Dynamical Systems
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Saturday 7 November
We propose a single formal object, the generalized particle system (GPS), under which self-propelled particle systems (SPPS) and agent-based models (ABMs) are instances of one construction. A GPS is a typed finite population, a per-particle state space with arbitrary internal structure, a global environment, a neighbourhood map that may depend on the configuration, and type-indexed local rules; these assemble into a generator whose iteration is a monoid action, i.e. an autonomous dynamical system. Update schedules and stochastic rules are added at the same generality, so sequential and noisy models (the original Primordial Particle System, Vicsek) are covered, not approximated. The construction isolates the one ingredient separating the two communities: in the classical ABM reading the dependency graph is fixed; in an SPPS it is a function of the state. Under an exchangeability axiom (rules see neighbours only through the multiset of their types and states) the generator is equivariant for type-preserving permutations and descends to a dynamical system on counting measures, the setting of kinetic limits of swarming models. We read the tuple as a metamodel: its structural components coordinatize a space of models inside which numerical parameters live. Six pre-registered experiments on one engine implementing the generator test the claims: freezing the neighbourhood map destroys alignment ($\varphi = 0.91 \to 0.04$) while yielding a different, non-aligned aggregation; the realized neighbourhood graph rewires up to 64% of its edges per step, its rewiring rate anti-correlated with order ($-0.99$); a topological instead of metric neighbourhood reshapes a flocking phase diagram; the same rule with cohesion and separation off shows a Vicsek-type transition whose susceptibility peak grows with system size. One prediction failed and one subsumption claim is unconfirmed; both are reported.
Physics Education
Hands-on fabrication of 2D material field-effect transistors for the classroom using off-the-shelf ionic gel
We introduce a hands-on experimental method that enables students to fabricate and operate van der Waals semiconductor-based field-effect transistors (FETs) using readily available materials and minimal infrastructure. The approach is based on ionic-gel gating using commercially available hydrogel conductors originally developed for muscle electrostimulation devices. For device fabrication we rely on commercially available thin-film interdigitated electrode arrays and for the semiconductor channel, we deposited MoS2 flakes produced by high-throughput mechanical exfoliation. By avoiding the need for cleanroom processing, vacuum deposition, or custom electrolytes, this method allows functional FETs based on van der Waals materials to be realized in a standard teaching laboratory. The experiment is well suited for undergraduate and early graduate courses and is intended to complement concepts introduced in electronics and solid-state physics by providing students with first-hand experience in the fabrication and electrical characterization of emerging electronic devices. Representative electrical measurements demonstrate gate-dependent modulation of the channel conductance and illustrate characteristic features of ionic-gated transistor operation. Overall, the proposed platform provides an accessible route for students to fabricate, characterize, and analyze field-effect devices without access to conventional semiconductor foundry.
Read the full abstract
Computational Engineering
The Market Is the Reward: Design Markets, Not Just Traders for Human-AI Collective Intelligence
Learning agents are becoming market participants, yet machine learning still treats the market as a fixed environment, although its payoff rule is the reward agents learn from. This position paper argues that financial AI should build HyperMarkets for human-AI collective intelligence, and not only the agents that trade in them: markets whose liquidity provision, information production, outcome determination, and trading rule are performed or adapted by learning agents and designed jointly toward explicit market-level objectives. In a pre-registered simulation with Claude agents trading for principals, joint design raises a market-level objective by 79% over separately tuned functions, through a trading rule that helps only when slow traders are present, and how evidence is paid for decides whether a stronger agent earns more for its principal. We state when fixed rules or non-market aggregation should be preferred and what findings would refute the case.
Read the full abstract
High Energy Astrophysical Phenomena
Standalone Validation of a Pulsar Evolution Subroutine for NBODY6++GPU: Seven Test Scenarios
Pulsars in globular clusters provide critical constraints on the evolution of compact objects. However, the treatment of neutron stars in standard N-body codes is outdated, and requires a better implementation of magnetic field decay, spin evolution, and accretion driven recycling that shape their observable properties. We present a standalone validation of a pulsar evolution subroutine for NBODY6++GPU, systematically tested across seven controlled scenarios prior to integration into full cluster simulations. These scenarios isolate distinct evolutionary regimes, including young pulsar spin down, millisecond pulsar weak torque evolution, accretion driven recycling in X-ray binaries, environmental torques in dense cores, wide binary systems, and compact binaries in the dipole spin-down regime. Through two separate numerical implementations, we achieve cross validation demonstrating better than 0.1% agreement across all physical quantities and timescales. Our validation confirms that analytical solutions are reproduced with high precision for isolated pulsars, gigayear scale evolution (tau_c ~ 1.0 Gyr) matches observations of millisecond pulsars in clusters like 47 Tucanae, and accretion driven recycling operates with physically realistic timescales, yielding a final period of 1.0 ms and a period derivative of approximately -5.9e-19 s/s, with accretion dominance by a factor of about 14. We demonstrate that environmental torques remain negligible even in the densest cluster environments, while compact binary spin down is validated; gravitational wave orbital decay is deferred to future work. All validation codes, test suites, and integration instructions are publicly available to support future applications.
Read the full abstract
Mathematical Software
Quadratic Bendixson criterion for delay systems: a numerical implementation
We present a software implementation (as a Python package) of approximation schemes for verifying frequency inequalities arising from the quadratic Bendixson criterion for systems of delay equations (with a single delay in the linear part). In applications to dissipative nonlinear autonomous systems, such inequalities ensure global stability. The theoretical foundations of this approach were developed in our related work. In particular, it was shown that the transfer operators involved into the frequency inequalities can be represented as integral operators with bounded kernels of the Hardy–Volterra type. Here, we use this representation to construct approximation schemes based on finite-dimensional truncations of the transfer operators. Examples of using the package are provided.
Read the full abstract
Quantum Physics
Time Running Backwards, and Where It Stops: Superoscillations of Proper Time and the Memory of the Environment
A quantum system prepared in a superposition of proper times - a clock whose worldline is superposed, or a clock at rest in a superposed gravitational potential - evolves, in each branch, by the same dynamics for a different duration. Post-selecting the control that sets the proper time turns the superposition into a superoscillation of the time-dilation factor: the system is found to have evolved for a proper time $\alpha_w t$, where the weak value $\alpha_w$ may exceed unity, vanish, or be negative - the quantum time-translation machine of Aharonov, Anandan, Popescu and Vaidman. We ask how far back such a machine can run an irreversible process. The test system is a sealed cell - excited atoms, the radiation field they emit into, and the absorbing walls, carried as one quantum object - and the witness is projective: atoms that had begun to decay before the run are found, in the post-selected sub-ensemble, more excited than when inserted. The answer is exact and depends only on the atoms' survival amplitude. With two branches the witness holds only while the decay is younger than the memory time of the environment, after which the emission becomes Markovian. Beyond it, the post-selection amplifies the minute difference between the contents of the cell in the two branches, set by the full energy spread of the atoms and their field rather than by the linewidth, and the witness fails. Reversing a decay past the memory time requires many branches, an environment with a strictly bounded spectrum, and a post-selection probability that falls exponentially with the number of environmental oscillations to be undone. The obstacle to reversal is therefore not entropy but bandwidth - the same bandwidth that makes the emission irreversible in the first place. We give four ways to superpose a clock rate, from hollow massive shells to inertial velocities.
Read the full abstract
Artificial Intelligence
TrustMod-SM: A Multi-Axis Benchmark for Evaluating Trustworthiness of LLMs in Social Media Content Moderation
Large Language Models (LLMs) show strong potential for social media content moderation, but their trustworthiness beyond overall accuracy remains underexplored. We introduce TrustMod-SM, a benchmark of roughly 29,000 instances across eight datasets, to evaluate LLM moderators on five dimensions: trustfulness, fairness, safety, robustness, and context integrity. Evaluating 16 open-weight models (0.5B-284B parameters, spanning text-only and vision-language architectures) and one closed frontier model, we find persistent trustworthiness gaps. Higher detection accuracy often coincides with greater demographic disparity, and models frequently comply with concealment or exaggeration prompts, assign high confidence to wrong predictions, rarely escalate ambiguous cases, and struggle to separate hate speech from counter-speech or reclaimed language (especially below 8B parameters). Scaling does not uniformly resolve these issues, and no single model excels across all five dimensions.
