Our research highlights serve as a collection of feature articles detailing recent scientific achievements on GCS HPC resources.
Polymers are a broad class of materials made up of long chain molecules of repeating units —anything from biopolymers such as plant cellulose to a variety of synthetic materials, used in everyday applications such as packaging or car tires, among countless other applications. The polymer length and structure play significant roles in how polymers are able to mix (or phase separate) with other polymers.
Connecting two polymers into a single, long chain molecule, also called a diblock copolymer, gives rise to the “self-assembly” of these polymers at the molecular level, meaning that the two polymer blocks spontaneously separate into different spatial regions or domains at the nanoscale, and these domains arrange in distinct ways that create patterns with cylindrical pores or other shapes.
As materials science has advanced, researchers are increasingly turning to such diblock copolymers to develop thin films and membranes to help separate complex chemical mixtures. These pores or channels can act as “gatekeepers,” as their dimensions can be tailored in such a way to only allow certain molecules to pass through.
Once science had a rough understanding of the process of fabricating polymer membranes, scientists and engineers started to implement it in certain industrial separation methods. Unfortunately, membrane fabrication (i.e., the process of forming a membrane) depends on a multitude of thermodynamic and kinetic parameters and is still not well understood at a fundamental level.
In recent years, researchers at the University of Göttingen have collaborated with the experimental group of Prof. Dr. Volker Abetz (Helmholtz-Zentrum Hereon and University of Hamburg) to advance our understanding of these processes at a fundamental level in the hopes of better understanding how process-directed self-assembly in the course of solvent evaporation from a polymer film or the exchange of solvent and nonsolvent can help tailor this nonequilibrium phase separation process.
“Our work in this area is motivated by the fact that solvent evaporation is a basic and ubiquitous real-world process,” said Prof. Dr. Marcus Müller, Professor of Theoretical Physics at the University of Göttingen. “Polymer solutions are frequently employed in the preparation of polymer films and membranes, yet this process is not well understood and simulation studies are in their infancy.” Müller and his collaborators have been using high-performance computing (HPC) resources at the Jülich Supercomputing Centre (JSC) to study the phase separation and self-assembly at the molecular level.
Pull yourself together
The fabrication of polymer membranes has made large advancements in recent decades, but is still an emergent field of study. While these self-assembled structures are very effective at filtering out desired molecules in a chemical mixture, they are also very fragile and prone to fouling—clogging of a polymer membrane’s pores that prevents it from filtering out the desired molecules.
Müller and a group of international collaborators are investigating a suite of potential methods to improve copolymer self-assembly and polymer membranes used in filtration or separation processes. Perhaps the most exciting and promising among them lies in the realm of process-directed self-assembly.
“As a theoretical physicist, I am really fascinated by process-directed self-assembly, where the processing pathway—in our case, the evaporation or solvent exchange—is utilized to fabricate a functional nanostructure,” Müller said. “This kind of processing is a well-known strategy in engineering. Japanese swordsmiths all the way back in feudal Japan knew that micrometre-sized grains in metals dictate their mechanical application properties, for instance, but processing on the molecular time and length scales is much less explored and understood.”
Experiments have been the driving force in finding new ways to use polymers in this research field, but rely heavily on prior knowledge followed by trial and error. To accelerate development of new fabrication processes, researchers like Müller and his collaborators pair experiment with simulation.
Experiments serve as the basis for simulations, as researchers want to recreate the experimental conditions in their computational models as closely as possible. Unlike experiments, though, simulations allow researchers to not only model the process of structure formation in time and space, but also to access conditions that are difficult to create in experiments in order to highlight the role of specific interactions or process parameters. This strategy provides insights and suggests how modifications impact the structure of a polymer membrane and its ability to effectively filter a mixture.
However, in order to accurately model a given system, simulations must be large enough to capture real-world conditions while detailed enough to accurately represent molecules’ interactions with one another. Further, researchers need simulations to follow the large spectrum of time scales involved in the membrane formation and usage. That means that researchers like Müller require access to HPC resources, such as JSC’s modular supercomputing system JUWELS.
Using a combination of theory and experiment, Müller and his collaborators have already gained new insights into how the relationship between polymer structure, thermodynamics, and process conditions influences membrane fabrication, but stressed that this field is still emergent and more experimental and computational work remains.
Next-generation technologies advance next-generation industrial processes
Müller indicated that advancements in both experimental techniques as well as computational power have already played a big role in furthering researchers’ understanding of polymer membranes. “These are exciting times, because by virtue of new techniques and resources like the JUWELS Booster module at JSC, the time and length scales of simulation and experiment are starting to truly converge,” he said. “One can anticipate many opportunities of fruitful collaboration in the context of nonequilibrium structure formation of polymer materials.”
Cutting-edge computational resources such as the JUWELS Booster—currently the fastest machine in Europe and among the top 10 most energy-efficient machines in the world—can only truly demonstrate their computational muscle if scientists and engineers can make effective use of them, though, and Müller indicated that JSC’s emphasis on training has helped improve his team’s performance significantly.
Specifically, he pointed to JSC’s GPU Hackathons as being beneficial for not only improving performance, but also generally sharing best practices across JSC’s user base. GPUs, a relatively recent addition to the many supercomputing ecosystems, were originally designed to quickly render graphics in high-end computer gaming, but have become powerful accelerators for simulations programmed to use them in concert with traditional CPUs. Müller indicated that by participating in the GPU Hackathons, the team was able to port and scale its code to run effectively across multiple GPUs on the JUWELS Booster module.
“Hackathons allow my group members and experts from JSC like Dr. Andreas Herten as well as NVIDIA experts like Markus Hrywniak to team up. This collaboration started for us with an event in 2016, and continued through the JUWELS Booster early-access program and Hackathon at the end of last year,” he said. “This personal contact to HPC experts at JSC and NVIDIA is essential. It allows us to acquire top-notch technical knowledge and is also a great team-building activity that brings together new and experienced group members – my group members dedicate a lot of work to these exciting events, but it also provides a lot of excitement and motivation and is also a lot of fun.”
Moving forward, the team plans to leverage its increased knowledge of using the JUWELS Booster to include even more processes in their simulations in the hopes of not just validating experimental hypotheses, but suggesting new potential materials. Müller indicated that by using JUWELS’ modular computing concept to draw from its ultra-fast GPUs and robust CPUs, the team could combine different modelling approaches to further accelerate the iteration between computational scientists and experimentalists.
Funding for JUWELS was provided by the Ministry of Culture and Research of the State of North Rhine-Westphalia and the German Federal Ministry of Education and Research through the Gauss Centre for Supercomputing (GCS).