[Pw_forum] vc-relax bug?

[Stefano de Gironcoli]

Dear Marcello Rosini,
  vc-relax uses the information on forces and STRESS to search for the 
minimum configuration.
  The stress is calculated as the derivative of the total nergy w.r.t. 
cell dimension AT FIXED NUMBER PW and in the evolution of the system the 
same number of PW is kept.
  When you minimize by a muraghan fit  you are instead comparing 
energies  calculated at FIXED CUTOFF .
  This is why you obtain different results.
  Obviously for a well converged calculation derivatives at  fixed 
number of PW or fixed cutoff should agree but 15 Ry is likely 
insufficient for converging stress calculation in InAs.

  stefano

Marcello Rosini wrote:
> Hallo everybody
> I am using pw.x version 4.1 and I noticed a very strange behaviour when 
> performing vc-relax calculation. I have done many tests, and I have 
> reduced to the simplest case possible wich I illustrate here:
>
> InAs fcc lattice:
> * the lattice parameter scan and the murnaghan fit lead me to 11.177au 
> lattice constant with total energy -16.91675895 Ry
> *starting from 11.177 fcc lattice I perform a vc-relax, and I find a new 
> unit cell with ~0.98% smaller lattice constant. All the symmetries are 
> preseved and the InAs distance is still (1/4,1/4,1/4) of the new lattice 
> parameter. Total energy is now -16.91886930 Ry
> *I run a new scf calculation with the new lattice parameters and now I 
> get Etot=-16.91485951... obviously higher, as it was or the murnaghnan 
> calculation.
>
> the question is:
> is there something wrong in the vc-relax procedure? (you can find my 
> input in the following)
>
> thank you
> Marcello
>
> -----------------------
>
> &control
>     calculation = 'vc-relax'
>     restart_mode='from_scratch',
>     nstep=100,
>     pseudo_dir = '~/AB_INITIO/PSEUDO/LDA',
>     outdir='./'
>     etot_conv_thr=1.0e-7,
>     forc_conv_thr=1.0e-5,
> /
> &system    
>       ibrav=2,
>       celldm(1)=11.177,
>       nat=2,
>       ntyp=2,
>       ecutwfc =15,
> /
> &electrons
>     electron_maxstep = 300,
>     diagonalization='david'
>     mixing_mode = 'plain'
>     mixing_beta = 0.7 
>     conv_thr =  1.0e-8,
> /
> &IONS
>     ion_dynamics='bfgs',
> /
> &CELL
>     cell_dynamics='bfgs',
> /
> ATOMIC_SPECIES
> In   114.818   in.cpi.UPF
> As   74.922   As.pz-bhs.UPF
> ATOMIC_POSITIONS
> In      0.00 0.00 0.00
> As      0.25 0.25 0.25
> K_POINTS automatic
> 6 6 6 1 1 1
>
>
>