NEB MEP convergence check

Hi,

I need some suggestions regarding Minimum energy path calculations using NEB implemented in LAMMPS. We will get some final conformation of each NEB replica after NEB calculations are over. Each replica must minimize its energy at each NEB step (except reparameterization). But how can we make sure the final set of conformations we obtain for each replica indeed represents conformations along the Minimum Energy Path?
I think one way could be to run the same simulation for longer minimization steps and see that the RMSD of each image is not changing much in the final steps. One other way is to make sure after some number of NEB steps, the energy of each replica should not decrease further.

Is there any other interesting way anyone can think of, to make sure we have converged to the correct MEP?

Thanks,
Srijan Singh

If by “longer min steps” you mean more iterations, then for the damped
dynamics minimizers (e.g. FIRE) which are used with NEB in LAMMPS,
that simply means set a lower tolerance. When a tolerance is reached
that also means your 2nd criterion is met, i.e. the energy (or forces
depending on how you set the tolerances) are decreasing by less than
that tolerance.

Steve

Hi Steve,

Thanks for your inputs. I agree with your suggestions completely.
Also, will you be able to help me out a bit with this? : I am using verbose keyword to get supplementary output during my NEB simulation to better understand what is going on. I had a small confusion regarding how “angletangrad” is calculated. The documentation says:

“angletangrad” is the angle between the 3N-length tangent vector and the 3N-length force vector at image i.

I just wanted to make sure whether this 3N-length force vector also contains the contribution of nudging forces?

If that is the case, I was wondering if it would be possible to get the angle between the tangent vector and the “natural” force vector (ie. not including spring forces) for each atom in each replica? As far as my understanding goes, for each atom in each replica, if the natural force has zero perpendicular component to the tangent, then it satisfies the condition for the Minimum Energy Path. So, I could use this data to validate that I have indeed converged to the MEP.

Thanks for your time,
Srijan Singh

Hi Steve,

Thanks for your inputs. I agree with your suggestions completely.
Also, will you be able to help me out a bit with this? : I am using verbose keyword to get supplementary output during my NEB simulation to better understand what is going on. I had a small confusion regarding how “angletangrad” is calculated. The documentation says:

“angletangrad” is the angle between the 3N-length tangent vector and the 3N-length force vector at image i.

I just wanted to make sure whether this 3N-length force vector also contains the contribution of nudging forces?

If that is the case, I was wondering if it would be possible to get the angle between the tangent vector and the “natural” force vector (ie. not including spring forces) for each atom in each replica? As far as my understanding goes, for each atom in each replica, if the natural force has zero perpendicular component to the tangent, then it satisfies the condition for the Minimum Energy Path. So, I could use this data to validate that I have indeed converged to the MEP.

Thanks for your time,
Srijan Singh

Hi all,

My apologies for asking again, but it would really be helpful if I could hear anyone’s thoughts on my idea described in the previous mail.

Thanks,
Srijan Singh

​What you’re proposing is a bit of a chicken-egg thing, because the ‘tangent’ vector at any given replica only approximates a vector tangent to the MEP in the limit of a system near convergence. So if you were to check convergence by verifying that the natural force vector was orthogonal to the tangent, you would also be making the assertion that the tangent line (which, for replica I is just the 3N line between replicas I-1 and I+1) is tangent to the real MEP (i.e. that your system is converged), and if your system is NOT converged, the ‘tangent’ is a very bad approximation to the vector that is tangent to the actual MEP.

Look at the max replica force as a function of step number. If it is still decreasing, the MEP hasn’t converged. Also look at the stability of the MEP itself. MEP stability is a good sign. If you use several similar starting points, do they all converge to the same MEP? If you geometry optimize one of your replicas, does it fall to either end of the MEP, or take a different path? All of these tests can help answer questions about how many local paths there are between reactants and products and which is the minimum energy one.