Refined upper bounds for the linear Diophantine problem of Frobenius

Matthias Beck,Shelemyahu Zacks

DOI: https://doi.org/10.48550/arXiv.math/0305420

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 g(a_1,...,a_d)) such that m_1 a_1 + ... m_d a_d = t has no solution in nonnegative integers m_1,...,m_d. We introduce a method to compute upper bounds for g(a_1,a_2,a_3), which seem to grow considerably slower than previously known bounds. Our computations are based on a formula for the restricted partition function, which involves Dedekind-Rademacher sums, and the reciprocity law for these sums.

Number Theory,Combinatorics

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the Frobenius problem of linear Diophantine equations. That is, given a set of relatively prime positive integers \(a_1, a_2, \ldots, a_d\), find the largest integer \(t\) (called the Frobenius number \(g(a_1, a_2, \ldots, a_d)\)) such that the equation

\[
t = m_1 a_1 + m_2 a_2 + \cdots + m_d a_d
\]

has no non - negative integer solutions \(m_1, m_2, \ldots, m_d\). Specifically, the goal of the paper is to provide more precise upper bounds for the Frobenius number \(g(a_1, a_2, a_3)\), and these upper bounds grow more slowly compared to other known upper bounds.

### Main contributions of the paper

1. **Introducing a new method**: The paper introduces a method based on the restricted partition function to calculate the upper bounds, which involves the Dedekind - Rademacher sum and its reciprocity law.

2. **Improving existing upper bounds**: Through this method, the paper derives new upper - bound formulas, which are more precise than the existing upper bounds, especially in the case of \(d = 3\).

3. **Numerical verification**: The paper compares the new and old upper bounds through a large number of randomly generated triples \((a, b, c)\). The results show that the new upper bounds are tighter than the known upper bounds in most cases and are closer to the actual Frobenius number.

### Mathematical background

For the case of two variables (\(d = 2\)), the Frobenius number has an explicit formula:

\[
g(a_1, a_2) = a_1 a_2 - a_1 - a_2
\]

But for three or more variables (\(d>2\)), finding an explicit formula becomes very difficult. Therefore, researchers turn to finding effective upper bounds. Some of the known upper bounds mentioned in the paper include:

- Erdős and Graham's upper bound:
\[
g(a_1, a_2, \ldots, a_d) \leq 2 a_d \left\lfloor\frac{a_1}{d}\right\rfloor - a_1
\]
- Selmer's upper bound:
\[
g(a_1, a_2, \ldots, a_d) \leq 2 a_{d - 1} \left\lfloor\frac{a_d}{d}\right\rfloor - a_d
\]
- Vitek's upper bound:
\[
g(a_1, a_2, \ldots, a_d) \leq \left\lfloor\frac{1}{2}(a_2 - 1)(a_d - 2)\right\rfloor - 1
\]

The new method proposed in the paper aims to provide tighter upper bounds, so as to better approximate the actual Frobenius number.