A proof of Perko's conjectures for the Bogdanov-Takens system (Q2637627): Difference between revisions

The authors consider the following expression of the Bogdanov-Takens system \[ \dot{x} \, = \, y, \quad \dot{y} \, = \, -m^2+by+x^2+xy \] with \(b,m \in \mathbb{R}\) and \(m>0\), and deal with its bifurcation diagram. This system has two finite critical points, \((\pm m,0)\) a saddle and a focus. In [\textit{C. Li} et al., Can. Math. Bull. 33, No. 1, 84--92 (1990; Zbl 0706.34026)] it is proved that this system has at most one limit cycle and that when it exists it is hyperbolic and unstable. Quoting the manuscript, it is also known that the limit cycle exists if and only if \(b^{*}(m) < b < m\), for an unknown function \(b^{*}(m)\). Fixing \(m\) and decreasing \(b\), a unique unstable limit cycle is born via Hopf bifurcation for \(b=m\), increases diminishing \(b\) and disappears in a saddle-loop connection for \(b=b^{*}(m)\). \textit{L. M. Perko} [SIAM J. Appl. Math. 52, No. 4, 1172--1192 (1992; Zbl 0758.58026)] gave several analytic properties about the saddle-loop bifurcation curve \(b=b^{*}(m)\) and established three conjectures. Two of these conjectures are proved in the present paper. The authors write Perko's conjectures in a convenient way. In Section 6 they show that this way is equivalent to Perko's. The main results of the paper are the following: (i) For \(m\) large enough, \(b^{*}(m) \, = \, m-1 + o \left(1/m\right)\). (ii) It holds that \[ \max\left( \frac{5m}{7}, m-1\right) \, < \, b^{*}(m) \, < \, \min\left( \frac{\left(5+\frac{37}{12}m\right)m}{7+\frac{37}{12}m}, m-1+\frac{25}{7m} \right). \] Statement \text{(ii)} is a significant improvement of the global lower and upper bounds stated in Perko's conjecture. The proof of these results is based on an adaptation of the method introduced by the authors in [Nonlinearity 23, No. 12, 2977--3001 (2010; Zbl 1215.34045)]. The idea is to construct suitable negatively or positively invariant regions and then to use Poincaré-Bendixson Theorem. For \(m\) small, the authors consider a self-intersecting algebraic curve whose vertex is on the saddle point and whose branches approximate its separatrices. The computational difficulties in order to choose a convenient algebraic curve and to show that it provides the boundary of a Poincaré-Bendixson region are shortened by means of a convenient rational parametrization of the straight line \(7b-5m=0\). For \(m\) large, a piecewise algebraic closed curve is considered. This piecewise closed curve without contact also approximates the separatrices at the saddle point and takes into account the region where both separatrices touch for the first time the negative \(x\)-axis. The ideas behind the proof also provide a way to tackle the difficult problem of bounding bifurcation curves. It is remarkable that the bounds are analytically proved, contrary to the usual situation in which the bifurcation curves are numerically found.