Efficient acceleration of the convergence of the Minimum Free Energy Path via path-planning generated initial guess

Yi Sun

DOI: https://doi.org/10.26434/chemrxiv-2024-58tt1-v2

2024-03-13

Abstract:We demonstrate how the combination of a shifted clustering algorithm and a fast-marching- based algorithm is able to generate good approximations of the Minimum Free Energy Path (MFEP) if a Free Energy Landscape (FEL) is given. Then, we show that using this kind of approximation as the MFEP’s first guess and the string method for further refinement (also called the FMT-string combined approach) cuts down on the number of iterations needed for the MFEP to converge by a large amount. This saves a lot of time compared to using a linear interpolation as the first guess. Such a method provides an efficient alternative to the growing string method for obtaining a good initial guess of the MFEP.

Chemistry

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to efficiently accelerate the convergence of the Minimum Free Energy Path (MFEP). Specifically, the author proposes a new method that combines the Fast Marching Tree (FMT) algorithm and the string method to generate a better initial guess path for MFEP. This method can significantly reduce the number of iterations and the computation time, thereby improving the computational efficiency.

### Main Problem Background

In theoretical chemistry, analyzing the most probable reaction path from reactants to products is one of the core tasks. Locating transition states (TS) is crucial for calculating the reaction barrier height. However, it is not easy to find a suitable TS, and it is usually more challenging than finding the energy minima of stable reactants and products. Traditional single - point TS optimization and Intrinsic Reaction Coordinate (IRC) calculations may not converge correctly because low - frequency modes affect the quality of the Hessian matrix.

To solve this problem, collective variables (CV) and molecular dynamics (MD) simulations can be used to evaluate the Free Energy Landscape (FEL) and then find TS and obtain MFEP. These methods only require gradient calculations and are not affected by low - frequency harmonic modes. To overcome the activation barrier, enhanced sampling methods such as umbrella sampling, MetaDynamics, and replica exchange are usually adopted.

### Limitations of Current Methods

The currently mainstream method is the "growing string method", which generates an initial path by linear interpolation and optimizes it using gradients or Hessians. However, this linear - interpolation method may not be accurate enough in complex systems and requires a large number of iterations to converge to the correct MFEP.

### Solution Proposed in the Paper

This paper proposes a new method that uses the FMT algorithm to generate an initial path and then further optimizes it using the string method. FMT is a path - planning algorithm based on the Fast Marching Method, originally used for path planning in the robotics field. By applying FMT to FEL, an initial path closer to the real MFEP can be generated, thereby reducing the number of iterations required for subsequent optimization.

### Specific Steps of the Method

1. **Construct FEL**: First, construct a convergent FEL through MD simulation or other methods.
2. **Path Planning**: Use the FMT algorithm to plan a path from the starting point to the ending point on FEL. The FMT algorithm will consider the cost of the path, where "speed" is inversely proportional to energy.
3. **Path Refinement**: Use the path generated by FMT as an initial guess and input it into the string method for further optimization until the path converges to MFEP.

### Experimental Results

By testing on simple model Free Energy Surfaces (FES) and complex Muller - Brown potential surfaces, the results show that, compared with the linear interpolation method, the initial path generated by FMT requires significantly fewer iterations and the computation time is also greatly shortened. For example, in some cases, the FMT path only needs 6 iterations to reach convergence, while the linear interpolation path requires 24 iterations.

### Conclusion

In summary, the method proposed in this paper provides a new way to efficiently generate the initial path of MFEP. Combining FMT and the string method can significantly improve the computational efficiency and reduce the computation time. This method is expected to play an important role in extracting MFEP on complex free energy surfaces.