Combinatorics
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Tuesday 10 November
A graph $G$ is called a brick if it is $3$-connected and $G-\{u,v\}$ has a perfect matching for every pair of distinct vertices $u,v\in V(G)$. A brick is minimal if deleting any edge results in a graph that is no longer a brick. A graph is claw-free if it contains no induced $K_{1,3}$. Plummer proved that every $3$-connected claw-free graph of even order is a brick. In this paper, we completely characterize all claw-free minimal bricks. Apart from $K_4$ and the six-vertex wheel $W_5$, they are precisely the line graphs of graphs obtained from loopless $3$-edge-connected cubic multigraphs by subdividing every edge outside a matching of even size exactly once. Consequently, every claw-free minimal brick of order $n$ contains at least $4n/5$ cubic vertices and at least $4n/5$ edges joining cubic vertices. Both bounds are sharp. This establishes the conjectures of Norine and Thomas and of Lovász for claw-free minimal bricks in sharp quantitative forms. We also show that, for a claw-free minimal brick of order at least $8$, deleting an edge leaves a matching covered graph if and only if the edge lies in a triangle.
Mesoscale and Nanoscale Physics
Geometry-Selective Higher-order topological insulators induced by $d_{xy}$ altermagnet
We show that proximity to a $d_{xy}$-wave altermagnet provides an effective route to engineering boundary-geometry-dependent higher-order topological phases. The boundary geometry plays a crucial role in determining the topological properties: axis-aligned edge states remain gapless, whereas diagonal edges acquire effective masses with alternating signs, giving rise to four corner modes localized at the mass domain walls. By projecting onto the edge-state subspace, we derive the effective edge Hamiltonians for different boundary geometries. Furthermore, $\pi$-flux defects introduced under periodic boundary conditions yield zero-energy bound states localized at the flux cores, providing complementary evidence for nontrivial topology without physical boundaries. The interplay between the Zeeman field and the altermagnetic order drives a topological transition through the closing and reopening of the edge gap, while the bulk gap remains open. These findings highlight the crucial role of boundary geometry in higher-order topology and establish proximity-induced $d_{xy}$-wave altermagnetism as a route to engineering higher-order topological phases.
Read the full abstract
Cryptography and Security
5G CATS: 5G-Based Cryptographic Authorization Against Trafficking of Semiconductor Devices
Enforcing export controls on AI accelerators requires verifying where a chip operates after deployment. Round-trip time measurements can bound a chip's distance from a trusted landmark, but any jitter in honest replies forces the verifier to loosen its timeout, giving an attacker slack to relay traffic over fiber from a restricted region. Prior approaches either cross the untrusted host stack or rely on noisy power-line channels that inflate this slack. We present 5G CATS, a location certification system that pairs a co-packaged 5G radio, cryptographic engine, and hardware interlock on the GPU package with 5G Ultra-Reliable Low-Latency Communication (URLLC) base stations. Terminating the radio link inside the package removes the host software, PCIe fabric, and baseboard management controller from the verification path. We map an authenticated challenge-response protocol onto 5G's slot structure, replace permanent key erasure with short-lived execution leases, and specify a cross-check of the chip's reported serving cell against the carrier's own attachment records. In ns-3 and 5G-LENA simulations of an unloaded cell, our URLLC-based radio profile narrows the 99.9th percentile round-trip spread to 2.60 ms, more than ten times narrower than default 5G. A tracer rejects every round of a fiber relay from 300 km, against 2,000 km for default 5G. Against a 3,000 km attacker, a tracer stays safe at up to 1,410 km from the chip, against at most 619 km for PowerCATS's published model, while default 5G is never safe.
