# Chernoff bound for high-dimensional expanders

Tali Kaufman, Ella Sharakanski

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

2 Scopus citations

## Abstract

We generalize the expander Chernoff bound to high-dimensional expanders. The expander Chernoff bound is an essential property of expanders, first proved by Gillman [9]. Given a graph G and a function f on the vertices, it states that the probability of f's mean sampled via a random walk on G to deviate from its actual mean, has a bound that depends on the spectral gap of the walk and decreases exponentially as the walk's length increases. We are interested in obtaining an analog Chernoff bound for high order walks on high-dimensional expanders. A naive generalization of the expander Chernoff bound from expander graphs to high-dimensional expanders gives a very poor bound due to obstructions that occur in high-dimensional expanders and are not present in (one-dimensional) expander graphs. Because of these obstructions, the spectral gap of high-order random walks is inherently small. A natural question that arises is how to get a meaningful Chernoff bound for high-dimensional expanders. In this paper, we manage to get a strong Chernoff bound for high-dimensional expanders by looking beyond the spectral gap. First, we prove an expander Chernoff bound that depends on a notion that we call the “shrinkage of a function” instead of the spectral gap. In one-dimensional expanders, the shrinkage of any function with zero-mean is bounded by λ(M). Therefore, the spectral gap is just the one-dimensional manifestation of the shrinkage. Next, we show that in good high-dimensional expanders, the shrinkage of functions that “do not come from below” is good. A function does not come from below if from any local point of view (called “link”) its mean is zero. Finally, we prove a high-dimensional Chernoff bound that captures the expansion of the complex. When the function on the faces has a small variance and does not “come from below”, our bound is better than the naive high-dimensional expander Chernoff bound.

Original language English Approximation, Randomization, and Combinatorial Optimization. Algorithms and Techniques, APPROX/RANDOM 2020 Jaroslaw Byrka, Raghu Meka Schloss Dagstuhl- Leibniz-Zentrum fur Informatik GmbH, Dagstuhl Publishing 9783959771641 https://doi.org/10.4230/LIPIcs.APPROX/RANDOM.2020.25 Published - 1 Aug 2020 23rd International Conference on Approximation Algorithms for Combinatorial Optimization Problems and 24th International Conference on Randomization and Computation, APPROX/RANDOM 2020 - Virtual, Online, United StatesDuration: 17 Aug 2020 → 19 Aug 2020

### Publication series

Name Leibniz International Proceedings in Informatics, LIPIcs 176 1868-8969

### Conference

Conference 23rd International Conference on Approximation Algorithms for Combinatorial Optimization Problems and 24th International Conference on Randomization and Computation, APPROX/RANDOM 2020 United States Virtual, Online 17/08/20 → 19/08/20