Peer Revenue Equity in science

Papers

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

All of arXiv

Thursday 27 August

Number Theory

Value distribution of multiplicative functions along linear fractional sequences

For $a,c\in\mathbb{N}$ and $b,d\in\mathbb{Z}$ such that the (non-empty) set \[ R_{a,b,c,d} :=\left\{\frac{an+b}{\,cn+d\,}: n\in\mathbb{N}\right\} \cap\bigl(\mathbb{Q}_{>0}\setminus\{1\}\bigr) \] is multiplicatively recurrent, we give a complete characterization of the set of limit points of every unimodular multiplicative function $f\in\mathcal{M}$ along $R_{a,b,c,d}.$ We show that the possible limit sets are either the finite subgroups of the unit circle or the entire circle, thereby extending the dichotomy of Klurman–Mangerel.
Read the full abstract

Computer Vision

Retrieval Heads Meet Vision: Uncovering How VLMs Locate and Extract Visual Information

Vision-language models (VLMs) can locate an image region referred to by a text prompt and route the corresponding visual evidence to the output, yet the internal mechanism behind this behavior is not understood. Inspired by retrieval heads in large language models, we ask whether VLMs contain an analogous mechanism for visual retrieval. We answer affirmatively by introducing Visual Retrieval Heads (VRHs), a small subset of attention heads (about 1.7-2.6%) that are causally responsible for grounding text descriptions to image regions. To find them, we recast existing head-scoring methods under a unified design space over query tokens, key aggregation, and cross-sample aggregation. We then show that scoring attention from output prediction tokens with a sum over the ground-truth referent region most reliably identifies causal heads. Across eleven VLMs and five referring-expression benchmarks, masking only the top 20 VRHs reduces grounding accuracy by up to 80 percentage points, while masking the same number of random heads has little effect. Beyond replicating the causal-sparse-universal triad established for text retrieval heads, VRHs exhibit several properties not previously reported: they generalize across visual reference tasks, remaining causal on attribute, spatial, counting, and visual-math benchmarks despite being discovered through bounding-box prediction; they are functionally specific, preserving output format while corrupting localization; and they are architecturally shared, transferring causally across VLMs that share an LLM backbone but differ in vision encoder, projector, and instruction tuning.
Read the full abstract

All authors Chanho Park, Daehyeon Choi, Jihyun Lee, Minhyuk Sung

Combinatorics

Refutation of the Non-Cancelling Intersections Conjecture

The Non-Cancelling Intersections (NCI) conjecture of Amarilli, Monet and Suciu [arXiv:2401.16210] states that the union of a finite family of sets can always be built from its algebraically non-cancelling intersections using only disjoint unions and subset complements. In Wilhelm [arXiv:2608.19414] the conjecture was shown to fail when the witnessing dot-algebra expression is required to be left-linear. Here we remove that restriction and show that the conjecture is false in general: there is a finite lattice admitting no dot-algebra representation of its top element whatsoever. The counterexample is a lattice $P_{p,\mathfrak{m}}$ as in Wilhelm [arXiv:2608.19414], and the argument differs in only two ways. First, we replace the sequential "toggle game" of Wilhelm [arXiv:2608.19414] by a corresponding tree-shaped object, the plane tree, which stands to dot-algebra trees as the toggle game stands to left-linear ones. Second, we use a marked plane in which there is no admissible set of any size between $2p$ and $4p$, which also removes the need for the Erdős–Beck theorem and for the arithmetic Nullstellensatz. Consequently $p$ need not be astronomically large: every prime $p \ge 10^{5}$ works.
Read the full abstract

Pattern Formation and Solitons

Wedge problems and dispersive shock waves in the two-dimensional Toda lattice

We study the formation and interaction of dispersive shock waves (DSWs) in the two-dimensional Toda lattice subject to wedge-type initial conditions, and show that their interaction gives rise to a discrete analog of Mach reflection for dispersive shock waves in discrete systems. The initial jump across each leg of the wedge acts locally as a Riemann problem for the one-dimensional Toda lattice, producing two oblique DSWs whose leading-edge soliton amplitude is determined explicitly by the one-dimensional Whitham modulation theory. The two-dimensional nature of the problem manifests when these oblique DSWs meet along the symmetry axis. We show that, for compressive wedges (i.e., when the initial conditions are such that two oblique DSWs that are generated propagate toward each other), a critical slope $q_{\mathrm{cr}}$ separates two regimes: in the subcritical regime ($q<q_{\mathrm{cr}}$) the interaction is resonant and it produces an expanding DSW whose amplitude, length and velocity are explicitly computed by using exact soliton solutions of the two-dimensional Toda lattice; in the supercritical regime ($q>q_{\mathrm{cr}}$) the interaction is ordinary and produces a localized peak whose amplitude is determined analytically. We also show qualitatively that a similar dichotomy between two regimes exists for expansive wedges (i.e., when the initial conditions are such that the two oblique DSWs propagate away from each other). We confirm all analytical predictions by comparing them with the results of direct numerical simulations. Finally, we show that the continuum limit of the result is consistent with the analogous theory for the Kadomtsev-Petviashvili equation, providing an independent validation of the analytical framework.
Read the full abstract

