How to Predict a Stable Material Without Experiments ✛ ✦
Searching for a new material is like assembling a dream team: you need to guess which metal atoms will get along and withstand the heat. In the past, this took years of random guessing. Now, scientists from France and Japan have applied machine learning and computer modeling based on the laws of quantum physics to predict the stability of mixtures in seconds.
The secret lies in entropy — a measure of disorder in the crystal. Ludwig Boltzmann was the first to realize that nature loves chaos, and in mixtures of metals with oxygen this chaos helps atoms stick together — just like in a team, a diversity of talents adds strength. The neural network was trained on five hundred successful examples (including iron, copper, zinc), and it evaluated all 4368 possible five-element combinations. The error turned out to be negligible — like a drop in the ocean. Most importantly: the computer not only confirmed known stable compositions but also found three new ones, which were then successfully synthesized in the lab.
Now materials for batteries and electronics are designed with purpose. The key is a balance between disorder and chemical affinity: the ability of metals to share electrons and the sizes of their atoms. Just like in a good team, both novelty and compatibility matter.
If you decided to test all 4368 possible combinations manually, spending a week on each, it would take you about 84 years of continuous work. The neural network does it in a fraction of a second.