Zaremba’s conjecture and Korobov’s optimal coefficients
Zaremba’s conjecture and Korobov’s optimal coefficients
Let $p$ be a prime number, $d$ be a positive integer, and $M\ge 1$ be a real parameter.
A tuple $(a_1,\dots, a_d) \in \mathbb{F}^d_p$ is called a tuple of
(Korobov) optimal coefficients if, for any nonzero $x\in \mathbb{F}_p$, one has \[ |x| |a_1 x| \dots |a_d x| \ge \frac{p^d}{M}.\]
Here for $y\in\mathbb{F}_p$ we write $|y|$ for the absolute value of
the representative of $y$ in $(-p/2,p/2]$.
These coefficients arise naturally in problems of numerical
integration.
Namely, if a tuple $(a_1, \dots, a_d)$ satisfying the above
condition is found, then any function $f:[0,1]^{d+1} \to \mathbb{R}$
can be integrated using the formula
\[\left | \int_{[0,1]^{d+1}} f(x)\,dx - \frac{1}{p} \sum_{x=0}^{p-1}
f\left(\frac{x}{p}, \left\lbrace \frac{a_1 x}{p} \right\rbrace, \dots, \left\lbrace
\frac{a_d x}{p} \right\rbrace \right) \right| \ll \frac{M \log^d p \cdot
\mathrm{V}(f)}{p}\]
where $\mathrm{V}(f)$ is the Hardy--Krause variation of the function $f$.
Korobov (1959--1963) proved that the case $M=O(\log^d p)$ is always
realizable, whereas the special case $d=1$, $M=O(1)$ is equivalent to
the well-known Zaremba conjecture (1972): for any $p$ one can find $1\le
a<p$ such that
\[\frac{a}{p} = \cfrac{1}{c_1 +\cfrac{1}{c_2 +\cdots +\cfrac{1}{c_s}}} \,,\qquad \text{all} \quad c_j \quad \text{are bounded.}\]
For $d>1$ and $M =o (\log^d p)$, only a few results are known.
We give an overview of the problems in this area and describe recent
advances and connections to other topics in number theory and
combinatorics. We also discuss the connection between this topic and
McMullen's arithmetic chaos conjecture.