Could anyone please suggest what could be the problem of strong
total energy oscillations while calculating the ground state
of the 16-atomic hexagonal Mn5Ge3 crystal?
I tried to increase input parameters like rgkmax, k-points mesh,
nempty, but the problem still exists and I cannot receive
the total energy convergence.
Here is my input file.
Any ideas will be greatly appreciated!
Regards,
Yvett
&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
! Ferromagnetic Mn5Ge3. Note the small global magnetic field, which
! is needed to break spin symmetry. Check the total moment of the cell in the
! file INFO.OUT.
tasks
0
spinpol
.true.
! small magnetic field (NOT in the z-direction)
bfieldc
0.0 0.0 0.01
! fairly large number of empty states required for magnetic cases
nempty
10
avec
13.575797890 0.000000000 0.000000000
-6.787898945 11.756985849 0.000000000
0.000000000 0.000000000 9.548789913
Thank you very much for your suggestions! And by the way, it was not your mistake,
it was me who mistakenly posted the new topic as an answer to the topic you mentioned.
Well, I tried to change mixing parameter to beta0=0.05, nut it did not help.
The set of total energies from the file TOTENERGY.OUT (not converged results) is the following:
The local field will be useful if you would like to converge antiferromagnetic structure or non collinear magnetic structure by pointing such local field in different directions for different atoms.
RMS change in effective potential (target) : 0.2938114421E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.4885447121E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7146768015E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7320332993E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7005066471E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7187715347E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7014510926E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7120874388E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7025873089E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7087804969E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7034197062E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7070344620E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7039825926E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7060559912E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7043423675E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7054848938E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7045633134E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7051453275E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7046937974E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7049431071E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047667528E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7048248425E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7048041931E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047582247E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7048204128E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047234120E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7048244045E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047077956E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7048217385E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047035036E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7048159838E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047054057E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7048088217E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047105764E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7048017042E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047172056E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047951022E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047240140E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047892333E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047304930E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047841575E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047363192E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047801171E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047414341E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047766182E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047457562E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047739345E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047495005E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047717478E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047525063E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047699927E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047549409E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047686519E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047570207E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047677065E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047586403E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047668469E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047598974E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047662415E-01 ( 0.1000000000E-05)
>The local field will be useful if you would like to converge antiferromagnetic structure or non collinear magnetic structure by >pointing such local field in different directions for different atoms.
OK, I will try to remember that, thank you.
As I understand, in calculations with non-collinear magnetism the directions of global and local
magnetic fields do not necessarily coincide, am I right?
And one more question, please: do the directions of local magnetic spins change during structural
relaxation or they remain the same as specified in the input file? The same concerns the global magnetic field.
Thank you again for your helpful hints and quick replies!
I appreciate it very much!
Regards,
Yvett
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Yes your system is not converging at all, I'm almost sure that is because you're using to few kpoints.
What happen to the atom moment in INFO.OUT?
For what concern your question about the magnetic field and non collinear:
In general a small magnetic field is needed to break the symmetry and make your calculation converge to a magnetic solution.
So if you would like to get an antiferromagnetic solution you point the local magnetic field in opposite directions on two atoms.
When you start you should already have an idea of the magnetic structure that you would like to converge. To see which magnetic structure is more stable you may compare total energy of each of them.
In a collinear calculation the hamiltonian is divided in two blocks < up | H | up > and < down | H | down >, where up and doown refer to the spin up and spin down part of the Kohn-Sham wavefunctions, by doing so we allow only two directions for the magnetization: direction A and the -A .
In a non collinear calculation we also include the hamiltonian blocks < up | H | down > < up | H | down >, consequently the magnetization is allowed to vary along every arbitray direction. This is a more realsitic case, but more expensive about 8 times that the collinear case, since you diagoalize an N by N matrix instead that twice an N/2 by N/2 matrix.
I can give you this reference that give how magnetism is implemented in Exciting:
PRB 66 014447 (2002)
During structure relaxation the system is relaxed at every relaxation step to the ground state, so the moment will converge to the most stable solution corresponding to the volume that you are at.
Of course if you provided a local field on the same direction on all the magnetic atoms, the system will stay ferromagnetic during the relaxation.
Hope this will help!
Best
Francesco
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I have used the parameters beta0=0.05 and betamax=0.1, and k-points mesh without
shift, but the system still cannot converge, and I obtain strange
message in the log file on every scf-loop (strange, because in INFO.OUT every
scf-loop ends with a correct value of total charge, but maybe I miss something):
Warning(charge): total charge density incorrect for s.c. loop 1
Calculated : 443.0112466
Required : 442.0000000
I don't know yet where exactly the problem comes from, but I have to solve it.
