Section: New Results

Towards parallel adaptive molecular simulations

Participants : Krishna Kant Singh, Benjamin Bouvier, Jean-Francois Mehaut, Stephane Redon.

The adaptive algorithms that we are developing have two main components. The first component determines when and how degrees of freedom can be deactivated and reactivated during a simulation. The second component takes advantage of the frozen degrees of freedom to accelerate the calculation of the potential energy and interatomic forces. Indeed, the potential energy and forces can often be expressed as a (potentially complex) sum of terms which only depend on relative atomic positions. When the relative positions do not change, it is not necessary to update the corresponding terms, which reduces the computation time. We have shown that it is possible to significantly speed up simulations using this approach, while being able to recover static equilibrium statistics [11] .

We have now begun to study the possibility of developing adaptive parallel simulation algorithms, and have begun to review and benchmark popular simulation packages (GROMACS, NAMD, OpenMM, etc.), depending on the number of atoms, the number of available cores, etc.