tangential force in TST calculation

[DIDO]

I'm trying 4-images TST search using NEB, but when i check OUTCAR of each individual images, the tangential force increased as calculation goes--which i think is abnormal. (one increased from -1. to + 27. )

Anybody have ideal what's been going on?

[graeme]

First, please post here for TST related problems, or on the vasp forum, but not both.

The force along the tangent can increase during a run. This is especially true for stiff bonds which are broken in a short distance along a minimum energy path. The tangential force is ignored during the NEB optimization, and you should not expect it to systematically drop. This is true for all images except for the highest energy image if you are using the climbing image. For the climbing images, the total force (tangential and perpendicular) should drop as the band converges.

Another simple thing to check is that the band is converging smoothly. If you are starting from a linearly interpolated path, do a few very conservative steps (i.e. ibrion=3, potim=0.01) until the forces drop below 1 eV/Ang. Then continue with potim=0.1 and/or ibrion=1. If the initial forces are too large for the optimizer, the initial band can go crazy.

To really figure out what's going on, we'll need to see the files.

[DIDO]

First, sorry about posting similar threads on both forum--i didn't know those two forums are related--i've been working on this TST thing for a while but couldn't get it work, so i was really in a rush to figure out the answer~~~

Thanks for answering--here are the INCAR setups--i'm working on small molecule dissociate over metal clusters.
IMAGES = 4;
SPRING = -5;
EDIFF = 0.1E-04 stopping-criterion for ELM
EDIFFG = 0.1E-03 stopping-criterion for IOM
IBRION = 3 ionic relax: 0-MD 1-quasi-New 2-CG
POTIM = 0.05

So basically what i'm gonna do is using those setups to calculate, until i got low force for each image, and then I'm gonna switch IBRION =2 and switch on LCLIMB-tag (maybe add more images in between) until i find an image with tangential force 0--is this right?


Thanks a lot!

[graeme]

I recommend a few changes. Use a force-based convergence criteria, such as ediffg=-0.01. Also, you can use LCLIMB=.TRUE. for the entire run. Finally, either switch to IBRION=1, or just increase the POTIM (to 0.1 or 0.2) and stick with IBRION=3. Don't use IBRION=2, because it has an energy based criteria, which does not work with the NEB.

I think there is some confusion about the tangential force. We usually talk about the tangential force as the real force projected along the band. This will not go to zero at convergence. The force which is brought to zero during optimization is the perpendicular real force plus the tangential spring force. It is this force which is reported in the OUTCAR file.

[szygmunt]

I have a question about the meaning of the forces reported in the OUTCAR file during a NEB optimization.
They are: TOTAL FORCE, CHAIN FORCE, CHAIN + TOTAL, and tangential force.

I have verified that indeed the values for CHAIN + TOTAL are just the sum of CHAIN and TOTAL forces.
My main question deals with the nature of CHAIN and TOTAL forces. Originally I thought that CHAIN
FORCE = the projection of the spring force parallel to the TANGENT direction, and TOTAL FORCE = the
projection of the true (potential) force perpendicular to the TANGENT direction. However, when I take
the dot product of these two forces for any atom in my system, I find that the result is NOT equal to zero.
So I am confused. Is CHAIN FORCE = spring force (before taking any projection) and TOTAL FORCE =
true force (before taking any projection)? If so it would seem odd to add these two forces, since that is
not the quantity being minimized in the NEB algorithm.

Can you help me understand these quantities better? Thanks!

[szygmunt]

[quote="szygmunt"]I have a question about the meaning of the forces reported in the OUTCAR file during a NEB optimization.
They are: TOTAL FORCE, CHAIN FORCE, CHAIN + TOTAL, and tangential force.

I have verified that indeed the values for CHAIN + TOTAL are just the sum of CHAIN and TOTAL forces.
My main question deals with the nature of CHAIN and TOTAL forces. Originally I thought that CHAIN
FORCE = the projection of the spring force parallel to the TANGENT direction, and TOTAL FORCE = the
projection of the true (potential) force perpendicular to the TANGENT direction. However, when I take
the dot product of these two forces for any atom in my system, I find that the result is NOT equal to zero.
So I am confused. Is CHAIN FORCE = spring force (before taking any projection) and TOTAL FORCE =
true force (before taking any projection)? If so it would seem odd to add these two forces, since that is
not the quantity being minimized in the NEB algorithm.

Can you help me understand these quantities better? Thanks![/quote]

===========================================================================================================
I don't know if it is kosher to reply to my own post, but after posting I learned something new. Apparently the CHAIN FORCE
written to OUTCAR is defined so that it is ALREADY parallel to the TANGENT direction. Assuming that the TOTAL
FORCE is the true force, when I calculate its projection perpendicular to the TANGENT direction and take the dot product
with the CHAIN FORCE, I get zero. This leads me to the following (tentative) conclusions, and two final questions:

Conclusions:
1. CHAIN FORCE is (already) parallel to the TANGENT direction.
2. TOTAL FORCE is the true force (before taking any projection).

Questions:
1. Am I correct above?
2. Why does VASP add the CHAIN FORCE to the TOTAL FORCE and write this to OUTCAR for each atom? It is the net force on
each atom in the NEB algorithm, but what is being minimized in NEB is NOT the net force but the sum of the parallel component of the
spring force and the perpendicular component of the true force.

Thanks for any help you can give me.

[graeme]

The code that you refer to is the NEB implemented by the vasp folks rather than in our vtstcode, which we use. Thus, I can only really give you my view looking at their code.

The built-in vasp NEB only works for a spring constant that is negative, otherwise it does an elastic band calculation without the force projections.

Assuming that we are talking about an NEB calculation, the CHAIN-FORCE appears to be the spring forces along the tangent plus the negative of the true force along the tangent, CHAIN-FORCE = (Fs * T - F * T) T where T is the tangent, Fs is the spring force and F is the true force.

Then, the CHAIN + TOTAL force is the sum of the NEB force and the true force, CHAIN + TOTAL = F + (Fs * T - F * T) T = (F - F*T T) + Fs *T

which is the proper NEB force with the true force perpendicular to the tangent and the spring force parallel to it.

I just don't understand why you would want the CHAIN-FORCE or to decompose the force in this way. Anyway, the final result (CHAIN + TOTAL) looks fine to me. Which is good; I was worried for a moment as I was trying to understand their code.

[szygmunt]

Graeme,

Thanks for the detailed explanation of the definition of these forces in the original VASP code, which
is the one I am currently using to run NEB calculations. I too find it puzzling that this particular force
decomposition is written to the OUTCAR file. I don't understand why the name "CHAIN FORCE" is given to
Fs - F along the direction of the "chain" or the band. But clearly this quantity has no intrinsic meaning,
and I assume it is only used to calculate the total NEB force, which as you mentioned is Fs along the tangent
+ F perpendicular to the tangent.

This is good to know for sure.

On a related note, I thought it was odd that the NEB force magnitudes are not written to the OUTCAR file, so I wrote
a simple script to calculate them for each atom and to write the maximum NEB force on any atom in the system. I
wanted to convince myself that the parameter EDIFFG was being applied correctly in the NEB optimization, which
I am doing with

SPRING = -5
IBRION = 3
POTIM = 0.005
SMASS = 0.4
EDIFFG = -5E-2

Anyway, I have noticed that sometimes the maximum force on an atom will increase for quite a few steps before decreasing
again. Is there any easy way to understand this behavior?

Thanks again for your help.