Beyond linear stability: Heterogeneity-induced fingering of crack fronts

Driven interfaces are ubiquitous and underlie phenomena as diverse as wetting [1], magnetism [2], solid phase transformation [3], growth processes in biology [4], multiphase flow [5, 6], dislocation loops [7, 8, 9], earthquakes [10, 11, 12, 13], and adhesion [14]. In all these physical situations, the geometrical stability of the interface, i.e., its ability to keep an overall straight configuration despite the destabilizing effects of material heterogeneities, loading conditions, thermal fluctuations, etc., is a central issue, as the interface behavior at small scales drives the transport or mechanical properties at large scales. For small amplitude perturbations, a linear stability analysis is generally sufficient to infer the interface stability and to some extent, characterize the unstable modes. For example, this approach accounts for the viscous Saffman-Taylor fingering between two immiscible fluids [5] or the destabilization of a peeling front in a confined adhesive layer [15, 16, 17, 18]. However, this method gives no hint on the interface response to finite-amplitude perturbations that may result from naturally strong heterogeneities or obstacles, as for instance those designed to achieve new properties [19, 20].

Here, we investigate this problem in the context of fracture, using crack fronts propagating in heterogeneous brittle solids [21, 22, 23] as a paradigmatic realization of elastic interfaces. To probe their nonlinear response, we consider penny-shaped cracks drove through periodic arrays of obstacles with tunable size and toughness. While this configuration was previously used to explore the relationship between local and global fracture properties [1], we focus here on the stability of the crack front.

At low toughness contrasts, we recover the results of the linear stability analysis that predicts stable crack growth [25, 26]. Beyond a critical toughness contrast, however, the crack front destabilizes into fingers that grow unstably. Near threshold, the crack-front dynamics displays the hallmark scaling signatures of a classical saddle-node bifurcation. Surprisingly, however, our analysis provides no evidence for the unstable branch required by the conventional fold scenario. Instead, the instability appears to result from the disappearance of Griffith-compatible energy-minimizing crack-front configurations. Finally, we discuss how this mechanism may extend to a broad class of elastic interfaces and how it could offer a route for rationalizing the brittle-to-quasibrittle transition in heterogeneous materials.

Penny-shaped cracks propagating in brittle solids provide a prototypical realization of elastic interfaces when dissipative failure processes remain confined to the vicinity of the crack tip [27, 28]. This looped interface behaves then like an elastic line with long-range elasticity that emerges from the material bulk elasticity [25, 29]. In the reference configuration with homogeneous material properties and remote uniform tensile loading (Fig. 1a), the crack front adopts a circular shape of radius $a$ and the driving force acting on it is uniform. Its magnitude is set by the Stress Intensity Factor (SIF), $K_{\circ}=2\,\sigma\,\sqrt{a/\pi}$ where $\sigma$ is the tensile stress applied remotely, or equivalently, by the rate $G_{\circ}=-\delta\mathcal{E}_{\mathrm{m}}/\delta\mathcal{S}=K_{\circ}^{2}\,(1-\nu^{2})/E$ of mechanical energy released for a uniform crack advance $\delta a=\delta\mathcal{S}/2\pi a$, where $\mathcal{S}$ is the crack surface area, and $E$ and $\nu$ are the Young’s modulus and the Poisson’s ratio, respectively [30]. The crack evolution is then inferred from the comparison of $K_{\circ}$ with the material toughness $K_{\mathrm{c}}$, or equivalently, of $G_{\circ}$ with the fracture energy $G_{\mathrm{c}}=\delta\mathcal{E}_{\mathrm{diss}}/\delta\mathcal{S}=K_{\mathrm{c}}^{2}\,(1-\nu^{2})/E$ where $\mathcal{E}_{\mathrm{diss}}$ is the energy dissipated by fracture through the so-called Griffith criterion $\dot{a}\,(G_{\circ}-G_{c})=0$ that derives from thermodynamics of irreversible processes [31]. This criterion implies that during crack propagation, $\dot{a}>0$, the mechanical energy released during crack growth is dissipated into fracture energy, a condition that writes as $G_{\circ}=G_{\mathrm{c}}$ [32, 31]. Note that we assume here quasi-static crack propagation and neglect any rate-dependence of the fracture energy. Rewritten in terms of SIF, the propagation criterion writes as $\dot{a}\,(K_{\circ}-K_{\mathrm{c}})=0$. The concepts of driving force and growth resistance, defined here within a thermodynamic framework, can be generalized to any interface problem.

Under the effect of material heterogeneities or heterogeneous loading conditions, the interface geometry becomes more complex and the driving force $K(s)$ varies with the position $s$ along the interface (Fig. 1b). The Griffith criterion then applies locally along the front [33, 34]. For cracks with small perturbations $\delta a(s)=a(s)-a$, $\left|\delta a(s)\right|\ll a$, where $a=\sqrt{\mathcal{S}/\pi}$, the corresponding variations of the SIF $\delta K(s)=K(s)-K_{\circ}$ can be determined using a perturbation-based formula $\delta K(s)=\mathcal{W}\ast\delta a(s)$ that involves a long-range kernel $\mathcal{W}\sim 1/s^{2}$ (see Refs. [35, 36] for reviews).

To explore the linear stability of moving interfaces in general, and crack fronts in particular, small harmonic perturbations $\delta a^{\mathrm{(k)}}(\theta)=-\Delta a/2\,\cos(k\theta)$ of amplitude $\Delta a\ll a$ and wavelength $\lambda=2\pi a/k$ where $k\geq 2$ is the wave number are considered. Applying the linearized formula that relates $\delta K(\theta)$ to $\delta a(\theta)$ using the expression of the kernel $\mathcal{W}$ for penny-shaped cracks [26] provides the driving force perturbations $\delta K(\theta)/K_{\circ}=(k-1)\,\delta a^{\mathrm{(k)}}(\theta)/2a$ that also follow sinusoidal variations, but with an opposite sign. As a result, the most advanced regions of the front located in $\theta=0,\,2\pi/k,...2\pi(k-1)/k$ are also those with the smallest driving force, ensuring that any small front perturbations, that can be decomposed as a sum of harmonic modes $\delta a^{\mathrm{(k)}}(\theta)$, vanish, accounting for the circular shape adopted by penny-shaped cracks growing in homogeneous materials [26].

To explore whether penny-shaped cracks remain stable in the presence of perturbations of finite amplitude, we must go beyond linear stability analysis. We propose herein to trigger large perturbations using a heterogeneous toughness field $K_{\mathrm{c}}(\theta)=\overline{K}_{\mathrm{c}}\left[1+\Delta\cos(k\theta)\right]$, that consists in $k$ obstacles of toughness contrast $\Delta\in[0\;1]$ (Fig. 1b), $\overline{K}_{\mathrm{c}}$ being the mean toughness [1]. Starting from a circular configuration of radius $a_{0}$, the crack grows and deforms in a flower shape with $k-$petals. The size of the crack can be described by its surface area $\mathcal{S}$ or equivalently by the radius $a=\sqrt{\mathcal{S}/\pi}$ of the circle of same surface [37], and its deformation by the petal amplitude $\Delta a$, as schematized in Fig. 1b. To ensure quasi-static propagation, the applied stress $\sigma$ is adjusted at each time step so that at least one point along the front (here the points located in the weak regions $\theta=\pi/k,$ $3\pi/k$, … $(2k-1)\pi/k$ of the fracture plane) verifies $K=K_{\mathrm{c}}$ while $K\leq K_{\mathrm{c}}$ elsewhere. To carry this non-linear stability analysis and compute $K(s)$ along the front, one must resort to either higher-order analytical theory [38] or numerical methods [39, 4]. These require the determination of the perturbations of the kernel $\mathcal{W}$ that is obtained by an analog formula to $\delta K(s)=\mathcal{W}\ast\delta a(s)$ [3]. In practice, linear approximations of $\delta K(s)$ and $\delta\mathcal{W}$ are added incrementally during the deformation of the front, starting from the initial circular configuration [5, 4, 6].

For weak obstacles $\Delta\ll 1$, the method is equivalent to a linear stability analysis. Indeed, the stationary modes of growth satisfying $K(\theta)=K_{\mathrm{c}}(\theta)$ correspond to the harmonic perturbations $\delta a^{\mathrm{(k)}}(\theta)$ introduced previously, with $\Delta a/a=4\Delta/(k-1)$. However, considering tougher obstacles, the method allows to study the stability of penny-shaped cracks to perturbations of any amplitude. The results of our numerical investigation, comprising hundreds of simulations, are compiled in the phase diagram of Fig. 1c in the normalized obstacle width $1/k=\lambda/2\pi a$ versus toughness contrast $\Delta$ plane. For small toughness contrasts, $\Delta<\Delta_{c}(1/k)$, the crack front adopts a stationary shape represented in green in the inset (a) of Fig. 1c. In this so-called weak pinning regime, the propagation threshold $K(s)=K_{\mathrm{c}}(s)$ is reached everywhere along the front, after an initial transient phase during which the crack advances in the weakest zones only (successive red front positions). Moreover, the effective toughness that defines the load required to propagate the crack through the relation $K^{\mathrm{eff}}_{\mathrm{c}}\equiv 2\sigma\sqrt{a/\pi}$ turns out to be nearly equal to the mean toughness $K^{\mathrm{eff}}_{\mathrm{c}}\simeq\overline{K}_{\mathrm{c}}$ [1]. Conversely, for large toughness contrasts, $\Delta>\Delta_{c}(1/k)$, a fingering regime sets in: Some regions of the front never reach the propagation threshold and thus remain trapped by the obstacles while the rest satisfies locally to the propagation criterion $K=K_{\mathrm{c}}$ and thus keeps propagating yielding infinitely growing fingers invading the weakest regions of the fracture plane, as shown in inset(b) of Fig. 1c. An interesting limit deserves attention: As ${k\to\infty}$, the critical toughness contrast goes to one. This corresponds to infinitely tough obstacles, highlighting the stability of semi-infinite cracks, even to perturbations of finite amplitude, as taking the limit $k\to\infty\Leftrightarrow\lambda\ll a$ amounts to considering a straight crack front. This is consistent with the observation that straight crack fronts stiffen as they deform beyond the linear regime [44, 45].

Both crack growth regimes can be differentiated kinematically from the evolution of the dimensionless petal length $\Delta a/a$ shown in Fig. 2a. In the weak pinning regime, $\Delta a/a$ saturates to a stationary value $(\Delta a/a)_{\mathrm{s}}$, while the normalized petal length grows continuously in the fingering regime. The underlying mechanism is studied in Fig. 2b that shows the difference of normalized SIF $\Delta K/\overline{K}_{\mathrm{c}}\equiv\left[K(B)-K(A)\right]/\overline{K}_{\mathrm{c}}$ between the regions $B$ of the front trapped by the obstacles and the regions $A$ growing in-between. As the initial crack configuration is circular, the SIF is initially uniform, so $\Delta K=0$. However, as the front propagates in $A$ in the weakest regions of the fracture plane, the SIF increases in $B$ owing to the front elasticity. In the weak pinning regime, this results in an increase of $\Delta K$ until it reaches a plateau, $\Delta K/\overline{K}_{\mathrm{c}}=2\Delta$, reminiscent of the stationary state.

Unstable cracks display a strikingly different behavior, as $\Delta K$ first increases and then starts decreasing before the untrapping threshold $\Delta K/\overline{K}_{\mathrm{c}}=2\Delta$ can be reached. This unexpected drop may result from the shielding effect produced by the large petals, which behave like individual cracks on their own (see the shape of the crack shown with solid line in inset of Fig. 2b). It explains why regions $B$ lag behind the rest of the crack and ultimately remain trapped by the obstacles as an array of nearly independent cracks (or petals) continues to grow. As a support to this interpretation, one notices that $\Delta K$ starts decreasing when $\Delta a\simeq a_{0}$, i.e. when daughter cracks reach a size similar to their mother crack. This suggests that the long-range elastic restoring forces that maintain penny-shaped cracks in an overall circular shape vanish for perturbations larger than the crack itself.

To get deeper insights into the nature of this instability, we now recast the formation of fingers in terms of energy landscape and variational approach to fracture [46]. We focus on the total energy ${\cal E}$ defined as the sum of the elastic energy and the fracture cost. For a circular crack in a homogeneous medium, it reads $\mathcal{E}_{\circ}=\pi a^{2}\overline{G}_{\rm c}-8/3\pi\,\sigma^{2}a^{3}(1-\nu^{2})/E$ [47]. It reaches a maximum at $a_{\circ}(\sigma)=\pi/4(\overline{K}_{\rm c}/\sigma)^{2}$ corresponding to the crack size satisfying the Griffith condition $G_{\circ}=G_{\mathrm{c}}$. We now introduce front perturbations and focus on the energy variations $\Delta\mathcal{E}_{\circ}=\mathcal{E}-\mathcal{E}_{\circ}$. We seek for the front shape that minimizes $\Delta\mathcal{E}_{\circ}$ at constant crack surface area under a constant applied stress $\sigma$ [47].

For small perturbations $\delta a(s)$, the variation of elastic energy is $\int_{\mathcal{F}}G(s)\delta a(s)ds$ while the variation of fracture cost is $\int_{\mathcal{F}}G_{c}(s)\delta a(s)ds$, where both integrals are taken along the unperturbed crack front $\mathcal{F}$. Hence, the energy variation $\Delta\mathcal{E}_{\circ}$ can be obtained incrementally using the same perturbation method used previously to update $K$. In the weak pinning regime, we force petals to grow beyond their stationary amplitude $(\Delta a/a)_{\mathrm{s}}$ by using a toughness contrast $\Delta_{\mathrm{driving}}>\Delta_{\mathrm{c}}$ once the front has reached its stationary shape while keeping the primary toughness field to evaluate the variations of fracture energy. The energy landscape obtained by this protocol turns out to be qualitatively independent of $\Delta_{\mathrm{driving}}$ [47].

The variations $\Delta{\cal E}_{\circ}$ of total energy normalized by the reference energy $\mathcal{E}_{\mathrm{ref}}=\overline{G}_{\mathrm{c}}\mathcal{S}$ is shown in Fig. 3 as a function of the petal aspect ratio $R\equiv\Delta a/\lambda$ where $\lambda=2\pi a/k$ is the petal width, for different toughness contrasts $\Delta$ and obstacle numbers $k$. Focusing first on the weak pinning regime $\Delta<\Delta_{\mathrm{c}}\Leftrightarrow k>k_{\mathrm{c}}$, we observe an initial decrease of the total energy for small petal size $\Delta a/a<(\Delta a/a)_{\mathrm{s}}$, followed by an increase of energy for larger petals. The energy-minimizing configurations marked by dots in Fig. 3 correspond to stationary crack fronts of amplitude $(\Delta a/a)_{\mathrm{s}}$ satisfying the Griffith criterion $G(s)=G_{\mathrm{c}}(s)$, highlighting the equivalence between a global energy minimization approach and Griffith’s local propagation criterion [21, 46, 33].

Strikingly, as we approach the fingering threshold $\Delta\rightarrow\Delta_{\mathrm{c}}\Leftrightarrow k\rightarrow k_{\mathrm{c}}$, the potential well progressively flattens and eventually disappears. As shown in the supplemental materials [47], the relaxation time required for a crack front to retrieve its stationary shape after being perturbed diverges as $\tau\propto(\Delta_{\mathrm{c}}-\Delta)^{-1/2}$, revealing a critical slowing down commonly observed near saddle-node bifurcations.

We now explore in Fig. 4 how the stationary petal size, i.e. the order parameter of the bifurcation, varies with the toughness contrast $\Delta$, its control parameter. Panel (a) shows the petal aspect ratio $R_{\mathrm{s}}$ of the stationary crack configurations as a function of the toughness contrast. For $\Delta\ll\Delta_{\mathrm{c}}$, the petal aspect ratio increases linearly, $R_{\mathrm{s}}=\mathcal{C}_{0}(k)\,\Delta$, where $\mathcal{C}_{0}(k)$ is the linear front compliance. The linear stability analysis predicts $\mathcal{C}_{0}(k)=2k/(\pi(k-1))$ [26], in agreement with our simulations as shown in inset of Fig. 4a. Interestingly, the compliance $\mathcal{C}_{0}^{\infty}=2/\pi$ of a semi-infinite crack [48, 49] is retrieved when $k\rightarrow\infty$, confirming that the behavior of straight cracks is recovered in this limit.

With increasing obstacle toughness, the crack displays a non-linear response featured by an increasing compliance $\mathcal{C}(\Delta)={\rm d}R_{\mathrm{s}}/{\rm d}\Delta>\mathcal{C}_{0}$, until it diverges for $R_{\mathrm{s}}\rightarrow R_{\mathrm{c}}\Leftrightarrow\Delta\rightarrow\Delta_{\mathrm{c}}$ at the onset of fingering. This non-linear response $R_{\mathrm{s}}(\Delta)$ is shown in Fig. 4b in the normalized form $(R_{\mathrm{c}}-R_{\mathrm{s}})/R_{\mathrm{c}}$ versus $(\Delta-\Delta_{\mathrm{c}})/\Delta_{\mathrm{c}}$ for different values of $k$. Remarkably, all data collapse on a single master curve, revealing a power-law behavior $\left(R_{\mathrm{c}}-R_{\mathrm{s}}\right)/R_{\mathrm{c}}\simeq\left[(\Delta_{\mathrm{c}}-\Delta)/\Delta_{\mathrm{c}}\right]^{\gamma}$ with exponent $\gamma=0.51\pm 0.02$. This implies that the crack front compliance diverges as $\mathcal{C}\propto 1/\sqrt{\Delta_{c}-\Delta}$ upon approaching the onset of fingering for $\Delta\to\Delta_{c}$, or equivalently, that the front stiffness vanishes as $\mathcal{K}\equiv 1/\mathcal{C}\propto\sqrt{\Delta_{\mathrm{c}}-\Delta}$.

To elucidate this behavior, we consider the total energy ${\cal E}(\Delta,R)$ introduced previously that is represented schematically in Fig. 4c in the (petal aspect ratio $R/R_{c}$, toughness contrast $\Delta/\Delta_{c}$) plane. Because all stationary crack-front configurations satisfy the Griffith criterion, moving from one nearby stationary configuration to another leaves the total energy unchanged. The stable branch therefore forms an effectively isoenergetic valley of energy-minimizing crack-front configurations. Incorporating this property into a Taylor expansion of the energy near the bifurcation point yields ${\cal E}(\Delta,R_{\mathrm{s}})\simeq{\cal E}(\Delta_{\mathrm{c}},R_{\mathrm{c}})+\,\partial^{2}{\cal E}/\partial\Delta\partial R\,(\Delta-\Delta_{\mathrm{c}})(R_{\mathrm{s}}-R_{\mathrm{c}})+1/6\,\partial^{3}{\cal E}/\partial R^{3}\,(R_{\mathrm{s}}-R_{\mathrm{c}})^{3}+...$ where only the leading-order terms have been retained and all derivatives are evaluated at the critical point [47]. Balancing the two contributions immediately gives $R_{\mathrm{c}}-R_{\mathrm{s}}\propto\sqrt{\Delta_{\mathrm{c}}-\Delta}$ in agreement with the simulation results of Fig 4b.

At first sight, the fingering instability exhibits the hallmark features of a classical saddle-node bifurcation, including the square-root scaling of both the perturbation amplitude and the relaxation time [50]. Such behavior would normally be associated with the collision and disappearance of stable and unstable branches. Yet, several observations suggest a fundamentally different picture.

First, all stationary crack-front configurations satisfying the Griffith criterion remain isoenergetic. As a consequence, a hypothetical Griffith-compatible unstable branch should lie at the same energy level as the stable branch while appearing as an ensemble of maxima on the energy landscape of Fig. 4c, two features that do not seem compatible with each other. Second, throughout the stable regime, the observed branch acts as a unique attractor with no identifiable unstable counterpart, as even strongly fingered crack fronts ultimately relax toward the Griffith-compatible stationary configuration [47]. Taken together, these observations suggest that fingering does not originate from a local stable–unstable branch annihilation, but rather from a global loss of Griffith-compatible equilibria. Thenon-local elasticity of the crack front plays a major role in this mechanism, as it imposes constraints on the global geometry of the stable configurations.

This suggests that nonlocal elastic interfaces may host a broader class of collective global instabilities, fundamentally distinct from the conventional local bifurcation paradigm of stable-unstable branch annihilation. Remarkably, near threshold, the crack front converges toward a universal critical shape shown in Fig. 4d  [47]. This indicates that at the onset of fingering, the crack front loses memory of the details of the toughness map and becomes mainly governed by the nonlocal elasticity of the interface.

To conclude, we emphasize the distinctive and potentially generic nature of the fingering instability uncovered in crack fronts. By introducing obstacles with tunable toughness contrast, we drove penny-shaped crack fronts far beyond their linear response regime. Above a critical threshold, the front becomes unstable at a finite deformation and develops long fingers that eventually form arrays of daughter cracks. From a local perspective, this transition originates from the progressive vanishing of the crack-front stiffness, leaving the interface unable to suppress large front distortions. From a global perspective, however, the instability corresponds to the disappearance of the entire branch of stationary Griffith-compatible crack-front configurations, without any identifiable unstable counterpart.

Taken together, these observations suggest that nonlocal elastic interfaces may undergo instabilities fundamentally distinct from conventional local bifurcations. Yet, despite their different origin, these transitions retains the hallmark phenomenology of classical saddle-node bifurcations, including the square root scaling of both the perturbation amplitude and relaxation time near threshold.

An important extension of the present work would consist in relaxing the assumption of a connected crack front and allowing for crack nucleation. This could provide a route toward describing the brittle-to-quasibrittle transition within the broader context of collective global bifurcations in nonlocal elastic interfaces.

Acknowledgments - The authors thank Prof. Jean-Baptiste Leblond, Dr. Mathias Lebihain, Dr. Patrick Ribeiro, and Prof. K. Ravi-Chandar for fruitful discussions.