Discrepancy of high-dimensional permutations

From MaRDI portal



Abstract: Let L be an order-n Latin square. For X,Y,Zsubseteq1,...,n, let L(X,Y.Z) be the number of triples iinX,jinY,kinZ such that L(i,j)=k. We conjecture that asymptotically almost every Latin square satisfies |L(X,Y,Z)−frac1n|X||Y||Z||leO(sqrt|X||Y||Z|) for every X,Y and Z. Let varepsilon(L):=max|X||Y||Z| when L(X,Y,Z)=0. The above conjecture implies that varepsilon(L)leO(n2) holds asymptotically almost surely (this bound is obviously tight). We show that there exist Latin squares with varepsilon(L)leO(n2), and that varepsilon(L)leO(n2log2n) for almost every order-n Latin square. On the other hand, we recall that varepsilon(L)geqOmega(n33/14) if L is the multiplication table of an order-n group. Some of these results extend to higher dimensions. Many open problems remain.











This page was built for publication: Discrepancy of high-dimensional permutations

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2826225)