1: \documentclass[aps,twocolumn]{revtex4}

2: \usepackage{epsfig}

3: \usepackage{bm}

4: %\input REVTeXDraft1.tex

5: \def\B.#1{{\bm{#1}}}

6: \def\C.#1{{\cal{#1}}}

7: \def\pop#1{{{\partial}\over{\partial#1}}}

8: \def\popa#1#2{{{\partial#1}\over{\partial#2}}}

9: \def\p#1#2{\partial#1/\partial#2}

10: \def\i{{\rm i}}

11: \def\d{{\rm d}} 

12: \def\e{{\rm e}} 

13: \def\D{{\rm D}} 

14: \def\invr{O\left(R^{-1}\right)}

15: \def\siminvr{\sim O\left(R^{-1}\right)}

16: \def\abs#1{\left|#1\right|}

17: \def\ep{{\epsilon}}

18: \def\be{\begin{equation}}

19: \def\ee{\end{equation}}

20: \def\etal{{\em et al.} }

21: \def\R{{\rm Re}} 

22: \def\k{k} 

23: \def\Rec {${\rm R}_{\rm cr}$}

24: \def\s{\left(k_+^2 + \beta^2\right)}

25: \def\A{\left[\i (\omega_+ - k_+ U) + \nu(D^2 - k_+^2 - \beta^2) +(D\nu)

26: D\right]}

27: \def\DA{\left[-\i k_+ (DU) + (D\nu)(D^2 - k_+^2 - \beta^2) + (D^2\nu)

28: D\right]}

29: \def\beta{k_z}

30: 

31: \begin{document} 

32: 

33: \title{Stabilization of Hydrodynamic Flows by Small Viscosity Variations}

34: \author{Rama Govindarajan$^{\dag}$, Victor S. L'vov$^*$ and Itamar

35: Procaccia$^*$}

36: \affiliation{$^\dag $ Fluid Dynamics Unit, Jawaharlal Nehru Centre for

37: Advanced 

38: Scientific Research, Jakkur, Bangalore 560064, India. \\

39: $^*$ Dept. of Chemical Physics, The Weizmann Institute of Science, Rehovot

40: 76100, Israel. }

41: %%%%%%%%%%%%%%%%%%%%%%

42:  

43: \begin{abstract}  

44: Motivated by the large effect of turbulent drag reduction by minute

45: concentrations of polymers we study the effects of a weakly

46: space-dependent viscosity on the stability of hydrodynamic flows. In a

47: recent Letter [Phys. Rev. Lett. {\bf 87}, 174501, (2001)] we exposed

48: the crucial role played by a localized region where the energy of

49: fluctuations is produced by interactions with the mean flow (the

50: ``critical layer"). We showed that a layer of weakly space-dependent

51: viscosity placed near the critical layer can have a very large

52: stabilizing effect on hydrodynamic fluctuations, retarding

53: significantly the onset of turbulence.  In this paper we extend these

54: observation in two directions: first we show that the strong

55: stabilization of the primary instability is also obtained when the

56: viscosity profile is realistic (inferred from simulations of turbulent

57: flows with a small concentration of polymers).  Second, we analyze the

58: secondary instability (around the time-dependent primary instability)

59: and find similar strong stabilization. Since the secondary instability

60: develops around a time-dependent solution and is three-dimensional,

61: this brings us closer to the turbulent case. We reiterate that the

62: large effect is {\em not} due to a modified dissipation (as is assumed

63: in some theories of drag reduction), but due to reduced energy intake

64: from the mean flow to the fluctuations. We propose that similar

65: physics act in turbulent drag reduction.

66: \end{abstract}

67: 

68: \maketitle

69: 

70: \section{Introduction}

71:  

72: This paper is motivated by the dramatic effects that are observed with

73: the addition of small amounts of polymers to hydrodynamic flows.

74: While interesting effects were discussed in the context of the

75: transition to turbulence, vortex formation and turbulent transport

76: \cite{95NH}, the phenomenon that attracted the most attention was, for

77: obvious reasons, the reduction of friction drag by up to 80\% when

78: very small concentrations of long-chain polymers were added to

79: turbulent flows \cite{69Lum,00SW}. In spite of the fact that the

80: phenomenon is robust and the effect huge, there exists no accepted

81: theory that can claim quantitative agreement with the experimental

82: facts. Moreover, it appears that there is no mechanistic

83: explanation. In the current theory that is due to de Gennes

84: \cite{86TG,90Gennes} one expects the Kolmogorov cascade to be

85: terminated at scales larger than Kolmogorov scale, leading somehow to

86: an increased buffer layer thickness and reduced drag, but how this

87: happens and what is the fate of the turbulent energy is not being made

88: clear.

89: 

90: In a recent Letter \cite{01GLP} we proposed that the crucial issue is

91: in the {\em production} of energy of hydrodynamic fluctuations by

92: their interaction with the mean flow. For the sake of concreteness we

93: examined a Poiseuille laminar flow and its loss of linear stability,

94: and showed how small viscosity contrasts lead to an order of magnitude

95: retardation in the onset of instability of ``dangerous"

96: disturbances. Specifically, we considered a flow in a channel of

97: dimensionless width 2, in which there are two fluids: one fluid of

98: viscosity $\mu_1$ flows near the walls, and the other fluid of

99: viscosity $\mu_2$ flows at the center, see Fig. \ref{f:scheme}. The

100: viscosities differ slightly, for example we considered (in

101: dimensionless units) $\mu_2=1$ and $m=\mu_1/\mu_2 =0.9$. The main

102: ingredient of the calculation was that all the viscosity difference of

103: 0.1 concentrated in a ``mixed" layer of width 0.10. The motivation

104: behind these numbers was the observation that the inferred effective

105: viscosity in polymer drag reduction increases towards the center by

106: about 30\% over about a 1/3 of the half-channel \cite{97SBH}. With our

107: choice we have comparable viscosity gradients in the mixed layer.

