Rajeev Surendran Assary on using artificial intelligence to boost scientific and technological discovery

Artificial intelligence (AI) and computing are becoming increasingly useful as assistants in the search for the next promising ingredient for advanced batteries, next-generation fuels and more. At the U.S. Department of Energy’s (DOE) Argonne National Laboratory, these tools are wielded by materials scientists who have spent decades exploring how different materials play a role in the future of energy. 

Rajeev Surendran Assary is one of those scientists. Starting at the lab as a postdoctoral researcher 16 years ago, he is now a chemist and group leader in Argonne’s Materials Science division, managing research on understanding and discovering new materials and interfaces that underpin new frontiers in energy conversion, storage and chemical transformation. Here, he talks about his work.

“Conventionally, people discover by trial and error, doing experiments. Our job is to enable that scientific understanding much faster.” — Rajeev Surendran Assary, Argonne chemist

Q: What is the focus of your research at Argonne?
A: My role as a group leader and a scientist is to use reliable predictive modeling at the atomistic scale and state-of-the-art AI approaches to accelerate materials discovery in energy storage and conversion. Conventionally, people discover by trial and error, doing experiments. Our job is to enable that scientific understanding much faster. I think of it like an expert handyman who comes to your house and can do a lot of different types of projects. We are the molecular material handymen of Argonne.

Q: How does your work relate specifically to energy storage and battery research?
A: If you want to build a better battery — one that stores more energy, stays charged longer or uses a different type of chemistry for supply chain reasons — then you need to start with the components. You might want to optimize the properties of the electrolyte, which allows the flow of electrical charge between the two ends of the battery, the electrodes. You might be focused on the electrodes. That’s when we start looking at what are the available materials, and what does the optimal material look like? 

One example is separators, which are semipermeable membranes that keep the battery’s two electrodes apart but allow them to charge and discharge energy. We need a membrane that will allow some charged particles — or ions — to go through and others to stick to the membrane. Let’s say there are 10,000 membranes we are interested in. So, which material would be ideal for a membrane? We can do simulations of one membrane after another with 10 ions or scenarios, then another 10 ions … that’s almost a million calculations. This would be impossible to do with real-world experiments, but we can do it in a matter of minutes with supercomputers and AI tools. 

Q: Can you talk more about material separation as a scientific challenge?
A: Separation — extracting a specific substance from a mix of others — is a huge problem for critical materials. Many of the materials we work with have similar shapes and properties at the molecular level. Being able to separate them out efficiently is essential for mining, building batteries, making drugs — any number of processes that different industries rely on. So, simulations can enable us to design and discover possible materials and processes — the idea of compute before you create. 

For computational discovery to enable practical materials discovery, it is essential to create effective and reliable datasets of materials. Note that there is not a lot of data regarding critical materials and ions such as neodymium, which is used in magnets for wind turbines. So, we are also working on ways to take what we know about such critical materials and use AI to predict how they will behave in certain scenarios. 

Q: What capabilities at Argonne do you use for your work?
A: Mostly we use the Argonne Laboratory Computing Resource Center (LCRC), but we also sometimes use the Argonne Leadership Computing Facility (ALCF), which is a DOE Office of Science user facility. Resources like the Aurora supercomputer at the ALCF, which just came online this year, provide us with a huge opportunity to explore a larger chemical space. 

Large-scale development of a simulated library of fundamental science is very important to the R&D community and educational folks. Therefore, I think targeted and reliable data generation using Aurora has a huge potential for enabling discovery in any given area, such as energy storage, separations and material properties.

Q: What is an example of how computing has accelerated discovery in your research?
A: I have many success stories, but one example is an exploratory project where we decided to look at almost all the curated molecules in the universe. So, we looked at 166 billion molecules as candidates for hydrogen carriers in liquid form. I’m not sure anyone in the world has done research like that. From billions of possibilities, we were able to pinpoint 40 molecules that one day could allow us to safely transport hydrogen in liquid form as fuel for cars, trucks and other vehicles.

Q: What makes Argonne unique in this space?
A: Argonne has world-class computing resources like LCRC and ALCF, but just having a powerful computer doesn’t enable the science. There is quite a lot of domain knowledge needed. For example, if we want to define a particular molecule and its properties, how do we know which properties to target? What types of reliable predictions do we want to make? How do we make sure we have the right data to generate these predictions? For this, you need a combination of leadership, computing and domain expertise across materials, chemistry, imaging and other specialties — and you need the ability to put them together on the fly. That cross-disciplinary expertise is what makes Argonne special.