Papers

We present calculations of nuclear shielding constants, a key ingredient in nuclear magnetic resonance spectroscopy, within a quantum-computing-suitable framework. We employ quantum linear response (qLR) theory in combination with the unitary coupled cluster (UCC) ansatz and its orbital-optimized (oo-UCC), factorized (fUCC), and orbital-optimized factorized (oo-fUCC) variants. The methods are assessed for seven small molecules against CASCI, CASSCF, CCSD, and CCSD(T) reference calculations. The different UCC-based qLR approaches yield shielding constants in good agreement with comparable classical methods. Factorization, which is necessary for quantum-computing implementations, has only a negligible effect on the calculated shielding constants. Orbital optimization, in contrast, leads to substantial improvements, with oo-UCC and oo-fUCC showing markedly better agreement with CCSD and CCSD(T). Our results establish oo-fUCC combined with qLR as a promising approach for computing nuclear shielding constants on quantum computers.

In this study, we revisit Lorenz's pioneering concept of analogues to investigate error growth and predictability in fully developed turbulence using experimental data. By applying the method to a 1D turbulent velocity time series, we successfully identify the three error growth regimes intuited by Lorenz and later formalized by turbulence theory: the chaotic exponential regime, the algebraic regime of spontaneous stochasticity, and the saturation regime. Our analysis reveals that the Lyapunov exponent and the exponent of the algebraic regime depend on the initial error. We interpret this dependence as a statistical bias induced by the intermittency of turbulence, where rare events, characterized by lower Lyapunov exponents and more singular velocity increments, are preferentially sampled when considering larger initial errors. These findings not only validate Lorenz's qualitative intuition but also provide new quantitative insights into the predictability of turbulent flows, paving the way for potential applications in the prediction of extreme events.

For $0<s\leq 1/2$ and $d\geq 1$, the $\widetilde H^s$-projection onto piecewise linear finite element spaces on quasi-uniform meshes of a bounded Lipschitz domain in $\mathbb{R}^d$ with a uniform exterior ball condition is almost stable in the maximum norm. The mesh domain need not coincide with the original domain.

Learning to act in unfamiliar environments requires agents to infer how the world works and revise that understanding as new evidence arrives. Yet limited observations can support multiple world models that explain past interactions but predict different outcomes in unseen states. We introduce Memento 3, building on the Memento series to enable frozen LLM agents to continually learn explicit world models through external memory. The agent maintains a natural-language rulebook as persistent semantic memory, recording revisable hypotheses about environment dynamics while leaving unknown aspects underspecified. It compiles this rulebook into executable code for prediction and planning. Through a continual loop of observation, reflection, rule revision, compilation, and verification, the agent uses prediction errors to refine both the rulebook and its code. Updated code is accepted only when the LLM judges it faithful to the rulebook and cell-exact replay reproduces the observed transitions. We investigate this process as a model-based route to recursive self-improvement (RSI): the agent autonomously explores the environment, revises its world model, and uses verified updates to guide subsequent interaction and learning, while the underlying LLM remains fixed. A population extension maintains multiple world models in parallel, sharing interaction evidence and using their predictions to guide exploration. On ARC-AGI-3, the single-model agent clears every level of all 25 public games, achieves a mean Relative Human Action Efficiency (RHAE) of 100.0, and uses 44% of the human action count. In an Atari Pong case study, a learned feedback controller wins 21:0 in each of three evaluated episodes with different openings, without further LLM calls.

For the quadratic family $f_c(z)=z^2+c$, we study the non-escaping locus $\KM=\{(c,z):c\in\M,\ z\in K_c\}$, the model space for cubic capture straightening in the framework of Inou and Kiwi. Here $\M$ is the Mandelbrot set, $K_c$ is the filled Julia set of $f_c$, and $J_c=\partial K_c$. We prove that $\partial\KM$ has full Hausdorff dimension $4$. The set $\KM$ is locally connected at every $(c,z)$ with hyperbolic $c\in\M$. At parameters on $\partial\M$, we construct mixed puzzle pieces from Yoccoz puzzles and parapuzzles. For non-degenerate pairs, shrinking of both planar puzzles implies local connectedness of $\KM$ at $(c,z)$. However, local connectedness fails at some $(c,z)$ with $c$ real parabolic and $z\in K_c\setminus J_c$, even though both the parameter space and the filled Julia fiber are locally connected there.

We present AuraLuxMuse, a novel system for automated aesthetic stage lighting design that integrates expert knowledge, representation learning, and preference-adaptive modeling. Lighting design in live performance settings requires the seamless translation of musical features into dynamic lighting behaviors. However, traditional workflows remain time-consuming, labor-intensive, and difficult to transfer. AuraLuxMuse encodes music and professional cue sequences into a shared retrieval space, estimates cue-event density, and retargets selected fixture commands to the destination stage. It assists pre-production authoring by returning editable cues rather than replacing the designer with an unconstrained generator. At the heart of AuraLuxMuse are two key modules: Lighting-Aligned Music Pretraining (LAMP), which performs contrastive learning between audio and lighting cues for alignment, and Preference-Adaptive Mixture of Experts (PAMoE), which conditions preference-aware cue retrieval and adaptation on designers' intent through a gated ensemble of style-specific expert networks. To support training and evaluation, we introduce Musilux, the first dataset of paired musical audio and professional lighting cue sequences under diverse performance scenarios. We evaluate AuraLuxMuse across both virtual simulation environments and professional-grade laboratories. Experimental results, including objective and subjective evaluation, demonstrate that AuraLuxMuse retrieves and adapts stage-lighting cues that are visually cohesive, semantically meaningful, and artistically expressive, showing its potential for AI-assisted aesthetic stage design.

Generating physically plausible videos for solid-gas dynamics is challenging as different phases exhibit distinct dynamics yet remain coupled through physical interactions. We present PAVG, a Phase-Aware Video Generator for solid-gas dynamics and interactions. It employs a dual-branch architecture to explicitly model the distinct dynamics of solids and gases, while spatiotemporal cross-attention captures their physical interactions. This design enables PAVG to preserve phasespecific motion characteristics while producing physically consistent responses across phases. To facilitate this task, we further construct a simulation corpus comprising over 700K physical trajectories across diverse solid, gas, and solid-gas interaction scenarios. Extensive evaluations demonstrate that our PAVG produces videos with improved motion adherence, physical plausibility, and visual quality compared with existing approaches.

Byte-level language models such as the Byte Latent Transformer (BLT) group bytes into patches and run their large global model once per patch. BLT starts a patch where a small model's next-byte entropy is high, so global compute goes where the next byte is hard to predict. We show that this rule has a systematic blind spot: positions whose type is predictable but whose value must be computed, such as the number after "=" in a worked math solution. Under tight patch budgets, entropy-triggered layouts skip these positions, and accuracy on them collapses. In Meta's BLT-1B with patch starts on 10% of bytes, the entropy rule puts a patch start at 16% of the computed results in GSM8K solutions and gets 19.0% of them exactly right; a boundary after each "=" at the same patch count gets 51.8%, and entropy combined with a label-free boundary-dependence signal gets 67.1% (default layout at 26% of bytes: 76.8%). The gap survives adapting BLT-1B to the budget with low-rank fine-tuning (32.9% vs 72.7%, three runs per rule, paired p < 1e-200) and grows with model size in byte models trained from scratch at a 10% budget: at 1M, 12M and 50M parameters, boundary dependence beats entropy on final answers by -1.6, +10.1 and +19.8 points, and at 50M it gets 35.9% of computed results against 13.9% (3 seeds each). BLT's entropy-jump rule helps neither target at 50M. The entropy trigger of Scratchpad Patching is likewise indistinguishable from random scratchpads on final answers (5.6% vs 5.9%, 5 seeds), while answer-start scratchpads give 38.1%. The effect is specific to computed values: copies and lookups gain little, and values the model cannot compute gain nothing. Boundary dependence, the rise in the model's own loss when a patch start is removed, measured per two-byte context, finds these positions without labels: combined with entropy it beats the hand-written rule on computed results.

Motivated by the recent JUNO indication favoring normal neutrino mass ordering, we study two realizations of the minimally modified Frampton–Glashow–Yanagida (MMFGY) framework in the minimal type-I seesaw with two right-handed neutrinos. Both retain one exact Yukawa texture zero, supplemented by either a single physical CP-violating phase or a remnant CP symmetry. The explicit-CP realization yields twelve high-energy Yukawa configurations that reduce to three low-energy classes, with correlated CP phases and predictions for neutrinoless double-beta decay. In both realizations, a strongly suppressed second Yukawa entry can emerge, although the full-matrix hierarchy remains moderate in the quasi-degenerate regime. We analyze leptogenesis for a non-degenerate benchmark, quasi-degenerate TeV-scale right-handed neutrinos, and lower masses where coherent production, oscillations, and electroweak sphaleron freeze-out become important. Successful leptogenesis is obtained across a broad range of the scanned benchmarks, with increased sensitivity to the initial right-handed-neutrino abundance at the lowest mass scale. Requiring successful leptogenesis restricts the allowed phases while preserving the possibility of an approximate second zero. The two realizations exhibit distinct approximate-zero patterns and leptogenesis predictions, demonstrating how different restrictions on the high-energy Yukawa structure affect baryogenesis.

AI-washing is the use of 'AI' as a marketing term where products' claims are exaggerated, likely to capture the hype from investors and consumers. In this paper, we examine how startups market AI and how people perceive those claims. We audited the German startup market, applying an annotation codebook of seven dimensions to 100 startups and conducting 63 expert interviews. Startups exaggerated AI capability in three forms: claims without inspectable evidence, performance numbers without methodology, and claims without acknowledged limits. The capabilities they claimed rarely matched the underlying system, and consumers who had encountered exaggerated claims before withdrew trust from later AI marketing. AI-washing has also been linked to the marketing of unethical products: 38% of the sample were flagged for moderate or serious ethical concern, in domains such as surveillance and automated decisions in regulated fields. Together these findings show how AI-washing happens: founders exaggerate capability to attract investors who reward AI branding they cannot verify, and consumers in unfamiliar domains take the marketing at face value. We recommend that startup AI claims be brought under disclosure requirements, with the codebook offered to regulators, investors, and consumer-protection bodies for routine audits.

Multimodal depression assessment offers a promising approach to analyzing behavioral patterns associated with depression. However, existing methods often rely on dense and opaque multimodal representations, making it difficult to interpret the behavioral patterns underlying their predictions. In this work, we introduce BehavDep, a sparse factor-based framework that decomposes multimodal behavioral representations into sparse latent factors and associates them with behaviorally meaningful concepts through a semantic bridge. To address the mismatch between user-level annotations and heterogeneous video-level behaviors, BehavDep further learns video-level depression tendency scores under weak supervision and aggregates information across multiple observations for user-level assessment. Extensive experiments demonstrate that BehavDep achieves the best overall assessment performance while revealing complementary modality contributions, heterogeneous behavioral patterns across observations, and prediction responses to concept-level editing. These results show that BehavDep provides a structured and interpretable approach to analyzing multimodal behavioral representations for depression assessment.

In this paper, we present a one-parameter extension of some supercongruences related to Van Hamme's $\mathrm{(E.2)}$ and $\mathrm{(F.2)}$ supercongruences. To deduce these results, we first establish the corresponding $q$-supercongruences modulo the third power of a cyclotomic polynomial. Our proofs rely on the creative microscoping method devised by Guo and Zudilin~\cite{guo2019q}, together with Jackson's ${}_{6}\phi_{5}$ summation formula.

Given a pair of complementary subgroups $P$ and $N$ of a finite group $G$, we first discuss how the same type of functions $\mathbb E\colon P\times N\to \mathbb T$, which we call bi-$1$-cocycles, appear naturally in the study of projective representations of $G$ of Heisenberg type and second cohomology of $G$ and of its dual quantum group $\widehat G$. We show then that such representations are irreducible and the corresponding ordinary and dual $2$-cocycles are nondegenerate if and only if the matrix $(\mathbb E(p,n))_{p,n}$ is invertible, if and only if it is Hadamard. We illustrate the result with examples arising from symplectic nilpotent Lie algebras over finite fields with two complementary Lagrangian subalgebras.

Tabular in-context learning (ICL) has emerged as a training-free and accurate paradigm for tabular prediction, but current approaches to compressing its in-context examples face an accuracy-throughput tradeoff: fixed subsets can sacrifice accuracy, while query-specific retrieval limits cache reuse and batching across queries, reducing throughput. We propose QCOC (Query-Calibrated Operator Compression), which exploits the exchangeability and repeated use of in-context examples by compiling their full KV cache once into compact memory shared across subsequent queries. Instead of retaining raw examples, QCOC clusters their states into joint-KV prototypes, preserves per-cluster multiplicities and the original example count, and calibrates prototype values against attention query vectors produced by the in-context examples through an anchored closed-form solution. Prototype compression drives the speedup, while value fitting helps preserve accuracy. On 64 held-out OpenML-CC18 datasets, QCOC achieves the highest mean accuracy among the compared compression and retrieval methods at both retained counts. Across 12 configurations on seven long tables, it ranks first among compressed methods in ten and averages 0.23 percentage points below full context. Compressing 8,192 in-context examples to 512 memory slots yields a 10.5x cache compression ratio; excluding one-time compilation, in a single-core CPU online-serving comparison over 1,000 queries, QCOC is up to 508x faster than dynamic retrieval baselines and 1.98x faster than full-context inference. These results show that QCOC enables compact-memory reuse and efficient inference across queries while retaining accuracy close to full context.

Shared e-scooter services have become a widely adopted urban transport mode. While most users ride responsibly, alcohol intoxication stands out among the factors contributing to severe crashes. Nonetheless, countermeasures remain limited to single-point reaction tests and night bans that suspend the service altogether. This paper proposes a new approach in which onboard sensors evaluate the rider as the trip unfolds, raising an alarm as soon as enough evidence of impairment has accumulated. Specifically, we introduce a detector that operates on inertial and throttle measurements, with a provable bound on the rate of false alarms. Experiments on sensor data from 141 rides, in which 25 participants rode while sober and at two target blood alcohol concentration levels, confirm that the bound holds, whereas baselines and ablations either exceed it or lose detection performance, and in some cases delay the alarm. At a bound of 0.023, the detector identifies 91% of the rides performed at the higher concentration and 50% of those at the lower one, with median detection times of 25 and 27 seconds, respectively. We further show that an embedded implementation meets the real-time requirement, making mitigation actions feasible onboard, without requiring data to leave the vehicle. Overall, this work lays the ground for interventions that reach impaired riders as soon as possible, sparing the sober ones the burden of a pre-ride test or the suspension of the service at night, while letting operators budget false alarms against user experience.

Interactive agents can turn experience into reusable skills, yet existing self-evolving skill libraries primarily improve by accumulating new knowledge. Failures may lead to new skills, while previously stored skills are less often revisited as new evidence arrives. However, growth alone does not ensure reliability, as a retrieved skill may be inapplicable under the current task conditions, and an existing skill may encode a mis-specified operational boundary. Reliable skill evolution therefore requires not only adding knowledge, but also testing and revising what is already stored. We introduce Skill-V, a verifiable self-evolving skill library. To make stored knowledge testable, we propose representing skills as versioned, falsifiable contracts that link semantic intent to observable behavioral criteria. We use environment outcomes to drive library evolution. Specifically, task failures motivate skill addition, while disagreements between contract evaluations and task outcomes guide revisions to existing skill boundaries. To validate these revisions, we require them to preserve protected semantic constraints and satisfy non-regression criteria for rubric-outcome metrics on historical replay evidence. Finally, we employ an applicability-aware filter to exclude candidates judged confidently inapplicable to the current task. Across ALFWorld and WebShop, Skill-V achieves success rates of 95.3% and 85.9%, respectively, while maintaining a more compact skill library than growth-oriented baselines. Applicability-aware filtering reduces incorrect skill invocations, and outcome-grounded revisions correct mis-specified skill boundaries without degrading performance on previously observed evidence. These results show that reliable skill evolution requires more than accumulating experience: the library must learn which knowledge to retain, when to revise it, and when it should be applied.

We consider weakly reinforced Pólya urns on a countable graph of uniformly bounded degree. The vertices have positive bounded firing rates, which need not be bounded away from zero. For every reinforcement exponent $\alpha\in[0,1)$ and every deterministic choice of finite positive initial weights, the normalized edge weights converge almost surely, coordinate-wise, to a deterministic vector independent of the initial weights. We identify this vector as the unique non-vanishing equilibrium of the limiting equation. This proves the homogenization conjecture of Couzinié and Hirsch, including its formulation for non-uniform firing rates in the MATRIX open-problems report. The main step is a Liouville theorem for the limiting equation, established by a logarithmic-norm comparison of complete solutions using the row-sum balance at each vertex.

In this note, we show that a locally conformally flat free boundary minimal hypersurface in $B^n$ must be either an equatorial ball or a critical catenoid for all $n \geq 4$. As a consequence, we conclude that the equatorial ball is unique among locally conformally flat free boundary minimal hypersurfaces in $B^n$ with the topology of $B^{n - 1}$.

In this paper, we establish several $q$-supercongruences for basic hypergeometric series modulo the third and fifth powers of a cyclotomic polynomial. These results extend several existing supercongruences related to the $\mathrm{(A.2)}$ and $\mathrm{(C.2)}$ supercongruences of Van Hamme. Our proofs employ the creative microscoping method developed by Guo and Zudilin \cite{guo2019q}, together with Watson's ${}_{8}\phi_7$ transformation formula.

We compute the entanglement entropy of quantum matter in spacetimes with a Killing horizon entirely on-shell, by integrating the first law of thermodynamics along a one-parameter family of equilibrium backgrounds labelled by the Hawking temperature. This yields a closed expression in terms of the renormalized stress-energy tensor and the Killing vector, finite by construction, with no replicas and no conical singularities. Its covariant form agrees with the matter terms of the Bardeen–Carter–Hawking mass variation formula. We show in details that for static patch of de Sitter space and for the non-rotating BTZ black hole it reproduces the conical entropy exactly, contact term $Q$ included.

Long-term personalization requires language models to use interaction history to track users' preferences across sessions. Parametric memory encodes this interaction history into model parameters or adapters, reducing the need to include it in the inference context. However, independent context compilation leaves cross-session integration unspecified, while recurrent updates can attenuate earlier evidence. To address these challenges, we propose Dual-Path Parametric Memory (DPPM). Its Evidence path directly pools representations of the interaction history to preserve earlier evidence, while its Delta path sequentially updates an associative state to capture changes. Fusing both outputs produces history-conditioned LoRA adapters that combine evidence accumulation with ordered revision. Across multiple backbones, DPPM outperforms the evaluated baselines, achieving 54.22% on PersonaMem-v2 and 86.79% on PrefEval. These results suggest that DPPM provides a simple and effective design choice for cross-session personalized parametric memory.

Sparse Mixture of Experts (MoE) models scale more efficiently than dense models by routing tokens to modular expert networks that are only active for processing a fraction of tokens. A leading hypothesis for the performance of MoE models is that each expert specialises in a single, coherent domain. However, interpretability efforts that assume this hypothesis have generally been unsuccessful. We propose and present evidence for an alternative account that we call the Superposed Specialisation Hypothesis (SSH): experts specialise in a disjoint union of fine-grained features rather than one broad domain. Leveraging the SSH, we introduce RouterInterp, a method for interpreting expert routing that identifies Sparse Autoencoder features most predictive of routing decisions and produces unified natural language explanations. On gpt-oss-20b, RouterInterp explains expert routing with ${\sim}65\%$ higher detection accuracy than prior token statistics based methods. This work provides a scalable method for generating more accurate explanations of expert routing and increases our understanding of a previously uninterpretable component of foundation models.

We consider two-level hybrid Schwarz domain-decomposition preconditioners for finite element discretizations of the high-frequency Helmholtz equation (with wavenumber $k$). To date, all existing $k$-explicit theory for these preconditioners considers coarse spaces in the $\textit{asymptotic regime}$, i.e., where the Galerkin solution is $k$-uniformly quasi-optimal. We consider the situation where the fine and coarse spaces consist, respectively, of piecewise polynomials of degree $p$ on meshes of width $h$ and $H_c$. We prove results for the coarse space in the $\textit{preasymptotic regime}$, under the familiar mesh threshold that $(H_c k)^{2p} \rho(k)$ is sufficiently small, where $\rho(k)$ is the norm of the solution operator (such that $\rho(k)\sim k$ for the Helmholtz equation in free space). We prove that, in this regime, the fixed-point iteration with this preconditioner converges in a number of iterations that grows at most like $\log k$. We give numerical experiments demonstrating this theory. Furthermore, these experiments indicate that our theoretical results are sharp, in that the number of iterations required for these preconditioners to converge with $(H_c k)^{2p} \rho(k) \gg 1$ grows rapidly with $k$.

Guard models are increasingly used to safeguard LLM-based agents, primarily by identifying actions that agents are forbidden to perform. However, identifying forbidden actions alone is insufficient to ensure agent safety. In this paper, we argue that agent safety also depends on identifying required yet unperformed safety-critical actions, which we call obligations. Our preliminary study on a popular benchmark for evaluating safety shows that 56.92% of GLM-5.3 trajectories contain unfulfilled obligations, compared with only 30.00% containing forbidden actions. This finding reveals unfulfilled obligations as a major and previously overlooked source of safety risk. However, to our knowledge, no existing benchmark evaluates whether guard models can identify these obligations. To close this gap, we introduce ObligationBench, the first benchmark for evaluating the capability of obligation identification, comprising 240 expert-validated trajectories covering issue resolution, feature development, and terminal operations. Our evaluation of 14 representative models reveals substantial limitations: the highest recall and exact-match rate are only 48.97% and 10.00%, respectively. To address these limitations, we develop ObligationGuard using 40,000 synthetic training examples. ObligationGuard achieves 57.52% recall and an exact-match rate of 21.67%, surpassing all evaluated models on both metrics. We call on the community to incorporate obligation identification into the design and evaluation of future guard models to improve agent safety.

Zador's celebrated theorem is a cornerstone of optimal quantisation: it establishes both the weak limit of the empirical distribution of an optimal $n$-point quantiser in $R^d$ and the decay rate of the associated $L_s$-mean quantisation error. In large dimension, however, observing this asymptotic behaviour requires an astronomically large sample size. We prove that, for spherically symmetric target distributions, optimisation over all spherically symmetric distributions is a convex problem and derive an equivalence theorem that both characterises global optimality and yields a constructive algorithm. We show that, for moderate $n$, random quantisers uniformly distributed on a sphere of suitably chosen radius $R$ perform exceptionally well and, over a broad range of values of $n$, are numerically certified to be optimal among all random quantisers. Their expected distortion has an explicit integral representation that can be evaluated to arbitrary precision, and we prove concentration across random quantisers: the distortion variance tends to zero as $n\to\infty$ for fixed $d$. For $s=2$, both the optimal radius and the associated minimum expected distortion admit exact expressions. For general $s$, the optimal radius can be determined efficiently, and extreme-value theory provides useful approximations when $n$ grows with $d$. Depending on this growth rate, $R$ either converges to zero or approaches a positive limit that is independent of $s$.

Vision-Language-Action (VLA) policies typically predict actions at fixed time intervals, coupling the route a robot follows with its execution pace. This coupling complicates adaptation from teleoperation: useful geometric guidance comes with timing shaped by interface delays and operator behavior. Our key insight is to bring the path-time parameterization of classical motion planning into the learned action representation of a VLA. We introduce PathTime-VLA, which represents motion as a progress-indexed interaction path $X(s)$ and a positive interval-time profile. The latter defines a monotone time law $t(s)$, yielding controller commands $X(s(t))$. For a given path, alternative executions are expressed through the time profile, allowing chunk-wise speed choices without changing the geometric prediction target. This representation supports a staged post-training procedure: demonstrations and DAgger interventions establish a target-domain prior, Speed-DQN learns execution multipliers from robot interaction, and Path-AWR uses rollout outcomes to refine the diffusion path generator. A path-conditioned action expert realizes the resulting motions while maintaining distinct learning interfaces for path generation and execution timing. Across three tasks, the complete method achieves $58/60$ successes versus $57/60$ for PathTime-VLA under BC + DAgger at fixed $1\times$, with approximately $39$-$52\%$ shorter mean completion times over successful trials.

AI-Native video creation is shifting from isolated video clips toward iterative video creation workflows. However, existing datasets remain largely video clips, representing video generation and editing as separate tasks rather than connected stages of a video creation workflow. In this paper, we introduce Sora100K, a dataset that represents the AI-Native video creation workflow as a structured video creation trajectory. Specifically, we first identify video creation trajectories and decompose them into three subsets according to their structural roles: text-to-video generation records as roots, single-turn video editing records as editing edges, and multi-turn video editing records as complete trajectories. Then, we use a VLM to assign semantic annotations for generation roots and editing-operation annotations for editing edges. A strict construction pipeline further reconstructs source-to-edit lineage, editing order, and intermediate video states while ensuring data quality. Finally, we perform lightweight adaptation on LTX-2 models to assess the supervision value of Sora100K. The results show improvements in visual quality, multi-shot generation, and cross-shot consistency, while successive-turn evaluation reveals that following multi-turn editing instructions remains challenging. Sora100K establishes a new data foundation for AI-Native video creation beyond isolated video clips and toward structured video creation trajectory. The dataset and supplementary materials are publicly available at https://huggingface.co/datasets/ysicong/Sora100K.

LM26 is a magnetized target fusion experiment built to demonstrate compressional heating of a plasma with a collapsing solid lithium liner. We report the design of a Thomson scattering diagnostic for LM26 and measurements from the LMC-12 compression shot where an electron temperature of $1090 \pm 40 \rm{eV}$ was measured with high confidence. The LM26 Thomson scattering diagnostic can measure temperature during compression at 8 times and 2 spatial points. Access limitations restricted the design to narrow f/30 views, and a forward scattering configuration with scattering angles < 20deg. The small solid angle of the scattered light view is largely compensated for by the 10 times increase in the density during compression to $> 10^{20}\rm{m}^{-3}$. Measuring at a small scattering angle significantly narrows the Thomson spectral broadening of the laser line. The narrow spectrum (1030-1063 nm) is advantageous for the design of optical components and to minimize noise from plasma background light. UKAEA polychromators specialized for small broadening measurements were used to analyze the scattered light. Results are also presented for an uncompressed plasma preceding LMC-12. Stray light was not a significant source of error. This design provides a strong design basis for Thomson scattering systems on future LM26 configurations.

Large language model (LLM) agents are beginning to operate industrial energy equipment, and what they get right depends on what they are told about the plant. Established building ontologies name many kinds of points across many sites, whereas an industrial equipment system needs few entities with much knowledge about each. This study proposes the ontology tower, a narrow-and-deep ontology of a single equipment system whose knowledge deepens in two ways: through quantities derived from the measured points by physical relations, and through lessons from the operating journal incorporated as knowledge nodes. On a real low-humidity air-handling test plant operated daily through a programmable logic controller, agents received a text projected from its tower in a preregistered evaluation of nine tasks replayed from the plant's records, using four open-weight models from 9 to about 750 billion parameters. This knowledge raised the rate at which the agents avoided the most plausible misjudgment of each task by about 20 percentage points, and the overall task score of the 9-billion-parameter model as much as that of the largest. Operating lessons were used when incorporated into the tower or placed in the prompt as records, but seldom when left in the journal behind a search tool. In live runs through an invariant safety layer, the agents brought the controlled variable into its target band in 12 of 14 runs. An ontology narrow in entities but deep in what is known about them can thus supply the knowledge that an agent for an industrial equipment system needs.

The frequency-dependent absorption and emission of light determine the colors of matter and underpin today's technologies, from solar cells to light-emitting diodes. Predicting optical properties from first principles remains a computational challenge, especially for large and disordered systems. The $GW$ plus Bethe-Salpeter equation ($GW$-BSE) approach provides predictive accuracy for optical properties, but conventional implementations become costly as system size grows. Here, we present a $GW$-BSE framework combining real-time propagation, atom-centered basis sets, and a locally optimized real-space resolution of the identity. The cost of each propagation step scales quadratically with system size and roughly equals the cost of one self-consistent-field iteration in a density-functional calculation with a semilocal functional. This computational efficiency allows us to calculate the absorption spectrum of a nanographene containing 4616 atoms and opens a path toward routine calculations of optical properties for large disordered systems.