arXiv subjects

Thomas Köhler

Publications and source records attributed to Thomas Köhler.

At least 19 recordsLinked to original sources

arXiv2026

Predicting critical temperature in quantum simulators for high-$T_c$ superconductivity: the matrix product state plus mean field approach

Quantum simulation based on ultra cold atomic lattice gases is one of the most promising platforms to investigate high-$T_c$ superconductivity beyond the limited capabilities of quantum many body numerics on classical computers. Yet, despite enormous progress since the field's inception, realizing a high-$T_c$ superconducting state still remains out of reach. The present work lays the groundwork to purpose the recently proposed, and already partly realized, mixed-dimensional (mixD) models, towards this end. These systems offer the proven capability to realize very high pairing energies while retaining appreciable mobility of pairs. We specifically investigate the potential of 2D mixD-models with anisotropic tunneling, using the matrix product state plus mean field theory (MPS+MF) for fermions, and show that these models may enter a high-$T_c$ superconducting phase. These simulations in turn are based on a comprehensive characterization of the 1D mixD-systems, which are the sub-units of which the 2D system is comprized. In this, we cover the range of currently experimentally relevant system sizes, and establish practical heuristics to determine when finite\hyp size effects preclude the use of a 1D mixD-system to build the 2D ones.

cond-mat.supr-con↗

arXiv2026

Confidence-Based Mesh Extraction from 3D Gaussians

Recently, 3D Gaussian Splatting (3DGS) greatly accelerated mesh extraction from posed images due to its explicit representation and fast software rasterization. While the addition of geometric losses and other priors has improved the accuracy of extracted surfaces, mesh extraction remains difficult in scenes with abundant view-dependent effects. To resolve the resulting ambiguities, prior works rely on multi-view techniques, iterative mesh extraction, or large pre-trained models, sacrificing the inherent efficiency of 3DGS. In this work, we present a simple and efficient alternative by introducing a self-supervised confidence framework to 3DGS: within this framework, learnable confidence values dynamically balance photometric and geometric supervision. Extending our confidence-driven formulation, we introduce losses which penalize per-primitive color and normal variance and demonstrate their benefits to surface extraction. Finally, we complement the above with an improved appearance model, by decoupling the individual terms of the D-SSIM loss. Our final approach delivers state-of-the-art results for unbounded meshes while remaining highly efficient.

cs.CV↗

arXiv2026

A LoD of Gaussians: Unified Training and Rendering for Ultra-Large Scale Reconstruction with External Memory

Gaussian Splatting has emerged as a high-performance technique for novel view synthesis, enabling real-time rendering and high-quality reconstruction of small scenes. However, scaling to larger environments has so far relied on partitioning the scene into chunks -- a strategy that introduces artifacts at chunk boundaries, complicates training across varying scales, and is poorly suited to unstructured scenarios such as city-scale flyovers combined with street-level views. Moreover, rendering remains fundamentally limited by GPU memory, as all visible chunks must reside in VRAM simultaneously. We introduce A LoD of Gaussians, a framework for training and rendering ultra-large-scale Gaussian scenes on a single consumer-grade GPU -- without partitioning. Our method stores the full scene out-of-core (e.g., in CPU memory) and trains a Level-of-Detail (LoD) representation directly, dynamically streaming only the relevant Gaussians. A hybrid data structure combining Gaussian hierarchies with Sequential Point Trees enables efficient, view-dependent LoD selection, while a lightweight caching and view scheduling system exploits temporal coherence to support real-time streaming and rendering. Together, these innovations enable seamless multi-scale reconstruction and interactive visualization of complex scenes -- from broad aerial views to fine-grained ground-level details.

cs.GR↗

arXiv2026

Adaptive Pseudoboson Density-Matrix Renormalization Group for Dilute 2D Systems

Simulating strongly correlated systems in two dimensions is notoriously challenging due to rapid entanglement growth and frustration. Here, we introduce the adaptive projected-purified pseudoboson density-matrix renormalization group (A3P-DMRG) tailored to explore the ground states of dilute lattice models. The method compresses cluster Hilbert spaces by retaining only the most probable low-occupation Fock states, identified via probabilistic bounds and refined through a self-consistent mean-field basis optimization. We demonstrate that A3P-DMRG is advantageous in low-filling and weak-coupling regimes for large system sizes where conventional DMRG struggles. This establishes the method as a versatile tool for studying dilute quantum many-body systems relevant to ultra-cold atom quantum simulators, photonic lattices, Moiré materials and quantum chemistry.

