betaFeedback

WorksPhysics and Astronomy

A non incremental variational principle for brittle fracture

Abstract

The aim of the paper is to propose a paradigm shift for the variational approach of brittle fracture. Both dynamics and the limit case of statics are treated in a same framework. By contrast with the usual incremental approach, we use a space-time principle covering the whole loading and crack evolution. The emphasis is given on the modelling of the crack extension by the internal variable formalism and a dissipation potential as in plasticity, rather than Griffith's original approach based on the surface area. The new formulation appears to be more fruitful for generalization than the standard theory.

Introduction

Many dynamical systems are subjected to energy loss resulting from dissipation, for instance collisions, surface friction, viscosity, plasticity, fracture and damage. On the other hand, deformations of solids and motions of fluids are modeled through constitutive laws. Due to collisions, brittle fracture and threshold effects, most dissipative laws are non smooth and multivalued. Moreover, experimental testing suggests that convexity is a keystone property of these phenomenological laws. Continuum mechanics with internal variables provides a convenient foundation to develop constitutive models that describe the inelastic behavior of various materials, despite the large differences in their physical nature and the relevant scale. The phenomenological law link the vector of internal variable rates α˙\dot{\alpha} to the dual force vector A\mathbf{A}. The theory of Generalized Standard Materials [41] is based on an hypothesis of normal dissipation.

α˙∈∂φ(A)\dot{\alpha} \in\partial\varphi(\mathbf{A})

where occurs the subdifferential of a convex and lower semicontinuous function φ\varphi, not everywhere differentiable. Equivalently, the law reads

A=argmax⁡α˙(⟨α˙,A⟩−φ(A))A = \operatorname*{argmax}_{\dot{\alpha}} \left(\langle\dot{\alpha}, A\rangle- \varphi(A)\right)

or, introducing the Fenchel polar φ∗\varphi^* [33]

(α˙,A)=argmin⁡A(φ(A)+φ∗(α˙)−⟨α˙,A⟩)(\dot{\alpha},A) = \operatorname*{argmin}_{A} \left(\varphi(A) + \varphi^*(\dot{\alpha}) - \langle\dot{\alpha},A\rangle\right)

that does not favour one of the two dual variables. In the present work, the latest formulation is chosen as starting point, leading to variational methods and unconstrained or constrained optimization problems. Among them, Brezis-Ekeland-Nayroles principle [13] [63], or in short BEN principle, is based on the time integration of the sum of dissipation potential ϕ\phi and its Fenchel polar (analogous of Legendre polar for convex functions). Although used a few in the literature, this principle is noteworthy in the sense that it allows covering the whole evolution of the dissipative system.

In a previous paper [15], Buliga proposed the formalism of Hamiltonian inclusions, able to model dynamical systems with 1-homogeneous dissipation potential (for laws such as brittle damage using Ambrosio-Tortorelli functional [3]). This formalism is a dynamical version of the quasi-static theory of rate-independent systems of Mielke ([55], [56], [57]).

Latter on, Buliga and the author merged in a symplectic framework this formalism with BEN principle to extend it to dynamics [16]. The key-idea is to decompose additively the time rate z˙\dot{z} into reversible part z˙R\dot{z}_{R} (the symplectic gradient) and dissipative or irreversible one z˙I\dot{z}_{I}, next to define the symplectic subdifferential ∂ωϕ(z)\partial^{\omega}\phi(z) of the dissipation potential. To release the restrictive hypothesis of 1-homogeneity (in particular to address viscoplasticity), we introduce in this work the symplectic Fenchel polar ϕ∗ω\phi^{*\omega}, that allows to build theoretical methods to model and analyse dynamical dissipative systems in a consistent geometrical framework with the numerical approaches not very far in the background. Numerical simulation with the BEN principle were performed for elastoplastic structures in statics [17, 19] and in dynamics [18].

Closer to this approach, we can cite the contributions of Aubin [5], Aubin, Cellina and Nohel [4], Rockafellar [66], Stolz [70], which considered various extensions of Hamiltonian and Lagrangian mechanics. In the article [7] by Bloch, Krishnaprasad, Marsden and Ratiu, Hamiltonian systems are explored with an added Rayleigh dissipation. A theory of quasistatic rate-independent systems is proposed by Mielke and Theil [55], Mielke [56], and developed towards applications in many papers, among them Mielke and Roubíček [57], see also Visintin [75]. In [39, 64], Grmela and Öttinger proposed the framework GENERIC (General Equation for Non-Equilibrium Reversible-Irreversible Coupling), a systematic method to derive thermodynamically consistent evolution equations. A variational formulation of GENERIC is proposed in [53]. In [58], Mielke proposed a GENERIC formulation for Generalized Standard Materials quite similar to the one in the present paper. It would be worth to make the present formalism thermodynamic. In [69], Stefanelli used Brezis-Ekeland-Nayroles variational principle to represent the quasistatic evolution of an elastoplastic material with hardening in order to prove the convergence of time and space-time discretizations as well as to provide some possible a posteriori error control. Finally, Ghoussoub and MacCann characterized the path of steepest descent of a non-convex potential as the global minimum of Brezis-Ekeland-Nayroles functional [37].

Moreover, another advantage of Brezis-Ekeland-Nayroles principle is the easiness to be generalized. Indeed, it is worth to know that many realistic dissipative laws, called non-associated, cannot be cast in the mould of the standard ones deriving of a dissipation potential. To skirt this pitfall, the author proposed in [24] a new theory based on a function called bipotential. It represents physically the dissipation and generalizes the sum of the dissipation potential and its Fenchel polar, reason for which extension of Brezis-Ekeland-Nayroles principle is natural. The applications of the bipotential approach to solid Mechanics are various: Coulomb’s friction law [25], non-associated Drücker-Prager [26] and Cam-Clay models [78] in Soil Mechanics, cyclic Plasticity ([25],[9], [10], [54], [11]) and Viscoplasticity [44] of metals with non linear kinematical hardening rule, Lemaitre’s damage law [8], the coaxial laws ([27],[73]). Such kind of materials are called implicit standard materials. A synthetic review of these laws can be found in the two later references. It is also worth to notice that monotone laws but which does not admit a convex potential can be represented by Fitzpatrick’s function [35] which is a bipotential.

There is an abundant literature on the crack propagation criteria and our intention is not to draw here an exhaustive picture. For a deep survey, the reader is referred to [61]. Let us cite only the most popular ones: the maximum tensile hoop stress criterion formulated in 1963 by Cherepanov [21], Erdogan and Sih [32], the criterion of minimum strain energy density by Sih [68] in 1973, the principle of local symmetry formulated in 1974 by Goldstein and Salganik [38], the maximum energy release rate criterion considered in 1974 by Hussain et al. [45] and in 1978 by Wu [76], and the generalized maximum energy release rate criterion [42, 40, 47, 48]. These criteria give predictions more or less closed to the experimental results and the possible link with the variational approach is puzzling, at least apparently. This is one of the challenge we have to address.

Our aim here is to study the propagation of already initiated cracks. Then we exclude of this paper the problem of the crack nucleation or initiation that would require extra information about stresses in the damaged region. Two models are used in the literature, the Coupled Criterion proposed par Leguillon [50] and the Cohesive Zone Model, originated in the pioneer works by Barenblatt [6] and Dugdale [30], next developed by Tvergaard and al [72], Xu [77] and, for the variational aspects, by Bourdin et al [12].

The paper is organized as follows.

  • In Section 2, the initial crack and the loading history being known, the problem is to find the crack evolution. Instead of the usual modelling of the crack extension as a time-parameterized family of surfaces, we propose to introduce a flow of a vector field on the final crack manifold.

  • Section 3 is devoted to the definition of the variable of the problem in the symplectic framework: the displacement and flow fields and the corresponding dynamic momenta.

  • In Section 4, the symplectic BEN principle is applied to the fracture mechanics, that leads to the definition of a driving force, dual of the crack flow rate.

  • In Section 5, we propose a constitutive law for brittle fracture in the form of a crack stability criterion and the normality law for the crack extension rule.

  • We prove in Section 6 for structure with uniform toughness the equivalence with the classical variational approach in the form of a Munford-Shah functional.

  • In Section 7, we deduce from the space-time principle an incremental version suitable for the usual step-by-step numerical approaches.

  • In Section 8, we show how a suitable interpretation of the experimental data prove the relevancy of the normality law and the link with the principle of local symmetry.

  • Section 9 is devoted to the calculation of the crack driving force using the calculus of variation on the jet space of order one in the general case of dynamics.

Modelling of the crack extension

Data

Let Ω⊂Rn\Omega\subset\mathbb{R}^{n} be a bounded, open set, with piecewise smooth boundary ∂Ω\partial\Omega occupied by the uncracked body (Figure 2.1). Our aim is to study a crack extension during a time interval [0,T][0,T]

Figure 2.1. Crack Flow

The data of the problem is:

  • the initial crack Γ0\Gamma_{0}

  • the imposed displacements uˉ\bar{u} on the part ∂Ω0\partial\Omega_{0} of the boundary (the supports)

  • the surface forces fˉ\bar{f} on the remaining part ∂Ω1\partial\Omega_{1}

  • the volume forces ff

All the external actions, uˉ\bar{u}, fˉ\bar{f} and ff, are prescribed during the time interval. No stress is transmitted through the crack.

Crack

A crack bifurcates when its shape suffers a change of topology. The most common example is a crack which develops in time new branches. During this phenomenon the number of crack fronts increases. In this paper we do not study the bifurcation of an existing crack.

We denote SS the set of admissible surfaces, i.e. closed countably 2 rectifiable subsets of Ω\Omega without change of topology. The crack evolution is described by a time-parameterized family of surfaces (Figure 2.1)

Γ:[0,T]→S:t↦Γt=Γ(t)\Gamma: [0,T] \to S : t \mapsto\Gamma_t = \Gamma(t)

such that Γ(0)=Γ0\Gamma(0) = \Gamma_0 and the map Γ\Gamma is monotone increasing, SS being equipped with the inclusion order. The crack front ctc_t at time tt is parameterized by the arc length ss.

The cornerstone of the formulation is to model the crack extension by a flow on the cracked surface ΓT\Γ0\Gamma_T \backslash\Gamma_0 during the time interval

[0,T]×c0→ΓT\Γ0:(t,x0)↦x=ψ(t,x0)(1)[0,T] \times c_0 \to\Gamma_T \backslash\Gamma_0 : (t,x_0) \mapsto x = \psi(t,x_0) \tag*{(1)}

such that

ψ({t}×c0)=ct,ψ([0,t]×c0)=Γt\Γ0\psi(\{t\} \times c_0) = c_t,\qquad\psi([0,t] \times c_0) = \Gamma_t \backslash\Gamma_0

It is worth to remark that:

  • From a calculus viewpoint, the field ψ\psi represents both the crack and its evolution. The main advantage is that it is easier working with fields living in a functional space than with surfaces.

