arXiv subjects

Xin Wu

Publications and source records attributed to Xin Wu.

At least 37 records · Page 2Linked to original sources

arXiv2026

Third type of spacetime with the coexistence of integrability and non-integrability

The integrability or non-integrability of a spacetime usually refers to whether the motion of massive or massless particles in the spacetime is integrable or not. The standard black hole spacetimes such as the Schwarzschild and Kerr metrics are always integrable for both timelike and null geodesics. They belong to a first type of spacetime. However, the Melvin type spacetimes as a second type of spacetime are non-integrable, regardless of whether they are for massive or massless particle motion. In this paper, we discover the possibility of a third type of spacetime with non-integrable dynamics of timelike geodesics and integrable dynamics of null geodesics. In fact, conformal transformations may transform type one solutions into type three. This is due to the conformal factors preventing the separation of variables from the Hamilton-Jacobi equation and leading to the absence of a fourth constant of motion for the massive particle dynamics. Nevertheless, the massless particle motion still remains integrable in these metrics for any conformal factors because the conformal factors have no effect on the null geodesics whatsoever. The conformal Kerr metric is an example of the third type of spacetime. In addition to the conformal transformation method, other paths may yield the third type of spacetime. The Kerr-Bertotti-Robinson black hole metric and the accelerating Schwarzschild spacetime are two examples of non-conformal solutions that are also of type three.

gr-qc↗

arXiv2026

Spatial4D-Bench: A Versatile 4D Spatial Intelligence Benchmark

4D spatial intelligence involves perceiving and processing how objects move or change over time. Humans naturally possess 4D spatial intelligence, supporting a broad spectrum of spatial reasoning abilities. To what extent can Multimodal Large Language Models (MLLMs) achieve human-level 4D spatial intelligence? In this work, we present Spatial4D-Bench, a versatile 4D spatial intelligence benchmark designed to comprehensively assess the 4D spatial reasoning abilities of MLLMs. Unlike existing spatial intelligence benchmarks that are often small-scale or limited in diversity, Spatial4D-Bench provides a large-scale, multi-task evaluation benchmark consisting of ~40,000 question-answer pairs covering 18 well-defined tasks. We systematically organize these tasks into six cognitive categories: object understanding, scene understanding, spatial relationship understanding, spatiotemporal relationship understanding, spatial reasoning and spatiotemporal reasoning. Spatial4D-Bench thereby offers a structured and comprehensive benchmark for evaluating the spatial cognition abilities of MLLMs, covering a broad spectrum of tasks that parallel the versatility of human spatial intelligence. We benchmark various state-of-the-art open-source and proprietary MLLMs on Spatial4D-Bench and reveal their substantial limitations in a wide variety of 4D spatial reasoning aspects, such as route plan, action recognition, and physical plausibility reasoning. We hope that the findings provided in this work offer valuable insights to the community and that our benchmark can facilitate the development of more capable MLLMs toward human-level 4D spatial intelligence. More resources can be found on our project page.

cs.CV↗

arXiv2026

SSR: Pushing the Limit of Spatial Intelligence with Structured Scene Reasoning

While Multimodal Large Language Models (MLLMs) excel in semantic tasks, they frequently lack the "spatial sense" essential for sophisticated geometric reasoning. Current models typically suffer from exorbitant modality-alignment costs and deficiency in fine-grained structural modeling precision.We introduce SSR, a framework designed for Structured Scene Reasoning that seamlessly integrates 2D and 3D representations via a lightweight alignment mechanism. To minimize training overhead, our framework anchors 3D geometric features to the large language model's pre-aligned 2D visual semantics through cross-modal addition and token interleaving, effectively obviating the necessity for large-scale alignment pre-training. To underpin complex spatial reasoning, we propose a novel scene graph generation pipeline that represents global layouts as a chain of independent local triplets defined by relative coordinates. This is complemented by an incremental generation algorithm, enabling the model to construct "language-model-friendly" structural scaffolds for complex environments. Furthermore, we extend these capabilities to global-scale 3D global grounding task, achieving absolute metric precision across heterogeneous data sources. At a 7B parameter scale, SSR achieves state-of-the-art performance on multiple spatial intelligence benchmarks, notably scoring 73.9 on VSI-Bench. Our approach significantly outperforms much larger models, demonstrating that efficient feature alignment and structured scene reasoning are the cornerstones of authentic spatial intelligence.

cs.CV↗

arXiv2026

Calculations of Di-Hadron Production via Two-Photon Processes in Relativistic Heavy-Ion Collisions

