Turbulent far wake of a body of arbitrary shape

M. Le Berre and Y. Pomeau

Laboratoire d’Hydrodynamique, CNRS, École polytechnique, Institut Polytechnique de Paris, 91120 Palaiseau, France

Abstract – Using a recently proposed closure for the Boussinesq-Reynolds turbulent stress we re-
derive Prandtl’s result for the turbulent wake width behind blunt bodies. We show that it is valid
for a body of arbitrary shape, but only for wakes due to a drag oriented parallel to the incoming
flow. If the geometry of the body resisting the flow lacks symmetries forbidding it, the force on
the body has also a component orthogonal to the incoming velocity, called lift, that allows, for
instance, planes to fly. Using our closure theory we show that this lift generates a wake wider
than the drag, with a width growing like the one-half power of the distance to the body, instead
of the one-third in the case of drag only. This has important consequences as it leads to a change
of structure of the wake that remains turbulent all the way to infinity.

Introduction . – An important field of application of same order of magnitude2 , and then obey the same scaling
theories of turbulence is the turbulent wake of blunt bodies law (1). This is the point we shall debate in this letter,
in a uniform flow at large Reynolds number. Wakes have because this assumption is unsupported by our analysis of
long been a topic of studies beginning with the derivation the balance of momentum.
of the quadratic law of drag by Newton. Although New- As in [2] we assume that the 3D wake becomes axially
ton assumed indirectly that the wake has constant width, symmetrical at x large, although we are aware that this
Prandtl showed this width grows slowly downstream and is still a debated point. This property could be improper
derived a law for this growth in an infinite medium. Both especially in the case of lift, however it is justified here
Prandtl and Landau explain the slow expansion of the since our aim is to derive order of magnitudes.
wake at long distances by using the property1 that at large Since Prandtl and Landau more stringent conditions
distances from the blunt body the change in flux of mo- have been set up to derive this law, see for instance [3]
mentum due to the blunt body and across a plane perpen- which predicts another scaling for a 2D plane wake be-
dicular to the wake is equal to the force on the blunt body, hind an obstacle of height a in the y-direction, infinitely
as a result of the conservation of momentum. Depending elongated in the z-direction
on the orientation of the force on the blunt body it is not
surprising that the law of widening differs if the force is h2D (x) ∼ (xa)1/2 (2)
perpendicular or parallel to the constant velocity at infin-
ity, as we show. According to Prandtl [1] and Landau [2], Generalizing the derivation to 3D axisymmetric wake,
the width of the wake behind an obstacle of cross section these authors confirm the scaling (1). Both results will be
area a2 , grows with distance x like called ”classical” in the following, although recent mea-
surements of the wake properties behind thin bodies hav-
h(x) ∼ (xa2 )1/3 (1) ing different shapes, from square to snowflakes [4], [5] were
interpreted as better in favor of the relation (2)3 .
where x is the (large) distance of the point of measurement Below we analyse the problem of the asymptotic turbu-
to the blunt body along the uniform velocity creating the lent wake properties of a body of arbitrary shape, either
wake. Let us precise that the scaling laws considered here blunt body or streamlined one, with the help of a recently
concern high Reynolds number conditions (above the drag presented closure expressing the turbulent (or Reynolds)
crisis value). Let us notice that Landau’s derivation as- 2 More precisely Landau [2] justified his derivation via two incom-
sumed (after Prandtl) that the lift and drag forces are of patible equations (u is the transverse component in Eq. 37.1 whereas
it is the longitudinal component of the velocity defect in Eq. 37.2).
1 named below the Newton condition, because Newton clearly in- 3 This result was deduced from dissipation measurements because

dicates that the force on the body results from the equality between it was impossible to discriminate relations (1) and (2) directly from
two vectors their width measurements.

p-1
M. Le Berre et al.

stress as a function of the average velocity field by an in- In these equations, a sum on the same index is implied,
tegral over space of this velocity field, see eq. (6). The and p is the time average of the molecular pressure (the
validity of this closure is not limited to 3D axially sym- pressure involved in Navier-Stokes equation) called be-
metrical wake (far from the body) and can be used for low ”molecular pressure” and labelled as pmol . Setting
∂
non axisymmetric wake due to a lift force on this body. ∂i = ∂x i
for the i-component of the gradient operator, the
We refer the interested reader to a recent publication ex- condition of incompressibility
plaining the basis of the closure equations used here [6].
This closure does not introduce extra physical parameters ∂i ui = 0. ∂i u′i = 0. (5)
besides the one of the geometry of the flow and the bound-
is satisfied thanks to a convenient choice of the pressure
ary conditions on the body (of arbitrary shape). Therefore
field p, a subtil point discussed below. The closure ex-
the computation of the far wake could be done explicitly
presses the Reynolds stress tensor σij = ρ u′i u′j as a func-
including cases where the force on the body is not parallel
tion of the averaged flow field to close the equations. Our
to the incoming flow. The most obvious example of that
”closure” set the Reynolds stress
is the lift felt by a flying aircraft that is perpendicular to
its velocity, plus an undesired drag. Therefore it makes
Z
sense to look for this situation of the wake generated by σij (x) = ργ|∇×u(x)| dx′ K(x, x′ )(∂i uj +∂j ui )(x′ ) (6)
a conjunction of what we call the ”drag”, parallel to the
incoming velocity, and the ”lift”, perpendicular to it. We for i ̸= j. For i = j we add to the latter expression the
show that, within our closure, the decay of the wake is positive term
different in the two cases (lift and drag). We find that Z
the width of the wake due to lift increases more (with the σ (x) = ργ |∇ × u(x)| dx′ K(x, x′ )|∇ × u(x′ )|. (7)
p ′

distance from the body) than the width of the wake due
to drag. This agrees also with an early observation [7]- where γ ′ may be different from γ, although we set γ ′ = γ
[8] that turbulent wakes are sensitive to the shape of the in the following because they are of same order of magni-
blunt body, including at quite large distances. This could tude, see the discussion below eq.(11). This term is added
be explained by contributions of the lift and drag part [6] in order to ensure that the Cauchy-schwarz inequal-
of the wake spreading with different power laws at large ity ( u′i u′j )2 ≤ (u′i u′i )( u′j u′j ) is fulfilled. In 3D free space,
distances. the kernel K for a pointwise source located at x = 0 is
For the drag case, our prediction agrees with the classi- the Green function from which 1/|x′ | can be subtracted
cal ones, its interest lies in the fact that the assumptions to make the integral over x′ converging
made in the classical derivation are here deduced from an
1 1
explicit closure and the momentum balance, for 3D and K(x, x′ ) = − (8)
2D flows. Moreover the main result of this work concerns |x − x′ | |x′ |
the lift characteristics deduced by the same procedure. Note that this expression represents also that the turbu-
Our approach. – In the frame of reference of the lent fluctuations are absent at x = 0. In presence of an
body let us set U + u as the total time averaged velocity obstacle the source of turbulence is not pointwise, there-
where U = U0 ⃗ex is the constant fluid velocity at infinity fore a more complex Kernel should be used to describe
supposed to be along the x axis and u vanishes at in- the drag originating from more or less unstable viscous
finity in any direction This total averaged velocity splits boundary layers formed on the blunt body surface. Lastly
into two components, one parallel to the x direction, that γ is a pure number related to the von Karman constant.
is dominated by the velocity at infinity, and two compo- The equation for the stress imposes also that wherever the
nents perpendicular to x, uy and uz , reducing to the radial average velocity is potential there is no turbulence and no
component of the average velocity ur in the axisymmetric Reynolds stress. This accounts for a basic feature of tur-
case without swirl (uθ = 0), or to uy for 2D flows. The bulence that is fundamentally linked to amplification of
velocity fluctuations are labelled as u′ . vorticity by various instabilities. The absolute value of
the rotational term in eqs. (6)-(7) is to represent that the
As classically our starting point are the equations for
Reynolds stress changes sign as the average velocity is re-
the average balance of momentum, written in cartesian
versed, as in the Newton’s drag law,
coordinates in the form
F = Cρ|U |U a2 . (9)
∂i Σij = 0. (3)
where U is the uniform velocity of the flow impinging on
where Σij is the stress tensor, a sum of three terms in the the blunt body in the x direction, the coefficient C is a
large Reynolds number limit where the viscous stress is pure number depending on the shape of the blunt body
negligible, and weakly of the Reynolds number or not at all of it in the
limit of very large values, and the traditional cross section
Σij = ρ(Ui + ui )(Uj + uj ) + ρ u′i u′j + p δij . (4) area of the blunt body is here replaced by a length square