  • In particular, our approach is well suited for numerical applications. s↦x0=f(s)s \mapsto x_0 = f(s) being the parameterization of the initial crack c0c_0 by the arc length $s,`

the cracked surface can be parameterized according to (t,s)↦x=ψ~(t,s)=ψ(t,f(s))(t,s) \mapsto x = \tilde{\psi}(t,s) = \psi(t,f(s)). The node xjk=ψ~(tj,sk)x_{jk} = \tilde{\psi}(t_j,s_k) being the position at time tjt_j of the node x0k=f(sk)∈c0x_{0k} = f(s_k) \in c_0 and (s,t)↦Njk(t,s)(s,t) \mapsto N_{jk}(t,s) the corresponding shape function, the cracked surface ΓT\Gamma_T is parameterized by

x=ψ~(t,s)=∑j∑kNjk(t,s)xjk(2)x = \tilde{\psi}(t,s) = \sum_j \sum_k N_{jk}(t,s)x_{jk} \tag*{(2)}

In the forthcoming variational formulation, the nodal values xjk=ψ~(tj,sk)x_{jk} = \tilde{\psi}(t_j,s_k) will be unknowns of the discretized problem.

  • A streamline t↦x=ψ(t,x0)t \mapsto x = \psi(t,x_0) does not correspond to the motion of a material particle initially at the position x0∈c0x_0 \in c_0. The flow is not Lagrangian but Eulerian in the sense it represents just the evolution of the crack front.

If the cracked surface is smooth enough, it is generated by a vector field ψ˙\dot{\psi}, solution on ΓT∖Γ0\Gamma_T \setminus\Gamma_0 of the ODE (Figure 2.1)

ddt(ψ(t,x0))=ψ˙(ψ(t,x0))withψ(0,x0)=x0\frac{\mathrm{d}}{\mathrm{d}t}\left(\psi(t,x_0)\right) = \dot{\psi}\left(\psi(t,x_0)\right) \qquad\text{with} \qquad\psi(0,x_0) = x_0

As in the FEM method, it is easy to compute the vector field ψ˙\dot{\psi} as the partial derivative of (2) with respect to tt.

Variables of the problem

To do not enter into too complex formulations, we consider the case of an elastic body in small strains, although the generalization to the elastoplasticity and the finite strains is rather straightforward. The cracked solid at time tt is denoted Ωt=Ω∖Γt\Omega_t = \Omega\setminus\Gamma_t. Within the body, the displacement field at time tt is

Ωt→R3:x↦u(t,x)\Omega_t \to\mathbb{R}^3 : x \mapsto u(t,x)

According to the symplectic BEN principle [16], the variables of the problem of the elastodynamics are:

  • the field couple ξ=(u,ψ)\xi= (u,\psi) of the displacement and the flow

  • the momenta η=(p,π)\eta= (p,\pi) where pp is the classical linear momentum and π\pi is the momentum associated to the flow

The system evolution is given by a curve

t↦z(t)=(ξ(t),η(t))t \mapsto z(t) = (\xi(t),\eta(t))

The symplectic BEN variational principle

The duality between the space XX of degrees of freedom ξ\xi and the space YY of momenta η\eta has the form

⟨ξ˙,η˙⟩=⟨u˙,p˙⟩+⟨ψ˙,π˙⟩=∫Ωtu˙⋅p˙ d3x+∫Ctψ˙⋅π˙ ds\langle\dot{\xi},\dot{\eta}\rangle=\langle\dot{u},\dot{p}\rangle+\langle\dot{\psi},\dot{\pi}\rangle=\int_{\Omega_t}\dot{u}\cdot\dot{p}\,\mathrm{d}^{3}x+\int_{C_t}\dot{\psi}\cdot\dot{\pi}\,\mathrm{d}s

The symplectic form is the 2-form field on X×YX\times Y defined by

ω(z˙,z˙′)=ω((ξ˙,η˙),(ξ˙′,η˙′))=⟨ξ˙,η˙′⟩−⟨ξ˙′,η˙⟩\omega(\dot{z},\dot{z}')=\omega((\dot{\xi},\dot{\eta}),(\dot{\xi}',\dot{\eta}'))=\langle\dot{\xi},\dot{\eta}'\rangle-\langle\dot{\xi}',\dot{\eta}\rangle

The total Hamiltonian of the structure is taken of the integral form

Ht=H(t,z)=∫Ωt{12ρ∥p∥2+w(∇su)−f(t)⋅u} d3x−∫∂Ω1f‾(t)⋅u d(∂Ω1)(3)H_t=H(t,z)=\int_{\Omega_t}\left\{\frac{1}{2\rho}\|\boldsymbol{p}\|^2+w(\nabla_s\boldsymbol{u})-\boldsymbol{f}(t)\cdot\boldsymbol{u}\right\}\,\mathrm{d}^{3}x-\int_{\partial\Omega_1}\overline{\boldsymbol{f}}(t)\cdot\boldsymbol{u}\,\mathrm{d}(\partial\Omega_1) \tag*{(3)}

The first term is the kinetic energy. The second one is the elastic strain energy ww depending on the strain tensor ε=∇su\boldsymbol{\varepsilon}=\nabla_s\boldsymbol{u} and the two latter terms are the works of the external forces. As usual, the stress tensor is

σ=∂w∂ε\boldsymbol{\sigma}=\frac{\partial w}{\partial\boldsymbol{\varepsilon}}

A crack being a material discontinuity, we must distinguish two material surfaces Γt+\Gamma_t^+ and Γt−\Gamma_t^- that occupy the same position as Γt\Gamma_t but are the two sides of the crack and have opposite unit normal vectors, exterior to Ωt\Omega_t: n+=−n−\boldsymbol{n}^+=-\boldsymbol{n}^-.

A curve is said admissible if it satisfies

  • the boundary conditions:

∀t∈[0,T],u=u‾ on ∂Ω0,σ⋅n=f‾ on ∂Ω1,σ⋅n±=0 on Γt±\forall t\in[0,T],\quad\boldsymbol{u}=\overline{\boldsymbol{u}}\ \text{on }\partial\Omega_0,\quad\boldsymbol{\sigma}\cdot\boldsymbol{n}=\overline{\boldsymbol{f}}\ \text{on }\partial\Omega_1,\quad\boldsymbol{\sigma}\cdot\boldsymbol{n}^{\pm}=0\ \text{on }\Gamma_t^{\pm}
  • the initial conditions: z(0)=z0z(0)=z_0

The SBEN formalism of dissipative media is based on the additive decomposition of the velocity into reversible and irreversible parts

z˙=z˙R+z˙I\dot{z}=\dot{z}_R+\dot{z}_I

As the variables of the problem are fields, we use functional derivatives

⟨δξ,DξH⟩+⟨DηH,δη⟩=lim⁡ϵ→01ϵ(H(t,z+ϵ δz)−H(t,z))\langle\delta\xi,D_{\xi}H\rangle+\langle D_{\eta}H,\delta\eta\rangle=\lim_{\epsilon\to0}\frac{1}{\epsilon}\left(H(t,z+\epsilon\,\delta z)-H(t,z)\right)

The reversible part of the velocity is given by the symplectic gradient of the Hamiltonian (or Hamiltonian vector field)

z˙R=∇ωH=(DηH,−DξH)=((DpH,DπH),(−DuH,−DψH))\dot{z}_R=\nabla^{\omega}H=(D_{\eta}H,-D_{\xi}H)=((D_pH,D_{\pi}H),(-D_uH,-D_{\psi}H))

For the Hamiltonian (3), we obtain

DpH=pρ,DπH=0D_p H = \frac{p}{\rho}, \qquad D_\pi H = 0
−DuH=∇⋅σ+f,−DψH=G-D_u H = \nabla\cdot\boldsymbol{\sigma} + \boldsymbol{f}, \qquad-D_\psi H = \boldsymbol{G}

The last equation is just a definition of the driving force G\boldsymbol{G} of which the explicit expression will be discussed afterwards.

In [16], we formulated a general variational principle for the dynamical dissipative systems which claims that

Symplectic BEN principle 1. The natural evolution of the system minimizes the functional

Π(z)=∫0T{ϕ(z˙)+ϕ∗ω(z˙I)−ω(z˙I,z˙)} dt(4)\Pi(\boldsymbol{z}) = \int_0^T \left\{\phi(\dot{\boldsymbol{z}}) + \phi^{*\omega}(\dot{\boldsymbol{z}}_I) - \omega(\dot{\boldsymbol{z}}_I,\dot{\boldsymbol{z}})\right\}\,\mathrm{d}t \tag*{(4)}

among the admissible curves t↦z(t)t \mapsto\boldsymbol{z}(t) and the minimum is zero.

In this principle occurs the symplectic form ω\omega, the Hamiltonian through the irreversible part of the velocity z˙I=z˙−∇ωH\dot{\boldsymbol{z}}_I=\dot{\boldsymbol{z}}-\nabla^\omega H, a convex dissipation potential ϕ\phi and its symplectic polar ϕ∗ω\phi^{*\omega}. We calculate now the detailed expression of this ingredients for the brittle fracture. We start with a simplifying hypothesis similar to the one introduced in [16] to recover the classical elastoplasticity. We claim that in the dissipation potential, the variables other than π˙\dot{\pi} are ignorable

ϕ(z˙)=φ(π˙)\phi(\dot{\boldsymbol{z}})=\varphi(\dot{\pi})

The symplectic polar is defined in [16] as

ϕ∗ω(z˙′)=sup⁡[ω(z˙′,z˙)−ϕ(z˙):z˙]\phi^{*\omega}(\dot{\boldsymbol{z}}')=\sup\left[\omega(\dot{\boldsymbol{z}}',\dot{\boldsymbol{z}})-\phi(\dot{\boldsymbol{z}}):\dot{\boldsymbol{z}}\right]

Under the previous hypothesis, one has

ϕ∗ω(z˙′)=sup⁡[⟨u˙′,p˙⟩+⟨ψ˙′,π˙⟩−⟨u˙,p˙′⟩−⟨ψ˙,π˙′⟩−φ(π˙):z˙]\phi^{*\omega}(\dot{\boldsymbol{z}}')=\sup\left[\langle\dot{\boldsymbol{u}}',\dot{\boldsymbol{p}}\rangle+\langle\dot{\psi}',\dot{\pi}\rangle-\langle\dot{\boldsymbol{u}},\dot{\boldsymbol{p}}'\rangle-\langle\dot{\psi},\dot{\pi}'\rangle-\varphi(\dot{\pi}):\dot{\boldsymbol{z}}\right]

that gives

ϕ∗ω(z˙′)=χ{0}(u˙′)+φ∗(ψ˙′)+χ{0}(p˙′)+χ{0}(π˙′)\phi^{*\omega}(\dot{\boldsymbol{z}}')=\chi_{\{0\}}(\dot{\boldsymbol{u}}')+\varphi^*(\dot{\psi}')+\chi_{\{0\}}(\dot{\boldsymbol{p}}')+\chi_{\{0\}}(\dot{\pi}')

where χK\chi_K is the indicatory function of the set KK, equal to 0 in KK and +∞+\infty otherwise, and φ∗\varphi^* is the Fenchel polar of φ\varphi. The second term in the functional (4) becomes:

