How to decide ewidth, edelt, and bzqlty

Posted on : April 21, 2017 (Fri) 01:20:57

by 池田敦俊

Dear AkaiKKR users,

I am very sorry for asking many questions.
Since I'm not a specialist of the calculations of electronic states, every calculation gives me another question.

In this great blog (http://gomisai.blog75.fc2.com/blog-entry-591.html), the total energy of copper as a function of bzqlty is shown, and we see that the energy remains somewhat unchanged with bzqlty >= 8.
I tried similar check on ewidth, edelt, and also bzqlty.
However, no such convergence of energy is obtained in my case.
I'm wondering if there is an objective way to decide the proper values for these parameters.
Maybe I am too nervous of the parameters.

Here I will explain what I did and what I got.
Detailed input file is attached in the end of this post.

First, I calculated with edelt=0.001, ewidth=4.0, and bzqlty=4.
Then I found that the bottoms of the energy bands are located at 2.7, 1.9, 1.5, 0.7, and 0.4 Ry below the Fermi energy.
(I don't know about the bands below -3 Ry, which are out of range of the DOS calculation.)

I fixed edelt=0.001 and bzqlty=4, and decreased ewidth.
The obtained total energies are as follows:
-------------------
ewidth energy
2.7 -31582.4764466
1.9 -31582.5312236
1.5 -31582.5662046
0.7 -31582.5865802
0.4 -31571.8323648
-------------------
If we neglect ewidth=0.4 Ry, the energy gets larger and larger if I increase ewidth.
I don't know where to stop.
To my eyes, the band just above -2.7 Ry looks fairly narrow, or the density of state is sharp, suggesting that this band is a core state.
Therefore I chose to set ewidth=1.9, but I'm not sure this is the best value or not.

Next, I fixed edelt=0.001 and ewidth=1.9, and increased bzqlty.
The obtained total energies are as follows:
------------------
bzqlty energy
4 -31582.5312017
5 -31582.5322586
6 -31582.5312146
7 -31582.5324347
8 -31582.5325331
9 -31582.5325039
10 -31582.5333232
11 -31582.5332628
------------------
Without bzqlty=10 and 11, I would think that bzqlty=8 is enough, but the energy suddenly drops at bzqlty=10.
I chose bzqlty=10, but I'm afraid that the energy may drop again at bzqlty=12, possibly 20 or even larger.

Finally, I fixed ewidth=1.9 and bzqlty=10, and changed edelt from 0.01 to 0.0001.
The obtained total energies are as follows:
----------------------
edelt energy
0.01 -31582.5609261
0.005 -31582.5517688
0.002 -31582.5355211
0.001 -31582.5308982
0.0005 -31582.5409966
0.0002 -31582.5634297
0.0001 -31582.5809119
----------------------
The energy has a maximum at edelt=0.001.
I have no idea why this maximum appears nor which edelt is the best.
Is the smaller the better?

I would really appreciate it if someone could answer this very long question.

Sincerely,
Ikeda


Below is an example of my calculation.
I changed edelt, ewidth, and bzqlty of this input.
c----------------------Sr3SnO--------------------------------
go data/Sr3SnO_ewidthScan
c------------------------------------------------------------
c brvtyp a c/a b/a alpha beta gamma
sc 9.71206 , , , , , ,
c------------------------------------------------------------
c edelt ewidth reltyp sdftyp magtyp record
0.001 2.7 srals pbe nmag 2nd
c------------------------------------------------------------
c outtyp bzqlty maxitr pmix
update 4 100 0.03
c------------------------------------------------------------
c ntyp
3
c------------------------------------------------------------
c type ncmp rmt field mxl anclr conc
Sr 1 1 0.0 2 38 100
Sn 1 1 0.0 2 50 100
O 1 1 0.0 2 8 100
c------------------------------------------------------------
c natm
5
c------------------------------------------------------------
c atmicx atmtyp
0 0 0 Sn
0 0.5 0.5 Sr
0.5 0 0.5 Sr
0.5 0.5 0 Sr
0.5 0.5 0.5 O
c------------------------------------------------------------

[Re:07] How to decide ewidth, edelt, and bzqlty

Posted on : April 25, 2017 (Tue) 00:48:58

by 池田敦俊

Dear Dr. Gomi,

Thank you for the answer.
OK, I will use large ewidth if I'm not sure.
That is safer because treating the valence bands as core states will be worse.
Thank you again for your kindness.

Sincerely,
Ikeda

[Re:06] How to decide ewidth, edelt, and bzqlty

Posted on : April 24, 2017 (Mon) 22:55:20

by Hitoshi GOMI

Hello Ikeda,

Yes, it is sometimes not certain.
If so, you can increase the ewidth to include the band.
AkaiKKR will automatically treat the band as valence.

