[QE-users] Relaxation: bfgs failed after 4 scf cycles and 3 bfgs steps, convergence not achieved

Paolo Giannozzi paolo.giannozzi at uniud.it
Thu Sep 19 12:07:40 CEST 2024


QE was not devised to reach the level of convergence that you would like 
to achieve. I don't think you can do that, without some insight of the 
origin of numerical errors in QE, and maybe not even with: numerical 
errors often arise in system and mathematical libraries

Paolo

On 18/09/2024 14:38, aleksandr.doma--- via users wrote:
> Dear QE users & devs!
> 
> We get the same error using different pseudopotentials (SSSP, LDA, 
> PBEsol) at the final steps of the structure relaxation. I mean the message:
> / bfgs failed after   4 scf cycles and   3 bfgs steps, convergence not 
> achieved/
> /     (criteria: energy <  1.0E-08 Ry, force <  1.0E-08 Ry/Bohr, cell <  
> 1.0E-01 kbar)/
> /
> /
> /     End of BFGS Geometry Optimization/
> 
> 
> Using the last geometry (lattice parameters and atomic coordinates) in 
> the next input file, QE takes 4 steps again and stops with the same problem.
> 
> We would like to achieve "ideal relaxation" for use in elastic and 
> phonon calculations. Highlights of the input file:
> 
> /&control/
> /    ..../
> /    calculation  = 'vc-relax',
> /
> /    forc_conv_thr = 1.0D-8,/
> /    etot_conv_thr = 1.0D-8,/
> /    //
> /&SYSTEM/
> /  celldm(1) =     15.89938814/
> /  celldm(2) =      1.30410009/
> /  celldm(3) =      1.47077836/
> /  ibrav       =  8,/
> /  nat         = 104,/
> /  ntyp        = 4,/
> /  ecutrho     = 800/
> /  ecutwfc     = 100/
> ///
> /&ELECTRONS/
> /  conv_thr    = 1.D-10, (or 1.D-11)/
> /  mixing_beta=0.5/
> ///
> /
> /
> /&IONS/
> / ion_dynamics = "bfgs"/
> ///
> /&CELL/
> / cell_dofree = "ibrav"/
> / cell_dynamics = "bfgs"/
> / press = 0.0/
> / press_conv_thr = 0.1/
> /
> /
> /K_POINTS AUTOMATIC/
> /6 5 4 0 0 0/
> /(8 6 6 0 0 0)/
> 
> Restarting the 'vc-relax' calculation several times, we got the 
> following "total force" results:
> Total force =     0.000080     Total SCF correction =     0.000018
> SCF correction compared to forces is large: reduce conv_thr to get 
> better values
> ----------------------------------------------------------------------------------------------
> Total force =     0.000084     Total SCF correction =     0.000037
> SCF correction compared to forces is large: reduce conv_thr to get 
> better values
> ----------------------------------------------------------------------------------------------
> Total force =     0.000072     Total SCF correction =     0.000002
> ----------------------------------------------------------------------------------------------
> Total force =     0.000072     Total SCF correction =     0.000003
>       Computing stress (Cartesian axis) and pressure
> 
>            total   stress  (Ry/bohr**3)                   (kbar)     P=  
>        0.00
>     0.00000000   0.00000000   0.00000000            0.00        0.00    
>      0.00
>     0.00000000   0.00000001   0.00000000            0.00        0.00    
>      0.00
>     0.00000000   0.00000000   0.00000002            0.00        0.00    
>      0.00
> 
>       Message from routine bfgs:
>       history already reset at previous step: exiting
>       Energy error            =      2.4E-09 Ry
>       Gradient error          =      1.3E-05 Ry/Bohr
>       Cell gradient error     =      3.1E-03 kbar
> ibrav =      8
>   celldm(1) =     15.89938798
>   celldm(2) =      1.30409950
>   celldm(3) =      1.47077720
> Input lattice vectors:
>       0.99999903     0.00000000     0.00000000
>       0.00000000     1.30409824     0.00000000
>       0.00000000     0.00000000     1.47077577
> New lattice vectors in INITIAL alat:
>       0.99999903     0.00000000     0.00000000
>       0.00000000     1.30409824     0.00000000
>       0.00000000     0.00000000     1.47077577
> New lattice vectors in NEW alat (for information only):
>       1.00000000     0.00000000     0.00000000
>       0.00000000     1.30409950     0.00000000
>       0.00000000     0.00000000     1.47077720
> Discrepancy in bohr =     0.000000    0.000000    0.000000
> 
>       bfgs failed after   4 scf cycles and   3 bfgs steps, convergence 
> not achieved
>       (criteria: energy <  1.0E-08 Ry, force <  1.0E-08 Ry/Bohr, cell <  
> 1.0E-01 kbar)
> 
>       End of BFGS Geometry Optimization
> *=======================================================*
> 
> *next file:*
> *
> *
