New techniques for the analysis of earthquake sources from local array data with an application to the 1993 Scotts Mills, Oregon, aftershock sequence

We analysed aftershocks recorded by a temporary digital seismic network following the moderate M _w = 5.5 1993, Scotts Mills, Oregon, earthquake. A technique to retrieve source moment tensors from local waveforms was developed, tested, and applied to 41 small earthquakes ( M _w ranging from 1.6 to 3.2). The derived focal mechanisms, although well resolved, are highly variable and do not share a common nodal plane. In contrast, the majority of the events, relocated with a joint hypocentre determination algorithm, collapse to a well-focused plane. The incompatibility of the nodal planes of most events with the plane defined by their locations suggests that the aftershocks did not occur on the fault plane, but tightly around it, outlining the rupture area rather than defining it. Furthermore, the moment tensors reveal stable P -axes, whereas T  -axes plunges are highly dispersed. We detect a rotation of average T  -axis plunge with depth, indicating a change from shallower, predominantly dip-slip mechanisms to deeper strike-slip mechanisms. These characteristics are difficult to explain by remnant stress concentrations on the main-shock rupture plane or asperity- and barrier-type models. We suggest that the aftershocks occurred under the ambient regional stress, triggered by a sudden weakening of the region surrounding the main-shock slip, rather than from a shear stress increase due to the main shock.     

A detailed analysis of the February 1996 aftershock sequence in the eastern Pyrenees, France

Following the 1996 February 18 M _L = 5.2 earthquake in the Agly massif in the eastern French Pyrenees, we installed a temporary network of seismometers around the epicentre. In this paper, we analyse 336 well-located aftershocks recorded from February 19 to February 23 by 18 temporary stations and two permanent stations located less than 35  km from the epicentre. Most aftershocks have been located with an accuracy better than 1.5  km in both horizontal and vertical positions. Their spatial distribution suggests the reactivation of a known fault system. We determined 39 fault-plane solutions using P -wave first motions. Despite their diversity, the focal mechanisms yield an E–W subhorizontal T-axis. We also determined fault-plane solutions and principal stress axes using the method developed by Rivera & Cisternas (1990 ) for the 15 best-recorded events. We obtain a pure-shear-rupture tectonic regime under N–S subhorizontal compression and E–W subhorizontal extension. These principal stress axes, which explain the focal mechanisms for at least 75 per cent of the 39 aftershocks, are different from the axes deduced from the main shock. The post-earthquake stress field caused by the main-shock rupture, modelled as sinistral strike slip on three vertical fault segments, is computed for various orientations and magnitudes of the regional stress field, assumed to be horizontal. The aftershock distribution is best explained for a compressive stress field oriented N30°E. Most aftershocks concentrate where the Coulomb failure stress change increases by more than 0.2  MPa. The diversity of aftershock focal mechanisms, poorly explained by this model, may reflect the great diversity in the orientations of pre-existing fractures in the Agly massif.     

What controls an earthquake's size? Results from a heterogeneous cellular automaton

The controls on an earthquake's size are examined in a heterogeneous cellular automaton that includes stress concentrations which scale with rupture size. Large events only occur when stress is highly correlated with strength over the entire fault. Although the largest events occur when this correlation is the highest, the magnitude of the correlation has no predictive value as events of all magnitudes occur during times of high stress/strength correlation. Rather, the size of any particular event depends on the local stress heterogeneity encountered by the growing rupture. Patterns of energy release with time for individual ruptures reflect this heterogeneity and many show nucleation-type behaviour, although there is no relation between the duration of nucleation phase and the size of the event. These results support the view that earthquake size is determined by complex interactions between previous event history and dynamic stress concentrations and suggest that deterministic earthquake prediction based on monitoring nucleation zones will not be possible.     

Non-Poissonian earthquake occurrence in coupled stress release models and its effect on seismic hazard

