Computer Vision
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Sunday 30 August
Visual aliasing, also known as the doppelganger problem, remains a key challenge for structure-from-motion (SfM): visually similar but physically distinct surfaces can produce incorrect image matches and degrade reconstruction quality. Previous work mitigates this issue with geometry-aware foundation-model features, but places a heavy transformer classifier on top of the backbone, making large-scale disambiguation expensive. We introduce XDG, an efficient visual disambiguation model designed for scalable SfM. Our key observation is that a 3D foundation model already performs the cross-view geometric reasoning necessary for visual disambiguation, so doppelganger classification should adapt the backbone representation directly rather than relearn pair reasoning in a separate heavy decoder. XDG fine-tunes Depth Anything 3 with lightweight LoRA adapters and repurposes its camera tokens as compact pair-level classification tokens. A compact MLP head predicts whether a candidate image pair observes the same 3D surface. Extensive experiments show that XDG provides a favorable accuracy-efficiency tradeoff: it remains competitive with the state-of-the-art disambiguation method across pairwise and reconstruction benchmarks and delivers more than a 3x inference speedup. On individual LaMAR scenes containing thousands of images, XDG saves more than 10 hours of visual disambiguation processing. Code is available at https://github.com/xtcpete/xdg.
Combinatorics
Orthogonal adjointness in posets with $0$
Motivated by the concept of polarity introduced by G. Birkhoff for a binary relation on a set, we introduce a concept of orthogonality in a poset with $0$. A pair of operators $f$, $g$ on a poset with $0$ is called orthogonally adjoint if $f(x)$ is orthogonal to $y$ if and only if $x$ is orthogonal to $g(y)$. We characterize the existence and the uniqueness of $g$ for given $f$ and describe basic properties of orthogonal adjointness. We present constructions of orthogonally adjoint pairs in pseudocomplemented posets. If a given operator $f$ is an order-isomorphism of a pseudocomplemented poset satisfying some natural properties then the corresponding adjoint $g$ can be described explicitly. Moreover, if $f$ and $f^{-1}$ are bijective $\perp$-morphisms then they are orthogonally adjoint, too. Finally we show that a given pair of orthogonally adjoint mappings on a poset $\mathbf P$ may not be extendable to the Dedekind-McNeille completion of $\mathbf P$ and we present sufficient conditions for the existence of such an extension. We also provide sufficient conditions for the existence of an extension of orthogonally adjoint mappings to the lattice of ideals. Our results are illustrated by numerous examples.
Read the full abstract
Dynamical Systems
Interconnection of port-Hamiltonian systems is not dynamical
The interconnection of port-Hamiltonian systems is usually understood as structure-preserving due to the geometric fact that the composition of the underlying Dirac structures produces again a Dirac structure. Pointwise, this is doubtless true, but when comparing dynamical systems, their trajectories are usually compared, and trajectories usually obey some regularity assumptions. It is shown that both classical and weak solutions do not necessarily exhibit a one-to-one correspondence between solutions of the interconnected port-Hamiltonian system, and the dynamical system realising the interconnection. In the mostly studied cases of constant Dirac structures or port-Hamiltonian ODE systems, both systems are behaviourally equivalent; for non-constant Dirac structures, a sufficient condition is given.
Read the full abstract
Category Theory
Models for dagger $(\infty,1)$-categories I: Dagger simplicial sets and unitary cores
We develop simplicial-set models for dagger $(\infty,1)$-categories. We first establish a model structure on dagger simplicial categories whose weak equivalences are local weak equivalences that are essentially surjective up to coherently unitary equivalence. We then construct the dagger Joyal model structure on dagger simplicial sets and prove that the dagger rigidification–nerve adjunction is a Quillen equivalence. We next introduce the unitary core and use it to characterize dagger Joyal equivalences intrinsically. Finally, we compare the dagger Joyal model structure with the DCH anti-involutive Joyal model structure.
Read the full abstract
Fluid Dynamics
Lagrangian particles and particle clusters in large-aspect-ratio turbulent Rayleigh-Bénard convection experiment
We perform Lagrangian particle-tracking experiments in turbulent Rayleigh-Bénard convection at aspect ratio Gamma = 10, Rayleigh number Ra = 9.3 x 10^7, and Prandtl number Pr = 4.8. Approximately 50,000 neutrally buoyant particles are tracked for durations of up to 150 Kolmogorov times. The resulting trajectories are sufficiently long to capture the mean-square displacement of single particles from the short-time ballistic regime to the long-time diffusive regime. We further analyse Lagrangian velocity structure functions and particle-pair dispersion, both of which exhibit scaling behavior consistent with theoretical expectations in their respective temporal regimes. In the Richardson-like regime, the pair-dispersion exponent exceeds 3, most clearly in the lateral direction. Multi-particle statistics are investigated by analysing particle clouds with initial radii of approximately 8 and 20 Kolmogorov lengths using principal component analysis. The clouds spread more strongly in the lateral directions than in the vertical direction and deform from initially spherical shapes into oblate ellipsoids within approximately 10 Kolmogorov times, revealing pronounced anisotropic dispersion. The dynamics is identified from the particle-connectivity network through a spectral analysis of the graph Laplacian followed by k-means clustering. Furthermore, we examine the probability density functions of all three acceleration components and assess the statistical convergence of their second and fourth moments. The lateral acceleration components are more intermittent than the vertical component, as indicated by their heavier tails. Acceleration statistics conditioned on the bulk region are slightly less intermittent than those obtained over the full measurement volume, which we attribute to the highly intermittent plume-ejection and plume-aggregation events occurring near the walls.
Read the full abstract
Soft Condensed Matter
Active embracement enables autonomous tweezing in active star polymers
The capacity for autonomous structural reconfiguration is a defining trait of living systems, yet it remains elusive in artificial active matter. Here, we report the discovery of active embracement, a non-equilibrium phenomenon where active star polymers, comprising a central core and self-propelled monomeric arms, transition from open configurations to tightly collapsed, “hugging” states. By combining polymer experiments using connected vibrobots with simulations, we demonstrate that internal self-propulsion fundamentally overrides the steric repulsion that keeps passive polymers dispersed. This active drive enables a suite of behaviors unattainable in equilibrium systems: individual star polymers undergo a globular-like self-collapse, multiple star polymers mutually intertwine and mutually embrace, and can spontaneously embrace and capture surrounding passive particles. Our findings reveal that active embracement is a distinct kinetic phase that allows star polymers to function as autonomous tweezers. By bridging the gap between macroscopic robotic collectives and microscopic polymer physics, this work provides a versatile blueprint for the design of smart materials capable of targeted cargo capture and self-directed assembly in complex environments.
Read the full abstract
Astrophysics of Galaxies
Primordial Black Hole Seeds for Little Red Dots in $f(R)$ Gravity
The little red dots (LRDs) discovered by the James Webb Space Telescope host black holes of up to $10^{9}\,M_{\odot}$ within the first billion years, posing a challenge to conventional seeding scenarios. We investigate whether such seeds can form in dense clusters of primordial black holes (PBHs), in which the central object grows by gravitational capture of the surrounding PBHs while accreting gas, within Hu-Sawicki $f(R)$ gravity, where the scalar field acts on the matter surrounding the PBHs. For the adopted cluster configuration, our results show that accretion and mergers reinforce one another: for time-averaged Eddington ratios $\lambda_{\rm Edd}\gtrsim0.3$, the seed consumes its cluster by $z\simeq5.5$-$7.3$ and subsequently grows into the LRD mass range. Calibrated on the observed black-hole mass function, the model reproduces the abundance of LRDs near the peak of their population without further adjustment, while it underpredicts the abundance at $z\simeq7.5$ and overpredicts it at $z\simeq2$. Our findings indicate that the earliest seeds may accrete about twice as efficiently as later ones, and that the finite cluster reservoir can imprint a lower edge on the mass function near the cluster mass. Within the adopted screening and halo-growth prescriptions, the scalar field enhances the abundance of active seeds by up to $\sim 13\%$ at lower redshifts and by $\sim 7\%$ at $z \simeq 7.5$ for the strongest model considered, modestly moving the high-redshift prediction towards the data at fixed normalization. Also, the remnant spins can further discriminate between feeding modes.
