1. Little Red Dots as Direct-collapse Black Hole Nurseries
Authors: Elia Cenci, Melanie Habouzit β’
Published: 2025-08-20 β’
Source: arXiv
The James Webb Space Telescope recently uncovered a population of massive black holes (BHs) in the first billion years after the Big Bang. Among these high-redshift BH candidates, observations have identified a class of active galactic nuclei candidates, dubbed Little Red Dots (LRDs), with extraordinarily compact gas reservoirs and peculiar spectral features. LRDs clearly emerge at redshift z<8 and their abundance declines by z<5. Recent theoretical studies have explored the link between LRDs and the formation of heavy BH seeds in the early Universe, such as direct-collapse BHs (DCBHs). Here we present results from preliminary runs for the MELIORA cosmological hydrodynamical simulations, where we implement an accurate model for DCBH formation, accounting for the Lyman-Werner radiation field and mass-inflow rates in the target host haloes. We aim to test whether or not DCBH formation could lead to systems resembling those hypothesized for LRDs. We find that the population of newly formed DCBHs in the simulations exhibits a steep decline at z<6, akin to the emergence of LRDs, primarily driven by reduced inflows. The birth of DCBHs is associated with a significant gas compaction event, followed by a phase of intense luminosity in the 200 Myr after their birth, and subsequently by the formation of the first PopIII stars in these very haloes. If these DCBHs nurseries are associated with LRDs, then it could explain their weak emission from X-rays and hot dust.
2. Quantization Meets dLLMs: A Systematic Study of Post-training Quantization for Diffusion LLMs
Authors: Haokun Lin, Haobo Xu, Yichen Wu, Ziyu Guo, Renrui Zhang, Zhichao Lu, Ying Wei, Qingfu Zhang, Zhenan Sun β’
Published: 2025-08-20 β’
Source: arXiv
Recent advances in diffusion large language models (dLLMs) have introduced a promising alternative to autoregressive (AR) LLMs for natural language generation tasks, leveraging full attention and denoising-based decoding strategies. However, the deployment of these models on edge devices remains challenging due to their massive parameter scale and high resource demands. While post-training quantization (PTQ) has emerged as a widely adopted technique for compressing AR LLMs, its applicability to dLLMs remains largely unexplored. In this work, we present the first systematic study on quantizing diffusion-based language models. We begin by identifying the presence of activation outliers, characterized by abnormally large activation values that dominate the dynamic range. These outliers pose a key challenge to low-bit quantization, as they make it difficult to preserve precision for the majority of values. More importantly, we implement state-of-the-art PTQ methods and conduct a comprehensive evaluation across multiple task types and model variants. Our analysis is structured along four key dimensions: bit-width, quantization method, task category, and model type. Through this multi-perspective evaluation, we offer practical insights into the quantization behavior of dLLMs under different configurations. We hope our findings provide a foundation for future research in efficient dLLM deployment. All codes and experimental setups will be released to support the community.
3. Global Anomalies in Sigma Models with Majorana--Weyl Fermions
Authors: Changha Choi β’
Published: 2025-08-20 β’
Source: arXiv
We investigate a global sigma model anomaly in two-dimensional sigma models with Majorana--Weyl fermions coupled to a sigma model field with target space~$G$. The anomaly originates from the nontrivial topology of the space of maps and manifests as a phase in the fermion path integral. Using the global anomaly formula expressed in terms of the reduced~$\eta$-invariant, we demonstrate that this anomaly modifies the standard quantization condition of the Wess--Zumino term on~$G$, in close analogy with the three-dimensional parity anomaly. However, our situation is more refined and highlights a qualitatively new phenomenon in two dimensions: whereas in three dimensions the anomalous quantization restricts the level to lie in a half-integer lattice, here it can force the level to take \emph{arbitrary real values}. Furthermore, our results support the consistency of the low-energy description proposed by Gaiotto, Johnson-Freyd, and Witten for three-dimensional $\mathcal{N}=1$ supersymmetric Yang--Mills theory on an interval, by highlighting a subtle and qualitatively distinct nature of the sigma-model anomalies.
4. Anyon superfluidity of excitons in quantum Hall bilayers
Authors: Zhaoyu Han, Taige Wang, Zhihuan Dong, Michael P. Zaletel, Ashvin Vishwanath β’
Published: 2025-08-20 β’
Source: arXiv
The charged anyons of a fractional quantum Hall fluid are necessarily dispersionless due to the continuous magnetic translation symmetry. Neutral anyons, however, can disperse, resulting in a much richer space of possible ``daughter'' states when doped to finite density. We discuss a natural realization of such physics in quantum Hall bilayers, where a finite density of excitons with fractional statistics is argued to give rise to `anyonic exciton superfluidity,' the charge-neutral analog of anyon superconductivity. In a balanced bilayer of two Laughlin $\nu = 1/3$ states, the minimal interlayer exciton carries anyonic exchange statistics. A finite density of these excitons is argued to yield an exciton superfluid stitched to a specific bulk topological order and edge spectrum. Such superfluidity should be most robust near the direct transition into the Halperin $(112)$ state, and near analogous transitions in the bilayer Jain sequence at total filling $\nu_\text{T} = 2\times \frac{n}{2n+1}$. These topological transitions can be described by Chern-Simons QED$_3$, from which we derive several novel and general properties of anyon superfluidity near such transitions, including an anomalously large superfluid stiffness of $\kappa_\text{s} \propto |\delta\nu|^{1/2}$ at layer imbalance fraction $\delta\nu$. A notable feature of the phase diagrams we construct is the prevalence of spatial symmetry breaking, driven by an underlying composite Fermi surface. Our results can be directly tested with currently available experimental techniques. We compare our theory with existing data and make concrete predictions for future measurements, including higher-pseudospin exciton superfluids when doping higher Jain fractions.
5. Virtual Community: An Open World for Humans, Robots, and Society
Authors: Qinhong Zhou, Hongxin Zhang, Xiangye Lin, Zheyuan Zhang, Yutian Chen, Wenjun Liu, Zunzhe Zhang, Sunli Chen, Lixing Fang, Qiushi Lyu, Xinyu Sun, Jincheng Yang, Zeyuan Wang, Bao Chi Dang, Zhehuan Chen, Daksha Ladia, Jiageng Liu, Chuang Gan β’
Published: 2025-08-20 β’
Source: arXiv
The rapid progress in AI and Robotics may lead to a profound societal transformation, as humans and robots begin to coexist within shared communities, introducing both opportunities and challenges. To explore this future, we present Virtual Community-an open-world platform for humans, robots, and society-built on a universal physics engine and grounded in real-world 3D scenes. With Virtual Community, we aim to study embodied social intelligence at scale: 1) How robots can intelligently cooperate or compete; 2) How humans develop social relations and build community; 3) More importantly, how intelligent robots and humans can co-exist in an open world. To support these, Virtual Community features: 1) An open-source multi-agent physics simulator that supports robots, humans, and their interactions within a society; 2) A large-scale, real-world aligned community generation pipeline, including vast outdoor space, diverse indoor scenes, and a community of grounded agents with rich characters and appearances. Leveraging Virtual Community, we propose two novel challenges. The Community Planning Challenge evaluates multi-agent reasoning and planning ability in open-world settings, such as cooperating to help agents with daily activities and efficiently connecting other agents. The Community Robot Challenge requires multiple heterogeneous robots to collaborate in solving complex open-world tasks. We evaluate various baselines on these tasks and demonstrate the challenges in both high-level open-world task planning and low-level cooperation controls. We hope that Virtual Community will unlock further study of human-robot coexistence within open-world environments.
6. Snap-Snap: Taking Two Images to Reconstruct 3D Human Gaussians in Milliseconds
Authors: Jia Lu, Taoran Yi, Jiemin Fang, Chen Yang, Chuiyun Wu, Wei Shen, Wenyu Liu, Qi Tian, Xinggang Wang β’
Published: 2025-08-20 β’
Source: arXiv
Reconstructing 3D human bodies from sparse views has been an appealing topic, which is crucial to broader the related applications. In this paper, we propose a quite challenging but valuable task to reconstruct the human body from only two images, i.e., the front and back view, which can largely lower the barrier for users to create their own 3D digital humans. The main challenges lie in the difficulty of building 3D consistency and recovering missing information from the highly sparse input. We redesign a geometry reconstruction model based on foundation reconstruction models to predict consistent point clouds even input images have scarce overlaps with extensive human data training. Furthermore, an enhancement algorithm is applied to supplement the missing color information, and then the complete human point clouds with colors can be obtained, which are directly transformed into 3D Gaussians for better rendering quality. Experiments show that our method can reconstruct the entire human in 190 ms on a single NVIDIA RTX 4090, with two images at a resolution of 1024x1024, demonstrating state-of-the-art performance on the THuman2.0 and cross-domain datasets. Additionally, our method can complete human reconstruction even with images captured by low-cost mobile devices, reducing the requirements for data collection. Demos and code are available at https://hustvl.github.io/Snap-Snap/.
7. GaussianArt: Unified Modeling of Geometry and Motion for Articulated Objects
Authors: Licheng Shen, Saining Zhang, Honghan Li, Peilin Yang, Zihao Huang, Zongzheng Zhang, Hao Zhao β’
Published: 2025-08-20 β’
Source: arXiv
Reconstructing articulated objects is essential for building digital twins of interactive environments. However, prior methods typically decouple geometry and motion by first reconstructing object shape in distinct states and then estimating articulation through post-hoc alignment. This separation complicates the reconstruction pipeline and restricts scalability, especially for objects with complex, multi-part articulation. We introduce a unified representation that jointly models geometry and motion using articulated 3D Gaussians. This formulation improves robustness in motion decomposition and supports articulated objects with up to 20 parts, significantly outperforming prior approaches that often struggle beyond 2--3 parts due to brittle initialization. To systematically assess scalability and generalization, we propose MPArt-90, a new benchmark consisting of 90 articulated objects across 20 categories, each with diverse part counts and motion configurations. Extensive experiments show that our method consistently achieves superior accuracy in part-level geometry reconstruction and motion estimation across a broad range of object types. We further demonstrate applicability to downstream tasks such as robotic simulation and human-scene interaction modeling, highlighting the potential of unified articulated representations in scalable physical modeling.
8. Estimating Initial Mass of Gaia-Enceladus Dwarf Galaxy with Chemical Evolution Model
Authors: Olcay Plevne, Furkan Akbaba β’
Published: 2025-08-20 β’
Source: arXiv
This work investigates the initial mass and chemical evolution history of the Gaia-Enceladus dwarf galaxy. We combine spectroscopic data from APOGEE with astrometric data from Gaia DR3 to identify Gaia-Enceladus candidate stars via a machine-learning pipeline using t-SNE and HDBSCAN. By focusing on kinematic and chemical parameters, especially $\mathrm{[Fe/H]}$, $\mathrm{[Mg/Fe]}$, $\mathrm{[Al/Fe]}$, and $\mathrm{[Mn/Fe]}$, we uncover a population of metal-poor, high-eccentricity stars that align with literature criteria for Gaia-Enceladus debris. We then apply the \textit{OMEGA+} chemical evolution model, incorporating MCMC fitting of the observed abundance trends in the $\mathrm{[Mg/Fe]\times[Fe/H]}$ plane. Our best-fitting model indicates a gas mass of $4.93_{-0.72}^{+0.32}\times10^9\,{M_{\odot}}$ for Gaia-Enceladus, placing it at the higher end of previously suggested mass ranges. The model scenario suggests a short star formation timescale, substantial outflows, and a rapid build-up of metals mainly driven by core-collapse supernovae, with a lesser contribution from Type~Ia supernovae. Comparison with observational data in other chemical planes (e.g., $\mathrm{[Mg/Mn]\times[Al/Fe]}$) supports this scenario, emphasizing a distinct evolution path relative to the Milky Way. Additionally, our results provide indirect evidence that star formation in Gaia-Enceladus likely ceased within the first 4 Gyr, consistent with earlier inferences of an early merger event. These findings highlight the power of chemical evolution modeling in reconstructing the origin and mass of ancient accreted systems. Overall, we show that Gaia-Enceladus, through a rapid star formation and strong outflows, contributed a significant fraction of the metal-poor stellar halo of the Milky Way.
9. MS-CLR: Multi-Skeleton Contrastive Learning for Human Action Recognition
Authors: Mert Kiray, Alvaro Ritter, Nassir Navab, Benjamin Busam β’
Published: 2025-08-20 β’
Source: arXiv
Contrastive learning has gained significant attention in skeleton-based action recognition for its ability to learn robust representations from unlabeled data. However, existing methods rely on a single skeleton convention, which limits their ability to generalize across datasets with diverse joint structures and anatomical coverage. We propose Multi-Skeleton Contrastive Learning (MS-CLR), a general self-supervised framework that aligns pose representations across multiple skeleton conventions extracted from the same sequence. This encourages the model to learn structural invariances and capture diverse anatomical cues, resulting in more expressive and generalizable features. To support this, we adapt the ST-GCN architecture to handle skeletons with varying joint layouts and scales through a unified representation scheme. Experiments on the NTU RGB+D 60 and 120 datasets demonstrate that MS-CLR consistently improves performance over strong single-skeleton contrastive learning baselines. A multi-skeleton ensemble further boosts performance, setting new state-of-the-art results on both datasets.
10. $\mathrm{GL}_n$ large sieves and density estimates via positive semi-definiteness
Authors: Alexandru Pascadi, Jesse Thorner β’
Published: 2025-08-20 β’
Source: arXiv
Let $\mathfrak{F}_n$ be the set of unitary cuspidal automorphic representations of $\mathrm{GL}_n$ over a number field $F$, and let $\mathcal{S}\subseteq\mathfrak{F}_n$ be a finite subset. Given $\pi_0\in\mathfrak{F}_{n_0}$, we establish large sieve inequalities for the families $\{L(s,\pi)\colon \pi\in\mathcal{S}\}$ and $\{L(s,\pi\times\pi_0)\colon \pi\in\mathcal{S}\}$ that, unlike previous results, are independent of progress towards the generalized Ramanujan conjecture and simultaneously handle the Dirichlet coefficients of $L$, $L^{-1}$, and $\log L$. Our approach is based on the duality principle and is sharp in ranges that are complementary to large sieve inequalities based on trace formulae. We apply our large sieve inequalities to establish several density estimates for families of $L$-functions, counting potential violations of the generalized Riemann hypothesis and the generalized Ramanujan conjecture. In particular, we remove all restrictions in the log-free zero density estimate of Brumley, Thorner, and Zaman for families of Rankin-Selberg $L$-functions.
11. Particle Injection in 3D Relativistic Magnetic Reconnection
Authors: Omar French, Gregory R. Werner, Dmitri A. Uzdensky β’
Published: 2025-08-20 β’
Source: arXiv
Relativistic magnetic reconnection has been proposed as an important nonthermal particle acceleration (NTPA) mechanism that generates power-law spectra and high-energy emissions. Power-law particle spectra are in general characterized by three parameters: the power-law index, the high-energy cutoff, and the low-energy cutoff (i.e., the injection energy). Particle injection into the nonthermal power law, despite also being a critical step in the NTPA chain, has received considerably less attention than the subsequent acceleration to high energies. Open questions on particle injection that are important for both physical understanding and astronomical observations include how the upstream magnetization~$\sigma$ influences the injection energy and the contributions of the known injection mechanisms (i.e., direct acceleration by the reconnection electric field, Fermi kicks, and pickup acceleration) to the injected particle population. Using fully kinetic particle-in-cell simulations, we uncover these relationships by systematically measuring the injection energy and calculating the contributions of each acceleration mechanism to the total injected particle population. We also present a theoretical model to explain these results. Additionally, we compare 2D and 3D simulations to assess the impact of the flux-rope kink and drift-kink instability on particle injection. We conclude with comparisons to previous work and outlook for future work.
12. Deep Reinforcement Learning Based Routing for Heterogeneous Multi-Hop Wireless Networks
Authors: Brian Kim, Justin H. Kong, Terrence J. Moore, Fikadu T. Dagefu β’
Published: 2025-08-20 β’
Source: arXiv
Routing in multi-hop wireless networks is a complex problem, especially in heterogeneous networks where multiple wireless communication technologies coexist. Reinforcement learning (RL) methods, such as Q-learning, have been introduced for decentralized routing by allowing nodes to make decisions based on local observations. However, Q-learning suffers from scalability issues and poor generalization due to the difficulty in managing the Q-table in large or dynamic network topologies, especially in heterogeneous networks (HetNets) with diverse channel characteristics. Thus, in this paper, we propose a novel deep Q-network (DQN)-based routing framework for heterogeneous multi-hop wireless networks to maximize the end-to-end rate of the route by improving scalability and adaptability, where each node uses a deep neural network (DNN) to estimate the Q-values and jointly select the next-hop relay and a communication technology for transmission. To achieve better performance with the DNN, selecting which nodes to exchange information is critical, as it not only defines the state and action spaces but also determines the input to the DNN. To this end, we propose neighbor node selection strategies based on channel gain and rate between nodes rather than a simple distance-based approach for an improved set of states and actions for DQN-based routing. During training, the model experiences diverse network topologies to ensure generalization and robustness, and simulation results show that the proposed neighbor node selection outperforms simple distance-based selection. Further, we observe that the DQN-based approach outperforms various benchmark schemes and performs comparably to the optimal approach.
13. Novel Knockoff Generation and Importance Measures with Heterogeneous Data via Conditional Residuals and Local Gradients
Authors: Evan Mason, Zhe Fei β’
Published: 2025-08-20 β’
Source: arXiv
Knockoff variable selection is a powerful framework that creates synthetic knockoff variables to mirror the correlation structure of the observed features, enabling principled control of the false discovery rate in variable selection. However, existing methods often assume homogeneous data types or known distributions, limiting their applicability in real-world settings with heterogeneous, distribution-free data. Moreover, common variable importance measures rely on linear outcome models, hindering their effectiveness for complex relationships. We propose a flexible knockoff generation framework based on conditional residuals that accommodates mixed data types without assuming known distributions. To assess variable importance, we introduce the Mean Absolute Local Derivative (MALD), an interpretable metric compatible with nonlinear outcome functions, including random forests and neural networks. Simulations show that our approach achieves better false discovery rate control and higher power than existing methods. We demonstrate its practical utility on a DNA methylation dataset from mouse tissues, identifying CpG sites linked to aging. Software is available in R (rangerKnockoff) and Python (MALDimportance).
14. Compute-Optimal Scaling for Value-Based Deep RL
Authors: Preston Fu, Oleh Rybkin, Zhiyuan Zhou, Michal Nauman, Pieter Abbeel, Sergey Levine, Aviral Kumar β’
Published: 2025-08-20 β’
Source: arXiv
As models grow larger and training them becomes expensive, it becomes increasingly important to scale training recipes not just to larger models and more data, but to do so in a compute-optimal manner that extracts maximal performance per unit of compute. While such scaling has been well studied for language modeling, reinforcement learning (RL) has received less attention in this regard. In this paper, we investigate compute scaling for online, value-based deep RL. These methods present two primary axes for compute allocation: model capacity and the update-to-data (UTD) ratio. Given a fixed compute budget, we ask: how should resources be partitioned across these axes to maximize sample efficiency? Our analysis reveals a nuanced interplay between model size, batch size, and UTD. In particular, we identify a phenomenon we call TD-overfitting: increasing the batch quickly harms Q-function accuracy for small models, but this effect is absent in large models, enabling effective use of large batch size at scale. We provide a mental model for understanding this phenomenon and build guidelines for choosing batch size and UTD to optimize compute usage. Our findings provide a grounded starting point for compute-optimal scaling in deep RL, mirroring studies in supervised learning but adapted to TD learning.
15. MedReseacher-R1: Expert-Level Medical Deep Researcher via A Knowledge-Informed Trajectory Synthesis Framework
Authors: Ailing Yu, Lan Yao, Jingnan Liu, Zhe Chen, Jiajun Yin, Yuan Wang, Xinhao Liao, Zhiling Ye, Ji Li, Yun Yue, Hansong Xiao, Hualei Zhou, Chunxiao Guo, Peng Wei, Jinjie Gu β’
Published: 2025-08-20 β’
Source: arXiv
Recent developments in Large Language Model (LLM)-based agents have shown impressive capabilities spanning multiple domains, exemplified by deep research systems that demonstrate superior performance on complex information-seeking and synthesis tasks. While general-purpose deep research agents have shown impressive capabilities, they struggle significantly with medical domain challenges, as evidenced by leading proprietary systems achieving limited accuracy on complex medical benchmarks. The key limitations are: (1) the model lacks sufficient dense medical knowledge for clinical reasoning, and (2) the framework is constrained by the absence of specialized retrieval tools tailored for medical contexts.We present a medical deep research agent that addresses these challenges through two core innovations. First, we develop a novel data synthesis framework using medical knowledge graphs, extracting the longest chains from subgraphs around rare medical entities to generate complex multi-hop question-answer pairs. Second, we integrate a custom-built private medical retrieval engine alongside general-purpose tools, enabling accurate medical information synthesis. Our approach generates 2100+ diverse trajectories across 12 medical specialties, each averaging 4.2 tool interactions.Through a two-stage training paradigm combining supervised fine-tuning and online reinforcement learning with composite rewards, our MedResearcher-R1-32B model demonstrates exceptional performance, establishing new state-of-the-art results on medical benchmarks while maintaining competitive performance on general deep research tasks. Our work demonstrates that strategic domain-specific innovations in architecture, tool design, and training data construction can enable smaller open-source models to outperform much larger proprietary systems in specialized domains.
16. MeshCoder: LLM-Powered Structured Mesh Code Generation from Point Clouds
Authors: Bingquan Dai, Li Ray Luo, Qihong Tang, Jie Wang, Xinyu Lian, Hao Xu, Minghan Qin, Xudong Xu, Bo Dai, Haoqian Wang, Zhaoyang Lyu, Jiangmiao Pang β’
Published: 2025-08-20 β’
Source: arXiv
Reconstructing 3D objects into editable programs is pivotal for applications like reverse engineering and shape editing. However, existing methods often rely on limited domain-specific languages (DSLs) and small-scale datasets, restricting their ability to model complex geometries and structures. To address these challenges, we introduce MeshCoder, a novel framework that reconstructs complex 3D objects from point clouds into editable Blender Python scripts. We develop a comprehensive set of expressive Blender Python APIs capable of synthesizing intricate geometries. Leveraging these APIs, we construct a large-scale paired object-code dataset, where the code for each object is decomposed into distinct semantic parts. Subsequently, we train a multimodal large language model (LLM) that translates 3D point cloud into executable Blender Python scripts. Our approach not only achieves superior performance in shape-to-code reconstruction tasks but also facilitates intuitive geometric and topological editing through convenient code modifications. Furthermore, our code-based representation enhances the reasoning capabilities of LLMs in 3D shape understanding tasks. Together, these contributions establish MeshCoder as a powerful and flexible solution for programmatic 3D shape reconstruction and understanding.
17. Lifespan Pancreas Morphology for Control vs Type 2 Diabetes using AI on Largescale Clinical Imaging
Authors: Lucas W. Remedios, Chloe Cho, Trent M. Schwartz, Dingjie Su, Gaurav Rudravaram, Chenyu Gao, Aravind R. Krishnan, Adam M. Saunders, Michael E. Kim, Shunxing Bao, Thomas A. Lasko, Alvin C. Powers, Bennett A. Landman, John Virostko β’
Published: 2025-08-20 β’
Source: arXiv
Purpose: Understanding how the pancreas changes is critical for detecting deviations in type 2 diabetes and other pancreatic disease. We measure pancreas size and shape using morphological measurements from ages 0 to 90. Our goals are to 1) identify reliable clinical imaging modalities for AI-based pancreas measurement, 2) establish normative morphological aging trends, and 3) detect potential deviations in type 2 diabetes. Approach: We analyzed a clinically acquired dataset of 2533 patients imaged with abdominal CT or MRI. We resampled the scans to 3mm isotropic resolution, segmented the pancreas using automated methods, and extracted 13 morphological pancreas features across the lifespan. First, we assessed CT and MRI measurements to determine which modalities provide consistent lifespan trends. Second, we characterized distributions of normative morphological patterns stratified by age group and sex. Third, we used GAMLSS regression to model pancreas morphology trends in 1350 patients matched for age, sex, and type 2 diabetes status to identify any deviations from normative aging associated with type 2 diabetes. Results: When adjusting for confounders, the aging trends for 10 of 13 morphological features were significantly different between patients with type 2 diabetes and non-diabetic controls (p < 0.05 after multiple comparisons corrections). Additionally, MRI appeared to yield different pancreas measurements than CT using our AI-based method. Conclusions: We provide lifespan trends demonstrating that the size and shape of the pancreas is altered in type 2 diabetes using 675 control patients and 675 diabetes patients. Moreover, our findings reinforce that the pancreas is smaller in type 2 diabetes. Additionally, we contribute a reference of lifespan pancreas morphology from a large cohort of non-diabetic control patients in a clinical setting.
18. Proof of a Generalized Ryu-Takayanagi Conjecture
Authors: Artem Averin β’
Published: 2025-08-20 β’
Source: arXiv
We derive a generalized version of the Ryu-Takayanagi formula for the entanglement entropy in arbitrary diffeomorphism invariant field theories. We use a recent framework which expresses the measurable quantities of a quantum theory as a weighted sum over paths in the theory's phase space. If this framework is applied to a field theory on a spacetime foliated by a hypersurface $\Sigma,$ the choice of a codimension-2 surface $B$ without boundary contained in $\Sigma$ specifies a submanifold in the phase space. For diffeomorphism invariant field theories, a functional integral expression for their density matrices was recently given and then used to derive bounds on phase space volumes in the considered submanifold associated to $B.$ These bounds formalize the gravitational entropy bound. Here, we present an implication of this derivation in that we show the obtained functional integral expression for density matrices to be naturally suited for the replica trick. Correspondingly, we prove a functional integral expression for the associated entanglement entropies and derive a practical prescription for their evaluation to leading order and beyond. An important novelty of our approach is the contact to phase space. This allows us both to obtain a prescription for entanglement entropies in arbitrary diffeomorphism invariant field theories not necessarily possessing a holographic dual as well as to use entanglement entropies to study their phase space structure. In the case of the bulk-boundary correspondence, our prescription consistently reproduces and hence provides a natural and independent proof of the Ryu-Takayanagi formula as well as its various generalizations. These include the covariant holographic entanglement entropy proposal, Dong's proposal for higher-derivative gravity as well as the quantum extremal surface prescription.
19. Holographic Extended Thermodynamics of deformed AdS-Schwarzschild black hole
Authors: Kamal L. Panigrahi, Balbeer Singh β’
Published: 2025-08-20 β’
Source: arXiv
We investigate the thermodynamics and phase structure of a deformed AdS-Schwarzschild black hole, generated via the gravitational decoupling (GD) method as a minimal geometric deformation. In the bulk canonical ensemble, our results exhibit a van der Waals-type first-order phase transition in addition to the Hawking-Page transition, in the suitable parameter regime. Further, we compute the critical exponents characterising the bulk transition, confirming their consistency with mean-field theory predictions. Exploiting the exact holographic dictionary between extended black hole thermodynamics and the dual conformal field theory (CFT), we extend this analysis to the boundary and uncover a rich array of phase transitions and critical phenomena across three distinct thermodynamic ensembles. In particular, for an ensemble of fixed volume and central charge, the dual CFT displays a van der Waals-like phase structure. Throughout, we emphasise the pivotal influence of the GD deformation parameter on the thermodynamic behaviour, and we elucidate its role in the confinement-deconfinement transitions characteristic of the deformed AdS-Schwarzschild geometry.
20. High to low temperature: $O(N)$ model at large $N$
Authors: Justin R. David, Srijan Kumar β’
Published: 2025-08-20 β’
Source: arXiv
We study the $O(N)$ vector model for scalars with quartic interaction at large $N$ on $S^1\times S^2$ without the singlet constraint. The non-trivial fixed point of the model is described by a thermal mass satisfying the gap equation at large $N$. We obtain the partition function and the energy density for the model as a series at low temperature in units of the radius of the sphere. We show these results agree with the Borel-Pad\'{e} extrapolations of the high temperature expansions of the partition function and energy density obtained in our previous work. This agreement validates both the expansions and demonstrates that low temperature expansions obtained here correspond to the same fixed point studied earlier at high temperature. We obtain the ratio of the partition function of the theory at the non-trivial fixed point to that of the Gaussian theory at all values of temperature. This ratio begins at $4/5$ when the temperature is infinity, decreases to a minimum value of $0.760937$, then increases and approaches unity as the temperature is decreased.
21. Squeezed Diffusion Models
Authors: Jyotirmai Singh, Samar Khanna, James Burgess β’
Published: 2025-08-20 β’
Source: arXiv
Diffusion models typically inject isotropic Gaussian noise, disregarding structure in the data. Motivated by the way quantum squeezed states redistribute uncertainty according to the Heisenberg uncertainty principle, we introduce Squeezed Diffusion Models (SDM), which scale noise anisotropically along the principal component of the training distribution. As squeezing enhances the signal-to-noise ratio in physics, we hypothesize that scaling noise in a data-dependent manner can better assist diffusion models in learning important data features. We study two configurations: (i) a Heisenberg diffusion model that compensates the scaling on the principal axis with inverse scaling on orthogonal directions and (ii) a standard SDM variant that scales only the principal axis. Counterintuitively, on CIFAR-10/100 and CelebA-64, mild antisqueezing - i.e. increasing variance on the principal axis - consistently improves FID by up to 15% and shifts the precision-recall frontier toward higher recall. Our results demonstrate that simple, data-aware noise shaping can deliver robust generative gains without architectural changes.
22. Universal winding properties of chiral active motion
Authors: Ion Santra, Urna Basu, Sanjib Sabhapandit β’
Published: 2025-08-20 β’
Source: arXiv
We propose the area swept $A(t)$ and the winding angle $\Omega(t)$ as the key observables to characterize chiral active motion. We find that the distributions of the scaled area and the scaled winding angle are described by universal scaling functions across all well-known models of active particles, parametrized by the chirality $\omega$, along with a self-propulsion speed $v_0$, and the persistence time $\tau$. In particular, we show that, at late times, the average winding angle grows logarithmically with time $\la\Omega \ra\sim(\omega\tau/2)\,\ln t$, while the average area swept has a linear temporal growth $\la A(t)\ra\simeq(\omega\tau D_{\text{eff}})\,t$, where $D_{\text{eff}}=v_0^2 \tau /[2(1+ \omega^2 \tau^2)]$ is the effective diffusion coefficient. Moreover, we find that the distribution of the scaled area $z=[A-\la A\ra]/(2D_{\text{eff}}t)$ is described by the universal scaling function $F_{\text{ch}}(z)=\text{sech}(\pi z)$. From extensive numerical evidence, we conjecture the emergence of a new universal scaling function $G_{\text{ch}}(z)=\mathcal {N}/[e^{\alpha z} + e^{-\beta z}]$ for the distribution of the scaled winding angle $z=\Omega/[\ln t]$, where the parameters $\alpha$ and $\beta$ are model-dependent and $\mathcal{N}$ is the normalization constant. In the absence of chirality, i.e., $\omega=0$, the scaling function becomes $G_{\text{ch}}(z)=(\alpha/\pi)\,\mathrm{sech}(\alpha z)$.
23. Graph Structure Learning with Temporal Graph Information Bottleneck for Inductive Representation Learning
Authors: Jiafeng Xiong, Rizos Sakellariou β’
Published: 2025-08-20 β’
Source: arXiv
Temporal graph learning is crucial for dynamic networks where nodes and edges evolve over time and new nodes continuously join the system. Inductive representation learning in such settings faces two major challenges: effectively representing unseen nodes and mitigating noisy or redundant graph information. We propose GTGIB, a versatile framework that integrates Graph Structure Learning (GSL) with Temporal Graph Information Bottleneck (TGIB). We design a novel two-step GSL-based structural enhancer to enrich and optimize node neighborhoods and demonstrate its effectiveness and efficiency through theoretical proofs and experiments. The TGIB refines the optimized graph by extending the information bottleneck principle to temporal graphs, regularizing both edges and features based on our derived tractable TGIB objective function via variational approximation, enabling stable and efficient optimization. GTGIB-based models are evaluated to predict links on four real-world datasets; they outperform existing methods in all datasets under the inductive setting, with significant and consistent improvement in the transductive setting.
24. Data Fusion for High-Resolution Estimation
Authors: Amy Guan, Marissa Reitsma, Roshni Sahoo, Joshua Salomon, Stefan Wager β’
Published: 2025-08-20 β’
Source: arXiv
High-resolution estimates of population health indicators are critical for precision public health. We propose a method for high-resolution estimation that fuses distinct data sources: an unbiased, low-resolution data source (e.g. aggregated administrative data) and a potentially biased, high-resolution data source (e.g. individual-level online survey responses). We assume that the potentially biased, high-resolution data source is generated from the population under a model of sampling bias where observables can have arbitrary impact on the probability of response but the difference in the log probabilities of response between units with the same observables is linear in the difference between sufficient statistics of their observables and outcomes. Our data fusion method learns a distribution that is closest (in the sense of KL divergence) to the online survey distribution and consistent with the aggregated administrative data and our model of sampling bias. This method outperforms baselines that rely on either data source alone on a testbed that includes repeated measurements of three indicators measured by both the (online) Household Pulse Survey and ground-truth data sources at two geographic resolutions over the same time period.
25. Single-click protocols for remote state preparation using weak coherent pulses
Authors: Janice van Dam, Emil R. Hellebek, Tzula B. Propp, Junior R. Gonzales-Ureta, Anders S. SΓΈrensen, Stephanie D. C. Wehner β’
Published: 2025-08-20 β’
Source: arXiv
Remote state preparation (RSP) allows one party to remotely prepare a known quantum state on another party's qubit using entanglement. This can be used in quantum networks to perform applications such as blind quantum computing or long-distance quantum key distribution (QKD) with quantum repeaters. Devices to perform RSP, referred to as a client, ideally have low hardware requirements, such as only sending photonic qubits. A weak coherent pulse source offers a practical alternative to true single-photon sources and is already widely used in QKD. Here, we introduce two new protocols to the previously known protocol for RSP with a weak-coherent-pulse-based device. The known technique uses a double-click (DC) protocol, where a photon from both the server and the client needs to reach an intermediate Bell state measurement. Here, we add to that a single-click (SC) RSP protocol, which requires only one photon to reach the Bell state measurement, allowing for better performance in certain regimes. In addition, we introduce a double-single-click (DSC) protocol, where the SC protocol is repeated twice, and a CNOT gate is applied between the resulting qubits. DSC mitigates the need for phase stabilization in certain regimes, lowering technical complexity while still improving performance compared to DC in some regimes. We compare these protocols in terms of fidelity and rate, finding that SC consistently achieves higher rates than DC and, interestingly, does not suffer from an inherently lower fidelity than the DC, as is the case for entanglement generation. Although SC provides stronger performance, DSC can still show performance improvements over DC, and it may have reduced technical complexity compared to SC. Lastly, we show how these protocols can be used in long-distance QKD using quantum repeaters.
26. Universal and Transferable Adversarial Attack on Large Language Models Using Exponentiated Gradient Descent
Authors: Sajib Biswas, Mao Nishino, Samuel Jacob Chacko, Xiuwen Liu β’
Published: 2025-08-20 β’
Source: arXiv
As large language models (LLMs) are increasingly deployed in critical applications, ensuring their robustness and safety alignment remains a major challenge. Despite the overall success of alignment techniques such as reinforcement learning from human feedback (RLHF) on typical prompts, LLMs remain vulnerable to jailbreak attacks enabled by crafted adversarial triggers appended to user prompts. Most existing jailbreak methods either rely on inefficient searches over discrete token spaces or direct optimization of continuous embeddings. While continuous embeddings can be given directly to selected open-source models as input, doing so is not feasible for proprietary models. On the other hand, projecting these embeddings back into valid discrete tokens introduces additional complexity and often reduces attack effectiveness. We propose an intrinsic optimization method which directly optimizes relaxed one-hot encodings of the adversarial suffix tokens using exponentiated gradient descent coupled with Bregman projection, ensuring that the optimized one-hot encoding of each token always remains within the probability simplex. We provide theoretical proof of convergence for our proposed method and implement an efficient algorithm that effectively jailbreaks several widely used LLMs. Our method achieves higher success rates and faster convergence compared to three state-of-the-art baselines, evaluated on five open-source LLMs and four adversarial behavior datasets curated for evaluating jailbreak methods. In addition to individual prompt attacks, we also generate universal adversarial suffixes effective across multiple prompts and demonstrate transferability of optimized suffixes to different LLMs.
27. Carrier mobilities and electron-phonon interactions beyond DFT
Authors: Aleksandr Poliukhin, Nicola Colonna, Francesco Libbi, Samuel PoncΓ©, Nicola Marzari β’
Published: 2025-08-20 β’
Source: arXiv
Electron-phonon coupling is a key interaction that governs diverse physical processes such as carrier transport, superconductivity, and optical absorption. Calculating such interactions from first-principles with methods beyond density-functional theory remains a challenge. We introduce here a finite-difference framework for computing electron-phonon couplings for any electronic structure method that provides eigenvalues and eigenvectors, and showcase applications for hybrid and Koopmans functionals, and $GW$ many-body perturbation theory. Our approach introduces a novel projectability scheme based on eigenvalue differences and bypasses many of the limitations of the direct finite difference methods. It also leverages symmetries to reduce the number of independent atomic displacements, thereby keeping computational costs manageable. This approach enables seamless integration with established first-principles codes for generating displaced supercells, performing Wannier interpolations, and evaluating transport properties. Applications to silicon and gallium arsenide show that advanced electronic-structure functionals predict different electron-phonon couplings and modify band curvatures, resulting in much more accurate estimates of intrinsic carrier drift mobilities and effective masses. In general, our method provides a robust and accessible framework for exploring electron-phonon interactions in complex materials with state-of-the-art electronic structure methods.
28. Correct Black-Box Monitors for Distributed Deadlock Detection: Formalisation and Implementation (Technical Report)
Authors: RadosΕaw Jan Rowicki, Adrian Francalanza, Alceste Scalas β’
Published: 2025-08-20 β’
Source: arXiv
Many software applications rely on concurrent and distributed (micro)services that interact via message-passing and various forms of remote procedure calls (RPC). As these systems organically evolve and grow in scale and complexity, the risk of introducing deadlocks increases and their impact may worsen: even if only a few services deadlock, many other services may block while awaiting responses from the deadlocked ones. As a result, the "core" of the deadlock can be obfuscated by its consequences on the rest of the system, and diagnosing and fixing the problem can be challenging. In this work we tackle the challenge by proposing distributed black-box monitors that are deployed alongside each service and detect deadlocks by only observing the incoming and outgoing messages, and exchanging probes with other monitors. We present a formal model that captures popular RPC-based application styles (e.g., gen_servers in Erlang/OTP), and a distributed black-box monitoring algorithm that we prove sound and complete (i.e., identifies deadlocked services with neither false positives nor false negatives). We implement our results in a tool called DDMon for the monitoring of Erlang/OTP applications, and we evaluate its performance. This is the first work that formalises, proves the correctness, and implements distributed black-box monitors for deadlock detection. Our results are mechanised in Coq. DDMon is the companion artifact of this paper.
29. EECT: an Eccentricity Evolution Consistency Test to distinguish eccentric gravitational-wave signals from eccentricity mimickers
Authors: Sajad A. Bhat, Avinash Tiwari, Md Arif Shaikh, Shasvath J. Kapadia β’
Published: 2025-08-20 β’
Source: arXiv
Eccentric compact binary coalescences (CBCs) are expected to be observed in current and future gravitational-wave (GW) detector networks. However, it has been recently pointed out that a number of other physical and beyond-GR effects, could imitate, or be mimicked by, eccentric CBCs. In this work, we propose a conceptually simple but powerful method to directly confirm or reject the eccentric hypothesis, without needing to compare the hypothesis with the plethora of other possible hypotheses. The key idea is that while spurious non-zero values of eccentricity, at some reference frequency, could be acquired when a non-eccentric CBC with additional physical/beyond-GR effects is recovered with an eccentric CBC waveform model, the {\it evolution} of eccentricity with frequency will in general not be mimicked. We accordingly formulate an eccentricity evolution consistency test (EECT). The method compares the eccentricity recovered at some low frequency value (e.g, $10$ Hz), evolved to higher frequencies assuming GR, with eccentricities recovered at those same higher frequencies. Discrepancy between the two eccentricities at any reference frequency would violate EECT and indicate the presence of a mimicker. As a proof of concept, assuming a few eccentric CBC systems, quasi-circular CBCs with additional physics mimicking eccentricity, and an O4-like three-detector-network configuration, we demonstrate that our proposed method is indeed able to reject mimickers at $\geq 68\%$ confidence, while ensuring that truly eccentric CBCs satisfy EECT.
30. Physics-Informed ML Exploration of Structure-Transport Relationships in Hard Carbon
Authors: Nikhil Rampal, Stephen E. Weitzner, Fredrick Omenya, Marissa Wood, David M. Reed, Xiaolin Li, Jonathan R. I. Lee, Liwen F. Wan β’
Published: 2025-08-20 β’
Source: arXiv
Sodium-ion batteries are a cost-effective and sustainable alternative to lithium-ion systems for large-scale energy storage. Hard carbon (HC) anodes, composed of disordered graphitic and amorphous domains, offer high capacity but exhibit complex, poorly understood ion transport behavior. In particular, the relationship between local microstructure and sodium mobility remains unresolved, hindering rational performance optimization. Here, we introduce a data-driven framework that combines machine-learned interatomic potentials with molecular dynamics simulations to systematically investigate sodium diffusion across a broad range of carbon densities and sodium loadings. By computing per-ion structural descriptors, we identify the microscopic factors that govern ion transport. Unsupervised learning uncovers distinct diffusion modes, including hopping, clustering, and void trapping, while supervised analysis highlights tortuosity and NaNa coordination as primary determinants of mobility. Correlation mapping further connects these transport regimes to processing variables such as bulk density and sodium content. This physics-informed approach establishes quantitative structure-transport relationships that capture the heterogeneity of disordered carbon. Our findings deliver mechanistic insights into sodium-ion dynamics and provide actionable design principles for engineering high-performance HC anodes in next-generation battery systems.
31. Power-Law Interactions Stabilize Time Crystals Realizing Quantum Energy Storage and Sensing
Authors: Ayan Sahoo, Debraj Rakshit β’
Published: 2025-08-20 β’
Source: arXiv
We study discrete time-crystalline (DTC) phases in one-dimensional spin-1/2 chains with power-law interactions under periodic Floquet driving. By generalizing Stark localization to power-law interaction profiles, we identify robust period-doubled dynamics across a wide range of interaction exponents, stabilized by the interplay between coherent driving and spatially varying coupling. Within the DTC phase, the energy stored in the system, interpreted as a quantum battery, increases superlinearly with system size, although no scaling advantage persists in normalized power. Beyond energy storage, we demonstrate that the DTC phase supports enhanced quantum sensing. The quantum Fisher information associated with estimating timing deviations in the drive scales superextensively with system size, surpassing the Heisenberg limit. The degree of quantum advantage can be tuned by varying the interaction exponent, though DTC behavior remains robust throughout. Our results position power-law interacting Floquet systems as robust platforms for storing quantum energy and achieving metrological enhancement.
32. Modeling tails of escaping gas in exoplanet atmospheres with Harmonica
Authors: Carlos GascΓ³n, Mercedes LΓ³pez-Morales, Shreyas Vissapragada, Morgan MacLeod, Hannah R. Wakeford, David Grant, Ignasi Ribas, Guillem Anglada-EscudΓ© β’
Published: 2025-08-20 β’
Source: arXiv
Exoplanets that reside close to their host stars, and therefore receive substantial amounts of X-ray and ultraviolet radiation, are prone to suffer from strong atmospheric escape. This can lead to the creation of an envelope of escaping gas along the planet's orbital trajectory, often referred to as a tail. When transiting in front of their host star, these tails can not only produce larger depths in the transit light curves, but also introduce significant asymmetries between ingress and egress. Using the publicly available software Harmonica, we present a method to model the light curves of transiting planets surrounded by extended envelopes of escaping gas, and subsequently infer the shape and size of the latter. We apply this method to the JWST NIRISS/SOSS observations of HAT-P-18b, which show pronounced helium tail features in its spectroscopic light curve of the metastable helium triplet at 10830 \r{A}. Our model reveals that, in order to fit the observed light curve of HAT-P-18b, the planet must possess a trailing helium tail of $15.79^{+1.14}_{-1.05}$ planetary radii. We carry out injection-recovery tests to validate the effectiveness of the proposed methodology. We demonstrate that, with sufficient precision, we would be able to fit a multi-layer envelope to the data, which would provide insight into the relative radial variations in the opacity profile.
33. A State-Space Representation of Coupled Linear Multivariate PDEs and Stability Analysis using SDP
Authors: Declan S. Jagt, Matthew M. Peet β’
Published: 2025-08-20 β’
Source: arXiv
Physical processes evolving in both time and space are often modeled using Partial Differential Equations (PDEs). Recently, it has been shown how stability analysis and control of coupled PDEs in a single spatial variable can be more conveniently performed using an equivalent Partial Integral Equation (PIE) representation. The construction of this PIE representation is based on an analytic expression for the inverse of the spatial differential operator, $\partial_s^{d}$, on the domain defined by boundary conditions. In this paper, we show how this univariate representation may be extended inductively to multiple spatial variables by representing the domain as the intersection of lifted univariate domains. Specifically, we show that if each univariate domain is well-posed, then there exists a readily verified consistency condition which is necessary and sufficient for existence of an inverse to the multivariate spatial differential operator, $D^\alpha=\partial_{s_1}^{\alpha_1}\cdots\partial_{s_N}^{\alpha_N}$, on the PDE domain. Furthermore, we show that this inverse is an element of a $*$-algebra of Partial Integral (PI) operators defined by polynomial semi-separable kernels. Based on this operator algebra, we show that the evolution of any suitably well-posed linear multivariate PDE may be described by a PIE, parameterized by elements of the PI algebra. A convex computational test for PDE stability is then proposed using a positive matrix parameterization of positive PI operators, and software (PIETOOLS) is provided which automates the process of representation and stability analysis of such PDEs. This software is used to analyze stability of 2D heat, wave, and plate equations, obtaining accurate bounds on the rate of decay.
34. Quantum mechanics, non-locality, and the space discreteness hypothesis
Authors: W. A. ZΓΊΓ±iga-Galindo β’
Published: 2025-08-20 β’
Source: arXiv
The space discreteness hypothesis asserts that the nature of space at short distances is radically different from that at large distances. Based on the Bronstein inequality, here, we use a totally disconnected topological space $\mathcal{X}$ as a model for the space. However, we consider the time as a real variable. In this framework, the formalism of Dirac-von Neumann can be used. This discreteness hypothesis implies that given two different points in space, there is no continuous curve (a world line) joining them. Consequently, this hypothesis is not compatible with the theory of relativity. We propose $\mathbb{R}\times(\mathbb{R}\times\mathcal{X})^{3}$ as a model of a space-time. For simplicity, we work out our models using $\mathbb{R}\times(\mathbb{R}\times\mathcal{X})$ as the configuration space. Quantum mechanics (QM), in the sense of Dirac-von Neumann, on the Hilbert space $L^{2}(\mathbb{R}\times\mathcal{X})$ is a non-local theory: the Hamiltonians are non-local operators, and thus, spooky action at a distance is allowed. The paradigm asserting that the universe is non-locally real implies that the proposed version of QM admits realism. This version of QM can be specialized to standard QM by using Hamiltonians acting on wavefunctions supported on the region $\mathbb{R}\times\mathbb{R}$. We apply the developed formalism to the measurement problem. We propose a new mechanism for the collapse of the wavefunction. The mechanism resembles the one proposed by Ghirardi, Ramini, and Weber, but there are significant differences. The most important feature is that the Schr\"{o}dinger equation describes the dynamics at all times, even at the moment of measurement. We also discuss a model for the two-slit experiment, where bright and dark states of light (proposed recently) naturally occur.
35. An Investigation Into Secondary School Students' Debugging Behaviour in Python
Authors: Laurie Gale, Sue Sentance β’
Published: 2025-08-20 β’
Source: arXiv
Background and context: Debugging is a common and often frustrating challenge for beginner programmers. Understanding students' debugging processes can help us identify the difficulties and misunderstandings they possess. However, we currently have limited knowledge of how secondary students debug in a text-based language, a medium through which millions of students will learn to program in the future. Objectives: In this paper, we investigate the debugging behaviour of K-12 students learning a text-based programming language, as part of an effort to shape how to effectively teach debugging to these students. Method: We collected log data from 73 students attempting a set of debugging exercises using an online code editor. We inductively analysed these logs using qualitative content analysis, generating a categorisation of the debugging behaviours observed. Findings: A range of behaviours were exhibited by students, skewed towards being ineffective. Most students were able to partially locate errors but often struggled to resolve them, sometimes introducing additional errors in the process. We argue that students struggling to debug possess fragile knowledge, a lens through which we view the results. Implications: This paper highlights some of the difficulties K-12 learners have when debugging in a text-based programming language. We argue, like much related work, that effective debugging strategies should be explicitly taught, while ineffective strategies should be discouraged.