Search arXiv⌕ Search

arXiv subjects

Zhijun Wang

Publications and source records attributed to Zhijun Wang.

At least 19 recordsLinked to original sources
arXiv2026

An Inspectable LLM Council for Multi-Model Research Answer Aggregation

Long-form research questions often require a single answer that combines details, reconciles conflicting estimates, and retains qualifications spread across several large language model (LLM) outputs. We present an inspectable LLM Council that separates synthesis into two stages: an analyst converts three independent candidate answers into a structured state, and a writer composes the final response from that state. We evaluate the workflow under closed-book conditions on 100 tasks from the Deep Research Accuracy, Completeness, and Objectivity (DRACO) benchmark, retaining 600 final answers, 100 analyst states, and 720 whole-answer judgments across the main and supplementary experiments. Under the primary rubric judge, Council scores 70.98, exceeding GPT and Gemini by 6.18 and 8.34 points; its 0.47-point difference from Claude has a 95% paired interval of $[-0.69,1.61]$. In the full supplementary comparison, Council scores 1.87 points above direct fusion (95% paired interval $[0.33,3.48]$) and covers 51.1% of positive rubric criteria covered by exactly one candidate, compared with 43.7% for direct fusion. On an exploratory 68-task subset with matching returned model labels, direct fusion scores higher, so the full comparison cannot isolate the analyst state's effect. Traced cases show how estimates, provenance qualifications, and selection decisions pass through the saved state into final answers.

cs.CL↗
arXiv2026

RexDrug: Reliable Multi-Drug Combination Extraction through Reasoning-Enhanced LLMs

Automated Drug Combination Extraction (DCE) from large-scale biomedical literature is crucial for advancing precision medicine and pharmacological research. However, existing relation extraction methods primarily focus on binary interactions and struggle to model variable-length n-ary drug combinations, where complex compatibility logic and distributed evidence need to be considered. To address these limitations, we propose RexDrug, an end-to-end reasoning-enhanced relation extraction framework for n-ary drug combination extraction based on large language models. RexDrug adopts a two-stage training strategy. First, a multi-agent collaborative mechanism is utilized to automatically generate high-quality expert-like reasoning traces for supervised fine-tuning. Second, reinforcement learning with a multi-dimensional reward function specifically tailored for DCE is applied to further refine reasoning quality and extraction accuracy. Extensive experiments on the DrugComb dataset show that RexDrug consistently outperforms state-of-the-art baselines for n-ary extraction. Additional evaluation on the DDI13 corpus confirms its generalizability to binary drugdrug interaction tasks. Human expert assessment and automatic reasoning metrics further indicates that RexDrug produces coherent medical reasoning while accurately identifying complex therapeutic regimens. These results establish RexDrug as a scalable and reliable solution for complex biomedical relation extraction from unstructured text. The source code and data are available at https://github.com/DUTIR-BioNLP/RexDrug

cs.CL↗
arXiv2026

AdaR: A Framework for Equipping LLMs with Adaptive Reasoning

Mathematical reasoning is a primary indicator of large language models (LLMs) intelligence. However, existing LLMs exhibit failures in robustness and generalization. This paper attributes these deficiencies to spurious reasoning, wherein generated reasoning traces are driven by superficial correlations, leading models to blindly reproduce memorized patterns from the training data. To address this challenge, we propose the AdaR framework to equip LLMs with adaptive reasoning, wherein models establish correct correlations between query templates and reasoning processes. AdaR automatically synthesizes logically equivalent queries by varying variable values and trains models using Reinforcement Learning with Verifiable Rewards (RLVR) to penalize spurious logic while encouraging adaptive logic. To ensure data quality, we extract the problem-solving logic from the original query, generate the corresponding answer through code execution, and then apply a sanity check. Experimental results demonstrate that AdaR achieves substantial improvements in mathematical reasoning while maintaining high data efficiency. Furthermore, even advanced LLMs still exhibit deficiencies in robustness and generalization, which our work effectively mitigates. Our project is available at https://github.com/NJUNLP/AdaR.

cs.AI↗
arXiv2026

Robust structure from motion for aerial-ground images via detector-free feature matching and multi-view track refinement

Integrated 3D reconstruction from aerial-ground images is essential for generating high-precision urban 3D models, yet severe variations in viewpoint, scale, and rotation make robust feature matching highly challenging. To address these limitations, this study introduces a rotation-robust detector-free matching network coupled with multi-view track refinement for incremental Structure from Motion (ISfM). The proposed workflow features four key modules. First, rotation-aware feature extraction replaces traditional convolutions with an Omnidirectional State Space Block (OSS Block) that selectively scans across eight symmetrical directions to model long-range spatial dependencies and synthesize rotation-invariant feature maps. Second, multi-scale attention transformation utilizes quadtree attention to build a hierarchical token pyramid that isolates high-association token regions and discards irrelevant areas, capturing long-range context with linear computational complexity. Third, bi-directional feature matching executes a symmetric coarse-to-fine matching scheme where coarse alignment computes dual-direction Softmax confidence matrices under mutual nearest neighbor constraints, and fine alignment uses a multi-layer perceptron to regress sub-pixel coordinate offsets. Finally, multi-view track refinement employs an integrated indexing structure to evaluate localized spatial proximity and link disjoint sub-tracks to the highest-confidence anchor point, ensuring stable feature repeatability across the ISfM pipeline. By using real aerial-ground datasets, experimental results demonstrate that the proposed method improves AUC at 5° pose error by 93.9% compared with LoFTR and achieves the highest precision in ISfM reconstruction, with the improved accuracy ranging from 27.6% to 32.7%. The proposed method provides a reliable solution for integrated 3D reconstruction of aerial-ground images.

cs.CV↗
arXiv2026

R3S: Refining and Recovering Reinforcement Signals for Multilingual Understanding and Reasoning

Large reasoning models often default to English reasoning when processing non-English questions, yet their performance drops substantially when reasoning in the question language. Even with the same reasoning language, semantically equivalent English and non-English questions still exhibit a clear performance gap. Together, these phenomena reveal two distinct bottlenecks: target-language question understanding and target-language reasoning. Existing methods typically optimize only one of these capabilities. However, simply combining them may not be sufficient to optimize both effectively, as answer correctness alone cannot distinguish failures in question understanding from those in reasoning. We propose R3S, a reinforcement learning framework that disentangles the optimization of the two capabilities. R3S refines translation rewards derived from downstream reasoning accuracy through English-solvability filtering and recovers target-language RLVR signals using self-generated English hints. Together, these designs require neither external model feedback nor external multilingual training data. Experiments across three backbone models and five languages show that R3S improves language-consistent accuracy over the target-language RLVR baseline on MMATH by an average of 10.3 percentage points, while maintaining near-perfect language consistency. Consistent gains on MMLU-ProX further demonstrate its generalization beyond math problems.

cs.CL↗
arXiv2026

Efficient Multilingual Reasoning Transfer via Progressive Code-Switching

Large reasoning models (LRMs) have achieved strong reasoning capabilities in English, yet their performance degrades significantly when required to reason in other languages. A natural solution is to transfer the model's English reasoning ability to target languages. However, existing transfer approaches typically rely on distilled target-language reasoning traces from stronger LRMs or online supervision from external judge models, which are costly and difficult to scale. In this paper, we propose PCS (Progressive Code-Switching), a more efficient transfer framework that requires only lightweight translation without any stronger model for distillation or judging. PCS first constructs code-switched reasoning traces by translating a subset of English reasoning steps into the target language, and uses them to initialize the model's code-switching ability via supervised fine-tuning. It then applies reinforcement learning with a step-level language consistency curriculum, progressively raising the target-language ratio until the model reasons entirely in the target language. This progressive design provides a smooth transfer path that avoids the instability and performance degradation commonly observed when directly enforcing target-language reasoning. Experiments on multiple benchmarks and five typologically diverse languages show that PCS substantially narrows the performance gap between target-language and English reasoning, yielding more language-consistent reasoning while maintaining competitive accuracy.

cs.CL↗
arXiv2026

A Possible Triple Formation Scenario of Binary Black Hole Merge With One In Pair-instability Supernova Mass Gap

Observations of binary black hole (BBH) mergers detected by LIGO -- such as GW170729, GW190620, GW190706, GW230107, GW230820, and GW230928 -- feature high effective spins and primary black holes that fall squarely into the pair-instability supernova (PISN) mass gap ($\sim 45-130 \, M_{\odot}$). These events pose a significant challenge to standard stellar and binary evolution theories. To address this, we propose an isolated hierarchical triple stellar evolution channel. In this framework, tidal synchronization in tight inner binaries drives chemically homogeneous evolution (CHE), entirely bypassing giant expansion. A subsequent triple common envelope (TCE) evolution, triggered by the tertiary companion, rapidly drives the inner BBH to coalescence. Our model can provide a detailed evolutionary pathway that elegantly reproduces the properties of these GWs, such as GW190706. Assuming a low-metallicity environment ($Z = 0.001$), our framework predicts a volumetric merger rate of approximately $0.011 \, \mathrm{Gpc}^{-3}\mathrm{yr}^{-1}$ at $z \approx 0.68$, accounting for $22\%$ of the empirical rate for this mass regime in the GWTC-4 catalog. This study demonstrates that primordial triple interactions are a highly efficient avenue for populating the PISN mass gap.

astro-ph.HE↗
arXiv2026

Seeing Is No Longer Believing: Frontier Image Generation Models, Synthetic Visual Evidence, and Real-World Risk

Frontier image generation has moved from artistic synthesis toward synthetic visual evidence. Systems such as GPT Image 2, Nano Banana Pro, Nano Banana 2, Nano Banana 2 Lite, Grok Imagine Image Quality, Qwen Image 2.0 Pro, and Seedream 5.0 Lite combine photorealistic rendering, readable typography, reference consistency, editing control, and in several cases reasoning or search-grounded image construction. These capabilities create large benefits for design, education, accessibility, and communication, yet they also weaken one of society's most common trust shortcuts: the belief that a plausible picture is a reliable record. This paper provides a source-grounded technical and policy analysis of synthetic visual risk. We first summarize the public capabilities of recent image models, then analyze public incidents involving fake crisis images, celebrity and public-figure imagery, medical scans, forged-looking documents, synthetic screenshots, phishing assets, and market-moving rumors. We introduce a capability-weighted risk framework that links model affordances to real-world harm in finance, medicine, news, law, emergency response, identity verification, and civic discourse. Our findings show that risk is driven less by photorealism alone than by the convergence of realism, legible text, identity persistence, fast iteration, and distribution context. We argue for layered control: model-side restrictions, cryptographic provenance, visible labeling, platform friction, sector-grade verification, and incident response. The paper closes with practical recommendations for model providers, platforms, newsrooms, financial institutions, healthcare systems, legal organizations, regulators, and ordinary users.

cs.CL↗
arXiv2026

The Rise of Verbal Tics in Large Language Models: A Systematic Analysis Across Frontier Models

As Large Language Models (LLMs) continue to evolve through alignment techniques such as Reinforcement Learning from Human Feedback (RLHF) and Constitutional AI, a growing and increasingly conspicuous phenomenon has emerged: the proliferation of verbal tics--repetitive, formulaic linguistic patterns that pervade model outputs. These range from sycophantic openers ("That's a great question!", "Awesome!") to pseudo-empathetic affirmations ("I completely understand your concern", "I'm right here to catch you") and overused vocabulary ("delve", "tapestry", "nuanced"). In this paper, we present a systematic analysis of the verbal tic phenomenon across eight state-of-the-art LLMs: GPT-5.5, Claude Opus 4.8, Gemini 3.1 Pro, Grok 4.3, Doubao-Seed-2.1-pro, Kimi K2.6, DeepSeek V4 Pro, and GLM-5.2. Utilizing a custom evaluation framework for standardized API-based evaluation, we assess 10,000 prompts across 10 task categories in both English and Chinese, yielding 160,000 model responses. We introduce the Verbal Tic Index (VTI), a composite metric quantifying tic prevalence, and analyze its correlation with sycophancy, lexical diversity, and human-perceived naturalness. Our findings reveal significant inter-model variation: Gemini 3.1 Pro exhibits the highest VTI (0.590), while DeepSeek V4 Pro achieves the lowest (0.295). We further demonstrate that verbal tics accumulate over multi-turn conversations, are amplified in subjective tasks, and show distinct cross-lingual patterns. Human evaluation (N = 120) confirms a strong inverse relationship between sycophancy and perceived naturalness (r = -0.87, p < 0.001). These results underscore the "alignment tax" of current training paradigms and highlight the urgent need for more authentic human-AI interaction frameworks.

cs.CL↗
arXiv2026

Entropy-Driven Structural Phase Transition in Nb$_3$Cl$_8$ via Density Functional Theory and an Effective Model

As a prototypical flat-band cluster Mott insulator on an effective triangular lattice, Nb$_3$Cl$_8$ is a potential candidate for hosting a quantum spin liquid (QSL) state. Nevertheless, a first-order structural phase transition around 90K transforms the high-temperature paramagnetic $α$ phase into the low-temperature nonmagnetic $β$ phase, suppressing the candidate QSL regime of the $α$ phase. To clarify the microscopic origin of this transition, we combine first-principles calculations with an extended Hubbard model to construct a unified free-energy framework. This framework reveals that the transition is jointly driven by phonon and spin entropy: the $α$ phase is stabilized by softer phonons and larger paramagnetic spin entropy, whereas the $β$ phase is favored by interlayer dimerization, which hardens the phonons and quenches the spin entropy through singlet formation. Furthermore, by evaluating the pressure-dependent generalized enthalpy, we provide a thermodynamic explanation for the suppression of the transition under c-axis uniaxial pressure, where stabilizing the $α$ phase may allow the candidate QSL regime of the $α$ phase to be explored at low temperatures.

cond-mat.mtrl-sci↗
arXiv2026

RGLD: Randomized Global-Local Density Estimation for Tabular Anomaly Detection

Unsupervised tabular anomaly detection requires methods that are accurate, robust across heterogeneous datasets, and computationally efficient. Classical statistical detectors are often efficient, but they usually rely on a fixed data view and a single notion of abnormality. Deep anomaly detectors can learn more flexible scoring functions, but they are substantially slower and difficult to tune in unsupervised settings due to the lack of a reliable supervisory signal. We propose RGLD, a randomized global-local density estimator for efficient unsupervised tabular anomaly detection. RGLD combines a global random-feature density branch, which identifies samples in broadly low-density regions, with a local neighbor branch, which detects samples that are weakly supported by nearby observations. Both branches operate over feature-bagged randomized views, allowing RGLD to expose anomaly evidence that may be hidden in any single representation. We conduct experiments on 47 tabular datasets against 23 statistical and deep anomaly detection baselines under fully unsupervised setting. RGLD achieves the strongest dataset-level AUROC performance, ranking 1st in dataset wins, and ranks 2nd in AUPRC wins. RGLD is also faster than all evaluated deep detectors, achieving 50x-580x speedups, and remains competitive with statistical methods in runtime, yielding a favorable accuracy-efficiency tradeoff.

cs.LG↗
arXiv2026

Discovery of d-orbital order in Tb2CoAl4Ge2

Orbital order describes a quantum state where occupied orbitals line up in a periodic pattern. While orbital physics plays a fundamental and universal role in strongly correlated electron systems, the existence and particularly the band structure fingerprint of orbital order remain a long-standing mystery. Here, we report the discovery of rare earth 5d-orbital order developed by the surface states of intermetallic compound Tb2CoAl4Ge2. Angle-resolved photoemission spectroscopy reveals characteristic nematic features like Fermi surface deformation and band split. These experimental observations can be described by a ferro-orbital order term in the mean-field Hamiltonian. The structural and magnetic origin of such order is excluded by systematic high-resolution neutron powder diffraction and scanning tunnelling microscopy measurements. Our results provide strong evidence for a pure surface orbital order scenario avoiding complications from structural distortion as in colossal magnetoresistance manganites, magnetic order as in iron-based superconductors, and charge transfer p-orbital order in cuprates.

cond-mat.str-el↗
arXiv2026

Coexistence of topologically nontrivial and trivial insulating states in topological Anderson Chern insulator

The interplay between disorder and topology has become a central theme in condensed matter physics. Disorder can not only destroy topological phases but also induce them, as exemplified by the topological Anderson insulator (TAI). Here we show that, in close analogy, disorder can drive the clean-limit, time-reversal-broken(T-broken) quantum spin Hall state of ferromagnetic(FM) monolayer MnBi4Te7 into a quantum anomalous Hall phase, which was called topological Anderson Chern insulator (TACI). Using density functional theory (DFT) and nonequilibrium Green's func tion (NEGF) calculations in the presence of disorder, we identify disorder induced phases-including T-broken TAI, TACI, Normal insulator, etc., then construct a comprehensive phase diagram. To discriminate multiple phases in the strong disorder regime, we further use the density of states computed within the self-consistent Born approximation (SCBA), which in particular distinguishes gapped and ungapped topological phases. We find that the two effective band inversions of Hamiltonian are suppressed at distinct critical disorder strengths; the survival of a single inversion over a finite disorder window stabilizes the TACI. Remarkably, at strong disorder, we further propose a zero Hall plateau insulating state characterized by an insulating bulk and edge channels subject to diffusive scattering that can coexist with the TACI. This behavior is distinct from a conventional band-gap Chern insulator and provides a clear experimental signature.

cond-mat.dis-nn↗
arXiv2026

IRSSG: An Open-Source Software Package for Spin Space Groups

We present an open-source software package IRSSG for investigating magnetic systems with spin space groups (SSGs). The package works within the density functional theory (DFT) framework and requires wavefunctions from DFT codes, such as VASP, Quantum ESPRESSO, as well as any other code that has an interface to Wannier90. We introduce a set of compact SSG international symbols by combining non-crystallographic point groups with the 230 crystallographic space groups. The program first identifies all SSG operations and determines the SSG international symbol for a given magnetic system. It then generates the SSG character tables of little groups at any $k$ point. Finally, it computes the traces of matrix representations of SSG operations and assigns irreducible corepresentation labels to magnetic energy bands. The program is not only timely but also essential for advancing research on the study of magnons, altermagnetism, magnetic topology, and novel high-degeneracy excitations in SSG systems.

cond-mat.mtrl-sci↗
arXiv2026

The shift-and-invert Arnoldi method for singular matrix pencils

A popular method for solving large sparse regular eigenvalue problem is the shift-and-invert Arnoldi method. This paper aims to use the method for large sparse singular pencils. In three recent papers, {\em Hochstenbach, Mehl, and Plestenjak, 2019, 2023, and 2024}, propose regularization of the singular pencil, using randomly chosen regularization matrices. We propose sparse regularization matrices obtained from the pivoting sequence of a sparse LU factorization. As a side effect, the LU factorization often is rank revealing, which facilitates finding a regularization. Numerical examples illustrate that the LU factorization mostly detects the normal rank and finds a suitable sparse regularization. A rank correction method is proposed for the cases where the normal rank is not determined correctly. For full rank rectangular eigenvalue problems, the pivoting sequence of existing sparse direct system solvers can be used. We compare with randomized regularization methods: preservation of sparsity is beneficial for performance, and often, the accuracy of the eigenvalue solver.

math.NA↗
arXiv2026

A Data-Efficient Path to Multilingual LLMs: Language Expansion via Post-training PARAM$Δ$ Integration into Upcycled MoE

Expanding Large Language Models~(LLMs) to new languages is a costly endeavor, demanding extensive Continued Pre-Training~(CPT) and data-intensive alignment. While recent data-free merging techniques attempt to bypass alignment by fusing a multilingual CPT-enhanced model with its instruct counterpart, they are plagued by a critical trade-off: mitigating parameter conflicts to preserve original abilities inevitably dilutes new language acquisition, and vice-versa. To resolve this conflict, we introduce \method, which upcycles a dense model into a Mixture-of-Experts~(MoE) architecture, allocating different experts to different languages. Alignment ability is then transferred by grafting a MoE-expanded parameter delta~($Δ_{\text{post}}$) to the CPT-enhanced base model, bypassing the complex alignment phase. Experiments demonstrate \method's superiority even against baselines with similar FLOPs or number of parameters; it improves performance on expanded languages while effectively preserving original capabilities. We further show our approach is highly applicable across different models and Post-training deltas.

cs.CL↗
arXiv2026

Three-dimensional topological ferroelectrics

Three-dimensional (3D) topological ferroelectric (FE) insulators, in which topological and FE orders naturally coexist, enable field-controlled spintronic devices. In this work, we predict a new structure of bismuth monohalides Bi4Br4 and Bi4I4, denoted $γ$ phase, and demonstrate that it is an ideal 3D topological FE insulator. Systematic first-principles calculations confirm the stability and synthesizability of $γ$-Bi4X4 (X=Br, I). Although the noncentrosymmetric $γ$ phase crystallizes in the space group $Cmc2_1$ with no symmetry-based classifications/indicators, the nontrivial topology can be characterized by the spin Chern number (SCN). Spin-resolved Wilson loops show the $s_z$ SCN $C_{s_z}=2$, indicating the spin-resolved topology of a 3D quantum spin Hall insulator state. The $z$-direction polarization can be switched by interlayer sliding, requiring only crossing a small energy barrier. Finally, we design an electrically controlled spin-filter device on bilayer films that can generate a switchable spin-polarized current. Combining a single-phase crystal, a sizable band gap, and robust band topology against FE switching, these bismuth monohalides serve as a prototype of intrinsic 3D topological FE insulators, providing an ideal platform for realizing new nonvolatile functionalities in spintronic devices.

cond-mat.mtrl-sci↗
arXiv2026

Percolation from Quantum Metric in Flat-Band Delocalization

The quantum metric is a fundamental ingredient of band quantum geometry and has recently at tracted intense interest, with most of its transport signatures appearing in the intrinsic second order nonlinear conductivity. In the clean limit, previous works argued that linear response conductivity is insensitive to the quantum metric, while the Berry curvature yields an intrinsic anomalous Hall con tribution. Here we combine analytic derivations with new numerics to show that disorder modifies the linear response conductivity dominated by geometric conductivity which is determined by the real space quantum metric. Focusing on a two dimensional multi-flatband stub-pyrochlore lattice, we identify a critical delocalized regime sandwiched between flat band localization and Anderson localization, characterized by finite geometric conductivity. Upon including spin orbit coupling, this regime evolves into a diffusive metallic phase, constituting a two dimensional inverse Anderson transition. Moreover, exploiting the connection between the real space quantum metric marker and the Wannier function spread, we construct a bond-percolation model on a square lattice. The resulting percolation region quantitatively coincides the critical delocalized regime, the exponent of which supports a classical percolation universality class. These findings suggest that flat band de localization can be understood as a classical percolation of quantum metric puddles. This advances our understanding of quantum geometric contributions to transport and establishes linear response measurements as a new avenue for accessing the quantum metric.

cond-mat.dis-nn↗