ϕ∗ω(z˙I)=χ{0}(u˙−pρ)+φ∗(ψ˙)+χ{0}(p˙−∇⋅σ−f)+χ{0}(π˙−G)\phi^{*\omega}(\dot{\boldsymbol{z}}_I)=\chi_{\{0\}}\left(\dot{\boldsymbol{u}}-\frac{\boldsymbol{p}}{\rho}\right)+\varphi^*(\dot{\psi})+\chi_{\{0\}}(\dot{\boldsymbol{p}}-\nabla\cdot\boldsymbol{\sigma}-\boldsymbol{f})+\chi_{\{0\}}(\dot{\pi}-\boldsymbol{G})

As the minimum of the functional is the finite value zero, it will be reached when the arguments of the indicatory functions vanish. Then we may take the zero value of the indicatory functions in the functional while we introduce extra corresponding constraints

p=ρu˙,∇⋅σ+f=p˙,π˙=G(5)\boldsymbol{p}=\rho\dot{\boldsymbol{u}}, \qquad\nabla\cdot\boldsymbol{\sigma}+\boldsymbol{f}=\dot{\boldsymbol{p}}, \qquad\dot{\pi}=\boldsymbol{G} \tag*{(5)}

Next we can transform the last term in the functional, taking into account the velocity decomposition, the linearity and the antisymmetry of the symplectic form

−ω(z˙I,z˙)=−ω(z˙−z˙R,z˙)=−ω(z˙,z˙)+ω(z˙R,z˙)=ω(z˙R,z˙)(6)-\omega(\dot{z}_{I},\dot{z})=-\omega(\dot{z}-\dot{z}_{R},\dot{z})=-\omega(\dot{z},\dot{z})+\omega(\dot{z}_{R},\dot{z})=\omega(\dot{z}_{R},\dot{z}) \tag*{(6)}

or in detail

−ω(z˙I,z˙)=⟨pρ,p˙⟩−⟨u˙,∇⋅σ+f⟩−⟨ψ˙,G⟩-\omega(\dot{z}_{I},\dot{z})=\left\langle\frac{p}{\rho},\dot{p}\right\rangle-\langle\dot{u},\nabla\cdot\sigma+f\rangle-\langle\dot{\psi},G\rangle

Owing to the constraints (5), it holds

−ω(z˙I,z˙)=⟨u˙,p˙−∇⋅σ−f⟩−⟨ψ˙,G⟩=−⟨ψ˙,G⟩-\omega(\dot{z}_{I},\dot{z})=\langle\dot{u},\dot{p}-\nabla\cdot\sigma-f\rangle-\langle\dot{\psi},G\rangle=-\langle\dot{\psi},G\rangle

Moreover, thanks to the former and latter constraints in (5), the momenta can be eliminated from the functional and the intermediate constraint becomes

∇⋅σ+f=ρu¨\nabla\cdot\sigma+f=\rho\ddot{u}

while the initial condition on the linear momentum can be transformed into an initial condition on the velocity because p(0)=ρu˙(0)p(0)=\rho\dot{u}(0). Taking into account these transformations, the minimum can be searched only on the space of the degrees of freedom uu and ψ\psi. Then we obtain a second version of the variational principle

Symplectic BEN principle 2. The natural evolution of the system minimizes

Π(ξ)=∫0T{φ(G)+φ∗(ψ˙)−⟨ψ˙,G⟩} dt(7)\Pi(\xi)=\int_{0}^{T}\left\{\varphi(G)+\varphi^{*}(\dot{\psi})-\langle\dot{\psi},G\rangle\right\}\,\mathrm{d}t \tag*{(7)}

among the admissible curves t↦ξ(t)t\mapsto\xi(t) such that ∇⋅σ+f=ρu¨\nabla\cdot\sigma+f=\rho\ddot{u} and the minimum is zero.

It could seem puzzling that the displacement does not appear explicitly in the expression of the functional but in fact uu and ψ\psi are coupled in the minimization problem:

  • ψ\psi is controlled by GG which depends on uu as derivative of the Hamiltonian.

  • uu must satisfy the constraint and the admissibility conditions that are defined on Ωt=Ω\Γt\Omega_t=\Omega\backslash\Gamma_t controlled by ψ\psi.

Constitutive laws for brittle fracture

As discussed in the Introduction, most of the crack stability criteria in the literature, in particular the most popular, are of local nature and not variational. Nevertheless, Strifors proposed in [71] to predict the onset of brittle fracture and the extension direction thanks to the crack extension force that can be identified to the driving force GG. Now we show how to cast it into the mold of our variational approach. Based on the works by Strifors [71] and Hellen et al. [43], we propose a simple constitutive law. At least for plane problems, it seems reasonable to think that the influence on the fracture. extension of the projection of GG onto the tangent to the crack front can be neglected. Then we introduce the deviatoric force

G⊥=G−(G⋅τ)τG_{\perp}=G-(G\cdot\tau)\tau

where τ\tau is the unit tangent vector to the crack front in the direction of increasing arc length (Figure 2.1). As in the FEM method, it is easy to compute τ\tau from the partial derivative of (2) with respect to the arc length ss of c0c_0. If we assume that the toughness of the material is isotropic, the critical energy release rate GcG_c being a parameter measuring it, we introduce the convex crack stability domain

K={G such that ∥G⊥∥≤Gc}K=\{G\ \text{such that}\ \lVert G_{\perp}\rVert\leq G_c\}

and the crack extension rule

ψ˙∈∂φ(G)=∂χK(G)\dot{\psi}\in\partial\varphi(G)=\partial\chi_K(G)

that is the normality law

ifG∈K˚ thenψ˙=0! crack stabilityelseG∈∂K and ∃λ≥0,ψ˙=λG⊥! crack extension\begin{array}{ll} \text{if} & G\in\mathring{K}\ \text{then} \\ & \dot{\psi}=0 \qquad\qquad\qquad\qquad\qquad\qquad\quad\text{! crack stability} \\ \text{else} & G\in\partial K\ \text{and}\ \exists\lambda\geq0,\qquad\dot{\psi}=\lambda G_{\perp}\quad\text{! crack extension} \end{array}

and ψ˙\dot{\psi} is perpendicular to τ\tau when the crack extends. The Fenchel polar is the support function

φ∗(ψ˙)=∫ctGc∥ψ˙∥ ds\varphi^*(\dot{\psi})=\int_{c_t}G_c\lVert\dot{\psi}\rVert\,\mathrm{d}s

where ss is the arc length of ctc_t. With this choice of constitutive law, we particularize the variational principle in the form

Symplectic BEN principle 3. The natural evolution of the system minimizes

Π(ξ)=∫0T{∫ctGc∥ψ˙∥ ds−⟨ψ˙,G⟩} dt(8)\Pi(\xi)=\int_0^T\left\{\int_{c_t}G_c\lVert\dot{\psi}\rVert\,\mathrm{d}s-\langle\dot{\psi},G\rangle\right\}\,\mathrm{d}t \tag*{(8)}

among the admissible curves t↦ξ(t)t\mapsto\xi(t) such that ∇⋅σ+f=ρu¨\nabla\cdot\sigma+f=\rho\ddot{u}, ∥G⊥∥≤Gc\lVert G_{\perp}\rVert\leq G_c and the minimum is zero.

Link with the classical variational approach to fracture

First we recall that the last term in (8) is in fact the last one of the integrand of (4). Owing to (6), we have

−ω(z˙I,z˙)=ω(z˙R,z˙)=⟨DηH,η˙⟩−⟨ξ˙,−DξH⟩=H˙−DtH-\omega(\dot{z}_I,\dot{z})=\omega(\dot{z}_R,\dot{z})=\langle D_\eta H,\dot{\eta}\rangle-\langle\dot{\xi},-D_\xi H\rangle=\dot{H}-D_tH

Besides the crack surface element comprised between the crack fronts ctc_t and ct+dtc_{t+dt}, and the streamlines of x0x_0 and x0+dx0x_0+dx_0 is an infinitesimal parallelogram with two sides supported by the vectors d1x=τ dsd_1x=\tau\,ds and d2x=ψ˙ dtd_2x=\dot{\psi}\,dt. Its area is

dΓ=∥τ×ψ˙∥ ds dt=∥τ∥ ∥ψ˙∥ ∣sin⁡(τ,ψ˙)∣ ds dt=∥ψ˙∥ ds dtd\Gamma=\lVert\tau\times\dot{\psi}\rVert\,ds\,dt=\lVert\tau\rVert\,\lVert\dot{\psi}\rVert\,\lvert\sin(\tau,\dot{\psi})\rvert\,ds\,dt=\lVert\dot{\psi}\rVert\,ds\,dt

because τ\tau is a unit vector and ψ˙\dot{\psi} is perpendicular to τ\tau.

If we assume that the toughness properties of the material are homogeneous, the critical energy release rate is uniform

Gc(x)=GcG_c(x)=G_c

The area of a surface will be denoted by ∣∙∣\lvert\bullet\rvert. The variational principle can be recast as

Symplectic BEN principle 4. The natural evolution of the system minimizes

Π(Γ,u)=Gc∣ΓT\Γ0∣−∫0TDtH dt+HT−H0(9)\Pi(\Gamma,u)=G_c\lvert\Gamma_T\backslash\Gamma_0\rvert-\int_0^T D_tH\,dt+H_T-H_0 \tag*{(9)}

among the admissible curves t↦(Γ(t),u(t))t\mapsto(\Gamma(t),u(t)) such that ∇⋅σ+f=ρu¨\nabla\cdot\sigma+f=\rho\ddot{u}, ∥G⊥∥≤Gc\lVert G_{\perp}\rVert\leq G_c and the minimum is zero.

Two cases must be considered in the applications:

  • If the loading is controlled by the forces, the structural evolution is in general dynamic and the functional to minimize is

Π(Γ,u)=Gc∣ΓT\Γ0∣−∫0TDtH dt+HT−H0\Pi(\Gamma,u)=G_c\lvert\Gamma_T\backslash\Gamma_0\rvert-\int_0^T D_tH\,dt+H_T-H_0
  • If the loading is slow and controlled by the displacements (for instance, fˉ\bar{f} is null, the support is divided into two disjoint parts, one of them is fixed to the foundation, the other one is loaded), the evolution is quasi-static and the functional to minimize is reduced to

Π(Γ,u)=Gc∣ΓT\Γ0∣+HT−H0\Pi(\Gamma,u)=G_c\lvert\Gamma_T\backslash\Gamma_0\rvert+H_T-H_0

Regardless of the constant H0H_0, we recover the Munford-Shah functional [62, 36, 14, 12]

Incremental method

Our goal now is to find an incremental or step-by-step formulation as a by-product of our general principle. As usual, the time interval [0,T][0,T] is divided into NN small subintervals [tn,tn+1][t_n,t_{n+1}] with t0=0t_0=0 and tN=Tt_N=T. The value of a variable aa at time tnt_n is denoted ana_n. The time step is Δt=tn+1−tn\Delta t=t_{n+1}-t_n and Δa=an+1−an\Delta a=a_{n+1}-a_n is the increment of aa.

