Emily Crabb

Massachusetts Institute of Technology

After Emily Crabb’s father received his chemistry doctorate decades ago, job interviewers expressed suspicion at his interest in some of the latest technology.

Computational chemistry was rare in those days, Crabb says. Some recruiters told her father “computers were just a fad. Why would you do a Ph.D. in this?”

Charles Crabb went on to a long career as a scientist for a chemical company near Philadelphia. His daughter, meanwhile, didn’t have to convince anyone that high-performance computing (HPC) simulations were useful for finding better energy storage materials. Her mother, Katherine – a mathematics teacher who trained as a chemical engineer – “sometimes claims I’ve become a clone of my father,” says Crabb, a Department of Energy Computational Science Graduate Fellowship (DOE CSGF) recipient.

At the Massachusetts Institute of Technology, Crabb works with advisor Jeffrey Grossman to improve the prospects for lithium-air batteries, devices that pack more power into less space than standard lithium-ion batteries because they use ambient oxygen in place of a standard cathode to move lithium ions when charging or discharging. “What I’m trying to do is understand on an atomic level what’s going on when you have a certain type of lithium-air battery,” Crabb says.

One problem Crabb and her colleagues hope to address is the buildup of lithium peroxide on electrodes during discharge. The substance can form as a thin film or as a doughnut-shaped structure, but it’s unclear why the deposit varies. Each affects the battery’s capacity differently.

They also study electrolytes – the substances, often fluids, that conduct ions as the battery charges or discharges. Crabb plans to use simulations to understand the chemical stability of various candidate materials.

Crabb: lithium

The group has first focused on comparing mathematical methods the simulations use. Computational chemistry researchers have several to choose from: molecular dynamics (MD), a classical physics technique that tracks how atoms or molecules interact; density functional theory (DFT), which includes the strange quantum-mechanical realm where electrons can be both particles and waves (an ab initio method, in which preset conditions or experimental data don’t inform the simulations); and ab initio molecular dynamics (AIMD), which uses DFT to calculate electron forces at every time step in a simulation. “It’s quite expensive and quite slow,” Crabb says, “while classical MD can access longer time scales and bigger atomic systems, but it’s not quantum.” Because AIMD is computationally expensive, researchers often use classical MD to calculate the best starting point for a limited number of ab initio simulations.

Crabb and Grossman have found that the quality of AIMD simulations often depends on choosing a good starting point – and thus a good method to calculate it. Unfortunately, not every researcher adequately reports their initial computations in published papers. “We’re trying to emphasize the importance of documenting all this,” Crabb says, because “maybe it doesn’t have an effect for certain systems, but maybe it does, and you can’t reproduce results unless you have background information.”

Crabb has run many of her calculations on the Cori supercomputer at DOE’s National Energy Research Scientific Computing Center and on the Comet system at the San Diego Supercomputer Center. She and Grossman will continue researching possible electrolytes and may expand their mathematical toolbox to include machine-learning methods.

Crabb gathered lessons in another computational technique while on her 2017 Argonne National Laboratory practicum. Working with materials scientist Olle Heinonen, she focused on the properties of hafnium dioxide. The compound is an important semiconductor material, but it can leak electric current when tiny transistors are packed into ever-smaller spaces. Crabb’s job was to calculate the characteristics of different hafnium dioxide polymorphs – atomic arrangements, such as cubic.

DFT performs poorly when modeling heavier elements because it must track an overwhelming number of electron-electron interactions. To compensate, Crabb first used a quantum Monte Carlo (QMC) technique, which reduces the problem size by randomly sampling a range of electron interactions to calculate the electrons’ minimal local energy. She plugged that value into a DFT method to calculate the hafnium dioxide polymorph properties. Finally, she used QMC again to calculate the polymorph at larger sizes.

The project was one of the first to run on Argonne’s Theta HPC system, which launched that summer. She contributed to a paper in the journal Physical Review Materials that reported calculations for both hafnium dioxide and zirconium dioxide.

Crabb expects to graduate in 2022. Because of her practicum experience, she’s added a national laboratory postdoctoral research post to her options, along with academic and industry positions.

Image caption: The solvation shell, where ions are surrounded by a layer of solvent molecules, for two different lithium ions at the end of an ab initio molecular dynamics simulation of a system of 50 acetonitrile molecules and three lithium bis(trifluoromethanesulfonyl)imide (LiTFSI) molecules. At left, the Li+ ion is coordinated with one negatively charged TFSI- ion and three electrolyte molecules. In the right figure, the Li+ ion is coordinated with four electrolyte molecules. The environment in which lithium ions form complexes with surrounding solvent molecules is an important factor governing the course of reactions in lithium-air batteries and depends on the electrolyte and lithium salt used in the system. Part of Emily Crabb’s research uses computation to probe the Li+ ion solvation environments in different systems. Credit: Emily Crabb, using the Medea®-3.0 software package.