abacus-develop
abacus-develop copied to clipboard
deepmodeling
Slow SCF in comparison with QE on CALYPSO generated random structures.
Details
I generated 12 structures (each in a different chemical composition) using CALYPSO (a structure-search software).
The overall average performance against QE is nearly 60% slower.
Please check the test result and files here: (Author Xingliang Peng) https://labs.dp.tech/projects/abacustest/?request=GET%3A%2Fapplications%2Fabacustest%2Fjobs%2Fjob-abacustest-v0.3.74-7cf445 Machine: 64 core 256G memory CPU, on paratera plaform. Number of Atoms: 20 (or 21 for PtAuAg)
Note that these structures are randomly generated. They are not 100% physically reasonable. But all tasks converge successfully. The structures may have very large pressures. These cases have a large number of k points
Task list for Issue attackers (only for developers)
- [ ] Reproduce the performance issue on a similar system or environment.
- [ ] Identify the specific section of the code causing the performance issue.
- [ ] Investigate the issue and determine the root cause.
- [ ] Research best practices and potential solutions for the identified performance issue.
- [ ] Implement the chosen solution to address the performance issue.
- [ ] Test the implemented solution to ensure it improves performance without introducing new issues.
- [ ] Optimize the solution if necessary, considering trade-offs between performance and other factors (e.g., code complexity, readability, maintainability).
- [ ] Review and incorporate any relevant feedback from users or developers.
- [ ] Merge the improved solution into the main codebase and notify the issue reporter.
@haozhihan, could you share more news about this issue here. I believe many developers and users are quite curious about the progress.
Update:
- This issue is originated from the fact that the dav method is slower than the one of QE. We are working on it.
- Based on test https://deepmodeling-activity.github.io/abacus-test.github.io/index.html?pname=dpa/dcu, we find the cost of DCU is much cheaper than CPU version. So we suggest the user to use DCU version now.
The code related to the new Davidson method has been merged into the current main repository code of abacus.
The new Davidson method is called dav_subspace, which has a significant performance improvement compared to previous Davidson methods.
And even in many cases, it has a faster calculation speed than QE's Davidson method.
@Carrotkingdom
@pxlxingliang could you make some test using dav_subspace method to check the efficiency.
I have tested one alloy case, and it seems that the new performance of dav_subspace is worse than old dav method.
scf_steps scf_time scf_time/step
dav/LiLaH_8 11 227.39 20.671818
dav-subspace/LiLaH_8 16 587.91 36.744375
The detail of SCF: dav
ITER ETOT(eV) EDIFF(eV) DRHO TIME(s)
DA1 -1.325515e+03 0.000000e+00 6.759e-01 6.460e+01
DA2 -1.326614e+03 -1.098380e+00 4.145e-01 2.133e+01
DA3 -1.327599e+03 -9.847214e-01 3.253e-01 1.348e+01
DA4 -1.328066e+03 -4.677166e-01 5.503e-03 1.235e+01
DA5 -1.328087e+03 -2.095905e-02 9.647e-05 2.294e+01
DA6 -1.328088e+03 -3.460400e-04 7.558e-05 1.963e+01
DA7 -1.328088e+03 -1.104796e-04 1.044e-06 1.221e+01
DA8 -1.328088e+03 -3.610688e-06 1.839e-07 1.960e+01
DA9 -1.328088e+03 -4.773487e-07 6.780e-08 1.424e+01
DA10 -1.328088e+03 -1.130312e-07 1.025e-08 1.179e+01
DA11 -1.328088e+03 -1.942902e-08 3.093e-10 1.522e+01
dav-subspace
ITER ETOT(eV) EDIFF(eV) DRHO TIME(s)
DS1 -1.319801e+03 0.000000e+00 6.081e-01 1.296e+02
DS2 -1.321970e+03 -2.168314e+00 2.709e-01 1.146e+01
DS3 -1.324139e+03 -2.169000e+00 2.677e-01 1.278e+01
DS4 -1.325194e+03 -1.054989e+00 2.845e-03 1.149e+01
DS5 -1.328076e+03 -2.882602e+00 4.846e-03 2.595e+02
DS6 -1.328064e+03 1.193250e-02 5.176e-03 2.206e+01
DS7 -1.328068e+03 -3.891617e-03 8.619e-04 1.171e+01
DS8 -1.328076e+03 -7.617757e-03 3.308e-05 2.322e+01
DS9 -1.328080e+03 -4.526490e-03 2.527e-05 2.424e+01
DS10 -1.328082e+03 -1.501581e-03 1.944e-05 1.244e+01
DS11 -1.328083e+03 -9.081024e-04 2.174e-05 1.167e+01
DS12 -1.328084e+03 -8.042899e-04 1.180e-06 1.157e+01
DS13 -1.328084e+03 -6.529929e-04 8.501e-07 1.162e+01
DS14 -1.328085e+03 -5.592638e-04 5.921e-08 1.159e+01
DS15 -1.328085e+03 -4.761198e-04 2.183e-08 1.147e+01
DS16 -1.328086e+03 -4.011135e-04 9.338e-09 1.149e+01
This is my test result, by adjusting the PW-DIAG-NDIM parameter to the same value as the old dav.
