LATEST RESEARCH RESULTS

Integral-asymmetric, isoporous block-copolymer membranes [1], which combine a thin selective skin of regular, nanometre-sized pores with a macroporous support, are highly attractive for separation applications. Their fabrication by self-assembly and nonsolvent-induced phase separation (SNIPS) is challenging because it couples several processes: evaporation-induced self-assembly (EISA), solvent–nonsolvent exchange, nonsolvent-induced phase separation (NIPS), and glassy arrest. Using JUWELS, JUWELS Booster, and JEDI, we developed a digital twin of the entire fabrication process, which combines particle-based simulations with a continuum description to reduce the computational cost. With it, we elucidated how material parameters and processing…

Some enzymes work like molecular machines with moving parts, they must physically reshape themselves to carry out their chemistry. The carboxylic acid reductase NoCAR does exactly this: a flexible swinging arm shuttles a reaction intermediate between two catalytic sites as the enzyme cycles through open and closed shapes. Researchers led by Prof. Dr. Holger Gohlke used 150 supercomputer simulations on the JUWELS Booster at Forschungszentrum Jülich to map this choreography in atomic detail for the first time, revealing unexpected shortcuts in the conformational pathway and identifying how cofactor binding orchestrates each step. The same team also uncovered the binding secrets of a second enzyme, SgMT, opening the door to engineering both…

How does a fruit know when to ripen? The answer involves the simple gas ethylene and a receptor called ETR1 that detects it, but only when armed with a copper ion delivered by specialized proteins. Researchers led by Prof. Dr. Holger Gohlke used quantum mechanical simulations and protein docking calculations on the JUWELS Booster supercomputer at Forschungszentrum Jülich to reveal how copper travels through a molecular relay chain to reach ETR1, and how ethylene binding then reconfigures the copper coordination to switch the receptor off. The findings explain at atomic resolution why related compounds block this switch and open new avenues toward controlling fruit ripening and plant stress responses.

Some bacteria export large toxins through a molecular machine that spans their entire cell envelope, powered by three ATP-driven motors working in concert. How these motors coordinate their activity to drive transport had been a long-standing mystery. Researchers led by Prof. Dr. Holger Gohlke used supercomputer simulations on the JUWELS Booster at Forschungszentrum Jülich to simulate million-atom models of the hemolysin A secretion system in all four possible motor configurations. The results reveal that only specific combinations of motor states are stable when fuelled with ATP, providing the first atomistic picture of how this bacterial export machine operates and pointing toward new strategies to interfere with bacterial toxin…

How should an AI model’s training change when it is given more data? Researchers at Forschungszentrum Jülich used JUWELS Booster to investigate how training settings and neural network building blocks affect performance at scale. The BeSL (Better models with Scaling Laws) project established practical relationships between data budgets, learning rates, and batch sizes. It developed a stable way to train networks with learnable activation functions. The results offer guidance for making better use of supercomputing resources when developing AI models.

The Chemical Lagrangian Model of the Stratosphere (CLaMS) is a Lagrangian Chemistry-Transport model, that was developed in Jülich since 1997. The specialty of this model is the Lagrangian principle. The simulations are not performed on a space-fixed grid but on an adaptive grid that follows the individual air masses. By now, CLaMS has become an internationally recognized model, particularly because the Lagrangian approach makes it is possible to realistically simulate small scale features and large trace gas gradients. Here, we report on two highlights of the extensive outcomes of the project CLaMS-ESM in which the Lagrangian approach has helped to understand long-standing problems and discrepancies in atmospheric modelling, that is the…

Glycine receptors help nerve cells in the spinal cord and brainstem control movement and process sensations such as pain. Experimental images and computer simulations disagreed on how the receptor's natural messengers, glycine and taurine, sit in their docking pocket. Using highly accurate quantum-mechanical simulations on the JUWELS supercomputer at the Jülich Supercomputing Centre, researchers from Forschungszentrum Jülich and the University of Strasbourg acted as referee. The verdict: the quantum results agree with the faster classical simulations, not with the blurry experimental snapshots, showing that standard simulation tools can be trusted for designing new drugs against disorders such as hyperekplexia and chronic pain.

Proteins do not always adopt a single fixed shape. Some, known as intrinsically disordered proteins (IDPs), continuously change their conformation and can form liquid-like condensates inside cells. This project used high-performance computing to investigate how the sequence and flexibility of such proteins determine their behaviour. Simulations showed that commonly used coarse-grained models differ significantly in their ability to reproduce protein conformations. The work also established a theoretical way of representing heterogeneous protein chains with simpler polymer models and provided initial insights into the phase separation of plant RS-proteins.

Deep learning models power everything from image recognition to language translation, but their performance depends heavily on small internal components called activation functions. Researchers at the University of Freiburg have developed a new activation function called GoLU, the Gompertz Linear Unit, that consistently outperforms existing alternatives across a wide range of tasks. Using high performance computing clusters, the team conducted over 112,000 GPU hours of experiments, demonstrating that this deceptively simple mathematical function can make neural networks more accurate and robust.