108: 

109: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

110: \begin{figure} 

111: \epsfxsize=7.5cm \epsfbox{Ramaj1.eps}

112: \caption{Schematic of the flow: the fluid near the walls has a viscosity

113: $\mu_1$, and that flowing at the center is of viscosity $\mu_2$. In

114: the mixed layer (of width $q$) the viscosity varies gradually between

115: $\mu_1$ and $\mu_2$. The parameter $p$ controls the position of the

116: mixed layer. For simplicity we neglect the down-stream growth in $q$.}

117: \label{f:scheme} 

118: \end{figure} 

119: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

120: 

121: In this model everything was explicitly calculable. The main point of

122: our analysis (see Sect. \ref{primary} for further details) was that

123: there exists a position in the channel where the velocity of the mean

124: flow is the same as the velocity of the most dangerous primary

125: instability. Below we refer to the layer around this position as the

126: ``critical layer". If we placed the mixed layer in the vicinity of the

127: critical layer, we got a giant effect of stabilization.  Analyzing

128: this phenomenon, we demonstrated that nothing special happened to the

129: dissipation.  Rather, it was the energy intake from the mean flow to

130: the unstable mode that was dramatically reduced, giving rise to a

131: large effect for a small cause. In this paper we extend these

132: observations in two directions. In Sect. \ref{primary}, after

133: reviewing the results of the simple model, we extend the analysis of

134: the primary instability to a case in which the viscosity profile is

135: that inferred from direct numerical simulations of turbulent channel

136: flow of a dilute polymeric solution \cite{97SBH}. We will see that

137: very similar effects are found. In other words, one does not need to

138: put by hand the region of viscosity variation in the vicinity of the

139: critical layer. When we have a continuous variation of the viscosity

140: in the region near the wall, the effect is the same, since it is only

141: crucial that there will be {\em some} space dependence of the

142: viscosity in the critical layer, which is usually not too far from the

143: wall.

144: 

145: A possible criticism of our results can be that the primary

146: instability is still too far from typical turbulent fluctuations. This

147: is in particular true since the most unstable primary modes are

148: 2-dimensional, whereas typical turbulent fluctuations are

149: 3-dimensional. For these reasons we present in Sect. \ref{secondary}

150: the analysis of the effect of small viscosity variations on the

151: secondary instability, for which the most ``dangerous" modes are

152: 3-dimensional. The tactics are similar to those taken for the primary

153: instability. First we discuss the effects of a mixed layer put at the

154: ``right" place in the channel, and second we show that continuous

155: viscosity profiles do exactly the same. We find again the giant effect

156: of stabilization for relatively small viscosity variations, lending

157: further support to our proposition that similar effects may very well

158: play a crucial role in turbulent drag reduction.  In Sect.

159: \ref{conclude} we present concluding remarks and suggestions for the

160: road ahead.

161: 

162: \section{Primary instability of Poiseuille flow}

163: \label{primary}

164: 

165: It is well known that parallel Poiseuille flow loses linear stability

166: at some threshold Reynolds number Re=Re$_{\rm th}$ (close to 5772).

167: It is also well known that the instability is ``convective", with the

168: most unstable mode having a phase velocity $c_p$. Analytically it has

169: the form \be \hat \phi_p(x,y,t) = \frac{1}{2}\big\{\phi_p(y) \exp\left

170: [\i \k_p(x-c_p \, t) \right ] + \mbox{c.c.}\big\} \exp(\gamma_p t) \ ,

171: \label{eq:inst} 

172: \ee where the subscript $p$ stands for the primary instability, $\hat

173: \phi(x,y,t)$ is the disturbance streamfunction and $\phi(y)$ is the

174: complex envelope of $\hat \phi(x,y,t)$. We have chosen $x$ and $y$ as

175: the streamwise and wall-normal coordinates respectively, $\k$ as the

176: streamwise wavenumber of the disturbance and $t$ as time. $\gamma_p$

177: is the growth rate of the primary instability.  What is not usually

178: emphasized is that the main interactions leading to the loss of

179: stability occur in a sharply defined region in the channel, i.e. at

180: the critical layer whose distance from the wall is such that the phase

181: velocity $c$ is identical to the velocity of the mean flow somewhere

182: within this layer. It is thus worthwhile to examine the effect on the

183: stability of Poiseuille flow of a viscosity gradient placed in the

184: vicinity of the critical layer. This will provide us with a very sharp

185: understanding of the mechanism of the stabilization of the flow by

186: viscosity variations. In the following subsection we will examine the

187: case of continuous viscosity profiles.

188: 

189: \subsection{Mixed Layer}

190: \label{primix}

191: 

192: A report of the results of this subsection was provided in

193: \cite{01GLP}. We examine a channel flow of two fluids with different

194: viscosities $\mu_1$ and $\mu_2$, see Fig. \ref{f:scheme}.

195: 

196: The observation that we want to focus on is shown in Fig. \ref{f:rcr}:

197: the threshold Reynolds number for the loss of stability of the mode as

198: in Eq.  (\ref{eq:inst}) depends crucially on the position of the mixed

199: layer. When the latter hits the critical layer, the threshold Reynolds

200: number for the loss of stability reaches as much as 88000. In other

201: words, one can increase the threshold of instability {\em for a given

202: mode} 15 times, and by making the mixed layer thinner one can reach

203: even higher threshold Reynolds values.

204: %%%%%%%%%%%%%%%%%%%%%%%

205: \begin{figure}

206: \epsfxsize=8.3

207: cm

208: \epsfbox{Ramarcr.eps}

209: \caption{The dependence of the threshold Reynolds

210: number Re$_{\rm th}$ on the position of the viscosity stratified layer

211: for $m=0.9$. The dashed line pertains to the neat fluid. Note the huge

212: increase in $\Re_{\rm th}$ within a small range. This occurs when the

213: stratified layer overlaps the critical layer.}

214: \label{f:rcr}

215: \end{figure}

216: %%%%%%%%%%%%%%%%%%%%%%%

217: In \cite{01GLP} we analyzed the physical origin of this huge

218: sensitivity of the flow stability to the profile of the viscosity.

219: 

220: The stability of this flow is governed by the modified Orr-Sommerfeld

221: equation \cite{white}

222: \begin{eqnarray}

223: \nonumber

224: && \i\k_p

225: \left[\left(\phi_p''-\k_p^2\phi_p\right)(\bar U-c_p-\i

226: \gamma_p) - \bar

227: U''\phi_p \right] \nonumber\\&&= {1\over {\rm Re}}\bigg[\mu

228: \phi_p^{\rm (4)}

229: + 2

230: \mu' \phi_p''' +\left(\mu'' - 2 \k_p^2 \mu \right)\phi_p''

231: \nonumber\\&&-

232: 2

233: \k_p^2 \mu' \phi_p' +

234: \left(\k_p^2 \mu'' + \k_p^4 \mu\right)

235: \phi_p\bigg]\,,

236: \label{modOS} 

237: \end{eqnarray} 

238: in which $\bar U(y)$ is the basic laminar velocity, and $\mu$ is a

239: function of $y$. The boundary conditions are $\phi_p(\pm 1) = \phi_p'

240: (\pm 1) = 0$. All quantities have been non-dimensionalised using the

241: half-width $H$ of the channel and the centerline velocity $U_0$ as the

242: length and velocity scales respectively.  The Reynolds number is

243: defined as Re$\equiv \rho U_0 H /\mu_2$, where $\rho$ is the density

244: (equal for the two fluids). The primes stand for derivative with

245: respect to $y$.  At $y=0$, we use the even symmetry conditions

246: $\phi(0)=1$ and $\phi'(0)=0$, as the even mode is always more unstable

247: than the odd.

