1: \documentclass[aps,prl,preprint,groupedaddress,floatfix]{revtex4}

2: %\documentclass[aps,prl,twocolumn,groupedaddress,floatfix]{revtex4}

3: 

4: %

5: %  Notation

6: %

7: \usepackage{graphicx}% Include figure files

8: \usepackage{dcolumn}% Align table columns on decimal point

9: \usepackage{bm}% bold math

10: 

11: \newcommand{\Rey}{\mbox{\it Re}}           % Reynolds number

12: \newcommand{\fmode}{\tilde{\bf u}}         % Floquet mode

13: \newcommand{\base}{{\bf U}}                % Base flow

14: \newcommand{\perb}{{\bf u}'}               % Perturbation

15: \newcommand{\Karman}{von~K\'{a}rm\'{a}n }

16: \newcommand{\eg}{e.g.\/}

17: \newcommand{\ie}{i.e.\/}

18: 

19: \begin{document}

20: 

21: \bibliographystyle{prsty}

22: 

23: %

24: %==================== TITLE ====================

25: %

26: 

27: \title{Confined three-dimensional stability analysis of the cylinder wake}

28: 

29: \author{Dwight Barkley}

30: \email[]{barkley@maths.warwick.ac.uk}

31: \homepage[]{www.maths.warwick.ac.uk/~barkley}

32: %\thanks{}

33: %\altaffiliation{}

34: \affiliation{Mathematics Institute, University of Warwick, 

35: Coventry CV4 7AL, United Kingdom}

36: 

37: \date{\today}

38: 

39: %

40: %==================== ABSTRACT ====================

41: %

42: \begin{abstract}

43: A spatially confined stability analysis is reported for the cylinder wake at

44: Reynolds numbers 190 and 260.  The two three-dimensional instabilities at

45: these Reynolds numbers are shown to be driven by the flow just behind the

46: cylinder.

47: \end{abstract}

48: 

49: \pacs{47.20.Ft, 47.27.Vf} 

50: 

51: \maketitle

52: 

53: %==================== INTRODUCTION ====================

54: 

55: Consider the \Karman vortex street generated by flow past an infinitely long

56: circular cylinder. This flow is time periodic and two dimensional for Reynolds

57: numbers between approximately 47 and

58: 189~\cite{Jackson87,Provansal87,Williamson88,Miller94}.  (The Reynolds number

59: is $\Rey \equiv U_\infty d /\nu$, where $U_\infty$ is the fluid velocity far

60: from the cylinder, $d$ is the cylinder diameter, and $\nu$ is the kinematic

61: viscosity.)  At Reynolds number $189$ the 2D vortex street becomes three

62: dimensionally unstable

63: \cite{Williamson88,Miller94,Barkley96,Henderson96,Williamson96b}.  A numerical

64: stability analysis of the flow up to Reynolds number 300 has determined two

65: separate bands of linearly unstable modes \cite{Barkley96}. The first appears

66: at $Re=189$ with a spanwise wavelength of $4$ cylinder diameters and the

67: second appears at $Re=260$ with a spanwise wavelength of $0.8$ diameters.

68: These linear instabilities are related to 3D shedding modes known as mode A

69: and mode B, first observed experimentally by Williamson \cite{Williamson88}.

70: 

71: In this paper I revisit the stability analysis of the vortex street. I show

72: that small regions of the full flow just behind the cylinder are responsible

73: for the 3D linear instabilities despite the fact that the actual linear modes

74: extend many cylinder diameters downstream of the cylinder.  This is important

75: for two separate reasons. The first is that this limits the regions which

76: should be analyzed either to clarify instability mechanisms or to suppress the

77: instabilities if so desired.  The second reason is that the stability analysis

78: of small flow regions is computationally very efficient compared with a

79: stability analysis of the full flow field. Thus this approach provides a

80: method for quickly obtaining approximate stability information about a complex

81: flow.

82: 

83: The computational approach is described fully elsewhere \cite{Barkley96}.  I

84: summarize briefly the main points with focus on the important aspects for the

85: current work.  First direct simulations of the incompressible Navier--Stokes

86: equations are used to obtain 2D, time-periodic wake flows: $\base(x,y,t+T) =

87: \base(x,y,t)$ where $T$ is the wake period. A spectral-element method is

88: employed on a computational domain shown in Fig.~\ref{fig:mesh}.  The length

89: scale is the cylinder diameter.  The size of the computational domain depends

90: on the Reynolds number and the accuracy required \cite{Barkley96}. The domain

91: shown gives sufficiently accurate results for the purposes of this work.

92: 

93: \begin{figure}

94: \begin{center}

95: \includegraphics[width=3.2in]{mesh.ps}

96: \end{center}

97: \caption{{} Spectral-element computational mesh.  The base flow is computed on

98: the full domain and stability computations are performed on the full domain and

99: on a variety of subdomains. The subdomain outlined in bold and shown as an

100: enlargement has inflow length $L_i=1.5$, outflow length $L_o=4$, and

101: cross-stream height $L_y=1.5$.  }

102: \label{fig:mesh} 

103: \end{figure}

104: 

105: The boundary conditions imposed for the computation of $\base$ are as follows.

