Hydrogen energy has long remained an elusive target of the renewable energy industry, promising clean, carbon-free energy that would allow for rapid refueling, unlike current battery-based electric vehicles. Hydrogen-based fuels are expensive to produce, however, rendering it non-viable against both combustion-engine and electric vehicles. Now, supercomputer-powered research from Penn State University (PSU) is aiming to identify materials to economize a crucial part of the hydrogen fuel production pipeline.
The process in question is photocatalysis, through which hydrogen is extracted from water when exposed to electricity or solar energy. Obviously, the less direct the energy, the less efficient the photocatalysis, so researchers have been hunting for a way to use solar energy to directly induce photocatalysis rather than relying on electricity.
“When Thomas Edison wanted to find materials for the light bulb, he looked at just about every material under the sun until he found the right material for the light bulb,” said Ismaila Dabo, an associate professor of materials science and engineering at PSU, in an interview with PSU’s Matt Swayne. “Here we’re trying to do the same thing, but in a way to use computers to accelerate that process.”
The researchers scoured an enormous database of more than 70,000 different compounds for materials that could enable solar-driven photocatalysis when added to water. To achieve this, they developed an algorithm to identify materials that possessed desirable characteristics, such as an ideal energy range or good chemical stability. The algorithm was run on Penn State’s own Roar supercomputer, which is operated through the university’s Institute for Computational and Data Sciences (ICDS) and delivers around 890 peak gigaflops of computing power.
“We believe the integrated computational-experimental workflow that we have developed can considerably accelerate the discovery of efficient photocatalysts,” said Yihuang Xiong, a graduate research assistant at PSU and co-first author of the paper, which was published in Energy and Environmental Science. “We hope that, by doing so, we will be able to reduce the cost of hydrogen production.”
Still, the team isn’t writing off physical experimentation just yet. “Computers can make the recommendations as to what materials will be the most promising and then you still need to do the experimental study,” Dabo said.
At the end of the day, the process identified six promising candidate photocatalysts. Now, the researchers are expanding the scope of their study to include non-oxide chemical compounds – and investigating whether machine learning-enabled algorithms could be used to improve the process.
“So far, we did one cycle of this process on oxides — essentially rusted metals — but there are a lot of compounds that could be made that aren’t based on oxygen,” Dabo said. “For example, there are compounds based on nitrogen or sulfur that we could explore.”