The number zeta(3), also written as 1 + 1/8 + 1/27 + 1/64 + ..., is known to be irrational, meaning it cannot be expressed as a simple fraction. But mathematicians care not just about whether a number is irrational, but how "hard" it is to approximate by fractions. The irrationality exponent measures this difficulty: a larger exponent means the number can be approximated by fractions surprisingly well. For ordinary irrational numbers like most that arise in practice, this exponent equals 2, but special constants like zeta(3) can have larger values, and bounding that value from above is a longstanding challenge in number theory.
For 25 years, the best known upper bound on the irrationality exponent of zeta(3) was 5.513891, established by Rhin and Viola in 2001. The current paper improves this to 5.5138800, a modest but meaningful advance given how long the previous record stood. The method follows a framework Rhin and Viola themselves developed, which turns the problem into finding a clever choice of a mathematical integral, described by eight numerical parameters. The key insight in the new work is that the best choice of parameters does not lie anywhere near the region that previous computer searches explored. The new parameters are roughly 52,000 times larger than those of Rhin and Viola, shifted only slightly, placing them far outside any reasonable search range.
The discovery was inspired by recent work of Bai on a similar problem involving the constant pi. In that case, Bai also found that the long-accepted "optimal" solution was actually not optimal, once a much larger region of parameter space was explored. The lesson from both results is that the standard approach of enlarging search grids gradually can miss dramatically better solutions hiding in unexpected corners of a very large space. This suggests that other longstanding bounds in number theory, obtained by similar optimization methods, might also be improvable by looking in entirely new directions rather than refining the same neighborhood.