Multiscale Modeling: The wide-angle lens of science
Walk into an Apple store and you’ll see that the newest phones now incorporate an extra camera to provide wide-angle capabilities, allowing us to capture both the sand we’re standing on as well as the waves crashing on the rocks in the distance together, all in gorgeous resolution. We value the ability to see as much as possible in as much detail as possible, and as a result our world is filled with similar technological marvels that let us do just that—IMAX theaters, aerial drone photography, 3-D TV, and more.
Scientists, especially those that study systems and phenomena at the nanoscale, are no different. They pursue a combined understanding of how things behave “far away” over long time and length scales and “up close” at the molecular level. To get a sense of how small this is, there are approximately eight septillion (10 to the 24th power) molecules of water in your mug. Achieving this combination of detail and scope is crucial. To design materials, we want to predict how changing certain inputs controls a material’s properties and how it will behave over time, but observing and characterizing the molecular-level behavior of a system is essential to ensure we have the fundamental understanding we need to make those predictions.
Unfortunately for scientists, getting a “wide-angle” view is not as easy as purchasing a new phone. To get a complete and detailed picture, they need to turn to a combination of different methods, since a single technique is often insufficient. This can be tricky, just like photoshopping together different images—sometimes the seams show. However, recently published works from the Center for Integrated Mesoscale Architectures for Sustainable Catalysis (IMASC) and the Center for Bio-Inspired Energy Science (CBES) show how explorations at multiple levels of resolution—termed “multiscale modeling”—can be executed elegantly. Such combinations of methodologies can be powerful for advancing materials design for sustainable energy technologies.
Modeling surface restructuring to design more efficient catalysts
One class of materials that is key for energy sustainability are catalysts, materials that accelerate reactions and thus make the process of converting one substance (such as a waste stream) to another substance (such as a valuable chemical product) more efficient. Especially promising within this field are bimetallic catalysts that comprise two different elements, each of which contributes its own advantages to the combination.
“There’s a tradeoff between selectivity [making a specific product] and activity [making a lot of product] that is addressed with this bimetallic composition,” said Boris Kozinsky, a principal investigator in IMASC. “You can have two metals that each do certain parts in the catalytic chain.”
David (Jin Soo) Lim, the lead graduate student on this project, along with Kozinsky and others, studied a bimetallic surface made of palladium and silver. This catalyst helps accelerate reactions that add hydrogen atoms to a material, such as the reaction that turns an unsaturated fat, like oil, into a saturated fat, like margarine. In industry, these reactions can help make products more valuable or less toxic. Despite the ubiquity of the reaction, not much is known about how the bimetallic alloy behaves, which is problematic, according to Kozinsky. “When you make a catalyst, you want to optimize its properties, but if you don’t know how the structure is changing, it’s very difficult to control its synthesis and reliability,” he says.
The IMASC team set out to study how the atoms on the surface can move over time. These mechanisms are crucial to understand because reactions take place at a catalyst’s surface and the surface may respond to the presence of these reactants and products by restructuring.
“Catalytic surfaces are never bystanders in the reaction, they are always participants,” Kozinsky explained.
Previous studies used a technique called density functional theory (DFT) that models the electron density of the atoms in order to find the energies of various surface arrangements. While modeling the system at the electron level helps to precisely capture which arrangements are more energetically favorable, it cannot access the time evolution of the restructuring process that converts one surface arrangement to another. Discovering these mechanisms through modeling requires many evaluations of the energies of different surface arrangements, a task that is too expensive for DFT.
Instead, Kozinsky and others combined a less precise, but faster model of the energy of the system, with an algorithm that can use those energy calculations to simulate how the system will behave over time. By using this approximate method, the researchers could model the system for long enough to observe restructuring events, which they detected automatically by having the computer flag times at which the atoms’ positions change sharply. The researchers could then focus in on these observed restructuring events, using the more precise and expensive DFT method to refine their calculations.
Through this combination of a rigorous method that allows precise investigation of specific events and a faster technique that “widens out” the scope and allows exploration of longer timescales, Kozinsky and others found four different mechanisms through which the palladium and silver surface restructures. Three of these mechanisms were completely novel and had not previously been characterized. The next step for Kozinsky and others is to use novel machine learning techniques to model the surface dynamics over a longer period of time: “We have started to utilize a machine-learned forcefield, developed by Jonathan Vandermause and co-workers in our group, to perform large-scale simulations of surfaces spanning microseconds. We recently used those methods to investigate restructuring in other palladium-silver systems,” explained Lim.
