Advanced Materials Through AI & Computational Materials Science

A recent Nature article examines how materials researchers are using artificial intelligence to accelerate quantum-mechanical calculations. Work that once required hours on supercomputers can now, in some cases, be completed in only a few seconds.

These computer-modeling and machine-learning techniques are making it possible to generate enormous libraries of potential materials. Researchers hope this approach will greatly increase both the speed and the practical value of materials discovery. British materials scientist Neil Alford captures the importance of this shift: “We are now seeing a real convergence of what experimentalists want and what theorists can deliver.”

So far, some of the most promising results have involved lithium compounds, which are important for batteries and other energy-storage applications.

The Nature article also argues that “artificial intelligence will help researchers comb through vast numbers of materials to find just the one they need for the application at hand.” In a typical workflow, researchers apply machine learning to laboratory data and computer models. The goal is to identify patterns, predict new materials, and search for candidates with specific properties. Once a promising theoretical material is found, chemists then try to synthesize it in the lab and test whether it behaves as expected.

Personally, I think these materials databases create enormous opportunities. Their potential is almost limitless, but only if researchers can connect computational predictions to useful physical results. The work reminds me of robotic discovery efforts at Dow, advances by Bristol-Myers Squibb and other pharmaceutical companies, and virus-based discovery work such as Angie Belcher’s research. These efforts have produced discoveries ranging from catalysts for the oxidative coupling of methane to battery electrode materials. They are the physical counterpart to the computational approaches now being used in materials databases.

Turning computer predictions into real-world technologies, however, remains difficult. Existing databases still cover only a small fraction of known materials, and an even smaller fraction of all possible materials. Researchers have also learned that data-driven discovery works well for some classes of materials but poorly for others. Even when a model identifies a promising candidate, it can take years for chemists to synthesize it, characterize it, and determine whether it can survive real operating conditions.

Despite these challenges, researchers remain confident that this approach will lead to useful materials for electronics, robotics, healthcare, energy storage, and other fields. In my opinion, the key is to avoid a scattershot approach. If scientists can screen millions of possibilities, they still need to decide which questions are worth asking first. Computational tools can expand the search space, but human judgment is still needed to choose the right goals, constraints, and applications. Otherwise, researchers risk trying to boil the ocean.

Success will require close collaboration between disciplines. Computational scientists may understand the models and data, while experimental scientists may better understand synthesis, measurement, and real material behavior. Neither side is enough on its own. The real value comes from combining these perspectives so that information inside a computer can be translated into useful materials in the physical world.