Vous êtes sur la page 1sur 23


Abstract LP Estimates for the Cauchy Problem

with Applications to the Navier-Stokes Equations
in Exterior Domains

Department of Mathematics, Hokkaido University,

Sapporo 060, Japan



Department of Mathematics, University of Paderborn,

D-4790 Paderborn, Germany

Communicated by the Editors

Received November 8, 1989; revised July 20, 1990

We apply an abstract perturbation theorem to derive global in time LY estimates

for the Cauchy problem and L4-L estimates for the nonstationary Stokes
equations in exterior domains. This will be applied to obtain various new global in
time estimates for weak solutions of the Navier-Stokes equations. f; 1991 Academic
Press, Inc.


We are mainly interested in global in time Ly estimates for the

nonstationary Stokes system in a domain Sz in UP, n > 2, with smooth
boundary &J (at least of class C2+P, 0 < p < 1):

div u = 0, UlZiCJ=O, u(x, 0) = u(x). (1.1)

Here u = (u(x, t), .... u(x, t)) and p = p(x, t) represent the unknown
velocity and pressure, respectively; f =f(x, t) represents a given external
force and a denotes the initial velocity. For the moment we assumea = 0.
When n = 3 and 52is a bounded or an exterior domain (i.e., domain having
a nonempty compact complement), for every JE Ly(sZ x (0, T))n with
OO22-1236/91 $3.00
Copyright Y 1991 by Academic Press. Inc
All rights of reproduclmn m any form reserved

0 < T< cc, 1 <q < co, Solonnikov [42] constructed a unique solution
(u, VP) of (1.1) in Q x [0, T) satisfying the Ly estimate


with C = C(T, 52, q) independent of f; 11.jly denotes the norm in Ly(Q) or

LY(Q)m = Ly(B) -x . . . x L4(Q) with m = n or n* and V*u = (ai aj~)i, i= 1,2,...,n
is the matrix of themsecond order derivatives of U. When L2 is unbounded,
Solonnikovs estimate (1.2) is not global in time because C( T, Q, q) may
tend to infinity as T-t a. His approach is based on methods in the theory
of potentials.
Our purpose is to strengthen such estimates in two directions. First, our
estimate is global in time, i.e., the above constant C is independent of T.
Second, the integral norms we use may have different exponents q, s in
space and time.
To derive such global Lq - L estimates for (1.1) we use and extend an
abstract perturbation result recently developed by Dore and Venni [ 131.
Such estimates are important in studying regularity and large time
behavior of weak or strong solutions of the nonstationary Navier-Stokes
system in an exterior domain Q:

;-du+(u,V)u+Vp=i, div u = 0, 2.dI ,?D= 0, u(x, 0) = u(x).

Here we set (u, V) = x:1=, U ai, di = a/axi. The system (i.3) has been
extensively studied (see, e.g., [4, 5, 18, 25, 28, 31, 32, 39-411). Although a
global weak solution under suitable assumptions on a and f is known to
exist, its regularity is not known unless n = 2. As an application of our
estimates we derive various global a priori estimates for a given weak
solution u and the associated pressure p; for example, when Q is an
exterior domain in R with it 2 3 we prove that

jz IlV*u(t)ll, dt + j
0 0
dt < cc (1.4)

ifn+l=n/q+2/s, l<q<cc, l<s<oo,j, Ilf(t)ll:dt<oo (anda=Ofor

simplicity); the pressure p can be chosen so that

,I Ildt)ll; dt < 00, (1.5)
where n = n/r + 2/s. We note that this pressure estimate for n = 3,
r = s = 5/3 improves the partial regularity theory of suitable weak solutions

in [9]. The global result (1.5) is new while 1: Il~(t)il: dt < CC for finite T is
known by [40].
From (1.4) we deduce a global result on U:

Iox Ilu(t) dt < ~0, (1.6)

where n - 1 = n/k + 2/s.

Although there are many results on large time behavior of weak solutions
(see, e.g., [3,4]), our result (1.6) is not contained in the literature because
(i) our estimate holds for weak solutions which need not satisfy energy
inequalities and (ii) we estimate the decay as t --) cc by an integral norm
while the algebraic decay over of Ilu(t) as t + co is studied in the
literature (see [4] ).
In order to explain the abstract perturbation result we need some
notations. Let X be a Banach space with norm I(. /Ix = I(. /I and let A be a
closed linear operator with domain D(A)cX and range R(A)cX. P(X)
is the Banach space of all bounded linear operators with D(A) = X,
R(A) - G X, equipped
- with the operator norm which is again denoted by II . II.
By D(A), Z?(A) we denote the closure of D(A), R(A). We say A is
nonnegative if the resolvent set p(A) = (2~ @; (I -A)- E Y(A)} of A
contains all negative real numbers and

sup lIt(t+A)-II Coo, (1.7)


where Z is the identity and t + A = tZ+ A. For a nonnegative operator A we

can define the complex power A for all z E C as a densely defined closed
linear operator in the closed subspace X, = D(A) n R(A) in X (see [26]).
Let 8 > 0, K 3 1. We say a closed nonnegative operator A in X belongs
to E!(X) if AiY~~(X,) for all YE R and if the norm in Y(XA) is
estimated as
I(AYlJ <K&- for all y E R. (1.8)

A closed nonnegative operator A in X belongs to SC(X) if both D(A)

and R(A) are dense in X (i.e., X, = X), if A E Y(X) for all y E R and if
(1.8) holds in Y(X). Here X is always a complex Banach space; a real
space is embedded in its complexification.
Now we consider two nonnegative operators A and B in X. A and B are
called resolvent commuting if
(t+A)-(t+B)-=(t+B)-(t+A)- for all t > 0. (1.9)
We are interested in conditions under which the sum A + B with
D(A + B) = D(A) n D(B) has an inverse in a certain sense. Dore and Venni

showed in [13] that (A+B))EZ(X) if X is c-convex, AEcF~(X) and

BE d,(X) are resolvent commuting with 8 + g < rc, and if both A and B
have an inverse in Y(X). The following Theorem 1.1 extends this result to
the case that A-l, BP1 E Y(X) is not needed. The first proof of this result
has been given by the authors in [24]. For the convenience of the reader
we give an outline of the proof in Appendix B; this proof is basically
parallel to the one in [13]; in [34] Priiss and the second author gave
another proof of essentially the same result in another context.
The concept of c-convexity is stronger than that of uniform convexity or
reflexivity (see [6, 7, 131). For the application above it is important that
X is c-convex if and only if the truncated Hilbert transform

