Accurate GW self-energies in a plane-wave basis using only a few empty states:: Towards large systems

The GW approximation to the electronic self-energy yields band structures in excellent agreement with experimental data. Unfortunately, this type of calculation is extremely cumbersome even for present-day computers. The huge number of empty states required both in the calculation of the polarizability and of the self-energy is a major bottleneck in GW calculations. We propose an almost costless scheme, which allows us to divide the number of empty states by about a factor of 5 to reach the same accuracy. The computational cost and the memory requirements are decreased by the same amount, accelerating all calculations from small primitive cells to large supercells.