Improving materials by understanding processes on the atomic scale

Stories about exploding smart phones or burning e-bike or e-car batteries occasionally make the headlines. Even if it does not happen as often as it seems, lithium-ion batteries, which are used very frequently, pose a certain danger. "They are susceptible to malfunctioning, which can sometimes be severe and lead to them catching fire and exploding. One promising alternative is to replace the flammable liquid electrolytes contained in batteries with safer solid-state electrolytes," says Blazej Grabowski, Professor at the Institute for Materials Science at the University of Stuttgart. He and his team want to improve these novel electrolytes and are conducting research on the question of what materials can be used for the electrolytes.

Fortunately, you do not have to simulate this unimaginable number of atoms directly. The now achievable number of one million atoms already makes it possible to simulate the determinative defects in crystals with nearly ab initio accuracy. These defects have a very strong influence on the material behavior.

However, several orders of magnitude have to be bridged before scientists can simulate the behavior from the atom to the respective application. To this end, they are developing suitable methods within a so-called multi-scale approach. This can be used to bridge the different orders of magnitude and physical worlds across many scales in the simulation. For example, in order to investigate the lithium movement on the micro and millimeter scales, the results from the atomistic simulations are passed on to SimTech colleagues Professor Felix Fritzen and early career researcher Lena Scholz from the continuum area. Here, too, machine learning approaches are helpful.

"The potential of machine learning is extremely important when bridging the scales. Ab initio computations are not new. However, it was only with the help of machine learning that we were able to bring the accuracy of the ab initio method to a larger scale to simulate, for example, lithium diffusion through a complex crystal with defects," explains Blazej Grabowski. Conventional ab initio methods can simulate only a few hundred atoms, as the simulations require an extremely large amount of computing power. Machine learning – especially in the form of so-called machine-learning interatomic potentials – reduces the necessary computing power and the time required. With this approach, the scientists have succeeded in simulating the behavior of up to one million atoms with a high degree of accuracy. 

In order to improve materials, it is essential to precisely understand the interaction between the individual atoms. These interactions determine, for example, the strongly temperature-dependent movement of dislocations, which in turn are responsible for the plastic deformation of materials. "For example, if you manage to make aircraft turbines operate at slightly higher temperatures, you can increase efficiency. Just a few degrees more that the materials in the turbine are able to withstand, can have a positive effect on fuel and therefore energy consumption," says Grabowski.

Nickel-based superalloys, for example, which are incorporated in the innermost turbine blades, have to withstand extremely high temperatures and pressures over long periods of time and over many cycles, so that the material does not simply melt away or deform. "These are fantastic high-tech materials that are used in turbines. They are not new in principle, but the exact processes that contribute to the unique material properties are not fully understood," says Grabowski. Together with Dr. Xi Zhang and early-career researcher Xiang Xu, he was able to contribute to a better understanding of the material behavior on the atomic scale with simulations of complex dislocations. This understanding is valuable for the further optimization and design of new materials for such extreme applications.

The scientists continuously check whether their simulations are transferable into practice by comparing them with experimental data from the scientific literature. "Even the ab initio methods contain approximations whose effects need to be carefully checked. Improvements in the approximations make it possible to achieve increasingly reliable results and to reproduce or predict experiments more accurately," says Grabowski. If experimental data on certain properties is not yet available, the scientists cooperate with other researchers from science and industry to obtain it.

However, he and his team do not work on real aircraft turbines – they leave that to the engineers. "We only work on the individual materials that later make up a turbine. The engineers who assemble the turbine need to know up to what temperature the material is durable and choose the optimum material based on various criteria." With their basic research, Blazej Grabowski and his team provide the input that the engineers need. With their simulations, the scientists contribute to a deeper understanding of the processes on the atomic scale and thus lay the foundation for the next generation of functional and structural materials.

Manuela Mild | SimTech Science Communication