Why higher cutoff energy gives higher energy in transition metal oxide system

[zongtan_fang1]

Hi,

I am trying to do cutoff test and find out an trend with higher cutoff energy giving slightly higher energy in transition metal oxide system.
The system includes 240 atoms with different transition metal such as Ni, Mn, and Co. The calculations with higher cutoff energy were carried out by reading wavefunctions and they should converge to the same state. the K points are decent using 2*2*2.

From the vasp manual, the kinetic errors contribute a little, while that should not be significant.

What is the physics reason behind that ?

Thanks a lot.


PBE+U

Ecut Energy
400 -1247.0595250
420 -1246.4514930
450 -1245.7843340
480 -1245.4574000
500 -1245.3868620
520 -1245.3921760
550 -1245.4529350


Zongtang

[alex]

Hi Zongtang,

VASP is adding an extra-bit of cutoff dependend energy per atom, e.g. for H2O (in OUTCAR), to compensate for the incomplete basis set:

kinetic energy error for atom= 0.0091 (will be added to EATOM!!)
kinetic energy error for atom= 0.0002 (will be added to EATOM!!)

If you remove that, you'll find the proper decrease in energy with increasing cutoff.
These corrections become smaller with better basis, too.

Cheers,

alex

[zongtan_fang1]

Hi Alex,

Thanks a lot for your response. Actually I noticed that from the manual.

Is "kinetic energy error for atom" for each atom of this type or all atoms for this type. ?

In my case, the energy difference is larger than 1.5 eV, seems much larger than the kinetic energy error.

Thanks.

Zongtang

[alex]

Hello Zongtang,

in your case you have to collect all contributions of your 240 atoms.

Hth,

alex

[zongtan_fang1]

Hi Alex,

Thanks a lot for your response.
While it looks a little bit weird if I sum the errors .
Not converged yet at 800 eV. Is there something wrong ?

Error_t = 120*O + 48*x + 6*y +6*z

Kinetic Error
Ee O x y z Error_t Ee+ Error_t
400 -1247.05953 0.11560 0.01010 0.00880 0.00930 14.46540 -1232.59413
420 -1246.45149 0.07270 0.00660 0.00580 0.00600 9.11160 -1237.33989
450 -1245.78433 0.03890 0.00380 0.00330 0.00360 4.89180 -1240.89253
480 -1245.45740 0.02580 0.00310 0.00250 0.00290 3.27720 -1242.18020
500 -1245.38686 0.02240 0.00300 0.00240 0.00280 2.86320 -1242.52366
520 -1245.39218 0.02080 0.00300 0.00240 0.00280 2.67120 -1242.72098
550 -1245.45294 0.01940 0.00270 0.00220 0.00260 2.48640 -1242.96654
600 -1245.67490 0.01680 0.00190 0.00160 0.00180 2.12760 -1243.54730
650 -1245.87048 0.01290 0.00100 0.00090 0.00100 1.60740 -1244.26308
680 -1245.97592 0.01050 0.00080 0.00070 0.00070 1.30680 -1244.66912
700 -1246.03819 0.00910 0.00070 0.00060 0.00060 1.13280 -1244.90539
750 -1246.11327 0.00590 0.00060 0.00050 0.00050 0.74280 -1245.37047
800 -1246.16477 0.00410 0.00060 0.00050 0.00050 0.52680 -1245.63797

[alex]

Hello Zongtang,

imho it looks very good, considering you have 240 atoms per cell, ending up with an average error of about 1meV/atom.
Don't forget that we normally take differencies of total energies to do some fancy chemistry. So most of the error is (hopefully) cancelled out.

Of course it depends on the quantity you are interested in. And this is up to you.

Cheers,

alex

[zongtan_fang1]

Hi Alex,

Thanks for your quick reply.

Actually the property I am interested in actually converged well (< 0.01 eV) from the test results using 480 and higher cutoff energies.

Just didn't expect somewhat significant difference of Ee when a larger cutoff energy is used.

Thanks again.

Zongtang