[Pw_forum] What is the exact criteria for scf convergence?

[MKondrin]

Hello!

I try to optimize quite large structure (several tens of atoms in the
unit cell) but I can not quite get that is the criteria of convergence
to proceed for the next optimization cycle. I use QE 5.01

I have in the preamble of input file something like:

&control
calculation='relax',
prefix='scf',
tstress=.true.,
tprnfor=.true.,
pseudo_dir = '../pseudo',
outdir='./tmp'
/
&system
ibrav=0,
celldm(1) = 12.60447,
ntyp=2,
nat=84,
ecutwfc=70,
ecutrho=800,
occupations='fixed'
/
&electrons
    mixing_beta = 0.7,
    conv_thr =  1.0d-6
/
&ions
/
....

I suppose that the next cycle of structure optimization should occur
than the accuracy of self consistent calculations of about 1e-6 Ry.

However in the output file I get this (this is the end of one scf cycle)

     iteration # 20     ecut=    70.00 Ry     beta=0.70
     Davidson diagonalization with overlap
     ethr =  5.68E-10,  avg # of iterations =  1.0

     total cpu time spent up to now is   230665.4 secs

     total energy              =    -823.97008715 Ry
     Harris-Foulkes estimate   =    -823.97008883 Ry
     estimated scf accuracy    <       0.00000279 Ry

     iteration # 21     ecut=    70.00 Ry     beta=0.70
     Davidson diagonalization with overlap
     ethr =  5.68E-10,  avg # of iterations =  4.0

     total cpu time spent up to now is   233303.9 secs

     total energy              =    -823.97008869 Ry
     Harris-Foulkes estimate   =    -823.97008855 Ry
     estimated scf accuracy    <       0.00000071 Ry

     iteration # 22     ecut=    70.00 Ry     beta=0.70
     Davidson diagonalization with overlap
     ethr =  3.37E-10,  avg # of iterations =  1.0

     total cpu time spent up to now is   236012.3 secs

     total energy              =    -823.97008898 Ry
     Harris-Foulkes estimate   =    -823.97008869 Ry
     estimated scf accuracy    <       0.00000092 Ry

     iteration # 23     ecut=    70.00 Ry     beta=0.70
     Davidson diagonalization with overlap
     ethr =  3.37E-10,  avg # of iterations =  3.5

     total cpu time spent up to now is   238645.5 secs

     total energy              =    -823.97008880 Ry
     Harris-Foulkes estimate   =    -823.97008904 Ry
     estimated scf accuracy    <       0.00000178 Ry

     iteration # 24     ecut=    70.00 Ry     beta=0.70
     Davidson diagonalization with overlap
     ethr =  3.37E-10,  avg # of iterations =  1.0

     total cpu time spent up to now is   241385.7 secs

     total energy              =    -823.97008829 Ry
     Harris-Foulkes estimate   =    -823.97008881 Ry
     estimated scf accuracy    <       0.00000147 Ry

     iteration # 25     ecut=    70.00 Ry     beta=0.70
     Davidson diagonalization with overlap
     ethr =  3.37E-10,  avg # of iterations =  3.9

     total cpu time spent up to now is   244038.6 secs

     total energy              =    -823.97008797 Ry
     Harris-Foulkes estimate   =    -823.97008835 Ry
     estimated scf accuracy    <       0.00000057 Ry

     iteration # 26     ecut=    70.00 Ry     beta=0.70
     Davidson diagonalization with overlap
     ethr =  2.72E-10,  avg # of iterations =  4.0

     total cpu time spent up to now is   246730.6 secs

     total energy              =    -823.97008832 Ry
     Harris-Foulkes estimate   =    -823.97008847 Ry
     estimated scf accuracy    <       0.00000043 Ry

     iteration # 27     ecut=    70.00 Ry     beta=0.70
     Davidson diagonalization with overlap
     ethr =  2.03E-10,  avg # of iterations =  1.0

     total cpu time spent up to now is   249273.6 secs

     total energy              =    -823.97008831 Ry
     Harris-Foulkes estimate   =    -823.97008833 Ry
     estimated scf accuracy    <       0.00000021 Ry

     iteration # 28     ecut=    70.00 Ry     beta=0.70
     Davidson diagonalization with overlap
     ethr =  9.87E-11,  avg # of iterations =  1.1

     total cpu time spent up to now is   252028.1 secs

     total energy              =    -823.97008826 Ry
     Harris-Foulkes estimate   =    -823.97008832 Ry
     estimated scf accuracy    <       0.00000017 Ry

     iteration # 29     ecut=    70.00 Ry     beta=0.70
     Davidson diagonalization with overlap
     ethr =  8.33E-11,  avg # of iterations =  6.2

     total cpu time spent up to now is   254596.4 secs

     End of self-consistent calculation


Aren't the cycles after #25 superfluous, because energy difference
between two successive steps and estimated accuracy is less than 1e-6
Ry? What is the control parameter that help me to reduce the number of
steps ?

Thank you in advance!

M. V. Kondrin (High Pressure Physics Institute RAS, Troitsk, Russia)