Work

Graph Intersections and Hypergraph Discrepancy

Year: 2026
Type: preprint
Abstract: Let $G$ and $H$ be graphs on $n$ vertices with $e(G)=p\binom{n}{2}$ and $e(H)=q\binom{n}{2}$, where $p,q\in(0,1)$ are fixed. Bollobás and Scott studied intersections of two graphs under random relabelling and asked for the minimum $L^2$ norm of the deviation of their intersection size from its expected value. We answer this question by proving that, for graphs of order $n$ with fixed edge densities $p,q\in(0,1)$, the minimum is \begin{equation*} \left(\sqrt{\frac{p(1-p)q(1-q)}{2}\cdot\frac{1-\sqrt R}{1+\sqrt R}}\right)n+o(n), \end{equation*} where $R=\max\{p,1-p,q,1-q\}$. The coefficient of the linear term is best possible. Bollobás and Scott defined the positive and negative discrepancies of two hypergraphs as the largest deviations of their intersection size above and below its expected value under relabelling, respectively. They asked for the minimum product of the positive and negative discrepancies of two $(n/2)$-uniform hypergraphs on $n$ vertices with $\frac{1}{2}\binom{n}{n/2}$ edges. We answer this question by proving that the minimum product of these discrepancies is $0$ if $4\mid\binom{n}{n/2}$ and $1/4$ otherwise.
Institution: Fuzhou University
Language: English
Open Access status: green
OpenAlex ID: W7234895091
Cites: 10
Cited by:
Related to: 10