See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/257347894

A dynamical systems approach to the control of chaotic dynamics in a

spatiotemporal jet flow

Article  in  Chaos (Woodbury, N.Y.) · September 2013

DOI: 10.1063/1.4820819 · Source: PubMed

CITATIONS READS

6 246

3 authors:

Satish Narayanan Gemunu Gunaratne

Raytheon Company University of Houston
77 PUBLICATIONS   1,735 CITATIONS    172 PUBLICATIONS   4,678 CITATIONS   

SEE PROFILE SEE PROFILE

Fazle Hussain
Texas Tech University
462 PUBLICATIONS   23,263 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

vortex dynamics View project

Microfluidic biomechanical differentiation of tumor metastatic cells View project

All content following this page was uploaded by Fazle Hussain on 08 January 2019.

The user has requested enhancement of the downloaded file.

 A dynamical systems approach to the control of chaotic dynamics in a spatiotemporal
jet flow
Satish Narayanan, Gemunu H. Gunaratne, and Fazle Hussain

Citation: Chaos 23, 033133 (2013); doi: 10.1063/1.4820819

View online: http://dx.doi.org/10.1063/1.4820819
View Table of Contents: http://scitation.aip.org/content/aip/journal/chaos/23/3?ver=pdfcov
Published by the AIP Publishing

Articles you may be interested in

The jet in crossflow
Phys. Fluids 26, 101303 (2014); 10.1063/1.4895900

Measurements of axial instability waves in the near exit region of a high speed liquid jet
Phys. Fluids 23, 124105 (2011); 10.1063/1.3671733

Viscous linear stability of axisymmetric low-density jets: Parameters influencing absolute instability
Phys. Fluids 22, 024103 (2010); 10.1063/1.3306671

Suppression of global modes in low-density axisymmetric jets using coflow

Phys. Fluids 19, 014102 (2007); 10.1063/1.2427091

On large streamwise structures in a wall jet flowing over a circular cylinder

Phys. Fluids 16, 2158 (2004); 10.1063/1.1703531

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.118.12.124 On: Thu, 07 Jan 2016 18:31:30
 CHAOS 23, 033133 (2013)

A dynamical systems approach to the control of chaotic dynamics

in a spatiotemporal jet flow
Satish Narayanan,1,a) Gemunu H. Gunaratne,2 and Fazle Hussain3
1
Systems & Controls Engineering, United Technologies Corporation, East Hartford, Connecticut 06108, USA
2
Department of Physics, University of Houston, Houston, Texas 77204, USA
3
Department of Mechanical Engineering, Texas Tech University, Lubbock, Texas 79409, USA

(Received 3 April 2013; accepted 26 August 2013; published online 9 September 2013)
We present a strategy for control of chaos in open flows and provide its experimental validation in
the near field of a transitional jet flow system. The low-dimensional chaotic dynamics studied here
results from vortex ring formation and their pairings over a spatially extended region of the flow
that was excited by low level periodic forcing of the primary instability. The control method
utilizes unstable periodic orbits (UPO) embedded within the chaotic attractor. Since hydrodynamic
instabilities in the open flow system are convective, both monitoring and control can be
implemented at a few locations, resulting in a simple and effective control algorithm. Experiments
were performed in an incompressible, initially laminar, 4 cm diameter circular air jet, at a Reynolds
number of 23 000, housed in a low-noise, large anechoic chamber. Distinct trajectory bundles
surrounding the dominant UPOs were found from experimentally derived, time-delayed embedding
of the chaotic attractor. Velocity traces from a pair of probes placed at the jet flow exit and farther
downstream were used to empirically model the UPOs and compute control perturbations to
be applied at the jet nozzle lip. Open loop control was used to sustain several nearly periodic states.
C 2013 AIP Publishing LLC. [http://dx.doi.org/10.1063/1.4820819]
V

Excitation of fluids to enhance flow-mixing rates1–3 often
generates instabilities resulting in chaotic or turbulent
flows.4 Nonlinearities play a significant role in such highly
driven systems1,2 and traditional control algorithms
designed for linear systems5 are not always effective. In
recent years, a solution was proposed based on the obser-
vation that unstable periodic orbits (UPOs) are contained
in (the closure of) chaotic orbits.6,7 Since these periodic
orbits are embedded in the chaotic flow, the control
required to force the system towards a periodic cycle is
small.8 The approach has been validated previously in
temporal dynamical systems ranging from model sys-
tems8 to cardiac rhythms.9 In this paper, we experimen-
tally validate the methodology for a periodically driven
open fluid flow. Periodic orbits embedded in a low-
dimensional chaotic attractor found in a jet flow are com-
puted, and the chaotic flow is controlled and stabilized to
periodic cycles.
I. INTRODUCTION
Methods to perturb chaotic motion to periodic dynamics
using “small” parameter changes yield a rich variety of states
that are highly relevant for practical applications.10,11 For
example, they can be used to reduce aerodynamic drag by
manipulating turbulence structure near walls and enhance the
efficiency of combustion in aero-engines by preventing flames
from being extinguished. Control of complex irregular dy-
namics has evolved to be an important application of
a)
Author to whom correspondence should be addressed. Electronic mail:
Satish.Narayanan@utc.com
nonlinear science, and the chaos control methodology has
been successfully implemented in systems as diverse as chem-
ical reactions, electronic communications, and lasers; see Ref.
12 for a comprehensive review. For example, chaos control
was used to show how the input power for a multimode laser
could be increased to levels for significantly higher steady
power output.13 Progress has been made extending chaos con-
trol methods to spatiotemporal and stochastic dynamical sys-
tems.12,14 Petrov et al.15 studied the stabilization of periodic
states in a premixed flame, a spatially extended system. An
algorithm based on the Kuramoto-Sivashinsky equation was
proposed to stabilize and track a steady two-cell front in a pa-
rameter range where the uncontrolled system undergoes a
bifurcation sequence that terminates in chaotic dynamics.
Algorithms based on linear control theory have been
widely used to control hydrodynamic flows.16,17 However,
their applications to turbulent flows have met with limited
success, perhaps due to the failure to account for the inherent
multi-scale organized motion. The presence of low-
dimensional chaotic dynamics in open flows including cylin-
der wakes,18 plane mixing layers,19 and axisymmetric jets20
suggest that hydrodynamic phenomena relevant to jet
exhaust noise, aerodynamic drag, mixing and chemical reac-
tion in combusting flows, and flow-induced structural vibra-
tions, are amenable to chaos control.
Prior control algorithms were applied to dynamical sys-
tems governed by a few coupled ordinary differential equa-
tions, one-/two-dimensional maps, or closed flows.8,21
Control of spatiotemporal chaos was attempted in homoge-
neous systems.22,23 However, most practically relevant flows
are inhomogeneous, e.g., spatially developing; even findings
of low-dimensional dynamics, let alone nonlinear control in

