For real numbers $x$, let $\|x\|$ denote the distance from $x$ to the nearest integer. Define a function $\ell:\mathbb{R}^2\rightarrow\mathbb{R}$ by $$\ell(\alpha,\beta)=\liminf_{n\rightarrow\infty}n\|n\alpha\|\|n\beta\|.$$ The Littlewood conjecture asserts that, for all $(\alpha,\beta)\in\mathbb{R}^2$, we have $\ell (\alpha,\beta)=0$.

Can anyone see how to prove the (seemingly much weaker) statement that, for all $(\alpha,\beta)\in\mathbb{R}$, $$\inf_{A\in\mathrm{SL}_2(\mathbb{Z})}\ell\left(A\left( \begin{array}{c} \alpha\\ \beta\\ \end{array} \right)\right)=0\quad ?$$ I wouldn't be surprised if this problem has an easy solution, but I haven't yet been able to find one. For the purposes of orientation note that, in the definition of $\ell$ if we replace $\liminf$ by $\inf$, then the problem becomes close to trivial.


I believe so. Let $K$ be a large positive integer, and $N$ an integer parameter going to infinity. By Dirichlet approximation we can find $n \in [N,2N]$ such that $\|n\alpha\|, \|n\beta\| = O(N^{-1/2})$; we can in fact assume that $\|n\alpha\|, \|n \beta\| \asymp N^{-1/2}$ since otherwise we are done. For sake of notation let's suppose that the signed fractional parts $\{n\alpha\}, \{n\beta\} \in (-1/2,1/2]$ are positive (and thus comparable to $N^{-1/2}$). Let $p_j/q_j$ be the last continued fraction approximant to $\{n\alpha\}/\{n\beta\}$ (a quantity comparable to 1) with $q_j \leq K$. (Note we may assume this ratio is irrational, since Littlewood's conjecture is easy in the commensurate case.) Standard continued fraction theory tells us that $$ p_{j-1} q_j - p_j q_{j-1} = \pm 1$$ and $$ p_j - q_j \{n\alpha\}/\{n\beta\} = O(1/q_{j+1})$$ $$ p_{j-1} - q_{j-1} \{n\alpha\}/\{n\beta\} = O(1/q_{j})$$ and so $\| n (p_j \beta - q_j \alpha) \| = O( N^{-1/2}/q_{j+1} ) = O( N^{-1/2}/K)$ and $\| n (p_{j-1} \beta - q_{j-1} \alpha) \| = O( N^{-1/2}/q_j ) = O(N^{-1/2})$. After pigeonholing we can find (for fixed $K$) an infinite sequence of $N$ such that the $p_j,q_j,p_{j-1},q_{j-1}$ are constant, and thus $\ell( p_j \beta - q_j \alpha, p_{j-1} \beta - q_{j-1} \alpha ) = O(1/K)$. Letting $K$ to infinity we obtain the claim.

  • $\begingroup$ Thank you, I think that is correct. Am I right in assuming that your reason for wanting $\{n\alpha\}/\{n\beta\}$ to be $\asymp 1$ is just so that the first partial quotient is bounded? $\endgroup$ – Alan Haynes Jun 19 '15 at 7:09
  • $\begingroup$ yes; it also keeps the $p_j$ of magnitude comparable to $q_j$, allowing the pigeonholing to work. $\endgroup$ – Terry Tao Jun 19 '15 at 14:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.