1: \documentclass{jfm}

2: %\documentclass[referee]{jfm}

3: \usepackage{epsf}

4: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

5: % begin JFM caption

6: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

7: 

8: % See if the author has AMS Euler fonts installed: If they have, attempt

9: % to use the 'upmath' package to provide upright math.

10: 

11: \ifCUPmtlplainloaded \else

12:   \checkfont{eurm10}

13:   \iffontfound

14:     \IfFileExists{upmath.sty}

15:       {\typeout{Found AMS Euler Roman fonts on the system,

16:                    using the 'upmath' package.}%

17:        \usepackage{upmath}}

18:       {\typeout{Found AMS Euler Roman fonts on the system, but you

19:                    dont seem to have the}%

20:        \typeout{'upmath' package installed. JFM.cls can take advantage

21:                  of these fonts, if you use 'upmath' package.}%

22:        \providecommand\upi{\pi}%

23:       }

24:   \else

25:     \providecommand\upi{\pi}%

26:   \fi

27: \fi

28: 

29: % See if the author has AMS symbol fonts installed: If they have, attempt

30: % to use the 'amssymb' package to provide the AMS symbol characters.

31: 

32: \ifCUPmtlplainloaded \else

33:   \checkfont{msam10}

34:   \iffontfound

35:     \IfFileExists{amssymb.sty}

36:       {\typeout{Found AMS Symbol fonts on the system, using the

37:                 'amssymb' package.}%

38:        \usepackage{amssymb}%

39:        \let\le=\leqslant  \let\leq=\leqslant

40:        \let\ge=\geqslant  \let\geq=\geqslant

41:       }{}

42:   \fi

43: \fi

44: 

45: % See if the author has the AMS 'amsbsy' package installed: If they have,

46: % use it to provide better bold math support (with \boldsymbol).

47: 

48: \ifCUPmtlplainloaded \else

49:   \IfFileExists{amsbsy.sty}

50:     {\typeout{Found the 'amsbsy' package on the system, using it.}%

51:      \usepackage{amsbsy}}

52:     {\providecommand\boldsymbol[1]{\mbox{\boldmath $##1$}}}

53: \fi

54: 

55: 

56: %%% Example macros (some are not used in this sample file) %%%

57: 

58: % For units of measure

59: \newcommand\dynpercm{\nobreak\mbox{$\;$dynes\,cm$^{-1}$}}

60: \newcommand\cmpermin{\nobreak\mbox{$\;$cm\,min$^{-1}$}}

61: 

62: % Various bold symbols

63: \providecommand\bnabla{\boldsymbol{\nabla}}

64: \providecommand\bcdot{\boldsymbol{\cdot}}

65: \newcommand\biS{\boldsymbol{S}}

66: \newcommand\etb{\boldsymbol{\eta}}

67: 

68: % For multi-letter symbols

69: \newcommand\Real{\mbox{Re}} % cf plain TeX's \Re and Reynolds number

70: \newcommand\Imag{\mbox{Im}} % cf plain TeX's \Im

71: \newcommand\Rey{\mbox{\textit{Re}}}  % Reynolds number

72: \newcommand\Pran{\mbox{\textit{Pr}}} % Prandtl number, cf TeX's \Pr product

73: \newcommand\Pen{\mbox{\textit{Pe}}}  % Peclet number

74: \newcommand\Ai{\mbox{Ai}}            % Airy function

75: \newcommand\Bi{\mbox{Bi}}            % Airy function

76: \newcommand\Ei{\mbox{Ei}}            % Exponential integral

77: \newcommand\Ci{\mbox{Ci}}            % Cosine integral

78: \newcommand\Si{\mbox{Si}}            % Sine integral

79: \newcommand\ue{\mathrm{e}}           % e

80: \newcommand\uexp{\mathrm{exp}}       % "exp"

81: \newcommand\ui{\mathrm{i}}           % I

82: \newcommand\ucc{\mathrm{c.c.}}      % complex conjugate

83: \newcommand\uhot{\mathrm{h.o.t.}}   % H.O.T.

84: \newcommand\ud{\mathrm{d}}          % differential sign

85: \newcommand\usign{\mathrm{sign}}    % sign function

86: \newcommand\usin{\mathrm{sin}}      % sine function

87: \newcommand\ucos{\mathrm{cos}}      % cosine function

88: \newcommand\bu{\boldsymbol{u}}

89: \newcommand\p{\ensuremath{\partial}}

90: \newcommand\D{\ensuremath{\Delta}}

91: \newcommand\eps{\epsilon}

92: \newcommand\Om{\Omega}

93: \newcommand\om{\omega}

94: %

95: % For sans serif characters:

96: % The following macros are setup in JFM.cls for sans-serif fonts in text

97: % and math.  If you use these macros in your article, the required fonts

98: % will be substituted when you article is typeset by the typesetter.

99: %

100: % \textsfi, \mathsfi   : sans-serif slanted

101: % \textsfb, \mathsfb   : sans-serif bold

102: % \textsfbi, \mathsfbi : sans-serif bold slanted (doesn't exist in CM fonts)

103: %

104: % For san-serif Roman use \textsf and \mathsf as normal.

105: %

106: \newcommand\ssC{\mathsf{C}}    % for sans serif C

107: \newcommand\sfsP{\mathsfi{P}}  % for sans serif sloping P

108: \newcommand\slsQ{\mathsfbi{Q}} % for sans serif bold-sloping Q

109: 

110: % Hat position

111: \newcommand\hatp{\skew3\hat{p}}      % p with hat

112: \newcommand\hatR{\skew3\hat{R}}      % R with hat

113: \newcommand\hatRR{\skew3\hat{\hatR}} % R with 2 hats

114: \newcommand\doubletildesigma{\skew2\tilde{\skew2\tilde{\Sigma}}}

115: %       italic Sigma with double tilde

116: 

117: % array strut to make delimiters come out right size both ends

118: \newsavebox{\astrutbox}

119: \sbox{\astrutbox}{\rule[-5pt]{0pt}{20pt}}

120: \newcommand{\astrut}{\usebox{\astrutbox}}

121: 

122: \newcommand\GaPQ{\ensuremath{G_a(P,Q)}}

123: \newcommand\GsPQ{\ensuremath{G_s(P,Q)}}

124: %\newcommand\p{\ensuremath{\partial}}

125: \newcommand\tti{\ensuremath{\rightarrow\infty}}

126: \newcommand\kgd{\ensuremath{k\gamma d}}

127: \newcommand\shalf{\ensuremath{{\scriptstyle\frac{1}{2}}}}

128: \newcommand\sh{\ensuremath{^{\shalf}}}

129: \newcommand\smh{\ensuremath{^{-\shalf}}}

130: \newcommand\squart{\ensuremath{{\textstyle\frac{1}{4}}}}

131: \newcommand\thalf{\ensuremath{{\textstyle\frac{1}{2}}}}

132: \newcommand\Gat{\ensuremath{\widetilde{G_a}}}

133: \newcommand\ttz{\ensuremath{\rightarrow 0}}

134: \newcommand\ndq{\ensuremath{\frac{\mbox{$\partial$}}{\mbox{$\partial$} n_q}}}

135: \newcommand\sumjm{\ensuremath{\sum_{j=1}^{M}}}

136: \newcommand\pvi{\ensuremath{\int_0^{\infty}%

137:   \mskip \ifCUPmtlplainloaded -30mu\else -33mu\fi -\quad}}

138: 

139: \newcommand\etal{\mbox{\textit{et al.}}}

140: \newcommand\etc{etc.\ }

141: \newcommand\eg{e.g.\ }

142: 

143: 

144: \newtheorem{lemma}{Lemma}

145: \newtheorem{corollary}{Corollary}

146: 

147: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

148: % end JFM caption

149: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

150: 

151: %\def\baselinestretch{2}

152: 

153: \title[Flow induced by a randomly ...]

154: {Flow induced by a randomly vibrating boundary}

155: \author[D. Volfson and J. Vi\~nals]{DMITRI VOLFSON$^{1}$ AND JORGE 

156: VI\~NALS$^{1,2}$}

157: 

158: \affiliation{$^1$Supercomputer Computations Research Institute,

159:              Florida State University, Tallahassee, Florida 32306-4130, USA\\

160:              $^{2}$ Department of Chemical Engineering,

161:              FAMU-FSU College of Engineering, Tallahassee, Florida

162:                 31310-6046, USA}

163: \pubyear{1999}

164: \volume{000}

165: \pagerange{000--000}

166: \date{\today}

167: \setcounter{page}{1}

168: 

169: \begin{document}

170: \maketitle

171: 

172: \begin{abstract}

173: We study the flow induced by random vibration of a solid

174: boundary in an otherwise quiescent fluid. The analysis is motivated by

175: experiments conducted under the low level and random effective

176: acceleration field that is 

177: typical of a microgravity environment. When the boundary is planar and

178: is being vibrated along its own plane, the variance of the velocity field

179: decays as a power law of 

180: distance away from the boundary. If a low frequency cut-off is introduced in

181: the power spectrum of the boundary velocity, the variance

182: decays exponentially for distances larger than a Stokes layer thickness

183: based on the cut-off frequency. Vibration of a gently curved boundary

184: results in steady streaming in the ensemble average of the tangential

185: velocity. Its amplitude diverges logarithmically with distance away

186: from the boundary, but asymptotes to a constant value instead

187: if a low frequency cut-off is considered. This steady component of the 

188: velocity is shown to depend

189: logarithmically on the cut-off frequency. Finally, we consider the case

190: of a periodically modulated solid boundary that is being randomly vibrated.

191: We find steady streaming in the ensemble average of the first order velocity,

192: with flow extending up to a characteristic distance of the order of the 

193: boundary wavelength. The structure of the flow in the vicinity of the 

194: boundary depends strongly on the correlation time of the boundary velocity.

195: \end{abstract}

196: 

197: \section{Introduction}

198: \label{sec:introduction}

199: 

200: This paper examines the formation and separation of viscous layers 

201: in a fluid which is in contact with a

202: solid boundary that is vibrated randomly. The analysis is motivated by the low

203: level and random acceleration field that affects a number of microgravity

204: experiments. We first study the case of a planar boundary to

205: generalize the classical result of \cite{re:stokes51} who considered

206: a boundary vibrated periodically along its own plane.

207: We next consider a slightly curved boundary, and show that steady streaming 

208: appears in the ensemble average at first order in the perturbed flow

209: variables. There are

210: several qualitative similarities and differences with the classical result by

211: \cite{re:schlichting79} for the case of periodic vibration. Finally, we

212: address the case of a modulated boundary that is vibrated randomly.

213: 

214: Our study is motivated by the significant levels of residual accelerations

215: (g-jitter) that have been detected during space missions in which microgravity 

216: experiments have been conducted

217: (\cite{re:walter87,re:nelson91,re:delombard97}). 

218: Direct measurement of these residual accelerations has shown that they have a 

219: wide frequency spectrum, ranging approximately from $10^{-4} Hz$ to $10^{2}

220: Hz$. Amplitudes range from  $10^{-6}g_{E}$ at the lowest end of

221: the frequency spectrum, and increase roughly linearly for high

222: frequencies, reaching values of $10^{-4}g_{E}\;-\;10^{-3}g_{E}$

223: at frequencies of around $10 Hz$ ($g_{E}$ is the intensity of

224: the gravitational field on the Earth's surface). Despite the efforts of a 

225: number of researchers over the last decade, there remain areas of uncertainty

226: about the potential effect of such a residual acceleration field on

227: typical microgravity fluid experiments, especially in quantitative terms. 

