We report the results of experiments on the initiation of subduction, using stratified analogue models in a large centrifuge, where the experiments are driven only by the enhanced gravity of the centrifuge without the effect of external lateral stresses. The scaled density of the stratified components resembles that of the asthenosphere and the continental and oceanic lithospheres. The experiments demonstrate that under the effect of enhanced gravity, the layers simulating the oceanic lithosphere detach from the front of the ‘continental lithosphere’ and plunge under it, pushing the more pliable asthenosphere downwards. Simultaneously, the ‘continental lithosphere’ is thrust over the downgoing slab; where friction is low, the ‘continental lithosphere’ extends considerably so that the ‘ductile continental lithosphere’ is exposed in some places. The rate of thrusting of the experimental continental slab over the oceanic one, as well as the amount of extension of the overriding slab and the extent of the rollback of the subduction zone that follows the initial lithospheric detachment, are controlled by friction and density differences between the subducting and the overthrust slabs. The analogue experiments emphasize the role of lateral density variations in incipient subduction and the effect of differential seismic friction along the subduction plane on the evolution of subduction zones, their shape and the evolution of their backarc basins. The morphological resemblance of the experimental results to various subduction systems seems to support their applicability to real subduction systems.
The detachment of the oceanic lithosphere from the continental one during the initial stages of subduction is critical in the development of the structure of the Earth and its chemical differentiation (Anderson 2001). In spite of this significance, initiation of subduction is rare, and the lithospheric structure of potential study areas such as the western Mediterranean off Algeria, the western coast of northern Honshu at the Japan Sea or the Hjort Trench south of New Zealand is not fully understood (Tamaki & Honza 1985; Collot et. al. 1995). Auzende et. al. (1973) suggested that structural inversion and initial subduction might take place off Algeria in the western Mediterranean and Bourgois et. al. (1992) and Déverchère (personal communication, 2004) support this notion. Structural and petrological evidence suggests that the initial subduction of oceanic lithosphere under continental lithosphere is not associated with a change in the mechanical and thermal regimes of the lithospheric slabs, except for the breakup of the lithological link that had welded the slabs together since the early evolution of the oceanic lithosphere (Bloomer et. al. 1995; Stein & Stein 1996). In spite of this, the mature evolution of the oceanic convergent margin leaves profound and prominent crustal marks such as oceanic trenches, subduction zones, volcanic arcs and backarc basins.
Nearly 40 oceanic trenches and subduction zones of oceanic crust are known, most of them located along the rims of the Pacific Ocean; a few are found in the Atlantic and Indian oceans, and the Mediterranean Sea (Jarrard 1986). In spite of the similarity in their configuration, surface signature and structural and petrological processes, the physical characteristics of subduction zones and their ages vary greatly. The length of their surface signature ranges from 100 to 800 km, and their depths—from 40 to 600 km. The dip of their shallow section ranges from 10° to 35°, the velocity of their subduction varies from 1 to 20 cm yr−1, and their age ranges from late Jurassic to the Miocene. Jarrard (1986) also showed a large variability in structural regimes associated with subduction: there is nearly an equal distribution in the number of trenches where the structural regime is extensional, compressional or neutral, irrespective of age or location (Table 1). Karig (1971) first observed an extensional regime in the western Pacific—extension in the overriding lithospheric slabs, and active spreading centres in marginal seas. The existence of backarc extension and the prolonged shrinkage of the Pacific Ocean implies that the encircling subduction zones must travel back, up the plates they consume, and thus some arcs will migrate away from the ocean-bounding continent (Elsasser 1971; Gordon et. al. 1978; Chase 1978). The simultaneous occurrence of zones of extension within a region of plate collision was explained by the concept of trench roll-back (Dewey 1980), and linked to the process of arc migration (Malinverno & Ryan 1986).
It is generally accepted now that the density of the asthenosphere is close to that of the cold oceanic lithosphere, namely approximately (3.2–3.4) × 103 kg m−3 (e.g. Turcotte & Schubert 1982, 2001), with the oceanic lithosphere being slightly denser (e.g. Cloos 1993). This naturally occurring conversion, where the cold and denser oceanic lithosphere overlies the lighter asthenosphere, is gravitationally unstable. Therefore, if, for some reason, the lithosphere were to plunge into the asthenosphere, the thermally driven density differences and the transformation of the basalts and gabbros of the oceanic slab material at about 100 km depth into the denser eclogite phase would sustain the plunging of the penetrating slab deeper into the mantle. Plunging of the increasingly dense slab is opposed by the elastic bending stresses of the oceanic plate, which act to keep the plate floating. McKenzie (1977) calculated that about 120 km of subduction is required before the increasing negative buoyancy of oceanic slabs would start a ‘runaway’ sinking process. Jull and Kelemen (personal communication, 2003) have calculated that a large root of oceanic slab transformed into eclogite, with a depth of approximately 60 km and thickness 120 km, is required before the additional effect of the denser, plunging oceanic lithosphere is sufficient to initiate a subduction instability. Therefore the process by which subduction initiates has remained a puzzling question: what forces cause the initial 100 km of subduction?
After the early suggestion that subduction is derived from some vertical force, which also shapes its commonly arcuate configuration (Frank 1968), was rejected (Tovish & Schubert 1978), ridge push and the negative buoyancy of the oceanic lithosphere were proposed as the primary forces that initiate subduction (see Shemenda 1993; Turcotte & Schubert 1982, 2001, for a review of analogue experiments and the analytical problem, respectively). Cloos (1993) used an analytical approach to show that slight negative buoyancy of the mature oceanic lithosphere would occur when thermal stability of the ocean lithosphere is attained, at 80 Ma, and proposed that subduction would then take place spontaneously. Another model, generated by Regenauer-Lieb et. al. (2001), proposed that subduction could be triggered by slow sedimentary loading. This model presumably follows the example of the Atlantic passive margin, where after steady sediment loading during 100 Myr the loaded lithosphere started to plunge down, and the water in the sediments might enhance the subsidence and start downgoing subduction. However, observations that subduction also takes place in relatively young lithosphere (e.g. Jarrard 1986, Table 1) cast some doubt on the applicability of these models. It appears that lateral density and topographic variations between the continental and oceanic lithospheres, coupled with the ductility of the lower lithosphere, may also play an important role in the initial subduction process.
In the present work we isolated and studied the role of lateral stresses arising at a divergent ocean–continent margin due to lateral density differences between the denser, old oceanic lithosphere and the lighter continental lithosphere, and also due to geometrical effects such as differences in elevation (e.g. Turcotte et. al. 1977; Faccenna et. al. 1999). Faccenna et. al. (1999) showed, in a set of analogue experiments conducted under compressive lateral stresses, that the interplay between the lateral forces acting on plate margins and the ductile strength of the lithosphere exercises an important control on the initiation of subduction: ductility of the lower lithosphere plays a major role in controlling subduction, in contrast with the notion that the stronger brittle part of the lithosphere would be the ‘strength-limiting factor’, and would control the initiation of the subduction. However, in the experiments of Faccenna et. al. (1999) gravity had its natural value, and was not scaled according to the viscous forces that constrained their experiments. In our set of experiments, performed in a centrifuge (Ramberg 1981), both viscous processes and gravitational forces were enhanced to investigate the dynamic interplay between these two major processes—gravitational and viscous. Furthermore, our experiments also investigated the role of friction in constraining the configuration, and possibly the shape, of the subduction zone as it forms (Scholz & Campos 1995).
If the continental lithosphere and the oceanic lithosphere were two immiscible liquids of differing densities poured into a container with a dam separating the two, and then the dam were lifted, what would happen? After some time, the lighter fluid (conceptually the continental lithosphere) would end up overlying the denser fluid (the ‘oceanic lithosphere’), in a two-layered pancake manner. A similar physical process might have taken place at a passive margin but for the high viscosity of both lithospheres. The continental and oceanic lithospheres lie on top of the asthenosphere, which has a lower viscosity than both lithospheres. While continent and ocean are each independently isostatically compensated relative to the asthenosphere, the 200–400 km wide marginal province, where the continental lithosphere tapers to merge with the oceanic lithosphere, is, in essence, gravitationally unbalanced. A body force arises due to lateral density and, in places, elevation contrasts, and its magnitude and direction depends on the geometry of the contact of the lithospheres across the continental margin.
To illustrate this point consider two end-members of margin geometries: in the first case, isostatically compensated continental and oceanic lithospheres are juxtaposed one against the other across a very narrow transitional zone (Fig. 1a). In this margin geometry, the continental crust towers above the oceanic crust. Looking at pressure as a function of depth in the continental lithosphere, one finds that the maximum pressure arises close to the top of the oceanic lithosphere, and is directed so that it drives a ‘continental collapse’ on top of the ocean floor. However, since the maximum pressure is encountered within the brittle lithosphere, the resistance to viscous flow is large in this case.
The other end-member to consider is presented in Fig. 1(b). In this case the continental and oceanic lithospheres have similar thicknesses at their contact zone in the margin, i.e. the continental lithosphere tapers towards the oceanic one, and the two are locked together, confining vertical motion and local isostatic compensation. Let us denote depth as z, the density of the oceanic and continental lithospheres as ρo and ρc, respectively, and gravitational acceleration as g. In this case the pressure at the bottom of the oceanic lithosphere increases with depth by δρgz, and is larger than that at the continental side, because δρ=ρo−ρc is positive on the oceanic side. Since the difference is greatest at the base of the lithosphere, then, in this case, lateral forces drive the deformation in the deeper, and more ductile, part of the lithospheres, on top of the relatively less viscous asthenosphere. If the density-derived body force overcame the strength of the plates, the lithospheres would deform so that the oceanic lithosphere would creep between the continental lithosphere and the asthenosphere, while the continental lithosphere would creep on top of the oceanic lithosphere.
The parameter that characterizes whether deformation processes can occur is the Argand number, Ar (England & McKenzie 1982, 1983), which measures the ratio between stresses driving deformation and the strength of the material, where Ar > 1 is required for deformation to occur. The behaviour of the lithospheres is controlled in our conceptual model by three different Ar numbers, characterizing the ratio between gravity-driven stresses and the different strength and rheology profiles along the lithospheric boundaries:
The first Argand number, Ar1, measures the ratio between the stress induced by the density differences at the bottom of the lithosphere and the ductile strength of the lithosphere (Faccenna et. al. 1999):Figs 2a and b).
The second Argand number, Ar2, measures the ratio between the stress at the bottom of the unsupported brittle continental lithosphere, generated by the difference in elevation between the continental and oceanic lithosphere, and the brittle strength of the lithosphere:Faccenna et. al. 1999), hu is the thickness of the continental brittle lithosphere, which is unsupported by a juxtaposed oceanic lithosphere, and subscript c stands for continent. Ar2 controls the evolution of brittle failure at the upper lithosphere along the interface of the plates (Fig. 2c).
Construction of an integrated Argand number, representing the combined resistance of the lithosphere to deformation by all brittle and ductile modes, is desirable, but not obvious; namely, the total lithospheric resistance to deformation is not simply the sum of different layer strengths integrated over the depth for which they are applicable. This is because the relative importance of the specific Argand numbers depends on the sequence of events that leads to initiation of subduction. This sequence, in turn, depends on the values of the various Argand numbers. To clarify, we suggest two combinations of values of Argand numbers and speculate on the resultant sequence of events:
In case of Ar1 < 1 while both Ar2 and Ar3 > 1, subduction will initiate by sliding along the ocean–continent interface plane. Ductile deformation will be avoided altogether, and thus the ductile strength becomes irrelevant to the possibility of subduction.
In case of both Ar1 and Ar2 > 1 while Ar3 < 1, collapse of the brittle portion of the continental plate allows overthrusting of the continental lithosphere over the oceanic lithosphere. This is augmented by ductile flow of the oceanic lithosphere under the continental lithosphere, both processes contributing to the resultant low-angle subduction. Here no sliding occurs along the fault plane, and low-angle subduction happens by ductile rotation of the fault plane. In this case frictional shear resistance becomes irrelevant and should not be considered.
In order to calculate Argand numbers in the lithosphere we discuss characteristic values for the lithosphere: the density difference between the oceanic and continental lithospheres is approximately 300 kg m−3. The thickness over which this density difference is effective is that of the lithosphere, taken as about 120 km. The viscosity of the ductile portion of the lithosphere is considered to be 1021–1022 Pa s (Davy & Cobbold 1988, 1991). The strain rate that we use is 1–2 cm yr−1. Using these estimates yields values of Ar1 in the range of 1 to 35, ensuring that ductile processes may play a major role in trench initiation.
Another controlling parameter is frictional resistance to sliding; as lithospheric slabs subduct they experience seismic friction, which depends upon several variables in the structure and composition of the lithospheres that slide and grind against each other (Scholz & Campos 1995). Considering the lateral density difference forces expected in the lithosphere, and using normal values of friction (see Faccenna et. al. (1999) for crustal rocks and Scholz (2002) for the mantle) will lead to values for Ar2 of the order of 10−1 and Ar3 of the order of 10−2. Therefore, either additional driving forces or mechanisms for shear strength reduction are required for the values of Ar2 and Ar3 to exceed 1.0.
There are several natural mechanisms which may significantly reduce resistance to sliding in subduction zones: for the upper lithosphere, the coefficient of friction may be reduced due to the occurrence of sediments (Scholz 2002) or serpentinization of peridotites (e.g. Escartin et. al. 2001). Alternatively, the low strength of densely fractured lithospheric slab, where the fracture zones trend normal to the strike of the subduction zone, would also reduce the slab rigidity significantly (e.g. Toth & Gurnis 1998; Hall et. al. 2003). For example, the reduced rate of seismicity off the apex of the Aleutian Arc was associated with a high concentration of fluid-mobile trace elements in the basalts of the central segment of this arc, suggesting the subduction of a fracture zone in which large quantities of sediments had accumulated (Singer et. al. 1996). Another mechanism for reduction of shear resistance in the upper lithosphere is lowering of the effective normal stress due to high pore pressure (Scholz 2002) or by a local extensional regime (Scholz & Campos 1995). The lower lithospheric shear resistance may also be reduced by more complex processes, such as a thermomechanical feedback mechanism (e.g. Branlund et. al. 2001; Regenauer-Lieb et. al. 2001). An additional force resisting subduction is the strength of the lithosphere, although it is generally assumed that the intrinsic strength required to break the rocks is negligible compared with frictional sliding resistance (e.g. Brace & Kohlstedt 1980; McNutt & Menard 1982).
Our experiments also suggest that friction plays a role in controlling the structure and shape of subduction zones. In some experiments we caused non-uniform ‘seismic friction’ along the subduction zone. The experiments show a correlation between the arcuate shape of emerging subduction zones and the variability in modelled interplate friction. Our observations suggest a new cause for the arcuate shape of many subduction zones. One classical explanation for that configuration was given by Frank (1968), who noted that subduction occurs along small circles. Frank remarked on the geometric similarity between the arcuate shape of most subduction zones and the way in which a ping-pong ball yields to vertical compression, and suggested that the resemblance has geological significance. The main reason for the eventual rejection of Frank's model was that the model suggests a correlation between the dimension of the arc and the dip of the deformed crust in ping-pong balls, but such correlation does not occur in nature (Le Pichon et. al. 1973; Tovish & Schubert 1978; Turcotte & Schubert 1982, 2001).
When determining the feasibility of the initiation of subduction, and calculating the Argand numbers, additional forces other than lateral density differences may be added to the driving stresses in the appropriate vectorial direction (see Mueller & Phillips 1991, for overview). These forces are not investigated in our experimental set-up, and are not added here to our calculations of Argand numbers. Such forces include ‘ridge push’ which results from magmatic accretion at the distal oceanic ridge, the negative buoyancy associated with cooling of the oceanic lithosphere, the basal drag that mantle motion exerts under the plates and sediment loading (e.g. Stein et. al. 1989). In our two sets of experiments none of these additional forces was modelled, but since Ar1 and Ar2 exceeded 1 (Tables 2 and 3) subduction was initiated in both cases.
We conducted analogue experiments in order to test the possibility of generating initial subduction under the stress of lateral density variations and gravity alone, without the effect of lateral push. To investigate ductile processes driven by gravity we used a model where viscous forces were scaled according to body stresses, which may be carried out in a centrifuge. We used the large and fast centrifuge of the Hans Ramberg Tectonic Laboratory of the University of Uppsala in Sweden (Ramberg 1981; Mulugeta 1988). In these sets of experiments we also tested the effects of variable friction on the configuration of the subduction plane.
We carried out two sets of experiments, the first comprised a basal homogeneous layer simulating the asthenosphere overlain by a composite layer simulating the oceanic and continental lithospheres. A very thin layer of black powder was introduced into the modelled asthenosphere to show flow patterns during the experiments. The section of the layer that simulated the oceanic lithosphere was built of two segments, one representing a young and relatively light lithosphere and the other an old, cold and dense lithosphere. The density of the old oceanic lithosphere was similar to that of the asthenosphere, and that of the young lithosphere was intermediate between the old oceanic and the continental lithosphere (Fig. 3). Unlike in some previous models, both analytical and analogue, which presumed a priori that subduction starts because the old oceanic lithosphere becomes denser than the underlying asthenosphere, the densities of the modelled old oceanic lithosphere and the asthenosphere were the same in this series of experiments. The friction along the simulated subduction was variable. We applied some Vaseline lubrication to the contact between the ‘old oceanic’ and ‘continental lithospheres’ along the entire contact plane in some experiments and along a part of it in others. The models were deformed at 500g, and the characteristics of this series of experiments are presented in Fig. 2 and Table 2.
The second series of experiments, in the same centrifuge, studied the deformation of a three-layered model simulating the asthenosphere as well as the ductile and brittle continental and oceanic lithospheres. The purpose of this series of experiments was to test the possibility of subducting a relatively light oceanic lithosphere and to further investigate the effects of differential lubrication on both the subducting and the overriding slabs. In addition, the behaviour of a coupled brittle–ductile layer was investigated. The ‘asthenosphere’ in these experiments had a density of 1.78 × 103 kg m−3 and viscosity of 3× 103 Pa s. The experimental ductile oceanic lithosphere had a density of 1.47 × 103 kg m−3, and the ductile continental lithosphere had a density of 1.28 × 103 kg m−3 (Table 3 and Fig. 3). Vaseline was used to lubricate the contact between the slabs. The ‘brittle lithospheres’ were designed to show brittle failure under deformation (Mulugeta 1988). These models were also deformed by a centrifugal gravity acceleration of 500g. The initial pressure difference along the profile of the experiment described in Table 3 takes its form from the model presented in Fig. 1(a), with a maximum pressure difference of 44 kPa at the base of the unsupported continent. As the brittle continental lithosphere collapsed towards the ocean at the start of the experiment the pressure difference profile changed and took the form of the model in Fig. 1(b), with maximum density difference of −9.4 kPa at the base of the lithospheres, where the minus sign represent the larger pressure at the ocean side of the model.
Consequently the Argand numbers of our experiments were calculated using the relevant densities and layer thicknesses (Tables 2 and 3). Since in our experimental set-up we did not impose a specific strain rate, its value was taken as uc/h, where uc is the scaled characteristic convergence velocity of 1.37 × 10−5 m s−1, equivalent to 1 cm yr−1 in reality. In the models where the continental lithosphere rises above the oceanic lithosphere, the values of Ar1 and Ar3 are calculated while ignoring the elevation gradient, assuming that continental collapse has occurred in either brittle deformation when Ar2 > 1 for the three-layer models or by ductile deformation for the two-layer models. Ar2 was calculated for continental failure; thus the brittle strength was taken as the continental strength with a friction coefficient of 0.5 and negligible scaled cohesion. Ar3 was calculated using a friction coefficient of ∼0.1, since the model plane interface was lubricated to simulate the existence of a weak zone.
The primary result of our experiments, conducted in an enhanced gravity field derived from the rotating centrifuge, is that subduction can indeed be initiated by lateral density differences between adjacent lithospheric slabs. We conclude that subduction initiation requires neither externally imposed lateral stresses (‘ridge push’) nor a negatively buoyant oceanic lithosphere. During our experiments we encountered the detachment of the ‘oceanic’ from the ‘continental lithospheres’, and noticed its gliding under the ‘continent’; and the stacked ‘lithospheres’ pushed the denser but more viscous ‘asthenosphere’ downwards. The subduction took place both where the density of the ‘oceanic lithosphere’ was similar to the asthenosphere but also where it was lighter. It seems that the density contrast between the two lithospheres is the prime driving mechanism that initiates the subduction process.
Where friction along the subduction zone was reduced, the overlapping segments of the ‘lithospheres’ stretched, thinned and extended considerably, the stretching overriding plate caused the ‘seaward’ migration of the ‘subduction trench’, and the gliding underthrust slab pushed the underlying ‘asthenosphere’ away (Fig. 4). Where the lubrication was applied to only a segment of the contact zone between the ‘oceanic’ and ‘continental lithospheres’, and not to the entire contact zone, the rate of underthrusting of both the denser slab and the overthrusting of the lighter slab in the segment of reduced friction was much faster than the thrusting in the unlubricated segments, leading to the development of an arcuate subduction zone. These fast subduction processes also led to the development of a backarc basin above the lubricated segment of the thrust fault (Fig. 5).
