JOURNAL OF THE
AERONAUTICAL SCIENCES
Volume 1 JANUARY,
Downloaded by UNIVERSITY OF CALIFORNIA - DAVIS on February 13, 2015 | http://arc.aiaa.org | DOI: 10.2514/8.5
Turbulence and Skin Friction
T H . VON K ARM AN, California Institute of Technology
(Received November 3, 1933)
1. HISTORICAL DATA
/ V
I HE first tests known on skin friction of flat
-*• surfaces in fluids are those carried out by
M. G. Beaufoy 1 in 1793. About eighty years
later W. Froude conducted a series of systematic
experiments, investigating the influence of the
velocity and the dimensions of the plate, espe-
cially t h a t of the length in the direction of the
motion, on the skin friction. In 1883 O. Reynolds
published his fundamental investigations on the
similarity of fluid motion and introduced the
dimensionless parameter known as Reynolds
number (referred to hereinafter as R.N.). In
1908 Lord Rayleigh pointed out t h a t for geo-
metrically similar arrangements, the skin friction
coefficient, i.e., the friction per unit area divided
by the head, must be a function of R.N. Suc-
ceeding investigations of Zahm, Wieselsberger,
Gibson, with air, of the British Froude National
Tank, Gebers, Kempf and the Washington Navy
Yard with water, were evaluated using the R.N.
as parameter.
Poiseuille gave in 1840 the theory of skin
friction for t h e case of laminar flow in pipes.
In 1850 G. Stokes published his calculations on
the resistance of bodies in pure viscous flow, i.e.,
neglecting the inertia forces in the fluid. Prandtl
and Blasius gave the mathematical theory of the
1 2
Beaufoy's results were published by his son H. Beau-
foy in 1834.
1
1934 Number 1
laminar skin friction of plates in a free stream in
1904 and 1907. The motion of the boundary
layer introduced by Prandtl proved itself useful
through the further developments of the theory
both for laminar and turbulent flow. The first
semi-empirical theory of turbulent skin friction
was given in 1921 by Prandtl and by the present
author. The recent development of the theory
presented in this paper started with the author's
publications in 1930.
The scientific and technical literature on skin
friction is very elaborate. An excellent bibli-
ography, extending to 1930, has been compiled
by A. F. Zahm and C. A. Ross. 2 Additional
references are given at the end of this paper.
2. S K I N FRICTION AND MOMENTUM
The modern theory of skin friction is based on
the theorem t h a t friction between a fluid and a
solid is accompanied by an equivalent change of
the momentum carried by the fluid. Let us
consider the case of a fluid moving along a flat
plate and compare the momentum carried by
the fluid through a cross section upstream from
the leading edge of the plate and through a
cross section at the distance x downstream from
the leading edge. We assume t h a t the velocity
is uniform over the first cross section and has
Tentative bibliography on skin friction and boundary
flow. Library of Congress, December, 1930.
 2 T H . VON K A R M A N
the value U. The velocity normal to the second
cross section at an arbitrary point therein shall
be denoted by u. T h e fluid mass passing in unit
time through an element dS of the second cross
section is evidently pudS, where p denotes the
density of the fluid; this fluid mass had the
momentum U per unit mass passing through
the first cross section and has the momentum u
per unit mass passing through the second cross
section. Hence the loss of momentum of the fluid
mass considered amounts to pu(U—u)dS and
the total loss of momentum of the fluid in unit
Downloaded by UNIVERSITY OF CALIFORNIA - DAVIS on February 13, 2015 | http://arc.aiaa.org | DOI: 10.2514/8.5
time to f pu{U~u)dS, t h e integral being taken
over the cross section downstream. This amount
is equal to the total frictional force acting on the
portion of the plate extending from the leading
edge to the distance x.
Let us consider a plate of t h e width b and
introduce the coordinates x parallel to the
direction of the main flow, y perpendicular to the
flow and the plate with origin at the leading edge
U U
U-u
IT T O T A L FORCE 3^ T<>
FIG. 1. Skin friction and boundary layer.
(Fig. 1). Then the friction Fx acting on the plate
between x = 0 (leading edge) and an arbitrary
value of x will be
pu(U — u)dy.
t/0
The friction acting on a strip bdx is obviously
equal t o dF; on the other hand, the same force
can be expressed by t h e local friction per unit area
or shearing stress at t h e wall r 0 in the form br^dx.
Hence we obtain
ldF d call it 0. Introducing the value of T0 and t h a t of
To=- I pu(U—u)dy
b dx dx\
In all practical cases it is sufficiently exact to
extend the integral from }/ = 0 to a finite value
y = 8; we choose 8 in such a way t h a t the velocity
defect U—u is very small for values of y superior
to <5. As a matter of fact, the influence of the
friction is restricted to a narrow region along the
plate; we call this region the "boundary layer"
and 8 "the thickness of the boundary layer."
The choice of 8 is, of course, somewhat arbitrary,
but as it will be seen, the results concerning the
laws and the values of the skin friction are
independent of the definition of 8.
The consideration sketched above replaces the
problem of the skin friction by the problem of
the velocity distribution in the boundary layer.
As a m a t t e r of fact, the friction itself can be
determined as well by direct measurement as by
measurement of the velocity distribution across
the boundary layer.
3. T H E T H E O R Y OF LAMINAR S K I N FRICTION
In the case of laminar motion, the velocity
distribution can be computed theoretically by
using the hydrodynamic equations in their some-
what simplified form due to Prandtl (1904). The
main result of this theory can be understood by
using very little calculation if we assume similar
velocity distributions over all cross sections of
the boundary layer. Let us write, accordingly,
u=U-f(y/8) = Uf(r]) where rj=y/8 is the "dis-
tance ratio" referred to the thickness 8 of the
boundary layer. The function f(rj) equals zero
for y = rj = 0 and is approximately equal to one
for y = 8 or rj = l. The "momentum integral"
f<Tpu(U—u)dy will be replaced by pU28jl1f
• a-/)<*„.
The integral Jo1/* (1 —/)d?7 is a pure number
which shall be denoted by a. On the other hand,
(i) the friction at the wall r0, according to the
general law of viscosity, equals the product of
the viscosity coefficient JJL and the slope of
velocity du/dy at the wall. Hence
r 0 = jJL(du/dy)y=o = M( U/8) (df/drj) v=o- (3)
The value (df/drj)v=0 is also a pure number, we
. (2) the momentum integral in Eq. (2) we obtain
U d d8 fi 1 fi
/x - — P =—(apU25) or 8— = •.
dx dx U a
 SKIN F ICTION 3

distance from the leading edge and proportional

By integration, §<52 = --x
to the geometrical mean value between viscosity
pUa
and and density.
S = Z(2(i/a)(nx/pU)y. (4) (c) The friction coefficients both for the local
and the mean friction are inversely proportional
Putting this value of 8 in Eq. (3) we obtain the to the square root of the R.N.
value of the local friction
4. P O W E R L A W S FOR T U R B U L E N T
T* = tf{U/$) = §aW.{HLlP/x)* (5)
S K I N FRICTION
and by integration, the value of the total friction The method of the last section can be extended
to the case of the turbulent boundary layer
Fx=(2aP)t-(pnxU*)*b. (6)
Downloaded by UNIVERSITY OF CALIFORNIA - DAVIS on February 13, 2015 | http://arc.aiaa.org | DOI: 10.2514/8.5

without investigating the actual mechanism of

We introduce the coefficient of the local friction
cf=r0/hU*
and t h a t of the total (or mean) friction
Cf = Fx/±'p-U*.b'X.
The mathematical theory gives for (2a/3)* the
numerical value 0.664. Hence with JJL/P = V
cf = 0.664/(Ux/v)*
and
Cf=1.328/(Ux/v)*.
We call Ux/v the R.N. of the plate Rx.
Summarizing the results given b y . t h e equa-
tions from (4) to (10), we obtain the following
statements (cf. Fig. 2):
turbulent motion and by using merely the method
of dimensional analysis.
(7) We assume, as before, t h a t the velocity
distribution is given by a function of the form
u = Uf(y/8), but for the wall friction we use an
(8) empirical law of the form ro = ^pU2g(U8/v).
Using a *'power law" as a convenient interpola-
tion formula, we put
T0 = yU2-const./(U8/v)m. (11)
(9)
Evidently the momentum integral Jo8pu(U—u)dy
(10) keeps the form apU28, where a is a numerical
value, which depends on the shape of the velocity
distribution in the boundary layer. Hence we
obtain from the fundamental Eq. (2)
apU2(dd/dx)=hU2-const./(U8/v)m (12)
and
8m{d8/dx) = const. -(v/U)m. (12a)

