QMCPACK
Inaccurate MPC energies
Describe the bug
Using the legacy code, physical=no MPC in the hamiltonian produces total energy values that are several Ha higher than Ewald and the gap between the Ewald and MPC gets larger with increasing supercell size. This is reported in bulk MnSe.
Analysis of the twists show that energies in each twist is consistent, indicating that the error could be related to density or some initialization step.
>qmca -e 20 -q ee *s002*scalar*
dmc.g000 series 2 ElecElec = 360.467753 +/- 0.207441
dmc.g001 series 2 ElecElec = 360.334781 +/- 0.157508
dmc.g002 series 2 ElecElec = 360.195652 +/- 0.147934
dmc.g003 series 2 ElecElec = 360.052337 +/- 0.133481
dmc.g004 series 2 ElecElec = 359.911319 +/- 0.211262
dmc.g005 series 2 ElecElec = 360.169259 +/- 0.160747
dmc.g006 series 2 ElecElec = 360.082586 +/- 0.118986
dmc.g007 series 2 ElecElec = 359.583074 +/- 0.246933
dmc.g008 series 2 ElecElec = 360.002621 +/- 0.141038
dmc.g009 series 2 ElecElec = 359.930153 +/- 0.188313
dmc.g010 series 2 ElecElec = 360.361782 +/- 0.273003
dmc.g011 series 2 ElecElec = 360.046368 +/- 0.092051
dmc.g012 series 2 ElecElec = 360.020397 +/- 0.182795
dmc.g013 series 2 ElecElec = 360.031820 +/- 0.184771
dmc.g014 series 2 ElecElec = 359.956056 +/- 0.184314
dmc.g015 series 2 ElecElec = 360.130444 +/- 0.087700
dmc.g016 series 2 ElecElec = 359.797406 +/- 0.122967
dmc.g017 series 2 ElecElec = 359.876725 +/- 0.264728
dmc.g018 series 2 ElecElec = 360.202026 +/- 0.181562
dmc.g019 series 2 ElecElec = 360.684932 +/- 0.110667
dmc.g020 series 2 ElecElec = 359.858839 +/- 0.169722
dmc.g021 series 2 ElecElec = 360.328722 +/- 0.095143
dmc.g022 series 2 ElecElec = 359.968411 +/- 0.260615
dmc.g023 series 2 ElecElec = 360.249228 +/- 0.206540
dmc.g024 series 2 ElecElec = 359.800285 +/- 0.238310
dmc.g025 series 2 ElecElec = 360.000984 +/- 0.221613
dmc.g026 series 2 ElecElec = 360.061534 +/- 0.130006
> qmca -e 20 -q MPC *s002*scalar*
dmc.g000 series 2 MPC = 625.248696 +/- 0.204685
dmc.g001 series 2 MPC = 625.104993 +/- 0.155363
dmc.g002 series 2 MPC = 624.972805 +/- 0.147327
dmc.g003 series 2 MPC = 624.841202 +/- 0.131456
dmc.g004 series 2 MPC = 624.690616 +/- 0.210045
dmc.g005 series 2 MPC = 624.961608 +/- 0.160649
dmc.g006 series 2 MPC = 624.866464 +/- 0.118222
dmc.g007 series 2 MPC = 624.367708 +/- 0.244609
dmc.g008 series 2 MPC = 624.782147 +/- 0.139274
dmc.g009 series 2 MPC = 624.703217 +/- 0.185981
dmc.g010 series 2 MPC = 625.129868 +/- 0.268086
dmc.g011 series 2 MPC = 624.820891 +/- 0.091476
dmc.g012 series 2 MPC = 624.795860 +/- 0.185064
dmc.g013 series 2 MPC = 624.814715 +/- 0.182802
dmc.g014 series 2 MPC = 624.736065 +/- 0.184180
dmc.g015 series 2 MPC = 624.910460 +/- 0.087743
dmc.g016 series 2 MPC = 624.582619 +/- 0.122511
dmc.g017 series 2 MPC = 624.662230 +/- 0.261837
dmc.g018 series 2 MPC = 624.978983 +/- 0.179683
dmc.g019 series 2 MPC = 625.460868 +/- 0.112656
dmc.g020 series 2 MPC = 624.642946 +/- 0.169742
dmc.g021 series 2 MPC = 625.105141 +/- 0.094158
dmc.g022 series 2 MPC = 624.759835 +/- 0.256939
dmc.g023 series 2 MPC = 625.034452 +/- 0.205712
dmc.g024 series 2 MPC = 624.579686 +/- 0.239591
dmc.g025 series 2 MPC = 624.788790 +/- 0.220319
dmc.g026 series 2 MPC = 624.842865 +/- 0.129755
To Reproduce Please see the input and some of the output files for the largest supercell in this calculation. Please let me know where you would like me to transfer these files, if you like to reproduce.
