The Role of Depth and 1/f Dynamics in Perceiving Reversible Figures

When confronted with a reversible figure, such as the Necker Cube, viewers experience a spontaneously changing percept. We assess the dynamic of how the human visual system resolves perceptual ambiguity in stimuli that offer multiple interpretations. Subjects observed the Necker cube for one of three viewing durations during which they pressed a key each time they perceived a change in the orientation of the cube. Manipulations of binocular disparity served as a parameter to control perceptual stability. Low-depth conditions yielded more perceptual reversals than high-depth conditions. A Fourier analysis performed on the time series of reversals show 1/f (pink) noise was evident in their power spectra. These results together with theoretical models of complex systems (e.g., Bak, Tang, & Wiesenfeld, 1987) suggest that depth information may guide our perceptual system into a self-organized state to assist us in resolving ambiguous information. Moreover, slopes of the spectra were steeper in high-depth and brief viewing conditions, suggesting that the visual system relies more on previous perceptual states and filters more white noise in these conditions.     

Excess thermal noise and low-frequency oscillations during capillary flow of aqueous solutions of poly(ethylene oxide)

The excess thermal noise generated in polymer solutions through narrow capillaries is studied in detail for aqueous solutions of poly(ethylene oxide), , of varying concentration. With increasing flow rate, the excess noise level increases, the noise spectrum assuming a 1/f -form with 1.5. Within a critical flow range, distinct peaks appear in the spectrum, their frequencies being multiples of a fundamental frequency. The latter frequency (f ₀) is found to increase with the flow rate; this variation, as well as that brought about by varying concentration and capillary dimensions, can be accommodated in a single curve correlatingf ₀ with the shear rate at the capillary wall. No such correlation was found for the total noise level. The value off ₀ appeared to be determined by transversal oscillations of the liquid stream entering the capillary. Addition of small amounts of silica particles (Aerosil) led to the disappearance of the peaks in the spectrum.     

Excess thermal noise generated during Poiseuille flow of certain polymer solutions

The previously reported results concerning the generation of excess thermal noise induced by capillary flow of aqueous solutions of poly(ethylene oxide) (PEO) are supplemented by measurements on the following solutions: PEO/DMF, PEO/i-PrOH, PS/THF, PVAC/cyclohexanone, and poly(acrylamide)/water. Similarly to the previous findings, a noise level increasing with the flow rate is recorded, the noise exhibiting a l/f -frequency spectrum. Within a certain flow range, distinct peaks are recorded in the spectrum (harmonics of a fundamental frequency,f ₀). Thef ₀-values of the various solutions under varying flow conditions arrange themselves along a commonf ₀-shear rate curve. They appear to be associated with transversal oscillations of the solution upstream the capillary entrance.     

Bifurcations of Heteroclinic Orbits

The search for traveling wave solutions of a semilinear diffusion partial differential equation can be reduced to the search for heteroclinic solutions of the ordinary differential equation ü − cu̇ + f(u) = 0, where c is a positive constant and f is a nonlinear function. A heteroclinic orbit is a solution u(t) such that u(t) → γ ₁ as t → −∞ and u(t) → γ ₂ as t → ∞ where γ ₁, γ ₂ are zeros of f. We study the existence of heteroclinic orbits under various assumptions on the nonlinear function f and their bifurcations as c is varied. Our arguments are geometric in nature and so we make only minimal smoothness assumptions. We only assume that f is continuous and that the equation has a unique solution to the initial value problem. Under these weaker smoothness conditions we reprove the classical result that for large c there is a unique positive heteroclinic orbit from 0 to 1 when f(0) = f(1) = 0 and f(u) > 0 for 0 < u < 1. When there are more zeros of f, there is the possibility of bifurcations of the heteroclinic orbit as c varies. We give a detailed analysis of the bifurcation of the heteroclinic orbits when f is zero at the five points −1 < −θ < 0 < θ < 1 and f is odd. The heteroclinic orbit that tends to 1 as t → ∞ starts at one of the three zeros, −θ, 0, θ as t → −∞. It hops back and forth among these three zeros an infinite number of times in a predictable sequence as c is varied.     

Convergence of biorthogonal series of biharmonic eigenfunctions by the method of titchmarsh

Canonical edge problems for the biharmonic equation can be solved by separating variables. The eigenvalues and eigenvectors arising in this separation are derived from a reduced system of ordinary differential equations along lines suggested in the excellent work of R. C. Smith (1952). We study the reduced system which is governed by a vector ordinary differential equation. A solution of the biharmonic problem, governed by a partial differential equation, can be found only if the prescribed data is restricted to a subspace of the space spanned by the eigenfunctions of the reduced problem. The theory leads to problems in generalized harmonic analysis which seek conditions under which arbitrary vector fields f(y) with values in ² can be represented in terms of eigenvectors of the reduced problem. This paper adds new theorems and conjectures to the theory. We extend Smith's generalization to fourth-order problems of the methods introduced by Titchmarsh (1946) to study eigenfunction expansions associated with second-order problems. We use this method to prove that, if f(y)=[(f ₁(y), f ₂y)], -1y1, f(y) C¹[-1, 1], f L₂[-1, 1], then the series expressing f(y) converges uniformly to f(y) in the open interval (-1, 1), uniformly in [-1, 1] if f ₁(±1)=0 and, in any case, to [0, f ₂(±1)-f ₁(±1)] at y=±1. This is unlike Fourier series, which converge to the mean value of the periodic extension of a function. The series exhibits a Gibbs phenomenon near the end points of discontinuity when f ₁(±1) 0.The Gibbs undershoot and overshoot for the step function vector [1, 0] and ramp function vector [y, 0] are computed numerically. The undershoot and overshoot are much larger than in the case of Fourier series and, unlike Fourier series, the Gibbs oscillations do not appear to be entirely suppressed by Féjer's method of summing Cesaro sums. We show that, when f(y) has interior points of discontinuity, the series for f(y) diverges and we present numerical results which indicate that, in this divergent case, the Cesaro sums converge to f(y) apparently with Gibbs oscillations near the point of discontinuity.     

Aeroacoustic Modelling of Low-Speed Flows

A new numerical algorithm for acoustic noise generation is developed. The approach involves two steps comprising an incompressible flow part and an inviscid acoustic part. The acoustic part can be started at any time of the incompressible computation. The formulation can be applied both for isentropic flows and non-isentropic flows. The model is validated for the cases of an isentropic pulsating sphere and non-isentropic flows past a circular cylinder. In the latter case the computations show that the generated acoustic field in addition to the dominant Strouhal frequency, f ₀, contains a slightly higher frequency, f ₂, and a modulating lower frequency, f ₁=f ₂−f ₀. Numerical experiments with different interpolation schemes, boundary conditions, etc., show that the appearance of these modes is not an artifact from the numerical discretization. Received 20 July 1998 and accepted 26 May 1999     

Modified flapping jet for increased jet spreading using synthetic jets

The present paper is an experimental investigation, using a PIV system, on modified rectangular jet flow co-flowing with a pair of synthetic jets placed symmetrically with respect to the geometric centerline of the main flow. The objective was to determine the optimal forcing conditions that would result in jet spreading beyond what would be obtained in a simple flapped jet. The main jet had an exit Re_h = 36,000, based on the slot height, h. The synthetic jets were operated in a periodic manner with a periodic momentum coefficient of about 3.3% and at a frequency of the main jet preferred mode. A short, wide angle diffuser of half angle of about 45° was attached to the main jet. Generally for the vectored jet, much of the flow features found here resembled those reported in the literature except that the deflection angle in this study increased with downstream distances inside the diffuser and then remained roughly unchanged thereafter. Larger jet spreading was achieved when the main jet was subjected to simultaneous actuation of the synthetic jets but the flow did not achieve the initial jet spreading that was observed in the vectored jet. Further jet spreading was achieved when the synthetic jets were alternately actuated in which each synthetic jet was actuated for a number of cycles before switching. This technique allowed the jet to flap across the flow between transverse positions larger than what would be obtained in a simple flip-flop jet. Under the present flow geometry and Reynolds number, it was found that when the ratio f_s/f_al, where f_s is the synthetic actuation frequency and f_al is the alternating frequency, was larger than 10, the mean streamwise velocity of the main jet had two peaks symmetrically placed with respect to the jet axis and the jet had the appearance of flowing into two streams each moving nearly parallel to the diffuser wall. For a value of f_s/f_al of about 10, the optimal value in this study, the desired flow properties were achieved in that, the mean velocity was nearly uniform with an increase in the jet width compared to the simultaneous actuations, and the jet flapping was more effective in redistributing and homogenizing the turbulent kinetic energy across the main jet.     

Signal-processing techniques for low signal-to-noise ratio laser Doppler velocimetry signals

A variety of methods have been developed to obtain acurate frequency estimates from laser Doppler velocimetry (LDV) signals. Rapid scanning and fiber optic LDV systems require robust methods for extracting accurate frequency estimates with computational efficiency from data with poor signal-to-noise ratios. These methods typically fall into two general categories, time domain parametric techniques and frequency domain techniques. The frequency domain approach is initiated by transforming the Doppler bursts into the frequency domain using the fast Fourier transform (FFT). From this basic transformation a variety of interpolation procedures (parabolic, Gaussian, and centroid fits) have been developed to optimize the frequency estimation accuracy. The time domain approaches are derived from the parametric form of a sinusoid. The estimation of constants in this relationship is performed to satisfy specific constraints, typically a minimization of a variance expression. A comparison of these techniques is presented using simulated signals and additive Gaussian and Poisson white noise. The statistical bias and random errors for each method are presented from 200 signal simulations at each condition. Frequency estimation via the FFT with zero-padding and a Gaussian interpolation scheme was found to produce the lowest bias and random errors.List of symbols A(z) eigenfilter or characteristic polynomial - a _{m + 1} eigenvector - f frequency, Hz - f normalized frequency f = f/f _s - _d Doppler frequency estimate - f _i frequency of FFT spectral bin - f ^s sampling frequency - N number of sample points in data set - P _i ith power spectral line from PSD - r _xx (i) autocorrelation coefficient for time lag i - R_{Mn + 1} autocorrelation matrix of order M+1 - T sampling period - f spectral resolution for FFT, f = 1/N t - t sampling interval     

Boundedness of Fatou Components of Holomorphic Maps

We shall discuss the topological properties of the Fatou sets for holomorphic maps. Let f ₁,f ₂,....,f _N be non-constant holomorphic maps in the plane, each having order less than 1/2. It is shown that if the lower order of f _j , is greater than 0 for some j{1,2,...,N}, then the Fatou set of the map h=f _N f _N-1 ... f ₁ has no unbounded components     

Connectivity and Design of Planar Global Attractors of Sturm Type. III: Small and Platonic Examples

Based on a Morse-Smale structure, we study planar global attractors A_f{{\mathcal A}_f} of the scalar reaction-advection-diffusion equation u _t = u _xx + f (x, u, u _x ) in one space dimension. We assume Neumann boundary conditions on the unit interval, dissipativeness of f, and hyperbolicity of equilibria. We call A_f{{\mathcal A}_f} Sturm attractor because our results strongly rely on nonlinear nodal properties of Sturm type. The planar Sturm attractor A_f{{\mathcal A}_f} consists of equilibria of Morse index 0, 1, or 2, and their heteroclinic connecting orbits. The unique heteroclinic orbits between adjacent Morse levels define a plane graph C_f{{\mathcal C}_f} , which we call the connection graph. Its 1-skeleton C¹_f{{\mathcal C}^1_f} is the closure of the unstable manifolds (separatrices) of the index-1 Morse saddles. We summarize and apply two previous results (Fiedler and Rocha, J. Diff. Equ. 244: 1255–1286, 2008, Crelle J. Reine Angew. Math. 26 pp., 2009, doi:) which completely characterize the connection graphs C_f{{\mathcal C}_f} and their 1-skeletons C¹_f{{\mathcal C}^1_f}, in purely graph theoretical terms. Connection graphs are characterized by the existence of pairs of Hamiltonian paths with certain chiral restrictions on face passages. Their 1-skeletons are characterized by the existence of cycle-free orientations. Such orientations are called bipolar in de Fraysseix et al. (Discrete Appl. Math. 56: 157–179, 1995). We describe all planar Sturm attractors with up to 11 equilibria. We also design planar Sturm attractors with prescribed Platonic 1-skeletons of their connection graphs. We present complete lists for the tetrahedron, octahedron, and cube. We provide representative examples for the design of dodecahedral and icosahedral Sturm attractors. Unlike previous examples, and in particular unlike the classification of Sturm attractors with up to nine equilibria, our present results are based on analytic insight rather than mindless computer-based enumeration.     

Regularity of the Inverse of a Planar Sobolev Homeomorphism

Let be a domain. Suppose that f ∈ W^1,1_loc(Ω,R²) is a homeomorphism such that Df(x) vanishes almost everywhere in the zero set of J_f. We show that f^-1 ∈ W^1,1_loc(f(Ω),R²) and that Df⁻¹(y) vanishes almost everywhere in the zero set of Sharp conditions to quarantee that f⁻¹ ∈ W^1,q(f(Ω),R²) for some 1<q≤2 are also given.     

On resolution to Wu＇s conjecture on Cauchy function＇s exterior singularities

This is a series of studies on Wu’s conjecture and on its resolution to be presented herein. Both are devoted to expound all the comprehensive properties of Cauchy’s function f(z) (z = x + iy) and its integral J[f(z) ] ≡(2πi) -1 C f(t)(t z) -1dt taken along the unit circle as contour C,inside which(the open domain D+) f(z) is regular but has singularities distributed in open domain Doutside C. Resolution is given to the inverse problem that the singularities of f(z) can be determined in analytical form in terms of the values f(t) of f(z) numerically prescribed on C(|t| = 1) ,as so enunciated by Wu’s conjecture. The case of a single singularity is solved using complex algebra and analysis to acquire the solution structure for a standard reference. Multiple singularities are resolved by reducing them to a single one by elimination in principle,for which purpose a general asymptotic method is developed here for resolution to the conjecture by induction,and essential singularities are treated with employing the generalized Hilbert transforms. These new methods are applicable to relevant problems in mathematics,engineering and technology in analogy with resolving the inverse problem presented here.     

Electrical fluctuations and entry flow instabilities during capillary flow of aqueous poly (ethylene oxide) solutions

Capillary flow of poly (ethylene oxide) solutions generates voltage fluctuations (noise) between electrodes placed on both sides of the capillary. The noise has a 1/f type spectral distribution, the value of increasing with (shear rate) to a limiting level. Within certain ranges, two sets of harmonic peaks appear in the spectra. It is demonstrated that these peaks are related to the frequency components of the instabilities in the entry flow region. The lower frequency set of harmonics corresponds to axial oscillations (pulsations) of the flow, while the higher frequency peaks are associated with the transverse oscillation of the stream lines in the vicinity of the entry. The corresponding frequencies were measured by visual counting and by spectral analysis of laser light transmitted through the entry region during injection of a coloured solution.The noise measurements were carried out using both platinum and reversible Ag/AgCl electrodes. In the latter case the streaming potential was also measured; its variation due to pressure fluctuations in the instable flow region appears to provide a plausible explanation of flow-induced noise phenomena as observed in elastic solutions.Dedicated to Prof. Dr. J. Schurz on the occasion of his sixtieth birthday.     

The Spectrum of the Frobenius–Perron Operator and Its Discretization for Circle Diffeomorphisms

To investigate the dynamical behaviour of a discrete dynamical system given by a map f, it is nowadays a standard method to look at the discretization of the Frobenius–Perron operator _f w.r.t. a box-partition of the state space resulting in a transition matrix P _{f, N} of a finite Markov chain. We are interested in information about the dynamics of f obtained from the spectrum of P _{f, N} , especially for circle diffeomorphisms. Therefore the spectra of _f and P _{f, N} are investigated.     

Piezo rotary and axial vibrator (PRAV) characterization of a fresh coating during its drying

Using a piezo rotary and axial vibrator (PRAV) the viscoelastic properties of a fresh Al₂O₃ coating that is deposited on a well-chosen plate can be measured simultaneously during drying. There are three types of vibrations available to be excited and detected in their resonance modes, by measuring resonance frequencies f _k and half-widths h _k, before and after coating as a function of time: axial vibration, bending vibration to get the evolving Young’s modulus and rotary vibration to follow the viscosity increase during drying of the thin coating. This information is contained in the complex frequency shifts Ω_k = 2Δf _k /f _k + iΔh _k /f _k of the three vibration modes caused by the coating on the plate. A derivation of the relationships, their validation on Newtonian liquids and experimental applications carried out using the PRAV are given in the paper.     

Variants of the Stokes Problem: the Case of Anisotropic Potentials

We investigate the smoothness properties of local solutions of the nonlinear Stokes problem$\begin{eqnarray*}-\diverg \{T(\eps(v))\} + \nabla \pi &=& g \msp \mbox{on $\Omega$,}\\\diverg v&\equiv & 0 \msp \mbox{on $\Omega$,}\end{eqnarray*}$where v: ⁿ is the velocity field, $\pi$: $ denotes the pressure function, and g: ⁿ represents a system of volume forces, denoting an open subset of ⁿ . The tensor T is assumed to be the gradient of some potential f acting on symmetric matrices. Our main hypothesis imposed on f is the existence of exponents 1 < p q < \infty such that\lambda (1+|\eps|^{2})^{\frac{p-2}{2}} |\sigma|^{2} \leq D^{2}f(\eps)(\sigma ,\sigma) \leq \Lambda (1+|\eps|^{2})^{\frac{q-2}{2}} |\sigma|^{2}holds with suitable constants , > 0, i.e. the potential f is of anisotropic power growth. Under natural assumptions on p and q we prove that velocity fields from the space W ¹ _{p, loc} (; ⁿ ) are of class C ^1, on an open subset of with full measure. If n = 2, then the set of interior singularities is empty.Dedicated to O. A. Ladyzhenskaya on the occasion of her 80th birthday     

An experimental investigation of the sources of propeller noise due to the ingestion of turbulence at low speeds

Noise radiation from a four bladed, 10 in. diameter propeller operating in air at a rotational speed of 3000 RPM and a freestream velocity of 33 ft/s was experimentally analyzed using hot-wire and microphone measurements in an anechoic wind tunnel. Turbulence levels from 0.2 to 5.5% at the propeller location were generated by square-mesh grids upstream of the propeller. Autobicoherence measurements behind the blade trailing edges near the hub and tip showed regions of high phase-coherence between the blade-passage harmonics and the broadband frequencies. Inflow turbulence reduced this coherence. By relating the fluctuation velocities in the propeller wake to the unsteady blade forces, the primary regions of tonal noise generation have been identified as the hub and tip regions, while the midspan has been identified as a region responsible for broadband noise generation. These measurements were complimented by cross-spectra between the propeller wake-flow and the measured sound. The effect of turbulence on the radiated noise level showed an overall increase of 2 dB in the broadband levels for every 1% increase in turbulence. This effect varied for different frequency bands in the acoustic spectrum.List of Symbols b ² (f _k, f _l) Bicoherence - B (k, l) Bispectrum - B Number of Blades - c Speed of Sound - C _T Thrust coefficient=T/n ²D⁴ - D Propeller diameter - E [ ] Expected value - f Frequency, Hz - G _xx (f) One sided autospectral density function. - G _xy (f) One sided cross-spectral density function - J Advance ratio, J=U/nD - j, k Fourier component indices - m Grid mesh length - M _o Rotational mach number at a radial location M _o =2nr/a _o - M _c Axial convection mach number - n Rotational speed, revolutions per second - r Propeller radial location - R Propeller radius - R _e Reynolds number - T Propeller thrust - U Freestream velocity - U _i Induced axial velocity from propeller - u, w RMS of fluctuating velocity, u=(U–u)² - X(f) Fourier transform of x(t) Symbols _xy ^/2 (f) Coherence function, - Observer angle, measured from propeller thrust direction - _f Longitudinal Eulerian dissipation length scale - _f Longitudinal Eulerian integral length scale - Air density - Blade azimuthal location This research was performed at the Hessert Center for Aerospace Research, Department of Aerospace and Mechanical Engineering, University of Notre Dame, and was sponsored by the U.S. Navy, Office of Naval Research, Arlington, Virginia under Contract No. N00014-89-J-1783. The authors would like to thank the program manager, and technical manager, Dr. E. P. Rood. The authors would also like to thank Dr. William Blake of the David Taylor Research Center and Dr. Flint O. Thomas and Dr. Huang-Chang Chu of Notre Dame for their help and comments at various stages of this research     

Reynolds' analogy for the thermal convection driven by nonisothermal stretching surfaces

In the present paper the steady boundary-layer flows induced by permeable stretching surfaces with variable temperature distribution are investigated under the aspect of Reynolds' analogy r = St _x /C _f(x). It is shown that for certain stretching velocities and wall temperature distributions, “Reynolds' function”r, i.e. the ratio of the local Stanton number St _x and the skin friction coefficient C _f(x) equals −1/2 for any value of the Prandtl number Pr and of the dimensionless suction/injection velocity f _w. In all of these cases, the dimensionless temperature field ϑ is connected to the dimensionless downstream velocity f ^′ by the simple relationship ϑ=(f ^′)^Pr. It is also shown that in the general case, Reynolds' function r may possess several singularities in f _w. The largest of them represents a critical value, so that for f _w<f _w,crit the solutions of the energy equation (although they still satisfy all the boundary conditions) become nonphysical.     

Feasibility study of three-dimensional PIV by correlating images of particles within parallel light sheet planes

One approach to obtain information about the out-of-plane velocity component from PIV recordings is to analyze the height of the peak in the correlation plane. This value depends on the portion of paired particle images, which itself depends on the out-of-plane velocity component and on other parameters. To circumvent problems with other influences (e.g. background light, amount and size of images), images from another light sheet plane parallel to the first one were also captured for peak height normalization. Our experimental results show the feasibility of an out-of-plane velocity estimation by analyzing images of particles within parallel light sheets by spatial cross-correlation.List of Symbols C particle density in the flow - d particle image diameter - f ₀, f ₁ frames containing images of particles within the first light sheet at t=t ₀ (frame f ₀) and at t=t ₀ + t (frame f ₁) - f ₂ frame containing images of particles within the second light sheet parallel to the first one at t=t ₀ + 2t - F ₁ estimator of the loss of image pairs due to in-plane motion - F ₀ estimator of the loss of image pairs due to out-plane motion - F convolution of the particle image intensity distributions - K factor containing constant parameters in the correlation plane - M imaging magnification (image size/object size) - n ₀ number of particles in the measurement volume at t=t ₀ - n _0,1 number of particle image pairs in interrogation windows of f ₀ andf ₁ - n _1,2 number of particle image pairs in interrogation windows off ₁ and f ₂ - O _z overlap of the light sheets - R _C (s) convolution of the mean intensity distributions - R _D (s) correlation which gives the image displacement - R _F (s) fluctuating noise component of the cross correlation estimator - R _0,1(s _D ) cross-correlation peak height of interrogation windows off ₀ and f ₁ - R _1,2(s _iuD) cross-correlation peak height of interrogation windows of f ₁ and f ₂ - s two-dimensional separation vector in the correlation plane - s _D mean particle image displacement in the interrogation cell - t _e light pulse duration - t _f frame-transfer time of the video camera - u three-dimensional local flow velocity vector (u,v,w) - X _i position of the center of an interrogation window in the image plane (2d) - x _i position of the center of an interrogation volume in the flow (3d) - (z ₂ — Z ₁) displacement of the light sheets in z-direction - t separation time of the light pulses - x ₀ x-extension of an interrogation volume - y ₀ y-extension of an interrogation volume - z ₀ light sheet thickness The authors would like to thank DLR for supporting Markus Raffel's and Olaf Ronneberger's visit to Caltech (Center for Quantitative Visualisation), and the Office of Naval Research through the URI grant ONR-URI-N00014-92-J-1610. Dr. Alexander Weigand's generous offer of his experimental set-up and stimulating discussions with Dr. Jerry Westerweel and Dr. Thomas Roesgen are greatly appreciated. Special thanks also to Dr. Christian Willert for his advice regarding the modifications to the DPIV software.     

Intermittency as a codimension-three phenomenon

We analyze the interaction of three Hopf modes and show that locally a bifurcation gives rise to intermittency between three periodic solutions. This phenomenon can occur naturally in three-parameter families. Consider a vector fieldf with an equilibrium and suppose that the linearization off about this equilibrium has three rationally independent complex conjugate pairs of eigenvalues on the imaginary axis. As the parameters are varied, generically three branches of periodic solutions bifurcate from the steady-state solution. Using Birkhoff normal form, we can approximatef close to the bifurcation point by a vector field commuting with the symmetry group of the three-torus. The resulting system decouples into phase amplitude equations. The main part of the analysis concentrates on the amplitude equations in R³ that commute with an action ofZ ₂+Z ₂+Z ₂. Under certain conditions, there exists an asymptotically stable heteroclinic cycle. A similar example of such a phenomenon can be found in recent work by Guckenheimer and Holmes. The heteroclinic cycle connects three fixed points in the amplitude equations that correspond to three periodic orbits of the vector field in Birkhoff normal form. We can considerf as being an arbitrarily small perturbation of such a vector field. For this perturbation, the heteroclinic cycle disappears, but an invariant region where it was is still stable. Thus, we show that nearby solutions will still cycle around among the three periodic orbits.