 both tensile and shear progressive failure behavior while allowing the independent assignment of
545 cohesive laws for each.

3 target cohesive law (Linear) 3 target cohesive law (Exponential)

numericial result h=0.01mm numericial result h=0.01mm
numericial result h=0.005mm numericial result h=0.005mm
2.5 2.5

2 2

1.5 1.5

1 1

0.5 0.5

0 0
0 0.02 0.04 0.06 0.08 0.1 0 0.05 0.1 0.15 0.2

(a) linear softening (b) Exponential softening

Figure 12: Pure shear test: predicted shear cohesive law vs. target results

546 6.3. Shear test on single-edge notched plate

547 Next, let us delve into the example of brittle fracture in a single-edge notched plate. As
548 illustrated in Fig. 13, this plate is square, measuring 1 mm long, with a unit thickness in the
549 out-of-plane direction. A straight horizontal notch, spanning a length of 0.5 mm, is precisely
550 introduced at the midpoint of the left edge. In this scenario, the bottom edge remains fixed, while
551 the top edge is subjected to a horizontal shear displacement in the case of shear testing. This
552 particular example has been previously explored in the research of Ambati et al. (2015), Miehe et al.
553 (2010), and Wu et al. (2020b). However, only the tensile failure mode was observed in these prior

Liang Xue et al.: Preprint submitted to Elsevier Page 36 of 68

 An extended gradient damage model for anisotropic fracture

554 investigations. This limitation can be attributed to the fact that these models did not incorporate an
555 energy decomposition approach that accounted for shear failure.

0.5

Initial notch

0.5

Figure 13: Shear test on single-edge notched plate: Geometry and boundary conditions. The unit of
length is [mm].

(a) 𝛽 = 10 (b) 𝛽 = 0.6 (c) 𝛽 = 0.2

Figure 14: Shear test on single-edge notched plate: Ultimate fracture modes at different tensile-shear
strength ratios 𝛽

556 In this example, we demonstrate that the failure mode can be readily manipulated using the
𝑓s
557 proposed model through the adjustment of the strength ratio 𝛽 = 𝑓t
. Both tensile and shear cohesive
558 laws employed exponential softening. The material parameters are as follows: tensile fracture
559 energy 𝐺I = 2.7𝑁∕mm, tensile strength 𝑓t = 2445.42MPa, shear fracture energy 𝐺II = 2.7𝑁∕mm,

Liang Xue et al.: Preprint submitted to Elsevier Page 37 of 68

 An extended gradient damage model for anisotropic fracture

(a) 𝛽 = 10 (b) 𝛽 = 0.6 (c) 𝛽 = 0.2

Figure 15: Shear test on single-edge notched plate: Ultimate deformation at different tensile-shear
strength ratios 𝛽

560 shear strength 𝑓s = 𝛽𝑓t , Young’s modulus 𝐸0 = 2100GPa, and Poisson’s ratio 𝜈 = 0.3. The
561 characteristic length 𝑙 = 0.002mm, and the mesh size ℎ = 0.001mm.
562 Figs. 14 and 15 display the trajectories of cracks and deformations, respectively, for various
563 strength ratios. Fig. 15 illustrates the predicted crack paths for varying strength ratios. It is worth
564 noting that when 𝛽 = 10, the predicted crack paths align with the expectations of the linear elastic
565 fracture mechanics (LEFM) with maximum hoop (circumferential) stress criterion (Erdogan and
566 Sih, 1963). This consistency is attributed to the fact that when shear strength significantly exceeds
567 tensile strength, cracks will develop perpendicular to the direction of the maximum principal stress.
568 Similar crack paths have been predicted by both the fracture phase field model (Ambati et al., 2015;
569 Wu et al., 2020b) and the integral nonlocal damage model (Ren et al., 2023). Conversely, when
570 𝛽 = 0.2, a straight shear crack forms, effectively dividing the specimen into two distinct sections.
571 This outcome underscores the notable difference between the material’s shear and tensile strengths.
572 We set 𝛽 to 0.6, and a fracture mode between tensile and shear fracture can be observed from Fig.
573 15(b).
574 The force-displacement curves for different strength ratios are shown in Fig. 16, and it can be
575 seen that different strength ratios not only affect the failure mode but also have an effect on the
576 global force-displacement curve.

Liang Xue et al.: Preprint submitted to Elsevier Page 38 of 68

 An extended gradient damage model for anisotropic fracture

350

300

250
Force Fx (N)
200

150

100

50

0
0 0.005 0.01 0.015 0.02
Displacement ux (mm)

Figure 16: Shear test on single-edge notched plate: Force–displacement curve for different strength
ratios

577 6.4. Double edge-notched concrete specimens under pure shear loading

Figure 17: Double edge-notched concrete specimens under pure shear loading: Geometry and boundary
conditions. The unit of length is [mm].

578 Next, we consider a pure shear test of double-edge-notched plain concrete. The geometry and
579 boundary conditions of the specimen are depicted in Fig. 17. The vertical displacement at the

Liang Xue et al.: Preprint submitted to Elsevier Page 39 of 68

 An extended gradient damage model for anisotropic fracture

(a) shear fracture band (b) shear-slip deformation

Figure 18: Double edge-notched concrete specimens under pure shear loading: Mode II failure. (The
domain is scaled by the displacement field with a factor of 20)

450
Linear cohesive law
Exponential cohesive law
400

350

300
Force Fx (N)

250

200

150

100

50

0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
Displacement ux (mm)

Figure 19: Double edge-notched concrete specimens under pure shear loading: Load versus displacement
curves under different cohesive laws

580 bottom of the specimen is constrained, as is the horizontal displacement at the lower right. The
581 load of the specimen is only subjected to the horizontal displacement (𝑢𝑥 ) applied at the upper
582 left. Throughout the loading process, both the horizontal displacement (𝑢𝑥 ) and the corresponding
583 horizontal force (𝐹𝑥 ) are recorded. The material parameters are as follows: Young’s modulus

Liang Xue et al.: Preprint submitted to Elsevier Page 40 of 68

 An extended gradient damage model for anisotropic fracture

584 𝐸0 = 30GPa, and Poisson’s ratio 𝜈 = 0.2, tensile fracture energy 𝐺I = 0.1𝑁∕mm, tensile
585 strength 𝑓t = 3MPa, shear fracture energy 𝐺II = 0.24𝑁∕mm, shear strength 𝑓s = 3MPa, and
586 the characteristic length 𝑙 = 2mm. The mesh size ℎ = 0.5mm. In this example, the shear cohesive
587 law uses linear and exponential forms, respectively.
588 Fig. 18 illustrates the predicted crack paths and deformations, showing that the proposed model
589 is capable of predicting the true mode II of crack extension. The load-displacement curves for
590 different shear cohesive laws are given in Fig. 19, indicating that the EGD model can flexibly control
591 the shear-dominated failure behavior without changing the tensile properties of the material, which
592 is a difficult problem to be overcome by other phase field models.

593 6.5. Double edge-notched concrete specimens under non-proportional loading

594 To validate the capabilities of our proposed approach in capturing complex crack paths, we
595 turn to a series of mixed-mode fracture tests performed by Nooru-Mohamed (1993) on double
596 edge-notched (DEN) plain concrete specimens.
597 Our analysis primarily focuses on experiments DEN-4a, DEN-4b, and DEN-4c, conducted by
598 Nooru-Mohamed (1993). The detailed specimen geometry and boundary conditions are illustrated
599 in Fig. 20. In these experiments, the vertical displacement at the bottom of the specimen was held
600 fixed, as was the horizontal displacement at the lower right. Initially, a rightward horizontal shear
601 force (𝐹𝑠 ) was applied to the top left of the specimen via a rigid steel frame. Subsequently, an upward
602 displacement was introduced at the specimen’s top while maintaining a constant 𝐹𝑠 throughout the
603 loading process. Notably, in the DEN-4a, DEN-4b, and DEN-4c tests, the horizontal shear force
604 𝐹𝑠 was maintained at 5 kN, 10 kN, and 27.5 kN, respectively. The material parameters for these
605 experiments are as follows: tensile fracture energy 𝐺I = 0.1𝑁∕mm, tensile strength 𝑓t = 2.4MPa,
606 shear fracture energy 𝐺II = 0.24𝑁∕mm, shear strength 𝑓s = 4MPa, Young’s modulus 𝐸0 = 30GPa,
607 and Poisson’s ratio 𝜈 = 0.2. The characteristic length is 𝑙 = 2mm, and the mesh size is ℎ = 0.5mm.
608 Both tensile and shear cohesive laws employed exponential softening.

