PBE band structures diagonalization of H(R) and nscf calculation are different

[1041176461]

Describe the Bug

@PeizeLin and I compare PBE band structures from diagonalization of H(R) matrix and nscf calculation, among three versions of ABACUS (21-12-04, 22-04-25 and 22-07-11). We found that results of version 22-07-11 are different from the other two, especically, hr-results of 22-07-11 are different from nscf-results of 22-07-11 while such a big difference do not appear in the old versions. Here we provide the pairwise MAEs: and input files: input.zip

[1041176461]

Test

I did SCF calculation first and got a charge file SPIN1_CHG. Then I used four settings to compare 'nscf' and 'HR' bands:

1. NSCF: out_band=1 based on SPIN1_CHG
2. SCF: out_band=1, scf_nmax=1 based on SPIN1_CHG
3. NSCF: out_band=1,out_mat_hs2=1 based on SPIN1_CHG
4. SCF: out_mat_hs2=1 + External diagonalization

Results

Band structure obtained from setting 1 and 2 are same, and setting 3 and 4 are same. However, setting 1(2) and 3(4) are different, their MAE is 0.003678632989845569 eV, which is a large error.

Possible causes

I guess there are some additional pre-processing when out_mat_hs2=1 is set.

Files

Here are my input and result files test.zip

[1041176461]

I retested the parameters, I found that the error originates from the keyword out_mat_r, rather than out_mat_hs2. Here I write the final band data as a tag

1. out_mat_hs2=0 and out_mat_r=0 -> 11.539922
2. out_mat_hs2=0 and out_mat_r=1 -> 11.550836
3. out_mat_hs2=1 and out_mat_r=0 -> 11.539922

Here are my new result files retest.zip

[jingan-181]

When setting out_mat_r=1, in module_orbital/ORB_read.cpp

if(out_mat_r)
{
    kmesh = kmesh * 4;
} 

This kemsh is the cause of the different band results.