Papers

Primordial magnetic fields (PMFs) are anticipated to be a cosmological candidate for generating a stochastic gravitational wave background (SGWB) through their anisotropic stress. In particular, if parity violation is present in the production of PMFs, it can induce a helical component in the PMFs, leading to circular polarization in the isotropic SGWB. In this study, we phenomenologically consider parity-violating inflationary magnetogenesis and compute the intensity and circular polarization of the SGWB for several assumed power spectra of helical PMFs. Using the planned sensitivity of the LISA-TAIJI network, we calculate the signal-to-noise ratio (SNR) and perform a Fisher forecast to assess the feasibility of constraining the parameters of helical PMFs. We conclude that it is possible to estimate the helical-to-non-helical power ratio~$r_H$ with $\mathrm{SNR}^{V}>2$ if the PMF strength is $\mathcal{B} \gtrsim 10~\mathrm{nG}$ ($50~\mathrm{nG}$) in the case of the delta-function-type (scale-invariant-type) PMF spectrum. Our results offer a useful criterion that indicates the future observational limit on helical PMFs at the small scale $k_\mathrm{LISA} \approx 10^{12}~\mathrm{Mpc}^{-1}$.

We present ASENA, an embodied agent system that enables general-purpose coding agents to program humanoid navigation, manipulation, and interaction. Through a unified robot harness, agents read sensor data, write and execute programs, inspect recorded outcomes, repair failures, and reuse notes and executable skills while keeping their model weights fixed. We also introduce ASENA-VLN, a 4B monocular navigation policy that serves as an optional tool within this programmable system. ASENA-VLN predicts body-frame trajectories for both extended routes and short-horizon behaviors. Its shared vision-language decoder is trained on route instructions, visual question answering, and a newly curated dataset of geometry-derived atomic navigation tasks. As a standalone policy, ASENA-VLN achieves state-of-the-art success rates of 68.7% on R2R and 70.2% on RxR. When integrated with a coding agent, learned navigation improves ASENA's success rate by 11 percentage points on both agentic benchmarks and reduces execution time. Persistent workspace evolution uses simulator feedback to improve success over ten passes through recurring 100-task subsets. Success increases from 72% to 98% on R2R and from 65% to 89% on RxR. On embodied question answering, ASENA achieves state-of-the-art accuracy with fewer interaction steps. On a Unitree G1, we demonstrate search, visual inspection, spatial reasoning, synthesized gestures, and mobile manipulation combining object pickup, carrying, and handoff without a pre-built map. These demonstrations illustrate how online programming extends robot behavior beyond route following and predefined skills.

In order to estimate the longitudinal trend, we often use the generalized multivariate analysis of variance (GMANOVA) model (Potthoff & Roy, 1964), when the longitudinal data is balanced data. Usually, we use this GMANOVA model with some polynomial at the time of measurement. However, when the longitudinal trend has flexible curve, we cannot derive good fitting estimated curve when we use some polynomial curves. Then, Nagai (2011) proposed the nonparametric GMANOVA model which uses on several known basis functions instead of using the polynomial curves. If we use several basis functions, then overfitting problem is occurred. Nagai (2011) also proposed the estimation method for avoiding overfitting and unstable problems, and reducing computational iterative algorithm by extending the generalized ridge regression model (Yanagihara, Nagai & Satoh, 2009). In the present paper, we extend one of the optimization methods in Nagai, Yanagihara and Satoh (2012) into the estimation method in Nagai (2011). Through numerical studies, we show some properties of each optimization method.

Experiments often detect only selected transitions, leaving the underlying state dynamics and dissipation hidden. We show that waiting-time distributions between directed visible events can reconstruct the boundary dynamics without observing the reverse of every detected transition. Under a simple source-target closure condition, they determine visible rates and the hidden excursions connecting observed states, yielding a rigorous lower bound on total entropy production that improves the standard waiting-time estimate when reverse events are available. Remarkably, observing only a spanning directed cycle is sufficient to identify the full Markov generator, even when all reverse cycle edges and other microscopic transitions remain unseen.

Despite the availability of excellent vaccines and antiviral medicines, hepatitis B virus (HBV) remains a major public health problem. In this work, we develop a mathematical vaccination-treatment model for HBV transmission to investigate threshold conditions for illness persistence as well as the level of intervention. The analysis gives the characterisation of both disease-free and endemic equilibria and determines the basic reproduction number, $\mathcal{R}_0$, by the next-generation matrix approach. The model indicates that $\mathcal{R}_0=1$ is a critical threshold dividing disease-free from endemic states and characterises a forward transcritical bifurcation regarding the transmission parameter. Numerical equilibrium continuation and stability assessments confirm the analytical results. A two-parameter study of transmission and vaccination rates identifies threshold geometries and assesses impacts of important parameters using sensitivity analysis. The model is calibrated with India-specific data, such as HBsAg prevalence, HBV incidence and mortality rates and fits epidemiological targets closely through limited nonlinear calibration. Additionally, multi-start optimisation and practical identifiability profiling are used to investigate parameter uncertainty. Robustness analyses reveal that the calibrated $\mathcal{R}_0^*=1.1744$ is associated with a super-threshold regime; however, parameter flexibility permits both sub-threshold and super-threshold regimes. This combined analytical and data-driven framework connects epidemic thresholds, bifurcation structures, intervention efficacy, and parameter uncertainty. It provides a quantitative framework for analysing HBV persistence and evaluating control options, especially in the absence of epidemiological evidence.

Accurate velocity measurement is a fundamental requirement for autonomous surface and underwater vehicles. Commonly, velocity is provided by a Doppler velocity log (DVL) sensor, yet it becomes unavailable due to operational altitude constraints. In such situations, electromagnetic logs (EMLogs) provide a critically robust alternative for continuous velocity estimation. However, raw EMLog measurements are inherently corrupted by systematic errors, which need to be calibrated prior mission begins. Currently, a benchmarking comparative evaluation of how different calibration models perform under rapidly changing dynamic sea conditions is missing in the literature. To bridge this gap, this paper presents a comparative model-based calibration methodology that evaluates four distinct calibration models using two different estimation pipelines. The proposed framework is rigorously validated on a unique 221 minutes of continuous real-world telemetry collected from the MARVEL surface vehicle during dynamic sea trials. The dataset contains two different EMLogs and DVL recordings. Experimental results demonstrate that the bias and scale error model implemented with the Kalman filter improves the speed estimation by 71%. We also demonstrate that dynamical manoeuvres further improve the accuracy compared to standard straight-line paths, ultimately delivering a validated, real-time online calibration EMLog approach for autonomous surface vehicles.

In knee arthroscopy surgery, accurate registration between preoperative and intraoperative anatomy is a critical step for patient-specific navigation. Achieving an accurate registration requires a reliable 3D reconstruction of the joint during surgery. Preoperative 3D models can be obtained from patient imaging through segmentation and reconstruction, but generating an intraoperative 3D representation remains particularly challenging. Arthroscopic imaging suffers from a limited field of view, low surface texture, and strong specular reflections, which make conventional feature-based 3D reconstruction methods unreliable. In this work, we investigate the application of MIS-NeRF (Minimally-Invasive Surgery Neural Radiance Fields) for reconstructing intraoperative knee 3D models from monocular arthroscopic images. The approach is evaluated on six simulated arthroscopic acquisitions representing six patient-specific knee 3D models. Both qualitative and quantitative results are presented to assess the reconstruction quality. The reconstructed knee 3D models were evaluated through their rendered images, achieving PSNR (Peak Signal-to-Noise Ratio) of 31.88 $\pm$ 2.82, SSIM (Structural Similarity Index) of 0.98 $\pm$ 0.004 and LPIPS (Learned Perceptual Image Patch Similarity) of 0.017 $\pm$ 0.006. These preliminary results suggest the feasibility of NeRF-based reconstruction in the challenging context of arthroscopy and may represent a promising step toward accurate in-silico preoperative-to-intraoperative 3D registration for computer-assisted orthopedic surgery. Further, validation on real arthroscopic data will be necessary to assess clinical applicability.

We present examples of classical time crystals (in a parabolic potential for the position) that admit simple exact analytic solutions. The kinetic energy in these crystals is piecewise parabolic as a function of speed and is minimised at nonzero values of the speed. The solutions are periodic, while the speed acquires definite periodic discontinuities. The solutions are stable along the branches where the action is lower. We also examine the classical time crystal with quartic kinetic energy. We show how the quartic kinetic energy can be emulated by a piecewise parabolic one, leading to accurate and transparent solutions.

We consider harmonic approximation on the torus $\mathbb{T}^d$, the sphere $\mathcal{S}^d$, and the rotation group $\mathrm{SO}(3)$. While several global approaches are available for computing a harmonic expansion from scattered data, their stability and, especially, their runtime depend strongly on the geometry of the nodes, in particular on a sufficiently small fill distance. Even a single large hole in the data may cause instability and long runtimes. We propose harmonic approximation via moving least squares (HAMLS), a global harmonic approximation scheme that avoids the ill-conditioned global solve and works well in both the overdetermined and the underdetermined settings. The main idea is an intermediate transition from the scattered data to a quadrature grid, which is realized via moving least squares (MLS). This replaces one large global problem by many small local ones that can be regularized individually. The global harmonic approximation is then obtained via quadrature using a fast Fourier transform. This makes the global stage fast, stable and non-iterative. For sufficiently smooth functions sampled at well-distributed nodes with small fill distance $h$, we bound the resulting $\mathrm{L}^2$-error by the error of the harmonic approximation obtained from exact values on the quadrature grid, plus a term of order $h^{K+1}$, where $K$ is the polynomial degree employed in the MLS step. Numerical experiments on $\mathcal{S}^2$ with ill-distributed nodes show that HAMLS achieves errors comparable to those of global least-squares approximation, while being significantly faster, especially for larger bandwidths.

Large Audio Language Models (LALMs) rely on effective audio encoders for multi-task performance. We introduce UniAE-MoE, a unified audio encoder designed to model cross-domain audio representations and achieve outstanding downstream understanding performance via a Mixture-of-Experts (MoE) architecture. Specifically, we explore mainstream audio encoders and integrate those from Qwen2-Audio and Audio-Flamingo 3, which demonstrate superior downstream capabilities. To facilitate effective model fusion, we improve our encoder using SwiGLU with shared experts to decouple encoder networks, and we further introduce a two-stage instruction-tuning strategy to better adapt the model to diverse downstream tasks. Moreover, we propose the task-specific data scaling (TSDS) technique to enhance UniAE-MoE's understanding capabilities. On the XARES-LLM benchmark, UniAE-MoE attains a score of 0.802, achieving state-of-the-art performance. It also delivers top-tier performance in the official Interspeech 2026 Audio Encoder Capability Challenge, further demonstrating robust generalization across diverse audio tasks. Together, these results validate the effectiveness of UniAE-MoE for unified audio understanding across speech, music, and general audio domains.

While Vision-Language-Action (VLA) models excel in static tasks, they struggle in dynamic environments where objects are in motion (e.g., conveyor belt manipulation). We identify three fundamental limitations hindering current VLAs in these scenarios: the perception gap, where static visual inputs lack temporal motion cues; the latency gap, where inference delays render actions obsolete; and the control gap, caused by the open-loop action chunk execution without real-time adjustment. In this work, we propose DSDyn-VLA, a Slow-Fast Dual-Stream Dynamic manipulation framework that integrates motion-aware foresighted planning with real-time residual correction. The slow Flow-Planner serves as a macro-planner. By enhancing the VLA with optical flow for temporal perception and a future state awareness mechanism to preemptively offset inference latency, it produces globally consistent, motion-aware action chunks. Complementing this, the fast Res-Refiner employs a lightweight RL policy to inject high-frequency, closed-loop corrections into the planned action chunks based on real-time observations. In addition, we introduce DynBench, a MuJoCo-based benchmark for dynamic object manipulation that comprises nine tasks. Extensive experiments demonstrate that DSDyn-VLA reduces the failure rate by over 76% compared to current SOTA method in high-latency setting on the Kinetix dynamic benchmark, while achieving about 6$\times$ the success rate of PI0.5 in real-world dynamic settings and about 5$\times$ on DynBench. We will open-source all the code and weights.

We propose an Improved version of the Lagrangian Finite Time Lyapunov Exponents(IFTLE) framework as a new quantifier for Comprehensive Synchronization analysis of multilayer networks undergo adaptive coupling scheme in order to incorporate the effects of deformation in synchrony due to the dynamic variations in the coupling strengths of intra-layer and inter-layer links. Theoretical formulation of the IFTLE along with its numerical validation has been provided using a three layer network of Chaotic Rössler oscillators. The numerical illustration clearly shows the efficacy of the proposed method revealing the need for comprehensive synchrony deformation analysis in adaptive coupling scenarios. Also we have stated a few properties of IFTLE useful for the quantification purpose.

Charged particle track reconstruction at the High-Luminosity Large Hadron Collider is challenged by dense hit environments and combinatorial ambiguities. Existing graph-based and Transformer approaches commonly rely on static geometric neighbourhoods, which cannot adapt as track-level representations evolve. We propose the Evolving-Geometry Transformer (EGT), a graph Transformer that reconstructs the hit-connectivity graph at successive encoder layers and incorporates relative hit coordinates as learnable attention biases. Across five simulation benchmarks spanning linear, helical, and reduced TrackML events with up to 200–500 simultaneous trajectories, EGT achieves the highest FitAccuracy on three of five settings under the shared benchmark protocol, reaching 79.6% on the most complex setting and exceeding the strongest baseline by 1.6 percentage points. Replacing the evolving topology with a static graph reduces the perfect-track rate from 81% to 29%, while removing the geometric bias reduces it to 73%. Under synthetic background injection equal to the number of signal hits, FitAccuracy decreases by 3.4 percentage points. The neural-network forward pass requires 31~ms per event on an NVIDIA A100 GPU. These results indicate that iterative refinement of hit associations improves complete-track recovery when local geometry alone is insufficient to determine trajectory membership.

Spatial intelligence requires maintaining a coherent understanding of the world as the embodied agent moves. Like humans, the agent must use its own motion to interpret changes across observations and update object locations and spatial relations accordingly. Despite spatial post-training having substantially broadened the spatial intelligence of vision-language models (VLMs), they still struggle with two atomic spatial capabilities: tracking self-motion and mapping the surrounding world during motion. To address this gap, we provide dense multi-turn supervision over interleaved atomic capabilities within each training episode, mimicking the visual experience of a continuously moving agent that reasons as it observes. To scale this up, we synthesize 11,738 motif-driven camera trajectories over a broad range of 3D scenes, supporting self-motion tracking, persistent object mapping, and rich spatial operations within each visual experience. By training models to reason over these atomic questions, our URUQI$_{\mathrm{Syn}}$-8B improves accuracy from 15.84% to 47.73% on our Uruqi benchmark comprising 52k questions across 2.7k episodes. URUQI-SI-Mix-8B further reaches 50.41%, comparable to the 50.08% achieved by GPT-6 Astra. Trained solely on our synthesized data, URUQI$_{\mathrm{Syn}}$-8B achieves an average relative accuracy improvement of 17.13% over its InternVL3-8B backbone across three external spatial benchmarks. These results highlight continuous visual experience as a scalable source of supervision for developing spatial cognition in VLMs.

Wireless split learning (SL) reduces on-device computation by offloading upper layers to a server, yet transmitting high-dimensional intermediate features at each iteration remains a major communication bottleneck. Existing methods select features at the client side using task-agnostic criteria such as magnitude, statistics, or clustering, which increases client-side processing and often degrades accuracy under non-independent and identically distributed (non-i.i.d.) client data. We propose importance-aware class-balanced sparsification (ICS), a lightweight approach in which the server ranks feature channels using Grad-CAM-based scores obtained from the true-class logit during backpropagation. The per-class scores are aggregated into a class-balanced, label-agnostic importance vector that mitigates head-class bias under label skew, and each client reuses this vector in the next round to retain the top-$N$ feature channels, incurring no additional client-side forward or backward passes. We further derive a non-asymptotic convergence bound that isolates the sparsification-induced error and characterizes how the sparsification ratio and mini-batch size jointly affect convergence under a fixed communication budget, and we analyze the communication and computational overhead of ICS against representative baselines. Beyond sequential CNN-based SL, we extend ICS to parallel split learning and to transformer-based models. Experiments show that ICS consistently outperforms the baselines, with larger gains under severe non-i.i.d. partitions.

We study self-interacting dark matter composed of heavy baryons in a confining dark $SU(N)$ gauge theory with a single heavy vector-like quark, $m_Q \gg \Lambda_d$. In this regime, the dark baryons are compact Coulombic bound states whose long-distance interactions can be described by a multipole expansion. We derive the leading baryon-baryon interaction by adapting the quarkonium operator-product-expansion formalism to heavy baryons. The resulting force contains non-retarded London dispersion, retarded Casimir-Polder, and Yukawa contributions from dark glueball exchange. We find that the self-interaction phenomenology depends strongly on the number of colors. For $N=2,3$, the attractive interaction is too weak to overcome the short-distance Pauli repulsion, and no phenomenologically relevant velocity dependence is obtained within our minimal set-up. For sufficiently large $N$, the attractive interaction remains competitive with the repulsive core and can generate strong velocity dependent cross sections. Taking $SU(10)$ as a representative finite-$N$ realization, we find regions compatible with self-interaction requirements inferred from dwarf galaxies and clusters, while allowing much larger cross sections at low velocities relevant for gravothermal evolution. These regions correspond to GeV-scale dark baryons and a confinement scale close to that of QCD, which naturally motivate an asymmetric dark matter origin.

Radchenko and Wheeler (RW) recently proved a finite pentagon relation for real quadratic special values of the modular quantum dilogarithm and used it to establish the rank-$1$ twisted convolution identity conjectured by the current authors. RW gave an explicit argument in the principal case and remarked that their proof holds for all rank-$1$ admissible tuples. We extend their proof to all rank-$r$ admissible tuples and provide an explicit dictionary between the modular quantum dilogarithm and the Shintani-Faddeev modular cocycle conventions in the respective papers. Thus, we establish that, if $d,r$ are positive integers such that $r<\frac{d-1}{2}$ and $\frac{d^2-1}{r(d-r)} \in \mathbb{Z}$, then there exist ghost $r$-SICs: i.e., configurations of $d^2$ rank-$r$ subspaces in $\mathbb{C}^d$ that satisfy a non-Hermitian equichordal condition. Under the Stark conjecture, these configurations are Galois conjugate to Hermitian equichordal configurations called $r$-SICs (or rank-$r$ SIC-POVMs).

The extreme therapeutic resistance of malignant gliomas motivates the development of multimodal treatment strategies. We present a mechanistic mathematical framework to investigate the coupled dynamics of radiotherapy, temozolomide, and chimeric antigen receptor (CAR) T-cell therapy. The model is formulated as a nine-dimensional impulsive dynamical system that captures proliferative-quiescent tumor states, treatment-specific resistance, tissue damage, TMZ pharmacokinetics, and the lymphotoxic interferences exerted by conventional regimens on engineered T cells. Mathematical analysis establishes the existence of biologically meaningful solutions, maps invariant structures, and identifies threshold conditions under which sustained therapy locally stabilizes the tumor-free equilibrium. After study-matched benchmarking against clinical trials, including the standard Stupp protocol, we deploy heterogeneous virtual patient cohorts to evaluate five combined CAR-T–Stupp sequences. At the population level, the immunotherapeutic adjunct yields a consistent but modest survival extension, increasing median overall survival by $0.8$–$0.9$ months with minimal sensitivity to temporal ordering or scheduling perturbations. Crucially, these population-level medians mask profound individual heterogeneity: under combined protocols, $19\%$–$22\%$ of virtual patients extend survival by more than $60$ days, and $\sim 3\%$ gain over a year. A low intrinsic tumor proliferation rate $r_1$ emerges as the primary determinant of response, bolstered by robust T-cell expansion and attenuated microenvironmental suppression. These findings suggest that the clinical promise of multimodal integration depends on identifying responsive tumor phenotypes for personalized treatments rather than searching for a single universally optimal timeline.

A deep classifier is defined not only by the decision it produces, but also by the sequence of transformations through which that decision is formed. Treating this evolution as a dynamical system across layers provides a natural framework for asking how decision geometry emerges through depth and how far back we can trace a boundary's dynamical signature. We model a feed-forward classifier as a finite, nonautonomous discrete dynamical system, with layers playing the role of discrete time steps. We study the Finite-Time Maximum Lyapunov Exponent (FTMLE) of the data samples' dynamical trajectory through depths of the classifier. The FTMLE measures the rate of convergence/divergence of nearby trajectories. We move the observation endpoint backward from probabilities to logits and then to hidden representations. For Gaussian classes, we prove that probability-level FTMLE carries a clear geometric signature of the decision boundary, with its dominant direction aligned with the boundary normal. Moving one step backward to the logits, we prove this relationship is no longer universal but depends critically on how the classifier is trained, particularly on the choice of loss function. Moving further backward to the hidden representation, the connection becomes more conditional: boundary-related FTMLE can persist, but only under identifiable structural conditions. We propose geometry-aware fine-tuning for restructuring the classifier's hidden FTMLE, and propose conditions for guaranteed concentration of high hidden FTMLE near the decision boundary. Through our numerical results, we show the generality and validity of our theoretical results. Understanding the evolution of data samples as traveling through the layers of classifier provides a principled foundation for identifying where boundary-relevant sensitivity emerges and for developing layer-aware regularization strategies.

Trustworthy Legal AI requires systems that can answer legal questions while grounding their responses in authoritative sources. However, existing Vietnamese legal benchmarks provide limited coverage of real-world legal consultations. We introduce ViLegalExpert, a large-scale benchmark constructed from authentic citizen–lawyer consultations, containing over 172K questions across 34 legal domains, together with professional answers and expert-verified legal evidence. ViLegalExpert supports legal information retrieval, extractive QA, and abstractive QA. Experiments with representative retrieval methods and language models reveal substantial challenges in evidence retrieval and grounded answer generation. While pretrained models perform strongly on QA, hybrid retrieval achieves the best retrieval performance. These results demonstrate the difficulty of mapping naturally expressed legal questions to authoritative provisions and establish ViLegalExpert as a challenging benchmark for reliable Vietnamese Legal AI.

Developmental psychology holds that certain priors are given to infants prior to experience rather than induced from data, and that the influence of such priors is suppressed under strong, well-constrained conditions but reasserts itself under weak ones. We ask whether an analogous principle holds for the Transformer: can the activation function given to attention heads serve as an intrinsic inductive bias? We propose Mixture of Function Attention (MoFA), a parameter-free modification to multi-head attention that fixes a ratio of softmax and sigmoid heads before training. Across five ratios, a 124M-parameter GPT-2 model, and five seeds, we find that this given ratio has little effect in-distribution – differences between ratios are statistically negligible for moderate mixtures and remain small even at the extremes – but its influence re-emerges sharply under zero-shot distribution shift across 15 out-of-distribution domains. Perplexity gaps between ratios widen by more than an order of magnitude on several domains, and the best-performing ratio tracks a single axis of domain structure, separating short, informal text (softmax-favoring) from technical, long-form text (sigmoid-favoring), that explains 78.3% of the variance in domain response. This reorganization is visible at the head level: sigmoid heads show an accelerating drop in attention entropy as their ratio increases, while softmax heads respond more modestly, yielding a consistent division of labor between the two head types. Our results suggest that activation choice functions as a given prior whose influence is masked in-distribution and re-emerges out-of-distribution.

Plate tectonics is described in two registers: a discrete one, in which rigid plates tile the sphere and meet at trivalent junctions, and a continuous one, in which Bercovici and Wessel (1994) replace hard plate outlines by smooth shape functions of finite boundary half-width $\delta^*$. We prove that the two registers are connected by a limit theorem. Generalizing the shape functions to a signed-geodesic-distance construction, we show that they converge almost everywhere, exponentially in $1/\delta^*$, to the plate indicator functions; that the sharp plate mosaic is a finite regular trivalent CW decomposition of the sphere under an explicit structural hypothesis; and that a component-counted weighted Euler characteristic $\chi_w^{\delta^*}(\tau)$, built from the super-level sets of the shape functions, equals $V-E+F=2$ for all $\delta^*$ below an explicit threshold $\delta_0(\tau)$, with its face, edge and vertex terms converging separately to the numbers of plates, boundary arcs and triple junctions. The theorem is proved unconditionally on the signed-distance offset cover under positive-reach, separation and bounded-sector hypotheses, and transferred to the normalized partition-of-unity field for thresholds $\tau<1/3$ by a comparison lemma whose junction-ball step, a planar single-crossing property, is certified numerically. The residual $|\chi_w-2|$ defines a topological diffuseness index for diffuse plate boundaries. For the present-day PB2002 network the hypotheses are measured to be non-vacuous, with $\delta_0(0.15)\approx2.8$ km. The theorem is the mathematical foundation of the topological audit of plate reconstructions presented in a companion paper (Kim, 2026, submitted to Geoscience Frontiers).

We investigate whether a spherically symmetric black hole can have a complete inner end at finite, nonzero areal radius while satisfying the standard energy conditions sufficiently far outside its horizon. Retaining two independent metric functions, we treat both null-type and spacelike-type asymptotic throats and derive a necessary and sufficient integral criterion for causal geodesic completeness. Under boundedness and finite-limit assumptions on the independent curvature components, null-type ends have limiting curvature fixed by the throat radius, whereas spacelike-type ends admit an additional nonnegative curvature parameter. We obtain criteria for bounded curvature in timelike and null parallel-propagated frames. Complete spacelike-type examples can have bounded scalar invariants and bounded timelike-frame curvature but unbounded null-frame curvature at infinite affine parameter. We also prove that the radial null energy condition must fail arbitrarily close to every complete inner end in the stated class. Nevertheless, explicit real-analytic constructions with either type of throat satisfy the null, weak, dominant, and strong energy conditions throughout a sufficiently distant exterior region. Smooth constructions can become exactly Schwarzschild beyond a finite radius. These results establish the compatibility of complete asymptotic throats with exterior energy conditions while distinguishing completeness from different levels of curvature regularity.

Mamba-style and hybrid language models compress their past into a fixed-size recurrent state that is rewritten at every generated token. Storing this state in low precision saves memory bandwidth, but every rounding error is fed back into the next update and can accumulate over long generations. Production systems round the state stochastically; we ask which rounding rule such caches should use. We find that a deterministic golden-ratio Weyl dither, which needs no random numbers, consistently brings the quantized model closer to the full-precision one than stochastic rounding, across pure and hybrid models, storage formats, and long decoding horizons, at no extra cost. Round-to-nearest behaves differently: because it discards small updates, its error keeps growing, so it can look best in short evaluations yet falls far behind over long generations. A discrepancy analysis explains this ordering, and we document implementation pitfalls that silently remove the benefit.

Fiber orientation and compartmental microstructure are central to the characterization of white matter tissue in diffusion MRI, yet existing methods either resolve fiber orientations without quantifying microstructure, or quantify microstructure while assuming a fixed number of compartments and a single fiber direction. Nonparametric approaches that recover both require tensor-valued diffusion encoding and computationally expensive Monte-Carlo inversion of an ill-posed inverse Laplace transform. We propose to reframe this problem as an object detection-like task, adopting the Detection Transformer (DETR) architecture to jointly predict mean diffusivity (MD), fractional anisotropy (FA), main fiber direction, and signal fraction for a variable number of compartments per voxel from standard multi-shell diffusion MRI with linear encoding. Hungarian matching during training resolves permutation invariance across compartments. We introduce mean Average Precision as a reproducible benchmark metric. Evaluated on synthetic test data with up to five compartments per voxel, our model achieves $R^2=0.95$ for MD, $R^2=0.88$ for FA, and a median angular error of 4.2{\deg}, with performance scaling naturally with compartmental signal fraction.

This paper studies thinking–answer consistency in vision-language models. We focus on Visual Intention Grounding, where a model infers a target object based on a human intention query and predicts a bounding box. We reveal that previous IoU-based reinforcement learning (RL) frameworks suffer from “thinking drift”, where the model produces a correct bounding box, despite having an incorrect reasoning process pointing to a different target object. Thus, we propose Rita (ReInforcing Thinking–Answer consistency) as a novel RL paradigm to tame the drift. Specifically, Rita introduces two reasoning-label-free RL rewards, constructed from the conditional probability of reference answers: a thinking reward and a consistency reward. It also adopts a difficulty-aware data filtering strategy that selects informative easy-to-medium samples for RL using rollout error rate and reward variance. Extensive experiments on EgoIntention and the new RefEgo-Int benchmarks show that Rita performs consistently superior to the supervised finetuning approaches and vanilla RL-finetuned frameworks.

World Models are appearing as the next major frontier in computer vision. However, their robustness is currently largely unexplored. We identify the phenomenon of hallucination in latent World Models: given a state and an action, the predicted next latent can decode to a scene that never occurs. Because the prediction is statistically ordinary and is fed back autoregressively by the model, the error is both silent and compounding. We study whether such latent hallucination can be detected, localised, and corrected at inference time, on a frozen self-supervised world model in the absence of ground-truth error labels. We introduce Masked Empirical-Bayes Neural Denoising (MEND), a single conditional score network trained by denoising score matching on real transitions, whose score field serves three roles: its magnitude detects hallucination, its per-token field localises it to specific image patches, and it defines an inference-time correction direction. On two navigation environments MEND detects hallucination with an AUROC of up to 0.80 without using actions, exceeding a single-Gaussian density baseline while also localising the error (per-token AUPRC up to 0.87) and correcting it, all from one score field. Our correction reliably reduces single-step latent error and improves predictions. We identify that a part of the error is tangent to the data manifold, hence, we focus on detection and localisation while highlighting promises of the correction.

In 1943, R. Salem asked whether the Fourier-Stieltjes transform of the Minkowski question-mark function vanishes at infinity. This problem was answered affirmatively by Jordan and Sahlsten in 2016 as a consequence of their general results on Gibbs measures for the Gauss map. In this note we give a self-contained proof for ?(x). Several aspects of our argument are substantially different, making a very short proof possible in the ?(x) case. Moreover, we obtain an explicit polynomial decay estimate, with exponent at least 0.0472. We then prove the existence of the correlation dimension of the Minkowski measure, thereby improving the previous upper bound for the optimal Fourier decay exponent from 0.4373 to 0.4223.

Over-the-air (OTA) radar target emulation is an important method for evaluating the sensing capabilities of integrated sensing and communication (ISAC) and radar devices. Conventional single-feed compact antenna test ranges (CATRs) are generally limited to a single target direction, while multi-feed angular-synthesis methods commonly used in radar testing still require far-field illumination over the device-under-test (DUT) aperture. To overcome these limitations, this paper proposes a radar target OTA emulation framework based on multi-feed plane wave generator (PWG). The key capability is the simultaneous emulation of spatially close targets from different directions with joint angle–range–Doppler control. Target range is emulated through the corresponding delay, while Doppler and the complex echo coefficient are controlled by a channel emulator. Meanwhile, a multi-input amplitude and phase matrix (APM) maps each independent target-signal path to a dedicated PWG excitation vector that synthesizes the corresponding incident plane wave over the same test zone. This enables multiple target directions to be reproduced as local plane waves at a test distance substantially shorter than the conventional far-field requirement. The framework is experimentally validated at 3~GHz using a real $1\times11$ phased-array DUT. The PWG–DUT separation is 1.6~m, substantially shorter than the 7.2~m Fraunhofer distance of the DUT. At this separation, the synthesized $0^{\circ}$ and $10^{\circ}$ incident waves each exhibit an amplitude peak-to-peak error of 1.0~dB. Across two four-target angle–delay cases and one two-target angle–velocity case, the estimated target parameters show angular deviations within $1^{\circ}$, interpolated delay errors within 2.4~ns, and velocity errors within 0.15~m/s.

High-precision locomotion combines motion-command tracking with precise regulation of task-relevant physical states, enabling robots to interact reliably with their surroundings during motion. Joint end-to-end optimization can leave precision objectives insufficiently optimized, while reactive residual control adjusts actions only after deviations become observable. We present LocoWM, a world-model-guided preactive residual adaptation framework for high-precision locomotion. A base policy provides command-following locomotion, while an action-conditioned world model predicts a sequence of future physical states from proprioceptive history and the proposed base action. A residual adapter conditions on this predicted sequence to generate additive action corrections that compensate for anticipated deviations. Two-stage training first learns locomotion and action-conditioned dynamics, then freezes both modules while training the adapter, separating locomotion acquisition from precision adaptation. Experiments spanning terrain leveling, acceleration compensation, and push recovery demonstrate improved control precision and disturbance robustness over end-to-end and reactive residual baselines. Demos and code are available at: https://zhaozijie2022.github.io/LocoWM