248: 

249: Since the flow is symmetric with respect to the channel centreline, we

250: restrict our attention to the upper half-channel. Fluid 2 occupies the

251: region $0 \le y \le p$. Fluid 1 lies between $p+q \le y \le 1$. The

252: region $p \le y \le p+q$ contains mixed fluid. The viscosity is

253: described by a steady function of $y$, scaled by the inner fluid

254: viscosity $\mu_2$:

255: \begin{eqnarray}\label{muin}

256: \hskip -0.4cm \mu(y) &=& 1

257: \,,\quad  \mbox{for} \quad 0 \le y \le p\,,\\

258: \label{vis5}

259: \hskip -0.4cm \mu(y) &=& 1 +

260: (m-1)\, \xi^3\left[10 - 15\, \xi + 6 \xi^2

261: \right], \ 0 \le \xi \le 1\,,

262: \\

263: \label{muout}

264: \hskip -0.4cm \mu(y) &=& m \,,\quad  \mbox{for} \quad p+q \le y \le

265: 1.

266: \end{eqnarray}

267: Here $\xi\equiv (y-p)/q$ is the mixed layer coordinate.  We have

268: assumed a 5th-order polynomial profile for the viscosity in the mixed

269: layer, whose coefficients maintain the viscosity and its first two

270: derivatives continuous across the mixed layer. The exact form of the

271: profile is unimportant. For a plot of the profile $m=0.9$, see

272: Fig.~\ref{f:thinprof}.

273: 

274: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

275: \begin{figure}

276: \epsfxsize=7.5cm  \epsfbox{RamaU.eps}

277: \caption{Profiles of the normalized

278: viscosity $\mu(y)$ and normalized velocity $\bar U(y)$ and the second

279: derivative $\bar U^{''}(y)$ for $m=0.9$ (solid lines) and $m=1.0$

280: (dashed lines). The mixed layer is between the vertical dashed lines.

281: }

282: \label{f:thinprof} 

283: \end{figure} 

284: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

285: 

286: The basic flow $\bar U(y)$ is obtained by requiring the velocity and

287: all relevant derivatives to be continuous at the edges of the mixed

288: layer:

289: \begin{eqnarray}\label{ui}

290: \hskip -0.5cm &&   \bar U(y) = 1-Gy^2/2\,, \quad

291: \mbox{for} \quad y \le p\,,\\

292: \hskip -0.5cm &&\bar U(y) = U(p) - G\int_p^ydy~ y/ \mu

293: \,,\quad  \mbox{for}

294: \quad p \le y \le p+ q,

295: \label{um}\\

296: \hskip -0.5cm && \bar U(y) =

297: G\left(1-y^2\right)/2m, \quad  \mbox{for}

298: \quad y \ge p+q\

299: .\label{uo}

300: \end{eqnarray}

301: Here $G$ is the streamwise pressure gradient.

302: 

303: It can be seen, comparing the mean profile $\bar U(y)$ to that of the

304: neat fluid (cf. Fig. \ref{f:thinprof}), that nothing dramatic happens

305: to this profile even when the mixed layer is chosen to overlap a

306: typical critical layer. Accordingly we need to look for the origin of

307: the large effect of Fig. 2 in the energetics of the disturbances. To

308: do so, recall that the streamwise and normal components of the

309: disturbance velocity $\hat u_p(x,y,t)$ and $\hat v_p(x,y,t)$ may be

310: expressed via streamfunction as usual: $\hat u_p(x,y,t) = \partial\hat

311: \phi_p/\partial y$, ${\rm and} \quad \hat v_p(x,y,t) = -\partial \hat

312: \phi_p/\partial x$.  These functions may be written in terms of

313: complex envelopes similar to Eq.~(\ref{eq:inst}):

314: \begin{eqnarray} 

315:   \label{eq:vel-env}

316:   \hat u_p(x,y,t) &\!\!\!=\!\!\!& \frac{1}{2}\big\{u_p(y)

317:   \exp\left[\i\k_p(x\!-\!c_p\, t)\right]\!+\!  \mbox{c.c.}\big\}

318:   \exp(\gamma_p t),\\ \nonumber \hat v_p(x,y,t) &\!\!=\!\!&

319:   \frac{1}{2}\big\{v_p(y) \exp\left[\i\k_p(x-c_p\, t)\right] +

320:   \mbox{c.c.}\big\} \exp(\gamma_p t) \ .

321: \end{eqnarray} 

322: The pressure

323: disturbance $\hat p_p$ is defined similarly.

324: 

325: Define now a disturbance of the density of the kinetic ene rgy of the

326: primary instability 

327: \be\label{kinen} \hat E_p(x,y,t) =

328: \frac{1}{2}\left[ \hat u_p(x,y,t)^2 + \hat v_p(x,y,t)^2\right] \ .

329: \ee We can express the mean (over $x$) density of the kinetic energy

330: as follows:

331: \begin{eqnarray}\label{averen}

332:   E_p(y,t)&\equiv& \left< \hat

333: E_p(x,y,t)\right>_x =\C.E_p (y)\exp\,(2

334: \gamma_p t)\,,

335: \\ \nonumber

336: \C.E_p

337: (y)&=&\frac{1}{4}\left( |u_p(y)|^2 + |v_p(y)|^2 \right)\ .

338: \end{eqnarray}

339: 

340: The physics of our phenomenon will be discussed in terms of the

341: balance equation for the averaged disturbance kinetic energy. Starting

342: from the linearized Navier-Stokes equations for $\hat u_p$ and $\hat

343: v_p$, dotting it with the disturbance velocity vector, averaging over

344: one cycle in $x$ and using Eqs. (\ref{eq:vel-env})-(\ref{averen})

345: leads to \be 2 \gamma_p \,\C.E_p (y) = \nabla \cdot J_p(y) + W_{p+}(y)

346: - W_{p-}(y) \,,

347: \label{enbal} 

348: \ee where the energy flux $J_p(y)$ in the $y$ direction, rates of

349: energy production (energy taken up by the primary instability from the

350: mean flow) $W_{p+}(y)$ and energy dissipation (by the viscosity)

351: $W_{p-}(y)$ are given by

352: \begin{eqnarray}\label{current}

353:   J_p(y) &\equiv&

354: {\left[u_p(y) p_p^*(y)+\mbox{c.c.}\right]

355: \over 4 \rho} + {1 \over \R} \mu

356: (y) \nabla \C.E_p(y)\,,\\

357: W_{p+}(y) &\equiv&  - {1 \over 4} \bar

358: U'(y)\left[u_p(y)

359:  v_p^*(y) +\mbox{c.c.}  \right]\,,

360: \label{prod}

361: \\

362: W_{p-}(y) &\equiv& {\mu(y) \over \R} \left\{2 \k_p^2 \C.E_p(y)

363: + {1 \over

364: 2}\left[|u_p'(y)|^2+|v_p'(y)|^2\right]\right\}\

365: .

