Define a comparison that answers the paper
For a phase-energy comparison, converge the energy difference using comparable cells and normalization. For relaxation, inspect force and stress stability. For a band edge, transport coefficient, or phonon mode, converge that observable directly.
A stable total energy does not imply every response property has converged. Keep the target tolerance in your project description or method notes.
Run a controlled series
- Duplicate a reference setup and give each variant a name containing the varied parameter and unit.
- Vary ecutwfc with a compatible ecutrho. Keep geometry, pseudopotentials, k mesh, spin, occupations, and SCF threshold fixed.
- At a sufficiently high cutoff, vary the uniform k mesh. Inspect smearing sensitivity independently for metals.
- For low-dimensional systems, vary actual vacuum separation without changing the slab geometry or in-plane cell.
- Run the necessary combinations near the chosen settings to check that separately converged choices remain sufficient together.
- Retain every run and calculate the difference from the most resolved reference. Export the observations as a numerical table for plotting.
Record a minimal results table
Replace placeholders with observed values; the table is a recording template, not simulated data. For the same composition and cell, delta per atom = (E - E_reference) / atom_count. Convert Ry to the unit reported in the paper consistently.
Do not subtract raw total energies of unrelated compositions as if they were a formation energy. Such comparisons require defined reference chemical potentials and stoichiometry.
cutoff_Ry,k_mesh,energy_Ry,atoms,delta_meV_per_atom,run_id
40,4x4x4,<measured>,2,<derived>,<run-id>
50,4x4x4,<measured>,2,<derived>,<run-id>
60,4x4x4,<measured>,2,<derived>,<run-id>Choose a stopping rule
| Observable | Inspect | Frequent mistake |
|---|---|---|
| Energy difference | Change between successively more resolved calculations | Choosing a universal threshold without regard to the effect being studied. |
| Forces and stress | Largest relevant component and equilibrium geometry | Relaxing at low cutoff, then assuming a high-cutoff single point fixes the geometry. |
| Band extrema | k sampling and the actual location of extrema | Assuming the band gap always lies on the plotted high-symmetry path. |
| Phonons | SCF/DFPT thresholds, k/q meshes, acoustic behavior | Removing negative modes cosmetically before understanding them. |
| Transport | Dense electronic sampling, interpolation, chemical potential and temperature | Treating an interpolated smooth curve as evidence of convergence. |