arXiv subjects

Zhiyuan Li

Publications and source records attributed to Zhiyuan Li.

At least 19 recordsLinked to original sources

arXiv2026

Bloch's conjecture for (anti-)symplectic automorphisms of K3 surfaces

We prove Bloch's conjecture for (anti-)symplectic automorphisms of smooth projective complex K3 surfaces of Picard number two. Together with the recent result for abelian surfaces and the known K3 cases, this settles the remaining case of Bloch's conjecture for (anti-)symplectic automorphisms of smooth projective complex surfaces with trivial canonical bundle. More generally, we prove Bloch's conjecture for (anti-)symplectic derived autoequivalences of projective K3 surfaces whose Picard number is at least three, or is two with anisotropic Néron--Severi lattice. The key new result is the congruence property of the subgroup generated by reflective involutions in $O(U\oplus N)$ for anisotropic even lattices $N$ of signature $(1,1)$. Combined with the local generation theorem and Chow-theoretic criterion of Li--Yu--Zhang, this gives the required reflective decompositions and their consequences for zero-cycles.

math.AG↗

arXiv2026

Special cycles and Picard groups of Baily--Borel compactifications

In this paper, we answer two basic questions concerning algebraic cycles on orthogonal Shimura varieties attached to lattices. First, we prove that Kudla's special cycles and Heegner cycles have the same rational span in every codimension. This fills a gap in the codimension-one argument of \cite[Remark 2.4]{BLMM17}, pointed out by Bruinier--Zuffetti \cite{BZ26}. Second, in dimension at least ten, we prove that the rational Picard group of the Baily--Borel compactification is generated by the Hodge class. The proof of the second result reduces, via isotropic induction on the discriminant form, to the K3-type case treated in our previous work.

math.AG↗

arXiv2026

The Influence of Hubble Expansion on the Future Merger of the Milky Way and Andromeda

Most existing idealized simulations of galaxy mergers neglect Hubble expansion, as the local Hubble flow is curved by overdensities and the expansion effect is weak at these scales. However, even in the gravity-dominated Local Group (LG), the expansion itself still persists, and it has been shown important in some LG-related studies. Here we take the future merger of the Milky Way (MW) and Andromeda (M31) as a case study to investigate the differences between simulations with and without Hubble expansion. We perform a suite of idealized N-body simulations of the MW--M31 system under pure Newtonian, $Λ$CDM, and $w_0w_a$CDM scenarios, testing different masses and proper motions of galaxies informed by current observational constraints. At low M31 tangential velocities, especially in cases where the two galaxies merge within a Hubble time, the expansion effect is negligible; while at high velocities (>66.5 km/s), the $Λ$CDM simulations introduce a noticeable delay in the merger time compared to the Newtonian ones and may even alter the final fate of the system, but the caveat here is that the isolated-pair idealization and the assumed expansion history break down. The delay is also present but less pronounced in the $w_0w_a$CDM runs. When M33 and the Large Magellanic Cloud are included, the differences in merger times are reduced. In summary, Hubble expansion can play a noticeable role for galaxy mergers with high tangential velocities, but its actual influence on the MW--M31 merger depends on the precise masses and proper motions to be determined in the future.

astro-ph.GA↗

arXiv2026

The Dancing Bonfire: An Unusual Persistent X-ray Source in M81 Exhibiting Recurrent Luminous Flares

Short-timescale X-ray variability provides a direct probe of unstable accretion onto compact objects, yet extragalactic X-ray binaries (XRBs) with recurrent flaring remain largely unexplored. Here we present a detailed X-ray study of M81F4 (2CXO~J095524.2+690957), a persistently luminous X-ray source towards M81 that exhibits recurrent short X-ray flares. Based on multi-mission X-ray observations spanning two decades, it is found that M81F4 remains persistently bright at $L_{0.5-8.0\rm~keV}\sim10^{38}-10^{39}\rm~erg~s^{-1}$, while producing flares lasting a few hundred seconds and reaching peak luminosities of a few $10^{39}\rm~erg~s^{-1}$, with an observed mean (minimum) recurrence time of $\sim0.5$ day ($\sim$6 ks). The stacked flare profile shows a rapid rise followed by a slower decay, but no significant hardness evolution. Spectral analysis reveals that both the flare and non-flare states are dominated by a hard power-law continuum, with marginal evidence of highly ionized Fe lines in the non-flare state spectrum. These properties disfavor classical Type I and Type II bursts and have no close analog among known XRBs. We favor a magnetized neutron star accreting from an intermediate mass donor at a persistently high rate with the flares arising from instabilities at the disk-magnetosphere boundary. If confirmed, M81F4 might represent a hidden population of XRBs in which sustained, magnetically-regulated accretion can lead to brief episodes of super-Eddington emission.

astro-ph.HE↗

arXiv2026

Inverse heat source problem from terminal data on a set of low Hausdorff dimension

We study the recovery of a spatial source in the heat equation from terminal observations on a compact interior set $E$ with positive $β$ dimensional Hausdorff content for some $β>d-1$. The prescribed temporal forcing vanishes on a nontrivial interval immediately preceding the observation time. Under a spectral Sobolev prior and a quantitative noncancellation condition on the terminal spectral multipliers, we establish conditional logarithmic stability. For scattered observations on such sets, we derive a sampling inequality under a quasi uniformity assumption. Under explicit sampling and independent noise assumptions, we prove a logarithmic convergence rate in probability for the corresponding Tikhonov reconstructions with a data driven parameter choice. Two numerical experiments using a finite candidate approximation of a fixed Hausdorff set assess discrete spectral sampling stability and illustrate the effects of the waiting interval and measurement noise on reconstruction.

math.AP↗

arXiv2026

Temporal Slice Learning for AI-Generated Video Detection with 400$\times$ Fewer FLOPs

Recent advances in generative video models have improved visual realism, making detection increasingly dependent on subtle temporal inconsistencies. However, these signals are weak and spatially localized, while most of the video content is redundant. This creates a fundamental mismatch: existing detectors rely on heavy backbones to model full videos, which leads to high computational cost and limited scalability. In this work, we rethink AI-generated video detection from a representation perspective. Instead of modeling entire videos, we restructure them into compact temporal slice representations that isolate informative temporal evolution. We introduce a lightweight framework using frequency-aware temporal slice consistency learning, where informative spatial rows and columns are aggregated over time to form slice-domain inputs. This reparameterization suppresses redundant appearance while preserving discriminative temporal dynamics, enabling efficient and targeted detection. By constructing slices from both high- and low-frequency regions, our method exposes complementary temporal artifacts, including unstable details and abnormal motion smoothness. We instantiate this idea in a Temporal Slice Consistency Network, which integrates task-driven slice localization, hierarchical positional encoding, and a lightweight Transformer to model cross-slice and cross-frequency dependencies. We further introduce AI-Artist, a new benchmark with artist-curated videos from recent high-fidelity generators, including Seedance and Kling. Experiments on GenVideo and AI-Artist show that our method achieves strong performance while requiring over 400$\times$ fewer model-forward FLOPs and 3.6$\times$ faster model inference than the strong video-based baseline ReStraV (2.9$\times$ faster end to end, including preprocessing), enabling scalable and practical deployment.

cs.CV↗

arXiv2026

NOEMA Observations of the Center of M33: A Parsec-scale Molecular Filament and an Ordered Gas-Star-Formation Sequence

We present NOEMA CO(1-0) observations of the inner $\sim$500 pc of M33 at a resolution of $0.91''\times0.37''$ ($3.7\times1.5$ pc), resolving the compact molecular gas structures and revealing their relation to recent massive star formation. The most prominent structure is a southeastern molecular filament (SE filament) that extends 76 pc along its principal axis, is 23 pc wide, and contains three CO-bright knots. Across eight independent transverse cuts, the adjacent H$α$ ridge is displaced preferentially to one side of the principal CO ridge by a median $10.0^{+0.9}_{-1.4}$ pc. Young OB candidates are more closely associated with H$α$ than with compact CO. The filament shows little velocity change along its length but a resolved gradient across its minor axis. Its elongated morphology, transverse kinematics, and ordered CO-H$α$ distribution are consistent with locally asymmetric molecular-gas compression and subsequent massive star formation. Across the wider field, 23 well-measured compact clumps have a median radius of 2.67 pc and a median velocity dispersion of 2.15 km s$^{-1}$. Their linewidth and apparent virial-parameter distributions show no clear difference between the SE filament and the rest of the field. Although the local spiral potential and pattern speed remain unconstrained, the observations provide a resolved view of a compression-compatible molecular-to-ionized-gas sequence in the center of M33.

astro-ph.GA↗

arXiv2026

LaWAM: Latent World Action Models for Efficient Dynamics-Aware Robot Policies

Vision-Language-Action models (VLAs) leverage large-scale vision-language pretraining for semantic robot control, but often lack explicit foresight into how robot actions change the scene. World-Action Models (WAMs) address this limitation by conditioning policies on predicted futures, yet existing approaches typically rely on computationally expensive video generation with substantial pixel-level redundancy. We present LaWAM, a Latent World Action Model that exposes predictive dynamics to robot policies through compact latent visual subgoals instead of reconstructed future video. At the core of LaWAM is a latent-action-conditioned Latent World Model (LaWM). We obtain LaWM by training a latent action model in the latent space of a pretrained vision foundation model and repurposing its forward decoder to predict future observation features for scene evolution. LaWAM then conditions action generation on these predicted latent visual subgoals to enable dynamics-aware robot control. LaWAM achieves state-of-the-art or competitive success rates (SRs) across LIBERO (98.6% SR), RoboTwin (91.22% SR), and real-world manipulation tasks while retaining low-latency inference. LaWAM runs in 187 ms per action-chunk prediction and achieves up to 24x lower wall-clock latency than pixel-space WAMs.

cs.RO↗

arXiv2026

Rethinking Scientific Discovery in the Agentic Era

Artificial intelligence has advanced scientific discovery, but most AI4Science systems remain fragmented tools that rely on humans to coordinate problem formulation, literature grounding, model use, simulation, validation, and knowledge reuse. This paper presents \textbf{SCION (Scientific Collaborative Innovation with Agentic Organizational Nexus)}, an agentic scientific operating system that acts as an \textbf{organizational nexus}. Through a Science Agent serving as a \textbf{Meta-Harness}, SCION connects scientific tasks, tools, agents, artifacts, and memory, transforming research into an executable, auditable, and reusable operational process. At its core is the \textbf{Research Execution Plan (REP)}, which compiles high-level scientific intent into staged objectives, dependencies, verification checkpoints, tool requirements, expected artifacts, and fallback conditions. SCION further integrates hierarchical multi-agent execution, profile-driven specialization, selective context construction, governed delegation, and layered epistemic memory to support long-horizon scientific work. We formulate discovery under SCION as \textbf{Target-conditioned Inverse Search} and extend it to hidden-target settings through batch active search under finite experimental budgets. Applications in materials analysis, molecule design, and protein or antibody screening, together with experiments on scientific reading, idea generation, molecule generation, and antibody screening, show that SCION outperforms existing autonomous research-agent baselines, especially in decomposition, verification, refinement, and memory reuse. Overall, SCION shifts AI from isolated tools toward a coordinated operational layer for traceable and reusable scientific innovation.

cs.CL↗

arXiv2026

Recovery of a source of anomalous diffusion from space-time sparse data

We study an inverse source problem for a time-fractional sub- and super-diffusive anomalous diffusion equation on the unit disk with sparse boundary observations. We prove the recovery of a general class of $L^2$ sources for anomalous diffusion, with separated variables, from boundary measurements that are sparse in both space and time, using observations at two carefully chosen boundary points on a discrete set of times arbitrarily close to the final time. The analysis relies heavily on the memory effect of the model, properties of complex analysis and the representation of solutions.

math.AP↗

arXiv2026

From Signals to Evidence: Interpretable Underwater Acoustic Target Sensing with Large Language Models - Part II: Passive Detection

Underwater acoustic passive detection is essential to maritime surveillance and intelligent ocean sensing. Existing detection systems typically collapse multichannel array observations into a binary decision or a scalar confidence score, offering limited insight into the physical evidence that supports or contradicts a target declaration. This limitation is acute at low signal-to-noise ratio (SNR), where isolated spectral peaks and transient beamforming responses are inherently ambiguous and physically verifiable targets are characterized by an ensemble of coupled reinforcing cues. Building on the unified from-signals-to-evidence framework introduced in the companion Part I, this paper instantiates the framework for passive target detection. A physics-aware multichannel acoustic encoder encodes the array observations into complementary acoustic evidence tokens through a direction-guided architecture that combines minimum variance distortionless response (MVDR) processing with time-frequency and fine-frequency representations. These tokens are combined with environmental priors and fed into a pretrained large language model (LLM) that evaluates target-present and noise-only hypotheses and generates a structured detection report detailing target presence, bearing, spectral-line locations, received SNR, range, and per-field confidence scores. A physics-parameterized data engine extended to realistic oceanic channels provides traceable supervision at both discrete-label and structured-description levels. Experiments under white-noise and mixed ocean-background conditions demonstrate reliable detection, accurate bearing estimation, and physically interpretable structured reports, validating the feasibility of evidence-grounded underwater acoustic detection.

cs.SD↗

arXiv2026

Copying Under Repetition: From Prefix Matching to Position Hashing

Frontier LLMs can copy random text almost flawlessly, yet they often overrun or truncate long blocks of repeated characters. In repeated strings, matching local content no longer reliably identifies where copying should continue or when it should stop. Nevertheless, we show that a two-layer transformer with $O(\log\log N)$-bit precision and $O(\log N/\log\log N)$ width can copy every string up to a chosen design horizon $N$ and terminate correctly under arbitrary repetition. We use RoPE-style rotations of attention values and outputs (VO-rotations) with small prime periods to encode distances from delimiter anchors modulo those primes. By the Chinese Remainder Theorem, these periodic encodings jointly distinguish all required distances, allowing one retrieval rule to select either the next source token or the end marker. In contrast, we prove a width-independent impossibility result for copying under sequential finite-precision softmax accumulation. It applies to transformers whose only positional input is an additive attention bias that eventually becomes constant, including NoPE and a recursive implementation of ALiBi. Once an identical-token run exceeds the resulting saturation threshold, further repetitions leave the logits unchanged. If this threshold grows sublinearly with $N$, these models cannot both complete a long all-identical copy and stop correctly, a pattern resembling the termination errors of frontier LLMs. In synthetic copying experiments, baseline transformer models fail on long repeated strings even within the training range. Trained VO models, by contrast, generalize beyond the training lengths with near-perfect accuracy up to length $\approx5005$, the selected primes' period product.

cs.LG↗

arXiv2026

Supersingular Tate conjecture for irreducible symplectic varieties of known types: I

We prove that for a supersingular irreducible symplectic variety admitting suitable lifting to characteristic zero has Tate Chow motive if it is of deformation type $K3^{[n]}$, OG6 (with Artin invariant $\neq$ 4), or OG10 (with Artin invariant $\neq$ 12), and has supersingular abelian Chow motive if it is of $\mathrm{Kum}^n$-type (with Artin invariant $\neq$ 3). In particular, any product of those irreducible symplectic varieties satisfies the supersingular Tate conjecture for the whole $\ell$-adic or crystalline cohomology ring.

math.AG↗

arXiv2026

Answer Delivery Under a Shared Token Budget: An Audited Study of Stopping and Finalization

An inference budget is useful only when the system delivers an answer within it. We study a fixed-budget policy combining first-valid-box stopping with a charged finalization cue. On 100 fresh, independently audited GSM8K test questions per model, it improves delivered endpoint accuracy over first-box stopping by 8.0 percentage points for Qwen3-4B-Thinking-2507 and 7.0 for DeepSeek-R1-Distill-Llama-8B. A prelocked repeat on the same questions gives 7.0 and 10.0 points; neither seed shows a gross harm relative to first-box stopping. The point estimates of incremental physical-slot savings are below one percent. A finalization-only ablation separates cue recovery from stopping, and a fixed scalar reader tests extraction as an alternative explanation. The cue is prior art; the contribution is a component and readout analysis with charged injections and verified paths. Results cover two checkpoints, one task, one prompt family, and two fixed seed conditions.

cs.AI↗

arXiv2026

Analysis and approximation for inverse potential problem of space-dependent variable-exponent sub-diffusion in heterogeneous medium

Anomalous transport in complex heterogeneous medium is often characterized by non-Markovian memory effects and spatially varying trapping properties. From the point of view of continuous-time random walks, such heterogeneity can be described by a position-dependent heavy-tailed waiting-time density $q(t \mid x)\sim C(x)t^{-1-α(x)}$, where $0<α(x)<1$ is the local anomalous exponent. This microscopic interpretation naturally leads to the sub-diffusion with a space-dependent variable exponent $α(x)$. Compared with constant-exponent models, the variable-exponent fractional operators generally do not commute with the spatial Laplacian, which makes both the forward and inverse analysis substantially more delicate. This paper investigates the inverse problem of recovering a spatially dependent radiative or potential coefficient in variable-exponent sub-diffusion from terminal observation data. We formulate the inverse problem as a nonlinear fixed-point equation and prove that the associated fixed-point operator is monotone such that the fixed-point iteration converges decreasingly to the unique solution. We further establish a Lipschitz-type conditional stability estimate for the terminal observation map, which also implies uniqueness of the inverse problem for sufficiently large terminal time. For numerical reconstruction, we develop a fully discrete finite element scheme and derive an error estimate for the inverse problem. Numerical simulations are conducted to validate the effectiveness of the method, even in the presence of noisy data.

math.NA↗

arXiv2026

Derived equivalences and isogenies of supersingular abelian varieties

We study (twisted) Fourier-Mukai partners of supersingular abelian varieties. We give a simple proof that any two supersingular abelian varieties of the same dimension at least two are derived isogenous, answering a question raised by Li, Lu, and Tang and generalizing the result of Li and Zou. Moreover, in every dimension greater than two, we construct derived-equivalent supersingular abelian varieties whose corresponding positive powers are never isomorphic. We also give conditions under which derived equivalence between two supersingular principally polarized abelian varieties implies that their underlying abelian varieties are isomorphic. In particular, this implication holds in dimensions at most three.

math.AG↗

arXiv2026

Towards Secure Agent Skills: Architecture, Threat Taxonomy, and Security Analysis

Agent Skills is an emerging open standard that defines a modular, filesystem-based packaging format enabling LLM-based agents to acquire domain-specific expertise on demand. Despite rapid adoption across multiple agentic platforms and the emergence of large community marketplaces, the security properties of Agent Skills have not been systematically studied. This paper presents a systematization of the security of the Agent Skills framework. We define a threat model for the framework and decompose the full lifecycle of an Agent Skill across four phases---Creation, Distribution, Deployment, and Execution---identifying the structural attack surface each phase introduces. Building on this lifecycle analysis, we construct a threat taxonomy comprising seven categories and nineteen scenarios organized across three attack layers, grounded in both architectural analysis and real-world evidence. Each scenario is annotated with whether the exposure follows from the specification or from the latitude the specification leaves to individual runtimes, with the strength of the public evidence supporting it, and with the feasibility of its mitigation. We map seven documented security incidents onto the taxonomy. Based on these findings, we discuss defense directions for each threat category, identify open research challenges, and provide actionable recommendations for stakeholders. The most severe threats arise from structural properties of the framework itself: the absence of a data-instruction boundary, a single-approval persistent trust model, and a capability declaration mechanism to which the specification attaches no enforcement semantics. None of these is addressable through incremental mitigations alone.

cs.CR↗

arXiv2026

NeoHorse-1: Towards Recursive Self-Improvement via Agentic Post-Training with Routing Harness

Recursive self-improvement (RSI) requires a concrete mechanism through which an AI system observes its capabilities and converts that evidence into the next round of learning. We present NeoHorse-1, a family of agent-native models developed to explore this path through agentic post-training. Our system combines a heterogeneous model pool with intelligent routing, recording the predicted capability demand, selected service tier, and subsequent interaction for each user turn. These records are converted into training examples that preserve interleaved reasoning, tool calls, and harness context, and are admitted through structural validation, six-dimensional semantic evaluation, and subscene-level labeling. Routing signals organize supervised fine-tuning into a three-stage curriculum and extend to routing-guided on-policy distillation, where a teacher supervises student-generated responses under the same progression. Capability-guided allocation then converts evaluation feedback into the next training mixture, closing an evaluation-selection-update loop in which what the system learns to do shapes what it learns from next. Across eleven benchmarks covering harness-based agents, tool use, coding, and instruction following, post-training raises the macro-average from 58.94 to 64.87 at 4B and from 65.60 to 69.04 at 9B, substantially narrowing the aggregate gap between the post-trained 4B model and the 9B base model. We further adapt NeoHorse-1-4B into NeoHorse-Jev-4B, a prefill-only model that returns structured decisions and probabilities for agent workflows. NeoHorse-1 provides an initial prototype of this feedback-driven process and a path toward harness-mediated RSI across successive iterations.

cs.CL↗