Skip to content

2.4 Converge the observable, not a familiar cutoff

These are unexecuted teaching inputs and starting models. Original diagrams are schematics, not calculated results. Validate version-specific syntax, licensed or authorized data, numerical convergence and the scientific model before using this workflow.

2.4.1 Model, units and provenance

PW cutoffs and energies use Ry, common force output uses Ry/bohr, and pressure uses kbar. Geometry cards state their coordinate units. Different executables have distinct grammars and time-unit conventions.

Shared inputs, conventions and evidence

Original schematic: Converge the observable, not a familiar cutoff. No numerical results are claimed.
Original schematic: Converge the observable, not a familiar cutoff. No numerical results are claimed.

2.4.2 Unexecuted inputs and explicit deltas

Use the accompanying instructions to identify the parent calculation and placement of every delta; a snippet is not automatically a standalone input. Preserve all blank-line and file-provenance requirements.

2.4.2.1 Input block 1

Stage A: fixed geometry and k mesh; ecutwfc = 40, 50, 60, 80 Ry
Stage B: selected ecutwfc; ecutrho = 4, 6, 8, 10 times ecutwfc
Stage C: selected cutoffs; Si meshes = 4^3, 6^3, 8^3, 10^3
Stage D: repeat the two strictest settings on a slightly strained cell

2.4.3 Worked investigation

2.4.3.1 Intuition and prerequisites

Electronic iteration, plane-wave basis, density grid and Brillouin-zone sampling are different error sources. Their errors can cancel accidentally. Use Si-base for an insulating example and Al-base for a metallic comparison. Decide first whether the target is an energy difference, force, stress or band separation; a cutoff adequate for one may fail another.

2.4.3.2 Original experiment matrix

See input block 1 above.

2.4.3.3 Outputs and decision

Store a CSV containing both cutoffs, mesh, smearing, energy per atom, pressure, selected energy difference, number of plane waves and wall time. Plot error relative to the strictest tested setting and say explicitly that this reference is finite, not exact. Accept a working setting only if a still stricter independent test preserves the scientific conclusion. Tightening everything at once may show agreement but cannot reveal which parameter controls the cost.

Exercise: write a stopping rule before looking at the curves, such as a chosen meV/atom tolerance plus a force tolerance for your application. Explain why a nonmonotonic metal sequence does not automatically imply a software bug. Compare a one-variable-at-a-time workflow with a small cross-check matrix and identify possible error cancellation.

2.4.3.4 Geometry and mechanics

2.4.5 Sources and further reading