A new study has managed to accurately represent water’s movement. The U.S Department of Energy’s Argonne National Laboratory conducted this experiment. The researchers used a machine learning model to derive characteristics of water using mathematical precision.
The team utilized extensive data at their disposal to uncover a highly accurate model. According to Subramanian Sankaranaryanan, an Argonne nanoscientists, the water model proved to be a challenge. There are more than 50 water models, he added.
Water is a simple substance to understand. However, modelling its behavior has been a different ball-game altogether. Till date, not a single model has thoroughly represents its wide range of characteristics. Hence, this study can mean a new lease of life for scientists studying water.
Supercomputers Must for Machine Learning
His team tried to understand the navigational path which ultimately led them capture a wide spectrum of water’s properties. According to Subramanian, there is no existing model which measures the water’s melting point, the density of ice, and water’s maximum density at the same time.
Additionally, the researchers also achieved the breakthrough at low computational costs. The research gained significantly by the use of supercomputers at the Argonne Leadership Computing Facility. The team later performed stimulations up to 8 million water molecules. This research helped them expand their understanding formation and growth of interfaces in polycrystalline ice.
According to Henry Chan, the lead author of the study, water’s behavior has puzzled researchers due to their computationally intensive models. Additionally, earlier models also failed to produce results and describe the many temperature-dependent properties of water. Mr. Chan believed, it is even more difficult for simple models like the one they had in mind.
Hopefully, the world of supercomputers and machine learning holds many brighter promises for researchers.