> Total force =     0.000071     Total SCF correction =     0.000008
> SCF correction compared to forces is large: reduce conv_thr to get 
> better values
> -------------------------------------------------------------------------------------------------
> Total force =     0.000077     Total SCF correction =     0.000009
> SCF correction compared to forces is large: reduce conv_thr to get 
> better values
> -------------------------------------------------------------------------------------------------
> Total force =     0.000078     Total SCF correction =     0.000000
> --------------------------------------------------------------------------------------------------
> Total force =     0.000082     Total SCF correction =     0.000001
> 
> 
>       Computing stress (Cartesian axis) and pressure
> 
>            total   stress  (Ry/bohr**3)                   (kbar)     P=  
>       -0.00
>    -0.00000001   0.00000000   0.00000000           -0.00        0.00    
>      0.00
>     0.00000000  -0.00000002   0.00000000            0.00       -0.00    
>      0.00
>     0.00000000   0.00000000  -0.00000001            0.00        0.00    
>     -0.00
> 
>       Message from routine bfgs:
>       history already reset at previous step: exiting
>       Energy error            =      1.5E-09 Ry
>       Gradient error          =      2.2E-05 Ry/Bohr
>       Cell gradient error     =      2.7E-03 kbar
> ibrav =      8
>   celldm(1) =     15.89938814
>   celldm(2) =      1.30410009
>   celldm(3) =      1.47077836
> Input lattice vectors:
>       1.00000001     0.00000000     0.00000000
>       0.00000000     1.30410010     0.00000000
>       0.00000000     0.00000000     1.47077837
> New lattice vectors in INITIAL alat:
>       1.00000001     0.00000000     0.00000000
>       0.00000000     1.30410010     0.00000000
>       0.00000000     0.00000000     1.47077837
> New lattice vectors in NEW alat (for information only):
>       1.00000000     0.00000000     0.00000000
>       0.00000000     1.30410009     0.00000000
>       0.00000000     0.00000000     1.47077836
> Discrepancy in bohr =     0.000000    0.000000    0.000000
> 
>       bfgs failed after   4 scf cycles and   3 bfgs steps, convergence 
> not achieved
>       (criteria: energy <  1.0E-08 Ry, force <  1.0E-08 Ry/Bohr, cell <  
> 1.0E-01 kbar)
> 
>       End of BFGS Geometry Optimization
> *==========================================================*
> 
> 
> We seem to have reached a threshold as the last calculation shows an 
> increase in "total force" and stops with the same problem after 4 steps:
> 
>   Total force =     0.000082     Total SCF correction =     0.000008
> ------------------------------------------------------------------------------------
> Total force =     0.000088     Total SCF correction =     0.000009
> SCF correction compared to forces is large: reduce conv_thr to get 
> better values
> ---------------------------------------------------------------------------------------
> Total force =     0.000093     Total SCF correction =     0.000000
> ---------------------------------------------------------------------------------------
> Total force =     0.000105     Total SCF correction =     0.000001
> 
> *==================================================*
> 
> Questions:
> 1) What is the acceptable threshold for further calculations of elastic 
> properties? (for example, with the relaxed "Total force = 0.000081" we 
> had error of 200% in elastic constants using thermo_pw for tetraborate 
> crystals)
> 2) Can the relaxation error be avoided to obtain a more accurate geometry?
> 
> Thanks in advance!
> Oleksandr
> 
> 
> _______________________________________________
> The Quantum ESPRESSO community stands by the Ukrainian
> people and expresses its concerns about the devastating
> effects that the Russian military offensive has on their
> country and on the free and peaceful scientific, cultural,
> and economic cooperation amongst peoples
> _______________________________________________
> Quantum ESPRESSO is supported by MaX (www.max-centre.eu)
> users mailing list users at lists.quantum-espresso.org
> https://lists.quantum-espresso.org/mailman/listinfo/users

-- 
Paolo Giannozzi, Dip. Scienze Matematiche Informatiche e Fisiche,
Univ. Udine, via delle Scienze 206, 33100 Udine Italy, +39-0432-558216
X SCUOLA ENERGIE RINNOVABILI 16-21/9/2024


More information about the users mailing list