Thank you very much for your help!
Best regards,
Yvett
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I'm running your input file right now after I have increased all relevant parameters, starting from number of kpoints. We will see in the next 24 hours what if the calculation is going to converge.
I'm curious to look at the eigenvalues in EIGVAL.OUT and at the core energies in EVALCORE.OUT, to see that all the local orbitals are set properly in the default species file.
By the way when something does not work in EXCITING I usually follow this procedure:
1) Check your structure, first of all by plotting it in Xcrysden (http://www.xcrysden.org)by using task 250 for example.
2) Increase all relevant numerical parameters, now I'm trying:
! mixing reduced
beta0
0.05
! number of plane waves increased
rgkmax
7.0
! default
gmaxvr
12.0
! number of empty states increased
nempty
10
! l-cutoff potential increased
lmaxvr
7
! l-cutoff hamiltonian increased
lmaxmat
7
! l-cutoff basis set increased
lmaxapw
10
! kpoint grid increaased
ngridk
7 7 10
! smearing reduced
swidth
0.005
3) Check the file EIGVAL.OUT and EVALCORE.OUT to see that everything is all right, you may check
the EXCITING FORUM topic "From valence to core electrons" in which I tried to give all the essential information on this topic and the setting of the species file.
Finally a trick to speed up convergence for magnetic systems
perform few iterations with a bigger magnetic field (local or global)
example:
task
0
maxscl
10
bfieldc
0.0 0.0 3.0
and then reduce it and converge starting from the previous STATE.OUT:
task
1
bfieldc
0.0 0.0 0.01
Concerning this you may also check the forum topic " bcc iron" answer number 2.
I'll write you as soon as I get some news.
In the meanwhile you may also check and plot your structure, if you didn't do it yet and trying to run with the same input setting I'm using.
Best
Francesco
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi again!
By the way do you know what is the experimetal ground state magnetic structure of your system, is it the ferromagnetic one, or antiferromagnetic?
Best
Francesco
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Sorry for this long silence, just wanted to check everything.
> By the way do you know what is the experimetal ground state magnetic structure of your system, is it the ferromagnetic one, or antiferromagnetic?
Well, in principle all I know from the previous pseudopotential calculations is that
collinear FM and non-collinear (with a small deviation [~6 deg] from collinearity between Mn spins
on different sites) are competing phases in the case of Mn5Ge3 (non-collinear phase is
0.001 ev/Mn stabler than that of a FM one).
But I tried to conduct non-collinear calculations also and met the same problem (I just chose
the direction of the local magnetic field to be (0.0, 0.1, 0.0)).
Now, concerning the structure. I checked the structure both by using task 250 and
by using spacegroup utility. Everithing seems fine with the structure (see below).
! Atomic positions generated by spacegroup version 1.1.02
! Hermann-Mauguin symbol : P63/mcm
! Hall symbol : -P 6c 2
! Schoenflies symbol : D6h^3
! space group number : 193
! lattice constants (a,b,c) : 13.57579254 13.57579254 9.548786149
! angles in degrees (ab,ac,bc) : 120.0000000 90.00000000 90.00000000
! number of conventional unit cells : 1 1 1
! reduction to primitive cell : F
! Wyckoff positions :
! species : 1, Mn.in
! 0.3333333300 0.6666666700 0.000000000
! 0.2390000000 0.000000000 0.2500000000
! species : 2, Ge.in
! 0.6030000000 0.000000000 0.2500000000
avec
13.57579254 0.000000000 0.000000000
-6.787896270 11.75698122 0.000000000
0.000000000 0.000000000 9.548786149
I think me and Fredrik Bultmark found a solution for your problem: the number of empty states.
nempty in your calculation is 10, definetely too low, to contain all the electrons. In fact you have 10 Mn atoms whose d electrons significantly "migrates" from one spin channel another in a spin polarized calculation.
If you check your EIGVAL.OUT witn nempty=10 and the relative occupancies you'll see that the highest energy occupied state will have an occupancy significantly different than zero.
Since Mn has 5 empty states in the d shell (check http://www.webelements.com/webelements/elements/text/Mn/econ.html\), in the extreme case we could say that all 5 electrons of one spin channel will migrate to the empty states of the other spin channel, this extreme case will require (Number of Mn)*5 = 50 empty states to contain all your electrons.