366: \nonumber

367: \end{eqnarray}

368: The superscript $*$ denotes complex conjugate. To plot these functions

369: we need to solve Eq. (\ref{modOS}) as an eigenvalue problem, to obtain

370: $c_p$, $\gamma_p$, and $\phi_p(y)$ at given $\R$ and $\k_p$.  The

371: value of $c_p$ determines the position of the critical layer. It is

372: convenient to compute and compare the space averaged production and

373: dissipation terms $\Gamma_{p+}$ and $\Gamma_{p-}$ defined by:

374: %%%%%%%%%%%%%%%%%%

375: \be\label{totpd} \Gamma_{p\pm} \equiv\int_0^1 W_{p\pm}(y) \d y \ \Big/

376: \int_0^1 \C.E_p(y) \d y \ .  \ee The local production of energy can be

377: positive or negative, indicative of energy transfer from the mean flow

378: to the primary disturbance and vice-versa respectively. The production

379: in one region (where $ W_{p+}(y)>0$) can be partly canceled out by a

380: ``counter-production'' in other region (where $ W_{p+}(y)<0$).

381: 

382: The use of these measures can be exemplified with the neat fluid

383: ($m=1.0$ here). The laminar flow displays its first linear instability

384: at a threshold Reynolds number of $\R_{\rm th}=5772$, which means that

385: the total production $\Gamma_{p+}$ across the layer becomes equal to

386: the total dissipation $\Gamma_{p-}$ at this value of $\R$. Examining

387: Fig.  \ref{f:far} we can see that the disturbance kinetic energy is

388: produced predominantly within the critical layer, where the basic flow

389: velocity is close to the phase speed of the disturbance, while most of

390: the dissipation is in the wall layer.

391: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

392: \begin{figure} 

393: \epsfxsize=7.5cm

394: \epsfbox{Ramas0.eps}

395: \epsfxsize=7.5cm \epsfbox{Ramafar.eps}

396: \caption{Energy

397: balance: production $W_{p+}(y)$, solid line; dissipation

398: $W_{p-}(y)$,

399: dot-dashed line, $\R=5772$. Top: $m=1$, $\Gamma_{p+}=

400: \Gamma_{p-}=0.0148$.

401: Bottom: $m=0.9$, $p=0.3$, $\Gamma_{p+}=0.0158$,

402: $\Gamma_{p-}=0.0148$. In

403: this and the two subsequent figures the solid

404: vertical lines show the

405: location $y_c$ of the critical lines, whereas the

406: region

407: between the dotted

408: lines is the mixed layer.}

409: \label{f:far} 

410: \end{figure}

411: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

412: The balance is not changed significantly when the viscosity ratio is

413: changed to $0.9$ so long as the mixed layer is not close to the

414: critical layer.  There is a small region of production and one of

415: counter-production within the mixed layer, whose effects cancel out,

416: leaving the system close to marginal stability.

417: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

418: \begin{figure}

419: \epsfxsize=7.5cm \epsfbox{Ramanear.eps}

420: \caption{Energy balance: production

421: $W_{p+}(y)$, solid line; dissipation

422: $W_{p-}(y)$, dot-dashed line.

423: $\R=5772$, $m=0.9$, $p=0.85$, $\Gamma_{p+}=

424: -0.0114$, $\Gamma_{p-}=0.0122$.

425: }

426:  \label{f:near} 

427: \end{figure} 

428: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

429: 

430: We now turn our attention to Fig. \ref{f:near}, in which our main

431: point is demonstrated.  The Reynolds number is the same as before, but

432: the mixed layer has been moved close to the critical layer. It is

433: immediately obvious that the earlier balance is destroyed. The

434: counter-production peak in the mixed layer is much larger than before,

435: making the flow more stable.  The wavenumber used is that at which the

436: flow is least stable for the given Reynolds number at this $p$. For

437: $m=0.9$, the threshold Reynolds number is $46400$.  Fig.  \ref{f:marg}

438: shows the energy balances at marginal stability - the picture is

439: qualitatively the same here as at $\R\approx 5772$ for the neat fluid.

440: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

441: \begin{figure}

442: \epsfxsize=7.5cm

443: \epsfbox{Ramamarg.eps}

444: \caption{Energy balance: production $W_{p+}(y)$,

445: solid line; dissipation

446: $W_{p-}(y)$, dot-dashed line. $\R=46400$, $m=0.9$,

447: $p=0.85$, $\Gamma_{p+}=

448: \Gamma_{p-}=0.0053$. }

449:  \label{f:marg} 

450: \end{figure}

451: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

452: \subsection{The mechanism of

453: stabilization}

454:  

455: The main factor determining the instability is the energy intake from

456: the mean flow, which is driven by the phase change caused by the

457: viscosity stratification. The dissipation on the other hand depends

458: only on Reynolds number and does not respond disproportionately to

459: changes in viscosity. In neat fluids, the term containing $\bar

460: U''(y)$ in (\ref{modOS}) is always of higher order within the critical

461: layer. However, with the introduction of a viscosity gradient within

462: the critical layer, the gradients of the basic velocity profile will

463: scale according to the mixed layer coordinate $\xi$. We show in the

464: analysis that follows that for $q \le O(\R^{-1/3})$, the term

465: containing $\bar U''$ is now among the most dominant. Since most of

466: the production of disturbance kinetic energy takes place within the

467: critical layer, we return to equation (\ref{modOS}) and isolate all

468: lowest-order effects within the critical layer. The relevant normal

469: coordinate in the critical layer is \be \eta \equiv {y-y_c \over \ep}

470: \label{eta}

471: \ee

472: where $y_c$ is the critical point defined by $U(y_c)=c$, and $\ep$ is the

473: critical layer thickness, which is a small parameter at large Reynolds

474: number. The basic channel flow velocity may be expanded in the vicinity of

475: the critical point as

476: \be

477: U(y) = c + (y-y_c) U'(y_c) + {(y-y_c)^2 \over 2!} U''(y_c) + \cdots.

478: \label{ucrit}

479: \ee

480: We use (\ref{ucrit}), and redefine $\phi_p(y)\equiv\Phi(\eta)$ and

481: $\mu(y)\equiv \nu(\xi)$, to rewrite

482: (\ref{modOS}) within the critical layer. We obtain

483: \be

484: \ep \sim \R^{-1/3} \equiv \left(\k_p\R\right)^{-1/3},

485: \label{epsilon}

486: \ee and the lowest order equation in the critical layer: \be \i \eta

487: {\d U \over \d y}\bigg|_c\Phi'' - {\i G p \over \nu^2} \chi\nu'\Phi =

488: \nu \Phi^{\rm (4)} + 2 \chi\nu' \Phi''' + \chi^2\nu''\Phi''\,,

489: \label{critlow}

490: \ee where $\chi \equiv \ep /q$ is $O(1)$ for the mixed layer. In the

491: absence of a viscosity gradient in the critical layer (i.e. $\nu=1$),

492: equation (\ref{critlow}) would reduce to \be \i \eta {\d U \over \d

493: y}\bigg|_c\Phi'' = \Phi^{\rm (4)}\,,

494: \label{lin}

495: \ee which is the traditional lowest-order critical layer equation for

496: a parallel shear flow \cite{45Lin}. The mechanism for the

497: stabilization now begins to be apparent: there are several new terms

498: which can upset the traditional balance between inertial and viscous

499: forces. In order to narrow down the search further, we resort to

500: numerical experimentation, because although all terms in

501: (\ref{critlow}) are estimated to be of $O(1)$, their numerical

502: contributions are different.  It transpires that the second term on

503: the left hand side of (\ref{critlow}) is particularly responsible: it

504: is straightforward to verify that it originates from the term

505: containing $\bar U''(y)$ in the modified Orr-Sommerfeld equation. As

506: testimony, note the dramatic effect on $\bar U''$ in Fig. 3. Any

507: reasonable viscosity gradient of the right sign will pick up this

508: term, leading to vastly enhanced stability.

509: 

510: Indeed, in the light of this discussion we can expect that the large

511: effect of retardation of the instability would even increase if we

512: make the mixed layer thinner. This is indeed so. Nevertheless, one

513: cannot conclude that instability can be retarded at will, since other

514: disturbances, differing from the primary mode, become unstable first,

515: albeit at a much higher Reynolds number than the primary mode; when we

516: stabilize a given mode substantially, we should watch out for other

517: pre-existing/newly destabilized modes which may now be the least

518: stable.

519: 

520: Finally, we connect our findings to the phenomenon of drag reduction

521: in turbulent flows. Since the total dissipation can be computed just

522: from the knowledge of the velocity profile at the walls, any amount of

523: drag reduction must be reflected by a corresponding reduction of the

524: gradient at the walls.  Concurrently, the energy intake by the

525: fluctuations from the mean flow should reduce as well. Indeed, the

526: latter effect was measured in both experiments \cite{97THKN} and

527: simulations \cite{01DSBH,00Ang}. The question is which is the chicken

528: and which is the egg. In our calculation we identified that the

529: reduction in production comes first. From Figs. 4 and 5 which are at

530: the same value of $\R$ we see that the dissipation does not change at

531: all when the mixed layer moves, but the production is strongly

532: affected. Of course, at steady state the velocity gradient at the wall

533: must adjust as shown in Fig. 6.

534: 

535: \subsection{Continuous Viscosity

536: Profile}

537: \label{contprim}

538: 

539: One could think that the strong stabilization discussed in the

540: previous subsection is only due to the precise positioning of the

541: mixed layer at the critical layer. If so, the result would have very

542: little generic consequence.  In this subsection we show that any

543: reasonable viscosity profile achieves the same effects. To this aim we

544: consider the effective viscosity profile reported in \cite{97SBH} (in

545: their Fig. 5) which is obtained from simulations of a turbulent

546: channel flow with polymer additive. It may be prescribed as

547: \begin{eqnarray}\label{cmuin}

548: \mu(y) &=& 1

549: \,,\quad  \mbox{for} \quad 0 \le y \le p\,,\\

550: \label{vis3}

551: \mu(y) &=& 1 +

552: (m-1)\, \left({y-p \over q}\right)^3,

553: \end{eqnarray}

554: with $q \sim 0.4$, and

555: $m \sim 0.7$, as shown in Fig.

556: \ref{f:contprof}.

557: \begin{figure}

558: \epsfxsize=7.5cm

559: \epsfbox{Ramacontprof.eps}

560: \caption{Prescribed continuous viscosity profile

561: (in accordance with that

562: obtained in direct numerical simulations of

563: polymeric flow). The

564: corresponding laminar velocity profile $\bar U(y)$ and

565: its second derivative

566: are also shown.}

567: \label{f:contprof}

568: \end{figure}

569: The energy balance for the least stable primary mode at $\R=5772$ for

570: this case (Fig. \ref{f:polyprim}) shows a large counter-production of

571: disturbance kinetic energy, which is in fact more pronounced than what

572: we obtained with a mixed layer (Fig. \ref{f:near}). Thus the strong

573: stabilization effect does not require careful placing of the viscosity

574: variation at a particular layer.  It is sufficient that there exist a

575: viscosity variation in the region of the critical layer (indicated as

576: the vertical line in Fig. \ref{f:polyprim}) to achieve the

577: stabilization.

578: 

579: It comes as no surprise that this continuous viscosity profile behaves

580: very similarly to the thin mixed-layer. If we return to equation

581: (\ref{critlow}), we will see that all we have now done is to increase

582: both $\nu'$ (which is proportional to $m-1$) and $q$ threefold (the

583: effective $q$ here is closer to 0.3 than 0.4, as we can see from

584: Fig. \ref{f:contprof}), so the ratio remains the same.

585: 

586: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

587: \begin{figure} 

588: \epsfxsize=7.5cm \epsfbox{Ramapolyprim.eps}

589: \caption{Energy balance: production $W_{p+}(y)$, solid line; dissipation

590: $W_{p-}(y)$, dot-dashed line. $\R=5772$, $m=0.7$, $p=0.6$, $q=0.4$,

591: $\Gamma_{p+}= -0.0345$, $\Gamma_{p-}=0.0138$. }

592:  \label{f:polyprim}

593: \end{figure} 

594: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

595: \section{Secondary Instabilities}

596: \label{secondary}  

597: 

598: A laminar flow through a channel is linearly unstable at $\R=5772$. In

599: all except the cleanest experiments, however, the flow becomes

600: turbulent at much lower Reynolds numbers, as low as $1000$

601: \cite{80OK,93TTRD}. This is because the linear stability analysis is

602: carried out on a steady laminar velocity profile, whereas a real flow,

603: except under carefully designed clean conditions, consists in addition

604: of small but finite disturbances (most of whom will decay at long

605: times). The stability behaviour of the real flow is quite different

606: from that of the steady profile: the actual flow is unstable to new

607: modes, often referred to as secondary modes. The secondary modes are

608: often three dimensional, and their signature is prominent in

609: fully-developed turbulence. As described below, the secondary

610: instabilities are studied by a Floquet analysis of the periodic

611: primary flow we obtained earlier.

612: 

613: As is usual in the analysis of secondary instabilities

614: \cite{88Herb,83Herb}, we begin by splitting the flow into a periodic

615: component (consisting of the mean laminar profile in addition to the

616: primary wave) and a secondary disturbance, e.g., \be \B.U_{\rm

617: total}(x,y,z,t) = \B.U(x,y,t) + \B.u_s(x,y,z,t), \ee where

618: \begin{eqnarray}

619: &&\!\!\B.U(x,y,t) = \bar U(y) \hat \B.x \label{basic}\\&&\!\!+ A_p(t)

620: \left\{\left[u_p(y) \hat \B.x  + v_p(y) \hat \B.y

621: \right] \exp \left[\i k_p (x - c_p t)\right] + {\rm c.c.} \right\}.

622: \nonumber

623: \end{eqnarray}

624: Here $\hat \B.x$ and $\hat \B.y$ are units vectors in the $x$

625: (steamwise) and $y$ (wall normal) directions.  The amplitude $A_p$ of

626: the primary disturbance changes very slowly with time, and $\d A_p /\d

627: t$ may be neglected during one time period. The spatial and temporal

628: dependence of the secondary disturbance is written in the form

629: \begin{eqnarray}

630: &&\B.u_s(y,\B.r_\perp,t)\equiv {\cal R}\!e\Big\{\B.u_{s+}(y) \exp\left[\i

631: \left(\B.k_+ \cdot \B.r_\perp - \omega_+ t\right)\right] \nonumber\\&&+

632: \B.u_{s-}(y) \exp\left[\i \left(\B.k_-\cdot \B.r_\perp - \omega_- t\right)

633: \right]\Big\},

634: \label{form}

635: \end{eqnarray}

636: where $\B.r_\perp\equiv x\hat \B.x +z\hat \B.z $, and $\B.k_\pm= k_\pm

637: \hat\B.x \pm k_z \hat\B.z$.  We substitute the above ansatz into the

638: Navier-Stokes and continuity equations, and retain linear terms in the

639: secondary. On averaging over $x$, $z$ and $t$, only the resonant modes

640: survive, which are related by \be \B.k_+ + \B.k_- = k_p \hat \B.x \ ,

641: \quad \text{therefore}~~\B.k_\pm=\pm \B.q+\frac{k_p}{2} \hat \B.x \ ,

642: \ee for any vector $\B.q$, and \be \omega_+ = \omega + \i\gamma_s

643: \qquad {\rm and} \qquad \omega_- = (\omega_p - \omega) + \i\gamma_s.

644: \label{om}

645: \ee Eliminating the disturbance pressure and streamwise component of

646: the velocity, we get the equations for the secondary disturbances

647: $v_s$ and $w_s$.  Using the operator $D$ for differentiation with

648: respect to the normal coordinate $y$, and the notation $f_\pm \equiv -

649: \i w_{s\pm}/k_z$, the equations read \cite{02SG}

650: \begin{eqnarray}

651: &&\A \nonumber\\&&

652: \times\left[\s f_+ - Dv_+\right] - \i k_+ U' v_+ \label{first}\\&&-

653: {A_p k_+ \over 2 k_-}\Bigg\{\left[\i k_+ u_p D + v_p D^2 +\i k_- Du_p

654: \right]v_-^*

655: \nonumber\\&&+ \left[\left(\beta^2-k_-k_+\right)v_pD + \i k_+

656: \left(k_-^2+\beta^2\right)u_p\right]f_-^*\Bigg\} = 0,

657: \nonumber

658: \end{eqnarray}

659: and

660: \begin{eqnarray}

661: &&\A 

662: \left(Df_+ - v_+\right) \nonumber\\&&+ \bigg[-\i k_+ (DU) + (D^2\nu) D +

663: (D\nu)(D^2 - k_+^2 - \beta^2)\bigg] f_+ \nonumber\\&&+ {A_p (k_p + k_-)

664: \over 2} \bigg[ \i u_p \left(v_-^* + Df_-^*\right)

665: -{v_p \over k_-}Dv_-^* \bigg] \nonumber\\&&+

666: {A_p \over 2}\left[v_p\left({k_p\beta^2 \over k_-} + D^2\right) -

667: \i k_-(Du_p) \right]f_-^* = 0,

668: \label{second}

669: \end{eqnarray}

670: The boundary conditions are

671: \be

672: \B.u_s=0 \qquad {\rm at} \qquad y = \pm 1.

673: \label{bcs}

674: \ee Equations (\ref{first}) and (\ref{second}), along with two

675: corresponding equations in $v_-^*$ and $f_-^*$, describe an eigenvalue

676: problem for the secondary instability. The four equations are solved

677: by a Chebychev collocation spectral method, details of the solution

678: procedure are available in \cite{02SG}.

679: 

680: The most unstable secondary mode in our case is found to be the

681: subharmonic, for which $\B.q=\beta \hat \B.z$.  The production and

682: dissipation are computed as before.

683: 

684: We survey in turn the thin mixed-layer profile, and the continuous

685: viscosity profile to see what viscosity variation does to the

686: secondary instability.

687: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

688: \subsection{Mixed Layer}

689: \label{thin-sec}

690: 

691: The velocity and viscosity profiles here are as given in Fig.

692: \ref{f:thinprof}, and the primary instability is that presented in

693: Sect.  \ref{primix}. Since the subharmonic ($k_+=k_-= k_p/2$) is the

694: least stable mode, we present this case alone. In Fig. \ref{f:thinmax}

695: a typical dependence of the growth rate of the secondary mode on the

696: spanwise wavenumber is shown. We can see that the viscosity variation

697: damps the secondary mode significantly, but it is still

698: unstable. However, there is a crucial difference in the {\em primary}

699: instabilities of the two: the primary is unstable for a constant

700: viscosity flow, but very stable in the mixed layer case. Therefore at

701: long times, the secondary mode, which feeds on the primary for its

702: existence, dies down in the latter case.  To compute the time

703: dependence of the amplitude of the secondary mode we computed the

704: growth rate $\gamma_s$ by neglecting the time dependence of the

705: amplitude of the primary words.  As a result we obtain the growth rate

706: $\gamma_s[A_p(t)]$, in which $A_p(t)$ can be an exponentially growing

707: or a decaying function of time. Having this growth rate we can present

708: the time dependence of the amplitude of the secondary mode, see Fig.

709: \ref{f:thinamp}. Without the viscosity contrast, the amplitude of the

710: secondary mode increases (essentially exponentially).  With the

711: viscosity contrast the amplitude decays in time.

712: 

713: We now observe the balances of energy initially and at a later time in

714: Figs.  \ref{f:thinprod0} and \ref{f:thinprod40} respectively. The

715: initial balance of energy is not very different from the constant

716: viscosity case.  At the later time, however, the production of

717: secondary kinetic energy is significantly lower. The location $y_c$ of

718: the critical point is seen from the figures to be close to the layer

719: of stratified viscosity. If the two were well-separated, the

720: stratification would do nothing to the secondary mode.

721: 

722: \begin{figure}

723: \epsfxsize=7.5cm \epsfbox{Ramathinmax.eps}

724: \caption{Dependence of growth rate on spanwise wavenumber. Solid line:

725: varying viscosity ($p=0.8, q=0.1, m=0.9$); dashed line: constant

726: viscosity ($m=1$). $k_p=1, A_p=0.005, \R=6000$. }

727: \label{f:thinmax}

728: \end{figure}

729: 

730: \begin{figure}

731: \epsfxsize=7.5cm \epsfbox{Ramathinamp.eps}

732: \caption{Amplitude of the secondary mode in logarithmic 

733: scale as a function of time. Dashed line: constant viscosity,

734: $m=1$. Here $\gamma_p=0.0003$, and the primary mode is unstable. Solid

735: line: varying viscosity; here $\gamma_p=-0.0206$, the primary mode is

736: stable. All conditions like in Fig. \ref{f:thinmax}, in particular

737: $A_p(t=0)=0.005$.}

738: \label{f:thinamp}

739: \end{figure}

740: 

741: \begin{figure}

742: \epsfxsize=7.5cm \epsfbox{Ramathinpr0.eps}

743: \caption{Production $W_{s+}$ and dissipation $W_{s-}$ of the 

744: kinetic energy

745: of the secondary disturbance at time=0. Solid line: $W_{s+}, m=0.9$;

746: dot-dashed line: $W_{s-}, m=0.9$; long dashes: $W_{s+}, m=1$; dotted

747: line: $W_{s-}, m=1$.  The vertical lines show $y_c$ (the critical

748: point location) for $m=0.9$ (solid) and $m=1$ (dotted).}

749: \label{f:thinprod0}

750: \end{figure}

751: 

752: \begin{figure}

753: \epsfxsize=7.5cm \epsfbox{Ramathinpr40.eps}

754: \caption{Production $W_{s+}$ and dissipation $W_{s-}$ of 

755: the kinetic energy of the secondary disturbance at time=40. Solid

756: line: $W_{s+}, m=0.9$; dot-dashed line: $W_{s-}, m=0.9, A_p=0.00215$;

757: long dashes: $W_{s+}, m=1$; dotted line: $W_{s-}, m=1, A_p=0.00506$.}

758: \label{f:thinprod40}

759: \end{figure}

760: 

761: A lowest-order analysis of the secondary stability equations is not as

762: straightforward as for the primary mode, since the secondary is highly

763: dependent on the amplitude of the primary \cite{02SG}. We may however

764: make the following observations from a critical layer analysis of

765: equations (\ref{first}) and (\ref{second}) and their

766: counterparts. When $A_p\gg \ep$, (cf. Eq.(\ref{epsilon})) only the

767: nonlinear terms appear at the lowest order, and the secondary mode is

768: completely driven by the primary.  When $A_p \sim O(\ep)$, both the

769: basic terms and the nonlinear terms contribute at the lowest order. It

770: may be numerically determined, however, that the secondary is slaved

771: to the primary here as well. When $A_p=o(\ep)$, the lowest-order

772: theory for the secondary is (not surprisingly) exactly that given by

773: (\ref{critlow}) for the primary.

774: 

775: A direct estimate of the effect of the viscosity stratification on the

776: secondary mode is obtained from the threshold amplitude $A_{\rm th}$ of

777: the primary for the instability. At a Reynolds number of 6000 and

778: primary wavenumber of $k_p=1$, for a neat fluid, all secondary modes

779: are damped if $A_{\rm th}<0.002$, while for the continuous viscosity

780: profile, all secondary modes continue to be damped even for larger

781: primary disturbances, up to $A_{\rm th}=0.005$. When the Reynolds number

782: is reduced to 2000, the threshold amplitudes are 0.012 and 0.016 for

783: the neat and viscosity-stratified fluids respectively.

784: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

785: \subsection{Continuous Viscosity Profile}

786: \label{cont-sec}

787: 

788: The velocity and viscosity profiles here are as given in Fig.

789: \ref{f:contprof}, and the primary instability is that presented in Sect.

790: \ref{contprim}. The counterparts for the continuous viscosity

791: profile of Figs. \ref{f:thinmax} to \ref{f:thinprod40} are presented

792: in Figs. \ref{f:max} to \ref{f:prod40} respectively. It is clear that

793: nothing has changed qualitatively.

794: 

795: \begin{figure}

796: \epsfxsize=7.5cm \epsfbox{Ramamax.eps}

797: \caption{Dependence of growth rate on spanwise wavenumber. Solid line:

798: varying viscosity [according to equation (\ref{vis3})]; dashed line:

799: constant viscosity. Wavenumbers and $\R$ as in Fig. \ref{f:thinmax}.}

800: \label{f:max}

801: \end{figure}

802: 

803: \begin{figure}

804: \epsfxsize=7.5cm \epsfbox{Ramaamp.eps}

805: \caption{Variation of the amplitude of the secondary instability mode with

806: time. Solid line: varying viscosity, $\gamma_p=-0.0244$; dashed line: 

807: constant viscosity, $\gamma_p=0.0003$.

808: Wavenumbers and $\R$ as in Fig. \ref{f:thinamp}.}

809: \label{f:amp}

810: \end{figure}

811: 

812: \begin{figure}

813: \epsfxsize=7.5cm \epsfbox{Ramaprod0.eps}

814: \caption{Production and dissipation at time=0. Solid line: varying

815: viscosity;

816: dashed line: constant viscosity, $A_p=0.005$ for both.}

817: \label{f:prod0}

818: \end{figure}

819: 

820: \begin{figure}

821: \epsfxsize=7.5cm \epsfbox{Ramaprod40.eps}

822: \caption{Production and dissipation at time=40. Solid line: varying

823: viscosity, $A_p=0.00184$;

824: dashed line: constant viscosity, $A_p=0.00506$.}

825: \label{f:prod40}

826: \end{figure}

827: 

828: Fig. \ref{f:ampdep} shows the dependence of the growth rate of the 

829: secondary mode on the amplitude of the primary disturbance. It is

830: clear that the instability is reduced by the stratification of

831: viscosity, but there is no dramatic effect in the secondary alone. We

832: may conclude that the large effect comes from the complete reliance 

833: of the secondary on the primary.

834: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

835: \begin{figure}

836: \epsfxsize=7.5cm \epsfbox{Ramaampdep.eps}

837: \caption{Dependence of the growth rate of the secondary mode on the

838: amplitude of the primary disturbance. Solid line: stratified 

839: viscosity; dashed line: constant viscosity. $k_p=1, k_+=0.5, \beta=1,

840: \R=6000$.}

841: \label{f:ampdep}

842: \end{figure}

843: %%%%%%%%%%%%%%%%%%%%%%%%%

844: \section{Concluding Remarks}

845: \label{conclude}

846: 

847: We addressed the primary and secondary instability of simple channel

848: flows, and examined the effects of small viscosity variations. We find

849: dramatic effects of stabilization when the viscosity variations exist

850: in the vicinity of the critical layers, in which the speed of

851: propagation of the modes coincided with the mean velocity of the basic

852: flow. With about 10\% viscosity changes we can have very large

853: increases in the threshold Reynolds numbers for instability. In all

854: cases we find that the main mechanism for the large effects is the

855: reduction of the intake of energy from the mean flow to the putative

856: unstabale modes, which therefore become stable. For the same Reynolds

857: numbers in Newtonian fluids there is no such mechanism for

858: stabilization and these flows will become turbulent. We would like to

859: propose that similar effects should be examined in the case of

860: turbulent drag reduction by polymer additives.

861: 

862: 

863: We recognize that in a turbulent flow there are many more modes that

864: interact, but we propose that a similar mechanism operates for each

865: mode at its critical layer, where both elastic and viscous effects

866: determine the mean flow. The advantage of the present calculation is

867: that we can consider all the putative unstable modes, and conclude

868: that with a viscosity gradient similar to that seen in polymeric

869: turbulent flows the linear threshold $\R_{\rm th}$ goes up five times

870: (to 31000). We note in passing that this effect had not been put to an

871: experimental test, and it would be exciting to have a confirmation of

872: our predictions by future experiments. For actual turbulent flows we

873: will need first to identify what are the main modes that interact

874: between themselves and with the mean flow. A significant numerical

875: effort is required, but appears worthwhile due to the importance of

876: the phenomenon of drag reduction, and its relative lack of

877: understanding.

878: 

879: We have demonstrated that the exact form of the viscosity profile is

880: immaterial; a continuous profile of viscosity in the critical region

881: behaves exaclty like a thin mixed layer. We have shown that the secondary

882: three dimensional modes of instability are ``slaved'' to the primary

883: linear mode of instability: the mechanism which stabilizes the primary

884: mode indirectly ensures that the secondary is damped out quickly.

885: 

886: Finally we note that a linear disturbance can rear its head either in

887: the form of the fastest growing (or slowest decaying) mode as

888: considered here; or in non-modal form with a transient growth followed

889: by long-term decay \cite{02Chap}. The former situation will correspond

890: to relatively high Reynolds numbers, or cleaner set-ups. We expect

891: similar conclusions in the latter situation as well.

892: %%%%%%%%%%%%%%%%%

893: \acknowledgments This work was supported by the German-Israeli

894: Foundation, the European Commission under a TMR grant, the Israeli

895: Science Foundation and the Naftali and Anna Backenroth-Bronicki Fund

896: for Research in Chaos and Complexity.  RG thanks the Defence R\&D

897: Organisation, India, for financial support.

898: 

899: 

900: \begin{references} 

901: 

902: \bibitem{95NH}

903: R.H. Nadolink and W.W. Haigh,  ASME Appl. Mech. Rev. {\bf 48}, 351 (1995).

904: 

905: \bibitem{69Lum}

906: J. L. Lumley, Ann. Rev. Fluid Mech. {\bf 1}, 367 (1969).

907: 

908: \bibitem{00SW}

909: K.R. Sreenivasan and C. M. White, J. Fluid Mech., {\bf 409}, 149 (2000).

910: 

911: \bibitem{86TG}

912: M. Tabor and P.G. de Gennes, Europhys. Lett., {\bf 2}, 519 (1986).

913: 

914: \bibitem{90Gennes}

915: P.G. de Gennes, {\em Introduction to Polymer Dynamics}, (Cambridge, 1990).

916: 

917: \bibitem{01GLP}

918: R. Govindarajan, V.S. L'vov and I. Procaccia, Phys. Rev. Lett.

919: {\bf 87}, 174501 (2001)

920: 

921: \bibitem{97SBH}

922: R. Sureshkumar, A. N.  Beris and R.A. Handler, Phys. Fluids {\bf 9}, 743

923: (1997).

924: 

925: \bibitem{white}

926: F.M White, {\em Viscous Fluid Flow} (McGraw Hill, 1991).

927: 

928: \bibitem{Rama} 

929: B.T. Ranganathan and R. Govindarajan, Phys. Fluids {\bf 13}, 1 (2001).

930: 

931: \bibitem{97THKN}

932: J.M.J den Toonder, M.A. Hulsen,G.D.C Kuiken and F.T.M Niewstadt,

933: J. Fluid Mech. {\bf 337}, 193 (1997).

934: 

935: \bibitem{01DSBH}

936: C.D. Dimitropoulos, R. Sureshkumar, A.N. Beris and R.A. Handler,

937: Phys. Fluids, {\bf 13}, 1016 (2001).

938: 

939: \bibitem{00Ang}

940: E. De Angelis, ``The Influence of Polymer Additives on the Structure

941: of Wall Turbulence", Thesis, Universita' di Roma, La Sapienza, 2000.

942: 

943: \bibitem{45Lin}

944: C.C. Lin, Quarterly of Appl. Math., {\bf 3}, 117-142 (1945).

945: 

946: \bibitem{80OK}

947: S.A. Orszag and L.C. Kells, J. Fluid Mech., {\bf 96} 159 (1980).

948: 

949: \bibitem{93TTRD}

950: L.N. Trefethen, A.E.Trefethen, S.C. Reddy and T.A. Driscoll, Science, {\bf

951: 261} 578 (1993).

952: 

953: \bibitem{88Herb}

954: T. Herbert, Ann. Rev. Fluid Mech., {\bf 20}, 487 (1988).

955: 

956: \bibitem{83Herb}

957: T. Herbert, Phys. Fluids, {\bf 26} (4), 871 (1983).

958: 

959: \bibitem{02SG}

960: A. Sameen and R. Govindarajan, preprint, 2002.

961: 

962: \bibitem{02Chap}

963: See for example: S.J. Chapman,  J. Fluid Mech., {\bf 451}, 35 (2002).

964: 

965: \end{references}  

966: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

967: 

968: \end{document}

969: 

970: \bibitem{33Squi}

971: H.B. Squire, Proc Roy. Soc. Lond. A, {\bf 142} 621-628 (1933).

972: