Comparative Study of Bond Dissociation Energies of Hydrides using Ab Initio, Quantum Monte Carlo, and Machine Learning Approaches

Bond dissociation energy (BDE) is a commonly quantified measurement of a chemical process, providing insights to its reactivity and stability. Hydride bond breaking, in particular, is of great interest due to its involvement in combination of catalytic and organometallic processes. Despite having many applications, accurate determination of BDEs, specifically metal-hydride bonds, remains a challenge due to the combined effects of electron correlation, relativistic contributions, and nuclear quantum effects. In this work, we present a comparative study of BDEs, starting from lighter molecules such as LiH, and OH, then moving on to TM-hydride: TiH, and NiH. The goal of this study is testing and understanding the cost and efficiency of neural network wavefuctions in prediction of the chemical properties while comparing the same with widely used ab initio quantum chemistry approaches and real space stochastic Quantum Monte Carlo methods. This study also throws light on the advantage of using pseudopontentials with increasing size of the system and benchmarks the results for the recently developed ccECPs. By analyzing these molecules, we assess the strengths and limitations of each method, highlighting the areas of agreement and discrepancy with available experimental data.