Abstract:The Lagrange and Markov spectra $L$ and $M$ describe the best constants of Diophantine approximations for irrational numbers and binary quadratic forms. In 1880, A. Markov showed that the initial portions of these spectra coincide: indeed, $L\cap (0,3) = M\cap (0,3)$ is a discrete set of explicit quadratic irrationals accumulating only at $3$.
In this article, we show that the statement above ceases to be true immediately after $3$: in particular, $L\cap (3,3+\varepsilon)\neq M\cap (3,3+\varepsilon)$ for all $\varepsilon>0$, and thus $\inf(M\setminus L)=3$. In fact, we derive this result as a by-product of lower bounds on the Hausdorff dimension of $(M\setminus L)\cap (3,3+\varepsilon)$ implying that $\liminf\limits_{\varepsilon\to 0} \frac{\dim_H((M\setminus L)\cap(3,3+\varepsilon))}{\dim_H(M\cap (3,3+\varepsilon))}\geq \frac{1}{2}$ and, as it turns out, these bounds are obtained from the study of projections of Cartesian products of almost affine dynamical Cantor sets via an argument of probabilistic flavor based on Baker--Wüstholz theorem on linear forms in logarithms of algebraic numbers.
What problem does this paper attempt to address?
The main problem that this paper attempts to solve is about the difference between the Lagrange spectrum \(L\) and the Markov spectrum \(M\) near the value \(3\). Specifically, the authors show that after the value \(3\), the Lagrange spectrum and the Markov spectrum no longer coincide, that is:
\[L\cap(3, 3 +\varepsilon)\neq M\cap(3, 3+\varepsilon)\]
for all \(\varepsilon> 0\), and thus conclude that:
\[\inf(M\setminus L)=3\]
### Specific problem description
1. **Background and known results**:
- The Lagrange spectrum \(L\) and the Markov spectrum \(M\) are defined by the best approximation constants of irrational numbers and the optimal values of binary quadratic forms.
- In 1880, A. Markov showed that below the value \(3\), these two spectra are the same and are discrete sets consisting of explicit quadratic irrational numbers, which accumulate only at \(3\).
2. **New findings**:
- This paper proves that the above conclusion no longer holds after the value \(3\): within the interval \((3, 3+\varepsilon)\), there are differences between the Lagrange spectrum and the Markov spectrum.
- This conclusion is derived by studying the Hausdorff dimension of the parts of the Markov spectrum and the Lagrange spectrum close to \(3\), especially by estimating the lower bound of the Hausdorff dimension of \((M\setminus L)\cap(3, 3+\varepsilon)\).
### Main results
- **Theorem 1.1**: For any \(\varepsilon>0\), we have
\[
M\cap(3, 3 +\varepsilon)\neq L\cap(3, 3+\varepsilon)
\]
Therefore,
\[
\inf(M\setminus L)=3
\]
- **Theorem 1.2**: For a sufficiently small \(\varepsilon>0\), we have
\[
\dim_H((M\setminus L)\cap(3, 3 +\varepsilon))\geq\frac{W(e^{c_0|\log\varepsilon|})}{|\log\varepsilon|}-O\left(\frac{\log|\log\varepsilon|}{|\log\varepsilon|^2}\right)
\]
where \(c_0 = -\log\log\left(\frac{3+\sqrt{5}}{2}\right)\approx 0.0383\), and \(W(x)\) is the Lambert \(W\) function.
### Method overview
To construct elements in \(M\setminus L\), the authors use a method based on non - semisymmetric words, combined with local uniqueness and self - replicating properties. Through this method, they are able to prove that there are many sequences belonging to \(M\setminus L\) near the value \(3\), and the Hausdorff dimension of these sequences has a specific lower bound.
### Conclusion
This paper solves the problem of the behavior of the Lagrange spectrum and the Markov spectrum near the value \(3\), reveals the significant differences between them, and provides new tools and methods for future research.