All authors Marco Calabrese, Gino Biondini, Christopher Chong, Panayotis Kevrekidis

Rings and Algebras

Very good gradings on structural matrix rings

Let $R$ be a nonzero associative unital ring, let $G$ be a group, and let $\rho$ be a preorder on $\{1,\ldots,n\}$. A $G$-grading on $\rho$ induces a very good $G$-grading on the structural matrix ring $M_n(\rho,R)$. We show that, for each of the properties trivial, symmetric, epsilon-strong and strong, the grading on $\rho$ has the property if and only if the induced ring grading does. The epsilon-crossed product and crossed product properties pass from $\rho$ to the ring, but the converses fail in general. We also give a concrete criterion for epsilon-strongness and show that a very good $G$-grading on $M_n(\rho,R)$ that is strong satisfies $|G|\leq n$. When $\rho$ is an equivalence relation and the neutral component is diagonal, very good gradings correspond bijectively to free partial actions of $G$ on $\{1,\ldots,n\}$ with orbit relation $\rho$. These gradings are epsilon-crossed products, and over a field the correspondence gives a classification up to graded algebra isomorphism.
Read the full abstract

All authors Patrik Lundström, Johan Öinert, Laura Orozco, Héctor Pinedo

Information Retrieval

Scaling Graph Neural Networks for Friend Recommendation: Multi-Hash User Embeddings and Temporal Neighbor Sampling

Friend recommendation is inherently graph-structured: the relevance of a potential connection depends on multi-hop social context rather than user attributes alone. However, deploying message-passing GNNs on a production-scale social graph with hundreds of millions of users and tens of billions of edges requires addressing numerous modeling and systems challenges. We present a scalable end-to-end GNN ranking system for production social graphs, focusing on two design choices that are critical in this setting: multi-hash ID embeddings and temporal neighbor sampling. Multi-hash embeddings are common for high-cardinality features, but industrial GNN systems typically either ignore trainable IDs or accept full embedding tables, exceeding 200 GB for our graph. We integrate multi-hash as the primary node representation, reducing the ID-embedding table size by more than 98 percent while preserving ranking quality. Temporal neighbor sampling is well understood in principle, but existing implementations scan full adjacency lists, which is a non-starter for users with tens of thousands of friends. We implement timestamp-sorted CSR storage with binary search, reducing the per-node temporal sampling cost from $O(deg(v) + k)$ to $O(\log(deg(v)) + k)$. Beyond these components, we show that this combination scales and yields measurable production impact. On a graph with 194M users and 28B edges, offline ablations isolate each design choice's contribution. In an online A/B test, our system increases friend additions from recommendations by 16 percent and unique friend adders by 11.5 percent over a strong production baseline. We release our framework for distributed training and inference on large temporal graphs.
Read the full abstract

High Energy Physics (Theory)

Casimir operators of $4D\,, \mathcal{N}=2$ supersymmetry in the harmonic approach

We construct Casimir operators of $4D, \,\mathcal{N}=2$ supersymmetry algebra in the harmonic superspace approach, and provide the explicit expressions for them in terms of covariant derivatives in the analytic basis. Specifically, the superisospin operator $C_I$ is expressed in terms of new 'long' harmonic derivatives $\nabla^{\pm\pm}, \nabla^0$ as $C_I = \frac{1}{4}\left(\nabla^0\right)^2 + \frac{1}{2}\left\{\nabla^{++},\nabla^{--}\right\}$. These derivatives involve non-local terms with the inverse Box operator $\Box^{-1}$ and satisfy modified $SU(2)$ algebra relations in which the eigenvalues of $\nabla^0$ on analytic superfields are shifted by -2 relative to the standard $U(1)$ harmonic charges. We also construct $\mathcal{N}=2$ superhelicity and super-isohelicity operators relevant to the massless case (with vanishing $\Box$) and find their eigenvalues for few instructive examples of $4D, \,\mathcal{N}=2$ superfield theories.
Read the full abstract

Strongly Correlated Electrons

A continuous confinement-deconfinement transition in a triangular quantum magnet

A continuous transition between phases hosting distinct excitations—bosonic magnons versus fermionic spinons—is a long-sought phenomenon in quantum magnetism, analogous to the confinement–deconfinement transition in quantum chromodynamics. We report evidence for such a transition in the triangular-lattice antiferromagnet TlYbS$_2$. Antiferromagnetic order develops below $T_\mathrm{N} \approx 0.53\,\mathrm{K}$. A $c$-axis field suppresses this order, driving the system into a gapless quantum spin liquid with a spinon Fermi surface above $\mu_0 H_\mathrm{c} \approx 3\,\mathrm{T}$, evidenced by a finite residual linear term in thermal conductivity, a Pauli-like susceptibility, and a temperature-independent NMR Knight shift. Approaching $H_\mathrm{c}$ from above, the scattering rate of itinerant excitations is strongly enhanced while their density of states shows no critical enhancement, atypical of conventional magnetic quantum criticality. These results point to a continuous confinement–deconfinement transition governed by fractionalized excitations beyond the Ginzburg–Landau paradigm.
Read the full abstract

All authors Suguru Hosoi, Sejun Park, Michihiro Hirata, Minseong Lee, Adam P Dioguardi, Joe D Thompson, Filip Ronning, Allen O Scheie, Kumpei Imamura, Kenichiro Hashimoto, Takasada Shibauchi, Bishnu P Belbase, Arjun Unnikrishnan, Johannes Knolle, Arnab Banerjee, Yuji Matsuda

High Energy Physics (Theory)

Doubly-scaled planar $\mathcal{N} = 4$ SYM & Carroll Holography

In their seminal paper, Okuda and Penedones (OP) put forward an intriguing proposal towards holography in asymptotically flat spacetimes (AFS). They showed that the flat space limit of 4 point bosonic tree-level string amplitude in AdS$_{5}\, \times\, S^{5}$ is dual to a specific double scaling limit of 4 point correlator in four dimensional (4d) $\mathcal{N} = 4$ super Yang-Mills (SYM) theory. In this paper, we show that the resulting boundary correlators can be used to define a 4d Carrollian conformal field theory (CFT) on future null infinity. This is done by mapping the set of doubly scaled correlators in SYM to null infinity, the conformal boundary of AFS. A Carroll CFT thus emerges in the infinite 't Hooft coupling limit of $\mathcal{N} = 4$ SYM. We extend the OP analysis to higher point functions and show how the Gram conditions on flat space amplitude put constraints on the double scaling limit of SYM correlators. We then use the soft factorization theorem for dilatonic string amplitude to write a recursion relation for the $({1}/{2})$-BPS sector of $\mathcal{N} = 4$ SYM in the double scaling limit. Finally, we show that the essential ideas underlying such a double scaling limit can be used as a tool-kit to build a class of Carrollian correlators, and hence define a Carrollian CFT, via certain integral transforms of flat space amplitudes of non-gravitational effective field theories.
Read the full abstract

All authors Arjun Bagchi, Prateksh Dhivakar, Alok Laddha, Partha Paul

Computation and Language

Consolidating RLVR Capabilities Across Domains: A Deep Dive into Fusion Paradigms

Reinforcement learning with verifiable rewards (RLVR) improves specific capabilities of large language models, but covering multiple capabilities often involves training separate domain experts and subsequently consolidating them. We organize three fusion paradigms by the artifacts they reuse: Merge combines expert task vectors, Mix RL pools their datasets, and multi-teacher on-policy distillation (MOPD) uses both. Because they have largely been studied in isolation, how they compare and how to choose among them remain unclear. We compare all three using shared experts and data across model scales and a multi-domain benchmark suite. Although their average performance differs by at most 1.4 points, the gap reaches 8.6 points on a single benchmark, with domain-level variation tracking cross-domain relations visible in task-vector geometry. Training dynamics expose distinct constraints: Mix RL depends on domain mixture proportions, MOPD remains bounded by its teachers, and Merge compresses all expert updates into one. All three improve single-sample accuracy without measurable gains in solution coverage or losses in held-out capabilities. These results yield a practical guideline: use Merge when experts already exist and cheap fusion is paramount; Mix RL when training a unified model without experts, with domain proportions adjusted for cross-domain transfer; and MOPD when preserving domain-specific gains matters more than surpassing teachers or minimizing end-to-end cost.
Read the full abstract

All authors Siye Wu, Kai Yang, Yuchen Cai, Xin Xu, Peng-Yuan Wang, Jiaxuan Wang, Jiashun Liu, Jiafei Lyu, Yangkun Chen, Saiyong Yang, Yanghua Xiao

Populations and Evolution

Reservoir: A Large-Scale Simulated Dataset for Training and Evaluating Epidemiological Models

Large-scale, standardized datasets have driven many advances in AI-based scientific modeling, from protein structure prediction to natural language processing. Infectious disease epidemiology is increasingly adopting AI methods for forecasting, surveillance, and outbreak analytics, but the time-series data available to train them remains orders of magnitude smaller than the corpora behind the advances seen in other fields. Because the scope of real-world epidemiological data cannot practically reach the scale needed to train truly large-scale AI methods, simulated data provides a possible alternative. Here we introduce Reservoir, a large open simulator and dataset of realistic epidemic simulations in which every trajectory carries complete ground-truth labels, including quantities that cannot be measured directly in a real outbreak, such as true infection counts, time-varying reproduction numbers, and counterfactual intervention effects. Reservoir is generated by a stochastic simulator with realistic noise and reporting artifacts, together with interventions with configurable timing, compliance, and age-dependent efficacy. The current release contains 500,000 outbreak trajectories spanning one billion simulated days across diverse pathogen characteristics, population structures, and intervention regimes. Reservoir enables counterfactual experiments, surveillance-design studies, and training of epidemic models at a scale real-world datasets cannot provide.
Read the full abstract

Computer Vision

Reconstructing Humans and Objects in Interaction using Large Reconstruction Models

Estimation of Human-Object Interactions in 3D (3D HOI) is a fundamental problem in 3D computer vision with applications in AR/VR, robotics, and embodied AI. However, reconstructing these interactions in 3D remains challenging due to depth ambiguities, occlusions, and object shape variability. Existing approaches are primarily concerned with reprojection and contact constraints, fitting parametric human models and object templates to 2D images. In this paper, we explore a different avenue. We present MILO, a framework that leverages the visual capabilities of Large Reconstruction Models (LRMs) to recover detailed 3D human-object interactions from a single image. Our key observation is that LRMs provide a powerful geometric scaffold that preserves relative human-object arrangement and proximity cues. This significantly simplifies the reconstruction procedure, reframing the problem as interpreting the LRM mesh: we segment it into human and object components, fit a parametric body model to the human part, and optionally align an object template to the object part (if such a template is available). MILO achieves strong reconstruction accuracy and outperforms existing baselines across multiple benchmarks and interaction scenarios. Our code is available at https://ac5113.github.io/MILO.
Read the full abstract

Robotics

CLAP: Cross-Embodiment Video World Models are Zero-Shot Physical Simulators

State-of-the-art action-conditioned video models are typically restricted to a single robot embodiment, preventing them from leveraging the vast corpus of heterogeneous video data that contains rich signals for learning generalizable physics. To bridge this gap, we introduce CLAP, a framework for cross-embodiment action-conditioned video generation capable of being trained on diverse, internet-scale videos across human and robotic agents. CLAP is grounded in the insight that universal physical laws govern spatiotemporal dynamics regardless of the actor. However, cross-embodiment learning is non-trivial because action representations vary sharply across robot platforms and are typically absent in human videos. CLAP addresses this fundamental challenge through the following core contributions. First, CLAP reconciles disparate action spaces using end-effector poses, language instructions, and latent actions. Second, to resolve their individual limitations, CLAP introduces a curriculum-based cross-embodiment learning recipe that first learns foundational physical priors across unlabeled video data using latent actions and subsequently grounds them in end-effector action spaces for zero-shot deployment to real-world tasks. Crucially, CLAP approaches or surpasses state-of-the-art single-embodiment video models in challenging environments like DROID. These performance advantages compound via few-shot adaptation to establish a novel paradigm for training single-embodiment video world models. Ultimately, CLAP delivers the most comprehensive suite of action-conditioned video world models to date - spanning diverse action-conditioning spaces (end-effector, language, and latent) and robot morphologies (including cross-embodiment, DROID, Bridge, bimanual YAM robots, and G1 humanoids). We open-source all code and models. Project Website at https://omni-clap.github.io .
Read the full abstract

Combinatorics

Asymptotic Bounds for Online Ramsey Numbers of Stars versus Long Paths and Cycles

The online Ramsey game for graphs $G$ and $H$ is played on the infinite complete graph $K_\mathbb{N}$. In each round, Builder chooses an edge, and Painter colors it red or blue. The online Ramsey number $\tilde{r}(G,H)$ is the smallest integer $t$ for which Builder has a strategy guaranteeing a red copy of $G$ or a blue copy of $H$ within $t$ rounds. For every fixed integer $k\ge4$, the best-known asymptotic lower bounds for $\tilde{r}(K_{1,k},P_n)$ and $\tilde{r}(K_{1,k},C_n)$ are $\left(\frac{k+3}{4}+o(1)\right)n$ as $n\to\infty$. We improve the corresponding asymptotic upper bounds from $\left(k+o(1)\right)n$ to $\left(\frac{k+3}{3}+o(1)\right)n$ as $n\to\infty$.
Read the full abstract

Combinatorics

The Erdős-Gallai bound for consecutive even cycle lengths

Erdős and Gallai in 1959 proved the seminal result that every $n$-vertex graph with no cycle of length at least $2t+2$ has at most $\tfrac{2t+1}{2}(n-1)$ edges. We prove the extension that, for every sufficiently large $t$, the same quantity is also the sharp extremal bound for graphs with no $t$ consecutive even cycle lengths, resolving a conjecture of Verstraëte. Thus, at the Erdős-Gallai threshold, forcing an entire interval of even cycle lengths costs no more than forcing its longest member. More precisely, every $n$-vertex graph $G$ with $e(G)\ge \tfrac{(2t+1)(n-1)}2$ $\bullet$ either contains $t$ consecutive even cycle lengths, $\bullet$ or equality holds and $G$ is connected with every block isomorphic to $K_{2t+1}$. As consequences, for every sufficiently large even $k$ we determine the sharp edge thresholds forcing a cycle of length $0\pmod k$ or $2\pmod k$, answering questions of Bai, Grzesik, Li, and Prorok and of Gao, Li, Ma and Xie, respectively, for sufficiently large even $k$. The proof develops a stability-enhanced sublinear expander method. Its main new ingredient is a dense-case decomposition that recovers the lengths lost in the expander extraction by combining a flexible dense core with rooted cycle families in the vertices outside the core.
Read the full abstract

All authors Yaobin Chen, Hong Liu, Xia Wang, Xin Wei, Fan Yang

Algebraic Geometry

Proper moduli spaces of isolated non-normal singularities

We study the moduli of isolated non-normal singularities by introducing the functor $\mathrm{Crimp}_S^d(X)$ of crimpings of $(X \to S)$ corank $d$. This globalizes Ishii's moduli functor of subrings of finite colength. By verifying Artin's criteria, we prove that if $X \to S$ is a separated morphism of finite presentation, then $\mathrm{Crimp}_S^d(X)$ is represented by a separated algebraic space of finite presentation over $S$, which is proper when $X \to S$ is proper. We use this to construct a semistability condition and moduli space for geometrically unibranch curves of genus zero.
Read the full abstract

All authors Jiucheng Dai, Daniel Halpern-Leistner, Mingjun Sun, Ni Tang, John Veliz

Computation and Language

How Language Models Organize and Structure Moral Knowledge

How do large language models (LLMs) organize moral knowledge? Models detect moral content broadly, but detection is a low bar. We ask whether they go further, distinguishing moral foundations from one another and organizing the relationships between them geometrically. We train six independent linear probes on open-weight language models, one per Moral Foundations Theory (MFT) category (care/harm, fair/cheat, lib/oppress, loy/betray, auth/subv, sanc/degrade), and examine how the resulting directions relate to each other in representation space. We find the directions neither collapse into a single moral detector nor isolate from one another. Rather, they span a near-maximal number of independent dimensions while sharing a positive common component. The shared component is the signature of integration, and it is moral-specific relative to a matched non-moral concept battery built identically (mean pairwise cosine 0.26 vs. 0.013). The geometry is consistent across architectures and scale and reaches its integration regime early in pre-training, well before probe accuracy saturates. The structure the model discovers shows no evidence of the individualizing/binding distinction predicted by Moral Foundations Theory (an underpowered test: only 10 distinct splits exist, so it cannot reject at the 0.05 level) but rather reflects corpus statistics. Extending to moral dilemmas, each dilemma direction partially composes from its component foundations, at 2.7x a mismatched-pair baseline, while the majority of its variance encodes conflict-specific structure. The model represents moral tension itself, not a pre-resolved judgment.
Read the full abstract

Earth and Planetary Astrophysics

An Equation of State for Supercritical Silicate-Hydrogen Mixtures at Sub-Neptune Interior Conditions

Many sub-Neptunes are expected to contain long-lived molten silicate interiors beneath dense H2-rich envelopes. At pressures and temperatures near the atmosphere-interior boundary, MgSiO3 and H2 may become fully miscible, forming a supercritical silicate-hydrogen fluid. Here we use density functional theory molecular dynamics simulations to construct a self-consistent equation of state for supercritical MgSiO3H4, corresponding to 3.86 wt% hydrogen expressed as equivalent H2. We fit the simulation results with a Helmholtz free-energy formulation that yields density, entropy, heat capacity, thermal expansivity, bulk modulus, and Gruneisen parameter from a single thermodynamic surface. We find that the hydrogen-bearing fluid is lower in density than dry MgSiO3 liquid and deviates significantly from ideal specific-volume mixing between MgSiO3 and H2. The resulting excess volume varies with pressure, indicating that the interaction between hydrogen and the silicate framework evolves with compression. Structural analysis shows increasing Si-H coordination and decreasing persistent H-H bonding at high pressure, consistent with a transition away from molecular H2-like bonding toward a more strongly coupled silicate-hydrogen fluid. We incorporate the MgSiO3H4 equation of state into a composition-dependent MgSiO3-H lookup table and apply it to representative sub-Neptune structure models. The models show that hydrogen partitioning between the atmosphere and condensed interior affects the planetary adiabat relative to the MgSiO3 liquidus. These results demonstrate that supercritical silicate-hydrogen fluids have distinct thermodynamic and structural properties that must be accounted for when modeling sub-Neptune interiors.
Read the full abstract

Astrophysics of Galaxies

Map of the Interstellar Continuum FUV Field within 800 Parsecs

The far-ultraviolet (FUV) field, consisting of photons with energies between 6 and 13.6 electron volts, plays an important role in the chemical and thermodynamical evolution of the interstellar medium. FUV continuum radiation is produced by bright stars in the galaxy and absorbed by interstellar dust. We used newly available maps of the dust distribution, in conjunction with catalogs of stellar positions and spectra, to create a map of the stellar FUV field within 800 parsecs of the Earth. The map shows a total FUV intensity in the solar neighborhood in line with existing measurements (0.7 times the Draine field), but substantial point-to-point fluctuations exist. Near the Galactic midplane, the FUV intensity decreases with increasing gas density up to a density of approximately 10 hydrogen atoms per cm$^3$, due to absorption of light by dust. Between 10 and 50 hydrogen atoms per cm$^3$ the UV field increases again, due to a positive correlation between high density gas and young hot stars. We find that the spectrum is generally harder where the UV field is higher, and softer in regions where dust extinction is more important. The resulting map, as well as our catalog of derived stellar properties, are both available for download for those wishing to study a nearby region of the ISM.
Read the full abstract

All authors David Lomeli, Kedron Silsbee, Alexei Ivlev, Marta Obolentseva

Quantum Physics

Krylov Break Times from an Inhomogeneous Lieb–Robinson Light Cone

Krylov and Lanczos approximations are used in quantum dynamics, quantum subspace methods, and Hamiltonian learning. A practical question is how long an $m$-dimensional Krylov truncation can be trusted. We argue that this time is fixed by causal propagation on the associated Jacobi chain. The relevant distance is not the Krylov index itself, but the inhomogeneous transport metric $\rho(m,n) = \sum_{j=\min(m,n)}^{\max(m,n)-1} 1/b_j$, where $b_j$ is the Lanczos hopping across the bond $j \leftrightarrow j+1$. We prove a Lieb–Robinson bound in this metric. Its small-weight limit gives the velocity $v_{\rm LR} = 2$, meaning that propagation is exponentially suppressed outside the cone $\rho(m,n) \simeq 2|t|$. The error of a finite Krylov approximation to the return amplitude is a round-trip effect: information has to travel from the probe to the truncation boundary and back. Combining the Lieb–Robinson bound with Duhamel's formula yields a lower bound on the error of the truncated dynamics. For a fixed tolerance $\epsilon$, let the break time $t_\ast(m;\epsilon)$ denote the longest time for which the $m$-dimensional truncation is guaranteed to reproduce the exact return amplitude within error $\epsilon$. We show that $t_\ast(m;\epsilon) \ge \tau_m[1-o(1)]$, where $\tau_m = \rho(0,m) = \sum_{j<m} 1/b_j$. When the probe spreads along the chain, this lower bound is also tight, so $t_\ast(m) \simeq \tau_m$. The situation is different when the probe excites only a localized part of the spectrum, or a part already resolved by the truncation. In this case, essentially no signal reaches the boundary. Beyond a state-dependent Krylov dimension $m_\ast$, the approximation can therefore remain accurate at all times, and the break time is effectively infinite. Numerical tests on spin chains and random Jacobi matrices support $t_\ast(m) \simeq \tau_m$ in the transport-limited regime.
Read the full abstract

All authors Shunji Matsuura, Yoji Kawamura, Joseph Salfi, Satoshi Iso

Differential Geometry

Strict Stability and Strict Minimality of Regular Area-Minimizing Hypercones: A Quantitative Characterization

Let $\mathbf{C}=\partial E\subset \mathbb{R}^{n+1}$ be a regular area-minimizing hypercone. We first prove that $\mathbf{C}$ is simultaneously strictly stable and strictly minimizing if and only if there exists $c_\mathbf{C}>0$ such that \[ \operatorname{Per}(F;B_R)-\operatorname{Per}(E;B_R) \geq c_\mathbf{C} \int_{F\mathbin\triangle E} \frac{\operatorname{dist}(x,\mathbf{C})}{|x|^2}\,dx \] for every $R>0$ and finite-perimeter set $F$ with $F\mathbin\triangle E\Subset B_R$. Thus this intrinsic distance-weighted inequality characterizes exactly the simultaneous strictness of the stability and minimizing properties. Second, without either strictness assumption, every regular area-minimizing hypercone satisfies the scale-invariant quadratic inequality \[ \frac{\operatorname{Per}(F;B_R)-\operatorname{Per}(E;B_R)}{R^n} \geq c_\mathbf{C} \left( \frac{|F\mathbin\triangle E|}{R^{n+1}} \right)^2. \] This extends the inequality previously established for area-minimizing Lawson cones. Finally, the area-minimizing assumption is unnecessary for our spectral result: for every stable regular minimal hypercone, the first Dirichlet eigenvalue $\lambda_\mathbf{C}^D(R)$ of the Jacobi operator on $\mathbf{C}\cap B_R$ is given exactly by \[ \lambda_\mathbf{C}^D(R) = \frac{j_{b_1,1}^2}{R^2}, \qquad b_1^2 = \frac{(n-2)^2}{4}+\mu_1, \] where $\mu_1$ is the first eigenvalue of the link Jacobi operator and $j_{b_1,1}$ is the first positive zero of the Bessel function $J_{b_1}$. In particular, this identifies the optimal Dirichlet spectral constant for every stable regular minimal hypercone.
Read the full abstract

Computation and Language

Making Clinical Language Models Auditable: Concept-Guided Fine-Tuning for Robust Prediction

Clinical language models can achieve strong in-hospital accuracy yet fail under deployment shifts because they exploit note-specific artifacts (e.g., templates, separators, boilerplate) that do not reflect patient state. We propose CAST (Concept-guided Artifact Suppression Tuning), an SAE-based framework for auditable clinical text classification. CAST uses Sparse Autoencoders to expose sparse, human-auditable features from intermediate Transformer activations, labels SAE latents with an LLM-assisted interpretation pipeline and ICD-10 retrieval constraints, suppresses verified artifact latents via residual subtraction during fine-tuning, and provides post-hoc per-concept attributions for auditing model decisions. On MIMIC-IV discharge-note mortality prediction, CAST improves over its corresponding fine-tuned encoder baselines and remains competitive with strong LLM baselines, while producing a feature-level audit trail of the clinical concepts that support each prediction and the artifact concepts suppressed during training.
Read the full abstract

Combinatorics

Sharp quadratic $\chi$-binding functions for powers of bipartite graphs

For every natural number $r\geq 2$, we construct $r^{th}$ powers of bipartite graphs whose chromatic number is quadratic in their clique number, showing that the straightforward quadratic upper bound is best possible. We thereby settle an open problem posed by Chakraborty, Chandran, Jacob and Pillai [J. Graph Theory 112(3) (2026), 235-254] by establishing the sharpness of the quadratic bound for squares of bipartite graphs.
Read the full abstract

Computer Vision

LeVJEPA: Efficient & Scalable Video Pretraining without the Heuristics

Video carries the temporal structure of the physical world, yet learning representations from it has remained computationally expensive: prevailing self-supervised methods either prevent representation collapse through architectural asymmetries, coupling an exponential-moving-average target encoder, a stop-gradient, and a capacity-limited predictor, or circumvent it by reconstructing masked content in pixel space. We introduce LeVJEPA, the first video encoder trained under LeJEPA's collapse-free objective, which dispenses with both. A single encoder is trained with an invariance loss over global and local views of a clip, regularized by SIGReg, which excludes collapse with a provable guarantee. The architecture reduces to an encoder and a projector, and the objective to a single hyperparameter. This formulation admits two properties. First, the cost of pretraining is governed by the number of tokens the encoder observes; uniform random token dropping renders this number small while simultaneously improving downstream accuracy. At matched epochs on identical data, LeVJEPA matches or surpasses V-JEPA 2 across ViT-S/B/L at 5.6 to 20.8x less pretraining compute, and at matched total FLOPs it exceeds the strongest video baseline by 7.6 points on ImageNet-1K while remaining competitive on motion-centric benchmarks. Second, since no asymmetry between branches is required, the encoder can be trained with block-causal attention at no measurable accuracy cost: temporal ordering becomes a property of the encoder itself. Against a compute-matched DINOv2 trained on frames of the same videos, LeVJEPA approaches the image-pretrained encoder on appearance-centric evaluation while nearly doubling its motion-centric accuracy. These results indicate that, once its computational overhead is removed, video becomes a viable and in several respects preferable substrate for general-purpose visual pretraining.
Read the full abstract

All authors Lukas Kuhn, Lucas Maes, Giuseppe Serra, Quentin Le Lidec, Yann LeCun, Randall Balestriero, Florian Buettner

Computation and Language

RATIO: A Benchmark for Retrieval Across Typed Ideation Operations in Scientific Literature

Retrieved scientific literature can serve as inspiration for both human and AI scientists. Inspiration can take different forms: prior work may directly suggest how to address a problem, or surface directions at different levels of abstraction—zooming out to a more general view or zooming in to a concrete realization. We introduce RATIO (Retrieval Across Typed Ideation Operations), a large-scale benchmark in which relevance is defined by three operations which we name ideation moves: Address retrieves potential approaches for stated problems, Broaden retrieves more general formulations, and Specify retrieves concrete instantiations. RATIO is constructed from millions of full-text scientific papers across CS literature via a general recipe that extends discourse-marker distant supervision—previously used only for classification—to corpus-scale retrieval, combined with extensive LLM and human vetting. Experiments show that operation-specific fine-tuning substantially boosts retrievers and LLM reranking helps further, but much room for improvement remains. Researchers blind to condition found the fine-tuned retriever's candidates more useful as inspirations. RATIO provides a scalable training and evaluation framework for retrieval components that support literature-grounded ideation, opening up new research avenues in scientific inspiration retrieval. We release full data, code, models, and annotations.
Read the full abstract

Systems and Control

Adversarially-Informed Node Criticality Identification in Power Grid Measurements

Power grid state estimation relies on sensor measurements that are increasingly vulnerable to adversarial corruption in cyberphysical environments, potentially leading to significant deviations in system observations. This motivates the need to identify critical measurement nodes whose compromise results in the most severe system-level impact. However, existing node criticality methods primarily rely on structural or steady-state analyses and do not explicitly account for adversarial effects on system behavior. To address this gap, this paper proposes an adversarially informed framework for identifying critical measurement nodes in linearized power systems. Within this framework, a structured attack generation mechanism is developed to construct stealthy and effective false data injection attacks (FDIAs) against an H-infinity resilient state estimator. Node criticality is then evaluated using coalition-based marginal contributions of compromised sensor subsets, estimated via permutation sampling over a prescribed set of admissible nodes, with the resulting importance scores mapped to the corresponding physical buses. Simulation results on the IEEE 14-bus system show that adversarially identified nodes induce larger deviations in frequency, voltage angle, and net power compared to randomly selected nodes, demonstrating the effectiveness of the proposed framework.
Read the full abstract

Signal Processing

Property-Specific Recoverability from Contact PPG to Camera rPPG under Heterogeneous Observation Conditions

Camera-derived remote photoplethysmography (rPPG) is commonly validated through endpoint accuracy, but endpoint performance does not establish whether other physiological properties of source contact photoplethysmography (PPG) remain preserved recording by recording. We evaluated property-specific PPG-to-rPPG recoverability on 655 recordings from the Multi-Domain Mobile Video Physiology Dataset using CHROM as a fixed camera-rPPG observation pathway. The pathway reproduced the published CHROM correlation regime, with heart-rate MAE of 15.26 bpm and Pearson correlation of 0.0801. Matched-versus-shuffled validation revealed modest recording-specific autocorrelation correspondence, while spectral and recurrence-rate measures showed little matched discrimination. Maximal Lyapunov exponents showed essentially no recording-specific PPG-to-rPPG correspondence, with correlation of 0.0231 and permutation p-value of 0.5584, despite population-level overlap. Endpoint discrepancy exhibited Fitzpatrick-associated heterogeneity after adjustment for lighting and motion, including a Fitzpatrick VI versus III contrast of 9.32 bpm, while dynamical discrepancy showed no corresponding gradient. Aggregate RGB signal-to-noise ratio did not materially account for the endpoint contrast. In subject-held-out analysis, adding motion and lighting consistently reduced MAE across linear, ridge, and random-forest learners relative to rPPG-HR-only calibration, with reductions up to 13.32 percent. These findings show that recoverability is property-specific: physiological properties differ in recording-specific preservation and dependence on observation conditions, and population-level plausibility does not establish preservation of individual recordings.
Read the full abstract

Artificial Intelligence

CorporateBench: Large-Scale Q&A Benchmarking with Temporal Knowledge Bases

LLMs are increasingly able to answer complex questions about enterprise-scale document collections. But evaluation is hard: companies don't want to share internal communications, and synthetic datasets have been overly simple. We present CorporateBench (CB), a human-validated multi-task Q&A benchmark whose scale approaches the conditions LLMs encounter in corporate communication networks, with evaluation corpora surpassing 230,000 documents. CB evaluates LLMs across two dimensions (information extraction and knowledge base querying) through four synthetically generated firms ranging from 12 to 10,000 employees. Each corpus is sampled from a temporally evolving knowledge base describing a consistent world, guaranteeing cross-document logical consistency even across hundreds of thousands of documents. We evaluate five LLMs on CB, revealing increasingly poor performance as input size approaches realistic scales. CB provides LLM developers a metric for corporate communication reasoning, filling a crucial gap in the benchmarking ecosystem.
Read the full abstract

All authors Sil Hamilton, Albert Yu Sun, Oscar J. Romero, Carl-Leander Henneking, David Mimno, Bishan Yang, Igor Labutov

Human-Computer Interaction

Beyond Harassment: Exploring the Harm Experienced by People with Disabilities in Social Virtual Reality

People with disabilities (PWD) are increasingly engaging in social virtual reality (VR) platforms, where immersive and embodied interactions can intensify negative experiences. While prior work has examined harassment in VR, little is known about the harms experienced by PWD and the perceived severity associated with different harassment and disability types. Unlike harassment, which represents behaviors, harm is more critical to designing effective protections, as it reflects the consequences and impact; the realism of VR and the vulnerability resulting from disability identity can further amplify such impact. To characterize and model harms for PWD, we conducted a literature review, followed by an online survey with 67 PWD to understand participants' harassment experiences and resulting harms in social VR. We identified 19 types of harm in 5 categories, and reported the severity perception of each type of harm. Finally, we analyzed our results from the critical disability theory perspective, summarized the uniqueness of harm in social VR, and discussed design implications for specialized safety mechanisms that mitigate harm for PWD.
Read the full abstract

All authors Xinran Adeline Li, Kexin Zhang, Yuhang Zhao, Yaxing Yao

Astrophysics of Galaxies

LightCube: A Parsec-Resolution 3D Model of the Local Far-Ultraviolet Interstellar Radiation Field

We present a three-dimensional model of the local interstellar radiation field (ISRF) in the ultraviolet (UV). Using UV flux measurements from the TD1 catalog and stellar distances from Gaia and Hipparcos, we construct a catalog of stars that we expect to dominate the UV flux in the nearby Galaxy. We use the radiative transfer code DIRTY to model the propagation of photons from these stars through a 3D dust map, including the effects of scattering and absorption. The result is LightCube, a model of the ultraviolet ISRF out to 1.25 kpc from the Sun, with a maximum linear resolution of 1 pc. We model the ISRF in the four TD1 bands (1565 A, 1965 A, 2365 A, and 2740 A), as well as a single value for the full FUV range, and calculate the ISRF at the Sun to be 5.65$\times$10$^{-14}$ erg cm$^{-3}$ from 912 A to 2000 A. The modeled ISRF is quite variable, with more than an order of magnitude variation seen in dense regions, and about half that in lower density regions. By comparing LightCube to an estimate of the 3D distribution of total-to-selective extinction ratio, $R_V$, we find a positive correlation between UV flux and $R_V$ in regions of low UV radiation.
Read the full abstract

All authors Hannah V. Bish, J.E.G. Peek, Catherine Zucker, Karl D. Gordon, Claire E. Murray, Susan E. Clark, Mathieu Willems, Erika Hamden