## Total energy increases for more k points[SOLVED]

Moderators: jzwanzig, jolafc

### Total energy increases for more k points

Hello,
I have one understanding problem which I even didn't recognise for a long time. My system converges very fast with for higher Ecut or more k-points. I did a converges study via a loop from 1 k-point + 2 k-point each loop. after 5 k-points it coveraged, the total energy rises like this:
etotal1 -3.0999171036E+01
etotal2 -3.0909749251E+01
etotal3 -3.0905728251E+01
etotal4 -3.0905582987E+01
etotal5 -3.0905496096E+01
etotal6 -3.0905541922E+01
etotal7 -3.0905462868E+01
etotal8 -3.0905522243E+01
etotal9 -3.0905474050E+01
etotal10 -3.0905492651E+01
Why...isn`t it a bit strange ... how should one interpret this result?
thx for help
BR
maths
maths

Posts: 23
Joined: Tue Oct 13, 2015 6:15 pm

### Re: Total energy increases for more k points

Hi!

Can you copy here your input file?

Best regards,

Igor
ilukacevic

Posts: 271
Joined: Sat Jan 16, 2010 1:05 pm
Location: Dept. of Physics, University J. J. Strossmayer, Osijek, Croatia

### Re: Total energy increases for more k points

# =======================================
# Input for PAW
# Polypropylen DOS
# =======================================

#Output of the total DOS
# prtdos 1

#Output of the projected DOS (with PAW contribs)
#prtdos 2
#natsph 1 iatsph 1 ratsph 1.5 0.9

#determine of latice parameters
#optcell 0
#ionmov 2
#ntime 1000
#toldfe 1.5d-7
#dilatmx 1.1

#Cell and atoms definition 1. 2.53573 2. 2.53428
#ndtset 20
#acell: 1. 5 5 angstrom acell+ 0.2 0 0 angstrom
acell 2.534276572 5 5 angstrom
rprim 1.0 0.0 0.0
0.0 1.0 0.0
0.0 0.0 1.0
ntypat 2 natom 6 typat 2*1 4*2
xangst
0.633933 0.000000 0.000000
1.901800 0.000000 0.855186
0.633933 -0.859273 -0.635829
0.633933 0.859273 -0.635829
1.901800 -0.859273 1.491010
1.901800 0.859273 1.491010

znucl 6 1 nband 14
ecut 40
ecutsm 0.5
pawecutdg 50
toldfe 1.5d-7
#SCF cycle parameters
nbdbuf 2 #nbdbuf:does not take care of empty bands
nstep 100

#K-points and sym
nsym 0
#occopt 7 tsmear 0.005
ndtset 20
ngkpt: 1 1 1 ngkpt+ 2 0 0
# nshiftk 4
# shiftk 0.5 0.5 0.5
# 0.5 0.0 0.0
# 0.0 0.5 0.0
# 0.0 0.0 0.5

#I/O parameters
#optforces 2 optstress 1
#prtwf 0 prtden 0 prteig 0
#getwfk -1

#%%<BEGIN TEST_INFO>
#%% [setup]
#%% executable = abinit
#%% [files]
#%% files_to_test =
#%% tpaw1_4.out, tolnlines= 0, tolabs= 0.000e+00, tolrel= 0.000e+00, fld_options = -easy
#%% psp_files = 6c.lda.atompaw
#%% [paral_info]
#%% max_nprocs = 4
#%% [extra_info]
#%% authors =
#%% keywords = PAW
#%% description =
#%%<END TEST_INFO>
maths

Posts: 23
Joined: Tue Oct 13, 2015 6:15 pm

### Re: Total energy increases for more k points

Any ideas?
thx lg
maths
maths

Posts: 23
Joined: Tue Oct 13, 2015 6:15 pm

### Re: Total energy increases for more k points

It would be really helpful if someone could help me with this topic!
many thx!
maths

Posts: 23
Joined: Tue Oct 13, 2015 6:15 pm

### Re: Total energy increases for more k points

Hi,

You mentionned 10 values for etotal while you have ndtset=20. What are the last 10 values ?
It seems to converge, you should continue the convergence study.

Also, make sure the y and z axis of the box does not change the energy. Maybe 5A is not enough.

Jordan
Jordan

Posts: 280
Joined: Tue May 07, 2013 9:47 am

### Re: Total energy increases for more k points

I know that it converges ... What I do not get is the fact that is converges vom a smaller total energy.
So why is Etot of k=1,2 < Etot of k=5,6,7,8....20. whereby the total energy of k=1 and k=2 are different but k=5 to k=20 are approx. the same
I calculated the DOS and it looks really good so it can't be to wrong I think. But I simply have problems with the interpretation.
many thx!
maths
maths

Posts: 23
Joined: Tue Oct 13, 2015 6:15 pm

### Re: Total energy increases for more k points  [SOLVED]

Hi maths,

the energy is not monotonous as a function of nkpt:
* For ecut, you keep the existing basis set and complete it - the variational principle implies Etot should always decrease (some exceptions possible with weird pseudopotentials or PAW densities etc...)
* For k-points, the equivalent would be doing grids and sub grids e.g. 2 4 8 16 etc... but even then monotony is not guaranteed. Depending on whether you hit k-points with large or small contributions to Etot you will get shifts up or down, which only converge asymptotically. Your grids are not even related to each other so no rules on Etot (nkpt)
Matthieu Verstraete
University of Liege, Belgium
mverstra

Posts: 618
Joined: Wed Aug 19, 2009 12:01 pm