Slip vectors from thrust earthquakes at subduction zones where convergence is oblique to the trench often point between the directions of relative plate convergence and normal to the trench axis, suggesting that oblique convergence is taken up by partial decoupling. Decoupling means that a component of arc-parallel motion of the leading edge of the upper plate results in less oblique thrusting at the trench. Partial decoupling is modeled by partitioning of oblique convergence into slip on thrust and strike-slip faults that are parallel to the trench and to each other and, starting with a force equilibrium condition, a relationship between the obliquity and the earthquake slip vector orientation is derived. Assuming that either fault slips when shear stress on it reaches a yield stress, oblique slip parallel to the plate vector should occur on the thrust fault when obliquity is smaller than a critical angle. For obliquity at or greater than this angle the stress on the strike-slip fault is large enough to start it slipping, and when both faults are active, the arc-parallel motion of the forearc deflects the slip vector back toward the trench-normal. If we assume that continued slip on either fault occurs at constant stress (but the two faults can be at different stresses), the slip vector will maintain a constant angle relative to the trench-normal even when obliquity is larger than the critical angle. This limiting angle of the slip vector, called &PSgr;max (measured relative to the trench-normal, is simply the arcsine of the ratio of the shear forces resisting slip on the strike-slip and thrust faults.

A consequence is that when the obliquity exceeds &PSgr;max, the slip vectors on the thrust fault are sensitive only to the thrust fault orientation and contain no information about the convergence direction between the plates. Slip vectors at the Java trench southwest of Sumatra show the relationship clearly with &PSgr;max=20¿¿5¿, while slip vectors at the Aleutian trench show the relationship less clearly with &PSgr;max=25¿ to 45¿. The greater angle at the Aleutian trench suggests that the upper plate is stronger in the Aleutian arc (relative to the thrust fault) than in the Sumatran arc, consistent with the Sumatran arc being continental and having a well-developed strike-slip fault while the Aleutian arc is oceanic and without a clear transcurrent fault. Slip vectors at the Philippine trench which, like Sumatra, has a large strike-slip fault inboard of it, tend to stay within 25¿ of the trench-normal when obliquity is as large as 50¿. If obliquity exceeds &PSgr;max and continues to increase along a subduction zone, the rate of motion of the forearc relative to the upper plate will vary with obliquity, in which case the forearc sliver should extend or contract parallel to the arc. From the geometry of modern island arcs, arc-parallel extension should be the more common and has been hypothesized for both Sumatra and the Aleutians on the basis of earthquake slip vectors and for these and other arcs from geological observations. From estimates of &PSgr;max and the arc-parallel gradients in obliquity, arc-parallel strain rates are estimated to be 1 to 3¿10-8/yr for the Sumatran forearc, 2 to 6¿10-8/yr for the Aleutian forearc, and 0.3 to 3¿10-8/yr for the Philippine forearc. Oblique convergence and subsequent arc-parallel extension, if accompanied by crustal thinning, may provide an important yet little appreciated mechanism for bringing high-grade metamorphic rocks to the surface of subduction complexes. ¿ American Geophysical Union 1992