Two-photon processes in relativistic heavy-ion collisions have emerged as a critical probe of quantum electrodynamics in ultra-intense electromagnetic fields, with recent focus extending beyond dileptons to hadronic final states. At present, quantitative studies of di-hadron production via two-photon interactions remain scarce. In this work, we employ the Equivalent Photon Approximation and the two-photon fusion measurements from $e^{+}e^{-}$ collisions to obtain differential cross-section predictions for $π^{+}π^{-}$, $K^{+}K^{-}$, and $p\bar{p}$ pairs produced in ultra-peripheral $\mathrm{Au{+}Au}$ collisions at $\sqrt{s_{NN}} = 200\,\text{GeV}$ within the STAR acceptance, as well as in $\mathrm{Pb{+}Pb}$ collisions at $\sqrt{s_{NN}} = 5.36\,\text{TeV}$ within typical LHC acceptance. The calculations deliver the unified baseline for light-meson and baryon pairs in this environment, supplying benchmarks for upcoming STAR and LHC measurements and guiding future systematic investigations of hadronic two-photon processes at RHIC and LHC facilities.

hep-ph↗

arXiv2026

Exploring Partial Multi-Label Learning via Integrating Semantic Co-occurrence Knowledge

Partial multi-label learning aims to extract knowledge from incompletely annotated data, which includes known correct labels, known incorrect labels, and unknown labels. The core challenge lies in accurately identifying the ambiguous relationships between labels and instances. In this paper, we emphasize that matching co-occurrence patterns between labels and instances is key to addressing this challenge. To this end, we propose Semantic Co-occurrence Insight Network (SCINet), a novel and effective framework for partial multi-label learning. Specifically, SCINet introduces a bi-dominant prompter module, which leverages an off-the-shelf multimodal model to capture text-image correlations and enhance semantic alignment. To reinforce instance-label interdependencies, we develop a cross-modality fusion module that jointly models inter-label correlations, inter-instance relationships, and co-occurrence patterns across instance-label assignments. Moreover, we propose an intrinsic semantic augmentation strategy that enhances the model's understanding of intrinsic data semantics by applying diverse image transformations, thereby fostering a synergistic relationship between label confidence and sample difficulty. Extensive experiments on four widely-used benchmark datasets demonstrate that SCINet surpasses state-of-the-art methods.

cs.CV↗

arXiv2026

Interference-induced entanglement in an effectively zero-lifetime particle pair

Quantum entanglement in high-energy collisions is often obscured by finite lifetimes, dynamical evolution, and final-state interactions, complicating the identification of genuinely quantum correlations. Ultra-peripheral heavy-ion collisions provide a clean benchmark via the Drell-S${\rm\ddot{o}ding}$ production of nonresonant pion pair, realizing an effectively zero-lifetime particle pair whose quantum correlations are fixed at production and remain robust against subsequent elastic scattering. The coherent superposition of photoproduction amplitudes from two indistinguishable nuclei encodes the linear polarization of quasi-real photons in the orbital motion of the pair, generating a nonfactorizable two-particle quantum state. This entanglement leaves a direct experimental imprint: a characteristic second-harmonic azimuthal modulation in momentum space arising from spin-dependent interference between the two sources. In this paper, we establish a quantitative framework for Drell-S${\rm\ddot{o}ding}$ pion-pair production in relativistic heavy-ion collisions and predict the magnitude and transverse-momentum dependence of the entanglement-induced azimuthal asymmetry. Our results provide experimentally accessible signatures of interference-induced entanglement and a controlled test of quantum coherence in relativistic environments.

hep-ph↗

arXiv2026

Machine-learning correction for the calorimeter saturation of cosmic-ray ions with the Dark Matter Particle Explorer: towards the PeV scale

The Dark MAtter Particle Explorer (DAMPE) instrument is a space-borne cosmic-ray detector, capable of measuring ion fluxes up to $\sim$500 TeV/n. This energy scale is made accessible through its calorimeter, which is the deepest currently operating in orbit. Saturation of the calorimeter readout channels starts occurring above $\sim$100 TeV of incident energy, and can significantly affect the primary energy reconstruction. Different techniques -- analytical and machine-learning based -- were developed to tackle this issue, focusing on the recovery of single-bar deposits, up to several hundreds of TeV. In this work, a new machine-learning technique is presented, which benefits from a unique model to correct the total deposited energy in DAMPE calorimeter. The described method is able to generalise its corrections for different ions and extend the maximum detectable incident energy to the PeV scale. This work is a continuation of the results presented in [1].

astro-ph.HE↗

arXiv2026

RemoteDet-Mamba: A Hybrid Mamba-CNN Network for Multi-modal Object Detection in Remote Sensing Images

Unmanned Aerial Vehicle (UAV) remote sensing, with its advantages of rapid information acquisition and low cost, has been widely applied in scenarios such as emergency response. However, due to the long imaging distance and complex imaging mechanisms, targets in remote sensing images often face challenges such as small object size, dense distribution, and low inter-class discriminability. To address these issues, this paper proposes a multi-modal remote sensing object detection network called RemoteDet-Mamba, which is based on a patch-level four-direction selective scanning fusion strategy. This method simultaneously learns unimodal local features and fuses cross-modal patch-level global semantic information, thereby enhancing the distinguishability of small objects and improving inter-class discrimination. Furthermore, the designed lightweight fusion mechanism effectively decouples densely packed targets while reducing computational complexity. Experimental results on the DroneVehicle dataset demonstrate that RemoteDet-Mamba achieves superior detection performance compared to current mainstream methods, while maintaining low parameter count and computational overhead, showing promising potential for practical applications.

cs.CV↗

arXiv2026

STEP3-VL-10B Technical Report

We present STEP3-VL-10B, a lightweight open-source foundation model designed to redefine the trade-off between compact efficiency and frontier-level multimodal intelligence. STEP3-VL-10B is realized through two strategic shifts: first, a unified, fully unfrozen pre-training strategy on 1.2T multimodal tokens that integrates a language-aligned Perception Encoder with a Qwen3-8B decoder to establish intrinsic vision-language synergy; and second, a scaled post-training pipeline featuring over 1k iterations of reinforcement learning. Crucially, we implement Parallel Coordinated Reasoning (PaCoRe) to scale test-time compute, allocating resources to scalable perceptual reasoning that explores and synthesizes diverse visual hypotheses. Consequently, despite its compact 10B footprint, STEP3-VL-10B rivals or surpasses models 10$\times$-20$\times$ larger (e.g., GLM-4.6V-106B, Qwen3-VL-235B) and top-tier proprietary flagships like Gemini 2.5 Pro and Seed-1.5-VL. Delivering best-in-class performance, it records 92.2% on MMBench and 80.11% on MMMU, while excelling in complex reasoning with 94.43% on AIME2025 and 75.95% on MathVision. We release the full model suite to provide the community with a powerful, efficient, and reproducible baseline.

cs.CV↗

arXiv2026

PaCoRe: Learning to Scale Test-Time Compute with Parallel Coordinated Reasoning

We introduce Parallel Coordinated Reasoning (PaCoRe), a training-and-inference framework designed to overcome a central limitation of contemporary language models: their inability to scale test-time compute (TTC) far beyond sequential reasoning under a fixed context window. PaCoRe departs from the traditional sequential paradigm by driving TTC through massive parallel exploration coordinated via a message-passing architecture in multiple rounds. Each round launches many parallel reasoning trajectories, compacts their findings into context-bounded messages, and synthesizes these messages to guide the next round and ultimately produce the final answer. Trained end-to-end with large-scale, outcome-based reinforcement learning, the model masters the synthesis abilities required by PaCoRe and scales to multi-million-token effective TTC without exceeding context limits. The approach yields strong improvements across diverse domains, and notably pushes reasoning beyond frontier systems in mathematics: an 8B model reaches 94.5% on HMMT 2025, surpassing GPT-5's 93.2% by scaling effective TTC to roughly two million tokens. We open-source model checkpoints, training data, and the full inference pipeline to accelerate follow-up work.

cs.LG↗

arXiv2025

Otus Supercomputer

Otus is a high-performance computing cluster that was launched in 2025 and is operated by the Paderborn Center for Parallel Computing (PC2) at Paderborn University in Germany. The system is part of the National High Performance Computing (NHR) initiative. Otus complements the previous supercomputer Noctua 2, offering approximately twice the computing power while retaining the three node types that were characteristic of Noctua 2: 1) CPU compute nodes with different memory capacities, 2) high-end GPU nodes, and 3) HPC-grade FPGA nodes. On the Top500 list, which ranks the 500 most powerful supercomputers in the world, Otus is in position 164 with the CPU partition and in position 255 with the GPU partition (June 2025). On the Green500 list, ranking the 500 most energy-efficient supercomputers in the world, Otus is in position 5 with the GPU partition (June 2025). This article provides a comprehensive overview of the system in terms of its hardware, software, system integration, and its overall integration into the data center building to ensure energy-efficient operation. The article aims to provide unique insights for scientists using the system and for other centers operating HPC clusters. The article will be continuously updated to reflect the latest system setup and measurements.

cs.DC↗

arXiv2025

POLAR-2 -- Latest Developments of the Next Generation GRB Polarimeter

Gamma-Ray Bursts (GRBs) are among the most energetic events in the Universe. Despite over 50 years of research and measurements their prompt emission remains poorly understood, with key questions surrounding the structure of relativistic jets, magnetic field configurations, and dominant radiation mechanisms. Polarization measurements are critical in resolving these uncertainties. The POLAR mission, operational in 2016-2017 on Tiangong-2, provided the most statistically significant GRB polarization data. Its results indicated low time-averaged polarization with hints of temporal evolution. However, POLAR's limited sensitivity, small effective area, and restricted energy range prevented more detailed time- and energy-resolved analyses in addition to a larger sample of GRB polarization measurements. POLAR-2 is designed to address these limitations by offering a fourfold increase in effective area (at least) and an extended energy range of 30-800 keV by utilizing Silicon Photomultipliers (SiPMs) and an updated module design, enabling the differentiation of competing GRB emission models. The instrument comprises of 100 polarimeter modules (each with 64 plastic scintillator bars), wherein the polarization angle is extracted through Compton Scattering of the gammas. The polarimeter module design was validated during an ESRF beam test campaign in 2023. The instrument was developed by a joint effort of Switzerland, China, Poland and Germany and is planned for launch in 2027. Currently, POLAR-2 is in its production phase with the first module targets being produced. We will provide an overview of the current status of the development.

astro-ph.IM↗

arXiv2025

Comprehensive Study of the Lunar Energetic Particle Environment with LunPAN

LunPAN (Lunar Particle Analyzer Network) is a three-year mission proposal designed to comprehensively map the particle spectra in the lunar radiation field. It aims to provide precise measurements of Galactic Cosmic Rays (GCR), Solar Energetic Particles (SEP), and albedo particles, including charged particles, neutrons, and gamma-rays, originating from the Moon's surface. Therefore it will contribute to fundamental space physics, lunar geology sciences, space weather prediction, and radiation risk assessment for future lunar explorations. This is achieved through two state-of-the-art instruments; Pix.PAN and NeuPix. Pix.PAN is a compact magnetic spectrometer designed for precise measurements of penetrating charged particles, ranging from 100 MeV to 10 GeV. Based on the Mini.PAN project, Pix.PAN employs thin silicon pixel sensors optimized for energy resolution and particle identification. NeuPix is a hybrid active pixel sensor system capable of detecting neutrons, gamma-rays, and lower-energy charged particles between 10MeV and 100 MeV. Utilizing innovative sensor-converter combinations, NeuPix will provide spectral measurements of lunar albedo neutrons and gamma-ray fluxes. Currently, the LunPAN mission is accepted by ESA's "Small Missions for Exploration - Destination the Moon" call for a pre-A phase study. We will discuss mission outline and expected scientific performance of the PixPAN and NeuPix.

astro-ph.IM↗

arXiv2025

Out-of-Distribution Generalization in Time Series: A Survey

Time series frequently manifest distribution shifts, diverse latent features, and non-stationary learning dynamics, particularly in open and evolving environments. These characteristics pose significant challenges for out-of-distribution (OOD) generalization. While substantial progress has been made, a systematic synthesis of advancements remains lacking. To address this gap, we present the first comprehensive review of OOD generalization methodologies for time series, organized to delineate the field's evolutionary trajectory and contemporary research landscape. We organize our analysis across three foundational dimensions: data distribution, representation learning, and OOD evaluation. For each dimension, we present several popular algorithms in detail. Furthermore, we highlight key application scenarios, emphasizing their real-world impact. Finally, we identify persistent challenges and propose future research directions. A detailed summary of the methods reviewed for the generalization of OOD in time series can be accessed at https://tsood-generalization.com.

cs.LG↗

arXiv2025

Implications of recent LHCb data on CPV in b-baryon four body decays

Motivated by the recent CPV observation, we investigate the CPV of b-baryon charmless four body decays under the U-spin symmetry. However, we find that only U-spin symmetry cannot provide effective predictions, particularly for $Λ_b$ decays. For giving more useful predictions, we also give a simple dynamic analysis. By counting the power($λ=\sqrt{\frac{Λ_{QCD}}{m_b}}$) of each topological diagram, we find that for the specific decay $B_b^2\to R(B_1^2 M^2 M^{\bar 2})M^2$, only one U-spin amplitude can contribute in the leading power, while for $B_b^2\to R(B_1^2 M^2)R( M^{\bar 2}M^2)$, only two U-spin amplitudes can contribute in this leading power. Then the most effective prediction can be given as \begin{align} &A_{CP}^{dir}(Λ_b^0 \to R( p π^- π^+ )π^-) = (-12.99 \pm 2.83\pm2.59\pm0.65)\%,\notag \end{align} Considering the $Λ_b$ can effectively produced in LHCb, we strongly encourage a more precise experimental investigation of it.

hep-ph↗

arXiv2025

Zero-Shot Large Language Model Agents for Fully Automated Radiotherapy Treatment Planning

Radiation therapy treatment planning is an iterative, expertise-dependent process, and the growing burden of cancer cases has made reliance on manual planning increasingly unsustainable, underscoring the need for automation. In this study, we propose a workflow that leverages a large language model (LLM)-based agent to navigate inverse treatment planning for intensity-modulated radiation therapy (IMRT). The LLM agent was implemented to directly interact with a clinical treatment planning system (TPS) to iteratively extract intermediate plan states and propose new constraint values to guide inverse optimization. The agent's decision-making process is informed by current observations and previous optimization attempts and evaluations, allowing for dynamic strategy refinement. The planning process was performed in a zero-shot inference setting, where the LLM operated without prior exposure to manually generated treatment plans and was utilized without any fine-tuning or task-specific training. The LLM-generated plans were evaluated on twenty head-and-neck cancer cases against clinical manual plans, with key dosimetric endpoints analyzed and reported. The LLM-generated plans achieved comparable organ-at-risk (OAR) sparing relative to clinical plans while demonstrating improved hot spot control (Dmax: 106.5% vs. 108.8%) and superior conformity (conformity index: 1.18 vs. 1.39 for boost PTV; 1.82 vs. 1.88 for primary PTV). This study demonstrates the feasibility of a zero-shot, LLM-driven workflow for automated IMRT treatment planning in a commercial TPS. The proposed approach provides a generalizable and clinically applicable solution that could reduce planning variability and support broader adoption of AI-based planning strategies.

physics.med-ph↗

arXiv2025

Chaos of charged particles near a renormalized group improved Kerr black hole in an external magnetic field

In a quantum theory of gravity, a renormalization group improved Kerr metric is obtained from the Kerr metric, where the Newton gravitational constant is modified as a function of the radial distance. The motion of neutral test particles in this metric is integrable. However, the dynamics of charged test particles is nonintegrable when an external asymptotically homogeneous magnetic field exists in the vicinity of the black hole. The transition from regular dynamics to chaotic dynamics is numerically traced as one or two dynamical parameters vary. From a statistical point of view, the strength of chaos is typically enhanced as both the particle energy and the magnetic field increase, but it is weakened with increasing the particle angular momentum and the black hole spin. In particular, an increase of the quantum corrected parameter weakens the extent of chaos. This is because the running Newton gravity constant effectively weakens the central gravitational attraction and results in decreasing sensitivity to initial conditions.

gr-qc↗

arXiv2025

ERIS: An Energy-Guided Feature Disentanglement Framework for Out-of-Distribution Time Series Classification

An ideal time series classification (TSC) should be able to capture invariant representations, but achieving reliable performance on out-of-distribution (OOD) data remains a core obstacle. This obstacle arises from the way models inherently entangle domain-specific and label-relevant features, resulting in spurious correlations. While feature disentanglement aims to solve this, current methods are largely unguided, lacking the semantic direction required to isolate truly universal features. To address this, we propose an end-to-end Energy-Regularized Information for Shift-Robustness (ERIS) framework to enable guided and reliable feature disentanglement. The core idea is that effective disentanglement requires not only mathematical constraints but also semantic guidance to anchor the separation process. ERIS incorporates three key mechanisms to achieve this goal. Specifically, we first introduce an energy-guided calibration mechanism, which provides crucial semantic guidance for the separation, enabling the model to self-calibrate. Additionally, a weight-level orthogonality strategy enforces structural independence between domain-specific and label-relevant features, thereby mitigating their interference. Moreover, an auxiliary adversarial generalization mechanism enhances robustness by injecting structured perturbations. Experiments across four benchmarks demonstrate that ERIS achieves a statistically significant improvement over state-of-the-art baselines, consistently securing the top performance rank.

cs.LG↗