Generic Poincaré-Bendixson Theorem for singularly perturbed monotone systems with respect to cones of rank-2

First, we consider the solutions on the invariant manifold \mathbb{C}_{\epsilon} satisfying

For brevity, we only focus on the x-direction on the invariant manifold \mathbb{C}_{\epsilon}, since y=h_{\epsilon}(x). Clearly, the limiting equation of (Proof) is (?) as \epsilon approaches zero. For system (?), the matrix U_{f_{0}}^{pq}(t) satisfies U_{f_{0}}^{pq}(t)[C\backslash \{0\}]\subset \text{Int }C for all p,q\in K_{0} and t>0. By the continuity of D_{\epsilon} and h_{\epsilon}(x) at \epsilon=0, we can pick an \epsilon_{2}<\epsilon_{1} small enough such that U_{f_{0}}^{pq}(t) satisfies U_{f_{0}}^{pq}(t)[C\backslash \{0\}]\subset \text{Int }C for p,q\in K_{\epsilon} and \epsilon \in (0,\epsilon_{2}), where K_{\epsilon} is the projection of \mathbb{C}_{\epsilon} \cap D_{\epsilon} to the x-axis. Note also that D_{\epsilon} is positively invariant under (Proof) and \mathbb{C}_{\epsilon} is an invariant manifold. Then, K_{\epsilon} is positively invariant under the flow \psi^{\epsilon}_{t} of (Proof). Applying Lemma 3.1, we obtain that there exist an \epsilon_3 \leq \epsilon_2 and some t_{*}>0 such that for each \epsilon \in (0,\epsilon_3), system (Proof) is C-cooperative for t\geq t_*.

By [Sanchez09, Proposition 1], it is clear that the flow of system (Proof) satisfies the assumption (FWW) (see [FWW19, p.4]) for t\geq t_*. Then [FWW19, Theorem 5.3] implies that there exists an open and dense subset M_{\epsilon}\subset K_{\epsilon} such that for any x\in M_{\epsilon}, the \omega-limit set \omega(x) containing no equilibrium is a single closed orbit.

It follows from the Lemma 7 in [WS08] that there exist \epsilon_4>0 and d\in (0,\delta_{0}) such that, for each \epsilon \in (0,\epsilon_4), (x_0,y_0) \in W^{s}(\mathbb{C}_{\epsilon}) whenever (x_0,y_0)\in D_{\epsilon} satisfies |y_{0} - h_{\epsilon}(x_{0})| < d. Moreover, Lemma 3.2 3) guarantees that (x_0,y_0)\in W^{s}_{\epsilon, d}(\xi), where \xi \in K_{\epsilon} and W^{s}_{\epsilon, d}(\xi) is the stable fiber restricted to the d neighborhood of \mathbb{C}_{\epsilon}.

Let N_{\epsilon}=\{(x,y)\in D_{\epsilon} : |y - h_{\epsilon}(x)| < d \}. In the following, we will show that the set \underset{\xi \in M_{\epsilon}}{\cup} W^{s}_{\epsilon, d}(\xi) is open and dense in N_{\epsilon}. By Lemma 3.2 6), the stable fiber can be characterized as W^{s}_{\epsilon, d}(\xi) = T_{\epsilon, d}(\xi, L_d), where L_d=\{ z\in \mathbb{R}^{m} : |z|<d \}. In other words, \underset{\xi \in M_{\epsilon}}{\cup} W^{s}_{\epsilon, d}(\xi)=T_{\epsilon, d}(M_{\epsilon},L_d). For any (x_0,y_0)\in N_{\epsilon}, there exists (\xi_0,\eta_0)\in K_{\epsilon} \times L_d such that (x_0,y_0)=T_{\epsilon, d}(\xi_0, \eta_{0}). Since M_{\epsilon} is dense in K_{\epsilon}, there exists a sequence \{ \xi_n\}\subset M_{\epsilon} such that \xi_n \to \xi_0. Let \eta_n= \eta_0 and (x_n,y_n)=T_{\epsilon, d}(\xi_n,\eta_n) for n=1,2,\cdots, then (\xi_n,\eta_n)\to (\xi_0,\eta_0) and (x_n,y_n)\in T_{\epsilon, d}(M_{\epsilon},L_d). Since the mapping T_{\epsilon, d} is continuous, one has (x_n,y_n)\to (x_0,y_0) as n\to \infty. Thus, the set T_{\epsilon, d}(M_{\epsilon},L_d) is dense in N_{\epsilon}. In order to prove the openess of T_{\epsilon, d}(M_{\epsilon},L_d), we only need to show the inverse of the mapping T_{\epsilon, d} is continuous, because both M_{\epsilon} and L_d are open sets. Clearly, the mapping T_{\epsilon, d} is both injective and surjective. And by the proof of the Claim 3.7 in [Saka90], the inverse of T_{\epsilon, d} is continuous (In fact, there exists a mapping \phi^*, see [Saka90, Lemma 3.9], such that \xi_0=x_0-\phi^*(0) and \eta_0=y_0-h_{\epsilon}(x_{0}) hold for (x_0,y_0)\in N_{\epsilon}). Thus, the set T_{\epsilon, d}(M_{\epsilon},L_d) is open and dense in N_{\epsilon}.

Moreover, there exist some positive \tau_{0} and an \epsilon_5 < \epsilon_4 such that |y(\tau_{0})-h_{\epsilon}(x(\tau_{0}))| < d for all \epsilon \in (0,\epsilon_5), whenever (x(\tau),y(\tau)) is the solution to (?) with (x(0),y(0))=(x_{0},y_{0})\in D_{\epsilon}. This implies that (x(\tau_{0}),y(\tau_{0})) \in N_{\epsilon}. The corresponding proof is similar as Lemma 8 in [WS08] with the positive invariance of D_{\epsilon} (In fact, |y(\tau)-h_{\epsilon}(x(\tau))| \leq |y(\tau)-h_{0}(x(\tau))| + |h_{0}(x(\tau)) - h_{\epsilon}(x(\tau))|, the assumption A5 and Lemma 3.2 1) imply that there exist a \tau_{0}>0 and \epsilon_5 < \epsilon_4 such that |y(\tau_{0})-h_{0}(x(\tau_{0}))|<\dfrac{d}{2} and |h_{0}(x(\tau_{0})) - h_{\epsilon}(x(\tau_{0}))|<\dfrac{d}{2}). Let \phi^{\epsilon}_{\tau} be the flow of (?), and F(D_{\epsilon}) = \{ \phi^{\epsilon}_{\tau_0}((x,y)):(x,y)\in D_{\epsilon} \} \subset N_{\epsilon}. Since \phi^{\epsilon}_{\tau_0} is diffeomorphism, the set G_{\epsilon}=F(\text{Int } D_{\epsilon})\cap T_{\epsilon, d}(M_{\epsilon},L_d) is open and dense in F(D_{\epsilon}). So, F^{-1}(G_{\epsilon}) is open and dense in D_{\epsilon}. Let \mathcal{P}_{\epsilon}=F^{-1}(G_{\epsilon}). For (x,y)\in \mathcal{P}_{\epsilon}, the fact that \phi^{\epsilon}_{\tau_0}((x,y))\in T_{\epsilon, d}(M_{\epsilon},L_d) implies the \omega-limit set \omega((x,y)) containing no equilibrium is a single closed orbit by the asymptotic phase property.

We have completed the proof of Theorem 2.3 by taking \epsilon^{*} = \min \{\epsilon_3,\epsilon_5 \}.

Pick t_*>0, we consider the matrices

and the solution U_{G}^{pq}(t) of

for any p,q\in W.

We first prove the positiveness of U_{G}^{pq}(t) for t\in [t_*, 2t_*], i.e., U_{G}^{pq}(t)[C\backslash \{0\}]\subset \text{Int }C for t\in [t_*, 2t_*]. Since system (?) is C-cooperative, there exists a \delta_{1}>0 such that B_{\delta_{1}}(U_{F}^{pq}(t)v)\subset \text{Int }C for all v\in C\backslash \{0\} with |v|=1. A positive \delta can be found such that if \parallel F(z)-G(z)\parallel +\parallel DF(z)-DG(z)\parallel<\delta then \parallel U_{F}^{pq}(t)-U_{G}^{pq}(t) \parallel \leq \frac{\delta_{1}}{2} hold for t\in [t_{*},2t_{*}].

In fact, by the definition of flow, we have

Then,

Since F and G are C^{1} and t_{*} \leq t \leq 2t_{*}, there exists some M > 0 such that \parallel F(\phi_{s}(z))-F(\psi_{s}(z))\parallel \leq M \parallel \phi_{s}(z)-\psi_{s}(z)\parallel. Then \parallel \phi_{t}(z)-\psi_{t}(z)\parallel \leq \int^{t}_{0}M \parallel \phi_{s}(z)-\psi_{s}(z)\parallel ds + \delta t.

Using the Gronwall’s inequality, we obtain

Hence, for s\in [0,1],

Note that U_{F}^{pq}(t) and U_{G}^{pq}(t) are the solutions of (?) and (Proof), respectively, then

Using the Gronwall’s inequality again, it follows that

There exist N>0 and L>0 such that \parallel U_{G}^{pq}(s) \parallel \leq N and \parallel A_{F}^{pq}(s) \parallel \leq L for all t\in [t_*, 2t_*]. Meanwhile,

Since F is C^{1} and \parallel [s\phi_{t}(p)+(1-s)\phi_{t}(q)]-[s\psi_{t}(p)+(1-s)\psi_{t}(q)]\parallel \leq \delta 2t_{*} e^{M2t_{*}}, one can choose \delta < \frac{1}{2}\cdot \frac{\delta_{1}}{2}\cdot \frac{1}{2Nt_{*}e^{2t_{*}L}} small enough such that \parallel DF(s\phi_{t}(p)+(1-s)\phi_{t}(q))-DF(s\psi_{t}(p)+(1-s)\psi_{t}(q))\parallel < \frac{1}{2}\cdot \frac{\delta_{1}}{2}\cdot \frac{1}{2Nt_{*}e^{2t_{*}L}}. So, \parallel A_{F}^{pq}(t)-A_{G}^{pq}(t)\parallel \leq \frac{\delta_{1}}{2}\cdot \frac{1}{2Nt_{*}e^{2t_{*}L}}, which implies that \parallel U_{F}^{pq}(t)-U_{G}^{pq}(t) \parallel \leq \frac{\delta_{1}}{2}. Thus, we obtain that U_{G}^{pq}(t)[C\backslash \{0\}]\subset \text{Int }C for t\in [t_{*},2t_{*}] and p,q\in W.

For t>2t_*, let us write t=t_0+kt_* with t_0\in [t_*,2t_*). Define t_j=jt_*, p_j=\psi_{t_j}(p), q_j=\psi_{t_j}(q) with j=1,2,\cdots,k. It is clear that p_j, q_j\in W if W is positively invariant. Since \psi_{t}(p)-\psi_{t}(q)=U_{G}^{pq}(t)(p-q), we obtain that

By the preceding proof, U_{G}^{pq}(t)[C\backslash \{0\}]\subset \text{Int }C for t>2t_*.

Thus, we have proved that the system (?) is C-cooperative for t\geq t_*.