Convergence problems for PdS2

Dear all,

I am trying to reproduce the band structure as well as the DOS of a PdS2 structure already relaxed in Quantum Expresso (QE) for the sake of obtaining the effective mass tensor; however, I am having convergence problems (Not convergung even after 200 iterations) and errors such as :

Warning(rhonorm): total charge density incorrect for s.c. loop 1
Calculated : 78.09296802
Required : 78.00000000

Warning(linengy): could not find 3 linearisation energies in s.c. loop 2

I have tried increasing the value of gmaxvr and the kpoints grid but the problems persist. Here is my input file :

! Band structure of silicon. The points listed after plot1d below are the
! vertices joined in the band structure.

tasks
0
20

! use Broyden mixing
mixtype
3

gmaxvr
14.0

xctype
20

autolinengy
.true.

autoswidth
.true.

msmooth
1

maxscl
500

nempty
8

avec
1.0 0.0 0.00
-0.5 0.866026 0.0
0.00 0.0 6.3

scale
6.70591

sppath
'/home/eamoujaes/elk-5.2.14/species/'

atoms
2 : nspecies
'Pd.in' : spfname
1 : natoms; atposl below
-0.001537 0.0 0.0

'S.in'
2
0.6651353 0.3332582 0.057769
0.3318771 -0.3332582 -0.057769

ngridk
14 14 1

! These are the vertices to be joined for the band structure plot
plot1d
4 700 : nvp1d, npp1d
0.0 0.0 0.0 : vlvp1d
0.3333 0.3333 0.00
0.5000 0.0000 0.000
0.0000 0.0000 0.0000

I know that QE and ELK are not equivalent but what is the best choice of xctype in ELK to reproduce similar results of QE, where a PBE Troullier Martins have been used?

Thanks in advance

Elio
University of Rondonia
Porto Velho
Brazil

mfechner - 2019-05-07

Dear Elio,

To Fix your computation you can

add first

demaxbnd
3.5

this will increase the search radius for the linearization energy and the message will be gone about this problem

next add

lradstp
2

which increases the radial meshes and hence the message with the charge density is gone. Still adding this the system behaves strangely. Are youre sure about the lattice constants ?
I found for PdS2 a=5.502 Ang == 10.3973 bohrradius. The units of lattice constants in elk are bohr ! Checking your structure I noticed scaleavec = 6.705911.0 bohr, which correspinds to a highly compressed state.

best regards
Michael

Elio - 2019-05-07

Dear Michael,

Thanks for your reply. I will use the flags you have suggested. I do know that lattice parameter is in Bohrs and not in agnstroms. My lattice parameter is around 3.51 angstroms (corresponding to around 6.63293 au). I have seen other pulications have a very close value. to this.

So what I have actually done is that in 'avec' section I have defined the lattice vectors for a monolayer trigonal symmetry as {1, 0, 0}, {-1/2, 0.866025,0}, {0, 0, c/a==6.3} and then defined the scale as the lattice parameter in bohrs.

However I am geting confused with the examples given in elk : For example, Al has a lattice parameter of ~ 4.04 angstroms. In the elk.in file , 'avec' and 'scale' are given by :

avec
1.0 1.0 0.0
1.0 0.0 1.0
0.0 1.0 1.0

scale
3.8267

where avec defined the FCC lattice vectors but the scale (which is is in fact the lattice parameter) is given in angstroms!

What am I missing here?

mfechner - 2019-05-07

Dear Elio,

In general you define the fcc lattice as

avec
0.5 0.5 0.0
0.5 0.0 0.5
0.0 0.5 0.5

scale
7.646==a

which is equivalent to the setting given in the example just one uses
avec
1 1 0
1 0 1
0 1 1

scale
3.8267==a/2

So both settings are the same, which you use is a matter of taste. You can check GEOMETRY.OUT for confirmation :-) if you like.
Back to your problem, I have run the computation further, and I have the feeling still something is not well here. For example, with "autolinengy=.true." the values you find in LINENGY get to out of control for my taste. Is this a 2D layer by the way ?

best regards
Michael

Elio - 2019-05-07

Right got your point about "a". Absolutely right.

Yes the values do get out of control. Yes it is a 2 D structure. Any special treatment for this?

Thanks once again for your reply.

Elio - 2019-05-08

The scf cycle didn't converge however the produced band structure looks correct :. So my only problem now is the convergence. How can that be solved?

Find attached the band structure

PdS2.jpg

Anonymous - 2019-05-08

Dear Elio,

is this an insulating material ?

My general suggestion would be first switch to a small linear mixing and increase gmaxvr.
like this

mixtype
1

beta0
0.01

betamax
0.20

gmaxvr
20.0

The increase of GMAXVR is just a feeling that it improves convergence.

best
Michael

Anonymous - 2019-05-08

There are two "anonymous" replies but they are still awaiting moderation. I cannot read the messages. Maybe someone posted a response and forgot to log in. Please post again if this is the case.

Elio - 2019-05-08

There are two anonymous replies awaiting moderation. I cannot see the messages. Maybe someone replied and forgot to log in. Please repost. Thanks

Elio - 2019-05-08

Dear Michael,

The material is a narrow gap semiconductor. I was able to converge by increasing demaxbnd to 10. However , the band strcuture was incorrect! I will try your suggestion and let you know.

Many thanks

Elio - 2019-05-11

I was able to produce the correct band structure. Thanks :-)

Myron Kapetanakis - 2019-06-10

Dear Elio,

Can you please explain how did you finally manage to get the correct band structure? I have a similar experience while trying to get the BS for a monolayer MoSe2. I could solve the non-converangence problem by increasing the demaxnbd parameter but I get unphysical results. For example I get a metallic MoSe2. So, I am not sure that the demaxnbd parameter can really solve the problem.

Thanks
Myron

Elio - 2019-06-12

Dear Myron,

Sorry for taking too long to reply. I just saw your message. In fact this is weird how a parameter changes the whole band structure and made me really skeptical about results obtained . What if the structure is new and one wants to obtain its properties. I have not actually resolved the PdSe2 problem. I ma getting it metalic as well! Maybe something is wrong with the Se file. It worked fine for PdS2. I am posting the PdS2 input file. Hope it helps. Try playing with the parameters. Sorry if I could not be of great help. I am using ELK just to obatin effective masses since QE does not have this feature.

! Band structure of silicon. The points listed after plot1d below are the
! vertices joined in the band structure.

tasks
25

! use Broyden mixing
mixtype
3

beta0
0.01

betamax
0.20

gmaxvr
20.0

xctype
20

demaxbnd
3.5

lradstp
2

stype
1

autoswidth
.true.

autolinengy
.true.

msmooth
1

maxscl
1500

nempty
8

avec
1.0 0.0 0.00
-0.5 0.866026 0.0
0.00 0.0 6.5

scale
6.63293

sppath
'/home/eamoujaes/elk-5.2.14/species/'

atoms
2 : nspecies
'Pd.in' : spfname
1 : natoms; atposl below
-0.001537 0.0 0.0

'S.in'
2
0.6651353 0.3332582 0.057769
0.3318771 -0.3332582 -0.057769

ngridk
14 14 1

! These are the vertices to be joined for the band structure plot
plot1d
4 700 : nvp1d, npp1d
0.0 0.0 0.0 : vlvp1d
0.3333 0.3333 0.00
0.5000 0.0000 0.000
0.0000 0.0000 0.0000

If you ever solve it , please be kind to let me know.

Regards

Myron Kapetanakis - 2019-06-13

Dear Elio,

Thank you for your response. I also tried MoS2 instead of MoSe2 and it seems to work fine. Again the calculation starts with the usual warning, about the linearisation energies, which disappears after just a couple of iterations. So, it seems the problem is at the Se atoms. I have tried the parameters you mentioned in your response and I have found that the most helpful one in the MoSe2 case is the lradstp=2. Also, I have seen that the parameter isgkmax=-4, which controls the MT radious to be used, improves the convergence too. However, I am not yet sure that the results are correct. I will let you know if I find something better.

Best,
Myron