Abelian integrals and limit cycles (Q5920602)

This paper deals with perturbations of the Hamiltonian system \[ X_{(\overline{\lambda},\varepsilon)}: \left\{ \begin{matrix} \dot{x}=- \frac{\partial H}{\partial y}+\varepsilon f, \vspace{0.2cm} \\ \displaystyle \dot{y}= \frac{\partial H}{\partial x}+\varepsilon g, \end{matrix} \right. \tag{1} \] where \(H(x,y)\), \(f(x,y,\overline{\lambda},\varepsilon)\) and \(g(x,y,\overline{\lambda},\varepsilon)\) are \(\mathcal{C}^{\infty}\) functions, \(H(x,y)\) is assumed only to have Morse-type critical points, \(\varepsilon\) is considered to take small positive values and \(\overline{\lambda} \in \mathbb{R}^p\), \(\lambda=(\overline{\lambda},\varepsilon)\). The results of the paper are related to the problem of studying the number of limit cycles that can be found when restricting \(\overline{\lambda}\) to a compact set, \(\varepsilon\) near zero and \((x,y)\) to some compact domain containing one or more periodic annuli. Taking a first-order approximation with respect to \(\varepsilon\), this problem requires the study of the number of zeroes (in \(h\)) of the function \[ I_{\overline{\lambda}} (h)=\int_{\Gamma_h} f \, dy-g \, dx, \] where \(\Gamma_h\) is the \(h\)-level curve of the Hamiltonian \(H\) representing one of the closed curves in the periodic annulus. When \(H\), \(f\) and \(g\) are polynomials, the function \(I_{\overline{\lambda}}(h)\) is called an Abelian integral, and the same name is used in this more general context. We refer the reader to [\textit{R. Roussarie}, Bifurcation of planar vector fields and Hilbert's sixteenth problem. Progress in Mathematics (Boston, Mass.). 164. (Basel): Birkhäuser. (1998; Zbl 0898.58039)] and [\textit{Y. Ilyashenko}, Bull. Am. Math. Soc., 39, 301--354 (2002; Zbl 1004.34017)] and the references therein for the complete definitions, history and settings of the weak or tangential Hilbert's 16th problem. The paper is not devoted to this problem but to the exact information given by the zero-set of \(I_{\overline{\lambda}}(h)\) in relation with the set of limit cycles of system (1). It is shown that there are examples of limit cycles bifurcating from the boundary of a periodic annulus which cannot be detected as zeroes of the corresponding Abelian integral. A result, proved in the aforementioned book, is that only the limit cycles that, when \(\varepsilon \to 0\), can be found in a neighborhood of the closure of the periodic annulus need to be studied. When, for instance, this closure is a nondegenerate singularity, the zeroes of the function \(I_{\overline{\lambda}}(h)\) are shown to unfold in terms of the parameters as elementary catastrophes and, moreover, results like implicit function theorem allow one to prove that the limit cycles of (1), for \(\varepsilon\) near zero and in a compact neighborhood of the point, behave exactly as the zeroes of \(I_{\overline{\lambda}}(h)\). This fact is proved using that the first approximation in \(\varepsilon\) to the return map is exactly given by \(I_{\overline{\lambda}}(h)\) and that the return map is well extended over the nondegenerate singularity. Therefore, the limit cycles of (1) near a nondegenerate singular point can be ``seen'' by the zeroes of the function \(I_{\overline{\lambda}}(h)\). However, this fact is not so obvious when the boundary of the periodic annulus contains singularities. To start with, the return map is not defined over a boundary formed by a polycycle, although a continuous extension can be considered. However, this extension is not differentiable enough so as to permit the use of implicit function theorem. In this case, one can think about which asymptotic expansion the function \(I_{\overline{\lambda}}(h)\) exhibits and, within some technicalities in the way of choosing \(H\), the expansion in monomials of the form \(h^n\) and \(h^n \log h\), with \(n \in \mathbb{N}\), is encountered. In the case the boundary is a homoclinic loop, R. Roussarie [\textit{R. Roussarie}, Bol. Soc. Bras. Mat. 17, No. 1, 67--101 (1986; Zbl 0628.34032)] proved that the configuration of the limit cycles of \(X_{(\overline{\lambda},\varepsilon)}\) is completely analogous to the configuration of the zeroes of \(I_{\overline{\lambda}}(h)\). In this paper, it is shown that for a \(2\)-saddle cycle, this result is no longer valid, even keeping one of the connections unbroken. The reason is that the asymptotic development of the equations describing the limit cycles contains monomials of the form \(h^n\), \(h^n \log h\) and \(h^n \log^m h\) with \(2 \leq m \leq n\) and these last terms cannot be encountered in the related Abelian integral. It is shown that within generic families of systems \(X_{(\overline{\lambda},\varepsilon)}\) depending on \(4\) parameters one can find vector fields with \(4\) limit cycles close to a \(2\)-saddle cycle while the related Abelian integral only depends on \(3\) parameters and, thus, can have at most three simple zeroes. It is shown in this way that the study of the limit cycles near a \(2\)-saddle cycle with one unbroken connection is much more complicated than the saddle loop case. The paper starts with a complete and useful introduction and generalities about the weak Hilbert's 16th problem which settles down the context. A summary of the results, taking into account the saddle loop including the definitions of asymptotic scale of functions and simple asymptotic scale deformation, is next given. These definitions correspond to the study of the unfolding of system (1) and the way the asymptotic expansion of \(I_{\overline{\lambda}}(h)\) is treated to prove the mentioned results. Moreover, these definitions allow one to give the precise definition of the numbers \(\text{cod}\, I_{\overline{\lambda}}\) and \(\text{cod}\, X_{\lambda}\). These two notions of codimension are shown to satisfy the following relation: Let \(\text{cod}\, I_{\overline{\lambda}} = q\). If \(q=2p\), then \(\text{cod}\, X_{\lambda} = 2p + p (p-1)/2\). If \(q=2p+1\), then \(\text{cod}\, X_{\lambda} = 2p + 1+ p (p+1)/2\). One of the main results of the paper states that: if \(\text{cod}\, X_{\lambda}\) is finite, then the cyclicity of \(X_{\lambda}\) in a neighborhood of a \(2\)-saddle cycle is lower or equal to \(\text{cod}\, X_{\lambda}\). In the view of these results, we see that when \(\text{cod}\, I_{\overline{\lambda}} = 3\), then \(\text{cod}\, X_{\lambda}=4\) and a section is devoted to study generic unfoldings with one unbroken connection of the \(2\)-saddle cycle of codimension \(4\) and to show the existence of such unfoldings with four simple limit cycles, while \(\text{cod}\, I_{\overline{\lambda}} = 3\) and thus it has at most three simple zeroes. The last section is concerned with unfoldings of a \(k\)-saddle cycle with \(k\) parameters also showing the existence of limit cycles escaping from the validity domain of the Abelian integral.