Abstract
In this paper we treat a particular class of planar Filippov systems which consist of two smooth systems that are separated by a discontinuity boundary. In such systems one vector field undergoes a saddle-node bifurcation while the other vector field is transversal to the boundary. The boundary-saddle-node (BSN) bifurcation occurs at a critical value when the saddle-node point is located on the discontinuity boundary. We derive a local topological normal form for the BSN bifurcation and study its local dynamics by applying the classical Filippov’s convex method and a novel regularization approach. In fact, by the regularization approach a given Filippov system is approximated by a piecewise-smooth continuous system. Moreover, the regularization process produces a singular perturbation problem where the original discontinuous set becomes a center manifold. Thus, the regularization enables us to make use of the established theories for continuous systems and slow-fast systems to study the local behavior around the BSN bifurcation.
1. Introduction
The study of Filippov system was motivated by its considerable applications in mechanical systems exhibiting dry friction [1–4], biological systems [5–7], and control systems [8–10]. In general, Filippov systems, or differential equations with discontinuous right-hand side, model physical processes which experience abrupt transitions between different modes. Modelling in such way we often idealize the transition to be instantaneous. The consequence of doing so is that the established discontinuous model focuses on the overall dynamics of the entire physical process while slightly ignoring the detailed dynamics in the transition stage. An efficient method to learn the dynamics in the transition stage is modelling this process in a small time scale such that the dynamics of one mode smoothly or continuously changes to the other mode. For a given Filippov system this can be realized by a smooth or continuous approximation which removes the discontinuities. This method is referred to as regularization. As a consequence, this enables us to make use of the established theories for smooth or continuous systems to study the dynamical properties of Filippov systems.
Different ways have been proposed to regularize a Filippov system; see [11–13]. In this paper we apply the regularization of [12] to a particular class of planar Filippov systems. Such systems consist of two smooth vector fields that are separated by a smooth discontinuity boundary. The vector field on one side undergoes a standard saddle-node bifurcation, while the vector field on the other side intersects the boundary transversally. Here a “standard saddle-node bifurcation” means a saddle-node bifurcation in smooth systems. We refer for a formal definition of Filippov system and the regularization approach of [12] to Sections 2.1 and 2.2, respectively.
Our aim of this work is to describe the local dynamics in the neighborhood of the codimension- boundary-saddle-node (BSN) bifurcation where two equilibria of one smooth vector field go through a saddle-node bifurcation while they lie on the boundary. We treat this particular class of Filippov systems by applying both Filippov’s convex method and the regularization approach to the BSN point to obtain its local bifurcation diagram, to understand how all the codimension- bifurcations interact. As we shall see in Section 3 the BSN point studied by Filippov’s convex method acts as organizing center for three families of codimension- bifurcations appearing in its neighborhood, which are standard saddle-node bifurcation, equilibrium transition, and nonsmooth fold bifurcation. However, after regularization, some of the discontinuity-induced bifurcations disappear, such as the equilibrium transition. The unique bifurcation that occurs in the regularized system is the saddle-node-like bifurcation; see Section 2.2. Here the extension “like” means that this bifurcation is not the standard one as known from smooth systems. We refer to [15] for a detailed description of the generic bifurcations in piecewise-smooth continuous systems.
We briefly outline the structure of this paper. In the next section we give an overview of Filippov dynamics and the regularization technique of [12]. Moreover, we explain how to transform the regularized system to a singular perturbation problem. In Section 3 we construct a topological normal form for the BSN bifurcation. Subsequently, we study the BSN bifurcation by Filippov’s convex method and obtain its bifurcation diagram and phase portraits in Section 4. After that, in Section 5 we apply the regularization approach to the topological normal form and investigate its local dynamics as a piecewise-smooth continuous system. Finally, in Section 6 we summarize the main results.
2. Preliminaries
In this section we introduce Filippov’s convex method and the regularization approach of [12].
2.1. Filippov’s Convex Method
We first introduce the basic concepts in Filippov systems.
2.1.1. Filippov Dynamics
A triplet defined on a 2-dimensional manifold is called a Filippov system, denoted by . Both and are ( or ) vector fields, which are extendable over a full neighborhood of the discontinuity boundary . The boundary is given as follows: where is a function and has as a regular value. Therefore, is a smooth 1-dimensional submanifold of . As a consequence, separates to two open subsets and :
For any , the dynamics of is given by
According to Filippov’s convex method [16], can be divided into the crossing setand the sliding setwhere denotes the directional derivative of with respect to at the point . That is, .
In particular, the dynamics of a point at the sliding set is defined by a unique convex linear combination of and , which is tangent to at . This dynamics is called the sliding vector field [16], denoted by and given as follows:when . We define when for .
2.1.2. Equilibria and Bifurcations
Here we give a brief description of the discontinuity-induced bifurcations (bifurcations where the discontinuity of the Filippov systems plays an essential role) of equilibria appearing in planar Filippov system.
Definition 1 (equilibria). A point is an ordinary equilibrium of (resp. ) if and (resp., and ). A point is called a boundary equilibrium of (resp. ) if (resp., ). A point is an ordinary (resp., boundary) equilibrium of if it is an ordinary (resp., boundary) equilibrium of or . Equilibria of in and of in are called virtual equilibria.
Definition 2 (pseudoequilibria). A point is a pseudoequilibrium of if is not a boundary equilibrium and . If for , then one calls a virtual pseudoequilibrium of .
When an ordinary equilibrium of collides with the discontinuity boundary , it generically gives either an equilibrium transition (ET) or a nonsmooth fold (NSF) bifurcation; compare [17]. For the ET an ordinary equilibrium collides with a virtual pseudoequilibrium and then becomes a virtual ordinary equilibrium and a pseudoequilibrium; see Figure 1(a). For the NSF bifurcation an ordinary equilibrium collides with a pseudoequilibrium and then they both become virtual; see Figure 1(b).
(a) Equilibrium transition
(b) Nonsmooth fold
The following theorem [14] gives conditions for the occurrence of an equilibrium transition or nonsmooth fold bifurcation in a Filippov system that smoothly depends on a parameter .
Theorem 3 (equilibrium transition and nonsmooth fold bifurcation [14]). Assume that and that is nonsingular. Moreover, assume that for an equilibrium branch of the vector field transversally crosses the discontinuity manifold at ; that is, Finally, assume that the nondegeneracy condition is satisfied. Then, at , there is an equilibrium transition if and there is a nonsmooth fold bifurcation if .
2.1.3. Equivalence between Filippov Systems
Here we define the equivalence that will be used in this work.
Definition 4 (topological equivalence [18]). Two Filippov systems and are topologically equivalent if there is a homeomorphism that sends the orbits of to and the sliding set to .
2.2. Regularization Approach
In this section we introduce the regularization approach of [12] and present its properties about equilibria.
According to [12] piecewise-smooth continuous regularization of the Filippov system is given in the following way. Letbe the extended manifold, where . That is, is obtained by cutting the manifold along and then glueing the two pieces together by adding the “strip” . The regularized vector field is defined on asHere is a strictly increasing, smooth function defined in with , . For simplicity we assume .
In this paper we are interested in local dynamics of the Filippov system. Since is a submanifold of , there are local coordinates in a neighborhood of such that locally . In this case the regularized system can be written as
For the regularized system we have the following theorem about its equilibria and stability type.
Theorem 5. The ordinary equilibria and pseudoequilibria of correspond to equilibria of with the same type of stability. Boundary equilibria of either or (but not both) correspond to equilibria of . Common boundary equilibria of and give rise to a curve of equilibria of in .
2.3. Regularization as a Singular Perturbation Problem
In this section we briefly review singular perturbation theory. Consider a general singular perturbation problem with the slow time systemor equivalently the fast time system, after time rescaling ,where , , .
The slow manifold is the set of equilibria of the fast time system for and is given by Then is said to be normally hyperbolic if for every the matrix has no eigenvalues on the imaginary axis.
For the slow time system (11) is reduced to the following system on :called the reduced problem.
The following result of Fenichel (see [19]) on the preservation of normally hyperbolic invariant manifolds plays a central role in geometric singular perturbation theory.
Theorem 6 (Fenichel’s invariant manifold theorem). If is a compact set and normally hyperbolic, then, for and sufficiently small, there exists a locally invariant manifold close to in the topology. The manifold is diffeomorphic to , and the fast flow on is close to the flow of the reduced equation on .
Now consider the Filippov system on . Recall that given a point we can find a local coordinate system such that is given by and the regularized system is given by (10). We now focus on the region and we perform the scaling transformation . This scaling induces in the systemRescaling time by we getSystem (15) is the slow time system while (16) is the fast time system. The slow manifold is given by
The slow time system (15) for defines a dynamical system on with dynamics given by Accordingly the slow time system (15) for ; that is,is called the reduced problem of the singularly perturbed system.
Comparing the singular perturbation problem (15) and (16) with the original Filippov system , we directly derive the following result.
Lemma 7. Suppose . Then there is a diffeomorphism , having the form , where at , that maps the flow of the sliding vector field to the flow of the reduced problem on .
Remark 8. The slow manifold and the sliding segment of the boundary manifold are homeomorphic and carry the same dynamics but their geometry is different.
3. The Boundary-Saddle-Node Bifurcation
Here we first define the boundary-saddle-node (BSN) bifurcation.
Definition 9 (boundary-saddle-node bifurcation). Consider a planar Filippov system which smoothly depends on parameters . In particular, for the vector field has an equilibrium . Then the Filippov system undergoes a boundary-saddle-node (BSN) bifurcation at when the following conditions hold true: (1)the equilibrium is located on the boundary; that is, ;(2), , and , where is the first component of ;(3) is transversal to at ; that is, ;(4) = 0.
Remark 10. Condition implies that the vector field goes through a saddle-node bifurcation at ; condition defines a degenerate situation for the boundary equilibrium at . From Theorem 3 we know that the system goes through a nonsmooth fold bifurcation for , while it goes through an equilibrium transition for .
Theorem 11 (topological normal form). Consider a general Filippov system which undergoes a BSN bifurcation at . Then at the equilibrium point there exist a local homeomorphism depending on parameters and a smooth invertible reparameterization such that is topologically equivalent to the following system:with where and . All higher order terms in , , and smoothly depend on .
Note that the vector filed undergoes a saddle-node bifurcation at . The parameter determines the passage of the equilibrium of through the boundary .
Proof. The method of proving this theorem is similar to [5]. Thus we omit it here.
4. Bifurcation Diagram and Phase Portraits
System (20a), (20b), (20c), and (20d) undergoes 3 different codimension-1 bifurcations: saddle-node bifurcation, equilibrium transition, and nonsmooth fold bifurcation. Next we will only describe these three bifurcations in detail for the case . The other cases follow in the same way.
The vector field undergoes a saddle-node bifurcation at . It has a saddle at and a stable node at for , while it has no equilibrium for . The sliding set is given by Therefore, the sliding set is empty for , while for . According to (5), at the sliding set , the sliding vector field is defined asIt is direct to check that is the unique attractive pseudoequilibrium.
Now we check that system (20a), (20b), (20c), and (20d) satisfies all the conditions of Theorem 3. For (20a), (20b), (20c), and (20d) it is direct to check that the equilibria transversally meets at or equivalently . Moreover, By Theorem 3, this implies that the equilibrium goes through an equilibrium transition for , while the equilibrium goes through a nonsmooth fold bifurcation for . Apart from these two boundary bifurcations, the line for in the parameter plane corresponds to a standard saddle-node bifurcation; see Figure 2. The phase portraits corresponding to the open regions I, II, and III labelled in Figure 2 are given in Figure 3.
(a) ,
(b) ,
(c) ,
5. Regularization of the BSN Bifurcation
In this section we apply the regularization approach (9) to the normal form of the BSN bifurcation and study the bifurcations of the regularized system.
Following (9) the regularization of (20a), (20b), (20c), and (20d) for is defined as follows:where the smooth function is defined in with , . For our computations we have chosen .
Now we consider the bifurcations that occur in the regularized system (24). Before that we first define the following bifurcation for piecewise-smooth continuous system.
Saddle-Node-Like Bifurcation (SNL). A “saddle” and a “node” collide at a certain parameter value and then both disappear.
We note that the “saddle” and “node” in the definition are not restricted to a standard saddle and a standard node. They can also be pseudosaddle and pseudonode. The extension “like” means that these bifurcations are not the standard ones as known from smooth systems. A detailed description of bifurcations in piecewise-smooth continuous systems is referred to in [15].
5.1. Equilibrium Transition
Recall that is an ordinary stable node for . It hits for and becomes a boundary node. Then it is replaced by a stable pseudonode on for . By Theorem 5, this equilibrium is still an equilibrium of after regularization with different position. It is located in for , on the line for , and in the region for . Thus, the regularized system does not experience any qualitative change as varies. This type of discontinuity-induced bifurcation disappears after regularization, see Figure 4.
(a)
(b)
(c)
5.2. Nonsmooth Fold
Recall that for , an ordinary saddle and a stable pseudonode coexist, colliding on for and then both disappear for . By Theorem 5 these two equilibria both preserve in their regularization with the same type of stability. To be precise, the saddle of is located in , while the stable node is located in the region for . The unique equilibrium of is located on the line for . However, there is no equilibrium in the regularized system for . Notice that is a piecewise-smooth continuous system; then a saddle-node-like bifurcation occurs in ; see Figure 5.
(a)
(b)
(c)
5.3. Standard Saddle-Node Bifurcation
Finally, the discontinuous system (20a), (20b), (20c), and (20d) has a standard saddle at and a standard stable node at for any , colliding for , and then both disappear for . By Theorem 5 the equilibria for different persist after regularization. Thus a saddle-node-like bifurcation occurs in the regularized system ; see Figure 6.
(a)
(b)
(c)
From the above results we find that the saddle-node-like bifurcation is the unique bifurcation occurring in the regularized system, which has codimension-1; see Figure 7. The phase portraits corresponding to the open regions I and II are presented in Figure 8.
(a)
(b)
6. Conclusions
This paper investigated the boundary-saddle-node (BSN) bifurcation by Filippov’s convex method and the regularization approach. This is a codimension- bifurcation of Filippov systems. We derived the topological normal form of the BSN bifurcation and gave a detailed description of its bifurcations and phase portraits. After regularization the Filippov system becomes a piecewise-smooth continuous system. Some of the discontinuity-induced bifurcations disappear, such as equilibrium transition. The unique bifurcation that occurs in the regularized system is the saddle-node-like bifurcation. The BSN point becomes a codimension-1 bifurcation in the piecewise-smooth continuous system, which makes our discussion easier. The regularization approach enables us to make use of the established theories for continuous systems and slow-fast systems to study the local behavior around the BSN bifurcation. We will apply the regularization approach to more complicated Filippov systems involving limit cycles in the future work, where more advantages of our approach in proving the existence and uniqueness of limit cycles will be presented.
Conflicts of Interest
The author declares that there are no conflicts of interest regarding the publication of this paper.
Acknowledgments
The author greatly appreciates the fruitful discussions with Ying Gao for her valuable suggestions. This research is supported by the Research Foundation of Education Bureau of Hebei Province under Grant no. QN2016026 and the Doctoral Starting up Foundation of Hebei GEO University under Grant no. B201612.