We performed few tests with your system here at Uppsala and it seems nempty=25 or 30 is enough.
In addition also the mixing seems to play an important role in order to get rid of the oscillations.
It seems we are going to achieve convergence with the following the input:
tasks
0
spinpol
.true.
betamax
0.1
beta0
0.05
! small magnetic field (NOT in the z-direction)
bfieldc
0.0 0.0 0.01
! large number of empty states
nempty
25
avec
13.575797890 0.000000000 0.000000000
-6.787898945 11.756985849 0.000000000
0.000000000 0.000000000 9.548789913
You may keep the kpoint offset since it usually helps convergence by respect to kpoint mesh size.
I'm performing also a test with few iterations with a big magnetic field and then reduce it to check if this is going to reduce the number of total iterations
Rememeber to converge all parameters, especially kpoint number and rgkmax, before productive calculation.
Hope this will finally solve the problem!
Best
Francesco and Fredrik
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
many-many thanks to you for your help!
I am impressed with such an enormous assistance!
I appreciate it very much!
>We performed few tests with your system here at Uppsala and it seems nempty=25 or 30 is enough.
Maximum which I could try to test the convergence in a reasonable time was nempty=20.
This could be the reason for poor convergence thus.
I will try to do calculations with the parameters
which you found to be appropriate and let you know
how it works. And of course, the "production"
calculations I will start only after I check the
convergence with respect to all important parameters.
Thank you very much once again!
Hope, once I can be helpful to you too. :-)
All my best wishes to you,
Yvett
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
But could you please show me the values of total energy
from the TOTENERGY.OUT file which you got
while testing the Mn5Ge3 system?
I tried to do some calculations with the parameters
which you found to be appropriate for convergence, but
it seems that even with a high value of nempty and ngridk
I still can't get the converged results.
Thank you very much in advance!
Kind regards,
Yvett
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hello Yvett,
I checked better, and I saw that you are perfectly right, with the file I sent you I get a convergence about 1.d-4 and then the system oscillates again.
This should be the mixing parameter to be decreased, let's try with decreased betamax and a larger field
Dear EXCITING users,
Could anyone please suggest what could be the problem of strong
total energy oscillations while calculating the ground state
of the 16-atomic hexagonal Mn5Ge3 crystal?
I tried to increase input parameters like rgkmax, k-points mesh,
nempty, but the problem still exists and I cannot receive
the total energy convergence.
Here is my input file.
Any ideas will be greatly appreciated!
Regards,
Yvett
&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
! Ferromagnetic Mn5Ge3. Note the small global magnetic field, which
! is needed to break spin symmetry. Check the total moment of the cell in the
! file INFO.OUT.
tasks
0
spinpol
.true.
! small magnetic field (NOT in the z-direction)
bfieldc
0.0 0.0 0.01
! fairly large number of empty states required for magnetic cases
nempty
10
avec
13.575797890 0.000000000 0.000000000
-6.787898945 11.756985849 0.000000000
0.000000000 0.000000000 9.548789913
!scale
!1.00
sppath
'../../species/'
atoms
2 : nspecies
'Mn.in' : spfname
10 : natoms
0.33333333 0.66666667 0.00000000 0.0 0.0 0.1 : atposl, bfcmt
-0.33333333 -0.66666667 0.00000000 0.0 0.0 0.1
0.66666667 0.33333333 0.50000000 0.0 0.0 0.1
-0.66666667 -0.33333333 -0.50000000 0.0 0.0 0.1
0.23900000 0.00000000 0.25000000 0.0 0.0 0.1
-0.23900000 0.00000000 -0.25000000 0.0 0.0 0.1
0.00000000 0.23900000 0.25000000 0.0 0.0 0.1
0.00000000 -0.23900000 -0.25000000 0.0 0.0 0.1
-0.23900000 -0.23900000 0.25000000 0.0 0.0 0.1
0.23900000 0.23900000 -0.25000000 0.0 0.0 0.1
'Ge.in' : spfname
6 : natoms
0.60300000 0.00000000 0.25000000 0.0 0.0 0.1
-0.60300000 0.00000000 -0.25000000 0.0 0.0 0.1
0.00000000 0.60300000 0.25000000 0.0 0.0 0.1
0.00000000 -0.60300000 -0.25000000 0.0 0.0 0.1
-0.60300000 -0.60300000 0.25000000 0.0 0.0 0.1
0.60300000 0.60300000 -0.25000000 0.0 0.0 0.1
autormt
.true.
rgkmax
6.0
ngridk
2 2 4
vkloff
0.5 0.5 0.5
Hi,
I posted my mistake my answer to this topic in the other topic "How accurate get the cohesive energy". You may check that.
