40 (2004), 12671290
A Survey on a Class of Exact Solutions of
the Navier-Stokes Equations and a Model
for Turbulence
By
Koji Ohkitani
Abstract
In this paper we review some classes of exact solutions of the Navier-Stokes
equations under a time-independent external straining ow, centering around the
celebrated Burgers vortex. The objectives are (i) to clarify the relationship between
them and (ii) to examine them as models of turbulence. Particularly we study the
Lundgren spiral model for turbulence in the presence of azimuthal vorticity (that is,
with axial velocity). The implication on linear stability of the Burgers vortex is briey
discussed.
1. Introduction
At present there is no known method of integrating the Navier-Stokes equa-
tions in a systematic fashion. In this sense one should not be surprised at the
paucity of exact solutions known to date. There are a number of attempts to
collect and tabulate the exact solutions, which are scattered rather sporadically
over a range of the literature [B63, W89, W91, O97]. Some of them are dis-
covered independently by dierent researchers and some of them belong to the
folklore. The purpose of this article is two-fold. First, we review a class of exact
solutions centering around the celebrated exact solution of the Burgers vortex
Communicated by H. Okamoto. Received January 19, 2004.
2000 Mathematics Subject Classication(s): 76D05, 76F02
Key words: uid dynamics, Navier-Stokes equations, exact solutions, turbulence model
Research Institute for Mathematical Sciences, Kyoto University, Kyoto 606-8502, Japan.
e-mail: ohkitani@kurims.kyoto-u.ac.jp
c 2004 Research Institute for Mathematical Sciences, Kyoto University. All rights reserved.
(k) associated with axial velocity in the spirit of the original paper [L82]
in Section 5,
1
where we consider the Lundgrens model using the general su-
perposed class of solutions. Finally, Section 6 will be devoted to summary and
discussion.
2. Some Classes of Stretched Solutions
2.1. Burgers vortex
The celebrated Burgers vortex tube [B48] is an exact solution under an
external time-independent strain eld. With the external eld the solution
dose not belong in L
2
, but the physical meaning of it is that it mimics the
eect of other vortices residing at far distances. The Burgers vortex solution
was independently found by Rott [R58]. There, some of its properties such as
pressure distribution were studied in detail.
Originally, the Burgers solution was given as a stationary solution of the
Navier-Stokes equations. In [R58] a non-stationary form is described. More
generally it can be realized as a large-time asymptotic state of non-stationary
solutions [K84, M86]. Consider a ow under a constant strain rate (> 0),
whose velocity and vorticity are respectively given by
u = (x +u
1
(x, y, t), y +u
2
(x, y, t), 2z) (2.1)
and
= (0, 0,
3
),
3
=
x
u
2
y
u
1
. (2.2)
The vorticity equation reads
3
t
+u
1
3
x
+u
2
3
y
= x
3
x
+y
3
y
3
+
2
3
, (2.3)
where
2
2
x
+
2
y
. If the ow is axisymmetric, we have
u
1
3
x
+u
2
3
y
=
(
3
, )
(x, y)
= 0,
where is the stream function. Introducing the transformations
3
=
3
e
t
and X = xe
t
, Y = ye
t
, T = (e
2t
1)/(2), the vorticity equations reduces
to a linear diusion equation
3
T
=
_
3
X
2
+
2
3
Y
2
_
. (2.4)
1
It should be emphasized that the same result has been obtained by [PL01] by a more
sophisticated treatment following [PS93, PBS94].
3
(X, Y, T) =
1
4T
__
3
(X
, Y
, 0) (2.5)
exp
_
(X X
)
2
+ (Y Y
)
2
4T
_
dX
dY
.
In the original variables we nd
3
(r, t) =
2
e
2t
e
2t
1
__
3
(X
, Y
, 0) (2.6)
exp
_
2
(e
t
x X
)
2
+ (e
t
y Y
)
2
e
2t
1
_
dX
dY
,
so we nd by dominated convergence that
3
(r, t)
2
exp
_
r
2
2
_
as t .
This is the Burgers vortex tube, obtained as a balance between stretching by
the external strain eld and viscous diusion.