Liang Xue et al.: Preprint submitted to Elsevier Page 41 of 68

 An extended gradient damage model for anisotropic fracture

609 According to the experimental results, two nearly symmetrical cracks propagate from the notch
610 towards opposite sides, with the curvature depending on the magnitude of the initially applied
611 shear force 𝐹𝑠 ; see Fig. 21. However, these experimental findings require reconsideration due
612 to several reasons: (i) The fracture process is actually a three-dimensional phenomenon, while
613 the numerical simulations usually take a two-dimensional assumption. In experiments, there are
614 notable differences between the cracks on the front and back surfaces of the specimen, and the
615 cracks at the top and bottom of the sample are not symmetric. Therefore, the sample does not
616 experience pure membrane action; (ii) It has been reported (Nooru-Mohamed et al., 1993) that there
617 is detachment between the rigid frame and the specimen, resulting in peak loads of the vertical force
618 significantly lower than the predicted values from existing numerical simulations, (iii) To check the
619 repeatability of the test results, a small number of tests were duplicated in a random manner. As
620 can be seen in Fig. 23(b), two tests of 𝐹s∗ were repeated. There are some deviations in the peak
621 behavior of the two specimens that can likely be attributed to the difference in the age of loading
622 (48 and 157 days, respectively). However, similar crack patterns were observed for both specimens.
623 Fig. 22 gives the crack paths predicted by the anisotropic EGD model, replicating the tendency
624 of the curvature to become larger as the shear force becomes larger. This consistency validates
625 the ability of the proposed model to capture complex crack paths. In Fig. 23, we present the
626 global load-displacement curves predicted for the DEN specimens. It is important to note that
627 the same set of material parameters was consistently applied across all shear force levels, and no
628 data calibration was performed. As previously emphasized, the central objective of this research
629 paper is to introduce and validate the proposed approach, demonstrating its capability to predict
630 intricate crack paths in mixed-mode fracture scenarios. Interestingly, in the DENS-4c specimen, it
631 is necessary to rely on the pressure to maintain the stability of the specimen despite the upward
632 displacement applied. Since the loading conditions in the experiments cannot be exactly replicated
633 in the numerical simulations, this undoubtedly leads to discrepancies between the predicted results
634 and the experimental observations. Nevertheless, the predicted results still show a reasonable trend.

Liang Xue et al.: Preprint submitted to Elsevier Page 42 of 68

 An extended gradient damage model for anisotropic fracture

Figure 20: Double edge notched specimens: geometry (unit: mm), boundary condition and experimental
crack paths (left); and finite element mesh (right)

Figure 21: Double edge notched specimens: Experimentally observed cracks

635 Furthermore, to investigate the distinctions between isotropic and anisotropic fracture models,
636 this study employs an isotropic damage model (Wu, 2017; Xue and Ren, 2024a) for predicting
637 crack propagation paths in DENS-4a and assesses the disparities between the Rankine criterion and
638 the Modified von Mises criterion (Wu, 2017). Fig. 24 illustrates that the predictive results of the
639 isotropic damage model are contingent upon the choice of the failure criterion, and an inappropriate
640 selection can result in inaccuracies in the predictions. In this numerical example, the crack paths
641 predicted using the Modified von Mises criterion are very close to those predicted by the anisotropic

Liang Xue et al.: Preprint submitted to Elsevier Page 43 of 68

 An extended gradient damage model for anisotropic fracture

(a) DENS-4a (b) DENS-4b (c) DENS-4c

Figure 22: Double edge notched specimens: The predicted final crack paths by the presented model

16 14 3
Experimental result Experimental result Experimental result
14 Numerical prediction 12 Numerical prediction Numerical prediction
2
12
10
1
Force F y (kN)

Force F y (kN)

Force F y (kN)
10
8
8 0
6
6
-1
4
4
2 -2
2

0 0 -3
0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2
Displacement u y (mm) Displacement u y (mm) Displacement u y (mm)

(a) DENS-4a (b) DENS-4b (c) DENS-4c

Figure 23: Double edge notched specimens: Comparisons between the numerically predicted
load–displacement curves and the experimental results of the tests

642 EGD model. The difference between the two is that the anisotropic damage model predicts a smaller
643 crack curvature, which is more in line with the real test results. However, this result does not prove
644 that the Modified von Mises criterion is a trustworthy failure criterion. In other numerical examples,
645 the Modified von Mises criterion may also lead to incorrect predictions. In contrast, the anisotropic
646 extended gradient damage models require only basic material parameters to accurately predict the
647 failure behavior of members. Therefore, it is imperative to develop anisotropic damage models that
648 align with the actual failure characteristics of materials.

649 6.6. Three-point bending beam with a two-segment notch

650 Next, we simulate a three-point bending beam with multiple notched segments in the span (Wu
651 et al., 2020c). The experimental results show that the ratio of the stress strength factor (𝐾II ∕𝐾I )

Liang Xue et al.: Preprint submitted to Elsevier Page 44 of 68

 An extended gradient damage model for anisotropic fracture

(a) Rankine criterion (b) Modified von Mises criterion

Figure 24: Double edge notched specimens: Prediction of crack paths using isotropic unified phase field
model

652 between mode II and mode I at the top of the notch depends on the inclination angle of the last
653 notch. The geometry and loading conditions of the test are shown in Fig. 25.

Figure 25: Three-point bending beam with a two-segment notch: geometry (unit: mm), boundary
condition.

654 This study selected three test samples with initial crack depths of 𝑎0 = 30mm and inclination
655 angles of 22.5°, 45°, and 67.5°, respectively. The material parameters are as follows: Young’s
656 modulus 𝐸0 = 27GPa, and Poisson’s ratio 𝜈 = 0.2, tensile fracture energy 𝐺I = 0.09𝑁∕mm, tensile
657 strength 𝑓t = 2.4MPa, shear fracture energy 𝐺II = 0.24𝑁∕mm, shear strength 𝑓s = 2.5MPa, and

Liang Xue et al.: Preprint submitted to Elsevier Page 45 of 68

 An extended gradient damage model for anisotropic fracture

658 the characteristic length 𝑙 = 1mm. The mesh size ℎ = 0.2mm. Both tensile and shear cohesive
659 laws employed exponential softening.
660 Fig. 26 compares the predicted crack paths with the tested crack paths for the three samples.
661 The predicted results obtained good agreement with the test results in all the samples with different
662 inclination angles.

(a) 𝛼 = 22.5◦

(b) 𝛼 = 45◦

(c) 𝛼 = 67.5◦

(d) comparison of predicted cracks with test cracks

Figure 26: Cracks in a three-point bending beam with a two-segment notch.

Liang Xue et al.: Preprint submitted to Elsevier Page 46 of 68

 An extended gradient damage model for anisotropic fracture

663 Fig. 27 presents the load vs crack mouth opening displacement (CMOD) curves. Despite
664 employing the same set of material parameters across different samples, the predicted results
665 align consistently with the experimental findings. The load vs. CMOD curves of the observed
666 experiments show differences before the samples, which may be due to the stochastic nature of
667 the mechanical properties of concrete. According to Liang et al. (2022, 2023); Wu et al. (2023),
668 the combination of damage model and random field can well characterize the random fracture
669 phenomenon observed in the tests.

1.6 1.8
Experimental results Experimental results Experimental results
Numerical Prediction 1.6 Numerical Prediction 2 Numerical Prediction
1.4

1.4
1.2
1.2 1.5
1
Force P (kN)

Force P (kN)

Force P (kN)
1
0.8
0.8 1
0.6
0.6
0.4
0.4 0.5

0.2 0.2

0 0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
CMOD (mm) CMOD (mm) CMOD (mm)

(a) 𝛼 = 22.5◦ (b) 𝛼 = 45◦ (c) 𝛼 = 67.5◦

Figure 27: Experimental and numerical load 𝑣𝑠. CMOD curves.

670 6.7. Mixed-model failure test of notched beams

671 Finally, two notched beam tests with mixed-mode failure are considered (Gálvez et al., 1998).
672 The geometry and loading conditions of the test are shown in Fig. 28. In the first beam, the stiffness
673 at the upper left support is zero (specifically, three-point bending with a stiffness value of 0),
674 whereas in the second beam, it is infinite (specifically, four-point bending with a stiffness value
675 of ∞). The load 𝑃 is applied using displacement control, and the CMOD is measured in both
676 scenarios.
677 The material parameters are as follows: Young’s modulus 𝐸0 = 38GPa, and Poisson’s ratio
678 𝜈 = 0.2, tensile fracture energy 𝐺I = 0.065𝑁∕mm, tensile strength 𝑓t = 2.8MPa, shear fracture
679 energy 𝐺II = 0.06𝑁∕mm, shear strength 𝑓s = 2.5MPa, and the characteristic length 𝑙 = 1mm. The
680 mesh size ℎ = 0.2mm. Both tensile and shear cohesive laws employed exponential softening.