The experiment series that replicated both the brittle and the ductile lithospheres and an ‘oceanic lithosphere’ that was significantly lighter than the ‘asthenosphere’ showed two patterns of deformation, as a function of the friction along the ‘subduction zone’. Where friction was high, collision occurred between the lithospheres, the slabs were folded and the amount of subduction was small (Fig. 6). However, where lubrication was applied, both ‘lithospheres’ extended considerably as they stretched one above the other, the ‘continental’ over the ‘oceanic’. The transition from the high-friction to the low-friction zone was gradual, and the subduction zone was arcuate—the low-friction zone migrated effectively seawards. Plastic deformation and thinning occurred in the ductile ‘lithospheres’ and a series of faults developed in the brittle ‘lithosphere’ segment of the overthrust slab that overlies the lubricated subduction zone. The faults trended normal to the direction of the extension, and did not replicate the arcuate orientation of the subduction zone. Exhumation of the ‘ductile continental lithosphere’ was encountered in places in the backarc basin that developed near the advancing front of the continental slab (Fig. 7).
Discussion and Geological Implications
The processes that pull a slab of oceanic lithosphere downwards into the asthenosphere and form a subduction zone comprise numerous parameters and various forces, as indicated by analyses of earthquakes associated with subduction zones (e.g. Mueller & Phillips 1991; Scholz & Campos 1995). The considerable variability of magmatic and metamorphic lithologies in many island arcs and backarc basins indicates that numerous factors contribute to the subduction processes (Hawkins 1995; Davidson 1996). The principal forces believed to drive subduction once it is ongoing are the increase in the density of the slab due to eclogitization of the plunging oceanic plate (McKenzie 1977), as well as the ridge push and the thermally driven density differences. Our present work suggests a reason why many subduction zones are formed at continental margins; it is possible that one of the major players in the set of forces that trigger the process which initiates subduction is derived from lateral density differences between the continental and oceanic lithospheres, and differences in elevation across the continental margin. The effects of other driving forces such as ridge push and negative buoyancy would supplement these lateral density variations to aid in overcoming the initial resisting forces, and enable the initiation of subduction.
The reported experiments showed that the simulated oceanic lithosphere subsided and was thrust beneath the lighter simulated continental lithosphere, under the effect of the enhanced gravity of the centrifuge. The ‘oceanic lithosphere’ did not sink into the more pliable, yet denser, ‘asthenosphere’, but squeezed its way under the ‘continental lithosphere’, displacing the ‘asthenosphere’ downwards, and, in places, also sideways. This type of motion is probably the result of the lateral density differences between the ‘continental’ and ‘oceanic lithospheres’ driving ductile deformation, and it demonstrates that, under favourable conditions (i.e. when Ar1 and Ar2 exceed 1), the initiation of subduction requires neither lateral push nor negative buoyancy of the oceanic slab. An open question remains about what happens when the ductile deformation may be active (i.e. Ar1 > 1) but the brittle deformation is inhibited (Ar2 < 1, Ar3 < 1). Since there is always a slow enough strain rate for which Ar1 > 1, we expect some deformation signature at the lower part of the lithosphere, with at least slow rotation of the contact zone towards horizontal orientation, following Fig. 2(d).
The results of our analogue model experiments further confirm the analytical model of Scholz & Campos (1995), which predicted that the rates of subduction and the convergent lateral motion of the ‘oceanic’ and ‘continental’ slabs depend on seismic friction. The experiments show that low friction is associated with fast roll-back of the subduction arc. The roll-back leads to rapid extension of the overriding slab, which, in turn, enhances the extensional growth of the backarc basin. In our experiments, low-angle principal overthrust faults and secondary normal faults characterized the extensional section of the overriding slab in the low-friction segment of the subduction zone. Thinning also was prevalent in the wedge-shaped front of the underthrust slab. The experiments suggest that the roll-back is constrained by low interplate coupling. One could argue further that the subduction of very old and probably very dense oceanic lithosphere slabs, such as Tonga or the Marianas (see Table 1 for details), would subduct irrespective of the seismic friction. However, the eclogitization of the deeper parts of such slabs probably matured long ago, and they cannot be considered as prime examples for initiation of subduction. The experiments show further that as friction increases, ‘seaward’ propagation and extension of the overthrust slab decrease, and so does the motion ‘landward’ of the underthrust slab. Under the constraints of high friction, the structural regime at the boundary between ‘oceanic’ and ‘continental lithospheres’, both the overriding and the underthrusting slabs acquire compressional structures, the principal overthrust fault becomes steeper and folding takes place at the edge of the convergent ‘continent’. We looked for evidence of ascending flow from the simulated asthenosphere into the lithosphere above the subducting slab, which could have been the mechanism of the development of the marginal basin as predicted by previous studies (e.g. Dvorkin et. al. 1993; Chemenda et. al. 1995), but encountered none. It seems that the ‘asthenosphere’ flowed away from the slab that penetrated the contact zone between the lighter continental lithosphere and the denser ‘asthenosphere’. Our experiments support the notion that had previously been suggested by Malinverno & Ryan (1986) and Scholz & Campos (1995) that the subduction of the oceanic lithosphere, the arc roll-back, the tectonic extension at the edge of the converging continental lithosphere and the development of the backarc basin are all parts of the same subduction process. Malinverno & Ryan (1986) indicate further that the extension along the converging continent might be accompanied by exhumation of the lower section of the brittle lithosphere, and possibly even the upper section of the ductile lithosphere.
The requirement of large compressional stress in the initial stages of subduction was presented in analytical models (Mueller & Phillips 1991). Some analogue models of the initiation of lithospheric subduction also pursued this line of thought, and considered the combined effects of ridge push and the negative buoyancy of old oceanic lithosphere as the two principal parameters in the experiments (e.g. Shemenda 1993; Chemenda et. al. 1995). Others combined centrifuge and piston to generate the effects of both horizontal push and vertical pull in their modelling (Mulugeta 1988). Faccenna et. al. (1999) combined both slab pull and ridge push together with inversion of the lithospheric buoyancy and elevation differences at the continental margin. Faccenna et. al. (1999) also scaled their model for gravity by using two non-dimensional numbers, namely Argand number and buoyancy forces. While the above experiments indicated that both the negative buoyancy of the oceanic lithosphere and ridge push may initiate subduction, our experiments show that subduction can be initiated without negative buoyancy and without ridge push. And indeed, had the asthenosphere under the margins of the oceans been gravimetrically metastable numerous peridotitic diapirs would have been expected in these domains.
The centrifuge-driven subduction experiments show that lateral density differences at continental margins can generate viscous flow and frictional sliding of the oceanic lithosphere if the respective Argand numbers are high enough. Our experiments show that the oceanic lithosphere does not necessarily need to be denser than the asthenosphere, and its negative buoyancy is not a prerequisite for generating and initializing subduction. The experiments suggest that in old continental margins the break-up of the bond between the oceanic and continental lithospheres would be followed by creep of the heavier oceanic plate under the lighter continental plate, and thus provide the mechanism for the initial plunge of a few tens of kilometres, and the initiation of subduction. After the initial stages of subduction, thermally driven density differences between the cold oceanic lithosphere and the hotter asthenosphere, and densification due to eclogitization occurring at depth, cause buoyancy inversion, and the oceanic slab would continue to subduct into the asthenosphere because it becomes increasingly denser.
Our experiments also demonstrated the dependence of the configuration of the subduction zone on differential seismic friction. Reduced friction along the subduction zone in our experiments caused structural evolution of subduction arcs, backarc basins, a high rate of trench roll-back and extensional structures in the overthrust slab. Differential reduced friction along the simulated subduction zone generated arcuate thrust planes with the concave side directed towards the continent.
We thank the Institute of Earth Sciences in the University of Uppsala for the permission to use the facilities of Hans Ramberg Tectonic Laboratory for our experiments. We are grateful to Christopher Talbot and Hemin Koyi for their support, advice and encouragement during experimentation, to Christopher Scholz and Jacques Déverchère for sharing with us their knowledge and to Kim Kastens for review of an earlier version of the manuscript. Thorough reviews by Claudio Faccenna and Marc-André Gutscher improved the manuscript considerably. The technical assistance of Einar Meland is appreciated.