Vorträge

The atmosphere, ocean, and sea ice components in Earth system models are coupled at the sea surface via boundary conditions. In essence, this amounts to coupled heat equations with discontinuous material parameters. However, the problem is special in two ways: First of all, the boundary conditions used in practice, so-called bulk interface conditions, allow for a temperature jump across the interface. Secondly, sea ice acts as a partially isolating layer and affects the boundary conditions seen by the atmosphere and ocean.
Theoretical analysis of this problem is missing, even with simplified models. For this reason, we propose a coupled toy model that describes the heat exchange of a partially ice-covered ocean with the atmosphere. This allows us to analytically derive convergence factors of the corresponding coupling iteration. We compare this to a numerical implementation of the same model using an open-source coupling software for climate applications, ClimaCoupler.jl. Our results show that the convergence behavior with bulk interface conditions is fundamentally different from using standard Dirichlet-Neumann or Robin-Robin interface conditions for conjugate heat transfer.

Organic redox polymers such as poly (TEMPO-acrylamide) (PTAM) are being investigated as cathode materials for post-lithium batteries, yet their electrochemical behavior is still not well characterized. For PTAM–Zn cells, internal states and dynamic processes remain unclear. To address this, we adapted the single-particle model with electrolyte (SPMe), originally developed for lithium-ion batteries, by re-deriving the governing equations for this chemistry and parametrizing the framework with experimental data. The model describes solid-state and electrolyte diffusion together with electrolyte potential, overpotential and voltage dynamics and is solved using finite-element discretization in space and Backward Euler time stepping with Newton iteration. Simulations quantitatively reproduce charging curves up to 50 C, capturing both capacity and voltage profiles. These results suggest that the adapted SPMe captures the dominant transport and kinetic behavior while highlighting the need for model refinement, targeted material studies, and improved validation through closer integration with experiments

Zoomlink:
https://tuhh.zoom.us/j/81920578609?pwd=TjBmYldRdXVDT1VkamZmc1BOajREZz09

This talk provides an overview of my dissertation, which analyzes flexible, high-performance algorithms for fundamental tensor operations used in tensor methods such as Alternating Least Squares (ALS) and the Higher-Order Singular Value Decomposition (HOSVD).

The first part of the talk presents layout-oblivious algorithms for elementwise tensor operations that perform equally well on tensors and subtensors, regardless of their memory layout. Next, we discuss the concept of multi-dimensional iterators that enable the design of generic algorithms without relying on pointer arithmetic. The final part focuses on efficient, layout-oblivious algorithms for tensor-vector and tensor-matrix multiplication. Various tensor slicing techniques and parallelization strategies will be presented. Their multi-core CPU performance is evaluated on a wide range of tensors with varying shapes and orders.

We demonstrate that the performance of the fastest algorithms remains independent of tensor layout and that the presented methods are competitive with, and in some cases outperform, state-of-the-art libraries such as LibTorch and Eigen.

Zoomlink:
https://tuhh.zoom.us/j/81920578609?pwd=TjBmYldRdXVDT1VkamZmc1BOajREZz09

Model Predictive Control (MPC) is based on receding-horizon solution of optimal control problems and it is among the most successful advanced control methods. Core reasons are its applicability to nonlinear systems with constraints as well as the variety of tailored numerical algorithms and powerful software tools enabling
 efficient real-time implementations [1]. The application of MPC to cyber-physical systems (or to multi-
agent systems) is of pivotal interest in many application domains such as energy systems, logistics and
 transport, and robotics [2]. In this talk we present recent results on collaborative distributed nonlinear MPC 
for cyber-physical systems. We discuss a family of algorithms which is based on the decomposition 
of primal-dual Newton steps arising from Sequential Quadratic Programming (SQP) [3]. We explore 
how the underlying partially separable problem structure translates into partially separable Newton
 steps which can then be computed in decentralized fashion, i.e., based only on neighbor-to-neighbor
 communication. Moreover, we show that this numerical framework for decentralized real-time iterations in distributed NMPC
 allows for closed-loop stability guarantees [4] and for scalability [5]. Our findings are illustrated with several examples including multiple 
real-time implementations [6,7].

1] Gros, S., Zanon, M., Quirynen, R., Bemporad, A., & Diehl, M. (2020). From linear to nonlinear MPC:
bridging the gap via the real-time iteration. International Journal of Control, 93(1), 62-80.
[2] Christofides, P. D., Scattolini, R., de la Pena, D. M., & Liu, J. (2013). Distributed model predictive 
control: A tutorial review and future research directions. Computers & Chemical Engineering, 51,
21-41.
[3] Stomberg, G., Engelmann, A., & Faulwasser, T. (2022). Decentralized non-convex optimization via
bi-level SQP and ADMM. In 61st Conference on Decision and Control (CDC), 273-278.
[4] Stomberg, G., Engelmann, A., Diehl, M., & Faulwasser, T. (2024). Decentralized real-time iterations
for distributed nonlinear model predictive control. arXiv preprint arXiv:2401.14898.
[5] Stomberg, G., Raetsch, M., Engelmann, A., & Faulwasser, T. (2025). Large problems are not necessarily hard: A case study on distributed NMPC paying off. European Control Conference
[6] Stomberg, G., Ebel, H., Faulwasser, T., & Eberhard, P. (2023). Cooperative distributed MPC via de-
centralized real-time optimization: Implementation results for robot formations. Control Engineering 
Practice, 138, 105579.
[7] Stomberg, G., Schwan, R., Grillo, A., Jones, C. N., & Faulwasser, T. (2025). Cooperative distributed model predictive control for embedded systems: Experiments with hovercraft formations. 2025 IEEE International Conference on Robotics and Automation

Zoomlink:
https://tuhh.zoom.us/j/81920578609?pwd=TjBmYldRdXVDT1VkamZmc1BOajREZz09

Axons are micrometer-thin cables that transmit signals across the brain. Their size affects how fast signals travel, making axon diameter a key determinant of brain function -- and, when altered, a potential marker of disease. In theory, MRI is sensitive to axon size through the physics of water diffusion, but this sensitivity has remained unproven in real-world settings for decades. In this talk, I'll present recent advances in validating MRI-based axon radius estimates using experimental MRI and high-resolution microscopy of more than 46 million axons across the human brain.

Zoomlink:
https://tuhh.zoom.us/j/87285771127?pwd=bjlWT3AyQncwajZQN0l3dVd1WXJmZz09