Many plate tectonic processes, such as subduction initiation, are embedded in long-term (> 100 Myr) geodynamic cycles often involving subsequent phases of extension, cooling without plate deformation and convergence. However, the impact of upper mantle convection on lithosphere dynamics during such long-term cycles is still poorly understood. We have designed two-dimensional upper mantle- scale (down to a depth of 660 km) thermo-mechanical numerical models of coupled lithosphere–mantle deformation. We consider visco–elasto–plastic deformation including a combination of diffusion, dislocation and Peierls creep law mechanisms. Mantle densities are calculated from petrological phase diagrams (Perple_X) for a Hawaiian pyrolite. Our models exhibit realistic Rayleigh numbers between 1e6 and 1e7, and the model temperature, density and viscosity structures agree with geological and geophysical data and observations. We tested the impact of the viscosity structure in the asthenosphere on upper mantle convection and lithosphere dynamics. We also compare models in which mantle convection is explicitly modelled with models in which convection is parameterized by Nusselt number scaling of the mantle thermal conductivity. Further, we quantified the plate driving forces necessary for subduction initiation in 2D thermo- mechanical models of coupled lithosphere–mantle deformation. Our model generates a 120 Myr long geodynamic cycle of subsequent extension (30 Myr), cooling (70 Myr) and convergence (20 Myr) coupled to upper mantle convection in a single and continuous simulation. Fundamental features such as the formation of hyperextended margins, upper mantle convective flow and subduction initiation are captured by the simulations presented here. Compared to a strong asthenosphere, a weak asthenosphere leads to the following differences: smaller value of plate driving forces necessary for subduction initiation (15 TN/m instead of 22 TN/m) and locally larger suction forces. The latter assists in establishing single-slab subduction rather than double-slab subduction. Subduction initiation is horizontally forced, occurs at the transition from the exhumed mantle to the hyperextended passive margin and is caused by thermal softening. Spontaneous subduction initiation due to negative buoyancy of the 400 km wide, cooled, exhumed mantle is not observed after 100 Myr in model history. Our models indicate that long-term lithosphere dynamics can be strongly impacted by sub-lithosphere dynamics. The first-order processes in the simulated geodynamic cycle are applicable to orogenies that resulted from the opening and closure of embryonic oceans bounded by magma-poor hyperextended rifted margins, which might have been the case for the Alpine orogeny.