1: %\documentstyle[aps,preprint]{revtex}
2: \documentstyle[multicol,aps,prl]{revtex}
3:
4: \begin{document}
5:
6: \widetext
7:
8: \draft
9:
10: \title{Chaotic flow and efficient mixing in microchannel with a polymer solution.}
11:
12: \author{Teodor Burghelea$^a$, Enrico Segre$^b$, Israel Bar-Joseph$^c$, Alex Groisman$^d$ and
13: Victor Steinberg$^a$}
14: \address{$^a$ Department of Physics of Complex Systems, Weizmann Institute of Science, Rehovot, 76100 Israel;\\
15: $^b$ Department of Physical Services, Weizmann Institute of Science, Rehovot, 76100 Israel;\\
16: $^c$ Department of Condensed Matter Physics, Weizmann Institute of Science, Rehovot, 76100 Israel;\\
17: $^d$ Department of Physics, UCSD, 9500 Gilman Dr., La Jolla, CA, 92093-0374.}
18:
19: \date{\today}
20:
21: \maketitle
22: \begin{abstract}
23:
24: Microscopic flows are almost universally linear, laminar and stationary because Reynolds number, $Re$,
25: is usually very small. That impedes mixing in micro-fluidic devices, which sometimes limits their
26: performance. Here we show that truly chaotic flow can be generated in a smooth micro-channel of a
27: uniform width at arbitrarily low $Re$, if a small amount of flexible polymers is added to the working
28: liquid. The chaotic flow regime is characterized by randomly fluctuating three-dimensional velocity
29: field and significant growth of the flow resistance. Although the size of the polymer molecules
30: extended in the flow may become comparable with the micro-channel width, the flow behavior is fully
31: compatible with that in a table-top channel in the regime of elastic turbulence. The chaotic flow
32: leads to quite efficient mixing, which is almost diffusion independent. For macromolecules, mixing
33: time in this microscopic flow can be three to four orders of magnitude shorter than due to molecular
34: diffusion.
35: \end{abstract}
36: \pacs {PACS numbers: 47.27.-i,47.50.+d,83.50.-v}
37:
38: \begin{multicols}{2}
39: \narrowtext
40:
41: \section{Introduction.}
42:
43: Flows of liquids in microscopic channels have been attracting increasing interest recently due to fast
44: development of microfluidics and soft lithography\cite{1,2}. The microfluidic devices, which are
45: becoming increasingly advanced and reliable, allow dramatic reduction of amounts of reagents required
46: for fine chemistry and biochemistry\cite{3}, well controlled manipulation and sophisticated
47: experiments on individual cells\cite{4,5,6} and macromolecules\cite{7}. The microscopic flows are
48: almost universally laminar, with linear dependence of the flow rate on the driving force. They also
49: usually remain steady as long as the driving force does not change. All of that has to do with low to
50: moderate values of the Reynolds number, $Re=Vd\rho/\eta$, which is a general measure of non-linear
51: inertial effects in the flow and of likelihood to find it being chaotic or turbulent\cite{8}. Here $V$
52: is the flow velocity, $d$ is the diameter of the channel, and $\rho$ and $\eta$ are the density and
53: the viscosity of the fluid, respectively. When $d$ is reduced, the flow velocity needed to reach a
54: given high $Re$, which is required to generate chaotic or turbulent flow, increases as $d^{-1}$ . The
55: driving pressure per unit length scales like $\Delta P/\Delta L\sim \eta V/d^2$ giving $\Delta
56: P\sim\Delta L/d^3$ at a given $Re$ in the channel. Therefore, if the channel proportions are
57: preserved, $\Delta P$ grows quadratically with $d^{-1}$ and when the channels are only a few tens of
58: microns wide, achieving sufficiently high $Re$ requires impractically high driving pressures.
59:
60: The laminar character of microscopic flows has multiple practical advantages including possibility of
61: precise control of flow velocity, chemical concentration profiles\cite{4,5,6}, and targeted delivery
62: of chemicals and particles\cite{5}. On the other hand, the laminar flows have an inherent problem of
63: inefficient mass transfer in directions perpendicular to the main flow, which occurs due to molecular
64: diffusion only. Diffusion time, $d^2/D$, across a typical micro-channel with a width of 100 $\mu$m is
65: on the order of 100 s even for moderate size proteins, such as bovine serum albumin with a diffusion
66: coefficient of $D\approx 3\cdot 10^{-7}$ cm$^2$/s in water\cite{9}.
67:
68: A few techniques have been proposed to generate stirring by a three-dimensional flow in order to
69: increase the rate of mixing in the micro-channels. They include application of time dependent external
70: forces\cite{10,11} and raising $Re$ to moderately high values in curvilinear three-dimensional
71: channels\cite{12,13}. An ingenious method of mixing has been suggested recently, which involves
72: special "herring-bone" patterning of a micro-channel wall to generate fluid motion perpendicular to
73: the main flow direction in the linear, low $Re$ regime\cite{14}. The fluid elements are continuously
74: stretched and folded in the flow as they advance along the channel. That separates closely spaced
75: fluid particles and brings distant particles together, dramatically reducing the characteristic length
76: scales and diffusion time and increasing the rate of homogenization of the mixture. The flow was
77: stationary in the laboratory frame, however. Therefore, the concentration profiles of the fluorescent
78: dye used as a tracer were uniquely defined by the entrance conditions and the channel
79: geometry\cite{14}. They did not change in time, and there was always some constant difference in
80: concentration between neighboring points.
81:
82: The basic condition of linearity in low $Re$ flows can be changed by adding flexible high molecular
83: weight polymers to the working liquid\cite{15}. Solutions of those polymers are known as non-Newtonian
84: visco-elastic fluids\cite{15}. Mechanical stress in these fluids depends on the flow history with some
85: characteristic relaxation time, $\lambda$, which for dilute solutions is a time of relaxation of
86: individual polymer molecules. Another specific property of the polymer solutions is the non-linear
87: dependence of the polymer contribution to the stress on the rate of deformation in the flow, $\nabla
88: V$ \cite{15}. This non-linearity usually becomes significant, when the Weissenberg number,
89: $Wi=\lambda\nabla V$, becomes on the order of unity. The non-linear growth of the elastic polymer
90: stresses is especially striking in extensional flows at $Wi>1/2$, where apparent viscosity of polymer
91: solutions can rise by up to three orders of magnitude as the total deformation increases\cite{16}. In
92: pure shear flows the major non-linear elastic effect shows up in the appearance of negative normal
93: stress along the flow direction. It leads to the well known effect of rod climbing\cite{15,17} and
94: causes purely elastic instabilities in curvilinear flows of viscous polymer solutions in table-top
95: set-ups, when inertial effects are virtually absent\cite{18}. The non-linear polymer stresses in
96: curvilinear shear flows can also lead to elastic turbulence, a random multi-scale three-dimensional
97: flow, which can develop at arbitrarily low $Re$\cite{19}. Elastic turbulence causes sharp growth of
98: the flow resistance\cite{19,20}, and it was found to generate efficient mixing in a table-top
99: curvilinear channel\cite{21}.
100:
101: Recently it was shown that the non-linear elasticity of the polymer solutions can cause a purely
102: elastic transition and non-linear growth of flow resistance in a microscopic channel with
103: contractions\cite{22}. Those non-linear effects, however, were due to regions of fast extensional flow
104: near the contractions, where individual polymer molecules are expected to partially unravel at
105: $Wi>1/2$ \cite{23,24}. The non-linearity in resistance was the most dramatic feature of the transition
106: and it became quite significant before any substantial changes in the flow pattern could be
107: seen\cite{22}. Therefore the non-linear resistance could be understood as a simple additive effect of
108: individual molecules forced through the contractions. Although the flow became rather irregular at
109: higher flow rates, the fluctuating flow regions were mostly near the contractions, and no detailed
110: study of those fluctuations was made.
111:
112: If the basic linear flow is a pure shear as in Ref.\cite{19,20,21}, a non-linear elastic transition
113: can only occur through a major reorganization of the flow structure. In Ref.\cite{19,20,21} the
114: secondary flow generated above the instability threshold had well expressed turbulent features and
115: involved irregular fluid motion in a broad range of temporal and spatial scales. That implies an
116: essentially collective effect of the polymer molecules on the flow, and the polymer solution behaving
117: as a visco-elastic continuum. If the size of the set-up is reduced to a micro-scale, non-homogeneity
118: in the polymer concentration and extension of polymer molecules stretched in the flow may become
119: comparable with the size of the set-up, and larger than size of some of the generated vortices.
120: Therefore, whether or not the purely elastic instability and elastic turbulence can be reproduced in a
121: microscopic shear flow is still an open question.
122:
123: Here we show that a fully developed chaotic flow similar to the elastic turbulence can be generated in
124: a flat curvilinear microscopic channel with smooth walls and uniform width at arbitrarily low $Re$, if
125: the working liquid contains a small amount of high molecular weight polymers. The flow is
126: characterized by significant non-linear growth in resistance, randomly fluctuating velocity field and
127: chaotic three-dimensional mixing patterns. It is further shown that stirring by the flow results in
128: efficient mixing in the micro-channel with characteristic mixing length significantly shorter compared
129: with the "herring-bone" patterning method reported before\cite{14}. The characteristic mixing times
130: for solutions of macromolecules are reduced by three to four orders of magnitude compared with
131: molecular diffusion.
132:
133: \section{ Materials and Methods.}
134:
135: \subsection{ Device fabrication.}The micro-channel devices consist of a silicon elastomer
136: (Sylgard 184 by Dow Corning) chip sealed to a \#1 microscope cover glass. The channel structure of the
137: chip was fabricated using the technique of soft lithography\cite{1}. First, a negative master mold was
138: fabricated in UV-curable epoxy (SOTEC micro-systems SU8-1070) by using conventional photolithography.
139: The epoxy was spun onto a silicon wafer at 1800 rpm for 60 s to create a 100 $\mu$m layer and
140: patterned by using a high-resolution negative transparency mask. Liquid elastomer was poured on the
141: mold to a thickness of $\approx 5$ mm and cured in an 80C oven for 1 hr 30 min. After that the
142: elastomer was peeled off the mold, trimmed to its final size and liquid feeding ports were punched by
143: using a 20-gauge luer stub. The patterned side of the chip was bonded to the cover glass by overnight
144: baking in the 80C oven.
145:
146: \subsection{Experimental set-up.} A snapshot of our first microfluidic set-up is shown in Fig.1. It has a
147: uniform thickness $d =100 \mu$m. Its main active element is a curvilinear channel with square
148: cross-section. It is a chain of 40 identical segments, which are couples of interconnected half-rings
149: with inner and outer radii $R_i=100 \mu$m and $R_o=200 \mu$m, respectively, Fig.1b. It has the same
150: proportions as the table-top channel, which was used in the elasticity induced mixing experiments
151: reported before\cite{21}, but its dimensions are reduced by a factor of 30. Because of the periodic
152: structure of the channel, it is convenient to use the number of a segment, $N$, starting from the
153: inlet as a discrete linear coordinate along the channel. The auxiliary rectilinear channel (b) has
154: width of 90 $\mu$m and total length of about 72.5 mm. Channel (b) and the comparator region (c) serve
155: to make differential in situ measurements of flux vs. pressure by the method described in
156: Ref.\cite{22}.
157:
158: \subsection{Flow control.} The flow in the micro-channels was generated and controlled by pressure
159: differences between the inlets and the outlet, Fig.1a. The pressures were generated hydrostatically
160: using long vertical rails with precise rulers and sliding stages. Working liquids were kept in 30 ml
161: plastic syringes, which were held upright, open to the atmosphere and connected to the two inlets and
162: the outlet by plastic tubing with internal diameter of 0.76 mm. The pressure drop in the tubing was
163: estimated below 1\% of the total. The two syringes feeding the inlets were attached to the sliding
164: stages. The difference in liquid elevation between these two syringes and the outlet syringe was
165: measured and adjusted with a precision of about 0.1 mm corresponding to 1 Pa in pressure. Dependence
166: of the volumetric low rate, Q, in the curvilinear channel on the pressure difference between inlet 1
167: and outlet, Fig.1, was determined with a relative precision of about 0.5\% using an in-situ
168: measurement technique described elsewhere\cite{22}. A syringe pump (PHD 2000 by Harvard Apparatus
169: Inc.) with a 50 ml gastight Hamilton syringe was used for an absolute flow rate calibration.
170:
171: \subsection{Polymer solutions.} The polymer used was polyacrylamide, PAAm, by Polysciences Inc. with high
172: molecular weight $M_w=1.8\cdot 10^7$, which was the same polymer sample as in Ref.\cite{21}. It was
173: dissolved at identical
174: concentrations of 80 ppm by weight in two Newtonian solvents with different viscosities. The solvent
175: for the low viscosity solution 1 was a 35\% solution of sucrose in water with 1\% of NaCl added to fix
176: the ionic contents. The Newtonian viscosity of the solvent, $\eta_s$, was 4.2 mPas at the room temperature
177: of 22C. The viscosity of solution 1, $\eta$, was 5.6 mPas at a shear rate of 50 s$^{-1}$, suggesting a dilute
178: polymer solution. The Newtonian solvent for the high viscosity solution 2 was sugar syrup containing
179: 64.4\% sucrose and 1\% of NaCl in water. The viscosity of solvent 2 was 114 mPas at 22C, and the
180: viscosity of solution 2 was 138 mPas at a shear rate of 2 s$^{-1}$. The polymer relaxation time, $\lambda$,
181: of solution 2 measured by phase shift between stress and shear rate in an oscillatory flow regime was 1.1 s.
182: The measurements of viscosity and relaxation time were made using a high precision rheometer
183: (AR1000 by TA Instruments). Relaxation time of solution 1 was estimated as 0.04 s with the assumption
184: that $\lambda$ scales linearly with $\eta_s$\cite{15}.\\
185: The overlap concentration $c^*$, taken as a concentration at which the viscosity ratio reached
186: $\eta/\eta_s=2$, was 200 ppm by weight for solvent 2, corresponding to molecular concentration of
187: $n=8.76\cdot 10^{12}$ cm$^{-3}$. Characteristic size of the polymer coils at rest can be estimated
188: from this as $n^{-1/3}\approx 0.5\mu$m, and characteristic distance between them at 80 ppm by weight
189: can be estimated as 0.7 $\mu$m. These estimates are well supported by the data on PAAm taken from the
190: literature\cite{25}. One can use the Mark-Houwink scaling relation for PAAm, $[\eta]=6.31\times
191: 10^{-3}M_w^{0.8}$ (in ml/g)\cite{25}, where $[\eta]$ is the intrinsic viscosity of the polymer defined
192: as $[\eta]=\frac{\eta-\eta_s}{c\eta_s}|_{c\rightarrow 0}$. At molecular weight $M_w=1.8\times 10^7$
193: one gets $[\eta]=4020$ ml/g. Defining $c^*=1/[\eta]$\cite{15} we obtain $c^*=250$ ppm by weight for an
194: aqueous solution in a good agreement with the above estimate. Further, we can compare the estimated
195: size of the PAAm coils with the data obtained from light scattering Ref\cite{25}. Plugging in
196: $M_w=1.8\times 10^7$ into an interpolation relation from Ref.\cite{15}, one gets $R_g\approx 0.4\mu$m
197: for the radius of gyration of the coils, which is rather consistent with the above estimate obtained
198: from $c^*$.
199:
200: It is worth noting that $c^*=200$ ppm is very close to the value found for $\lambda$-phage
201: DNA\cite{26}, which has a comparable molecular weight of $3.1\times 10^7$. Radius of gyration of the
202: $\lambda$-phage DNA was found to be $0.73\mu$m\cite{27}, which is quite in line with the above
203: estimate for the PAAm coil size. The full contour length of a PAAm molecule having $M_w=1.8\times
204: 10^7$ and thus consisting of $2.5\times 10^5$ monomers (having molecular weight 71.08 g/mol) can be
205: estimated as about $50\mu$m, if a monomer length of $0.2$ nm is assumed. It is significantly larger
206: than the contour length of the $\lambda$-phage DNA (which is equal to about $16\mu$m) and twice
207: smaller than the micro-channel diameter.
208:
209: The experiments on mixing
210: were carried out with polymer solution 2, and then fluorescent dyes with low diffusivities were added
211: to the solution and used as tracers. Those were a few different samples of fluorescein-conjugated
212: Dextran, FITCD, by Sigma with average molecular weights, $M$, varying from 10 kDa to 2 MDa. In spite
213: of the relatively high molecular weight of FITCD it did not have any measurable influence on the
214: solution rheology due to high rigidity of the polysaccharide molecules. Diffusion coefficients of the
215: FITCD samples in water were estimated using the data in Ref.\cite{28}, giving values from $9.1\cdot
216: 10^{-7}$ to $7.4\cdot 10^{-8}$ cm$^2$/s that corresponded to a broad range of biological
217: macromolecules. The diffusion coefficients in solvent 2 were estimated with assumption that $D\sim
218: 1/\eta_s$, resulting in $D_1=6.6\cdot 10^{-9}$ and $D_2=5.4\cdot 10^{-10}$ cm$^2$/s for 10 kDa and 2
219: MDa FITCD, respectively.
220:
221: \subsection{Measuring flow velocity.} Measurements of the flow velocity in the micro-channel were carried
222: out using custom developed microscopic particle image velocimetry, micro-PIV. The polymer solution was
223: seeded with 0.2 $\mu$m yellow-green fluorescent beads (Polysciences), and epi-fluorescent imaging of
224: the flow in the micro-channel, Fig. 1, was made with an inverted microscope (Olympus IMT2) and narrow
225: band excitation and emission filters in the dichroic filter cube. The objective was a LWD $20\times$,
226: N.A. = 0.40 , and the images were projected onto a CCD array with $640\times 480$ pixel resolution
227: (PixelFly camera by PCO, Germany) and digitized to 12 bits. The snapshots of the flow were taken with
228: even time intervals of 40 ms, and digitally post-processed. Images of out-of-focus particles were
229: disregarded. Velocity field was found by cross-correlating positions of the particles in two
230: consecutive snapshots, and the particle velocity vectors were neighbor-validated. (The calculated
231: velocity field corresponded to the time interval between the two snapshots.) The collected time series
232: represented velocity values measured at equal distances from interconnections of two half-rings at
233: $N=35$, and averaged over a $20\times 20\mu$m square region at the middle of the channel and over 4
234: $\mu$m across the channel mid-plane.
235:
236: \subsection{Measuring tracer concentration profiles.} Concentrations of the fluorescent dyes, which were
237: used as passive tracers in the experiments on mixing in the channel, were measured using a commercial
238: confocal microscope (Fluoview FV500 by Olympus). It was equipped with a $40\times$ N.A.= 0.85
239: infinity corrected objective and a 12 bit photomultiplier. The scanning was done at a rate of 56 lines
240: per second and 512 pixels per line corresponding to a step of 0.18 $\mu$m per pixel .
241:
242: \section{ Results.}
243:
244: We measured volumetric flux rate, $Q$, of solution 1 through the curvilinear channel, Fig.1, in a
245: broad range of applied pressures and calculated the resistance factor, $Z = P /Q$, where $\Delta P$ is
246: pressure drop per segment. The resistance factor is a constant proportional to viscosity for Newtonian
247: fluids in linear, low $Re$ regime, and it can be used as a measure of turbulent flow resistance in
248: large channels at high $Re$. Fig. 2 shows dependence of $Z$ on $Q$ for solution 1, after $Z$ is
249: divided by a resistance factor, $Z_0$, found for a Newtonian liquid with the same viscosity, $\eta$.
250: The ratio $Z/Z_0$ is constant equal to unity in the linear regime at low $Q$. At $Q$ of about 8.5
251: nl/s, however, a non-linear transition occurs; $Z/Z_0$ starts to grow and reaches a factor of about
252: 2.8 at high $Q$. The Reynolds number for the channel flow can be defined as $Re=Q\rho/(\eta d)$ . It
253: was 0.017 at the transition point and 0.14 at highest $Q$, that we tried, so that the inertial effects
254: were always negligible. The Weissenberg number can be defined as $Wi=4\lambda Q/d^3$, and its
255: estimated value at the transition point is 1.4, which is comparable with $Wi$ found at purely elastic
256: transitions in macroscopic set-ups\cite{18,19,21,29}.
257:
258: In order to get detailed information about structure of the flow above the non-linear transition we
259: used solution 2 with high viscosity and large $\lambda$. The polymer relaxation time defines both
260: characteristic time of changes in the flow and the inverse of flow velocity in the elastic non-linear
261: regime\cite{15,29}. Therefore, the non-linear flow in solution 2 was expected to be much slower, and
262: measurements of its characteristics were expected to be more feasible with the standard video
263: microscopy techniques. Using the micro-PIV we measured flow velocity in the middle of the curvilinear
264: channel. Dependence of RMS of fluctuations of the longitudinal component of the flow velocity,
265: $V_1^{rms}$, (which is the velocity component along the main flow) on $\Delta P$ is shown in Fig.3.
266: The fluctuations are virtually absent in the linear regime at low pressure. At $\Delta P$ of about 50
267: Pa, however, $V_1$ starts to fluctuate, and $V_1^{rms}$ begins to grow quickly and non-linearly. It
268: can be learned from the inset in Fig.3 that the longitudinal component of the flow velocity,
269: $\bar{V_1}$, grows linearly at low $\Delta P$, but its growth slows down at the same critical $\Delta
270: P$ of about 50 Pa, which is another evidence for a non-linear elastic flow transition taking place. At
271: the transition point, the average longitudinal flow velocity, estimated from the micro-PIV
272: measurements in different points across the channel, was $\bar{V_1}\approx 80 \mu$m/s. That gives
273: estimates of $Q\approx 0.8$ nl/s, $Re\approx 8\cdot 10^{-5}$ and $Wi\approx 3.5$ for the elastic
274: non-linear transition. This value of the Weissenberg number is very close to $Wi_c\approx 3.2$ found
275: for the transition to chaotic flow in the table-top curvilinear channel in Ref.\cite{21}
276:
277: A typical time series of $\bar{V_1}$ above the transition is shown in Fig.4a. The velocity is
278: strongly fluctuating and its time dependence has well expressed chaotic appearance. The chaotic
279: character of the velocity fluctuations is confirmed by analysis of its time correlations. The velocity
280: autocorrelation function shown in Fig. 4b does not have distinct peaks and decays uniformly.
281:
282: For the experiments on mixing the design of the micro-channel was slightly modified to enable side by
283: side injection of two streams of solution 2, one with and one without FITCD, to the channel inlet,
284: Fig.5a. Apart from $c_0=280$ ppm by weight of FITCD added to one of them the polymer solutions were
285: identical and were injected at equal flow rates by careful adjustment of the driving pressures, Fig.
286: 5a. The set-up was first tested with the plain solvents without PAAm added. The flow appeared laminar
287: at all $\Delta P$ that we applied, and the interface between the streams with and without FITCD
288: remained smooth and sharp along the whole channel with only minor smearing by diffusion, Fig. 5b.
289:
290: The situation was similar with the polymer solutions in the linear regime at low $\Delta P$. However,
291: when the driving pressure was raised above the non-linear transition threshold, fluctuating flow
292: velocity produced significant stirring and complex and chaotically changing tracer concentration
293: profiles, Fig. 5c, d. We studied mixing in the channel in detail at $\Delta P=134$ Pa, corresponding
294: to a flow rate about twice above the non-linear transition threshold (cf. Ref.\cite{21}) and
295: $\bar{V_1}$ of about 173 $\mu$m/s. Variation of the tracer concentration profiles with time at
296: different distances from the inlet is illustrated by the space-time plots in Fig. 6a,b. One can
297: observe that the tracer concentration appears to fluctuate quite randomly without any apparent scale
298: in time or space. Next, one can see in Fig. 6a, taken at $N=12$, that the left side of the channel,
299: where the tracer was initially injected, looks much brighter and has much higher average concentration
300: of the tracer. Although also noticeable in Fig.6b taken further downstream, at $N=18$, this feature is
301: clearly weaker there. Thus, stirring by the fluctuating velocity field seems to create a more
302: symmetric distribution of the tracer between the two sides of the channel. In order to validate this
303: observation, we measured time averages of the tracer concentration, $\bar{c}$, at different positions
304: across the channel and at different $N$, Fig.7a. One can see that the cross-channel distribution of
305: $\bar{c}/c_0$ close to the inlet, at $N=7$, is strongly influenced by the asymmetric conditions at the
306: channel entrance. As one can learn from the curve at $N=11$, however, the imprint of the initial
307: conditions is clearly fading as the liquid advances downstream and being stirred. Further downstream,
308: at $N=41$, asymmetry in the tracer distribution introduced by the initial conditions disappears
309: completely. Fading of the initial condition influence with time and restoration of symmetry in flow in
310: statistical sense are both distinct features of chaotic and turbulent flows. Therefore, the curves in
311: Fig. 7a provide further evidence for truly chaotic nature of the flow in the micro-channel.
312:
313: A natural parameter characterizing inhomogeneity of the mixture is standard deviation of the
314: instantaneous local tracer concentration, $c$, from its overall average value $\langle c\rangle=c_0$.
315: It is convenient
316: to divide it by $\langle c\rangle$ and to introduce a dimensionless standard deviation $c_{std}=\sqrt{\langle
317: (c-\langle c\rangle)^2\rangle}/\langle c\rangle$. At the channel entrance $c_{std}$
318: is equal to unity, and it becomes zero, when the liquid is perfectly mixed and homogeneous.
319: Dependence of $c_{std}$ on $N$ for two tracers with diffusion coefficients of $D_1=6.6\cdot 10^{-9}$ and
320: $D_2=5.4\cdot 10^{-10}$
321: cm2/s is shown in Fig.7b in semi-logarithmic scale. One can see that the mixture becomes increasingly
322: homogeneous as the liquid advances downstream, and the parameter $c_{std}$ decays exponentially with
323: $N$ for the both tracers. An exponential decay was also found in the case of elastic turbulence in a
324: macro-channel \cite{21}, and it agrees very well with theoretical predictions for the so-called
325: Batchelor regime\cite{30,31,32,33} of mixing. The latter corresponds to a flow, which is chaotic in
326: time but essentially "smooth" in space, in the sense that small eddies are rare, and the main
327: contribution to mixing comes from the largest eddies having the size of the whole system.
328:
329: The rates of the exponential decay for the two dependencies shown in Fig. 7b were 0.217 and 0.137,
330: corresponding to $\Delta N$ of 4.61 and 7.30, and lengths $L_1=4.34$ mm and $L_2=6.88$ mm,
331: respectively. The latter can be considered as characteristic mixing distances along the channel for
332: the two tracers. Characteristic mixing times, which can be estimated as $t_{mix}=L/\bar{V}$, are then
333: found to be $t_{mix,1}=25$ s and $t_{mix,2}=40$ s, respectively. In the absence of active stirring,
334: mixing would only occur through molecular diffusion, with characteristic diffusion times across the
335: channel given by $t_{diff}=d^2/D$ resulting in $t_{diff,1}=1.5\cdot 10^4$ s and $t_{diff,2}=1.9\cdot
336: 10^5$ s, respectively. Therefore, the stirring produced by the chaotic flow in the channel reduces the
337: mixing times for the FITCD macromolecules by three to four orders of magnitude.
338:
339: \section{ Discussion.}
340:
341: We studied flow of two dilute polymer solutions in a microscopic curvilinear channel. We observed a
342: non-linear transition to occur in the flow at $Re\approx 1.7\cdot 10^{-2}$, $Wi\approx 1.4$, and
343: $Re\approx 8\cdot 10^{-5}$, $Wi\approx 3.5$, for solution 1 and solution 2, respectively. The very
344: small and different values of $Re$, and comparable, order of unity values of $Wi$ at the transition
345: threshold for the two solutions indicate that the transition is of purely elastic nature and can occur
346: at arbitrarily low $Re$. A possible explanation of the difference in estimated critical $Wi$ for the
347: two polymer solutions is that the actual polymer relaxation time of solution 1 may be significantly
348: higher than the estimate based on the assumption that $\lambda\sim\eta_s$.\\
349: A major global feature of
350: the flow above the non-linear elastic transition is fast growth of the resistance, Fig.2. It increases
351: by up to a factor of 2.8 above the resistance for a Newtonian fluid with the same shear flow
352: viscosity, $\eta$, at the same $Q$. Since $Re$ is very low, the whole growth of the flow resistance
353: should be due to increase in the elastic polymer stresses\cite{20}. In a pure shear flow with
354: $\dot{\gamma}=240$ s$^{-1}$, corresponding to $Q=60$ nl/s (and $Wi\approx 10$), viscosity ratio for
355: polymer solution 1 was $\eta/\eta_s=1.22$. Hence, the average increase in the polymer shear stresses
356: due to the secondary flow at $Q=60$ nl/s (Fig.2) can be estimated as a factor of about 11. Suggesting
357: linear deformation of the flexible polymer molecules in a shear flow at low to moderate $Wi$\cite{34},
358: we can estimate extension of the molecules at $Wi=10$ as $10\cdot 0.5 \mu$m = $5 \mu$m. (Flexible
359: polymer molecules were found to extend linearly with the shear rate, until the extension reached
360: 15-20\% of their full length\cite{34}.) Additional extension of the polymers due to the irregular
361: secondary flow, as suggested by the stress growth by the factor of $11$, can be estimated as a factor
362: of $\sqrt{11}$ (suggesting that the secondary flow results in an isotropic polymer unravelling and the
363: stress grows as a square of the polymer extension\cite{15}). That brings characteristic size of the
364: extended polymer molecules to $20\mu$m range, less than an order of magnitude smaller than the
365: diameter of the channel, $d=100\mu$m.
366:
367: The growth of the elastic stresses due to the fluctuating secondary flow can only occur as a result
368: of significant reorganization of the flow structure and spontaneous generation of regions with strong
369: extensional flow, as in the case of the elastic turbulence\cite{19,20}. (In shear flows polymer contribution
370: to resistance increases slower than linearly with the shear rate, which is called shear thinning.)
371: This suggestion is quite corroborated by the direct flow velocity measurements in solution 2 above the
372: non-linear transition. Velocity is found to be strongly fluctuating, Figs.3,4, with RMS of the
373: fluctuations reaching as much as 10\% of the mean longitudinal velocity at the center of the channel,
374: Fig.3, just as in the table-top channel in Ref.\cite{21}. The velocity appears to vary randomly in time,
375: Fig.4a, and its auto-correlation function decays quite quickly and does not have any distinct peaks,
376: Fig.4b. These all are clear indications of chaotic nature of the flow in the channel in the
377: non-linear regime at high $Wi$. Another evidence of chaos comes from the experiments on mixing,
378: Figs.6,7. Asymmetry in distribution of $\bar{c}/c_0$ across the channel, which is imposed by the
379: conditions
380: at the channel entrance, decays with the distance from the entrance, Fig7a. Further, $c_{std}$
381: exponentially
382: decays with $N$, as it is supposed to be in the chaotic Batchelor regime of mixing\cite{21,32,33}.
383:
384: Experimental results presented in Figs.2-4 and Figs.6,7 and discussed above are fully consistent with
385: a suggestion that the regime of elastic turbulence\cite{19} is being realized in the micro-channel. In
386: fact, the results in Figs.2-4 and Figs.6,7 agree rather well with the measurements in macroscopic
387: systems reported before \cite{19,20}. We find it rather remarkable that although the size of extended
388: molecules may become comparable with the channel diameter, it does not seem to cause any significant
389: new effects in the polymer solution dynamics. An essential feature of turbulent flows is a broad range
390: of spatial scales, at which fluid motion is excited. Unfortunately, limited resolution of the
391: micro-PIV technique did not allow us to explore properties of the flow down to sufficiently small
392: spatial scales and to obtain conclusive data about its spatial structure. Therefore we can only refer
393: to the flow in the micro-channel as being chaotic.
394:
395: In the experiments with passive tracers, we found that stirring by the chaotic flow results in
396: efficient mixing in the channel, Figs.5-7. So, mixing times for the macromolecules, which we used as
397: tracers, were reduced by three to four orders of magnitude compared with passive molecular diffusion,
398: Fig.7b. It is worth noting that the characteristic distance of $\Delta N=7.3$ required for mixing of
399: low diffusivity FITCD is comparable with $\Delta N\approx 15$ found in the chaotic flow in the
400: table-top channel Ref.\cite{21}. We further notice that whereas the diffusion time scaled as $1/D$,
401: the time of mixing in the chaotic flow depended on $D$ very
402: weakly. It increased by only 60\% between the lower and upper curve in Fig.7b (from $t_{mix,1}=25$ s
403: and $t_{mix,2}=40$ s),
404: whereas $D$ dropped by a factor of 12 (from $D_1=6.6\cdot 10^{-9}$ to $D_2=5.4\cdot 10^{-10}$ cm2/s).
405: The weak dependence
406: of $t_{mix}$ on $D$ is quite consistent with theoretical predictions for chaotic flows and the
407: Batchelor regime
408: \cite{30,31,32,33}. It implies that the elasticity induced chaotic stirring in the micro-channel can be
409: efficiently used for mixing of liquids with additives of any low diffusivity, such as large molecules of
410: DNA, viruses, particles, and possibly living cells.
411:
412: The relatively long times of mixing in the chaotic flow of solution 2 are due to high viscosity of the
413: solvent, large $\lambda$ and low flow velocity at the elastic instability threshold. In a much more
414: practically relevant case of water based solutions with the viscosity on the order of 0.001 Pas,
415: $\lambda$ is expected to be about hundred times lower and the flow velocities in the chaotic regime
416: should be about hundred times higher\cite{15,29}. Suggesting qualitatively similar flow
417: conditions\cite{29}, we expect characteristic mixing times for the water based solutions to be on 100
418: ms scale. Since diffusion coefficients for macromolecules are proportional to $1/\eta_s$\cite{15}, the
419: ratio between $t_{diff}$ and $t_{mix}$ for the chaotic flow in the water based solutions should be in
420: the same range of $10^3$ to $10^4$.
421:
422: The characteristic mixing length along the channel for water based solutions should also remain in the
423: same range of about 4-7 mm. It is about 2-3 times shorter than characteristic lengths in the
424: "herring-bone" patterned channel of Ref.\cite{14} at comparable values of Peclet number, $Pe=Vd/D$.
425: Although the method of mixing by elasticity induced chaotic flow necessarily requires addition of
426: polymers to the working liquid, it may be quite practical for many biochemical assays, taking into
427: account the very low concentration of the high molecular weight polymers used. It does not rely on any
428: special patterning of the micro-channels and should be readily compatible with rounded channel
429: profiles used for integrated micro-valves and peristaltic pumps\cite{35}, allowing efficient mixing in
430: closed loop microscopic flows\cite{36}.\\
431:
432: We are grateful to M. Chertkov and V. Lebedev for many useful and illuminating discussions and to D.
433: Mahalu for valuable help with the microfabrication. One of us (T. B.) thanks V. Kiss for his support
434: in the confocal set-up measurements. This work is partially supported by an Israel Science Foundation
435: grant, Binational US-Israel Foundation, and by the Minerva Center for Nonlinear Physics of Complex
436: Systems.
437:
438: \begin{references}
439:
440: \bibitem{1} { Y. N. Xia, G.M. Whitesides, {\sl Ann. Rev. Material Sc.} {\bf28}, 153(1998).}
441: \bibitem{2} {G. M. Whitesides, A.D. Stroock, {\sl Physics Today} {\bf 54}(6), 42 (2001). }
442: \bibitem{3} { C. L. Hansen, E. Skordalakes, J.M. Berger, \& S.R. Quake, {\sl PNAS} {\bf 99}, 16531 (2002).}
443: \bibitem{4}{ N. L. Jeon, H. Baskaran, S.K.W. Dertinger, G.M. Whitesides, L. Van de Wate, \& M. Toner,
444: {\sl Nature Biotech.} {\bf 20}, 826 ( 2002).}
445: \bibitem{5}{S. Takayama, E. Ostuni, P. LeDuc, K. Naruse, D.E. Ingber, \& G.M. Whitesides,
446: {\sl Nature} {\bf 411}, 1016 (2001).}
447: \bibitem{6}{ H. B. Mao, P.S. Cremer, \& M.D. Manson, {\sl PNAS} {\bf 100}, 5449 (2003).}
448: \bibitem{7}{ H. P. Chou, C. Spencer, A. Scherer, \& S. Quake, {\sl PNAS} {\bf 96}, 11 (1999).}
449: \bibitem{8}{ L. D. Landau, \& E.M. Lifshitz, {\sl Fluid Mechanics} (Pergamon Press)(1987).}
450: \bibitem{9}{ L. Reyes, J. Bert, J. Fornazero, R. Cohen, \& L. Heinrich, {\sl Colloids and
451: surfaces} {\bf B 25}, 99 (2002). }
452: \bibitem{10}{ M. H. Oddy, J. G. Santiago, \& J. C. Mikkelsen, {\sl Anal. Chem.} {\bf 73}, 5822 (2001).}
453: \bibitem{11}{J. H. Tsai, \& L.W. Lin, {\sl Sens. Actuat.} {\bf A 97}, 665 (2002).}
454: \bibitem{12}{ D. Therriault, S.R. White, \& J.A. Lewis, {\sl Nature Materials} {\bf 2}, 265 (2003).}
455: \bibitem{13}{ R. A. Vijayendran, K.M. Motsegood, D.J. Beebe, \& D.E.Leckband, {\sl Langmuir } {\bf 19},
456: 1824 (2003).}
457: \bibitem{14}{ A. D. Stroock, S. K. Dertinger, A. Ajdari, I. Mezic, H. A. Stone, \&
458: G. M. Whitesides, {\sl Science} {\bf 295}, 647 (2002).}
459: \bibitem{15}{ R. B. Bird, Ch. Curtiss, R.C. Armstrong, \& O. Hassager {\sl Dynamics of Polymeric
460: Liquids}, (Wiley, New York)(1987).}
461: \bibitem{16}{ V. Tirtaatmadja \& T. Sridhar, {\sl J. Rheology} {\bf 37}, 1081 (1993).}
462: \bibitem{17}{ D. V. Boger, K. Walter, {\sl Rheological phenomena in focus} (Elsevier) (1993). }
463: \bibitem{18}{ R. G. Larson, E.S.G. Shaqfeh, \& S.J. Muller, {\sl J. Fluid Mech.} {\bf 218}, 573 (1990).}
464: \bibitem{19}{ A. Groisman, \& V. Steinberg, {\sl Nature} {\bf 405}, 53 (2000).}
465: \bibitem{20}{ A. Groisman, \& V. Steinberg, {\sl Phys. Rev. Lett.} {\bf 86}, 934 (2001).}
466: \bibitem{21}{ A. Groisman, \& V. Steinberg, {\sl Nature} {\bf 410}, 905 (2001).}
467: \bibitem{22}{ A. Groisman, M. Enzelberger, \& S.R. Quake, {\sl Science} {\bf 300}, 955 (2003).}
468: \bibitem{23}{P. G. DeGennes, {\sl J. Chem. Phys.} {\bf 60}, 5030 (1974). }
469: \bibitem{24}{ T. T. Perkins, D.E. Smith, \& S. Chu, {\sl Science} {\bf 276}, 2016 (1997).}
470: \bibitem{25}{W.-M. Kulicke, M. Kotter, \& H. Grager, in {\sl Advances in Polymer Science 89}, {\sl
471: Polymer characterization/Polymer solutions}, (Springer-Verlag, Berlin), 1989.}
472: \bibitem{26}{ P.J. Shrewbury, S. J. Muller, \& D. Liepmann, {\sl Biomedical Microdevices} {\bf 3}, 225 (2001).}
473: \bibitem{27}{ D.E. Smith, T. Perkins, \& S. Chu, {\sl Macromolecules} {\bf 29}, 1372 (1996).}
474: \bibitem{28}{ C. Wu, {\sl Macromolecules} {\bf 26}, 3821 (1993).}
475: \bibitem{29}{ A. Groisman, \& V. Steinberg, {\sl Europhys. Lett.} {\bf 43}, 165 (1998).}
476: \bibitem{30}{G. K. Batchelor, {\sl J. Fluid Mech.} {\bf 5}, 113 (1959).}
477: \bibitem{31}{ M. Chertkov, G. Falkovich, I. Kolokolov, \& V. Lebedev, {\sl Phys. Rev. E} {\bf 51}, 5609 1995).}
478: \bibitem{32}{T. D. Son, {\sl Phys. Rev E} {\bf 59}, R3811 (1999).}
479: \bibitem{33}{ E. Balkovsky, \& A. Fouxon, {\sl Phys. Rev E} {\bf 60}, 4164 (1999).}
480: \bibitem{34}{ D.E. Smith, H. P. Babcock, \& S. Chu, {\sl Science} {\bf 283}, 1724 (1999).}
481: \bibitem{35}{ M. A. Unger, H.P. Chou, T. Thorsen, A. Scherer, \& S.R. Quake, {\sl Science}
482: {\bf 288}, 113 (2000).}
483: \bibitem{36}{H. P. Chou, M.A. Unger, \& S.R. Quake, {\sl Biomedical Microdevices} {\bf 3}, 323 (2001).}
484: \end{references}
485:
486: \begin{figure}
487:
488: \caption{(A) Photograph of the micro-fluidic device. The micro-channels were filled with ink for
489: better contrast. (B) Magnified image of a section of the functional curvilinear channel.}
490:
491: \label{figa}
492: \end{figure}
493:
494: \begin{figure}
495:
496: \caption{Dependence of normalized resistance, $Z/Z_0$, in flow of solution 1 through the curvilinear
497: channel on the volumetric flow rate, $Q$, in semi-logarithmic scale.}
498:
499: \label{figb}
500: \end{figure}
501:
502: \begin{figure}
503:
504: \caption{Dependence of RMS of fluctuations of the longitudinal component of flow velocity, $V_1^{rms}$
505: , on pressure drop per segment, $\Delta P$, in the center of the micro-channel for solution 2. Inset:
506: Dependence of time average of the longitudinal flow velocity, $\bar{V_1}$ , in the center of the
507: micro-channel on $\Delta P$.}
508:
509: \label{figc}
510: \end{figure}
511:
512: \begin{figure}
513:
514: \caption{(A) Time series of the longitudinal flow velocity, $V_1$ , in the center of the micro-channel
515: at $\Delta P=100$Pa. (B) Auto-correlation function for $V_1$ based on about 6000 individual velocity
516: measurements. }
517:
518: \label{figd}
519: \end{figure}
520:
521: \begin{figure}
522:
523: \caption{ (A) Epi-fluorescent micro-photograph of the entrance area of a micro-channel used in
524: experiments on mixing. Wide triangular region in front of a curvilinear channel allows to adjust equal
525: flow rates for polymer solutions with (from below) and without FITCD. (B) Confocal photograph of flow
526: in the micro-channel without polymers added. Right wall of the channel is shown by a dotted line from
527: below. (C) Confocal image of mixing in chaotic flow in the micro-channel with solution 2 in it. (D)
528: Confocal image of cross-section of the micro-channel with chaotic flow in solution 2.}
529:
530: \label{fige}
531: \end{figure}
532:
533: \begin{figure}
534:
535: \caption{ Space-time plots of FITCD distribution across the channel taken at (A) $N=12$ and (B)
536: $N=18$. Confocal scanning was done along the same line across the channel in the mid-plane at equal
537: distances from half-ring interconnections, with even time intervals of 0.0177 sec. Profiles of FITCD
538: concentration in consecutive moments of time are plotted from top to bottom.}
539:
540: \label{figf}
541: \end{figure}
542:
543: \begin{figure}
544:
545: \caption{(A) Time average of FITCD concentration, $\bar{c}$ , divided by $c_0$, as a function of
546: position, $x$, along a line across the channel taken at different distances from the inlet: $N=7$,
547: $N=11$, and $N=41$. The lines across the channel were in the mid-plane at equal distances from
548: half-ring interconnections, just as for the space-time plots in Fig.6. (B) Standard deviation of FITCD
549: concentration from its average value, $c_{std}$, as a function of distance, $N$, from the inlet for
550: two kinds of FITCD with average molecular weights of 10 kDa (triangles) and 2 MDa (squares).}
551:
552: \label{figg}
553: \end{figure}
554:
555:
556: \end{multicols}{2}
557:
558:
559: \end{document}
560: