Magic Dice
Let \(X\), \(Y\) be random variables, each with range \(\{1,\dots, k\}\) and \(P\) \((X= i,Y=j)= p(i,j)\). Is it possible, when \(k^2p(i,j)\neq 1\) for some \((i,j)\), to find linear combinations \(aX+ bY\), each with distribution \((*)\) the same as when \(k^2p(i,j)= 1\) for all \((i,j)\)? The paper starts with an example \(X\), \(Y\), \(X+ Y\), \(X-Y\), \(X+ 2Y\) for \(k= 6\) so that a magician may offer seemingly fair bets on a pair of dice to an audience that observes the distribution of each of these combinations. Looking for such \((a,b)\) may be restricted to the set \({\mathcal J}(k)\) of pairs of integers \((a,b)\) such that \(0\leq a\leq k-1\), \(|b|\leq k-1\); \(ab= 0\) only if \(a= 1\) or \(b=1\); when \(ab\neq 0\), then \(a\) and \(b\) are relative prime. The maximum cardinality of sets \({\mathcal S}\subset{\mathcal J}(k)\) such that \((a,b)\in{\mathcal S}\) satisfy \((*)\) for some \(p(i,j)\) not all equal to \(k^{-2}\), the same \(p(i,j)\) for all \((a,b)\in{\mathcal S}\), is called the magic number \(m(k)\) for \(k\)-sided dice. The \(m(k)\), \(k= 1,2,\dots, 24\), are tabulated. For \((a,b)\in{\mathcal J}(k)\) to satisfy \((*)\) a set of linear equations in the unknowns \(k^2p(i,j)- 1\) should have a nonzero solution. To find \({\mathcal S}\subset{\mathcal J}(k)\) as above the system of equations for all \((a,b)\) in \({\mathcal S}\) should have a coefficient matrix with \(\text{rank}< k^2\). A recurrence for these ranks is derived. By studying the equations the authors find several characterizations of \(m(k)\) by which it may be found without actually solving them. One of these expresses \(m(k)\) in terms of Euler's totient function. The asymptotic behaviour of \(m(k)\) for \(k\to\infty\) is derived.
This page was built for publication: Magic Dice
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2757363)