Researchers at Stony Brook University’s Materials Science and Chemical Engineering Department have been using computers that are capable of learning to recognize various steps in the complex movement of atoms which occurs with many chemical reactions. They hope that the insights gathered from machine learning will aid them in raising the performance of catalysts in order to drive reactions toward desired chemical products more speedily.
The new system was invented by several chemists, computational scientists, and physicists, showing the importance of working at across scientific disciplines.
The new research and machine application shows how neural networks and machine learning can be used to ‘teach’ software to decode information that was hitherto inaccessible. In the studies, such information related to X-ray data. The collected data was then used by the computer to decipher 3D nanoscale structures.
The reason such information is obscured is because when studying key chemical reactions with the creation of products like fertilizers is that such reactions occur at high temperatures and under extreme pressures. This makes the information difficult to track. However, information can be picked up via X-rays, since X-rays can show the atomic-level structures by causing atoms that absorb their energy to emit electronic waves. While X-rays can demonstrate this, such information is difficult to decipher.
Commenting on the success of the machine learning application, lead researcher Professor Anatoly Frenkel told Laboratory Manager magazine: “The main challenge in developing catalysts is knowing how they work—so we can design better ones rationally, not by trial-and-error.”
The chemist goes onto explain: “The explanation for how catalysts work is at the level of atoms and very precise measurements of distances between them, which can change as they react. Therefore it is not so important to know the catalysts’ architecture when they are made but more important to follow that as they react.”
The research has been published in the Journal of Physical Chemistry Letters. The paper is titled “Supervised Machine-Learning-Based Determination of Three-Dimensional Structure of Metallic Nanoparticles.”
