(E) High Reynolds-Number Scaling of Channel & Pipe Flow

We shall now discuss the scaling of statistical quantities in turbulent pipe and channel flow in

the limit of high Reynolds numbers. Although an old subject, it is also a very active one at the

present time, because of new developments in experiment, simulation and theory. Many of the

key issues are still hotly debated and the subject is in a state of flux. For an excellent current

review, we recommend

“Scaling and structure in high Reynolds number wall-bounded flows,” Theme Issue

in Philos. Trans. Roy. Soc. A 365 (2007) 635-876

a collection of fourteen articles by leading researchers in the field. We shall discuss some of the

findings and conclusions in those papers further below.

We shall first consider the scaling of the two primary quantities, the mean velocity ū(y) and the

Reynolds stress u! v ! , in turbulent channel flow. To begin, we shall follow closely the traditional

treatment in Tenekes & Lumley, Section 5.2. For a more careful treatment along the same lines,

using methods of asymptotic analysis, see:

R.L. Panton, “Composite asymptotic expansions and scaling wall turbulence,” Phi-

los. Trans. Roy. Soc. A 365 733-754 (2007)

Starting from the basic equation of the total stress τxy

tot

−u! v ! + ν ∂∂yū = u2∗ (1 − hy ) (#)

we see that u! v ! ∼ O(u2∗ ) at high Reynolds number, if the viscous term can be neglected at

finite distances from the wall. Also, the turbulent length-scale away from the walls must be set

by the channel half-height h. This suggests the dimensionless outer scaling of variables, as
∂ ū/∂y −u" v "
η = hy , dF
dη = u∗ /h , G= u2∗

in terms of which (#) becomes

G+ 1 dF
Re∗ dη =1−η (O)

37
with

Re∗ = u∗ h
ν

the Reynolds number based on friction velocity u∗ and half-height h. The dimensionless quan-

tities F and G are some functions of η and Re∗ :

dF
dη = dη (η, Re∗ ),
dF
G = G(η, Re∗ )

Taking Re∗ # 1, the equation (O) suggests that

G(η, Re∗ ) ∼ G0 (η) = 1 − η, Re∗ # 1

as the leading-order asymptotics, with η fixed. Note, however, that

G(0, Re∗ ) = 0

because of the stick b.c. at the wall, whereas G0 (0) = 1! This is an indication that the high

Re∗ -asymptotics is a singular perturbation problem and that a layer with different physics exists

near the wall. In particular, there will be influence from viscosity at distance from the wall of

the order

d = ν/u∗ .

This motivates us to consider a different set of dimensionless variables corresponding to that

layer at the wall, the inner scaling:

y yu∗ −u" v "
y+ = d = ν , f= ū
u∗ , g= u2∗

In these variables, the Reynolds stress equation becomes

df y+
g+ dy +
=1− Re∗ (I)

with

f = f (y + , Re∗ ), g = g(y + , Re∗ ), Re∗ = h+ .

The equation (I) suggests that for Re∗ # 1,

f (y + , Re∗ ) ∼ f0 (y + ), g(y + , Re∗ ) ∼ g0 (y + ), Re∗ # 1

with
df0
g0 + dy +
= 1.

In the inner layer or wall layer, where this approximation is accurate, the stress is constant and

does not vary perceptibly with the distance to the wall. The relations

38
 ū(y) ∼
= f0 (y + )
u∗

−u! v ! /u2∗ ∼
= g0 (y + )

are called the law of the wall. We shall discuss later the test of these relations and empirical

results for the scaling functions f0 , g0 .

We have, so far, no information on the mean velocity F (η, Re∗ ) in the outer layer or core region

of the flow. For this purpose, we may consider the turbulent energy balance

−u! v ! ∂∂yū = ε + ∂y [(p

∂ ! + 12 q 2 )v ! − ν ∂y ( 2 q )]
∂ 1 2

We can expect that

3
ε ∼ O( uh∗ ), (p! + 12 q 2 )v ! ∼ O(u3∗ ), 1 2
2q ∼ O(u2∗ )

which motivates us to define

(p" + 12 q 2 )v " 1 2
q
&= ε
u3∗ /h
, J= u3∗
, K= 2
u2∗

so that, in the outer scaling,

dη = & +
G dF dη [J + Re∗ dη ]
d 1 dK

we therefore expect that for Re∗ # 1

dη (η, Re∗ ) dη (η)

dF dF0
∼

in order to make both sides of order one. To get information on the mean velocity in the outer

layer, this equation, written as

∂ ū ∼
= u∗ dF0
∂y h dη

must be integrated from the centerline η = 1, to yield

ū(y) − ūc = u∗ F0 (η)

where ūc = ū(h) is the velocity at the center of the channel. This relation, usually written as
ū−ūc
u∗ = F0 (η),

is called the velocity-defect law in the outer layer.