dav_subspace:
ABACUS v3.6.1
Atomic-orbital Based Ab-initio Computation at UStc
Website: http://abacus.ustc.edu.cn/
Documentation: https://abacus.deepmodeling.com/
Repository: https://github.com/abacusmodeling/abacus-develop
https://github.com/deepmodeling/abacus-develop
Commit: 43cde6d9e (Sun Apr 14 14:53:29 2024 +0800)
Tue Apr 16 19:18:26 2024
MAKE THE DIR : OUT.ABACUS/
RUNNING WITH DEVICE : CPU / Intel(R) Xeon(R) Gold 6130 CPU @ 2.10GHz
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Warning: the number of valence electrons in pseudopotential > 1 for Li: [He] 2s1
Warning: the number of valence electrons in pseudopotential > 3 for La: [Xe] 5d1 6s2
Pseudopotentials with additional electrons can yield (more) accurate outcomes, but may be less efficient.
If you're confident that your chosen pseudopotential is appropriate, you can safely ignore this warning.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
UNIFORM GRID DIM : 45 * 45 * 96
UNIFORM GRID DIM(BIG) : 45 * 45 * 96
DONE(0.0414046 SEC) : SETUP UNITCELL
DONE(0.0863787 SEC) : INIT K-POINTS
---------------------------------------------------------
Self-consistent calculations for electrons
---------------------------------------------------------
SPIN KPOINTS PROCESSORS
1 592 16
---------------------------------------------------------
Use plane wave basis
---------------------------------------------------------
ELEMENT NATOM XC
Li 2
La 1
H 5
---------------------------------------------------------
Initial plane wave basis and FFT box
---------------------------------------------------------
DONE(0.141307 SEC) : INIT PLANEWAVE
MEMORY FOR PSI (MB) : 127.477
DONE(0.263062 SEC) : LOCAL POTENTIAL
DONE(0.420346 SEC) : NON-LOCAL POTENTIAL
DONE(0.420392 SEC) : INIT BASIS
-------------------------------------------
SELF-CONSISTENT :
-------------------------------------------
START CHARGE : atomic
DONE(0.658796 SEC) : INIT SCF
ITER ETOT(eV) EDIFF(eV) DRHO TIME(s)
DS1 -1.325148e+03 0.000000e+00 6.714e-01 1.269e+02
DS2 -1.326307e+03 -1.159468e+00 3.099e-01 1.834e+01
DS3 -1.327350e+03 -1.042559e+00 2.857e-01 2.049e+01
DS4 -1.327867e+03 -5.175793e-01 8.317e-03 1.842e+01
DS5 -1.328083e+03 -2.152396e-01 2.184e-03 6.982e+01
DS6 -1.328083e+03 -6.010596e-04 7.426e-04 2.428e+01
DS7 -1.328086e+03 -3.213118e-03 2.674e-05 2.745e+01
DS8 -1.328087e+03 -8.680951e-04 8.392e-06 2.283e+01
DS9 -1.328087e+03 -2.311163e-04 1.420e-06 1.808e+01
DS10 -1.328088e+03 -1.017140e-04 2.372e-07 1.807e+01
DS11 -1.328088e+03 -5.239065e-05 2.240e-08 1.823e+01
DS12 -1.328088e+03 -3.283545e-05 5.809e-09 1.836e+01
----------------------------------------------------------------
TOTAL-STRESS (KBAR)
----------------------------------------------------------------
265.1634021231 -31.0062037473 60.8166564208
-31.0062037473 229.3627558415 105.3388149814
60.8166564208 105.3388149814 195.5904956706
----------------------------------------------------------------
TOTAL-PRESSURE: 230.038885 KBAR
TIME STATISTICS
-------------------------------------------------------------------------------------
CLASS_NAME NAME TIME(Sec) CALLS AVG(Sec) PER(%)
-------------------------------------------------------------------------------------
total 421.24 15 28.08 100.00
Driver reading 0.01 1 0.01 0.00
Input Init 0.00 1 0.00 0.00
Input_Conv Convert 0.00 1 0.00 0.00
Driver driver_line 421.22 1 421.22 100.00
UnitCell check_tau 0.00 1 0.00 0.00
PW_Basis_Sup setuptransform 0.01 1 0.01 0.00
PW_Basis_Sup distributeg 0.00 1 0.00 0.00
mymath heapsort 0.00 3 0.00 0.00
PW_Basis_K setuptransform 0.02 1 0.02 0.01
PW_Basis_K distributeg 0.00 1 0.00 0.00
PW_Basis setup_struc_factor 0.00 1 0.00 0.00
ppcell_vnl init 0.01 1 0.01 0.00
ppcell_vl init_vloc 0.11 1 0.11 0.03
ppcell_vnl init_vnl 0.16 1 0.16 0.04
WF_atomic init_at_1 0.00 1 0.00 0.00
wavefunc wfcinit 0.00 1 0.00 0.00
Ions opt_ions 420.76 1 420.76 99.89
ESolver_KS_PW run 401.55 1 401.55 95.33
H_Ewald_pw compute_ewald 0.05 1 0.05 0.01
Charge set_rho_core 0.00 1 0.00 0.00
Charge atomic_rho 0.17 1 0.17 0.04
PW_Basis_Sup recip2real 0.10 81 0.00 0.02
PW_Basis_Sup gathers_scatterp 0.08 81 0.00 0.02
Potential init_pot 0.01 1 0.01 0.00
Potential update_from_charge 0.15 13 0.01 0.04
Potential cal_fixed_v 0.00 1 0.00 0.00
PotLocal cal_fixed_v 0.00 1 0.00 0.00
Potential cal_v_eff 0.15 13 0.01 0.04
H_Hartree_pw v_hartree 0.01 13 0.00 0.00
PW_Basis_Sup real2recip 0.06 108 0.00 0.01
PW_Basis_Sup gatherp_scatters 0.04 108 0.00 0.01
PotXC cal_v_eff 0.13 13 0.01 0.03
XC_Functional v_xc 0.13 13 0.01 0.03
Potential interpolate_vrs 0.00 13 0.00 0.00
Charge_Mixing init_mixing 0.00 1 0.00 0.00
ESolver_KS_PW hamilt2density 401.09 12 33.42 95.22
HSolverPW solve 401.09 12 33.42 95.22
Nonlocal getvnl 8.73 7104 0.00 2.07
pp_cell_vnl getvnl 10.17 8288 0.00 2.41
Structure_Factor get_sk 1.97 27232 0.00 0.47
DiagoIterAssist diagH_subspace 6.36 592 0.01 1.51
Operator hPsi 250.68 40963 0.01 59.51
Operator EkineticPW 2.25 40963 0.00 0.53
Operator VeffPW 218.10 40963 0.01 51.78
PW_Basis_K recip2real 125.91 640657 0.00 29.89
PW_Basis_K gathers_scatterp 82.86 640657 0.00 19.67
PW_Basis_K real2recip 112.66 491473 0.00 26.75
PW_Basis_K gatherp_scatters 82.21 491473 0.00 19.52
Operator NonlocalPW 30.19 40963 0.00 7.17
Nonlocal add_nonlocal_pp 8.97 40963 0.00 2.13
DiagoIterAssist diagH_LAPACK 0.16 592 0.00 0.04
Diago_DavSubspace diag_once 330.99 7104 0.05 78.57
Diago_DavSubspace first 83.42 7104 0.01 19.80
Diago_DavSubspace cal_elem 19.90 40371 0.00 4.72
Diago_DavSubspace diag_zhegvx 45.74 40371 0.00 10.86
Diago_DavSubspace cal_grad 181.18 33267 0.01 43.01
Diago_DavSubspace check_update 0.03 33267 0.00 0.01
Diago_DavSubspace last 3.94 10338 0.00 0.93
Diago_DavSubspace refresh 1.38 3234 0.00 0.33
ElecStatePW psiToRho 33.01 12 2.75 7.84
Charge_Mixing get_drho 0.01 12 0.00 0.00
Charge_Mixing inner_product_recip_rho 0.00 12 0.00 0.00
Charge mix_rho 0.02 11 0.00 0.00
Charge Broyden_mixing 0.01 11 0.00 0.00
Charge_Mixing inner_product_recip_hartree 0.01 104 0.00 0.00
Forces cal_force_loc 0.00 1 0.00 0.00
Forces cal_force_ew 0.00 1 0.00 0.00
Forces cal_force_nl 3.02 1 3.02 0.72
Forces cal_force_cc 0.00 1 0.00 0.00
Forces cal_force_scc 0.18 1 0.18 0.04
Stress_PW cal_stress 16.01 1 16.01 3.80
Stress_Func stress_kin 0.21 1 0.21 0.05
Stress_Func stress_har 0.00 1 0.00 0.00
Stress_Func stress_ewa 0.00 1 0.00 0.00
Stress_Func stress_gga 0.01 1 0.01 0.00
Stress_Func stress_loc 0.46 1 0.46 0.11
Stress_Func stress_cc 0.00 1 0.00 0.00
Stress_Func stress_nl 15.32 1 15.32 3.64
ModuleIO write_istate_info 0.05 1 0.05 0.01
-------------------------------------------------------------------------------------
START Time : Tue Apr 16 19:18:26 2024
FINISH Time : Tue Apr 16 19:25:27 2024
TOTAL Time : 421
SEE INFORMATION IN : OUT.ABACUS/
dav:
ABACUS v3.6.1
Atomic-orbital Based Ab-initio Computation at UStc
Website: http://abacus.ustc.edu.cn/
Documentation: https://abacus.deepmodeling.com/
Repository: https://github.com/abacusmodeling/abacus-develop
https://github.com/deepmodeling/abacus-develop
Commit: 43cde6d9e (Sun Apr 14 14:53:29 2024 +0800)
Tue Apr 16 18:56:05 2024
MAKE THE DIR : OUT.ABACUS/
RUNNING WITH DEVICE : CPU / Intel(R) Xeon(R) Gold 6130 CPU @ 2.10GHz
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Warning: the number of valence electrons in pseudopotential > 1 for Li: [He] 2s1
Warning: the number of valence electrons in pseudopotential > 3 for La: [Xe] 5d1 6s2
Pseudopotentials with additional electrons can yield (more) accurate outcomes, but may be less efficient.
If you're confident that your chosen pseudopotential is appropriate, you can safely ignore this warning.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
UNIFORM GRID DIM : 45 * 45 * 96
UNIFORM GRID DIM(BIG) : 45 * 45 * 96
DONE(0.0874296 SEC) : SETUP UNITCELL
DONE(0.128312 SEC) : INIT K-POINTS
---------------------------------------------------------
Self-consistent calculations for electrons
---------------------------------------------------------
SPIN KPOINTS PROCESSORS
1 592 16
---------------------------------------------------------
Use plane wave basis
---------------------------------------------------------
ELEMENT NATOM XC
Li 2
La 1
H 5
---------------------------------------------------------
Initial plane wave basis and FFT box
---------------------------------------------------------
DONE(0.181191 SEC) : INIT PLANEWAVE
MEMORY FOR PSI (MB) : 127.477
DONE(0.303691 SEC) : LOCAL POTENTIAL
DONE(0.463623 SEC) : NON-LOCAL POTENTIAL
DONE(0.463668 SEC) : INIT BASIS
-------------------------------------------
SELF-CONSISTENT :
-------------------------------------------
START CHARGE : atomic
DONE(0.702524 SEC) : INIT SCF
ITER ETOT(eV) EDIFF(eV) DRHO TIME(s)
DA1 -1.325515e+03 0.000000e+00 6.759e-01 1.112e+02
DA2 -1.326614e+03 -1.098380e+00 4.145e-01 4.754e+01
DA3 -1.327599e+03 -9.847214e-01 3.253e-01 2.952e+01
DA4 -1.328066e+03 -4.677166e-01 5.503e-03 2.752e+01
DA5 -1.328087e+03 -2.095905e-02 9.647e-05 5.057e+01
