JOURNAL OF MATHEMATICAL PHYSICS 48, 065202 共2007兲
Estimates for the two-dimensional Navier–Stokes
equations in terms of the Reynolds number
J. D. Gibbon and G. A. Pavliotisa兲
Department of Mathematics, Imperial College London, London SW7 2AZ,
United Kingdom
共Received 17 May 2006; accepted 25 August 2006; published online 4 June 2007兲
The tradition in Navier–Stokes analysis of finding estimates in terms of the Grashof
number Gr, whose character depends on the ratio of the forcing to the viscosity ,
means that it is difficult to make comparisons with other results expressed in terms
of Reynolds number Re, whose character depends on the fluid response to the
forcing. The first task of this paper is to apply the approach of Doering and Foias
关C. R. Doering and C. Foias, J. Fluid Mech. 467, 289 共2002兲兴 to the twodimensional Navier–Stokes equations on a periodic domain 关0 , L兴2 by estimating
quantities of physical relevance, particularly long-time averages 具·典, in terms of the
Reynolds number Re= U ᐉ / , where U2 = L−2具储u储22典 and ᐉ is the forcing scale. In
particular, the Constantin–Foias–Temam upper bound 关P. Constantin, C. Foias, and
R. Temam, Physica D 30, 284 共1988兲兴 on the attractor dimension converts to
a2ᐉRe共1 + ln Re兲1/3, while the estimate for the inverse Kraichnan length is 共a2ᐉRe兲1/2,
where aᐉ is the aspect ratio of the forcing. Other inverse length scales, based on
time averages, and associated with higher derivatives, are estimated in a similar
manner. The second task is to address the issue of intermittency: it is shown how
the time axis is broken up into very short intervals on which various quantities have
lower bounds, larger than long time averages, which are themselves interspersed by
longer, more quiescent, intervals of time. © 2007 American Institute of
Physics. 关DOI: 10.1063/1.2356912兴
I. INTRODUCTION
A. General introduction
In the last two decades the notion of global attractors in parabolic partial differential equations
has become a well-established concept.1–4 The general nature of the dynamics on the attractor A,
in a time averaged sense, can roughly be captured by identifying sharp estimates of the Lyapunov
共or fractal or Hausdorff兲 dimension of A, or the number of determining modes,5 with the number
of degrees of freedom. Introduced by Landau,6 this latter idea says that in a dynamical system of
spatial dimension d of scale L, the number of degrees of freedom N is roughly defined to be that
number of smallest eddies or features of scale and volume d that fit into the system volume Ld,
N⬃
冉冊
L
d
.
共1.1兲
This is the origin of the much-quoted N ⬃ Re9/4 result associated with the three-dimensional
Navier–Stokes equations which rests on taking ⬃ k ⬃ L Re−3/4, where k is the Kolmogorov
length scale. In the absence of a proof of existence and uniqueness of solutions of the threedimensional Navier–Stokes equations, at best this is no more than a rule of thumb result. It rests
on a more solid and rigorous foundation, however, for the closely related three-dimensional
a兲
Electronic mail: g.pavliotis@imperial.ac.uk
0022-2488/2007/48共6兲/065202/14/$23.00
48, 065202-1
© 2007 American Institute of Physics
Downloaded 07 Jun 2007 to 155.198.190.103. Redistribution subject to AIP license or copyright, see http://jmp.aip.org/jmp/copyright.jsp
065202-2
J. Math. Phys. 48, 065202 共2007兲
J. D. Gibbon and G. A. Pavliotis
LANS-␣ equations for which Foias, Holm, and Titi7 have proved existence and uniqueness of
solutions. Following on from this, Gibbon and Holm8 have demonstrated that the dimension of the
global attractor for this system has an upper bound proportional to Re9/4. An important milestone
has been passed recently in another closely related problem with the establishment by Cao and
Titi9 of an existence and uniqueness proof for Richardson’s three-dimensional primitive equations
for the atmosphere.
For the Navier–Stokes equations the idea sits more naturally in studies in the two-dimensional
context. The existence and uniqueness of solutions has been a closed problem for many decades
and the nature of the global attractor has been well-established.1–5,10–14 While the two- and threedimensional equations have the same velocity formulation, in reality, the former have a tenuous
connection with the latter because of the absence of the drastic property of vortex stretching. As a
result, the presence of vortex stretching in three dimensions, and perhaps other more subtle
properties, have set up seemingly unsurmountable hurdles even on periodic boundary conditions.
For problems on non-periodic boundaries, such as lid-driven flow, solving the two-dimensional
Navier–Stokes equations is a technically more demanding problem—see some references in Refs.
10, 15, and 16.
The sharp estimate found by Constantin, Foias, and Temam1 for the Lyapunov dimension of
the global attractor A expressed in terms of the Grashof number Gr
dL共A兲 艋 c1Gr2/3共1 + ln Gr兲1/3 ,
共1.2兲
has been one of the most significant results in two-dimensional Navier–Stokes analysis on a
2
periodic domain ⍀ = 关0 , L兴per
. The traditional length scale in the two-dimensional Navier–Stokes
equations is the Kraichnan length, k, which plays an equivalent role in two dimensions to that of
the Kolmogorov length, k, which is more important in three dimensions. In two dimensions, k
and k are defined respectively in terms of the enstrophy and energy dissipation rates ⑀ens and ⑀,
⑀ens = L−2
冓冕
⍀
冔
⑀ = L−2
兩ⵜ兩2dV ,
冓冕 冔
⍀
兩兩2dV ,
共1.3兲
where the pair of brackets 具·典 denote a long-time average defined as2,3,10–13
具g共·兲典 = lim lim sup
t→⬁
g共0兲
1
t
冕
t
g共兲d .
共1.4兲
0
−1
The inverse Kraichnan length −1
k and the inverse Kolmogorov length k are defined in terms of
⑀ens and ⑀ as
−1
k =
冉 冊
⑀ens
3
1/6
,
−1
k =
冉冊
⑀
3
1/4
.
共1.5兲
It has been shown by Constantin, Foias, and Temam1 that instead of using an estimate for ⑀ens in
terms of Gr, the upper bound for dL can be re-expressed in terms of L−1
k 共see other literature on
this topic17–19兲
−1 1/3
2
dL 艋 c2共L−1
k 兲 兵1 + ln共Lk 兲其 .
共1.6兲
If dL is identified with the number of degrees of freedom N, this result is consistent with the idea
expressed in Eq. 共1.1兲 that in a two-dimensional domain, the average length scale of the smallest
vortical feature can be identified with the Kraichnan length k, to within log corrections. The
result in Eq. 共1.2兲 has also been improved by Foias et al.20,21 to an estimate proportional to Gr1/2
共to within logarithmic corrections兲 provided Kraichnan’s theory of fully developed turbulence is
implemented.22
While these results display a pleasing convergence between rigorous estimates and scaling
methods in the two-dimensional case, the tradition in Navier–Stokes analysis of finding estimates
Downloaded 07 Jun 2007 to 155.198.190.103. Redistribution subject to AIP license or copyright, see http://jmp.aip.org/jmp/copyright.jsp
065202-3
Estimates for the 2D Navier-Stokes equations
J. Math. Phys. 48, 065202 共2007兲
in terms of the Grashof number Gr, whose character depends on the ratio of the forcing to the
viscosity , means that it is difficult to compare with the results of scaling theories whose results
are expressed in terms of Reynolds number. One of the tasks of this paper is to estimate quantities
of physical relevance, particularly long-time averages, in terms of the Reynolds number, whose
character depends on the fluid response to the forcing, and which is intrinsically a property of
Navier–Stokes solutions. Doering and Foias23 have addressed this problem and have shown that in
the limit Gr→ ⬁, solutions of the d-dimensional Navier–Stokes equations must satisfy 关This result
is not advertised in Ref. 23 but follows immediately from their Eq. 共48兲.兴
Gr 艋 c共Re2 + Re兲,
共1.7兲
while the energy dissipation rate ⑀ has a lower bound proportional to Gr. The problem, however,
is not as simple as replacing standard estimates in terms of Gr by Re2 from Eq. 共1.7兲. Estimates
such as that for dL in Eq. 共1.2兲 and the inverse Kraichnan and Kolmogorov lengths defined in Eq.
共1.5兲, depend upon long time averages of the enstrophy and energy dissipation rates defined in Eq.
共1.3兲. Other estimates of inverse length scales 共to be discussed in Sec. I B兲 also depend upon long
time averages. When estimated in terms of Re all these turn out to be better than straight substitution using Eq. 共1.7兲. These results are summarized in Sec. I B and worked out in detail in
Sec. II.
The second topic to be addressed in this paper is that of intermittency. Originally this important effect was considered to be a high Reynolds number phenomenon associated with threedimensional Navier–Stokes flows. First discovered by Batchelor and Townsend,24 it manifests
itself in violent fluctuations of very short duration in the energy dissipation rate ⑀. These violent
fluctuations away from the average are interspersed by quieter, longer periods in the dynamics.
This is a well established, experimentally observable phenomenon;25–27 its appearance in systems
other than the Navier–Stokes equations has been discussed in an early and easily accessible paper
by Frisch and Morf.28 One symptom of its occurrence is the deviation of the “flatness” of a
velocity signal 共the ratio of the fourth order moment to the square of the second order moment兲
from the value of 3 that holds for Gaussian statistics.
Recent analysis discussing intermittency in three-dimensional Navier–Stokes flows shows that
while it may be connected with loss of regularity, the two are subtly different issues.29 This is
reinforced by the fact that although solutions of the two-dimensional Navier–Stokes equations
remain regular for arbitrarily long times, nevertheless many of its solutions at high Re are known
to be intermittent.30–34 While three-dimensional analysis of the problem is based on the assumption that a solution exists,29,35 so that the higher norms can be differentiated, no such assumption
is necessary in the two-dimensional case where existence and uniqueness are guaranteed. The
result in both dimensions is such that the time-axis is broken up into good and bad intervals: on the
latter there exist large lower bounds on certain quantities, necessarily resulting in their extreme
narrowness and thus manifesting themselves as spikes in the data. This is summarized in Sec. I B
and worked out in detail in Sec. IV.
B. Summary and interpretation of results
For simplicity the forcing f共x兲 in the two-dimensional Navier–Stokes equations 共div u = 0兲
ut + u · u = ⌬u − ⵜp + f共x兲
共1.8兲
is taken to be divergence-free and smooth of narrow-band type, with a characteristic single lengthscale ᐉ such that23,29,35
储nf储2 ⬇ ᐉ−n储f储2 .
共1.9兲
Moreover, the aspect ratio of the forcing length scale to the box scale is defined as
Downloaded 07 Jun 2007 to 155.198.190.103. Redistribution subject to AIP license or copyright, see http://jmp.aip.org/jmp/copyright.jsp
065202-4
J. Math. Phys. 48, 065202 共2007兲
J. D. Gibbon and G. A. Pavliotis
aᐉ = L/ ᐉ .
共1.10兲
With f rms = L−d/2 储 f储2, the usual definition of the Grashof number Gr appearing in Eq. 共1.7兲 in d
dimensions is
Gr =
ᐉ3 f rms
.
2
共1.11兲
The Reynolds number Re in Eq. 共1.7兲 is defined as
Re =
Uᐉ
,
U2 = L−d具储u储22典,
共1.12兲
where 具·典 is the long-time average defined in Eq. 共1.4兲. One of the main results of this paper is the
following theorem whose proof is given in Sec. II A. All generic constants are designated as c.
Theorem 1.1: Let u共x , t兲 be a solution of the two-dimensional Navier–Stokes equations 共1.8兲
on a periodic domain 关0 , L兴2, and subject to smooth, divergence-free, narrow-band forcing f共x兲.
Then estimates in terms of the Reynolds number Re and the aspect ratio aᐉ for the inverse
−1
Kraichnan length −1
k , the attractor dimension dL, and the inverse Kolmogorov length k are
given by
2
1/2
L−1
k 艋 c共aᐉRe兲 ,
共1.13兲
dL 艋 ca2ᐉRe关1 + ln Re兴1/3 ,
共1.14兲
5/8
L−1
k 艋 caᐉRe .
共1.15兲
In the short proof of this theorem in Sec. II A, the estimate for dL in Eq. 共1.14兲 is not reworked
from first principles but is derived from a combination of Eqs. 共1.13兲 and 共1.14兲. The result in Eq.
共1.15兲 comes from a Re5/2 bound on 具H1典 and has also recently been found by Alexakis and
Doering.36 It implies that
L⑀
艋 caᐉRe−1/2 ,
U3
共1.16兲
whereas in three dimensions the right hand side is O共1兲. The estimate in Eq. 共1.14兲 is also
consistent with the result of Foias et al.20 when their Gr1/2 estimate is converted to one proportioanl to Re. Their estimate, however, was based on the implementation of certain features of the
Kraichnan model,22 while Eq. 共1.14兲 is true for all solutions and requires no assumption of fully
developed turbulence.
1/2
1/2
The estimates for −1
k and dL are consistent with the long-standing belief that Re ⫻ Re
grid points are needed to numerically resolve a flow; indeed, when the aspect ratio is taken into
account, Theorem 1.1 is consistent with aᐉRe1/2 ⫻ aᐉRe1/2. However, both these estimates are
dependent upon only the time average of low moments of the velocity field. For non-Gaussian
flows, low-order moments are not sufficient to uniquely determine the statistics of a flow. Thus it
is necessary to find ways of estimating small length scales associated with higher-order moments.
In Sec. II B we follow the way of defining inverse length scales associated with derivatives higher
than 2, introduced elsewhere,18 by combining the forcing with higher derivatives of the velocity
field such that
Fn =
冕
⍀
共兩nu兩2 + 2兩nf兩2兲dV,
共1.17兲
where = ᐉ2−1关Gr共1 + ln Gr兲兴−1/2 is a characteristic time: this choice of is discussed in Appendix
A. The gradient symbol ⵜn within Eq. 共1.17兲 refers to all derivatives of every component of u of
Downloaded 07 Jun 2007 to 155.198.190.103. Redistribution subject to AIP license or copyright, see http://jmp.aip.org/jmp/copyright.jsp
065202-5
J. Math. Phys. 48, 065202 共2007兲
Estimates for the 2D Navier-Stokes equations
order n in L2共⍀兲. The Fn are used to define a set of time-dependent inverse length scales
n,r共t兲 =
冉 冊
Fn
Fr
1/2共n−r兲
共1.18兲
.
2n
Actually, n,0
behaves as the 2nth moment of the energy spectrum as shown by
冕
冕
⬁
2n
n,0
=
2/L
⬁
k2n共兩û兩2 + 2兩f̂兩2兲dVk
2/L
.
共1.19兲
共兩û兩 + 兩f̂兩 兲dVk
2
2
2
2共n−1兲
behaves as the 2共n − 1兲th moment of the enMore relevant to the two-dimensional case, n,1
strophy spectrum. Using Landau’s argument the dimension of the global attractor dL共A兲 was
identified with the number of degrees of freedom N. In Ref. 19 a definition was introduced to
represent the number of degrees of freedom associated with all higher derivatives of the velocity
field represented by n,r, which is itself an inverse length. This naturally leads to the definition of
the infinite set
2
典.
Nn,r = L2具n,r
共1.20兲
Using the definition of the quantities ⌳n,0 and ⌳n,1 共n 艌 2兲,
⌳n,0 =
3n − 2
,
2n
⌳n,1 =
3n − 4
,
2共n − 1兲
共1.21兲
the second main result of the paper is a theorem whose proof is given in Sec. II B.
Theorem 1.2: Let n,r be the moments of a two-dimensional Navier–Stokes velocity field
defined in Eq. 共1.18兲. Then in a two-dimensional periodic box of side L the numbers of degrees of
freedom Nn,1 and Nn,0 defined in Eq. 共1.20兲 are estimated as 共n 艌 2兲,
Nn,1 艋 cn,1共a2ᐉRe兲⌳n,1共1 + ln Re兲1/2 ,
共1.22兲
Nn,0 艋 cn,0共a2ᐉRe兲⌳n,0共1 + ln Re兲1/2 ,
共1.23兲
where ⌳n,0 and ⌳n,1 are defined in Eq. 共1.21兲.
Note that ⌳2,0 = ⌳2,1 = 1. Thus the estimate for the first in each sequence, N2,1 and N1,0, are of
the same order as the estimate for dL, namely a2ᐉRe共1 + ln Re兲1/3 except in the exponent of the
logarithm. The exponents in Eqs. 共1.22兲 and 共1.23兲 provide an estimate of the extra resolution that
is needed to take account of energy at sub-Kraichnan scales. Notice that in the limit n → ⬁ both
exponents converge to 3 / 2.
The intermittency results of Sec. IV show that there can exist small intervals of time where
2
that are much larger than the upper bound on the long-time
there are large lower bounds on n,1
2
average for 具n,1典. Translated into pictorial terms, Fig. 1 in Sec. IV is consistent with the existence
of spiky data whose duration must be very short. Estimates are found for the width of these spikes
which turn out to be in terms of a negative exponent of Re.
II. TIME AVERAGE ESTIMATES IN TERMS OF Re
A. Proof of Theorem 1.1
The first step in the proof of Theorem 1.1, which has been expressed in Sec. I B, is to find an
upper bound on 具H2典 in terms of Re. Consider the equation for the two-dimensional Navier–Stokes
vorticity = k̂,
Downloaded 07 Jun 2007 to 155.198.190.103. Redistribution subject to AIP license or copyright, see http://jmp.aip.org/jmp/copyright.jsp
065202-6
J. Math. Phys. 48, 065202 共2007兲
J. D. Gibbon and G. A. Pavliotis
+ u · = ⌬ + curl f,
t
共2.1兲
and let Hn be defined by 共n 艌 0兲,
Hn =
冕
⍀
兩nu兩2dV.
共2.2兲
For a periodic, divergence-free velocity field u,
H1 =
冕
⍀
兩u兩2dV =
冕
兩兩2dV.
共2.3兲
· curl fdV
共2.4兲
⍀
Then the evolution equation for H1 is
1
Ḣ1 = − H2 +
2
冕
⍀
艋− H2 + 储u储2储2f储2
共2.5兲
艋− H2 + ᐉ−2储u储2储f储2 ,
共2.6兲
where the forcing term has been integrated by parts in Eq. 共2.4兲 and the narrow-band property has
been used to move from Eq. 共2.5兲 to Eq. 共2.6兲. Using the definitions of Re, Gr, and aᐉ in Eqs.
共1.12兲, 共1.11兲, and 共1.10兲, the long-time average of H2 is estimated as
具H2典 艋 L2ᐉ−62Re Gr
艋ca2ᐉᐉ−42Re3 + O共Re2兲.
共2.7兲
共2.8兲
This holds the key to the three results in Theorem 1.1.
3
−2
The inverse Kraichnan length −6
k = ⑀ens / with ⑀ens = L 具H2典, can now be estimated by
noting that
L6⑀ens 艋 ca6ᐉ3Re3
共2.9兲
2
1/2
L−1
k 艋 c共aᐉRe兲 ,
共2.10兲
and so
which is Eq. 共1.13兲 of Theorem 1.1. The estimate for dL in Eq. 共1.14兲 then follows immediately
from the relation between the estimate for dL in Eqs. 共1.6兲 and 共2.10兲.
Finally, we turn to proving the estimate for 具H1典 in Eq. 共1.15兲 which turns around the use of
the simple inequality H21 艋 H2H0. The next step is to use the fact that
具H1典 艋 具H2典1/2具H0典1/2
= aᐉRe具H2典1/2 .
共2.11兲
共2.12兲
Using the upper bound in Eq. 共2.7兲 gives
具H1典 艋 c2a2ᐉᐉ2Re5/2 ,
共2.13兲
which then gives Eq. 共1.15兲 in Theorem 1.1. In fact, Eq. 共2.13兲 is an improvement in the bound for
具H1典 from Re3 to Re5/2. This result has also been found recently by Alexakis and Doering.36
Downloaded 07 Jun 2007 to 155.198.190.103. Redistribution subject to AIP license or copyright, see http://jmp.aip.org/jmp/copyright.jsp
065202-7
J. Math. Phys. 48, 065202 共2007兲
Estimates for the 2D Navier-Stokes equations
B. Proof of Theorem 1.2
Having introduced the notation for Hn in Eq. 共2.2兲, similar quantities are used that contain the
forcing,35,29 namely
Fn =
冕
⍀
共兩nu兩2 + 2兩ⵜnf兩2兲dV,
共2.14兲
defined first in Eq. 共1.17兲, and the moments n,r defined in Eq. 共1.18兲,
n,r共t兲 ª
冉 冊
Fn
Fr
1/2共n−r兲
.
共2.15兲
The parameter in Eq. 共2.14兲 is a time scale and needs to be chosen appropriately. The idea is that
it should be chosen in such a way that the forcing does not dominate the behavior of the moments
of the velocity field. Defining 0 = ᐉ−2, it is shown in Appendix A that this end is achieved if −1
is chosen as
−1 = 0关Gr共1 + ln Gr兲兴1/2
共2.16兲
艋c0Re共1 + ln Re兲1/2 .
共2.17兲
As a preliminary to the proof of Theorem 1.2, we state the ladder theorem proved in Refs. 35 and
29.
Theorem 2.1: The Fn satisfy the differential inequalities
1
2 Ḟ0
艋 − F1 + c−1F0 ,
共2.18兲
1
2 Ḟ1
艋 − F2 + c−1F1 ,
共2.19兲
艋 − Fn+1 + cn,1共储u储⬁ + −1兲Fn ,
共2.20兲
艋 − 21 Fn+1 + cn,2共−1储u储⬁2 + −1兲Fn .
共2.21兲
and, for n 艌 2, either
1
2 Ḟn
or
1
2 Ḟn
The L⬁ inequalities in Theorem 2.1, particularly 储u储⬁ in Eq. 共2.20兲, can be handled using a
modified form of the L⬁ inequality of Brezis and Gallouet that has already been proved in Ref. 18.
Lemma 2.1: In terms of the Fn of Eq. 共2.14兲 and 3,2 of Eq. 共2.15兲, a modified form of the
two-dimensional L⬁ inequality of Brezis and Gallouet is
1/2
储u储⬁ 艋 cF1/2
2 关1 + ln共L3.2兲兴 .
共2.22兲
2
典
具n,r
for r 艌 2.
This lemma directly leads to an estimate for
Lemma 2.2: For n ⬎ r 艌 2, to leading order in Re,
2
典 艋 c共a2ᐉRe兲3/2共1 + ln Re兲1/2 .
L2具n,r
共2.23兲
Proof: By dividing Eq. 共2.20兲 by Fn and time averaging, we have
2
具n+1,n
典 艋 cn,1具储u储⬁典 + c0Re共1 + ln Re兲1/2 .
共2.24兲
However, because n,r 艋 n+1,n for r ⬍ n, for every 2 艋 r ⬍ n, in combination with Lemma 2.1, we
have
Downloaded 07 Jun 2007 to 155.198.190.103. Redistribution subject to AIP license or copyright, see http://jmp.aip.org/jmp/copyright.jsp
065202-8
J. Math. Phys. 48, 065202 共2007兲
J. D. Gibbon and G. A. Pavliotis
2
1/2
1/2
具n,r
典 艋 c具F1/2
2 关1 + ln共L3,2兲兴 典 + c0Re共1 + ln Re兲 .
共2.25兲
The logarithm is a concave function and 3,2 艋 n,r so Jensen’s inequality gives
2
2
典 艋 L2−1c具F2典1/2具关1 + ln兵L2具n,r
典其兴典1/2 + ca2ᐉRe共1 + ln Re兲1/2 .
L2具n,r
共2.26兲
is no more than
The estimate for 具F2典 can be found from 具H2典 in Eq. 共2.7兲; the extra term 储
䊏
O共Re2兲. Standard properties of the logarithm turn inequality Eq. 共2.26兲 into Eq. 共2.23兲.
2
典 for r 艌 2. These are used in the following theorem to give
Lemma 2.2 gives estimates for 具n,r
better estimates for the cases r = 0 and r = 1. Prior to this, it is necessary to state the results that
immediately derive from Eqs. 共2.18兲 and 共2.19兲 by, respectively, dividing through by F0 and F1
before time averaging
2
2
典 艋 ca2ᐉRe共1 + ln Re兲1/2,
N1,0 ⬅ L2具1,0
2
f储22
2
N2,1 ⬅ L2具2,1
典 艋 ca2ᐉRe共1 + ln Re兲1/2 .
共2.27兲
With the estimates in Eq. 共2.27兲 we are now ready to complete the proof of Theorem 1.2.
Proof of Theorem 1.2: Let us return to Eq. 共2.23兲 in Lemma 2.2 and use the fact that
2
具n,1
典=
冓冉 冊 冉 冊 冔
Fn
F2
1/n−1
F2
F1
1/n−1
2共n−2兲/n−1 2/n−1
= 具n,2
2,1 典 ,
共2.28兲
and thus
2
2 n−2/n−1 2 1/n−1
典 艋 具n,2
典
具2,1典
.
具n,1
共2.29兲
Using Eq. 共2.23兲 in Lemma 2.2, together with Eq. 共2.27兲, for n 艌 2,
2
典 艋 cn,1共a2ᐉRe兲共3n−4兲/2共n−1兲关1 + ln Re兴1/2 ,
Nn,1 = L2具n,1
共2.30兲
which coincides with a2ᐉRe共1 + ln Re兲1/2 at n = 2 but converges to Re3/2共1 + ln Re兲1/2 as n → ⬁. The
exponent ⌳n,1 is defined in Eq. 共1.21兲.
Likewise, in the same manner as Eq. 共2.28兲 we have
2
2 共n−1兲/n 2 1/n
典 艋 具n,1
典
具1,0典 .
具n,0
共2.31兲
2
典 艋 cn,0共a2ᐉRe兲共3n−2兲/2n关1 + ln Re兴1/2 .
Nn,0 = L2具n,0
共2.32兲
Thus we find that for n 艌 2,
The exponent ⌳n,0 is defined in Eq. 共1.21兲.
䊏
III. POINT-WISE ESTIMATES
Let us consider the differential inequalities for H0 and H1:
1
2 Ḣ0
1
2 Ḣ1
艋 − H1 + 储f储2H1/2
0 ,
共3.1兲
艋 − H2 + ᐉ−2储f储2H1/2
0 ,
共3.2兲
having used the narrow-band property on Eq. 共3.2兲. Upon combining Poincaré’s inequality with
Lemmas B.1 and B.2 in Appendix B we obtain
limt→⬁H0 艋 ca6ᐉ2Gr2 艋 ca6ᐉ2Re4 ,
共3.3兲
limt→⬁H1 艋 cᐉ−2a6ᐉ2Gr2 艋 cᐉ−2a6ᐉ2Re4 .
共3.4兲
and
The additive forcing terms in F1 and F0 are of a lower order in Re so we end up with
Downloaded 07 Jun 2007 to 155.198.190.103. Redistribution subject to AIP license or copyright, see http://jmp.aip.org/jmp/copyright.jsp
065202-9
J. Math. Phys. 48, 065202 共2007兲
Estimates for the 2D Navier-Stokes equations
limt→⬁F0 艋 ca6ᐉ2Re4 + O共Re2兲,
共3.5兲
limt→⬁F1 艋 cᐉ−2a6ᐉ2Re4 + O共Re2兲.
共3.6兲
The estimate for F1 enables us to obtain point-wise estimates on Fn, n 艌 2 共Ref. 18, Sec. 7.2兲. In
fact we have the following lemma.
Lemma 3.1: As Gr→ ⬁,
4n
limt→⬁Fn 艋 cn2ᐉ−2na6n
ᐉ Re .
共3.7兲
Proof: Applying a Gagliardo–Nirenberg inequality in two-dimensions to u we obtain
a/2 共1−a兲/2
,
储u储⬁ 艋 c储nu储a2储u储1−a
2 艋 cFn F1
共3.8兲
with a = 1 / 共n − 1兲. Using this in Eq. 共2.20兲 gives
1
2 Ḟn
艋 − Fn+1 + cnF1+a/2
F1−a/2
+ c0Re共1 + ln Re兲1/2Fn .
1
n
共3.9兲
Moreover, the following inequality can easily be proved using Fourier transforms:
q
p
FN+q
,
FNp+q 艋 FN−p
共3.10兲
from which, with N = n, p = n − 1, q = 1, it can be deduced that
− Fn+1 艋 −
Fn/共n−1兲
n
F1/共n−1兲
1
共3.11兲
.
We now use Eq. 共3.11兲 in Eq. 共3.9兲 to obtain
Fn/共n−1兲
1
n
F1−a/2
+ c0Re共1 + ln Re兲1/2Fn ,
Ḟn 艋 − 1/共n−1兲 + cnF1+a/2
1
n
2
F1
共3.12兲
with a = 1 / 共n − 1兲. We use now estimate Eq. 共3.6兲 in Eq. 共3.12兲 with the further use of Lemma B.2
to obtain
2n
limt→⬁Fn 艋 cn2ᐉ−2na6n
ᐉ Gr ,
which leads to the result.
The above Lemma enables us to obtain an estimate on the wave numbers n,r.
Lemma 3.2: For n ⬎ r 艌 0, as Gr→ ⬁,
limt→⬁共Ln,r兲 艋 cna共4n−r−1兲/共n−r兲
Re共2n−1兲/共n−r兲共1 + ln Re兲1/2共n−r兲 .
ᐉ
共3.13兲
䊏
共3.14兲
Proof: Essentially one uses the upper bound on Fn and the lower bound on Fr which can be
calculated from the forcing part in terms of Gr, leading to the result 共see also Ref. 18, Chap. 7兲.䊏
IV. INTERMITTENCY: GOOD AND BAD INTERVALS
The issue of intermittency in solutions of the two-dimensional Navier–Stokes equations is
now addressed. While the Fn and n,r are bounded from above for all time, nevertheless it is
possible that their behaviour could be spiky in an erratic manner. To show how this might come
about, consider the definition of n,r in Eq. 共1.18兲 from which we find
2
Fn+1 = n,r
冉 冊
n+1,r
n,r
2共n+1−r兲
Fn .
共4.1兲
Now consider inequality 共3.9兲 rewritten as
Downloaded 07 Jun 2007 to 155.198.190.103. Redistribution subject to AIP license or copyright, see http://jmp.aip.org/jmp/copyright.jsp
065202-10
J. Math. Phys. 48, 065202 共2007兲
J. D. Gibbon and G. A. Pavliotis
冉 冊 冉 冊
1 Ḟn
n+1,1
2
艋 − n,1
2 Fn
n,1
n+1,1 n
1/2
n,1F1/2
1 + c0Re共1 + ln Re兲 ,
n,1
2n
+ cn
共4.2兲
where we have used Eq. 共4.1兲 and the fact that n,1 艋 n+1,1 in the middle term. Using Young’s
inequality on this same term we end up with
冉 冊
1 2 n+1,1
1 Ḟn
艋 − n,1
2 Fn
2
n,1
2n
+ cn−1F1 + c0Re共1 + ln Re兲1/2 .
共4.3兲
The main question is whether, for Navier–Stokes solutions, the lower bound on
n+1,1
艌1
n,1
共4.4兲
can be raised from unity. A variation on the interval theorem proved in Ref. 29 is used.
Theorem 4.1: For any value of the parameter 苸 共0 , 1兲, the ratio n+1,1 / n,1 obeys the
long-time averaged inequality 共n 艌 2兲,
冓冋 冉 冊 册 冋
cn
−
册 冔
1/−1
2
共L2n,1
兲
2 1/−1
n+1,1
n,1
共a2ᐉRe兲⌳n,1共1 + ln Re兲1/2
艌 0,
共4.5兲
where the cn are the same as those in Theorem 1.2. Hence there exists at least one interval of time,
designated as a “good interval”, on which the inequality
cn
冉 冊
n+1,1
n,1
2
共L2n,1
兲
2
艌
共4.6兲
共a2ᐉRe兲⌳n,1共1 + ln Re兲1/2
holds. Those other parts of the time-axis on which the reverse inequality
cn
冉 冊
n+1,1
n
2
共L2n,1
兲
2
⬍
共4.7兲
共a2ᐉRe兲⌳n,1共1 + ln Re兲1/2
holds are designated as “bad intervals”.
Remark: In principle, the whole time-axis could be a good interval, whereas the positive time
average in Eq. 共4.5兲 ensures that the complete time axis cannot be “bad”. This paper is based on
the worst-case supposition that bad intervals exist, that they could be multiple in number, and that
the good and the bad are interspersed. The precise distribution and occurrence of the good/bad
intervals and how they depend on n remains an open question. The contrast between the twodimensional and three-dimensional Navier–Stokes equations is prominent; while no singularities
can occur in the n,1 in the two-dimensional case, in three dimensions it is within these bad
intervals that they can potentially occur.
Proof: Take two parameters 0 ⬍ ⬍ 1 and 0 ⬍ ␣ ⬍ 1 such that + ␣ = 1. The inverses −1 and
−1
␣ will be used as exponents in the Hölder inequality on the far right-hand side of
2␣
2␣
典 艋 具n+1,1
典=
具n,1
thereby giving
冓冉 冊 冔 冓冉 冊 冔
n+1,1
n,1
2␣
2␣
n,1
艋
冓冉 冊 冔 冉 冊
n+1,1
n,1
2␣/
艌
2␣
具n,1
典
2 ␣
具n,1
典
1/
n+1,1
n,1
2␣
= 具n,1
典
2␣/
冉 冊
2␣
具n,1
典
2
具n,1
典
2 ␣
具n,1
典 ,
共4.8兲
␣/
.
共4.9兲
Two-dimensional Navier–Stokes information can be injected into these formal manipulations: the
2
典 from Theorem 1.2 and the lower bound Ln,1 艌 1 are used in the ratio on the
upper bound on 具n,1
䊏
far right-hand side of Eq. 共4.9兲 to give Eq. 共4.5兲, with the same cn as in Theorem 1.2.
Downloaded 07 Jun 2007 to 155.198.190.103. Redistribution subject to AIP license or copyright, see http://jmp.aip.org/jmp/copyright.jsp
065202-11
Estimates for the 2D Navier-Stokes equations
J. Math. Phys. 48, 065202 共2007兲
Now consider what must happen on bad intervals. It is always true that n+1,1 / n,1 艌 1, so Eq.
共4.7兲 implies that on these intervals there is a lower bound
2
L2n,1
⬎ cn共a2ᐉRe兲⌳n,1/共1 + ln Re兲1/2 .
共4.10兲
This lower bound cannot be greater than the upper point-wise bound in Eq. 共3.14兲, which means
that is restricted by
冉 冊
2n − 1
⌳n,1
⬍2
.
n−1
共4.11兲
Moreover, the factor of 1 / in the exponent makes the lower bound in Eq. 共4.10兲 much larger than
2
the upper bound on the average 具n,1
典 given in Theorem 1.2. These intervals must therefore be
very short. To estimate how large they can be requires an integration of Eq. 共4.3兲 over short times
⌬t = t − t0 which, in turn, requires the time integral of H1 for short times ⌬t. We use the notation
兰⌬t = 兰tt , with the definition 0 = ᐉ−2.
0
Lemma 4.1: To leading order in Re
冕
⌬t
F1dt 艋 a4ᐉ关c1a2ᐉ + c20⌬t兴Re4 .
共4.12兲
Proof: Integrating Eq. 共3.1兲 over a short time ⌬t gives
冕
1
H1dt 艋 H0共t0兲 + ⌬t关ᐉ−23a4ᐉGr2兴 艋 c1a6ᐉ2Re4 + ⌬t关c2ᐉ−23a4ᐉRe4兴,
2
⌬t
共4.13兲
䊏
having used Eq. 共3.3兲 for the 21 H0共t0兲 term. The forcing term in F1 is only O共Re2兲.
Now we wish to estimate 0⌬t in terms of Re. Integrating Eq. 共4.3兲, using 共4.13兲 and the
lower bound Eq. 共4.10兲 and multiplying by ᐉ2, we have
1 2
2 ᐉ 关ln
2
⌳n,1/
Fn共t兲 − ln Fn共t0兲兴 + 21 cna−2
共1 + ln Re兲1/2⌬t 艋 ᐉ2a4ᐉ关c1a2ᐉ + c20⌬t兴Re4
ᐉ 共aᐉRe兲
+ cᐉ20⌬tRe共1 + ln Re兲1/2 .
共4.14兲
As Gr→ ⬁, the dominant terms are
2
⌳n,1/
0⌬t兵a−2
共1 + ln Re兲1/2 − a6ᐉRe4其 艋 c1a6ᐉRe4 .
ᐉ 共aᐉRe兲
共4.15兲
Choosing in the range, to leading order we have
⬍ 41 ⌳n,1 ,
共4.16兲
0⌬t 艋 c共a2ᐉRe兲4−⌳n,1/ .
共4.17兲
then ⌬t must satisfy
Because the exponent in Eq. 共4.17兲 is necessarily negative these intervals are very small and
decreasing with increasing Re. Combining Eq. 共4.11兲 with Eq. 共4.16兲 we have
1
共n − 1兲
⌳n,1 ⬍ ⬍ ⌳n,1 ,
4
2共2n − 1兲
共4.18兲
which actually holds for every n 艌 1. Figure 1 is a cartoon-like figure displaying the lower bound
on the bad intervals of width 共⌬t兲b and also the maximum of n,1 allowed by Eq. 共3.14兲 in Lemma
3.2. The full dynamics of two-dimensional Navier–Stokes is actually determined by the intersection of all cartoons for every n 艌 3 on the grounds that the position and occurrence of the bad
intervals varies with n. Thus we are interested in the limit n → ⬁ which determines that the range
of is squeezed between
Downloaded 07 Jun 2007 to 155.198.190.103. Redistribution subject to AIP license or copyright, see http://jmp.aip.org/jmp/copyright.jsp
065202-12
J. Math. Phys. 48, 065202 共2007兲
J. D. Gibbon and G. A. Pavliotis
FIG. 1. A cartoon, not to scale, of good/bad intervals for some value of n 艌 3.
冉 冊
冉 冊
5
3
1
3
1−
⬍⬍ 1−
.
8
8
6n
3n
共4.19兲
Thus, in the limit, takes a value just under 3 / 8. We conclude that the interval theorem 共Theorem
4.1兲 reproduces the effects of intermittency in a two-dimensional flow by manifesting very large
lower bounds within bad intervals and suppressing spiky behavior within the good intervals which
must be quiescent for long intervals, otherwise the long-time average would be violated.
ACKNOWLEDGMENTS
The authors would like to thank Matania Ben-Artzi, Charles Doering, Darryl Holm, Haggai
Katriel, and Edriss Titi for comments and suggestions. J.D.G. would also like to thank the Mathematics Departments of the Weizmann Institute of Science and the Hebrew University of Jerusalem for their hospitality during December 2005 and January 2006 when some of these ideas were
conceived.
APPENDIX A: FORCING AND THE FLUID RESPONSE
For technical reasons, we must address the possibility that in their evolution the quantities Hn
might take small values. Thus we need to circumvent problems that may arise when dividing by
these 共squared兲 seminorms. We follow Doering and Gibbon35 who introduced the modified quantities
Fn = Hn + 2储ⵜnf储22 ,
共A1兲
where the “time scale” is to be chosen for our convenience. So long as ⫽ 0, the Fn are bounded
2
away from zero by the explicit value 2L3ᐉ−2n f rms
. Moreover, we may choose to depend on the
parameters of the problem such that 具Fn典 – 具Hn典 as Gr→ ⬁. To see how to achieve this, let us define
= ᐉ2−1关Gr共1 + ln Gr兲兴−1/2 .
共A2兲
Then the additional term in Eq. 共A1兲 is
2
2储nf储22 = L3−2ᐉ4−2n f rms
关Gr共1 + ln Gr兲兴−1 = 2ᐉ−共2n+2兲L3Gr共1 + ln Gr兲−1 .
23
Now Doering and Foias
bound of the form
共A3兲
proved that in d dimensions, the energy dissipation rate ⑀ has a lower
Downloaded 07 Jun 2007 to 155.198.190.103. Redistribution subject to AIP license or copyright, see http://jmp.aip.org/jmp/copyright.jsp
065202-13
J. Math. Phys. 48, 065202 共2007兲
Estimates for the 2D Navier-Stokes equations
⑀ 艌 c3ᐉ−3L−1Gr.
共A4兲
Using this on the far right-hand side of Eq. 共A3兲 we arrive at
2储ⵜnf储22 艋 c6⑀ᐉ−共2n−1兲L4−1共1 + ln Gr兲−1 = c6
冉冊
L
ᐉ
共2n−1兲
L−2共n−1兲具H1典 ⬎ 共1 + ln Gr兲−1 .
共A5兲
Using Poincaré’s inequality in the form H1 艋 共2L兲2共n−1兲Hn, as Gr→ ⬁ we have
2储ⵜnf储22
艋 c6a共2n−1兲
共1 + ln Gr兲−1 .
ᐉ
具Hn典
共A6兲
Hence, the additional forcing term in Eq. 共A1兲 becomes negligible with respect to 具Hn典 as Gr
→ ⬁, so the forcing does not dominate the response.
APPENDIX B: COMPARISON THEOREMS FOR ODE
We present a comparison theorem for ODE which is useful for obtaining various estimates.
We start with the following classical result.
Lemma B.1: Let f : 关0 , T兴 ⫻ R → R be a continuous function which is locally Lipschitz
uniformly in t: for all intervals 关a , b兴 傺 R there exists a constant such that 兩f共s , x兲 − f共s , y兲 兩
艋 C 兩 x − y兩 for all x , y 苸 关a , b兴 and all s 苸 关0 , T兴. Furthermore, let x 苸 AC共关0 , T兴 , R兲 be such that
ẋ共t兲 艋 f关t,x共t兲兴
for all t 苸 关0 , T兴 and let y共t兲 be the solution of ẏ共t兲 = f关t , y共t兲兴 on 关0 , T兴. Assume further that x共0兲
艋 y共0兲. Then, x共t兲 艋 y共t兲 for all t 苸 关0 , T兴.
We can use this Lemma to prove the following useful result.
Lemma B.2: Let x : 关0 , T兴 → 关0 , ⬁ 兲 be an absolutely continuous function with x共0兲 ⬎ 0 which
satisfies
ẋ 艋 ⌬0x + Fxn1 − Exn2 ,
共B1兲
where ⌬0 , F , E ⬎ 0 and 1 ⬍ n1 ⬍ n2. Then
lim sup x共t兲 艋 共4⌬0E−1兲1/n2−1 + 共2FE−1兲1/n2−n1 .
t→⬁
共B2兲
P. Constantin, C. Foias, and R. Temam, Physica D 30, 284 共1988兲.
P. Constantin and C. Foias, Navier-Stokes Equations 共The University of Chicago Press, Chicago, 1988兲.
3
C. Foias, O. Manley, R. Rosa, and R. Temam, Navier-Stokes Equations and Turbulence 共Cambridge University Press,
Cambridge, 2001兲.
4
R. Temam, Infinite Dimensional Dynamical Systems in Mechanics and Physics, Applied Mathematical Sciences, Vol. 68
共Springer-Verlag, New York, 1988兲.
5
D. A. Jones and E. S. Titi, Indiana Univ. Math. J. 42, 875 共1993兲.
6
L. D. Landau and E. M. Lifshitz, Fluid Mechanics 共Pergamon, Oxford, 1986兲.
7
C. Foias, D. D. Holm and E. S. Titi, J. Dyn. Differ. Equ. 14, 1 共2002兲.
8
J. D. Gibbon and D. D. Holm, Physica D 220, 6978 共2006兲.
9
C. Cao and E. S. Titi, Ann. Math. 共to be published兲 , Vol. 165 共2007兲.
10
R. Temam, Navier-Stokes Equations and Non-Linear Functional Analysis, CBMS-NSF Regional Conference Series in
Applied Mathematics, Vol. 66, 2nd ed. 共SIAM, Philadelphia, 1995兲.
11
C. Foias, Rend. Sem. Mat. Univ. Padova 48, 219 共1972兲.
12
C. Foias, Rend. Sem. Mat. Univ. Padova 49, 9 共1973兲.
13
C. Foias and G. Prodi, Ann. Mat. Pura Appl. 111, 307 共1976兲.
14
O. A. Ladyzhenskaya, The Mathematical Theory of Viscous Incompressible Flow 共Gordon and Breach, New York, 1963兲.
15
E. J. Dean, R. Glowinski, and O. Pironneau, Comput. Methods Appl. Mech. Eng. 81, 117–156 共1991兲.
16
M. Ben-Artzi, D. Fishelov, and S. Trachtenburg, Math. Modell. Numer. Anal. 35, 313 共2001兲.
17
C. R. Doering and J. D. Gibbon, Physica D 48, 471 共1991兲.
18
C. R. Doering and J. D. Gibbon, Applied Analysis of the Navier–Stokes Equations 共Cambridge University Press, Cambridge, 1995兲.
19
J. D. Gibbon, Physica D 92, 133 共1996兲.
1
2
Downloaded 07 Jun 2007 to 155.198.190.103. Redistribution subject to AIP license or copyright, see http://jmp.aip.org/jmp/copyright.jsp
065202-14
J. D. Gibbon and G. A. Pavliotis
J. Math. Phys. 48, 065202 共2007兲
C. Foias, M. S. Jolly, O. P. Manley, and R. Rosa, J. Stat. Phys. 111, 1017–1019 共2003兲.
C. Foias, M. S. Jolly, O. P. Manley, and R. Rosa, J. Stat. Phys. 108, 591–645 共2002兲.
22
R. H. Kraichnan, Phys. Fluids 10, 1417–1423 共1967兲.
23
C. R. Doering and C. Foias, J. Fluid Mech. 467, 289 共2002兲.
24
G. K. Batchelor and A. A. Townsend, Proc. R. Soc. London, Ser. A 199, 238 共2002兲.
25
A. Y.-S. Kuo and S. Corrsin, J. Fluid Mech. 50, 285 共1971兲.
26
C. Meneveau and K. Sreenivasan, J. Fluid Mech. 224, 429 共1991兲.
27
U. Frisch, Turbulence: The Legacy of A. N. Kolmogorov 共Cambridge University Press, Cambridge, 1995兲.
28
U. Frisch and R. Morf, Phys. Rev. A 23, 2673 共1991兲.
29
J. D. Gibbon and C. R. Doering, Arch. Ration. Mech. Anal. 177, 115 共2005兲.
30
K. Schneider, M. Farge, and N. Kevlahan, http://www.l3m.univ-mrs.fr/site/sfk-woodshole2004.pdf.
31
S. Chen, R. E. Ecke, G. L. Eyink, M. Rivera, M. Wan, and Z. Xiao, Phys. Rev. Lett. 96, 084502 共2006兲.
32
J. Paret and P. Tabeling, Phys. Fluids 10, 3126 共1998兲.
33
C. Jullien, P. Castiglione, and P. Tabeling, Phys. Rev. E 64, 035301 共2001兲.
34
J. Paret, A. Babiano, T. Dubos, and P. Tabeling, Phys. Rev. E 64, 036302 共2001兲.
35
C. R. Doering and J. D. Gibbon, Physica D 165, 163 共2002兲.
36
A. Alexakis and C. R. Doering, Phys. Lett. A, available online 2 August 共2006兲.
20
21
Downloaded 07 Jun 2007 to 155.198.190.103. Redistribution subject to AIP license or copyright, see http://jmp.aip.org/jmp/copyright.jsp