Best
Francesco
Hi, Francesco!
Thank you very much for your suggestions! And by the way, it was not your mistake,
it was me who mistakenly posted the new topic as an answer to the topic you mentioned.
Well, I tried to change mixing parameter to beta0=0.05, nut it did not help.
The set of total energies from the file TOTENERGY.OUT (not converged results) is the following:
1 -24066.08414
2 -24066.96831
3 -24102.15895
4 -24095.39668
5 -24136.98513
6 -24115.86685
7 -24153.89259
8 -24123.80533
9 -24164.29959
10 -24128.27558
11 -24169.10954
12 -24130.57663
13 -24171.27200
14 -24131.91787
15 -24172.45172
16 -24132.82167
17 -24173.22575
18 -24133.47018
19 -24173.76275
20 -24133.94181
21 -24174.14356
22 -24134.28610
23 -24174.43747
-24134.53580
-24174.65445
-24134.70841
-24174.79720
-24134.82199
-24174.89311
-24134.89216
-24174.94208
-24134.93006
-24174.96967
-24134.94568
-24174.97758
-24134.94388
-24174.96673
-24134.92980
-24174.94488
-24134.90818
-24174.92329
-24134.88076
-24174.88402
-24134.84935
-24174.85052
-24134.81809
-24174.82149
-24134.78592
-24174.77929
-24134.75357
-24174.74447
-24134.72324
-24174.71402
-24134.69387
-24174.68193
-24134.66506
-24174.64836
-24134.63875
-24174.62043
-24134.61457
-24174.59610
-24134.59195
-24174.57354
-24134.57011
-24174.54921
-24134.54912
-24174.52687
-24134.52889
-24174.50221
-24134.51018
-24174.48049
-24134.49334
-24174.46263
-24134.47726
............
And so on....
But I will try to play with k-mesh more (without shift), and see what's going to happen.
Thank you once again!
Kind regards,
Yvett
Hi Yvett!
Yes try by increasing number of kpoints and no shift, by the way what is the output of the command
grep RMS INFO.OUT to check the convergence?
You may also try to increase rgkmax to 6.5, and by the way when you specify the atoms coordinates
atoms
2 : nspecies
'Mn.in' : spfname
10 : natoms
0.33333333 0.66666667 0.00000000 0.0 0.0 0.1 : atposl, bfcmt
-0.33333333 -0.66666667 0.00000000 0.0 0.0 0.1
etc...
you do not need to apply a local field since you already applied a global field
bfieldc
0.0 0.0 0.01
so the atom coordinates will look like
atoms
2 : nspecies
'Mn.in' : spfname
10 : natoms
0.33333333 0.66666667 0.00000000 0.0 0.0 0.0 : atposl, bfcmt
-0.33333333 -0.66666667 0.00000000 0.0 0.0 0.0
etc...
The local field will be useful if you would like to converge antiferromagnetic structure or non collinear magnetic structure by pointing such local field in different directions for different atoms.
For example
atoms
2 : nspecies
'Mn.in' : spfname
10 : natoms
0.33333333 0.66666667 0.00000000 0.0 0.0 -0.01 : atposl, bfcmt
-0.33333333 -0.66666667 0.00000000 0.0 0.0 0.01
etc...
Best
Francesco
Hi, Francesco!
> grep RMS INFO.OUT to check the convergence?
It gives the following (far from convergence):
RMS change in effective potential (target) : 0.2938114421E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.4885447121E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7146768015E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7320332993E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7005066471E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7187715347E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7014510926E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7120874388E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7025873089E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7087804969E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7034197062E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7070344620E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7039825926E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7060559912E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7043423675E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7054848938E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7045633134E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7051453275E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7046937974E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7049431071E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047667528E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7048248425E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7048041931E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047582247E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7048204128E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047234120E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7048244045E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047077956E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7048217385E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047035036E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7048159838E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047054057E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7048088217E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047105764E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7048017042E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047172056E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047951022E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047240140E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047892333E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047304930E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047841575E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047363192E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047801171E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047414341E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047766182E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047457562E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047739345E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047495005E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047717478E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047525063E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047699927E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047549409E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047686519E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047570207E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047677065E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047586403E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047668469E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047598974E-01 ( 0.1000000000E-05)
RMS change in effective potential (target) : 0.7047662415E-01 ( 0.1000000000E-05)
>The local field will be useful if you would like to converge antiferromagnetic structure or non collinear magnetic structure by >pointing such local field in different directions for different atoms.
OK, I will try to remember that, thank you.
As I understand, in calculations with non-collinear magnetism the directions of global and local
magnetic fields do not necessarily coincide, am I right?
And one more question, please: do the directions of local magnetic spins change during structural
relaxation or they remain the same as specified in the input file? The same concerns the global magnetic field.
Thank you again for your helpful hints and quick replies!
I appreciate it very much!
Regards,
Yvett
Hi again!
Yes your system is not converging at all, I'm almost sure that is because you're using to few kpoints.
What happen to the atom moment in INFO.OUT?
For what concern your question about the magnetic field and non collinear:
In general a small magnetic field is needed to break the symmetry and make your calculation converge to a magnetic solution.
So if you would like to get an antiferromagnetic solution you point the local magnetic field in opposite directions on two atoms.
When you start you should already have an idea of the magnetic structure that you would like to converge. To see which magnetic structure is more stable you may compare total energy of each of them.
In a collinear calculation the hamiltonian is divided in two blocks < up | H | up > and < down | H | down >, where up and doown refer to the spin up and spin down part of the Kohn-Sham wavefunctions, by doing so we allow only two directions for the magnetization: direction A and the -A .
In a non collinear calculation we also include the hamiltonian blocks < up | H | down > < up | H | down >, consequently the magnetization is allowed to vary along every arbitray direction. This is a more realsitic case, but more expensive about 8 times that the collinear case, since you diagoalize an N by N matrix instead that twice an N/2 by N/2 matrix.
I can give you this reference that give how magnetism is implemented in Exciting:
PRB 66 014447 (2002)
During structure relaxation the system is relaxed at every relaxation step to the ground state, so the moment will converge to the most stable solution corresponding to the volume that you are at.
Of course if you provided a local field on the same direction on all the magnetic atoms, the system will stay ferromagnetic during the relaxation.
Hope this will help!
Best
Francesco
Hi, Francesco!
Sorry for the late answer.
>What happen to the atom moment in INFO.OUT?
Well, the total moment from file MOMENT.OUT behaves the following:
-0.2492779377
-0.2617233067
-0.2983269619
-0.3205680582
-0.3434758464
-0.3719576227
-0.3681208006
-0.4179813196
-0.3998820885
-0.4640267947
-0.4407561651
-0.5117366578
-0.4852470782
-0.5613140656
-0.5313533733
-0.6120465696
-0.5783744821
-0.6635646869
-0.6261129397
-0.7158394090
-0.6745939910
-0.7689913037
-0.7234357815
-0.8224466744
-0.7730150510
-0.8767588719
-0.8230412266
-0.9315275515
-0.8738125827
-0.9872042240
-0.9250102252
-1.043316378
-0.9767857761
-1.100088218
-1.028960929
-1.157190900
-1.081888036
-1.215595354
-1.135450277
-1.274269294
-1.189198434
-1.333312015
-1.243697554
-1.393373175
-1.298775417
-1.453672189
-1.354374300
-1.514892128
-1.410355615
-1.576565463
-1.467340710
-1.639031937
-1.524408072
-1.701772270
-1.581979157
-1.765107594
-1.640329542
-1.829334286
-1.698699481
-1.893535685
-1.757827874
-1.958569527
-1.817697950
-2.023729730
-1.876762629
-2.089785543
-1.937052912
-2.156109237
-1.998122518
-2.222410926
-2.058521976
-2.290059984
-2.120446230
-2.357287655
-2.181638446
-2.425819633
-2.244040753
-2.493472833
-2.305877385
-2.562764545
-2.369122210
-2.631305757
-2.432027736
-2.701936114
-2.495647760
-2.770774880
-2.558821596
-2.840670262
-2.622050608
-2.911609312
-2.686256270
-2.980836705
-2.750271636
-3.051655975
-2.813102338
-3.122249489
-2.878460364
-3.192808308
-2.942627774
-3.263784183
-3.006484682
-3.334347929
-3.070227896
-3.406003553
-3.136402383
-3.477421004
-3.200429343
-3.548232502
-3.264229178
-3.618665045
-3.327706201
-3.688710520
-3.392257445
-3.759841343
-3.455650054
-3.829865707
-3.519529562
-3.899259445
-3.581519282
-3.969043909
-3.643548746
-4.037346366
-3.706095377
-4.105089433
-3.766820873
-4.173449530
-3.829683417
