In briefGPU engine for sparse matrix-vector products on compressed k²-tree graphs
SummaryThis paper presents a GPU engine that performs sparse matrix-vector multiplication directly on k²-trees, a compact representation of Boolean matrices. It descends the tree level by level, obtaining child positions from a parallel prefix sum or a sampled directory, and processes it in tiles so that temporary memory is bounded by the tree depth. Tested on NVIDIA GB10 and H100 with fourteen graphs of up to 1.2 billion edges, the tree is 2.3–27 times smaller than cuSPARSE's CSR, and a product on the largest graph needs 1.1 GB of device memory, 9 times less than cuSPARSE.
In briefUniformly generating linear codes with a set hull size
SummaryThis paper presents exactly uniform samplers for linear codes whose hull dimension is specified in advance, covering the Euclidean, Hermitian, and symplectic inner products over finite fields of every characteristic. Apart from the Euclidean case over fields of odd characteristic, these are the first uniform samplers for arbitrary prescribed hull dimension. Each sample requires an expected O(k²N + k log q) field operations.
In briefAn information-geometric theory of distributed computation
SummaryThe paper introduces ΦID^IG, a geometric theory of information dynamics that captures irreducible collective computation in complex systems. It decomposes multi-input multi-output operations into interpretable computational modes, and validates the framework on discrete and Gaussian systems with known ground truth. The authors apply it to chaotic maps, the brain anatomy of over 100 mammalian species, and financial markets, reporting distinct computational profiles in each case.
Alberto Liardi, George Blackburne, Pedro A.M. Mediano
In briefBeyond GNSS: Using LEO satellite signals for positioning via Doppler shifts
SummaryLow Earth orbit satellites move rapidly, causing significant Doppler shifts that can be used to locate user terminals, especially where GNSS is unavailable. The paper reviews the principles, challenges, and enabling technologies of signals-of-opportunity Doppler positioning. It also proposes SionnaNTN-RT, a ray tracing framework built on Sionna for simulating non-terrestrial networks. Simulations show high positioning accuracy in both urban and open-sky settings.
Md. Ali Hasan, M. Humayun Kabir, Sangmin Han et al.
In briefExaTract: scaling global brain tractography to large cohorts and ultra-high-resolution scans
SummaryThis paper presents ExaTract, a parallel spin-glass tractography framework built for high-performance computing, implemented as an open-source MPI+Kokkos application. It splits the tractography domain into spatially overlapping regions processed across distributed memory nodes, with periodic synchronization of boundary changes to keep global convergence. The method is tested on in vivo and ex vivo diffusion MRI data, showing population-scale throughput on the full HCP dataset and producing the first mesoscopic structural connectome from global tractography on the high-resolution Chenonceau dataset.
Simon Legeay, Felix Matuschke, Bastien Herlin et al.
In briefOn inverse aggregate network assignment games: estimating how transport operators and travelers behave on a shared network from observed data.
SummaryThe paper extends aggregate assignment games to shared transport networks, offering path-based and link-based formulations with solution algorithms. It proposes an inverse optimization estimator for the link-based model, extended to capacitated two-sided markets. Calibrated on a Long Island Rail Road network of 897 nodes and 8,320 directed links, the method fits observed link flows well even when only 10-50% of links are sampled.
In briefAt What Cost? Inefficiency in First-Price Procurement Auctions
SummaryThis paper studies the efficiency of Bayes-Nash equilibria in first-price procurement auctions, measured by the Price of Anarchy, which compares equilibrium expected social cost to the optimal minimum. It shows that the standard format has no bounded efficiency guarantee. It then introduces a phantom bidder mechanism to restore a finite bound.
WhyThe authors show unbounded PoA for standard first-price procurement, derive a tight bound, and propose phantom-bidder mechanisms with matching lower bounds.
In briefNon-coherent over-the-air distributed consensus: algorithm design and power control optimization
SummaryThis paper presents a non-coherent over-the-air (NCOTA) consensus framework for distributed average consensus in wireless networks, requiring neither instantaneous channel state information nor phase alignment. It proposes a decentralized projected gradient descent algorithm with row-stochastic mixing (DPGD-RS) and proves its mean-square convergence to the exact average of the agents' initial states. The authors also derive a closed-form bound on the mean-square consensus error and a spectral-reshaping power control scheme that jointly optimizes transmit powers and receive-scaling factors, which converges faster than full-power transmission and spectral-radius-minimization benchmarks in numerical results.
In briefClassifying good rational functions of degree up to six
SummaryThis paper studies locally recoverable codes, which underpin modern distributed storage systems. It completes the classification of good rational functions of degree up to 6 in characteristic 2, using algebraic function field theory to determine their minimal arithmetic and geometric Galois groups. The authors prove that the exact number of totally split places over extension fields follows a linear recurring sequence. This provides an efficient way to compute the exact lengths and dimensions of the corresponding optimal LRCs.
In briefuTECH Turbo-MOVES: Faster Vehicle Emissions Modeling Through Software Architecture
SummaryThis paper presents uTECH Turbo-MOVES, a software architecture that speeds up computation in the U.S. EPA's Motor Vehicle Emission Simulator (MOVES) without changing its Calculator modules or input/output structure. It combines RunSpec partitioning, container isolation, in-memory intermediate storage, and concurrent workload orchestration. Across six benchmark cases, it reports a measured 5.5x and an estimated 15.3x per-run speedup for Default-scale workloads, and an 84.8x campaign speedup for an eight-county Emission Rates analysis.
Cafer Avc{\i}, Alireza Yazdiani, Roopak Srinivasan et al.
