Papers

Self-blended images are widely used to train face-swap detectors, but primarily capture blending artifacts. We investigate whether adding illumination inconsistencies improves detection. Temporal Self-Blended Images (T-SBI) transfer lighting statistics between frames of the same video, with the mismatch controlled by luminance difference ({\Delta}L). Using five training regimes and a three-seed comparison of high- and low-{\Delta}L training, we find no evidence of illumination-specific improvements. AUC differences remain within seed variability across four datasets, and an analysis of 506,328 attribute-binned samples shows no preferential reduction in errors under harsh lighting. Instead, T-SBI shifts prediction scores, changing optimal thresholds by approximately 0.34 on FaceForensics++ and 0.30 on Celeb-DF, making comparisons at a fixed threshold misleading. However, T-SBI improves robustness to heavy JPEG compression on DFDC (AUC 0.780 versus 0.696), potentially reflecting greater reliance on low-frequency cues. These findings highlight the importance of evaluating training methods against their intended targets and accounting for threshold effects.

We study Killing vector fields, Ricci solitons and hyperbolic Ricci solitons on the hyperbolic strip $S_a=\{(x,y)\in\mathbb{R}^2:0<y<a\}$ endowed with the metric $g_{S_{a}}=\frac{\pi^2}{a^2\sin^2(\pi y/a)}(dx^2+dy^2)$, and on its unit tangent bundle $T^1S_a$ endowed with the Sasaki metric $g^S$. On the strip, we show that the Gauss curvature is constant equal to $-1$, that the Lie algebra of Killing fields is isomorphic to $\mathfrak{sl}(2,\mathbb{R})$, that every Ricci soliton is trivial (Killing potential, $\lambda=-1$), and that a $2$-Killing field which is conformal, or which is the potential of a hyperbolic Ricci soliton with $\lambda\neq0$, is a Killing field. On the unit tangent bundle, we prove that the Lie algebra of Killing fields of $(T^1S_a,g^S)$ is four-dimensional and isomorphic to $\mathfrak{sl}(2,\mathbb{R})\oplus\mathbb{R}$, that $g^S$ is not Einstein but is $\eta$-Einstein with constant coefficients, $\operatorname{Ric}^S=-\tfrac32\,g^S+2\,\eta\otimes\eta$, and that $(T^1S_a,g^S)$ admits no Ricci soliton. For hyperbolic Ricci solitons we prove non-existence for Killing, left-invariant, and $\theta$-invariant potentials with conformal projection, and we reduce the general $\theta$-invariant case to a system of equations on the strip; the classification for an arbitrary potential is left open. Finally, we show that the $\eta$-Ricci solitons of $g^S$ have a Killing potential and $\lambda=\tfrac32$, $\mu=-2$, and that the $\eta$-hyperbolic Ricci solitons obtained in the same classes of potentials are the trivial ones, with a Killing potential, $\mu=-\tfrac32$ and $\nu=2$.

Complete orthogonal predictive coordinates do not by themselves bind a latent direction to a named intervention. We present a mathematical and synthetic audit motivated by vascular device-vessel suitcordance. Capacity-matched least-squares predictors were exactly equivalent under complete fixed output transforms, whereas an anchor-only observer recovered interpretations only within the span of known perturbation signatures. Six three-dimensional configurations across 64 seeds gave a maximum paired prediction discrepancy of 6.7e-15 but a median untransported edit error of 1.513. Coordinate transport removed that error. Noisy and weak anchors constrained calibration stability, and changing the representation basis required recalibration or verified transport. Across 256 additional fits in dimensions 3-24, prediction equivalence persisted within 4.0e-15. We then evaluated nine deliberate runnable fault classes across 64 seeds. All 576 faulty executions completed, but each violated at least one reconstruction, prediction, delivered-edit or scope contract; all 320 valid control records passed. Repeating a faulty implementation gave exact self-agreement despite error against the separately computed simulator expectation. For one omitted-direction defect, probe coverage followed its analytic law, and rank-aware abstention protected unsupported interpretations. Scalar-noise experiments exposed both missed weak faults and excessive rejection under narrow relative tolerances. These controls provide an executable separation of prediction, semantic support and scientific acceptance. They are synthetic numerical audits, not clinical validation, neural JEPA-Anything replication, agent learning or patient treatment-effect estimation.