At sufficiently large Re∗ , there is a region where both outer layer and inner layer descriptions

are simultaneously valid, with η → 0 and y + → ∞, respectively. Since

y + /η = u∗ h
ν = Re∗

39
one may take, for example,

y + ∼ Reα∗ , η ∼ Reα−1
∗

with any 0 < α < 1. In such a range, y + → ∞, η → 0 as Re∗ → ∞. In this range, called the

region of overlap or matched layer, one can match the two descriptions, giving
u2∗ df0
u∗ dF0
h dη = ∂ ū
∂y = ν dy +

Multiplying this equation by y/u∗ gives

+ df0
dη = y dy + .
η dF 0

If we consider finite (but large) y + , then η ∼

= 0 as Re∗ → ∞. Likewise, if we consider finite

(but small) η, then y + ∼

= ∞, Re∗ → ∞. Assuming that one of the limits limη→0 η dF
dη or
0

df0
limy+ →∞ y + dy + is a finite number, then the other must be the same number. We then see that

the above expression must be nearly constant in the matching region:

+ df0
dη = y dy + ∼
η dF 0 1
κ

These relations may be integrated to yield the logarithmic law of the wall:

F0 (η) ∼ 1
κ ln η + A, η&1

f0 (y + ) ∼ 1
κ ln y + + B, y+ # 1

with A and B constants. The constant κ is the so-called von Kármán constant. This logarithmic

behavior was first suggested by

T. von Kármán, “Mechanische Ähnlichkeit und Turbulenz,” Nach. Ges. Wiss.

Göttingen, Math.-Phys. Kl. 58-76 (1930)

and

L. Prandtl, “Zur turbulenten Strömung in Röhren und längs Platten,” Ergebn.

Aerodyn. Versuchsanst, Göttingen 4 18-29 (1932)

by different theoretical arguments. The above derivation using asymptotic matching is an

(oversimplified) version of that given by

40
 A. Isakson, “On the formula for the velocity distribution near the walls,” Zh. Eksper.

Teor. Fiz. 7 919-924 (1937)

and

C. B. Millikan, “A critical discussion of turbulent flows in channels and circular

tubes,” Proc. 5th. Int. Congr. Appl. Mech. (Cambridge, MA 1939), pp. 386-392.

Because of the logarithmic variation of the mean velocity, the overlap region is often called

the logarithmic layer. A corresponding result for the Reynolds stress can be obtained from the

relation
df0
g0 + dy +
= 1,

which yields
" "
− uuv2 = g0 (y + ) = 1 − 1
κy +
, y+ # 1
∗

The logarithmic region is thus a range of approximately constant Reynolds stress and relatively

small viscous stress ∼ O(1/y + ). For this reason, the overlap region is often called also the

inertial sublayer.

Returning to the logarithmic defect law

ū−ūc
u∗ = 1
κ ln η + A, η & 1

and the logarithmic law of the wall

ū
u∗ = 1
κ ln y + + B

we see by subtraction that

ūc
u∗ = 1
κ ln Re∗ + C

with C = B − A. This is called Prandtl’s logarithmic friction law. It implies that if u∗ (or

∂ p̄/∂x) is held fixed as Re∗ → ∞, then ūc → ∞. Equivalently, if ūc is held fixed as Re∗ → ∞,

then u∗ → 0 at a weak (logarithmic) rate. The above derivation has assumed that
∂ ū/∂y
dη (η) = O(1),
dF dF0
dη ≡ u∗ /h → Re∗ → ∞.

However, the friction law shows that it would not be consistent to assume that

41
 F!(η) ≡ ū
u∗ → F̃0 (η) = O(1), Re∗ → ∞

Instead ū must be scaled by a logarithmically different velocity scale, e.g. ūc = ū(h). In that

case,
ū(y)
F!(η, Re∗ ) ≡ ūc = 1 + ( uū∗c )F (η, Re∗ ).

See Panton (2007) and especially

R. L. Panton, “Review of wall turbulence as described by composite expansions,”

Appl. Mech. Rev. 58 1-36 (2005)

for a comprehensive account. These papers review also the notion of “composite expansion”

in the theory of matched asymptotics applied to wall-bounded turbulence. It is useful to say a

word about this here. The logarithmic profile

F0 (η) ∼ 1
κ ln η + A ≡ [F0 (η)]cp , η & 1

is called the common part of F0 (η) since it matches onto the similar logarithmic behavior of

f0 (y + ) ∼ 1
κ ln y + + B for y + # 1. One can thus define a composite expansion
ū(y) ∼
= f0 (y + ) + F0 (η) − [F0 (η)]cp
u∗

which should be uniformly valid over the whole width of the channel. This is usually expressed

in terms of the wake function.

W0 (η) = F0 (η) − [F0 (η)]cp = F0 (η) − [ κ1 ln η + A]

which measures the deviation from the logarithmic profile. It satisfies W0 (η) → 0 for η → 0, but

can be sizable for η = O(1), near the centerplane or core of the flow. The composite expansion

then becomes
ū(y) y+
u∗ = f0 (y + ) + W0 (η), η = Re∗

which should be uniformly valid for all 0 < y < h as Re∗ → ∞.

Physical discussion: energetics

Let us consider some physical consequences of our asymptotic analysis. We have seen that in

the inertial sublayer the Reynolds stress is approximately equal to u2∗ and the mean velocity-

42
gradient equal to u∗ /κy. Hence, the turbulence production is
u3∗
−u! v ! ∂∂yū ∼
= κy

If turbulence production is mainly balanced by viscous dissipation, then (Taylor, 1935)

u3∗
ε∼
= κy .

The picture is one of a Kolmogorov-type energy cascade in approximately homogeneous tur-

bulence under weak local shear, with a spatial flux of energy transported across it from the

outer flow to the wall. However, this spatial transport is expected to be a relatively small

perturbation of the energy cascade in the inertial sublayer. For example,

N. Marati, C. M. Casciola & R. Piva, “Energy cascade and spatial fluxes in wall

turbulence,” J. Fluid Mech. 521 191-215 (2004)

have studied the issue in DNS. They find that energy that is produced at the shear length-scale
"
*s = |∂ ū/∂y|
ε
3

is locally dissipated around the location y at the Kolomogrov microscale

3 3 3/4
ηK = ( νε )1/4 ∼
= ( κyν
u3
)1/4 = κ1/4 y/y+
∗

Of course, this cannot be true arbitrarily close to the wall, because the integral length scale

must be approximately

L∼
= κy,

since the largest eddies can have size, at most, the distance to the wall. This decreases with

decreasing y even faster than does ηK . Thus,

L ∼
= (κy + )3/4 = (Rey )3/4
ηK

as y + decreases. The existence of a turbulent energy cascade requires that L/ηK be sufficiently

large, above some critical value (roughly of order 10 − 100). Thus, there is some region with

distance y + = O(1) from the wall, with no energy cascade and where viscous energy dissipation

dominates. This is called the viscous sublayer. The velocity field in this range still fluctuates,

since the local Reynolds number Rey is transitional, but does not support an energy cascade.

The mean velocity profile can be inferred in this range from

43
 = −u! v ! + ν ∂∂yū ∼
u2∗ ∼ = ν ∂∂yū

which implies that

ū/u∗ ∼
= y+

in the viscous sublayer. At somewhat larger distances y + neither one of the stresses can be

neglected, in a region called the buffer layer. The turbulence production −u! v ! ∂∂yū reaches a

maximum here. This may be seen by writing the production in dimensionless form
df0
g0 dy +

and using
df0
g0 + dy +
=1

which implies that the maximum value 1/4 is reached when Reynolds stress and viscous stress
df0
exactly balance, i.e. g0 = dy +
= 2.
1
A schematic representation of the different ranges in

channel flow is shown below:

Figure 1.

The energy dissipation in the inertial sublayer (overlap region) ε(y) ∼

= u3∗ /κy also peaks at its

inner limit, near the buffer layer. This leads to a logarithmic divergence in the total dissipation.

If we integrate over the entire logarithmic layer, from a point near the outer edge O(h) to a

point near the inner edge O(ν/u∗ ), one obtains

44
 # # O(h) u3∗
= u3∗ dy ∼
= ln(Re∗ )
log−layer ε(y)dy O(ν/u∗ ) κy κ

The source of the energy is mean-flow kinetic energy transferred into the wall layer by Reynolds

stresses, i.e. through the spatial energy transport term

−u! v ! · ū(y)

at the outer edge of the inertial sublayer. For y = O(h) this gives
u3∗
−u! v ! · ū(y) ∼
= u2∗ ūc ∼
= κ ln(Re∗ ),

matching the dissipation rate in the log-layer. Note that there is relatively little direct dissipa-

tion in the outer layer. This may be estimated by production, as

∂ ū(y) 3
−u! v ! · ∂y = O(u2∗ uh∗ ) = O( uh∗ )

which integrated across the height O(h) of the core gives a net dissipation

O(u3∗ ) & O(u3∗ ln(Re∗ )).

