# Rank One Chaos

Post-publication activity

Curator: Quidong Wang

The theory of chaotic rank one attractors originated from the theory of Benedicks and Carleson on Hénon maps ((Benedicks and Carleson, 1985), (Benedicks and Carleson, 1991)) and the development that followed ((Benedicks and Young, 1993), (Benedicks and Young, 1999)). This theory was introduced by Lai-Sang Young and myself in a sequence of articles ((Wang and Young, 2001), (Wang and Young, 2008), (Wang and Young, 2010)). It takes the theory of Hénon-like maps ((Mora and Viana, 1993), (Benedicks and Viana, 2001)) as a particular case. The motivations for this theory are as follows:

(1) There was a geometric structure (namely the structure of critical regions) that, if added, could make more transparent the overall purpose of the different components of the very complicated tour de force analysis of Benedicks and Carleson. It could also induce non-trivial technical simplifications. Considering the profoundness of the theory in terms of impact, there was also a need to fill out, with care, the various technical details skipped in (Benedicks and Carleson, 1991).

(2) It is highly desirable to have a theory on non-uniformly hyperbolic maps that could be applied to the analysis of specific differential equations. Such a goal could be achieved in two steps. First, we could introduce a flexible setting on non-uniformly hyperbolic maps and construct a comprehensive chaos theory by generalizing previous analysis based on (Benedicks and Carleson, 1991). Second, since many natural and engineered systems are described by concrete differential equations of rank one type, they offer many possibilities for potential applications for this body of ideas.

We note that before this theory was introduced, the scope of application of the theory on the Hénon maps had been extended by Mora and Viana ((Mora and Viana, 1993)) to maps with transversal homoclinic tangency through the Newhouse theory ((Newhouse, 1974), (Palis and Takens, 1993)). See also (Diaz et al., 1996).

The rest of this article is divided into two sections. In the first we discuss the content of the theory and in the second we discuss its application to differential equations.

## A dynamics theory on chaotic rank one attractors

### Admissible family of rank one maps

For $$m \geq 2\ ,$$ let $${\mathcal A} = S \times D^{(m-1)}$$ where $$S$$ is the unit circle and $$D^{(m-1)}$$ is the unit disk of dimension $$m-1\ .$$ We consider a 2-parameter family of maps $$\{ T_{a, b}, \ (a, b) \in (a_0, a_1) \times (0, b_0) \}$$ where $$T_{a, b}: {\mathcal A} \to {\mathcal A}$$ is a diffeomorphism from $${\mathcal A}$$ to its image. Let $$(x_1, y_1) = T_{a, b}(x, y)$$ be such that $\tag{1} x_1 = F(x, y, a) + b u(x, y, a, b),\ y_1 = b v(x, y, a, b)$

where $$x \in S\ ,$$ $$y \in D^{(m-1)}\ .$$ Assume that $$F(x, y, a), u(x, y, a, b), v(x, y, a, b)$$ are $$C^3$$ in $$(x, y, a, b)\ ,$$ and their $$C^3$$-norms are uniformly bounded by a constant $$K_0$$ that is independent of $$a$$ and $$b\ .$$

We call the 1D map $$f_a: S \to S$$ defined by $$f_a(x) = F(x, 0, a)$$ the 1D singular limit for $$T_{a, b}\ .$$ We obtain $$\{f_a \}$$ by letting $$b = 0$$ in (1). For small $$b$$ we could regard $$T_{a, b}$$ as an m-dimensional map unfolded from the 1D singular limit $$f_a\ .$$ Let $$C(a) = \{ x: f'_a(x) = 0 \}$$ be the set of critical points for $$f_a\ .$$ We assume that $$|\partial_y F(x, 0, a)| \neq 0$$ for all $$x \in C(a)\ .$$ This is to say that the unfolding from $$f_a$$ to $$T_{a, b}$$ is non-degenerate in the direction of $$y$$ on $$C(a)\ .$$ We also assume $\frac{\det(DT_{a, b}(z))}{\det(DT_{a, b}(z'))} < K$ for all $$z, z' \in {\mathcal A}$$ where $$K$$ is a constant independent of $$b\ .$$ The last is a regularity condition.

The rest of our conditions are imposed on the 1D singular limit $$\{ f_a \}\ .$$ We assume that there exists an $$a_* \in (a_0, a_1)\ ,$$ so that $$f_{a_*}$$ is a Misiurewicz map. Misiurewicz maps are the simplest 1D maps with non-uniform expansion (Misiurewicz, 1981). For a Misiurewicz map, all critical orbits are kept a fixed distance away from the critical set. We also impose a parameter transversality condition on $$f_{a}$$ at $$a = a_*$$ to assure that changing $$a$$ around $$a_*$$ changes effectively the dynamics of $$f_a\ .$$ The precise formulation of these two conditions are a little tedious and they are fully laid out in (Wang and Young, 2008), to which we would refer the reader who is interested in such detail. What is important for us is that these are verifiable conditions. For instance, let $\tag{2} f_a(\theta) = \theta + a + L \Phi(\theta)$

and assume that $$\Phi(\theta)$$ is periodic in $$\theta$$ and all critical points of $$\Phi(\theta)$$ are non-degenerate. It is proved in (Wang and Young, 2002a) that, for $$L$$ sufficiently large, there exists an $$a_*$$ so that $$f_{a_*}$$ is a Misiurewicz map and the parameter transversality condition holds at $$a = a_*\ .$$

A two parameter family $$T_{a, b}$$ that assumes the form of (1) and satisfies all assumptions listed above is an admissible family of rank one maps.

### Existence of chaotic rank one attractors

Figure 1: (a) Axiom A solenoid and (b) rank one attractor.

If the 1D singular limit $$f_a$$ were uniformly expanding, then $$T_{a, b}$$ would represent an Axiom A solenoid (Smale, 1967). This is to say that $$T_{a, b}$$ would stretch $${\mathcal A}$$ in the $$x$$-direction, compress it in the $$y$$-direction. The image would then be wrapped a few times and put back in $${\mathcal A}\ .$$ In this case, we have a global splitting of stable and unstable directions hence a uniformly hyperbolic attractor. See Figure 1(a).

For an admissible family of rank one maps, however, the 1D singular limit is not uniformly expanding. See Figure 1(b). For rank one maps, the direction in which the image is wrapped in $${\mathcal A}$$ reverses at the critical values of $$f_a\ ,$$ making it impossible for us to attain a global splitting of stable and unstable directions.

The dynamics of the maps of Figure 1(b) is very different from that of the Axiom A solenoid of Figure 1(a) because of the lacking of a global splitting of stable and unstable directions. For the maps of Figure 1(b), periodic sinks, representing stable dynamical behavior, often co-exist with Smale's horseshoe. From a measure theoretic stand point, periodic sinks are directly observable but horseshoes are not. This is to say that a sink attracts an open, hence a positive measure, set of orbits but the basin of attraction of a horseshoe is typically a measure zero set. Is there chaos in a directly observable form for the maps of Figure 1(b)? How about the existence of such maps that admit a positive Lyapunov exponent on a positive Lebesgue measure set? Smale's horseshoe is now insufficient to offer affirmative answers to these questions.

Historically, these questions were asked for the Hénon maps and were long standing until the ground breaking analysis of Benedicks and Carleson emerged. Our first major task, in order to acquire a more flexible theory on maps for future applications, was to reconstruct the analysis of Benedicks and Carleson for an admissible family of rank one maps. Therefore the following statement is not only the first, but also the main theorem for the theory of chaotic rank one attractors.

Main Theorem  Let $$\{ T_{a, b}, \ (a, b) \in (a_1, a_2) \times (0, b_0)\}$$ be an admissible family of rank one maps. Then there exists a positive measure set of parameters $$\Delta \subset (a_1, a_2) \times (0, b_0)\ ,$$ such that for all $$(a, b) \in \Delta\ ,$$ $$T_{a, b}$$ admits a positive Lyapunov exponent Lebesgue almost everywhere in $${\mathcal A}\ .$$

The contents of the theory of chaotic rank one attractors are (a) to construct a positive measure set of parameters that fulfills this theorem, and (b) to further understand the dynamics of these chaotic rank one attractors by systematically building a dynamical profile for the maps of the good parameters constructed.

### Inductive construction of good parameters

Figure 2: Structure of critical regions.

Let $$\{ T_{a, b} \}$$ be an admissible family of rank one maps, and for $$T = T_{a, b}\ ,$$ let $\Omega = \cap_{n=0}^{+\infty} \ T^n({\mathcal A})$ be the attractor for $$T\ .$$ Our main theorem was proved by constructing a positive measure set $$\Delta$$ of good parameters through an elaborate inductive process. The good parameter set $$\Delta$$ is so constructed that, for $$(a, b) \in \Delta\ ,$$ the map $$T = T_{a, b}$$ possesses a well-defined critical set, which we denote as $${\mathcal C} \subset \Omega\ .$$ The critical set $${\mathcal C}$$ is such that (i) every point in $${\mathcal C}$$ is a point of quadratic tangency of well-defined local stable and unstable manifolds, and (ii) the critical orbits $$\cup_{n=-\infty}^{\infty} T^n({\mathcal C})$$ are all the tangency in $$\Omega$$ for $$T\ .$$

For the maps of good parameters, the critical set $${\mathcal C}$$ is a well structured Cantor set located around the critical points of the 1D singular limit. Let $$x_0$$ be a critical point of $$f_{a_*}\ ,$$ and let $$Q(x_0) = [x_0 - \delta, x_0+ \delta] \times D^{(m-1)}$$ where $$\delta > 0$$ is a fixed small number. A conceptual way to comprehend the structure of that Cantor set is to forever replace every cylinder by a finite collection of thinner and shorter cylinders inside, starting with $$Q(x_0)$$ for all $$x_0 \in C(a_*)\ .$$ See Figure 2.

The set of good parameters and the critical sets $${\mathcal C}$$ are constructed side by side inductively. Each inductive step starts with a temporary set of good parameters and a collection of small cylinders, each of which we denote as a critical region $$Q^{(k-1)}$$. We construct inductively new critical regions $$Q^{(k)}$$ inside of $$Q^{(k-1)}\ ,$$ deleting parameters that could potentially ruin the intended quadratic tangency of local stable and unstable manifolds. We prove that, at each step of the induction, the measure of the deleted parameters declines exponentially. At the end we obtain a positive measure set of parameters with a well-defined critical set in $$Q(x_0)$$ for all $$x_0 \in C(a_*)\ .$$

The induction described above and the proof of the main theorem are complicated and long. See (Wang and Young, 2001), (Wang and Young, 2008) and (Wang and Young, 2010).

### Dynamics of chaotic rank one attractors

Through the inductive process discussed in the previous subsection we have obtained a new class of chaotic attractors. These chaotic attractors are genuinely non-uniformly hyperbolic in the sense that the stable and unstable directions are allowed to mix. Our focus is now to built a dynamical profile so we could claim a comprehensive understanding on the geometric and dynamical structure of these chaotic rank one attractors. In what follows we let $$T = T_{a, b}$$ where $$(a, b) \in \Delta$$ is a good parameter that came out of the long inductive construction. Let $$\Omega = \cap_{n=0}^{+\infty} \ T^n({\mathcal A})$$ be the chaotic rank one attractor, and $${\mathcal C}$$ be the critical set for $$T\ .$$ The theorems listed in this subsection are proved for 2D chaotic rank one attractors in (Wang and Young, 2001), and the corresponding version in higher dimensions is proved in (Wang and Young, 2008) and (Wang and Young, 2010).

Theorem 1

For any give $$\varepsilon > 0\ ,$$ $\Lambda_{\varepsilon} = \{ z = (x, y) \in \Omega, \ d(T^n(z), {\mathcal C}) \geq \varepsilon, \ {\rm for \ all} \ n \in {\mathbb Z} \}$ is a uniformly hyperbolic invariant subset in $$\Omega$$ for $$T\ .$$

This theorem assures that the critical set $${\mathcal C}$$ is in fact the sole source of non-uniform hyperbolicity in $$\Omega\ .$$

We next turn to a special coding of orbits on $$\Omega\ .$$ The following notation will be used$x_1 < x_2 < \cdots < x_q < x_{q+1} = x_1$ are the critical points of $$f_{a_*}\ ,$$ and $${\mathcal C}_i$$ is the part of $$\mathcal C$$ near $$(x_i, 0) \in {\mathcal A}\ .$$ Let $$\Sigma_q= \Pi_{- \infty}^\infty \{1,2, \cdots, q\}$$ be the space of sequences of $$q$$-symbols, and let $$\sigma: \Sigma_q \to \Sigma_q$$ be the shift map.

Theorem 2

(a) There is a natural partition of $$\Omega \setminus {\mathcal C}$$ into disjoint sets $$A_1, A_2, \cdots, A_q$$ so that $$z \in A_i$$ can be thought of as being "to the right" of $${\mathcal C}_i$$ and "to the left" of $${\mathcal C}_{i+1}\ .$$

(b) Under the additional assumption that $$f_{a_*}([x_j, x_{j+1}]) \not \supset S^1$$ for any $$j\ ,$$ there is a closed subset $$\Sigma \subset \Sigma_q$$ with $$\sigma(\Sigma) \subset \Sigma$$ and a map $$\pi: \Sigma \to \Omega$$ with the property that

- for all $$\mathbf{s}=(s_i) \in \Sigma\ ,$$ $$\pi(\mathbf{s})=z$$ implies that $$T^iz \in \bar A_{s_i}$$ for all $$i\ ;$$

- $$\pi$$ is a continuous surjection that is 1-1 except on $$\cup_{i=-\infty}^\infty T^i {\mathcal C}\ ,$$ where it is 2-1.

If the assumption in Part (b) does not hold, the statement can be amended by increasing the number of symbols as follows: For $$f_{a_*}\ ,$$ use the partition given by $$C(a_*) \cup f_0^{-1}(C(a_*))$$ instead of that by $$C(a_*)\ ,$$ and do likewise with $$T\ .$$ See also (Wang and Young, 2010) for more results that follow immediately from the symbolic coding of Theorem 2.

B. Statistical properties  Inside a chaotic rank one attractor, orbits jump around in a seemingly random fashion and the future of individual orbits appears entirely unpredictable. For these chaotic attractors, however, there are, laws of statistics that control the asymptotic distributions of Lebesgue almost all orbits in $${\mathcal A}\ .$$

The precise meaning of the last statement is as follows. Let us first partition $${\mathcal A}$$ into a collection of disjoint sub-regions. For a finite orbit of $$T\ ,$$ we use this partition to define a histogram by counting the proportion of points on this finite orbit in each of the sub-regions. We then take two limits: first we let the orbit go infinitely long and second we refine the partition so the maximal size of all sub-regions goes to zero. In principle, there is no reason a priori for the histogram to converge to anything in any one of the two limit processes. For chaotic rank one attractors, however, not only the histograms converge, but also they converge to only a finite number of probability distributions for Lebesgue almost all orbits in $${\mathcal A}\ .$$ These limit distributions are the Sinai-Ruelle-Bowen measures for $$T\ .$$

Theorem 3

There are at least one, and at most $$q\ ,$$ ergodic SRB measures for $$T$$ where $$q$$ is the number of critical points of the 1D singular limit.

See (Young, 2002) and the reference therein for more on the theory of SRB measures. In addition to Theorem 3, we have also proved many refined dynamical properties of chaotic rank one attractors, including a central limit theorem, an exponential mixing rate and a principle of large deviations by applying the theories developed previously in (Young, 1998) and (Rey-Bellet and Young, 2008). See (Wang and Young, 2010).

## Application to differential equations

The theory of chaotic rank one attractors has been applied to the analysis of many naturally and engineered systems described by concrete differential equations. In this section we present three applications. The first application is on periodically perturbed homoclinic solutions (Wang and Ott, 2007). The second and the third applications are both under the framework of periodically kicked stable limit cycles (Wang and Young, 2002a). The second is on the dynamics of periodically kicked equations with Hopf bifurcation (Wang and Young, 2002b) and the third is an application to PDE arena \cite{LWY}. For more examples of application see also (Lin and Young, 2010), (Lin, 2006), (Guckenheimer et al., 2006), (Wang and Oksasoglu, 2008).

### Periodically perturbed homoclinic solutions

Figure 3: (a) Homoclinic intersection and (b) separated invariant manifold.

Periodically-forced second order equations, including the forced nonlinear pendulum, Duffing equation, and van der Pol oscillator, have been studied extensively in history ((Duffing, 1918), (Cartwright and Littlewood, 1945), (Levinson, 1949) (Guckenheimer and Holmes, 1983), (van der Pol and van der Mark, 1927)). When a homoclinic solution in a given second order equation is periodically perturbed, transversal intersections of stable and unstable manifolds occur within a certain range of the forcing parameter, generating homoclinic tangles and chaotic dynamics ((Poincaré, 1899), (Smale, 1965)). There also exist values of parameter for which the stable and unstable manifolds of the perturbed saddle are pulled apart. For these two cases, Figure 3 schematically illustrates the time-$$T$$ maps for the perturbed equations, where $$T$$ is the period of the perturbation.

Historically, the homoclinic tangles of Figure 3(a) have been regarded as a major venue for chaos to be manifested in differential equations. Proving chaos through the intersection of the stable and unstable manifolds ((Melnikov, 1963), (Smale, 1965)) has become a standard practice in the study of differential equations. This is to say that we often prove chaos in a given equation via the elimination of the scenario of Figure 3(b).

Afraimovich and Shilnikov, on the other hand, argued for chaos in the scenario of Figure 3(b) in (Afraimovich and Shil'nikov, 1977). They proposed to study the dynamics of periodically perturbed homoclinic solutions through the separatrix map, a return map around the unperturbed homoclinic solution in the extended phase space. Afraimovich and Shilnikov illustrated that, when a homoclinic solution is periodically perturbed, then horseshoe exists as well in the scenario of Figure 3(b) provided that the forcing frequency is sufficiently large.

Following the method proposed in (Afraimovich and Shil'nikov, 1977), Ott and I rigorously derived the separatrix map for periodically perturbed equation with a homoclinic saddle in (Wang and Ott, 2007). We verified that, under the assumption that the saddle fixed point is dissipative and non-resonant, the separatrix map is a family of rank one maps with a 1D singular limit in the form of $f_a (\theta) = \theta + a + \frac{\omega}{\beta} \ln M(\theta)$ where $$\omega$$ is the forcing frequency; $$\beta$$ is the unstable eigenvalue of the saddle fixed point; $$M(\theta)$$ is the classical Melnikov function, which is explicitly computable for a given equation; and $$a$$ is such that $a = \frac{\omega}{\beta} \ln \mu^{-1} + K,$ in which $$\mu$$ is a parameter representing the magnitude of the perturbation and $$K$$ is a constant from the given equation. Therefore if the Melnikov function $$M(\theta)$$ is such that $$\min_{\theta} M(\theta) > 0$$ (which puts us directly into the scenario of Figure 3(b)), and all critical points of $$M(\theta)$$ are non-degenerate, then the separatrix map is an admissible family of rank one maps provided that the forcing frequency is sufficiently large. In this case we obtain chaotic rank one attractors for a positive measure set of $$\mu\ .$$

As a concrete example, we applied the theory of chaotic rank one attractors to the Duffing equation in the form of $\tag{3} \frac{d^2 q}{dt^2} + (\lambda - \gamma q^2) \frac{d q}{dt} - q + q^3 = \mu \sin \omega t.$

It is well-known that, for every $$\lambda = \lambda_0 > 0$$ sufficiently small, there exists a $$\gamma = \gamma_0$$ such that equation (3) has a homoclinic solution when $$\mu = 0\ .$$ This homoclinic solution we denote as $$\ell_0\ .$$ Our next theorem asserts the existence of a directly observable global chaotic attractor in the neighborhood of $$\ell_0\ .$$

Theorem 4

Let $$\lambda_0 > 0$$ be sufficiently small and $$\gamma_0\ ,$$ $$\ell_0$$ be as in the above. Then there exists a positive measure set $$\Delta$$ for $$(\gamma, \omega, \mu)$$ close to $$(\gamma_0, \infty, 0)$$ such that for $$(\gamma, \omega, \mu) \in \Delta\ ,$$ the Duffing equation (3) admits a unique chaotic rank one attractor in an open neighborhood of $$\ell_0\ .$$

The results reported in this subsection are due to Ott and myself (Wang and Ott, 2011).

### Periodically kicked super-critical Hopf bifurcations

Our next application is about periodically kicked super-critical Hopf bifurcations. Hopf bifurcation is a dynamical scenario of historical and practical importance in both theory and application. External actions in the form of an impulse function are often modelled by periodic kicks in various applications, including in switch controlled circuits and in neuroscience and biology.

Let us start with the theory on Hopf bifurcation. For $$x \in {\mathbb R}^m, \ m \geq 2\ ,$$ consider the following $$\mu$$-dependent system of differential equations $\tag{4} \frac{d x}{d t} = A_{\mu} x + f_{\mu}(x)$

where $$A_{\mu}$$ is an $$m \times m$$ matrix and $$f_{\mu}(x)$$ is a vector valued function of order $$\geq 2$$ at $$x = 0\ .$$ We assume that, around $$\mu = 0\ ,$$ all eigenvalues of $$A_{\mu}$$ except a conjugating pair, which we denote as $$\lambda_{1, 2}\ ,$$ are with negative real part. We assume that $$\lambda_{1, 2} = a(\mu) \pm \omega(\mu) \sqrt{-1}$$ are such that $$a(0) = 0, \ \omega(0) \neq 0\ .$$ Corresponding to $$\lambda_{1, 2}\ ,$$ equation (4) has a 2-dimensional local center manifold $$W^c$$ at $$x = 0\ ,$$ and the equation induced on $$W^c$$ can be written in a complex variable $$z$$ in a normal form as $\tag{5} \dot z = (a(\mu) + i \omega(\mu)) z + k_1(\mu) z^2 {\bar z} + k_2(\mu) z^3 {\bar z}^2 + \cdots$

where $$k_1(\mu), k_2(\mu), \cdots$$ are complex numbers. We have a generic Hopf bifurcation for equation (4) if $$Re(k_1(0)) \neq 0\ .$$ A Hopf bifurcation is super-critical if $$Re(k_1(0)) < 0$$ and it is sub-critical if $$Re(k_1(0)) > 0\ .$$ In the rest of this subsection we assume $$Re(k_1(0)) < 0$$ so we are interested in a super-critical Hopf bifurcation, in which a stable periodic solution comes out of $$x = 0\ .$$

We now add a time-dependent forcing term to equation (4) to form a new equation $\tag{6} \frac{dx}{dt} = A_{\mu} x + f_{\mu}(x) + \varepsilon \Phi(x) \sum_{n = -\infty}^{+ \infty} \delta(t - nT)$

where $$\varepsilon$$ is a small parameter representing the magnitude of the forcing, $$\Phi(x)$$ is such that $$\Phi(0) = 0\ ,$$ and $$\delta(t)$$ is the standard $$\delta$$-function.

We computed the time-T map around the stable periodic solution coming out of $$x = 0$$ for small $$\mu\ .$$ It turned out that the time-T map is a family of rank one maps with a singular 1D limit, as $$T \to \infty\ ,$$ in the form of $f_a(\theta) = \theta + a + L \phi(\theta)$ where $$a \approx \omega(0)T \ mod(2 \pi)\ ;$$ $$\phi(\theta) = \phi_0(\theta) + {\mathcal O}(\sqrt{\mu}) + {\mathcal O}(\varepsilon)$$ in which $$\phi_0(\theta)$$ is a function of $$\theta$$ determined by $$A_0$$ and $$D\Phi(0)\ ;$$ and $$L \approx \varepsilon \cdot \tau$$ and $\tau = \left| \frac{Im(k_1(0))}{Re(k_1(0))}\right|.$ So to find chaotic rank one attractors for equation (6),

• (i)] first we compute $$\phi_0(\theta)\ ,$$ and verify that all its critical points are non-degenerate;
• (ii)] second we compute the normal form (5) to find $$k_1(0)\ ,$$ and we look for equations for which $$\tau$$ is large;
• (iii)] let $$\mu$$ and $$\varepsilon$$ be sufficiently small so that $$\phi(\theta)$$ is close to $$\phi_0(\theta)\ ;$$ we also let $$L = \tau \cdot \varepsilon$$ be sufficiently large;
• (iv)] we then conclude that chaotic rank one attractors exist for a positive measure set of $$T\ .$$

We skip the precise statement and refer the reader to (Wang and Young, 2002b), \cite{LWY} and (Wang and Oksasoglu, 2008) for more details.

Figure 4: (a) Switch controlled Chua circuit and (b) a chaotic rank one attractor.

To provide a concrete application we applied this theory to a periodically kicked Chua's circuit. The circuit is shown inFigure 3(a) and the periodic kicks are implemented by switches being turned on and off periodically. We analyzed the differential equation for the circuit following the steps (i)-(iv) above, and found parameters for which the chaotic rank one attractors are likely to occur. We then physically built the circuit. Figure 3(b) is a picture of chaotic rank one attractor that came out of lab simulations. See (Wang and Oksasoglu, 2008) and (Demirkol et al., 2009).

The theory on periodically kicked Hopf bifurcation is due to Young and myself (Wang and Young, 2002b); the application to Chua's circuit is due to Oksasoglu and myself (Wang and Oksasoglu, 2008); and the circuit is built by Demirkol, Ozoguz, and Akgul (Demirkol et al., 2009).

### Application to a parabolic partial differential equation

The Brusselator, as described by the equations below, is a simplified model of an autocatalytic chemical reaction with diffusion (Lefever and Prigogine, 1968): $\tag{7} u_t = d_1 \Delta u + a - (b+1)u + u^2v,\ v_t = d_2 \Delta v + bu -u^2v.$

We consider this model in one physical dimension, i.e. $$u = u(x, t)$$ and $$v=v(x, t)$$ for $$x \in [0, 1]\ ,$$ and let $$\Delta =\partial_{xx}\ .$$ Here $$a$$ and $$b$$ are constants representing concentrations of certain initial substances, $$u$$ and $$v$$ are variables representing concentrations of two intermediates, and $$d_1, d_2 >0$$ are their respective diffusion coefficients. The term $$u^2v$$ represents the autocatalytic step in the reaction.

One sees immediately that $$(u(t),v(t)) \equiv (a, ba^{-1})$$ is a stationary solution. Letting $$U=u-a\ ,$$ $$V=v-ba^{-1}\ ,$$ and $$\mathbf{u}=(U,V)\ ,$$ we write (7) as an evolutionary equation $\tag{8} \dot \mathbf{u}=\mathcal{A}_{a,b} \mathbf{u} + f_{a,b}(\mathbf{u}),$

where $\mathcal{A}_{a,b}=\begin{pmatrix} d_1 \frac{\partial^2}{\partial x^2} + (b-1)& a^2\\ -b & \theta d_1\frac{\partial ^2}{\partial x^2} - a^2 \end{pmatrix}, \qquad \theta = \frac{d_2}{d_1}, \ \ d_1 \neq 0;$ and $f_{a, b}(\mathbf{u}) = \begin{pmatrix} UV^2 + ba^{-1} U^2 + 2aUV \\ -UV^2 - ba^{-1} U^2 - 2a UV \end{pmatrix}.$ It is well known that Hopf bifurcation occurs as the parameters $$a$$ and $$b$$ are varied. See (Hassard et al., 1981).

We study periodically kicked equations satisfying the Neumann boundary condition and the Dirichlet boundary condition. The forcing functions are chosen to respect the corresponding boundary condition.

Neumann boundary condition. $u_t = d_1 \Delta u + a - (b+1)u + u^2v + \rho (1+\cos \pi x) \ p_{T,\iota}(t),$ $v_t = d_2 \Delta v + bu -u^2v;$ $\tag{9} \partial_x u(0,t) = \partial_x u(1,t) = 0, \qquad \partial_x v(0,t) = \partial_x v(1,t) = 0.$

Dirichlet boundary condition. $u_t = d_1 \Delta u + a - (b+1)u + u^2v + \rho \sin \pi x \ p_{T,\iota}(t),$ $v_t = d_2 \Delta v + bu -u^2v;$ $\tag{10} u(0,t) = u(1,t) = a, \qquad v(0,t) = v(1,t) = ba^{-1}.$

In both cases, $$\rho \geq 0$$ and $$0<\iota<<1<<T$$ are constants and $p_{T,\iota}(t)= \sum_{n=-\infty}^{\infty} p_\iota (t - nT) \quad {\rm with} \quad p_{\iota}(t) = \left\{ \begin{array}{lll} \iota^{-1} & & 0 \leq t < \iota, \\ 0 & & \rm{elsewhere}. \end{array} \right.$ We use $$p_{T,\iota}(t)$$ in the place of a periodic delta function to avoid certain regularity problems. Our plan in \cite{LWY} is to first develop a theory on chaotic rank one attractors for periodically kicked super-critical Hopf bifurcation in the context of infinite dimensional systems. We then apply this theory to equations (9) and (10).

Theorem 5 (Neumann Boundary condition)

Let $$d_1>0\ ,$$ $$\theta<1$$ be fixed. Then for $$a$$ sufficiently large, there is an open set of $$\rho, \iota$$ and $$b$$ (depending on $$d_1\ ,$$ $$\theta$$ and $$a$$), $$b \approx a^2+1\ ,$$ for which the time-T map of equation (9) has a chaotic rank one attractor for a positive measure set of large $$T\ .$$

The computation needed for the Dirichlet boundary condition is considerably more involved than those for Neumann boundary condition. We limited ourselves in \cite{LWY} to the special case $$d_1=\pi^{-2}\ .$$

Theorem 6 (Dirichlet boundary condition)

Let $$d_1=\pi^{-2}\ ,$$ $$0< \theta <<1\ .$$ Then there are open sets of $$a, b, \rho$$ and $$\iota\ ,$$ for which the time-T map of equation (10) has a chaotic rank one attractor for a positive measure set of large $$T\ .$$

We refer the interested reader to \cite{LWY} for details on the infinite dimensional theory and the application. The results reported in this subsection are due to Lu, Young and myself \cite{LWY}.

## References

• V.S. Afraimovich and L.P. Shil'nikov, The ring principle in problems of interaction between two self-oscillating systems, PMM Vol 41(4), (1977), 618-627.
• M. Benedicks and L. Carleson, On iterations of $$1-ax^2$$ on $$(-1, 1)$$, Ann. Math. 122 (1985), 1-25.
• M. Benedicks and L. Carleson, The dynamics of the Hénon map, Ann. Math. 133 (1991), 73-169.
• M. Benedicks and M. Viana, Solution of the basin problem for Hénon attractors, Invent. Math. 143 (2001), 375-434.
• M. Benedicks and L.-S. Young, Sinai-Bowen-Ruell measure for certain Hénon maps, Invent. Math. 112 (1993), 541-576.
• M. Benedicks and L.-S. Young, Markov extensions and decay of correlations for certain Hénon maps, Asterisque (1999).
• K.L. Cartwright and J.E. Littlewood, On non-linear differential equations of the second order, I: The equation $$\ddot{y} + k (1-y^2) \dot{y} + y = b \lambda k \cos{(\lambda t + a)}, k$$ large, J. London. Math. Soc. 20, (1945), 180-189.
• L.J. Diaz, J. Rocha and M. Viana. Strange attractors in saddle-node cycles: prevalence and globality, Invent. Math. 125 (1996), 37-74.
• G. Duffing, Erzwungene Schwingungen bei Veranderlicher Eigenfrequenz, F. Wieweg u. Sohn: Braunschweig. (1918).
• J. Guckenheimer and P. Holmes, Nonlinear oscillations, dynamical systems and bifurcation of vector fields, Springer-Verlag, Appl. Math. Sciences 42 (1983).
• J. Guckenheimer, M. Weschelberger and L.-S. Young, Chaotic Attractors of Relaxation Oscillators, Nonlinearity, 19, (2006), 701-720.
• B. Hassard, N. Kazarinoff, and Y-H. Wan, Theory and Applications of Hopf Bifurcation, Cambridge University Press, (1981).
• N. Levinson, A Second-order differential equation with singular solutions, Ann. Math. 50 (1949), 127-153.
• K. Lin, Entrainment and chaos in a pulse-driven Hodgkin-Huxley oscillator, SIAM J. Appl. Dyn. Systt. 5 (2006), 179-204.
• K. Lin and L.-S. Young, Dynamics of periodically-kicked oscillators, to appear, Journal of Fixed Point Theory and Applications (2010).
• R. Lefever and I. Prigogine. Symmetry-breaking instability in dissipative systems II, J. Chem. Phy., 48 (1968), 1695-1700.
• K. Lu, Q.D. Wang, and L.-S. Young, Strange attractors for periodically forced parabolic equations, 2007 preprint.
• V.K. Melnikov, On the stability of the center for time periodic perturbations, Trans. Moscow Math. Soc., 12 (1963), 1-57.
• M. Misiurewicz, Absolutely continues invariant measures for certain maps of an interval, Publ. Math. IHES. 53 (1981), 17-51.
• L. Mora and M. Viana, Abundance of strange attractors, Acta. Math. 171 (1993), 1-71.
• S. E. Newhouse, Diffeomorphisms with infinitely many sinks, Topology, 13 (1974), 9-18.
• A.S Demirkol, S. Ozoguz, A. Oksasoglu and T. Akgul and Q. Wang, Experimental verification of rank one chaos in switch-controlled smooth Chua's circuit, Chaos 19(1) (2009).
• J. Palis and F. Takens, Hyperbolicity & sensitive chaotic dynamics at homoclinic bifurcations, Cambridge studies in advanced mathematics, 35 Cambridge University Press, Cambridge, (1993).
• H. Poincaré, Les Méthodes Nouvelles de la Mécanique Céleste, 3 Vols. Gauthier-Villars: Paris (1899).
• L. Rey-Bellet and L.-S. Young, Large Deviations in nonuniformly hyperbolic dynamical systems, Ergodic Theory Dynam. Systems 28(2) (2008), 587-612.
• S. Smale, Diffeomorphisms with many periodic points, Differential and Combinatorial Topology (A Symposium in Honor of Marston Morse), Princeton University Press, (1965), 63-80.
• S. Smale, Differentiable Dynamical Systems, Bull. Amer. Math. Soc. 73 (1967), 747-817.
• B. van der Pol and J. van der Mark, Frequency demultiplication, Nature 120 (1927), 363-364.
• Q.D. Wang and W. Ott, Dissipative homoclinic loops and rank one chaos, preprint (2007).
• Q.D. Wang and Ali Oksasoglu, Rank one chaos: theory and applications, IJBC 185 (2008) 1261-1319.
• Q.D. Wang and L.-S. Young, Strange attractors with one direction of instability, Commun. Math. Phys., 218 (2001), 1-97.
• Q.D. Wang and L.-S. Young. Towards a theory of rank one attractors, Annal. Math., 167 (2008) 349-480.
• Q.D. Wang and L.-S. Young. Dynamical Profile of a Class of Rank One Attractors, preprint, (2010).
• Q.D. Wang and L.-S. Young. From invariant curves to strange attractors, Commun. Math. Phys., 225 (2002a), 275-304.
• Q.D. Wang and L.-S. Young. Strange attractors in periodically-kicked limit cycles and Hopf bifurcations, Commun. Math. Phys., 240 (2002b), 509-529.
• L.-S. Young, Statistical properties of dynamical systems with some hyperbolicity, Ann. of Math. 147 (1998), 585-650.
• L.-S. Young, What are SRB measures, and which dynamical systems have them?, J. Stat. Phys. 108 (2002), 733-754.