arXiv:2605.28637v2 Announce Type: replace
Abstract: Variational excited-state density functional theory (DFT) enables the calculation of excited states at a cost comparable to ground-state calculations, but single-configuration approaches often suffer from spin contamination. We implement restricted open-shell Kohn-Sham (ROKS) DFT, which recovers spin-pure singlet excitation energies via the variational minimization of a weighted combination of mixed-spin and triplet configurations, within the plane-wave projector augmented-wave framework of VASP. The energy functional is optimized using a preconditioned conjugate-gradient or a direct inversion in the iterative subspace algorithm, and analytical atomic forces are derived. The implementation is validated for eight organic molecules by comparison to the Q-Chem quantum chemistry code, yielding mean deviations of approximately $30\,\mathrm{meV}$. As a solid-state application, we investigate the three lowest lying excitations of MgO with a neutral oxygen vacancy. For a dielectric-dependent hybrid functional, vertical excitation energies from ROKS and time-dependent density functional theory (TDDFT) differ on average by about $0.21\,\mathrm{eV}$. The Franck-Condon shifts deviate on average by $0.14\,\mathrm{eV}$ between the two methods and mass-weighted displacements between the excited states and the ground state by $0.12\,\mathrm{amu}^{1/2}$ Ang. Additional calculations at the PBE level reveal that these properties depend less strongly on the DFT functional for ROKS than for TDDFT. These results demonstrate that ROKS provides excitation energies and excited-state forces with an accuracy similar to TDDFT while retaining the favorable scaling of ground-state DFT, making it a promising approach for affordable excited-state simulations in extended systems.
Science Journals
arXiv:2606.06340v1 Announce Type: cross
Abstract: Alfv\'enic turbulence is vital to powering the solar wind and corona, yet eludes a comprehensive understanding of the kinetic processes by which it dissipates. Minor ions are sensitive tracers of these processes, showing extreme perpendicular temperatures and mass-weighted temperature trends that can either correlate or anticorrelate with mass-to-charge ratio, $A_i/Z_i$. We use a combination of quasilinear theory and 3D hybrid-kinetic simulations to explain these features and their correlations with properties of turbulence in the fast solar wind. When Alfv\'enic turbulence is imbalanced, its cascade to ion-Larmor scales is throttled by the helicity barrier. This barrier ultimately leads to high-frequency proton-cyclotron waves (PCWs), both oblique and parallel, the latter of which produce very flat electric-energy spectra ($\mathcal{E}_{E_{\perp}}\sim k_\parallel^{-\eta}$ with $\eta<2$) over the range of scales that are cyclotron resonant with minor ions. While steeper spectra lead to a positive correlation of heating with $A_i/Z_i$, the shallower spectra cause the dependence to invert, with $Q_i\propto Q_{\mathrm{p}}A_i(A_i/Z_i)^{\eta-2}$. Six simulations of balanced and imbalanced turbulence spanning $\beta_{\rm p0}=\{1,0.3,1/16\}$ corroborate this prediction, showing minor-ion heating rates that follow $(A_i/Z_i)^a$. Minor-ion heating is strongest and most perpendicular in our lowest $\beta_{\rm p0}=1/16$ simulation of imbalanced turbulence, reaching $T_{\perp{\rm O}^{5+}}/T_{\perp{\rm p}}\approx40$ and $T_{\perp{\rm O}^{5+}}/T_{\parallel{\rm O}^{5+}}\approx10$, consistent with low-coronal observations. Future minor-ion measurements should test whether intervals in which minor-ion thermal speeds decrease with increasing mass-to-charge ratio are associated with a history of large cross helicity, enhanced power in parallel PCWs, and a steep transition-range spectrum.
arXiv:2606.06373v1 Announce Type: cross
Abstract: Wireless foundation models have emerged as a promising alternative to building separate models for each wireless task. However, existing approaches rely on masked input reconstruction, which can bias representations toward low-level signal details. In this paper, we propose LatentWave, a wireless foundation model pretrained using a Joint-Embedding Predictive Architecture (JEPA) on diverse wireless spectrograms and channel state information (CSI). By predicting masked regions in latent space, LatentWave learns representations that are more transferable out of the box across diverse downstream tasks. The proposed architecture employs per-channel patch embeddings with stochastic channel sampling during pretraining, allowing it to process variable antenna counts and improving usability across heterogeneous wireless configurations. We evaluate LatentWave on four downstream tasks: RF signal classification, 5G NR positioning, beam prediction, and LoS/NLoS classification, comparing against a masked-modeling baseline (WavesFM) pretrained on the same data. Additionally, we show that the masking geometry introduces a task-dependent inductive bias: frequency masking strongly favors channel-related tasks such as positioning and beam prediction, while region masking better preserves discriminability for signal classification.
arXiv:2606.06469v1 Announce Type: cross
Abstract: Let $S$ be the set of unit norm linear classifiers $\theta \in \mathbb{R}^d$ which correctly classify every point of a labeled dataset $(X_i,y_i)_{i=1}^n$, $X_i \in \mathbb{R}^d$, $y_i \in \{-1,+1\}$, with a possibly negative margin $\kappa$ fixed in advance. Under two natural data-generating distributions of the $(X,y)$ pairs -- a Gaussian mixture model and a logistic model with Gaussian features -- and in the proportional regime $n/d \to \alpha$ with small enough $\alpha$, we establish a large deviation principle on the event that a point $\theta$ chosen uniformly at random from $S$ achieves a given generalization error, with high probability over the choice of the data. The associated large deviation rate function is deterministic and describes the proportion, at the exponential scale in $d$, of interpolating classifiers having a given desired performance. As a consequence, we establish the following concentration phenomenon: all but an exponentially small fraction of interpolating classifiers have approximately the same generalization performance given by the unique maximizer of this rate function.
We numerically compare this maximizer to the performance of empirical risk minimization by gradient descent and to the performance of a natural linear program, both finding a point in $S$, and deduce that in the overparametrized regime of small $\alpha$, these efficient procedures outperform the vast majority of interpolators, pointing to their nontrivial benign overfitting in this setting.
arXiv:2606.06301v1 Announce Type: new
Abstract: Agentic AI is increasingly being integrated into software engineering workflows. In crowdsourced testing, however, the large volume and uneven quality of submitted reports still create a substantial review burden for developers. In prior work, we developed and validated a multi-agent assessment backbone based on the LLM-as-a-Judge paradigm. That backbone assesses reports along three dimensions--textuality, adequacy, and competitiveness--and was shown to align well with human consensus while substantially reducing assessment effort. Yet reliable automated judging does not by itself show whether agent outputs can improve human work when embedded into workflow. This paper studies that missing question in the context of crowdsourced testing. We investigate whether assessment-derived, actionable feedback can improve how testers revise reports, perform on later tasks, and transfer reporting practices across applications. To do so, we conducted a controlled four-stage human-subject study with 20 testers across three real-world applications. The results show that agent-generated feedback supports immediate improvements in revised reports, better first submissions on a new task after prior feedback exposure, and evidence of partial but meaningful transfer to a later application. A post-task questionnaire completed by 17 participants complements these artifact-based findings by suggesting that the feedback was generally understandable, acted upon in revision, and carried into later tasks, while also revealing remaining friction in specificity and execution. Overall, the study provides empirical evidence that, in the studied crowdsourced testing setting, assessment agents can serve not only as post-hoc judges but also as workflow-integrated feedback providers that support upstream report-quality improvement.
arXiv:2509.05528v2 Announce Type: replace
Abstract: We report the in-situ 3D reconstruction of through-going muons in the CUORE experiment, a cryogenic calorimeter array searching for neutrinoless double beta ($0\nu\beta\beta$) decay, leveraging the segmentation of the detector. Due to the slow time response of the detector, time-of-flight estimation is not feasible. Therefore, the track reconstruction is performed using a multi-objective optimization algorithm that relies on geometrical information from the detector as a whole. We measure the integral flux of cosmic-ray muons underground at the {\it Laboratori Nazionali del Gran Sasso}, and find our value to be in good agreement with other experiments that have performed a similar measurement. To our knowledge, this work represents the first demonstration of 3D particle tracking and reconstruction of through-going muons with per-event angular determination in a millikelvin cryogenic detector array. The analysis performed for this work will be critical for validating the muon-related background in CUPID, a next-generation $0\nu\beta\beta$ experiment, and for follow-up studies on detector response and on delayed products induced by cosmic-ray muons.