DA6 -1.328088e+03 -3.460438e-04 7.557e-05 4.132e+01
DA7 -1.328088e+03 -1.104754e-04 1.044e-06 2.616e+01
DA8 -1.328088e+03 -3.610345e-06 1.834e-07 4.222e+01
DA9 -1.328088e+03 -4.759293e-07 6.776e-08 3.105e+01
DA10 -1.328088e+03 -1.130938e-07 1.026e-08 7.592e+01
DA11 -1.328088e+03 -1.948084e-08 3.094e-10 3.199e+01
----------------------------------------------------------------
TOTAL-STRESS (KBAR)
----------------------------------------------------------------
265.3484013815 -31.0439570392 60.8170545107
-31.0439570392 229.5075403446 105.3378429642
60.8170545107 105.3378429642 195.6297038382
----------------------------------------------------------------
TOTAL-PRESSURE: 230.161882 KBAR
TIME STATISTICS
-------------------------------------------------------------------------------------
CLASS_NAME NAME TIME(Sec) CALLS AVG(Sec) PER(%)
-------------------------------------------------------------------------------------
total 535.57 15 35.70 100.00
Driver reading 0.03 1 0.03 0.01
Input Init 0.02 1 0.02 0.00
Input_Conv Convert 0.00 1 0.00 0.00
Driver driver_line 535.54 1 535.54 99.99
UnitCell check_tau 0.00 1 0.00 0.00
PW_Basis_Sup setuptransform 0.01 1 0.01 0.00
PW_Basis_Sup distributeg 0.00 1 0.00 0.00
mymath heapsort 0.00 3 0.00 0.00
PW_Basis_K setuptransform 0.02 1 0.02 0.00
PW_Basis_K distributeg 0.00 1 0.00 0.00
PW_Basis setup_struc_factor 0.00 1 0.00 0.00
ppcell_vnl init 0.01 1 0.01 0.00
ppcell_vl init_vloc 0.11 1 0.11 0.02
ppcell_vnl init_vnl 0.16 1 0.16 0.03
WF_atomic init_at_1 0.00 1 0.00 0.00
wavefunc wfcinit 0.00 1 0.00 0.00
Ions opt_ions 535.01 1 535.01 99.90
ESolver_KS_PW run 515.29 1 515.29 96.21
H_Ewald_pw compute_ewald 0.05 1 0.05 0.01
Charge set_rho_core 0.00 1 0.00 0.00
Charge atomic_rho 0.17 1 0.17 0.03
PW_Basis_Sup recip2real 0.09 75 0.00 0.02
PW_Basis_Sup gathers_scatterp 0.08 75 0.00 0.01
Potential init_pot 0.01 1 0.01 0.00
Potential update_from_charge 0.15 12 0.01 0.03
Potential cal_fixed_v 0.00 1 0.00 0.00
PotLocal cal_fixed_v 0.00 1 0.00 0.00
Potential cal_v_eff 0.15 12 0.01 0.03
H_Hartree_pw v_hartree 0.01 12 0.00 0.00
PW_Basis_Sup real2recip 0.06 100 0.00 0.01
PW_Basis_Sup gatherp_scatters 0.04 100 0.00 0.01
PotXC cal_v_eff 0.13 12 0.01 0.02
XC_Functional v_xc 0.13 12 0.01 0.02
Potential interpolate_vrs 0.00 12 0.00 0.00
Charge_Mixing init_mixing 0.00 1 0.00 0.00
ESolver_KS_PW hamilt2density 514.84 11 46.80 96.13
HSolverPW solve 514.84 11 46.80 96.13
Nonlocal getvnl 8.75 6512 0.00 1.63
pp_cell_vnl getvnl 10.18 7696 0.00 1.90
Structure_Factor get_sk 1.91 26640 0.00 0.36
DiagoDavid diag_mock 453.26 6512 0.07 84.63
DiagoDavid first 113.31 6512 0.02 21.16
DiagoDavid SchmitOrth 45.71 463494 0.00 8.53
Operator hPsi 279.52 41223 0.01 52.19
Operator EkineticPW 2.38 41223 0.00 0.44
Operator VeffPW 240.84 41223 0.01 44.97
PW_Basis_K recip2real 130.86 600246 0.00 24.43
PW_Basis_K gathers_scatterp 88.20 600246 0.00 16.47
PW_Basis_K real2recip 128.19 463494 0.00 23.93
PW_Basis_K gatherp_scatters 97.99 463494 0.00 18.30
Operator NonlocalPW 36.12 41223 0.00 6.74