Hitoshi GOMI

[Re:05] How to decide ewidth, edelt, and bzqlty

Posted on : April 24, 2017 (Mon) 16:48:01

by 池田敦俊

Dear Dr. Gomi,

Well, it doesn't seem that I'm making my question understood.

When I look at the density of states, Sr3SnO has five bunches of bands down to -3 Ry.
Bch 1: from 0 to -0.31 Ry
Bch 2: from -0.48 to -0.59 Ry
Bch 3: from -1.1 to -1.4 Ry
Bch 4: from -1.65 to -1.84 Ry
Bch 5: from -2.5 to -2.64 Ry
(It's a pity that I cannot show a picture here.)

Is bch 1 core state or valence band? -> I'm sure it's valence.
Is bch 2 core state or valence band? -> Not sure.
Is bch 3 core state or valence band? -> Probably valence.
Is bch 4 core state or valence band? -> Not sure.
Is bch 5 core state or valence band? -> Probably core.

If I could clearly answer these questions, where to set ewidth would be trivial.
As you and the manual say, it's between the core states and valence bands.
My problem is that I cannot answer the questions, like indicated by "Not sure."

How do you know a certain bunch is core state or valence bands?
Is it written somewhere in the output "below here are the core states and above here are the valence bands"?
Is there any threshold of the bunch width?
For example, if bunches narrower than 0.1 Ry can be treated as core states and broader than 0.1 Ry should be treated as valence bands, the threshold is 0.1 Ry.

Sincerely,
Ikeda

[Re:04] How to decide ewidth, edelt, and bzqlty

Posted on : April 22, 2017 (Sat) 18:17:50

by Hitoshi GOMI

Hello Ikeda,

> However, when there are four or five bands, it is not so easy for me to tell where the boundary between core and valence is.

If core states is very close to the bottom of valence bands, you can increase the ewidth to include the core states.

Hitoshi GOMI

[Re:03] How to decide ewidth, edelt, and bzqlty

Posted on : April 22, 2017 (Sat) 01:11:15

by 池田敦俊

Dear Dr. Gomi and KKR Administrators,

Thank you for the answers.

So, Dr. Gomi suggests that I should focus on the convergence of the physical properties of my interest rather than the convergence of the energy, which is less meaningful by itself.
That is reasonable.
I'm happy that the convergence of the properties is obtained more easily.

Both of you know from experience that edelt=0.001 Ry is good.
Then I will also use edelt=0.001 Ry unless I want to calculate some properties related to fine structures.
It is interesting that edelt corresponds to the energy resolution.

ewidth seems to be a problem since its proper value is material dependent.
I know that ewidth should be between the core states and valence bands.
However, when there are four or five bands, it is not so easy for me to tell where the boundary between core and valence is.
Is it the only way to look at DOS and compare the band width of each band?

Sincerely,
Ikeda

[Re:02] How to decide ewidth, edelt, and bzqlty

Posted on : April 21, 2017 (Fri) 20:28:16

by Administrator

Dear Ikeda,

The smaller the delta, the larger the number of k-points needed. The advantage of using a smaller delta is that can retain the finer structure of the band (delta corresponds to the inverse life time of the states or the energy resolution). However, using small delta obviously requires large number k-points to obtain the overall structure of energy bands. As Hitoshi wrote, delta=0.001 would be a good choice in most cases. On the other hand ewidth is not a parameter that controls the accuracy or convergence. It is a quantity that must be set properly: in some cases 0.7Ry would be fine but another case may requires as large as 2Ry.

Best regards,

KKR Administrators

[Re:01] How to decide ewidth, edelt, and bzqlty

Posted on : April 21, 2017 (Fri) 18:09:05

by Hitoshi GOMI

Hello Ikeda,

Ideally speaking, a larger value is preferable for bzqlty. For edelt, a smaller value is preferable.
But practically speaking, you can use moderate value, depending on your purpose. For example, the equilibrium lattice parameter and the bulk modulus of Cu can be reasonably calculated with bzqlty=3, even though the total energy is not well converged with respect to the bzqlty value.
http://gomisai.blog75.fc2.com/blog-entry-593.html

I usually use edelt=0.001 Ry. Then, I calculate the physical properties (e.g. bulk modulus) with changing the bzqlty, to see the convergence.



For ewidth, neither too large nor too small value is preferable.
As is written in the manual, the range of ewidth must contain all of the valence band. In other words, the bottom of ewidth should locate within a band gap.
http://gomisai.blog75.fc2.com/blog-entry-708.html
http://gomisai.blog75.fc2.com/blog-entry-709.html



In most cases, the absolute value of the total energy is not matter. We only interested in the relative change with changing something (e.g. lattice parameters).
Therefore, the most important point is to keep calculation settings unchanged throughout the series of calculations.

Hitoshi GOMI