Peer Revenue Now accepting positions

Papers

Fresh from arXiv, typeset properly. Highlighted words trade on the market.

All of arXiv

Friday 6 November

Computation and Language

Are These Programmers Ripe Enough to Eat? Evaluating How LLMs Handle Lexical Knowledge Conflicts in Translation

External lexical resources can help large language models (LLMs) translate texts in low-resource languages and specialized domains, but these resources may conflict with models' prior lexical knowledge. To isolate this conflict, we construct a controlled Mandarin-to-English translation task in which familiar source words are assigned new meanings within and across semantic categories. We treat these prompt-supplied meanings as contextual knowledge (CK) and systematically evaluate the effects of word frequency, sentence context, and prompt framing on the translated outputs. Across nine models, CK reliance is greater for less frequent source words and in sentence contexts that support the supplied meanings. Prompt framing can reduce CK reliance, but seven of nine models favor CK even when explicitly instructed to ignore the supplied meanings. Within model families, larger models are more likely than smaller models to utilize CK in their outputs. Three follow-up experiments further show that modal weakening lowers CK reliance, the effect of repetition depends on prompt layout, and the effect of reasoning varies across models. Overall, our controlled study reveals how CK reliance varies across conditions and models, motivating new methods to detect and evaluate lexical conflicts and guide models' lexical choices in LLM-based translation.
Read the full abstract

Optics

Time-gated optical-parametric-amplification and field-resolved detection of mid-infrared free-induction decay signals

Infrared spectroscopy is a key tool to study complex structural dynamics in molecular systems. However, resolving weak or broad vibrational modes is challenging due to their rapid decay. Here, we demonstrate a technique to significantly extend the detectable lifetime - and thereby spectral resolution - of molecular signatures by employing synchronized time-gated optical parametric amplification of mid-infrared free-induction decay signals. Utilizing a Cr:ZnS laser system and a nonlinear ZnGeP2 crystal, we achieve a 10-fold increase in the electric field strength of the infrared molecular signal, yielding a two-orders of magnitude enhancement in dynamic range and a 2-fold increase in the signal-to-noise ratio. Using isopropyl alcohol as a model system, we demonstrate that OPA-enhanced detection extends the observable window of vibrational modes up to 72 %. This approach allows for the reconstruction of high-fidelity molecular fingerprints that resolve complex vibrational relaxation pathways and weak structural dynamics, previously hidden within the background and noise, providing a pathway for new high-sensitivity and time-resolved studies of complex chemical structures.
Read the full abstract

All authors Sergio Revuelta, Gerard Faustino, Ka Fai Mak, Ferenc Krausz, Nicholas Karpowicz

Statistics Applications

From Point Anomalies to Anomalous Driving Episodes: Interpreting Abnormal Driving from Vehicle Telematics

Vehicle telematics provides detailed observations of driving behaviour that can be useful for automobile insurance risk assessment. Existing approaches often summarize these data through counts of harsh events or individually unusual observations, although driving movements naturally develop over time. Building on a previously proposed hidden Markov model (HMM) framework that identifies pointwise anomalies, we identify short intervals in which a sequence of observations is unusual when considered together. We then examine the original acceleration signals to measure the magnitude and temporal concentration of these episodes and identify recurring patterns of vehicle movement. Temporal concentration describes whether acceleration activity is distributed evenly across an episode or concentrated in relatively few observations. Using a controlled telematics dataset containing normal, aggressive, and drowsy driving by the same drivers, we evaluate the proposed measures through comparisons matched by driver and road type. We find that statistical anomaly measures depend on the driving population used to train the HMM. More importantly, mean temporal concentration alone distinguishes normal from abnormal driving more effectively than the original pointwise anomaly indices across different reference populations. The recurring movement patterns further reveal different characteristics of abnormal driving, with stronger longitudinal acceleration during aggressive driving and more frequent lateral corrections or excursions during drowsy driving. The framework therefore goes beyond pointwise anomaly detection, providing interpretable behavioural features that can be calculated for unlabeled trips, used for trip-level assessment, aggregated across a driver's trip history, and subsequently evaluated alongside existing telematics and rating variables for insurance risk modelling and ratemaking.
Read the full abstract

Analysis of PDEs

The variational formulation of the follow-the-leader model

We provide a variational formulation of a finite-dimensional Follow-the-Leader system through a minimizing-movement scheme based on a discrete transport distance with nonlinear mobility. Despite the degeneracy of the kinetic action at maximal density, we prove finiteness of the distance, $C^1$ regularity of geodesics, and suitable Euler–Lagrange relations. For a decreasing external potential, we show that the JKO scheme yields an approximate implicit Euler discretization of the Follow-the-Leader dynamics. Consequently, as $\tau\to0^+$, the piecewise affine interpolations converge, up to a subsequence, to a $C^1$ solution of the continuous Follow-the-Leader system. This complements the many-particle convergence theory for the discrete nonlinear-mobility transport distance by establishing the continuous-time limit at fixed number of particles.
Read the full abstract

Game Theory

Strategic Competition and Convergence in LLM Leaderboards

Leaderboards are a popular way to compare large language model (LLM) capabilities and can influence how model developers invest resources. We ask when competition on leaderboards leads developers to produce models with similar capability profiles, which we call monoculture, or similar performance levels. We study this by formulating leaderboard competition as a game in which model developers invest resources across capabilities to maximize their win rate. We show that similar production functions can induce similar capability profiles, while comparative advantage can sustain distinct capability profiles. We then study convergence in performance levels and show that diminishing returns can lead to an increasingly crowded competitive frontier. Using data from LM Arena and OpenRouter, we find that developers improve more in categories where they lag competitors, while still retaining category-specific advantages. At the same time, we find that the frontier of models becomes increasingly crowded, alongside increasingly diversified model usage patterns. Our results suggest that increasing competition does not need to produce interchangeable models, but can instead sustain competitive models with distinct strengths.
Read the full abstract

Quantum Physics

Differentiable Digital Twins for Optimization, Calibration, and Characterization of Neutral-Atom- and Ion-based Quantum Computers

Quantum technologies based on individually controlled atoms are among the most promising platforms for scalable quantum computing, owing to their favorable coherence properties and controllability. Nonetheless, achieving reliable and high-fidelity gate operations remains challenging due to the sensitivity of these systems to noise and hardware imperfections. In this work, we introduce a comprehensive software toolset designed to model atomic quantum systems with high precision and simulate qubit gate dynamics. By integrating quantum optimal control and benchmarking techniques with automatic differentiation for efficient and accurate gradient-based optimization, the toolset enables significant improvements in gate performance. We demonstrate its capabilities through detailed case studies involving both neutral atoms and trapped ions, highlighting its effectiveness in enhancing quantum gate fidelities, and providing valuable insights for experimental implementation.
Read the full abstract

All authors Marco Rossignolo, Lorena C. Bianchet, Anurag Saha Roy, Shai Machnes

Machine Learning

Quasipolynomial-Time Planning for Decodable POMDPs

We study planning for finite-horizon partially observable Markov decision processes (POMDPs) with finite state, action, and observation spaces. Both exact and approximate planning are known to be computationally hard in this setting, so prior work has explored natural assumptions that make planning easier. Recent work [Golowich et al., 2022] gave quasipolynomial-time approximate planning algorithms for observable POMDPs. These are POMDPs whose observation channels satisfy an $\ell_1$-observability condition, but notably, no assumptions are required on the transition dynamics. In a similar vein, for every fixed $\rho>0$, we give quasipolynomial-time approximate planning algorithms for $\rho$-decodable POMDPs. We show that decodability is implied by both $\ell_1$- and $\ell_2$-observability and is strictly weaker than observability. Moreover, we show that the optimal decodability constant can be computed in polynomial time, whereas computing the optimal $\ell_1$-observability constant is NP-hard in the worst case. We also show that the fast filter-forgetting guarantee underlying the approach of [Golowich et al., 2022] does not extend directly from $\ell_1$-observability to constant $\ell_2$-observability. Thus, decodability is an easy-to-check, unifying assumption under which quasipolynomial-time approximate planning for POMDPs is possible.
Read the full abstract

Robotics

Autonomous Whole-body Loco-manipulation Using a Snake Robot

Snake robots can leverage their hyper-redundant morphology for both locomotion and manipulation through distributed body contact but whole-body loco-manipulation is particularly challenging since the entire body serves simultaneously as propulsive mechanism and manipulation tool Prior work on wheeled legged and snake robot platforms has decoupled these functions through dedicated appendages body segmentation or grippers We present closed-loop loco-manipulation with our snake robot COBRA without this distinction such that any part of the body may make contact We develop a contact-implicit dynamic model unifying robot-ground robot-object and object-ground interactions within a Moreau-Jean time-stepping framework and formulate a trajectory optimization over the full-order dynamics guided by a geometric planner and nominal proportional controller to autonomously push an object to a goal location during sidewinding We validate the planner and controller on hardware across varied configurations and present simulation results for a simplified trajectory optimization that resolves robot-object and object-ground contacts
Read the full abstract

All authors Adarsh Salagame, Henry Noyes, Haroon Hublikar, Alireza Ramezani

Probability

The Localization Scale of the Block Anderson Model in One Dimension at Arbitrary Coupling and Volume

We study the one-dimensional block Anderson model with independent $W\times W$ Gaussian unitary ensemble potentials and deterministic nearest-neighbor hopping, with hopping-to-disorder ratio $g$. The system consists of $L$ blocks, with total dimension $N=WL$. As $W\to\infty$, we identify the bulk localization length, up to an arbitrarily small power of $W$, with $\min\{N,\max\{W,W^2g^2\}\}$, uniformly for deterministic couplings $g\ge0$ and cycle lengths $L\ge3$. For every fixed $D>0$ and all sufficiently large $W$, the expected fraction of eigenvectors in the selected energy interval satisfying these bounds is at least $1-W^{-D}$. In polynomial volume the bounds hold simultaneously for all eigenvectors in that interval with probability at least $1-W^{-D}$. The localization length is the number of coordinates in the shortest cyclic interval containing at least half of an eigenvector's mass.
Read the full abstract

High Energy Physics (Phenomenology)

Uncovering TeV-scale dark matter with Z-fusion at a muon collider

We demonstrate the advantage of the $Z$-boson fusion channel at a high-energy, high-luminosity muon collider in eliciting signals of a TeV-scale scalar dark matter (DM) particle. The utility of such a set-up lies in (a)the attainability of high centre-of-mass energy ($\approx 10 ~\rm TeV$), (b)the likely availability of large integrated luminosity ($\gtrsim 10~ab^{-1}$), and (c)the facility for efficient detection of very forward tagged muons. Our analysis is based on the process $\mu^+ \mu^- \rightarrow \mu^+ \mu^-$ + missing energy, driven by $Z$-fusion. We present the results in the context of (i) a scenario characterised by dimension-6 effective couplings, and (ii) an inert doublet model. After filtering each scenario through its associated constraints, we find that it is possible to achieve a signal significance of about 6.7$\sigma$ in case (i) and about 5.5$\sigma$ in case (ii), for DM particle masses $\approx$ 1 TeV and an integrated luminosity of $10 ~ ab^{-1}$.
Read the full abstract

Algebraic Geometry

Exploring the Regular and Singular Loci of Group Representation Varieties

Let $B$ be the Borel Subgroup of $\mathrm{SL}_2(\mathbb{C})$, and let $G_m=\langle x,y,z\mid x^2y^3=z^{2m+1}\rangle$, $m\geq2$. The $G_m$ are parafree of rank two and deviation one. We compute the integral homology of the regular and singular loci of $X_m=\mathrm{Hom}(G_m,B)$. Each $X_m$ is a four-dimensional irreducible affine variety with the homotopy type of a two-dimensional torus. Its singular locus is a disjoint union of $16m$ affine lines. The regular locus has Betti numbers $b_0=1$, $b_1=2$, $b_2=16m+1$, and $b_3=b_5=16m$, with all other Betti numbers zero. Thus, although each $X_m$ has the homotopy type of a torus and its singular locus is a disjoint union of contractible lines, removing the singular locus creates $16m$ additional independent homology classes in each of degrees $2$, $3$, and $5$, arising from the transverse link $S^2\times S^3$. We also establish that the regular loci are not affine.
Read the full abstract

Computational Complexity

A proof of the Aaronson-Ambainis conjecture

Aaronson and Ambainis conjectured that every bounded polynomial of low degree on the discrete cube has an influential variable. They showed that the conjecture implies that the acceptance probability of every quantum algorithm making $T$ queries can be approximated on most inputs by a classical algorithm making $\mathrm{poly}(T)$ queries. We prove their conjecture: every function $f\colon\{-1,1\}^n\to[-1,1]$ of degree at most $d\ge1$ has a variable of influence at least $c\,\operatorname{Var}(f)^{5/2}/d^{6}$, where $c>0$ is an absolute constant. In the original formulation, for polynomials of degree at most $d$ with values in $[0,1]$ on $\{0,1\}^n$ and variance at least $\varepsilon$, this gives a variable of influence at least $(\varepsilon/d)^{7}$. The proof is elementary, with steps inspired by works of others. Thus, as Aaronson and Ambainis anticipated, superpolynomial quantum speedups in the query model require structure, such as a promise on the inputs.
Read the full abstract

Probability

Malliavin structure for conditionally independent random variables

We consider a countable family of random variables which are conditionally independent given a latent random variable $Z$. Given $Z$, the random variables are independent. Hence the Malliavin calculus for independent random variables applies for each value of $Z$. We show that these constructions fit together. We obtain the chaos decomposition, the domains of the Malliavin operators, the Mehler formula and the carré du champ operator, with conditional expectations given $Z$ in place of ordinary expectations. For normal approximation, conditioning on $Z$ and the infinite de Jong theorem of Döbler give mixed normal approximations of degenerate U-statistics, in Wasserstein and Kolmogorov distances. Working under the unconditional probability, we obtain a de Jong type bound for U-statistics whose kernels are products. In both cases, the fluctuations of the conditional variance appear as an additional term, which cannot be removed. As an application, we consider subhypergraph counts in a random $3$-uniform hypergraph built on a latent random graph. For the count centred at its conditional mean, we obtain an explicit rate in Wasserstein distance. It reduces to the known rate for the Erdős–Rényi hypergraph when the latent graph is complete. The count centred at its mean is a pattern count in a non-uniform hypergraph with independent edges, which gives a necessary and sufficient condition for its asymptotic normality.
Read the full abstract

Cryptography and Security

A Cross-Domain TARA-to-NAC Rule Mapping Methodology: From Automotive OEM Networks to Energy Storage Systems

Threat analysis and risk assessment under ISO/SAE 21434 ends with cybersecurity goals, but existing methods give little structured guidance for turning those goals into enforceable network access control rules. Compliance standards such as IEC 62443, ISO/SAE 21434, and ISO/IEC 27001 are usually applied as parallel efforts rather than as one enforcement architecture. This paper presents a six-step methodology that maps cybersecurity goals from a threat analysis and risk assessment to network access control rules, each placed at a named enforcement point and traceable to the threat records behind it. It also presents a unified compliance framework that treats IEC 62443 and ISO/IEC 27001 as a fixed substrate and domain-specific regulation as an interchangeable overlay. The methodology is applied to an automotive original equipment manufacturer plant network, addressing multi-tier supplier authorization and legacy operational technology devices, and then to utility-scale battery energy storage, where North American Electric Reliability Corporation Critical Infrastructure Protection standards and ISO/IEC 27019 replace the automotive-specific standard while the substrate stays constant. In the automotive example, 37 threat entries and 24 cybersecurity goals yield 23 rules covering 11 of the 13 in-scope goals; the energy example yields six rules. Three derived rules have no equivalent in the cited standards. Both examples are desk-based: the methodology is applied to them, not tested on them. The result is a reusable pattern for translating risk assessment into enforcement across critical infrastructure sectors, with empirical validation left as future work.
Read the full abstract

Computation and Language

ROSE: Risk-Aware Orthogonal Subspace Navigation for Lifelong Knowledge Editing in Multimodal Large Language Models

Multimodal Large Language Models (MLLMs) require continuous factual updates to stay current, yet lifelong knowledge editing remains a daunting challenge. The primary obstacle is the severe Locality Erosion and Catastrophic Forgetting, where sequential edits inevitably interfere with pre-trained capabilities and prior updates. In this paper, we propose ROSE, a Risk-aware Orthogonal Subspace navigation framework for lifelong knowledge Editing. ROSE addresses forgetting by aiming to restrict parameter updates to the orthogonal complement of previously edited subspaces. Crucially, it introduces a risk-guided mechanism that utilizes a dynamically computed risk mask to safeguard specific factual parameter subspaces essential for model stability. During inference, ROSE employs a multi-granularity knowledge integration strategy, featuring prototype-anchored semantic gating and perturbation-driven subspace fusion to precisely activate edited knowledge for relevant queries while limiting interference on unrelated inputs. Comprehensive experiments on two prominent multimodal benchmarks show that ROSE is competitive with state-of-the-art approaches across five key metrics, while maintaining outstanding stability over $1,000$ sequential updates. Our code is available at https://github.com/sunlitsong/ROSE.
Read the full abstract

All authors Lingyun Song, Ziyao Chen, Kang Pan, Xiaolin Han, Yifei Zhang, Xiaoqi Wang, Yudai Pan, Xuequn Shang

Software Engineering

AnimWeb-Bench: Benchmarking Animated Webpage Code Generation from Videos

Animation is pervasive on the Web, the CSS animation property alone appears on over three-quarters of webpages. Yet existing webpage code generation benchmarks target static screenshots, user-triggered interactions, or isolated motion graphics, leaving the ability of multimodal large language models (MLLMs) and coding agents to reconstruct complete animated webpages largely unexplored. To fill this gap, we introduce AnimWeb-Bench, the first benchmark for animated webpage code generation from videos. It comprises 124 real-world webpages spanning 7 application domains and 33 animation types, with 91.9% of pages combining multiple types, each paired with a reference video and the static assets of the original webpage. We further propose a human-aligned, two-dimensional evaluation protocol: Static Appearance Fidelity measures structure, text, color, and visual similarity on aligned keyframes, while Animation Fidelity tracks UI components across frames and compares their position, size, velocity, and appearance with those in the reference video. Evaluating 14 MLLMs and three coding agents reveals: (1) reproducing webpage animations is substantially more difficult than reproducing static appearance; (2) explicit frame inspection and render feedback improve the reconstruction quality of coding agents; and (3) rapid, continuous changes are harder to reproduce than discrete state transitions. Our code and dataset are available at https://starry1001.github.io/AnimWeb-Bench-website/ .
Read the full abstract

All authors Zhongyi Zhang, Jingyu Xiao, Shuqing Li, Michael R. Lyu

Computation and Language

Language as Trait, Politeness as Strategy: Controlled Tests of Linguistic Effects in LLM Social-Deduction Agents

Large language model (LLM) agents coordinate through conversation, and the same message can be phrased in simpler or more complex language, bluntly or politely. We test whether such differences change how other agents judge and act, using the LURK Werewolf framework, which separates each utterance's speech-act predicate from its natural-language realization. We rewrite only the realization, check that the game content is preserved, and replay each game day from saved checkpoints against a re-sampled control. We vary linguistic range consistently across a speaker's messages and politeness strategy per utterance. Across 30 games and 4,335 replays, neither manipulation changes the trust other agents place in the speaker, and no effect on votes or night actions survives correction for multiple comparisons; the largest uncorrected effect is that shifting range away from a neutral paraphrase changes about 4 percentage points more of the other players' votes. In natural play, 88% of face-threatening speech is bald on-record regardless of role.
Read the full abstract

All authors Yuchen Li, Lisa Liu, Runqi Zou, Shuxuan Kuang, Qingquan Zhang, Julian Togelius

Machine Learning

From PAC bounds on the generalization error to PAC bounds on the parameter estimation error

Probably Approximately Correct (PAC) bounds on the generalization error, the difference between the true prediction error and the empirical prediction error, are an established tool in statistical machine learning, and there have been several extensions of these bounds to dynamical systems. In parallel, in recent years, there has been growing interest in finite-sample bounds for parameter estimation error in system identification. The latter bounds tend to be algorithm-specific, relying on reformulating the identification algorithm as a least-squares regression problem. This is in contrast to classical PAC bounds, which are algorithm-agnostic and can be applied to any identification algorithm based on minimizing a cost function involving the empirical loss. In this paper, we show how oracle bounds derived from PAC bounds can be used to obtain finite-sample bounds for dynamical systems. As an illustration of this approach, we present finite-sample parameter-estimation bounds for linear and bilinear systems with bounded inputs and noise. These bounds apply to identification algorithms based on minimizing the empirical loss function.
Read the full abstract

All authors Mihaly Petreczky, John Leth, Emma Lukacs, Balint Daroczy

Fluid Dynamics

Bathyspheres, dust devils, lunar plumes and Martian drones: a generalised mathematical model for particle-fluid flow fields generated by static particle bed entrainment across macroscopic scales

We consider a recently derived mathematical model for the mechanism by which a wind driven flow entrains particles into a fluid from an otherwise static particle bed, a process that is fundamental in initiating and driving the phenomenon of particle cloud formation. This model was previously applied to the "helicopter brownout" problem, in which, as a helicopter descends towards a bed of sand, the downwash and swirling flow from the rotors entrain particles into the air, forming a high-velocity particle-laden cloud around the helicopter. Here, we define a new measure of particle cloud formation potential, outline how the model can be incorporated into existing Computational Fluid Dynamics software, and extend the model to scenarios including: formation of particle clouds in the absence of swirl, such as those caused by a bathysphere descending to the ocean bed, or a rocket landing on the lunar surface, in which case we can precisely predict the critical height above the surface at which particle cloud formation is initiated; generation of dust clouds by less powerful downwash and swirling flows caused by drones or other smaller rotorcraft, such as NASA's Ingenuity Mars helicopter, in which case gravitational effects are included in the model; formation of dust devils (dust clouds formed within and around an upwardly circulating columnar air flow), in which case downwash is replaced by updraft. In each case, predictive models are derived through minor adaptations to key modelling assumptions.
Read the full abstract

Probability

Limit laws for the friendship paradox in random geometric graphs

The friendship paradox says that, on average, the neighbours of a vertex have more neighbours than the vertex itself. We study the size of this effect, the average over all vertices of the mean degree of the neighbours minus the degree of the vertex, in the random geometric graph on the torus with the maximum-norm distance. We determine its limit law in every regime of the connection radius and quantify the approximation by non-asymptotic error bounds in the Kolmogorov distance or the total variation distance. We also give its exact mean for every $n$ and every radius below one half. For a fixed radius the limit is not Gaussian but a weighted sum of centred exponential variables; when the radius shrinks and the expected degree grows, the statistic is asymptotically Gaussian, with a Berry–Esseen bound; for bounded expected degree it is Gaussian at the classical rate; and when the expected degree tends to zero it is asymptotically Gaussian, Poisson or identically zero, depending on how fast the radius shrinks. When the expected degree is large, the proofs rest on the vanishing of the first Hoeffding projection, which makes a degenerate $U$-statistic of pairs of points the source of the fluctuations; it is analysed by Fourier series for a fixed radius and by Stein's method of exchangeable pairs for a shrinking radius. For bounded expected degree we use normal approximation for local functionals, and for vanishing expected degree the Chen–Stein method.
Read the full abstract

Data Structures and Algorithms

The k-Clique hypothesis is dead, for good

We present an algorithm that solves the $k$-Clique problem for any large enough $k$ and any $n$-vertex graph in $n^{\left(\frac{\omega}{3}-c\right)k}$ time, where $\omega\geq 2$ is the matrix multiplication exponent and $c>0$ is a positive constant independent of both $\omega$ and $k$. This refutes the long-standing $k$-Clique conjecture, regardless of the value of the matrix multiplication constant, even if $\omega=2$. In fact, our algorithm solves the more general problem of counting weighted $k$-Cliques.
Read the full abstract

Networking

Link-Level Performance Comparison of 5G Beamforming: Reciprocity-Based SVD vs. Codebook-Based Precoding

Downlink (DL) precoding in mid-band MIMO relies on uplink (UL) sounding reference signals (SRS) with reciprocity-based singular value decomposition (SVD) or channel state information reference signals (CSI-RS) with codebook-based precoding matrix indicator (PMI) feedback. Although SVD precoding offers unquantized beamforming, it depends on a channel estimate from SRS transmitted at the limited user equipment (UE) power, whereas Type-I PMI is quantized but derived from a high-power CSI-RS based channel estimate. In this paper, we compare SRS-SVD and CSIRS-PMI for single-user MIMO through an end-to-end, 3GPP-compliant link-level simulation at 3.75 GHz with 100 MHz bandwidth, using a 192 antenna element gNB mapped into 32 ports, the 3GPP Release 19 UE antenna model, and ray-traced (RT) channels in Manhattan and Chicago. We show that neither scheme is uniformly superior: CSIRS-PMI serves weak and shadowed users better and reduces outage, whereas SRS-SVD achieves higher spectral efficiency (SE) and enables more multi-layer transmission for users with strong channels. The results on the Clustered Delay Line (CDL)-C statistical channel also confirm this crossover. We further show that by utilizing a simple logistic regression based on the received power of the standard CSI reference signal (CSI-RSRP) and the channel quality indicator (CQI), the gNB can select the optimal transmission scheme with around 90% accuracy, achieving an SE that closely approaches an ideal per-user selection.
Read the full abstract

Probability

Percolation on grandparent graphs of Bienaymé trees

We consider Bernoulli bond percolation on the grandparent graph of a Bienaymé tree, obtained by adding to the tree an edge between every vertex and its grandparent. We show that the critical parameter $p_c$ is the unique solution of an explicit equation involving the generating function of the offspring distribution, and we study how $p_c$ depends on this distribution beyond its mean $m$. When $m\geq 1.569$, the offspring distribution concentrated on the (one or two) integers closest to $m$ is the unique maximiser of $p_c$ among all offspring distributions with mean $m$, and its critical parameter is $\frac1{m^2}-\frac1{m^3}+O\left(\frac1{m^4}\right)$ as $m\to\infty$; more generally, for two distributions that are comparable in the convex order, $p_c$ is larger for the smaller of the two distributions under the condition that its support bounded some explicit $N(m)$ of order $m^4$. These restrictions cannot be removed: the convex order alone does not determine the order of the critical parameters, even among distributions with support bounded by $O(m^4)$, and the extremal property fails for $m$ close to $1$. Finally, for every $m>1$, the critical parameter $p_c$ can be arbitrarily close to $0$ for general offspring distributions with mean $m$. If we only consider offspring distributions that do not charge $0$, so that the obtained tree is almost surely infinite, this remains true if and only if $m\geq 2$.
Read the full abstract

High Energy Physics (Theory)

Remark on residual solutions in the $KBc$ algebra

In this paper we study residual solutions of open bosonic string field theory in the $KBc$ algebra with the auxiliary identities $(\partial^mc)^2=0$ of the ghost field. These solutions are not covered by the classification based on the defining relations alone. We show that the operator which controls the cohomology on the residual family depends only on $s=\gamma_1+\gamma_2$. At positive level $n$, cohomology can appear only on the lines $s=\pm n$. We compute it through level four and find one class at every point of each line $s=-3$ and $s=-4$. These classes give exact polynomial deformations of the residual family. At level six on $s=-6$, the cohomology instead vanishes generically and has dimension two at $(-1,-5)$ and $(-5,-1)$. We also compute the failure of the standard generator redefinition to preserve the auxiliary identities, and discuss the regularization of an explicit residual solution.
Read the full abstract

Quantum Physics

Scalable Unitary State-Encoding for Generic Large Data

Efficient encoding of data into quantum states remains a fundamental bottleneck in many quantum computing protocols, which would otherwise offer advantages over their classical counterparts. Here, we introduce deterministic and scalable amplitude-encoding methods with low two-qubit gate costs and tunable trade-offs between circuit depth and encoding fidelity. Our approaches require no amplitude amplification or mid-circuit measurement and are capable of encoding arbitrary complex data or queryable functions by using tensor-networks to go beyond what can be held in classical memory. Our principle contributions are two complementary algorithms for simplifying fully controlled multiplexer circuits to one- and two-qubit gates. In the first, we formally map the elimination of redundant gate controls to a Boolean logic minimization problem, while in the second we utilize a single ancilla qubit to map the multi-controlled rotation gates to a sequence of one- and two-qubit operations via the Walsh-Hadamard spectrum of the gate angles. We benchmark these methods across a broad range of function classes, characterizing their total depth in an all-to-all connectivity graph against system size and target fidelity. Hybridizing our methods gives up to orders of magnitude improvement in circuit depth at equivalent state fidelity compared to the default state-preparation method in the Qiskit package. We furthermore demonstrate our method on IQM hardware with a square lattice connectivity, finding excellent fidelity, even with minimal noise mitigation. By significantly reducing gate depth for encoding, this work enables the experimental execution of applications on current and near-term hardware which would otherwise be reduced to noise before the first calculation step, shortening the time horizon to robust quantum utility.
Read the full abstract

All authors Vittorio Pagni, Peter K. Schuhmacher, Michael Felderer, Michael Epping, Kevin Lively

Machine Learning

A Tight Hierarchy for Chain of Thought

Chain-of-thought (CoT) reasoning has emerged as a powerful paradigm for extending the computational capabilities of transformer-based models. A growing body of recent work has begun to uncover the fine-grained computational power of transformers augmented with CoT, as well as their connections to classical complexity theory. In this work, we advance this line of research by establishing the first hierarchy theorem for CoT-based computation. Our main result shows that for any reasonable resource bound $r(n) \geq n$ and any $\varepsilon > 0$, \[ \mathrm{CoT}(r(n)) \subsetneq \mathrm{CoT}(r(n)^{1+\varepsilon}), \] where $\mathrm{CoT}(r(n))$ denotes the class of decision problems solvable by transformers using $O(r(n))$ CoT reasoning steps on inputs of length $n$. Our result shows that increasing the number of CoT reasoning steps provably yields strictly greater computational power, establishing CoT steps as a fundamental resource in transformer-based computation. Unlike classical hierarchy theorems, which are typically proved via diagonalization, our approach leverages known coarse-grained simulations between CoT-augmented transformers and Turing machines in both directions. Building on these connections, we adapt the classical padding technique from complexity theory to this setting to obtain a tight hierarchy. A key technical contribution of our work is the implementation of padding within the transformer-based CoT model.
Read the full abstract

Machine Learning

Diffusion-free Score Matching for Exact Conditional Sampling

The success of modern score-based generative models is largely driven by diffusion-based training, which learns a collection of time-indexed score fields corresponding to progressively noisier versions of the data distribution. While such models achieve impressive sample quality, leveraging them for sound probabilistic inference—particularly sampling from arbitrary conditionals of the model—remains challenging. This difficulty arises because conditioning information is only available for clean data and not for the higher noise levels required to generate exact conditional samples. In this paper, we introduce DIffusion-free SCOre matching (DISCO), a principled approach that sidesteps the need for time-dependent score fields altogether. Our method is based on an objective that, while reminiscent of diffusion-based training, estimates only a single score of the (slightly perturbed) data distribution, for which the exact conditional score is readily obtained. In experiments, DISCO is competitive with state-of-the-art diffusion models in terms of sample quality. More importantly, DISCO yields a more faithful representation of the underlying data distribution and—crucially—enables accurate sampling from arbitrary conditionals, outperforming standard heuristic samplers. This capability opens the door to principled and flexible probabilistic reasoning with score-based models.
Read the full abstract

General Relativity and Quantum Cosmology

Generalized Schwarzian thermodynamics from gravitational perturbations of near-extremal Kerr

We derive the generalized Schwarzian thermodynamics of near-extremal Kerr directly from four-dimensional gravitational perturbations of the near-horizon geometry. We construct two families of exact linear perturbations of the near-NHEK background: a non-normalizable mode that deforms the asymptotic boundary, and a normalizable mode supported near the horizon. Both perturbations are smooth solutions of the vacuum Einstein equations reduced to the Jackiw–Teitelboim equation for the dual scalar, without conical singularities. We study their asymptotic symmetries and show that they generate an algebra of time reparametrizations; requiring the gauge parameters to depend on the Schwarzian mode, the isometry condition selects the stationary KdV hierarchy. The same hierarchy is independently enforced by the vacuum Einstein equations on the perturbation coefficient. In the Euclidean regime, we evaluate the Hamiltonian action and show that each mode independently generates a tower of odd-power corrections to the near-extremal entropy.
Read the full abstract

Computer Vision

A lightweight convolutional neural network for target detection in an edge environment

This paper proposes a novel, lightweight convolutional neural network (CNN) designed for edge device deployment. The CNN is developed to recognize and classify threats or targets of interest. The model is equipped with a real-time hotspot detection algorithm, which enables it to process information quickly and accurately, resulting in improved inference time of up to 20 fps and a detection range of more than 400 meters with typical thermal situational awareness sensors. The proposed CNN has been rigorously tested and validated to demonstrate its high accuracy, sensitivity, and low false positive rate, providing a reliable solution for edge devices to detect potential targets. Additionally, the lightweight design of the CNN allows for easy deployment, making it an ideal solution for extending the security bubble for edge devices. The proposed model can be used for a variety of applications, including surveillance, security, and object recognition.
Read the full abstract