4.1 Optimize geometry and prove what converged
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.
4.1.1 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
4.1.2 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.
4.1.2.1 Input block 1
&MOTION
&GEO_OPT
TYPE MINIMIZATION
OPTIMIZER BFGS
MAX_ITER 150
MAX_FORCE 1.0E-4
RMS_FORCE 5.0E-5
MAX_DR 1.0E-3
RMS_DR 5.0E-4
&END GEO_OPT
&END MOTION
4.1.3 Worked investigation
Intuition. An optimizer finds a stationary structure near its starting point; it does not certify a global minimum. Electronic convergence is nested inside nuclear optimization, so noisy forces can make a geometry optimizer appear erratic. Start with A, a grid validated for forces and at least two different initial water distortions.
Input delta. Change GLOBAL/RUN_TYPE to GEO_OPT and add this root-level MOTION block. The explicitly noted force units are Hartree/bohr and displacement units bohr; the numeric thresholds below are educational targets and must be tightened for frequency work.
See input block 1 above.
Workflow. 1. Inspect the initial geometry for overlaps and the intended connectivity. 2. Run the optimization and follow each step's energy, maximum/RMS force and displacement criteria. 3. Distinguish convergence from exhaustion of MAX_ITER. 4. Extract the final coordinates from the completed step, not simply the last frame of a partially written file. 5. Recalculate a single point at the final structure with tighter SCF and grid to check residual forces. 6. Repeat from a second distortion and compare the resulting structure after accounting for overall rotation and translation. 7. Use case 25's Hessian check before describing a structure as a verified local minimum.
Interpretation. A small energy change alone is insufficient; a shallow direction can have small energy increments while retaining nonnegligible force. Conversely, finite grid noise can prevent thresholds from being reached even when the geometry has stopped moving sensibly. Constraints remove degrees of freedom, so convergence refers only to the allowed subspace. For flexible molecules, multiple conformers require separate starting structures and relative free-energy reasoning.
Checks and pitfalls. Verify bond lengths and angles, not only a success banner. If a constrained atom carries force, that may be the reaction force of the constraint rather than an unconverged free degree. Do not change functional midway and join the energy history as one optimization. For large systems LBFGS can avoid dense-Hessian storage; this is an optimizer choice, not a different potential-energy model.
Exercise. Compare two initial geometries and two force tolerances. Prepare a convergence certificate listing all tested criteria and a separate statement of what remains unproven, such as global minimality or vibrational stability.
Diagram. A funnel into one of two local basins, with separate electronic and nuclear convergence gates. EN: starting geometry, SCF accuracy, force criterion, local minimum candidate, Hessian check.
4.1.4 Related calculations
- 3.4 Brillouin-zone sampling and supercell logic
- 4.2 Cell optimization and an independent equation of state