Skip to content

1.6 Converge the quantity you intend to publish

The inputs on this page are educational templates that have not been executed for this course. Use an authorized installation, verify version-specific options and establish your own convergence.

1.6.1 Four knobs answer four questions

ecutwfc controls orbital basis, ecutrho density/potential representation including augmentation requirements, k mesh Brillouin sampling and conv_thr electronic stopping. Tightening only the last can precisely solve an under-resolved model. Choose the property and quantitative acceptance rule first; paper settings using a different UPF are not a transferable certificate.

1.6.2 A staged protocol

A — Wavefunction basis. Use a preliminary dense mesh and scan around/above the UPF recommendation. 30/40/50/60/70 Ry is only an example if appropriate; hard data may require a different range. Initially use suitable dual-cutoff recommendations. If ecutrho changes alongside ecutwfc, call it a coupled scan, not an isolated test.

B — Density representation. Fix selected ecutwfc and increase ecutrho separately, checking forces/stress as well as energy. A familiar ratio is only a starting convention. Augmentation and observable determine sufficient resolution; large untested ratios waste resources and small assumed ratios risk error.

C — Sampling. Fix both cutoffs and compare \(4^3,6^3,8^3,10^3\) for this cell, recording shifts. Even/odd or shifted/unshifted grids need not vary monotonically. For anisotropic cells use reciprocal dimensions, not equal integers by habit.

D — Electronic threshold. At the candidate basis/mesh try conv_thr=1d-6, 1d-8 and 1d-10 Ry. Confirm actual convergence and changes in the observable. This is a total-system electronic error criterion, not automatic per-atom uncertainty. Cross-check stronger combinations because errors can interact. PW input reference.

1.6.3 Decide using data, not an invented plateau

For two atoms:

delta_meV_per_atom = abs(E_test_Ry - E_reference_Ry) * 13.605693 * 1000 / 2

Identify a densest successful provisional reference and test beyond the apparent plateau. Record cutoff pair, mesh/shift, threshold, energy convention, property and success state in CSV. A failed/missing point remains failed/missing, not zero. Plot values and differences. Converged energy alone does not establish equilibrium lattice accuracy; stress has its own sensitivity.

1.6.4 Metals and smearing form another branch

Do not transplant silicon fixed occupations to a metal. Investigate empty bands, k density and occupation broadening together. pw.x degauss is Ry: 0.01 Ry is about 0.136 eV, not 0.01 eV. Broadening changes energy-related quantities while helping iteration; document scheme, width and dependence. It is not universally a physical electronic temperature unless the occupation model and interpretation justify that statement.

1.6.5 Exercise and sources

Write quantity, tolerance, reference and stronger test points before running. Design pressure as well as energy tests; explain their difference. Submit plots only from observed values, or a blank planned protocol. PARADIM convergence lessons; SSSP property-aware validation research. These give further study rather than substitute numerical data.


Previous:Complete silicon SCF calculation · Next:Relax atoms and cells with distinct criteria