Papers

We study deterministic bounded-noise state reconstruction for nonlinear systems when the measurement schedule drives the sampled-output map through a rank-deficient configuration. Local reconstruction is governed by the leading pairwise separation order and its corresponding higher-order modulus, which together determine the leading minimax noise-to-state law. For corank-one cubic singular schedules, including overdetermined sampled maps, we derive the exact physical $\ell_\infty$ cubic modulus through quotient geometry in $\mathbb{R}^q/\operatorname{Im}DF$ and show how the required finite jet is computed from trajectory sensitivities. We then prove a uniform signed timing crossover: a transversal timing perturbation has transition scale $\varepsilon=O(\delta^{2/3})$, linking the regular, cubic-singular, and fold regimes. A damped pendulum provides closed-form crossover coefficients and exact nonlinear minimax validation, while a four-state downward cart–pole verifies the rectangular theory. The results show that measurement timing can change not only reconstruction conditioning but the exponent governing bounded-noise state uncertainty.

Quantum decoherence does not permit the operation of (new) miniaturised devices or protocols that can otherwise run unassisted in classical settings, without explicit user agent intervention in the quantum realm. Classical driving signals are inadequate to model a truly quantum device built entirely in the quantum realm. Synchronisation endures miniaturisation without sacrificing quantum coherence, unlike thermalisation. We quantise the Kuramoto model and the Van der Pol oscillator and couple it with qubit(s) to show a robust limit cycle even in typical quantum thermodynamical settings. Novel effects such as Entanglement State Revivals, Zeno locking and thermal current interference are observed. We propose toy model devices that can be implemented harnessing the Synchronisation of a qubit (or an arbitrary collection of qubits) with a nonlinear phase oscillator. The framework theory used here, Spectral Mechanics, is a fresh look at the foundations of conventional quantum mechanics and is extensible to completely new forms of interactions and models in experimental settings.

In this study, an artificial intelligence-assisted evaluation of the refractive indices of different materials is performed. The evaluation of the refractive index is necessary to design photonic devices for various classical and quantum photonic applications, and to theoretically analyse experimental results. The refractive index is one of the crucial parameters that affect the interaction of light with matter. Although the refractive index can be measured experimentally, its estimation via computational means provide the advantage of not requiring costly instrumentation, trained researchers and could avoid potential laboratory-based uncertainties. Therefore, in this study, artificial intelligence is implemented to estimate the optical constants of two types of materials, i.e., noble metals, and semiconductors. To this end, an artificial neural network is used to estimate the real and the imaginary parts of the refractive index of a specific material for various wavelengths of light. Then, the knowledge of the trained neural network is employed to initiate and train new neural networks to estimate the refractive indices of other materials. For the training of a neural network, various methods can be deployed. However, the application of the training can be a time-consuming process. A transfer learning approach is here implemented to provide a faster training solution for the problem under study. The implementation of transfer learning allows training the neural networks for a new material with higher speed or accuracy when compared to non-knowledge transfer-based training. Our results show the efficiency of the proposed technique in accurately evaluating the real and imaginary parts of the refractive index of different materials.

Many bug reports lack the necessary information required to reproduce issues. In early studies, large language models have been employed to generate follow-up questions to enhance incomplete reports. However, they ask a fixed number of questions regardless of whether the report is incomplete. We propose BRQGen, an iterative framework that identifies missing or ambiguous information in bug reports and generates targeted follow-up questions, only if necessary. Reports with no identified gaps are classified as sufficient. When the author responds, BRQGen evaluates whether the answers provide the requested information and generates further questions for any exposed gaps. We curated a novel dataset of 429 incomplete bug reports from Home Assistant and Brave Browser, each containing at least one maintainer follow-up question. Each question is mapped to the author's answer and assigned a label indicating whether the requested information could have been provided directly through the issue template. Earlier maintainer requests were associated with author responses, whereas delayed requests were associated with no response. Among responding authors, requests within one hour received median replies after 0.52 hours in Home Assistant and 1.45 hours in Brave, versus 9.76 and 13.49 hours after four hours. BRQGen covered 35.4% (Qwen 3.8 Flash) to 51.8% (DeepSeek V4.1 Flash) of the 1,073 ground-truth follow-up questions. DeepSeek V4.1 Flash managed to ask 45.7% of all detailed questions and 63.9% of all generic questions. Gemini 3 Flash Preview reached the highest macro-F1 for completeness assessment of 49.3%, but with a post-hoc analysis by getting only the gaps with high severities, Qwen 3.8 Flash reaches 73.6% macro-F1 score. Completeness assessment improved when at least one identified gap had high severity, and nearly all ground-truth questions were captured within the first two iterations.

Forensic analysis of low-resolution Word thumbnails presents a significant challenge because these artifacts may constitute the only remaining visual evidence of deleted documents. Standard Optical Character Recognition (OCR) tools perform poorly on such severely degraded images, creating a critical bottleneck in forensic evidence recovery. Although deep learning architectures, such as U-Net, can improve visual image quality, models optimized solely using pixel-based loss functions tend to favor visual smoothness over the structural fidelity required for accurate text recognition. To address this limitation, we propose a text-aware restoration framework designed for forensic triage. The proposed approach employs a U-Net architecture guided by a hybrid optimization strategy that integrates pixel-level fidelity with machine-readable text constraints. Specifically, an OCR-based Normalized Levenshtein Distance (NLD) metric is used during model selection to prioritize the recovery of an accurate character topology. Experimental results demonstrate that the proposed method significantly reduces the NLD, enabling reliable recovery of machine-readable text from severely degraded thumbnails.

Charting the evolution of star-formation, stellar mass and dust build-up over cosmic time relies on the accurate characterisation of the stellar population properties of high redshift galaxies. We present a catalogue of spectroscopically derived physical properties for ~12,000 galaxies at z>2, with archival JWST/NIRSpec PRISM spectroscopy and JWST/NIRCam imaging from the DAWN JWST Archive. We present a custom version of the SED fitting code BAGPIPES, used to perform joint spectrophotometric modelling of PRISM spectra and broadband NIRCam photometry. In doing so, we publicly release a science-ready catalogue of the physical properties of galaxies beyond cosmic noon. We report key physical properties, such as stellar masses, star-formation rates (SFR) and timescales, and V-band dust attenuation (Av). We further probe the recent star-formation histories (SFH) of galaxies via their SFRs on 3 Myr, 10 Myr, and 100 Myr timescales, capturing both rising and falling recent SFHs. Through dust-corrected ugi colour-colour diagrams of massive galaxies, we show how quiescent and dusty galaxies populate this colour space at z>2, with galaxies experiencing a recent downturn in star-formation occupying distinct parameter space. We also study the dustiest galaxies in our sample, showing that they are preferentially more massive and strongly star-forming relative to the parent sample, residing on the high-mass end of the star-forming main-sequence. The slope of the attenuation curve has a strong dependence on Av, varying from steep to shallow as the dust attenuation increases and the specific star-formation rate decreases. Our findings show that the unparalleled number statistics of this catalogue of spectroscopically derived stellar population properties uniquely enables the study star-formation, quenching and dust properties of galaxies from cosmic noon to cosmic dawn.

We present Gen Through the Lens, a camera-controllable image-to-video generation framework that synthesizes videos from a given reference image while enabling realistic and physically consistent camera operations. Unlike previous models that treat camera motion as abstract conditioning or rely on a single coordinate system, our approach explicitly decomposes and fuses control signals across three coordinate spaces: the camera coordinate (modeled by intrinsic parameters), the world coordinate (represented by spatial geometry hints), and the object coordinate (described by 3D bounding boxes of target subjects). Through cross-coordinate conditioning fusion, these heterogeneous cues are unified to guide the generative process, enabling precise control of camera motion, focus, and framing while preserving temporal and spatial coherence. Extensive experiments demonstrate that our method produces controllable and visually consistent results across diverse camera operations, outperforming existing approaches in both realism and editability.

Large language model serving is increasingly limited not only by model size and GPU speed, but also by scheduling during autoregressive decoding, where continuous batching interleaves requests at token granularity as sequences arrive and finish. We study whether lightweight scheduling policies can improve latency-throughput tradeoffs without changing model architectures, kernels, or deploying learned schedulers. We implement a Python/PyTorch continuous-batching testbench with real Hugging Face execution and evaluate eight policies across 1,152 GPU-model-scheduler-workload runs using ShareGPT-derived traffic, six Qwen, Gemma, and Mistral models, and Modal L40S, A100, and H100 GPUs. Scheduler choice changes throughput, TTFT, completion latency, and tail behavior under fixed model, GPU, workload, batch size, and prefill budget. Aging shortest-prompt-first is the strongest practical baseline in many L40S and mixed-workload settings, while oracle remaining-length scheduling shows that job-size knowledge alone is insufficient because batching and hardware effects also shape performance. The custom dataset has been uploaded to Hugging Face.

Light reflected from a deforming solid depends on boundary motion and strain-induced modulation of the dielectric response. For a propagating acoustic field, scattering from different depths interferes coherently, so the measured optical amplitude and phase depend jointly on the strain profile, dielectric constant, and photoelastic constant. Optical tomography often makes such reconstructions tractable through independent projections. To probe mechanical strain, however, the optical measurement also depends on the material properties that mediate its observation, requiring those properties to be known or assumptions about the field. Here we show that broadband picosecond ultrasonic echoes in an opaque solid can supply the missing information. Fourier transformation of the measured complex relative reflectance change obtained with monochromatic light at normal incidence gives access to many acoustic frequency components. We recover the complex dielectric and photoelastic constants together with the nanoscale depth-dependent strain profile by formulating the reconstruction as a multiparameter nonlinear eigenvalue problem. We demonstrate the method using experimental echoes from a chromium film. The recovered refractive index agrees with ellipsometry, whereas the recovered strain profile closely follows a two-temperature-model prediction including electron and thermal diffusion and frequency-dependent ultrasonic attenuation. Synthetic echoes test robustness to added noise. Broadband response thus replaces additional optical projections, making the measurement self-calibrating and enabling simultaneous recovery of an unknown physical field and the material properties that mediate its optical observation.

Qwen3-8B reviewing Llama-3.1-8B in math adds 33.2 points, yet Qwen3-8B alone is 6.4 points more accurate than the result. Reviewing another model's answer is usually judged by its gain over the generator. We show that this confuses verification, which adds accuracy beyond either model alone, with capability transfer, where a stronger reviewer supplies answers it could have given on its own. Measured against the stronger model alone, across 48 pairs of 12 open-weight models (3B to 72B) in math, knowledge, code and reasoning, review beats it in only 9 of 96 review directions, and does worse the larger the capability gap. The reason is anchoring: the reviewer mostly follows the generator, discarding its own correct answers. Review costs 2.6 to 2.8 times as much as the stronger model alone, while a selector using the same two calls is 2.7 to 4.2 points more accurate than review and cheaper. Studies of verification should report the stronger model alone, not only the gain over the generator.

In sequential learning, a group of agents sequentially receives a noisy private signal and observes the predictions of prior agents to estimate a hidden ground truth. While extensively studied for binary signals, we formalize and analyze sequential learning when the hidden ground truth is an ordering of $n$ items, and both private signals and inferences involve pairwise comparisons. A key focus in sequential learning is herding, where agents disregard their private signals to follow the crowd. We show that, while local herding in our model may lead agents to make incorrect pairwise inferences, the global hidden order can still be approximated. Specifically, by aggregating all pairwise comparison outputs, the highest-ranked items can be retrieved. We investigate top-$k$ retrieval across three scenarios: (1) passive aggregation, where information is gathered from the social learning process without intervention; (2) strategic selection, where agents choose their pairwise comparisons dynamically; and (3) aggregation with stubborn nodes, where a very small set of stubborn agents, who base their inferences solely on their private signals, join the mechanism. With query complexity as the number of agents needed for a high-probability estimation, we propose top-$k$ retrieval algorithms with query complexities sublinear in $O(n^2)$, the total number of pairs.

Context. White-light observations of the solar corona contain contributions from both the electron-scattered K-corona and the dust-scattered F-corona. Because the diffuse F-corona dominates the measured intensity at altitudes of approximately 4 $R_{\odot}$ and above, background subtraction techniques are routinely employed to isolate the electron corona in coronagraph observations. Aims. Since no direct observational "ground truth" for the K-corona exists, the amount of coronal signal that is inadvertently removed by background subtraction methods remains unknown. We aim to quantify, for the first time, the impact of background subtraction on the inferred large-scale structure and brightness of the solar corona. Methods. We use a 32-day simulation based on a novel near-real-time, data-assimilative, time-evolving magnetohydrodynamic model to generate synthetic white-light observables. Because these synthetic observations are free from dust and instrumental stray-light contributions, they provide direct access to the "true" K-corona and enable a quantitative assessment of the effects of background subtraction. Results. We find that background subtraction generally preserves the large-scale morphology of the electron corona, including the locations and shapes of major coronal structures. However, the procedure systematically modifies the relative brightness of different features and removes a non-negligible fraction of the K-coronal signal. Conclusions. Although standard background subtraction techniques successfully recover the overall appearance of the electron corona, they can alter its quantitative photometric properties. These findings should be considered when interpreting coronal brightness measurements and deriving physical properties from processed white-light observations.

Large language model (LLM) code-repair systems fix faulty programs by repeatedly generating and testing revisions. Each revision is costly, making the choice of which repair to pursue important for efficiency. Existing search policies rely on test pass rates and search statistics, ignoring the details of test failures and proposed edits. We introduce CodeTriage, a learned policy that prioritizes edits before their revised programs are generated and tested. Its 0.6B-parameter model reads the test output, failure diagnosis, and proposed edit, then predicts a distribution over the number of repair steps needed to reach a fix through that edit. These predictions prioritize all pending edits across programs before generating and testing the revised code. On held-out repair tasks, CodeTriage finds fixes with 21% fewer program revisions and 12% fewer output tokens than the strongest baseline on each metric. Without retraining, it achieves the highest average success across search budgets for each of 11 unseen repair LLMs. Ablations show that both the failure context and edit text, as well as comparisons across programs, improve search. We also release a corpus of 12.2 million program versions from four LLMs, organized into repair trees with proposed edits and test results.

The protection of quantum information in systems leveraging more than two basis states (qudits) is an ongoing challenge. A tool for transforming the plethora of qubit stabilizer codes into qudit stabilizer codes is the local-dimension-invariant (LDI) extension to the stabilizer formalism. A drawback of the LDI method is that generally in order to ensure the distance of the resultant code is at least preserved the newly selected local-dimension will need to be any prime bigger than a large cutoff value. In this work we consider the repercussions of this large cutoff value, then by focusing on quantum low-density parity-check (qLDPC) codes we are able to bring down this cutoff value. Notably, we are able to reduce it in some cases to the initial local-dimension and for arbitrary CSS qubit-derived codes to singly exponential in $d$ with no dependence on the number of logicals $k$. This reduces the size of the caveat in LDI forms in their distance preservation condition.

We study eigenpair estimation for the tensorized, density-weighted Laplace–Beltrami operator from $n$ independent data points sampled from a multi-manifold mixture distribution where manifolds may intersect. Let $d$ denote the largest intrinsic dimension of the manifold components and $s\geq2$ the smoothness of the sampling densities. For simple, nontrivial eigenpairs, we establish the minimax rate $n^{-s/(d+2s)}$ for approximating the eigenvalue and eigenfunction of the tensorized Laplace–Beltrami operator, and show that a continuum plug-in estimator based on kernel density estimation attains the same rate when the supporting manifolds are known. We further analyze a manifold-agnostic graph Laplacian estimator and prove that its eigenpairs converge to those of the tensorized Laplace–Beltrami operator at rate $n^{-1/(2.5d-1)}$, up to logarithmic factors, improving upon previously known spectral convergence rates for multi-manifold learning. Numerical experiments on synthetic multi-manifold data corroborate the rates guaranteed by our theorems. These results characterize the information-theoretic difficulty of multi-manifold spectral estimation and quantify the additional cost of learning without oracle component information.

Predictive video representations do not by themselves specify how a multimodal system should handle missing sensors. We transfer frozen V-JEPA2 features to a set of heterogeneous sensors comprised by RGB video, depth video, and radar time series. A shared fusion model uses modality queries, fixed sensor slots, and explicit availability to classify breathing activity and pose from all seven non-empty sensor subsets. The contribution is a controlled evaluation of subset supervision, pooling capacity, and representation transfer on 39 OMuSense-23 subjects with five subject-group folds and three seeds. With matched downstream parameter counts, the final model reaches recording-level activity macro-F1 of $75.23%\pm0.60$ averaged across seven subsets. Supervising all subsets in each batch improves the matched query-pooling control by 1.923 percentage points (conditional 95% subject-bootstrap interval [0.647, 3.209]), while query pooling shows no clear advantage over matched mean pooling under balanced supervision. Radar probes support transfer beyond pooled wavelet pixels, 64-token inference approximately preserves mean activity performance, and the tested Consensus-JEPA loss adds no measured benefit against its exact control.

In this paper we prove that the classical thermoelastic Timoshenko and Bresse systems with thermal coupling acting only on the bending moment are not exponentially stable under full homogeneous Dirichlet boundary conditions if the equal wave speed condition is not satisfied. Combined with their well-known converses, our results provide complete characterizations of exponential stability in terms of the coefficients of each system. To the best of our knowledge, these characterizations were previously known only under mixed boundary conditions. The proof is based on the introduction of a boundary correction term in the standard trigonometric ansatz so that the full homogeneous Dirichlet conditions are satisfied.

We study the $p$-Bures-Wasserstein distance $D_p(A,B)=(\mathrm{tr}\,A^p+\mathrm{tr}\,B^p-2\|A^{1/2}B^{1/2}\|_p^p)^{1/2}$ on positive semidefinite matrices; it equals the Bures-Wasserstein distance for $p=1$, and the Hilbert-Schmidt distance for $p=2$. We prove that $D_p$ is a metric for every $1\le p\le2$, by proving a stronger 2014 conjecture of Lin and van den Driessche on $3\times3$ block matrices. Our main tool is a partition inequality for Schatten norms, proved by complex interpolation. We also characterize Fréchet means for $D_p$ by a fixed-point equation and prove geometric convergence of the associated iteration.

Reinforcement learning with verifiable rewards spends much of its rollout budget on states that yield no learning signal. We recast this as a question about the sampling distribution itself: under a fixed rollout budget, which distribution over states minimizes the variance of the policy-gradient estimator? The answer admits a closed-form Neyman allocation, $\mu_{\mathrm{var}}^\pi(s) \propto d^\pi(s)\,\sigma_\pi(s)$, which connects naturally to GRPO's group-relative advantage and induces an emergent reverse curriculum as the policy improves, with no externally imposed schedule. Building on this principle, we propose VSR (Variance-driven State Resampling), a plug-in modification to GRPO using a prefix replay buffer and a zero-cost variance proxy. On mathematical reasoning benchmarks and long-horizon agentic tasks (Sokoban, BabyAI), VSR outperforms GRPO and GiGPO on most benchmarks with fewer generated tokens per update.

We prove the geometric halves of Hilbert's Nullstellensatz and of the real Nullstellensatz for submodules of finitely generated modules and for coherent sheaves, complementing our earlier result that semiprime (resp. real) submodules are intersections of prime (resp. real prime) submodules. First, we pass from prime to maximal submodules: over a Jacobson ring, a prime submodule $P$ of $M$ is an intersection of maximal submodules whenever $M/P$ is a torsionless $R/(P:M)$-module, in particular whenever $M/P$ is finitely generated (the latter case is due to Maani Shirazi and Sharif). Consequently, if $R$ is a finitely generated algebra over an algebraically closed field and $M$ is finitely generated, the smallest semiprime submodule containing a submodule $N$ of $M$ consists of all $f\in M$ such that, for every point $a$, the value $f(a)$ (i.e. the image of $f$ in the fibre $M/\mathfrak{m}_aM$) lies in $N(a)$ (i.e. the image of $N$ in $M/\mathfrak{m}_aM$). We prove the real analogues over real Jacobson rings; for finitely generated algebras over a real closed field they are determined by the real points. The module classifications and directional vanishing formulas yield both Nullstellensätze for coherent sheaves on schemes of finite type over a field. We also prove compatibility of semiprimeness, reality and both radicals with arbitrary localization, and the resulting stalk criteria.

Workflow scheduling in user device-centric edge-cloud computing has emerged as a challenging problem due to the need to jointly optimize execution performance, energy consumption, and monetary cost. Modern workflows consist of heterogeneous tasks, where compute-intensive components are well-suited for Infrastructure-as-a-Service (IaaS) resources, while fine-grained, event-driven tasks are more efficiently executed using the Function-as-a-Service (FaaS) model. We formulate this scheduling problem for application workflows over a hybrid User Device (UD)-IaaS-FaaS computing continuum as a multi-objective Mixed-Integer Linear Program (MILP). The formulation models the workflow execution mechanism by incorporating cumulative capacity constraints for IaaS resources and elastic resource provisioning for FaaS. To overcome the intractability of MILP, we propose a learning-based framework that combines graph neural networks, surrogate performance modeling, and a multi-sample REINFORCE approach. The proposed approach is workflow-aware, leveraging task pairing to improve data locality and structured task placement decisions. Agent training is organized into three stages: small-scale learning of UD-IaaS trade-offs, intermediate locality aware co-location, and large-scale handling of parallelism and aggregation bottlenecks in workflows. The proposed framework is validated through real-world experiments using the Pegasus Montage workflow on a custom UD-IaaS-FaaS testbed comprising a user device, multiple IaaS nodes, and serverless FaaS instances. Extensive experimental results demonstrate the improved trade-offs across makespan, energy consumption, and cost. Compared with the genetic algorithm and greedy scheduling approaches, the proposed approach reduces makespan by 10-17.5%, energy consumption by 3-9%, and cost by 4-10%, while remaining within 3.5-5.6% of the MILP optimum.

Computational fluid dynamics (CFD) is computationally expensive for iterative building design. Deep learning surrogates offer faster predictions, but generalization across varying room dimensions remains challenging. This study compares two geometry-conditioned neural network surrogate models for rapid indoor airflow and temperature field prediction under varying room sizes. During the architectural design stage, this method is relevant for real-time assessment of whether conditioned supply air will effectively provide fresh and cool air throughout rooms with varying heights and lengths. The canvas-based neural network (CaNN) explicitly encodes room size by preserving different occupied regions within a fixed canvas, whereas the coordinate-based neural network (CoNN) uses normalized coordinate channels. Both employ U-Net autoencoders to separately predict airflow and temperature fields. Evaluation uses simplified rooms with varying heights and lengths as a controlled benchmark. The results show that CaNN achieves higher agreement with CFD results than CoNN, demonstrating its potential to support performance-driven building design.

False retry permission (FRP) occurs when an agent retries a timed-out tool call without support from the visible contract. RetryBench separates this error from realized recovery. Successful protected recovery does not imply reliable rejection of unsupported retries. In a 720-response key control, valid keys usually enabled safe progress, but offered changed keys elicited unsupported retries in Sol (1/80), Terra (55/80), and Luna (60/80). A later 240-response control removed the candidate and required models to supply a key. Luna still made 8/80 unsupported dispatches; Sol and Terra made none in 80 responses each. Accepting a displayed bad key therefore cannot explain all observed errors, although changed output requirements and conditions preclude a pure display-effect comparison. Removing the weak-contract label increased retries in Sol and Terra's coded-action conditions in a six-endpoint study. A separate 1,536-response canonical-action block also found a positive Terra contrast (115/128 versus 124/128 retries), concentrated in two task/order conditions; the other five contrasts remained unresolved after multiplicity correction. An explicit-contract cue control elicited deferral in all 762 returned responses, leaving recovery unresolved. These results support reporting permission and recovery together across declared interface variants. The evidence covers four purposive tasks, fixed GPT endpoints, and one recovery decision; it identifies neither internal mechanisms nor episode-level safety.

Text wrapped in a trigger can be made behaviorally invisible to a language model: omitted from summaries and quotations, ignored in question answering, and left unflagged by downstream filtering. We study this selective blindness as a dual-use information-flow primitive, a supply-chain backdoor, and a white-box auditing problem. We first implement it exactly as a serving-time span veto: blocking information flow out of a marked span makes every outside activation bit-identical under same-length substitution of the span's content, a noninterference property we verify across ten models, seven mixer families, and 0.5B-120B parameters. The veto serves as a deployable quarantine, an exact reference for calibrating leakage, and a teacher. Distilled into weights, it yields a backdoor: checkpoints trained on six model families from 4B to 120B ignore marked context at little benchmark cost, and their outputs show no measurable semantic trace of the suppressed content above the veto reference on our held-out suite; an in-domain variant suppresses planted vulnerabilities during AI code review. Detection is therefore a white-box problem. Simpler backdoors remain internally transparent – linear probes recover suppressed content at 0.57-0.79 accuracy versus 0.025 chance – but hardening the attack until probes reach chance collapses attention from outside the span into it below the range of base models. Across the attacks we evaluate, probe visibility trades off against routing visibility; we do not establish that this tradeoff is fundamental.

A stored fact may stop being valid when a user changes a preference or an event replaces an earlier one. We study whether a language model can learn these changes in its parameters and answer correctly from only a question and the current time. Engram Injection stores facts in LoRA adapters and supplies the clock through FiLM conditioning on a frozen Qwen2.5-3B-Instruct model. We compare it with additive clock conditioning, prompting with the full dated history, and retrieving four dated records. We train six matched adapters across three seeds after fixing a sampler that could admit test-age examples into training. The training audit reports no such examples among 127,760 evaluated training inputs. On 62 pairs of questions asked before and after an update, at an elapsed age of seven days excluded from training, both adapter methods answer zero pairs correctly under the primary blinded assistant review with order code 1 held fixed. Full-history prompting answers 59/62 pairs correctly, and retrieval answers 62/62. Separate experiments show that adapters can store facts and that removable memory units can support local deletion. Combining fact selection with time-based version selection remains unsuccessful. These results separate learning a fact from learning when to use it.

Given $1 < p < \infty$ and a function $m \colon \mathbf{N}_0 \to \mathbf{C}$, we provide a characterization of radial Schur multipliers $M(x,y) = m(d(x,y))$ on the Schatten $p$-class of a tree $(X,d)$ equipped with its graph metric. That is, the operator norm of the radial Schur multiplier $S_M$ is controlled by the norm of the Fourier multiplier $T_m$ on the torus: \[ \big\| S_M \colon S_p(X) \to S_p(X) \big\|_{\mathrm{cb}} \lesssim \frac{p^4}{(p-1)^2} \, \big\| T_m \colon L_p(\mathbf{T}) \to L_p(\mathbf{T}) \big\|_{\mathrm{cb}}, \] with both quantities being comparable when the tree $X$ has infinite diameter. This is a rather unexpected inequality, which necessarily fails at the endpoints $p=1,\infty$ due to the work of Haagerup, Steenstrup and Szwarc. As an application, we construct the first examples of completely $S_p$-bounded Herz-Schur multipliers in the free group that are unbounded as Fourier multipliers. This result disproves Fourier-Schur transference—a longstanding open problem since the work of Neuwirth and Ricard in 2011. In line with this, our main result also implies the surprising fact that free balls Schur multipliers are uniformly $S_p$-bounded, in sharp contrast to their Fourier behavior. In addition, we disprove Fourier-Schur transference over the algebraic groups $\mathrm{PGL}_2(\mathbf{Q}_q)$ and $\mathrm{PSL}_2(\mathbf{R})$ with $\mathrm{K}$-biinvariant counterexamples, the latter involving additional geometric and probabilistic techniques.

Local out-of-time-order correlators (OTOCs) measure the growth of initially small commutators. For bounded local operators, this growth ends when the local commutator becomes of order one, but in an extended system the region over which it is large can continue to expand. We study this spatial proliferation using three related observables: the local squared commutator $C_r(t)$, its spatial sum $\bar C(t) = \sum_r C_r(t)$, and the symmetrically regularised connected dOTOC $D(t)$ of two extensive observables. We show that Lieb–Robinson locality gives $\bar C(t) \lesssim t^d$. We then prove that, in a Gibbs state with uniform exponential clustering, the full dOTOC obeys the same bound, $D(t) \lesssim t^d$, improving our earlier suboptimal $t^{3d}$ bound. We illustrate these results in two quantum many-body systems: a Brownian coupled-cluster model and a finite-temperature SYK chain. In both cases, an exact disorder symmetry relates $D(t)$ to $\bar C(t)$ and allows us to follow the same extensive observable from the pre-scrambling Lyapunov regime into the post-scrambling growth associated with an expanding saturated region. This late-time power-law growth is the concrete realisation of the weak quantum chaos introduced in our earlier work.

Zero-added-loss multiplexing (ZALM)~[Phys. Rev. Appl. 19, 054029 (2023)] is a midpoint-source scheme for entanglement distribution. Its quantum transmitter (QTX) employs two Sagnac sources of polarization-entangled pairs plus partial Bell-state measurements (BSMs) on their idler outputs to herald entanglement in their signal outputs, thus avoiding the switching losses suffered by other multiplexing approaches. When its Sagnac sources' spontaneous parametric downconverters (SPDCs) utilize $\chi^{(2)}$ crystals with phase-matched spectral islands, ZALM avoids the tradeoff between heralding efficiency and output-state fidelity incurred when conventional-crystal SPDCs' outputs are channelized by dense wavelength-division multiplexing filters. This paper reports an end-to-end fidelity and rate analysis for islands-based ZALM that includes: SPDC multipair emissions of all orders and partial-BSM losses in the ZALM QTX; loss in propagation from the ZALM QTX to Alice and Bob's quantum receivers (QRXs); loss and mode mismatch in the mode converters employed at Alice and Bob's QRXs; and loss in the partial-BSM loading of Alice and Bob's emissive quantum memories.

This paper proposes a fully autonomous execution model for multi-GPU applications that completely excludes the involvement of the CPU beyond the initial kernel launch. In a typical multi-GPU application, the host serves as the orchestrator of execution by directly launching kernels, issuing communication calls, and acting as a synchronizer for devices. We argue that this orchestration, or control flow path, causes undue overhead and can be delegated entirely to devices to improve performance in applications that require communication among peers. For the proposed CPU-free execution model, we leverage existing techniques such as persistent kernels, thread block specialization, device-side barriers, and device-initiated communication routines to write fully autonomous multi-GPU code and achieve significantly reduced communication overheads. We demonstrate our proposed model on two broadly used iterative solvers, 2D/3D Jacobi stencil and Conjugate Gradient(CG). Compared to the CPU-controlled baselines, the CPU-free model can improve 3D stencil communication latency by 58.8% and provide a 1.63x speedup for CG on 8 NVIDIA A100 GPUs. The project code is available at https://github.com/ParCoreLab/CPU-Free-model.

We study the approximate channel simulation problem in the normal approximation regime. Our starting point is the fact established by Cao et al.~\cite{Cao_channel_simulation}, that the minimal codebook size $M_\epsilon^*(W)$ for a fixed error tolerance $\epsilon>0$ is closely related to the $\epsilon$-smooth channel max-information, $I_{\max}^\epsilon(W)$. Using this, we refine earlier large-blocklength approximations to $M_\epsilon^*(W^{\otimes n})$ for discrete memoryless channels. Specifically, we provide tight third-order expansions for Gallager-symmetric channels and demonstrate that the third-order term is channel-dependent. As in corresponding channel coding expansions, we identify a substantial rate gap between singular and nonsingular channels. Notably, the direction of this gap is reversed relative to channel coding: The third-order term is $-\frac{1}{2}\log n$ for singular channels but $O(\log\log n)$ for nonsingular channels.