On Communication Complexity of Fixed Point Computation

09/24/2019
by   Anat Ganor, et al.
0

Brouwer's fixed point theorem states that any continuous function from a compact convex space to itself has a fixed point. Roughgarden and Weinstein (FOCS 2016) initiated the study of fixed point computation in the two-player communication model, where each player gets a function from [0,1]^n to [0,1]^n, and their goal is to find an approximate fixed point of the composition of the two functions. They left it as an open question to show a lower bound of 2^Ω(n) for the (randomized) communication complexity of this problem, in the range of parameters which make it a total search problem. We answer this question affirmatively. Additionally, we introduce two natural fixed point problems in the two-player communication model. ∙ Each player is given a function from [0,1]^n to [0,1]^n/2, and their goal is to find an approximate fixed point of the concatenation of the functions. ∙ Each player is given a function from [0,1]^n to [0,1]^n, and their goal is to find an approximate fixed point of the interpolation of the functions. We show a randomized communication complexity lower bound of 2^Ω(n) for these problems (for some constant approximation factor). Finally, we initiate the study of finding a panchromatic simplex in a Sperner-coloring of a triangulation (guaranteed by Sperner's lemma) in the two-player communication model: A triangulation T of the d-simplex is publicly known and one player is given a set S_A⊂ T and a coloring function from S_A to {0,... ,d/2}, and the other player is given a set S_B⊂ T and a coloring function from S_B to {d/2+1,... ,d}, such that S_A∪̇S_B=T, and their goal is to find a panchromatic simplex. We show a randomized communication complexity lower bound of |T|^Ω(1) for the aforementioned problem as well.

READ FULL TEXT

Please sign up or login with your details

Forgot password? Click here to reset
Success!
Error Icon An error occurred

Sign in with Google

×

Use your Google Account to sign in to DeepAI

×

Consider DeepAI Pro