Nudged Elastic Band (NEB)

NEB discretizes the reaction path into a series of replicas (images) connected by harmonic springs. Each image represents an intermediate atomic configuration between the reactant and product states. The spring forces maintain uniform spacing along the path, preventing images from sliding down to the minima, while the true force projection ensures the chain converges to the minimum energy path.

The core algorithmic innovation of NEB is the decoupling of the true force (derived from the PES gradient) and the spring force (maintaining image spacing). The true force perpendicular to the path and the spring force parallel to the path are retained, while the parallel true force and perpendicular spring force are projected out. This prevents the spring forces from corner-cutting and the true forces from collapsing the band.

The Climbing Image NEB is a modification that drives the highest-energy image directly to the saddle point. After a few standard NEB iterations, the image with the maximum energy ceases to feel spring forces. Its true force is inverted along the path tangent, propelling it uphill to the first-order saddle point. This yields the transition state geometry and activation energy with high precision without increasing computational cost.

Accurate path tangent calculation is critical for force projection. The improved tangent method uses a bisection algorithm that switches between forward and backward finite differences based on the energy and force of adjacent images. This avoids kinks and instabilities at regions of high curvature, ensuring robust convergence even for complex, asymmetric reaction pathways.

Advanced NEB implementations employ variable spring constants that adapt based on the local energy landscape. Stronger springs are applied in flat regions to maintain resolution, while weaker springs in steep regions allow images to naturally concentrate near the saddle point. This adaptive scheme improves the resolution of the barrier region without increasing the total number of images.

Modern NEB calculations leverage neural network potentials (NNPs) as the underlying energy and force engine. Unlike traditional DFT-based NEB, which requires thousands of self-consistent field calculations, NNP-driven NEB evaluates energies and forces in milliseconds per image. This enables routine barrier calculations on systems of thousands of atoms with near-DFT accuracy, making ab initio kinetic modeling practical for complex catalytic and diffusive processes.

The fundamental 3N-dimensional landscape defining system energy as a function of atomic coordinates. NEB operates directly on this surface. Key aspects:

• Minima: Represent stable reactant and product states
• Saddle Points: First-order saddle points are the transition states NEB locates
• Curvature: The Hessian matrix at the saddle point defines the reaction coordinate
• NEB requires a smooth, continuous PES, typically from DFT or a Neural Network Potential

The steepest-descent path connecting two local minima on a PES, passing through the saddle point. The MEP is the most probable reaction pathway. NEB discretizes this path into a chain of replicas (images). The converged chain represents the MEP, with the highest-energy image approximating the transition state. The energy profile along the MEP directly yields the activation barrier for rate calculations.

The theoretical framework that uses the NEB result to calculate reaction rate constants. Key equation: k = (k_B T / h) * exp(-ΔG‡ / RT). NEB provides the critical input:

• ΔE‡: Electronic energy barrier from the converged band
• Vibrational Frequencies: From a Hessian calculation at the saddle point
• The single imaginary frequency confirms a true first-order saddle point
• Variational TST refines the dividing surface for more accurate rates

The core algorithmic innovation preventing images from sliding to minima. Each image experiences two orthogonal forces:

• True Force: The negative gradient of the PES perpendicular to the path
• Spring Force: A harmonic force parallel to the path maintaining even image spacing
• The nudging projection decouples these, ensuring the chain converges to the MEP without corner-cutting. Variants like the Climbing Image (CI-NEB) modify the highest image's behavior to converge exactly on the saddle point.

A critical modification where the highest-energy image is driven to the exact saddle point. After a few regular NEB iterations, the climbing image:

• Ignores spring forces entirely
• Inverts the true force along the path direction, pushing it uphill
• Converges rigorously to the first-order saddle point
• Provides a more accurate barrier than interpolating between images. This is the standard production method for reaction barrier calculations.

An alternative saddle-point search algorithm that requires only a single initial configuration, not the full reactant-product path. It uses two nearby images (a dimer) to follow the lowest curvature mode uphill. Often used in conjunction with NEB:

• NEB provides a global path overview
• Dimer method refines the saddle point from a guess
• Useful when the product state is unknown or the reaction coordinate is ambiguous