Computation and Language
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Wednesday 2 September
Large language models can look capable of logical reasoning, but correct or incorrect answers alone tell us little about what the model represents internally. We study logical verification in five open-weight transformer models using matched valid–invalid premise–claim pairs that vary across inference families, semantic domains, templates, and difficulty levels. Despite near-chance behavioral performance, logical validity is often almost perfectly decodable from hidden states and remains strongly decodable under held-out templates, domains, and inference families. Validity also remains highly decodable on behaviorally incorrect examples in the conditions where correctness-conditioned evaluation is well defined. At the same time, exhaustive leave-one-out tests reveal clear limits to this generalization, and interventions along probe-derived validity directions have only weak, nonspecific effects compared with random controls. Our results suggest that representing validity, expressing it in behavior, and using it causally are distinct. Validity related information can be strongly decodable from a model's hidden states without being reliably expressed in its output.
High Energy Physics (Theory)
An Infinite Family of Non-Rational VOAs from Strongly Coupled 4d Higgsless SCFTs
We study an infinite family of strongly coupled four-dimensional $\mathcal N=2$ superconformal field theories (SCFTs) distinguished by a trivial Higgs branch, known as the $(A_2,D_{3m+1})$ Argyre-Douglas theories. We propose that their associated vertex operator algebras (VOAs) are the doublet algebras $\mathcal A(4m+2)$, an infinite family of non-rational vertex operator superalgebras with a remarkably simple strong generating set of only three fields. We provide several highly nontrivial checks of this proposal. In particular, we reproduce the four-dimensional conformal anomalies $a$ and $c$ from the VOA and analytically prove the exact equality between the Schur index of the SCFT and the supercharacter of the VOA. A key ingredient is a diagonal-gauging realization of the $(A_2,D_{3m+1})$ theories in terms of two simpler building blocks, which makes the Schur-index computation tractable. Remarkably, we find that the resulting Schur index of the $(A_2,D_{3m+1})$ theory coincides with that of $\mathcal N=4$ $SU(2)$ Super-Yang-Mills theory, up to an overall prefactor and an appropriate identification of fugacities. We also discuss a generalization to the two-parameter family $(A_{2s},D_{(2s+1)m+1})$, whose members likewise have trivial Higgs branches and admit diagonal-gauging realizations. A particularly interesting subfamily is $(A_{2s},D_{2s+2})$, for which the four-dimensional conformal anomalies coincide, $a=c$. As a byproduct, we also show that the doublet VOA $\mathcal A(4m+2)$ is not classically free for positive integer $m $, and we discuss its implications for the Schur and Macdonald indices. Our results reveal a systematic connection between Higgsless SCFTs, diagonal gauging, and strongly finite but non-rational vertex operator algebras.
Read the full abstract
Human-Computer Interaction
Decoding Decision Correctness from EEG Under High Cognitive Workload in Virtual Reality: Implications for Collaborative Brain-Computer Interface Teams
Collaborative Brain-Computer Interfaces (cBCIs) offer a promising mechanism to augment team decision-making, but existing approaches rely exclusively on evidence available only after a decision has been made and reported, such as reaction time or stated confidence. This limits their use to explaining or discounting a decision after the fact, rather than informing a team's response before it is finalised. We tested whether spatial-covariance EEG features could instead provide a genuinely pre-emptive signal of an operator's decision correctness, available within the response window itself, and whether such a signal depends on cognitive workload. Using a continuous virtual reality target-detection task, participants (N = 23) completed a within-subject workload manipulation (High vs. Low). At the team level, weighting votes by this pre-emptive neural signal, available before a response is committed, produced substantial accuracy gains on contested (evenly-split) trials under High Workload (57% to 88% as team size increased from 2 to 16), but was actively detrimental under Low Workload. Critically, this advantage held even against post-hoc behavioural signals: confidence was the strongest single team-level signal overall, but by definition cannot inform a decision still in progress, whereas the neural signal can. These findings indicate that EEG-based decision-reliability signals are not a general-purpose team augmentation tool, but a workload-conditional one, with clear implications for when and how cBCI systems should be deployed in operational teams.
Read the full abstract
Materials Science
Chemical pressure and vacancies in crystals under nonhydrostatic stress
Inhomogeneous stress is a driving force for diffusion of vacancies in crystals. The other way around a non-uniform distribution of vacancies induces stress. The accepted theory of composition-stress coupling in crystals is Larché-Cahn (LC) theory. Composition is defined in terms of the population of lattice sites with a limit of one-particle per site. Coupling is modelled by adding a compositional strain term to the elastic strain in the constitutive relation for stress. An alternative mechanism, proposed here, is letting the site binding energy vary with spatial (deformed) density. This generates a chemical pressure adding to the elastic stress. The governing equations for this model are derived using non-equilibrium continuum thermodynamic methods. The gradient of the chemical pressure has a dual function acting both as the drift force for migration and as an effective internal one body force in the Cauchy equation for the elastic stress. The practical evaluation presented in the paper is restricted to equilibrium properties. We examine the effect of externally applied non-hydrostatic stress, either in the form of surface tractions or a one-body force density (gravitation). The model system is a one component crystal with a fixed number of lattice sites. The number of particles can be variable but is always smaller than the number of lattice sites. The linear elastic response is modelled by the standard Lamé stress tensor. The results for systems deformed by surface tractions are in qualitative agreement with LC theory allowing for differences in the expression for the elastic moduli. Deviations are more serious for the crystal deformed by gravitation.
Read the full abstract
Computer Vision
Uncertainty-Guided Adverse Weather Restoration via Gated Transformer Network
Restoring images degraded by adverse weather remains challenging due to spatially heterogeneous degradations. Many existing weather-specific restoration models rely on weather-agnostic global aggregation, naive cross-scale fusion, and deterministic objectives, which struggle to handle heterogeneous degradations in all-in-one adverse-weather settings. To address these limitations, we propose an Uncertainty-guided Adverse-weather Restoration Network (UAR-Net), a weather-specific AiO framework that integrates a gated transformer with balanced multi-scale skip connections. Specifically, we employ Gated Dual-scale Transformer Blocks (GDTB) to jointly model selective global interactions and multi-scale local structures, a progressive Balanced Multi-scale Skip Connection (BMSC) for balanced multi-scale feature integration, and an Uncertainty-Aware Refinement Head (URH) that performs artifact removal, detail enhancement, and predictive uncertainty estimation. The model is supervised by a Brightness-Aware Energy Loss (BAE-Loss) to encourage accurate reconstruction with well-calibrated uncertainty. Extensive experiments demonstrate that our method achieves state-of-the-art performance across multiple adverse-weather benchmarks. The codes will open source upon acceptance.
Read the full abstract
Numerical Analysis
Analysis of the Virtual Element Approximation for the Smagorinsky Model
In this paper, we consider the Smagorinsky model for the Navier-Stokes equations within a virtual element framework. Under the standard assumption of small data, we prove the existence and uniqueness of a solution. Assuming more regularity to the solutions, we derive the known \del{suboptimal} convergence rates $h$, using $k>1$, for the a priori error estimates of the Smagorinsky model in two dimensional domains. We additionally prove that divergence-free virtual discretizations provide improved convergence orders, with weaker regularity assumptions than in the finite element literature. We conclude the paper with numerical results that corroborate the theory.
Read the full abstract
Physics Education
Student responses to a modified Force Concept Inventory: The Newtonian Mechanics Quiz
Even in the era of modern physics, understanding and applying Newton's classical Laws of Motion are key skills for physicists and form a core component of instructional programmes. Students arrive in the classroom with intuitive ideas about force and motion that are often misaligned with Newtonian thinking. Prior research has shown that these incorrect intuitive ideas are remarkably persistent, and can stay with students throughout their education and beyond. Reliable instruments, that can accurately evaluate student understanding of Newtonian mechanics, are required to track student progress and evaluate the efficacy of instructional programmes. The Force Concept Inventory (FCI), a conceptual quiz composed of multiple-choice questions, is a well-established instrument, developed to address these challenges. The FCI, and how students respond to its questions, has been widely investigated and discussed in the literature. Notably, research has shown that: correct responses to questions do not always correspond to correct application of Newtonian concepts; and the limited response options given for multiple-choice questions often do not capture the full range of student thinking. There is therefore a burgeoning need to develop new instruments that probe students' understanding at a deeper level, by interrogating conceptual understanding from multiple perspectives and incorporating different question formats that require students to construct and consider their responses. We introduce the Newtonian Mechanics Quiz (NMQ), a modified version of the FCI that incorporates follow-up questions in a variety of formats, each with different diagnostic functions. This article describes the development and validation of the NMQ instrument, and illustrates several novel analyses enabled by the student response dataset that can offer deep insight into students' conceptual understanding of Newton's Laws.