Nonlocal add_nonlocal_pp 9.63 41223 0.00 1.80
DiagoDavid cal_elem 50.52 41223 0.00 9.43
DiagoDavid diag_zhegvx 42.21 41223 0.00 7.88
DiagoDavid cal_grad 247.13 34711 0.01 46.14
DiagoDavid check_update 0.03 34711 0.00 0.01
DiagoDavid last 5.98 8252 0.00 1.12
DiagoDavid refresh 2.63 1740 0.00 0.49
ElecStatePW psiToRho 30.96 11 2.81 5.78
Charge_Mixing get_drho 0.01 11 0.00 0.00
Charge_Mixing inner_product_recip_rho 0.00 11 0.00 0.00
Charge mix_rho 0.02 10 0.00 0.00
Charge Broyden_mixing 0.01 10 0.00 0.00
Charge_Mixing inner_product_recip_hartree 0.01 88 0.00 0.00
Forces cal_force_loc 0.00 1 0.00 0.00
Forces cal_force_ew 0.00 1 0.00 0.00
Forces cal_force_nl 3.02 1 3.02 0.56
Forces cal_force_cc 0.00 1 0.00 0.00
Forces cal_force_scc 0.17 1 0.17 0.03
Stress_PW cal_stress 16.53 1 16.53 3.09
Stress_Func stress_kin 0.26 1 0.26 0.05
Stress_Func stress_har 0.00 1 0.00 0.00
Stress_Func stress_ewa 0.00 1 0.00 0.00
Stress_Func stress_gga 0.01 1 0.01 0.00
Stress_Func stress_loc 0.45 1 0.45 0.08
Stress_Func stress_cc 0.00 1 0.00 0.00
Stress_Func stress_nl 15.81 1 15.81 2.95
ModuleIO write_istate_info 0.08 1 0.08 0.01
-------------------------------------------------------------------------------------
START Time : Tue Apr 16 18:56:05 2024
FINISH Time : Tue Apr 16 19:05:00 2024
TOTAL Time : 535
SEE INFORMATION IN : OUT.ABACUS/
We can see an efficiency improvement of approximately 20%
I try to set pw_diag_ndim =4 and do the test on some alloy systems. The energy/force/stress calculated by new dav have a large different with the results of old dav for some cases. For most cases, the energy difference on the last SCF step is larger thant that in the old dav method.
https://app.bohrium.dp.tech/abacustest/?request=GET%3A%2Fapplications%2Fabacustest%2Fjobs%2Fjob-abacustest-v0.3.112-65e576
@pxlxingliang Is abacus test result now accessible to public? I try to look into the examples offered in https://labs.dp.tech/projects/abacustest/?request=GET%3A%2Fapplications%2Fabacustest%2Fjobs%2Fjob-abacustest-v0.3.74-7cf445
but it seems I am refused to check the results.
Messages are shown as follows:
I am now working on dav(#4874 ) and want to test the performance, so I may need some reference. Thanks!
Hi @Cstandardlib，please access link：https://app.bohrium.dp.tech/abacustest/?request=GET%3A%2Fapplications%2Fabacustest%2Fjobs%2Fjob-abacustest-v0.3.74-7cf445 Domain labs.dp.tech is not used anymore.
@pxlxingliang Thank you! Now I have access to these examples via the new link.
After investigating math of Davidson, we find duplicate function of overlap matrix scc and orthogonalization.
#4874 removes useless scc of origin Davidson method.
#3903 adds a new Davidson iteration method called subspace davidson for pw basis, and #5199 updates a new version of dav_subspace with higher performance.
Davidson is the traditional method with orthogonalization, while Dav_Subspace is the one without orthogonalization and uses a generalized projected eigenproblem instead.