Some experimental results on the Frobenius problem

Matthias Beck,David Einstein,Shelemyahu Zacks
DOI: https://doi.org/10.48550/arXiv.math/0204036
2005-01-02
Abstract:We study the Frobenius problem: given relatively prime positive integers $a_1,...,a_d$, find the largest value of t (the Frobenius number) such that $\sum_{k=1}^d m_k a_k = t$ has no solution in nonnegative integers $m_1,...,m_d$. Based on empirical data, we conjecture that except for some special cases the Frobenius number can be bounded from above by $\sqrt{a_1 a_2 a_3}^{5/4} - a_1 - a_2 - a_3$.
Number Theory,Combinatorics
What problem does this paper attempt to address?
The problem that this paper attempts to solve is a specific aspect of the Frobenius problem. That is, given relatively prime positive integers \(a_1, a_2,\ldots, a_d\), find the largest value \(t\) (called the Frobenius number \(g(a_1, a_2,\ldots, a_d)\)) such that the equation \[ \sum_{k = 1}^{d}m_k a_k=t \] has no non - negative integer solutions \(m_1, m_2,\ldots, m_d\). Specifically, the focus of the paper is to establish an upper bound for the case of \(d = 3\) and verify the rationality of this upper bound through experimental data. ### Main objectives of the paper: 1. **Study the Frobenius problem**: For given three relatively prime positive integers \(a, b, c\), find the largest integer \(t\) that cannot be expressed as \(am+bn + cp\). 2. **Establish an upper bound**: Based on experimental data, the author speculates that, except for some special cases, the Frobenius number \(g(a, b, c)\) can be estimated by the following formula for the upper bound: \[ g(a, b, c)\leq\sqrt{abc}^{\frac{5}{4}}-a - b - c \] ### Background and motivation: - For the case of \(d = 2\), it is known that the Frobenius number \(g(a_1, a_2)=a_1a_2 - a_1 - a_2\). - For the case of \(d>2\), explicit formulas are difficult to obtain, so researchers focus more on finding reasonable upper and lower bounds. - The existing upper bounds in the literature are usually proportional relationships of the products of two or more \(a_i\), and the lower bound was proposed by Davison: \[ g(a_1, a_2, a_3)\geq\sqrt{3a_1a_2a_3}-a_1 - a_2 - a_3 \] ### Experimental results: - By calculating more than 10,000 randomly selected triples \((a, b, c)\), it was found that in most cases the Frobenius number \(f(a, b, c)\) is close to the lower bound \(\sqrt{3abc}\). - Statistical results show that in about 75% of cases \(R = f/z<2.30\), where \(z=\sqrt{abc}\). ### Conclusions and conjectures: - The author proposes a conjecture that for all acceptable triples \((a, b, c)\), there exists an upper bound: \[ g(a, b, c)\leq\sqrt{abc}^p - a - b - c \] where \(p<\frac{4}{3}\). More specifically, they believe that: \[ g(a, b, c)\leq\sqrt{abc}^{\frac{5}{4}}-a - b - c \] This is more compact than the upper bounds in the existing literature.