Hi, all,
I am runing the NEB calculation with 5 images which are interpolated between the stable initial and final states with the "nebmake.pl".
During the calculation, the total energy of one of the images is sharply decreased. The result for typing the "nebef.pl" is listed as follows:
0 0.00745200 -232.74888200 0.00000000
1 0.38143300 -231.69333000 1.05555200
2 0.17449300 -230.72317300 2.02570900
3 0.71103600 -241.32736800 -8.57848600
4 0.14234000 -232.08724700 0.66163500
5 0.15633100 -232.29338400 0.45549800
6 0.00572200 -232.58569400 0.16318800
The third one of the five images are decreased by more than 8 eV during the NEB run.
However,when I use the CONTCAR of the third image to run a geometry optimization, the total energy has no such apparent changes. The result for typing the "vef.pl" is listed as follows:
0 5.01542200 -229.990503 0
1 4.86494900 -230.022566 -0.032063
2 2.40174700 -230.557999 -0.567496
3 2.32226700 -230.573555 -0.583052
4 1.22182000 -230.764234 -0.773731
5 1.12853300 -230.772393 -0.78189
6 0.92472400 -230.818847 -0.828344
7 0.85344300 -230.824242 -0.833739
8 0.61843100 -230.855713 -0.86521
9 0.57509200 -230.859326 -0.868823
10 0.58101300 -230.885104 -0.894601
11 0.50909900 -230.887728 -0.897225
12 0.49776800 -230.896062 -0.905559
13 0.49186100 -230.898238 -0.907735
14 0.59830400 -230.918450 -0.927947
15 0.60066700 -230.920308 -0.929805
16 1.09548300 -230.986531 -0.996028
17 1.04800300 -230.994544 -1.00404
18 1.51027200 -231.046402 -1.0559
19 1.49587100 -231.052427 -1.06192
20 1.53153800 -231.223304 -1.2328
21 1.53516600 -231.231918 -1.24142
22 1.36029900 -231.521355 -1.53085
23 1.35429800 -231.531425 -1.54092
24 1.23371900 -231.858837 -1.86833
25 1.22632900 -231.868908 -1.87841
26 1.71389600 -232.098057 -2.10755
27 1.68669100 -232.106878 -2.11638
28 0.66578000 -232.350663 -2.36016
29 0.62894000 -232.357417 -2.36691
30 0.91421000 -232.442329 -2.45183
31 0.89359700 -232.446982 -2.45648
32 0.31639200 -232.543880 -2.55338
33 0.28296700 -232.546689 -2.55619
34 0.08335200 -232.559654 -2.56915
35 0.05339400 -232.560545 -2.57004
36 0.06604600 -232.561377 -2.57087
37 0.03029500 -232.561762 -2.57126
38 0.01867000 -232.561870 -2.57137
39 0.02034700 -232.561939 -2.57144
40 0.01262800 -232.561947 -2.57144
41 0.01801400 -232.561931 -2.57143
42 0.00438600 -232.561972 -2.57147
my INCAR for the NEB calculation is also attached.
SYSTEM = default
Start parameter for this Run:
ISTART = 0
ICHARG = 2
INIWAV = 1
Electronic Relaxation:
ENCUT = 400
ENAUG = 650
PREC = Accurate
IALGO = 38
NELM = 600
NELMIN = 4
EDIFF = 1E-07
GGA = PE
LREAL = Auto
WEIMIN = 0
Ionic Relaxation:
EDIFFG = -0.01
NSW = 1000
IBRION = 3
ISIF = 2
POTIM = 0.0
IOPT = 2
DOS related values:
SIGMA = 0.2
ISMEAR = 1
Spin polarized:
ISPIN = 2
MAGMOM = 16*2.7 16*1 1*1
File writing
LWAVE = .FALSE.
LCHARG = .FALSE.
NEB
ICHAIN=0
IMAGES = 5
SPRING = -5
LCLIMB=.TRUE.
Can anybody tell me what was wrong? Thanks.
What is wrong with my NEB calculation?
Moderator: moderators
Re: What is wrong with my NEB calculation?
This is very strange. I suggest that you look at the electronic convergence. Did the 3rd image converge within the 600 electronic iterations that you allowed? If not, the energy can be nonsense.
What about the spin state? Has the 3rd image in the NEB is converging to a different spin state than the separate minimization?
Another test you could run is to load the CHGCAR/WAVCAR files from the 3rd NEB image in your separate minimization. This would give you a better chance at getting the same electronic structure.
What about the spin state? Has the 3rd image in the NEB is converging to a different spin state than the separate minimization?
Another test you could run is to load the CHGCAR/WAVCAR files from the 3rd NEB image in your separate minimization. This would give you a better chance at getting the same electronic structure.
Re: What is wrong with my NEB calculation?
Hello,
I am also facing similar problems with my NEB calculation. I am calculating the migration barrier for a simple vacancy diffusion between nearest neighbour sites and am using 5 images interpolated with nebmake.pl. During the first run using IBRION = 3 and POTIM = 0.1, I get the following for nebef.dat.
0 0.20919300 -591.65842400 0.00000000
1 0.07313900 -591.20524600 0.45317800
2 0.04847200 -590.71855300 0.93987100
3 0.25234500 -603.48226200 -11.82383800
4 0.15143000 -591.16896100 0.48946300
5 0.10612800 -590.89956200 0.75886200
6 0.20957300 -591.65842200 0.00000200
The forces on the 1st and the 5th image were converged to around 0.05 eV/Ang whereas for the other images the max forces was around 0.1 eV/Ang. I checked the electronic convergence and did not find any errors. I restarted the calculation by setting IBRION = 3 POTIM=0.01 and the results was:
run2:
0 0.20919300 -591.65842400 0.00000000
1 0.00600200 -591.21747400 0.44095000
2 0.00731000 -590.72107400 0.93735000
3 0.03174400 -591.55570900 0.10271500
4 0.04987600 -591.08175400 0.57667000
5 0.12130600 -590.93059700 0.72782700
6 0.20957300 -591.65842200 0.00000200
The forces in the first and the 2nd and 3rd image is converged well to 0.003 eV/Ang while for the 4th and the 5th image the forces are only converged to 0.05 eV/Ang.
Since the forces were lower than 0.1 eV/Ang, I switched to a different optimizers to find the saddle points. I used IOPT = 7 and restarted the calculation. the third run gave the following output.
run3:
0 0.20919300 -591.65842400 0.00000000
1 0.00282300 -591.24454100 0.41388300
2 0.00640500 -590.72235700 0.93606700
3 0.02782000 -591.64736200 0.01106200
4 0.04576900 -591.16547100 0.49295300
5 0.17172900 -591.05444300 0.60398100
6 0.20957300 -591.65842200 0.00000200
The energy of the third image is lower than the energies of the end points. The forces of the first and the 2nd image is fine and converged upto 0.002 eV/Ang, whereas the 4th image the forces oscillate between 0.045 and 0.01 eV/Ang and similarly for the 5th image the maximum force is 0.09 eV/Ang.
Since this is a simple nearest neighbour jump, I expect only one saddle point and this is confirmed in literature as well. In all the calculations, I verified for electronic convergence. I am not sure what the error is? Do I need to keep running more iterations to get a smooth MEP or do I have to switch to a different optimizer?
Regards,
PG.
I am also facing similar problems with my NEB calculation. I am calculating the migration barrier for a simple vacancy diffusion between nearest neighbour sites and am using 5 images interpolated with nebmake.pl. During the first run using IBRION = 3 and POTIM = 0.1, I get the following for nebef.dat.
0 0.20919300 -591.65842400 0.00000000
1 0.07313900 -591.20524600 0.45317800
2 0.04847200 -590.71855300 0.93987100
3 0.25234500 -603.48226200 -11.82383800
4 0.15143000 -591.16896100 0.48946300
5 0.10612800 -590.89956200 0.75886200
6 0.20957300 -591.65842200 0.00000200
The forces on the 1st and the 5th image were converged to around 0.05 eV/Ang whereas for the other images the max forces was around 0.1 eV/Ang. I checked the electronic convergence and did not find any errors. I restarted the calculation by setting IBRION = 3 POTIM=0.01 and the results was:
run2:
0 0.20919300 -591.65842400 0.00000000
1 0.00600200 -591.21747400 0.44095000
2 0.00731000 -590.72107400 0.93735000
3 0.03174400 -591.55570900 0.10271500
4 0.04987600 -591.08175400 0.57667000
5 0.12130600 -590.93059700 0.72782700
6 0.20957300 -591.65842200 0.00000200
The forces in the first and the 2nd and 3rd image is converged well to 0.003 eV/Ang while for the 4th and the 5th image the forces are only converged to 0.05 eV/Ang.
Since the forces were lower than 0.1 eV/Ang, I switched to a different optimizers to find the saddle points. I used IOPT = 7 and restarted the calculation. the third run gave the following output.
run3:
0 0.20919300 -591.65842400 0.00000000
1 0.00282300 -591.24454100 0.41388300
2 0.00640500 -590.72235700 0.93606700
3 0.02782000 -591.64736200 0.01106200
4 0.04576900 -591.16547100 0.49295300
5 0.17172900 -591.05444300 0.60398100
6 0.20957300 -591.65842200 0.00000200
The energy of the third image is lower than the energies of the end points. The forces of the first and the 2nd image is fine and converged upto 0.002 eV/Ang, whereas the 4th image the forces oscillate between 0.045 and 0.01 eV/Ang and similarly for the 5th image the maximum force is 0.09 eV/Ang.
Since this is a simple nearest neighbour jump, I expect only one saddle point and this is confirmed in literature as well. In all the calculations, I verified for electronic convergence. I am not sure what the error is? Do I need to keep running more iterations to get a smooth MEP or do I have to switch to a different optimizer?
Regards,
PG.
Re: What is wrong with my NEB calculation?
graeme wrote :
> This is very strange. I suggest that you look at the electronic
> convergence. Did the 3rd image converge within the 600 electronic
> iterations that you allowed? If not, the energy can be nonsense.
>
> What about the spin state? Has the 3rd image in the NEB is converging to a
> different spin state than the separate minimization?
>
> Another test you could run is to load the CHGCAR/WAVCAR files from the 3rd
> NEB image in your separate minimization. This would give you a better
> chance at getting the same electronic structure.
Thanks very much for your suggestions, I will try.
> This is very strange. I suggest that you look at the electronic
> convergence. Did the 3rd image converge within the 600 electronic
> iterations that you allowed? If not, the energy can be nonsense.
>
> What about the spin state? Has the 3rd image in the NEB is converging to a
> different spin state than the separate minimization?
>
> Another test you could run is to load the CHGCAR/WAVCAR files from the 3rd
> NEB image in your separate minimization. This would give you a better
> chance at getting the same electronic structure.
Thanks very much for your suggestions, I will try.
Re: What is wrong with my NEB calculation?
Hello Andy,
Thanks for the suggestion. I did check the electronic convergence. There is no error there. The energy is converged upto 1e-5 eV in all the runs. Will changing the IOPT help?
PG
Thanks for the suggestion. I did check the electronic convergence. There is no error there. The energy is converged upto 1e-5 eV in all the runs. Will changing the IOPT help?
PG