Professional Documents
Culture Documents
EMPLISSANT LESPACE1
Par
JEAN LERAY
`
a Rennes
Introduction.2
I. La theorie de la viscosite conduit `a admettre que les mouvements des liquides
visqueux sont regis par les equations de Navier; il est necessaire de justifier a posteriori cette hypoth`ese en etablissant le th`eoreme dexistence suivant: il exist une solution des
`equations de Navier qui correspond `a un etat de vitesse donne arbitrairement `a linstant
initial. Cest ce qua cherche `a demontrer M. Oseen3 ; il na reussi `a etablir lexistence
dune telle solution que pour une duree tres br`eve succedant `a linstant initial. On peut
verifier en outre que lenergie cinetique totale du liquide reste bornee4 ; mais il ne semble
pas possible de deduire de ce fait que le mouvement lui-meme reste regulier; jai meme
indique une raison qui me fait croire `a lexistence de mouvements devenant irreguliers au
bout dun temps fini5 ; je nai malheureusement pas reussi `a forger un exemple dune telle
singularite.
1
Ce memoire a ete resume dans une note parue aux Comptes rendus de lAcademie des
Sciences, le 20 fevrier 1933, T. 196, p. 527.
2
Les pages 5963 de ma These (Journ. de Math. 12, 1933) announcent ce memoire et en
completent lintroduction.
3
Voir Hydrodynamik (Leipzig, 1927), 7, p. 66. Acta mathematica T. 34. Arkiv f
or
matematik, astronomi och fysik. Bd. 6, 1910. Nova acta reg. soc. scient. Upsaliensis Ser. IV,
Vol. 4, 1917.
4
l. c. 2, p. 5960.
5
l. c. 2, p. 6061. Je reviens sur ce sujet an 20 du present travail (p. 224).
Introduction.2
I. The theory of viscosity leads one to allow that motions of a viscous liquid are
governed by Naviers equations. It is necessary to justify this hypothesis a posteriori by
establishing the following existence theorem: there is a solution of Naviers equations which
corresponds to a state of velocity given arbitrarily at an initial instant. That is what Oseen
tried to prove3 . He only succeeded in establishing the existence of such a solution for a
possibly very short time after the initial instant. One can also verify that the total kinetic
energy of the liquid remains bounded4 but it does not seem possible to deduce from this
fact that the motion itself remains regular . I have indicated a reason which makes me
believe there are motions which become irregular in a finite time5 . Unfortunately I have
not succeeded in creating an example of such a singularity.
1
This paper has been summarized in a note which appeared in Comptes rendus de lAcademie
des Sciences, February 20 1933, vol. 196, p. 527.
2
Pages 5963 of my Thesis (Journ. de Math. 12, 1933) announce this paper and complement
this introduction.
3
See Hydrodynamik (Leipzig, 1927), 7, p. 66. Acta mathematica vol. 34. Arkiv f
or
matematik, astronomi och fysik. Bd. 6, 1910. Nova acta reg. soc. scient. Upsaliensis Ser. IV,
Vol. 4, 1917.
4
l. c. 2, p. 5960.
5
l. c. 2, p. 6061. I return to this subject in 20 of the present work (p. 224).
194
Jean Leray.
Il nest pas paradoxal de supposer en effet que la cause qui regularise le mouvement
la dissipation de lenergie ne suffise pas `a maintenir bornees et continues les derivees
secondes des conposantes de la vitesse par rapport aux coordonnees; or la theorie de Navier
suppose ces derivees secondes bornees et continues; M. Oseen lui-meme a dej`a insiste sur
le caract`ere peu naturel de cette hypothese; il a montre en meme temps comment le
fait que le mouvement obeit aux lois de la mecanique peut sexprimer `a laide dequations
integro- differentielles1 , o
u figurent seulement les composantes de la vitesse et leurs derivees
premieres par rapport aux coordonnees spatiales. Au cours du present travail je consid`ere
justement un systeme de relations2 qui equivalent aux equations integro-differentielles de
M. Oseen, completes par une inegalite exprimant la dissipation de lenergie. Ces relations
se deduisant dailleurs des equations de Navier `a laide dintegrations par parties qui font
disparaitre les derivees dordres les plus eleves. Et, si je nai pu reussir `a etablir le theor`eme
dexistence enonce plus haut, jai pu neammoins demontrer le suivant3 : les relations en
question poss`edent toujours au moins une solution qui correspond `a un etat de vitesse
donne initialement et qui est definie pour une duree illimite, dont lorigine est linstant
initial. Peut-etre cette solution est-elle trop peu reguli`ere pur posseder `a tout instant des
derivees seconds bornees; alors elle nest pas, au sens propre du terme, une solution des
equations de Navier; je propose de dire quelle en constitue <<une solution turbulent>>.4
Il est dailleurs bien remarquable que chaque solution turbulente satisfait effectivement les equations de Navier proprement dites, sauf `a certaines epoques dirregularite; ces
epoques constituent un ensemble ferme de mesure nulle; `a ces epoques sont seules verifiees
certaines conditions de continuite extremement larges.
1
194
Jean Leray.
In fact it is not paradoxical to suppose that the thing which regularizes the motion
dissipation of energydoes not suffice to keep the second derivatives of the velocity components bounded and continuous. Naviers theory assumes the second derivatives bounded
and continuous. Oseen himself had already emphasised that this was not a natural hypothesis. He showed at the same time how the fact that the motion obeys the laws of mechanics
could be expressed by means of integro-differential equations1 which contain only the velocity components and their first spatial derivatives. In the course of the present work I
consider a system of relations2 equivalent to Oseens integro-differential equations complemented by an inequality expressing dissipation of energy. Moreover, these relations may
be deduced from Naviers equations, using an integration by parts which causes the higher
order derivatives to disappear. And, if I have not succeeded in establishing the existence
theorem stated above, I have nevertheless proved the following3 : the relations in question
always have at least one solution corresponding to a given initial velocity and which is defined for an unlimited time of which the origin is the initial instant. Perhaps that solution
is not sufficiently regular to have bounded second partial derivatives at each instant, so it
is not, in a proper sense of the term, a solution to Naviers equations. I propose to say
that it constitutes a turbulent solution.
It is moreover quite remarkable that each turbulent solution actually satisfies Naviers
equations, properly said, except at certain times of irregularity. These times constitute a
closed set of measure zero. At these times alone must the continuity of the solution be
interpreted in a very generous sense.
1
195
Une solution turbulent a donc la structure suivante: elle se compose dune succession de
solutions regulieres.
Si javais reussi `a construire des solutions des equations de Navier qui deviennent
irregulieres, jaurais le droit1 daffirmer quil existe effectivement des solutions turbulent
ne se reduisant pas,tout simplement, `a des solutions regulieres. Meme si cette proposition
etait fausse, la notion de solution turbulente, qui naurait des lors plus `a jouer aucun role
dans letude des liquides visqueux, ne perdrait pas son interet: il doit bien se presenter des
problems de Physique mathematique pour lesquels les causes physiques de regularite ne
suffisent pas `a justifier les hypotheses de regularite faites lors de la mise en equation: `
a ces
problems peuvent alors sappliquer des considerations semblables `a celles que jexpose ici.
Signalons enfin les deux faits suivants:
Rien ne permet daffirmer lunicite de la solution turbulente qui correspond `a un etat
initial donne. (Voir toutefois Complements 1o , p. 245; 33).
La solution qui correspond `a un etat initial suffisamment voisin du repos ne devient
jamais irreguli`ere. (Voir les cas de regularite que signalent les 21 et 22, p. 226 et 227).
II. Le travail present concerne les liquides visqueux illimites. Les conclusions en sont
extremement analogues `a celles dun autre memoire2 que jai consacre aux mouvements
plans des liquides visqueux enfermes dans des parois fixes convexes; ceci autorise `a croire
que ces conclusions setendent au cas general dun liquide visqueux `a deux ou trois dimensions que limitent des parois quelconques (meme variables).
Labsense de parois introduit certes quelques complications concernant lallure `a linfini
des fonctions inconnues3 , mais simplifie beaucoup lexpose et met mieux en lumi`ere les difficultes essentielles; le role important que joue lhomogeneite des formules est plus evident;
(les equations aux dimensions permettent de prevoir a priori presque toutes les inegalites
que nous ecrirons).
1
195
In virtue of the existence theorem of 31 (p. 241) and of the uniqueness theorem of 18 (p.
222).
2
196
Jean Leray.
Thesis, Journal de Math`ematiques 12, 1933; chapitre IV p. 64-82. (On peut donner
une variante interessante au procede que nous y employons en utilisant la notion detat initial
semiregulier quintroduit le memoire present.)
2
On peut dans ce cs baser letude du probl`eme sure la propriete que poss`ede alors le maximum
du tourbillon `
a un instant donne detre une fonction decroissante du temps. (Voir: Comptes rendus
de lAcademie des Sciences, T. 194; p. 1893; 30 mai 1932). M. Wolibner a lui aussi fait cette
remarque.
196
Jean Leray.
Recall that we have already treated the case of unlimited motions in the plane1 . These
are special2 because the motion is regular.
Summary of the paper.
Chapter I recalls a series of propositions of analysis which are important, but which
are not entirely classical.
Chapter II establishes several preliminary inequalities easily deduced from properties
of Oseens fundamental solution.
Chapter III applies the inequalities to the study of regular solutions of Naviers equation.
Chapter IV states several properties of regular solutions to be used in Chapter VI.
Chapter V establishes that for each initial state, there is at least one turbulent solution
defined for unlimited time. The proof of this existence theorem is based on the following
principle: one doesnt directly approach the problem of solving Naviers equations; instead one treats a neighboring problem which can be proved to have a regular solution of
unlimited duration; we let the neighboring problem tend toward the original problem to
construct the limit (or limits) of these solutions. There is an elementary method to apply
this principle: it is the same one which I used in my study of planar motion of a viscous
liquid within walls, but it is intimately bound with the structure of turbulent solutions
which we have previously pointed out. Without this structure the method may not apply.
Here we proceed in another fashion whose range is very likely larger, and which justifies
the notion of turbulent solution, but which requires calling on some of the less ordinary
theorems of chapter I.
Chapter VI studies the structure of turbulent solutions.
1
Thesis, Journal de Math`ematiques 12, 1933; chapter IV p. 64-82. (One can give an
interesting variation on the process used there, by using the notion of semi-regular initial state
introduced in the present paper.)
2
In this case one can base the study on the property that the maximum swirl is a decreasing
function of time. (See: Comptes rendus de lAcademie des Sciences, T. 194; p. 1893; 30 mai
1932). Wolibner has also made this remark.
[translators note: (reference provided by Shigeru Masuda, Tokyo): Wolibner, Un
Theor`eme sur lexistence du mouvement plan dun fluide parfait, homog`ene, incompressible, pendant un temps infin`ement long, Math. Z. 37 (1933), 698-726, (footnote: mark (**) in p.698)
197
I. Pr
eliminaires
1. Notations
Nous utiliserons la lettre 0 pour designer un domaine arbitraire de points de lespace;
0
pourra etre lespace tout entier, que nous designerons par ; $ designera un domaine
borne de points de , dont la fronti`ere constitue une surface reguli`ere .
Nous representerons un point arbitraire de par x, ses cooronnees cartesiennes par
xi (i = 1,2,3) sa distance `a lorigine r0 , un element de volume quil engendre par x, un
element de surface quil engendre par x1 , x2 , x3 . Nous designerons de meme par y un
second point arbitraire de ; r representera toujours la distance des points nommes x et
y.
Nous utileserons la convention << de lindice muet>>: un term o`
u un indice figure
deux fois representera la somme des termes obteunus en donnant `a cet indice successivement les valeurs 1, 2, 3.
A partir du chapitre II le symbole A nous servira `a designer les constantes dont nous
ne preciserons pas la valeur numerique.
Nous representerons systematiquement par de grandes lettres les fonctions que nous
supposerons seulement mesurables; par de petites lettres les fonctions qui sont continues
ainsi que leurs derivees premi`eres.
2. Rappelon linegalite de Schwarz:
Z Z Z
(1.1)
U (x)V (x) x
ZZZ
ZZZ
2
V 2 (x) x
U (x) x
On est assure que le premier membre a un sens quand le second est fini.
Cette inegalite est `a la base de toutes les proprietes enoncees au cours de ce chapitre.
Premi`ere application:
Si
U (x) = V1 (x) + V2 (x)
on a:
sZ Z Z
sZ Z Z
sZ Z Z
U 2 (x) x
0
V12 (x) x +
V22 (x) x;
197
I. Preliminaries
1. Notation
We use the letter for an arbitrary domain of points in space. may be the entire
space, denoted by . $ will be a bounded domain in which has as boundary a regular
surface . We represent an arbitrary point of by x, which has cartesian coordinates xi
(i = 1, 2, 3) and distance r0 to the origin, and generates volume element x and surface
element x1 , x2 , x3 . Similarly we use y for a second arbitrary point of . r will always
represent the distance between points named x and y.
We use the silent index convention: a term containing the same index twice represents the sum of terms obtained by successively giving that index the values 1, 2, 3.
Beginning with chapter II the symbol A denotes constants for which we do not specify
the numerical value.
We systematically use capital letters for measurable functions and lower case letters
for functions which are continuous and have continuous first partial derivatives.
2. Recall the Schwarz inequality:
Z Z Z
(1.1)
U (x)V (x) x
ZZZ
ZZZ
2
V 2 (x) x
U (x) x
sZ Z Z
sZ Z Z
sZ Z Z
U 2 (x) x
V12 (x) x +
V22 (x) x;
198
Jean Leray.
alors:
sZ Z Z
Z
U 2 (x) x
(1.2)
0
sZ Z Z
dt0
V 2 (x, t0 ) x
0
#2
p U (x +
~ p)
x <
"p=n
X
p=1
#2
|p |
ZZZ
U 2 (x) x;
p=1
(cette inegalite se demontre aisement en developpant les deux carres qui y figurent
et en utilisant linegalite de Schwarz). On en deduit la suivante qui nous sera tres utile:
Soit une fonction H(z): nous designerons par H(y x) la fonction que lon obtient en
substituant aux coordonees zi de z les composantes yi xi du vecteur xy;
~ nous avons:
Z Z Z Z Z Z
(1.3)
Z Z Z
x <
|H(z)| z
ZZZ
U 2 (y) y;
et on est assure que le premier membre est fini quand les deux integrals qui figurent
au second membre le sont.
3. Forte convergence en moyenne.1
Definition: On dit quune infinite de fonctions U (x) a pour forte limite en moyenne
sur un domaine 0 une fonction U (x) quand:
1
198
Jean Leray.
V (x, t0 ) dt0
U (x) =
0
then
sZ Z Z
sZ Z Z
dt0
U 2 (x) x
(1.2)
0
V 2 (x, t0 ) x
0
the left sides of these inequalities being finite when the right sides are.
Second application:
Consider n constants p and n constant vectors
~ p . Write x +
~ p for the translation
of x by the vector
~ p . We have
Z Z Z "p=n
X
#2
p U (x +
~ p)
"p=n
X
p=1
#2
|p |
ZZZ
U 2 (x) x;
p=1
(this inequality is easily proved by expanding the two squares and using the Schwarz
inequality). From it, one deduces the following very useful one. Let H(z) be a function.
We denote by H(y x) the function obtained by substituting for coordinates zi of z the
components yi xi of the vector xy.
~ We have
Z Z Z Z Z Z
(1.3)
Z Z Z
|H(z)| z
ZZZ
U 2 (y) y;
and the left side is finite when the two integrals on the right are finite.
3. Strong convergence in mean.1
Definition: One says that an infinity of functions U (x) have function U (x) as strong
limit in mean on a domain 0 when:
1
Translators note: The last two inequality signs appeared as < in the original.
199
ZZZ
(1.4)
[U (x) U (x)]2 x = 0.
lim
0
ZZZ
lim
U (x)A(x) x =
U (x)A(x) x.
ZZZ
2
lim
U 2 (x) x.
U (x) x =
0
lim
U (x)A(x) x =
U (x)A(x) x.
0
Exemple I. La suite sin x1 , sin 2x1 , sin 3x1 , . . . converge faiblement vers zero sur tout
domaine $ .
Exemple II. Soit une infinite de fonctions U (x) admettant une fonction U (x) comme forte
limite en moyenne
tout domaine $ , elle ladmet comme faible limite en moyenne sur quand
RRR sure
2
les quantites
U
(x)
x sont bournees dans leur ensemble.
0
Exemple III. Soit une infinite de fonctions U (x) qui sur un domaine 0 convergent presque
partout
vers une fonction U (x); cette fonction est leur faible limite en moyenne quand les quantites
RRR
2
U (x) x sont bournees dans leur ensemble.
0
On a:
(1.7) Z Z Z
ZZZ
ZZZ
lim
ZZZ
2 0
2 0
199
ZZZ
(1.4)
[U (x) U (x)]2 x = 0.
lim
0
ZZZ
lim
U (x)A(x) x =
U (x)A(x) x.
ZZZ
2
lim
U 2 (x) x.
U (x) x =
0
lim
U (x)A(x) x =
U (x)A(x) x.
0
Example I. The sequence sin x1 , sin 2x1 , sin 3x1 , . . . converges weakly to zero on all
domains $.
Example II. If an infinity of functions U (x) have strong limit U (x) in mean on
all
$, then they admit a weak limit in mean on when the set of quantities
RRR domains
2
U
(x)
x
is bounded.
0
Example III. Let an infinity of functions U (x) on a domain 0 converge almost everywhere to
U (x). That function is their weak limit in mean when the set of
RRRa function
2
quantities 0 U (x) x is bounded.
One has
(1.7) Z Z Z
lim
1 0
ZZZ
ZZZ
ZZZ
2 0
200
Jean Leray.
quand on suppose que les U (x) convergent faiblement en moyenne vers U (x) sur 1 0 , les
V (x) vers V (x) sur 2 0 et que lintegral
ZZZ ZZZ
A2 (x) x y
1 0
2 0
est fini.
On a:
ZZZ
(1.8)
ZZZ
lim
quand on suppose, sur 0 , A(x) borne , U (x) forte limite des U (x) et V (x) faible limite
des V (x).
Il est dautre part evident que lon a, si les fonctions U (x) convergent faiblement en
moyenne vers U (x) sur un domaine 1 0 :
ZZZ
ZZZ
lim{
ZZZ
[U (x) U (x)] x
0
U 2 (x) x} = 0;
U (x) x +
0
do`
u resultent linegalite:
ZZZ
(1.9)
ZZZ
2
lim inf
U 2 (x) x,
U (x) x
0
ZZZ
2
lim sup
U 2 (x) x.
U (x) x
0
De meme: Les composantes Ui (x) dun vecteur convergent fortement en moyenne sur
le domaine 0 vers celles dun vecteur Ui (x) quand elles convergent faiblement en moyenne
vers ces composantes sur ce domaine et quen outre1 :
Pi=3
1
Rappelons que le symbole Ui (x)Ui (x) represent lexpression
i=1 Ui (x)Ui (x).
200
Jean Leray.
when the U (x) converge weakly in mean to U (x) on 1 0 and V (x) to V (x) on 2 0 and
the integral
ZZZ ZZZ
A2 (x) x y
1 0
2 0
ZZZ
lim
when on 0 , A(x) is bounded, U (x) is the strong limit of the U (x) and V (x) is the weak
limit of the V (x).
It is also evident that, if the functions U (x) converge weakly in mean to U (x) on a
domain 1 0
ZZZ
ZZZ
2
lim{
[U (x) U (x)] x
0
ZZZ
2
U 2 (x) x} = 0
U (x) x +
ZZZ
2
lim inf
U 2 (x) x
U (x) x
0
ZZZ
2
lim sup
U 2 (x) x.
U (x) x
0
Equivalently, the components Ui (x) of the vector converge strongly in mean on domain
to those of a vector Ui (x) when they converge weakly in mean to the components on
the domain and in addition1
Pi=3
1
Recall that the symbol Ui (x)Ui (x) represents the expression
i=1 Ui (x)Ui (x).
0
ZZZ
0
(1.10 )
201
ZZZ
lim sup
0
Ui (x)Ui (x) x
Ui (x)Ui (x) x.
elements de cette premi`ere suite une seconde suite n21 , n22 , n23 . . ., telle que les quantites bn2 , bn2 ,
1
2
bn2 , . . .
3
ZZZ
0
(1.10 )
201
ZZZ
lim sup
0
Ui (x)Ui (x) x
Ui (x)Ui (x) x.
The weak convergence criteria applied in Example III gives the following.
Lemma 1. If an infinity of functions U (x) [or vectors Ui (x)] converge almost everywhere on domain 0 to a function U (x) [or a vector Ui (x)] and satisfy inequality (1.10)
[or (1.10)], then they converge strongly in mean.
Theorem of F. Riesz: An infinity of functions U (x) have a weak limit in mean on
domain 0 if the two following conditions are satisfied:
RRR
a) the set of numbers 0 U 2 (x) x is bounded;
RRR
U (x)A(x) x
b) for each square summable function A(x) on 0 the quantities
0
have a single limiting value.
Condition b) may be replaced by the following:
b) forRRR
each cube c with sides parallel to the coordinate axes and rational vertices, the
quantities c U (x) x have a single limiting value.
The proof of this theorem makes use of the work of Lebesgue on summable functions.
the first sequence a second sequence n21 , n22 , n23 . . ., such that the quantities bn2 , bn2 , bn2 , . . .
1
202
Jean Leray.
Lintegral
U 2 (x, t0 ) x est donc une fonction continue de t. Inversement le lemme I
nous apprend quune fonction U (x, t) continue par rapport aux variables x1 , x2 , x3 , t est
fortement continue en t quand lintegral precedent est une fonction continue de t.
202
Jean Leray.
converge to a limit; etc. One then chooses mp equal to npp , which is the p-th term of the diagonal
j
Application: The following results from the theorem of the preceeding paragraph.
Fundamental theoremRRRof F. Riesz Let an infinity of functions U (x) on a domain 0 be
such that the quantities
U 2 (x) x are bounded. Then one can always extract from
0
them a sequence of functions which have a weak limit in mean.
In fact condition a) is satisfied and Cantors diagonal process allows construction of a sequence
of functions U (x) which satisfy condition b).
implies
ZZZ
[U (x, t) U (x, t0 )]2 x < .
RRR
203
nous avons:
ZZZ
(r0 ) =
S
a(y) u(y)
u(y)
+
a(y) y.
yi
yi
La second expression de (r0 ) prouve que cette quantite tend vers une limite () quand
r0 augmente indefiniment. La premi`ere expression de (r0 ) nous donne:
ZZ
xi xi
|(r0 )|
|u(x)a(x)|
r0
s
do`
u:
Z
ZZZ
|(r0 )| dr0
0
|u(x)a(x)| x.
a(y) u(y)
+
a(y) y = 0;
u(y)
yi
yi
ZZZ
(1.12)
$
ZZ
a(y) u(y)
u(y)
+
a(y) y =
u(y)a(y) yi .
yi
yi
Choisissons comme domaine $ une sph`ere de rayon infiniment petit dont nous nom1
1 ( r )
merons le centre x; faisons1 dans (1.12) a(y) = 4
yi ; ajoutons les
1
203
We have
ZZZ
(r0 ) =
S
a(y) u(y)
u(y)
+
a(y) y.
yi
yi
The second expression shows that (r0 ) tends to a limit () when r0 increases indefinitely. The first expression for (r0 ) gives us
ZZ
xi xi
|u(x)a(x)|
|(r0 )|
r0
s
from which
Z
ZZZ
|(r0 )| dr0
0
|u(x)a(x)| x.
a(y) u(y)
u(y)
+
a(y) y = 0
yi
yi
ZZZ
(1.12)
$
ZZ
a(y) u(y)
u(y)
+
a(y) y =
u(y)a(y) yi .
yi
yi
Choose as domain $ a sphere of infinitely small radius and center x and take1 in
1
1 ( r )
(1.12) a(y) = 4
yi ; add the
1
204
Jean Leray.
(1.13)
ZZZ
( 1r ) u
y.
yi yi
i
Si maintenant nous faison dans (1.11) a(y) = yirx
u(y) et si nous ajoutons les relations
2
qui correspondent aux valeurs 1, 2, 3 de i, il vient:
ZZZ
ZZZ
1 2
yi xi u
u(y) y =
u (y) y;
2
2
2
r
yi
r
1 2
u (y) y 4
r2
ZZZ
u u
y.
yi yi
7. Quasi-derivees.
Soit une infinite de fonctions u (x) possedant des derivees premi`eres continues qui
soient, comme ces fonctions elles-memes, de carres sommable sur . Supposon que les
u u
derivees u
x1 , x2 , x3 convergent faiblement en moyenne sur vers des fonctions U,1 ,
U,2 , U,3 . Soit U (x) la fonction mesurable definie presque partout par la relation:
1
U (x) =
4
ZZZ
( 1r )
U,i (y) y.
yi
Nous avons:
(1.15)
ZZZ
ZZZ ZZZ
[u (x)U (x)] x =
$
u
Kij (y, y )
U,i (y)
yi
en posant1 :
1
u
0
U,j (y ) y y 0
yj 0
204
Jean Leray.
(1.13)
ZZZ
( 1r ) u
y.
yi yi
i
We now take a(y) = yirx
u(y) in (1.11) and add these relations for values 1, 2, 3 of
2
i, giving
ZZZ
ZZZ
1 2
yi xi u
u(y) y =
u (y) y.
2
2
2
r
yi
r
By applying the Schwarz inequality to the left side we get the useful inequality
ZZZ
(1.14)
1 2
u (y) y 4
r2
ZZZ
u u
y.
yi yi
7. Quasi-derivatives.
Let u (x) be an infinity of square summable functions with continuous square summ
u u
able first derivatives on . Suppose that the derivatives u
x1 , x2 , x3 converge weakly in
mean on to functions U,1 , U,2 , U,3 . Let U (x) be the measurable function defined almost
everywhere by
1
U (x) =
4
ZZZ
( 1r )
U,i (y) y.
yi
We have
(1.15)
ZZZ
ZZZ ZZZ
[u (x)U (x)] x =
$
u
Kij (y, y )
U,i (y)
yi
where1
1
u
0
U,j (y ) y y 0
0
yj
1
Kij (y, y ) =
16 2
ZZZ
205
( 1r ) ( r10 )
x;
yi yj 0
est finie; le seconde membre de (1.15) a donc bien un sens; et il tend vers zero dapr`es
la relation (1.7). Donc les fonctions u (x) ont surRRRtout domaine $ une forte limite en
moyenne: la fonction U (x). Et, si les integrals
U 2 (x) x sont bornees dans leur
ensemble, U (x) est sur faible limite en moyenne des fonctions u (x) (Cf. 3. Exemple
II, p. 199); on deduit alors de (I.11) legalite:
ZZZ
(1.16)
a
U (y)
+ U,i (y)a(y) y = 0.
yi
fonctions u (x) convergent en moyenne vers une fonction U (x), dont les fonctions U,i (x)
sont des quasi-derivees; cette convergence est forte sure tout domaine $; elle est faible1
sur .
1
Ou forte.
1
Kij (y, y ) =
16 2
ZZZ
205
( 1r ) ( r10 )
x.
yi yj 0
is finite, so the right side of (1.15) is defined. It tends to zero by (1.7). Therefore
u (x)
RRR the
have U (x) as strong limit in mean on all domains $. And, if the integrals U 2 (x) x
are bounded, U (x) is the weak limit in mean of the u (x) on (Cf. 3. Example II, p.
199). One then gets from (1.11) the equality
ZZZ
(1.16)
a
U (y)
+ U,i (y)a(y) y = 0.
yi
u (x)
derivatives xi has a weak limit in mean U,i (x) on . Then the u (x) converge in
mean to a function U (x) for which the U,i (x) are the quasi-derivatives. This convergence
is strong on all domains $. It is weak1 on .
1
Or strong.
206
Jean Leray.
De meme que nous avons defini les quasi-derivees, nous allons definir comme suit la
quasi-divergence (x) dun vecteur Ui (x) dont les composants sont de carres sommables
sur : cest, quand elle existe, une fonction de carre sommable verifiant la relation:
ZZZ
(1.17)
a
Ui (y)
+ (y)a(y) y = 0.
yi
8. Approximation dune fonction mesurable par une suite de fonctions reguli`eres. Soit
une quantite positive arbitraire . Choisissons1 une fonction (s) continue, positive, definie
pour 0 s, identique `a zero pour 1 s, possedant des derivees de tous les ordres et telle
que:
Z
( 2 ) 2 d = 1.
U (x) etant une fonction sommable sur tout domaine $, nous poserons
1
U (x) = 3
(1.18)
ZZZ
r2
2
U (y) y
U (x)
= 3
U (y) y.
n
n
m
m
l
l
x1 x2 x3
x1 x2 x3
l+m+n
(1.19)
(1.20)
Suppossons U (x) de carre sommable sur ; linegalite (1.3) appliquee `a (1.18) nous donne:
ZZZ
ZZZ
U (x)2
(1.21)
U 2 (x) x;
x <
Pour fixer les idees nous prendrons (s) = Ae s1 , A etant une constante convenable, pour
0 < s < 1.
206
Jean Leray.
a
Ui (y)
+ (y)a(y) y = 0.
yi
( 2 ) 2 d = 1.
(1.18)
1
U (x) = 3
ZZZ
r2
2
U (y) y
(1.19)
2
l+m+n
Z
Z
Z
r2
l+m+n U (x)
1
= 3
U (y) y.
n
l
m
n
xl1 xm
x1 x2 x3
2 x3
(1.20)
(1.21)
U 2 (x) x.
U (x) x
ZZZ
207
appliquee `a (1.19) elle prouve que les derivees partielles de U (x) sont de carres sommables
sur .
Notons enfin que nous avons, si U (x) et V (x) sont de carres sommables sur :
ZZZ
(1.22)
ZZZ
U (x)V (x) x =
U (x)V (x) x.
Si V (x) est continue, V (x) tend uniformement vers V (x) sur tout domaine $ quand
tend vers zero; on a alors dapr`es (1.22):
ZZZ
ZZZ
U (x)V (x) x;
lim
U (x)V (x) x =
on en deduit que sur U (x) convergent faiblement en moyenne vers U (x) quand tend
vers zero; linegalite (1.21) et le crit`ere de forte convergence enonce p. 200 autorisent meme
une conclusion plus precise:
Lemme 3. Soit une fonction U (x) de carres sommables sur ; U (x) converge sur
fortement en moyenne vers U (x) quand tend vers zero.
On etablit de meme la proposition suivante:
Generalization du lemme 3. Soit une suite de fonctions U (x) qui sur convergent
fortement (ou faiblement) vers une limite U (x) quand tend vers zero; les fonctions U (x)
convergent fortement (ou faiblement) ver cette meme limite.
9. Quelques lemmes concernant les quasi-derivees.
Soit une fonction U (x) de carre sommable sur ; supposons
ZZZ
U (x)a(x) x = 0
quelle que soit la fonction a(x) de carre sommable sur dont les derivees de tous les ordres
existent et sont de carres sommables sur ; nous avons alors:
ZZZ
U (x)U (x) x = 0
207
The same applied to (1.19) proves that the partial derivatives of U (x) are square summable
on .
Finally note that we have, if U (x) and V (x) are square summable on
ZZZ
(1.22)
ZZZ
U (x)V (x) x =
U (x)V (x) x.
If V (x) is continuous V (x) tends uniformly to V (x) on all domains $ when tends
to zero. One therefore has from (1.22)
ZZZ
ZZZ
U (x)V (x) x.
lim
U (x)V (x) x =
From this one deduces that U (x) converges weakly in mean to U (x) on when approaches
zero. Inequality (1.21) and the criteria for strong convergence on p. 200 similarly give a
more precise conclusion:
Lemma 3. Let U (x) be square summable on . U (x) converges strongly in mean to
U (x) on when tends to zero.
Similarly one establishes the following proposition.
Generalization of lemma 3. Suppose a sequence of functions U (x) converge strongly
(or weakly) in mean on to a limit U (x) as tends to zero. Then the functions U (x)
converge strongly (or weakly) to the same limit.
9. Some lemmas on quasi-derivatives.
Let U (x) be square summable on . Suppose that for all square summable functions
a(x) having square summable derivatives of all orders
ZZZ
U (x)a(x) x = 0
then
ZZZ
U (x)U (x) x = 0
208
Jean Leray.
do`
u, en faisant tendre vers zero:
ZZZ
U 2 (x) x = 0.
= U,i (x).
ZZZ
U (x)
a(x) x =
xi
ZZZ
U,i (x)a(x) x.
a(x)
a(x)
=
;
xi
xi
cette formule et les formules (1.11), (1.16), (1.22) justifient les transformations:
ZZZ
U (x)
a(x) x =
xi
ZZZ
a(x)
U (x)
x =
xi
ZZZ
a(x)
U (x)
x =
xi
ZZZ
ZZZ
a(x)
U (x)
x =
xi
ZZZ
U,i (x)a(x) x =
U,i (x)a(x) x.
C.Q.F.D.
Lemme 5. Soient deux fonctions de carres sommables sur , U (x) et V (x), qui
poss`edent les quasi-derivees U,i (x) et V,i (x); je dis que:
ZZZ
(1.23)
ZZZ
[U (x)V,i (x) + U,i (x)V (x)] x = 0.
208
Jean Leray.
U (x)
xi
= U,i (x).
ZZZ
U (x)
a(x) x =
xi
ZZZ
U,i (x)a(x) x.
a(x)
a(x)
=
xi
xi
and this formula with (1.11), (1.16), and (1.22) justify the transformations
ZZZ
U (x)
a(x) x =
xi
ZZZ
ZZZ
a(x)
x =
U (x)
xi
a(x)
U (x)
x =
xi
ZZZ
ZZZ
a(x)
U (x)
x =
xi
ZZZ
U,i (x)a(x) x =
U,i (x)a(x) x.
Q.E.D.
Lemma 5. Suppose that two square summable functions U (x) and V (x) have quasiderivatives Ui (x) and Vi (x) on . I claim that
ZZZ
(1.23)
ZZZ
[U (x)V,i (x) + U,i (x)V (x)] x = 0.
209
Lemme 6. Soit un vecteur Ui (x) admettant une quasi-divergence (x); on a: divergence Ui (x) = (x).
(La demonstration de ce lemme est tr`es analogue `
a celle du lemme 4).
quel que soit le vecteur ai (x), de divergence nulle, dont les composantes ainsi que leurs
derivees de tous les ordres sont de carres sommables sur . Je dis que Ui (x) = 0.
En RRR
effet le lemme 4 nous autorise `
a choisir ai (x) = Ui (x); or quand tend vers zero la
relation
U
(x)U
(x)
x
=
0
se
r
e
duit
`
a la suivante:
i
i
ZZZ
Ui (x)Ui (x) x = 0.
derivees de tous les oedres, sont de carres sommables sur , les quantites Ui (x)ai (x) x
ont une seule valeur limite.
Sinon le Theorem fondamental de M. F. Riesz (p. 202) permettrait dextraire de la suite
Ui (x) deus suites partielles possedant deux faibles limites distinctes, dont lexistence contredireit
le lemme 7.
(2.1)
ui (x, t) 1 p(x, t)
2
ui (x, t)
= Xi (x, t) =
t
xi
xk xk
uj (x, t)
= 0.
xj
209
Lemma 6. If a vector Ui (x) has quasi-divergence (x) then the divergence of Ui (x) =
(x).
(The proof is very much analogous to that of lemma 4.)
for all square summable vectors ai (x) which have 0 divergence and square summable derivatives of all orders on . Then I claim that Ui (x) = 0.
RRR In fact lemma 4 allows us to choose ai (x) = Ui (x) because when tends to 0 the relation
U (x)Ui (x) x = 0 reduces to
i
ZZZ
Ui (x)Ui (x) x = 0.
Ui (x)
(2.1)
ui (x, t) 1 p(x, t)
2
ui (x, t)
= Xi (x, t) =
t
xi
xk xk
uj (x, t)
= 0.
xj
210
Jean Leray.
et sont les constantes donnees, Xi (x, t) est un vecteur donne qui represente les forces
exterieures; p(x, t) represente la pression, ui (x, t) la vitesse des molecules du liquide.
Le probleme que pose la theorie des liquides visqueux est le suivant:
Construire pour t > 0 la solution de (2.1) qui correspond `a des valeurs initiales
donnees, ui (x, 0).
Nous allons rappeler la solution de ce probl`eme et quelques-unes des proprietes quelle
poss`ede. Nous posserons:
ZZZ
W (t) =
ui (x, t)ui (x, t) x
ZZZ
2
Jm
(t) =
m ui (x, t) m ui (x, t)
x.
xk xl . . . xk xl . . .
p
V (t) = Maximum de ui (x, t)ui (x,
`a linstant
t.
t)
m ui (x,t)
Dm (t) = Maximum des fonctions xh xk xl `a linstant t (h + k + l = m).
1
Nous ferons les hypoth`eses suivantes relativement aux donnees: Les fonctions ui (x, t)
uj (x,0)
et leurs derivees premi`eres sont continues;
= 0; les quantites W (0) et V (0)
xj
1
2
sont
RRR finies; |Xi (x, t) Xi (y, t)| < r C(x, y, t), C(x, y, t) etant une fonction continue;
X (x, t)Xi (x, t) x est une fonction continue de t, ou est inferieure `a une fonction
i
continue de t.
Les lettres A et Am nous serviront desormais `a designer les constantes et les fonctions
de lindice m dont nous ne preciserons pas les valeurs numeriques.
(2.2)
u0i (x, t)
1
= 3
(2 )
ZZZ
r2
e 4t
(t)
3
2
ui (y, 0) y;
p0 (x, t) = 0.
Les integrales u0i (x, t) sont uniformement continues en t (cf. 5, p. 202) pour 0 < t,
et lon a:
210
Jean Leray.
and are given constants, Xi (x, t) is a vector which represents external forces, p(x, t) is
the pressure, and ui (x, t) the speed of the molecules of the liquid.
The problem posed by the theory of viscous liquids is the following: Construct for
t > 0 the solution of (2.1) which has given initial values ui (x, 0).
We recall the solution of this problem and some of its properties. Write
ZZZ
W (t) =
ui (x, t)ui (x, t) x
ZZZ
2
Jm
(t) =
m ui (x, t) m ui (x, t)
x.
xk xl . . . xk xl . . .
p
V (t) = Maximum of ui (x, t)ui (x, t)
time t.
at
m ui (x,t)
Dm (t) = Maximum of the function xh xk xl at time t (h + k + l = m).
1
We make the following assumptions: The functions ui (x, t) and their first derivatives
uj (x,0)
are continuous, x
= 0, the quantities W (0) and V (0) are finite, |Xi (x, t)Xi (y, t)| <
j
RRR
1
r 2 C(x, y, t), where C(x, y, t) is a continuous function, and Xi (x, t)Xi (x, t) x is a continuous function of t, or is less than a continuous function of t.
From now on the letters A and Am denote constants and functions with index m for
which we do not specify the numerical value.
11. First case: Xi (x, t) = 0.
The theory of heat gives the following solution to system (2.1):
(2.2)
u0i (x, t)
1
= 3
(2 )
ZZZ
r2
e 4t
(t)
3
2
ui (y, 0) y;
p0 (x, t) = 0.
The integrals u0i (x, t) are uniformly continuous in t (cf. 5, p. 202) for 0 < t, and from
this one has
[translators note: where r = |y x|.
(2.3)
211
Quand J1 (0) est fini, lappliction de linegalite (1.14) et de linegalite de Schwarz (1.1)
a (2.2) permet dobtenir une seconde borne de V (t):
`
2
V (t) <
4J12 (0)
1
(4)3
ZZZ
r2
e 2t 2
r y,
(t)3
cest-`
a-dire:
(2.4)
V (t) <
AJ1 (0)
1
(t) 4
les integrales u0i (x, t) sont fortement continues en t (cf. 5, p. 202) meme pour t = 0.
Appliquee `a la relation:
" r2 #
ZZZ
u0i (x, t)
1
e 4t
ui (y, 0) y
= 3
3
xk
(2 )
(t) 2
xk
cette inegalite (1.3) prouve que:
(2.6)
les derivees premi`eres xk sont fortement continues en t, meme pour t = 0 si J1 (0) est fini.
Pour les raisons analogues les derivees de tous les ordres des integrales u0i (x, t) sont
uniformement et fortement continues en t pour t > 0; et plus precisement:
Am
W (0)
(2.7)
Dm (t) <
(2.8)
p
Am W (0)
Jm (t) <
.
m
(t) 2
(t)
2m+3
4
(2.3)
211
If J1 (0) is finite, inequality (1.14) and the Schwarz inequality (1.1) applied to (2.2)
give a second bound on V (t):
2
V (t) <
4J12 (0)
1
(4)3
ZZZ
r2
e 2t 2
r y,
(t)3
which is to say
(2.4)
AJ1 (0)
V (t) <
(t) 4
the integrals
(1.3) applied to
ZZZ
xk
"
r2
e 4t
3
(t) 2
#
ui (y, 0) y
proves that
(2.6)
Dm (t) <
Am
W (0)
,
(t)
p
Am W (0)
Jm (t) <
.
m
(t) 2
(2.8)
2m+3
4
tjk =
(0)
1 E(r,t t)
jk 2
t(0) t
2
xj xk ,
1
r
Rr
0
E(, t
(0)
t) d, E(r, t
(0)
t) =
r2
4(t(0) t)
t(0) t
.]
212
Jean Leray.
(2.9)
Z
u00i (x, t)
ZZZ
dt
0
p (x, t) =
4 xj
ZZZ
00
1
Xj (y, t) y
r
Nous avons:
(2.10)
|Tij (x y, t t0 )| <
m
Tij (x y, t t0 )
<
xh xk xl
1
A
3
[r2 + (t t0 )] 2
Am
[r2
+ (t
t0 )]
m+3
2
Nous remarquerons en premier lieu que les integrals (1.2) et (1.3) appliquees en meme
temps que (2.10) `
a la formule:
(2.11)
u00i (x, t)
=
xk
ZZZ
dt
0
u00
i
xk
Tij (x y, t t0 )
Xj (y, t0 ) y
xk
dt0
sZ Z Z
p
(t t0 )
J1 (t) < A
0
dt0
D1 (t) < A
0
(t t0 )
max
cette inegalite (2.13) peut etre completee comme suit: nous avons
212
Jean Leray.
Z
(2.9)
u00i (x, t)
ZZZ
dt
0
p (x, t) =
4 xj
ZZZ
00
1
Xj (y, t) y
r
We have
(2.10)
|Tij (x y, t t0 )| <
m
Tij (x y, t t0 )
<
xh xk xl
1
A
[r2
+ (t t0 )] 2
Am
[r2
+ (t
t0 )]
m+3
2
We remark in the first place that integrals (1.2) and (1.3) applied with (2.10) to the
formula
(2.11)
u00i (x, t)
=
xk
Tij (x y, t t0 )
Xj (y, t0 ) y
xk
dt
0
u00
i
xk
(2.12)
ZZZ
dt0
sZ Z Z
p
(t t0 )
J1 (t) < A
0
pThis done, we add to previously stated hypotheses the assumption that the maximum
of Xi (x, t0 )Xi (x, t0 ) at time t is a continuous function of t, or is less than a continuous
u00
function of t. Then there is no difficulty in deducing from (2.9) that u00i (x, t) and xik are
uniformly continuous in t for t 0, and more precisely for example
Z
(2.13)
dt0
D1 (t) < A
0
(t
t0 )
max
213
Z t
ZZZ
u00i (x, t) u00i (y, t)
Tij (x z, t t0 )
0
=
dt
Xj (z, t0 ) z
xk
yk
xk
0
$
Z t
ZZZ
Tij (y z, t t0 )
dt0
Xj (z, t0 ) z
y
k
0
$
Z
+
ZZZ
dt
0
0
$
Tij (x z, t t0 ) Tij (y z, t t0 )
Xj (z, t0 ) z,
xk
yk
$ etant le domaine des points z situes `a une distance de x ou de y inferieure `a 2r; appliquons
la formule des accroissements finis au crochet:
Tij (x z, t t0 ) Tij (y z, t t0 )
xk
yk
et majorons les trois integrales precedentes en remplacant les diverses fonctions qui y
figurent par des majorantes de leurs valeurs absolues; nous verifions aisement que:
00
xk yk <
(2.14)
Z
< Ar
1
2
dt0
[(t
t0 )]
3
4
max
Nous dirons quune fonction U (x, t) satisfait une condition H quand elle verifie une
inegalite analogue `a la precedente:
(2.15)
o
u C(t) est inferieur `a une fonction continue de t. Nous nommerons coefficient de la
condition H celle des fonctions C(t) dont les valeurs sont les plus faibles possibles.
Supposons maintenant que les fonctions Xi (x, t) satisfassent une telle condition H, de
2 u00 (x,t)
coefficient C(t); les derivees secondes xik xl , qui sont donnees par les formules:
213
Z t
ZZZ
u00i (x, t) u00i (y, t)
Tij (x z, t t0 )
0
=
dt
Xj (z, t0 ) z
xk
yk
xk
0
$
Z t
ZZZ
Tij (y z, t t0 )
dt0
Xj (z, t0 ) z
y
k
0
$
Z
+
ZZZ
dt
0
0
$
Tij (x z, t t0 ) Tij (y z, t t0 )
Xj (z, t0 ) z,
xk
yk
$ being the domain of points at distance less than 2r to x or y. We apply the formula of
finite differences to the bracket
Tij (x z, t t0 ) Tij (y z, t t0 )
xk
yk
and majorize the preceeding three integrals by replacing the various functions there by the
majorants of their absolute values. We easily verify
00
xk yk <
(2.14)
Z
< Ar
1
2
dt0
[(t
t0 )]
3
4
max
(2.15)
where C(t) is smaller than a continuous function of t. We call the weakest possible C(t),
the condition H coefficient.
Now suppose that the functions Xi (x, t) satisfy condition H with coefficient C(t).
2 u00 (x,t)
Then the second derivatives xik xl , given by the formulas
214
Jean Leray.
2 u00i (x, t)
=
xk xl
ZZZ
0
dt
0
2 Tij (x y, t t0 )
[Xj (y, t0 ) Xj (x, t0 )] y
xk xl
C(t0 ) dt0
D2 (t) < A
[(t t0 )] 4
Plus generalement:
Supposons que les derivees dordre m des fonctions Xi (x, t) par rapport `a x1 , x2 , x3
existent, soient continues et soient inferieures en valeur absolue `a une fonctions continue
m (t). Alors les derivees dordre m + 1, par rapport `a x1 , x2 , x3 , des fonctions u00i (x, t)
existent, sont uniformement continues en t; on a:
Z
(2.17)
(t0 ) dt0
pm
(t t0 )
m (t0 ) dt0
[(t t0 )] 4
Si de plus:
ZZZ
m Xi (x, t)
xh1 xk2 xl3
2
2
x < m
(t),
m (t) etant une fonction continue (positive), alors les derivees dordre m + 1 par rapport
a x1 , x2 , x3 des fonctions ui (x, t) sont fortement continues en t et verifient linegalite:
`
Z
(2.19)
(t0 ) dt0
pm
(t t0 )
Supposons maintenant que les derivees dordre m des fonctions Xi (x, t) par rapport `a x1 ,
x2 , x3 existent, soient inferieures en valeur absolue `a une fonction continue de t et verifient
une condition H de coefficient m (t). Alors les derivees dordre m + 2 des fonctions ui (x, t)
par rapport `a x1 , x2 , x3 existent, sont uniformement continues en t et verifient linegalite:
214
Jean Leray.
2 u00i (x, t)
=
xk xl
ZZZ
0
dt
0
2 Tij (x y, t t0 )
[Xj (y, t0 ) Xj (x, t0 )] y,
xk xl
C(t0 ) dt0
D2 (t) < A
[(t t0 )] 4
More generally:
Suppose the m-th order derivatives of the Xi (x, t) with respect to x1 , x2 , x3 exist, are
continuous, and are smaller than some continuous functions m (t). Then the derivatives
of order m + 1 of the u00i (x, t) with respect to x1 , x2 , x3 exist and are uniformly continuous
in t. We have
Z
(2.17)
(t0 ) dt0
pm
(t t0 )
m (t0 ) dt0
[(t t0 )] 4
If further
ZZZ
m Xi (x, t)
xh1 xk2 xl3
2
2
x < m
(t),
where m (t) is a (positive) continuous function, then the derivatives of order m + 1 with
respect to x1 , x2 , x3 of the ui (x, t) are strongly continuous in t and satisfy the inequality
Z
(2.19)
(t0 ) dt0
pm
(t t0 )
Now suppose that the derivatives of order m of the functions Xi (x, t) with respect to x1 ,
x2 , x3 exist, are smaller in absolute value than a continuous function of t, and satisfy
condition H with coefficient m (t). Then the derivatives of order m + 2 of the ui (x, t) with
respect to x1 , x2 , x3 exist, are uniformly continuous, and satisfy the inequality
Z
(2.20)
m (t0 ) dt0
3
[(t t0 )] 4
215
tivement comme (xi xi ) 2 , (xi xi )2 et (xi xi )1 ; et il suffit dintegrer les deux membres de
legalite:
ui ui
1
1 p
(ui ui ) ui
= ui Xi
2 t
xi
1
1
+ W (t) W (0) =
2
2
ZZZ
dt
0
do`
u resulte linegalite:
1
1
W (t) W (0) +
2
2
sZ Z Z
p
Xi (x, t0 )Xi (x, t0 ) x.
dt0 W (t0 )
cest-`
a-dire:
(2.22)
Z t sZ Z Z
W (t)
0
W (0).
Z
(2.20)
m (t0 ) dt0
3
[(t t0 )] 4
215
1 p
1
(ui ui ) ui
= ui Xi
2 t
xi
1
1
+ W (t) W (0) =
2
2
ZZZ
dt
0
sZ Z Z
p
dt0 W (t0 )
Xi (x, t0 )Xi (x, t0 ) x.
W (t) is therefore less than or equal to the solution (t) of the equation
sZ Z Z
Z t
p
1
1
(t) = W (0) +
dt0 (t0 )
Xi (x, t0 )Xi (x, t0 ) x
2
2
0
which is to say
(2.22)
Z t sZ Z Z
W (t)
0
W (0).
216
Jean Leray.
Quand Xi (x, t) nest pas nul hors dun domaine $, on peut approcher les fonctions
Xi (x, t) par une suite de fonctions Xi (x, t) nulles hors de domaines $ , et par ce procede
etablir ques les relations (2.21) et (2.22) sont encore valables. La relation (2.21) prouve
que W (t) est continue; les fonctions ui (x, t) sont donc fortement continues en t pour t 0.
14. ui (x, t) = u0i (x, t) + u00i (x, t) est la seule solution du probl`eme pose au parapraphe
10 pour laquelle W (t) est inferieure `a une fonction continue de t; cette proposition resulte
de la suivante
Theor`eme dunicite: Le syst`eme
(2.23)
ui (x, t)
ui (x, t) 1 p(x, t)
uj (x, t)
= 0;
=0
t
xi
xj
admet une seule solution, definie et continue pour t 0, nulle pour t = 0, telle que W (t)
soit inferieure `a une fonction continue de t; cest ui (x, t) = 0.
En effet les fonctions
vi (x, t) =
ui
(x, t0 ) dt0 ,
q(x, t) =
p(x, t0 ) dt0
m vi (x, t)
xh1 xk2 xl3
et
m+1 vi (x, t)
txh1 xk2 xl3
(vi ) = 0;
t
la Theorie de la Chaleur permet
RRR den deduire vi = 0. Dautre part les inegalitees (1.2) et
(1.21) prouvent que lintegral
v (x, t)vi (x, t) x est finie. Donc vi (x, t) = 0. Et par suite
i
ui (x, t) = 0.
vi
ui (x, t) 1 p(x, t)
Xik (x, t) uj (x, t)
=
;
= 0.
t
xi
xk
xj
2 Xik (x,t)
xj xl
216
Jean Leray.
When Xi (x, t) is not zero outside a domain $, one can approach the functions Xi (x, t)
by a sequence of functions Xi (x, t) zero outside domains $ , and establish by the preceeding that relations (2.21) and (2.22) still hold. Then (2.21) shows that W (t) is continuous.
The ui (x, t) are therefore strongly continuous in t for t 0.
14. ui (x, t) = u0i (x, t)+u00i (x, t) is the only solution to the problem posed in paragraph
10, for which W (t) is less than a continuous function of t. This proposition results from
the following
Uniqueness theorem The system
(2.23)
ui (x, t)
ui (x, t) 1 p(x, t)
uj (x, t)
= 0;
=0
t
xi
xj
has just one solution defined and continuous for t 0, zero for t = 0, such that W (t) is
less than a continuous function of t. This solution is ui (x, t) = 0.
In fact the functions
vi (x, t) =
ui
(x, t0 ) dt0 ,
q(x, t) =
p(x, t0 ) dt0
m vi (x, t)
xh1 xk2 xl3
and
m+1 vi (x, t)
txh1 xk2 xl3
exist and are continuous. One evidently has q = 0 and it follows that
(vi ) = 0.
t
The theory of heat
RRRallows us to deduce that vi = 0. Further, inequalities (1.2) and (1.21) show
that the integral
v (x, t)vi (x, t) x is finite. Therefore vi (x, t) = 0. And then ui (x, t) = 0.
i
vi
ui (x, t) 1 p(x, t)
Xik (x, t) uj (x, t)
=
;
= 0.
t
xi
xk
xj
2 Xik (x,t)
xj xl
ZZZ
217
ZZZ
Xik (x, t)Xik (x, t) x,
0
Tij (x y, t t0 )Xjk (y, t) y;
dt
xk t0
ZZZ
1
p(x, t) =
Xik (y, t) y; ( t0 < t < T ).
4 xk
r
III. Mouvements r
eguliers.
15. Definitions: Les mouvements des liquides visqueux sont regis par les equations
de Navier:
(3.1)
ui (x, t)
ui (x, t) 1 p(x, t)
ui (x, t) uk (x, t)
= uk (x, t)
;
= 0,
t
xi
xk
xk
o`
u r et sont des constantes, p la pression, ui les composantes de la vitesse. Nous poserons:
ZZZ
W (t) =
ui (x, t)ui (x, t) x,
V (t) = max
Nous dirons quune solution ui (x, t) de ce syst`eme est reguliere dans un intervalle
de temps1 < t < T si dans cet intervalle de temps les fonctions ui , la fonction p
2
p
ui
i
, x kuxi l , u
a x1 , x2 ,
correspondante et les derivees x
t , xi sont continues par rapport `
k
x3 , t et si en outre les fonctions W (t) et V (t) sont inferieures `a des fonctions de t continues
pour < t < T .
Nous utiliserons les conventions suivantes:
La fonction Dm (t) sera definie pour chaque valeur de t au voisinage de
1
Le cas o`
u T = + nest pas exclu.
ZZZ
217
ZZZ
Xik (x, t)Xik (x, t) x,
0
Tij (x y, t t0 )Xjk (y, t) y;
dt
xk t0
ZZZ
1
p(x, t) =
Xik (y, t) y; ( t0 < t < T ).
4 xk
r
III. Regular motions.
15. Definitions: Motions of viscous liquids are governed by Naviers equations
(3.1)
ui (x, t)
ui (x, t) 1 p(x, t)
ui (x, t) uk (x, t)
= uk (x, t)
;
= 0,
t
xi
xk
xk
where and are constants, p is the pressure, ui the components of the velocity. We set
ZZZ
W (t) =
ui (x, t)ui (x, t) x,
V (t) = max
We say that a solution ui (x, t) of this system is regular in an interval of time1 < t <
2
ui
T if in this interval the functions ui , the corresponding p and the derivatives x
, x kuxi l ,
k
p
ui
t , xi are continuous with respect to x1 , x2 , x3 , t and if in addition the functions W (t)
and V (t) are less than some continuous functions of t for < t < T .
We use the following conventions.
The function Dm (t) will be defined for each value of t in a neighborhood in
1
218
Jean Leray.
m
ui (x,t)
laquelle les derivees x
ement continues en t; elle sera egale
h xk xl existent et sont uniform
1
2
3
a la borne superieure de leurs valeurs absolues.
`
La fonction C0 (t) [ou Cm (t)] ser definie pour toutes les valeurs de t au voisinage
m ui (x,t)
erifient une meme condition
desquelles les fonctions ui (x, t) [ ou les derivees x
h xk xl ] v
3
1
2
H; elle en sera le coefficient.
Enfin la fonction Jm (t) sera definie pour chaque valeur de t au voisinage de Laquelle
m ui (x,t)
les derivees x
h xk xl existent et sont fortement continues en t; nous poserons:
3
2
1
ZZZ
m ui (x, t) m ui (x, t)
2
Jm (t) =
x.
xk xl . . . xk xl . . .
Le lemme 8 (p. 216) sapplique aux solutions reguli`eres du syst`eme (3.1) et nous
donne les relations:
1
ui (x, t) = 3
(2 )
(3.2)
xk
(3.3)
ZZZ
r2
e 4t
3
(t) 2
ui (y, t0 ) y+
ZZZ
0
dt
t0
2
p(x, t) =
4 xk xj
ZZZ
1
uk (y, t)uj (y, t) y; ( < t0 < t < T ).
r
218
Jean Leray.
m
ui (x,t)
which the derivatives x
h xk xl exist and are uniformly continuous in t; it will be the
1
2
3
upper bound of their absolute values.
The function C0 (t) [or Cm (t)] will be defined for all values of t in a neighborhood in
m ui (x,t)
which the functions ui (x, t) [ or the derivatives x
h xk xl ] satisfy the same condition H;
3
1
2
it will be the coefficient.
Finally the function Jm (t) will be defined for each value of t in a neighborhood in
m ui (x,t)
which the derivatives x
h xk xl exist and are strongly continuous in t. We set
3
2
1
ZZZ
m ui (x, t) m ui (x, t)
2
Jm (t) =
x.
xk xl . . . xk xl . . .
Lemma 8 (p. 216) applies to regular solutions to system (3.1) and gives us the
relations
1
ui (x, t) = 3
(2 )
(3.2)
xk
(3.3)
ZZZ
r2
e 4t
3
(t) 2
ui (y, t0 ) y+
ZZZ
dt
t0
2
p(x, t) =
4 xk xj
ZZZ
1
uk (y, t)uj (y, t) y; ( < t0 < t < T ).
r
Paragraphs 11 and 12 allow us to conclude from (3.2) that the functions ui (x, t) are
uniformly and strongly continuous in t for < t < T , C0 (t) is defined for < t < T and
we have [cf. (2.7) and (2.18)]
p
Z t
A W (t0 )
V 2 (t0 ) dt0
C0 (t) <
+A
.
0 3
(t t0 )
t0 [(t t )] 4
This result with (3.2) shows that D1 (t) exists for < t < T and gives the inequality [cf.
(2.7) and (2.16)]
p
Z t
A W (t0 )
V (t0 )C0 (t0 ) dt0
D1 (t) <
+
A
.
5
3
[(t t0 )] 4
[(t t0 )] 4
t0
translators note: Danny Goodman of Harvard has pointed out that there appears
to be a sign discrepancy here.
219
Am
p
W (t0 )
[(t t0 )]
m+3
2
+ Am
t0
P
+=m+1
[(t
D (t0 )D (t0 )
3
t0 )] 4
dt0 .
Lexistence de D1 (t), . . . , Dm+1 (t), C0 (t), . . . , Cm+1 (t) assure celle de Dm+2 (t) et lon
peut majorer cette derni`ere fonction `a laide des precedentes [cf. (2.7) et (2.20)].
Les fonctions Dm (t) et Cm (t) sont donc definies pour < t < T , quelque grand que
soit m.
Dautre part les paragraphes 11 et 12 permettent de deduire de (3.2) lexistence de
J1 (t) pour toutes ces valeurs de t; et nous avons [cf. (2.8 et (2.19)]:
p
Z t
A W (0)
W (t0 )D1 (t0 ) 0
p
J1 (t) <
dt .
1 + A
(t t0 )
[(t t0 )] 2
t0
Plus generalement lexistence de D1 (t), . . . , Dm (t), J1 (t), . . . , Jm1 (t) assure celle de Jm (t)
[cf. (2.8) et (2.19)].
Il nous est maintenant aise detablir par lintermediaire de (3.3) que la fonction p(x, t)
m p(x,t)
et ses derivees x
sont uniformement et fortement continues en t pour < t < T .
k xj ...
m+1
u
i
Dapr`es les equations de Navier il en est de meme pour les fonctions u
t , txk xj ... .
Plus generalement les equations (3.1) et (3.3) permettent de ramener letude de
derivees qui sont dordre n + 1 par rapport `a t `a letude des derivees qui sont dordre
n par rapport `a t. Et lon aboutit finalement au theoreme suivante:
Si les fonctions ui (x, t) constituent une solution des equations de Navier reguli`ere
pour < t < T , alors toutes leurs derivees partielles existent; ces derivees partielles
et les fonctions ui (x, t) elles-memes sont uniformement et fortement continues en t pour
< t < T.
219
We proceed by recurrence:
The existence of D1 (t), . . . , Dm+1 (t) guarantees that of Cm+1 (t) and one has [cf. (2.7)
and (2.18)]
Am
p
W (t0 )
[(t t0 )]
m+3
2
+ Am
P
+=m+1
[(t
t0
3
t0 )] 4
D (t0 )D (t0 )
dt0 .
The existence of D1 (t), . . . , Dm+1 (t), C0 (t), . . . , Cm+1 (t) guarantees that of Dm+2 (t)
and one can majorize this last function using the preceeding [cf. (2.7) and (2.20)].
The functions Dm (t) and Cm (t) are therefore defined for < t < T , however large m
may be.
Further, paragraphs 11 and 12 allow us to deduce from (3.2) the existence of J1 (t) for
all values of t and we have [cf. (2.8) and (2.19)]
p
Z t
A W (0)
W (t0 )D1 (t0 ) 0
p
J1 (t) <
+
A
dt .
1
(t t0 )
[(t t0 )] 2
t0
More generally the existence of D1 (t), . . . , Dm (t), J1 (t), . . . , Jm1 (t) guarantees that of
Jm (t) [cf. (2.8) and (2.19)].
It is now easy for us to establish by the intermediary (3.3) that p(x, t) and its deriva m p(x,t)
tives x
are uniformly and strongly continuous in t for < t < T . By Naviers
k xj ...
m+1
u
i
equations it is the same for the functions u
t , txk xj ... .
More generally, equations (3.1) and (3.3) allow us to reduce the study of the order
n + 1 derivatives with respect to t to the study of the order n derivatives with respect to
t. So finally we achieve the following theorem.
If the functions ui (x, t) are a regular solution of Naviers equations for < t < T , then
all their partial derivatives exist, and the derivatives as well as the ui (x, t) are uniformly
and strongly continuous in t for < t < T .
220
Jean Leray.
16. Le paragraphe precedent nous apprend plus: il nous apprend `a majorer les fonctions ui (x, t) et leurs derivees partielles de tous les ordres au moyen des seules fonctions
W (t) et V (t). Il en resulte:
Lemme 9. Soit une infinite de solutions des equations de Navier, ui (x, t), reguli`eres
dans un meme intervalle de temps (, T ). Supposons les diverses fonctions V (t) et W (t)
inferieures `a une meme fonction de t, continue dans (, T ). De cette infinite de solutions
on peut alors extraire une suite partielle telle que les fonctions ui (x, t) de cette suite et
chaqune de leurs derivees convergent respectivement vers certaines fonctions ui (x, t) et
vers leurs derivees. Chacune de ces convergences est uniforme sur tout domaine $ pour
+ < t < T ( > 0). Les fonctions ui (x, t) constituent une solution des equations
de Navier reguli`ere dans (, T ).
En effet le Proceede diagonal de Cantor (4, p. 201) permet dextraire une suite de fonctions
ui (x, t) telle que ces fonctions ui (x, t) et leurs derivees convergent pour tous les syst`emes rationnels de valeurs donnees `
a x1 , x2 , x3 , t. Cette suite partielle poss`ede les proprietes quenonce
le lemme.
17. La quantite W (t) et la quantite J1 (t) que desormais nous designerons pour
simplifier par J(t) sont liees par une relation importante; elle sobtient en remplacant
ui
dans (2.21) Xi par uk x
et en remarquant que:
k
ZZZ
ui (x, t0 )
1
ui (x, t )uk (x, t )
x =
xk
2
ZZZ
uk (x, t0 )
t0
1
1
J 2 (t0 ) dt0 + W (t) = W (t0 ).
2
2
Cette relation et les deus paragraphes ci-dessus prouvent que les fonctions W (t), V (t), et
J(t) jouent un role essentiel. Aussi retiendrons-nous de toutes les inegalites quon peut
deduire du chapitre II uniquement quelques-unes o`
u figurent ces trois fonctions, sans plus
nous occuper des quantites Cm (t), Dm (t), . . .
Avant decrire ces quelques inegalites fondamentales posons une definition:
Une solution ui (x, t) des equations de Navier sera dite reguli`ere pour t < T
quand elle sera reguli`ere pour < t < T et quen outre les circonstances suivantes
220
Jean Leray.
16. The preceeding paragraph teaches us more: we learn that it is possible to bound
the functions ui (x, t) and their partial derivatives of all orders by means of just W (t) and
V (t). The result is:
Lemma 9. Let ui (x, t) be an infinity of solutions to Naviers equations, all regular in
the same interval (, T ). Suppose the various V (t) and W (t) all less than one function of
t, continuous in (, T ). Then one can extract a subsequence such that the ui (x, t) and each
of their derivatives converge respectively to certain functions ui (x, t) and their derivatives.
Each of the convergences is uniform on all domains $ for + < t < T ( > 0). The
functions ui (x, t) are a regular solution of Naviers equations in (, T ).
In fact, Cantors diagonal method (4, p. 201) allows the extraction of a sequence of functions
ui (x, t) which, with their derivatives, converge for any given rational values of x1 , x2 , x3 , t. This
subsequence has the properties stated in the lemma.
17. The quantities W (t) and J1 (t) which from now on we write simply as J(t)
ui
are linked by an important relation. It is obtained by replacing Xi in (2.21) by uk x
k
and remarking that
ZZZ
ui (x, t0 )
1
ui (x, t )uk (x, t )
x =
xk
2
ZZZ
uk (x, t0 )
t0
1
1
J 2 (t0 ) dt0 + W (t) = W (t0 ).
2
2
This relation and the two paragraphs above show that the functions W (t), V (t), and J(t)
play an essential role. We will especially point out, of all the inequalities one can deduce
from chapter II, some which involve these three functions without any longer occupying
ourselves with the quantities Cm (t), Dm (t), . . .
Before writing the fundamental inequalities, we make the definition:
A solution ui (x, t) of Naviers equations will be called regular for t < T when it
is regular for < t < T and if in addition the following conditions
221
i (x,t)
se presenteront: les fonctions ui (x, t) et ux
sont continues par rapport aux variables
j
x1 , x2 , x3 , t meme pour t = ; elles sont fortement continues en t meme pour t = ; les
fonctions ui (x, t) restent bornees quand t tend vers .
Dans ces conditions la relation (3.2) vaut pour t0 < t < T (la valeur etait jusq`a
present interdite `a t0 ); le chapitre II permet den deduire deux inegalites fondamentales;
ce sont, le symbole {B; C} nous servant `a representer la plus petite des deux quantites B
et C; A0 , A00 , A000 etant des constantes numeriques:
Z
(3.5)
V (t) < A
W (t0 )
V 2 (t0 )
A000 J(t0 )
0
;
{p
}
dt
+
{V
(t
);
0
1 }
(t t0 ) [(t t0 )]2
[(t t0 )] 4
t0
Z
(3.6)
J(t) < A
00
t0
J(t0 )V (t0 ) 0
p
dt + J(t0 )
(t t0 )
( t0 < t < T ).
vi
1 q
ui
vi
= vk
+ (uk + vk )
;
t
xi
xk
xk
vk
= 0.
xk
Posons:
ZZZ
w(t) =
ZZZ
2
j (t) =
vi (x, t) vi (x, t)
x.
xk
xk
Appliquons la relation (2.21) qui nous a dej`a fourni la relation fondamentale (3.4); elle
donne ici:
ZZZ
ZZZ
1 dw
ui
vi
2
j (t) +
=
vi vk
x +
vi (uk + vk )
x.
2 dt
xk
xk
Or nous avons:
ZZZ
1
vi
vi (uk + vk )
x =
xk
2
ZZZ
(uk + vk )
(vi vi )
x = 0;
xk
et
ZZZ
ui
x =
vi vk
xk
ZZZ
p
vi
vk ui x < j(t) w(t)V (t).
xk
221
i (x,t)
are satisfied: The functions ui (x, t) and ux
are continuous with respect to x1 , x2 , x3 ,
j
t also for t = , they are strongly continuous in t also for t = , and the ui (x, t) remain
bounded when t approaches .
In these conditions the relation (3.2) holds for t0 < t < T (the value was not
allowed to be t0 until now). Chapter II allows us to deduce two fundamental inequalities.
In these, the symbol {B; C} is the smaller of B and C, and A0 , A00 , A000 are numerical
constants. The inequalities are
Z
(3.5)
V (t) < A
W (t0 )
V 2 (t0 )
A000 J(t0 )
0
;
}
dt
{p
+
{V
(t
);
0
1 }
(t t0 ) [(t t0 )]2
[(t t0 )] 4
t0
Z
(3.6)
J(t) < A
00
t0
J(t0 )V (t0 ) 0
p
dt + J(t0 )
(t t0 )
( t0 < t < T ).
1 q
ui
vi
vi
= vk
+ (uk + vk )
;
t
xi
xk
xk
vk
= 0.
xk
Let
ZZZ
w(t) =
ZZZ
2
j (t) =
vi (x, t) vi (x, t)
x.
xk
xk
We apply (2.21) which has already given us the fundamental relation (3.4). Here it gives
ZZZ
ZZZ
1 dw
ui
vi
2
j (t) +
=
vi vk
x +
vi (uk + vk )
x.
2 dt
xk
xk
Now we have
ZZZ
vi
1
vi (uk + vk )
x =
xk
2
ZZZ
(uk + vk )
(vi vi )
x = 0
xk
and
ZZZ
ui
vi vk
x =
xk
ZZZ
p
vi
vk ui x < j(t) w(t)V (t).
xk
222
Jean Leray.
Donc:
p
1 dw
< j(t) w(t)V (t)
2 dt
j 2 (t) +
do`
u:
2
dw
< w(t)V 2 (t)
dt
et finalement:
(3.7)
1
2
Rt
t0
V 2 (t0 ) dt0
1
= 3
(2 )
ZZZ
r2
e 4t
3
(t) 2
ui (y, 0) y,
....................................
(n+1)
ui
(x, t)
=
xk
ZZZ
(n)
dt
0
(n)
....................................
Ecrivons en premier lieu les inegalites, deduites de (2.3) et (2.13):
V 0 (t) V (0)
Z
V
(n+1)
(0)
(t) A
222
Jean Leray.
Therefore
p
1 dw
< j(t) w(t)V (t)
2 dt
j 2 (t) +
from which
2
dw
< w(t)V 2 (t)
dt
and finally
(3.7)
1
2
Rt
t0
V 2 (t0 ) dt0
1
= 3
(2 )
ZZZ
r2
e 4t
3
(t) 2
ui (y, 0) y,
....................................
(n+1)
ui
(x, t)
=
xk
ZZZ
(n)
dt
0
(n)
....................................
First we write inequalities which follow from (2.3) and (2.13):
V 0 (t) V (0)
Z
V
(n+1)
(0)
(t) A
223
pour 0 t ,
si (t) est une fonctions continue qui verifie pour ces valeurs de t linegalite:
Z t
2 (t0 )
0
p
(t) A
dt0 + V (0);
0
(t t )
0
nous choisirons (t) = (1 + A)V (0); la valeur `a donner `a est:
= AV 2 (0).
(3.8)
Posons alors:
v
(n)
q
(n)
(n+1)
(n)
(n+1)
(x, t)][ui (x, t) ui
(x, t)]
(t) = max [ui (x, t) ui
a linstant t.
`
Nous avons:
(1)
Z
v
(n+1)
(t) < A
0
V 2 (0)
p
dt0 = AV (0);
0
( t )
0
Z
v (n) (t0 )
(t0 )v (n) (t0 ) 0
p
p
dt = AV (0)
dt0 ;
0
0
( t )
( t )
0
(t) < A
(n)
do`
u resulte que, pour 0 t , les fonctions ui (x, t) convergent uniformement vers des
limites continues, ui (x, t).
On demontre sans difficulte qu`a linterieur de lintervalle (0, ) chacune des derivees
(n)
des fonctions ui (x, t) converge uniformement vers la derivee correspondante des fonctions
ui (x, t); les raisonnements sont trop proches de ceux du paragraphe 15 pour que nous les
reproduisions. Les fonctions ui (x, t) satisfont donc les equations de Navier pour 0 < t < .
RRR
Verifions que lintegrale W (t) =
u (x, t)ui (x, t) x est inferieure `a une fonction
i
continue de t: les inegalites (2.5) et (2.12) fournissent les suivantes, o`
u A0 represente une
constante:
q
p
W (0) (t) W (0)
p
Z t
q
(t0 ) W (n) (t0 ) 0 p
(n+1)
p
W
(t) A0
dt + W (0);
(t t0 )
0
223
for 0 t ,
(3.8)
Then let
q
v
(n)
(n)
(n+1)
[ui (x, t) ui
(t) = max
at time t.
We have
(1)
(t) < A
0
(n+1)
(x, t)]
V 2 (0)
p
dt0 = AV (0)
0
( t )
0
Z
v (n) (t0 )
(t0 )v (n) (t0 ) 0
p
p
dt = AV (0)
dt0
0
0
( t )
( t )
0
(t) < A
Z
(n+1)
(n)
(n)
From this we get that the functions ui (x, t) converge uniformly to continuous limits
ui (x, t) for 0 t .
One shows without difficulty that in the interior of the interval, each of the derivatives
(n)
of the ui (x, t) converges uniformly to the corresponding derivative of the ui (x, t); the
reasoning is too close to that of paragraph 15 to repeat it. The functions ui (x, t) therefore
satisfy Naviers equations for 0 < t < .
RRR
We verify that the integral W (t) =
u (x, t)ui (x, t) x is less than a continuous
i
function of t. Inequalities (2.5) and (2.12) give the following, where A0 is a constant
q
p
W (0) (t) W (0)
p
Z t
q
0
(t
)
W (n) (t0 ) 0 p
p
W (n+1) (t) A0
dt + W (0).
(t t0 )
0
224
Jean Leray.
la theorie des equations lineaires nous apprend lexistence dune fontion positive (t) solution de lequation:
Z t
p
(t0 )(t)
p
(t) = A0
dt0 + W (0);
(t t0 )
0
nous avons W (n) (t) 2 (t); donc W (t) 2 (t).
Il nous reste `a preciser comment les fonctions ui (x, t) se comportent quand t tend
vers zero. Nous savons deja quelles se reduisent alors aux donnees ui (x, 0), en restant
continues meme pour t = 0. Pour prouver quelles demeurent fortement continues en t
quand t sannule, il suffit dapr`es le lemme 1 detablir que:
lim sup W (t) W (0);
t0
or cette inegalite a manifestement lieu, puisque 2 (0) = W (0). On prouve de meme que
i (x,t)
les fonctions ux
sont fortement continues en t, meme pour t = 0.
k
D`es lors la demonstration du theorem dexistence enonce ci-dessus est achevee.
Mais la formule (3.8) nous fournit un second resultat: Convenons de dire quune
solution des equations de Navier, reguli`ere dans un intervalle (, T ), devient irreguli`ere `
a
lepoque T quand T est fini et quil est impossible de definir cette solution reguli`ere dans
un intervalle (, T 0 ) plus grand que (, T ). La formule (3.8) rev`ele:
Un premier caract`ere des irregularites: Si une solution des equations de Navier devient
irreguli`ere `a lepoque T , alors V (t) augmente indefiniment quand t tend vers T ; et plus
precisement:
r
(3.9)
V (t) > A
.
T t
20. Il serait important de savoir sil existe des solutions des equations de Navier qui
deviennent irreguli`eres. Sil ne sen trouvait pas, la solution reguli`ere unique qui correspond
a un etat initial regulier, ui (x, 0), existerait pour toutes les valeurs positives de t.
`
Aucune solution ne pourrait devenir irreguli`ere si linegalite (3.9) etait incompatible
avec les relations fondamentales (3.4), (3.5) et (3.6); mais il nen est rien, comme en le voit
en choisissant:
224
Jean Leray.
V (t) > A
.
T t
20. It will be important to know whether there are solutions which become irregular.
If these cannot be found to exist, then the regular solution corresponding to a regular
initial state ui (x, 0) will exist for all positive values of t.
No solution can become irregular if inequality (3.9) is incompatible with the fundamental relations (3.4), (3.5) and (3.6), but this is not an issue as one sees by choosing
(3.10)
V (t) =
A00 [(T
12
t)]
; W (t) =
A000 [(T
225
p 00
A0
1
[(T t)] 4 ;
t)] ; J(t) =
2
1
2
et en verifiant que pour des valeurs suffisamment fortes des contantes A00 et A000 linegalite
(3.9) et la relation (3.4) sont verifiees, ainsi que les deux inegalites suivantes, qui sont plus
strictes que (3.5) et (3.6):
Z t
W (t0 )
V 2 (t0 )
A000 J(t0 )
0
0
;
V (t) < A
{p
}
dt
+
{V
(t
);
0
1 }
(T t0 ) [(T t0 )]2
[(t t0 )] 4
t0
Z t
J(t0 )V (t0 ) 0
00
p
J(t) < A
dt + J(t0 )
(t0 < t < T ).
(T t0 )
t0
Les equations de Navier poss`edent s
urement une solution qui devient irreguli`ere et
pour laquelle les fonctions W (t), V (t) et J(t) sont du type (3.10) si le syst`eme:
(3.11)
Ui (x)
Ui (x)
1 P (x)
Ui (x) Ui (x) + xk
= Uk (x)
;
xk
xi
xk
Uk (x)
= 0,
xk
o`
u designe une constante
RRR positive, poss`ede une solution non nulle, les Ui (x, t) etant
bornes et les integrales
U (x, t)Ui (x, t) x finies; la solution des equations de Navier
i
dont il sagit est:
(3.12)
(t < T )
225
p
(3.10)
V (t) =
A00 [(T
12
t)]
; W (t) =
A000 [(T
1
2
t)] ; J(t) =
A000
1
[(T t)] 4
2
and from this check that for all sufficiently large values of the constants A00 and A000 inequality (3.9) and relation (3.4) are satisfied, as well as the following two inequalities which
are stronger than (3.5) and (3.6)
Z t
W (t0 )
V 2 (t0 )
A000 J(t0 )
0
0
;
V (t) < A
{p
}
dt
+
{V
(t
);
0
1 }
(T t0 ) [(T t0 )]2
[(t t0 )] 4
t0
Z t
J(t0 )V (t0 ) 0
00
p
J(t) < A
dt + J(t0 )
(t0 < t < T ).
(T t0 )
t0
Naviers equations certainly have a solution which becomes irregular and for which
W (t), V (t) and J(t) are of the type (3.10) if the system
(3.11)
Ui (x)
Ui (x)
1 P (x)
Ui (x) Ui (x) + xk
= Uk (x)
;
xk
xi
xk
Uk (x)
= 0,
xk
where is
RRRa positive constant, has a nonzero solution with the Ui (x, t) bounded and the
integrals Ui (x, t)Ui (x, t) x finite. It is
(3.12)
(t < T )
226
Jean Leray.
Z
(3.13)
V (t) < A
V 2 (t0 )
W (t0 )
A000 J(t0 )
0
;
{p
}
dt
+
{V
(t
);
0
1 }
(t t0 ) [(t t0 )]2
[(t t0 )] 4
t0
( t0 < t < T );
supposons quune fonction (t), continue pour 0 < t , verifie pour ces valeurs de t
linegalite:
Z
(3.14)
(t) A
2 (t0 )
W (t0 )
A000 J(t0 )
0
{p
;
}
dt
+
{V
(t
);
0
1 },
(t t0 ) [(t t0 )]2
[(t t0 )] 4
nous avons alors pour les valeurs de t communes aux deux intervalles (t0 , T ) et (t0 , t0 + ):
(3.15)
V (t) < (t t0 );
t0 + < T.
00
(t) A
(t0 )(t0 ) 0
p
dt + J(t0 )
(t t0 )
(0 < t );
t0 < t t0 + .
et
= AV 2 (t0 ).
226
Jean Leray.
Z
(3.13)
V (t) < A
V 2 (t0 )
W (t0 )
A000 J(t0 )
0
{p
;
}
dt
+
{V
(t
);
0
1 }
(t t0 ) [(t t0 )]2
[(t t0 )] 4
t0
( t0 < t < T ).
We suppose there is a continuous function (t) in 0 < t , satisfying the inequality
Z
(3.14)
(t) A
2 (t0 )
W (t0 )
A000 J(t0 )
0
{p
;
}
dt
+
{V
(t
);
0
1 }.
(t t0 ) [(t t0 )]2
[(t t0 )] 4
We then have
(3.15)
V (t) < (t t0 )
for values of t common to the two intervals (t0 , T ) and (t0 , t0 + ). Then the first characterisation of irregularity implies
(3.16)
t0 + < T.
00
(t) A
(t0 )(t0 ) 0
p
dt + J(t0 )
(t t0 )
(0 < t ).
t0 < t t0 + .
and = AV 2 (t0 ).
227
certaine constante A soit `a linstant initial, soit `a tout autre instant anterieurement auquel
cette solution nest pas devenue irreguli`ere.
On peut satisfaire (3.14) et (3.17) par un choix du type
(3.19)
(t) = (1 + A)J(t0 );
= A 3 J 4 (t0 ).
J(t) >
A 4
1
(T t) 4
Les inegalites (3.15) et (3.19) prouvent quune solution reguli`ere `a un instant t reste
reguli`ere jusq`a linstant t0 + et que lon a:
V (t0 + ) < A 1 J 2 (t0 ).
La relation fondamentale (3.4) donne dautre part:
W (t0 + ) < W (t0 ).
Donc:
3 W (t0 + )V (t0 + ) < A 4 W (t0 )J 2 (t0 ).
Lapplication du premier cas de regularite `a lepoque t0 + fournit d`es lors:
Un second cas de regularite: On est assure quune solution reguli`ere ne devient
jamais irreguli`ere quand la quantite 4 W (t)J 2 (t) se trouve etre inferieure `a une certaine
constante A soit `a linstant initial, soit `a tout autre instant anterieurement auquel cette
solution nest pas devenue irreguli`ere.
22. On etablit de meme les resultats suivants dont les precedent peuvent dailleurs
etre consideres comme des cas particuliers:
Caract`ere des irregularites: Si une solution devient irreguli`ere `a lepoque T , on a:
1
3
ZZZ
A(1 p3 ) 2 (1+ p )
p
1
{
[ui (x, t)ui (x, t)] 2 x} p >
1
3
(T t) 2 (1 p )
Cas de regularite:
(p > 3).
227
certain constant A either initially or at any other instant at which the solution has not
become irregular.
One can satisfy (3.14) and (3.17) by a choice of the type
(3.19)
(t) = (1 + A)J(t0 );
= A 3 J 4 (t0 ).
This gives
A second characterisation of irregularities: If a solution of Naviers equations becomes
irregular at time T , then J(t) grows indefinitely as t tends to T ; and more precisely
3
J(t) >
A 4
1
(T t) 4
Inequalities (3.15) and (3.19) show that a solution regular at t remains regular until
t0 + and that one has
V (t0 + ) < A 1 J 2 (t0 ).
The fundamental relation (3.4) further gives
W (t0 + ) < W (t0 ).
Therefore
3 W (t0 + )V (t0 + ) < A 4 W (t0 )J 2 (t0 ).
An application of the first case of regularity to the time t0 + now gives
A second case of regularity: A regular solution never becomes irregular if
4 W (t)J 2 (t)
is less than a certain constant A either initially or at all other previous instants at which
the solution has not become irregular.
22.
One similarly establishes the following results, of which the preceeding are
particular cases.
Characterisation of irregularities: If a solution becomes irregular at time T , one has
ZZZ
p
2
1
p
A(1 p3 ) 2 (1+ p )
1
(T t) 2 (1 p )
(p > 3).
228
Jean Leray.
ZZZ
p3
[AW (t)]
p
3
[ui (x, t)ui (x, t)] 2 x} < A(1 )3 3(p2)
p
(p > 3).
Les cas de regularite que nous signalons montrent comment une solution reste toujours
reguli`ere quand son etat initial de vitesse est suffisamment voisin du repos. Plus generalement
considerons un etat de vitesse auquel correspnd une solution ne devenant jamais irreguli`ere; `
a tout
etat initial suffisamment voisin correspond une solution qui elle aussi ne devient jamais irreguli`ere.
La demonstration de ce fait utilise ceux des resultats du paragraphe 34 qui concernent lallure
dune solution des equations de Navier pour les grandes valeurs de t.
228
Jean Leray.
ZZZ
p3
[AW (t)]
p
3
[ui (x, t)ui (x, t)] 2 x} < A(1 )3 3(p2)
p
(p > 3).
The case of regularity which we are pointing out shows how a solution always remains regular
if its initial velocity state is sufficiently near rest. More generally, consider a velocity state to which
corresponds a solution which never becomes irregular. For all initial states sufficiently near there
corresponds a solution which also never becomes irregular. The proof makes use of those results
of paragraph 34 which concern behavior of solutions to Naviers equations for large values of t.
229
Posons:
ZZZ
ZZZ
W (0) =
Ui (x)Ui (x) x
et
J (0) =
Les fonctions Ui (x) constituent un etat initial regulier, comme le prouvent le lemme 6 et
le paragraphe 8 (p. 209 et 206); soit ui (x, t) la solution reguli`ere qui correspond `a letat
initial Ui (x); nous avons en vertu de lin egalite (1.21) et de la relation de dissipation de
lenergie (3.4):
W (t) < W (0).
(4.1)
Le lemme 4 nous apprend que
Ui (x)
xj
les relations (3.15), (3.18), et (3.19) permettent den deduire que sur un meme intervalle
(0, ) les diverses solutions ui (x, t) sont reguli`eres et verifient les inegalites:
(4.2)
(4.3)
Les inegalites (4.1) et (4.2) nous autorisent `a appliquer le lemme 9 (p. 220): dans la
formule de definition (1.18) de U (x) figure une longueur ; il est possible de la faire tendre
vers zero en sorte que pour 0 < t < les fonctions ui (x, t) et chacune de leurs derivees
convergnet respectivement vers certaines fonctions ui (x, t) et vers leurs derivees. Ces
fonctions ui (x, t) constituent une solution des equation de Navier reguli`ere pour 0 < t < ;
dapr`es (4.1) et (4.2) cette solution satisfait les trois inegalites:
(4.4)
W (t) W (0);
V (t) AJ(0)(t) 4 ;
J(t) (1 + A)J(0).
Rt
Lintegrale 0 V 2 (t0 ) dt0 est donc finie pour 0 < t < . Il nous reste `a preciser comment les
fonctions ui (x, t) se comportent quand t tend vers zero.
229
Let
ZZZ
ZZZ
W (0) =
Ui (x)Ui (x) x
et
J (0) =
The functions Ui (x) constitute a regular initial state, as shown by lemma 6 and paragraph
8 (p. 209 et 206). Let ui (x, t) be the regular solution which corresponds to the initial state
Ui (x). We have, in virtue of inequality (1.21) and the energy dissipation relation (3.4) that
W (t) < W (0).
(4.1)
Lemma 4 shows us that
Ui (x)
xj
Relations (3.15), (3.18), and (3.19) allow us to deduce from this that in some interval (0, )
the various solutions ui (x, t) are regular and satisfy inequalities
1
(4.2)
We have further
= A 3 J 4 (0).
(4.3)
Inequalities (4.1) and (4.2) let us apply lemma 9 (p. 220). There is a length in the
definition (1.18) of U (x). It is possible to make this tend to zero in such a way that for
0 < t < the functions ui (x, t) and all their derivatives converge respectively to certain
functions ui (x, t) and to their derivatives. These ui (x, t) are a regular solution to Naviers
equations for 0 < t < . By (4.1) and (4.2) this solution satisfies the three inequalities
(4.4)
W (t) W (0);
V (t) AJ(0)(t) 4 ;
J(t) (1 + A)J(0).
Rt
The integral 0 V 2 (t0 ) dt0 is therefore finite for 0 < t < . Now we must specify how the
ui (x, t) behave as t tends to zero.
230
Jean Leray.
Soit ai (x) un vecteur quelconque, de divergence nulle, dont les composantes, ainsi que
toutes leurs derivees, sont de carres sommables sur . Des equations de Navier resulte la
ralation:
ZZZ
ZZZ
ZZZ
dt
0
do`
u, en passant `a la limite:
ZZZ
ZZZ
Ui (x)ai (x) x+
ZZZ
0
Z
0
dt
ai (x)
x;
xk
ZZZ
dt
ZZZ
dt
0
ai (x)
x.
xk
ZZZ
tend vers
Ui (x)ai (x) x
quand t tend vers zero. Dans ces conditions ui (x, t) a une faible limite en moyenne unique,
qui est Ui (x) (cf, Corollaire du lemme 7, p. 209). Mais linegalite W (t) W (0) nous
permet dutiliser le crit`ere de forte convergence enonce p. 200; et nous constatons ainsi
que les fonctions ui (x, t) convergent fortement en moyenne vers les fonctions Ui (x) quand
t tend vers zero.
ui (x, t) est donc une solution semi-reguli`ere1 pour 0 t < et elle correspond `a letat
initial Ui (x).
25. On peut par des raisonnements analogues traiter les deux autres cas que signale
le theor`eme ci-dessous:
Theor`eme dexistence: Soit un vecteur Ui (x), de quasi-divergence nulle, dont
1
ui (x,t)
xj
230
Jean Leray.
Let ai (x) be any vector of divergence zero, for which the components as well as all
their derivatives are square summable on . From Naviers equations we get
ZZZ
ZZZ
ZZZ
dt
0
ZZZ
Ui (x)ai (x) x+
Z
0
dt
ZZZ
0
ai (x)
x.
xk
ZZZ
dt
ZZZ
dt
0
ai (x)
x.
xk
ZZZ
tends to
Ui (x)ai (x) x
when t tends to zero. In these conditions ui (x, t) has a unique weak limit in mean, which
is Ui (x) (cf. Corollary to lemma 7, p. 209). But the inequality W (t) W (0) allows us
to use the criteria for strong convergence announced on p. 200, and we also note that the
ui (x, t) converge strongly in mean to the Ui (x) as t tends to zero.
ui (x, t) is therefore a semi-regular solution1 for 0 t < and it corresponds to the
initial state Ui (x).
25. By analogous reasoning one can treat the two other cases pointed out in the
theorem below.
Existence theorem: Let the vector Ui (x) have quasi-divergence zero, with
1
ui (x,t)
xj
231
les composantes sont de carres sommables sur ; on peut affirmer que letat initial de
vitesses quil definit est semi-regulier:
a) quand les fonctions Ui (x) poss`edent des quasi-derivees de carres sojmmables sur ;
b) quand les fonctions Ui (x) sont
RRR bornees;
p
c) ou enfin quand lintegrale
[U (x)Ui (x)] 2 x est finie pour une valeur de p
i
superieure `a 3.
N. B. Ce theor`eme dexistence du paragraphe 19 nepuisent evidemment pas letude
de lallure que presente au voisinage de linstant initial la solution qui correspond `a un etat
initial donne.
V. Solutions turbulentes.
26. Soit un etat initial regulier ui (x, 0). Nous navons pas reussi `a prouver que la
solution reguli`ere des equations de Navier qui lui correspond est definie pour toutes les
valeurs de t posterieures `a linstant initial t = 0. Mais considerons le syst`eme:
(5.1)
ui (x, t)
ui (x, t) 1 p(x, t)
ui (x, t)
= uk (x, t)
;
t
xi
xk
uj (x, t)
= 0.
xj
Cest un syst`eme qui est tr`es voisin des equations de Navier quand la longeur1
est tr`es courte. Tout ce que nous avons dit au cours du chapitre III sur les equations de
Navier lui est applicable sans modification, hormis les considerations non concluantes du
paragraphe 20. Par l`a se trouve etablie toute une categorie de proprietes du syst`eme (5.1),
dans lesquelles ne figure pas la longueur . Dautre part linegalite de Schwarz (1.1) nous
donne:
3p
uk (x, t) < A0 2 W (t),
A0 etant une constante numerique. Cette nouvelle inegalite et la relation de dissipation
de lenergie (3.4) autorisent `a ecrire `a cote de linegalite (3.5) la suivante: si une solution
du syst`eme (5.1) est reguli`ere pour 0 t < T , alors:
1
Rappelons que cette longueur a ete introduite au 8 (p, 206), quand nous avons defini le
symbole U (x).
231
the components are square summable on . One can verify that the initial velocity state
which it defines is semi-regular
a) if the functions Ui (x) have square summable quasi-derivative on ;
b) if the functions Ui (x) are
RRR bounded;
p
c) or finally if the integral [Ui (x)Ui (x)] 2 x is finite for some value of p larger than
3.
N. B. This theorem and the existence theorem of paragraph 19 evidently do not allow
a study of the attraction which is presented in the neighborhood of the initial instant by
the solution which corresponds to a given initial state.
V. Turbulent solutions.
26. Let ui (x, 0) be a regular initial state. We have not succeeded in proving that
the corresponding regular solution to Naviers equations is defined for all values of t after
the initial instant t = 0. But consider the system
(5.1)
ui (x, t)
ui (x, t) 1 p(x, t)
ui (x, t)
= uk (x, t)
;
t
xi
xk
uj (x, t)
= 0.
xj
This system is very near Naviers equations when the length1 is very short. All we
have said in Chapter III on Naviers equations is applicable without modification, other
than the inconclusive considerations of paragraph 20. Thus we know many properties of
system (5.1) which are independent of . Further, the Schwarz inequality (1.1) gives us
3p
uk (x, t) < A0 2 W (t),
A0 being a numerical constant. This new inequality and the energy dissipation relation
(3.4) allows us to write the following beside inequality (3.5) if a solution to system (5.1) is
regular for 0 t < T , then
1
Recall this length was introduced in 8 (p, 206), when we defined the symbol U (x).
232
Jean Leray.
V (t) < A A0
32
W (0)
0
V (t0 ) dt0
p
+ V (0)
(t t0 )
(0 < t < T ).
De l`a resulte que sur tout intervalle de regularite (0, T ) V (t) reste inferieur `a la fonction
(t), continue pour 0 t, qui satisfait lequation integrale lineaire du type de Volterra:
Z t
p
(t0 ) dt0
0
32
p
W (0)
(t) = A A0
+ V (0);
(t t0 )
0
V (t) reste donc borne quand, T etant fini, t tend vers T ; ceci contredit le premier caract`ere
des irregularites (p. 224); en dautres termes lunique solution des equations (5.1) qui
correspond `
a un etat initial regulier donne est definie pour toutes les valeurs du temps
posterieures `
a linstant initial.
27. Etant
donne un mouvement qui satisfait les equations (5.1), nous aurons besoin de resultats concernant la repartition de son energie cinetique: 21 ui (x, t)ui (x, t). Ces
resultats devront etre independants1 de .
Soit deux longueurs constantes R1 et R2 (R1 < R2 ); introduisons la fonction f (x)
suivante:
f (x) = 0
f (x) =
r0 R1
R2 R1
pour
pour
r0 R1 ;
(r02 = xi xi )
R1 r0 R2 ;
f (x) = 1
pour
R2 r0 .
Un calcul analogue `a celui qui fournit la relation de dissipation de lenergie (2.21) nous
donne:
Z
ZZZ
0
dt
0
1
=
2
1
1
ui (x, t0 ) ui (x, t0 )
x +
f (x)
xk
xk
2
ZZZ
Z
f (x)ui (x, 0)ui (x, 0) x
ZZZ
f (x)ui (x, t)ui (x, t) x =
ZZZ
0
dt
0
f (x)
ui (x, t)
ui (x, t0 )
x+
xk
xk
Ils vaudront egalement pour les solutions reguli`eres des equations de Navier.
232
Jean Leray.
V (t) < A A0
32
W (0)
0
V (t0 ) dt0
p
+ V (0)
(t t0 )
(0 < t < T ).
From this we get that on all intervals of regularity (0, T ), V (t) remains less than the
continuous function (t) on 0 t, which satisfies the Volterra linear integral equation
Z t
p
(t0 ) dt0
0
32
p
W (0)
(t) = A A0
+ V (0).
(t t0 )
0
V (t) therefore remains bounded when, T being finite, t tends to T . That contradicts
the first caracterization of irregularity (p. 224). In other words, the unique solution to
equations (5.1) corresponding to a given regular initial state is defined for all time after
the initial instant.
27.
Given a motion which satisfies equations (5.1), we will need results on its
repartition of kinetic energy: 12 ui (x, t)ui (x, t). These must be independent1 of .
Consider two constant lengths R1 and R2 (R1 < R2 ) and introduce the following
function f (x)
f (x) = 0
f (x) =
r0 R1
R2 R1
for
for r0 R1 ;
(r02 = xi xi )
R1 r0 R2 ;
f (x) = 1
for R2 r0 .
A calculation analogous to that giving the energy dissipationn relation (2.21) here gives
Z
ZZZ
0
dt
0
1
=
2
1
ui (x, t0 ) ui (x, t0 )
1
f (x)
x +
xk
xk
2
ZZZ
Z
f (x)ui (x, 0)ui (x, 0) x
ZZZ
f (x)ui (x, t)ui (x, t) x =
ZZZ
0
dt
0
f (x)
ui (x, t)
ui (x, t0 )
x+
xk
xk
1
+
2
1
+
Z t
ZZZ
f (x)
p(x, t0 )ui (x, t0 ) x+
xi
dt
0
ZZZ
f (x)
uk (x, t0 )ui (x, t0 )ui (x, t0 ) x.
xk
dt
0
(5.2)
W (0)
+
R2 R1
s
ZZZ
dt
2
1
0
0
ui (x, t )ui (x, t ) x.
2
W (0)
t.
2
(5.4)
ZZZ
( 1r ) ui (y, t0 )
uk (y, t0 ) y,
xj
yk
do`
u
1
2
ZZZ
1
p (x, t ) x =
4
ZZZ ZZZ
uk (x, t0 )
ui (x, t0 ) 1
ui (y, t0 )
uj (y, t)
x y;
xk r
yj
i
en outre:
233
1
+
2
1
+
Z t
ZZZ
f (x)
p(x, t0 )ui (x, t0 ) x+
xi
dt
0
ZZZ
f (x)
uk (x, t0 )ui (x, t0 )ui (x, t0 ) x.
xk
dt
0
(5.2)
W (0)
+
R2 R1
s
ZZZ
dt
2
1
0
0
ui (x, t )ui (x, t ) x.
2
W (0)
t.
2
(5.4)
ZZZ
( 1r ) ui (y, t0 )
uk (y, t0 ) y,
xj
yk
from which
1
2
ZZZ
1
p (x, t ) x =
4
ZZZ ZZZ
uk (x, t0 )
ui (x, t0 ) 1
ui (y, t0 )
uj (y, t)
x y.
xk r
yj
i
Further
233
234
Jean Leray.
X Z Z Z
ui (x, t0 )
0
uk (x, t )
x
xk
2
< W (t0 )J 2 (t0 );
donc:
1
2
par suite:
p2 (x, t0 ) x <
1 p
W (t0 )J 3 (t0 );
2
(5.5)
ZZZ
sZ Z Z
dt
0
[W (0)] 4
p2 (x, t0 ) x <
2
W (0)
2(2)
3
4
t4 .
De (1.13) resulte:
1
1
ui (x, t0 )ui (x, t0 ) =
2
4
ZZZ
( 1r )
ui (y, t0 )
ui (y, t0 )
y;
xk
yk
cette formule analogue `a (5.4) conduit par des calculs analogues aux precedents `a linegalite:
Z
(5.6)
s
ZZZ
0
dt
0
1
ui (x, t0 )ui (x, t0 )
2
2
x <
W (0)
2(2)
3
4
t4 .
Tenons compte dans (5.2) des majorantes (5.3), (5.5) et (5.6); nous obtenons:
ZZZ
ZZZ
1
1
ui (x, t)ui (x, t) x <
ui (x, 0)ui (x, 0) x+
2
2
r0 >R2
r0 >R1
3
1
W 2 (0)t 4
W (0) t
+ 1 1 3
.
+
2(R2 R1 ) 2 4 2 4 (R2 R1 )
(5.7)
1
Z
3
2
J (t ) dt <
J (t ) dt
34
t4
Z t
Z t
p1 Z t
q1
p
0
0
0
0
0
q
0
0
(t ) dt
(t )(t ) dt <
(t ) dt
1 1
( + = 1; 1 < p, 1 < q).
p q
234
Jean Leray.
X Z Z Z
uk
(x, t0 )
ui (x, t0 )
x
xk
2
< W (t0 )J 2 (t0 );
therefore
1
2
ZZZ
p2 (x, t0 ) x <
1 p
W (t0 )J 3 (t0 );
2
and it follows
1
(5.5)
sZ Z Z
p2 (x, t0 ) x
dt
0
[W (0)] 4
<
2
W (0)
2(2)
3
4
t4 .
ZZZ
0
( 1r )
0 ui (y, t )
ui (y, t )
y.
xk
yk
This formula is analogous to (5.4). By calculations like the preceeding it leads to the
inequality
Z
(5.6)
s
ZZZ
0
dt
0
1
ui (x, t0 )ui (x, t0 )
2
2
x <
W (0)
2(2)
3
4
t4 .
3
1
W (0) t
W 2 (0)t 4
+
+ 1 1 3
.
2(R2 R1 ) 2 4 2 4 (R2 R1 )
(5.7)
1
Z
3
2
J (t ) dt <
J (t ) dt
34
t4
Z t
Z t
p1 Z t
q1
p
0
0
0
0
0
q
0
0
(t ) dt
(t )(t ) dt <
(t ) dt
1 1
( + = 1; 1 < p, 1 < q).
p q
235
Cette inegalite renseigne sur la facon dont lenergie cinetique reste localisee `a distance finie.
28. Donnons nous `a linstant initial t = 0 un etat initial constitue par un vecteur
quelconque, Ui (x), dont les composantes sont de carres sommables sur et dont la quasidivergence est nulle. Le vecteur Ui (x) constitue un etat initial regulier (cf. lemme 6 et
paragraphe 8); nommons ui (x, t) la solution reguli`ere des equations (5.1) qui lui correspond; elle est definie pour toutes les valeurs de t. Le but de ce chapitre est detudier les
limites que peut avoir cette solution reguli`ere ui (x, t) du systeme (5.1) quand tend vers
zero.
Les proprietes des fonctions ui (x, t) dont nous ferons usage sont les trois suivantes:
1o ) Soit ai (x, t) un vecteur quelconque de divergence nulle, dont toutes les composantes et toutes leurs derivees sont uniformement et fortement continues en t; nous
avons dapr`es (5.1):
ZZZ
ZZZ
Z
(5.8)
ZZZ
dt
0
Z
+
ui (x, t0 )
ZZZ
0
dt
0
ai (x, t0 )
x+
ai (x, t ) +
t0
0
ai (x, t0 )
x.
xk
t0
(5.10)
1
1
1
J 2 (t0 ) dt0 + W (t) = W (t0 ) < W (0).
2
2
2
ZZZ
Par definition W (0) =
Ui (x)Ui (x) x .
235
This inequality shows how kinetic energy remains localized at finite distance.
28. Suppose we have given at t = 0 an arbitrary initial vector Ui (x), with components
square summable on and quasi-divergence zero. The vector Ui (x) is a regular initial state
(cf. lemma 6 and paragraph 8). Write ui (x, t) for the corresponding regular solution to
equations (5.1). It is defined for all t. The object of this chapter is to study the limits
which this regular solution may have as tends to zero.
We will use the following three properties of the ui (x, t).
1o ) Let ai (x, t) be an arbitrary vector of divergence zero, of which all components and
all their derivatives are uniformly and strongly continuous in t. By (5.1):
ZZZ
ZZZ
Z
(5.8)
ZZZ
dt
0
Z
+
ui (x, t0 )
ZZZ
0
dt
0
ai (x, t0 )
ai (x, t ) +
x+
t0
0
ai (x, t0 )
x.
xk
t0
(5.10)
1
1
1
J 2 (t0 ) dt0 + W (t) = W (t0 ) < W (0).
2
2
2
ZZZ
By definition W (0) =
Ui (x)Ui (x) x .
3o ) Inequality (5.7) and the inequality W (0) < W (0) justify the following proposition
236
Jean Leray.
Ui (x)Ui (x) x = ,
2
r0 >R1 ()
designons par s(, t) la sph`ere, qui depend contin
ument de et de t, dont le centre est
lorigine des coordonees et dont le rayon est:
"
#
3 1
4 W (0) t
W 2 t4
R2 (, t) = R1 (, t) +
+ 1 1 3 ,
2
24 2 4
nous avons:
ZZZ
(5.11)
lim sup
0
s(,t)
29. Faisons tendre vers zero par une suite denombrable de valeurs: 1 , 2 , . . . Considerons les fonctions W (t) qui leur correspondent; elles sont bornees dans leur ensemble
et chacune delles est decroissante. Le Procede diagonal de Cantor (4) permet dextraire
de la suite 1 , 2 , . . . une suite partielle l1 , l2 , . . . telle que les fonctions W (t) conrrespondantes convergent pour chaque valeur rationnelle de t. Ces fonctions W (t) convergent
alors vers une fonction decroissante, sauf peut-etre en des points de discontinuite de cette
derni`ere. Les points de discontinuite dune fonction decroissante sont denombrable. Une
seconde application du Procede diagonal de Cantor permet donc dextraire de la suite
l1 , l2 , . . . une suite partielle m1 , m2 , . . . telle que les fonctions W (t) correspondants
convergent1 quel que soit t. Nous nommerons W (t) la fonction decroissante qui est leur
limite. (Cette definition ne contredit pas (5.10).)
Linegalite W (t) < W (0) prouve que chacune des integrals:
Z t2
ZZZ
Z t2
ZZZ
0
0
0
dt
ui (x, t ) x;
dt
uk (x, t0 )ui (x, t0 ) x
t1
t1
est inferieure `a une borne independente de . Par une troisi`eme application du Procede
diagonal de Cantor nous pouvons donc extraire de la suite m1 , m2 , . . . une suite partielle
n1 , n2 , . . . telle que chacune de ces integrals ait une limite
1
236
Jean Leray.
Ui (x)Ui (x) x =
2
r0 >R1 ()
and write s(, t) for the sphere with center at the origin with radius
"
#
3 1
4 W (0) t
W 2 t4
R2 (, t) = R1 (, t) +
+ 1 1 3 .
2
24 2 4
We have
ZZZ
(5.11)
lim sup
0
s(,t)
29. Let tend to zero through a countable sequence of values 1 , 2 , . . . Consider the
corresponding functions W (t). This is a bounded set of functions and each is decreasing.
Cantors diagonal method (4) allows us to extract from the sequence 1 , 2 , . . . a subsequence l1 , l2 , . . . such that the W (t) converge for all rational values of t. The W (t)
therefore converge to a decreasing function, except maybe at points of discontinuity of the
limit. The points discontinuity of a decreasing function are countable. A second application of Cantors method allows us to extract from l1 , l2 , . . . a subsequence m1 , m2 , . . .
such that the corresponding W (t) converge1 for all t. We write W (t) for the decreasing
function which is their limit. (This definition does not contradict (5.10).)
The inequality W (t) < W (0) shows that each of the integrals
Z t2
ZZZ
Z t2
ZZZ
0
0
0
dt
ui (x, t ) x;
dt
uk (x, t0 )ui (x, t0 ) x
t1
t1
is less than a bound independent of . By a third use of Cantors diagonal method we can
therefore extract from the sequence m1 , m2 , . . . a subsequence n1 , n2 , . . . such that each
of these integrals has a unique
1
237
unique quand t1 et t2 sont rationnels et que $ est un cube daretes parall`eles aux axes et de
sommets `a coordonees rationnelles. Linegalite W (t) < W (0) et les hypoth`eses faite sur
les fonctions ai (x, t) permettent daffirmer que dans ces conditions chacune des integrales:
ZZZ
Z t
ai (x, t0 )
0
0
0
dt
ui (x, t ) ai (x, t ) +
x;
t0
0
Z t
ZZZ
ai (x, t0 )
0
dt
uk (x, t0 )ui (x, t0 )
x
xk
0
a une limite unique. Ce resultat, porte dans (5.8), nous apprend que lintegrale
ZZZ
ui (x, t)ai (x, t) x
converge vers une limite unique, quels que soient ai (x, t) et t. Donc (cf. Corollaire du
lemme 7) les fonctions ui (x, t) convergent faiblement en moyenne vers une limite Ui (x, t)
pour chaque valeur de t.
Ainsi, etant donnee une suite de valeurs de qui tendent vers zero, on peut en extraire
une suite partielle telle que les fonctions W (t) convergent vers une limite unique W (t)
et que les fonctions ui (x, t) aient pour chaque valeur de t une faible limite en moyenne
unique: Ui (x, t). Nous supposerons desormais que tend vers zero par une suite de valeurs
telle que ces deux circonstances se produisent.
Remarque I Nous avons dapr`es (1.9):
ZZZ
W (t)
Ui (x, t)Ui (x, t) x.
De ces hypoth`eses resulte en effet quetant donnes t, un nombre (> 0) et une fonction
(x, t) egale `a lune des derivees des fonctions ai (x, t) on peut trouver un entier N et deux fonctions discontinues (x, t) et (x, t) qui possedent les proprietes suivantes: ces fonctions (x, t),
(x, t) restent constantes quand x1 , x2 , x3 , t varient sans atteindre aucune valeur multiple de
1
N ; chacune delles est nulle hors dun domaine $ ; on a:
ZZZ
0
dt
0
237
limit when t1 and t2 are rational and $ is a cube with sides parallel to the axes and with
vertices having rational coordinates. The inequality W (t) < W (0) and the hypotheses
made on the ai (x, t) imply that the integrals
ZZZ
Z t
ai (x, t0 )
0
0
0
dt
ui (x, t ) ai (x, t ) +
x;
t0
0
Z t
ZZZ
ai (x, t0 )
0
dt
uk (x, t0 )ui (x, t0 )
x
xk
0
converges to a unique limit, for all ai (x, t) and t. Therefore (cf. Corollary to lemma 7) the
ui (x, t) converge weakly in mean to some limit Ui (x, t) for each value of t.
Also, given a sequence of values of which tend to zero, one can extract from them
a subsequence such that the W (t) converge to a unique limit W (t) and that the ui (x, t)
have for each value of t a unique weak limit in mean: Ui (x, t). We suppose from here on
that tends to zero through a sequence of values such that these two conditions hold.
Remark I By (1.9)
ZZZ
W (t)
In fact these hypotheses imply the following. Given t, a number (> 0) and a function
(x, t) equal to one of the derivatives of the ai (x, t), one can find an integer N and two discontinuous functions (x, t) and (x, t) with the following properties. (x, t) and (x, t) remain
1
constant when x1 , x2 , x3 , t vary without hitting(?) any multiple of N
, and each of them is zero
outside of a domain $ , and(?)
ZZZ
0
dt
0
238
Jean Leray.
t0
1
1
1
J 2 (t0 ) dt0 + W (t) W (t0 ) W (0)
2
2
2
(0 t0 t).
Le lemme 2 nous apprend ensuite que sur tout domaine $ les fonctions u
i (x, t1 )
convergent fortement en moyenne vers les fonctions Ui (x, t);
ZZZ
ZZZ
lim
ui (x, t1 )ui (x, t1 ) x =
Ui (x, t1 )Ui (x, t1 ) x.
Cette suite partielle que nous choisissons est fonction de lepoque t1 envisagee.
238
Jean Leray.
t0
1
1
1
J 2 (t0 ) dt0 + W (t) W (t0 ) W (0)
2
2
2
(0 t0 t).
ZZZ
239
ZZZ
lim sup
u
i (x, t1 )ui (x, t1 ) x
Do`
u, puisque est arbitrairement faible et que W (t1 ) a une valeur limite:
ZZZ
(5.13)
ZZZ
lim
Appliquons le crit`ere de forte convergence enonce p. 200; nous constatons que sur les
fonctions ui (x, t) convergent fortement en moyenne vers les fonctions Ui (x, t) pours toutes
les valeurs t1 de t qui nappartiennent pas `
a lensemble de mesure nulle sur lequel la limite
W (t1 ) =
Les fonctions ui (x, t1 ) elles aussi convergent fortement en moyenne vers Ui (x, t1 ) (cf.
Generalisation du lemme 3, p. 207). Lintegrale qui figure dans (5.8):
ZZZ
ai (x, t0 )
uk (x, t0 )ui (x, t0 )
x
xk
ai (x, t0 )
x
xk
pour presque toutes les valeurs de t0 (cf. (1.8)); cette integrale est dautre part inferieure `a
3W (0) max |
ai (x, t0 )
|;
xk
permet
ZZZ
239
ZZZ
lim sup
u
i (x, t1 )ui (x, t1 ) x
From this, since is arbitrarily small and since W (t1 ) has a limit
ZZZ
(5.13)
ZZZ
lim
We apply the strong convergence criterion from p. 200. Note that on the ui (x, t)
converge strongly in mean to the Ui (x, t) for all values t1 of t not belonging to the set of
measure zero on which lim inf J (t) = +.
For all these values of t the two sides of (5.13) are equal, i.e.
ZZZ
(5.14)
W (t1 ) =
The functions ui (x, t1 ) also converge strongly in mean to Ui (x, t1 ) (cf. Generalisation
of lemma 3, p. 207). The integral which figures in (5.8)
ZZZ
ai (x, t0 )
uk (x, t0 )ui (x, t0 )
x
xk
therefore converges to
ZZZ
Uk (x, t0 )Ui (x, t0 )
ai (x, t0 )
x
xk
for almost all values of t0 (cf. (1.8)). Further, this integral is less than
3W (0) max |
ai (x, t0 )
|
xk
sign gives
240
Jean Leray.
Z
lim
ZZZ
dt
ZZZ
dt
0
ai (x, t0 )
x =
xk
ai (x, t0 )
x,
xk
le second membre de cette relation pouvant etre mis, dapr`es lemme 5, sous la forme:
Z t
ZZZ
0
dt
Uk (x, t0 )Ui,k (x, t0 )ai (x, t0 ) x.
D`es le debut de ce paragraphe nous avions le droit daffirmer que les autres termes
qui figurent dans (5.8) convergent de meme ver des limites qui sobtiennent en substituant
Ui (x, t) `a ui (x, t), Ui (x) `a Ui (x). Par suite:
ZZZ
ZZZ
Ui (x, t)ai (x, t) x =
Ui (x)ai (x, 0) x
Z
(5.15)
ZZZ
dt
ai (x, t0 )
Ui (x, t) ai (x, t ) +
x
t0
ZZZ
0
dt
0
31. Les resultats ainsi obtenus conduisent `a la definition suivante: Nous dirons quun
vecteur Ui (x, t), defini pour t 0, constitue une solution turbulente des equations de
Navier quand les conditions que nous allons enoncer se trouveront realisees, les valeurs de
t que nous nommerons singuli`eres constituant un ensemble de mesure nulle:
Pour chacque valeur positive de t les fonctions Ui (x, t) sont de carres sommables sur
, et le vecteur Ui (x, t) a une quasi-divergence nulle.
La fonction:
240
Jean Leray.
Z
lim
ZZZ
dt
ZZZ
dt
0
ai (x, t0 )
x =
xk
ai (x, t0 )
x,
xk
By lemma 5 the right hand side of this can be put into the form
Z t
ZZZ
0
dt
Uk (x, t0 )Ui,k (x, t0 )ai (x, t0 ) x.
From the beginning of this paragraph we can claim that the other terms in (5.8)
similarly converge. We obtain the limits by substituting Ui (x, t) for ui (x, t) and Ui (x) for
Ui (x). This gives
ZZZ
ZZZ
Ui (x)ai (x, 0) x
Ui (x, t)ai (x, t) x =
Z
(5.15)
ZZZ
dt
ai (x, t0 )
Ui (x, t) ai (x, t ) +
x
t0
ZZZ
0
dt
0
31. These results lead to the following definition. We say that a vector Ui (x, t)
defined for t 0 constitutes a turbulent solution to Naviers equations when the following
conditions are realised, where values of t that we call singular form a set of measure zero.
For each positive t the functions Ui (x, t) are square summable on and the vector
Ui (x, t) has quasi-divergence zero.
The function
ZZZ
0
dt
0
241
ZZZ
ai (x, t0 )
0
Ui (x, t ) ai (x, t ) +
x
Ui (x, t)ai (x, t) x
t0
ZZZ
0
dt
0
est constante (t 0). (Autrement dit la relation (5.15) a lieu.) Pour toutes les valeurs
positives de t, sauf eventuellement pour certaines valeurs singuli`eres, les fonctions Ui (x, t)
poss`edent des quasi-derivees Ui,j (x, t), de carres sommables sur .
Nous poserons:
ZZZ
2
J (t) =
Ui,j (x, t)Ui,j (x, t) x,
J(t) se trouvant donc defini pour presque toutes les valeurs positives de t.
Il existe une fonction W (t), definie pour t 0, qui poss`ede les deux proprietes suivantes:
Z t
1
la fonction
J 2 (t0 ) dt0 + W (t) est non croissante;
2
0
RRR
on a:
U (x, t)Ui (x, t) x W (t), linegalite nayant lieu qu`a certaines epoques sin i
guli`eres, dont lepoque initiale t = 0 ne fait pas partie.
Nous dirons quune telle solution turbulente correspond `
a letat initial Ui (x) quand
nous aurons: Ui (x, 0) = Ui (x).
La conclusion de ce chapitre peut alors se formuler comme suit:
Theor`eme dexistence: Supposons donne `
a linstant initial un etat initial Ui (x) tel que
les fonctions Ui (x) soient de carres sommables sur et que le vecteur de composantes
Ui (x) poss`ede une quasi-divergence nulle. Il correspond `
a cet etat initial au moins une
solution turbulente, que est definie pour toutes les valeurs du temps post`erieures `
a linstant
initial.
VI. Structure dune solution turbulente.
32. Il nous reste `a etablir quel liens existent entre les solutions reguli`eres el les
solutions turbulentes des equations de Navier. Il est tout dabord
ZZZ
0
dt
0
241
ZZZ
ai (x, t0 )
0
Ui (x, t ) ai (x, t ) +
x
Ui (x, t)ai (x, t) x
t0
ZZZ
0
dt
0
is constant (t 0). (Equivalently, (5.15) holds.) For all positive values of t except possibly
for certain singular values, the functions Ui (x, t) have quasi-derivatives Ui,j (x, t) which are
square summable on .
Set
ZZZ
2
J (t) =
Ui,j (x, t)Ui,j (x, t) x,
242
Jean Leray.
evident que toute solution reguli`ere constitue a fortiori une solution turbulente. Nous
allons chercher dans quels cas une solution turbulente se trouve constituer une solution
reguli`ere. Generalisons `a cet effet les raisonnements du paragraphe 18 (p. 221).
Comparaison dune solution reguli`ere et dune solution turbulente: Soit une solution
des equations de Navier, ai (x, t), definie et semi-reguli`ere pour t < T ; nous supposons quelle devient irreguli`ere quand t tend vers T , `a moins que T ne soit egal `a +.
Considerons une solution turbulente, Ui (x, t), definie pour t, lepoque netant pas
singuli`ere. Les symboles W (t) et J(t) se rapporteront `a cette solution turbulente. Nous
poserons:
ZZZ
ZZZ
w(t) = W (t) 2
Ui (x, t)ai (x, t) x +
ai (x, t)ai (x, t) x
ZZZ
2
j (t) = J (t) 2
ai (x, t)
Ui,j (x, t)
x +
xj
ZZZ
ai (x, t) ai (x, t)
x.
xj
xj
dt
x +
ai (x, t)ai (x, t) x
xj
xj
2
1
J 2 (t0 ) dt0 + W (t)
2
1
j (t ) dt + w(t) + 2
2
2
ZZZ
0
Ui,k (x, t0 )
dt
ai (x, t0 )
x+
xk
ZZZ
Ui (x, t)ai (x, t) x
242
Jean Leray.
clear that any regular solution is a fortiori a turbulent solution. We are going to look for
those cases in which a turbulent solution is regular. To this end we generalize the reasoning
of paragraph 18 (p. 221).
Comparison of a regular solution and a turbulent solution: Let ai (x, t) be a solution to
Naviers equations, defined and semi-regular for t < T . We suppose that it becomes
irregular when t tends to T , at least in the case when T is not equal to +. Consider a
turbulent solution Ui (x, t) defined for t, where is not a singular time. The symbols
W (t) and J(t) correspond to the turbulent solution. Set
ZZZ
ZZZ
w(t) = W (t) 2
Ui (x, t)ai (x, t) x +
ai (x, t)ai (x, t) x
ZZZ
2
j (t) = J (t) 2
ai (x, t)
Ui,j (x, t)
x +
xj
ZZZ
ai (x, t) ai (x, t)
x.
xj
xj
dt
xj
xj
2
1
j (t ) dt + w(t) + 2
2
2
ZZZ
0
Ui,k (x, t0 )
dt
ai (x, t0 )
x+
xk
ZZZ
Ui (x, t)ai (x, t) x
243
est une solution semi-reguli`ere des equations de Navier: nous constatons que la fonction
non croissante (6.1) est `a une constante pr`es egale `a la suivante:
Z
(6.2)
1
j (t ) dt + w(t) +
2
2
ZZZ
dt
ZZZ
1
ai (x, t0 )ai (x, t0 )
[ak (x, t0 ) Uk (x, t0 )]
x = 0;
2
xk
lintegrale
ZZZ
[ak (x, t0 ) Uk (x, t0 )]Ui,k (x, t0 )ai (x, t0 ) x
ai (x, t0 )
[ak (x, t ) Uk (x, t )] Ui,k (x, t )
ai (x, t0 ) x
x
k
j (t ) dt + w(t)
w(t0 )j(t0 )V (t0 ) dt0 .
2
Or
Z
Z tp
Z t
1
0
0
0
0
j (t ) dt
w(t )j(t )V (t ) dt +
w(t0 )V 2 (t0 ) dt0
4
243
is a semi-regular solution to Naviers equations. Note that the nonincreasing function (6.1)
is up to a constant nearly equal to
Z
(6.2)
1
j (t ) dt + w(t) +
2
2
ZZZ
dt
ZZZ
1
ai (x, t0 )ai (x, t0 )
[ak (x, t0 ) Uk (x, t0 )]
x = 0.
2
xk
The integral
ZZZ
[ak (x, t0 ) Uk (x, t0 )]Ui,k (x, t0 )ai (x, t0 ) x
ZZZ
ai (x, t0 )
0
0
0
[ak (x, t ) Uk (x, t )] Ui,k (x, t )
ai (x, t0 ) x
x
k
where V (t0 ) is the greatest length of the vector ai (x, t) at time t0 . Since (6.2) is not
increasing, it is a fortiori the same for the function
Z t
Z tp
1
2 0
0
j (t ) dt + w(t)
w(t0 )j(t0 )V (t0 ) dt0 .
2
Now
Z
Z tp
Z t
1
0
0
0
0
j (t ) dt
w(t )j(t )V (t ) dt +
w(t0 )V 2 (t0 ) dt0
4
244
Jean Leray.
1
1
w(t)
2
4
(6.3)
w(t) w()e 2
Rt
V 2 (t0 ) dt0
( < t < T ).
244
Jean Leray.
1
1
w(t)
2
4
(6.3)
w(t) w()e 2
Rt
V 2 (t0 ) dt0
( < t < T ).
Suppose in particular that the solutions Ui (x, t) and ai (x, t) correspond to the same
initial state. Then w() = 0 and by (6.3) w(t) = 0. therefore Ui (x, t) = ai (x, t) for
t < T . The uniqueness theorems of paragraphs 18 and 23 (p. 222 and 228) are special
cases of this result.
33. Regularity of a turbulent solution in certain time intervals.
Consider a turbulent solution Ui (x, t) defined for t 0. For each nonsingular time the
vector Ui (x, t) is a semi-regular initial state (cf. p.231 existence theorem, case a)). The
uniqueness theorem that we are going to establish will have the following consequence.
Consider a nonsingular time, i.e. a time chosen outside of a certain set of measure zero.
Then this is the origin of an interval of time in the interior of which the turbulent solution
coincides with a regular solution of Naviers equation, and this coincidence does not end as
long as the regular solution remains so. This result, complemented by some others which
are easy to establish, gives us the next theorem.
Structure theorem.
For a vector Ui (x, t) to be a turbulent solution to Naviers equations for t 0, it is
necessary and sufficient that it have the following three properties.
a) By an interval of regularity we mean any interval l Tl of time in the interior of
which the vector Ui (x, t) is a regular solution to Naviers equations, and such that this
is true for no interval containing l Tl . Let O be the open set which is the union of the
intervals of regularity
1
I have not been able to establish a uniqueness theorem stating that to a given initial state,
there corresponds a unique turbulent solution.
245
(qui sont deux `a deux sans point commun). O ne doit differer du demi-axe t 0 que par
un ensemble de mesure nulle.
RRR
b) La fonction Ui (x, t)Ui (x, t) x est decroissante sur lensemble que constitue O
et linstant initial t = 0.
c) Quand t0 tend vers t les fonctions Ui (x, t0 ) doivent converger faiblement en moyenne
vers les fonctions Ui (x, t).
Complements:
1) Une solution turbulente correspondant `a un etat initial semi-regulier coincide avec
la solution semi-reguli`ere qui correspond `a cet etat initial aussi longtemps que celle-ci
existe.
2) Faisons tendre en croissant t vers lextremite Tl dun intervalle de regularite. La
solution Ui (x, t), qui est reguli`ere pour l < t < Tl devient alors irreguli`ere.
Ce theor`eme de structure nous permet de resumer notre travail en ces termes: Nous
avons essaye detablir lexistence dune solution des equations de Navier correspondant `a
un etat initial donne: nous ny avons reussi quen renoncant `a la regularite de la solution
en certains instants, convenablement choisis, dont lensemble est de mesure nulle; en ces
instants la solution nest assujettie qu`a une condition de continuite tr`es large (c) et `a une
condition exprimant la non-croissance de lenergie cinetique (b).
Remarque: Si le syst`eme (3.11) poss`ede une solution non nulle Ui (x) cette solution
permet de construire un exemple tr`es simple de solution turbulente cest le vecteur Ui (x, t)
egal `a
h
i
12
12
[2(T t)] Ui [2(T t)] x pour t < T
et `a 0 pour t > T ; il existe une seule epoque dirregularite: T .
34. Complements relatifs aux intervalles de regularite et `
a lallure dune solution des
equations de Navier pour les grandes valeurs du temps.
Le chapitre IV nous fournit, outre le theor`eme dexistence utilise au paragraphe
precedent, linegalite (4.3); do`
u resulte la proposition suivante: considerons
245
(no two of which have a point in common). O must only differ from the half axis t 0 by
a set of measure zero.
RRR
b) The function
U (x, t)Ui (x, t) x is decreasing on the set consisting of O and
i
t = 0.
c) As t0 tends to t the Ui (x, t0 ) must converge weakly in mean to the Ui (x, t).
Supplementary information
1) A turbulent solution corresponding to a semi-regular initial state coincides with
the semi-regular solution having that initial state, for as long a time as the semi-regular
solution exists.
2) Make t increase to Tl in an interval of regularity. Then the solution Ui (x, t) which
is regular for l < t < Tl becomes irregular.
This structure theorem allows us to summarize our work in these terms: We have
tried to establish the existence of a solution to Naviers equations corresponding to a given
initial state. We have had to give up regularity of the solution at a set of times of measure
zero. At these times the solution is only subject to a very weak continuity condition (c)
and to condition (b) expressing the nonincrease of kinetic energy.
Remark: If system (3.11) has a nonzero solution Ui (x) then we can very simply construct a turbulent solution Ui (x, t) equal to
h
i
12
12
[2(T t)] Ui [2(T t)] x for t < T
and to 0 for t > T . This has a single irregular time T .
34. Supplementary information on intervals of regularity and behavior of solutions to
Naviers equations for large time.
Chapter IV gives inequality (4.3) in addition to the existence theorem used in the
previous paragraph. This results in the following proposition. Consider
246
Jean Leray.
une solution turbulente Ui (x, t); soit une epoque non singuli`ere t et une epoque Tl posterieure; nous avons:
3
Tl
1
W (0),
2
il vient:
1
W (0).
2
Toutes les epoques singuli`eres sont donc anterieures `
a lepoque:
5
(6.4)
W 2 (0)
.
16A41 5
Autrement dit, il existe un intervalle de regularite qui contient cette epoque et qui setend
jusqu`
a +. Un mouvement regulier jusqu`a lepoque ne devient jamais irregulier.
Il est aise de preciser ce resultat:
Soit un intervalle de regularite de longueur finie l Tl ; toute epoque t interieure `a cet
intervalle est non singuli`ere; nous avons donc dapr`es (4.3):
3
2
l Tl
l
il vient, le signe
(6.5)
P0
l
2A21 2
X0 p
1
(Tl l ) < W (0).
2
l
246
Jean Leray.
a turbulent solution Ui (x, t). Let t0 be a nonsingular time and Tl a later time. We have
3
Tl
we get
1
W (0)
2
1
W (0).
2
(6.4)
W 2 (0)
.
16A41 5
In other words, there is an interval of regularity that contains and which extends to +.
A motion which is regular up to time never becomes irregular.
It is easy to make this more precise.
Let l Tl be an interval of regularity of finite length. All times t interior to this interval
are nonsingular. By (4.3)
3
2A21 2
(6.5)
X0 p
1
(Tl l ) < W (0).
2
l
247
1
J(t)}.
1+A
{A21 2 (t t0 ) 2 ;
1
1
J 2 (t)} dt0 W (0).
2
(1 + A)
2
Cette inegalite (6.6) fournit pour les valeurs de t superieures `a une majorante de J(t);
cette majorante a dailleurs une expression analytique assez compliquee.
Nous nous contenterons de remarquer que de (6.6) resulte linegalite moins precise:
Z t
1
3
1
1
{A21 2 t 2 ;
J 2 (t)} dt0 W (0);
2
(1 + A)
2
0
cette derni`ere exprime tout simplement que
(6.7)
J 2 (t) <
(1 + A)2 W (0)
W 2 (0)
pour t >
.
2
t
4A41 5
Completons ce resultat, qui est relatif `a lallure asymptotique de J(t), par un autre
relatif `a lallure asymptotique de V (t): Les inegalites (4.3) et (4.4) nous apprennent que:
1
247
In Chapter IV we found the pair of inequalities (4.3) (4.4). These imply the following.
Consider a turbulent solution, a nonsingular time t0 , and a later time t. We have
either t t0 > A41 3 J 4 (t0 ), or J(t) < (1 + A)J(t0 )
in other words1
1
1
J(t)}.
1+A
{A21 2 (t t0 ) 2 ;
1
1
2
0
W (0).
J
(t)}
dt
(1 + A)2
2
This gives an upper bound for J(t) for values of t larger than . However this bound has
a rather complicated analytic expression.
We content ourselves by remarking that (6.6) gives the less precise
Z t
1
3
1
1
{A21 2 t 2 ;
J 2 (t)} dt0 W (0).
2
(1 + A)
2
0
This can most simply be expressed as
(6.7)
J 2 (t) <
(1 + A)2 W (0)
W 2 (0)
.
for t >
2
t
4A41 5
248
Jean Leray.
2
Dapr`es (6.7) cette derni`ere inegalite est satisfaite pour t0 = 12 t quand on prend t > A W 5(0) ;
on a donc pour ces valeurs de t:
p
3
V (t) < A W (0)(t) 4 .
En resume il existe des constantes A telles que lon ait:
J(t) < A
p
p
1
3
W 2 (0)
W (0)(t) 2 et V (t) < A W (0)(t) 4 pour t > A
.
5
248
Jean Leray.
2
By (6.7) this last inequality is satisfied for t0 = 12 t if one takes t > A W 5(0) . One therefore
has for these t
p
3
V (t) < A W (0)(t) 4 .
In summary there exist some constants A such that
J(t) < A
p
p
1
3
W 2 (0)
W (0)(t) 2 and V (t) < A W (0)(t) 4 for t > A
.
5