228: A better understanding of the response of a fluid to such 

229: disturbances would enable improved experiment design to minimize 

230: or compensate for their influence. In addition, it would also be useful to 

231: have error estimates of quantities measured in the presence of residual 

232: accelerations, including whenever possible some methodology for extrapolation 

233: to ideal zero gravity.

234: 

235: The formation of viscous layers around solid boundaries 

236: when the flow amplitude has a random component has not been addressed yet

237: despite its potential relevance for a number of microgravity experiments. 

238: Among them we

239: mention the dynamics of colloidal suspensions, coarsening studies of 

240: solid-liquid mixtures in which purely diffusive controlled transport

241: is desired,

242: or the interaction between the viscous layer produced by bulk flow of random 

243: amplitude and the morphological instability of a crystal-melt interface.

244: Our study represents the first step in this direction, and focuses on simple

245: geometries in order to elucidate those salient features of the flow that arise 

246: from the random nature of the vibration.

247: 

248: Previous theoretical work on the influence of g-jitter on fluid flow

249: ranges from order of magnitude estimates to

250: detailed numerical calculations. For example, the order of magnitude of

251: the contributions to fluid flow from the residual acceleration field

252: may be estimated from the length and time scales of a particular

253: experiment, and the values of the relevant set of dimensionless numbers

254: (\cite{re:alexander90}). Such studies are of interest as a first approximation,

255: but are not very accurate. Other studies have modeled the residual

256: acceleration field by some simple analytic function in which the acceleration 

257: is typically decomposed into steady and time dependent components, the latter 

258: being periodic in time

259: (\cite{re:gershuni76,re:kamotani81,re:alexander91,re:farooq94,re:grassia98a,re:grassia98b,

260: re:gershuni98}).

261: A few studies have also addressed the consequences of isolated pulses

262: of short duration (\cite{re:alexander97}).

263: 

264: \cite{re:zhang93} and \cite{re:thomson97} adopted a statistical 

265: description of the residual acceleration field onboard spacecraft, and modeled 

266: the acceleration time series as a stochastic process in time.

267: The main premise of this approach is that a statistical

268: description is necessary in those cases in which the characteristic

269: time scales of the physical process under investigation are long

270: compared with the correlation time of $g$-jitter, $\tau$ (the acceleration

271: amplitudes and orientations at two different times are statistically 

272: independent if separated by an interval larger than $\tau$).

273: Progress has been achieved through the consideration of a specific stochastic 

274: model according to which each Cartesian component of the residual acceleration 

275: field $\vec{g}(t)$ is modeled as a narrow band noise. This noise is a Gaussian 

276: process defined by three independent parameters: its intensity 

277: $\left<g^2\right>$, a dominant angular frequency $\Omega$, and a 

278: characteristic spectral width $\tau^{-1}$. Each realization of narrow band 

279: noise can be viewed as a temporal sequence of periodic functions of angular 

280: frequency $\Omega$ with amplitude and phase that remain constant only for a 

281: finite amount of time ($\tau$ on average). At random intervals, new values of 

282: the amplitude and phase are drawn from prescribed distributions. This model is 

283: based on the following mechanism underlying the residual acceleration field: 

284: one particular natural frequency of vibration of the spacecraft structure 

285: ($\Omega$) is excited by some mechanical disturbance inside the spacecraft, 

286: the excitation being of random amplitude and taking place at a sequence of 

287: unknown (and essentially random) instants of time. 

288: 

289: Narrow band noise has been 

290: shown to describe reasonably well many of the features of g-jitter time 

291: series measured onboard Space Shuttle by \cite{re:thomson97}. Actual 

292: $g$-jitter data collected during the SL-J mission were analyzed, and a time 

293: series of roughly six hours was studied in detail. A scaling analysis revealed 

294: the existence of both deterministic and stochastic components in the time 

295: series. The deterministic contribution appeared at a frequency of $17$ Hz, 

296: with an amplitude $\left<g^2\right>^{1/2}= 3.56 \times 10^{-4} g_{E}$. 

297: Stochastic components included two well defined spectral features with a 

298: finite correlation time; one at 22 Hz with $\left<g^2\right>^{1/2}= 3.06 

299: \times 10^{-4}g_{E}$ and $\tau = 1.09$ s, and one at 44 Hz with 

300: $\left<g^2\right>^{1/2}= 5.20 \times 10^{-4} g_{E}$ and $\tau = 0.91$ s. White 

301: noise background is also present in the series with an intensity

302: $D = 8.61 \times 10^{-4} {\rm cm}^{2}/{\rm s}^{3}$.

303: 

304: A further theoretical advantage of narrow band noise is that it provides a 

305: convenient way of

306: interpolating between monochromatic noise (akin to studies involving a

307: deterministic and periodic gravitational field), and white noise (in which

308: no frequency component is preferred). In the limit $ \tau\rightarrow 0$ with

309: $D = \left< g^{2}\right> \tau$ finite, narrow band noise reduces to white noise

310: of intensity $D$; whereas, for $\tau\rightarrow\infty$ with $\left< g^{2}

311: \right>$ finite, monochromatic noise is recovered. 

312: 

313: We discuss in this paper the flow induced in an otherwise quiescent fluid by 

314: the random vibration of a solid boundary. The velocity of the boundary 

315: $u_{0}(t)$ is assumed prescribed, and modeled as a narrow band stochastic 

316: process. First, we consider an infinite planar boundary that is being vibrated 

317: along its own plane to generalize the classical problem studied 

318: by \cite{re:stokes51}. In the monochromatic limit, the variance of the

319: velocity field decays exponentially away from the wall, with a

320: characteristic decay

321: length given by the Stokes layer thickness $\delta_s=(2\nu/\Om)^{1/2}$,

322: where $\nu$ is the kinematic viscosity of the fluid, and $\Om$ is the angular

323: frequency of vibration of the boundary. Since the equations governing the flow

324: are linear, we are able to obtain an analytic solution describing transient

325: layer formation in the stochastic case, but only in the neighborhood of the 

326: white and monochromatic noise limits. We then show that for any finite 

327: correlation time the stationary variance of the tangential velocity 

328: asymptotically decays as 

329: the inverse squared distance from the wall, in contrast with the exponential

330: decay in the deterministic case. This asymptotic behavior originates from

331: the low frequency range of the power spectrum of the boundary velocity. The

332: crossover from power law to exponential decay is explicitly computed by

333: introducing a low frequency cut-off in the power spectrum.

334: 

335: We next investigate two additional geometries in which the equations governing

336: fluid flow are not linear, and show that several of the generic features

337: obtained for the case of a

338: planar boundary still hold. In the first case, we generalize the analysis of

339: \cite{re:schlichting79} concerning secondary steady streaming.

340: He found that the oscillatory motion of the boundary

341: induces a steady secondary flow outside of the viscous boundary layer 

342: even when the velocity of the boundary averages to zero. If the thickness of 

343: the Stokes layer, $\delta_s$, and the amplitude of oscillation, $a$, are small

344: compared with a characteristic length scale of the  boundary $L$ 

345: ($\delta_s \ll L$, $a\ll L$), then the generation of secondary steady 

346: streaming may be described as follows. Vibration of the rigid boundary gives 

347: rise to an oscillatory and nonuniform motion of the fluid. The flow is

348: potential in the bulk, and rotational in the boundary layer because of

349: no-slip conditions 

350: on the boundary. The bulk flow applies pressure at the outer edge of boundary 

351: layer, which does not vary across the layer. The non uniformity

352: of the flow leads to vorticity convection in the boundary layer through 

353: nonlinear terms. Both convection and the applied pressure drive

354: vorticity diffusion, and thus induce secondary steady motion which does not 

355: vanish outside of the boundary layer. In the simplest case in which the far 

356: field velocity is a standing wave $U(x,t)=U(x)\ucos(\Om t)$, the

357: tangential component of the secondary steady velocity is,

358: \begin{equation}

359: u^{(s)}=-\frac{3}{4\Om}U\frac{\ud U}{\ud x},

360: \label{sch0}

361: \end{equation}

362: where $x$ is a curvilinear coordinate along the boundary. In fact, Eq. 

363: (\ref{sch0}) serves as the boundary condition for the stationary part of the 

364: flow in the bulk. Similar conclusions have been later reached by 

365: \cite{re:batchelor67} who studied sinusoidal oscillations of

366: nonuniform phase, and by \cite{re:gershuni98} who studied

367: monochromatic oscillations of a general form.

368: 

369: The second geometry that we address is the so called wavy wall 

370: (\cite{re:lyne71}). The deterministic limit in which a wavy boundary is being 

371: periodically vibrated has been studied by a number of authors, mainly to

372: address the interaction between the flow above the sea bed and ripple

373: patterns on it (\cite{re:lyne71,re:kaneko79,re:vittori89,re:blondeaux94} and 

374: references therein). \cite{re:lyne71} deduced the existence of steady

375: streaming in the limit in which the amplitude of the wall deviation

376: from planarity is small compared with the thickness of the Stokes

377: layer. He introduced a conformal transformation and obtained an explicit

378: solution in the limit of small $k\Rey$, where $k$ is the wavenumber of

379: the wall profile scaled by the thickness of the Stokes layer, and 

380: $\Rey$ is the Reynolds number. The detailed structure of the secondary

381: flow depends on the ratio between the wavelength of the boundary

382: profile and the thickness of the Stokes layer.

383: 

384: In Sections \ref{sec:sch} and \ref{sec:wavy_wall}, we discuss how the results 

385: for these two geometries generalize to the case of stochastic vibration.

386: Section \ref{sec:sch} addresses the flow created by a gently curved solid

387: boundary that is being vibrated randomly. The perturbation parameter that we use

388: is the ratio between the amplitude of vibration and the characteristic inverse curvature

389: of the wall. The ensemble average of the stream function is not zero, 

390: and hence there exists stationary streaming in the

391: stochastic case as well. The average velocity diverges logarithmically away 

392: from the boundary because of the low frequency range of the power spectrum. 

393: We again introduce a low frequency cut-off $\om_c$ in the spectrum, and study the

394: dependence of the stationary streaming on the cut-off frequency. 

395: We compute the stationary tangential velocity as a function of $\om_c \ll 1$

396: and arbitrary $\beta$, and find a weak (logarithmic) singularity as 

397: $\om_c \rightarrow 0$.

398: 

399: Section \ref{sec:wavy_wall} discusses the formation of a boundary layer around

400: a wavy boundary that is vibrated randomly. Positive and negative vorticity

401: production in adjacent regions of the boundary introduces a natural

402: decay length

403: in the solution, thus leading to exponential decay of the flow away from the

404: boundary, even in the absence of a low frequency cut-off in the power spectrum

405: of the boundary velocity. Steady streaming is found at second order comprising

406: two or four recirculating cells per period of the boundary profile. The number

407: of cells depends on the scaled correlation time $\Om \tau$.

408: 

409: \section{Randomly vibrating planar boundary}

410: \label{sec:plane}

411: 

412: We first examine the case of a planar boundary that is being vibrated along 

413: its own plane. In this case the governing equations are considerably simpler 

414: then in the more general geometries discussed in Sections \ref{sec:sch} 

415: and \ref{sec:wavy_wall}. In particular, the Navier-Stokes equation is linear, 

416: fact that allows a complete solution of the flow. Nevertheless, this simple 

417: solution still exhibits several of the qualitative features that are present in

418: the case of random forcing by a curved boundary, namely asymptotic power law 

419: decay of the velocity field away from the boundary, and sensitive dependence 

420: on the low frequency range of the power spectrum of the boundary velocity.