Similarly, there is relatively little dissipation in the viscous sublayer. The local magnitude is

very large
4
ν( ∂∂yū )2 = O( uν∗ ),

but this is concentrated in a narrow layer of height O(ν/u∗ ). Thus, the net dissipation is again

O(u3∗ ) & O(u3∗ ln Re∗ ). The final conclusion is that most of the energy dissipation occurs in

the log-layer near the inner edge (buffer layer). This energy is provided by mean-flow energy

transported to the walls by Reynolds stress and the mean-flow energy is, in turn, provided by

pressure head.

It is interesting to consider Onsager’s conjecture on inviscid energy dissipation for wall-

bounded flows. If the channel flow is forced with ūc fixed, then, as we have seen earlier, the

standard scaling theory implies that

u∗ ∼ ūc / ln Re∗ , Re∗ # 1. (∗)

Thus the energy dissipation ε(y) → 0 pointwise in y as Re∗ → ∞ and also the energy dissipation

integrated over the entire channel is predicted to vanish as Re∗ → ∞. Of course, the energy

dissipation goes to zero only very weakly (logarithmically). It is not really surprising, because

45
the mechanism of turbulence generation—the viscous boundary layer at the wall—is Reynolds-

number dependent. Notice that if the pressure gradient ∂ p̄/∂x is held fixed instead, then energy

dissipation becomes Re∗ -independent (but in that case ūc → ∞ as Re∗ → ∞!) As we shall

see in subsection (g), empirical evidence from simulations and experiments generally supports

the scaling (*). We know of no direct study of local energy dissipation ε(y) in channel flow or

pipe flow, which investigates systematically its scaling with Re∗ . In Chapter O, we mentioned

the experimental work of Cadot et al. (1997) in Taylor-Couette cells with smooth walls, which

found that energy dissipation in the boundary layer of such flows also decreases with Reynolds

number but that the energy dissipation in the bulk appears to satisfy Taylor’s relation ε ∼ U 3 /L

and to be independent of Reynolds number. It is therefore suggested that, at extremely high

Reynolds numbers, the “inviscid” energy dissipation in the bulk may become greater than the

dissipation in the near-wall log-layer and and buffer layer.

In this respect, there is an important mathematical result of Kato on wall-bounded flows:

T. Kato, “Remarks on the zero viscosity limit for nonstationary Navier-Stokes flows

with boundary,” In: Seminar on Nonlinear Partial Differential Equations, S.S.

Chern, eds. (Springer, NY, 1984).

since elaborated and extended by others, e.g.

X. Wang, “A Kato type theorem on zero viscosity limit of Navier-Stokes flows,”

Indiana U. Math. J. 50 223–241 (2001)

Assuming that a smooth Euler solution exists satisfying the no flow-through condition at the

wall, Kato proved that the following two conditions are equivalent: (ii) integrated energy dis-

sipation vanishes in a very tiny boundary layer of thickness cν/u for a fixed constant c, and

(ii) any Navier-Stokes solution with stick b.c. at the wall converges in strong L2 -sense to the

Euler solution as ν → 0. The latter statement implies that energy dissipation must, in fact,

vanish everywhere in the domain. Notice that the boundary layer in Kato’s theorem goes to

46
zero thickness even in wall units ν/u∗ , if indeed u∗ /u → 0 as ν → 0. Thus, if energy dissi-

pation vanishes very near the boundary—as present experiments suggest is true—then energy

dissipation can remain in the bulk only if the Euler solution becomes singular in finite time.

vorticity transport

We have seen that energy dissipation/drag in wall-bounded flow requires a cross-stream flow

of vorticity, and in channel flow (or pipe flow) a constant flow in the wall-normal direction

Σyz = v ! ωz! − w! ωy! − ν ∂∂y

ω̄z
= −σ∗ .

Because the vorticity flux is related in general to the divergence of the stress (minus the isotropic

part), the vorticity flux in channel flow is related to the Reynolds and viscous stresses as:

Σyz = − ∂y
∂ tot
τxy

with τxy
tot = u! v ! − ν ∂ ū . Thus, our previous scaling analyses can be applied to the vorticity
∂y

flux. The most important conclusion is that, in the near-wall region, the nonlinear advective

transport contribution has the wrong (positive) sign. This can be inferred from
3
u∗
v ! ωz! − w! ωy! = − d1 ∂y∂+ u! v ! ∼ + νκ(y + )2 > 0.

Thus, the vorticity transport is dominated by viscous diffusion close to the wall.

How close? As the above relation makes clear, the advective vorticity transport changes

sign at the maximum of the Reynolds stress in the wall normal direction. An extremum in the

stress cannot be seen either in inner or in outer scaling separately, since in the former the stress

is strictly increasing and in the latter strictly decreasing. To get a prediction for the location

of the maximum stress, we must employ a uniformly valid expression. This has been done by

Panton (2005), who used a composite expansion for Reynolds stress constructed from

1
g0 (y + ) ∼ 1 − , G0 (η) ∼ 1 − η, [G0 (η)]cp = 1,
κy +

so that
1 y+
−u! v ! /u2∗ ) g0 (y + ) + G0 (η) − [G0 (η)]cp = 1 − − .
κy + Re∗

47
An easy calculation (Panton, 2005) shows that the peak Reynolds stress occurs for

yp+ ∼ (Re∗ /κ)1/2 .

This result was already known empirically by data analyses of

R. R. Long and T.-C. Chen, “Experimental evidence for the existence of the meso-

layer in turbulent systems,” J. Fluid Mech. 105 19-59 (1981).

K. R. Sreenivasan, “A unified view of the origin and morphology of the turbulent

boundary layer structure,” in Turbulence Management and Relaminarisation, edited

by H. W. Liepmann and R. Narasimha (Springer- Verlag, Berlin, 1987), pp. 3761.

who used the results to argue for a “mesolayer” or “critical layer” in turbulent boundary layers.

Here is the compilation of data from Long & Chen (1981):

Evidence for the existence of the ‘mesolayer’ 23

6

f
-w 4

2 I I I I I I
5 6 I 8 9 10
In R
FIGURE 2. Distance z,, of maximum of Reynolds stress from wall for pipes and boundary layers.
The line is z, = 1.89R*, where R = u,a/v for a pipe and u7Sd/v for a boundary layer where 6,
is boundary-layer thickness. x , Nikuradse, pipe; 0 ,Laufer, pipe; 0, Ueda & Mizushina, pipe;
A, Gupta &, Kaplan, boundary layer; V , Klebanoff, boundary layer; 0,Schildknecht et al.,
boundary layer.

where S = zU,and we use the exact equation (Monin & Yaglom 1971, p. 269)
T = 1-zR-l-ZZ.
These results were originally quite controversial, because it was claimed(5)that they invalidated
Let us now consider the essential nature of the classical assumptions. One assumes
that there is a region M, far above the sublayer but far also from the centre of the
the traditional scaling
pipeanalyses
in which for leading
large R meantoquantities
the log suchlayer.
as TdareThe argument
independent of v and of
H asPanton (2005) shows
v -f 0 and H + a.As we move in M, toward larger and larger zd, we experience ulti-
clearly that this is not thea small
mately case.
-
but sensible and growing importance of H . This should be felt first at
some zd of order H or z R,because v is unimportant in the region M, and further out.

begins to compete with the z-l term and48 -

We see from (4),however, that the first effects of H (or R )are felt when the zR-l term
that this occurs a t z Ra (which is also the
observed location of the maximum of Tdor T )and we find a basic contradiction in the
classical arguments. -f Evidently, the effects of viscosity are non-negligible in the
transition region as our present theory reveals. Furthermore, (4) shows that the first
effects of Hor R for the curvature, T,, are felt in a different region, z R, and that the
N
 However, there is a contradiction with some traditional physical interpretations of the log

layer as a region where viscosity is negligible. Quoting Tennekes & Lumley (1971), p.156, “The

matched layer is called inertial sublayer because of this absence of local viscous effects.” As

a matter of fact, the peak of the Reynolds stress occurs in the middle of the log-layer! (More

precisely, intermediate length scales at high Re∗ with y + ∼ Reα∗ → ∞ and η ∼ Reα−1
∗ → 0 for
1/2
any 0 < α < 1 define the matching region.) As we have seen, however, for y + < yp+ ∼ Re∗

(and some distance beyond) viscous diffusion dominates in the vorticity transport.

effects of roughness

All of our discussion up until this point has assumed perfectly smooth walls. The surfaces of

real pipes (and channels) will instead be rough at macroscopic scales:

Figure 2.

The rms variation of the y-coordinate of the wall around the mean value y = 0 is called

the roughness height k. This introduces a new fundamental length-scale into the problem, in

addition to the friction length d = ν/u∗ and the pipe radius R. Thus, dimensional analysis for

the mean velocity gives (in inner scaling)

ū(y)
u∗ = f (y + , Rek , Re∗ )

49
or
ū(y)
u∗ = f!( ky , Rek , Re∗ )

where Rek = k + = ku∗ /ν is the roughness Reynolds number.

If f has a limit for Re∗ → ∞, then

ū(y)
u∗ = f0 (y + , Rek )

or
ū(y)
u∗ = f!0 ( ky , Rek )

for pipe radius R # y, k. It is reasonable to expect that the velocity defect will be independent

of Rek for k & R:

ū(y)−ūc
u∗ = F ( η, Re∗ ).

If we follow the traditional log-layer theory, then for Re∗ → ∞, F (η, Re∗ ) → F0 (η) so that
ū(y)−ūc
u∗ = F0 (η) ∼ 1
κ ln η + A, η & 1.

This must be matched to the inner functions, which requires

ū(y)
u∗ = 1
κ ln y + + B(Rek ), y + # 1

or
ū(y) !
u∗ = 1
κ ln( ky ) + B(Rek ), y/k # 1

A similar analysis for the Reynolds stress gives

" "
− uuv2 = g(y + , Rek , Re∗ ) = g!(yk , Rek , Re∗ )
∗

with yk ≡ y/k. Scaling lengths with k,

1 dfe
g! + Rek dyk = 1 − (R
k
)yk .

For Re∗ → ∞
1 dfe0
g!0 (yk , Rek ) + Rek dyk (yk , Rek ) =1

This equation implies that viscous stress is small at positions yk of order 1 if Rek → ∞. Phys-

ically, the roughness elements at large Rek produce turbulent wakes that generate substantial

Reynolds stress on scale y ∼ O(k). It is only for y & k that the viscous stresses come to

dominate.

50
 These considerations suggest that the limit Rek → ∞ is well-defined at yk ∼ O(1) so that
ū
= 1 ! yk # 1, Rek # 1
ln yk , +B,
u∗ κ

! = limRe →∞ B(Re
with B ! k ). If one subtracts the above relation and the defect law, then one
k

obtains a friction law in the presence of roughness

= ln( R !
k)+C
ūc 1
u∗ κ

!=B
with C ! − A. This result is independent of the Reynolds number Re∗ ! Note in this case

that energy dissipation will also become independent of Reynolds number, asymptotically for

Re∗ # 1. This is in agreement with the findings of Cadot et al. (1997) for a Taylor-Couette

flow with wall-riblets, who observed in that case that energy dissipation was insensible to the

Reynolds number even in the near-wall boundary layer.

The above result should merge with Prandtl’s logarithmic friction law as k → 0. Based on

empirical data available at the time

C. F. Colebrook, “Turbulent flow in pipes, with particular reference to the transi-

tional region between smooth and rough wall laws,” J. Inst. Civil Eng. 11 133-156

(1939)

developed a transitional curve of the form

!− C"
ūc
u∗ =C 1
κ ln ( R
k
+ Re∗ )

which interpolated between the two results, depending upon the relative magnitudes of R/k and

Re∗ (or, equivalently, of k and d = ν/u∗ . Colebrook’s formula predicts a monotonic decrease

of the friction coefficient λ = 2(u∗ /ūm )2 toward the totally rough result as Re∗ → ∞ (or

Re → ∞). In fact, this disagrees with the experiment of

J. Nikuradse, “Laws of flow in rough pipes,” VDI Forschungsheft 361 (1933); En-

glish translation in NACA Tech. Memo. 1292 (1950)

or, for a presentation of the results,

51
 Modern Developments in Fluid Dynamics, ed. S. Goldstein, (Clarendon Press,

Oxford, 1938), vol. II, Section VIII, 167.

Nikuradse’s measurements of λ show a dip below the completely rough results before eventually

saturating at the rough value of λ as Re → ∞. We shall discuss later the modern experimental

confirmation of this effect. The results are still not understood very well theoretically, however.

For a recent attempt, see

G. Gioia & P. Chakraborty, “Turbulent friction in rough pipes and the energy

spectrum of the phenomenological theory,” Phys. Rev. Lett. 96 044502 (2006)

52