Papers

Training a single LLM agent jointly across diverse interactive environments has attracted increasing attention as a route to generalist agents. Existing curriculum and data-selection strategies often allocate training at the environment level or prioritize local reward-based signals, without explicitly considering relationships between current rollouts across environments for prompt-group selection. Meanwhile, as environments are learned at different rates, all-failure and all-success rollout groups can coexist within a batch, leaving those data without group-relative reward signals. Both challenges highlight limitations of relying solely on scalar rewards in multi-environment RL: they provide limited information about cross-environment relationships and no within-group reward contrast when rewards are identical. This motivates richer textual feedback, such as rubrics describing rollout behaviours, to guide learning. Beyond rubrics' usage as reward, we repurpose rubrics to guide both online data selection and policy supervision. An LLM judge tags each rollout using a predefined rubric vocabulary shared across environments. The resulting profiles guide the selection of data that aligns with the overall behavioural composition of the mixed-environment batch while limiting overlap with already-selected data. Available positive rubrics (describing desired behaviours) provide privileged context for an on-policy self-distillation teacher, supplying additional token-level supervision, while negative rubrics (describing undesired behaviours) guide subsequent rollout generation away from recurring failure modes. Together, these components form RISED. Across model backbones, RISED achieves the highest mean pass rate across environments and ranks first or second in every individual environment. Rubric-based analysis of RISED can further characterize the behavioural changes accompanying these gains.

Time-series anomaly detection (TSAD) is difficult to generalize across datasets because heterogeneous temporal dynamics imply different notions of normality and favor different detection criteria. While time-series foundation models provide transferable representations, coupling them with a fixed anomaly-scoring mechanism can overlook this variation. This motivates a different perspective on foundation-model-based TSAD: using foundation models to coordinate specialized anomaly criteria rather than directly imposing a universal one. Based on this view, we propose TS-Router, a generalist-representation, specialist-detection framework that estimates the relative competence of heterogeneous anomaly detectors from pretrained temporal representations and selects suitable specialists for each target series. To avoid relying on specialist-performance labels from real tasks, we derive soft competence supervision from specialists' relative performance on labeled simulated tasks. At deployment, routing requires no target anomaly labels, and only the selected specialists are fitted unsupervisedly on the target series. We bound Top-\(k\) set-competence regret under representation coverage and conditional competence stability. Across 16 real-world benchmarks and four complementary evaluation metrics, TS-Router achieves the best overall average rank. Controlled ablations with multiple frozen TSFM encoders further support the use of pretrained representations for competence estimation and adaptive specialist selection. The code is available at https://anonymous.4open.science/r/TS-Router-D8FF.

Broken access control, the failure of authorization, is one of the most prevalent web security risks. Unlike injection, a flow of untrusted input into a dangerous operation, authorization is a relation: who may act on what, not how data moves. Each application decides that relation for itself, so no rule written in advance carries to the next. An LLM agent can infer it from the code, but with no systematic way to cover the application and prioritize what to inspect, its search stays undirected and access-control flaws go undetected. We present ABSENTIA, a security scaffolding that turns general LLM agents into systematic vulnerability detectors for the backend of web applications, run as an audit by the developers and security engineers who maintain the code. Under its direction, the agents build a graph that maps the application's routes to the code behind them. ABSENTIA then works route by route, applying invariant falsification: it infers the properties the code is meant to satisfy, and where one is not enforced, reports the route for maintainer review. We also release BAC-Bench, a benchmark of 30 broken access control advisories across 25 repositories, 3 languages, and 9 frameworks, each published in 2025 or later, verified by a human auditor, and paired with its fixing commit, so credit requires flagging the vulnerable version and not the fixed one. ABSENTIA recalls 19 of them, 17 under paired credit, and an LLM verifier confirms 51% of its findings. CodeQL and Semgrep recall none, and an unstructured agent on the same model recalls 3. In the OWASP Benchmark injection categories, ABSENTIA leads the dedicated analyzers in Python and trails only CodeQL and IRIS in Java.

