Differential Geometry
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Tuesday 1 September
The classical Heisenberg Pauli Weyl (HPW) inequality exhibits a strong rigidity phenomenon on Riemannian manifolds i.e. on Cartan Hadamard manifolds and those with non-negative Ricci curvature, the validity of the Euclidean HPW inequality or the existence of extremizers strictly forces the manifold to be isometric to Euclidean space, $\mathbb{R}^n$. This geometric discrepancy motivates the study of curvature dependent corrections and their associated stability properties. In this article, we investigate the geometric stability of the HPW inequality. Specifically, given a sequence of pointed Riemannian manifolds and appropriately normalized functions with a vanishing HPW deficit, we address whether the sequence converges to the corresponding model space in the pointed Gromov Hausdorff topology. We prove that for pinched Cartan Hadamard manifolds with sectional curvature bounded above by $c < 0$, the manifolds converge to the model hyperbolic space $\mathbb{H}^n_c$. In the non-negative Ricci curvature setting, we establish convergence to $\mathbb{R}^n$, up to a metric rescaling factor governed by the sequence's moment term. Consequently, we deduce that the known HPW inequality formulated via the asymptotic volume ratio (AVR) is suboptimal for non negatively Ricci curved manifolds not isometric to Euclidean space. To resolve this, we introduce a curvature corrected HPW inequality for this setting, analogous to the Cartan Hadamard case. Finally, we establish quantitative rigidity estimates in both curvature regimes, demonstrating that the HPW deficit when evaluated at Gaussian profiles which explicitly controls an appropriately defined distance to the respective model space.
Algebraic Topology
Parametrized LS category and group actions
We introduce a new invariant associated to an action of a group $G$ on a (possibly non-commutative) group $K$. The invariant is based on the $G$-parametrized LS-category of the semi-direct product $K\rtimes G$ but it also depends on a choice of a derivation $d\colon G\to K$. The invariant is thus denoted $\operatorname{tc}(d)$ and it takes integer values between $\operatorname{cd}(K)$ and $\operatorname{cd}(K)+\operatorname{cd}(G)$ (where $\operatorname{cd}(G)$ stands for the cohomological dimension of a group). We develop a general theory and then compute its values for a number of important examples, including fundamental groups of mapping tori, of pure braid groups and of iterated semi-direct products of free groups. To achieve this, we relate $\operatorname{tc}(d)$ to the cohomological dimension of its factors, to the sectional category of the inclusion of $G$ in $K\rtimes G$, and to the properties of torsors that are associated to the derivation $d$.
Read the full abstract
Human-Computer Interaction
Better Situational Awareness in AR-HRC? A Comparative Study of Augmented Reality and Mobile Interfaces for Human-Robot Collaboration
Augmented reality (AR) facilitates human-robot collaboration (HRC) by enabling in-situ spatial visualizations of the robot and the joint task. However, in safety-critical HRC scenarios such as search-and-rescue, spatial visualizations may also reshape visual attention in ways that create competing situational awareness (SA) demands, potentially introducing new safety concerns. While prior AR-HRC work suggests potential benefits for SA, rigorous evaluations that jointly consider robot and environmental awareness across multiple levels of SA remain limited. We address this through a between-subjects study with 30 participants comparing custom AR and mobile interfaces presenting equivalent information, measuring robot and environmental SA with the Situation Awareness Global Assessment Technique (SAGAT) across all three levels, with concurrent eye tracking to identify the attentional mechanisms underlying any SA differences. Both interfaces achieved high usability; relative to the mobile baseline, AR improved perception-level awareness of the robot but yielded no gains in higher-level robot awareness or in environmental awareness at any level. Gaze analysis explained this: AR freed attention from the map, but that attention was re-invested in the conformal visuals rather than the physical environment. Freeing the eyes from a screen is not the same as directing them to the world, a distinction AR interfaces for safety-critical HRC must design around.