-4.242707246
-3.888386143
-4.305948600
-3.947348251
-4.372338334
-4.004462549
-4.433704953
-4.061376190
-4.497706523
-4.116840152
-4.557277457
-4.171069848
-4.616954618
-4.225257409
-4.677908669
-4.277862705
-4.734163077
-4.331581164
-4.793311766
-4.378914417
-4.846663271
-4.431019762
-4.902230207
-4.480085276
-4.956193979
-4.528319433
-5.009230063
-4.575673727
-5.061131164
-4.616892428
-5.106078177
-4.663164373
-5.156802396
-4.706132919
-5.203891480
-4.744956546
-5.246332467
-4.787461174
-5.292813399
-4.829751269
-5.339214170
-4.872864375
-5.386636429
-4.915124419
-5.432839313
-4.952595383
-5.473792353
-4.991182829
-5.515868832
-5.027619380
-5.555656040
-5.065671707
-5.596091939
-5.102942241
-5.638109026
-5.140662664
-5.679245833
-5.176298012
-5.716988184
I have used the parameters beta0=0.05 and betamax=0.1, and k-points mesh without
shift, but the system still cannot converge, and I obtain strange
message in the log file on every scf-loop (strange, because in INFO.OUT every
scf-loop ends with a correct value of total charge, but maybe I miss something):
Warning(charge): total charge density incorrect for s.c. loop 1
Calculated : 443.0112466
Required : 442.0000000
I don't know yet where exactly the problem comes from, but I have to solve it.
Thank you very much for your help!
Best regards,
Yvett
Hi Yvett!
I'm running your input file right now after I have increased all relevant parameters, starting from number of kpoints. We will see in the next 24 hours what if the calculation is going to converge.
I'm curious to look at the eigenvalues in EIGVAL.OUT and at the core energies in EVALCORE.OUT, to see that all the local orbitals are set properly in the default species file.
By the way when something does not work in EXCITING I usually follow this procedure:
1) Check your structure, first of all by plotting it in Xcrysden (http://www.xcrysden.org)by using task 250 for example.
2) Increase all relevant numerical parameters, now I'm trying:
! mixing reduced
beta0
0.05
! number of plane waves increased
rgkmax
7.0
! default
gmaxvr
12.0
! number of empty states increased
nempty
10
! l-cutoff potential increased
lmaxvr
7
! l-cutoff hamiltonian increased
lmaxmat
7
! l-cutoff basis set increased
lmaxapw
10
! kpoint grid increaased
ngridk
7 7 10
! smearing reduced
swidth
0.005
3) Check the file EIGVAL.OUT and EVALCORE.OUT to see that everything is all right, you may check
the EXCITING FORUM topic "From valence to core electrons" in which I tried to give all the essential information on this topic and the setting of the species file.
Finally a trick to speed up convergence for magnetic systems
perform few iterations with a bigger magnetic field (local or global)
example:
task
0
maxscl
10
bfieldc
0.0 0.0 3.0
and then reduce it and converge starting from the previous STATE.OUT:
task
1
bfieldc
0.0 0.0 0.01
Concerning this you may also check the forum topic " bcc iron" answer number 2.
I'll write you as soon as I get some news.
In the meanwhile you may also check and plot your structure, if you didn't do it yet and trying to run with the same input setting I'm using.
Best
Francesco
Hi again!
By the way do you know what is the experimetal ground state magnetic structure of your system, is it the ferromagnetic one, or antiferromagnetic?
Best
Francesco
Hi, Francesco!
Sorry for this long silence, just wanted to check everything.
> By the way do you know what is the experimetal ground state magnetic structure of your system, is it the ferromagnetic one, or antiferromagnetic?
Well, in principle all I know from the previous pseudopotential calculations is that
collinear FM and non-collinear (with a small deviation [~6 deg] from collinearity between Mn spins
on different sites) are competing phases in the case of Mn5Ge3 (non-collinear phase is
0.001 ev/Mn stabler than that of a FM one).
But I tried to conduct non-collinear calculations also and met the same problem (I just chose
the direction of the local magnetic field to be (0.0, 0.1, 0.0)).
Now, concerning the structure. I checked the structure both by using task 250 and
by using spacegroup utility. Everithing seems fine with the structure (see below).
! Atomic positions generated by spacegroup version 1.1.02
! Hermann-Mauguin symbol : P63/mcm
! Hall symbol : -P 6c 2
! Schoenflies symbol : D6h^3
! space group number : 193
! lattice constants (a,b,c) : 13.57579254 13.57579254 9.548786149
! angles in degrees (ab,ac,bc) : 120.0000000 90.00000000 90.00000000
! number of conventional unit cells : 1 1 1
! reduction to primitive cell : F
! Wyckoff positions :
! species : 1, Mn.in
! 0.3333333300 0.6666666700 0.000000000
! 0.2390000000 0.000000000 0.2500000000
! species : 2, Ge.in
! 0.6030000000 0.000000000 0.2500000000
avec
13.57579254 0.000000000 0.000000000
-6.787896270 11.75698122 0.000000000
0.000000000 0.000000000 9.548786149
atoms
2 : nspecies
'Mn.in' : spfname
10 : natoms; atposl, bfcmt below
0.33333333 0.66666667 0.00000000 0.00000000 0.00000000 0.00000000
0.66666667 0.33333333 0.00000000 0.00000000 0.00000000 0.00000000
0.66666666 0.33333333 0.50000000 0.00000000 0.00000000 0.00000000
0.33333334 0.66666667 0.50000000 0.00000000 0.00000000 0.00000000
0.23900000 0.00000000 0.25000000 0.00000000 0.00000000 0.00000000
0.76100000 0.00000000 0.75000000 0.00000000 0.00000000 0.00000000
0.23900000 0.23900000 0.75000000 0.00000000 0.00000000 0.00000000
0.76100000 0.76100000 0.25000000 0.00000000 0.00000000 0.00000000
0.00000000 0.23900000 0.25000000 0.00000000 0.00000000 0.00000000
0.00000000 0.76100000 0.75000000 0.00000000 0.00000000 0.00000000
'Ge.in' : spfname
6 : natoms; atposl, bfcmt below
0.60300000 0.00000000 0.25000000 0.00000000 0.00000000 0.00000000
0.39700000 0.00000000 0.75000000 0.00000000 0.00000000 0.00000000
0.60300000 0.60300000 0.75000000 0.00000000 0.00000000 0.00000000
0.39700000 0.39700000 0.25000000 0.00000000 0.00000000 0.00000000
0.00000000 0.60300000 0.25000000 0.00000000 0.00000000 0.00000000
0.00000000 0.39700000 0.75000000 0.00000000 0.00000000 0.00000000
I'm currently doing ground-state calculations for FM phase with dense k-mesh,
but its really time consuming, I have to wait what happens.
Also I'm checking the files EIGVAL.OUT and EVALCORE.OUT according to your advice.
So, this is my current situation.
Thank you very much for your help and interest to the problem!
I will let you know as soon as I found the solution.
Kind regards,
Yvett
Dear Yvett,
I think me and Fredrik Bultmark found a solution for your problem: the number of empty states.
nempty in your calculation is 10, definetely too low, to contain all the electrons. In fact you have 10 Mn atoms whose d electrons significantly "migrates" from one spin channel another in a spin polarized calculation.
If you check your EIGVAL.OUT witn nempty=10 and the relative occupancies you'll see that the highest energy occupied state will have an occupancy significantly different than zero.
Since Mn has 5 empty states in the d shell (check http://www.webelements.com/webelements/elements/text/Mn/econ.html\), in the extreme case we could say that all 5 electrons of one spin channel will migrate to the empty states of the other spin channel, this extreme case will require (Number of Mn)*5 = 50 empty states to contain all your electrons.
We performed few tests with your system here at Uppsala and it seems nempty=25 or 30 is enough.
In addition also the mixing seems to play an important role in order to get rid of the oscillations.
It seems we are going to achieve convergence with the following the input:
tasks
0
spinpol
.true.
betamax
0.1
beta0
0.05
! small magnetic field (NOT in the z-direction)
bfieldc
0.0 0.0 0.01
! large number of empty states
nempty
25
avec
13.575797890 0.000000000 0.000000000
-6.787898945 11.756985849 0.000000000
0.000000000 0.000000000 9.548789913
! default species
sppath
'../../species/'
atoms
2 : nspecies
'Mn.in' : spfname
10 : natoms
0.33333333 0.66666667 0.00000000 0.0 0.0 0.0 : atposl, bfcmt
-0.33333333 -0.66666667 0.00000000 0.0 0.0 0.0
0.66666667 0.33333333 0.50000000 0.0 0.0 0.0
-0.66666667 -0.33333333 -0.50000000 0.0 0.0 0.0
0.23900000 0.00000000 0.25000000 0.0 0.0 0.0
-0.23900000 0.00000000 -0.25000000 0.0 0.0 0.0
0.00000000 0.23900000 0.25000000 0.0 0.0 0.0
0.00000000 -0.23900000 -0.25000000 0.0 0.0 0.0
-0.23900000 -0.23900000 0.25000000 0.0 0.0 0.0
0.23900000 0.23900000 -0.25000000 0.0 0.0 0.0
'Ge.in' : spfname
6 : natoms
0.60300000 0.00000000 0.25000000 0.0 0.0 0.0
-0.60300000 0.00000000 -0.25000000 0.0 0.0 0.0
0.00000000 0.60300000 0.25000000 0.0 0.0 0.0
0.00000000 -0.60300000 -0.25000000 0.0 0.0 0.0
-0.60300000 -0.60300000 0.25000000 0.0 0.0 0.0
0.60300000 0.60300000 -0.25000000 0.0 0.0 0.0
autormt
.true.
