[Pw_forum] Convergence of forces

[sbraccia carlo]

On Sunday 25 April 2004 04:00 pm, Sergei Lisenkov wrote:
> Dear PWscf authors and users,
>
> I performed the relaxation of the structure using PWscf. I looked into
> output file for forces convergence:
>
> .....
>      Total force =     0.002421     Total SCF correction =     0.001485
>
>      bfgs converged in  10 scf cycles and   8 bfgs steps
> .....
>
> By default,  convergence threshold on forces is 1.d-3 (a.u). But the value
> of the total force is bigger than this threshold. May be, there is another
> criterion for the total relaxation than forces?
>
> Thank you very much,
>  Best wishes,
>   Sergey
>
Dear Sergey,
the convergence threshold on forces is evaluated on each component of the 
force acting on each atom: when all the components are less the the 
threshold, convergence is achieved. Instead Total force is the norm of the 
global force vector (a 3N dimensional vector) that, being a quantity that 
depends on the number of atoms involved, is not suited to be used for test 
pourpose (unless you divide that number by SQRT(3N)).
carlo sbraccia