2.2. A class of solutions studied by Takaoka
This class is complementary to the Burgers vortex, in that vorticity has two
non-zero components and velocity one non-zero component. Axisymmetric case
can be obtained by applying Lundgrens transformation to a class of quasi 2D
solutions (see below) and is only marginally novel. In [T91] this class has been
used as a model for vortex reconnection, see also, [T90] for the two-dimensional
case.
The velocity and vorticity are
u = (x, y, z +u
3
(x, y, t)) (2.7)
and
= (
y
u
3
,
x
u
3
, 0), (2.8)
where , > 0, + + = 0.
The vorticity equations read
1
t
= x
1
x
+y
1
y
1
+
2
1
, (2.9)
2
t
= x
2
x
+y
2
y
2
+
2
2
, (2.10)
x
u
3
+
2
y
u
3
= 0.
By transforming dependent and independent variables as
2
1
=
1
e
t
,
2
=
2
e
t
,
U
3
= u
3
e
t
,
X = xe
t
, Y = ye
t
, T = t
we nd
1
T
=
e
2T
2
X
2
+e
2T
2
Y
2
1
. (2.11)
Noting that
1
=
U
3
Y
we verify that the axial velocity obeys the same equation
U
3
T
=
e
2T
2
X
2
+e
2T
2
Y
2
U
3
. (2.12)
This can be solved as
U
3
(X, Y, T) =
(e
2T
1)(e
2T
1)
(2.13)
u
3
(X
, Y
, 0) exp
(X X
)
2
2(e
2T
1)
(Y Y
)
2
2(e
2T
1)
dX
dY
,
that is,
u
3
(x, y, t) =
e
t
(e
2t
1)(e
2t
1)
(2.14)
u
3
(X
, Y
, 0) exp
(xe
t
X
)
2
2(e
2t
1)
(ye
t
Y
)
2
2(e
2t
1)
dX
dY
.
Unlike the Burgers vortex, it does not converge to a nontrivial stationary solu-
tion in the limit t . Rather it decays to zero as
u
3
(x, y, t)
u
3
(x
, y
, 0)dx
dy
exp
x
2
+y
2
2
exp(2t),
where = ( +) < 0.
2
In [T91] a slightly dierent scaling was used for U
3
.
y
u
1
). (3.2)
Recall that
u
1
=
y
, u
2
=
x
,
1
=
y
u
3
,
2
=
x
u
3
,
3
=
x
u
2
y
u
1
.
The vorticity equations read
1
t
=
1
+
_
x
1
x
+y
1
y
_
_
u
1
1
x
+u
2
1
y
_
(3.3)
+
_
1
u
1
x
+
2
u
1
y
_
+
2
1
,
2
t
=
2
+
_
x
2
x
+y
2
y
_
_
u
1
2
x
+u
2
2
y
_
(3.4)
+
_
1
u
2
x
+
2
u
2
y
_
+
2
2
,
3
t
=
3
+
_
x
3
x
+y
3
y
_
_
u
1
3
x
+u
2
3
y
_
(3.5)
+
_
1
u
3
x
+
2
u
3
y
_
+
2
3
.
We have, as in the Takaoka vortex,
u
3
=
1
x
u
3
+
2
y
u
3
= 0,
3
x
+u
2
3
y
(
3
, )
(x, y)
= 0.
The same equation results as the one for the Burgers vortex for axial vorticity
3
t
=
3
+
3
x
+y
3
y
+
2
3
.
It should be noted that u
3
,
1
,
2
are functions of x, y, t but u
1
, u
2
,
3
are
functions of r, t only.
We can make use of the transformation of variables (writing for the gen-
eral case = to show how the time-dependent coecients appear in the
equations)
X = xe
t
, Y = ye
t
,
U
1
= e
t
u
1
, U
2
= e
t
u
2
, U
3
= e
t
u
3
,
1
= e
t
1
,
2
= e
t
2
,
3
= e
t
3
.
Dening
1
=
U
3
Y
,
2
=
U
3
X
,
3
=
U
2
X
U
1
Y
,
we may write the rst two equations as
1
t
=
e
2t
U
1
1
X
+e
2t
U
2
1
Y
+e
2t
1
U
1
X
+
2
U
1
Y
(3.6)
+
e
2t
1
X
2
+e
2t
1
Y
2
2
t
=
e
2t
U
1
2
X
+e
2t
U
2
2
Y
+e
2t
1
U
2
X
+
2
U
2
Y
(3.7)
+
e
2t
2
X
2
+e
2t
2
Y
2
.
