In briefFTN Signaling: How Spectral Efficiency Varies from BPSK to 16-QAM
SummaryThis paper studies faster-than-Nyquist (FTN) signaling, which sends symbols closer together than the Nyquist limit to raise spectral efficiency at the cost of added inter-symbol interference. It analyzes achievable information rates for BPSK, QPSK, and 16-QAM using an MMSE channel-shortening detector under a power constraint. Lower-order constellations tolerate more aggressive acceleration, while 16-QAM performs best closer to the Nyquist limit, with optimal factors of about 0.65, 0.75, and 0.90 at 6 dB SNR.
In briefA Tighter Capacity Upper Bound for the Primitive Diamond Relay Channel
SummaryThis paper studies the capacity of the primitive diamond relay channel, a network with one source, two relays, and one destination. Using the Csiszár–Körner–Marton sum identity and Gallager-type identification, it derives an upper bound that improves on existing ones. For Gaussian channels, the bound matches a previously known strengthened Gaussian upper bound. For discrete memoryless channels, it can be strictly tighter than both the cut-set bound and the existing bound, and the improvement is assessed on several channel examples.
In briefThresholds for local coordinate-wise linear properties of random AG codes
SummaryThis paper studies local coordinate-wise linear (LCL) properties of random one-point algebraic geometry (AG) codes. It proves that under explicit alphabet and sampling conditions, these codes have the same LCL threshold rates as random linear codes. The authors also identify a flaw in a previously claimed above-threshold proof for random Reed-Solomon codes and provide a different argument. The results yield new list decoding, list recovery, and proximity gap guarantees for random one-point AG codes.
In briefExact second-order asymptotics for the discrete Wyner–Ziv problem
SummaryThis paper revisits the discrete Wyner–Ziv problem and establishes exact second-order asymptotics. It derives a second-order converse bound using posterior decomposition, concentration inequalities, and a conditional Gaussian approximation for martingales. It also extends an earlier achievability result by Li and Li by letting the test channel depend on the type of the source sequence. Numerical examples show the earlier bound is not always optimal, with the new achievability result reducing the second-order coding rate by 10.77% at excess-distortion probability 0.1 and by 21.35% at 0.2 for a quaternary source.
In briefParameter estimation when the estimator also gets a quantized version of the parameter
SummaryThe paper studies a parameter estimation problem in which the estimator observes independent, identically distributed samples and also a quantized version of the parameter. The number of quantization levels is limited, but the quantization can otherwise be optimized. The authors give upper and lower bounds on the minimax risk, which decay at similar rates when both the number of samples and the number of quantization levels are large.
In briefTwo Communicating Robots Catching a Moving Target on Star Graphs
SummaryThis paper studies searching for a moving target on an m-ray star graph, a generalization of linear search to several directions. The target is placed adversarially on one ray and moves at constant speed. Two robots cooperate under either a face-to-face communication model or an asymmetric sender-receiver model. The authors give upper bounds on the competitive ratio under three knowledge assumptions and derive lower bounds in some cases.
In briefCovering radius of rank-metric codes via covering lifts and clubs
SummaryThe paper introduces a generator-matrix-based geometric method for studying the covering radius of linear rank-metric codes over finite fields. Using covering lifts derived from the code's q-system, the covering radius is expressed through hyperplane weights, giving an exact criterion for the maximal value and an upper bound otherwise. For one-dimensional codes, the authors determine the covering radius for several boundary minimum distances and for all such codes with extension degree m in {3, 4, 5}.
Giuseppe Marino, Alessandro Neri, Alessia Vallefuoco
In briefNon-linear maximum sum-rank distance codes built from switched curves
SummaryThe paper builds a family of scalar-closed, non-additive maximum sum-rank distance (MSRD) codes. It switches norm components of a σ-rational normal curve and lifts the point set through a linearized Reed–Solomon syndrome map. Partitions satisfying Condition (◇) yield codes with minimum sum-rank distance δ, and these are non-additive when a curve component is kept.
Daniele Bartoli, Giovanni Giuseppe Grimaldi, Giovanni Longobardi
In briefNosRacer: Dynamic Detection of Race Conditions in On-Device Network Operating Systems
SummaryThis paper presents NosRacer, a dynamic analysis framework that detects race conditions in industrial on-device network operating systems, where asynchronous message ordering can cause configuration errors. It uses a two-phase design: a concentration phase that narrows analysis to relevant components and state variables, and a perturbation phase that alters message asynchrony to expose race-prone interleavings. The framework was implemented in a commercial NOS and tested on five key control-plane components.
In briefFederated zeroth-order optimization using direction aggregation and variance reduction
SummaryThis paper studies constrained nonsmooth nonconvex stochastic optimization in federated settings, where clients can only access stochastic function evaluations. It proposes a framework based on direction aggregation with two local sampling schemes: FedZOO builds local directions from a minibatch shared across multiple spherical queries, while FedVRZO uses independent sample-direction pairs. Both perform a single projected server update using a weighted aggregate of client directions, and the authors provide convergence guarantees for each. Experiments on black-box adversarial attacks demonstrate the methods' effectiveness.
In briefNew contraction bounds for multidimensional consensus in dynamic networks
SummaryThis paper studies multidimensional asymptotic agreement in dynamic networks, where nodes must converge to each other's inputs while staying in their convex hull. It gives a new upper bound on the contraction rate in the non-split network model, obtained by adapting the BallMidpoint algorithm. The paper also proves a multidimensional lower bound and shows that BallMidpoint is asymptotically optimal for a single round.
Zeremonienmeister Berber Aab, Darya Melnyk, Stefan Schmid
In briefWZIP and WLZ4: LZ77 compressors whose window size depends on match length
SummaryThe paper presents two LZ77-based compression formats, WLZ4, a byte-aligned LZ4-class format, and WZIP, an entropy-coded Zstandard-class format. Each match length gets its own sliding window, and the decoder infers the allowed offset range from the decoded length. On the Silesia corpus, WLZ4 reaches a compression ratio up to 16% above LZ4HC but decodes more slowly. WZIP exceeds the compression ratio of Zstandard 22, with its trusted decoder running at about 91% of Zstandard's speed.
In briefDistributed resource management for 6G networks using a scalable hybrid learning framework
SummaryThis paper addresses radio resource management in 6G wireless networks, where dynamic demands and interference make centralized optimization impractical. It proposes a two-phase hybrid framework that first allocates minimum resources to meet users' QoS needs based on channel statistics, then uses a multi-agent system to distribute remaining resources for throughput maximization. Simulations with 50MHz bandwidth report up to 40% higher throughput and 100% constraint satisfaction compared with existing baselines.
In briefOutage-aware robust sensing for drone tracking with ISAC
SummaryThis paper studies power allocation for tracking unmanned aerial vehicles using integrated sensing and communications across multiple base stations, where dynamic blockage makes available sensing links uncertain. The authors propose a framework that predicts next-slot link availability with a history-aware predictor, keeps plausible patterns through adaptive conformal prediction sets, and enforces tracking constraints only on non-outage patterns. In paired closed-loop simulations over five random seeds, the method reports a 0.01% serviceable-slot constraint-violation rate and 47.0% less power than protecting all non-outage patterns.
In briefDelegating Pricing to Algorithms: Why Slower Learning Can Win
SummaryThis paper studies how firms choose learning rates for pricing algorithms in a continuous-time pricing game. The authors model a meta-game where payoffs are the discounted sum of profits along the learning path. They find that under strategic substitutability firms prefer faster learning, while under strategic complementarity overly fast algorithms speed up rival responses and reduce transitional profits. The paper proves that firms design sluggish algorithms in equilibrium, even though convergence to the competitive price is guaranteed.
Galit Ashkenazi-Golan, Edward Plumb, Clemens Possnig et al.
In briefAn entropy-based proof of Ruzsa and Plünnecke sumset inequalities in abelian groups
SummaryThis paper gives a direct information-theoretic proof of the Plünnecke–Ruzsa inequalities for finite subsets of abelian groups. The argument uses an elementary coupling inequality obtained by maximizing joint entropy, avoiding tensor powers and asymptotic constructions. It also derives Ruzsa's sumset triangle inequality and extends the result to non-abelian groups.
In briefThree Improved Lower Bounds for Binary Error-Correcting Codes
SummaryThis paper gives explicit binary codes with parameters (n,M,d) = (22,84,9), (24,196,9), and (25,65,11), where n is code length, M is the number of codewords, and d is minimum distance. These establish lower bounds A_2(22,9) ≥ 84, A_2(24,9) ≥ 196, and A_2(25,11) ≥ 65, improving the public table values of 80, 192, and 64. The authors release all codewords and an exhaustive verifier for independent checking.
WhyExplicit constructions improve three long-standing binary code lower bounds, verified exhaustively, giving concrete value to coding theory.
In briefDRL-Based AoI Minimization for RSMA in Finite-Blocklength MU-MISO Networks
SummaryThis paper studies minimizing age of information (AoI) in multi-user wireless networks operating with finite blocklengths, which suits short state-update packets. The authors propose an actor-critic deep reinforcement learning framework to learn dynamic resource allocation policies for rate-splitting multiple access in MU-MISO broadcast channels. Simulations show the method achieves lower AoI than state-of-the-art benchmarks, with larger gains at low SNR and short blocklengths. At high SNR it matches benchmark performance while requiring only a single neural-network forward pass during execution.
Enes Kaya, Mehmet Alp Demircio\u{g}lu, Elif Tugce Ceran et al.
In briefLLM Agents Automating Checkpoint/Restart for Scientific Applications
SummaryThis paper studies whether frontier LLM coding agents can automatically add checkpoint/restart resilience to HPC scientific applications. The authors built a benchmark of 16 MPI applications and ran a no-human-in-the-loop generate-validate-revise pipeline, which produced 41 working resilient implementations. Agent-driven resilience engineering worked well when critical state was visible or accessible through coherent abstractions, but modularized and fragmented state remained a major limitation.
In briefZepp: Faster distributed MoE serving with relaxed balance constraints
SummaryThis paper presents Zepp, a system that speeds up serving Mixture-of-Experts (MoE) models across many devices. Instead of treating load balancing as the main optimization goal, Zepp treats balance as a constraint and directly reduces the inter-node communication bottleneck. It optimizes expert placement, token routing, and execution in stages. Evaluated against 7 state-of-the-art MoE serving systems, Zepp achieves up to 6.68x MoE layer speedup and a 1.86x geometric mean speedup over the fastest baseline.
In briefA framework of information-based drivers of intelligence and their interactions, modeled on n-body dynamics
SummaryThe paper examines how neural systems respond to different patterns of information, using concepts from informatics and algorithmic thermodynamics. It introduces "information enthalpy" as internal structured information available for prediction, and defines an information enthalpy potential (IEP) metric to quantify a signal's capacity to increase it. The authors also propose a testable, n-body-inspired model describing how these drivers move a neural system through a high-dimensional state space.
Brett J. Kagan, Johnson Zhou, Forough Habibollahi et al.
In briefIAPRepair: Proactive Repair for Erasure-Coded Storage Using In-Network Aggregation
SummaryErasure-coded storage tolerates failures with less overhead than full replication, but repairing lost or at-risk blocks needs heavy cross-node data transfer. This paper presents IAPRepair, which jointly plans repair batches, assigns reconstruction and replacement nodes by normalized transmission load, and selectively enables in-network aggregation for blocks that would overload healthy nodes. Implemented on a Tofino programmable switch with 16 storage nodes and evaluated on a prototype and large-scale simulations, it reduces repair time by at least 47.83% compared with the evaluated state-of-the-art methods.
In briefRevenue-maximizing auctions for multiple items with identically distributed values
SummaryThis paper studies auctions that sell several items and aim to maximize revenue while keeping truthful bidding optimal and buyers no worse off. Buyers value items additively, and each item's value is drawn independently from the same finite distribution. For two-point distributions, the authors give an explicit deterministic optimal mechanism and a polynomial-time algorithm for the optimal revenue. For general distributions, computing the exact optimal revenue is shown to be #P-hard.
WhyGives explicit optimal mechanisms for two-point distributions and proves #P-hardness for general finite supports, with rigorous theory and a PTAS.
In briefA general framework explaining crystal formation in maximal hard-core particle models
SummaryThe paper studies general maximal hard-core models on periodic lattice-type graphs. The authors introduce a unified set of assumptions, based on volume allocation, that imply the Peierls condition and a convergent cluster expansion, so that the conclusions of Pirogov-Sinai theory apply. They verify these assumptions on several examples and prove crystallization at both high and low activity, and also estimate activity values bounding phase coexistence regions.
WhyProposes unified volume-allocation assumptions yielding crystallization even at low activity, with rigorous proofs covering several periodic lattice examples.
In briefOptimal Public Project Payment Redistribution for Real-Valued Types via Complex Analysis
SummaryThe paper studies VCG payment redistribution for a public project, aiming to return as much payment as possible without a deficit. Earlier results reached efficiency near 1 only under a bounded-precision assumption on types. This work removes that assumption and builds a deterministic, anonymous, strategy-proof, non-deficit mechanism for real types in [0,1] with worst-case efficiency ratio 1-O(1/log log n). The proof relies on complex analysis.
WhyRemoves a bounded-precision assumption, reaching 1-O(1/log log n) with a novel complex-analytic proof; impact confined to mechanism design theory.
In briefQoS-aware joint design of user concurrency and HARQ limit for HARQ-CC-aided slow fluid antenna multiple access
SummaryThis paper studies downlink slow fluid antenna multiple access (sFAMA) that uses HARQ with Chase combining to improve reliability. It proposes a joint design of the number of concurrent users and the HARQ transmission limit, using a two-stage optimization that first maximizes feasible concurrency and then picks the smallest feasible limit meeting outage and mean-service constraints. Numerical results match Monte Carlo simulations, and the adaptive design supports six concurrent users, more than fixed-HARQ FAS (three to five) and HARQ-adaptive fixed-position antennas (two).
In briefA G-invariant Erdős–Hanani theorem and its use for optical orthogonal and constant-weight codes
SummaryThe paper studies packings of t-subsets by k-subsets that are invariant under a permutation group G acting on v points. Under two conditions on G, it shows that as v grows there exist free G-invariant packings with asymptotically the optimal number of blocks, binom(v,t)/binom(k,t), giving a G-invariant version of the Erdős–Hanani theorem. The results are applied to optical orthogonal codes, their multidimensional and signature pattern versions, and cyclic and quasi-cyclic constant-weight codes, where the Johnson bound is asymptotically attained. The method also determines the asymptotic maximum size of two-dimensional optical orthogonal codes with at most one pulse per wavelength.
WhyExtends Erdős–Hanani to group-invariant packings with full proofs, determining asymptotic maximum sizes of optical orthogonal and constant-weight codes.
In briefDistributed Satellite Antenna Modeling: Space Is Closer Than You Think
SummaryThis paper studies distributed satellites that cooperate to form a large synthetic antenna aperture, providing broadband connectivity to commercial handsets. It combines spherical-wave and plane-wave models with Earth and orbital curvature to quantify how simplifying assumptions affect throughput and outage probability. Numerical results show that local plane-wave approximations within each satellite array closely match the spherical-wave reference, while a global plane-wave approximation can cause severe losses, including full outage.
Giacomo Bacci, Riccardo De Gaudenzi, Marco Luise et al.
In briefLow-complexity equalization for Zak-OTFS using a Neumann series
SummaryThe paper proposes a general method for choosing an orthonormal carrier waveform basis aligned with a wireless channel's delay and Doppler characteristics, enabling low-complexity equalization. It illustrates the approach on satellite links and ground-to-UAV links, and reports robustness to fractional delays and Doppler shifts. When channel taps lie within the chosen subgroup, a single-tap equalizer suffices for reliable communication.
Vineetha Yogesh, Saif Khan Mohammed, Sandesh Rao Mattu et al.
In briefWhy estimating a chain's average mixing time can be impossible
SummaryThe paper estimates how many samples from one stationary trajectory are needed to learn a Markov chain's average mixing time. It proves nearly matching converse bounds showing when such estimation is impossible. These bounds match the key dependencies of existing achievability bounds up to logarithmic factors. Together they give nearly tight PAC guarantees for the estimation task.
In briefA constant-factor approximation to multidimensional consumer utility, meaning a simple, provably near-optimal way to maximize how much consumers gain.
SummaryThe paper studies mechanisms that maximize consumer utility when consumers pay with non-transferable ordeals, with multiple unit-demand buyers and heterogeneous items. The authors show simple mechanisms achieve a constant-factor approximation to optimal utility, with 5.67+ε for general independent values and 2e/(e-1) for i.i.d. values. The mechanisms are BIC, ex-post individually rational, and computable in polynomial time.
WhyResolves whether simple mechanisms achieve constant-factor utility approximation, improving prior logarithmic gaps via a general upper bound on optimal constrained utility.
In briefLOCAA: An Agent System That Automatically Tunes Lossy Compressors
SummaryLarge scientific simulations produce huge data volumes, so lossy compression is used to cut storage and data-movement costs, but users set it through error bounds whose effects are hard to predict, often requiring slow exhaustive search. The paper presents LOCAA, a large-language-model-based agent that searches in the loop with the compressor using tool execution, compressor-aware guidance, and persistent memory, supporting user-defined objectives and constraints. Evaluated across three scientific applications and five compressors, it needs fewer trials than binary search and FRaZ for fixed-ratio tuning, and cuts average trials from 61.5 to 17 under joint quality constraints.
Khondoker Mirazul Mumenin, Dong Dai, Sheng Di et al.
In briefPolarization-based beamforming for covert MIMO wireless transmission
SummaryThis paper studies covert MIMO communication using polarization-reconfigurable antennas, where the transmitter and receiver each perform polarforming. The authors jointly optimize the transmit covariance matrix and phase shifts to maximize the receiver's achievable rate under covertness and power constraints. They derive the detector's optimal threshold and minimum detection error probability for finite observation length and propose an alternating optimization algorithm. Numerical results show the rate gain of up to 53.8% over benchmark schemes.
In briefRisk-minimizing GUE implementations of truthful reporting in dynamic social choice
SummaryThis paper studies the Guaranteed Utility Mechanism (GUM), originally designed to achieve efficiency in dynamic social choice problems. The authors show it has a previously unknown link to risk minimization. With an efficient decision policy fixed, they identify variance-minimizing, budget-balanced transfer rules among GUE implementations of truthful reporting.
In briefSUSpMV: A High-Frequency Sparse Matrix-Vector Multiplier on HBM FPGAs, Written in SUS
SummarySUSpMV is a sparse matrix-vector multiplication accelerator written in the SUS hardware description language and run on an Alveo U280 FPGA. It uses all 32 HBM channels at 400 MHz to stream matrix data to 32 compute units, with input and output vectors kept in DDR memory. The paper reports a 79% geometric mean improvement over prior work on the same platform, reaching a peak of 144.9 GFLOPs.
Lennart Van Hirtum, David Volz, Andreas Koch et al.
In briefExact sample complexity of Rényi entropy estimation for non-integer orders
SummaryThis paper studies how many samples are needed to estimate the Rényi entropy of a non-integer order α>0 for an unknown discrete distribution over an alphabet of size d, within additive error ε. The authors give a tight Θ bound, completing the complexity picture that previously covered only integer orders. The upper bound uses a refined polynomial approximation estimator to reduce bias for large probabilities, while the lower bound relies on a new hard instance built by a moment matching construction.
WhyTightly characterizes noninteger-order Rényi entropy estimation sample complexity with matching bounds and a new moment-matching construction.
In briefHeuristics and machine learning for analyzing faults in distributed self-stabilizing programs
SummaryThe paper studies consistency violating faults, where a process reads stale shared-memory data, in self-stabilizing distributed programs. It proposes a conflicting-edges-based state-space exploration method for (Δ+1)-Graph Coloring and a machine learning model for a Maximal Independent Set program. The goal is to characterize when and where such faults occur to support runtime measures that improve efficiency.
In briefSettling PROPm and PROPavg in Graphical Resource Allocation: PROPm Always Exists, PROPavg May Not
SummaryThis paper studies proportional fairness in graphical resource allocation, where agents are vertices and indivisible items are edges that must go to one of their two endpoints. The authors prove that PROPm orientations always exist for goods on multigraphs and can be computed efficiently, extending to chores and mixtures. In contrast, PROPavg orientations may not exist even for simple graphs with binary valuations, and deciding their existence is NP-complete. The paper also shows the price of PROPm for goods is exactly 2 and gives hardness results for welfare optimization under PROPm.
WhyResolves open existence questions for PROPm and PROPavg, with efficient construction and NP-completeness proofs; theoretical results are rigorous.
In briefAntenna coding with pixel antennas to add a new degree of freedom for wireless communications
SummaryConventional wireless antennas have fixed radiation patterns, and adding more antennas raises cost and size limits. This article reviews antenna coding using pixel antennas, which adjust radiation patterns by controlling variable lumped elements between adjacent pixels to gain new degrees of freedom in the beamspace domain. It introduces the concept and classification of antenna coding and highlights gains in channel gain and multiplexing gain. It also surveys optimization approaches for antenna coding design and outlines remaining challenges and opportunities.
In briefTighter bounds on thresholds for sensitive and Blackwell optimality in stochastic games
SummaryThis paper studies Blackwell and sensitive optimality in perfect-information two-player stochastic games, which generalize mean-payoff and discount optimality to capture more farsighted preferences. It gives bounds on the Blackwell threshold and the d-sensitive thresholds, the smallest discount factors above which discount optimal policies match Blackwell or d-sensitive optimal policies. By focusing on reduced families of polynomials, the bounds improve prior work and are tight in controlling degrees and heights of minimal polynomials. Using root separation based on Mahler's and Cauchy's bounds, the authors derive the strongest known upper and lower bounds on the Blackwell threshold and are the first to bound d-sensitive thresholds in the multichain setting.
Xavier Allamigeon, St\'ephane Gaubert, Julien Grand-Cl\'ement et al.
In briefMacKay-Neal Codes Reach Channel Capacity Under MAP Decoding
SummaryThis paper proves a prediction from statistical-mechanics analyses that MacKay-Neal codes with fixed degrees can achieve channel capacity. For every binary-input memoryless symmetric channel with capacity above 3/ℓ, the average block error probability under MAP decoding is O(log N/N), and the actual rate converges to 3/ℓ. The proof relies on an entropy inequality for six even-parity bits.
WhyProves a statistical-mechanics prediction that MacKay-Neal codes achieve capacity under MAP decoding, with explicit error bounds; a notable theoretical result mainly relevant to coding theory.
In briefAsymptotic Subsidies for Envy-Free Allocation: how much money is needed so indivisible goods can be split with no one envious
SummaryThis paper studies the minimum monetary subsidy needed to allocate m_n indivisible goods among n agents so that the result is exactly envy-free. Valuations are additive, item values are i.i.d., and the analysis covers the asymptotic regime as n grows. The authors give a cubic-time algorithm that always returns an envy-free outcome and derive asymptotic bounds on the subsidy under a random model. For the square case q=1, they report that the minimum subsidy is of order Θ_p(log n).
In briefq-ary GRAND (Guessing Random Additive Noise Decoding) for error-correcting codes
SummaryThis paper develops a q-ary version of GRAND, a decoder for linear codes over alphabets with q symbols. It sorts symbol likelihoods and uses three local rules to generate symbol deviation patterns. The authors prove these rules form a monotone spanning tree, so best-first search achieves maximum-likelihood decoding when search is unlimited. Reed-Solomon simulations confirm agreement with ML decoding and show the performance-complexity tradeoff under finite search budgets.
In briefPrivate DNA data retrieval using random sampling across multiple colluding servers
SummaryThis paper studies sequential random sampling private information retrieval for DNA-based data storage, where the database is split across several equal-sized servers that may collude. It analyzes the tradeoff between download cost, measured as the expected number of queries, and privacy leakage, measured by mutual information. Lower bounds are derived, including one given by an optimization problem, and this bound is shown to be tight when each server stores two files. For servers of arbitrary size, schemes are constructed that apply a single-server scheme to a randomly chosen subset of servers.
