Section: New Results

Algorithmic speed up of the ARPS method

Participants : Zofia Trstanova, Gabriel Stoltz, Stephane Redon.

Adaptively Restrained Particles Simulations (ARPS) allow to save computational time at each time step since particles do not move and forces need not be updated. The associated gain can be quantified by an algorithmic speed-up factor $S_{\mathrm{algo}}\ge 1$. Intuitively, freezing more particles leads to larger algorithmic speed-ups, but also larger correlations in time.

We analyzed the algorithmic speed up with respect to the standard methods. Since the ARPS algorithm is based on adding and subtracting of the forces between active particles, the gain with respect to the standard method, where only one complete computation of all interactions is performed at each time step, is achieved only if the percentage of restrained particles is big enough. Hence we studied the necessary conditions, under which the computational complexity of the forces updating in the ARPS method is lower than the one of the standard method. This allows to achieve an algorithmic speed up that is always bigger than one.

We also propose a simple strategy for choosing optimal simulation parameters.