Fermi energy shift upon change of k-mesh

[physicalattraction]

I changed EDIFF to 1E-6 instead of 1E-4 and recalculated the k=5x5x1 and k=9x9x1 calculations, but the results are the same.

POTCAR order: Ag, C, H, S
I use PAW_GGA POTCAR files.
VRHFIN =Ag : s1 d10
VRHFIN =C: s2p2
VRHFIN =H: ultrasoft test
VRHFIN =S : s2p4

I have the surface normal in the z-direction.

POSCAR: (it ends with all the zeros, because it is a copy of a previous CONTCAR file)
Thiophenol fcc hollow - polar angle 70 d
1.00000000000000000
4.4001000000000001 -2.5404000000000000 0.0000000000000000
4.4001000000000001 2.5404000000000000 0.0000000000000000
0.0000000000000000 0.0000000000000000 52.6086000000000027
21 12 10 2
Direct
-0.0095857250752562 0.0095879626736480 0.3636931194127489
0.3369278092716918 0.3189018624169284 0.3628597835994536
0.6810994399487706 0.6630754118739857 0.3628596041038925
0.0020054177132850 0.3337711578374746 0.4082048604646082
0.6662310971170391 -0.0020031574191989 0.4082048325005797
0.3341125468284605 0.6658897431522336 0.4084552015445388
0.3329852186034474 0.0009125650244796 0.4542698932996296
-0.0009090266664059 0.6670179844384491 0.4542698285511225
0.6682975278295269 0.3317060872545711 0.4540068471980333
-0.0000036921957533 0.0000032537733881 0.5000058964332599
0.3339121418405372 0.3339400838654661 0.5000083557766075
0.6660639752002545 0.6660920409902774 0.5000083633925277
0.0008505416736009 0.3330057983255288 0.5457465636506377
0.6669969464900575 -0.0008473539379818 0.5457466586524736
0.3317003500822112 0.6683026885268412 0.5460073756034589
0.3337197961069501 0.0020464901326024 0.5918073935150181
-0.0020458408770580 0.6662814576478319 0.5918074343168879
0.6658744263087496 0.3341268715680350 0.5915588495367213
0.0095636300946455 -0.0095636078532585 0.6363232600432384
0.3188685610341569 0.3369654381134383 0.6371509146500192
0.6630355206341340 0.6811301891723189 0.6371512197890561
0.4758925711152586 0.5241062533050319 0.2945471283686625
0.3750198485391763 0.3502417196415261 0.2816097334788393
0.4385869491189057 0.2882105477145116 0.2558262930690601
0.6054344726523615 0.3945544453831792 0.2427430480262714
0.7117891398512575 0.5614102437310344 0.2558256908107858
0.6497553785116039 0.6249787009584395 0.2816099586109923
0.5240912861590162 0.4759042518930113 0.7054543854378976
0.6250344371674010 0.6497774598587090 0.7183862355261093
0.5614581116545136 0.7118158814538522 0.7441676131426483
0.3945858520999809 0.6054028117000950 0.7572368709358206
0.2881866088156482 0.4385372034608510 0.7441663281271388
0.3502156764544531 0.3749553223198469 0.7183857260051513
0.2434101706667859 0.2676604643367048 0.2916584367739812
0.3558696540383371 0.1552125293258743 0.2460195495091990
0.6541619264221725 0.3458369111440738 0.2225400345626481
0.8447932581470643 0.6441311247744667 0.2460184386774967
0.7323511940945523 0.7566043093033895 0.2916590609881372
0.7567141729933166 0.7323269152457149 0.7083271918023034
0.6441454018415844 0.8447652114270864 0.7539683603339455
0.3458545205344160 0.6541500036766812 0.7774347059033891
0.1552440663378221 0.3558526564801618 0.7539662919239895
0.2676911219600935 0.2433070234147084 0.7083281599011556
0.3602001731039098 0.6398004113306462 0.3264673731826105
0.6398474358916418 0.3601487206776662 0.6735071288672517

0.00000000E+00 0.00000000E+00 0.00000000E+00
[...]
0.00000000E+00 0.00000000E+00 0.00000000E+00
<span class='smallblacktext'>[ Edited Fri Oct 23 2009, 03:01PM ]</span>

[forsdan]

Using VASP version 5.2.2 together with the input files you provided I obtained the following:

5x5x1: energy(sigma->0) = -208.619016
9x9x1: energy(sigma->0) = -208.528016
13x13x1: energy(sigma->0) = -208.559939

So it appears that I do not get the large variations as you do. I was a bit concerned at first when I noticed the high Ewald and Hartree energies (~ 0.45 MeV) when doing the 5x5x1 run, which might suggest a compressed system. I did however compare to systems where the size was the same but the elements were different. So now I'm more or less convinced that higher energies must be due to the heavier Ag ions (and different number of valence electrons), since I compared to a system with mostly Fe. But please double-check that all distances are reasonable everywhere just to be sure.

This last point does not however explain why you get the results that you do. Which VASP version are you using? Have you tried the latest version? However, I seriously doubt that this is something that can occur in previous versions. So it's more likely that it's a compiler issue. Maybe it can be some over-optimization of a sub-routine or similar. This my only idea since I can't reproduce your results. Maybe someone else has additional suggestions.


/Dan

EDIT (24/10 2009): Added the 13x13x1 result






<span class='smallblacktext'>[ Edited Sat Oct 24 2009, 08:48PM ]</span>

[physicalattraction]

Thanks a lot Dan!

I recalculated the ionic convergence with a 3x3x1 and a 7x7x1 grid, and I found E = -208.6 eV for both of them. Did you also do another ionic convergence, or just an electronic relaxation? I still don't understand why I get these wrong results if I use the CONTCAR from the 5x5x1 calculation, increase the number of KPOINTS and calculate the electronic calculation.

I did use the old WAVECAR file though, which is set for a different number of k-points. I thought that this would still be a better guess than the atomic densities as a starting point and I thought that the end results would not rely on the input wave functions, but could that be wrong? If you increase the number of k-points and read WAVECAR, it gives a warning in the output (might fail), but also after that that it has succeeded (WAVECAR file read successfully).

[alex]

Have you ever cross-checked that? You might very well be caught by your own trap ...

alex

[physicalattraction]

I tried a 9x9x1 electronic relaxation, with the CONTCAR of a 5x5x1 ionic optimization, but without reading the WAVECAR file, and now the total energy is the same as the original 5x5x1 calculation. So indeed, this turned out to be the problem. :-s