My Community

I have tried to solve the strange problem in VASP5.2.12 for a long time and now I found what the problem is in VASP5.2.12.
First, I start the calculations using VASP5.2.2 and VASP5.2.12 with the same input to check the problem. Using VASP5.2.2, the structure can easily converge to reach our criteria (EDIFFG = -0.01) but using VASP5.2.12 the structure cannot relax to reach the criteria and the calculation always stop with the following massage;

ZBRENT: fatal error in bracketing
please rerun with smaller EDIFF, or copy CONTCAR
to POSCAR and continue

After comparing the results obtained from VASP5.2.2 and VASP5.2.12, I found that the results of n-local (FORCE) are too different. Therefore, I believed that this must be the cause of problem. After checking codes, I found that n-local was calculated by calling subroutine in nonlr.F and nonl.F. Therefore, I try to replace nonlr.F and nonl.F in VASP5.2.12 with those in VASP5.2.2.
After modifying the codes, the structure can easily converge and reach the criteria as VASP5.2.2 did.
I thought that this may be a problem in VASP5.2.12 and hope you can fix this problem.
Now, I'm not sure that our calculation result is still valid.

I have to note that you will found this problem when you perform the calculation in defect cell. For bulk or perfect cell, there is no problem because of the symmetry.
<span class='smallblacktext'>[ Edited ]</span>

For more information, I showed you the results
VASP5.2.2

FORCE on cell =-STRESS in cart. coord. units (eV):
Direction XX YY ZZ XY YZ ZX
--------------------------------------------------------------------------------------
Alpha Z 1664.76740 1664.76740 1664.76740
Ewald -6756.37105 -6738.85708 -5452.80711 -7.43256 -0.03339 -0.00486
Hartree 1974.50912 1937.38658 2170.44220 1.20172 -0.06349 -0.03001
E(xc) -1830.94018 -1832.79279 -1843.27934 0.03943 -0.00020 -0.00076
Local -1360.34398 -1331.85698 -2729.98437 8.87463 0.10618 0.03648
n-local -1131.44601 -1128.67533 -1097.89377 0.12044 0.00847 0.00452
augment 448.30590 447.90643 440.27205 -0.27147 -0.00119 -0.00012
Kinetic 6985.69835 6965.42520 6836.31201 -0.05531 -0.01322 0.00797
Fock 0.00000 0.00000 0.00000 0.00000 0.00000 0.00000

vasp5.2.12

FORCE on cell =-STRESS in cart. coord. units (eV):
Direction XX YY ZZ XY YZ ZX
--------------------------------------------------------------------------------------
Alpha Z 1664.76740 1664.76740 1664.76740
Ewald -6756.37105 -6738.85708 -5452.80711 -7.43256 -0.03339 -0.00486
Hartree 1974.50897 1937.38644 2170.44212 1.20172 -0.06349 -0.03001
E(xc) -1830.94020 -1832.79281 -1843.27937 0.03943 -0.00020 -0.00076
Local -1360.34389 -1331.85689 -2729.98443 8.87463 0.10618 0.03648
n-local -1132.79022 -1128.67527 -1097.89372 -1.22400 0.00847 1.34893
augment 448.30584 447.90639 440.27200 -0.27153 -0.00121 -0.00019
Kinetic 6985.69818 6965.42504 6836.31181 -0.05531 -0.01322 0.00797
Fock 0.00000 0.00000 0.00000 0.00000 0.00000 0.00000


You can see a large difference in n-local part indicating that there is something wrong.

After modifying VASP5.2.12

FORCE on cell =-STRESS in cart. coord. units (eV):
Direction XX YY ZZ XY YZ ZX
--------------------------------------------------------------------------------------
Alpha Z 1664.76740 1664.76740 1664.76740
Ewald -6756.37105 -6738.85708 -5452.80711 -7.43256 -0.03339 -0.00486
Hartree 1974.50897 1937.38644 2170.44212 1.20172 -0.06349 -0.03001
E(xc) -1830.94020 -1832.79281 -1843.27937 0.03943 -0.00020 -0.00076
Local -1360.34389 -1331.85689 -2729.98443 8.87463 0.10618 0.03648
n-local -1131.44595 -1128.67527 -1097.89372 0.12044 0.00847 0.00452
augment 448.30584 447.90637 440.27203 -0.27152 -0.00121 -0.00017
Kinetic 6985.69818 6965.42504 6836.31181 -0.05531 -0.01322 0.00797
Fock 0.00000 0.00000 0.00000 0.00000 0.00000 0.00000


Now, the results are similar to VASP5.2.2 and the structure can relax without any problem (EDIFFG = -0.01).
<span class='smallblacktext'>[ Edited Fri Aug 26 2011, 03:30PM ]</span>

this behaviour cannot be be reproduced by us, the non-local forces are calculated
consistently with the nonl.F and nonr.F routines of vasp.4.6, vasp.5.2.2 and vasp.5.2.12, as delivered with the respective code versions

After sending email to the Admin, admin also found the same problem on the relaxation of defect cell. The problem arise clearly in the calculation of defect cell, not bulk or perfect cell. Hope to hear from admin how to solve this problem.