A direct proof of the Nekhoroshev theorem for nearly integrable symplectic maps (Q707502)

The renowned Nekhoroshev stability theorem [\textit{N. N. Nekhoroshev}, Russ. Math. Surv. 32, 1--65 (1977; Zbl 0389.70028)] for nearly integrable hamiltonian flows provides conditions under which the motions of an integrable hamiltonian system and those of the same system subject to a perturbation of norm \(\varepsilon \) remain close for a time of order \(t \simeq 1/ \varepsilon^k\). A corresponding result, under conditions which extend naturally those holding for flows, was conjectured by Nekhoroshev [loc. cit.] for nearly integrable sysmplectic maps. Such a result was then proved by \textit{S. Kuksin} [Russ. J. Math. Phys. 1, 191--207 (1993; Zbl 0931.37030)] by constructing a nearly integrable (non autonomous) hamiltonian flow which interpolates the map. This paper provides a direct proof of Nekhoroshev theorem for maps, without resorting to auxiliary continuous flows. Consider a \(2n\) dimensional symplectic manifold \(M\) and the symplectic map \(\mu : M_0 \to M\) generated by the function \(S(I,\phi) := I \cdot \phi + h (I) + \varepsilon f (I,\phi) \) and defined on a neighbourhood \(M_0\) of the origin, \(M_0 = B \times {\mathbb T}^n\). This corresponds to the map \(\mu : (I,\phi) \mapsto (\widehat{I} , \widehat{\phi} ) \), defined for \(I \in B \subset {\mathbb R}^n\), with \[ I \;= \;\widehat{I} \, + \, \varepsilon {\partial f \over \partial \phi } (\widehat{I} , \phi ) \;\;; \;\;\widehat{\phi } \;= \;\phi \, + \, {\partial h \over \partial I} (\widehat{I}) \, + \, \varepsilon {\partial f \over \partial I } (\widehat{I} , \phi ) \;. \] Write \(\widetilde{B} (a,d,\epsilon) := \{ I \in B \;: \;\text{ dist} (I,\partial B) > 2 d \epsilon^a \}\). Then the following is the main result of the paper. Theorem. Write \((I(k),\phi(k)) := \mu^k (I(0),\phi (0))\). If \(h\) is convex, then there exist positive constants \(\epsilon_0, a, b , k_0 , d_0\) such that for any \(\varepsilon < \epsilon_0\) and for any \(I(0), \phi(0) \in [\widetilde{B} (a,d_0 , \epsilon_0 )] \times {\mathbb T}^n \), and for any \(k \in {\mathbb Z}\) satisfying \( | k| \leq k_0 \exp [ (\epsilon_0 / \varepsilon )^b ] \), it results \(| I(k) - I(0) | \leq d_0 \varepsilon^a\). This theorem differs from the result conjectured by Nekhoroshev in that convexity of \(h\) is assumed (as opposed to \(p\)-steepness; convex functions are also \(p\)-steep and provide a significant class of Nekhoroshev stable Hamiltonians, see [\textit{G. Benettin, L. Galgani} and \textit{A. Giorgilli}, Celest. Mech. 37, 1--25 (1985; Zbl 0602.58022)]). It is observed in the paper that the result also extends to the \(p\)-steep case, but cannot be extended to the case where \(h\) is quasi-convex, as shown by the case \(S(I,\phi) = I \phi + 2 \pi I + \varepsilon \cos (\phi )\). The author presents a discussion of differences of his results and Kuksin's result, pointing out advantages and disadvantages -- in particular, for what concerns estimates of the limiting parameters -- in the two approaches. The proof is based on a lemma of indipendent interest concerning the normalization (in the sense of Poincaré-Birkhoff normal forms) of a symplectic map. Consider a near-identity symplectic map \(\sigma : (I,\phi) \mapsto (I'\phi')\) with generator \(\Sigma (I,\phi) = I \cdot \phi + \varepsilon \chi (I,\phi)\); this is defined by \[ I \;= \;I' \, + \, \varepsilon {\partial \chi \over \partial \phi } (I' , \phi ) \;\;; \;\;\phi' \;= \;\phi \, + \, \varepsilon {\partial \chi \over \partial I } (I' , \phi ) \;. \] Under this, the symplectic map \(\mu\) is mapped to a symplectic map \(\mu' := \sigma^{-1} \circ \mu \circ \sigma \); this will correspond to a generating function \(S' (I,\phi) := I \cdot \phi + h (I) + \varepsilon f' (I,\phi) \). Then Guzzo proves that: Lemma. Let \(S,\chi,\mu\) and \(\sigma\) be as above; let \(\chi\) be defined and analytic in \(B \times {\mathbb T}^n\). For small \(\varepsilon\), it results \( S' (I,\phi) = I \cdot \phi + h (I) + \varepsilon f (I,\phi) + \varepsilon [ \chi (I,\phi ) - \chi ( I,\phi + \omega (I) )] + \varepsilon^2 R (I,\phi ) \); where \(\omega (I) := (\partial h / \partial I) (I) \) and the remainder \(R (I,\phi)\) is real analytic and uniformly bounded in \(\varepsilon\) in the domain \(B \times {\mathbb T}^n\).