Read the full abstract
Accelerator Physics
Compact plasma-based extreme compression of relativistic electron beams
Accelerated particle beams are essential tools of modern science and technology with applications ranging from medicine [1], to light sources [2], to high-energy physics [3]. Over time, as accelerator-based experiments become more sophisticated and probe natural phenomena with greater temporal resolution, there is an increasing demand on beam parameters, in particular on bunch duration. Experiments have been proposed that require beams with 1-2 orders of magnitude shorter durations than what is possible with current state-of-the-art technology. However, state-of-the-art temporal bunch compressors force a trade-off between how short an electron bunch can be made and how well its transverse quality survives the strong bending magnets required to do so – a limitation that stands as a grand challenge for accelerator physics to-date [4]. In this experiment, we demonstrate a new, compact compression technique for generating ultrashort, high-peak-current electron bunches that circumvents this trade-off using plasma wakefield acceleration. Just as chirped pulse amplification unlocked ultrashort, high-power laser pulses and the technologies that followed from them, here we demonstrate an analogous chirp-and-compress strategy applied to relativistic electron beams by using the extreme electric fields of a plasma wakefield in place of a stretcher-amplifier, thereby unlocking the path to ultrashort, extremely high-current electron beams. We demonstrate the first application of this plasma-compressed bunch using it to drive single-microbunch free electron lasing at optical wavelengths in a single, meter-scale undulator module, producing radiation pulses roughly four orders of magnitude more energetic than the first beam-driven, plasma-based free electron laser demonstration at the same accelerator facility [5, 6].
Read the full abstract
Analysis of PDEs
A uniqueness theorem for Allen–Cahn global minimizers asymptotic to the Simons cone
For every integer $m\ge4$, we prove that the saddle-shaped solution of the Allen–Cahn equation in $\R^{2m}$ constructed by Cabré and Terra \cite{CabreTerra2009} is a global minimizer. Moreover, up to sign, it is the unique global minimizer whose nodal set at infinity is a normal graph over the Simons cone with a height that decays faster than that of the Hardt–Simon leaves.
Read the full abstract
Information Theory
Minimum Synthesis Time for Single-Deletion Correcting Codes of Prescribed Size
DNA is a promising medium for data storage, but the high cost of DNA synthesis remains a major obstacle to its widespread deployment. In the reference-based synthesis model, synthesis time is measured by the length of the reference sequence. Previous studies have analyzed code sizes under synthesis-time constraints and constructed cost-constrained codes, including codes that correct deletions. In this paper, we study the minimum synthesis time for codes of prescribed size, considering both unrestricted codes and single-deletion correcting codes. We establish finite-length lower bounds valid for arbitrary reference sequences and upper bounds using existing $q$-ary differential VT codes. For fixed $q\geq3$, $A>0$, and $R\in(0,1)$, we determine the minimum synthesis time up to an $O(1)$ term when the code size is $M_n=\lfloor Aq^{Rn}\rfloor$. The corresponding redundancy is $(1-R)n\log_2q-\log_2A+o(1)$ bits. In particular, for these code sizes, the minimum synthesis time for single-deletion correcting codes exceeds that for unrestricted codes by a term proportional to $\ln n$, and we determine its coefficient. We also derive upper and lower bounds for code sizes $\lfloor q^n/(nq^{s_n})\rfloor$, where $s_n\to\infty$ and $s_n=o(n)$. These sizes correspond to redundancy $\log_2 n+s_n\log_2q+o(1)$ bits. For code sizes $\lfloor q^n/(nq^c)\rfloor$ with fixed constant $c$, the redundancy is $\log_2 n+c\log_2q+o(1)$ bits. We give conditions under which single-deletion correcting codes exist or do not exist for all sufficiently large $n$, together with bounds on the minimum synthesis time. These results show how code size and the single-deletion correcting constraint affect synthesis time and provide theoretical guidance for the construction of cost-constrained codes.
Read the full abstract
Statistical Methodology
Regression Modeling for the Bloc Basu Bivariate Exponential Distribution: Marginal and Conditional Means
The bivariate absolutely continuous exponential distribution introduced by \citet{block1974continuous} provides a model for dependent lifetimes in reliability and survival analysis. Its marginal means depend jointly and nonlinearly on the rate parameters, which complicates regression modeling and the interpretation of covariate effects on these means. In this paper, we develop two regression approaches based on a region-wise multivariate exponential-family representation. In the full-distribution approach, link functions relate the cumulant derivatives in a fixed reference region to linear predictors. These derivatives do not directly equal the marginal means, which are instead probability-weighted combinations of the regional derivatives. This relationship allows covariate effects on the marginal means to be examined. Likelihood derivatives are obtained through the chain rule, and the regression coefficients are estimated by maximum likelihood. The rate parameter estimates are then recovered through the inverse link functions and an explicit inverse mapping, with the admissible solution selected to ensure positive rates. As an alternative, we propose regression models based on the normalized conditional distribution within each region. In this setting, the cumulant derivatives coincide with the conditional means, allowing covariates to be linked directly to these means. Both approaches are evaluated through a simulation study.