Liang Xue et al.: Preprint submitted to Elsevier Page 47 of 68

 An extended gradient damage model for anisotropic fracture

Figure 28: Mixed-model failure test of notched beams: Geometry (unit of length: mm), loading and
boundary conditions.

681 As shown in Fig. 29, the predicted crack paths for the three-point and four-point bending tests
682 agree quite well with the experimental crack patterns. The cracks propagated to the top of the beam
683 at different angles of inclination under different boundary conditions.
684 Fig. 30 depicts the prediction curves of load versus CMOD, which are in good agreement with
685 the experimental results. The above predictions argue for the ability of the EGD model to capture
686 complex fracture modes.

Figure 29: Mixed-model failure test of notched beams: Damage profiles. Left column (three-point
bending): Experimental crack pattern (top); Numerical damage profile (bottom). Right column (four-
point bending): Experimental crack pattern (top); Numerical damage profile (bottom).

687 7. Conclusion

688 This paper adeptly integrates orthogonal anisotropic damage and cohesive fracture theories
689 into a unified framework. It introduces an extended gradient damage model to address the issue of
690 damage unloading resulting from the mismatch of cohesive laws in the phase field model (Pham

Liang Xue et al.: Preprint submitted to Elsevier Page 48 of 68

 An extended gradient damage model for anisotropic fracture

Experimental results Experimental results

6
Numerical Prediction 12 Numerical Prediction

5
10
Force P (kN)

Force P (kN)
8

3 6

2 4

1 2

0 0
0 0.1 0.2 0.3 0.4 0.5 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08
CMOD (mm) CMOD (mm)

(a) three-point bending (b) four -point bending

Figure 30: Mixed-model failure test of notched beams: Curves of load 𝑣𝑠. CMOD.

691 and Marigo, 2013; Xue and Ren, 2024a). By decoupling the cohesive law and the damage evolution
692 strategy, this model guarantees that the choice of the cohesive law no longer impacts the irreversibil-
693 ity of damage. Moreover, the paper introduces a shear-normal energy decomposition to construct
694 a driving force that aligns with cohesive fracture theory. This energy decomposition characterizes
695 traction separation and dislocation glide behaviors on the potential crack surface, enabling precise
696 representation of the cohesive law for both tensile and shear failures. It is important to note that
697 the EGD model is not limited to shear-normal energy decomposition; it is also compatible with
698 other decompositions, including tensile-compression decomposition (Wu et al., 2020b), multiscale
699 energy decomposition (Feng and Li, 2023), and spherical-deviation decomposition (Amor et al.,
700 2009). Consequently, anisotropic damage can be seamlessly incorporated into the EGD model.
701 This study illustrates, through seven numerical examples, that the proposed model can effec-
702 tively capture the complex fracture failure behavior of solids. Examples 1 and 2 validate the ability
703 of the EGD model to predict tensile and shear cohesive laws. Example 3 controls the failure mode
704 of the material by adjusting the shear strength. Example 4 controls the load-displacement curve of
705 a specimen by adjusting the shear cohesive law. Example 5 compares the results predicted by the
706 isotropic fracture model with those of the anisotropic EGD model, highlighting the limitations of
707 the isotropic damage model in predicting complex fracture behavior. Therefore, the incorporation

Liang Xue et al.: Preprint submitted to Elsevier Page 49 of 68

 An extended gradient damage model for anisotropic fracture

708 of anisotropic damage into cohesive fracture theory becomes imperative. Example 6 and Example
709 7 validate the ability of the EGD model to simulate mixed fracture modes. Furthermore, the model
710 introduced in this paper differentiates itself from previous anisotropic damage models. For instance,
711 Miehe et al. (2010); Wu et al. (2020b) concentrated solely on tensile failure behavior, restricting
712 their ability to predict shear failure. Feng and Li (2023) mandated consistency between shear
713 and tensile cohesive laws, thereby limiting the model’s capacity to assign distinct properties to
714 materials under various loading conditions. In quasi-brittle materials, the damage direction may
715 gradually shift as the stress direction changes during loading, causing the stress-strain principal axis
716 to no longer be coaxial. Conversely, current anisotropic damage models still assume that the stress-
717 strain axis remains coaxial, which may not align with the actual physical behavior of engineering
718 materials (Sellier and Bary, 2002). The shear-normal energy decomposition proposed in this paper
719 permits the stress-strain principal axis to deviate from coaxial, enhancing the model’s alignment
720 with the physical properties of engineering materials.
721 Finally, it is important to note that the thermodynamic framework of the proposed model
722 is established following the principles of energy conservation and stability conditions (Pham
723 et al., 2011; Feng and Li, 2022a). A thermodynamically consistent anisotropic fracture model is
724 established through carefully constructed dissipative forces.

725 A. A thermodynamic framework for considering strain rate and temperature effects

726 Under the action of heat and force, the total strain 𝜖 𝑒 usually consists of the elastic strain 𝜖 𝑒 and
727 the free thermal strain 𝜖 𝜃 .

𝜖 = 𝜖𝑒 + 𝜖𝜃 (A.1)

728 The free thermal strain rate is caused by the thermal expansion of concrete and is generally
729 expressed as:

Liang Xue et al.: Preprint submitted to Elsevier Page 50 of 68

 An extended gradient damage model for anisotropic fracture

𝜖̇ 𝜃 = 𝛼E (𝜃) ⋅ 𝜃̇ ⋅ 𝟏 (A.2)

730 where 𝜖 = ∇sym 𝑢, 𝟏 is the second-order identity tensor, 𝛼E is the coefficient of thermal expansion
731 of the material, 𝜃 is the temperature field.
732 The simulation of concrete damage must comply with the first law of thermodynamics. When
733 considering strain rate and temperature effects, the rate of change of the internal energy of the
734 system (including kinetic energy, potential energy, dissipated energy, and thermal energy) should
735 equal the sum of the rate of work done by external forces and the rate of external thermal energy
736 input (Wang et al., 2020, 2022; Zhou et al., 2024). This can be expressed as:

K̇ + Ė + Ḋ + Q̇ = Ḟ + Ẇ (A.3)

737 where K is the kinetic energy, E is the potential energy, D is dissipated energy, Q is the stored
738 thermal energy, F is the external work, and is the input thermal energy.
739 In the dynamic cracking process, accounting for both macroscopic kinetic energy and the
740 influences of rate-dependent and micro-inertial effects is essential. Therefore, the kinetic energy
741 (Hai and Li, 2021; Hai et al., 2024) of the system can be defined as:

ˆ ˆ
𝜌𝑢̇ 2 𝑘𝜙̇ 2
K = d𝑉 + d𝑉 (A.4)
 2  2

742 where 𝜌 is the mass density; 𝑢̇ is the displacement field rate; 𝜙̇ is the damage field rate; 𝑘 is
743 the hypothetical microscopic mass density to quantify the micro-inertial effect of the dynamic
744 expansion of cracks, considered as a material constant.
745 The potential energy E of the system of a cracked solid usually includes the Helmholtz free
746 energy and the fracture energy. Therefore, the internal energy function can be expressed as:

ˆ
E = 𝛹 (𝜖 e , 𝜙, ∇𝜙, 𝜃) d𝑉 (A.5)


Liang Xue et al.: Preprint submitted to Elsevier Page 51 of 68

 An extended gradient damage model for anisotropic fracture
[ ]
747 In time period 𝑡1 , 𝑡2 , the dissipative energy can generally be expressed as a function of damage
748 and strain processes:

ˆ 𝑡2 ˆ
D= ̇ d𝑡
𝑓diss 𝜙d𝑉 (A.6)
𝑡1 

749 The thermal energy stored in a solid can be expressed as:

ˆ
O= 𝜌𝑐p 𝜃d𝑉 (A.7)


750 where 𝑐p is the specific heat capacity of the concrete.

[ ]
751 The total heat input to the solid in time period 𝑡1 , 𝑡2 consists of the heat flowing in/out through
752 the boundaries 𝜕𝜃 and 𝜕𝐽 and the energy emitted by the heat source 𝛶 , i.e.