Since the release of ChatGPT in December 2022, artificial intelligence (AI) has garnered unprecedented media attention. Mainstream media outlets, such as Tagesschau, now regularly report AI-related news, and public discourse on the subject has permeated schools and universities. This widespread attention is the result of significant developments in Deep Learning (DL), which has proven incredibly successful at solving previously unsolvable tasks and outperforming known solutions to established problems. Deep Neural Networks (DNNs) are very powerful, but designing their architectures and training pipelines for novel tasks has traditionally required extensive trial-and-error by human experts. Thus, DNNs have been hard to adopt in new fields.

Describing the behavior of quarks and gluons—fundamental particles of the Standard Model of particle physics occurring at length scales of less than a femtometer – is of prime importance in understanding the formation of hadrons and nuclei, such as the neutron, proton, or pion, and their interactions. A deeper understanding of their interactions in turn allows us to better predict the behavior of nuclear matter in extreme conditions, like those that occur in compact astrophysical environments like neutron stars or during type II supernovae, and allows us to disentangle and identify physics beyond the Standard Model (BSM) in experiments occurring, for example, at CERN

Quantum chemical density functional calculations have been carried out to determine the way actinide(IV) ions bind to calcium silicate hydrate, the essential mineral phase of hardened cement. We showed for U4+ and Th4+ that the preferred binding occurs via replacement of Ca ions in the mineral and not at the surface or between the mineral sheets. This result favorably explains measured interatomic distances. In addition, we find that the binding strength of the actinide(IV) ions can vary between the various actinide elements in dependence of the acidity of the surrounding solution. These results contribute to the understanding of the interaction of radioactive waste with construction materials in a final geological repository and are useful…

Many promising solar and battery materials are not static crystals: their atoms constantly move around with locally breaking symmetries. In this project, Prof. David Egger’s team used the JUWELS computing facility and machine-learning-accelerated simulations to follow these atomic motions and show how they shape light absorption, ion transport and more. Simulating materials accurately from first principles could be a very computationally demanding task. This is where machine-learning algorithms come into play. The work helps explain experimental measurements and points to faster ways of screening energy materials under realistic conditions.

The THESAN-ZOOM project follows the birth and growth of galaxies during the first roughly two billion years of cosmic history, the era when the very first galaxies formed and their light gradually transformed the gas filling the Universe. Using a new generation of computer models that track both matter and radiation, the team followed 14 representative galaxies from their initial formation to their later, more settled states, resolving the individual clouds in which stars are born. The simulations, carried out on the SuperMUC-NG supercomputer at the Leibniz Supercomputing Centre, help astronomers interpret the distant galaxies now being observed in detail by the James Webb Space Telescope.

Predicting how long a drug stays bound to its protein target — the drug residence time — is one of the key challenges in computational pharmacology. Standard molecular dynamics (MD) simulations are inherently serial in time and therefore cannot exploit modern exascale supercomputers to their full potential. This project implemented, optimized, and benchmarked the Metadynamics of Paths (MoP) algorithm — a highly parallelizable path-sampling method — within the widely-used GROMACS biomolecular simulation package.

How can cleaner engines, recyclable fuels, and better battery materials be developed without costly trial-and-error experiments? Researchers at the University of Duisburg-Essen used the HAWK supercomputer to recreate combustion and particle formation processes in unprecedented detail. Their simulations explored hydrogen combustion in engines, recyclable iron fuels, cleaner coal and ammonia combustion, and the formation of nanoparticles for advanced batteries. The project also helped prepare scientific software for the next generation of supercomputers, opening new opportunities for sustainable energy and industrial technologies.

What if the familiar rules of water chemistry break down inside the tiny pockets of a living cell? A team at the University of Duisburg-Essen used the Hawk supercomputer at HLRS Stuttgart to track single protons inside reverse micelles — water droplets just a few nanometres across that mimic the confined environments found in enzymes, fuel-cell membranes and catalysts. The simulations reveal that confinement changes the rules: a proton can stick to the surfactant walls of the droplet, something that does not happen in ordinary bulk water, opening competing pathways that may explain why nature so often does its chemistry in tight spaces.

Roughly one in three new medicines contains fluorine — yet science still cannot fully explain why swapping hydrogen for fluorine in a molecule makes it shun water so strongly. A team at the University of Duisburg-Essen, led by Dr. Ana Vila Verde, used the Hawk supercomputer at HLRS Stuttgart to simulate, atom by atom, how water rearranges itself around fluorinated and non-fluorinated cousins of common solvents. The first results overturn a long-standing assumption: simple gases such as methane are not, in fact, reliable miniature models for understanding fluorine’s water-repelling effect in tomorrow’s drugs and materials.

The matter we are made of are atomic nuclei, which are hold together by the strong force. While the fundamental theory underlying the strong interactions, Quantum Chromodynamics, is well studied at high energies, in the realm of nuclear physics we are dealing with protons and neutrons, that are build of quarks and gluons. However, at the energy scales pertinent to nuclear physics, we can not resolve this substructure. Thus, the proper first principles description of these strongly-interacting many-body systems is based on protons and neutrons, with their interactions consistently given in terms of chiral effective field theory, as initiated by Nobel laureate Steven Weinberg more than three decades ago.