arXiv subjects
Publications and source records attributed to Xiaoyang Chen.
This book provides a comprehensive and accessible introduction to the emerging field of AI for mathematics. It covers the core principles and diverse applications of using artificial intelligence to advance mathematical research. Through clear explanations, the text explores how AI can discover hidden mathematical patterns, assist in proving complicated theorems, and even construct counterexamples to challenge conjectures.
The discovery of superconductivity in the Ruddlesden-Popper (RP) nickelates has opened a new chapter in the search for high superconducting transition temperatures ($T_\mathrm{c}$) materials. A central and puzzling feature of this family is the wide variation in $T_\mathrm{c}$ despite their common NiO$_2$ building blocks, as highlighted by the recent observation of superconductivity at $\sim$ 30 K in trilayer $\mathrm{La_4Ni_3O_{10}}$, significantly lower than 80 K reported in bilayer $\mathrm{La_3Ni_2O_7}$. Understanding the factors that control $T_\mathrm{c}$ in this family is therefore of paramount importance. Here, we use resonant inelastic x-ray scattering (RIXS) to investigate the electronic and magnetic excitations of $\mathrm{La_4Ni_3O_{10}}$ in direct comparison with its bilayer counterpart. Our results reveal a markedly different landscape. $\mathrm{La_4Ni_3O_{10}}$ exhibits a more itinerant character, evidenced by broader Ni $dd$ orbital excitations and a strong Ni 3$d$ fluorescence continuum, suggesting weaker electronic correlations than in the bilayer. Despite this, well-defined collective spin excitations persist, including dispersive acoustic and optical magnon branches alongside an incommensurate spin density wave. Using linear spin wave theory, we extract the interlayer superexchange interaction ($J_z$) to be $\sim$ 22 meV, much smaller than that in $\mathrm{La_3Ni_2O_7}$. The weaker correlation and reduced interlayer exchange together provide a consistent explanation for the substantially lower $T_\mathrm{c}$ in the trilayer compound. Our findings establish interlayer magnetic coupling and electronic correlation as key parameters governing superconductivity in layered nickelates and offer critical constraints for understanding the pairing mechanism in this emerging family.
This paper is the second one in a series of work on the mathematical creativity of LLM. In the first paper, the authors proposed three criteria for evaluating the mathematical creativity of LLM and constructed a benchmark dataset to measure it. This paper further explores the mathematical creativity of LLM, with a focus on investigating whether LLM can generate valuable and cutting-edge mathematical research problems. We develop an agent to generate unknown problems and produced 665 research problems in differential geometry. Through human verification, we find that many of these mathematical problems are unknown to experts and possess unique research value.
One of the most remarkable discoveries of JWST is a population of compact, red sources at z > 4, commonly referred to as Little Red Dots (LRDs). Spectroscopic identifications reported that most LRDs are active galactic nuclei (AGNs), which are preferentially found around z ~ 6 and could imply a key phase in the formation and growth of black holes (BHs) in the early universe. Photometric surveys at lower redshift have recently been carried out to trace their evolution across cosmic time, and a small number of LRD-like galaxies have been spectroscopically identified at both Cosmic Noon and in the local universe. Here we report the discovery of one of the lowest-z LRD-like galaxies, J204837.26-002437.2 (hereafter J2048) at z = 0.4330, using new Gemini-N/GMOS IFU observations combined with archival multi-band photometric SED data. The GMOS data reveal extended blue emission from starburst with a star formation rate of 400 solar mass per year, together with an extended, highly fast ionized outflow. This is the first spectroscopic confirmation of extended host emission and outflow in an LRD-like galaxy, providing a unique laboratory for understanding the nature of their high-redshift counterparts. Moreover, J2048 would host an extremely overmassive BH with a BH-to-stellar mass ratio of ~ 60%, with the BH mass and host stellar mass estimated to be 10^10.2 and 10^10.4 solar masses, respectively. We discuss the origin and evolutionary fate of J2048, and the implications that such low-z analogs have for interpreting the properties of high-z LRDs.