Read the full abstract
Probability
Bi-parameter local linearization and parameter estimation for the damped stochastic Klein–Gordon equation with spatially colored noise
Consider the one-dimensional nonlinear damped stochastic Klein–Gordon equation driven by Gaussian noise that is white in time and fractional in space, with Hurst parameter $H\in(1/2,1)$. For a globally Lipschitz diffusion coefficient, we approximate characteristic mixed increments by those of the additive equation, multiplied by the coefficient frozen at the base point. A critical-kernel reduction and localized Riesz covariance estimates control the additional kernel terms and spatial correlations. We also establish $L^p$ error bounds for normalized quadratic variations and, under a nondegeneracy condition, construct jointly strongly consistent estimators of the Hurst parameter and a positive diffusion parameter.
Read the full abstract
Machine Learning
You Don't Need Feature Learning to Do In-Context Learning
How should we understand memorization vs. generalization in in-context learning (ICL) tasks, and when does one happen vs. the other? To investigate, we study how rotation-invariant kernel machines learn the ICL linear regression task proposed by Raventos et al. (2023). By decomposing the task in the kernel eigenbasis, we obtain a spectral picture of learning in this ICL setting. This picture reveals that as sample size grows, kernel machines learn successively-higher-degree components of the memorizing solution, which depend upon successively-higher-degree statistics of the training task set. The generalizing solution is approximately a low-degree truncation of the memorizing solution, and so the model first generalizes, then degrades into memorization as training proceeds. The takeaways are threefold. First, if this case is indicative, learning a "generalizing" solution on an ICL task is really just underfitting the memorizing solution! Second, the fact that kernel machines can generalize on this task means that feature learning is inessential to ICL. Third, transformers empirically trained to do ICL are shown to memorize tasks through the higher-order statistics of the training tasks that the kernel-based theory predicts, suggesting it applies even to models capable of feature learning.
Read the full abstract
Audio and Speech Processing
Off-target posterior mass: A training-free diagnostic for fine-tuning spoken language diarization probes
Spoken Language Diarization (SLD) models provide frame-level language information of a speech recording. SLD modeling pipeline includes feature extraction from pre-trained speech models followed by clustering. Fine-tuning pre-trained speech models has shown performance gains for SLD with its magnitude being dataset dependent. Off-target Posterior Mass (OTPM), computed based on posterior language identification probabilities is proposed as a diagnostic metric to forecast performance gains from SLD fine-tuning. OTPM obtains positive correlation with probe gains, 0.56 on unseen synthetic bilingual data and 0.23 on real-world corpora. These results indicate promising potential of OTPM as a diagnostic metric for SLD.
Read the full abstract
High Energy Physics (Phenomenology)
Composite spectroscopy of a chiral $SU(4)$ model
We consider a QCD-inspired chiral $SU(4)$ model, with a new fundamental sector ($q_{\rm HC}$) charged under the confining hypercolor gauge group $SU(3)_{\rm HC}$ as a template for electroweak dynamics and compositeness. The $(\mathrm{SU}(4)_L\times\mathrm{SU}(4)_R)/\mathrm{SU}(4)_V$ breaking leads to a rich structure of spin-0 and spin-1 resonances which can be produced by the hadronization of $q_{\rm HC}$ in Drell-Yan processes at FCC-hh. Working in the true vacuum at a general misalignment angle $\theta$, the fermion currents are derived for the fifteen techni-rho ($\rho_{\rm HC}$) resonances, the techni-omega ($\omega_{\rm HC}$), and the techni-eta-prime ($\eta^{\prime}_{\rm HC}$) using the Hidden Local Symmetry formalism. The estimated branching fractions pave the way for studying the model spectroscopy. The analysis of the production by hadronization of $\rho_{\rm HC}$, $\omega_{\rm HC}$ and $\eta^{\prime}_{\rm HC}$ states is presented for a 100 TeV future hadronic collider, FCC-hh. The projected discovery of $\rho_{\rm HC}$ and $\eta^{\prime}_{\rm HC}$ is complementary to the direct production by Drell-Yan or Vector Boson Fusion, providing a smoking gun signature for Higgs compositeness.
