arXiv subjects

Wei Xiong

Publications and source records attributed to Wei Xiong.

At least 37 records · Page 2Linked to original sources

arXiv2026

Aurora: Unified Video Editing with a Tool-Using Agent

Recent video editing models have converged on a unified conditioning design: a single diffusion transformer jointly consumes text, source video, and reference images, and one set of weights covers replacement, removal, style transfer, and reference-driven insertion. The design is flexible, but it assumes that the user already provides model-ready text, reference images, and spatial grounding for local edits, which real requests often omit. We present Aurora, an agentic video editing framework that pairs a tool-augmented vision-language model (VLM) agent with a unified video diffusion transformer. The VLM agent maps a raw user request to a structured edit plan aligned with the transformer's conditioning channels, thereby resolving textual and visual underspecification before generation. We train the VLM agent with supervised data for complete edit planning and reference-image selection, together with preference pairs for robust tool use and instruction refinement. We introduce AgentEdit-Bench to evaluate agent-enhanced video editing under textual and visual underspecification. Experiments on AgentEdit-Bench and two existing video editing benchmarks show that Aurora improves over instruction-only baselines and that the VLM agent transfers to compatible frozen video editing models. Project page: https://yeates.github.io/Aurora-Page

cs.CV↗

arXiv2026

A new construction of permutation polynomials over $\mathbb{F}_{q^3}$

We determine all permutation polynomials among several families of polynomials over $\mathbb{F}_{q^3}$ for arbitrary prime powers $q$. We obtain some new families of permutation polynomials over $\mathbb{F}_{q^3}$ with simple coefficients for infinitely many characteristics. As a specific consequence, our results resolve the generalization of conjectures of Zhang, Zheng, Wang, Peng, and Li in the even characteristic. Our proofs are conceptually short and involve no complicated computations, in contrast to the proofs of results on permutation polynomials which were published previously. Moreover, we develop a totally new systematic method in this paper for the study of permutation polynomials.

math.CO↗

arXiv2026

Hybrid-order topology in two-dimensional nonsymmorphic antiferromagnets

We theoretically demonstrate hybrid-order topology in a two-dimensional nonsymmorphic antiferromagnet. Utilizing a generic antiferromagnetic Dirac model with a symmetry-allowed, momentum-dependent spin-density-wave (SDW) mass, we show that a single bulk insulating phase exhibits distinct topological boundary manifestations governed solely by the termination geometry. For screw-compatible edges, nonsymmorphic screw symmetry protects gapless first-order edge states. In contrast, for a $45^\circ$ diamond-shaped termination, the screw symmetry is broken at the boundary, resulting in gapped edges. However, the finite geometry still preserves magnetic mirror symmetries $\mathcal{M}_x\mathcal{T}$ and $\mathcal{M}_y\mathcal{T}$, which enforce an alternating pattern of edge masses, thereby binding zero-dimensional corner states. This second-order phase is characterized by a quantized quadrupole moment, with corner states pinned to zero energy by the chiral symmetry. We further demonstrate that explicit lattice perturbations can selectively gap the first-order edge modes while robustly preserving the corner states. Our work establishes a symmetry-based route to a termination-controlled duality between first- and second-order topology in magnetic nonsymmorphic systems.

cond-mat.mes-hall↗

arXiv2026

Rethinking Importance Sampling in LLM Policy Optimization: A Cumulative Token Perspective

Reinforcement learning, including reinforcement learning with verifiable rewards (RLVR), has emerged as a powerful approach for LLM post-training. Central to these approaches is the design of the importance sampling (IS) ratio used in off-policy policy-gradient estimation. Existing methods face a fundamental bias-variance dilemma: token-level IS ratios, as adopted by PPO (Schulman et al., 2017) and GRPO (Shao et al., 2024), introduce bias by ignoring prefix state distribution mismatch; full sequence ratios provide exact trajectory-level correction but suffer from high variance due to the multiplicative accumulation of per-token ratios, while GSPO (Zheng et al., 2025) improves numerical stability via length normalization at the cost of deviating from the exact full-sequence IS correction. In this work, we identify the cumulative token IS ratio, the product of per-token ratios up to position $t$, as a theoretically principled solution to this dilemma. We prove that, under the token-level policy-gradient formulation, this ratio provides an unbiased prefix correction for each token-level gradient term and has strictly lower variance than the full sequence ratio. Building on this insight, we propose CTPO (Cumulative Token Policy Optimization), which combines the cumulative token IS ratio with position-adaptive clipping that scales log-space clip bounds according to the natural $\sqrt{t}$ growth of the cumulative log-ratio. This yields more consistent regularization across token positions. We implement and evaluate CTPO in the tool-integrated reasoning setting on several challenging mathematical reasoning benchmarks, achieving the best average performance across both model scales compared with strong GRPO and GSPO baselines. Code will be available at https://github.com/horizon-llm/CTPO.

cs.LG↗

arXiv2026

Multidimensional semiclassical single- and double-quantum spectroscopy of anharmonic molecular polaritons

We present a general and efficient approach to compute phase-resolved multidimensional spectra of anharmonic molecular polaritons, based on a semiclassical evolution of the molecular Hamiltonian and cavity field in the large-$\mathcal{N}$ limit of many molecules coupled to a confined photonic mode. By systematically expanding the response in both amplitudes and phases of the input fields, our method enables a transparent and computationally simple construction of phase-cycled two-dimensional single- and double-quantum polariton spectra from the underlying nonlinear signal components. Here, phase cycling acts as an analogue of phase matching with oblique pulses, allowing for the isolation of the contributing nonlinear pathways in Liouville space. We specialize to vibrational polaritons and benchmark the method through direct comparison with experimentally measured single-quantum spectra, providing an explanation for the longstanding puzzle of the polariton bleach effect observed at short waiting times. Further, we show how the imprint of various types of anharmonicities on the double-excitation manifold can be directly probed and analyzed through double-quantum coherence spectroscopy. Taken together, our results establish a practical and powerful framework for the modeling and interpretation of nonlinear spectroscopic experiments on strongly coupled light-matter platforms and for guiding the design of cavity-enhanced molecular platforms.

quant-ph↗

arXiv2026

PixelDiT: Pixel Diffusion Transformers for Image Generation

Latent-space modeling has been the standard for Diffusion Transformers (DiTs). However, it relies on a two-stage pipeline where the pretrained autoencoder introduces lossy reconstruction, leading to error accumulation while hindering joint optimization. To address these issues, we propose PixelDiT, a single-stage, end-to-end model that eliminates the need for the autoencoder and learns the diffusion process directly in the pixel space. PixelDiT adopts a fully transformer-based architecture shaped by a dual-level design: a patch-level DiT that captures global semantics and a pixel-level DiT that refines texture details, enabling efficient training of a pixel-space diffusion model while preserving fine details. PixelDiT achieves 1.61 FID on ImageNet 256 and 1.81 FID on ImageNet 512, surpassing existing pixel generative models. We further extend PixelDiT to text-to-image generation and pretrain it at the 10242resolution in pixel space. It achieves 0.74 on GenEval and 83.5 on DPG-bench, approaching the best latent diffusion models. Code: https://github.com/NVlabs/PixelDiT

cs.CV↗

arXiv2026

Tidal Love numbers for regular black holes

Tidal Love numbers (TLNs) characterize the response of compact objects to external tidal fields and vanish for classical Schwarzschild and Kerr black holes in general relativity. Nonvanishing TLNs therefore provide a potential observational window into beyond-classical physics. In this work, we present a unified and fully analytic study of the TLNs of three representative classes of regular black holes -- the Bardeen black hole, the black hole with sub-Planckian curvature, and the black hole arising in asymptotically safe gravity -- under scalar, vector, and axial gravitational perturbations. Employing a Green's function method combined with systematic perturbative expansions, we show that TLNs of regular black holes are generically nonzero and exhibit strong model and mode dependence. In many cases, higher-order corrections develop logarithmic scale dependence, closely resembling renormalization-group running in quantum field theory and revealing a scale-dependent tidal response absent in classical black holes. Our analysis demonstrates that the internal structure of regular black holes, including de Sitter or Minkowski cores and quantum-gravity-inspired modifications, leaves distinct fingerprints in their tidal properties. These results provide a comparative theoretical benchmark for assessing regular black-hole models and establish a basis for future phenomenological and observational studies with gravitational-wave detectors.

gr-qc↗

arXiv2026

Accelerating 4D Hyperspectral Imaging through Physics-Informed Neural Representation and Adaptive Sampling

High-dimensional hyperspectral imaging (HSI) enables the visualization of ultrafast molecular dynamics and complex, heterogeneous spectra. However, applying this capability to resolve spatially varying vibrational couplings in two-dimensional infrared (2DIR) spectroscopy, a type of coherent multidimensional spectroscopy (CMDS), necessitates prohibitively long data acquisition, driven by dense Nyquist sampling requirements and the need for extensive signal accumulation. To address this challenge, we introduce a physics-informed neural representation approach that efficiently reconstructs dense spatially-resolved 2DIR hyperspectral images from sparse experimental measurements. In particular, we used a multilayer perceptron (MLP) to model the relationship between the sub-sampled 4D coordinates and their corresponding spectral intensities, and recover densely sampled 4D spectra from limited observations. The reconstruction results demonstrate that our method, using a fraction of the samples, faithfully recovers both oscillatory and non-oscillatory spectral dynamics in experimental measurement. Moreover, we develop a loss-aware adaptive sampling method to progressively introduce potentially informative samples for iterative data collection while conducting experiments. Experimental results show that the proposed approach achieves high-fidelity spectral recovery using only $1/32$ of the sampling budget, as opposed to exhaustive sampling, effectively reducing total experiment time by up to 32-fold. This framework offers a scalable solution for accelerating any experiments with hypercube data, including multidimensional spectroscopy and hyperspectral imaging, paving the way for rapid chemical imaging of transient biological and material systems.

eess.IV↗

arXiv2026

Quasinormal modes and grey-body factors of axial gravitational perturbations of regular black holes in asymptotically safe gravity

In this paper, we present a detailed study of axial gravitational perturbations of the regular black hole solution in asymptotically safe gravity, as proposed in \cite{Bonanno:2023rzk}. We analyze the quasinormal mode (QNM) spectrum of this black hole using two numerical techniques: the Bernstein spectral method and the asymptotic iteration method (AIM). These approaches allow us to compute QNM frequencies with high accuracy, even for higher overtones. Our results show that the fundamental mode is only weakly affected by the deviation parameter, whereas notable deviations from the Schwarzschild case emerge for higher overtones. Additionally, we examine the correspondence between grey-body factors and QNMs using the sixth-order WKB approximation, finding excellent agreement, especially for larger multipole numbers l.

gr-qc↗

arXiv2026

Nonreciprocal superradiant quantum phase transition induced by the magnon Kerr effect

Recently, proposals for realizing a nonreciprocal superradiant quantum phase transition (SQPT) have been put forward, based on either nonreciprocal interactions between two spin ensembles or the Sagnac-Fizeau shift in a spinning cavity. However, experimental implementation of such a nonreciprocal SQPT remains challenging. This motivates the search for new mechanisms capable of producing a nonreciprocal SQPT. Here, we propose an alternative approach to realize a nonreciprocal SQPT, induced by the magnon Kerr effect (MKE), in a cavity magnonic system, where magnons in a yttrium iron garnet (YIG) sphere are coupled to cavity photons. The MKE coefficient is positive ($K>0$) when the bias magnetic field is aligned along the crystallographic axis [100], but negative ($K<0$) when aligned along the axis [110]. We show that the steady-state phase diagram for $K > 0$ differs markedly from that for $K < 0$. This contrast is the origin of the nonreciprocal SQPT. By further studying the steady-state magnon occupation and its fluctuations versus the parametric drive strength, we demonstrate that the SQPT becomes nonreciprocal, characterized by distinct critical thresholds for $K > 0$ and $K < 0$. Moreover, we introduce a bidirectional contrast ratio to quantify this nonreciprocal behavior. Our work provides a new mechanism for realizing the nonreciprocal SQPT, with potential applications in designing nonreciprocal quantum devices.

quant-ph↗

arXiv2026

Quantum Coulomb drag signatures of Majorana bound states

Majorana bound states (MBSs), with their non-Abelian statistics and topological protection, are key candidates for fault-tolerant quantum computation. However, their unambiguous identification in solid-state systems remains a fundamental challenge. Here, we present a theoretical study demonstrating that drag transport in a capacitively coupled double quantum dot system offers a robust and nonlocal probe of weakly coupled MBSs. Using the master equation approach, we investigate both steady-state and transient dynamics and uncover a distinctive signature of MBSs, namely the emergence of pronounced split peaks in the drag transconductance, directly linked to inter-MBS coupling. We further show that the dynamics of quantum coherence is correlated with the emergence and enhancement of MBS-induced split peaks in the drag transconductance. A comparative analysis with trivial subgap states reveals key differences, that is, MBS-induced transconductance peaks are symmetric and exhibit characteristic splitting, while trivial-state features are generally asymmetric and lack such robust splitting behavior. These findings establish experimentally accessible criteria for distinguishing MBSs from trivial subgap states and provide a practical framework for probing Majorana physics through nonlocal transport.

cond-mat.mes-hall↗

arXiv2026

PhysAlign: Physics-Coherent Image-to-Video Generation through Feature and 3D Representation Alignment

Video Diffusion Models (VDMs) offer a promising approach for simulating dynamic scenes and environments, with broad applications in robotics and media generation. However, existing models often generate temporally incoherent content that violates basic physical intuition, significantly limiting their practical applicability. We propose PhysAlign, an efficient framework for physics-coherent image-to-video (I2V) generation that explicitly addresses this limitation. To overcome the critical scarcity of physics-annotated videos, we first construct a fully controllable synthetic data generation pipeline based on rigid-body simulation, yielding a highly-curated dataset with accurate, fine-grained physics and 3D annotations. Leveraging this data, PhysAlign constructs a unified physical latent space by coupling explicit 3D geometry constraints with a Gram-based spatio-temporal relational alignment that extracts kinematic priors from video foundation models. Extensive experiments demonstrate that PhysAlign significantly outperforms existing VDMs on tasks requiring complex physical reasoning and temporal stability, without compromising zero-shot visual quality. PhysAlign shows the potential to bridge the gap between raw visual synthesis and rigid-body kinematics, establishing a practical paradigm for genuinely physics-grounded video generation. The project page is available at https://physalign.github.io/PhysAlign.

cs.CV↗

arXiv2026

Logarithmic Regret for Online KL-Regularized Reinforcement Learning

Recent advances in Reinforcement Learning from Human Feedback (RLHF) have shown that KL-regularization plays a pivotal role in improving the efficiency of RL fine-tuning for large language models (LLMs). Despite its empirical advantage, the theoretical difference between KL-regularized RL and standard RL remains largely under-explored. While there is a recent line of work on the theoretical analysis of KL-regularized objective in decision making \citep{xiong2024iterative, xie2024exploratory,zhao2024sharp}, these analyses either reduce to the traditional RL setting or rely on strong coverage assumptions. In this paper, we propose an optimism-based KL-regularized online contextual bandit algorithm, and provide a novel analysis of its regret. By carefully leveraging the benign optimization landscape induced by the KL-regularization and the optimistic reward estimation, our algorithm achieves an $\mathcal{O}\big(η\log (N_{\mathcal R} T)\cdot d_{\mathcal R}\big)$ logarithmic regret bound, where $η, N_{\mathcal R},T,d_{\mathcal R}$ denote the KL-regularization parameter, the cardinality of the reward function class, number of rounds, and the complexity of the reward function class. Furthermore, we extend our algorithm and analysis to reinforcement learning by developing a novel decomposition over transition steps and also obtain a similar logarithmic regret bound.

cs.LG↗

arXiv2026

Magnon squeezing near a quantum critical point in a cavity-magnon-qubit system

Preparing magnon nonclassical states is a central topic in the study of quantum magnonics. Here we propose to generate magnon squeezed states in a hybrid cavity-magnon-qubit system by engineering an effective Rabi-type magnon-qubit interaction. This is achieved by adiabatically eliminating the cavity mode and driving the qubit with two microwave fields, of which the driving frequencies and amplitudes are properly selected. By operating the system around the critical point associated with the ground-state superradiant phase transition in the normal phase, a magnon parametric amplification-like interaction is induced, leading to a dynamical magnon squeezing. We further analyze the effects of the dissipation, dephasing, and thermal noise on the magnon squeezing. Our results indicate that a moderate degree of squeezing can be produced using currently available parameters in the experiments.

quant-ph↗

arXiv2026

Retrieval Augmented Generation of Literature-derived Polymer Knowledge: The Example of a Biodegradable Polymer Expert System

Polymer literature contains a large and growing body of experimental knowledge, yet much of it is buried in unstructured text and inconsistent terminology, making systematic retrieval and reasoning difficult. Existing tools typically extract narrow, study-specific facts in isolation, failing to preserve the cross-study context required to answer broader scientific questions. Retrieval-augmented generation (RAG) offers a promising way to overcome this limitation by combining large language models (LLMs) with external retrieval, but its effectiveness depends strongly on how domain knowledge is represented. In this work, we develop two retrieval pipelines: a dense semantic vector-based approach (VectorRAG) and a graph-based approach (GraphRAG). Using over 1,000 polyhydroxyalkanoate (PHA) papers, we construct context-preserving paragraph embeddings and a canonicalized structured knowledge graph supporting entity disambiguation and multi-hop reasoning. We evaluate these pipelines through standard retrieval metrics, comparisons with general state-of-the-art systems such as GPT and Gemini, and qualitative validation by a domain chemist. The results show that GraphRAG achieves higher precision and interpretability, while VectorRAG provides broader recall, highlighting complementary trade-offs. Expert validation further confirms that the tailored pipelines, particularly GraphRAG, produce well-grounded, citation-reliable responses with strong domain relevance. By grounding every statement in evidence, these systems enable researchers to navigate the literature, compare findings across studies, and uncover patterns that are difficult to extract manually. More broadly, this work establishes a practical framework for building materials science assistants using curated corpora and retrieval design, reducing reliance on proprietary models while enabling trustworthy literature analysis at scale.

cs.CE↗

arXiv2025

$\bf A^2$-robust superradiant phase transition in hybrid qubit-cavity optomechanics

The $\mathbf{A}^2$ term presents a fundamental challenge to realizing the superradiant phase transition (SPT) in cavity quantum electrodynamics. Here, we propose a hybrid quantum system enabling SPT regardless of the presence of the $\mathbf{A}^2$ term. The system consist of a qubit, a mechanical mode, and an optical cavity, where the qubit and mechanical mode constitute a quantum Rabi model, while the mechanical mode and cavity form an optomechanical system. Crucially, the auxiliary cavity introduces a switchable $\mathbf{A}^2$ term that effectively counteracts or even fully eliminates the original $\mathbf{A}^2$ effect. This allows controllable observation of SPT, diagnosed via the second-order equal-time correlation function $g^{(2)}(0)$ of phonons. Furthermore, the auxiliary cavity exponentially reduces the critical coupling strength, significantly relaxing experimental requirements. In addition, we show that phonons in the normal phase display bunching, but coherent in the superradiant phase. Interestingly, higher-order squeezing is found in both phases, with near-perfect higher-order squeezing achieved at the SPT point, establishing it as a probe for SPT behavior. Our work demonstrates that hybridizing optomechanics and cavity quantum electrodynamics provides a promising route to accessing SPT physics in the presence of the $\mathbf{A}^2$ term.

quant-ph↗

arXiv2025

Exponentially Enhanced Tripartite Coupling in Quantum Nonlinear Magnonics

Strong and controllable tripartite interactions play a pivotal role in quantum information and nonlinear quantum optics, yet challenging to realize. In this work, we propose a hybrid system consisting of a nitrogen-vacancy (NV) center coupled to Kerr magnons (magnons with Kerr nonlinearity) in two yttrium-iron-garnet spheres. By adiabatically eliminating the ground state of the NV qutrit in the dispersive regime, an effective tripartite interaction among magnons and an NV qubit encoded in its excited states is obtained. In the strong driving limit, Kerr magnons can be linearized and give rise to degenerate parametric amplification for squeezing magnons. As a result, both the tripartite interaction and cooperativity are exponentially enhanced twice, which is about $\exp(ξ)$ times than schemes only involving single-squeezing. Hence, our proposal is more experimentally feasible because modest squeezing parameter is sufficient. With this amplified tripartite coupling strength, the system dynamics are greatly accelerated, leading to fast generation of tripartite entanglement. In addition, noise-resilient perfect magnon blockade can be achieved, well predicted by both the analytical approach and numerical simulation with quantum master equation. Our results suggest that the NV center represents a promising interface for engineering many-body interactions in quantum magnonics, offering a versatile platform for exploring fundamental quantum phenomena such as entanglement and correlations.

quant-ph↗

arXiv2025

Nonreciprocal Bistability in Coupled Nonlinear Cavity Magnonics

We propose a coupled nonlinear cavity-magnon system, consisting of two cavities, a second-order nonlinear element, and a yttrium-iron-garnet (YIG) sphere that supports Kerr magnons, to realize the sought-after highly tunable nonreciprocity. We first derive the critical condition for switching between reciprocity and nonreciprocity in the absence of magnon driving, and then numerically demonstrate that strong magnonic nonreciprocity can be achieved by violating this critical condition. When magnons are driven, we show that strong magnonic nonreciprocity can also be attained even within the critical condition. Compared to previous studies, the introduced nonlinear element not only relaxes the critical condition in both the weak and strong coupling regimes, but also offers an alternative means to tune magnonic nonreciprocity. Our work provides a promising avenue for realizing highly tunable nonreciprocal devices based on Kerr magnons.

quant-ph↗