ˆ 𝑡2 ( ˆ ˆ ˆ )
F = − 𝐽 ⋅ 𝑛d𝑆 − ∗
𝐽 d𝑆 + 𝛶 d𝑉 d𝑡 (A.8)
𝑡1 𝜕𝜃 𝜕𝐽 

753 By substituting Fourier’s law of heat conduction 𝐽 = −𝑘ef f ⋅ ∇𝜃 into Eq. (A.8) and applying
754 the Gauss divergence theorem with consideration of 𝜕𝐽 ∪ 𝜕𝜃 = 𝜕 and 𝜕𝐽 ∩ 𝜕𝜃 = ∅, one
755 obtains:

ˆ 𝑡2 (ˆ ˆ ˆ ˆ )
F = 𝑘ef f ∇ ⋅ ∇𝜃d𝑉 − 𝑘ef f ∇𝜃 ⋅ 𝑛d𝑆 − ∗
𝐽 d𝑆 + 𝛶 d𝑉 d𝑡 (A.9)
𝑡1 𝜕 𝜕𝐽 𝜕𝐽 

756 where 𝑘ef f is the effective heat conductivity.

[ ]
757 During the time period 𝑡1 , 𝑡2 , the work done by external forces, including the body force, is
758 expressed as:

ˆ 𝑡2 (ˆ ˆ ˆ )
W = ∗
𝑏 ⋅ 𝑢d𝑉
̇ + ∗
𝑡 ⋅ 𝑢d𝐴
̇ − 𝜂 ⋅ 𝜙̇ ⋅ 𝜙d𝑉
̇ d𝑡 (A.10)
𝑡1  𝜕 

Liang Xue et al.: Preprint submitted to Elsevier Page 52 of 68

 An extended gradient damage model for anisotropic fracture

759 where 𝜂 is a material constant used to describe the influence of viscosity-dependent damping effects
760 on the evolution of the damage, which has a negligible effect on the results when a small value is
761 taken.
762 Substituting Eqs. (A.4), (A.5), (A.6), (A.7), (A.9), and (A.10) into Eq. (A.3), one obtains:

ˆ ˆ ˆ ˆ ˆ ( )
𝜕𝛹 𝜕𝛹
𝜌𝑢̈ 𝑢d𝑉
̇ + 𝑘𝜙̈ 𝜙d𝑉 ̇ + 𝜂 𝜙̇ 𝜙d𝑉 ̇ + e
⋅ 𝑛 ⋅ 𝑢d𝐴
̇ − ∇ ⋅ ⋅ 𝑢d𝑉
̇
   𝜕 𝜕𝜖  𝜕𝜖 e
ˆ ˆ ˆ
( ) 𝜕𝛹
+ ̇
𝛹 + 𝑓diss 𝜙d𝑉 + ̇
⋅ 𝑛 ⋅ 𝜙d𝐴 + 𝜌𝑐p 𝜃d𝑉 ̇ =
 𝜕 𝜕∇𝜙 
ˆ ˆ ˆ ˆ (A.11)
∗
𝑘ef f ∇ ⋅ ∇𝜃d𝑉 − 𝑘ef f ∇𝜃 ⋅ 𝑛d𝑆 − 𝐽 d𝑆 + 𝛶 d𝑉
𝜕 𝜕𝐽 𝜕𝐽 
ˆ ˆ
+ 𝑏∗ ⋅ 𝑢d𝑉 ̇ + 𝑡∗ ⋅ 𝑢d𝐴
̇
 𝜕

763 The thermodynamics laws assert that Eq. (A.11) is applicable for any 𝑢,̇ 𝜙,
̇ and 𝜃.
̇ Therefore,

764 the strong form of the governing equations for each physical field and the corresponding boundary
765 conditions can be derived as follows:

⎧ ∇ ⋅ 𝜎 + 𝑏∗ = 𝜌𝑢̈ in (A.12a)

⎪
⎪ ̈
⎨ 𝑘𝜙 + 𝜂 𝜙̇ + 𝛹 + 𝑓diss = 0 in (A.12b)
⎪
⎪ 𝑘ef f ∇ ⋅ ∇𝜃 + 𝛶 = 0 in (A.12c)
⎩

766 The corresponding boundary conditions are:

⎧ 𝑛 ⋅ 𝜎 = 𝑡∗ on (A.13a)
⎪
⎪ 𝜕𝛹
⎨ 𝜕∇𝜙 ⋅ 𝑛 = 0 on (A.13b)
⎪
⎪ 𝑘ef f ∇𝜃 ⋅ 𝑛 + 𝐽 ∗ = 0 on (A.13c)
⎩

Liang Xue et al.: Preprint submitted to Elsevier Page 53 of 68

 An extended gradient damage model for anisotropic fracture

767 The above content provides a thermodynamic framework considering the influence of strain rate
768 and temperature on fracture behavior and outlines the derivation process of governing equations
769 and boundary conditions. It is evident that the extended gradient damage model can be effectively
770 integrated with multiple physical fields. It’s worth noting that in the aforementioned process, certain
771 material parameters such as elastic modulus, strength, fracture energy, and heat conductivity may
772 be functions related to temperature or damage rather than fixed values. Interested readers can refer
773 to the works of Wang et al. (2020, 2022); Zhou et al. (2024) for further exploration of this aspect.

774 B. Crack surfaces under 3D stress conditions

775 When the stress state is 3-dimensional, it is extremely complicated to find the potential crack
776 surface and the corresponding stress state. The relationship between the stress tensor and the
777 principal stress tensor on a potential crack surface can be characterized as:

𝑇
⎡ ⎤ ⎡ ⎤
⎢𝑙11 𝑙12 𝑙13 ⎥ ⎢𝑙11 𝑙12 𝑙13 ⎥
⎢ ⎥ ⎢ ⎥
𝜎̄ crack = ⎢𝑙21 𝑙22 𝑙23 ⎥ 𝜎̄ p ⎢𝑙21 𝑙22 𝑙23 ⎥ (B.1)
⎢ ⎥ ⎢ ⎥
⎢𝑙31 𝑙32 𝑙33 ⎥ ⎢𝑙31 𝑙32 𝑙33 ⎥
⎣ ⎦ ⎣ ⎦
778 where 𝜎̄ crack is the stress tensor when the potential crack surface is used as the coordinate system,
779 and 𝜎̄ P is the stress tensor when the direction of the principal stress is used as the coordinate
⟨ ⟩
780 system, and 𝑙𝑖𝑗 = cos 𝑃𝑖 , 𝑒𝑗 . 𝜎̂̄ 𝑖 is the 𝑖-th principal stress, 𝑃𝑖 is the unit vector of the eigenvector
781 corresponding to the 𝑖-th principal stress. 𝑒𝑖 is the unit vector corresponding to the 𝑖-th coordinate
782 axis when the potential crack surface is used as the coordinate system.
783 In general, 𝑃𝑖 is easy to find. While the derivation of 𝑒𝑖 needs to be completed with the assistance
784 of Rodrigues’ rotation formula (Murray et al., 2017). Similar to the 2D stress state, the constraint
785 on the potential crack surface is that the normal stress on at least one of the other two surfaces
786 perpendicular to the crack surface is zero.
787 In the following, the authors give the solution procedure of 𝑒𝑖 for various stress states.

Liang Xue et al.: Preprint submitted to Elsevier Page 54 of 68

 An extended gradient damage model for anisotropic fracture

788 B.1. 𝜎̂̄ 1 ≥ 𝜎̂̄ 2 ≥ 0 ≥ 𝜎̂̄ 3

789 According to Rodrigues’ rotation formula, 𝑒𝑖 can be expressed as:

⎧ 𝑒 = 𝑝 cos 𝜃 + (𝑝 × 𝑝 ) sin 𝜃 + 𝑝 (𝑝 ⋅ 𝑝 ) (1 − cos 𝜃 ) (B.2a)

⎪ 3 2 1 1 2 1 1 1 2 1
⎪ ( ) ( )( )
⎨ 𝑒2 = 𝑎2 cos 𝜃2 + 𝑒̄3 × 𝑎2 sin 𝜃2 + 𝑒̄3 𝑒̄3 ⋅ 𝑎2 1 − cos 𝜃2 (B.2b)
⎪
⎪𝑒 = 𝑒 ×𝑒 (B.2c)
⎩ 1 2 3

790 where 𝑎2 = 𝑝1 ×𝑒3 , ⋅ is the dot product of vectors, and × is the cross product of vectors. 𝑒̄𝑖 is the unit
791 vector of vector 𝑒𝑖 . 𝜃𝑖 is the angle of rotation between the principal stress space and the potential
792 crack surface.
793 The values of the rotation angle 𝜃𝑖 under different stress conditions are as follows:
794 Case 1: when 𝜎̂̄ 2 ≥ ||𝜎̂̄ 3 ||, the relationship between the potential crack surface and principal stress
795 is shown in Fig. 31.

Figure 31: Relationship between potential crack surface and principal stress.

