PoiseuilleFlowofTwoImmiscibleFluidsBetweenFlatPlateswithApplicationsto

Microfluidics

BruceA.Finlayson
ProfessorEmeritusofChemicalEngineering
UniversityofWashington
Seattle,WA981951750

Abstract

An analytic solution is derived for fully developed flow of two immiscible fluids
betweentwoflatplateswhenthefluidshavearbitraryviscosityratiosandarbitrary
flowrateratios.ThisgeneralizesthesolutioninBird,etal.1whenthetwofluidstake
up exactly the same space (i.e. the thickness of each fluid is onehalf the total
thickness).Thissolutionisuseful,forexample,incertainmicrofluidicdeviceswhen
twofluidsflowinlaminarcontact;knowingthepositionofthedividingstreamlineis
critically important to accurate modeling of diffusion across the interface. Since
diffusion is often slow in comparison to the channel transit time, an approximate
expression is provided to identify the limits of diffusion about the dividing
streamline. Thus, it is possible to obtain a priori a good estimate of the
concentrationbandasthefluidsmovedownthedevice.

1.AnalyticalSolution

Fully developed flow of two immiscible fluids between two flat plates is solved in
Bird,etal.1whenthetwofluidstakeupexactlythesamespace(i.e.thethicknessof
eachfluidisonehalfthetotalthickness).Thatsolutionisgeneralizedheretoallow
any fraction, which ultimately will depend upon the flow rate ratio. The notation
follows that of Bird, et al.1 and the domain is illustrated in Figure 1. Let the total
distancebetweenthetwoflatplatesbeH,andthefractionfilledwiththelowerfluid
(I)bef.Thedomainstretchesfromx=fHtox=(1f)H,i.e.theinterfaceisatx=0.
Theviscositiesofthefluidsare1and2,andtheflowratesareQ1andQ2.

Figure1.Flowoftwoimmisciblefluidsbetweenparallelplates

Thestressprofilesare

1 =

p
p
x + C1, 2 =
x + C1
L
L

The constant C1 is the same in both fluids since the stresses are the same at the
interface,x=0.ForNewtonianfluidswehave

p
dv
dv p
dv
= , or 1 1 =
x + C1, and 2 2 =
x + C1
dx
dx
L
L
dx
Theseequationscanbeintegratedtogive

v1 =

p 2 C1
p 2 C1
x x + C2 , v 2 =
x x + C2
1
2
21L
22 L

Theboundaryconditionsrequire
v1 ( fH) = 0, v 2 [(1 f )H] = 0

ApplyingtheseconditionstothevelocitiesandsolvingforC1andC2gives
f (1 f )
pH 2
pH 2 f 2 1 (1 f ) 2

C1 =
, C2 =
2L f2 + (1 f )1
2L f2 + (1 f )1

Theflowratesare
(1 f )H

Q1 =

v1dx, Q2 =

fH

Q1 =

Q2 =

v dx
2

p
C
( fH) 3 + 1 ( fH) 2 + C2 ( fH)
21
61L

pH 3
C H2
(1 f ) 3 1 (1 f ) 2 + C2 H(1 f )
62 L
22

Onecanspecifythefraction,f,andcalculatetheratioofflowratesdirectly.Whenf=
0.5itgivesthesolutioninBird,etal.Alternatively,theflowrateratio,R=Q1/Q2,can
and the value of f is determined to meet this ratio. When the two
be specified,
viscosities are equal, this should give the equation for a fully developed flow
betweentwoparallelplates,whichitdoes.AMATLABprogramsisusedtosolvefor
fwhengivenR=Q1/Q2.

While this solution is for immiscible fluids, it may be a good approximation for
misciblefluidsinamicrofluidicchannelwhentheReynoldsnumberissmall,which
itusuallyis,sincethennorecirculationorinstabilitieswillbepresent.Onerelevant
situationisthelaminardiffusioninterface,whichisestablishedwhentwofluidsof
different chemical compositions are brought into laminar contact. Since they have

differentconcentrations,diffusiontakesplaceperpendiculartotheinterfaceasthe
fluidsmovedownthechannel.Whilethecompletesolutiontotheprobleminvolves
solving the convective diffusion equation, and including the possible effect of
concentration on viscosity, there are situations in which an approximation is
feasible.Ifonefluidcontainsasubstancethatincreasesitsviscosityrelativetothe
other fluid, but does not diffuse appreciably, then the viscosities of both fluids
remain essentially constant as they move down the channel. Other chemicals,
though,maydiffusionmorerapidly.Forexample,2ifonefluidisserumandtheother
is water, one fluid (serum) contains proteins (albumin representing the most
concentrated among them), that diffuses slowly, and metabolites (creatinine being
oneofmany),whichdiffusesmorerapidly.Theproteinsaffecttheviscosity,butthe
creatinine hardly does. In order to make use of such a channel to separate
metabolites from proteins, one would like to know how far the albumin diffuses
sidewaysastheflowproceedsdownthechannel,giventheflowratesforserumand
water. This modeling capability would then permit identification of flow rates for
which the ratio of metabolite to protein is particularly high. This is still an
approximation, since there is some axial diffusion, but it is a good approximation
andcangiveguidanceinthedesignandoperationofdevices.

Dz

Figure2.Diffusionproblemarounddividingstreamline

Thus,considerthefollowingproblemfordiffusionaboutthedividingstreamlineatx
= f, as sketched in Figure 2. Now the coordinate x is taken from 0 to H and the
dividingstreamlineisatx=fH.

c
2c c
= D 2 , (,z) = 0

z
x x
c(x,0) =1 for x f H,= 0 for x > f H
v( f )

The velocity is the velocity at the dividing streamline; this assumption can be

removedbyusinganumericalsolution.Withthisassumption,thesolutioncentered
onx=fHis

c = 0.5erfc(),

x f H

Dz
4
v( f )

The thickness of the diffusion layer is taken as the distance over which the
concentration falls from 0.5 to 0.01 (2% criterion). This happens when the

=1.645.
complementaryerrorfunctionis0.01/0.5=0.02,oratanargumentof

= 10.8

Dz

v( f )

Thissolutionobviouslycannotbeusedoncethediffusionthicknessreachesoneof
thewalls.Thisisthethicknessthatthefluiddiffusesfromonefluidtoanotherina
3

distancezdownstream.Anapproximatesolutiontothisproblemis
c = 0.5(1 2 ), =

x f H
Dz

, = 12

v( f )

Eitherofthesefunctionscanbeusedtoestimatethediffusionthicknessaboveand
below the dividing streamline under the stated assumptions. Other numerical

experiencehasshownthattheseassumptionsarequitegoodinrealisticcases.Inthe
caseofalbuminandcreatinine,inalengthof2cmwithavelocityof2mm/sec,the
creatinine diffusion length (D = 9.2 1010 m2/s) was 315 m, but the albumin
diffusion length (D = 6.7 1011 m2/s) was 85 m. The microfluidic device thus
providedonestreamthathadverylittlealbumininitprovideditwasobtainedata
distancefH+abovethebottom.
2.NumericalResults
Figure 3 shows the dimensionless velocity profiles when the flow rate ratios vary
from0.2to5andtheviscosityratiosis5.Thecoordinatexisdimensionless,scaled
by the height, H, with the dividing streamline at x = 0. The velocity is scaled such
thatpH2/2L=1.Asexpected,whentheviscositiesaredifferent,theslopesofthe
velocity profiles are different at the dividing streamline. Of more interest is the
difference between the flow rate ratios and the physical location of the actual
streamline. Figure 4 shows the difference between the f predicted by the analysis
and the value of f that would be predicted based upon the ratio of flow rates,
Q1/(Q1+Q2). As can be seen, there is considerable difference in those values when
theflowrateratioissmall,regardlessoftheviscosityratio,upto100%differencein
some cases. Thus, the ratio of flow rates does not give a good estimate of the
locationofthedividingstreamline.
Consider next a numerical solution of the diffusion problem given above with v(f)
replacedbytheactualvelocity,scaledsothatthestreamlinevelocityis2mm/sec.
The finite difference method4 is used to solve the convective diffusion equation
whentheheightis380m,thediffusivitiesareasgiven,theviscosityratiois5and

the flow ratio is 1.0. In this case the f = 0.5735. Figure 5 shows the numerical
solutionattheexitusingxscaledbyHandwiththedividingstreamlineatx=f.The
velocityprofilehasadramaticeffectontherapidlydiffusingspecies(creatinine)but
almost no effect on the albumin because it diffuses so slowly. If one takes the
diffusionlayerasthedistancebetweenthestreamlinevalueandthepointwherec=
0.01or0.99,thediffusionlayerforalbuminis85mononesideand95monthe
other side. These numbers are comparable to the estimates given above but show
theeffectofthevelocityprofileoneachsideofthedividingstreamline.
Simulations for twodimensional and threedimensional configurations, including
entryandexitgeometries,areavailableelsewhere.5

Figure3.Velocityprofilefortwoimmisciblefluidsflowwhen 1/ 2=5;all
curvesareforthesamepressuredropperlength


Figure4.Fractionoftheheightshowinglocationofdividingstreamline;solid
lineistheratioofflowrates,Q1/(Q1+Q2)


Figure5.Numericalsolutionattheexit;viscosityratio=5,flowrateratio=1;
(a,b)creatininewithavelocityprofile(a)andconstantvelocity(b);
(c,d)albuminwithavelocityprofileandconstantvelocity

3. Conclusions

Ananalyticalsolutionisderivedfortheflowoftwoimmisciblefluidsbetweentwo
flat plates. It is valid for any ratio of viscosities and any ratio of flow rates, thus
generalizingtheresultsofBird,etal.1Whencombinedwithananalyticalsolutionof
diffusion in the transverse direction, it can be used to estimate the thickness of
diffusionlayerswhenthetwofluidshavedifferentchemicalcompositions.

4. References

1. Bird, R. B., W. E. Stewart, E. N. Lightfoot, Transport Phenomena, 2nd ed., Wiley

(2002),pp.5668.

2.Shaw,A.,B.Schattka,M.Alexander,S.LowYing,A.Man,Metabolicfingerprinting
ofbiofluidsbyinfraredspectroscopy:Modelingandoptimizationofflowratesfor
laminarfluiddiffusioninterfacesamplepreconditioning,submitted.

3. Finlayson, B. A., The Method of Weighted Residuals and Variational Principles,

AcademicPress(1972),p.45.
4.Finlayson,B.A.,NonlinearAnalysisinChemicalEngineering,McGrawHill(1980).
5. Finlayson, B. A., R. A. Shaw, Modeling Microfluidic Separations using Comsol
Multiphysics,ComsolConference2010,Boston.

5. Acknowledgment

The author would like to thank R. Anthony Shaw, National Research Council of
Canada, Institute for Biodiagnostics, Winnipeg, Manitoba, Canada, for introducing
himtothisproblem.