Integrating (12a) and assuming 5 = 0 for x = 0,

we obtain

5 m+1 /'(w + l ) = const. '(v/U)m-x

or
8=const.'X'(v/Ux)m^m+1K (13)

This equation determines the ''development of

FIG. 2. Distribution of laminar skin friction and thickness
of laminar boundary layer. the boundary layer" along the plate, if the
"elementary friction law" (11) is known.
(a) The increase of the thickness of the lami- Introducing the value of 8 in Eq. (11) for r 0 ,
nar boundary layer 8 is proportional to the we write
square root of the distance from the leading edge TO = hU2- const. /{Ux/v)mKm+l)
and inversely proportional to the square root of or
t h e relative velocity between plate and fluid. cf = const. /(Ux/v)mKm+1). (14)
(b) The local friction per unit area is pro-
portional to the 3/2 power of the velocity, We started with Eq. (11) which expressed the
inversely proportional to the square root of the relation between the friction and the R.N. of the
 4 T H . VON K A R M A N
boundary layer. It is logical to assume t h a t this
relation is similar to t h a t obtained for a channel
or pipe with constant cross section. For laminar
flow we have therefore, according to Poiseuille's
law, m = 1 and therefore m/m + 1 =% so t h a t Eq.
(13) leads to the results of the last section. For
turbulent flow tests on pressure drop in smooth
pipes give in the range 3000< Ur/v<70,000
(f = radius of circular pipe) good accordance with
the empirical Eq. (11) choosing the exponent
m = \. Using this exponent, Prandtl and the
present author obtained from (14) the formula
Downloaded by UNIVERSITY OF CALIFORNIA - DAVIS on February 13, 2015 | http://arc.aiaa.org | DOI: 10.2514/8.5
for the turbulent skin friction coefficient of a
smooth plate:
cf = const. /(Ux/v)115 (15)
and with the value of the numerical constant
deduced from the pipe tests
cf = 0.059/(Ux/v)^K
It is easy to show t h a t the coefficient of the
mean friction is equal to Cf = cf-(m + l)/m and
therefore
C / = 0.074/(C/xA) 1/5 - (16)
This formula checks the tests with smooth plates
in the range extending to about Ux/v^3-106.
Extension of the tests to larger R.N. made it
evident t h a t the power formula with a constant
exponent could not serve for a larger range of
R.N., either in the case of the pipe or in the case
of the plate. It was found t h a t the exponent m
decreases with increasing R.N. This proved the
assumption of similar velocity distributions to
be incorrect. In order to find a law valid in the
whole range and especially valid for large R.N.
allowing extrapolation to full scale conditions, a
more systematic investigation on the nature of
turbulent flow became inevitable. This new
development of the theory achieved in the last
few years is shortly reviewed in the following
sections.
5. T H E MECHANISM OF T U R B U L E N T FLOW
In laminar flow the interaction between ad-
jacent fluid layers consists of so-called "molecular
friction"; the shearing stress, i.e., the friction
per unit area, is equal to the product of the rate
of shear (slope of velocity in the case of parallel
flow) and the viscosity coefficient. The latter is a
characteristic physical parameter of the fluid,
dependent chiefly on the temperature, practically
independent of the velocity. In turbulent flow
an additional interaction is due to the momentum
transfer from layer to layer due to the velocity
fluctuations. Let us consider a flow, consisting
of parallel mean motion in the x-direction, the
temporal mean value u of the velocity being a
function of y only. O. Reynolds has shown t h a t
the momentum transport in unit time and
through unit area due to the fluctuations repre-
sents an apparent shearing stress for the mean
motion. This apparent shearing stress, or "turbu-
lent friction," has the magnitude r=—pu'v'
where u' and vf are the fluctuations of the velocity
components in the x- and ^-direction, and the
dash denotes temporal mean value of the product.
In fact the fluid volume passing through a unit
area perpendicular to the ^-direction is equal to
v' and the excess of momentum in the direction
of the mean flow carried by the fluid per unit
volume is equal to pu''. The negative sign
corresponds to the definition of positive friction
in such a way t h a t r is considered as positive if
the fluid layer at the distance y from the wall is
accelerated by the outside flow; accordingly, the
momentum transfer produces friction of positive
amount if the fluid portions moving toward the
wall (i.e., having negative vf components) carry
positive excess of momentum.
In the case of velocity fluctuations with a
random distribution of the u' and vf components,
the product u'v' has the mean value zero. The
shearing stress is different from zero only in the
case of a certain correlation between u' and vf.
It is easy to see t h a t a correlation can be expected
in all cases in which the mean velocity u is
variable with y. Let us assume for instance t h a t
u(y) is increasing with y (Fig. 3). Then the
FIG. 3. Turbulent exchange of momentum.
 SKIN
fluid masses moving in the negative ^-direction
( V < 0 ) , i.e., coming from a region of higher
velocity, probably have larger ^-components
than the masses with positive vf values, i.e.,
coming from a region of lower velocity. Now
evidently larger ^-velocities than the mean
velocity at the place considered appear as posi-
tive, smaller w-velocities as negative u' fluctua-
tions ; hence negative v' values will be connected
with positive u' values and vice versa. Therefore,
we have to expect a negative value for the mean
product u'vf and a positive value for the shearing
Downloaded by UNIVERSITY OF CALIFORNIA - DAVIS on February 13, 2015 | http://arc.aiaa.org | DOI: 10.2514/8.5
stress. Fig. 4 shows schematically the random
RANDOM WITH
DISTRIBUTION CORRELATION
FIG. 4. Diagram of velocity fluctuations.
and correlated distribution of the velocity fluctu-
ations.
T h e main characteristics of the turbulent flow
at a certain point are the magnitude of the
fluctuations uf, v', w', and the correlations be-
tween them. The theory of correlations uses a
certain ratio between mean values as "correlation
factor." For instance, the correlation between
the components ur and vf is_measured_by the
ratio wV/(V 2 )*(y 2 )* where (V 2 )* and (z/2)* are
the square roots of the mean squares of the
particular components. This correlation factor
is zero for random distribution and is equal to
unity if u' and v' are proportional. The ellipse
shown in Fig. 4 becomes a circle in the first case
and a straight line in the latter.
Considering the "flow p a t t e r n " in the neigh-
borhood of a certain point, we have to introduce
a further characteristic parameter of the turbu-
lent flow: namely a length which is characteristic
for the size of the region involved in the turbulent
exchange, i.e., in the turbulent momentum trans-
fer. The corresponding parameter in the molecu-
lar mechanism of friction is the mean free path.
Prandtl uses for t h e length-parameter of the
turbulent friction-mechanism the expression
"mixing length." His particular picture of the
FRICTION 5
turbulent exchange is based on the assumption
t h a t the fluid masses displaced perpendicular to
the direction of mean flow carry their momentum
invariably over a certain length perpendicular
to the mean flow. The actual mechanism may
be more complicated; in any case the conception
of a characteristic length is necessary for the
understanding of the fundamental laws of turbu-
lent flow. Let us compare, for instance, the
turbulent fluctuations in the boundary layer of
an airfoil placed in the wind tunnel and those in
the boundary layer produced by the natural
wind on the ground. Mean velocities and ve-
locity fluctuations may have the same order of
magnitude; it is obvious t h a t the length scale of
the flow pattern in the two cases will be very
different.
There is a further factor to be considered in
connection with the characteristic length or with
the length scale of the flow pattern, namely the
time-factor or the time-scale of the fluctuations.
"Large scale turbulence" will be connected in
general with fluctuations of large period, and
"small scale turbulence" with fast fluctuations.
The characteristic features of turbulent flow
have been dealt with in some detail and from a
general point of view, because the author feels
t h a t clear notions on this question are of im-
portance for understanding the phenomena and
for further research. Common language, even in
technical papers, uses the term "turbulence" in a
rather indefinite way, including in it for instance
"vortex motion" in general. The scientific term
"turbulence," as used in this paper, implies
"irregular fluctuations," governed by laws of
some statistical equilibrium. To be sure, turbu-
lence may be originated by vortices as in the
case represented in Fig. 5, showing the flow
through a honeycomb; but we call such a
motion "turbulent" only if the regular pattern,
because of the intermingling of a great number
of vortices, disappears, as happens farther down-
stream from the honeycomb. Similarly, vortices
produced by obstacles at the ground contribute
essentially to atmospheric turbulence, but also
in this case we restrict the term turbulence to the
statistical phenomenon of the mass exchange and
keep this case apart from other conceptions, such
as individual vortices (tornados, vortex sheets,
etc.) or regular atmospheric waves.
 6 TH. VON K AR M AN

mum. Iwf.lt
MBSHS

gg§n
fe^^s^^^^^R
^^jffe§jp

i^S?^^^^
Downloaded by UNIVERSITY OF CALIFORNIA - DAVIS on February 13, 2015 | http://arc.aiaa.org | DOI: 10.2514/8.5

