You are on page 1of 9

Journalof

Non-Newtoain
Fluid
Mechanics
ELSEVIER J. Non-Newtonian Fluid Mech., 59 (1995) 93 101

Short communication

On the mechanism of drag reduction


Michael R e n a r d y
Department of Mathematics, Virginia Tech, Blacksburg, VA 24061-0123, USA
Received 17th October 1994; in revised form 15 Febuary 1995

Abstract

Recent studies indicate that growing streamwise perturbations driven by streamwise


vortices and the resulting Reynolds stresses play a significant role in both the transition to
turbulence and the near-wall turbulent flow. In this note, we show that viscoelasticity
introduces a negative feedback, so that the streamwise flow driven by a streamwise vortex
would have the effect of diminishing the vortex. The terms responsible for this effect are the
same as these that lead to a second normal stress difference.

Keywords: Drag reduction; Turbulence

I. Introduction

It is well known that viscous shear flows develop instabilities at Reynolds


numbers well below those where linear stability analysis would predict them. While
nonlinearities are important in sustaining those instabilities, there are no initial
disturbances of sufficient magnitude that nonlinear terms would dominate over
linear ones, and therefore linear effects are largely responsible for the initial growth
of disturbances. Several recent works have elucidated this mechanism, which
involves the interaction of Orr-Sommerfeld and Squire modes [1-3]. Recall that
Squire's transformation transforms the three-dimensional linear stability problem
for parallel shear flow to the two-dimensional Orr-Sommerfeld equation. This
two-dimensional equation is for a velocity field which is in the plane spanned by the
wave vector of the disturbance and the wall-normal direction. We refer to the part
of the velocity field governed by the Orr-Sommerfeld equation as the Orr Som-

0377-0257/95/$09.50 1995- Elsevier Science B.V. All rights reserved


SSDI 0377-0257(95)013 57-1
94 M. Renard)' /J. Non-Newtonian Fluid Mech. 59 (1995) 93 101

merfeld mode. The third velocity component, perperdicular to the disturbance wave
vector, is governed by a separate equation, which is known to produce only stable
eigenvalues and is therefore often ignored in the study of linear stability. We refer
to this velocity component as the Squire mode.
At the root of the studies in Refs. [1 3] lies the observation that eigenvalues alone
can give a quite misleading picture of linear dynamics. This is because exponential
decay as time goes to infinity can mask a large transient growth at finite times, in
cases where the underlying linear operator is far from normal. In this context, it is
important that there is a coupling term between Orr-Sommerfeld and Squire modes;
the Orr Sommerfeld mode appears as a forcing term in the equation for the Squire
mode. This interaction vanishes for streamwise wave vectors and is largest for
spanwise wave vectors. Even though the interaction term has no effect on eigenval-
ues, it leads to linear growth of the Squire mode in the inviscid case [4,5]. For large
Reynolds numbers, this linear growth is still present over a large time interval and
it leads to amplification of initial disturbances by a factor on the order of the
Reynolds number. Essentially, the disturbances most affected involve a streamwise
vortex (this is the Orr Sommerfeld mode), which then drives a streamwise velocity
perturbation (the Squire mode) by moving high-speed fluid inward on one side
("sweep") and low-speed fluid outward on the other side ("ejection").
Some recent studies on turbulent boundary layers indicate that streamwise or
"hairpin" vortices continue to play a significant role after the transition to
turbulence has occurred. The ejection and sweep motions generated by streamwise
vortices are believed to contribute significantly to the Reynolds stress. Pairs of
streamwise vortices generate low-speed streaks, which are one of the principal
features seen in flow visualizations, and these low-speed streaks eventually lead to
"bursting" (see Ref. [6] for a review). Even though the evolution of hairpin vortices
and in particular their breakup involves highly nonlinear effects, the production of
drag by streamwise vortices can still be understood in terms of the linear mechanism
outlined above.
It is a consequence of Squire's transformation that the coupling between O r r -
Sommerfeld and Squire modes is in one direction only. The Orr-Sommerfeld mode
drives the Squire mode, but there is no coupling in the other direction. It is precisely
this feature which leads to the linear growth. Viscoelastic fluids, on the other hand,
do not necessarily have a Squire's transformation [7]. At the second-order fluid level,
the terms destroying Squire's transformation are the same as those that lead to a
second normal stress difference. In this note, we shall show that these terms destroy
the linear growth property. In a "pseudo-inviscid" limit, we shall find periodic rather
than linearly growing solutions, and we can generally expect a transient growth
factor on the order of x / - ~ W , where R is the Reynolds number and W is the
Weissenberg number. Even at small Weissenberg numbers x/R/W may be small
relative to R, so that there is an effective stabilization.
Qualitatively, we can describe the inviscid instability mechanism as follows: A
streamwise vortex generates a flow away from the wall on one side, and a flow
towards the wall on the other side. Where the flow is away from the wall, low-speed
fluid is convected outward, and the downstream flow decelerates. Where the flow is
M. Renardy /J. Non-Newtonian Fluid Mech. 59 (1995) 93-101 95

towards the wall, high-speed fluid is convected inward and the downstream flow
accelerates. In both cases a contribution to the Reynolds shear stress results from the
product of wall-normal and streamwise velocity, which enhances the drag. A second
normal stress difference creates a tension in the spanwise direction and a pressure in
the wall-normal direction. As we saw above, the effect of a streamwise vortex is to
decelerate the flow, i.e. to decrease the shear rate, where the flow is away from the
wall, and to increase the shear rate where the flow is towards the wall. With the
shear rate, the second normal stress is decreased, so that there is a reduced pressure
in the wall-normal direction which counteracts the flow of the streamwise vortex and
thus provides a negative feedback. This negative feed- back leads to an oscillation
instead of the unlimited linear growth in the inviscid case.
It is natural to wonder whether the mechanism identified here plays a role in drag
reduction. It has been known since the 1940s [8] that even very small concentrations
of polymers have a strong effect on the suppression of turbulence. There has been an
extensive literature on the subject, but the basic mechanisms are still not understood.
See Ref. [9], Section 10.7.3, for an overview. Conventional wisdom is that the
resistance of polymers to extension is important. The mechanism discussed here may
give an alternative explanation.

2. Liner stability of viscoelastic shear flows

The analysis which follows is based on the second-order fluid, i.e. it pressumes
that the product of the shear rate and the relaxation time of the fluid is small. We use
the notation
A = v u + ( v u ) T, (1)
and we put the constitutive law in the dimensionless form
~d
T= A- Wt --~--~- W2A 2 (2)

Here ~ / ~ t denotes the upper convected derivative (in contrast to the usual
convention in the literature, where the lower convected derivative is used), and the
signs in front of the terms in (2) reflect the observed signs of normal stress
differences. We are interested in cases where W~ and W2 are small, and we only
want to carry terms which couple Squire modes into the equation for O r r - S o m m e r -
feld modes. Since the term involving @ A / ~ t preserves Squire's transformation, it
leads to no such coupling, and we shall drop it. Thus, we have the Reiner-Rivlin
law
T = A - W A 2, (3)
and we shall call W = W2 the Weissenberg number. Note that this is a Weissenberg
number based on second rather than first normal stresses, and we can therefore
expect it to be about one order of magnitude smaller than the usual Weissenberg
number.
96 M. Renardy / J. Non-Newtonian Fluid Mech. 59 (1995)93-101

The momentum equation reads, in dimensionless form

< )
R ~-~ + (u V)u = div T - Vp = z~u - W A Au - W ( A : ~2)lg -- Vff, (4)

where R is the Reynolds number, A :8 2 stands for Eij A u 82/8xg axj, and fi is a
modified pressure term; the tilde will henceforth be suppressed. In addition to the
momentum equation, we have the incompressibility condition

div u = 0. (5)

We are interested in the linearization at parallel shear flow given by u = (Uo(z),0,0).


The linearized equations for the perturbations read

R - ~ -~- NO(Z ) ~X + I'l~'g(z)Wl

,,/~u
= A.- WVo(A e3 + A w < , ) - WVo[, + Vu)\
O2u ~w
-- 2 WU'o ~x ~----~- 2 W U ; ~ z e, - Vp. (6)

Here we have used components u = (u,v,w), and ei denotes the vector in the ith
coordinate direction.
We now consider modes proportional to ~ + i f l y ) and transform
u* = (~u + flv)/~*, v* = (~v - flu)/c~*, where c~* -- x/c~2 + f12. This yields the follow-
ing equations:

R( + U o i c ~ u * + ~ U'ow ) =Au*--i~*p

au* ~ aw
-- W ~ U o A w - 2 WU~ic~u* + WU~iflv* - 2 W U o i e ~ 2 W U ; ~ , Oz '

R~- + U0i~v* - - - U'ow = Av*


(7)
__ ~v* fl ~w
+ W fl U ' o A W - W U ' ~ i c ~ v * - 2 W U ' o i ~ - ~ z + 2 W U ' ~ c~* az

\ at + Uoi~w = A w - ~ - 7; Au*+ WU'o - - Av*

au* flav* aw
- WUDi~w- WU[~ - g _z 8+ WUo ~ , az 2WU'oi~ ~--7 "

Only those non-Newtonian terms which couple v* into the other variables will be
important. This simplifies (7) to
M. Renard), / J. Non-Newtonian Fluid Mech. 59 (1995) 93 I01 97

/'OU* ~ / "' *
R ~ - ~ + Uoieu* +~-g U ' o w / = au* - ie*p + mUolflV ,

R~,--~-- + Uoi~v* - - - U;w = Av*, (8)

(~w ) ~p
R~-+ U0ic~w = A W - ~ z + WU'o - - Av* + W U ; ~fl, aV*az

We can combine the equations for u* and w by exploiting the incompressibility


condition ie*u* + wz = 0. In the usual way, we then obtain the system

~+i~U0 Aw-i~Uow =A2w-c~*flW(UoAv *-Uo'v*),

~+U0ic~ v* = A v * + R ~, U;w. (9)

We get further simplifications by making the following specializations: (1) We focus


on plane Couette flow, where Uo = 1. (2) Since spanwise wave vectors are the ones
most affected by transient linear growth, we concentrate on the case fl = ~*, ~ = 0.
With these simplifications, (9) becomes
~Aw
R -- f l 2 W A v * + AAw,
~t
(lO)
i~v*
R~=Av*+Rw.

In the inviscid limit, we simply get


~w ~v*
-0, =w, (11)
0t ~t
which implies that w stays constant while v* grows linearly. However, dissipation
leads to decay at a rate proportional to l/R, so that the linear growth will persist
only up to times of order R, followed by decay thereafter. We thus have transient
growth, with v* reaching a magnitude on the order of Rw(O). Let us now consider
the effect of the viscoelastic terms, and, for simplicity, we shall ignore dissipation.
Then the equations are
~Aw__ ~2 m
~t ~- Av*,
Or* (12)
~t
If w and v* satisfy the same boundary conditions (for example, zero at walls), we
can simplify the first equation to
~W f12 W
et = - ~- v*, (13)
98 M. Renardy / J. Non-Newtonian Fluid Mech. 59 (1995) 93 101

The solution of (12) is then

W= /' v(0) sin(fl ~ / ~ t ) ,

v = v(o) cos
(g) 1
/~ t + w(0) s i n ( f l ~ - ~ t).
(14)

Thus, even if viscous dissipation is ignored, transient growth is limited to a factor


of order fl-lx/-R/W. We can expect this effect to be important if fl ~x/-R/W is
comparable to R, i.e. if flRRW is at least of order 1. Modes with fl on the order of
1 are the ones of interest. If fl >> 1, then viscous dissipation becomes large; on the
other hand, if fl << 1, then the incompressibility condition implies that u* is large
relative to w, i.e. only a small part of the disturbance is the w-component of the
velocity, which is responsible for driving the Squire mode. Thus the mechanism
discussed here is important if RW reaches order of magnitude 1.
The above analysis is rather crude and neglects many terms. The guiding
principle has been that inviscid terms dominate, viscous terms are small and
primarily act to provide a small dissipation, and viscoelastic terms are even smaller
than the viscous terms. Therefore, the viscoelastic terms have been taken into
account only in the one place where the inviscid and viscous terms make no
contribution, and this is in coupling the Squire mode back into the Orr Sommer-
feld equation. A more complete study of the linear dynamics, which takes into
account the additional terms, would require a numerical investigation.

3. Potential relevance to drag reduction

As is well known, polymers are effective in at least partially suppressing turbu-


lence at very small concentrations. A number of explanations for this effect have
been offered. Some early works tried to explain the phenomenon by a geometric
effect of long polymer molecules on small-scale eddies. This explanation, however,
does not seem to hold up to quantitative scrutiny, since polymer molecules are
about two orders of magnitude too small. More recent attempts have therefore
focused on "elasticity". There are, however, many ways in which elasticity may be
relevant (linear memory effects, elongational stresses, normal stresses in shear flow).
There is no consensus in the literature about the questions of which aspect of
elasticity and which feature of the turbulent flow are important. Many hypotheses
exist, and most "explanations" are vague; often they do not go much beyond a
comparison of time or velocity scales.
A much more detailed and quantitative theory has been developed by Goldshtik
et al. [10]. Their analysis is based on the equations of linearization at laminar flow.
However, they take account of the nonlinear contribution to the Reynolds shear
stress, and this nonlinearity enters the boundary conditions, which are formulated
on the basis of some ad hoc assumptions. For the velocity field, they take a
traveling wave with essentially streamwise propagation, superposed on the laminar
M. Renardy /J. Non-Newtonian Fluid Mech. 59 (1995) 93-101 99

flow. In the Newtonian case, their prediction of the logarithmic velocity profile
agrees with experiments, but the fluctuation amplitudes which they predict are
much too large. They study the viscoelastic case assuming Maxwell and Oldroyd B
rheology. They find a reduction of drag, which is primarily due to a reduction in
the wall-normal fluctuations. Such a reduction is consistent with experimental
trends. On the other hand, it has been claimed [11] that the primary effect in drag
reduction is not a reduction of fluctuation amplitudes, but a decrease in the
correlation between axial and radial components. The fluctuations considered in
Ref. [10] have a correlation which is too low; this is clearly evidenced by the fact
that the fluctuation amplitudes are much larger than the experimental values, while
the Reynolds shear stresses are in agreement with the experiment. On the other
hand, the mechanism considered above, in which a streamwise vortex generates a
streamwise speed, would lead to fluctuations with a high correlation. The effect of
viscoelasticity, as described in Section 2, would be to discourage these highly
correlated motions and also to reduce their correlation.
Observations of turbulent near-wall flows support the existence of streamwise
vortices, which generate low-speed "streaks". These streaks are intermittent and are
destroyed in a process known as "bursting." This bursting process is often linked
with the formation of inflection points in the mean profile and resulting instabilities.
A very clear and concise exposition of these ideas for near-wall turbulence is given
in the article by Clark [12]. I also refer to the survey article by Robinson [6].
Starting with the work of Black [13], there has been a substantial literature trying
to link drag reduction to a decrease in bursting activity. Experiments [14,15] show
an increase in the lateral streak spacing and a decrease in the frequency of bursts.
In terms of the analysis of Section 2, we can interpret the increased streak spacing
as a shift to smaller values of ft. At the same time, the Reynolds stresses produced
by the streamwise vortices are decreased, which would be expected to delay the
bursting. The suggestion of this paper is therefore that bursts are reduced not
because drag-reducing polymers inhibit bursting, as has often been suspected, but
because they interfere with the sequence of events which leads up to bursting. In
Ref. [15], the specific hypothesis is made that drag reduction results from the
inhibition of the formation of streaks. This conjecture is based purely on experi-
mental results, and no mechanism is offered. Interestingly, some streaks are
observed to disappear and then reform. This is consistent with the transition from
linear algebraic growth to oscillatory behavior which was found above.
The analysis of Section 2 gives a criterion when elastic effects should become
important. We try to apply this criterion to the viscous layer. The Reynolds number
of the pipe flow is defined by R=pUL/I~, and the wall shear stress is
=fp U2 =flt2R2/(pL2). Here L is the pipe diameter, U is the average speed, and
f i s the friction factor. Moreover, the Weissenberg number is a multiple of the shear
stress, and we can express this as W = E~. We expect drag reduction to become
relevant at a critical value of WRb, where R b is a Reynolds number associated with
the viscous sublayer. Since Rb is approximately a constant (on the order of 100), we
would expect significant drag reduction to start when W is on the order of 1/100.
Since our Weissenberg number was defined on the basis of second rather than first
100 M. Renardy / J. Non-Newtonian Fluid Mech. 59 (1995) 93 101

normal stresses, a conventionally defined Weissenberg number would be about one


order of magnitude larger. Since f varies slowly the Reynolds number, a fixed value
of W = ev translates into a critical Reynolds number R which is proportional to
L/x/-~, where E is presumably proportional to the concentration of the polymer.
These scalings appear to be roughly consistent with observations. The proportional-
ity with L can, for example, be seen in Figure 10.8 of Ref. [9], which gives a
schematic representation of experiments. The literature is more ambiguous on the
concentration dependence. I think it is important whether one defines "significant
drag reduction" in absolute or relative terms. That is, does "significant drag
reduction" commence when the drag is reduced by a given percentage, or by a
certain fraction of the maximum that can be achieved for the same concentration?
The absolute standard is roughly consistent with the square root dependence
predicted here; see e.g. the data reported in Figure 3.11-1 of Ref. [16]. On the other
hand, the relative interpretation leads to the conclusion that the wall shear stress at
which drag reduction commences is independent of concentration [17].
The use of the second-order fluid model in the analysis is open to question. Even
if elastic effects are "small", this does not necessarily mean that the use of the
second-order fluid is justified. For that, one needs the wall shear rate to be small
relative to the relaxation time of the polymer. Polymer relaxation times in aqueous
solutions are estimated to be on the order of 10 4 - 1 0 - 3 S (see e.g. Table I of Ref.
[18]), and inverse wall shear rates in drag reduction experiments are on both sides
of this range. Outside of the second-order fluid range, we cannot expect that the
second normal stress will continue to grow quadratically as a function of shear rate,
and we would therefore expect a relatively smaller effect. This is consistent with
observations showing that smaller concentrations, which need a higher wall shear
rate to become effective, have small drag reductions. Even for higher concentra-
tions, drag reduction has a tendency to "bottom out" at large wall shear rates.
Since Reynolds stress production from streamwise vortices is believed to be
important in transition to turbulence, drag-reducing polymers should also show a
tendency to extend the laminar regime. Indeed, some experiments show this; e.g. in
Ref.[18] this is observed for concentrations in excess of 1500 ppm. At lower
concentrations, the polymer is effective only at the higher wall shear rates which
occur in turbulent flow.
Joseph [19,20] has emphasized that the fluid speeds in the drag reducing regime
are comparable to the speed of shear wave propagation. Such an observation is not
inconsistent with the criterion found here; actually, it is consistent with a number
of mechanisms which do not necessarily involve wave speeds in an intrinsic way.
Regardless of the mechanism for drag reduction, it would be useful to have more
accurate values for relaxation times, and wave speeds can be helpful here. Litera-
ture values for relaxation times in dilute solution are often not measured, but
estimated from molecular theory.
" O f all the unusual phenomena observed with flow of non-Newtonian systems,
none has caused more scientific and economic speculation than drag reduction"
[21]. In this note, I have advanced a theory which provides a rational and specific
explanation of the way in which polymers might act to reduce drag. I believe that
M. Renardy / J. Non-Newtonian Fluid Mech. 59 (1995) 93 101 101

f u r t h e r e x p l o r a t i o n o f these ideas is needed. N u m e r i c a l s i m u l a t i o n s , in p a r t i c u l a r ,


w o u l d be useful.

Acknowledgement

T h i s research was s u p p o r t e d b y the N a t i o n a l Science F o u n d a t i o n u n d e r G r a n t


D M S - 9 3 0 6 6 3 5 a n d b y the Office o f N a v a l R e s e a r c h u n d e r G r a n t N00014-92-J-1664.
I a m grateful to the n o t so a n o n y m o u s referees for their c o n s t r u c t i v e c o m m e n t s . I n
p a r t i c u l a r , I w o u l d like to a c k n o w l e d g e E.J. H i n c h , w h o p r o v i d e d the v e r b a l
d e s c r i p t i o n o f the m e c h a n i s m w h i c h is p r e s e n t e d in the i n t r o d u c t i o n a b o v e .

References

[1] K.M. Butler and B.F. Farrell, Phys. Fluids, A4 (1992) 1637.
[2] D.S. Henningson, A. Lundbladh and A.V. Johansson, J. Fluid Mech., 250 (1992) 169.
[3] L.N. Trefethen, A.E. Trefethen, S.C. Reddy and T.A. Driscoll, Science, 261 (1990) 578.
[4] T. Ellingsen and E. Palm, Phys. Fluids, 18 (1975) 487.
[5] M.T. Landahl, J. Fluid Mech., 98 (1980) 243.
[6] S.K. Robinson, Ann. Rev. Fluid Mech., 23 (1991) 601.
[7] G. Tlapa and B. Bernstein, Phys. Fluids, 13 (1970) 565.
[8] B.A. Toms, Proc. Int. Congr. Rheol., North-Holland, Amsterdam, 1949.
[9] R.I. Tanner, Engineering Rheology, Clarendon, Oxford, 1988.
[10] M.A. Goldshtik, V.V. Zametalin and V.N. Shtern, J. Fluid Mech., 119 (1982) 423.
[11] P.S. Virk, AIChE J. 21 (1975), 625.
[12] D.G. Clark, in Turbulent Drag Reduction by Passive Means, Royal Aeronautical Society, London,
1988, p. 457.
[13] T.J. Black, in C.S. Wells (ed), Viscous Drag Reduction, Plenum, New York, 1969, p. 383.
[14] B.U. Achia and D.W. Thompson, J. Fluid Mech., 81 (1977) 439.
[15] D.K. Oldaker and W.G. Tiederman, Phys. Fluids, 20 (1977) S133.
[16] R.B. Bird et al., Dynamics of Polymeric Liquids I, Wiley, New York, 1977.
[17] P.S. Virk, J. Fluid Mech., 45 (1971) 417.
[18] C.B. Wang, Ind. Eng. Chem. Fund., 11 (1972) 546.
[19] D.D. Joseph, Fluid Dynamics of Viscoelastic Liquids, Springer, New York, 1990.
[20] D.D. Joseph and C. Christodoulou, J. Non-Newtonian Fluid Mech., 48 (1993) 225.
[21] W.R. Schowalter, Mechanics of Non-Newtonian Fluids, Pergamon, Oxford, 1978.

You might also like