In briefComputational Complexity of Exact and Approximate Nash Equilibria in Anonymous Games
SummaryThe paper proves that, for any fixed rational c>0, computing a 2^{-n^c}-approximate Nash equilibrium in an n-player two-strategy anonymous game is PPAD-complete, even under restricted game structures. It simulates circuit constraints approximately using coefficient estimates. For exact computation, a three-strategy construction uses polynomial roots to implement circuit gates exactly. This establishes FIXP-completeness for exact equilibrium computation and FIXP_a-completeness for approximation in distance to an exact equilibrium.
WhyProves PPAD-completeness for exponentially precise approximate equilibria and FIXP-completeness for exact computation, a significant advance with rigorous proofs.
In briefInformation-Theoretic Limits of Spiked Matrix Models with Rotationally Invariant Noise
SummaryThis paper studies the information-theoretic limits of inference in rectangular rank-one spiked matrix models with additive rotationally invariant noise, relaxing the usual assumption of independent noise entries. Using the replica method, the authors derive a variational formula for the mutual information and saddle point equations predicting the minimum mean squared error. In the Bayes-optimal setting, these equations match the fixed point equations of state evolution for orthogonal approximate message passing, and numerical and finite-size simulations confirm the predictions.
In briefGenerative deep learning for CSI feedback in massive MIMO via channel distribution prediction
SummaryThis paper addresses channel state information (CSI) feedback in massive MIMO systems, where existing history-based deep-learning methods can become unreliable when user motion changes suddenly. The authors propose a prediction-aided coding framework that predicts the conditional distribution of next-frame CSI tokens from historical tokens and then uses variable-length entropy coding with that distribution. Experiments reported in the paper show about 72% to 84% lower feedback overhead than history-aware baselines, with higher reconstruction accuracy and maintained accuracy under random-walk trajectories.
In briefFaster parallel algorithms for approximate shortest paths in large graphs
SummaryThis paper presents randomized algorithms for approximate shortest path computation in weighted undirected graphs under the Massively Parallel Computation (MPC) model. The authors build O(1)-approximate single-source shortest path oracles that answer distance queries in O(1) rounds. Construction takes O(log log n) rounds in the near-linear memory regime and O(log² log n) rounds in the heterogeneous regime, improving on prior work by Dory and Matar.
In briefRIS Integration in Satellite Systems Based on the ITU-R VSAT Radiation Pattern
SummaryThis paper proposes a simple engineering method to optimize the position and orientation of a Reconfigurable Intelligent Surface (RIS), using the antenna radiation patterns and system topology. It aims to ease double fading on the reflected link without active elements by exploiting the RIS Fresnel zone. Reported results show gains of up to 9 dB at 2 GHz (S band) and up to 14 dB at 4 GHz (C band) for a 1 m VSAT. The paper also refines the RIS channel equation to include tropospheric attenuation and wave polarization.
In briefStageBreaker: Scheduling video RL post-training by data dependencies instead of stages
SummaryThis paper presents StageBreaker, a fine-grained scheduling framework for reinforcement learning post-training of video models. Rather than following coarse stages, it schedules the actual dataflow of each RL step, overlapping policy training with reward evaluation and merging two trajectory replays into one. It also balances GPU load at a finer granularity. The evaluation reports a speedup of up to 1.41 times over stage-centric execution.
In briefMeasuring Dissimilarity Between Ordered Probability Distributions
SummaryThis paper studies measures of dissimilarity and similarity on the majorization simplex, the set of probability vectors sorted in non-increasing order. It examines a Rényi-entropy based distance proposed by Yadav and Shkel, showing it is a metric for α in [1, ∞) but generally not otherwise, and gives necessary and sufficient conditions for equality in its triangle inequality when α > 1. It also derives bounds and equality conditions for total variation distance and Hellinger affinity, noting that a Hellinger-affinity-based upper bound on total variation can be smaller than those from Pinsker's and Bretagnolle–Huber inequalities.
In briefVoter Model with Opinion Graph: An Opinion Dynamics Model with Bounded Confidence
SummaryThis paper studies the voter model with opinion graph, a generalization of the standard voter model that adds bounded confidence. Agents adopt another agent's opinion only when the two opinions are neighbors in an undirected opinion graph. The authors analyze the ordinary and productive EdgeVOG processes, bounding stabilization time, opinion survival probabilities, and the maximum number of agents any single opinion can acquire.
Petra Berenbrink, Thorsten G\"otte, Martin Hoefer et al.
In briefAnalysis of array factor properties for reconfigurable intelligent surfaces in radio communications
SummaryThis paper examines the array factor of reconfigurable intelligent surfaces (RIS) used in antenna design, emphasizing how phase shifts shape the reflected path. It also studies spatial positions that require identical phase shifts, conditions for nullifying the array factor, the total number of maxima, the amplitude of the second lobe, and the calculation of the 3dB bandwidth. These results are presented as useful for RIS-assisted radio communication systems.
In briefCovering radius and deep holes of hyperderivative Reed-Solomon codes under the NRT metric
SummaryThis paper studies the covering radius and deep holes of Hyperderivative Reed-Solomon (HRS) codes under the Niederreiter-Rosenbloom-Tsfasman (NRT) metric, a generalization of the Hamming-metric setting. The authors determine the NRT covering radius of HRS codes and characterize their deep holes using MDS extensions and polynomial remainder conditions. They also give three classes of explicit deep-hole constructions, propose a Cheng-Murray type conjecture, and derive exact enumeration formulas in several special cases.
In briefPRB-WM: A Physics-Regularized World Model for Predicting Link Adaptation on Real Underwater Acoustic Channels
SummaryThis paper addresses the problem that underwater acoustic links receive stale channel feedback because sound travels slowly. It proposes PRB-WM, a world model that learns channel evolution with stochastic latent dynamics, estimates bit-error distributions for each modulation and coding scheme, and uses a chance-constrained controller to maximize spectral efficiency under a bounded BER-violation probability. In closed-loop replay of six measured sea-trial links, its mean goodput was 87% above the strongest baseline.}
Shengjie Zhong, Zhongliang Zhao, Jingxuan Chen et al.
In briefDeterministic distributed (Δ+1)-coloring of graphs in O(√Δ+log* n) rounds
SummaryThe paper presents a deterministic distributed algorithm that (Δ+1)-colors an n-node graph of maximum degree Δ in O(√Δ+log* n) rounds in both the LOCAL and CONGEST models. It combines Barenboim's polynomial coloring with arbdefective decompositions, degree reduction, and a short family of hash functions. An explicit variant based on lossless expanders avoids the exhaustive search needed to compute the hash family.
WhyImproves the known deterministic round bound for distributed (Δ+1)-coloring from O(√Δ log Δ) to O(√Δ) with proofs and an explicit variant.
In briefPrivacy-preserving state estimators when attackers know prior information
SummaryThis paper studies privacy-preserving state estimation for discrete-time linear stochastic systems when an adversary has prior knowledge, measuring privacy with pointwise maximal leakage (PML). It shows that the steady-state Kalman filter leaks the most PML privacy among the considered stable linear estimators when no release noise is added, and derives a lower bound on steady-state error under (ε, δ)-PML privacy. Sufficient release-noise conditions are given, and a convex semidefinite program jointly designs the estimator gain and noise covariance. Simulations on a source-load power system illustrate the privacy-utility tradeoff.
In briefSingle-Round Multi-Function Approximated Coded Computing with Error Bounds
SummaryThis paper studies coded computing, which reduces the impact of stragglers in distributed computation. It addresses single-round evaluation of multiple arbitrary, including non-polynomial, functions in the analog domain using Berrut approximated coded computing. The authors extend an existing multi-function approach and derive an error bound, then propose a Berrut-basis-based construction with its own error bound. Numerical comparisons indicate the proposed construction achieves lower relative errors than the existing approach in the tested settings.
In briefCALIPER: A method for calibrating HPC/AI network simulators
SummaryThis paper presents CALIPER, a method for systematically calibrating and validating simulators of HPC/AI networks. It separates directly observable system parameters from auxiliary parameters set through calibration, constraining the latter to ranges consistent with the target system. Evaluated with collective communication benchmarks on a reference system of up to 64 nodes at the Jülich Supercomputing Centre, the calibrated configurations predicted performance with less than 5% mean error across scale-up and scale-out networks.
Seyyidahmed Lahmer, Daniel Seibel, Srijeet Basu et al.
In briefComplete ω-powers of one-counter languages in the difference hierarchy
SummaryThe paper studies finitary languages accepted by one-counter automata whose ω-powers are complete for classes of the difference hierarchy. For any natural numbers n and k, it shows the existence of such a language whose ω-power is D_{2n-1}(Π^0_k)-complete. It also gives a conditional result for the dual classes D^-_{2n}(Π^0_k) and a conditional result about Wadge degrees of rational Δ^0_2 ω-powers.
In briefSpace AI Infrastructure: Linking Edge Computing and Orbital Cloud
SummaryThis paper proposes a roadmap for Space AI infrastructure (SpaceAI-Infra), which extends AI capabilities from onboard satellite payloads to a global infrastructure. It describes three stages of development: onboard AI, Space Edge AI, and Space Cloud AI. The proposed architecture combines communication-focused free-flying LEO constellations with computing-focused formation-flying satellites arranged as space data centers in dawn-dusk sun-synchronous orbits. A case study indicates the architecture can support data-center-scale foundation-model inference for millions of concurrent users worldwide.
In briefSP-Allreduce: Symmetric Tree Schedules for Torus and Other Networks
SummaryThis paper introduces SP-allreduce, a communication algorithm for allreduce operations on d-dimensional torus networks, which are common in high-performance computing. It exploits network symmetry and uses only shortest paths, and the authors prove it needs the minimum number of rounds among fully bandwidth-optimal allreduce algorithms on tori. The approach is also extended to other symmetric topologies. Experiments report speedups of up to 1.19x on square tori, 1.65x on rectangular tori, 14.82x on HyperX networks, and 1.72x on spine-leaf networks over the evaluated baselines.
WhyProves minimum rounds among fully bandwidth-optimal allreduce on tori, halving Bucket's count, with speedups shown across several topologies.
Mikhail Cherniavskii, Andrey Kupavskii, Georgy Mironov et al.
In briefMeasuring Deliverable Compute: A Dynamic Metering System for Distributed Networks
SummaryThis paper proposes D-PIN Standardized Compute Units (D-PIN SCU), a system for measuring the compute that heterogeneous distributed networks can actually deliver, rather than nominal chip specifications. It combines resource normalization, topology-aware and runtime-corrected metering, and a calibration algorithm. The framework was recomputed end to end on one production node in Zhangjiang Science Gate, covering ten 8-GPU servers totaling 59.5 D-PIN SCU. The authors note that validation is limited to this single node and that calibrated chip values differ materially from datasheet projections.
In briefDecoding polar codes under strong phase noise using log-density spectrum recursion
SummaryThis paper proposes a decoding method for polar codes over channels with strong phase noise. Unlike traditional successive cancellation (SC) decoding, it recurses over probability density functions of the phase-noise variable, represented through their log-density spectra. The method is called log-density spectrum recursion (LDSR). Numerical results under strong phase noise and AWGN show that LDSR achieves clearly better error performance than traditional SC and Kalman-based methods.
In briefMore Interactions, More Efficient Communication: A Directed Information Bottleneck for Task-Oriented Communication
SummaryThis paper proposes VDIB-JSCC, a variational directed information bottleneck joint source-channel coding framework that transmits only information the receiver has not yet acquired, reducing redundancy with previously sent content. The authors use directed information to characterize the causal flow of task-relevant incremental information. Experiments on CIFAR-10 and CIFAR-100 video streams report higher classification accuracy with less transmitted information than state-of-the-art semantic communication methods.
In briefWeight structure of twisted Reed-Solomon codes and their subcode supports
SummaryThis paper determines the generalized Hamming weight hierarchies of two classes of twisted Reed-Solomon codes. For single-twist codes, it gives all possible weight values and characterizes the hierarchy via the dual code's minimum distance. For double-twist codes, the complete hierarchy is determined. It also derives explicit formulas counting subcodes with prescribed support sizes for single-twist codes.
In briefHeteroVDiff: Sharing Weights to Run Video Diffusion Across a Phone's NPU and GPU
SummaryThis paper presents HeteroVDiff, a system that runs a multi-expert video diffusion model on both the NPU and GPU of a mobile phone. It stores base weights once in INT8 for both engines and splits the denoising process at one block boundary, with the GPU running the first part and the NPU the rest. On a Galaxy S25 it reportedly achieves 1.35x the stream throughput of NPU-only execution and uses 48% less weight memory than separate copies, with VBench-2.0 total scores matching FP32.
In briefRPA Decoding Using Few Subspace Projections to Correct Structured Errors in RM Codes
SummaryThis paper studies the Recursive Projection Aggregation (RPA) decoder for Reed-Muller codes when the number of errors is less than half the minimum distance. It shows that for structured error patterns, choosing projection subspaces carefully can reduce the number needed at each recursion level. For a single burst of length below half the minimum distance, three subspaces per level are sufficient for correction.
In briefLearning Rules and Memorizing Facts: An Information-Theoretic View of Hallucination
SummaryThis paper examines how generalization and hallucination relate to memorization, using an information-theoretic Selector-Density Conditional Mutual Information framework. It decomposes model-data mutual information into rule learning, atomic-fact memory, and excess memorization, showing competition between rule learning and factual memory. The authors derive a joint budget bound on atomic hallucination and use an additive Gaussian model and controlled relational experiments to explore how training exposure, rule complexity, and model capacity affect these trade-offs.
In briefA New Lens on Multi-GPU Communication: API-Level Memory Models for Portable Optimization
SummaryThis paper argues that device-side communication libraries such as rocSHMEM, MSCCL++, and NCCL/RCCL GIN lack rigorous memory models, which makes optimized multi-GPU communication brittle. The authors propose API-level memory models for these device APIs and define one for rocSHMEM, AMD's recently open-sourced library. They report that relaxing the model to support cache bypassing for network data improves performance by up to 2.48x on a collective workload, 1.77x on a fused compute-plus-communication workload, and 1.18x on an expert-parallel communication workload.
Johnathan Alsop, Bradford Beckmann, Brandon Potter et al.
In briefBinary LCD codes built from double circulant codes over F2+uF2
SummaryThis paper studies double circulant linear complementary dual (LCD) codes over the ring F2+uF2, where u²=0. The authors derive an explicit enumeration formula using the Chinese Remainder Theorem and the factorization of x^n-1, treating the x-1 component separately. By combining Gray images with a known extension construction, they obtain a binary LCD code with parameters [29,15,7] and two inequivalent binary LCD codes with parameters [45,23,9].
In briefSubstep Linearizability: Relaxing Atomicity While Keeping Sequential Specs
SummaryThis paper proposes substep linearizability, a relaxation of linearizability, the standard correctness condition for concurrent objects, to handle operations that cannot be specified atomically, such as atomic exchangers. The authors note that interval linearizability, the current state of the art, requires interval-sequential specifications that are hard to generate and understand. They show substep linearizability is equivalent to interval linearizability and evaluate it on an atomic exchanger, a non-linearizable queue, and snapshot isolation between database transactions.
Finn Hackett, Evan Wrench, Ivan Beschastnikh et al.
In briefFourier entropy bounds for threshold functions and their compositions
SummaryThe paper proves the Fourier entropy–influence inequality for every linear threshold function, with a constant below 8.937 and no restrictions on weights or threshold. The results extend to regular Boolean functions and their combinations. Symmetric thresholds under arbitrary product measures also satisfy a strengthened version of the inequality needed for read-once composition.
WhyProves the Fourier entropy-influence inequality for all linear threshold functions with explicit constant, extending to compositions and multiplexers; substantial theoretical advance.
In briefRobust quickest change detection with cost-efficient experiment design under unknown post-change models
SummaryThis paper studies quickest change detection, where a decision-maker can choose only one of several experiments at a time, and the post-change distribution under each experiment is unknown but belongs to a known family. Each family is assumed to contain a least favorable distribution, which is used to design the proposed algorithm. The authors show the algorithm is first-order asymptotically optimal as the false alarm rate goes to zero. It minimizes the worst-case average detection delay under constraints on the average run length to false alarm and on the fraction of pre-change time each experiment is performed.
In briefDelay-Doppler Duality: Simpler Exact Equalization for Zak-OTFS in Time or Frequency
SummaryThis paper studies equalization for Zak-OTFS wireless signals, where inter-carrier interference raises the cost of standard equalizers. The authors show that the delay-Doppler channel matrix is equivalent, via the discrete Zak transform, to a banded matrix in either the time or frequency domain. They design a receiver that picks the cheaper domain and solves the system exactly, with complexity growing linearly in frame size. Simulations report error rates matching direct delay-Doppler LMMSE with three to four orders of magnitude fewer multiplications.
In briefExact privacy-distortion design for support-constrained randomized response
SummaryThe paper studies randomized response whose outputs are limited to a radius-t neighborhood of the private input, since full-support mechanisms cause unnecessary distortion on ordered alphabets. Under approximate local privacy for input pairs at distance at most H, it derives the exact hockey-stick defect and jointly designs the truthful probability, off-truth probability, and support radius. The authors also show that strict approximate privacy forces the support size to grow as Ω(H/δ).
Amirreza Zamani, Parastoo Sadeghi, Mikael Skoglund
In briefLearning-Based Scalable Multi-Objective Workflow Scheduling Across User Device, IaaS and FaaS Systems
SummaryThis paper studies how to schedule workflows across a hybrid edge-cloud setup of user devices, IaaS nodes and serverless FaaS instances, jointly optimizing makespan, energy and cost. The authors formulate the problem as a multi-objective mixed-integer linear program and propose a learning framework combining graph neural networks, surrogate performance modeling and multi-sample REINFORCE. On a real testbed running the Pegasus Montage workflow, the approach reduces makespan by 10-17.5%, energy by 3-9% and cost by 4-10% versus genetic and greedy baselines, staying within 3.5-5.6% of the MILP optimum.
In briefMulti-GPU communication for iterative solvers without CPU control
SummaryThis paper proposes a fully autonomous multi-GPU execution model in which the CPU is involved only in the initial kernel launch. It combines existing techniques such as persistent kernels, thread block specialization, device-side barriers, and device-initiated communication to reduce communication overhead. Tested on 2D/3D Jacobi stencil and Conjugate Gradient solvers, the model reduced 3D stencil communication latency by 58.8% and gave a 1.63x speedup for CG on 8 NVIDIA A100 GPUs.
Ismayil Ismayilov, Javid Baydamirli, Do\u{g}an Sa\u{g}bili et al.
In briefPrecise large-blocklength asymptotics for approximate channel simulation of symmetric channels
SummaryThis paper studies approximate channel simulation in the normal approximation regime, where the minimal codebook size is linked to the smooth channel max-information. For Gallager-symmetric discrete memoryless channels, the authors derive tight third-order expansions and show the third-order term depends on the channel. Unlike channel coding, the rate gap between singular and nonsingular channels is reversed: the third-order term is -(1/2)log n for singular channels and O(log log n) for nonsingular ones.
Valentinian Lungu, Aaron B. Wagner, Ioannis Kontoyiannis
In briefHigher-order nonhomomorphicity of Boolean functions restricted to affine subspaces, with results on threshold functions
SummaryThis paper studies how the k-th-order nonhomomorphicity of Boolean functions changes when they are restricted to affine subspaces, introducing the t-restriction measure P_{k,t}(f). For a direct sum of m monomials of degree at least two, the authors determine this value exactly: when t<m, the minimum comes from setting one variable to zero in each of the t lowest-degree monomials, and when m≤t<n it is zero. The work also gives a Walsh-transform formula for restrictions and shows that every non-affine function has a hyperplane restriction with smaller odd-order nonhomomorphicity. Finally, the normalized nonlinearity of all threshold functions with n≥3 is determined, resolving two open conjectures.
Serge Feukoua, Ana S\u{a}l\u{a}gean, Pantelimon St\u{a}nic\u{a}
In briefFast Quantum LDPC Decoding via Learned Update Scheduling
SummaryThis paper proposes bucket RL-S (B-RL-S), a belief-propagation decoder for quantum LDPC codes that quantizes reinforcement-learned Q-values into integer priorities and schedules variable-node updates with first-in, first-out buckets. It requires no retraining and no changes to the BP message updates. For the [[288,12,18]] bivariate bicycle code over the depolarizing channel at p=0.05, the eight-level version achieves roughly one fifth the block error rate of flooding BP followed by order-10 ordered-statistics decoding, without an extra post-processing stage.
In briefGeneralization bounds when only residual information is privacy-constrained
SummaryThe paper studies i.i.d. pairs (X,Y) where a learner observes both, leaks at most ε_n nats about X^n, and incurs loss on Y. A chain-rule argument yields the universal certificate I(W;Y^n) ≤ min{nH(Y), nH(Y|X)+ε_n}, which is combined with mutual-information generalization bounds. For deterministic-task and independent-residual source families, the resulting √(ε_n/n) rate is shown to be tight up to a universal constant.
Amirreza Zamani, Parastoo Sadeghi, Mikael Skoglund
In briefReed-Solomon codes over binary extension fields can need superpolynomial list sizes at every fixed rate
SummaryThis paper studies list decoding of ordinary Reed-Solomon codes over binary extension fields. For every fixed rate 0<R<1, it shows there is a constant ε_R>0 such that, for large enough h, some received word over F_{2^h} has n^{Ω_R(log n)} codewords within relative distance 1-R-ε_R. This contrasts with prime fields, where polynomial list sizes hold, so the list-size behavior depends on the underlying field.
WhyBuilds superpolynomial Reed-Solomon list-size lower bounds at every fixed rate via a rigorous BCH weight-enumerator correspondence and extension arguments, a notable contrast to prime-field results.
In briefFair allocation of item bundles where agents value complementary items together
SummaryThis paper studies fair allocation when agents have dichotomous submodular preferences over groups of items rather than single items, a natural extension useful for settings like course allocation. The authors show that computing utilitarian or Nash-optimal allocations is NP-hard in this setting. When complementary bundles have size at most k, polynomial-time algorithms achieve approximate efficiency, and approximately maximum utilitarian welfare EF-X allocations can be computed efficiently with per-agent utility guarantees.
WhyExtends dichotomous preferences to complementary item bundles, proving NP-hardness and giving polynomial approximation algorithms with provable EF-X fairness guarantees.
Soumyajit Pyne, Vignesh Viswanathan, Wenhao Yang et al.
In briefAverage list size of VT codes under two insertions or deletions
SummaryThis paper studies list decoding of binary VT codes under insertion and deletion errors, adapting McEliece's average list size problem. For length-n VT codes, the average number of codewords returned under two insertions is at most (n+12)/4, while under two deletions it is at least n/2 − 10 on average. The nearly factor-of-two gap suggests that, from an average-case view, insertions are easier to list-decode than deletions.
Anisha Banerjee, Roni Con, Antonia Wachter-Zeh et al.
In briefDual Pools or Shared Paging? Memory Layouts for Concurrent Hybrid LLM Serving
SummaryThis paper compares four memory organization schemes for serving hybrid attention and state-space model LLMs, implemented in a modified vLLM runtime. Progressive refinements from shared paging to recurrent-state paging and convolution-slot packing improve completed-output throughput and tail completion time at the main operating point. The authors find that recurrent-checkpoint availability limits joint reuse in the tested runs, while the correctness of retained-state continuation remains unresolved.
In briefSharp Rogers–Shephard inequalities for Rényi entropy: an entropy bound on differences of independent log-concave vectors
SummaryThe paper proves sharp weighted Rogers–Shephard inequalities for Rényi differential entropy. For independent log-concave random vectors X and X', it shows h_p(aX−bX') ≤ h_p(X) + n·log(a+b) for all a, b > 0 and 1 ≤ p ≤ ∞, where p=1 is Shannon entropy and p=∞ is min-entropy. The Shannon case resolves the entropic Rogers–Shephard conjecture of Madiman and Kontoyiannis. For 0 < p < 1, uniform laws on high-dimensional simplices violate the unweighted bound, so p=1 is the critical order.
WhyProves sharp Rényi-entropy Rogers–Shephard inequalities, settling the Madiman–Kontoyiannis conjecture with complete proofs and characterized equality cases.
In briefDualColl: A Throughput-Optimal Method for Bandwidth-Limited Collective Communication Using Fractional Tree Packing
SummaryThis paper addresses bandwidth-bound AllReduce collective communication in large-scale distributed training, where network bandwidth is a central bottleneck. It proposes DualColl, which optimizes AllReduce directly rather than composing AllGather and ReduceScatter. Using a multiplicative-weights algorithm with an exact minimum-cost collective-tree oracle, it computes schedules whose throughput is at least (1-ε)² of optimal. The framework is also extended to packet-sprayed routing, networks without explicit per-tree bandwidth control, other collective operations, and multiple concurrent groups.
In briefCorrecting Burst Errors and Their Use in DNA Data Storage
SummaryThis paper studies permutation burst errors, which scramble the entries of a substring, motivated by molecular data storage read with nanopore sequencing. It shows how to reduce the redundancy of classical burst-correcting codes such as Fire and Reed-Solomon codes by exploiting the fact that the error preserves substring weight. The authors give redundancy lower bounds, asymptotic upper bounds, and constructions for permutation and cyclic permutation bursts.