Ss*S^^

^^^S
i^EsaEsss?*^" ^ E f ^

FIG. 5. Vortex rows produced by a honeycomb. (Photograph by Ahlborn.)

6. T H E BASIC L A W S FOR THE M E A N VELOCITY
DISTRIBUTION IN T U R B U L E N T FLOW
Adding the momentum transfer by turbulent
fluctuations to the laminar friction, O. Reynolds
obtained the following expression for the total
friction (shearing stress):
r = JJL • du/dy — pufvr (17)
Calculating r from the measured values of the
pressure drop in pipes or the skin friction of
plates, the values n(du/dy) from the measured
mean velocity distribution, we find t h a t in the
practically important range of high R.N. the
laminar friction is a negligible part of the total
friction except in the immediate neighborhood of
the wall. Hence we conclude t h a t except for
this region the turbulent exchange is practically
independent of the viscosity of the fluid. This
assumption is essentially substantiated by experi-
mental facts concerning the velocity distribution
in pipes and channels. Let us consider, for
instance, turbulent flow between two parallel
walls (rectangular pipe with great aspect ratio).
The velocity in the center will be denoted by U,
the velocity at a distance y from the wall by u,
the distance between center and wall by r, the
wall friction by T0. The difference (U—u) will
be called the 'Velocity defect." We find experi-
mentally t h a t for high R.N. the ratio (U—u)/
(r 0 /p)* is a general function of the ratio y/r or
(r — y)/r, i.e., independent of R.N. and also
independent of the roughness of the wall.
A simple interpretation of this result can be
given by comparing cases in which both r 0 and r
have the same values, but the roughness of the
wall is different (Fig. 6). The velocity defect
(U—u) is the relative velocity between the fluid
moving in the center layer and the fluid moving
in a layer at a distance (r — y) from the center.
Rough Smooth Wall
FIG. 6. Comparison of velocity distributions in a smooth
and in a rough channel.
 SKIN FRICTION 7

Considering t h e flow relative t o t h e fluid moving
in t h e center layer, t h e fact t h a t (U—u) is
independent of the roughness of t h e wall and is
proportional t o (ro/p)^ means t h a t t h e action of
the wall friction is independent of t h e way the
friction is produced: whether b y slow displace-
ment of a very rough wall or by fast displacement
of a smooth wall. T h e mechanism of the turbu-
lent exchange seems t o be t h e same in both
cases, except in t h e immediate neighborhood of
the walls.
The same experimental result is true for the
Downloaded by UNIVERSITY OF CALIFORNIA - DAVIS on February 13, 2015 | http://arc.aiaa.org | DOI: 10.2514/8.5
y will contain not only t h e wall friction r 0 and
the density p, but also t h e kinematic viscosity v.
Let us restrict t h e investigation t o the case of
perfectly smooth walls. In this case it can be
expected t h a t near t h e wall u is in fact fully
determined by r 0 , y, p and v. Dimensional
analysis shows easily t h a t t h e only possible
combination is given by t h e formula
u = (T0/p)ig((To/p)iy/v), (19)
where g is another universal function.

flow in a pipe with circular cross section. De-
noting the radius of t h e pipe by r we obtain
(U-u)/(r0/p)i=f(y/r), (18)
where f(y/r) is a universal function.
The quantity (T0/P)^ has evidently t h e di-
mension of a velocity. In aerodynamics it is usual
to characterize a velocity by the corresponding
pressure head; in t h e inverse way we can charac-
terize a certain value of the pressure p by t h e
corresponding velocity (2p/p)^. In a similar
The parameter {r^/p)^y/v, introduced first b y
L. Prandtl, is built u p in the same way as the
R.N.; it contains t h e friction velocity, t h e
distance and t h e kinematic viscosity. We call it
the friction-distance parameter.
T- / x 1 • . velocity
Hence Eq. (19) says the ratio —:—: ——
friction velocity
near the wall is a universal function of the friction-
distance parameter.
7. SIMILARITY OF T H E T U R B U L E N T

way we can assign the velocity (2r 0 /p)^ t o the FLOW PATTERN

shearing stress TO. T h e pressure head corre-
sponding t o this velocity is numerically equal t o
the shearing stress. Instead of (2r 0 /p)^ we will
use ( T 0 / P ) V To be sure it would be more logical
to use t h e former quantity, b u t (rQ/p)^ has been
used already in so many publications t h a t a
change would likely produce confusion. We call
the quantity (r0/p)^ t h e friction velocity. T h e
ratio y/r can be called the relative distance from
the center.
Hence t h e statement expressed by Eq. (18)
can be worded in the following way: T h e ratio
( velocity defect \ .
In order t o determine t h e function f(y/r)
representing t h e velocity distribution over t h e
cross section of a pipe on a channel, t h e author
introduced t h e assumption t h a t the flow pattern
of t h e turbulent interchange in t h e neighborhood
of different points is similar and only different
as far as length and time scale are concerned.
This assumption appears t o be justified in all
cases in which t h e characteristic length of t h e
turbulent exchange is small in comparison with
the dimensions of the cross section.
The following conclusions are deduced from

—:—;
. .

T—:— I is a universal function of the

the above-mentioned assumption:
(a) T h e velocity fluctuations are proportional
friction velocity / to a characteristic length I of the flow pattern
relative distance from the center. and t o t h e slope of t h e mean velocity du/dy.
Let us now consider t h e velocity distribution (b) T h e shearing stress r, variable over t h e
relative t o t h e wall. At t h e wall itself t h e velocity cross section, is a t every point proportional t o
is zero or equal t o t h e velocity of the wall, if the
pi2(du/dy)2. Hence / is heretofore only a relative
wall is moving. Hence, a t the wall t h e velocity
components u' and v' vanish, and the momentum measure of the flow pattern; we can absorb the
transfer is zero, so t h a t t h e friction is merely proportionality factor and define / by writing
2 2
laminar friction. Thus we obtain a "laminar r = pl (du/dy) .
sublayer" adjacent t o t h e wall with gradual (c) The characteristic length is proportional
transition t o the turbulent region. Therefore, the to the ratio | (du/dy)/ (d2u/dy2)\. W e write
velocity distribution as function of the distance I — K\ (du/dy)/(d2u/dy2)\, where K is a numerical
 8 T H . VON K A R M A N

(c) The curve (b) drops near the wall. How-

u v' , fa) l ' • —V ._ • 8 1
ever, Reichardt's experiments are carried out
^ . 3 -*
/ ^ —-y
~&r J 2.
.
11
at a relatively low R.N., so t h a t the range of
2 ^T^
viscosity influence is comparatively large. The
A' / j]\|| estimated range in which fidu/dy cannot be
neglected in comparison with —pu'v' is indicated
°< l ? .:} \
* ."S .( 7 .*} .<3 1.0 by the dotted lines for both cases. The way we
1-y/r
can estimate the thickness of the range influenced
FIG. 7. Correlation between turbulent velocity fluctuations, by laminar friction is shown in the latter part
(a) Wattendorf (C.I.T.). (b) Reichardt (Gottingen). of this section.
The assumption of similarity of the turbulent
flow pattern gives the following formula for the
Downloaded by UNIVERSITY OF CALIFORNIA - DAVIS on February 13, 2015 | http://arc.aiaa.org | DOI: 10.2514/8.5

factor. It will be seen t h a t the constant K occurs

velocity distribution:
in most of the formulae given in the later
sections; it appears as a l'universal constant'' of
the turbulent exchange.
The formula r = pl2(du/dy)2 was first given by
Prandtl in his theory of the mixing length,
H'-0-;)V0-;)'J
assuming a special kind of mechanism for the The formula checks the measured velocity
turbulent exchange of momentum. distributions very well (Fig. 8).
The assumption of similar flow patterns is
identical with the assumption of constant corre-
lation between t h e components of the velocity 16
fluctuations. There is not much experimental
evidence on the correlation of the fluctuations, i5
f~]
but in some cases the magnitude of the fluctua- n T

tions has been measured. Recently F. Wattendorf 1

13
of the California Institute of Technology meas-
ured the value of un over the cross section of a 12
m
rectangular channel with aspect ratio 1 : 18,
11
using a hot wire anemometer suitable for fast
fluctuations. In this case the value of u'v' could 10 i
be deduced from the measurement of t h e pressure Una*'" • o k =0

drop, assuming r= — pu'v' and taking into con- v% * ® k^ 0,1mm

a km0,2mm Je
sideration t h a t the shearing stress r in such a I t © k^O^mm
channel is almost exactly proportional to the ® k^ 0,8 mm 7e
7
distance from the center. Thus the ratio u'v1'/u'2
could be calculated; the results are given in Fig. 6
7 (curve (a)). The points connected by t h e curve /»
5
(b) correspond to similar measurements by
/' >
Reichardt (Gottingen). The curves show the 1/

following results: j
(a) The ratio u'v'/u'2 is almost constant over
2
the main part of the cross section, corresponding
to constant correlation. 1
(b) For the center u'v'/u'*->0; this is evident b*^
by symmetry considerations, since in the center Cy o, 1 0,2 0, 3 0. 9 0,5 0, $ 0,7 0,S G>9 1,0
- i-l
no reason for correlation exists. Accordingly, in ' r

the center part the similarity assumption cannot FIG. 8. Measured velocity distributions compared with
be correct. the theoretical curve.
 SKIN FRICTION
Downloaded by UNIVERSITY OF CALIFORNIA - DAVIS on February 13, 2015 | http://arc.aiaa.org | DOI: 10.2514/8.5

F I G . 9. Universal velocity distribution near a smooth wall.

For small values of y/r, we obtain pattern involves u'/u" = u"'/u'" = • • 3>. There-
fore we put
Umax-u= -(1/K)(T0/P)* log (yM
or (20) 1 = K\ U'/U" I = ny, r 0 = K2pu/2y2
u = const. + (1 A ) (ro/p)* log 3/
and obtain
This result can be seen directly in the following
way: tt = (T 0 /p)*(const. + (l/*) logy). (21)
Let us consider t h e idealized case t h a t a
Comparing Eq. (21) with the general Eq. (19),
constant shearing stress is transferred in a
we obtain
parallel flow along the wall. In this case an
exact similarity of the flow pattern can be tt=(ro/p)*[>+(lA) ^g ((ro/p^A)]. (22)
expected; in fact looking at the flow from the
This equation was given by the present author
point of view of two observers travelling with
in 1930. The numerical value found by the author
the fluid with a velocity equal to the mean
for the universal constant is 0.38. Nikuradse
velocity of the fluid, at two different distances
found later t h a t K = 0.40 fit the experimental
3/1 and 3>2 from the wall, nothing can be different
values better. In Fig. 9 are plotted Nikuradse's
in the flow pictures, conceived by the two
velocity distribution measurements using
observers, except the length and scale of the
u+ = u/(rQ/p)^ and logio j>+ = logio ((r0/p)iy/v)
flow pattern. Thus the characteristic length of
as coordinates. For -y + <110 individual points
the flow pattern is proportional to y. The
are shown, for higher values of y+ a certain
distribution of the mean velocity is, according to
number of measured points are replaced by their
Taylor's theorem, given by
centers of gravity. It is seen t h a t the equation
u(y + Ay) =u(y)+u'Ay + ^u"Ay2+ • • •. u+ = 5.5 + 5.75 logio y+ fits the measurements very
well. The factor 5.75 corresponds to K = 0.40.
It is evident t h a t the similarity of the flow There is a systematic deviation for values of
 10 TH. VON K A R M A N

y+ = (ro/p)^y/v<30. Obviously for small values outside of the laminar range, but near enough
of the friction distance the similarity assumption to the wall so t h a t the logarithmic formula (22)
leading t o t h e logarithmic law cannot be correct. for the velocity distribution is sufficiently exact.
We remember t h a t at the wall itself the flow is In such a region the two Eqs. (18) and (22)
laminar. If we assume a pure laminar sublayer overlap, so t h a t both are valid. Hence we write
corresponding t o the constant value r 0 of the
friction, the corresponding velocity distribution u 1 (ro/p^y
= a+- log, •, (26)
would be given by u= (T 0 /M) -y or
(r 0 /p)^ K v
« = ((Vp)*(Vp)*:yA). i.e., u+ = y^ (23) and at the same time

The two lines representing (22) and (23) meet U-u-u /r~y\
(27)
at t h e point (r0/p)^y/v = 11.5. Hence if there ( T O / pp \ r /
Downloaded by UNIVERSITY OF CALIFORNIA - DAVIS on February 13, 2015 | http://arc.aiaa.org | DOI: 10.2514/8.5

were an abrupt transition between pure laminar

and turbulent regions, the "thickness of the If both equations hold, Eq. (27) for small values
laminar sublayer" would be given by ;y+ = 11.5 of y, must have the form:
or y — di = 11.5v/(ro/p)\ Actually the transition U—u 1 y
is gradual and the value y+ = 30 appears as a = const. — log - , (28)
reasonable limit for the region influenced by (r 0 /p)* K r
laminar friction.
and adding (26) and (28), we obtain
Before the author found Eq. (22), power laws
were used for Eq. (19). If we put U 1 /(ro/p)*r\
- = const.. + - l o g ( - ) . (29)
W=(T 0 /p)*-i4-((r 0 /p)*yA) n » (24) (ro/p) K \ V J

where A is a numerical constant, it is easy to see Let us introduce the friction coefficient referred
t h a t a relation must prevail between the expo- to the maximum velocity by the relation
nent n of the velocity distribution and the Cf=T0/yu* (30)
exponent m in the friction law, mentioned in
Section 4. In fact if we solve Eq. (24) for r 0 , and the R.N. referred to the same velocity
we obtain R=Ur/v. (31)
2 1
r 0 = \pul . (25) Then from Eq. (30) follows immediately
^2/Cn+l) ( w : y/„)2n/(n+l)
(2/ C / )* = const. + (l/iO log, {R{cfY). (32)
Therefore m = 2n/(n-\-l), or n — m/{2—m). For
Eq. (32) can easily be checked by plotting the
w = l / 4 we obtain n = l/7. We mentioned t h a t
values of \/cf* as function of log (R(cf)*). If the
for large R.N. m diminishes, as also does n. In
equation holds, 1/c/* appears as a linear function
Fig. 9 is shown how far the power law formulae
of log (R(Cf)%), and the slope of the straight line
with w = 1/7 and 1/8 represent an approximation
representing 1/c/* as a function of log (i?(c/)*)
to the logarithmic formula in certain ranges.
determines the value of the universal constant /c.
This plotting is shown in Fig. 10. The circles
8. S K I N FRICTION IN SMOOTH P I P E S
represent centers of gravity of groups of Niku-
Combining Eqs. (18) and (22), i.e., the de- radse's measured points. The crosses are repre-
velopments for the velocity distribution starting sentative of mean values taken from measure-
from the center and from the wall, the author ments of Stanton and Pannell. The straight line
obtained in 1930 a rather simple formula con- corresponds to the equation
necting the frictional loss in pipes and R.N., the
1 / ^ = 3.60+4.15 logio (2?fa)*). (33)
first formula which at the same time is based on
theory and checks the experiments. The factor 4.15 in the last term corresponds to
Let us consider the velocity u at a point y the value of the constant K = 0 . 3 9 .
 SKIN
£L\S
in
Ivy
Downloaded by UNIVERSITY OF CALIFORNIA - DAVIS on February 13, 2015 | http://arc.aiaa.org | DOI: 10.2514/8.5
0 same constant is right for the skin friction of
() 2 3 4 5
logjRftj)
FIG 10. Plotting of 1/c/* against log]0 (R(cf)%) for
smooth pipes.
It is easy to show t h a t a formula of the same
form with another additive constant is valid for
the skin friction coefficient referred to the mean
velocity U instead of to the maximum velocity
U. This remark is due to L. Prandtl.
According to Eq. (27), with the exception of a
small region near to the wall, (U—U)/(TQ/P)* is
a universal function of y/r, independent of the
R.N.; integrating (U—u) over the cross section
we obtain
f(U-u)dA = (U-U)-A
so t h a t
. ( E / - t / ) / ( r o / p ) * = const.
or
l / c / * - l / c / * = const.
On the other hand the value of R(cf)$ is
invariable if both R and cf are referred to the
same parameter, i.e., both to the maximum or
both t o t h e mean velocity.
The friction coefficient X used in the hydraulic
practice is defined by the relation
h= \(l/d)(U*/2g),
where h denotes the pressure loss measured in
height, I the length, d the diameter of the pipe.
The factor X is the double of the coefficient Cf.
Nikuradse and Prandtl gave, following the
author's deductions, the formula
l / \ i = - 0 . 8 + 2 logioCRX*).
FRICTION 11
The factor 2 in this formula corresponds to the
value 0.403 for t h e universal constant K. T h e
value of K can be determined from (a) velocity
distribution measurements, (b) tests in smooth
pipes, (c) tests in rough pipes, (d) friction
measurements on plates. Nikuradse gives formu-
lae identical in form to those given by the
author previously for the three cases (a)-(c),
b u t he determines the numerical coefficients
independently for each case. Comparing his
formulae, we find K = 0 . 4 0 for (a), K = 0.403 for
(b) and (c). Plotting his results and those of
Stanton and Pannell and Fromm, we find t h a t
K = 0 . 3 9 fits both items (b) and (c) well; the
6
plates (d). For the velocity distribution near to
the wall /c = 0.40 checks the experiments better,
for the velocity distribution in greater distance
from the wall /c = 0.38 fits better. In view of the
scattering of the measured values and the quite
considerable differences between the measure-
ments of different observers, it seems to the
author t h a t the accordance between the K values
obtained in different ways is quite satisfactory
and substantiates the existence of such a uni-
versal constant entering in all four problems
mentioned.
9. S K I N FRICTION OF SMOOTH P L A T E S
The results obtained and checked by com-
parison with pressure drop measurements have
been applied by t h e author to the computation
of the skin friction of smooth plates. The basic
(34) assumption is t h a t Eq. (32) holds for the turbu-
lent boundary layer, if the R.N. involved in the
equation is replaced by the R.N. of the boundary
layer, R^=Ub/v. The calculation leads to a
formula of the form
2*/c/* = const. + (l/ic) log (Rxcf) (35)
both for the coefficient of the local friction
cf=To/yu2 (36)
and for the coefficient of the total or mean
friction
Cf = F/yiPA, (37)
where A is the surface area of the portion
considered. The R.N. is referred to the velocity
 12 T H , VON K A R M A N

U outside of the boundary layer and the length corresponding to K = 0.395, or

x of the portion of t h e plate considered.
The Eq. (35) settles some questions often 1/C/* = 4.15 logio (iJxC/).
discussed without decisive answer in previous Between the coefficients of the local and the
technical papers on skin friction. It shows for mean friction, Schoenherr found the relation
instance that t h e contribution of equal areas to
the skin friction of a plate diminishes slowly but c / = 0.558C//0.558 + 2(C/)*, (40)
indefinitely with increasing distance from the
where the numerical constant 0.558 represents
leading edge. Also the coefficient of the mean
K(2)K
friction diminishes with increasing R.N., i.e.,
The calculation leading to Eq. (35) can be
with increasing velocity and increasing length.
found in the author's paper presented at the
Formerly it was assumed t h a t both the local
International Congress for Applied Mechanics
and the mean friction coefficients have certain
Downloaded by UNIVERSITY OF CALIFORNIA - DAVIS on February 13, 2015 | http://arc.aiaa.org | DOI: 10.2514/8.5

(Stockholm 1930) or in his paper on the theory

finite inferior limits.
of skin friction in the Proceedings of the Ham-
Concerning the numerical values of the con-
burg conference on problems of ship propulsion
stants involved in Eq. (35), the same method of
(1932). The Eq. (35) is obtained by neglecting
plotting can be used as shown above for the
terms of the order C/* in comparison with
case of the pipes. In Fig. 11 are plotted the
1/Cf*. For moderate values of the R.N. the
terms omitted are not quite negligible, but it
30i T 1 1 1 1 appears t h a t by slight changes in the constants
of the Eq. (35) the whole practical range can be
covered with sufficient exactitude by this simpler
25 y/\
formula. Prandtl repeated the calculation in a
1
4 \ 20
\A \ slightly different way and gave the results of
his calculation in the Gottinger Ergebnisse of
\ X 1932. His results are very similar to those ob-
l5J tained by the present author.
Eqs. (38) and (39) have the disadvantage t h a t
they cannot be analytically resolved for cf and
Cf. The Table I gives the values of cf and Cf as
function of R.
F I G . l i . Plotting of.l/c/$ against logio (Rcf) for
TABLE I.
smooth plates.

R* C/ Cf R* Cf Cf
values of the local friction coefficients according 5
0.00614 0.00480 7
0.00222
2X10 2X10 0.00263
to measurements of G. Kempf. These experi- 3 562 444 3 247 216
ments are very remarkable because the values 4 529 419 4 237 210
5 506 402 5 229 196
of the local friction were obtained by direct 7 473 379 7 218 187
measurements of t h e force acting on small 10° 441 357 108 207 179
1.5X106 408 332 2X10 8 189 163
movable plates arranged at different points of a 2 387 317 3 179 156
very long pontoon. 3 360 296 5 167 145
4 342 283 109 155 135
The equation of the straight line representing 5 330 274 2X10 9 141 125
the measurements is 7 311 260 5 127 113
107 293 245 1010 118 105
1/C7* = 1.7+4.15 logio (RxCf). (38)

E. Schoenherr found t h a t the coefficient of the Schlichting suggested that the Eq. (39) can
mean friction Cf is given by be replaced by an empirical formula C/ = 0.455/
(logio Rx)2'r°8- This formula replaces the Eq. (39)
0.242/C/* = log10 (1?*C/), (39) fairly well in the range 1 0 6 < i ^ < 109.
 SKIN FRICTION 13

0.007

• THEORY OF TURBULENT
BOUNDARY LAYER (KARMAN 1930)
0.006
Downloaded by UNIVERSITY OF CALIFORNIA - DAVIS on February 13, 2015 | http://arc.aiaa.org | DOI: 10.2514/8.5

0.002

FIG. 12. Skin friction of smooth plates compared with the theory.

10. DISCUSSION OF T H E E X P E R I M E N T S . The transition between laminar and turbulent

TRAN-
SITION B E T W E E N LAMINAR AND boundary layers along a flat plate was investi-
TURBULENT REGION gated by Burgers, Hegge Zynen, Hensen, Fage,
Fig. 12 represents t h e results of t h e most and recently by Schoenherr. The ideal case of a
important test on skin friction of smooth plates. mathematically thin plate can be approximated
The experimental points are, in general, replaced by using plates with sharp leading edges. I t is
by curves connecting t h e mean values. In to be expected t h a t under similar conditions the
addition, t h e theoretical curve for the coefficient transition depends on the R.N. of the boundary
of t h e laminar skin friction and t h e curve for layer R8 = Ub/v. Hegge Zynen and Burgers found
turbulent skin friction corresponding t o Eq. (39) that in their wind tunnel experiments the critical
are drawn. value of R?> was between 1650 and 3500; Hansen
T h e tests of Wieselsberger, Zahm and Gibson found the value R$c = 3100. I t appears t h a t there
were wind tunnel tests on stationary models; all is no definite characteristic critical value of R&
other tests were made by towing flat plates in for the transition; the value R8c depends on t h e
water. In all tests except those made by G. amount of perturbations, for instance on t h e
Kempf the total force acting on the model plate turbulence in t h e outside stream. Schoenherr
was measured and a correction was made for the found t h a t t h e transition was substantially
additional drag due t o the finite thickness of the facilitated by producing artificial turbulence in
plate. In K e m p f s tests t h e local friction was the water upstream from t h e leading edge. H e
measured and t h e total friction computed by obtained the transition curves (b) and (c) shown
integration. in Fig. 13 under normal conditions and t h e
It is interesting t h a t in t h e range 2 X l 0 5 < i £ curve (a) for artificial turbulence produced by
<10 6 the results of the wind tunnel tests check roughening t h e forward edge of t h e plates. H e
with the theoretical values for turbulent friction also found t h a t if artificial turbulence is produced
and the tests of the Froude national tank check in the water upstream from the leading edge t h e
approximately with t h e theoretical values for skin friction values, from R = 2XlOb on, fall on
laminar friction. Geber's results in t h e range the curve for pure turbulent friction, as in t h e
1 0 6 < i ? < 1 0 7 show transitions between t h e two tests of Zahm and Wieselsberger. In general,
theoretical curves. rounding the leading edge has a similar influence.
 14 TH. VON K A R M A N
.vvy»v-q
Downloaded by UNIVERSITY OF CALIFORNIA - DAVIS on February 13, 2015 | http://arc.aiaa.org | DOI: 10.2514/8.5
FIG. 13. Transition between laminar and turbulent friction (smooth plates).
The range of critical values of R§ can be
assumed to be between 1600 and 6000.
Assuming t h a t the transition takes place
abruptly at a certain value of Rs a family of
"transition curves" for Cf can be calculated as
function of the R.M. of the plate with the
critical value R?>c as parameter. Such curves are
drawn in Fig. 13. Curves I and II correspond
to somewhat different assumptions. In both cases
a laminar layer is assumed, until a certain value
of Rs is reached. T h e turbulent layer beyond t h e
transition point is calculated in case I, as if it
had started with t h e thickness zero at the
leading edge; in case II it is assumed t h a t at the
transition point the momentum carried by the
boundary layer remains unchanged. Geber's
experimental curves fit the calculated curves
quite well. However, in view of the Schoenherr
curves it must be remarked t h a t the transition
phenomenon is probably too unstable to be
represented by a simple equation.
11. PRACTICAL APPLICATIONS
Predictions of t h e maximum speed of airplanes
based on wind tunnel tests involve a knowledge
of the "scale effect" for the drag. Induced drag
has practically no scale effect, also the drag of
protruding parts follows fairly well the so-called
quadratic law, at least beyond their critical R.N.,
which is connected with a more or less sudden
drop of the drag coefficients. Thus the greatest
uncertainty is t h a t due to the scale effect on
skin friction, i.e., the gradual drop of the friction
coefficient with increasing R.N. The range of
extrapolation is comparatively large in the case
of modern fast airplanes. The comparison can
be based on the R.N. of the wing defined in terms
of the speed, the mean chord of the wing and the
kinematic viscosity. Present atmospheric wind
tunnels, with the exception of the Langley Field
full scale tunnel, reach values of about R = 2
X10 6 , t h e variable density tunnel about 3X10 6 .
The full scale values for large modern transport
airplanes may reach 25X10 6 , somewhat smaller
values corresponding to fast military planes.
Hence there is a large change in the skin friction
coefficient which influences the predicted speed
very noticeably, especially because, in the case
of modern clean airplanes, t h e skin friction may
represent the major portion of the total drag.
An exact calculation is made difficult by two
facts:
 SKIN
(a) The separation of the skin friction and the
drag of protruding parts is hardly possible.
(b) The skin friction acts partly on the wings,
partly on the fuselage and tail; these parts have
different R.N. and so somewhat different scale
effects.
At the California Institute of Technology the
following procedure, suggested by Clark B.
Millikan, is in use: Two different values for the
full scale drag coefficient are calculated, a first
value based on the assumption t h a t the total
parasite drag follows the skin friction law for
Downloaded by UNIVERSITY OF CALIFORNIA - DAVIS on February 13, 2015 | http://arc.aiaa.org | DOI: 10.2514/8.5
smooth plates, and a second value corresponding
to the assumption t h a t the profile drag of the
wings follows the law mentioned, while for the
remainder of the drag the scale effect can be
neglected. I t is believed t h a t these values consti-
t u t e limits between which the value of t h e actual
drag should lie.
As an example the following data are quoted,
taken from the results of actual wind tunnel and
flight tests on an observation type, wire braced,
gull wing monoplane.
Model results scaled up without R.N. correction:
CDp = 0.0396, calculated F m a x = 180 m.p.h.
Wing profile drag only corrected to full scale:
CD =0.0364, calculated F m a x = 186 m.p.h.
Total parasite drag considered as turbulent skin
friction and corrected to full scale:
CD =0.0268, calculated F m a x = 205 m.p.h.
Flight tests on the actual airplane gave
T / max = 196 m.p.h., i.e., almost exactly halfway
between the two extrapolations. The relative
distance from the two limits should, in general,
be capable of estimation from the general
* 'cleanness" of the airplane investigated. More
refined methods of making such estimates have
recently been developed by Millikan and will be
presented in a paper t o be published by him in
the near future.
Another important application of the skin
friction theories is an approximate prediction of
the boundary layer thickness or of the velocity
distribution near the surface of plates or stream-
line bodies. To be sure in the case of streamline
bodies only a rough approximation can be
obtained, because the ''development of the
boundary layer" is influenced by pressure distri-
FRICTION 15
bution and curvature along the surface. How-
ever, in many cases, an approximate calculation
is of great value, for instance for an estimate of
the drag of protruding parts in the boundary
layer region, for an estimate of the influence of
small roughness, or of t h e influence of the
velocity field near the hull on t h e behavior of
fins, rudders and propellers of airships.
A simple formula for the boundary layer
thickness can be obtained by comparing the two
expressions for the coefficient of the local skin
friction of a plate at the distance x from the
leading edge, according to the Eqs. (33) and (38).
We have the following two relations:
l/C/* = 1.7+4.15 1ogio (i?,C/) (41)
according to (38) and replacing in (33) R by RB
1/CV = 3.6+4.15 log (U«(C/)*). (42)
By subtracting the two expressions it is seen
t h a t the value R8/Rx(Cf)i2 is constant and
R8 = 0.38Rx(Cf)* or 5 = 0.38x(C/)l (43)
The value of Cf as function of Rx is tabulated
on page 12; hence t h e value of 5 can easily be
computed. The velocity distribution in the
boundary layer is approximately given by the
formula
tt=C7[l-4.15(C/)Mog10(5/y)]. (44)
Let us calculate, as an example, the boundary
layer thickness at the trailing edge of an airfoil;
with chord of 3 m = 9.8 ft. for the speed U= 100
m/sec. =223.7 m.p.h. The corresponding R.N.
(for "standard" air) =20,000,000 = 2 X10 7 ; Cf
= 0.00222, accordingly 6 = 5.4 cm.
12. S K I N FRICTION OF BODIES OF REVOLUTION
(AIRSHIPS)
The application of the boundary layer theory
to bodies of revolution is interesting for analyses
and predictions of airship drags. The resistance
of a bare airship hull is almost entirely due to
skin friction, the pressures at the bow and the
stern being practically balanced. Early wind
tunnel tests, carried out at comparatively low
R.N., appeared to be inconsistent with one
another because the transition between laminar
and turbulent flow was not taken into consider-
ation. The airstreams in different wind tunnels
 16 T H . VON K A R M A N
have very different degrees of turbulence and so
the transition occurs in quite different ways in
different wind tunnels.
B. M. Jones and H. L. Dryden independently
showed t h a t such differences in airstream turbu-
lence could account qualitatively for the wind
tunnel discrepancies. In their considerations a
figure of revolution was replaced by an "equiva-
lent flat plate" and the existing information on
the flow along flat plates was thus applied to
their problem. Shortly afterwards C. B. Millikan
applied the boundary layer integral relation
Downloaded by UNIVERSITY OF CALIFORNIA - DAVIS on February 13, 2015 | http://arc.aiaa.org | DOI: 10.2514/8.5
(analogous to Eq. (2), but taking the variation
of pressure along the surface into account) to
the case of axial flow about a figure of revolu-
tion and gave expressions for the variation of
boundary layer thickness and skin friction along
such a body, assuming a parabolic velocity
profile for the laminar portion of the boundary
layer and a " l / 7 t h power" profile for the
turbulent portion. The calculations were carried
out in detail for the so-called N.P.L. models and
curves of drag coefficient vs. R.N. for various
degrees of airstream turbulence were obtained.
These gave a satisfactory quantitative agreement
with many of the experimental curves-, so t h a t
the explanation of the latter's inconsistencies
was in this way successfully accomplished.
Miss H. M. Lyon has carried the experimental
side of the problem still further in an elaborate
investigation of the correlation between the drag
coefficient vs. R.N. curves for two airship hull
models and the degree of turbulence produced
artificially in the airstream.
Freeman repeated Millikan's calculations for
the hull of the Akron dirigible and obtained a
remarkably close agreement with his beautiful
measurements on a l/40th scale wind tunnel
model, both as to drag coefficient and distribu-
tion of boundary layer thickness. For the R.N.
of his measurements the l / 7 t h power law used
in Millikan's theory was shown to be a very
satisfactory approximation to reality. However,
for the extrapolation of drag values to full scale
and for full scale boundary layer calculations
the l / 7 t h power law will almost certainly be a
much less satisfactory approximation, so t h a t it
would be very desirable to repeat the analysis
by using our improved knowledge of the laws of
skin friction.
13. T H E INFLUENCE OF R O U G H N E S S
The problem of the influence of surface rough-
ness on the skin friction is rather complex.
Although many experiments have been con-
ducted concerning the friction in rough-walled
pipes, considerably fewer tests are available on
skin friction of rough plates in a free stream.
In pipes with different diameters but with
identical or geometrically similar surface condi-
tions the coefficient of skin friction depends on
the R.N. and the "relative roughness." The
relative roughness is the ratio between a length
which is characteristic of the roughness and a
length characteristic of the cross section (for
instance, the hydraulic radius). It is not possible
to define a priori which dimension is to be
considered characteristic of the roughness, except
t h a t it must be the same for identical surfaces
and for walls with geometrically similar kinds
of roughness proportional to the scale of the
protruding parts (for instance, to their mean
height). Denoting this parameter by k and the
radius by ry the friction coefficient will be
cf = F(R,k/r). (45)
The general form of this function is not known,
in fact it is not even sure t h a t one universal
function covers the case of all kinds of rough
walls. It is more probable t h a t it depends on the
shape of the roughness. Fortunately, the experi-
mental evidence shows t h a t the conditions are
relatively simple for high R.N. and compara-
tively large relative roughness. It appears t h a t
with increasing R.N. the skin friction coefficient
becomes independent of the viscosity and in turn
independent of the R.N. In this respect the
behavior of rough surfaces is fundamentally
different from t h a t of a smooth surface, since in
the case of the smooth surface the friction
coefficient decreases indefinitely with increasing
R.N.
For the range in which the friction coefficient
is independent of the R.N., a simple relation
between the friction coefficient and the relative
roughness can be found, based on the velocity
distribution laws given in Section 6. We re-
member t h a t the ratio (velocity defect/friction
velocity) was found to be a universal function
of the ratio (distance/radius), and it has been
 SKIN
noted t h a t this theorem holds also in the case
of rough surfaces. Accordingly, denoting again
t h e maximum velocity by U, t h e velocity at the
distance y from t h e wall by u, and the wall
friction by r 0 , we write
(U-u)/(r0/P)i=f(y/r).
T h e development for t h e velocity distribution
starting from t h e wall will be different in the
cases of smooth and rough walls. For smooth
walls we assume t h a t u is a function of y, r 0 , p
and fji. In t h e general case of a rough surface u
Downloaded by UNIVERSITY OF CALIFORNIA - DAVIS on February 13, 2015 | http://arc.aiaa.org | DOI: 10.2514/8.5
will be a function of these quantities and also
of the roughness parameter k. However, the
fact t h a t for large R.N. the skin friction coeffi-
cient is independent of the viscosity indicates
t h a t the velocity distribution is also independent
of v and hence u depends only on y, r 0 , p and k.
Dimensional analysis shows t h a t the only form
possible for this functional relation is
* = (TO/P)W*).
On the other hand, we obtained from the
similarity of the turbulent flow pattern the
relation
w=(T 0 /p)*(const. + (l/K') logy)
so t h a t due to (43) the development for the
velocity distribution starting from the wall will be
« = (T 0 /p)*(const. + ( l A ) log (y/k)).
Combining Eqs. (46) and (49) we obtain, by a
reassumption similar to t h a t applied in Section 8
tf=(r0/p)*(const.+(1/K) log (r/k))
and
(2/C/)* = a + ( l A ) l o g ( r / i f e ) .
It is easy to show (compare Section 8), t h a t the
same equation holds for the coefficient referred
to t h e mean velocity.
T h e Eq. (51) was found by the author in 1930,
by a somewhat different reasoning from t h a t
given above. T h e equation can be checked
independently of the definition of k by tests in
pipes or channels with walls of identical surface
conditions.
Let us write Eq. (51) in the form
2Vc/* = a - ( l A ) logib + (l/ic)logr.
Then obviously, if we plot 1/c/* against log r, we
FRICTION 17
get parallel straight lines, each line corresponding
to a given roughness. The slope of these lines is
determined by t h e universal constant /c, which
was encountered also in the case of smooth pipes
and smooth plates.
In Fig. 14 are plotted results of tests made by
(46)
Fromm and Nikuradse. With one exception they
agree with t h e foregoing conclusions. The ex-
(47)
(48)
10 mm 50 100
r
(49)
FIG. 14. Plotting of 1/C/i for channels of different widths.
(50) ceptional case refers to a test in which a surface
with extreme roughness (sawteeth of 1.5 mm
height) was used.
(51)
The vertical distance between any two such
straight lines is determined by the difference of
the terms (l//c) log*, i.e., by the ratio of the
roughnesses. Thus, by choosing one definite
roughness as a basis for comparison, it is possible
to assign a definite value to any surface. Niku-
radse produced geometrically similar rough sur-
faces for all of his tests, making use of sand
grains of various diameters. He found t h a t by
determining the roughness by Eq. (52), the
roughnesses of the surfaces used by him are
simply proportional to the diameter of the grains
(52)
used. Thus we can tentatively use as a unit of
roughness the grain diameter of an equivalent
 18 TH. VON K A R M A N

1 .005

° FROM M
— .004 x/te 4,000

o NINUR ADSE
gpoo
iapoo
.003
}ff C
sspoa
f zmw\
.002-
ojajjl
.00!
10

10° I0' R I0 8 I0"

FIG. 16. Local friction coefficients for rough plates.

Downloaded by UNIVERSITY OF CALIFORNIA - DAVIS on February 13, 2015 | http://arc.aiaa.org | DOI: 10.2514/8.5

These values give a fair estimate of the influence

5
5 10 50 100 bt\o 1000 of roughness on the coefficient of skin friction
for plates. Fig. 16 shows these values of cf for
FIG. 15. Plotting of 1/C/* against relative roughness. various roughnesses, compared with the curve
for smooth plates. For a more exact calculation,
it must be taken into account t h a t in general
granular surface. Fig. 15 shows the results of the the local skin friction will be a function both of
Fromfn and Nikuradse tests combined into one the boundary layer R.N. and of the relative
straight line, which has the equation roughness. Experimental evidence for the range
in which both the R.N. and the roughness have
l/ C / * = 3.0+4.15 log (r/k). (53)
an influence on the friction is somewhat contra-
The next step is the application of Eq. (51) to dictory, even in the case of pipes. However, the
the problem of skin friction of rough plates, in interpretation of the experimental results is
the same manner as we applied Eq. (33), deduced relatively easy in the case of the artificial
for the skin friction in smooth tubes, to the case granular roughness used by Nikuradse. He found:
of smooth plates. We must replace the radius r (a) t h a t the influence of roughness is, for all
in Eq. (51) by the boundary layer thickness. practical purposes, null below a certain R.N.,
First, we notice t h a t the * 'relative roughness" and t h a t the limiting value of the R.N. increases
k/8 which determines the friction is variable for with decreasing k.
constant k, it is large for small values of <5 (i.e., (b) t h a t beyond this limit the friction coeffi-
near the leading edge), and decreases with cient, for a given roughness, as a function of the
increasing distance from the leading edge. Hence R.N., first decreases, then slowly increases and
the validity of such a calculation is restricted to becomes constant at a certain value of the R.N.,
cases in which the roughness k is so large t h a t which value also depends on the roughness.
the assumption of a friction independent of the Considering the two limits (the limit at which
R.N. holds. the influence of the roughness appears, and the
We obtain, under these conditions, the equa- limit beyond which the friction coefficient is
tion for the local friction coefficient: constant) Nikuradse's tests show t h a t both
correspond to definite values of (ro/p)^k/v. This
1 / V = 5.8+4.15 log ((x/k)cfi). (54)
can easily be understood: Let us remember t h a t
As might be expected, the friction depends on in the case of a smooth wall, we have in the
the ratio k/x (roughness relative to the length). immediate neighborhood of the wall a laminar
In the following table are calculated a few region, whose outer limit or "the thickness of the
corresponding values of cf and x/k: laminar layer" 81 is proportional to P/(TO/P)*;
thus the parameter (r0/p)^k/v is proportional to
C/ = 0.0040 0.0035 0.0030 0.0025 0.0020
k/81 or to the ratio between the roughness and
3 3 3 3 3
X / £ = 4X10 8X10 18X10 53X10 220X10 the thickness of the laminar sublayer.
 SKIN FRICTION 19

According to the tests, the influence of the TABLE III.

roughness starts with (ro/p)^k/v = 3, and the
friction coefficient becomes constant for (r 0 /p)^fe/ X k X k
i>>60. We remember t h a t 8 ^ 1 1 . 5 V / ( T O / P ) * gives 0.075 m 0.0230 m m 2 m 0.0306 m m
0.1 0.0237 2.5 0.0311
about t h e "thickness" of t h e laminar sublayer. 0.25 0.0257 3 0.0316
Thus it appears t h a t the influence of roughness 0.5 0.0276 10 0.0344
1.0 0.0290 50 0.0385
becomes noticeable if t h e height of the roughness 1.5 0.0299 100 0.0400
has the same order of magnitude as the laminar 200 0.0418
sublayer {k/di>\) and t h a t t h e viscosity has no
more influence if t h e roughness is large in harmful at t h e rear portion of the surface. This
comparison with this thickness ( K / 5 Z > 6 ) . has been proved experimentally by G. Kempf
Downloaded by UNIVERSITY OF CALIFORNIA - DAVIS on February 13, 2015 | http://arc.aiaa.org | DOI: 10.2514/8.5

Accepting this evidence for t h e influence of and corresponds to common experience.

roughness with t h e limiting value (TQ/p)^k/v = 3, Another consequence which can be drawn from
we can carry out a quite interesting calculation Eq. (55) is t h e fact t h a t t h e allowable rough-
indicating t h e magnitude of "allowable rough- ness k depends chiefly on t h e speed and only
ness" in practical cases—"allowable" in t h e to a smaller degree on t h e R.N. This enhances
sense t h a t roughness below this limit does not the reliability of wind tunnel tests, since it shows
influence the friction of a plate. t h a t in model scale-reduction it is not necessary
According t o this condition the limiting value to reduce k in the ratio of the length scale, but
of k is given by only in the ratio in which cp is increased.
Unfortunately the evidence of tests on pipes
* = 3 y /(ro/p)* = 3(v/t7)(2A/)*. (55)
with different roughnesses produced by various
Now, for smooth plates cf is a function of the paints or by different manufacturing methods is
Rx and is calculated on page 12. Using these not so consistent as are Nikuradse's tests on
tabulated values, we obtain for the case of an artificially roughened pipes. Some of these other
airfoil with 3 m chord and 100 m/sec. speed tests are very exact, hence it is probable t h a t
(x = distance from t h e leading edge), Table I I . Nikuradse's conclusion as to the existence of a
universal "roughness function" does not cover
TABLE II.
all kinds of roughness. I t is probably t h a t the
one parameter k is not sufficient in itself to
X k X k characterize the surface; at least one more
0.075 m 0.0100 m m 1.5 m 0.0129 m m
parameter, such as the mean distance between
0.1 0.0103 2.0 0.0131 protruding parts, must enter into the picture.
0.25 0.0111 2.5 0.0133
0.5 0.0118 3.0 0.0135
At the present time it is not possible to compute
1.0 0.0125 the skin friction for a plate with a given rough-
ness.
The author believes t h a t the main problems
Similarly, for an airship of 200 m length and of future research in t h e field of skin friction are:
speed of 40 m / s e c , we have Table I I I . (a) to work out a method of determining a
It is interesting t o notice t h a t recent experi- reliable "roughness scale" for different kinds of
ments of t h e N.A.C.A. give t h e same order of materials used in practice,
magnitude for t h e allowable roughness of an (b) to investigate, by systematic experiments
airfoil. on rough surfaces, t h e transition range, i.e.,
It is evident t h a t a certain small roughness is the range in which both the R.N. and t h e
more harmful near the leading edge and less relative roughness influence t h e friction.
 20 T H.
S O M E P A P E R S ON S K I N F R I C T I O N AND BOUNDARY L A Y E R
Abell and Lamble, The Resistance of Lapped Butt Joints of
Ships Shell Plating to Motion through Water, Liverpool,
1931.
Buri, Dissertation, Zurich, 1931.
Cuno, Z.F.M. 23, No. 7 (1932).
Dearborn, N.A.C.A. Tech. Note No. 461.
Ebert, Z . F . M . 1933, p. 629.
Eisner, Hydromechanische Probleme des Schiffsantriebs,
H a m b u r g (1932), pp. 1-49.
Fage, J. Roy. Aero. Soc. 37, 573-600 (1933).
Freeman, N.A.C.A. Report No. 430.
Gruschwitz, Ing.-Arch. 2 (1931).
H e r m a n n a n d Schiller, Ing.-Arch. 1, No. 4 (1930).
Downloaded by UNIVERSITY OF CALIFORNIA - DAVIS on February 13, 2015 | http://arc.aiaa.org | DOI: 10.2514/8.5
Hooker, N.A.C.A. Tech. Note No. 457.
von K a r m a n , Proc. I l l Intern. Cong. Appl. Mech.,
Stockholm, 1930. Hydromechanische Probleme des Schiff-
santriebs, pp. 50-73.
von K a r m a n and Millikan, paper read before t h e Aero-
nautic Division of the A.S.M.E., J u n e , 1933.
VON KARMAN
SINCE 1930
Kempf, Hydromechanische Probleme des Schiffsantriebs,
pp. 74-82.
Lerbs, Werft-Reederei-Haften, 17 (1930).
Lyon, Brit. Rep. & Mem. No. 1511.
Millikan, Trans. A.S.M.E. 1932.
Mueller, Werft-Reederei-Haften, No. 4 (1932).
Nikuradse, Proc. I l l I n t e r n . Cong. Appl. Mech., Stock-
holm, 1930. V.D.L Forschungsheft, 356 (1932). V . D . I .
Forschungsheft, 361 (1933).
Prandtl, Zeits. V.D.L 77, No. 5 (1933). Hydromechanische
Probleme des Schiffsantriebs, p. 87. Gottinger Ergebnisse,
IV Lieferung.
Schiller, Aachener Vortrage (Springer, Berlin), 1930. H a n d -
buch d. Exp. Physik 4, 2 and 4 (1932).
Schoenherr, paper read before Soc. Naval Arch. • and
Marine Eng., Nov., 1932.
Schrenk, Luftfahrtforschung 2, 1 (1928).
Tollmien, T r a n s . A.S.M.E., 1933.