Where two tectonic plates converge, if one or both of the plates is oceanic lithosphere, a subduction zone will form. An oceanic plate will sink back into the mantle. Remember, oceanic plates are formed from mantle material at midocean ridges. Young oceanic lithosphere is hot and buoyant (low density) when it forms at a midocean ridge. But as it spreads away from the ridge and cools and contracts (becomes denser) it is able to sink into the hotter underlying mantle. Three key features are associated with subduction zones: a deep ocean trench, a volcanic arc on the overriding plate parallel to the trench, and a plane of earthquakes, shallow near the trench and descending beneath and beyond the volcanic arc.
The deep ocean trench occurs where the oceanic plate bends downward for its descent into the mantle. The bending of the lithosphere also produces an outer bulge, seaward of the trench.
Benioff Zones: Shallow focus earthquakes near the deep ocean trenches and in the overriding plate are principally produced by motions on thrust faults, indicating compression (converging plates). A plane of earthquake foci descends beneath the overriding plate. The farther from the trench, the deeper the earthquakes are. This was first recognized by Kiyoo Wadati in Japan in the 1920s. Hugo Benioff studied this in the years after World War II and in 1954 he proposed that the plane of descending earthquakes was the result of the seafloor subducting beneath the continent. These earthquakes of the Benioff zone (or Wadati-Benioff zone) are believe to delineate the upper surface of the descending plate (or slab).
In some cases, for example beneath Japan, near the trench there are two planes of earthquake foci, one at the top of the plate, one partway down into the plate. Apparently, the upper plane is produced by tension in the upper surface of the downward bending plate and the lower plane of earthquakes is produced by compression lower in the bending slab [note: if you bend a beam, the upper surface is stretched while the lower surface is compressed].
The angle of subduction (slope of the Benioff zone) varies from one subduction zone to another. The angle of subduction is largely related to the age and therefore temperature and density of the subducting slab. The rate of convergence is also a factor. The older and colder the slab, the steeper the angle of subduction because the faster it sinks. For example, at the Mariana subduction zone, where old, cold lithosphere of the Pacific plate subducts beneath the Mariana Islands, the subduction angle becomes very steep, nearly vertical. Along sections of the Peru-Chile Trench the subduction angle beneath the Andes Mountains, which are much closer to the East Pacific Rise, is not nearly as steep. In some places where very young lithosphere is being subducted the subduction angle is nearly flat.
Benioff zone earthquakes occur down to depths of around 670 km at some subduction zones. These deep focus earthquakes are intriguing. First, it is interesting that no earthquakes have ever been recorded below the 670 km seismic discontinuity (the boundary between the upper mantle and the denser perovskite lower mantle). Second, earthquakes are normally produced by motion on brittle faults. Rocks ordinarily behave in a brittle fashion only down to 10 to 50 km (depending on the minerals and the temperature). Yet Benioff zone earthquakes occur down to around 670 km. Why? One reason is that subducting crust remains "cold" and brittle well down into the mantle before it gradually is heated by the surrounding mantle. So perhaps subducting crust may remain brittle at intermediate depths, perhaps down to 300 km. But it couldn't remain brittle all the way to 670 km. The proposed cause of deep earthquakes (Harry Green, 1990s) is due to sudden transformation of minerals no longer stable at depth.
Do subducting slabs stop at the base of the upper mantle? Seismic tomography is a method for studying the Earth's interior. Using many earthquakes throughout the globe whose waves pass through a given point in the Earth's interior, the seismic velocity of the rocks at that point can be determined. Any differences in seismic velocity are the result of small differences in density of the rocks. Since the composition of the mantle is all ultramafic, density differences are the result of temperature differences from place to place. So seismic tomography measures temperature variations in the mantle.
Images of the mantle produced in this way show subducting slabs as zones of high velocity (lower temperature rocks) angling downward from their respective trenches. In some subduction zones, these high velocity zones pass through the 670 km seismic discontinuity and continue downward through the lower mantle, in some cases nearly to the core mantle boundary. In other cases the high velocity zone flattens out just above 670 km. Clearly, some slabs pass well down into the lower mantle. And clearly, slabs remain intact for hundreds or thousands of kilometers as they pass through the mantle without melting or somehow mixing with the shallow mantle below the trench.
Volcanic Arcs: Chains of volcanoes, all concurrently active, form on the overriding plate parallel to deep ocean trenches. Volcanic island arcs like the Mariana Islands and Aleutian Islands form where an oceanic plate subducts beneath another oceanic plate. Continental volcanic arcs, like the Andes and Cascades, form near continental margins parallel to deep oceanic trenches. Volcanic arcs form on the overriding plate where the subducting slab (Benioff zone) is around 100 km beneath the surface. Magma continuously forms and collects into diapirs which slowly rise through the mantle wedge above the subducting slab, into the overriding crust, and to the surface, feeding the volcanic arc. Beneath the active volcanic arc lie intrusive igneous rocks formed from magma that didn't make it all the way to the surface before crystallizing.
Generation of Arc Magmas: With the discovery of seafloor spreading and the realization that Benioff (1954) must have been right about subduction of ocean crust at deep ocean trenches, it was proposed that arc magmas might be produced by melting of the subducting ocean crust as a result of shear heating of the crust (this hypothesis got into all of the textbooks as if fact). However it was soon determined that shearing between the downgoing crust and the mantle wedge could not produce enough heat to melt the crust. Moreover, because rocks are poor conductors of heat, the subducting slab (crust and lithospheric mantle) remains colder than the mantle through which it sinks. Also, the detailed composition of arc lavas (and rocks) is geochemically distinct from midocean ridge basalts. This indicates that the mechanism by which arc magmas form is distinct from the decompression melting that produces midocean ridge magmas and that arc magmas are not simply made of melted ocean crust. And remember, the volcanic arc lies where the subducting slab is around 100 km beneath the surface, but Benioff zone earthquakes continue continue to hundreds of kilometers depth and hundreds of kilometers landward from the trench; therefore the crust, or at least the slab has not melted away!
Recall that the basaltic ocean crust contains hydrous minerals like amphiboles, some of which formed by hydrothermal alteration as seawater seeped through hot, fractured, young ocean crust at the midocean ridge. As the subducting ocean crust sinks deeper into the mantle, the pressure increases. At depths of around 100 km beneath the surface, the pressure is great enough for the hydrous minerals to undergo metamorphism. The resulting metamorphic minerals are denser and they don't contain the bonded water. This metamorphic dewatering process liberates water from the descending crust. The water slowly diffuses upward into the overlying wedge of hot mantle. The addition of water to the already hot mantle rocks lowers their melting temperature resulting in partial melting of ultramafic mantle rocks to yield mafic magma. Melting aided by the addition of water or other fluid is called flux melting. Ringwood proposed this model in the 1970s. It is probably somewhat more complicated than this, but metamorphic dewatering of subducting crust and flux melting of the mantle wedge appears to account for most of the magma at subduction zones. In rare locations where the descending plate is very young and hot, the chemistry of the the arc rocks (adakite) appears to indicate that the magma was produced, at least in part, by melting of the subducting ocean crust.
Parent magmas are also modified as they rise. As the rising magma cools, minerals with high melting temperatures begin to form in the magma. Olivine is an early-forming, high temperature mineral. It is denser than the melt and will gradually settle out as the magma continues to ascend if the ascent is not too fast. But the crystallization of olivine uses more magnesium and iron than it does silica, so the remaining melt becomes more felsic as olivine separates. This is called fractional crystallization. Rising, hot, mafic magma may also incorporate some of the country rocks of the crust that it rises through (especially if they are more felsic and therefore have a lower melting temperature. So volcanic arc rocks can range in composition from basalt to rhyolite (mafic to felsic), though diorites (intermediate) are the most abundant. Young island arcs, where magmas rise through normal thickness basaltic ocean crust, are dominated by basalt and basaltic andesite. In mature island arcs like Japan and Indonesia, and continental volcanic arcs like the Andes and Cascades, where rising magma must pass through thicker crust including more felsic rocks, andesites are most abundant, though basalt through rhyolite are erupted.
Other Fetures of Subduction Zones...
subduction complex - accretionary wedge: oceanic sediments scraped off from a subducting oceanic plate and piling up as a series of thrust plates forming a bathymetric high.
forearc basins: the low-lying region between the volcanic arc and the accretionary wedge into which sediments, mostly from the arc, are deposited.
backarc basins - marginal basins: basins lying behind many volcanic island arcs. They frequently comprise young, newly forming oceanic crust with composition similar to midocean ridge basalts. Some backarc basins have lineated marine magnetic anomalies like those that parallel the midocean ridges. But others do not. Several models have been proposed to account for the formation of backarc basins including secondary convection behind the arc produced by the drag of the descending slab, trench suction produced by slab rollback, and abandoned ocean crust as a result in a step-back of the position of subduction.
remnant arcs: The Mariana Islands and Lesser Antiles of the Caribbean are active volcanic island arcs located above a subducting slab. Both have backarc basins. Beyond the backarc basins are extinct volcanic ridges - the remnant of former volcanic island arcs. The site of subduction has apparently stepped back to it current position.