cond-mat.str-el↗

arXiv2026

Pi-GS: Sparse-View Gaussian Splatting with Dense π^3 Initialization

Novel view synthesis has evolved rapidly, advancing from Neural Radiance Fields to 3D Gaussian Splatting (3DGS), which offers real-time rendering and rapid training without compromising visual fidelity. However, 3DGS relies heavily on accurate camera poses and high-quality point cloud initialization, which are difficult to obtain in sparse-view scenarios. While traditional Structure from Motion (SfM) pipelines often fail in these settings, existing learning-based point estimation alternatives typically require reliable reference views and remain sensitive to pose or depth errors. In this work, we propose a robust method utilizing π^3, a reference-free point cloud estimation network. We integrate dense initialization from π^3 with a regularization scheme designed to mitigate geometric inaccuracies. Specifically, we employ uncertainty-guided depth supervision, normal consistency loss, and depth warping. Experimental results demonstrate that our approach achieves state-of-the-art performance on the Tanks and Temples, LLFF, DTU, and MipNeRF360 datasets.

cs.GR↗

arXiv2026

Matrix-product-state-based band-Lanczos solver for quantum cluster approaches

We present matrix-product state (MPS) based band Lanczos method as solver for quantum cluster methods such as the variational cluster approximation. While a naïve implementation of MPS as cluster solver would barely improve its range of applicability, we show that our approach makes it possible to treat cluster geometries well beyond the reach of exact diagonalization methods. The key modifications we introduce are a continuous energy truncation combined with a convergence criterion that is more robust against approximation errors introduced by the MPS representation and provides a bound to deviations in the resulting Green's function. The potential of the resulting cluster solver is demonstrated by computing the self-energy functional for the single-band Hubbard model at half filling in the strongly correlated regime, on different cluster geometries. Here, we find that only when treating large cluster sizes, observables can be extrapolated to the thermodynamic limit, which we demonstrate at the example of the staggered magnetization. Treating clusters sizes with up to $6\times 6$ sites we obtain significant improvement over the extrapolation accessible with exact diagonalization solvers when comparing to quantum Monte Carlo results. Finally, we illustrate the applicability of the MPS cluster solver to more complex models by calculating spectral properties as relevant for the electron-doped cuprate CaCuO$_2$.

cond-mat.str-el↗

arXiv2025

NeuralPVS: Learned Estimation of Potentially Visible Sets

Real-time visibility determination in expansive or dynamically changing environments has long posed a significant challenge in computer graphics. Existing techniques are computationally expensive and often applied as a precomputation step on a static scene. We present NeuralPVS, the first deep-learning approach for visibility computation that efficiently determines from-region visibility in a large scene, running at approximately 100 Hz processing with less than $1\%$ missing geometry. This approach is possible by using a neural network operating on a voxelized representation of the scene. The network's performance is achieved by combining sparse convolution with a 3D volume-preserving interleaving for data compression. Moreover, we introduce a novel repulsive visibility loss that can effectively guide the network to converge to the correct data distribution. This loss provides enhanced robustness and generalization to unseen scenes. Our results demonstrate that NeuralPVS outperforms existing methods in terms of both accuracy and efficiency, making it a promising solution for real-time visibility computation.

cs.GR↗

arXiv2025

AAA-Gaussians: Anti-Aliased and Artifact-Free 3D Gaussian Rendering

Although 3D Gaussian Splatting (3DGS) has revolutionized 3D reconstruction, it still faces challenges such as aliasing, projection artifacts, and view inconsistencies, primarily due to the simplification of treating splats as 2D entities. We argue that incorporating full 3D evaluation of Gaussians throughout the 3DGS pipeline can effectively address these issues while preserving rasterization efficiency. Specifically, we introduce an adaptive 3D smoothing filter to mitigate aliasing and present a stable view-space bounding method that eliminates popping artifacts when Gaussians extend beyond the view frustum. Furthermore, we promote tile-based culling to 3D with screen-space planes, accelerating rendering and reducing sorting costs for hierarchical rasterization. Our method achieves state-of-the-art quality on in-distribution evaluation sets and significantly outperforms other approaches for out-of-distribution views. Our qualitative evaluations further demonstrate the effective removal of aliasing, distortions, and popping artifacts, ensuring real-time, artifact-free rendering.

cs.GR↗

arXiv2024

A logical qubit-design with geometrically tunable error-resistibility

Breaking the error-threshold would mark a milestone in establishing quantum advantage for a wide range of relevant problems. One possible route is to encode information redundantly in a logical qubit by combining several noisy qubits, providing an increased robustness against external perturbations. We propose a setup for a logical qubit built from superconducting qubits (SCQs) coupled to a microwave cavity-mode. Our design is based on a recently discovered geometric stabilizing mechanism in the Bose-Hubbard wheel (BHW), which manifests as energetically well-separated clusters of many-body eigenstates. We investigate the impact of experimentally relevant perturbations between SCQs and the cavity on the spectral properties of the BHW. We show that even in the presence of typical fabrication uncertainties, the occurrence and separation of clustered many-body eigenstates is extremely robust. Introducing an additional, frequency-detuned SCQ coupled to the cavity yields duplicates of these clusters, that can be split up by an on-site potential. We show that this allows to (i) redundantly encode two logical qubit states that can be switched and read out efficiently and (ii) can be separated from the remaining many-body spectrum via geometric stabilization. We demonstrate at the example of an X-gate that the proposed logical qubit reaches single qubit-gate fidelities $>0.999$ in experimentally feasible temperature regimes $\sim10-20\,\mathrm{mK}$.

quant-ph↗

arXiv2023

LORD: Leveraging Open-Set Recognition with Unknown Data

Handling entirely unknown data is a challenge for any deployed classifier. Classification models are typically trained on a static pre-defined dataset and are kept in the dark for the open unassigned feature space. As a result, they struggle to deal with out-of-distribution data during inference. Addressing this task on the class-level is termed open-set recognition (OSR). However, most OSR methods are inherently limited, as they train closed-set classifiers and only adapt the downstream predictions to OSR. This work presents LORD, a framework to Leverage Open-set Recognition by exploiting unknown Data. LORD explicitly models open space during classifier training and provides a systematic evaluation for such approaches. We identify three model-agnostic training strategies that exploit background data and applied them to well-established classifiers. Due to LORD's extensive evaluation protocol, we consistently demonstrate improved recognition of unknown data. The benchmarks facilitate in-depth analysis across various requirement levels. To mitigate dependency on extensive and costly background datasets, we explore mixup as an off-the-shelf data generation technique. Our experiments highlight mixup's effectiveness as a substitute for background datasets. Lightweight constraints on mixup synthesis further improve OSR performance.

cs.CV↗

arXiv2023

Stable bipolarons in open quantum systems

Recent advances in numerical methods significantly pushed forward the understanding of electrons coupled to quantized lattice vibrations. At this stage, it becomes increasingly important to also account for the effects of physically inevitable environments. In particular, we study the transport properties of the Hubbard-Holstein Hamiltonian that models a large class of materials characterized by strong electron-phonon coupling, in contact with a dissipative environment. Even in the one-dimensional and isolated case, simulating the quantum dynamics of such a system with high accuracy is very challenging due to the infinite dimensionality of the phononic Hilbert spaces. For this reason, the effects of dissipation on the conductance properties of such systems have not been investigated systematically so far. We combine the non-Markovian hierarchy of pure states method and the Markovian quantum jumps method with the newly introduced projected purified density-matrix renormalization group, creating powerful tensor-network methods for dissipative quantum many-body systems. Investigating their numerical properties, we find a significant speedup up to a factor $\sim 30$ compared to conventional tensor-network techniques. We apply these methods to study dissipative quenches, aiming for an in-depth understanding of the formation, stability, and quasi-particle properties of bipolarons. Surprisingly, our results show that in the metallic phase dissipation localizes the bipolarons, which is reminiscent of an indirect quantum Zeno effect. However, the bipolaronic binding energy remains mainly unaffected, even in the presence of strong dissipation, exhibiting remarkable bipolaron stability. These findings shed light on the problem of designing real materials exhibiting phonon-mediated high-$T_\mathrm{C}$ superconductivity.

cond-mat.str-el↗

arXiv2023

Resolving competition of charge-density wave and superconducting phases using the MPS+MF algorithm

Materials with strong electronic correlations may exhibit a superconducting (SC) phase when tuning some parameters, but they almost always also have multiple other phases, typically insulating ones, that are in close competition with SC. It is highly challenging to resolve this competition with quantitative numerics for the group of quasi-two-dimensional materials such as the cuprates. This is the case even for the simplified minimal models of these materials, the doped 2D Hubbard model with repulsive interactions, where clusters of sufficient size to determine the phase in the thermodynamic limit can be hard-to-impossible to treat in practice. The present work shows how quasi-one-dimensional systems, 2D and 3D arrays of weakly coupled 1D correlated electrons, are much more amenable to resolve the competition between SC and insulating orders on an equal footing using matrix-product states (MPS). Using the recently established MPS plus mean field (MPS+MF) approach for fermions, we demonstrate that large systems are readily reachable in these systems, and thus the thermodynamic regime by extrapolation. Focusing on basic model systems, 3D arrays of negative-U Hubbard chains with additional nearest-neighbor interaction V, we show that despite the MF component of the MPS+MF technique we can reproduce the expected coexistence of SC and charge-density wave at V=0 for density n=1. We then show how we can tune away from coexistence by both tuning V and doping the system. This work paves the way to deploy two-channel MPS+MF theory on some highly demanding high-$T_c$ SC systems, such as 3D arrays of repulsive-U doped Hubbard ladders, where we have recently characterized the properties of such arrays in single-channel MPS+MF calculations. The present approach could thus conclusively show that this SC order would actually be obtained, by explicitly comparing SC against its insulating competitors.

cond-mat.str-el↗

arXiv2022

Transient superconductivity in three-dimensional Hubbard systems by combining matrix product states and self-consistent mean-field theory

We combine matrix-product state (MPS) and Mean-Field (MF) methods to model the real-time evolution of a three-dimensional (3D) extended Hubbard system formed from one-dimensional (1D) chains arrayed in parallel with weak coupling in-between them. This approach allows us to treat much larger 3D systems of correlated fermions out-of-equilibrium over a much more extended real-time domain than previous numerical approaches. We deploy this technique to study the evolution of the system as its parameters are tuned from a charge-density wave (CDW) phase into the superconducting (SC) regime, which allows us to investigate the formation of transient non-equilibrium SC. In our ansatz, we use MPS solutions for chains as input for a self-consistent time-dependent MF scheme. In this way, the 3D problem is mapped onto an effective 1D Hamiltonian that allows us to use the MPS efficiently to perform the time evolution, and to measure the BCS order parameter as a function of time. Our results confirm previous findings for purely 1D systems that for such a scenario superconductivity forms in a transient state.

cond-mat.str-el↗

arXiv2022

Solving 2D and 3D lattice models of correlated fermions -- combining matrix product states with mean field theory

Correlated electron states are at the root of many important phenomena including unconventional superconductivity (USC), where electron-pairing arises from repulsive interactions. Computing the properties of correlated electrons, such as the critical temperature $T_c$ for the onset of USC, efficiently and reliably from the microscopic physics with quantitative methods remains a major challenge for almost all models and materials. In this theoretical work we combine matrix product states (MPS) with static mean field (MF) to provide a solution to this challenge for quasi-one-dimensional (Q1D) systems: Two- and three-dimensional (2D/3D) materials comprised of weakly coupled correlated 1D fermions. This MPS+MF framework for the ground state and thermal equilibrium properties of Q1D fermions is developed and validated for attractive Hubbard systems first, and further enhanced via analytical field theory. We then deploy it to compute $T_c$ for superconductivity in 3D arrays of weakly coupled, doped and repulsive Hubbard ladders. The MPS+MF framework thus enables the reliable, quantitative and unbiased study of USC and high-$T_c$ superconductivity - and potentially many more correlated phases - in fermionic Q1D systems from microscopic parameters, in ways inaccessible to previous methods. It opens the possibility of designing deliberately optimized Q1D superconductors, from experiments in ultracold gases to synthesizing new materials.

cond-mat.str-el↗

arXiv2022

Exploring the Open World Using Incremental Extreme Value Machines

Dynamic environments require adaptive applications. One particular machine learning problem in dynamic environments is open world recognition. It characterizes a continuously changing domain where only some classes are seen in one batch of the training data and such batches can only be learned incrementally. Open world recognition is a demanding task that is, to the best of our knowledge, addressed by only a few methods. This work introduces a modification of the widely known Extreme Value Machine (EVM) to enable open world recognition. Our proposed method extends the EVM with a partial model fitting function by neglecting unaffected space during an update. This reduces the training time by a factor of 28. In addition, we provide a modified model reduction using weighted maximum K-set cover to strictly bound the model complexity and reduce the computational effort by a factor of 3.5 from 2.1 s to 0.6 s. In our experiments, we rigorously evaluate openness with two novel evaluation protocols. The proposed method achieves superior accuracy of about 12 % and computational efficiency in the tasks of image classification and face recognition.

cs.LG↗

arXiv2021

Superconducting pairing from repulsive interactions of fermions in a flat-band system

Fermion systems with flat bands can boost superconductivity by enhancing the density of states at the Fermi level. We use quasiexact numerical methods to show that repulsive interactions between spinless fermions in a one-dimensional (1D) flat-band system, the Creutz ladder, give a finite pairing energy that increases with repulsion, though charge quasi-order (QO) remains dominant. Adding an attractive component shifts the balance in favor of superconductivity and the interplay of two flat bands further yields a remarkable enhancement of superconductivity, well outside of known paradigms for 1D fermions.

cond-mat.str-el↗

arXiv2021

Merging-ISP: Multi-Exposure High Dynamic Range Image Signal Processing

High dynamic range (HDR) imaging combines multiple images with different exposure times into a single high-quality image. The image signal processing pipeline (ISP) is a core component in digital cameras to perform these operations. It includes demosaicing of raw color filter array (CFA) data at different exposure times, alignment of the exposures, conversion to HDR domain, and exposure merging into an HDR image. Traditionally, such pipelines cascade algorithms that address these individual subtasks. However, cascaded designs suffer from error propagation, since simply combining multiple steps is not necessarily optimal for the entire imaging task. This paper proposes a multi-exposure HDR image signal processing pipeline (Merging-ISP) to jointly solve all these subtasks. Our pipeline is modeled by a deep neural network architecture. As such, it is end-to-end trainable, circumvents the use of hand-crafted and potentially complex algorithms, and mitigates error propagation. Merging-ISP enables direct reconstructions of HDR images of dynamic scenes from multiple raw CFA images with different exposures. We compare Merging-ISP against several state-of-the-art cascaded pipelines. The proposed method provides HDR reconstructions of high perceptual quality and it quantitatively outperforms competing ISPs by more than 1 dB in terms of PSNR.

eess.IV↗

arXiv2021

Comparative Study of State-of-the-Art Matrix-Product-State Methods for Lattice Models with Large Local Hilbert Spaces

Lattice models consisting of high-dimensional local degrees of freedom without global particle-number conservation constitute an important problem class in the field of strongly correlated quantum many-body systems. For instance, they are realized in electron-phonon models, cavities, atom-molecule resonance models, or superconductors. In general, these systems elude a complete analytical treatment and need to be studied using numerical methods where matrix-product states (MPS) provide a flexible and generic ansatz class. Typically, MPS algorithms scale at least quadratic in the dimension of the local Hilbert spaces. Hence, tailored methods, which truncate this dimension, are required to allow for efficient simulations. Here, we describe and compare three state-of-the-art MPS methods each of which exploits a different approach to tackle the computational complexity. We analyze the properties of these methods for the example of the Holstein model, performing high-precision calculations as well as a finite-size-scaling analysis of relevant ground-state obervables. The calculations are performed at different points in the phase diagram yielding a comprehensive picture of the different approaches.

cond-mat.str-el↗