A solution to a problem of Cassels and Diophantine properties of cubic numbers (Q640747)

The author proves that almost any pair of real numbers \(\alpha,\beta\) satisfies the following inhomogeneous uniform version of \textit{Littlewood's conjecture}: \[ \forall \gamma,\delta\in \mathbb R, \;\lim\inf_{|n|\to\infty} |n|<n\alpha-\gamma><n\beta-\delta>=0, \tag{C1} \] where \(<.>\) denotes the distance from the nearest integer. The existence of even a single pair that satisfy (C1) solves a problem of Cassels from the 50's. The author then proves that if \(1,\alpha,\beta\) span a totally real cubic number field, then \(\alpha,\beta\) satisfy (C1). This generalizes a result of Cassels and Swinnerton-Dyer, which says that such pairs satisfy Littlewood's conjecture. It is further shown that if \(\alpha,\beta\) are two real numbers such that \(1,\alpha,\beta\) are linearly independent over \(\mathbb Q\), they cannot satisfy (C1). The results are derived from rigidity results concerning hyperbolic actions of higher rank commutative groups on homogeneous spaces. \textbf{Definitions-Notations:} let \(X_d\) denote the space of \(d\)-dimensional unimodular lattices in \(\mathbb R^d\) and \(Y_d\) the space of translates of such lattices. Points of \(Y_d\) are called \textit{grids}; hence for \(x\in X_d, v\in \mathbb R^d, y=x+v\in Y_d\) is the grid obtained by translating the lattice \(x\) by the vector \(v\). Let us denote by \(\pi\) the natural projection \(X_d\overset{\pi}{\to} Y_d, {x+v} x\). For each \(x\in X_d\) the fiber \(\pi^{-1}(x)\) in \(Y_d\) is identified with the torus \(\mathbb R^d/x\). Let \(N: \mathbb R^d\to \mathbb R\) denote the function \(N(w)=\prod_1^dw_i\). For a grid \(y\in Y_d\), let us define the \textit{product set} of \(y\) to be \(P(y):=\{N(w) : w\in y\},\) and \(N(y):=\inf \{|N(w)| : w\in y \}\). The inhomogeneous minimum of a lattice \(x\in X_d\) is defined by \(\mu(x):=\sup\{ N(y) : y\in \pi^{-1} (x)\}.\) \textbf{Results:} the author proves: For \(d\geq 3\), almost any lattice \(x\in X_d\) (with respect to the natural probability measure) satisfies \(\mu(x)=0\). Let \(v\in\mathbb R^{d-1}\) be a column vector. Denote the matrices \(h_v= \begin{pmatrix} I_{d-1}&v \\ 0&1 \end{pmatrix}\) and \(g_v= \begin{pmatrix} 1&v^t \\ 0&I_{d-1} \end{pmatrix}\), where \(I_{d-1}\) denotes the identity matrix of dimension \(d-1\) and the \(0\)'s denote the corresponding trivial vectors. Let \(x_v,z_v\in X_d\) denote the lattices spanned by the columns of \(h_v\) and \(g_v\) respectively. For \(\gamma\in\mathbb R\), denote by \(<\gamma>\) the distance from \(\gamma\) to the nearest integer. The statements \[ \begin{aligned}& (C1): \;\;\forall \overset{\to}{\gamma}\in \mathbb R^{d-1}\;\;\lim\inf_{|n|\to\infty} |n|\prod_1^{d-1}<nv_i-\gamma>\;=0,\\ & (C2): \;\;\forall \gamma\in \mathbb R \;\;\lim\inf_{\prod|n_i|\to\infty}\prod_1^{d-1}|n_i| <|\sum_1^{d-1}<n_iv_i-\gamma>\;=0, \end{aligned} \] imply that \(\mu(x_v)=0\) and \(\mu(z_v)=0\). A vector \(v\in\mathbb R^{d-1}\) is said to have the property \(C1\) (resp. \(C2\)) if statement \(C1\) (resp \(C2\)) is satisfied. The author then proves the following diophantine approximations theorems. Let \(d\geq 3\): Almost all \(v\in \mathbb R^{d-1}\)(with respect to Lebesgue measure) has property \(C1\) and \(C2\). In particular, \(\mu(x_v)=\mu(z_v)=0,\) nonetheless, if \(\dim\text{ span}_{\mathbb Q}\{1,v_1,\dots,v_{d-1}\}\leq 2\) then \(\mu(x_v),\mu(z_v)\) are positive. And the most important theorem of the article: If \(1,\alpha,\beta\) form a basis for a totally real cubic number field, then \[ \forall \gamma,\delta\in \mathbb R, \lim\inf_{|n|\to\infty} |n|<n\alpha-\gamma><n\beta-\delta>\;=0, \] \[ \forall \gamma\in \mathbb R, \lim\inf_{|nm|\to\infty} |nm|<n\alpha+m\beta-\gamma>\;=0, \] that is the vector \((\alpha,\beta)^t\) has properties \(C1\) and \(C2\).