This paper examines job waiting times on the Jean Zay supercomputer at IDRIS (Institut du Développement et des Ressources en Informatique Scientifique http://www.idris.fr/), driven by the need to improve scheduling strategies for crossfacility scientific workflows. Drawing on large workload traces, we show that job waiting times are not independent of one another but are instead correlated through job bursts-batches of identical jobs submitted together, also called “flurries”. We introduce a method for quantifying the effect of these bursts on waiting times, and find a strong Pearson correlation between waiting time and a job's rank within its burst. This correlation is above 0.8 for more than 40% of the jobs. We further derive an exact formula for the Slurm fairshare factor based on a convolution product, showing that the burst penalty operates on a much shorter timescale than fairshare itself-pointing to a distinct throttling mechanism specifically targeting bulk submissions. Once burst effects are accounted for, we use Welch's t-test to find that partition occupancy and number of requested cores have a significant effect on waiting time, though the effect is not significant for all partitions.

Neural Radiance Caching (NRC) provides an online neural representation of scene illumination for real-time path tracing. This paper presents an NRC variant tailored to specular lighting through a reflection-direction parameterization and a roughness-dependent radiance target. Our network predicts prefiltered incoming radiance at individual surface points, which is combined with a precomputed BRDF integration map using the split-sum approximation to estimate outgoing radiance. On Bunny and Specular Sponza, the reported comparisons indicate faster convergence to the accumulated path-tracing reference and improved specular-lighting quality relative to the evaluated NRC baselines. Our online-trained method maintains real-time performance while adapting its cache during rendering.

The centralization of web information raises legal and ethical concerns, particularly in social, healthcare, and education applications. Decentralized architectures offer a promising alternative by keeping data closer to its source, yet efficient query processing remains a significant challenge. Link Traversal Query Processing (LTQP) enables querying across decentralized networks but often suffers from long execution times and high data transfer costs due to the large number of HTTP requests involved. Many queries are highly selective with respect to the data model objects distributed across the network. For example, in a social media application where users store heterogeneous data, a query may target only users' posts and comments, ignoring their other information. We refer to such queries as data-model selective. We propose a shape-based pruning approach that relies on shape indexes and a query-shape subsumption algorithm to reduce the search space and thus the number of HTTP requests. We formalize this approach as a link pruning mechanism for LTQP and evaluate it on social media queries from the SolidBench benchmark across multiple metrics. Our results show that shape-based pruning substantially improves query execution time, first-result arrival time, diefficiency, and network usage for data-model selective queries, while having a negligible impact on non-selective data-model queries. These gains cost only a minor increase in triples per shape-index instance. Our approach is also resilient, retaining its benefits even when some data providers do not supply shape indexes. This work demonstrates that shape-based metadata can significantly optimize LTQP in decentralized knowledge graphs for an important class of queries. By exposing such metadata, data providers not only enhance data quality and interoperability but also improve the efficiency of traversal-based query processing.

We introduce a smooth radial approximation of the covariogram of a convex body. The construction replaces the intersection \(K\cap(K+x)\) by the radial \((-p)\)-mean of \(K\) and \(K+x\), producing a family \(g_{K,p}\) which converges pointwise to the classical covariogram as \(p\to\infty\). For each fixed \(p>0\), the function \(g_{K,p}\) is smooth near the origin, and its Hessian is computed explicitly in terms of the operator defining the polar \(L_2\)-projection body. We study the simultaneous limit in which \(p\to\infty\) and the level deficit is \(\delta=p^\alpha\), \(\alpha<0\). The normalized level sets exhibit three regimes: the polar projection body appears for \(\alpha>-1\), the polar \(L_2\)-projection body appears for \(\alpha<-1\), and a critical transition body appears at \(\alpha=-1\). This transition body is described by an explicit \(\log\cosh\)-type boundary integral, placing the critical regime naturally within the framework of Orlicz projection bodies.