Read the full abstract
Quantum Physics
Sample complexity of learning fermionic and bosonic Gaussian states with single-copy and entangled measurements
With single-copy measurements, every mixed fermionic Gaussian state on $m$ modes, pure modes included, is learned to trace distance $\epsilon$ from $\Theta(m^3/\epsilon^2)$ copies, with no logarithmic factor, against the known $\Theta(m^2/\epsilon^2)$ for collective measurements. We prove the upper bound, resolving Conjecture 10.1 of arXiv:2609.23189 (lower bound for $\epsilon\le1/3600$), and fill most open entries of the sample-complexity table of fermionic and bosonic Gaussian states, for single-copy, $k$-copy and collective measurements. With blocks of $k$ copies, the covariant highest-weight measurement of arXiv:2611.07271, read out through the Weyl-shift estimator, is exactly unbiased for every state; it gives $\Theta(\max\{m^3/k,m^2\}/\epsilon^2)$ in Frobenius norm on all states and in trace distance on every class at a fixed distance from the pure boundary. The gain $1/k$ ($k\le m$) away from the pure boundary, against $1/\sqrt k$ on cold number-conserving states, is therefore attained. Learning a quadratic fermionic Hamiltonian in Frobenius norm from its Gibbs state at bounded $\beta\|h\|$ has the same rate. For bosonic Gaussian states with bounded covariance, any displacement and squeezing, and occupations $n_1\le\dots\le n_m$, $O((m^2+r^3+\sum_{i,j>r}1/(n_i+n_j))/\epsilon^2)$ single-copy measurements suffice, $r$ the largest index with $r\,n_r\le2$; this is the minimax rate on dyadic-interval classes, with no spectral gap or cold-mode count assumed, and holds with no covariance bound for displaced passive states. With no bound on energy or squeezing, non-Gaussian measurements learn centred one-mode states, mixed included, from $O(\epsilon^{-2})$ and centred pure $m$-mode states from $O(m^2/\epsilon^2)$ single copies, free of the $\log\log E$ term of Gaussian measurements, and $m$-mode states from $O(m^3/\epsilon^2)$ copies, entangled only in a first round on blocks of $4m+6$.
Read the full abstract
Statistics Applications
Reconciling District Prediction Markets and Seat-Count Buckets in U.S. Midterm Elections
The prediction market platform Kalshi offers individual markets for the winner in each U.S. congressional district and an aggregate market for the total number of Democratic House seats. We develop an applied statistical framework for reconciling these two market layers, since the aggregate seat count is determined by the collection of individual district outcomes. The framework combines a latent-Gaussian dependence structure with a log-odds adjustment to district probabilities and a power adjustment to aggregate seat-count probabilities, with parameters estimated by divergence minimization. Applied to a July 30, 2026 snapshot, the three-parameter primary model achieves close reconciliation, with forward and reverse KL divergences both approximately 0.0013, while yielding a positive latent national correlation, a negative district log-odds shift, and flattening of the aggregate target distribution. Daily analysis of 236 dates from January through August 2026 reveals two relatively stable periods separated by a spring transition period exhibiting substantial volatility in both parameter and divergence values. The framework provides a diagnostic approach for assessing cross-market consistency in hierarchically structured prediction or financial markets.
Read the full abstract
Computation and Language
BioMem: Agentic Memory for Biomedical LLM Agents
Large language models (LLMs) have demonstrated considerable potential in biomedical applications, while agentic memory systems have emerged as a promising approach to preserving and utilizing information across tasks and interactions. However, a fundamental question remains: how effectively and reliably do these systems retain, retrieve, update, and utilize biomedical information? In this study, we systematically investigate agentic memory in the biomedical domain, focusing on task effectiveness, retrieval quality, information fidelity, and long-term robustness. We evaluate five memory approaches across four LLMs in three representative settings: biomedical knowledge reasoning, clinical prediction, and long-term conversational memory. Our analyses reveal how memory construction, evidence relevance, and conversational history shape downstream biomedical performance, highlighting the distinct challenges of retrieving relevant information, preserving its fidelity, and effectively utilizing it for reasoning. Building on these findings, we introduce Hybrid Memory Retrieval and Utilization (HMRU), a training-free framework that jointly enhances evidence accessibility and task-aware utilization to improve biomedical reasoning and long-term conversational memory. HMRU achieves an average relative improvement of 8.6% across biomedical question-answering and conversational memory benchmarks, with additional gains in selected clinical prediction tasks. It further reduces conversational answer errors by 24.9% and closes 48.0% of the performance gap to full-context inference. Our findings provide a systematic characterization of biomedical agentic memory and practical insights into developing more effective and reliable memory-augmented LLM systems.
Read the full abstract
Computer Vision
Prompt2Echo (P2Echo): Text-Prompted Multi-View Echocardiography Segmentation with One Model
Automated segmentation in multi-view echocardiography is essential for comprehensive cardiac assessment, yet remains challenging due to substantial anatomical and appearance variability across standard acoustic windows. In this paper, we present P2Echo, a unified framework capable of segmenting diverse cardiac structures across multiple standard echocardiographic views using a single model controlled by structured view-aware text prompts. Unlike recent foundation model adaptations that rely on implicit latent priors, P2Echo explicitly conditions the segmentation process via a novel Dynamic Image-Text Alignment (DyITA) module. DyITA integrates a differentiable routing mechanism and token differential operators to dynamically modulate decoder features according to the requested view and target structure. This allows the network to adaptively switch its processing pathways for different views without architectural changes. Extensive experiments on a large-scale internal dataset (seen during training) and held-out evaluation on three external datasets demonstrate that P2Echo matches or exceeds both view-specific specialists and state-of-the-art universal models at a fraction of their computational cost, offering an efficient, controllable solution for multi-plane cardiac analysis. Code is available at: https://github.com/moeinheidari7829/P2Echo.
Read the full abstract
Operator Algebras
Ostrik's Theorem for Operator Algebras in a Unitary Tensor Category
We define Hilbert modules for $\mathrm{C^*}$ and $\mathrm{W^*}$ algebra objects in the ind-completion of a unitary tensor category. We then prove an operator algebraic Ostrik's theorem in an ind-category: the 2-category of $\mathrm{C^*}$ or $\mathrm{W^*}$ algebras objects, their modules, and adjointable intertwiners is $\dagger$-equivalent to the 2-categories of $\mathrm{C^*}$ or $\mathrm{W^*}$ module categories.
Read the full abstract
Statistical Methodology
High-Dimensional Nonsparse Regression under Covariate Shift via Importance-Weighted Partial Least Squares
This paper studies high-dimensional nonsparse regression when the data exhibit both strong predictor dependence and covariate shift. Many existing regression methods struggle with highly correlated predictors, while covariate shift can make components learned from source data less effective for target-domain prediction. To address these joint challenges, we develop an importance-weighted partial least squares (IWPLS) method that reweights source observations to reflect the target covariate distribution and extracts response-relevant components for coefficient estimation. The resulting estimator depends on weighted sample covariances, which generally lack the usual Wishart structure even when the target-domain covariates are Gaussian. We therefore combine a perturbation analysis of weighted PLS moments with an influence-function decomposition to establish consistency for the one- and multi-component estimators and asymptotic normality for the one-component estimator. Simulation studies and real-data analyses show that IWPLS outperforms the benchmark methods considered.
Read the full abstract
Algebraic Geometry
Gromov-Witten Nonvanishing and Symplectic Deformations of Toric Varieties
Let $A$ be nonzero in the cohomology ring of a smooth projective toric variety $X$. In this note, we observe the existence of a nonzero Gromov-Witten invariant of the form $\langle A, A, \ldots\rangle_{0, n,\beta}^X$ as a consequence of Iritani's work on the big quantum cohomology ring of $X$. This implies Kollár's conjecture on the symplectic deformation invariance of rational connectedness in the case where $X$ is toric. If $X$ is moreover Fano and $B$ is also nonzero in the cohomology ring, we prove the existence of a nonzero Gromov-Witten invariant of the form $\langle A,B,*\rangle_{0,3,\beta}^X$ using Batyrev's presentation of the small quantum cohomology ring of $X$. We conclude with a few consequences of this nonvanishing result.
Read the full abstract
Optimization and Control
Boundary-controlled irreversible port–Hamiltonian modeling and structure-preserving discretization of tubular reactors
This paper presents a boundary-controlled irreversible port-Hamiltonian model and a structure-preserving spatial discretization for a non-isothermal tubular reactor with axial convection, variable-density momentum dynamics, chemical reaction, mass diffusion, heat conduction, and jacket heat transfer. Reaction, diffusion, and heat conduction are described by irreversible port-Hamiltonian channels, whereas convection and momentum are represented by a weighted Eulerian interconnection. The boundary ports recover the axial energy flux and the corresponding entropy flux. A partitioned finite-element method is then formulated in conservative coordinates using a power-preserving change of variables. The semi-discrete model satisfies the Hamiltonian balance and a discrete entropy balance with non-negative entropy production due to reaction, diffusion, heat conduction, and jacket heat transfer. Numerical results for an open tubular reactor are validated against independently computed continuous steady boundary-value solutions and show the transient temperature, composition, and velocity profiles in the very-low-Mach regime. The same discretization is also tested with reciprocal Soret–Dufour coupling, for which the discrete entropy production remains non-negative.
Read the full abstract
Dynamical Systems
Minimal dynamics with polynomial and intermediate complexity
We study how complicated the orbits of a minimal system can be when the topological entropy vanishes, by looking at the exact order of growth of the number of distinguishable orbit segments, and not only at its exponent. For a broad class of growth rates, including all polynomial ones, we construct minimal homeomorphisms of one-dimensional continua realizing them; in particular, two such systems may have the same polynomial entropy and yet incomparable growth. We also show that passing from a minimal subshift to the maps obtained by running its suspension flow for a fixed time only rescales time in the growth law, and we characterize exactly when this rescaling has no effect. Combined with a construction of Cassaigne, this yields minimal homeomorphisms of continua whose growth lies strictly between all polynomial and all exponential rates.
Read the full abstract
Numerical Analysis
Neumann Series of Bessel Functions for a Quadratic Sturm–Liouville Pencil
We construct Neumann series of Bessel functions representations for solutions and their first derivatives of the equation $-y^{\prime\prime}(x)+(q(x)+2\rho p(x))y(x)=\rho^2y(x)$. The series arise from Legendre expansions of two transmutation operator kernels. The obtained series converge uniformly on horizontal strips in the complex plane of the parameter $\rho$. We develop a recursive integration procedure for computing the coefficients of these representations. The knowledge of the first coefficients allows one to recover the potentials $p$ and $q$ which is useful for solving inverse problems. For separated boundary conditions, we estimate the characteristic functions whose zeros determine the eigenvalues. Two numerical examples illustrate the practical applicability of the obtained series representations to the solution of spectral problems.
Read the full abstract
High Energy Physics (Phenomenology)
A dual-moment phenomenological framework for gravitational,electromagnetic and strong interactions
The unification of fundamental interactions remains an outstanding open problem in fundamental physics. Most existing unification schemes rely on extra dimensions, exotic symmetries, or unobserved dynamical degrees of freedom. Here we construct a minimal classical phenomenological framework built on primitive charges and motion-induced kinetic dual moments. Gravitational, electromagnetic, and strong interactions are coherently formulated within a unified inverse-square-law structure. We postulate a universal coupling relation connecting source-charge couplings and moment-mediated couplings across these interaction sectors. Although weak-charge degrees of freedom can be formally accommodated within the formalism, full quantitative treatment of weak-interaction dynamics is left for future work. The model recovers well-established low-energy limits for planetary orbits, atomic systems, and heavy-quark hadrons. Importantly, it yields distinguishable corrections to light deflection by rapidly spinning compact objects, and three independent experimental schemes are proposed to test these predictions.
Read the full abstract
Combinatorics
The Isomorphism Problem for Cayley Graphs on Monogenic Semigroups
In this paper, we investigate the structural properties of Cayley graphs of monogenic semigroups and present a complete characterization of their weakly connected components. Several essential properties of these components are rigorously proven. Building upon these structural results, we introduce the notion of labelled colour-permuting isomorphisms between Cayley graphs. Using this new isomorphism framework, we address the Cayley graph isomorphism problem for monogenic semigroups, commonly known as the CI problem. In particular, we establish the necessary and sufficient conditions under which two Cayley graphs over monogenic semigroups are labelled colour-permuting isomorphic.
Read the full abstract
Dynamical Systems
Local Periodic Orbit Penetration in Dynamical Systems
We introduce two complementary frameworks for studying local periodicity and recurrence in dynamical systems. In the topological setting, we consider continuous maps with nonempty periodic-point sets, restricted to the closures of their periodic points. We define the periodic penetration time of an open set as the least period of a periodic point it contains. Using dynamically refined finite generating covers, we construct asymptotic mean and variance invariants that describe local periodic penetration relative to the logarithms of minimum subcover cardinalities. We establish their invariance under topological conjugacy and investigate their behavior for finite periodic systems, full shifts, shifts of finite type, and zero-entropy systems with dense periodic points. We also extend the construction to invertible systems using symmetric dynamical refinements. In the measure theoretic setting, we define local return times through positive measure intersections of measurable sets with their iterated preimages. Using dynamically refined finite generating partitions, we construct corresponding asymptotic mean and variance invariants and establish their invariance under measure-theoretic isomorphism. We derive explicit formulas for finite probability-preserving systems and study finite periodic approximations of uniform Bernoulli shifts on $q\geq 2$ symbols, proving that their system-level mean and variance invariants converge to $1/\log q$ and $0$, respectively. Our results demonstrate that local periodic penetration and return-time statistics provide information about dynamical systems beyond that captured by classical entropy.
Read the full abstract
Combinatorics
Hamilton cycles in vertex-transitive graphs of prime-power order
A graph is vertex-transitive if, for any two vertices, there is an automorphism mapping one to the other. We prove that every connected vertex-transitive graph of order $p^n$, where $p$ is a prime and $p^n\geq 3$, has a hamiltonian cycle. Our main tool extends Witte's theorem [JCTB (1986)] on Cayley digraphs of $p$-groups.
Read the full abstract
Machine Learning
Free Lunch Uncertainty: Sequence-Level Rényi Entropy from Particle Decoding for Adaptive LLM Reasoning
Quantifying uncertainty over an entire LLM completion is critical for selective prediction and adaptive compute. Zero-cost signals such as mean token entropy and log-probability are local to tokens or to one completion, and self-consistency needs many extra completions. We show that Sequential Monte Carlo (SMC) decoding for power sampling already estimates the normalizing constant of the power distribution, which gives the sequence-level R'enyi entropy (SRE) of the model's completion distribution at zero marginal cost, with a guarantee at every particle budget. A R'enyi chain rule extends this readout to a family of stepwise entropies of any order. We prove that orders above one are tied to the most likely token of each step, which explains why they track log-probability, while the order-1/2 member accumulates the log of per-step search effort. This member is the best zero-cost selective-prediction signal in 10 of 13 settings. On MATH500 its AUROC is 0.022 to 0.044 above log-probability with all four models and within length quartiles, and the same order is selected when each setting is held out. An 8-particle probe identifies the problems that benefit from more particles, and a two-pass policy matches 64-particle accuracy on MATH500 with 34.2% fewer particles over three seeds.
Read the full abstract
Distributed Computing
Placing LLM Lookup Tables by Per-Layer Prefetch Deadlines
DeepSeek-V4.1's 203 GB of Engram n-gram tables sit in host DRAM today, taking 190 of the 223 GiB a 4-GPU serving process uses. Row indices depend only on input tokens, so reads can start with each step and overlap earlier layers. V4.1's two Engram layers have windows that differ by 30x (0.3 vs. 9.3 ms). We show that this deadline, together with session reuse and contention on a shared device, decides which tier each table can use and which layers need cache, and we give a model that predicts the resulting slowdown from measured read times. Agent and chat traffic reuse rows within sessions, and static hot sets transfer poorly. We build SlackTier, a table tier for SGLang that issues all lookups at step start. It reads the short-window layer first and splits a small GPU row cache across layers. The path runs inside the decode CUDA graph and is bit-exact with the original gather. Layer 14's table can go to NVMe uncached with no measurable decode cost. With both tables on four NVMe disks, mean time per output token on agent and chat replays rises at most 1.4% over unmodified SGLang, and host memory falls from 212 to 34 GiB. Both tables need no cache when read over RDMA (remote direct memory access) from another node's memory; one copy serves two instances. On a KV-constrained agent workload, using the freed memory as host KV cache at the same memory budget raises median throughput by 36% and cuts median p90 time to first token from 38 to 24 s.
Read the full abstract
Distributed Computing
Bursts That Outlast Themselves: Persistent Degradation in Prefix-Cached Agent Serving
LLM agents drive inference engines with closed-loop sessions whose context grows every turn, and serving them economically depends on prefix caching: a hit costs an incremental prefill of about a thousand tokens, a miss recomputes a 10–40K-token context. We show that this makes prefix-cached agent serving metastable. With an identical base workload, a few minutes of extra arrivals leaves a vLLM server in a degraded regime–a ~10% hit ratio, about half the throughput, and a backlog that grows for the remaining 70 minutes–while its no-burst twin stays healthy (9 of 12 pairs; none in the reverse direction). The regime appears at loads the cache-warm server sustains, is absent without prefix caching, and also appears with Qwen3-32B on SWE-bench agent trace shapes and on SGLang, which tolerates larger triggers. Slower service keeps more sessions in the system, their working set outgrows the KV pool, and every turn becomes a recomputation; a first-order workload-balance condition, calibrated independently, brackets when the regime can drain. Getting out is the harder problem: throttling new admissions recovered 1 of 7 matched collapsed runs, while suspending admitted sessions against a capacity-based limit recovered all of them within minutes. A CPU offload tier delays the collapse but does not prevent it. A budgeter that bounds the resident KV demand of active sessions to the pool, with thresholds that need no per-workload tuning, prevented the failure in every run we tested.
Read the full abstract
Systems and Control
Residual-Corrected Physics-Informed Neural Network KKL Observers with Input-to-State Stability Guarantees
Physics-informed neural network (PINN) Kazantzis–Kravaris–Luenberger (KKL) observers require an approximate solution to a KKL partial differential equation (PDE) and an approximate inverse of this solution. Either approximation can leave a nonzero estimation bias even without measurement noise. In this paper, we remove both errors from the implemented observer by construction. We build an encoder containing an explicitly invertible neural map. Its closed-form inverse provides the internal state reconstruction used to evaluate the learned KKL residual and correct the latent dynamics. This construction makes the synchronization manifold exactly invariant for every value of the network weights. Based on contraction of the residual-corrected observer dynamics, we prove an input-to-state stability (ISS) result for the implemented error dynamics with respect to measurement noise. A direct consequence is that the estimation error converges exponentially to zero in the noise-free setting. We provide two numerical examples, the Van der Pol oscillator and the chaotic Lorenz system, in which the residual-corrected observers achieve exact noise-free convergence and improved error responses compared with released PINN–KKL baselines.
Read the full abstract
Artificial Intelligence
SteerDistill: Quantizing LLM Agents via Closed-Loop Teacher Steering
Low-bit quantization reduces the inference cost of LLM agents but can compromise task success. Examining failed rollouts, we find that quantized agents overlook task requirements or environmental conditions, producing environment transitions that diverge from successful full-precision references. Correcting these behavioral divergences remains difficult: offline data may miss the student's failure contexts, while its own rollouts may provide few correct responses in those contexts. We introduce SteerDistill, a quantization-aware distillation method in which a full-precision teacher guides the student to revise failed responses and iteratively refines its guidance using feedback from student execution. This closed-loop teacher steering produces locally verified behavioral corrections. We internalize these corrections by training the student without guidance to match the steered student's predictions at corrected turns and the full-precision teacher's at other turns, enabling unassisted inference. Across five agentic benchmarks and three models, SteerDistill exceeds the strongest quantized baseline by 1.67–2.18 points in average score under W4A16 and stays within one point of BF16. On $\tau^2$-Bench, it outperforms QeRL by 1.84 percentage points in success rate using 67% fewer training GPU hours and produces shorter successful trajectories than the quantized baselines.