Read the full abstract
Probability
A Priori Estimates for Singular Fractional Stochastic Burgers Equations
We study the periodic fractional stochastic Burgers equation $(\partial_t+\Lambda^\gamma)u=\partial_x(u^2)+|\partial_x|^{1-\alpha}\xi$, where $1<\gamma\leq 2$ and $\xi$ is space-time white noise. Under the condition $\alpha>\max{(7-4\gamma)/2,(15-8\gamma)/6}$, we establish pathwise $L^1$, energy, and Besov estimates for a transformed remainder. The argument combines a response decomposition with a conservative Zvonkin transformation. For $1<\gamma<2$, this produces a compensated nonlocal diffusion operator whose coercivity is coupled to a cubic-increment estimate adapted from the modified Karman–Howarth–Monin argument.
Read the full abstract
Signal Processing
A New Paradigm of 6G Networks: Proactive Channel Cognition and Reconfiguration
The sixth-generation (6G) wireless networks are expected to enable the deep integration of communication, sensing, computing, control, and intelligence in highly dynamic environments. This evolution drives a fundamental transition from conventional passive channel adaptation to proactive channel cognition and reconfiguration, wherein wireless channels are no longer regarded as uncontrollable propagation media but as network resources that can be learned, predicted, and actively reconfigured. This paper presents a comprehensive overview of this emerging paradigm. We first review channel cognition through the channel knowledge map (CKM) as a systematic framework for learning and exploiting channel characteristics across space, time, and frequency domain. The definitions, construction methods, and applications in wireless networks of CKMs are comprehensively reviewed. Building upon channel cognition, we then review channel reconfiguration technologies from two complementary perspectives: transceiver-side reconfiguration enabled by movable antennas (MAs) and environment-side reconfiguration enabled by intelligent reflecting surfaces (IRSs). For both MA- and IRS-enabled wireless systems, we review their architectures, performance advantages, and key design challenges. Finally, we discuss several promising research directions to inspire further innovations in this burgeoning field.
Read the full abstract
Mesoscale and Nanoscale Physics
Harnessing the skyrmion Hall effect for low-power skyrmion transport in a tubular synthetic antiferromagnet
We show that optimal control can turn the skyrmion Hall effect from a source of typically unwanted transverse motion into a mechanism for reducing Ohmic losses in skyrmion transport. We consider a pair of antiferromagnetically coupled skyrmions confined to a tubular geometry, where their relative transverse displacement becomes a periodic internal coordinate. The skyrmion Hall effect drives this coordinate, while finite interlayer coupling makes the resulting relation between applied current and longitudinal velocity nonlinear, allowing different current protocols to produce the same prescribed average velocity. We determine which of these protocols minimizes Joule heating. Depending on the interlayer coupling and spin-transfer-torque parameters, the optimal current is either constant, with the skyrmion pair maintaining a fixed relative position, or time dependent, with the pair undergoing periodic relative motion around the tube. In either case, the additional internal degree of freedom created by the skyrmion Hall effect and interlayer coupling enables skyrmion transport at a lower power than in the uncoupled limit.
Read the full abstract
General Relativity and Quantum Cosmology
Peeling property for the Einstein scalar field equations with the nonzero cosmological constant
Inspired by interaction of gravitational waves, dark matters and dark energy, we study the Einstein scalar field equations with the nonzero cosmological constant for the Bondi-Sachs metrics. We provide the asymptotic expansions under the outgoing radiation condition. Unlike the case of the zero cosmological constant, the asymptotic expansions depend on four additional $\Lambda$-independent functions $B(u, \theta, \phi)$, $\tilde{X}(u, \theta, \phi)$, $\tilde{Y}(u, \theta, \phi)$ and $\tilde{I}(u)$. We show that, under suitable coordinate transformation, we can make $B(u, \theta, \phi)=0$. We prove the peeling property for the Einstein scalar field equations if $\tilde{I}=0$, and derive a loss formula of the Bondi energy-momentum for the nonzero cosmological constant. We remark that, for certain real data from gravitational waves, the loss formula is likely to indicate the loss property of the Bondi energy-momentum.
Read the full abstract
Functional Analysis
Singular inner eigenfunctions of composition operators
This paper characterizes all the singular inner eigenfunctions of the composition operators $C_\phi$ that arise from discrete measures, when $\phi$ is an automorphism of unit disk. By establishing a connection between Beurling and model invariant subspaces, we classify all the inner functions so that the corresponding Beurling subspace is invariant under the composition operators induced by non-elliptic automorphisms. This classification involves solving the eigenfunction equation for the composition operator. Further, we present some applications of the above-mentioned connection.
Read the full abstract
Analysis of PDEs
A Uniqueness theorem for the Vortex-Wave system with non-constant vorticity near the point vortex
We study the uniqueness of the vortex wave system, i.e.\ the 2D incompressible Euler equations with vorticity consisting of a point vortex and a bounded part. We show that if the initial vorticity and point vortex location $(w_0,X_0)$ satisfy a constraint of the type $|w_0(x)-w_0(X_0)|=O(|x-X_0|^\alpha)$ for $\alpha> 2$ and the vorticity is in $L^\infty$, then solutions are unique, globally in time, even though the velocity is not log-Lipschitz.
Read the full abstract
Robotics
VeloBins: Learning Velocity and Its Uncertainty via Bins and Error-Conditioned Gaussian Labels for Aerial Inertial Odometry
Inertial odometry (IO) is critical for aerial robots, where aggressive maneuvers and poor lighting degrade visual sensors. Recent learning-based IO methods improve traditional integration-based approaches by learning motion priors from IMU and platform-specific sensors, then fusing the predictions within an extended Kalman filter. However, learning velocity through regression is difficult, while jointly estimating uncertainty with a separate decoder and negative log-likelihood (NLL) loss further complicates training and can lead to over-confident estimates. We introduce VeloBins, which reformulates velocity regression as classification over discretized velocity bins. We decode both the velocity from the bin distribution's expectation and the uncertainty from its variance, removing the need for a separate uncertainty decoder. We further supervise the uncertainty explicitly using an error-conditioned Gaussian label centered at the ground-truth velocity, with a standard deviation set to the velocity error. We evaluate VeloBins on four aerial datasets, ranging from free-form aggressive flights and a 27 g nano-quadrotor to drone racing at over 21~m/s. VeloBins achieves the lowest average errors on all four datasets, reducing velocity, relative trajectory, and absolute trajectory errors by 3-27%, 8-40%, and 6-53%, respectively, compared with the strongest baseline. Notably, the proposed supervision achieves the lowest NLL and best filter consistency despite never optimizing an NLL loss. The code will be available upon acceptance.
Read the full abstract
Algebraic Topology
Persistence Pairing Graphs: Tracking Changes of Selected Homology Bases
Persistent homology summarizes a filtration by recording when homological features appear and disappear, but the resulting barcode does not determine a preferred homology basis or preferred cycle representatives. We study the additional information carried by a basis selected through matrix reduction and introduce the persistence pairing graph, a reduction-dependent directed graph that records how selected homology classes are re-expressed when their associated persistence intervals end. After fixing a filtered finite CW complex, a coefficient field, a filtration-compatible cell order, and a reduced factorization of the boundary matrix, the reduction determines selected birth cycles. When an associated persistence interval ends, the image of its selected birth class has unique coordinates in the basis formed by selected classes whose intervals survive beyond the same filtration value, and the nonzero coordinates define the graph edges. These coordinates can also be recovered through the Kronecker pairing with a dual cohomology basis. We prove that the active selected birth cycles form a homology basis at every cellwise filtration stage, characterize the kernel and image of each homological transition, and show that the graph is directed and acyclic, with every edge pointing toward an interval containing the source interval. In dimension zero, under the standard reduction convention, the graph agrees with the elder rule merge tree. We also prove that its edges can change while the persistence pairs and barcode remain fixed. Finally, we define a bounded fused graph discrepancy and separate the contribution controlled by ordinary persistence stability from the reduction-dependent graph structure. The graph is therefore a descriptor relative to a fixed reduction convention, rather than an invariant of the persistence module alone.
Read the full abstract
General Relativity and Quantum Cosmology
Testing $f(Q)$ Gravity with Logarithmic Equation of State Using Latest Cosmological Data
In this paper, we investigate the late-time accelerated expansion of the Universe in power law $f(Q)$ gravity, where a logarithmic dark energy parametrization is considered: $\omega_{de}(z)=\omega_0+\omega_1\ln(1+z)$. This description gives a smooth deviation from a constant equation of state within the complete range of redshifts. We obtain an analytical expression for the Hubble parameter and we use a Markov Chain Monte Carlo to compare the model with the most recent observational data obtained from Pantheon$^+$ Type Ia supernovae, Baryon Acoustic Oscillation measurements from the second data release (DR2) of the Dark Energy Spectroscopic Instrument, and Cosmic Chronometers. We then perform a statistical comparison between our model and the standard $\Lambda$CDM model using the Akaike Information Criterion and the Bayesian Information Criterion. From our results, we conclude that the use of logarithmic parametrization in the $f(Q)$ gravity model $f(Q)=Q+6\gamma H_0^2\left(\frac{Q}{Q_0}\right)^n$ is a valid and more flexible alternative to standard dark energy models, as it provides a richer phenomenology at low redshifts.
Read the full abstract
Physics and Society
Open, Reproducible Per-Bidding-Zone Carbon Intensity for Nordic Power Systems
Carbon-aware scheduling of electrical loads depends on accurate grid carbon-intensity (CI) signals, yet existing tools commonly apply a uniform placeholder emission factor that misrepresents regional variation. This work presents an open, reproducible method for computing per-bidding-zone CI across the Nordic power system. Generation-by-type data from ENTSO-E is combined with IPCC AR5 life-cycle emission factors, duration-weighted per bidding zone, with an explicit materiality threshold for missing data. Applied to nine bidding zones (NO1-NO5, SE1-SE4) for 2025, the method yields CI values ranging from 14.5 to 39.6 gCO2eq/kWh. The results reveal that the uniform 18.0 gCO2eq/kWh placeholder is wrong in both directions: it understates CI for most zones, by a factor of approximately 2.2 for NO4, yet overstates CI for the nuclear-dominated zone SE3, whose production-based CI is the lowest of the nine. Drift is characterised against a pre-registered threshold of 15% year-on-year change, including a borderline case in SE4 and event-driven variation in NO4. Forecasting experiments do not reject the null hypothesis that simple baselines suffice, in either the Norwegian or Swedish zones. The full method, data, and results are released under open licences and archived at DOI 10.5281/zenodo.21042581, with the per-zone values contributed upstream to the codecarbon library. Per-bidding-zone resolution is necessary for accurate carbon accounting in Nordic grids, where country-level averages obscure substantial intra-country variation.
Read the full abstract
Software Engineering
Agent-Driven Verification of Memory Safety for liblzma Decoder Components with VST
We report on the verification of memory safety for decoder components of liblzma, the compression library underlying xz-utils: the LZMA2 state machine, the LZMA1 decoder it controls, the outer decoding path, and the shared sliding-window dictionary. Built with the Verified Software Toolchain (VST), machine-checked body theorems establish memory safety and partial functional correctness. Across 27 completed body proofs, the largest covers lzma decode, whose 338 source lines expand to 1,934 lines of C after preprocessing; its proof comprises 183,268 lines of proof script over 775,768 lines of mechanically extracted goal statements. The verification exposed undefined behavior in raw LZMA1 zero-input handling, where range-decoder macros add zero to a null pointer and subtract two null pointers. Unlike similar work that synthesizes verified code, we verify pre-existing, production-scale C. AI agents complete proof goals and propose refinements; humans write and review models and specifications, and approve semantic changes; the Rocq kernel checks the proof terms. With agents constructing the proof scripts, the main engineering problems lay in translating and modeling production C, building a robust harness for driving Rocq, and providing feedback for proving agents. VST's assertion logic expressed every contract required by the development. We describe the pipeline, coordination mechanisms, and proof-engineering techniques that resolved these frictions.
Read the full abstract
Machine Learning
Higher-Dimensional Rotary Position Embedding
Transformers rely on position embedding mechanisms in long context modeling in most cases. Rotary Position Embedding (RoPE) embeds positional information with independent 2D rotations, forming relative position terms in self-attention. However, its pairwise, block-based, and decoupled structure limits deep mixing and robustness across channels. We propose HD-RoPE, which extends RoPE from independent 2D rotations to higher-dimensional rotations and introduces a Paley-I orthogonal basis to obtain balanced, isotropic, and dense phase mixing within each rotation subspace. This significantly enhances channel coupling and rotational degrees of freedom while maintaining orthogonal stability and the relative position closure property. Furthermore, HD-RoPE is easily optimized for engineering efficiency without introducing additional trainable parameters. We have conducted extensive evaluation results demonstrating that HD-RoPE achieves significant performance improvements over standard RoPE across various popular benchmarks and in both long and short contexts.
Read the full abstract
Machine Learning (Statistics)
Neural ODE enhanced linear mixed effect models for estimating complex association patterns of time-varying covariates with the marker trajectory
Longitudinal cohort studies produce repeated data that enable the assessment of time-varying association patterns between exposures and health outcomes. Classical linear mixed-effects models (LMMs) can accommodate a large variety of association patterns while accounting for the irregularly spaced, partially observed measurement. But they require the analyst to pre-specify the functional form linking the exposure history to the outcome. We propose the Neural ODE-LMM, which embeds a Neural Ordinary Differential Equation (Neural ODE) within the linear mixed-effects framework: a learned vector field encodes covariate trajectories into a continuous-time latent state that drives both the fixed- and random-effect design, while preserving the standard LMM observation model. This retains classical likelihood-based inference while learning complex, potentially cumulative, covariate effects flexibly. All parameters are estimated by maximising a penalised marginal likelihood. To quantify covariate effects, we introduce contrasts of counterfactual predictions that compare the expected outcome under alternative covariate trajectories with variance estimated via the delta method. In simulations, the model recovers both instantaneous and cumulative-burden effects without prior specification of the functional form. Applied to the Trois-Cités (3C) cohort, a population-based study of 7{,}324 participants, the method reveals trajectory-dependent associations of BMI and fasting glucose with cognitive decline.
Read the full abstract
Differential Geometry
A Cartan geometry from a correspondence space in Riemannian geometry
Cartan geometries are curved analogues of homogeneous spaces, and the correspondence space construction produces, from a Cartan geometry of one type, another of a different type over a larger base manifold. The unit tangent bundle of a Riemannian manifold is, in this language, a Cartan geometry of type $(\operatorname{Euc}(n),O(n-1))$ obtained in this way; we study Cartan geometries of this type in general, whether or not they arise as correspondence spaces. We show that every such geometry induces on its base manifold an almost contact metric structure, together with an orthogonal splitting of the tangent bundle and a compatible linear connection; conversely, these data determine the geometry, so that the correspondence is a bijection. When $n\ge3$, we single out a distinguished Cartan connection, which we call normal, by a condition on its torsion formulated in terms of the Spencer differential. This normalization is dictated by the correspondence space construction: the Cartan connection induced on the unit tangent bundle of a Riemannian manifold is normal.
Read the full abstract
Materials Science
High-Field Terahertz Spin Resonance in Cr$_2$O$_3$ above the Spin-Flop Transition
We report single-shot terahertz time-domain spectroscopy of Cr$_2$O$_3$ in pulsed magnetic fields up to 30 T. Well above the spin-flop field, in the 20-30 T range, the resonance frequency exhibits a nearly linear field dependence with a slope of ~22 GHz/T, smaller than the 28 GHz/T reported from low-field measurements. This reduction is insensitive to temperature and to a 15{\deg} field tilt, suggesting an intrinsic high-field property.
Read the full abstract
Computation and Language
DVBench: Benchmarking MLLMs for Understanding Dynamic Charts and Narratives in Data Videos
While MLLMs have made significant strides in chart comprehension and video understanding, current evaluations largely isolate these capabilities, leaving a critical gap in understanding temporally evolving structured visual information. To address this gap, we introduce DVBench, a benchmark for evaluating MLLMs on data videos, a storytelling medium that integrates dynamic charts with structured narratives. We decompose data video understanding into five dimensions. DVBench comprises 300 real-world data videos and 1,000 human-verified QA pairs curated through a rigorous semi-automated pipeline. Extensive evaluations of nine MLLMs show that Gemini-3.1-Pro achieves the best overall performance, while Kimi-k2.5 is the strongest open-source model. We further identify two notable phenomena: open-source model performance does not scale strictly with parameter size, and narrative proficiency does not guarantee visual capability. Fine-grained analyses and ablation studies further reveal dimension-specific weaknesses and the effects of frame configurations and subtitle inputs, informing future MLLM development. DVBench is publicly available at https://bomiaowang.github.io/DVBench/.
Read the full abstract
Statistical Mechanics
Phase Diagram and Critical Behaviour of the Two-Dimensional Potts Model with Long-Range Quenched Disorder
We study the phase diagram and critical properties of the $q = 3$ and $q = 8$ Potts models with spatially correlated disorder governed by a power-law decay with exponent $a$. Building on the phase diagram proposed by Chippari et al. [3], where a transition from finite-disorder to infinite-disorder fixed points was identified as a function of a, we refine this picture by using wrapping probabilities of Fortuin-Kasteleyn clusters. Furthermore, using magnetic observables, we clarify the physical origin of the double-peak structure in the magnetic susceptibility and establish the validity of the hyper-scaling relation across all investigated regimes.
Read the full abstract
Computational Physics
Characteristic Mode Analysis of Composite Nanostructures using a Coupled System of Volume Integral and Hydrodynamic Equations
Full-structure and sub-structure characteristic mode analysis (CMA) formulations are developed for composite metallic–dielectric nanostructures based on a coupled system of volume integral equations (VIE) and the hydrodynamic equation (HDE). In the full-structure CMA, the generalized eigenvalue equation (GEE) is constructed from the matrix of the complete coupled system, and the resulting characteristic currents describe the response of the entire composite nanostructure. In the sub-structure CMA, the GEE is constructed from a reduced system, derived from the coupled system by eliminating the dielectric-region unknowns, so that it is expressed only in terms of the metallic-region currents. This isolates the resonances of the metallic region while still incorporating the effect of the dielectric region through the reduced system. Because the reduced system has a smaller dimension, the sub-structure CMA is computationally more efficient than the full-structure CMA and, when the dielectric does not resonate in the frequency range of interest, identifies the same resonances. Both formulations are validated against extinction cross-section (ECS) results and are used to characterize how a dielectric environment reshapes the metallic resonances, including substrate-induced red-shifts and, for high-contrast substrates in direct contact, hybridized modal responses.
Read the full abstract
Logic
Intuitionistic and Constructive Modal Logics for Classical Modal Logicians
This paper gives a survey of propositional modal logic in an intuitionistic setting. Mainly intended for readers already familiar with classical modal logic, we discuss why there are two prominent families of such logics, intuitionistic modal logics (IMLs) and constructive modal logics (CMLs), what motivates their study, and how they compare with each other. We then take a look at the logics lying between the CML and IML traditions, and survey various extensions and $\Diamond$/$\Box$-free fragments. We further record several observations which seem to be implicit or absent in the literature; we point out a small defect in the original semantics for $\textbf{CK}$ and prove Lyndon interpolation theorems for $\textbf{CK}$ and $\textbf{WK}$.
Read the full abstract
Quantum Physics
Kempe-Invariant Ordering of Three-Qubit Pairwise Entanglement under Local Depolarization
Pure three-qubit states with the same pairwise concurrences and three-tangle can still have different Kempe invariants. We show that the Kempe invariant orders the entanglement dynamics of all three pairs under identical, independent local depolarizing channels, with smaller values giving greater robustness. The same ordering also holds for the sum of squared pairwise entanglements of formation and the sudden-death thresholds of pairwise entanglement.
Read the full abstract
Computation and Language
The Depth Flow of Token Representations Is Nonlinear and Does Not Descend Its Own Density
A token's representation is carried through the network layer by layer. The whole vocabulary carried together forms a flow. We fit this flow's equation of motion as a discrete Langevin model over corpus-mean trajectories of Pythia-160M and Pythia-410M, and score the predicted steps on held-out tokens. Linear maps are often used as cheap surrogates for a layer. The flow they summarize is not linear: a quadratic drift beats the linear map at every transition of both models, and the Kramers-Moyal estimator agrees wherever its neighborhoods stay local. At full width, the step of a mean trajectory is nearly affine, 3% non-affine, while the step of an individual state is not, 17%. We then characterize the flow further. First, we show that it does not descend its own log-density, whatever the diffusion matrix. The drift instead descends a potential that is not the density. Second, the rotational component is not negligible, 4 to 45% of the explainable drift, and it is the antisymmetric part of the linear map. Finally, a token keeps its angular rank and its local-density rank across all thirteen layers, on two corpora and both models.
Read the full abstract
Computer Vision
A Calibration Audit of Confidence in Feed-Forward 3D Reconstruction Models
Feed-forward 3D reconstruction models output a per-pixel confidence that is used by downstream systems as an uncertainty signal. The confidence is trained to serve as a weight in the training loss of models. Whether the confidence can be used as an uncertainty magnitude has not been measured. We audit seven backbones on 13 datasets and score the confidence on four properties, i.e., ranking of error, ratio of error to uncertainty on average, slope of this ratio across the confidence range, and coverage of the implied error distribution. Although the confidence ranks error quite well, the uncertainty decoded from the confidence is too small compared to the actual error. The uncertainty has the right size only under the exact training conditions. The median case is off by at least 2.4x across all seven models, while the uncertainty is further off the more confident the model is. Our work shows that the overconfidence appears on unseen scenes even when the model reaches its loss's optimum. As a post-hoc repair we fit a power law on the confidence with two constants per backbone–dataset pair. The repair brings all four audited properties to target at the dataset level, while leaving ranking untouched. Fitted with the target dataset held out, the constants bring the median case from 2.4x off to 1.35x. The repair does not hold below the dataset level, where two-thirds of held-out scenes are still more than five points off in coverage. We attribute what the repair cannot reach to the model, which carries neither the scale of the error nor the shape of its distribution across predictions. We release the audit protocol, its results, and the fitted constants per backbone-dataset pair.
Read the full abstract
Analysis of PDEs
Existence of Periodic Solutions to Steady Viscous Burgers Equation with a General Force
In this paper, we will construct periodic solutions to the viscous steady Burgers equation with an external force $f(x)$, based on the following formal expansion $u^\varepsilon(x)=u_0(x)+\varepsilon u_1(x)+\cdots+\varepsilon^n u_n(x)+\cdots$, where $\varepsilon \ge 0$ represents the viscosity and $u_0(x)$ is a solution to non-viscous steady Burgers equation with the external force $f(x)$. We will focus on the solutions which are uniformly bounded with respect to the viscosity. In our previous work, starting from $u_0=-(2+\cos x)$, the authors constructed the periodic solutions to the viscous steady Burgers equation with the external force $f=u_0(x)u_{0x}=-2\sin x-\sin x\cos x$. In this paper, we will extend the main result obtained in our previous work and construct the solutions starting from general $u_0$ and $f$ satisfying the non-viscous steady Burgers equation $u_0(x)u_{0x}=f$. It will be shown that there exists a $\varepsilon_0>0$, which depends on $n$, such that for any $0<\varepsilon<\varepsilon_0$, the viscous steady Burgers equation with the external force $f$ has a periodic solution $u^\varepsilon(x) \in C^2([0,2\pi])$, satisfying $|u^\varepsilon(x)-u_0(x)-\varepsilon u_1(x)-\cdots-\varepsilon^n u_n(x)| \leq C\varepsilon^{n+1}$, where $C>0$ is a constant which depends on $n$, but is independent of $\varepsilon$. Compared with Jauslin-Kreiss-Moser's result, we present a new approach to construct periodic solutions to the viscous steady Burgers with a general external force. The constructed solutions will tend to the ones of the non-viscous Burgers equation with a sharper convergence rate (up to higher order) when the viscosity vanishes.