Differential Geometry
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Wednesday 2 September
Let $(M,\mathcal{D},g)$ be a smooth equiregular sub-Riemannian manifold equipped with a smooth positive measure $\mu$. We study the small-scale limit of the metric-ball mean-value operator \[ A_hf(x)=\frac{1}{h^{2}\mu(B(x,h)) }\int_{B(x,h)}(f(q)-f(x))\, d\mu(q). \] Exact homogeneity yields the pointwise limit on Carnot groups. On a general equiregular manifold, convergence for every smooth test function is equivalent to convergence of the rescaled horizontal first moments of metric balls in first-kind privileged coordinates. This criterion is independent of $\mu$; when it holds, the principal symbol is determined by the normalized second-moment tensor of the tangent unit ball, and the drift satisfies an explicit change-of-measure formula. In step at most two, we verify the criterion by combining a real-analytic finite-jet reduction with tame integration, which rules out oscillation of the normalized moments. We also compute the limit on Lie groups and prove unconditional distributional convergence of the volume-weighted operators on every equiregular manifold.
Materials Science
From Goldene to Noblene: exhaustive enumeration of the ordered Au-Ag-Cu monolayer alloys
Two-dimensional metals became a laboratory reality with the isolation of Goldene, a gold sheet one atom thick released by chemical exfoliation, which raises the question of what alloying can achieve on the same close-packed lattice. Here we combine an exhaustive enumeration of derivative superstructures with density functional theory to map the ordered Au-Ag-Cu monolayer alloys, relaxing every symmetry-inequivalent arrangement up to four atoms per cell and extending the set with larger cells selected by a fitted model. We find that the arrangement of the atoms, and not their proportion, controls the mixing energy, since the spread among the orderings of a single composition is several times the step between neighboring compositions, one composition spans more than 50~meV/atom between two arrangements of the same three atoms, and at two compositions the arrangement decides the sign. The energetics is carried by a competition between Au-Cu contacts, which bind, and Ag-Cu contacts, which do not, and bulk enthalpies computed under the same protocol show that this chemistry is not uniformly rescaled in two dimensions, since Au-Cu orders more strongly in the monolayer than in the crystal while the other two weaken. A pair model completed by an elastic size-mismatch term reproduces the computed energies to a few meV/atom and returns a composition landscape whose convex hull no balanced ternary composition reaches. One equimolar ordering, which we name Noblene, is the only structure in which every atom is surrounded exclusively by unlike species, and it is dynamically stable and metallic, and carries a Poisson's ratio above one half, higher than that of any of the three pure monolayers. Since Noblene sits on the lattice that Goldene already realizes, the route that produced Goldene is a plausible starting point for its synthesis.
Read the full abstract
Earth and Planetary Astrophysics
The Hill sphere transits of Beta Pictoris c and the search for another 1981-like event
Beta Pictoris is a young and nearby planetary system hosting an edge-on debris disk and at least three gas-giant planets. Their circumplanetary environments could host exomoons and rings, a detection of which would be highly informative for moon and planet formation theories. A photometric fluctuation of about 4% was seen towards Beta Pictoris in 1981, indicating the transit of dust in the system, which could be associated with the Hill spheres of the two inner planets. We search for the origin of the 1981 event by searching for an analogous event in multi-epoch photometry from 2017 to 2023, and look for signs of circumplanetary material in transits of the Hill sphere of Beta Pictoris c. Observations from the BRITE satellite, and the bRing and ASTEP observatories are fitted to a model of the 1981 event to search for a similar event, and also search for a signal consistent with a circumplanetary disk transit using a simple flat disk model. No compelling evidence for a 1981 event during a Hill sphere primary transit is found, although we do find a candidate event around 2019 July 26. Due to the uncertainty in the time of closest projected separation of Beta Pictoris c, a search for a disk during the primary transit of 2018 could not be robustly determined, but bRing photometry for the second primary transit places an upper limit on the dust content of the Hill sphere of ~ $10^{22}$ grams of material. With no compelling detection of an event similar to that seen in 1981 during the transits of Beta Pictoris c in the photometric time series, we rule out the hypothesis that this event was related to the Hill spheres of Beta Pictoris c and Beta Pictoris b. Future observations of the next Hill sphere transit in 2028 can be realised with both ground-based observatories and with PLATO, whose first long duration science pointing will include Beta Pictoris.
Read the full abstract
Artificial Intelligence
Door-in-the-Face Requests and Refusal Behaviour in Large Language Models
Does the door-in-the-face technique work on language models? In humans, a large request that is refused makes a smaller follow-up request more likely to be granted. We test this on nine production models from three providers: each model refuses a large request, then receives a smaller version of the same request, and we compare its compliance with asking directly. The answer depends on the model. On Anthropic's frontier models the technique works: Opus 5 answers the smaller request 65.8% of the time after refusing the larger one, against 29.3% when asked directly. On the frontier models of OpenAI and Google, and on Haiku 4.5, it backfires, lowering compliance by 15.5 to 23.0 points. A control locates the effect: a refused large request on an unrelated topic does less than the related one on all nine models, so the concession itself matters everywhere, while the reaction to having just refused something differs by model family. The technique does not transfer to refusals drawn from public benchmarks. What decides whether a retreat can work is what the request asks for: rewriting 265 refused requests for usable instructions into requests for explanations of the same topic removed the refusal in 263 cases. Human influence techniques port to language models one model family at a time.
Read the full abstract
Geometric Topology
On the length conjecture for twist knots
Using the newly developed $3$d quantum trace map, we compile more evidence for the length conjecture, a refinement of the all-order volume conjecture that incorporates insertions of additional links in the skein module of the knot complement. In particular, we prove that the length conjecture holds up to first order for an infinite family of twist knots. Along the way, we form a conjecture that the $3$d quantum trace map behaves naturally with respect to Dehn filling.
Read the full abstract
Computer Vision
A Top-Down Framework for Metric-Scale Athlete Localization from Single Broadcast Frames
Accurate world-coordinate localization of athletes from single-frame broadcast footage is inherently challenging due to extreme scale disparities in ultra-high-resolution imagery. In this paper, we propose a top-down framework for metric-scale athlete localization from a single calibrated frame. Our approach centers on three key contributions. First, we propose Boundary-Aware Adaptive Tiling, a semantics-guided extension of standard sliced inference. By iteratively expanding tile boundaries based on coarse bounding-box predictions, it systematically ensures full object containment, effectively mitigating boundary-splitting artifacts through a lightweight pipeline adaptation without architectural modifications. By substantially mitigating recall degradation under extreme scale variance, Boundary-Aware Adaptive Tiling enables us to isolate perspective distortion as the primary source of residual localization error. Second, we adapt the RTMPose-X architecture into a specialized two-keypoint estimator (pelvis and ground projection), employing a reformulated Gated Attention Unit optimized for this geometrically coupled point pair, and then deterministically lift the 2D ground projections into world coordinates via camera-calibrated ray casting. On the public test set, our method achieves a LocSim score of 97.44 and an mAP of 0.9128, outperforming the baseline by over 21 % and establishing a robust solution for high-resolution scale variance.
Read the full abstract
Nuclear Theory
Halo-EFT: an effective and efficient tool to study reactions with halo nuclei
Halo nuclei are exotic nuclei, which exhibit a much larger matter radius than their isobars. This unusual size is now understood as a threshold effect, in which one or two valence nucleons are loosely bound to the core of the nucleus. Thanks to this loose binding, halo nuclei exhibit a strongly clusterised structure with a diffuse neutron halo surrounding a compact core. Being short lived, halo nuclei are mostly studied through reactions, such as breakup. In this talk we show that including Halo Effective Field Theory within existing reaction codes is both effective and efficient to analyse experimental data. We first apply this idea to study the Coulomb breakup of 19C within a Bayesian approach. Second, we extend Halo-EFT to include the core's excitation to test the sensitivity of the Coulomb breakup of 11Be on spectroscopic factors.
Read the full abstract
Logic
A choice-free proof of the Erdős–Dushnik–Miller theorem
The Erdős–Dushnik–Miller theorem states that for every aleph $\kappa$, \[ \kappa\to(\kappa,\omega); \] that is, every coloring $c:[\kappa]^2\to2$ has either a $0$-homogeneous set of cardinality $\kappa$ or a $1$-homogeneous set of cardinality $\omega$. In this article, we present a purely combinatorial proof of this theorem in $\mathsf{ZF}$ (i.e., Zermelo–Fraenkel set theory without the axiom of choice), avoiding any metamathematical considerations.
Read the full abstract
Computation and Language
Trace as State: Reasoning Traces as Conditional States for Long-Context Transformers
Transformers process information causally, but long-context reasoning may depend on task state discovered only later. We formalize this mismatch through conditional state update tasks. For causal state update processors, providing the condition first can require exponentially less memory in the worst case than providing it last. Motivated by this principle, we introduce Trace as State. We use collected reasoning traces as a textual proxy for task state and place it before the long-context block on a fresh pass, allowing information derived previously to guide rereading. We conduct extensive experiments on Trace as State and Trace Append, a matched control that uses the same task state proxy but put it after the context. Across three models and three long-context datasets, Trace as State outperforms Trace Append in 26 of 27 reported combinations of model, task, and metric. On GraphWalks Parents, exact match lifts DeepSeek V4 Pro Preview from 29.2% on the initial pass and 43.0% with Trace Appendto 81.8% with Trace as State, and from 66.4% and 83.2% to 100.0% for GLM-5.2. These results show that placing traces before the context can improve long-context reasoning while retaining the causal transformer structure.
Read the full abstract
Geometric Topology
Circuit Decomposition for Triangulations of Surfaces
An Euler circuit of a graph is a closed path that visits every edge of the graph exactly once. Euler circuit and circuit decomposition problems can also be formulated for higher dimensional simplicial complexes. An Euler k-circuit in K is a cyclic sequence of vertices v_1...v_n such that every k+1 adjacent terms { v_i,v_{i+1},...,v_{i+k} } (indexed modulo n) form a k-simplex, and every k-simplex of K appears exactly once in the sequence v_1v_2...v_n(v_1v_2...v_k). We investigate the 2-circuit decomposition problem for triangulated closed compact surfaces. For an orientable triangulated surface we use interior angles of paths to define an obstruction that lives in the first cohomology of the surface. It vanishes if and only if the surface has a 2-circuit decomposition. We also show that a non-orientable triangulated surface has a 2-circuit decomposition if and only if its orientable 2-fold cover does.
Read the full abstract
Human-Computer Interaction
The PIONEER Project: A PrIvacy companion for mOtivatioN and knowlEdge transfER
Remaining control over their private data is one of the key challenges in this century for users. We know from prior work that users are often neither in a position to fully grasp the content of the usually complicated texts, nor are they motivated to spend the time necessary to do so. We report on the progress made by the PIONEER project on a privacy support tool that combines knowledge transfer and persuasive elements to increase users' privacy awareness and motivation; thus empowering them to more privacy sovereignty. Throughout the research and design process, we consider user group specifics that may result in different requirements, e.g., for children, adolescents, parents, or elderly people. We further target sustainable behavior change by addressing different states of change, precisely: spark initial motivation, facilitate the creation of new habits, and encourage habituation of these habits in the long term (volition). Finally, we provide a privacy support tool demonstrator that can be utilized for research and education purposes, e.g., in school contexts.
Read the full abstract
Differential Geometry
Gauge theory and symplectic structures
Motivated by the problem of finding obstructions to the existence of symplectic structures in dimensions higher than four, we introduce an elliptic system of equations on almost Hermitian manifolds that reduces to the Seiberg–Witten equations with Taubes' perturbation in dimension four. To define the equations, one needs to choose a spin$^{\mathbb C}$ structure on the manifold. We prove that the system has index zero for the canonical spin$^{\mathbb C}$ structure. Moreover, on almost Kähler (symplectic) manifolds, we construct a canonical solution and prove its transversality with large values of the perturbation parameter, while in dimension six we obtain uniqueness of the canonical solution under a certain assumption. We also show that on Kähler manifolds the equations reduce, under a natural ansatz, to the vortex equations. These results provide evidence toward a possible higher dimensional gauge theory capable of producing invariants of almost complex structures and ultimately, obstructions to compatible symplectic forms.
Read the full abstract
Optics
Random-mapped intensity optical neural network: all-optical two-layer computing for multimodal optical-field inference
Free-space optical neural networks offer distinct advantages for computational imaging and machine vision because they can compute directly on incident optical fields. However, conventional ONNs composed of cascaded linear optical components are bound to a general linear input-output relation with square-law detection between the input field $\mathbf{x}$ and output score $\mathbf{y}$, $\mathbf{y}=|\mathbf{T}\mathbf{x}|^2$, where at best every complex-valued element of the transmission matrix $\mathbf{T}$ is trainable. This restricts each output score to a quadratic form $\mathbf{x}^\dagger\mathbf{A}\mathbf{x}$ with rank-one decision matrix $\mathbf{A}=\mathbf{t}^\dagger\mathbf{t}$. Here, we present a random-mapped intensity optical neural network (RMI-ONN) as an all-optical two-layer computational network that lifts this rank-one limit. We numerically demonstrate that a high-dimensional feature projection by a disordered medium, a programmable nonnegative intensity mask, and segmented spatial power summation together can surpass the rank-one ceiling through the expressivity of higher-rank quadratic decision boundaries. Furthermore, exploiting the vectorial coherent wave-mixing nature of the disordered medium, we experimentally validate multimodal classification of amplitude, phase, and polarization on MNIST, Fashion-MNIST, and Quick Draw with the RMI-ONN, under a single optical configuration across all encoding domains. These results provide a practical and conceptual basis for scalable direct-field optical processors capable of exploiting amplitude, phase, and polarization information within a unified intensity-based inference framework.
Read the full abstract
Computer Vision
Generating Medical Image Counterfactuals using Causal Explanations
Deep learning models have achieved impressive performance in medical image diagnosis, yet their deployment in clinical settings remains constrained by limited explainability. Counterfactual images provide one means of auditing model behavior by showing how an image would need to change for a classifier to produce a different prediction. Existing approaches typically generate such explanations using auxiliary models, including generative adversarial networks and diffusion models. While often capable of producing visually realistic images, these methods explain one black-box model using another, making it difficult to separate the classifier's decision-making process from the inductive biases of the generator. We propose a novel counterfactual-generation framework that requires no generative model. Instead, counterfactuals are constructed directly from causal evidence extracted from the classifier. The resulting approach is deterministic, requires no additional model training, and enables controllable edits within user-specified regions of interest. Experiments on real-world medical imaging datasets demonstrate that the proposed method successfully changes classifier predictions while remaining closer to the original image than generative baselines, providing a more direct and transparent view of the classifier's decision boundary.
Read the full abstract
Biological Physics
A viscoelastic theory for ultrasound-induced intracellular streaming
Ultrasound is increasingly used to control biological cells, but its multiple physical effects, including radiation force, streaming, heating, and cavitation, require theoretical frameworks that can quantify their relative strengths. Here we develop a semi-analytical model to predict flow patterns and energy distributions of ultrasound-induced intracellular streaming. Cells are modeled as viscoelastic droplets immersed in a fluid with different viscoelastic properties. For this geometry, the momentum equation with an Oldroyd-B constitutive law is solved using a perturbation expansion commonly applied in acoustofluidics. The resulting equations for the different orders of the expansion are solved using partial wave expansion and explicit integration. We investigate our model for a range of parameters which is relevant both for synthetic polymer and for protein solutions. We find a series of flow reversals that we can connect to the distribution of energy into the different modes. We also consider density and compressibility contrasts, which shift these transitions and introduce new ones. Finally, we discuss the potential biological relevance of our mathematical results for cellular sensing and signaling.
Read the full abstract
Numerical Analysis
Positivity loss in bandlimited spectral reproduction on spheres
How small can the positivity loss be for an $N$-bandlimited spectral operator that exactly reproduces all modes up to degree $L$? For spherical polynomial approximation on $\mathbb S^d$, we prove that the smallest possible excess of the uniform operator norm above 1, equivalently the least positivity loss, is of sharp order $\left({L}/{(N+1)}\right)^2$ when $1\le L< N$. The lower bound follows from a Fejér peak test and a concentration estimate for bandlimited kernels, while a matching upper bound is obtained by correcting a positive Jackson operator with a smooth filter. We illustrate the result in three settings. On the circle, taking $N=sL-1$, this determines the sharp order of the generalized-projection constant above 1 and identifies the gap between the $s^{-1}$ excess of delayed de la Vallée–Poussin means and the optimal $s^{-2}$ order. For filtered hyperinterpolation, whose operator norm has long been known to be uniformly bounded, we give a quantitative lower bound on its separation from the positivity threshold 1. Finally, we identify an operator-level obstruction to maximum principles.
Read the full abstract
Computational Physics
RBMD 2.0: Random batch molecular dynamics package for large-scale simulations on multi-GPU architectures
Large-scale molecular dynamics simulations of particle systems on multi-GPU architectures are often constrained by the computational and communication costs of nonbonded force evaluation. We present RBMD 2.0, a major new release of the random batch molecular dynamics package designed for cross-node multi-GPU simulations of large-scale systems. It combines the improved random batch Ewald method with three-dimensional domain decomposition and ghost-particle communication to accelerate multi-GPU nonbonded force evaluation, while the DTK CUDA framework facilitates portability across heterogeneous accelerator architectures. Numerical experiments on multiple benchmark systems demonstrate both the accuracy and efficiency of simulations with RBMD 2.0. For simulations involving up to hundreds of millions of particles across multiple accelerator devices, one achieves speedups ranging from severalfold to approximately two orders of magnitude in nonbonded force evaluation while exhibiting over $97.5\%$ weak-scaling behavior. These results demonstrate the promising nature of RBMD 2.0 as a computational engine for future exascale molecular dynamics simulations.
Read the full abstract
Earth and Planetary Astrophysics
Galactic tides in the Solar System within a non-axisymmetric Milky Way model adjusted to Gaia data
Context. Galactic tides are external differential forces acting on extended systems immersed in the Galactic potential. They play a key role in the dynamics of comets in the Oort cloud. Aims. We aim to establish the strength of Galactic tides from the non-axisymmetric potential of the Milky Way revealed by the Gaia mission, and how this strength evolves along the Galactic trajectory of the Solar System. Methods. We derived expressions for Galactic tide parameters independently of any simplified trajectory for a star in the Galaxy or any specific symmetry of the Galactic potential. We obtained six parameters, $G_1$ to $G_6$, that quantify the influence of tides. Using the most up-to-date Galactic potential model, extended here to three dimensions, we studied the time evolution of these parameters along the Sun's trajectory using a statistical approach. Results. Even for Galactic trajectories featuring modest radial and vertical excursions, as investigated here, the dominant Galactic parameter, $G_3$, is found to vary by an order of magnitude along the solar trajectory. $G_1$ and $G_2$ reach up to about one half and one quarter, respectively, of $G_3$ along the solar trajectory, whilst parameters $G_4$ to $G_6$ reach up to one tenth of $G_3$. Conclusions. Contrary to what is often assumed, all Galactic tide parameters vary widely along the Galactic trajectory of the Solar System. Two consequences can be expected: First, $G_3$ directly affects the flux of observable long-period comets and the extent of the fossilised Sednoids region; second, $G_1$ and $G_2$, and to a lesser extent, $G_4$ to $G_6$, break the integrability of the dynamics, potentially affecting the long-term structure of the Oort cloud. Additionally, the new parameters $G_4$ to $G_6$, while small for the Solar System, may strongly impact extrasolar systems with a large out-of-the-plane excursion in the Galaxy.
Read the full abstract
Soft Condensed Matter
Response of a Model Glass to Athermal Quasistatic Pinching
We numerically investigate the mechanical response of amorphous solids to localized force dipoles, referred to as pinching, using athermal quasistatic simulations of model glasses with varying degrees of stability. We employ a control parameter, the imposed extension, corresponding to the increase in rest length between the two pinched particles relative to their initial separation. Increasing this extension allows us to continuously tune the system from the elastic to the plastic regime. For small extensions, the response remains elastic. In this regime, the displacement field induced by pinching is well described by linear elasticity and exhibits a long-range power-law decay consistent with dipolar forcing. Averaged responses display anisotropic, quadrupolar-like patterns, with quantitative agreement between simulations and analytical predictions. This continuum description remains valid down to particle-scale distances. As the imposed extension increases, the response becomes plastic. Pinching can then trigger either localized or system-spanning rearrangements, depending on glass stability. Well-annealed glasses exhibit localized plastic events, whereas poorly annealed systems display delocalized cascades. We introduce a method to extract the principal axis of plastic deformation and analyze the associated displacement fields and plastic activity. Overall, our results demonstrate that pinching provides a minimal local probe of amorphous solids. The resulting response, governed by both glass stability and imposed extension, offers insight into the interplay between elasticity, elementary rearrangements, and the emergence of collective plasticity.
Read the full abstract
High Energy Physics (Phenomenology)
Entanglement entropy minimization and global symmetry violation in scatterings
The concept of quantum information has provided new ways to investigate the theoretical structure of particle interactions. In particular, it has been shown that extrema of quantum entanglement in scattering and decay processes can be related to symmetries and characteristic properties of particles. We focus on a more specific question: whether the minimization of entanglement entropy generated in scattering processes is systematically related to the suppression of interactions that violate global symmetries or selection rules. We conjecture that the minimization of the entanglement entropy can select the symmetry-preserving point when a new interaction opens a symmetry-violating scattering channel. We investigate the plausibility of this conjecture by considering interactions beyond the Standard Model (BSM) that violate global symmetries or the corresponding selection rules, including lepton number violation, baryon number violation, lepton flavor violation, and flavor-changing neutral currents. We show that, when symmetry-violating interactions open final-state sectors that are orthogonal to the symmetry-preserving sector, the symmetry-preserving point becomes a local minimum of the entanglement entropy. Our results support the possibility that the minimization of entanglement provides a common information-theoretic principle underlying the suppression of BSM phenomena relevant to symmetry-violating interactions.
Read the full abstract
Cryptography and Security
ACLE-MCP: Attested Capability Leases for Execution-Time Trust in Remote LLM Tool Use
Remote Model Context Protocol (MCP) services enable large language model agents to invoke external tools, but OAuth authorization alone does not ensure that a later tool call is executed by the provider-side workload that the relying party intended to trust. An endpoint may remain authorized even after execution shifts to a substituted workload, relies on stale appraisal state, reuses authority transferred from another sender, or traverses an undeclared downstream component. We call this problem the post-authorization execution trust gap. We present ACLE-MCP, an invocation-scoped architecture that couples delegated authorization, workload appraisal, and resource-side execution admission. For protected calls, ACLE-MCP issues a short-lived, sender-constrained capability lease that binds the expected workload, freshness requirement, operation, object and parameter bounds, downstream constraints, and receipt obligations. A provider-side Execution Gate consumes the lease immediately before protected tool logic begins. We implement a runnable prototype with Keycloak/OIDC validation, an MCP Python SDK server, and an optional vTPM quote-verification backend. Controlled security experiments and an agent tool-use extension show that weaker authorization or connect-time attestation modes leave distinct post-authorization attacks open, whereas full ACLE-MCP blocks all evaluated attack families while preserving all benign tasks. In the locally simulated agent extension, the complete design increases request-level pooled p95 latency on normal allowed calls by 25.7% relative to OAuth-only. These results indicate that invocation-time binding between call authority and current workload state is a practical complement to OAuth-protected remote tool use.
Read the full abstract
Computer Vision
GaLe: memory-efficient Global Approximate and Local Exact features
Embedded devices typically lack the resources of GPU-equipped machines, and existing inference methods suffer from either high computational overhead (patch-based) or accuracy loss (approximation-based). We propose GaLe, a memory-efficient technique that enables the deployment of pretrained networks on constrained devices without retraining. GaLe partitions feature maps into two components: a local exact (Le) representation that preserves fine details and a global approximate (Ga) representation that retains long-range dependencies. Unlike standard tiling, GaLe supports global operations and attention mechanisms found in hybrid CNN-transformer models. Validated on ImageNet, our method matches exact-inference performance while achieving up to 65% speedup and 90% RAM reduction on a Cortex-M33 compared to patch-based inference. We further demonstrate GaLe's versatility across classification, detection, and generation tasks, highlighting its potential as a foundation for resource-efficient architecture design.
Read the full abstract
Robotics
From Proxy Learning to Driving Decisions: A Transfer-Based Framework for Evaluating Future-Aware Autonomous Driving Planners
Future-aware representations and world models are increasingly used in proposal-based autonomous-driving planners to improve trajectory selection. However, improvements in proxy objectives or restricted subsets are often interpreted as planning gains without verifying proposal ordering, selected trajectories, full-scale utility, and critical driving components. We propose the Proxy-to-Decision Transfer (PDT) Framework, an analysis framework that evaluates when learned future information supports a reliable driving-performance improvement claim. Its Decision-Transfer Decomposition Module localizes value loss through score margins, switch-conditioned utility, and support-versus-selection regret. Its Reliability-Constrained Validation Module requires exact pairing, a minimum meaningful effect, scale-expanded confirmation, safety non-compensation, sequential comparability, and family-level robustness. On a representative future-aware planner evaluated with NAVSIM-v1, component BCE decreases from 0.705 to 0.530 while held selected PDM decreases from 0.963 to 0.961. A separate candidate improves a 512-record prefix by 0.00909, with a scene-bootstrap 95% interval of [0.000744, 0.0177], but its 2048-record and complete-support intervals include zero. A proposal-level replay further confirms the switch-utility decomposition, yet none of 432 screened configurations passes the two-half, two-seed robustness gate. PDT therefore identifies where decision transfer fails or remains indeterminate across proxy, subset, aggregate, and selection evidence.
Read the full abstract
Quantum Physics
Quantum Meta-Complexity Is All You Need: Characterizing One-Way Puzzles via Time-Bounded Kolmogorov Complexity
We initiate the time-bounded meta-complexity program for quantum cryptography. Recent work characterizes one-way puzzles, the minimal search primitive of quantum cryptography without one-way functions, by the average-case hardness of approximating the plain, uncomputable Kolmogorov complexity over quantumly samplable distributions; the classical program of Liu and Pass, by contrast, lives at polynomial time bounds. We define a probabilistic time-bounded quantum program complexity pKq^t for classical strings and prove two unconditional theorems. First, a quantum coding theorem: any string output by a quantum polynomial-time sampler with probability delta admits a description of the information-theoretically optimal length log(1/delta) plus logarithmic terms, decodable by a quantum machine in time O(sqrt(1/delta)) times a polynomial, via amplitude amplification over the coherently executed sampler. Second, an exact characterization at subexponential time: one-way puzzles exist if and only if the gap problem for pKq at time bound 2^(n/2) poly(n) is weakly quantum-average-hard, refining the plain-complexity characterizations. We then isolate the polynomial-time coding theorem as the single load-bearing open conjecture of the program, prove that it implies the full polynomial-time characterization, analyze why the classical derandomization proof resists quantization, and formulate a relativized barrier conjecture delimiting string-valued meta-complexity at one-way puzzles. Conjectures are labeled as such throughout.
Read the full abstract
Differential Geometry
The Bonnet-Myers theorem on Finsler manifolds with integral weighted Ricci curvature bounds
In this paper, we derive some new relative volume comparison theorems and Bishop-Gromov volume comparisons on Finsler metric measure manifolds, all of which are controlled by the integral weighted Ricci curvature. In particular, we establish a Bishop-Gromov volume comparison theorem for nonconcentric balls. Based on these, we prove a theorem of Bonnet-Myers type on Finsler metric measure manifolds with integral weighted Ricci curvature bounds.
Read the full abstract
Computation and Language
DKL: Decoupled Knowledge Learning for Instruction-Tuned Language Models
RAG has become the de facto method for incorporating new, corpus-specific knowledge into an instruction following LLM (Instruct LLM). Although RAG-based prompting improves factual grounding, it fails when retrieval is incorrect or incomplete, leading to hallucinations. Finetuning methods such as RAFT and PA-RAG enhance RAG by injecting new knowledge into the model's parameters, but require generating a massive amount of synthetic QA that covers the entire corpus. Extended Pre-Training (EPT) on the text corpus avoids the need for comprehensive synthetic data generation but compromises an Instruct LLM's instruction-following capabilities, necessitating instruction fine-tuning (IFT) after pre-training. However, IFT is costly and may be infeasible due to the unavailability of an instruction-tuning corpus. In this work, we propose DKL-Decoupled Knowledge Learning for Instruction-Tuned Language Models. Instead of doing EPT on the Instruct LLM, DKL performs EPT on its corresponding base LLM to infuse new knowledge. These knowledge infused weights are then merged with the Instruct LLM, imparting new knowledge without affecting their instruction-following capabilities. DKL is a lightweight method that avoids expensive instruction fine-tuning and relies on model merging to infuse the new knowledge into the Instruct LLM without destroying its instruction following capabilities. Empirical results show that DKL improves RAG accuracy from 54.17 to 79.26 on retrieval failure cases, while outperforming prior approaches with substantially less training data.
Read the full abstract
Machine Learning
H3DNAS: Hardware-Aware ONNX-Native 3D Point Cloud Model Compression
Deploying 3D point cloud models on edge hardware such as the NVIDIA Jetson Orin Nano is severely constrained by compute and memory budgets. Existing compression methods require access to the model's original source code, rendering them inapplicable to the Open Neural Network Exchange (ONNX) binaries commonly distributed by vendors and model repositories. We present H3DNAS, a hardware-aware model compression framework that operates directly on ONNX computational graphs without requiring original source code, architecture class definition, or gradient access during search. H3DNAS makes three contributions: (1) a Channel Dependency Graph (CDG) that classifies ONNX operators into four constraint classes and formally establishes that the free parameter fraction $\rho_f$ is topological invariant, a provable compression ceiling computable in $\mathcal{O}(|V|+|E|)$; (2) a Two-Stage Hierarchical Search that prunes candidate architectures by $L_1$-importance channel selection, ranks them by output fidelity as a zero-shot label-free proxy, and applies GhostConv structural mutation to Pareto-optimal candidates; and (3) the first source-code-free compression pipeline for 3D point cloud models, operating entirely via ONNX graph surgery with no original architecture definition required. On ModelNet40, H3DNAS reduces the number of parameters in PointNet, PointNet++, and PointMLP by $65.5\%$, $43.2\%$, and $49.1\%$, respectively, while achieving $1.99\times$, $1.29\times$, and $1.67\times$ inference speedups with negligible loss in accuracy. The source code is publicly available\footnote{https://github.com/ClarityLab-Org/h3dnas}.
Read the full abstract
Computer Vision
Genesis: A Generative Engine for Hierarchical Satellite Image Synthesis
Earth observation is fundamentally multi-scale; geospatial tasks span varied resolutions, and satellite imagery is organized into cascading tile pyramids that nest fine detail within wide coverage. Current generative models of satellite imagery, however, operate along a single axis: they either zoom to enhance a single tile's resolution or pan to extend imagery at a fixed scale. As a result, no existing method produces a complete pyramid that stays consistent across both scale and space, where a high-zoom tile must agree with the coarse context it refines and with the neighbors it meets. Motivated by this gap, we introduce a new task, multi-scale tile completion: given a sparse set of seed tiles at arbitrary zoom levels and positions, synthesize a complete, uniform quadtree that is globally consistent across both scale and space. We approach this task with Genesis, a generative engine that brings both axes together by composing two specialized operators over the quadtree: a vertical super-resolution model and a horizontal mask-based outpainting model, producing pyramids that are consistent across zoom levels and seamless across neighboring tiles. Each operator achieves state-of-the-art results on its subtask, and the engine propagates sparse seeds into seamless, multi-resolution maps from any initial configuration. To evaluate the task and benchmark Genesis, we introduce dense500, a fully observed multi-scale pyramid dataset spanning diverse geographic regions, together with a suite of pyramid-level metrics. Code, models, and our dataset are available at https://github.com/mvrl/genesis.
Read the full abstract
Probability
LDP for Tensor Forms
In this paper, we study the large deviation principle (LDP) for a tensor-weighted functional of i.i.d. random variables, when the sequence of tensors converges under a variant of the "bad" cut norm. Using the LDP, we analyze a Gibbs measure with a tensor-valued Hamiltonian, and characterize the optimizers of the limiting variational problem in terms of a functional fixed point equation. As applications, we focus on several concrete examples, which include monochromatic subgraph counts in sparse random graphs, Erdős-Rényi hypergraphs, and a generalized Potts statistic of order $v\ge 2$. Studying the optimization problem, we give sufficient conditions for uniqueness of the optimizer, as well as for existence of constant optimizers (replica symmetry). Our results demonstrate universal weak laws for a large class of tensor Gibbs models with approximately regular tensors.
Read the full abstract
High Energy Astrophysical Phenomena
Multiwavelength View of Black Hole X-ray Binary Swift J151857.0-572147
We investigate multiwavelength timing and spectral properties of the newly discovered black hole X-ray binary Swift J151857.0-572147 during its 2024 outburst using observations from AstroSat, NuSTAR, NICER, Swift, and MeerKAT. The AstroSat/LAXPC20 power density spectrum reveals a $\sim$8 Hz low-frequency QPO with a quality factor of $\sim2.8$. The broadband X-ray spectrum of the source exhibits prominent reflection features, which are modeled to constrain the geometry and physical properties of the accretion disk. The spectral analysis indicates that the source was in the soft/intermediate spectral state, characterized by a steep power law with a photon index of $\Gamma \approx 2.6$ and a relatively high disk temperature of $\sim0.9$ keV. The reflection modeling suggests a disk inclination of $23-31^{\circ}$ and a highly ionized accretion disk with $\log\xi \sim 3.3$-4.0. The MeerKAT monitoring revealed a strong radio variability that closely tracked the hard X-ray evolution, including a bright radio flare during the hard-to-soft state transition, supporting close coupling between the accretion flow and jet activity. Using the quasi-simultaneous AstroSat, Swift/XRT and MeerKAT observations, we investigated the radio/X-ray correlation and found that Swift J151857.0-572147 follows the established $L_{\rm X}$-$L_{\rm R}$ relation for Galactic black hole X-ray binaries. Its location in the $L_{\rm X}$-$L_{\rm R}$ plane is consistent with that of Galactic black hole X-ray binaries, further supporting its black hole nature.