(H,f)(t&j f&%JT, .fEL(R X) (1.10)


converges as E --t 0 for almost all t E IF! and there is a constant C = C(s, X)
independent off such that

IlHf IILyR,x)G Cllfll LS(Iw,X), l<S<co, (1.11)

where (Hf)(t)=lim,,, (H,f)(t). More precisely, (1.10) and (1.11) for

some 1 < s < co imply c-convexity and c-convexity implies (1.10) and (1.11)
for all 1 < s < a3 [6, 71. By L(.Y; X) where 1 < s < cc and 9 is a real
interval, we denote the usual Banach space of L functions on 9 with
values in X and norm llfll LS(,fl:X,= (i,p I/f(t)11 Iy tit)..
Furthermore, for two operators AEC$(X), BEG> we let 6(A + B)
be the completion of D(A) n D(B) under the norm llAullx + IIBullx. Since
A, B may not have bounded inverses, &A + B) is not a subspace of X in
general. However, if A ~ E P(X) or BP E 2(X) we get @A + B) c X. It is
easy to see that these operators A and B are injective and so 11 Au11x + 1)Bull x
is a norm; indeed, (1.7) implies that t(t + A)- f -+ 0 as t + 00 for
~ER(A)=X and Au=0 yields u=t(t+A)-u and letting t+O implies
24= 0.
The operator A + B: @A + B) + X is now defined by closure, i.e., by

(A+ B)u=Au+ Bu=;i~l (Au,+ Bu,)

for all u E @A + B) where (u,) is a Cauchy sequence which represents U. In

the same way we define A : 6( A + B) -+ X and B: @A + B) + X by Au =
lim, _ m Auj and Bu = lim,, ;u Bu,. Evidently, &A + B) is a Banach space
equipped with the norm

IIAuII + IIWI =iT, (IIAuiII + IIBujlI ).


The following theorem extends Theorem 2.1 of [ 131.

THEOREM 1.1. Let X be a [-convex Banach space. Suppose that

AE&~(X), BE&~(X) are resolvent commuting with 9, a>O, K> 1,
8 + a < 7~.Then the operator A + B: 8(A + B) + X is bijective and boundedly
invertible; for every f EX there is a unique u E&A + B) such that
Au+ Bu=f and
II4 + IIBull d Cllf II
with a positive constant C= C(0, a, K, X).
Remarks. Let X, A, and B be as in Theorem 1.1. Then we easily
conclude that A + B: D(A) n D(B) + X is an injective operator with dense
range R(A + B) G X and it holds

lI4l + IlBull d Cll(A + Bbll

for all u ED(A + B) = D(A) n D(B) where C= C(O, 0, K, X) > 0. If
in addition A-ET(X) or BP Ed, then D(A+ B)=6(A+ B),
R(A + B) = X, and (A + B)) E Y(X). If both A- and BP are Y(X),
then we obtain the assertion of Theorem 2.1 in [ 131.
For some applications it is interesting to extend Theorem 1.1 to the case
that D(A) or R(A) is not dense in X, i.e., X, # X. In this case we consider
the restriction A, of A on

D(A,)= {uED(A)~ X,; AuED(A)}.

It turns out that AEE~(X) implies A,E&L(X,) since A$= AiY on X, by

Proposition 4.12 in [26]. For such operators, it is easy to extend Theorem 1.1
as follows :

THEOREM 1.2. Let X be [-convex. Suppose that A EEL(X), BE E;(X)

are resolvent commuting with 0, a > 0, K> 1, 8 + a < n, and X, =X,. Then
the operator A, + B,: fi(A, + B,) --* X0, defined as above with X replaced by
X,, = X, = X, is bijective and boundedly invertible, and

IlAull + IIWI d Cll(A + B)ull

for all u E @A0 + B,) with C = C(O, a, K, X) > 0.


The application of Theorem 1.1 yields a new result for the abstract
Cauchy problem. Let X be a i-convex Banach space, 1 <s < cc and

0 < T< co. By L(0, T; X) we mean the Banach space of all L functions on
the interval (0, T) with values in X; the norm is (IuII~~(,,~;~) =
(JOTIl4t)llx 41s. For UE L(0, T; X), U E L(0, T; X) means that the
derivative U = du/dt in the distribution sense exists as an element of
L(0, T; X); then u(O)EX is the well defined trace. For A E&$(X), D(A) is
regarded as a Banach space with the graph norm I(u(l,+ llAullx.
The following theorem extends the result by Dore and Venni [ 13,
Theorem 3.21 on abstract Cauchy problems in two directions: (i) the class
of operators A is larger since A PI E 9(X) is not needed and (ii) the
constant C does not depend on T and therefore T= cc is included; this
leads to global in time results.

THEOREM 2.1. Let X be [-convex. Assume that O< T< a, 1 <s< co,
and that A E 8;(X) for some K3 1 and some 0 with 0 d 0 < n/2. Then for
every f E L(0, T; X) there exists a unique solution u of the Cauchy problem

u + Au = f for a.e. t E (0, T), u(0) = 0 (2.1)

satisfying the properties :

UE L(0, T,; D(A)) for all T, d T and T, < co (2.2)

u E L(0, T; X) (2.3)

0 l,u(r)ll;dt+/ llAu~~)ll;,dMj-T Ilf(t)ll,dt (2.4)

s 0 0

with C= C(s, 0, K, X) independentqf f and T.

Remark 2.2. If A E E:(X) with 0 < 8 < nJ2, then -A generates a

bounded analytic semigroup e-IA, t > 0 [34]. So one can construct a
strong solution if f is Holder continuous in time with values in X (cf. [ 16,
Lemma 2.141). However, as is pointed out in [ 133, analyticity of eprA is
not enough to deduce the existence of solutions of (2.1) with (2.2)(2.4) for
general f o L(0, T; X) unless X is a Hilbert space (cf. [2, 371). According
to [2, 373, if (2.4) holds for some s (1 <s< co) it holds for all s provided
that -A generates a bounded analytic semigroup in X (cf. [lo] for a
weaker version). We thank Professor Kato for letting us know the results
in [2, 371.
It is not difficult to extend Theorem 2.1 when the initial data u(0) = a is
not zero. For this purpose we introduce

VEX; IIuIIop4~=Ilull,+ jox ~~t~Ae~vl~~~)< a}, (2.5)


where 0 < CY < 1. This space agrees with a real interpolation space
(D(A), A), --a,s when eprA is an analytic semigroup; see [S, Sect. 3.51. If
a ED~ rks we seethat u = ePIAu satisfies (2.2), (2.3) and solves u + Au = 0
with $0) = a. If u solves (2.1), then u + u solves (2.6),with (U + u)(O) = a.
This leads from Theorem 2.1 to the next more general result.

THEOREM 2.3. Assume the same assumptionsas in Theorem 2.1. Thenfor

every f EL(0, T; X) and aE DflP1ls, there exists a unique solution u of
u + Au = f ,for a.e. t E (0, T), u(0) = a (2.6)
satisfying (2.2), (2.3) and

I,,,u(t),,;dt+J 0
llAu(t)l/;dt<C(jO Ilf(t)ll,dt+ lid;;-,~~) (2.7)

with C= C(s, 9, K, X) independent off and T.

Remark 2.4. In Theorem 2.3 we may replace 02 by the homogeneous
space B:, the completion of D(A) under the norm

The space fi; agrees with (@A), X),Pa,s. Although this is not explicitly
written in [S, Chap. 31, it can be proved in a similar way as in the proof
of 02~ (D(A), X), ~Z,,,. All statements in Theorem 2.3 still hold if we
replace (2.2) by
u EL(0, T; b(A)), (2.2)

here b(A) is he completion of D(A) under llAullx.

Remark 2.5. For c1= 1 - l/s the space .(A+) is continuously
embedded in D:. with E > 0. Indeed,

< : llA1isPCemrAA+EuIISdt+ C, s, f ~~~~~


dc,,s0 4t IIA+uII+ Cllul/ (E, s small)

C,= sup llAyePrall tY and y = l/S-&,

where jlull = IlullX; this leads to the desired conclusion.


Proof That Theorem 1.1 Implies Theorem 2.1. First we study the
operator d/dt in (2.1). Set 8= L(0, T; X) with S, T, and X as in Theorem
2.1 and define B: D(B) + w by

D(B)= {uE&E~, u(O)=O), Bu = u = du/dt. (2.8)

Then B is nonnegative and D(B) and R(B) are both dense in $ B has a
bounded inverse if T is finite, and
IIB~vlIF< C(1 + y2)elvi12 for all y E R, (2.9)
where C= C(s, X). This has been proved essentially by Dore and Venni
[13, Theorem 3.11 if T<o3. However, the constant C in (2.13) of [13]
does not depend on T; therefore (2.9) holds even if T= a3. Indeed, we have
the explicit representation

((r+B)-,f)(x)=jOe --f(y)dlJ, x6(0, T)

for each r > 0, f E 2. From this it is clear that B is nonnegative and that
B- E P(g) if T is finite. The density of D(B) is clear and the density of
R(B) follows by a duality argument. Indeed, X is i-convex and therefore
reflexive. Therefore we have 8* = L(0, T; X*) with l/s + l/s = 1 for the
dual spaces 8*, X* of 2, A. Th e a djomt. operator B* on x* is defined by
B*u= -du/dt with D(B*)= {us%*; uE~*, u(T)=O} if T<cc and
D(B*) = {u~f*; u~z*} if T= co. Since B* is injective we see that R(B)
is dense in ,!?.
Next we observe that 8= L(0, T; X) is again i-convex for 1 < s < cc
(see [13, 353). Using (2.9) we see that BE&,(~) for every u> 7r/2 and
some L 2 1. We take 0 so close to n/2 that 8 + (T< rc. We may take K in
Theorem 2.1 so large that Ka L. We define a closed linear operator
A:D(d)-+$ D(A)~f,by(l!if)(t)=Af(t), D(A)={f~f;f(t)~D(A)for
a.e. tE (0, T), sl IIAf(t)(l,dt < co}. Since A E&~(X) we obtain AE&:(~).
Now we apply Theorem 1.1 to $ A, and B and conclude that for every
f~ 2 there is a unique UE 6(A + B) such that (A + B)u= f and IIAu]lm +
IIBull~<Cllfll~ with C=C(s,&K,X)>O. This implies (2.1) and (2.4).
From u~fi(A + B) we easily conclude (2.2), (2.3), and u(O) = 0. This
proves Theorem 2.1.
Now we apply Theorem 2.1 and Theorem 2.2 to the Stokes system (1.1);
this yields new global in time Lq - L estimates.
Let 0 be a domain in R with n>2. We assume that the boundary 812
is (at least) of class C2+P with O</.< 1. Let l<q<cc. By L:(Q) we
denote the closure of CCJSZ) in Lq(Q) where
C~JQ) = {u E CF(i2)n; div u = 0)

and C,(Q) denotes the space of all smooth functions compactly supported
in 0. If R is either
(Ql) R
(522) a bounded domain
(Q3) a halfspace
(524) an exterior domain in KY (i.e., a domain whose complement in
R is a nonempty compact set) and n > 3
then each vector field f~ Ly(Q) is uniquely decomposed as

f =fo+VP (2.10)

with some f. E L:(Q), p E L;b,(@, Vp E Ly(Q), and

Ilwl,~cllfll, and IIPII LqRnB)G Cllf Ilq (2.11)

with C independent off; where B, is an open ball in R and /Iglly denotes

the norm of g in L4(Q)n or Ly(Q); see [17, 3, 311 for the proofs. The
decomposition (2.10) is called the Helmholtz decomposition. By (2.11) the
mapping f++ f. defines a continuous projection P, from Lq(sZ) onto
L:(Q) which we call the Helmholtz projection. P, is independent of q on
C,(Q); therefore we sometimes suppress the subscript q.
We now recall important properties of the Stokes operator A, = -P, A
with dense domain

D(A,)= {~EL~(sz);~~~,~EL~(Q)~; i di, j+2, wlaa=O)

in L;(B), where ai= a/&xi and d = C= i a,? is the Laplacian. By a priori

estimates [l] we know that the Stokes operator A = A, is a densely
defined closed operator in the Banach space X= L:(Q) when one of
(al)-(524) holds. When Q is bounded, we know A has a bounded inverse
(cf. [ll, 271). Since the dual operator AZ = A,. with l/q + l/q= 1 (see
[ 17, 3, 31]), the injectivity of A,, (see [3, 231) implies that R(A,) is dense in
X (cf. Corollary 3.6 in [23] when (Ql) or (524) holds). The nonnegativity
of A is proved by [19, 421 under (522), by [30, 31 under (Q3), and by
[S, 231 under (524). When Q = R, one can express (t + A)) explicitly by
Fourier transforms, so the proof is easy; see, e.g., [23]. Actually these
authors proved the stronger result that -A generates a bounded analytic
semigroup in X which in particular implies the nonnegativity of A. We also
know that
IlAfsll < Cell (2.12)

holds for every E > 0 with some constant C = C(Q, q, E) independent of s.

The estimate (2.12) is first proved by [20] under (522) and by [23] under

(04). We show in the Appendix that (2.12) holds under (523) using results
in [3]. Again, the proof of (2.12) under (Ql) is easy by estimating the
explicit expression of AiS for R; see [23]. Summarizing these results we see
the Stokes operator A satisfies the assumptions of Theorem 2.2.

LEMMA 2.6. For every e > 0 the Stokes operator A, is in &E,(X) with
some K> 1 and with X= L:(R), 1 <q < n3, provided that Q c R satisfies
one of (521))(524).
If Y is a i-convex Banach space, 9 z R an interval, and 1 <q < co, we
know by [35] (see also [13]) that Ly(4, Y) is again [-convex; from (l.lO),
(1.11) we see that a closed subspace of a i-convex Banach space is again
c-convex. Since R is i-convex ((1.10) (1.11) is valid if X=R), we
conclude that Ly( R) = Ly( IR, R) is [-convex. From Ly( I??) =
L4(R, Ly(R- )I) we see by induction that Ly(R) is [-convex. Since
Ly(Q) may be regarded as a closed subspace of Ly(Rn)n, Ly(Q) is
c-convex. Therefore, L;(Q) is c-convex as a closed subspace of Ly(Q). By
Lemma 2.6 we may apply Theorem 2.3 with X= L:(Q), A = A, and obtain
Ly - L estimates of solutions of the Stokes system (1.1).

THEOREM 2.7. Let 1 < q < co, 1 <s < 00, 0 < Tb co and assumefor
Q c R one of the cases(Ql)-(524). Then, for every f~ L(0, T; L:(Q)) and
a E w 1A,sthere exists a unique solution u of the Stokes equation
u+A,u=f for a.e. tE(0, T), u(O)=a (2.13)
satisfying the properties

u E L(0, T,; D(A,)) for all O< T,< T with T,< 00, (2.14)
u E L(0, T; L;(Q)), (2.15)

JOTIlu(t)lj, dt +IoT IIA,u(t)ll.; dt d C(joT Ilf(t)ll$dt+ ll4$~~-) (2.16)

with C = C(q, s, Q) > 0 independent off, a, and T; in particular, (2.16) also

holds with T= co. Here D: denotes the space D: defined by (2.5) with
The equations (2.13) can be written in the following form:
z-Au+Vp=s, div u=O in Q x (0, T), (2.17)

u=O on an, 24(x,0) =a(x) on 52 (2.18)

(see, e.g., [16, 22, 42, 441); p is the associated pressure.

Next we use an embedding property for the second order derivatives

v2v = tai ajv)t, j= I.2 ,..., n:

llV2~ll,6 w%/VIl,~ UED(A,). (2.19)

This holds with C = C(q, Q) > 0 for 1 < q -K 00 if SzG [w is one of the cases
(521)-(Q3) and for 1 < q<n/2 in the case (524). Indeed, the case (Ql)
follows from [23, Lemma 3.31, the case (Q2) follows from [20, 11, 421, see
also [27], the case (523) is proved in [3], and for the case (524) see [42]
for n = 3 and [IS, 231 for n 2 3.
Applying (2.19) to (2.16) and using Vp = f -u+ Au we obtain the
following variant of Theorem 2.7.
THEOREM 2.8. Assume for Q z Iw one of the cases (Szl )-(Q4), ler
1 < s < co and assume 1 < q < co in the cases (Ql )-(Q3) and 1 < q < n/2 in
the case (524). Then for every f E L(0, T; L:(Q)) and aE Di-L/s,s there
exists a unique solution (u, Vp) of the Stokes system (2.17), (2.18) satisfying
u E L(0, T,; Wz*q(a)n) forall O<T,<T with T,<cQ, (2.20)
au/at, Vp E L(0, T; Ly(Q)), (2.21)

jT I[au/atll;dt + j IIV2ull; dt + j IlVpll; dt Q C ( joT Ilf II; dt + IlalKp~)

0 0 0
with C= C(q, s, 0) > 0, where W2~4(Q) is the usual Lq Sobolev space.
Remark 2.9. Solonnikov [42] proved for n = 3 and q =s a weaker
form of (2.22) since his constant C in the right hand side of (2.22) also
depends on T; so T= cc is impossible in [42]. Our estimates (2.16) (2.22)
have two significant features: (i) the norms have different exponents in
space and time direction and (ii) we may take T = cc which leads to decay
estimates for the solution u as t --) co. If n = 3, the estimate for different
exponents q and s also follows from [42] by using [2, 371 or [lo], but
only with C depending on T. The dependence of C on T is an important
question only for unbounded domains.
Remark 2.10. In [42] Solonnikov proved that Di- J agrees with the
completion of D(A,) in the norm of the Sobolev space W2P2/r,s(Q)n when
n = 3, q = s # 312; if q = s = 312, a weighted norm appears (see [42,
p. 4871). For general q, s the norm of D:- WJ seems to be expressed by
certain Besov space norms (see [43, p. 3211).
Remark 2.11. By Remark 2.4 we may replace Di-, in Theorems 2.7
and 2.8 by the homogeneous space a:-,= 61- i/, with A = A,. In this
case, we have to replace (2.14) and (2.20) by
u E L(0, T; &A,))

ueLS(O, T; @'z,y(Q)n),

respectively, since (2.2) replaces (2.2). Here P?2,q(Q) denotes the

homogeneous Sobolev space, the completion of Cc(Q) in the norm
IIV24q+ II4lLU(U)where U is a bounded neighborhood of the boundary
&Q. When 52 is bounded, it is easy to see that @2,q(Q)= W2*y(Q). The
space f@*~~(Q) is important when Q is an exterior domain since it is
included in the space of distributions.


We apply our Lq - Lp estimates (2.16) and (2.22) to weak solutions of

the Navier-Stokes system

g-Au+(u.V)u+Vp=/, divu-0 in sZx(0, co) (3.1)

u=O on a52, u(x, 0) = u(x) (3.2)

and derive new global a priori estimates for the velocity u and the pressure
p. Our estimates describe a large time behavior of general weak solutions
which need not satisfy energy inequalities. We say u is a weak solution of
(3.1)-(3.2) if

UEL"(0, co;L,@)), VUE L2(0, m;L2(Q)"),

u=o on asz for a.e. t 3 0,

(u,$)di-tj:(Vu,VL;)d~+jom (u.Vu,o)dT

= (a,40))+jmCLu>dz
for all v E Ch( [0, co); CCJQ)),


<gN=J R g(x).Nx)dx for vector fields g and h.

For the existence of weak solutions we need to assume

a E L;(Q) and fE L2(0, co;L@q). (3.5)

We now state the following global in time a priori estimates for weak

THEOREM 3.1. Assume that one of (Ql)-(Q4) holds. Let u be a weak

solution of (3.1)-(3.2) with (3.5). Assume that 1 <q, s< co satisfies n+ 1 =
n/q + 2/s and that f E L w = L(0, co; Ly(Q)) and a E Dip ,.
(i) There is a constant C= C(s, q, 52) such that


IIII + /Iv2~lIy,.~+ llwq,.s d CM
z 4,s


M= Ilf IIqJ+ llw::; ll4:2 + Ildy~~~

where I/f I/q,.7denotes the norm off in Ly,. (A4 is finite by assumptions of J;
u, and a.)
(ii) The pressure p can be chosen so that

llPIlr,3dC~ if 1 + n/r = n/q. (3.8)

Moreover, we have

l141h,pd CM9 n/h*+2/p=n-1, p>s, h*>q (3.9a)

IlWlh,~ 6 CM n/h + 2/p = n, p 2 s, h > q. (3.9b)

Proof: (i) Since u satisfies (3.3) we apply the Gagliardo-Nirenberg

inequality [33] (see also [S, 231 for exterior domains)

lI~ll.dww~ IIW
with nlr=n/2-9, 0~8~ 1, 1 <r< co to get

Il(u, V)ull,,,~ cllw:j; I14:,-,2.y

n + 1 = n/q + 2/s,

where C is independent of u.
Since P, is bounded, (3.10) yields

Il~llq,,~ aw;i; 114:,22/+ Ilf llq.., for F=f- P(u, V)U. (3.11)

By (3.3) and (3.4) we conclude that u solves

u+A,u=F, u(0) = a.

Actually, its Yosida Approximation uk= J,u is in L(0, T, D(A,)) and

u; E L(0, T, L:(Q)) for all 0 < T< co and uk solves

u; + A,u, = J, F, uk() = JkQ,

where J, = (I+ A,/k)- (cf. [38, 321). Applying (2.16) to uk yields

Ilu;ll,,,r+ llAy4y,.,~~ C(IIFIl,,,,+ ll4~:-4, 1 <q,s<oo

since llJkil < C with some constants C, C independent of k, T; here

Ilf IIq,sT denotes the norm of f in L(0, T; Ly(Q)). Letting T-+ cc and
k + cc now yields (3.6) since F is estimated by (3.11). We also obtain (3.7)
if we apply (2.22) to uk instead of (2.16). The restriction 1 <q<n/2 under
(Q4) is satisfied since q satisfying (3.10) is smaller than n/(n - 1) and n > 3.
(ii) Since 1 < q < n/(n - 1) by (3.10), the Poicare-Sobolev inequality
II~+k,ll,dCIIWy, 1 + n/r = n/q (3.12)

with a constant k,. This is well known if Q is bounded or Q = R. When

Q is an exterior domain, this is found in Corollary 2.2 in [23]; see also
[S]. The case where Q is a halfspace can be reduced to the caseQ = R by
symmetric extension of p w.r. to the boundary. The estimates (3.7) and
(3.12) yield (3.8) where p should be replaced by p + k,.
The estimates (3.9a), (3.9b) follow from (3.6) and the interpolation
inequalities stated below. This proves Theorem 3.1.

LEMMA 3.2. Assumethatl<q<n/2,q<h<h*<oo,andl<s<p<co.

Assume that
n/q + 2/s = 2 + n/h * + 2/p, n/h*=n/h-1. (3.13)

Then there are constants Ci = C, (s, q, Q) (j = 1, 2) such that

II4l/2*.pd C, IlWlh,, d C~(Il~ll~,~+IIA4,,,+ IlallD:-~~~~~) (3.14)

for all u satisfying u E L(0, T; D(A,)) for all 0 < T< co with u E Ly,,
A,u E Ly., and u(O) = a EDim /*..
Proof: Step 1. We first assumea = 0. In this case u(t) is expressed as

u(t)=jeP~(u+Au)dr. (3.15)

Since eprA is a bounded analytic semigroupsin L:(Q) [3, 30, 42, 19, 5, 231
we have
IIAaePA II < C,tr, t > 0, (3.16)

where c(> 0. Applying A to (3.15) now yields

For 0 <a < 1 the Hardy-Littlewood inequality (see, e.g., [43, p. 1401)

II~~4~)llY.P6 Ca(l141y,s+ II4l,,.J (3.17)

with 1 - CI+ l/p = l/s; this inequality is also valid for p = s and a = 1. By
(3.13) we obtain 2c( + n/h* = n/q.
We now apply an embedding inequality (Corollary 6.7 in [23])

llwllh d CIIA- %,, 1 < q < n/2,

with 2a+n/h=n/q- 1, 1/2<a< 1, and obtain

lIA12Ullh< CIIA%llq. (3.18)

Since 1 < q < n/2 we see 1 < h < n, so applying

IIWI, 6 CllA*4~, for 1 <h<n (3.19)

to (3.18) yields

IlWlhd Cww,. (3.20)

The estimate (3.19) was first proved by [23] for 1 <h < max(n/2,2) and
later extended by [4] to the case 1 < h < n. The estimates (3.17) and (3.20)
yield (3.14); the first inequality of (3.14) follows from the Sobolev
inequality since U= 0 on 852. For the Sobolev inequality for exterior
domains, see, e.g., Corollary 2.2 in [23].
We note that we do not need to apply (3.19) if we are only interested in
the estimates for U, not in the derivatives Vu. Indeed, instead of (3.20) we

llull~ d CIIA4,, 1 < q < n/2

by Corollary 6.7 in [23]. This together with (3.17) yields

lI4/r* G C(ll~ll~,,+ llA4,.J

Step 2. For every i E D~ Is, there exists some w E L( - l,O; D(A,))
with w E L( - l,O; X), w(0) 2 a, w( - 1) = 0, and

IIwIl.~~-,,,;,j+ lI~y~II~~~~~,~;~~~~lI~ll~~-~~~~~ (3.21)


with c=c(q,s,Q)>O, where X=L;(Q). Let 0~C([w) satisfy e(t)=0 for

- co<t<l,0(t)=lfor06t<coand0<8<1andset

w(t) = 19(t)erAu, A=A,. (3.22)

This w satisfies the desired properties. Indeed, we see

which yields (3.21). By the definition (3.22) we see w(O) = a and w( - 1) = 0.

Step 3. We now prove (3.14) for general a in Dip. We construct
an extension ii E L( - 1, 00 ; D(A,)) of u by defining ii(t) = u(t) for t b 0 and
ii(t) = w(t) for - 1 6 t < 0 with w in (3.22). Applying Step 1 to ii where the
interval [IO, co) is now replaced by [ - 1, co), we now obtain (3.14) since
w and Aw are controlled by (3.21).
Remark 3.3. As is well known (see [21, 36, 38]), when n =2 our
estimate (3.9a) is enough to conclude everywhere regularity of weak
solutions. When n 3 3 we do not known everywhere regularity. In the case
n = 3 our estimate (3.8) with s = 5/3 and q = 15/14 yields

p E L53(Q x (0, cc )). (3.23)

As is pointed out in [9] the estimate (3.23) for the pressure plays an
important role to improve the partial regularity theory of suitable weak
solutions, i.e., weak solutions satisfying a local version of energy inequality.
In [9], (3.23) is proved when .Q = R3 and conjectured for a general
domain. A weaker estimate
p E L53(sz x (0, T)) (3.24)

for every T-C cc is proved by [40] when Q is bounded or an exterior

domain. Since (3.3) and (3.24) imply that u and p are small for large 1x1,
using the partial regularity theory [9] yields regularity of suitable weak
solutions for large 1x1 for exterior domains (see [39, 281). Our estimates
also yield the regularity of suitable weak solutions for large t apart from
the boundary even when Sz is an exterior domain or a halfspace. Smoothness
for large t without restriction near the boundary is proved in [32] by a
different method. See also [ 18, 281 where also regularity results are proved
for large 1x1.

Remark 3.4. We now compare our estimates with other known a priori
estimates for weak solutions when n = 3. For a weak solution satisfying an
energy inequality, Foias, Guillope, and Temam [ 151 proved, among other


for T < co when Q is bounded (see also [ 143 ) and derived

s, IMt)ll ccdt < ~0.

Interpolating (3.26) and llz.l12,3c< cc in (3.3) yields

soTIlu(t) dt < ~0, 4/r+2/p=2, i-22, 2ap31.

As a local estimate, this is better than our (3.9a). However, ours are global
in time while T cannot be infinite in (3.25), (3.26). Moreover our estimate
holds for every weak solution while (3.25) and (3.26) only hold for weak
solutions satisfying an energy inequality. Recently, Constantin [ 121
constructed a weak solution satisfying
llMt)ll: dt < ~0, (3.27)

by studying the vorticity equation under periodic boundary conditions in

the space variables. As a local estimate this is better than (3.7):

scIkW)ll; dt < ~0, q=5/4;

however, the bound in (3.27) is not uniform in T, so the estimate (3.27) is

not global in time.
Our Theorem 3.1 describes the large time behavior of weak solutions.
Even when Q is unbounded, many interesting results are now available;
see [4] and references cited there, for example, [ 18, 28, 291. However,
there is a big difference between ours and known results and neither results
imply the others because

(i) there are no global in time estimates for the pressure in the
(ii) our estimates hold for any weak solutions which may not satisfy
energy inequalities;

(iii) the external force f is allowed in our theory;

(iv) in the literature the algebraic decay order of Ilu(t)lih* as t + co
is studied while we estimate its decay by an integral norm.
Let us compare our estimate (3.9a) with a result in the literature. For
UE L(Q)n L:(Q) with 1 <r <n/(n - l), r < 2n/(n +2), there is a weak
solution u(x, t) of (3.1t(3.2) (assumingf=O) such that Ilu(t)ll,=~(t-~) as
f -+ 00 with y = (n/r - n/k)/2 for r 5 k < 2; see [4, Theorem C]. If we take
r=n/(n - I), this implies

Ib(t) = 4-q
with n/k + 2/p = n - 1. However, it does not imply (3.9a).



We shall prove a bound (2.12) when Q is a halfspace

KY; = {(x,, .... x,)ER;X,>O), n 3 2.

THEOREM Al. Let A, be the Stokes operator in X = LY,(R; ). Then

A$ E -Y(X) and for every F> 0 there is C = C(n, q, E) such that

IIA i /I 6 Ce

holdsfor all s E R.
We first list some properties of fractional powers of general nonnegative

LEMMA A2. Let A be a nonnegative operator in a Banach space X. It

(i) (aA) = &A for a > 0, s E R
(ii) (6 + A)f-t Af in X as 6 -0 (6 >0) provided that
j-c D(A) n R(A).
Proof of Theorem Al. Our argument is based on a scaling transforma-
(s,S)(x)=f(XlP)7 .fEcp",), P>O
and the estimate (3.12) in [3]:

]/(A, + I)[/ < Cetm, -ldRez<O.


This estimate yields

II(A, + Z)ll < Ce, SER (1)

since (A4 + Z)zf is analytic in z, -l<Rez<l, forfED(A,)nR(A,).
By the definition of A, we observe that

p*A, = S,A,S,, (p2A,)p=S,A,S,

(1+p*AJ=SJl+ A,)- S,

as in the proof of (3.13) in [3]. These relations yield

(p2A, + I) = S,(A, + I) S/y

by the definition [26, (1.3) and (4.11)] of A. Since

IIs, II = P, IlS,II =p--q

the estimate (1) yields
JI(p*A, + Z)ll < Cell.
($A,+ z)S=~*(Ay+~~*);.~

by Lemma A2(i) we now obtain

II(A, + pp)II < Cell.

Applying Lemma A2(ii), we have

IIA~fll = lim ll(A, + p-)fll d CeBis llfll


for all f E D(A,) n R(A,) which completes the proof of Theorem Al since
D(A,) n R(A,) is dense in X= Lz(Iw: ).
Proof of Lemma A2. We recall the definition of Af for


where m is an arbitrary fixed positive number; see [26, p. 3153. Of course,

the definition is independent of the choice of m. We replace A by aA in (2)

and change the variable of integration by 2 = az. A calculation shows (i) on

D(A) n R(A). By a density argument we obtain (i) as operators.
It remains to prove (ii). We replace A by A + 6 in (2) and observe that
the norms of all integrals are dominated by integrable functions independent
of 6. Since we have

(6+A)-f+A-f, (A+d+A)pf-+(i+A)pf in X

as 6 -+ 0 for f~ R(A), Lebesguesconvergence theorem yields (ii).


The following proof is basically parallel to that of Theorem 2.1 in [ 131.

Step 1. Using the notations in Theorem 1.1, we first observe a density
property. The set

Do= { [A(t+A)-]2(t+A)-*[B(z+A)-1]*(T+B)--2u; t>o, z>o, UEX}

is a dense subset of X. This easily follows by the following standard

arguments: (t+A))XcX is dense for t>O, t(t+A)-o-,0, and
A (t + A ) ~ u -+ u as t --) 0 for all u EX, the corresponding facts hold with A
replaced by B.
Step 2. Using Komatsus representation for A and B ([26]; see
also [23]) it easily follows that AZBWu= BAzu for all u ED,, and for all z,
we@ with -l<RezQl, -l<Rew<l.
Step 3. We use the same integrals as in [13],
A - iy B ~ 1 + i,
u dy,
sin rciy
A ~ 1+ iJBi
@,;,u=f j u d! (YER)

with E>O, UED,. The convergence of these integrals follows from the
assumptions which lead to the estimate

[sin njyl PI I(A-JII IIBYII < Ce(+o)lyl,-nlyl, I Yl2 1

where 8 + 0 < 7~.The argument on the Hilbert transform in [13] shows
that the limits
lim @o,Eu, s-0lim @,,cv,
lim BQo,, u, lim A@,,,u
s-0 c-0

exist in X and that

iye0 B@O,EVll
G Cll4l, II60lim A@,,,LJIId Cll4l (1)

for all u E Do, where C = C( 8, cr, K, X) > 0.

Step 4. An elementary calculation as in [13] shows that

B@,,,v = -A@,,,u, (2)

(~i,mo@o,,v)+~B~lu=(lim @,,,~)+fA~o (3)
c- 0

for all UED,. The same relation also holds for all u = Ah or u = Bh with
h E Do. The expression in (3) will be denoted by So in these cases.
Step 5. We use the closedness of A and B and Step 2 to obtain

lim @o,Eu E D(B), B( lim @O,Etl)= lim @o,EBu,

i:- 0 EO s-0

lim @l,c~ E D(A), A( lim @,,E~) = lim @,,,Au

c-0 c+ 0 1.- 0

for all UE Do. So we conclude that SUE D(A) n D(B) and BSv = SBu,
ASu = SAu for all u E Do. From (1) it follows that

IlBSuIId Cllull, IlASull d Cl14

for all u E Do, where C = C(0, (T, K, X) > 0.
Step 6. Let u E Do. Then by (2), (3), and Step 5 we obtain
= (;Fo A@,,,v) + iv + (,li_mo
B@,,,u) + ;u

= ;u + fu = u

and SAv + SBv = (A + B) Su = u. Let u = So, v E Do. Then by Step 5 it


IlAull = IIAW dCII4 = Cll(A +B)Svll =Cll(A +@4,

IlBuIl=llBSvIl ~C~IY~~=CII(A+B)S~I~=C~~(A+B)~~~.

Step 7. Let f~ X. Since Do is dense in X we find a sequence

USEDo,j= 1, 2, .... such that lim,, a, ui = f in the norm of X. Setting ui = Su,
we see by Step 5 that uj E D(A) n D(B) and by Step 6 that

IIA(u, - uj)ll + IIB(u, - uj)ll G 2CIl(A + B)(u/c- Uj)ll


for k, j= 1, 2, ... . So (3) is a Cauchy sequence in the norm liAull + llBull

and there is some u E D(A + B) such that Au + Bu =f and IlAull + IIBull <
2Cllfll. The theorem is proved.


1. S. AGMON, A. DOUGLIS, AND L. NIRENBERG, Estimates near the boundary for solutions of
elliptic partial differential equations satisfying general boundary conditions, I, Comm.
Pure Appl. Math. 12 (1959), 623-121; II, Comm. Pure Appl. Math. 17 (1964), 35-92.
2. A. BENEDEK, A. P. CALDER~N, AND R. PANZONE, Convolution operators on Banach space
valued functions, Proc. Nat. Acad. Sci. U.S.A. 48 (1962), 356-365.
3. W. BORCHERS AND T. MIYAKAWA, L*-decay for the Navier-Stokes flow in halfspaces,
Math. Ann. 282 (1988), 139-155.
4. W. BORCHERS AND T. MIYAKAWA, Algebraic L-decay for Navier-Stokes flows in exterior
domains, Acla Math. 165 (1990), 189-227.
5. W. BORCHERS AND H. SOHR, On the semigroup of the Stokes operator for exterior
domains, Math. Z. 196 (1987), 415425.
6. J. BOURGAIN, Some remarks on Banach spaces in which martingale difference sequences
are unconditional, Ark. Mat. 21 (1983), 163-168.
7. D. L. BURKHOLDER, A geometric condition that implies the existence of certain singular
integrals of Banach-space-valued functions, in Conference on Harmonic Analysis in
Honor of Antoni Zygmund, Chicago, 1981, pp. 27&286, Wadsworth, Belmont, 1983.
8. P. BUTZER AND H. BERENS, Semi-Groups of Operators and Approximations, Springer,
Berlin/Heidelberg/New York, 1967.
9. L. CAFFARELLI, R. KOHN, AND L. NIRENBERG, Partial regularity of suitable weak solutions
of the Navier-Stokes equations, Comm. Pure Appl. Mafh. 35 (1982), 771-831.
10. P. CANNARSA AND V. VESPRI, On maximal Lp regularity for the abstract Cauchy problem,
Boll. Un. Mat. Ital. (6) 5-13 (1986), 165-175.
11. L. CATTABRIGA, Su un problema al contorno relative al sistema di equazioni di Stokes,
Rend. Sem. Mat. Univ. Padova 31 (1961), 308-340.
12. P. CONSTANTIN, Remarks on the Navier-Stokes equations, preprint.
13. G. DORE AND A. VENNI, On the closedness of the sum of two closed operators, Math. Z.
196 (1987), 189-201.
14. G. F. D. DUFF, Derivative estimates for the Navier-Stokes equations in a three
dimensional region, Acta Mafh. 164 (1990), 145-210.
15. C. FOIAS, C. GUILLOP~, AND R. TEMAM, New a priori estimates for Navier-Stokes
equations in dimension 3, Comm. Partial Differential Equations 6 (1981), 329-359.
16. H. FUJITA AND T. KATO, On the Navier-Stokes initial value problem, I, Arch. Rational
Mech. Anal. 16 (1964), 269-315.
17. D. FUJIWARA AND H. MORIMOTO, An L,-theorem of the Helmholtz decomposition of
vector fields, .I. Far. Sci. Univ. Tokyo Sect. IA Math. 24 (1977), 685-700.
18. G. GALDI AND P. MAREMONTI, Monotic decreasing and asymptotic behavior of the
kinematic energy for weak solutions of the Navier-Stokes equations in exterior domains,
Arch. Raiional Mech. Anal. 94 (1986), 253-266.
19. Y. GIGA, Analyticity of the semigroup generated by the Stokes operator in L, spaces,
Math. Z. 178 (1981), 297-329.
20. Y. GIGA, Domains of fractional powers of the Stokes operators in L, spaces, Arch.
Rational Mech. Anal. 89 (1985), 251-265.
21. Y. GIGA, Solutions for semilinear parabolic equations in Lp and regularity of weak
solutions of the Navier-Stokes system, J. Differenfial Equations 62 (1986), 186212.

22. Y. GIGA AND T. MIYAKAWA, Solutions in L, to the Navier-Stokes initial value problem,
Arch. Rational Mech. Anal. 89 (1985), 267-281.
23. Y. GIGA AND H. SOHR, On the Stokes operator in exterior domains, J. Fat. Sci. Univ.
Tokyo Sect. IA Math. 36 (1989), 103-130.
24. Y. GIGA AND H. SOHR, Note on the Cauchy problem in Banach spaces with applications
to the Navier-Stokes equation in exterior domain, unpublished preprint, 1988.
25. H. IWASHITA, Ly- L estimates for solutions of non-stationary Stokes equations in
exterior domains and the Navier-Stokes initial value problems in L, spaces, Math. Ann.
285 (1989), 2655288.
26. H. KOMATSU, Fractional powers of operators, Pacific J. Math. 19 (1966), 2855346.
27. 0. A. LADYZHENSKAYA, The Mathematical Theory of Viscous Incompressible Flow,
Gordon & Breach, New York, 1969.
28. P. MAREMONTI, Partial regularity of a generalized solution to the Navier-Stokes equations
in exterior domains, Comm. Math. Phys. 110 (1987), 75-87.
29. K. MASUDA, Weak solutions of Navier-Stokes equations, T6hoku Mafh. J. 36 (1984),
30. M. MCCRACKEN, The resolvent problem for the.Stokes equations on halfspaces in L,,
SIAM J. Math. Anal. 12 (1981), 201-228.
31. T. MIYAKAWA, On nonstationary solutions of the Navier-Stokes equations in exterior
domains, Hiroshima Math. J. 12 (1982), 115-140.
32. T. MIYAKAWA AND H. SOHR, On energy inequality, smoothness and large time behavior
in L2 for weak solutions of the Navier-Stokes equations in exterior domains, Math. Z. 199
(1988), 455478.
33. L. NIRENBERG, On elliptic partial differential equations, Ann. Scuola Norm. Sup. Piss Cl.
Sci. (3) 13 (1959), 115-162.
34. J. P~iiss AND H. SOHR, On operators with bounded imaginary powers in Banach spaces,
Math. Z. 203 (1990), 429452.
35. J. L. RUBIO DE FRANCIA, Martingale and integral transforms of Banach space valued
functions, in Probability and Banach Spaces (Proceedings, Zaragoza, 1985),
pp. 195-222, Lecture Notes in Math., Vol. 1221, Springer, Berlin/Heidelberg/New York,
36. J. SERRIN, On the interior regularity of weak solutions of the Navier-Stokes equations,
Arch. Rational Mech. Anal. 9 (1962), 187-195.
37. P. E. SOBOLEVSKII, Coerciveness inequalities for abstract parabolic equations, Dokl. Akad.
Nauk SSSR 157 (1964), 52-55. [In Russian]
38. H. SOHR, Zur Regularitltstheorie der instationaren Gleichungen von Navier-Stokes,
Math. Z. 184 (1983), 3599375.
39. H. SOHR AND W. VON WAHL, A new proof of Lerays structure theorem and the
smoothness of weak solutions of Navier-Stokes equations for large 1x1, Bayreuth. Math.
Schr. 20 (1985), 153-204.
40. H. SOHR AND W. VON WAHL, On the regularity of the pressure of weak solutions of
Navier-Stokes equations, Arch. Mafh. 46 (1986), 428-439.
41. H. SOHR, W. VON WAHL, AND M. WIEGNER, Zur Asymptotik der Gleichungen von
Navier-Stokes, Nachr. Akad. Wiss. Giittingen 3 (1986) 1-15.
42. V. A. SOLONNIKOV, Estimates for solutions of nonstationary Navier-Stokes equations,
J. Soviet Math. 8 (1977), 467-523.
43. H. TRIEBEL, Interpolation Theory, Function Spaces, Differential Operators, North-
Holland, Amsterdam/New York/Oxford, 1978.
44. W. VON WAHL, The Equations of Navier-Stokes and Abstract Parabolic Equations,
Aspects of Math., Vol. E8, Vieweg, Braunschweig/Wiesbaden, 1986.