Communication Complexity

The first section starts with the basic definitions following mainly the notations of the book written by E. Kushilevitz and N. Nisan. At the end of the first section I examine tree-balancing. In the second section I summarize the well-known lower bound methods and prove the exact complexity of certain functions. In the first part of the third section I introduce the random complexity and prove the basic lemmas about it. In the second part I prove a better lower bound for the complexity of all random functions. In the third part I introduce and compare several upper bounds for the complexity of the identity function. In the fourth section I examine the well-known Direct-sum conjecture. I introduce a different model of computation then prove that it is the same as the original one up to a constant factor. This new model is used to bound the Amortized Time Complexity of a function by the number of the leaves of its protocol-tree. After this I examine the Direct-sum problem in case of Partial Information and in the Random case. In the last section I introduce the well-known hierarchy classes, the reducibility and the completeness of series of functions. Then I define the class PSPACE and Oracles in the communication complexity model and prove some basic claims about them.

💡 Research Summary

The paper presents a comprehensive survey and several original contributions to the theory of communication complexity. It begins by establishing the basic definitions and notation, largely following the conventions of Kushilevitz and Nisan. The authors then examine tree‑balancing, showing how the depth of a protocol tree can be related to the distribution of its leaves and why a balanced tree often yields the smallest communication cost.

In the second section the authors systematically review the classic lower‑bound techniques. They discuss partition arguments, the publish‑subscribe method, and information‑theoretic approaches based on mutual information and cross‑entropy. Using these tools they re‑prove the well‑known Ω(n) lower bounds for Disjointness and Equality, and they identify a handful of functions for which the lower and upper bounds match exactly, thereby determining their precise communication complexity.

The third section introduces randomized communication complexity. Two models are defined: public‑coin and private‑coin protocols with error probability ε < 1/2. The authors prove the fundamental lemmas that underpin this area, including Yao’s Minimax Principle and the Amplification Lemma, which allow error reduction at a modest cost. They then establish a stronger lower bound that applies to “almost all” random functions, arguing that because the output distribution of a random Boolean function is essentially uniform, any protocol must exchange Ω(n) bits on average. For the identity function, several upper‑bound constructions are compared—fingerprinting, hash‑based protocols, and others—highlighting the regimes where each method is near‑optimal. In particular, fingerprinting achieves O(log n) communication with arbitrarily small error, illustrating the power of randomness.

The fourth section tackles the Direct‑Sum conjecture, a central open problem that asks whether solving k independent instances of a function requires k times the communication needed for a single instance. The authors first propose an alternative computational model that they prove is equivalent to the standard model up to a constant factor. This model enables them to bound the amortized time complexity of a function by the number of leaves in its protocol tree. They then analyze Direct‑Sum in two specialized settings: partial‑information protocols and randomized protocols. In the partial‑information setting they exhibit counter‑examples showing that Direct‑Sum can fail, whereas in the randomized setting they prove that for a broad class of functions—including all random functions—the Direct‑Sum property holds, i.e., the total communication scales linearly with the number of instances.

The final section surveys the hierarchy of communication‑complexity classes, introduces notions of reducibility and completeness, and situates PSPACE‑complete functions within this framework. The authors define an oracle communication model and prove several basic facts about how oracles affect communication cost. For instance, they show that granting access to a PSPACE oracle can reduce the communication required for PSPACE‑complete problems from exponential to polynomial. Throughout the paper, the authors interleave rigorous proofs with intuitive explanations, thereby providing both a solid reference for seasoned researchers and an accessible entry point for newcomers. The new model for Direct‑Sum, the strengthened lower bound for random functions, and the oracle results together push the frontier of what is known about communication complexity, suggesting several promising directions for future work.