106: On the top, bottom, and left boundaries uniform flow is imposed $\base = (U,V)

107: = (1,0)$.  At the right edge a (Neumann) outflow boundary condition for the

108: velocity and pressure $P$ is used:

109: %

110: \begin{equation}

111: \label{bcs}

112: 	\partial U/\partial x = 0, \quad

113: 	\partial V/\partial x = 0, \quad 

114: 	                    P = 0. 

115: \end{equation}

116: %

117: No-slip conditions are imposed at the cylinder surface.

118: 

119: The next step is a Floquet stability analysis of 3D disturbances to the 2D

120: wake. This analysis is based on the evolution of infinitesimal 3D

121: perturbations $\perb(x,y,z,t)$ of the 2D flow $\base(x,y,t)$. The equation for

122: such a perturbation is obtained by linearizing the Navier--Stokes equations

123: about $\base$:

124: %

125: % equation

126: %

127: \begin{equation}

128: \label{eqn:linear}

129: \frac{\partial{\bf u}'}{\partial t} =

130:      -(\base\cdot\nabla)\perb - (\perb\cdot\nabla)\base

131:      -\frac{1}{\rho}\nabla p' + \frac{1}{\Rey}\nabla^2\perb,

132: \end{equation}

133: %

134: where $p'$ is the perturbation to the pressure that enforces $\nabla\cdot\perb

135: = 0$.  

136: 

137: In this work I shall primarily consider Eq.~(\ref{eqn:linear}) posed on

138: subdomains such as that illustrated in Fig.~\ref{fig:mesh}. These subdomains

139: are characterized by their inflow and outflow lengths $L_i$ and $L_o$,

140: measured with respect to the cylinder center, and their cross-flow range $-L_y

141: \le y \le L_y$.  The base flow $\base$ appearing in Eq.~(\ref{eqn:linear}) is

142: obtained by restricting the base flow from the full domain to the appropriate

143: subdomain.

144: 

145: Boundary conditions on ${\bf u}'$ are required by Eq.~(\ref{eqn:linear}).

146: Consider the standard case in which ${\bf u}'$ is computed on the same domain

147: as $\base$. Then the requirement that $\base + \perb$ satisfy the same

148: boundary conditions as $\base$ gives homogeneous boundary conditions on ${\bf

149: u}'$: ${\bf u}'= (0,0,0)$ on all domain boundaries except the outflow boundary

150: where ${\bf u}'$ satisfies the analogy of Eq.~(\ref{bcs}).  One can view these

151: boundary conditions as the requirement that the perturbation be zero at the

152: inflow and lateral boundaries of the domain but that perturbations can advect

153: out of the domain.  Based on this reasoning we impose these homogeneous

154: boundary conditions when computing ${\bf u}'$ on subdomains. We are

155: interested in examining perturbations ${\bf u}'$ which are zero on all

156: boundaries of the subdomain except the outflow boundary.  Perturbations are

157: allowed to advect out of the subdomain.

158: 

159: From the linearized equations the spectrum of Floquet multipliers $\mu$ and

160: corresponding Floquet modes $\fmode$ can be found \cite{Barkley96}.

161: Exponentially growing perturbations correspond to multipliers outside the unit

162: circle in the complex plane ($|\mu|>1$).  Because the geometry is homogeneous

163: in the spanwise direction, Floquet modes decouple in spanwise wavenumber

164: $\beta$.  The Floquet multipliers are thus computed as a function of $\beta$

165: on a number of subdomains for $\Rey =190$ and $\Rey = 260$, just above the

166: critical Reynolds numbers for the 3D instabilities.

167: 

168: %\section{Results}

169: 

170: Figure~\ref{fig:mults} summarizes the study in terms of multiplier spectra

171: while Figs.~\ref{fig:R190}-\ref{fig:R260} shows representative Floquet modes.

172: A large number of subdomains have been studied but only representative cases

173: are shown near the minimum dimensions necessary to capture the 3D

174: instabilities.

175: 

176: Figure~\ref{fig:mults} primarily demonstrates the effect of outflow length

177: $L_o$ on the Floquet multiplier spectrum.  The inflow length and cross-stream

178: height are fixed at minimal values of $L_i=0$ and $L_y=1.5$.  One sees that

179: $L_o=3$ is sufficient to capture the destabilizing Floquet mode. Increasing

180: $L_o$ further has minimal effect.  For $\Rey=190$, decreasing $L_o$ to $2.25$

181: causes a substantial change to the spectrum -- the subdomain is evidently too

182: small to capture all the relevant flow features which drive the instability.

183: In the case of $\Rey=260$, decreasing $L_o$ to $2.25$ causes only a small

184: change to the spectrum and the subdomain is essentially still large enough to

185: capture relevant flow features.  Decreasing the outflow length to $L_o = 1.5$

186: causes the spectrum in both cases to deviate wildly from the correct behavior

187: (at $\Rey=260$ the multipliers become complex with large magnitude and cannot

188: be plotted on the scale of Fig.~\ref{fig:mults}).

189: Figs.~\ref{fig:R190}-\ref{fig:R260} shows just how well the Floquet modes are

190: captured by the subdomain computations even though the Floquet mode on the

191: full domain extends far downstream of the cylinder \cite{Barkley96}.

192: 

193: Increasing the inflow length $L_i$ has some effect on the spectrum of

194: multipliers in the case of $\Rey=190$. However, the effect is very small at

195: the critical wavenumber where the multiplier branch has largest

196: magnitude. Increasing the inflow length has almost no effect in the case

197: $\Rey=260$.  Increasing the cross-stream height has very little effect in

198: either case.  It is possible, in fact, that the cross-stream height could be

199: reduced slightly from that used here without compromising the accuracy of the

200: multipliers. 

201: 

202: It is reasonable to conclude that the physical region responsible for the

203: instability is localized within the subdomain in those cases where the Floquet

204: multipliers and Floquet modes from subdomains are close to those from the full

205: domain.  This localization of the instability is similar to the transition

206: from steady flow to vortex shedding which takes place at the primary wake

207: instability. In that case there is a local, absolutely unstable region

208: extending approximately three to four diameters downstream

209: \cite{Hannemann89,Oertel90}, very similar to the scales observed here.  In

210: fact the current computational analysis is similar in spirit to that of

211: Hannemann and Oertel \cite{Hannemann89} and can be thought of as situated

212: between the local analysis of a parallel wake profile and the global analysis

213: previously reported for the secondary instability to three dimensionality

214: \cite{Barkley96}.

215: 

216: \begin{figure}

217: \begin{center}

218: \includegraphics[width=3.0in]{mults.ps}

219: \end{center}

220: \caption{{}Summary of Floquet multipliers at $\Rey =190$ and $\Rey = 260$ (as

221: labeled) for $\beta$ near the critical value in each case.  Solid curves are

222: from computations on the full domain. Points are from subdomains with

223: $L_i=0$, $L_y=1.5$, and $L_o=3$ (squares), $L_o=2.25$ (crosses), and $L_o=1.5$

224: (dots).  The multipliers for $\Rey=260$, $L_o=1.5$ are off the scale of the

225: figure.  }

226: \label{fig:mults} 

227: \end{figure}

228: 

229: Finally, it is worth considering the stability of the downstream portion of

230: the flow.  In this case the base flow is projected onto the subdomain with

231: $2.25 \le x \le 25$ and $-4 \le y \le 4$.  Both at $\Rey=190$ and $\Rey=260$

232: the flow in this downstream region is found to be very stable. The maximum

233: modulus of the Floquet multipliers are $|\mu| \simeq 0.28$ at $\Rey=190$,

234: $\beta=1.6$ and $|\mu| \simeq 0.19$ at $\Rey=260$, $\beta=7.5$.

235: 

236: \begin{figure}

237: \begin{center}

238: \includegraphics[width=3.2in]{R190.ps}

239: \end{center}

240: \caption{{}Floquet modes at $\Rey=190$ with $\beta = 1.6$. The main figure

241: shows (a portion of) the mode computed on the full domain. Insets show modes

242: computed on the subdomains indicated. (The inflow boundary $L_i=0$ and the

243: lateral boundaries $L_y=1.5$ are the same for all subdomains.)  All modes are

244: plotted as spanwise vorticity contours at the same fixed time. The spanwise

245: vorticity contours $\pm 0.4$ of the base flow are also plotted.  }

246: \label{fig:R190} 

247: \end{figure}

248: 

249: \begin{figure}

250: \begin{center}

251: \includegraphics[width=3.2in]{R260.ps}

252: \end{center}

253: \caption{{}Floquet modes at $\Rey=260$ with $\beta = 7.7$ using the same

254: conventions as Fig.~\ref{fig:R190}}

255: \label{fig:R260} 

256: \end{figure}

257: 

258: In this paper I have shown how a local Floquet stability analysis of small

259: regions of the wake just behind a cylinder are sufficient to capture both 3D

260: instabilities observed in wake transition.  The specific results are of

261: interest for future attempts to understand better the physical mechanisms of

262: 3D transition \cite{Williamson96b,Brede96,Persillon98,Leweke98,Thompson01}

263: because the results set limits to the regions which could drive the

264: instabilities.  The general method is of interest for future computational

265: studies of flow instabilities.

266: 

267: Much of this work was performed while visiting the Courant Institute of

268: Mathematical Sciences, NYU.  I thank members of the Institute for their

269: hospitality.

270: 

271: \bibliography{pre4}

272: % Included from pre4.bbl

273: 

274: %\begin{thebibliography}{10}

275: 

276: %\bibitem[\dagger]{byline}  email:barkley@maths.warwick.ac.uk.

277: 

278: %\end{thebibliography}

279: 

280: \bigskip

281: \bigskip

282: 

283: \end{document}

284: 

285: 

286: The downstream scales are somewhat smaller than those found at similar

287: Reynolds numbers for absolute instability.  

288: 

289: