A theoretical preprint reports that fixed-order Khatri-Rao sketches can achieve a near-linear subspace-embedding dimension in the subspace size, improving the stated dependence on that size over earlier bounds. The result is a proof under sub-Gaussian assumptions, not an experimental performance claim.
A theoretical preprint proposes algorithms for flexible graph connectivity across several nested tiers. Its main randomized method succeeds with probability at least one third for a fixed number of tiers, while a related multi-graph problem receives a factor-two approximation.
The arXiv work claims a deterministic polynomial-time certificate for every nonnegative rational square matrix, with an exponential approximation base below the canonical Bethe guarantee.
A computational study reports near-complete hypervolume from a parallel multi-objective MaxCut solver in 0.9 seconds on two benchmark instances, with important limits on what the result shows.
A methods paper extends path abstraction from reachability probabilities to expected rewards in finite Markov reward models. It shows that the abstraction remains a Markov reward model, behaves consistently across nested state sets, and can be computed using a linear-equation recipe with a high-level PARI/GP reference implementation.
A theoretical protocol combines locally consistent parsing with invertible Bloom lookup tables to reconcile separately held strings while tying communication to their edit distance.
A theoretical preprint reports faster exact procedures for two axis-parallel problems, while showing that the balance limits in one problem are still not fully settled.
A theoretical preprint reports quadratic-time all-pairs shortest paths, fast matrix operations, and near-linear detection of several small graph patterns in structured graph classes.
The methods paper combines tapering, key uniformization and XOR amplification, then claims NC1 pseudorandom functions under LWE, LPN, CDH and GapSVP conditions.
A theoretical preprint presents exact acceptance-cancellation identities and warm-start mixing bounds for Metropolis-adjusted Dikin walks on polytopes and spectrahedra, alongside exact-arithmetic implementation results and a family-specific warning about unscaled proposals.
A theoretical paging study gives a formal counterexample to a cache-victim rule, while reporting exact results for restricted configurations, a unit-cost factor-five approximation, and weight-sensitive bounds.
A theoretical study reports different sample needs for learning a nearly second-best bilateral-trade mechanism under additive and multiplicative guarantees.
Two exact methods reached the same optimal completion time on all 35 small scheduling instances, but the re-evaluation also found historical results that were sometimes infeasible.
A formal study reports polynomial-time algorithms for selected continuum attacks, while constructive discrete k-Approval-Swap Bribery with additively separable nonnegative rational costs is NP-complete for every fixed k at least 2 and general-cost Borda remains conditionally hard in the continuum.
The mathematical preprint reports an inverse-linear spectral-gap lower bound and logarithmic mixing-time upper bound for proper colorings on graphs with girth at least five, then extends its local-to-global framework to the anti-ferromagnetic Potts model.
A theoretical data structure reuses a compact spectral sparsifier across stable update periods to return approximate all-pairs graph queries under stated asymptotic guarantees.
A theoretical preprint examines dimension-independent strong row coresets for ℓp subspace approximation. It reports kε⁻² size dependence up to suppressed factors for 1 ≤ p < 2 and k^(p/2)ε⁻² for p > 2, while the p > 2 upper bound does not match the known lower bound jointly in rank and accuracy.
An arXiv preprint describes a distributed method for games that mix cooperation and competition. In one simulated energy example, the method was reported to reach the same accuracy in less computation time than two Euclidean comparison methods.
A preprint reports that the QisMC model checker handled some 100-qubit quantum programs within seconds, while circuit structure and numerical precision limited its reach.
A mathematical preprint reports parameterized hardness for exact Lp maximization over generator-represented zonotopes and a corresponding two-layer ICNN Lipschitz-constant problem, while leaving W[1] membership open.
A new scheduling preprint maps how the difficulty of finding maximally fair schedules changes with the number of days and the measure used to judge service.
A mathematical analysis finds that GREEDY can produce a common superstring at least twice as long as the optimum when input strings have length six or more. It also establishes an exact worst-case ratio of 9/5 for strings of length three.
An arXiv modeling paper studies fair allocation as indivisible items arrive over time. It finds constructive guarantees in restricted settings, but shows that TEF1 implies no more than a tight 1/n temporal maximin-share guarantee for additive goods.
A preprint on Seg-Agony charts the computational boundary for temporal directed-graph instances. It reports fixed-parameter tractability for the combined parameters n + ℓ, a polynomial-time algorithm for two ranks, and hardness results in unweighted three- and four-rank regimes.
An arXiv preprint develops a formal probability framework for a six-valued logical system representing gaps, gluts and reliability. It connects several probability representations through inverse mappings and reports equivalent semantic and syntactic update rules. The work is mathematical rather than empirical.
A methods preprint studies how allowing relative error changes the guarantees for differentially private continual release. Its bounds improve additive error for several tasks and stream settings, while adaptive MinSelect remains subject to a large lower bound.
An arXiv preprint finds a sharp divide in Product-Gap, a randomized facility-location rule on the real line. Its expected social cost is at most 2k times optimal for every k ≥ 2, but strategyproofness in expectation holds only for two and three facilities and fails from four onward.