For the symmetric case = , by
T =
exp(2t) 1
2
3
Without axisymmetry the equation for
3
is decoupled, but is not linearized.
1
T
=
_
U
1
1
X
+U
2
1
Y
_
+
_
1
U
1
X
+
2
U
1
Y
_
+
_
1
X
2
+
2
1
Y
2
_
,
(3.8)
2
T
=
_
U
1
2
X
+U
2
2
Y
_
+
_
1
U
2
X
+
2
U
2
Y
_
+
_
2
X
2
+
2
2
Y
2
_
.
(3.9)
Or, using
= , U
1
=
Y
, U
2
=
X
,
we can write alternatively
1
T
=
(,
1
)
(X, Y )
+
(U
1
, U
3
)
(X, Y )
+
_
1
X
2
+
2
1
Y
2
_
,
2
T
=
(,
2
)
(X, Y )
+
(U
2
, U
3
)
(X, Y )
+
_
2
X
2
+
2
2
Y
2
_
.
together with
3
T
=
_
3
X
2
+
2
3
Y
2
_
.
The axial vorticity
3
is obtained by solving a diusion equation. Then
U
1
, U
2
(or ) is known explicitly and we may solve the linear equations for
1
and
2
.
3.1. An alternative interpretation of the superposed solutions
A method is known for constructing a class of quasi two-dimensional
(/z = 0) Navier-Stokes ows, see, e.g. [M85]. It has axisymmetric azimuthal
velocity and axial velocity.
Lemma 3.1. Consider a ow in cylindrical coordinates (r, , z),
u
r
= 0, u
= u
(r, t), u
z
= u
z
(r, , t).
The Navier-Stokes equations can be written as
u
t
=
_
2
u
r
2
+
1
r
u
r
u
r
2
_
, (3.10)
r
u
z
=
_
2
u
z
r
2
+
1
r
u
z
r
+
1
r
2
2
u
z
2
_
, (3.11)
and
u
2
r
=
p
r
. (3.12)
Proof. This is a direct consequence of the axisymmetric Navier-Stokes
equations.
Remark. The components of vorticity are
r
=
1
r
u
z
=
u
z
r
,
z
=
1
r
(ru
)
r
.
Remark. When = 0, we may solve them easily as
u
r
= 0, u
= u
(r), u
z
= u
z
_
r, t
u
(r)
r
_
.
As time elapses, u
z
has an increasingly oscillating structure. It is well known
that as t such a solution does not converge to anything in the normal
sense, but it displays weak convergence. In this case as t , u
z
converges
in a weak sense to the -average of its initial value, see e.g. [Y00, BG98]. If
> 0, u
z
eventually decays to zero by viscous diusion.
Proposition 3.2. For the axisymmetric case = , the superposed
class of solutions mentioned above actually belongs to this category.
Proof. To see this we write the equations for velocity
u
1
t
= u
1
+
_
x
u
1
x
+y
u
1
y
_
_
u
1
u
1
x
+u
2
u
1
y
_
p
2
x
+
2
u
1
,
(3.13)
u
2
t
= u
2
+
_
x
u
2
x
+y
u
2
y
_
_
u
1
u
2
x
+u
2
u
2
y
_
p
2
y
+
2
u
2
,
(3.14)
u
3
t
= u
3
+
_
x
u
3
x
+y
u
3
y
_
_
u
1
u
3
x
+u
2
u
3
y
_
+
2
u
3
, (3.15)
2
x
2
+
2
y
2
+
2
z
2
_
+p
2
(x, y, t).
Under the set of transformations, the last equation (3.15) becomes
U
3
T
+U
1
U
3
X
+U
2
U
3
Y
=
_
2
U
3
X
2
+
2
U
3
Y
2
_
,
or, in cylindrical coordinates
U
3
T
+
U
R
U
3
=
_
2
U
3
R
2
+
1
R
U
3
R
+
1
R
2
2
U
3
2
_
,
which is the same as (3.11).
Note that U
3
behaves like a passive scalar because of the decoupling of
axial and planar components of velocity. We also note that
R
=
1
R
U
3
=
U
3
R
.
4. Lundgrens Spiral Model for Turbulence
In a seminal paper Lundgren considered a spiral model for small-scale
turbulent motion on the basis of the Burgers vortex. Lundgrens asymptotic
solution of the 2D Navier-Stokes equations takes the following form
(r, , t) =
n=
n
(r, t) exp(in), (4.1)
n
= e
2t
f
n
(R) exp(in (R)T n
2
(R)T
3
/3), (n = 0) (4.2)
0
(r, t) =e
2t
_
f
0
(R, T) +g
(R, T)
_
, (4.3)
R(r, t) =re
t
, T = (e
2t
1)/(2), (4.4)
where
1
R
d
dR
_
R
2
(R)
_
=f
0
(R) +g
(R). (4.5)
The expression (4.5) relates vorticity and circulation in cylindrical coordinates
=
1
r
r
(ru
(r)).
0
(R, T) describes -averaged vorticity of the spiral component and
g
n
() = f
() (independent of
n) we have
(r, , t) = 2e
2t
f
(R)( (R)T) + e
2t
g
(R),
where is the Dirac delta function. Therefore the underlying mechanism in this
case is rolling-up of a nearly circular vortex layer due to Kelvin-Helmholtz in-
stability by dierential rotation . For general f
n
, such a simple interpretation
is not available.
Using this asymptotic solution, Lundgren computed the averaged energy
spectrum E(k) as
E(k) = E
0
(k) + E
(k),
where E
0
(k) k
1
is the contribution from the vortex tube,
4
whose unaveraged
form is given by
E
0
(k) =
N
2
4k
exp
k
2
. (4.6)
The spectrum E
(k) = Ak
5/3
exp
k
2
3
, (4.7)
A =
4
3
N(2)
1/3
n=1
n
4/3
0
|f
n
()|
2
d
(
())
4/3
(4.8)
Under the circumstances where the prefactor of the core spectrum is smaller
than the one of the spiral, we nd E(k) k
5/3
. We observe that this result
is independent from the choice of initial conditions, because f
n
() is included
in A.
This model consists of three parts, each of them are of importance in their
own right in applied analysis of the Navier-Stokes equations. The rst two
are required for the approximate solutions and the nal for estimation of the
spectrum.
4
Care should be taken in averaging E
0
(k), as a cut-o is needed for time integration
[PS93, PBS94].
t
r
r
+
1
r
(, )
(r, )
= 2 +, (4.9)
where
u
r
=
1
r
, u
r
. (4.10)
We introduce transforms for independent variables
R = e
t
r, = , T =
e
2t
1
2
and those for dependent variables
(R, , T) = e
2t
(r, , t),
(R, , T) = (r, , t).
Then we can reduce the dynamical equations to
T
+
1
R
(, )
(R, )
=
_
R
2
+
1
R
R
+
1
R
2
2
_
, (4.11)
which is a 2D ow without strain elds. Note that
U
R
=
1
R
, U
R
. (4.12)
2. Asymptotic axisymmetrization of advecting velocity
This means that at large times (t ), dynamics of vorticity is well
approximated by
T
+
U
0
(R)
R
= 0, (4.13)
where U
0
is the axisymmetric azimuthal velocity induced by vorticity
0
.
For simplicity, viscosity has been discarded.
It means that as the vorticity winds up with excitation on small scales,
its non-axisymmetric perturbations become rapidly varying in radial di-
rection, which makes vorticity-stream function coupling cancel at leading
order [BL94, BG98]. Thus, as T the vorticity behaves as if it were a
passive scaler advected by a ow induced by
0
. This large time asymp-
totic analysis has been recast using a kind of WKB analysis [BL94] under a
linear approximation. While a rigorous mathematical justication for the
fully-nonlinear case seems to be lacking, validity of the asymptotic theory
has been obtained by numerical simulations, see e.g. [L93, BG98]. See also
related works [LB95, BG99, MBG02, HBG03].
3. Statistical average of the energy spectrum
It is assumed that there are a large number of vortices in the box. In that
box each vortex has the identical structure but is in dierent stages of time
evolution. These vortices are assumed to be supplied at some constant
rate N, the rate of creation of vortex length per unit time per unit volume
[L
2
T
1
] to take into account the increasing length of the structures by
vortex stretching. This may be viewed as a kind of ergodic hypothesis.
The Lundgrens model assumes that azimuthal vorticity is absent. If it is
present, it might inuence the shape of the spectrum E(k).
A phenomenological theory based on the cascade argument (essentially the
so-called -model for turbulence) is given by Gilbert in [G93]. According to it,
the spectrum due to axial vorticity leads to
E
(k) k
5/3
,
consistent with Lundgrens model. If we apply the same phenomenology to
azimuthal component of vorticity, it is readily veried that
(k) k
7/3
,
whose slope is steeper than 5/3.
It may be in order to briey describe the derivation here. Consider a blob
of vortex in an axisymmetric eld u = (0, u
0
t,
r
0
,
z
0
, where r
r
(u
/r) is the
dierential rotation. If we turn on the straining led u = (r, 0, 2z), the
vorticity is stretched as
r
(r, , z, t) = e
t
R
(R, , Z, T),
(r, , z, t) = e
t
(R, , Z, T)
and
z
(r, , z, t) = e
2t
R
(Z, , Z, T),
where
R = e
t
r, = , Z = e
2t
z, T =
e
2t
1
2
.
Thus, axial and azimuthal components of vorticity are transformed as
z
0
0
e
2t
and
0
t
0
e
t
e
2t
1
2
0
e
t
.
On the other hand, the height, length and thickness of the blob undergo re-
spectively transformations as
l
0
l
0
e
2t
, l
0
t
l
0
e
t
,
l
0
t
l
0
e
3t
( l).
Now, considering the eddy turn-over time O(1/) and reading the smallest scale
by l(= 1/k), the enstrophy spectra Q
(k) and Q
(k)
l
3
0
(
0
e
2t
)
2
l
4/3
k
4/3
and
kQ
(k)
l
3
0
e
t
_
2
l
2/3
k
2/3
.
These are Gilberts phenomenological predictions.
= u
(r, t), u
z
( u
3
) = 2z +u
z
(r, , t), ( > 0), (5.1)
r
=
1
r
u
z
=
u
z
r
,
z
=
1
r
(ru
)
r
. (5.2)
At large time t , an asymptotic solution is given by
u
3,n
S
1
h
n
(S
1/2
r) exp
_
in(S
1/2
r)P
n
2
3
(S
1/2
r)
2
P
3
_
,
,n
f
n
(S
1/2
r)S
1/2
r
(S
1/2
r)P exp
_
in(S
1/2
r)P
n
2
3
(S
1/2
r)
2
P
3
_
,
and
r,n
S
1/2
f
n
(S
1/2
r) exp
_
in(S
1/2
r)P
n
2
3
(S
1/2
r)
2
P
3
_
for Fourier expansions such as
u
3
(r, , t) =
n=
u
3,n
(r, t) exp(in).
Here we have introduced
(r) =
u
(r, 0)
r
, S = exp(2t), P =
exp(2t) 1
2
,
and f
n
(r) = inh
n
(r)/r with h
n
(r) is an arbitrary function of r. An asymptotic
solution of axial vorticity =
z
is still given by (4.1).
R
U
3
=
_
2
U
3
R
2
+
1
R
U
3
R
+
1
R
2
2
U
3
2
_
, (5.3)
which has the same form as the equation for axial vorticity. We dene its
Fourier coecient by
U
3
=
n=
U
3,n
(R, T)e
in
and
(R, T) =
U
(R, T)
R
.
The axial component of velocity U
3
is a passive scalar. As in [L82] we
may treat (R, T) as a constant; (R, T) = (R, 0), see also [BG98, BG99].
The validity of such an approximation is dierent from the (more dicult,
nonlinear) vorticity problem, where asymptotic axisymmetrization is the key
property. Here for a passive scalar problem, the large time asymptotic analysis
holds because dynamics is linear by denition [RW83].
