Deep learning can almost perfectly predict how ice forms

The lead author of the study says that the properties of matter come from how electrons behave. It's a way to capture more rich physical phenomena.

This is the first time that this method has been used to model ice nucleation. The first step in the formation of clouds is here.

Xiaohong Liu, a professor of atmospheric sciences at Texas A&M University who was not involved in the study, says ice crystals are half of all precipitation events. Ice nucleation could give a big boost to weather prediction.

Laboratory experiments are used to predict ice nucleation. Data on ice formation is collected in a laboratory and fed into models for predicting weather. Sometimes the method works, but often it ends up being incorrect because of the number of variables involved. The results can be vastly different if a few factors are not the same.

It's only valid for a certain region, temperature, or type of laboratory setting.