We investigate the randomized and quantum communication complexities of the well-studied Equality function with small error probability ε, getting the optimal constant factors in the leading terms in various different models. The following are our results in the randomized model: - We give a general technique to convert public-coin protocols to private-coin protocols by incurring a small multiplicative error at a small additive cost. This is an improvement over Newman’s theorem [Inf. Proc. Let.'91] in the dependence on the error parameter. - As a consequence we obtain a (log(n/ε²) + 4)-cost private-coin communication protocol that computes the n-bit Equality function, to error ε. This improves upon the log(n/ε³) + O(1) upper bound implied by Newman’s theorem, and matches the best known lower bound, which follows from Alon [Comb. Prob. Comput.'09], up to an additive log log(1/ε) + O(1). The following are our results in various quantum models: - We exhibit a one-way protocol with log(n/ε) + 4 qubits of communication for the n-bit Equality function, to error ε, that uses only pure states. This bound was implicitly already shown by Nayak [PhD thesis'99]. - We give a near-matching lower bound: any ε-error one-way protocol for n-bit Equality that uses only pure states communicates at least log(n/ε) - log log(1/ε) - O(1) qubits. - We exhibit a one-way protocol with log(√n/ε) + 3 qubits of communication that uses mixed states. This is tight up to additive log log(1/ε) + O(1), which follows from Alon’s result. - We exhibit a one-way entanglement-assisted protocol achieving error probability ε with ⌈log(1/ε)⌉ + 1 classical bits of communication and ⌈log(√n/ε)⌉ + 4 shared EPR-pairs between Alice and Bob. This matches the communication cost of the classical public coin protocol achieving the same error probability while improving upon the amount of prior entanglement that is needed for this protocol, which is ⌈log(n/ε)⌉ + O(1) shared EPR-pairs. Our upper bounds also yield upper bounds on the approximate rank, approximate nonnegative-rank, and approximate psd-rank of the Identity matrix. As a consequence we also obtain improved upper bounds on these measures for a function that was recently used to refute the randomized and quantum versions of the log-rank conjecture (Chattopadhyay, Mande and Sherif [J. ACM'20], Sinha and de Wolf [FOCS'19], Anshu, Boddu and Touchette [FOCS'19]).

The multiple knapsack problem is a well-studied combinatorial optimization problem with several practical and theoretical applications. It consists of packing some subset of n items into m knapsacks such that the total profit of the chosen items is maximum. A new formulation of the problem is presented, where a Lagrangian relaxation is derived, and we prove that it dominates the commonly used relaxations for this problem. We also present a Dantzig-Wolfe decomposition of the new formulation that we solve to optimality using a branch-and-price algorithm, where its main advantage comes from the fact that it is possible to control whether an item is included in some knapsack or not. An improved algorithm for solving the resulting packing subproblems is also introduced. Computational experiments then show that the new approach achieves state-of-the-art results. History: Accepted by Andrea Lodi, Area Editor for Design & Analysis of Algorithms–Discrete. Funding: This work was supported by the Canadian Natural Sciences and Engineering Research Council (NSERC) [Grants 2017-06054 and 2021-04037]. This support is gratefully acknowledged.