p-2
 Title

a2 . In the following the couple (C, a) becomes (Cd , ad ) below that both (A) and (B) methods lead in fine to the
or (Cℓ , aℓ ) whether it concerns the drag or the lift force. same solutions for the velocity field. Note also that there
Let us emphasize that the non analyticity reflected by the is a way to bypass the problem arising from the unknown
absolute value is a consequence of the occurence of sin- pressure term, which consists in taking the rotational of
gularities in the solutions of the Euler equations, a basic the vector equations (3). This yields equations without
feature of our closure model. pressure at the cost of higher order derivatives. Here we
To derive explicit consequences of our theory for the shall not do that. Our choice is to look at the balance
turbulent stress, one should solve eq. (3) for the balance equation. Moreover we consider both (A) and (B) cases,
of momentum, something that is not easy to do: this is in order to understand the role of σ p .
a nonlinear integral equation and it has to be solved to-
3D axisymmetric far wake . – We consider the 3D
gether with the incompressibility condition. However this
case of a body with comparable size in the y and z direc-
equation carries important properties of invariance under
tions, and assume that the wake becomes axisymmetric
change of scale that allows to perform a scaling analysis
at large distance x from the body with a radial expan-
of its solutions just by inspection of the way the various
sion h(x). Therefore we use cylindrical coordinates (r, x).
terms depend of the physical parameters. We shall search
Without swirl let us define U0 +u and v as the longitudinal
a solution for the velocity field in the far wake which dis-
and transverse components of the time averaged velocity,
plays some characteristic features. First it is expected that
U0 the incident velocity along the x-axis, and u′ , vr′ , vθ′ the
its width h(x) is much smaller than the distance x from
components of the fluctuations.
the body and this has to be in agreement with the results
Our derivation of the evolution of the turbulent domain
derived from this assumption4 .
width consists in writing the equations (3) for the balance
Our aim is to investigate the wake characteristics in
of momentum, as is done classically, and derive the order
the frame of our closure. At this stage it is important to
of magnitude of the different terms help to our closure,
precise the role of the term σ p in eq. (7), named ”turbulent
using successively (A) and (B) methods. For an axisym-
pressure” in the following. There are two ways to interpret
metric flow, the cross-stream momentum equation is
this term,
(A) σ p can be included in the definition of the diagonal 2
∂r p v ′ 2 − vθ′
terms of the Reynolds stress tensor (as defined above) (U0 +u) ∂x v +v ∂r v = − −∂r vr′ 2 −∂x (vr′ u′ )− r
ρ r
1 (12)
u′i u′i = (σii + σ p ) (10) and the stream-wise momentum equation is
ρ
and the pressure term in Eq.(4) is restricted to the molec- ∂x p 1
(U0 + u) ∂x u + v ∂r u = − − ∂r (r vr′ u′ ) − ∂x (u′ 2 ) (13)
ular pressure p. This description corresponds to what is ρ r
experimentally measured.
(B) σ p can be added to the molecular pressure. In The incompressibility conditions for the averaged and
this case the Reynolds stress diagonal components are re- fluctuating velocities
stricted to σii defined in Eq.(6), and the pressure p in the
∂x (r u) + ∂r (r v) = 0 ∂x (ru′ ) + ∂θ vθ′ + ∂r (r vr′ ) = 0.
momenta equations becomes ptot which is the sum of the
(14)
molecular pressure coming from Navier-equations, plus the
lead to different magnitudes for the longitudinal and trans-
turbulent pressure. In summary we have
verse components. Because u changes as a function of
1 position with a typical length scale of order x in the x-
u′i u′i = σii p → ptot = p + σ p (11) direction and h(x) ≪ x in the perpendicular direction,
ρ
one has
Although not explicitly written in this form, the de- h(x) h(x) ′
v∼ u vr′ ∼ vθ′ ∼ u (15)
scription (A) is the one made in classical textbooks which x x
assume that all components of the Reynolds stress are of which means that the longitudinal velocity is much bigger
same order of magnitude, with a coefficient Cu′ v′ ≈ 0.4 than the transverse ones. This condition allows to esti-
for free turbulent shear flows like wakes, mixing layer and mate the components of the Reynolds stress as defined by
jets. our integral model (6)-(7). They are all proportional to
We shall see that the solution (B) violates the Cauchy- the curl modulus of u. Without swirl there is only one
Schwarz inequalities, whereas it is just another way of component, ∇ × u(x) = (∂x v − ∂r u)e⃗θ , that gives
grouping the terms in the balance equations (3). We prove
u
4 Notice that such an assumption is not always correct. For in- |∇ × u| ∼ . (16)
stance the flow induced by a jet, namely a point source of momentum
h
in a fluid, is such that the width of the perturbed region h(x) has
Inside the integrals of eq.(6) (written in cylindrical co-
an aperture of about 25° [2], which corresponds to a width not much
smaller than x, that implies different scaling than the ones valid for ordinates) the main contribution of the strain components
a wake. are ∂x u ∼ u/x for σxx , ∂r u ∼ u/h for both σxr and σ p ,

p-3
M. Le Berre et al.

∂r v ∼ v/h for both σrr and σθθ . The volume of integra- Wake due mainly to the drag. As written above, the
tion is the turbulent volume, h2 x, the kernel is of order drag case was originally covered by Prandtl and Landau
K(x, x′ ) ∼ 1/x, then using eq.(16) we get [1] [2]. We shall get back the same result, but using the
method just described, based on scaling laws derived from
σxx ∼ σrr ∼ σθθ ∼ ργ u2 hx the balance of momentum with our integral closure. For
(17) the drag we have first to integrate the leader terms of the
σxr ∼ σ p ∼ ργ u2 (B) stream-wise equation (13) over the whole turbulent cross-
section, we get
when the turbulent pressure is added to the molecular one
(B standpoint), whereas all components are of order Z ∞
rdrU0 ∂x u = 0. (22)
σxx ∼ σrr ∼ σθθ ∼ σxr ∼ ργ u2 (A). (18) 0

when using the classical description (A). The parameter γ provided that the Reynolds stress component vr′ u′ tends
is kept in the equations, because it was found of order 0.03 to zero faster than 1/r as r → ∞. It follows that the total
in pipes, nevertheless in the case of wakes flux of the stream-wise momentum across a plane perpen-
p it is currently R∞
dicular to x-axis, which is 0 rdrU0 u(x, r) ∼ h2 U0 u, is
accepted that the turbulence amplitude u′2 is of order
constant (with respect to x) and equal to the Newton drag
of the velocity defect u in far wakes. From eq.(18), this
force Cd a2d U02 . Using the relation (20), we get
observation implies γ ∼ 1 for wakes.
(A) standpoint. When the diagonal components of u 1/3 ad
h3 ∼ x(γ Cd a2d ) ∼ Cd ( )2/3 (23)
the Reynolds tensor include the turbulent pressure, it is U0 γx
easy to understand that our derivation of the wake scal-
ings mainly follows the classical textbooks one, since both Notice that within our closure (6), the scaling law (23) can
are based upon the hypothesis that all σij have same or- be used to calculate the time averaged velocity field u in
der of magnitude. Looking at the cross-wise momentum an axisymmetric wake of a blunt body as a function of the
equation (12), and assuming that the left hand side (l.h.s.) dimensionless radial length ζ = r/(x a2d )1/3 . This method
terms are smaller than the right hand side (r.h.s.) ones which was used in our paper dealing with the turbulent
(see the condition (21) below), the leading order terms mixing layer [6] is clearly adapted here.
with respect to the small ratio h/x, are the first and sec- Wake due in part to the lift. Aa written above we
ond ones in the r.h.s., because the third and fourth terms define as ”lift” the force generated by turbulent friction,
are smaller by a factor h/x (σ p is cancelled in the differ- applied on a body of arbitrary shape, the force being per-
2 pendicular to an incoming fast flow, whereas the drag is
ence vr′ 2 − vθ′ ). Assuming that the pressure vanishes far
from the wake in the transverse plane, we get parallel to this velocity. Of course bodies symmetrical
with respect to the horizontal plane are not subject to
pmol such a lift, like a sphere for instance. However such lift
∼ −vr′ 2 ∼ γu2 . (19)
ρ forces are obvious in other cases, like the force allowing a
plane to fly by balancing the downward pull of gravity5 .
Now let us consider the stream-wise momentum equa- Whereas the derivation by Newton of his drag law (9) does
tion (13). The leading order terms are U0 ∂x u on the l.h.s. not apply as it is to the lift force, it can be said that, from
and and − 1r ∂r (r (vr′ u′ ) on the r.h.s., the pressure term be- simple scaling arguments, this lift force is proportional to
ing of higher order. Using eq.(17) and matching the two the velocity square times a factor with the same structure
leader terms, we get the relation as Newton’s drag. However extending this equation to the
h general situation is not a trivial matter. For instance the
γu ∼ U0 (20) reversal of velocity U does more than change the sign of
x
the force, drag or lift. This is particularly obvious for the
which is valid for both drag and lift. Let us return to lift since a plane flying backward will likely not be subject
the assumption made above, that the l.h.s. terms of eq. to a downward lift exactly equal to the upward lift ob-
(12) are smaller than those of the r.h.s. This is true if served when it flies in the direction it has been designed
U0 ∂x v ≪ ∂r vr′ 2 is fulfilled, or using (18) and (20), for. Changing the sign of the velocity does more than
change the sign of the drag and lift force, it changes also
h their magnitudes if the body has no symmetry between
x≫ (21)
γ its front and back part. This does not affect however the
which is achieved for wakes, since γ is of order unity (see 5 In the case of self-propelled bodies, the Newton condition as

the discussion below eq.(18)) and x ≫ h was assumed. invoked above is not applicable to calculate the drag which requires
to consider the contribution of the propulsor to the integration of
At this step the analysis bifurcates depending if one
the stream-wise momentum. However our predictions for the lift are
deals with the case of the wake generated by the drag or meaningful when the propulsor does not contribute to the cross-wise
by the lift. momentum.

p-4
 Title

coming development that only assumes that the body is a in cylindrical coordinates. Notice however that all dimen-
source of momentum in a direction perpendicular to x. sionless parameter have been defined, so that the solution
The wake due to the lift is not axi-symmetrical, how- of this problem is universal in some sense.
ever we shall use eqs.(12) and (13), written in cylindri- Let us now compare the Reynolds number in the drag
cal coordinates but dropping the polar angle θ, that is and lift cases. Because the largest component of the per-
justified because the contribution of θ would not change turbed velocity is u, this is the one pertinent for estimating
the order of magnitude of the scalings. In this frame the the Reynolds number in the wake. It is given in general
cross-stream momentum is equal to the integral of eq.(12) as Re = uhν where ν is the kinematic viscosity of the fluid.
(integrated on x and on the transverse plane). The lead- In the case of the drag, using the estimate in eq.(23) one
ing order terms of eq.(12), ∂r (pmol + vr′ 2 ), do not con- gets
1/3
U0 ad ad Cd2

tribute to the cross-wise momentum since they are linked
Red = (27)
by the relation pmol + vr′ 2 = 0, see eq.(19). The lift is ν γx
then given by the integration of the next order terms which slowly decays at large distance from the body. No-
2
v ′ 2 −v ′
(U0 ∂x v + ∂x (vr′ u′ ) + r r θ ). Using eqs.(18) and (20) tice that a transition obviously exists between the turbu-
we deduce that all these terms are of order U02 h2 /(γx3 ). lent drag and the laminar regime further. The transitory
Integrating them over x and the transverse plane, we get domain should display a regime of localized directed per-
the relation FL = h4 U02 /(γx2 ), then using Newton’s ex- colation [10] with specific scaling laws which should be
pression (9), we get interesting to observe and explain in details.
In the lift case, inserting eq.(24) into the expression for
u Cℓ aℓ the Reynolds number one finds
h2 ∼ (xaℓ )(γ Cℓ )1/2 ∼ ( )1/4 ( )1/2 (24)
U0 γ γx
U0 aℓ Cℓ 1/2
Reℓ = ( ) (28)
2
where (aℓ ) is of order of magnitude of the cross section ν γ
area of the body times a dimensionless geometrical pa-
which doesn’t decrease at large distance, contrary to the
rameter characterizing the asymmetry of the blunt body
case of the drag.
(against the horizontal plane). Recall that the scalings
It is of interest to give a qualitative idea of the flow
have been derived under the condition (21), which be-
structure inside the far wake when it is due to lift. As is
comes for drag and lift respectively
well known large (and unwarranted) vortices are present
p p in the wake of aircrafts at take off [11]. From long ob-
x/ad ≫ Cd /γ x/aℓ ≫ Cℓ /γ 3 (25)
servations those (trailing) vortices are made generally of
which gives the distance x/ad,ℓ from the body where our pair of counter-rotating vortices which are quasi-parallel
scalings are valid. It is few tens if γ ∼ 1. to direction of motion of the plane. They are the more
stronger when the aircraft is the heavier. The flow struc-
Let us discuss the main result of our paper which pre-
ture is compatible with the expected symmetries of the
dicts that the order of magnitude of the width of the wake
wake with respect to the vertical plane containing the lift
due to lift is different from the one of the drag, eqs.(23)-
force.
(24). If one assumes, as reasonable to do, that ad and
We expect that this flow structure should agree with the
aℓ are of similar order of magnitude, the width of the lift
explicit solution of our closure model calculated with the
is bigger than the drag one, both becoming (in principle)
variable η introduced in eq. (26). In a rigorous theory, the
large at large x. However the difference could be hard to
pair of counter-rotating vortices (which yields a downward
measure because of the small difference between the two
flow) formed below the wings of an aircraft AND the wake
exponents 1/3 and 1/2 and the constraint to investigate
formed further behind, should result both from a correct
enough longitudinal distance with large Reynolds number
analytical description of the σxz component of the stress.
and significant mean intensity defect u.
In other words σxz should determine the geometry of those
Concerning a fully explicit solution of the far wake in
trailing vortices and the direction of their rotation”.
the case where the lift is bigger than the drag, one can in
Another point which deserves to be put forward con-
principle transform the integral equation (6) into an inte-
cerns an apparent paradox related to the above results.
gral equation for a velocity field of the wake depending on
When one looks at a plane flying overhead, the width of
the dimensionless radius (in cylindrical coordinates with
its wake seems to get a constant value downstream (far be-
reference axis along x)
hind the aircraft) although one expects it grows as h(x).
r This could be explained as follows. The visible part of the
η= (26) wake is not the whole part of the turbulent wake which has
(xaℓ )1/2
a width h(x) deduced from the balance of the forces and
and on the polar angle θ in the plane perpendicular to x. could be measured, but it is the part of the turbulent wake
This makes likely a quite hard problem, because, on top where ice crystals are radially convected by the transverse
of that, one has to impose the incompressibility condition velocity v. One can estimate the asymptotic width of this

p-5
M. Le Berre et al.

cloud-like condensation trail formed in the sky in the two Returning to eq. (31), and looking at the last two equiv-
cases of drag and lift wakes. alent terms, we get eq. (20) as expected.
The radial displacement of the ice crystals at time t after For the drag we have to integrate eq.(13) (over the trans-
the plane crossing is δr = vt where x = U0 t and v = hu/x. verse plane and over x). In the stream equation the two
It gives leading order terms are U0 ∂x u on the left and 1r ∂r (r vr′ u′ )
hu on the right (as found within (A) standpoint), because
δr(x) = (29)
U0 v∂r u is smaller by a factor u/U0 and ∂x ptot /ρ as well as
which is proportional to the Reynolds number, hu/U0 = ∂x (u′ 2 ) are smaller by a factor (h/x)2 . It follows that the
Re(ν/U0 ), therefore it should decrease for the drag, and drag obeys the scaling (23), as found within (A) stand-
be constant for the lift. More precisely we get point. Now for the lift, we have to integrate (on x and the
Cℓ 1/2 transverse plane) most of the terms of eq.(12) which are
δr(x) ∼ aℓ (
) (30) of leading order U02 h2 /x3 . We recover the relation (24).
γ
1/3 In summary we find that both standpoints lead to the
a C2

for the lift and δr ∼ ad dγx d for the drag. Note same characteristics of the far-wake behind a 3D blunt
body, as expected. Furthermore we have shown that the
that the latter relation (drag case) might be questionable
turbulent pressure ( giving the scale of the diagonal terms
for a self propelled body because the total stream-wise
of the Reynolds stress) has same order of magnitude as
momentum is zero, that could affect the drag scalings but
the molecular pressure, but is opposite to it.
not the lift scalings. However one may conjecture that the
product hu for the drag is also a decreasing function of x 2D turbulent wake. – Let us finally consider the
for self propelled bodies. In summary the lift explains the case of a blunt body whose transverse dimensions are very
constant width of the cloud-like condensation trail formed different, say ay ≪ az . Assuming that the body size along
behind a plane. z is quasi-infinite, the average velocity field has two com-
(B) standpoint. It is of interest to investigate the or- ponents u(x, y) and v(x, y) depending on x, y only. In this
der of magnitude of the terms in eqs.(12)-(13) within the case the cross- stream momentum equation is
(B) standpoint, because it gives a supplementary result
∂y p
concerning the order of magnitude of the molecular pres- (U0 + u) ∂x v + v ∂y v = − − ∂y (v ′ 2 ) − ∂x (u′ v ′ ), (33)
sure as compared to the turbulent pressure. In this frame ρ
the term σ p is added to the usual (molecular) pressure. In the stream-wise momentum equation is
other words the quantity p in the momentum equations
(12) and (13) is now the total pressure (11), whereas all ∂x p
(U0 + u) ∂x u + v ∂y u = − − ∂x (u′ 2 ) − ∂y (u′ v ′ ) (34)
components of the Reynolds stress are given by eq.(6), ρ
without any additive term for i = j. Equation (17) shows
that the diagonal components are now smaller than the and the mass conservation constraint is
off-diagonal ones, in other words the Cauchy-Schwarz con-
straint is (apparently) violated, unlike (A) standpoint. ∂x u + ∂y v = 0 ∂x u′ + ∂y v ′ + ∂z w′ = 0 (35)
Consider first the cross-wise equation (12) within the
where the components of the velocity fluctuations are
(B) frame. First we note that the 2nd, third and fourth
u′ , v ′ , w′ along x, y, z. Let us consider the (A) standpoint.
terms in the r.h.s. are all of order γu2 /x. It follows that
We shall show that Reynolds stress components scale as
the total pressure, if balanced by them, must be of the
same order, namely smaller than the molecular pressure σxx ∼ σyy ∼ σxy ∼ p ∼ ργ u2 (36)
(19) by a factor h/x, see below eq.(32). This is really plau-
sible because the molecular pressure is negative, whereas Consider for instance the component σxy . In the integral
the turbulent one is positive, then the sum may be one or- expressions (6) the component z plays no role, therefore
der smaller. The leader order term of the l.h.s. is U0 ∂x v ∼ the turbulent volume around a given point (x, y) is now of
U0 hu/x2 . This term is a good candidate to balance the order hx, the 2D kernel
r.h.s. ones. Indeedusing successively the two conditions
|x − x′ |
2

∂r ptot ′ 2 ′ ′ vr′ 2 −vθ′ ′
U0 ∂x v ∼ ρ ∼ ∂r vr + ∂x (vr u ) + r , we get K(x, x ) = ln (37)
|x′ |
the two relations
1 tot h h is of order ln(1 + h/x) ∼ h/x, the curl is still u/h, as
p ∼ ( )2 U0 u ∼ γu2 (31) well as the strain component Sxy ∼ u/h,. It follows that
ρ x x
σxy ∼ γu2 . Similarly same scalings are found for the diag-
The left part of these three equivalent terms leads to the onal components σxx and σyy . Moreover the scaling of the
interesting result that ptot is smaller than pmol by a factor pressure (p = pmol ) is deduced from the cross-stream equa-
h/x, see eq.(19), although at leading order one has tion (33) by balancing the leading order terms, ∂y pmol /ρ
pmol ∼ −σ p ≫ ptot (32) and −∂y (v ′ 2 ).

p-6
 Title

The balance of the two leading order terms of the as for the axisymmetric case of a pure drag to derive the
stream-wise equation, gives U0 u/x ∼ ∂y (u′ v ′ ), that gives scaling laws, because we deal with orders of magnitude.
h
u ∼ U0 γx as in the 3D case, see eq.(20). Now let us derive We also derive the scalings in the 2D case.
the drag scaling as above, by integrating the leading order We emphasize that a computational study based on our
terms of the stream momentum equation over the vari- closure model (with a kernel adapted to each case), should
ables y and x. That amounts to integrate U0 u/x which be able to treat rigorously any case. We hope that the
gives hU0 u ∼ Cd ad U02 . Then using (20), we get present derivation of the scaling laws will encounter the
 1/2 same sort of recognition as the classical result proposed
u C d ad almost a century ago by Prandtl in the case of boundary
h2 ∼ γCd ad x ∼ (38)
U0 γx layers created by the drag force.
We emphasize that our results are obtained bypassing
and the product hu is constant in the far wake. We notice the difficulty of defining ”in general” a length scale (the
that the law h ∼ x1/2 for the drag agrees with the works ”Mischungsweg” or mixing length in Prandtl’s own words)
of [7], [12], [13]. related to the distance to the solid boundaries, supposed
Then, let us derive the lift scaling from the cross-stream much less than the length along the boundary.
momentum equation. Consider first the integration over
the radial variable y.The term ∂y (p/ρ + v ′ 2 ) does not con- ∗∗∗
tribute because the turbulence and the pressure vanish at
infinity, the remaining terms provide We thank very much Christophe Josserand, Thierry
Z +∞  Lehner, Luc Pastur and Sergio Rica for their interest in
∂
x dy(U v + u′ v ′ ) = 0
0 (39) this work and stimulating discussions.
−∞

Integration over x leads to a constant which is the lift force REFERENCES

(perpendicular to the x axis) per unit depth
[1] Prandtl L., Proceedings of the Second International
Z +∞ Congress for Applied Mechanics, edited by Emmanuel E.
dy(U0 v + u′ v ′ ) ∼ Fℓ,2D (40) Gdutos, Vol. 8 (Springer, Zurich) 1926, p. 62
−∞ [2] Landau L. and Lifshitz E.M., Course of theoretical
physics, Fluid Mechanics (Pergamon Press,Oxford) 1987,
where
sect. 37
Fℓ,2D = Cℓ aℓ U02 . [3] Schlichting H. and Gersten K., , Boundary Layer the-
Because the two terms in the integrand are of same or- ory, (Springer, Berlin) 1999
der, one may consider either. The first one is of order [4] Nedić J., Vassilicos J.C. and Ganapathisubramani B.
U0 (hu/x)h. Balancing with the lift gives (u/U0 )h2 /x ∼ , Axisymmetric turbulent wakes with new non-equilibrium
similarity scalings,.Phys.Rev.Let., 11 (2013) 144503
aℓ Cℓ , and using (20) we get
[5] Obligado M., Dairay T., and Vassilicos J. C.,
 1/3 Nonequilibrium scalings of turbulent wakes. Phys. Rev.
3 2 u Cℓ aℓ Fluids, 1 (2016) 044409
h ∼ γCℓ aℓ x ∼ (41)
U0 γ2x [6] Pomeau Y. and Le Berre M., Turbulence in a wedge:
The case of the mixing layer, Phys. Rev. Fluids, 6 (2021)
In conclusion we find that the width of the wake due to the 074603
lift increases more rapidly than the one due to the drag, [7] Wygnanski W., Champagne F. and Marasli B., On
both in 2D and 3D flows. Note that one may associate the large-scale structures in two-dimensional, small-deficit,
Cℓ aℓ to the y-dimension of a fixed airfoil (of infinite span) turbulent wakes. J. Fluid Mech., 168 (1986) 31-71
by using the Kutta-Joukovski theorem.HIt states that the [8] Gunasekaran S. and Altman A., Self-Preserved Turbu-
lift per unit span is ρU0 Γ, where Γ = u · ds is the cir- lent Wake and Its Relation to Aerodynamic Efficiency En-
culation around a close contour enclosing the airfoil. The ergies American Institute of Aeronautics and Astronautics,
circulation being of order ay U0 , we get Cℓ aℓ ∼ ay . 14 (2021) 3641 MDPI Energies, 14 (2021) 3641
We find that the Reynolds number is constant for the [9] Pomeau Y. and Le Berre M. , Turbulent plane Poiseuille
flow Eur.Phys. Journ. Plus, , (136) 2021 11
drag, a classical result for 2D flows, we have Re2D,drag =
[10] Pomeau Y., Front motion, metastability and subcritical
Cd U0 ad /ν, although Re slowly increases with x for the lift, bifurcations in hydrodynamics, Physica D, 23 (1986) 3-11
as x1/3 . [11] see for instance the aerospace web site
Conclusion. – Thanks to our explicit closure for- http://www.aerospaceweb.org/question/nature/q0237.shtml
mula, we have first revisited the original Prandtl’s result [12] Sreenivasan, K., Approach to Self-Preservation in Plane
Turbulent Wakes, AIAA J., 19 (1981) 1365–1367
on the asymptotic wake aperture, and secondly extended
[13] Sreenivasan, K. and Narasimha R., Equilibrium Pa-
it to the case where the wake is due in part to the lift force rameters for Two-Dimensional Turbulent Wakes, ASME J.
on the moving body. In the latter case the wake could be Fluids Eng., 104 (1982) 167–169
asymmetric, nevertheless we use the same basic equations

p-7