Papers

Radio frequency fingerprint identification (RFFI) enables device identification and authentication by exploiting transmitter-specific hardware impairments embedded in received RF signals. Existing RFFI methods are typically designed for a specific operating regime, such as closed-set identification, open-set recognition (OSR), or discovery of new devices, and often treat these tasks as separate algorithmic components. In this paper, we propose a holistic RFFI framework that jointly supports closed-set identification, OSR, and novel class discovery (NCD) within a single expandable architecture. The proposed method formulates RFFI as vector quantization in a learned latent feature space, where registered devices are represented by labeled codewords. This representation enables closed-set identification through codeword assignment, OSR through uncertainty measures induced by the codebook geometry, and NCD by using the latent representations and open-set uncertainty of rejected samples to organize them into new device identities. We further develop a multi-stage training procedure that combines supervised representation learning, latent-space refinement, codebook initialization, and vector-quantization-aware optimization. Experimental results on LoRa and WiFi RFFI datasets demonstrate that the proposed framework achieves accurate identification of registered devices, reliable detection of unseen transmitters, and effective discovery and registration of new device classes, highlighting the benefits of coupling the different RFFI tasks.

Tokenizer-free language models remove the inductive bias of fixed tokenizers by modeling text directly as bytes, but the resulting longer sequences substantially increase computation and eliminate explicit text abstractions. We ask whether this additional computation can be useful, and whether standard Transformers can learn the abstractions that tokenization provides. We study these questions on Transformers without specialized tokenization-related architectures. With token-superposition training and hash embeddings, byte Transformers consistently outperform subword Transformers as model size scales. We further find that byte Transformers build local text abstractions as external tokenizers: a set of segmentation-like positions are used to collect local context representations, and restricting up to $25\%$ of intermediate layers to these local representations preserves downstream performance. Finally, these learned structures induce highly non-uniform generation difficulty, with uncertainty concentrated near local structure boundaries; exploiting them for speculative decoding yields $3.4\times$ more accepted tokens than in subword Transformers.

Serving a large language model (LLM) across a fleet of deployments requires several weight-precision operating points. Multi-precision formats serve them all from one stream whose prefixes are valid lower-precision codes, instead of storing multiple copies. We present StagQ, a multi-precision weight format whose main stream is a 2-bit group-wise affine base followed by a configurable number of 1-bit refinement planes on a dyadic step schedule. Every supported precision is a readable prefix, decoded by an affine map derived from metadata shared across all precisions, with no per-weight lookup. A sparse side record, filled both before and after the grid is fitted, holds out the few weights the grid serves worst. We report two configurations of the encoder. At two bits the cheaper one leads the strongest multi-precision baseline on Llama-3.1-8B, Phi-4, and OLMo-2-7B by 3.1 to 7.0 MMLU points, at a slightly lower logical rate. At three bits it leads on Llama-3.1-8B, leads on Phi-4 at a higher rate, and ties on OLMo-2-7B. At four bits it ties on all three, at a higher rate. In a batch-one matrix-vector product on an NVIDIA A100 GPU, timed on synthetic weights, our kernel is faster than the two baseline kernels in most shape-precision cases.

Verifier-guided decoding can prevent harmful reasoning steps from contaminating subsequent generation, but typically relies on an external learned verifier. We ask whether a language model can instead reject its own bad reasoning steps. We define a prefix's recoverability as the probability that the frozen generator can complete it correctly. Diagnostics show that adjacent recoverability changes are often difficult to resolve with practical Monte Carlo budgets, while same-prefix candidates exhibit a sparse low-recoverability tail. We introduce Self-Step Rejection (SSR), which trains a lightweight LoRA acceptance gate on the generator backbone while keeping the base model frozen. SSR uses confidence-qualified first-passage supervision: steps before the first resolved crossing of a root-relative recoverability barrier are accepted, the crossing step is rejected, and unresolved steps and suffixes are excluded. Training combines pointwise classification, same-prefix pairwise learning, and group-relative policy refinement using final-answer correctness. At inference, SSR accepts candidates or resamples from the unchanged prefix under rejection budgets, without an external learned verifier. Across three reasoning models and five mathematical reasoning benchmarks, SSR improves macro-average accuracy over single-pass decoding by 5.4–10.1 points using 1.21–1.40x as many generated tokens, and achieves the highest macro-average accuracy among evaluated step-level methods. Full-solution scaling methods require 4.47–8.27x the single-pass token cost for comparable performance.

The goal of this article is to study optimal high-order Taylor-based polynomial quasi-Trefftz spaces for first-order time-harmonic Maxwell's equations. Compared with standard polynomial spaces, the purpose of quasi-Trefftz spaces is to retain the same high-order best approximation property but more efficiently: their dimension is considerably smaller. To achieve this goal, they approximate only smooth solutions to the governing equation rather than to approximate general smooth functions. While there is a natural definition of Taylor-based polynomial quasi-Trefftz spaces for scalar equations, the situation is fundamentally different for vector-valued equations, as the naive quasi-Trefftz property does not systematically define the smallest space with the desired best approximation property. This article identifies the optimal quasi-Trefftz spaces for first-order time-harmonic Maxwell's equations, proposes two approaches to construct quasi-Trefftz bases, and proves the spaces' best approximation property. The theoretical results, which strongly leverage polynomial Helmholtz decompositions and the exactness of the polynomial de Rham sequences, are validated with numerical experiments.

Reinforcement learning can substantially improve a reasoning teacher, but it is unclear which of those improvements survive when the teacher supervises a smaller on-policy student. We study this question in mathematical reasoning by comparing teacher lineages before and after GRPO, multiple student scales, direct GRPO, and several on-policy distillation objectives. The central finding is that transfer is structured rather than scalar: teacher strength alone does not make dense distillation competitive, while an RL-improved teacher creates useful but metric-dependent student gains. This motivates Transfer-Stratified On-Policy Distillation (TS-OPD), which screens training problems by the joint sampled success of the student and teacher, routes acquisition problems to gated forward KL, routes consolidation problems to gated reverse KL, and adds an entropy brake to protect sampled coverage. Across the main comparison, TS-OPD is the strongest student objective for macro average correctness with the GRPO-improved teacher, while pass@K remains more mixed. Ablations show that the gains come from routing and token gating rather than skipping problems. These results support a transfer-aware view of OPD: stronger teachers help when the supervision direction and token budget match the student's observed ability, not merely because the teacher endpoint is stronger.

We establish transference results for multilinear Fourier multiplier operators from Euclidean spaces to product spaces of the form $\mathbb R^{d_1}\times\mathbb T^{d_2}$. We prove strong and weak-type transference principles that preserve the corresponding Euclidean operator bounds. As applications, we obtain bounds for certain multilinear operators, as well as for some multiplier operators arising naturally in PDEs.

We give an example of a proper, regular, integral arithmetic surface over $\mathrm{Spec}\ \Z$ endowed with a section, and whose étale fundamental group admits an infinite pro-$2$ quotient. The construction is a geometric analog of the Golod–Shafarevich towers for rings of integers of number fields.

Robot delivery studies can overstate transport capacity when travel to pickups and human support fall outside the modeled schedule. We formulate a location aware dispatch model that couples robot admission to transporter support and shared elevators, charging, and cleaning. The model replays 33,079 observed hospital requests; missing contents, deadlines, staffing, and completion times remain explicit scenario assumptions. Two full grids compare human dispatch, a resource aware controller, and a proximity and workload benchmark in 30 paired replications per scenario. An exploratory extension tests a simpler deadline admission rule under the same operating model. Pickup travel increases demand on a pooled elevator bank. In one high staffing case, raising assumed elevator capacity from two to four reduces human only lateness from 55.11% to 3.18%, exceeding the dispatch differences. Direct policy comparisons and robot coverage distinguish admission selectivity from system service. The analysis explains why a robot can pass a deadline admission test yet delay service through staff handoffs and shared facilities. Its contribution is a reproducible evaluation of coupled dispatch workflows, with a clear separation between admission, completion, and return to availability. The results are conditional comparisons, not estimates of hospital benefit or released clinical capacity.

The rolling of two-dimensional (2D) van der Waals (vdW) magnetic monolayers into cylindrical architectures provides an ideal platform for confining magnetic domain walls (DWs) without edge-pinning effects. Here, we present a combined analytical and numerical investigation of DW confinement and interaction in ferromagnetic (FM) nanotubes. Using an analytical Heisenberg model on a 2D rectangular lattice, we show that transversal curvature introduces an azimuthal exchange energy penalty that intrinsically compresses the DW width. This model demonstrates excellent quantitative agreement with atomistic Landau-Lifshitz-Gilbert (LLG) spin dynamics simulations. By extending our study to realistic CrSBr (CSB) nanotubes, we model a bound state of two interacting DWs arising from a balance between long-range dipolar attraction and short-range exchange repulsion. Micromagnetic simulations reveal a non-monotonic evolution of the inter-wall distance with the tube perimeter. These findings highlight vdW magnetic nanotubes as robust platforms for stable domain confinement in quasi-one-dimensional spintronic architectures.

Diffusion policies provide expressive action distributions for continuous-control reinforcement learning. However, safety-aware online diffusion policy optimization remains underexplored, particularly methods that use predictive reachability information without an explicit dynamics model. We propose Reachability-Aware Diffusion Policy Optimization (RADPO), a model-free method that combines predictive first-hit safety estimation with cumulative-cost budget feedback. RADPO learns a discounted first-hit reachability value that captures the discounted risk of a cost event, assigns larger weight to events that occur sooner, and uses this signal to shape the reward. A separate dual-like multiplier adjusts the shaping strength according to realized episodic costs relative to a prescribed budget. The diffusion actor improves through weighted denoising regression on candidate actions scored by the reward critic. Our approach requires neither a learned dynamics model, action gradients through the critics, nor differentiation through the reverse diffusion sampler. We establish theoretical properties of the reachability value and show that accumulated reachability penalty provides a conservative surrogate for future discounted cumulative cost. Across ten continuous-control safety tasks, RADPO achieves competitive reward-cost trade-offs, with substantial reductions in constraint violations on several tasks relative to the compared baselines. Our theoretical and empirical analysis supports that combining reachability with cumulative budget feedback is a viable approach to safety-aware diffusion policies.

We study last-iterate convergence in unknown two-player zero-sum discounted Markov games with bandit feedback. The players learn independently along a single trajectory without observing each other's actions. We develop Adaptive Regularized TD Learning (ARTD), which achieves a $\widetilde{\mathcal{O}}(t^{-1/4})$ duality gap bound for the current policies under a uniform hitting time assumption, with high probability simultaneously over all rounds and starting states. This improves the $\widetilde{\mathcal{O}}(t^{-1/(9+\nu)})$ rate of Cai et al. (2023), for any fixed $\nu>0$, under the same feedback model and hitting time assumption. Our algorithm requires no knowledge of the hitting time bound, the time horizon, or the confidence level. To stabilize policy learning as value estimates change, we separate fast temporal difference averaging from bounded value updates. We adapt log-barrier regularization to the progress of value estimation, controlling both policy and value errors throughout learning. Together, these mechanisms enable fast convergence of the policies actually played, even when the players learn independently from bandit feedback.

Image compression, restoration, and generation all require a way to measure how different two images look to a person. Pixel error ignores how people see, while the most accurate perceptual distances are typically fitted to human judgments, tying them to a fixed data and resolution. For example, when image resolution is doubled, the correlation of DISTS with human scores on TID2013 drops from 0.815 to 0.717. We introduce the Jacobian Lens Distance (JLD), which derives its perceptual geometry from a frozen vision encoder rather than from human labels. JLD combines the locality of early patch features with the perceptual sensitivity captured by later encoder representations. Specifically, we use the encoder Jacobian to identify directions in the early feature space that most strongly affect the encoder output, producing a fixed metric tensor, $E[J^\top J]$, which we call the Jacobian lens. The lens is fitted only once from 100 unlabeled images, taking about 35 seconds. Locally, this construction defines a pullback metric in pixel space, giving JLD a clear geometric interpretation that can be directly analyzed on real images. Across four standard perceptual databases, JLD achieves state-of-the-art performance and consistently outperforms LPIPS, DISTS, PieAPP, and DreamSim. JLD is also robust to changes in image resolution, on TID2013, its lens-term correlation remains nearly unchanged when the resolution is doubled, decreasing only from 0.850 to 0.845. We further introduce JLD-fast, which is $4\times$ faster than LPIPS-VGG while achieving a mean correlation of 0.911. Finally, JLD naturally extends to video, reaching a correlation of 0.786 on Waterloo IVC 4K compared with 0.611 for VMAF.

Domain adaptation aims to turn a general-purpose large language model (LLM) into an expert for a target domain. While the dominant SFT+RL pipeline offers a convenient cold start, it may reduce exploration diversity and introduces additional complexity through multi-stage optimization. These limitations motivate RL-only adaptation. However, pure on-policy RL suffers from a cold-start problem, while mixed-policy RL still falls short: informative tokens in teacher outputs are learned too slowly in early training, and stale teacher outputs can hinder later improvement. We identify these two failure modes as Gradient Starvation and Teacher-Distribution Anchoring. To address them, we propose One-stage Policy Optimization (OnePO), which treats teacher outputs as transient guidance for policy improvement. OnePO combines Adaptive Objective Evolution to strengthen learning on informative low-probability teacher tokens and Teacher Retirement to discard teacher outputs once the current policy can surpass them. On medical adaptation, OnePO achieves 67.2 on HealthBench (Total) with only 20K training samples, outperforming SFT+RL and pure RL by 2.7 and 7.4 points, respectively. We further scale OnePO to produce HuatuoGPT-3, an open-source medical LLM series whose 27B variant reaches 70.1 on HealthBench (Total) and 71.4 on HealthBench Professional, surpassing frontier models such as GPT-6 Astra. Models and code are available at https://github.com/FreedomIntelligence/HuatuoGPT-3.

We study the Cauchy problem for the one-dimensional isothermal compressible Navier–Stokes system with the van der Waals equation of state, and analyze the well-posedness and asymptotic stability of steady-state phase-transition solutions at subcritical temperatures. Working in Lagrangian coordinates, we use variational techniques to construct piecewise smooth steady-state solutions corresponding to two-phase equilibrium states under finite-mass perturbations. Introducing artificial-viscosity regularization, we first establish existence of the global minimizer of the regularized variational problem, unique within each fixed symmetry class; passing to the vanishing artificial-viscosity limit then yields a physically admissible steady-state solution with two phase interfaces. For initial perturbations that are small with respect to energy, piecewise-gradient, total-variation, and integral norms but may exhibit large pointwise jumps across phase boundaries, we derive uniform a priori bounds for weak solutions near the two-phase equilibrium profile. We further prove that these global weak solutions converge to the steady-state two-phase equilibrium in $L^\infty(\mathbb{R})$ as $t\to\infty$, thereby characterizing their large-time asymptotic behavior. Physically, finite mass perturbations break the uniqueness of the uniform steady state and generate infinitely many piecewise constant two-phase equilibria satisfying the Maxwell equal-area rule. Artificial viscosity acts as a selection principle that singles out the physically relevant two-interface steady-state configuration, and this two-phase equilibrium is asymptotically stable under small integral perturbations that allow large-amplitude discontinuities at phase interfaces.

Natural face-to-face human–robot interaction requires a robot to understand an evolving social situation, decide when to engage, and express its intent through coordinated physical behavior. Yet existing approaches rarely close this loop: foundation-model agents provide increasingly capable multimodal reasoning and memory but remain largely disembodied, while expressive virtual agents do not face the physical constraints of real robots, and physical social robots typically address social reasoning and embodied expression only partially. We present ARISE, a unified framework that bridges Agentic Reasoning and Interactive Social Embodiment on the Sophia humanoid robot. ARISE integrates multimodal context understanding, long-term memory, and reactive and proactive interaction to determine when and what to communicate, and translates social intent into robot-native gestures coordinated with speech and mechanical facial expressions through streaming execution. Extensive evaluations on Sophia demonstrate strong perceived interaction quality, expressive and well-coordinated embodied behavior, and substantial latency reductions through streaming execution. These results highlight the importance of jointly reasoning about what to communicate, when to engage, and how to physically express social intent for natural interaction with humanoid robots. Project Page: https://robosocial.github.io/

Smart ports increasingly depend on IoT-enabled digitalization to turn heterogeneous operational processes into measurable, optimizable workflows. Accordingly, this paper first provides a concise review of IoT-enabler technologies for port digital transformation, emphasizing how sensing, connectivity, and data integration capabilities support real-time decision-making in port environments. Motivated by the growing emphasis on operational decarbonization reflected in the European Green Deal and the Sustainable Development Goals (SDGs), we then present a data-driven fleet scheduling framework for landside port operations at the Port of HaminaKotka (Mussalo terminal), Finland. The scheduling framework reallocates vehicle operating hours to minimize total CO2eq emissions while preserving the same aggregate workload. The study uses a two-year dataset from 77 port-operating vehicles, analyzed at monthly and annual aggregation levels, to derive vehicle-specific intensity metrics (CO2eq per operating hour) and identify reallocation opportunities. We formulate the scheduling task as a linear programming workload allocation problem that minimizes the weighted sum of assigned hours while keeping total required hours fixed, enforcing fleet-level caps on CO2eq and energy use, and respecting per-vehicle operating limits. The optimization is solved using the HiGHS linear programming solver via SciPy's linprog. Results show that shifting utilization away from high-intensity vehicles and toward more efficient units yields substantial system-level improvements with no reduction in total operating hours: the optimized schedule achieves an approximately 31% reduction in total CO2eq and a simultaneous 9.4% reduction in total energy consumption.

We study four of Manton's five vortex equations on a Riemann surface $X$ in relation to anti-self-dual connections. From a solution of each vortex equation, we construct an equivariant holomorphic bundle over $X$$\times$$\mathbb{P}^1$, $X$$\times$$\Delta$, or $X$$\times$$\mathbb{C}$, together with an invariant Hermitian or pseudo-Hermitian metric whose associated connection is anti-self-dual. The equivariant structures are given by the actions of $SU(2)$, $SU(1,1)$, and $SE(2)$, respectively. Conversely, we show that an equivariant holomorphic bundle equipped with an invariant (pseudo-)Hermitian metric gives rise to a solution of the corresponding vortex equation when the associated connection is anti-self-dual.

Valuing player actions in football, a complex multi-agent system, requires accounting for strategic interactions among 22 players, including off-ball movements and defensive positioning. Existing reinforcement-learning-based valuation methods commonly incorporate other players through the observed state while retaining independent-agent value formulations. To our knowledge, we introduce the first Markov perfect equilibrium (MPE)-inspired framework for valuing the actions of all 22 players using real football tracking and event data, formulating each possession as a finite-horizon multi-agent dynamic game. To improve interpretability, the framework uses Expandable Decision-Making States (EDMS) and decomposes each Q-value into a successor-feature basis and a linear reward-weight vector. The value basis is estimated through a two-stage procedure combining linear temporal-difference initialization with nonlinear iterative refinement from observed state-action transitions. Using tracking and event data from 95 J1 League matches, we compare the proposed formulation with an independent reinforcement learning baseline under both Position and Velocity States (PVS) and EDMS representations. With EDMS held fixed, the independent baseline assigns the highest value to forward movement in 99.21% of evaluated off-ball states, whereas the most frequent direction under the proposed formulation accounts for 17.63%. Qualitative analyses illustrate context-dependent off-ball valuations and highlight limitations of the current state representation for defensive positioning. Overall, the framework provides an MPE-inspired approach to interpretable action valuation in football, enabling analysis under alternative tactical reward weights.

The global spatial structure of the universe is a fundamental question in cosmology. Although the standard model usually assumes a trivial topology, a multiply connected universe can break statistical isotropy and generate off-diagonal covariance between the spherical-harmonic coefficients of cosmic microwave background fluctuations. In this work, we investigate whether CMB B-mode polarization improves constraints on the cubic three-torus, or $E_1$ topology. We construct the full harmonic-space covariance matrices of the temperature, E-mode, and primordial B-mode anisotropies, including correlations between different $(\ell,m)$ modes, and evaluate Gaussian likelihoods using random realizations generated from a statistically isotropic $\Lambda$CDM model with trivial topology. To isolate the intrinsic topological information carried by each CMB observable, we consider the ideal cosmic-variance-limited case, neglecting instrumental noise and gravitational lensing. We find that adding primordial B-mode polarization yields only a marginal improvement in the ability to distinguish the $E_1$ topology from the trivial topology, even under these ideal conditions. The different constraining powers of the temperature, E-mode, and B-mode anisotropies can be understood through the combined effects of their transfer-function responses and the topology-induced off-diagonal correlations in harmonic space. }

Recent neural power-flow solvers, including emerging foundation models, achieve accurate voltage predictions, yet such accuracy does not necessarily imply physically consistent solutions. Even small complex voltage errors can yield large AC power-balance residuals. We study this accuracy-consistency gap across PIGNN-GC, GridSFM, gridfm-graphkit, and LUMINA on realistic 2224-bus Great Britain network (GBnetwork) scenarios, with cross-grid evaluation of GridSFM over 31 systems. Using a singular value decomposition (SVD) basis fitted to training AC power-flow solutions, we find that neural prediction errors contain substantial components outside the dominant solution subspace. To address this mismatch, calibrated solution-subspace projection (CSP) suppresses off-subspace prediction components after train-only bias calibration, reducing Mean PB by 67.0%, 37.8%, 40.5%, and 68.9% for PIGNN-GC, GridSFM, gridfm-graphkit, and LUMINA, respectively, relative to calibrated predictions, while improving voltage-magnitude accuracy in all four models. These results identify output-error geometry as an important factor in physics-consistent neural AC power flow. Code: https://github.com/Kimchangheon/neural-acpf-error-geometry

Motivated by the deviations between experimental data and SM predictions for the branching fractions of $\bar{B}_{(s)}^{0} \to D_{(s)}^{(*)+}L^-$ decays with $L=\pi,K^{(*)},\rho$, we investigate whether the type-IV scalar diquark can offer a plausible explanation for them. By matching the renormalized amplitudes in the full theory onto those in the low-energy effective theory formulated in the diquark operator basis, we extract the corresponding Wilson coefficients up to NLO in $\alpha_s$ at the NP scale $\mu_\phi=M_{\phi}$. We then rewrite the obtained operators in the SM basis and apply the renormalization-group running to evolve the Wilson coefficients down to the hadronic scale $\mu_b=m_b$. For the hadronic matrix elements of the relevant NP operators, we employ the QCD factorization approach to evaluate the non-factorizable one-loop vertex corrections, and confront our theoretical predictions to the latest data to constrain the diquark parameter space. In the single-coupling case, we find that the purely left-handed coupling $\Delta_{d(s)}^{LL}$ can reconcile the discrepancies at $1\sigma$, while the mixed right-left coupling $\Delta_{d(s)}^{RL}$ remains compatible with data only at about $2\sigma$. Extending to the two-coupling case, we find viable parameter regions as long as $\Delta_{d(s)}^{LL}\neq 0$. Incorporating the LHCb measurement of the ratio $R_{\pi/\mu\nu_{\mu}}$ further reduces the allowed range of $\Delta_{d}^{LL}$, but does not yield meaningful extra restriction on $\Delta_{d}^{RL}$. Hence, while either $\Delta_{d(s)}^{LL}$ or $\Delta_{d(s)}^{RL}$ can separately provide an acceptable NP interpretation, our phenomenological results favor a scenario dominated by purely left-handed diquark-quark couplings. Finally, we present updated predictions for the branching fractions and the ratios $R_{(s)L}^{(*)}$, which can be probed by forthcoming LHCb and Belle II measurements.

In this work we investigate the behaviour of early inflationary cosmic correlation functions under the flow of the infrared scale for interacting spectator scalar field theories. We start with the derivation of the one-loop effective potentials for $\phi^4$, $\phi^4$-Yukawa and broken phase $O(N)$ model in the inflationary de Sitter spacetime, first using the local Schwinger-DeWitt expansion propagator, and next using the exact de Sitter propagator. The Schwinger-DeWitt expansion being valid in a sub-Hubble length scale, furnishes suitable framework to determine the leading curvature corrections to high energy phenomena having sub-Hubble Compton wavelengths. These effective potentials show spontaneous symmetry breaking pattern, and are next used as ultraviolet boundary conditions in the Wetterich equation generating the desired IR flow. The flow is shown to drive the theory towards symmetry restoration. Using these IR-improved effective potentials, we next compute the stochastic power spectra and spectral indices for the temporally coincident two point correlation function of the spectator scalar. We show that the results corresponding to the aforementioned two kind of propagators are roughly in qualitative agreement with each other, but may differ quantitatively significantly. Both of them show increase in the spectral index in the deep IR for $\phi^4$ and $\phi^4$-Yukawa. The effect of fermionic interaction has been emphasised. Also for the case of the $O(N)$ scalar, we demonstrate that the effective potential's shape becomes more and more insensitive to the flow as $N$ gets large. Implication of this on the correlation function has been discussed.

We study the complexity of Nash equilibrium computation in discrete Colonel Blotto games. For two-player general-sum Colonel Blotto with monotonic piecewise-constant battlefield payoffs, we show that computing an inverse-polynomial approximate Nash equilibrium is PPAD-hard, even with only two battlefields and a constant number of pieces per battlefield, thereby resolving an open question of [KPF+25]. Allowing more battlefields, the hardness persists when every battlefield payoff is $2$-piecewise constant with respect to either player's allocation. A general PPAD-membership theorem for multiplayer Blotto with arbitrary local reward functions then implies PPAD-completeness for both hardness results. Motivated by fixed-rank bimatrix games, we then identify a tractable frontier within two-player general-sum Colonel Blotto. If the social payoff has rank-$r$ structure, then, for every fixed $r$, an $\varepsilon$-Nash equilibrium can be computed in time polynomial in $B_1,B_2,k$, and $1/\varepsilon$. Thus constant-rank two-player Colonel Blotto remains tractable despite its exponentially large pure strategy spaces. Finally, we study multiplayer winner-takes-all Colonel Blotto with player-specific battlefield values and uncover a sharp tie-breaking frontier. If every highest bidder receives her full battlefield value, an exact pure Nash equilibrium can be computed in polynomial time for arbitrary binary-encoded budgets. In contrast, under ordinary equal splitting among highest bidders, we prove that computing an inverse-polynomial-accuracy Nash equilibrium is PPAD-hard even with only five players. This resolves an open question posed in concurrent work by Bichler and Ghosh [BG26], who established PPAD-hardness in the same player-specific equal-splitting setting when the number of players grows with the instance and asked whether hardness persists for a constant number of players.

Coherent radio emission from magnetically active M dwarfs is largely observed in two varieties: stochastic solar-like bursts and beamed, rotationally modulated emission including auroral pulses. Detailed monitoring of these stars is critical for probing the plasmas and magnetic fields driving the emission. We conducted a search for polarised radio emission from stars within 85 pc, covering $2.2\times10^4$ h of stellar dynamic spectra from the ASKAP Deep Investigation of Neutral Gas Origins (DINGO) survey. We report the discovery of coherent radio bursts from the M4V dwarf WW PsA. In addition to quiescent emission, we detect three long-duration ($\gtrsim$30 min) bursts, each ~100% circularly polarised, in 283.3 h of data. We also report a fourth burst in an epoch not covered by our search, detected using the VASTER pipeline. One of the bursts features a polarity reversal midway through, based on which we favour the electron cyclotron maser (ECM) mechanism over plasma emission. We tentatively propose the bursts are due to rotationally modulated ECM pulses of variable intensity. We infer the required physical properties of WW PsA, including a high obliquity $\beta = 74^{+6}_{-22}$ deg under an oblique rotator model.

Reward post-training of flow models either reweights the model's own samples under a KL penalty or a frozen reference, often for thousands of updates, or backpropagates the reward and moves every sample without checking that the move is worth its size. We introduce MEND, a reinforcement learning method built on proximal velocity matching. MEND caps rewards within each prompt group, so samples that already score well receive no move. Below the cap, it proposes moves along the reward gradient and accepts one only when its capped reward gain exceeds a quadratic displacement price. The model then regresses onto the resulting velocity targets, with no KL term, frozen reference model, or advantage weights. In 100 updates, MEND outperforms Flow-GRPO (about 4k updates) on five of six evaluators at the same distance to base-model images. Under an equal-budget protocol, it surpasses ReFL and DiffusionNFT at every evaluated update across four training rewards, reaching PickScore 24.03 versus 23.92 and 23.43, respectively. A 300-update three-reward run also surpasses the five-reward DiffusionNFT model on all three rewards it trains on. MEND is general and easy to adopt: it applies to any flow backbone with a differentiable reward.

The Sachdev-Ye-Kitaev (SYK) model is a strongly interacting fermionic system. Its dynamics are difficult to analyze with Lieb-Robinson bounds and cluster expansions due to its all-to-all disordered interactions. We prove that, with high probability over the disorder, the SYK model admits a quantum Gibbs sampler with a system-size-independent spectral gap at sufficiently high constant temperatures. This yields polynomial-time preparation of SYK Gibbs states on a quantum computer. While prior non-rigorous computations that suggested SYK thermalization were limited to studying local correlators, we emphasize that our result holds up to arbitrary inverse polynomial trace distance. Our proof introduces the pseudo-Lindbladian approach to fermions then applies two key new ideas: a Bochner inequality for quantum dynamics, and an especially simple SYK analysis based on a comparison to classical dynamics.

We construct linearly twisted spectral triples on $C^\ast$-crossed products $\mathfrak{A}\rtimes_\alpha\mathbb{Z}^d$, starting from a spectral triple $(\mathfrak{A},\mathcal{H},D)$ and an action $\alpha$ admitting an adapted weight. The weight compensates for the growth of the base Dirac operator under the dynamics while preserving the canonical differential structure in the $\mathbb{Z}^d$-directions. The construction takes place directly on the original crossed product and does not require the action to be equicontinuous. We prove that, whenever the base spectral triple is metric, the resulting twisted spectral triple is $\kappa$-metric for every $\kappa>0$: its associated Monge–Kantorovich metric metrizes the weak* topology on the state space of the crossed product. Our framework applies, in particular, to every action preserving the domain of the Lipschitz seminorm of the base triple. In this case, an exponential weight may be chosen so that the induced metric geometry restricts exactly to the original metric geometry on the base algebra. In the equicontinuous case, the weight may be chosen constant, the twist disappears, and the standard crossed-product spectral triple is recovered. We also establish a general finite-mode criterion for convergence in the spectral propinquity. As an application, we prove joint continuity, with respect to the deformation parameter and suitable fluctuations of the flat metric, for the twisted spectral triples associated with crossed products of quantum $2$-tori by fixed automorphisms in $\operatorname{SL}(2,\mathbb{Z})$, including Anosov automorphisms.

Exoplanetary magnetic fields govern atmospheric escape and circulation but have eluded detection in close-in planets. In the ultra-hot Jupiter KELT-9b, atmospheric dynamo is predicted to generate atmospheric fields of 400–4000 G. We search for the circular polarization in atomic lines that a magnetic field would imprint, through the Zeeman effect, on the transmission spectrum of KELT-9b. We observed three transits with SPIRou at the CFHT, on 07 September 2025 and 03 and 06 July 2026, and applied least squares deconvolution to 1398 atomic lines to map the transit-averaged Stokes $V$ and null spectra against orbital semi-amplitude $K_{\rm p}$ and systemic velocity $v_{\rm sys}$. The 2025 transit shows a Stokes $V$ signature of $V/I_c = 9.3 \times 10^{-5}$ at $K_{\rm p} = 219 \pm 11$ km s$^{-1}$, $v = -34.0 \pm 1.2$ km s$^{-1}$, close to the expected planetary velocity. Its false-alarm probability is $1.9 \times 10^{-4}$, formally a marginal detection. Both 2026 transits are of comparable sensitivity but neither exhibits a signal at that location, setting $3\sigma$ limits of $4.9$ and $6.0 \times 10^{-5}$; the 2025 amplitude would have been recovered at $4.9\sigma$ and $4.0\sigma$. The 2025 signature is robust to the tests we applied to it, but qualifies as a marginal detection at best. Spurious signals of comparable strengths occur elsewhere in our Stokes $V$ and null maps, and we cannot exclude that our marginal detection is excess noise. It is not reproduced ten months later.

LLM-driven evolution enables iterative model development, but two practical goals remain underexplored: finding model designs that transfer across related tasks and sustaining improvement when training is expensive. We introduce Population Evolution (PE), a collaborative, hierarchical framework that connects ongoing local searches through shared experimental evidence. PE evaluates code changes across related training instances and shares the results to guide subsequent proposals and promotion to larger training scales. For expensive targets, PE searches small training subsets and screens candidates through peer and intermediate evaluations before full-target training. We introduce RMD-Bench to evaluate both settings across ranking, watch-time prediction, RL algorithm discovery, and LLM/VLM pretraining. Compared with standalone evolution at matched source iterations, PE raises mean best local gains from 7.01% to 8.97% in ranking and from 2.84% to 3.85% in watch-time, while improving the best larger-scale outcome in all three joint-discovery families. In watch-time discovery, PE improves best larger-scale gains with four of five harnesses and all four proposers. On new recommendation datasets under shared target-side calibration, every evaluated PE design improves over the reference in mean performance. Under matched total GPU compute, completed LLM discovery runs yield a best relative accuracy gain of 2.48% and 13 successful candidates for PE, versus 0.92% and none for direct evolution. VLM loss reduction reaches 8.78% versus 5.05% under matched total GPU compute.