Nudged elastic bands

The nudged elastic band (NEB) method is a computational technique to study the energy landscape and reaction pathways of chemical reactions or phase transitions. It involves the construction of an initial path connecting the initial and final states of a system, with a series of images representing intermediate configurations. These images are connected by springs, which create an elastic band. The method then iteratively adjusts the positions of the images along the band, while minimizing the energy, until a minimum energy pathway (the "nudged" path) is found.

How-to set up a NEB calculation

First, carefully converge the fixed structures of your elastic band, i.e., the initial and the final state. Remember that in the subsequent steps, the elastic band will be attached to these fixpoints so any error will affect the transition path you obtain.

Secondly, construct an initial guess for the intermediate structures. You may use a script like in the [[1]] or come of with your method. This script in the tutorial also offers the option to remove center-of-mass motion. These intermediate images should be somewhat close to the real transition path otherwise the optimization of the elastic band may fail. Place the POSCAR files corresponding to these intermediate structures in subdirectories 01, 02, ...

Mind: It is highly recommended to minimize the number of images used to an absolute minimum. Convergence to the ground state is faster with fewer images. Starting with a single image between the two endpoints and increasing the number of images after the initial run has converged is often a prudent approach.

Thirdly, create an INCAR file in the root directory and set the tag IMAGES to the number of intermediate structures. This will introduce tangential springs to maintain equidistance among images during the relaxation process. You can control the strength with the SPRING tag, where negative values (like the default of -5) activate the NEB method. It is important not to use excessively large values for SPRING, as it can hinder convergence. The default value generally provides reliable results. You should also set IBRION, ISIF, NSW, and other relaxation related tags.

Fourthly, create the remaining input files KPOINTS and POTCAR. For the NEB method, we recommend that all input files, except the POSCAR, WAVECAR and CHGCAR file, reside in the root directory. Then run VASP until the path is optimized.

Possible issues and advice how to address it

One challenge with the NEB method arises from its non-linear constraint, where movements are restricted to the hyper-plane perpendicular to the current tangent. This characteristic can lead to convergence issues with the Conjugate Gradient (CG) algorithm (IBRION=2). In such cases, it is advisable to use alternative algorithms like the RMM-DIIS algorithm (IBRION=1) or the quick-min algorithm (IBRION=3). Additionally, the equidistant images tend to deviate from this constraint in the initial steps. To address this, applying a low dimensionality parameter (IBRION=1, NFREE=2) in the initial steps or using steepest descent minimization without line optimization (IBRION=3, SMASS=2) can help pre-converge the images.

If all degrees of freedom are allowed to relax (e.g., in isolated molecules or surfaces), it is crucial to ensure that the sum of all positions remains consistent across all cells. Failing to do so introduces artificial forces, causing the images to drift apart. While this does not affect the VASP calculations, it can complicate result visualization.

Tip: If you want to run more advanced calculations, you may consider the Transition State Tools for VASP.

Related tags and articles

IMAGES, SPRING, IBRION

Collective jumps of a Pt adatom on fcc-Pt (001): Nudged Elastic Band Calculation

TS search using the NEB Method