Check that all implementions set same kpoints mesh at least for "precise" protocol.

[bosonie]

Consensus have been reach that the "converged" set of kpoints should be independent on the method and code. Agreement on the kpoints used have been reach just before the submission of the first paper, however a check is needed and the choice must be translated to the new codes. I will self assign myself to monitor the status of this issue.

[PhilippRue]

So would this mean we take the 'precise' parameter, then run a series of calculations for different k-points and see how well the energy is converged with respect to, for example, the k-point density?

If we tried this for at least one representative example we should be able to get a feeling if a similar k-point density makes sense in the different algorithms and implementations.

[bosonie]

Sorry @PhilippRue, I should have clarified what the decision was. The decision we took is just to impose always a kpoints distance of 0.1 1/Ang for the "precise" protocol. So we fix the density without convergence check. This choice was surely enough to have converged calculation for the crystal elements. I would expect this is enough also for the oxides. In addition, the common interface also defines an input electronic_type and one might thing to "reduce" the density for "INSULATOR" calculations. We will see how it goes. We can always re-discuss this choice if needed.

[bosonie]

Apparently the kpoints mesh with distance of 0.1 1/Ang is not sufficiently dense for the calculations on oxides. The abinit and quantum-espresso team found out that kpoints distance of 0.04 1/Ang is required to have converged results for several cases (together with smearing of 0.005 Ry). We should then review our choice and see if it is feasible for all codes to double the kpoints in the stringent protocol.