Nudged-Elastic-Band Calculations

• Nudged-Elastic-Band (NEB) Calculations are chain-of-states methods that discretize reaction paths with images and springs to locate minimum-energy paths and transition states.
• Enhanced variants, such as climbing-image and multi-climbing NEB, improve saddle point accuracy by adjusting forces to navigate complex, curved energy landscapes.
• Practical implementations rely on precise tangent estimation and stringent force convergence criteria, resulting in robust activation barrier quantification and transition state detection.

The Nudged-Elastic-Band (NEB) method is a widely used chain-of-states algorithm for identifying minimum-energy paths (MEPs) and locating transition states (TS) on high-dimensional energy surfaces. Given fixed initial and final configurations, NEB generates a sequence of intermediate “images” coupled by artificial springs and relaxes them under projected forces to trace the MEP between metastable states. The method plays a crucial role in quantifying activation barriers, elucidating atomic-scale mechanisms in solids and molecules, and supporting rate calculations in transition-state theory.

1. Mathematical Formulation of NEB

NEB discretizes a reaction path by $N$ movable images $R_1, ..., R_N$ (in $3N$-dimensional configuration space if considering $N$ atoms), with fixed endpoints $R_0$ and $R_{N+1}$. The total NEB energy functional is

$$E_{\mathrm{NEB}} = \sum_{i=0}^{N+1} V(R_i) + \sum_{i=1}^{N} \frac{k}{2}\|R_{i} - R_{i-1}\|^2,$$

where $V(R_i)$ is the physical potential energy and $k$ the spring constant. Forces on each image $i$ are decomposed via projection operators onto the local path tangent $\tau_i$:

• Spring force along the tangent:

$$F_{\text{spring},i} = k \left( \|R_{i+1} - R_{i}\| - \|R_{i} - R_{i-1}\| \right)\, \tau_i$$

• True force perpendicular to the tangent:

$$F_{\perp, i} = -\nabla V(R_i) + [\nabla V(R_i) \cdot \tau_i] \tau_i$$

The total force is $F_i = F_{\perp,i} + F_{\text{spring},i}$ (Zarkevich et al., 2014).

Convergence is checked via a norm of the NEB forces, typically requiring $\max_i \vert F_i \vert < 0.01 - 0.03$ eV/Å for ab initio calculations.

2. Climbing-Image and Multi-Climbing Variants

To accurately locate transition states, NEB is extended by the climbing-image protocol (CI-NEB), which modifies the forces for the highest-energy image: $$F_M = -\nabla V(R_M) + 2 [\nabla V(R_M) \cdot \tau_M] \tau_M$$ where $M := \arg\max_i V(R_i)$, suppressing the spring force and inverting the parallel component (Zarkevich et al., 2014).

The two-climbing-image NEB (C2-NEB) generalizes this further for complex energy landscapes, such as serpentine MEPs. Here, the immediate neighbors of the highest-energy image, $M-1$ and $M+1$, become climbing images: $$F^{\rm C2}_i = \begin{cases} -\nabla V(R_i) + 2 [\nabla V(R_i)\cdot\tau_i] \tau_i, & i = M-1 \text{ or } M+1 \ F_{\perp,i} + F_{\text{spring},i}, & \text{otherwise} \end{cases}$$ with the highest-energy image $M$ being “nudged” per the standard NEB scheme. C2-NEB improves stability and accuracy for transition-state searches where the path tangent is poorly aligned with the true MEP—such as in solid-state martensitic transformations—by bracketing the saddle point and minimizing tangent misalignment and re-parametrization errors (Zarkevich et al., 2014).

Selection criteria require reverting to C1-NEB when the highest-energy image is adjacent to a fixed endpoint.

3. Tangent Estimation and Path Projection

Accurate path tangents are vital, especially in regions of strong curvature. The tangent at image $i$ is typically approximated as

$$\tau_i = \frac{R_{i+1} - R_{i-1}}{\|R_{i+1} - R_{i-1}\|}$$

Improvements, such as the Henkelman-Jónsson “energy-weighted” tangent specification, use energy differences to avoid kinks and discontinuities. This is especially important in climbing-image NEB, where correct projection of the forces is required to prevent divergence near saddle points (Zarkevich et al., 2014).

4. Practical Implementation and Algorithmic Steps

A standard NEB workflow comprises:

• Linear interpolation of $N$ images between endpoint configurations.
• Selection of spring constants (typically $k=1$–$2$ eV/Å$^2$ for DFT or higher for empirical potentials).
• Loop over optimization steps:
  • Evaluate energies and forces for all images.
  • Compute tangents.
  • Project spring and true forces for each image.
  • Apply climbing-image modifications as needed.
  • Propagate all images using an optimizer (velocity-Verlet, quasi-Newton, FIRE, or LBFGS).
• Convergence declared when all NEB force norms fall below tolerance (Zarkevich et al., 2014).

The C2-NEB logic requires tracking the position of the highest-energy image and, if appropriate, enabling climbing-only for its neighbors.

5. Theoretical Rationale for Multi-Climbing NEB

The necessity of C2-NEB arises in energy landscapes where the MEP direction near a transition state is significantly curved (“serpentine paths”). With sparse image placement, single climbing-image NEB can misalign the path tangent with the true MEP, causing instability—images swap and saddle points are missed. By climbing from both sides, C2-NEB approaches the saddle along the true geodesic, provides localized bracketing, and allows a triadic accuracy estimate (M–1, M, M+1) (Zarkevich et al., 2014).

This modification is especially relevant in fixed-cell and generalized solid-state NEB (SS-NEB) protocols, where atomic and cell degrees of freedom must be treated consistently.

6. Applications, Performance, and Convergence

Benchmarks in martensitic transition pathways (e.g., NiTi austenite→martensite, with 324 DOF) showed standard NEB and C1-NEB failing or overestimating barriers, while C2-NEB with sufficient images (e.g., 8) converged to barriers within 1 meV/atom. In simple transitions (e.g., BCO→B19′), C2-NEB yields smoother convergence and a direct measure of TS accuracy via the energy triad (Zarkevich et al., 2014).

Spring constants and force tolerances must be tuned to balance stiffness and true-force fidelity. Force convergence below 0.01–0.03 eV/Å is typical for accurate saddle localization.

C2-NEB logic is equally applicable to SS-NEB cases where both atomic positions and cell strain are variable.

7. Connections, Limitations, and Further Generalizations

The C2-NEB advance builds directly on prior tangent-estimation and spring-force definitions [Henkelman & Jónsson, JCP 113, 9901 (2000); Sheppard et al., JCP 136, 074103 (2012)]. While C1-NEB is adequate for simple landscapes, broader uptake of C2-NEB is advised in high-curvature MEPs with complex geodesic structure or in high-dimensional solid-state transformations. Theoretical justification, algorithmic pseudocode, and recommendations for parameter selection are detailed in (Zarkevich et al., 2014).

In summary, C2-NEB provides systematically improved transition-state detection, robust convergence in general energy landscapes, and better internal accuracy diagnostics relative to classical NEB and single climbing-image variants.