In this contribution, we will present a new discretization and implementation for the Conductor-like screening model (COSMO), a popular continuum solvation model. Such a new discretization is based on Schwarz's domain decomposition and we refer to it as ddCOSMO.
After summarizing some formal aspects of the procedure, we will describe the algorithm and focus on the numerical details and on the modular implementation. We will show, including with some examples, that our implementation scales linearly in computational cost and memory requirements with respect to the size of the solute and allows to treat large and very large systems without significantly increasing the simulation cost with respect to the same system treated in vacuo.
Compared to the most efficient existing implementations, ddCOSMO is two to three orders of magnitude faster, more robust and defined by a smaller number of discretization parameters: these features allow us to extend the applicability of continuum solvation models to large and complex biological and chemical systems in combination with classical, QM and hybrid descriptions of the solute.

Co-authors: Benjamin Stamm, Louis Lagardère, Eric Cancès, Yvon Maday, Jean-Philip Piquemal and Benedetta Mennucci