3.2 Orbital transformation as a controlled solver choice
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.
3.2.1 Model, units and provenance
Keep basis/potential files and executable versions traceable. Grid controls use Ry; common energy/force outputs use hartree and hartree/bohr. Read the unit in each output heading.
Shared inputs, conventions and evidence
3.2.2 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.
3.2.2.1 Input block 1
3.2.3 Worked investigation
Intuition. Solving the same Kohn–Sham problem faster should not change the scientific model. Orbital transformation optimizes the occupied subspace and is particularly useful for insulating systems. Its preconditioner changes the geometry of the numerical search, not the physical band gap. Start from A and a converged grid; use the identical geometry for each timing comparison.
Input delta. In DFT/SCF, compare the following OT block against the baseline. ENERGY_GAP here is a preconditioning estimate in atomic energy units; it is not an imposed physical excitation or a measured gap. Keep EPS_SCF and MAX_SCF fixed while testing MINIMIZER CG versus DIIS. Do not combine this introductory OT recipe with the metallic smearing recipe of case 11; specialized OT occupation variants need their own documentation and validation.
See input block 1 above.
Workflow. 1. Run a fresh atomic guess and record iteration count, SCF time and final energy/forces. 2. Switch only the minimizer to DIIS and repeat from the same type of guess. 3. For a difficult geometry, compare robust CG behavior against DIIS without hiding failed runs. 4. Test restart guesses separately: a calculation starting from a converged nearby wavefunction is a different performance experiment from one starting atomically. 5. If convergence stalls, inspect electron count, geometry and gap character before increasing MAX_SCF. 6. Verify that both successful solvers reach matching observables within a tolerance tighter than the chemical question.
Interpretation. Fewer iterations do not guarantee lower total time; building and applying a preconditioner can be expensive. A stable residual decline is useful, but the final state must also be chemically appropriate. If repeated guesses reach different electronic minima, report that model ambiguity instead of averaging their energies. The OT energy-gap estimate is a tuning parameter with a documented relationship to preconditioning; picking a large value to force convergence does not demonstrate a large physical gap.
Checks and pitfalls. Compare forces as well as energies, because small electronic errors can have different derivative effects. Time more than one identical run when system noise is appreciable, and record process/thread layout. Do not include first-time file caching effects in one solver but exclude them from the other. Increasing MAX_SCF merely grants more attempts. Metallic and nearly degenerate systems often need diagonalization, extra orbitals and occupation control; a failure there is not evidence that all CP2K calculations are unreliable.
Exercise. Create a solver report with a failed case, a successful case, the same accuracy target and total wall time. Explain which change addressed conditioning and which would have changed the physical problem. Propose one justified next diagnostic for a stubborn run.
Diagram. Two schematic paths over the same energy landscape, one stable and one oscillatory, converge toward the same target. A separate clock icon accounts for preconditioner cost. EN: occupied-subspace optimization, conditioning, residual, total time. No invented iteration histories.
3.2.4 Related calculations
- 3.1 A two-dimensional grid-convergence experiment
- 3.3 Diagonalization and finite-temperature occupations