[Pw_forum] relax vs multiple SCF calculations inconsistency

sarashs sarashs at ece.ubc.ca
Tue Apr 18 21:54:22 CEST 2017


They use exactly the same cuttoffs, K-points and everything 
(unfortunately.) but the near equilibrium energies are slightly less 
than equilibrium one. For instance for SiOZr angle at equilibrium I get:

!    total energy              =    -245.22924923 Ry
      Harris-Foulkes estimate   =    -245.22924923 Ry
      estimated scf accuracy    <          3.9E-12 Ry

      The total energy is the sum of the following terms:

      one-electron contribution =    -715.79581276 Ry
      hartree contribution      =     369.08543112 Ry
      xc contribution           =     -57.16353294 Ry
      ewald contribution        =     158.66552815 Ry
      Dispersion Correction     =      -0.02086281 Ry

And for SiOZr angle at equiliberium-25 I get:

!    total energy              =    -245.23454839 Ry
      Harris-Foulkes estimate   =    -245.23454839 Ry
      estimated scf accuracy    <          1.7E-12 Ry

      The total energy is the sum of the following terms:

      one-electron contribution =    -739.45020087 Ry
      hartree contribution      =     380.81843421 Ry
      xc contribution           =     -57.19470926 Ry
      ewald contribution        =     170.61520075 Ry
      Dispersion Correction     =      -0.02327322 Ry

Which is similar in terms of total energy but slightly lower and that's 
weird. Is there anything I can do to force QE to use the same basis set 
throughout SCF calculations?


> On Tuesday, 18 April 2017 17:10:25 CEST sarashs wrote:
>> the other SCF's not have higher energy than the
>> equilibrium angle regardless of them being optimized with a 
>> constraint?
>> I mean if the structure is originally relax (which it is) then one
>> expects other near equilibrium structures to have higher energies. Am 
>> I
>> wrong there?
> 
> You are right.
> 
> They should be higher, which usually means less negative. As long as 
> the same
> pseudopotentials, cutoffs, k-points and everything else is used.
> 
> Do they?



More information about the users mailing list