Computational Modeling Streamlines Hunt for Battery Electrolytes
April 11, 2023
By Elizabeth Ball
Using computing resources at the National Energy Research Scientific Computing Center (NERSC) at Lawrence Berkeley National Laboratory (Berkeley Lab), researchers from the Joint Center for Energy Storage Research have identified new, more efficient ways to find improved electrolytes for batteries. By computationally modeling molecules and virtually observing their properties, researchers can identify the most promising ones and save experimental scientists from spending time and resources on those that won’t work. The results of this work were published in The Journal of Physical Chemistry in 2022.
Currently, most electric vehicles and consumer electronics are powered by lithium-ion batteries, which are monovalent, meaning each lithium ion in the battery is only able to bond with one other atom at a time. There are, however, other options. Batteries such as magnesium-ion and calcium-ion batteries are multivalent, meaning they use ions that can bond with more than one other atom at a time; this may allow them to achieve higher energy density than monovalent batteries. They’re also lightweight and can be derived from minerals with higher annual production than lithium-ion and associated metals. However, more information about the stability of their electrolytes is needed before those batteries are ready for widespread use.
“Multivalent batteries have a number of challenges associated with them; there’s a reason we don’t have a working magnesium or calcium battery in any of our devices,” said Berkeley Lab materials scientist Kristin Persson, an author on the paper.
According to Persson, one of those challenges is finding the right electrolytes. The hunt for new, more effective electrolytes has so far taken place experimentally, a laborious process of testing one type of molecule and all its variations at a time and observing the results – though, as Persson notes, researchers often aren’t sure why a molecule fails as an electrolyte, only that it does. That’s because the scale of these reactions is so small and their duration is so short that they cannot be measured with current experimental tools.
This paper proposes another way: using computational models to assess molecules and their properties before screening them experimentally.
“We’re interested in doing computational-based prediction first, and then giving some guidance to the experimental side so they don’t have to synthesize all the possibilities, and just try out the things that are computationally predicted to be good,” said the paper’s primary author, Xiaowei Xie, who was a graduate student at UC Berkeley when she did this research.
The method allows researchers to computationally “pause” the battery reaction midstream to visualize and understand processes that can’t currently be observed or measured experimentally. This window into how each molecule behaves during a reaction is a powerful screening tool and may give researchers a better idea of how to proceed in the future.
“We can ‘freeze’ it computationally in the computer,” said Persson. “And we can look at these molecules to see whether they’re stable, and examine whether they have the potential to be useful. In the past, you’d test molecule after molecule experimentally and just see that they didn’t work. In this case, you can see where the process is failing.”
In the paper, Xie assessed a series of salts and determined that two of them, the hexafluoro-tert-isopropoxy (hfip) ligand for borates and the trifluoro-tert-butoxy (tftb) ligand for aluminates, have promise as anions; the borate salt is a confirmation of previous experimental work and of the effectiveness of computational modeling. The predicted aluminum salt hasn’t been synthesized before and is currently being tested experimentally at Argonne National Laboratory.
Xie ran her simulations on the Cori supercomputer at NERSC, in addition to other systems, using the Q-Chem software for ab initio prediction of molecular structures. Molecular screening is a good candidate for supercomputing, said Xie, because the vibrational frequency calculations to get the free energy estimate are very memory-intensive – “the more computing nodes with the more memory, the better,” she said. “We used a ton of computing to do it!” added Persson.
Xie’s work took place before NERSC’s current GPU-enabled system, Perlmutter, was available, but in the future, she says, this kind of work – and battery science as a whole – will benefit from the new system: “Q-Chem released a new version that works on GPUs, and some people have benchmarked it to compare with the CPU version and found that it was faster in some cases,” she said. “I definitely think it would be promising to use GPUs for this work.”
Zooming In and Scaling Up
Now that computational modeling has been proven to improve screening efficiency, researchers can begin testing a practically endless number of potentially useful molecules and exploring all their forms and variations, as well as how they behave under a variety of conditions. “Do they decompose, and what do they form when they do?” said Persson. “What happens if you add other molecules and additives to the mix?”
Additionally, researchers will continue to use the “pause” functionality to observe reactions in progress and gain understanding of why some molecules fail as electrolytes.
Ultimately, refining and expanding computational modeling for molecules may streamline the process of finding better electrolytes for all kinds of batteries, since researchers may not need to spend as much time testing molecules that lack the properties they’re looking for; instead, they can start with a smaller subset.
“It’s exciting,” said Persson, “because before the Joint Center for Energy Storage research that we’ve done now for ten years, I wasn’t aware of a single case of computational screening and design of novel molecules – literally being able to examine the molecule in the computer, having the design metrics that agree with what we see experimentally, and saying, ‘OK, now, you don’t have to make these molecules anymore.’ We’re going to prescreen the molecules and give a smaller selection of more useful, more-likely-to-succeed candidates that you can then synthesize and make.”
NERSC is a U.S. Department of Energy Office of Science user facility.
About NERSC and Berkeley Lab
The National Energy Research Scientific Computing Center (NERSC) is a U.S. Department of Energy Office of Science User Facility that serves as the primary high-performance computing center for scientific research sponsored by the Office of Science. Located at Lawrence Berkeley National Laboratory, the NERSC Center serves more than 7,000 scientists at national laboratories and universities researching a wide range of problems in combustion, climate modeling, fusion energy, materials science, physics, chemistry, computational biology, and other disciplines. Berkeley Lab is a DOE national laboratory located in Berkeley, California. It conducts unclassified scientific research and is managed by the University of California for the U.S. Department of Energy. »Learn more about computing sciences at Berkeley Lab.