1054-1500/2013/23(3)/033133/11/$30.00 23, 033133-1 C 2013 AIP Publishing LLC

V

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.118.12.124 On: Thu, 07 Jan 2016 18:31:30
 033133-2 Narayanan, Gunaratne, and Hussain
such flows have been scarce. Keefe’s24 demonstration of
control with the strategy suggested in Ref. 8 using simula-
tions of the spatially homogeneous Ginzburg-Landau equa-
tion necessitates velocity field information at several
locations. Techniques have also been evaluated to control
chaotic dynamics in direct numerical simulations of two-
dimensional turbulence governed by the incompressible
Navier–Stokes equations.25 Global and local feedback strat-
egies were used to monitor the targeted periodic states and to
provide control input to track the desired state, demonstrat-
ing control of periodic states in the model system. However,
as for Ref. 24, control required a network of spatially distrib-
uted sensing and actuation, making it difficult to realize in
practice. Despite significant advances, an experimental dem-
onstration of chaos control in an open flow is still lacking.
Two approaches to control are prevalent: feedback-based
and open-loop. The merit of feedback-based methods (e.g., see
Refs. 26 and 27) lies in their use of simple models and their
applicability to non-stationary systems. Since real-time sensing
of a local variable is used to modify the control perturbations,
sensing must be paired with nearly instantaneous actuation. In
contrast, open-loop control techniques do not require elaborate
(and cumbersome) sensing/actuation and real-time processing,
but need accurate models, may not achieve the desired dynam-
ics when the system contains multiple solutions, and are inef-
fective for non-stationary systems.28,29
A jet flow was selected for proof-of-concept since it
embodies key features associated with spatiotemporal dynam-
ical systems: the flow is spatially inhomogeneous, dispersive,
and has linear and nonlinear instabilities. Furthermore, it has
been shown that low levels of periodic forcing at a frequency
corresponding to the primary jet instability can result in low-
dimensional chaotic dynamics. The choice of the jet flow is
also motivated by its technological relevance, and the avail-
ability of many prior studies on the instabilities and dynamics
of this shear flow.30,31
Periodic forcing has been used extensively to control
large-scale vortex dynamics in jet flows.30–32 These methods
use pre-determined single/multiple frequency excitation to
induce a periodic response in the flow. In contrast, using dy-
namical invariants of the low-dimensional (chaotic) flow,
namely the periodic orbits and their eigenvalues, the chaos
control approach provides forcing to achieve periodic flow
states. Since the periodic orbit is embedded in the chaotic
attractor, the forcing required is guaranteed to be smaller than
that needed for traditional brute-force approaches used to
force the system to a pre-specified solution. Furthermore, to
obtain a specified flow response, conventional forcing meth-
ods probe the parameter space of several excitation frequen-
cies, their amplitudes and relative phase differences among
them.31,33 The excitation needed to generate a periodic
response is determined experimentally for each desired flow
state. The chaos control approach presented here is objective
and systematic in the development of the controller. Once the
low-dimensional chaotic attractor of interest is determined,
unstable periodic orbits, desired periodic states to be con-
trolled, and the perturbations for their control can be obtained
without any further testing. Detailed flow-field measurements
of the mean velocity and turbulence characteristics of the
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.118.12.124 On: Thu, 07 Jan 2016 18:31:30
Chaos 23, 033133 (2013)
controlled flow to quantify the effectiveness of the approach
are addressed elsewhere.34
The paper is organized as follows: Sec. II describes the
chaotic attractor obtained for the jet flow and the underlying
UPOs. In Sec. III, the chaos control approach and implemen-
tation method is described. Experimental results from chaos
control in the jet flow system are then discussed. Concluding
remarks are presented in Sec. IV. Appendix describes the
adaptive filtering technique used in Sec. III to determine con-
trol perturbations.
II. ANALYSIS OF CHAOTIC DYNAMICS IN THE
CIRCULAR JET
Using a coherence-based spatial coupling measure, it
was shown that the dynamics of the forced jet are coupled
for a region extending from the jet nozzle exit to more than 4
jet diameters downstream.35 Spatial coupling implies the
predictability of dynamics at one location from observables
at another. Consequently, temporal attractors, reconstructed
from single-point measurements, suffice to describe the
low-dimensional dynamics. The physical mechanism for
coupling in the jet flow is verified to be feedback from large-
scale vortex motion comprising vortex pairings, caused by
upstream propagating pressure perturbations;20 such internal
feedback can be detected in induced velocity signals at the
nozzle exit. Evidence of such coupling is an important factor
in the analyses presented below.
A. The chaotic attractors
Single-frequency bulk excitation of a 4 cm diameter axi-
symmetric jet (housed in a low-noise anechoic chamber)
reveals at least two periodic and two chaotic attractors in the
parameter space of fundamental forcing amplitude af ( u0f/
Ue) and forcing frequency StD ( f D/Ue),20 where u0f is the
exit-centerline rms velocity fluctuation at the excitation fre-
quency f, Ue is the exit velocity, and D is the jet diameter.
Two tungsten-rhodium hot-wires of 4 lm diameter were
used (with 25 mm and 5 mm long prongs) at a 1.4 overheat
ratio for the longitudinal flow velocity component measure-
ments. To maintain precise amplitude ratios of various fre-
quency components in the control signal at the nozzle exit,
an additional excitation facility was installed.34 A 1500 acous-
tic speaker is enclosed in a box packed with fiberglass wool
and connected through 12 tygon tubes of equal length (ffi 30 )
to twelve 0.7500 diameter holes in a casing surrounding the
nozzle, terminating in a 0.6 mm wide uniform slit around the
nozzle lip. Measurements of the transfer function of the exci-
tation facility were used to confirm its uniformity over the
frequencies of interest, i.e., less than 10 dB variation and no
discrete peaks over the entire frequency range of interest
[10–500 Hz]. A comprehensive description of the facility
and data acquisition procedures employed is provided in
Refs. 20 and 34.
Figure 1 schematically displays the spatiotemporal jet
dynamics: vortex ring formation and multiple pairings,
which feedback perturbations to the nozzle lip to trigger and
sustain subsequent structure formation. Such vortex dynam-
ics are of considerable technological interest, e.g., mixing
 033133-3 Narayanan, Gunaratne, and Hussain
and noise generation.32,36 The transition region extends from
the jet nozzle exit plane (the point-of-receptivity for instabil-
ities) to 4–6D, where the flow develops significant three-
dimensionality, leading to the generation of fine-scale
random vorticity and turbulence.
Due to the lack of evidence of low-dimensional dynam-
ics in the unforced jet, small levels of periodic forcing at a
single frequency were needed to reveal low-dimensional
chaotic dynamics.20 In addition to periodic vortex formation
(sustained by the low level forcing), the chaotic attractor is
found to involve up to two self-sustained quasi-periodic/cha-
otic vortex pairings.20 A chaotic attractor termed quarter-
harmonic chaotic attractor (QCA) was selected for this study,
since it has low-dimensional dynamics of both the first and
the second pairings.
1. QCA
This chaotic attractor is found for 0.008  af  0.02 and
1.1  StD  1.3 with an embedding dimension of 4 revealing a
