1: %% p2b.tex resubmitted in response to round 1 of refs comments Sept 2006

2: %% p2.tex is version submitted to PRL 26th June 2006

3: %% p3.tex is version submitted to PRE March 2007

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

5: \documentclass[twocolumn,showpacs,amsmath,amssymb,bibnotes]{revtex4}

6: %\documentclass[preprint,showpacs,amsmath,amssymb,bibnotes]{revtex4}

7: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

8: \usepackage{bm}

9: \usepackage{graphicx,epsfig,subfigure,afterpage}

10: \usepackage{amsfonts}

11: %\usepackage{mcite}

12: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

13: 

14: \newcommand{\be}{\begin{equation}}

15: \newcommand{\ee}{\end{equation}}

16: \newcommand{\bea}{\begin{eqnarray}}

17: \newcommand{\eea}{\end{eqnarray}}

18: \newcommand{\gdot}{\dot{\gamma}}

19: \newcommand{\gdotbar}{\overline{\dot{\gamma}}}

20: \newcommand{\ie}{{\it i.e.\/}}

21: \newcommand{\eg}{{\it e.g.\/}}

22: \newcommand{\etc}{{\it etc.\/}}

23: \newcommand{\versus}{{\it vs.\/}}

24: \newcommand{\etal}{{\it et al.\/}}

25: \newcommand{\bw}{\begin{widetext}}

26: \newcommand{\ew}{\end{widetext}}

27: \newcommand{\lae}{\stackrel{<}{\scriptstyle\sim}}

28: \newcommand{\gae}{\stackrel{>}{\scriptstyle\sim}}

29: 

30: \newcommand{\tmax}{t_{\rm max}}

31: \newcommand{\Tband}{T_{\rm b}}

32: \newcommand{\Nbase}{N_{\rm base}}

33: 

34: \newcommand{\ommax}{\omega_{\rm max}}

35: 

36: \newcommand{\xhat}{\vecv{\hat{x}}}

37: \newcommand{\yhat}{\vecv{\hat{y}}}

38: \newcommand{\zhat}{\vecv{\hat{z}}}

39: 

40: \newcommand{\vecv}[1]{\mathbf{{#1}}}

41: \newcommand{\tens}[1]{\mathbf{{#1}}}

42: \newcommand{\nablu}{{\bf \nabla}}

43: 

44: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

45: \begin{document}

46: 

47: \title{Vorticity structuring and Taylor-like velocity rolls triggered by gradient shear bands}

48: \author{Suzanne M. Fielding}

49: \email{suzanne.fielding@manchester.ac.uk}

50: \affiliation{School of Mathematics, University of Manchester, Booth

51:   Street East, Manchester M13 9EP, United Kingdom }

52: 

53: \date{\today}

54: \begin{abstract}

55: 

56: We suggest a novel mechanism by which vorticity structuring and

57: Taylor-like velocity rolls can form in complex fluids, triggered by

58: the linear instability of one dimensional gradient shear banded

59: flow. We support this with a numerical study of the diffusive Johnson-Segalman

60: model. In the steady vorticity structured state, the thickness of the

61: interface between the bands remains finite in the limit of zero stress

62: diffusivity, presenting a possible challenge to the accepted theory of

63: shear banding.

64: 

65: 

66: \end{abstract}

67: \pacs{{47.50.+d}, %{ Non-Newtonian fluid flows}--

68:      {47.20.-k}, %{ Hydrodynamic stability}--

69:      {36.20.-r}.%{ Macromolecules and polymer molecules}

70:      } 

71: \maketitle

72: 

73: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

74: 

75: \section{Introduction}

76: \label{sec:intro}

77: 

78: Many complex fluids exhibit flow instabilities that result in

79: spatially heterogeneous, shear banded states.  Examples include

80: wormlike~\cite{BritCall97c,berret-pre-55-1668-1997,BecManCol04} and

81: onion~\cite{diat93,wilkinsOlms2006,SalManCol03} surfactants;

82: side-chain liquid crystalline polymers~\cite{pujolle01}; viral

83: suspensions~\cite{lettinga-jpm-16-S3929-2004,dhont-fd-123-157-2003};

84: telechelic polymers~\cite{berret-prl-8704--2001}; soft

85: glasses~\cite{CRBMGHJL02}; polymer solutions~\cite{HilVla02}; and

86: colloidal suspensions~\cite{ChenZAHSBG92}. In many cases, the

87: instability is explained by a region of negative slope

88: $dT_{xy}/d\gdot<0$ in the constitutive relation $T_{xy}(\gdot)$

89: between shear stress and shear rate for homogeneous flow, as shown in

90: Fig~\ref{fig:schematic}a. In this regime, homogeneous flow is unstable

91: ~\cite{Yerushalmi70} with respect to the formation of bands of

92: differing shear rates $\gdot_1$ and $\gdot_2$, with layer-normals in

93: the flow-gradient direction.  Force balance requires that the shear

94: stress $T_{xy}$ is uniform across the gap, and therefore common to

95: both the bands. Any change in the overall applied shear rate

96: $\gdotbar$ causes a change in the relative volume fraction $f$ of the

97: bands according to a lever rule $\gdotbar=f\gdot_1+(1-f)\gdot_2$,

98: while $\gdot_1,\gdot_2$ and $T_{xy}$ remain constant. In bulk

99: rheometry, this leads to a plateau in the steady state flow curve at

100: some stress $T_{xy}=T^*$ (Fig.~\ref{fig:schematic}a), the value of

101: which is selected by accounting for spatial non-locality in the

102: constitutive dynamics of the viscoelastic

103: stress~\cite{lu-prl-84-642-2000}.  In what follows, we shall refer to

104: the effect just described as gradient shear banding, or simply

105: gradient banding.

106: 

107: In some systems, flow induced heterogeneity has been reported in the

108: vorticity direction. By analogy with the above discussion, this effect

109: is often termed vorticity banding. It has been observed in onion

110: surfactants~\cite{diat93,Bonn+98} and colloidal

111: crystals~\cite{ChenZAHSBG92}, accompanied by a steep stress ``cliff''

112: in the flow curve. Multiple turbid and clear vorticity bands also

113: occur in some polymeric~\cite{HilVla02} and

114: micellar~\cite{fischer-ra-41-35-2002,fischer-ra-39-234-2000}

115: solutions, accompanied by shear thickening. In these cases, the bands

116: not only alternate in space but also oscillate in time. In shear

117: thinning viral suspensions, multiple stationary vorticity bands can

118: arise in the regime of isotropic-nematic microphase

119: separation~\cite{lettinga-jpm-16-S3929-2004,dhont-fd-123-157-2003}.

120: 

121: In comparison with gradient banding, vorticity banding is poorly

122: understood theoretically.  To date, the main attempts to model it have

123: invoked the analogy with gradient

124: banding~\cite{goveas-epje-6-79-2001}. As discussed above, gradient

125: bands have different shear rates $\gdot_1,\gdot_2$, and coexist at a

126: common shear stress $T^*$. For vorticity bands, the moving rotor

127: imposes a shear rate that is common to each band. Pursuing the analogy

128: with gradient banding, it is natural to invoke an underlying

129: constitutive curve of the form in Fig.~\ref{fig:schematic}b, in the

130: case of shear thickening systems.  This allows bands (A and B) of

131: differing shear {\em stresses} to coexist, having layer normals in the

132: vorticity direction.  In steady state, one then expects a flow curve

133: with a steep stress cliff, consistent with the experimental

134: observations discussed above~\cite{diat93,Bonn+98,ChenZAHSBG92}. This

135: scenario can be adapted to shear thinning systems by invoking a

136: constitutive curve of the form sketched in Fig.~\ref{fig:schematic}c.

137: In fact, at the level of 1D calculations performed separately in the

138: flow-gradient and vorticity directions, such a curve can support

139: either gradient or vorticity banded

140: states~\cite{olmsted-pre-60-4397-1999}.  Which of these (if either)

141: would be selected in a full 3D calculation remains an outstanding

142: question.  Indeed, any concrete calculation of vorticity banding to

143: date has taken a simplified one-dimensional (1D) approach, allowing

144: spatial variations only in the direction of the layer normals, and

145: thereby imposing axial and radial symmetry.

146: 

147: Beyond the traditional shear banding literature, other flow

148: instabilities are well known to trigger structuring in the vorticity

149: direction, in both simple and complex fluids.  For simple liquids in

150: curved Couette flow, the unstable centripetal interplay of fluid

151: inertia with cell curvature gives rise to Taylor velocity

152: rolls~\cite{drazin} stacked in the vorticity direction. An analogous

153: inertia-free instability occurs in viscoelastic fluids, here triggered

154: by viscoelastic hoop stresses~\cite{larson-jfm-218-573-1990,Lars92c},

155: and again leading to velocity rolls stacked along the axis of the

156: Couette cylinder.  The rolls are typically observed as bandlike

157: structures, imaged by seeding the fluid with mica flakes.

158: 

159: In contrast to the 1D vorticity banding scenario discussed above, both

160: traditional and viscoelastic Taylor-Couette instabilities are

161: inherently 2D (at least), with velocity rolls comprising a circulation

162: of fluid in the flow-gradient/vorticity

163: plane~\cite{drazin,larson-jfm-218-573-1990}.  Not unexpectedly, given

164: this roll-like structure, the wavelength of the associated banding in

165: the vorticity direction is roughly set by the width of the gap in the

166: flow-gradient direction~\cite{drazin}.

167: 

168: %

169: \begin{figure}[tb]

170: \subfigure{

171:   \includegraphics[scale=0.40]{./flowCurvea0.3.eps}

172: }

173: \hspace{-0.3cm}

174: \subfigure{

175:   \includegraphics[scale=0.35]{./schematic.eps}

176: }

177: \caption{a) Homogeneous constitutive curve and steady state flow curve for 1D planar gradient banded flow in the  DJS model at $a=0.3, \eta=0.05$. Inset: schematic arrangement of the bands in curved Couette flow. b) Schematic constitutive curve giving shear thickening vorticity banding. c) A shear thinning curve allowing gradient or vorticity banding in 1D. }

178: \label{fig:schematic}

179: \vspace{-0.5cm}

180: \end{figure}

181: %

182: 

183: In view of the above discussion, it is natural to ask whether any link

184: exists between traditional (1D) vorticity banding and 2D viscoelastic

185: Taylor-Couette instabilities; or whether the two effects are entirely

186: distinct. Given that both can be accompanied by shear thickening in

187: the bulk flow curve, this is a difficult question to address

188: experimentally. Even if they are distinct in theory, it seems feasible

189: that some experimental observations that have traditionally been

190: interpreted as 1D vorticity banding in fact comprise 2D viscoelastic

191: Taylor Couette rolls. Likely candidates include those systems in which

192: the wavelength of the alternating vorticity bands is comparable to the

193: width of the cell in the flow-gradient direction, suggesting an

194: underlying roll

195: structure~\cite{HilVla02,fischer-ra-41-35-2002,fischer-ra-39-234-2000,lettinga-jpm-16-S3929-2004,dhont-fd-123-157-2003}.

196: This was recently suggested in the context of viral suspensions in

197: Ref.~\cite{kang-pre-74--2006}. In other systems, particularly those

198: showing a very marked stress cliff in flow curve, the traditional 1D

199: scenario of Fig.~\ref{fig:schematic}b or c remains more likely.

200: 

201: In this paper, we suggest a novel mechanism by which vorticity

202: structuring can emerge in complex fluids. A key feature of our

203: approach is that, to some extent, it unifies traditional 1D (gradient)

204: banding descriptions with those of 2D roll-like instabilities. The

205: basic idea is as follows. The constitutive curve of

206: Fig.~\ref{fig:schematic}a gives rise initially to 1D gradient bands,

207: via the conventional instability in the region of negative slope,

208: $dT_{xy}/d\gdot<0$. These then undergo a secondary linear

209: instability~\cite{fielding-prl-95--2005,wilson-jnfm-138-181-2006}, due to the

210: action of normal stresses across the interface between the bands. This

211: leads finally to pronounced undulations along the interface, with

212: wavevector in the vorticity direction. These are accompanied by 2D

213: Taylor-like velocity rolls stacked in the vorticity direction, and

214: undulatory vorticity stress structuring superposed on the underlying

215: gradient bands.  In contrast to the conventional

216: inertial~\cite{drazin} and viscoelastic~\cite{larson-jfm-218-573-1990}

217: Taylor mechanisms, the vorticity instability introduced here does not

218: rely on cell curvature, but occurs even in the limit of planar shear,

219: to which our calculations are confined for simplicity.

220: 

221: The results to be presented are in good agreement with recent

222: experiments in which a gradient banded solution of wormlike micelles

223: was found to be unstable with respect to interfacial undulations with

224: wavevector in the vorticity direction~\cite{lerouge-prl-96--2006}. We

225: will return to a detailed comparison with these experiments later in

226: the manuscript.  Our results might also apply to systems in which

227: shear thickening~\cite{decruppe-pre-73--2006} and/or vorticity

228: banding~\cite{fischer-ra-41-35-2002,fischer-ra-39-234-2000,HilVla02,Bonn+98}

229: is reported to set in at the right hand edge of a stress plateau in

230: the flow curve (suggestive of underlying gradient banding, as in

231: Fig.~\ref{fig:schematic}a); or in which gradient and vorticity banding

232: have actually been observed

233: concomitantly~\cite{britton-prl-78-4930-1997}.

234: 

235: The paper is structured as follows. In Sec.~\ref{sec:model} we

236: introduce the diffusive Johnson Segalman model, within which the

237: subsequent calculations are to be performed. In Sec.~\ref{sec:1D} we

238: discuss 1D calculations, confined to the flow-gradient direction.

239: These predict gradient banding for applied shear rates in the regime

240: of negative slope in the homogeneous constitutive curve. In

241: Sec.~\ref{sec:linear} we switch to two dimensions -- the

242: flow-gradient/vorticity plane -- and show an initially 1D gradient

243: banded ``base state'' to be linearly unstable with respect to

244: undulations along the interface with wavevector in the vorticity

245: direction. We then perform a full 2D nonlinear numerical study of the

246: subsequent growth and eventual saturation of these undulations.

247: Details of the numerical method are discussed in

248: Sec.~\ref{sec:numerics}, followed by presentation of the results in

249: Sec.~\ref{sec:nonlinear}. Finally we summarise our findings and

250: discuss some directions for future study.

251: 

252: \section{Model and geometry}

253: \label{sec:model}

254: 

255: The generalised Navier--Stokes equation for a viscoelastic material in

256: a Newtonian solvent of viscosity $\eta$ and density $\rho$ is

257: %

258: \begin{equation} \label{eqn:NS} \rho(\partial_t +

259:   \vecv{v}.\nablu)\vecv{v} = \nablu .(\tens{\Sigma} + 2\eta\vecv{D}

260:   -P\tens{I}),

261: \end{equation} 

262: %

263: where $\vecv{v}(\vecv{r},t)$ is the velocity field and $\tens{D}$ is

264: the symmetric part of the velocity gradient tensor, $(\nablu

265: \vecv{v})_{\alpha\beta}\equiv

266: \partial_\alpha v_\beta$. The pressure field

267: $P(\vecv{r},t)$ is determined by enforcing incompressibility,

268: %

269: \be

270: \label{eqn:incomp}

271: \vecv{\nabla}\cdot\vecv{v}=0.

272: \ee

273: %

274: The quantity $\vecv{\Sigma}(\vecv{r},t)$ in Eqn.~\ref{eqn:NS} is the

275: extra stress contributed by the viscoelastic component. In principle,

276: the dynamics of this quantity should be explicitly derived by

277: averaging over the underlying microscopic dynamics of the viscoelastic

278: component. This was done in Refs.~\cite{cates90} for wormlike

279: micelles.  For simplicity, however, we use the phenomenological

280: Johnson-Segalman (JS) model~\cite{johnson-jnfm-2-255-1977}

281: %

282: \begin{widetext}

283: \be

284: \label{eqn:DJS}

285: (\partial_t

286: +\vecv{v}\cdot\nablu )\,\tens{\Sigma} 

287: = a(\tens{D}\cdot\tens{\Sigma}+\tens{\Sigma}\cdot\tens{D}) +

288: (\tens{\Sigma}\cdot\tens{\Omega} + \tens{\Omega}\cdot\tens{\Sigma})  

289:  + 2 G\tens{D}-\frac{\tens{\Sigma}}{\tau}+ \frac{\ell^2}{\tau }\nablu^2 

290:  \tens{\Sigma}.

291: \ee

292: \end{widetext}

293: %

294: In this equation, $G$ is a plateau modulus, $\tau$ is the viscoelastic

295: relaxation time, and $\tens{\Omega}$ is the antisymmetric part of the

296: velocity gradient tensor.  For $a=1$ and $\ell=0$, Eqn.~\ref{eqn:DJS}

297: reduces to the Oldroyd B model, which can be derived by considering

298: the dynamics of an ensemble of Hookean dumbbells in solution. For

299: $|a|<1$, the JS model captures non-affine slip between the dumbbells

300: and the solvent, leading to the drastic shear thinning of

301: Fig.~\ref{fig:schematic}a. Accordingly, $a$ is called the slip

302: parameter. The JS model is the simplest tensorial model to exhibit a

303: regime of negative slope in the homogeneous constitutive curve, and so

304: to predict a shear banding instability. As discussed further below,

305: the diffusive term $\nablu^2 \tens{\Sigma}$ in Eqn.~\ref{eqn:DJS} is

306: needed to correctly describe the ultimate shear banded

307: flow~\cite{lu-prl-84-642-2000}.

308: 

309: Within this model, we study planar shear between parallel plates at

310: $y=0,L$, with the top plate driven at velocity $V \xhat$.  At the

311: plates we assume boundary conditions of

312: $\partial_y\Sigma_{\alpha\beta}=0\;\forall\;\alpha,\beta$ for the

313: viscoelastic stress, with no slip and no permeation for the fluid

314: velocity.  In the linear stability analysis of Sec.~\ref{sec:linear},

315: we consider small values of the Reynolds number $Re=\rho L^2/\eta$. In

316: the nonlinear study of Secs.~\ref{sec:numerics}

317: and~\ref{sec:nonlinear}, we set $Re=0$ at the outset.  Throughout we

318: use units in which $G=1,\tau=1$ and $L=1$. 

319: 

320: %

321: \begin{figure}[b]

322:   \centering

323: \includegraphics[scale=0.4]{./dispersion.eps}

324: \includegraphics[scale=0.42]{./peak.eps}

325:  \caption{a) Dispersion relation for perturbations about a 1D banded

326:  state for $\ell=0.00375,\, 0.00250,\, 0.00125$. $a=0.3$, $\eta=0.05$,

327:  $\rho/\eta=0.02$. Inset: Linear dynamics of 2D code, starting from a

328:  flat interface. Solid lines: weight in modes, $q_zL_z/2\pi=1,2$;

329:  dashed: analytical prediction. b) Peak $\omega^*$ ($q^*$ in inset) in

330:  dispersion relation {\it vs.} $\gdotbar$ across the plateau of

331:  Fig.~\ref{fig:schematic}a. Vertical dashed lines denote the edges of

332:  the stress plateau. }

333: \label{fig:dispersion}

334: \vspace{-0.5cm}

335: \end{figure}

336: %

337: 

338: 

339: \section{1D gradient bands}

340: \label{sec:1D}

341: 

342: As noted above, to capture shear thinning the DJS model invokes a slip

343: parameter $a$ with $|a|<1$, giving non affine deformation of the

344: viscoelastic component~\cite{johnson-jnfm-2-255-1977}. The homogeneous

345: constitutive curve $T_{xy}=\Sigma_{xy}(\gdot, a)+\eta\gdot$ is then

346: capable of non-monotonicity, as in Fig.~\ref{fig:schematic}a.  For an

347: imposed shear rate $\gdotbar\equiv V/L$ in the region of decreasing

348: stress, homogeneous flow is unstable with respect to fluctuations with

349: wavevector in the flow-gradient direction

350: $y$~\cite{yerushal.j-ces-25-1891-1970}. A 1D calculation then predicts

351: separation into gradient bands of differing shear rates $\gdot_1,

352: \gdot_2$, with a flat interface in between. The diffusive term in

353: Eqn.~\ref{eqn:DJS} is needed to account for spatial gradients of the

354: shear rate and viscoelastic stress across the interface, which has a

355: characteristic thickness $O(\ell)$. It also ensures a unique,

356: history-independent banding stress

357: $T_{xy}=T^*$~\cite{lu-prl-84-642-2000}, as seen experimentally.  We

358: expect $\ell=O(10^{-4})$, set by the typical micellar mesh size, in

359: units of the (typical) gap size.

360: 

361: %$\gdot_1=0.66,\gdot_2=7.09$, at a shear stress $T^{\ast}_{xy}=0.506$.

362: 

363: \section{Linear instability}

364: \label{sec:linear}

365: 

366: In Refs.~\cite{fielding-prl-95--2005,wilson-jnfm-138-181-2006}, we

367: considered the linear stability of this 1D gradient banded state with

368: respect to 3D ($x, y, z$) perturbations of infinitesimal amplitude. In

369: the flow direction $\xhat$ and vorticity direction $\zhat$ these are

370: decomposed into Fourier modes with wavevectors

371: $\vecv{q}=q_x\xhat+q_z\zhat$.  (Ref.~\cite{renardy} had previously

372: considered $\vecv{q}=q_x\xhat$ in the pathological limit $\ell=0$,

373: assuming ``top jumping''.)  For diffuse interfaces, $l\gae 0.015$, the

374: 1D state is linearly stable. For $l\lae 0.015$, we find it to be

375: linearly unstable with respect to modes with wavevector

376: $\vecv{q}=q_x\xhat$ in the flow

377: direction~\cite{fielding-prl-95--2005,wilson-jnfm-138-181-2006}.  The

378: associated eigenfunction essentially corresponds to undulations along

379: the interface. For $l\lae 0.005$, the 1D state is also unstable with

380: respect to undulations with wavevector $\vecv{q}=q_z\zhat$ in the

381: vorticity direction. However, these modes are predicted to grow much

382: more slowly than those with $\vecv{q}=q_x\xhat$.  Accordingly, in

383: Refs.~\cite{fielding-prl-95--2005,wilson-jnfm-138-181-2006}, we

384: focused mainly on the dominant modes, $\vecv{q}=q_x\xhat$.

385: Subsequently in Ref.~\cite{fielding-prl-96--2006}, however, we showed

386: that these undulations are cut off, once they attain finite amplitude,

387: by the nonlinear effects of shear. In contrast, the vorticity

388: direction is neutral with respect to the shear. Accordingly, the modes

389: with $\vecv{q}=q_z\vecv{\hat{z}}$ should not suffer this cutoff and

390: are therefore likely to contribute significantly to the ultimate

391: nonlinear state, despite their much slower initial growth rate. With

392: this motivation, in this paper we study the dynamics of the model in

393: the flow-gradient/vorticity $(y-z)$ plane. For simplicity and

394: computational efficiency, we will assume uniformity in the flow

395: direction $x$, returning in Sec.~\ref{sec:nonlinear} below to comment

396: on the validity of this simplification.  It corresponds to taking a

397: vertical slice through one side of an axisymmetric flow state in the

398: planar limit of a Couette device.

399: 

400: The growth rates $\omega$ of the vorticity modes

401: $\vecv{q}=q_z\vecv{\hat{z}}$ are shown in Fig.~\ref{fig:dispersion}a,

402: at a single value of the imposed shear rate. States with thinner

403: interfaces (smaller $\ell$) are more unstable (larger $\omega>0$).

404: Fig.~\ref{fig:dispersion}b shows the growth rate of the maximally

405: unstable mode for shear rates across the stress plateau of

406: Fig.~\ref{fig:schematic}a. The corresponding wavelength

407: $\lambda^*=O(1)$ is of order the rheometer gap $L\equiv 1$ (inset). At

408: small $\ell$, the instability persists across most of the plateau, so

409: is likely to be unavoidable experimentally.  The mechanism of

410: instability is not fully understood, but is likely to stem from steep

411: gradients in the normal stress and shear rate across the

412: interface~\cite{hinch-jnfm-43-311-1992,mcleish87,fielding-prl-95--2005}.

413: 

414: \section{Numerical method}

415: \label{sec:numerics}

416: 

417: To study the undulations once they have grown to attain a finite

418: amplitude, beyond the regime of linear instability, we solve the

419: model's full nonlinear dynamics numerically.  In this section we

420: discuss the details of our numerical method, which is adapted from

421: that of Refs.~\cite{chilcott-jnfm-29-381-1988,fielding-prl-96--2006}.

422: Readers who are not interested in these issues can skip straight to

423: Sec.~\ref{sec:nonlinear} without loss of thread.

424: 

425: The model equations have already been specified in

426: Sec.~\ref{sec:model}, together with the flow geometry, boundary

427: conditions and choice of adimensionalisation. For computational

428: efficiency, our numerical study is confined to the $(y-z)$ plane,

429: assuming translational invariance along the flow direction $x$. In the

430: vorticity direction we take a cell of length $L_z$, with periodic

431: boundary conditions.

432: 

433: We consider the limit of zero Reynolds number, in which

434: Eqn.~\ref{eqn:NS} reduces to

435: %

436: \be

437: \label{eqn:zeroRey}

438: 0 = \nablu .(\tens{\Sigma} + 2\eta\vecv{D} -P\tens{I}).

439: \ee

440: %

441: To ensure that the incompressibility constraint of

442: Eqn.~\ref{eqn:incomp} is satisfied always, we express the velocity in terms

443: of stream functions $\phi$ and $\psi$:

444: %

445: \be

446: \label{eqn:stream}

447: v_x=\partial_y\phi,\;\; v_y=\partial_z\psi,\;\; v_z=-\partial_y\psi.

448: \ee

449: %

450: In this way, Eqn.~\ref{eqn:incomp} need no longer be considered: it

451: remains only solve Eqns.~\ref{eqn:DJS} and~\ref{eqn:zeroRey}, with the

452: velocity expressed as in Eqn.~\ref{eqn:stream}.

453: 

454: 

455: To solve these, the basic strategy is to step along a grid of time

456: values $t^n=n\Delta t$ for $n=1,2,3\cdots$, at each step updating

457: $\tens{\Sigma}^n, \phi^n, \psi^n\to \tens{\Sigma}^{n+1}, \phi^{n+1},

458: \psi^{n+1}$.  Discretization with respect to time of any quantity $f$

459: is denoted $f(t^n)=f^n$, or sometimes below by $f|^n$. At each

460: time-step, we first update the viscoelastic stress $\tens{\Sigma}^n \to

461: \tens{\Sigma}^{n+1}$ using the constitutive equation (\ref{eqn:DJS})

462: with fixed, old values of the stream-functions $\phi^n,\psi^n$. We

463: then update $\phi^n,\psi^n \to \phi^{n+1},\psi^{n+1}$ using the force

464: balance equation (\ref{eqn:zeroRey}) with the new values of

465: $\tens{\Sigma}^{n+1}$.

466: 

467: The update $\tens{\Sigma}^n\to\tens{\Sigma}^{n+1}$ using the

468: viscoelastic constitutive equation (\ref{eqn:DJS}) is performed as

469: follows. As a preliminary step, we rewrite (\ref{eqn:DJS}) in the form

470: %

471: \be

472: \label{eqn:DJSsep}

473: \partial_t\tens{\Sigma}= 

474:  f(\nablu \vecv{v},\tens{\Sigma}) - \vecv{v}\cdot\nablu \tens{\Sigma}

475:  + \ell^2 \nablu^2 \tens{\Sigma}, 

476: \ee

477: %

478: in which $f(\nablu \vecv{v},\tens{\Sigma})$ comprises the

479: non-diffusive terms from the right hand side of Eqn.~\ref{eqn:DJS}.

480: In what follows, the three terms on the right hand side of

481: Eqn.~\ref{eqn:DJSsep} are referred to as the local, advective and

482: diffusive terms respectively. Numerically, they are dealt with in

483: three successive partial updates $\tens{\Sigma}^n \to

484: \tens{\Sigma}^{n+1/3}$, $\tens{\Sigma}^{n+1/3} \to

485: \tens{\Sigma}^{n+2/3}$ and $\tens{\Sigma}^{n+2/3} \to

486: \tens{\Sigma}^{n+1}$.

487: 

488: In the first of these, the local term is handled using an explicit

489: Euler algorithm~\cite{PreTeuVetFla92}, checked for consistency against

490: a fourth order Runge-Kutta algorithm~\cite{PreTeuVetFla92}.

491: Temporarily setting aside the issue of spatial discretization, the

492: Euler algorithm can be written

493: %

494: \be

495: \label{eqn:local}

496: \tens{\Sigma}^{n+1/3}(y,z)=\tens{\Sigma}^n + \Delta t \; f(\nablu

497: \vecv{v}^n,\tens{\Sigma}^n).

498: \ee

499: %

500: In terms of the stream-functions $\phi$ and $\psi$, the

501: velocity-gradient tensor $\nablu \vecv{v}$ has Cartesian components

502: %

503: \be

504: \label{eqn:localM}

505: \nablu \vecv{v}= \left(\begin{array}{ccc}

506:             0 & 0 & 0 \\

507:             \partial_y^2\phi & \partial_y\partial_z\psi &

508:             -\partial_y^2\psi \\

509:             \partial_y\partial_z\phi & \partial_z^2\psi &

510:             -\partial_y\partial_z\psi \end{array}\right),

511: \ee

512: %

513: in which we have omitted the superscripts $n$ for clarity.

514: Eqns.~\ref{eqn:local} and~\ref{eqn:localM} are then spatially

515: discretized on a rectangular grid in real space. In some runs of the

516: code, the grid points are linearly spaced, with $z_i=i\Delta z$ for

517: $i=1\cdots N_z$ and $y_j=j\Delta y$ for $j=1\cdots N_y$.  In others,

518: we used a nonlinear mapping in the $y$ direction to focus attention on

519: the region explored by the interface. For simplicity, most of the

520: description of this section will concern the linear grid, though we

521: will return briefly at the end of the section to discuss the nonlinear

522: modification.  In either case, any spatially discretized function $f$

523: is denoted $f(y^j,z^i)=f_{ij}$, or sometimes $f|_{ij}$.  (The

524: apparently unusual order of the indices is a historical convention on

525: the part of the author, stemming from a previous study in the $x-y$

526: plane.)  Eqn.~\ref{eqn:local} then becomes

527: %

528: \begin{widetext}

529: %

530: \be

531: \label{eqn:localD}

532: \tens{\Sigma}^{n+1/3}_{ij}=\tens{\Sigma}^n_{ij} + \Delta t \; f(\nablu

533: \vecv{v}^n_{ij},\tens{\Sigma}^n_{ij}),

534: \ee

535: %

536: The derivatives in the components of $\nablu\vecv{v}^n_{ij}$ are

537: discretized (in the case of a rectangular grid) as follows:

538: %

539: \be

540: \label{eqn:p2y}

541: \partial_y^2\psi|^n_{ij}=\frac{1}{ \Delta

542:   y^2}\left[\psi_{i(j+1)}^n-2\psi_{ij}^n+\psi_{i(j-1)}^n\right],

543: \ee

544: %

545: \be

546: \partial_z^2\psi|^n_{ij}=\frac{1}{ \Delta

547:   z^2}\left[\psi_{(i+1)j}^n-2\psi_{ij}^n+\psi_{(i-1)j^n}\right],

548: \ee

549: %

550: and

551: %

552: \be

553: \partial_y\partial_z \psi|^n_{ij}=\frac{1}{4\Delta x\Delta

554:   y}\left[\psi_{(i+1)(j+1)}-\psi_{(i+1)(j-1)}-\psi_{(i-1)(j+1)}+\psi_{(i-1)(j-1)}\right].

555: \ee

556: %

557: Corresponding derivatives of $\phi$ are obtained in the same way,

558: replacing $\psi$ by $\phi$ in the above equations. For values of $ij$

559: at the edges of the flow domain, these formulae link to values of the

560: flow variables at ``phantom'' grid points that lie just outside the

561: domain.  These values are specified by imposing the boundary

562: conditions, the spatial discretization of which is discussed at the

563: end of this section.

