Research

Current Research Topics

Large-scale rock friction experiments

My current research activity mainly involves conducting rock friction experiments at NIED, Japan (see photo above for the scale of our apparatus). I am interested in understanding (1) how fault slip mode evolves with the on-going or cumulative fault displacement, (2) possible scale and loading-rate dependence of rock friction, (3) rupture nucleation process, (4) selection of rupture mode between sub-Rayleigh and supershear, (5) governing conditions for producing long-lived or short-lived supershear ruptures, (6) bimaterial ruptures, (7) relation between macroscopic fault behavior and microscopic process. Some preliminary results have been reported in recent meetings and workshops (abstract titles listed in my CV).

Past Research Topics

Dynamic ruptures with off-fault plastic yielding

With Yehuda Ben-Zion and Jean-Paul Ampuero, we implemented the Spectral Element code SEM2DPACK (originally developed by Ampuero) with a Mohr-Coulomb type plasticity to study off-fault plastic yielding during 2-D in-plane dynamic ruptures. In the framework of the commonly used elasto-plasticity, the elastic moduli do not change with the accumulation of plastic strain while the plastic (hardening) moduli or other related properties (e.g. internal friction angle, rock cohesion) can evolve with the on-going plastic deformation. For simplicity, a rate-dependent visco-plasticity with constant plastic properties was employed to produce smoothly distributed plastic strain. Detailed results on plastic yielding zone properties during dynamic ruptures and comparison to other studies focusing on quasi-static fault propagation or fault roughness can be found in Xu et al. (GJI, 2012a, 2012b).

Figure above shows the distribution of off-fault plastic yielding zone (top panel) and spatiotemporal evolution of slip as well as slip rate (bottom panel) for a crack-like rupture (a, c) and a pulse-like rupture (b, d). Notice that both the plastic yielding zone geometry and the slip profile show approximately self-similarity for the crack rupture mode, but are almost translation-invariant for the pulse rupture mode.

Dynamic ruptures with off-fault brittle damage

Unlike plasticity, the brittle damage model can account for dynamic changes of elastic moduli inside the off-fault yielding zone. The basic theory was formulated by Vladimir Lyakhovsky, Yehuda Ben-Zion and their collaborators on the basis of thermodynamics accounting for internal irreversible process, and based on laboratory observation that the relevant elastic moduli of rocks are larger under compressive loading than under tensile loading. The model was first used to study evolution of fault zone structure under quasi-static loading, and then extended to study problems during propagation of dynamic ruptures. To simplify, the model explicitly assumes that the elastic moduli are dependent on a scalar damage variable, which collectively describes the density of distributed microcracks at some mesoscopic scale. It further assumes how the damage variable can evolve with time as a function of current deformation state, by obeying the principle of non-negative entropy production. Prominent features of this brittle damage model include spontaneous generation of (1) a material contrast across the fault, (2) dynamic normal stress change along the fault, and (3) trapped/reflected waves on the damaged side(s) of the fault. These features in turn can greatly modify the rupture behavior, which would not be expected in purely elastic simulations with initially homogeneous rock properties. More details can be found in Xu et al. (PAGEOPH, 2015) and the cited references therein.

Figure above shows examples of how the spontaneous generation of off-fault brittle damage can lead to (a) reduction of rock elastic moduli (expressed as reduction of shear wave speed), and a detached pulse front from an original crack-like rupture (indicated by the black curve representing slip rate profile), (b) asymmetric particle velocity across the fault, and (c) alternating normal stress change surrounding the fault and trapped wave signals on the damaged side of the fault.

Off-fault damage, shear localization, and implications for rupture dynamics

Shear localization (or delocalization) is well known to have a strong dependence on rock rheology, material inhomogeneity or anisotropy, loading condition, and many other factors. Moreover, the preferred mode of shear localization is found to depend on some non-local process that requires to consider changing configuration of the loading (e.g. propagation of the rupture front) and finite growth of shear localization (e.g. whether subject to the same order of stress variation). In the work of Xu and Ben-Zion (GJI, 2013), we first introduced a concept of stress non-locality in space and time, and then applied such concept to understand off-fault damage pattern produced by earthquake ruptures. We particularly investigated presence vs. absence of off-fault damage, and synthetic vs. antithetic shear bands related to earthquake ruptures. Our study suggested that (1) relative absence of off-fault damage may indicate abrupt rupture acceleration or jump, expected near fault releasing bends or locations with reduced fracture energy, (2) prominent off-fault damage particularly well-developed antithetic shear band may indicate abrupt rupture deceleration or arrest, expected near fault restraining bends or segment ends.

Figures above show anticipated (left panel) and numerically simulated (right panel) large-scale shear bands (representing splay faults, fault branches) generated by earthquake ruptures in typical (a) thrust regime remotely stressed at low angles, and (b) strike-slip regime remotely stressed at moderate-to-high angles. The branching scenario in (a) provides an alternative view to understand forethrusts and backthrusts often observed in accretionary subduction zones. The anticipated antithetic branching in the strike-slip regime is not well developed during smooth rupture propagation shown by numerical simulations. This can be explained by a faster stress decaying gradient along that direction and by rapid moving out of the examined path from a forward propagating rupture front (see illustration by the following figure).

Figure above illustrates how the "directivity effect" can influence the preferred mode of growing bands. Here we simply assume an isotropic loading stress field that has decaying magnitude away from some reference point. When such stress field does not change its configuration (i.e. it remains in place), virtual bands growing to different radial directions (e.g., marked by red, black, and blue) will be subject to the same stress decaying gradient. However, if the stress field is moving, say to the right direction, then factors of "direction cosine" and "relative growth speed of bands with respect to that of the moving stress field" will influence the stress gradient evaluated in an absolute coordinate system. Clearly, a virtual band growing to the "red" direction will be subject to a slower stress decaying gradient and can continue benefiting from the loading, while a virtual band to the "blue" direction will be subject to a faster stress decaying gradient and will be quickly removed from the active loading area. In general, a forward direction with a smaller inclination angle will be more favored for band growth than a backward direction with a larger inclination angle. To recover band growth along a backward direction, one has to temporarily or fully halt the moving stress field, as shown by the following two examples.

Left figure shows rupture speed variation along a heterogeneous strike-slip fault marked by piecewise constant fracture energy (here reflected by a variation in slip-weakening distance Dc). Red and blue bars highlight the transition boundaries into a stronger patch and a weaker patch, respectively. Right figure shows the simulated damage pattern near two transition boundaries: with abrupt rupture deceleration around the red bar, enhanced damage magnitude and antithetic shear band are observed; with abrupt rupture acceleration around the blue bar, a damage gap is observed.

Figure above shows well-developed antithetic (left-lateral) shear band near rupture termination end along a strike-slip fault (note its absence before rupture was terminated by a strong barrier). This band makes a high angle (near orthogonal) to the main strike-slip fault on its extensional side, which may represent a high-angle aftershock cluster or conjugate earthquake triggering in a complex fault network. Examples include the section between the San Andreas Fault and the San Jacinto Fault (the 1987 Superstition Hills earthquake sequence), the Eastern California Shear Zone (high-angle aftershock cluster near the northern end of the 1992 Landers earthquake), junction between the North Anatolian Fault and the East Anatolian Fault (see Stein et al., GJI 1997), the 1997 Kagoshima earthquake doublet in Japan, oceanic intraplate earthquakes and their aftershocks (the 1998 Antarctic Plate earthquake), cross-cutting fault system in Greece (the 2001 Skyros earthquake), obliquely stressed diffuse zone (containing magnetic anomaly, fracture zones, transform faults, etc) near some subduction plate boundaries (the 1987-1992 earthquake sequence in Gulf of Alaska, earthquake sequence in 2000 and in 2012 southwest of Sumatra).

Subduction zone megathrust earthquakes and dynamic reactivation of splay faults

Inspired by recently increased seismic activity along some backthrust faults in Taiwan, Sumatra and South America, we conducted numerical simulations to investigate how splay faults, particularly backthrusts, can be reactivated by megathrust earthquakes in subduction zones. Background review and detailed parameter-space study can be found in Xu et al. (Tectonophysics, 2015).

Before probing how backthrusts can be reactivated, it is useful to first understand how they can be initiated. Figure above unifies a range of barrier models (defined in a general sense) that can promote initiation of backthrusts. They include the fault-bend fold model or the subducting seamount model that emphasizes the effect of a geometric restraining bend (Type-I), the hybrid model combining the Coulomb wedge theory and limit analysis that emphasizes an increasing basal friction towards the trench (Type-II), and the fault propagation/termination model that emphasizes a slipped/locked basal boundary condition (Type-III). The shared kinematics by all three barrier models is shown by the plot on the right.

The configuration of geometric irregularities or the frictional strength along the basal fault (the megathrust) is not necessarily stationary, but could evolve with time. Figure above summarizes several mechanisms that can produce slip reversal along a backthrust splay by a switching behavior between "barrier-like" (left column) and "asperity-like" (right column) along the megathrust. Scenario (a) emphasizes the opposite stressing regimes dominated by compression on the leading edge and by extension on the trailing edge of a subducting seamount, respectively. Doing work against gravity during uplift on the leading edge may enhance the "barrier-like" behavior in addition to the restraining-bend effect, while release of the gravitational energy (slumping) on the trailing edge may enhance the "asperity-like" behavior in addition to the releasing-bend effect. Scenario (b) emphasizes the alternating polarity of strength contrast between the down-dip aseismic zone and the seismogenic zone. During the inter-seismic period, stable creep in the aseismic zone can load the locked seismogenic zone (appearing stronger) near its down-dip edge, which together with the remote loading drives the nearby upper plate into compression. However, during the co-seismic period, the seismogenic zone slips and becomes weaker than its down-dip aseismic zone, which drives the nearby upper plate into extension. Scenario (c) focuses on different stages of a single megathrust rupture. A (statically) strong patch near the trench may initially halt the rupture, making the frontal wedge under compression at an earlier stage. But such patch may eventually fail due to dynamic weakening, making the frontal wedge under extension at a later stage.