Real-time sequential control agents are often bottlenecked by inference latency. Even modest per-step planning delays can destabilize control and degrade overall performance. We propose a speculation-and-correction framework that adapts the predict-then-verify philosophy of speculative execution to model-based control with TD-MPC2. At each step, a pretrained world model and latent-space MPC planner generate a short-horizon action queue together with predicted latent rollouts, allowing the agent to execute multiple planned actions without immediate replanning. When a new observation arrives, the system measures the mismatch between the encoded real latent state and the queued predicted latent. For small to moderate mismatch, a lightweight learned corrector applies a residual update to the speculative action, distilled offline from a replanning teacher. For large mismatch, the agent safely falls back to full replanning and clears stale action queues. We study both a gated two-tower MLP corrector and a temporal Transformer corrector to address local errors and systematic drift. Experiments on the DMC Humanoid-Walk task show that our method reduces the number of planning inferences from 500 to 282, improves end-to-end step latency by 25 percent, and maintains strong control performance with only a 7.1 percent return reduction. Ablation results demonstrate that speculative execution without correction is unreliable over longer horizons, highlighting the necessity of mismatch-aware correction for robust latency reduction.
Reasoning-intensive ranking models built on Large Language Models (LLMs) have made notable progress. However, existing approaches often rely on large-scale LLMs and explicit Chain-of-Thought (CoT) reasoning, resulting in high computational cost and latency that limit real-world use. To address this, we propose \textbf{TFRank}, an efficient pointwise reasoning ranker based on small-scale LLMs. To improve ranking performance, TFRank effectively integrates CoT data, fine-grained score supervision, and multi-task training. Furthermore, it achieves an efficient ``\textbf{T}hink-\textbf{F}ree" reasoning capability by employing a ``think-mode switch'' and pointwise format constraints. Specifically, this allows the model to leverage explicit reasoning during training while delivering precise relevance scores for complex queries at inference without generating any reasoning chains. Experiments show that TFRank achieves performance comparable to models with four times more parameters on the BRIGHT benchmark and demonstrates strong competitiveness on the BEIR benchmark. Further analysis shows that TFRank achieves an effective balance between performance and efficiency, providing a practical solution for integrating advanced reasoning into real-world systems. Our code and data are released in the repository: https://github.com/JOHNNY-fans/TFRank.
We report the multiwavelength properties of eFEDS J084222.9+001000 (hereafter ID830), a quasar at $z=3.4351$, identified as the most X-ray luminous radio-loud quasar in the eROSITA Final Equatorial Depth Survey (eFEDS) field. ID830 shows a rest-frame 0.5-2 keV luminosity of $\log (L_\mathrm{0.5-2\,keV}/\mathrm{erg}~\mathrm{s}^{-1}) = 46.20 \pm 0.12$, with a steep X-ray photon index ($Γ=2.43 \pm 0.21$), and a significant radio counterpart detected with VLA/FIRST 1.4 GHz and VLASS 3 GHz bands. The rest-frame UV to optical spectra from SDSS and Subaru/MOIRCS $J$-band show a dust reddened quasar feature with $A_\mathrm{V} = 0.39 \pm 0.08$ mag and the expected bolometric AGN luminosity from the dust-extinction-corrected UV luminosity reaches $L_\mathrm{bol,3000}= (7.62 \pm 0.31) \times 10^{46}$ erg s$^{-1}$. We estimate the black hole mass of $M_\mathrm{BH} = (4.40 \pm 0.72) \times 10^{8} M_{\odot}$ based on the MgII$λ$2800 emission line width, and an Eddington ratio from the dust-extinction-corrected UV continuum luminosity reaches $λ_\mathrm{Edd,UV}=1.44 \pm 0.24$ and $λ_{\mathrm{Edd,X}} = 12.8 \pm 3.9$ from the X-ray luminosity, both indicating the super-Eddington accretion. ID830 shows a high ratio of UV-to-X-ray luminosities, $α_\mathrm{OX}=-1.20 \pm 0.07$ (or $α_\mathrm{OX}=-1.42 \pm 0.07$ after correcting for jet-linked X-ray excess), higher than quasars and little red dots in super-Eddington phase with similar UV luminosities, with $α_\mathrm{OX}<-1.8$. Such a high $α_\mathrm{OX}$ suggests the coexistence of a prominent radio jet and X-ray corona, in this high Eddington accretion phase. We propose that ID830 may be in a transitional phase after an accretion burst, evolving from a super-Eddington to a sub-Eddington state, which could naturally describe the high $α_\mathrm{OX}$.
To understand and identify the unprecedented risks posed by rapidly advancing artificial intelligence (AI) models, this report presents a comprehensive assessment of their frontier risks. Drawing on the E-T-C analysis (deployment environment, threat source, enabling capability) from the Frontier AI Risk Management Framework (v1.0) (SafeWork-F1-Framework), we identify critical risks in seven areas: cyber offense, biological and chemical risks, persuasion and manipulation, uncontrolled autonomous AI R\&D, strategic deception and scheming, self-replication, and collusion. Guided by the "AI-$45^\circ$ Law," we evaluate these risks using "red lines" (intolerable thresholds) and "yellow lines" (early warning indicators) to define risk zones: green (manageable risk for routine deployment and continuous monitoring), yellow (requiring strengthened mitigations and controlled deployment), and red (necessitating suspension of development and/or deployment). Experimental results show that all recent frontier AI models reside in green and yellow zones, without crossing red lines. Specifically, no evaluated models cross the yellow line for cyber offense or uncontrolled AI R\&D risks. For self-replication, and strategic deception and scheming, most models remain in the green zone, except for certain reasoning models in the yellow zone. In persuasion and manipulation, most models are in the yellow zone due to their effective influence on humans. For biological and chemical risks, we are unable to rule out the possibility of most models residing in the yellow zone, although detailed threat modeling and in-depth assessment are required to make further claims. This work reflects our current understanding of AI frontier risks and urges collective action to mitigate these challenges.
To advance the mathematical proficiency of large language models (LLMs), the DeepMath team has launched an open-source initiative aimed at developing an open mathematical LLM and systematically evaluating its mathematical creativity. This paper represents the initial contribution of this initiative. While recent developments in mathematical LLMs have predominantly emphasized reasoning skills, as evidenced by benchmarks on elementary to undergraduate-level mathematical tasks, the creative capabilities of these models have received comparatively little attention, and evaluation datasets remain scarce. To address this gap, we propose an evaluation criteria for mathematical creativity and introduce DeepMath-Creative, a novel, high-quality benchmark comprising constructive problems across algebra, geometry, analysis, and other domains. We conduct a systematic evaluation of mainstream LLMs' creative problem-solving abilities using this dataset. Experimental results show that even under lenient scoring criteria -- emphasizing core solution components and disregarding minor inaccuracies, such as small logical gaps, incomplete justifications, or redundant explanations -- the best-performing model, O3 Mini, achieves merely 70% accuracy, primarily on basic undergraduate-level constructive tasks. Performance declines sharply on more complex problems, with models failing to provide substantive strategies for open problems. These findings suggest that, although current LLMs display a degree of constructive proficiency on familiar and lower-difficulty problems, such performance is likely attributable to the recombination of memorized patterns rather than authentic creative insight or novel synthesis.
Recent high-$z$ quasar observations strongly indicate that super-Eddington accretion is a crucial phase to describe the presence of supermassive black holes (SMBHs) with $M_\mathrm{BH} \gtrsim 10^9 M_\odot$ at $z \gtrsim 7$. Motivated by theoretical predictions that the super-Eddington phase efficiently produces outflows and jets bright in radio bands, we identify a super-Eddington radio-loud dust-obscured galaxy (DOG) J1406+0102 at $z=0.2367$. This source is discovered by cross-matching the infrared-bright DOGs of Nobriguchi et al. (2019) with the VLA/FIRST 1.4 GHz radio survey data and the SDSS optical spectral catalog. J1406+0102 shows broad components in the Balmer lines, and by assuming that those lines are from the broad line region, they give BH mass estimations of $\log (M_\mathrm{BH}/M_\odot)=7.30 \pm 0.25$. Combined with an AGN luminosity of $\log (L_\mathrm{bol,[OIII]}/\mathrm{erg}~\mathrm{s}^{-1}) = 45.91\pm0.38$ estimated from the intrinsic [OIII] luminosity, this implies a super-Eddington accretion rate of $λ_\mathrm{Edd} \simeq 3$. We show that 1) J1406+0102 shows strong AGN feedback, with the [OIII] outflow velocity exceeds the escape velocity of the host galaxy halo and a kinetic efficiency of $\approx$ 8%, sufficient to quench the host galaxy; 2) its expected growth trajectory places it on an over-massive BH track; and 3) if representative of radio-loud DOGs, such sources can contribute significantly to the high-energy ($\gtrsim$ 100 TeV) cosmic neutrino background.
3D Affordance detection is a challenging problem with broad applications on various robotic tasks. Existing methods typically formulate the detection paradigm as a label-based semantic segmentation task. This paradigm relies on predefined labels and lacks the ability to comprehend complex natural language, resulting in limited generalization in open-world scene. To address these limitations, we reformulate the traditional affordance detection paradigm into \textit{Instruction Reasoning Affordance Segmentation} (IRAS) task. This task is designed to output a affordance mask region given a query reasoning text, which avoids fixed categories of input labels. We accordingly propose the \textit{3D-AffordanceLLM} (3D-ADLLM), a framework designed for reasoning affordance detection in 3D open-scene. Specifically, 3D-ADLLM introduces large language models (LLMs) to 3D affordance perception with a custom-designed decoder for generating affordance masks, thus achieving open-world reasoning affordance detection. In addition, given the scarcity of 3D affordance datasets for training large models, we seek to extract knowledge from general segmentation data and transfer it to affordance detection. Thus, we propose a multi-stage training strategy that begins with a novel pre-training task, i.e., \textit{Referring Object Part Segmentation}~(ROPS). This stage is designed to equip the model with general recognition and segmentation capabilities at the object-part level. Then followed by fine-tuning with the IRAS task, 3D-ADLLM obtains the reasoning ability for affordance detection. In summary, 3D-ADLLM leverages the rich world knowledge and human-object interaction reasoning ability of LLMs, achieving approximately an 8\% improvement in mIoU on open-vocabulary affordance detection tasks.
The development of Natural Language Interfaces to Databases (NLIDBs) has been greatly advanced by the advent of large language models (LLMs), which provide an intuitive way to translate natural language (NL) questions into Structured Query Language (SQL) queries. While significant progress has been made in LLM-based NL2SQL, existing approaches face several challenges in real-world scenarios of natural language querying over massive databases. In this paper, we present DBCopilot, a framework that addresses these challenges by employing a compact and flexible copilot model for routing over massive databases. Specifically, DBCopilot decouples schema-agnostic NL2SQL into schema routing and SQL generation. This framework utilizes a single lightweight differentiable search index to construct semantic mappings for massive database schemata, and navigates natural language questions to their target databases and tables in a relation-aware joint retrieval manner. The routed schemata and questions are then fed into LLMs for effective SQL generation. Furthermore, DBCopilot introduces a reverse schema-to-question generation paradigm that can automatically learn and adapt the router over massive databases without manual intervention. Experimental results verify that DBCopilot is a scalable and effective solution for schema-agnostic NL2SQL, providing a significant advance in handling natural language querying over massive databases for NLIDBs.
Powerful, galactic outflows driven by Active Galactic Nuclei (AGNs) are commonly considered as a main mechanism to regulate star formation in massive galaxies. Ultra- and hyper-luminous IR galaxies (U/HyLIRGs) are thought to represent a transition phase of galaxies from a rapidly growing period to a quiescent status as gas swept out by outflows, providing a laboratory to investigate outflows and their feedback effects on the hosts. In this paper we report recent Gemini and ALMA observations of a HyLIRG, J1126 at $z=0.46842$, which has been identified with a puzzling co-existence of a fast ionized outflow ($>2000$ km s$^{-1}$) and an intense starburst (star formation rate of 800 $M_{\odot}$ yr$^{-1}$). The Gemini observation shows the fast ionized outflow is extended to several kpc with a mass-loss rate of 180 $M_{\odot}$ yr$^{-1}$. A massive molecular outflow with a high mass-loss rate (2500 $M_{\odot}$ yr$^{-1}$) is revealed by ALMA. The multi-phase outflows show large factors of momentum boost and loading of kinetic power, indicating a driving by thermal pressure of a nuclear hot wind and/or radiation pressure of a highly obscured AGN. In addition to ejection of kinetic energy, it is also found that the powerful outflow can induce an ionizing shock in the galaxy disk and enhance the excitation and dissociation of molecular gas. The powerful outflow probably results in an instantaneous negative feedback and shows potential to regulate the host growth in a long term.
Entity matching (EM) is a critical step in entity resolution (ER). Recently, entity matching based on large language models (LLMs) has shown great promise. However, current LLM-based entity matching approaches typically follow a binary matching paradigm that ignores the global consistency among record relationships. In this paper, we investigate various methodologies for LLM-based entity matching that incorporate record interactions from different perspectives. Specifically, we comprehensively compare three representative strategies: matching, comparing, and selecting, and analyze their respective advantages and challenges in diverse scenarios. Based on our findings, we further design a compound entity matching framework (ComEM) that leverages the composition of multiple strategies and LLMs. ComEM benefits from the advantages of different sides and achieves improvements in both effectiveness and efficiency. Experimental results on 8 ER datasets and 10 LLMs verify the superiority of incorporating record interactions through the selecting strategy, as well as the further cost-effectiveness brought by ComEM.
The discovery of interface-enhanced superconductivity in single-layer FeSe/oxides has generated intensive research interests. Beyond the family of FeSe interfaced with various TiO$_2$ terminated oxides, high pairing temperature up to 80~K has been recently observed in FeSe interfaced with FeO$_x$-terminated LaFeO$_3$. Here we successfully extend the FeSe/FeO$_x$ superconducting interface to FeSe/NdFeO$_3$, by constructing 1uc-FeSe/6uc-NdFeO$_3$/Nb:SrTiO$_3$ heterostructures. Intriguingly, well-annealed FeSe/NdFeO$_3$ exhibits a low doping level of 0.038$\sim$0.046 ~e$^-/$Fe which deviates universally magic doping level (0.10$\sim$0.12 e$^-/\rm{Fe}$) and provides a new playground for studying the FeSe/oxide interface in the low electron-doped regime. Comparing it with thick FeSe films at the comparable electron doping level induced by surface potassium dosing, FeSe/NdFeO$_3$ shows a larger superconducting gap and the absence of a nematic gap, indicating an enhancement of the superconductivity and suppression of nematicity by the FeSe/FeO$_x$ interface. These results not only expand the FeSe/FeO$_x$ superconducting family but also enrich the current understanding on the roles of the oxide interface.
Monolayer FeSe/TiO$_x$ and FeSe/FeO$_x$ interfaces exhibit significant superconductivity enhancement compared to bulk FeSe, with interfacial electron-phonon coupling (EPC) playing a crucial role. However, the reduced dimensionality in monolayer FeSe, which may drive superconducting fluctuations, complicates the understanding of the enhancement mechanisms. Here we construct a new superconducting interface: monolayer FeSe/SrVO$_3$/SrTiO$_3$, in which the itinerant electrons of highly metallic SrVO$_3$ films can screen all the high-energy Fuchs-Kliewer phonons, including those of SrTiO$_3$, making it the first FeSe/oxide system with screened interfacial EPC while maintaining the monolayer FeSe thickness. Despite comparable doping levels, the heavily electron-doped monolayer FeSe/SrVO$_3$ exhibits a lower pairing temperature ($T_\mathrm{g}$ $\sim$ 48 K) than FeSe/SrTiO$_3$ and FeSe/LaFeO$_3$. Our findings disentangle the contributions of interfacial EPC from dimensionality on enhancing $T_\mathrm{g}$ in FeSe/oxide interfaces, underscoring the importance of interfacial EPC in $T_\mathrm{g}$ enhancement. This FeSe/VO$_x$ interface also provides a platform for studying the interfacial superconductivity.
We present Atacama Large Millimeter/submillimeter Array observations of the [C II] 158 $μ$m line and the underlying continuum emission of TN J0924$-$2201, which is one of the most distant known radio galaxies at $z>5$. The [C II] line and 1-mm continuum emission are detected at the host galaxy. The systemic redshift derived from the [C II] line is $z_{\rm [C II]}=5.1736\pm0.0002$, indicating that the Ly$α$ line is redshifted by a velocity of $1035\pm10$ km s$^{-1}$, marking the largest velocity offset between the [C II] and Ly$α$ lines recorded at $z>5$ to date. In the central region of the host galaxy, we identified a redshifted substructure of [C II] with a velocity of $702\pm17$ km s$^{-1}$, which is close to the CIV line with a velocity of $500\pm10$ km s$^{-1}$. The position and the velocity offsets align with a model of an outflowing shell structure, consistent with the large velocity offset of Ly$α$. The non-detection of [C II] and dust emission from the three CO(1--0)-detected companions indicates their different nature compared to dwarf galaxies based on the photodissociation region model. Given their large velocity of $\sim1500$ km s$^{-1}$, outflowing molecular clouds induced by the AGN is the most plausible interpretation, and they may exceed the escape velocity of a $10^{13}\,M_{\odot}$ halo. These results suggest that TN J0924$-$2201, with the ongoing and fossil large-scale outflows, is in a distinctive phase of removing molecular gas from a central massive galaxy in an overdense region in the early universe. A dusty HI absorber at the host galaxy is an alternative interpretation.
The practice of Retrieval-Augmented Generation (RAG), which integrates Large Language Models (LLMs) with retrieval systems, has become increasingly prevalent. However, the repercussions of LLM-derived content infiltrating the web and influencing the retrieval-generation feedback loop are largely uncharted territories. In this study, we construct and iteratively run a simulation pipeline to deeply investigate the short-term and long-term effects of LLM text on RAG systems. Taking the trending Open Domain Question Answering (ODQA) task as a point of entry, our findings reveal a potential digital "Spiral of Silence" effect, with LLM-generated text consistently outperforming human-authored content in search rankings, thereby diminishing the presence and impact of human contributions online. This trend risks creating an imbalanced information ecosystem, where the unchecked proliferation of erroneous LLM-generated content may result in the marginalization of accurate information. We urge the academic community to take heed of this potential issue, ensuring a diverse and authentic digital information landscape.