Social gathering & Poster pitches

There is a pressing need to develop new rechargeable battery technologies that offer higher energy density, faster charging, longer lifetimes, and lower costs, particularly in the context of the global transition toward renewable energy and electrified transportation. Lithium-ion batteries have already transformed portable electronics by enabling safe, long-lasting, and rechargeable operation, and they are expected to play a key role in large-scale energy storage and mobility applications. Meeting these growing demands requires interdisciplinary efforts to discover and understand new materials, especially cathode materials, whose performance is governed by complex electronic structures. Computational methods, particularly electronic structure simulations, have become essential tools for studying battery components and for predicting key properties such as ground-state energies that determine equilibrium cell voltage, ionic mobility, and thermal stability. These calculations are also critical for understanding degradation mechanisms, including solid electrolyte interphase growth and phase instability. However, accurately modeling charging and discharging processes remains highly challenging, as it involves exploring a vast configurational space with up to 10^9-10^15 possible ionic arrangements even in moderately sized systems [1]. Classical approaches, especially density functional theory (DFT), are widely used but face intrinsic limitations: they often fail to capture strong electron correlations in transition metal oxides and become computationally prohibitive when high accuracy or extensive sampling is needed, while more accurate post-Hartree–Fock methods are not tractable for realistic periodic materials. Consequently, there is a pressing need for quantum-assisted approaches that can handle the exponential scaling of materials science problems. This project addresses the simulation of charging/discharging characteristics in cathode materials, shifting the focus from simple electrolyte molecules to the complex, periodic structures of solid-state battery materials.

Quantum kernel methods offer compelling possibilities for enhanced machine learning [1,2]; however, the extent to which the specific noise characteristics of near-term quantum computers impact their performance remains an open question. To accurately assess the feasibility of near-term quantum machine learning applications, it is fundamental to characterize the behavior of quantum kernels under practical hardware conditions.
In the noisy intermediate-scale quantum (NISQ) regime, computations are unavoidably affected by hardware imperfections and measurement errors, which distort the quantum kernel's feature space and directly impact the performance of quantum support vector machines (qSVMs). Since quantum kernels encode pairwise similarities between data points, these distortions can compromise the reliability of the learning process.
Understanding whether these noise-induced deviations translate into reduced classification performance is therefore a key open problem. Addressing this challenge is fundamental to accurately assess the performance and applicability of quantum kernel methods in practical scenarios. 
In this work, we investigate this issue by applying realistic hardware noise models to qSVMs, evaluating their performance relative to ideal, noiseless simulations. We assess diverse quantum feature maps, contrasting ideal baselines against realistic noise models on simulated backends. Furthermore, we systematically analyze both quantum kernel bandwidth scaling [3] and M3 [4] measurement error mitigation, studying their individual and combined effects, and revealing how hyperparameter tuning and error mitigation jointly influence performance in noisy quantum environments.

Power modules built on silicon carbide and gallium nitride run at junction temperatures above 200 °C, which rules out conventional solder in the die-attach layer. Sintered silver has taken its place, because a joint melting at 962 °C sits at a low homologous temperature, and the porous network left behind by sintering absorbs the thermal-expansion mismatch between die and substrate [1, 2]. The as-sintered network is however metastable, and sustained service temperature drives pore coalescence and grain-boundary migration that thin the load-bearing skeleton and degrade the thermal and mechanical response [3]. Qualifying a module for a field life of years therefore means predicting how the microstructure moves, not the final equilibrium state. The homogenised stiffness and conductivity that finite-element lifetime models consume are properties of the intermediate states.
Mesoscale kinetic models supply that trajectory today at a characteristic cost. Monte Carlo Potts models and hybrid Potts phase-field models advance the microstructure through long sequences of accepted local updates [4, 5], each acting on the configuration the previous one produced, so the length of a run is set by the number of update events rather than by the number of microstructures the study needs to resolve.
 Ising machines invert that structure by resolving an entire configuration in a single hardware pass [6, 7], which makes a quadratic unconstrained binary optimization (QUBO) encoding of the microstructure an obvious target. One obstacle has kept materials applications to static problems [8]. A global minimiser returns the equilibrium configuration, whereas coarsening is a chain of local rearrangements. The concrete problem we solve is how to make annealing hardware emit an ordered microstructural trajectory instead of a single relaxed endpoint.

The two-dimensional cutting stock problem deals with the task of cutting out required rectangular pieces of certain fixed sizes and quantities out of some base material. In principal one can optimize the process in many different ways. The simplest objective is to minimize the waste area, but one can also have different arbitrary values of the ready pieces. Sometimes it is important to cut out pieces in as little cuts as possible and sometimes the rest pieces can be stored for later use and have to be cut down as little as possible. In each cutting stage the direction of the cuts switches between horizontal and vertical cuts. We consider here an arbitrary number of cutting stages but have a restriction in the possible width or length of the cuts. Work on the two-dimensional cutting stock problem is relevant for multiple companies that cut sheets of material into smaller rectangles. This task is quite common for plywood, paper, metal plate or glass sheet processing.

The design of a transport aircraft requires a large number of load computations for an optimal structural sizing. Different flight points, maneuver and gust load cases, as well as different mass and failure cases, and control laws need to be taken into account, easily summing up to hundreds of thousands of simulations per design cycle. Current simulations are therefore based on unsteady linearized aerodynamic methods [1,2] as these are computationally efficient. However, when it comes to modern aircraft design, more accurate methods are necessary in order to exploit the full design space and optimize the aircraft's weight. Especially when it comes to large-amplitude excitations as they need to be computed for gust encounters, the currently used time-linearized methods naturally have their limitations. Time-linearized predictions might result, e.g., in an overprediction of actually occurring loads [3] due to an insufficient modeling of, e.g., unsteady flow separation [4]. Therefore, unsteady nonlinear aerodynamic methods, e.g., based on the Unsteady Reynolds-Averaged Navier-Stokes (URANS) equations, are increasingly applied for gust load computations. One of their biggest drawback, however, is the enormous increase in computational time when compared to a RANS-based time-linearized formulation such as, e.g., a time-linearized frequency-somain solver [5].

This work addresses this specific bottleneck by exploiting the advantages of a reservoir computing-based (RC) surrogate approach for the time-domain prediction of URANS-based gust responses: In general, RC [6,7] leverages a fixed nonlinear dynamical system (the reservoir) to project inputs into a high-dimensional space such that complex nonlinear relationships become easier to represent and extract. Unlike deep neural networks, RC requires training only of the output layer, typically via simple linear regression, mapping the high-dimensional space to the training targets. The training is therefore computationally very efficient and stable. The reservoirs intrinsic dynamics naturally retain memory of past inputs and capture temporal dependencies, making RC particularly suited for modeling dynamical systems and time series data.

Using quantum reservoir computing (QRC) [8] might even enhance the strengths of classical RC: First, quantum systems naturally evolve in high-dimensional Hilbert spaces, so a relatively small number of qubits can generate a rich set of nonlinear transformations, potentially boosting the expressive power of the reservoir without increasing system size in the classical sense. Secondly, quantum dynamics inherently include features such as superposition, entanglement, and interference. These properties may provide richer temporal and nonlinear processing capabilities than classical reservoirs, potentially improving performance on complex physical systems such as the one described by the current use case. 
This leads to the idea of QRC as a data-driven surrogate to complement costly URANS computations, thus reducing the total computation time while balancing the approximation error: Use the high-fidelity model only in a subset of the parameter space, and the cheaper surrogate for the rest. This is explored exemplarily for the problem of computing gust responses of a two-dimensional airfoil in terms of two global coefficients across different gust configurations.

Recent works on quantum lattice Boltzmann methods (QLBM) have begun to move beyond model constructions [1, 2] toward optimization-oriented scientific-computing applications [3]. In particular, the emerging perspective of quantum search based on QLBM, treats multiple lattice configurations as candidate solutions, and combines coherent time evolution with amplitude estimation and minimum finding [3]. This direction is promising for future quantum computational fluid dynamics, but it also highlights a practical bottleneck: once the algorithmic core becomes deeper and more coherent, the front-end cost of quantum state preparation (QSP) becomes increasingly important. 
The current QLBM software frameworks in Refs. [1-3] have already separated InitialConditions, algorithm, postprocessing, and measurement into distinct components, which makes the initial-state block a natural insertion point for a dedicated state-preparation compiler. At the same time, current QLBM software frameworks focus on algorithm-specific initialization routines, rather than a general-purpose method for compiling arbitrary non-uniform initial states.
However, existing QSP methods each address this challenge only partially. Exact synthesis routines such as Qiskit's initialization method [4] are general-purpose but tend to produce deep circuits. The genetic algorithm for state preparation (GASP) proposed by Creevey et al. (2023) [5], improves resource usage through evolutionary search over circuit structures and angle optimization, but it remains essentially fidelity-driven and does not explicitly optimize the trade-off between circuit depth and total gate count while achieving a high-fidelity solution.

Accurate climate prediction requires the numerical simulation of nonlinear, multiscale, and often chaotic dynamical systems over long time periods. Despite major progress in climate modelling, substantial uncertainty persists in key quantities such as climate sensitivity, extremes, and long-time variability [1–6]. High-resolution simulations and larger ensembles could reduce biases and improve uncertainty quantification, but they remain computationally too expensive over climate timescales [7,8].
It is therefore important to investigate how future quantum-computing methods could accelerate nonlinear differential-equation solvers for climate-relevant models. Quantum linear-systems algorithms (QLSAs) [9,10] can in principle accelerate certain high-dimensional linear problems, but they do not directly solve the nonlinear equations that arise in climate dynamics [6,11]. We therefore examine which reformulation provides the more promising preprocessing step for a QLSA workflow. Specifically, we compare trajectory-based Carleman linearization [12–15] with ensemble-based Liouville and Fokker–Planck reformulations [16–18]. Our hypothesis is that, for strongly nonlinear and chaotic dynamics, ensemble reformulations may be more informative than single-trajectory reformulations because they directly target uncertainty quantification and long-time statistical observables that are central to climate science.