796 According to Fig. 31, the rotation angle 𝜃𝑖 is,

Liang Xue et al.: Preprint submitted to Elsevier Page 55 of 68

 An extended gradient damage model for anisotropic fracture

⎧ √
⎛ ⎞
⎪ 1 ⎜ 2 𝜎̂̄ 2 ||𝜎̂̄ 3 || ⎟
⎪ 𝜃1 = arc cos ⎜ ⎟ (B.3a)
⎨ 2 ⎜ 𝜎̂̄ 3 − 𝜎̂̄ 2 ⎟
⎪ ⎝ ⎠
⎪𝜃 = 0 (B.3b)
⎩ 2

797 Case 2: when 𝜎̂̄ 1 + 𝜎̂̄ 2 ≥ ||𝜎̂̄ 3 || ≥ 𝜎̂̄ 2 , the relationship between the potential crack surface and
798 principal stress is shown in Fig. 32.

Figure 32: Relationship between potential crack surface and principal stress.

799 According to Fig. 32, the rotation angle 𝜃𝑖 is,

⎧ √
⎛ ⎞
⎪ ⎜ 2 𝜎̄ 2 ||𝜎̄ 3 || ⎟
1
⎪ 𝜃1 = arc cos ⎜ ⎟ (B.4a)
⎪ 2 ⎜ 𝜎̄ 2 − 𝜎̄ 3 ⎟
⎪ ⎝ ⎠
⎨ √
⎪ ⎛ ⎞
⎜ 2 𝜎̄ 1 ||𝜎̄ n1 || ⎟
⎪ 𝜃 = arc cos
1
(B.4b)
⎪ 2 2 ⎜ 𝜎̄ − 𝜎̄ ⎟
⎪ ⎜ n1 1 ⎟
⎩ ⎝ ⎠

√( )2
− 𝜎̂̄ 2 ||𝜎̂̄ 3 ||.
𝜎̂̄ 2 −𝜎̂̄ 3
800 where 𝜎̂̄ n1 = − 2

Liang Xue et al.: Preprint submitted to Elsevier Page 56 of 68

 An extended gradient damage model for anisotropic fracture

801 Case 3: when ||𝜎̂̄ 3 || ≥ 𝜎̂̄ 1 + 𝜎̂̄ 2 , the relationship between the potential crack surface and principal
802 stress is shown in Fig. 33.

Figure 33: Relationship between potential crack surface and principal stress.

803 According to Fig. 33, the rotation angle 𝜃𝑖 is,

⎧ √
⎛ ̂ | ̂ |⎞
⎪ 1 ⎜ 2 𝜎̄ 2 |𝜎̄ 3 | ⎟
⎪ 𝜃1 = arc cos ⎜ ⎟ (B.5a)
⎪ 2 ⎜ 𝜎̂̄ 2 − 𝜎̂̄ 3 ⎟
⎪ ⎝ ⎠
⎨ √
⎪ ⎛ ⎞
2 𝜎̂̄ 1 ||𝜎̂̄ n1 || ⎟
⎪ 𝜃 = 1 arc cos ⎜ (B.5b)
⎪ 2 2 ⎜ ̂ ⎟
⎜ 𝜎̄ 1 − 𝜎̂̄ n1 ⎟
⎪ ⎝ ⎠
⎩

√( )2
− 𝜎̂̄ 2 ||𝜎̂̄ 3 ||.
𝜎̂̄ 2 −𝜎̂̄ 3
804 where 𝜎̂̄ n1 = − 2

805 B.2. 𝜎̂̄ 1 ≥ 0 ≥ 𝜎̂̄ 2 ≥ 𝜎̂̄ 3

806 According to Rodrigues’ rotation formula, 𝑒𝑖 can be expressed as:

Liang Xue et al.: Preprint submitted to Elsevier Page 57 of 68

 An extended gradient damage model for anisotropic fracture

⎧ 𝑒 = 𝑝 cos 𝜃 + (𝑝 × 𝑝 ) sin 𝜃 + 𝑝 (𝑝 ⋅ 𝑝 ) (1 − cos 𝜃 ) (B.6a)

⎪ 3 1 1 3 1 1 3 3 1 1
⎪ ( ) ( )( )
⎨ 𝑒2 = 𝑎2 cos 𝜃2 + 𝑒̄3 × 𝑎2 sin 𝜃2 + 𝑒̄3 𝑒̄3 ⋅ 𝑎2 1 − cos 𝜃2 (B.6b)
⎪
⎪𝑒 = 𝑒 ×𝑒 (B.6c)
⎩ 1 2 3

807 where 𝑎2 = 𝑝1 × 𝑒3 .
808 The values of the rotation angle under different stress conditions are as follows:
809 Case 1: when ||𝜎̂̄ 2 || ≥ 𝜎̂̄ 1 , the relationship between the potential crack surface and principal stress
810 is shown in Fig. 34.

Figure 34: Relationship between potential crack surface and principal stress.

811 According to Fig. 34, the rotation angle 𝜃𝑖 is,

⎧ √
⎛ ̂ | ̂ |⎞
⎪ 1 ⎜ 2 𝜎̄ 1 |𝜎̄ 2 | ⎟
⎪ 𝜃1 = arc cos ⎜ ⎟ (B.7a)
⎨ 2 ⎜ 𝜎̂̄ 1 − 𝜎̂̄ 2 ⎟
⎪ ⎝ ⎠
⎪𝜃 = 0 (B.7b)
⎩ 2

812 Case 2: when ||𝜎̂̄ 2 + 𝜎̂̄ 3 || ≥ 𝜎̂̄ 1 ≥ ||𝜎̂̄ 2 ||, the relationship between the potential crack surface and
813 principal stress is shown in Fig. 35.

Liang Xue et al.: Preprint submitted to Elsevier Page 58 of 68

 An extended gradient damage model for anisotropic fracture

Figure 35: Relationship between potential crack surface and principal stress.

814 According to Fig. 35, the rotation angle 𝜃𝑖 is,

⎧ √
⎛ ̂ | ̂ |⎞
⎪ 1 ⎜ 2 𝜎̄ 1 |𝜎̄ 2 | ⎟
⎪ 𝜃1 = arc cos ⎜ ⎟ (B.8a)
⎪ 2 ⎜ 𝜎̂̄ 2 − 𝜎̂̄ 1 ⎟
⎪ ⎝ ⎠
⎨ √
⎪ ⎛ ⎞
⎜ 2 𝜎̂̄ n1 ||𝜎̂̄ 3 || ⎟
⎪ 𝜃 = arc cos
1
(B.8b)
⎪ 2 2 ⎜ ̂ ⎟
⎪ ⎜ 𝜎̄ n1 − 𝜎̂̄ 3 ⎟
⎩ ⎝ ⎠

√( )2
− 𝜎̂̄ 1 ||𝜎̂̄ 2 ||.
𝜎̂̄ 1 −𝜎̂̄ 2
815 where 𝜎̂̄ n1 = 2

816 Case 3: when 𝜎̂̄ 1 ≥ ||𝜎̂̄ 2 + 𝜎̂̄ 3 ||, the relationship between the potential crack surface and principal
817 stress is shown in Fig. 36.
818 According to Fig. 36, the rotation angle 𝜃𝑖 is,

Liang Xue et al.: Preprint submitted to Elsevier Page 59 of 68

 An extended gradient damage model for anisotropic fracture

Figure 36: Relationship between potential crack surface and principal stress.

⎧ √
⎛ ̂ | ̂ |⎞
⎪ 1 ⎜ 2 𝜎̄ 1 |𝜎̄ 2 | ⎟
⎪ 𝜃1 = arc cos ⎜ ⎟ (B.9a)
⎪ 2 ⎜ 𝜎̂̄ 2 − 𝜎̂̄ 1 ⎟
⎪ ⎝ ⎠
⎨ √
⎪ ⎛ ⎞
⎜ 2 𝜎̂̄ n1 ||𝜎̂̄ 3 || ⎟
⎪ 𝜃 = arc cos
1
(B.9b)
⎪ 2 2 ⎜ ̂ ⎟
⎪ ⎜ 𝜎̄ 3 − 𝜎̂̄ n1 ⎟
⎩ ⎝ ⎠

√( )2
− 𝜎̂̄ 1 ||𝜎̂̄ 2 ||.
𝜎̂̄ 1 −𝜎̂̄ 2
819 where 𝜎̂̄ n1 = 2

820 Once all the rotation angles have been determined, the constitutive relations for the three-
821 dimensional stress state can be obtained by following the steps presented in Section 3.3.

822 C. Required 𝑠 (𝜙∗ ) and 𝜔 (𝜙) for some commonly used cohesive laws

823 In this section, we provide explicit expressions for 𝑠 (𝜙∗ ) and 𝜔 (𝜙) for various commonly
824 employed cohesive laws, encompassing both linear and exponential forms. Additionally, we offer
825 fitting methods for more intricate cohesive laws, such as polynomial and Cornelissen cohesive laws.

826 C.1. Linear softening

827 The parametric equation set for the linear softening cohesive law is:

Liang Xue et al.: Preprint submitted to Elsevier Page 60 of 68

 An extended gradient damage model for anisotropic fracture

{
𝛿n = 𝑠 (𝜙∗ ) 𝛿ch,n = 𝜙∗ 𝛿ch,n (C.1a)

𝜎 = (1 − 𝜙∗ ) 𝑓t (C.1b)

828 By substituting Eq. (C.1a) into Eq. (51), we obtain 𝜔 (𝜙) for the linear softening cohesive law
829 as:

1 (2𝜙 − 𝜙2 )
𝜔 (𝜙) = √ (C.2)
𝜋 2 (1 − 𝜙)2

830 C.2. Exponential softening

831 The parametric equation set for the exponential softening cohesive law is:

⎧ 1
∗ ∗
⎪ 𝛿n = 𝑠 (𝜙 ) 𝛿ch,n = − ln (1 − 𝜙 ) 𝛿ch,n (C.3a)
⎨ 2
⎪ 𝜎 = (1 − 𝜙∗ ) 𝑓 (C.3b)
⎩ t

832 By substituting Eq. (C.3a) into Eq. (51), we obtain 𝜔 (𝜙) for the exponential softening cohesive
833 law as:

√
(2𝜙 − 𝜙2 ) (√ )
𝜔 (𝜙) = √ arc tanh 2𝜙 − 𝜙2 (C.4)
2 2𝜋 (1 − 𝜙)2
834 C.3. Polynomial softening

835 The parametric equation set for the polynomial softening law (Feng et al., 2021) is:

⎧ ∑𝑁
⎪ 𝛿n = 𝑠 (𝜙∗ ) 𝛿ch,n = −𝛿ch,n 𝑏𝑛 (1 − 𝜙∗ )𝑛 (C.5a)
⎨ 𝑛=0
⎪
⎩ 𝜎 = (1 − 𝜙 ∗
) 𝑓t (C.5b)

836 where

Liang Xue et al.: Preprint submitted to Elsevier Page 61 of 68

 An extended gradient damage model for anisotropic fracture

𝛿f ∑ ∑
𝑁 𝑁
𝑛 1
𝑏0 = − , 𝑏𝑛 = 0 and 𝑏𝑛 = (C.6)
𝛿ch 𝑛=0 𝑛=1
𝑛+1 2

837 where 𝑏𝑛 is the parameter to be fitted, 𝛿f represents the final crack opening, 𝑁 ∈ ℤ+ denotes the
838 polynomial order, and ℤ+ represents the set of positive integers. In Eq. (B.6), the first condition
839 occurs when the crack opening reaches 𝛿f , and the fracture zone attains maximum damage, i.e.,
840 𝛿 ∣𝜙̄ ∗ =1 = 𝛿f . The second condition ensures that the initial crack opening is zero, i.e., 𝛿 ∣𝜙∗ =0 = 0.
841 The third condition ensures that

ˆ 𝛿f ˆ 1
d𝛿n ∗
𝐺I∕II = 𝜎d𝛿n = 𝜎 d𝜙 (C.7)
0 0 d𝜙∗

842 By substituting Eq. (C.5a) into Eq. (51), we obtain 𝜔 (𝜙) for the exponential softening cohesive
843 law as:

√ ⎡ ˆ ⎤
2𝜙 − 𝜙2 ∑ ⎢ −𝑏𝑛 𝜙
𝑁
(1 − 𝜙∗ )𝑛 ⎥
𝜔 (𝜙) = √ ⎢ d𝜙 √ d𝜙∗ ⎥ (C.8)
2𝜋 (1 − 𝜙) 𝑛=0 ⎢ 0
(1 − 𝜙∗ )2 − (1 − 𝜙)2 ⎥
⎣ ⎦
844 The parameters of the 3rd-order polynomial cohesive law in Section 5.2 of this paper can be
845 calibrated to 𝑏0 = −1.25, 𝑏1 = 3.5, 𝑏2 = −5.25, and 𝑏3 = 3 according to the above steps.

846 C.4. Cornelissen cohesive law

847 Following the method presented in Appendix C.3, the Cornelissen cohesive law can be fitted
848 to a polynomial form where the parameter 𝑏𝑛 can be fitted as 𝑏0 = −2.5193, 𝑏1 = 13.6107, 𝑏2 =
849 −30.3544, 𝑏3 = 21.5536, 𝑏4 = 21.2455, 𝑏5 = −39.4196, 𝑏6 = 15.8834. When combined with Eq.
850 (C.8), the Cornelissen cohesive law can be explicitly expressed as follows:

√
2𝜙 − 𝜙2 ∑
6
𝜔 (𝜙) = √ 𝑏𝑛 𝛽𝑛 (𝜙) (C.9)
2𝜋 (1 − 𝜙) 𝑛=0
851 with (Feng et al., 2021)

Liang Xue et al.: Preprint submitted to Elsevier Page 62 of 68

 An extended gradient damage model for anisotropic fracture

⎧
⎪ 𝛽0 (𝜙) = − 1
⎪ 𝑎𝑠
⎪ 𝑎
⎪ 𝛽1 (𝜙) = − 𝑠
⎪
⎪ 𝛽 (𝜙) = − 𝑎 + 𝑎 ⋅ arc tanh (𝑠)
⎪ 2 𝑠
⎪
⎨ 𝛽3 (𝜙) = 𝑎−2𝑎
3
(C.10)
⎪ 𝑠
⎪ 𝑎−3𝑎3 3𝑎3
⎪ 𝛽4 (𝜙) = 2𝑠 + 2 arc tanh (𝑠)
⎪
⎪ 𝛽 (𝜙) = 𝑎+4𝑎3 −8𝑎5
⎪ 5 3𝑠
⎪ 2𝑎+5𝑎3 −15𝑎5 15𝑎5
⎪ 𝛽6 (𝜙) = + arc tanh (𝑠)
⎩ 8𝑠 8

√
852 where 𝑎 = 1 − 𝜙 and 𝑠 = 1 − 𝑎2 .

853 C.5. Park–Paulino–Roesler cohesive law

854 The Park-Paulino-Roesler (PPR) (Park et al., 2009) cohesive law represents the failure behavior
855 of most materials, with its softening curve taking different shapes depending on the selected
856 cohesive law parameter 𝛼. Specifically, when 𝛼 < 2, the softening curve is convex; when 𝛼 = 2,
857 the softening curve is linear; and when 𝛼 > 2, the softening curve is concave. The PPR cohesive
858 law is expressed as follows,

( )𝛼−1
𝑓t
𝜎 = 𝑓t 1 − 𝛿 (C.11)
𝛼𝐺I n

859 For such complex cohesive laws, we can use polynomial fitting. We selected the cases where
860 𝐺I = 0.1, 𝛼 is 1.3, 2.0, and 2.8 to fit the parameters of the polynomial. When 𝛼 = 1.3,
861 𝑏0 = −0.6504, 𝑏1 = 0.0066, 𝑏2 = −0.0511, 𝑏3 = 0.4672, 𝑏4 = 0.2903, 𝑏5 = −0.0746, 𝑏6 = 0.0122.
862 When 𝛼 = 2, 𝑏0 = −1, 𝑏1 = 1, 𝑏2 = 0, 𝑏3 = 0, 𝑏4 = 0, 𝑏5 = 0, 𝑏6 = 0. When 𝛼 = 2.8,
863 𝑏0 = −1.3446, 𝑏1 = 4.8779, 𝑏2 = −19.599, 𝑏3 = 55.7066, 𝑏4 = −85.3668, 𝑏5 = 65.2377, 𝑏6 =
864 −19.5117.

Liang Xue et al.: Preprint submitted to Elsevier Page 63 of 68

 An extended gradient damage model for anisotropic fracture

865 References
866 Alnæs, M., Blechta, J., Hake, J., Johansson, A., Kehlet, B., Logg, A., Richardson, C., Ring, J., Rognes, M.E., Wells, G.N., 2015. The fenics project

867 version 1.5. Archive of numerical software 3.

868 Ambati, M., Gerasimov, T., De Lorenzis, L., 2015. A review on phase-field models of brittle fracture and a new fast hybrid formulation.

869 Computational Mechanics 55, 383–405.

870 Amor, H., Marigo, J.J., Maurini, C., 2009. Regularized formulation of the variational brittle fracture with unilateral contact: Numerical experiments.

871 Journal of the Mechanics and Physics of Solids 57, 1209–1229.

872 Aydiner, I.U., Tatli, B., Yalçinkaya, T., 2024. Investigation of failure mechanisms in dual-phase steels through cohesive zone modeling and crystal

873 plasticity frameworks. International Journal of Plasticity , 103898.

874 Bazant, Z.P., Pijaudier-Cabot, G., 1988. Nonlocal continuum damage, localization instability and convergence. Journal of Applied Mechanics 55,

875 287–293.

876 Bazant, Z.P., Planas, J., 1997. Fracture and size effect in concrete and other quasibrittle materials. volume 16. CRC press.

877 de Borst, R., Verhoosel, C.V., 2016. Gradient damage vs phase-field approaches for fracture: Similarities and differences. Computer Methods in

878 Applied Mechanics and Engineering 312, 78–94.

879 Bourdin, B., Francfort, G.A., Marigo, J.J., 2000. Numerical experiments in revisited brittle fracture. Journal of the Mechanics and Physics of Solids

880 48, 797–826.

881 Cai, W., Sun, C., Wang, C., Qian, L., Li, Y., Fu, M., 2022. Modelling of the intergranular fracture of twip steels working at high temperature by

882 using czm–cpfe method. International Journal of Plasticity 156, 103366.

883 de Carvalho, M.V., Lopes, I.R., Pires, F.A., 2023. A multi-scale formulation for polycrystalline materials accounting for cohesive micro-cracks:

884 Homogenisation of the traction-separation law. International Journal of Plasticity 171, 103780.

885 Chaboche, J.L., 1988. Continuum damage mechanics: Part i—general concepts. Journal of Applied Mechanics 55, 59–64.

886 Chen, J., Ren, Y., Lu, G., 2021. Meso-scale physical modeling of energetic degradation function in the nonlocal macro-meso-scale consistent

887 damage model for quasi-brittle materials. Computer Methods in Applied Mechanics and Engineering 374, 113588.

888 Elices, M., Guinea, G., Gomez, J., Planas, J., 2002. The cohesive zone model: advantages, limitations and challenges. Engineering fracture mechanics

889 69, 137–163.

890 Erdogan, F., Sih, G., 1963. On the crack extension in plates under plane loading and transverse shear .

891 Fei, F., Choo, J., 2021. Double-phase-field formulation for mixed-mode fracture in rocks. Computer Methods in Applied Mechanics and Engineering

892 376, 113655.

893 Feng, Y., Fan, J., Li, J., 2021. Endowing explicit cohesive laws to the phase-field fracture theory. Journal of the Mechanics and Physics of Solids

894 152, 104464.

895 Feng, Y., Freddi, F., Li, J., Ma, Y.E., 2024. Phase-field model for 2d cohesive-frictional shear fracture: An energetic formulation. Journal of the

896 Mechanics and Physics of Solids , 105687.

897 Feng, Y., Li, J., 2022a. Phase-field cohesive fracture theory: A unified framework for dissipative systems based on variational inequality of virtual

898 works. Journal of the Mechanics and Physics of Solids 159, 104737.

899 Feng, Y., Li, J., 2022b. Phase-field method with additional dissipation force for mixed-mode cohesive fracture. Journal of the Mechanics and Physics

900 of Solids 159, 104693.

901 Feng, Y., Li, J., 2023. A unified regularized variational cohesive fracture theory with directional energy decomposition. International Journal of

902 Engineering Science 182, 103773.

Liang Xue et al.: Preprint submitted to Elsevier Page 64 of 68

 An extended gradient damage model for anisotropic fracture

903 Francfort, G.A., Marigo, J.J., 1998. Revisiting brittle fracture as an energy minimization problem. Journal of the Mechanics and Physics of Solids

904 46, 1319–1342.

905 Freddi, F., Iurlano, F., 2017. Numerical insight of a variational smeared approach to cohesive fracture. Journal of the Mechanics and Physics of

906 Solids 98, 156–171.

907 Freddi, F., Royer-Carfagni, G., 2010. Regularized variational theories of fracture: a unified approach. Journal of the Mechanics and Physics of

908 Solids 58, 1154–1174.

909 Gálvez, J., Elices, M., Guinea, G., Planas, J., 1998. Mixed mode fracture of concrete under proportional and nonproportional loading. International

910 Journal of Fracture 94, 267–284.

911 Gao, L.T., Feng, X.Q., Gao, H., 2009. A phase field method for simulating morphological evolution of vesicles in electric fields. Journal of

912 Computational Physics 228, 4162–4181.

913 Gao, Y., Li, B., Wang, J., Feng, X.Q., 2021. Fracture toughness analysis of helical fiber-reinforced biocomposites. Journal of the Mechanics and

914 Physics of Solids 146, 104206.

915 Goldstein, H., Poole, C., Safko, J., 2002. Classical mechanics.

916 Hai, L., Li, J., 2021. A rate-dependent phase-field framework for the dynamic failure of quasi-brittle materials. Engineering Fracture Mechanics

917 252, 107847.

918 Hai, L., Wriggers, P., Huang, Y.j., Zhang, H., Xu, S.l., 2024. Dynamic fracture investigation of concrete by a rate-dependent explicit phase field

919 model integrating viscoelasticity and micro-viscosity. Computer Methods in Applied Mechanics and Engineering 418, 116540.

920 Hibbeler, R.C., 2006. Mecánica de materiales. Pearson educación.

921 Hughes, T.J., 2012. The finite element method: linear static and dynamic finite element analysis. Courier Corporation.

922 Kardar, M., 2007. Statistical physics of particles. Cambridge University Press.

923 Kong, L., Xie, H., Li, C., 2023. Coupled microplane and micromechanics model for describing the damage and plasticity evolution of quasi-brittle

924 material. International Journal of Plasticity 162, 103549.

925 Lawn, B., 1993. Fracture of brittle solids. Cambridge University Press , 194.

926 Le, J.L., Manning, J., Labuz, J.F., 2014. Scaling of fatigue crack growth in rock. International Journal of Rock Mechanics and Mining Sciences 72,

927 71–79.

928 Li, W.J., Zhu, Q.Z., 2024. An innovative quasi-bond approach to bridge continuity, anisotropic damage and macroscopic fracture of solids and

929 structures. International Journal of Plasticity 172, 103829.

930 Li, Z., Lu, Y., Huang, D., Rabczuk, T., 2024. Nonlocal anisotropic model for deformation and fracture using peridynamic operator method.

931 International Journal of Mechanical Sciences 268, 109023.

932 Li, Z., Shen, Y., Han, F., Yang, Z., 2021. A phase field method for plane-stress fracture problems with tension-compression asymmetry. Engineering

933 Fracture Mechanics 257, 107995.

934 Liang, Y.P., Feng, D.C., Ren, X., Li, J., 2023. Three-stage non-gaussian homogeneous random field representation over manifolds. Computer-Aided

935 Civil and Infrastructure Engineering 38, 1462–1482.

936 Liang, Y.P., Ren, X., Feng, D.C., 2022. Efficient stochastic finite element analysis of irregular wall structures with inelastic random field properties

937 over manifold. Computational Mechanics , 1–17.

938 Liu, S., Hao, S., Shen, Y., 2024. Asymptotic homogenization for phase field fracture of heterogeneous materials and application to toughening.

939 Composite Structures 339, 118134.

Liang Xue et al.: Preprint submitted to Elsevier Page 65 of 68

 An extended gradient damage model for anisotropic fracture

940 Lu, G., Chen, J., 2020. A new nonlocal macro-meso-scale consistent damage model for crack modeling of quasi-brittle materials. Computer Methods

941 in Applied Mechanics and Engineering 362, 112802.

942 Mandal, T.K., Nguyen, V.P., Heidarpour, A., 2019. Phase field and gradient enhanced damage models for quasi-brittle failure: A numerical

943 comparative study. Engineering Fracture Mechanics 207, 48–67.

944 Miehe, C., Hofacker, M., Welschinger, F., 2010. A phase field model for rate-independent crack propagation: Robust algorithmic implementation

945 based on operator splits. Computer Methods in Applied Mechanics and Engineering 199, 2765–2778.

946 Mosler, J., Meschke, G., 2004. Embedded crack vs. smeared crack models: a comparison of elementwise discontinuous crack path approaches with

947 emphasis on mesh bias. Computer methods in applied mechanics and engineering 193, 3351–3375.

948 Mozaffari, N., Voyiadjis, G.Z., 2015. Phase field based nonlocal anisotropic damage mechanics model. Physica D: Nonlinear Phenomena 308,

949 11–25.

950 Mozaffari, N., Voyiadjis, G.Z., 2016. Coupled gradient damage–viscoplasticty model for ductile materials: Phase field approach. International

951 Journal of Plasticity 83, 55–73.

952 Murray, R.M., Li, Z., Sastry, S.S., 2017. A mathematical introduction to robotic manipulation. CRC press.

953 Nguyen-Thanh, N., Rabczuk, T., 2024. Phase-field modeling of anisotropic crack propagation based on higher-order nonlocal operator theory.

