TCSLib

44 Communication Complexity — Det Complexity

44.1 Overview

This module defines the deterministic communication complexity of a two-party function \(f : X \to Y \to \alpha \) as the infimum of the complexity (number of bits exchanged) over all deterministic protocols that compute \(f\). It also establishes the key characterizations of when this complexity is at most a given natural number \(n\), both in terms of general protocols and in terms of finite-message protocols.

44.2 Declarations

Theorem 44.1 Infimum over ENat-valued family bounded by coercion

Let \(\iota \) be a type and \(f : \iota \to \mathbb {N}_\infty \) an \(\mathbb {N}_\infty \)-valued family. Then \(\inf _{i} f(i) \le n\) if and only if there exists an index \(i\) such that \(f(i) \le n\). Here \(n : \mathbb {N}\) is coerced into \(\mathbb {N}_\infty = \mathrm{ENat}\).

Definition 44.2 Deterministic communication complexity

The deterministic communication complexity of \(f : X \to Y \to \alpha \) is defined as

\[ D(f) \; =\; \inf _{\substack {p \; :\; \mathrm{Protocol}\; X\; Y\; \alpha \\ p \text{ computes } f}} p.\mathrm{complexity} \; \in \; \mathbb {N}_\infty . \]

The infimum is taken over all deterministic protocols \(p\) that compute \(f\), and the result lives in \(\mathrm{ENat}\) to accommodate the case where no finite protocol exists.

Theorem 44.3 Upper bound characterization of communication complexity

For \(f : X \to Y \to \alpha \) and \(n : \mathbb {N}\), we have \(D(f) \le n\) if and only if there exists a protocol \(p : \mathrm{Protocol}\; X\; Y\; \alpha \) such that \(p\) computes \(f\) and \(p.\mathrm{complexity} \le n\).

For \(f : X \to Y \to \alpha \) and \(n : \mathbb {N}\), we have \(D(f) \le n\) if and only if there exists a finite-message protocol \(p : \mathrm{FiniteMessage.Protocol}\; X\; Y\; \alpha \) such that \(p.\mathrm{run} = f\) and \(p.\mathrm{complexity} \le n\). This equivalence follows from the mutual simulation between general protocols and their finite-message counterparts.

Theorem 44.5 Lower bound characterization of communication complexity

For \(f : X \to Y \to \alpha \) and \(n : \mathbb {N}\), we have \(n \le D(f)\) (in \(\mathbb {N}_\infty \)) if and only if every protocol \(p : \mathrm{Protocol}\; X\; Y\; \alpha \) that computes \(f\) satisfies \(n \le p.\mathrm{complexity}\).