核心概念
Bounded integral functions in polytopes have sub-exponential extension complexity.
摘要
This article discusses the deterministic communication complexity of bounded integral functions in polytopes. It explores the relationship between rank, non-negative matrices, and extension complexity. The key results include theorems on communication protocols, non-negative rank, positive semidefinite rank, and their implications on extension complexity. The article also delves into finding large monochromatic rectangles in matrices and provides a proof for the main theorem.
-
Introduction to Communication Complexity
- Two-player Boolean function computation.
- Deterministic communication protocol definition.
- Log-rank conjecture by Lovász and Saks.
-
Main Results on Communication Complexity
- Theorem 1: Communication protocol for bounded integral functions.
- Non-negative rank and its relation to quantum communication complexity.
-
Extension Complexity and Polytopes
- Definition of extension complexity in polytopes.
- Connection between extension complexity and deterministic communication complexity.
-
Communication Protocol Design
- Recursive reduction of matrix to compute submatrices efficiently.
- Balancing techniques for protocol tree optimization.
-
Finding Monochromatic Rectangles
- Reduction to finding almost-monochromatic rectangles.
- Lemmas on factorization norm, Gaussian distributions, and hyperplane rounding.
-
Proof of Main Theorem
- Application of Lemmas to derive the bound on communication complexity.
-
Acknowledgment
- Thanks to Sam Fiorini for discussions and reviewers for feedback.
統計資料
"The positive semidefinite rank of M... has an analogous relationship to quantum communication complexity [FGP+15]."
"The positive semidefinite rank of M... is defined as the minimum r such that there are positive semidefinite matrices U1, ... UA..."
"Barvinok [Bar12] showed that if M has at most k distinct entries... then rankpsd(M) ≤ k−1+rank(M) k−1."
引述
"Any matrix M ∈RA×B satisfies γ2(M) ≤∥M∥∞· p rank(M)."
"For any matrix M ∈RA×B and s ≥ p γ2(M), there are vectors {ua}a∈A, {vb}b∈B such that Ma,b = ⟨ua, vb⟩and ∥ua∥2 = ∥vb∥2 = s..."