Distributed control requires choosing how far information travels and how long it is retained. For networked linear quadratic regulation (LQR), we use System Level Synthesis to bound these resources for a specified performance loss per node relative to centralized control. For locally coupled systems under uniform regularity assumptions, including stabilizability and detectability, we prove that the centralized optimal state and input responses to disturbance impulses satisfy exponential decay bounds in both time and spatial distance. On networks with polynomial neighborhood growth, sufficient communication ranges and memory horizons grow logarithmically with the inverse tolerance, independently of network size. Truncating these responses gives local finite-memory filters with stability and performance guarantees under a computable small-gain condition. Under an additional uniform local feedback condition, we also obtain controllers that confine nominal disturbance effects in space and eliminate them in finite time. Numerical experiments illustrate how performance and disturbance containment requirements guide architecture selection.

Extracting spectral properties such as energy position, broadening, and spectral weight from dielectric spectra is critical for interpreting collective electronic excitations in many-body physics. Conventional approaches typically rely on nonlinear fitting procedures that can be computationally demanding and highly sensitive to initialization and fitting choices. In this work, we develop a convolutional neural-network framework for the direct inversion of dielectric spectra into their underlying plasmonic pole structures using a multipole-Padé representation. Rather than training on a constrained database of spectra associated with a specific set of materials, the network is trained entirely on synthetic spectra generated from the analytical multipole-Padé expression with randomized parameters. This enables the network to learn the general mapping between the spectra and their features in an unbiased way, without requiring large, material-specific datasets derived from real materials. We demonstrate that the synthetically trained network generalizes to complex first-principles and experimental dielectric spectra, extracting the underlying pole parameters with high accuracy in a single forward pass. This approach provides an efficient and robust alternative to conventional nonlinear fitting, enabling high-throughput, automated analysis of dielectric spectra across diverse materials and applications.

Unsupervised learning is one of the main branches of machine learning. Here I argue that unlike the other branches of machine learning (supervised and reinforcement learning), unsupervised learning is a rather heterogenous field that can serve several different goals. It seems futile to try to define one single goal for unsupervised learning. I identify four different goals for unsupervised learning: 1) Estimating the distribution, 2) Generating new data points, 3) Extracting features for downstream tasks, and 4) Understanding the data.

Puzzles are widely used to evaluate the reasoning capabilities of artificial intelligence (AI) systems for sequential decision making, yet approaches originating from different paradigms are rarely compared under unified conditions. To address this gap, we introduce a three-dimensional characterization framework that enables the analysts of AI methods by 1) projecting them to the Markov decision process (MDP) sequential decision making formalism, 2) degree of autonomy through human prior ranking of their designs and, 3) skill and computational cost. Using this framework, we analyze how representative graph-based, reinforcement learning, and large language model (LLM)-based approaches differ in their design choices and performance characteristics, instantiated respectively by Neurosolver, forward-backward reinforcement learning (FBRL), and automated thought-of-search (AutoToS), including a double-agent extension of thought-of-search (DA-ToS). The analysis relies on the Tower of Hanoi puzzle that provides a controlled benchmark with well-defined rules and scalable complexity, enabling consistent comparison across increasing problem sizes. The 3D characterization reveals that LLM-based methods, due to their weakly constrained action-space design, shift complexity from architecture to inference-time verification, leading to substantially higher memory and runtime costs than Neurosolver and FBRL.

This study addresses the problem of structured sentiment analysis, whose goal is to obtain a fine-grained sentiment graph where the nodes represent spans of sentiment holders, targets, and expressions, while the arcs define the relationships among them. Our proposed approach casts the task as dependency graph parsing, but departs from traditional parsing methods by solving it through sequence labeling. To do so, we leverage recent advances in linearized graph encodings that allow each word in the input to be assigned a label, effectively capturing the structure of the dependency graph. We conducted experiments on seven datasets spanning five languages (English, Spanish, Norwegian, Basque, and Catalan), showing performance competitive with leading, more complex single-model approaches.

