degaussw convergency problem

[m.alidoosti]

Dear EPW users,

I am working on phosphorene nanoribbons which are passivated by H atoms. I have some problems with my calculations.
1) I tried to accurately optimize degaussw and nkf1(1D). I changed nkf1 from 12000 to 180000 kpoints for each value of degaussw. The results showed that nkf1=12000 is sufficient for a certain value of degaussw, while decreasing the value of degaussw leads to a substantial drop in the calculated value of lambda. The main drop occurs for degaussw= 0.0001. The following data are obtained results of calculations :

#dgaussw #lambda(nkf1=12000) #lambda(nkf1=60000) #lambda(nkf1=180000)
0.0025 # # 14.1790055
0.001 # # 14.3567454
0.0005 14.3829390 14.3567454 14.3829390
0.0004 14.3719485 14.3719485 14.3719485
0.0003 13.9956692 13.9956692 13.9956692
0.00025 # # 12.9528112
0.0002 10.2579702 10.2579702 10.2579702
0.0001 1.3090617 1.3090616 1.3090616

To what order of degaussw should I go down? I mean that how small the degaussw should be?

Thanks in advance for your time

P.S. I used delta_approx = true in all my inputs and Fsthick was converged at 0.01.

[sponce]

Dear m.alidoosti,

A small value would be 1meV (0.001).

Note that you need larger grids to converge smaller degauss values.

Best wishes,
Samuel

[m.alidoosti]

Dear Dr. Ponce,

Thanks for your reply. As you recommended, I did some calculations for the different values of degaussw such as 90, 70, 50, 30, 10 meV. Surprisingly by decreasing the value of degaussw, the value of the lambda diverges from 3 to 13 . That is, lambda is degaussw-dependent strikingly. ?while we expect to when we decrease degausw and increase nkf1 Simultaneously ( degaussw \propto nkpoint^{-\frac{1}{3}}), all results lead to an about constant value for lambda. Here are our results for different nkf1 and nqf1 at each degaussw:

degausw= 10 meV nqf1= 5000 nkf1 = 25000 lambda : 13.3604018
degausw= 30 meV nqf1= 5000 nkf1 = 25000 lambda : 5.7027918
degausw= 50 meV nqf1= 5000 nkf1 = 25000 lambda : 3.8236447
degausw= 70 meV nqf1= 5000 nkf1 = 25000 lambda : 3.2729306
degausw= 90 meV nqf1= 5000 nkf1 = 25000 lambda : 3.1053587

degausw= 10 meV nqf1= 10000 nkf1 = 50000 lambda : 13.3438190
degausw= 30 meV nqf1= 10000 nkf1 = 50000 lambda : 5.6977669
degausw= 50 meV nqf1= -10000 nkf1 = 50000 lambda : 3.8209117
degausw= 70 meV nqf1= 10000 nkf1 = 50000 lambda : 3.2710280
degausw= 90 meV nqf1= 10000 nkf1 = 50000 lambda : 3.1038707

degausw= 10 meV nqf1= 20000 nkf1 = 100000 lambda : 13.3520595
degausw= 30 meV nqf1= 20000 nkf1 = 100000 lambda : 5.7002592
degausw= 50 meV nqf1= 20000 nkf1 = 100000 lambda : 3.8222639
degausw= 70 meV nqf1= 20000 nkf1 = 100000 lambda : 3.2719569
degausw= 90 meV nqf1= 20000 nkf1 = 100000 lambda : 3.1045785

degaussq= 0.05 meV (default) is used for above results.
Are these values sufficient for nkf1 and nqf1 to do this kind of calculation? Should I consider many more points (nkf1 and nqf1) for achieving a convergent lambda?
If the answer to the above question is NO, what is the cause of lambda divergency when we decrease the value of degaussw despite increasing nkf1?

[sponce]

Dear m.alidoosti,

Indeed it seems quite an unexpected behavior. However I have never done any 1D calculations so it might be due to this.

In your original post, what was your q-point grids?
Your results there seems quite different from the results you have now.
You got a lambda of 1.3 with dense k-point grid and a delta of 0.1 meV, much lower than the current \lambda=13.3 that you have with a delta of 1 meV.

Best wishes,
Samuel

[m.alidoosti]

In the original post, nqf1 =1000 was used as a converged value. If we calculate \lambda with degaussw 0.00005 and high kgrid, we will find the value of about 9. So, it has strange behaviour. It is worthwhile to note that there is a sharp peak in DOS at the VBM and I didn't obtain converged DOS despite using the high dense Kpoint grid for different values of the degaussw.
degausw= 10 meV epw-10000q-50000k.out: DOS = 15.467486 states/spin/eV/Unit Cell
degausw= 30 meV epw-10000q-50000k.out: DOS = 8.517337 states/spin/eV/Unit Cell
degausw= 50 meV epw-10000q-50000k.out: DOS = 6.396111 states/spin/eV/Unit Cell
degausw= 70 meV epw-10000q-50000k.out: DOS = 5.064148 states/spin/eV/Unit Cell
degausw= 90 meV epw-10000q-50000k.out: DOS = 4.249077 states/spin/eV/Unit Cell
degausw= 10 meV epw-20000q-100000k.out: DOS = 15.467486 states/spin/eV/Unit Cell
degausw= 30 meV epw-20000q-100000k.out: DOS = 8.517337 states/spin/eV/Unit Cell
degausw= 50 meV epw-20000q-100000k.out: DOS = 6.396111 states/spin/eV/Unit Cell
degausw= 70 meV epw-20000q-100000k.out: DOS = 5.064148 states/spin/eV/Unit Cell
degausw= 90 meV epw-20000q-100000k.out: DOS = 4.249077 states/spin/eV/Unit Cell

when we used smaller degaussw for example 0.0001, 0.0003,...... , a converged DOS (about 22) obtained, but the values of \lambda didn't.
Is there any method to calculate \lambda independent from degauss?