arXiv:2606.05265v1 Announce Type: new
Abstract: Near-real-time flood depth prediction demands surrogate models that are accurate, fast, and transferable across watersheds. Supervised surrogates can match physics-based simulators in accuracy but need millions of training rows per watershed and cannot extrapolate beyond their original mesh. We propose a domain-aware coreset construction pipeline that conditions a tabular foundation model at inference time. The pipeline stratifies storms by return period and most-affected watershed, then samples hexagons with a target-aware spatial selector. With 0.7% of the per-watershed training pool, the model attains a mean $R^2$ of 0.663 across nine Houston-area watersheds, within 98.5% of the supervised reference ($R^2$ = 0.673). It transfers to held-out watersheds without task-specific retraining, staying ahead of a coreset-trained supervised baseline. On real storms it exceeds the supervised reference on a far out-of-distribution case and trails it on a mostly in-distribution one. Domain-aware coreset construction lets tabular foundation models deliver data-efficient, watershed-transferable flood predictions without per-watershed training.
arXiv:2606.05262v1 Announce Type: new
Abstract: Uncrewed aerial vehicles (UAVs) are increasingly considered as aerial platforms capable of providing both sensing and communication services, representing a promising paradigm for intelligent transportation systems. This paper investigates the optimal time allocation for a UAV-enabled integrated sensing and communication (ISaC) system operating in the X-band for vehicular networks. We analyze the trade-off between sensing accuracy and communication performance under practical UAV constraints and fading effects, considering both single-shadowing and double-shadowing channel models. An optimization framework is developed to allocate time between sensing and communication while guaranteeing minimum communication rates and sufficient sensing reliability. Simulation results demonstrate adaptive time allocation strategies, highlighting how UAV-to-ground channel conditions and target distances influence the balance between sensing and communication in smart mobility scenarios.
arXiv:2606.04332v2 Announce Type: replace
Abstract: Electron spin polarization in radiative plasmas with ultrarelativistic kinetic turbulence under highly magnetized conditions is investigated using particle-in-cell simulations. We observe that a significant spin polarization can be sustained when the leptons undergo energetic photon emission accompanied by spin flips during the nonequilibrium turbulent evolution. By analyzing the time evolution of spatially dependent spin polarization, we identify an electromagnetic (EM) regime of kinetic turbulence, distinct from the well-known density-dominated regime characterized by vortex currents and magnetic islands. While in the latter regime the spin polarization exists only transiently, in the EM regime significant anisotropic net polarization emerges and persists in non-dissipative scenarios. The correlation between spin signals and turbulence features is leveraged to introduce the characteristic parameter delimiting the EM regime via the ratio of electric and magnetic energy densities and to gain insight into complex plasma turbulence. This study demonstrates the versatility of a spin-resolved study of the plasma turbulence in extreme environments, such as black holes and magnetar magnetospheres.
arXiv:2606.05257v1 Announce Type: new
Abstract: Foundation models are increasingly trained on sequences of user actions in recommendation, payments, fraud, and commerce, but these models still lack the kind of compute calibration that scaling laws provide for language models. We study a common two-part behavioral-model architecture: a feature-based event embedder maps each multi-modal item to a vector, and a decoder-only transformer predicts the next event from the resulting sequence. Across roughly 600 runs on real interaction data, spanning $10^{15}$-$10^{19}$ training FLOPs, we jointly vary four deployment-relevant axes: the two-part parameter split, critical batch size, model/data allocation, and the number of sampled negatives used after freezing the embedder. A small embedder ($s^{\star}\!\approx\!2\%$ of parameters) is compute-optimal at every budget we test because embedder parameters are both more expensive per step and exposed to far more repeated items than contextualizer parameters. Compute-optimal training is data-heavy relative to text at low compute, but its $D/N$ ratio moves toward the Chinchilla heuristic as compute increases. The sampled training objective and deployed ranking metrics disagree in ways that themselves scale: critical batch size, optimal negative count after freezing, and the agreement between loss and ranking quality all shift with compute and with the chosen evaluation metric. For negative sampling, larger budgets increasingly prefer more negatives; by $10^{19}$ FLOPs the active constraint is candidate-axis memory rather than FLOPs. In behavioral foundation models, the evaluation metric is therefore part of the scaling law: changing it can change the compute-optimal recipe.
arXiv:2606.05350v1 Announce Type: new
Abstract: Emergency landing flight envelope analysis traditionally adopts a binary notion of safety, whereby a trajectory is safe only if state constraints are satisfied pointwise in time. In practice, ensuring a successful landing requires recognizing that aircraft operation spans a continuum in the state space from the nominal to the critical regime. Between these regimes lies a degraded regime of states outside nominal operation that may be visited only for limited durations. Safety is therefore inherently graded, in the sense that limited exposure to degraded states may be tolerated, and must be assessed using a trajectory-dependent criterion rather than a purely pointwise-in-time one. This paper develops a Hamilton-Jacobi reachability framework for analyzing emergency landing flight envelopes under this graded notion of safety. Safety is encoded through a soft constraint defined by a designer-specified continuous violation cost function that assigns zero cost in the nominal regime and larger cost to more safety-critical off-nominal states. We introduce a general class of state- and time-dependent violation cost functions and establish monotonicity and continuity properties that characterize how the flight envelope varies with the cost of off-nominal operation. These results provide a principled sensitivity analysis linking safety conservativeness to operational capability. Building on this analysis, we propose a synthesis algorithm for parameterized violation cost functions in this class. The algorithm provably converges to the least conservative parameter under which a prescribed off-nominal safety requirement is satisfied. Numerical results for a fixed-wing emergency landing scenario under propulsion failure demonstrate the sensitivity properties and validate the algorithm.
arXiv:2507.00460v3 Announce Type: replace
Abstract: Open Large Language Model (LLM) benchmarks, such as HELM and BIG-Bench, provide standardized and transparent evaluation protocols that support comparative analysis, reproducibility, and systematic progress tracking in Language Model (LM) research. Yet, this openness also creates substantial risks of data leakage during LM testing--deliberate or inadvertent, thereby undermining the fairness and reliability of leaderboard rankings and leaving them vulnerable to manipulation by unscrupulous actors. We illustrate the severity of this issue by intentionally constructing cheating models: smaller variants of BART, T5, and GPT-2, fine-tuned directly on publicly available test-sets. As expected, these models excel on the target benchmarks but fail terribly to generalize to comparable unseen testing sets. We then examine task specific simple paraphrase-based safeguarding strategies to mitigate the impact of data leakage and evaluate their effectiveness and limitations. Our findings underscore three key points: (i) high leaderboard performance on limited open, static benchmarks may not reflect real-world utility; (ii) private or dynamically generated benchmarks should complement open benchmarks to maintain evaluation integrity; and (iii) a reexamination of current benchmarking practices is essential for reliable and trustworthy LM assessment.
arXiv:2507.02758v2 Announce Type: replace
Abstract: In complex systems research, the study of higher-order interactions has exploded in recent years. Researchers have formalized various types of group interactions, such as public goods games, biological contagion, and information broadcasting, showing how higher-order networks can capture group effects more directly than pairwise models. However, equating hyperedges-edges involving more than two agents-with groups can be misleading, as it obscures the polysemous nature of ``group interactions''. For instance, many models of higher-order interactions focus on the internal state of the hyperedge, specifying dynamical rules at the group level. These models often neglect how interactions with external groups can influence behaviors and dynamics within the group. Yet, anthropologists and philosophers remind us that external norms, factors, and forces governing intergroup behavior are essential to defining within-group dynamics. In this paper, we synthesize concepts from social ontology relevant to the emerging physics of higher-order networks. We propose a typology for classifying models of group interactions based on two perspectives. The first focuses on individuals within groups engaging in collective action, where shared agency serves as the binding force. The second adopts a group-first approach, emphasizing institutional facts that extend beyond the specific individuals involved. Building on these perspectives, we introduce four dimensions to classify models of group interactions: persistence, coupling, reducibility, and alignment. For the physics of higher-order networks, we provide a hierarchy of nested mathematical models to explore the complex properties of social groups. We highlight social interactions not yet explored in the literature on higher-order networks and propose future research avenues to foster collaboration between social ontology and the physics of complex systems.
arXiv:2606.05259v1 Announce Type: new
Abstract: We introduce VideoKR, the first large-scale training corpus specifically designed to strengthen knowledge- and reasoning-intensive video understanding. It comprises 315K video reasoning examples over 145K newly collected, CC-licensed, expert-domain videos. We develop a human-in-the-loop, skill-oriented example generation pipeline that targets progressively deeper video reasoning capabilities while ensuring the difficulty, diversity, and reliability of both the examples and their CoT rationales. We also curate VideoKR-Eval, a new expert-annotated benchmark where questions require genuine video understanding and knowledge-intensive reasoning rather than textual shortcuts. Our experiments show that, under a standard SFT$\rightarrow$GRPO pipeline, models post-trained on VideoKR outperform prior post-training approaches on knowledge-intensive video reasoning while remaining competitive on general video reasoning, highlighting data design as a key driver of progress in video reasoning. We further conduct comprehensive ablations to isolate the contributions of VideoKR, providing actionable insights for future work.
arXiv:2606.05348v1 Announce Type: new
Abstract: Inference in probabilistic programs generally requires evaluating many possible program executions to find those of high posterior density. To scale inference to large datasets, it is crucial that expensive intermediate results are shared across these many evaluations, rather than recomputed from scratch. This paper presents a new approach to realizing this sharing, based on \textit{incremental computation}, a technique for efficiently recomputing (deterministic) program outputs when program inputs change. First, we show how expressive probabilistic programs can be compiled to deterministic ones that compute their density functions. Then, building on the incremental $\lambda$-calculus, we develop a general technique for compositionally incrementalizing expressive functional programs, and apply it to these densities. The resulting incremental densities can be used to accelerate a broad range of Monte Carlo inference algorithms, including for nonparametric models not well supported by existing systems. Furthermore, our decomposition of incremental density computation into separate density and incrementalization steps allows for modular reasoning about correctness -- a key pain point in existing systems, where ad-hoc incrementalization features are a known source of soundness bugs. We develop denotational logical relations arguments for the correctness of each step independently, and implement the approach in a Julia prototype, finding that it leads to asymptotic runtime improvements in the size of the dataset on a range of models and inference algorithms.
arXiv:2508.00775v2 Announce Type: replace
Abstract: The design of many classical optimization algorithms is driven by the certification of linear convergence rates over classes of optimization problems. In this paper, we consider the problem of improving the average-case performance of an algorithm over a specific distribution of problem instances. While this task can be tackled by embedding trainable components into the algorithm updates, a key challenge is to preserve worst-case guarantees across the entire problem class. For classes of composite optimization problems, we show that all linearly convergent algorithms can be parametrized in terms of a baseline linearly convergent algorithm, and a set of trainable, exponentially-decaying modifications to its update rule; crucially, this parametrization excludes all-and only-the algorithms that do not converge linearly. Our results apply to improving the average-case performance of classical algorithms such as gradient descent for nonconvex, gradient-dominated functions; Nesterov's accelerated method for smooth, strongly convex functions; and projected gradient methods for optimization over polyhedral feasible sets. We illustrate how our characterization can be used for learning to optimize with linear convergence and feasibility guarantees. Numerical results showcase benefits over classical optimizers when solving ill-conditioned systems of linear equations and running a model predictive control scheme on a linear dynamical system.
arXiv:2508.06249v3 Announce Type: replace
Abstract: Fine-tuning lets practitioners repurpose aligned large language models (LLMs) for new domains, yet recent work reveals emergent misalignment (EM): Even a small, domain-specific fine-tune can induce harmful behaviors far outside the target domain. Even in the case where model weights are hidden behind a fine-tuning API, this gives attackers inadvertent access to a broadly misaligned model in a way that can be hard to detect from the fine-tuning data alone. We present the first systematic study of in-training safeguards against EM that are practical for providers who expose fine-tuning via an API: We evaluate whether they a) prevent broad misalignment, b) allow narrow misalignment, c) learn well on benign tasks, and d) remain coherent. We investigate five training regularization interventions: (i) KL-divergence regularization toward a safe reference model, (ii) $\ell_2$ distance in feature space, (iii) preventive steering with an evil persona vector, (iv) interleaving training examples from a general instruct-tuning dataset and (v) inoculation prompting. We demonstrate that selecting interleaving data by the perplexity gap between aligned and misaligned models yields the best results overall.
arXiv:2410.04907v2 Announce Type: replace-cross
Abstract: In this paper we contribute to the frequently studied question of how to decompose a continuous piecewise linear (CPWL) function into a difference of two convex CPWL functions. Every CPWL function has infinitely many such decompositions, but for applications in optimization and neural network theory, it is crucial to find decompositions with as few linear pieces as possible. This is a highly challenging problem, as we further demonstrate by disproving a recently proposed approach by Tran and Wang [Minimal representations of tropical rational functions. Algebraic Statistics, 15(1):27-59, 2024]. To make the problem more tractable, we propose to fix an underlying polyhedral complex determining the possible locus of nonlinearity. Under this assumption, we prove that the set of decompositions forms a polyhedron that arises as intersection of two translated cones. We prove that irreducible decompositions correspond to the bounded faces of this polyhedron and minimal solutions must be vertices. We then identify cases with a unique minimal decomposition, and illustrate how our insights have consequences in the theory of submodular functions. Finally, we improve upon previous constructions of neural networks for a given convex CPWL function and apply our framework to obtain results in the nonconvex case.
arXiv:2606.05250v1 Announce Type: new
Abstract: Most top-performing autonomous data-science agents rely on frontier cloud models and lack persistent, cross-session memory. This paper addresses two open gaps: (1) the underexplored use of formally structured, quality-controlled Case-Based Reasoning (CBR) case bases coupling symbolic case records with executable code artefacts; and (2) the untested viability of Small Language Models (SLMs) as locally deployable agent backbones. We present CBR-augmented R&D-Agent, integrating a persistent CBR layer into Microsoft's R&D-Agent framework with a custom backend for Gemma 4 31B Dense -- the first published end-to-end evaluation of Gemma 4 as an autonomous data-science agent backbone. The CBR layer overrides three R&D loop phases via a surgical subclass toggled by a single environment variable. Cases are stored as structured records with executable code snapshots and quality metadata; a five-gate quality filter and a heuristic reuse-detection mechanism assess knowledge transfer by combining embedding similarity, code-fingerprint overlap, and injection provenance. Evaluated on two Kaggle competitions (NOMAD 2018, Spaceship Titanic) with four seeds over eight improvement loops each, CBR achieves directionally higher accuracy than the CBR-disabled baseline on Spaceship Titanic (0.8147 vs. 0.8098, d = -1.41) with substantially lower variance. Heuristic reuse detection across 108 retrieval events shows high semantic relevance (mean embedding similarity 0.882) alongside variable structural proximity (mean code-fingerprint similarity 0.305), consistent with conceptual guidance rather than verbatim code copying.
arXiv:2606.05828v1 Announce Type: new
Abstract: As Large Language Model (LLM) capabilities advance, locally deployed personal agents relying on API-based remote models and external skills have emerged as a novel paradigm. With the rapid expansion of available skills, enabling personal agents to learn and adapt to implicit user preferences becomes a critical challenge. However, local deployment constraints preclude complex centralized selection algorithms, creating an urgent need for a lightweight local preference harness. This paper explores the implementation of such a harness through a novel architecture that strictly decouples statistical preference learning from semantic intent parsing. Specifically, we leverage localized statistical results to influence and modulate the selection decisions of the remote LLM. Extensive evaluations demonstrate that our decoupled approach achieves the lowest cumulative regret and highest test accuracy, significantly outperforming traditional memory-augmented agents.
arXiv:2606.05341v1 Announce Type: new
Abstract: Instabilities driven by energetic particles are central to the physics of a burning plasma. The majority of kinetic simulations and reduced models assume that the unstable distribution is already fully established when energetic-particle-driven modes grow unstable. In realistic scenarios, however, energetic particles may accumulate in the resonance on an effective timescale comparable to the growth rate of the instability, meaning that the formation of the resonant distribution and the growth of the unstable mode must be treated concurrently. We study the behavior of these instabilities in the presence of such a dynamically forming distribution, evaluating two distinct metrics which measure how close a mode is to its linear stability threshold and how close a mode remains to its nonlinear stability threshold. It is found that saturation at large $\omega_b/\nu_\text{eff}$ (where $\omega_b$ is the bounce frequency of deeply trapped particles and $\nu_\text{eff}$ is the effective scattering rate at a resonance), normally associated with strongly driven excitation, can be achieved even if dynamically the mode remains at all times near its nonlinear stability threshold. We extend existing analytic models for near-marginal and far from marginal modes allowing for a time-dependent linear growth rate, deriving explicit expressions for the mode amplitude evolution. These formulas are shown to agree with nonlinear kinetic simulations. The discrepancies between the case of a dynamically forming distribution and the case of a fully formed distribution are shown to be particularly pronounced for energetic particle distributions which relax diffusively.
arXiv:2606.05400v1 Announce Type: new
Abstract: Long-horizon autoformalization of research mathematics fails not only at hard lemmas, but at scale: statements drift, dependencies tangle, context decays, and local repairs corrupt distant work. We present LeanMarathon, a multi-agent harness for reliable research-level Lean autoformalization. Its core abstraction is an evolving blueprint: a Lean file that serves simultaneously as formal proof skeleton, natural-language proof graph, and shared system of record. Four contract-scoped agents construct, audit, prove, and repair this blueprint. These agents are coordinated by a two-stage orchestrator that first stabilizes target fidelity through adversarial review and then discharges the proof directed acyclic graph (DAG) from its dynamic leaves upward in parallel CI-gated rounds. LeanMarathon turns one brittle multi-hour run into many local, recoverable, parallel transactions. We evaluate LeanMarathon on two recent research papers spanning four Erd\H{o}s problems (#1051, #1196, #164, #1217). Across three autonomous runs, it formalizes all seven target theorems with no sorry, proving 258 lemmas and theorems. These results show that reliable AI co-mathematics requires not only stronger provers, but durable harnesses that preserve target fidelity across long mathematical developments. The code can be found at https://github.com/YuanheZ/LeanMarathon.
arXiv:2606.05464v1 Announce Type: new
Abstract: Verifiable reward training has improved mathematical and coding reasoning, but these domains capture only part of step-by-step decision making. Many real-world tasks require finding a high-value feasible plan among many valid alternatives. We introduce OPT*, a scalable family of optimization-style tasks for training and evaluating LLM step-by-step optimization-like reasoning along a complexity axis: each task provides a feasibility checker and evaluator, while a complexity parameter expands the search space without requiring new human labels. This motivates studying these tasks in two regimes: (i) solver-guided online policy optimization, which uses a solver as a value oracle for partial states and applies rank-based reward shaping to reinforce better next steps, and (ii) search-based offline RL when such solvers are unavailable. Theoretically, we relate success in large search spaces to the information a reasoner extracts per unit of search budget. Empirically, we ablate the ingredients that make search efficient on OPT* and show that training on OPT* improves step-by-step optimization-like reasoning.
arXiv:2606.05253v1 Announce Type: new
Abstract: Large language models (LLMs) have shown increasing promise in generating
functionally correct register-transfer-level (RTL) hardware designs.
Recent systems improve further through EDA-integrated reinforcement
learning with syntax, simulation, and PPA rewards, but train a general
RTL generator before deployment while test-time approaches search with
a frozen policy. We instead perform reinforcement learning at test time,
allowing the LLM policy to adapt to executable EDA feedback for the
specific RTL problem at hand. We propose TTT-RTL, to our knowledge the
first per-design test-time training framework that closes the loop
between an LLM policy and an EDA pipeline for RTL optimization. TTT-RTL
samples candidate implementations, verifies them through syntax checking
and simulation, scores valid designs using synthesis-derived PPA product,
reuses high-reward variants through a PUCT-indexed design-state pool,
and updates the policy with an entropic policy-gradient objective. To
stabilize policy updates under sparse or plateaued rewards, we introduce
an adaptive KL-budget controller that adjusts the entropy constraint
using reference KL, effective sample size, and reward saturation signals.
On RTLLM v2.0 under Nangate 45nm, TTT-RTL reduces the geometric-mean
PPA product by 65.1% over the reference, outperforming the strongest
published frozen-policy agent baseline at 26.1%. On an industrial
XuanTie C910 FPU leading-zero-anticipation unit under Sky130, TTT-RTL
achieves a 59.4% ADP reduction, and ablations confirm that policy
adaptation, state reuse, and KL-budget control each contribute. These
results suggest that test-time training with executable EDA feedback can
move LLM-based RTL generation beyond functional correctness toward
physically optimized hardware.
arXiv:2606.03189v2 Announce Type: replace
Abstract: Large Language Models (LLMs) as judges across various scenarios such as assessing model responses is becoming an increasingly accepted paradigm. However, existing judgment approaches often rely on trained judgers using fixed preference data, which tend to overlook diverse user preferences and struggle to adapt to real-world human-AI dialogue scenarios. To address these limitations, we propose SenseJudge, a customizable judgment framework driven by human preferences and SenseBench, a diverse and challenging instruction-following benchmark derived from real-world multi-turn interactions. We applied the automatic judgment framework and benchmark to two tasks: (1) LLMs as personalized judges, and (2) model ranking. We conducted extensive experiments, and the results demonstrate that the SenseJudge framework surpasses other judgment methods and models in the LLMs-as-personalized-judges task and achieves model ranking that aligns with real human sense. Additionally, we conducted analyses on position bias and consistency, alongside ablation studies, which affirmed the robustness of SenseJudge.