Developing robots using particle and continuum-level models
In addition to controlling how materials react, scientists at CBES—directed by principal investigator Samuel I. Stupp—are working on how to control how materials move in an effort to mimic living matter with synthetic structures. Researchers at CBES, led by Monica Olvera de la Cruz, are looking into using magnetic fields to control how materials deform and move. Their recent work focuses on understanding how changing the strength of a magnetic field controls the behavior of sheets of superparamagnetic particles, which are particles that respond immediately to a magnetic field. Using this understanding, they will be able to design robots comprised of these superparamagnetic particles that move by converting magnetic energy into movement, or kinetic energy. This conversion is much cleaner than how we typically get things to move, such as by burning fossil fuels to power cars. Furthermore, these robots would be controllable in a non-invasive way, which is promising for applications in medicine:
“The main advantage of this type of system would be that it’s totally bio-orthogonal. It can penetrate into the tissue extremely easily without affecting the body,” explained Chase Brisbois, a researcher who was part of this study.
To explore this system, Brisbois and others used a set of models with varying levels of detail. One of them, the “particle-level model,” had variables and equations for each individual superparamagnetic particle in the sheet. Simulating this model involves solving all those equations and adjusting the values of all those variables simultaneously using an algorithm called molecular dynamics, which is computationally expensive. Another model, the “continuum model,” treated the sheet as a single entity with variables describing the sheet as a whole, and was therefore easier to simulate and also much easier to scale to larger length scales.
Through these models, the researchers were able to show that the factors that govern the behavior of the membrane can be combined into a single parameter, termed a “magnetoelastic parameter.” Knowledge of this parameter is useful because it summarizes how different factors, such as the magnetic field and membrane size, affect membrane behavior. Both levels of models—“particle” and “continuum”—predicted the same trends in membrane behavior with changes in the magnetoelastic parameter, but each technique was also uniquely important. The particle-level molecular dynamics captured some phenomena that the continuum model could not.
“That’s why the molecular dynamics model was valuable,” said Brisbois. “It was able to look at a wider range of parameters.” On the other hand, the continuum model was scalable to longer length scales, which is critical for these researchers in their collaboration with experimental groups that work at millimeter and centimeter scales.
“The [techniques] all give you a deeper understanding of the problem,” said Olvera de la Cruz. “Extending all the way from nano- to micro- to millimeters is the nice thing about using multiscale modeling.”
For science, getting a “wide-angle” picture by integrating different types of models at different levels of resolution is important. On the molecular level, models can capture details of the mechanisms that are crucial to understanding next-generation energy technologies such as more efficient catalysts and remotely controllable robots. When used together, they can help develop design rules for time and length scales that are more comparable to the macroscopic world we live in, and that therefore can ultimately help us engineer and build those groundbreaking, new materials.
Lim JS, N Molinari, K Duanmu, P Sautet, and B Kozinsky. “Automated Detection and Characterization of Surface Restructuring Events in Bimetallic Catalysts.” J. Phys. Chem. C 123:16332. DOI: 10.1021/acs.jpcc.9b04863.
Brisbois CA, M Tasinkevych, P Vázquez-Montejo, and M Olvera de la Cruz. 2019 “Actuation of Magnetoelastic Membranes in Precessing Magnetic Fields.” Proc. Natl. Acad. Sci. 116:2500. DOI:10.1073/pnas.1816731116.
Lim et al. This work was supported by the Integrated Mesoscale Architectures for Sustainable Catalysis (IMASC), an Energy Frontier Research Center funded by the US Department of Energy, Office of Science, Basic Energy Sciences under award no. DE-SC0012573. This research used the following computational resources: (1) the Odyssey cluster, FAS Division of Science, Research Computing Group at Harvard University; (2) the Oak Ridge Leadership Computing Facility, a DOE office of Science User Facility supported under contract no. DE-AC05-00OR22725, through allocation CHP106; (3) the National Energy Research Scientific Computing Center (NERSC), a DOE Office of Science User Facility supported under contract no. DE-AC02-05CH11231, through allocation m3275; and (4) the Extreme Science and Engineering Discovery Environment (XSEDE), supported by National Science Foundation grant number ACI-1548562, through allocation che170060. We acknowledge enlightening discussions with M. Stamatakis, C. M. Friend, R. J. Madix, E. Kaxiras, W. Chen, L. Sun, and R. Réocreux.
Brisbois et al. This work was supported as part of the Center for Bio-Inspired Energy Science, an Energy Frontier Research Center funded by the U.S. Department of Energy, Office of Science, Basic Energy Sciences under Award DE-SC0000989. M.T. acknowledges financial support from Portuguese Foundation for Science and Technology Contract IF/00322/2015. P.A.V.-M. acknowledges support from Consejo Nacional de Ciencia y Tecnología Grant Fondo Institucional de Fomento Regional para el Desarrollo Científico, Tecnológico y de Innovación (FORDECYT) 265667.