The trenches of oceanic crust interact with the convecting mantle. At Sumatra the pattern looks like this (viewed from south):
The oceanic crust starts bending over the division line between rolls A and B. The crust bends. The outer arc forms over B. The oceanic crust material is then further subducted and enters the down-welling part of convection roll C, and enters the up-welling part between C and D, ending down below the volcano.
Between C and D, geoscientists have already found out that up-welling convection must be taking place underneath the back-arc basin.
This is not a scientific report, as the section is not to scale and no relevant measurements have been made. It is the methodology of comparing convection currents with surface features that is important.
This is based on Wikipedia https://en.wikipedia.org/wiki/Sumatra_Trench
The large scale convection does fix the point between F and G very accurately. The convection rolls have to be of equal height and width (Rayleigh-Bénard convection). The arrangement of the rolls is therefore not disturbed by the interaction between oceanic crust and rolls C and D.