ξn=(un,ψn)\xi_n=(u_n,\psi_n) being known from the initial conditions for n=0n=0 or from the previous increment otherwise and the time step being prescribed, the problem is to calculate ξn+1\xi_{n+1} or equivalently Δξ\Delta\xi. The idea is to apply the symplectic BEN principle 3 to the subinterval [tn,tn+1][t_n,t_{n+1}]. The time step being small, the functional can be approximated by considering the integrand is constant. As for the radial return algorithm in plasticity, we use the implicit scheme by evaluating the crack extension rate ψ˙\dot{\psi} and the driving force G\mathbf{G} at the end of the step. The crack extension increment is

Δψ=ψ˙n+1Δt=λGn+1Δt(10)\Delta\psi=\dot{\psi}_{n+1}\Delta t=\lambda G_{n+1}\Delta t \tag*{(10)}

Hence the incremental principle reads

Symplectic BEN principle 3*. The natural evolution of the system minimizes

ΔΠ(ξn+1)=∫ctn+1Gc∥Δψ∥ ds−⟨Δψ,Gn+1⟩(11)\Delta\Pi(\boldsymbol{\xi}_{n+1})=\int_{c_{t_{n+1}}}G_c\left\Vert\Delta\psi\right\Vert\,\mathrm{d}s-\langle\Delta\psi,\mathbf{G}_{n+1}\rangle \tag*{(11)}

among the ξn+1\boldsymbol{\xi}_{n+1} such that ∇⋅σn+1+fn+1=ρu¨n+1\nabla\cdot\boldsymbol{\sigma}_{n+1}+\mathbf{f}_{n+1}=\rho\ddot{\mathbf{u}}_{n+1}, ∥(Gn+1)⊥∥≤Gc\left\Vert(\mathbf{G}_{n+1})_{\perp}\right\Vert\leq G_c and the minimum is zero.

Likewise, the last version of the principle in the previous section gives rise to the incremental form

Symplectic BEN principle 4*. The natural evolution of the system minimizes

ΔΠ(Γn+1,un+1)=Gc∣ΔΓ∣−(DtH)n+1Δt+ΔH(12)\Delta\Pi(\Gamma_{n+1},\mathbf{u}_{n+1})=G_c\left|\Delta\Gamma\right|-(D_tH)_{n+1}\Delta t+\Delta H \tag*{(12)}

among the couples (Γn+1,un+1)(\Gamma_{n+1},\mathbf{u}_{n+1}) such that ∇⋅σn+1+fn+1=ρu¨n+1\nabla\cdot\boldsymbol{\sigma}_{n+1}+\mathbf{f}_{n+1}=\rho\ddot{\mathbf{u}}_{n+1}, ∥(Gn+1)⊥∥≤Gc\left\Vert(\mathbf{G}_{n+1})_{\perp}\right\Vert\leq G_c and the minimum is zero.

If the loading is slow and controlled by the displacements, the evolution is quasistatic and the functional to minimize is reduced to

ΔΠ(Γn+1,un+1)=Gc∣ΔΓ∣+ΔH\Delta\Pi(\Gamma_{n+1},\mathbf{u}_{n+1})=G_c\left|\Delta\Gamma\right|+\Delta H

and we recover Buliga’s incremental formulation [14].

Comparison of the constitutive law with experiments

The aim of this section is to check the validity of the normality law with respect to the experimental results on PMMA specimens [65]. A good test is to considered an initial straight crack in a plate which, under mixed mode loading, may extend suddenly in a direction deviating of an angle β\beta from the original one by kinking (Figure 8.1). The length of the kink crack is taken very small. The idea is to use the incremental formulation of the previous section with only one step. To simplify the notations, the initial value a0a_0 of a given quantity is denoted simply aa while the value a1a_1 at the end of the step is denoted a∗a^*. In Hellen et al. [43], Euler explicit scheme is used that can read with the simplified notations

Figure 8.1. Kinked crack

Δψ=ψ˙Δt=λGΔt\Delta\psi=\dot{\psi}\Delta t=\lambda G\Delta t

In the frame of origin OO at the initial crack tip, the axis Ox1Ox_1 in the direction ahead the initial crack and the axis Ox2Ox_2 perpendicular to Ox1Ox_1 within the plane (Figure 8.1), the driven force is given in terms of stress intensity factors (SIFs) by

G1=1E(KI2+KII2),G2=−1E2KIKIIG_1=\frac{1}{E}\left(K_I^2+K_{II}^2\right),\qquad G_2=-\frac{1}{E}2K_IK_{II}

where κ=3−4ν\kappa= 3 - 4\nu in plane strain, κ=(3−ν)/(1+ν)\kappa= (3-\nu)/(1+\nu) in plane stress and

1E=(1+ν)(1+κ)4E\frac{1}{E} = \frac{(1+\nu)(1+\kappa)}{4E}

The deviation angle β\beta of the kink crack is given by the slope of the driven force

β=tan⁡−1(−2KIKIIKI2+KII2)\beta= \tan^{-1}\left(-\frac{2K_I K_{II}}{K_I^2 + K_{II}^2}\right)

In mode II, the formula gives β=0∘\beta= 0^\circ, according to the experience. For KI=KIIK_I = K_{II}, β=−45∘\beta= -45^\circ but the experimental values are above −49∘-49^\circ. In mode IIII, β=0∘\beta= 0^\circ while the experimental values are in the range from −70∘-70^\circ to −74∘-74^\circ. Clearly, the predictions are disastrous. Our opinion is that the constitutive law must not be a priori rejected but it is the explicit scheme which is problematic.

The problem of the kinked crack were studied by many authors. It is an awkward problem of Elasticity and only approximated formula were proposed to express the SIFs at the kink crack tip in terms of the SIFs at the initial crack tip. At the limit of vanishing length of the kink crack, we adopt in the sequel the expression proposed in [23]

KI∗(β)=cos⁡3β2KI−3sin⁡β2cos⁡2β2KIIK_I^*(\beta) = \cos^3\frac{\beta}{2} K_I - 3\sin\frac{\beta}{2}\cos^2\frac{\beta}{2} K_{II}
KII∗(β)=sin⁡β2cos⁡2β2KI+cos⁡β2(1−3sin⁡2β2)KIIK_{II}^*(\beta) = \sin\frac{\beta}{2}\cos^2\frac{\beta}{2} K_I + \cos\frac{\beta}{2}\left(1-3\sin^2\frac{\beta}{2}\right)K_{II}

However, it is noticeable that a correction is proposed in [34] for the largest values of the ∣β∣\lvert\beta\rvert and the dependency with respect to the length of the kink crack is taken into account in [49, 2] but these improvements will be not considered here. With the simplified notations, the implicit scheme (10) reads

Δψ=ψ˙∗Δt=λG∗Δt(13)\Delta\psi= \dot{\psi}^{*}\Delta t = \lambda G^{*}\Delta t \tag*{(13)}

In the frame of origin KK at the kink crack tip, the axis Kx1∗Kx_1^* in the direction ahead the kink crack and the axis Kx2∗Kx_2^* perpendicular to Kx1∗Kx_1^* within the plane (Figure 8.1), the driven force is given in terms of stress intensity factors (SIFs) by

G1∗=1Eˉ[(KI∗)2+(KII∗)2],G2∗=−1Eˉ2KI∗KII∗G_1^* = \frac{1}{\bar{E}}\left[(K_I^*)^2 + (K_{II}^*)^2\right], \qquad G_2^* = -\frac{1}{\bar{E}}2K_I^*K_{II}^*

The normality law (13) entails

EˉG2∗=−2KI∗KII∗=0,Eˉ∥G∗∥=(KI∗)2+(KII∗)2,(14)\bar{E}G_2^* = -2K_I^*K_{II}^* = 0, \qquad\bar{E}\left\lVert\mathbf{G}^* \right\rVert= (K_I^*)^2 + (K_{II}^*)^2, \tag*{(14)}

To satisfy the former condition, two scenarios may be considered for the crack extension:

  1. Scenario 1. The kink crack inclination is β1\beta_1, solution of KII∗(β1)=0K_{II}^*(\beta_1) = 0, and Eˉ∥G∗∥=(KI∗(β1))2\bar{E}\left\lVert\mathbf{G}^* \right\rVert= (K_I^*(\beta_1))^2

  1. Scenario 2. The kink crack inclination is β2\beta_2, solution of KI∗(β2)=0K_I^*(\beta_2) = 0, and Eˉ∥G∗∥=(KII∗(β2))2\bar{E}\left\lVert\mathbf{G}^* \right\rVert= (K_{II}^*(\beta_2))^2

According to the stability criterion, the crack extends for the inclination βi\beta_i with the maximum driven force magnitude. Then scenario 1 is realized if (KI∗(β1))2>(KII∗(β2))2(K_I^*(\beta_1))^2 > (K_{II}^*(\beta_2))^2 and scenario 2 otherwise.

The admissible solutions must be searched in the interval ]−π,π[]-\pi,\pi[. The mode II singularity disappearing in compression, KIK_I is positive. In mode IIII, the SIF KIIK_{II} may be positive or negative, depending on the sign of the shear loading. Before tackling the mixed mode, let us consider the limit case:

  • Mode I. For scenario 1, there is a unique admissible solution β1=0∘\beta_1 = 0^\circ and (KI∗(β1))2=KI2(K_I^*(\beta_1))^2 = K_I^2. For scenario 2, there is no admissible solution. Then the scenario 1 occurs, according to the experimental observations.

Let us examine now the general case (KIKII≠0)(K_I K_{II} \ne0). For scenario 1, there are two admissible solutions

β1±=sin⁡−1(KIIKI±3KIIKI8(KIIKI)2+19(KIIKI)2+1)\beta_{1\pm} = \sin^{-1}\left(\frac{\frac{K_{II}}{K_I} \pm3\frac{K_{II}}{K_I}\sqrt{8\left(\frac{K_{II}}{K_I}\right)^2+1}}{9\left(\frac{K_{II}}{K_I}\right)^2+1}\right)

For scenario 2, there is a unique admissible solution

β2=2tan⁡−1(KI3KII)\beta_2 = 2\tan^{-1}\left(\frac{K_I}{3K_{II}}\right)

In particular, let us discuss the case:

  • Mixed mode KI=KII>0K_I = K_{II} > 0. For scenario 1, there are two admissible solutions: β1−=sin⁡−1(−4/5)=−53.13∘\beta_{1-} = \sin^{-1}(-4/5) = -53.13^\circ then (KI∗(β1−))2=(1.847KI)2(K_I^*(\beta_{1-}))^2 = (1.847K_I)^2 and β1+=sin⁡−1(1)=90∘\beta_{1+} = \sin^{-1}(1) = 90^\circ then (KI∗(β1+))2=(−1.14KI)2(K_I^*(\beta_{1+}))^2 = (-1.14K_I)^2. For scenario 2, there is a unique admissible solution β2=36.87∘\beta_2 = 36.87^\circ and (KII∗(β2))2=(0.948KI)2(K_{II}^*(\beta_2))^2 = (0.948K_I)^2. Comparing the values of (KI∗(β1−))2(K_I^*(\beta_{1-}))^2, (KI∗(β1+))2(K_I^*(\beta_{1+}))^2 and (KII∗(β2))2(K_{II}^*(\beta_2))^2, we conclude that the scenario 1 occurs with the angle β=β1−=−53.13∘\beta= \beta_{1-} = -53.13^\circ closed to the experimental values (−49∘-49^\circ for KII=0.874KIK_{II} = 0.874K_I and values in the range from −56∘-56^\circ to −59∘-59^\circ for KII=1.142KIK_{II} = 1.142K_I).

In this respect, it is worth to remark that, if the sign of KIIK_{II} is reversed, the sign of the kink angles are reversed too: β1−=53.13∘\beta_{1-}=53.13^\circ, β1+=−90∘\beta_{1+}=-90^\circ and β2=−36.87∘\beta_2=-36.87^\circ but the corresponding values of (KI∗(β1−))2(K_I^*(\beta_{1-}))^2, (KI∗(β1+))2(K_I^*(\beta_{1+}))^2 and (KII∗(β2))2(K_{II}^*(\beta_2))^2 are the same. We conclude that the sign of the kink angle is reversed: β=53.13∘\beta=53.13^\circ.

Finally, let us deal with the other limit case:

  • Mode II. For scenario 1, there are two admissible solutions β1±=2sin⁡−1(±1/3)=±70.47∘\beta_{1\pm}=2\sin^{-1}(\pm1/\sqrt{3})=\pm70.47^\circ and (KI∗(β1±))2=4KII2/3(K_I^*(\beta_{1\pm}))^2=4K_{II}^2/3. For scenario 2, there is a unique admissible solution β2=0∘\beta_2=0^\circ and (KII∗(β2))2=KI2(K_{II}^*(\beta_2))^2=K_I^2. Then the scenario 1 occurs and the sign of the kink angle depends of the one of mode IIII. For KII>0K_{II}>0, the prediction β=−70.47∘\beta=-70.47^\circ is just above the experimental values.

Comment 1. On the basis of the experimental data, Richard proposed an empirical criterion in terms of the SIFs at the original crack tip

KIKIc+(KIIKIIc)2=1\frac{K_I}{K_{Ic}}+\left(\frac{K_{II}}{K_{IIc}}\right)^2=1

According to (14), the crack extends when

(KI∗)2+(KII∗)2=EˉGc(K_I^*)^2+(K_{II}^*)^2=\bar{E}G_c

that gives, for the scenario 1 of mode I, KI2=EˉGcK_I^2=\bar{E}G_c then

KIc2=EˉGcK_{Ic}^2=\bar{E}G_c

and, for the scenario 1 of mode II, 4KII2/3=EˉGc=KIc24K_{II}^2/3=\bar{E}G_c=K_{Ic}^2 then

KIIcKIc=32=0.86\frac{K_{IIc}}{K_{Ic}}=\frac{\sqrt{3}}{2}=0.86

a value better than 2/3=0.81\sqrt{2/3}=0.81 proposed in [34], by comparison with the experimental data covering the range from 0.88 to 0.95. Moreover, our prediction for the mixed mode KI=KIIK_I=K_{II} leads to values KIKIc=0.54\frac{K_I}{K_{Ic}}=0.54 and KIIKIIc=0.62\frac{K_{II}}{K_{IIc}}=0.62 closed to the curve of Richard criterion.

Comment 2. At least in this example, the relevant scenario corresponds always to the Principle of Local Symmetry KII∗(β)=0K_{II}^*(\beta)=0 proposed in 1974 by Goldstein and Salganik [38]. This principle is valid at least for a kinked straight crack with vanishing kink crack but there can be no assurance that it is valid for arbitrary cracks and we recommend to replace it by the normality law proposed in this paper.

Comment 3. The present criterion should not be confused with the Maximum Energy Release Rate criterion, proposed in 1974 by Hussain et al. [45], for which the maximum must be searched among all the directions of the crack extension while with our criterion the maximum is found only among the directions satisfying the normality law, according to the different scenarios. As argued in [20], the questions of when and how a crack propagates should be simultaneously investigated and the energy conservation is not sufficient for such a task.

Comment 4. This example shows also that the implicit scheme must be preferred to the explicit scheme in fracture mechanics.

Calculation of the crack driving force

As we are working at time tt in this Section, the dependency with respect to the time will be omitted. In order to obtain the expression of GG, we use a special form of the calculus of variation performed on the jet space of order one ([1], [31], [52]). For more details about the jet spaces, the reader is referred for instance to [67]. The first jet prolongation of the smooth function u:Ωt→R3\boldsymbol{u} : \Omega_t \to\mathbb{R}^3 is the function j1uj^1\boldsymbol{u} from Ωt\Omega_t into the jet bundle J1(Ωt,R3)J^1(\Omega_t,\mathbb{R}^3) such that

j1u(x)=(x,u(x),∇u(x))j^1\boldsymbol{u}\left(\boldsymbol{x}\right) = \left(\boldsymbol{x},\boldsymbol{u}\left(\boldsymbol{x}\right),\nabla\boldsymbol{u}(\boldsymbol{x})\right)

The Hamiltonian (3) at time tt has the form

H=∫Ωth(x,u,∇u,p) d3x−∫∂Ω1f‾⋅u d(∂Ω1)H = \int_{\Omega_t} h(\boldsymbol{x},\boldsymbol{u},\nabla\boldsymbol{u},\boldsymbol{p})\,\mathrm{d}^3x - \int_{\partial\Omega_1} \overline{\boldsymbol{f}}\cdot\boldsymbol{u}\,\mathrm{d}(\partial\Omega_1)
H=∫Ωth(j1u,p) d3x−∫∂Ω1f‾⋅u d(∂Ω1)H = \int_{\Omega_t} h(j^1\boldsymbol{u},\boldsymbol{p})\,\mathrm{d}^3x - \int_{\partial\Omega_1} \overline{\boldsymbol{f}}\cdot\boldsymbol{u}\,\mathrm{d}(\partial\Omega_1)

The new viewpoint which consists in replacing the original field u\boldsymbol{u} by its first jet prolongation j1uj^1\boldsymbol{u} leads to perform variations not only on the field and its derivatives but also on the variable x\boldsymbol{x}. We want to calculate the functional derivative of HH in the direction δψ\delta\psi defined on the crack front ctc_t. The idea is to prolong it locally by a field δx\delta\boldsymbol{x} on an open cylindrical neighbourhood CηC_\eta of radius η\eta around the crack front, the field vanishing at radius η\eta and beyond (Figure 9.1). To avoid the singularity of the stress field, we replace CηC_\eta by Cη,r=Cη\C‾rC_{\eta,r}=C_\eta\backslash\overline{C}_r where CrC_r is an open cylindrical neighbourhood of radius r<ηr<\eta around the crack front. Next, we pass to the limit η→r→0\eta\to r\to0.

We consider now a new parameterization given by a regular map x=θ(y)\boldsymbol{x}=\theta(\boldsymbol{y}) of class C1C^1 on Cη,r′=θ−1(Cη,r)C'_{\eta,r}=\theta^{-1}(C_{\eta,r}) and we perform the variation of the function θ\theta, the new variable being y\boldsymbol{y}. The variation of the action is

δH=δ∫Cη,r′h(θ(y),u,∇yu⋅∇xy,(∇xy)T⋅p′)det⁡(∇yx) d3y(15)\delta H=\delta\int_{C'_{\eta,r}}h(\theta(\boldsymbol{y}),\boldsymbol{u},\nabla_y\boldsymbol{u}\cdot\nabla_x\boldsymbol{y},(\nabla_x\boldsymbol{y})^{T}\cdot\boldsymbol{p}')\det(\nabla_y\boldsymbol{x})\,\mathrm{d}^{3}y \tag*{(15)}

where p′\boldsymbol{p}' is the linear momentum in y\boldsymbol{y} coordinates. We do not give here the details of the calculations that can be found in Appendix A but only indicate the sketch of the method and the result. As usual in calculus of variation, we integrate by part. Next we consider the limit case where the function θ\theta approaches the identity of Ωt\Omega_t, y\boldsymbol{y} approaches x\boldsymbol{x}, p′\boldsymbol{p}' approaches p\boldsymbol{p} and Cη,r′C'_{\eta,r} approaches Cη,rC_{\eta,r}. Finally, the variation of the Hamiltonian has the form

δH=∫∂Cη,rn⋅T⋅δx d(∂Cη,ρ)−∫Cη,r(∇⋅T+f)⋅δx d3x(16)\delta H=\int_{\partial C_{\eta,r}}\boldsymbol{n}\cdot\boldsymbol{T}\cdot\delta\boldsymbol{x}\,\mathrm{d}(\partial C_{\eta,\rho})-\int_{C_{\eta,r}}(\nabla\cdot\boldsymbol{T}+\boldsymbol{f})\cdot\delta\boldsymbol{x}\,\mathrm{d}^{3}x \tag*{(16)}

where

T=h1R3−σ⋅∇u−u˙⊗p(17)\boldsymbol{T}=h\mathbf{1}_{\mathbb{R}^{3}}-\boldsymbol{\sigma}\cdot\nabla\boldsymbol{u}-\dot{\boldsymbol{u}}\otimes\boldsymbol{p} \tag*{(17)}

The contribution on the cylindrical part ∂Cη,r\partial C_{\eta,r} of radius η\eta vanishes as δx\delta\boldsymbol{x}. When η\eta approaches rr, the contributions on the remaining parts of ∂Cη,r\partial C_{\eta,r} excepted the cylindrical part of radius rr, vanish too. Let n′\boldsymbol{n}' the outward unit vector to ∂Cr\partial C_r, opposite to the outward unit vector n\boldsymbol{n} to ∂Cη,r\partial C_{\eta,r}, Figure 9.1

Cylindrical neighbourhood

Figure 9.1. Cylindrical neighbourhood

n′=−n\boldsymbol{n}'=-\boldsymbol{n}

Then using the polar coordinates r,θr,\theta, one has

δH=−lim⁡r→0∫∂Cr∩∂Cη,rn′⋅T⋅δx r dθ ds\delta H=-\lim_{r\to0}\int_{\partial C_r\cap\partial C_{\eta,r}}\boldsymbol{n}'\cdot\boldsymbol{T}\cdot\delta\boldsymbol{x}\,r\,\mathrm{d}\theta\,\mathrm{d}s

Finally, recalling that the functional derivative is a density on the crack front, the value of the driving force corresponding to the arc length ss is

G=−DδxH=lim⁡r→0∫cr,sn′⋅T r dθ(18)G=-D_{\delta\boldsymbol{x}}H=\lim_{r\to0}\int_{c_{r,s}}\boldsymbol{n}'\cdot\boldsymbol{T}\,r\,\mathrm{d}\theta \tag*{(18)}

where cr,sc_{r,s} is the line at the intersection between the cylindrical-shaped surface ∂Cr∩∂Cη,r\partial C_r\cap\partial C_{\eta,r} and the plane normal to the crack front ctc_t at the arc length ss. This limit is called by Gurtin et al. [40] a tip integral and denoted

G=∮tipn′⋅TG=\oint_{\mathrm{tip}}\boldsymbol{n}'\cdot\boldsymbol{T}

but we do not recover the last term of (17) in Gurtin’s and Stolz’s [70] expressions of T\boldsymbol{T}. In this form, the driving force appears as a generalization of Rice-Eshelby JJ integral.

In statics, this quantity has the well known property of path-independence and it is relevant to wonder whether this property remains true in dynamics. The elastic stress field being singular at the crack front, the volume force field may be neglected when rr approaches zero. In Appendix B, it is shown that for the natural evolution of the system

∇⋅T=(∇u)⋅p˙−(∇⋅u˙)p−2(∇sp)⋅u˙(19)\nabla\cdot\mathbf{T} = (\nabla\mathbf{u}) \cdot\dot{\mathbf{p}} - (\nabla\cdot\dot{\mathbf{u}})\mathbf{p} - 2(\nabla_{\mathrm{s}}\mathbf{p}) \cdot\dot{\mathbf{u}} \tag*{(19)}

Unfortunately, this expression does not vanish in general but only in the limit case of statics where we recover the classical Rice-Eshelby identity

∇⋅T=∇⋅[w 1R3−σ⋅∇u]=0\nabla\cdot\mathbf{T} = \nabla\cdot[w\,\mathbf{1}_{\mathbb{R}^{3}} - \boldsymbol{\sigma} \cdot\nabla\mathbf{u}] = 0

Then except the particular case of statics, there is no path-independence of the integral in dynamics and passing to the limit is required in (18).

Conclusions and perspectives

In this work, we presented a particular version of the non incremental symplectic BEN principle. It can be generalized in various ways:

  • The Hamiltonian HH depends on u\mathbf{u}, p\mathbf{p} and ψ\psi (through Ωt=Ω∖Γt\Omega_t = \Omega\setminus\Gamma_t) but not on π\boldsymbol{\pi}. It was a bias motivated by the wish to recover classical formulations of the fracture mechanics but we could give up this restrictive assumption by introducing an explicit dependence of the Hamiltonian on π\boldsymbol{\pi}.

  • For the same motivation, the dissipation potential ϕ\phi depends on π˙\dot{\boldsymbol{\pi}} but not on u˙\dot{\mathbf{u}}, p˙\dot{\mathbf{p}} and ψ˙\dot{\psi}. Nevertheless there is nothing preventing us from lifting this restriction, in particular concerning ψ˙\dot{\psi}.

  • The potential φ\varphi could fully depend on G\mathbf{G}, not only on G⊥\mathbf{G}_{\perp}. It could be also generalized to anisotropic behaviours with symmetry group techniques.

  • For easiness, we considered only 1-homogeneous potential φ\varphi in this work to illustrate the method but we can add extra terms to represent for instance the dependence of the critical energy release rate with respect to the crack extension vector as in [22, 46, 47, 48]. The crack extension by fatigue can be also modelled thanks to homogeneous potentials of degree linked to the crack growth law ([51], [12]).

  • We could easily take into account more sources of dissipation such as damping, plasticity, damage and so on, by considering in the symplectic formalism new internal variables and the corresponding momenta, and by simply adding terms in the dissipation potential ϕ\phi.

  • Sometimes there is no getting away from experimental facts and the kinetic law is non associated. For such atypical events, the present principle can be generalized, replacing the sum of the dissipation potential and its Fenchel polar by a bipotential.

We are convinced of the interest of this non incremental principle in computational structural mechanics because the error can be controlled uniformly on the whole evolution, in contrast to incremental methods which accumulate the errors and degrade the accuracy over time.

We hope that our application of the normality law to the kinked crack allows to take forward the discussion in the literature on the choice of the crack stability criterion.

In the future, we plan to develop numerical methods based on the present approach. We already have given a hint about the discretization of the crack flow at the end of § 2.2. In order to avoid a cumbersome remeshing, an efficient technique is the XFEM where the standard displacement-based approximation is enriched by incorporating both discontinuous fields and the near tip asymptotic fields ([59], [28], [29], [60]).

Acknowledgements

The author would like to thank Marius Buliga, Djimedo Kondo, Abdelbacet Oueslati, Pierre Gosselet, Céline Bouby and Long Cheng for their questions and comments during seminars that allowed to improve the paper.

Appendix

To calculate the expression of the crack driven force, we start from the variation of the Hamiltonian (15)

δH=δ∫Cn,r′h(θ(y),u,∇ ⁣yu⋅∇ ⁣xy,(∇ ⁣xy)T⋅p′)det⁡(∇ ⁣yx) d3y\delta H = \delta\int_{C'_{n,r}} h(\theta(\mathbf{y}),\mathbf{u},\nabla_{\!y}\mathbf{u}\cdot\nabla_{\!x}\mathbf{y},(\nabla_{\!x}\mathbf{y})^{T}\cdot\mathbf{p}')\det(\nabla_{\!y}\mathbf{x})\,\mathrm{d}^{3}y

and we take the variation with respect to the fields u\mathbf{u} and θ\boldsymbol{\theta}

δH=∫Cn,r′{Tr⁡(σ δ(∇ ⁣yu⋅∇ ⁣xy))+u˙⋅δ((∇ ⁣xy)T⋅p′)−f⋅δx}det⁡(∇ ⁣yx)+h δ(det⁡(∇ ⁣yx)) d3y(20)\delta H = \int_{C'_{n,r}} \left\{ \operatorname{Tr}\left(\boldsymbol{\sigma}\,\delta(\nabla_{\!y}\mathbf{u}\cdot\nabla_{\!x}\mathbf{y})\right)+\dot{\mathbf{u}}\cdot\delta\left((\nabla_{\!x}\mathbf{y})^{T}\cdot\mathbf{p}'\right)-\mathbf{f}\cdot\delta\mathbf{x}\right\}\det(\nabla_{\!y}\mathbf{x}) + h\,\delta(\det(\nabla_{\!y}\mathbf{x}))\,\mathrm{d}^{3}y \tag*{(20)}

Next we calculate the variation of the field derivative in terms of the derivative of the variation of the field

δ(∇ ⁣yu⋅∇ ⁣xy)=−∇ ⁣yu⋅∇ ⁣xy⋅∇ ⁣y(δx)⋅∇ ⁣xy\delta(\nabla_{\!y}\mathbf{u}\cdot\nabla_{\!x}\mathbf{y})=-\nabla_{\!y}\mathbf{u}\cdot\nabla_{\!x}\mathbf{y}\cdot\nabla_{\!y}(\delta\mathbf{x})\cdot\nabla_{\!x}\mathbf{y}
δ(det⁡(∇ ⁣yx))=Tr⁡(∇ ⁣y(δx)⋅adj⁡(∇ ⁣yx))\delta(\det(\nabla_{\!y}\mathbf{x}))=\operatorname{Tr}\left(\nabla_{\!y}(\delta\mathbf{x})\cdot\operatorname{adj}(\nabla_{\!y}\mathbf{x})\right)

where adj⁡(A)\operatorname{adj}(A) is the transposed of the matrix of cofactors of AA

adj⁡(A)=det⁡(A) A−1\operatorname{adj}(A)=\det(A)\,A^{-1}

Likewise, one has

δ((∇ ⁣xy)T⋅p′)=−(∇ ⁣xy⋅∇ ⁣y(δx)⋅∇ ⁣xy)T⋅p′=−(∇ ⁣xy)T⋅(∇ ⁣y(δx))T⋅(∇ ⁣xy)T⋅p′\delta\left((\nabla_{\!x}\mathbf{y})^{T}\cdot\mathbf{p}'\right)=-\left(\nabla_{\!x}\mathbf{y}\cdot\nabla_{\!y}(\delta\mathbf{x})\cdot\nabla_{\!x}\mathbf{y}\right)^{T}\cdot\mathbf{p}'=-\left(\nabla_{\!x}\mathbf{y}\right)^{T}\cdot\left(\nabla_{\!y}(\delta\mathbf{x})\right)^{T}\cdot\left(\nabla_{\!x}\mathbf{y}\right)^{T}\cdot\mathbf{p}'

Inserting the three previous expression into (20) and after simple manipulations, we obtain

δH=∫Cη,r′{Tr⁡(adj⁡(∇yx)⋅(h1R3−σ⋅∇yu⋅∇xy−(u˙⊗p′)⋅∇xy)⋅∇y(δx))−det⁡(∇yx) f⋅δx} d3y\delta H = \int_{C'_{\eta,r}} \left\{ \operatorname{Tr}\left(\operatorname{adj}(\nabla_y x)\cdot\left(h\mathbf{1}_{\mathbb{R}^3}-\boldsymbol{\sigma}\cdot\nabla_y\mathbf{u}\cdot\nabla_x y-(\dot{\mathbf{u}}\otimes\mathbf{p}')\cdot\nabla_x y\right)\cdot\nabla_y(\delta x)\right)-\det(\nabla_y x)\,\mathbf{f}\cdot\delta x\right\}\,\mathrm{d}^3y

Integrating by parts leads to

δH=∫∂Cη,r′n⋅adj⁡(∇yx)⋅(h1R3−σ⋅∇yu⋅∇xy−(u˙⊗p′)⋅∇xy)⋅δx d(∂Cη,r′)−∫Cη,r′{∇⋅(adj⁡(∇yx)⋅(h1R3−σ⋅∇yu⋅∇xy−(u˙⊗p′)⋅∇xy))+det⁡(∇yx)f}⋅δx d3y\delta H = \int_{\partial C'_{\eta,r}}\mathbf{n}\cdot\operatorname{adj}(\nabla_y x)\cdot\left(h\mathbf{1}_{\mathbb{R}^3}-\boldsymbol{\sigma}\cdot\nabla_y\mathbf{u}\cdot\nabla_x y-(\dot{\mathbf{u}}\otimes\mathbf{p}')\cdot\nabla_x y\right)\cdot\delta x\,\mathrm{d}(\partial C'_{\eta,r}) -\int_{C'_{\eta,r}}\left\{\nabla\cdot\left(\operatorname{adj}(\nabla_y x)\cdot\left(h\mathbf{1}_{\mathbb{R}^3}-\boldsymbol{\sigma}\cdot\nabla_y\mathbf{u}\cdot\nabla_x y-(\dot{\mathbf{u}}\otimes\mathbf{p}')\cdot\nabla_x y\right)\right)+\det(\nabla_y x)\mathbf{f}\right\}\cdot\delta x\,\mathrm{d}^3y

Considering the limit case where the function θ\theta approaches the identity of Cη,rC_{\eta,r}, y\mathbf{y} approaches x\mathbf{x}, p′\mathbf{p}' approaches p\mathbf{p} and Cη,r′C'_{\eta,r} approaches Cη,rC_{\eta,r}, we obtain (16)

δH=∫∂Cη,rn⋅T⋅δx d(∂Cη,ρ)−∫Cη,r(∇⋅T+f)⋅δx d3x\delta H=\int_{\partial C_{\eta,r}}\mathbf{n}\cdot\mathbf{T}\cdot\delta\mathbf{x}\,\mathrm{d}(\partial C_{\eta,\rho})-\int_{C_{\eta,r}}(\nabla\cdot\mathbf{T}+\mathbf{f})\cdot\delta\mathbf{x}\,\mathrm{d}^3\mathbf{x}

where occurs the tensor (17)

T=h1R3−σ⋅∇u−u˙⊗p\mathbf{T}=h\mathbf{1}_{\mathbb{R}^3}-\boldsymbol{\sigma}\cdot\nabla\mathbf{u}-\dot{\mathbf{u}}\otimes\mathbf{p}

Appendix B

Because of the stress singularity, the volume force that are regular may be consider as uniform near the crack front. In one hand, the variation of the Hamiltonian density hh in the direction δx\delta\mathbf{x} is

δh=∇h⋅δx=∇⋅(h1R3)⋅δx(21)\delta h=\nabla h\cdot\delta\mathbf{x}=\nabla\cdot(h\mathbf{1}_{\mathbb{R}^3})\cdot\delta\mathbf{x} \tag*{(21)}

In the other hand, according to (3), hh is a function of x\mathbf{x} through j1u\mathbf{j}^{1}\mathbf{u} and p\mathbf{p}

h=12ρ∥p∥2+w(∇su)−f(t)⋅uh=\frac{1}{2\rho}\lVert\mathbf{p}\rVert^2+w(\nabla_s\mathbf{u})-\mathbf{f}(t)\cdot\mathbf{u}

Using the chain rule and assuming the volume force are uniform, one has

δh=Tr⁡(σ⋅∇(δu))+u˙⋅δp−f⋅u\delta h=\operatorname{Tr}(\boldsymbol{\sigma}\cdot\nabla(\delta\mathbf{u}))+\dot{\mathbf{u}}\cdot\delta\mathbf{p}-\mathbf{f}\cdot\mathbf{u}

Using the linear momentum balance (intermediate constraints in (5) which is satisfied for the natural evolution of the system), the first term of the right hand member becomes

Tr⁡(σ⋅∇(δu))=∇⋅(σ⋅δu)−(∇⋅σ)⋅δu=∇⋅(σ⋅δu)+(f−p˙)⋅δu\operatorname{Tr}(\boldsymbol{\sigma}\cdot\nabla(\delta\mathbf{u}))=\nabla\cdot(\boldsymbol{\sigma}\cdot\delta\mathbf{u})-(\nabla\cdot\boldsymbol{\sigma})\cdot\delta\mathbf{u}=\nabla\cdot(\boldsymbol{\sigma}\cdot\delta\mathbf{u})+(\mathbf{f}-\dot{\mathbf{p}})\cdot\delta\mathbf{u}

Hence

δh=∇⋅(σ⋅δu)−p˙⋅δu+u˙⋅δp\delta h = \nabla\cdot(\boldsymbol{\sigma} \cdot\delta\mathbf{u}) - \dot{\mathbf{p}} \cdot\delta\mathbf{u} + \dot{\mathbf{u}} \cdot\delta\mathbf{p}

The infinitesimal variations of the displacement and linear momentum fields resulting from an infinitesimal arbitrary variation δx\delta\mathbf{x} being

δu=∇u⋅δx,δp=∇p⋅δx\delta\mathbf{u} = \nabla\mathbf{u} \cdot\delta\mathbf{x}, \qquad\delta\mathbf{p} = \nabla\mathbf{p} \cdot\delta\mathbf{x}

and in particular if the field δx\delta\mathbf{x} is uniform

δh=[∇⋅(σ⋅∇u)−p˙⋅∇u+u˙⋅∇p]⋅δx(22)\delta h = \left[\nabla\cdot(\boldsymbol{\sigma} \cdot\nabla\mathbf{u}) - \dot{\mathbf{p}} \cdot\nabla\mathbf{u} + \dot{\mathbf{u}} \cdot\nabla\mathbf{p}\right] \cdot\delta\mathbf{x} \tag*{(22)}

The expressions (21) and (22) being equal for every δx\delta\mathbf{x}, we obtain

∇⋅(h1R3−σ⋅∇u)=(∇u)T⋅p˙−(∇p)T⋅u˙\nabla\cdot\left(h \mathbf{1}_{\mathbb{R}^{3}} - \boldsymbol{\sigma} \cdot\nabla\mathbf{u}\right) = (\nabla\mathbf{u})^{T} \cdot\dot{\mathbf{p}} - (\nabla\mathbf{p})^{T} \cdot\dot{\mathbf{u}}

Taking into account the identity

∇⋅(u˙⊗p)=(∇⋅u˙)p+(∇p)⋅u˙\nabla\cdot(\dot{\mathbf{u}} \otimes\mathbf{p}) = (\nabla\cdot\dot{\mathbf{u}})\mathbf{p} + (\nabla\mathbf{p}) \cdot\dot{\mathbf{u}}

we obtain (19)

∇⋅T=(∇u)⋅p˙−(∇⋅u˙)p−2(∇sp)⋅u˙\nabla\cdot\mathbf{T} = (\nabla\mathbf{u}) \cdot\dot{\mathbf{p}} - (\nabla\cdot\dot{\mathbf{u}})\mathbf{p} - 2(\nabla_s \mathbf{p}) \cdot\dot{\mathbf{u}}

where occurs the tensor (17)

T=h1R3−σ⋅∇u−u˙⊗p\mathbf{T} = h\mathbf{1}_{\mathbb{R}^{3}} - \boldsymbol{\sigma} \cdot\nabla\mathbf{u} - \dot{\mathbf{u}} \otimes\mathbf{p}

References

  1. [1]Aldaya, V., Azcárraga, J. A., 1978. Variational principles on rth order jets of fibre bundles in field theory, J. Math. Phys. 19(9), 1869-1975.DOI
  2. [2]Amestoy, M., Leblond, J.B., 1992. Crack path in plane situation – II. Detailed form of the expansion of the stress intensity factors. Int J Solids Struct. 29, 465–501.DOI
  3. [3]Ambrosio, L., Tortorelli, V., 1992. On the approximation of free discontinuity problems. Bollett UMI. 7(6B), 105–123.
  4. [4]Aubin, J.P., Cellina, A.,Nohel, J., 1997. Monotone trajectories of multivalued dynamical systems. Annali di Matematica Pura ed Appl. 115, 99-117.
  5. [5]Aubin, J.P., 2002. Boundary-Value Problems for Systems of Hamilton-Jacobi-Bellman Inclusions with Constraints. SIAM J. Control. 41, 425-456.
  6. [6]Barenblatt, G.I., 1959. The formation of equilibrium cracks during brittle fracture: general ideas and hypotheses, Axially-symmetric Cracks. Prikladnaya Matematika i Mekhanika. 23, 622-636.DOI
  7. [7]Bloch, A.M., Krishnaprasad, P.S., Marsden, J.E., Ratiu, T.S., 1994. Dissipation induced instabilities. Ann. de l’Institut Henri Poincaré. Analyse non linéaire. 11(1), 37-90.
  8. [8]Bodovillé, G., 1999. On damage and implicit standard materials. C. R. Acad. Sci. Paris Série IIB. 327(8), 715-720.
  9. [9]Bobovillé, G. de Saxcé, G., 2001. Plasticity with non linear kinematic hardening : modelling and shakedown analysis by the bipotential approach. Eur. J. Mech. A/Solids. 20, 99-112.
  10. [10]Bouby, C., de Saxcé, G., Tritsch, J.B., 2009. On shakedown of structures under variable loads with a kinematic non linear and non associated hardening rule, in: Weichert, D., Ponter, A. (Eds.), Limit States of Materials and Structures: Direct Methods. Springer.DOI
  11. [11]Bouby, C., Kondo, D., de Saxcé, G., 2015. A comparative analysis of two formulations for non linear hardening plasticity models: Application to shakedown analysis. European Journal of Mechanics A/Solids. 53, 48-61.DOI
  12. [12]Bourdin, B., Francfort, G., Marigo, J.J., 2008. The variational approach to fracture. J of Elasticity. 91, 5-148.
  13. [14]Buliga, M., 1999. Energy minimizing brittle fracture propagation. J. of Elasticity. 52, 201-238.DOI
  14. [15]Buliga, M., 2009. Hamiltonian inclusions with convex dissipation with a view towards applications. Math Appl 1(2), 228–225.arxiv.org/abs/0810.1419
  15. [16]Buliga, M., de Saxcé, G., 2016. A symplectic Brezis-Ekeland-Nayroles principle. Mathematics and Mechanics of Solids. 22(6), 1-15.DOI
  16. [17]Cao, X., Oueslati, A., Nguyen, A.D., de Saxcé, G., 2020. Numerical simulation of elastoplastic problems by Brezis-Ekeland-Nayroles non-incremental variational principle. Computational Mechanics. 65(4), 1006-1018DOI
  17. [18]Cao, X., Oueslati, A., Shirafkan, N., Bamer, F., Markert, B., de Saxcé, G., 2021. A non-incremental numerical method for dynamic elastoplastic problems by the symplectic Brezis-Ekeland-Nayroles principle. Computer Methods in Applied Mechanics and Engineering. 384, 11908.DOI
  18. [19]Cao, X., Oueslati, A., de Saxcé, G., 2021. A non-incremental approach for elastoplastic plates basing on the Brezis-Ekeland-Nayroles principle. Applied Mathematical Modelling. 99, 359-379.DOI
  19. [20]Chambolle, A., Francfort, G.A., Marigo, J.J., 2009. When and how do cracks propagate? J of the Mechanics and Physics of Solids. 57, 1614-1622.DOI
  20. [21]Cherepanov, G.P., 1963. One problem of indentor testing with the formation of cracks. Prikladnaya Matematika i Mekhanika. 27(1) (in Russian).
  21. [22]Chopin, J., Bhaskar, A., Jog, A., Ponson, L., 2018. Deppinning dynamics of crack fronts. Phys. Rev. Lett. 121, 235501.
  22. [23]Cotterell, B., Rice, J.R., 1980. Some remarks. on elastic crack-tip stress fields. Int. J. of Fracture. 16, 155-169
  23. [24]de Saxcé, G., Feng, Z.Q., 1991. New inequation and functional for contact with friction : the implicit standard material approach. Int. J. Mech. of Struct. and Machines. 19(3), 301-325.DOI
  24. [25]de Saxcé, G., 1992. Une généralisation de l’inégalité de Fenchel et ses applications aux lois constitutives. C. R. Acad. Sci. Paris Série II. 314, 125-129.
  25. [26]de Saxcé, G., Bousshine, L., 1998. Limit Analysis Theorems for the Implicit Standard Materials: Application to the Unilateral Contact with Dry Friction and the Non Associated Flow Rules in Soils and Rocks. Int. J. Mech. Sci. 40(4), 387-398.DOI
  26. [27]de Saxcé, G., 2002. Implicit standard materials, in: Weichert, D., Maier, G. (Eds.), Inelastic Behaviour of Structures under Variable Repeated Loads, CISM International Centre for Mechanical Sciences, Courses and Lectures N° 432. Springer.DOI
  27. [28]Dolbow, J., Moës, N., Belytschko, T., 2000. Discontinuous enrichment in finite elements with a partition of unity method. Finite Elements in Analysis and Design. 36, 235-260.DOI
  28. [29]Dolbow, J., Moës, N., Belytschko, T., 2000. Modeling fracture in Mindlin-Reissner plates with the extended finite element method. Int. J. of Solids and Structures. 37, 7161-7183.DOI
  29. [30]Dugdale, D.S., 1960. Yielding of steel sheets containing slits. J. Mech. Phys. Sol. 8, 100-104.DOI
  30. [31]Edelen, D.G.B, 1985. Applied Exterior Calculus. John Wiley and sons.
  31. [32]Erdogan, F., Sih, G.C., 1963. On the crack extension in plates under plane loading and transverse shear. Trans. ASME, ser. D. 85(4), 519-527.DOI
  32. [33]Fenchel, W., 1949. On conjugate convex functions. Canadian Journal of Mathematics. 1, 1105.
  33. [34]Fett, T., Munz, D., 2002. Kinked cracks and Richard fracture criterion. Int. J. of Fracture. 115, L69-L73DOI
  34. [35]Fitzpatrick, S., 1988. Representing monotone operators by convex functions, in: Workshop/Miniconference on Functional Analysis and Optimization. Canberra, Proc. Centre Math. Anal. Austral. Nat. Univ., 20, Austral. Nat. Univ, pp. 59-65.
  35. [36]Francfort, G.A., Marigo, J.J., 1998. Revisiting brittle fracture as an energy minimization problem. J. Mech. Phys. Solids. 46, 1319-1342.DOI
  36. [37]Ghoussoub, N., MacCann, R.J., 2004. A least action principle for steepest descent in non-convex landscape. Contemporary Mathematics. 362, 177-187.DOI
  37. [38]Goldstein, R.V., Salganik, R.L., 1974. Brittle fracture of solids with arbitrary cracks. Int. J. of Fracture. 10, 507–23.DOI
  38. [39]Grmela, M., Öttinger, H.C., 1997. Dynamics and thermodynamics of complex fluids. I. Development of a general formalism. Phys. Rev. E 56(6), 6620-6632.DOI
  39. [40]Gurtin, M., Podio-Guidugli, P., 1998. Configurational forces and a constitutive theory for crack propagation that allows for kinking and curving. J. Mech. Phys. Solids. 46, 1343–1378.DOI
  40. [41]Halphen, B., Nguyen Quoc, S., 1975. Sur les matériaux standard généralisés. Journal de Mécanique. 14, 39-63.
  41. [42]He, M., Hutchinson, J., 1989. Crack deflection at an interface between dissimilar elastic materials. Int. J. Solids Struct. 25, 1053–1067.
  42. [43]Hellen, T.K., Blackburn, W.S., 1975. The calculation of stress intensity factors for combined tensile and shear loading. Int J Fract. Mech. 11, 605–17.DOI
  43. [44]Hjiaj, M., Bobovillé, G. de Saxcé, G., 2000. Matériaux viscoplastiques et loi de normalité implicites. C. R. Acad. Sci. Paris Série IIb. 328, 519-524.DOI
  44. [45]Hussain, M.A., Pu, S.L., Underwood, J.H., 1974. Strain energy release rate for a crack under combined mode I and mode II. Fract Anal ASTM STP. 560, 2–28.
  45. [46]Kolvin, I., Cohen, G., Fineberg, J., 2015. Crack front dynamics: the interplay of singular geometry and crack instabilities. Phys. Rev. Lett. 114, 175501.DOI
  46. [47]Lebihain, M., Leblond, J.B., Ponson, L., 2020. Effective toughness of periodic heterogeneous materials: the effect of out-of-plane excursions of cracks. J of the Mechanics and Physics of Solids. 137, 103876.
  47. [48]Lebihain, M., Ponson, L., Leblond, J.B., Kondo, D., 2021. Effective toughness of disordered brittle solids: A homogenization framework. J of the Mechanics and Physics of Solids. 153, 104463
  48. [49]Leblond, J.B., 1989. Crack paths in plan situation – I. General form of the expansion of the stress intensity factors. Int J Solids Struct. 25, 1311–25.DOI
  49. [50]Leguillon, D., 2002. Strength or toughness? A criterion for crack onset at a notch. Eur. J. Mech. A/Solids. 21, 61-72.DOI
  50. [51]Lemaitre, J., Chaboche, J.L., 2012. Mechanics of solid materials, Cambridge University Press.
  51. [52]Mangiarotti, L., Modugno, M., 1983. Some results on the calculus of variation on jet spaces, Ann. Inst. H. Poincaré. 23(1), 29-43.
  52. [53]Manh Hong D., Peletier, M.A., Zimmer, J., 2013. GENERIC formalism of a Vlasov-Fokker-Planck equation and connection to large-deviation principles. Non-linearity. 26, 2951-2971.
  53. [54]Magnier, V., Charkaluk, E., Bouby, C., de Saxcé, G., 2006. Bipotential Versus Return Mapping Algorithms: Implementation of Non-Associated Flow Rules, in: Topping, B.H.V., Montero, G., Montenegro, R. (Eds.), Proceedings of The Eighth International Conference on Computational Structures Technology (las Palmas de Gran Canaria, sept. 12-15, 2006). Civil-Comp Press, Stirling-shire, United Kingdom.
  54. [55]Mielke, A., Theil, F., 1999. A mathematical model for rate-independent phase transformations with hysteresis, in: Alber, H.D., Balean, R., Farwig, R. (Eds.), Workshop on Models of Continuum Mechanics in Analysis and Engineering, Shaker-Verlag, pp.117-129.
  55. [56]Mielke, A., 2005, in: Feireisl, E. (Ed.), Handbook of Differential Equations, Evolutionary Equations, vol. 2. Elsevier, pp.461-559.
  56. [57]Mielke, A., Roubíček, T., 2006. Rate-independent damage processes in nonlinear elasticity. Mathematical Models and Methods in Applied Sciences (M3AS). 16(2), 177-209.
  57. [58]Mielke, A., 2011. Formulation of thermo-elastic dissipative material behavior using GENERIC. Contin. Mech.Thermodyn. 23, 233-256.DOI
  58. [59]Moës, N., Dolbow, J., Belytschko, T., 1999. A finite element method for crack growth without remeshing. Int. J. for Numerical Methods in Engineering 46:131-150.
  59. [60]Moés, N., Belytschko, T., 2002. Extended finite element method for cohesive crack growth. Engineering Fracture Mechanics. 69, 813-833.DOI
  60. [61]Mróz, K.P., Mróz, Z., 2010. On crack path evolution rules. Engineering Fracture Mechanics. 77, 1781-1807DOI
  61. [62]Mumford, D., Shah, J., 1989. Optimal approximation by piecewise smooth functions and associated variational problems. Comm. on Pure and Appl. Math. 42(5), 577–685.DOI
  62. [63]Nayroles, B., 1976. Deux théorèmes de minimum pour certains systèmes dissipatifs. C. R. Acad. Sci. Paris Série A-B. 282, A1035-A1038.
  63. [64]Öttinger, H.C., Grmela, M., 1997. Dynamics and thermodynamics of complex fluids. II. Illustrations of a general formalism. Phys. Rev. E. 56(6), 6633-6655.DOI
  64. [65]Richard, H.A., 1984. Examination of brittle fracture criteria for overlapping mode I and II loading applied to cracks, in: Sih, G.C. et al. (Eds.), Applications of Fracture Mechanics to Materials and Structures. Nijhoff Publ., The Hague, pp. 309-316.
  65. [66]Rockafellar, R.T., 1970. Generalized Hamiltonian equations for convex problems of Lagrange. Pacific J. of Math. 33(2), 411-427.DOI
  66. [67]Saunders, D.J., 1989. The Geometry of Jet Bundles, Cambridge University Press.
  67. [68]Sih, G.C., 1973. Some basic problems in fracture mechanics and new concepts. Engineering Fracture Mechanics. 5, 365-377DOI
  68. [69]Stefanelli, U., 2008. A variational principle for hardening elasto-plasticity. SIAM J. Math. Anal. 40(2), 623-652.
  69. [70]Stolz, C., 1995. Functional approach in nonlinear dynamics. Arch. Mech. 47, 421-435.
  70. [71]Strifors, H.C., 1973. A generalized force measure of conditions at crack tips. Int. J. Solids Struc. 10,1389–404.DOI
  71. [72]Tvergaard, V., Hutchinson, J.W., 1992. The relation between crack growth resistance and fracture process parameters in elastic-plastic solids. J. Mech. Phys. Sol. 40, 1377-1397.DOI
  72. [73]Vallée, V., Lerintiu, C., Fortuné, D., Ban, M., de Saxcé, G., 2005. A bipotential expressing simultaneous ordered spectral decomposition between stress and strain rate tensor, in: International conference New Trends in Continuum Mechanics (ed Theta), Constanta (Romania), 8-12 September 2003. Published under the title ”Hill’s bipotential”, New Trends in Continuum Mechanics, pp.339-351.
  73. [75]Visintin, A., 2013. Structural stability of rate-independent nonpotential flows. Discrete and Continuous Dynamical Systems Series S. 6, 257-275.DOI
  74. [76]Wu, C.H., 1978. Fracture under combined loads by maximum-energy-release-rate criterion. ASME J Appl Mech. 45, 553-8.DOI
  75. [77]Xu, X., Needleman, A., 1994. Numerical simulations of fast crack growth in brittle solids. J. Mech. Phys. Sol. 42, 1397-1434.DOI
  76. [78]Zouain, N., Pontes Filho, I., Vaunat, J., 2010. Potentials for the modified Cam-Clay model. European Journal of Mechanics A/Solids. 29, 327-336.DOI

Paper details

vlnc:W_96q8f383
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents