Papers

Large Language Models increasingly serve as interfaces for knowledge-intensive information seeking tasks across languages by synthesizing multilingual evidence. Prior work has shown that they often exhibit query-language preference – the tendency to favor sources written in the language of the query – but has largely examined this behavior in settings where equivalent knowledge is available across languages. However, this bias becomes consequential when sources in different languages provide incomplete or inconsistent accounts of the same fact, since the information users receive then depends on the sources a model selects to use. To characterize query-language preference under such cross-lingual knowledge disparities, we introduce Waldo, a multilingual Question-Answering (QA) benchmark constructed from Wikipedia. Waldo contains 12K QA pairs targeting knowledge gaps, where a fact is available in one language but absent in another, and knowledge conflicts, where language editions provide conflicting versions of the same fact. Evaluating eight models across five languages, we find that when one language edition merely lacks the relevant fact, models generally use evidence from the other language regardless of the query language. Under conflicting accounts, however, model responses strongly align with the document in the query language, causing semantically equivalent queries to elicit different accounts depending on the user's language. Finally, we explore two different approaches that could mitigate this preference under knowledge conflicts: a mechanistic intervention that ablates attention heads associated with query-language preference, and LoRA-based training, which reduces the preference gap by up to 61.5%.

The rapid proliferation of data centers (DCs), driven by cloud computing and artificial intelligence (AI), has led to massive energy demand and carbon emissions, posing significant sustainability challenges. Carbon-aware optimization in geographically distributed data centers has been widely studied. Most existing approaches mainly focus on operational carbon emissions from server usage. However, existing literature often ignores workload-induced thermal stress, which accelerates nonlinear hardware degradation. This leads to more frequent server replacements and ultimately increases embodied carbon emissions. To address these limitations, we propose a comprehensive carbon life-cycle modeling framework for distributed data centers. Apart from operational carbon emissions, this work combines workload scheduling with a utilization-dependent exponential aging model to evaluate long-term carbon costs from server degradation. In order to solve the proposed optimization model in an online and privacy-preserving manner, an enhanced Lyapunov framework with time-varying queue shifting (TVQS) is first introduced to handle system uncertainties. Then, a zero-sum perturbation-based alternating direction method of multipliers (ZSP-ADMM) framework is developed to enable distributed coordination across geographically separated data centers while protecting locally exchanged workload information. Simulation results demonstrate that the proposed approach achieves up to 13.0% lower carbon emissions and 12.6% lower operational costs compared with benchmarks.

Vision-language-action policies often see only one or a few recent frames, which makes it difficult to evaluate how they use information that disappears during a task. We introduce MIKASA-Robo-VLA, a benchmark of 90 language-conditioned manipulation tasks. All but 10 hide the cue an action depends on. Those 10 are reactive controls. MIKASA-Robo, the suite it rebuilds, has 32 tasks and uses language only in a representative VLA subset. Here every task provides an instruction, while memory-dependent tasks hide a task-relevant cue and reactive controls keep it available. For 70 tasks, environment phase timings specify an information gap, and for 28 of them the gap exceeds the 16-frame window of the widest fixed-context VLA we survey. The gap counts only the interval the cue is provably absent, not the full duration a policy must retain it, so every memory-dependent task still requires memory by construction, including the ones whose measured gap is short. We release 22,500 oracle trajectories across 10 memory types in RLDS and LeRobotDataset v3. A reference $\pi_{0.5}$ baseline with current images and proprioception, but no observation history or explicit memory module, is fine-tuned on 14 tasks and achieves 0.211 $\pm$ 0.044 mean task success. Its lower success on the evaluated Long-split tasks is confounded by open-loop chunking and the memory types represented in that subset. Project page: https://mikasarobo.github.io/

In quantum mechanics, the properties of an unstable system are closely related to its corresponding energy distribution. One of the predictions of the theory is the emergence of a late-time power law in the survival probability $P(t)$ of the unstable state. However, in complex systems, similar effects may emerge due to other phenomena. This is the case for nonextensive statistical models, for some continuous lifetime distribution models, and also for certain energy-migration models applied to fluorescence. In this work, these alternative interpretations are confronted with data from time-resolved fluorescence spectroscopy experiments with erythrosine B and eosine Y as fluorophores. It turns out that some of the features expected within the quantum-mechanical description, such as the sharp transition from the exponential to the power-law regime, cannot be reproduced by the specific statistical and energy-migration parametrizations fitted in the analysis.

Computational pathology models infer biologically and clinically meaningful outcomes from histology, but their predictions are shaped by complex, intertwined tissue signals whose roles are important to understand. Common pixel- and feature-space perturbations can produce implausible tissue, making model responses difficult to interpret. We introduce CPathOGen, a conditional latent-diffusion framework for generating paired H&E counterfactuals with explicit controls over cellular spatial organization, nuclear morphology, and stain appearance. Cellular maps condition spatial structure through a spatial encoder, while a morphology/appearance vector modulates denoising through blockwise feature-wise linear modulation (FiLM). On held-out H&E tiles, CPathOGen generates visually plausible tissue with improved distributional agreement after spatially guided selection, as reflected by lower Fréchet Inception Distance (FID) and Kernel Inception Distance (KID); generated cells track requested abundance and position, and measured morphology and color vary monotonically with their controls. We use these verified interventions to probe pathology encoders with endpoint heads, task-specific classifiers, and survival models. Responses are quantified using total variation distance and prediction-flip rate. We further introduce the Biology-Nuisance Sensitivity Ratio, a metric that contrasts model sensitivity to biologically motivated morphology and spatial factors with sensitivity to non-biological, stain-related nuisance variation. CPathOGen provides a practical, fidelity-audited framework for evaluating robustness and controlled feature sensitivity in computational pathology model. \href{https://github.com/a12dongithub/PathOGen}{GitHub} and \href{https://huggingface.co/a12donhf/CPathOGen}{Hugging~Face}.

Reasoning-enabled VLA policies expose chain-of-thought (CoT) traces that appear to explain and guide their actions, creating a potential interface for runtime safety through reasoning monitoring and correction. In this work, we define and operationalize two evaluation axes for assessing when this interface can improve embodied behavior: correctability, which measures whether unreliable reasoning can be detected and improved during generation, and actionability, which measures whether reasoning corrections produce behaviorally meaningful changes in the intended direction. To enable correctability, we introduce Token-level Reward for Utility-Steered Chain-of-Thought (TRUST), an offline-trained value model that predicts eventual reasoning correctness from partial prefixes and uses these estimates to monitor and selectively steer reasoning generation in frozen VLA policies. On the Alpamayo 1.5 driving VLA, TRUST monitors correctness with 88.9% accuracy and improves reasoning correctness from 75.9% to 90.0%. On a baseline-defined challenging subset in AlpaSim, TRUST reduces collision rate by 30.4% and maximum trajectory error by 11.5% relative to the unsteered policy, outperforming a compute-matched Best-of-4 baseline. On the DeepThinkVLA manipulation VLA, TRUST improves the correctness of grasp-state claims from 69.3% to 90.2% and action-choice claims from 68.8% to 85.9%, yet closed-loop task performance on LIBERO-Plus remains largely unchanged. Empirical analysis reveals intent-consistent behavioral effects in Alpamayo 1.5 but limited effects in DeepThinkVLA, helping interpret these different task-level outcomes. Together, our results show that gains in reasoning correctness do not automatically imply gains in embodied performance, motivating evaluation of correctability and actionability when using CoT as a runtime safety interface.

Representation alignment has become an effective way to accelerate diffusion training, but its benefits do not transfer reliably to pixel-space clean-image prediction. In JiT, we find that auxiliary feature alignment can improve access to semantic features while reducing access to image variation needed for clean-image prediction, creating a mismatch between the auxiliary objective and the denoising task. This suggests a different principle: auxiliary supervision should improve the prediction target itself rather than impose a separate representation target. We introduce JAx (Just Align x), a prediction-supervision method that aligns clean-image predictions across noise levels. JAx couples a noisier student observation with a cleaner observation through a Markov degradation that preserves the original JiT input distribution. Under this coupling, the oracle prediction from the cleaner state has the same conditional mean as the optimal JiT target, while its conditional target covariance is no greater. Thus, oracle prediction alignment preserves the population JiT objective up to a constant while providing a lower-variance training target. To make this construction practical with an imperfect EMA teacher, JAx combines ground-truth supervision with a reliability-gated coupling band that selects nearby teacher states based on prediction risk. On ImageNet 256x256, JAx consistently improves FID and accelerates convergence across JiT-B/16, L/16, and H/16, without an external encoder or changes to the architecture or sampling procedure. Gradient diagnostics further show reduced minibatch gradient variance, while ablations demonstrate that the gains cannot be explained by time reweighting alone. These results show that prediction-space supervision provides a simple and principled alternative to representation alignment for pixel-space generative models.

In this paper, we give exact combinatorial expressions for all CUE joint moments on the unit circle, in the case that all exponents are even positive integers, yielding the corresponding asymptotics. We offer two equivalent expressions for these quantities: the first is a sum involving Kostka numbers; the second is related to counting integer matrices (magic squares). As an application of our results, we unify many results in the literature, which were previously disconnected. Moreover, we prove a conjecture of Basor et al. (2018) regarding an irregular conformal block structure.

RR Lyraes (RRLs) are important tracers of old stellar populations and crucial distance indicators. To complement large variability surveys such as Gaia and Rubin-LSST, the topology of the RRL instability strip (IS) and their period-luminosity-metallicity (PLZ) and period-Wesenheit-metallicity (PWZ) relations need to be updated using state-of-the-art stellar evolution and pulsation models. We employ the updated set of BaSTI horizontal-branch (HB) evolutionary models, with finer sampling in effective temperature and metallicity, and compute the corresponding pulsation models using the mesa-rsp module building on the approach of Marconi et al. (2015). We compute a grid of linear non-adiabatic radial pulsation models with mesa-rsp, covering -2.2 <= [Fe/H] <= +0.06 dex. Seven convection configurations are explored including the recently empirically calibrated prescription for RRLs by Kovacs et al. (2024). Analytical IS boundaries are derived for RRab and RRc pulsators and transformed into several photometric systems (Gaia, OGLE, 2MASS, and LSST) and the Kiel diagram. The updated evolutionary models do not substantially change the pulsation predictions: RRab IS edges agree well with previous results, while the RRc red edge is slightly redder. The width of the IS decreases systematically with increasing alpha_MLT, and RRc is relatively more sensitive to the convection prescription than RRab. The PLZ and PWZ relations are also in good agreement with previous empirical and theoretical calibrations across the NIR bands; however, the convection prescription giving the closest match is wavelength dependent. The predicted IS boundaries and PLZ/PWZ relations exhibit a coupled dependence on pulsation mode, metallicity, and convective treatment, with no single prescription performing uniformly best across all observational planes.

Neuronal networks exhibit complex dynamics shaped by connectivity and stochastic input. Empirical studies show that neuronal networks contain recurring subgraphs, or motifs, but the collective influence of different motif types after embedding in large stochastic networks remains less well understood. We construct a spiking network composed of six representative structural classes and examine how intrinsic noise, coupling strength, inter-motif connectivity, network size, and neuronal heterogeneity shape synchronization. Both motif- and network-level coherence peak at intermediate noise intensities, consistent with coherence resonance. Bidirectionally coupled pairs (M2) and the type-2 recurrent feed-forward loop (M3c) consistently exhibit high motif-level coherence. Rewiring these motifs while preserving local synapse number and strength produces some of the largest reductions in network coherence, showing that connection arrangement contributes beyond strong local coupling alone. M2 and M3c also exhibit frequent spike doublets, linking short inter-spike intervals with elevated coherence. Relative to a degree- and weight-matched random control, the motif-structured network reaches greater coherence at weaker noise. Increasing network size enhances coherence until the response begins to saturate, whereas applied-current heterogeneity lowers peak coherence and shifts the optimum toward stronger noise without erasing the relative differences among motif classes. These results show that local connection arrangement remains dynamically consequential after embedding and can shape noise-driven coherence at the population level.

In frequency division duplexing (FDD) massive multiple-input multiple-output (MIMO) systems, the uplink (UL)-to-downlink (DL) channel covariance matrix (CCM) conversion problem is studied to relieve the heavy burden of DL training and feedback required for channel estimation. Learning- based methods perform well up to a certain array size, but for a fixed dataset size their accuracy deteriorates with the number of antennas, to the point where simple model-based methods outperform them. This paper identifies a key cause of this behavior and addresses it. The mean angle of arrival (AoA) induces a phase ramp along the lags of the CCM. Since the oscillation rate of this ramp grows with the number of antennas, a dataset of fixed size becomes increasingly sparse relative to the variation that must be captured. We propose estimating the slope of this ramp from the UL CCM separately and mapping it to the DL band in closed form, leaving the learner with a residual that is largely insensitive to the mean AoA, which substantially reduces the performance degradation with an increasing number of antennas. The proposed scheme, termed deramping, is a combination of pre- and post-processing steps that applies to learning-based conversion methods without altering their internal structure, as demonstrated on three structurally different learners. Simulation results show that deramping reduces the covariance estimation error of all three learners under uniform, Laplacian, and Gaussian angular power spectra,keeps the interpolation-based learners ahead of a model-based benchmark at large array sizes, and improves downlink channel estimation.

Bell nonlocality allows secret key to be certified without trusting the internal operation of the devices, and even against eavesdroppers constrained only by the no-signaling principle. We ask whether generalized contextuality, a notion of nonclassicality available in prepare-and-measure experiments on a single system, can play an analogous role. In our approach, Alice and Bob first establish operational equivalences among their local preparation and measurement procedures, which are then treated as theory-independent constraints that any physical intervention on the communication channel, including that of an adversary, must respect. Eve is otherwise unrestricted, that is, her intervention on the transmitted system is not assumed to be described by quantum theory. Within this adversarial model, we show that Eve's optimal guessing probability against individual attacks is the solution of a finite linear program, and hence that a worst-case asymptotic key rate follows from the observed statistics and the operational equivalences alone. We prove that contextuality is necessary for a positive key rate. However, we also show that it is not sufficient. Concretely, the (3,2)-parity-oblivious random access code, known to yield secret key against a quantum eavesdropper, yields none certifiable key rate against a post-quantum one. Conversely, operational equivalences with no analogue among no-signaling constraints can strictly increase the asymptotic key rate. We illustrate this with the prepare-and-measure version of the standard CHSH protocol with an aligned key-generating setting.

This article is the third in a series investigating ill-posedness in critical Sobolev spaces for Camassa-Holm-type equations. Following results for the $b$-Novikov and Fokas-Olver-Rosenau-Qiao equations, we study the hyperelastic rod equation at the critical Sobolev regularity $H^{3/2}(\mathbb R)$. We prove ill-posedness at this endpoint, complementing the local well-posedness theory in $H^s(\mathbb R)$ for $s>3/2$ and the recently established norm-inflation result for $1<s<3/2$. The nature of the critical ill-posedness depends on the parameter $\gamma$. For $0<\gamma\leq3$, we construct arbitrarily small smooth initial data whose corresponding solutions develop finite-time blow-up, with the maximal lifespan tending to zero with the size of the initial data. For $\gamma<0$ or $\gamma>3$, we construct solutions with arbitrarily small $H^{3/2}$ initial data whose $H^{3/2}$ norm becomes arbitrarily large in arbitrarily short time.

The discrete geometry on an $n$-dimensional simplicial manifold are studied, in order to incorporate torsion into Regge calculus. In each simplex, the edge vectors are assigned to the edges to encode the information of their lengths and directions. In addition, the holonomies along the curves across the interfaces of two adjacent simplices are represented by the internal gauge group elements. The torsion manifests itself as the difference between an edge vector on an interface belonging to one simplex and the parallel transported edge vector, via the holonomy, of the same edge but belonging to the adjacent simplex. The simplicial Einstein–Cartan actions are then constructed as functions of the edge vectors and holonomies in three and four dimensions with Euclidean and Lorentzian signatures, respectively. It is shown that they return to the corresponding Regge actions for the torsion-free cases. The variations of the 4-dimensional discrete action with respect to the edge vectors and holonomies, respectively, give two equations of motion. It is shown that the former is consistent with the corresponding equation in the continuum theory, and the torsion-free holonomies satisfy the latter equation as in the continuous case. Thus, the resulting theory on the simplicial manifold can be regarded as a discrete analogue of Einstein–Cartan theory.

In partially observable reinforcement learning (RL), a later observation can make stored information obsolete or change what it implies for the next decision. Memory architectures and benchmarks for RL mostly test retention, the ability to keep information unchanged until it is needed. We formalize two further requirements. Rewriting sets the decision-relevant content to a value independent of the old one, and experience fusion transforms the old content by a rule that a later observation specifies. For tasks built from such updates, we count the memory states that a solution needs, and several baselines reach their lowest success rates on compositions that need more states. We introduce ALER (Adaptive Learnable Experience Rewriting), an agent that pairs an LSTM with a slot memory. An independently addressed Gumbel-Softmax write that concentrates its weight on one slot overwrites that slot, and a learned gate fuses the retrieved content with the recurrent state before the policy and value heads. We also introduce Rune-Mazes, three environments in which rune observations invert, cancel, reset, or repeat updates of a hidden cue under vector and pixel observations. Against seven baselines, ALER reaches a success rate of at least $0.82$ in all sixteen Endless T-Maze configurations and at least $0.99$ on all five Rune T-Maze compositions, and it has the highest mean success rate on four-branch Rune Multi-Corridor with an Invert rune. On pixel-based Rune MiniGrid Memory, it has a higher mean success rate than PPO-LSTM in eight of ten configurations. Project page: https://quartz-admirer.github.io/ALER-Adaptive-Learnable-Experience-Rewriting/.

Let $X$ be a nodal curve, and $G$ be the graph dual to $X$. A stability condition on a graph $G$ is an assignment of integers to the biconnected subsets of vertices of $G$ satisfying some desired properties. Stability conditions yield degeneracy sets, which are certain collections of biconnected subsets of $G$, but degeneracy sets can also be described without a reference stability condition. We show that the degeneracy sets of a graph $G$ correspond to the single-element extensions of the graphic matroid $M(G)$. Some single-element extensions of the graphic matroid $M(G)$ can be oriented to be single-element extensions of the oriented graphic matroid $\mathcal{M}(G)$. Single-element extensions of oriented matroids are in bijection with subdivisions of the Lawrence polytope of the dual oriented matroid. We construct stability conditions on $G$ from subdivisions of the Lawrence polytope of the cographic matroid $\mathcal{M}^\ast(G)$. Finally, we show that any degeneracy set $\mathcal{D}$ of a graph $G$ that has the set of all biconnected subsets of $G$ as a lower bound in the poset of degeneracy sets corresponds to an orientable extension of $M(G)$.

An autonomous cyber defender trained with reinforcement learning (RL) is typically tied to the network on which it was trained, limiting its ability to generalize as network scale changes. Hierarchical RL reduces decision complexity by separating strategic targeting from tactical execution, but it does not eliminate this retraining dependence. We investigate whether frozen, zero-shot large language models (LLMs) can provide retraining-free control in hierarchical cyber defense and how performance changes as LLM control is extended from planning to execution. We formulate a controller-agnostic planner-executor hierarchy in which the planner selects a subnet to defend over a fixed horizon and the executor selects defensive actions within that subnet. Using the high fidelity Cyberwheel environment, with its built-in automated red team agent mapped to the MITRE ATT&CK framework, we compare RL+RL, LLM+RL, and LLM+LLM configurations using six models ranging from 3B to 70B parameters, including two cybersecurity-specialized models, across small, medium, and large networks. Replacing only the planner with an LLM yields limited gains as network size increases. In contrast, extending LLM control to execution produces notable improvements for sufficiently capable models. For instance, a frozen general purpose 70B model holds successful lateral movement to approximately 1% of steps and attacker impact near zero across all three network scales using the same model weights, while the RL baseline is retrained for each scale. Our results show that sufficiently capable frozen LLMs can maintain strong defensive performance across the evaluated network scales without task-specific retraining, while also indicating that strong tactical execution is important to realizing the benefits of LLM-based control.

Time-series foundation models such as Google's TimesFM-3 forecast series they were never trained on, but whether they should replace classical methods such as ARIMA and exponential smoothing is hard to settle on public benchmarks, because few benchmark datasets are absent from every model's pre-training corpus. This exploratory study compares TimesFM-3 with classical methods on data generated from nine known processes, which the model cannot have seen. Across 7200 series and a pre-registered protocol with a 12-step horizon, neither family dominates. TimesFM-3's error is never more than 1.31 times that of the best method in a scenario, whereas every classical method's is at least 2.2 times somewhere; the largest classical failures occur with only two seasonal cycles, where the automatic methods fall back to non-seasonal models. With four or more cycles automatic ARIMA is 21-24% more accurate than TimesFM-3 on seasonal ARIMA data. Among five foundation models this robustness is specific to TimesFM-3, and it holds only at short horizons: at steps 25 to 48 the three foundation models tested there forecast a decline on a saturating process whose level stays flat, and automatic ARIMA has the smaller worst case elsewhere. On intermittent demand TimesFM-3 shows no advantage over simple benchmarks built from the history. Randomised parameters, heavy-tailed noise and outliers leave these patterns qualitatively unchanged; on the official test period of 1000 M4 monthly series TimesFM-3 is level with the fifth- to seventh-ranked M4 entries, behind the four best, and on 101 macroeconomic series observed after the models' documented training data it is statistically level with the classical methods. The study characterises the behaviour of a black-box model, not the reasons for it, and its findings are conditional on the processes and horizons examined.

Principal component analysis and factor analysis are foundational to the study of high-dimensional data, yet their efficacy depends entirely on correctly identifying the latent dimension $r$. While parallel analysis (PA) is widely regarded as the gold standard for this task, its performance suffers in high-dimensional regimes consisting of mixed signals and/or heavy-tailed error distributions. This arises because under the preceding conditions, column-wise permutation mechanism (or any modern variations) applied to the data matrix $\textbf{X}$ fails to destroy the signal, leading to an inflated null spectrum. To address this, we propose a nonparametric solution. We show that the latent dimension is uniquely preserved in the degree of order within the rank-transformed data matrix, which restores spectral exchangeability under permutation. We prove the consistency of Rank PA under relaxed conditions, requiring neither the pervasiveness of factors nor restrictive fourth moment assumptions. We establish that Rank PA can consistently estimate the number of factors up to the maximum of $O(n^{1/3 - \tau})$ where $\tau >0$. The robust performance of our remedy is demonstrated through numerical simulations and an analysis of the FRED-MD dataset.

The corpus of worldwide news is important for natural language processing, knowledge graphs, large language models, and other technical efforts. Additionally, this corpus is important for understanding the people, places, organizations, and events that interact in real-time every day. This paper compares two news datasets used for these tasks today, namely the Global Database of Events, Language, and Tone (GDELT) and Common Crawl News. Our research highlights the strengths and limitations of each dataset, analyzing their content and coverage. Notably, while GDELT relies on broadcasts, prints, and web news from across the globe, Common Crawl focuses on news sites from around the world gathered through web crawling. Our analysis revealed considerable differences in where the two datasets gather their news sources.

Activation memory, not compute, limits CNN inference on constrained hardware such as microcontrollers. Direct in-place convolution removes the dual-buffer cost, but the memory-optimal formulation of Gural and Murmann assumes valid padding, unit stride, unit dilation, and odd square kernels, and needs a non-sequential traversal costing $2\times$ inference time in transposes. We identify two regimes in which their published closed form does not hold: (1) an under-allocation of exactly $(k-1)C_{in} \bmod (C_{out}-C_{in})$ scalars, active on every convolutional layer of their own deployed network and manifesting as a silent corruption of still-live input; (2) an unbounded overestimate, up to $2{,}432\times$, once the critical leg leaves the output grid. We correct both and generalize to arbitrary stride, dilation, padding, and rectangular kernels. We then propose Right In-Place (RiP) convolution, a bit-identical operation in which every layer reads its input right-aligned in a shared workspace and writes its output left-aligned from index zero. The debt is piecewise affine in the output pixel index, so evaluating its breakpoints in $O(1)$ yields the minimum safe gap without enumerating the output grid, with row-major access preserved. Across $10{,}000$ random layers RiP produced no corruption, and across 84 convolutional layers from 25 architectures it matches the herringbone workspace exactly on 58 and within 5% on 81, using 24.8% less memory than dual buffering on average. Written into TinyEngine's kernels and deployed to a Raspberry Pi Pico 1 and Pico 2, it cuts peak activation memory across eleven MCUNet models by 12.5 to 33.3% at unchanged cycle counts and bit-identical outputs, raising the number of models that fit the Pico 1's 256 KB SRAM from six to nine.

Universal machine-learning interatomic potentials now reach held-out energy and force errors so small that they no longer predict how a model behaves in simulation. Here we show that a potential can be graded without any reference calculation, against properties the exact Born-Oppenheimer surface satisfies by mathematical or physical necessity. We organise these properties into four families, symmetry and invariance, self-consistency and integrability, statistical-mechanical equilibrium and regularity, and turn them into fourteen inexpensive probes, each certifying against a target that is exact and independent of chemistry and reference method. Every probe therefore returns an absolute, architecture-comparable score from the energies, forces and stresses a potential already exposes, and reveals failures a fixed test set cannot. Applied to 64 pretrained potentials, the suite resolves two orders of magnitude of certified quality at comparable reported accuracy, and flags a fine-tuned potential whose reference errors improve while its surface degrades until molecular dynamics fails. An interactive leaderboard of all 64 models is available at https://jhaens.github.io/pescert-bench.

In this paper, in addition to establishing several new sufficient or equivalent conditions for the well-known Separable Quotient Problem (SQP), we show that every infinite-dimensional Banach space lacking the Lotz property admits a separable quotient, thereby revealing a surprising connection between the SQP and semigroup theory.

People are increasingly delegating tasks to AI agents, and those agents are increasingly encountering other people's agents over shared resources such as a codebase, a calendar, or a budget. When each agent acts for a different user with different goals, coordination often fails, and the group ends up worse off than if a single agent had acted for everyone. We study this multi-user, multi-agent setting across five frontier models and 77 scenarios in four environments: an API key environment in which agents share a compute budget, a clinic in which they share a calendar, a personal assistant environment in which they share a group order or booking, and a merge queue in which they share a release cutoff. In each scenario, we compare a single agent that serves every user (a coordinator) to a team in which each agent serves one user, with and without a communication channel between the agents. Teams deliver worse group outcomes than the coordinator in every environment: without a channel, they completely collapse in two environments, and even with one, coordination overhead creates substantial gaps. For example, in the personal assistant environment, the coordinator fulfills a targeted user request about twice as often as teams. We identify distinct behaviors associated with this poor group-level performance, including stalling as teams grow, overriding each other's actions, and fabricating claims. We find effective but environment-specific mitigations, such as a team lead, explicit procedural instructions, and a platform check that makes an agent read its peers' messages before committing. We will release the API key, clinic, and personal assistant environments as MAMUBench, comprising 74 scenarios for evaluating multi-user, multi-agent coordination.

Aesthetic image cropping aims to identify the optimal crop of an image in terms of aesthetics and composition. While supervision based on annotated data is fundamental, the field has been hindered by a long-standing problem: existing datasets suffer from (1) human subjectivity and (2) rigid discreteness confined to fixed sampling grids. These flawed annotations not only limit the accuracy and generalization of trained models but also severely distort fair evaluation. To overcome this, we propose to model human cropping preference as a multi-peaked, continuous, and sharp field over the crop space. We introduce the Continuous Preference Field (CPF), which recovers a dense preference landscape from discrete annotations through (1) peak clustering, (2) off-lattice refinement, (3) negative shaping, and (4) field assembly. Based on this, we train CPIC, a VLM-based cropping model optimized via GRPO with the CPF reward, which overcomes template collapse, achieving state-of-the-art performance and exceptional out-of-domain generalization. Finally, to resolve the long-standing benchmark evaluation crisis, we introduce CPICD, a comprehensive recalibration of existing ground-truth boxes. By leveraging the CPF to correct grid-bound artifacts across mainstream benchmarks, CPICD establishes a rigorous and reliable foundation for future cropping research. Extensive experiments and user studies demonstrate the superiority of our CPF, CPIC, and CPICD. Code, model, and data are available at https://github.com/zzqingz/CPIC.

We study the anatomy of commodity risk by distinguishing micro, market-level, and economy-wide sources. We develop a two-stage "divide-and-conquer" framework that allows sensitivities to these risk sources to vary across commodities while treating economy-wide risk as latent. The first stage uses defactored instrumental-variable estimation to recover commodity-specific sensitivities to micro and market conditions. The second combines principal components with high-dimensional variable selection to identify an observable representation of macro-financial risk. We then construct Risk Intensity Indices (RIIs), which combine estimated sensitivities with prevailing risk conditions to quantify the relative importance of each risk source on a common scale. Market risk is the largest component on average, accounting for about two fifths of total risk intensity and more than half for energy commodities. Risk intensity is also highly concentrated across individual commodities: the top 20% account for approximately half of micro and market risk intensity, whereas macro risk is more broadly dispersed. The composition of risk varies substantially across sectors and over time, with market risk becoming particularly prominent during episodes of commodity-market stress. Micro and market RIIs also contain information about future volatility and absolute returns. These findings provide investors, risk managers, and policymakers with a diagnostic of where commodity risk is concentrated, which risk layers are most important, and how their importance changes over time. More broadly, our divide-and-conquer framework provides a flexible approach to decomposing layered risk in settings where common risk is latent.

In collaborative foundation model fine-tuning, client data is rarely homogeneous. Instead, clients typically possess unknown mixtures of distinct data distributions, or tasks. Conventional federated learning primarily addresses heterogeneity across clients without explicitly resolving latent task mixtures within each client. We study this setting as compound heterogeneity, where data is heterogeneous both across and within clients. We study adaptation over a common frozen representation and show that, when tasks share the same feature geometry, the optimal model for a client's task mixture under squared loss is a convex combination of the optimal models for its underlying tasks. Thus, a single locally trained model represents the client's overall task mixture, while individual inputs may be drawn from different underlying task distributions. This motivates routing inputs to specialized experts, and we show that, when the task optima form a simplex, task-aligned routing achieves lower risk than any single adapted model for genuinely mixed clients. With access to a small set of task-labeled public samples, we derive a convex program to recover task experts and match them to their corresponding tasks. Our routing analysis shows that effective specialization requires input-dependent expert selection aligned with each client's task mixture. Motivated by this analysis, we propose FedSEE. Across our experiments, FedSEE avoids the negative transfer observed in the evaluated baselines and improves performance by 2.9 points overall and 3.7 points for the worst-served quartile.

It is well-known that every n-qubit unitary can be decomposed into a product of controlled single-qubit unitaries. We show that the control conditions in such a decomposition naturally define a quasi-process function, an object that originally came up in the study of indefinite causal order. Here quasi-process functions yield a representation of every n-qubit unitary in a canonical form, namely, as a quasi-process function unitary (qPFU). We use this representation to show how quantum nonlocality without entanglement (QNLWE) can be understood in terms of unitaries that encode the trade-off between causal order and locality noted in R. Kunjwal and Ä. Baumeler, arXiv:2202.00440. Specifically, we obtain two sufficient (but not necessary) criteria—mutual exclusivity of control conditions (MECC) and Unambiguity—for a unitary to be a process function unitary (PFU), i.e., it admits a qPFU representation where the quasi-process function is a process function, an object that can model paradox-free classical causal loops. We also show the inequivalence of these criteria, neither implying the other. We then show that the circuit depth of MECC unitaries admits a characterization in terms of clique-partitioning problems on Hamming graphs and use this correspondence to provide bounds on the circuit depth of n-qubit MECC unitaries. Finally, we show how MECC unitaries provide a recipe for constructing a unitary purification of the associated process function, thereby providing a large class of unitary processes of interest in quantum causality. The generality of this causality-inspired representation of unitaries makes it a versatile tool for analyzing the structure of quantum circuits in other domains.

Cosmological simulations allow us to study galaxy-progenitor stream populations, allowing comparisons with observations to constrain galaxy formation physics. Previous work found that present-day galaxy-progenitor streams in FIRE have larger pericenters than observed in the MW, which might indicate that they tidally disrupt at larger distances, where the Galactic tides are weaker, than in the MW. One possible explanation is that they tidally disrupt too quickly because of limited resolution. At lower resolution, low-mass dwarf galaxies can be puffier and more susceptible to disruption. We build and compare dwarf-galaxy stream catalogs in two FIRE-2 cosmological zoom-in simulations with identical initial conditions (m12i) but different particle mass resolution (7070 $M_\odot$ and 880 $M_\odot$). The higher-resolution run yields a larger dwarf galaxy stream population (22 streams versus 14) using the same cutoff of 100 star particles, primarily because it resolves lower-mass streams (down to $\sim10^{5}$ $M_\odot$). Both simulations form streams at galactocentric radii < 50 kpc, but the higher-resolution Triple Latte simulation produces more streams at these small radii. We find that dwarf galaxies in the higher-resolution simulation begin disrupting at smaller pericenters than in the lower-resolution simulation. By cross-matching stream candidates, we identify counterparts for $\sim$89$\%$ of the systems in the lower-resolution simulation. For matched systems that remain on similar orbits, the higher-resolution counterparts remain intact for longer. The orbital properties of streams in the higher-resolution simulation are broadly consistent with the MW dwarf-galaxy stream population, in contrast to the lower-resolution simulation.

In a district-based election, N voters are partitioned into k districts, and each voter votes for one of m candidates. Each district elects a winner using the plurality rule (i.e. the candidate getting the largest number of votes is declared the winner, breaking ties as per some fixed rule), and the overall winner is determined by applying plurality to the district winners; we assume that there is a unique winner amongst the district winners. The margin of victory of such an election is the minimum number of votes that must be altered so that the current winner ceases to be the unique district winner. We study the problem of predicting the winner of a district-based election in the query complexity model, where one has query access to individual votes. The objective is to minimise the number of queries. This setting captures exit polling, where queries correspond to interviewing voters, and is closely related to problems in query complexity and property testing. Assuming that the margin of victory of the election is at least eps N, Dey, Kar and Sanyal (AAMAS 2023) gave algorithms for the case of two candidates with error probability del and query complexity tilde{O}(1/eps^6 log^2 1/del), which improves to tilde{O}(1/eps^4 log^2 1/del) under the additional assumption that district populations are balanced. Our main result is an adaptive randomised algorithm that, for an arbitrary district-based election and any error parameter del, with probability at least 1-del, predicts the winner correctly using tilde{O}(1/eps^2 log m/del log 1/del) queries. In particular, we improve the bounds of Dey et al. for arbitrary district populations and extend their results to any number of candidates. Furthermore, for constantly many candidates, our algorithm nearly matches a lower bound of Omega(1/eps^2 log 1/del) on the query complexity that holds even for two candidates and a single district.