UT theoretical chemistry code forum

Hi

I am trying to use CI-NEB method. After going through the forum messages and some literature, I have some questions. I will be thankful if anybody can share his experience!

I am trying to calculate reaction barrier for water molecules on a metal oxide surface. I already have some favorable configurations with full static relaxation by VASP code. In the next step I chose an intermediate geometry as the image (which I believe is a precursor state) . I am just using one image.

My questions:

1) Can I use an intermediate geometry as image or I have to use the script to generate the image files?

2) The vasp manual on NEB method (pg 81) states “ If all degrees of freedom are allowed to relax (isolated moleules, no surfaces, etc.), make sure that the sum of all positions is the same for each cell ” I could not properly get the statement, can anyone say a bit more ! In my case I do allow some surface atoms along with adsorbates to relax.

3) In my initial calculations the output is like this
NEB: the previous image is higher in energy: F
NEB: the next image is higher in energy : F
NEB: image is at an extrema
NEB: diff energy (min, max): 762.479528 762.479528
NEB: forces: par spring, perp REAL, dneb 304.963231 0.478778 0.000000
NEB: distance to prev, next image, angle between 69.443272 8.450626 91.535402
NEB: projections on to tangent (spring, REAL) -304.963231 -0.322032

FORCES: max atom, RMS 0.184192 0.052673
FORCE total and by dimension 0.577003 0.171505
Quasi-Newton relaxation of ions (Broydens 2nd method)
g(Force) = 0.140E-01 g(Stress)= 0.000E+00

In order to make myself sure; i just shared the above output. Is there some problem when it says image is at an extrema?

4) I am a bit worried about the total drift shown at the end of forces
0.018779 0.013595 -0.016822. As per vasp manual the drift should be low.
Ofcourse the calculation is not fully over but I will appreciate if some one suggests the intervals of this drift.

INCAR looks the following
LCLIMB=.TRUE.
ICHAIN=0
IMAGES=1
SPRING=-5
LSCALAPACK= .FALSE.
NSW = 130
IBRION = 1
ISIF = 0
ISYM = 0
POTIM = 0.5
ISMEAR = 0
SIGMA = 0.2

1) You can try a single intermediate image in a NEB calculation. However, if the path is curved, the vector between your initial and final states may not be a good approximation to the negative mode at the saddle. If the image does not converge to a saddle, try using 3 or 5 images. To the extent that the path is linear, a single image will be fine.

It is a little strange that you think the intermediate is a precursor. A minimization would tell you if this is a local minimum. The climbing image is used to find a saddle - it will not find an intermediate minimum.

2) If you have a surface, some of the lower atoms should probably be frozen. Then, don't worry about drift; just make sure that the frozen atom are in the same location in each image.

3) If you only have one image, it will be considered an extrema and used to climb to a saddle.

Dear Admin

Thanx for the reply. As you said if the path is linear , one image is acceptable. How can one know if the path is linear?

For simple elementary processes we generally use 3-5 images. While you might get away with a single image, we don't usually do this because it's too much trouble to redo bands using more images if a single image does not converge. Also, using more images gives information about intermediate minima along the path.

We don't have any way to estimate how linear a path is without calculating it.

Also, there are some elementary reactions for which more (about 8) images are required. The dissociative adsorption reaction of a molecule on a surface, for example, can be a long overall path with a short curved segment near the adsorbed state.

I recommend starting with about 4 images. If you find that your paths are linear, you could get away with fewer, but if convergence of the 4-image band is slow or the path is too long to adequately resolve it, go up to 6 or 8. If you find any intermediate minima along the path, it can be faster to break up your calculation into two NEB calculations instead of using a long single band.

Dear Admin and Users

Thank you for the previous reply. I indeed started using 4 images.

I have the following questions:

1) Regarding the linearity of the path and distance between images:
I just ran one scf calculation nsw=0 and want to make sure if I have correctly chosen the images.
trying to calculate 2 cases.