Probabilistic forecasting is important for predicting complex dynamical systems because intrinsic randomness and incomplete observations can cause the same observed state to evolve into multiple plausible futures. While flow matching is a flexible approach for probabilistic forecasting, it is computationally expensive. Streaming flow (SF) reformulates this approach to model temporal evolution efficiently by learning a continuous velocity field directly in physical time. However, SF learns a deterministic velocity field. Thus, it provides only a single future trajectory for a given fixed initial state and observation history. To overcome this limitation, we introduce Variational Streaming Flow (VSF). Our approach learns a latent distribution that is conditioned on the dynamics of interest. In turn, this enables probabilistic forecasting. Importantly, we retain the computational efficiency of SF by generating in physical time. Across deterministic and stochastic dynamical systems, VSF demonstrates superior predictive accuracy and distributional fidelity. We demonstrate the advantage for both long-horizon rollouts exceeding 1,000 steps, and settings with bifurcating dynamics. Moreover, VSF can be integrated into existing Joint-Embedding Predictive Architecture (JEPA)-based world models as a plug-and-play predictor to improve temporal dynamics and goal-directed success rate in navigation, motion planning, and manipulation.

Tropical cyclone landfall angle strongly modulates storm-surge potential, but how it changes with climate remains unknown. From 1980 to 2024, the global fraction of near-perpendicular landfalls shows a weak, statistically insignificant increase, with no significant basin-scale trends. PepC-Global simulations driven by reanalyses reproduce observed landfall-angle and pre-landfall approach-angle distributions and broad geographic variations in the near-perpendicular fraction. When driven by output from 14 CMIP6 models, the simulations show a similarly weak and statistically insignificant increase during 1980-2024, suggesting a forced contribution not yet detectable above internal variability. Under high emissions, however, the global near-perpendicular fraction increases significantly through 2099, with the strongest and most spatially coherent changes along the eastern North Pacific, whereas no significant global trend emerges under intermediate emissions. The projected increase arises mainly from changes in near-perpendicular fractions at individual coastal locations rather than from a redistribution of landfalls. These results identify landfall angle as an overlooked dimension of changing tropical cyclone coastal risk, one whose evolution this century depends on the emissions pathway.

This paper proposes a physics-guided Bayesian optimization for high-dimensional mixed-variable multiple-input and multiple-output (MIMO) base station (BS) design. The considered problem jointly selects a subset of candidate sites for BS deployment and optimizes the azimuth angles, downtilt angles, and transmit power spectral densities of the BSs, while each configuration is evaluated using computationally expensive site-specific ray tracing. To efficiently optimize the system configuration, the proposed method constructs a low-cost physics-based proxy from precomputed propagation information. The proxy-estimated communication coverage is used as the Gaussian process (GP) prior mean, and a residual GP with three-dimensional physical features learns the discrepancy between the proxy and full evaluations. Ray-tracing-based evaluations in two urban scenarios show that the proposed method achieves up to approximately 15 percentage points higher coverage than conventional and high-dimensional optimization baselines under the same evaluation budget.

Science teachers frequently search for documentary excerpts not by describing what appears on screen, but by querying the abstract concepts they intend to teach. This use case exposes a limitation of existing language-based video moment retrieval methods, which typically assume that queries describe observable events, whereas instructional search requires retrieving concrete visual phenomena that instantiate an underlying scientific principle. We study this setting as concept-to-example video retrieval, an abstract-needle-in-a-haystack problem where compact curriculum concepts must be grounded in temporally sparse documentary evidence. To bridge this abstraction gap, we propose Concept-Driven Domain Adaptation (CDDA), a three-stage framework for adapting two-tower vision-language models to concept-level retrieval. CDDA treats concepts as intermediate semantic anchors: it first structures the textual embedding space with textbook and teacher-handbook example-concept pairs, then transfers this concept-aware geometry to documentary visuals under a frozen visual encoder, and finally jointly adapts both encoders with sparse visual concept supervision. From a geometric perspective, this staged alignment reduces text-concept and vision-concept angular gaps, thereby encouraging concept-level adaptation while preserving the pretrained model's concrete image description alignment. On a curated middle-school physics retrieval benchmark, CDDA achieves stronger pedagogically oriented concept retrieval than several competitive multimodal baselines, including Qwen3-VL-Embedding-2B, while maintaining concrete image-text matching after adaptation.