System:
-
system name: CADES
-
modules loaded:module purge; source $MODULESHOME/init/bash; module load PE-intel/3.0; module swap intel intel/2021.1; module load intel/2021.1; module swap openmpi openmpi/4.1.0; module load gcc/10.2.0; module load python ;module load fftw/3.3.5; module load boost/1.70.0; module load libxml2/2.9.9; module list; LD_LIBRARY_PATH=/software/tools/compilers/intel_2021/mkl/2021.1.1/lib/intel64:$LD_LIBRARY_PATH
-
other systems where this is reproducible: None
Additional context files.zip
Q1: Have you verified that all the tests pass with this compilation, particularly all the *mpc* ones?
Q2: Does MPC behave for you in other systems or is it broken in only this case?
Q3: Does VMC also misbehave?
Q4: What does the batched code give for the same systems?
No need to answer all of these at once. Any info will help give an idea of where things are wrong.
Thanks Paul,
Q2: I ran 4 different materials. For 1 material, MPC and Ewald agreed for the finite size extrapolation, but the issue existed in the other three. Those three have net non-zero magnetization. Here is the example where it worked as expected for Bi2Te2Se1:
Q3: Yes
I will get to Q1 and Q4 as soon as possible
Q1: The MPC tests below have passed with ctest -R short: Start 545: short-diamondC_1x1x1_pp-vmc_sdj-1-16-mpc 196/570 Test #545: short-diamondC_1x1x1_pp-vmc_sdj-1-16-mpc ...................................... Passed 0.07 sec Start 555: short-diamondC_1x1x1_hybridrep_pp-vmc_sdj-1-16-mpc 206/570 Test #555: short-diamondC_1x1x1_hybridrep_pp-vmc_sdj-1-16-mpc ............................ Passed 0.06 sec Start 565: short-diamondC_1x1x1_pp-vmc_sdj-meshf-1-16-mpc 216/570 Test #565: short-diamondC_1x1x1_pp-vmc_sdj-meshf-1-16-mpc ................................ Passed 0.06 sec Start 591: short-diamondC_1x1x1_pp-dmc_sdj-1-16-mpc 242/570 Test #591: short-diamondC_1x1x1_pp-dmc_sdj-1-16-mpc ...................................... Passed 0.05 sec Start 601: short-diamondC_1x1x1_pp-vmc_sdj_excited-1-16-mpc 252/570 Test #601: short-diamondC_1x1x1_pp-vmc_sdj_excited-1-16-mpc .............................. Passed 0.06 sec Start 609: short-diamondC_1x1x1_pp-dmc_sdj_excited-1-16-mpc 260/570 Test #609: short-diamondC_1x1x1_pp-dmc_sdj_excited-1-16-mpc .............................. Passed 0.05 sec Start 619: short-diamondC_1x1x1_pp-vmc-dmc-allp_sdj-1-16-mpc 270/570 Test #619: short-diamondC_1x1x1_pp-vmc-dmc-allp_sdj-1-16-mpc ............................. Passed 0.05 sec Start 628: short-diamondC_1x1x1_pp-vmc-dmc-allp_sdj-1-16-1-mpc 279/570 Test #628: short-diamondC_1x1x1_pp-vmc-dmc-allp_sdj-1-16-1-mpc ........................... Passed 0.05 sec Start 635: short-diamondC_1x1x1_pp-vmc-dmc-allp_sdj-1-16-2-mpc 286/570 Test #635: short-diamondC_1x1x1_pp-vmc-dmc-allp_sdj-1-16-2-mpc ........................... Passed 0.05 sec