Papers

We revisit the $e^+e^- \to \phi \pi^+\pi^-$ reaction within the $K_1$-$\bar{K}$-$K$ triangle dynamics framework, in which the $e^+e^-$ pair annihilates through one-photon exchange approximation to produce a $K_1\bar{K}$ pair, followed by the $K_1 \to\phi K$ decay and the final-state $K\bar{K} \to \pi^+\pi^-$ rescattering. With the same theoretical formalism and model parameters, the $e^+e^- \to \phi K^+K^-$ reaction is investigated, and it is found that the predicted total cross sections for the $e^+ e^- \to \phi K^+ K^-$ reaction are in good agreement with the existing BESIII measurements. Our study shows that the triangle singularity in the $\phi K^+K^-$ channel is strongly suppressed, because the higher $K^+K^-$ mass threshold shifts the kinematics away from the triangle singularity condition and the phase space near the $K_1\bar{K}$ threshold is very limited. Moreover, the interference between the tree-level and loop amplitudes eliminates the remaining signal. These combined effects naturally explain the absence of a distinct $\phi(2170)$ signal in the $e^+ e^- \to \phi K^+ K^-$ reaction, provide a strong test of the model, and reinforce the picture that both reactions are governed by the same underlying mechanism, in which the $\phi(2170)$ state is produced in the $e^+ e^-$ annihilation from the $K_1$-$\bar{K}$-$K$ triangle loop.

Massive stars, including Wolf-Rayet stars, are predominantly found in binary systems. These systems are known to emit both thermal emission from stellar winds and, occasionally, non-thermal emission produced by relativistic electrons accelerated in the wind-wind interaction region. We intend to provide the most complete census of radio emission from WR stars in the Galactic plane, using the SARAO MeerKAT Galactic Plane Survey (SMGPS) complemented by the MeerKAT Galactic Center Survey. Our main motivation is to identify hints of synchrotron radio emission indicative of particle-accelerating colliding-wind binaries (PACWBs). We compiled an input catalogue of 428 WR stars positionally covered by the SMGPS and the MGCS. Using the survey data, we measured the radio emission at 1.3 GHz for detected WR stars. We also measured the upper limits for objects located in sufficiently low radio background regions. For detected objects, we searched for a radio excess by comparing the measured fluxes to two different evaluators of the thermal emission from massive star winds. We detected 23 targets and determined the upper limits for 279 WR stars. Among the detected objects, 15 display a (significant or potential) radio excess that cannot be explained by unresolved circumstellar emission of any kind. After removing already known PACWBs, we report the identification of 12 potential new PACWB candidates. Our study has led to the compilation of the most extensive catalogue of Galactic WR radio emission to date, based on a homogeneous dataset covering the Galactic plane. The low detection rate indicates either a low occurrence rate of synchrotron emission or substantial attenuation by turnover processes that are clearly dominated by free-free absorption from the WR wind material. Our results open the door to dedicated follow-up observations aimed at ascertaining the nature of the identified radio excesses.

We construct variational integrators to study the dynamics of stochastic Hamiltonian systems using their Wong-Zakai approximation. This approach can be used to interpret and justify previous work of Wang et al (L. Wang, J. Hong, R. Scherer, F. Bai. Dynamics and variational Integrators of stochastic Hamiltonian systems. International Journal of Numerical Analysis and Modeling, 6(4), 2009) on stochastic variational integrators. Numerical examples are provided.

The total decay width of the Higgs boson, $\Gamma_H$, encodes all of its decay modes; any decay into particles beyond the Standard Model (SM) would increase $\Gamma_H$ above the SM prediction of 4.1 MeV. In this Letter, a direct, on-shell measurement of $\Gamma_H$ is reported, exploiting the interference between the resonant $H \to \gamma\gamma$ amplitude and the continuum diphoton background, which distorts the diphoton invariant mass distribution in a way that depends on $\Gamma_H$ without requiring assumptions on the Higgs boson couplings. Using 140 fb$^{-1}$ of proton-proton collision data at $\sqrt{s}=13$ TeV recorded by the ATLAS detector during Run 2 of the LHC, an observed (expected) upper limit of $\Gamma_H < 307$ (258) MeV is set at 95% confidence level, approximately 75 (63) times the SM prediction. Unlike indirect off-shell constraints on $\Gamma_H$, this measurement does not rely on assumptions about the Higgs boson couplings, providing a novel complementary probe sensitive to potential Higgs boson decays into new particles.

LLM-based AI agents process user requests through iterative reasoning and tool execution, often involving the invocation of remote LLM APIs with local tool containers. This execution model can make the optimization of agent serving difficult because latency, local resource demand, and container bottlenecks inter-mix across requests. However, the current agent ecosystem runs without much consideration of resource dynamics, which results in significant waste of the precious resources. This paper analyzes the resource inter-mix of AI agents for three representative tasks: retrieval-augmented question answering, web search, and software coding. To this end, we characterize the latency with respect to the resource dynamics of processing multiple requests and tasks concurrently. Our measurements show that agents have a wide range of behaviors depending on tasks, so that even the same tool can differ substantially in resource dynamics. We also find that running multiple requests concurrently exposes task-dependent bottlenecks in resource dynamics such as CPU, disk I/O, and memory. Furthermore, we uncover that faster LLM responses or more CPU cores do not always accelerate agents. Based on these observations, we demonstrate new optimization opportunities that exploit the resource dynamics of tasks: CPU-aware tool admission and task-aware CPU allocation. Our results show that the latency of CPU-sensitive agent tasks improves $\sim$5.4$\times$, and the average latency across multiple tasks is reduced $\sim$32% compared to native agents.

A geometrically valid placement can still be difficult to execute because the selected grasp changes the required end-effector pose, collision geometry, and transport motion. Placement is formulated as a pre-execution ranking problem in which supplied grasp-placement candidates are scored before planning. The model combines a typed target-conditioned point cloud with three pose descriptors and hierarchical heads for planning success and execution success conditioned on planning. On a 30-object, 1,235-scene dataset with scene-group-held-out splits, three-seed top-1 success on covered test groups reaches 85.63 +/- 1.08% for joint selection and 79.84 +/- 0.16% for fixed-target ranking. For the designated frozen seed-42 checkpoint, top-1 success improves from 72.84% to 85.78% over full-pool cuMotion for joint ranking and from 59.65% to 79.67% for fixed-target ranking. Frozen transfer to xArm7/MoveIt requires no xArm-specific retraining. Across 27 locked cases, 13 complete end to end (48.15%). Of the 16 cases that pass Top-5 preflight and begin execution, 13 succeed (81.25%). Candidate-level deployment-feasibility prediction reaches 81.25% recall, 85.20% specificity, and 83.23% balanced accuracy.

We prove that the moment-angle manifold $\mathcal Z_{\mathcal K}$ is diffeomorphic to a connected sum of products of spheres when $\mathcal K$ is a starshaped (in particular, polytopal) 3-dimensional simplicial sphere with exactly two missing edges that are not adjacent to each other. One of the summands of the connected sum is a product of three spheres. For neighbourly starshaped simplicial spheres $\mathcal K$ of odd dimension, we prove the diffeomorphism $\mathcal Z_{\mathcal K} \cong M_1\#\cdots\# M_k$, where each $M_i$ is a product of two spheres. We give an explicit description of the moment-angle manifolds corresponding to non-polytopal Barnette and Brückner spheres.

Large language models (LLMs) have been applied to causal discovery, but candidate-graph generation rarely treats premature omission of potentially relevant causal relations as an explicit design objective. We propose MaSCoD, a multi-agent framework that organizes candidate third variables and local structural patterns before direct-edge judgment. We evaluate MaSCoD on Auto-MPG, DWD, and Sachs using GPT-5.4 as the primary backbone and GPT-4o for replication. MaSCoD exhibits a dataset- and backbone-dependent retention-selectivity profile rather than uniform superiority. Across all six dataset-backbone settings, Full, which supplies structural hypotheses before direct-edge judgment, achieved higher mean Recall and F1 than No Phase 1, which instead constructs them within the judgment procedure, while also increasing false-positive rates. Additional reference-edge retention over all evaluated baselines was observed on DWD with GPT-5.4 and on Sachs with GPT-4o, rather than uniformly across settings. Partial ablations showed that supplying both information components did not always outperform supplying only one. For GPT-5.4, stage-wise analysis showed that the Full-No Phase 1 retention gap was already present after direct-edge judgment, while reconciliation introduced additional reference-edge loss for Full on Sachs. These findings support structural pre-organization as an explicit design and evaluation target for omission control and motivate evaluating context construction jointly with its utilization in judgment.

Most existing graph streaming algorithms assume the ideal scenario where each edge arrives only once. Real-world graph streams, such as communication or transaction logs, often contain many repeated occurrences of the same edge. In general, the algorithms developed for the single-edge arrival case can fail when edges can arrive multiple times. Motivated by this, we study the repeated-edge arrival graph streaming model where an edge is allowed to arrive multiple times. In this work, we study the triangle counting problem in the repeated-edge arrival model: approximate the number of triangles in the underlying simple graph despite arbitrary edge repetitions. We design the first algorithms for triangle counting with optimal space complexity. In particular, we present a single-pass algorithm that computes an $(\varepsilon,\delta)$-approximation of the number of triangles with optimal space complexity. We introduce right-to-be-forgotten graph streaming (RFGS) model, where a forget operation can cause all previous occurrences of an edge to disappear. We show that our single-pass algorithm can be extended to the RFGS model with optimal space complexity. Finally, we present optimal constant-pass algorithms that compute an $(\varepsilon,\delta)$-approximation of the number of triangles and cliques for the repeated-edge arrival graph streams.

In extractive document question answering whose questions were generated from the passages that contain their answers – so that retrieval recovers 92-99.8% of what any mode combination could reach, whatever its absolute accuracy – confidence-driven mechanisms have little to gain. Fine-tuning an open language model on a specialized domain corpus yields a model whose own confidence is a tempting control signal: it could decide which queries warrant further adaptation, and which answers to trust. We evaluate both uses under criteria fixed before the runs were executed, across four 7-9B model families whose adaptation moved closed-book F1 by at most +0.03, and both fail: a distillation trigger on all four families, under its pre-specified three-step transfer budget, and a routing-and-abstention policy in its single-model pilot. Retrieval alone recovers 92-99.8% of best-case combined accuracy under every correctness criterion we test, leaving routers no meaningful gain. The sequence-likelihood signal is insufficient relative to that mode – area under the receiver operating characteristic curve 0.65-0.81 under the registered criterion – before adaptation as well as after, unchanged by scalar recalibration and not consistently improved by token-level temperature rescaling. And the finer diagnostics depend on the correctness criterion and on answer length; on the three adapted combinations where we could test it, selector ablations show no statistically detectable downstream benefit from the confidence term on any seed; on Gemma, removing it changes the selector from failing to passing both registered criteria. The usable product is a set of pre-specified negatives with their dependencies made explicit.

A soliton microcomb requires increasing pump power as its optical bandwidth is broadened. Resonant pumping through an auxiliary microresonator can reduce the power required to sustain a soliton, but simultaneously increases the power required for soliton formation. We show that this competition leads to optimal inter-resonator coupling, and the predicted minimum input pump power scales as the two-thirds power of the comb bandwidth, in contrast to the quadratic scaling under direct pumping. Experiments support the opposing power trends, providing a design rule for power-efficient, ultra-broadband soliton microcombs.

In previous papers, we began the study of sequence prediction algorithms adapted to stringological word complexity measures. In particular, we defined a complexity measure called Arithmetic Repetition Complexity (ARC) which admits a polynomial-time prediction algorithm with a mistake bound quasilinear in the complexity. Here, we show a weaker complexity measure related to ARC that admits an especially efficient prediction algorithm: an algorithm that runs in quasilinear time and polylog space for appropriate highly-structured sequences. The complexity measure is defined via a restricted class of "zipline programs" (a variant of straight-line programs), which we call layered. We thus get a less expressive measure with a more efficient algorithm (compared to our results for ARC), demonstrating a possible tradeoff.

Digital Educational Systems (DES) enable millions of individuals to learn or to acquire new skills. Their promises vary from one platform to another: some grant genuine degrees, others pledge rapid progress, drawing for this purpose on a range of techniques, from the recommendation of learning resources through complex models to a plain course textbook uploaded onto a learning management system. In this article, we argue that categorizing DES along two dimensions allows to map differences in the way these DES structurally conceptualise the learning process. These underlying dimensions are at once dimensions of the student model and mechanisms built in the DES's designs; we identify them as agency and legitimacy. Indeed, DES are, on the one hand, anchored in an institutionalised regime that grants them legitimacy and, in turn, legitimises the learning undertaken on these platforms; on the other hand, they make design choices as to the share of decision left to the learner in the learning process. We map a curated sample of common DES onto a plane composed of these two dimensions translated into indicators. This allows us to highlight underpopulated regions of the plane and to envisage how a DES might regulate these two dimensions algorithmically, in the interest of learning.

We show that the cover graph of the face lattice of every simplicial polytope admits a Hamilton cycle. Moreover, any two distinct facets can be prescribed as the neighbors of the full polytope on this cycle. By duality, the face lattice of every simple polytope is Hamiltonian as well. This proves the simplicial and simple cases of a conjecture of Behrooznia, Brenner, Merino, Mütze, Rieck, and Verciani (SODA26). The proof uses a line shelling in both directions. The forward shelling partitions the faces into Boolean intervals, whose cover graphs are cubes. The reverse shelling gives a second partition, on which we construct a perfect matching by adding or removing a fixed vertex in each interval. This matching connects the cubes of the first partition. A theorem of Gregor then completes it to a Hamilton cycle using only edges within these cubes. The same argument applies to shellable simplicial spheres, although without the prescribed neighbors of the full complex in this case. Using Chari's shelling construction of perfect matchings, we also obtain Hamiltonicity for polytopes with at most one facet that is not a simplex.

Recent studies have shown that smooth functions can be well approximated by ReLU neural networks with path norm constraint on the weights. We extend these results from uniform approximation to approximation in Sobolev norm. Specifically, we analyze how well Sobolev functions in $W^{n,p}$ can be approximated by neural networks with width $W$, depth $L$ and path norm bounded by $K$, when the approximation error is measured in the $W^{1,p}$-norm. For shallow networks with depth $L=1$, we derive the approximation error bound $\mathcal{O}(\max\{W^{-(n-1)/d}, K^{-(n-1)/(s-n)}\})$, when the smoothness index satisfies $n<s=(d+3)/2$ and the input is $d$-dimensional. For deep networks, we remove the restriction on the smoothness by showing that the approximation bound $\mathcal{O}(K^{-(n-1)/(d+d/p+1)})$ holds if the width $W$ and depth $L$ are sufficiently large.

Memory over-provisioning results in resource underutilization when HPC workloads run on Kubernetes. The default Vertical Pod Autoscaler (VPA) cannot anticipate phase-driven memory spikes for first-run HPC jobs. In this work, we present a reinforcement learning (RL) recommender VERA that formulates vertical memory scaling as a Markov Decision Process and trains an agent on 3353 real Prometheus traces. Evaluated on a live Google Kubernetes Engine cluster using LAMMPS, graph analytics, in-memory analytics, and MLPerf 3D-UNet, the RL agent reclaims 31.6% of the available memory headroom and incurs at most one OOM event while VPA reclaims -7.9% over the same runs, raising memory provisioning, and its recommendation would have been insufficient to avoid OOM in 30 runs. The results demonstrate that an observation-driven RL recommender could outperform retrospective heuristics for dynamic memory scaling.

Stably stratified wall-bounded turbulence governs the dynamics of many environmental and engineering flows. A key challenge is characterizing how stratification modifies mean and turbulent profiles. Monin–Obukhov similarity theory (MOST) is the dominant modelling framework, although it has rarely been rigorously validated against well-controlled direct numerical simulation (DNS) data over a wide range of stratification levels. In this study, we exploit the temporally developing turbulent boundary layer (TTBL) framework to investigate stratified turbulent boundary layers from the weakly stable to the very stable regime, spanning a range of Reynolds and Richardson numbers, and isolating the effects due to buoyancy from other mechanisms such as flow rotation. We demonstrate that the TTBL set-up faithfully reproduces classical similarity theory results and that surface-based scaling of the mean velocity gradient holds over a wider range of $z/L$ ($L$ being the Obukhov length) than previously reported. This result is attributed to the similar decay rate of turbulent shear stress and heat flux in this canonical flow. Next, we show that, as stratification intensifies, the intercept of the mean velocity profile increases, until the separation of scales required for a logarithmic region to exist can no longer be sustained. We propose an empirical closure for this intercept shift in terms of the Reynolds number based on the Obukhov length. Finally, a simple damping of the MOST contribution to the mean velocity profile is proposed and validated, enabling accurate prediction of the wall friction coefficient ($C_f$) across the investigated regimes.

Depth-recurrent language models iteratively apply a small layer stack, decoupling per-token compute from distinct parameter count. To determine whether such a model genuinely utilizes its depth, both recurrence and layer-pruning literatures rely on a shared evaluation: truncating depth at inference time, plotting quality against retained depth fraction, and reading off the slope. While cheap and training-free, this metric suffers from an unexamined flaw: it extracts a single scalar from an intervention that alters multiple model properties simultaneously. Depth truncation concurrently reduces the number of block applications, decreases the volume of distinct computation performed, and pushes the readout head onto an out-of-distribution residual stream. The observed slope conflates all three factors, yet is conventionally interpreted as reflecting solely the second. We propose the Depth Control Protocol (DCP), a diagnostic suite that disentangles these three quantities. DCP comprises three positive controls that isolate each factor while varying the others, a negative control applying the identical interventions to dense transformers to ensure the effect is not an artifact of the measurement protocol, and a controlled training intervention to verify causality. The linchpin control, running the full budget of block applications while executing only a single distinct iteration, is strictly realizable only in depth-wise weight-sharing architectures, since in a dense network repeating a layer yields an entirely different model rather than the same model in an alternative configuration.

Transverse modes beyond the fundamental mode are increasingly being explored to increase information capacity in both mode-division multiplexing and high-dimensional quantum information processing. The ability to combine and separate these modes at arbitrary ratios over a broad bandwidth is central to these applications. Motivated by the lower bending loss and intermodal crosstalk of partial Euler bends in multimode waveguides, we develop a geometric framework based on partial Euler bends for broadband power coupling between the first two transverse electric modes on a silicon-on-insulator platform. Our methodology uses a hybrid of eigenmode expansion and finite-difference time-domain methods. Our designs feature a mode (de)multiplexer, as well as unbalanced couplers (2/3 and 1/3 splitters). The simulated 1-dB bandwidths of our directional couplers range from $101$ nm to $134$ nm. The best-performing fabricated mode (de)multiplexer has a measured 1-dB bandwidth of $90$ nm, a conversion efficiency of $94.0\%$ and crosstalk below $-18$ dB at $1550$ nm. Our results validate the use of partial Euler bends for broadband transverse mode directional couplers and provide a general framework for their design.

The von Neumann-Morgenstern axioms are uncontroversial desiderata for individual decision-making. We say that a bargaining solution is rational if it can be interpreted as the most preferred alternatives under these axioms. Yet, neither the Nash nor the Kalai-Smorodinsky bargaining solution is rational in this sense. We formalize two consequences of rationality, namely that one can neither be strictly better off nor strictly worse off from randomizing over different actions. These two axioms, together with other standard axioms, characterize the relative utilitarian bargaining solution. We then implement this bargaining solution in sub-game perfect equilibrium.

Recent advances in quantum computing hardware toward fault-tolerance have increased interest in evaluating near-term quantum platforms on application-relevant quantum algorithms. In this work, we introduce Quantum WalkScore (QWS), a scalable application-oriented benchmark designed to assess the performance of NISQ and future fault-tolerant quantum computers in executing essential quantum routines –discrete-time quantum walks and quantum amplitude amplification– to solve the graph nodefinding problem (marked-vertex search). QWS quantifies performance by scoring the largest problem size for which a designated target node can be found with success probability above a defined threshold. In addition to the complete protocol description, we provide example parameter-selection scenarios designed from noiseless simulation results and demonstrate QWS evaluation through experimental runs on multiple generations of IBM real quantum processors.

We report the results of the first numerical lattice study using domain-wall fermions in the Sp(4) gauge theory coupled to two flavours of (Dirac) fermions, transforming in the fundamental representation of the gauge group. This theory plays a prominent role in the literature on extensions of the Standard Model with composite dynamics. It provides a short-distance completion for a class of composite Higgs models, or, alternatively, of dark matter models based on the strongly interacting massive particle paradigm. We adopt the Möbius formulation of domain-wall fermions (MDWF), implemented within the Grid software environment. We report the results of extensive tests of the algorithm implementation, and of the optimisation of the choices of algorithmic parameters appearing in the MDWF action. We then measure masses and decay constants of the lightest flavoured mesons in ensembles with moderately large fermion masses and several choices of lattice coupling, and perform an extrapolation to the continuum. We compare our results for the physical observables to published measurements obtained in the same field theory, but derived on the lattice by employing Wilson fermions. We demonstrate that, with the deployment of moderate computational resources, the MDWF formulation can yield order-of-magnitude gains in the approach to the continuum limit, in regions of physical parameter space relevant to phenomenological applications of this theory.

Leakage attacks on secret sharing schemes exploit partial information about individual shares to recover the underlying secret. In coding theory, linear exact repair schemes (LERSs) enable the recovery of one codeword symbol from a small amount of information obtained from the remaining symbols, provided that the code has sufficiently low rate. This can be interpreted as recovering the secret from partial information, namely subfield symbols, of the shares. Recently, a randomized construction based on subfield subcodes was proposed for constructing LERS-derived leakage attacks against Massey secret sharing schemes based on general linear codes. We extend this framework to multiple shared secrets whose corresponding shares are related through linear computations, with leakage also allowed on the computation outcomes. More precisely, we consider N secrets, of which K $\le$ N are linearly independent input values and the remaining N -K secrets are determined by linear computations on these inputs. We analyse the existence of LERS-derived leakage that exploits this structure. We first study the case of addition and then generalize our construction to arbitrary linear computations. Our analysis applies to general linear codes of length n+1 and dimension k over F\_{q^m} with k $\le$ N n/(Km), and supports arbitrary linear computations, whereas the previous subfield subcode construction only applies to k $\le$ n/m -1. Consequently, exploiting the linear relations enables LERS based leakage which extend the range of code parameters vulnerable to such attacks. Finally, identical leakage functions can arise for certain linear relations, making this a more realistic yet still potentially powerful attack model. Finally, simulations indicate that identical leakage functions can be used for certain linear relations, yielding a more realistic attack model.

Per-user LLM inference on transaction histories binds the inference budget linearly to user count, which becomes prohibitive at applied scale. We re-cast attribute inference from per-user to per-transaction-pattern. The pipeline runs in three phases: Resolve abstracts item names with optional web grounding, Profile infers attributes for each frequent pattern, and Tag clusters free-text attributes into a queryable database. In Profile, a single LLM call per pattern emits predefined categorical labels, free-text attributes, and per-attribute prevalence estimates. Because inference runs over patterns rather than users, the budget grows with the pattern count rather than the user count. On the public Open e-commerce corpus, the database is statistically indistinguishable from an LLM that reads each user's raw history directly in AUC across the evaluated attributes, and the prevalence estimates carry discriminative signal between positive and negative users. The pipeline is deployed at a major Japanese bank profiling on the order of tens of millions of users, with close to a three-order-of-magnitude reduction in LLM inference targets versus a per-user pipeline. The code is publicly available on https://github.com/CyberAgentAILab/profiling-agent-open-ecommerce.

Optical motion capture delivers high-fidelity human motion, but its reliance on strict marker layouts and clean trajectories severely limits its real-world applicability. In practice, tracking systems frequently output unconstrained markers: sparse, noisy, and unordered point clouds with unknown or varying configurations. To bridge the gap between corrupted raw markers and parametric human models, we introduce DirtyMoCap, a robust, marker-layout-free framework. Our core insight is to map unordered marker observations to a fixed set of "proxy anchors" comprising skeletal joints and body surface points, which serve as a stable intermediate representation. We first initialize and track these anchors over long sequences using a recurrent sliding-window architecture. Then, a custom differentiable Gauss-Newton solver fits the SMPL-H model to the tracked anchors to recover full-body pose, translation, and shape. By explicitly deriving geometric residuals, our solver learns adaptive observation confidence, smoothness, and prior weights end-to-end, adapting dynamically to the reliability of the input data. Extensive experiments on diverse, noisy marker configurations demonstrate that DirtyMoCap successfully generalizes across arbitrary layouts using only a single trained model. It consistently outperforms state-of-the-art configuration-specific baselines in both joint and vertex reconstruction accuracy, while our custom CUDA solver achieves up to a 100x speedup over standard PyTorch implementations. We further apply DirtyMoCap to heterogeneous raw optical MoCap recordings of traditional Chinese martial arts, yielding a Kung Fu motion dataset of temporally coherent SMPL-H reconstructions. Code and data are available at https://wanglongzju.github.io/DirtyMoCap-Project-Page.

Trilinear hexahedral mesh elements are used in finite element analysis to simulate volumetric phenomena. Realism of the simulation mandates that mesh elements through the trilinear map maintain positive volume, or mathematically, that the trilinear map maintains positive Jacobian determinant (jacdet). While jacdet positivity generally needs to be verified in the full element volume, we prove Knupp's conjecture which posits that for trilinear hexahedra, jacdet positivity on the boundary of the element is sufficient to guarantee positivity in the entire element. We further prove that for any valid hexahedron, its globally minimal jacdet must reside on the boundary. Lastly, we prove that the globally minimal jacdet on the boundary of a hexahedron, valid or not, must be achieved within a finite set of candidate points that can be determined by quartic root finding. Combining these results, we can algorithmically validate a hexahedron through evaluation of its jacdet on a finite set of boundary points.

Present models of the zodiacal light (scattered sunlight from interplanetary dust in the visible) and zodiacal emission (thermal emission in the mid-infrared) are primarily constrained by surface-brightness measurements obtained at relatively large solar elongations (e >~ 60 deg). As a result, their predictive accuracy at smaller elongations remains uncertain. This is particularly relevant for solar elongations of e ~ 30-60 deg, where next-generation near-Earth-object discovery missions such as NEO Surveyor and NEOMIR are designed to operate. To evaluate the reliability of these models at small solar elongations, we compile and reanalyze the available small-elongation zodiacal-light and zodiacal-emission measurements and provide the first systematic validation of modern zodiacal-light models in a regime where they have never been independently tested. We compare two models derived from the COBE/DIRBE survey: one constrained primarily by the spatial and temporal variation of the monochromatic sky brightness, and another that additionally uses interband color information and fixes the absolute normalization by requiring the minimum high-latitude residual at 25um to vanish. We find that the former reproduces the observations significantly better. At wavelengths most relevant to infrared near-Earth-object surveys (8-10 um), it typically agrees with measurements to within ~10% at e ~ 40-60 deg. Although it also performs better at e ~ 30 deg, significant discrepancies remain. These results show that zodiacal emission at small solar elongations is still only weakly constrained and that its absolute brightness remains uncertain. New mid-infrared observations are required to improve background estimates for future infrared survey missions.

Large Language Models (LLMs) transform vast collections of unstructured text into semantic patterns used for language generation and reasoning tasks. Behind their ease of use lies a complex process: words become tokens, tokens become vectors, and vectors ultimately give rise to streams of bits that flow through High-Performance Computing (HPC) systems. As modern LLMs grow to billions or trillions of parameters, this path increasingly unfolds across thousands of interconnected accelerators, making the underlying communication fabric a critical and often opaque component of model training. This tutorial aims to walk the reader through the journey from words to network traffic, shedding light on how language is translated into communication flows within HPC training systems. Using concrete examples from Dante's Divine Comedy, we illustrate how model architecture, tokenization, embeddings, and parallelization strategies shape the volume, structure, and timing of data exchanged across the network. We combine architectural analysis with analytical traffic models and numerical examples to characterize the communication requirements of LLM training. We try to demystify how words travel across the network and provide practical insights into the network requirements needed to support the journey from text to trained model.

Vision-language-action models (VLAs) have emerged as a promising paradigm for general-purpose robot learning, with performance improving as models and datasets scale. Scaling robot data collection, however, remains challenging because data are naturally distributed across robots, tasks, and locations, making centralization costly or impractical. Federated learning offers a way to train on decentralized robot data, but applying it to VLAs requires accounting for heterogeneous robot client data distributions. We present Co-VLA, which applies consensus optimization using the Alternating Direction Method of Multipliers~(ADMM) to federated VLA training. We show that the same algorithm supports both full-model training and parameter-efficient fine-tuning with both fixed-rank and rank-adaptive adapters. The name Co-VLA reflects both consensus and collaboration: clients with different local robot datasets collaboratively train a shared model without sharing their data. Our experiments demonstrate that Co-VLA achieves performance comparable to centralized training in both full-model training and parameter-efficient fine-tuning settings. The project website and videos of our real-world experiments are available at https://embodiedvision.github.io/co-vla/.

Weak interactions are fundamental in the merger of binary neutron stars (BNS). They provide cooling, change the composition of matter outflows, and can lead to significant secular outflows through neutrino-driven winds. Hence, the inclusion of neutrino transport has become standard in numerical-relativity simulations of mergers involving neutron stars. In more detail, the interactions of neutrinos with matter in a truncated two-moments scheme (M1) are described through interaction rates, calculated from a set of standard interactions. Among those are charged-current reactions, for which the elastic approximation is commonly assumed. More sophisticated alternatives are provided by, e.g., the publicly available NuLib library, or rates computed within a full kinematics approach. In this work, we compare merger simulations of a BNS systems employing each of these three prescriptions for the computation of charged-current weak rates, included in the M1 equations with an implicit-explicit time integrator. We find that the elastic approximation slightly overestimates the dynamical ejecta mass, while underestimating the amount of high electron fraction material therein. Additionally, it overestimates the neutrino luminosities at early times. The effects of full kinematics in comparison to the NuLib rates appear negligible within the uncertainties of the simulation data. Similarly, the post-merger gravitational-wave spectra for the three prescriptions were found to be indistinguishable within the numerical uncertainties. Nucleosynthetic yields, computed with a nuclear network, are broadly consistent, except in the atomic mass range $A\in[90,125]$ where the elastic approximation shows larger abundances. Finally, the different interaction rates seem to affect the remnant disk properties, hinting at the importance of more accurate interaction rates for simulations spanning secular timescales.