[QE-users] Strange total force with ESM on gpu-QE 6.5

[Michele Re Fiorentin]

Dear QE users,

I'm currently trying to run the gpu-accelerated version of QE 6.5 (on the
brand new CINECA Marconi100 cluster).
I have tried to simulate a slab with assume_isolated='esm', esm=bc='bc3'
and zero total charge. I'll attach the input file below.
Though starting with a previously relaxed geometry, when computing the
forces on atoms, I get very strange results


     Forces acting on atoms (cartesian axes, Ry/au):
>      atom    1 type  1   force =   -79.36264506   58.50564927  193.21862563
>      atom    2 type  1   force =   151.48848433  -95.71164712  156.42915271
>      atom    3 type  1   force =  -157.75453419   75.13372814  255.59098655
>      atom    4 type  1   force =    57.72287892  -37.54582029   50.45982033
>      atom    5 type  2   force =   -36.81532728   18.06988603   22.16454482
>      atom    6 type  2   force =    46.90240002  -28.89575882   59.34996427
>      atom    7 type  1   force =    92.73157800  -66.52542242  226.46207068
>      atom    8 type  1   force =  -143.89224269   92.18337180  266.85779603
>      atom    9 type  1   force =   112.08721764  -53.83773401  359.49265886
>      atom   10 type  1   force =   -83.72081022   59.21267373  281.55587127
>      atom   11 type  2   force =    57.99370828  -32.75316399   86.22952848
>      atom   12 type  2   force =   -17.38070774   12.16423769
>  174.19405690



     Total force =   664.738235     Total SCF correction =     0.000155


This does not happen if I use esm_bc= 'pbc'.
Could you please help me by pointing out what I am missing?
Thank you very much!
All the best,


Michele


INPUT FILE:

&CONTROL
   calculation      = 'scf'
   wf_collect       = .true.
   nstep            = 1000
   outdir           = './Cu2Sb_esm3/'
   prefix           = 'Cu2Sb_esm3'
   etot_conv_thr    = 0.0001
   forc_conv_thr    = 0.001
   pseudo_dir       = '/m100/home/userexternal/mrefiore/espresso/pseudo'
   tprnfor   = .true.
/
&SYSTEM
   ibrav            = 6
   celldm(3)        = 7.43772
   celldm(1)        = 7.6222
   ntyp             = 2
   nat              = 12
   ecutwfc          = 34
   ecutrho          = 340
   occupations      = 'smearing'
   degauss          = 0.002
   smearing         = 'gaussian'
   nosym = .true.
   assume_isolated  = 'esm'
   esm_bc           = 'bc3'
   esm_w            = 0.
   tot_charge       = 0.
/
&ELECTRONS
   electron_maxstep = 200
   conv_thr         = 1e-08
   mixing_mode      = 'local-TF'
   mixing_beta      = 0.3
/


ATOMIC_SPECIES
Cu 63.546 Cu.pbe-dn-rrkjus_psl.1.0.0.UPF
Sb 121.76 Sb.pbe-n-rrkjus_psl.1.0.0.UPF

ATOMIC_POSITIONS angstrom
Cu 0.0000000000 0.0000000000 -3.1585705000 0 0 0
Cu 2.0172032430 2.0172032300 -3.1585704420 0 0 0
Cu 2.0164553070 0.0007474290 -1.5374613730 0 0 0
Cu 0.0007492090 2.0164535650 -4.7798157180 0 0 0
Sb 2.0170967170 0.0001063070 -5.0090138370 0 0 0
Sb 0.0000000000 2.0171039870 -1.3080844460 0 0 0
Cu 0.0002346720 0.0002478990 3.0428563360
Cu 2.0169301130 2.0169338950 3.0428562230
Cu 2.0170039260 0.0001702910 4.5612774790
Cu 0.0001708180 2.0169938130 1.4073471570
Sb 2.0169996630 0.0001688030 1.1709124870
Sb 0.0002762610 2.0169228960 5.0110138370

K_POINTS automatic
8 8 1  1 1 0




--
Michele Re Fiorentin, PhD

Istituto Italiano di Tecnologia (IIT)
Center for Sustainable Future Technologies @ PoliTO
via Livorno 60
10144 Torino (Italy)
tel: +39 0110904333
email:    michele.refiorentin at iit.it
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20200513/c9f9f0bf/attachment.html>