Understanding the Dielectric Relaxation of Liquid Water Using Neural Network Potential and Classical Pairwise Potential

Understanding the role of hydrogen bond networks in determining the relaxation dynamics is essential for understanding natural phenomena in liquid water. Classical pairwise additive models have been widely utilized for elaborating the underlying mechanism behind the relaxation phenomena. However, they have shown their limits due to either the absence or inaccurate descriptions of many-body and medium-to-long-range interactions. This work demonstrates that the Deep Potential Molecular Dynamics (DPMD) model trained with SCAN functional help calculate the dielectric constant at the accuracy of the first-principles simulations. The DPMD model outperforms the classical force fields (SPC/Fw and TIP4P/epsilon) in predicting dielectric spectra especially in replicating high-frequency excesses, attributed to its adeptness in simulating intricate hydrogen bond networks. Through a comprehensive analysis of the simulation results, it becomes evident that only the DPMD model effectively accommodates a wide range of hydrogen bond coordination scenarios thereby characterizing the intricate nature of the hydrogen bond network. This adaptability stems from the intricate interplay of many-body interactions and intramolecular dynamics. In addition, orientation defects within the DPMD model play a significant role in shaping the potential energy barrier due to the adaptability.