correlation dimension in the range 2.5    3 and a positive
Lyapunov exponent.20 The longitudinal velocity time trace
u(t) and its power spectrum Guu(f) in Figs. 2(a) and 2(b) are
from a sensor located at x/D ffi 2 (on the jet flow-field center-
line) for af ffi 1% and StD ffi 1.2; here, the Reynolds number
ReD ffi 23 000; such chaotic attractors were also found for
higher ReD (76 000). (Analog-to-digital (A/D) converter units
are used for u(t).) The frequency peaks at the fundamental f,
the sidebands around the sub-harmonic, and the quarter-
harmonic f/4 are footprints of vortex formation, modulated
first pairing, and chaotic second pairing. The nearly periodic
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.118.12.124 On: Thu, 07 Jan 2016 18:31:30
Chaos 23, 033133 (2013)
FIG. 1. Schematic of the near field jet
flow illustrating the axisymmetric vor-
tex dynamics, internal feedback (re-
sponsible for spatial coupling), and the
placement of the velocity probe.
sub-harmonic modulations and the broadband f/4 indicate vor-
tex pairings whose locations shift in x due to aperiodic dy-
namics. Such associations between centerline velocity signals/
spectra and the vortex dynamics are supported by flow visual-
ization and measurements of the spatial vorticity distribu-
tion.30 For this chaotic attractor, vortex formation is
completed by x/D ffi 0.5, the first vortex pairing is typically
completed by x/D ffi 2, and the second pairing is completed
by x/D ffi 4; these events were verified by smoke-visualization
to be nominally axisymmetric.34 Henceforth, discussion of
“downstream dynamics” will be based on data acquired on the
centerline at x/D ffi 2, where the effects of vortex formation
and both pairings are comparable.
It is well known that chaotic attractors contain UPOs.29
A chaotic orbit that reaches sufficiently close to a periodic
orbit can follow it closely for a time larger than the period of
the cycle, and thus make a close return to the original. A cha-
otic orbit can reach close to and follow many UPOs embed-
ded in the strange attractor. The eigenvalues of these UPOs
can be used to estimate the dynamical invariants (such as
fractal dimension, topological entropy, and Lyapunov expo-
nent) of the chaotic motion.37 The orbits can also be used for
control.8 For fixed StD, as af is increased from very low back-
ground levels, the chaotic attractor appears first followed by
a periodic attractor at higher values of af (5–10 times large
forcing amplitude than for the chaotic attractor). Figure 3
displays a 2D Poincarè section for QCA in a plane intersect-
ing the time-delayed reconstruction of the velocity signal
data from the flow with scattered crossings that are typical of
chaotic orbits crossings.
FIG. 2. (a) Measured velocity-time
trace for QCA at x/D ¼ 2; (b) Velocity
power spectrum for QCA shows the
dynamically significant fundamental,
sub-harmonic sidebands and broadband
quarter-harmonic frequency components.
 033133-4 Narayanan, Gunaratne, and Hussain
FIG. 3. Positive (A) and negative (B) direction crossings of the chaotic orbit
with a Poincarè section.
B. The UPOs
1. UPO identification
The procedure to detect UPOs by analyzing returning tra-
jectories in phase space is similar to that developed in Ref. 38;
subsequent developments have used return plots and unstable
periodic orbits to characterize chaotic dynamics.39 The delay
co-ordinates are used to construct an m-dimensional represen-
tation of the velocity; u(t) is defined as an m-dimensional vec-
tor with components u(t), u(t þ s), u(t þ 2s),… u(t þ (m  1)s).
The time delay s corresponds to that for the first minimum of
the mutual information.40 A false-nearest-neighbor algorithm41
is used to estimate the attractor’s minimum embedding dimen-
sion m. We consider the m-dimensional orbits and study their
crossings with a suitable Poincarè section: specifically, we con-
sider the intersection when the orbits pass from a specified side
of the Poincarè section to the other. Iterates of a UPO will
return to the initial point on the Poincarè section. We assume
that close returns of the orbit on the Poincarè section indicate
the presence of a UPO. Thus, we search for the smallest recur-
rence time T such that jj u(t þ T)  u(t) jj < e1.
The uneven scatter in the 2D projection of the Poincarè
section from the chaotic attractor (Fig. 3) is expected due to
the low-dimensional dynamics; the time series is 300 000
data points and, for well-resolved Poincarè sections, is
sampled at 20 times the fundamental frequency. The u(t þ 2s)
¼ u(t) þ u(t þ s) plane (i.e., passing through the u(t þ 2s) axis
and intersecting the [u(t), u(t þ s)] plane at 45 ) is used to
obtain the 3D Poincarè section. The histogram of UPOs in
Fig. 4 is determined from the Poincarè section with e1 ffi 5%,
a fraction of the range of the Poincarè section coordinates;
the UPOs shown in Fig. 4 comprise 53% of all the observed
crossings of the Poincarè section. The recurrence times are
normalized by the fundamental period, the smallest signifi-
cant time scale in the transition region. A detailed exploration
of the size of the return neighborhoods e1 across different
embedding dimensions for the chaotic attractor was per-
formed. The embedding dimension of m ¼ 4 was found to be
the lowest dimension in which unambiguous evidence of
low-dimensional chaos was obtained (also see Ref. 20 for
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.118.12.124 On: Thu, 07 Jan 2016 18:31:30
Chaos 23, 033133 (2013)
FIG. 4. Histogram displays the dominant UPOs in QCA. The inset shows
the number of UPOs found by varying e1, for period-4, period-11, period-13,
and period-15 orbits.
embedding dimension sensitivity studies), resulting in the
choice of a 3D Poincarè section. e1 ffi 5% was chosen since
the number of UPOs of a given period levels off for e1  8%
(see inset in Fig. 4). This choice provides a sufficiently large
sample of periodic orbits without introducing potentially spu-
rious effects of averaging across multiple trajectories. The
selection of the return neighborhood size (e1) was also verified
by analysis of the Jacobian associated with the unstable fixed
point in the Poincare section (see UPO stability analysis pro-
vided later) to ensure a single positive Lyapunov exponent.
For a selected period, Poincarè section crossings are
found to cluster;34 i.e., successive orbit returns are confined
to a few neighborhoods in the Poincarè section. For most of
the UPOs studied here, two to three clusters were found, sug-
gesting orbits of a certain period and within one of these
clusters that were similar to each other.34 The time traces
corresponding to the close returns of a certain period reveal
striking similarity (discussed further in Sec. II B 1).
2. UPO stability
Since a positive Lyapunov exponent is a necessary con-
dition for chaos, periodic orbits within the chaotic attractor
will be unstable. The local stability of a UPO in phase space
can be determined from successive linear mappings along
the orbit until it returns to its initial neighborhood, or more
easily by a linear approximation near the corresponding
unstable fixed point in a Poincarè map.37,42,43 In terms of
the Jacobian A (an (m  1)  (m  1) matrix), we have
[xiþn  x*] ffi A [xi  x*], where xi denotes a crossing at the
(m  1) dimensional Poincarè map and x* is a period-n fixed
point. When more than one cluster in the Poincarè map exists
with the same return period, A is computed for the centroid
x* of each cluster. We compute the eigenvalues kj of A,
which is estimated using a least-mean-square-error algo-
rithm;44 at least 10 realizations around each fixed point were
used. All the fixed points in the 3D QCA Poincarè map are
found to be of the saddle type, i.e., one unstable (jk1j > 1), as
expected for chaotic dynamics, and two stable (jk2j, jk3j < 1)
eigenvalues.
 033133-5 Narayanan, Gunaratne, and Hussain Chaos 23, 033133 (2013)

3. Physical significance of UPO
The transitional jet dynamics can be interpreted in terms
of the UPO obtained from centerline velocity signals, assum-
ing nominally axisymmetric vortex dynamics. The spatial
sequence of the vortex dynamics can be inferred from the ve-
locity trace along the time axis in the reverse direction, in
the spirit of Taylor’s hypothesis. This correspondence is
qualitative since the relative orientation, spacing, and
strengths of vortices could change during the structure pas-
sage over a stationary sensor.
The period-9 orbit realization in Fig. 5(a) displays domi-
nant subharmonic and quarter-harmonic components in the
induced velocity from advecting once-paired vortices and
vortices undergoing a second pairing. This orbit corresponds
to a sequence of four once-paired vortices, undergoing two
second pairings, and an unpaired vortex (numbered 4 in Fig.
5(a) inset). The second pairing process is evident from the
alternating peak-to-peak amplitudes of adjacent sub-
harmonic period segments in the centerline signal; the ampli-
tude variation arises from the different radial locations of the
vortex ring cores during pairing (see inset in Fig. 5(a)). After
9 periods, the unstable nature of the chaotic orbit causes the
divergence of this period-9 trajectory into the neighborhoods
of other UPOs (with the same or different periods); i.e., the
flow undergoes a different sequence of vortex ring interac-
tions. Period-9 orbits belonging to the other clusters resem-
ble that in Fig. 5(a). The signal corresponding to a period-11
orbit in Fig. 5(b) reflects a sequence of five first pairings fol-
lowed by a single unpaired vortex. The period-11 signals in
Fig. 5(c) (from a second Poincarè section cluster) also reflect
a similar sequence of dynamics, but the relatively stronger
quarter-harmonic component implies advanced stages of sec-
ond pairing (similar to the inset of Fig. 5(a)). Snapshots of
period-15 UPOs (Fig. 5(d)) indicate seven first pairings, six
of which undergo second pairings, and an unpaired vortex.
The significance of the multiple realizations and the symbols
shown in Figs. 5(c) and 5(d) are discussed in Sec. II C.
In summary, the chaotic dynamics sweep through sev-
eral unstable, nearly periodic states. Note that these can be
realized as stable periodic attractors but require much higher
af (5–10 times larger than required to establish chaotic

FIG. 5. (a) Period-9 UPO; the inset illustrates advecting paired vortices and the process of second vortex pairings (responsible for the quarter-harmonic
component in the signal); (b) A period-11 UPO after crossing a Poincare section cluster shows a dominant sub-harmonic period; (c) The solid lines are all the
period-11 close returns from a different cluster, indicating resemblance of trajectories within a bundle; (d) The solid and dashed lines are three realizations of
period-15 UPOs from a Poincare section cluster; the result of averaging UPOs crossing cluster II is denoted by the symbols. The UPO time periods have been
normalized by the fundamental period.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.118.12.124 On: Thu, 07 Jan 2016 18:31:30
 033133-6 Narayanan, Gunaratne, and Hussain
FIG. 6. Distribution of UPOs detected in the upstream signal is similar to
that in Fig. 4.
dynamics) and different StD.12 Control seeks to stabilize
these unstable states for lower, fixed af and StD.
C. The averaged UPO representation
A key feature of the UPOs is that the close returns are
found to be nearly identical, with only slight variations in
amplitude and/or phase [see Figs. 5(c) and 5(d)]. The orbits
crossing the clusters remain in “bundles” and evolve in a
nearly identical manner in phase space. At most three dis-
tinct clusters in the Poincarè section are found surrounding
all UPO crossings with periods lower than 15. Studies
have shown that low period UPOs yield the coarsest fea-
tures of chaotic attractors and successively higher period
UPOs that tend to be scattered reveal finer details.37 Thus,
UPOs with low periods visit only few neighborhoods of
the attractor.
After distinct clusters in a single Poincarè section are
identified, UPOs are recorded for the time interval between
their origin from a cluster and their return. Figure 5(c) dis-
plays all period-11 UPOs crossing a cluster in the Poincarè
section, revealing striking similarities. The differences
between the UPOs are most pronounced just after 6 funda-
mental time periods, when a relatively distinct isolated fun-
damental period is seen. The physical configuration
comprising an unpaired vortex between two stronger, once-
paired vortices is extremely sensitive to ambient perturba-
tions. This may explain why it is impossible to repeat the
same sequence of vortex dynamics in every sequence. In
phase space, the underlying UPOs display slight differences.
Use of smaller e1 reduces the differences but does not elimi-
nate them. Only three period-15 orbits a Poincarè section
cluster are shown in Fig. 5(d) to better display their average.
Similar results are obtained for other UPOs. Orbits crossing
distinct clusters are averaged, removing slight variations in
amplitude/phase, to reveal the underlying UPO. Figure 5(d)
also displays the averaged period-15 UPO for QCA (indi-
cated by symbols), which is virtually identical to other
orbits from the cluster. Such will obviously be the case in
Fig. 5(c) too.
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.118.12.124 On: Thu, 07 Jan 2016 18:31:30
Chaos 23, 033133 (2013)
These results suggest that orbits neighboring a UPO
evolve in a nearly identical way. The clustering of UPO
crossings signifies preferential occurrence of certain sequen-
ces of vortex formation and interactions; thus, low forcing
levels will suffice to maintain these UPOs for control (see
Sec. III B). Since UPOs of smaller periods are typically the
dominant ones (see Fig. 4), these represent the dominant
flow dynamics. The spirit of averaging UPOs described here
resembles ensemble averaging of multi-point vorticity sig-
nals to describe organized vortical motion,30 but differs in
that the procedure employed here does not require explicit
phase-alignment and thresholding of the signals (both sub-
jective features of the conventional flow eduction proce-
dure). Such averaging of trajectories near UPOs—phase-
space eduction—can be used to describe the evolution of
dynamically significant flow structures (as in Sec. II B). In
particular, using a weighted average of these UPOs (e.g.,
based on the number of close returns found or the magnitude
of the most unstable eigenvalue), one can construct a
dynamic model.
III. THE CHAOS CONTROL STRATEGY
Convective instabilities in open flows are generally
undesirable due to their noise-sensitivity. However, such
instabilities can be advantageous for control. By providing
localized perturbations at the point-of-receptivity with suffi-
cient amplitude to dominate over ambient disturbances, one
can effect significant changes in the downstream dynamics;
conventional flow control methods exploit this feature.32
Having chosen the UPO to be stabilized, we achieve
control by perturbing the flow at the flow origin to maintain
the attractor dynamics observed downstream near the
selected UPO. Since a chaotic trajectory visits many UPOs,
given sufficient time, the attractor dynamics can be expected
to approach the low order UPOs (which typically have
smaller eigenvalues than orbits of larger period); such
“dominant” UPOs are visited frequently. Since evolution of
orbits close to a specific UPO are similar (see Sec. II C), in
the absence of “noise” (i.e., background disturbances within
the instability bandwidth), the desired dynamics will be self-
sustained for at least one orbital period. Low-level control
perturbations are then needed to prevent the dynamics from
wandering away due to the unstable nature of the orbit. For a
convectively unstable flow, continuous control is needed to
overpower ambient perturbations, since the growth and
advection of disturbances leaves the flow origin susceptible
to new perturbations, i.e., “noise.” As noted earlier, pairing
feedback induces spatial coupling in this open flow; i.e.,
feedback perturbations generally dominate over broadband
ambient disturbances. It is expected that, for a chaotic state,
although perturbations from UPOs are fed back to the origin
(primarily from vortex pairings downstream), they are not
strong enough to sustain phase-locking, i.e., periodic behav-
ior. The control provides additional energy at the appropriate
frequencies to enable locking to a specified flow state, i.e.,
stabilize the selected UPO. Appropriate upstream perturba-
tions at the receptivity point are needed for control, and
approaches to determine this are described next.
 033133-7 Narayanan, Gunaratne, and Hussain Chaos 23, 033133 (2013)

A. The upstream control perturbations 2. Adaptive filtering

One approach to obtain a unique relationship between This technique is used in signal-processing for modeling
the upstream and downstream signals is a “nonlinear transfer control systems, and filter design.46 It was used here for
function,”45 which involves computationally intensive, error- “inverse system modeling:” using a set of weights, the rela-
prone procedures (even for quadratic nonlinearity). Spatial tionship between a pair of simultaneously sampled signals
coupling in the flow guarantees that an upstream velocity [x(t), y(t)] is approximated. Figure 7 shows this schemati-
signal contains footprints of all the UPOs. Thus, analyses of cally. The least-mean-square algorithm to determine the fil-
close-returns in the upstream signal, sampled simultaneously ter weights is described in Appendix. Using the estimated
with the downstream signal, should reveal (unstable) peri- adaptive filter and the averaged downstream UPO to be
odic orbits as well. For two-point measurements, a long- controlled xUPO(t), the corresponding upstream perturbation
prong hot-wire probe is positioned at an angle at the nozzle xorigin(t) is evaluated. Signals sampled for QCA (of at least
exit (near r/D ffi 0.25), minimizing probe interference effects, 300 000 data points) are used to evaluate the filter weights,
and a second probe is stationed, aligned with the flow, on the taking simultaneously sampled data at the nozzle exit plane
centerline at x/D ¼ 2. and at x/D ¼ 2. The stepping size l (¼ 106) and the number
of weights Nw (¼ 10 000) are chosen to yield early error con-
1. Upstream signal analysis vergence (within 5000 fundamental periods); these parame-
ters are discussed in Appendix. The error e(t) remains
The distribution of UPO signatures at the nozzle exit
converged even after further samples of x(t) and y(t) are pro-
plane (using e1 ¼ 5%) in Fig. 6 resembles Fig. 4 (for the
vided. The upstream perturbation evaluated is then averaged
downstream signal). The difference in the number of UPOs
over the time period of the UPO selected for control (ignor-
detected and the absence of some UPOs in Fig. 6 (com-
ing initial transients during convergence). This removes
pared to Fig. 4, at x/D ¼ 2) is due to the inability of
effects of slight variations in the filter weights, which con-
upstream measurements to resolve the corresponding close
tinue even after e(t) appears to converge. The averaged
returns; the additional frequency components (associated
upstream perturbation in Fig. 8 is determined by applying
with these UPOs) are much weaker (at x/D ¼ 0) in compari-
adaptive filtering on an averaged downstream period-11
son to the dominant fundamental frequency component (see
UPO. The small variations in the peak-to-peak amplitudes of
also Fig. 8). Clusters as well as trajectory bundles like those
the signal are footprints of the dynamics observed down-
analyzed in Sec. II B are found for this upstream signal as
stream; the fundamental frequency (the dominant periodic
well.
component) is evident. Since bulk-excitation (by single-
To identify the upstream signal corresponding to a
frequency forcing) is already provided to establish QCA, the
UPO obtained downstream, signals were sampled simulta-
fundamental frequency component must be eliminated from
neously at the two points. Whenever a UPO crosses a
the averaged upstream perturbation for control.
Poincarè section cluster observed in the downstream attrac-
tor, a realization is simultaneously extracted from the
B. Control experiments in the axisymmetric jet
upstream signal for the UPO period, assuming that feedback
is instantaneous and is dominant compared to background The effectiveness of chaos control is assessed by com-
disturbances (evidenced by strong spatial coupling). These paring the velocity signals u(t) from the controlled and the
realizations are then averaged from the upstream signal (as desired flows. After selecting a UPO to be controlled, the
in Sec. II C) to obtain a perturbation that will excite the averaged upstream perturbation is used to generate a periodic
flow and keep the downstream attractor dynamics in the vi- control signal (see Sec. III A). After the fundamental fre-
cinity of the desired downstream UPO. For control, this quency is filtered, this signal is provided through a digital-to-
perturbation is provided continuously with sufficient ampli- analog (D/A) converter and an adjustable external gain, to an
tude at the nozzle lip. acoustic speaker, connected through several tubes to a thin
Spatial coupling for several commonly encountered slit around the nozzle lip (see Sec. II A).
open flows (e.g., channel flows, far wakes, and boundary
layers) may not exist or are not known. However, the preva-
lence of large-scale vortices in these flows suggests low-
dimensional dynamics that can be manipulated for control.
For these flows, selective amplification of certain perturba-
tions from a broadband of local disturbances results in
nonlinear dynamics downstream. Adaptive filtering was
explored as a means to obtain the upstream control perturba-
tion. This technique can easily be extended to other open
flows such as mixing layers, wakes and boundary layers.
Note that the instantaneous feedback invoked here to
obtain the control perturbation from the upstream and down-
stream signals is not a hard requirement since any delays are
FIG. 7. Schematic of the empirical modeling of the spatiotemporal dynamics
automatically accounted for in the adaptive filtering proce- by adaptive filtering, using upstream x(t) and downstream y(t) velocity sig-
dure used. nals; z(t) is the adaptive filter prediction and e(t) indicates the prediction error.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.118.12.124 On: Thu, 07 Jan 2016 18:31:30
 033133-8 Narayanan, Gunaratne, and Hussain
FIG. 8. Adaptive filter prediction of upstream control perturbations for a
period-11 UPO.
The period-11 UPO from one of the two possible
Poincarè section clusters (also see Fig. 5(b)) is selected to be
controlled for two reasons: (i) controlling a period-11 orbit
demonstrates the effectiveness of the chaos control strategy in
stabilizing UPOs of relatively large periods, overcoming a
common shortcoming in prior chaos control methods;8 and
(ii) controlling the orbits from the selected cluster, with the
least quarter-harmonic frequency content, will suppress the
second vortex pairing. The latter results in the suppression of
quantities such as turbulence intensities and Reynolds stresses
with potential implications for reducing noise radiated from
the jet flow system.36 Evidence of low-dimensional chaotic
dynamics in the plane mixing layer19 and its similarity with
jet chaos suggests the use of pairing suppression to control
large-scale flow separation in wall-bounded flows.
The control signal is synthesized from realizations simi-
lar to that shown in Fig. 8 and is then low-pass filtered to
minimize the otherwise dominant fundamental component,
reduced here by more than two orders of magnitude. The
bulk excitation signal and that provided at the exit are phase-
locked at the fundamental frequency, which is not com-
pletely eliminated from the exit control signal. Single-
frequency bulk forcing is first employed (at af ffi 1% and
StD ffi 1.2) to establish QCA. The control signal is then con-
tinuously applied at the nozzle lip with an external gain
which is increased until the desired UPO is sustained,
observed by a downstream centerline probe (as in Fig. 1);
the flow system dynamics locks on to the selected UPO after
a transient. Ceasing control causes the flow to revert back to
the chaotic dynamics of QCA, showing that the low-level
control perturbations merely keep the dynamics in the vicin-
ity of the selected UPO. For much larger control signal am-
plitude, a period-11 flow state is observed which does not
resemble the desired UPO; i.e., a new flow state (not intrinsic
to the chaotic dynamics) is established.
Figure 9(a) displays the close match between a velocity
time trace from the controlled flow at x/D ¼ 2 and repeated
segments from a signal obtained downstream for QCA asso-
ciated with realizations of the desired period-11 UPO
(denoted by symbols), e.g., Fig. 5(b). (All signals have been
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.118.12.124 On: Thu, 07 Jan 2016 18:31:30
Chaos 23, 033133 (2013)
time-shifted to best align the desired and the controlled states
but no amplitude re-scaling.) Note that, the desired period-11
UPO is observed for only one orbital period in the chaotic
state, while it is sustained indefinitely in the controlled flow.
Interestingly, the nozzle exit signal in the uncontrolled cha-
otic flow (see dashed lines) and the control perturbation,
denoted by solid line, also match very closely (Fig. 9(b)).
Thus, the amplitude of the perturbations to achieve the
period-11 dynamics is not much larger than that in the
uncontrolled flow; this is because forcing is needed only to
preserve the period-11 flow state intrinsic to the chaotic dy-
namics. Also note the low-level control energy provided in
the power spectrum of an upstream signal from the con-
trolled flow (solid line in Fig. 9(b)). The power spectrum of
the downstream signal in the controlled flow is shown in Fig.
9(c). In comparison to the QCA power spectrum (see Fig.
2(b)), the sidebands around the sub-harmonic frequency are
sharper, the modulation frequency, now at 24 Hz, is stronger
(30 dB above background), and the broadband quarter-
harmonic is uniformly suppressed by at least an order of
magnitude (except for weak modulation frequencies at 48 Hz
and 72 Hz). The absence of the quarter-harmonic component
is also evident in the time trace in Fig. 9(a) (e.g., compare
with Fig. 9(c)). Moreover, the broadband spectral back-
ground is uniformly suppressed by at least 10 dB. The broad-
band pedestal remaining below the sidebands around the
subharmonic frequency is due partly to slight aperiodicity in
the controlled flow and partly to spectral leakage when
resolving nearby frequencies of discrete power spectra.
Figure 9(d) shows the mutual information from QCA
(dotted line) and from the controlled flow. This displays the
rapid loss of information in the chaotic flow and the recovery
of information at a delay corresponding to the period-11 orbit
in the controlled flow; the increase of the floor to almost 1 bit
is expected for the periodic controlled flow. Similar results for
period-11 control were also obtained by using the upstream
control perturbation estimated by an adaptive filter approxi-
mation (see Sec. III A). A period-11 UPO intersecting a differ-
ent cluster was chosen to assess control when multiple phase-
space bundles of the same period are present. An upstream
control perturbation resembling that in Fig. 8 was used to sta-
bilize the selected period-11 UPO. The close match between
the controlled flow and the desired period-11 UPO in Fig. 9(e)
demonstrates that the control strategy can be effective even in
the presence of more than one UPO of the same period.
The control of period-13 UPOs, Fig. 9(f), shows signals
from the controlled flow and the corresponding averaged
downstream UPO. The absence of the quarter-harmonic com-
ponent in the signal indicates nearly complete suppression of
the second vortex pairing. Due to the shorter instability wave-
lengths at these large StD (> 1), the second pairing is usually
unavoidable in the uncontrolled jet. The periodicity of these
controlled states persists up to about 4 jet diameters, follow-
ing which the jet transitions to turbulence, as also ascertained
by spatial coherence measurements.34 The selected period-13
UPO belongs to one of two Poincarè section clusters found
(not shown). The control of the period-13 UPO from the sec-
ond cluster (with dominant quarter-harmonic) was unsuccess-
ful: the flow frequently wandered away from the UPO to the
 033133-9 Narayanan, Gunaratne, and Hussain Chaos 23, 033133 (2013)

FIG. 9. (a) Controlled state closely matches the desired state (indicated by symbols) for a period-11 UPO, where “*” denotes the isolated fundamental period;
(b) Power spectra from the controlled (solid line) and the uncontrolled (dashed line) flows at x/D ¼ 0 shows the small control energy used; (c) Power spectrum
measured from the controlled flow at x/D ¼ 2 shows increased order (compared to Fig. 2(b)) and suppression of the quarter-harmonic component; (d) Mutual
information from the controlled flow signal shows an expected peak at a delay of 220 (corresponding to a period-11 UPO), compared to the rapid decay in the
chaotic flow (see dotted line); (e) Signal from the controlled flow and that corresponding to the period-11 UPO (indicated by symbols) from cluster II in Fig.
5(a) match closely; (f) Signal from the controlled flow and that corresponding to the desired period-13 UPO (indicated by symbols) match closely.

flow state shown in Fig. 9(f). This loss of control seems to be are more spread out than those for the period-11 UPO.
caused by the relative stability of the orbits in the neighboring “Adaptive” control could resolve this problem.
cluster, having a smaller magnitude of the unstable eigen- The discrepancies between the controlled flow and the
value. Moreover, the clusters for period-13 UPO crossings desired UPO [see signals in Figs. 9(a), 9(e), and 9(f) and the

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.118.12.124 On: Thu, 07 Jan 2016 18:31:30
 033133-10 Narayanan, Gunaratne, and Hussain
spread in the spectral peaks (Fig. 9(c))] indicate the occa-
sional drift of the dynamics away from the desired UPO.
These could result from: (i) the variations among orbits in
a UPO bundle, (ii) uncertainty in determining the control
perturbations, and (iii) ambient disturbances (e.g., airflow
blower oscillations, wind tunnel acoustics) within the band
of frequencies present in the exit control perturbations,
requiring further investigation.
IV. CONCLUDING REMARKS
An effective approach to control chaotic dynamics in an
open flow is proposed and demonstrated experimentally in a
circular jet flow. Inherent periodicities underlying the chaotic
dynamics are revealed, viz., UPOs, which are amenable to
control. The strategy is to provide low-amplitude perturba-
tions to maintain the dynamics in the neighborhood of a
selected UPO. The approach involves two steps. The first is
to determine the size of the region that can be driven to the
UPO. The second step, required for spatially developing
flows, is to estimate the mapping of the UPO dynamics
between the control and the monitoring locations. When the
chaotic dynamics at the two locations are strongly correlated,
perturbations at the control location associated with a selected
UPO at the monitoring location can be detected (to within a
phase delay) by simultaneous sampling. Experimental evi-
dence of UPO stabilization demonstrates success in achieving
the desirable periodic state. For long period UPOs, the accu-
mulation of ambient noise causes the orbits to be scattered on
the attractor. Open loop control is ineffective for such orbits.
Detailed flow field measurements revealing the spatial extents
over which the periodic states are maintained are described in
Ref. 34. It should be noted that control was enabled by
evidence of low dimensional chaos in the flow system for
which low levels of periodic forcing were needed. To the best
of our knowledge, this is the first experimental demonstration
of chaos control in an open flow system.
The use of our methodology hinges on the presence of
chaotic dynamics in the flow. Evidence for the existence of
low-dimensional chaotic attractors in wakes18 and transi-
tional and turbulent boundary layers47,48 is therefore promis-
ing; i.e., a desirable flow response can be obtained by
tracking an unstable periodic orbit underlying the chaotic
attractor. For instance, Coller et al.49 have demonstrated
feedback control of an unstable fixed point via simulations in
a “reduced model” of boundary layer turbulence, believed to
be associated with near-wall longitudinal vortices. Using
simulations of a low-dimensional temporal dynamical sys-
tem model representing a shear flow bounded in a channel,
chaos control was also demonstrated for unstable fixed points
believed to be associated with streamwise vortices and wavy
streaks.50 For boundary layers, due to the spatiotemporal na-
ture of the flow and the lack of a single (well defined) point
of receptivity (as for free shear flows), in practice actuation
at several downstream locations may be necessary for con-
trol. The spatial extent of correlation in the flow, which can
be quantified using coherence measurements,35 will dictate
the sensor placement for reconstructing the chaotic attractor
and hence the actuator locations. The reconstruction of a
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.118.12.124 On: Thu, 07 Jan 2016 18:31:30
Chaos 23, 033133 (2013)
low-dimensional dynamical system from multipoint meas-
urements is needed; e.g., see Ref. 51 for such a study.
There are several technologically beneficial applications
of chaos control in open shear flows. Jet control can be used
to enhance mixing in combusting flows and to suppress tur-
bulence for aerodynamic noise reduction.36,52 Controlling
vortex pairings in free shear flows enable mixing control53
and separation control during slot injection over airfoils.54
Chaos control of the organized near-wall flow structures in
wall-bounded shear flows could be used to achieve drag
reduction and heat transfer enhancement.55
APPENDIX: LEAST-MEAN-SQUARE ALGORITHM
FOR ADAPTIVE FILTERING
The adaptive filter weights (see Sec. III) are computed
using a least-mean-square algorithm;45 these weights are
used to evaluate upstream control perturbations for desirable
downstream dynamics, i.e., as an “inverse” model. The sim-
plest gradient estimate for the kth iteration of this steepest
descent type algorithm is: Wkþ1 ¼ Wk þ 2 l ek Xk, where l
is the gain constant that regulates the speed and stability of
the adaptation, ek is the error (ek ¼ dk  XTk Wk), Wk is the
weight vector with elements {w1, w2, …, wNw}, Xk is the
input vector, dk is the actual (output) response for testing the
adaptive filter prediction, and Nw is the number of weights.
The two criteria to evaluate the accuracy of an adaptive filter
are: (i) convergence of the envelope of ek (within 610% of
peak-to-peak signal amplitude), and (ii) convergence of the
rms of all the weights (R wi2)k. The adaptive filter prediction
is also inspected (expected to be periodic for each UPO) to
ensure that no major changes occur in its envelope. To con-
tinue refining the approximated weights, the input/output
pair of signals is provided to the algorithm even after achiev-
ing acceptable convergence of the weights. The actual
upstream signal (from a second set of two-point data) was
also compared with that predicted using the adaptive filter
weights (computed using the originally sampled two-point
QCA data) to ensure reliable adaptive filter predictions.
1
S. Chandrasekhar, Hydrodynamic and Hydromagnetic Stability (Dover
Publications, Inc., New York, 1961).
2
P. G. Drazin and W. H. Reid, Hydrodynamic Stability (Cambridge
University Press, Cambridge, 1981).
3
A. C. Noble, G. B. King, N. M. Laurendeau, J. R. Gord, and S. Roy,
Combust. Sci. Technol. 184, 293 (2012).
4
M. Cross and P. Hohenberg, Rev. Mod. Phys. 65, 851 (1993).
5
G. F. Franklin, J. D. Powell, and A. Emami-Naeini, Feedback Control
of Dynamic Systems, 5th ed. (Prentice Hall, Upper Saddle River, NJ,
6
2005).
D. Auerbach, P. Cvitanovic, J. P. Eckmann, G. Gunaratne, and I.
Procaccia, Phys. Rev. Lett. 58, 2387 (1987).
7
P. Cvitanovic, Phys. Rev. Lett. 61, 2729 (1988).
8
E. Ott, C. Grebogi, and J. Yorke, Phys. Rev. Lett. 64, 1196 (1990).
9
A. Garfinkel, M. L. Spano, W. L. Ditto, and J. N. Weiss, Science 257,
1230 (1992).
10
G. Chen and X. Dong, Int. J. Bifurcation Chaos 3, 1363 (1993).
11
T. Shinbrot, Adv. Phys. 44(2), 73 (1995).
12
E. Sch€oll and H. G. Shuster, Handbook of Chaos Control, 2nd ed. (John
Wiley & Sons, 2008).
13
Z. Gills, C. Iwata, R. Roy, I. B. Schwartz, and I. Triandaf, Phys. Rev. Lett.
69, 3169 (1992).
14
O. Beck, A. Amann, E. Sch€ oll, J. E. S. Socolar, and W. Just, Phys. Rev. E
66, 016213 (2002).
 033133-11 Narayanan, Gunaratne, and Hussain Chaos 23, 033133 (2013)

15 37
V. Petrov, M. F. Crowley, and K. Showalter, J. Chem. Phys. 101(8), 6606 P. Cvitanovic, G. H. Gunaratne, and I. Procaccia, Phys. Rev. A 38, 1503
(1994). (1988).
16 38
S. C. Huang, J. Kim, and J. S. Gibson, “Adv. Turb. X,” in Proc. 10th Eur. D. P. Lathrop and E. J. Kostelich, Phys. Rev. A 40, 4028 (1989).
39
Turbulence Conf., Trondheim, Norway, edited by H. I. Andersson and P.- E. Bradley and R. Mantilla, Chaos 12, 596 (2002).
40
A. Krogstad (2004). A. Fraser and H. L. Swinney, Phys. Rev. A 33, 1134 (1986).
17 41
J. Kim and T. R. Bewley, Ann. Rev. Fluid Mech. 39, 383 (2007). M. Kennel, R. Brown, and H. Arbabanel, Phys. Rev. A 45, 3403 (1992).
18 42
C. W. Van Atta and M. Gharib, J. Fluid Mech. 174, 113 (1987). A. L. Belmonte, M. J. Vinson, J. A. Glazier, G. H. Gunaratne, and G.
19
S. Narayanan and F. Hussain, J. Fluid Mech. 320, 71 (1996). Kenny, Phys. Rev. Lett. 61(5), 539 (1988).
20 43
G. Broze and F. Hussain, J. Fluid Mech. 263, 93 (1994). G. H. Gunaratne, P. S. Linsay, and M. J. Vinson, Phys. Rev. Lett. 63(1), 1
21
J. Singer, Y.-Z. Wang, and H. H. Bau, Phys. Rev. Lett. 66, 1123 (1991). (1989).
22 44
D. Auerbach, Phys. Rev. Lett. 72, 1184 (1994). M. Sano and Y. Sawada, Phys. Rev. Lett. 55, 1082 (1985).
23 45
I. Aranson, J. Levine, and L. Tsimring, Phys. Rev. Lett. 72, 2561 (1994). Ch. P. Ritz and E. J. Powers, Physica D 20, 320 (1986).
24 46
L. R. Keefe, Phys. Fluids A 5(4), 931 (1993). B. Widrow and S. D. Stearns, Adaptive Signal Processing (Prentice-Hall,
25
S. Guan, Y. C. Zhou, G. W. Wei, and C.-H. Lai, Chaos 13(1), 64 (2003). Inc., 1985).
26 47
K. Pyragas, Phys. Lett. A 170, 421 (1992). N. Aubry, P. Holmes, J. L. Lumley, and E. Stone, J. Fluid Mech. 192, 115
27
C. M. Postlethwaite and M. Silber, Phys. Rev. E 76(5), 056214 (2007). (1988).
28 48
E. L€uscher and A. W. H€ ubler, Helv. Phys. Acta 62, 544 (1989). J. J. Healey, J. Fluid Mech. 255, 667 (1993).
29 49
R. Shermer, A. H€ubler, and N. Packard, Phys. Rev. A 43, 5642 (1991). B. D. Coller, P. Holmes, and J. L. Lumley, Physica D 72, 135 (1994).
30 50
K. B. M. Q. Zaman and F. Hussain, J. Fluid Mech. 101, 449 (1980). M. Pausch and B. Eckhardt, J. Phys.: Conf. Ser. 318, 032005 (2011).
31 51
C. O. Paschereit, I. Wygnanski, and H. E. Fiedler, J. Fluid Mech. 283, 365 S. Narayanan, S. Roy, and F. Hussain, in Proceedings of the Seventh Asian
(1995). Congress of Fluid Mechanics, Madras, India, December 8–12 (Allied
32
M. Gad-el-Hak and A. Pollard, “Flow control,” in Lecture Notes in Publishers Ltd., 1997), Vol. 1, p. 327.
52
Physics (Springer, Berlin, 1998), Vol. 53, p. 335. F. Hussain and M. A. Z. Hasan, J. Fluid Mech. 150, 159 (1985).
33 53
C.-M. Ho and P. Huerre, Ann. Rev. Fluid Mech. 16, 365 (1984). P. A. McMurtry, W.-H. Jou, J. J. Riley, and R. W. Metcalfe, AIAA J.
34
S. Narayanan, “Chaos control of an open shear flow: A novel control strat- 24(6), 962 (2012).
egy and its experimental demonstration in an axisymmetric jet,” Ph.D. dis- 54
S. Narayanan and A. Banaszuk, in Proceedings of 41st Aerospace Sciences
sertation (University of Houston, 2013). Meeting & Exhibit, Reno, NV, U.S.A., January 2003, AIAA 2003-0060
35
G. Broze, S. Narayanan, and F. Hussain, Phys. Rev. E 55, 4179 (1997). (2003).
36 55
K. B. M. Q. Zaman, J. Fluid Mech. 152, 83 (1985). J. Kim, Phys. Fluids 15, 1093 (2003).

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
129.118.12.124 On: Thu, 07 Jan 2016 18:31:30
View publication stats