Section: New Results
Adaptive Algorithms for Orbital-Free Density Functional Theory
Participants : Francois Rousse, Stephane Redon.
The SAMSON App developed to simulate molecular systems with an adaptive version of OF-DFT has been continued. It has been tested on several small systems : atoms, dimers, etc. The errors found on the energies and the bond length found were coherent with the predictive characteristics of OF-DFT and with other OF-DFT softwares like PROFESS.
The pseudopotentials computed by the Carter Group of Princeton (who developed PROFESS) have been implemented in the SAMSON App. The electronic densities became smoother and the predictions were improved, but it restricted the applicability of the SAMSON App since the pseudopotentials were computed only for the elements of the columns III (like aluminum) and V (like Potassium) of the periodic table.
Several optimization algorithms have been tried : projected gradient, Primal-Dual, Lagrangian multiplier improved with a penalization, different nonlinear conjugate gradient minimization algorithms ... None of them showed a clear superiority on the other in both stability and speed. Currently, we use the projected gradient since it is the most stable.
We have implemented an interaction model in SAMSON based on the OF-DFT code and tested its ability to predict the geometry of system on a small crystal of aluminum. The crystal contracted itself, which is coherent with the OF-DFT theory, since it tends to underestimate bond lengths, and with the surface tension, since it tends to minimize the surface of the system. The next step will be to make this interaction model adaptive and measure how much time is gained.