Research

Foundation models have revolutionized natural language processing through a ``train once, deploy anywhere'' paradigm, where a single pre-trained model adapts to countless downstream tasks without retraining. Access to a Physics Foundation Model (PFM) would be transformative -- democratizing access to high-fidelity simulations, accelerating scientific discovery, and eliminating the need for specialized solver development. Yet current physics-aware machine learning approaches remain fundamentally limited to single, narrow domains and require retraining for each new system. We present the General Physics Transformer (GPhyT), trained on 1.8 TB of diverse simulation data, that demonstrates foundation model capabilities are achievable for physics. Our key insight is that transformers can learn to infer governing dynamics from context, enabling a single model to simulate fluid-solid interactions, shock waves, thermal convection, and multi-phase dynamics without being told the underlying equations. GPhyT achieves three critical breakthroughs: (1) superior performance across multiple physics domains, outperforming specialized architectures by up to 29x, (2) zero-shot generalization to entirely unseen physical systems through in-context learning, and (3) stable long-term predictions through 50-timestep rollouts. By establishing that a single model can learn generalizable physical principles from data alone, this work opens the path toward a universal PFM that could transform computational science and engineering.

The use of gas diffusion electrodes (GDEs) enables efficient electrochemical CO2 reduction and may be a viable technology in CO2 utilization after carbon capture. Understanding the spatio-temporal phenomena at the triple-phase boundary formed inside GDEs remains a challenge; yet it is critical to design and optimize industrial electrodes for gas-fed electrolyzers. Thus far, transport and reaction phenomena are not yet fully understood at the microscale, among other factors, due to a lack of experimental analysis methods for porous electrodes under operating conditions. In this work, a realistic microfluidic GDE surrogate is presented. Combined with fluorescence lifetime imaging microscopy (FLIM), the methodology allows monitoring of wetting and local pH, representing the dynamic (in)stability of the triple phase boundary in operando. Upon charging the electrode, immediate wetting leads to an initial flooding of the catalyst layer, followed by spatially oscillating pH changes. The micromodel presented gives an experimental insight into transport phenomena within porous electrodes, which is so far difficult to achieve. The methodology and proof of the spatio-temporal pH and wetting oscillations open new opportunities to further comprehend the relationship between gas diffusion electrode properties and electrical currents originating at a given surface potential.

Flow-electrode capacitive deionization (FCDI) is an electrically driven water desalination technology promising for many applications, such as industrial wastewater treatment. FCDI exploits the pumpability of carbon slurries, enabling a continuous process suitable for a wide range of feed salinities. Previously, we demonstrated the applicability of FCDI processes with incorporated ion-exchange membranes for the desalination and concentration of saline brines containing 60–120 g/L NaCl. At such elevated salinities, steep concentration gradients occur across the membranes of an FCDI cell. Hence, the characteristics of the membranes become crucial for the overall process performance. It is not yet fully understood, which physical phenomena dominantly influence the ion transport. In this article, we present the first FCDI process model focusing on brine treatment. While our previously published FCDI model (Rommerskirchen et al., 2018) was suitable for simulating the treatment of low salinity solutions, the model at hand includes more non-idealities and focuses on the ion transport through the ion-exchange membranes. We introduce a constriction factor (sigma-factor) for the diffusion coefficients to model the electrical double layer behavior within the membranes at steep concentration gradients. Hence, the model is now also suitable for the simulation of continuous FCDI processes for brine treatment and shows good correlation with experimental results.