Peer Revenue Returns subject to peer review

Papers

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

All of arXiv

Tuesday 10 November

Optics

Revisiting the dispersions and quality factors of nanoshell plasmons

The dispersions and quality factors of a metallic nanoshell's plasmons may be readily tuned by varying the thickness of the nanoshell. This is because 1. Thickness determines the hybridization between the two nanoshell boundaries and 2. Thickness affects the electronic surface scattering rate and, therefore, the non-radiative decay rates of the plasmonic resonances. We find that prior theoretical works, focused primarily on the thin-nanoshell and retarded regimes, obtained qualitatively incorrect results for both the plasmonic dispersions and the associated plasmonic quality factors. Some of these errors have been pointed out in subsequent works, though the corrected results have not been reported, to our knowledge. Beyond obtaining the correct results (correct dispersions and quality factors as a function of nanoshell thickness), we scrutinize the assumption (used by many prior works) that Matthiessen's rule may be used to combine the effects of bulk and surface-mediated plasmonic scattering processes. To do so, we extend prior work by Dingle to the nanoshell geometry. Although we find qualitative agreement with the Matthiessen's rule approach, we find that Dingle's theory systematically predicts higher quality factors across the range of nanoshell thicknesses we consider.
Read the full abstract

Numerical Analysis

Two nodally bound-preserving and mass-conservative discontinuous Galerkin methods for reaction-diffusion problems

We propose and analyze two high-order discontinuous Galerkin methods that preserve prescribed nodal bounds and total mass for reaction-diffusion problems with anisotropic and heterogeneous diffusion. Both methods combine a weighted interior-penalty prediction step with a convex correction in either the dG energy or the $L^2$ metric. We construct a pointwise bound-preserving and cellwise mass-preserving comparison function with optimal local approximation properties. This construction yields optimal dG-energy error estimates for the energy correction and, under a dual regularity assumption, optimal $L^2$ and dG-energy error estimates for the $L^2$ correction on quasi-uniform meshes. We further establish a general a posteriori estimate that combines a predictor estimator with the computable correction norm. The resulting estimators are globally reliable and efficient up to data oscillation, with parameter-dependent efficiency constants. Numerical experiments compare the two corrections and illustrate their use in adaptive computations.
Read the full abstract

Plasma Physics

Drive and damping govern resonant electron energy redistribution in a kinetic model informed by helium discharges

Energetic electrons exchange energy with electrostatic waves in low-temperature plasmas, but wave amplitude alone does not determine the direction or cumulative effect of that exchange. We analyse nonlinear electron redistribution using a single-mode kinetic model parameterized by published helium-discharge measurements. Frequency and pressure comparisons are combined with a controlled damping scan and explicit particle-current and energy diagnostics. Across the frequency cases, the inferred resonant drive varies by a factor of 2.7 while background damping varies by only about 9%. Reducing pressure from 33.3 to 10.66 Pa at 37 MHz increases the net drive by a factor of 2.24. In the damping scan, the peak wave energy and residual parallel-energy deficit both decrease by approximately 93%, consistent with quadratic field-amplitude scaling at the reported endpoints. An integrated energy identity connects retained wave energy, background loss and distribution recovery, and distinguishes cessation of wave growth from reversal of energy transfer. Quantitative timescale and directional-distribution tests define the connection between this reduced model and discharge kinetics. The analysis provides a common interpretation of resonant redistribution, coherent current and energy exchange, together with observables for testing that interpretation in kinetic simulations and phase-resolved measurements.
Read the full abstract

All authors Zhao-Ju Bo, Lei Chang, Shi-Jie Zhang, Ying-Xin Zhao, Zi-Chen Kan, Ji-Kai Sun, Ilya Zadiriev, Elena Kralkina, Shin-Jae You, Jing-Jing Ma, Géza Husi

Combinatorics

Maximal Triply Even Binary Codes from Elliptic Orthogonal Graphs

For a finite set $X$, a binary linear code $C\subseteq\mathbb{F}_2^X$ is called a triply even binary code if $\operatorname{wt}(c)\equiv 0\pmod{8}$ for every $c\in C$. Let $q$ be an odd prime power and $n$ be an integer with $n\ge 2$. Let $E^-_{2n+2}(q)$ denote the binary code spanned by all pairwise sums of rows of the adjacency matrix of the elliptic orthogonal graph arising from a $(2n+2)$-dimensional elliptic quadratic space over $\mathbb{F}_q$. We prove that $E^-_{2n+2}(q)$ is a triply even binary code if and only if $n=2$ and $q\equiv 3\pmod{4}$. In this case, we also prove that $E^-_6(q)$ is a maximal triply even binary code.
Read the full abstract

Data Structures and Algorithms

Approximating graphic TSP by b-matchings

We give a polynomial-time algorithm for graphic TSP that returns a tour of length at most $\frac{89}{64}=\frac{7}{5}-\frac{3}{320}$ times the value of the subtour relaxation. This improves the long-standing ratio $\frac75$ of Sebő and Vygen and shows that the integrality gap of the subtour LP on graphic instances is at most $\frac{89}{64}\approx 1.3906$. We optimize a non-linear objective function over an odd-join polytope, which uses (i) a new removable system of edges, and (ii) reconnection of subtours if the graph is disconnected. This approach gives a $\frac{13}{10}$-approximation algorithm for $4$-regular $4$-vertex-connected graphs, improving the $\frac43$ of Newman, and a $\frac43$-approximation for $k$-vertex-connected graphs of maximum degree at most $k+1$, for every $k\ge 3$. For the general graphic TSP, we construct a feasible point in the odd-join polytope, using an optimal vertex from the Subtour linear programming relaxation of TSP, which helps us bound the approximation factor by 89/64.
Read the full abstract

Optimization and Control

Exact (Non-)Accelerations of Heavy-Ball Discretizations

Momentum methods with the same slow dynamics can have different worst-case rates on smooth strongly convex functions. We identify a half-step phase barrier in the gradient query that governs this difference. For momentum with extrapolation $\tau_\kappa\to\tau$, $\beta=1-\theta/\sqrt\kappa+o(\kappa^{-1/2})$ and $hL\to y$, where $\kappa=L/\mu$ and $\theta,y>0$, the exact uniform distance factor is $1-c/\sqrt\kappa+o(\kappa^{-1/2})$, with $c=\min\{\theta/3,c_Q(\theta,y)\}$, when $\tau>1/2$ and $y(1+2\tau)<4$; $c_Q$ is the slowest quadratic decay exponent. Quadratically accelerated families with $\limsup\tau_\kappa<1/2$ have divergent trajectories on the full function class. Upper bounds use finite-lag certificates: nonnegative combinations of inequalities for values and gradients at finitely separated query points, together with comparisons to the minimizer. For any prescribed accuracy, spiral mixtures or quadratic trajectories of one finite-dimensional objective give matching lower bounds at all iteration horizons. We resolve the parameter layer near $\tau=1/2$ and classify all tunings of this family attaining the optimal leading constant 1. The flow also bounds the leading constant of every second-order stationary linear family with the same slow parameters $(\theta,y)$ and fixed finite memory.
Read the full abstract

Physics Education

Active learning at a graduate summer school: feasibility, student response, and the role of instructor training

Active learning is becoming standard practice in undergraduate STEM instruction in North America and is slowly being adopted in graduate courses. Graduate summer schools, one- to two-week programs in which students from many institutions gather for intensive, specialized instruction in their field, are common in physics; they are taught by researchers chosen for their expertise, and their pedagogy remains almost entirely traditional lecturing. We report a study at a one-week graduate school in neutrino and astroparticle physics attended by 38 students from 8 countries. Six lecturers with limited or no prior experience of teaching with active learning introduced Peer Instruction-style active learning into their lectures after approximately four hours of preparation led by a physics-education specialist. Each of the three main lecture streams was taught traditionally on one day and in the active format on the other. Students were surveyed after each of the first two days and at the end of the week; all six instructors were surveyed as well. In every active lecture, spanning three instructors and three topics, 82-100% of respondents reported that the interactive segments helped them stay engaged and check their understanding, and in every stream the difficulty was rated as better matched on the active day. At the end of the week, 33 of 36 wanted such teaching in future graduate courses, and not one respondent disagreed. All six instructors judged the approach productive and intend to use it again; only two found implementation straightforward. With brief expert-led preparation, lecturers new to active learning implemented it successfully for a heterogeneous audience, and both students and instructors want more. The school was also the first time any of the six instructors implemented structured active learning: a summer school can develop its lecturers as well as its students.
Read the full abstract

All authors Nicholas Kamp, Louis Deslauriers, Carlos A. Argüelles, Gonzalo Herrera, William G. Thompson

Group Theory

On Distinct Powers in Finite Groups

Sun conjectured that if a group $G$ has no elements of orders $2,3,...,n+1$, then every $n$-element subset $A$ can be ordered as $a_1,...,a_n$ so that $a_1,a_2^2,...,a_n^n$ are pairwise distinct. We prove the conjecture for finite groups when $n=4,5,6$. The proofs combine collision-counting arguments with reductions to finite-field assignment problems; the remaining finite cases for $n=6$ are verified exhaustively. We also prove the conjecture for every $n$ when no four elements of $A$ generate the same cyclic subgroup.
Read the full abstract

Computer Vision

Lightweight Deep Learning Framework for Astrocyte-Centric 3D Reconstruction across Normal and Stress-associated Ultrastructural States

The morphology of astrocytes is closely associated with their ability to support and modulate neural circuit functions; however, its reconstruction from volume electron microscopy (vEM) data remains labor-intensive and computationally resource-intensive, placing substantial barriers on its broader adoption and scalability despite the high spatial resolution of vEM. We present AstroSeg-vEM, a lightweight, astrocyte-centric framework for morphology-preserving astrocyte segmentation, 3D reconstruction, and quantitative morphological analysis in vEM images. AstroSeg-vEM combines transformer-based 2D segmentation with dedicated post-processing methods and operates directly on high-resolution EM sections, making it well suited to anisotropic vEM data and downstream proofreading workflows while preserving the full in-plane resolution of the original EM data. Across four mouse hippocampal datasets representing normal and morphologically defined stress-associated astrocyte ultrastructural states, AstroSeg-vEM demonstrates robust performance and generalization, achieving average Dice and weighted Dice values of 0.92 and 0.95, respectively, on a held-out test dataset. Its lightweight architecture enables efficient, high-fidelity reconstruction on consumer-grade hardware, providing a scalable and accessible platform for large-scale astrocyte ultrastructure analysis and future astrocyte-neuron interaction studies.
Read the full abstract

All authors Bowen Deng, Gwang-Jin Jeong, Seulgi Noh, Ji Young Mun, Na Young Kim

Combinatorics

The minimum number of cycles of length $0$ modulo $k$ in graphs with no proper $k$-coloring

Consider an integer $k\geq 3$. Toft and Tuza observed that every graph with no proper $k$-coloring has a cycle of length $0$ modulo $k$. Brewster, McGuinness, Moore, and Noel conjectured the following quantitative strengthening: every graph with no proper $k$-coloring has at least $\frac{(k+1)!}{2k}$ cycles whose lengths are divisible by $k$. The conjecture is sharp, as demonstrated by the complete graph on $k+1$ vertices. Kim and Picollelli confirmed the conjecture for $k=3$. We establish the conjecture for every $k\geq 3$. In other words, we determine the minimum number of cycles of length $0$ modulo $k$ in a graph with no proper $k$-coloring.
Read the full abstract

Materials Science

Hidden interfacial magnetism in RuO$_2$ heterostructures revealed by a superconducting spin switch

RuO$_2$, once the flagship altermagnetic candidate, is now believed to be nonmagnetic in the bulk, yet weak, static and spatially inhomogeneous magnetism emerges at its surfaces and interfaces. Where this magnetism resides, what stabilises it, and whether it is connected to these states remain open questions. Here we address these questions at the Nb/RuO$_2$ interface using a superconducting spin switch. A GdN/Nb/RuO$_2$ trilayer allows interfacial exchange-field alignment to tune the superconducting transition temperature of Nb, probing interfacial magnetism inaccessible to conventional magnetometry. The switch reveals a hidden magnetic subsystem, which we attribute to a thin RuO$_x$ redox layer formed as Nb scavenges oxygen from RuO$_2$. This layer behaves as a weak and magnetically hard ferromagnet with vacancy-induced anisotropy. Furthermore, supercurrent amplifies the switching efficiency, making the probe sensitivity tunable in situ. Superconducting spin switches thus emerge as a quantitative low-temperature probe of hidden magnetism at buried interfaces.
Read the full abstract

All authors Tetsuma Mandokoro, Hisakazu Matsuki, Shutaro Karube, Rodrigo de las Heras, Andrei Mazanik, Akihiro Yamada, Fugo Tokoro, Susumu Hirata, Ryusuke Hisatomi, Yoichi Shiota, Vitaly N. Golovach, F. Sebastian Bergeret, Teruo Ono

Cryptography and Security

POEM: Output-Relative Block-Sparse Attention for Long-Context Prefilling

Large language models (LLMs) have shown strong capabilities across diverse applications, making both safety and utility essential for reliable deployment. Jailbreak attacks use diverse forms of adversarial prompting to bypass alignment safeguards and induce LLMs to generate harmful responses. Some representation-based defenses regulate internal activations using predefined steering directions or learned safety regions. However, adapting intervention to each input without a predefined representational target while limiting over-refusal remains a challenge. In this paper, we propose Target-free Entropy-guided Residual Modulation (TERM), an inference-time defense that uses prompt-calibrated attention entropy deviations to guide adaptive modulation of the model's own residual updates. Attention patterns vary across prompts and decoding steps, making raw entropy values difficult to interpret as a consistent signal for intervention. We therefore calibrate attention entropy against each prompt's own prefill baseline to establish an input-specific reference for reducing unnecessary intervention on benign requests. Adaptive scale normalization and temporal accumulation then account for differences in deviation scale and persistence to determine intervention strength. The resulting signal controls bounded scaling of the model's own residual updates, without relying on a predefined safety direction. We compare our method with seven defenses on seven jailbreak attacks and two over-refusal benchmarks across Llama, Qwen, and Gemma models, including Qwen3 models from 0.6B to 32B. Results show our method improves jailbreak robustness while maintaining low over-refusal on benign safety-sensitive requests with modest inference overhead.
Read the full abstract

All authors Haoyi Wu, Yunkai Dang, Wanyu Zhang, Lei Wang, Wenbin Li

Cryptography and Security

TERM: Target-Free Jailbreak Defense via Entropy-Guided Residual Modulation

Large language models (LLMs) have shown strong capabilities across diverse applications, making both safety and utility essential for reliable deployment. Jailbreak attacks use diverse forms of adversarial prompting to bypass alignment safeguards and induce LLMs to generate harmful responses. Some representation-based defenses regulate internal activations using predefined steering directions or learned safety regions. However, adapting intervention to each input without a predefined representational target while limiting over-refusal remains a challenge. In this paper, we propose Target-free Entropy-guided Residual Modulation (TERM), an inference-time defense that uses prompt-calibrated attention entropy deviations to guide adaptive modulation of the model's own residual updates. Attention patterns vary across prompts and decoding steps, making raw entropy values difficult to interpret as a consistent signal for intervention. We therefore calibrate attention entropy against each prompt's own prefill baseline to establish an input-specific reference for reducing unnecessary intervention on benign requests. Adaptive scale normalization and temporal accumulation then account for differences in deviation scale and persistence to determine intervention strength. The resulting signal controls bounded scaling of the model's own residual updates, without relying on a predefined safety direction. We compare our method with seven defenses on seven jailbreak attacks and two over-refusal benchmarks across Llama, Qwen, and Gemma models, including Qwen3 models from 0.6B to 32B. Results show our method improves jailbreak robustness while maintaining low over-refusal on benign safety-sensitive requests with modest inference overhead.
Read the full abstract

All authors Yunkai Dang, Wanyu Zhang, Meiyi Zhu, Qi Fan, Tianyu Ding, Wenbin Li

Computer Vision

PhyRubric: Rubric-Grounded Diagnostic Evaluation of Physical Commonsense in Text-to-Video Generation

As text-to-video (T2V) models are increasingly positioned as world simulators, it is essential to evaluate whether their videos conform to physical commonsense. However, existing benchmarks rely on annotators who rate each video with a single score based on intuition, yielding ratings that are subjective and do not tell which physical principle is violated. We introduce PhyRubric, a rubric-grounded diagnostic framework for evaluating physical commonsense in T2V generation. For each caption, PhyRubric constructs a caption-specific rubric before any video is seen, decomposing the caption into semantic and physical items, and derives scores from item-level verdicts grounded in timestamped evidence. We instantiate this framework as PhyRubric-Bench, which spans 34 categories across five physical domains, and scale it up to PhyRubric-Train with 3,788 videos, on which we fine-tune the open-source PhyRubric-Judge. Empirical results across ten T2V models show that physical commonsense remains far from solved, and that PhyRubric-Judge correlates more closely with human judgments than existing open-source judges while matching the frontier proprietary Gemini 4 Argon. Further analysis provides actionable insights for T2V models.
Read the full abstract

All authors Huiyi Chen, Weitai Kang, Shicheng Fan, Xinshuo Wang, Zhixiang Shen, Yan Yan, Philip S. Yu, Baoxin Li

Data Structures and Algorithms

Greedy+Singleton for knapsack-constrained monotone submodular maximization has approximation ratio exactly $\beta^\star \approx 0.42944$

Greedy+Singleton is the textbook heuristic for monotone submodular maximization under a knapsack constraint: it returns the better of the cost-benefit greedy set and the best single item. Its worst-case approximation ratio, known only to lie in $[0.427, 0.42945]$ (Feldman-Nutov-Shoham; Kulik-Schwartz-Shachnai), is exactly an explicit constant $\beta^\star\approx 0.42944$. In the extremal instances, many cheap "filler" items wear down the value of two large optimal items while greedy spends the budget, until neither fits. This picture gives a sharper factor-revealing relaxation; we bound it analytically near the extremal parameters, up to finitely many interval-arithmetic checks, and certify it elsewhere by an exact branch-and-bound. The matching bad instances are tie-free and checkable by hand. For the version that stops at the first item that does not fit, Wolsey's bound $1-e^{-\beta_W}\approx 0.35780$, where $e^{\beta_W}=2-\beta_W$, is tight; Wolsey stated this for his variant without giving the instances, and we give an explicit family with no ties that also covers comparison with the best feasible singleton. The CELF rule also has ratio exactly $\beta^\star$.
Read the full abstract

Signal Processing

Joint Air Corridor and BS Coverage Planning for Large-Scale High-Density Low-Altitude Aerial Networks via Channel Knowledge Maps

This paper studies the joint air corridor and base station (BS) coverage planning problem for high-density low-altitude aerial networks supported by terrestrial cellular infrastructures. In the considered system, unmanned aerial vehicles (UAVs) operate within a structured air corridor planned over a grid-based airspace, and each BS serves selected corridor grid cells by sharing its communication resources among all UAVs therein. Each BS's traffic load depends on both the number of covered corridor cells and the UAV density, thereby affecting the communication rate available to each UAV. We exploit the channel knowledge map (CKM) to characterize the channel gains between BSs and candidate airspace cells and evaluate desired signals and inter-cell interference for environment-aware offline planning. Based on CKM, we jointly optimize the corridor layout and BS coverage to minimize a weighted objective balancing corridor length and the supportable average UAV density along the entire corridor, subject to minimum communication rate requirements for all UAVs along the air corridor. The joint design is formulated as a mixed-integer linear program (MILP), which can be optimally solved using off-the-shelf solvers. For large maps, we develop a scalable suboptimal algorithm based on Lagrangian relaxation and A* search. Specifically, we incorporate BS resource constraints into A* corridor search via Lagrangian relaxation, and then jointly optimize BS coverage and UAV density along the generated corridors under the original rate constraints. Numerical results demonstrate the benefits of the proposed joint design in supporting high-density UAV operations and show that the proposed algorithm obtains high-quality solutions with substantially reduced computation time.
Read the full abstract

All authors Jiaxuan Li, Yilong Chen, Yuan Guo, Fan Liu, Jie Xu

Information Theory

Third-Order Asymptotics for Fixed-Length Lossy Compression of Discrete Memoryless Sources

This paper determines the third-order asymptotics of the logarithm of the minimum codebook size for fixed-length lossy compression of discrete memoryless sources. For an excess-distortion probability constraint $\varepsilon\in(0,1)$ and every interior distortion level, we establish a third-order correction of $\frac12\log n+O(1)$, where $n$ is the blocklength. This sharpens the general bounds of Kostina and Verdú [IEEE Trans. Inf. Theory, 2012]. The main technical innovation lies in the converse for positive dispersion. Under a reverse change of measure, we retain the exponential weights associated with tilted information and distortion slack and control them jointly using the multivariate weighted-strip estimate of Scarlett, Martinez, and Guillén i Fàbregas [Proc. IEEE ISIT, 2015]. A pairing argument verifies the required covariance nondegeneracy uniformly over all reconstruction words whose distortion balls meet a typical set, yielding the matching converse.
Read the full abstract

Machine Learning

Market-Conditioned JEPA: Learning from Unpredictable Market Shocks

Much of the variation in financial time series is driven by market-wide shocks that cannot be predicted from an asset's own history. Per-asset self-supervised learning treats this variation as irreducible noise. Under a latent factor model, we show that conditioning prediction on the realized market factor makes the asset's market exposure necessary for optimal prediction, while reducing the noise against which the asset-specific component is learned. Building on this insight, we introduce Market-Conditioned JEPA (MC-JEPA), a joint-embedding predictive architecture for financial time series. During training, we construct a leave-one-out market condition for each asset from the future target embeddings of other assets and incorporate it through factor loadings derived from the asset's own representation. On 11 years of minute-level Chinese A-share data, MC-JEPA improves over per-asset JEPA across nine targets spanning returns, market exposure, risk, and liquidity, with the strongest and most statistically significant gains in exposure, risk, and liquidity. It achieves the best RankIC on all liquidity targets and on return horizons from two hours to five days, outperforming time-series foundation models and self-supervised baselines, and delivers the best portfolio performance in a historical backtest. Ablations show that the return and exposure gains require the condition to be aligned with the realized market movement. After restoring the volatility scale removed by normalization, the predictor's factor loadings rank stocks by market beta nearly as well as a supervised probe, without labels. A synthetic factor-model study with known parameters shows that conditioning improves the recovery of both market exposure and the asset-specific component.
Read the full abstract

All authors Yuqi Wang, Fengyuan Liu, Mukai Li, Haochen Luo, Lei Li, Qi Liu

Data Structures and Algorithms

Improved Bounds for Online Matching with Random Edge Arrivals

We study the unweighted online matching problem with random edge arrivals, where each edge arrives at an independent time drawn uniformly from $[0,1]$, and the algorithm must immediately and irrevocably accept or reject it while maintaining a matching. Guruganesh and Singla (IPCO, 2017) first broke the $1/2$ barrier of the naive Greedy algorithm by giving a $(1/2+\varepsilon)$-competitive algorithm for a small constant $\varepsilon>0$, a guarantee that has remained the best known for nearly a decade. We give a simple three-phase algorithm that achieves a competitive ratio of $0.5445$ by deliberately rejecting selected early edges to reserve their endpoints for later matching opportunities. With a different choice of parameters, the same algorithm is $0.5535$-competitive on triangle-free graphs, including bipartite graphs. We also establish competitive ratio upper bounds below $0.6964$ for general graphs and $0.7043$ for bipartite graphs via exact rational evaluation of an optimal-policy recurrence on fixed graphs, even when the input graph is known up to isomorphism.
Read the full abstract

Optics

Development of a versatile modular system to study the ablation and plume expansion in a thin film

Time-resolved investigation of laser ablation of thin films requires stable, repeatable, and efficient experimental environment. Conventional approaches often rely on multiple laser-ablation shots acquired at different delay times to reconstruct the temporal evolution of the process, thereby assuming shot-to-shot reproducibility. This work presents a novel modular experimental setup that addresses these challenges and enables a systematic pump–probe study of laser–matter interaction. The setup provides reliable single-pulse measurements with controlled sample positioning inside the high vacuum chamber, as well as integrated data recording, which improves redundancy and ease with which experiments can be performed. Further, the setup serves as an advanced diagnostic tool for investigating the time evolution of laser–matter interactions, ablation dynamics and laser-produced plasmas in a single laser shot, in various configurations with a temporal resolution better than 15 ns. Using this capability, we show that for a fixed ablation geometry, the transient reflectivity and transmission response depends strongly on the probing side, indicating that probing direction is an independent diagnostic variable whose omission can lead to incomplete or misleading picture of the ablation dynamics. This experimental technique can advance the understanding of surface modification and plume expansion, which are relevant to fundamental studies and also for range of applications, including laser cleaning of tokamak viewports and mirrors, pulsed laser deposition (PLD), surface structuring and laser-induced forward transfer (LIFT), among others. The present setup is versatile enough to perform these studies under different ablation configurations without implementing major changes in the experimental configuration.
Read the full abstract

All authors Renjith Kumar R, Pabitra Kumar Mishra, B R Geethika, Hem C Joshi, Jinto Thomas

General Relativity and Quantum Cosmology

Quasinormal overtones and grey-body factors of Frolov black holes

We investigate scalar quasinormal modes (QNMs), with particular attention to overtones, and grey-body factors of Frolov black holes, which interpolate between the Reissner–Nordström (RN) and Hayward limits. Over the allowed ranges of charge $q$ and regularization parameter $\alpha_0$, all QNM frequencies have negative imaginary parts, confirming stability against scalar perturbations. Increasing either $q$ or $\alpha_0$ can trigger overtone outbursts, with the Frolov threshold shifted to smaller parameter values than in the RN or Hayward cases. Angular momentum does not eliminate the outburst but postpones its onset from $n=1$ at $l=0$ to $n\geq 2$ at $l=1$. Comparisons of the effective potential show that, for $l=1$, charge raises the barrier peak much more strongly than $\alpha_0$, while the clearest deviations from the Hayward and RN references are localized near the horizon. Using the QNM–grey-body correspondence at $l=1$, we find that overtone corrections are essential, and the amended formula substantially outperforms the eikonal truncation. At fixed frequency, both $q$ and $\alpha_0$ suppress transmission, with charge being the dominant factor, consistent with the stronger barrier enhancement under the charge scan.
Read the full abstract

All authors Ying Huang, Huajie Gong, Li-Gang Zhu, Jian-Pin Wu

Data Structures and Algorithms

Pair-Merge Tokenisation Is Hard to Approximate on Binary Strings and Strongly NP-Hard on Unary Datasets

Tokenisers such as Byte-Pair Encoding (BPE) shorten text by repeatedly merging an adjacent pair of symbols into a new one; fewer tokens mean less computation. We ask whether an efficient algorithm can choose $k$ merges that leave a string of length $n$ almost as short as possible; doing nothing is already within a factor $n$. We prove that, unless P = NP, no polynomial-time algorithm comes within a factor $c\,n/\log n$, for some constant $c>0$, already for one binary string and for BPE-style merges that replace every occurrence. The key idea is that $k$ merges form a grammar with $k$ rules: a vertex-cover gap becomes a grammar-size gap, which we amplify with relabelled copies, encode in binary and realise with explicit merges. For merges that may skip occurrences, a simple algorithm achieves $O(n/\log\log n)$, and no Karp reduction can prove hardness beyond order $n/\log n$ unless coNP $\subseteq$ NP/poly. Over a one-letter alphabet, both tokenisation problems left open by Kastreva et al. are strongly NP-complete on datasets with repeated strings.
Read the full abstract

Cosmology and Nongalactic Astrophysics

The unified bound dark energy–bound dark matter model confronted with DESI DR2, ACT and SPT-3G, and the Unite supernovae measurements

The unified bound dark energy-bound dark matter (BDE-BDM) model derives the whole dark sector from a single supersymmetric SU(3) dark gauge group, whose lightest meson is the dark energy and whose heavier bound states, created with a velocity $v_c$ when the group condenses, are the dark matter. Because the condensation scale $\Lambda_c$ is fixed by gauge-coupling unification, the dark-energy sector has no free parameter beyond those of $\Lambda$CDM, and the dark-matter sector adds only $v_c$. We confront the model with DESI DR2 baryon acoustic oscillations (BAO), the Planck, ACT DR6 and SPT-3G D1 CMB spectra, the combined ACT and SPT CMB lensing and the Unite supernovae. The measured condensation scale, $\Lambda_c=43.74\pm0.11$ eV, lies inside the window $34^{+16}_{-11}$ eV predicted by unification, and the dark energy thaws without ever crossing the phantom divide, reaching $w_0=-0.9267\pm0.0003$ today; it thus reproduces the DESI distances that $w_0w_a$CDM fits with a phantom past. BDE-BDM gives the best fit to the DESI DR2 BAO of the four models, reducing $\chi^2_{\rm BAO}$ by 20% with respect to $\Lambda$CDM, more than $w_0w_a$CDM, and over the full data combination it is statistically equivalent to both $\Lambda$CDM and $w_0w_a$CDM, with $\ln B=-0.50$ and $-0.15$, while having two fewer dark-energy parameters than $w_0w_a$CDM. Within bound dark energy, bound dark matter is moderately preferred over cold dark matter, $\ln B=+3.2$ with $\Delta\chi^2=-9.0$, mostly from the CMB. The data now constrain the bound velocity from both sides, $v_c=0.37\pm0.14$, which places the free-streaming cutoff of the matter power spectrum at $k_{1/2}=1.49\,h$ Mpc$^{-1}$ and lowers the amplitude of structure to $S_8=0.807\pm0.007$. Future high-redshift and Lyman-$\alpha$ forest data, together with growth and lensing measurements, will therefore provide the decisive tests of the unified dark sector.
Read the full abstract

Machine Learning

Metamorphic testing of learned AC optimal power flow surrogates reveals representation dependence

Evaluating a learned AC optimal power flow (AC-OPF) surrogate on one grid serialization cannot show whether its accuracy depends on that encoding. We test this dependence with thirteen metamorphic relations, each specifying when a transformation is valid and how to compare its outputs. Our audit covers 17 model and pipeline configurations: two released trained checkpoints and 15 random-weight architecture or mechanism controls. Every configuration violates at least six relations, even at reporting tolerances ten times tighter or looser. For GridSFM-Open v1.1, source inspection and intermediate-tensor interventions point to an indexing bug and three noncanonical or numerically unstable preprocessing choices; a repaired checkpoint remains untested. In a separate case500 training experiment, augmentation lowers transformed-view error in three of four models and median discrepancy and constraint-violation P95 in all four. These benefits depend on the training protocol: some relations worsen, and every evaluated prediction remains AC infeasible. A GPU/CPU eigensolver mismatch accounts for most of the large raw-Laplacian-baseline gain. We recommend reporting representation consistency alongside accuracy, feasibility, and cost. These results establish neither deployment readiness nor universal invariance.
Read the full abstract

Machine Learning

Investigating Composite Modeling in Multi-Objective Bayesian Optimization for Scientific Design

Scientific design often requires balancing competing objectives that are expensive to evaluate through experiments or simulations. Multi-objective Bayesian optimization (MOBO) addresses this challenge by using probabilistic models to guide the selection of promising designs. In many applications, tunable inputs map to observable intermediate responses, which are then combined through known functions to compute the final objectives. Prior work on Bayesian optimization of composite functions shows that learning from these intermediate responses, which we call Composite modeling, can substantially improve sampling efficiency in single-objective optimization, compared to methods that condition only on black-box observations of the objective function. The benefits of Composite modeling in MOBO, however, remain understudied. In this work, we study the benefits of Composite modeling in MOBO and investigate the factors driving variation in its gains. We first establish single-objective conditions under which intermediate observations provide no additional predictive information beyond objective values, where composite and black box methods use the same warped Gaussian process prior. Controlled numerical experiments with versions of four MOBO methods across eight synthetic and application-inspired scientific-design benchmarks examine how these conditions relate to optimization gains in the multi-objective setting, where the black box approach models the objective functions with a standard multi-output Gaussian process. Both investigations support the claim that composite modeling tends to help more when the function mapping intermediate responses to final objectives is non-bijective and non-affine.
Read the full abstract

All authors Ricky Ye, Sahas Goli, Muhammad Ali, Poompol Buathong, Peter I. Frazier

Computation and Language

A Concept-Based Interpretable Framework for Clinical Text-based Prediction via Token-Concept Attention without Concept Supervision

Objective: Concept-based interpretation has emerged as a promising approach for explaining blackbox models using human-understandable, high-level concepts. However, most existing methods require annotated concept labels to train the model to identify concepts in the input, and obtaining such annotations in clinical text is often expensive. This study aims to enable concept-based interpretation for language models without concept supervision on clinical prediction tasks. Methods: We propose a novel framework that enables concept-based interpretation without concept supervision through a tailored cross-attention bottleneck. Clinical text and the text descriptions of medical concepts are first tokenized and embedded by a pretrained language model. A cross-attention module then estimates the presence of each concept for each text token, producing concept relevance scores that capture how concepts are involved throughout the note. These scores are pooled and used in a linear prediction layer to model clinical outcomes in an interpretable manner. Results: The performance of the proposed framework was compared against blackbox models across two tasks: chest trauma severity stratification using a private dataset and an in-hospital mortality prediction task using the publicly available MIMIC-III dataset. The proposed framework achieved AUROC values within 0.02 of the corresponding blackbox models. The framework also enabled cohort- and patient-level interpretation, capturing the overall concept importance across the cohort and patient-specific concept contributions. Conclusion: Our proposed framework realizes concept-based interpretation without requiring ground-truth concept presence labels. It improves the transparency of clinical text predictions while maintaining strong performance in clinical prediction tasks.
Read the full abstract

All authors Jifan Gao, Junjie Hu, John Caskey, Madeline Oguss, Simona Cristea, Anoop Mayampurath, Matthew Churpek, Majid Afshar, Guanhua Chen

Artificial Intelligence

Benchmarking Vendor-Guided Optimizations on Edge NPUs: From NPU Only to End-to-End Performance. Extended technical report

Edge NPU performance depends not only on hardware capabilities, but also on model transformation and the overall deployment pipeline. We propose a benchmarking methodology that separates these effects by comparing three configurations: Baseline, Vendor Guided NPU Only (VG-NPU), and Vendor Guided End-to-End (VG-E2E), where VG-NPU preserves common model I/O and host-side processing while VG-E2E follows the vendor-native deployment pipeline. We apply this methodology to four edge NPUs and three vision workloads. VG-NPU improves both accuracy and Runtime-call latency in 10 of 12 NPU–workload combinations, whereas VG-E2E reduces System-level latency in all combinations but improves accuracy in only four. Moreover, NPU-level performance rankings do not always match End-to-End rankings, and VG-E2E provides an additional 10.4–70.7% reduction in System-level latency over VG-NPU. These results demonstrate the importance of evaluating vendor-guided optimization separately at the controlled model-transformation and full deployment levels when comparing heterogeneous commercial Edge NPU platforms in practice.
Read the full abstract

All authors Youngjin Jo, Jihwan Kim, Minseok Choi, Kwanghoon Choi, Jusung Kang, Jaein Kim

Computation and Language

An evidence-traceable benchmark of clinical reasoning in large language models

Benchmark scores for medical large language models (LLMs) are difficult to interpret when scoring criteria cannot be traced to clinical evidence. Here we developed Med Evidence Bench (MEB; https://www.medevidencebench.org.cn/), comprising 1,536 physician-constructed items across three levels: guideline knowledge, standardized clinical scenarios and real-world laboratory scenarios. Each criterion stating a medical claim is linked to a versioned source-text span. Across ten LLMs, the Evidence Trace Score (ETS) was 0.975-0.997 for guideline knowledge, 0.292-0.658 for standardized scenarios and 0.182-0.447 for real-world scenarios, and rankings varied across levels. For real-world scenarios, scores for identifying clinical information (mean, 0.770) exceeded those for matching evidence (0.165) or formulating recommendations (0.160). On 1,296 criterion-level judgments, automated grading agreed with the majority vote of three physicians (macro-F1, 0.930; 95% confidence interval, 0.915-0.945), although agreement was lower for evidence application. These findings from Chinese-language materials indicate that guideline recall does not ensure patient-specific evidence use and do not establish clinical effectiveness.
Read the full abstract

All authors Zhiqun Zhao, Xiaodong Yang, Bruce Gu, Ming Li, Yi Fu, Song Li, Qi Mei, Zhi Liu, Kai Zhang, Kexue Fu, Jing Wu, Xiaofei Lu, Yule Li, Zheng Wang, Haiquan Lu, Xiaoming Wu

Machine Learning

Discovering Structurally Plausible and Interpretable Cognitive Models with Large Language Models

Computational cognitive modeling has traditionally relied on manual model construction, which is a time-intensive process that requires substantial domain expertise. Although recent LLM-based approaches have begun to automate model generation, they often prioritize predictive accuracy at the expense of structural plausibility and parameter identifiability. We introduce DisCo, an LLM-guided framework that discovers interpretable symbolic cognitive models through evolutionary search with two structural components: a Validator that filters structurally invalid models, and a Cleaner that promotes parameter identifiability by removing redundant parameters. Across three behavioral tasks, DisCo consistently discovers models that achieve performance competitive with task-specific baselines. Ablation analyses show that the Validator is important for recovering psychologically meaningful structure. Without this component, models tend to violate behavioral constraints and show less consistent associations with external cognitive measures. These results suggest that LLM-driven model discovery, when coupled with appropriate structural constraints, can yield cognitively plausible models that go beyond predictive fit.
Read the full abstract

Computation and Language

Are Dense Rewards Worth Their Tokens in GRPO?

RLVR trains language models to reason by rewarding correct final answers. A widely used algorithm, group-relative policy optimization (GRPO), gives every token of a solution the same credit, so dense process rewards have been proposed to credit intermediate steps, for example with a Monte-Carlo reward that scores each step by sampling continuations from it. Comparing such a reward with GRPO per training update, or vetting it with a gradient test on a single batch, leaves out what it costs: in our setting, the continuations alone take 91% of all generated tokens. We introduce a matched-compute protocol that gives every training intervention the same budget, measured in generated tokens and training GPU-hours, and tests it in preregistered studies with controls for side effects. Across 1,668 runs on six models, matching compute reverses the ranking. On Qwen2.5-1.5B, the Monte-Carlo reward leads standard GRPO per update, yet ends 14.3 percentage points lower at equal tokens and 7.8 points lower at equal training time, and none of four dense-reward variants outperforms GRPO even with as much training time. A per-token gradient test orders these rewards as training does but overestimates each by one to two points, as much as the gains it is meant to detect. In contrast, training on the prompts whose predicted success is nearest 1/2 improves accuracy at matched compute, by 1.7 points on Qwen3-0.6B and 0.8 on Qwen2.5-1.5B. These gains replicate on fresh seeds and unseen problems, and a simple rule, that the candidate prompts are predicted to fail more often than not, predicted where they appear in five of six preregistered tests. Both gains come early in training, and standard GRPO with twice the initial learning rate comes within half a point of them. We argue that training interventions should be validated at matched compute, which overturns some apparent gains and confirms others.
Read the full abstract

All authors Vikash Singh, Weicong Chen, Debargha Ganguly, Yanyan Zhang, Mohsen Hariri, Sreehari Sankar, Wang Yang, Chaoda Song, Alexander Nemecek, Osama Zafar, Yuting Shao, Kai Ye, Biyao Zhang, Xinpeng Li, Shouren Wang, Erman Ayday, Jing Ma, Sanmukh Rao Kuppannagari, Vipin Chaudhary