564: 

565: The advective term is also handled using an explicit Euler

566: algorithm~\cite{PreTeuVetFla92}, on the same real space grid:

567: %

568: \bea

569: \label{eqn:advect}

570: \tens{\Sigma}^{n+2/3}_{ij}&=&\tens{\Sigma}^{n+1/3}_{ij} - \Delta t

571: \left(

572: v_{y}|_{ij}^n \, \partial_y\tens{\Sigma}|^n_{ij}+v_{z}|_{ij}^n\,\partial_z\tens{\Sigma}|^n_{ij}\right)\nonumber\\

573:   &=&\tens{\Sigma}^{n+1/3}_{ij} - \Delta t

574: \left(

575: \partial_z\psi |_{ij}^n \, \partial_y\tens{\Sigma}|^n_{ij}-\partial_y\psi|_{ij}^n\,\partial_z\tens{\Sigma}|^n_{ij}\right).

576: \eea

577: %

578: The derivatives of $\psi$ in this equation are discretized as follows:

579: %

580: \be

581: \label{eqn:p1y}

582: \partial_y\psi|_{ij}^n=\frac{1}{2\Delta

583:   y}\left[\psi_{i(j+1)}-\psi_{i(j-1)}\right]\;\;\;\textrm{and}\;\;\; \partial_z\psi|_{ij}^n=\frac{1}{2\Delta

584:   z}\left[\psi_{(i+1)j}-\psi_{(i-1)j}\right].

585: \ee

586: %

587: %

588: The derivative of $\tens{\Sigma}$ with respect to $y$ in

589: Eqn.~\ref{eqn:advect} was discretized using third-order

590: upwinding~\cite{Pozrikidis}:

591: %

592: \be

593: \partial_y\tens{\Sigma}_{ij}^n=\frac{1}{6\Delta

594:   y}\left[\tens{\Sigma}^n_{i(j-2)} -6 \tens{\Sigma}^n_{i(j-1)} +3

595:   \tens{\Sigma}^n_{ij} +2 \tens{\Sigma}^n_{i(j+1)}

596: \right]\;\;\;\textrm{if}\;\;\; v_y|^n_{ij}>0,

597: \ee

598: %

599: while

600: %

601: \be

602: \partial_y\tens{\Sigma}_{ij}^n=\frac{1}{6\Delta

603:   y}\left[-\tens{\Sigma}^n_{i(j+2)} +6 \tens{\Sigma}^n_{i(j+1)} -3

604:   \tens{\Sigma}^n_{ij} -2 \tens{\Sigma}^n_{i(j-1)}

605: \right]\;\;\;\textrm{if}\;\;\; v_y|^n_{ij}<0,

606: \ee

607: %

608: with analogous expressions for the derivative of $\tens{\Sigma}$ with

609: respect to $z$.

610: 

611: The diffusive term is handled by discretizing on $y$ in real space as

612: above, taking a fast Fourier transform $z\to q_i$ in the vorticity

613: dimension using a standard NAG routine~\cite{NAG}, and solving the

614: resulting problem using a semi-implicit Crank-Nicolson

615: algorithm~\cite{PreTeuVetFla92}:

616: %

617: \be

618: \label{eqn:Crank}

619: \tens{\Sigma}^{n+1}_{ij}-\tens{\Sigma}^{n+2/3}_{ij} =\frac{1}{2} l^2

620: \Delta t

621: \left(\partial_y^2\tens{\Sigma}_{ij}^{n+2/3}-q_i^2\tens{\Sigma}_{ij}^{n+2/3}\right)+\frac{1}{2}

622: l^2 \Delta t

623: \left(\partial_y^2\tens{\Sigma}_{ij}^{n+1}-q_i^2\tens{\Sigma}_{ij}^{n+1}\right),

624: \ee

625: %

626: in which the index $i$ now labels the Fourier mode number. The

627: derivatives $\partial_y^2$ are discretized as in Eqn.~\ref{eqn:p2y}

628: above. Note that Eqn.~\ref{eqn:Crank} contains no mixing of the

629: Cartesian components $\Sigma_{\alpha\beta}$ for any

630: $\alpha\beta=xx,xy,yy\cdots$, so can be solved for each one

631: separately. Bringing all terms in the unknown

632: $\tens{\Sigma}^{n+1}_{ij}$ across to the left hand side, and putting

633: all terms in the known $\tens{\Sigma}^{n+2/3}_{ij}$ on the right hand

634: side, we obtain a sparse set of linear equations characterized by a

635: tridiagonal matrix on the left hand side. These are then solved for

636: the $\tens{\Sigma}^{n+1}_{ij}$ using standard NAG routines~\cite{NAG}.

637: %

638: 

639: Having updated $\tens{\Sigma}^n \to \tens{\Sigma}^{n+1}$ using the

640: viscoelastic constitutive equation, we now update the stream-functions

641: $\phi^n,\psi^n \to \phi^{n+1},\psi^{n+1}$ using the $x,y$ and $z$

642: components of the force balance equation (\ref{eqn:zeroRey}). Again,

643: we work in real flow-gradient space and reciprocal vorticity space. To

644: eliminate the pressure from Eqn.~\ref{eqn:zeroRey}, we subtract

645: $\partial_y$ of the $z$ component from $\partial_z$ of the $y$

646: component to get the following equations, written separately for

647: $q_i=0$ and $q_i\neq 0$:

648: %

649: \be

650: \label{eqn:fb1}

651: \partial_y^3\phi|^{n+1}_{0(j+1/2)}=-\frac{1}{\eta}\partial_y\Sigma_{xy}|_{0(j+1/2)}^{n+1}\;\;\;\textrm{for}\;\;\; q_i=0,

652: \ee

653: %

654: \be

655: \label{eqn:fb2}

656: \partial^3_y\phi|^{n+1}_{i(j+1/2)}-q_i^2 \partial_y\phi|^{n+1}_{i(j+1/2)}=-\frac{1}{\eta}\left(\partial_y\Sigma_{xy}|^{n+1}_{i(j+1/2)}  + iq_i \Sigma_{xz}|^{n+1}_{i(j+1/2)}\right)\;\;\;\textrm{for}\;\;\; q_i\neq 0,

657: \ee

658: %

659: \be

660: \label{eqn:fb3}

661: \partial_y^3\psi|^{n+1}_{0(j+1/2)}=\frac{1}{\eta}\partial_y\Sigma_{yz}|^{n+1}_{0(j+1/2)}\;\;\;\textrm{for}\;\;\;q_i=0,

662: \ee

663: %

664: \be

665: \label{eqn:fb4}

666: \nablu^4\psi|^{n+1}_{ij}=-\frac{1}{\eta}\left[iq_i\partial_y\left(\Sigma_{yy}-\Sigma_{zz}\right)|^{n+1}_{ij}-\left(q_i^2+\partial_y^2\right)\Sigma_{yz}|^{n+1}_{ij}\right]\;\;\;\textrm{for}\;\;\;

667: q_i\neq

668: 0,

669: \ee

670: %

671: with $\nablu^2=(\partial_y^2-q_i^2)$. The real and imaginary parts of

672: these equations are treated separately. The third order equations

673: (\ref{eqn:fb1}), (\ref{eqn:fb2}) and (\ref{eqn:fb3}) are discretized

674: at staggered half grid points $y_{j+1/2}$ for $j=1\cdots N_y-1$, with

675: derivatives calculated as follows:

676: %

677: \be

678: \partial_y f|_{j+1/2}=\frac{1}{\Delta y}\left(f_{j+1}-f_j\right),

679: \ee

680: %

681: and 

682: %

683: \be

684: \partial_y^3 f|_{j+1/2}=\frac{1}{\Delta

685:   y^3}\left(f_{j+2}-3f_{j+1}+3f_{j}-f_{j-1}\right),

686: \ee

687: %

688: for any quantity $f$. (For clarity, the subscript $i$ and the

689: superscript $n$ have been omitted from these expressions.)  The fourth

690: order equation is implemented at full grid points $y_j$, excluding

691: those at the very edge of the domain ($y_0$ and $y_N$). In it,

692: $\partial^2_y$ and $\partial_y$ are discretized as in

693: Eqns.~\ref{eqn:p2y} and~\ref{eqn:p1y} respectively, and $\partial_y^4$

694: according to

695: %

696: \be

697: \partial_y^4 f|_{j}=\frac{1}{\Delta y^4}\left(f_{j+2}-4f_{j+1}+6f_j

698:   -4f_{j-1}+f_{j-2}\right).

699: \ee

700: %

701: Each of Eqns.~\ref{eqn:fb1} to~\ref{eqn:fb4}, for each mode index $i$,

702: then takes the form of a sparse set of linear equations for

703: $\phi_{ij}^{n+1}$ or $\psi_{ij}^{n+1}$. These equations are solved

704: using standard NAG routines~\cite{NAG}.

705: 

706: \end{widetext}

707: 

708: It remains finally to specify the spatial discretization of the

709: boundary conditions. In turn, this will prescribe the values of the

710: flow variables on the phantom grid points that lie just outside the

711: flow domain.

712: 

713: In the vorticity direction $z$, the boundary conditions are periodic.

714: For any quantity $f$ on the grid $z_1\cdots z_{N_Z}$ in real space

715: $z_i$, we thus have $f_{-1}=f_{{N_z}-1}$, $f_{0}=f_{{N_Z}}$, $f_{{N_z}+1}=f_0$,

716: $f_{{N_z}+2}=f_1$. In reciprocal space $q_i$, the periodic boundary conditions

717: are always satisfied.

718: 

719: In the flow-gradient direction $y$, the boundary conditions for the

720: fluid velocity at the plates $y=0,1$ are those of no slip and no

721: permeation. In terms of the stream-function $\phi$, the no slip

722: condition gives $\partial_y\phi=0$ at $y=0$ and

723: $\partial_y\phi=\gdotbar$ at $y=1$. We also note that $\phi$ is only

724: defined up to an arbitrary additive constant, and accordingly choose

725: $\phi=0$ at $y=0$. The third order Eqns.~\ref{eqn:fb1}

726: and~\ref{eqn:fb2} then have three boundary conditions, as required.

727: After discretizing real flow-gradient space $y_j$ in reciprocal

728: vorticity space $q_i$, we then have

729: %

730: $$\phi_{i1}=0,\;\;\;\phi_{i0}=\phi_{i2},\;\;\;\textrm{and}\;\;\;\phi_{i({N_y}+1)}=\phi_{i({N_y}-1)}+\delta_{i0}\gdotbar \Delta y,$$

731: %

732: in which $\delta_{ij}$ is the usual Kronecker delta function.

733: 

734: In terms of the stream-function $\psi$, the no-slip condition gives

735: $\partial_y\psi=0$ at $y=0,1$, and the no-permeation condition gives

736: $\partial_z\psi=0$ at $y=0,1$. We also note that $\psi$ is only

737: defined up to an arbitrary additive constant, and choose $\psi=0$ at

738: $y=0$. In the $q_i=0$ equation (\ref{eqn:fb3}), we then have

739: %

740: $$\psi_{i1}=0,\;\;\;\psi_{i0}=\psi_{i2}\;\;\;\textrm{and}\;\;\;\psi_{i({N_y}+1)}=\psi_{i({N_y}-1)}.$$

741: %

742: These also hold for the $q_i\neq 0$ equation (\ref{eqn:fb4}), which

743: obeys the additional condition

744: %

745: $$\psi_{i{N_y}}=0.$$

746: %

747: The zero-gradient boundary condition for the viscoelastic stress, after

748: discretization on the flow-gradient grid $y_1,y_2\cdots y_{N_y}$, gives

749: $f_{i0}=f_{i2}$ and $f_{i({N_y}+1)}=f_{i({N_y}-1)}$ for all components

750: $f=\Sigma_{xx}, \Sigma_{xy}, \Sigma_{yy}\cdots$. These apply in both

751: real $z_i$ and reciprocal $q_i$ vorticity spaces.

752: 

753: 

754: Extension to the case of a nonlinear grid in the $y$ direction is

755: straightforward in principle, but cumbersome in detail. Discretized

756: derivatives are calculated via the usual Taylor expansions. For

757: example, to first order accuracy, centred second derivatives with

758: respect to $y$ become

759: %

760: \be

761: \partial_y^2

762: f_j=\frac{2}{y_{j+1}-y_{j-1}}\left[\frac{f_{j+1}-f_j}{y_{j+1}-y_j}-\frac{f_j-f_{j-1}}{y_j-y_{j-1}}\right].

763: \ee

764: %

765: We checked our nonlinear mapping carefully by performing a few runs

766: for identical parameter sets with both linear and highly nonlinear

767: grids.  All the numerical results in this paper are converged with

768: respect to grid and timestep, to within the accuracy resolvable on the

769: plots presented.

770: 

771: \begin{figure}[t]

772: \vspace{-0.8cm}\includegraphics[scale=0.63,

773: angle=90]{./gdot2.0_eta0.05_a0.3_l0.00375_Lz4.0_state_time506.34999999494_OPWxx_border150.ps}

774: \includegraphics[scale=0.44, angle=90]{./flowMap.eps}

775: \caption{Steady state at $a=0.3$, $\eta=0.05$, $\gdotbar=2.0$, $\ell=0.00375$, $L_z=4.0$. Left: greyscale of $\Sigma_{xx}$ in $y-z$ plane. Middle: $(y-z)$ velocity vectors, showing Taylor-like rolls. Right: vorticity banding of viscoelastic shear stress.}

776: \label{fig:steady}

777: \vspace{-0.5cm}

778: \end{figure}

779: 

780: 

781: 

782: \section{Nonlinear steady state}

783: \label{sec:nonlinear}

784: 

785: In each simulation run, we input as an initial condition the 1D

786: gradient-banded state discussed in Sec.~\ref{sec:1D}, superposed with

787: Fourier perturbations of tiny random amplitudes.  As expected, under

788: conditions where the linear analysis of Sec.~\ref{sec:linear} predicts

789: the 1D initial state to be unstable with respect to perturbations with

790: wavevectors $\vecv{q}=q_z\zhat$ in the vorticity direction, we find

791: that these initial disturbances grow in time. Full agreement between

792: (i) the early-time growth rate and functional form of the fastest

793: growing mode and (ii) the most unstable eigenvalue and eigenfunction

794: of the linear stability analysis provides a stringent check of our

795: numerical method.

796: 

797: During the instability, the initially flat interface between the bands

798: develops undulations that grow in time.  At long times, once nonlinear

799: effects become important, these saturate in a finite amplitude

800: interfacial undulation to give a 2D steady state

801: (Fig.~\ref{fig:steady}).  The wavelength of the steady undulations

802: corresponds to that of the maximally unstable mode of the linear

803: analysis. For the parameters of Fig.~\ref{fig:steady}, this is roughly

804: twice the gap width.  Associated with these undulations are

805: Taylor-like velocity rolls stacked in the vorticity direction

806: (Fig.~\ref{fig:steady}, middle), accompanied by undulations of the

807: stress along the wall (Fig.~\ref{fig:steady}, right).  The results of

808: Fig.~\ref{fig:steady} could be tested experimentally as follows.

809: Optical measurements should reveal birefringent stripes stacked in the

810: vorticity direction, each of height comparable to the gap width.

811: Likewise, the velocity rolls could potentially be measured using

812: velocimetry. This is a challenging task, however, because the highest

813: speed in Fig.~\ref{fig:steady}c is only $O(0.01)$.

814: 

815: Our results capture recent experimental observations in which an

816: initially 1D gradient banded state of a wormlike surfactant solution

817: was found to destabilise with respect to interfacial undulations with

818: wavevector in the vorticity direction~\cite{lerouge-prl-96--2006}.

819: Indeed, several features of our results can be directly compared with

820: these experiments, as follows. In the ultimate steady state, the

821: wavelength $O(L)$ and amplitude $O(L/10)$ of the undulations in our

822: Fig.~\ref{fig:steady} are comparable to those in Fig.  2 of

823: Ref.~\cite{lerouge-prl-96--2006}, measured in units of the gap width

824: $L$.  The wavelength of these undulations was furthermore reported to

825: increase with increasing average applied shear rate

826: $\gdotbar$~\cite{lerouge-prl-96--2006}, consistent with the inset of

827: our Fig.~\ref{fig:dispersion}b.

828: %

829: \begin{figure}[t]

830:   \centering

831: \includegraphics[scale=0.42]{./selected.eps}

832:  \caption{Selected stress: 1D initial state and 2D steady

833:  state. $L_z=2.0$, $a=0.3$, $\eta=0.05$, $\gdotbar=2.0$. At $\ell=0$, the

834:  stress varies erratically about an average that depends on the

835:  grid. Inset: evolution to steady state at $l=0.00375$. }

836: \label{fig:selected}

837: \vspace{-0.5cm}

838: \end{figure}

839: %

840: 

841: The kinetics of the instability can also be compared, via the temporal

842: evolution of the stress signal. In the

843: experiments~\cite{lerouge-prl-96--2006}, a shear startup protocol was

844: followed. Accordingly, the stress signal showed an initial overshoot

845: followed by a decay (at $\gdotbar=30s^{-1}$) on a timescale $O(\tau)$

846: to a plateau value. This part of the dynamics corresponded to the

847: initial formation of 1D gradient bands.  It is absent from our

848: simulations, because we take as our initial condition an already 1D

849: gradient banded flow.  Subsequently, the stress signal in

850: Ref.~\cite{lerouge-prl-96--2006} slowly increased by about $1\%$ on a

851: timescale $O(100\tau)$.  This part of the dynamics was associated with

852: the 1D gradient banded state destabilising to exhibit vorticity

853: undulations. As shown in Fig.~\ref{fig:selected}, it is captured very

854: well by our simulations: we find a slow stress increase $O(1\%)$ on a

855: time-scale $O(100\tau)$, consistent with the experiments.

856: 

857: Some differences between our work and the experiments of

858: Ref.~\cite{lerouge-prl-96--2006} are noted as follows.  In

859: Ref.~\cite{lerouge-prl-96--2006}, the instability was studied using

860: light scattering techniques, which couple to concentration

861: fluctuations. In the present manuscript, we do not consider

862: concentration coupling. In future work, it might be interesting to

863: perform analogous simulations in the concentration coupled model of

864: Ref.~\cite{fielding-epje-11-65-2003}.  However, an important finding

865: of the present work is that concentration coupling is not actually

866: needed to trigger the basic undulatory instability.  Indeed, we

867: believe this to stem instead from normal stress effects, with

868: concentration coupling a sub-dominant feature. Finally, we have not

869: seen the exotic dynamics reported at the edges of the stress plateau

870: in Ref.~\cite{lerouge-prl-96--2006}. We cannot access small enough

871: $\ell$ for the instability to persist here.  (Recall

872: Fig.~\ref{fig:dispersion}b.)

873: 

874: \begin{figure}[t]

875: \subfigure{

876:   \includegraphics[scale=0.47]{./interface2.eps}

877: }

878: \caption{Profile $\Sigma_{xx}$ normal to the interface; $\ell=0.00375$ (thick lines), $\ell=0.00250, 0.00125$ (thin). $a=0.3$, $\eta=0.05$, $\gdotbar=2.0$, $L_z=2.0$. a,b) 1D state with interfacial thickness $d\propto \ell$. (c,d), (e,f) 2D  state at the $z-$coordinate where the interface has maximal $+$ and $-$ $y$ displacements. The  thickness $d$ appears {\em independent} of $\ell$, but note the  bump of thickness $O(l)$. The offset $y_0(\ell)$ is chosen to centre the profile at the origin in each case.}

879: % results files as follows

880: % gdot2.0_eta0.05_a0.3_l0.00375_Lz2.0_temperature1.0e-4_Dt0.001_Nz200_Nin1000_Nout100_adapt1_tmax800.0 

881: % gdot2.0_eta0.05_a0.3_l0.0025_Lz2.0_temperature1.0e-4_Dt0.001_Nz400_Nin1600_Nout160_adapt1_tmax800.0

882: %gdot2.0_eta0.05_a0.3_l0.00125_Lz2.0_temperature1.0e-4_Dt0.002_Nz600_Nin2000_Nout200_adapt1_tmax600.0

883: \label{fig:interface}

884: \vspace{-0.5cm}

885: \end{figure}

886: 

887: We return to comment on the validity of restricting our study to the

888: $y-z$ plane, which was done mainly for computational efficiency. As

889: seen from our results, the ultimate amplitude of the interfacial

890: undulation in this ($y-z$) plane is in fact comparable to that

891: reported in the ($x-y$) plane in

892: Ref.~\cite{fielding-prl-96--2006} (to within $10\%$ at $a=0.3$,

893: $\eta=0.05$, $\gdotbar=2.0$, $\ell=0.00375$, $L_{\{x,z\}}=2$). The

894: present study therefore shows that vorticity structuring is indeed

895: important, but also that a full 3D simulation should be performed in

896: future work.

897: 

898: Finally, we discuss briefly the thickness $d$ of the interface between

899: the bands.  In the 1D initial state, $d=O(\ell)$. In the limit

900: $\ell\to 0$, this gives an unphysically sharp interface $d\to

901: 0$. Associated with this is a pathological steady state that strongly

902: depends strongly on the flow history~\cite{olmsted-jr-44-257-2000}. In

903: 2D, in contrast, $d$ appears virtually independent of $\ell$, as shown

904: in Fig.~\ref{fig:interface}c-f.  This is an important finding that

905: could potentially obviate the gradient term $l^2\nabla^2\tens\Sigma$

906: in Eqn.~\ref{eqn:DJS}, which is needed to give a finite interfacial

907: thickness in 1D~\cite{lu-prl-84-642-2000}.  Nonetheless, the

908: interfacial profile does retain a small bump of thickness $O(\ell)$,

909: Fig.~\ref{fig:interface}e-f, suggesting that the local case $\ell=0$

910: remains pathological even in 2D. Indeed, at $\ell=0$ the stress signal

911: varies erratically about an average that varies between runs,

912: Fig.~\ref{fig:selected}, though purely numerical instability cannot be

913: ruled out. This important issue will be pursued further in future

914: work.

915: 

916: To summarise, we have identified a mechanism by which vorticity stress

917: bands and Taylor-like velocity rolls can form in a complex fluid,

918: triggered by the instability of gradient shear banded flow with

919: respect to interfacial undulations. In any real startup experiment, we

920: would expect the vorticity instability to commence during the final

921: stages of the initial band formation: above we assumed a complete

922: separation of timescales between the processes. In future work, we

923: will study the true dynamics of shear startup experiment in curved

924: Couette flow. We will also extend to 3D, to study the interplay of

925: vorticity banding with the dynamics of

926: Ref.~\cite{fielding-prl-96--2006}. Robustness of the mechanism in

927: other models will also be studied.

928: 

929:  The author thanks Peter Olmsted, Paul Callaghan, Georgina Wilkins and

930:  Helen Wilson for discussions, and the UK's EPSRC for financial

931:  support, GR/S29560/01.

932: 

933:  \begin{thebibliography}{10}

934: 

935: \bibitem{BritCall97c}

936: M.~M. Britton and P.~T. Callaghan, Phys. Rev. Lett. {\bf 78},  4930  (1997).

937: 

938: \bibitem{berret-pre-55-1668-1997}

939: J.~F. Berret, G. Porte, and J.~P. Decruppe, Phys. Rev. E {\bf 55},  1668

940:   (1997).

941: 

942: \bibitem{BecManCol04}

943: L. Becu, S. Manneville, and A. Colin, Phys.\ Rev.\ Lett. {\bf 93},  art. no.

944:   (2004).

945: 

946: \bibitem{diat93}

947: O. Diat, D. Roux, and F. Nallet, J.~Phys.~II (France) {\bf 3},  1427  (1993).

948: 

949: \bibitem{wilkinsOlms2006}

950: G.~M.~H. Wilkins and P.~D. Olmsted, Submitted for publication  (2006).

951: 

952: \bibitem{SalManCol03}

953: J.~B. Salmon, S. Manneville, and A. Colin, Phys.\ Rev.\ E {\bf 68},  art. no.

954:   (2003).

955: 

956: \bibitem{pujolle01}

957: C. Pujolle-Robic and L. Noirez, Nature {\bf 409},  167  (2001).

958: 

959: \bibitem{lettinga-jpm-16-S3929-2004}

960: M.~P. Lettinga and J.~K.~G. Dhont, J. Physics-condensed Matter {\bf 16},  S3929

961:    (2004).

962: 

963: \bibitem{dhont-fd-123-157-2003}

964: J.~K.~G. Dhont, M.~P. Lettinga, Z. Dogic, T.~A.~J. Lenstra, H. Wang, S.

965:   Rathgeber, P. Carletto, L. Willner, H. Frielinghaus, and P. Lindner, Faraday

966:   Discussions {\bf 123},  157  (2003).

967: 

968: \bibitem{berret-prl-8704--2001}

969: J.~F. Berret and Y. Serero, Phys. Rev. Lett. {\bf 8704},    (2001).

970: 

971: \bibitem{CRBMGHJL02}

972: P. Coussot, J.~S. Raynaud, F. Bertrand, P. Moucheront, J.~P. Guilbaud, H.~T.

973:   Huynh, S. Jarny, and D. Lesueur, Phys.\ Rev.\ Lett. {\bf 88},  art. no.

974:   (2002).

975: 

976: \bibitem{HilVla02}

977: L. Hilliou and D. Vlassopoulos, Ind.\ Eng.\ Chem.\ Res. {\bf 41},  6246

978:   (2002).

979: 

980: \bibitem{ChenZAHSBG92}

981: L.~B. Chen, C.~F. Zukoski, B.~J. Ackerson, H.~J.~M. Hanley, G.~C. Straty, J.

982:   Barker, and C.~J. Glinka, Phys. Rev. Lett. {\bf 69},  688  (1992).

983: 

984: \bibitem{Yerushalmi70}

985: J. Yerushalmi, S. Katz, and R. Shinnar, Chemical Engineering Science {\bf 25},

986:   1891  (1970).

987: 

988: \bibitem{lu-prl-84-642-2000}

989: C.~Y.~D. Lu, P.~D. Olmsted, and R.~C. Ball, Phys. Rev. Lett. {\bf 84},  642

990:   (2000).

991: 

992: \bibitem{Bonn+98}

993: D. Bonn, J. Meunier, O. Greffier, A. Alkahwaji, and H. Kellay, Phys. Rev. {\bf

994:   E} {\bf 58},  2115  (1998).

995: 

996: \bibitem{fischer-ra-41-35-2002}

997: P. Fischer, E.~K. Wheeler, and G.~G. Fuller, Rheologica Acta {\bf 41},  35

998:   (2002).

999: 

1000: \bibitem{fischer-ra-39-234-2000}

1001: P. Fischer, Rheologica Acta {\bf 39},  234  (2000).

1002: 

1003: \bibitem{goveas-epje-6-79-2001}

1004: J.~L. Goveas and P.~D. Olmsted, European Phys. J. E {\bf 6},  79  (2001).

1005: 

1006: \bibitem{olmsted-pre-60-4397-1999}

1007: P.~D. Olmsted and C.~Y.~D. Lu, Phys. Rev. E {\bf 60},  4397  (1999).

1008: 

1009: \bibitem{drazin}

1010: P.~G. Drazin and W.~H. Reid, {\em Hydrodynamic stability} (Cambridge University

1011:   Press, Cambridge, 2004).

1012: 

1013: \bibitem{larson-jfm-218-573-1990}

1014: R.~G. Larson, E.~S.~G. Shaqfeh, and S.~J. Muller, J. Fluid Mechanics {\bf 218},

1015:    573  (1990).

1016: 

1017: \bibitem{Lars92c}

1018: R.~G. Larson, Rheol. Acta. {\bf 31},  213  (1992).

1019: 

1020: \bibitem{kang-pre-74--2006}

1021: K.~G. Kang, M.~P. Lettinga, Z. Dogic, and J.~K.~G. Dhont, Phys. Rev. E {\bf

1022:   74},    (2006).

1023: 

1024: \bibitem{fielding-prl-95--2005}

1025: S.~M. Fielding, Phys. Rev. Lett. {\bf 95},    (2005).

1026: 

1027: \bibitem{wilson-jnfm-138-181-2006}

1028: H.~J. Wilson and S.~M. Fielding, J. Non-newtonian Fluid Mechanics {\bf 138},

1029:   181  (2006).

1030: 

1031: \bibitem{lerouge-prl-96--2006}

1032: S. Lerouge, M. Argentina, and J.~P. Decruppe, Phys. Rev. Lett. {\bf 96},

1033:   (2006).

1034: 

1035: \bibitem{decruppe-pre-73--2006}

1036: J.~P. Decruppe, O. Greffier, S. Manneville, and S. Lerouge, Phys. Rev. E {\bf

1037:   73},    (2006).

1038: 

1039: \bibitem{britton-prl-78-4930-1997}

1040: M.~M. Britton and P.~T. Callaghan, Phys. Rev. Lett. {\bf 78},  4930  (1997).

1041: 

1042: \bibitem{cates90}

1043: M.~E. Cates, J.~Phys. Chem. {\bf 94},  371  (1990).

1044: 

1045: \bibitem{johnson-jnfm-2-255-1977}

1046: M.~W. Johnson and D. Segalman, J. Non-newtonian Fluid Mechanics {\bf 2},  255

1047:   (1977).

1048: 

1049: \bibitem{yerushal.j-ces-25-1891-1970}

1050: W. Yerushal.j, S. Katz, and R. Shinnar, Chem. Engineering Science {\bf 25},

1051:   1891  (1970).

1052: 

1053: \bibitem{renardy}

1054: Y.~Y. Renardy, The. Comp. Fl.~Dyn. {\bf 7},  463  (1995).

1055: 

1056: \bibitem{fielding-prl-96--2006}

1057: S.~M. Fielding and P.~D. Olmsted, Phys. Rev. Lett. {\bf 96},    (2006).

1058: 

1059: \bibitem{hinch-jnfm-43-311-1992}

1060: E.~J. Hinch, O.~J. Harris, and J.~M. Rallison, J. Non-newtonian Fluid Mechanics

1061:   {\bf 43},  311  (1992).

1062: 

1063: \bibitem{mcleish87}

1064: T.~C.~B. McLeish, J.~Poly. Sci. B-Poly. Phys. {\bf 25},  2253  (1987).

1065: 

1066: \bibitem{chilcott-jnfm-29-381-1988}

1067: M.~D. Chilcott and J.~M. Rallison, J. Non-newtonian Fluid Mechanics {\bf 29},

1068:   381  (1988).

1069: 

1070: \bibitem{PreTeuVetFla92}

1071: W.~H. Press, S.~A. Teukolsky, W.~T. Vetterling, and B.~P. Flannery, {\em

1072:   Numerical Recipes in C (2nd ed.)} (Cambridge University Press, Cambridge,

1073:   1992).

1074: 

1075: \bibitem{Pozrikidis}

1076: C. Pozrikidis, {\em Introduction to Theoretical and Computation Fluid Dynamics}

1077:   (Oxford University Press, New York, 1997).

1078: 

1079: \bibitem{NAG}

1080: Numerical Algorithms Group Ltd., Wilkinson House, Jordan Hill Road, Oxford, OX2

1081:   8DR, UK.

1082: 

1083: \bibitem{fielding-epje-11-65-2003}

1084: S.~M. Fielding and P.~D. Olmsted, European Phys. J. E {\bf 11},  65  (2003).

1085: 

1086: \bibitem{olmsted-jr-44-257-2000}

1087: P.~D. Olmsted, O. Radulescu, and C.~Y.~D. Lu, J. Rheology {\bf 44},  257

1088:   (2000).

1089: 

1090: \end{thebibliography}

1091: 

1092: 

1093: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1094: %\bibliographystyle{prsty}

1095: %\bibliography{ackerson,actin,articles,banding,barham,Berret,berrportdecruppe,berthier,books,bray,callaghan,cates,chandcits,cook,crystal,crystal_theory,Decruppe,dhont,dnatheory,elasticTurbulence,fielding,fischer,Fischer,flowcryst,fredrickson,gelbart,goveas,graham,Groisman,head,hebraud,helfrich,HinchRallison,hsiao,Kadoma,larson,LCtheory,leal,lerougeDecruppe,lifshitz,line_tension,maffettone,malkus,master,mccoy,membs,Mexican,noirez,notes,olmsted,onions,otherRelated,phanThien,phd1,phd,pine,PineHu,pomeau,poon,pratt,psolutions,ramaswamy,recent,rheochaos,rheofolks,ryan,salmon,SalmonManneville,savedrecs.txt,schoot,semenov,sgrband,shaqfeh,sood,sriram,stein,sureshkumar,vansaarloos,vorticity,Wang,weitz,wilson,worms2,worms3,worms,yuan,zubarev}

1096: 

1097: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1098: 

1099: \end{document}

1100: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1101: 

1102: