betaFeedback

WorksMathematics

Nonlocal regularization of sharp-crack energetics: a variational bridge from Griffith geometry to emergent constitutive degradation

Abstract

We develop a variational diffuse-crack model based on nonlocal averaging of tensile strain energy. A bounded potential acts on the scaled average, while the compressive energy remains local. The initial slope preserves the averaged tensile strain-energy density to leading order. The first variation yields a constitutive stress in which the tensile contribution is multiplied by the adjoint average of the derivative of the saturation potential. The resulting degradation is therefore determined by the stored-energy functional, without introducing an independent fracture field or prescribing a separate degradation law. For fixed smooth displacement fields with bounded strain, the energy density converges to the classical elastic strain-energy density as the averaging radius tends to zero. The Griffith surface-energy coefficient is recovered for planar opening transitions. A structure-preserving scalar auxiliary variable (SAV) scheme satisfies an exact discrete balance between the modified-energy increment and external work. We compare SAV and velocity-Verlet solutions for dynamic crack branching and Kalthoff–Winkler impact.

Introduction

Fracture couples solid deformation to the creation of an internal surface. Griffith’s energetic description balances elastic-energy release against the energy required to create new crack surfaces [1]. Variational fracture treats displacement and crack geometry as unknowns [2, 3], allowing crack paths to emerge from energetic competition. Computing this evolution requires a numerical representation of displacement jumps across the growing crack surface. Enriched finite elements accommodate these jumps without a crack-conforming mesh [4], but still require a separate representation of the crack geometry that is updated during propagation.

Diffusive fracture methods instead approximate the crack-surface energy by a volume integral of a fracture-energy density concentrated in a narrow region around the crack. Building on elliptic approximations of free-discontinuity functionals [5], phase-field fracture couples displacement to a scalar crack field [6]. In this framework, a crack-surface density defines the regularized fracture energy, while a degradation function couples the elastic energy to the crack field. Cohesive phase-field extensions use these functions to represent traction–separation laws [7, 8]. Phase-field models have been applied to a broad range of fracture problems, including dynamic fracture [9, 10], ductile fracture in elastoplastic solids [11], finite-strain fracture in hyperelastic solids [12], and multiphysics fracture [13, 14, 15]. Across these applications, phase-field models avoid explicit crack-surface tracking by solving for an additional fracture field coupled to displacement.

Nonlocal and gradient-enhanced damage models introduce spatial scales to regularize softening and localization [16, 17], and variational gradient damage connects these processes to brittle-fracture energetics [18]. Other routes include virtual and geometric internal-bond models [19, 20], peridynamics [21, 22], and nonlocal geometric fracture theory [23]. Eliminating a local damage variable can also produce a saturating effective energy [24]. The elimination of a local damage variable suggests a formulation with displacement as its sole primary variable. The aim here is to combine this feature with nonlocal spatial regularization in a variational diffuse-crack model grounded in sharp-crack energetics.

Nonlocal approximations of free-discontinuity energies connect spatial averages of gradient- or strain-dependent energy densities to limiting functionals with surface-energy contributions on the jump set. Braides and Dal Maso [25] established a nonlocal Γ\Gamma-approximation of the scalar Mumford–Shah functional. For unsplit elastic energies based on linearized strain, Griffith-type Γ\Gamma-limits have been established for nonlocal energies in both continuum and finite-element settings [26, 27].

We develop a variational diffuse-crack model for dynamic fracture based on nonlocal averaging of tensile strain energy. A bounded potential acts on the scaled tensile-energy average, while the compressive energy remains local. Tensile stress degradation emerges from the exact first variation, with displacement as the sole primary field. We further construct a structure-preserving time discretization for the resulting dynamics. The specific contributions are as follows:

  1. We specify the potential Φ\Phi to recover the classical elastic strain-energy density for fixed smooth displacement fields as the averaging radius tends to zero, and the Griffith fracture energy per unit area for planar opening. With its initial slope and limiting value fixed, the shape of Φ\Phi determines the homogeneous tensile strength at a fixed averaging radius.

  1. We introduce tension–compression splitting into the nonlocal energy, saturating only the averaged tensile contribution and retaining local compressive energy. Constitutive degradation emerges from the exact first variation of the energy, without an independent fracture field or a separately prescribed degradation law.

  1. We construct a linearly implicit, structure-preserving scalar auxiliary variable (SAV) scheme that satisfies an exact discrete balance between the modified-energy increment and the external work.

We compare the SAV and velocity-Verlet solutions for two standard planar dynamic-fracture benchmarks: dynamic crack branching and Kalthoff–Winkler impact. The split–unsplit comparison indicates that tension–compression splitting provides a more physically plausible description of mixed-mode dynamic cracking.

Section 2 introduces the model with tension–compression splitting and derives its constitutive stress and governing equations. Section 3 examines the elastic and opening-crack limits, emergent constitutive degradation and homogeneous strength. Section 4 develops the SAV scheme and its discrete energy law. Section 5 presents the spatial discretization and the two dynamic benchmarks.

Variational formulation and governing equations

We regularize the Griffith energy by applying a bounded potential to the scaled nonlocal average of tensile strain energy and retaining the compressive energy locally. We then derive the constitutive stress and governing equations.

The Griffith energy of sharp cracks

Let Ω⊂Rd\Omega\subset\mathbb{R}^{d} be a bounded open set, d≥2d \ge2, and let u:Ω→Rd\mathbf{u} : \Omega\to\mathbb{R}^{d} denote the displacement field. The infinitesimal strain tensor is defined by

e(u)=sym⁡∇u=12(∇u+(∇u)T).(1)\mathbf{e}(\mathbf{u}) = \operatorname{sym}\nabla\mathbf{u} = \frac{1}{2}\left(\nabla\mathbf{u} + (\nabla\mathbf{u})^{\mathsf{T}}\right). \tag*{(1)}

The jump set JuJ_{\mathbf{u}} of the displacement field is defined through its one-sided approximate limits. For x∈Ω\mathbf{x} \in\Omega, ρ>0\rho> 0, and ν∈Sd−1\boldsymbol{\nu} \in\mathbb{S}^{d-1}, let

Bρ±(x,ν):={y∈Bρ(x):±(y−x)⋅ν>0}.(2)B_{\rho}^{\pm}(\mathbf{x},\boldsymbol{\nu}) := \left\{\mathbf{y} \in B_{\rho}(\mathbf{x}) : \pm(\mathbf{y}-\mathbf{x})\cdot\boldsymbol{\nu} > 0\right\}. \tag*{(2)}

Then

Ju:={x∈Ω:∃ u±(x)∈Rd, νu(x)∈Sd−1, u+(x)≠u−(x), lim⁡ρ→01ρd∫Bρ±(x,νu(x))(1∧∣u(y)−u±(x)∣) dy=0}.(3)J_{\mathbf{u}} := \left\{\mathbf{x} \in\Omega: \exists\,\mathbf{u}^{\pm}(\mathbf{x}) \in\mathbb{R}^{d},\ \boldsymbol{\nu}_{\mathbf{u}}(\mathbf{x}) \in\mathbb{S}^{d-1},\ \mathbf{u}^{+}(\mathbf{x}) \ne\mathbf{u}^{-}(\mathbf{x}),\ \lim_{\rho\to0}\frac{1}{\rho^{d}}\int_{B_{\rho}^{\pm}(\mathbf{x},\boldsymbol{\nu}_{\mathbf{u}}(\mathbf{x}))}\left(1\wedge\left|\mathbf{u}(\mathbf{y})-\mathbf{u}^{\pm}(\mathbf{x})\right|\right)\,\mathrm{d}\mathbf{y}=0\right\}. \tag*{(3)}

where a∧b:=min⁡{a,b}a \wedge b := \min\{a,b\} denotes the minimum. The total potential energy, including the Griffith fracture energy, is defined by

Π0[u]=∫Ω∖JuW(e(u)) dx+GcHd−1(Ju)−ℓext[u],(4)\Pi_{0}[\mathbf{u}] = \int_{\Omega\setminus J_{\mathbf{u}}} W(\mathbf{e}(\mathbf{u}))\,\mathrm{d}\mathbf{x} + G_{c}\mathcal{H}^{d-1}(J_{\mathbf{u}}) - \ell_{\mathrm{ext}}[\mathbf{u}], \tag*{(4)}

where WW is the elastic strain-energy density, Gc>0G_{c} > 0 is the critical energy-release rate, ℓext[u]\ell_{\mathrm{ext}}[\mathbf{u}] is the external work functional, and Hd−1(Ju)\mathcal{H}^{d-1}(J_{\mathbf{u}}) is the Hausdorff measure of the sharp-crack interface. The natural setting for (4) is the space of generalized special functions of bounded deformation, GSBDp(Ω)GSBD^{p}(\Omega), which admits displacement jumps [28]. The main difficulty is that the crack geometry JuJ_{\mathbf{u}} is unknown [1, 2].

Nonlocal regularization with a tension–compression split

To retain compressive stiffness in the regularized model, we first separate the tensile and compressive contributions to the elastic strain energy. For an isotropic linearly elastic solid, the strain-energy density is

W(e)=λ2(tr⁡e)2+μ e:e,(5)W(\mathbf{e}) = \frac{\lambda}{2}(\operatorname{tr}\mathbf{e})^2 + \mu\,\mathbf{e}:\mathbf{e}, \tag*{(5)}

where λ≥0\lambda\ge0 and μ>0\mu> 0 are Lamé parameters. Following the strain-spectral decomposition [29], define

e=∑α=1dηαnα⊗nα,e±=∑α=1d⟨ηα⟩±nα⊗nα.(6)\mathbf{e} = \sum_{\alpha=1}^{d} \eta_\alpha\mathbf{n}_\alpha\otimes\mathbf{n}_\alpha,\qquad\mathbf{e}^{\pm} = \sum_{\alpha=1}^{d} \langle\eta_\alpha\rangle_{\pm}\mathbf{n}_\alpha\otimes\mathbf{n}_\alpha. \tag*{(6)}

where ηα\eta_\alpha and nα\mathbf{n}_\alpha are the eigenvalues and corresponding orthonormal eigenvectors of e\mathbf{e}, respectively, and ⟨r⟩+=max⁡(r,0)\langle r\rangle_{+} = \max(r,0) and ⟨r⟩−=min⁡(r,0)\langle r\rangle_{-} = \min(r,0). The elastic strain-energy density then admits the decomposition

W(e)=W+(e)+W−(e),(7)W(\mathbf{e}) = W^{+}(\mathbf{e}) + W^{-}(\mathbf{e}), \tag*{(7)}

where W±(e)=λ2⟨tr⁡e⟩±2+μ e±:e±W^{\pm}(\mathbf{e}) = \frac{\lambda}{2}\langle\operatorname{tr}\mathbf{e}\rangle_{\pm}^{2} + \mu\,\mathbf{e}^{\pm}:\mathbf{e}^{\pm}. For a scalar field ff, define the nonlocal averaging operator by

(Aεf)(x)=1mε(x)∫Bε(x)∩Ωf(y) dy,(8)(\mathcal{A}_{\varepsilon}f)(\mathbf{x}) = \frac{1}{m_{\varepsilon}(\mathbf{x})}\int_{B_{\varepsilon}(\mathbf{x})\cap\Omega} f(\mathbf{y})\,\mathrm{d}\mathbf{y}, \tag*{(8)}

where Bε(x)B_{\varepsilon}(\mathbf{x}) is the open ball centered at x\mathbf{x} with radius ε\varepsilon and mε(x)=∣Bε(x)∩Ω∣m_{\varepsilon}(\mathbf{x}) = |B_{\varepsilon}(\mathbf{x})\cap\Omega|. To regularize tensile fracture, we apply a bounded saturation potential to the scaled nonlocal average of tensile strain energy. The lower-semicontinuous, nondecreasing potential Φ:[0,∞)→[0,∞)\Phi: [0,\infty) \to[0,\infty) satisfies

Φ(0)=0,lim⁡s→0Φ(s)s=1,lim⁡s→∞Φ(s)=Gc2.(9)\Phi(0)=0,\qquad\lim_{s\to0}\frac{\Phi(s)}{s}=1,\qquad\lim_{s\to\infty}\Phi(s)=\frac{G_c}{2}. \tag*{(9)}

The small-ss condition implies Φ(s)=s+o(s)\Phi(s)=s+o(s) and preserves the leading-order tensile elastic contribution. The saturation value Gc/2G_c/2 recovers the Griffith surface energy in the planar-opening limit. We use the exponential saturation potential

Φexp⁡(s)=Gc2(1−exp⁡[−2sGc]).(10)\Phi_{\exp}(s)=\frac{G_c}{2}\left(1-\exp\left[-\frac{2s}{G_c}\right]\right). \tag*{(10)}

Appendix A lists other saturation potentials.

Remark 2.1. In three dimensions, WW has units J/m3\mathrm{J}/\mathrm{m}^{3}, so s=εAεWs=\varepsilon\mathcal{A}_{\varepsilon}W has units J/m2\mathrm{J}/\mathrm{m}^{2}, the same as GcG_c. Thus Φ\Phi also has units J/m2\mathrm{J}/\mathrm{m}^{2} and Φ(s)/ε\Phi(s)/\varepsilon is an energy density.

The split stored-energy functional is

Fεsplit[u]=∫ΩW−(e(u)) dx+1ε∫ΩΦ(sε+[u]) dx.(11)\mathcal{F}_{\varepsilon}^{\mathrm{split}}[\mathbf{u}]=\int_{\Omega}W^{-}(\mathbf{e}(\mathbf{u}))\,\mathrm{d}x+\frac{1}{\varepsilon}\int_{\Omega}\Phi\left(s_{\varepsilon}^{+}[\mathbf{u}]\right)\,\mathrm{d}x. \tag*{(11)}

Here, sε+[u]=εAε[W+(e(u))]s_{\varepsilon}^{+}[\mathbf{u}]=\varepsilon\mathcal{A}_{\varepsilon}[W^{+}(\mathbf{e}(\mathbf{u}))] is the scaled nonlocal average of the tensile strain-energy density. Figure 1 illustrates the nonlocal regularization of an opening crack and the associated saturation potential.

Nonlocal regularization of fracture

Figure 1. Nonlocal regularization of fracture in Ω\Omega, with prescribed displacements on ΓD\Gamma_D and tractions on ΓN\Gamma_N: (a) a sharp opening crack; (b) diffuse tensile localization and averaging over Bε(x)B_{\varepsilon}(\mathbf{x}), with W−W^{-} retained locally; (c) the exponential saturation potential, with unit initial slope and plateau Gc/2G_c/2. The opening and localization widths are schematic; ε\varepsilon denotes the averaging radius.

Let Ω\Omega be a fixed Lipschitz reference domain, with disjoint Dirichlet and Neumann boundary parts ΓD\Gamma_D and ΓN\Gamma_N. For t∈[0,T]t\in[0,T], impose u=uˉ(t)\mathbf{u}=\bar{\mathbf{u}}(t) on ΓD\Gamma_D and define

Πεsplit[u;t]=Fεsplit[u]−ℓext(t)[u],(12)\Pi_{\varepsilon}^{\mathrm{split}}[\mathbf{u};t]=\mathcal{F}_{\varepsilon}^{\mathrm{split}}[\mathbf{u}]-\ell_{\mathrm{ext}}(t)[\mathbf{u}], \tag*{(12)}

where the external work functional associated with the prescribed body force per unit reference volume b\mathbf{b} and traction tˉ\bar{\mathbf{t}} is

ℓext(t)[u]=∫Ωb(x,t)⋅u dx+∫ΓNtˉ(x,t)⋅u da.(13)\ell_{\mathrm{ext}}(t)[\mathbf{u}] = \int_{\Omega} \mathbf{b}(\mathbf{x},t)\cdot\mathbf{u}\,\mathrm{d}x + \int_{\Gamma_N} \bar{\mathbf{t}}(\mathbf{x},t)\cdot\mathbf{u}\,\mathrm{d}a. \tag*{(13)}

Energy variations

The constitutive stress follows from the first variation of the stored-energy functional (11). To identify the stress conjugate to the strain variation, we first introduce the adjoint of the nonlocal averaging operator. The integral kernel of (8) is

Kε(x,y)=1{∣x−y∣<ε}mε(x).(14)K_{\varepsilon}(\mathbf{x},\mathbf{y})=\frac{\mathbf{1}_{\{|\mathbf{x}-\mathbf{y}|<\varepsilon\}}}{m_{\varepsilon}(\mathbf{x})}. \tag*{(14)}

Accordingly, the adjoint of the averaging operator is

(Aε∗g)(y)=∫Bε(y)∩Ωg(x)mε(x) dx.(15)(\mathcal{A}_{\varepsilon}^{*}g)(\mathbf{y})=\int_{B_{\varepsilon}(\mathbf{y})\cap\Omega}\frac{g(\mathbf{x})}{m_{\varepsilon}(\mathbf{x})}\,\mathrm{d}x. \tag*{(15)}

In general, Aε∗≠Aε\mathcal{A}_{\varepsilon}^{*}\ne\mathcal{A}_{\varepsilon} because the normalization depends on the centre x\mathbf{x} of the truncated ball.

Proposition 2.2 (First variation and constitutive stress). Let u\mathbf{u} be an admissible displacement and v\mathbf{v} an admissible variation, with u,v∈W1,p(Ω;Rd)\mathbf{u},\mathbf{v}\in W^{1,p}(\Omega;\mathbb{R}^{d}), p>1p>1, and let v=0\mathbf{v}=0 on ΓD\Gamma_D. Assume that W±:Rsymd×d→[0,∞)W^{\pm}:\mathbb{R}^{d\times d}_{\mathrm{sym}}\to[0,\infty) are C1C^{1} and

∣DEW±(E)∣≤CW(1+∣E∣p−1)for all E∈Rsymd×d,(16)\left|D_EW^{\pm}(\mathbf{E})\right|\le C_W\left(1+|\mathbf{E}|^{p-1}\right)\quad\text{for all }\mathbf{E}\in\mathbb{R}^{d\times d}_{\mathrm{sym}}, \tag*{(16)}

that Φ∈C1([0,∞))\Phi\in C^{1}([0,\infty)) has bounded derivative, and that ℓext\ell_{\mathrm{ext}} is differentiable at u\mathbf{u}. Define

σ±(u)=DEW±(e(u)),qε+[u](x)=Φ′(sε+[u](x)),(17)\boldsymbol{\sigma}^{\pm}(\mathbf{u})=D_EW^{\pm}(\mathbf{e}(\mathbf{u})),\qquad q_{\varepsilon}^{+}[\mathbf{u}](\mathbf{x})=\Phi'\left(s_{\varepsilon}^{+}[\mathbf{u}](\mathbf{x})\right), \tag*{(17)}

and

aε+[u](y)=(Aε∗qε+[u])(y).(18)a_{\varepsilon}^{+}[\mathbf{u}](\mathbf{y})=\left(\mathcal{A}_{\varepsilon}^{*}q_{\varepsilon}^{+}[\mathbf{u}]\right)(\mathbf{y}). \tag*{(18)}

Then

DΠεsplit[u;t](v)=∫Ω(σ−(u)+aε+[u]σ+(u)):e(v) dy−ℓext(t)[v],(19)D\Pi_{\varepsilon}^{\mathrm{split}}[\mathbf{u};t](\mathbf{v})=\int_{\Omega}\left(\boldsymbol{\sigma}^{-}(\mathbf{u})+a_{\varepsilon}^{+}[\mathbf{u}]\boldsymbol{\sigma}^{+}(\mathbf{u})\right):\mathbf{e}(\mathbf{v})\,\mathrm{d}y-\ell_{\mathrm{ext}}(t)[\mathbf{v}], \tag*{(19)}

and the variationally consistent stress is

σεsplit=σ−+aε+σ+,σ±=λ⟨tr⁡e⟩±I+2μe±(20)\boldsymbol{\sigma}_{\varepsilon}^{\mathrm{split}}=\boldsymbol{\sigma}^{-}+a_{\varepsilon}^{+}\boldsymbol{\sigma}^{+},\qquad\boldsymbol{\sigma}^{\pm}=\lambda\langle\operatorname{tr}\mathbf{e}\rangle_{\pm}\mathbf{I}+2\mu\mathbf{e}^{\pm} \tag*{(20)}

for (6).

Proof. Differentiating sε+[u]s_\varepsilon^{+}[\mathbf{u}] in the direction v\mathbf{v} gives

Dsε+[u](v)=εAε[σ+(u):e(v)].(21)D s_\varepsilon^{+}[\mathbf{u}](\mathbf{v})=\varepsilon A_\varepsilon\left[\boldsymbol{\sigma}^{+}(\mathbf{u}):\mathbf{e}(\mathbf{v})\right]. \tag*{(21)}

Applying the chain rule and then the adjoint relation yields

DFεsplit[u](v)=∫Ωσ−:e(v) dx+∫Ωqε+[u]Aε[σ+(u):e(v)] dx(22)D F_\varepsilon^{\mathrm{split}}[\mathbf{u}](\mathbf{v})=\int_{\Omega}\boldsymbol{\sigma}^{-}:\mathbf{e}(\mathbf{v})\,\mathrm{d}x+\int_{\Omega}q_\varepsilon^{+}[\mathbf{u}]A_\varepsilon\left[\boldsymbol{\sigma}^{+}(\mathbf{u}):\mathbf{e}(\mathbf{v})\right]\,\mathrm{d}x \tag*{(22)}
=∫Ω(σ−+(Aε∗qε+)σ+):e(v) dy,(23)=\int_{\Omega}\left(\boldsymbol{\sigma}^{-}+(A_\varepsilon^{*}q_\varepsilon^{+})\boldsymbol{\sigma}^{+}\right):\mathbf{e}(\mathbf{v})\,\mathrm{d}y, \tag*{(23)}

which proves (19).

For the exponential potential, we have

qε+=exp⁡(−2sε+Gc),aε+=Aε∗[exp⁡(−2εGcAεW+(e(u)))].(24)q_\varepsilon^{+}=\exp\left(-\frac{2s_\varepsilon^{+}}{G_c}\right),\qquad a_\varepsilon^{+}=A_\varepsilon^{*}\left[\exp\left(-\frac{2\varepsilon}{G_c}A_\varepsilon W^{+}(\mathbf{e}(\mathbf{u}))\right)\right]. \tag*{(24)}

Elastodynamic equations and energy balance

Let ρ(x)>0\rho(\mathbf{x})>0 denote the reference mass density. The kinetic energy and action functional are defined by

T[u˙]=12∫Ωρ∣u˙∣2 dx,S[u]=∫0T(T[u˙]−Πεsplit[u;t]) dt.(25)\mathcal{T}[\dot{\mathbf{u}}]=\frac{1}{2}\int_{\Omega}\rho|\dot{\mathbf{u}}|^{2}\,\mathrm{d}x,\qquad S[\mathbf{u}]=\int_{0}^{T}\left(\mathcal{T}[\dot{\mathbf{u}}]-\Pi_\varepsilon^{\mathrm{split}}[\mathbf{u};t]\right)\,\mathrm{d}t. \tag*{(25)}

Stationarity of SS for variations vanishing on ΓD\Gamma_D and at the temporal endpoints, followed by integration by parts in time, gives

∫Ωρu¨⋅v dx+∫Ωσεsplit:e(v) dx=ℓext(t)[v]for all v=0 on ΓD.(26)\int_{\Omega}\rho\ddot{\mathbf{u}}\cdot\mathbf{v}\,\mathrm{d}x+\int_{\Omega}\boldsymbol{\sigma}_\varepsilon^{\mathrm{split}}:\mathbf{e}(\mathbf{v})\,\mathrm{d}x=\ell_{\mathrm{ext}}(t)[\mathbf{v}]\quad\text{for all }\mathbf{v}=0\text{ on }\Gamma_D. \tag*{(26)}

Remark 2.3 (Crack indicator). We use 1−qε+[u]1-q_\varepsilon^{+}[\mathbf{u}] as a crack indicator. For the exponential potential,

1−qε+[u]=1−exp⁡(−2sε+[u]Gc)∈[0,1).(27)1-q_\varepsilon^{+}[\mathbf{u}]=1-\exp\left(-\frac{2s_\varepsilon^{+}[\mathbf{u}]}{G_c}\right)\in[0,1). \tag*{(27)}

It vanishes at zero averaged tensile energy and approaches one in regions where qε+q_\varepsilon^{+} tends to zero. The indicator is computed from displacement and introduces no additional unknown.

For sufficiently regular solutions, (26) corresponds to the momentum balance

ρu¨=div⁡σεsplit+bin Ω×(0,T),(28)\rho\ddot{\mathbf{u}} = \operatorname{div}\boldsymbol{\sigma}_{\varepsilon}^{\mathrm{split}} + \mathbf{b} \quad\text{in } \Omega\times(0,T), \tag*{(28)}

subject to the prescribed displacement and traction data, with initial conditions u(⋅,0)=u0\mathbf{u}(\cdot,0)=\mathbf{u}_0 and u˙(⋅,0)=v0\dot{\mathbf{u}}(\cdot,0)=\mathbf{v}_0 compatible with the Dirichlet data.

Remark 2.4 (Time-dependent Dirichlet data). For sufficiently regular solutions with prescribed displacement u=uˉ(t)\mathbf{u}=\bar{\mathbf{u}}(t) on ΓD\Gamma_D, the total mechanical energy

E(t)=T[u˙(t)]+Fεsplit[u(t)](29)\mathcal{E}(t)=\mathcal{T}[\dot{\mathbf{u}}(t)]+\mathcal{F}_{\varepsilon}^{\mathrm{split}}[\mathbf{u}(t)] \tag*{(29)}

satisfies

dEdt=ℓext(t)[u˙]+PD(t),(30)\frac{\mathrm{d}\mathcal{E}}{\mathrm{d}t}=\ell_{\mathrm{ext}}(t)[\dot{\mathbf{u}}]+\mathcal{P}_D(t), \tag*{(30)}

where PD(t)=∫ΓD(σεsplitn)⋅uˉ˙ da\mathcal{P}_D(t)=\int_{\Gamma_D}(\boldsymbol{\sigma}_{\varepsilon}^{\mathrm{split}}\mathbf{n})\cdot\dot{\bar{\mathbf{u}}}\,\mathrm{d}a is the power of the reactions associated with the prescribed motion, and n\mathbf{n} is the outward unit normal.

If the Dirichlet data are time-independent, PD=0\mathcal{P}_D=0; then E\mathcal{E} is conserved in the absence of external power, and H=E−ℓext[u]\mathcal{H}=\mathcal{E}-\ell_{\mathrm{ext}}[\mathbf{u}] is conserved for a time-independent load functional. These conservation properties motivate the SAV time discretization in Section 4.

Sharp-crack consistency and emergent tensile stress degradation

The potential Φ\Phi determines the opening-crack energy and tensile stress degradation. We examine the elastic limit, planar displacement transitions and homogeneous tensile response of the split model.

Elastic limit and displacement jumps

For a fixed smooth displacement with bounded strain, Φ(s)=s+o(s)\Phi(s)=s+o(s) as s↓0s\downarrow0 gives

W−(e(u))+1εΦ(εAεW+(e(u)))⟶W(e(u))(ε↓0).(31)W^{-}(\mathbf{e}(\mathbf{u}))+\frac{1}{\varepsilon}\Phi\left(\varepsilon A_{\varepsilon}W^{+}(\mathbf{e}(\mathbf{u}))\right)\longrightarrow W(\mathbf{e}(\mathbf{u})) \quad(\varepsilon\downarrow0). \tag*{(31)}

Thus, the classical elastic energy density is recovered as ε→0\varepsilon\to0.

To examine displacement jumps, consider the planar profile uδ(x)=aζ((x⋅n)/δ)\mathbf{u}_{\delta}(\mathbf{x})=\mathbf{a}\zeta((\mathbf{x}\cdot\mathbf{n})/\delta). Here a\mathbf{a} is the displacement jump across the plane x⋅n=0\mathbf{x}\cdot\mathbf{n}=0, n\mathbf{n} is its unit normal, and δ>0\delta>0 is the transition thickness. The profile ζ\zeta is smooth and nondecreasing, with ζ(r)=0\zeta(r)=0 for r≤−1/2r\le-1/2 and ζ(r)=1\zeta(r)=1 for r≥1/2r\ge1/2. The matrix a⊙n=(a⊗n+n⊗a)/2\mathbf{a} \mathbin{\odot} \mathbf{n} = (\mathbf{a} \otimes\mathbf{n} + \mathbf{n} \otimes\mathbf{a})/2 has eigenvalues (a⋅n±∣a∣)/2(\mathbf{a} \cdot\mathbf{n} \pm|\mathbf{a}|)/2, with any remaining eigenvalues zero. Consequently,

W−(a⊙n)=λ2⟨a⋅n⟩−2+μ4(∣a∣−a⋅n)2,(32)W^{-}(\mathbf{a} \mathbin{\odot} \mathbf{n}) = \frac{\lambda}{2}\langle\mathbf{a} \cdot\mathbf{n}\rangle_{-}^{2} + \frac{\mu}{4}(|\mathbf{a}|-\mathbf{a}\cdot\mathbf{n})^{2}, \tag*{(32)}

and the compressive energy per unit interface area is

Eδ−=CζδW−(a⊙n),Cζ=∫R∣ζ′(r)∣2 dr.(33)\mathcal{E}_{\delta}^{-} = \frac{C_{\zeta}}{\delta}W^{-}(\mathbf{a} \mathbin{\odot} \mathbf{n}), \qquad C_{\zeta} = \int_{\mathbb{R}}|\zeta'(r)|^{2}\,\mathrm{d}r. \tag*{(33)}

For a fixed opening jump a=γn\mathbf{a}=\gamma\mathbf{n}, γ>0\gamma>0, (33) gives Eδ−=0\mathcal{E}_{\delta}^{-}=0, so only the nonlocal tensile term contributes. Let the transition thickness δ=δε\delta=\delta_{\varepsilon} satisfy δε/ε→0\delta_{\varepsilon}/\varepsilon\to0 as ε→0\varepsilon\to0. For an interior planar patch, we use the infinite planar extension and full averaging balls. With t=x⋅nt=\mathbf{x}\cdot\mathbf{n} the normal coordinate, the energy per unit interface area is

Eε,δεopen:=1ε∫RΦ(εAε[W+(e(uδε))](tn)) dt\mathcal{E}_{\varepsilon,\delta_{\varepsilon}}^{\mathrm{open}} := \frac{1}{\varepsilon}\int_{\mathbb{R}}\Phi\left(\varepsilon\mathcal{A}_{\varepsilon}\left[W^{+}(\mathbf{e}(u_{\delta_{\varepsilon}}))\right](t\mathbf{n})\right)\,\mathrm{d}t
⟶2Φ(∞)=Gc.(34)\longrightarrow2\Phi(\infty)=G_{\mathrm{c}}. \tag*{(34)}

The limit follows because averaging produces a layer of thickness 2ε+o(ε)2\varepsilon+o(\varepsilon), within whose interior Φ\Phi approaches its upper limit. In contrast, tangential sliding or normal compression gives W−(a⊙n)>0W^{-}(\mathbf{a}\mathbin{\odot}\mathbf{n})>0, so Eδ−→∞\mathcal{E}_{\delta}^{-}\to\infty as δ↓0\delta\downarrow0. Appendix B gives these derivations.

Remark 3.1. Scilla and Solombrino [27] established a Griffith Γ\Gamma-limit for the unsplit reference energy

Fε[u]=1ε∫ΩΦ(εAε[W(e(u))]) dx.(35)\mathcal{F}_{\varepsilon}[\mathbf{u}] = \frac{1}{\varepsilon}\int_{\Omega}\Phi\left(\varepsilon\mathcal{A}_{\varepsilon}[W(\mathbf{e}(\mathbf{u}))]\right)\,\mathrm{d}\mathbf{x}. \tag*{(35)}

This result does not transfer directly to the split model: W+W^{+} is not coercive, and the local term W−W^{-} changes the energy of transitions involving sliding or compression. A full Γ\Gamma-convergence analysis of the split functional is beyond the scope of this work.

Emergent constitutive degradation and tensile strength

The constitutive stress in (20) determines the tensile stress coefficient through q(s)=Φ′(s)q(s)=\Phi'(s) and aε+=Aε∗[q(sε+)]a_{\varepsilon}^{+}=\mathcal{A}_{\varepsilon}^{*}[q(s_{\varepsilon}^{+})]; no independent degradation law is prescribed [26]. If Φ∈C1([0,∞))\Phi\in C^{1}([0,\infty)) is also concave and satisfies (9), then qq is nonnegative and nonincreasing, with

q(0)=1,lim⁡s→∞q(s)=0.(36)q(0)=1,\qquad\lim_{s\to\infty}q(s)=0. \tag*{(36)}

Moreover, Gc=2∫0∞q(r) drG_c = 2\int_0^\infty q(r)\,\mathrm{d}r. The potential therefore determines the opening-crack energy and, through the adjoint average of its derivative, the tensile stress degradation.

For a fixed averaging radius, we determine the tensile strength from a homogeneous scalar tensile mode.

Proposition 3.2 (Homogeneous scalar tensile response). Let W+(η)=W(η)=Eη2/2W^+(\eta) = W(\eta) = E\eta^2/2 and W−=0W^- = 0 for η≥0\eta\ge0, where E>0E > 0 is the stiffness of this mode. Let ε>0\varepsilon> 0 and Φ∈C1([0,∞))∩C2((0,∞))\Phi\in C^1([0,\infty)) \cap C^2((0,\infty)) satisfy (9). With q=Φ′q = \Phi', the stress conjugate to the homogeneous strain is

T(η)=Eηq(s),s=εEη22.(37)T(\eta) = E\eta q(s), \qquad s = \frac{\varepsilon E\eta^2}{2}. \tag*{(37)}

Any interior stationary point sc>0s_c > 0 satisfies

Φ′(sc)+2scΦ′′(sc)=0.(38)\Phi'(s_c) + 2s_c\Phi''(s_c) = 0. \tag*{(38)}

For a peak at scs_c, setting zc=sc/Gcz_c = s_c/G_c gives

ηc=2GczcεE,Tc=CΦEGcε,(39)\eta_c = \sqrt{\frac{2G_c z_c}{\varepsilon E}}, \qquad T_c = C_\Phi\sqrt{\frac{E G_c}{\varepsilon}}, \tag*{(39)}

where CΦ=2zc Φ′(Gczc)C_\Phi= \sqrt{2z_c}\,\Phi'(G_c z_c).

Proof. For a homogeneous field, AεW=W\mathcal{A}_\varepsilon W = W, so the energy per unit volume is ψε(η)=ε−1Φ(εEη2/2)\psi_\varepsilon(\eta) = \varepsilon^{-1}\Phi(\varepsilon E\eta^2/2). Differentiation gives (37) and

dTdη=E(Φ′(s)+2sΦ′′(s)).\frac{\mathrm{d}T}{\mathrm{d}\eta} = E\left(\Phi'(s) + 2s\Phi''(s)\right).

Setting this derivative to zero gives (38); using sc=εEηc2/2s_c = \varepsilon E\eta_c^2/2 then gives (39). □

For the exponential potential (10), the peak occurs at sc=Gc/4s_c = G_c/4, giving

ηc=Gc2εE,Tc=EGc2ε.(40)\eta_c = \sqrt{\frac{G_c}{2\varepsilon E}}, \qquad T_c = \sqrt{\frac{E G_c}{2\varepsilon}}. \tag*{(40)}

The coefficients for other potentials are listed in Table A.5. The scaling Tc∝EGc/εT_c \propto\sqrt{E G_c/\varepsilon} shows that the averaging radius also sets the homogeneous strength scale.

Structure-preserving time discretization

We apply the SAV method to the entire stored-energy functional (11), including its tensile and compressive contributions. The formulation uses the exact variational derivative of this functional and introduces a single scalar auxiliary variable.

Equivalent SAV formulation

Let V={w∈H1(Ω;Rd):w=0 on ΓD}V = \{\mathbf{w} \in H^1(\Omega;\mathbb{R}^d) : \mathbf{w} = \mathbf{0}\text{ on }\Gamma_D\} and define

m(v,w)=∫Ωρ v⋅w dx,∥v∥ρ2=m(v,v).(41)m(\mathbf{v},\mathbf{w}) = \int_{\Omega} \rho\,\mathbf{v}\cdot\mathbf{w}\,\mathrm{d}x,\qquad\|\mathbf{v}\|_{\rho}^{2} = m(\mathbf{v},\mathbf{v}). \tag*{(41)}

We first consider time-independent Dirichlet data; prescribed motion is addressed below. The internal-force functional is defined by the first variation of the stored energy:

⟨Gε(u),w⟩:=DFεsplit[u](w)=∫Ω(σ−(u)+aε+[u]σ+(u)):e(w) dx.(42)\begin{aligned} \langle\mathcal{G}_{\varepsilon}(\mathbf{u}),\mathbf{w}\rangle:= D\mathcal{F}_{\varepsilon}^{\mathrm{split}}[\mathbf{u}](\mathbf{w}) \\ &= \int_{\Omega}\left(\boldsymbol{\sigma}^{-}(\mathbf{u})+a_{\varepsilon}^{+}[\mathbf{u}]\boldsymbol{\sigma}^{+}(\mathbf{u})\right):\mathbf{e}(\mathbf{w})\,\mathrm{d}x. \tag*{(42)} \end{aligned}

Thus (26) becomes

u˙=v,m(v˙,w)+⟨Gε(u),w⟩=ℓext(t)[w],w∈V.(43)\dot{\mathbf{u}}=\mathbf{v},\qquad m(\dot{\mathbf{v}},\mathbf{w})+\langle\mathcal{G}_{\varepsilon}(\mathbf{u}),\mathbf{w}\rangle=\ell_{\mathrm{ext}}(t)[\mathbf{w}],\qquad\mathbf{w}\in V. \tag*{(43)}

Following the SAV construction [30, 31, 32], set

A(u)=Fεsplit[u]+C0,r(t)=A(u(t)),C0>0.(44)A(\mathbf{u})=\mathcal{F}_{\varepsilon}^{\mathrm{split}}[\mathbf{u}]+C_{0},\qquad r(t)=\sqrt{A(\mathbf{u}(t))},\qquad C_{0}>0. \tag*{(44)}

Here C0C_{0} has units of energy, and A≥C0A\ge C_{0} because Fεsplit≥0\mathcal{F}_{\varepsilon}^{\mathrm{split}}\ge0. The chain rule gives the equivalent system

u˙=v,m(v˙,w)+rA(u)⟨Gε(u),w⟩=ℓext(t)[w],w∈V,r˙=12A(u)⟨Gε(u),v⟩.(45)\begin{aligned} \dot{\mathbf{u}}&=\mathbf{v},\\ m(\dot{\mathbf{v}},\mathbf{w})+\frac{r}{\sqrt{A(\mathbf{u})}}\langle\mathcal{G}_{\varepsilon}(\mathbf{u}),\mathbf{w}\rangle&=\ell_{\mathrm{ext}}(t)[\mathbf{w}],\qquad\mathbf{w}\in V,\\ \dot{r}&=\frac{1}{2\sqrt{A(\mathbf{u})}}\langle\mathcal{G}_{\varepsilon}(\mathbf{u}),\mathbf{v}\rangle. \tag*{(45)} \end{aligned}

With r(0)=A(u0)r(0)=\sqrt{A(\mathbf{u}_{0})}, the last equation implies d(r−A(u))/dt=0\mathrm{d}(r-\sqrt{A(\mathbf{u})})/\mathrm{d}t=0. Hence r/A(u)=1r/\sqrt{A(\mathbf{u})}=1 at the continuous level. Define

E~(t)=12∥v(t)∥ρ2+r(t)2−C0.(46)\widetilde{\mathcal{E}}(t)=\frac{1}{2}\|\mathbf{v}(t)\|_{\rho}^{2}+r(t)^{2}-C_{0}. \tag*{(46)}

Testing the momentum equation with v\mathbf{v} gives dE~dt=ℓext(t)[v]\frac{\mathrm{d}\widetilde{\mathcal{E}}}{\mathrm{d}t}=\ell_{\mathrm{ext}}(t)[\mathbf{v}]. For compatible initial data, the modified energy E~\widetilde{\mathcal{E}} equals the physical energy E\mathcal{E} in the continuous formulation. This equality is generally not preserved by SAV time discretization [33].

Structure-preserving SAV scheme

In the discrete formulas below, AA denotes Ah(uh)=Fε,hsplit[uh]+C0A_h(\mathbf{u}_h)=\mathcal{F}^{\mathrm{split}}_{\varepsilon,h}[\mathbf{u}_h]+C_0, and Gε\mathcal{G}_{\varepsilon} denotes its exact variational derivative. Let τ>0\tau>0, tn=nτt_n=n\tau, and write

Dtzn+1=zn+1−znτ,δzn=zn+1−zn,zn+1/2=zn+1+zn2.(47)D_tz^{n+1}=\frac{z^{n+1}-z^n}{\tau},\qquad\delta z^n=z^{n+1}-z^n,\qquad z^{n+1/2}=\frac{z^{n+1}+z^n}{2}. \tag*{(47)}

For the time-independent Dirichlet data considered here, the displacement belongs to the affine space with prescribed trace, and the velocity and test functions have zero Dirichlet trace. Prescribed motion is treated in Remark 4.2. Initialize r0=A(u0)r^0=\sqrt{A(\mathbf{u}^0)} and define

E~n=12∥vn∥ρ2+(rn)2−C0.(48)\widetilde{\mathcal{E}}^n=\frac{1}{2}\lVert\mathbf{v}^n\rVert_{\rho}^{2}+(r^n)^2-C_0. \tag*{(48)}

For n≥1n\geq1, extrapolate the displacement to the midpoint u∗n+1/2=32un−12un−1\mathbf{u}_*^{n+1/2}=\frac{3}{2}\mathbf{u}^n-\frac{1}{2}\mathbf{u}^{n-1}, and define

A∗n+1/2=A(u∗n+1/2),Gε,∗n+1/2=Gε(u∗n+1/2).(49)A_*^{n+1/2}=A(\mathbf{u}_*^{n+1/2}),\qquad\mathcal{G}_{\varepsilon,*}^{n+1/2}=\mathcal{G}_{\varepsilon}(\mathbf{u}_*^{n+1/2}). \tag*{(49)}

With

ξn+1/2=rn+1/2A∗n+1/2,ℓextn+1/2=ℓext(tn+τ/2),(50)\xi^{n+1/2}=\frac{r^{n+1/2}}{\sqrt{A_*^{n+1/2}}},\qquad\ell_{\mathrm{ext}}^{n+1/2}=\ell_{\mathrm{ext}}(t_n+\tau/2), \tag*{(50)}

the update is

Dtun+1=vn+1/2,m(Dtvn+1,w)+ξn+1/2⟨Gε,∗n+1/2,w⟩=ℓextn+1/2[w],Dtrn+1=12A∗n+1/2⟨Gε,∗n+1/2,Dtun+1⟩.(51)\begin{aligned} D_t\mathbf{u}^{n+1}&=\mathbf{v}^{n+1/2},\\ m(D_t\mathbf{v}^{n+1},\mathbf{w})+\xi^{n+1/2}\langle\mathcal{G}_{\varepsilon,*}^{n+1/2},\mathbf{w}\rangle&=\ell_{\mathrm{ext}}^{n+1/2}[\mathbf{w}],\\ D_tr^{n+1}&=\frac{1}{2\sqrt{A_*^{n+1/2}}}\langle\mathcal{G}_{\varepsilon,*}^{n+1/2},D_t\mathbf{u}^{n+1}\rangle. \tag*{(51)} \end{aligned}

For the first step, use

u∗1/2=u0+τ2v0.(52)\mathbf{u}_*^{1/2}=\mathbf{u}^0+\frac{\tau}{2}\mathbf{v}^0. \tag*{(52)}

Theorem 4.1 (Discrete modified-energy balance). For time-independent Dirichlet data on a fixed spatial discretization, (51) satisfies, for every τ>0\tau> 0.

E~n+1−E~n=ℓextn+1/2[δun].(53)\widetilde{\mathcal{E}}^{n+1}-\widetilde{\mathcal{E}}^n=\ell_{\mathrm{ext}}^{n+1/2}[\delta\mathbf{u}^n]. \tag*{(53)}

Thus E~\widetilde{\mathcal{E}} is conserved for zero loading, and H~n+1=H~n\widetilde{\mathcal{H}}^{n+1}=\widetilde{\mathcal{H}}^n for a time-independent load functional, where H~n=E~n−ℓext[un]\widetilde{\mathcal{H}}^n=\widetilde{\mathcal{E}}^n-\ell_{\mathrm{ext}}[\mathbf{u}^n].

Proof. Test the momentum equation with τvn+1/2=δun\tau\mathbf{v}^{n+1/2}=\delta\mathbf{u}^n and multiply the scalar equation by 2τrn+1/22\tau r^{n+1/2}. Apply 2(a−b,(a+b)/2)=∥a∥2−∥b∥22(a-b,(a+b)/2)=\lVert a\rVert^2-\lVert b\rVert^2 to obtain (53).

Remark 4.2 (Discrete work of prescribed motion). For time-dependent Dirichlet data, the discrete power supplied by the displacement constraints is

PD,hn+1/2=(RDn+1/2)TVDn+1/2,(54)\mathcal{P}_{D,h}^{n+1/2}=(\mathbf{R}_D^{n+1/2})^{\mathsf{T}}\mathbf{V}_D^{n+1/2}, \tag*{(54)}

where VDn+1/2=δU‾Dn/τ\mathbf{V}_D^{n+1/2}=\delta\overline{\mathbf{U}}_D^n/\tau contains the prescribed boundary velocities and RDn+1/2\mathbf{R}_D^{n+1/2} is the reaction exerted by the displacement constraint on the body:

RDn+1/2=[MallVn+1−Vnτ+ξn+1/2∇Fε,hsplit(U∗n+1/2)−fextn+1/2]D.(55)\mathbf{R}_D^{n+1/2}=\left[\mathbf{M}_{\mathrm{all}}\frac{\mathbf{V}^{n+1}-\mathbf{V}^n}{\tau}+\xi^{n+1/2}\nabla\mathcal{F}_{\varepsilon,h}^{\mathrm{split}}\left(\mathbf{U}_*^{n+1/2}\right)-\mathbf{f}_{\mathrm{ext}}^{n+1/2}\right]_D. \tag*{(55)}

Here [⋅]D[\cdot]_D selects the prescribed degrees of freedom. Mall\mathbf{M}_{\mathrm{all}} is the full consistent mass matrix, and U\mathbf{U}, V\mathbf{V} and fext\mathbf{f}_{\mathrm{ext}} contain the components associated with both free and prescribed degrees of freedom. The exact discrete balance for the modified energy becomes

E~n+1−E~n=ℓextn+1/2[δun]+τPD,hn+1/2.(56)\widetilde{\mathcal{E}}^{n+1}-\widetilde{\mathcal{E}}^n=\ell_{\mathrm{ext}}^{n+1/2}[\delta\mathbf{u}^n]+\tau\mathcal{P}_{D,h}^{n+1/2}. \tag*{(56)}

This is the discrete counterpart of (30). For fixed Dirichlet data, PD,hn+1/2=0\mathcal{P}_{D,h}^{n+1/2}=0, recovering Theorem 4.1.

With fixed Dirichlet data, the midpoint velocity system on the free degrees of freedom is

(M+τ28b∗b∗T)Vn+1/2=MVn+τ2(fextn+1/2−rnb∗),(57)\left(\mathbf{M}+\frac{\tau^2}{8}\mathbf{b}_*\mathbf{b}_*^{\mathsf{T}}\right)\mathbf{V}^{n+1/2} =\mathbf{M}\mathbf{V}^n+\frac{\tau}{2}\left(\mathbf{f}_{\mathrm{ext}}^{n+1/2}-r^n\mathbf{b}_*\right), \tag*{(57)}
b∗=∇Fε,hsplit(U∗n+1/2)A∗n+1/2,\mathbf{b}_*=\frac{\nabla\mathcal{F}_{\varepsilon,h}^{\mathrm{split}}\left(\mathbf{U}_*^{n+1/2}\right)}{\sqrt{A_*^{n+1/2}}},

where M\mathbf{M} is the mass matrix restricted to the free degrees of freedom, U\mathbf{U} and V\mathbf{V} are the corresponding coefficient vectors, and the gradient is restricted to the same degrees of freedom. The new state follows from

Un+1=Un+τVn+1/2,Vn+1=2Vn+1/2−Vn,rn+1=rn+τ2b∗TVn+1/2.(58)\begin{aligned} \mathbf{U}^{n+1} &= \mathbf{U}^{n}+\tau\mathbf{V}^{n+1/2},\\ \mathbf{V}^{n+1} &= 2\mathbf{V}^{n+1/2}-\mathbf{V}^{n},\\ r^{n+1} &= r^{n}+\frac{\tau}{2}\mathbf{b}_{*}^{\mathsf{T}}\mathbf{V}^{n+1/2}. \tag*{(58)} \end{aligned}

Remark 4.3 (Modified and physical energies). The exact balance (53) concerns the modified energy E~n\widetilde{\mathcal{E}}^{n}. The physical discrete energy and the difference between the two energies are

Ehn=12∥vn∥ρ2+Fε,hsplit[un],ΔSAVn=E~n−Ehn=(rn)2−A(un).(59)\begin{aligned} \mathcal{E}_{h}^{n} &= \frac{1}{2}\lVert\mathbf{v}^{n}\rVert_{\rho}^{2}+\mathcal{F}_{\varepsilon,h}^{\mathrm{split}}[\mathbf{u}^{n}],\\ \Delta_{\mathrm{SAV}}^{n} &= \widetilde{\mathcal{E}}^{n}-\mathcal{E}_{h}^{n}=(r^{n})^{2}-A(\mathbf{u}^{n}). \tag*{(59)} \end{aligned}

Although initialization gives ΔSAV0=0\Delta_{\mathrm{SAV}}^{0}=0, the scalar rnr^{n} is advanced by its own discrete equation, which does not enforce (rn)2=A(un)(r^{n})^{2}=A(\mathbf{u}^{n}) at subsequent steps. Hence this SAV consistency defect need not remain zero. Under the assumptions of Theorem 4.1, the discrete physical energy satisfies

Ehn+1−Ehn=ℓextn+1/2[δun]−(ΔSAVn+1−ΔSAVn).(60)\mathcal{E}_{h}^{n+1}-\mathcal{E}_{h}^{n}=\ell_{\mathrm{ext}}^{n+1/2}[\delta\mathbf{u}^{n}]-(\Delta_{\mathrm{SAV}}^{n+1}-\Delta_{\mathrm{SAV}}^{n}). \tag*{(60)}

Thus the change in ΔSAVn\Delta_{\mathrm{SAV}}^{n} quantifies the discrepancy between the physical-energy increment and the external work.

Numerical validation

We assess the crack-path predictions and propagation dynamics of the model, together with the SAV scheme of § 4, using two benchmarks: dynamic crack branching and Kalthoff–Winkler impact. For both benchmarks, the explicit velocity-Verlet algorithm [34], in kick–drift–kick form, serves as the reference integrator. Table 1 summarizes the material parameters and discretization settings.

QuantityBranchingKalthoff–Winkler
EE (GPa)32190
ν\nu0.200.30
GcG_c (N/mm)0.00322.13
Thickness tpt_p (mm)11
Density ρ\rho (kg/m3^3)24508000
KinematicsPlane strainPlane strain
ElementsAdaptive Q1Q_1Adaptive Q1Q_1
Coarse-mesh size h0h_0 (mm)0.667, 0.444, 0.2221.5, 1, 0.5
Nonlocal radiusεK=5hK\varepsilon_K = 5h_KεK=3hK\varepsilon_K = 3h_K
IntegratorsVelocity-Verlet, SAVVelocity-Verlet, SAV

Table 1. Material parameters, spatial discretizations and time integrators for the two dynamic benchmarks.

Both integrators use adaptive meshes of bilinear quadrilateral (Q1Q_1) elements [35], with a consistent mass matrix for SAV and a lumped mass matrix for velocity-Verlet. The coarsest element size is denoted by h0h_0; two refinement levels yield hmin⁡=h0/4h_{\min}=h_0/4. The nonlocal averaging radius in each element KK is set to εK=MhK\varepsilon_K=Mh_K.

Dynamic crack branching

We simulate dynamic crack branching using the geometry, material data and loading of Borden et al. [9]. With the origin at the lower-left corner, the plate and initial slit in Figure 2 are

Dynamic branching specimen and step loading

Figure 2. Dynamic branching specimen and step loading.

ΩB=((0,100)×(0,40))∖Γ0,B,Γ0,B=[0,50]×{20},(61)\begin{aligned} \Omega_B &= \left((0,100) \times(0,40)\right) \setminus\Gamma_{0,B},\\ \Gamma_{0,B} &= [0,50] \times\{20\}, \tag*{(61)} \end{aligned}

with dimensions in millimetres. At t=0+t = 0^{+}, equal and opposite 1 MPa tensile tractions are applied to the horizontal edges and held constant; the remaining boundaries are traction-free. Each simulation runs to 80 μs80\,\mu\mathrm{s}.

Figure 3 compares the crack patterns when the connected crack region first reaches the external boundary. Both integrators consistently reproduce an initially straight extension followed by two approximately symmetric branches at all three mesh sizes. The dominant branching pattern is retained as h0h_0 decreases, while the indicator band narrows. The hydrostatic stress fields near 70 μs70\,\mu\mathrm{s} (Figure 4) show tensile stress concentrated near the advancing branch tips.

Figure 3. Dynamic branching when the connected crack region 1−q+≥0.951-q_{+} \ge0.95 first reaches the external boundary. (a,c,e) Velocity-Verlet; (b,d,f) SAV. The initial mesh size h0h_{0} is indicated beside each row. Height and colour both represent 1−q+∈[0,1]1-q_{+} \in[0,1] over the reference plane.

Figure 4. Hydrostatic stress σhyd=(σxx+σyy+σzz)/3\sigma_{\mathrm{hyd}} = (\sigma_{xx} + \sigma_{yy} + \sigma_{zz})/3 near 70 μ\mus. The actual saved times are indicated above each panel. (a,c,e) Velocity-Verlet; (b,d,f) SAV. The initial mesh size h0h_0 is indicated beside each row. The in-plane display coordinates are X+50uX + 50u. Height and colour represent the signed stress, with a common height scale across the six panels. Dashed lines show the initial boundary and notch in the zero-stress plane. Elements with crack indicator 1−q+=11 - q_{+} = 1 at all four nodes are hidden. The common colour scale is [−2,2][-2, 2] MPa. The ranges of nodal stress values, from top to bottom, are [−3.95,2.60][-3.95, 2.60], [−5.20,3.33][-5.20, 3.33] and [−8.39,4.78][-8.39, 4.78] MPa in (a,c,e), and [−3.74,2.66][-3.74, 2.66], [−5.01,3.50][-5.01, 3.50] and [−9.83,4.69][-9.83, 4.69] MPa in (b,d,f). Values outside the colour scale use its endpoint colours.

Across the three mesh sizes, the two integrators differ by at most 0.59 μ\mus in branching time and 1.2° in initial branch angle (Table 2). SAV predicts branch-angle magnitudes of approximately 28°–32°. The table also lists the first resolved extension time, tdett_{\mathrm{det}}. These measurements use the common threshold and procedure in Appendix C.

h0h_0 (mm)Integratortdett_{\mathrm{det}} (μ\mus)tbrt_{\mathrm{br}} (μ\mus)θupper\theta_{\mathrm{upper}} (°)θlower\theta_{\mathrm{lower}} (°)
2/3Velocity-Verlet19.10337.34027.34−27.34
SAV16.57936.80028.03−27.97
4/9Velocity-Verlet15.56636.53828.77−28.77
SAV17.50036.00029.53−29.95
2/9Velocity-Verlet15.14135.69830.36−30.84
SAV15.14335.11030.06−31.74

Table 2. Dynamic branching: onset, branching times and early branch angles.

Figure 5 shows that the two integrators also give comparable crack-tip speeds during propagation. The parent-crack tip speeds obtained with velocity-Verlet reach approximately 1.0–1.2 km/s. The SAV upper-branch speeds are approximately 0.96–1.10 km/s over 50–60 μ\mus, close to the corresponding velocity-Verlet speeds, and then decrease as the tips approach the outer boundary.

Figure 5. Dynamic branching: apparent parent-tip and upper-branch speeds for velocity-Verlet and SAV at (a) h0=2/3h_0 = 2/3 mm, (b) h0=4/9h_0 = 4/9 mm, and (c) h0=2/9h_0 = 2/9 mm.

For both integrators and all three mesh sizes, the energy increments closely follow the external work through branching (Figure 6). Figure 7 reports the SAV consistency defect; the stepwise modified-energy balance does not require this defect to vanish.

Figure 6. Dynamic branching: physical-energy increments Ehn\mathcal{E}_h^n and external work for velocity-Verlet and SAV at (a) h0=2/3h_0 = 2/3 mm, (b) h0=4/9h_0 = 4/9 mm, and (c) h0=2/9h_0 = 2/9 mm.

Figure 7. SAV consistency defect ΔSAVn\Delta_{\mathrm{SAV}}^n in dynamic branching. The inset enlarges the h0=2/9h_0 = 2/9 mm curve. The maximum absolute defects max⁡n∣ΔSAVn∣\max_n |\Delta_{\mathrm{SAV}}^n| are 3.28×10−63.28 \times10^{-6}, 8.35×10−68.35 \times10^{-6} and 1.13×10−71.13 \times10^{-7} J for h0=2/3h_0 = 2/3, 4/94/9 and 2/92/9 mm, respectively.

Kalthoff–Winkler impact

This benchmark examines brittle crack kinking under dynamic shear loading [36]. The full plate is modelled, with the loading idealization of Borden et al. [9]. With the origin at the lower-left corner, xx horizontal and yy vertical, the geometry in Figure 8 is

Kalthoff–Winkler geometry and loading

Figure 8. Kalthoff–Winkler geometry and loading. The velocity in (63) is prescribed on the upper edge between the initial slits; all other boundaries are traction-free.

ΩKW=((0,200)×(0,100))∖(Γ0,l∪Γ0,r),Γ0,l={75}×[50,100],Γ0,r={125}×[50,100],(62)\Omega_{\mathrm{KW}} = \left((0, 200) \times(0, 100)\right) \setminus\left(\Gamma_{0,\mathrm{l}} \cup\Gamma_{0,\mathrm{r}}\right), \qquad \Gamma_{0,\mathrm{l}} = \{75\} \times[50, 100], \quad\Gamma_{0,\mathrm{r}} = \{125\} \times[50, 100], \tag*{(62)}

with dimensions in millimetres. A downward velocity is prescribed on the upper edge of the central region between the initial slits, Γv=[75,125]×{100}\Gamma_{\mathrm{v}} = [75, 125] \times\{100\}:

U˙(t)=−u˙y(t)=v0{t/tr,0≤t≤tr,1,t>tr,v0=16.5 m/s,tr=1 μs.(63)\dot{U}(t) = -\dot{u}_y(t) = v_0 \begin{cases} t/t_r, & 0 \le t \le t_r,\\ 1, & t > t_r, \end{cases} \qquad v_0 = 16.5\ \mathrm{m/s}, \qquad t_r = 1\ \mu\mathrm{s}. \tag*{(63)}

Here U(0)=0U(0)=0. The horizontal displacement on Γv\Gamma_{\mathrm{v}} is unconstrained; all remaining boundaries are traction-free. Loading is applied directly to the specimen, without modelling projectile contact.

Figure 9 shows that both integrators reproduce two approximately symmetric outward-kinking cracks across all three initial mesh sizes. This dominant pattern persists as h0h_0 decreases, while the indicator bands narrow. The coarsest cases also show additional localization at the opposite boundary, y=0y = 0. Figure 10 compares the hydrostatic stress fields near 75 μ\mus. All time-history comparisons below end at 83 μ\mus, when the cracks are near the outer boundaries.

Figure 9. Kalthoff–Winkler impact when the connected crack region 1−q+≥0.951 - q_{+} \ge0.95 reaches the external boundary. (a,c,e) Velocity-Verlet; (b,d,f) SAV. The initial mesh size h0h_0 is indicated beside each row. Height and colour both represent 1−q+∈[0,1]1 - q_{+} \in[0,1] over the reference plane; height is a visualization coordinate.

Figure 10. Hydrostatic stress σhyd=(σxx+σyy+σzz)/3\sigma_{\mathrm{hyd}} = (\sigma_{xx} + \sigma_{yy} + \sigma_{zz})/3 near 75 μ\mus. The actual saved times are indicated above each panel. (a,c,e) Velocity-Verlet; (b,d,f) SAV. The initial mesh size h0h_0 is indicated beside each row. The in-plane display coordinates are X+5uX + 5u. Height and colour represent the signed stress, with a common height scale across the six panels. Dashed lines show the initial geometry in the zero-stress plane. Fully saturated cells are hidden. The common scale is [−600, 600] MPa; nodal ranges from top to bottom are [−2045, 545], [−2339, 583] and [−3284, 968] MPa in (a,c,e), and [−2010, 604], [−2350, 660] and [−3272, 1125] MPa in (b,d,f). Out-of-range values use endpoint colours.

h0h_0 (mm)Integratortdett_{\mathrm{det}} (μ\mus)θl\theta_l (°)θr\theta_r (°)
1.5Velocity-Verlet24.241−74.89-74.89+74.89+74.89
SAV23.997−74.85-74.85+74.85+74.85
1Velocity-Verlet24.522−72.52-72.52+72.52+72.52
SAV24.301−73.19-73.19+73.19+73.19
0.5Velocity-Verlet23.998−72.65-72.65+72.65+72.65
SAV23.925−73.45-73.45+73.45+73.45

Table 3. Kalthoff–Winkler: onset and kink angles.

Table 3 quantifies the close agreement: at each resolution, the two integrators differ by less than 0.25 μ\mus in the first resolved extension time and at most 0.8° in kink-angle magnitude. SAV predicts extension near 24 μ\mus and angle magnitudes of 73.19°–74.85°, close to the approximately 70° angle reported for this benchmark [9]. Negative and positive signs denote clockwise and counterclockwise kinking from the downward slit direction. Both integrators use the same measurement procedure (Appendix C).

Figure 11 shows comparable apparent right-tip speed histories for velocity-Verlet and SAV at each initial mesh size. Both integrators capture acceleration to approximately 1.6–1.8 km/s over 35–50 μ\mus, followed by a decrease as the cracks approach the outer boundaries. Both integrators give similar histories of energy increments and external work over the 83 μ\mus comparison interval (Figure 12).

Figure 11. Kalthoff–Winkler: apparent right-tip speed up to 83 μ\mus for velocity-Verlet and SAV at (a) h0=1.5h_0 = 1.5 mm, (b) h0=1h_0 = 1 mm, and (c) h0=0.5h_0 = 0.5 mm.

Figure 12. Kalthoff–Winkler: physical-energy increments Ehn\mathcal{E}_h^n and external work for velocity-Verlet and SAV up to 83 μ\mus at (a) h0=1.5h_0 = 1.5 mm, (b) h0=1h_0 = 1 mm, and (c) h0=0.5h_0 = 0.5 mm.

We compare the model with tension–compression splitting against the unsplit model (35) at h0=0.5h_0 = 0.5 mm, using otherwise identical physical parameters and numerical settings.

Near 30 μ\mus, the split model reproduces outward kinking, whereas the unsplit model develops an inward localization band between the notch tips and markedly reduces the compressive stress in the impacted central region (Figure 13).

Figure 13. Kalthoff–Winkler, h0=0.5h_0 = 0.5 mm: fields computed with velocity-Verlet for the split and unsplit models near 30 μs on the reference geometry. (a,c) Split model; (b,d) unsplit model. (a,b) Saturation indicators 1−q+1 - q_+ and 1−q1 - q, respectively; (c,d) hydrostatic stress. All elements are shown. Stress ranges in (c,d) are [−3559, 949] and [−940, 1002] MPa, respectively; values outside [−600, 600] MPa use endpoint colours.

Figure 14 shows the stored energy and impact reaction Fimpn=∑a∈Dv(mau¨an+fint,an)F_{\mathrm{imp}}^n = \sum_{a \in D_v} (m_a \ddot{u}_a^n + f_{\mathrm{int},a}^n), where DvD_v denotes the prescribed impact degrees of freedom, mam_a the lumped masses, and fint,af_{\mathrm{int},a} the internal forces, with components taken positive in the impact direction. Both the stored energy and impact reaction are substantially smaller in the unsplit model.

Figure 14. Kalthoff–Winkler, h0=0.5h_0 = 0.5 mm: time histories computed with velocity-Verlet for the split and unsplit models. (a) Impact reaction FimpF_{\mathrm{imp}}; (b) stored energy Fε,h[unh]F_{\varepsilon,h}[u_n^{\mathrm{h}}].

The outward crack paths and retained compressive response support tension–compression splitting for physically plausible mixed-mode dynamic cracking.

Conclusions and outlook

We have developed a variational diffuse-crack model based on nonlocal averaging of tensile strain energy. A bounded potential acts on the scaled tensile-energy average, while the compressive energy remains local. The first variation of the energy determines tensile stress degradation through the adjoint averaging operator, without an independent fracture field or a separately prescribed degradation law. At a fixed averaging radius, the shape of the potential determines the homogeneous tensile strength.

The stored-energy functional recovers classical elasticity for fixed smooth displacement fields and the Griffith surface energy for the planar pure-opening profiles considered here.

The linearly implicit SAV scheme satisfies an exact discrete balance between the modified-energy increment and the external work. The two dynamic benchmarks illustrate crack branching and kinking. In the Kalthoff–Winkler benchmark, the unsplit model predicts inward localization between the notch tips and a smaller compressive reaction. The split model preserves the compressive elastic response by retaining the compressive energy locally. Further work will address Γ\Gamma-convergence of the split functional and extend the model to multiphysics fracture, fracture in elastoplastic and hyperelastic solids, and fracture of shells and rods.

Representative saturation potentials

The following saturation potentials satisfy Φ′(0)=1\Phi'(0)=1 and Φ(∞)=Gc/2\Phi(\infty)=G_c/2. Their derivatives decrease from unity to zero at rates determined by the form of the potential.

FamilyΦ(s)\Phi(s)q(s)=Φ′(s)q(s)=\Phi'(s)
ExponentialGc2(1−e−2s/Gc)\frac{G_c}{2}\left(1-e^{-2s/G_c}\right)e−2s/Gce^{-2s/G_c}
Rationals1+2s/Gc\frac{s}{1+2s/G_c}(1+2s/Gc)−2\left(1+2s/G_c\right)^{-2}
ArctangentGcπtan⁡−1(πsGc)\frac{G_c}{\pi}\tan^{-1}\left(\frac{\pi s}{G_c}\right)11+(πs/Gc)2\frac{1}{1+\left(\pi s/G_c\right)^2}
Hyperbolic tangentGc2tanh⁡(2sGc)\frac{G_c}{2}\tanh\left(\frac{2s}{G_c}\right)sech⁡2(2sGc)\operatorname{sech}^{2}\left(\frac{2s}{G_c}\right)

Table A.4. Saturation potentials satisfying (9).

Figure A.15 compares the approach of the potentials to their common saturation limit and the corresponding decay of qq.

Figure A.15. Comparison of (a) the normalized saturation potential 2Φ(s)/Gc2\Phi(s)/G_c and (b) the corresponding degradation function q(s)=Φ′(s)q(s)=\Phi'(s) for the four potentials in table A.4.

The dimensionless peak locations zc=sc/Gcz_c=s_c/G_c and strength coefficients CΦC_\Phi in (39), obtained from equation (38), are listed in table A.5.

Familyzc=sc/Gcz_c=s_c/G_cCΦC_\Phi
Exponential1/41/41/2e≈0.4288821/\sqrt{2e}\approx0.428882
Rational1/61/69/(163)≈0.3247609/(16\sqrt{3})\approx0.324760
Arctangent1/(π3)1/(\pi\sqrt{3})342/(π3)≈0.454696\frac{3}{4}\sqrt{2/(\pi\sqrt{3})}\approx0.454696
Hyperbolic tangent≈0.260907\approx0.2609072zc sech⁡2(2zc)≈0.556558\sqrt{2z_c}\,\operatorname{sech}^{2}(2z_c)\approx0.556558

Table A.5. Homogeneous uniaxial peak locations and strength coefficients in (39) for the potentials in table A.4. For the hyperbolic-tangent potential, zcz_c is the unique positive root of 8zctanh⁡(2zc)=18z_c\tanh(2z_c)=1.

Planar displacement transitions in the split model

For the profile introduced in § 3.1, set t=x⋅nt=\mathbf{x}\cdot\mathbf{n}. Differentiation gives

e(uδ)=ζ′(t/δ)δA,A=a⊙n:=a⊗n+n⊗a2.(64)\mathbf{e}(\mathbf{u}_\delta)=\frac{\zeta'(t/\delta)}{\delta}\mathbf{A},\qquad\mathbf{A}=\mathbf{a}\mathbin{\odot}\mathbf{n}:=\frac{\mathbf{a}\otimes\mathbf{n}+\mathbf{n}\otimes\mathbf{a}}{2}. \tag*{(64)}

Decompose a=ann+aτ\mathbf{a}=a_n\mathbf{n}+\mathbf{a}_{\tau}, where an=a⋅na_n=\mathbf{a}\cdot\mathbf{n} and aτ⋅n=0\mathbf{a}_{\tau}\cdot\mathbf{n}=0. In a basis aligned with n\mathbf{n} and aτ\mathbf{a}_{\tau}, the nonzero part of A\mathbf{A} lies in the block

A(2)=(an∣aτ∣/2∣aτ∣/20),det⁡(A(2)−ηI)=η2−anη−∣aτ∣24.(65)\mathbf{A}_{(2)}=\begin{pmatrix} a_n & |\mathbf{a}_{\tau}|/2 \\ |\mathbf{a}_{\tau}|/2 & 0 \end{pmatrix},\qquad\det(\mathbf{A}_{(2)}-\eta\mathbf{I})=\eta^2-a_n\eta-\frac{|\mathbf{a}_{\tau}|^2}{4}. \tag*{(65)}

An arbitrary tangential axis is used when aτ=0\mathbf{a}_{\tau}=\mathbf{0}. The two eigenvalues are

η±=a⋅n±∣a∣2,η−≤0≤η+,(66)\eta_{\pm}=\frac{\mathbf{a}\cdot\mathbf{n}\pm|\mathbf{a}|}{2},\qquad\eta_{-}\le0\le\eta_{+}, \tag*{(66)}

and any remaining eigenvalues vanish. Hence tr⁡A=a⋅n\operatorname{tr}\mathbf{A}=\mathbf{a}\cdot\mathbf{n} and A−:A−=η−2\mathbf{A}^{-}:\mathbf{A}^{-}=\eta_{-}^{2}, which gives (32). Since ζ′≥0\zeta'\ge0 and W−W^{-} is quadratic under nonnegative scaling, integration along the normal yields

∫RW−(ζ′(t/δ)δA) dt=W−(A)δ∫R∣ζ′(r)∣2 dr.\int_{\mathbb{R}}W^{-}\left(\frac{\zeta'(t/\delta)}{\delta}\mathbf{A}\right)\,\mathrm{d}t=\frac{W^{-}(\mathbf{A})}{\delta}\int_{\mathbb{R}}|\zeta'(r)|^{2}\,\mathrm{d}r.

which proves (33). For pure opening, a=γn\mathbf{a}=\gamma\mathbf{n} with fixed γ>0\gamma>0, this term vanishes. For the local calculation per unit interface area, we use the infinite planar extension with full averaging balls, representing an interior patch away from the exterior boundary and interface edges. For δε=o(ε)\delta_{\varepsilon}=o(\varepsilon), the argument of Φ\Phi diverges at each fixed rescaled normal position t/ε∈(−1,1)t/\varepsilon\in(-1,1). The affected layer lies within ∣t∣≤ε+δε/2|t|\le\varepsilon+\delta_{\varepsilon}/2; hence boundedness of Φ\Phi and dominated convergence give

Eε,δεopen=1ε∫RΦ(εAεW+(e(uδε))) dt⟶∫−11Φ(∞) dr=Gc.(67)\mathcal{E}_{\varepsilon,\delta_{\varepsilon}}^{\mathrm{open}}=\frac{1}{\varepsilon}\int_{\mathbb{R}}\Phi\left(\varepsilon\mathcal{A}_{\varepsilon}W^{+}\left(\mathbf{e}(\mathbf{u}_{\delta_{\varepsilon}})\right)\right)\,\mathrm{d}t\longrightarrow\int_{-1}^{1}\Phi(\infty)\,\mathrm{d}r=G_{\mathrm{c}}. \tag*{(67)}

Finally, W+(−rI)=0W^{+}(-r\mathbf{I})=0 for every r>0r>0, which verifies the lack of coercivity stated in § 3.1.

Crack-path identification and apparent tip speed

Initiation, branching and angles are extracted from crack-connected Gauss points satisfying 1−q+≥0.951-q_{+}\ge0.95. The recorded tip rk\mathbf{r}_{k} at time tkt_{k} is the ridge tip for Kalthoff–Winkler and the tip of the connected front of the nodally projected Q1Q_{1} indicator for branching, using the same threshold. In the branching test, the recorded tip follows the parent crack up to branching and the upper branch thereafter. For t∈[tk,tk+1]t\in[t_{k},t_{k+1}], define

r^(t)=(1−α)rk+αrk+1,α=t−tktk+1−tk.(68)\hat{\mathbf{r}}(t)=(1-\alpha)\mathbf{r}_{k}+\alpha\mathbf{r}_{k+1},\qquad\alpha=\frac{t-t_{k}}{t_{k+1}-t_{k}}. \tag*{(68)}

The apparent tip speed is

vapp(t;τ)=∥r(t+τ/2)−r(t−τ/2)∥τ,τ=2 μs,(69)v_{\mathrm{app}}(t;\tau) = \frac{\left\lVert\mathbf{r}(t+\tau/2)-\mathbf{r}(t-\tau/2)\right\rVert}{\tau}, \qquad\tau= 2\,\mu\mathrm{s}, \tag*{(69)}

evaluated at centres spaced by 0.5 μs0.5\,\mu\mathrm{s}.

Declaration of competing interest

The authors declare no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

References

  1. [1]A. A. Griffith, The phenomena of rupture and flow in solids, Philosophical Transactions of the Royal Society of London. Series A 221 (1921) 163–198. doi:10.1098/rsta.1921.0006.DOI
  2. [2]G. A. Francfort, J.-J. Marigo, Revisiting brittle fracture as an energy minimization problem, Journal of the Mechanics and Physics of Solids 46 (1998) 1319–1342. doi:10.1016/S0022-5096(98)00034-9.DOI
  3. [3]B. Bourdin, G. A. Francfort, J.-J. Marigo, The variational approach to fracture, Journal of Elasticity 91 (2008) 5–148. doi:10.1007/s10659-007-9107-3.DOI
  4. [4]N. Moës, J. Dolbow, T. Belytschko, A finite element method for crack growth without remeshing, International Journal for Numerical Methods in Engineering 46 (1999) 131–150. doi:10.1002/(SICI)1097-0207(19990910)46:1<131::AID-NME726>3.0.CO;2-J.doi.org/10.1002/(sici)1097-0207(19990910)46:1<131::aid-nme726>3.0.co;2-j
  5. [5]L. Ambrosio, V. M. Tortorelli, Approximation of functionals depending on jumps by elliptic functionals via Γ-convergence, Communications on Pure and Applied Mathematics 43 (1990) 999–1036. doi:10.1002/cpa.3160430805.DOI
  6. [6]B. Bourdin, G. A. Francfort, J.-J. Marigo, Numerical experiments in revisited brittle fracture, Journal of the Mechanics and Physics of Solids 48 (2000) 797–826. doi:10.1016/S0022-5096(99)00028-9.DOI
  7. [7]J.-Y. Wu, A unified phase-field theory for the mechanics of damage and quasi-brittle failure, Journal of the Mechanics and Physics of Solids 103 (2017) 72–99. doi:10.1016/j.jmps.2017.03.015.DOI
  8. [8]S. Conti, M. Focardi, F. Jurlano, Phase field approximation of cohesive fracture models, Annales de l’Institut Henri Poincaré C, Analyse non linéaire 33 (2016) 1033–1067. doi:10.1016/j.anihpc.2015.02.001.DOI
  9. [9]M. J. Borden, C. V. Verhoosel, M. A. Scott, T. J. R. Hughes, C. M. Landis, A phase-field description of dynamic brittle fracture, Computer Methods in Applied Mechanics and Engineering 217–220 (2012) 77–95. doi:10.1016/j.cma.2012.01.008.DOI
  10. [10]Q. Cheng, Y. Sun, X. Xu, A variational nonlocal phase-field model for dynamic fracture in elastic solids, arXiv preprint arXiv:2607.01881, 2026. doi:10.48550/arXiv.2607.01881.DOI
  11. [11]M. Ambati, T. Gerasimov, L. De Lorenzis, Phase-field modeling of ductile fracture, Computational Mechanics 55 (2015) 1017–1040. doi:10.1007/s00466-015-1151-4.DOI
  12. [12]C. Miehe, L.-M. Schänzel, Phase field modeling of fracture in rubbery polymers. Part I: Finite elasticity coupled with brittle failure, Journal of the Mechanics and Physics of Solids 65 (2014) 93–113. doi:10.1016/j.jmps.2013.06.007.DOI
  13. [13]C. Miehe, L.-M. Schänzel, H. Ulmer, Phase field modeling of fracture in multi-physics problems. Part I. balance of crack surface and failure criteria for brittle crack propagation in thermo-elastic solids, Computer Methods in Applied Mechanics and Engineering 294 (2015) 449–485. doi:10.1016/j.cma.2014.11.016.DOI
  14. [14]C. Miehe, H. Dal, L.-M. Schänzel, A. Raina, A phase-field model for chemo-mechanical induced fracture in lithium-ion battery electrode particles, International Journal for Numerical Methods in Engineering 106 (2016) 683–711. doi:10.1002/nme.5133.DOI
  15. [15]E. Martínez-Pañeda, A. Golahmar, C. F. Niordson, A phase field formulation for hydrogen assisted cracking, Computer Methods in Applied Mechanics and Engineering 342 (2018) 742–761. doi:10.1016/j.cma.2018.07.021.
  16. [16]G. Pijaudier-Cabot, Z. P. Bažant, Nonlocal damage theory, Journal of Engineering Mechanics 113 (1987) 1512–1533. doi:10.1061/(ASCE)0733-9399(1987)113:10(1512).
  17. [17]R. H. J. Peerlings, R. de Borst, W. A. M. Brekelmans, J. H. P. de Vree, Gradient enhanced damage for quasi-brittle materials, International Journal for Numerical Methods in Engineering [18] K. Pham, H. Amor, J.-J. Marigo, C. Maurini, Gradient damage models and their use to approximate brittle fracture, International Journal of Damage Mechanics 20 (2011) 618–652. doi:10.1177/1056789510386852.
  18. [19]H. Gao, P. Klein, Numerical simulation of crack growth in an isotropic solid with randomized internal cohesive bonds, Journal of the Mechanics and Physics of Solids 46 (1998) 187–218. doi:10.1016/S0022–5096(97)00047–1.DOI
  19. [20]Y. Sun, H. Yu, A geometric internal bond model for fracture of brittle materials, Engineering Fracture Mechanics 325 (2025) 111329. doi:10.1016/j.engfracmech.2025.111329.DOI
  20. [21]S. A. Silling, Reformulation of elasticity theory for discontinuities and long-range forces, Journal of the Mechanics and Physics of Solids 48 (2000) 175–209. doi:10.1016/S0022–5096(99)00029–0.
  21. [22]S. A. Silling, M. Epton, O. Weckner, J. Xu, E. Askari, Peridynamic states and constitutive modeling, Journal of Elasticity 88 (2007) 151–184. doi:10.1007/s10659–007–9125–1.DOI
  22. [23]H. Yu, Y. Sun, Nonlocal geometric fracture theory: A nonlocal–local asymptotically compatible framework for continuous–discontinuous deformation of solid materials, International Journal for Numerical Methods in Engineering 124 (2023) 4740–4768. doi:10.1002/nme.7330.DOI
  23. [24]H. Ren, T. Rabczuk, X. Zhuang, Variational damage model: A new paradigm for fractures, Frontiers of Structural and Civil Engineering 19 (2025) 1–21. doi:10.1007/s11709–025–1144–0.
  24. [25]A. Braides, G. Dal Maso, Non-local approximation of the Mumford–Shah functional, Calculus of Variations and Partial Differential Equations 5 (1997) 293–322. doi:10.1007/s005260050068.DOI
  25. [26]L. Lussardi, M. Negri, Convergence of nonlocal finite element energies for fracture mechanics, Numerical Functional Analysis and Optimization 28 (2007) 83–109. doi:10.1080/01630560701189994.DOI
  26. [27]G. Scilla, F. Solombrino, Non-local approximation of the Griffith functional, NoDEA: Nonlinear Differential Equations and Applications 28 (2021) 17. doi:10.1007/s00030–021–00682–y.
  27. [28]G. Dal Maso, Generalised functions of bounded deformation, Journal of the European Mathematical Society 15 (2013) 1943–1997. doi:10.4171/JEMS/410.DOI
  28. [29]C. Miehe, M. Hofacker, F. Welschinger, A phase field model for rate-independent crack propagation: Robust algorithmic implementation based on operator splits, Computer Methods in Applied Mechanics and Engineering 199 (2010) 2765–2778. doi:10.1016/j.cma.2010.04.011.DOI
  29. [30]J. Shen, J. Xu, J. Yang, The scalar auxiliary variable (SAV) approach for gradient flows, Journal of Computational Physics 353 (2018) 407–416. doi:10.1016/j.jcp.2017.10.021.DOI
  30. [31]J. Shen, J. Xu, J. Yang, A new class of efficient and robust energy stable schemes for gradient flows, SIAM Review 61 (2019) 474–506. doi:10.1137/17M1150153.DOI
  31. [32]Q. Cheng, C. Liu, J. Shen, Generalized SAV approaches for gradient systems, Journal of Computational and Applied Mathematics 394 (2021) 113532. doi:10.1016/j.cam.2021.113532.DOI
  32. [33]Q. Cheng, C. Liu, J. Shen, A new Lagrange multiplier approach for gradient flows, Computer Methods in Applied Mechanics and Engineering 367 (2020) 113070. doi:10.1016/j.cma.2020.113070.DOI
  33. [34]W. C. Swope, H. C. Andersen, P. H. Berens, K. R. Wilson, A computer simulation method for the calculation of equilibrium constants for the formation of physical clusters of molecules: Application to small water clusters, The Journal of Chemical Physics 76 (1982) 637–649. doi:10.1063/1.442716.DOI
  34. [35]Z.-C. Shi, X. Xu, Z. Zhang, The patch recovery for finite element approximation of elasticity problems under quadrilateral meshes, Discrete and Continuous Dynamical Systems - B 9 (2008) 163–182. doi:10.3934/dcdsb.2008.9.163.DOI
  35. [36]J. F. Kalthoff, Modes of dynamic shear failure in solids, International Journal of Fracture 101 (2000) 1–31. doi:10.1023/A:1007647800529.DOI

Paper details

vlnc:W_ubnruvad
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC0-1.0
Usage

views · 0

downloads · 0

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

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents