Water evaporation with machine learning interatomic potentials trained on MB-pol

Evaporation plays a critical role in ecological and environmental processes, yet computational investigations have thus far been limited by the lack of water models with quantum-mechanical accuracy that are also computationally efficient. To address this challenge, we employ the recently developed neuroevolution potential (NEP), which is trained on extensive many-body polarization (MB-pol) reference data and achieves a favorable balance between accuracy and efficiency. Using NEP-MB-pol in molecular dynamics simulations, we perform a systematic study of water evaporation. We first establish the vapor–liquid equilibrium, finding that the liquid and vapor densities at different temperatures, as well as the fitted critical points, are in excellent agreement with reference values, underscoring the predictive capability of the employed model. We then revisit the microscopic mechanism of evaporation. Our MD simulations show that an evaporating molecule must remain in a highly energetic pre-evaporation state for several 100 fs. A successful evaporation involves the cooperative interactions of at least four water molecules, with the last collision occurring within a short time window of ∼56 fs before evaporation. Finally, motivated by recent intriguing experimental observations, such as the photomolecular effect, we investigated the impact of external electric fields on water evaporation. In contrast to experimental findings, we did not observe a consistent effect from green light on water evaporation, i.e., the photomolecular effect was not reproduced. This may be attributed to the negligence of quantum effects in our simulation. Overall, our study provides new microscopic insight into the evaporation process and offers valuable guidance for experimental studies and potential industrial applications.