dyzheng
commented
2 years ago I have retest Se-dimer with 1.9 distance and find that ks_solver cg lead to -513.726514779eV and ks_solver dav lead to -512.815324635547eV.
Note: I used nbands 20 for enhance robust of diagonalization.
(Screen output)
-------------------------------------------
SELF-CONSISTENT :
-------------------------------------------
ITER ETOT(eV) EDIFF(eV) DRHO TIME(s)
DA1 -5.129074e+02 0.000000e+00 5.686e-01 1.708e+01
DA2 -5.132766e+02 -3.692763e-01 2.566e-01 2.247e+00
DA3 -5.092154e+02 4.061274e+00 2.380e+00 4.831e+00
DA4 -5.126262e+02 -3.410865e+00 1.612e-01 2.568e+00
DA5 -5.125761e+02 5.009499e-02 1.215e-01 1.975e+00
DA6 -5.127251e+02 -1.489932e-01 8.579e-04 2.138e+00
DA7 -5.127622e+02 -3.704226e-02 2.215e-03 7.909e+00
DA8 -5.128021e+02 -3.994833e-02 1.974e-03 1.972e+00
DA9 -5.128077e+02 -5.573774e-03 3.679e-04 3.873e+00
DA10 -5.128100e+02 -2.264354e-03 3.571e-04 2.283e+00
DA11 -5.128064e+02 3.555817e-03 1.578e-04 2.496e+00
DA12 -5.128107e+02 -4.335449e-03 1.389e-04 3.007e+00
DA13 -5.128129e+02 -2.146815e-03 6.801e-05 5.311e+00
DA14 -5.128147e+02 -1.833987e-03 1.673e-05 3.140e+00
DA15 -5.128154e+02 -6.958200e-04 4.094e-06 3.230e+00
DA16 -5.128154e+02 -3.001393e-05 6.727e-07 2.666e+00
DA17 -5.128155e+02 -1.552457e-05 4.030e-07 3.420e+00
DA18 -5.128155e+02 -2.462081e-06 1.362e-07 2.817e+00
DA19 -5.128155e+02 -1.326276e-06 1.358e-08 2.618e+00(Total energy and band energys)
STATE ENERGY(eV) AND OCCUPATIONS NSPIN == 1
1/1 kpoint (Cartesian) = 0 0 0 (33761 pws)
1 -20.9341 2.00000
2 -14.5414 2.00000
3 -8.61194 2.00000
4 -8.21075 2.00000
5 -8.21070 2.00000
6 -4.08544 1.00005
7 -4.08543 0.999946
8 -3.49056 6.27934e-10
9 -1.97658 0.00000
10 -1.97655 0.00000
11 -1.56123 0.00000
12 -1.56121 0.00000
13 -1.53215 0.00000
14 -1.53212 0.00000
15 -1.31344 0.00000
16 -1.31343 0.00000
17 -0.302143 0.00000
18 -0.0417651 0.00000
19 0.0526112 0.00000
20 0.883383 0.00000
--------------------------------------------
!FINAL_ETOT_IS -512.8154568691700 eV
--------------------------------------------
mohanchen
Describe the bug
ABACUS v2.3.0 and 2.0.0 produced different converged PW energes for Se dimer structure.
for v2.0.0: charge density convergence is achieved final etot is -512.815324635547 eV
STATE ENERGY(eV) AND OCCUPATIONS. 1/1 kpoint (Cartesian) = 0.00000 0.00000 0.00000 (13507 pws) [spin1_state] 1 -20.934722 2.000000 [spin1_state] 2 -14.542070 2.000000 [spin1_state] 3 -8.612504 2.000000 [spin1_state] 4 -8.211223 2.000000 [spin1_state] 5 -8.211148 2.000000 [spin1_state] 6 -4.085818 1.000268 [spin1_state] 7 -4.085753 0.999732 [spin1_state] 8 -3.489782 0.000000 [spin1_state] 9 -1.978077 0.000000 [spin1_state] 10 -1.977633 0.000000
for 2.3.0: charge density convergence is achieved final etot is -513.726514779 eV
STATE ENERGY(eV) AND OCCUPATIONS NSPIN == 1 1/1 kpoint (Cartesian) = 0 0 0 (13506 pws) 1 -18.7319 2.00000 2 -12.8708 2.00000 3 -7.20274 2.00000 4 -7.20257 2.00000 5 -6.95123 2.00000 6 -3.32170 1.28991 7 -3.19708 0.355689 8 -3.19684 0.354400 9 -0.284140 0.00000 10 0.369519 0.00000
Expected behavior
No response
To Reproduce
The input fils and output files: spin1_abacus-pw_2.0vs3.0.zip
Environment
The following modulefiles were used in Hanhai20@USTC:
for abacus 2.0.0: module load intel/2018.update4 intelmpi/2018.update4 mkl/2018.update4 module load /opt/Modules/lib/elpa/2016.05.004/intel-2018.update4
for abacus 3.0.0: module load intel/2019.update5 intelmpi/2019.update5 mkl/2019.update5 gcc/9.2.0 elpa/2021.05.002/intelmpi2018
Additional Context
No response