[CPMD-list] how to do constraint optimization?

[Salomon Billeter]

Dear Lan-Feng,

version 3.10 is not affected by the bug. With the linear-scaling geometry
optimizer, the optimization will converge quickly (it converged quickly
here, while the conjugate gradient optimization did not converge either).
To switch to L-BFGS, change the keyword 'CONJUGATE GRADIENTS IONS' to
'LBFGS' in the &CPMD section. You best also switch on diagnostics using
PRINT LSCAL ON.

May I comment on two other portions of your input:
- You have an isolated system but not SYMMETRY=0. Your periodic images will
be electrostatically coupled. Also, your cell size is rather at the lower
limit.
- You have very tight convergence criteria, espeically for the electrons.
If you start with a large nuclear gradient (as you do), adaptive tolerance
saves you many cycles (keywords CONVERGENCE ADAPT, CONVERGENCE INITIAL, and
- in your case probably not too helpful - CONVERGENCE ENERGY, CONVERGENCE
CALFOR, and CONVERGENCE RELAX). Values around 0.01 and 0.0001 for the first
two options could speed up the calculation in your case.

When reaching convergence, please make sure the constraint is maintained
accurately enough. As written before, CPMD does not determine Lagrange
multipliers during geometry optimization for most optimizers, and the
penalty could be too small.

Best regards,
Salomon