One of the main challenges in optimal scaling of large language models (LLMs) is the prohibitive cost of hyperparameter tuning, particularly learning rate $\eta$ and batch size $B$. While techniques like $\mu$P (Yang et al., 2022) provide scaling rules for optimal $\eta$ transfer in the infinite model size limit, the optimal scaling behavior in the infinite data size limit remains unknown. We fill in this gap by observing for the first time an intricate dependence of optimal $\eta$ scaling on the pretraining token budget $T$, $B$ and its relation to the critical batch size $B_\mathrm{crit}$, which we measure to evolve as $B_\mathrm{crit} \propto T$. Furthermore, we show that the optimal batch size is positively correlated with $B_\mathrm{crit}$: keeping it fixed becomes suboptimal over time even if learning rate is scaled optimally. Surprisingly, our results demonstrate that the observed optimal $\eta$ and $B$ dynamics are preserved with $\mu$P model scaling, challenging the conventional view of $B_\mathrm{crit}$ dependence solely on loss value. Complementing optimality, we examine the sensitivity of loss to changes in learning rate, where we find the sensitivity to decrease with increase of $T$ and to remain constant with $\mu$P model scaling. We hope our results make the first step towards a unified picture of the joint optimal data and model scaling.

To ensure resilience against the unavoidable noise in quantum computers, quantum information needs to be encoded using an error-correcting code, and circuits must have a particular structure to be fault-tolerant. Compilation of fault-tolerant quantum circuits is thus inherently different from the non-fault-tolerant case. However, automated fault-tolerant compilation methods are widely underexplored, and most known constructions are obtained manually for specific codes only. In this work, we focus on the problem of automatically synthesizing fault-tolerant circuits for the deterministic initialization of an encoded state for a broad class of quantum codes that are realizable on current and near-term hardware. To this end, we utilize methods based on techniques from classical circuit design, such as satisfiability solving, resulting in tools for the synthesis of (optimal) fault-tolerant state preparation circuits for near-term quantum codes. We demonstrate the correct fault-tolerant behavior of the synthesized circuits using circuit-level noise simulations. We provide all routines as open-source software as part of the Munich Quantum Toolkit (MQT) at this https URL.

Recurrent neural networks (RNNs) are valued for their computational efficiency and reduced memory requirements on tasks involving long sequence lengths but require high memory-processor bandwidth to train. Checkpointing techniques can reduce the memory requirements by only storing a subset of intermediate states, the checkpoints, but are still rarely used due to the computational overhead of the additional recomputation phase. This work addresses these challenges by introducing memory-efficient gradient checkpointing strategies tailored for the general class of sparse RNNs and Spiking Neural Networks (SNNs). SNNs are energy efficient alternatives to RNNs thanks to their local, event-driven operation and potential neuromorphic implementation. We use the Intelligence Processing Unit (IPU) as an exemplary platform for architectures with distributed local memory. We exploit its suitability for sparse and irregular workloads to scale SNN training on long sequence lengths. We find that Double Checkpointing emerges as the most effective method, optimizing the use of local memory resources while minimizing recomputation overhead. This approach reduces dependency on slower large-scale memory access, enabling training on sequences over 10 times longer or 4 times larger networks than previously feasible, with only marginal time overhead. The presented techniques demonstrate significant potential to enhance scalability and efficiency in training sparse and recurrent networks across diverse hardware platforms, and highlights the benefits of sparse activations for scalable recurrent neural network training.

Graph partitioning has many applications in powersystems from decentralized state estimation to parallel simulation. Focusing on parallel simulation, optimal grid partitioning minimizes the idle time caused by different simulation times for the sub-networks and their components and reduces the overhead required to simulate the cuts. Partitioning a graph into two parts such that, for example, the cut is minimal and the subgraphs have equal size is an NP-hard problem. In this paper we show how optimal partitioning of a graph can be obtained using quantum annealing (QA). We show how to map the requirements for optimal splitting to a quadratic unconstrained binary optimization (QUBO) formulation and test the proposed formulation using a current D-Wave QPU. We show that the necessity to find an embedding of the QUBO on current D-Wave QPUs limits the problem size to under 200 buses and notably affects the time-to-solution. We finally discuss the implications on near-term implementation of QA in combination to traditional CPU or GPU based simulation.

Understanding the physical nature of the D-Wave annealers remains a subject of active investigation. In this study, we analyze the sampling behavior of these systems and explore whether their results can be replicated using quantum and Markovian models. Employing the standard and the fast annealing protocols, we observe that the D-Wave annealers sample states with frequencies matching the Gibbs distribution for sufficiently long annealing times. Using Bloch equation simulations for single-qubit problems and Lindblad and Markovian master equations for two-qubit systems, we compare experimental data with theoretical predictions. Our results provide insights into the role of quantum mechanics in these devices.

One-dimensional (1D) topological superconductivity is a state of matter that is not found in nature. However, it can be realised, for example, by inducing superconductivity into the quantum spin Hall edge state of a two-dimensional topological insulator. Because topological superconductors are proposed to host Majorana zero modes, they have been suggested as a platform for topological quantum computing. Yet, conclusive proof of 1D topological superconductivity has remained elusive. Here, we employ low-temperature scanning tunnelling microscopy to show 1D topological superconductivity in a van der Waals heterostructure by directly probing its superconducting properties, instead of relying on the observation of Majorana zero modes at its boundary. We realise this by placing the two-dimensional topological insulator monolayer WTe$_2$ on the superconductor NbSe$_2$. We find that the superconducting topological edge state is robust against magnetic fields, a hallmark of its triplet pairing. Its topological protection is underpinned by a lateral self-proximity effect, which is resilient against disorder in the monolayer edge. By creating this exotic state in a van der Waals heterostructure, we provide an adaptable platform for the future realization of Majorana bound states. Finally, our results more generally demonstrate the power of Abrikosov vortices as effective experimental probes for superconductivity in nanostructures.

In the past decades many density-functional theory methods and codes adopting periodic boundary conditions have been developed and are now extensively used in condensed matter physics and materials science research. Only in 2016, however, their precision (i.e., to which extent properties computed with different codes agree among each other) was systematically assessed on elemental crystals: a first crucial step to evaluate the reliability of such computations. We discuss here general recommendations for verification studies aiming at further testing precision and transferability of density-functional-theory computational approaches and codes. We illustrate such recommendations using a greatly expanded protocol covering the whole periodic table from Z=1 to 96 and characterizing 10 prototypical cubic compounds for each element: 4 unaries and 6 oxides, spanning a wide range of coordination numbers and oxidation states. The primary outcome is a reference dataset of 960 equations of state cross-checked between two all-electron codes, then used to verify and improve nine pseudopotential-based approaches. Such effort is facilitated by deploying AiiDA common workflows that perform automatic input parameter selection, provide identical input/output interfaces across codes, and ensure full reproducibility. Finally, we discuss the extent to which the current results for total energies can be reused for different goals (e.g., obtaining formation energies).

The electric dipole form factors and moments of the ground state baryons are calculated in chiral perturbation theory at next-to-leading order. We show that the baryon electric dipole form factors at this order depend only on two combinations of low-energy constants. We also derive various relations that are free of unknown low-energy constants. We use recent lattice QCD data to calculate all baryon EDMs. In particular, we find d_n = -2.9\pm 0.9 and d_p = 1.1\pm 1.1 in units of 10^{-16} e \theta_0 cm. Finite volume corrections to the moments are also worked out. We show that for a precision extraction from lattice QCD data, the next-to-leading order terms have to be accounted for.

We currently observe a disconcerting phenomenon in machine learning studies in psychiatry: While we would expect larger samples to yield better results due to the availability of more data, larger machine learning studies consistently show much weaker performance than the numerous small-scale studies. Here, we systematically investigated this effect focusing on one of the most heavily studied questions in the field, namely the classification of patients suffering from major depressive disorder (MDD) and healthy control (HC) based on neuroimaging data. Drawing upon structural magnetic resonance imaging (MRI) data from a balanced sample of $N = 1,868$ MDD patients and HC from our recent international Predictive Analytics Competition (PAC), we first trained and tested a classification model on the full dataset which yielded an accuracy of $61\,\%$. Next, we mimicked the process by which researchers would draw samples of various sizes ($N = 4$ to $N = 150$) from the population and showed a strong risk of misestimation. Specifically, for small sample sizes ($N = 20$), we observe accuracies of up to $95\,\%$. For medium sample sizes ($N = 100$) accuracies up to $75\,\%$ were found. Importantly, further investigation showed that sufficiently large test sets effectively protect against performance misestimation whereas larger datasets per se do not. While these results question the validity of a substantial part of the current literature, we outline the relatively low-cost remedy of larger test sets, which is readily available in most cases.

