My Community

Dear Vasp users and experts,

I have a problem I like to share in order to find a solution
I am optimizing one strucutre which consists in one molecule on top
of a metal surface(Au(111)). I am using PBE-PAW and 1-kpoint mesh
After the first iteration the systems explotes and the final geometry is a chaos
and of course is not converging.
I kept the last layers frozen and the rest is without constraints.
I wonder what I am doing wrong. Many thanks.

Here is my INCAR file

Electronic Relaxation 1
PREC = low
ENMAX = 450.00 eV
NELM = 99
EDIFF = 1E-04
INIWAV = 0 random initial wavefunctions
MAXMIX = 30

Ionic Relaxation
EDIFFG = -0.03 stopping-criterion for IOM
NSW = 20 number of steps for IOM
NBLOCK = 1; KBLOCK = 40 inner block; outer block
IBRION = 2 ionic relax: 0-MD 1-quasi-New 2-CG
ISIF = 2 stress and relaxation
IWAVPR = 1 prediction: 0-non 1-charg 2-wave 3-comb
ISYM = 1 0-nonsym 1-usesym
SYMPREC = 0.001
LCORR = F Harris-correction to forces
ISMEAR = 0
LREAL = F
ISPIN = 1
ALGO = Fast
GGA = PR
SIGMA =0.05
ISMEAR = 2
and my POSCAR file looks like as follow

C S Au N H
1.00000000000000
8.6999999999999993 0.0000000000000000 0.0000000000000000
-4.3499999999999996 7.5344210000000000 0.0000000000000000
0.0000000000000000 0.0000000000000000 90.0000000000000000
3 1 45 1 4
Selective dynamics
Direct
0.4290965381394014 0.3560043659719025 0.5127194069577761 T T T
-0.7248216271167025 -0.1976256221208859 0.5095208925330077 T T T
-0.4075294615998533 0.2119056486200086 0.5567394088998638 T T T
0.1850670150260940 -0.6914127047793990 0.5444674984041180 T T T
0.5316646527496547 0.5674883483619202 0.1153543570591613 T T T
0.5053455568620613 0.8240784702059892 0.1744576897074481 T T T
0.1165755554386121 0.8838682103938699 0.6583812256728619 T T T
0.0664920180219664 0.8335795274058958 0.7840081489501964 T T T
0.6516042047009085 0.0446677786383010 0.8121904985143596 T T T
0.4103964064136809 0.7851677329001953 0.7072580589708616 T T T
0.1029570111070183 0.6965212345087467 0.6728996587192420 T T T
0.2537572672220864 0.5669481009538331 0.1974142902239842 T T T
0.2425059470947027 0.2788511053243620 0.1608120029499518 T T T
-0.1518764386369267 0.2815501148943392 0.1977450719967915 T T T
0.3662838513397113 0.1781256321644807 0.6629966274021958 T T T
-0.0022998104193551 0.0823901326080133 0.7648960501062855 T T T
0.2269712036583987 0.0952099441523911 0.8244384330941473 T T T
0.1879770588661437 0.4619843733097340 0.7795876924992933 T T T
0.3165758625483630 0.8911283730404812 0.7842166187474486 T T T
-0.7644938111358088 -0.1696120806840566 0.1418928162927671 T T T
-0.1797514970765919 -0.0465455642104604 0.2003718171761786 T T T
0.6453278603167304 0.5258587251557625 0.6261918080263706 T T T
0.8281903625392468 0.6208258597968508 0.7518996033590950 T T T
-0.3388837620151846 -0.4887349485287514 0.2391866681864716 T T T
0.2030898957090683 0.3845041153548371 0.7055309764030510 T T T
0.7063626042111096 0.3219842721404028 0.8167975836229364 T T T
0.8348029461589057 0.1529173542455831 0.8362527935886300 T T T
0.3984081145627085 -0.3575838880392048 0.6843166757490726 T T T
0.4381122672221545 0.6659493922435780 0.2400992037126497 T T T
0.3678130395675220 -0.0742198251816783 0.7549606906820394 T T T
0.8673617573067337 -0.1314456145847929 0.8124448701998523 T T T
0.5221000000000018 0.9419469999999990 0.4711580000000026 F F F
0.8535439999999994 0.9419430000000020 0.4711580000000026 F F F
0.7430599999999998 0.7209780000000023 0.4449979999999982 F F F
......etc.....

one of the reasons may be that you choose the parameters too inaccurate to obtain reliable forces:
1) set EDIFF to a higher precision (at least 1e-5). Please also check if your convergence criterium was reached for each of the ionic steps in your run (or if the electronic scf-procedure was stopped after 99 steps). If not, the calculated forces are not reliable.
2) set PREC to accurate
3) you can safely set LREAL to Auto, this gives very good results (the relaxed structures are the same as if you set them to false) at a significantly decreased computing time!
4) if the first relaxation step still gives unreasonable results, do a few steps with decreased POTIM or switch to a different ionic optimization algorithm (like damped-MD, IBRION=3)
5) for extremely 'long' supercells, it helps to decrease AMIN from the default value (=0.1) to achieve better electronic convergence for the ionic steps.
6) if possible, use an nxnx1 k-mesh