A modified nudged elastic band algorithm with adaptive spring lengths

Davide Mandelli, Michele ParrinelloThe Journal of Chemical Physics 2021 · DOI · arXiv

Abstract

We present a modified version of the nudged elastic band (NEB) algorithm to find minimum energy paths connecting two known configurations. We show that replacing the harmonic band-energy term with a discretized version of the Onsager-Machlup action leads to a NEB algorithm with adaptive spring lengths that automatically increase the resolution of the minimum energy path around the saddle point of the potential energy surface. The method has the same computational cost per optimization step of the standard NEB algorithm and does not introduce additional parameters. We present applications to the isomerization of alanine dipeptide, the elimination of hydrogen from ethane, and the healing of a 5-77-5 defect in graphene.

Introduction

In the fields of chemistry and materials science, determining the rate of transition between stable states—such as molecular conformational changes or defect migrations—requires mapping the Minimum Energy Path (MEP). The kinetic bottleneck of such a transition is the saddle point, the highest energy point along the MEP.

The standard Nudged Elastic Band (NEB) method is the most common tool for finding these paths. It works by simulating a chain of "images" (snapshots of atomic coordinates) connected by fictitious harmonic springs. However, standard NEB springs typically enforce equidistant spacing. This is computationally inefficient, as it allocates equal resources to flat, stable regions and the critical, high-curvature transition region. Consequently, the saddle point is often poorly resolved unless a large number of images are used.

Methodology: Adaptive Springs via Onsager-Machlup Action

The researchers propose a modified NEB algorithm that achieves adaptive resolution by replacing standard harmonic springs with forces derived from a discretized Onsager-Machlup (OM) action.

In statistical mechanics, the OM action describes the probability of stochastic trajectories. When repurposed for path optimization, the "natural length" of the fictitious springs becomes proportional to the gradient of the potential energy. Since the energy gradient is zero at a saddle point, the springs naturally shrink as images approach the transition state. This causes images to "bunch up" exactly where they are needed most, effectively doubling the resolution at the bottleneck without increasing the total number of images.

Crucially, the authors employ a gradient approximation that avoids calculating the Hessian (second derivatives), ensuring that the computational cost per optimization step remains identical to the standard NEB algorithm.

Results and Discussion

The algorithm was validated on three distinct physical systems: alanine dipeptide isomerization, ethane dehydrogenation, and graphene defect healing.

Alanine Dipeptide Isomerization

As a benchmark for molecular conformational changes, the OM-based NEB was applied to alanine dipeptide. As shown in Figure 1, the modified algorithm significantly increases the density of images near the energy peak compared to standard NEB. This effect is particularly pronounced when using a small number of images (Figure 1b), which is vital for computationally expensive ab initio calculations.

Comparison of energy profiles for alanine dipeptide isomerization using standard NEB and the proposed OM method.

Ethane Dehydrogenation

The method was also tested on the chemical reaction of ethane breaking down into ethylene and hydrogen. Figure 4 illustrates how the OM method clusters images around the transition state, providing a much clearer picture of the energy barrier compared to the standard approach.

Energy profiles for ethane dehydrogenation, demonstrating image clustering at the transition state.

Key Insights and Robustness

  • Physics-Informed Geometry: The algorithm inherently identifies the transition region without prior user input or parameter tuning.
  • Parameter Robustness: While the method introduces a mass-dependent spring constant parameter ($\nu$), the resulting path geometry is remarkably insensitive to it, working reliably across several orders of magnitude.
  • Consistency: The modified algorithm identifies the same MEPs and saddle points as standard NEB, ensuring physical accuracy is maintained while improving efficiency.

Conclusion

By integrating the Onsager-Machlup action into the NEB framework, the authors have developed a method that automatically focuses computational effort on the most critical parts of a reaction path. This adaptive resolution provides a significant advantage for complex simulations where every image calculation is costly. While the current implementation uses a fixed image count, future work could combine this adaptive spacing with algorithms that dynamically add or remove images to further optimize resource allocation.

Summary updated

Abstract

We present a modified version of the nudged elastic band (NEB) algorithm to find minimum energy paths connecting two known configurations. We show that replacing the harmonic band-energy term with a discretized version of the Onsager-Machlup action leads to a NEB algorithm with adaptive spring lengths that automatically increase the resolution of the minimum energy path around the saddle point of the potential energy surface. The method has the same computational cost per optimization step of the standard NEB algorithm and does not introduce additional parameters. We present applications to the isomerization of alanine dipeptide, the elimination of hydrogen from ethane, and the healing of a 5-77-5 defect in graphene.
BibTeX
@article{mandelli2021modified,
  author = {Davide Mandelli and Michele Parrinello},
  title = {{A modified nudged elastic band algorithm with adaptive spring lengths}},
  journal = {The Journal of Chemical Physics},
  year = {2021},
  doi = {10.1063/5.0059593},
}
Figure