Dan Roy

Super-Link Fragility: Why Stronger Entanglement Links Collapse Faster Under Noise

The asymmetric three-qubit W-class state creates an isosceles entanglement network where vertex-base (VB) links are stronger than base-base (BB) links. Counterintuitively, this structural asymmetry is a liability under amplitude damping: the stronger VB link is more vulnerable to energy dissipation, resulting in faster entanglement loss than even the symmetric |W⟩ state. The 'Super-Link Fragility Effect' demonstrates that entanglement robustness under noise depends not on initial concurrence magnitude, but on the joint interplay of network geometry, excitation sector, and noise symmetry.

Analytic Solution for Suppressed Implosion in 1D Compressible Euler Equations

This paper presents an exact analytic solution to the one-dimensional compressible Euler equations with an initial data cutoff, modifying Kidder's formulation. The solution demonstrates the emergence of a non-centered rarefaction from the cutoff, which effectively suppresses the implosion. The theoretical findings are corroborated by numerical simulations, addressing the practical computational challenges associated with previously studied smooth imploding solutions.

Long-Term and Short-Term Transistor Aging in Deep Neural Networks: Impact and Mitigation

Magnetic field free nonreciprocity in tapered atomic cladded nano waveguide

TeleHunt: A Framework and Tool for Efficient Cybercriminal Community Discovery on Telegram

Spectral Analysis Yields Global Well-Posedness for Coupled Navier–Stokes–Biot Systems with Poro-Viscoelastic Media

Binz, Hieber, and Roy establish strong global well-posedness for a 3D fluid–poro-viscoelastic interaction system coupling incompressible viscous flow (Navier–Stokes) with a poro-viscoelastic medium (Biot) via Beavers–Joseph–Saffman interface conditions. The proof strategy is spectral-analytic rather than energy-based, which enables derivation of a unique strong global solution under small initial data. Alongside existence and uniqueness, the authors derive a Serrin-type blow-up criterion, providing a sharp condition under which regularity can fail for large data.

MeshGuard: Scaling MUD-Based IoT Access Control to Multi-Border-Router Thread Networks via SDN

IETF's Manufacturer Usage Description (MUD) standard for IoT network access control breaks down in Thread mesh networks because Thread lacks a full TCP/IP stack and existing MUD-over-Thread extensions only support single-border-router topologies. MeshGuard addresses this gap by extending the Mesh Link Establishment (MLE) protocol to propagate MUD profiles from constrained devices to any border router, and uses Software-Defined Networking to synchronize ACLs across all routers in heterogeneous deployments. Proof-of-concept experiments on real hardware (nRF5340, nRF52833, Raspberry Pi 3) show enhanced security, minimal overhead, and linear scalability over the state of the art.

Barrier Lyapunov + Time-Delay Estimation Fusion Enables Safe, Model-Free Control of Robotic Manipulators

This paper presents an adaptive control framework for Euler-Lagrange robotic systems that simultaneously handles state-dependent uncertainties and enforces time-varying state constraints — a combination largely unaddressed by prior art. The approach fuses Time-Delay Estimation (TDE) for model-free uncertainty compensation with a Barrier Lyapunov Function (BLF) to guarantee hard bounds on both position and velocity. A key contribution is an analytically derived, state-dependent upper bound on TDE approximation error, whose parameters are estimated online via an adaptive law, eliminating the need for prior model knowledge. Lyapunov-based stability proofs and experimental validation on a 5-DOF manipulator confirm constraint adherence under dynamic uncertainties.

A cavity-less architecture for high-power integrated frequency combs

Why LLMs Fail at Causal Discovery and How Interventional Agents Escape

DecomposeRL: Learning to Ask Useful, Informative, and Diverse Questions for Semi-Supervised, Traceable Claim Verification

Efficient Shapley-Based Influence Attribution in Social Networks

Magnetic order, magnons, and crystal fields in van der Waals CeSiI

Comparative Analysis of Military Detection Using Drone Imagery Across Multiple Visual Spectrums

To Select or not to Select, that is the Question: Distilling Robot Skill Prediction into a Small Ensemble

On Matricial Order Operator Spaces

Ising surface defects can get dirty

Pretraining Objective Matters in Extreme Low-Data FGVC: A Backbone-Controlled Study

Reweighting free energy profiles between universal machine learning interatomic potentials for fast consensus building

Binarity at LOw Metallicity (BLOeM): massive star variability revealed using a novel software tool for point-spread function fitting of TESS images

Coarse-grained local available potential energy

Energy Evolution from the Chromosphere to the Heliosphere in the 2021 October 28 Solar Eruption

Human Demonstrations as Exploration Bootstraps for Multi-Agent Equilibrium Solving in Imperfect Information Games

DAGS (Data-Augmented Game Starts) addresses the sparse-reward, long-horizon exploration bottleneck in large-scale imperfect information games by initializing RL data collection at intermediate states sampled from offline human demonstrations rather than from fixed game starts. The core hypothesis is that skilled human play provides good coverage of strategically relevant subgames, enabling regularized policy-gradient methods to reach lower exploitability under fixed compute budgets. A key failure mode is identified — biased starting-state distributions can induce biased equilibria — and mitigated via multi-task observation flags. The authors also release extended OpenSpiel benchmark environments that increase exploration difficulty while keeping exploitability analytically tractable.

Morel-Voevodsky Localization Extended to Complex Analytic Stacks, Enabling a Full 6-Functor Motivic Formalism

Roy Magen proves the analog of the Morel-Voevodsky localization theorem in the setting of complex analytic stacks, a foundational result previously established only in algebraic geometry. This localization theorem serves as a key ingredient for the 6-functor formalism of complex analytic motivic homotopy theory developed in a companion paper (arXiv:2511.09371), and enables construction of an analytification map compatible with the six operations. The techniques developed are deliberately general and transfer to other geometric settings, including algebraic stacks and differentiable stacks.

Encoder-Decoder Framework Enables Online Adaptive Dynamics for Aerial Manipulators Under Nonstationary Conditions

Aerial manipulators suffer from residual dynamics that are cross-coupled, history-dependent, and nonstationary — stemming from quadrotor-manipulator interaction, aerodynamic lag, and payload/configuration changes — which cause both physics-based and offline-learned models to degrade at deployment. This paper proposes a structured encoder-decoder architecture where a nonlinear latent encoder captures cross-variable temporal coupling, and a linear-in-parameter decoder enables closed-form Bayesian online adaptation with covariance inflation for handling regime shifts. The design is explicitly made compatible with real-time MPC, and is validated on a physical aerial manipulation platform showing gains in residual prediction accuracy, adaptation speed, and trajectory tracking.

Abstract Image Editing Remains an Unsolved Challenge: New Benchmark Exposes Fundamental Gaps in Current Models

Current image editing benchmarks overwhelmingly focus on literal, explicit instructions, leaving abstract intents (e.g., "make it feel melancholic") largely unmeasured. This paper introduces AbstractEdit, the first benchmark targeting abstract image editing across diverse real-world scenes, alongside Entity-Rubrics, an evaluation framework that decomposes abstract edits into atomic entity-level assessments correlated with human judgment. Evaluation of 11 leading models reveals a systemic failure mode: models consistently struggle to balance semantic intent fulfillment against content preservation, defaulting to under- or over-editing. Meaningful gains are tied to integrating advanced LLM text encoders and iterative reasoning, with the entity-based paradigm extendable to reward modeling and test-time critique.

Exploring the stellar streams and satellites around the giant low surface brightness galaxy Malin 1

Averaging principle for a slow-fast stochastic nonlinear fractional Schrödinger equation

GW240925 and GW250207: Astrophysical Calibration of Gravitational-wave Detectors

Necessary conditions for causality from linearized stability at ultra-high boosts