arXiv subjects

Ke Wang

Publications and source records attributed to Ke Wang.

At least 19 recordsLinked to original sources

arXiv2026

Chiral Superconductivity Adjacent to a Wigner Crystal in Rhombohedral Graphene

It was experimentally observed that Wigner-crystal and chiral superconducting phases occur adjacent to each other in rhombohedral graphene. In this Letter, we explore electronic correlations underlying the two phases and identify the pairing mechanism of chiral superconductivity. Exchange correlations play an essential role in both phases. First, we use self-consistent Hartree-Fock calculations to identify a Wigner-crystal state. Second, to investigate the pairing mechanism, we go beyond charge screening and construct a particle-hole-renormalized interaction through the coupled parquet and Bethe-Salpeter equations. This interaction incorporates exchange correlations while preserving crossing symmetry. Within this framework, we find that electronic correlations in the quarter metal are strongly enhanced near the crystal phase boundary, generating an effective attraction that favors superconducting pairing. Consequently, chiral superconductivity (CSC) emerges in a narrow sliver adjacent to the crystal. Furthermore, we identify the winding number of the superconducting order parameter as $-3$, indicating its $f$-wave nature. These results connect the emergence of chiral pairing and its location in the phase diagram to the electronic correlations favoring crystallization.

cond-mat.supr-con↗

arXiv2026

Investigating nucleon resonances in the $pp\to K^+K^+ΛΛ$ reaction

In this work, we investigate the $pp \to K^+ K^+ ΛΛ$ reaction using an effective Lagrangian approach and the isobar model. The calculation includes the exchange of $π$, $η$, $σ$, $ρ$, $ω$, and $K$ mesons, as well as contributions from eight nucleon resonances: $N(1535)$, $N(1650)$, $N(1710)$, $N(1720)$, $N(1895)$, $N(1900)$, $N(2060)$ and $N(2120)$. The double-resonance excitation mechanism ($NN \to N^* N^*$) is found to dominate this reaction over the entire energy range considered. Among the exchanged mesons, the $η$ exchange provides the largest contribution at lower energies, whereas the $K$ exchange dominates at higher energies. Near the threshold, $N(1535)$ and $N(1650)$ play the most important roles, while at higher energies $N(1710)$ and $N(1720)$ become dominant. Due to isospin conservation, this reaction is predominantly coupled to nucleon resonances, providing a relatively clean channel for $N^*$ spectroscopy. Our results indicate the unique potential of this reaction for studying nucleon resonances and provide a theoretical reference for future experimental studies at HIAF.

hep-ph↗

arXiv2026

Human-Supervised Artificial Intelligence Researchers in Electric Power Systems: Knowledge, Physics, Simulation, and Governance for Large Language Models

Conclusions in power-system research require a clear evidential basis. Researchers need to understand the data sources, modelling assumptions, and solution procedures to assess the conditions under which a conclusion holds. Large language models (LLMs) can help researchers extract information from documents and translate research requests into code and tool calls, but their outputs still require physical validation. Focusing on human-supervised artificial intelligence (AI) researchers, this review examines power-system LLM applications and related scientific workflows, together with physics-grounded power AI studies, to discuss the conditions for using LLMs in scientific research. In the proposed knowledge-physics-simulation-governance framework, LLMs translate information and coordinate tasks, simulators and validation tools test models, and experts retain authority over critical assumptions, safety boundaries, and final conclusions. Research processes should preserve materials for review so that claims can be traced to their sources, computations can be reproduced, and failed attempts and approvals can be examined. On this basis, a staged development roadmap and requirements for research reporting are proposed. Evaluation should focus on research artifacts and their validation to assess whether LLMs can improve the efficiency of power-system research within clearly defined permissions.

cs.AI↗

arXiv2026

Sharp Injective Norm Bounds for Inhomogeneous Gaussian Tensors and Sparse Extensions

We study the Euclidean injective norm of inhomogeneous random tensors with independent entries. For Gaussian tensors, we establish optimal two-sided bounds for the expected norm, with constants depending only on the tensor order. These bounds describe the interplay between variance accumulation along fibres and the contribution of the largest individual entries. In the matrix case, our result is consistent with the characterizations in~\cite{LVHY18}. We then extend the analysis to sparse Gaussian tensors with arbitrary amplitudes and activation probabilities. The expected norm is characterized by the expected largest absolute entry and the largest fibre standard deviation after truncating amplitudes at this scale. Our proofs combine geometric chaining with estimates for averaging over the random support.

math.PR↗

arXiv2026

VL-Dedup: Graded Multimodal Entity Resolution for Image-Text Data Deduplication

Image-text corpora contain copies of one image, captions that describe one image in different ways, captions reused for different images, and captions that do not match their images. Each case calls for a different curation action. Deduplication and multimodal entity resolution (ER) only decide whether records are the same, and they are evaluated on labeled pairs. Yet pair scores do not tell which records end up in the same cluster or which useful captions get deleted. We therefore formulate image-text deduplication as graded multimodal ER, which decides visual identity, caption handling, and mismatches for each pair, and we evaluate it at three levels: pairs, clusters, and the retained set. Our method, VL-Dedup, groups records of the same visual entity into clusters and caps the size of each cluster. Within each cluster, a vision-language model (VLM) and CLIPScore decide which captions to keep and which to flag. We also build VL-DedupBench, with 50,000 COCO and SOP records and 66,788 labeled pairs. On it, a classifier that selects no false edge among the labeled pairs still joins millions of record pairs from different entities. Spending VLM calls on the pairs inside clusters, rather than on pairs with uncertain scores, gives better caption decisions from the same model. Even when almost all clusters are correct, keeping one record per cluster loses facts in many deletions, while the deletions of VL-Dedup lose much less information.

cs.CV↗

arXiv2026

Three omitted values and non-Blaschke point divisors in half-planes

We construct a real meromorphic function $F$ on $\mathbb C$ such that $F^{-1}(\{0,1,\infty\})\subset\mathbb R$, while $F$ is not of bounded type in either half-plane. More strongly, for every $a\in\widehat{\mathbb C}\setminus\{0,1,\infty\}$, the $a$-point divisor in either half-plane fails the Blaschke condition. Thus the construction gives a negative answer to the half-plane bounded-type question arising from Nevanlinna's 1925 work. Postcomposition gives the analogous counterexample for any prescribed triple of distinct values in the Riemann sphere. The core construction and proof were generated during an autonomous run of GPT-5.6 Sol Ultra.

math.CV↗

arXiv2026

SCALES. I. Bridging Lensed Clump - Giant Clump Scales in Lensed Galaxies at $z=1-4$

Dense star-forming regions in $z>1$ galaxies, often termed 'clumps', have been studied across two distinct regimes of telescope resolution, each motivating interpretations in tension with the other. In the SCALES series (Star-Forming Clumps As Layered Emergent Structures), we aim to connect star-forming structures across scales, from the $\sim100\,\mathrm{pc}$ clumps in lensed studies to the $\sim\mathrm{kpc}$-scale ''giant'' clumps in unlensed surveys. In this first paper, we present a difference-of-Gaussian (DoG) decomposition that simultaneously models $\approx 50-400\,\mathrm{pc}$ (lensed) clumps and $\sim\mathrm{kpc}$-scale component (the extended component, or EC), which would together constitute the giant clumps seen at lower resolution. Here the EC is treated as a scale-separated local background rather than a physical object, absorbing all light that is not a compact clump. Applying this to 23 moderately lensed ($μ\approx 2-9$) galaxies at $z_{spec} = 1-4$ to evaluate clump properties (deferring ECs and giant clumps to SCALES-II), we measure a clump stellar mass function slope of $1.88 \pm 0.04$, consistent with the $\approx2$ expected for hierarchical star formation, and with that of $\sim\mathrm{kpc}$-scale clumps in unlensed galaxies. We also show that empirical correlations of clump surface densities and clump-to-host mass ratio with redshift are exaggerated by redshift-dependent detection bias, so $\approx 50-400\,\mathrm{pc}$ clumps are largely co-similar across our range ($z = 1-4$). Finally, combining clump mass, size and age estimates with literature gas-dispersion measurements at similar scales, we argue that $\sim100\,\mathrm{pc}$ clumps are themselves composed of star-cluster-like sub-components at much smaller scales.

astro-ph.GA↗

arXiv2026

SCALES. II. The Internal Composition of Giant Star-Forming Clumps at $z = 1 - 4$

Star-forming structures of scale $\sim0.5-1\,\mathrm{kpc}$, dubbed 'giant' clumps, are routinely detected in unlensed surveys of $z > 1$ galaxies. Explaining them requires characterizing their internal structure. In this work, part of the Star-Forming Clumps As Layered Emergent Structures (SCALES) series, we hence ask: What is inside a giant clump? We analyze 23 lensed galaxies at $z=1-4$ with JWST/NIRCam using a difference-of-Gaussian decomposition to model (lensed) clumps at $\approx 50-400\,\mathrm{pc}$ scales, and the surrounding extended component (EC) at giant clump scales. Associating clumps to ECs, we identify 72 giant clumps. We find the constituent clumps above our mass completeness limit ($10^{7.1}\,\mathrm{M_\odot}$) in a giant clump to be consistent with random sampling of the clump stellar mass function (CSMF), hence holding no information about their host. This agrees qualitatively with both prevailing scenarios for giant clumps: (I) long-lived structures with constantly renewed constituents, and (II) blended random clump associations. As most models lack predictions for their internal mass distribution, our observations provide an empirical constraint. Finally, the CSMF follows a power law of slope $1.88\pm0.04$ above a break at $\sim10^{7.5}\,\mathrm{M_\odot}$, close to our completeness limit. The giant clumps follow a consistent slope ($1.95\pm0.15$), giving the same scale-free mass function slope at (lensed) clump and giant clump scales. Within giant clumps, both the EC mass, if made up of undetected clumps, and the mass share of the most massive clump are consistent with the CSMF slope continuing down to $\approx10^{5.5}-10^{6}\,\mathrm{M_\odot}$, while the latter rejects an extension below $10^{5}\,\mathrm{M_\odot}$.

astro-ph.GA↗

arXiv2026

Draining Without Emptying a Stringy Black Hole: Repetitive Electric Penrose Process

This paper studies the repetitive electric Penrose process in the Gibbons-Maeda-Garfinkle-Horowitz-Strominger (GMGHS) black hole under the optimal condition for maximum energy extraction. Unlike the Reissner-Nordström (RN) black hole, the effective potential curve of the incident particle in the GMGHS black hole remains unchanged during the iterations. The iteration termination conditions are determined by two aspects: the escape condition requires that particle 0 can overcome the effective potential peak and escape to infinity, and this condition is independent of the black hole charge; the existence condition for negative-energy particle 1 requires the black hole charge to be sufficiently large, indicating that the repetitive process cannot completely consume the charge of the GMGHS black hole to zero. Numerical simulations show that the return on investment of energy extraction can exceed $100\%$, but the energy utilization efficiency does not exceed $50\%$, and most of the extractable energy is converted into an increase in the irreducible mass of the black hole. In addition, a smaller decay radius and a larger absolute value of the charge-to-mass ratio of particle 1 can improve the return on investment and energy utilization efficiency of energy extraction; the larger the specific energy of the incident particle, the smaller the return on investment of energy extraction, while the energy utilization efficiency remains unchanged; for an initially non-extremal black hole, as the initial charge increases, the return on investment of energy extraction increases monotonically, while the energy utilization efficiency first increases and then decreases. After the iterations end, the black hole still retains extractable energy, and neither the charge nor the mass vanishes.

gr-qc↗

arXiv2026

PriFT: Guiding Supervised Fine-Tuning with Pretrained Priors

Supervised fine-tuning (SFT) is an efficient approach for downstream task adaptation and often serves as the initialization stage for reinforcement learning (RL), but it can show weaker generalization than RL. A key limitation is its off-policy objective: SFT fits fixed demonstrations token by token, including targets poorly aligned with the model's pretrained distribution, which can lead to overfitting. A recent line of work addresses this issue by assigning larger training weights to tokens better aligned with the current model's predictive distribution, with the intuition that fitting these tokens are less distortive to the model's knowledge and representations. However, computing the token weights from the model that is currently fine-tuned entangles token weights with the optimization trajectory, inducing a self-reinforcing loop which concentrates the supervision signal to a narrow set of initially preferred tokens. To address this, we propose PriFT (Prior-support guided Fine-Tuning), which derives token weights from a frozen pretrained reference to obtain a stable reweighting signal unaffected by fine-tuning. This signal estimates prior support: the extent to which each target token is supported by the pretrained distribution. By reweighting tokens according to their prior support, PriFT limits excessive adaptation to targets requiring substantial deviation from the pretrained distribution and improves SFT generalization. We introduce two instantiations: PriFT-prob uses pretrained target-token probability, while PriFT-mass selects tokens by cumulative probability mass under the pretrained distribution. Across mathematical reasoning, code generation, and medical QA, PriFT achieves state-of-the-art results among SFT baselines, while also providing a stronger initialization for subsequent RL.

cs.CL↗

arXiv2026

Sublinear thermal state and ground state preparation for free fermions via system-bath interaction

Dissipative algorithms for preparing thermal and ground states typically require a total Hamiltonian simulation time that grows at least linearly with system size $N$, even when their mixing time grows only polylogarithmically. We achieve sublinear simulation time for free fermions by engineering local interactions with a bath whose size grows in proportion to the system size. Starting from arbitrary initial states, our thermal state preparation algorithm achieves trace-norm error $ε$ in total simulation time $\widetilde{O}(\sqrt{N}/ε)$ at fixed inverse temperature and bounded single-particle bandwidth, while ground state preparation of gapped systems requires only $\mathrm{polylog}(N/ε)$ time. We further establish lower bounds for thermal state preparation within the bath protocols considered, identifying two obstacles to faster preparation, namely fourth-order deviations from the target thermal state and fluctuations induced by a random frequency shared by all bath qubits. Our results demonstrate that the linear simulation time barrier in dissipative state preparation can be overcome and clarify the bath and filter-design constraints that govern further complexity improvements.

quant-ph↗

arXiv2026

Repetitive Magnetic Penrose Process

The nonlinear iterative repetitive Penrose process established recently by Ruffini et al.\ is generalized to the magnetic Penrose process. The spacetime background is the Kerr metric, the external magnetic field is described by the Wald solution, the electromagnetic field is treated as a test field, and the charge is treated as a test charge. We analyze the optimal conditions for maximum energy extraction on the equatorial plane and find that under these conditions the core mechanism of the iterative process is charging feedback. After charging, the black hole charge is approximately half of the Wald charge. When the neutral parent particle decays, charge conservation forces the infalling fragment to carry charge while the other fragment carries the opposite charge, so the black hole charge accumulates monotonically. The monopole voltage of the Wald solution decreases linearly with the charge $Q$ and crosses zero to reverse sign at a certain point, so the magnetic gain is gradually neutralized by the charge. The repetitive magnetic Penrose process is therefore forced to terminate and eventually falls back to the mechanical Penrose process. The energy extraction efficiency $>100\%$ predicted by the single magnetic Penrose process will therefore be prevented.

gr-qc↗

arXiv2026

Replan for Error Recovery: Training Embodied VLMs through Verifiable Interaction with Action-Grouped Turn-GSPO

Embodied VLMs can execute increasingly complex manipulation tasks, yet execution errors can leave the environment in states that require replanning for recovery. Existing recovery methods either rely on inference-time mechanisms with fixed policies or learn from failure examples and reference corrections rather than verified action outcomes. To enable recovery learning through verifiable interaction, we introduce the Embodied Replanning for Recovery Environment (ERR-Env), which provides feedback on task progress and execution errors. We further construct ERR-Bench, a benchmark with held-out tasks spanning nominal, intermediate, and error states for evaluating recovery capability and execution reliability. We then propose Action-Grouped Turn-GSPO (AGT-GSPO), an online reinforcement learning method combining turn-level rewards for task progress and error states with action-grouped advantage estimation, and use it to train ERR-VLM, a compact 2B embodied VLM. Evaluation on ERR-Bench reveals a mismatch between task progress and execution reliability across diverse VLMs. AGT-GSPO improves ERR-VLM by 10.25 percentage points in Strict Success Rate for error-free completion and 22.95 percentage points in Recovery Rate, while ablations and zero-shot transfer further validate the effectiveness of our method.

cs.RO↗

arXiv2026

Global well-posedness of the periodic Majda-Biello systems in the resonant case

We study the Cauchy problem for the following Majda-Biello system in the resonant case $α\in (0,4]\setminus\{1\}$ on the periodic domain $\mathbb{T}$. \[ \left\{ \begin{array}{rcl} u_{t} + u_{xxx} & = & - v v_x, v_{t} + αv_{xxx} & = & - (uv)_{x}, (u,v)|_{t=0} & = & (u_0,v_0), \end{array} \right. \quad x \in \mathbb{T}, \, t \in \mathbb{R}. \] Here, the initial data $u_0$ and $v_0$ live in the Sobolev space $H^{s}(\mathbb{T})$ and $u_0$ has zero mean over $\mathbb{T}$. We establish the optimal global well-posedness for the above system when $s\geq \min\{1, s_α+\}$, where $s_α$ is a constant depending only on $α$, and $s_α= \frac12$ for almost every $α\in (0,4]\setminus\{1\}$. These indices match the sharp ones in the local theory obtained by Oh (Int. Math. Res. Not., 18:3516-3556, 2009). The main novelty of our proof is to introduce a third modified energy in the I-method by performing a partial correction to the second modified energy around the resonant region. This partial correction avoids the singularity in the full correction, and therefore makes it possible to pointwise bound the quintilinear multipliers arising in the time derivative of the third modified energy. Another feature in our proof is the combination of the Diophantine approximation theory with the sparsity of nearly resonant frequency pairs on the periodic lattice to obtain the key quadrilinear and quintilinear estimates, leading to an almost conservation law sufficient for the global iteration.

math.AP↗

arXiv2026

Atomistic prediction of magnetic behaviour across chemistry, structure and temperature

Controlling magnetic phases and properties through composition and external conditions is central to the development of magnetic materials. Atomistic foundation models offer a route to predicting such behaviour across diverse materials, but magnetic materials remain challenging because their atomic structure and local moments are strongly coupled, while temperature and disorder drive magnetic moments into non-collinear states with evolving orientations and magnitudes. Broadly pretrained atomistic models still sample only limited magnetic diversity, while finite-temperature descriptions of coupled lattice and magnetic fluctuations remain largely material-specific. Here we introduce MagPT, pretrained across 48 elements on 912,492 spin-lattice configurations spanning structural, orientational and longitudinal magnetic fluctuations over 0 $\sim$ 1,000 K and 0 $\sim$ 10 GPa. By treating species, atomic positions and vector moments together, MagPT distinguishes magnetic states that share the same atomic structure but differ in spin configuration and energy. It captures magnetic behaviour across composition, temperature and length scales, from Slater-Pauling-type composition-magnetism relations in transition-metal alloys and finite-temperature phase evolution in elemental ferromagnets and magnetic shape-memory Heusler alloys, to emergent long-wavelength non-collinear spin textures in chiral magnets and local chemistry-magnetism coupling in high-entropy alloys. MagPT moves atomistic studies of magnetic materials beyond material-specific approaches, establishing a general framework for exploring and designing magnetic matter across diverse chemistries and conditions.

cond-mat.mtrl-sci↗

arXiv2026

Selm: Self-Evolving Library Management for Multi-Skill Agent Optimization

Large language model (LLM) agents increasingly use skills as a lightweight way to acquire task-specific knowledge. Recent work enables skills to evolve from task trajectories. However, these methods often lack systematic management of evolved skills, making it difficult for agents to reuse and accumulate skills across domains, thus limiting performance on subsequent tasks.To reliably accumulate, reuse, and maintain useful skills across tasks, we present SELM (Self-Evolving Library Management), a framework that formulates skill evolution as library-level management. SELM jointly manages skill retrieval, updating, and lifecycle by maintaining skill confidence and utility signals. It introduces utility-aware retrieval to select useful skills, a Map-Reduce framework to generate targeted updates from task-specific experience, and lifecycle controls to retain or remove skills based on their effectiveness. These mechanisms enable the skill library to continuously accumulate useful experience while preserving and pruning skills across tasks, improving agent performance on complex tasks across diverse domains. We evaluate SELM on four agent benchmarks covering diverse tasks and execution environments, and SELM consistently outperforms in most settings.

cs.AI↗

arXiv2026

LEGS: Fine-Tuning Teleop-Free VLAs for Humanoid Loco-manipulation in an Embodied Gaussian Splatting World

Training vision-language-action (VLA) policies for humanoid loco-manipulation is constrained by the high cost and complexity of collecting human teleoperation demonstrations. VLA policies fine-tuned in simulators have so far transferred poorly to humanoid loco-manipulation tasks. We present LEGS (Loco-manipulation via Embodied Gaussian Splatting), a hybrid simulator that composites a mesh foreground (robot, objects, props) over a photorealistic 3D Gaussian Splatting (3DGS) background reconstructed from a handheld scene capture. LEGS uses a procedural motion-primitive generator to synthesize labeled demonstrations at scale without human teleoperation, and a two-stage color calibration that aligns renders to the deployment camera. On a Unitree G1 humanoid robot, across three pick-and-place tasks of increasing whole-body difficulty and three VLA backbones ($ψ_0$, $π_{0.5}$, GR00T N1.6), a policy trained purely on LEGS data matches or exceeds one trained on human teleoperation demos on every experiment. It also outperforms a mesh-only baseline trained on the same motions without the calibrated 3DGS frontend. Because motion is recorded independently of appearance, the same demonstrations can be re-rendered for a target scene and objects at over 15x lower cost than teleoperation. When both the objects and the scene change, a policy trained on LEGS-AUG data re-rendered for that target maintains task success, while the teleop baseline fails entirely. Code and videos are available at https://hojjunekim.github.io/legs/.

cs.RO↗

arXiv2026

What Makes a Social Agent's Interactions Effective? Learning from Verifiable Outcomes in Paired Worlds

Effective social interaction requires more than fluent responses: agents must reach plans that fit the participants' needs and circumstances. Existing evaluations of social agents mainly rely on LLM judges that rate the conversation, without checking the resulting plan. We introduce HICO-Bench, a benchmark of 180 everyday coordination scenarios, such as resource allocation, scheduling, and authorization, in which two agents with private information converse and then independently record the plan they agreed on. A program checks Feasibility (the two records match and satisfy both participants' private conditions), and an LLM judge checks Integrity (the conversation supports the recorded plan); passing both is a Verified Success. Each scenario has two versions that share the same public context but differ in one private condition, so the correct plan differs between them. Across thirteen LLMs, 36% of conversations rated highly for goal completion fail Verified Success. To learn from outcomes that arrive only at the end of a conversation, we propose Paired-Credit Policy Optimization (PaCPO). After each turn, PaCPO asks the agent for a tentative plan, checks it in both versions of the scenario, and rewards the turns after which the tentative plans fit their respective versions. PaCPO raises Qwen3-8B's Verified Success from 20.4% to 28.3%.

cs.CL↗