SeqQuest input: Cell optimization

Table of Contents

1. Overview
2. Input options
3. Troubleshooting

Overview

The default unit for the cell relaxation section is GPa for bulk 3D, 10^-7 N/m for 1D, 10⁺² N/m² for 2D.

Current constraints (as of version 2.57):

• cell relaxation implemented in v2.53
• atomic coordinates must be in LATTICE units
• no scaling functions are allowed in the input
• spin-polarized GGA stress code implemented in v2.57

Input options

The following are the common input keywords in this section. All the keywords are optional, and can appear in any order within this section. The keywords must be left-justified, but all data input is free-format.

cell optimization data - begin cell opt input section

...

ucsteps - number of cell steps

n_step_uc

uchistory - max cell steps to include in blending history

n_hist_uc

uc_convergence - max stress (GPa) convergence criterion

stress_conv

pressure - external pressure (GPa)

P

uniaxial_pressure - external x,y,z axial pressures

P_x P_y P_z

stress - external stress (ndim X ndim symmetric tensor)

s(1,1) s(2,1) s(3,1)
s(1,2) s(2,2) s(3,2)
s(1,3) s(2,3) s(3,3)

ucmethod - method for stress elimination

method_uc (BROYDEN or STEEPEST)

str_broyden - cutoff for switch to broyden (GPa)

stress_broy

bulk_modulus - (in GPa; can improve initial cell update)

B

...

end cell optimization - end of cell opt input section

Details

• Important: The external stress is input as a 2x2 matrix for a 2D (slab) problem and 1x1 in a 1D-chain problem.
• The default method for cell relaxation is a combination method: starts with STEEPEST (descent) and switches to BROYDEN when the maximum stress goes below stress_broy.
• Choose, at most, one of (pressure, uniaxial, stress) - all modify the same external stress tensor, and, used in combination, will do so unpredictably.

Troubleshooting

The cell shape and stress are converging very slowly.

The calculation may be making small steepest descent steps. The best option depends on the situation. Giving a smaller bulk_modulus (default ~100) will expand the first cell update step size (and subsequent step sizes). If the stress is decreasing, then the calculation is likely in a basin (and converging) and stuck doing small steepest descent steps. If the stress is monotonically decreasing from the beginning, select BROYDEN as the ucmethod, or modify the stress (using str_broyden) at which the default combination method switches from steepest descent to Broyden.

The cell "blew up" on the first cell update step.

Increase the bulk_modulus input (default ~100) to reduce the step size on the first cell update.

Later (after the first step) the cell "blew up".

If you are not in a potential well, the second derivative matrix might be ill-behaved and Broyden will fail. Try using STEEPEST as the minimization method (starting with the last "good" cell shape).

The minimization stalls and the cell will not converge further.

Numerical noise - the convergence criterion might be smaller than the inaccuracy in the computed stress. Tighten force convergence criterion, and/or use a denser grid (to get more accurate stress).