421: 

422: Consider an infinite solid boundary located at $z=0$, and an

423: incompressible fluid that 

424: occupies the region $z > 0$. The Navier-Stokes equation, and boundary 

425: conditions are,

426: \begin{equation}

427:   \partial_t{u} = \nu\partial_z^2{u},

428: \label{plane1}

429: \end{equation}

430: \begin{equation}

431:   u(0,t)=u_0(t), \quad u(\infty,t) < \infty,

432: \label{plane2}

433: \end{equation}

434: where $z$ is the coordinate normal to the boundary, $u(z,t)$ is the $x$

435: component of the velocity, and $u_{0}(t)$ is the prescribed velocity of the 

436: boundary. The solution for harmonic vibration $u_0(t)=u_0\ucos(\Om t)$

437: was given

438: by \cite{re:stokes51}. It is a transversal wave that propagates into the bulk

439: fluid with an exponentially decaying amplitude,

440: \begin{equation}

441: u(z,t)=u_0\ue^{-z/\delta_s}\ucos(\Om t - z/\delta_s),

442: \label{stokes}

443: \end{equation}

444: where $\delta_s=(2\nu/\Om)^{1/2}$ is the Stokes layer thickness.

445: 

446: \subsection{Narrow band noise}

447: 

448: As discussed in the introduction, the main topic of this paper is to 

449: examine how the nature of the bulk flow changes when the boundary velocity 

450: $u_{0}(t)$ is a random process. Specifically, we consider a Gaussian process 

451: defined by

452: \begin{equation}

453:   \left< u_0(t)\right>=0, \quad

454:   \left<u_{0}(t)u_{0}(t')\right>=\left<{u_0}^{2}\right>

455:   \ue^{-|t-t'|/\tau}\ucos\Om(t-t').

456: \label{nbn2}

457: \end{equation}

458: This process is known as narrow band noise (\cite{re:stratonovich67}).

459: It is defined by three independent parameters: its variance 

460: $\left<u_0^2\right>$, its dominant angular frequency $\Om$, and the 

461: correlation time $\tau$. Each realization of this random process can be viewed 

462: as a sequence periodic functions of frequency $\Om$, with amplitude and phase 

463: that remain constant for a time interval $\tau$ on average. White noise is 

464: recovered when $\Om\tau\rightarrow 0$ while $D = \left<{u_0}^{2}\right>\tau$

465: remains finite, whereas the monochromatic noise limit corresponds to

466: $\Om\tau\rightarrow \infty$, with $\left<{u_0}^{2}\right>$ finite.

467: Monochromatic noise is akin to a single frequency periodic signal of the same

468: frequency, but with randomly drawn amplitude and phase. The relationship 

469: between the two can be illustrated by considering a deterministic function 

470: $x(t)=x_0\ucos(\Om t)$ and defining the temporal average as,

471: \begin{equation}

472:   \left<x(t)x(t')\right> =

473:   \lim_{T\rightarrow\infty}\frac{1}{T}\int_0^T\ud t\,x(t)x(t')=

474:   \frac{x_0^2}{2}\ucos(\Om(t-t')).

475:   \label{nbn21}

476: \end{equation}

477: This average coincides with the ensemble average of the noise when 

478: $\left<u_0^2\right>=x_0^2/2$. The power spectrum corresponding to the 

479: correlation function (\ref{nbn2}) is

480: \begin{equation}

481:   P(\omega)=\frac{\left<u_0^2\right>}{2\upi}\left[\frac{\tau}

482:   {1+\tau^2(\omega-\Omega)^2}+\frac{\tau}{1+\tau^2(\omega+\Omega)^2}\right].

483:   \label{nbn3}

484: \end{equation}

485: We will also use the spectral density of the process $u_0(t)$,

486: \begin{equation}

487:   \hat{u}_0(\om)=\frac{1}{2\upi}\int_{-\infty}^{\infty}

488:   \ud t\,u_0(t)\ue^{-\ui\om t},

489:   \label{spec_dens}

490: \end{equation}

491: so that its ensemble average and correlation function are respectively given by

492: \refstepcounter{equation}

493: $$

494:   \left<\hat{u}_0(\om)\right>=0,\quad

495:   \left<\hat{u}_0(\om)\hat{u}^{*}_{0}(\om')\right>=\delta(\om-\om')P(\om).

496:   \eqno{(\theequation{\mathit{a},\mathit{b}})}

497:   \label{spec_dens_cor}

498: $$

499: We will often use dimensionless variables in which $\left<u_0^2\right>/\Om$ 

500: is the scale of $P(\om)$, and $\Om$ is the angular frequency scale. In

501: dimensionless form,

502: \begin{equation}

503:   P(\omega,\beta)=\frac{1}{2\upi}\left[\frac{\beta}{1+\beta^2(\omega-1)^2}+

504:   \frac{\beta}{1+\beta^2(\omega+1)^2}\right],

505:   \label{nbn6}

506: \end{equation}

507: where $\beta=\Omega\tau$.

508: We have $\int_{-\infty}^{\infty}\ud\om P(\om,\beta)=1$, independent of $\beta$,

509: and also $\lim_{\beta \rightarrow \infty} P(\omega)

510: = \left[\delta(\omega-1)+\delta(\omega+1)\right]/2$.

511: Note that the power spectrum does not vanish at small frequencies. Instead,

512: $P(0,\beta)=\beta/(\upi(1+\beta^2)$, which for large and small $\beta$

513: behaves as $P(0,\beta)\sim 1/(\upi \beta)$ and $P(0,\beta)\sim \beta/\upi$

514: respectively. We will discuss separately the effect of this

515: low frequency contribution on the results presented in the remainder of the

516: paper.

517: 

518: \subsection{Transient layer formation}

519: \label{subsec:transient}

520: 

521: In the two limiting cases of white and monochromatic noise, it is possible

522: to find an analytic solution for the transient flow starting from an initially

523: quiescent fluid. The solution can be found, for example, by introducing

524: the retarded, infinite space Green's function  corresponding to

525: equation (\ref{plane1}), with boundary conditions (\ref{plane2}),

526: \begin{equation}

527:   G(z,t|z',t')=\frac{1}{(4\upi\nu(t-t'))^{1/2}}\left[

528:   \ue^{-(z-z')^2/4\nu(t-t')}-\ue^{-(z+z')^2/4\nu(t-t')}\right],

529:   \quad t>t',

530: \label{greenplane}

531: \end{equation}

532: and $G(z,t|z',t')=0$ for $t<t'$. If the fluid is initially quiescent,

533: $u(z,0)=0$, we find

534: \begin{equation}

535: u(z,t)=\nu\int_{0}^{t}\ud t'\,u_0(t')\left(\p_{z'}G\right)_{z'=0},

536: \label{quis_sol}

537: \end{equation}

538: with

539: \begin{equation}

540:   \left(\p_{z'}G\right)_{z'=0}=\frac{z}{[4\upi\nu^3(t-t')^3]^{1/2}}\,

541:   \ue^{-z^2/4\nu(t-t')}.

542: \label{dif_green}

543: \end{equation}

544: Equations (\ref{quis_sol}-\ref{dif_green}) determine the transient behavior for

545: any given $u_0(t)$. 

546: 

547: If $u_0(t)$ is a Gaussian, white noise process, the ensemble average of 

548: Eq.(\ref{quis_sol}) yields $\left<u(z,t)\right>=0$. The corresponding equation 

549: for the variance reads

550: \begin{equation}

551: \left<u^2(z,t)\right>=2D\nu^2\int_0^t\ud t'\,\left[\left(\p_{z'}G\right)_{z'=0}

552: \right]^2=

553: \frac{2D\nu}{\upi z^2}\left(1+\frac{z^2}{2\nu t}\right)\,\ue^{-z^2/2\nu t}.

554: \label{wn_var}

555: \end{equation}

556: The variance of the induced fluid velocity propagates into the fluid 

557: diffusively.

558: Saturation occurs for $t\gg z^2/2\nu$, at which point the variance does 

559: not decay exponentially far away from the wall, but rather as a power law.

560: \begin{equation}

561: \left<u^2(z,\infty)\right>=\frac{2D\nu}{\upi z^2}.

562: \label{wn_var_longtime}

563: \end{equation}

564: Ascertaining whether random vibration can induce flows far away from

565: the boundary in more general geometries is one of the main 

566: motivations for this paper.

567: 

568: Consider now the opposite limit of monochromatic noise with

569: correlation function

570: \begin{equation}

571: \left<u_0(t)u_0(t')\right>=\left<u_0^2\right>\ucos\left[\Om(t-t')\right] .

572: \label{monocf}

573: \end{equation}

574: Now using Eqs. (\ref{quis_sol}) and (\ref{monocf}) we find, 

575: (\cite{re:carslaw59}),

576: \begin{equation}

577: \frac{\left<u^2(z,t)\right>}{2\left<u_0^2\right>}=\frac{2}{\upi}

578: \int_{\kappa}^{\infty}\ud\sigma\,\ue^{-\sigma^2}

579: \int_{\kappa}^{\infty}\ud\mu\,\ue^{-\mu^2}

580: \ucos\left[\frac{z^2}{2\delta_s^2}\left(\frac{1}{\mu^2}-\frac{1}{\sigma^2}

581: \right)\right],

582: \label{monovar}

583: \end{equation}

584: with $\kappa = z/(4\nu t)^{1/2}$. A closed form solution can only be obtained 

585: for long times. We find,

586: \begin{equation}

587: \frac{\left<u^2(z,t)\right>}{2\left<u_0^2\right>}=\frac{\ue^{-2z/\delta_s}}{2}

588: +\frac{2\kappa^3\delta_s^2}{\upi^{1/2}z^2}\ue^{-z/\delta_s}\usin(\Om\,t-z/

589: \delta_s)+{\cal O}(\kappa^5(t)).

590: \label{monovarcor}

591: \end{equation}

592: At long times,  the variance propagates into the bulk with phase velocity 

593: $(2\nu\Om)^{1/2}$, while its amplitude decays exponentially in space over the 

594: scale of the Stokes layer, and as $t^{-3/2}$ in time. In summary, the flow

595: created by the vibration of the boundary propagates diffusively for 

596: white noise $(z^2\propto 2\nu t)$, and as a power law 

597: $(z^2\propto \upi\nu\Om^2 t^3)$ in the monochromatic limit.

598: 

599: \subsection{Stationary variance for narrow band noise}

600: 

601: We have been unable to obtain a closed analytic solution for the transient 

602: evolution of the variance $\left< u(z,t)^{2} \right>$ when the

603: vibration of the 

604: boundary is given by a general narrow band process. It is possible, 

605: however, to obtain the stationary variance of the velocity. Equation 

606: (\ref{plane1}) can be rewritten in Fourier space as

607: \begin{equation}

608: \ui\om\hat{u}(z,\om)=\nu\p^2_z\hat{u}(z,\om)

609: \label{fourier2}

610: \end{equation}

611: with

612: \begin{equation}

613: u(z,t)=\int_{-\infty}^{\infty}\ud\om\,\hat{u}(z,\om)\ue^{\ui\om t},

614: \label{fourier1}

615: \end{equation}

616: The boundary conditions are,

617: $\hat{u}(0,\om)=\hat{u}_0(\om)$, and $\hat{u}(z,\om)<\infty$ at $z \rightarrow

618: \infty$. The solution of Eq. (\ref{fourier2}) with these boundary conditions is

619: \begin{equation}

620:   \hat{u}(z,\om)=\ue^{-\alpha z}\hat{u}_0(\om),

621:   \quad \alpha=(1+\ui\,\usign(\om))(\om/2\nu)^{1/2}

622: \label{fourier4}

623: \end{equation}

624: 

625: We next choose $1/\Om$ as the time scale, and $\delta_s$ as the length scale, 

626: and after some straightforward algebra we find,

627: \begin{equation}

628:   \frac{\left<u^2(z,\beta)\right>}{2 \left<u_{0}^{2}\right>} = I(z,\beta)=

629: \int_{0}^{\infty} \ud\om P(\om,\beta)e^{-z\om^{1/2}}.

630: \label{plane3}

631: \end{equation}

632: We have also used the fact that the power spectrum (\ref{nbn6}) is even 

633: in frequency.

634: 

635: We next analyze the asymptotic dependence of $I(z,\beta)$ at large $z$.

636: In this limit, $I(z,\beta)$ mainly depends on the low frequency region of the

637: power spectrum; higher frequencies are suppressed by the exponential factor.

638: By using Watson's lemma (\cite{re:nayfeh81}), we find,

639: \begin{equation}

640: I(z,\beta)=\frac{2\beta}{\upi(1+\beta^2)}\frac{1}{z^2} +

641: \frac{240\beta^3(3\beta^2-1)}{\upi(1+\beta^2)^3}\frac{1}{z^6} +

642: {\cal O}(z^{-10}).

643: \label{plane4}

644: \end{equation}

645: This asymptotic form at large $z$ is valid for all $\beta$.

646: In particular, the dominant behavior for small and large $\beta$ is

647: \begin{displaymath}

648: I(z,\beta) \sim \frac{2\beta}{\upi}\frac{1}{z^2},\quad

649: I(z,\beta) \sim \frac{2}{\upi\beta}\frac{1}{z^2},

650: \end{displaymath}

651: respectively. Hence we recover the power law decay of Eq. 

652: (\ref{wn_var_longtime}).

653: 

654: We can also find the asymptotic behavior at large $\beta$ that is 

655: uniformly valid in $z$,

656: \begin{equation}

657: I(z,\beta)=\frac{\ue^{-z}}{2}-\frac{z}{2\upi\beta}\left(

658: \Ci(z)\sin(z)-\Si(z)\cos(z)-\frac{1}{2}(\ue^{-z}\Ei(z)-\ue^z\Ei(-z))

659: \right) + {\cal O}(\beta^{-3}),

660: \label{plane5}

661: \end{equation}

662: where $\Ci$ and $\Si$ denote the integral sine and cosine functions, 

663: and $\Ei$ stands for the exponential integral function 

664: (\cite{re:gradshteyn80}).

665: For $z\lesssim 1$, the variance decreases exponentially. At larger $z$,

666: the exponential terms in (\ref{plane5}) become small, so that the remaining 

667: asymptotic dependence for large $z$ is given by Eq. (\ref{plane4}).

668: The quantity $I(z,\beta)\,z^2$ computed both from (\ref{plane3}) and the

669: uniform expansion (\ref{plane5}) is presented in Fig.\ref{fig:plane_var}.

670: For fixed $\beta$, $I(z,\beta)$ asymptotes to $2\beta/\upi(1+\beta^2)$

671: outside of Stokes layer. This value is the coefficient of the

672: leading term in the asymptotic expansion (\ref{plane4}). The expansion 

673: (\ref{plane5}) is a good approximation even for moderate values of $\beta$.

674: 

675: To summarize, the variance of the velocity field does not decay

676: exponentially away from the wall for finite $\beta$, but rather as the 

677: inverse squared distance. The crossover length separating exponential

678: and power law decay increases with increasing $\beta$.

679: 

680: \subsection{Low frequency cut-off in the power spectrum}

681: \label{subsec:cutplane}

682: 

683: The coefficient of the leading term in (\ref{plane4}) is in fact

684: twice the value of $P(0,\beta)=\beta/\upi(1+\beta^2)$. The algebraic decay of 

685: $\left<u_0^2(z,t)\right>$ follows from the diffusive nature of Eq. 

686: (\ref{plane1}), and a non vanishing value of $P(\om,\beta)$ as $\beta

687: \rightarrow 0$. Before we analyze in Sections \ref{sec:sch} and 

688: \ref{sec:wavy_wall} how this behavior is modified by nonlinearities in the

689: governing equations, we explicitly address here the consequences of a low 

690: frequency cut-off in the power spectrum. Of course, there always exists in 

691: practice a low frequency cut-off because of limited observation time. 

692: Furthermore, the low frequency range of the power spectrum of the residual 

693: acceleration field in microgravity ($\Om/2 \upi < 10^{-3}$Hz) is

694: fairly difficult to measure reliably. We therefore introduce an

695: effective cut-off frequency in the power spectrum, $\om_c\ll1$, and

696: study the dependence of $\left<u_0^2(z,t)\right>$ 

697: on $\om_c$. The stationary value of variance of the velocity is now given by,

698: \begin{equation}

699:   \frac{\left<u^2(z,\beta,\om_c)\right>}{2\left<u_0^2\right>}

700:   = I_c(z,\beta,\om_c)=\int_{\om_c}^{\infty}

701:   \ud\om P(\om,\beta)e^{-z\om^{1/2}}.

702: \label{plane1_cut}

703: \end{equation}

704: By using Watson's lemma, we find for large $z$,

705: \begin{equation}

706:  I_c(z,\beta,\om_c)=\frac{2\beta\ue^{-z\om_c^{1/2}}}{\upi(1+\beta^2)}

707: \left(\frac{1}{z^2}+\frac{\om_c^{1/2}}{z}+\uhot\right),

708: \label{plane2_cut}

709: \end{equation}

710: where $\uhot$ stands for terms which are of higher order than terms 

711: retained under the assumption that both $1/z$ and $\om_c^{1/2}$

712: are small but independent. For $z \gg 1$, but $z \om_c^{1/2} \ll 1$ 

713: the dominant term in (\ref{plane2_cut}) is 

714: \begin{equation}

715:   I_c(z,\beta,\om_c)\sim

716:   \frac{2\beta}{\upi(1+\beta^2)}\frac{1}{z^2},\quad z\om_c^{1/2}<<1.

717: \label{plane3_cut}

718: \end{equation}

719: On the other hand, if $z\om_c^{1/2}\geq 1$, the leading order term is now

720: a function of $\zeta=z\om_c^{1/2}$

721: \begin{equation}

722:   I_c(z,\beta,\om_c)\sim

723:   \frac{2\beta\om_c\ue^{-\zeta}}{\upi(1+\beta^2)}

724:   \left(\frac{1}{2\zeta}+\frac{1}{\zeta^2}\right),\quad \zeta\geq 1

725: \label{plane4_cut}

726: \end{equation}

727: Equations (\ref{plane3_cut}) and (\ref{plane4_cut}) show that at 

728: distances that are large compared with the thickness of the Stokes

729: layer based on the dominant frequency $\Om$, $\left<u^2(z,t)\right>$

730: decays algebraically with $z$. There exists, however, a length scale 

731: $z\om_{c}^{1/2}$ beyond which the decay is exponential. 

732: This new characteristic length scale is the thickness of the Stokes

733: layer based on the cut-off frequency. This conclusion appears natural 

734: given the principle of superposition for the linear differential

735: equation (\ref{plane1}).

736: 

737: \section{Streaming due to random vibration}

738: \label{sec:sch}

739: 

740: Next we investigate to what extent the results of Section \ref{sec:plane} hold

741: in configurations in which the governing equations are not linear. We examine 

742: in this section the flow induced by a gently curved solid boundary

743: that is being randomly vibrated. The boundary velocity is assumed to be

744: described by a narrow band stochastic process, and hence our results will

745: reduce to Schlichting's in the limit of infinite correlation time. 

746: However, for finite values of $\beta$ the results are qualitatively different.

747: The mechanism of secondary steady streaming generation described in the

748: introduction is no longer valid because there is no boundary layer solution at 

749: zeroth order. Vorticity produced at the vibrating boundary penetrates 

750: into the bulk fluid, at least perturbatively for small curvature. 

751: This results in a logarithmic divergence of the ensemble average of

752: the first order velocity with distance away from the wall. A cut-off 

753: analysis is also presented, and similarly to that of Section

754: \ref{subsec:cutplane}, it reveals the existence of an effective boundary

755: layer of thickness based on the cut-off frequency. We then find 

756: an expression analogous to Eq. (\ref{sch0}) as a function of $\beta$ and 

757: $\om_c$.

758: 

759: Define the following dimensionless quantities,

760: \begin{equation}

761: \left.

762: \begin{array}{l}

763: \displaystyle

764: z=\tilde{z}[(\nu/\Om)^{1/2}],\quad x=\tilde{x}[L],\quad

765: t=\tilde{t}[\Om^{-1}],\quad\psi=\tilde{\psi}[(2\left<u_0^2\right>

766: \nu/\Om )^{1/2}] , \\[16pt]

767: \displaystyle

768: \quad \eps=(2\left<u_0^2\right>)^{1/2}/\Om L,\quad \Rey_p=2\left<u_0^2\right>/

769: \Om\nu,

770: \quad\tilde{\Delta}=\p_{\tilde{z}}^2+\eps^2/\Rey_p\p_{\tilde{x}}^2

771: \end{array}

772: \right\}

773: \end{equation}

774: Assume now that the characteristic scale of the boundary $L$ is large so that

775: $\eps$ is a small quantity. If the Reynolds number $Re_{p}$ is assumed to

776: remain finite, both conditions imply $\nu / \Om L^{2} \ll 1$. We next write the

777: governing equations and boundary conditions in the frame of reference co-moving

778: with the solid boundary and obtain for a two dimensional geometry (tildes are

779: omitted),

780: \begin{equation}

781:   \partial_t{\Delta\psi}+\epsilon\frac{\p(\psi,\Delta\psi)}{\p(z,x)}=

782:   \Delta^2\psi

783: \label{sch1}

784: \end{equation}

785: \refstepcounter{equation}

786: $$

787:   \psi=0, \quad \p_z\psi=0 \quad \mbox{at}\quad y=0,\qquad

788:   \eqno{(\theequation{\mathit{a},\mathit{b}})}

789:   \label{sch2}

790: $$

791: \begin{equation}

792:   \p_z\psi=2^{-1/2}U(x)u_0(t) \quad \mbox{at}\quad z=\infty,

793:   \label{sch3}

794: \end{equation}

795: where $x,z$ are the tangential and normal coordinates along the boundary, 

796: and $\psi$ is the stream function $\bu=(\p_z\psi,-\p_x\psi)$. We have also 

797: used the notation $\p(a,b)/\p(z,x) = (\p_za)(\p_xb)-(\p_xa)(\p_zb)$ for the

798: nonlinear term. The far field boundary condition is a nonuniform and random

799: velocity field, of the order of  $\left<u_0^2\right>^{1/2}$, with a spatially 

800: nonuniform amplitude $U(x)$, and a stochastic modulation $u_0(t)$ which is a 

801: Gaussian stochastic process with zero mean, and narrow band power spectrum.

802: We first expand the stream function as a power series of $\eps$, 

803: $ \psi=\psi_0+\eps\psi_1+\ldots $ and solve (\ref{sch1}) order by

804: order. At order $\eps^0$ we obtain the following equation,

805: \begin{equation}

806:   (\p_t{\p_z^2}-\p_z^4)\psi_0=0

807:   \label{sch5}

808: \end{equation}

809: with boundary conditions,

810: \refstepcounter{equation}

811: $$

812:   \psi_0=0, \quad \p_z\psi=0 \quad \mbox{at}\quad z=0,\qquad

813:   \eqno{(\theequation{\mathit{a},\mathit{b}})}

814:   \label{sch6}

815: $$

816: \begin{equation}

817:   \p_z\psi=2^{-1/2}U(x)u_0(t) \quad \mbox{at}\quad z=\infty .

818:   \label{sch7}

819: \end{equation}

820: At this order, the equations effectively describe the flow induced

821: above a planar boundary with a far field velocity boundary condition

822: that is not uniform. The solution can be found by Fourier

823: transformation. We define,

824: \begin{eqnarray}

825:   \psi_0(x,z,t)=\int_{-\infty}^{\infty}\ud\om\,\hat{\psi}_0(x,z,\om)

826: \ue^{\ui\om t}.

827:   \label{sch8}

828: \end{eqnarray}

829: The transformed Eq. (\ref{sch5}) and the transformed boundary conditions 

830: (\ref{sch6}) allow separation of variables. We define,

831: $$

832:   \hat{\psi}_0(x,z,\om)=2^{-1/2}U(x)\hat{u}_0(\om)\hat{\zeta}_0(z,\om),

833: $$

834: so that Eq. (\ref{sch5}) leads to,

835: \begin{equation}

836:   (\ui\om{\p_z^2}-\p_z^4)\hat{\zeta}_0=0,

837:   \label{sch51}

838: \end{equation}

839: with boundary conditions

840: $\hat{\zeta}_0=0$, $\p_z\hat{\zeta}_0=0$, at $z=0$ and

841: $\p_z\hat{\zeta}_0=1$ at $z=\infty$. The solution is,

842: \begin{equation}

843:        \hat{\zeta}_0(z,\om)=-1/\alpha+z+1/\alpha\ue^{-\alpha z},\quad

844:        \alpha(\om) = (1+\ui\,\usign(\om))(|\om|/2)^{1/2}).

845: \label{sch81}

846: \end{equation}

847: %Note that $\hat{\zeta}_0$ grows linearly with $z$ at large $z$.

848:  

849: At order $\eps$ we find,

850: \begin{equation}

851:   (\p_t{\p_z^2}-\p_z^4)\psi_1=\p_x\psi_0\p_z^3\psi_0 - 

852:   \p_z\psi_0\p_x\p_z^2\psi_0

853:   \label{sch9}

854: \end{equation}

855: with boundary conditions $ \psi_1=0, \; \p_z\psi_1=0$ at $y=0$.

856: The remaining boundary condition for $\psi_1$ needs to be discussed separately.

857: Consider first the classical deterministic limit which can be 

858: formally obtained by setting $\beta=\infty$. Then, the right hand side of 

859: Eq. (\ref{sch9}) involves stationary terms (of zero frequency), and sinusoidal 

860: terms with twice the frequency of the far field flow. Since the equation is 

861: linear, the solution $\psi_1$ has exactly the

862: same temporal behavior. In this case, it is known that it is not possible to 

863: find a solution for $\p_z\psi_{1}$ that vanishes at large $z$, but

864: only one that simply remains bounded as $z \rightarrow \infty$.  By

865: analogy, we introduce a similar requirement in the stochastic case of 

866: $\beta < \infty$. Since the

867: zeroth order stream function diverges linearly, this condition simply

868: amounts to requiring that the expansion in powers of $\eps$ remains consistent.

869: 

870: We now take the ensemble average of Eq. (\ref{sch9}), and consider the long

871: time stationary limit of the average 

872: ($\psi_1^{(s)}=\lim_{t\rightarrow\infty}\left<\psi_1\right>$), to find,

873: \begin{equation}

874:   \p_z^4\psi_1^{(s)}=\left<\p_z\psi_0\p_x\p_z^2\psi_0-\p_x\psi_0\p_z^3\psi_0

875:   \right>.

876:   \label{sch101}

877: \end{equation}

878: The right hand side of this equation can be integrated from infinity 

879: to $z$. We obtain,

880: \begin{equation}

881:   \p_z^3\psi_1^{(s)}=\frac{U}{2}\frac{\ud U}{\ud x}F(z,\beta),

882:   \label{sch11}

883: \end{equation}

884: where

885: $$

886: F(z,\beta)=\int_{0}^{\infty}\ud\om P(\om,\beta)Q(\om,z),

887: $$

888: and

889: $$

890:   Q(\om,z)=  (-2+2\hat{\zeta}_0\p_z\hat{\zeta}_0^* -

891: 	     \hat{\zeta}_0\p_z^2\hat{\zeta}_0^* -

892: 	     \hat{\zeta}_0^*\p_z^2\hat{\zeta}_0).

893: $$

894: The power spectrum $P(\om,\beta)$ is defined in Eq. (\ref{nbn6}). The constant 

895: that appears in the expression for $Q(\om,z)$ comes from the pressure gradient 

896: imposed at infinity.

897: 

898: We now proceed to solve Eq.(\ref{sch11}) subject to the boundary conditions

899: $ \psi_1^{(s)}=\p_z\psi_1^{(s)}=0$ on the solid boundary, and 

900: $\p_z\psi_1^{(s)}<\infty$ as $z\rightarrow \infty$.

901: This is a boundary value problem for $\psi_1^{(s)}$. 

902: In the limit $\beta \rightarrow \infty$, it can be solved 

903: analytically, and the result obtained by Schlichting is recovered, namely that 

904: the solution may be bounded at infinity simply by setting the homogeneous 

905: part of the solution equal to zero to satisfy the principle of minimal 

906: singularity (\cite{re:vandyke64}). Otherwise $\p_z\psi_1^{(s)}$ is 

907: singular a fortiori. Since we cannot find an complete analytic

908: solution for finite $\beta$, we proceed as follows. We recast the 

909: boundary value problem as an initial value 

910: problem, and search for a boundary condition on $\p_z^2\psi_1^{(s)}$ at $z=0$ 

911: so that the homogeneous part of the solution remains bounded. This boundary 

912: condition can be found analytically by integrating (\ref{sch11}) from 

913: $0$ to $z$. We find,

914: $$

915:   \p_z^2\psi_1^{(s)}(z)-\p_z^2\psi_1^{(s)}(0)=

916:   \frac{U}{2}\frac{\ud U}{\ud x}\int_0^{z}\ud z'

917:   \int_{0}^{\infty}\ud\om P(\om,\beta)Q(\om,z'),

918: $$

919: or after changing the order of integration,

920: $$

921: \p_z^2\psi_1^{(s)}(z)-\p_z^2\psi_1^{(s)}(0)= 

922: \frac{U}{2}\frac{\ud U}{\ud x}

923: \int_{0}^{\infty}\ud\om P(\om,\beta)

924: \left[

925: \int_z \ud z' Q(\om,z')-\left(\int_z \ud z' Q(\om,z')\right)_{z=0}

926: \right].

927: $$

928: The second integral within brackets equals $(1/2\om)^{1/2}$. In order to 

929: avoid a linear divergence of $\p_z\psi_1^{(s)}(z)$ we equate the

930: constant terms on both sides, thus obtaining the third initial condition

931: \begin{eqnarray}

932: \p_z^2\psi_1^{(s)}(0,\beta) & = & 

933: U\frac{\ud U}{\ud x}

934: \frac{2^{1/2}}{4}\int_0^{\infty}\ud\om\,\om^{-1/2}P(\om,\beta) =

935: \nonumber\\

936: & = & U\frac{\ud U}{\ud x}\frac{(2\beta)^{1/2}}{8q}

937: ((2(q-\beta))^{1/2}+(q+1)^{1/2}+(q-1)^{1/2}),

938: \label{sch12}

939: \end{eqnarray}

940: where $q = (1+\beta^2)^{1/2}$, and the dependence of initial condition 

941: on $\beta$ is shown explicitly. Equation (\ref{sch11}) with its

942: original boundary

943: conditions, supplemented with Eq. (\ref{sch12}) is an initial 

944: value problem, which we have solved numerically. 

945: 

946: Before presenting the numerical results, we study the

947: asymptotic behavior of the solution for large $z$ which

948: is determined by the asymptotic form of $F(z,\beta)$ at large $z$.

949: By explicit substitution of the zeroth order solution we find

950: \begin{equation}

951: Q(z,\om)=(-4\ucos(Z)-2Z(\ucos(Z)+\usin(Z))+2\usin(Z))\ue^{-Z}+2\ue^{-2Z},

952: \label{a2}

953: \end{equation}

954: where $Z = z(\om/2)^{1/2}$. The leading contribution to $F(z,\beta)$

955: as $z\rightarrow\infty$ originates from the zero frequency limit of

956: $P(\om,\beta)$. Thus $ F(z,\beta)\sim

957: P(0,\beta)\int_{0}^{\infty}\ud\om Q(\om,z)$.

958: The remaining integral may be easily calculated to yield the

959: asymptotic form,

960: \begin{equation}

961: F(z,\beta)\sim \frac{6\beta^2}{\upi(1+\beta^2)}\frac{1}{z^2} .

962: \label{sch13}

963: \end{equation}

964: Therefore the stationary mean

965: first order velocity $u_1^{(s)}=\p_z\psi_1^{(s)}$ has a logarithmic 

966: asymptotic form (see Eq. (\ref{sch11})). This is to be contrasted with 

967: the power law decay of the stationary variance for the case of a

968: randomly vibrating planar boundary.

969: 

970: Finally, we have numerically calculated $\partial_{z}\psi_{1}^{(s)}(z)$ for a 

971: range of values of $\beta$. The results are presented in Fig. 

972: \ref{fig:sc_u_prof}. Equation (\ref{sch11}) was integrated numerically

973: with no-slip boundary conditions for $\psi_1^{(s)}$, and Eq.

974: (\ref{sch12}). The numerical results support our conclusion

975: about the logarithmic divergence of $u_1^{(s)}$ with distance for any 

976: finite $\beta$. As $\beta$ increases the stationary mean velocity

977: profile approaches a limiting

978: form that corresponding to the monochromatic limit of 

979: $\beta \rightarrow \infty$.

980: In this limit we recover the Schlichting result, according to which 

981: $\partial_{z} \psi_{1}^{(s)}(z)$ asymptotes to a constant value over a 

982: distance of the order of the deterministic Stokes layer. 

983: 

984: \subsection{Low frequency cut-off in the power spectrum}

985: \label{subsec:cut_sch}

986: 

987: In Section \ref{sec:plane}, we showed that a low frequency cut-off in 

988: $P(\om,\beta)$ led 

989: to an exponential decay of the velocity outside an effective boundary

990: layer of thickness determined by the cut-off frequency. We therefore

991: examine here the consequences of a low frequency cut-off on the

992: divergent behavior of the stationary average of the first order stream

993: function. In order to find the asymptotic form of $\p_z\psi_1^{(s)}$, 

994: we first integrate (\ref{sch11}) twice 

995: from $z=0$ to $z$. By using the low frequency cut-off defined in Section 

996: (\ref{subsec:cutplane}), we write,

997: \begin{equation}

998: \p_z\psi_1^{(s)}(z,\beta,\om_c)=\frac{U}{2}\frac{\ud U}{\ud x}

999: \left[G_c(z,\beta,\om_c)-G_c(0,\beta,\om_c)\right] ,

1000: \label{sch_cut_asym}

1001: \end{equation}

1002: with,

1003: $$

1004: G_c(z,\beta,\om_c)=\int \ud z'\int \ud z' F_c(z',\beta,\om_c),

1005: $$

1006: and,

1007: $$

1008:  F_c(z,\beta,\om_c)=\int_{\om_c}^{\infty}\ud\om P(\om,\beta)Q(\om,z).

1009: $$

1010: If we set $\om_c = 0$ and consider the monochromatic limit of $\beta 

1011: \rightarrow \infty$, we find that $G_c(z,\infty,0)$ contains an exponential

1012: factor that vanishes at $z\sim{\cal O}(1)$, and that $G_c(0,\infty,0) = 3/4$.

1013: Therefore the Schlichting result is recovered. An explicit form of 

1014: $G_c(z,\beta,\om_c)$ for any $\beta$ and $\om_c$ can be obtained

1015: analytically, but it is far too complicated and we do not quote it here.

1016: It has a similar functional dependence as in the cut-off analysis of

1017: the planar boundary, and contains exponential terms involving

1018: $(-z\om_c^{1/2})$. We find that $\p_z\psi_1^{(s)}(\infty,\beta,\om_c)=

1019: -U\ud U/2\ud x\, G_c(0,\beta,\om_c).$ For finite but small $\om_c$, we find,

1020: \begin{equation}

1021: \p_z\psi_1^{(s)}(\infty,\beta,\om_c)=-\frac{3}{4}U\frac{\ud U}{\ud x}

1022: \frac{\beta}{\upi(1+\beta^2)}

1023: \left(2\beta\arctan(\beta)-\ln\left(\frac{\beta^2\om_c^2}{1+\beta^2}\right) 

1024: \right) + {\cal O}(\om_c^2)

1025: \label{sch_cut_asym2}

1026: \end{equation}

1027: This asymptotic formula is compared with the numerically computed

1028: value of the tangential mean stationary velocity at large distances

1029: in Figure \ref{fig:usc_cut}. Computations were done as described in the

1030: previous section, and for different values of $\om_c\ll 1$ and

1031: $\beta$. In all cases

1032: $\p_z\psi_1^{(s)}$ reached constant values at large enough $z$ (the numerical

1033: value of infinity, $z_{\infty}$, was chosen so that the change in velocity for

1034: $z\geq z_{\infty}$ was less than a prescribed tolerance). We also checked

1035: that any change in the boundary condition

1036: Eq.(\ref{sch12}) leads to a linear divergence in the

1037: tangential mean stationary velocity, thus confirming the adequacy of this 

1038: boundary condition. The figure also shows that the computed values

1039: of $\p_z\psi_1^{(s)}(z=\infty,\beta,\om_c)$ for small $\om_c$  are in a good

1040: agreement with (\ref{sch_cut_asym2}).

1041: 

1042: In summary, Eq. (\ref{sch_cut_asym2}) shows that the tangential velocity away

1043: from the boundary asymptotes to a constant that is a function of

1044: $\beta$, and

1045: has a weak (logarithmic) dependence on the cut-off frequency $\om_c$. 

1046: Therefore the asymptotic dependence in the stochastic case (with a cut-off) 

1047: and the deterministic case is

1048: qualitatively similar, although the value of the asymptotic velocity of the

1049: former depends on $\beta$. Note also that this asymptotic behavior only

1050: sets in for distances larger than $(\nu/\om_{c}\Om)^{1/2}$, a value that can be

1051: quite large in practical microgravity conditions.

1052: 

1053: \section{Randomly vibrating wavy boundary}

1054: \label{sec:wavy_wall}

1055: 

1056: In the two previous sections we considered cases in which the characteristic

1057: longitudinal length scale of the solid boundary was either infinite or large 

1058: compared with

1059: the characteristic amplitude of boundary vibrations. We examine here

1060: the case of a wavy boundary and study how comparable length scales in

1061: both directions influence the flow away from the boundary. In contrast

1062: with the Schlichting problem, the external applied flow is now uniform

1063: or, alternatively, the length scale over which the flow is not uniform

1064: is much larger that the wavelength of the boundary. Thus one

1065: anticipates that the normal component of the flow that appears

1066: is caused by the wall profile. This flow interacts through

1067: nonlinear terms with the externally forced flow that is parallel 

1068: to the average profile of the boundary and gives rise

1069: to stationary streaming. Even for stochastic vibration we show that positive 

1070: and negative vorticity production in 

1071: adjacent regions of the boundary introduces a natural decay length in the

1072: zeroth order solution, thus leading to exponential decay of the flow

1073: away from the boundary, even in the absence of a low frequency cut-off in the

1074: power spectrum of the boundary velocity.

1075: 

1076: Consider a rigid wavy wall being washed by a uniform oscillatory flow parallel 

1077: to the wall wave vector,

1078: \begin{equation}

1079: \mathbf{u}(x,z=\infty)=(u_0(t),0).

1080: \label{rvwb_uinf}

1081: \end{equation}

1082: We now assume that $u_0(t)$ is a narrow band Gaussian process.

1083: Assume also that the amplitude of the boundary profile $l$ is small

1084: compared with both the Stokes layer $\delta_s$ and the

1085: wavelength $L$, with $\delta_{s}/L$ finite. The following dimensionless 

1086: quantities are introduced,

1087: \begin{equation}

1088: \left.

1089: \begin{array}{l}

1090: \displaystyle

1091: z=\tilde{z}[(\nu/\Om)^{1/2}],\quad x=\tilde{x}[(\nu/\Om)^{1/2}],\quad

1092: t=\tilde{t}[\Om^{-1}],\quad\psi=\tilde{\psi}[(2\left<u_0^2\right> 

1093: \nu/\Om )^{1/2}], \\[16pt]

1094: \displaystyle

1095: \eps=l/(\nu/\Om)^{1/2},\quad \Rey=\left[2\left<u_0^2\right>/

1096: \Om\nu\right]^{1/2},

1097: \quad k=2\upi(\nu/\Om)^{1/2}/L,\quad\tilde{\Delta}=\p_{\tilde{z}}^2+

1098: \p_{\tilde{x}}^2

1099: \end{array}

1100: \right\}

1101: \end{equation}

1102: referred to the Cartesian coordinate system sketched in

1103: Fig. \ref{fig:wavy_geom}. The solid boundary is located at

1104: \begin{equation}

1105: \tilde{\eta}(\tilde{x})=\hat{\eta}\epsilon\uexp(\ui k \tilde{x})+\ucc

1106: \label{profile}

1107: \end{equation}

1108: with constant complex amplitude $\hat{\eta}$ so that $|\hat{\eta}|=1/2$. 

1109: The dimensionless (and two dimensional) Navier-Stokes equation 

1110: (tildes are omitted in what follows) reads,

1111: \begin{equation}

1112:   \partial_t{\Delta\psi}+\Rey\frac{\p(\psi,\Delta\psi)}{\p(z,x)}=

1113:   \Delta^2\psi ,

1114: \label{rvwb1}

1115: \end{equation}

1116: with no-slip conditions at the boundary,

1117: \refstepcounter{equation}

1118: $$

1119:   \psi=0, \quad \p_z\psi=0 \quad \mbox{at}\quad y=\eta(x),\qquad

1120:   \eqno{(\theequation{\mathit{a},\mathit{b}})}

1121:   \label{rvwb2}

1122: $$

1123: and the imposed uniform flow at infinity,

1124: \refstepcounter{equation}

1125: $$

1126:   \p_x\psi=0, \quad \p_z\psi=2^{-1/2}u_0(t) \quad \mbox{at}\quad z=\infty.

1127:   \eqno{(\theequation{\mathit{a},\mathit{b}})}

1128:   \label{rvwb3}

1129: $$

1130: These equations depend only on three dimensionless parameters: $\eps$,

1131: the ratio of the amplitude of the wavy wall to the boundary layer

1132: width; $\Rey$, the Reynolds number (the square root of $\Rey_p$ used in 

1133: Section \ref{sec:sch}); and $k$, the wavenumber of the wall profile in units

1134: of boundary layer width. We assume $\eps \ll 1$ and expand the

1135: stream function in a power series of $\eps$,

1136: \begin{equation}

1137:   \psi=\psi_0+\eps\psi_1+\ldots

1138:   \label{rvwb5}

1139: \end{equation}

1140: The boundary conditions are likewise expanded in power series of

1141: $\eps$,

1142: \begin{equation}

1143: \psi(z)|_{z=\eta}=\psi_0(z)|_{z=0}+\eps(\psi_1(z)|_{z=0}+

1144: \eta\p_z\psi_0(z)|_{z=0})+\ldots

1145: \label{rvwb6}

1146: \end{equation}

1147: We now solve (\ref{rvwb1}) order by order in $\eps$. 

1148: 

1149: At zeroth order the wall is effectively planar. We decompose the Fourier

1150: transform of the stream function as,

1151: $ \hat{\psi}_0(x,z,\om)=2^{-1/2}\hat{u}_0(\om)\hat{\zeta}_0(z,\om)$.

1152: The function $\hat{\zeta}_0(z,\om)$ is given in Eq. (\ref{sch81}). At

1153: this order, the solution is identical to that found for a planar boundary.

1154: 

1155: At first order we seek a solution of the form,

1156: \begin{equation}

1157:   \psi_1=\hat{\eta} \; \uexp(\ui kx)\phi(z,t)+\ucc

1158: \label{rvwb7}

1159: \end{equation}

1160: so that the amplitude $\phi(z,t)$ satisfies the Orr-Sommerfeld equation,

1161: \begin{equation}

1162:   (\p_t{\cal D}-{\cal D}^2)\phi=\ui k\Rey (\p_z^3\psi_0-\p_z\psi_0{\cal D})

1163:   \phi,\quad {\cal D}=\p_z^2-k^2

1164: \label{rvwb8}

1165: \end{equation}

1166: with non-homogeneous boundary conditions,

1167: \refstepcounter{equation}

1168: $$

1169:   \phi=0, \quad \p_z\phi=-\p_z^2\psi_0 \quad \mbox{at}\quad z=0,\quad

1170:   \eqno{(\theequation{\mathit{a},\mathit{b}})}

1171:   \label{rvwb9}

1172: $$

1173: \refstepcounter{equation}

1174: $$

1175:   \phi=0, \quad \p_z\phi=0 \quad \mbox{at}\quad z=\infty,\quad

1176:   \eqno{(\theequation{\mathit{a},\mathit{b}})}

1177:   \label{rvwb10}

1178: $$

1179: 

1180: The linear operator in the left hand side of Eq. (\ref{rvwb8})

1181: contains a significant difference with respect to that of Eq. 

1182: (\ref{sch9}), the equation governing the first order stream function

1183: for the case of a slightly curved boundary. Both equations describe

1184: vorticity diffusion, but the biharmonic equation (\ref{rvwb8}) contains

1185: a cut-off through the parameter $k$. Is is precisely this term that

1186: will lead to an asymptotic exponential decay of the velocity field

1187: sufficiently far away from the boundary for any finite $\beta$. The 

1188: exponential decay at long distances arises from the screening

1189: introduced by the simultaneous positive and negative vorticity

1190: produced at the troughs and crests of the wavy wall.

1191: 

1192: In order to obtain a solution of the Orr-Sommerfeld equation (Eq. 

1193: (\ref{rvwb8})),

1194: we further expand the amplitude $\phi(z,t)$ in power series of 

1195: $k\Rey=2\upi\left(2 \left<u_0^2\right>\right)^{1/2}/L\Om$. This is the

1196: ratio between the amplitude of oscillation of the flow at infinity and 

1197: the wall wavelength. We write,

1198: \begin{equation}

1199: \phi=\phi_0+\ui k \Rey\phi_1+\ldots

1200: \label{rvwb11}

1201: \end{equation}

1202: The function $\phi_0$ obeys the linearized Eq. (\ref{rvwb8}) with  

1203: boundary conditions as in Eqs. (\ref{rvwb9}-\ref{rvwb10}) with

1204: $\phi$ replaced by $\phi_{0}$. The Fourier transform of $\phi_{0}$ is

1205: given by,

1206: \begin{equation}

1207: \hat{\phi}_0(z,\om)=(2)^{-1/2}\hat{u}_0(\om)

1208: \frac{\alpha}{\rho-k}\left(\ue^{-\rho z}-\ue^{-kz}\right),

1209: \label{rvwb_phi01}

1210: \end{equation}

1211: with $\rho\equiv(\alpha^2+k^2)^{1/2}$, and the principal branch of the 

1212: square root is assumed $(\Re\{\rho\}>0)$. Recall that 

1213: $\alpha = ( 1 + \ui{\rm sign}(\om))(\om/2\nu)^{1/2}$. The field

1214: $\phi_0$ describes

1215: vorticity diffusion near the wavy wall caused by the uniform but

1216: oscillatory far field flow. Both the spatial and ensemble averages of 

1217: this flow are zero. However, the flow non-uniformity at this order

1218: induces mean flow at the next order, as it is readily apparent from the 

1219: equation for $\phi_1$,

1220: \begin{equation}

1221:   (\p_t{\cal D}-{\cal D}^2)\phi_1=(\p_z^3\psi_0-\p_z\psi_0{\cal D})\phi_0,

1222: \label{rvwb_phi1}

1223: \end{equation}

1224: with boundary conditions $\phi_1=0, \; \p_z\phi_1=0$ at $z=0,\infty$.

1225: The field $\phi_1$ describes vorticity diffusion forced by the 

1226: nonlinear interaction between $\phi_0$ and $\psi_0$. As was the case 

1227: in Sec. \ref{sec:sch}, we focus on the long time limit of the ensemble

1228: average of Eq. (\ref{rvwb_phi1}) $\phi_1^{(s)} = \lim_{t \rightarrow 

1229: \infty} \left<\phi_1\right> = \chi + \ucc$, where $\chi$ is given by, 

1230: \begin{equation}

1231:   {\cal D}^2\chi=-\frac{1}{2}G(z,\beta),

1232: \label{rvwb12}

1233: \end{equation}

1234: with,

1235: \begin{eqnarray}

1236:   G(z,\beta)=\int_{0}^{\infty}\ud\om\,P(\om,\beta)\,Q(z,\om,\beta),\nonumber\\

1237:   Q(z,\om,\beta)=\alpha(\rho+k)(2\ue^{-(\alpha^*+\rho)z}- 

1238:   \ue^{-(\alpha^*+k)z}-\ue^{-\rho z}).

1239: \label{rvwb13}

1240: \end{eqnarray}

1241: The corresponding boundary conditions are homogeneous,

1242: $ \chi=0, \; \p_z\chi=0$ at $z=0,\infty$. The solution is,

1243: \begin{eqnarray}

1244: \chi(z,\beta) & = &\int_{0}^{\infty}\ud\om\,P(\om,\beta)\hat{\chi}(z,\om),

1245: \nonumber\\

1246: \hat{\chi}(z,\om) & = &

1247: A_1\ue^{-\rho z}+A_2\ue^{-(\alpha^{*}+\rho)z}+A_3\ue^{-(\alpha^*+k)z}+

1248: (B_1+z B_2)\ue^{-kz}

1249: \label{rvwb_chi}

1250: \end {eqnarray}

1251: where the functions $A_i,\,B_i$ depend on frequency and wavenumber,

1252: \begin{eqnarray}

1253: A_1=D/\alpha^4,\quad A_2=D/(2\alpha^2\rho^2),\quad 

1254: A_3=-D/(\alpha^2(\alpha^*+2k)^2,\quad D=\alpha(\rho+k)/2,\qquad\nonumber\\

1255: B_1=-(A_1+A_2+A_3),\quad B_2=(\rho-k)A_1+(\rho+\alpha^*-k)A_2 + 

1256: \alpha^*A_3.\qquad

1257: \label{rvwb_chi_const}

1258: \end{eqnarray}

1259: Therefore the stationary part of the averaged first order stream

1260: function is given by,

1261: \begin{equation}

1262: \psi_1^{(s)}=\ui k \Rey\hat{\eta}\uexp(\ui kx)(\chi+\chi^*)+\ucc

1263: \label{rvwb17}

1264: \end{equation}

1265: This solution shows that $\psi_1^{(s)}$

1266: has a phase advance of $\upi/2$ with respect to the wall profile, and hence 

1267: the flow in the vicinity of the boundary is directed from trough to crest 

1268: ($\p_z^2(\chi+\chi^*)|_{z=0} > 0 )$.

1269: By shifting the coordinate system along the $x$ axis we can change the

1270: phase of the wall profile so as to make it a simple cosine function.

1271: We consider $\eta(x)=\eps\ucos(kx)$ in what follows.

1272: 

1273: Following \cite{re:lyne71}, we now proceed to study the limits of

1274: $k$ large and small, while $k\Rey \ll 1$. For $k \gg 1$ the wavelength of the 

1275: boundary profile is much smaller that 

1276: the thickness of the viscous layer. In this case, a boundary layer 

1277: appears of characteristic thickness $1/k$. Screening between regions

1278: producing positive and negative vorticity occurs over a distance

1279: much smaller that the Stokes thickness based on the frequency of

1280: oscillation. The net vorticity does not diffuse even to

1281: distances of order $z\sim\cal {O}$(1), hence

1282: giving rise to exponential decay with an $\uexp(-kz)$

1283: factor. The region in which the stream function is not exponentially

1284: small depends on $Z = kz$. The explicit form of $\psi_1^{(s)}$ may be 

1285: obtained by direct expansion of the solution (\ref{rvwb17}) in power 

1286: series of $1/k$, keeping $Z$ fixed. 

1287: The leading contribution to the steady part of 

1288: the tangential component of the velocity is given by,

1289: \begin{equation}

1290: u_1^{(s)}=k\p_Z\psi_1^{(s)}\sim-\frac{\Rey}{24k^2}\usin(kx)\,\ue^{-Z}Z(6-Z^2)

1291: \label{wavy_klarge}

1292: \end{equation}

1293: The  boundary layer comprises two recirculating cells per wall 

1294: period, located  within $0<z\lesssim1/k$. The separation point is given 

1295: by $z^2=6/k^2$.

1296: 

1297: In the opposite limit of $k \ll 1$ one formally recovers the Schlichting 

1298: problem in that the characteristic longitudinal length scale is much 

1299: larger than the Stokes layer thickness. There is one fundamental

1300: difference, however, which can be seen from the solution,

1301: Eq. (\ref{rvwb_chi}). It has two contributions: the first one is

1302: proportional $A_i$, arises from the particular solution,and serves to

1303: balance the non-homogeneity in Eq. (\ref{rvwb12}). This contribution

1304: decays within the Stokes layer. The second one is proportional to

1305: $B_i$, and arises from the general solution of the homogeneous part of 

1306: the equation. This contribution decays over the stretched scale

1307: $Z$. It turns out that this second

1308: contribution introduces an additional separation of the composite

1309: boundary layer when $0<k\leq0.23$. The lines of zero value of the

1310: longitudinal component of the steady velocity profile

1311: are shown Fig. \ref{fig:wavy_uzero}. The location of the second

1312: separation point is entirely determined by that part of the solution

1313: that is proportional to $B_{i}$, and occurs at $z \sim {\cal{O}} (1/k) \gg 1$.

1314: 

1315: The steady velocity may be obtained by expanding the 

1316: exact solution (Eq. (\ref{rvwb_chi})) in power series of $k$, first

1317: keeping $z$ fixed (inner solution, $u_{1i}^{(s)}$),

1318: and second keeping $Z$ fixed (outer solution $u_{1o}^{(s)}$).

1319: To leading order, we find,

1320: \begin{eqnarray}

1321: u_{1i}^{(s)}(z') & \sim &  -\Rey k^2\usin(kx)

1322:   \left\{ \big[\frac{z'}{2}(\usin(z')-\ucos(z'))+2\usin(z')+\frac{1}{2}

1323: \ucos(z')\big]\ue^{-z'}

1324:   \right. \nonumber\\

1325: && \left.\mbox{}\hspace{70pt}+  

1326:    \frac{\ue^{-2z'}}{4} -\frac{3}{4}\astrut\right\},\quad 

1327: z' = z/2^{1/2}\sim {\cal{O}}(1),\\

1328: u_{1o}^{(s)}(Z) & \sim & -\Rey k^2\usin(kx)\big[\frac{3}{4}(Z-1)\ue^{-Z}\big],

1329: \qquad Z\sim {\cal{O}}(1) .

1330: \label{wavy_io_sols}

1331: \end{eqnarray}

1332: The solution for the inner and outer steady velocities was already obtained by

1333: \cite{re:lyne71} by a conformal transformation technique. We further note that

1334: the inner and outer solutions can now be matched by requiring that

1335: $u_{1i}^{(s)}(\infty)=u_{1o}^{(s)}(0)=3/4\Rey k^2\usin(kx)$. Hence it

1336: is possible to construct a uniformly valid solution by adding the

1337: inner and outer solutions, and subtracting the first term of the

1338: inner expansion of the outer solution. We find, 

1339: \begin{eqnarray}

1340: u_{1c}^{(s)}(z') & \sim &  -\Rey k^2\usin(kx)

1341:   \left\{ \big[\frac{z'}{2}(\usin(z')-\ucos(z'))+2 \usin(z') +

1342:   \frac{1}{2}

1343:   \ucos(z')\big]\ue^{-z'} \right. \nonumber\\

1344: && \left.\mbox{} \hspace{70pt}+  

1345:    \frac{\ue^{-2z'}}{4} + \frac{3}{4}(2^{1/2}kz'-1)\ue^{-2^{1/2}kz'}

1346:   \astrut \right\}

1347: \label{wavy_csol}

1348: \end{eqnarray}

1349: 

1350: We now turn to a numerical study of the case of finite $\beta$.

1351: The boundary value problem (\ref{rvwb12}) has been solved numerically

1352: by using a multiple shooting method for non stiff and linear boundary 

1353: value problems (\cite{re:mattheij84}). The method has the advantage

1354: that the necessary intermediate shooting points are determined by the method itself,

1355: and that it can give the solution on a preset and nonuniform grid of points.

1356: The code was tested on the analytically known solution of the deterministic

1357: limit, Eqs. (\ref{rvwb_chi}) and (\ref{rvwb_chi_const}). Our results are

1358: summarized in Figs. \ref{fig:wavy_uzero}, \ref{fig:wavy_ks} and \ref{fig:wavy_kl}.

1359: 

1360: Figure \ref{fig:wavy_uzero} shows the separation points of the stationary

1361: velocity as a function of the boundary wavenumber $k$ for a range of values of

1362: $\beta$, including for reference the deterministic limit of $\beta = \infty$

1363: (separation points are defined to be the zeros of the stationary tangential velocity).

1364: This figure shows that for fixed $k$, the mean stationary velocity field may 

1365: be comprised

1366: of two or four recirculating cells per wall period depending on $\beta$. 

1367: The first separation point is largely independent of $\beta$, whereas

1368: the deviation of the second relative to its value in the monochromatic

1369: limit is proportional to $\beta/\upi(1+\beta^2)$, the value of $P(0,\beta)$.  

1370: 

1371: Our results in the limit $k\ll1$ are presented in Fig. \ref{fig:wavy_ks}, where

1372: profiles of tangential component of the mean stationary velocity 

1373: are plotted for $k=0.1$ and different values of $\beta$. At large

1374: $\beta$ (close to the monochromatic limit) the flow is comprised of four

1375: recirculating cells per boundary period. Upon decresing $\beta$, the

1376: second separation point moves to infinity (see also Fig.

1377: \ref{fig:wavy_uzero}), so that beyond some critical value of $\beta$,

1378: only two recirculating cells remain. Further decrease in $\beta$ results

1379: in the reappearance of the second separation point at infinity, which

1380: then continues to move towards decreasing $z$. The intensity of the

1381: recirculating modes does not change monotonically with $\beta$ as we

1382: further discuss below. In the opposite limit of $k \gg 1$ 

1383: (Fig. \ref{fig:wavy_kl} shows the case $k=10.0$; note that

1384: $u_1^{(s)}$ is now normalized by $\Rey/k^2$), the qualitative structure of 

1385: the flow is largely independent of $\beta$. The streaming flow has

1386: two recirculating cells per wall wavelength, and their intensity increases

1387: monotonically with decreasing $\beta$.

1388: 

1389: The complex dependence of the flow on $\beta$ and $k$ can be

1390: qualitatively understood from the interplay between the width 

1391: of the power spectrum (given by $1/\beta$), the viscous damping of each 

1392: elementary excitation that depends on its frequency, and the penetration 

1393: depth of the flow field which is primarily dictated by the boundary wavelength. 

1394: For small $k$, large frequency modes are damped close to the boundary and

1395: do not penetrate much into the recirculating layers. Reducing $\beta$

1396: introduces high frequency components into the driving terms at first

1397: order, but they are dynamically damped.

1398: At the same time, the power in the dominant frequency components (around

1399: $\Omega$) decreases. Overall, a decrease in $\beta$ then leads to a decrease in

1400: recirculation strength. As $k$ increases, larger

1401: frequencies contribute to the flow over the entire range of the

1402: recirculating cells. Decreasing $\beta$ decreases the strength of the

1403: dominant components, but increases the range of high frequencies that

1404: can contribute to the flow. From Eq. (\ref{rvwb12}) one can show that the

1405: driving contribution from higher frequencies which is contained in $Q$ 

1406: increases faster

1407: with frequency than the decreasing weight given to them by the power

1408: spectrum $P(\omega,\beta)$. Consequently, decreasing $\beta$ (which

1409: amounts to moving towards the white noise limit) leads to increasing

1410: amplitude of the recirculation.

1411: 

1412: In summary, for any value of $\beta$, finite or infinite, the vorticity 

1413: produced by vibration of the wavy boundary does not penetrate into the bulk

1414: farther than a distance of order of the wavelength of the boundary. 

1415: However, there are qualitative differences with the deterministic

1416: limit in the character of the flow within that layer. In particular,

1417: the structure and the intensity of the stationary secondary flow

1418: strongly depend on $\beta$.

1419: 

1420: \section{Summary}

1421: 

1422: We have addressed the flow induced by a randomly vibrating solid

1423: boundary in an otherwise quiescent fluid. This analysis has been 

1424: motivated by the random residual acceleration field in which microgravity

1425: experiments are conducted. The salient features of the flow are

1426: summarized below.

1427: 

1428: When the solid boundary is planar, the flow field averages to zero

1429: (the average velocity of the boundary has been taken to be zero in all

1430: cases investigated), but its variance decays algebraically with

1431: distance away from the wall. This dependence follows from a

1432: non vanishing power spectrum of the boundary velocity at zero

1433: frequency. Introducing a low frequency cut-off in the power spectrum

1434: leads back to the classical exponential decay, with a rate that is

1435: determined by the cut-off frequency, Eq. (\ref{plane4_cut}). 

1436: The amplitude of the decaying variance depends explicitly on the 

1437: correlation time of the boundary velocity, $\beta = \Omega \tau$,

1438: where $\Omega$ is the dominant angular frequency of the power spectrum

1439: of the boundary velocity, and $\tau$ is inverse spectral width ($\tau$

1440: is the correlation time of the boundary velocity).

1441: 

1442: If the solid boundary is curved, steady streaming is generated in

1443: analogy with the classical analysis of Schlichting. The stationary part

1444: of the ensemble average of the secondary velocity is nonzero, even though

1445: the boundary velocity averages to zero. In this case, we find that the 

1446: leading contribution to the average stationary velocity diverges 

1447: logarithmically with distance away from the boundary. 

1448: In analogy to the planar case, the introduction of a low frequency cut-off

1449: in the power spectrum of the boundary velocity changes the asymptotic behavior

1450: qualitatively. The average stationary velocity asymptotes now to a constant, 

1451: given by Eq. (\ref{sch_cut_asym2}). The asymptotic velocity explicitly 

1452: depends on $\beta$ and logarithmically on the cut-off frequency. This asymptotic

1453: behavior is not reached until a length scale of the order of the Stokes layer 

1454: thickness that is based on the cut-off frequency.

1455: 

1456: We have finally analyzed the case of a periodically modulated solid

1457: boundary in the limit in which the scale of the wall modulation is

1458: small compared to the thickness of the Stokes layer, and also when the

1459: spatial amplitude of the boundary oscillation is small compared with

1460: the wavelength of the wall profile. Cancellation of vorticity

1461: production over the wall boundary leads to exponential decay of the

1462: fluid velocity away from the boundary, with a decay length which is

1463: proportional to the wall wavelength, even if the zero

1464: frequency value of the power spectrum of the boundary velocity is

1465: nonzero. If the boundary wavelength is much larger than the Stokes

1466: layer thickness, we find steady streaming in secondary flow with two or

1467: four recirculating cells per wall period depending on $\beta$.

1468: On the other hand, if the wavelength is

1469: much smaller than the Stokes layer thickness, only two recirculating

1470: cells are formed regardless of the value of $\beta$. Somewhat unexpectedly,

1471: the intensity of the recirculation can both increase or decrease with $\beta$.

1472: 

1473: \begin{acknowledgments}

1474: This research has been supported by the Microgravity Science and Applications

1475: Division of the NASA under contract No. NAG3-1885, and also in part

1476: by the Supercomputer Computations Research Institute, which is

1477: partially funded by the U.S. Department of Energy, contract No.

1478: DE-FC05-85ER25000.

1479: \end{acknowledgments}

1480: 

1481: \newpage

1482: \bibliographystyle{$HOME/tex/jfm}

1483: \bibliography{dv1}

1484: 

1485: \newpage

1486: \begin{figure}

1487:   \vspace{1.5pc}

1488:   \centerline{\epsffile{plane_var.eps}}

1489:   \caption{Normalized variance of the tangential velocity for the case

1490: of a planar boundary computed by numerical integration of Eq. (\ref{plane3})

1491: (symbols), and its uniform asymptotic expansion, Eq. (\ref{plane5}), 

1492: (solid lines). The function $I(z,\beta)\,z^2$ asymptotes to a constant

1493: value outside of the classical Sokes layer based on $\Om$. The uniform expansion remains 

1494: a good approximation even for moderate $\beta$.}

1495:   \label{fig:plane_var}

1496: \end{figure}

1497: %\vskip 5cm

1498: \newpage

1499: \newpage

1500: %

1501: \begin{figure}

1502:   \vspace{1.5pc}

1503:   \centerline{\epsffile{sc_u_prof.eps}}

1504: \caption{Stationary first order velocity as a function of distance for a

1505: range of values of $\beta$. All curves diverge logarithmically at large $z$, 

1506: except for $\beta=\infty$ (monochromatic limit), in which the velocity 

1507: asymptotes to a constant within the Stokes layer. This latter behavior

1508: reproduces the classical result of Schlichting.}

1509:   \label{fig:sc_u_prof}

1510: \end{figure}

1511: %\vskip 5cm

1512: \newpage

1513: %

1514: \begin{figure}

1515:   \vspace{1.5pc}

1516:   \centerline{\epsffile{sc_cut_prof.eps}}

1517: \caption{Stationary first order velocity as a function of distance for a

1518: range of values of $\beta$. The power spectrum of the boundary velocity

1519: has a low frequency cut-off at $\om_c=0.05$. The velocity asymptotes to

1520: a constant that depends on the value of $\beta$.}

1521:   \label{fig:usc_cut_prof}

1522: \end{figure}

1523: %\vskip 5cm

1524: \newpage

1525: %

1526: \begin{figure}

1527:   \vspace{1.5pc}

1528:   \centerline{\epsffile{sc_cut.eps}}

1529: \caption{Asymptotic dependence of the stationary velocity as a function of 

1530: the cut-off frequency $\om_c$. We show the case $\beta=10$ given by 

1531: Eq. (\ref{sch_cut_asym2}) along with the numerically obtained solution.}

1532: \label{fig:usc_cut}

1533: \end{figure}

1534: %\vskip 5cm

1535: \newpage

1536: %

1537: \begin{figure}

1538:   \vspace{1.5pc}

1539:   \centerline{\epsffile{wavy_geom.eps}}

1540:   \caption{Schematic view of the geometry of the wavy wall studied

1541:   		in Section \ref{sec:wavy_wall}.}

1542:   \label{fig:wavy_geom}

1543: \end{figure}

1544: %\vskip 5cm

1545: \newpage

1546: %

1547: \begin{figure}

1548:   \vspace{1.5pc}

1549:   \centerline{\epsffile{wavy_uzero.eps}}

1550: \caption{Separation points of the boundary layer over the wavy boundary

1551: as a function of its wavenumber $k$, for different values of $\beta$.

1552: The separation points are the loci of zero tangential velocity. At fixed

1553: $k$, the location of the second separation point depends strongly on $\beta$.}

1554:   \label{fig:wavy_uzero}

1555: \end{figure}

1556: \newpage

1557: %

1558: \begin{figure}

1559:   \vspace{1.5pc}

1560:   \centerline{\epsffile{wavy_ks.eps}}

1561: \caption{Tangential component of the mean stationary velocity as a function

1562: of $z$ for $k=0.1$ and a range of values of $\beta$. The case $\beta=\infty$

1563: corresponds to analytic solution obtained by Lyne. The other

1564: curves are the numerical solutions of the boundary value problem

1565: defined by Eq. (\ref{rvwb12}) and corresponding boundary conditions.}

1566: \label{fig:wavy_ks}

1567: \end{figure}

1568: \newpage

1569: %

1570: \begin{figure}

1571:   \vspace{1.5pc}

1572:   \centerline{\epsffile{wavy_kl.eps}}

1573: \caption{Tangential component of mean stationary velocity as a function

1574: of $z$ for $k=10$ and a range of values of $\beta$. The case $\beta=\infty$

1575: corresponds to analytic solution obtained by Lyne. The other

1576: curves are the numerical solutions of the boundary value problem

1577: defined by Eq. (\ref{rvwb12}) and corresponding boundary conditions.}

1578: \label{fig:wavy_kl}

1579: \end{figure}

1580: 

1581: \end{document}

1582: 

1583: