Papers

We prove a longstanding conjecture of András Bezdek on the stability of plank covers of the planar disk $D$. If a sufficiently small disk $\varepsilon D$ is removed, then every finite family of planks covering the resulting annulus can be rearranged to cover the whole disk. For the proof, we establish a stronger dimension-independent stability theorem. If $K \subset {\mathbb R}^n$ is an origin-symmetric convex body for $n \geq 2$, and the uncovered region lies in $\varepsilon K$ and has volume $v$, then the excess overlap is at least $c v/\varepsilon$, where $c>0$ is an absolute constant. The proof introduces a pruning procedure and a flattening mechanism. Pruning peels away curvature through local removals whose cumulative effect is guaranteed to be substantial. Flattening trades straightness for bounded multiplicity and geometrically exposes the overlap that pays for each pruning step.

Accurately modeling $\beta$-decay half-lives of highly neutron-rich nuclei is a critical challenge for understanding $r$-process nucleosynthesis and the resulting kilonova light curves. We introduce a data-driven approach to model $\beta$ decay that uses Mixture Density Networks (MDNs) trained on the latest experimental measurements to extrapolate the half-lives of neutron rich nuclei. Unlike standard deterministic regression, the MDN directly parameterizes the probability distribution of the half-lives, providing intrinsic (aleatoric) uncertainty. We contrast the intrinsic uncertainty of a single model with the range of models produced by using random variations of the input training data samples. While all models show excellent agreement with experimental data, we find that varying the training data sets produces a wide range of extrapolations, particularly along data-poor regions such as the $N=126$ isotonic chain. We then propagate a selection of model results into $r$-process network calculations to evaluate their impact on isotopic abundances. Finally, by coupling the resulting isotopic abundances with thermalization efficiencies, we translate the effective nuclear heating rates into bolometric light curves. We compare the ranges of outcomes produced from the intrinsic uncertainty of a single model to those produced by multiple independent models trained on different training data sets.

The theory of Ramanujan sums has been playing a fundamental role in several subfields of mathematics. They appear in the theory of special values of zeta functions, spectral graph theory, representation theory, and analytic number theory. Recent work has shown that classical Ramanujan sums can also be interpreted as a super-Fourier transform via the theory of supercharacters for the ring $\mathbb{Z}/n$. In this article, building upon our recent work on supercharacters over an arbitrary finite Frobenius ring, we explore additional arithmetical properties of Ramanujan sums. Our approach provides a unified framework that generalizes various results in the literature regarding these sums. As a by-product, we also describe a new criterion for determining when a finite commutative ring is Frobenius, which could be of independent interest to the algebra community.

We study planar openbooks obtained by lifting braids through branched covers of D^2, together with the quantum operations in the Ising representation. We give a criterion for the Stein fillings to be Calabi-Yau (CY). Among positive factorizations of a fixed monodromy, a CY one has minimal length and its filling minimizes \chi and b_2. Applied to Baykur's recent examples, this gives a planar contact 3-manifold with infinitely many non-homeomorphic CY fillings, the cover there having degree \ge 6. At degree 4 a single CY filling forces every filling to be CY, and in one case the filling is unique; at degree \le 3 it is unique, and CY under a mild condition. Admissible cuts of D^2 decompose the openbook into subopenbooks. Every positive factorization then localizes, and the CY condition holds exactly when it holds locally. The state space is then a direct sum of tensor products indexed by the compatible parity choices, with at most two qubits per factor when the pieces have degree $\le 4$, the same range in which the CY condition depends only on the monodromy. For one degree 4 cover, the points, lines and flags of the two-qubit doily are realized by its subopenbooks. Fifteen liftable braids there share the same quantum operation and Stein filling, and are separated only by the subopenbook systems they admit. A choice of tensor-product structure is thus carried by the lift, not by the braid group representation, suggesting a link between contact topology and quantum entanglement.

Large language models (LLMs) are increasingly deployed in settings where responses must reflect multiple, potentially interacting social norms and human values. Activation steering offers a lightweight alternative to training-based alignment by modifying internal activations at inference time. However, prior human-value steering methods have largely considered values in isolation, while direct composition of multiple directions relies on fixed intervention strengths that cannot respond to the model's evolving internal state. Motivated by this key observation, we introduce AIMES, a framework for adaptive multi-value activation steering. AIMES constructs layer-specific bipolar directions for moral-foundation values and uses intermediate-layer vocabulary readouts as online observers. An observer-guided controller then adapts the strength of each requested value intervention at every decoding step based on its current observed state, without training a separate value-state estimator. Across multiple instruction-tuned model families, value combinations, and intervention depths, we find that multi-value controllability varies across both value combinations and intervention locations. Compared with fixed joint steering and prompt-based steering, AIMES shows depth-dependent advantages that are broadly supported across two independent evaluators, with some variation in the precise depth at which specific control effects emerge. These advantages come with smaller realized activation-space interventions than fixed-joint steering and comparable response quality. Overall, our results suggest that online observer feedback can provide lightweight, state-aware adaptation for single-pass multi-value steering.

We present POIL, a point-based one-shot imitation learning framework with stable dynamical systems. While one-shot imitation avoids collecting extensive demonstrations, successful one-shot manipulation requires not only transferring a demonstrated trajectory to a novel object but also executing it robustly under changing scene conditions, grasp configurations, and external disturbances. POIL addresses both problems through a shared representation: a set of 3D points on the object's functional part, used jointly for trajectory transfer and closed-loop execution. The one-shot transfer from the demonstrated trajectory is enabled with point correspondences. POIL grounds the shared functional part with a multi-modal large language model, and transfers the trajectory across viewpoint, pose, and object category changes. During execution, multi-view tracking observes the same points online, and Point-set BCSDM drives them in closed loop by projecting per-point velocities onto a single rigid-body twist computed from the tracked points alone. This extends stable dynamical models from an SE(3) pose to a point set without requiring a known 3D model or pose estimator. We show that at the goal the controller becomes a gradient flow on the classical SO(3) potential, so its terminal phase inherits the almost-global convergence of that potential under a rigid-object assumption. Across simulation and real-robot experiments, POIL transfers a single demonstration across object category, grasp pose, and goal geometry, while recovering from external disturbances during execution. Project page: https://sangminkim-99.github.io/poil

Detailed Geant4 simulation of calorimeter showers is the largest single computing cost of collider experiments, and the High-Luminosity LHC will need about ten times more simulated events than are currently produced. We summarise recent progress in generative point cloud fast simulation for highly granular calorimeters. CaloClouds3 generates photon (electromagnetic) showers, is geometry-independent, and runs on average about 120x faster than Geant4 on a single CPU. CaloHadronic uses transformer attention to extend the point cloud diffusion approach to pion (hadronic) showers spanning the electromagnetic and hadronic calorimeters. AllShowers unifies twelve particle types in a single model with far fewer parameters than the specialised baselines while matching or exceeding their fidelity. We close with cross-geometry transfer learning, which needs two to three orders of magnitude fewer training showers while preserving the generative performance.

Multimodal misinformation is increasingly crafted to look convincing by pairing a textual claim with an image that appears to "prove" it. Yet in practice, building effective detectors often hinges on a small set of design choices that are rarely examined in a controlled way. In this paper, we conduct a large-scale study of multimodal design choices for misinformation detection with over 3,375 experiments- spanning three benchmark datasets and a broad range of pre-trained vision and language backbones. Through systematic comparisons and targeted robustness analyses, we distill practical guidance on which design choices help, when do they fail silently, and what aspects of the pipeline most strongly shape model behavior, answering 4 key Research Questions (RQs). We aim to provide a reliable foundation for designing stronger and more dependable multimodal misinformation detection systems, thus contributing to the broader research community.

We study the three-dimensional (3D) incompressible Euler equation linearized about a smooth compactly supported stationary flow of Gavrilov type. Baldi's volume-preserving action–angle chart allows a reduction of the high-frequency amplitude equation on invariant tori. Using Baldi's chart, we identify a critical torus ensured by the compact cutoff. We show that, for a fixed axisymmetric covector with sufficiently large normal component, the reduced trace-free periodic cocycle on this torus is hyperbolic and has a positive Floquet exponent. A localized geometric-optics construction then lifts this Floquet instability to the linearized Euler equation, yielding exponential essential-norm growth in space $H^s$, $s\geq 0$.

Graph states are a fundamental resource for measurement and fusion-based quantum computing, quantum networks, and sensing. Preparing them in a photonic system deterministically is, in principle, possible, but finding efficient schemes to prepare them was a long-standing problem addressed recently. Additionally, heuristic optimization schemes for reducing the required number of two-qubit gates were developed. However, the problem of reducing the number of emitters by optimizing the emission ordering was not addressed, due to its computational complexity, as it is connected to a well-known NP-hard problem from graph theory, the linear rank width computation. In this work, we focus on developing heuristic polynomial algorithms to reduce the number of emitters required. In total, we propose four distinct algorithms, which demonstrate up to $30\%$ emitter reduction on random graphs. Furthermore, we provide numerical and statistical evidence that the combination of our optimization schemes with the preexisting algorithms for optimizing the two-qubit gates of the preparation protocol can further reduce them by around $20\%$. Finally, we examine the developed algorithms for various useful graph state families, such as graphs useful for measurement-based quantum algorithms, and cluster states and graph codes used for quantum error correction, to determine the performance of each algorithm.

The encoding of 3-uniform edge-ordered weighted hypergraphs into multiqubit quantum states is proposed. We derive an analytical expression for the geometric measure of entanglement of hypergraph quantum states and establish its relation to hypergraph properties. The results establish a direct connection between the local structure of hypergraphs, hyperedge weights, and the entanglement of the corresponding quantum states. For equally weighted hypergraphs, it is shown that the entanglement depends explicitly on the vertex degree. The dependence of entanglement on hypergraph parameters is studied analytically and using quantum computing. Hypergraph states associated with chain, star, regular lattice, and binary-tree hypergraphs are examined, and their entanglement is quantified using quantum computing. These results open up the possibility of studying hypergraph properties with quantum programming.

The most general strategy for estimating parameters of a quantum channel is to probe the channel many times sequentially with the help of coherent ancillary systems. However, finding the strategy that minimizes the mean-squared error (MSE) is a notoriously hard problem, requiring optimization over the initial probe state, intermediate evolutions, final measurement, and classical postprocessing. In this work, we study how the minimum attainable MSE can be bounded from below in this setting. We propose a lower bound for the MSE of any physically implementable sequential sensing strategy that applies to estimating parameters of any physical quantum evolution, including quantum channels probed over multiple time steps under temporally correlated noise and general temporally correlated quantum processes. We show that this lower bound can be formulated as a semidefinite program (SDP). Our bound generalizes the Nagaoka-Hayashi lower bound (that applies to parameter estimation tasks with separable measurements) to the regime where the quantum channel can be accessed sequentially multiple times. For regular finite-dimensional single-parameter sequential estimation tasks admitting an optimal strategy, we show that our SDP bound is tight and provide a procedure to obtain a locally optimal strategy from the SDP optimizer. Using our optimization bound, we compute the MSE lower bound and derive new protocols for several other physically motivated problems including error-corrected quantum sensing under an unknown correlated noise model, multiparameter estimation, and estimation with qudit systems.

Evaluating explainable Artificial Intelligence (XAI) methods is a challenging task due to the lack of reliable evaluation procedures and, in particular, the absence of ground truth explanations. In the literature, existing evaluation approaches typically assess explanations by measuring their fidelity with respect to the predictions of a black-box model. However, such evaluation strategies only quantify the degree to which an explanation reproduces the model's output, without ensuring that the explanation correctly reflects the underlying decision process. As a consequence, different explanations may achieve similar fidelity scores while providing inconsistent or misleading interpretations of the model behavior. In this paper, we propose a framework for the evaluation of XAI methods based on synthetic ground truth. The proposed approach relies on controlled interventions to generate synthetic datasets in which the importance of input components can be determined by design. This enables the construction of ground truth explanations that are directly aligned with the behavior of the model under analysis. The framework is instantiated across three data domains, namely binary images, tabular data, and time series, allowing a comprehensive assessment of explanation methods in heterogeneous settings. Experimental results obtained by evaluating nine widely used XAI methods show significant limitations in current techniques and highlight the importance of synthetic, intervention-based benchmarks for a reliable assessment of explanation quality.

This paper presents an end-to-end causal machine learning (ML) pipeline designed for real-world applications with continuous treatments. The proposed framework consists of six sequential steps: dimensionality reduction, causal identification, positivity assumption violation handling, estimation, refutation and evaluation, and policy optimization. We introduce practical contributions not currently available in existing causal ML toolkits, specifically: (1) a method for detecting and quantifying positivity violations in continuous treatment settings (2) a novel, scalable two-stage dimensionality reduction framework tailored for causal inference with high-dimensional data; (3) the adaptation of sensitivity analysis and estimation methods originally designed for binary treatments to the continuous treatment space and (4) an end-to-end integration of these components into a modular, reproducible workflow. These innovations address real-world challenges in causal inference that are often not covered in theoretical frameworks but frequently encountered in industrial applications. The methodology is validated with a synthetic dataset inspired in a real-world financial debt collection use case, however its design can be applied to analogous problems across different industries. Results demonstrate that the proposed methodology offers a more computationally efficient approach and produces less biased estimates compared to standard methods for problems with continuous treatment and high-dimensional data. A fully functional GitHub repository with documented code and numbered notebooks is made available ensuring reproducibility and practical implementation. The pipeline presented is intended to contribute to closing the gap between academic approaches and practical application in industry contexts where causal ML can be highly beneficial such as the financial sector.

Real-time tiny object detection in aerial imagery is constrained by the weak spatial evidence of very small objects and the loss of high-resolution detail in lightweight detectors. This study presents Cross-Scale Channel-wise Knowledge Distillation (CSCWD), a training-time framework that transfers high-resolution spatial representations from a YOLO11m-P2 teacher to a compact YOLO11n student without altering the student's inference architecture. Unlike conventional same-scale feature distillation, CSCWD transfers supervision from teacher P2 to student P3 after feature alignment while retaining same-scale distillation at deeper pyramid levels. Under the unified seven-sequence Drone-vs-Bird validation protocol, YOLO11n-CSCWD achieves 50.17% mean average precision at an intersection-over-union threshold of 0.5 (mAP@0.5) and 59.73% recall, improving the matched CA-YOLO11n baseline by 2.92 percentage points in mAP@0.5 and 3.55 points in recall. Cross-scale alignment further increases mAP@0.5 by 2.09 points over the corresponding same-scale channel-wise distillation configuration. In zero-shot evaluation on DUT-Anti-UAV, mAP@0.5 increases from 48.29% to 50.06% without target-domain fine-tuning. This domain was included because its challenging small targets make low-latency, computationally efficient detection particularly relevant. On Raspberry Pi 5 using NCNN-FP16 at 640x640 resolution, the 2.58-million-parameter student achieves 50.32% mAP@0.5 at 82.32 ms mean wall-clock latency, or 12.15 frames per second, while retaining essentially the same runtime and memory requirements as the matched baseline. The results support cross-scale distillation for improving tiny-target detection without increasing inference-time model complexity.

Many cloud-based applications are organized as loosely coupled microservices, where invoking a service's API triggers a cascade of APIs across many services and leads to inter-service exchange of API parameters and output responses. Current tools for monitoring microservice safety properties have limited expressiveness for properties that describe the flow of data through API calls. To this end, we present SafeNom, a specification and monitoring framework for microservices based on nominal languages. SafeNom policies can express both the desired order of API calls and how the data carried in requests and responses should or should not flow between the APIs. Policies are enforced using a nominal automaton-based distributed runtime monitor which can be applied in a blackbox and non-invasive manner, without access to the service implementation and without making changes to the service implementation. Our experiments show that our monitor can efficiently enforce rich data-aware properties while incurring minimal latency overhead, on the order of a few milliseconds.

Selecting informative camera views is critical for efficient training and adaptive refinement in 3D Gaussian Splatting, where each observation significantly influences model parameters. However, information-driven view-selection strategies can require repeated evaluations of expensive information-gain oracles as the number of candidate views increases. We propose LiTe-GS, an oracle-efficient method for next best view selection in 3D Gaussian Splatting. LiTe-GS reduces the number of information-oracle evaluations by performing randomized subset evaluation of candidate views rather than exhaustively scoring the full candidate pool. The resulting approach achieves expected $O(M\log(1/\epsilon))$ oracle complexity with respect to the number of candidate views $M$, independent of the selection cardinality $K$, while providing an explicit trade-off between oracle efficiency and approximation quality through $\epsilon$. We provide theoretical guarantees on oracle complexity and approximation performance under the proposed selection scheme. Experiments on Blender and Mip-NeRF 360 demonstrate that LiTe-GS maintains reconstruction quality comparable to Fisher-information-based baselines while substantially reducing the number of Fisher-oracle evaluations across different acquisition settings.

We study jet formation from a magnetically diffusive, sub-Keplerian accretion disk threaded by different magnetic field configurations. We present two dimensional, axisymmetric MHD simulations in spherical coordinates employing the PLUTO code. Our results show that the reference simulation, in which the disk is threaded by an initial large scale poloidal magnetic field, reaches a quasi steady state and produces a well collimated jet.Both kinetic and magnetic torques are significant and evolve smoothly over time. The accretion process is established smoothly, with about 26% of the accreting material being delivered into the outflow. Alternatively, applying a purely stellar dipolar magnetic field produces weak and unstable outflows, characterized by inefficient angular momentum transport and strongly suppressed accretion. Additionally, in simulations where the stellar dipolar magnetic field is present alongside the disk's initial poloidal magnetic field, the disk magnetic field predominantly stabilizes the system. It also facilitates angular momentum transport from the disk and sustains a significant outflow mass flux. Furthermore, we performed additional simulations that explicitly include an inner gap associated with the disk truncation region. These simulations were designed as a robustness test of the treatment of the innermost disk region and confirm that our main conclusions are not sensitive to the presence of this gap. Overall, our findings demonstrate that although the magnitude of stellar dipole plays an important role in the evolution of the accretion-ejection structure, the large scale poloidal magnetic field threading the disk is essential for launching and sustaining a stable and well collimated magnetized jet.

Existing in-context reinforcement learning methods mainly pretrain Transformers with supervised behavior-prediction objectives. This enables task inference from context, but makes the learned policy strongly depend on the quality of offline actions: when trajectories are weak or suboptimal, imitation itself becomes a biased learning signal. We propose Q-Target Pretrained Transformers (QTPT), which keeps the context-conditioned Transformer architecture but replaces behavior cloning with a Bellman-style Q-target objective. QTPT therefore learns to use rewards and transitions in the context to estimate action values, rather than simply imitating the behavior policy. We theoretically analyze QTPT in stochastic linear bandits and finite-horizon MDPs, showing stronger robustness to data quality than supervised pretraining. Empirically, QTPT improves over supervised behavior prediction on controlled RL benchmarks with random or suboptimal data, and we examine extensions to D4RL Kitchen and AntMaze. Supplementary experiments evaluate backbone robustness, meta-RL comparisons, task-coherent context, and unsupported-action value overestimation. These comparisons distinguish the benefits of Q-target pretraining from the remaining limitations of offline coverage.

Cosmological $\alpha$-attractor models are renowned for their universal predictions for inflationary observables, largely independent of the detailed form of the inflaton potential. In this work, we investigate the extent to which this universality persists for polynomial $\alpha$-attractor models, with particular emphasis on the scalar spectral index ($n_s$) and the post-inflationary reheating history. We derive an extended analytical expression for the scalar spectral index ($n_s$) and the tensor-to-scalar ratio ($r$), retaining sub-leading contributions beyond the commonly used universal approximation. Interestingly, we find that these corrections introduce an explicit dependence of $n_s$ on the parameter characterizing the polynomial attractor potential $\mu$, a dependence that is absent in the leading universal expression. We systematically investigate how this model dependence modifies the predictions in the $(n_s-r)$ plane and affects the model parameter ($\mu$). We further incorporate the Duration of reheating ($\Delta N_{rh}$) and effective equation of state parameter of the reheating epoch ($w_{rh}$), accounting for their impact on the number of e-folds between horizon crossing and the end of inflation ($\Delta N_{\mathrm{CMB}}$). By confronting the resulting predictions with current CMB observations, we identify the viable parameter space and examine the interplay between the model parameter ($\mu$) and reheating. Our results demonstrate that sub-leading corrections can provide a window into the underlying structure of polynomial attractors, offering a possible way to distinguish different members of the attractor class observationally.

Computational Science and Engineering (CSE) combines expertise at the intersection of engineering, applied mathematics, and computer science to form powerful methods for model-based design and model-based decision support across disciplines. Today, CSE methods and tools have an impact as an enabling technology in the development of increasingly sustainable products, processes, and operations. However, sustainability is rarely considered holistically in the context of CSE itself. This paper develops a perspective on sustainability in CSE by distinguishing CSE as an enabler of sustainability in application domains from sustainability within CSE itself, which rests on two complementary pillars: sustainable computing and sustainable software. We present illustrative examples for these pillars and derive recommendations and best practices for CSE stakeholders. Developing an understanding of how sustainability can create added value in CSE is an important future direction for the field. This calls for a concerted effort within the CSE community to define measurable outcomes and best practices that embed sustainability as a core design principle rather than an afterthought.

Eliciting rich data in semi-structured interviews is cognitively demanding, prompting recent work to explore real-time AI assistance for interviewers. However, introducing AI into the interviewer-interviewee interaction creates a triadic context whose social dynamics remain underexplored. We investigated how interviewers experience AI assistance for probing during semi-structured interviews. To elicit rich participant reflections, we implemented two variants of AI assistance differing in initiation and granularity in a high-fidelity prototype, ProbeAssist. We conducted a qualitative-first comparative structured observation study where 18 participants each completed three simulated interviews: one without AI and two with different AI variants. Findings showed that participants leveraged AI as a supportive tool but resisted it as an assessor or competitor. As they navigated AI's benefits and interaction costs, tensions emerged around agency, ownership, creativity, and interpersonal communication. We propose three implications for AI-assisted human-to-human interaction: managing social pressure, balancing idea alignment with inspiration, and preserving interpersonal presence.

We prove Laurent positivity for skew-symmetrizable quantum cluster algebras of rank three, with arbitrary integral compatible quantizations and invertible frozen variables. The proof uses mutation-acyclic positivity from a companion manuscript.

Extracting operationally meaningful quantities such as von Neumann entropy and mutual information is central to characterizing many-body quantum systems, yet experiments and numerics often provide direct access only to integer Rényi entropies. Whereas conventional extrapolation relies on a prescribed fitting ansatz, in PRL 137, 100202 we introduced an alternative approach based on stabilized analytic continuation (SAC), which avoids such an ansatz and is inherently robust to noise. Here, we further develop this framework in the noiseless setting and benchmark its performance across several nontrivial many-body problems. We first show that, besides the intrinsic ambiguity of reconstructing the von Neumann entropy from finitely many Rényi samples, analytic continuation can also fail because of genuine nonanalyticities in the Rényi function arising from zeros of $\text{Tr}\rho^z$. We derive universal zero-free domains for $\text{Tr}\rho^z$ and systematically sharpen these bounds using additional spectral information about $\rho$. We then benchmark the performance of SAC in three settings: ($i$) extracting topological entanglement entropy from DMRG Rényi data for the toric code and Kagome Heisenberg models; ($ii$) reconstructing the mutual information between disjoint intervals in a $(1+1)d$ compact-boson CFT; and ($iii$) recovering finite-time ballistic growth of the von Neumann entropy from integer Rényi entropies that cross over toward subballistic growth in diffusive quantum dynamics. In the latter setting, analytic continuation is expected to fail in the long time limit. We illustrate this mechanism with a physically motivated two-sector model, where the asymptotic separation between von Neumann and higher-Rényi growth is accompanied by a zero of $\text{Tr}\rho^z$ pinching the real axis at $z=1$, thereby producing a first-order Rényi phase transition.

Haar randomness is the standard null model for quantum typicality, scrambling, and random-matrix formulations of thermalization. In many physical settings, however, randomness is naturally constrained to subspaces of Hilbert space. This structure arises, in particular, both in systems with symmetry-resolved sectors and in the Eigenstate Thermalization Hypothesis (ETH), where mesoscopic energy windows define subspaces in which nearby energy eigenstates are mixed. Motivated by these settings, we consider a Block-Unitary Ensemble consisting of independently Haar-random unitaries acting on each subspace. We develop a constructive Weingarten Calculus for this ensemble, providing a unified framework across quantum information and many-body physics. This construction reveals a symmetry-resolved Schur-Weyl duality, where the relevant commutant is not the ordinary permutation algebra, but the one of color-preserving permutations. We apply this framework to quantum states under symmetry constraints and to energy-resolved ensembles for ETH.

The recent discovery of a population of intermediate-mass helium stars ($\sim$2-8 $M_\odot$) in the Magellanic Clouds has opened a new observational window into massive binary evolution. These hot, helium-rich stars are thought to be products of binary interaction, and some show no evidence for a luminous companion in their optical spectra. Here, we present multi-epoch radial velocity (RV) measurements for eight of these systems obtained with Magellan/MagE from 2018-2023. Five helium stars exhibit significant RV variability consistent with binary motion, while three do not. The five helium stars with evidence of binarity have orbital periods from 16 hours to 300 days, and eccentricities consistent with zero, implying companion mass lower limits from $\sim$0.6-2.9 $M_\odot$. We discuss implications of their orbits for the nature of the unseen companions, and find that low-mass late-pre/early-main sequence stars ($\lesssim$1.5-3.1 $M_\odot$), low-mass stripped stars and compact objects are currently viable. We identify an unexpected correlation between surface hydrogen abundance and binarity within this sample: helium stars in binaries are hydrogen-poor ($X_{\rm H,surf}$ $\sim$ 0.3-0.4), whereas helium stars without detected binary motion are nearly hydrogen-free ($X_{\rm H,surf}$ $\lesssim$ 0.05). This bimodality suggests that low RV variability in a subset of the sample is not simply due to inclination, but rather a distinct evolutionary pathway. We consider three possible explanations for these objects: wide binaries, disrupted binaries, and stellar mergers. The binary properties of this first sample of intermediate-mass helium stars is diverse, and provides important empirical constraints for binary evolution and compact object formation.

A skill is a modular package of natural-language instructions, executable scripts, and reference resources that an agent can load at runtime to extend its capabilities for a specific task. Skill-based agent systems therefore enable flexible reuse of third-party capabilities, but the openness of this skill ecosystem also opens up a new attack surface. Prior work has focused on vulnerabilities within individual skills, but little attention has been paid to risks that arise from interactions across skills. In this paper, we introduce skill cascading attacks, a threat paradigm in which a malicious objective is distributed across multiple skills so that each modification looks benign in isolation, yet their combined execution is harmful. For instance, in a prescription-review pipeline, the first skill weakens signals of recently discontinued medications in the extracted history, the second downgrades the severity of any drug interaction tied to them, and the third suppresses the resulting low-priority alert in the final summary, so that a severe drug-interaction warning silently disappears before reaching the physician. To systematically study this safety blind spot, we develop SkillCascade, an automated multi-agent red-teaming framework, and release SkillCascade-Bench, a benchmark of 213 validated cascading test cases across multiple agent systems and domains. Across representative agents (e.g., OpenClaw, Claude Code, Codex) and LLM backbones, cascaded interactions reliably induce harmful behaviors while evading existing per-skill scanners and runtime monitors. Our findings highlight a gap between component-level integrity and system-level safety, and call for defenses that reason over cross-skill interactions rather than individual skills in isolation.

Microwave Kinetic Inductance Detectors (MKIDs) are superconducting photon counting detectors that simultaneously measure the energy and arrival time of individual photons and are suitable for large-format arrays. The energy resolving power R = E/$\Delta$E is the primary figure of merit for spectrophotometric applications. It is limited in practice by the assumptions underlying the Wiener optimal filter used to estimate photon energy: stationary noise, linear response, and energy independent pulse shapes, all of which MKIDs structurally violate. We present Venom (Very Efficient Neural Optimal-filter for MKIDs), a selective state space model based on the Mamba architecture that replaces the coordinate transform and optimal filter, operating directly on raw in-phase/quadrature timestream data, carrying 3,314 trainable parameters, and targeting deployment on the MKIDGen3 and future system-on-chip readout platforms. To deal with the small calibration datasets typical of MKID experiments, we train on synthetic pulses drawn from a streaming principal component analysis that samples photon energy continuously between calibration wavelengths. All numbers are reported on a stratified 15% held-out validation set, with R measured from a kernel density estimate. On an InHf bilayer resonator (ten wavelengths, 254-1310nm), Venom reaches a mean R of 26.4 versus 24.2 for the published per-wavelength optimal filter baseline, a 9% improvement from one model in place of ten per-laser filter templates. The R of 38.6 at 254 nm is the highest ever recorded for an ultraviolet-to-near-infrared MKID suitable for use in a dense array. On a PtSi resonator (five wavelengths, 808-1310nm), where the published analysis used one shared 920nm template, Venom matches the optimal filter, with a mean R of 9.1 versus 8.8 at the three interior wavelengths.

We consider a tunable hybrid device composed of four quantum dots and a floating superconducting island – the so-called poor man's tetron – close to its effective particle-hole symmetric point and coupled to four external leads. We predict the onset of a charge Kondo effect mediated by Cooper pairs, resulting from the strong crossed Andreev reflections that dominate its low-energy behavior. This two-channel Kondo effect makes the poor man's tetron behave as a double Cooper pair splitter, turning uncorrelated incoming electrons into spatially separated Cooper pairs. Based on renormalization group techniques, we analyze both its weak and strong coupling regimes and we determine its non-Fermi liquid signatures. Our estimates indicate that the fractional conductance and transport features associated to this two-channel charge Kondo effect can be observed over a broad range of temperatures and anisotropies in realistic experimental conditions.

Motivated by recent optical measurements in twisted MoTe$_2$ homobilayers, we numerically study the binding of itinerant anyons to ordinary electronic bound states. Using a minimal few-body model of an anyon alongside holes and electrons in a Gaussian trapping potential, we show that a fractionally charged bound state comprised of an anyon and an ordinary electronic trion can become favorable nearby local electrostatic defects. The resulting object is an anyonic molecule wherein the anyon is bound to the trion “nucleus,” which in turn is localized to the defect and can be excited optically. We find that the resulting anyon-trion can exhibit a photoluminescence red-shift relative to the ordinary trapped trion, with binding on the meV scale for parameters relevant to twisted MoTe$_2$. We find that the value of this red-shift depends on $\textit{both}$ the anyon's fractional charge and its kinetic energy, thereby encoding both universal and non-universal anyon properties. Our results are qualitatively consistent with observations in twisted MoTe$_2$.