The ever increasing availability of supercomputing resources led computer-based materials science into a new era of high-throughput calculations. Recently, Pizzi et al. [Comp. Mat. Sci. 111, 218 (2016)] introduced the AiiDA framework that provides a way to automate calculations while allowing to store the full provenance of complex workflows in a database. We present the development of the AiiDA-KKR plugin that allows to perform a large number of ab initio impurity embedding calculations based on the relativistic full-potential Korringa-Kohn-Rostoker Green function method. The capabilities of the AiiDA-KKR plugin are demonstrated with the calculation of several thousand impurities embedded into the prototypical topological insulator Sb2Te3. The results are collected in the JuDiT database which we use to investigate chemical trends as well as Fermi level and layer dependence of physical properties of impurities. This includes the study of spin moments, the impurity's tendency to form in-gap states or its effect on the charge doping of the host-crystal. These properties depend on the detailed electronic structure of the impurity embedded into the host crystal which highlights the need for ab initio calculations in order to get accurate predictions.

Ab initio calculations play an essential role in our fundamental understanding of quantum many-body systems across many subfields, from strongly correlated fermions to quantum chemistry and from atomic and molecular systems to nuclear physics. One of the primary challenges is to perform accurate calculations for systems where the interactions may be complicated and difficult for the chosen computational method to handle. Here we address the problem by introducing a new approach called wavefunction matching. Wavefunction matching transforms the interaction between particles so that the wavefunctions up to some finite range match that of an easily computable interaction. This allows for calculations of systems that would otherwise be impossible due to problems such as Monte Carlo sign cancellations. We apply the method to lattice Monte Carlo simulations of light nuclei, medium-mass nuclei, neutron matter, and nuclear matter. We use high-fidelity chiral effective field theory interactions and find good agreement with empirical data. These results are accompanied by new insights on the nuclear interactions that may help to resolve long-standing challenges in accurately reproducing nuclear binding energies, charge radii, and nuclear matter saturation in ab initio calculations.

Increasing HPC cluster sizes and large-scale simulations that produce petabytes of data per run, create massive IO and storage challenges for analysis. Deep learning-based techniques, in particular, make use of these amounts of domain data to extract patterns that help build scientific understanding. Here, we demonstrate a streaming workflow in which simulation data is streamed directly to a machine-learning (ML) framework, circumventing the file system bottleneck. Data is transformed in transit, asynchronously to the simulation and the training of the model. With the presented workflow, data operations can be performed in common and easy-to-use programming languages, freeing the application user from adapting the application output routines. As a proof-of-concept we consider a GPU accelerated particle-in-cell (PIConGPU) simulation of the Kelvin- Helmholtz instability (KHI). We employ experience replay to avoid catastrophic forgetting in learning from this non-steady process in a continual manner. We detail challenges addressed while porting and scaling to Frontier exascale system.

Mapping quantum approximate optimization algorithm (QAOA) circuits with non-trivial connectivity in fixed-layout quantum platforms such as superconducting-based quantum processing units (QPUs) requires a process of transpilation to match the quantum circuit on the given layout. This step is critical for reducing error rates when running on noisy QPUs. Two methodologies that improve the resource required to do such transpilation are the SWAP network and parity twine chains (PTC). These approaches reduce the two-qubit gate count and depth needed to represent fully connected circuits. In this work, a simulated annealing-based method is introduced that reduces the PTC and SWAP network encoding requirements in QAOA circuits with non-fully connected two-qubit gates. This method is benchmarked against various transpilers and demonstrates that, beyond specific connectivity thresholds, it achieves significant reductions in both two-qubit gate count and circuit depth, surpassing the performance of Qiskit transpiler at its highest optimization level. For example, for a 120-qubit QAOA instance with 25% connectivity, our method achieves an 85% reduction in depth and a 28% reduction in two-qubit gates. Finally, the practical impact of PTC encoding is validated by benchmarking QAOA on the ibm_fez device, showing improved performance up to 20 qubits, compared to a 15-qubit limit when using SWAP networks.

By introducing an additional operator into the action and using the Feynman-Hellmann theorem we describe a method to determine both the quark line connected and disconnected terms of matrix elements. As an illustration of the method we calculate the gluon contribution (chromo-electric and chromo-magnetic components) to the nucleon mass.

Electrons which are slowly moving through chiral magnetic textures can effectively be described as if they where influenced by electromagnetic fields emerging from the real-space topology. This adiabatic viewpoint has been very successful in predicting physical properties of chiral magnets. Here, based on a rigorous quantum-mechanical approach, we unravel the emergence of chiral and topological orbital magnetism in one- and two-dimensional spin systems. We uncover that the quantized orbital magnetism in the adiabatic limit can be understood as a Landau-Peierls response to the emergent magnetic field. Our central result is that the spin-orbit interaction in interfacial skyrmions and domain walls can be used to tune the orbital magnetism over orders of magnitude by merging the real-space topology with the topology in reciprocal space. Our findings point out the route to experimental engineering of orbital properties of chiral spin systems, thereby paving the way to the field of chiral orbitronics.

We report on the implementation of the Bogoliubov-de Gennes method into the JuKKR code [https: //jukkr.fz-juelich.de], an implementation of the relativistic all-electron, full-potential Korringa-Kohn-Rostoker Green function method, which allows a material-specific description of inhomogeneous super-conductors and heterostructures on the basis of density functional theory. We describe the formalism and report on calculations for the s-wave superconductor Nb, a potential component of the materials platform enabling the realization of the Majorana zero modes in the field of topological quantum computing. We compare the properties of the superconducting state both in the bulk and for (110) surfaces. We compare slab calculations for different thicknesses and comment on the importance of spin-orbit coupling, the effect of surface relaxations and the influence of a softening of phonon modes on the surface for the resulting superconducting gap.

Interfacing a topological insulator (TI) with an $s$-wave superconductor (SC) is a promising material platform that offers the possibility to realize a topological superconductor through which Majorana-based topologically protected qubits can be engineered. In our computational study of the prototypical SC/TI interface between Nb and Bi$_2$Te$_3$, we identify the benefits and possible bottlenecks of this potential Majorana material platform. Bringing Nb in contact with the TI film induces charge doping from the SC to the TI, which shifts the Fermi level into the TI conduction band. For thick TI films, this results in band bending leading to the population of trivial TI quantum-well states at the interface. In the superconducting state, we uncover that the topological surface state experiences a sizable superconducting gap-opening at the SC/TI interface, which is furthermore robust against fluctuations of the Fermi energy. We also show that the trivial interface state is only marginally proximitized, potentially obstructing the realization of Majorana-based qubits in this material platform.

Combinatorial optimization problems are one of the target applications of current quantum technology, mainly because of their industrial relevance, the difficulty of solving large instances of them classically, and their equivalence to Ising Hamiltonians using the quadratic unconstrained binary optimization (QUBO) formulation. Many of these applications have inequality constraints, usually encoded as penalization terms in the QUBO formulation using additional variables known as slack variables. The slack variables have two disadvantages: (i) these variables extend the search space of optimal and suboptimal solutions, and (ii) the variables add extra qubits and connections to the quantum algorithm. Recently, a new method known as unbalanced penalization has been presented to avoid using slack variables. This method offers a trade-off between additional slack variables to ensure that the optimal solution is given by the ground state of the Ising Hamiltonian, and using an unbalanced heuristic function to penalize the region where the inequality constraint is violated with the only certainty that the optimal solution will be in the vicinity of the ground state. This work tests the unbalanced penalization method using real quantum hardware on D-Wave Advantage for the traveling salesman problem (TSP). The results show that the unbalanced penalization method outperforms the solutions found using slack variables and sets a new record for the largest TSP solved with quantum technology.

Resource allocation of wide-area internet networks is inherently a combinatorial optimization problem that if solved quickly, could provide near real-time adaptive control of internet-protocol traffic ensuring increased network efficacy and robustness, while minimizing energy requirements coming from power-hungry transceivers. In recent works we demonstrated how such a problem could be cast as a quadratic unconstrained binary optimization (QUBO) problem that can be embedded onto the D-Wave AdvantageTM quantum annealer system, demonstrating proof of principle. Our initial studies left open the possibility for improvement of D-Wave solutions via judicious choices of system run parameters. Here we report on our investigations for optimizing these system parameters, and how we incorporate machine learning (ML) techniques to further improve on the quality of solutions. In particular, we use the Hamming distance to investigate correlations between various system-run parameters and solution vectors. We then apply a decision tree neural network (NN) to learn these correlations, with the goal of using the neural network to provide further guesses to solution vectors. We successfully implement this NN in a simple integer linear programming (ILP) example, demonstrating how the NN can fully map out the solution space that was not captured by D-Wave. We find, however, for the 3-node network problem the NN is not able to enhance the quality of space of solutions.