Read the full abstract
Computer Vision
Evaluation protocols for historical script classification: the cost of manuscript-level leakage and the role of soft labels
A paleographer applies a script type classifier to unseen manuscripts. We measure how much testing on other pages of training manuscripts inflates accuracy. On VML-HP-ext (715 pages, 171 Hebrew manuscripts, 14 script classes) we compare page-level and manuscript-level splits. Page-level accuracy drops from 97% to 54% (59% at manuscript level), and is 39% on the official blind test set. The gap of 40 to 47 percentage points (pp) at patch, page and manuscript level is statistically significant and reproduces with ResNet18; on binarized patches (ink and background only) it is 34 to 42 pp, also significant. On unseen manuscripts the model is overconfident: mean confidence 70% against 54% correct pages; patch-level expected calibration error (ECE, mean gap between confidence and accuracy) rises from 0.065 to 0.444. Binarization leaves patch-level overconfidence but cuts it markedly at page level. Confidence still helps triage: in the more confident half of pages 67% are correct, 41% in the rest. Training on expert soft labels showed no evidence of an accuracy difference, though a small page-level loss cannot be ruled out. The predicted distributions are closer to the expert's assessment. ECE is lower at patch level (0.21 to 0.31) and in most runs at page level, but not significantly at manuscript level. Label smoothing (blurring the label, no expert input) also lowers ECE but comes no closer to the expert. Temperature scaling (post-training confidence correction) helps significantly only at patch level. On 97 pages of 16 manuscripts with mixed expert assessment, soft labels cut divergence from the expert by 22 to 30% (significant in four of five runs); the point estimate of accuracy is 16.5 to 23.7 pp below hard labels, but with 16 manuscripts the loss can neither be confirmed nor ruled out. We propose recommendations for the evaluation protocol; code will be released with the paper
Read the full abstract
Materials Science
Wafer-scale MoS2 and HZO-based Ferroelectric Field Effect Transistors with 4.4 V Memory Window for Back-End Compatible Multibit Memory
Ferroelectric field-effect transistors (FeFETs) with low-power operation and non-volatile data retention enabled by stable ferroelectric polarization switching, have emerged as promising alternatives to conventional floating-gate transistors for next-generation low-power memory applications. A FeFET with two-dimensional (2D) channel provides the opportunity for better electrostatic gate control and scalability beyond silicon. In this work, we report wafer scale 2D semiconductor-based FeFETs with Hf0.5Zr0.5O2 (HZO) gate stack showing unusual p-type charge transport and stable multibit memory properties. FeFETs based on 2-3 monolayer Molybdenum disulfide (MoS2) channel and a 20 nm HZO back gate oxide layer exhibit a large hysteresis with a memory window spanning 4.4 V under a operating voltage range of +3.8 to -3.8 V, and an ION/IOFF ratio of 10^5. We demonstrate that the polarization charge-induced modification of Schottky barrier width strongly affects the hole transport and memory properties of MoS2-HZO FeFETs. We further demonstrate several advantages and unique features of the FeFETs such as nondestructive data readout, non-volatile data retention under zero gate bias, low operation voltage, high write endurance and ultra-fast switching speed. These 1T memory cells can simplify the dense 3D integrated array architecture and offer additional practical functionalities, such as fast programming and erasing of large data arrays that are absent for many conventional memories. The demonstrated MoS2 and HZO-based FeFET shows great potential for on-chip complementary metal-oxide semiconductor (CMOS)-back-end-of-line (BEOL) compatible circuits for multifunctional devices, especially for 3D monolithically integrated architectures for AI workload.
Read the full abstract
Classical Physics
A constrained mixed variational formulation with Legendre-dual defects for incremental irreversible thermodynamics
We present a constrained mixed variational formulation for incremental irreversible thermodynamics. A single augmented action combines a linearized contact-increment penalty, equilibrium and dissipative Fenchel–Young defects, a storage–transport functional, and multipliers enforcing the constitutive increments. Under frozen-coefficient and boundary assumptions, stationarity yields the linearized contact condition, the Tisza relation, the Onsager flux–force law, and the implicit balance equation. An exact-elimination theorem shows that the contact and multiplier fields can be reconstructed algebraically from a solution of the reduced weak balance problem. We also prove off-shell invariance of the equilibrium–storage–multiplier sector under an equilibrium-sector partial Legendre transformation; invariance of the full augmented action is not claimed. For Fourier heat conduction, we derive a weighted drift identity and a per-step energy-drift bound for the energy-representation three-term incremental functional. In one spatial dimension, a small-time-step estimate yields a uniform $H^1$ bound and a discrete $L^2(0,T_f;H^2)$ bound, leading to cumulative $O(\delta t)$ drift over fixed time intervals. A conservative mixed discontinuous Galerkin (DG0)–Raviart–Thomas (RT0) discretization with fully implicit conductivity $\kappa(T)=1+T^2$ preserves discrete energy to roundoff. A separate lumped-mass piecewise-linear benchmark reproduces the predicted finite-step drift.
Read the full abstract
Artificial Intelligence
BuySideFlow: A Point-in-Time Benchmark for LLM Agents in Investment Research
Large language models (LLMs) can generate SQL and SQL–Python workflows for investment research, but ensuring point-in-time (PIT) validity remains challenging. Agents must enforce disclosure cutoffs and select the appropriate historical states and record versions throughout data retrieval and financial analysis. We introduce BuySideFlow, a benchmark of 408 practitioner-derived Chinese-language tasks spanning equities, funds, fixed income, and macroeconomics over a fixed snapshot from a licensed production financial database. We provide executable references and a taxonomy of temporal errors, pairing result checks with expert audits of executed workflows. We evaluate nine agent configurations on all tasks, with result accuracy ranging from 16.2% to 36.8%. An audit of 900 runs on 100 shared tasks confirms PIT violations in 193 of the 380 runs that pass result checks (50.8%), showing that output agreement alone does not establish temporal validity.
Read the full abstract
Fluid Dynamics
A Beltrami Mode in Extreme Fractional Hydrodynamics under Stochastic Forcing
We construct an exact stochastic solution of the extreme order endpoint of adaptive fractional Navier Stokes (AFNS) hydrodynamics on the three-dimensional periodic torus. A helical Beltrami eigenfield reduces the nonlinear vector equation to a scalar Ornstein Uhlenbeck process when the white-in-time forcing is aligned with that eigenfield. The resulting invariant Gaussian law yields explicit formulas for the stationary energy, its fluctuations, temporal correlations, circulation, and helicity. Under the same forcing, the corresponding classical Navier Stokes mode has a stationary energy that diverges as the Reynolds number tends to infinity, while the Euler limit exhibits linear growth of the mean energy and has no invariant probability law. The construction isolates, without a cascade hypothesis, the statistical effect of the finite fractional dissipation retained at the extreme endpoint of AFNS.
Read the full abstract
Machine Learning
Exploration via Exploitation: The Blessing of Reward Diversity in Personalized Federated RL
While federated reinforcement learning (RL) has traditionally guaranteed high efficiency in homogeneous environments, collaboration is often hindered when agents possess distinct goals. In this paper, we flip this paradigm, framing reward heterogeneity not as a hurdle to consensus, but as a structural asset that catalyzes exploration. We first propose Personalized Federated Upper Confidence Bound Value Iteration (PF-UCBVI), which achieves optimal linear speedup by decoupling shared dynamics from personalized objectives. To eliminate the risks of explicit exploration, we then introduce Personalized Federated Exploration-Free Value Iteration (PF-EFVI), a purely greedy algorithm that leverages reward diversity to ensure state-action coverage. We prove that PF-EFVI attains logarithmic regret without explicit exploration bonuses under sufficient reward diversity. Our results reveal that agent disagreement shifts the exploration burden from temporal complexity to spatial diversity, enabling efficient collective learning without explicit exploration.
Read the full abstract
Fluid Dynamics
Neural Networks with Local Converging Inputs for Solving the 3D Stokes Equations on Unstructured Meshes: Novel Techniques and Advances
Neural-network surrogates for fluid-flow PDEs usually learn the solution globally, which requires large networks and many high-fidelity training simulations. Neural networks with local converging inputs (NNLCI) instead work locally. They predict the high-fidelity solution at a point from two converging coarse-grid solutions sampled near that point, which keeps the network small, needs little training data, and generalizes across geometries. Here, the two low-fidelity meshes, LF1 and LF2, have on average about 300 and 730 tetrahedra, compared with about 155,000 in the high-fidelity mesh (HF). In this work, we develop an NNLCI framework for three-dimensional Stokes equations on unstructured tetrahedral meshes and introduce two techniques for building local inputs near complex curved boundaries. First, input patches may extend beyond the boundary, with the outside values obtained from a local least-squares fit, so that every high-fidelity point can be predicted, including those next to the wall. Second, the local mesh spacing is estimated from neighboring centroids, which sets the stencil size consistently on unstructured meshes. For pipe flows with one to three bumps, including a circumferentially rotated bump, NNLCI reduces the relative $L^2$ error of the finer low-fidelity solution by $90\%$ to $94\%$, evaluated at all high-fidelity points. Both techniques carry over to other NNLCI problems with smooth solutions on point-cloud discretizations.
Read the full abstract
Physics and Society
Evolution of cooperation in enterprise resource-sharing networks with dynamic participation and adaptive decision-making
Resource sharing among enterprises is an important mechanism for improving resource utilization efficiency and facilitating collaborative development. However, the emergence and sustainability of cooperative behavior depend critically on whether enterprises choose to participate in resource-sharing activities and how they balance cooperation and competition in their decision-making processes. To investigate these issues, we develop an evolutionary cooperation model on the enterprise resource-sharing network that integrates participation status, decision-making modes, and cooperative strategies within a unified framework. Specifically, enterprises dynamically switch between active and dormant states according to their aspiration levels, while Q-learning is employed to enable adaptive selection between payoff-driven and conformist decision-making modes. Meanwhile, interactions between resource-sharing and non-sharing in active enterprises are characterized by the snowdrift game. Through extensive simulations, we systematically examine the effects of the payoff parameter, the reinforcement-learning parameters, the participation and Q-learning mechanism, and network size on enterprise participation and the evolution of cooperation. The results show that an intermediate aspiration level is most favorable for sustaining enterprise participation and facilitating the emergence of cooperation. Moreover, the dynamic activation mechanism effectively mitigates the decline in cooperation under high cost-to-benefit ratios, allowing the system to sustain a decent level of cooperative behavior. These findings highlight the important roles of dynamic participation and adaptive decision-making in shaping the evolution of resource-sharing cooperation and provide a new modeling framework to understand the emergence and persistence of cooperation in complex enterprise networks.
Read the full abstract
Artificial Intelligence
RBLA+: Efficient Rank-Specific LoRA Adapter Extraction for Federated Fine-Tuning with Heterogeneous Clients
Federated fine-tuning with low-rank adaptation (LoRA) reduces communication and computational overhead by training and exchanging compact trainable LoRA factors. Clients may use different LoRA ranks to meet adapter communication budgets and accommodate different local adaptation needs. The server must process these differences during aggregation and construct adapters at each client's rank for local training. We distinguish pre-aggregation alignment, which handles factor dimensions and scaling conventions before aggregation and rank-specific LoRA adapter extraction, which constructs client-sized pairs from the aggregated global model. Two common approaches to rank-specific extraction are direct rank truncation and Singular Value Decomposition (SVD) of the dense update. Direct prefix truncation keeps the leading columns of one aggregated factor and the corresponding rows of the other. It is cheap but can increase reconstruction error. The dense SVD approach first forms the product of aggregated LoRA factors. It then decomposes this matrix and retains the leading singular components to construct adapters at the target rank. This gives the optimal rank-constrained approximation in Frobenius norm but incurs significant computational overhead. To address this trade-off, we propose RBLA+ as an extension of RBLA. It uses established compact QR/core-SVD decomposition to minimize Frobenius error for the extracted LoRA update without forming the dense update. It also accounts for client and server scaling conventions during upload and broadcast. Across vision and language tasks, RBLA+ improves accuracy over RBLA in multiple heterogeneous settings. In CPU benchmarks with synthetic adapters, RBLA+ has lower extraction time than the dense partial-SVD baseline. RBLA+ preserves each client's prescribed adapter size and per-round communication payload.
Read the full abstract
Artificial Intelligence
IIT Patna at "ICPR 2026 TVRID Competition on Privacy-Preserving Person Re-Identification from Top-View RGB-Depth Camera"
Privacy-aware person re-identification (Re-ID) from top-view cameras replaces frontal appearance with anthropometric shape, while depth sensing further reduces color-based identity cues. The ICPR 2026 TVRID competition evaluates this setting using 86 identities captured by four synchronized Intel RealSense D455 cameras with RGB and Depth streams across flat, ascent, descent, and oblique viewpoints. Performance is measured using mAP and CMC@1 across within-camera in/out, ascent-descent, and flat-oblique scenarios. The competition had 147 registered participants and 33 final submissions. Three teams from IIT Patna participated: 'abhinavkumar03' and 'sjkdgfhl_' ranked 5th and 7th among 31 ranked participants, and 'vivek katuri' reached 77.67% CMC@1 / 85.20% mAP on the RGB track. This paper presents the methods of the two top-10 teams. Methodology 1 (abhinavkumar03) achieves 93.51% mAP and 90.42% CMC@1 on Depth, ranking 4th of 28, while Methodology 2 (sjkdgfhl_) reaches 99.63%/99.35% on RGB and 86.26%/80.34% on Depth. Both substantially outperform the 57.19% track-average Depth mAP, with Methodology 1 yielding a 7.3-point gain over Methodology 2. These results demonstrate the importance of depth representation and supervision for privacy-preserving Re-ID.
Read the full abstract
Artificial Intelligence
Monitoring One Channel Is Not Communication Control: Causal Evidence for Adaptive Reallocation of Joint Message–Route Codes
Monitoring a visible channel can change how learned agents communicate without removing why communication is useful. We test this distinction in an end-to-end game where a sender and receiver learn a discrete protocol and route policy without communication supervision. In two independent 20-seed batches, 20/20 and 19/20 protocols form. Online partial coverage increases shadow-route share over no monitoring by .701 [.574,.826] and .575 [.412,.737], while task accuracy remains .905/.913 and signed frozen message contribution remains .592/.563. Complete coverage removes most selective migration without removing communication. Frozen interventions, including a fresh pre-outcome confirmation, show that route is semantic: joint message–route shuffling changes .668 of receiver predictions and route alone retains .388 accuracy versus .25 chance. A corrected explicit-silence cost makes 18/20 policies fully silent and lowers mean sending to .050, but its strict complete-suppression gate fails because two seeds retain communication. Failed semantic-evasion and operational gates further bound the result. The evidence supports adaptive reallocation of a joint message–route code under partial monitoring, not open-world route discovery, emergent encryption, or a universal defense.
Read the full abstract
Artificial Intelligence
Embodied Science Agents: Making Embodied Experimental Design Learnable
Recent advances in large models have accelerated AI for Science toward agentic science, where AI agents autonomously conduct scientific discovery. Realising this vision requires closing the loop between computational reasoning and physical experimentation, positioning self-driving laboratories (SDLs) as critical enablers. However, existing SDLs exhibit a fundamental paradox: foundation model integration has increased linguistic sophistication while severing physical grounding. By treating models as disembodied planners that output symbolic workflows for separate robotic systems, current architectures inherently block the development of embodied experimental intuition. We argue that overcoming this paradox requires embedding learnability along three interdependent axes: cognitive learnability (learning what to investigate), experimental learnability (learning where to search), and embodied learnability (learning how to act). Building on this view, we introduce the Embodied Science Agent (ESA), which unifies digital intelligence with physical embodiment through three synergistic components: a Digital Agent that provides cognitive and experimental learnability via memory-based reasoning and Bayesian optimisation, an Embodied Agent that provides embodied learnability via generalisable instrument control, and an Embodied Unified Interface for seamless coordination. We validated ESA in autonomous wastewater treatment experiments, navigating over 200 million possible conditions. Through 15 continuous rounds, ESA raised the final solution transmittance from 82% to 99.7%, corresponding to a 98.5% reduction in residual dye concentration and near-complete decolourisation, while its learned visuomotor policies generalised across varying instrument placements, a promising step toward next-generation SDLs for autonomous scientific discovery.
Read the full abstract
Information Retrieval
DocFS: File Systems as Agentic Reasoning Environments for Complex Multimodal Document Question Answering
VLM-based agents solve complex tasks by iteratively interacting with environments, and their effectiveness depends heavily on how the environment exposes information and provides tools for acting on it. However, constructing an effective environment for complex multimodal document question answering remains challenging: documents usually span many pages, with multimodal contents distributed across hierarchical sections and linked by cross-modal associations, while existing methods organize documents into structured representations that address only the information side of the environment and lack tools aligned with the structure, limiting agents' ability to fully exploit it. We observe that file systems naturally combine structure organization with aligned tools: directories and files encode document structure, while file operations provide rich tools to access file contents. Building on this insight, we propose DocFS, a file-system-based agentic framework that reorganizes raw documents as file systems, where directories mirror the section hierarchy and explicit links connect text with figures, tables, and equations. Based on well-established file system tools such as navigation, search, and reading, agents can iteratively explore the document structure, locate relevant evidence, and understand fine-grained local information. Paired with Qwen3.5-Plus, DocFS achieves state-of-the-art performance on MMLongBench-Doc and FinRAGBench-V, outperforming the strongest baselines by 2.9 and 4.3 percentage points, respectively, and is up to 10.3x faster with up to 10.1x fewer tokens than the strongest baseline during reasoning.
Read the full abstract
Machine Learning
ADIndex: Layer-Adaptive Sparse Attention Indexing via Group-Level Distillation
DeepSeek Sparse Attention (DSA) reduces the computational cost of attention through token-level selection. However, its indexer still scores the entire causal prefix for each query, making indexing a dominant source of latency in long-context inference. Notably, the indexer alone can account for more than 50% of total prefill GPU time at long context lengths. To address this bottleneck, we propose ADIndex, an Adaptive and Distillation-driven Indexing method for token-level sparse attention. ADIndex integrates three key components: (i) a layer-wise calibration strategy that adaptively accommodates cross-query retrieval overlap, motivated by our observation of pronounced inter-layer heterogeneity; (ii) a group-level distillation scheme that leverages this heterogeneity to efficiently train adaptive layer-wise group indexers, capturing salient retrieval signals from individual queries; and (iii) a low-latency inference strategy that reduces repeated full-prefix scans through token bank sharing within query groups while retaining query-specific refinement. Experiments on DeepSeek-V3.2 and GLM-5.1 demonstrate that ADIndex preserves near-lossless task performance on LongBench, LongBench v2, and RULER while delivering substantial efficiency gains over standard DSA, including prefill and decoding kernel speedups of up to 6.22x and 3.40x, respectively, at a context length of 256K across the layer-wise reuse configurations, as well as an end-to-end serving speedup of 1.54x at a context length of 128K.
Read the full abstract
Machine Learning
Predictive Surprise as Self-Grounding Concept Bottleneck for Interpretable Time Series
Time-series classifiers deployed in risk-sensitive domains must be (1) accurate, (2) interpretable, and (3) able to abstain when evidence is weak. Existing methods do not deliver all 3 at once, and the obstacle is structural, rather than logistical. Meaningful temporal patterns cannot be defined without knowing segment boundaries, yet meaningful boundaries cannot be placed without knowing which patterns to expect. Existing approaches break this circularity by using fixed windows or expert-annotated vocabularies, thereby sacrificing at least one of the 3 properties. We introduce ConceptTime, which closes the loop through a single self-supervised signal. A frozen probabilistic forecaster predicts a distribution over the next observation at every timestep of the sequence. We call the mismatch between its prediction and the realized signal the predictive surprise. This signal both locates segment boundaries and characterizes each resulting segment through the forecaster. Segments are thus grounded by how they behaved relative to expectation, not by what they look like. These summaries cluster into a vocabulary of dynamical regimes that we call concepts. A lightweight head classifies from the concept sequence, and the distance to the nearest concept yields reliability by design. Across 17 public benchmarks, ConceptTime is competitive with state-of-the-art black-box accuracy, beats every interpretable baseline, and retains near-98% accuracy on Epilepsy with 1% of labels. It further remains competitive with softmax, prototype-distance, and SHAP/TimeX/TimeX++ on OOD detection and deletion faithfulness, without a single annotation, language model call, or domain expert. Code is available at: https://github.com/Sachithx/ConceptTime.git.
Read the full abstract
Artificial Intelligence
Physics2Weight: Direct Neural Network Architecture and Weight Generation from Physical Laws Using Large Language Models
Recent advances in large language models (LLMs) have enabled weight-space programming (WSP), which directly generates neural network weights without gradient-based optimization. However, extending WSP to physical problems remains challenging, particularly in achieving both solution accuracy and physical consistency. In this paper, we propose Physics2Weight (PtW), a framework that extends WSP to solving partial differential equations by directly generating neural network weights and editing architectures based on governing equations and initial and boundary conditions. PtW introduces physics-constrained prompting and physics-guided architecture editing, combined with iterative weight refinement from WSP, to improve solution accuracy and physical consistency without gradient-based optimization. Experiments on four PDEs, namely the heat, advection, Burgers', and Allen–Cahn equations, demonstrate lower solution errors than conventional physics-informed neural networks, while revealing limitations in satisfying the governing equations for nonlinear PDEs.
Read the full abstract
Artificial Intelligence
Agent Architecture Cost-Success Trade-offs: A Controlled Evaluation across Enterprise-Style SOPs
Enterprises increasingly use large language model agents to execute standard operating procedures (SOPs): long, branching, tool-intensive workflows. Yet there is little controlled evidence for choosing among competing agentic architectures. Holding the model, tasks, and tools fixed and varying only the architecture, we compare four designs (a function-calling single agent, a skill-augmented agent, centralized multi-agent orchestration, and a triage swarm; eight arms) across 13 enterprise SOPs, and introduce a skill-definition quality rubric to build matched high and low-quality skill corpora. Three takeaways emerge. (1)~There is no universal best architecture; the right choice is SOP-dependent. On a given SOP, task success ranges from under $10\%$ on the hardest to near $100\%$ on the easiest; no architecture is statistically separable in aggregate, and added structure only plausibly pays off on complex, branching SOPs (a directional, not yet significant, trend). (2)~Added architecture cost must be justified by a task-success gain. Every coordinated arm (multi-agent or skill-augmented) spends $1.4$–$4.7\times$ the tokens and $2.6$–$5.8\times$ the latency of the single agent for gains that are usually small and rarely significant, making the single agent the efficient default; high-quality skills do lift success (to $73.1\%$ from $67.4\%$) while degraded skills fall below it. (3)~Implementation quality outweighs the single- versus multi-agent choice. A plausible multi-agent implementation loses to the single agent, yet a targeted redesign of it becomes the only arm to significantly beat the baseline ($75.6\%$ vs.\ $67.4\%$).
Read the full abstract
Information Theory
Reliability Analysis for Ground-Air Cooperative Wireless-Powered Internet-of-Things
Battery-constrained Internet of Things (IoT) devices face significant challenges in maintaining long-term uplink data collection, especially in network-limited scenarios. This paper investigates a Ground-Air cooperative Wireless-Powered IoT network (GA-WPIoT), where ground power beacons (GPBs) provide radio-frequency energy supply and unmanned aerial vehicles (UAVs) serve as flexible aerial access points for IoT data collection. Unlike conventional UAV-enabled wireless-powered solutions, GA-WPIoT decouples energy transfer and data collection, reducing the hardware and energy burden of UAVs. To capture the device-level performance variations among IoT devices, we develop a stochastic geometry-based analytical framework, where UAVs are modeled by a Matérn hard-core point process, IoT devices by a Poisson point process, and GPBs by a finite annular binomial point process. The conditional success probabilities of the GPB-device link (for energy transfer) and the device-UAV link (for data collection) are characterized, and device-level reliability distributions are obtained through the Meta-Distribution based analysis. Numerical results under suburban and urban environments reveal the impacts of UAV altitude and annular GPB deployment on reliability distributions, providing useful guidelines for GA-WPIoT.
Read the full abstract
Combinatorics
On the average eccentricity of Eulerian and bipartite Eulerian digraphs
Let $D$ be an Eulerian digraph of order $n$ and minimum degree $\delta$. We prove that the average eccentricity of $D$ satisfies $\avec(D) \leq \frac{3n}{\delta+1}$, and that $\avec(D) \leq \frac{2n}{\delta}+2$ if $D$ is bipartite. We show that both bounds are best possible apart from an additive constant and the $\delta+1$ versus $\delta+2$ gap (resp.\ the $\delta$ versus $\delta+1$ gap in the bipartite case), where the average eccentricity of the extremal examples shows that $\avec = \frac{3n}{\delta+2}-2$ in general and $\avec = \frac{2n}{\delta+1}-\frac{5}{2}$ in the bipartite case for $\delta \geq 2$. We further conjecture that these values are the true maxima up to an additive constant (for $\delta \geq 3$ in the bipartite case).
Read the full abstract
Artificial Intelligence
SetPrint: Black-Box LLM Fingerprinting via Setwise Correspondence of Response Changes
Black-box LLM fingerprinting aims to determine whether a target model was derived from a source model when only the output responses to input queries are observable. One way to reveal this relationship is to perturb each query and observe the resulting responses of both models, yielding two corresponding sets of response variations. These sets provide two complementary sources of information: the relations among response changes within each set, and the correspondence of response changes between the two sets. Existing approaches, however, have not fully exploited both. We introduce SetPrint, a training-free method that jointly captures both within-set relations and between-set correspondence. SetPrint expresses each response change relative to the mean change of its own set, and then compares changes from the two models that were induced by the same perturbation. On LeaFBench, SetPrint consistently outperforms existing training-free black-box fingerprinting methods. Further analyses attribute these gains to both within-set relations and between-set correspondence.
Read the full abstract
Dynamical Systems
Forward and pullback attraction under worst-case and ergodic exponents in a predator-prey model with bounded random refuge
We consider a predator-prey system with logistic prey, Holling type II response and a prey refuge fluctuating randomly within a band, driven by a bounded transformation of an Ornstein-Uhlenbeck process. Bounded noise yields a deterministic compact set attracting forward uniformly in the realisation, alongside the random pullback attractor, and we compare the two. The smallest such deterministic set, the minimal uniform attracting set, is the kernel of the control system in which the accessible fraction is any measurable band-valued function, and it meets the coexistence region exactly when the upper worst-case invasion exponent of the predator is positive, whereas the pullback attractor is governed by the ergodic invasion exponent, which decides the predator's fate pathwise. When the predator becomes extinct, the closure of the union of the fibres of the pullback attractor equals the minimal uniform attracting set if and only if the upper worst-case exponent is nonpositive, and otherwise the inclusion is strict and the latter is invariant under no realisation. When the predator persists, the essential closure of the union of the fibres contains, for every frozen environment in which the predator invades, the equilibrium or a periodic orbit of the frozen system, and every state the control system can reach from them. The predator is uniformly persistent when the lower worst-case exponent is positive and almost surely not when it is negative. The ergodic exponent lies below its deterministic counterpart by an amount comparable to the curvature of the response times the variance of the accessible fraction.
Read the full abstract
Logic
Weak $\lambda$-sets and almost-normal Isbell-Mrówka spaces
We determine the least cardinality of a set of reals that is not a weak $\lambda$-set, proving that it is $\max\{\mathfrak b,\mathfrak s\}$, where $\mathfrak b$ and $\mathfrak s$ are the bounding and splitting numbers. We prove that if $\mathfrak s<\mathfrak b$, then every first countable weak $\lambda$-space is a $\lambda$-space. We also show, relative to the consistency of $\mathsf{ZFC}$, that the existence of a weak $\lambda$-set that is not a $\lambda$-set is independent of $\mathsf{ZFC}+\mathfrak b=\mathfrak s$. In particular, we show that in the model obtained by adding $\omega_2$ Cohen reals over a model of $\mathsf{CH}$, every weak $\lambda$-set is a $\lambda$-set. Finally, we construct an almost-normal MAD family under $\mathsf{CH}$ and show that no such family exists under $\mathsf{OCA}_{\mathrm T}+\mathfrak{ap}>\omega_1$. Thus, the existence of an almost-normal MAD family is independent of $\mathsf{ZFC}$.
Read the full abstract
Machine Learning
Phase-Conditioned Reward Shaping for Token-Efficient LLM Reasoning
Extended reasoning can help language models solve complex problems, but long responses often contain both useful computations and redundant checks. Conventional length penalties encourage shorter responses without distinguishing the functions of intermediate steps. We propose Phase-Conditioned Reward Shaping (PCRS), a framework that combines phase identification with training-time reward allocation. We investigate two classification routes: PCRS-H with a hidden-state classifier and PCRS-T with a frozen text classifier. Our main text classifier implementation assigns overlapping reasoning, execution, and revision labels to native chunks, then uses these labels and exact repetition to adjust existing token penalties. We compare the completed configurations on six mathematical and scientific reasoning benchmarks and examine four reward-weight settings with the text classifier. Compared with a baseline using token-level length penalties, the default text classifier configuration reduces mean response length by 3.13%; sampled Pass@1 is approximately 62.07% for both methods, while oracle Pass@16 rises from 86.08% to 87.67%. These results show how phase-informed penalty allocation can reduce output volume and improve multi-attempt coverage without inference-time classification.