rgkmax
6.0
ngridk
3 3 6
vkloff
0.05 0.05 0.05
You may keep the kpoint offset since it usually helps convergence by respect to kpoint mesh size.
I'm performing also a test with few iterations with a big magnetic field and then reduce it to check if this is going to reduce the number of total iterations
Rememeber to converge all parameters, especially kpoint number and rgkmax, before productive calculation.
Hope this will finally solve the problem!
Best
Francesco and Fredrik
Dear Francesco and Frederic,
many-many thanks to you for your help!
I am impressed with such an enormous assistance!
I appreciate it very much!
>We performed few tests with your system here at Uppsala and it seems nempty=25 or 30 is enough.
Maximum which I could try to test the convergence in a reasonable time was nempty=20.
This could be the reason for poor convergence thus.
I will try to do calculations with the parameters
which you found to be appropriate and let you know
how it works. And of course, the "production"
calculations I will start only after I check the
convergence with respect to all important parameters.
Thank you very much once again!
Hope, once I can be helpful to you too. :-)
All my best wishes to you,
Yvett
Dear Francesco,
sorry for being somewhat annoying.
But could you please show me the values of total energy
from the TOTENERGY.OUT file which you got
while testing the Mn5Ge3 system?
I tried to do some calculations with the parameters
which you found to be appropriate for convergence, but
it seems that even with a high value of nempty and ngridk
I still can't get the converged results.
Thank you very much in advance!
Kind regards,
Yvett
Hello Yvett,
I checked better, and I saw that you are perfectly right, with the file I sent you I get a convergence about 1.d-4 and then the system oscillates again.
This should be the mixing parameter to be decreased, let's try with decreased betamax and a larger field
tasks
1
10
atoms
2 : nspecies
'Mn.in' : spfname
10 : natoms
0.33333333 0.66666667 0.00000000 0.0 0.0 0.0 : atposl, bfcmt
-0.33333333 -0.66666667 0.00000000 0.0 0.0 0.0
0.66666667 0.33333333 0.50000000 0.0 0.0 0.0
-0.66666667 -0.33333333 -0.50000000 0.0 0.0 0.0
0.23900000 0.00000000 0.25000000 0.0 0.0 0.0
-0.23900000 0.00000000 -0.25000000 0.0 0.0 0.0
0.00000000 0.23900000 0.25000000 0.0 0.0 0.0
0.00000000 -0.23900000 -0.25000000 0.0 0.0 0.0
-0.23900000 -0.23900000 0.25000000 0.0 0.0 0.0
0.23900000 0.23900000 -0.25000000 0.0 0.0 0.0
'Ge.in' : spfname
6 : natoms
0.60300000 0.00000000 0.25000000 0.0 0.0 0.0
-0.60300000 0.00000000 -0.25000000 0.0 0.0 0.0
0.00000000 0.60300000 0.25000000 0.0 0.0 0.0
0.00000000 -0.60300000 -0.25000000 0.0 0.0 0.0
-0.60300000 -0.60300000 0.25000000 0.0 0.0 0.0
0.60300000 0.60300000 -0.25000000 0.0 0.0 0.0
! larger field
bfieldc
0.0 0.0 -3.0
avec
13.575797890 0.000000000 0.000000000
-6.787898945 11.756985849 0.000000000
0.000000000 0.000000000 9.548789913
spinpol
.true.
autormt
.true.
beta0
0.02
betamax
0.02
rgkmax
6.5
gmaxvr
12.0
nempty
25
lmaxvr
7
lmaxmat
7
lmaxapw
10
epspot
1.e-6
ngridk
3 3 6
swidth
0.005
if this converges we will decrease the field starting from the converged solution with a large field.
I will try this myself also!
Best
Francesco