Peer Revenue Equity in science

Papers

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

All of arXiv

Saturday 29 August

Software Engineering

Structure-Behavior Coalescence and the Limits of Traditional Systems Theory

This paper examines a foundational assumption in traditional systems theory, namely that structure (the organization of components) and behavior (the evolution of system activity over time) can be treated as separable analytical dimensions. It argues that this separation contributes to persistent difficulties in explaining system identity, particularly in cases involving change, emergence, and boundary specification. To address this issue, the paper introduces Structure-Behavior Coalescence (SBC) as a reframing principle. SBC proposes that structure and behavior should not be understood as independently existing entities that are subsequently related through modeling constructs, but as mutually constitutive aspects of a single systemic process. From this perspective, system identity is understood as arising from the sustained co-determination of structural organization and behavioral dynamics, rather than from their external correspondence or alignment. This reframing provides a unified way of understanding system identity, emergence, and boundary formation within a cybernetically informed systems perspective.
Read the full abstract

General Physics

Cosmological Perturbations and Observational Constraints on Spinor Field Quintessence Dark Energy

We investigate the linear perturbation cosmology of a spinor-field realization of quintessence dark energy by implementing the model in the Einstein–Boltzmann solver CLASS. The model parameters are constrained using a Markov Chain Monte Carlo analysis with Pantheon$^{+}$ Type Ia supernovae, cosmic chronometers, DESI DR2 baryon acoustic oscillations, redshift-space distortions, and Planck 2018 CMB distance-prior data. For the full combined dataset, we obtain $w_{\rm de}=-0.9710^{+0.0206}_{-0.0206}$, $\Omega_{m0}=0.2939^{+0.0047}_{-0.0047}$, and $H_0=66.23^{+0.55}_{-0.53},\mathrm{kms^{-1}Mpc^{-1}}$. The corresponding perturbation quantities are $\sigma_8=0.7537^{+0.0076}_{-0.0075}$ and $S_8=0.7459^{+0.0067}_{-0.0065}$. We find that the spinor-quintessence model closely reproduces the predictions of a phenomenological constant-$w$ model in the matter power spectrum, growth-rate observable $f\sigma_8$, growth index, and CMB temperature anisotropy spectrum, with deviations generally below the percent level. A comparison with $\Lambda$CDM and $w$CDM shows statistically indistinguishable fits, although the Bayesian information criterion favors the simpler $\Lambda$CDM model. Our results demonstrate that the spinor-field quintessence scenario remains consistent with current expansion-history and structure-growth observations when its perturbation evolution is treated consistently, providing a field-theoretically motivated realization of constant-$w$ dark energy at the background and linear perturbation levels.
Read the full abstract

Mathematical Physics

Fock-Space Formulation of the Boltzmann Collision Operator for Maxwell Molecules

We develop a representation-independent symmetric-Fock formulation of the nonlinear Boltzmann collision operator for Maxwell molecules. Starting from the standard velocity-space operator, we give a short derivation of Bobylev's Fourier identity and isolate its characteristic structure: two linearly related Fourier arguments followed by multiplication. We then prove, independently of collision geometry and of any coordinate realization, a canonical lift–fusion intertwining theorem: one-particle linear maps lift to the two incoming Fock legs and symmetric-algebra multiplication fuses them into a single outgoing state. Bargmann coordinates make this intrinsic construction especially transparent, but do not define it; Bobylev's collision maps subsequently select the Maxwell vertex from the already graded lift–fusion class. The resulting bilinear map obeys the exact grading \[ \widehat{\mathcal Q}(\mathscr H_p,\mathscr H_q)\subseteq\mathscr H_{p+q}, \] which gives an intrinsic origin for the triangular structures of Maxwell kinetics. Its one-vacuum restriction yields the complete Wang Chang–Uhlenbeck spectrum, while moment adaptation makes the nonequilibrium hierarchy strictly triangular and every finite compatible Fock jet an exact autonomous factor with global exponential relaxation. For the inhomogeneous problem, we extend realization covariance to the nonlinear two-leg vertex and evaluate the moving-frame connection independently through Hermite-function and Fourier realizations. The two routes give the same abstract Fock propagation and the same first Chapman–Enskog source; its traceless level-$2$ and vector level-$3$ sectors are relaxed by the exact Maxwell rates, giving $\Pr=2/3$. This provides an explicit computational demonstration that coordinate realizations may be chosen for economy without entering the structural Fock formulation.
Read the full abstract

General Mathematics

Radical defects, Wieferich primes, and the $abc$ conjecture

For coprime $a+b=c$, the parity class in which $a$ and $b$ are odd and $c$ is even admits an exact treatment. Triples are measured by the radical excess $E_{\varepsilon}=\log c-(1+\varepsilon)\log\operatorname{rad}(abc)$, and called transgressive at fixed $\varepsilon>0$ when it is non-negative. The $abc$ conjecture asserts that such triples are finite in number. The excess is written exactly in terms of the defect, the mass of repeated primes the radical discards, and rearranges into a linear threshold in which the smaller summand $s=\min\{a,b\}$ appears explicitly. Either $s$ stays bounded along a subsequence, or the defect must overshoot the threshold and rejoin it at an amplified exponent. For the Mersenne family $\mathcal{V}_m = (1, 2^m-1, 2^m)$ transgression at $\varepsilon=0$ holds precisely when $2^m-1$ fails to be squarefree, on a set of exponents of density $47/210$. The defect obeys the exact law $\Delta_m = \Omega_m + \log G_m$, where $\Omega_m$ collects the Wieferich primes dividing $2^m-1$ and $G_m$ is the largest divisor of $m$ whose prime factors divide $2^m-1$. This divisor is trivial on prime-power exponents and largest on $m_k = \operatorname{lcm}(1,\dots,k)$, where the margin $q(\mathcal{V}_{m_k}) - 1 \geq (1-o(1))\log m_k / (m_k\log 2)$ is the most any Wieferich-free mechanism can give. An $abc$ counterexample family along the Mersenne line would require infinitely many Wieferich primes with exponential order–defect growth. A final section separates two elliptic curves attached to the class: the Frey curve, whose minimal discriminant expresses the total defect with a bounded correction at $2$, and a congruent-number Jacobian, whose Szpiro quotient stays below $3$.
Read the full abstract

Computation and Language

trajectory-judge: What Outcome-Only LLM Judges Miss on Agent Trajectories

A direct test of an LLM judge of agent trajectories injects faults into correct runs and reports recall, per fault type or by whether the fault broke the environment outcome (loud) or not (silent). Such recall can credit a judge with detection it does not have; paired discrimination, its flag rate on the faults minus its rate on the clean runs they came from, exposes this. Our testbed, a deterministic support desk with a scripted oracle and a one-step fault injector, labels all 400 trajectories exactly. A 14B judge shown only the request and final reply scores 34% to 76% recall on four fault types that leave the reply unchanged. There its input is the clean run's, so its paired discrimination is zero and that recall is its flag rate on clean runs. Splitting by outcome survival does not fix this: its loud recall of 84% is a paired +0.393 and its silent recall of 45% a paired +0.048, all from the two fault types that change the reply. Told to check each step, the same model flags every fault of those four types and 0 of 100 clean runs (95% CI up to 3.6%). It does not reliably check the reply: of four invented promises it flags one every time and the other three once in 42 faults. Shown every step but asked only about the reply, it still reaches a paired +0.69 on reply-unchanged faults, against +1.00 when told to check each step. We recommend reporting paired discrimination against clean parents, split by whether the fault reaches the judge's input and by outcome survival, and release the testbed, raw verdicts and analysis pipeline.
Read the full abstract

Functional Analysis

On the $\rho$-numerical radius of rank-one operators and parametrized Buzano-type inequalities

Let $\mathcal{H}$ be a complex Hilbert space and $\rho>0$. We establish the explicit formula \[ \omega_\rho (a\otimes b)=\frac{1}{\rho}\|a\| \|b\|+\left|1-\frac{1}{\rho}\right||\langle a,b\rangle|,\qquad a,b\in\mathcal{H}, \] for the $\rho$-numerical radius of rank-one operators, a unified expression interpolating between the operator norm, the numerical radius, and the spectral radius, which correspond to $\rho=1,2$, and $\rho\to\infty$, respectively. As an application, we derive a parametrized family of Buzano-type inequalities for four vectors. We extract from it an explicit closed-form bound and, under a reality condition which is automatically satisfied in real Hilbert spaces, we determine the best bound in the family in closed form. Examples show that the resulting bounds can be strictly sharper than both the Cauchy–Schwarz and the Buzano inequality, and the classical Buzano inequality is recovered as a boundary case, which yields an alternative proof of it.
Read the full abstract

Computer Vision

A Globally Convergent Algorithm for Total Scaled-Gradient Variation via Cone-Constrained Bilinear Decomposition

The total scaled-gradient variation (TSGV) regularizer, derived from sparse modeling of piecewise-linear structures, has been shown to preserve edges and corners in image restoration. However, its highly nonconvex and nonlinear nature poses severe computational challenges, as existing methods often suffer from parameter sensitivity or lack convergence guarantees. To overcome this, we propose a tailored bilinear decomposition that decouples the nonlinear weighted gradient in the TSGV regularizer. This approach yields an equivalent optimization problem governed by cone or sphere constraints, depending on the chosen scaling function. In particular, the cone constraint plays a central role in characterizing edge- and corner-preserving behavior. We solve this reformulation using the alternating minimization method (AMM) equipped with a majorization–minimization strategy, ensuring a monotonic decrease in energy without step-size tuning. Furthermore, we provide a geometric interpretation of the edge-preserving properties of these constraints by analyzing their asymptotic behavior near image singularities. We establish the global convergence of the proposed method to a critical point within the Kurdyka–Łojasiewicz framework. Extensive numerical experiments on Gaussian denoising and non-line-of-sight (NLOS) imaging show that the proposed method achieves PSNR and SSIM competitive with or superior to representative variational methods, especially at high noise levels, and improves the structural reconstruction under dense and sparse scanning.
Read the full abstract

All authors Haibin Su, Chunlin Wu, Huibin Chang, Zhifang Liu

Information Retrieval

SilentProbe: Measuring Silent Failure in Production APIs Used as Agent Tools

An LLM agent calling a production API cannot distinguish a query that matched nothing from a query the server did not understand. Both return HTTP 200 with a parsable body, no exception to catch and no field to branch on. We ask what predicts which one occurred, and what it does to the agent. Auditing 721,320 parameters across 2,501 independently published OpenAPI documents, we find that 7.5% declare an enumeration and 15.2% declare any machine-checkable constraint at all, while 40.1% of documents state at least one constraint in prose that their schema does not encode. Executing 219 schema-derived perturbations against live commercial endpoints from 27 vendors, reached through a single aggregation layer (Monid) that publishes a schema and returns a run identifier for every call, we find that constraint form, not vendor identity, predicts honesty: machine-checkable constraints yielded an honest error in 111 of 111 cases, prose-only constraints failed silently in 44 of 61 (p = 2e-13). Twelve models across eight families then met these endpoints on ordinary tasks. A vocabulary that the description merely exemplifies was missed by every model on 88 of 88 attempts, while vocabularies written out in full were used correctly 88 to 91% of the time. Running the full agent loop, models detected the resulting silent failure in 12% of cases, repaired it in 0%, asserted a false negative to the user in 41%, and invented a figure in 12%. Promoting the vocabulary into the schema removes the failure, from 88 of 88 to 0 of 89. The fix is one line of schema rather than a better model. Code, schemas, perturbation sets, agent transcripts and per-call run identifiers are released at https://github.com/Jasper0122/silentprobe.
Read the full abstract

All authors Zongrong Li, Shengkun Ye, Feiyou Guo, Zuoyou Dang

Mathematical Physics

Alternative-mean trace divergences: geometry, data processing, and barycenters

Let $f:(0,\infty)\to(0,\infty)$ be a nontrivial normalized operator monotone function and set $s=f'(1)$. We introduce the alternative-mean trace functional $$ \altPhi_f(A,B) :=\Tr(A\nabla_s B) -\Tr\!\left( f(A^{-1}\sharp B)\,A\,f(A^{-1}\sharp B) \right). $$ on the positive definite cone. We prove that $\altPhi_f$ is a quantum divergence in the sense of Bhatia–Gaubert–Jain whose diagonal Hessian induces a positive multiple of the Bures–Wasserstein Riemannian metric. We also establish the sharp comparison $$ s(1-s)d_{\rm BW}(A,B)^2 \le \altPhi_f(A,B) \le (1-s+s^2)d_{\rm BW}(A,B)^2. $$
Read the full abstract

Machine Learning

Knowledge Distillation under Teacher Misspecification: An Order-Parameter Analysis of the Gap between Teacher Mimicry and Task Performance

Knowledge distillation trains a small student model to reproduce the outputs of a large teacher model, and its progress is typically monitored through the teacher–student discrepancy. The quantity of ultimate interest, however, is the student's error with respect to the true task. We study the relation between these two objectives in a minimal three-party model, a true teacher (generative model), a teacher, and a student, all soft committee machines, in which the true teacher contains a shared latent factor that the teacher cannot represent, with mismatch strength controlled by a single scalar $\dmiss$. Within an order-parameter description of online distillation, and exploiting closed-form (arcsine-type) expressions for all errors under error-function activations, we prove that the learning dynamics and the distillation error $\Ets$ are exactly invariant to $\dmiss$, whereas the true error $\Etzs$ and the gap $\Delta=\Etzs-\Ets$ are strictly increasing in $\dmiss$, with a rate that is amplified linearly by the complexity $M_0$ of the true teacher. Numerical phase diagrams over the plane spanned by true-teacher complexity and student capacity confirm the predicted deformation: the contours of $\Ets$ do not move while the landscape of $\Etzs$ rises systematically, and a teacher-miss regime, where mimicry succeeds but the task fails, expands with $\dmiss$. The results give a quantitative warning against evaluating distillation solely through teacher-mimicry metrics and identify the gap $\Delta$ as a minimal diagnostic for distinguishing teacher-miss from capacity-limited failure.
Read the full abstract

Atmospheric and Oceanic Physics

Is the Atmosphere Really Such a Good Blanket?

The incorrect application of the current methodology for treating and averaging Earth's surface temperatures has contributed to contradictions and misinterpretations in different scientific disciplines, particularly in climate change and global warming research. The methodology previously proposed by the author addresses these shortcomings and provides a physically based framework for further analysis. This paper demonstrates that the contribution attributed to the atmosphere in determining the warming of the Earth's surface may be substantially overestimated, with the estimated effect differing by approximately 10 degrees Celsius from the conventional interpretation. The analysis is based on the ERA5 reanalysis dataset produced by the European Centre for Medium-Range Weather Forecasts (ECMWF), using hourly skin temperature data for 1950-2025 on a global grid of approximately 1.04 million cells representing the Earth's surface. The proposed method enables an absolute assessment of global warming, confirms the long-term warming trend, and provides a comparative analysis with results obtained using the conventional methodology.
Read the full abstract

Number Theory

Kato's main conjecture for nonordinary modular forms

We prove Kato's main conjecture for modular forms at nonordinary primes for weights in the Fontaine-Laffaille range. The key ingredients in the proof are a reformulation of the conjecture in terms of signed Selmer groups due to Lei-Loeffler-Zerbes, certain $p$-adic families of Rankin-Eisenstein classes arising from the work of Lei-Loeffler-Zerbes and Kings-Loeffler-Zerbes, and the lower bound divisibility in an Iwasawa-Greenberg main conjecture for Rankin-Selberg $p$-adic $L$-functions obtained in our earlier work \cite{CLW}.
Read the full abstract

Optics

Artificial Intelligence in a Photonic Temporal Processor

Optical neural networks (ONNs) promise high-throughput and energy-efficient artificial intelligence, yet essentially all implementations so far encode information across space either in free-space arrays or in integrated waveguide meshes, tying the number of neurons to the number of physical components and fixes the routing topology at fabrication. Here we show that moving the computation into time decouples computational dimension from hardware dimension. Exploiting space-time duality, we implement optical diffraction and interference entirely in time domain, using thin-film lithium niobate modulators as time lenses and temporal masks, with chromatic dispersion providing the coupling between successive temporal neurons. We experimentally verify high-order, complex-valued matrix-matrix multiplications using just a single optical input/output port, scaling the computational dimensions far beyond the channel count. By incorporating optical feedback, we extend this platform into versatile neural networks, where the network layers, neuron numbers, and synaptic connections are fully programmable and in-situ trainable. Our temporal diffractive neural networks are successfully validated on various classification benchmarks, alongside image and video generation tasks. Notably, using this platform we demonstrate an all-analogue generative pipeline in which the latent variable is drawn directly from amplified spontaneous emission, so that no digital sampling or electronic modulation appears anywhere in the generative path. Furthermore, high-resolution images and videos are generated at high frame rates, outperforming state-of-the-art modulator-refresh-limited optical generative systems. These results establish a unified photonic temporal computing framework, providing a scalable and deployable pathway toward next-generation machine intelligence.
Read the full abstract

All authors Youlve Chen, Jinlong Xiang, Yimin Hu, Yuchen Yin, Chaojun Xu, Zhengshun Lei, Xin Wang, Yufeng Zhang, Yixiao Zhu, Qunbi Zhuge, Junwen Zhang, Wei Chu, Tao Lin, Yikai Su, Zhipei Sun, Xuhan Guo

Trading and Market Microstructure

The Convergence Rate of Stochastic Tracking with Application to Optimal Execution

We study the quadratic tracking problem of a general stochastic target process with absolutely continuous controls, with and without terminal constraint. We derive explicit, non-asymptotic upper bounds in terms of a Besov-type modulus of the target. These bounds yield sharp explicit rates that specialize to the square-root order for semimartingale targets. We then apply these results to a generalized Obizhaeva–Wang execution model with random terminal inventory. We first develop a Hilbert-space approach to characterize its optimal strategy, which includes jumps. To avoid such trading spikes, one regularizes the problem by a quadratic trading-rate penalty with coefficient $\varepsilon$. We then show that the regularized optimal execution cost—and therefore the excess price impact cost of the regularized optimal strategy—converges at the sharp rate $O(\sqrt{\varepsilon})$. Since the regularized optimal strategy is not available in closed form, we further construct a nearly optimal strategy which is readily implementable and shares the same approximation rate.
Read the full abstract

Computer Vision

Co-Evolutionary Prompt Optimization with Cross-Category Transfer for Zero-Shot Anomaly Detection

Zero-shot anomaly detection (ZSAD) has gained significant attention for its practical value in industrial inspection. Recently, CLIP-based approaches have been widely adopted in ZSAD due to their strong vision-language generalization capabilities. However, existing methods commonly employ continuous prompt embeddings for prompt optimization and encode semantics in latent vectors, which lack interpretability and scalability. To this end, we propose CoEvoAD, a co-evolutionary framework for discrete prompt selection. CoEvoAD performs prompt search in the discrete natural-language space using an evolutionary algorithm. Candidate prompts are iteratively generated, evaluated, and selected throughout population evolution, thus preserving the interpretability and composability of natural language. Furthermore, we introduce a Cross-Category Transfer Objective (CCTO), which treats held-out source categories as proxies for unseen categories and scores prompt rules based on their estimated cross-category transferability, effectively improving cross-category generalization. Extensive experiments are conducted to validate the effectiveness of CoEvoAD, and the results show that it achieves state-of-the-art performance across multiple anomaly detection datasets. The code is available at https://github.com/rstao-bjtu/CoEvoAD.
Read the full abstract

All authors Sisi Zhu, Changwei Yu, Renshuai Tao, Zhenliang Ni

General Relativity and Quantum Cosmology

How much of the black hole ringdown is sourced within the light ring?

Not much. It has been proposed that applying a rational filter to a black hole ringdown gravitational waveform reveals a direct wave component sourced near the event horizon. By solving the Teukolsky equation for a point particle plunging into a Kerr black hole, I show that the maximum amplitude of the waves sourced within the prograde light ring is $\lesssim 10\%$ that of those sourced by the full trajectory no matter if the waves are filtered or not. This agrees with the interpretation of quasinormal modes as waves orbiting the light ring and the direct wave as the power of these modes redistributed in time. Therefore, both quasinormal modes and direct waves are more a probe of the light ring than the near-horizon region. For GW250114, I estimate that the component of the waves sourced within the light ring have a signal-to-noise ratio $\rho \lesssim 1$ measured after the peak of the filtered waves.
Read the full abstract

Machine Learning (Statistics)

Deciding When to Decide: Testing Operational Suboptimality Under Distributional Shift

Deployed decisions are often optimized once and retained because updates impose operational, regulatory, or switching costs. As operating conditions change, when should such decisions be re-optimized? We study this question for stochastic optimization when the objective's functional form is known but the decision maker's trade-offs are encoded by an unknown preference parameter. Standard distribution-shift tests are poorly aligned with this goal: they can flag detectable yet decision-irrelevant changes without determining whether the incumbent decision has become materially suboptimal. We propose RADAR (Regret-based Assessment of Decision Adequacy and Risk), a decision-focused framework that uses inverse optimization to infer latent preferences and tests the deployed decision's optimality gap under the current distribution. By targeting regret, RADAR ignores decision-irrelevant shifts while detecting changes that warrant re-optimization. We develop two-sample and sequential changepoint procedures and establish asymptotic guarantees for Type-I error and power. Across synthetic optimization problems, a semi-synthetic capacity allocation task, and police-zone planning, RADAR more reliably distinguishes harmful from harmless shifts than decision-agnostic alternatives.
Read the full abstract

Computation and Language

Chain-of-Thought Faithfulness of Reasoning Models Varies with Where and How Preference Cues Are Delivered

Chain-of-thought (CoT) monitoring assumes that reasoning traces faithfully record the information that shapes a model's answer. Existing faithfulness tests often place explicit bias cues in the user message, while agents may encounter preferences through tool returns or raw artifacts. We introduce FACE-Eval (Faithful Attribution of Cue Effects Evaluation), a 5,100-sample evaluation that varies cue location (user message or tool return) and explicitness (direct summary or raw artifact). We measure verbalized commitment among cue-following answers and unverbalized adoption among all cued samples. We evaluate 15 open-weight models from eight families, with total parameters ranging from 4B to 1.60T. Every model has lower verbalized commitment for tool-return than user-message cues and for implicit than explicit cues. Unverbalized adoption is higher for tool-return cues on all 15 models and for implicit cues in 28 of 30 model-channel comparisons. A source-attribution prompt narrows the channel gap on seven models, sometimes by increasing user-channel unverbalized adoption, while telling models that their reasoning will be monitored does not reliably close the gap. We also use two transcript monitors (GPT-5.6-Luna and GPT-4o-mini) to detect preference adoption in the largest model of each family. Across 32 model-channel-explicitness cells, higher unverbalized adoption is associated with lower detection ability for both monitors (Pearson r=-0.54 and r=-0.78, respectively). These results suggest that CoT monitoring may be less reliable when preference information arrives through tools or must be inferred from raw artifacts, within the single-call, prefilled-tool setting tested here.
Read the full abstract

All authors Aryo Pradipta Gema, Neel Rajani, Rohit Saxena, Wai-Chung Kwan, Pasquale Minervini

Cryptography and Security

Benchmark Contamination: A Taxonomy Organized by Defeated Mitigation

A benchmark score is a joint property of the model, the evaluation harness, the elicitation budget, the sampled population, and contamination status. Leaderboards publish the model and the score, so capability and leakage stay observationally equivalent. Existing taxonomies classify contamination for automated detection, not the question a reporter faces at publication: given the mitigations already applied, which validity threats remain open? We introduce a taxonomy organized by the mitigation each type defeats – direct, derivative, temporal, distributional, and acquired – spanning training-time and evaluation-time leakage. Holding out a private test set closes the first alone. The fifth is acquired during the evaluation itself; because it is a property of one run, it must be recorded with the reported score rather than with the benchmark release. We operationalize it as a four-field disclosure protocol in which "unknown" is a valid entry, released under CC BY 4.0 with a JSON Schema, a validator, and worked examples. Two coders external to the design team applied a pre-registered instrument to 41 documents. Per-variable linear-weighted $\kappa$ runs from 0.00 to 0.35 (median 0.21) over 29 main-pass documents against a single-coder test-retest ceiling of 0.84, collapsing under the class skew the registration anticipated; pooling raises it to 0.46 through chance correction rather than better agreement. Two variables fall below the prevalence-robust threshold registered in advance: strata reporting and the acquired type introduced here. Disagreement concentrates on when a variable applies rather than on what a document states. Elicitation budgets are reported in 13% of documents, and no document addresses all five types. The contribution is the taxonomy, the score-side artifact that follows from it, and a pre-registered measurement of instrument reliability and current disclosure.
Read the full abstract

Analysis of PDEs

A Riemannian Extension of a Quadrature Surface Free Boundary Problem: Stability and Optimality

We study a quadrature surface free boundary problem on a smooth compact finite-dimensional Riemannian manifold $(M,g)$. The problem is formulated as a shape optimization problem involving a Dirichlet problem for the Laplace–Beltrami operator, with a geometric condition on the free boundary. Under suitable uniform geometric assumptions on the admissible class, we establish its compactness and prove the stability of the corresponding Dirichlet problems, including strong convergence of the associated states in $H_0^1(M)$. We derive the first-order optimality condition for the associated shape functional. Finally, we establish a Riemannian comparison principle for the second fundamental forms and the mean curvatures of tangent boundaries at a contact point. These results provide a rigorous extension of the quadrature surface free boundary framework from the Euclidean setting to compact Riemannian manifolds.
Read the full abstract

Machine Learning

A Causal Model for Locating and Unlocking Sandbagging in Model Organisms

Sandbagging models strategically underperform on evaluations while retaining the capabilities being measured. The evaluations that guide frontier-model deployment and governance then understate what these models can do. To understand the mechanism, we propose a causal model of how sandbagging is carried in the residual stream. Early layers write the sandbagging intent onto a single axis of the stream, and a later layer reads that axis and commits the answer. We study three instruction-tuned models (Qwen2.5-7B, Llama-3-8B, and Mistral-7B) and four ways of installing a sandbagging lock (prompting, fine-tuning, reinforcement learning, and circuit breaking). Each lock decides from the prompt whether to sandbag, and the fine-tuned and circuit-broken locks answer honestly whenever a password appears. The causal model predicts a window of layers, after the last sandbagging write and before the answer commit, in which a single-layer reference graft of the sandbagging axis to its honest value restores the full capability. The single-layer graft recovers the capability in 28 of the 33 runs of the prompted, fine-tuned, and RL-trained locks, with a median held-out recovery of 96%. The circuit-broken lock rewrites the whole state through a band of layers, and the single-layer graft fails at every layer. We therefore introduce a second intervention, context grafting, which replays the password's cached key/value activations so that every layer's attention reads them as additional context. Context grafting provably and empirically restores the full capability on all three models, and the recovery is surprisingly insensitive to the exact password content. More broadly, an auditor can use this causal model to design interventional auditing techniques for sandbagging models.
Read the full abstract

Artificial Intelligence

Can escalation channels redirect reward hacking toward defect disclosure?

When coding agents encounter defective test infrastructure they may reward-hack: hardcoding outputs or editing test files to pass tests they cannot legitimately satisfy, a pattern that has now appeared outside benchmarks, in a coordinated multi-agent intrusion of a major AI platform's production infrastructure. The same capability that lets an agent detect and exploit a defect could let it report one, given the right decision environment. We evaluate escalation channels, structured reporting tools available to the agent at the point of conflict, as a decision-environment intervention that both reduces reward hacking and surfaces the infrastructure defects that trigger it. A $2 \times 2$ factorial separates the contributions of an escalation tool, a standalone anti-reward-hacking policy, and their combination. Across 8 frontier models spanning 5 families, the combined intervention reduces reward hacking from 23.6% to 5.3% (mixed-effects logistic OR = 9.2, 95% CI 5.0–16.8, $p < 10^{-12}$) with no detectable cost or performance overhead, eliminating it entirely for 6 of 8 models. Escalation and hacking are near-perfectly mutually exclusive, with 98.7% of escalations involving no hacking (100% under the combined intervention). Beyond reduction, escalation channels function as diagnostic infrastructure: on top of monitoring, escalation adds +10.1 percentage points of defect detection coverage and is more accurate once it fires (99.4% vs 85.8%). Unlike containment-based approaches that risk outpacing growing model capabilities, escalation channels redirect capability toward disclosure rather than exploitation.
Read the full abstract

Artificial Intelligence

Toward Latent Language Model Skills Steering and Optimization: An Empirical Study

Skills, as a useful abstraction for the procedural capabilities of large language models (LLMs), capture how models perform structured, multi-step reasoning and program execution. Existing approaches typically treat skills as explicit, surface-level constructs specified through prompts or programs, leaving open the question of how such procedural capabilities are represented inside the model and whether they can be manipulated as structured objects in latent space. In this empirical study, we investigate whether procedural LLM skills can be represented as directions in activation space and whether vector-space operations over these directions can express skill-level behaviors. We find that procedural skills admit a vector-space representation: individual skill directions can be activated to shift model behavior; independently extracted directions can compose to form higher-level skills. Contrastive directions yield context-conditioned algorithmic personalization and optimization trajectories over skill directions evolve non-monotonically, with intermediate states often surpassing fully optimized solutions. These results support a representation-level view of procedural LLM skills: they admit a latent vector-space organization that allows direct manipulation through internal interventions.
Read the full abstract

All authors Xunyi Jiang, Junda Wu, Yuxin Xiong, Sheldon Yu, Tong Yu, David Arbour, Ritwik Sinha, Julian McAuley, Hongyi Wen

Machine Learning

Reference-Grafting Matches Fine-Tuning at Eliciting Sandbagged Capabilities

Sandbagging, in which a model deliberately underperforms on an evaluation despite retaining the underlying capability, threatens the safety evaluations that frontier-model governance depends on. The Elicitation Game found that fine-tuning elicits hidden capability from sandbagging model organisms whereas additive activation steering fails. We revisit that verdict with reference-grafting, which sets an activation's coordinate along a contrast direction to the value it takes in an honest reference, at a small set of circuits chosen by active learning. Across eleven fine-tuned password-locked models (1.5-32B, three architecture lineages), it recovers +94 to +101% of the honest-sandbagging gap, matching fine-tuning elicitation without weight updates or training labels; two to five paired examples suffice to fit the direction. Similar recovery holds for reinforcement-learning-induced sandbagging and for password-locked code generation. Grafting works because the fine-tuned lock is a thresholded gate: held-out accuracy stays at the sandbagged level until the grafted coordinate crosses a threshold near the honest reference, which is why additive steering and zeroing the coordinate often fail. The direction tracks how the sandbagging was induced rather than what is withheld – fit on grade-school science (ARC) it elicits withheld hazardous knowledge (WMDP), yet prompting, fine-tuning, and reinforcement learning each carry a different direction. Circuit-breaking marks the boundary: it reroutes activations on every forward pass, so the fixed edits we test are re-broken downstream and do not restore coherent generation.
Read the full abstract

Numerical Analysis

Dynamic subscales and unconditional semi-discrete stability for non-residual VMS approximations of generalised Newtonian flows

We analyse a non-residual variational multiscale finite element formulation with dynamic subscales for incompressible generalised Newtonian Navier–Stokes flows. The apparent viscosity is assumed to be bounded and Lipschitz continuous with respect to the shear rate, covering several regularised rheological laws. The method separates, through orthogonal projections, the unresolved pressure-divergence, convective, and pressure-gradient contributions. For the linearised semi-discrete problem, we prove well-posedness and an unconditional stability estimate in an anisotropic VMS norm. The estimate controls viscous dissipation, subscale energies, discrete divergence, and a coupled acceleration-convection-pressure balance. For the non-linear formulation, a fixed-point argument yields existence and optimal-order a priori bounds under suitable regularity and smallness assumptions. The analysis also provides weak-in-time control of the discrete pressure-convection balance. A key point is that the dynamic pressure-related subscale provides the time-derivative contribution needed to control the projected pressure-gradient error, a mechanism unavailable in the corresponding quasi-static setting.
Read the full abstract

All authors G. R. Barrenechea, E. Castillo, R. Codina, J. Pastene, P. Vega

Computer Vision

Text-Guided Diffusion-Based Adversarial Attacks on Chest X-Ray Images

As artificial intelligence is increasingly integrated into chest X-ray (CXR) interpretation, triage, and clinical decision support, understanding its vulnerability to adversarial manipulation is critical for safe deployment. Existing robustness evaluations, however, predominantly rely on pixel-space attacks that introduce numerically constrained perturbations but may not represent plausible radiographic variation. This limitation is particularly important in multi-disease CXR classification, where models simultaneously evaluate multiple overlapping pathologies and adversarial failures may alter several diagnostic predictions. We propose a text-guided diffusion-based adversarial framework that optimizes learnable text conditioning while keeping the diffusion generator and target classifier frozen, enabling adversarial generation through a learned image prior rather than direct pixel manipulation. We evaluate the framework across multiple classifier architectures in both binary atelectasis and multi-disease CXR classification and compare it with FGSM, PGD, and Carlini-Wagner attacks. Our approach consistently produced the greatest degradation in classifier performance, reducing AUROC to 0.3885-0.5646 in binary classification and 0.4441-0.4878 in the multi-disease setting, while achieving superior image fidelity (SSIM 0.9080, LPIPS 0.1670, FID 51.23). Importantly, clinician interpretation remained unchanged for 95.9% of binary and 73.8% of multi-disease adversarial images despite substantial changes in model predictions. These findings reveal a clinically important discrepancy between human and machine interpretation and demonstrate the need to extend medical AI robustness evaluation beyond conventional pixel-space attacks toward generative threat models that can expose failures under visually and clinically plausible image variations.
Read the full abstract

All authors Basudha Pal, Arjun Narayanan, Neha Ajith, Vikas R Bhat, Muhammad Umair

Computation and Language

Item-Mean Surrogates: Why Richer Personas Fail to Deliver Individual Fidelity in LLMs as Human Surrogates

LLMs are increasingly used to simulate and study human behavior. Such studies assume richer personas improve predictions of specific people's answers. We test this premise across four public datasets. First, we split each person's answer to an item into the item mean (the average human answer to that item) and the person's deviation from the item mean. Much of the accuracy that surrogate studies report comes from predicting item means. LLM predictions explain little of the deviations. We call this pattern item-mean surrogacy. Second, we split each deviation into stable signal and noise using a dataset in which the same participants answered the same questions twice. Stable signal is the part of a deviation that repeated answers share. Noise is the part that changes. In this dataset, more than half of the variance in deviations persists across occasions. Most of what the LLM misses is stable signal rather than noise. Third, we split stable signal into the person signal, a person's average deviation across questions, and the person-by-item signal, their remaining question-specific deviations. The person-by-item signal accounts for about 8.5 times as much variance as the person signal. The LLM recovers about a tenth of it. Richer personas, different models, and fine-tuning on human answers still leave most of the person-by-item signal unrecovered. Both fine-tuned models recover less of this signal than their untuned versions. These findings call for closer empirical scrutiny of LLMs as human surrogates.
Read the full abstract

Optimization and Control

The method of ellipcenters with momentum and relaxation for convex quadratic minimization

The method of ellipcenters (ME) is a recent technique developed for unconstrained minimization. Its iteration relies only on first order information and consists of building a suitable two-dimensional ellipse that tries to capture intrinsic ill-conditioning of the problem. The center of the ellipse is taken as the next iterate, which justifies the name of the method. ME was already shown to converge linearly when the objective function is smooth and strongly convex. The special case when the objective is quadratic was studied in the first paper on ME and is also the subject of our work here. In this paper, we propose two variants of ME: RelaxME which adds in ME an additional relaxation step at each iteration and MomME which embeds momentum in ME. We prove linear convergence of RelaxME and MomME and, in particular, we show convergence of RelaxME and MomME (and also of ME) in one iteration when the matrix of the quadratic form has only two distinct eigenvalues. Finally, we provide the results of numerical experiments which compare on five optimization problems (and different combinations of parameters defining these problems) ME, RelaxME, and MomME, with 3 other optimizers: the conjugate gradient method [10], Barzilai and Borwein gradient method with long step [2], and the gradient method with adaptive spectral step length [7]. On most instances, MomME provides the smallest number of iterations and conjugate gradient and MomME provide the smallest CPU times and similar CPU times. This opens optimistic possibilities for ME with momentum in broader settings.
Read the full abstract

Computation and Language

AI Can Be Easily Persuaded in Clinical Decision Making

As AI becomes increasingly integrated into clinical practice, it is playing a growing role in medical decision making. Medicine, however, is a high stakes and evidence based field, where decisions can directly affect patients' lives. It is therefore important to understand whether AI can maintain objective judgment when others try to persuade it. In this paper, we study how easily AI can be persuaded through controlled experiments. We find that professional authority, national background, institutional affiliation, claimed past performance, multiple physicians, supported clinician views, and repeated pressure can all affect AI decisions. Surprisingly, the same persuasive input changes about 10% more cases when it comes from a senior clinician than from a medical student. Simply claiming a better performance history consistently makes the physician more persuasive. More strikingly, a plausible clinician view can persuade AI away from a correct decision even when it is fabricated to support an incorrect answer. This indicates that AI can be strongly influenced by convincing support without reliably determining whether this view from the clinician is correct. Together, these findings suggest that AI can be easily persuaded by what people say, who says it, and how the opinion is presented. Therefore, it is essential for AI to maintain sound judgment under persuasion, enabling its safe and reliable use in high stakes medical decision making.
Read the full abstract

All authors Jiayuan Zhu, Jiazhen Pan, Fenglin Liu, Minhao Hu, Junde Wu

Statistics Applications

Rescheduled, not redefined: The moving plateau of old-age mortality

Whether the risk of death keeps climbing at extreme ages or levels off has divided researchers for a century. We show this conflict reflects a moving target. Using cohort data from twelve low-mortality populations, we find that mortality deceleration and plateau onset shift steadily later across cohorts born from the mid-19th to the early-20th century. The data support a plateau across the full cohort range, though evidence weakens for the youngest, incompletely observed cohorts. In these younger cohorts, deceleration begins near age 100, and the fitted plateau begins beyond age 108. Because past studies focused on different cohorts and fixed age ranges, they sampled different phases of the exact same shift. The mortality plateau has no fixed age: it has been rescheduled.
Read the full abstract