954 International Journal of Solids and Structures 289, 112632.

955 Nooru-Mohamed, M., Schlangen, E., van Mier, J.G., 1993. Experimental and numerical study on the behavior of concrete subjected to biaxial

956 tension and shear. Advanced cement based materials 1, 22–37.

957 Nooru-Mohamed, M.B., 1993. Mixed-mode fracture of concrete: An experimental approach. praca doktorska. Delft University of Technology,

958 Delft .

959 Park, K., Paulino, G.H., Roesler, J.R., 2009. A unified potential-based cohesive model of mixed-mode fracture. Journal of the Mechanics and

960 Physics of Solids 57, 891–908.

961 Peerlings, R.H., de Borst, R., Brekelmans, W.M., de Vree, J., 1996. Gradient enhanced damage for quasi-brittle materials. International Journal for

962 numerical methods in engineering 39, 3391–3403.

963 Pham, K., Marigo, J.J., 2013. From the onset of damage to rupture: construction of responses with damage localization for a general class of gradient

964 damage models. Continuum Mechanics and Thermodynamics 25, 147–171.

965 Pham, K., Marigo, J.J., Maurini, C., 2011. The issues of the uniqueness and the stability of the homogeneous response in uniaxial tests with gradient

966 damage models. Journal of the Mechanics and Physics of Solids 59, 1163–1190.

967 Poh, L., Swaddiwudhipong, S., 2009. Gradient-enhanced softening material models. International Journal of Plasticity 25, 2094–2121.

968 Poh, L.H., Sun, G., 2017. Localizing gradient damage model with decreasing interactions. International Journal for Numerical Methods in

969 Engineering 110, 503–522.

970 Polyanin, P., Manzhirov, A.V., 2008. Handbook of integral equations. Chapman and Hall/CRC.

971 Ren, Y., Chen, J., Lu, G., 2023. A structured deformation driven nonlocal macro-meso-scale consistent damage model for the compression/shear

972 dominate failure simulation of quasi-brittle materials. Computer Methods in Applied Mechanics and Engineering 410, 115945.

973 Ren, Y., Chen, J., Lu, G., 2024. Mesoscopic simulation of uniaxial compression fracture of concrete via the nonlocal macro-meso-scale consistent

974 damage model. Engineering Fracture Mechanics , 110148.

975 Rots, J.G., Nauta, P., Kuster, G., Blaauwendraad, J., 1985. Smeared crack approach and fracture localization in concrete. HERON, 30 (1), 1985 .

976 Scherer, J.M., Brach, S., Bleyer, J., 2022. An assessment of anisotropic phase-field models of brittle fracture. Computer Methods in Applied

977 Mechanics and Engineering 395, 115036.

Liang Xue et al.: Preprint submitted to Elsevier Page 66 of 68

 An extended gradient damage model for anisotropic fracture

978 Sellier, A., Bary, B., 2002. Coupled damage tensors and weakest link theory for the description of crack induced anisotropy in concrete. Engineering

979 Fracture Mechanics 69, 1925–1939.

980 Shao, Y., Zhao, H.P., Feng, X.Q., Gao, H., 2012. Discontinuous crack-bridging model for fracture toughness analysis of nacre. Journal of the

981 Mechanics and Physics of Solids 60, 1400–1419.

982 Tandogan, I., Yalcinkaya, T., 2022. Development and implementation of a micromechanically motivated cohesive zone model for ductile fracture.

983 International Journal of Plasticity 158, 103427.

984 Tripathi, A., Mantell, S., Le, J.L., 2017. Modeling of cohesive fracture interacting with a stationary capillary fluid. Engineering Fracture Mechanics

985 182, 19–31.

986 Voyiadjis, G.Z., Pekmezi, G., Deliktas, B., 2010. Nonlocal gradient-dependent modeling of plasticity with anisotropic hardening. International

987 Journal of Plasticity 26, 1335–1356.

988 Wang, L., Su, H., Zhou, K., 2024. A phase-field model for mixed-mode cohesive fracture in fiber-reinforced composites. Computer Methods in

989 Applied Mechanics and Engineering 421, 116753.

990 Wang, T., Han, H., Wang, Y., Ye, X., Huang, G., Liu, Z., Zhuang, Z., 2022. Simulation of crack patterns in quasi-brittle materials under thermal

991 shock using phase field and cohesive zone models. Engineering Fracture Mechanics 276, 108889.

992 Wang, T., Ye, X., Liu, Z., Liu, X., Chu, D., Zhuang, Z., 2020. A phase-field model of thermo-elastic coupled brittle fracture with explicit time

993 integration. Computational Mechanics 65, 1305–1321.

994 Wong, L., Einstein, H., 2009. Crack coalescence in molded gypsum and carrara marble: part 1. macroscopic observations and interpretation. Rock

995 Mechanics and Rock Engineering 42, 475–511.

996 Wu, J.Y., 2017. A unified phase-field theory for the mechanics of damage and quasi-brittle failure. Journal of the Mechanics and Physics of Solids

997 103, 72–99.

998 Wu, J.Y., Nguyen, V.P., 2018. A length scale insensitive phase-field damage model for brittle fracture. Journal of the Mechanics and Physics of

999 Solids 119, 20–42.

1000 Wu, J.Y., Nguyen, V.P., Nguyen, C.T., Sutula, D., Sinaie, S., Bordas, S.P., 2020a. Phase-field modeling of fracture. Advances in applied mechanics

1001 53, 1–183.

1002 Wu, J.Y., Nguyen, V.P., Zhou, H., Huang, Y., 2020b. A variationally consistent phase-field anisotropic damage model for fracture. Computer

1003 Methods in Applied Mechanics and Engineering 358, 112629.

1004 Wu, J.Y., Yao, J.R., Le, J.L., 2023. Phase-field modeling of stochastic fracture in heterogeneous quasi-brittle solids. Computer Methods in Applied

1005 Mechanics and Engineering 416, 116332.

1006 Wu, Z.M., Rena, C.Y., Sun, C.Y., Wang, Y.J., Zhang, X.X., Fei, X.D., 2020c. A new test method for the complete load-displacement curve of

1007 concrete under mixed mode i-ii fracture. Theoretical and Applied Fracture Mechanics 108, 102629.

1008 Xue, L., Ren, X., 2024a. Achieving irreversibility in damage evolution: Extended gradient damage model with decoupled damage profile and

1009 cohesive law. Journal of the Mechanics and Physics of Solids 183, 105524.

1010 Xue, L., Ren, X., 2024b. A tensorial energy-release-rate based anisotropic damage-plasticity model for concrete. Mechanics of Materials , 105025.

1011 Xue, L., Ren, X., Ballarini, R., 2023. Damage-plasticity modeling of shear failure in reinforced concrete structures. Engineering Fracture Mechanics

1012 290, 109536.

1013 Xue, L., Ren, X., Freddi, F., 2024. Analytical solution of a gradient-enhanced damage model for quasi-brittle failure. Applied Mathematical

1014 Modelling .

Liang Xue et al.: Preprint submitted to Elsevier Page 67 of 68

 An extended gradient damage model for anisotropic fracture

1015 Yuan, S., Tong, T., Liu, Z., Yang, P., 2022. Explicit double-phase-field formulation and implementation for bending behavior of uhpc-nc composite

1016 beams. Journal of Building Engineering 57, 104802.

1017 Zambrano, J., Toro, S., Sánchez, P., Duda, F., Méndez, C., Huespe, A., 2022. Interaction analysis between a propagating crack and an interface:

1018 Phase field and cohesive surface models. International Journal of Plasticity 156, 103341.

1019 Zeng, Q., Wang, T., Zhu, S., Chen, H.s., Fang, D., 2022. A rate-dependent phase-field model for dynamic shear band formation in strength-like and

1020 toughness-like modes. Journal of the Mechanics and Physics of Solids 164, 104914.

1021 Zhou, H., Tian, X., Wu, J., 2024. Cracking and thermal resistance in concrete: Coupled thermo-mechanics and phase-field modeling. Theoretical

1022 and Applied Fracture Mechanics 130, 104285.

1023 Zhu, Q.Z., Yuan, S.S., Shao, J.F., 2023. Microporomechanics of quasi-brittle rocks: Theoretical formulations and analytical simulations.

1024 International Journal of Plasticity 171, 103789.

1025 Zienkiewicz, O.C., Taylor, R.L., Zhu, J.Z., 2005. The finite element method: its basis and fundamentals. Elsevier.

Liang Xue et al.: Preprint submitted to Elsevier Page 68 of 68