Existing reaction models primarily learn molecular transformations, whereas enzy- matic reactions depend jointly on molecular structure and catalytic function. We formulate this problem as learning an enzymatic reaction space linking reactants, products, and Enzyme Commission (EC) annotations. To characterize this space, we introduce VenusRX-Bench, a unified benchmark for forward reaction prediction, single-step retrosynthesis, and EC-number prediction. VenusRX-Bench integrates reactions from multiple biochemical databases with standardized curation, leakage- controlled splits, and consistent evaluation. Benchmarking representative chemical and enzymatic models reveals a clear chemical-to-enzymatic domain gap, driven by limited domain data, catalytic-context dependency, and the difficulty of modeling large biomolecular structures. To bridge this gap, we develop VenusRX, a unified T5-style sequence-to-sequence model for enzymatic reactions. VenusRX jointly learns forward prediction, ret- rosynthesis, and reaction reconstruction, with two-stage training on millions of template-expanded reactions followed by real biochemical reactions. In addition, optional EC conditioning incorporates catalytic context, while Molecule Library- Constrained Decoding improves the generation of complex biomolecules. Across benchmark tasks and challenging generalization splits, VenusRX achieves the best or competitive performance on most evaluated settings over representative chem- ical and enzymatic baselines. Moreover, EC information consistently improves reaction prediction, while learned reaction representations support accurate EC prediction, revealing a bidirectional relationship between reaction structure and catalytic function. Together, VenusRX-Bench and VenusRX provide a unified framework for elucidating and modeling enzymatic reaction space

We study second-order methods for convex stochastic optimization, where gradients and Hessians are available only through stochastic estimates with variances $\sigma_1^2$ and $\sigma_2^2$, respectively. First, we propose the Stochastic Contracting Cubic Newton method. At each iteration, it minimizes a cubic model with additional quadratic regularization and then contracts the step toward the current point. After $T$ iterations, the method achieves the expected convergence rate $\mathcal{O}(\sigma_1/\sqrt{T}+\sigma_2/T+1/T^2)$. Building on this construction, we develop an accelerated variant achieving $\mathcal{O}(\sigma_1/\sqrt{T}+\sigma_2/T^2+1/T^{7/2})$, matching the known lower bounds of Agafonov et al. (2024) in all three terms.

Foundation models supply reinforcement learning (RL) with priors that mitigate its longstanding weaknesses in sample efficiency and transfer, but their token-by-token generation makes queries sequential and costly. Jev, a recently released decision model, generates nothing and returns calibrated, typed answers in a single forward pass. Existing work studies foundation models in RL either as models to be trained or as generators to be prompted, and Jev belongs to neither category, having so far served only as a black box in single domains. How well such a model decides on its own in RL environments, and how it can improve RL as a component of training, therefore remain unaddressed. To this end, in this paper we first examine the requirements that the objects of an RL system place on the answers they consume, and establish that Jev can fulfill all of them except the cardinal use of a value function. The remaining objects form positions that admit several roles each. We then construct algorithms that employ Jev at three of these positions, as a reference policy, an exploration judge, and a replay rater, to improve sample efficiency, exploration, and learning performance. Across nine MiniGrid tasks and three Atari games, training with Jev outperforms a standard RL learner, including where the learner makes no progress alone, while the model itself remains untrained. To our knowledge, we present the first use of Jev within the RL learning process and establish a frozen decision model as a usable component of RL training, inviting further exploration of how Jev and other advanced decision models can improve RL.

Structural price diffusivity explains many empirical regularities of market impact including the “square-root law” and its crossover to a linear regime for low trading rates \citep{bonart2026diffusive}. One of its central predictions is that an information-neutral trading strategy generates a diffusive impact state. We microfound this result in an economy with a trader and many arbitrageurs who progressively eliminate predictable returns. Each arbitrageur observes realized returns and a private signal of one component of the fundamental return. As aggregate private information becomes complete and under suitable convergence to a stationary limit, the trader's impact law is of the form $j=U\cdot w$, where $w$ is the innovation in the trading schedule and $U$ is causal all-pass. Impact returns are therefore white, even though no individual arbitrageur can reconstruct the underlying trading strategy. We then investigate the economic meaning of the impact phase. We argue that the $U$ which has minimum distance from the filter $L$ generating the trade flow is an especially interesting candidate: Trade flow is then minimally distorted under impact and arbitrage, and it always guarantees positive impact costs. Under a trader flow generated by a biexponential $L$ (the simplest form allowed in our model) an interesting result emerges: Impact decline is confined to a narrow band of $50\%$–$60\%$, lower than some empirical estimates and quite consistent with others. More complicated flow models can lead to different decays. Finally, we argue that in real markets, the impact phase is probably somewhat “distorted” in its long-term tails which allows for a full relaxation of the impact propagator. The effect is weak, long-term mean-reversion of the impact state.

In the graph model for conflict resolution (GMCR), a decision maker (DM) either moves the conflict to another state or does nothing. The basic definitions leave inaction implicit, so every action that leaves the state unchanged is treated as doing nothing. Yet announcements, exercises, leaks and selective disclosures are neither moves nor inaction: they leave the state unchanged but change what other DMs believe about which moves are available and which moves others would want to make. We introduce such state-preserving actions by augmenting states with the DMs' epistemic states: a physical move changes the physical state, a state-preserving action changes only the epistemic state, and inaction is the absence of a transition. Actions generate evidence through observer-specific interpretation maps. Building on a four-valued extension of GMCR from the author's earlier work, which separates evidence for and against, we show that evidence for a move can only enable perceived moves and evidence against can only disable them, that two of the four reduction operators ignore one kind of evidence, and that contradictory assessments are absorbing under monotone accumulation. With the monotonicity of stability in move sets, this fixes the direction in which any action moves a DM's stability judgements and characterizes when actions can enable provocation or deterrence. Capability assessments affect all sanction-based stability concepts, and on the DM's own side also Nash stability, whereas intention assessments affect only sequential stability. Hedging between two candidate types weakly expands or shrinks an observer's sequentially stable set according to how it reads contradiction. In the 1995 DVD format negotiation, general metarationality cannot distinguish its phases, since the computer industry group could always sanction; sequential stability, which asks whether it would, can.

YRu$_3$Si$_2$ and LaRu$_3$Si$_2$ host the highest charge density wave (CDW) transition temperatures ever reported in kagome metals, $T_{\mathrm{CDW}}\approx 800$ and 400~K, with propagation vectors $Q_{\mathrm{CDW}}\approx(1/2,0,0)$ and $(1/4,0,0)$, respectively. The microscopic mechanism behind these record values has remained unresolved. Here, we combine anharmonic phonon renormalization extracted from molecular dynamics driven by a machine-learned force field with mode-resolved electron-phonon coupling (EPC) analysis to identify this mechanism. The CDW is driven by mode-selective EPC: the Ru out-of-plane phonon modes couple the Ru-$d_{xz}/d_{yz}$ and Si-$p_x/p_y$ orbitals, and the phonon linewidth peaks sharply at $Q_{\mathrm{CDW}}$, while the featureless electronic susceptibility rules out Fermi-surface nesting. Phonon anharmonicity melts the CDW, and the anharmonic phonon spectra reproduce both $Q_{\mathrm{CDW}}$ and $T_{\mathrm{CDW}}$ in good agreement with experiments. Molecular dynamics simulations also visualize the CDW melting in real space. Chemical bonding analysis further shows that the smaller Y$^{3+}$ radius strengthens the Ru-Ru bonds, enhancing lattice rigidity and accounting for the factor-of-two higher $T_{\mathrm{CDW}}$ of YRu$_3$Si$_2$. Our results establish a unified microscopic picture of CDW formation and melting in kagome metals YRu$_3$Si$_2$ and LaRu$_3$Si$_2$.

We develop a localized hybrid finite element–random feature method (FEM–RFM) for elliptic interface problems on unfitted background meshes. The central idea is to use the random feature method (RFM) only in the narrow band formed by interface-cut cells of a uniform Cartesian background mesh,while retaining standard $Q_p$ finite elements on the remaining uncut cells. Within the band, local RFM representations are used on the two physical sides of the interface. The prescribed jumps in the solution and normal flux are incorporated through residual equations involving the two RFM representations on the physical interface $\Gamma$, while the finite element method (FEM) and RFM components are coupled across a separate grid-aligned artificial interface $\Gamma_c$ through trace and variational normal-flux residuals. A Schur reduction eliminates the interior finite element unknowns through sparse solves while retaining their Galerkin equations as equality constraints. The remaining band and interface residuals are minimized in a reduced least-squares problem involving only the coupling nodal values and random-feature coefficients. The method is assessed numerically on a range of two- and three-dimensional interface problems involving high coefficient contrasts, nonsmooth and multiple interfaces, and irregular outer boundaries, together with an application-oriented thermal-homogenization problem for a periodic bicontinuous composite.

The study of the distribution of real sequences in the unit torus has a long and rich history. There are many notions which describe the “pseudo-random” behavior of a given deterministic sequence (often of arithmetic origin), such as equidistribution (which describes the distribution on a “global scale”) or the pair correlation (which is a “local-scale” statistics). Another popular and important statistics, which can be applied to all ranges from global via intermediate to local, is the number variance $V_N(s)$, which describes the mean-square fluctuation of the number of elements of an initial segment of the sequence in an interval of length $s$, when moving this interval around the torus. In the terminology of quantum chaology, where such statistics are used to describe the distribution of the energy levels of a system, pseudo-random behavior is called Poissonian behavior. Since the number variance is an important indicator to qualify pseudo-random behavior, it is quite surprising that the literature does not seem to contain a precise description of the distribution and of the almost sure asymptotic behavior of the number variance in the truly random case (i.i.d.\ random points), even in the comparatively simple setup of uniformly distributed points on the unit torus. In the present paper we calculate the distribution of the number variance with high precision throughout a wide range of the parameter $s$, and determine the precise threshold (in terms of $s$) where the “Poissonian” asymptotics $V_N(s) \sim N s$ breaks for a typical realization of a random sequence. Key technical input comes from the Komlós–Major–Tusnády theorem and the Karhunen-Loève decomposition.

We present a non-perturbative framework for the quantum treatment of vibrational excitation in molecular systems driven by electric fields with arbitrary time dependence. For harmonic potentials with linear dipole coupling, the field-induced dynamics is exactly described by a displaced coherent state, whose evolution is completely determined by a complex phase-space coordinate. Closed-form expressions for this coordinate reveal a clear distinction between resonant and non-resonant regimes. The resulting vibrational populations follow a Poisson distribution determined by a dimensionless time-dependent phase-space displacement, for both resonant and non-resonant driving. Resonant driving results in an enhanced phase-space displacement, whereas finite detuning between the driving and vibrational frequencies leads to an exponential suppression of the phase-space displacement and vibrational excitation. While the approach reduces to the standard time-dependent perturbation theory in the weak-field limit, it remains valid in the strong-field regime, where perturbative treatments fail. Our results provide a transparent phase-space picture of vibrational control and enable non-perturbative state preparation in ultrafast spectroscopy.

Image-to-3D generation has become increasingly capable of producing objects that closely resemble the input image, and an outstanding challenge is to reproduce the depicted object itself, including the specific geometry that defines it. Inscriptions, brand marks, and repeated structures are frequently distorted or lost, despite being critical to object identity. We present Hi3D 3.0, an image-to-3D generation system targeting object-specific fidelity, with Twinkle3D as its geometry model for generating watertight triangle meshes at $2048^{3}$ resolution. Twinkle3D advances high-fidelity geometry generation along four dimensions. First, while O-Voxel/FaithC offers high representational precision, it often suffers from poor surface quality and non-watertight geometry. We address both issues while retaining its $2048^{3}$-level precision. Second, we scale diffusion generation to sequences of up to 300K geometric tokens through a redesigned DiT architecture and large-scale distributed training optimizations, reducing training time per step from approximately ten minutes to ten seconds. Third, subsequent refinement cannot fully compensate for errors introduced during initial generation; we therefore strengthen both global shape and local detail in the initial generation stage, and the resulting single-stage model surpasses prior two-stage pipelines with $512^{3}$ refinement. Finally, we introduce a fine-grained image-3D cross-modal interaction mechanism that strengthens correspondence between visual evidence and geometric tokens, improving the recovery of object-specific structures. We evaluate geometric fidelity using alignment metrics derived from silhouettes and normal fields. Hi3D 3.0 outperforms four commercial systems across all reported metrics, recovering 82.1% of inscribed characters at 98.2% precision, compared with 21.7% recall for the strongest competitor.

Selective databases such as Scopus periodically re-evaluate the titles they index and remove those that no longer meet their quality criteria. Once a journal is removed, its subsequent output is no longer recorded, and the consequences of this process have therefore received little attention. This study analyzes the publication output and citation impact of 514 sources delisted from Scopus between 2018 and 2024, based on 678,162 documents retrieved from OpenAlex over a window of three years before and after delisting. Differences between periods are tested with the Wilcoxon signed-rank test for paired samples, globally and by geographic region, field of knowledge and SJR quartile. Output grows until the year of delisting and falls sharply afterwards. The median annual output per journal drops from 35 to 22 documents (-37.1%; p < 0.001; r = 0.30), 61.7% of journals publish less, and 84 (16.3%) have no documents recorded in OpenAlex after delisting. Citations per document, measured over a three-year window, barely change when each journal is compared with itself (-2.4%; p = 0.079), and journals are almost evenly split between those whose impact falls and those whose impact holds steady or rises. The results indicate that delisting makes these journals less attractive as publication venues but does not penalize the citation of the work they publish to the same extent, which raises concerns from a research integrity perspective.

Fusing a learned camera-noise fingerprint with an RGB appearance backbone is an appealing route to generator-independent deepfake detection, because the noise residual is grounded in image-formation physics rather than in the texture statistics of a particular generator. We test, on FaceForensics++, whether a Noiseprint++ residual channel carries information complementary to an RGB Xception backbone for face-swap detection. A three-model ablation (RGB-only, residual-only, late-fusion) shows that fusion does not improve over RGB alone and that the residual branch alone is near chance. A seven-level bottleneck diagnostic localizes the cause: the noise maps do carry a discriminative signal, but it is statistical—carried by the per-sample first and second moments (mean, variance, energy) of the residual—and the per-sample InstanceNorm layer placed at the noise-branch input, following the TruFor template, standardizes exactly those moments away (five-fold cross-validated AUC drops from $0.747$ to $0.554$). A context-crop control rules out cropping geometry, and two fixed-fusion variants that remove the bottleneck recover the statistical signal yet still fail to beat RGB on every dataset. We conclude that, on this manipulation distribution, the noise residual is redundant with RGB rather than complementary, and we give concrete guidance for practitioners adopting noise-residual fusion for face-swap detection.

Magnetic skyrmions are typically nucleated via localized excitation methods, such as laser heating, spin-orbit torques, or engineered defects, and their controlled transport is often limited by the skyrmion Hall effect. Here, we demonstrate that spatial gradients in rare-earth–transition-metal alloys, achievable through patterned laser writing, enable both deterministic nucleation and controlled transport. The strategy leverages engineered compositional gradient to cross the magnetization compensation point, defining regions with opposite magnetization. An out-of-plane magnetic field selectively reverses the magnetization, generating a skyrmion. Crucially, the same compositional design enables a channel tuned to the spin angular momentum compensation point, where skyrmions are driven by spin-orbit torques at high velocity with vanishing skyrmion Hall effect. Local composition variations act as pinning sites, enabling threshold-controlled trapping and release.

We study near-surface tension instabilities in bilayered cylinders composed of an incompressible core coated with a thin compressible film. In theory, circumferential wrinkles with axial wavefronts may appear on the outer surface when the cylinder is stretched. We analyze these instabilities through modelling, numerics, finite element simulations, and controlled experiments. We show that the wrinkle amplitude and wavelength depend on the Poisson ratio of the coating and the ratio of the film and substrate shear moduli. We find that the lower the film's Poisson ratio, the earlier the wrinkles emerge under tension. This observation motivates us to use auxetic films (with negative Poisson ratios) with bespoke microstructures in the finite element simulations and the experiments. For the latter, we fabricated 3D-printed soft cylinders with stiff auxetic outer layers and subjected them to axial loading. Although simulations predicted clear wrinkle formation, experimental observation proved more elusive due to the small amplitude of the wrinkles.

We study intermediate coverings of spectral families, Cantor sets, and projections of measures. Two perturbation families of a fixed ReLU tangent kernel have identical source thresholds. Their Brownian images have equal box thresholds and distinct intermediate thresholds, almost surely along prescribed geometric scales. For Cantor sets with delayed contractions, we obtain entropy formulas for the intermediate dimensions of the sets and their fractional Brownian images. These formulas yield endpoint asymptotics and a sharp separation theorem for the first phase transition. An oscillating variant has distinct lower and upper intermediate dimensions while preserving the Assouad spectrum. Finally, almost every orthogonal projection preserves both intermediate dimensions of a nonzero finite compactly supported measure whose quasi-Assouad dimension does not exceed the target dimension.

We introduce positively transverse reaction networks and show that, with mass-action kinetics, no bounded trajectory with positive initial condition has a boundary omega-limit point. These networks are characterized by a geometric property of a collection of polytopes defined by a subset of complexes of the network.

Verifier scores now serve as both benchmark metrics and training rewards for large language model (LLM) agents, and a change in score is routinely read as a change in capability. It may instead reflect a change in the evaluation. We introduce TRACE, a protocol that turns a score change from a verdict into a testable diagnosis: it applies a targeted change to one part of an evaluation, compares paired runs, checks whether the agent's behavior changed, and rescores unchanged trajectories to test whether the scoring rule is responsible. In a controlled suite of 25 synthetic tasks, renaming tools lowers a scripted agent's score by 0.250 even though it performs exactly the same operations; restoring the original names at scoring time closes the entire gap, while the same mutation exposes a genuine behavioral failure in a second agent. On public $\tau^2$-bench tasks with four LLM agents, an initial 30-task study finds mixed reward changes whose one clear effect does not replicate. In a larger follow-up on 88 new tasks with repeated runs per condition, renaming tools or reformatting tool outputs leaves reward unchanged to within $\pm$0.10 for seven of eight agent-change pairs, whereas tool names that deliberately mislead lower every agent's reward by 0.20-0.44, showing that the setup can detect real effects. Identical reruns flip 15-36% of task outcomes, so single-run comparisons cannot separate presentation effects from run-to-run variation. Two frontier LLM judges give consistent verdicts when a fixed trajectory is presented differently, yet disagree with each other on 57% of the same records, largely because one grades procedure rather than outcome. TRACE thus separates what a score change says about the agent from what it says about the measurement.

We investigate weighted perimeters of convex bodies in $\mathbb{R}^n$. Our main result states that the weighted perimeter is continuous with respect to the Hausdorff metric for every radial weight function. We also establish monotonicity under inclusion under suitable assumptions on the radial weight, and show that these assumptions are sharp. These results are applied to show the existence of minimizers for a shape optimization problem.