High-Throughput Approach for Minimum Energy Pathway Search Using the Nudged Elastic Band Method with Efficient Data Handling and Parallel Computing

The Nudged Elastic Band (NEB) method is critical for mapping chemical reaction pathways but is a computationally and data-intensive workflow involving a large number of single-point (SP) calculations. Additionally, due to the complexity of the NEB method, understanding how variations in the protocol (algorithm, levels of theory, and parameters) impact performance is challenging. To address these issues, we developed and tested a high-throughput approach on the QCArchive cloud-based infrastructure, utilizing two open-source projects, QCFractal and geomeTRIC, to enhance the NEB efficiency. This approach parallelizes SP energy and gradient calculations and stores results in a database, facilitating data organization and retrieval. To evaluate its performance, we optimized four elementary reactions from the RGD1 data set of organic reactions using the B3LYP/6-31G(d), B3LYP-D3/def2-TZVP methods, and the PM7 semiempirical model. We tested 72 different combinations of chain optimization parameters and three types of band forces: conventional NEB, a hybrid band that projects out the perpendicular energy gradient as in NEB but retains the full spring force, and a plain band that does not project any forces. The highest-energy images of the optimized chains were used as the initial structures for transition state (TS) optimization to locate the first-order saddle points. The NEB and TS steps may be performed at different levels of theory, allowing us to perform NEB calculations with either DFT or PM7, followed by TS optimizations at the DFT level. The final TS structures were compared with reference geometries from the data set, which were further optimized at the corresponding level of theory. The convergence rates of TS and NEB are reported to demonstrate how the parameters influence the performance. Next, we performed NEB calculations on 118 diverse chemical reactions from a compilation of seven barrier height data sets from the literature using two selected protocols: one uses the NEB method, while the other employs the hybrid band. Notably, the hybrid band yielded consistently higher convergence rates across reactions from both data sets. Lastly, three elementary reactions from our previous work involving molecular transition metal catalysts were optimized using the hybrid band, successfully reproducing the earlier results. This study demonstrates that the high-throughput approach can perform a large number of NEB calculations concurrently in parallel while storing all calculation results in a database. The results presented here also confirm the reliability and correctness of the new implementation.