To begin addressing this problem, we have developed the bu ered-force QM/MM (bf-QM/MM) method based on a similar approach used for solid state systems [6,7]. e bf-QM/MM method allows for adaptivity of the set of QM atoms, letting functional groups and molecules move into and out of the QM region. e challenge in adaptive simulations is to prevent the inevitable errors at the interface between the two methods from building up over time by systematically transporting atoms into or out of the QM region. By reducing the errors in the forces on atoms near the interface, we reduce the size of artifacts in observables computed from the molecular dynamics trajectories. e resulting method produces stable, long-running simulations with adaptively changing QM regions. We note that we do not present dynamic simulation results for proteins here, only an initial evaluation of the performance and accuracy of the bf-QM/MM method when applied to small solutes in water. For proteins we only present calculations of force errors at the center of the QM region for congurations taken from an equilibrium MM trajectory as a function of QM region size, to show the e ect of the QM-MM interface in this type of system and to guide the selection of appropriate QM regions in future work.