Read the full abstract
Functional Analysis
On cost-induced Santaló-type inequalities in Polish measure spaces
We introduce a framework for establishing Blaschke-Santaló-type inequalities on $m$-tuples of Polish measure spaces coupled together by a continuous cost function. Central to our approach is a transference principle, which provides a mechanism to lift geometric weighted inequalities involving cost-polar sets into functional integral inequalities of Santaló-type. We call these equivalent inequalities cost-Santaló inequalities. This definition expands and includes previous notions in the literature. We apply this principle to deduce several new versions of functional Santaló inequalities, including on the space of rectangular matrices and a functional sine Santaló inequality. A surprising development is that probability spaces with log-concave isoperimetric functions fit into our framework, for example, Gauss space and spherical space, leading to new functional Santaló inequalities in these settings. In particular, we obtain results for $\operatorname{RCD}(K,\infty)$ spaces. As a discrete application, we obtain an inequality for the Hamming cube. Finally, we explore applications to optimal transport, utilizing our functional framework to establish generalized transport-entropy inequalities on arbitrary Polish spaces satisfying a cost-Santaló inequality, which we explicitly instantiate for matrix spaces.
Read the full abstract
Chemical Physics
Reference-Density Hartree Screening for Gausslet Hamiltonians
Gausslets are among the few electronic-structure bases that permit the four-index electron–electron interaction to be replaced by an accurate two-index integral diagonal approximation (IDA). Near a many-electron nucleus, however, the nuclear attraction and core-electron Hartree field are individually large and substantially cancel. Treating the first as a full finite-basis matrix while treating the second with IDA leaves an avoidable imbalance. We introduce reference-density Hartree screening: the Hartree field of a chosen reference density is represented accurately, and IDA is applied only to density fluctuations about it. Tests on He, Ne, atomic F, F$_2$, and Cr$_2$ show large reductions in direct Hartree errors, including transfer of fitted neutral-atom fields to molecules. For Cr$_2$, atomic-core screening prevents the spurious HF collapse found with the unscreened $q=5$ and $q=7$ Hamiltonians, whereas finite-reference matching without screening does not. Screening leaves exchange and residual correlation unchanged. We therefore also introduce a low-rank one-particle correction that uses an accurate conventional Gaussian-basis Hartree–Fock calculation to match either occupied-space exchange information or the complete occupied Fock vectors. In F$_2$ and Cr$_2$, $X_{\rm HF}$ reproduces the finite-reference energy and occupied Fock vectors to numerical precision and the selected states return after orbital perturbations. For Cr$_2$, the corrected $q=5$ basis uses one quarter as many functions as the $q=7$ control while retaining sub-mHa mean-field accuracy. Screening provides the physical improvement to the direct field; the state-specific correction then restores the remaining accuracy of the Gaussian-basis mean-field reference.
Read the full abstract
Number Theory
Confluence relations for $q$-analogues of multiple zeta values
In this paper, we construct confluence relations for $q$-analogues of multiple zeta values by adapting the classical construction to the $q$-setting. Our construction uses $q$-analogues of multiple polylogarithms depending on an auxiliary variable $z$, functional relations arising from their $q$-differential equations, and the (regularized) limit as $z$ tends to $1$. We prove that the resulting family of relations contains the stuffle product relations and, assuming a certain conjectural identity, the duality relations for both the Bradley–Zhao and Schlesinger–Zudilin models.
Read the full abstract
Distributed Computing
Just Talk Once: Communication-Efficient Split Federated LLM Fine-Tuning on Edge Devices
Large language model (LLM) fine-tuning is increasingly shifting toward data generated on edge devices, where memory, computation, bandwidth, and connectivity constraints make conventional federated learning difficult to sustain. Split federated fine-tuning (SFT) improves client-side efficiency by offloading most model parameters and computation to the server but requires step-by-step bidirectional communication loop across the split interface and forces continuous client involvement throughout training. In this paper, we present L-shaped SFT, a split fine-tuning framework that removes this bidirectional bottleneck. Our key insight is that weight tying in modern LLMs enables server-side hidden activations to be directly supervised using target embeddings, allowing the training loss to be computed on the server without returning server outputs to the client. To further eliminate the need for continuous client participation, based on L-shaped SFT, we introduce one-shot SFT, in which clients upload activations once and then go offline while the server continues optimization over cached representations. We implement our design in a real system testbed with heterogeneous edge clients, including commercial smartphones and NVIDIA developer boards. Experiments demonstrate that our schemes significantly reduce communication costs and client online time compared with existing SFT baselines.
Read the full abstract
Information Retrieval
AutoConcept: Training-Free Concept-Guided Reranking for Metadata-Available Composed Image Retrieval
Composed image retrieval (CIR) retrieves a target image from a reference image and a text modification. This paper studies metadata-available CIR reranking, where a fixed CIR model first returns a candidate pool and gallery metadata is then used for second-stage concept-guided scoring. We introduce AutoConcept, a training-free reranker that converts concept evidence into an interpretable memory. AutoConcept filters noisy concepts, activates query-relevant positive constraints with an auxiliary negative penalty, and combines base retrieval scores with metadata-based concept-candidate alignment through inference-time calibration. On FashionIQ, AutoConcept yields significant early-rank improvements over WeiMoCIR and consistent plug-in gains on LinCIR candidate pools. Metadata-aware controls show that structured concept memory adds signal beyond direct query-text and extracted-attribute matching, while a query-only variant further supports the effectiveness of concept-level reranking. A supplementary real-human concept-label study indicates that the same memory interface can consume participant-provided evidence. These results position AutoConcept as an interpretable concept-memory reranker for product-style CIR galleries with available metadata.
Read the full abstract
Cryptography and Security
When Safety Routing Breaks: Understanding Alignment Fragility under Benign Fine-Tuning
Benign fine-tuning severely weakens the safety alignment of large language models (LLMs), so we study why refusal behavior is so fragile. While prior work often attributes this failure to gradient conflict, we propose a fundamentally different Fisher-geometric explanation: safety Fisher is low-rank, and alignment makes the safety geometry flatter while preserving an output-routing pathway. After 100 benign fine-tuning examples, this pathway is selectively re-sharpened in output-side MLP modules, explaining the asymmetric fragility: safety can collapse to high attack success rates, while general utility degrades mildly. The routing view also explains why few safety examples can restore refusal behavior, indicating that internal safety-relevant representations are preserved. Finally, we show that LoRA and ASAM mitigate early collapse by suppressing output-side sharpness, but their protection weakens at larger fine-tuning scales. Overall, safety failure is best understood as a disruption of a low-rank output-routing mechanism
Read the full abstract
High Energy Physics (Theory)
Lagrangian varieties from $q$-matrix models
We consider three specific $q$-deformed matrix models (the Chern-Simons, $q$-Laguerre, and $q$-Gaussian matrix models) which can be solved explicitly using the property of superintegrability. We show that one- and two-point functions of inverse characteristic polynomials can be interpreted as quantizations of Lagrangian subvarieties in $(\mathbb{C}^*)^2$ and $(\mathbb{C}^*)^4$, respectively. While such a geometric picture is expected for the Chern-Simons matrix model, our results for the $q$-Laguerre and $q$-Gaussian models are new and exhibit additional features. In particular, specific anti-symplectic birational involutions play an essential role in the construction. This reformulation provides a geometric framework for analyzing the semiclassical, large-$N$ expansion of matrix-model correlators. This picture is suggestive of the geometry underlying open topological strings, although the two constructions are not identical and the precise relation between them remains to be understood.
Read the full abstract
Robotics
Does Imitation Learning Preserve Temporal Robustness in Dexterous Manipulation? An Expert-Learner Comparison Across Task Execution Speeds
Dexterous manipulation policies learned by imitation are typically evaluated for robustness to variation in scenes, objects, or instructions, but their performance across task execution speeds is less often examined. This leaves open how much temporal robustness a learner retains relative to the expert it imitates. We compare an expert and learner under the same task conditions, initial-condition draws, and speedup factors. We instantiate the evaluation in ParcelStow, a contact-rich task in which the robot acquires, reorients, and inserts a parcel. The demonstrations span the speedup range for the manipulation phases after parcel acquisition. A scripted expert and an Action Chunking with Transformers (ACT) policy trained from the expert's demonstrations both achieve 100 percent task success at nominal speed. Their success rates diverge within the demonstrated range: at its maximum, expert success is 84 percent and ACT success is 53 percent. Two ACT policies with different parameter initializations show similar degradation, decreasing by 34 and 48 percentage points from nominal speed to the maximum demonstrated speed, compared with 16 points for the expert. Stage-level analysis shows that 35 of ACT's 47 failures at the maximum demonstrated speed are insertion misalignments. Under the relative-motion handoff, every ACT acquisition retains the parcel through reorientation and transfer in free space, but only 64 percent complete the overall task, compared with 95 percent after expert acquisition. Across all evaluated policies and speeds, none of the 414 acquisitions without force closure completes the task. Equal nominal task success therefore does not imply preservation of expert performance across execution speeds. Code, data, and evaluation scripts are available at https://github.com/coenwerem/parcelstow.
Read the full abstract
Instrumentation and Methods for Astrophysics
Project management of the Canadian Hydrogen Observatory and Radio-transient Detector
The Canadian Hydrogen Observatory and Radio-transient Detector (CHORD) is a new drift-scan radio interferometer operating from 300-1500 MHz, optimized for 21cm intensity mapping, 21cm galaxy detection, pulsars, and transient detection and localization. The full instrument will feature a core array of 512 six-meter dishes, plus two 64-dish outriggers at several thousand kilometer baselines. The project is nearing completion of the first 64-dish sub array, which will enable commissioning activities to begin in parallel with the remaining construction. In this presentation, we discuss the project management framework that has enabled us to navigate the logistical challenges of a fixed-budget, fixed-timeline research infrastructure project through a period of significant global economic and technological volatility.
Read the full abstract
Formal Languages and Automata
Behavioral Memory under Symmetry in One-Way Quantum Automata
Under compact symmetry, observable behavior reduces to an invariant operator algebra, but its dimension is not yet classical memory: some coordinates are dynamically frozen, some invisible to threshold tests, and some already classical. We develop an operator-algebraic theory that separates these effects through three filters. For one automaton, behavior is the Hilbert–Schmidt pairing between prefix-reachable states and suffix-observable effects, whose rank equals the real Hankel rank without controllability or observability assumptions. Maximizing this invariant over a symmetry-constrained dynamical class gives a structural capacity controlled by the symmetry commutant: its center stores isotypic populations frozen by reversible dynamics, its traceless multiplicity blocks carry movable noncommutative coordinates, dissipation removes the unary spectral loss inside those blocks, and covariant mobility releases relative populations subject to component conservation. Operational realization then determines which surviving coordinates force probabilistic states. For a fixed nontrivial invariant readout, full mobility gives an exact dichotomy in worst-case state cost: a commutative invariant algebra costs exactly its dimension, whereas a noncommutative multiplicity block raises the unrestricted cost by exactly one state. Thus noncommutativity has a one-state worst-case classical price. The known four-letter quadratic-plus-one law at trivial symmetry is the fully mobile endpoint of this principle. Schur–Weyl duality further shows that different preserved symmetries on the same tensor-power Hilbert space can change the worst memory scale from polynomial to exponential, while fixed-weight modules give an exact Catalan law at half filling, with structural capacity equal to the Catalan count minus its central-sector correction.
Read the full abstract
Dynamical Systems
On free minimal constant speedups violating continuous orbit equivalence in $p$-adic $\mathbb{Z}^d$-odometers of adding type
Let $\mathbb{Z}_p$ be the ring of $p$-adic integers with respect to a prime $p$ and let $d$ be a positive integer. For each $\mathbf{z}=(z_1, z_2,..., z_d) \in \mathbb{Z}_p^d$, let $T_{\mathbf{z}}: \mathbb{Z}^d \times \mathbb{Z}_p \to \mathbb{Z}_p$ be an adding-type $\mathbb{Z}^d$-action on $\mathbb{Z}_p$ defined by $T^{\mathbf{n}}_{\mathbf{z}}(x):=x+ \sum_{i=1}^d n_i z_i$ for $\mathbf{n}=(n_1,n_2, \cdots, n_d) \in \mathbb{Z}^d$ and $x \in \mathbb{Z}_p$. Under some mild assumptions on $\mathbf{z}$, the action $T_{\mathbf{z}}$ is a free $\mathbb{Z}^d$-odometer (by odometer, we mean a minimal and equicontinuous action on a Cantor space). In this paper, we derive a necessary condition for continuous orbit equivalence between such $\mathbb{Z}^d$-odometers by constructing algebraic models for them. We then study the free minimal constant speedups of these $\mathbb{Z}^d$-odometers. It turns out that such a speedup of $T_{\mathbf{z}}$ is again an adding-type $p$-adic $\mathbb{Z}^d$-odometer $T_{\mathbf{w}}$ for some $\mathbf{w} \in \mathbb{Z}_p^d$. However, the necessary condition above may not hold for the speedup. This provides the first known examples of free minimal bounded speedups (of free $\mathbb{Z}^d$-odometers) which are not continuously orbit equivalent to the original ones and hence disproves a conjecture by Johnson and McClendon. Our result also indicates that continuous orbit equivalence is a rare phenomenon for free minimal constant speedups of $p$-adic $\mathbb{Z}^d$-odometers of adding type when $d \geqslant 2$.
Read the full abstract
Machine Learning
Diffusion as a Training Curriculum for Timestep-Free Iterative Reasoning
Diffusion models and recursive reasoners are both iterative, but they carry information across iterations differently. We add a persistent hidden state to a diffusion denoiser and remove its timestep conditioning, leaving a single shared update that can be run to arbitrary depth. The result is an anytime solver: accuracy keeps improving with inference depth far beyond the rollout lengths and backpropagation window used in training, reaching 99.90% exact solve on Sudoku-Extreme. We also obtain 98.93% solve rate on Maze-Unique. Surprisingly, progressive denoising is unnecessary at inference: holding corruption at its maximum by replacing every non-clue variable with fresh Gaussian noise at each step retains near-perfect solving and converges to stable solutions. This simple noise-injection mechanism enables a single trajectory to efficiently explore the solution space and settle on the correct answer without parallel rollouts, candidate selection, or external verifiers required by prior reasoning models. Nonetheless, ordered annealed corruption remains critical during training, which suggests that diffusion's primary contribution to our anytime solver is not a sampling procedure at inference, but a denoising training curriculum.
Read the full abstract
Quantum Physics
Verifiable quantum advantage in extremely low depth
We give a sampling problem that is solvable by shallow quantum circuits, hard for polynomial-time classical algorithms under lattice-based assumptions, and efficiently verifiable by a classical computer. The quantum sampler admits two implementations: one uses log-logarithmic-depth quantum circuits with one- and two-qubit gates, i.e., $\mathsf{QNC}^0[\log\log]$ circuits, while the other uses constant-depth quantum circuits with unbounded fan-in gates, i.e., $\mathsf{QAC}^0$ circuits. Our construction can be seen as compiling the Learning with Errors (LWE)-based single-round proof of quantumness of Arabadjieva et al. (2025) to very low depth. The price paid for this compilation is the reliance on less standard, though well-motivated, assumptions: in addition to the lattice knowledge assumption used by Arabadjieva et al. (2025), we require a strengthened variant of the adaptive-hardcore-bit property of LWE, for which we provide supporting evidence. Unlike previous low-depth proofs of quantumness, the quantum computation here requires no mid-circuit measurements or feed-forward: it consists only of running a shallow circuit and sampling from its output distribution. This shows that shallow quantum circuits have sufficient structure to solve certain classically hard tasks whose solutions can be verified efficiently.
Read the full abstract
Functional Analysis
Kazhdan's Property $(T)$ for Subspaces and Quotients of $L_p$-Spaces
We prove a uniform spectral-gap estimate at every even exponent for isometric representations on arbitrary closed subspaces of real $L_p$-spaces. As an application, this result, combined with the work of Bader, Furman, Gelander and Monod, gives an affirmative answer to their Question 4.1: Kazhdan's property $(T)$ implies property $(T_X)$ whenever $X$ is a closed subspace or a quotient of a real $L_p$-space, for every $1<p<\infty$.
Read the full abstract
Materials Science
Polar nanoregions and reentrant-like ferroelectric behavior in SrTiO$_3$
Recent real-space imaging in quantum paraelectric SrTiO$_3$ [Nature 656, 54 (2026)] reveals that local polar textures do not continuously grow upon cooling, but reach a maximum intensity at intermediate temperatures around 60-65 K and weaken again toward the quantum paraelectric ground state. Such a reentrant-like weakening of local polar textures challenges the conventional paradigm in which ordering tendencies generally continue to strengthen as temperature decreases. Here we employ a self-consistent phase-field theory showing that this anomalous behavior naturally arises from the coupling between intrinsic polar and antiferrodistortiv (AFD) fluctuations. We demonstrate that the flexoelectric-like coupling strongly hybridizes the polar and AFD modes. As the uncoupled polar and AFD modes cross near 52 K, their hybridization is maximized, driving the lower hybridized branch to develop a minimum on a finite-wave-vector shell. This finite-$q$ softening triggers a Brazovskii-type instability, strongly enhancing polarization correlations and producing nanoscale polar textures. Away from the crossing temperature, the two modes become increasingly detuned, weakening their hybridization and the associated finite-$q$ softening. These results reaveal the origin of the formation of polar nanoregions in SrTiO$_3$, naturally explaining the unexpected confinement to an intermediate-temperature window and providing a mechanism beyond the quantum-fluctuation-based interpretation suggested by experiment. Furthermore, we predict an unconventional reentrant-like sequence in weakly strained SrTiO$_3$, evolving from ferroelectric to paraelectric, polar-nanoregion, and eventually paraelectric regimes upon heating from zero temperature.
Read the full abstract
Rings and Algebras
Global representations of algebras with a near-unanimity term
Global representations are subdirect representations satisfying a sheaf-like local-to-global patching principle. We develop a unified and simplified theory of such representations. For quasivarieties with a near-unanimity term, this framework recovers the main classical representation theorems through a common argument and yields a general representation by factors of bounded subdirect width. When the relatively subdirectly irreducible members form a universal class, we obtain an optimal result: every relatively congruence-distributive algebra admits a global representation by relatively globally indecomposable factors, which we characterize explicitly. We also provide a converse: global representations by factors of bounded relative subdirect width force the existence of a near-unanimity term. Adding semisimplicity to the hypotheses of the main theorem yields optimal representation results for filtral quasivarieties and dual discriminator varieties, together with a simple description of the globally indecomposable algebras. The technical engine behind these results is a new infinitary extension of the congruence-system component of the Baker–Pixley theorem.
Read the full abstract
Fluid Dynamics
Macroscopic simulations of thin film wetting/dewetting using a precursor film model
Numerical simulation of dynamic wetting processes remains challenging due to the multiscale feature of the moving contact line. This paper presents a finite element method for two- or three-dimensional dynamic wetting problems within the framework of the lubrication equation and the precursor film model. By reconstructing the disjoining pressure formulation, a mesoscopic precursor film model is implemented to enable a relatively thick precursor film to reproduce the intermediate-region interfacial behavior of a physically thin precursor film. The proposed model reduces the requirement for excessive spatial resolution near the contact line, thereby significantly reducing the computational cost while preserving the macroscopic flow dynamics. This method does not incorporate any moving boundary and easily handles topological changes. Its capability and accuracy are validated through various simulations, including the spreading, retraction, sliding, and coalescence of drops, as well as the breakup of liquid ridges. Numerical results (with the mesoscopic precursor film model adopted) show good agreement with the available exact solutions and asymptotic theories.
Read the full abstract
Mathematical Physics
$A_1$ - Coulomb branches over finite fields and Selberg Character sums
Motivated by mirror symmetry, we study exponential sums over the $\mathbb{F}_q$-points of the $A_1$ Coulomb branch. We show that their fiberwise structure is governed by polynomial Gauss sums, providing a geometric realization of Selberg character sums studied by Evans. Using Evans's evaluation, we obtain explicit formulas for the Coulomb-branch exponential sums as products of classical Gauss sums.
Read the full abstract
Differential Geometry
Finite Time Singularities of Collapsing Kähler Ricci Flow on Ruled Surfaces
We prove that any finite time collapsing Kähler Ricci flow on ruled surfaces develops a Type I singularity, such singularity is modeled on the standard product shrinker $\mathbb{P}^1\times \mathbb{C}$. As an application, we obtain the optimal collapse rate of fibers on ruled surfaces.
Read the full abstract
Machine Learning
Edge-Girth as a Structural Edge Feature for Graph Neural Networks
Graph neural networks (GNN) based on message passing are provably no more powerful than the one-dimensional Weisfeiler–Leman colour-refinement test (1-WL): two graphs it cannot tell apart receive identical representations, however deep or wide the network. A common remedy augments node or edge features with precomputed structural descriptors, most often counts of a fixed small subgraph such as triangles or longer cycles, but such counts require committing in advance to the size of the substructure counted, a choice usually made blind to the data. We study a descriptor that avoids this choice. The edge-girth of an edge is the length of a shortest cycle through it, and its multiplicity is the number of such shortest cycles; together they form a per-edge invariant that reports cycles of arbitrary length, computable exactly by a single breadth-first search per edge. Injected into a gated message-passing architecture, EGAGNN, it reaches a test MAE a factor three below the closest gated comparator on the ZINC-12k regression benchmark at 104k parameters; against bounded cycle-counting descriptors under the same architecture, it matches only a dictionary counting cycles up to length eight, using twice as many channels, while a dictionary capped at length four performs no better than no structural information at all. On graph discrimination we prove a matching limitation: on graphs where every edge sees the same number of shortest cycles of the same length, the descriptor becomes constant and any model built on it collapses back to the 1-WL bound. This holds without exception across all 400 pairs of the BREC benchmark: not one of the 90 such pairs is distinguished.
Read the full abstract
Representation Theory
Diagrammatic Okada monoid and cellularity of the Okada algebra
It is well known that the Young lattice is the Bratelli diagram of the symmetric groups, expressing how irreducible representations restrict from $\mathfrak{S}_{N}$ to $\mathfrak{S}_{N-1}$. In 1975, Stanley discovered a similar lattice called the Young-Fibonacci lattice which was identified as the Bratelli diagram of a family of algebras $\{\mathbf{O}_N(X,Y)\}_{N \geq 0}$ by Okada in 1994. In this paper, we first realize the Okada algebra $\mathbf{O}_N(X,Y)$ and the associated monoid $\mathbf{O}_N$ using a labelled version of non-crossing arc-diagrams appearing in the description of the Temperley-Lieb algebra and Jones monoid. We establish, for general parameters $(X,Y)$, that the dimension of the Okada algebra $\mathbf{O}_N(X,Y)$ is $N!$, noting that Okada proved this result only in the semisimple case. We interpret a natural bijection between permutations and labelled arc-diagrams as an incarnation of Fomin's version of the Robinson-Schensted correspondence associated to the Young-Fibonacci lattice. The arc-diagram formalism allow us to probe the structure of the Okada monoid and algebra. In particular we prove that the Okada monoid is a regular, aperiodic $*$-monoid and we describe its Green relations and order. These results allow us to construct a cellular basis of the Okada algebra and to show that $\{\mathbf{O}_N(X,Y)\}_{N \geq 0}$ forms a coherent tower of cellular algebras in the sense of Goodman and Graber. We present some conjectures expressing the Gram determinant of the invariant bilinear form attached to each cell module in terms of Okada's clone Schur functions. We conclude the paper by presenting two follow-up, ongoing projects along with a series of questions pushing further the analogy between the symmetric groups and the Okada algebras.
Read the full abstract
Combinatorics
Box-Delaunay graphs of large chromatic number
For every $n$, we construct a 2-dimensional $n$-element poset whose Hasse diagram has independence number $n\exp(-\Omega(\sqrt{\log n}))$, and consequently, chromatic number $\exp(\Omega(\sqrt{\log n}))$. This also yields an $n$-point planar set whose box-Delaunay graph satisfies the same bounds.
Read the full abstract
Human-Computer Interaction
Cross-Modal Guidance for Out-of-View Object Search in Simulated Prosthetic Vision
Out-of-view guidance is well established in virtual and augmented reality, but its effectiveness may depend on the visual bandwidth available to the user. We test this under simulated prosthetic vision (SPV), where visual guidance must share the same sparse representation used to inspect the scene. Nineteen participants performed object search under two SPV conditions differing in electrode density and phosphene spread (10x10 and 20x20) and four guidance conditions (no guidance, visual, haptic, audio) all driven by the same horizontal target-offset variable. All three modalities reduced search time and head movement. The tested auditory and haptic cues produced approximately 25% faster overall search and 11-13% faster target acquisition than the visual cue, despite similarly direct orienting trajectories. The tested haptic and auditory cues also shortened post-acquisition search. Final head-target angular offset was reduced substantially more in the 10x10 SPV condition; there, all three cues also reduced vertical localization error by approximately 45-58% despite providing no elevation information. Under severe visual constraints, guidance performance depended on cue implementation and search stage.
Read the full abstract
Software Engineering
HarnessDev: Can LLMs Create and Evolve Their Own Agent Harness?
As agents move from research prototypes to deployed tools, their capability increasingly depends on model-external execution infrastructure, commonly termed the agent harness. Changing this harness while holding model weights fixed can substantially alter task performance. Current agent evaluations typically report downstream performance under a chosen harness, leaving a model's ability to develop the harness itself comparatively underexplored. We introduce HarnessDev, a benchmark that shifts the unit of evaluation from task outputs to runnable infrastructure. HarnessDev covers two stages. In Creation, the agent starts from a minimal seed and a small number of cases, then builds a complete execution system. In Evolution, it starts from its own created harness and iteratively revises it using downstream execution feedback, with the goal of improving benchmark performance. We then evaluate each constructed harness on capability (task success on held-out benchmarks) and efficiency (execution-token cost). The reported Creation results cover six creator LLMs, four domains, and five downstream benchmarks totaling 2,207 unique downstream instances, with hidden evaluation tasks withheld from development. We find that generated harnesses remain substantially behind mature human-engineered references on code and on search and research, while matching or exceeding the selected references on writing and machine-learning experimentation, with large variation in execution cost. Evolution produces some performance gains, but they are unstable and transfer only partially to held-out tasks. Experiments with a fixed runtime model further show that the gains depend strongly on the model executing the harness, indicating limited transfer across models.
Read the full abstract
Analysis of PDEs
Bose–Einstein Condensation without an Initial Low-Energy Concentration Assumption
We prove that semi-strong convergence to a Bose–Einstein equilibrium below the critical temperature implies finite-time Bose–Einstein condensation (BEC) for conservative isotropic measure solutions of the spatially homogeneous quantum Boltzmann equation with low-momentum scattering degeneracy exponent $0\leq\eta<1$. For every admissible initial measure with $\overline T/\overline T_c<1$, there exists a conservative solution that has a positive zero-energy atom after a finite time without any assumption of initial concentration near zero energy. We also show that it persists and converges to the equilibrium condensate mass, while the full solution converges strongly. The proof introduces a method to obtain condensation from relaxation estimate to Bose–Einstein equilibrium. The relaxation estimate first gives a fixed amount of mass below a small energy level. We then use collision estimates at smaller and smaller energy scales and add their effects in a weighted sum functional with a fixed upper bound. If no atom forms at zero energy, enough mass remains in the positive-energy shells to force this sum to exceed its upper bound in finite time, which leads to condensation in finite time.
Read the full abstract
Statistical Methodology
Gaussian Processes on Directed Metric Graphs
We introduce a statistical framework for Gaussian fields indexed at arbitrary edge locations on general compact directed metric graphs. The construction is based on a stochastic differential equation with a first-order operator and conditions at the vertices. We characterise well-posedness and identify the covariance reproducing kernel Hilbert space. We also connect the proposed framework to earlier stream-network models, showing that these arise from the same system under particular boundary conditions, and introduce new boundary conditions that yield more physically realistic processes. The differential-equation representation enables computationally efficient inference and prediction. This makes the method applicable to large data sets without approximation. Applications to temperature modelling on river networks and traffic speeds on road networks illustrate the framework, including the computational efficiency and improved performance under physically informed vertex conditions.
Read the full abstract
Numerical Analysis
Score-Based Generative Data Assimilation for Integrating Aggregated Surveillance Data into Agent-Based Models in Epidemic Tracking
Reliable epidemic monitoring often requires inferring regional infection burden and transmission heterogeneity from noisy, spatially aggregated, and potentially sparse surveillance data. Agent-based models (ABMs) are attractive for this task because they represent individual behavior, contact heterogeneity, and localized interventions, but these same features make them difficult to calibrate online. We develop a generative AI-based data-assimilation (GenDA) framework for partially observed epidemic ABMs that estimates both the epidemic state and a heterogeneous parameter field while respecting the gap between observable macrostates and latent agent-level microstates. GenDA combines a training-free, score-based generative update for macrostate correction with a direct parameter update based on macrostate discrepancies, followed by a macro-micro reassignment step that restores consistency with the ABM. In controlled and geographically explicit synthetic experiments, the framework recovers regional epidemic burden, dominant hotspot structures, and effective transmission heterogeneity from aggregated observations, while improving post-assimilation forecasts relative to state-only assimilation.