Case 1)
Image 01 : distance to prev, next image, angle between 1.141581 1.223549 170.528040
Image 02 : distance to prev, next image, angle between 1.223549 1.309415 163.147499
NEB: the previous image is higher in energy: F
NEB: the next image is higher in energy : F
NEB: image is at an extrema
Image 03) distance to prev, next image, angle between 1.309415 1.215930 165.459866
Image 04) NEB: distance to prev, next image, angle between 1.215930 1.141263 169.546181

Case 2)

Image 01 distance to prev, next image, angle between 1.626267 1.682289 172.292782
Image 02 distance to prev, next image, angle between 1.682289 1.695003 170.378778
NEB: the previous image is higher in energy: F
NEB: the next image is higher in energy : F
NEB: image is at an extrema
Image 03 distance to prev, next image, angle between 1.695003 1.668943 170.671637
Image 04 distance to prev, next image, angle between 1.668943 1.628874 172.995077

Are the distances between the images acceptable? I used nebmake.pl and nebavoid.pl. As some of the atoms were coming very near I had to use nebavoid.pl.

2) Regarding nebmake.pl : If I have an initial and final states where the water molecule on the surface is still in molecular form except that the molecules are oriented in a different way towards the final state. While using nebmake.pl I see that the images created are showing dissociated molecule. Whereas I expected molecular form in the path. Can you comment if I am making an error!

3) I am using QM optimiser
IOPT=3
TIMESTEP = 0.01
MAXMOVE = 0.2
My max forces are around 8 . I read some previous posts on convergence. where in QM was suggested in the first place. Do I need to use timestep of 0.001 for this force of 8 or 0.01 is enough? Is there any other way to increase the speed of the calculation.

4) Regarding ENCUT. Is it possible to use the default ENCUT as in the POTCAR file first and then later switching to higher ENCUT. As in the static relaxation one can use the default and then switch to a proper ENCUT. Is it possible to do the same with CI-NEB method as well?

I will be thankful for your comment and suggestion to my queries.

Thanks
Narasimham

(1) Distances between images approaching 2 Angstroms is quite large. It may converge, but for a long path like this, using 6 images would be safer.

(2) A linear initial path can be very poor if you have a rotating molecule. You would probably do better by constructing a path yourself. One way to do this is to make an intermediate structure with the molecule rotated half way. Then you can connect two linear paths (initial to intermediate) and (intermediate to final) as your initial path.

(3) A time step of 0.01 is pretty small. If you run for 10 or 20 iterations and the forces drop to below 1 eV/Ang and your path looks reasonable, then you can be more aggressive. If you find that the molecule does strange things along the path, then try again with a new initial path.

(4) Yes, it is generally safe to start with a low encut to optimize your path and then increase accuracy to get a precise saddle point energy.

Dear Admin

I used 6 images as suggested in the last reply. After doing 50 Iteration I find the following trend in my results

case1) showing one of the images after doing four runs

) NEB: forces: par spring, perp REAL, dneb 0.026925 23.740550 0.000000
NEB: distance to prev, next image, angle between 1.211576 1.206191 172.147523
NEB: projections on to tangent (spring, REAL) -0.026925 3.948477

FORCES: max atom, RMS 7.475838 2.167207
FORCE total and by dimension 23.740566 7.377401

2) NEB: forces: par spring, perp REAL, dneb 0.028677 14.196154 0.000000
NEB: distance to prev, next image, angle between 1.218880 1.213145 172.225753
NEB: projections on to tangent (spring, REAL) -0.028677 3.285779

FORCES: max atom, RMS 4.581112 1.295928
FORCE total and by dimension 14.196183 3.669166

3) NEB: forces: par spring, perp REAL, dneb 0.022870 8.971875 0.000000
NEB: distance to prev, next image, angle between 1.240267 1.235693 167.754067
NEB: projections on to tangent (spring, REAL) -0.022870 3.199844

FORCES: max atom, RMS 3.672535 0.819019
FORCE total and by dimension 8.971904 2.571256
4) NEB: forces: par spring, perp REAL, dneb 0.488858 2.522188 0.000000
NEB: distance to prev, next image, angle between 1.422571 1.324800 160.772192
NEB: projections on to tangent (spring, REAL) -0.488858 0.869763

FORCES: max atom, RMS 0.855265 0.234528
FORCE total and by dimension 2.569127 0.635048

The distance between the images when started was near to 1 Ang. but as the iterations proceeded the distance between images became larger. Even though the forces dropped. Now that the images are not equidistant will it be problem in the calculation? Can you comment on the path? I also checked with the xzy movie that the atoms are not behaving in any crazy way.

case2) one of the images after 4 runs

1) NEB: forces: par spring, perp REAL, dneb 0.206460 18.661628 0.000000
NEB: distance to prev, next image, angle between 0.901633 0.942925 168.140902
NEB: projections on to tangent (spring, REAL) 0.206460 -0.670620

FORCES: max atom, RMS 8.445910 1.796875
FORCE total and by dimension 18.673674 7.777686

2) NEB: forces: par spring, perp REAL, dneb 0.173290 11.131505 0.000000
NEB: distance to prev, next image, angle between 0.910607 0.945265 167.162140
NEB: projections on to tangent (spring, REAL) 0.173290 -0.333970

FORCES: max atom, RMS 5.029063 1.071612
FORCE total and by dimension 11.136514 4.650458
3) NEB: forces: par spring, perp REAL, dneb 0.272741 5.289487 0.000000
NEB: distance to prev, next image, angle between 0.908944 0.963492 165.600451
NEB: projections on to tangent (spring, REAL) 0.272741 0.443641

FORCES: max atom, RMS 1.789087 0.510768
FORCE total and by dimension 5.308059 1.713840

In this case, the distance between the images are less than 1 Ang. and forces have come down. Here as well I checked the movie which seems ok. I am using the QM optimiser with time step 0.01. As in last reply you suggested one can be more aggressive in relaxation can I use a larger time step in this case to make the calculation faster?

Regarding the linearity of the path: for the case1) I find the angle between the paths decreases from 170 - 140 degrees whereas in the case 2, the angle between the images decreases from 170-156 degrees after roughly 50 iterations. The path is not strictly linear as I think an angle of near to 180 between the images is required. Can you comment on linearity of a path? Also I will thankful if you can suggest whether the path seems ok or should I change?

Thanking you
Narasimham

Hi, Admin,

In the same spirit of climbing image, why not add some codes to determine the intermediate minimum and let it descend to
the precise minimal location? I think it will help in some situations that researchers care about the configuration of the meta-stable state. A separate relaxation can also do this but it needs more time.

This is a good suggestion and I see the potential value in it. It would be a good option to include as a flag and see how it works. The only potential drawback would be if it slows down convergence of the climbing image at all. Finding saddle points is a lot harder than finding minima. Once you have the saddle, it is relatively easy to find intermediate minima along the path. Sure, it is a separate calculation but it involves only 1 image. But anyway, this is no reason not to add it as an option -- we'll add it to the list.

I am going to have to disagree. To add a falling image would hurt convergence. It would pull images away from the saddle possibly compromising the tangent of the saddle. Not to mention a system that has in intermediate minima has multiple saddle points so it would require multiple climbing images. Also adding this method would encourage longer pathways and use of more images. The NEB works well for single transitions and we have tried to encourage use of only several images in an NEB run. If high resolution of the minimum energy path is required it is more efficient to find the saddle using as few images as possible and then use a steepest decent method to descend and map out the MEP. I believe adding this functionality will do more harm then good. It would be better to stop any run that shows signs of intermediate minima, minimize the intermediate, and then restart two NEB runs to map the entire path. Don't hold your breath waiting for this feature.