A Diophantine duality applied to the KAM and Nekhoroshev theorems (Q2435083)

Let \(\alpha = (1,\alpha_1,\ldots,\alpha_{n-1}) \in \mathbb{R}^n\), \(|\alpha_j| \leq 1\), \(n>1\) be a vector with \(d\) \(\mathbb{Z}\)-independent entries. Two left-continuous, non-decreasing functions \(\psi_k\) and \(\psi'_k\) that characterize the quantitative Diophantine's properties of \(\alpha\) are: \(\psi_k (Q) = \inf \big\{ K>0:\) there exist \(k\) linearly independent vectors \(x=(x_0,\ldots,x_{n-1})\in\mathbb{Z}^n\) with \(\alpha \!\cdot\! x \not= 0, |x|\leq K, |x_0\alpha-x|\leq Q^{-1} \big\}\) and \(\psi_k' (Q) = \inf \big\{ K>0:\) there exist \(k\) linearly independent vectors \(x=(x_0,\ldots,x_{n-1}) \in \mathbb{Z}^n\) with \(|\alpha \!\cdot\! x| \leq K^{-1}, |x|\leq Q \big\}\), where \(Q\geq 1\), \(\alpha \!\cdot\! x\) denotes the Euclidean scalar product, \(1 \leq k \leq d\). In Theorem 2.1, the authors prove that \(\psi\) and \(\psi'\) are equivalent: \(c_1 \psi'_{d+1-k}(c_2Q) \leq \psi_k(Q) \leq c_3 \psi'_{d+1-k}(c_4Q)\) for certain explicitly found constants \(c_j\). As a first application, consider real-analytic vector fields \(X = X_\alpha + P\) on \(\mathbb{T}_s^n = \{ \theta \in \mathbb{C}^n/\mathbb{Z}^n:\,|\mathrm{Im}(\theta)|<s\}\), \(s\) being an arbitrary positive number, \(X_\alpha=\alpha\) being a constant vector field, while \(P\) is a small perturbation, i.e., \(|P|_{\mathbb{T}_s^n}<\varepsilon\). Theorem 3.1 states the following: for each \(s>0\), there exists a real-analytic embedding \(\Phi:\mathbb{T}_{s/2}^n \rightarrow \mathbb{T}_s^n\), such that \(\Phi^*X=X_\alpha+N+R\), where the vector fields \(X\) and \(N\) commute, the reminder \(R\) is small, \(|R|<\varepsilon \exp (-c_5sQ(\varepsilon))\), and \(\Phi\) is close to the identity map \(|\Phi -\mathrm{Id}|<c_6 Q(\varepsilon)^{-1}\). The function \(Q(\varepsilon)\) is constructed with the help of \(\psi_k\) and \(\psi'_k\) to increase with \(\varepsilon^{-1}\), and \(c_j\) are always explicit constants. The second application (Theorem 3.3) is: if \(d=n\) and if \(Q(\varepsilon)\) satisfies some stronger condition, then there exists a unique constant \(\beta \in \mathbb{C}^n\) as well as a real-analytic embedding \(\Phi:\mathbb{T}_{s/2}^n \rightarrow \mathbb{T}_s^n\), also close to the identity, such that \(\Phi^*(X+X_\beta)=X_\alpha\). In other words, the modified vector field \(X_\alpha + P + X_\beta\) can be analytically conjugated to \(X_\alpha\). Combining these last two theorems gives the classical KAM theorem for constant vector fields, proved by V. Arnold and J. Moser. They guarantee stability for infinite time but not for the original vector field.