As LLM agents undertake increasingly complex, long-horizon tasks, verifying their outputs becomes increasingly challenging. We study how verification capability can be strengthened with a fixed base model, without access to reference answers or grading rubrics at test time. Repeated sampling yields multiple rollouts that can contain complementary correct claims, but we need a reliable verification mechanism to determine which claims to trust. We first find that disagreement often exposes correct alternatives, while consensus can conceal errors. These observations motivate VeriHarness, which turns the underlying LLM a generator uses into an agentic verifier by giving it a workspace, evidence tools, and reusable verification skills. A disagreement resolver checks competing claims against environmental evidence, while a consensus challenger tests shared claims and searches for omitted requirements. Their findings guide the selection and revision of the final artifact. Across five long-horizon workspace benchmarks and two frontier models, VeriHarness achieves the highest selection scores among the evaluated baselines. Evidence-backed revision further improves average performance, bringing gains over a single rollout to 6.2 points with Gemini 3.5 Flash and 6.4 points with Claude Opus 4.8. We further show that verification skills can self-improve from failure feedback, demonstrating VeriHarness as a novel and critical approach for scaling long-horizon agentic verification. We release the full pool of approximately 26,000 rollouts across all five benchmarks and both models, produced at a cost of over $100,000, to support future research on agentic verification.

Moving a workflow developed and tested at one HPC site to another rarely succeeds without some amount of trial and error. Package managers rebuild software environments, containers ship whole filesystems, and workflow specifications such as backpacks package a workflow with its software, data, and resource requirements. These approaches address one half of workflow portability: what a workflow needs. But none describes how a given HPC site must be used, and that missing half is why even a portable workflow requires manual adjustment at each new site. That gap includes the site's resource shape, storage configuration, network permissions, and operating policies. This information may be explicit in the batch system, hidden in the prose of documentation, or buried deep within a router's configuration, making it difficult for an automated deployment tool to turn site knowledge into useful deployment decisions. We propose the HPC site profile, a structured, evidence-backed document that makes this knowledge actionable. We automatically construct it in three steps that mirror where the information lives: measuring the login node, extracting typed fields from documentation with a bounded language-model agent, and submitting pilot jobs for eligible unresolved fields. Every field is verified against its evidence or discarded, so a rule, not the model, decides what enters the profile. The profile then preflights a workflow into an execution plan or an early, explainable failure. We build profiles at Purdue Anvil, TACC Stampede3, and Notre Dame CRC and present a case study of preflighting a real workflow.

Recent advances in 3D reconstruction have progressed from per-scene optimization to feed-forward inference, and semantic scene understanding has followed suit – yet existing methods remain confined to object-centric perception, neglecting spatial relations between objects. We formulate 3D spatial relation segmentation in a feed-forward, pose-free multi-view setting: given a visually specified subject and a relational text query, the model segments the target across views without receiving its category name. To this end, we propose RelationVGGT, a novel feed-forward framework that integrates semantic features from a visual foundation model with geometry-aware representations from a 3D geometry foundation model and leverages a relation transformer for subject-conditioned, cross-view relation prediction – requiring neither per-scene optimization nor known camera poses. We additionally provide a fully automated annotation pipeline built on ScanNet++ with VLMs and LLMs, enabling scalable training data generation for this new task.

Large language models (LLMs) have been increasingly used for financial document analysis, including earnings call transcripts (ECTs). Beyond generating standalone claims, users increasingly prefer grounded analyses that pair claims with verifiable citations from source documents to enable independent validation. However, evaluating such analytical claims typically requires extensive expert annotation, which is costly and difficult to scale, and real-world financial analysis commonly involves long context-question-answer triplets, further increasing task complexity. To address these challenges and benchmark the current landscape of grounded analysis by LLMs, we propose a numeric evidence evaluation method that enables groundedness assessment without reliance on expert annotation. We also introduce an automated dataset construction pipeline and construct ECTs-100 from the top 100 constituents of the S&P 500 to support benchmark of both groundedness and correctness. In addition, we examine conscious incompetence, a practical failure mode in financial analysis in which LLMs must detect when available evidence is insufficient and refrain from producing unsupported hallucinations. Empirical results show that LLMs perform well in groundedness but face notable limitations in correctness, with informational insufficiency presenting an additional challenge.

Causal representation learning aims to discover robust features by exploiting the causal structure underlying data generation. Existing methods require specifying the causal structure a priori, yet different structures demand fundamentally incompatible invariance constraints, and misspecification leads to representations that discard predictive information. We introduce SaCRL, a framework that jointly identifies the causal structure and learns the corresponding invariant representation without prior structural knowledge. Our approach formulates structure selection as a soft optimization over candidate invariances using HSIC-based violation metrics, with adaptive weights that automatically concentrate on the achievable structure. We provide theoretical guarantees for structure identification, including under random-feature approximation, invariance satisfaction, and out-of-distribution generalization. Empirically, SaCRL recovers the true structure on synthetic and semi-synthetic Bayesian-network benchmarks, outperforms fixed-invariance baselines on Colored MNIST, achieves state-of-the-art accuracy on three DomainBed benchmarks (PACS, VLCS, OfficeHome), and degrades gracefully under structural misspecification and limited environment diversity. Code is available at: https://github.com/ArmanBehnam/sacrl.

We construct indefinite theta representations for a family of signed generalized Nahm sums of arbitrary rank \(r\geq2\). After explicit normalization and boundary correction, one signed branch gives the even components of a \((2r-1)\)-component holomorphic theta vector associated with a lattice of signature \((1,1)\). Euler-type multiple sums provide the odd components, yielding formulas uniform in both the rank and the component index. Using Zwegers' theory, we determine the weight-one modular transformation laws of the real-analytic completion, with an explicit finite sine matrix for the \(S\)-transformation. We decompose the nonholomorphic correction into a finite sum of products of unary theta functions and one-dimensional error-function series, and compute the antiholomorphic derivative of the completion. Together with the completed transformation laws, these formulas give the inhomogeneous transformation law of the holomorphic vector. We also evaluate a linear combination of two signed branches as a difference of Jacobi product quotients. This identity and further evaluations for two families of symmetrizable matrices and their duals, combined with known results, give multiple-sum representations for every component of four associated product vectors. These representations are finite combinations of generalized Nahm sums with the prescribed matrices and symmetrizers; in the dual families, half-period translates are absorbed into signs and constant phase factors.

We prove the real-rootedness conjecture of Lin, Ma, Ma, and Zhou (2021) for multiset Eulerian-Narayana polynomials, including simplicity and negativity of all zeros after removing the factor t. Using a forest decomposition and Lagrange inversion, we derive a differential formula and a first-order factorization of the differential operator. We use these factors to show that the nonconstant gamma-coefficients and ordered gamma-zeros are nondecreasing, with corresponding bounds given by the Narayana and Eulerian polynomials. The same factorization gives totally nonnegative coefficient transition matrices. We also prove strict interlacing of the zeros of generating polynomials arising from adjacent columns of the transition matrices.

Tidal disruption events (TDEs) can launch sub-relativistic outflows that drive shocks into the surrounding circumnuclear material (CNM), providing a natural site for cosmic ray (CR) acceleration and high-energy neutrino production through hadronuclear ($pp$) interactions. We model this emission for optically bright and infrared (IR)-only TDEs, the latter motivated by a recently identified population of luminous IR transients with weak or absent optical counterparts, consistent with TDEs embedded in dusty, obscured nuclear environments. We semi-analytically compute the shock dynamics, CR acceleration and transport, and neutrino production, including the radiative cooling and compression in dense environments. While optical TDEs remain inefficient $pp$ neutrino sources with comparatively small neutrino yields, radiative compression in dense IR-only TDEs enhances the target density and allows the pp efficiency to approach the calorimetric regime, yielding up to $\sim 10\%$ of the observed diffuse neutrino flux at $100$ TeV, given the rate uncertainties. The long-lasting neutrino emission and low individual source yield motivate joint electromagnetically informed stacking searches. We forecast such searches for IceCube, IceCube-Gen2, KM3NeT, and HUNT, and conclude that a future $\sim 30\ {\rm km^3}$ neutrino observatory can reach $3\sigma$ sensitivity for nearby ($z \lesssim 0.4$) IR-only TDEs with multi-year search windows. Growing optical and IR TDE samples can therefore enable population-resolved searches that can test whether dense, obscured nuclear environments are efficient high-energy neutrino sources.

For small merchant businesses (SMBs) whose catalogs fit within a long-context LLM, full-catalog prompting offers a compelling alternative to multi-stage retrieval designed primarily for large marketplaces with millions of items. However, fitting the full catalog into the context window does not ensure that the model can use it effectively, since LLMs do not exploit long contexts uniformly. We therefore study in-context catalog search through two complementary questions: (1) how to curate and present catalogs to the LLM, and (2) how to adapt the LLM for product selection on curated contexts. We propose RPTune, an end-to-end framework that couples learned catalog curation with LLM post-training using automatically generated, catalog-grounded supervision. An encoder-reorganizer curator orders and prunes products guided by downstream LLM feedback, while the resulting curated catalogs in turn improve the effectiveness of LLM post-training with a context-relative reward. We evaluate RPTune on 7 real merchants spanning distinct retail verticals, using 100 complex conversational queries per merchant. RPTune consistently improves search accuracy across both proprietary and open-weight LLMs, with context curation yielding gains of up to 31.4 percentage points and post-training adding a further 10.3 points on average.

Band topology is usually assigned to a single crystal structure, yet at finite temperature a crystal is an ensemble of thermally disordered configurations with a fluctuating band gap. We ask whether the thermally averaged structure is a faithful proxy for this ensemble, taking Bi$_2$Se$_3$ as a representative topological insulator and classifying band ordering in each ab initio molecular dynamics snapshot by scaling the spin-orbit coupling. The averaged structure fails in two ways: it misses the Rashba-like spin splitting that instantaneous symmetry breaking produces in individual configurations, and it substantially underestimates the band-gap renormalization, whereas a small ensemble of harmonic Monte Carlo configurations reproduces both the mean and the spread of the gap. Using a band-inversion survival probability, we find that the band inversion of Bi$_2$Se$_3$ holds in every sampled configuration up to 600 K, above which individual configurations cross into normal band ordering well before the average gap closes. Topological robustness at finite temperature is thus a property of the distribution over configurations rather than of any single structure.

Pleiotropy, the phenomenon in which a single genetic variant influences multiple traits, is a fundamental feature of genome biology and an important consideration in statistical genetics. Many statistical methods exist that leverage pleiotropy to increase the power of association tests, but they are often designed to identify SNPs which are associated with at least one of the underlying traits. Only a limited number of methods exist for identification of “strict pleiotropy”, i.e. those associated with more than one trait, but these methods are either restricted to a small number of traits or become computationally challenging as the number of traits increases. In this paper, we develop two testing procedures to identify SNPs under strict pleiotropy and determine the corresponding minimum number of associated traits. Both methods rely only on z-scores available from genome-wide association studies (GWAS) and are applicable to independent as well as correlated traits. The first procedure is frequentist in nature and is designed to control the family-wise-error rate. The second procedure is based on the local false discovery rate and provides improved power while maintaining FDR control. We conduct extensive simulation studies to compare the performance of the proposed methods with alternative methods for testing strict pleiotropy. Finally, we demonstrate an application of the proposed methods to Alzheimer's disease-related biomarkers for the goal of distinguishing SNPs that may represent biomarker-specific variations from those that are related to disease-relevant biology.

As code generation is increasingly delegated to AI systems, the bottleneck is shifting from writing code to supervising the systems that write it — a shift CS-education researchers have begun to name. This shift exposes a vocabulary gap: the field asks for "human oversight" without a working distinction between the two things language does in a delegation channel — coordinate action (cybernetic: words succeed when the world comes to match them) and coordinate understanding (epistemic: they succeed when they answer to the world and a hearer can check that they do). The failure this names is not cybernetic language but epistemic-form language doing cybernetic work: explanation-shaped output calibrated for approval rather than truth. Oversight that checks only whether an output was approved is satisfiable by rubber-stamping; oversight that holds an agent accountable requires the reasoning behind its work be retrievable and checkable. We present three delegation episodes — illustrations, not controlled evidence — in which epistemic engagement proved practicable while remaining auditable, one public record where a recommendation was withdrawn on its own stated terms, and one failure case illustrating oversight that requires no reasons for its discretionary choices. We propose a criterion for agentic-system governance, alongside existing technical trust properties: every consequential choice should carry the condition under which it would have gone otherwise, in a form a third party can test. Without such a condition, a third party cannot distinguish a decision from a rubber stamp. We give the criterion an operational form — a two-part reconstruction test scoring a delegation record by whether a second reader can predict what the agent does under a perturbation — and a deliberation-recording convention, ORRCF, that makes the condition a required component of every recorded choice.

A video benchmark should reward the capability it claims to measure, yet models can exploit answer options, question text, or partial visual evidence. We introduce the attack pyramid, five levels of shortcut attacks with increasing access to each item, and audit 115 video benchmarks with it. On 35 benchmarks, attackers that never see a frame approach full-video accuracy. On 51 benchmarks with temporal probes, shuffled frames keep a median 96% of full-video accuracy. Near-duplicate questions make up at least half the items in 63 benchmarks. We screen 505,518 question-answer pairs from 112 of them into an audited pool. Agents turn evaluation requests into specifications, and a deterministic selector with a red-team gate composes reproducible benchmarks. We release Video-Index, the 210 hardest verified items under these attacks in each of four capability groups, 840 items from 76 sources. With the same fixed input, Claude Opus 5 outscores every open-source model by over 37 percentage points, and agent tools add about 20 more, yet all systems leave room to improve efficiency and accuracy. Blog: https://www.enxinsong.com/blog/video-index/ GitHub: https://github.com/Espere-1119-Song/Video-Index Hugging Face: https://huggingface.co/datasets/Video-Index/Video-Index

For the $N$-body problem in dimension $d \ge 3$, we give an alternative proof of a theorem of da Luz and Maderna on the nonexistence of free-time minimizers defined on the entire real line, and extend it from the Newtonian potential to a larger class of singular subharmonic pairwise interaction potentials. The proof is based on the following generalization of the Cheeger-Gromoll splitting theorem: let $(M,g)$ be a connected Riemannian manifold and let $U$ be a smooth positive function such that $\mathrm{Ric}_g \ge \tfrac{\Delta U}{2U}g$, and the metrics $Ug$ and $U^{-1}g$ are both complete; if the metric $Ug$ admits a line, then the function $U$ is constant and $(M,g)$ splits off that line. We apply this splitting theorem to the Jacobi-Maupertuis metric associated with the $N$-body problem; a quantitative version of Marchal's theorem enables us to reduce to the case of smooth potentials.

Large language models often lose track of persistent instructions and relevant information as multi-turn conversations grow. We study this cumulative contextual decay through three related failure modes: attention pollution, dilution, and drift. We propose REA (Role-aware Heuristic Episodic Attention), a context-management framework that assigns different persistence and representation policies to instructions and episodic interactions. Instructional Memory retains identified global constraints in a dedicated prefix. Episodic Memory preserves user inputs and compresses model replies, while heuristic retrieval selects raw text, compressed representations, or omission for each historical turn. On Long-MT-Bench+, REA improves the judge score from 6.32 to 7.36 on a 10-point scale, a 16.5% relative gain over the Vanilla baseline, and reduces average latency by 2.91$\times$. Additional evaluations show aggregate gains on three backbones spanning 1.7B-7B parameters and on Chinese and English role-playing tasks. These results support role-aware context management as a practical approach to maintaining conversational continuity and instruction adherence.

We study a real-valued max-plus particle system on a finite periodic lattice. Its three-dimensional fundamental diagram relates the mean flow to two densities: the conserved particle density and a second density defined by a local state function. We prove that the deviation from the diagram tends to zero for arbitrary initial data with all state values in [0,1]. The proof uses two nonnegative local quantities and their evolution relations. For rational initial data with all state values in $[0,1]$, the orbit is eventually periodic, and the fundamental-diagram relation holds exactly on the periodic orbit.

Atomic layer deposition (ALD) offers precise, conformal control of thin-film growth and is a workhorse for semiconductor manufacturing, but its use in superconducting quantum circuits is constrained by the need to simultaneously control crystallinity, stoichiometry and interfacial disorder at cryogenic temperatures. Here we develop hydrogen plasma assisted ALD to epitaxially grow superconducting titanium nitride (TiN) thin films, using a hydrogen plasma step to modify the surface chemistry to suppress precursor-derived impurities during each deposition cycle. Synchrotron X-ray scattering reveals semi-coherent epitaxy of TiN on c-plane sapphire, with discrete crystallographic domains and minimal long-range structural disorder. Complementary X-ray spectroscopy and neutron reflectometry show negligible oxygen-related disorder throughout the film and no substantial hydrogen incorporation. The resulting TiN films exhibit a superconducting transition at 2.2 K, a kinetic inductance of 15 pH/sq and a superconducting coherence length of 14.6 nm, comparable to the measured crystal coherence length. These results establish hydrogen plasma assisted ALD as a route to structurally ordered superconducting TiN thin film epitaxy and demonstrate that the nanoscale control of ALD can be extended to materials suitable for cryogenic quantum technologies.

This paper studies the focusing Calogero–Moser derivative nonlinear Schrödinger (CMdNLS) equation under the nonzero boundary condition \(|u(t,x)|\to 1\) as \(|x|\to+\infty\). In sharp contrast to the classical paradigm for the cubic nonlinear Schrödinger equation, namely that focusing nonlinearities support only bright solitons on a zero background, we show that the focusing CMdNLS equation admits a rich family of dark multi-soliton solutions on a nonzero background, whose existence is driven by a spectral gap \([-1,0]\) in the point spectrum of the Lax operator, together with the Hardy space structure. The main results are threefold. First, we establish an explicit formula for general solutions in the Hardy–Zhidkov space \(\mathcal Z^2_+\), via a semigroup method based on new commutator estimates between the Lax–Beurling shift semigroup and Toeplitz operators. This approach circumvents the non-commutativity of the unbounded generator with the time derivative and requires no weighted hypothesis on solutions or on the domain of validity of commutator formulas. Second, the algebraic definition of dark multi-soliton potentials is shown to be equivalent to two spectral conditions on the Lax operator. Together with the conservation of a generating functional, this yields the invariance of the dark \(N\)-soliton set \(\mathcal U_N\) under the CMdNLS flow. Third, the action-angle variables on \(\mathcal U_N\) are constructed, which establishes the complete integrability of the focusing CMdNLS equation on \(\mathcal U_N\). Combining these results, we obtain an inverse spectral formula, from which it follows that the poles of dark multi-soliton solutions satisfy a complexified Calogero–Moser system. Long time asymptotics and uniform Sobolev estimates are also derived.

As large language models (LLMs) move from standalone assistants into agentic workflows, evaluation must extend beyond scalar leaderboard accuracy to account for operational reliability, cost, latency, and token efficiency. We use an agentic ensemble of small language models (SLMs) with an SLM-judge-mediated feedback loop as a case study for such beyond-leaderboard evaluation. On the 541-prompt IFEval benchmark, the best ensemble achieves 97.34% strict prompt accuracy, exceeding the strongest standalone LLM baseline, gpt-5.4, by 5.81 percentage points while operating in a lower-cost regime. We then analyze the tokenomics and operational behavior behind this gain, including cost per sample, token composition, useful-output goodput, feedback-loop recovery, latency decomposition, and performance across instruction categories and constraint counts. Our results show that agentic SLM ensembles can trade additional test-time tokens and orchestration overhead for improved instruction-following fidelity, motivating multi-dimensional evaluation protocols for future agentic AI systems.

An answer candidate in a masked diffusion MLLM can stabilize while its rationale is still unfolding. We distinguish retrospective stabilization of the logged candidate from token commitment, and examine these two clocks relative to rationale generation. Analyzing our results across three visual question-answering benchmarks, we find that 89.4-98.1% of the rationale-side canvas remains unwritten at stabilization in single-block, EOS-suppressed LaViDa runs. On V*Bench, reducing block length from 128 to 8 changes this fraction from 89.4% to 1.7%, together with answer coverage and the eligible observation window. Under EOS-enabled prompting, direct instructions improve Nemotron's overall accuracy by 15.0 and 19.5 percentage points on M3CoT and ScienceQA, but reduce LaViDa/V*Bench accuracy by 11.0 points. A symmetric decomposition associates the larger absolute component of each change with coverage rather than conditional accuracy. Matched-canvas image ablations measure visual sensitivity alongside answer stabilization, separating the two temporal readouts. Together, these measurements distinguish answer stabilization, rationale unfolding, and visual sensitivity, and identify coverage as the larger component of the prompting differences.

Recaptioned image-text corpora are now standard for text-to-image (T2I) training, with vision–language model (VLM) captioners replacing sparse alt-text by dense descriptions. A recaptioned corpus is a supervision distribution induced by a documented captioning policy ($\pi$), captioner ($V_c$), and source corpus ($C$). Length-correlated proxies miss caption-register artifacts and downstream T2I benchmarks entangle the corpus with training choices, so this distribution is hard to audit at corpus scale. We introduce a reusable matched-budget audit framework for recaptioned supervision distributions $D_{\pi,V_c,C}$: at a fixed text budget of $B = 64$ it reports a five-axis profile spanning prompt-side coverage, image-conditioned faithfulness, and caption-surface health, with claimed controllable basic units (CBU) as the common claim unit. We instantiate the framework on seven paired comparisons over five public source corpora. Across the four cross-corpus pairs, the released surface raises supported CBU per caption by $+3.39$ to $+6.36$ under both Qwen and Gemma Judges, and on CC12M the same framework exposes a long-vs-dense frontier that is consistent across both judges and four budgets. We release the audited multi-source recap corpus ($\approx$ 490M) together with the audit-artifact bundle.

On April 20, 2020, the May front-month WTI oil futures contract, one day before its expiration date, opened near $\$17/$barrel and dropped far below zero in a single trading day, reaching an intraday low of $-\$40.32$ and settling at $-\$37.63$. Such market behavior was unforeseen at the time. This event, and the March 2022 nickel squeeze, illustrate how constraints on physical delivery create pressure to close futures positions and distort futures prices. We develop a feedback model that connects the resulting price distortion to a roll option, and to the imbalance between delivery-constrained long and short positions. Under lognormal benchmark dynamics, the roll-option value is shown to satisfy a pricing PDE which is nonlinear because its payoff depends on the observed futures price, which itself includes the feedback correction. Remarkably, in both cases, there is an explicit solution in terms of classical Black-Scholes-Margrabe exchange option formulas, up to solving a scalar equation. We show this allows that prices may go negative under distortion of a positive-price lognormal model. An analogous construction applies when the lognormal base is switched to Bachelier normal dynamics. Illustrations based on WTI and nickel show the imbalance required to reproduce the extreme event prices under the benchmark assumptions.

Discretizing electronic degrees of freedom into atomic orbitals is the default choice to accurately model electrons in molecules in a compact and flexible way. We consider the task of simulating such electronic structure models on a quantum computer for $N$ orbitals, keeping the number of orbitals per atom constant. Atomic orbitals give rise to Hamiltonians with potentially up to $\mathcal{O}(N^4)$ terms and little structure, which made it challenging to match the complexity of methods based on plane waves and real-space grids. In this work, we show that a Trotter step for atomic orbitals can be implemented in depth $\mathcal{O}(\mathrm{polylog}(N))$ without ancillas by combining a rigorous treatment of orbital localization, a hierarchical Hamiltonian decomposition based on the Fast Multipole Method, and shallow quantum Fourier arithmetic circuits. By bounding the Trotter error, we obtain sublinear simulation depth, and our total gate count $N^{5/3 + o(1)}$ matches the best scaling known for any basis.