arXiv subjects

Wei Cui

Publications and source records attributed to Wei Cui.

At least 19 recordsLinked to original sources

arXiv2026

Resolving the thermal component of stellar feedback with the "anchor-and-refine" scheme in galactic simulations

(abridged) Stellar feedback regulates galactic baryon cycles, but galaxy-scale simulations often cannot resolve supernova cooling radii. Momentum-based subgrid models can consequently underresolve the hot gas around young stars. We develop an "anchor-and-refine" feedback scheme that anchors a host cell near each star particle, preventing its displacement and preserving a thermal-feedback reservoir. When the cooling radius is resolved, all supernova energy is injected into the host cell as thermal energy; otherwise, the host cell is refined and terminal momentum is deposited into surrounding gas cells. Idealized tests agree with high-resolution simulations, yielding consistent superbubble properties and preserving hot interiors without artificial cooling. In an isolated Milky Way-mass galaxy, the scheme captures finer X-ray-emitting structures and a sharply defined, X-ray-bright central mid-plane. It also produces a smoother star formation history than pure hybrid feedback. Hot gas dominates outflow mass loading, with the mass-loading factor plateauing at approximately 0.2 at heights z>1 kpc. Phase decomposition indicates that hot gas sustains a steady, energy-driven wind, whereas cooler outflows decline with height because of radiative cooling and gravitational fallback. These results highlight the role of thermal feedback in driving galactic winds. By preserving hot superbubble interiors and the partition of thermal and kinetic energy, anchor-and-refine improves the treatment of hot gas, star formation, and feedback at limited resolution, supporting future applications to galaxy-scale and cosmological zoom-in simulations.

astro-ph.GA↗

arXiv2026

CanLegalRAGBench: Evaluating Retrieval-Augmented Generation on Canadian Case Law

RAG-based legal assistants have been growing in popularity, but LLM hallucinations remain a key issue and potentially undermines justice. While benchmarks have been developed to evaluate progress, many rely on synthetic queries rather than realistic legal scenarios. Moreover, Canadian law remains underrepresented in existing evaluations. To address this gap, we introduce CanLegalRAGBench, a Canadian legal QA benchmark based on realistic queries and expert-annotated answers grounded in case law. Our evaluation shows that retrieval performance is sensitive to design choices and that open-source embedding models are competitive with closed source models. However, it also reveals the limitation of automatic evaluations that penalize systems for retrieving alternative relevant documents. We also find that generated answers often diverge from gold responses, either with hallucinations or by producing overly detailed or irrelevant content, with 8-29% of claims not being supported by the retrieved documents. We hope this benchmark will help drive continued progress in addressing limitations of legal RAG systems.

cs.CL↗

arXiv2026

Broadband THz spectroscopy system beyond 25 THz using BNA crystals and a tunable single-ring-fiber compressor

We present a terahertz (THz) time-domain spectroscopy (THz-TDS) system which accesses a broadband spectrum, efficiently covering the so-called "new THz gap" between 5 and 15 THz and extending beyond 25 THz. The system exploits nonlinear interactions within the organic crystal BNA (N-benzyl-2-methyl-4-nitroaniline) to generate and detect THz radiation upon excitation by a near-infrared (NIR) pulse centered at 1.03 $μ$m. To enable broadband THz spectral monitoring, the NIR pulse from a Yb-based solid-state laser undergoes spectral broadening in a gas-filled single-ring hollow-core photonic-crystal fiber, followed by a pulse compression to achieve durations as short as 31 fs. This approach paves the way for broadband spectroscopy in hard-to-access THz regions using widely available near-infrared ultrafast sources.

physics.optics↗

arXiv2026

X-ray measurements of the elemental abundances for the diffuse emission in the nuclear starburst galaxy NGC 3079

Outflows driven by feedback leave imprints on a galaxy's diffuse X-ray emission. Elemental abundances provide key diagnostics of feedback-driven chemical enrichment. We present a spectroscopic study of the diffuse X-ray emission in NGC 3079 using XMM-Newton RGS and EPIC, focusing on the nuclear region and the extended galactic-scale superbubble (GSB). We analyzed the high-resolution RGS spectrum of the central $\sim4~\rm kpc$ region, encompassing the $\sim1~\rm kpc$ nuclear superbubble (NSB), and the EPIC spectrum of the GSB out to $5'~(\sim25~\rm kpc)$. The nuclear spectrum is best described by a multi-temperature thermal-plasma model plus a charge-exchange (CX) component, which contributes approximately $13\%$ of the total observed energy flux in the $0.2-2~\rm keV$ band. The best-fit O/Fe and Ne/Fe ratios are $0.47$ and $0.77$ in solar units. Comparing these ratios with IMF-weighted core-collapse supernova (SNcc) yields gives an upper progenitor-mass cutoff of $M_{\rm u}\simeq13-15~M_{\odot}$ for pure SNcc enrichment. For the best-fit spectral model, including an SNIa contribution increases the inferred cutoff, but its $1σ$ upper limit remains below $25~M_\odot$ for SNIa fractions up to $10\%$ within the adopted yield models. The characteristic ages of the NSB and GSB are $0.92^{+0.39}_{-0.46}~\rm Myr$ and $29.6^{+1.6}_{-2.9}~\rm Myr$, respectively, suggesting episodic feedback in NGC 3079.

astro-ph.HE↗

arXiv2026

Categorical spin-networks: state-sum invariants of 4-manifolds from higher-gauge theory

In this work, we define $(3+1)$-dimensional topological quantum field theories (TQFTs) and their lattice realizations with input data given by a pivotal fusion 2-category; more precisely, our framework applies in wider generality to presemisimple locally fusion pivotal tensor $2$-categories, which are equipped with a so-called \textit{shadow trace}. We introduce a decoration of triangulated manifolds by the input data, and develop a $20j$-symbol formalism for computing 4-simplex scattering amplitudes. We verify the topological invariance of 20$j$-symbol summations under the 4-dimensional Pachner moves. Furthermore, we also construct explicitly a 3-dimensional Levin-Wen-type lattice Hamiltonian model realizing the state-sum invariant. Our construction can be understood as a higher-dimensional analogue of the spin-networks construction for the Turaev-Viro invariant. Examples arising from $2$-groups are discussed from the perspective of higher gauge theory.

math-ph↗

arXiv2026

Half-Spacetime Gauging of 2-Group Symmetry in 3d

We construct a class of non-invertible duality defects, in (2+1)d quantum field theories, arising from half-spacetime gauging of a 2-group symmetry. Starting from a parent theory with two discrete and Abelian 0-form symmetries and a prescribed mixed anomaly, we show that gauging one factor produces a theory with a 2-group symmetry, while gauging the other yields a theory with a non-invertible 0-form symmetry, whose fusion rules we derive explicitly. When the parent theory possesses three such symmetries with a cyclic anomaly structure, gauging different factors can produce mutually dual theories and the half-spacetime gauging of the 2-group is implemented by a non-invertible duality defect, whose fusion rules we obtain. We illustrate the construction with explicit examples, including a $U(1)\times U(1)\times U(1)$ gauge theory and a general class of product theories. We also include a self-contained pedagogical introduction to the cohomological tools employed throughout the article.

hep-th↗

arXiv2026

SOLO: Stable Omni-terrain Long-Horizon Perceptive Humanoid Locomotion

Humans traverse complex terrain over long distances without losing balance, whereas perceptive humanoid policies become fragile as perception and control errors accumulate. We present SOLO, a unified framework addressing two compounding causes of this long-horizon fragility: dense terrain reconstruction smooths action-critical details, and pointwise imitation lacks temporal credit assignment. Its Query Reconstructor (QR) uses Fourier-encoded cell queries to retrieve spatially specific evidence from depth-proprioception tokens, preserving sharp terrain boundaries. Trajectory-Aware MSE (TA-MSE) Distillation adds next-state teacher-student disagreement to the PPO reward, enabling Generalized Advantage Estimation to propagate future disagreement penalties to preceding actions. In simulation, QR reduces height-map L1 error by factors of 3.3-4.0, while TA-MSE surpasses PPO and MSE+PPO in curriculum progression. On stress-test terrains, SOLO achieves 97.5% mean traversal success and 96% stepping-stone success, versus 75.0-75.6% and 0-3% for dense-reconstructor variants. Deployed zero-shot with only a chest-mounted depth camera and proprioception, SOLO completes a continuous 1.5-km outdoor route and an indoor mixed-terrain course. Project page: https://sunpihai-up.github.io/solo/

cs.RO↗

arXiv2026

Detector Development for HUBS I: Initial Testing of Small-Area TES Microcalorimeters

We report progress on the ongoing development of microcalorimeter detector technology for the Hot Universe Baryon Surveyor (HUBS) mission. We show the results from testing and characterizing selected pixels in a 10$\times$10 microcalorimeter array. The microcalorimeter is based on a Mo/Cu transition-edge sensor (TES) coupled to an Au absorber. To better understand the properties of the devices, we have first measured the energy resolution of a selected pixel in a TES array of the same design with a pulsed laser system that produces 3 eV photons, and found that individual photon peaks are easily resolved with the TES, indicating good performance. We have then exposed the microcalorimeter array to radiation from a $^{55}$Fe source, and found that the pixels tested show energy resolutions as good as 3.7$\pm$0.1 eV at 5.9 keV. The energy resolution is found to vary monotonically with the bias point for all the devices, showing little evidence for the presence of the so-called excess noise. This is consistent with the results from modeling the measured noise spectrum. The effects of thermal crosstalk are evident, leading to the degradation of energy resolution.

astro-ph.IM↗

arXiv2026

DRESS: Disentangled Representation-based Self-Supervised Meta-Learning for Diverse Tasks

Meta-learning represents a strong class of approaches for solving few-shot learning tasks. Nonetheless, recent research suggests that simply pre-training a generic encoder can potentially surpass meta-learning algorithms. In this paper, we first discuss the reasons why meta-learning fails to stand out in these few-shot learning experiments, and hypothesize that it is due to the few-shot learning tasks lacking diversity. We propose DRESS, a task-agnostic Disentangled REpresentation-based Self-Supervised meta-learning approach that enables fast model adaptation on highly diversified few-shot learning tasks. Specifically, DRESS utilizes disentangled representation learning to create self-supervised tasks that can fuel the meta-training process. Furthermore, we also propose a class-partition based metric for quantifying the task diversity directly on the input space. We validate the effectiveness of DRESS through experiments on datasets with multiple factors of variation and varying complexity. The results suggest that DRESS is able to outperform competing methods on the majority of the datasets and task setups. Through this paper, we advocate for a re-examination of proper setups for task adaptation studies, and aim to reignite interest in the potential of meta-learning for solving few-shot learning tasks via disentangled representations.

cs.LG↗

arXiv2026

Machine Learning Topological Order from Defect Partition Functions

We introduce a machine learning framework for extracting Ising topological order from defect partition functions of the two-dimensional Ising model on a torus. Restricted Boltzmann Machines (RBMs) are trained on Ising model data sampled at criticality across topological sectors. We take a component-wise square-root map of the learned distributions which naturally produces candidate wavefunctions for the (2+1)-dimensional Ising TQFT. As a nontrivial consistency check, we extract the modular S-matrix from overlaps of the resulting states and recover the expected Ising modular data. Our results demonstrate that neural network representations can capture both critical fluctuations and emergent topological structure, providing a data-driven route from lattice statistical mechanics to topological quantum field theory.

cond-mat.dis-nn↗

arXiv2026

DIffuse X-ray Explorer (DIXE): Sky Survey Strategy and Collimator Response Demodulation

DIffuse X-ray Explorer (DIXE) is a proposed high-resolution X-ray spectroscopic surveyor aimed at studying large structures of hot gas in the Milky Way. Its payload is designed to have a field of view (FoV) of $10^\circ$ (half-power diameter) and an energy resolution of better than 6 eV, covering an energy range of 0.1-10 keV. It will be mounted on the China Space Station (CSS) and follow the CSS orbit to conduct the survey with fixed zenith pointing in order to optimize the coverage of key science targets. The payload will avoid the Sun passively via an operable sunshade, where a minimum $25^\circ$ angular separation between the pointing axis and the direction of the Sun is required. Two Sun-avoidance strategies are considered: one focusing on minimizing mechanical risk and the other on maximizing exposure time. The one-year exposure maps indicate that DIXE will cover approximately $72.5\%$ of the sky, with typical exposure times of 26 ks and 68 ks for the two strategies, respectively. Although mechanically collimated, the imaging performance of the payload can be enhanced with a demodulation method based on Markov Chain Monte Carlo sampling using the collimator response. Through simulation, we found that the method could achieve a localization accuracy of $1^\circ$ for point-like sources and a spatial resolution of $3^\circ$ for the extended sources of complex surface brightness distribution, both of which are significantly smaller than the FoV.

astro-ph.IM↗

arXiv2026

Conf-Gen: Conformal Uncertainty Quantification for Generative Models

Conformal prediction (CP) and its extension, conformal risk control (CRC), are established frameworks for quantifying uncertainty in supervised machine learning through formal guarantees. However, recent breakthroughs in artificial intelligence (AI) have been driven by unsupervised generative models, such as large language models (LLMs) and image generators, which are not directly compatible with CP or CRC. In this work we introduce conformal generation (Conf-Gen), a general framework adapting CRC to generative tasks while relaxing its theoretical assumptions. Conf-Gen unifies and generalizes previous attempts to apply CP to LLMs, and extends conformal methodology to entirely new domains. We demonstrate the flexibility of Conf-Gen through some novel applications, including obtaining conformal guarantees on: image generators producing non-memorized images, conversational AI systems having asked enough clarifying questions, and the output of AI agents being correct.

cs.LG↗

arXiv2026

Central Charges and Vacuum Moduli of 2d $\mathcal{N}=(0,4)$ Theories from Class $\mathcal{S}$

We investigate 2d $\mathcal{N}=(0,4)$ supersymmetric theories obtained from a topologically-twisted reduction of 4d $\mathcal{N}=2$ class $\mathcal{S}$ theories on a Riemann surface. This study addresses subtle aspects of central charges, unbroken gauge groups, and emergent superconformal R-symmetries of these theories. Focusing on infrared vacuum structures, we propose conjectural formulas for the central charges. For theories with the gauge group $SU(2)$, we use a Lagrangian description to analyze the vacuum moduli spaces. In particular, we examine two distinct branches -- the special Higgs branch and the twisted Higgs branch -- by computing their Hilbert series, and find agreement with the proposed central charge formulas.

hep-th↗

arXiv2026

Room-temperature THz photon detection via nonlinear upconversion with 2% full-system efficiency

Sensitive detection of terahertz (THz) radiation is fundamental to progress in spectroscopy, advanced wireless communication, and the realization of emerging quantum technologies. However, the intrinsically low photon energies in the THz range combined with thermal background radiation tend to constrain detector performance when operating at ambient temperatures. Here, we demonstrate efficient room-temperature THz detection based on nonlinear upconversion in the organic crystal N-benzyl-2-methyl-4-nitroaniline (BNA) to resolve frequencies from 1 to 7.5 THz. The system encompassing spectral filters and a single-photon counter achieves an overall detection efficiency of 2% for sum-frequency generated photons. This enables the detection of a train of 50 000 terahertz pulses carrying, on average, fewer than 0.04 photons per pulse, with a signal-to-noise ratio of unity. At a higher flux, when ~60 photons per pulse impinge on the BNA crystal, the per-pulse detection probability reaches 50%. After accounting for loss mechanisms in the setup, the nonlinear THz-to-near-infrared conversion efficiency in BNA exceeds 75%. These results demonstrate the feasibility of quantum experiments relying on single-photon-level THz detection via upconversion in nonlinear crystals in ambient conditions.

physics.optics↗

arXiv2026

RobotPan: A 360$^\circ$ Surround-View Robotic Vision System for Embodied Perception

Surround-view perception is increasingly important for robotic navigation and loco-manipulation, especially in human-in-the-loop settings such as teleoperation, data collection, and emergency takeover. However, current robotic visual interfaces are often limited to narrow forward-facing views, or, when multiple on-board cameras are available, require cumbersome manual switching that interrupts the operator's workflow. Both configurations suffer from motion-induced jitter that causes simulator sickness in head-mounted displays. We introduce a surround-view robotic vision system that combines six cameras with LiDAR to provide full 360$^\circ$ visual coverage, while meeting the geometric and real-time constraints of embodied deployment. We further present \textsc{RobotPan}, a feed-forward framework that predicts \emph{metric-scaled} and \emph{compact} 3D Gaussians from calibrated sparse-view inputs for real-time rendering, reconstruction, and streaming. \textsc{RobotPan} lifts multi-view features into a unified spherical coordinate representation and decodes Gaussians using hierarchical spherical voxel priors, allocating fine resolution near the robot and coarser resolution at larger radii to reduce computational redundancy without sacrificing fidelity. To support long sequences, our online fusion updates dynamic content while preventing unbounded growth in static regions by selectively updating appearance. Finally, we release a multi-sensor dataset tailored to 360$^\circ$ novel view synthesis and metric 3D reconstruction for robotics, covering navigation, manipulation, and locomotion on real platforms. Experiments show that \textsc{RobotPan} achieves competitive quality against prior feed-forward reconstruction and view-synthesis methods while producing substantially fewer Gaussians, enabling practical real-time embodied deployment.

cs.RO↗

arXiv2026

One-shot learning for the complex dynamical behaviors of weakly nonlinear forced oscillators

Extrapolative prediction of complex nonlinear dynamics remains a central challenge in engineering. This study proposes a one-shot learning method to identify global frequency-response curves from a single excitation time history by learning governing equations. We introduce MEv-SINDy (Multi-frequency Evolutionary Sparse Identification of Nonlinear Dynamics) to infer the governing equations of non-autonomous and multi-frequency systems. The methodology leverages the Generalized Harmonic Balance (GHB) method to decompose complex forced responses into a set of slow-varying evolution equations. We validated the capabilities of MEv-SINDy on two critical Micro-Electro-Mechanical Systems (MEMS). These applications include a nonlinear beam resonator and a MEMS micromirror. Our results show that the model trained on a single point accurately predicts softening/hardening effects and jump phenomena across a wide range of excitation levels. This approach significantly reduces the data acquisition burden for the characterization and design of nonlinear microsystems.

cs.LG↗

arXiv2026

Simulation of non X-ray background for the DIffuse X-ray Explorer (DIXE) mission

DIffuse X-ray Explorer (DIXE) is a proposed high-resolution spectroscopic survey mission onboard the China Space Station. Equipped with microcalorimeters based on the Transition-edge sensor technology, it aims to survey the hot gas in the Milky Way. The performance of DIXE depends on the understanding of non X-ray background (NXB), which can strongly affect observations of diffuse X-ray emission. In this work, we simulated the NXB of DIXE in a low-earth orbit (LEO) using \textsc{Geant4}. A detailed mass model of the payload was constructed, and the major sources of NXB were identified, including cosmic rays, albedo neutrons and albedo photons. These components were implemented in \textsc{Geant4} with realistic angular and spectral distributions. We simulated the relevant physical processes of space radiation interacting with the instrument and calculated the resulting NXB. We also evaluated the delayed background from trapped protons in the South Atlantic Anomaly (SAA). Our simulations show that, at the geomagnetic equator and under solar minimum conditions, the NXB is on average $4.46 \times 10^{-2} ~\mathrm{counts~s^{-1}~cm^{-2}~keV^{-1}}$ in 0.1--10 keV energy band, with dominant contributions from the induced particles generated by primary cosmic protons. The NXB increases toward higher geomagnetic latitudes, reaching a maximum of $1.55 \times 10^{-1} ~\mathrm{counts~s^{-1}~cm^{-2}~keV^{-1}}$. The delayed background induced by the SAA decays rapidly after exiting the anomaly and becomes negligible within approximately 5 minutes. The simulated NXB is consistent with that of similar X-ray observatories in LEOs.

astro-ph.IM↗

arXiv2026

SARMAE: Masked Autoencoder for SAR Representation Learning

Synthetic Aperture Radar (SAR) imagery plays a critical role in all-weather, day-and-night remote sensing applications. However, existing SAR-oriented deep learning is constrained by data scarcity, while the physically grounded speckle noise in SAR imagery further hampers fine-grained semantic representation learning. To address these challenges, we propose SARMAE, a Noise-Aware Masked Autoencoder for self-supervised SAR representation learning. Specifically, we construct SAR-1M, the first million-scale SAR dataset, with additional paired optical images, to enable large-scale pre-training. Building upon this, we design Speckle-Aware Representation Enhancement (SARE), which injects SAR-specific speckle noise into masked autoencoders to facilitate noise-aware and robust representation learning. Furthermore, we introduce Semantic Anchor Representation Constraint (SARC), which leverages paired optical priors to align SAR features and ensure semantic consistency. Extensive experiments across multiple SAR datasets demonstrate that SARMAE achieves state-of-the-art performance on classification, detection, and segmentation tasks. Code and models will be available at https://github.com/MiliLab/SARMAE.

cs.CV↗