Read the full abstract
Systems and Control
Multivariable Geometric Laplace Transform and Fault Detection in Distributed-Converter Lines
Monitoring a DC line with many distributed power converters is a genuinely spatio-temporal problem: the information about a localized fault travels along the whole conductor and reaches a few measurement points mixed with the dynamics of the line itself. This paper develops a two-dimensional geometric Laplace transform (t,x) -> (s_t,s_x) over a commutative subalgebra of the geometric algebra Cl(4,0), isomorphic to Segre's bicomplex numbers, in which two bivectors B_t and B_x act as independent imaginary units for the temporal and the spatial phase. Because the two phases live in algebraically distinguishable planes, a fault at position x_f leaves a transformed residual that factorizes as F_f(s_t) e^{-s_x x_f}: its temporal nature stays in the first factor and its location can be read as a geometric argument of the second. On this representation we build a transmission-line model of the converter line and its space-time dispersion relation, a distributed control by admittance shaping, including an exact treatment of discrete converter sites (spatial sampling, aliasing, and a per-converter droop realization that is exact on the sub-Nyquist band), and a fault diagnosis chain that detects, localizes and classifies injection-loss, shunt, sensor and local-controller faults, extends to multiple simultaneous faults with automatic order selection, and distinguishes the outage of a plant from a cable defect. As an integral object the transform is known in bicomplex analysis, and with a single independent variable it reduces to the complex Laplace transform; the contribution lies in its geometric embedding and in its operational use for fault diagnosis in distributed-converter networks. All results are reproduced by an accompanying open implementation.
Read the full abstract
Differential Geometry
Mixed Radial Volume Comparison under Spectral Ricci Bounds
Let $(M^n,g)$ be complete, let $u>0$, and assume $\operatorname{Ric}_g-\alpha\frac{\Delta_g u}{u}g\ge (n-1)\kappa g$. We introduce mixed radial balls associated with the conformal metric $u^{2\alpha}g$. For these mixed radial balls, we obtain a space-form-sharp model comparison and polynomial weighted volume growth without pointwise bounds on $u$. As applications, we give a radial derivation of the spectral Bonnet–Myers and sharp volume theorem of Antonelli–Xu, and a short volume growth derivation of the stable Bernstein theorem in $\mathbb{R}^4$.
Read the full abstract
Mesoscale and Nanoscale Physics
Rank-Selective Optical Tomography of Higher-Wave Altermagnetism
Identifying the spatial rank of higher-wave altermagnetic order optically is challenging because local electric-dipole response does not uniquely resolve distinct continuum harmonics. We show that finite photon momentum turns one-photon spin-resolved absorption into a rank-selective tomography. For the planar $|m|=\ell$ sector of an even-parity $\ell$-wave component, a joint Fourier projection in polarization and momentum angle isolates $\mathcal T_\ell\propto\eta_\ell q^{\ell-2}$, yielding the hierarchy $d:q^0$, $g:q^2$, and $i:q^4$; phase changes track rotations of the selected magnetic harmonic. A Ward-consistent finite-$q$ microscopic calculation reproduces these powers without imposing them. Although discrete crystal symmetry can generate lower-order local aliases, they are orthogonal to the selected momentum harmonic and cannot contaminate it below $q^{\ell-2}$. In MnTe, whose nonrelativistic parent order is three-dimensional $g$ wave while spin–orbit coupling lowers the exact relativistic spin-momentum-locking symmetry, first-principles calculations show that more than $99.9\%$ of the Fourier power of the Néel-projected $A$-region spin-energy contrast remains in the parent $g$-wave-derived $m=3$ harmonic. Structured near fields place the required momentum window within experimental reach.
Read the full abstract
Quantum Physics
Circuit-Level Loss Performance of RHG and Foliated Floquet Color Codes in a Compound Photon–Atom Quantum Architecture
A central question for fault-tolerant quantum computing is which quantum error-correcting codes are best suited to a given hardware architecture. Here we compare the Raussendorf–Harrington–Goyal (RHG) code, the Foliated Floquet Color Code (FFCC), and the reduced FFCC in a compound photon–atom architecture that directly generates measurement-based quantum computation (MBQC) resources with near-deterministic photon–atom CZ gates. RHG serves as a natural benchmark, while the FFCC variants allow us to study whether reduced graph degree improves performance under an architecture-aware circuit-level loss model with delayed heralding and correlated bond-loss propagation. We construct two generation schemes compatible with the compound hardware and evaluate circuit-level thresholds under periodic boundary conditions. RHG achieves the highest circuit-level threshold, 2.75%, and its threshold falls below that of reduced FFCC only for large excess loss on intermodule CZ connections. RHG also achieves the lowest logical error rate in most resource-matched comparisons, but some low-loss windows favor reduced FFCC. Overall, we show that when the hardware supports the native gates and connectivity required for MBQC, the benefits of lower graph degree must be weighed against each code's intrinsic IID loss tolerance, generation-scheme details, and hardware-aware resource overhead.
Read the full abstract
Logic
Ampleness in the Farey graph
We show that the first-order theory of the Farey graph is $1$-ample but not $2$-ample. Along the way, we prove that it weakly eliminates imaginaries and observe that forking is trivial.
Differential Geometry
Deformations of the $\mathfrak{osp}(n|2)$-action on the superspace of symbols of differential operators on $\R^{1|n}$
We study formal deformations of the natural $\osp(n|2)$-action, $n\geq 3$, on the superspace $\Sc^n_d=\bigoplus_{k\geq 0}\Fc^n_{d-\frac{k}{2}}$ of symbols of linear differential operators on weighted densities over $\R^{1|n}$. Starting from the first cohomology space computed in \cite{10}, we compute the cup-product $\Hd^1\vee \Hd^1\to \Hd^2$ which carries the quadratic obstructions. The answer is governed by the $\osp(n|2)$-invariant operators $A_k=\eta_1\cdots\eta_n\partial_x^{k-1}$: the two cocycles $h_k$ and $\widetilde{h}_k$ spanning the off-diagonal part of $\Hd^1$ are exactly the two derivatives of the coboundary of $A_k$ with respect to the two weights. Consequently, all the products of two off-diagonal classes and all the products of two diagonal classes vanish, and the whole obstruction is carried, for each $k$, by a single non-trivial 2-cocycle $\Omega_k$. If $2d\notin\N$ the space $\Hd^1\vee\Hd^1$ is identically zero, so every infinitesimal deformation is integrable. If $2d=m\in\N$ we obtain exactly $m$ quadratic integrability conditions, $\tau_{2-n-k}(t_k-\widetilde{t}_k)+\tau_k\widetilde{t}_k=0$, $1\leq k\leq m$, and we prove that they are also sufficient: no condition of order $\geq 3$ occurs and the versal deformation is of degree one in the parameters. In particular every integrable formal deformation is equivalent to its infinitesimal part.
Read the full abstract
High Energy Physics (Theory)
Localization and Abelianization of Strings on Group Manifolds: The Simply Connected Case
We study the localization formula for the partition function of the Wess–Zumino–Witten (WZW) model on the torus for compact, connected, and simply connected simple Lie groups. We identify a missing factor in the original localization treatment of Murthy and Witten and show that it agrees with the result obtained from the Hamiltonian formulation. We trace its origin to the abelianization of the Wess–Zumino amplitude, which reduces to a flat Kalb–Ramond B–field holonomy in a Narain lattice CFT associated with the maximal torus. This establishes a direct relation between the localization result and the lattice description of the corresponding abelian theory. We further analyze the point–particle limit of the torus partition function, in which the WZW model reduces to quantum mechanics on the group manifold, and reproduce Frenkel's heat–kernel trace formula.
Read the full abstract
Numerical Analysis
Geometry-dependent rank defect in $C^1$ cubic spline space
Determining the dimension of the $C^1$ cubic spline space $S_3^1(\mathcal{T})$ on an arbitrary nondegenerate planar triangulation has remained unresolved since the 1970s. Schumaker's lower bound includes a local correction $\sigma$ for singular interior four-stars, and it was conjectured that this bound is always attained. We disprove this conjecture by constructing a one-parameter family of nondegenerate realizations of a fixed 18-triangle complex, with only the central vertex moving as $v_6(t)=(t,0)$ on the admissible interval $I=(-3/4,24/55)$. The family exhibits three distinct cases. For $t\in I\setminus\{1/5,3/83\}$, the lower bound is attained and $\dim S_3^1(\mathcal{T}(t))=33$. At $t=3/83$, the central four-star is singular, $\sigma=1$, and the resulting dimension 34 is exactly accounted for by the classical local correction. At $t=1/5$, however, all interior vertices are nonsingular and $\sigma=0$, yet $\dim S_3^1(\mathcal{T}(1/5))=34>P_{\mathcal{T}(1/5)}(1,3)=33$. The smoothing-cofactor calculation shows that the dependence at $t=3/83$ is confined to the central vertex block, whereas the dependence at $t=1/5$ couples all seven interior vertex cycles even though every individual block has full row rank. A complementary Bernstein–Bézier calculation gives the same dimension profile. Thus the singular-four-star correction does not capture every geometry-dependent contribution to $\dim S_3^1(\mathcal{T})$; genuinely global compatibility must also be taken into account.
Read the full abstract
Analysis of PDEs
On the local well-posedness of the Benjamin-Ono-Zakharov-Kuznetsov equation
We study the Cauchy problem for the Benjamin–Ono–Zakharov–Kuznetsov equation on $\mathbb R^2$. Following the strategy of Kenig and Ziesler, we establish new maximal-function estimates adapted to the BO–ZK equation and use them to implement the Kenig–Koenig method. As a result, we improve the best previously known isotropic result of Nascimento (2020), lowering the local well-posedness threshold from $s>5/4$ to $s>19/16$. At the BO–ZK endpoint, the resulting isotropic data class also contains the anisotropic $E^{5/4+}$ class of the preceding theory. On bounded subsets of $H^s(\mathbb R^2)$, the lifespan may be chosen so that $ T\gtrsim_s \bigl(1+\norm{u_0}_{H^s}\bigr)^{-8}. $ The proof combines a sharp dyadic mixed maximal-function estimate with an anisotropic local-smoothing mechanism that exploits the complementary behavior of the longitudinal and transverse group velocities. In particular, transverse dispersion compensates for the degeneration of longitudinal smoothing near the characteristic region. Together with refined short-time Strichartz estimates and a modified energy argument, these ingredients close the nonlinear estimates at the stated regularity. Existence, uniqueness, and continuous dependence on the initial data are then established in the corresponding solution class. The resulting threshold reflects the present optimization of the method and is not claimed to be sharp.
Read the full abstract
Machine Learning
IFW-BLS: Dual-Robust Broad Learning System with Intuitionistic Fuzzy Wave Loss
Broad Learning System is an efficient randomized learning model that expands network width through feature and enhancement nodes and estimates the output weights without deep backpropagation. Its standard least-squares training, however, is vulnerable in two different ways: (i) large residuals caused by noise, outliers, or corrupted labels can dominate the objective, and (ii) all samples are treated as equally reliable even when some lie in ambiguous or locally conflicting regions. This paper proposes IFW-BLS, an Intuitionistic Fuzzy Wave Broad Learning System that addresses these two sources of fragility within one optimization model. The first robustness mechanism is residual-level protection, obtained by replacing the squared loss with the bounded, smooth, and asymmetric wave loss. Boundedness prevents extreme residuals from receiving unbounded influence, while asymmetry allows positive and negative deviations to be penalized differently when the dominant error direction varies. The second mechanism is sample-level credibility control, obtained through intuitionistic fuzzy scores that combine global class-center consistency with local neighborhood conflict. The resulting model evaluates the wave loss on credibility-weighted residuals, so unreliable samples are down-weighted before the bounded loss further limits the effect of extreme errors. A Nesterov accelerated gradient based optimizer is used to solve the proposed objective, avoiding the explicit matrix inversion used in conventional BLS. Experiments on UCI benchmark datasets validate the superiority of the proposed IFW-BLS model over the baseline models; additional corruption experiments also show more stable performance than BLS under noise and outlier contamination.
Read the full abstract
Artificial Intelligence
UTP-Bench: Uncertainty-aware Travel Planning Benchmark
Large Language Models (LLMs) have recently demonstrated strong capabilities in automated travel itinerary generation. However, real- world travel planning is inherently uncertain: transportation delays, crowd fluctuations, and unexpected stochastic delays frequently inval- idate otherwise feasible schedules. Existing benchmarks like TravelPlanner and TripCraft assume deterministic environments, evaluating only static constraint satisfaction and ignoring whether generated plans remain robust when such uncertainties arise. To address this limitation, we introduce UTP-Bench1 , a large-scale benchmark for uncertainty-aware travel planning. The dataset integrates real-world travel data spanning 504 cities of India, including attractions, restau- rants, accommodations, and multi-modal trans- portation networks. To model realistic disrup- tions, UTP-Bench incorporates empirical delay distributions and crowd-density patterns col- lected from major cities, enabling evaluation of travel plans under stochastic conditions. We further propose three evaluation metrics, namely Buffer Adequacy Score (BAS), Crowd- Aware Timing Score (CATS), and Transport Delay Absorption Score (TDAS), which quan- tify the ability of generated itineraries to main- tain robustness against transit delays and crowd variability. Experiments with state-of-the-art LLMs like GPT-5, Qwen3, Mistral and Phi-4 re- veal substantial gaps between model-generated and human-authored plans, particularly in tem- poral buffering, delay-aware transportation scheduling, and crowd-sensitive planning.
Read the full abstract
Numerical Analysis
A Multiscale Finite Element Method for reaction-diffusion eigenproblems arising from neutronics
We consider reaction-diffusion eigenproblems with oscillatory diffusion and reaction coefficients. The reaction coefficient magnitude is large: the corrector equation identified by periodic homogenization involves both the diffusion and the reaction operators. We study the numerical approximation of this problem using the Multiscale Finite Element Method (MsFEM). This now classical method is a finite element type method that performs a Galerkin approximation of the oscillatory problem on a specific, problem dependent, basis set. The basis functions are precomputed in an offline stage. Inspired by homogenization theory and using some filtering ideas, we show how to define these basis functions in order to obtain an efficient method. The comprehensive set of numerical experiments that we present, in periodic and non-periodic cases, for the scalar-valued version of the problem (which is then self-adjoint) and for the vector-valued version of the problem (which is then in general non self-adjoint), demonstrates the performance of the approach. Some theoretical arguments complement the numerical observations.
Read the full abstract
Chaotic Dynamics
An active Lorentz gas: walking droplets in periodic media
The Lorentz gas is a paradigmatic model in dynamical systems theory for understanding the origin of nonequilibrium transport in terms of microscopic deterministic chaos. In the periodic setting, a point particle scatters elastically off disks arranged on a two-dimensional lattice. Here we replace the disks by smooth potentials and the particle by the widely studied walking droplet, which propels itself on a vertically vibrating fluid. In the low-memory limit, this droplet reduces to a particle with nonlinear active friction. We call this system an active Lorentz gas. Using extensive numerical simulations, we analyze how dissipation generated by the active deterministic dynamics alters the phase-space structure of the corresponding conservative Lorentz gas. We find that islands of stability collapse into attracting and repelling sets. To characterize these structures, we introduce an energy-variance filtering method that distinguishes localized periodic, quasi-ballistic periodic, and chaotic trajectories, enabling the construction of bifurcation diagrams in a non-conservative setting. We identify parameter regimes exhibiting strong bifurcation cascades, anomalous diffusion, and significant phase-space contraction. Our results establish the active Lorentz gas as a rich framework for studying transport in dissipative dynamical systems and provide a bridge between active matter and classical models of chaotic transport, with potential implications for hydrodynamic quantum analogs in periodic media.
Read the full abstract
Optimization and Control
On the Duality between Feature and Sample Screening
Feature and sample screening reduce the cost of machine learning by eliminating irrelevant features and noninformative samples, respectively. Although recognized as primal-dual counterparts, their relationship remains informal and model-dependent. Viewing screening and duality as transformations of objective functions, we introduce Fenchel-Rockafellar (FR) representations, a class of convex problems encompassing the Lasso and SVM that is closed under both transformations. We then prove that feature and sample screening form an equivariant pair: dualization followed by feature screening is equal to sample screening followed by dualization.
Read the full abstract
Machine Learning
Coverage, Not Targeting: A Structural Regime in Multi-Turn Agent Credit Assignment
Multi-turn agentic RL increasingly treats credit assignment as a targeting problem: given a terminal verifiable reward, per-turn methods localize credit onto the turns that mattered. We identify the structural quantity that predicts when this is the right move, the verifier information density V_d = k/C (the fraction of an agent's C-step causal chain whose per-turn correctness the verifier exposes), and show that terminal-state verifiers sit deep in a low-V_d regime where targeting is the wrong axis. In controlled shared-rollout comparisons on tau^2-bench that separate reward density from credit geometry, a continuous dense reward spread uniformly beats the sparse binary outcome reward (net-harmful on 4/5 seeds), while concentrating the same advantage on progress turns or on random turns is equally harmful: targeting is second-order. The mechanism is coverage: terminal-state verification collapses the observable signal to a single final-write turn (k=1 in 98% of rollouts) while success requires a 5-8 step chain of prerequisite tool calls. A synthetic phase boundary places the crossover at V_d* ~ 0.8, whereas measured V_d is ~0.15 on tau^2-bench and ~0.4 on BFCL V3; uniform also wins on BFCL, where a matched-concentration shuffled control is negative on 8/8 seeds. The effect reproduces across model families on ToolACE-2-8B (Delta = -0.048 over 32 pre-registered seeds; an independent 20-seed replication is itself significant), and a pre-registered matched-budget breadth sweep traces a monotone dose-response whose deficit vanishes only at full chain coverage, with a reward-to-go arm reaching full-coverage parity. Uniform redistribution is the zero-information coverage default that per-turn schemes must beat; we contribute the matched-concentration shuffled control that any targeting claim should clear.
Read the full abstract
Analysis of PDEs
Principal eigenvalues of nonlocal operators with advection: sharp scaling limits and spectral phase transitions
We consider generalized principal eigenvalues of one-dimensional nonlocal dispersal operators with a drift of fixed sign. A principal difficulty in this non-self-adjoint setting is the absence of a variational characterization of the principal value. In the symmetric drift-free problem, the critical nonlocal-to-local limit can be treated through a quadratic variational structure and Sobolev-seminorm approximation as Berestycki-Coville-Vo \cite{BCV}. The drift destroys this structure and, on a bounded interval, introduces at the same time a one-sided inflow–outflow boundary geometry. Replacing the missing variational argument by estimates which remain stable under singular rescaling is therefore a central technical issue. Our approach is direct : we work with a single generalized principal value and prove the maximum principle, simplicity in the positive cone and the scaling limits from the equation itself, without passing through auxiliary generalized principal eigenvalues analogous to $\lambda_1'$ and $\lambda_1''$ in Berestycki–Rossi \cite{BR}. The replacement mechanisms are directional Harnack inequalities and coefficient barriers, direct–adjoint identities, logarithmic Collatz–Wielandt transforms, Fourier coercivity of zero extensions and scale-dependent localization. At the critical diffusive scale, Fourier compactness and an exact energy–transport identity recover the missing inflow Dirichlet condition and identify the local Dirichlet limit without a Rayleigh quotient. For variable coefficients we determine the complete small-range phase diagram on bounded intervals and on the line, and we obtain sharp large-range three-term asymptotics by a rank-one reduction and a corner Laplace analysis in the advective travel-time variable. In the homogeneous whole-line problem the critical correction is of order $\sigma^2$.
Read the full abstract
Algebraic Geometry
Nef Cones of Hilbert Schemes of Orthogonal Grassmannians
We show the number of connected components of the Hilbert scheme of orthogonal Grassmannians under certain condition, and use this result to describe the geometry of the Hilbert scheme. Subsequently, we determine the Nef cone of the Hilbert scheme by identifying curves dual to the generators of its Neron-Severi group. Our results generalize those of ordinary Grassmannians by Seong, and our approach adapts his proof technique alongside relevant Schubert calculus.
Read the full abstract
Computation and Language
Before the Script, Set the Stage: How Worldview Simulation Amplifies Psychologically Grounded Persuasion in Multi-Turn Jailbreaking
Multi-turn jailbreak attacks demonstrate that harmful intent can be distributed across dialogue, yet existing methods obscure what conversational mechanisms drive vulnerability. We introduce BLUEPRINT, a safety-evaluation framework separating a factorized social-influence strategy space from WORLDVIEWSIM, a cross-turn situational context module. Monte Carlo Tree Search optimizes turn-level combinations of 18 theory-grounded influence factors across a four-turn trajectory. Across six frontier models, BLUEPRINT achieves near-ceiling ASR on major open-weight and proprietary models, while requiring the fewest average queries (2.46). The resulting trajectories further reveal model-specific vulnerability among resistant targets: each responds to distinct influence factors and strategy transitions, yet all share a common recovery pathway-shifting toward concrete, executable task framing consistently escapes hard-refusal states. Ablations confirm operational cues matter most: making requests actionable has the largest impact, gain framing is unusually potent, and some legitimacy appeals can backfire. These findings suggest robust multi-turn safety requires monitoring not only harmful content, but also how dialogue state makes unsafe requests appear concrete and locally executable.
Read the full abstract
Graphics
WildFab: Multi-Axis 3D Printing from Models in the Wild
Multi-axis 3D printing enables support-free fabrication and improved part quality, but robustly processing real-world geometries remains challenging. Models from design workflows or direct data acquisition often contain solid–shell combinations and non-manifold structures. Handling such models in the wild typically requires time-consuming geometry repair, which may alter the intended geometry. In this work, we present WildFab, a computational framework for multi-axis 3D printing that directly computes spatial toolpath and global collision-free motion from input models. Our pipeline builds on a hybrid query representation that combines a neural unsigned distance field (UDF) with a regularized generalized winding number field (reg-GWN). The UDF supplies differentiable surface-distance and direction queries, while the reg-GWN resolves near-surface ambiguity in the fitted UDF by providing reliable surface localization and a solid-void indicator. Based on this representation, we introduce a high-precision spatial toolpath computation algorithm that iteratively projects points between optimized guidance-field level sets and reg-GWN gradient-magnitude ridges. Subsequently, we develop an efficient and robust coarse-to-fine collision checking scheme for motion planning: UDF-based rejection first identifies potential collisions, while time-varying reg-GWN verification accurately resolves collision pairs for both solid and shell components. We validate WildFab on diverse inputs, demonstrating successful computation from non-manifold parametric surfaces, voxelized topology-optimization results, implicit models, raw scanned point clouds, and non-watertight meshes. The fabrication results highlight our method's ability to advance end-to-end design-to-3DP workflows.
Read the full abstract
Number Theory
A Two-Variable Zeta Function for a Parity-Perturbed Hofstadter Q-Recursion: The Exceptional t = -1 Slice and Gaussian Boundary Layers
We study the parity-perturbed Hofstadter $Q$-recursion $$ \widetilde Q(1)=\widetilde Q(2)=1,\qquad \widetilde Q(n)=\widetilde Q(n-\widetilde Q(n-1)) +\widetilde Q(n-\widetilde Q(n-2))+(-1)^n, $$ and the associated two-variable Dirichlet series $$ Z_{\widetilde Q}(s,t)=\sum_{n\ge1}n^{-s}\widetilde Q(n)^{-t}. $$ The estimate $\widetilde Q(n)=n/2+O(n/\sqrt{\log n})$ gives the exact domain of absolute convergence $\operatorname{Re}(s+t)>1$. With $w=s+t$, we separate the universal term $2^t\zeta(w)$ and derive exact transport, frequency-position, and dyadic renormalization identities. The main result concerns $t=-1$. For $E(n)=2\widetilde Q(n)-n$ and $A(X)=\sum_{n\le X}E(n)$, the binary-arch clock yields $$ A(X)=X\log_2X+X\Omega\!\left(\log_2\frac{3X}{32}\right) +O\!\left(\frac{X}{\sqrt{\log X}}\right), $$ where $\Omega$ is an explicit continuous periodic function. This continues the normalized correction to $\operatorname{Re}w>0$ and yields a boundary resonance lattice: a double resonance at $w=0$ and simple resonances at $2\pi i m/\log2$. After subtracting the full-slice order-$X$ skeleton, we analyze the negative-even arch channel. Its companion-forest layers have a weak Gaussian limit, and a canonical subsequence realizes the optimal $n/\sqrt{\log n}$ pointwise scale with an explicit signed constant. The negative-arch mass satisfies $$ A_r=\frac{512}{9\sqrt{2\pi}}\frac{16^r}{\sqrt r} \left(1-\frac{13}{16r}+O(r^{-2})\right). $$ We do not claim a full-slice continuation across $\operatorname{Re}w=0$.
Read the full abstract
Computer Vision
The Diagnosis a Reporter Leaves Unspoken: Surfacing Frozen Tumor Features for Brain-Tumor MRI Reporting
A capable brain-MRI report generator can still be, in effect, diagnostically silent. When a multi-chain chain-of-thought (CoT) reporter built on a medical Mistral-7B backbone is evaluated on held-out cohorts, it names most meningiomas and almost all metastases "glioma" (diagnosis recall 0.44/0.07). Yet the answer is not absent from the model: a supervised linear probe applied to its frozen segmentation features recovers the three tumour cohorts at 0.82 macro-F$_1$ (5-fold cross-validation; chance $\approx$0.33). We introduce NeuroFusion, an assistive reporter that surfaces this latent signal rather than overriding it: discriminative field-classifier heads over per-lesion features condition a fast, single-pass draft-then-review decoder on their committed outputs. Built on the identical Mistral backbone, this restores the diagnosis (meningioma 0.92, metastasis 0.75) and wins 8 of 9 prose-content comparisons across three held-out cohorts (RaTEScore, RadGraph-F$_1$, GREEN; Holm-corrected paired BCa), with no significant loss on the ninth, at 5-6x lower latency ($\approx$80 vs. 457 s/case). A controlled negative result sharpens the mechanism: a learned diagnosis pin that overrides the decoder instead of merely informing it collapses out-of-distribution metastasis recall to 0.03. Grammar-constrained decoding keeps 92.3% of records schema-valid, making every sentence entailment-checkable (7.5% contradicted vs. 36.8% for the direct baseline). In a blinded nine-case pilot, two board-certified neurologists independently rated NeuroFusion highest in every tumour type, the only system with zero critical errors, and gave it the top-rated sign-off in eight of nine cases (six outright, two ties).
Read the full abstract
Systems and Control
Leveraging Time-Causal State Variable Aggregation for Real-Time Schedule of Massive Air Conditioners
Air conditioner (AC) loads offer promising flexibility for active distribution networks to manage uncertainties, such as those in renewable energy generation, electricity prices, and load demand. However, real-time scheduling of ACs is challenging due to their massive temporal coupling constraints and time-causal uncertainties. To address this, a novel time-causal aggregation-based approximate dynamic programming (TCA-ADP) algorithm is proposed for efficient scheduling. The time-causality requirements for aggregating state variables are first analyzed to align with the real-time sequential decision-making process. Subsequently, an enhanced aggregation model is developed to ensure both high accuracy and adherence to time causality. The aggregation process is further reformulated as a linear program to optimize aggregation parameters and enable tractable computation. Accordingly, the TCA-ADP leverages aggregated state variables to approximate the value function as a new way, balancing computational efficiency and economy against the large value function space of massive ACs. By training the value function offline using historical data, the TCA-ADP efficiently achieves near-optimal real-time scheduling of massive ACs through parallel and closed-form disaggregation. Case studies demonstrate the effectiveness and scalability of the TCA-ADP, highlighting its aggregation accuracy, uncertainty handling, and the trade-off between economy and tractability.
Read the full abstract
Algebraic Geometry
Membranes and Maps
In positive degree, equivariant Gromov-Witten invariants of Calabi-Yau fivefolds are expected to admit an interpretation in terms of M2-branes, while we conjecture that constant maps are governed by the corresponding supergravity index. We make the first expectation precise by proposing a modular interpretation of M2-branes supported on several smooth curves meeting at an $n$-fold point. Together, the two pictures yield conjectural formulas for the Gromov-Witten invariants, which we translate into closed formulas for pointed and unpointed quintuple Hodge integrals. We show that these conjectures imply a K-theoretic Gromov-Witten/Pairs correspondence for local curves in degree one and link the generating series of constant maps to Donaldson-Thomas theory of points. We also prove the conjectures in two limits of the equivariant parameters. Along the way, we obtain new closed formulas for certain triple Hodge integrals.