To absorb rapidly varying exponentials, we introduce h
n
(R, T) =
U
3,n
exp(inT) then it satises
h
n
T
=
_
2
h
n
R
2
+
_
1
R
2in
d
dR
T
_
h
n
R
+
_
in
d
2
dR
2
T n
2
_
d
dR
_
2
T
2
in
R
d
dR
T
n
2
R
2
_
h
n
_
.
Retaining the dominant T
2
-term at large times in the viscous term we nd
h
n
T
n
2
_
d
dR
_
2
T
2
h
n
.
This can be solved to give
U
3,n
h
n
(R) exp
_
in(R)T
n
2
3
_
d
dR
_
2
T
3
_
, (5.4)
with an arbitrary function h
n
(R) of R. So we nd
U
3
n=
h
n
(R) exp
_
in(R)T
n
2
3
_
d
dR
_
2
T
3
+in
_
. (5.5)
R
=
1
R
U
3
n=
f
n
(R) exp
_
in(R)T
n
2
3
_
d
dR
_
2
T
3
+in
_
(5.6)
and
(5.7)
=
U
3
R
n=
_
g
n
(R) +Rf
n
(R)
d
dR
T +
in
3
Rf
n
(R)
d
dR
_
d
dR
_
2
T
3
_
exp
_
in(R)T
n
2
3
_
d
dR
_
2
T
3
+in
_
,
where
f
n
(R)
inh
n
(R)
R
and
g
n
(R) = h
n
(R).
For the Fourier coecients, we nd
R,n
f
n
(R) exp
_
in(R)T
n
2
3
_
d
dR
_
2
T
3
_
, (5.8)
,n
_
g
n
(R) +f
n
(R)R
d
dR
T +
3i
n
Rf
n
(R)
d
dR
_
d
dR
_
2
T
3
_
(5.9)
exp
_
in(R)T
n
2
3
_
d
dR
_
2
T
3
_
,
where
R
=
n=
R,n
(R, T)e
in
,
n=
,n
(R, T)e
in
.
Finally, after performing the Lundgrens transforms for independent and
dependent variables, we nd in terms of the original coordinates
r,n
S
1/2
f
n
(S
1/2
r) exp
_
in(S
1/2
r)P
n
2
3
(S
1/2
r)
2
P
3
_
(5.10)
,n
_
S
1/2
g
n
(S
1/2
r) +f
n
(S
1/2
r)S
1/2
r
(S
1/2
r)P (5.11)
+S
3i
n
_
()
2
_
n
()P
3
_
exp
_
in(S
1/2
r)P
n
2
3
(S
1/2
r)
2
P
3
_
,
where
S = exp(2t), P =
exp(2t) 1
2
.
As expected,
r,n
is negligible compared with
,n
. We also note that the third
term (with explicit -dependence) in the brackets in
,n
is small when the
Reynolds number is large.
Remark. As T we have U
3
0. Therefore addition of axial
velocity does not give rise to another stationary solution dierent from the
Burgers vortex.
5.2. Estimation of the energy spectrum
We will derive the energy spectrum associated with axial velocity following
the method in [L82].
Theorem 5.1. Consider the superposed class of solutions of the propo-
sition 5.1. The energy spectrum due to planar components of vorticity is given
by
E
(k) = Bk
7/3
exp
_
k
2
3
_
, (5.12)
B =
4
3
N
(2)
7/3
n=1
n
2/3
_
0
|f
n
()|
2
(
())
4/3
3
d, (5.13)
where f
n
() represents an amplitude of Fourier coecient of planar vorticity.
Proof. We will closely follow the footsteps of Lundgren to evaluate the
spectral from azimuthal component. We write
E(k, t) = E
0
(k, t) +E
(k, t) +E
(k, t) = 2k
n=1
|I
n
(k, t)|
2
, (5.15)
where
I
n
(k, t) =
0
J
n
(kr)u
3,n
(r, t)rdr (5.16)
is the Hankel transform. Using an integral representation of Bessel functions
J
n
(kr) =
1
2
2
kr
1/2
((i)
n+1/2
e
ikr
+i
n+1/2
e
ikr
),
we nd
I
n
(k, t) =
0
1
2
2
kr
1/2
((i)
n+1/2
e
ikr
+i
n+1/2
e
ikr
)
,n
(r, t)rdr.
Inserting the approximate solution obtained above
I
n
(k, t)
0
W(r
n
, t)
(i)
n+1/2
exp
ikr in(S
1/2
r)P
+i
n+1/2
exp
ikr in(S
1/2
r)P
dr,
where
W
n
(r, t)
1
2
2
kr
1/2
S
1
h
n
(S
1/2
r) exp
n
2
3
(S
1/2
r)
2
P
3
.
Assuming
(S
1/2
r) < 0,
(S
1/2
r) > 0,
we can evaluate the above integral using the method of stationary phase
5
as
I
n
(k, t) i
n1/2
e
i/4
2
n
(S
1/2
r
n
)PS
1/2
W
n
(r
n
, t) exp
ikr
n
in(S
1/2
r
n
)P
,
5
Note that
e
iax
2
dx =
a
e
i/4
, a > 0.
(S
1/2
r
n
)PS
1/2
= 0.
The azimuthal spectrum becomes
E
(k, t) = 2k
n=1
2
n
(S
1/2
r
n
)PS
(W
n
(r
n
, t))
2
= 2
n=1
(S
1/2
r
n
)PS
r
n
_
S
1
h
n
(S
1/2
r
n
)
_
2
exp
_
2
n
2
3
(S
1/2
r)
2
P
3
_
.
Inserting
n =
k
S
1/2
P
(S
1/2
r
n
)
,
we nd
E
(k, t) =
2
k
n=1
(S
1/2
r
n
)
(S
1/2
r
n
)
S
1/2
r
n
_
S
1
h
n
(S
1/2
r
n
)
_
2
exp
_
2
n
2
3
(S
1/2
r)
2
P
3
_
.
This is a deterministic form of the azimuthal spectrum.
It is assumed that there are a large number of vortices in a box of size L
3
,
then the averaged energy spectrum can be written as
E
(k, t) =
j
l
j
L
3
E
(k, t),
where l
j
denotes a segment along which each vortex. We replace the above
ensemble average with a long-time average
j
= N
c
_
0
dt
together with l
j
= S(t)l
0
. The averaged azimuthal spectrum is
E
(k, t) =
2
k
N
c
l
0
L
3
_
0
n=1
(S
1/2
r
n
)
(S
1/2
r
n
)
S
1/2
r
n
_
S
1
h
n
(S
1/2
r
n
)
_
2
exp
_
2
n
2
3
(S
1/2
r)
2
P
3
_
Sdt.
Changing variables to = S
1/2
r
n
and noting Sdt = dP and
()
()
dP =
2
3
Pd,
(k) =
4
3
N
2
k
1
exp
_
k
2
3
_
n=1
_
0
|h
n
()|
2
PS
3
d,
with
P =
_
k
n
()(2)
1/2
_
2/3
,
where we have dened N = N
c
l
0
/L
3
. Because P/S 1/(2), we obtain
E
(k) = Bk
7/3
exp
_
k
2
3
_
, (5.17)
B =
4
3
N
(2)
7/3
n=1
n
4/3
_
0
|h
n
()|
2
(
())
4/3
d. (5.18)
In terms of an amplitude f
n
() of planar vorticity components we may write
B =
4
3
N
(2)
7/3
n=1
n
2/3
_
0
|f
n
()|
2
(
())
4/3
3
d. (5.19)
6. Summary and Discussion
If we extend Gilberts phenomenological argument to planar components
of vorticity in a straightforward fashion, it predicts
E
(k) k
7/3
.
The above asymptotic solutions is consistent with it, in that they give the same
exponent. It is of interest to note that there is a subtle dierence between E
(k)
and the Lundgrens spectrum
E
(k) = Ak
5/3
exp
_
k
2
3
_
, A =
4
3
N(2)
1/3
n=1
n
4/3
_
0
|f
n
()|
2
d
(
())
4/3
.
(6.1)
While we can take f
n
() = f
() in E
n
() = f
() in
E
(k), because the summation w.r.t. n would be divergent. This suggests that
the spatial structure of planar vorticity cannot be a simple roll-up mechanism