Most seismic hazard estimations are based on the assumption of a Poisson process for earthquake occurrence, even though both observations and models indicate a departure of real seismic sequences from this simplistic assumption. Instrumental earthquake catalogues show earthquake clustering on regional scales while the elastic rebound theory predicts a periodic recurrence of characteristic earthquakes on longer timescales for individual events. Recent implementations of time-dependent hazard calculations in California and Japan are based on quasi-periodic recurrences of fault ruptures according to renewal models such as the Brownian Passage Time model. However, these renewal models neglect earthquake interactions and the dependence on the stressing history which might destroy any regularity of earthquake recurrences in reality. To explore this, we investigate the (coupled) stress release model, a stochastic version of the elastic rebound hypothesis. In particular, we are interested in the time-variability of the occurrence of large earthquakes and its sensitivity to the occurrence of Gutenberg–Richter type earthquake activity and fault interactions. Our results show that in general large earthquakes occur quasi-periodically in the model: the occurrence probability of large earthquakes is strongly decreased shortly after a strong event and becomes constant on longer timescales. Although possible stress-interaction between adjacent fault zones does not affect the recurrence time distributions in each zone significantly, it leads to a temporal clustering of events on larger regional scales. The non-random characteristics, especially the quasi-periodic behaviour of large earthquakes, are even more pronounced if stress changes due to small earthquakes are less important. The recurrence-time distribution for the largest events is characterized by a coefficient of variation from 0.6 to 0.84 depending on the relative importance of small earthquakes.     

Estimating rupture scenario likelihood based on dynamic rupture simulations: the example of the segmented Middle Durance fault, southeastern France

The Middle Durance fault system, southeastern France, is a slow active fault that produced moderate-size historical seismic events and shows evidence of at least one   M _w ≳ 6.5  event in the last 29 000 yr. Based on dynamic rupture simulation, we propose earthquake scenarios that are constrained by knowledge of both the tectonic stress field and of the 3-D geometry of the Durance fault system. We simulate dynamic rupture interaction among several fault segmentations of different strikes, dips and rakes, using a 3-D boundary integral equation method. 50 combinations of reasonable stress field orientations, stress field amplitudes and hypocentre locations are tested. The probability of different rupture evolutions is then computed. Each segment ruptures mainly as a single event (44 per cent of the 50 simulations test in this paper). However, the probability that an event triggers simultaneously along three segments is high (26 per cent), leading to a potential rupture length of 45 km. Finally, 2 per cent of the simulations occur along four adjacent segments, producing the greatest total rupture length of 55 km. The simulation results show that the southernmost segment is most easily ruptured (40 per cent), because of its favourable orientation with respect to the tectonic stress and of its favourable location for interaction with the other segments. South-bound unilateral propagation is slightly preferable (41 per cent), compared to north-bound unilateral and bilateral propagation modes. Although, these rupture scenarios cannot be directly translated into probabilities of occurrence, they do provide a better insight as to which rupture scenarios are more likely, an important element to better estimate near-field strong ground motion and seismic hazard.     

Modelling the effect of atmospheric pressure variations on gravity

Summary. The 1973 Hawaii earthquake occurred north of Hilo, at a depth of 40 to 50km. The location was beneath the east flank of Mauna Kea, a volcano dormant historically, but active within the last 4000 yr. Aftershocks were restricted to a depth of 55–35km. The event and its aftershock sequence are located in an area not normally associated with the seismicity of the Mauna Loa and Kilauea calderas. The earthquake was a double event, the epicentres trending NE-SW. The events were of similar size and faulting mechanism. The fault plane solutions obtained by seismic waveform analysis are a strike-slip fault striking EW and dipping 55° S, the auxiliary plane a NS vertical plane with a faulting plunge of 35°. The axis of maximum compressive stress is aligned with the direction of the gravity gradient associated with the island of Hawaii. The fault plane striking EW parallels a surface feature, the Mauna Kea east rift zone. The earthquakes were clearly not associated with volcanic activity normally associated with Mauna Loa and Kilauea and may indicate a deep seated prelude to a resumption of activity at Mauna Kea.     

Evidence for different scaling of earthquake source parameters for large earthquakes depending on faulting mechanism

Scaling relationships between seismic moment, rupture length, and rupture width have been examined. For this purpose, the data from several previous studies have been merged into a database containing more than 550 events. For large earthquakes, a dependence of scaling on faulting mechanism has been found. Whereas small and large dip-slip earthquakes scale in the same way, the self-similarity of earthquakes breaks down for large strike-slip events. Furthermore, no significant differences in scaling could be found between normal and reverse earthquakes and between earthquakes from different regions. Since the thickness of the seismogenic layer limits fault widths, most strike-slip earthquakes are limited to rupture widths of between 15 and 30 km while the rupture length is not limited. The aspect ratio of dip-slip earthquakes is similar for all earthquake sizes. Hence, the limitation in rupture width seems to control the maximum possible rupture length for these events. The different behaviour of strike-slip and dip-slip earthquakes can be explained by rupture dynamics and geological fault growth. If faults are segmented, with the thickness of the seismogenic layer controlling the length of each segment, strike-slip earthquakes might rupture connected segments more easily than dip-slip events, and thus could produce longer ruptures than dip-slip events of the same width     

Post-seismic reloading and temporal clustering on a single fault

Geological studies show evidence for temporal clustering of large earthquakes on individual fault systems. Since post-seismic deformation due to the inelastic rheology of the lithosphere may result in a variable loading rate on a fault throughout the interseismic period, it is reasonable to expect that the rheology of the non-seismogenic lower crust and mantle lithosphere may play a role in controlling earthquake recurrence times. We study this phenomenon using a 2-D, finite element method continuum model of the lithosphere containing a single strike-slip fault. This model builds on a previous study using a 1-D spring-dashpot-slider analogue of a single fault system to study the role of Maxwell viscoelastic relaxation in producing non-periodic earthquakes. In our 2-D model, the seismogenic portion of the fault slips when a predetermined yield stress is exceeded; stress accumulated on the seismogenic fault is shed to the viscoelastic layers below and recycled back to the seismogenic fault through viscoelastic relaxation. We find that random variation of the fault yield stress from one earthquake to the next can cause the earthquake sequence to be clustered; the amount of clustering depends on a non-dimensional number, W , called the Wallace number defined as the standard deviation of the randomly varied fault yield stress divided by the effective viscosity of the system times the tectonic loading rate. A new clustering metric based on the bimodal distribution of interseismic intervals allows us to investigate clustering behaviour of systems over a wide range of model parameters and those with multiple viscoelastic layers. For models with   W ≥ 1  clustering increases with increasing W , while those with   W ≤ 1  are unclustered, or quasi-periodic.     

Quantitative mapping of a precursory seismic quiescence to the Izu–Oshima 1990 (M6.5) earthquake, Japan

A highly significant seismic quiescence with a standard deviate Z = 10.1. corresponding to a 99 per cent confidence level, lasted from 1987.7 up to the 1990 February 20 Izu-Oshima M 6.5 earthquake. The quiescent volume had dimensions of 30 km N-S and 10 km E-W and was centred below 14 km depth. Within the recently upgraded seismograph network of the Earthquake Research Institute (ERI), this main shock was the only one with a magnitude M > 5.8 in the upper 30 km of the crust for which the precursory quiescence hypothesis could be tested. Within a radius of 50 km, and during the observation period (1983.5–1995.9), there were no other 1.5 yr or longer periods of quiescence that were rated Z > 6.5 in the declustered earthquake catalogue, except one that was associated with volcanic activity. The total space-time covered by alarms, including the volcanic one, was less than 1 per cent at the Z = 6.5 level. The rarity of highly significant episodes of quiescence, and the correlation in space and time suggest that a precursory seismic quiescence started 2.5 yr before the Izu-Oshima 1990 earthquake in its source volume and to the north of it, and that it can be recognized with an alarm level of Z = 6.0, generating no false alarms. During the 1.5 yr quiescence window, only 10 earthquakes occurred in the quiet volume, whereas 50 events were expected based on the rate seen at other times. In randomly selected volumes containing 50, 100 and 200 events, the anomaly scored Z = 6.1 to 10.1. On the basis of the data from May 1983 to 1995, there is no highly significant quiescence currently present in the Izu-Oshima area.     

Dynamical rupture process on a three-dimensional fault with non-uniform frictions and near-field seismic waves

Summary. Dynamical rupture process on the fault is investigated in a quasi-three-dimensional faulting model with non-uniform distributions of static frictions or the fracture strength under a finite shearing pre-stress. The displacement and stress time functions on the fault are obtained by solving numerically the equations of motion with a finite stress—fracture criterion, using the finite difference method.
If static frictions are homogeneous or weakly non-uniform, the rupture propagates nearly elliptically with a velocity close to that of P waves along the direction of pre-stress and with a nearly S wave velocity in the direction perpendicular to it. The rise time of the source function and the final displacements are larger around the centre of the fault. In the case when the static frictions are heavily non-uniform and depend on the location, the rupture propagation becomes quite irregular with appreciably decreased velocities, indicating remarkable stick-slip phenomena. In some cases, there remain unruptured regions where fault slip does not take place, and high stresses remain concentrated up to the final stage. These regions could be the source of aftershocks at a next stage.
The stick—slip faulting and irregular rupture propagation radiate high-frequency seismic waves, and the near-field spectral amplitudes tend to show an inversely linear frequency dependence over high frequencies for heavily non-uniform frictional faults.     

Source history of the 1905 great Mongolian earthquakes (Tsetserleg, Bolnay)

Two great Mongolian earthquakes, Tsetserleg and Bolnay, occurred on 1905 July 9 and 23. We determined the source history of these events using body waveform inversion. The Tsetserleg rupture (azimuth N60°) correspond to a N60° oriented branch of the long EW oriented Bolnay fault.
Historical seismograms recorded by Wiechert instruments are digitized and corrected for the geometrical deformation due to the recording system. We use predictive filters to recover the signals lost at the minute marks.
The total rupture length for the Tsetserleg earthquake may reach up to 190 km, in order to explain the width of the recorded body waves. This implies adding 60 km to the previously mapped fault. The rupture propagation is mainly eastward. It starts at the southwest of the central subsegment, showing a left lateral strike-slip with a reverse component. The total duration of the modelled source function is 65 s. The seismic moment deduced from the inversion is 10²¹ N m, giving a magnitude   M _w = 8  .
The nucleation of the Bolnay earthquake was at the intersection between the main fault (375 km left lateral strike-slip) and the Teregtiin fault (N160°, 80 km long right lateral strike-slip with a vertical component near the main fault). The rupture was bilateral along the main fault: 100 km to the west and 275 km to east. It also propagated 80 km to the southeast along the Teregtiin fault. The source duration was 115 s. The moment magnitude Mw varies between 8.3 and 8.5.
The nucleation and rupture depths remain uncertain. We tested three cases: (1) nucleation and rupture depth limited to the seismogenic zone; (2) nucleation in the seismogenic zone and rupture propagation going to the base of the crust and (3) nucleation within the crust–upper mantle interface and rupture propagation within the upper mantle.     

High-frequency radiation from crack (stress drop) models of earthquake faulting

We present a theory for the radiation of high-frequency waves by earthquake faults. We model the fault as a planar region in which the stress drops to the kinematic friction during slip. This model is entirely equivalent to a shear crack. For two-dimensional fault models we show that the high frequencies originate from the stress and slip velocity concentrations in the vicinity of the fault's edges. These stress concentrations radiate when the crack expands with accelerated motion. The most efficient generation of high-frequency waves occurs when the rupture velocity changes abruptly. In this case, the displacement spectrum has an ω^-2 behaviour at high frequencies. The excitation is proportional to the intensity of the stress concentration near the crack tips and to the change in the focusing factor due to rupture velocity. We extend these two-dimensional results to more general three-dimensional fault models in the case when the rupture velocity changes simultaneously on the rupture front. Results are similar to those described for two-dimensional faults. We apply the theory to the case of a circular fault that grows at constant velocity and stops suddenly. The present theory is in excellent agreement with a numerical solution of the same problem.
Our results provide upper bounds to the high-frequency radiation from more realistic models in which rupture velocity does not change suddenly. The ω^-2 is the minimum possible decay at high frequencies for any crack model of the source.     

Dynamic stress variations due to shear faults in a plane-layered medium

A complete set of expressions is presented for the computation of elastic dynamic stress in plane-layered media. We use a discrete-wavenumber reflectivity method to compute the stress field radiated by arbitrary moment-tensor sources. The expressions derived here represent an interesting tool for both-the observational and theoretical analysis of dynamic stress changes associated with earthquake phenomena. Dynamic stress changes associated with a strike-slip fault having unilateral rupture are shown. This modelling, which is similar to the 1992 Landers California earthquake, illustrates the effects of distance, directivity and depth on transient stress changes.     

A possible rupture process of slow earthquakes on a frictional fault

Summary. A possible mechanism for the occurrence of slow earthquakes is investigated by calculating numerical solutions for the dynamical rupture process on a quasi-three-dimensional fault with heterogeneous frictional strengths. Experimental friction laws for the dependence of sliding frictional stress on slip velocity, which are based on the cohesive properties of fault asperities, are taken into considerations.
It is found that the applied stress does not drop very rapidly with time and the rupture velocity remarkably decreases as the dependence on slip-velocity becomes smaller. These deceleration effects for the rupture propagation are greatly enhanced with increasing heterogeneities in the distribution of frictional strength and as the initial shear stress has lower levels with respect to the average strength. For these cases, the growth of rupture is extremely slow in a nucleus region with the dimension as large as 10 times the initial rupture length, and gains a terminal velocity dependent on the above factors. The displacement-time function becomes noticeably extended in these cases, and indicates a stick—slip-like phenomena in the extended time interval for a strongly heterogeneous fault.
It seems that these results could explain the characteristic features of slow earthquakes.     

The Zihuatanejo, Mexico, earthquake of 1994 December 10 (M= 6.6): source characteristics and tectonic implications

An analysis of the Zihuatanejo, Mexico, earthquake of 1994 December 10 ( M = 6.6), based on teleseismic and near-source data, shows that it was a normal-faulting, intermediate-depth ( H = 50 ± 5 km) event. It was located about 30 km inland, within the subducted Cocos plate. The preferred fault plane has an azimuth of 130°, a dip of 79° and a rake of −86°. The rupture consisted of two subevents which were separated in time by about 2 s, with the second subevent occurring downdip of the first. The measured stress drop was relatively high, requiring a Δσ of about a kilobar to explain the high-frequency level of the near-source spectra. A rough estimate of the thickness of the seismogenic part of the oceanic lithosphere below Zihuatanejo, based on the depth and the rupture extent of this event, is 40 km.
This event and the Oaxaca earthquake of 1931 January 15 ( M = 7.8) are the two significant normal-faulting, intermediate-depth shocks whose epicentres are closest to the coast. Both of these earthquakes were preceded by several large to great shallow, low-angle thrust earthquakes, occurring updip. The observations in other subduction zones show just the opposite: normal-faulting events precede, not succeed, updip, thrust shocks. Indeed, the thrust events, soon after their occurrence, are expected to cause compression in the slab, thus inhibiting the occurrence of normal-faulting events. To explain the occurrence of the Zihuatanejo earthquake, we note that the Cocos plate, after an initial shallow-angle subduction, unbends and becomes subhorizontal. In the region of the unbending, the bottom of the slab is in horizontal extension. We speculate that the large updip seismic slip during shallow, low-angle thrust events increases the buckling of the slab, resulting in an incremental tensional stress at the bottom of the slab and causing normal-faulting earthquakes. This explanation may also hold for the 1931 Oaxaca event.     

Progressive failure on the North Anatolian fault since 1939 by earthquake stress triggering

10 M ≥ 6.7 earthquakes ruptured 1000 km of the North Anatolian fault (Turkey) during 1939–1992, providing an unsurpassed opportunity to study how one large shock sets up the next. We use the mapped surface slip and fault geometry to infer the transfer of stress throughout the sequence. Calculations of the change in Coulomb failure stress reveal that nine out of 10 ruptures were brought closer to failure by the preceding shocks, typically by 1–10 bar, equivalent to 3–30 years of secular stressing. We translate the calculated stress changes into earthquake probability gains using an earthquake-nucleation constitutive relation, which includes both permanent and transient effects of the sudden stress changes. The transient effects of the stress changes dominate during the mean 10 yr period between triggering and subsequent rupturing shocks in the Anatolia sequence. The stress changes result in an average three-fold gain in the net earthquake probability during the decade after each event. Stress is calculated to be high today at several isolated sites along the fault. During the next 30 years, we estimate a 15 per cent probability of a M ≥ 6.7 earthquake east of the major eastern centre of Ercinzan, and a 12 per cent probability for a large event south of the major western port city of Izmit. Such stress-based probability calculations may thus be useful to assess and update earthquake hazards elsewhere.     

Dynamic rupture and stress change in a normal faulting earthquake in the subducting Cocos plate

A large nearly vertical, normal faulting earthquake ( M _w = 7.1) took place in 1997 in the Cocos plate, just beneath the ruptured fault zone of the great 1985 Michoacan thrust event ( M _w = 8.1). Dynamic rupture and resultant stress change during the 1997 earthquake have been investigated on the basis of near-source strong-motion records together with a 3-D dynamic model.
Dynamically consistent waveform inversion reveals a highly heterogeneous distribution of stress drop, including patch-like asperities and negative stress-drop zones. Zones of high stress drop are mainly confined to the deeper, southeastern section of the vertical fault, where the maximum dynamic stress drop reaches 280 bars (28 MPa). The dynamically generated source time function varies with location on the fault, and yields a short slip duration, which is caused by a short scalelength of stress-drop heterogeneities. The synthetic seismograms calculated from the dynamic model are generally consistent with the strong-motion velocity records in the frequency range lower than 0.5 Hz.
The pattern of stress-drop distribution appears, in some sense, to be consistent with that of coseismic changes in shear stress resulting from the 1985 thrust event. This consistency suggests that the stress transfer from the 1985 event to the subducting plate could be one of the possible mechanisms that increased the chance of the occurrence of the 1997 earthquake.     

Dynamic stress field of a kinematic earthquake source model with k-squared slip distribution

Source models such as the k -squared stochastic source model with k -dependent rise time are able to reproduce source complexity commonly observed in earthquake slip inversions. An analysis of the dynamic stress field associated with the slip history prescribed in these kinematic models can indicate possible inconsistencies with physics of faulting. The static stress drop, the strength excess, the breakdown stress drop and critical slip weakening distance D _c distributions are determined in this study for the kinematic k -squared source model with k -dependent rise time. Several studied k -squared models are found to be consistent with the slip weakening friction law along a substantial part of the fault. A new quantity, the stress delay, is introduced to map areas where the yielding criterion of the slip weakening friction is violated. Hisada's slip velocity function is found to be more consistent with the source dynamics than Boxcar, Brune's and Dirac's slip velocity functions. Constant rupture velocities close to the Rayleigh velocity are inconsistent with the k -squared model, because they break the yielding criterion of the slip weakening friction law. The bimodal character of D _c / D _tot frequency–magnitude distribution was found. D _c approaches the final slip D _tot near the edge of both the fault and asperity. We emphasize that both filtering and smoothing routinely applied in slip inversions may have a strong effect on the space–time pattern of the inferred stress field, leading potentially to an oversimplified view of earthquake source dynamics.     

Simulations of strong ground motion for earthquakes in the Mexicali–Imperial Valley region

Summary. In this paper computer modelling is used to test simple approximations for simulating strong ground motions for moderate and large earthquakes in the Mexicali–Imperial Valley region. Initially, we represent an earthquake rupture process as a series of many independent small earthquakes distributed in a somewhat random manner in both space and time along the rupture surface. By summing real seismograms for small earthquakes (used as empirical Green's functions), strong ground motions at specific sites near a fault are calculated. Alternatively, theoretical Green's functions that include frequencies up to 20 Hz are used in essentially similar simulations. The model uses random numbers to emulate some of the non-deterministic irregularities associated with real earthquakes, due either to complexities in the rupture process itself and/or strong variations in the material properties of the medium. Simulations of the 1980 June 9 Victoria, Baja California earthquake ( M _L= 6.1) approximately agree with the duration of shaking, the maximum ground acceleration, and the frequency content of strong ground motion records obtained at distances of up to 35 km for this moderate earthquake. In the initial stages of modelling we do not introduce any scaling of spectral shape with magnitude, in order to see at what stage the data require it. Surprisingly, such scaling is not critical in going from M = 4–5 events to the M = 6.1 Victoria earthquake. However, it is clearly required by the El Centro accelerogram for the Imperial Valley 1940 earthquake, which had a much higher moment ( Ms ∼ 7). We derive the spectral modification function for this event. The resulting model for this magnitude ∼ 7 earthquake is then used to predict the ground motions at short distances from the fault. Predicted peak horizontal accelerations for the M ∼ 7 event are about 25–50 per cent higher than those observed for the M = 6.1 Victoria event.     

Dynamic nucleation process of shallow earthquake faulting in a fault zone

Two distinct phases are commonly observed at the initial part of seismograms of large shallow earthquakes: low-frequency and low-amplitude waves following the onset of a P wave ( P ₁) are interrupted by the arrival of the second impulsive phase P₂ enriched with high-frequency components. This observation suggests that a large shallow earthquake involves two qualitatively different stages of rupture at its nucleation.
We propose a theoretical model that can naturally explain the above nucleation behaviour. The model is 2-D and the deformation is assumed to be anti-plane. A key clement in our model is the assumption of a zone in which numbers of pre-existing cracks are densely distributed; this cracked zone is a model for the fault zone. Dynamic crack growth nucleated in such a zone is intensely affected by the crack interactions, which exert two conflicting effects: one tends to accelerate the crack growth, and the other tends to decelerate it. The accelerating and decelerating effects are generally ascribable to coplanar and non-coplanar crack interactions, respectively. We rigorously treat the multiple interactions among the cracks, using the boundary integral equation method (BIEM), and assume the critical stress fracture criterion for the analysis of spontaneous crack propagation.
Our analysis shows that a dynamic rupture nucleated in the cracked zone begins to grow slowly due to the relative predominance of non-coplanar interactions. This process radiates the P₁ phase. If the crack continues to grow, coalescence with adjacent coplanar cracks occurs after a short time. Then, coplanar interactions suddenly begin to prevail and crack growth is accelerated; the P₂ phase is emitted in this process. It is interpreted that the two distinct phases appear in the process of the transition from non-coplanar to coplanar interaction predominance.