Lecture Notes 156.01556

Lecture Notes OxPDE-14/01
Lecture Notes on Regularity Theory for the

Navier-Stokes equations
by
G. Seregin
University of Oxford
Oxford Centre for Nonlinear PDE

Mathematical Institute
University of Oxford
Andrew Wiles Building
ROQ, Woodstock Road
Oxford, UK
OX2 6GG March 2014
Lecture Notes on Regularity Theory for the
Navier-Stokes equations
G. Seregin
Oxford University
March 2, 2014
2
Context
Chapter I Preliminaries
1.1. Notation
1.2. Newtonian potential
1.3. Equation div u = b
1.4. Nečas imbedding theorem
1.5. Spaces of solenoidal vector fields
1.6. Linear functionals vanishing on divergence free vector fields
1.7. Helmholtz-Weyl decomposition
Chapter II. Linear stationary problem
2.1. Existence and uniqueness of weak solutions
2.2. Coercive estimates
2.3. Local regularity
2.4. Further local regularity results, n = 2, 3
2.5. Stokes operator in bounded domains
Chapter III. Non-linear stationary problem
3.1. Existence of weak solutions
3.2. Regularity of weak solutions
Chapter IV. Linear non-stationary problem
4.1. Derivative in time
4.2. Explicit solution
4.3. Cauchy problem
4.4. Pressure field. Regularity
4.5. Uniqueness results
4.6. Local interior regularity
4.7. Local boundary regularity
Chapter V. Non-linear non-stationary problem
5.1. Compactness results for non-stationary problems
5.2. Auxiliary problem
3
5.3. Weak Leray-Hopf solutions

5.4. Multiplicative inequalities and related questions
5.5. Uniqueness of Weak Leray-Hopf solutions. 2D case
5.6. Further properties of weak Leray-Hopf solutions
5.7. Strong solutions
Chapter VI. Local regularity theory for non-stationary Navier-
Stokes equations
6.1. Notation
6.2. ε-regularity theory
6.3. Bounded ancient solution
6.4. Mild bounded ancient solutions
6.5. Liouville type theorems
6.5.1. LPS quantities
6.5.2. 2D case
6.5.3. Axially symmetric case with no swirl
6.5.4. Axially symmetric case
6.6 Axially symmetric suitable weak solution
6.7. Backward uniqueness for Navier-Stokes equations
Chapter VII. Behaviour of L3 -Norm
7.1. Main result
7.2. Estimates of scaled solutions
7.3. Limiting procedure
Appendix A. Backward Uniqueness and Unique Continuation
A.1. Carleman-type inequalities
A.2. Unique continuation across spatial boundaries
A.3. Backward uniqueness for heat operator in half space
Appendix B. Lemarie-Riesset Local Energy Solutions
B.1. Introduction
B.2. Proof of Theorem 1.4
B.3. Regularised problem
B.4. Passing to limit and proof of Proposition 1.6
4
B.5. Proof of Theorem 1.5

B.6. Density
Bibliography
Foreword
The Lecture Notes are based on the TCC course given by me in Trinity
Terms of 2009-2011. Chapters I-III contains material discussed in Trinity
Term of 2009 (16 hours in total), Chapters IV-V contains lectures of 2010
(16 hours), and finally, lectures of 2011 are covered by Chapter VI (16 hours).
Chapters I-V can be regarded as an Introduction to the Mathematical
Theory of the Navier-Stokes equations, relying mainly on the classical PDE’s
approach. First, the notion of weak solutions is introduced, then their exis-
tence is proven (where it is possible), and, afterwards, differentiability prop-
erties are analyzed. In other words, we treat the Navier-Stokes equations
as a particular case, maybe very difficult, of the theory of nonlinear PDE’s.
From this point of view, the Lectures Notes do not pretend to be a complete
mathematical theory of the Navier-Stokes equations. There are different ap-
proaches, for example, more related to harmonic analysis, etc. Corresponding
list of references (incomplete, of course) is given at the end of the Lecture
Notes.
Finally, Chapters VI and VII contains more advanced material, which
reflects my scientific interests.
Chapter 1
Preliminaries
1.1 Notation
Let us denote by Ω a domain (open connected set) in Rn . Then, C0∞ (Ω; Rm ) is
set of all infinitely differentiable functions from Ω into Rm , having a compact
support in Ω. If m = 1, we use abbreviation C0∞ (Ω).
The Lebesgue space is be denoted as Lp (Ω) and it is be endowed with the
standard norm Z p1
p
kf kp,Ω = |f (x)| dx
Ω
if 1 ≤ p < ∞ and
kf k∞,Ω = ess sup |f (x)|
x∈Ω
if p = ∞.
Lemma 1.1. Let 1 ≤ p < ∞. Then, Lp (Ω) = [C0∞ (Ω)]Lp (Ω) , i.e., Lp (Ω) is
the completion of C0∞ (Ω) in Lp (Ω).
In what follows, we always assume that the exponent of integrability is
finite unless otherwise is specially indicated.
We say that a distribution u, defined in Ω, belongs to the Sobolev space
Wsk (Ω) if and only if all its weak derivatives up to order k are integrable in
Ω with the power s. The norm in this space is defined as
k
X
kukWsk (Ω) = k∇i uks,Ω < ∞.
i=0
5
6 CHAPTER 1. PRELIMINARIES
We also let ◦
k ∞ W (Ω) k
W s (Ω) = [C0 (Ω)] s .
It is said that a distribution u, defined in Ω, belongs to the space Lks (Ω)
if and only if all its weak derivatives of order k are integrable in Ω with the
power s. The norm in this space is defined as
kukLks (Ω) = k∇k uks,Ω < ∞.
A non-trivial statement about spaces Lks (Ω) is as follows.
Theorem 1.2. Lks (Ω) ⊂ Ls,loc (Ω).
Proof We have a distribution T such that
Z
g(x)ϕ(x)dx = −T (∇ϕ)
Ω
for ϕ ∈ C0∞ (Ω) with g = (gi ) ∈ Ls (Ω). Our aim is to show that is in fact a
regular distribution and moreover, there exists a function u ∈ Ls,loc (Ω) such
that T = Tu .
Consider a subdomain Ω0 b Ω, i.e., a bounded subdomain Ω0 ⊂ Ω such
that the closure of Ω0 belongs to Ω. Let 0 < % < dist(Ω0 , ∂Ω). Define a
linear functional l : L1 (Ω0 ) → R in the following way
l(ψ) := T (ψ% )
for ψ ∈ L1 (Ω0 ) where
Z
ψ% (x) = $% (x − y)ψ(y)dy
Ω0
and $% is a standard mollifier. It is easy to check that l is a bounded

functional and thus there exists a function u% ∈ L∞ (Ω0 ) such that
Z
l(ψ) = u% (x)ψ(x)dx
Ω0
for all ψ ∈ L1 (Ω0 ). Next, for any ϕ ∈ C0∞ (Ω0 ) we have

Z Z
−T (∇ϕρ ) = gϕρ dx = g% ϕdx.
Ω Ω
1.1. NOTATION 7
On the other hand, by known properties of mollification we find

Z
−T (∇ϕρ ) = −T ((∇ϕ)ρ ) = − u% ∇ϕdx.
Ω0
The latter means that

g% = ∇u%0
where u%0 = u% − [u% ]Ω0 and
Z
% 1
[u ]Ω0 := u% dx.
|Ω0 |
Ω0
By Poincare inequality
ku%0 ks,Ω0 ≤ c(n, s, Ω0 )kg % ks,Ω0 ≤ c(n, s, Ω0 )kgks,Ω .
Without loss of generality, we may assume that
u%0 * u0
in Ls (Ω0 ). And thus

g = ∇u0
in Ω0 .
Next, we take a sequence of domains such that Ωk b Ω, Ωk b Ωk+1 for
any natural k and
[∞
Ω= Ωk .
k=0
Now, let us described a required function u. We let u = u0 in Ω0 . Then,

repeating the above procedure, we find a function u1 ∈ Ls (Ω1 ) such that
g = ∇u1 in Ω1 . It is easy to see that u1 − u = C0 on Ω0 . Then we let
u = u1 − C0 on Ω1 \ Ω0 . Obviously, g = ∇u in Ω1 and u ∈ Ls (Ω1 ). In the
same way, we define function u in Ω2 and so on.
◦ k
We also can introduce the subspace Lks (Ω) = [C0∞ (Ω)]Ls (Ω) of the space
Lks (Ω). In general, the following statement
◦
{u ∈ Lks (Ω) and kukLks (Ω) = 0} ⇒ u = 0
is NOT true. To see that, let us consider a standard cut-off function ϕ ∈

C0∞ (Rn ), having the following properties:
0 ≤ ϕ(x) ≤ 1 x ∈ Rn ,
ϕ(x) = 1 x ∈ B(1), ϕ(x) = 0 x ∈

/ B(2),
ϕR (x) = ϕ(x/R).
Here, B(R) is a ball of radius R centered at the origin.
Lemma 1.3. ◦
1 ∈ L22 (R3 ).
Proof
Z Z
2 2 1 3
|∇ ϕR | dx = |∇2 ϕR |2 dx ≤ c R →0
R4
R3 B(2)\B(1)
as R → ∞. ◦
As usual, the equivalence classes are introduced in L22 (Ω) so that
u1 ∼ u2 ⇔ ku1 − u2 kLks (Ω) = 0.
Proposition 1.4. For bounded domains Ω, the Friedrichs inequality is valid:
kukWsk (Ω) ≤ ckukLks (Ω)
for any u ∈ C0∞ (Ω) with a positive constant independent of u, i.e.,

◦ ◦
k k
W s (Ω) = Ls (Ω).
1.2 Newtonian Potential

It is known that the fundamental solution to the Laplace equation is
1 1
E(x) =
ωn n(n − 2) |x|n−2
if n ≥ 3 and
1 1
E(x) = ln
2π |x|
1.2. NEWTONIAN POTENTIAL 9
if n = 2, where ωn is the volume of unit ball in Rn .

Given function f : Rn → R, we define the Newtonian potential of f as
the following convolution:
u=E?f
or Z
u(x) = E(x − y)f (y)dy.
Rn
Proposition 2.1. Let f ∈ Lp (Rn ) with 1 < p < ∞ and u = E ? f . The

following statements are true:
R R
(i) |∇2 u|p dx ≤ c(n, p) |f |p dx,
Rn Rn
◦
(ii) u ∈ L2p (Rn ),
(iii) 4u = −f
n
in R .
Proof (i) follows from the theory of singular integrals. (iii) follows from
(i),(ii), and from the classical PDE theory.
Let us prove (ii), assuming that n ≥ 3. By Lemma 1.1, there exists a
sequence fm ∈ C0∞ (Rn ) such that fm → f in Lp (Rn ). Since supp fm is a
compact set in Rn ,
c(m, i)
|∇i um (x)| ≤ n−2+i
|x|
for all x ∈ Rn , for i = 0, 1, 2, for all m = 1, 2, ..., and for some positive c(m, i).
Here, um = E ? fm .
We claim that
◦
um ∈ L2p (Rn ).
Indeed, we have
Z h Z
2 p
|∇ (ϕR um − um )| dx ≤ c |∇2 um |p dx+
Rn Rn \B(R)
Z Z
1 p 1 i
+ p |∇um | dx + 2p |um |p dx ≤
R R
B(2R)\B(R) B(2R)\B(R)
Z h 1 Rn 1 Rn i
≤c |∇2 um |p dx + C(m) p (n−1)p + 2p (n−2)p → 0
R R R R
Rn \B(R)
as R → ∞ for each fixed m.

On the other hand, by (i), we have
k∇2 u − ∇2 um kp,Rn ≤ ckf − fm kp,Rn → 0
as R → ∞. This implies (ii).

Particular cases
1. Let f (x0 , xn ) = −f (x0 , −xn ), where x0 = (x1 , x2 , ..., xn−1 ). Then
u(x0 , xn ) = −u(x0 , −xn )
and u, u,n = ∂u/∂xn , and u,nn are in Lp,loc (Rn ) that implies u(x0 , 0) = 0.
So, the Newtonian potential u solves the following Dirichlet problem in half-
space:
4u = −f (1.2.1)
in Rn+ = {x = (x0 , xn ) : xn > 0},
u(x0 , 0) = 0
for any x0 .
2. The same arguments show that if f (x0 , xn ) = f (x0 , −xn ) then u solves
the Neumann boundary value problem, i.e., it satisfies (1.2.1) and the Neu-
mann boundary condition
u,n (x0 , 0) = 0
for any x0 .
1.3 Equation div u = b

We start with the simplest case Ω = Rn .
Proposition 3.1. Let 1 < s < ∞. Given b ∈ Ls (Rn ), there exists u ∈
◦
L1s (Rn ) with the following properties:
(i) div u = b
in Rn ,
1.3. EQUATION DIV U = B 11
(ii) k∇uks,Rn ≤ c(s, n)kbks,Rn .

◦
Proof We let h = E ? b. By Proposition 2.1, ∇h ∈ L1s (Rn ). If we let
u = −∇h, then, by the same statement, we have
k∇uks,Rn = k∇2 hks,Rn ≤ c(s, n)kbks,Rn
and
div u = −div ∇h = −4h = b.
In the case of the half-space, i.e., Ω = Rn+ := {x = (x0 , x3 ) : x0 ∈ R2 , x3 >
0}, we have
Proposition 3.2. Let 1 < s < ∞. Given b ∈ Ls (Rn+ ), there exists u ∈

◦
L1s (Rn+ ) with the following properties:
(i) div u = b
in Rn+ ,
(ii) k∇uks,Rn+ ≤ c(s, n)kbks,Rn+ .
Remark 3.3. The above vector-valued function u satisfies the homogeneous

boundary condition u|xn =0 = 0 in the sense of traces in Sobolev spaces.
Proof of Proposition 3.2 To show the essence of the matter, let us

consider a special case n = 3 and s = 2.
Let b ∈ C0∞ (R3+ ) and b̂ is the even extension of b to R3 . Clearly, b̂ ∈
C0∞ (R3 ). Letting h = −E ? b̂, we see that
4h = b̂ = b in R3+ , h,3 |x3 =0 = 0 (1.3.1)
and
k∇2 hk2,R3+ ≤ ckbk2,R3+ . (1.3.2)
The idea is to look for u in the form
u = ∇h + rot A,
where A is unknown vector field. Obviously,
div u = 4h = b in R3+ .
Equations for A is coming from condition u|x3 =0 = 0 that leads to the fol-
lowing relations
rot A = −∇h
at x3 = 0. We are seeking A, satisfying additional assumptions:
A|x3 =0 = 0, A3 ≡ 0 in R3+ .
So, the main equations for A = (A1 , A2 , 0) are:
Aα,3 (x0 , 0) = Bα (x0 ) α = 1, 2,
where B1 (x0 ) = h2 (x0 , 0) and B2 (x0 ) = −h,1 (x0 , 0) are known functions.
Now, we are going to exploit arguments, which are quite typical for the
theory of traces for functions from Sobolev spaces. This theory suggests to
seek A in the form:
Z2
Aα (x0 , x3 ) = x3 Bα (x0 + y 0 x3 )K(y 0 )dy 0 ,
R
where a function K ∈ C0∞ (R2 ) is supposed to obey the following conditions:

Z
0 0 0 0 0
K(x ) = 0 x ∈ 2
/ B = {x ∈ R : |x | < 1}, K(y 0 )dy 0 = 1.
R2
Now, our aim to show that functions
u1 = h,1 + A2,3 ,
u2 = h,2 − A1,3 ,
u3 = h,3 + A1,2 − A2,1 ,
with the described above A, satisfy all the requirements. This can be done
by direct calculations.
Indeed,
Z Z
∂Bα 0 0
Aα,3 (x) = x3 (x +y x3 )yβ K(y )dy + Bα (x0 +y 0 x3 )K(y 0 )dy 0 . (1.3.3)
0 0
∂zβ
R2 R2
Observing that
∂Bα ∂Bα 1
=
∂zβ ∂yβ x3
and integrating by parts with respect yβ , we can transform the right hand
side of (1.3.3) to the form
Z h i ∂
Aα,3 (x) = − Bα (x0 + y 0 x3 ) − Bα (x0 ) (yβ K(y 0 ))dy 0 +
∂yβ
R2
Z
+ Bα (x0 + y 0 x3 )K(y 0 )dy 0 .
R2
By the choice of K, this immediately implies Aα,3 (x0 , 0) = Bα (x0 ), α = 1, 2.

Now, our aim is prove the estimate
k∇2 Aα k2,R3+ ≤ ckbk2,R3+ , α = 1, 2, (1.3.4)
for some universal constant c.

Let us demonstrate one fact from the theory of traces in Sobolev spaces.
Lemma 3.4. For any smooth function f : R3+ → R, vanishing for sufficiently
large |x|, the following inequality is valid:
Z Z 0 0 Z
2 0 0 2 dx dy
kf k 21 := |f (x , 0) − f (y , 0)| 0 0 3
≤ c |∇f |2 dx (1.3.5)
2
L2 (R ) |x − y |
R2 R2 R3+
with some universal constant c.
Proof By the shift in variables,
dz 0
Z Z
2
kf k 21 = |f (x0 + z 0 , 0) − f (x0 , 0)|2 dx0 .
L2 (R2 ) |z 0 |3
R2 R2
Applying the triangle inequality, we find
|f (x0 + z 0 , 0) − f (x0 , 0)| ≤ |f (x0 + z 0 , |z 0 |) − f (x0 , |z 0 |)|+
+|f (x0 + z 0 , |z 0 |) − f (x0 + z 0 , 0)| + |f (x0 , |z 0 |) − f (x0 , 0)|. (1.3.6)

According to (1.3.6), we shall evaluate three integrals. In the first one, the
polar coordinates z 0 = (% cos ϕ, % sin ϕ) are used so that:
dz 0
Z Z
I1 = |f (x0 , |z 0 |) − f (x0 , 0)|2 dx0 =
|z 0 |3
R2 R2
Z Z∞
0 1
= 2π dx 2
|f (x0 , %) − f (x0 , 0)|2 d%.
%
R2 0
The right hand side of the latter inequality can be bounded from above with
the help of Hardy’s inequality
Z∞ p p Z∞
−p
t |g(t) − g(0)| dt ≤ |g 0 (t)|p dt
p
p−1
0 0
with 1 < p < ∞. Then

Z Z∞ Z
∂
2
0 0
I1 ≤ 2π dx 4 f (x , %) d% ≤ 8π |∇f |2 dx.

∂%
R2 0 R3+
Now, let us treat I2 , using the same arguments. As a result, we have

dz 0
Z Z
I2 = |f (x0 + z 0 , |z 0 |) − f (x0 + z 0 , 0)|2 dx0 =
|z 0 |3
R2 R2
dz 0
Z Z Z
0 0 0 2 0
= |f (y , |z |) − f (y , 0)| dy ≤ 8π |∇f |2 dx.
|z 0 |3
R2 R2 R3+
To estimate the third term, we exploit the following simple inequality
Z1 ∂
|f (x0 + z 0 , |z 0 |) − f (x0 , |z 0 |)| = (x0 + tz 0 , |z 0 |)dt ≤

∂t
0
Z1 21
0 0 0 0 2
≤ |z | |∇x f (x + tz , |z |)| dt ,
0
which give us
dz 0
Z Z
I3 = |f (x0 + z 0 , |z 0 |) − f (x0 , |z 0 |)|2 dx0 ≤
|z 0 |3
R2 R2
Z1
dz 0
Z Z
≤ dt |∇x0 f (x0 + tz 0 , |z 0 |)|2 dx0 =
|z 0 |
R2 0 R2
dz 0
Z Z
= |∇y0 f (y 0 , |z 0 |)|2 dy 0
|z 0 |
R2 R2
It remains to make the change of variables z 0 = (% cos ϕ, % sin ϕ)

Z
I3 ≤ 2π |∇f |2 dx
R3+
and complete our proof.

Now, let us show that (1.3.5) is true for any function f of class C 2 , having
the decay
c
|f (x)| ≤ , (1.3.7)
|x|
where c is a positive constant.
Denoting by B 0 (R) is the unit disk centered at the origin and letting
fR = f ϕR , where ϕR is a standard cut-off function, we have
|f (x0 , 0) − f (y 0 , 0)|2 0 0
Z Z
dx dy =
|x0 − y 0 |3
B 0 (R) B 0 (R)
|fR (x0 , 0) − fR (y 0 , 0)|2 0 0

Z Z
= dx dy ≤
|x0 − y 0 |3
B 0 (R) B 0 (R)
Z Z
c 2
≤c ϕR |∇f | dx + 2 |f |2 dx ≤
R
R3+ (B(2R)\B(R))∩R3+
Z
c
≤c |∇f |2 dx + .
R
R3+
Passing to the limit as R → ∞ and using Fatou’s lemma, we establish (1.3.5).

Next, we observe that it is sufficient to show
k∇2 Aα k2,R3+ ≤ ck∇h(·, 0)k 1 . (1.3.8)

L22 (R2 )
Indeed, since
c(b)
|∇h(x)| ≤
|x|2
for |x| 1 and, by (1.3.5) and by (1.3.2), one can conclude that
k∇h(·, 0)k 1 ≤ ck∇2 hk2,R3+ ≤ ckbk2,R3+ .

L22 (R2 )
Then statement (ii) of Proposition 3.2 for this particular class of b follows.
Now, let us prove (1.3.8), directly working out the second derivatives of
A, Z
0 ∂ ∂
Aα,β3 (x , x3 ) = − Bα (x0 + y 0 x3 ) (yγ K(y 0 ))dy 0 +
∂xβ ∂yα
R2
Z
∂
+ Bα (x0 + y 0 x3 )K(y 0 )dy 0 .
∂xβ
R2
Obviously,
∂ ∂ 1
Bα (x0 + y 0 x3 ) = Bα (x0 + y 0 x3 ) .
∂xβ ∂xβ x3
Then Z
1
S = Aα,β3 (x) = δBα (x0 , y 0 , x3 )Kβ (y 0 )dy 0 ,
x3
R2
where
δBα (x0 , y 0 , x3 ) := Bα (x0 + y 0 x3 ) − Bα (x0 )
and
Kβ (y 0 ) := (yγ K(y 0 )),βγ − K,β (y 0 ).
Now, we have
Z Z∞ Z 1 Z 2
2 0 0 0 0 0
S dx ≤ c dx3 dx |δBα (x , y , x3 )||Kβ (y )|dy
x3
R3+ 0 R2 R2
and, by Hölder inequality,

Z Z∞ Z Z Z
1 0 0 0
2
S dx ≤ c dx3 dx 2 |Kβ (z )|dz |δBα (x0 , y 0 , x3 )|2 |Kβ (y 0 )|dy 0 ≤
x3
R3+ 0 R2 R2 R2
Z Z∞ Z
0 0 dx3
≤c |Kβ (y )|dy |Bα (x0 + y 0 x3 ) − Bα (x0 )|2 dx0 .
x23
R2 0 R2
Introducing polar coordinates y 0 = %(cos ϕ, sin ϕ), we find
Z Z1 Z2π Z∞ Z
dx3
2
S dx ≤ c %d%dϕ |Bα (x0 + %(cos ϕ, sin ϕ)x3 ) − Bα (x0 )|2 dx0 .
x23
R3+ 0 0 0 R2
If we set z 0 = x3 (cos ϕ, sin ϕ), then
Z1
dz 0
Z Z Z
2
S dx ≤ c %d% |Bα (x0 + z 0 %) − Bα (x0 )|2 dx0 .
|z 0 |3
R3+ 0 R2 R2
Letting y 0 = z 0 %, we show
Z1
|Bα (x0 + y 0 ) − Bα (x0 )|2 0 0
Z Z Z
2
S dx ≤ c d% dx dy ≤
|y 0 |3
R3+ 0 R2 R2
|Bα (z 0 ) − Bα (y 0 )|2 0 0
Z Z
≤ 0 0 3
dz dy ≤ ck∇h(·, 0)k2 21 . (1.3.9)
|z − y | L2 (R2 )
R2 R2
With the remaining second derivatives, we proceed as follows:

Z
1
Aα,33 (x) = (Bα (x0 + y 0 x3 ) − Bα (x0 ))K(y
e 0 )dy 0 ,
x3
R2
where
e 0 ) := (yγ (yβ K(y 0 )),β )),γ − (yγ K(y 0 )),γ ,
K(y
and Z
1
Aα,βγ (x) = (Bα (x0 + y 0 x3 ) − Bα (x0 ))(K(y 0 )),βγ dy 0 .
x3
R2
So, the same arguments as above leads to the required bound

Z
(A2α,33 + A2α,βγ )dx ≤ ck∇h(·, 0)k2 21 .
L2 (R2 )
R3+
Hence, inequality (1.3.8) for smooth compactly supported b is proven.

Now, our aim is to show
◦
u ∈ L12 (R3+ ) (1.3.10)
for any b ∈ C0∞ (R3+ ). The proof of (1.3.10) consists of two parts.
Step 1. Let us show first that ∇h ∈ L2 (R3+ ). Indeed, since ∇2 h ∈
L2 (R3+ ), ∇h ∈ L2 (B+ (1)), where B+ (R) := {x ∈ B(R) : x3 > 0}. We know
that
c(b)
|∇h(x)| ≤
|x|2
for |x| 1. So,
Z Z Z ZR
d%
|∇h|2 dx ≤ |∇h|2 dx + |∇h|2 dx ≤ ... + c(b) ≤ c(b)
%2
B+ (R) B+ (1) B+ (R)\B+ (1) 1
for any R > 1.

Step 2. Let us show ∇Aα ∈ L2 (R3+ ). We know
Z
Aα,3 (x) = Bα (x0 + y 0 x3 )K0 (y 0 )dy 0 ,
R2
where
K0 (y 0 ) = K(y 0 ) − (yβ K(y 0 )),β .
Let α = 1. Then B1 (x0 ) = h,2 (x0 , 0) and
∂ 1
B1 (x0 + y 0 x3 ) = h(x0 + y 0 x3 , 0) .
∂y2 x3
So, Z
1 ∂
A1,3 (x) = − (h(x0 + y 0 x3 , 0) − h(x0 , 0)) K0 (y 0 )dy 0 .
x3 ∂y2
R2
Repeating the evaluation of Aα,β3 , we find
Z
A21,3 dx ≤ ckh(·, 0)k 1 .
L22 (R(2)
R3+
Since |h(x)| ≤ c(b)/|x| for |x| 1, one can derive with the help of Lemma
3.4 the inequality
kh(·, 0)k 12 ≤ ck∇hk2,R3+ .
L2 (R(2)
This means that, by Step 1, A1,3 ∈ L2 (R3+ ). The same is true for A2,3 . The
proof of the fact that Aα,β ∈ L2 (R3+ ) is an exercise. So, it has been proven
that
u ∈ L2 (R3+ ) (1.3.11)
provided b ∈ C0∞ (R3+ ).
Now, we wish to finish the proof of (1.3.10). Letting uR = ϕR u, we
observe that
Z Z Z
2 2 1
|∇(u − uR )| dx ≤ c |∇u| dx + c 2 |u|2 dx → 0
R
R3+ R3+ \B+ (R) B+ (2R)\B+ (R)
as R → ∞.
The function uR is not compactly supported in R3+ and we need to cut
it in the direction of x3 . To this end, let us introduce the following cut-off
function: χ(t) = 0 if −∞ < t ≤ ε/2, χ(t) = 2(t − ε/2)/ε if ε/2 < t ≤ ε, and
χ(t) = 1 if t > ε. Considering uR,ε (x) = uR (x)χ(x3 ), we have
Z Zε Z Z Zε
2 2 1 0 0
|∇(uR − uR,ε )| dx ≤ c dx3 |∇uR | dx + 2 dx |uR |2 dx3 .
ε
R3+ 0 R2 R2 0
The first integrals on the right hand side of the last inequality tends to zero
as ε → 0. To show that the second term does the same, we are going to use
two facts. Firstly, uR (x0 , 0) = 0 and secondly, by the Friedreich inequality,
Zε Zε
∂
2
0 2 2 0
|uR (x , x3 )| dx3 ≤ cε uR (x , x3 ) dx3 .

∂x3

0 0
So, combining the latter inequalities, we show that

Z Zε Z
2
|∇(uR − uR,ε )| dx ≤ c |∇uR |2 dx → 0
R3+ 0 R2
as ε → 0 for each fixed R > 0.

It remains to mollify uR,ε . The mollification (uR,ε )τ belongs to C0∞ (R3+ )
for 0 < τ ≤ τ (R, ε) and
Z
|∇(uR,ε − (uR,ε )τ )|2 dx → 0
R3+
as τ → 0 for each fixed R and ε. So, (1.3.10) is proven.

Now, we are going to extend our result to functions b ∈ L2 (R3+ ). Given
b ∈ L2 (R3+ ), there exists b(m) ∈ C0∞ (R3+ ) such that kb(m) − bk2,R3+ → 0 as
◦
m → ∞. We know that there is u(m) ∈ L12 (R3+ ) having the properties:
div u(m) = b(m)
in R3+ and
k∇u(m) k2,R3+ ≤ ckb(m) k2,R3+ .
Moreover, by construction
k∇u(m) − ∇u(k) k2,R3+ ≤ ckb(m) − b(k) k2,R3+ ,
which implies that

u(m) → u
in L12 (R3+ ).
Let us mention some consequences and generalizations.
Theorem 3.5. Let Ω ⊂ Rn be a bounded domain with Lipschitz boundary,

1 < p < ∞, and let
Z
L̄p (Ω) := {b ∈ Lp (Ω) : b(x)dx = 0}.
Ω
1.4. NEČAS IMBEDDING THEOREM 21
◦
Then, for any b ∈ L̄p (Ω), there exists u ∈ L12 (Ω) with the following properties:
div u = b
in Ω and
k∇ukp,Ω ≤ c(p, n, Ω)kbkp,Ω .
Remark 3.6. For bounded domain domains, we need the restriction on b:
Z
b(x)dx = 0.
Ω
It is a kind of compatibility condition.

Remark 3.7. Proof of Theorem 3.5 is based on Propositions 3.1 and 3.2,
decomposition of the unity, and changes of coordinates. It is quite involved
but does not contain new ideas.
Remark 3.8. There is a different approach to the proof of Theorem 3.5,
which is due to Bogovkii. It is simpler than the proof above. But it relies
upon the theory of singular integrals.
1.4 Nečas Imbedding Theorem

We start this section with recalling known facts related to duality between
function spaces. For are a given Banach space V , let V 0 be its dual one,
i.e., the space of all bounded linear functionals on V . Very often, we need
to identify V 0 with a particular function space. The choice might depend on
the problem under consideration. There is a relatively general construction
that is very popular in the theory of evolution problems. To describe it, let
us state the corresponding standing assumptions. We are given a reflexive
Banach space V with the norm k · kV and a Hilbert space with the scalar
product (·, ·). It is supposed that V is continuously
p imbedded into H, i.e.,
there exists a constant c such that kvkH = (v, v) ≤ ckvkV for any v ∈ V
and let V be dense in H.
As usual, we identify H 0 with H itself, i.e., H 0 = H (in usual functional
analysis sense). Now let us f ∈ H, then v 7→ (f, v) is a bounded linear
functional on V and thus there exists vf0 ∈ V 0 with the properties
< vf0 , v >= (f, v) ∀v ∈ V

and
kvf0 kV 0 ≤ ckf kH ∀f ∈ H.
So, we have a bounded linear operator τ : H → V 0 (one-to-one by density)
defined by the identity τ f = vf0 for f ∈ H.
Obviously, τ (H) is a linear manifold V 0 . Moreover, it is dense in V 0 .
To see that it is true, assume it is not, i.e., there exists v00 ∈ V 0 but v00 ∈ /
V0 00 00 0 0
[τ (H)] . By the Hahn-Banach theorem, there exists v ∈ V := (V ) with
the properties < v 00 , v00 >= 1 and < v 00 , v 0 >= 0 for any v 0 ∈ τ (H). Since
V is reflexive, there should be v ∈ V so that < v 00 , v 0 >=< v 0 , v > for any
v 0 ∈ V 0 . This gives us: < v00 , v >= 1 and < vf0 , v >= (f, v) = 0 for any
f ∈ H. Therefore, v = 0 and we get a contradiction. The latter allows us to
identify V 0 with the closure of τ (H) in V 0 . But we can go further and identify
duality relation between V and V 0 with the scalar product (·, ·) on H. Very
often, we call such an identification of V 0 the space dual to V relative to the
Hilbert space H.
So, under our standing assumptions, v 0 ∈ V 0 means that there exists a
sequence sequence fm ∈ H such that
sup{|(fk − fn , v)| : kvkV = 1} → 0
as k, n → ∞ and (v 0 , v) is just notation for lim (fk , v) that exists for all
k→∞
v ∈ V . Moreover,
kv 0 kV 0 = sup{|(v 0 , v)| : kvkV = 1}.

◦ 0
We let us denote by L−1
r (Ω) an identification of the space L 1
r 0 (Ω) ac-
◦
cording to the aforesaid scheme with V = L1r0 (Ω) and H = L2 (Ω).
Theorem 4.1. Let 1 < r < ∞ and let Ω be a domain in Rn . Assume that
that a distribution p, defined in Ω, has the property:
∇p ∈ L−1
r (Ω).
The following statements are valid:

(i) p ∈ Lr,loc (Ω) and for any Ω0 b Ω there exists a constant c(r, n, Ω0 , Ω) such
that Z
|p − a|r dx ≤ ck∇pkrL−1
r (Ω)
Ω0
for some constant a;

(ii) if Ω = Rn or Rn+ , then p ∈ Lr (Ω) and there exists a constant c(r, n) such
that Z
|p|r dx ≤ ck∇pkrL−1
r (Ω)
;
Ω
(iii) if Ω is a bounded Lipschitz domain, then p ∈ Lr (Ω) and there exists a

constant c(r, n, Ω) such that
Z
|p − a|r dx ≤ ck∇pkrL−1
r (Ω)
for some constant a.

Proof
(i) Without less of generality, we may assume that a bounded domain Ω0 has
Lipschitz boundary. We claim that there exists a constant c(r, n, Ω0 , Ω) such
that
| < p, q > | ≤ cKkqkr0 ,Ω0 (1.4.1)
for any q ∈ C0∞ (Ω0 ) with [q]Ω0 = 0. The latter would imply that p is a regular
distribution. Here, the following notion is used:
K := k∇pkL−1
r (Ω)
and Z
1
[q]ω = q(x)dx.
|ω|
ω
◦
By Theorem 3.5, there exists u ∈ L1r0 (Ω0 ) such that
div u = q
in Ω0 and
k∇ukr0 ,Ω0 ≤ c(r, n, Ω0 )kqkr0 ,Ω0 .
Functions u and q are supposed to be extended by zero outside Ω0 . Let us
mollify u in a standard way
Z Z
(u)% (x) = ω% (x − y)u(y)dy = ω% (x − y)u(y)dy
Ω0 Rn
with the help of a smooth mollifier ω% . So, (u)% ∈ C0∞ (Ω) for 0 < % <
%0 (Ω0 , Ω). Moreover, we know that
∇(u)% = (∇u)%
and thus
div (u)% = (q)% ,
k∇(u)% kr0 ,Ω ≤ k∇ukr0 ,Ω ≤ ckqkr0 ,Ω ,
by the known mollification properties.
Now, we have (in the sense of distributions)
< ∇p, (u)% >= − < p, div (u)% >= − < p, (q)% >,
which implies
| < p, (q)% > | ≤ Kk∇(u)% kr0 ,Ω ≤ cKkqkr0 ,Ω .
It worthy to notice that there exists a compact K such that Ω0 ⊂ K ⊂ Ω,

supp ∇k (q)% and supp ∇k q belong to K, and
∇k (q)% → ∇k q
uniformly in K for any k = 0, 1, ... as % → 0 and thus
< p, (q)% >→< p, q >
as % → 0. Tending % → 0, we then find (1.4.1).

It follows from Banach and Riesz theorems that there exist P ∈ Lr (Ω0 )
such that kP kr,Ω ≤ cK and
Z
< p, q >= P qdx
Ω0
for any q ∈ C0∞ (Ω0 ) with [q]Ω0 = 0.

Now, let us test the latter identity with q = div u for an arbitrary u ∈
C0∞ (Ω0 ) (as usual all the functions are extended by zero to the whole domain
Ω). As a result, we have
Z
< ∇p, u >= − < p, div u >= − P div udx.
Ω0
This means that ∇(p|Ω0 − P ) = 0. And thus, by Theorem 1.2, p − P =

constant on Ω0 . So, part (i) is proven.
(ii) According to (i), our distribution v is regular and, therefore,
Z
< p, q >= p(x)q(x)dx
Ω
for any q ∈ C0∞ (Ω).

◦
Given q ∈ C0∞ (Ω), we find u ∈ L1r0 (Ω) such that div u = q in Ω and
k∇ukr0 ,Ω ≤ ckqkr0 ,Ω .
◦
By the definition of L1r0 (Ω), there exists a sequence u(m) ∈ C0∞ (Ω) such that
∇u(m) → ∇u in Lr0 (Ω)
and thus
q (m) := div u(m) → q in Lr0 (Ω)
as m → ∞. Then, as it is pointed out above, we should have
Z Z
(m)
< ∇p, u >= − pdiv u dx = − pq (m) dx ≤ cKk∇u(m) kr0 ,Ω .
(m)
Ω Ω
Passing to the limit, we find the estimate

Z
− pqdx ≤ cKkqkr0 ,Ω , (1.4.2)
Ω
which allows us to state that
kpkr,Ω ≤ cK.
(iii) Here, it is enough to repeat the same arguments as in (i), replacing Ω0

with Ω, under the additional restriction on q that is [q]Ω = 0. As a result, we
arrive at estimate (1.4.2), being valid for any q ∈ C0∞ (Ω) provided [q]Ω = 0.
Repeating arguments, used at the end of the proof of the statement (i), we
conclude that there exists P ∈ Lr (Ω) such that kP kr,Ω ≤ cK and P = p − a
for some constant a.
1.5 Spaces of Solenoidal Vector Fields

First, we introduce the set of all smooth compactly supported in Ω vector
fields, which are divergence free there,
∞
C0,0 (Ω) := {v ∈ C0∞ (Ω) : div v = 0 in Ω}.
Next, for 1 ≤ r < ∞, we define the following ”energy” spaces
◦
1 ∞ L1r (Ω)
J r (Ω) := [C0,0 (Ω)]
and ◦
Jˆ1r (Ω) := {v ∈ L1r (Ω) : div v = 0 in Ω}.
In general,
◦
Jˆ1r (Ω) ⊇ J 1r (Ω).
For r = 2, we use abbreviations:
◦
V (Ω) := J 12 (Ω), V̂ (Ω) := Jˆ12 (Ω).
Here, it is an example of a domain in R3
Ω? = R3 \ {x = (0, x2 , x3 ), x22 + x23 ≥ 1},
for which
V̂ (Ω∗ ) \ V (Ω∗ ) 6= ∅.
This example is due to J. Heywood.
Proposition 5.1. V (Ω) = V̂ (Ω) if Ω = Rn or Rn+ , n = 2, 3.
Proof of Proposition 5.1 is based upon two Leray’s inequalities.
Lemma 5.2. If n = 2, then
|w(x)|2
Z Z
2 2 dx ≤ 4 |∇w(x)|2 dx
|x| ln |x|
|x|>1 |x|>1
for any w ∈ C0∞ (|x| > 1).

If n = 3, then
|w(x)|2
Z Z
dx ≤ 4 |∇w(x)|2 dx
|x − y|
R3 R3
for any w ∈ C0∞ (R3 ) and for any y ∈ R3 .

1.5. SPACES OF SOLENOIDAL VECTOR FIELDS 27
Proof Let n = 2. Integration by parts gives:
|w|2
Z Z Z
xk 2 xk
2 ww,k 2 dx = (|w| ),k 2 dx = dx ≤
|x| ln |x| |x| ln |x| |x|2 ln2 |x|
|x|>1 |x|>1 |x|>1
Z |w|2 21 Z 21
2
≤2 dx |∇w(x)| dx
|x|2 ln2 |x|
|x|>1 |x|>1
The case n = 3 is an exercise .

Proof of Proposition 5.1 We shall prove the statement for n = 2.
The case n = 3 is treated in a similar way and can be regarded as a good
exercise.
Let Ω = R2 . We introduce a special cut-off function ψR such that ψR (x) =
1 if |x| < R, ψR (x) = 0 if |x| > R3 , and
− ln ln |x| + ln ln R3
ψR (x) =
− ln ln R + ln ln R3
if R ≤ |x| ≤ R3 .
Given v ∈ V̂ (R2 ), let
Z Z
IR := |∇(v − vψR )| dx = |∇(1 − ψR )v|2 dx.
2
Ω Ω
Assuming that R > 3, we have
|v(x)|2
Z Z
IR ≤ c dx + c |∇v|2 dx. (1.5.1)
|x|2 ln2 |x|
R<|x|<R3 R3 \B(R)
If
|v(x)|2
Z
dx < ∞ (1.5.2)
|x|2 ln2 |x|
3<|x|
is valid, then the right hand side of (5.2) tends to zero as R → ∞ and thus
vψR → v (1.5.3)
in L12 (R2 ).
◦
Remark 5.3. Show that 1 ∈ L12 (R2 ).
In order to prove the validity of (1.5.2), we find a sequence v (k) ∈ C0∞ (R2 )
such that
∇v (k) → ∇v
in L2 (R2 ) and
v (k) → v
a.e. in R2 , and sequence v (k) is bounded in L2 (B(3)). Now, let a smooth
cut-off function η satisfy the properties: η(x) = 0 if |x| < 2 and η(x) = 1 if
|x| > 3. Then, by Leray’s inequality, we have
|v (k) (x)|2 η 2 (x)|v (k) (x)|2

Z Z
dx ≤ dx ≤
|x|2 ln2 |x| |x|2 ln2 |x|
3<|x| 1<|x|
Z Z
(k) 2
≤c |∇v | dx + c |v (k) |2 dx.
1<|x| 3>|x|
The right hand side of the latter inequality is bounded uniformly in k. And
then (1.5.2) follows from Fatou’s lemma. So, (1.5.3) is proven.
Unfortunately, vψR is not divergence free, in fact, div (vψR ) = v · ∇ψR ,
and we should correct it with he help of results of the previous section. We
apply Theorem 3.5 in the ball B(R3 ), which reads that there exists a function
◦ ◦
wR ∈ L12 (B(R3 )) = W 12 (B(R3 ))
such that
div wR = v · ∇ψR
in B(R3 ) and
k∇wR k2,B(R3 ) ≤ ckv · ∇ψR k2,B(R3 ) .
It is worthy to notice that a constant c in the latter inequality is independent
of v and of R (by scaling). We extend wR by zero to the whole R2 . By the
choice of cut-off function ψR , we have
|v(x)|2
Z
2
k∇wR k2,R2 ≤ c dx → 0
|x|2 ln2 |x|
R<|x|<R3
1.5. SPACES OF SOLENOIDAL VECTOR FIELDS 29
as R → ∞.
We let uR = v · ∇ψR − wR . New function has two important properties:
uR ≡ 0 out of B(R3 ), div uR = 0 in R2 , and
k∇(v − uR )k2,R2 ≤ ck∇(1 − ψR )vk2,R2 + ck∇wR k2,R2 → 0
as R → ∞.
∞
On the other hand, the mollification (uR )ε ∈ C0,0 (R2 ) and
k∇(uR )ε − ∇uR k2,R2 → 0
as ε → 0 for each fixed R > 0. So, the case Ω = R2 is done.

◦
Let Ω = R2+ . Given u ∈ L12 (R2+ ), denote by ũ the extension of u by zero
◦
to the whole R2 . Obviously, ũ ∈ L12 (R2 ). For λ > 0, uλ is made of u by shift
in direction of x2 : uλ = ũ(x1 , x2 − 2λ). This function is zero if x2 < 2λ and
we know that
Z Z
|∇ũ − ∇uλ | dx = |∇u − ∇uλ |2 dx → 0
2
R2 R2+
as λ → 0.
Next, we introduce the new function vR,λ = uλ ψR,λ , where ψR,λ (x) =
ψR (x1 , x2 − 2λ). Then
Z Z
|∇vR,λ − ∇uλ | dx = |∇vR,λ − ∇uλ |2 dx =
2
R2 R2+
Z
|∇(ψR ũ − ũ)|2 dx → 0
R2
as R → ∞ for fixed λ > 0 as

|ũ(x)|2 |ũ(x)|2
Z Z
dx = dx < ∞.
|x|2 ln2 |x| |x|2 ln2 |x|
3<|x| {3<|x|}∩R2+
One should correct the divergence of uR,λ as in the first part. We state that
there exists ◦
wR,λ ∈ L12 (B+ ((0, 2λ), R3 )),
where B+ ((0, 2λ), r) = {x ∈ B((0, 2λ), r) : x2 > 2λ}, such that
div wR,λ = uλ · ∇ψR,λ
in B+ ((0, 2λ), R3 ) and
k∇wR,λ k2,B+ ((0,2λ),R3 ) ≤ ckuλ · ∇ψR,λ k2,B+ ((0,2λ),R3 )
with a constant c independent of R and λ.

As in the previous case, we extend wR,λ to the whole R2 by zero and then
let
uR,λ = vR,λ − wR,λ .
What we know about this function uR,λ is as follows:
∇uR,λ → ∇uR
in L2 (R2+ ) as λ → 0,
div uR,λ = 0
in R2 ,
uR,λ = 0
out of semi-disk B+ ((0, 2λ), R3 ),
∇uR,λ ∈ L2 (R2 ).
∞
Then it remain to take the mollification (uR,λ )ε that belongs to C0,0 (R2+ ) for
sufficiently small ε, for example, 0 < ε < λ. The well-known property of the
mollification
∇(uR,λ )ε → ∇uR,λ in L2 (R2+ )
completes the proof.
In fact, we have the following statement
Theorem 5.4. Let 1 < m < ∞ and let Ω be Rn , or Rn+ , or a bounded domain
with Lipschitz boundary. Then
◦
J m (Ω) = Jˆm (Ω).
1 1
Proof See next section.

1.6. LINEAR FUNCTIONALS 31
1.6 Linear Functionals Vanishing on Diver-

gence Free Vector Fields
Proposition 6.1. Let Ω = Rn or Rn+ or be a bounded Lipschitz domain.
◦
Assume that 1 < s < ∞. Let, further, l : L1s (Ω) → R be a linear functional
having the following properties:
|l(v)| ≤ ck∇vks,Ω
◦
for any v ∈ L1s (Ω) and
l(v) = 0
for any v ∈ Jˆ1s (Ω).
Then there exists a function p ∈ Ls0 (Ω), s0 = s/(s − 1), such that
Z
l(v) = pdiv vdx
Ω
◦
for any v ∈ L1s (Ω).
Proof Let us consider case Ω = Rn or Rn+ .

We define a linear functional G : Ls (Ω) → R as follows. Given q ∈ Ls (Ω),
◦
take any u ∈ L1s (Ω) such that div u = q and let G(q) = l(u). By Proposition
3.2, there is at least one u with this property. Next, one should show that
◦
functional G is well-defined, i.e., for any v ∈ L1s (Ω) with div v = q, we have
l(u) = l(v). Indeed, u − v ∈ Jˆs1 (Ω) and by our assumptions l(u − v) = 0 =
l(u) − l(v).
It is not a difficult exercise to verify that G is linear functional.
◦
Now, we can select a special u ∈ L1s (Ω), for which we have the identity
div u = q and the estimate
k∇uks,Ω ≤ ckqks,Ω .
The latter implies
G(q) = l(u) ≤ ck∇uks,Ω ≤ ckqks,Ω

for any q ∈ Ls (Ω). So, the functional G is bounded on Ls (Ω) and by Riesz
theorem, there exists p ∈ Ls0 (Ω) such that
Z
G(q) = pqdx
Ω
◦
for any q ∈ Ls (Ω). Now, for any u ∈ L1s (Ω), we have
Z
l(u) = G(div u) = pdiv udx.
Ω
For bounded Lipschitz domains, one should replace the space Ls (Ω) with
its subspace
L̄s (Ω) = {q ∈ Ls (Ω) : [q]Ω = 0}
and use the same arguments as above.
◦
However, we can assume that our functional vanish on J 1s (Ω) only.
Theorem 6.2. Let Ω = Rn or Rn+ or be a bounded Lipschitz domain. Assume

◦
that 1 < s < ∞. Let, further, l : L1s (Ω) → R be a linear functional having
the following properties:
|l(v)| ≤ ck∇vks,Ω
◦
for any v ∈ L1s (Ω) and
l(v) = 0
◦
for any v ∈ J 1s (Ω).
Then there exists a function p ∈ Ls0 (Ω), s0 = s/(s − 1), such that
Z
l(v) = pdiv vdx
Ω
◦
for any v ∈ L1s (Ω).
Proof Indeed, let us consider a sequence of bounded smooth domains

Ωm , m = 1, 2, ..., with the following properties:
Ωm ⊂ Ωm+1
1.6. LINEAR FUNCTIONALS 33
and ∞
[
Ω= Ωm .
m=1
◦
Given v ∈ L1s (Ωm ), define v m = v in Ωm and v m = 0 outside Ωm . Obvi-
◦ ◦
ously, v m ∈ L1s (Ω). We also define a linear functional lm : L1s (Ωm ) → R as
follows:
lm (v) := l(vm )
◦
for any ∈ L1s (Ωm ). It is a bounded functional, i.e.,
|lm (v)| ≤ k∇vks,Ωm
with a constant c independent of m.
By mollification, we can show the following fact: if v ∈ Jˆ1s (Ωm ), then
◦
v m ∈ J 1s (Ω). This immediately implies that lm (v) = 0 for any v ∈ Jˆ1s (Ωm ).
According to Proposition 6.1, there exists pm ∈ Ls0 (Ωm ) such that
Z
lm (v) = pm div vdx
Ωm
◦
for any v ∈ L1s (Ωm ). Obviously, pm is defined up to an arbitrary constant.
Moreover, pm+1 − pm = c(m) = constant in Ωm if k > m. So, we can change
pm+1 adding a constant to achieve the identity pm+1 = pm in Ωm . This
allow us to introduce a function p ∈ Ls0 ,loc (Ω) so that p = pm on Ωm . By
construction, it satisfies identity
Z
l(v) = pdiv vdx, (1.6.1)
Ω
and the inequality

l(v) ≤ ck∇vks,Ω
for any v ∈ C0∞ (Ω), which, by definition, means that ∇p ∈ L−1 s0 (Ω) and as it
follows from Theorem 4.1, p ∈ Ls0 (Ω). If so, identity (1.6.1) can be extended
◦
to all function v ∈ L1s (Ω) by density arguments.
Proof of Theorem 5.4 Indeed, assume that there exists v∗ ∈ Jˆ1s (Ω)
◦
/ J 1s (Ω). By Banach theorem, there exists a functional
but v∗ ∈
◦ 0
1
l∗ ∈ Ls (Ω)
with the following properties:
l∗ (v∗ ) = 1
and
l∗ (v) = 0
◦
for any v ∈ J 1s (Ω). By Theorem 6.2, there exists p ∈ Ls0 (Ω) such that
Z
l∗ (v) = pdiv vdx
Ω
◦
for all v ∈ L1s (Ω). However,
Z
l∗ (v∗ ) = pdiv v∗ dx = 0
Ω
since v∗ ∈ Jˆ1s (Ω). This is a contradiction.
1.7 Helmholtz-Weyl Decomposition

Let ◦
∞ L2 (Ω)
J (Ω) := [C0,0 (Ω)]
and
G(Ω) := {v ∈ L2 (Ω; Rn ) : v = ∇p for some distribution p}.
Remark 7.1. We know that if a distribution p ∈ G(Ω), then in general

p ∈ L2,loc (Ω). However, if Ω is a bounded Lipschitz domain, then in fact
p ∈ L2 (Ω).
Theorem 7.2. (Ladyzhenskaya) For any domain Ω ∈ Rn ,

◦
L2 (Ω) := J (Ω) ⊕ G(Ω).
Proof Obviously, our statement is equivalent to the following formula

◦
G(Ω) = (J (Ω))⊥ .
1.7. HELMHOLTZ-WEYL DECOMPOSITION 35
Step 1 Let Ω be a bounded Lipschitz domain. It is easy to see that

◦
G(Ω) ⊆ (J (Ω))⊥ ,
since Z
v · ∇pdx = 0
Ω
∞
for any p ∈ G(Ω) and for any v ∈ C0,0 (Ω). Now, assume
◦
u ∈ (J (Ω))⊥ ,
i.e., u ∈ L2 (Ω) and Z

u · vdx = 0
Ω
∞
for any v ∈ C0,0 (Ω). By Poincaré inequality,
Z Z 1 Z 1
2 2 2 2
l(v) = u · vdx ≤ |u| |v| ≤
Ω Ω Ω
≤ c(Ω)kuk2,Ω k∇vk2,Ω
◦ ◦ ◦
for any v ∈ L1s (Ω). So, l : L12 (Ω) = W 12 (Ω) → R is bounded and l(v) = 0 for
◦
any v ∈ J 12 (Ω) =: V (Ω).
By Theorem 6.2, there exists p ∈ L2 (Ω) such that
Z
l(v) = pdiv vdx
Ω
◦
for any v ∈ L1s (Ω). Therefore, u = ∇p and thus p ∈ G(Ω) and
◦
(J (Ω))⊥ ⊆ G(Ω).
Step 2 We proceed is a similar way as in the proof of Theorem 6.2.

Consider a sequence of domains Ωj with the properties: Ωj ⊂ Ωj+1 and
∞
[
Ω= Ωj ,
j=1
where Ωj is a bounded Lipschitz domain.

Since v ∈ L2 (Ω) ⇒ v ∈ L2 (Ωj ), we can state that, for any j,
v = u(j) + ∇p(j) ,
where ◦
u(j) ∈ J (Ωj ), p(j) ∈ W21 (Ωj ).
We know that p(j) is defined up to a constant, which is going to be fixed by
the condition Z
p(j) dx = 0,
B∗
where B∗ is a fixed ball belonging to Ω1 .

Here, we are going to make use of the following version of Poincaré’s
inequality in a bounded Lipschitz domain Ω̃ containing the ball B∗ :
Z hZ Z 2 i
2 2
|q| dx ≤ c(Ω̃) |∇q| dx + qdx . (1.7.1)

Ω̃ Ω̃ B∗
A proof of (1.7.1) is based on standard compactness arguments and can be

regarded as a good exercise.
We further let ũ(j) = u(j) in Ωj and ũ(j) = 0 outside Ωj . It is easy to
◦
check that ũ(j) ∈ J (Ω) and
kũ(j) k2,Ω = ku(j) k2,Ωj ≤ kvk2,Ωj ≤ kvk2,Ω
and, hence, without loss of generality, we may assume that
ũ(j) * u
in L2 (Ω) and
◦
u ∈ J (Ω).
Next, by (1.7.1), we have for j ≥ s
Z Z
|p | dx ≤ c(s) |∇p(j) |2 dx ≤ c(s)kvk22,Ω .
(j) 2
Ωs Ωs
Letting s = 1, we find a subsequence {jk1 }∞

k=1 so that
1 1
p(jk ) * p1 , ∇p(jk ) * ∇p1 ,
1.7. HELMHOLTZ-WEYL DECOMPOSITION 37
in L2 (Ω1 ). Then we let s = 2 and select a subsequence {jk2 }∞ 1 ∞

k=1 of {jk }k=1
such that
2 2
p(jk ) * p2 , ∇p(jk ) * ∇p2
in L2 (Ω2 ). Obviously, p2 = p1 in Ω1 . Proceeding in the same way, we find a
l−1 ∞
a subsequence {jkl }∞
k=1 of {jk }k=1 such that
l l
p(jk ) * pl , ∇p(jk ) * ∇pl
in L2 (Ωl ). For the same reason, pl = pl−1 in Ωl−1 . And so on.

Next, the function p, defined
p = pl
in Ωl , is well-defined. Using the celebrated diagonal Cantor process, we find

a subsequence p(js ) such that
p(js ) * p, ∇p(js ) * ∇p
in L2 (ω) for each ω b Ω. Moreover, we have the estimate

Z Z
|∇p| dx ≤ |v|2 dx
2
ω Ω
for any ω b Ω. So, it is easy to deduce from here that p ∈ G(Ω).

Now, fix w ∈ C0∞ (Ω). We have
Z Z Z
(js )
v · wdx = ∇p · wdx + u(js ) · wdx.
Ωjs Ωjs Ωjs
For sufficiently large s0 , supp w ∈ Ωjs0 and thus for s > s0

Z Z Z
(js )
v · wdx = ∇p · wdx + u(js ) · wdx.
Ωjs Ωjs Ωjs
0 0 0
Passing s → ∞, we show that v = u + ∇p. Orthogonality and uniqueness

can be proven in standard way (exercise).
1.8 Comments
The main aim for writing up Chapter I is to show author’s preferences how
the theory of functional spaces related to Navier-Stokes equations can be
developed. In our approach, the basic things are estimates of certain solutions
to the equation div u = f and their applications to the derivation of the Nečas
embedding theorem. Each part of this theory can be given in either more
compact way or even in a different way. For example, in Section 3, one could
apply very nice Bogovskii’s approach, see [1], based on the theory of singular
integrals. For more generic and detailed investigation of spaces arising in the
Navier-Stokes theory, we refer the reader to the monographs [29], [68], and
[17].
Chapter 2
Linear Stationary Problem
2.1 Existence and Uniqueness of Weak Solu-

tions
Let us consider the Dirichlet problem for the Stokes system

 −4u + ∇p = f
in Ω, (2.1.1)
div u = 0

u|∂Ω = 0 (2.1.2)
and if n = 3 and Ω is unbounded then u(x) → u0 as |x| → ∞.
In what follows, we always consider the case
u0 = 0.
Let Z
(f, g) := f (x)g(x)dx.
Ω
∞
If u and p are smooth, then, for any v ∈ C0,0 (Ω), integration by parts
gives the following identity:
Z Z
(−4u + ∇p) · vdx = ∇u : ∇vdx = (∇u, ∇v) = (f, v),
Ω Ω
which shows how weak solutions can be defined.
39
40 CHAPTER 2. LINEAR STATIONARY PROBLEM
Let us list our standing assumptions: n = 2 or 3 and
f ∈ L−1
2 (Ω).
Definition 1.1. A function u ∈ V̂ (Ω) is called a weak solution to boundary

value problem (2.1.1) and (2.1.2) if and only if
(∇u, ∇v) = (f, v)

∞
for any v ∈ C0,0 (Ω).
Remark 1.2. Boundary conditions are understood in the sense of trace, see
◦
the definition of spaces L12 (Ω) and V̂ (Ω) in Sections 1 and 5 of Chapter I. If
Ω is unbounded and n = 3, then condition u(x) → 0 as |x| → ∞ holds in the
following sense:
Z 61 Z 21
6 2
|u| dx ≤ c |∇u| dx .
Ω Ω
Lemma 1.3. (Existence). Given f , there exist at least one weak solution to
boundary value problem (2.1.1) and (2.1.2) that satisfies the estimate
k∇uk2,Ω ≤ kf kL−1
2 (Ω)
.
Proof It is easy to see that [u, v] = (∇u, ∇v) is a scalar product in

V̂ (Ω). On the hand, l(v) = (f, v) defines the linear functional on V (Ω) that
is bounded:
|l(v)| ≤ kf kL−1
2 (Ω)
kvk ◦ 1 .
L2 (Ω)
Now, the required existence is an easy consequence of the Banach extension

theorem and Riesz representation theorem.
Lemma 1.4. (Uniqueness). Assume in addition that
V̂ (Ω) = V (Ω). (2.1.3)
Then the weak solution to (2.1.1) and (2.1.2) is unique.
Proof Assume that there are two different solutions u1 and u2 . Then
(∇(u1 − u2 ), ∇v) = 0
2.2. COERCIVE ESTIMATES 41
∞
for and v ∈ C0,0 (Ω) and, by assumption (2.1.3),
k∇(u1 − u2 )k2,Ω = 0.
For n = 3, this immediately implies u1 = u2 . This is also true if n = 2 and

Ω 6= R2 . If n = 2 and Ω = R2 , the uniqueness takes place in the equivalence
classes, i.e., u1 − u2 ∈ [0]. [0] consists of functions that are constant in R2 .

To recover the pressure, let us assume that Ω satisfies conditions of The-
orem 6.2 of Chapter I and consider the following linear functional
l(v) = (∇u, ∇v) − (f, v).

◦
It is bounded on L12 (Ω) and vanishes on V (Ω). By Theorem 6.2 of Chapter
I, there exists a function p ∈ L2 (Ω) such that
(∇u, ∇v) − (f, v) = (p, div v)

◦
for any v ∈ L12 (Ω). In other words, functions u and p satisfy the Stokes
system in the sense of distributions.
2.2 Coercive Estimates

Proposition 2.1. Let Ω be a domain with smooth boundary (Ω = Rn or Rn+
or bounded domain). Let functions
f ∈ L2 (Ω), g ∈ W21 (Ω), u ∈ V (Ω), p ∈ L2 (Ω),
with [g]Ω = 0 if Ω is bounded, satisfy the nonhomogeneous Stokes system


 −4u + ∇p = f
in Ω.
div u = g

Then ∇2 u, ∇p ∈ L2 (Ω) and the coercive estimate
k∇2 uk2,Ω + k∇pk2,Ω ≤ c(n, Ω)(k∇gk2,Ω + kf k2,Ω )
holds.
Proof To demonstrate the essence of the matter, we restrict ourselves

to the case Ω = R3+ .
Step 1 Here, we are going to estimate tangential derivatives of u, i.e.,
derivatives with respect to xα , α = 1, 2. Let h = (h1 , h2 , 0) be a vector in R3
and 4h f (x) := f (x + h) − f (x). We have

 −44h u + ∇4h p = 4h f
div 4h u = 4h g

with 4h u ∈ V (R3+ ). According to Proposition 3.2 of Chapter I, there exists

wh ∈ V (R3+ ) such that div wh = 4h g and
k∇wh k2,R3+ ≤ ck4h gk2,R3+
with a constant c independent of h. Then the previous system can be trans-

formed to the following form:
 −4(4h u − wh ) + ∇4h p = 4h f + 4wh ∈ L−1

 n
2 (R+ )
in R3+ .
div (4h u − wh ) = 0

From the first equation of the above system, it follows that
k∇(4h u − wh )k2,R3+ ≤ k4h f + 4wh kL−1 3 .

2 (R+ )
Now, our aim is to evaluate the right hand side of the latter inequality.
By the definition, we have
1
k4h f + 4wh kL−1 n =
2 (R+ )
|h|
n 1 Z 1
Z o
= sup 4h f · vdx + 4wh · vdx : v ∈ C0∞ (R3+ ), k∇k2,R3+ ≤ 1 =
|h| |h|
R3+ R3+
Z Z
n 1 1 o
= sup − f · 4h vdx − ∇wh : ∇vdx : ... ≤
|h| |h|
R3+ R3+
n 1 Z 12 o
2 ∞ 3
≤ kf k2,R3+ sup |4h v| dx : v ∈ C0 (R+ ), k∇k2,R3+ ≤ 1 +
|h|
R3+
2.2. COERCIVE ESTIMATES 43
1 1
+ k∇wh k2,R3+ ≤ kf k2,R3+ + k∇wh k2,R3+ .
|h| |h|
In the last line, there has been used the following fact (exercise)
n 1 Z 12 o
sup |4h v|2 dx : v ∈ C0∞ (R3+ ), k∇vk2,R3+ ≤ 1 ≤ 1.
|h|
R3+
So, we have
1 h 1 i
k∇(4h u)k2,R3+ ≤ c kf k2,R3+ + k4h gk2,R3+ .
|h| |h|
Tending h to zero, we find the bound for tangential derivatives of v
k∇u,α k2,R3+ ≤ cI,
where α = 1, 2 and
I = kf k2,R3+ + k∇gk2,R3+ .
Step 2 Let us start with evaluation of terms u3,33 and p,3 . u3,33 can be
estimated simply with the help of the equation div u = uα,α + u3,3 = 0. This
gives us u3,33 = −uα,3α and thus
ku3,33 k2,R3+ ≤ cI.
As to the second term, the above estimate, the equation p,3 = f3 + 4u3 , and
bounds for tangential derivatives lead to the inequality
kp,3 k2,R3+ ≤ cI.
Next, it is easy to verify (exercise) that:
g ∈ L2 (Ω) ⇒ k∇gkL−1
2 (Ω)
≤ ckgkL2 (Ω) .
With this observation, we shall estimate tangential derivatives of the pressure

p. Indeed, by the Stokes equations p,α , α = 1, 2, we have p,α = fα + 4uα
and thus
p,αβ = fα,β + uα,iiβ = fα,β + (uα,iβ ),i ∈ L−1 3
2 (R+ )
and
p,α3 = f3,α + u3,iiα = f3,α + (u3,iα ),i ∈ L−1 3
2 (R+ ).
So, by the aforesaid observation and by the latter estimates, we can conclude
that
k∇p,α kL−1 3
2 (R+ )
≤ cI.
On the other hand, by Theorem 4.1 of Chapter I,
kp,α k2,R3+ ≤ ck∇p,α kL−1 3

2 (R+ )
≤ cI.
So, the full gradient of the pressure p obeys the estimate
k∇pk2,R3+ ≤ cI.
The remaining part of the second derivatives can be estimated with the
help of the equation uα,33 = −uα,ββ + p,α − fα and previous bounds. Indeed,
we have
k∇2 uk2,R3+ ≤ cI
and this completes the proof.
Remark 2.2. The Stokes system holds a.e. in Ω provided assumptions of

Proposition 2.1.
In fact, we have more general statement, which are called Cattabriga-

Solonnikov estimates.
Theorem 2.3. Assume that all assumptions of Proposition 2.1 are fulfilled.
Let Ω be a bounded domain with sufficiently smooth boundary. In addition,
assume that
f ∈ Wrk (Ω), g ∈ Wrk+1 (Ω)
with [g]Ω = 0 and with integer k. Then
h i
k∇2 ukWrk (Ω) + k∇pkWrk (Ω) ≤ c(n, r, k, Ω) kf kWrk (Ω) + k∇gkWrk (Ω) .
2.3 Local Regularity

Proposition 3.1. Assume that we are given functions
v ∈ W21 (B+ ), q ∈ L2 (B+ ), f ∈ L2 (B+ ), g ∈ W21 (B+ ),

2.3. LOCAL REGULARITY 45
satisfying the stokes system


 −4v + ∇q = f
in B+
div v = g

and the boundary condition

v|x3 =0 = 0.
Then, for any τ ∈]0, 1[,
∇2 v, ∇q ∈ L2 (B+ (τ ))
and the following estimate is valid:

h
2
k∇ vk2,B+ (τ ) + k∇qk2,B+ (τ ) ≤ c(τ ) kf k2,B+ + kqk2,B+ +
i
+kgkW21 (B+ ) + kvkW21 (B+ ) .
Proof Let a cut-off function ϕ ∈ C0∞ (R3 ) possess the properties: 0 ≤

ϕ ≤ 1, ϕ ≡ 1 in B(τ ), and ϕ ≡ 0 outside B(1). Introducing new functions
u = ϕv and p = ϕq, we can verify that they satisfy the following system:
 −4u + ∇p = f˜ = ϕf − 2∇v∇ϕ − v4ϕ + q∇ϕ ∈ L2 (R3+ )


in R3+ ,
div v = g̃ = ϕg + v · ∇ϕ ∈ W21 (R3+ )

◦
By assumptions, u ∈ L12 (R3+ ) and p ∈ L2 (R3+ ) and thus we are in a position
to apply Proposition 2.1, which reads that ∇2 u, ∇p ∈ L2 (R3+ ) and
h i
k∇2 uk2,R3+ + k∇pk2,R3+ ≤ c k∇g̃k2,R3+ + kf˜k2,R3+ .
Then all the statements of Proposition 3.1 follow.

The statement below can be proven in the same way as Proposition 3.2.
Proposition 3.2. Assume that we are given functions
v ∈ W21 (B), q ∈ L2 (B), f ∈ L2 (B), g ∈ W21 (B),

satisfying 
 −4v + ∇q = f
in B.
div v = g

Then, for any τ ∈]0, 1[,
∇2 v, ∇q ∈ L2 (B(τ ))
and the following estimate is valid:

h
k∇2 vk2,B(τ ) + k∇qk2,B(τ ) ≤ c(τ ) kf k2,B + kqk2,B +
i
+kgkW21 (B) + kvkW21 (B) .
2.4 Further Local Regularity Results, n = 2, 3

Proposition 4.1. Assume that a divergence free vector field v ∈ W21 (B)
obeys the identity Z
∇v : ∇wdx = 0
B
∞
for any w ∈ C0,0 (B). Then
Z
2
sup |∇v(x)| ≤ c(n) |∇v|2 dx.
x∈B(1/2)
B
Proof As explained earlier, one can introduce the pressure q ∈ L2 (B)

with [q]B = 0 such that

 −4v + ∇q = 0
in B
div v = 0

in the sense of distributions and
kqk2,B ≤ ck∇vk2,B . (2.4.1)

2.4. FURTHER LOCAL REGULARITY RESULTS, N = 2, 3 47
Step 1 Let v̄ = v − [v]B and fix 1/2 < τ1 < 1. By previous results, see
Proposition 3.2,
Z hZ Z Z i
2 2 2
(|∇ v| + |∇q| )dx ≤ c(τ1 , n) |v̄| dx + |∇v| dx + |q|2 dx .
2 2
B(τ1 ) B B B
According to Poincaré’s inequality

Z Z
|v̄| dx ≤ c(n) |∇v|2 dx
2
B B
and to estimate (2.4.1),

Z Z
(|∇ v| + |∇q| )dx ≤ c(τ1 , n) |∇v|2 dx ≡ cI.
2 2 2
B(τ1 ) B
Step 2 Now, obviously, functions v,k and q,k obey the system

 −4v,k + ∇q,k = 0
in B(τ1 )
div v,k = 0

in the sense of distributions. Repeating the same arguments in two balls

B(τ2 ) and B(τ1 ) with 1/2 < τ2 < τ1 , we find
Z Z
2 2 2
(|∇ v,k | + |∇q,k | )dx ≤ c(τ2 , τ1 , n) (|∇2 v|2 + |∇q|2 )dx ≤ c(τ2 , τ1 , n)I.
B(τ2 ) B(τ1 )
As a result,
Z l
X l−1
X
i 2
|∇ v| + |∇i q|2 dx ≤ c(l, n)I.
i=1 i=1
B(1/2)
Taking l = 3 and using Sobolev’s imbedding theorem, we show
sup (|∇v(x)|2 + |q(x)|2 ) ≤ c(n)I.

x∈B(1/2)
The proof of the following statement is slightly more complicated but still
can be made along similar lines.
Proposition 4.2. Assume that a divergence free vector field v ∈ W21 (B+ )
satisfies the boundary condition
v|x3 =0 = 0
and the identity Z

∇v : ∇wdx = 0
B+
∞
for any w ∈ C0,0 (B+ ). Then
Z
2
sup |∇v(x)| ≤ c(n) |∇v|2 dx.
x∈B+ (1/2)
B+
Proof First, we recover the pressure q ∈ L2 (B+ ) with [q]B+ = 0 such

that 
 −4v + ∇q = 0
in B+
div v = 0

in the sense of distributions with the estimate
kqk2,B+ ≤ ck∇vk2,B+ . (2.4.2)
Fix 1/2 < τ1 < 1. By Proposition 3.1, we have additional regularity so that
Z Z h i
2 2 2
(|∇ v| + |∇q| )dx ≤ c(τ1 , n) |v|2 + |∇v|2 + |q|2 dx.
B+ (τ1 ) B+
Since v|x3 =0 = 0, Poincaré type inequality ensures the bound:

Z Z
|v| dx ≤ c(n) |∇v|2 dx
2
B+ B+
and, by (2.4.2),
Z Z
2 2 2
(|∇ v| + |∇q| )dx ≤ c(τ1 , n) |∇v|2 dx ≡ cI.
B+ (τ1 ) B+
2.4. FURTHER LOCAL REGULARITY RESULTS, N = 2, 3 49
Tangential derivatives of v and q satisfy the same equations and boundary

conditions: 
 −4v,α + ∇q,α = 0
in B+
div v,α = 0

and
v,α |x3 =0 = 0.
Assume that n = 3 (n = 2 is an exercise). We have for 1/2 < τ2 < τ1
Z Z
(|∇ v,α | + |∇q,α | )dx ≤ c(τ1 , n) |∇v|2 dx ≡ cI.
2 2 2
B+ (τ2 ) B+
It remains to evaluate vi,333 and q,33 . To this end, we are going to exploit
the incompressibility condition: v3,333 = −vα,α33 ∈ L2 (B+ (τ2 )), which gives
us the bound Z
|∇3 v3 |2 dx ≤ cI.
B+ (τ2 )
To estimate vα,333 , α = 1, 2, one can make use of the equation

q,3i = 4v3,i
and conclude that Z
|∇q,3 |2 dx ≤ cI.
B+ (τ2 )
Now, we are going to use the equations −4vα,3 + q,α3 = 0 one more time and
find
vα,333 = −vα,ββ3 + qα3 ∈ L2 (B+ (τ2 )).
The latter implies Z
|vα,333 |2 dx ≤ cI.
B+ (τ2 )
So, the final estimate

Z
(|∇3 v|2 + |∇2 v|2 + |∇v|2 + |∇2 q|2 + |∇q|2 + |q|2 )dx ≤ cI
B+ (τ2 )
comes out and it implies

sup |∇v(x)|2 ≤ c(n)I.
x∈B+ (1/2)
2.5 Stokes Operator in Bounded Domains

In this section, we always assume that Ω ∈ Rn has smooth boundary, and
n = 2 or 3.
By Ladyzheskaya’s theorem, given f ∈ L2 (Ω), there exists a unique f1 ∈
◦
J (Ω) such that
f = f1 + ∇q
with q ∈ W21 (Ω). We let P f := f1 . The operator P : L2 (Ω) → L2 (Ω) is
called the Leray projector.
It is worthy to notice that the Dirichlet problem

 −4u + ∇p = f ∈ L2 (Ω)
in Ω,
div u = 0

u|∂Ω = 0
can be transformed in the following way
 ◦
 −4u + ∇p1 = f1 ∈ J (Ω)
in Ω,

div u = 0
u|∂Ω = 0,
where f1 = P f and p1 = p − q. So, without loss of generality, we always may
◦
assume that the right hand side in the Stokes system belongs to J (Ω).
We know that
k∇2 uk2,Ω + k∇pk2,Ω ≤ ckf k2,Ω .
We can also re-write the Dirichlet problem in the operator form
4u
e = f,
where ◦ ◦
4
e := P 4 : J (Ω) → J (Ω)
is unbounded operator with the domain of the definition

◦
e := {u ∈ W 2 (Ω) : div u = 0, u|∂Ω = 0 } = J 1 (Ω) ∩ W 2 (Ω).
dom 4 2 2 2
2.5. STOKES OPERATOR IN BOUNDED DOMAINS 51
It is called the Stokes operator.

The Stokes operator 4 e has the similar properties as the Laplace operator
under the Dirichlet boundary conditions. Let us list these properties:
(i) the Stokes operator has a discrete spectrum
◦
−4u
e = λu, u ∈ J (Ω), u 6= 0,
0 < λ1 < λ2 < ... < λm < ..., λm → ∞,
(ii) dim ker (−4

e − λk I) is finite for each k ∈ N,
(iii) the set {ϕk }∞
k=1 of eigenvectors (eigenfunctions) of the Stokes operator is
◦
an orthogonal basis in J (Ω) so that (ϕk , ϕj ) = δij ,
◦
(iv) the set {ϕk }∞ 1
k=1 is an orthogonal system in J 2 (Ω) as well as in dom 4 so
e
that λk = k∇ϕk k22,Ω = k4ϕ e k k2,Ω ,
◦
(v) if fP∈ J (Ω), then kf k22,Ω = ∞ 2
P
k=1 |ck | < ∞, where ck = (f, ϕk ), and the
series ∞ k=1 ck ϕk converges to f in L2 (Ω),
◦
if f ∈ J 12 (Ω), then k∇f k22,Ω = ∞
P 2
P∞
k=1 |ck | λk < ∞ and series k=1 ck ϕk
converges to f in W21 (Ω),
if f ∈ dom 4, e then k∇f k2 = P∞ |ck |2 λ2 < ∞ and series P∞ ck ϕk
2,Ω k=1 k k=1
converges to f in W22 (Ω).
The proof of all above statements is based on the Hilbert-Shmidt theorem
and the compactness of the embedding of W21 (Ω) into L2 (Ω).
◦
Let us describe extension of 4 e to J 1 (Ω). We know that
2
◦ ◦
e : J 1 (Ω) ∩ W 2 (Ω) → J (Ω)
4 2 2
◦
is a bijection. Given u ∈ J 12 (Ω) ∩ W22 (Ω), we have
(−4u,
e v) = (−4u + ∇p, v) = (−4u, v) = (∇u, ∇v)
∞
for any v ∈ C0,0 (Ω). From the latter identity, we immediately derive the
following estimate
k4uk
e ◦ ≤ k∇uk2,Ω = kuk ◦ 1 .
(J 12 (Ω))0 J 2 (Ω)
◦
Here, we use the identification of the dual space (J 12 (Ω))0 described in Section
◦ ◦
4 of Chapter I with V = J 12 (Ω) and H = J (Ω) and in what follows we are
not going to use any special notation for this particular identification. Since
◦ ◦
the space J 12 (Ω) ∩ W22 (Ω) is dense in J 12 (Ω), there exists a unique extension of
◦ ◦
the Stokes operator 4 e (denoted again by 4) e from J 1 (Ω) ∩ W 2 (Ω) to J 1 (Ω).
2 2 2
Moreover, we have the following statement:
◦ ◦
e : J 1 (Ω) → (J 1 (Ω))0 is a bijection.
Proposition 5.1. (i) The extension 4 2 2
◦
(ii) If f ∈ (J 12 (Ω))0 , then
∞
X
kf k2◦ 1 = fk2 /λk ,
(J 2 (Ω))0
k=1
where fk = (f, ϕk ).
◦ ◦
e : J 1 (Ω) → 4(
Proof of Proposition 5.1 Obviously, 4 e J 1 (Ω)) is a
2 2
bijection. Our aim is to show that
◦ ◦
e J 1 (Ω)) = (J 1 (Ω))0 .
4( (2.5.1)
2 2
◦
Lemma 5.2. (i) for f ∈ (J 12 (Ω))0 , we have
∞
X
2
kf k ◦ ≤ fk2 /λk .
(J 12 (Ω))0
k=1
(ii) if
∞
X
fk2 /λk < ∞,
k=1
P∞ ◦ ◦
then the series k=1 fk ϕk converges to f in (J 12 (Ω))0 , f ∈ 4(
e J 1 (Ω)), and
2
∞
X
2
kf k ◦ = fk2 /λk .
(J 12 (Ω))0
k=1
2.5. STOKES OPERATOR IN BOUNDED DOMAINS 53
◦
Proof Fix an arbitrary function a ∈ J 12 (Ω)), then
N
X
N
a = ak ϕk → a
k=1
◦
in J 12 (Ω)). So,
N
X
N N
(f, a) = lim (f, a ) = lim (f, a ) = lim f k ak ≤
N →∞ N →∞ N →∞
k=1
N
X N
21 X 12 ∞
X 12
≤ fk2 /λk a2k λk ≤ fk2 /λk k∇ak2,Ω .
k=1 k=1 k=1
The latter certainly implies (i)

P∞ ◦
(ii) First, let us show that the series k=1 fk ϕk converges in (J 12 (Ω))0 .
Indeed, let fN = N
P
k=1 fk ϕk and then, by (i),
N
X
2
kfN − fM k ◦ ≤ fk2 /λk → 0
(J 12 (Ω))0
k=M +1
as M, N → 0.
◦
We denote by f ∈ (J 12 (Ω))0 the sum of our series. Then, by (i),
∞
X
kf − fN k2◦ 1 ≤ fk2 /λk → 0
(J 2 (Ω))0
k=N +1
and thus
kfN k ◦ → kf k ◦ .
(J 12 (Ω))0 (J 12 (Ω))0
◦
e J 1 (Ω)). Indeed, we have
Now, we are going to prove that f ∈ 4( 2
N
X N
X
fN = fk ϕk = 4
e fk ϕk /λk = 4u
e N,
k=1 k=1
where
N
X ◦
uN = fk ϕk /λk ∈ J 12 (Ω) ∩ W22 (Ω).
k=1
By direct computations,
N
X
k∇uN − ∇uM k22,Ω = fk2 /λk → 0.
k=M +1
Then by definition of the extension of 4,

e
4u
e N → 4u
e = f.
Next, we have
N
X
kfN k2◦ 1 e N k2 ◦
= k4u 1
= k∇uN k22,Ω = fk2 /λk → kf k2◦ 1 .
(J 2 (Ω))0 (J 2 (Ω))0 (J 2 (Ω))0
k=1
Lemma 5.3.
∞
◦ ◦ X
e J 1 (Ω)) = {f ∈ (J 1 (Ω))0 :
4( fk2 /λk < ∞} =: U.
2 2
k=1
Proof According to Lemma 5.2 (ii), we have

◦
e J 1 (Ω)).
U ⊆ 4( 2
◦ ◦
e J 1 (Ω)), i.e., f = 4
Now, assume that f ⊆ 4( e u for some u ∈ J 1 (Ω). Then
2 2
we have
fk = (f, ϕk ) = (4e u, ϕk ) = (u, 4
e ϕk ) = λk uk .
Since ∞
X
k∇uk22,Ω = u2k λk < ∞,
k=1
we find ∞
X
fk2 /λk < ∞.
k=1
◦
e J 1 (Ω)).
So, f ∈ U and thus U ∈ 4( 2
Now, we proceed with the proof of Proposition 5.1. We are done, if show
that ∞
◦ X
1 0
f ∈ (J 2 (Ω)) ⇒ fk2 /λk < ∞.
k=1
2.6. COMMENTS 55
Next, we let
N
X
N
ak = fk /λk , a = ak ϕk .
k=1
Then
N
X N
X
k∇aN k22,Ω = |ak | 2
k∇ϕk k22,Ω = fk2 /λk .
k=1 k=1
Next, we have
N
X N
X 21
N
(f, a ) = fk2 /λk ≤ kf k ◦
N
k∇a k2,Ω = kf k ◦ fk2 /λk
(J 12 (Ω))
0 (J 12 (Ω))0
k=1 k=1
which implies
N
X
fk2 /λk ≤ kf k2◦ 1
(J 2 (Ω))0
k=1
for any N ∈ Rn . This completes our proof of Proposition 5.1.
2.6 Comments
Chapter 2 contains standard results on linear stationary Stokes system in-
cluding the notion of Stokes operator in smooth bounded domains. In ad-
dition, various global and local interior and boundary regularity results are
discussed.
Chapter 3
Non-Linear Stationary Problem
3.1 Existence of Weak Solutions

Consider the Dirichlet boundary value problem for the classical stationary
Navier-Stokes system

 −ν4u + u · ∇u + ∇p = f
in Ω, (3.1.1)
div u = 0

u|∂Ω = 0 (3.1.2)
and if n = 3 and Ω is unbounded then u(x) → 0 as |x| → ∞. Here, ν is a
positive parameter called viscosity. We always assume that
f ∈ L−1
2 (Ω).
Definition 1.1. A function u ∈ V̂ (Ω) is called a weak solution to boundary

value problem (3.1.1) and (3.1.2) if
ν(∇u, ∇v) = (u ⊗ u, ∇v) + (f, v)
∞
for any v ∈ C0,0 (Ω).
For n = 2 or 3, the imbedding theorems ensure that
u ∈ L4,loc (Ω).
So, the first term on the right hand side in the identity of Definition 1.1 is
well-defined.
57
58 CHAPTER 3. NON-LINEAR STATIONARY PROBLEM
If domain Ω is bounded and has Lipschitz boundary, then
u ∈ L4 (Ω).
Proposition 1.2. Let Ω be a bounded Lipschitz domain. Then boundary

value problem (3.1.1) and (3.1.2) has at least one weak solution.
Proof Let us reduce our boundary value problem to a fixed point prob-
lem and try to apply the celebrated Leray-Schauder principle.
Theorem 1.3. (Leray-Schauder principle) Let X be a separable Banach

space, B : X → X be a continuous operator. Assume that the operator
B has the following additional properties:
(i) B is compact operator, i.e., the operator that maps bounded sets of
X into precompact sets in X, in other words, B is a completely continuous
operator;
(ii) all possible solutions to the equation
u = λB(u)
satisfy the inequality kukX < R with R independent of λ ∈ [0, 1].

Then the operator B has at least one fixed point u, i.e., u = B(u).
We define, as usual, [u, v] := (∇u, ∇v) a scalar product on V (Ω) that

coincides with V̂ (Ω) under assumptions of the proposition. It is not difficult
to prove that, for any w ∈ V (Ω),
div w × w ∈ L−1
2 (Ω).
According to our results in Chapter II, given w ∈ V (Ω), there exists a unique
u ∈ V (Ω) such that
ν(∇u, ∇v) = (w ⊗ w, ∇v) + (f, v)
for any v ∈ V (Ω). By Riesz representation theorem, we can define an oper-

ator A : V (Ω) → V (Ω) so that
[A(w), v] := (w × w, ∇v)
and
[F, v] := (f, v).
3.1. EXISTENCE OF WEAK SOLUTIONS 59
So, the previous identity can be re-written in the operator form

1
u= (A(w) + F ).
ν
Then, the existence of weak solutions is equivalent to the existence of fixed
point of the latter operator equation.
First, let us show that A is a completely continuous operator. To this
end, we take an arbitrary weakly converging sequence such that
w(k) * w
in V (Ω). Then, the compactness of the imbedding of V (Ω) into L4 (Ω) gives
us:
w(k) ⊗ w(k) → w ⊗ w
in L2 (Ω). From the main identity, it follows that
ν[u(k) − u(m) , v] = (w(k) ⊗ w(k) − w(m) ⊗ w(m) , ∇v) = 0
for any v ∈ V (Ω). It remains to plaque v = u(k) −u(m) into the latter relation
and make use of the fact that
kw(k) ⊗ w(k) − w(m) ⊗ w(m) k2,Ω → 0
as k, m → ∞. So, complete continuity of A has been proven.

Now, we need to get estimates of all possible solutions to the equation
νuλ = λA(uλ ) + F,
depending on a parameter λ ∈ [0, 1]. Since (uλ ⊗ uλ , ∇uλ ) = 0, we have
ν[uλ , uλ ] = (uλ ⊗ uλ , ∇uλ ) + (f, uλ ) ≤ kf kL−1

2 (Ω)
k∇uλ k2,Ω
and thus
1
k∇uλ k2,Ω ≤kf kL−1
2 (Ω)
.
ν
The right hand side of the latter inequality is independent of λ and thus
the existence of at least one fixed point follows from the Leray-Schauder
principle.
Regarding the uniqueness of weak solutions, we have the following
Lemma 1.4. Assume that all assumptions of Proposition 1.2 hold. Let in
addition
c20 (n, Ω)|
kf kL−1
2 (Ω)
< 1,
ν2
where c0 (n, Ω) is a constant in the inequality
kvk4,Ω ≤ c0 (n, Ω)k∇vk2,Ω (3.1.3)

◦
for any v ∈ L12 (Ω).
Then, our boundary value problem (3.1.1) and (3.1.2) has a unique weak
solution.
Proof Let u1 and u2 be two different solutions to boundary value prob-

lem (3.1.1), (3.1.2). Then, we have
ν[u1 − u2 , u1 − u2 ] = (u1 ⊗ u1 − u2 ⊗ u2 , ∇(u1 − u2 )) =
(u1 ⊗ (u1 − u2 ), ∇(u1 − u2 )) + ((u1 − u2 ) × u2 , ∇(u1 − u2 )) =

= (u1 ⊗ (u1 − u2 ), ∇(u1 − u2 )) ≤ ku1 k4,Ω ku1 − u2 k4,Ω k∇(u1 − u2 )k2,Ω .
Applying inequality (3.1.3) twice and taking into account the last estimate
in the proof of Proposition 1.2 for u1 , i.e.,
1
k∇u1 k2,Ω ≤ kf kL−1 (Ω),
ν 2
we find
νk∇(u1 − u2 )k22,Ω ≤ c20 k∇u1 k22,Ω k∇(u1 − u2 )k22,Ω ≤
c20
≤ k∇(u1 − u2 )k22,Ω kf kL−1 (Ω).
ν 2
This, by contradiction, implies the statement of the lemma.
Proposition 1.5. Assume that unbounded domain Ω satisfies the condition

V (Ω) = V̂ (Ω). Then problem (3.1.1) and (3.1.2) has at least one weak
solution satisfying the estimate
1
k∇uk2,Ω ≤ kf kL−1 (Ω).
ν 2
3.1. EXISTENCE OF WEAK SOLUTIONS 61
Proof Let R 1. Consider problem (3.1.1), (3.1.2) in ΩR := B(R) ∩ Ω.

By Proposition 1.2, there exists uR ∈ V (ΩR ), satisfying the identity
ν(∇uR , ∇v)ΩR = (uR ⊗ uR , ∇v)ΩR + (f, v)ΩR
∞
for any v ∈ C0,0 (ΩR ). Extending uR by zero to the whole domain Ω, we
notice that
1 1
k∇uR k2,Ω = k∇uR k2,ΩR ≤ kf kL−1 (ΩR ) ≤ kf kL−1 (Ω).
ν 2 ν 2
The latter allows us to select a subsequence, still denoted by uR , with the

following properties:
∇uR * ∇u
in L2 (Ω) and
uR → u
in L2,loc (Ω). It remains to pass to the limits as R → ∞ in the identity for
uR and show that
ν(∇u, ∇v) = (u ⊗ u, ∇v) + (f, v)
∞
for any v ∈ C0,0 (Ω), which means that u is a required weak solution.
Now, the question is whether we can recover the pressure? We shall
consider two cases.
Case 1 Here, we assume that Ω is a bounded domain with Lipschitz
boundary. Since, for v ∈ C0∞ (Ω),
l(v) := ν(∇u, ∇v) − (u ⊗ u, ∇v) − (f, v) ≤
≤ Ck∇vk2,Ω ,
with a positive constant C = C(ν, k∇uk2,Ω , kuk4,Ω , kf kL−1
2
(Ω)), and l(v) = 0
∞
for any v ∈ C0,0 (Ω), we can use the same arguments as before to recover the
pressure. According to them, there exists p ∈ L2 (Ω) such that
ν(∇u, ∇v) = (u ⊗ u, ∇v) + (f, v) + (p, div v)
for any v ∈ C0∞ (Ω).
Case 2 Here, we can use a familiar procedure, described in Section 1,
where ∞
[
Ω= Ωm , Ωm ⊂ Ωm+1 ,
m=1
and Ωm is a bounded Lipschitz domain. Since u ∈ L4,loc (Ω) implies u ∈

L4 (Ωm ), one can state that there exists pm ∈ L2 (Ωm ) such that
ν(∇u, ∇v) = (u ⊗ u, ∇v) + (f, v) + (pm , div v)
for any v ∈ C0∞ (Ωm ). Moreover, we can fix pm so that pm = pm+1 in Ωm . So,
now, if we introduce a function p, letting p = pm in Ωm , then p ∈ L2,loc (Ω)
and the following identity is valid:
ν(∇u, ∇v) = (u ⊗ u, ∇v) + (f, v) + (p, div v)
for any v ∈ C0∞ (Ω).
3.2 Regularity of Weak Solutions

We need the following known auxiliary statement.
Lemma 2.1. Let a non-decreasing function Φ :]0, R0 ] → R+ satisfy the

following condition:
% m
Φ(%) ≤ c + ε Φ(R) + CRs (3.2.1)
R
for any 0 < % < R ≤ R0 , for some positive constants c, C, ε > 0, and for
some m > s > 0.
There exist positive numbers ε0 = ε0 (m, s, c) and c1 = c1 (m, s, c) such
that if ε < ε0 , then h % s i
Φ(%) ≤ c1 + C%s (3.2.2)
R0
for any 0 < % ≤ R0 .
Proof Let % = τ R, 0 < τ < 1, ε0 = τ m . So, if ε < ε0 , then

m−s m+s
Φ(τ R) ≤ 2cτ m Φ(R) + CRs = 2cτ 2 τ 2 Φ(R) + CRs ≤
m+s
≤τ 2 Φ(R) + CRs .
m−s
If we select ε0 so that 2cτ 2 ≤ 1, then, after iterations, we have
m+s m+s m+s
Φ(τ k R0 ) ≤ τ k 2 Φ(R0 ) + Cτ s R0s (1 + τ 2 + ... + τ 2
(k−1)
)≤
3.2. REGULARITY OF WEAK SOLUTIONS 63
m+s 1
≤ τk 2 Φ(R0 ) + Cτ s R0s
m−s .
1−τ 2
Given 0 < % ≤ R0 , we find an integer number k such that
R0 τ k+1 < % ≤ R0 τ k .
Then
1 % s % s 1
k
Φ(%) ≤ Φ(τ R0 ) ≤ Φ(R0 ) + C .
τR τ 1 − τ m−s
2
Lemma 2.2. Let a divergence free vector-valued function u ∈ W21 (B(R)) and
a tensor-valued function F ∈ Lr (B(R)), with r > n = 3, satisfy the identity
Z Z Z
∇u : ∇vdx = u ⊗ u : ∇vdx + F : ∇vdx
B(R) B(R) B(R)
∞
for any v ∈ C0,0 (B(R)).
Then,
Z % 3 Z 31 Z
2 6
|∇u| dx ≤ c +R |u| dx |∇u|2 dx+
R
B(%) B(R) B(R)
2
Z r2
+cR3(1− r ) |F |r dx
B(R)
for 0 < % ≤ R. Here, c is a universal positive constant.

Lemma 2.3. Let a divergence free vector-valued function u ∈ W21 (B+ (R)),
with u|x3 =0 = 0, and a tensor-valued function F ∈ Lr (B+ (R)), with r > n =
3, satisfy the identity
Z Z Z
∇u : ∇vdx = u ⊗ u : ∇vdx + F : ∇vdx
B+ (R) B+ (R) B+ (R)
∞
for any v ∈ C0,0 (B+ (R)).
Then,
Z % 3 Z 31 Z
2 6
|∇u| dx ≤ c +R |u| dx |∇u|2 dx+
R
B+ (%) B+ (R) B+ (R)
Z r2
3(1− r2 ) r
+cR |F | dx
B+ (R)
for 0 < % ≤ R. Here, c is a universal positive constant.
Proof of Lemma 2.2 We know that div u ⊗ u ∈ L−1 2 (B(R)). Hence,

there exist ◦
ũR ∈ J 12 (B(R)), p̃R ∈ L2 (B(R)),
with [p̃R ]B(R)) = 0, so that

 −ν4ũR + ∇p̃R = −div u ⊗ u − div F
in B(R). (3.2.3)
div ũR = 0

Multiplying the first equation in (3.2.3) by ũR and integrating the product
by parts, we find
Z Z Z
2
|∇ũR | dx = (u ⊗ u − [u ⊗ u]B(R) ) : ∇ũR dx + F : ũR dx
B(R) B(R) B(R)
and, therefore, after application of the Cauchy-Schwartz inequality

Z Z Z
2 2 2
|∇ũR | dx ≤ 2 |u ⊗ u − [u ⊗ u]B(R) | dx + |F | dx .
B(R) B(R) B(R)
Next, we estimate the first term on the right hand side of the latter relation
with the help of Galliardo-Nirenberg inequality and, then, with the help of
Hölder inequality. As a result, we have
Z Z 6
35
2
|u ⊗ u − [u ⊗ u]B(R) | dx ≤ c |∇(u ⊗ u)| dx ≤
5
B(R) B(R)
Z 6 6
53 Z 31 Z 3
43
6
≤c |u| |∇u| dx
5 5 ≤c |u| dx |∇u| dx2 ≤
B(R) B(R) B(R)
Z 31 Z
≤c |u|6 dx R |∇u|2 dx,
B(R) B(R)
where c is a universal constant.

Let uR = u − ũR . This function satisfies the identity
Z
∇uR : ∇vdx = 0
B(R)
∞
for any v ∈ C0,0 (B(R)). By the results of Chapter II, see Section 2, we have
the following estimate
Z % 3 Z
2
|∇uR | dx ≤ c |∇uR |2 dx,
R
B(%) B(R)
which, in turn, implies another one:

Z % 3 Z Z
2 2
|∇u| dx ≤ c |∇u| dx + c |∇ũR |2 dx.
R
B(%) B(R) B(R)
At first, we apply our earlier estimates for

Z
|∇ũR |2 dx
B(R)
and, then, Hölder’s inequality for the term, containing F , in order to get the
estimate of Lemma 2.2.
Lemma 2.4. (Ch.-B. Morrey) Let u ∈ Wm1 (Ω) satisfy the condition
Z
|∇u|2 dx ≤ K%n−m+mα
B(%)
for some 0 < α < 1 and for any B(x0 , %) ⊂ Ω such that 0 < % < %0 with two
positive constants K and %0 .
α
Then u ∈ Cloc (Ω), i.e., u ∈ C α (Ω1 ) for any subdomain Ω1 b Ω.
Lemma 2.5. Assume that all assumptions of Lemma 2.2 hold with R = a.
Then
1− 3
u ∈ Cloc r (B(a)).
Proof We remind that the case n = 3 is considered only. Fix Ω1 b B(a)

and find Ω such that Ω1 b Ω b B(a). By shift, we have, for any B(x0 , R) ⊂
Ω, the following estimate
h % 3 i
Φ(x0 , %) ≤ c + RA Φ(x0 , R) + CR3−2+2α ,
R
with α = 1 − 3/R,
Z Z 13 Z r2
Φ(x0 , R) := |∇u|2 dx, A := 6
|u| dx , C := |F |r dx .
B(x0 ,R) B(a) B(a)
Now, we apply Lemma 2.1 for m = 3, s = 3 − 2 + 2α = 1 + 2α. Let us

suppose that RA < ε0 for all 0 < R ≤ R0 . Then
h % 3−2+2α i
3−2+2α
Φ(x0 , %) ≤ c1 Φ(x0 , R) + C%
R0
for any x0 ∈ Ω and for any 0 < % ≤ R0 , and
1 n ε0 o
R0 := min dist (∂B(a), Ω), .
2 A
So, we have Z
|∇u|2 dx ≤ K%3−2+2α
B(x0 ,%)
for any 0 < % ≤ R0 , where K = K(r, kukW21 (B(a)) , kF kr,B(a) , R0 ).
Lemma 2.6. Assume that all assumptions of Lemma 2.3 hold with R = a.
Then
u ∈ C α (B̄+ (b))
for any 0 < b < a with α = 1 − 3/r.
Proof We have two types of estimates. The first one is so-called ”inte-
rior”. For any b1 ∈]b, a[, the following estimate is valid:
Φ(x0 , %) ≤ K%1+2α (3.2.4)
for x0 ∈ B+ (b1 ), x30 ≥ 21 (a − b1 ), and 0 < % ≤ R0 = 21 min{a − b1 , %0 } with

%0 = ε0 /A. Here, K depends on r, R0 , kukW21 (B+ (a)) , and kF kr,B+ (a) .
The second estimate is ”boundary” one:

Z
Φ+ (x0 , %) := |∇u|2 dx ≤ K+ %1+2α (3.2.5)
B+ (a)
for x0 = (x00 , 0), |x00 | < 21 (a − b1 ), 0 < % ≤ R0 , and K+ depends on the same
arguments as K.
Now, let us denote by ũ extension of u to the whole ball B(a) by zero
and let Z
Φ̃(x0 , %) := |∇ũ|2 dx
B(a)
with 0 < % ≤ R0 and x0 ∈ B(b1 ).

Consider two cases: x30 ≥ 21 (a − b1 ) and x30 < 12 (a − b1 ). In the first
case, we may use our ”interior” estimate (3.2.4) and the definition of ũ. As
a result, we arrive at the inequality
Φ̃(x0 , %) ≤ K%1+2α . (3.2.6)

In the second case, we first assume that x30 > 0 and if x30 ≥ %, we still
have estimate (3.2.6). Now, suppose that x30 < %. Then, by (3.2.5), we have
Z Z
2
Φ̃(x0 , %) = |∇u| dx ≤ |∇u|2 dx ≤
B(x0 ,%)∩B+ (a) B+ ((x00 ,0),%+x30 )
≤ K+ (% + x30 )1+2α ≤ 21+2α K+ %1+2α .

Now, assume that x30 < 0. If |x30 | ≥ 0, then, obviously, Φ̃(x0 , %) = 0. So,
let us suppose that −x30 < %. Here,
Z Z
2
Φ̃(x0 , %) = |∇u| dx ≤ |∇u|2 dx ≤
B(x0 ,%)∩B+ (a) B+ ((x00 ,0),%)
≤ K+ %1+2α .
So, the statement of the lemma follows from Morrey’s condition on Hölder
continuity, see Lemma 2.4.
Proposition 2.7. Let u ∈ W21 (B(2a)) be a divergence free function and

satisfy the identity
Z Z
(∇u : ∇v − u ⊗ u : ∇v)dx = f · vdx
B(2a) B(2a)
∞
for any v ∈ C0,0 (B(2a)). If f is of class C ∞ in B(2a), then u is of class C ∞
in B(a).
Proof It is not so difficult to check that there exists a tensor-valued
function F of class C ∞ such that f = −div F . Then, the identity from the
statement of the proposition can be re-written in the following way
Z Z Z
∇u : ∇vdx = u ⊗ u : ∇v + F : ∇vdx
B(2a) B(2a) B(2a)
∞
for any v ∈ C0,0 (B(2a)). From Lemma 2.5, it follows that u belongs, at least,
to C(B(3a/2)). Using the same arguments as in Section 1, we can recover a
pressure p ∈ L2 (B(2a)) (exercise) so that

 −4u + ∇p = −div G := −div (u ⊗ u + F )
in B(2a).
div u = 0

Since div G ∈ L2 (B(3a/2)), we can apply results of Chapter II on properties

of solutions to the Stokes system and find
∇2 u ∈ L2 (B(a1 )) ⇒ ∇u ∈ L6 (B(a1 )), ∇p ∈ L2 (B(a1 ))
for any a < a1 < 23 a.

Next, we know that, for k = 1, 2, 3, functions u,k ∈ W21 (B(a1 )) and
p,k ∈ L2 (B(a1 )) satisfy the system

 −4u,k + ∇p,k = −div G,k
in B(a1 ).
div u,k = 0

Since ∇2 G ∈ L2 (B(a1 )), we can use the linear theory one more time and get:
∇3 u ∈ L2 (B(a2 )) ⇒ ∇2 u ∈ L6 (B(a2 )), ∇2 p ∈ L2 (B(a2 ))

for any a < a2 < a1 .

Then, for k, s = 1, 2, 3, functions u,ks ∈ W21 (B(a1 )) and p,ks ∈ L2 (B(a1 ))
satisfy the system

 −4u,ks + ∇p,ks = −div G,ks
in B(a3 ),
div u,ks = 0

where ∇3 G ∈ L2 (B(a2 )). The similar arguments allow us to deduce that
∇4 u ∈ L2 (B(a3 )) ⇒ ∇3 u ∈ L6 (B(a3 )), ∇3 p ∈ L2 (B(a3 ))
for any a < a3 < a2 . Proceeding, further, in the same way, we complete the
proof of the lemma.
Proposition 2.8. Let u ∈ W21 (B(2a)) be a divergence free function and

satisfy the conditions: u|x3 =0 = 0 and
Z Z
(∇u : ∇v − u ⊗ u : ∇v)dx = f · vdx
B+ (2a) B+ (2a)
∞
for any v ∈ C0,0 (B+ (2a)). If f is of class C ∞ in B(2a) ∩ {x3 ≥ 0}, then u is
∞
of class C in B(a) ∩ {x3 ≥ 0}.
Proof We start with our proof in a similar way as in the latter propo-
sition, i.e., we find F of class C ∞ in B(2a) ∩ {x3 ≥ 0} so that f = −div F .
Then, we recover the pressure p ∈ L2 (B+ (2a)), which gives us:

 −4u + ∇p = −div G := −div (u ⊗ u + F )
in B+ (2a),
div u = 0

u|x3 =0 = 0.
By Lemma 2.6, u ∈ C(B + (3a/2)) and, by the linear theory,
∇2 u ∈ L2 (B+ (a1 )) ⇒ ∇u ∈ L6 (B+ (a1 )), ∇p ∈ L2 (B+ (a1 ))
for any a < a1 < 23 a.

Next, for α = 1, 2 and for k = 1, 2, 3, we have functions u,k ∈ W21 (B+ (a1 ))
and p,k ∈ L2 (B+ (a1 )) satisfying the system

 −4u,k + ∇p,k = −div G,k
in B+ (a1 )
div u,k = 0

and the boundary condition
u,α |x3 =0 = 0,
where ∇2 G ∈ L2 (B+ (a1 )). Then, again, we apply the linear theory and
conclude that
∇2 u,α ∈ L2 (B+ (a2 )) ∇p,α ∈ L2 (B+ (a2 ))
for any a < a2 < a1 . We need to establish the same properties for ∇2 u,3
and ∇p,3 . To achieve this goal, it is sufficient to evaluate uk,333 and p,33 for
k = 1, 2, 3, which is, in fact, not so difficult. Indeed, denoting gik := −Gij,jk ,
we first use the incompressibility condition:
u3,333 = −uα,α33 ∈ L2 (B+ (a2 )).
For other derivatives, we use the equations:
p,33 = g33 + u3,kk3 ∈ L2 (B+ (a2 ))
and
uα,333 = −gα3 + p,α3 − uα,ββ3 ∈ L2 (B+ (a2 )).
So, we can state
∇3 u ∈ L2 (B+ (a2 )) ⇒ ∇2 u ∈ L6 (B+ (a2 )), ∇2 p ∈ L2 (B+ (a2 )).
Next, for α, β = 1, 2 and for k, j = 1, 2, 3, we have functions
u,kj ∈ W21 (B+ (a2 )) p,kj ∈ L2 (B+ (a2 ))
satisfying the conditions


 −4u,kj + ∇p,kj = −div G,kj
in B+ (a2 ),
div u,kj = 0

u,αβ |x3 =0 = 0,
where ∇3 G ∈ L2 (B+ (a2 )). Here, we are going to proceed as in the case of
the third derivatives. We let
hijk = −Gim,mjk .
From the linear theory, one can deduce that
∇2 u,αβ ∈ L2 (B+ (a3 )), ∇p,αβ ∈ L2 (B+ (a3 ))
for a < a3 < a2 . We start again with the incompressibility condition:
u3,333α = −uβ,β3α ∈ L2 (B+ (a3 )).
So,
∇3 u3,α ∈ L2 (B+ (a3 )).
Then,
p,33α = h33α + u3,jj3 ∈ L2 (B+ (a3 ))
and thus
∇2 p,α ∈ L2 (B+ (a3 )).
Next,
uβ,333α = −uβ,γγ3α + p,β3α − hβα3 ∈ L2 (B+ (a3 )).
So, we have
∇3 u,α ∈ L2 (B+ (a3 )).
Now, let us go back to the incompressibility condition:
u3,3333 = −uβ,β333 ⇒ ∇4 u3 ∈ L2 (B+ (a3 )).
For the pressure, we have
p,333 = h333 + 4u3,33 ⇒ ∇3 p ∈ L2 (B+ (a3 )).
Finally,
uα,3333 = −uα,ββ33 + p,α33 − hα33 ⇒ ∇4 uα ∈ L2 (B+ (a3 )).
Proceeding in a similar further, we complete the proof of the proposition.

Theorem 2.9. Let Ω be Rn , or Rn+ , or a bounded with smooth boundary.

Let u ∈ V̂ (Ω) be a weak solution to the stationary Navier-Stokes equations,
see Definition 1.1. Assume that the right hand side in these equations is of
class C ∞ in the closure of the domain Ω. Then u is also of class C ∞ in the
closure of the domain Ω.
Proof For Ω = Rn or Rn+ , the statement follows from Propositions 2.7

and 2.8.
3.3 Comments
Chapter 3 contains standard results on the existence and regularity of solu-
tions to non-linear stationary boundary value problem. The main point of
the chapter is the local regularity theory, which differs a bit from the theory
for standard elliptic systems.
Chapter 4
Linear Non-Stationary Problem
4.1 Derivative in Time

Let us recall some definitions from the theory of distributions. D(Ω) is a
vector space that consist of all elements, belonging to C0∞ (Ω), where the
convergence of a sequence of functions ϕk ∈ C0∞ (Ω) to a function ϕ ∈ C0∞ (Ω)
is understood in the following sense. There exists a compact K ⊂ Ω such
that suppϕk , suppϕ ⊂ K and ∇m ϕk → ∇m ϕ uniformly on K for any m ≥ 0.
The space of all linear functionals on D(Ω), being continuous with respect
to the above convergence in D(Ω), is denoted by D0 (Ω). Elements of D0 (Ω)
are called distributions.
We may consider the space D0 (a, b; D0 (Ω)). Given T ∈ D0 (a, b; D0 (Ω)), let
us denote by ∂t T or even by dtd T the following distribution
(∂t T (ϕ))(χ) = −T (ϕ)(∂t χ)
for any ϕ ∈ D(Ω) and for any χ ∈ D(a, b).

It is too general definition for our purposes and we are going to use
somewhat more specific. Let V be a Banach space, V ∗ be its dual space with
duality relation < v ∗ , v >.
Definition 1.1. Let v ∗ ∈ L1,loc (a, b; V ∗ ) (t 7→ v ∗ (·, t) ∈ V ∗ and t 7→ kv ∗ (·, t)kV ∗
is in L1,loc (a, b)). We call u∗ ∈ D0 (a, b; V ∗ ) derivative of v ∗ in t if and only if
Zb
< u∗ , v > (χ) = − < v ∗ (·, t), v(·) > ∂t χ(t)dt
a
73
74 CHAPTER 4. LINEAR NON-STATIONARY PROBLEM
for any v ∈ V and for any χ ∈ C0∞ (a, b). We let u∗ = ∂t v ∗ .
As usual, the left hand side of the above identity is written in the same
way as the right hand side, i.e.,
Zb Zb
< ∂t v ∗ (·, t), v(·) > χ(t)dt = − < v ∗ (·, t), v(·) > ∂t χ(t)dt
a a
for any v ∈ V and for any χ ∈ C0∞ (a, b), although the left hand side might
make no sense as Lebesgue’s integral.
Let us discuss the relationship between the introduced notion and and
the Sobolev derivatives. Assume that
Z
∗
V, V ∈ L1,loc (Ω), C0∞ (Ω) ⊂ V, ∗
< v , v >= v ∗ vdx,
Ω
∗
v ∈ L1,loc (a, b; L1,loc (Ω)) = L1,loc (Ω×]a, b[), (4.1.1)
∂t v ∗ ∈ L1,loc (Ω×]a, b[).
Then ∂t v ∗ is a usual Sobolev derivative of v ∗ in the domain Ω×]a, b[. To

understand why, we are going to use the following simple statement.
Lemma 1.2. Given ε > 0 and ϕ ∈ C0∞ (Ω×]a, b[), there exist positive integer
number N and functions ϕk ∈ C0∞ (Ω), χk ∈ C0∞ (a, b), k = 1, 2, ..., N such
that
N
X
kϕ − ϕk χk kC 1 (Ω×[a,b]) < ε.
k=1
Let us assume that Lemma 1.2 has been proved. Suppose that ∂t v ∗ is
the derivative in the sense of Definition 1.1 and satisfies assumptions (4.1.1).
Our aim is to show that it is Sobolev’s derivative as well. Take an arbitrary
ε > 0 and an arbitrary function ϕ ∈ C0∞ (Ω×]a, b[) and fix them. Clearly,
ϕ ∈ C0∞ (Ω0 ×]a0 , b0 [) for some Ω0 b Ω and for some a < a0 and b0 < b. Let
number N (ε) and functions ϕk and χk be from Lemma 1.2 for the domain
4.1. DERIVATIVE IN TIME 75
Ω0 ×]a0 , b0 [. Then we have

0
Zb Z Zb0 Z
− v ∗ (x, t)∂t ϕ(x, t)dxdt − ∂t v ∗ (x, t)ϕ(x, t)dxdt ≤

a0 Ω0 a0 Ω0
b0 N (ε)
Z Z X
∗
≤ v (x, t) ∂t ϕ(x, t) − ∂t ϕk (x)χk (t) dxdt+

a0 Ω0 k=1
b0 N (ε)
Z Z X
∗
+ ∂t v (x, t) ϕ(x, t) − ϕk (x)χk (t) dxdt ≤

a0 Ω0 k=1
N (ε)
X
≤ ckϕ − ϕk χk kC 1 (Ω0 ×[a0 ,b0 ]) kv ∗ kL1 (Ω0 ×]a0 ,b0 [) + k∂t v ∗ kL1 (Ω0 ×]a0 ,b0 [)
k=1

≤ cε kv ∗ kL1 (Ω0 ×]a0 ,b0 [) + k∂t v ∗ kL1 (Ω0 ×]a0 ,b0 [) .
Tending ε to zero, we get

Zb0 Z Zb0 Z
∂t v ∗ (x, t)ϕ(x, t)dxdt = − v ∗ (x, t)∂t ϕ(x, t)dxdt (4.1.2)
a0 Ω0 a0 Ω0
for any ϕ ∈ C0∞ (Ω×]a, b[). So, ∂t v ∗ is Sobolev’s derivative as well.

Regarding an inverse statement, we argue as follows. Suppose that ∂t v ∗
is Sobolev’s derivative, i.e., it satisfies identity (4.1.2) with v ∗ and ∂t v ∗ from
L1,loc (Ω×]a, b[). If we assume in addition that C0∞ (Ω) is dense in V , then
∂t v ∗ is a derivative of v ∗ in the sense of Definition 1.1.
Proof of Lemma 1.2 We may extend ϕ by zero to the whole Rn × R
(Ω ⊂ Rn ). Take a huge cube Cl ×] − l, l[ so that Cl ×] − l, l[⊃ suppϕ. Here,
Cl = {x ∈ Rn : |xi | < l, i = 1, 2, ..., n}. Then we expand ϕ as the Fourier
series in spatial variable x
∞ X
x·m
X
ϕ(x, t) = cm (t)eiπ l ,
k=0 |m|=k
where Z
1 x·m
cm (t) = ϕ(x, t)e−iπ l dx.
(2l)n
Cl
The Fourier series converges very well. So, after taking real and imaginary
parts, given ε > 0, we find the number N (ε) such that
kϕ − ΦN (ε) kC 1 (C l ×[−l,l]) < ε,
PN (ε)
where ΦN (ε) (x, t) = k=1 ϕk (x)χk (t). Assume that there exist functions
∞ ∞
ϕ0 ∈ C0 (Ω), χ0 ∈ C0 (]a, b[) with the following property
ϕ0 (x)χ0 (t) = 1 (4.1.3)
if (x, t) ∈ suppϕ. We may let then
Φ
e N (ε) = ΦN (ε) ϕ0 χ0
and show
kϕ − Φ C (Ω×[a,b]) = k(ϕ − ΦN (ε) )ϕ0 χ0 kC 1 (Ω×[a,b]) ≤ c(Ω, a, b, l)ε.

e N (ε) k 1
To justify (4.1.3), let us introduce the following sets
(suppϕ)t = {x ∈ Ω : (x, t) ∈ supp ϕ },
Λ = {t ∈ [a, b] : (supp ϕ)t 6= ∅}.
Let t1 = inf t and t2 = sup t. We claim that a < t1 ≤ t2 < b. Assume that
t∈Λ t∈Λ
t2 = b. Then, by the definition, there exists a sequence (xk , tk ) ∈ supp ϕ
with tk → b as k → ∞. Selecting if necessary a subsequence, we have a
contradiction for the limit point (x, b) ∈ supp ϕ. Now, let us show that
K = ∪t1 ≤t≤t2 (supp ϕ)t
is a closed set of R3 . Assume that xk ∈ K and xk → x as k → ∞. For each

k, one can find tk ∈ [t1 , t2 ] such that (xk , tk ) ∈ supp ϕ. We may assume that
tk → t ∈ [t1 , t2 ] and then, by the definition of the support, (x, t) ∈ supp ϕ.
So, x ∈ K and thus K is closed. It remains to find an open set Ω1 b Ω such
that K ⊂ Ω1 . So, supp ϕ ⊂ Ω1 × [t1 , t2 ]. The rest of the proof is easy.
4.2. EXPLICIT SOLUTION 77
4.2 Explicit Solution

Consider a bounded domain Ω ⊂ Rn with smooth boundary and the following
initial-boundary value problem
∂t u − ∆u = f − ∇p and div u = 0 in QT = Ω×]0, T [,

u = 0 on ∂Ω × [0, T ], (4.2.1)
u(x, 0) = a(x) x ∈ Ω.
Assume that ◦
a ∈J (Ω). (4.2.2)
This problem can be written in the operator form
◦
e = f ∈ L2 (0, T ; (J 12 (Ω))0 ),
∂t u − ∆u
◦
u|t=0 = a ∈J (Ω), (4.2.3)
see notation for the Stokes operator ∆e and for the dual space in the last
section of Chapter 2.
Our task is to construct an explicit solution provided eigenvalues and
eigenfunctions of the Stokes operator ∆
e in the domain Ω are known. So, we
have
−∆ϕ
e k = λk ϕk in Ω,
ϕk = 0 on Ω, (4.2.4)
where k = 1, 2, ....
First, we expand functions f and a, using eigenfunctions and eigenvalues
of the Stokes operator,
∞
X
f (x, t) = fk (t)ϕk (x),
k=1
where
fk (t) = (f (·, t), ϕk (·))
and ∞
X
a(x) = ak ϕk (x), ak = (a, ϕk ).
k=1
By our assumptions,
ZT X
∞
2 1
kf k ◦ = |fk (t)|2 dt,
L2 (0,T ;(J 12 (Ω))0 )
k=1
λk
0
∞
X
kak22,Ω = a2k . (4.2.5)
k=1
We are looking for a solution to (4.2.3) of the form
∞
X
u(x, t) = ck (t)ϕk (x). (4.2.6)
k=1
Assume that
ck (0) = ak , k = 1, 2, .... (4.2.7)
Our further calculations are going to be formal. Later on, we will explain
in what sense the formal solution is a solution to problem (4.2.3). So, if we
plague (4.2.6) into (4.2.3), then the identity
∞
X ∞
X
c0k ϕk + λk ck ϕk = fk ϕk
k=1 k=1
comes out, which is satisfied if one lets
c0k (t) + λk ck (t) = fk (t),

ck (0) = ak , (4.2.8)
where k = 1, 2, .... System (4.2.8) has a unique solution
Zt
−λk t
ck (t) = e ak + eλk τ fk (τ )dτ . (4.2.9)
0
So, we have got a formal solution of form (4.2.6). Let us analyze its
properties, starting with estimates for coefficients given by (4.2.9). We have
Zt 2
2 −2λk t 2 −λk (t−τ )
ck (t) ≤ 2e ak + 2 e fk (τ )dτ

0
Zt Zt
≤ 2e−2λk t a2k + 2 e−2λk (t−τ ) dτ fk2 (τ )dτ
0 0
Zt Zt
1 1 −2λk t
≤ 2e−2λk t a2k + fk2 (τ )dτ − e fk2 (τ )dτ.
λk λk
0 0
So, finally,
Zt
1
c2k (t) ≤ 2e−2λk t a2k + fk2 (τ )dτ. (4.2.10)
λk
0
Summing up the above inequalities, we establish the estimate:
∞ ∞ ∞ Zt
X
−2λ1 t
X X 1
ku(·, t)k22 = c2k (t) ≤ 2e a2k + fk2 (τ )dτ
k=1 k=1 k=1
λk
0
−2λ1 t
≤ 2e kak22 + kf k 2
◦ (4.2.11)
L2 (0,T ;(J 12 (Ω))0 )
or
kuk2L∞ (0,T ;L2 (Ω)) ≤ 2kak22 + kf k2 ◦ . (4.2.12)

L2 (0,T ;(J 12 (Ω))0 )
To get the second estimate, we multiply the first equation in (4.2.8) by

ck (t) and apply Young’s inequality
c0k (t)ck (t) + λk c2k (t) = fk (t)ck (t)

1 fk2 (t) 1
≤ + λk c2k (t).
2 λk 2
So,
fk2 (t)
(c2k (t))0 + λk c2k (t) ≤ .
λk
Integration in t gives us:
ZT ZT
fk2 (t)
c2k (T ) + λk c2k (t)dt ≤ c2k (0) + dt
λk
0 0
ZT
fk2 (t)
= a2k + dt.
λk
0
Then, after summation, we arrive at the second estimate
ZT X
∞
k∇uk2L2 (0,T ;L2 (Ω)) = λk c2k (t)dt
0 k=1
≤ kak22 + kf k2 ◦ . (4.2.13)
L2 (0,T ;(J 12 (Ω))0 )
The estimates (4.2.12) and (4.2.13) are called energy estimates.

The final estimate will be derived from (4.2.8) in the following way
ZT X
∞
2 |c0k (t)|2
k∂t uk ◦ = dt
L2 (0,T ;(J 12 (Ω))0 )
k=1
λk
0
ZT X
∞ ZT X
∞
fk2 (t)
≤2 λk c2k (t)dt +2 dt
k=1
λk
0 0 k=1
≤ 2k∇uk2L2 (0,T ;L2 (Ω)) + 2kf k 2
◦ .
L2 (0,T ;(J 12 (Ω))0 )
So, applying (4.2.13), we find the third estimate
k∂t uk2 ◦ ≤ 2kak22 + 4kf k2 ◦ . (4.2.14)

L2 (0,T ;(J 12 (Ω))0 ) L2 (0,T ;(J 12 (Ω))0 )
Now, we wish to figure out in which sense (4.2.3) holds. Let us take an
◦
arbitrary function w ∈ L2 (0, T ; J 12 (Ω)) and expend it as a Fourier series
∞
X
w(x, t) = dk (t)ϕk (x).
k=1
Obviously,
ZT X
∞
k∇wk2L2 (0,T ;L2 (Ω)) = λk d2k (t)dt < ∞.
0 k=1
Hence,
ZT Z ZT X
∞
∂t u · wdxdt = c0k (t)dk (t)dt,
0 Ω 0 k=1
ZT Z ZT Z X
∞
∇u : ∇wdxdt = ck (t)dk (t)|∇ϕk |2 dxdt =
0 Ω 0 Ω k=1
ZT X
∞
= λk ck (t)dk (t)dt,
0 k=1
ZT Z ZT X
∞
f · wdxdt = fk (t)dk (t)dt,
0 Ω 0 k=1
and, by (4.2.8),
ZT Z
∂t u · w + ∇u : ∇w − f · w dxdt =
0 Ω
ZT X
∞
= c0k (t) + λk ck (t) − fk (t) dk (t)dt = 0.
0 k=1
◦
Taking w(x, t) = χ(t)v(x) with v ∈ J 12 (Ω) and χ ∈ C01 (0, T ), we get that, for
a.a. ∈]0, T [, the identity
Z Z
∂t u(x, t) · v(x) + ∇u(x, t) : ∇v(x) dx = f (x, t) · v(x)dx (4.2.15)
Ω Ω
◦
holds for all v ∈ J 12 (Ω). To be more precise, (4.2.15) is fulfilled at all
Lebesgue’s points of the following functions t 7→ ∂t u(·, t), t 7→ ∇u(·, t), and
t 7→ f (·, t). Identity (4.2.15) is called the weak form of the first equation in
(4.2.3).
It remains to establish in what sense the initial data in (4.2.3) are satisfied.
◦
Lemma 2.1. Function t 7→ u(·, t) ∈ (J 12 (Ω))0 can be modified on a zero-
◦
measure subset of [0, T ] so that, for each v ∈ J 12 (Ω), the function
Z
t 7→ u(·, t) · v(·)dx
Ω
is continuous on [0, T ].
◦
Proof Since u ∈ L2 (0, T ; (J 12 (Ω))0 ), a.a. points t0 ∈ [0, T ] are Lebesgue’s
points of t 7→ u(·, t) in the following sense
tZ0 +ε
1
ku(·, t) − u(·, t0 )k ◦ dt → 0
2ε (J 12 (Ω))0
t0 −ε
as ε → 0.
Denote by S the set of al Lebesgue’s points of t 7→ u(·, t). We know that
|S| = T . Let t0 < t1 be two points from S. By the definition of the derivative
∂t u,
ZT Z ZT Z
∂t u(x, t) · v(x)χ(t)dxdt = − u(x, t) · v(x)∂t χ(t)dxdt
0 Ω 0 Ω
◦
for any v ∈ J 12 (Ω) and for any χ ∈ C01 (0, T ). We can easily extend the latter
◦
identity to functions χ ∈ W 12 (0, T ). Take function χ = χε so that χε (t) = 0
if 0 < t ≤ t0 − ε or t1 + ε ≤ t < T , χε (t) = 1 if t0 + ε ≤ t ≤ t1 − ε,
χε (t) = (t − t0 + ε)/(2ε) if t0 − ε < t < t0 + ε, and χε (t) = (t1 + ε − t)/(2ε)
if t1 − ε < t < t1 + ε. Then, we have
ZT Z tZ1 +εZ
1
∂t u(x, t) · v(x)χε (t)dxdt = u(x, t) · v(x)dxdt
2ε
0 Ω t1 −ε Ω
tZ0 +εZ
1
− u(x, t) · v(x)dxdt. (4.2.16)
2ε
t0 −ε Ω
Obviously,
1 tZ0 +εZ
u(x, t) − u(x, t0 ) · v(x)dxdt

2ε

t0 −ε Ω
tZ0 +ε
1
≤ ku(·, t) − u(·, t0 )k ◦ dtkv(·)k ◦ 1 →0
2ε (J 12 (Ω))0 ) J 2 (Ω)
t0 −ε
as ε → 0. So, after taking the limit, we have

Z Z Zt1 Z
u(x, t1 ) · v(x)dx = u(x, t0 ) · v(x)dx + ∂t u(x, τ ) · v(x)dxdτ
Ω Ω t0 Ω
for a.a. t1 ∈ [0, T ]. Since the right-hand side of the latter identity is a
continuous function with respect to t1 , the left-hand side is continuous in t1
as well.
Now, coming back to our function u, we note that u ∈ L∞ (0, T ; L2 (Ω)).
Therefore, we can state that
Z
t 7→ u(x, t) · v(x)dx is continuous in t on [0, T ]
Ω
◦
for each function v ∈J (Ω) and even for each function v ∈ L2 (Ω). The
latter follows from the fact that for any v ∈ L2 (Ω) we have Helmoholtz-
Weyl decomposition in the Ladyzhenskaya form so that v = v1 + ∇p, where
◦
v1 ∈J (Ω) and p ∈ W21 (Ω). Moreover,
Z Z
u(x, t) · v(x)dx = u(x, t) · v1 (x)dx,
Ω Ω
◦
since u(·, t) ∈J (Ω).
So, our initial data are satisfied at least in the following sense. Since
u(·, 0) = a(·) by construction u, we have
Z Z
lim u(x, t) · v(x)dx = a(x) · v(x)dx
t→+0
Ω Ω
for any v ∈ L2 (Ω).

However, in our particular case, we can claim even more.
Theorem 2.2. Assume that

◦ ◦
u ∈ L2 (0, T ; J 12 (Ω)), ∂t u ∈ L2 (0, T ; (J 12 (Ω))0 ).
Then u ∈ C([0, T ]; L2 (Ω)) and
Zt Z
1 1
∂t u · udxdt = ku(·, t)k22,Ω − ku(·, t1 )k22,Ω (4.2.17)
2 2
t1 Ω
for all t, t1 ∈ [0, T ].

Proof So, we are given
ZT X
∞
kuk2 ◦ = λk d2k (t)dt,
L2 (0,T ;J 12 (Ω))
0 k=1
where ∞
X
u(x, t) = dk (t)ϕk (x),
k=1
and
ZT X
∞
2 1 0
k∂t uk ◦ = (d (t))2 dt.
L2 (0,T ;(J 12 (Ω))0 )
k=1
λk k
0
In a view of Lemma 2.1, it is sufficient to show that the function t 7→
ku(·, t)k2,Ω is continuous. We know that functions t 7→ dk (t) are continuous
on [0, T ]. Therefore, the function t 7→ gN (t) = N 2
P
k=1 dk (t) is continuous on
2
[0, T ] as well. We know also that gN (t) → ku(·, t)k2,Ω as N → ∞. So, we need
to show that the sequence gN (t) is uniformly bounded and the convergence
is uniform.
First, we show uniform boundedness. We have
Zt X
N
gN (t) − gN (t1 ) = 2 d0k (τ )dk (τ )dτ (4.2.18)
t1 k=1
for any t, t1 ∈ [0, T ] and thus

ZT X∞
1 0 12 ZT X∞ 21
2
gN (t) ≤ gN (t1 ) + 2 (dk (t)) dt λk d2k (t)dt
k=1
λk k=1
0 0
≤ gN (t1 ) + 2k∇ukL2 (0,T ;L2 (Ω)) k∂t uk ◦ .
L2 (0,T ;(J 12 (Ω))0 )
The latter inequality can be integrated with respect to t1
ZT
1
gN (t) ≤ gN (t1 )dt1 + 2k∇ukL2 (0,T ;L2 (Ω)) k∂t uk ◦
T L2 (0,T ;(J 12 (Ω))0 )
0
1
≤ k∇uk2L2 (0,T ;L2 (Ω)) + 2k∇ukL2 (0,T ;L2 (Ω)) k∂t uk ◦ .
T λ1 L2 (0,T ;(J 12 (Ω))0 )
So, uniform boundedness follows.

From (4.2.18), one can deduce uniform continuity. Indeed,
Zt 12
2
|gN (t) − gN (t1 )| ≤ 2 k∇u(·, t)k2,Ω dt k∂t uk ◦ .
L2 (0,T ;(J 12 (Ω))0 )
t1
Now, gN (t) converges to ku(·, t)k22,Ω uniformly, which means that the function
t 7→ ku(·, t)k22,Ω is continuous on [0, T ].
At last, (4.2.17) follows directly from (4.2.18) if N → +∞.
Actually, the analogue of Theorem 2.2 takes places. Moreover, an abstract
version of it is valid:
Theorem 2.3. Let H be a Hilbert space, V be a reflexive Banach space, and

V is continuously imbedded into H. Let V ∩ H contain a countable set S
which is dense in V and in H, i.e.,
V = [S]V , H = [S]H .
Let V ∗ be a dual space to V with respect to scalar product in H, i.e.,
kv ∗ kV ∗ = sup{(v ∗ , v)H : v ∈ V, kvkV = 1}.
Assume that v ∈ Lp (0, T ; V ) ∩ L2 (0, T ; H) and ∂t v ∈ Lp0 (0, T ; V ∗ ) with p0 =

p/(p − 1) and p > 1.
Then, v ∈ C([0, T ]; H) and
Zt
kv(·, t)k2H − kv(·, t1 )k2H =2 (∂t v(·, τ ), v(·, τ ))H dτ
t1
for any t, t1 ∈ [0, T ].

Proof We start with general facts. Let t 7→ v(·, t), where v ∈ Lp (0, T ; V ).
We extend v by zero outside [0, T ]. The first fact is the integral continuity:
for any ε > 0, there is a number δ(ε) > 0 such that
ZT
kv(·, t + h) − v(·, t)kpV dt < ε
0
whenever |h| < δ(ε). This property provides the following. Let
ZT
vε (·, t) = ωε (t − τ )v(·, τ )dτ,
0
where ωε is a standard mollifying kernel. We then have
ZT
kvε (·, t) − v(·, t)kpV dt ≤
0
Z∞
≤ kvε (·, t) − v(·, t)kpV dt =
−∞
Z∞ Z∞
= k ωε (t − τ )(v(·, τ ) − v(·, t))dτ kpV dt ≤
−∞ −∞
Z∞ Z∞
≤ ωε (t − τ )kv(·, τ ) − v(·, t))kpV dτ dt =
−∞ −∞
Z Z∞
∞
= ωε (t1 )kv(·, t1 + t) − v(·, t))kpV dτ dt ≤ γ

−∞ −∞
provided ε < 21 δ(γ). This means that vε → v in Lp (0, T ; V ).

We know that
∂t vε (·, t) = (∂t v)ε (·, t)
provided 0 < ε ≤ t ≤ T − ε. The proof is an exercise. So, we can claim
∂t vε → ∂t v in Lp,loc (0, T ; V ∗ ).
Further, we can use the same trick as in the case of star-shaped domains.
Without loss of generality, we may replace the interval ]0, T [ with ] − 1, 1[.
We take λ > 1 and define
t
v λ (·, t) = v(·, ), |t| ≤ λ
λ
and thus
1
∂t v λ (·, t) = ∂s v(·, s)|s= λt .
λ
Here, the crucial things are as follows:
kv λ − vkLp (−1,1;V ) + kv λ − vkL2 (−1,1;H) → 0
and
k∂t v λ − ∂t vkLp0 (−1,1;V ∗ ) → 0
as λ → 1. Moreover, for fixed λ > 1,
kv λ − (v λ )ε kLp (−1,1;V ) + kv λ − (v λ )ε kL2 (−1,1;H) → 0
and
k∂t v λ − ∂t (v λ )ε kLp0 (−1,1;V ∗ ) → 0
as ε → 0. Summarizing these two properties, we may construct a sequence
v (k) that is differentiable in t and satisfies:
kv (k) − vkLp (−1,1;V ) + kv (k) − vkL2 (−1,1;H) → 0
and
k∂t v (k) − ∂t vkLp0 (−1,1;V ∗ ) → 0
as k → ∞.
Now, let u = v (k) − v (m) , we have the identity
Z t
2
ku(·, t)kH = 2 (∂t u(·, τ ), u(·, τ )H dτ + ku(·, t1 )k2H , (4.2.19)
t1
which implies the bound

1
sup ku(·, t)k2H ≤ 2k∂t ukLp0 (−1,1;V ∗ ) kukLp (−1,1;V ) + kukL2 (−1,1;H) .
−1<t<1 2
In turn, the latter inequality yields that the v (k) is a Cauchy sequence in
C([0, T ]; H) and thus v (k) converges to v in C([0, T ]; H). The identity of
Theorem 2.3 can be derived from (4.2.19) with u = v (k) and k → ∞.
◦ ◦
Theorem 2.4. Assume that a ∈J (Ω) and f ∈ L2 (0, T ; (J 12 (Ω))0 ). There
exists a unique function u called weak solution to (4.2.1) such that:
◦ ◦
u ∈ L2 (0, T ; J 12 (Ω)), ∂t u ∈ L2 (0, T ; (J 12 (Ω))0 ); (4.2.20)
for a.a. t ∈ [0, T ],
Z h i Z
∂t u(x, t) · v(x) + ∇u(x, t) · ∇v(x) dx = f (x, t) · v(x)dx (4.2.21)
Ω Ω
◦
for any v ∈ J 12 (Ω);
u(·, 0) = a(·) (4.2.22)
and (4.2.22) is fulfilled in the L2 -sense, i.e., ku(·, t)−a(·)k2,Ω → 0 as t → +0.
Moreover,
u ∈ C([0, T ]; L2 (Ω)).
Proof Existence has been already proven. It remains to show unique-
ness. Assume that u1 is another solution satisfying (4.2.20)–(4.2.22). Then
for w = u − u1 we have
Z Z
∂t w(x, t) · v(x)dx + ∇w(x, t) : ∇v(x)dx = 0
Ω Ω
◦
for a.a. t ∈ [0, T ] and for any v ∈ J 12 (Ω) and thus
Z Z
∂t w(x, t) · w(x, t)dx + |∇w(x, t)|2 dx = 0.
Ω Ω
Integrating the latter identity with respect to t in [0, t0 ], we get, by Theorem

2.2,
kw(·, t0 )k22,Ω ≤ kw(·, 0)k22,Ω = 0
for any t0 ∈ [0, T ].
4.3 Cauchy Problem
∂t u − ∆u = f − ∇p and div u = 0 in QT = Rn ×]0, T [,

u(x, 0) = a(x) x ∈ Rn . (4.3.1)
4.4. PRESSURE FIELD. REGULARITY 89
Assume that ◦ ◦
a ∈J ≡J (Rn ). (4.3.2)
It is supposed also that
◦ ◦
f ∈ L2 (0, T ; (J 12 )0 ) := L2 (0, T ; (J 12 (Rn ))0 ), divf = 0. (4.3.3)
In this case, the Cauchy problem can be reduced to the Cauchy problem for
the heat equation
∂t u − ∆u = f in QT ,
u(x, 0) = a(x) x ∈ Rn . (4.3.4)
Indeed, assume that u is a solution to the Cauchy problem (4.3.4). Take the
divergence of equations in (4.3.4). Then we have
∂t divu − ∆divu = 0 in QT ,
divu(x, 0) = 0 x ∈ Rn .
By the unique solvability of the Cauchy problem for the heat equation, one
can claim that divu = 0 in QT . The pressure field is an arbitrary function of
t.
Solution to (4.3.4) can be given in an explicit form with the help of the
fundamental solution to the heat equation:
Z Zt Z
u(x, t) = Γ(x − y, t)a(x)dx + Γ(x − y, t − τ )f (y, τ )dydτ,
Rn 0 Rn
where
1 |x|2
− 4t
Γ(a, t) =
n e
(4πt) 2
n
for x ∈ R and t > 0. This formula is a good source for understanding
properties of solutions to (4.3.1).
4.4 Pressure Field. Regularity

Let us go back to initial-boundary value problem (4.2.1) and its functional
formulation (4.2.2)
◦
e = f ∈ L2 (0, T ; (J 1 (Ω))0 ),
∂t u − ∆u 2
◦
u|t=0 = a ∈J (Ω). (4.4.1)
Assuming that our domain Ω is bounded, we have constructed a weak solution

to (4.4.1) with the help of eigenfunctions of the Stokes operator, i.e.,
∞
X
u(x, t) = ck (t)ϕk (x).
k=1
For unknown coefficient ck (t), we have the following system of equations
c0k (t) + λk ck (t) = fk (t),

ck (0) = ak , (4.4.2)
where k = 1, 2, ... and where

Z
fk (t) = f (x, t) · ϕk (x)dx = (f (·, t), ϕk (·)), ak = (a, ϕk ).
Ω
Now, we are going to assume additionally that

◦ ◦
a ∈ J 12 (Ω), f ∈ L2 (0, T ; J (Ω)). (4.4.3)
Then,
ZT X
∞
2
kf k ◦ = kf k22,QT = fk2 (t)dt < ∞,
L2 (0,T ;J (Ω))
0 k=1
∞
X
k∇ak22,Ω = λk a2k < ∞. (4.4.4)
k=1
Next, let us multiply the first equation in (4.4.2) by ck , sum up the result
from 1 to N , integrate the sum in time over the interval ]0, t[, and find
N Zt X
N
1X
λk c2k (t) + |c0k (τ )|2 dτ
2 k=1 k=1
0
N Zt N
1X X
= λk a2k + fk (τ )c0k (τ )dτ,
2 k=1 k=1
0
which yields the bound
N
X Zt X
N N
X Zt X
N
λk c2k (t) + |c0k (τ )|2 dτ ≤ λk a2k + fk2 (τ )dτ
k=1 0 k=1 k=1 0 k=1
≤ k∇ak22,Ω + kf k22,QT
Passing to the limit as N → ∞, we derive the following important estimate
Zt
k∇u(·, t)k22,Ω + k∂t u(·, τ )k22,Ω dτ ≤ k∇ak22,Ω + kf k22,QT (4.4.5)
0
being valid for any t ∈ [0, T ].

Now, our aim is to recover the pressure field. To this end, we proceed as
follows. Consider the linear functional
Z
lt (v) = (∇u(x, t) : ∇v(x) + ∂t u(x, t) · v(x) − f (x, t) · v(x))dx
Ω
◦
for any L12 (Ω). For it, we have the estimate
|lt (v)| ≤ k∇u(·, t)k2,Ω k∇vk2,Ω + (k∂t u(·, t)k2,Ω + kf (·, t)k2,Ω )kvk2,Ω .
According to Poincare’s inequality, kvk2,Ω ≤ c(Ω)k∇vk2,Ω . So, the functional

v 7→ lt (v) is bounded for a.a. t ∈ [0, T ] and a bound of its norm is:
klt k ≤ c(Ω)(k∇u(·, t)k2,Ω + k∂t u(·, t)k2,Ω + kf (·, t)k2,Ω ) (4.4.6)

◦
Moreover, lt (v) = 0 for any v ∈ J 12 (Ω). For bounded domains with Lipschitz
boundary, there exists a function t 7→ p(·, t) ∈ L2 (Ω), see Chapter I, such
that Z
lt (v) = p(x, t)divvdx, kp(·, t)k2,Ω ≤ klt k.
Ω
It follows from (4.4.6) that

h i
kpk2,QT ≤ c k∇uk2,QT + k∂t uk2,QT + kf k2,QT . (4.4.7)
So, we have
Z Z
∇u(x, t) : ∇v(x)dx − p(x, t)divv(x)dx
Ω
Z ZΩ
= f (x, t) · v(x)dx − ∂t u(x, t) · v(x)dx (4.4.8)
Ω Ω
◦
for any v ∈ L12 (Ω) and for a.a. t ∈ [0, T ].
For those t, i.e., for which (4.4.8) holds, we may apply the regularity
theory developed for the linear stationary Stokes system. More precisely,
one can estimate higher derivatives in spatial variables:
k∇2 u(·, t)k2,Ω + k∇p(·, t)k2,Ω ≤ c(kf (·, t)k2,Ω + k∂t u(·, t)k2,Ω
and thus
k∇2 uk2,QT + k∇pk2,QT ≤ c(kf k2,QT + k∂t uk2,QT ).
Combining the latter estimate with (4.4.5), we get the final bound:
k∂t uk2,QT + k∇2 uk2,QT + k∇pk2,QT ≤ c(kf k2,QT + k∇ak2,Ω ). (4.4.9)
Summarizing mentioned above, one can formulate the following result.
Theorem 4.1. Assume that the boundary of a bounded domain Ω is smooth

and conditions (4.4.3) holds. Then,
u ∈ W22,1 (QT ), p ∈ W21,0 (QT ),
with estimate (4.4.9). In addition, ∇u ∈ C([0, T ]; L2 (Ω)) and equations
∂t u − ∇u = f − ∇p, divu = 0
are satisfied a.e. in QT .
Here, we have used the following notion:

2,1
Ws,l (QT ) = {v ∈ Ll (0, T ; Ws2 (QT ), ∂t v ∈ Ll (0, T ; Ls (Ω)},
1,0
Ws,l (QT ) = {v ∈ Ll (0, T ; Ws1 (QT )},
and
Ws2,1 (QT ) = Ws,s
2,1
(QT ), Ws1,0 (QT ) = Ws,s
1,0
(QT ).
Proof of Theorem 4.1 We need to show ∇u ∈ C([0, T ]; L2 (Ω)). To
◦
this end, we are going to use Theorem 2.3. Introducing H = J 12 (Ω) with
◦
scalar product (u, v)H = (∇u, ∇v), V = J 12 (Ω) ∩ W22 (Ω) with the norm
◦
e 2,Ω , let us verify that V ∗ = J (Ω) is dual to V with respect H.
kvkV = k∆vk
Indeed, let l ∈ V ∗ . So, we have |l(v)| ≤ ck∆vk
e 2,Ω for any v ∈ V . Since
◦ ◦
e ) = J (Ω), for any p ∈ J (Ω), one can define G(p) = l(v), where p = −∆v.
∆(V e
Obviously, |G(p)| ≤ kpk2,Ω klk. By Riesz theorem, there exists a unique
◦
v ∗ ∈ J (Ω) such that Z
G(p) = v ∗ · pdx,
Ω
∗
kv k2,Ω = klk and
Z Z
∗
l(v) = − v · ∆(v)dx
e = ∇v ∗ : ∇vdx.
Ω Ω
◦
Now, any v ∗ from J (Ω) defines a linear functional on V by formula
Z Z
− v · ∆(v)dx = ∇v ∗ : ∇vdx ≤ kv ∗ k2,Ω k∆(v)k
∗ e e ∗
2,Ω ≤ kv k2,Ω kvkV .
Ω Ω
◦
It is easy to prove that in fact klk = kv ∗ k2,Ω . So, V ∗ ' J (Ω), i.e., spaces are
isometrically isomorphic.
By Theorem 2.3, ∇u ∈ C([0, T ]; L2 (Ω)). Theorem 4.1 is proved.
Theorem 4.2. Let Ω be a bounded domain with smooth boundary. Consider
the following initial boundary value problem
∂t u − ∆u = f − ∇p and div u = 0 in QT = Ω×]0, T [,

Z
1
p(x, t)dx ≡ [p(·, t)]Ω = 0, t ∈ [0, T ], (4.4.10)
|Ω|
Ω
u|∂ 0 QT = 0,
where ∂ 0 QT is the parabolic boundary of QT .

Let f ∈ Ls,l (QT ) := Ll (0, T ; Ls (Ω)) for some finite numbers s > 1 and
2,1
l > 1. Then problem (4.4.10) has a unique solution such that u ∈ Ws,l (QT )
1,0
and p ∈ Ws,l (QT ), satisfying the following coercive estimate
kukW 2,1 (QT ) + kpkW 1,0 (QT ) ≤ c(Ω, s, l, n)kf kLs,l (QT ) .
s,l s,l
4.5 Uniqueness Results

◦
Lemma 5.1. Let v ∈ L1 (0, T ; J m (Ω)) with m > 1 and Ω be a bounded
domain in Rn with sufficiently smooth boundary. If 1 < m < 2, assume in
addition that n = 2 or 3. Assume, further, that
Z
v · (∂t w + ∆w)dxdt = 0
QT
for w(x, t) = χ(t)W (x) with any function χ ∈ C 1 ([0, T ]) and any divergence
free field W ∈ C 2 (Ω) subject to the end condition χ(T ) = 0 and to the
boundary condition W = 0 on ∂Ω, respectively.
Then v is identically zero in QT . Here, QT = Ω×]0, T [.
Proof Take as a test function w = χ(t)ϕk (x), where ϕk is the k-th

eigenfunction of the Stokes operator. ϕk is a smooth function since Ω is
a domain with smooth boundary. This follows from imbedding theorems,
regularity theory for the linear Stokes problem, and bootstrap arguments.
χ(t) is a smooth function, satisfying the end condition χ(T ) = 0. Then, we
have Z
v(x, t) · (χ0 (t)ϕk (x) − χ(t)λk ϕk (x))dxdt = 0
QT
and thus
ZT
vk (t)(χ0 (t) − χ(t)λk ) = 0 (4.5.1)
0
for any χ ∈ C 2 ([0, T ]) with χ(T ) = 0, where

Z
vk (t) = v(x, t) · ϕk (x)dx.
Ω
4.5. UNIQUENESS RESULTS 95
From (4.5.1), it follows that
vk0 (t) + λk vk (t) = 0

vk (0) = 0.
The latter immediately implies that vk (t) = 0 for t ∈ [0, T ].

Now, we wish to show that v(x, t) = 0 for any x ∈ Ω and t ∈ [0, T ].
Let us start with the simplest case m ≥ 2. Obviously, for bounded
◦ ◦ ◦
domains, J m (Ω) ⊆ J 2 (Ω). Hence, v(·, t) ∈ J 2 (Ω) and
∞
X
kv(·, t)k22,Ω = vk2 (t) = 0.
k=1
Let us consider now the case, in which
1 < m < 2. (4.5.2)
First, we shall show Z

v(x, t) · u(x)dx = 0 (4.5.3)
Ω
∞
for and u ∈ C0,0 (Ω) and for a.a. t ∈ [0, T ]. To this end, fix an arbitrary test
∞
function u ∈ C0,0 (Ω) and let
N
X
SN = ck ϕk ,
k=1
where Z
ck = u · ϕk dx.
Ω
We know that SN → u in L2 (Ω) as N → ∞. But it is not sufficient to justify
(4.5.3) by taking the limit below
Z Z
0 = v(x, t) · SN (x)dx → v(x, t) · u(x)dx, N → ∞,
Ω Ω
for a.a. t ∈ [0, T ]. However, assuming additionally that n = 2 or n = 3, we

will be able to show that sequence SN is bounded in Lm0 (Ω) and this will
imply (4.5.3) in the case 1 < m < 2 (Exercise).
Let us consider first n = 2. Then, by embedding theorems, we have

Z 10 Z 21
m0 m 2
|SN | dx ≤ c(m, Ω) |∇SN | dx
Ω Ω
N
X 1
2 2
≤ c(m, Ω) λk ck ≤ c(m, Ω)k∇uk2,Ω .
k=1
So, required boundedness follows.

In the case n = 3, we have
Z 10 Z 21
m0 m 2 2 2
|SN | dx ≤ c(m, Ω) (|∇SN | + |∇ SN | )dx .
Ω Ω
If we let −∆S
e N = fN ∈ L2 (Ω), then simply, by definition of the Stokes
operator, the partial sum S N solves the following boundary value problem
−∆SN + ∇pN = fN , divSN = 0 in Ω

SN |∂Ω = 0.
Now, we are again in a position to apply the regularity theory, developed for
the stationary Stokes system, that gives the estimate
k∇2 SN k2,Ω + k∇pN k2,Ω ≤ c(Ω)kfN k2,Ω .
So, we have
N
X
k∇2 SN k2,Ω ≤ c(Ω)k∆S
e N k2,Ω ≤ c(Ω) λ2k c2k
k=1
e 2,Ω ≤ c(Ω)k∆uk2,Ω ≤ c(Ω)k∇2 uk2,Ω .
≤ c(Ω)k∆uk
And thus boundedness of kSN km0 ,Ω has been proven in the case n = 3 as
well.
Now, the aim is to show that v is identically zero in QT . Fix t ∈ [0, T ]
and consider a linear functional
Z
◦
l(w) = v · wdx, w ∈ L1m0 (Ω).
Ω
4.5. UNIQUENESS RESULTS 97
◦
By Poincaré inequality, it is bounded in L1m0 (Ω) and, by (4.5.3), vanishes on
◦
J 1m0 (Ω), i.e.,
◦
l(w) = 0 ∀w ∈ J 1m0 (Ω).
As we know, any functional, possessing the above properties, can be presented
in the form Z
◦
l(w) = p div wdx, ∀w ∈ L1m0 (Ω)
Ω
for some p ∈ Lm (Ω). The latter means that v = −∇p.
Our next step is to show that p is a solution to the classical Neumann
problem: ∆p = 0 in Ω and ∂p/∂ν = 0 on ∂Ω, where ν is the unit outward
normal to the surface ∂Ω, in the following sense
Z
∇p · ∇qdx = 0 ∀q ∈ Wm1 0 (Ω). (4.5.4)
Ω
◦
∞
Indeed, ∇p = −v ∈ J m (Ω). Therefore, there exists a sequence w(k) ∈ C0,0 (Ω)
(k)
such that w → ∇p in Lm (Ω). So,
Z Z
(k)
w · ∇qdx = 0 → ∇p · ∇qdx = 0
Ω Ω
for any q ∈ Wm1 0 (Ω).

Now, our problem has been reduced to the following uniqueness question:
Let p ∈ Wm1 (Ω), 1 < m < 2, and
Z
∇p · ∇qdx = 0 ∀q ∈ Wm1 0 (Ω).
Ω
Then p must be a constant in Ω.

Assume that f is a smooth function on ∂Ω and satisfies compatibility
condition Z
f (s)ds = 0.
∂Ω
Consider the classical Neumann problem: ∆q = 0 in Ω and ∂q/∂ν = f
on ∂Ω. There exists a smooth solution to this problem. For it, we have
Z Z Z
∂q
0 = ∇p · ∇qdx = p ds = pf ds.
∂ν
Ω ∂Ω ∂Ω
Since f is a smooth function satisfying the compatibility condition only, we

can claim that
p = c1
on ∂Ω for some constant c1 .
We let further p1 = p−c1 and p1 is a solution to the homogeneous Dirichlet
boundary problem: ∆p1 = 0 in Ω and p1 = 0 on ∂Ω. To show that p1 is in
fact identically zero, we find for any q ∈ Wm2 0 (Ω) with q = 0 on ∂Ω
Z Z
∇p1 · ∇qdx = − p1 ∆qdx.
Ω Ω
We may select a function q in a special way so that ∆q = |p1 |m−1 sign p1 =

f ∈ Lm0 (Ω) with q = 0 on ∂Ω. It is well known that such a function exists
and belongs to Wm2 0 (Ω). Hence,
Z Z
0 = ∇p1 · ∇qdx = − |p1 |m dx
Ω Ω
and thus p is a constant in Ω and v = 0 in QT . Lemma 5.1 is proved.

We have another uniqueness result.
◦
Theorem 5.2. Let v ∈ L1 (0, T ; J 1m (Ω)) with m > 1 and Ω be a bounded
domain with sufficiently smooth boundary. Assume
Z
(v · ∂t w − ∇v : ∇w)dz = 0 (4.5.5)
QT
for any w(x, t) = χ(t)W (x), where χ ∈ C 1 ([0, T ]) such that χ(T ) = 0 and
∞
W ∈ C0,0 (Ω).
Then v is identically zero in QT .
Proof By density arguments, (4.5.5), of course, holds for any W ∈
◦
J 1m0 (Ω).
Take any function W ∈ C 2 (Ω) with W = 0 on Ω and div W = 0 in Ω.
◦
We know that W ∈ J 1m0 (Ω), see Chapter I, Theorem 4.3. So, v satisfies all
assumptions of Lemma 5.1 and therefore v ≡ 0 in QT .
The above proof works well under additional assumption on n if 1 <
m < 2. However, we can give an alternative proof that does not need extra
4.6. LOCAL INTERIOR REGULARITY 99
assumptions on the spatial dimension. We assume that (4.5.3) has been

already proved. Then we can take test function in (4.5.3) in the following way
W = ∇ ∧ w with arbitrary function w ∈ C0∞ (Ω). This implies ∇ ∧ v(·, t) = 0
in Ω. Taking into account the fact that v is divergence free, we deduce that
◦
v(·, t) is a harmonic function in Ω belonging to J 1m (Ω). The rest of the proof
is more or less the same as the final part of the proof of Lemma 5.1, see
arguments providing p1 = 0 there. Theorem 5.2 is proved.
4.6 Local Interior Regularity

In this section, we shall restrict ourselves to the 3D case simply to reduce
a number of parameters. Although it is clear that the extension to other
dimensions is straightforward.
The problem of local interior regularity can be formulated as follows.
Consider the Stokes system in a canonical domain, say, in Q = B×] − 1, 0[
∂t u − ∆u = f − ∇p, div u = 0. (4.6.1)
We always assume that functions u and p have some starting differentiability

properties. Keeping in mind the 3D non-stationary non-linear problem, we
supposed that
1,0
u ∈ Wm,n (Q), p ∈ Lm,n (Q) (4.6.2)
for some finite m and n being greater than 1.
Assuming that some additional information about the right-hand side f
is given, we shall try to make some conclusions about smoothness of u and
p in smaller parabolic balls Q(r) = B(r)×] − r2 , 0[.
It is known that, for stationary Stokes system as well as for heat equation,
solutions are smooth locally as long as f is smooth. However, in the case of
non-stationary Stokes system, we have smoothing is spatial variables but not
in time. This can be seen easily from the following example, in which f = 0
and
u(x, t) = c(t)∇h(x), p(x, t) = −c0 (t)h(x).
Here, h is a harmonic function in B and c is a given function, defined on
[0, T ]. This solution is infinitely differentiable inside B but, under assump-
tions (4.6.2), it is just Hölder continuous in time. There is no smoothing in
time despite the smoothness of f .
In general, we have the following statement.
Proposition 6.1. Assume that u and p satisfy (4.6.1), conditions (4.6.2),

and let
f ∈ Ls,n (Q) (4.6.3)
with s ≥ m.
2,1 1,0
Then u ∈ Ws,n (Q(1/2)) and p ∈ Ws,n (Q(1/2)) and we have the estimate
k∂t uks,n,Q(1/2) + k∇2 uks,n,Q(1/2) + k∇pks,n,Q(1/2)
≤ c(kf ks,n,Q + kukm,n,Q + k∇ukm,n,Q + kpkm,n,Q ). (4.6.4)
Proof It is sufficient to prove this proposition for case s = m. General

case can be deduced from it by embedding theorems and bootstrap argu-
ments.
Fix a non-negative cut-off function ϕ ∈ C0∞ (B×] − 1, 1[) so that ϕ = 1
in B(1/2)×] − (1/2)2 , (1/2)2 [. For any t ∈] − 1, 0[, we determine a function
w(·, t) as a unique solution to the boundary value problem
∆w(·, t) − ∇q(·, t) = 0, div w(·, t) = v(·, t) · ∇ϕ(·, t)
in B and Z
q(x, t)dx = 0, w(·, t) = 0
B
on ∂B. It satisfies the estimate
k∇2 w(·, t)ks,B + kq(·, t)ks,B + k∇q(·, t)ks,B ≤
≤ ck∇(v(·, t) · ∇ϕ(·, t))ks,B . (4.6.5)

Letting
V = ϕv − w, P = ϕp − q,
F = ϕf + v∂t ϕ − 2∇v∇ϕ − v∆ϕ + p∇ϕ − ∂t w,
we observe that new functions V and P are a unique solution to the following
initial boundary value problem
∂t V − ∆V = F − ∇P, div V = 0
in Q,
V =0
4.6. LOCAL INTERIOR REGULARITY 101
on ∂ 0 Q. Taking into account the statement of Theorem 4.2 and estimate

there, we find
k∂t vks,n,Q(1/2) + k∇2 vks,n,Q(1/2) + k∇pks,n,Q(1/2) ≤ cA + ck∂t wks,n,Q , (4.6.6)
where
A = kf ks,n,Q + kvks,n,Q + k∇vks,n,Q + kpks,n,Q .
So, our task is to evaluate the last term on the right hand side of (4.6.6).
The key point here is duality arguments proposed by V. A. Solonnikov.
Introducing new notation u = ∂t w and r = ∂t q, we can derive from the
equations for w and q
∆u(·, t) − ∇r(·, t) = 0
div u(·, t) = ∂t v(·, t) · ∇ϕ(·, t) + v(·, t) · ∇ ∂t ϕ(·, t) (4.6.7)
in B, Z
r(x, t)dx = 0, u(·, t) = 0 (4.6.8)
B
on ∂B.
Given g ∈ Ls0 (B) with s0 = s/(s − 1), let us define a function u
e as a
unique solution to the boundary value problem
u − ∇e
∆e r = g, div u
e=0 (4.6.9)
in B, Z
re(x)dx = 0, u
e=0 (4.6.10)
B
on ∂B. Function re obeys the estimate
ke
rks0 ,B + k∇e
rks0 ,B ≤ ckgks0 ,B . (4.6.11)
Now, from (4.6.7)–(4.6.11), we find

Z Z
u(x, t) · g(x)dx = u(x, t) · (∆e
u(x) − ∇e
r(x))dx
B B
Z Z
= re(x)div u(x, t)dx = re(x)(∂t v(x, t) · ∇ϕ(x, t) + v(x, t) · ∇ ∂t ϕ(x, t))dx.
B B
Expressing ∂t v via the Navier-Stokes equations, we derive from the previous

identity
Z Z
u(x, t) · g(x)dx = re(x)(∆ v(x, t) − ∇ p(x, t) + f (x, t)) · ∇ϕ(x, t)dx
B B
Z
+ re(x)v(x, t) · ∇ ∂t ϕ(x, t)dx.
B
It remains to integrate by parts and use estimate (4.6.11). As a result, we
have
Z
u(x, t) · g(x)dx ≤ ckgks0 ,B (kv(·, t)ks,B + k∇ v(·, t)ks,B + kp(·, t)ks,B )
B
and thus
k∂t wks,n,Q ≤ cA.
Proposition 6.1 is proved.
Keeping in mind the 3D non-stationary non-linear problem, one cannot
expect that the number n is big. In such cases, the following embedding
result can be useful.
2,1
Proposition 6.2. Assume that v ∈ Ws,n (Q) with
2 3
1 < n ≤ 2, µ=2− − > 0.
n s
Then
1
|v(z) − v(z 0 )| ≤ c(m, n, s)(|x − x0 | + |t − t0 | 2 )µ (kvks,n,Q
+k∇vks,n,Q + k∇2 vks,n,Q + k∂t vks,n,Q )
for all z = (x, t) ∈ Q(1/2) and for all z 0 = (x0 , t0 ) ∈ Q(1/2). In other words,
v is Hölder continuous with exponent µ relative to parabolic metric in the
closure of Q(1/2).
Finally, using bootstrap arguments, we can prove the following statement
which in a good accordance with the above example.
Proposition 6.3. Assume that conditions (4.6.2) hold with 1 < n < 2 and
f = 0. Let u and p be an arbitrary solution to system (4.6.1). Then for any
0 < τ < 1 and for any k = 0, 1, ..., the function (x, t) 7→ ∇k u(x, t) is Hölder
continuous with any exponent less than 2 − 2/n in the closure of the set Q(τ )
relative to the parabolic metric.
4.7. LOCAL BOUNDARY REGULARITY 103
4.7 Local Boundary Regularity

To describe the results of this section, we are going to exploit the following
notation:
x = (x0 , x3 ), x0 = (x1 , x2 ),
Q+ (r) = C+ (r)×] − r2 , 0[⊂ R3 × R, C+ (r) = b(r)×]0, r[∈ R3 ,
b(r) = {x0 ∈ R2 : |x0 | < r }.
The complete analogue of Proposition 6.1 is as follows.
Proposition 7.1. Assume that we are given three functions

1,0
u ∈ Wm,n (Q+ (2)), p ∈ Lm,n (Q+ (2)), f ∈ Lm1 ,n (Q+ (2))
with m1 ≥ m satisfying the system
∂t u − ∆ u = f − ∇ p, div v = 0 in Q+ (2),
and the homogeneous Dirichlet boundary condition
u(x0 , 0, t) = 0.
Then u ∈ Wm2,11 ,n (Q+ (1)) and p ∈ Wm1,01 ,n (Q+ (1)) with the estimate
k∂t ukLm1 ,n (Q+ (1)) + k∇2 ukLm1 ,n (Q+ (1)) + k∇pkLm1 ,n (Q+ (1)) ≤
≤ c(kukLm,n (Q+ (2)) + k∇ukLm,n (Q+ (2)) + kpkLm,n (Q+ (2)) + kf kLm1 ,n (Q+ (2)) ).
If we assume f = 0 and 1 < n < 2, then, by embedding theorem similar

to Proposition 4.6.2, u is Hölder continuous in the closure of the space-time
cylinder Q+ (1). Hölder continuity is defined with respect to the parabolic
metrics and the corresponding exponent does not exceed 2−2/n. However, in
general, the analogue of Proposition 6.3 is not true in the boundary regularity
theory, i.e., in general there is no further smoothing even in spatial variables.
Let us describe the corresponding counter-example.
We are looking for non-trivial bounded solutions to the following homoge-
neous initial boundary value problem
∂t v − ∆ v = −∇ q, div v = 0 (4.7.1)
in R3+ ×] − 4, 0[ under the homogeneous Dirichlet boundary condition
v(x0 , 0, t) = 0 x0 ∈ R2 , −4 < t < 0, (4.7.2)
and under homogeneous initial data
v(x, −4) = 0 x ∈ R3+ . (4.7.3)
Here R+3
= {x = (x0 , x3 ) : x3 > 0 }.
Taking an arbitrary function f (t), we seek a non-trivial solution to (4.7.1)
–(4.7.3) in the form of shear flow, say, along x1 -axis:
v(x, t) = (w(x3 , t), 0, 0), q(x, t) = −f (t)x1 .
Here, a scalar function u solves the following initial boundary value problem
∂t w(y, t) − wyy (y, t) = f (t), (4.7.4)
w(0, t) = 0, (4.7.5)
w(y, −4) = 0, (4.7.6)
where 0 < y < +∞ and −4 < t < 0 and wyy = ∂ 2 w/∂y 2 .
It is not so difficult to solve (4.7.4)–(4.7.6) explicitly:
√ y
Zt Z4(τ +4)
2 2
w(y, t) = √ f (t − τ − 4)dτ e−ξ dξ. (4.7.7)
π
−4 0
Keeping in mind that our aim is to construct irregular but summable solution,
we choose the function f as follows
1
f (t) = , 0 < α < 1/2. (4.7.8)
|t|1−α
Then, direct calculations give us:

(i) w is a bounded smooth function in the strip ]0, +∞[×] − 4, 0[ satisfying
boundary and initial conditions;
1
(ii) wy (y, t) ≥ c(α) y1−2α for y and t subject to the inequalities y 2 ≥ −4t,
0 < y ≤ 3, and −9/8 ≤ t < 0.
4.8. COMMENTS 105
(iii) Let s, s1 , l, and l1 be numbers greater than 1 and satisfy the condition
n1 1 1o 1
K = max 1− ,1 − <α< . (4.7.9)
2 s l1 2
Then
1,0
v ∈ Ws,l (C+ (3)×] − 9/4, 0[), q ∈ Ls1 ,l1 (C+ (3)×] − 9/4, 0[).
Assume we are given numbers 1 < m < +∞ and 1 < n < 2. Letting
s = s1 = m and l = l1 = n and choosing α so that inequality (4.7.9)
holds. The functions v and q constructed above for the chosen α meet all
the conditions of Proposition 7.1 with f = 0. However, ∇v is unbounded
in any neighborhood of the space-time point z = (x, t) = 0. This is a
counter-example of Seregin-Sverak, which is an essential simplification of the
counter-example given by K. Kang.
4.8 Comments
Chapter 4 contains standard material about existence, uniqueness, and regu-
larity of solutions to the non-stationary Stokes system. A bit new results for
introductory course are in the last three sections. In particular fine unique-
ness theorems and local regularity issues are discussed in Sections 5–7. They
are needed for the local regularity analysis in Chapter 6.
Chapter 5
Non-linear Non-Stationary
Problem
5.1 Compactness Results for non-Stationary

Problems
Our standing assumptions are as follows. We are given a triple of Banach
spaces V0 , V , and V1 , having the following properties
(i) V0 ⊂ V ⊂ V1 , V0 is a reflexive space;
(ii) imbedding V0 ⊂ V is compact;
(iii) imbedding V ⊂ V1 is continuous;
(iv) v ∈ V0 and kvkV1 = 0 imply kvkV = 0.
Lemma 1.1. Given η > 0, there exists C(η) > 0 such that
kvkV ≤ ηkvkV0 + C(η)kvkV1 (5.1.1)
for any v ∈ V0 .
Proof Usual compactness arguments work. Assume that the statement
is wrong. Then for any n ∈ N there exists vn ∈ V0 such that
kvn kV > ηkvn kV0 + nkvn kV1 .
Then after normalization, we have
kvn0 kV = 1 > ηkvn0 kV0 + nkvn0 kV1 ,
107
108 CHAPTER 5. NON-LINEAR NON-STATIONARY PROBLEM
where vn0 = vn /kvn kV . The sequence vn0 is bounded in a reflexive space, and
thus without loss of generality we may assume that
vn0 * v0
in V0 and thus
vn0 → v0
in V and V1 . Since nkvn0 kV1 is bounded and therefore kvn0 kV1 → 0 = kv0 kV1 .
Hence, by assumption (iv), kv0 kV = 0. However, 1 = kvn kV → kv0 kV . This
leads to contradiction. Lemma 1.1 is proved.
Proposition 1.2. (Aubin-Lions lemma) Let 1 < p0 , p1 < ∞, V1 is reflexive,
and define
n o
W ≡ kvkW = kvkLp0 (0,T ;V0 ) + k∂t vkLp1 (0,T ;V1 ) < ∞ .
Then W is compactly imbedded into Lp0 (0, T ; V ).

Proof Suppose that sequence u(j) is bounded in W . Then, without loss
of generality, we may assume that
u(j) * u
in Lp0 (0, T ; V0 ) and

∂t u(j) * ∂t u
in Lp1 (0, T ; V1 ). Setting v (j) = u(j) − u, we need to show that
v (j) → 0
in Lp0 (0, T ; V ). By Lemma 1.1, we have for arbitrary number η > 0
kv (j) (·, t)kV ≤ ηkv (j) (·, t)kV0 + C(η)kv (j) (·, t)kV1
and thus
kv (j) kLp0 (0,T ;V ) ≤ ηkv (j) kLp0 (0,T ;V0 ) + C(η)kv (j) kLp0 (0,T ;V1 )
≤ cη + C(η)kv (j) kLp0 (0,T ;V1 ) .

So, it is enough to show
v (j) → 0 (5.1.2)
5.1. COMPACTNESS RESULTS 109
in Lp0 (0, T ; V1 ). To this end, we are going first to prove that
sup sup kv (j) (·, t)kV1 < ∞. (5.1.3)

j 0<t<T
Indeed, if (5.1.3) would be true, then (5.1.2) would follow from
kv (j) (·, t)kV1 → 0 (5.1.4)
for a.a. t ∈ [0, T ] and Lebesgue’s theorem. So, our goal is to prove (5.1.3)
and (5.1.4).
To prove (5.1.3), we exploit the following formula (it is a simple conse-
quence of the definition of ∂t v)
Zt
(j)
v (·, t) = ∂t v(·, τ )dτ + v (j) (·, s) (5.1.5)
s
for any 0 ≤ s, t ≤ T , which implies

Zt
kv (j) (·, t)kV1 ≤ kv (j) (·, s)kV1 + k∂t v(·, τ )kV1 dτ
s
1
(j) 0
≤ kv (·, s)kV1 + T p1 k∂t vkLp1 (0,T ;V1 ) .
The latter inequality can be integrated in s. As a result, we get (5.1.3).

Now, we wish to expalain validity of (5.1.4). To this end, let us integrate
(5.1.5) in s over the interval ]t, s1 [
Zs1 Zt Zs1
(j)
(s1 − t)v (·, t) = ds ∂t v(·, τ )dτ + v (j) (·, s)ds.
t s t
After integration by parts in s in the first term of the right hand side, we
find
v (j) (·, t) = a(j) (·, t) + b(j) (·, t),
where
Zs1
1
a(j) (·, t) = v (j) (·, s)ds
s1 − t
t
and
Zs1
1
b(j) (·, t) = (s1 − s)∂t v (j) (·, s)ds.
s1 − t
t
Now, take any ε > 0 and fix it. Then

Zs1 10
1 0 p
kb(j) (·, t)kV1 ≤ |s1 − s|p1 ds 1
|s1 − t|
t
ZT p1 1
(j) p1 0
× k∂t v (·, s)kV1 ds 1 ≤ c|s1 − t| p1 < ε
0
for any j and for s1 sufficiently closed to t.

Next, we wish to show that for each fixed s1 (for given t)
ka(j) (·, t)kV1 → 0. (5.1.6)
To this end, we first notice that
a(j) (·, t) * 0
in V1 . Then if we would show boundedness in V0 , (5.1.6) would follows from

compactness of imbedding V0 into V .
We have
Zs1
(j) 1
ka (·, t)kV0 ≤ kv (j) (·, s)kV0 ds
|s1 − t|
t
1 1
0
1
0 −1
≤ |s1 − t| p0 kv (j) kLp0 (0,T ;V0 ) ≤ c|s1 − t| p0 .
|s1 − t|
So, given s1 , (5.1.6) is true and we may find N (s1 , t) such that
ka(j) (·, t)kV1 ≤ ε
for any j ≥ N1 (s1 , t). This proves (5.1.4) and completes the proof of Propo-
sition 1.2.
5.2. AUXILIARY PROBLEM 111
5.2 Auxiliary Problem

Assume that ◦
a ∈ J (Ω) (5.2.1)
and ◦
f ∈ L2 (0, T ; (J 12 (Ω))0 ) (5.2.2)
Proposition 2.1. Let QT = Ω×]0, T [ and
w ∈ L∞ (QT ), div w = 0 in QT . (5.2.3)
There exists a unique solution v to the initial boundary value problem
∂t v − ∆v + div v ⊗ w + ∇q = f, div v = 0 in QT ,
v|∂Ω×[0,T ] = 0, (5.2.4)
v|t=0 = a
in the following sense:

◦ ◦
v ∈ C([0, T ]; L2 (Ω)) ∩ L2 (0, T ; J 12 (Ω)), ∂t v ∈ L2 (0, T ; (J 12 (Ω))0 );
for a.a. t ∈ [0, T ]

Z
(∂t v(x, t) · ve(x) + ∇v(x, t) : ∇e
v (x))dx
Ω
Z
= (v(x, t) ⊗ w(x, t) : ∇e
v (x) + f (x, t) · ve(x))dx (5.2.5)
Ω
◦
for all ve ∈ J 12 (Ω);
kv(·, t) − a(·)k2,Ω → 0 (5.2.6)
as t → +0.
Proof We are going to apply the Leray-Schauder principle, see Theorem

1.3 of Chapter III. We let
◦
X = L2 (0, T ; J (Ω)).
Given u ∈ X, define v = A(u) as a solution to the following problem:

◦ ◦
v ∈ C([0, T ]; L2 (Ω)) ∩ L2 (0, T ; J 12 (Ω)), ∂t v ∈ L2 (0, T ; (J 12 (Ω))0 ); (5.2.7)
for a.a. t ∈ [0, T ]

Z
(∂t v(x, t) · ve(x) + ∇v(x, t) : ∇e
v (x))dx
Ω
Z
= fe(x, t) · ve(x)dx (5.2.8)
Ω
◦
kv(·, t) − a(·)k2,Ω → 0 (5.2.9)
as t → +0. Here, fe = f − u ⊗ w. Such a function v exists and is unique (for
given u) according to Theorem 2.3 of Chapter 4 since
◦
fe ∈ L2 (0, T ; (J 12 (Ω))0 ).
So, operator A is well defined. Let us check that it satisfies all the assump-
tions of Theorem 1.3 of Chapter III.
Continuity: Let v 1 = A(u1 ) and v 2 = A(u2 ). Then
Z Z
1 2 1 2
v )dx = (u1 − u2 ) ⊗ w : ∇e
(∂t (v − v ) · ve + ∇(v − v ) : ∇e v dx
Ω Ω
and letting ve = v 1 − v 2 , we find

1
∂t kv 1 − v 2 k22,Ω + k∇v 1 − ∇v 2 k22,Ω ≤ c(w)ku1 − u2 k2,Ω k∇v 1 − ∇v 2 k2,Ω
2
and thus
sup kv 1 − v 2 k2,Ω ≤ c(w)ku1 − u2 k2,QT
0<t<T
The latter implies continuity.

Compactness: As in the previous case, we use the energy estimate
sup kvk22,Ω + k∇vk22,Ω ≤ c(w)kuk22,Ω + ckf k2 ◦

0<t<T L2 (0,T ;(J 12 (Ω))0 )
The second estimate comes from (5.2.8)
k∂t vk2 ◦ ≤ k∇vk2QT + c(w)kuk22,Ω + ckf k2 ◦ .

L2 (0,T ;(J 12 (Ω))0 ) L2 (0,T ;(J 12 (Ω))0 )
Combining the latter estimates, we see that sets which are bounded in X
remain to be bounded in
◦ ◦
W = {w ∈ L2 (0, T ; J 12 (Ω)), ∂t w ∈ L2 (0, T ; (J 12 (Ω))0 )}.
◦ ◦ ◦
By Proposition 1.2 for V0 = J 12 (Ω), V = J (Ω), and V1 = (J 12 (Ω))0 , such a set
is precompact.
Now, we wish to verify the second condition in Theorem 1.3 of Chapter
III. For v = λA(v), after integration by parts, we find that, for a.a. t ∈ [0, T ],
Z Z
(∂t v · ve + ∇v : ∇e
v )dx = λ (f · ve − (w · ∇v) · ve)dx
Ω Ω
◦
for any ve ∈ J 12 (Ω). If we plague ve(·) = v(·, t) into the latter relation, then
the identity Z
(w · ∇v) · vdx = 0,
Ω
ensures following estimate:

Z Z
1
|v| dx + |∇v|2 dx ≤ kf k ◦ 1 0 k∇vk2,Ω
2
2 J 2 (Ω))
Ω Ω
and thus
Z
kvk2QT ≤ T sup |v(x, t)|2 dx ≤ cT (kf k2 ◦ + kak22,Ω ) = R2 .
0<t<T L2 (0,T ;(J 12 (Ω))0 )
Ω
Now, all the statements of Proposition 2.1 follow from the Leray-Schauder
principle. Proposition 2.1 is proved.
Let ω% be the usual mollifying kernel and let
Z
(v)% (x, t) = ω% (x − x0 )v(x0 , t)dx0 .
Ω
◦
It is easy to check if t 7→ v(·, t) ∈ J (Ω) then div(v)% (·, t) = 0 (Exercise).
Now, we wish to show that there exists at least one function v % such that:
◦ ◦
v % ∈ C([0, T ]; L2 (Ω)) ∩ L2 (0, T ; J 12 (Ω)), ∂t v % ∈ L2 (0, T ; (J 12 (Ω))0 ); (5.2.10)
for a.a. t ∈ [0, T ]

Z
(∂t v % (x, t) · ve(x) + ∇v % (x, t) : ∇e
v (x))dx
Ω
Z
= (v % (x, t) ⊗ (v % )% (x, t) : ∇e
v (x) + f (x, t) · ve(x))dx (5.2.11)
Ω
◦
kv % (·, t) − a(·)k2,Ω → 0 (5.2.12)
as t → +0.
We note that (5.2.10)-(5.2.12) can be regarded as a weak form of the
following initial boundary value problem
∂t v % − ∆v % + (v % )% · ∇v % + ∇q % = f, div v ρ = 0 in QT ,
v % |∂Ω×[0,T ] = 0, (5.2.13)
v % |t=0 = a.
Proposition 2.2. There exists at least one function v % satisfying (5.2.10)-

(5.2.12). In addition, it satisfies the energy estimate
|v % |22,QT ≡ sup kv % (·, t)k22,Ω + k∇v % k22,QT

0<t<T
≤ c(kf k2 ◦ + kak22,Ω ) (5.2.14)
L2 (0,T ;(J 12 (Ω))0 )
with constant independent of %.
Proof We drop upper index %. The idea is the same as in Proposition

2.1: to use the Leray-Schauder principle. The space X is the same as in
Proposition 2.1. But the operator A will be defined in the different way:
given u ∈ X, we are looking for v = A(u) so that
◦ ◦
v ∈ C([0, T ]; L2 (Ω)) ∩ L2 (0, T ; J 12 (Ω)), ∂t v ∈ L2 (0, T ; (J 12 (Ω))0 ); (5.2.15)
for a.a. t ∈ [0, T ]

Z
(∂t v(x, t) · ve(x) + ∇v(x, t) : ∇e
v (x))dx
Ω
Z
= (v(x, t) ⊗ (u)% (x, t) : ∇e
v (x) + f (x, t) · ve(x))dx (5.2.16)
Ω
◦
kv(·, t) − a(·)k2,Ω → 0 (5.2.17)
as t → +0. By Proposition 2.1, such a function exists and is unique. We
need to check that all the assumptions of Theorem 1.3 of Chapter III hold
for our operator A.
Continuity: Do the same as in Proposition 2.1:
1
∂t kv 2 − v 1 k22,Ω + k∇(v 2 − v 1 )k22,Ω
Z 2

= v 2 ⊗ (u2 )% − v 1 ⊗ (u1 )% : ∇(v 2 − v 1 )dx
Ω
Z
= (v 2 − v 1 ) ⊗ (u2 )% : ∇(v 2 − v 1 )dx
Ω
Z
+ v 1 ⊗ (u2 − u1 )% : ∇(v 2 − v 1 )dx.
Ω
The first integral in the right hand side of the latter is zero and the second
one I can be bounded as follows
I ≤ sup |(u2 − u1 )% (x, t)|kv 1 k2,Ω k∇(v 2 − v 1 )k2,Ω .

x∈Ω
So, by Hölder inequality, we have
∂t kv 2 − v 1 k22,Ω + k∇(v 2 − v 1 )k22,Ω ≤ c(%)kv 1 k22,Ω ku2 − u1 k22,Ω
and thus
|v 2 − v 1 |22,QT ≤ c(%) sup kv 1 (·, t)k22,Ω ku2 − u1 k22,QT .

0<t<T
The latter gives us continuity.

Compactness: Now, we wish to write down the energy estimate for v
Z Z
1
∂t kvk22,Ω + k∇vk22,Ω = f · vdx − v ⊗ (u)% : ∇vdx
2
Ω Ω
Z
= f · vdx. (5.2.18)
Ω
As in the proof of the previous proposition, (5.2.18) implies the required

energy estimate
|v|22,QT ≤ c(kf k2 ◦ + kak22,Ω ) = cS, (5.2.19)

L2 (0,T ;(J 12 (Ω))0 )
where a constant c is independent of %.

Now, we need to evaluate the derivative in time. We have
Z
2 2 2
k∂t vk ◦ 1 0 ≤ ck∇vk2,Ω + ckf k ◦ 1 0 + c |v|2 |(u)% |2 dx
(J 2 (Ω)) (J 2 (Ω))
Ω
Z 2
≤ ck∇vk22,Ω + ckf k2◦ 1 + c(%) |u|dx kvk22,Ω .
(J 2 (Ω))0
Ω
After integration in time, the following estimate comes out:

Z
2 2
k∂t vk ◦
1 0
≤ c(%) S + S |u| dz . (5.2.20)
L2 (0,T ;(J 2 (Ω)) )
QT
Making use of the same arguments as in the proof of Proposition 2.1, we

conclude that for each fixed % > 0 the operator A is compact.
Let us check the second condition of Theorem 1.3. As in the proof of
(5.2.18) and (5.2.19), we can get
|v|22,QT ≤ c(λ2 kf k2 ◦ + kak22,Ω ) < 2cS = R2 .

L2 (0,T ;(J 12 (Ω))0 )
So, the existence is established.

The energy estimate can be proved along the lines of the proof of (5.2.19).
5.3. WEAK LERAY-HOPF SOLUTIONS 117
5.3 Weak Leray-Hopf Solutions

Now, we consider full non-stationary Navier-Stokes system in a bounded
domain Ω ⊂ Rn (n=2,3)
∂t v − ∆v + div v ⊗ v + ∇q = f, div v = 0 in QT ,
v|∂Ω×[0,T ] = 0, (5.3.1)
v|t=0 = a
We always assume that
f ∈ L2 (0, T ; V 0 ) (5.3.2)
and
a ∈ H, (5.3.3)
◦ ◦
where V = J 12 (Ω) and H = J (Ω).
Definition 3.1. A function v is called a weak Leray-Hopf solution to initial

boundary value problem (5.3.1)-(5.3.3) if it has the following properties:
(i) v ∈ L∞ (0, T ; H)
R ∩ L2 (0, T ; V );
(ii) function t 7→ v(x, t) · w(x)dx is continuous on [0, T ] for each w ∈
Ω
L2 (Ω); R
(iii) (−v · ∂t w − v ⊗ v : ∇w + ∇v : ∇w − f · w)dz = 0 for any test
Ω
∞
functionRC0,0 (QT ) = {v ∈ C0∞ (QT ) : div w = 0 in QT };
(iv) |v(x, t) − a(x)|2 dx → 0 as t → +0;
Ω
Rt R Rt R
1
|∇v|2 dxdt0 ≤ 1
f · vdxdt0
R R
(v) 2
|v(x, t)|2 dx + 2
|a(x)|2 dx +
Ω 0 Ω Ω 0 Ω
for all t ∈ [0, T ].
Theorem 3.2. Under assumptions (5.3.2) and (5.3.3), there exists at least
one weak Leray-Hopf solution to (5.3.1).
Proof By Proposition 2.2, for any positive %, there exists a function v %

such that
v % ∈ C([0, T ]; H) ∩ L2 (0, T ; V ), ∂t v % ∈ L2 (0, T ; V 0 ); (5.3.4)

for a.a. t ∈ [0, T ]

Z
(∂t v % (x, t) · ve(x) + ∇v % (x, t) : ∇e
v (x))dx
Ω
Z
= (v % (x, t) ⊗ (v % )% (x, t) : ∇e
v (x) + f (x, t) · ve(x))dx (5.3.5)
Ω
for all ve ∈ V ;
kv % (·, t) − a(·)k2,Ω → 0 (5.3.6)
%
as t → +0. Moreover, v has uniformly bounded energy
|v % |22,QT ≡ sup kv % (·, t)k22,Ω + k∇v % k22,QT ≤ A, (5.3.7)
0<t<T
where a constant A depends only on T , kf kL2 (0,T ;V 0 ) , and kak2,Ω .

Now, let us see what happens if % → 0. To apply Proposition 1.2 from
Section 1 on compactness, we need to estimate the derivative of v in t. To
this end, we are going to use the following imbedding theorem
◦
3 1
J 2 (Ω) ⊂ C (Ω),
which is true provided n = 2, 3. Then from (5.3.5) it follows
Z Z
%
v k∞,Ω |v % ||(v % )% |dx + k∇e
∂t v · vedx ≤ k∇e v k2,Ω k∇v % k2,Ω
Ω Ω
+kf kV 0 k∇e
v k2,Ω
◦
for any v ∈ J 32 (Ω) and thus

k∂t v % k ◦ 3 0 ≤ c(Ω) kv % k2,Ω k(v % )% k2,Ω + k∇v % k2,Ω + kf kV 0
(J 2 (Ω))

% % %
≤ c(Ω) |v |2,QT kv k2,Ω + k∇v k2,Ω + kf kV 0 .
Therefore

k∂t v % k ◦ ≤ c(Ω) |v % |2,QT kv % k2,QT + |v % |2,QT + kf kL2 (0,T ;V 0 ) .
L2 (0,T ;(J 32 (Ω))0 )
1
Since kv % k2,QT ≤ T 2 |v % |2,QT , we get
k∂t v % k ◦ ≤ A1 , (5.3.8)
L2 (0,T ;(J 32 (Ω))0 )
where a constant A1 depends only on T , kf kL2 (0,T ;V 0 ) , and kak2,Ω .

Now, we can apply Proposition 1.2 with the following choice of subspaces
◦
V ⊂ H ⊂ (J 32 (Ω))0 ,
◦ ◦
where the space (J 32 (Ω))0 is the space dual to J 32 (Ω) relative to H, and state
that after selecting a subsequence
∗
v % *v in L∞ (0, T ; H), (5.3.9)
v% * v in L2 (0, T ; V ), (5.3.10)
v% → v in L2 (0, T ; H). (5.3.11)
Now, let us show
Z
D% = |v % ⊗ (v % )% − v ⊗ v|dz → 0
QT
as % → 0. Indeed, from (5.3.11), it follows that

Z Z
D% ≤ |(v − v) ⊗ (v )% |dz + |v ⊗ ((v % )% − v)|dz
% %
QT QT
Z
% %
≤ kv − vk2,QT k(v )% k2,QT + |v ⊗ (v % − v)% |dz
QT
Z
1
+ |v ⊗ ((v)% − v)|dz ≤ kv % − vk2,QT T 2 |v % |2,QT
QT
1 1
+T |v|2,QT kv % − vk2,QT + T 2 |v|2,QT k(v)% − vk2,QT → 0
2
as % → 0 by (5.3.11) and (5.3.7).

∞
Setting w(x)
e = w(x, t) in (5.3.5), with w ∈ C0,0 (QT ), integrating in t by
parts in (5.3.5), and passing to the limit, we deduce that v satisfies (iii) of
Definition 3.1. So, (i) and (iii) of Definition 3.1 have been verified.
◦
Now, let us take any function ve ∈ J 32 (Ω) and consider functions
Z
%
t 7→ fve (t) = v % (x, t) · ve(x)dx.
Ω
Now, our goal is to show that for each fixed ve, the set of functions fve% is
precompact in C([0, T ]). Indeed, it is uniformly bounded since
sup |fve% (t)| ≤ |v % |2,QT ke
v k2,Ω ≤ c|v % |2,QT ke
vk ◦3 ≤ cAke
vk ◦3 .
0<t<T J 2 (Ω) J 2 (Ω)
Its equicontinuity follows from (5.3.8):

t+∆t
Z Z
|fve% (t + ∆t) − fve% (t)| = ∂t v % (x, τ ) · ve(x)dxdτ

t Ω
t+∆t
Z
≤ k∂t v % (·, τ )k ◦ ke
v (·)k ◦ 3 dτ
(J 32 (Ω))0 J 2 (Ω)
t
p p
≤ |∆t|k∂t v % k ◦ ke
vk ◦3 ≤c |∆t|A1 ke
vk ◦3 .
L2 (0,T ;(J 32 (Ω))0 ) J 2 (Ω) J 2 (Ω)
◦
Now, let ve(k) be a countable set that is dense in J 32 (Ω). Applying thedi-
agonal Cantor procedure, we can select a subsequence such that
Z Z
v (x, t) · ve (x)dx → v(x, t) · ve(k) (x)dx
% (k)
Ω Ω
in C([0, T ]). By boundedness of

sup sup kv % (·, t)k2,Ω ,
%>0 0<t<T
one can show (by density arguments)

Z Z
%
v (x, t) · ve(x)dx → v(x, t) · ve(x)dx
Ω Ω
◦ ◦
in C([0, T ]) for any v ∈ J 32 (Ω) and then for any v ∈ J (Ω).
Now, a given w ∈ L2 (Ω) can be decomposed as w = u + ∇p, where
◦
u ∈ J (Ω). Then
Z Z
%
v (x, t) · w(x)dx = v % (x, t) · u(x)dx
Ω Ω
Z Z
→ v(x, t) · u(x)dx = v(x, t) · w(x)dx (5.3.12)
Ω Ω
in C([0, T ]) as % → 0. So, (ii) of Definition 3.1 has been proved as well.

Now, we would like to justify (v) of Definition 3.1. To achieve this goal,
let us pick up ve(x) as v % (x, t) in (5.3.5) and integrate the equality in t. Since
Z Z Z
% % % 1
% % %
v ⊗ (v )% : ∇v dx = vi (vj )% vi,j dx = (vj% )% |v % |2,j = 0,
2
Ω Ω Ω
we have
Z Zt Z Z
1 % 2 % 2 1 0
|v (x, t)| dx + |∇v | dxdt = |a(x)|2 dx
2 2
Ω 0 Ω Ω
Zt Z
+ f · v % dxdt0 (5.3.13)
0 Ω
for all t ∈ [0, T ] and for all % > 0.

By (5.3.12),
Z Z
lim inf |v (x, t)| dx ≥ |v(x, t)|2 dx
% 2
(5.3.14)
%→0
Ω Ω
for any t ∈ [0, T ] and by (5.3.10)

Zt Z Zt Z
lim inf |∇v % |2 dxdt0 ≥ |∇v|2 dxdt0 (5.3.15)
%→0
0 Ω 0 Ω
and
Zt Z Zt Z
lim inf f · v % dxdt0 = f · vdxdt0 (5.3.16)
%→0
0 Ω 0 Ω
for all t ∈ [0, T ]. So, (v) of Definition 3.1 follows from (5.3.13)-(5.3.16).
It remains to prove validity of (iv) of Definition 3.1. To this end, notice
that by (5.3.12)
a(·) = v % (·, 0) * v(·, 0)
in L2 (Ω). So, v(·, 0) = a(·). Moreover, according to (ii) of Definition 3.1
v(·, t) * v(·, 0) = a(·)

in L2 (Ω) as t → +0. So,
lim inf kv(·, t)k2,Ω ≥ kak2,Ω .

t→+0
However, from the energy inequality it follows that
lim sup kv(·, t)k2,Ω ≤ kak2,Ω .

t→+0
The latter implies

lim kv(·, t)k2,Ω = kak2,Ω
t→+0
which together with week convergence gives (iv) of Definition 3.1. Theorem
3.2 is proved.
5.4 Multiplicative Inequalities and Related

Questions
Case 1: n=2
Lemma 4.1. (Ladyzhenskaya’s inequality)
kuk44,Ω ≤ 2kuk22,Ω k∇uk22,Ω
for any u ∈ C0∞ (Ω).
Proof. Oviously, it is enough to prove this inequality for Ω = R2 . We

have
Zx1
|u(x1 , x2 )|2 = 2 u(t, x2 )u,1 (t, x2 )dt
−∞
Z∞ 21 Z∞ 21
2 2
≤2 |u(t, x2 )| dt |u,1 (t, x2 )| dt .
−∞ −∞
5.4. MULTIPLICATIVE INEQUALITIES 123
And then
Z∞ Z∞ Z∞ Z∞
4
|u(x1 , x2 )| dx1 dx2 = |u(x1 , x2 )|2 |u(x1 , x2 )|2 dx1 dx2
−∞ −∞ −∞ −∞
Z∞ n Z∞ 21 Z∞ 21
2
≤4 dx1 dx2 |u(t, x2 )| dt |u,1 (t, x2 )|2 dt
−∞ −∞ −∞
Z∞ 21 Z∞ 12 o
× |u(x1 , s)|2 ds |u,2 (x1 , s)|2 ds
−∞ −∞
Z∞ Z∞ 21 Z∞ 21
2 2
= |u(t, x2 )| dt |u,1 (t, x2 )| dt dx2
−∞ −∞ −∞
Z∞ Z∞ 12 Z∞ 12
× |u(x1 , s)|2 ds |u,2 (x1 , s)|2 ds dx1
−∞ −∞ −∞
≤ 4kuk2 ku,1 k2 kuk2 ku,2 k2 ≤ 2kuk22 (ku,1 k22 + ku,2 k22 )

= 2kuk22 k∇uk22 . (5.4.1)
Corollary 4.2. Let u ∈ L∞ (0, T ; H) ∩ L2 (0, T ; V ). Then

1
kuk4,QT ≤ 2 4 |u|2,QT .
Proof We have
ku(·, t)k44,Ω ≤ 2ku(·, t)k22,Ω k∇u(·, t)k22,Ω ≤ 2|u|22,QT k∇u(·, t)k22,Ω .
After integration in t, we get the required inequality.

Case 2: n=3
3(s−2)
Lemma 4.3. Let 2 ≤ s ≤ 6 and α = 2s
. Then, for any u ∈ C0∞ (Ω),
1−α
kuks,Ω ≤ c(s)kuk2,Ω k∇ukα2,Ω .
Proof The Galiardo-Nirenberg inequality in dimension 3 reads
kuk6,Ω ≤ ck∇uk2,Ω , u ∈ C0∞ (Ω),

with a constant c independent of Ω. It remains to use interpolation in Ls
kuks,Ω ≤ kuk1−α α
2,Ω kuk6,Ω
3(s−2)
with α = 2s
.
kuks,l,QT ≤ c(s)|u|2,QT
for 2 ≤ s ≤ 6 and l satisfying

3 2 3
+ = .
s l 2
Here, kuks,l,QT = kukLs (0,T ;Ll (Ω)) .
Proof (Exercise) By Lemma 4.3,
ku(·, t)ks,Ω ≤ c(s)|u|1−α α

2,QT k∇u(·, t)k2,Ω
and
ZT 1l ZT 1l
1−α
ku(·, t)kls,Ω dt ≤ c(s)|u|2,Q T
k∇u(·, t)k αl
2,Ω dt .
0 0
If αl = 2, then the required inequality follows and

3 2 3 3 3(2 − s) 3
+ = +α= + = .
s l s s 2s 2
ku · ∇uks,l,QT ≤ c(s)|u|22,QT
with s and l greater than one and subject to the identity

3 2
+ = 4.
s l
Proof (Exercise) By Hölder inequality,
Z Z 2s Z ss
s 2 s1
|u · ∇u| dx ≤ |∇u| dx |u| dx 1
Ω Ω Ω
5.5. UNIQUENESS 125
2s
with s1 = 2−s
and thus after integration in t and application of Hölder
inequality
ZT Z s 2l 2−l
s1 1 (2−l) 2l
ku · ∇uks,l,QT ≤ k∇uk2,QT |u| dx dt .
0 Ω
It is easy to verify
3 2 3 2l
+ = , l1 = . (5.4.2)
s1 l1 2 l−2
The required inequality follows from Corollary 4.4.
Let us discuss some consequences of (5.4.2):
3 2 3 2−l 3(2 − s) 2 3
+ = 2s + = + −1= ,
s1 l1 2−s
l 2s l 2
which implies
3 2
+ = 4.
s l
5.5 Uniqueness of Weak Leray-Hopf

Solutions. 2D Case
Theorem 5.1. (O. Ladyzhenskaya) Let n = 2. Then, under assumptions
(5.3.2), (5.3.3), a weak Leray-Hopf solution to initial boundary value problem
(5.3.1) is unique.
Proof We let fe = −div v ⊗ v + f . By Corollary 4.2, fe belongs to

L2 (0, T ; V 0 ) (since v ∈ L4 (QT )). By Theorem 2.3 of Section 4, we know that
there exists a unique function u having the following properties:
u ∈ C([0, T ]; H) ∩ L2 (0, T ; V ), ∂t u ∈ L2 (0, T ; V 0 ); (5.5.1)
for a.a. t ∈ [0, T ],

Z h i
∂t u(x, t) · w(x) + ∇u(x, t) : ∇w(x) − fe(x, t) · w(x) dx = 0 (5.5.2)
Ω
for all w ∈ V ;
u(x, 0) = a(x), x ∈ Ω. (5.5.3)
Recalling Definition 3.1, part (ii), we get from (5.5.2) that v = v − u satisfies
the identity
Z h i
− v(x, t) · W (x)∂t χ(t) + ∇v(x, t) : ∇W (x)χ(t) dz = 0 (5.5.4)
QT
∞
for all W ∈ C0,0 (Ω) and for any χ ∈ C01 (0, T ). Our aim is to get rid of the
assumption that χ vanishes in a neighborhood of 0. To this end, we are going
to use the following fact:
kv(·, t)k2,Ω → 0 (5.5.5)
as t → +0. Take any function χ ∈ C 1 ([0, T ]) so that χ(T ) = 0 and a function
ϕε having the properties: ϕε (t) = 1 if t ≥ ε, ϕε (t) = 0 if 0 < t ≤ ε/2, and
ϕε (t) = (2t − ε)/ε if ε/2 < t < ε. Then, by (5.5.4), with ϕε χ as χ,
Z h i
ϕε (t) − v(x, t) · W (x)∂t χ(t) + ∇v(x, t) : ∇W (x)χ(t) dz
QT
Zε Z
2
= v(x, t) · W (x)χ(t)dz = Iε .
ε
ε Ω
2
For the right hand side, we have

p
|Iε | ≤ sup |W (x)| sup χ(τ ) |Ω| sup kv(·, t)k2,Ω → 0
x∈Ω τ ∈[0,T ] 0<t≤ε
as ε → 0.
So,
Z h i
− v(x, t) · W (x)∂t χ(t) + ∇v(x, t) : ∇W (x)χ(t) dz = 0
QT
∞
for all W ∈ C0,0 (Ω) and any χ ∈ C 1 ([0, T ]) with χ(T ) = 0. This means that
v ≡ 0 in QT and any weak Leray-Hopf solution v has the following properties:
v ∈ C([0, T ]; H) ∩ L2 (0, T ; V ), ∂t v ∈ L2 (0, T ; V 0 ); (5.5.6)
for a.a. t ∈ [0, T ],
Z h i
∂t v(x, t) · w(x) + ∇v(x, t) : ∇w(x) dx
Ω
Z h i
= v(x, t) ⊗ v(x, t) : ∇w(x) + f (x, t) · w(x) dx
e (5.5.7)
Ω
5.5. UNIQUENESS 127
for all w ∈ V ;
v(x, 0) = a(x), x ∈ Ω. (5.5.8)
Now, assume that we have two different solutions v 1 and v 2 . Letting
u = v 2 − v 1 , one deduce from (5.5.7) that:
Z h i
∂t u(x, t) · w(x) + ∇u(x, t) : ∇w(x) dx =
Ω
Z
= (v 2 (x, t) ⊗ v 2 (x, t) − v 1 (x, t) ⊗ v 1 (x, t)) : ∇w(x)dx
Ω
for any w ∈ V . Taking w(x) = u(x, t) in the above identity,

1
∂t ku(·, t)k22,Ω + k∇u(·, t)k22,Ω =
Z 2
= (u(x, t) ⊗ v 2 (x, t) : ∇u(x, t) + v 1 (x, t) ⊗ u(x, t) : ∇u(x, t))dx ≤
Ω
≤ 0 + k∇u(·, t)k2,Ω ku(·, t)k4,Ω kv 1 (·, t)k4,Ω .
By using Ladyzhenskaya’s inequality twice,
kv 1 (·, t)k44,Ω ≤ 2kv 1 (·, t)k22,Ω k∇v 1 (·, t)k22,Ω ≤ 2|v 1 |22,QT k∇v 1 (·, t)k22,Ω ≤
≤ c(a, f )k∇v 1 (·, t)k22,Ω
and
ku(·, t)k44,Ω ≤ 2ku(·, t)k22,Ω k∇u(·, t)k22,Ω .
And thus
1
∂t ku(·, t)k22,Ω + k∇u(·, t)k22,Ω ≤
2
3 1 1
≤ c(a, f )k∇u(·, t)k2,Ω
2
ku(·, t)k2,Ω
2
k∇v 1 (·, t)k2,Ω
2
.
Applying Young’s inequality, we find
∂t ku(·, t)k22,Ω + k∇u(·, t)k22,Ω ≤ c0 ku(·, t)k22,Ω k∇v 1 (·, t)k22,Ω
and thus
∂t ku(·, t)k22,Ω ≤ c0 y(t)ku(·, t)k22,Ω ,
where y(t) := k∇v 1 (·, t)k22,Ω . From this differential inequality, it is not diffi-
cult to derive
−c0 Rt y(τ )dτ
∂t e 0 ku(·, t)k22,Ω ≤ 0
and
Rt
−c0 y(τ )dτ
e 0 ku(·, t)k22,Ω ≤ ku(·, 0)k22,Ω = 0.
Therefore, ku(·, t)k22,Ω ≡ 0.
Let us discuss further regularity of 2D weak Leray-Hopf solutions.
Theorem 5.2. Assume that a ∈ V , and f ∈ L2 (QT ). Let v be a unique
solution to initial boundary value problem (5.3.1). Then
v ∈ W22,1 (QT ), ∇v ∈ C([0, T ]; L2 (Ω)).
Moreover, there exists q ∈ W21,0 (QT ) such that
∂t v + v · ∇v − ∆v = f − ∇q, div v = 0
a.e. in QT . It is supposed that Ω is a bounded domain with smooth boundary.

Proof Let us go back to problem (5.3.4)-(5.3.6), where a function v %
defined by the following relations:
v % ∈ C([0, T ]; H) ∩ L2 (0, T ; V ), ∂t v % ∈ L2 (0, T ; V 0 ); (5.5.9)
for a.a. t ∈ [0, T ]

Z
(∂t v % (x, t) · ve(x) + ∇v % (x, t) : ∇e
v (x))dx
Ω
Z
= fe(x, t) · ve(x))dx (5.5.10)
Ω
for all ve ∈ V , where
fe = f − (v % )% · ∇v % ∈ L2 (QT );
kv % (·, t) − a(·)k2,Ω → 0 (5.5.11)

as t → +0.
5.5. UNIQUENESS 129
According to Theorem 4.1 of Chapter 4, there exists a functions u% ∈

W22,1 (QT ) with ∇u% ∈ C([0, T ]; L2 (Ω)) and p% ∈ W21,0 (QT ) such that
∂t u% − ∆u% = fe − ∇p% , div u% = 0
a.e. in QT and u% (·, 0) = a(·). Using the same arguments as in the proof
of the previous statement, we can claim that v % = u% . Multiplying then the
equation
∂t v % + (vk% )% v,k
%
− ∆v % = f % − ∇ p%
e % and integrating each term in the product with respect to x, we find
by ∆v
Z Z
% e % e % |2 dx,
∆v · ∆v dx = |∆v
Ω Ω
Z
∇ p% · ∆v
e % dx = 0,
Ω
Z Z
% % 1
∂t v · ∆v
e dx = ∂t v % · ∆v % dx = − ∂t k∇ v % k22,Ω
2
Ω Ω
e % ∈ H and ∂t v % ∈ H for each fixed t. So, we derive the inequality

since ∆v
1
∂t k∇v % k22,Ω + k∆v
e % k22,Ω ≤ kf k2,Ω k∆v e % k2,Ω +
2
+k(v % )% k4,Ω k∇v % k4,Ω k∆v
e % k2,Ω ≤
e % k2,Ω + kv % k4,Ω k∇v % k4,Ω k∆v
≤ kf k2,Ω k∆v e % k2,Ω .
To estimate the first and the second factors in the last term of the right
hand side of the above inequality, we are going to exploit Ladyzhenskaya’s
inequality one more time
kv % k44,Ω ≤ 2kv % k22,Ω k∇v % k22,Ω ≤ c(a, f )k∇v % k22,Ω
and
k∇v % k44,Ω ≤ c(Ω)k∇v % k22,Ω (k∇2 v % k22,Ω + k∇v % k22,Ω ).
We also need the Cattabriga-Solonnikov inequality
k∇2 v % k22,Ω ≤ c(Ω)k∆v

e % k2,Ω .
So, combining the latter results, we find the basic estimate

1
∂t k∇v % k22,Ω + k∆v
e % k2 ≤ kf k2,Ω k∆v
2,Ω
e % k2,Ω
2
1 1 1 1
+c(a, f, Ω)k∇v % k 2 k∇v % k 2 (k∆v
2,Ω 2,Ω
e % k 2 + k∇v % k 2 )k∆v
2,Ω 2,Ω
e % k2,Ω
Now, if we apply Young’s inequalities in an appropriate way, we can arrive

at the following differential inequality
e % k2 ≤ c(a, f, Ω)(kf k2 + y + k∇v % k2 y),
∂t y + k∆v 2,Ω 2,Ω 2,Ω
where function y(t) = k∇v % (·, t)k22,Ω obeys the initial condition y(0) = k∇ak22,Ω .
The latter, together with energy estimate (5.3.7), gives two estimates
sup k∇v % (·, t)k22,Ω + k∇2 v % k22,QT ≤ c(a, f, Ω) < ∞. (5.5.12)

0<t<T
To get all remaining estimates, we make use of Theorem 4.1 of Section 4,

which reads
h i
% 2 % %
k∂t v k2,QT + k∇ v k2,QT + k∇p k2,QT ≤ c kf k2,QT + k∇ak2,Ω .
e
But
kfek2,QT ≤ kf k2,QT + k(v % )% · ∇v % k2,QT ≤
≤ kf k2,QT + k(v % )% k4,QT k∇v % k4,QT ≤
h i
% %
≤ c(a, f ) 1 + kv k4,QT k∇v k4,QT .
The right hand side of the above inequality can be evaluated with the help
of Ladyzhenskaya’s inequality and (5.5.12).
Finally, we can pass to the limit as % → 0 and get all the statements of
Theorem 5.2. Theorem 5.2 is proved.
5.6 Further Properties of Weak Leray-Hopf

Solutions
Theorem 6.1. Let Ω be a bounded domain with smooth boundary. Assume
f ∈ L2 (QT ), a ∈ H. (5.6.1)
5.6. FURTHER PROPERTIES 131
I. Let v be an arbitrary weak Leray-Hopf solution in QT with the right

hand side f and initial data a satisfying (5.6.1). Then, for each δ > 0 and
for any numbers s, l > 1 subject to the condition
3 2
+ = 4,
s l
we have
2,1
v ∈ Ws,l (Qδ,T ),
where Qδ,T = Ω×]δ, T [. Moreover, there exists a function q (pressure) which
belongs to the following spaces
1,0
q ∈ Ws,l (Qδ,T ) ∩ Ls0 ,l0 (Qδ,T ) (5.6.2)
with the same δ, s and l as above and

3s
s0 = , l0 = l
3−s
so that
3 2
0
+ 0 = 3.
s l
The Navier-Stokes equations
∂t v + div v ⊗ v − ∆ v = f − ∇q, div v = 0
hold in the sense of distributions and a.e. in QT .

II. There exists at least one weak Leray-Hopf solution v and a pressure
q with the properties mentioned in Part I such that, for any t0 ∈]0, T ], the
local energy inequality
Z Zt0 Z Zt0 Z
2 2
|v(x, t0 )| ϕ(x, t0 )dx + ϕ|∇v| dxdt ≤ |v|2 (∂t ϕ + ∆ϕ)
Ω 0 Ω 0 Ω

2
+v · ∇ϕ(|v| + 2q) + 2f · v dxdt (5.6.3)
holds for any non-negative function ϕ ∈ C0∞ (R3 ×]0, ∞[).

Proof I. Take χ ∈ C0∞ (0, inf ty) and let u = χv. Obviously,
u ∈ L∞ (0, T ; H) ∩ L2 (0, T ; V ).
∞
Next, plague χw with w ∈ C0,0 (QT ) into identity (iii) of Definition 3.1 and
get for u: Z Z
(−u · ∂t w + ∇u : ∇w)dz = fe · wdz,
QT QT
where
fe = χf − χdiv v ⊗ v − ∂t χv = χf − χv · ∇v − ∂t χv.
By Corollary 4.5,
fe ∈ Ls,l (QT )
for any s, l > 1 satisfying 3/s + 2/l = 4. Moreover, u = 0 for sufficiently
small t. On the other hand, the linear theory ensures that, for such fe, there
exist functions ve and qe such that
2,1 1,0
ve ∈ Ws,l (QT ), qe ∈ Ws,l (QT )
with finite numbers s, l > 1 satisfying 3/s + 2/l = 4 and
∂t ve − ∆e
v + ∇e
q = fe, div ve = 0
in QT , Z
ve(x, t) = 0 x ∈ ∂Ω, qe(x, t)dx = 0
Ω
for t ∈ [0, T ],
ve(·, 0) = 0
in Ω.
Using essentially the same arguments as in 2D-case, we can show that for
vb = u − ve Z
v · ∂t w − ∇b
(b v : ∇w)dz = 0
QT
∞
for w = χeW with W ∈ C0,0 (Ω) and with χ e ∈ C 2 ([0, T ]) and χ
e(T ) = 0.
Now, the uniqueness results for the linear theory imply that u b = 0. Hence,
2,1
χv ∈ Ws,l (QT ) and
χ(∂t v + v · ∇v − ∆v − f ) = −∇e
q.
Next, take any δ > 0 and assume that χ(t) = χδ (t) = 1 if t > δ and
χ(t) = χδ (t) = t/δ if < t < δ. A pressure qe corresponding to chosen δ is
1,0
denoted by qδ . Obviously, qδ ∈ Ws,l (Qδ,T ) with required s and l. Assuming
that δ1 > δ2 then qδ1 = qδ2 in Qδ1 ,T . This allows us to introduce the function
q so that
q(·, t) = qδ (·, t)
if t > δ > 0. It is well defined and satisfies the required properties. So, the
first part of the theorem is proved.
Part II. Now, let us go back to the proof of Theorem 3.2 on the existence
of weak Leray-Hopf solutions and try to apply the procedure, described in
the proof of Part I, to regularized problem. Letting u% = χv % , where χ is a
function of t from C02 (0, ∞), we state that u% is a solution to the problem:
u% ∈ C([0, T ]; H) ∩ L2 (0, T ; V ), ∂t u% ∈ L2 (0, T ; V 0 ); (5.6.4)
for a.a. t ∈ [0, T ]

Z
(∂t u% (x, t) · ve(x) + ∇u% (x, t) : ∇e
v (x))dx
Ω
Z
= fe% (x, t) · ve(x))dx (5.6.5)
Ω
for all ve ∈ V ;
ku% (·, t)k2,Ω → 0 (5.6.6)
as t → +0.
Here,
fe% = fe1% + fe2%
with
fe1% (x, t) = χ(t)f (x, t) + ∂t χ(t)v % (x, t),
fe2% (x, t) = −χ(t)(v % )% (x, t) · ∇v % (x, t).
Simply repeating the proof of Corollary 4.4 and Corollary 4.5, we estimate
the second part
kfe2% ks0 ,l0 ,QT ≤ c(s0 )|v % |22,QT ≤ c(s0 , a, f )
with s0 , l0 > 1 and 3/s0 + 2/l0 = 4. Moreover, we can claim that the whole
right hand side fe% is estimated similarly:
kfe% ks0 ,l0 ,QT ≤ c(s0 , a, f, χ, Ω).

e%
Now, according to Theorem 5.2 of Chapter 4, there exists functions u
and qχ% satisfy the relations:
e% ∈ Ws2,1
u 0 ,l0 (QT ), qχ% ∈ Ws1,0
0 ,l0 (QT );
e% − ∆ u
∂t u e% = fe% − ∇ qχ% , e% = 0
div u
in QT ;
e% |∂ 0 QT = 0,
u [qχ% (·, t)]Ω = 0.
By the same theorem, these functions have the bound
k∇2 u
e% ks0 ,l0 ,QT + k∂t u
e% ks0 ,l0 ,QT + k∇qχ% ks0 ,l0 ,QT
≤ c(s0 , Ω)kfe% ks0 ,l0 ,Q ≤ c(s0 , Ω, a, f, χ)
T
Applying Theorem 6.2 of Chapter 4 on uniqueness and the same arguments

as in 2D case, we can state
u% = u
e% .
Next, we consider the sequence of function χδ with δ = T /k, k ∈ N, such
that χδ (t) = t/δ if 0 < t < δ and χδ (t) = 1 if t ≥ δ. And thus
k∇2 v % ks0 ,l0 ,Qδ,T + k∂t v % ks0 ,l0 ,Qδ,T + k∇qδ% ks0 ,l0 ,Qδ,T
≤ c(s0 , Ω, a, f, δ). (5.6.7)
Let v % be a sequence constructed in the proof of Theorem 3.2, i.e.,

∗
v % *v
in L∞ (0, T ; H),
∇v % * ∇v
in L2 (0, T ; V ),
v% → v
in L2 (QT ), Z Z
%
v (x, t) · w(x)dx → v(x, t) · w(x)dx
Ω Ω
in C([0, T ]) for each w ∈ L2 (Ω).

In addition, we know that
kv % ks00 ,l00 ,QT ≤ c|v % |2,QT ≤ c(a, f )

with 3/s00 + 2/l00 = 3/2, see Corollary 4.4. In particular, we may assume that
(take s00 = l00 = 10/3)
v% * v
in L 10 (QT ) and therefore
3
v% → v (5.6.8)
in L3 (QT ) and
(v % )% · ∇v % * v · ∇v
in L1 (QT ).
Given s, l > 1 with 3/s+2/l=3, we find l0 = l and s0 = 3s/(s + 3). It
is easy to check, 3/s0 + 2/l0 = 4. Using the diagonal Cantor procedure and
bounds (5.6.7), one can ensure that
∂t v % * ∂t v, ∇2 v % * ∇2 v
in Ls0 ,l0 (Qδ,T ) for each δ > 0. As to the pressure, the diagonal Cantor proce-
dure can be used one more time to show that
∇qδ% * ∇qδ
in Ls0 ,l0 (Qδ,T ) and

qδ% * qδ
in Ls,l (Qδ,T ) for each δ. Here [qδ (·, t)]Ω = 0.
It is worthy to note that qδ1 = qδ2 in Q%2 ,T if δ1 < δ2 . Indeed, it follows
from two identities for δ = δ1 and δ = δ2
∇qδ = f − ∂t v − v · ∇v + ∆v
in Qδ,T and [qδ (·, t)]Ω = 0.

So, the function, defined as
q(z) = qδ (z), z ∈ Qδ,T ,
belongs Ws1,0
0 ,l0 (QT ) and Ls,l (QT ) for each δ > 0.
The last thing is to check validity of the local energy inequality. It

is known that the regularized solution is smooth enough and, therefore,
obeys the local energy identity. So, a given non-negative function ϕ ∈
C0∞ (R3 ×]0, ∞[, we choose δ so small that spt ϕ ∈ R3 ×]δ, ∞[. After mul-
tiplication of the equation
∂t v % + (v % )% · ∇v % − ∆v % = f − ∇qδ%
by ϕv % and integration of the product by parts (which is legal for the regu-
larized solution)
Z Zt0 Z Zt0 Z
|v % (x, t0 )|2 ϕ(x, t0 )dx + ϕ|∇v % |2 dxdt = |v % |2 (∂t ϕ + ∆ϕ)+
Ω 0 Ω 0 Ω

+(v % )% · ∇ϕ(|v % |2 + 2qδ% ) + 2f · v dxdt (5.6.9)
for any t ∈ [δ, T ].
We also know that
qδ% * q
in L 3 (Qδ,T ). (Indeed, qδ% * q in L 5 (Qδ,T ) since 3/(5/3) + 2/(5/3) = 3).
2 3
Taking into account (5.6.8) and using the same arguments as in the proof of
Theorem 3.2, one can pass to the limit in (5.6.9) as % → 0 and get required
local energy inequality (5.6.3).
5.7 Strong Solutions

Definition 7.1. A weak Leray-Hopf solution is called a strong solution, if
∇v ∈ L∞ (0, T ; L2 (Ω)). (5.7.1)
Theorem 7.2. (Global existence of strong solutions for ”small” data). There
exists a constant c0 (Ω) such that if
π
arctan(k∇ak22,Ω ) + c0 (Ω)(kak22,Ω + kf k22,QT ) < , (5.7.2)
2
then there exists a strong solution to initial boundary value problem (5.3.1)
Proof Let us go back to problem (5.3.4)-(5.3.6), see the proof of Theorem
3.2,
v % ∈ C([0, T ]; H) ∩ L2 (0, T ; V ), ∂t v % ∈ L2 (0, T ; V 0 ); (5.7.3)
5.7. STRONG SOLUTIONS 137
for a.a. t ∈ [0, T ]

Z
(∂t v % (x, t) · ve(x) + ∇v % (x, t) : ∇e
v (x))dx =
Ω
Z
= (v % (x, t) ⊗ (v % )% (x, t) : ∇e
v (x) + f % (x, t) · ve(x))dx (5.7.4)
Ω
for all ve ∈ V ;
kv % (·, t) − a(·)k2,Ω → 0 (5.7.5)
as t → +0. Here,
f % = f − (v % )% · ∇v % ∈ L2 (QT ).
Using the similar arguments as in the proof of Theorem 5.1 of this section
and Theorems 6.2, 5.1 of Section 4 on uniqueness and regularity for non-
stationary Stokes problem, we may conclude that
v % ∈ W22,1 (QT ), ∇v % ∈ C([0, T ]; L2 (Ω)).
Moreover, there exists a pressure field p% ∈ W21,0 (QT ) such that the regu-
larised Navier-Stokes equations
∂t v % − ∆v % = f % − ∇p% , div v % = 0
hold a.e. in QT . This is the starting point for the proof of our theorem.
We know that sequence v % converges to a weak Leray-Hopf solution to corre-
sponding initial boundary value problem (5.3.1). So, what we need is to get
uniform estimates of ∇v % . Let
Z
y(t) := |∇v % (x, t)|2 dx.
Ω
e %
We proceed as in the proof of Theorem 5.2, multiplying the equation by ∆v
and arguing exactly as it has been done there. As a result, after obvious
applications of Cauchy and Hölder inequalities, we find
Z Z
0
y + |∆v | dx ≤ c |(v % )% |2 |∇v % |2 dx + ckf k22,Ω ≤
e % 2
Ω Ω
≤ ckv % k26,Ω k∇v % k23,Ω + ckf k22,Ω . (5.7.6)
The first term on the right hand side of the latter inequality can be evaluated
with the help of the Galiardo-Nirenberg inequality
kv % k26,Ω ≤ cy,
and the multiplicative inequality

1
Z 21
k∇v % k23,Ω ≤ c(Ω)y 2 |∇2 v % |2 dx + y .
Ω
In addition, the Cattabriga-Solonnikov inequality of the form

Z Z
2 % 2 e % |2 dx.
|∇ v | dx ≤ c(Ω) |∆v (5.7.7)
Ω Ω
is needed. So, from (5.7.6) and (5.7.7), it follows that:

Z
0 e % |2 dx ≤ c(Ω)y 3 + ckf k22,Ω .
y + |∆v (5.7.8)
Ω
Recalling the properties of eigenvalues λk and eigenfunctions ϕk of the

Stokes operator, we observe that:
Z X∞ ∞
X
% 2 2 2
|∆v | =
e ck λ k ≥ λ 1 c2k λk = λ1 y,
Ω k=1 k=1
where Z
ck (t) = v % (x, t) · ϕk (x)dx.
Ω
So, (5.7.8) yields the final differential inequality
y 0 (t) + λ1 y(t) ≤ c1 (Ω)(y 3 (t) + g(t)), (5.7.9)
with
y(0) = k∇ak22,Ω , g(t) = kf (·, t)k22,Ω
A weaker versions of (5.7.9) is
y 0 (t) h i
≤ c1 (Ω) y(t) + g(t)
1 + y 2 (t)
so that after integration of it and application of the energy inequality, we

arrive at the bound
h ZT ZT i
arctan(y(t)) ≤ arctan(k∇ak22,Ω ) + c1 (Ω) y(t)tdt + g(t)tdt
h0 0
i π
≤ arctan(k∇ak22,Ω ) + c0 (Ω) kak22,Ω + kf k22,QT <
2
for any t ∈ [0, T ], which implies
y(t) ≤ C, t ∈ [0, T ],
with a constant c independent of % and t.

Theorem 7.3. Assume that
a ∈ V, f ∈ L2 (QT ).
Then there exists T 0 ∈]0, T ] such that initial boundary value problem (5.3.1)
has a strong solution in QT 0 .
Proof Arguing as in the proof of Theorem 7.2, let us go back to inequal-
ity (5.7.9). We need to show that there exists T 0 ≤ T , where y(t) has an
upper bound independent of % and t ∈ [0, T 0 ]. To achieve this goal, we make
a substitution z(t) = y(t) − y(0) and, after application of Young’s inequality,
arrive at the following modification of estimate (5.7.9)
z 0 (t) + λ1 z(t) ≤ c1 (Ω)(z 3 (t) + y 3 (0) + g(t)).
An equivalent form of it is:

c1 2
z 0 + λ1 (1 − z )z ≤ c1 (y 3 (0) + g(t))
λ1
with z(0) = 0. By continuity,
c1 2
z (t) < 1 (5.7.10)
λ1
for small positive t. Assume that t% is the first point, at which inequality
(5.7.10) is violated. So, inequality (5.7.10) holds for 0 < t < t% and
c1 2
z (t% ) = 1.
λ1
Next, we take the largest value T 0 ∈]0, T ] so that
ZT 0 h r
3
i 1 λ1
c1 y (0) + g(t) dt ≤ .
2 c1
0
From inequality (5.7.10) and the definition of t% , it follows that:
z 0 (t) ≤ c1 (y 3 (0) + g(t))
for all 0 ≤ t ≤ t% . Therefore,

Zt
z(t) ≤ c1 (y 3 (0) + g(s))ds
0
for all 0 ≤ t ≤ t% . And thus for t = t% , we have

r Zt%
λ1
z(t% ) = ≤ c1 (y 3 (0) + g(s))ds
c1
0
and, in a view of the definition of T 0 ,

ZT 0 h i Zt% h i
3
y (0) + g(t) dt < y 3 (0) + g(s) ds.
0 0
The latter implies t% ≥ T 0 for all % > 0 and the required estimate
ZT 0 h r
i 1 λ1
y(t) ≤ k∇ak22,Ω + c1 3 2
y (0) + g(t) dt ≤ k∇ak2,Ω +
2 c1
0
for all 0 < t ≤ T 0 and for all % > 0.

Remark 7.4. If f = 0, the lower bound for T 0 can be improved
c4 (Ω)
T0 ≥
k∇ak42,Ω
and this is the celebrated Leray estimate.

Proof In this case, one can deduce from inequality (5.7.9) the following:
y0
≤ c1
y3
for 0 ≤ t ≤ T . The integration gives us:
1 1
− ≤ 2c1 t
y 2 (0) y 2 (t)
and thus
y 2 (t)(1 − 2c1 ty 2 (0)) ≤ y 2 (0).
√
We let T00 = 4c1 y12 (0) . Then 1 − 2c1 ty 2 (0) ≥ 1/2 and y(t) ≤ 2y(0) for
0 < t ≤ T00 which implies T 0 ≥ T00 and we get the required estimate with an
appropriated constant.
Remark 7.5. Solutions, constructed in Theorems 7.2 and 7.3, have the fol-
lowing regularity properties: v ∈ W22,1 (QT ) and there exists a function q such
that ∇q ∈ L2 (QT ) and
∂t v + v · ∇v − ∆v = f − ∇q, div v = 0
a.e. in QT . (One should be replace QT with QT 0 in the case of Theorem 7.3).

Proof In fact, from (5.7.8) and from the Cattabriga-Solonnikov inequal-
ity (5.7.7), it follows that:
k∇2 v % k22,QT ≤ c(k∇ak2,Ω , kf k2,QT )
and, in a view of derivation (5.7.8), we get

Z
|(v % )% |2 |∇v % |2 dz ≤ c(k∇ak2,Ω , kf k2,QT ).
QT
The linear theory, applied to the initial boundary value problem
∂t v % − ∆v % + ∇q % = f − (v % )% · ∇v % , div v % = 0 in QT
% %
v |∂Ω×[0,T ] = 0, v |t=0 = a,
leads to all other statements of Remark 5.7.5.

The main result of this section is:
Theorem 7.6. (Uniqueness of strong solutions in the class of weak Leray-

Hopf solutions) Assume that u1 and u2 are weak Leray-Hopf solutions to the
initial boundary value problem
∂t v + v · ∇v − ∆v = f − ∇q, div v = 0 in QT
v|∂Ω×[0,T ] = 0, v|t=0 = a.
with a ∈ V and f ∈ L2 (QT ). Let u2 be a strong solution then u1 = u2 ,
We start with several auxiliary propositions.
Proposition 7.7. (Uniqueness of strong solutions in the class of strong so-

lutions) Assume that u1 and u2 are strong solutions to the initial boundary
value problem
∂t v + v · ∇v − ∆v = f − ∇q, div v = 0 in QT
v|∂Ω×[0,T ] = 0, v|t=0 = a,
with a ∈ V and f ∈ L2 (QT ). Then u1 = u2 .
Proof First we note that
∂t u ∈ L2 (0, T ; V 0 ).
Indeed,
Z Z
(u · ∇u) · wdx = u ⊗ u : ∇wdx ≤ kuk24,Ω k∇wk2,Ω

Ω Ω
≤ c(Ω)k∇uk22,Ω k∇wk2,Ω ≤ C(Ω, u)k∇wk2,Ω
for any w ∈ V . So, fe = f − u · ∇u ∈ L2 (0, T ; V 0 ) and thus ∂t u ∈ L2 (0, T ; V 0 ).

Then, by the definition of weak solution, we have
Z h
χ(t) ∂t u(x, t) · w(x) − (u ⊗ u)(x, t) : ∇w(x)
QT
i Z
+∇u(x, t) : ∇w(x) dz = χ(t)f (x, t) · w(x)dz
QT
∞
for any w ∈ C0,0 (Ω) and for any χ ∈ C0∞ (0, T ). It is easy to see
Z h
∂t u(x, t) · w(x) − (u ⊗ u)(x, t) : ∇w(x)
Ω
i Z
+∇u(x, t) : ∇w(x) dx = f (x, t) · w(x)dx
Ω
for any w ∈ V and for a.a. t ∈ [0, T ].

So, assume that u1 and u2 are two different strong solutions and let v =
u − u2 . Then, we have
1
Z
1
∂t kvk2,Ω + k∇vk2,Ω = (u1 ⊗ u1 − u2 ⊗ u2 ) : ∇vdx
2 2
2
Ω
Z Z
= (v ⊗ u + u ⊗ v) : ∇vdx = u2 ⊗ v : ∇vdx
1 2
Ω Ω
Z
=− v ⊗ v : ∇u2 dx ≤ k∇u2 k2,Ω kvk24,Ω . (5.7.11)
Ω
Let us recall the following 3D multiplicative inequality

1 3
kvk4,Ω ≤ ckvk2,Ω
4
k∇vk2,Ω
4
.
So, after application of Young’s inequality
∂t kvk22,Ω + k∇vk22,Ω ≤ ck∇u2 k42,Ω kvk22,Ω .
Since k∇u2 k2,∞,QT ≤ C1 ,
∂t kvk22,Ω ≤ C14 kvk22,Ω ,
which implies
Z Z
−C1 t 2
e |v(x, t)| dx ≤ |v(x, 0)|2 dx = 0.
Ω Ω

Proposition 7.8. (Smoothness of strong solutions) Let a vector field u be a

strong solution to the following initial boundary value problem:
∂t u + u · ∇u − ∆u = f − ∇p, div u = 0 in QT
u|∂Ω×[0,T ] = 0, u|t=0 = a,
with a ∈ V and f ∈ L2 (QT ). Then u ∈ W22,1 (QT ) and p ∈ W21,0 (QT ).
Proof Let us denote k∇uk2,∞,QT by A. Coming back to our proof of

Theorem 7.3, define a positive number TA so that
h Z i 1rλ
3 1
c1 TA A + g(s)ds ≤ (5.7.12)
2 c1
]0,T [∩]t,t+TA [
for any t ∈ [0, T ].

By (5.7.12) and by Theorem 7.3, there exists strong solution u1 in QTA
with initial data u1 |t=0 = a. This solution belongs to W22,1 (QTA ) and the
corresponding pressure q 1 belongs to W21,0 (QTA ), see Remark 7.5. By Propo-
sition 7.7, u1 = u in QTA . We know also ∇u ∈ C(0, T ; L2 (Ω)). So, we can
apply Theorem 7.3 one more time in QTA /2,3TA /2 and find a strong solution
u2 there with initial data u(·, TA /2). By Proposition 7.7, u2 = u QTA /2,3TA /2
and ∇q 1 = ∇q 2 in QTA /2,TA . After a finite number of steps, we find that
u ∈ W22,1 (QT ) and can easily recover a function p ∈ W21,0 (QT ) such that
∇p = ∇q k on QT ∩ QTA k/2,TA (k+1)/2 , where k = 1, 2, ....
Proof of Theorem 7.6 Since u2 is a strong solution, it satisfies the
identity
Z h
∂t u2 (x, t) · w(x) + (u2 (x, t) · ∇u2 (x, t)) · w(x)
Ω
i Z
2
+∇u (x, t) : ∇w(x) dx = f (x, t) · w(x)dx (5.7.13)
Ω
for any w ∈ V and for a.a. t ∈ [0, T ]. Regarding to u1 , we have a weaker

identity
Z
(−u1 · ∂t w − u1 ⊗ u1 : ∇w + ∇u1 : ∇w − f · w)dz = 0 (5.7.14)
QT
∞
for any w ∈ C0,0 (QT ).
We would like to test (5.7.14) with u2 but it should be justified. Indeed,
we know that
u1 ∈ L 10 (QT )
3
and, by density arguments, (5.7.14) must be true for w(x, t) = χ(t)v(x) with
◦ ◦
v ∈ J 15 (Ω) and χ ∈ W 11 (0, T ) = {χ ∈ W11 (0, T ) : χ(0) = 0, χ(T ) = 0}. For
2
domains with sufficiently smooth domains
◦ ◦
J 5 (Ω) = Jˆ5 (Ω) = {v ∈ L 5 (Ω) : div v = 0}.
1 1 1
2 2 2
So, we observe that (5.7.14) holds for χϕk , where ϕk is the

P∞kth eigenfunction
2 2,1
of the Stokes operator. Since u ∈ W2 (QT ), the series k=1 P ck (t)ϕk (x) con-
verges to u in W2 (QT ). This, in turn, implies that the series ∞
2 2,1
k=1 ck (t)∇ϕk (x)
converges to ∇u2 in L 5 (QT ). The latter follows from the multiplicative in-
2
equality
7 3
k∇v(·, t)k 5 ,Ω ≤ c(Ω)k∇v(·, t)k2,Ω
10
(k∇2 (·, t)k2,Ω + k∇(·, t)k2,Ω ) 10 ,
2
which is valid for sufficiently smooth v. Hence,
7 3
k∇vk 5 ,QT ≤ c(Ω, T ) sup k∇vk2,Ω
10
(k∇2 vk2,QT + k∇vk2,QT ) 10 .
2
0<t<T
Having this inequality in hands, it is not so difficult to show that (5.7.14)

can be tested with χα,β u2 , where χα,β (t) = t/α if 0 ≤ t ≤ α, χα,β (t) = 1 if
α < t < β, χα,β (t) = (t0 + β − t)/β if t0 ≤ t ≤ t0 + β, and χα,β (t) = 0 if
t0 + β < t ≤ T . Plaguing w = χα,β u2 into (5.7.14), we find
Z
χα,β (−u1 · ∂t u2 − u1 ⊗ u1 : ∇u2 + ∇u1 : ∇u2 − f · u2 )dz
QT
Z
=− u1 · u2 χ0α,β dz = Iα + Iβ ,
QT
where
Zα Z Zα Z
1 1
−Iα = u1 · u2 dxdt = (u1 − a) · (u2 − a)dxdt
α α
0 Ω 0 Ω
Zα Z Zα Z
1 2 1
+ a · (u − a)dxdt + a · (u1 − a)dxdt
α α
0 Ω 0 Ω
Zα Z
1
+ |a|2 dxdt.
α
0 Ω
Since ku1 (·, t) − a(·)k2,Ω and ku2 (·, t) − a(·)k2,Ω go to zero as t → 0, we can
observe that Z
Iα → − |a|2 dx
Ω
as α → 0.
The analogous result takes place at the right end point:
tZ
0 +βZ
1
Iβ = u1 · u2 dxdt
β
t0 Ω
tZ
0 +βZ
1
= u1 (x, t) · (u2 (x, t) − u2 (x, t0 ))dxdt
β
t0 Ω
tZ
0 +βZ
1
+ (u1 (x, t) − u1 (x, t0 )) · u2 (x, t0 )dxdt
β
t0 Ω
Z
+ u1 (x, t0 ) · u2 (x, t0 )dx
Ω
By strong continuity in L2 (Ω) of the strong solution u2 , the first term on the
right hand side goes to zero and by weak continuity of weak solution u1 the
second term there goes to zero as well. So,
Z
Iβ → u1 (x, t0 ) · u2 (x, t0 )dx
Ω
as β → 0. Finally, we have
Zt0 Z
(−u1 · ∂t u2 − u1 ⊗ u1 : ∇u2 + ∇u1 : ∇u2 − f · u2 )dz
0 Ω
Z Z
1 2
= u (x, t0 ) · u (x, t0 )dx − |a|2 dx (5.7.15)
Ω Ω
for any t0 ∈ [0, T ].

Now, we going to test (5.7.13) with w(x) = u1 (x, t)−u2 (x, t), which, after
integration over ]0, t0 [, gives us:
Zt0 Z Z Z
2 1 1 2 2 1
∂t u (x, t) · u (x, t)dxdt − |u (x, t0 ) dx + |a|2 dx
2 2
0 Ω Ω Ω
Zt0 Z
+ − u2 ⊗ u2 : ∇(u1 − u2 ) + ∇u1 : ∇(u1 − u2 )
0 Ω

1 2
−f · (u − u ) dxdt = 0.
So, adding the latter to (5.7.15), we find

Z h
1 1 i
− |a(x)|2 + u1 (x, t0 ) · u2 (x, t0 ) − |u2 (x, t0 )|2 dx
2 2
Ω
Zt0 Z h
− u1 ⊗ u1 : ∇u2 − u2 ⊗ u2 : ∇(u1 − u2 ) − |∇u1 |2
0 Ω
i
+2∇u1 · ∇u2 + f · u1 dxdt = 0. (5.7.16)
We also know that weak solution satisfies the energy inequality
Z Zt0 Z Z Zt0 Z
1
|u1 (x, t0 )|2 dx + |∇u1 |2 dxdt ≤ |a|2 dx + f · u1 dxdt.
2
Ω 0 Ω Ω 0 Ω
Subtracting (5.7.16) from the energy inequality, we show

Z Zt0 Z
1 1 2 2
|u (x, t0 ) − u (x, t0 )| dx + |∇(u1 − u2 )|2 dxdt
2
Ω 0 Ω
Zt0 Z
≤− (u2 ⊗ u2 − u1 ⊗ u1 ) : ∇(u1 − u2 )dxdt = I. (5.7.17)
0 Ω
The rest of the proof is similar to the proof of Theorem 7.2. Indeed,
Zt0 Z
I= (u2 ⊗ (u1 − u2 )) : ∇(u1 − u2 )dxdt
0 Ω
Zt0 Z
− (u1 − u2 ) ⊗ (u1 − u2 ) : ∇u2 dxdt
0 Ω
Zt0
≤ k∇u2 k2,Ω ku2 − u1 k24,Ω dt
0
Zt0
≤ sup k∇u2 k2,Ω ku2 − u1 k24,Ω dt.
0<t<T
0
Since u2 is a strong solution, the quantity sup0<t<T ku2 k4,Ω = k∇u2 k2,∞,QT is
finite. Applying multiplicative inequality, we have
Zt0 1 3
I ≤ ck∇u2 k2,∞,QT ku2 − u1 k2,Ω
2
k∇(u2 − u1 )k2,Ω
2
dt
0
Zt0 Z 41 Zt0 Z 34
2 2 1 2 2 1 2
≤ ck∇u k2,∞,QT |u − u | dxdt |∇(u − u )| dxdt .
0 Ω 0 Ω
From here and from (5.7.17), it follows that:

Z
0 2
y (t0 ) ≤ ck∇u k2,∞,QT y(t0 ), y(t0 ) = |u1 (x, t0 ) − u2 (x, t0 )|2 dx
Ω
for all t0 ∈ [0, T ] with y(0) = 0. This immediately implies u1 = u2 .

Theorem 7.9. (Ladyzhenskaya-Prodi-Serrin condition) Let a ∈ V and f ∈

L2 (QT ). Assume that we have two weak Leray-Hopf solutions u1 and u2 .
Assume that u2 obeys the Ladyzhenskaya-Prodi-Serrin condition, i.e.,
u2 ∈ Ls,l (QT )
with s, l ≥ 1, satisfying
3 2
+ = 1.
s l
Then u1 = u2 .
Proof Just for simplicity let us assume f = 0. We also suppose that
s > 3. The case s = 3 and l = ∞ is much more complicated and will be
discussed later. Our aim is to show that a weak Leray-Hopf solution, satis-
fying Ladyzhenskaya-Prodi-Serrin condition, is in fact a strong one. Then,
the statement of the theorem follows from Theorem 7.6.
We know that, by Theorem 7.3, there exists a strong solution to our
initial-boundary value problem on a small time interval [0, T ], which, by
Theorem 7.6, coincides with any weak solution and, in particular, with u ≡
u2 . Let us denote by T0 (≤ T ) the first instance of time, for which u is not a
strong solution on [0, T0 ]. By Proposition 7.8, we have for any T 0 < T0 ,
u ∈ W22,1 (QT 0 ), ∇u ∈ C([0, T 0 ]; L2 (QT 0 ))
and there exists a pressure field p ∈ W21,0 (QT 0 ) so that
∂t u + u · ∇u − ∆u = −∇p, div u = 0
a.e. in QT 0 .
By Remark 7.4,
c4
T0 − t ≥
k∇u(t)k42,Ω
for all t < T0 . So, we have
lim k∇u(t)k2,Ω = ∞. (5.7.18)
t→T0 −0
We proceed as in the proof of (5.7.6) simply replacing v % with u. As a

result,
Z Z
0
y + |∆u| dx ≤ c |u|2 |∇u|2 dx,
e 2
(5.7.19)
Ω Ω
where y(s) = k∇uk22,Ω .

Applying consequently Hölder inequality, an appropriated multiplicative
inequality, and Cattabriga-Solonnikov inequality to the right hand side of
(5.7.19), we have
Z
|u|2 |∇u|2 dx ≤ kuk2s,Ω k∇uk22s ,Ω
s−2
Ω
2(1− 3s ) 6
≤ c(Ω, s)kuk2s k∇uk2 (k∇2 uk2 + k∇uk2 ) s
4 6
≤ c(Ω, s)kuk2s,Ω k∇uk2,Ω
l
k∆uk
e s
2,Ω
Using Young’s inequality, we arrive at the final inequality
y 0 (t) ≤ c(Ω, s)ku(·, t)kls,Ω y(t)
for all t < T0 . Integrating it, we find

c(Ω,s)kukls,l,Q
y(t) ≤ y(0)e T
for all t < T0 . This contradicts (5.7.18).
Remark 7.10. Unfortunately, we do not know whether any weak Leray-Hopf

solution u satisfies Ladyzhenskaya-Prodi-Serrin condition. What we know is
that
u ∈ Ls0 ,l0 (QT )
with s0 , l0 ≤ 1 and
3 2 3
+ = .
s0 l 0 2
So, there is a finite gap.
The problem of uniqueness of weak solution is still open.

If we show that any weak (Leray-Hopf) solution is smooth (for example,
it is strong), then we have uniqueness in the class of weak solutions.
The problem of smoothness of weak solutions is one of seven Millennium
problems.
Chapter 6
Local Regularity Theory for

Non-Stationary Navier-Stokes
Equations
6.1 Notation
In this chapter, we are going to exploit the following notation:
R+ = {t ∈ R : t > 0}, R− = {t ∈ R : t < 0};
Rd+ = {x = (x0 , xd ) : x0 = (xi ), i = 1, 2, ..., d − 1, xd > 0};
Q− = Rd × R− , Q+ = Rd × R+ ;
Qδ,T = Ω×]δ, T [, QT = Ω×]0, T [, Ω ⊂ Rd ;
B(x, r) is the ball in Rd of radius r centered at the point x ∈ Rd , B(r) =
B(0, r), B = B(1);
B+ (x, r) = {y = (y 0 , yd ) ∈ B(x, r) : yd > xd } is a half ball, B+ (r) = B+ (0, r),
B+ = B+ (1);
Q(z, r) = B(x, r)×]t−r2 , t[ is the parabolic ball in Rd ×R of radius r centered
at the point z = (x, t) ∈ Rd × R, Q(r) = Q(0, r), Q = Q(1);
Q+ (r) = Q+ (0, r) = B+ (r)×] − r2 , 0[;
Ls (Ω) and Ws1 (Ω) are the usual Lebesgue and Sobolev spaces, respectively;
Ls,l (QT ) = Ll (0, T ; Ls (Ω)), Ls (QT ) = Ls,s (QT );
151
152 CHAPTER 6. LOCAL REGULARITY THEORY
1,0 1,0
Ws,l (QT ) = {|v| + |∇v| ∈ Ls,l (QT )} and Ws,l (QT ) = {|v| + |∇v| + |∇2 v| +
|∂t v| ∈ Ls,l (QT )} are parabolic Sobolev spaces;
∞
C0,0 (Ω) = {v ∈ C0∞ (Ω) : div v = 0};
◦ ◦
∞
J (Ω) is the closure of the set C0,0 (Ω) in the space L2 (Ω), J 12 (Ω) is the closure
of the same set with respect to the Dirichlet integral;
BM O is the space of functions having bounded mean oscillation;
C(Ω) is the space of continuous function, C α (QT ) is the space of Hölder
continuous with respect to the parabolic metrics;
c is a generic positive constant.
6.2 ε-Regularity Theory

The aim of this section is the so-called suitable weak solutions to the Navier-
Stokes equations and their smoothness. Those solutions were introduced in
[2], see also [42]-[45], [38], and [31]. Our version is due to [38].
Definition 2.1. Let ω be a open set in R3 . We say that a pair u and p is a

suitable weak solution to the Navier-Stokes equations in ω×]T1 , T [ if u and p
obey the conditions:
u ∈ L2,∞ (ω×]T1 , T [) ∩ L2 (T1 , T ; W21 (ω)); (6.2.1)
p ∈ L 3 (ω×]T1 , T [); (6.2.2)

2
∂t u + u · ∇u − ∆u = −∇p, div u = 0 (6.2.3)

in the sense of distributions;
the local energy inequality

ϕ|∇u|2 dxdt0
R R
ϕ(x, t)|u(x, t)|2 dx + 2 

ω ω×]T1 ,t[ 

(6.2.4)
0
R 2 2
≤ (|u| (∆ϕ + ∂t ϕ) + u · ∇ϕ(|u| + 2q)) dxdt 



ω×]T1 ,t[
holds for a.a. t ∈]T1 , T [ and all nonnegative functions ϕ ∈ C0∞ (ω×]T1 , ∞[).
One of the main results of the theory of suitable weak solutions reads
6.2. ε-REGULARITY THEORY 153
Lemma 2.2. There exist absolute positive constants ε0 and c0k , k = 1, 2, ...,
with the following property. Assume that a pair U and P is a suitable weak
solution to the Navier-Stokes equations in Q and satisfies the condition
Z
3
3
|U | + |P | dz < ε0 .
2 (6.2.5)
Q
Then, for any natural number k, ∇k−1 U is Hölder continuous in Q( 21 ) and

the following bound is valid:
max |∇k−1 U (z)| < c0k . (6.2.6)

z∈Q( 12 )
Remark 2.3. For k = 1, Lemma 2.2 has been proven essentially in [2], see
Corollary 1. For alternative approach, we refer the reader to [31], see Lemma
3.1. The case k > 1 was treated in [40], see Proposition 2.1, with the help of
the case k = 1 and regularity results for linear Stokes type systems.
In turn, if k = 1, Lemma 2.2 is a consequence of the following statement.
Proposition 2.4. Given numbers ϑ ∈]0, 1/2[ and M > 3, there are two
constants ε1 (ϑ, M ) > 0 and c1 (M ) > 0 such that, for any suitable weak
solution v and q to the Navier-Stokes equations in Q, satisfying the additional
conditions
|(v),1 | < M, Y1 (v, q) < ε1 , (6.2.7)
the following estimate is valid:
2
Yϑ (v, q) ≤ c1 ϑ 3 Y1 (v, q). (6.2.8)
Here and in what follows, we use the notation:
Y (z0 , R; v, q) = Y 1 (z0 , R; v) + Y 2 (z0 , R; q),
1 Z 31
1
Y (z0 , R; v) = |v − (v)z0 ,R |3 dz ,
|Q(R)|
Q(z0 ,R)
1 Z
3
23
2
Y (z0 , R; q) = R |q − [q]z0 ,R | 2 dz ,
|Q(R)|
Q(z0 ,R)
Z Z
1 1
(v)z0 ,R = v dz, [q]x0 ,R = q dx,
|Q(R)| |B(R)|
Q(z0 ,R) B(x0 ,R)
Yϑ1 (v) = Y 1 (0, ϑ; v), Yϑ2 (q) = Y 2 (0, ϑ; q),

Yϑ (v, q) = Y (0, ϑ; v, q), (v),ϑ = (v)0,ϑ , [q],ϑ = [q]0,ϑ .
Proof of Proposition 2.4 Assume that the statement is false. This
means that a number ϑ ∈]0, 1/2[ and a sequence of suitable weak solutions
v k and q k (in Q) exist such that:
Y1 (v k , q k ) = ε1k → 0 (6.2.9)
as k → +∞,
3
Yϑ (v k , q k ) > c1 ε1k ϑ 2 (6.2.10)
for all k ∈ N. The constant c1 will be specified later. Let us introduce
functions
uk = (v k − (v k ),1 )/ε1k , pk = (q k − [q k ],1 )/ε1k .
They obey the following relations
Y1 (uk , pk ) = 1, (6.2.11)
2
Yϑ (uk , pk ) > c1 ϑ 3 , (6.2.12)
and the system
1
∂t uk + div ((v k ),1 + ε1k uk ) ⊗ ((v k ),1 + ε1k uk )

ε1k in Q (6.2.13)
−∆ uk = −∇ pk , div uk = 0
in the sense of distributions.

Without loss of generality, one may assume that:
 k
 u *u in L3 (Q)
k
p *p in L 3 (Q) (6.2.14)
 k 2
(v ),1 → b in R3
and
∂t u + div u ⊗ b − ∆ u = −∇ p
in Q (6.2.15)
div u = 0
in the sense of distributions. By (6.2.11) and (6.2.14) , we have
|b| < M, Y1 (u, p) ≤ 1, [p(·, t)],1 = 0 for all t ∈] − 1, 0[. (6.2.16)
Choosing a cut-off function ϕ in an appropriate way in the local energy

inequality, we find
kuk k2,∞,Q(3/4) + k∇ uk k2,Q(3/4) ≤ c2 (M ) (6.2.17)
and thus, for the limit function,
kuk2,∞,Q(3/4) + k∇ uk2,Q(3/4) ≤ c2 (M ).
It is easy to check that p is a harmonic function depending on t as a

parameter. After application of bootstrap arguments, we find

sup |∇u(z)| + |∇2 u(z)|
z∈Q(2/3)
Z0 23
3
+ sup |∂t u(x, t)| dt ≤ c3 (M ).
2
x∈B(2/3)
−(2/3)2
From the above estimate, a parabolic embedding theorem and scaling, it

follows that 1 Z 31
|u − (∇u),τ x − (u),τ |3 dz
|Q(τ )|
Q(τ )
1
Z
3 3
23
2 2
≤ cτ (|∇ u| + |∂t u| )dz
2 2
|Q(τ )|
Q(τ )
1 23 2
≤ cτ 2 C(M ) + 2 C(M ) ≤ C(M )τ 3
τ
for all 0 < τ < 2/3. The latter estimates gives us:
2
Yϑ1 (u) ≤ e
c1 (M )ϑ 3 . (6.2.18)
Using the known multiplicative inequality, see the previous chapter, we

derive from (6.2.17) another estimate
kuk k 10 ,Q(3/4) ≤ c4 (M ). (6.2.19)

3
Let us estimate the first derivative in time with the help of duality argu-
ments. Indeed, we have from (6.2.13) and (6.2.16)
k∂t uk k ◦ ≤ c5 (M ). (6.2.20)
L 3 (−(3/4)2 ,0;(W 22 (B(3/4)))0 )
2
◦
Here, W 22 (B(3/4)) is the completion of C0∞ (B(2/3)) in W22 (B(2/3)). By the
compactness arguments used in the previous section, a subsequence can be
selected so that
uk → u in L3 (Q(3/4)). (6.2.21)
Now, taking into account (6.2.21) and (6.2.18), we pass to the limit in
(6.2.12) and find
2 2
c1 ϑ 3 + lim sup Yϑ2 (pk ).
c1 ϑ 3 ≤ e (6.2.22)
k→∞
To pass to the limit in the last term of the right hand side in (6.2.22), let us
decompose the pressure pk so that (see [46, 51, 52])
pk = pk1 + pk2 . (6.2.23)
Here, the first function pk1 is defined as a unique solution to the following
boundary value problem: find pk1 (·, t) ∈ L 3 (B) such that
2
Z Z
pk1 (x, t)∆ψ(x) dx = −ε1k uk (x, t) ⊗ uk (x, t) : ∇2 ψ(x) dx
B B
for all smooth test functions ψ subjected to the boundary condition ψ|∂B = 0.
It is easy to see that
∆pk2 (·, t) = 0 in B (6.2.24)
and, by the coercive estimates for Laplace’s operator with the homogeneous
Dirichlet boundary condition, we get the bound for pk1 :
Z 3
Z
3
|p1 (x, t)| 2 dx ≤ cε1k |uk (x, t)|3 dx.
k 2
(6.2.25)
B B
Passing to the limit in (6.2.22), we show with the help of (6.2.25) that:
2 2
c1 ϑ 3 + lim sup Yϑ2 (pk2 ).
c1 ϑ 3 ≤ e (6.2.26)
k→∞
By Poincare’s inequality, (6.2.26) can be reduced to the form

2 2
1 Z 23
2 k 23
c1 ϑ ≤ e
3 c1 ϑ + cϑ lim sup
3 |∇ p2 | dz . (6.2.27)
k→∞ |Q(ϑ)|
Q(ϑ)
We know that the function pk2 (·, t) is harmonic in B and thus the following
estimate is valid:
Z
3 3
k
sup |∇ p2 (x, t)| ≤ c |pk2 (x, t)| 2 dx
2
x∈B(3/4)
B
which in turns implies

Z Z
1 3 c 3
|∇ pk2 | 2 dz ≤ 2 |pk2 | 2 dz
|Q(ϑ)| ϑ
Q(ϑ) Q
1 Z
1 3

≤c + |pk1 | 2 dz .
ϑ2 ϑ2
Q
The latter inequality, together with (6.2.25), allows us to take the limit in
(6.2.27). As a result, we show that
2 2 2
c1 ϑ 3 ≤ e
c1 ϑ 3 + cϑ 3 . (6.2.28)
If, from the very beginning, c1 is chosen so that
c1 = 2(e
c1 + c),
we arrive at the contradiction. Proposition 2.4 is proved.
Proposition 2.4 admits the following iterations.
Proposition 2.5. Given numbers M > 3 and β ∈ [0, 2/3[, we choose ϑ ∈
]0, 1/2[ so that
2−3β
c1 (M )ϑ 6 < 1. (6.2.29)
Let ε1 (ϑ, M ) = min{ε1 (ϑ, M ), ϑ5 M/2}. If
|(v),1 | < M, Y1 (v, q) < ε1 , (6.2.30)
then, for any k = 1, 2, ...,
ϑk−1 |(v),ϑk−1 | < M, Yϑk−1 (v, q) < ε1 ≤ ε1 ,
2+3β (6.2.31)
Yϑk (v, q) ≤ ϑ 6 Yϑk−1 (v, q).
Proof We use induction on k. For k = 1, this is nothing but Proposition

2.4.
Assume now that statements (6.2.31) are valid for s = 1, 2, ..., k ≥ 2. Our
goal is to prove that they are valid for s = k + 1 as well. Obviously, by
induction,
Yϑk (v, q) < ε1 ≤ ε1 ,
and
|(v k ),1 | = ϑk |(v),ϑk | ≤ ϑk |(v),ϑk − (v),ϑk−1 | + ϑk |(v),ϑk−1 |
1 1 1
≤ 5 Yϑk−1 (v, q) + ϑk−1 |(v),ϑk−1 | < 5 ε1 + M/2 ≤ M.
ϑ 2 ϑ
Introducing scaled functions
v k (y, s) = ϑk v(ϑk y, ϑ2k s), q k (y, s) = ϑ2k q(ϑk y, ϑ2k s)
for (y, s) ∈ Q, we observe that v k and q k are a suitable weak solution in Q.
Since
Y1 (v k , q k ) = ϑk Yϑk (v, q) < ε1 ≤ ε1
and
|(v k ),1 | = ϑk |(v),ϑk | < M,
we conclude
2 2+3β
Yϑ (v k , q k ) ≤ c1 ϑ 3 Y1 (v k , q k ) < ϑ 6 Y1 (v k , q k ),
which is equivalent to the third relation in (6.2.31). Proposition 2.5 is proved.
A direct consequence of Proposition 2.5 and the Navier-Stokes scaling
v R (y, s) = Rv(x0 + Ry, t0 + R2 s), q R (y, s) = R2 q(x0 + Ry, t0 + R2 s)
is the following statement.
Proposition 2.6. Let M , β, ϑ, and ε1 be as in Proposition 2.5. Let a pair
v and q be an arbitrary suitable weak solution to the Navier-Stokes equations
in the parabolic cylinder Q(z0 , R), satisfying the additional conditions
R|(v)z0 ,R | < M, RY (z0 , R; v, q) < ε1 . (6.2.32)
Then, for any k = 1, 2, ..., the estimates
2+3β
Y (z0 , ϑk R; v, q) ≤ ϑ 6
k
Y (z0 , R; v, q) (6.2.33)
hold.
Proof of Lemma 2.2 We start with the case k = 1. Define

Z
3
A= |U |3 + |P | 2 dz.
Q
Then, let M = 2002, β = 1/3, and let ϑ be chosen according to (6.2.29) and
fix.
First, we observe that
Q(z0 , 1/4) ⊂ Q if z0 ∈ Q(3/4)
and
1 1 2 1 1
Y (z0 , 1/4; U, P ) ≤ c(A 3 + A 3 ), |(U )z0 , 1 | ≤ cA 3 .
4 4 4
Selecting ε0 so that
1 2 1
c(ε03 + ε03 ) < ε1 , cε03 < 2002.
Then, by (6.2.5), we have

1 1
Y (z0 , 1/4; U, P ) < ε1 , |(U )z0 , 1 | < M,
4 4 4
and thus, by Proposition 2.6,

k k
Y (z0 , ϑk /4; U, P ) ≤ ϑ 2 Y (z0 , 1/4; U, P ) ≤ ϑ 2 ε1
for all z0 ∈ Q(3/4) and for all k = 1, 2, .... Hölder continuity of v on the set
Q(2/3) follows from Campanato’s type condition. Moreover, the quantity
sup |v(z)|
z∈Q(2/3)
is bounded by an absolute constant.

The case k > 1 is treated with the help of the regularity theory for the
Stokes equations and bootstrap arguments, for details, see [40], Proposition
2.1. Lemma 2.2 is proved.
In what follows, the scaled energy quantities will be exploited
Z Z
1 2 1
A(v; z0 , r) ≡ sup |v(x, t)| dx, E(v; z0 , r) ≡ |∇ v|2 dz,
2
t0 −r ≤t≤t0 r r
B(x0 ,r) Q(z0 ,r)
Z Z
1 3 1 3
C(v; z0 , r) ≡ 2 |v| dz, D0 (q; z0 , r) ≡ 2 |q − [q]x0 ,r | 2 dz.
r r
Q(z0 ,r) Q(z0 ,r)
We are also going to use abbreviations like A(r) = A(v; 0, r), etc.
Our aim is to prove a version of the Caffarelli-Kohn-Nireberg theorem
(Here, we follow F.-H. Lin’s arguments, see [38]).
Theorem 2.7. Let v and q be a suitable weak solution to the Navier-Stokes
equations in Q. There exists a positive universal constant ε such that if
sup E(r) < ε,
0<r<1
then z = 0 is regular point of v, i.e., v is Hölder continuous in the closure of

the parabolic cylinder Q(%) with some positive % < r.
Let us start with the proof of auxiliary lemmata. The first statement is
actually the scaled version of a particular multiplicative inequality.
Lemma 2.8. For all 0 < r ≤ % ≤ 1,
h r 3 3 % 3 3 3
i
C(r) ≤ c A 2 (ρ) + A 4 (%)E 4 (%) . (6.2.34)
ρ r
Proof We have
Z Z Z
2 2 2
|v| dx = |v| − [|v| ],% dx + [|v|2 ],% dx ≤
B(r) B(r) B(r)
Z r 3 Z
2 2
≤ |v| − [|v| ],% dx + |v|2 dx.
%
B(%) B(%)
By the Poincaré-Sobolev inequality,

Z Z
2 2
|v| − [|v| ],% dx ≤ c% |∇ v| |v| dx,
B(%) B(%)
where c is an absolute positive constant. So, we get

R R 12 R 21 
|v|2 dx ≤ c% |∇ v|2 dx |v|2 dx + 



B(r)
B(%) B(%) 

3 R 

+ %r |v|2 dx ≤ (6.2.35)
B(%) 

3 1
R 21 3 

2 r
≤ c% A (%) |∇ v| dx

2 2 + %
%A(%). 


B(%)
Using the known multiplicative inequality, one can find

Z h Z 43 Z 34
3 2 2
|v| dx ≤ c |∇ v| dx |v| dx +
B(r) B(r) B(r)
1
Z 23 i
2
+ 3 |v| dx ≤ see (6.2.35) ≤
r2
B(r)
n 3 3 Z 34
≤ c % 4 A 4 (%) |∇ v|2 dx +
B(r)
1h 3 1
Z 21 r 3 i 32 o
+ 3 c% 2 A 2 (%) |∇ v|2 dx + %A(%) ≤
r2 %
B(%)
n r 3 Z 43 h 9
3
2 3 %4 i 3
o
≤c A (%) +
2 |∇ v| dx % +
4
3 A 4 (%) .
% r2
B(%)
Integrating the latter inequality in t on ]t0 − r2 , t0 [, we establish

Z n r 3 3
|v| dz ≤ c r2
3
A 2 (%) +
%
Q(r)
h 3 % 94 i 3 Zt0 Z 43 o
+ % 4 + 3 A 4 (%) dt |∇ v|2 dx ≤
r2 2 t0 −r B(x0 ,%)
n r 3 3 h 3 % 94 i 3 1
Z 34 o
2
≤c r A 2 (%) + % 4 + 3 A 4 (%)r 2 |∇ v|2 dz ≤
% r2
Q(%)
n r 3 3 h 3 % 94 i 3 1 3 3
o
2
≤c r A (%) + % + 3 A (%)r E (%) % .
2 4 4 2 4 4
% r2
It remains to notice that
9 h % 32
h 3 %4 i 1 3
% 3 i % 3
% + 4
3 r % =
2 4 + r2 ≤ 2 r2
r2 r r r
and then complete the proof of Lemma 2.8.
Lemma 2.9. For any 0 < R ≤ 1,

h 2 1 2
A(R/2) + E(R/2) ≤ c C 3 (R) + C 3 (R)D03 (R)
1 1 1
i
+A 2 (R)C 3 (R)E 2 (R) . (6.2.36)
Proof Picking up a suitable cut-off function in energy inequality (6.2.4),

we get the following estimates
n 1 Z
A(R/2) + E(R/2) ≤ c |v|2 dz +
R3
Q(R)
Z
1 2 2

+ 2 |v| − [|v| ] ,R |v| dz +

R

Q(R)
1
Z
3
23 Z 31 o
+ 2 |q − [q],R | 2 dz |v|3 dz .
R
Q(R) Q(R)
Since Z
1 2
|v|2 dz ≤ cC 3 (R),
R3
Q(z0 ,R)
we find
2
n 2 
1
A(R/2) + E(R/2) ≤ c C 3 (R) + C 3 (R) D03 (R)+ 

R 2 (6.2.37)
o
+ R12 |v| − [|v|2
] |v| dz .

,R 

Q(z0 ,R)
Application of Hölder inequality to the last term on the right-hand side of

(6.2.37) gives:
Z
2 2
S≡ |v| − [|v| ],R |v| dz ≤

Q(R)
Z0 Z 3 2 Z 13
2 2 2 3
≤ dt |v| − [|v| ],R dx |v|3 dx .
−R2 B(R) B(R)
By the Galiardo-Nirenberg inequality

Z 3 2 Z
2 2 2 3
|v| − [|v| ],R dx ≤ c |∇ v| |v| dx,
B(R) B(R)
we have
Z0 Z 12 Z 21 Z 31
2 2
S≤c dt |∇ v| dx |v| dx |v|3 dx ≤
−R2 B(R) B(R) B(R)
Z0 Z 21 Z 31
1 1
2 3
≤ c R A (R)
2 2 dt |∇ v| dx |v| dx ≤
−R2 B(R) B(R)
1 1
Z 13 Z0 Z 43 23
3 2
≤ c R A (R)
2 2 |v| dz dt |∇ v| dx ≤
Q(R) −R2 B(R)
1
Z 21
+ 32 1 1 1
≤ R 2 A (R)C (R)R
2 3 3 |∇ v|2 dz ≤
Q(R)
1 1 1
2
≤ c R A 2 (R)C 3 (R)E 2 (R).
Now, (6.2.36) follows from the latter relation and from (6.2.37). Lemma 2.9
is proved.
Now, our goal is to work out an estimate for the pressure.
Lemma 2.10. Let 0 < % ≤ 1. Then
h r 52 % 2 1 i
D0 (r) ≤ c D0 (%) + A 2 (%)E(%) (6.2.38)
% r
for all r ∈]0, %].
Proof We split the pressure in two parts
q = p1 + p2 (6.2.39)
in B(%) so that p1 is a unique solution to the variational identity

Z Z
p1 ∆ϕdx = − (τ − τ% ) : ∇2 ϕdx, (6.2.40)
B(%) B(%)
in which ϕ is an arbitrary test function of W32 (B(%)) satisfying the boundary

condition ϕ|∂B(%) = 0 and
τ := (v − c% ) ⊗ (v − c% ), τ% := [(v − c% ) ⊗ (v − c% )],% , c% := [v],% .

Here, time t is considered as a parameter. Clearly,

∆p2 = 0 (6.2.41)
in B(%).
We can easily find the bound for p1 (by a suitable choice of the test
function in (6.2.40))
Z Z
3 3
|p1 | 2 dx ≤ c |τ − τ% | 2 dx.
B(%) B(%)
The Galiardo-Nirenberg inequality

Z
3
Z 23
|p1 | 2 dx ≤ c |v − c% ||∇v|dx
B(%) B(%)
and Hölder inequality imply

Z
3
Z 43 Z 43
2
|p1 | 2 dx ≤ c |v − c% | dx |∇v|2 dx .
B(%) B(%) B(%)
On the other hand, Poincaré’s inequality

Z Z
2 2
|v − c% | dx ≤ c% |∇v|2 dx
B(%) B(%)
and the minimality property of c%

Z Z
2
|v − c% | dx ≤ |v|2 dx
B(%) B(%)
lead to the estimate

Z0 Z
1 3 1
|p1 | 2 dz ≤ cE(%)A 2 (%). (6.2.42)
%2
−%2 B(%)
By the mean value theorem for harmonic function p2 , we have for 0 <
r ≤ %/2
3 3 3
sup |p2 (x, t) − [p2 ],r (t)| 2 ≤ cr 2 sup |∇p2 (x, t)| 2
x∈B(r) x∈B(%/2)
r Z 32
≤c 4 |p2 (x, t) − [p2 ],% (t)|dx (6.2.43)
%
B(%)
c r 23
Z
3
≤ 3 |p2 (x, t) − [p2 ],% (t)| 2 dx.
% %
B(%)
Next, by (6.2.39) and (6.2.43),

Z Z
c 3 c 3
D0 (r) ≤ 2 |p1 − [p1 ],r | dz + 2
2 |p2 − [p2 ],r | 2 dz
r r
Q(r) Q(r)
Z0
c 1 r 32
Z Z
c 3
3 3
≤ 2 |p1 | dz + 2 3
2 r |p2 (x, t) − [p2 ],% (t)| 2 dx
r r % %
Q(r) −r2 B(%)
% 2 1
r 52 1 Z 3
≤c E(%)A 2 (%) + c |p 2 − [p 2 ],% | 2 dz
r % %2
Q(%)
% 2 1
r 52 h 1 Z 3
≤c E(%)A (%) + c
2
2
|q − [q],% | 2 dz
r % %
Q(%)
Z
1 3
i
+ 2 |p1 − [p1 ],% | dz 2
%
Q(%)
h r 25 % 2 1
i
≤c D0 (%) + E(%)A 2 (%)
% r
So, inequality (6.2.38) is shown. Lemma 2.10 is proved.
Proof of Theorem 2.7 It follows from (6.2.34), (6.2.38), and the as-
sumptions of Theorem 2.7 that:
h % 3 3 3
r 3 3 i
C(r) ≤ c A 4 (%)ε 4 + A 2 (%) (6.2.44)
r %
and h r 25 % 2 i
1
D0 (r) ≤ c D0 (%) + A (%)ε .
2 (6.2.45)
% r
Introducing the new quantity
3
E(r) = A 2 (r) + D02 (r),
we derive from local energy inequality (6.2.36) the following estimate

h 1 3 1 3
i
E(r) ≤ c C(2r) + C (2r)D0 (2r) + A (2r)C (2r)ε + D02 (r)
2 4 2 4
h 3 1 3
i
≤ c C(2r) + D02 (2r) + A 4 (2r)C 2 (2r)ε 4 . (6.2.46)
Now, let us assume that 0 < r ≤ %/2 < % ≤ 1. Replacing r with 2r in

(6.2.44) and (6.2.45), we can reduce (6.2.46) to the form
h % 3 3 3
r 3 3
E(r) ≤ c A (%)ε +
4 4 A 2 (%)
r %
r 5 % 4
+ D02 (%) + A(%)ε2
% r
3
% 3 3 3
r 3 3 12 3 i
+A 4 (2r) A 4 (%)ε 4 + A 2 (%) E04
r %
h r 3 3 r 5 r 23 3 3 3
% 34
≤c A 2 (%) + D02 (%) + A 4 (%)ε 4 A 4 (%)
% % % r
3 3
%2+4 3 3 % 4 % 3 3 i
3 3 3
+ A 4 + 8 (%)ε 4 + 8 + A(%)ε2 + A 4 (%)ε 4 .
r r r
Here, the obvious inequality A(2r) ≤ c%A(%)/r has been used. Applying
Young inequality with an arbitrary positive constant δ, we show that
r 34 3
E(r) ≤ c (ε 4 + 1)E(%) + cδE(%)
%
% 6 3
% 12 % 9 9
i
+c(δ) ε2 + ε6 + ε2 .
r r r
Therefore,
h r 43 3
i % 12 9 3
E(r) ≤ c (ε 4 + 1) + δ E(%) + c(δ) (ε6 + ε 2 + ε 2 ). (6.2.47)
% r
Inequality (6.2.47) holds for r ≤ %/2 and can be rewritten as follows.
h 3 3 i 9 3
E(ϑ%) ≤ c ϑ (ε + 1) + δ E(%) + c(δ)ϑ−12 (ε6 + ε 2 + ε 2 )
4 4 (6.2.48)
for any 0 < ϑ ≤ 1/2 and for any 0 < % ≤ 1.

Now, assuming that ε ≤ 1, let us fix ϑ and δ to provide fulfilment of the

conditions:
1 1
2cϑ 4 < 1/2, 0 < ϑ ≤ 1/2, cδ < ϑ 2 /2. (6.2.49)
Obviously, ϑ and δ are independent of ε. So,
1
E(ϑ%) ≤ ϑ 2 E(%) + G (6.2.50)
for any 0 < % ≤ 1, where G = G(ε) → 0 as ε → 0.

Iterations of (6.2.50) give us
k
E(ϑk %) ≤ ϑ 2 E(%) + cG
for any natural numbers k and for any 0 < % ≤ 1. Letting % = 1, we find
k
E(ϑk ) ≤ ϑ 2 E(1) + cG (6.2.51)
for the same values of k. It can be easily deduced from (6.2.51) that
1
E(r) ≤ c(r 2 E(1) + G(ε)) (6.2.52)
for all 0 < r ≤ 1/2. Now, (6.2.44) and (6.2.45) imply

h 3 3 3
i 1 1 1
C(r) + D0 (r) ≤ c A 4 (2r)ε 4 + A 2 (2r) + c(r 4 E 2 (1) + G 2 (ε))
h 3 3
i 1 1 1
≤ c A 2 (2r) + ε 2 + c(r 4 E 2 (1) + G 2 (ε))
h 1 3
i 1 1 1
≤ c c((2r) E(1) + G(ε)) + ε + c(r 4 E 2 (1) + G 2 (ε)).
2 2
Now we see that, for sufficiently small ε and sufficiently small r0 ,
C(r0 ) + D0 (r0 ) < ε0 ,
where ε0 is a number of Lemma 2.2. Since v and q − [q],r0 are a suitable weak
solution in Q(r0 ), Lemma 2.2 and the Navier-Stokes scaling yield required
statement. Theorem 2.7 is proved.
Now, we are in a position to speculate about ε-regularity theory. Quan-
tities that are invariant with respect to the Navier-Stokes scaling
v λ (y, s) = λv(x0 + λy, t0 + λ2 s),

q λ (y, s) = λ2 q(x0 + λy, t0 + λ2 s) (6.2.53)
play the crucial role in this theory. By the definition, such quantities are
defined on parabolic balls Q(r) and have the property
F (v, q; r) = F (v λ , q λ ; r/λ).
There are two types of statements in the ε-regularity theory for suitable
weak solutions to the Navier-Stokes equations and the first one reads:
Suppose that v and q are a suitable weak solution to the Navier-Stokes
equations in Q. There exist universal positive constants ε and {ck }∞ k=1 such
that if F (v, q; 1) < ε then |∇k v(0)| < ck , k = 0, 1, 2, .... Moreover, the
function z 7→ ∇k v(z) is Hölder continuous (relative to the parabolic metric)
with any exponent less 1/3 in the closure of Q(1/2).
An important example of such kind of quantities appears in Lemma 2.2
and is as follows:
Z
1 3

F (v, q; r) = 2 |v|3 + |q| 2 dz.
r
Q(r)
In the other type of statements, it is supposed that our quantity F is

independent of the pressure q:
Let v and q be a suitable weak solution in Q. There exists a universal
positive constant ε with the property: if sup0<r<1 F (v; r) < ε then z = 0 is
a regular point. Moreover, for any k = 0, 1, 2, ..., the function z 7→ ∇k v(z)
is Hölder continuous with any exponent less 1/3 in the closure of Q(r) for
some positive r.
Dependence on the pressure in the above statement is hidden. In fact,
the radius r is determined by the L 3 -norm of the pressure over the whole
2
parabolic cylinder Q.
To illustrate the second statement, let us consider several examples. In
the first one, we deal with the Ladyzhenskaya-Prodi-Serrin type quantities
Z0 Z sl
F (v; r) = Ms,l (v; r) = kvkls,l,Q(r) = |v|s dx dt
−r2 B(r)
provided
3 2
+ =1
s l
and s ≥ 3. Local regularity results connected with those quantities have been
proved partially by J. Serrin in [63] and then by M. Struwe in [65] for the
6.3. BOUNDED ANCIENT SOLUTIONS 169
velocity field v having finite energy even with no assumption on the pressure.
However, in such a case, we might loose Hölder continuity.
Energy scale-invariant quantities present an important example of the
second kind of quantities. Some of them have been listed above. For more
examples of scaled energy quantities, we refer to the paper [19]. It is worthy
to note that the second statement applied to the scaled dissipation E is the
famous Caffarelli-Kohn-Nirenberg theorem, which is Theorem 2.7. It gives
the best estimate for Hausdorff’s dimension of the singular set for a class of
weak Leray-Hopf solutions to the Cauchy problem. A certain generalization
of the Caffarelli-Kohn-Nirenber theorem itself has been proved in [50] and is
formulated as follows.
Proposition 2.11. Let v and q be a suitable weak solution to the Navier-
Stokes equations in Q. Given M > 0, there exists a positive number ε(M )
having the property: if two inequalities lim supr→0 E(r) < M and
lim inf E(r) < ε(M )
r→0
hold, then z = 0 is a regular point of v.

Typical examples of the third group of quantities invariant to the Navier-
Stokes scaling are:
G1 (v; r) = sup |x||v(z)|,
z=(x,t)∈Q(r)
√
G2 (v; r) = sup −t|v(z)|.
z=(x,t)∈Q(r)
A proof of the corresponding statements has been presented in [61], see

also [66], [26], and [6] for similar results.
6.3 Bounded Ancient Solutions

Definition 3.1. A bounded divergence free field u ∈ L∞ (Q− ; Rn ) is called
a weak bounded ancient solution (or simply bounded ancient solution) to the
Navier-Stokes equations if
Z
(u · ∂t w + u ⊗ u : ∇w + u · ∆w)dz = 0
Q−
∞
for any w ∈ C0,0 (Q− ).
Without loss of generality, we may assume that |u(z)| ≤ 1 a.e. in Q− .

If not, the function uλ (x, t) = λu(λx, λ2 t) with λ = 1/kuk∞,Q− will be a
bounded ancient solution satisfying the condition |uλ (z)| ≤ 1 a.e. in Q− .
Our aim is differentiability properties of an arbitrary bounded ancient
solution. Before stating and proving the main result, let us formulate several
auxiliary lemmata.
Lemma 3.2. For any F = L∞ (Rn ; Mn×n ), there exists a unique function
qF ∈ BM O(Rn ) such that [qF ]B(1) = 0 and
∆qF = −div divF = −Fij,ij in R3
in the sense of distributions. Moreover, the following estimate is valid
kqF kBM O(Rn ) ≤ c(n)kF k∞,Rn .
To state Lemma 3.2, the following notation has been used. [f ]Ω is the
mean value of a function f over a spatial domain Ω ∈ Rn . The mean value
of a function g over a space-time domain Q is denoted by (g)Q .
Lemma 3.3. Assume that functions f ∈ Lm (B(2)) and q ∈ Lm (B(2)) satisfy

the equation
∆q = −divf in B(2).
Then
Z Z Z
m m
|∇q| dx ≤ c(m, n) |f | dx + |q − [q]B(2) |m dx .
B(1) B(2) B(2)
Lemma 3.4. Assume that functions f ∈ Lm (Q(2)) and u ∈ Wm1,0 (Q(2))

satisfy the equation
∂t u − ∆u = f in Q(2).
Then u ∈ Wm2,1 (Q(1)) and the following estimate is valid:
h i
k∂t ukm,Q(1) + k∇2 ukm,Q(1) ≤ c(m, n) kf km,Q(2) + kukWm1,0 (Q(2)) .
Lemma 3.2 is proved with the help of the singular integral theory, see
[64]. Proof of Lemmata 3.3 and 3.4 can be found, for example, in [34] and
[33].
If we let
F (·, t) = u(·, t) ⊗ u(·, t),
then, by Lemma 3.2, there exists an unique function
pu⊗u ∈ L∞ (−∞, 0; BM O(Rn ))
which satisfies the condition [pu⊗u ]B(1) (t) = 0 and the equation
∆pu⊗u (·, t) = −div divF (·, t) in Rn
for all t ≤ 0.
To state the main result of this section, we introduce the space
Lm (Q− ) := { sup kf km,Q(z0 ,1) < ∞}.

z0 ∈Q−
Theorem 3.5. Let u be an arbitrary bounded ancient solution. For any

number m > 1,
|∇u| + |∇2 u| + |∇pu⊗u | ∈ Lm (Q− ).
Moreover, for each t0 ≤ 0, there exists a function bt0 ∈ L∞ (t0 − 1, t0 ) with
the following property
sup kbt0 kL∞ (t0 −1,t0 ) ≤ c(n) < +∞.

t0 ≤0
If we let ut0 (x, t) = u(x, t) + bt0 (t) in Qt0 = Rn ×]t0 − 1, t0 [, then, for any
number m > 1 and for any point x0 ∈ Rn , the uniform estimate
kut0 kWm2,1 (Q(z0 ,1)) ≤ c(m, n) < +∞, z0 = (x0 , t0 ),
is valid and, for a.a. z = (z, t) ∈ Qt0 , functions u and ut0 obey the system of
equations
∂t ut0 + divu ⊗ u − ∆u = −∇pu⊗u , divu = 0.
Remark 3.6. The first equation of the latter system can be rewritten in the
following way
∂t u + divu ⊗ u − ∆u = −∇pu⊗u − b0t0 , b0t0 (t) = dbt0 (t)/dt,
in Qt0 in the sense of distributions. So, the real pressure field in Qt0 is the
following distribution pu⊗u + b0t0 · x.
Remark 3.7. We can find a measurable vector-valued function b defined on

] − ∞, 0[ and having the following property. For any t0 ≤ 0, there exists a
constant vector ct0 such that
sup kb − ct0 kL∞ (t0 −1,t0 ) < +∞.

t0 ≤0
Moreover, the Navier-Stokes system takes the form
∂t u + divu ⊗ u − ∆u = −∇(pu⊗u + b0 · x), divu = 0
in Q− in the sense of distributions.

Remark 3.8. In most of our applications, we shall have some additional
global information about the pressure field, which will make it possible to
conclude that b0 = 0. For example, it is true if the pressure field belongs to
L∞ (−∞, 0; BM O(Rn )), i.e., u is a mild bounded ancient solution, see the
next section for details and definitions.
We can exclude the pressure field completely by considering the equation
for vorticity ω = ∇ ∧ u. Differentiability properties of ω are described by the
following theorem.
Theorem 3.9. Let u be an arbitrary bounded ancient solution. For any
m > 1, we have the following statements. If n = 2, then
ω = ∇⊥ u = u2,1 − u1,2 ∈ Wm
2,1
(Q− ) := {ω, ∇ω, ∇2 ω, ∂t ω ∈ Lm (Q− )}
and
∂t ω + u · ∇ω − ∆ω = 0 a.e. in Q− .
If n = 3, then
2,1
ω = ∇ ∧ u ∈ Wm (Q− ; R3 )
and
∂t ω + u · ∇ω − ∆ω = ω · ∇u a.e. in Q− .
Remark 3.10. We could analyse smoothness of solutions to the vorticity
equations further and it would be a good exercise. However, regularity results
stated in Theorem 3.9 are sufficient for our purposes.
Remark 3.11. By the embedding theorems, see [33], functions ω and ∇ω
are Hölder continuous in Q− and uniformly bounded there.
Proof of Theorem 3.5: Step 1.Energy estimate. Fix an arbitrary

number t0 < 0. Let kε (z) be a standard smoothing kernel. We use the
following notation for mollified functions:
Z
ε
F (z) = kε (z − z 0 )F (z 0 )dz 0 , F = u ⊗ u,
Q−
Z
ε
u (z) = kε (z − z 0 )u(z 0 )dz 0 .
Q−
◦
t0 t0
Assume that w ∈ C ∞ n
0 (Q− ), where Q− = R ×] − ∞, t0 [. For sufficiently
◦
small ε (0 < ε < ε(t0 )), wε belongs to C ∞ 0 (Q− ) as well. Then using well
known properties of smoothing kernel and Definition 3.1, we find
Z
◦
w · (∂t uε + divF ε − ∆uε )dz = 0, ∀w ∈ C ∞ t0
0 (Q− ).
Q−
It is easy to see that in our case there exists a smooth function pε with the
following property
∂t uε + divF ε − ∆uε = −∇pε , divuε = 0 (6.3.1)
in Qt−0 . Let us decompose pε so that
pε = pF ε + peε . (6.3.2)
It is not difficult to show that the function ∇pF ε is bounded in Qt−0 (exercise).
So, it follows from (6.3.1) and (6.3.2) that
∆peε = 0 in Qt−0 , ∇peε ∈ L∞ (Qt−0 ; Rn ).
By the Liouville theorem for harmonic functions, there exists a function

aε : [−∞, t0 [→ Rn such that
∇peε (x, t) = aε (t), x ∈ Rn , −∞ < t ≤ t0 .
So, we have
∂t uε + divF ε − ∆uε = −∇pF ε − aε , divuε = 0 (6.3.3)

in Qt−0 .
Now, let us introduce new functions
Zt
bεt0 (t) = aε (τ )dτ, t0 − 1 ≤ t ≤ t0 ,
t0 −1
vε (x, t) = uε (x, t) + bεt0 (t), z = (x, t) ∈ Qt0 .

Using the notion above, we may rewrite system (6.3.3) so that
∂t vε − ∆vε = −divF ε − ∇pF ε , divvε = 0 (6.3.4)
in Qt−0 .
Fix an arbitrary cut-off function ϕ so that
0 ≤ ϕ ≤ 1, ϕ ≡ 1 in B(1), suppϕ ⊂ B(2).
And then let ϕx0 (x) = ϕ(x − x0 ).

Now, we can derive the energy identity from (6.3.4), multiplying the latter
by ϕ2x0 vϕ and integrating the product by parts. As a result, we have
Z Zt Z
I(t) = ϕ2x0 (x)|vε (x, t)|2 dx +2 ϕ2x0 |∇vε |2 dxdt0 =
Rn t0 −1 Rn
Z Zt Z
= ϕ2x0 (x)|vε (x, t0 − 1)| dx + 2
∆ϕ2x0 |vε |2 dxdt0 +
Rn t0 −1 Rn
Zt Z
+ (pF ε − [pF ε ]B(x0 ,2) )vε · ∇ϕ2x0 dxdt0 +
t0 −1 Rn
Zt Z
+ (F ε − [F ε ]B(x0 ,2) ) : ·∇(ϕ2x0 vε )dxdt0 .
t0 −1 Rn
Introducing the quantity

Z
αε (t) = sup |vε (x, t)|2 dx
x0 ∈Rn
B(x0 ,1)
and taking into account that vε (·, t0 − 1) = uε (·, t0 − 1) and |uε (·, t0 − 1)| ≤ 1,
we can estimate the right hand side of the energy identity in the following
way
Zt
I(t) ≤ c(n) + c(n) αε (t0 )dt0 +
t0 −1
Zt0 Z 12 Z t 12
+c(n) |pF ε − [pF ε ]B(x0 ,2) | dxdt 2
αε (t0 )dt0 +
t0 −1 B(x0 ,2) t0 −1
Zt0 Z 21 Z t Z
+c(n) ε ε
|F − [F ]B(x0 ,2) | dxdt 2
ϕ2x0 |∇vε |2 dxdt0 + (6.3.5)
t0 −1 B(x0 ,2) t0 −1 Rn
Zt 21
+ αε (t0 )dt0 , t0 − 1 ≤ t ≤ t0 .
t0 −1
ε
Next, since |F | ≤ c(n), we find two estimates
Zt0 Z
|F ε − [F ε ]B(x0 ,2) |2 dxdt ≤ c(n)
t0 −1 B(x0 ,2)
and
Zt0 Z
|pF ε − [pF ε ]B(x0 ,2) |2 dxdt ≤ c(n)kpF ε k2L∞ (−∞,t0 ;BM O(Rn ))
t0 −1 B(x0 ,2)
≤ c(n)kF ε k2L t0 ≤ c(n).

∞ (Q− )
The latter estimates, together with (6.3.5), imply the inequalities

Zt
αε (t) ≤ c(n) 1 + αε (t0 )dt0 , t0 − 1 ≤ t ≤ t0
t0 −1
and
Zt0 Z Zt0
sup |∇vε |2 dxdt ≤ c(n) 1 + αε (t)dt .
x0 ∈Rn
t0 −1 B(x0 ,1) t0 −1
Usual arguments allows us to conclude that:

Zt0 Z
sup αε (t) + sup |∇uε |2 dxdt ≤ c(n). (6.3.6)
t0 −1≤t≤t0 x0 ∈Rn
t0 −1 B(x0 ,1)
It should be emphasised that the right hand size in (6.3.6) is independent of

t0 . In particular, estimate (6.3.6) gives:
sup bεt0 (t) ≤ c(n).

t0 −1≤t≤t0
Now, let us see what happens if ε → 0. Selecting a subsequence if neces-

sary and taking the limit as ε → 0, we have the following facts:
?
bεt0 * bt0 in L∞ (t0 − 1, t0 ; Rn );
the estimate
Zt0 Z
kbt0 kL∞ (t0 −1,t0 ) + sup |∇u|2 dxdt ≤ c(n) < +∞ (6.3.7)
x0 ∈Rn
t0 −1 B(x0 ,1)
is valid for all t0 < 0;

the system
∂t ut0 + divu ⊗ u − ∆u = −∇p − u ⊗ u, divu = 0
holds in Qt0 in the sense of distributions.

The case t0 = 0 can be treated by passing to the limit as t0 → 0.
Step 2, Bootstrap Arguments By (6.3.7),
f = divF = u · ∇u ∈ L2 (Q− ; Rn ).
Then Lemma 3.3 in combination with shifts shows that
∇pu⊗u ∈ L2 (Q− ; Rn ).
Next, obviously, the function ut0 satisfies the system of equations
∂t ut0 − ∆ut0 = −u · ∇u − ∇pu⊗u ∈ L2 (Q− ; Rn ).

Using the invariance with respect to shifts and Lemma 3.4, one can conclude
that
ut0 ∈ W22,1 (Q(z0 , τ2 ); Rn ), 1/2 < τ2 < τ1 = 1,
and, moreover, the estimate
kut0 kW 2,1 (Q(z0 ,τ2 )) ≤ c(n, τ2 )

2
holds for any z0 = (x0 , t0 ), where x0 ∈ Rn and t0 ≤ 0. According to the

embedding theorem, see [33], we can state that
∇ut0 = ∇u ∈ Wm1,02 (Q(z0 , τ2 ); Rn )
for
1 1 1
= − , m1 = 2.
m2 m1 n + 2
By lemma 3.3, by shifts, and by scaling, for 1/2 < τ30 < τ2 , we have the
following estimate
Z h Z i
m2 0
|∇pu⊗u (·, t)| dx ≤ c(n, τ2 , τ3 ) |∇u(·, t)|m2 dx + 1 .
B(x0 ,τ30 ) B(x0 ,τ30 )
In turn, Lemma 3.4 provides to two statements:
ut0 ∈ Wm2,12 (Q(z0 , τ3 ); Rn ), 1/2 < τ3 < τ30
and
kut0 kWm2,1 (Q(z0 ,τ3 )) ≤ c(n, τ3 , τ30 ).
2
Then, again, by the embedding theorem, we find
∇ut0 = ∇u ∈ Wm1,03 (Q(z0 , τ3 ); Rn )
for
1 1 1
= − .
m3 m2 n + 2
Now, let us take an arbitrary large number m > 2 and fix it. Find α as
an unique solution to the equation
1 1 α
= − .
m 2 n+2
Next, we let k0 = [α] + 1, where [α] is the entire part of the number α. And
then we determine the number mk0 +1 satisfying the identity
1 1 k0
= − .
mk0 +1 2 n+2
Obviously, mk0 +1 > m. Setting

1 1
τk+1 = τk − , τ1 = 1, k = 1, 2, , ,
4 2k
and repeating our previous arguments k0 times, we conclude that:
ut0 ∈ Wm2,1k +1 (Q(z0 , τk0 +1 ); Rn )

0
and
kut0 kWm2,1 (Q(z0 ,τk0 +1 )) ≤ c(n, m).
k0 +1
Thanks to the inequality τk > 1/2 for any natural numbers k, we complete
the proof of Theorem 3.5.
Proof of Theorem 3.9 Let us consider the case n = 3. The case n = 2
is in fact easier. So, we have
∂t ω − ∆ω = ω · ∇u − u · ∇ω ≡ f.
Take an arbitrary number m > 2 and fix it. By Theorem 3.5, the right hand
side has the following property
|f | ≤ c(n)(|∇2 u| + |∇u|2 ) ∈ Lm (Q(z0 , 2))
and the norm of f in Lm (Q(z0 , 2)) is dominated by a constant depending

only on m and being independent of z0 . It remains to apply Lemma 3.4 and
complete the proof of Theorem 3.9.
6.4 Mild Bounded Ancient Solutions

In this section, we assume that z = 0 is a singular point. Making use of
the space-time shift and the Navier-Stokes scaling, we can reduce the general
problem of local regularity to a particular one that in a sense mimics the first
time singularity.
6.4. MILD BOUNDED ANCIENT SOLUTIONS 179

Stokes equations in Q and z = 0 be a singular point of v. There exist two
functions ve and qe having the following properties:
(i) ve ∈ L3 (Q) and qe ∈ L 3 (Q) obey the Navier-Stokes equations in Q in
2
the sense of distributions;
(ii) ve ∈ L∞ (B×] − 1, −a2 [) for all a ∈]0, 1[;
(iii) there exists a number 0 < r1 < 1 such that ve ∈ L∞ ({(x, t) : r1 <
|x| < 1, −1 < t < 0}).
Moreover, functions ve and qe are obtained from v and q with the help of
the space-times shift and the Navier-Stokes scaling and the origin remains to
be a singular point of ve.
We recall z = 0 is a regular point of v if there exists a positive number

r such that v is Hölder continuous in the closure Q(r). A point z = 0 is a
singular point if it is not a regular one.
Proof Consider now an arbitrary suitable weak solution v and q in Q.
Let S ⊂ B×] − 1, 0] be a set of singular points of v. It is closed in Q. As
it was shown in [2], P 1 (S) = 0, where P 1 is the one-dimensional parabolic
Hausdorff measure. By assumptions, S 6= ∅. We can choose number R1
and R2 satisfying 0 < R2 < R1 < 1 such that S ∩ Q(R1 ) \ Q(R2 ) = ∅ and
S ∩ B(R2 )×] − R22 , 0] 6= ∅. We put
t0 = inf{t : (x, t) ∈ S ∩ B(R2 )×] − R22 , 0]}.
Clearly, (x0 , t0 ) ∈ S for some x0 ∈ B(R2 ). In a sense, t0 is the instant of time

when singularity of our suitable weak solution v and q appears in Q(R1 ).
Next, the one-dimensional Hausdorff measure of the set
St0 = {x∗ ∈ B(R2 ) : (x∗ , t0 ) is a singular point }
is zero as p well. Therefore, given x0 ∈ St0 , we can find sufficiently small

0 < r < R22 + t0 such that B(x0 , r) b B(R2 ) and ∂B(x0 , r) ∩ St0 = ∅.
Since the velocity field v is Hölder continuous at regular points, we can ensure
that all statements of Proposition 4.1 hold in the parabolic ball Q(z0 , r) with
z0 = (x0 , t0 ). We may shift and re-scale our solution if z0 6= 0 and r 6= 1.
In what follows, it is always deemed that such a replacement of v and q
with ve and qe has been already made. Coming back to the original notation, we
assume that functions v and q satisfy all the properties listed in Proposition
4.1 and z = 0 is a singular point of v.
One of the most powerful methods to study potential singularities is a

blowup technique based on the Navier-Stokes scaling
u(k) (y, s) = λk v(x, t), p(k) (y, s) = λ2k q(x, t)

with
x = x(k) + λk y, x = tk + λ2k s,
where x(k) ∈ R3 , −1 < tk ≤ 0, and λk > 0 are parameters of the scaling and
λk → 0 as k → +∞. It is supposed that functions v and q are extended by
zero to the whole R3 × R. A particular selection of scaling parameters x(k) ,
tk , and λk depends upon a problem under consideration.
Now, our goal is to describe a universal method that makes it possible to
reformulate the local regularity problem as a classical Liouville type problem
for the Navier-Stokes equations. To see how things work, let us introduce
the function
M (t) = sup kv(·, τ )k∞,B(r)
−1<τ ≤t
for some r ∈]r1 , 1[. It tends to infinity as time t goes to zero from the left
since the origin is a singular point of v. Thanks to the obvious properties of
the function M , one can choose parameters of the scaling in a particular way
letting λk = 1/Mk , where a sequence Mk is defined as
Mk = kv(, tk )k∞,B(r) = |v(x(k) , tk )|
with x(k) ∈ B(r1 ) for sufficiently large k. Before discussing what happens if
k tends to infinity, let us introduce a subclass of bounded ancient (backward)
solutions playing an important role in the regularity theory of the Navier-
Stokes equations.
Definition 4.2. A bounded vector field u, defined on R3 ×] − ∞, 0[, is called

a mild bounded ancient solution to the Navier-Stokes equation if there exists
a function p in L∞ (−∞, 0; BM O(R3 )) such that u and p satisfy the Navier-
Stokes system
∂t u + div u ⊗ u − ∆u + ∇p = 0,
div u = 0
in R3 ×] − ∞, 0[ in the sense of distributions.

The notion of mild bounded ancient solutions has been introduced in

[27]. It has been proved there that u has continuous derivatives of any order
in both spatial and time variables. Actually, the definition accepted in the
present paper is different but equivalent to the one given in [27]. Here, we
follow [59].
The statement below proved in [59] shows how mild bounded ancient
solutions occur in the regularity theory of the Navier-Stokes equations.
Proposition 4.3. There exist a subsequence of u(k) (still denoted by u(k) )
and a mild bounded ancient solution u such that, for any a > 0, the sequence
u(k) converges uniformly to u on the closure of the set Q(a) = B(a)×]−a2 , 0[.
The function u has the additional properties: |u| ≤ 1 in R3 ×] − ∞, 0[ and
|u(0)| = 1.
Proof of Proposition 4.3 Our solution v and q has good properties
inside Q1 = B1 ×] − 1, 0[ with B1 = {r1 < |x| < 1}. Let us list them. Let
Q2 = B2 ×] − τ22 , 0[, where 0 < τ2 < 1, B2 = {r1 < r2 < |x| < a2 < 1}. Then,
for any natural k,
z = (x, t) 7→ ∇k v(z) is Hölder continuous in Q2 ;
q ∈ L 3 (−τ22 , 0; C k (B 2 )).
2
The corresponding norms are estimated by constants depending on kvk3,Q ,

kqk 3 ,Q , kvk∞,Q1 , and numbers k, r1 , r2 , a2 , τ2 . In particular, we have
2
Z0
3
max |∇q(x, t)| 2 dt ≤ c1 < ∞. (6.4.1)
x∈B 2
−τ22
Proof of the first statement can be done by induction and found in [13], [31],
and [40]. The second statement follows directly from the first one and the
pressure equation: ∆q = −vi,j vj,i .
Now, let us decompose the pressure q = q1 + q2 . For q1 , we have
h i
∆q1 (x, t) = −div div χB (x)v(x, t) ⊗ v(x, t) , x ∈ R3 , −1 < τ < 0,
where χB (x) = 1 if x ∈ B and χB (x) = 0 if x ∈

/ B. Obviously, the estimate
Z0 Z Z
3
|q1 (x, t)| dxdt ≤ c
2 |v|3 dz
−1 R3 Q
holds and it is a starting point for local regularity of q1 . Using differentiability

properties of v, we can show
Z0
3
max |∇q1 (x, t)| 2 dt ≤ c2 < ∞, (6.4.2)
x∈B 3
−τ22
where B3 = {r2 < r3 < |x| < a3 < a2 }. From (6.4.1) and (6.4.2), it follows
that
Z0
3
max |∇q2 (x, t)| 2 dt ≤ c3 < ∞. (6.4.3)
x∈B 3
−τ22
However, q2 is a harmonic function in B, and thus, by the maximum principle,

we have
Z0
3
max |∇q2 (x, t)| 2 dt ≤ c3 < ∞, (6.4.4)
x∈B(r4 )
−τ22
where r4 = (r3 + a3 )/2.

Let us re-scale each part of the pressure separately, i.e.,
pki (y, s) = λ2k qi (x, t), i = 1, 2,
so that pk = pk1 + pk2 . As it follows from (6.4.4), for pk2 , we have
Z0 5
3
sup |∇y pk2 (y, s)| 2 ds ≤ c3 λk2 . (6.4.5)
y∈B(−xk /λk ,r4 /λk )
−(τ22 −tk )/λ2k
The first component of the pressure satisfies the equation
∆y pk1 (y, s) = −divy divy (χB(−xk /λk ,1/λk ) (y)u(k) (y, s) ⊗ u(k) (y, s)), y ∈ R3 ,
for all possible values of s. For such a function, we have the standard estimate
kpk1 (·, s)kBM O(R3 ) ≤ c (6.4.6)
for all s ∈]−(1−tk )/λ2k , 0[. It is valid since |u(k) | ≤ 1 in B(−xk /λk , 1/λk )×]−
(1 − tk )/λ2k , 0[.
We slightly change pk1 and pk2 setting
pk1 (y, s) = pk1 (y, s) − [pk1 ]B(1) (s) pk2 (y, s) = pk2 (y, s) − [pk2 ]B(1) (s)
so that [pk1 ]B(1) (s) = 0 and [pk2 ]B(1) (s) = 0.

Now, we pick up an arbitrary positive number a and fix it. Then from
(6.4.5) and (6.4.6) it follows that for sufficiently large k we have
Z Z
k 32 3
|p1 | de + |pk2 | 2 de ≤ c4 (c2 , c3 , a).
Q(a) Q(a)
Using the same bootstrap arguments, we can show that the following estimate
is valid:
ku(k) kC α (Q(a/2) ≤ c5 (c2 , c3 , c4 , a)
for some positive number α < 1/3. Indeed, the norm ku(k) kC α (Q(a/2)) is
estimated with the help of norms ku(k) kL∞ (Q(a))) and kpk kL 3 (Q(a)) , where
2
pk = pk1 + pk2 . Hence, using the diagonal Cantor procedure, we can select
subsequences such that for some positive α and for any positive a
u(k) → u in C α (Q(a)),
pk1 * p1 , in L 3 (Q(a)), [p1 ]B(1) (s) = 0,

2
pk2 * p2 in L 3 (Q(a)), [p2 ]B(1) (s) = 0.

2
So, |u| ≤ 1 in Q− and u and p = p1 + p2 satisfy the Navier-Stokes system

in Q− in the sense of distributions. Moreover, at it is follows from (6.4.6),
p1 ∈ L∞ (−∞, 0; BM O(R3 )).
Next, for sufficiently large k, we get from (6.4.5) that
Z 5
3
|∇pk2 (y, s)| 2 ds ≤ c3 λk2 .
Q(a)
Hence, ∇p2 = 0 in Q(a) for any a > 0. So, p2 (y, s) is identically zero. This
allows us to conclude that the pair u and p1 is a solution to the Navier-Stokes
equations in the sense of distributions and thus u is a nontrivial mild bounded
ancient solution satisfying the condition |u(0, 0)| = 1 and the estimate |u| ≤ 1
in Q− .
It is worthy to notice that the trivial bounded ancient solution of the

form
u(x, t) = c(t), p(x, t) = −c0 (t) · x,
with arbitrary bounded function c(t), is going to be a mild bounded ancient
solution if and only if c(t) ≡ constant. This allows us to make the following
plausible conjecture, see [59].
Conjecture Any mild bounded ancient solution is a constant.
To explain what consequences of the conjecture could be for regularity
theory of the Navier-Stokes equations, let us formulate a question which can
be raised in connection with the ε-regularity theory: what happens if we
drop the condition on smallness of scale-invariant quantities, assuming their
uniform boundedness only, i.e, sup0<r<1 F (v, r) < +∞. For Ladyzheskaya-
Prodi-Serrin type quantities with s > 3, the answer is still positive, i.e.,
z = 0 is a regular point. It follows from scale-invariance and the fact that
the assumption Ms,l (v; 1) = sup0<r<1 Ms,l (v; r) < +∞ implies Ms,l (v; r) → 0
as r → 0 if s > 3. Although in the marginal case s = 3 and l = +∞, the
answer remains positive, the known proof is more complicated and will be
outlined later.
Let us recall certain definitions and make some general remarks about
relationships between some scale-invariant quantities. Boundedness of
sup G2 (v; r) = G2 (v, 1) = G20 < +∞

0<r<1
can be rewritten in the form

G20
|v(z)| ≤ √
−t
for all z = (x, t) ∈ Q. If v satisfies the above inequality and z = 0 is still a

singular point of v, we say that a singularity of Type I or Type I blowup takes
place at t = 0. All other singularities are of Type II. The main feature of
Type I singularities is that they have the same rate as potential self-similar
solutions. The important properties connected with possible singularities of
Type I have been proved in [48], [61], and [59] and are as follows.
Proposition 4.4. Let functions v and q be a suitable weak solution to the

Navier-Stokes equations in Q.
6.5. LIOUVILLE TYPE THEOREMS 185
(i) If min{G1 (v; 1), G2 (v; 1)} < +∞, then
g = sup {A(v; r) + C(v; r) + D(q; r) + E(v; r)} < +∞.

0<r<1
(ii) If
g 0 = min{ sup A(v; r), sup C(v; r), sup E(v; r)} < +∞,
0<r<1 0<r<1 0<r<1
then g < +∞.
This proposition admits many obvious generalizations.

If we assume that v possesses uniformly bounded energy scale-invariant
quantities, then, by Proposition 4.3, the same type of quantities will be
bounded for the ancient solution, which is not trivial if z = 0 is a singular
point of v. However, by the conjecture, the above ancient solution must be
zero. So, the origin z = 0 cannot be a singular point of v. This would be a
positive answer to the question formulated above. In particular, according
to Proposition 4.4, validity of the conjecture would rule out Type I blowups.
6.5 Liouville Type Theorems

6.5.1 LPS Quantities
Theorem 5.1. Let u be a mild bounded ancient solution to the Navier-Stokes
equations, i.e., u ∈ L∞ (Q− ) is divergence free and satisfies the identity
Z
(u · ∂t w + u ⊗ u : ∇w + u · ∆w)dz = 0 (6.5.7)
Q−
for any divergence free function w from C0∞ (Q− ). Assume that
Z0 Z sl
sup Ms,l (u; r) = |u(x, t)|s dx <∞
0<r<∞
−∞ R3
with 3/s + 2/l = 1 and l < ∞. Then u ≡ 0 in Q− .

Proof Let us consider the simplest case of the regular LPS quantity
M5,5 . By the pressure equation, we may assume
Z0 Z
5
(|u|5 + |p| 2 )dxdt < +∞.
∞ R3
Given ε > 0, we can find T < 0 such that

ZT Z
5
(|u|5 + |p| 2 )dxdt < ε.
∞ R3
Then, by Hölder inequality, we have

Zt0 Z
1 3 3
(|u|3 + |p| 2 )dxdt < cε 5
R2
t0 −R2 B(x0 ,R)
for any x0 ∈ R3 , any R > 0, and any t0 ≤ T with some universal constant c.
In turn, the ε-regularity theory ensures the inequality
c
|u(x0 , t0 )| <
R
with another universal constant c. Tending R → ∞, we get u(·, t) = 0 as
t ≤ T . One can repeat more or less the same arguments in order to show
that in fact u is identically zero on R3 ×] − ∞, 0].
6.5.2 2D case
In two-dimensional case, we have the following Liouville type theorem.
Theorem 5.2. Assume that n = 2 and u is an arbitrary bounded ancient
solution. Then u(x, t) = b(t) for any x ∈ R2 .
To prove the above statement, we start with an auxiliary lemma.
Lemma 5.3. Let functions
2,1 2,1
ω ∈ Wm (Q− ) = {u ∈ Wm,loc (Q− ) : sup kukWm2,1 (Q(z0 ,1)) < ∞},
z0 ∈Q−
with m > 3, and u ∈ L∞ (Q− ) satisfy the equation
∂t ω + u · ∇ω − ∆ω = 0 in Q−
and the inequality

|u| ≤ 1 in Q− .
Then, for any positive numbers ε and R, there exists a point z0 = (x0 , t0 ),
x0 ∈ R2 and t0 ≤ 0, such that
ω(z) ≥ M − ε, z ∈ Q(z0 , R),
where M = sup ω(z).

z∈Q−
Remark 5.4. By the embedding theorem, M < +∞.

In order to prove Lemma 5.3, we need the strong maximum principle.
Here, it is.
Theorem 5.5. Strong maximum principle Let functions
w ∈ Wm2,1 (Q(z0 , R)) with m > n + 1 and a ∈ L∞ (Q(z0 , R); Rn ) satisfy the
equation
∂t w + a · ∇w − ∆w = 0 in Q(z0 , R).
Let, in addition,
w(z0 ) = sup w(z).
z∈Q(z0 ,R)
Then
w(z) = w(z0 ) in Q(z0 , R).
Proof of Lemma 5.3 ( [27]) In fact, we shall prove even a stronger
result. Let zk be a sequence of points in Q− such that
ω(zk ) → M.
We state that
inf ω(z) → M.
z∈Q(zk ,R)
Indeed, assume that this statement is false. Then, we can find a number
ε > 0 and a sequence of points zk0 ∈ Q(zk , R) such that
ω(zk0 ) ≤ M − ε.
Now, let us consider shifted functions
ω k (x, t) = ω(xk + x, tk + t), uk (x, t) = u(xk + x, tk + t),

2,1
for z = (x, t) ∈ Q(R). By the definition of the space Wm (Q− ), these new
functions are subject to the estimates
kω k kWm2,1 (Q(R)) ≤ c,
|uk | ≤ 1 in Q(R),
with a constant c that is independent of k. Moreover, we have
∂t ω k + uk · ∇ω k − ∆ω k = 0 in Q(R),
ω k (zk00 ) ≤ M − ε, zk00 ∈ Q(R),

ω k (z) ≤ M z ∈ Q(R).
Using standard compactness arguments, we show
ωk * ω in Wm2,1 (Q(R)),
?
uk *u in L∞ (Q(R); R2 ),
ωk → ω in C(Q(R)),
ω(z) ≤ ω(0) = M z ∈ Q(R), (6.5.8)
ω(z∗ ) ≤ M − ε, (6.5.9)
where z∗ ∈ Q(R). Clearly, ω ∈ Wm2,1 (Q(R)) and
∂t ω + u · ∇ω − ∆ω = 0 in Q(R).
By (6.5.8) and by the strong maximum principle,
ω(z) = M z ∈ Q(R),
which is in a contradiction with (6.5.9).

Proof of Theorem 5.2 We are going to apply Lemma 5.3 to the
vorticity equation. Let us show first that
sup ω(z) = M ≤ 0.
z∈Q−
To this end, assume that the latter statement is wrong and in fact
M > 0.
Take a cut-off function ϕ ∈ C0∞ (B(R)) with the following properties:
0 ≤ ϕ ≤ 1, |∇ϕ| ≤ c/R in B(R),
ϕ≡1 in B(R/2).
By Lemma 5.3, for an arbitrary number R > 0, there exists a point z0R =
(x0R , t0R ) with t0R ≤ 0 such that
|ω(z)| ≥ M − M/2 = M/2 > 0, z ∈ Q(z0R , R).
If we let ϕx0R (x) = ϕ(x − x0R ), then

Z
M 2 M
A(R) = ϕx0R (x)ω(z)dz ≥ R |B(x0R , R)| = πR4 . (6.5.10)
2 2
Q(z0R ,R)
On the other hand, since ω = u2,1 − u1,2 , we have after integration by parts
Z
A(R) = (ϕx0R ,2 u1 − ϕx0R ,1 u2 )dz ≤
Q(z0R ,R)
≤ cR3 ,
where c is a universal constant. The latter inequality contradicts (6.5.10) for

sufficiently large R. So, M ≤ 0. In the same way, one can show that m ≥ 0,
where
−∞ < m = inf ω(z).
z∈Q−
So, ω ≡ 0 in Q− . Since u(·, t) is a divergence free function in R2 , we can state

that u(·, t) is a bounded harmonic function in R2 . Therefore, u(x, t) = b(t),
x ∈ R2 . Theorem 5.2 is proved.
6.5.3 Axially symmetric case with no swirl

In the case of axial symmetry, it is convenient to introduce the cylindrical
coordinates %, ϕ, x3 so that x1 = % cos ϕ, x2 = % sin ϕ, x3 = x3 . The velocity
components are going to be u% , uϕ , u3 . By the definition of axial symmetry,
u%,ϕ = ∂u% /∂ϕ = 0, uϕ,ϕ = 0, u3,ϕ = 0, p,ϕ = 0.
For the vorticity components, we have simple formulae
ω% = uϕ,3 , ωϕ = u%,3 − u3,% , ω3 = uϕ,% + uϕ /%.
Now, assume that vector field u is an arbitrary axially symmetric bounded

ancient solution with zero swirl, i.e., uϕ = 0. This, in particular, leads to the
representation
1
∇ω = − ωϕ e% ⊗ eϕ + ωϕ,% eϕ ⊗ e% + ωϕ,3 eϕ ⊗ e3 .
%
We know that |∇ω| is a bounded function, which implies boundedness of
functions ωϕ,% , ωϕ,3 , and %1 ωϕ . Regarding ∇2 ω, we can state
Z0 Z h i m2
2 2 2 2 2
|ωφ,%% | + 2|ωϕ,%ϕ | + |ωϕ,33 | + 2|(ωϕ /%),% | + 2|(ωϕ /%),3 | %d%dϕ ≤
−T C(a)
≤ c(a, T, p) < +∞,

where a > 0, for any T > 0, and for any m > 1. Here, we recall that:
C(a) = {x = (x0 , x3 ) ∈ R3 : |x0 | < a, |x3 | < a}
and x0 = (x1 , x2 ) so that |x0 | = %.

Theorem 5.6. Let u be an arbitrary axially symmetric bounded ancient so-
lution with zero swirl. Then u(x, t) = b(t) for any x ∈ R3 and for any t ≤ 0.
Moreover, u1 (x, t) = 0 and u2 (x, t) = 0 for the same x and t or, equivalently,
u% (%, x3 , t) = 0 for any % > 0, for any x3 ∈ R, and for any t ≤ 0.
Proof We let η = ωϕ /%. It is not difficult to verify that η satisfies the
equation
2
∂t η + u% η,% + u3 η,3 − (∆η + η,% ) = 0, % > 0, −∞ < x3 < +∞, t < 0,
%
where
1 1
∆η = (%η,% ),% + η,33 = η,%% + η,33 + η,% .
% %
Let us make the change of variables
y = (y 0 , y5 ) ∈ R5 , y 0 = (y1 , y2 , y3 , y4 ),
q
% = |y 0 | = y12 + y22 + y32 + y42 , y5 = x3 .
Then after simple calculations, we see that a new function
f (y, t) = f (y1 , y2 , y3 , y4 , y5 , t) = η(%, ϕ, t)
obeys the equation

∂t f + U · ∇5 f − ∆5 f = 0 (6.5.11)
in Q5− = R5 ×] − ∞, 0[. Here, ∇5 and ∆5 are usual gradient and Laplacian
with respect to the Cartesian coordinates in R5 and
U (y, t) = (U1 (y, t), U2 (y, t), U3 (y, t), U4 (y, t), U5 (y, t)),
where
u% (%, x3 , t)
Ui (y, t) = yi , i = 1, 2, 3, 4, U5 (y, t) = u3 (%, x3 , ).
%
Obviously, the function U is bounded in Q5− . However, previous arguments
show that ∇5 U is a bounded function as well. Indeed,we have
|∇5 U (y, t)| ≤ c(|∇u(x, t)| + |u% (%, x3 , t)|/%) ≤ c|∇u(x, t)| ≤ c < +∞
for any y ∈ R5 and any t ≤. So,
|U |, |∇5 U | ∈ L∞ (Q5− ).
For bounded f , weak solution to (6.5.11) can be defined as follows

Z h i
f ∂t g + f U · ∇5 g + f g div5 U + f ∆5 g dy dt = 0
Q5−
for any g ∈ C0∞ (Q5− ). In the way, explained in the previous section, one can
show that, for any m > 1,
2,1
f ∈ Wm (Q5− )
and the norm can be dominated by a positive constant, depending on m,

supQ5− |f |, and supQ5− (|U | + |∇5 U |) only.
We let
M = sup sup f (y, t) = sup sup η(|x0 |, x3 , t).
y∈R5 t≤0 x∈R3 t≤0
We wish to show that M ≤ 0. Assume that it is not so, i.e., M > 0, and we
can apply Lemma 5.3 in our five-dimensional setting. Then, for any R > 0,
there exists a point yR in R5 and a moment of time tR ≤ 0 such that
f (y, t) ≥ M/2, (y, t) ∈ Q((yR , tR ), R) = B(yR , R)×]tR − R2 , tR [,
where B(yR , R) = {|y − yR | < R}.
By our assumptions,
0 < M0 = sup ωϕ (|x0 |, x3 , t) < +∞.
x∈R3 , t≤0
We may choose a number R so big that

2M0
R > 100
M
and then let
2M0
y∗ = (y∗0 , y5R ), y∗0 = 50 l + yR0 ,
M
where
l ∈ R4 , |l| = 1, (l, yR0 ) = l1 y1R + l2 y2R + l3 y3R + l4 y4R = 0.
It is not difficult to check that y∗ ∈ B(yR , R) and, moreover,
2M0
|y∗0 | ≥ 50 .
M
Then we find
M M0 M0 M
≤ f (y∗ , tR − R2 /2) ≤ 0 < 2M0 = .
2 |y∗ | 50 M 100
This means that in fact M ≤ 0. In the same way, one can show that m ≥ 0
and then conclude that f ≡ 0 in Q5− , which in turn implies
ωϕ (|x0 |, x3 , t) = 0, ∀(x, t) ∈ Q− ,
and therefore
ω≡0 in Q− .
The rest of the proof is the same as in Theorem 5.2.
6.5.4 Axially Symmetric Case

We are going to prove the following statement.
Theorem 5.7. Let u be an arbitrary axially symmetric bounded ancient so-
lution satisfying assumption
A
|u(x, t)| ≤ , x = (x0 , x3 ) ∈ R3 , −∞ < t ≤ 0, (6.5.12)
|x0 |
where A is a positive constant independent of x and t. Then u ≡ 0 in Q− .
Proof Let us explain our strategy. First, we are going to show that,
under condition (6.5.12), the swirl is zero, i.e., uϕ = 0. Then we apply
Theorem 5.6 and state that u(x, t) = b(t). But condition (6.5.12) says b(t) =
0 for all t ≤ 0. So, our aim now is to show that uϕ ≡ 0 in Q− .
Let us introduce the additional notation:
e = R+ × R,
R R+ = {% ∈ R, % > 0}, Q e − ∞, 0[,
e− = R×]
Π(%1 , %2 ; h1 , h2 ) = {%1 < % < %2 , h1 < x3 < h2 },

Q(%
e 1 , %2 ; h1 , h2 ; t1 , t2 ) = Π(%1 , %2 ; h1 , h2 )×]t1 , t2 [.
Now, our aim is to show that
M = sup %uϕ ≤ 0.
Q
e−
Assume that it is false, i.e., M > 0, and let

g = %uϕ /M.
The new scaled function g satisfies the equation
∂t g + u% g,% + u3 g,3 − (∆g − 2g,% /%) = 0 in Q
e− .
By the assumptions,
q
sup g = 1, u2% + u23 ≤ A/%, |g| ≤ A/M in Q
e− (6.5.13)
Q
e−
and
(%u% ),% + (%u3 ),3 = 0 in Q
e− .
To formulate the lemma below, we abbreviate
Π = Π(%1 , %2 ; h1 , h2 ), e = Π×]t − t1 , t[.
Q
Lemma 5.8. For any ε > 0, there exists a positive number
δ = δ(Π, t, t1 , A, M, ε) ≤ ε
such that if
sup g(x, t) > 1 − δ,
x∈Π
then
inf g(z) > 1 − ε.
z∈Q
e
Proof If we assume that the statement of the lemma is false, then there
must exist a number ε0 > 0 such that, for any natural k, one can find
sequences with the following properties:
δk > δk+1 , δk → 0, sup g k (x, t) > 1 − δk , inf g k (z) ≤ 1 − ε0 , (6.5.14)

x∈Π z∈Q
e
functions uk and g k satisfy the equations
(%uk% ),% + (%uk3 ),3 = 0, ∂t g k + uk% g,%

k
+ uk3 g,3
k
− (∆g k − 2g,%
k
/%) = 0
in Q
e− and the relations
q
sup g k = 1, |uk% |2 + |uk3 |2 ≤ A/%, |g k | ≤ A/M in Q
e− .
Q
e−
By (6.5.14), there are points (%k , xk3 , t), with (%k , xk3 ) ∈ Π, and (%0k , x0k3 , t0k ) ∈
Q
e such that
g k (%k , xk3 , t) > 1 − 2δk , g k (%0k , x0k3 , t0k ) ≤ 1 − ε0 /2. (6.5.15)
Weak form of the equations for uk and g k is as follows:

Z h i
g k ∂t f + g k (uk% f,% + uk3 f,3 ) + g k (∆f + 2f,% /%) %d%dx3 dt = 0
Q
e−
for any f ∈ C0∞ (Q

e− ). Routine arguments show
?
uk * u e− ; R2 ),
in L∞ (Q
gk * g in Wm2,1 (Q
e2 ),
e2 = Π2 ×]t − t2 , t[, Π2 = Π(%2 , %2 ; h2 , h2 ) c Π, t2 > t1 , and m >> 1.
where Q 1 2 1 2
Then we have
gk → g in C(Q e2 ) (6.5.16)
and
q
sup g ≤ 1, |u% |2 + |u3 |2 ≤ A/%, |g| ≤ A/M in Q
e2 , (6.5.17)
Q
e2
and
∂t g + (u% + 1/%)g ,% + u3 g ,3 − g ,%% − g ,33 = 0 in Q
e2 .
According to (6.5.15) and (6.5.16),
g(%0 , x03 , t) = 1, g(%00 , x003 , t00 ) ≤ 1 − ε0 /2, t00 ≤ t, (6.5.18)
where
(%k , xk3 , t) → (%0 , x03 , t), (%0k , x0k3 , t0k ) → (%00 , x003 , t00 )
and points (%0 , x03 , t) and (%00 , x003 , t00 ) belong to the closure of the set Q.
e
Clearly, by (6.5.17),
g(%0 , x03 , t) = sup g(z) = 1.

z∈Q
e2
By the strong maximum principle, g ≡ 1 in Q

e2 . But this contradicts (6.5.18).

Now, we proceed with the proof of Theorem 5.7. Take arbitrary positive
numbers R, L, T , and 0 < ε ≤ 1/2. We can always assume that
1−ε≤g ≤1 e0 = Π0 ×] − T, 0[,
on Q (6.5.19)
where Π0 = Π(1, R; −L, L). To explain this, we let δ∗ = δ(Π0 , 0, −T, A, M, ε).
Obviously, there exists a point (%0 , x03 , t0 ) ∈ Q
e− such that
1 − g(%0 , x03 , t0 ) < δ∗ ≤ ε ≤ 1/2.
It is easy to see
1/2 ≤ %0 uϕ (%0 , x03 , t0 )/M = g(%0 , x03 , t0 ) ≤ %0 /M

and, therefore, %0 > M/2 > 0. Then one can scale our functions so that
gλ (r, y3 , s) = g(λr, x03 + λy3 , t0 + λ2 s), λ = %0 ,
uλr (r, y3 , s) = λu% (λr, x03 + λy3 , t0 + λ2 s),

uλ3 (r, y3 , s) = λu3 (λr, x03 + λy3 , t0 + λ2 s).
For scaled functions, we have
1 − gλ (1, 0, 0) < δ∗ ,
∂s gλ + uλr gλ,r + uλ3 gλ,3 − (∆gλ − 2gλ,r /r) = 0 in Q

e− ,
(ruλr ),r + (ruλ3 ),3 = 0 in Q

e− ,
q
sup gλ = 1, |uλr |2 + |uλ3 |2 ≤ A/r, |gλ | ≤ A/M in Q
e− .
Q
e−
By Lemma 5.8,
1 − ε ≤ gλ ≤ 1 on Q
e0 .
It is always deemed that this operation has been already made and script λ
is dropped. It is important to note two things. Numbers R, L, T , and ε are
in our hands and we cannot use the fact |u| ≤ 1 any more since after scaling
|u| ≤ %0 (R, T, L, A, M, ε).
We choose a cut-off function
Φ(%, x3 , t) = ψ(%)η(x3 )χ(t),
where functions ψ, η, and χ have the following properties:
ψ(%) = 1 0 ≤ % ≤ R − 1, ψ(%) = 0 % ≥ R,
|ψ 0 (%)| + |ψ 00 (%)| ≤ c 0 ≤ % < +∞;

η(x3 ) = 1 |x3 | ≤ L − 1, η(x3 ) = 0 |x3 | ≥ L,
|η 0 (x3 )| + |η 00 (x3 )| ≤ c |x3 | < +∞;
χ(t) = 1 − T + 1 < t ≤ −1, χ(t) = 0 t < −T,
χ(t) = t + T − T ≤ t ≤ −T + 1,
χ(t) = −t − 1 < t ≤ 0.
So, we have
Z
I0 = ∂t g + u% g,% + u3 g,3 − ∆g Φ%d%dx3 dt = I00 =
Q
e−
Z
g,%
= −2 Φ%d%dx3 dt (6.5.20)
%
Q
e−
We replace g with g − 1 in the left hand side of (6.5.20) and, after integration
by parts, have
Z
1
I0 = ∂t Φ + u% Φ,% + u3 Φ,3 + ∆Φ (1 − g)dxdt.
4π
Q−
We know that 1 − g ≤ ε in Q
e0 . Then, by 6.5.13,
Z0 ZL Z1
I0 ≥ (1 − g) ∂t Φ + u3 Φ,3 + Φ,33 %d%dx3 dt + εC0 (R, T, L, A, M )
−T −L 0
≥ −(L + T )C1 (A, M ) + εC0 (R, T, L, A, M ). (6.5.21)

Next, let us evaluate the right hand side in (6.5.20). Integration by parts
gives:
Z0 ZL Z0 ZL ZR
I00 = −2 g(0, x3 , t)Φ(0, x3 , t)dx3 dt + 2 Φ,% gd%dx3 dt.
−T −L −T −L 0
The first term on the right hand side of the latter identity is equal to zero.
The second one is estimated from above
Z0 ZL ZR Z0 ZL ZR
I00 =2 Φ,% d%dx3 dt + 2 Φ,% (g − 1)d%dx3 dt ≤
−T −L 0 −T −L 0
Z0 ZL ZR
≤2 Φ,% d%dx3 dt + εC00 (R, T, L, A, M ) =
−T −L 0
Z0 ZL
= −2 Φ(0, x3 , t)dx3 dt + εC00 (R, T, L, A, M ) <
−T −L
< −2(L − 1)(T − 2) + εC00 (R, T, L, A, M ).

The latter, together with identity (6.5.20) and (6.5.21), implies the following
inequality
2(L − 1)(T − 2) ≤ (L + T )C1 (A, M ) + εC000 (R, T, L, A, M ).
This leads to contradiction for large L and T and sufficiently small ε.

So, the assumption M > 0 is wrong. In the same way, one shows that
inf %uϕ = m ≥ 0.
Q
e−
This means that the swirl is zero.
6.6 Axially Symmetric Suitable Weak

Solutions
In this section, just for convenience, we replace balls B(r) with cylinders
C(r) = {x = (x0 , x3 ), x0 = (x1 , x2 ), |x0 | < r, |x3 | < r} and then Q(r) =
C(r)×] − r2 , 0[. As usual, let us set
v = v% e% + v3 e3 vb = vϕ eϕ
for v = v% e% + vϕ eϕ + v3 e3 .
Here, we follow paper [59], where results are stated for the canonical
domain Q = Q(1). The general case can be deduced by re-scaling.
Theorem 6.1. Assume that functions v ∈ L3 (Q) and q ∈ L 3 (Q) are an

2
axially symmetric weak solution to the Navier-Stokes equations in Q. Let, in
addition, a positive constant C exists such that
C
|v(x, t)| ≤ √ (6.6.1)
−t
for almost all points z = (x, t) ∈ Q. Then z = 0 is a regular point of v.

6.6. AXIALLY SYMMETRIC SUITABLE WEAK SOLUTIONS 199
Theorem 6.2. Assume that functions v ∈ L3 (Q) and q ∈ L 3 (Q) are an

2
addition,
v ∈ L∞ (C×] − 1, −a2 [) (6.6.2)
for each 0 < a < 1 and
C
|v(x, t)| ≤ (6.6.3)
|x0 |
for almost all points z = (x, t) ∈ Q with some positive constant C. Then
z = 0 is a regular point of v.
According to the Caffarelli-Kohn-Nirenberg theorem if v and q are an

axially symmetric suitable weak solution and z = (x, t) is singular (i.e., not
regular) point of v, then there must be x0 = 0. In other words, all singular
points must seat on the axis of symmetry, which in our case is the axis x3 .
The following estimate is obtained with help of Mozer’s iterations. Its
proof is not complicated, see, for example, [59].
Lemma 6.3. Assume that functions v ∈ L3 (Q) and q ∈ L 3 (Q) are an

2
addition, condition (6.6.2) hold. Then following estimate is valid:
Z 10
103
ess sup |%vϕ (z)| ≤ C(M ) |%vϕ | 3 dz , (6.6.4)
z∈Q(1/2)
Q(3/4)
where Z 103
10
M= |v| dz
3 + 1.
Q(3/4)
Remark 6.4. Under the assumptions of Lemma 6.3, the pair v and q is a
suitable weak solution to the Navier-Stokes equations in Q. Hence, the right
hand side of (6.6.4) is bounded from above.
With some additional notation
C(x0 , R) = {x ∈ R3 k x = (x0 , x3 ), x0 = (x1 , x2 ),
|x0 − x00 | < R, |x3 − x03 | < R}, C(R) = C(0, R), C = C(1);
Q(z0 , R) = C(x0 , R)×]t0 − R2 , t0 [, Q(R) = Q(0, R), Q = Q(1),
we recall the definition of certain scaled energy quantities:

Z
1
A(z0 , r; v) = ess sup |v(x, t)|2 dx,
2
t0 −r <t<t0 r
C(x0 ,r)
Z Z
1 2 1 3
E(z0 , r; v) = |∇v| dz, D(z0 , r; q) = 2 |q| 2 dz,
r r
Q(z0 ,r) Q(z0 ,r)
Z Z
1 3 1
C(z0 , r; v) = 2 |v| dz, H(z0 , r; v) = 3 |v|2 dz,
r r
Q(z0 ,r) Q(z0 ,r)
Zt0 Z sl
1 s
Ms,l (z0 , r; v) = κ |v| dx dt,
r
t0 −r2 C(x0 ,r)
where κ = l( 3s + 2l − 1) and s ≥ 1, l ≥ 1.
The following statement is proven in a similar way as Proposition 4.4, see
details in [59].
Lemma 6.5. Under assumptions of Theorem 6.1, we have the estimate
A(zb , r; v) + E(zb , r; v) + C(zb , r; v) + D(zb , r; q) ≤ C1 < +∞ (6.6.5)
for all zb and for all r satisfying conditions
1 1
zb = (be3 , 0), b ∈ R,
|b| ≤ , 0<r< . (6.6.6)
4 4
A constant C1 depends only on the constant C in (6.6.1), kvkL3 (Q) , and
kqkL 3 (Q) .
2
To prove Theorem 6.2, we need an analogue of Lemma 6.5. Here, it is.

Lemma 6.6. Under assumptions of Theorem 6.2, estimate (6.6.5) is valid as
well with constant C1 depending only on the constant C in (6.6.3), kvkL3 (Q) ,
and kqkL 3 (Q) .
2
Lemma 6.6 is proved along the same lines as Lemma 6.5.

As it follows from conditions of Theorem 6.2 and the statement of Lemma
6.3, the module of the velocity field grows not faster than C/|x0 | as |x0 | → 0.
Moreover, the corresponding estimate is uniform in time. However, it turns
out that the same is true under conditions of Theorem 6.1. More precisely,
we have the following.
Proposition 6.7. Assume that all conditions of Theorem 6.1 hold. Then
C1
|v(x, t)| ≤ (6.6.7)
|x0 |
for all z = (x, t) ∈ Q(1/8). A constant C1 depends only on the constant C

in (6.6.1), kvkL3 (Q) , and kqkL 3 (Q) .
2
Proof In view of (6.5), we can argue essentially as in [62].

Let us fix a point x0 ∈ C(1/8) and put r0 = |x00 |, b0 = x03 . So, we have
r0 < 81 and |b0 | < 18 . Further, we introduce the following cylinders:
Pr10 = {r0 < |x0 | < 2r0 , |x3 | < r0 }, Pr20 = {r0 /4 < |x0 | < 3r0 , |x3 | < 2r0 }.
Pr10 (b0 ) = Pr10 + b0 e3 , Pr20 (b0 ) = Pr20 + b0 e3 ,

Q1r0 (b0 ) = Pr10 (b0 )×] − r02 , 0[, Q2r0 (b0 ) = Pr20 (b0 )×] − (2r0 )2 , 0[.
Now, let us scale our functions so that
x = r0 y + b0 e3 , t = r02 s, u(y, s) = r0 v(x, t), p(y, s) = r02 q(x, t).
As it was shown in [62], there exists a continuous nondecreasing function

Φ: R+ → R+ , R+ = {s > 0}, such that
Z
sup |u(u, s)| + |∇u(y, s)| ≤ Φ sup |u(y, s)|2 dy
(y,s)∈Q11 (0) −22 <s<0
P12 (0)
Z Z Z
3
2 3
+ |∇u| dy ds + |u| dy ds + |p| 2 dy ds . (6.6.8)
Q21 (0) Q21 (0) Q21 (0)
After making inverse scaling in (6.6.8), we find

2
sup r0 |u(x, t)| + r0 |∇u(x, t)| ≤ Φ cA(zb0 , 3r0 ; v) + cE(zb0 , 3r0 ; v)+
z∈Q1r0 (b0 )

+cC(zb0 , 3r0 ; v) + cD(zb0 , 3r0 ; q) ≤ Φ 4cC1 .
It remains to apply Lemma 6.5 and complete the proof of the proposition.
Now, we proceed with proof of Theorems 6.1 and 6.2. Using Lemmata 6.3,
6.5, 6.6, Remark 6.4, Proposition 6.7 and scaling arguments, we may assume
(without loss of generality) that our solution v and q have the following
properties:

sup A(0, r; v) + E(0, r; v) + C(0, r; v) + D(0, r; q) = A1 < +∞, (6.6.9)
0<r≤1
ess sup |x0 ||v(x, t)| = A2 < +∞. (6.6.10)

z=(x,t)∈Q
We may also assume that the function v is Hölder continuous in the closure
of the set C×] − 1, −a2 [ for any 0 < a < 1.
Introducing functions
H(t) = sup |v(x, t)|, h(t) = sup H(τ ),

x∈C −1<τ ≤t
let us suppose that our statement is wrong, i.e., z = 0 is a singular point.

Then there are sequences xk ∈ C and −1 < tk < 0, having the following
properties:
h(tk ) = H(tk ) = Mk = |v(xk , tk )| → +∞ as k → +∞.
We scale our functions v and q so that scaled functions keeps axial symmetry:
1
uk (y, s) = λk v(λk y 0 , x3k + λk y3 , tk + λ2k s), λk = ,
Mk
pk (y, s) = λ2k q(λk y 0 , x3k + λk y3 , tk + λ2k s).

These functions satisfy the Navier-Stokes equations in Q(Mk ). Moreover,
|uk (yk0 , 0, 0)| = 1, yk0 = Mk x0k . (6.6.11)
According to (6.6.10),
|yk0 | ≤ A2
for all k ∈ N. Thus, without loss of generality, we may assume that
yk0 → y∗0 as k → +∞. (6.6.12)
Now, let us see what happens as k → +∞. By the identity
sup |uk (e)| = 1 (6.6.13)

e=(y,s)∈C(Mk )
and by (6.6.9), we can select subsequences (still denote as the entire sequence)
such that
?
uk * u in L∞ (Q(a)), (6.6.14)
and
pk * p in L 3 (Q(a))
2
for any a > 0. Functions u and p are defined on Q− = R3 ×] − ∞, 0[.

Obviously, they possess the following properties:
ess sup |u(e)| ≤ 1,

e∈Q−

sup A(0, r; u) + E(0, r; u) + C(0, r; u) + D(0, r; p) ≤ A1 ,
0<r<+∞
ess sup |y 0 ||u(y, s)| ≤ A2 . (6.6.15)

e=(y,s)∈Q−
Now, our aim is to show that u and p satisfy the Navier-Stokes equations
Q− and u is smooth enough to obey the identity
|u(y∗0 , 0, 0)| = 1. (6.6.16)
To this end, we fix an arbitrary positive number a > 0 and consider numbers
k so big that a < Mk /4. We know that uk satisfies the nonhomogeneous heat
equation of the form
∂t uk − ∆uk = −div F k in Q(4a),
where F k = uk ⊗ uk + pk I and
kF k k 3 ,Q(4a) ≤ c1 (a) < ∞.

2
This is implies the following fact, see [33],
k∇uk k 3 ,Q(3a) ≤ c2 (a) < ∞.

2
Now, we can interpret the pair uk and pk as a solution to the nonhomogeneous

Stokes system
∂t uk − ∆uk + ∇pk = f k , div uk = 0 in Q(3a), (6.6.17)

where f k = −uk · ∇uk is the right hand side having the property
kf k k 3 ,Q(3a) ≤ c2 (a).
2
Then, according to the local regularity theory for the Stokes system, see
Chapter IV, we can state that
k∂t uk k 3 ,Q(2a) + k∇2 uk k 3 ,Q(2a) + k∇pk k 3 ,Q(2a) ≤ c3 (a).

2 2 2
The latter, together with the embedding theorem, implies
k∇uk k3, 3 ,(Q(2a)) + kpk k3, 3 ,Q(2a) ≤ c4 (a).

2 2
In turn, this improves integrability of the right hand side in (6.6.17)
kf k k3, 3 ,Q(2a) ≤ c4 (a).

2
Therefore, by the local regularity theory,
k∂t uk k3, 3 ,Q(2a) + k∇2 uk k3, 3 ,Q(2a) + k∇pk k3, 3 ,Q(2) ≤ c5 (a).
2 2 2
Applying the imbedding theorem once more, we find
k∇uk k6, 3 ,Q(2a) + kpk k6, 3 ,Q(2a) ≤ c6 (a).

2 2
The local regularity theory leads then to the estimate
k∂t uk k6, 3 ,Q(a) + k∇2 uk k6, 3 ,Q(a) + k∇pk k6, 3 ,Q(a) ≤ c7 (a).
2 2 2
By the embedding theorem, sequence uk is uniformly bounded in the parabo-

1
lic Hölder space C 2 (Q(a/2)). Hence, without loss of generality, one may
assume that
1
uk → u in C 4 (Q(a/2)).
This means that the pair u and p obeys the Navier-Stokes system and (6.6.16)
holds. So, the function u is the so-called bounded ancient solution to the
Navier-Stokes system which is, in addition, axially symmetric and satisfies
the decay estimate (6.6.15). As it was shown in Section 5 of this chapter, such
a solution must be identically zero. But this contradicts (6.2.43). Theorems
(6.1) and (6.2) are proved.
6.7. BACKWARD UNIQUENESS 205
6.7 Backward Uniqueness for Navier-Stokes

Equations
In this section, we deal with another subclass of ancient solutions u possessing
the following property: there exists a function p defined on R3 ×] − ∞, 0[
such that functions u and p are a suitable weak solution to the Navier-Stokes
equations in R3 ×] − ∞, 0[, i.e., they are a suitable weak solution on each
parabolic ball of the form Q(a) = B(a)×] − a2 , 0[ with < a < +∞. We call
u a local energy ancient solution. Certainly, mild bounded ancient solutions
belong to this subclass.
Local energy ancient solutions can be obtained from a given suitable weak
solution v and q defined in Q with the help of the scaling mentioned in the
previous section provided boundedness of g 0 takes place, see the definition of
g 0 in Proposition 4.4.

Stokes equations in Q with g 0 < +∞ and let u(k) (y, s) = λk v(λk y, λ2k s) and
p(k) (y, s) = λ2k q(λk y, λ2k s) with λk → 0 as k → +∞. Then there exist sub-
sequences of u(k) and p(k) still denoted by u(k) and p(k) such that, for each
a > 0,
u(k) → u
in L3 (Q(a)) ∩ C([−a2 , 0]; L 9 (B(a))) and
8
p(k) * p
in L 3 (Q(a)), where u is a local energy ancient solution with the corresponding

2
pressure p. For them, the scaled energy quantities are uniformly bounded, i.e.,
sup {A(u; a) + C(u; a) + D(p; a) + E(u; a)} < +∞.

0<a<+∞
Moreover, if z = 0 is a singular point of the velocity field v, then

Z
|u|3 dz > c (6.7.1)
Q(3/4)
with a positive universal constant c, i.e., u is not identically equal to zero.

A proof of this proposition and similar facts can be found in [13], [54],
[59], and [55]. Let us comment the last statement of Proposition 7.1. Indeed,
if z = 0 is a singular point of v, the ε-regularity theory gives us
Z
1 3
2
(|v|3 + |q| 2 )dz > ε > 0
r
Q(r)
for all 0 < r < 1 and for some universal constant ε. Making the inverse
change of variables, we find
1
R 3
a2
(|u(k) |3 + |p(k) | 2 )dyds =
Q(a)
1
R 3
λ2k a2
(|v|3 + |q| 2 )dxds > ε > 0
Q(λk a)
for each fixed radius a > 0 and for sufficiently large natural number k. We
cannot simply pass to the limit in the latter identity since it is not clear
whether the pressure p(k) converges strongly. This is quite typical issue when
one works with sequences of weak solutions to the Navier-Stokes equations.
In order to treat this case, let us split the pressure p(k) into two parts. The
first part is completely controlled by the velocity field u(k) while the second
one is a harmonic function with respect to the spatial variables. This, to-
gether with a certain boundedness of the sequence p(k) , implies (6.7.1). For
more details, we recommend papers [54] and [55].
We do not know whether local energy ancient solutions with bounded
scaled energy quantities are identically equal to zero. However, there are
some interesting cases for which the answer is positive. Let us describe them.
Our additional standing assumption of this section can be interpreted as
a restriction on the blowup profile of v and has the form
Z
1 9
15 |v(x, 0)| 8 dx → 0 (6.7.2)
r8
B(r)
as r → 0. The most important consequence of (6.7.2) is that
u(·, 0) = 0, (6.7.3)
where u is a local energy ancient solution that is generated by the scaling

described by Proposition 7.1. Indeed, for any a > 0, we have

1
R 9
15 |u(y, 0)| 8 dy ≤
a 8
B(a)
1
R 9 1
R 9
c 15 |u(y, 0) − u(k) (y, 0)| 8 dy + c 15 |u(k) (y, 0)| 8 dy =
a8 a8
B(a) B(a)
1
R 9
αk (a) + c 15 |v(x, 0)| 8 dx.
(λk a) 8
B(λk a)
Now, by Proposition 7.1 and by (6.7.2), the right hand side of the latter
inequality tends to zero and this completes the proof of (6.7.3).
In a view of (6.7.3), one could expect that our local energy ancient solution
is identically equal to zero. We call this phenomenon a backward uniqueness
for the Navier-Stokes equations. So, if the backward uniqueness takes place
or at least our ancient solution is zero on the time interval ] − 3/4, 0[, then
(6.7.1) cannot be true and thus, by Proposition 7.1, the origin z = 0 is not a
singular point of the velocity field v.
The crucial point for understanding the backward uniqueness for the
Navier-Stokes equations is a similar phenomenon for the heat operator with
lower order terms. The corresponding statement for the partial differential
inequality involving the backward heat operator with lower order terms has
been proved in [13] and reads:
Theorem 7.2. Assume that we are given a function ω defined on Rn+ ×]0, 1[,
where Rn+ = {x = (xi ) ∈ Rn , xn > 0}. Suppose further that they have the
properties:
ω and the generalized derivatives ∇ω, ∂t ω, and ∇2 ω are square integrable
over any bounded subdomain of Rn+ ×]0, 1[;
|∂t ω + ∆ω| ≤ c(|ω| + |∇ω|) (6.7.4)
on Rn+ ×]0, 1[ with a positive constant c;
|ω(x, t)| ≤ exp{M |x|2 } (6.7.5)
for all x ∈ Rn+ , for all 0 < t < 1, and for some M > 0;
ω(x, 0) = 0 (6.7.6)
for all x ∈ Rn+ .

Then ω is identically zero in Rn+ ×]0, 1[.
The interesting feature of Theorem 7.2 is that there has been made no
assumption on ω on the boundary xn = 0. In order to prove the theorem,
two Carleman’s inequalities have been established, see details in [13] and
[12] and the appendix. For the further improvements of the above backward
uniqueness result, we refer to the interesting paper [10].
Theorem 7.2 clearly indicates what one should add to (6.7.3) in order
to get the backward uniqueness for ancient solutions to the Navier-Stokes
equations. Obviously, we need more regularity for sufficiently large x and
a right decay at infinity. One can hope then to apply Theorem 7.2 to the
vorticity equation
∂t ω − ∆ω = ω · ∇u − u · ∇ω, ω = ∇ ∧ u,
which could be interpreted as a perturbation of the heat equation by lower

order terms. To make it possible, it is sufficient to show boundedness of u
and ∇u outside of the Cartesian product of some spatial ball and some time
interval. The rest of the section will be devoted to a certain situation, for
which it is really true.
Let us assume that
|u(x, t)| + |∇u(x, t)| ≤ c < +∞ (6.7.7)
for all |x| > R, for all −1 < t < 0, and for some constant c and try to
figure out what follows from (6.7.7). It is not difficult to see that (6.7.3) and
(6.7.7) implies (6.7.6) and (6.7.4), (6.7.5), respectively. At last, the linear
theory ensures the validity of first condition in Theorem 7.2, see details in
[47]. So, Theorem 7.2 is applicable and by it, ω(x, t) = 0 for all |x| > R and
for −1 < t < 0. Using unique continuation across spatial boundaries, see,
for instance, [13], we deduce ω(x, t) = ∇ ∧ u(x, t) = 0 for all x ∈ R3 and,
say, for −5/6 < t < 0. Since u is divergence free, it is a harmonic functionp in
3
R depending on t ∈] − 5/6, 0[ as a parameter. Therefore, for any a > 5/6
and for any x0 ∈ R3 , by the mean value theorem for harmonic functions, we
have Z
2 1
sup |u(x0 , t)| ≤ c sup 3
|u(x, t)|2 dx
−5/6<t<0 −5/6<t<0 a
B(x0 ,a)
a + |x0 |
Z
1
≤c sup 3
|u(x, t)|2 dx ≤ c A(u, a + |x0 |).
−5/6<t<0 a a3
B(|x0 |+a)
Thanks to boundedness of scaled energy quantities stated in Proposition 7.1,

the right hand side of the latter inequality tends to zero as a goes to infinity.
By arbitrariness of x0 , we conclude that u(x, t) = 0 for all x ∈ R3 and for
−5/6 < t < 0, which contradicts (6.7.1). Hence, the origin z = 0 cannot be
a singular point of v.
Coming back to a marginal case of Ladyzhenskaya-Prodi-Serrin condition,
which is called L3,∞ -case, and show that it can be completely embedded into
the above scheme. So, we assume that functions v and q are a suitable
weak solution to the Navier-Stokes equations in Q and satisfy the additional
condition
kvk3,∞,Q < +∞. (6.7.8)
With the help of Proposition 4.4, it is not so difficult to show that g 0 < +∞.
So, for v, all the assumptions of Proposition 7.1 hold and thus our blowup
procedure produces a local energy ancient solution u with the properties
listed in that proposition. Exploited the ε-regularity theory once more, we
can show further that v(·, 0) ∈ L3 (B(2/3)), which in turn implies (6.7.2).
Now, in order to prove regularity of the velocity v at the point z = 0, it is
sufficient to verify the validity of (6.7.7). Indeed, by scale-invariance,
kuk3,∞,R3 ×]−∞,0[ < +∞.
Applying Proposition 7.1 once again and taking into account properties of
harmonic functions, one can conclude that
kpk 3 ,∞,R3 ×]−∞,0[ < +∞.

2
Combining the latter estimates, we show that for any T > 0

Z0 Z
3
(|u|3 + |p| 2 )dxdt < +∞. (6.7.9)
−T R3
Our further arguments rely upon the ε-regularity theory. Indeed, letting,
say, T = 4, one can find R > 4 so that
Z0 Z
3
(|u|3 + |p| 2 )dxdt < ε.
−4 R3 \B(R/2)
The rest of the proof of (6.7.7) is easy.

6.8 Comments
The section is essentially the context of my lectures on the local regularity
theory given in Summer School, Cetraro, Italy, 2010, see [58]. It contains
an introduction to the so-called ε-regularity theory in the spirit of the paper
[13], see also [50] for some generalizations. A big part of this section is
an alternative approach to derivation of mild bounded ancient solutions and
Liouville type theorems for them presented in [27]. Here, we follow the paper
[59] although proofs of Liouville type theorems is essentially the same as in
[27].
Chapter 7
Behaviour of L3-Norm
7.1 Main Result

Let us consider the Cauchy problem for the classical Navier-Stokes system
∂t v + v · ∇v − ∆v = −∇q, divv = 0 (7.1.1)
with the initial condition
v|t=0 = v0 (7.1.2)
in R3 . For simplicity, assume
∞
v0 ∈ C0,0 (R3 ) ≡ {v ∈ C0∞ (R3 ) : divv = 0}. (7.1.3)
In 1934, J. Leray proved certain necessary conditions for T to be a blow
up time. They can be stated as follows. Suppose that T is a blow up time,
then, for any 3 < m ≤ ∞, there exists a constant cm , depending on m only,
such that
Z m1 cm
m
kv(·, t)km ≡ kv(·, t)km,R3 ≡ |v(x, t)| dx ≥ m−3 (7.1.4)
3
(T − t) 2m
R
for all 0 < t < T .

However, for the scale-invariant L3 -norm, a weaker statement
lim sup kv(·, t)k3 = ∞ (7.1.5)
t→T −0
has been proven in the previous chapter. The aim of this chapter is to
improve (7.1.5). At the moment, the best improvement of (7.1.5) is given by
the following theorem.
211
212 CHAPTER 7. BEHAVIOUR OF L3 -NORM
Theorem 1.1. Let v be an energy solution to the Cauchy problem (7.1.1)

and (7.1.2) with the initial data satisfying (7.1.3). Suppose that T > 0 is a
finite blow up time. Then
lim kv(·, t)k3 = ∞ (7.1.6)

t→T −0
holds true.
Let us briefly outline our proof of Theorem 1.1 that relays upon ideas
developed in [54]-[56]. In particular, in [54], a certain type of scaling has
been invented, which, after passing to the limit, gives a special non-trivial
solution to the Navier-Stokes equations provided there is a finite time blow
up. In [55] and [56], it has been shown that the same type of scaling and
blowing-up can produce the so-called Lemarie-Rieusset local energy solutions,
introduced and carefully studied in the monograph [35], see Appendix B for
details. It turns out to be that the backward uniqueness technique is still
applicable to those solutions. Although the theory of backward uniqueness
itself is relatively well understood, its realization is not an easy task and based
on delicate regularity results for the Navier-Stokes equations. Actually, there
are two main points to verify: solutions, produced by scaling and blowing-up,
vanish at the last moment of time and have a certain spatial decay. The first
property is easy when working with L3 -norm while the second one is harder.
However, under certain restrictions, the required decay is a consequence of the
Lemarie-Rieusset theory. So, the main technical part of the whole procedure
is to show that scaling and blowing-up lead to local energy solutions. On that
way, a lack of compactness of initial data of scaled solutions in L2,loc is the
main obstruction. This is why the same theorem for a stronger scale-invariant
1
norm of the space H 2 is easier. The reason for that is a compactness of the
corresponding embedding, see [41] and [55].
In this chapter, we are going to show that, despite of a lack of compactness
in L3 -case, the limit of the sequence of scaled solutions is still a local energy
solution, for which a spatial decay takes place. Technically, this can be done
by splitting each scaled solution into two parts. The first one is a solution
to a non-linear problem but with zero initial data while the second one is a
solution of a linear problem with weakly converging nonhomogeneous initial
data.
We also prove (7.1.4) as a by-product of the proof of Theorem 1.1, see
Section 4.
7.2. ESTIMATES OF SCALED SOLUTIONS 213
7.2 Estimates of Scaled Solutions

Assume that our statement is false and there exists an increasing sequence
tk converging to T as k → ∞ such that
sup kv(·, tk )k3 = M < ∞. (7.2.1)

k∈N
By the definition of a blow up time for energy solutions, there exists at

least one singular point at time T . Without loss of generality, we may assume
that it is (0, T ). Moreover, the blow-up profile has the finite L3 -norm, i.e.,
kv(·, T )k3 < ∞. (7.2.2)

Let us scale v and q so that
u(k) (y, s) = λk v(x, t), p(k) (y, s) = λ2k q(x, t), (7.2.3)
for (y, s) ∈ R3 ×] − λ−2

k T, 0[, where
x = λk y, t = T + λ2k s,
r
T − tk
λk =
S
and a positive parameter S < 10 will be defined later.
By the scale invariance of L3 -norm, u(k) (·, −S) is uniformly bounded in
L3 (R3 ), i.e.,
sup ku(k) (·, −S)k3 = M < ∞. (7.2.4)
k∈N
Let us decompose our scaled solution u(k) into two parts: u(k) = v (k) +w(k) .
Here, w(k) is a solution to the Cauchy problem for the Stokes system:
∂t w(k) − ∆w(k) = −∇r(k) , div w(k) = 0 in R3 ×] − S, 0[,

w(k) (·, −S) = u(k) (·, −S). (7.2.5)
Obviously, (7.2.5) can be reduced to the Cauchy problem for the heat equa-
tion so that the pressure r(k) = 0 and w(k) can be worked out with the help
of the heat potential. The estimate below is well-known, see, for example
[24],
sup{kw(k) kL5 (R3 ×]−S,0[ + kw(k) kL3,∞ (R3 ×]−S,0[ } ≤ c(M ) < ∞. (7.2.6)
k
It is worthy to note that, by the scale invariance, c(M ) in (7.2.6) is indepen-

dent of S.
As to v (k) , it is a solution to the Cauchy problem for the following per-
turbed Navier-Stokes system
∂t v (k) + div(v (k) + w(k) ) ⊗ (v (k) + w(k) ) − ∆v (k) = −∇p(k) ,

div v (k) = 0 in R3 ×] − S, 0[, (7.2.7)
v (k) (·, −S) = 0.
Now, our aim is to show that, for a suitable choice of −S, we can prove
unform estimates of v (k) and p(k) in certain spaces, pass to the limit as k → ∞,
and conclude that the limit functions u and p are a local energy solution to the
Cauchy problem for the Navier-Stokes system in R3 ×] − S, 0[ associated with
the initial data, generated by the weak L3 -limit of the sequence u(k) (·, −S).
Let us start with estimates of solution to (7.2.7). First of all, we know
the formula for the pressure:
Z
(k) 1 (k) 2 1
p (x, t) = − |u (x, t)| + K(x − y) : u(k) (y, t) ⊗ u(k) (y, t)dy, (7.2.8)
3 4π
R3
where K(x) = ∇2 (1/|x|).

Next, we may decompose the pressure in the same way as it has been
done in [25], see Appendix B. For x0 ∈ R3 and for x ∈ B(x0 , 3/2), we let
p(k) (k) (k) 1(k) 2(k)

x0 (x, t) ≡ p (x, t) − cx0 (t) = px0 (x, t) + px0 (x, t), (7.2.9)
where
Z
1 1
px1(k)
0
(x, t) = − |u(k) (x, t)|2 + K(x − y) : u(k) (y, t) ⊗ u(k) (y, t)dy,
3 4π
B(x0 ,2)
Z
1
p2(k)
x0 (x, t) = (K(x − y) − K(x0 − y)) : u(k) (y, t) ⊗ u(k) (y, t)dy,
4π
R3 \B(x0 ,2)
Z
1
c(k)
x0 (t) = K(x0 − y) : u(k) (y, t) ⊗ u(k) (y, t)dy.
4π
R3 \B(x0 ,2)
Using the similar arguments as in [35], see also Appendix B, one can
1(k) 2(k)
derive estimates of px0 and px0 . Here, they are:
kp1(k)
x0 (·, t)kL 3 (B(x0 ,3/2)) ≤ c(M )(kv
(k)
(·, t)k2L3 (B(x0 ,2)) + 1), (7.2.10)
2
sup |px2(k)
0
(x, t)| ≤ c(M )(kv (k) (·, t)k2L2,unif + 1), (7.2.11)
B(x0 ,3/2)
where
kgkL2,unif = sup kgkL2 (B(x0 ,1)) .
x0 ∈R3
We further let
α(s) = α(s; k, S) = kv (k) (·, s)k22,unif ,
Zs Z
β(s) = β(s; k, S) = sup |∇v (k) |2 dydτ.
x∈R3
−S B(x,1)
From (7.2.10), (7.2.11), we find the estimate of the scaled pressure
h Z0 i
3
δ(0) ≤ c(M ) γ(0) + (1 + α 2 (s))ds , (7.2.12)
−S
with some positive constant c(M ) independent of k and S. Here, γ and δ are
defined as
Zs Z
γ(s) = γ(s; k, S) = sup |v (k) (y, τ )|3 dydτ
x∈R3
−S B(x,1)
and
Zs Z
3
δ(s) = δ(s; k, S) = sup |p(k) (y, τ ) − cx(k) (τ )| 2 dy dτ,
x∈R3
−S B(x,3/2)
respectively. It is known that an upper bound for γ can be given by the

known multiplicative inequality
Zs 14 Zs 43
3
γ(s) ≤ c α (τ )dτ β(s) + α(τ )dτ . (7.2.13)
−S −S
Fix x0 ∈ R3 and a smooth non-negative function ϕ such that
ϕ = 1 in B(1), spt ⊂ B(3/2)
and let ϕx0 (x) = ϕ(x − x0 ).

Since the function v (k) is smooth on [−S, 0[, we may write down the
following energy identity
Z Zs Z
ϕ2x0 (x)|v (k) (x, s)|2 dx +2 ϕ2x0 |∇v (k) |2 dxdτ =
R3 −S R3
Zs Z h i
= |v (k) |2 ∆ϕ2x0 + v (k) · ∇ϕ2x0 (|v (k) |2 + 2px(k)
0
) dxdτ +
−S R3
Zs Z h
+ w(k) · ∇ϕ2x0 |v (k) |2 + 2ϕ2x0 w(k) ⊗ (w(k) + v (k) ) : ∇v (k) +
−S R3
i
+2w(k) · v (k) (w(k) + v (k) ) · ∇ϕ2x0 dxdτ = I1 + I2 .
The first term I1 is estimated with the help of the Hölder inequality,
multiplicative inequality (7.2.13), and bounds (7.2.10), (7.2.11). So, we find
h Zs 3
I1 ≤ c(M ) (1 + α(τ ) + α 2 (τ ))dτ +
−S
Zs 14 Zs 34 i
3
+ α (τ )dτ β(s) + α(τ )dτ .
−S −S
Now, let us evaluate the second term

Zs
I2 ≤ c kv (k) (·, τ )k2L3 (B(x0 ,3/2)) kw(k) (·, τ )kL3 (B(x0 ,3/2)) dτ +
−S
Zs Z 51 Z 54
5 5
(k) 5
+c |w | dx |v (k) | 4 |∇v (k) | 4 dx dτ +
−S B(x0 ,3/2) B(x0 ,3/2)
Zs
1
Z 12
+cβ (s) 2 |w(k) |4 dxdτ dτ +
−S B(x0 ,3/2)
Zs
+c kv (k) (·, τ )kL3 (B(x0 ,3/2)) kw(k) (·, τ )k2L3 (B(x0 ,3/2)) dτ.
−S
Taking into account (7.2.6) and applying Hölder inequality several times, we
find
2 1
I2 ≤ c(M )γ 3 (s)(s + S) 3 +
Zs Z 15 Z 12
(k) 5 (k) 2
+c |w | dx |∇v | dx ×
−S B(x0 ,3/2) B(x0 ,3/2)
Z
10
103 1 1
(k)
× |v | dx
3 dτ + c(M )β 2 (s)(s + S) 10 +
B(x0 ,3/2)
1 2
+c(M )γ 3 (s)(s + S) 3 .
It remains to use another known multiplicative inequality
Z 10
103 Z 51
(k) (k) 2
|v (x, s)| dx
3 ≤c |v (x, s)| dx ×
B(x0 ,3/2) B(x0 ,3/2)
Z 103
× (|∇v (k) (x, s)|2 + |v (k) (x, s)|2 dx
B(x0 ,3/2)
and to conclude that

2 1 1 1 1 2
I2 ≤ c(M )γ 3 (s)(s + S) 3 + c(M )β 2 (s)(s + S) 10 + c(M )γ 3 (s)(s + S) 3 +
Zs 54 Zs 51
(k) 5
+c β(s) + α(τ )dτ ) × α(τ )kw (·, τ )kL5,unif dτ .
−S −S
Finally, we find h 1
α(s) + β(s) ≤ c(M ) (s + S) 5 +
Zs i
+ α(τ )(1 + kw(k) (·, τ )k5L5,unif ) + α3 (τ ) dτ , (7.2.14)
−S
which is valid for any s ∈ [−S, 0[ and for some positive constant c(M ) inde-
pendent of k, s, and S.
It is not so difficult to show that there is a positive constant S(M ) such
that
1
α(s) ≤ (7.2.15)
10
for any s ∈] − S(M ), 0[. In turn, the latter will also imply that
1
α(s) ≤ c(M )(s + S) 5 (7.2.16)
for any s ∈] − S(M ), 0[.

To see how this can be worked out, let us assume
α(s) ≤ 1 (7.2.17)
for −S ≤ s < s0 ≤ 0. Then (7.2.14) yields

1
α(s) ≤ c(M )((s + S) 5 + y(s)) (7.2.18)
for the same s. Here,

Zs
y(s) = α(τ )(2 + g(τ ))dτ, g(s) = kw(k) (·, s)k2L5 (R3 ) .
−S
The function y(s) obeys the differential inequality

1
y 0 (s) ≤ c(M )(2 + g(s))((s + S) 5 + y(s)) (7.2.19)
for −S ≤ s < s0 ≤ 0. After integrating (7.2.19), we find

Zs n Zs o
1
y(s) ≤ c(M ) (τ +S) (2+g(τ )) exp c(M ) (2 + g(ϑ)) dϑ dτ (7.2.20)
5
−S τ
for −S ≤ s < s0 ≤ 0. Taking into account estimate (7.2.6), we derive from

(7.2.20) the following bound
1
y(s) ≤ c1 (M )(s + S) 5 (7.2.21)
7.3. LIMITING PROCEDURE 219
for −S ≤ s < s0 ≤ 0 and thus

1
y(s) ≤ c1 (M )S 5 (7.2.22)
for the same s.

Now, let us pick up S(M ) > 0 so small that
1 1
c(M )(1 + c1 (M ))S 5 (M ) = . (7.2.23)
20
We claim that, for such a choice of S(M), statement (7.2.15) holds true.
Indeed, assume that it is false. Then since α(s) is a continuous function on
1
[−S, 0[ and α(0) = 0, there exists s0 ∈] − S, 0[ such that 0 ≤ α(s) < 10 for
1
all s ∈] − S, s0 [ and α(s0 ) = 10 . In this case, we may use first (7.2.22) and
then (7.2.18), (7.2.23) to get
1 1
α(s) ≤ c(M )(1 + c1 (M ))S 5 (M ) =
20
for s ∈] − S, s0 [. This leads to a contradiction and, hence, (7.2.15) has been
proven. It remains to use (7.2.18) and (7.2.21) with s0 = 0 in order to
establish (7.2.16).
7.3 Limiting Procedure

As to w(k) , it is defined by the solution formula
1
Z |x − y|2
(k)
w (x, t) = 3 exp − u(k) (y, −S)dy.
(4π(s + S)) 2 3 4(s + S)
R
Moreover, by standard localization arguments, the following estimate can be

derived:
sup sup kw(k) (·, s)k2L2 (B(x0 ,1)) +
−S<s<0 x0 ∈R3
Z0 Z
+ sup |∇w(k) (y, s)|2 dyds ≤ c(M ) < ∞.
x0 ∈R3
−S B(x0 ,1)
Obviously, w(k) and all its derivatives converge to w and to its correspond-
ing derivatives uniformly in sets of the form B(R) × [δ, 0] for any R > 0 and
for any δ ∈] − S, 0[. The limit function satisfies the same representation
formula
1
Z |x − y|2
w(x, t) = 3 exp − w0 (y)dy,
(4π(s + S)) 2 4(s + S)
R3
in which w0 is the weak L3 (R3 )-limit of the sequence u(k) (·, −S). The function
w satisfies the uniform local energy estimate
sup sup kw(·, s)k2L2 (B(x0 ,1)) +

−S<s<0 x0 ∈R3
Z0 Z
+ sup |∇w(y, s)|2 dyds ≤ c(M ) < ∞.
x0 ∈R3
−S B(x0 ,1)
The important fact, coming from the solution formula, is as follows:
w ∈ C([−S, 0]; L3 (R3 )) ∩ L5 (R3 ×] − S, 0[). (7.3.1)
Next, the uniform local energy estimate for the sequence u(k) (with respect
to k) can be deduced from the estimates above. This allows us to exploit
the limiting procedure explained in [25], see Appendix B, in details. As a
result, one can selected a subsequence, still denoted by u(k) , with the following
properties:
for any a > 0,
u(k) → u (7.3.2)
weakly-star in L∞ (−S, 0; L2 (B(a))) and strongly in L3 (B(a)×] − S, 0[) and
in C([τ, 0]; L 9 (B(a))) for any −S < τ < 0;
8
∇u(k) → ∇u (7.3.3)
weakly in L2 (B(a)×] − S, 0[);

Z Z
(k)
t 7→ u (x, t) · w(x)dx → t 7→ u(x, t) · w(x)dx (7.3.4)
B(a) B(a)
strongly in C([−S, 0]) for any w ∈ L2 (B(a)). The corresponding sequences

v (k) and w(k) converge to their limits v and w in the same sense and of course
u = v + w. For the pressure p, we have the following convergence: for any

(k)
n ∈ N, there exists a sequences cn ∈ L 3 (−S, 0) such that
2
pe(k)
n ≡p (k)
− c(k)
n * p (7.3.5)
in L 3 (−S, 0; L 3 (B(n))).
2 2
So, arguing in the same way as in [25], see Appendix B, one can show
that u and p satisfy the following conditions:
Z0 Z
sup sup ku(·, s)k2L2 (B(x0 ,1)) + sup |∇u(y, s)|2 dyds < ∞; (7.3.6)
−S<s<0 x0 ∈R3 x0 ∈R3
−S B(x0 ,1)
p ∈ L 3 (−S, 0; L 3 ,loc (R3 ); (7.3.7)

2 2
the function Z
s 7→ u(y, s) · w(y)dy (7.3.8)
R3
is continuous on [−S, 0] for any compactly supported w ∈ L2 (R3 );
∂t u + u · ∇u − ∆u = −∇p, div u = 0 (7.3.9)
in R3 ×] − S, 0[ in the sense of distributions;
for any x0 ∈ R3 , there exists a function cx0 ∈ L 3 (−S, 0) such that
2
p(x, t) − cx0 (t) = p1x0 (x, t) + p2x0 (x, t) (7.3.10)

for all x ∈ B(x0 , 3/2), where
Z
1 1
p1x0 (x, t) = − |u(x, t)|2 + K(x − y) : u(y, t) ⊗ u(y, t)dy,
3 4π
B(x0 ,2)
Z
1
p2x0 (x, t) = (K(x − y) − K(x0 − y)) : u(y, t) ⊗ u(y, t)dy;
4π
R3 \B(x0 ,2)
for any s ∈] − S, 0[ and for ϕ ∈ C0∞ (R3 ×] − S, S[),

Z Zs Z
ϕ2 (y, s)|u(y, s)|2 dy + 2 ϕ2 |∇u|2 dydτ ≤
R3 −S R3
Zs Z
2 2 2 2 2
≤ |u| (∆ϕ + ∂ϕ ) + u · ∇ϕ (|u| + 2p) dydτ. (7.3.11)
−S R3
Passing to the limit in (7.2.16), we find

1
sup kv(·, s)k2L2 (B(x0 ,1)) ≤ c(M )(s + S) 5
x0 ∈R3
for all s ∈ [−S, 0]. And thus

v(·, s) → 0 in L2,loc (R3 )
as s ↓ −S. Then, taking into account (7.3.1), we can conclude that
u(·, s) → w0 in L2,loc (R3 ). (7.3.12)
as s ↓ −S.
By definition accepted in [25], see Apendix B, the pair u and p, satisfying
(7.3.6)–(7.3.12), is a local energy solution to the Cauchy problem for the
Navier-Stokes equations in R3 ×] − S, 0[ associated with the initial velocity
w0 .
Now, our aim is to show that u is not identically zero. Using the inverse
scaling, we observe that the following identity takes place:
Z Z
1 (k) 3 (k) 23 1 3
2
(|u | + |ep | )dy ds = 2
(|v|3 + |q − b(k) | 2 )dx dt
a (aλk )
Q(a) Q(zT ,aλk )
p p
for all 0 < a < a∗ = inf{1, S/10, T /10} and for all λk ≤ 1. Here,
(k) (k)
zT = (0, T ), pe(k) ≡ pe2 , and b(k) (t) = λ−2k c2 (s). Since the pair v and q − b
(k)
is a suitable weak solution to the Navier-Stokes equations in Q(zT , λk a∗ ), we

find Z
1 3
2
(|u(k) |3 + |e
p(k) | 2 )dy ds > ε (7.3.13)
a
Q(a)
for all 0 < a < a∗ with a positive universal constant ε.

Now, by (7.3.2) and (7.3.5),
Z Z
1 (k) 3 1
|u | dy ds → 2 |u|3 dy ds (7.3.14)
a2 a
Q(a) Q(a)
for all 0 < a < a∗ and

Z
1 3
sup 2 (|u(k) |3 + |e
p(k) | 2 )dy ds = M1 < ∞. (7.3.15)
k∈N a∗
Q(a∗ )
To treat the pressure pe(k) , we do the usual decomposition of it into two

parts, see similar arguments in [55]. The first one is completely controlled
by the pressure while the second one is a harmonic function in B(a∗ ) for all
admissible t. In other words, we have
(k) (k)
pe(k) = p1 + p2
(k)
where p1 obeys the estimate
(k)
kp1 (·, s)k 3 ,B(a∗ ) ≤ cku(k) (·, s)k23,B(a∗ ) . (7.3.16)
2
For the harmonic counterpart of the pressure, we have

Z
(k) 3 (k) 3
sup |p2 (y, s)| 2 ≤ c(a∗ ) |p2 (y, s)| 2 dy
y∈B(a∗ /2)
B(a∗ )
Z
3
≤ c(a∗ ) p(k) (y, s)| 2 + |u(k) (y, s)|3 )dy
(|e (7.3.17)
B(a∗ )
−a2∗
for all < s < 0.
For any 0 < a < a∗ /2,
Z
1 3
ε≤ 2 p(k) | 2 + |u(k) |3 )dy ds ≤
(|e
a
Q(a)
Z
1 (k) 3 (k) 3
≤c (|p1 | 2 + |p2 | 2 + |u(k) |3 )dy ds ≤
a2
Q(a)
Z
1 (k) 3
≤c 2 (|p1 | 2 + |u(k) |3 )dy ds+
a
Q(a)
Z0
1 3 (k) 3
+ca 2 sup |p2 (y, s)| 2 ds.
a y∈B(a∗ /2)
−a2
From (7.3.15)–(7.3.17), it follows that

Z Z0 Z
1 (k) 3 3
ε≤c 2 |u | dy ds + ca ds p(k) (y, s)| 2 + |u(k) (y, s)|3 )dy ≤
(|e
a
Q(a∗ ) −a2 B(a∗ )
Z Z
1 (k) 3 3
≤c 2 |u | dy ds + ca p(k) | 2 + |u(k) |3 )dy ds ≤
(|e
a
Q(a∗ ) Q(a∗ )
Z
1
≤c |u(k) |3 dy ds + cM1 aa2∗
a2
Q(a∗ )
for all 0 < a < a∗ /2. After passing to the limit and picking up sufficiently
small a, we find Z
2
0 < cεa ≤ |u|3 dy ds (7.3.18)
Q(a∗ )
for some positive 0 < a < a∗ /2. So, the limit function u is non-trivial.
Proof Theorem 1.1 The limit function w0 ∈ L3 and, hence,
kw0 k2,B(x0 ,1) → 0
as |x0 | → ∞. The latter, together with Theorem 1.4 from Appendix B, and
ε-regularity theory for the Navier-Stokes equations, gives a required decay at
infinity. To be more precise, there are positive numbers R, T ∈]a∗ , S[, and
ck with k = 0, 1, ... such that
|∇k u(x, t)| ≤ ck (7.3.19)
for any x ∈ R3 \ B(R/2) and for any t ∈] − T, 0[.

The second thing to be noticed is that the following important property
holds true:
u(·, 0) = 0. (7.3.20)
This follows from (7.2.2) and (7.3.2), see the last statement in (7.3.2). Since
vorticity ω = ∇ ∧ u vanishes at t = 0 as well, we can apply the backward
uniqueness result from Appendix A to the vorticity equation and conclude
that ω = 0 in (R3 \ B(R/2))×] − T, 0[. Now, our aim is to show that in
fact ω = 0 in R3 ×] − T, 0[. If so, u(·, t) is going to be a bounded harmonic
function with the additional property ku(·, t)kL2 (B(x0 ,1)) → 0 as |x0 | → ∞
and thus we may conclude that u = 0 in (R3 \ B(R/2))×] − T, 0[. The latter
contradicts (7.3.18) and, hence, zT is not a singular point.
The idea of the proof of the above claim is more or less the same as in
paper [13]. However, in the present case, we have less regularity and no
global finite norm for the pressure. The way out is to use decomposition
(7.3.10) in order to get better estimates for the pressure, say, in the domain
(R3 \ B(R))×] − T, 0[. Indeed, using estimates of type (7.2.10) and (7.2.11)
for the parts of the pressure p1x0 and p2x0 in (7.3.10), we show
kp(·, t) − cx0 (t)kL3/2 (B(x0 ,3/2)) < c
provided B(x0 , 2) ∈ R3 \ B(R/2). Here, a constant c is independent of

x0 and t. Then, local regularity theory, applied to the pressure equation
∆p = −div div u ⊗ u, together with estimate (7.3.19), implies
|∇k p(x, t)| < c1k
for any x ∈ R3 \ B(R), any t ∈]0, T [, and any k = 1, 2, .... If we replace

the pressure p with p − [p]B(4R)\B(R) , then from Poincare’s inequality, from
previous estimates, and from the equation ∂t u + u · ∇u = −∇p, it follows
that
|∇k u(x, t)| + |∇k p(x, t)| + |∇k ∂t u(x, t)| < c2k (7.3.21)
for all x ∈ B(4R) \ B(R), for all t ∈] − T, 0[, and for all k = 0, 1, ....
Next, we pick up a smooth cut-off function ϕ such that ϕ = 0 out of B(3R)
and ϕ = 1 in B(2R) and introduce auxiliary functions w e and re obeying the
equations
∆we = ∇er, divwe = u · ∇ϕ
in B(4R) and the additional conditions
Z
w|
e ∂B(4R) = 0, redx = 0.
B(4R)
In a view of (7.3.19), the regularity theory for the stationary Stokes system
gives the estimates
|∇k w(x,
e t)| + |∇k re(x, t)| + |∇k ∂t w(x,
e t)| < c2k (7.3.22)
being valid for all x ∈ B(4R), for all t ∈] − T, 0[, and for all k = 0, 1, ....
Letting U = w − w e and P = r − re, where w = ϕu and r = ϕp, we find
∂t U + div(U ⊗ U ) − ∆U + ∇P = F = −div(U ⊗ w e ⊗ U ) + G,
e+w
div U = 0
in Q∗ = B(4R)×] − T, 0[,
U |∂B(4R)×]−T,0[ = 0.
Here, G = −div(w
e ⊗ w)
e + g − ∂t w
e and
g = (ϕ2 − ϕ)div(u ⊗ u) + uu · ∇ϕ2 + p∇ϕ − 2∇ϕ · ∇u − u∆ϕ.
Since u and p are a local energy solution, it follows from its definition that
there exists a set Σ ⊂] − T, 0[ of full measure, i.e., |Σ| = T , such that U is a
weak Leray-Hopf solution to initial boundary problem for the above system in
B(4R)×]t0 , 0[ for each t0 ∈ Σ. The rest of the proof is based upon estimates
(7.3.21) and (7.3.22) and unique continuation across spatial boundaries for
parabolic differential inequalities and goes along lines of arguments in the
last section of Chapter 6. Theorem 1.1 is proved.
Let us outline the proof of (7.1.4), which is much easier than the proof
of Theorem 1.1. Indeed, arguing as in the main case, we find a sequence
tk → T − 0 such that
m−3
lim kv(·, tk )km (T − tk ) 2m = 0.
k→∞
The scaling implies ku(k) (·, −S)km → 0 and thus

ku(k) (·, −S)k2,unif → 0. (7.3.23)
For solutions u(k) , we may use local energy estimates proved in Appendix B.
In particular, the y give the estimate
ku(k) (·, t)k22,unif ≤ 2cku(k) (·, −S)k22,unif
for any t ∈] − S, 0[. And S should be chosen independently of k so that

ln 2
0<S<
c(1 + (2cku(k) (·, −S)k 2,unif
)2 )
for all k ∈ N. It is possible because of (7.3.23).

So, we can claim that
Z0 Z
sup ku(k) (·, t)k22,unif + sup x0 ∈ R3 |∇u(k) |2 dxdt → 0
−S<t<0
−S B(x0 ,1)
7.4. COMMENTS 227
as k → ∞. This means that the limit solution must be identically zero.

However, using the same arguments as in the previous section, we can show
that the limit solution is not a trivial one provided that the original solution
blows up at time T .
7.4 Comments
This section is essentially based on my paper [57], which in turn summarizes
all previous attempts made in [54]-[56] to solve the problem about behaviour
of L3 -norm of the velocity field as time approaches possible blow up time.
Appendix A
Backward Uniqueness and

Unique Continuation
A.1 Carleman-Type Inequalities

We start with the first Carleman type inequality which has been already used
in [11] and [12] (see also [9], [14], and [67]).
Proposition 1.1. For any function u ∈ C0∞ (Rn ×]0, 2[; Rm ) and for any
positive number a, the inequality
|x|2

h−2a (t)e− a
R
4t
t
|u|2 + |∇u|2 dxdt
Rn ×]0,2[
(A.1.1)
|x|2
− 4t
h−2a (t)e
R
≤ c0 |∂t u + ∆u|2 dxdt.
Rn ×]0,2[
1−t
is valid with an absolute positive constant c0 and a function h(t) = te 3 .
Proof of Proposition 1.1 Our approach is based on the L2 -theory of

Carleman inequalities developed essentially in [21], see also [67].
Let u be an arbitrary function from C0∞ (Rn ×]0, 2[; Rm ). We let φ(x, t) =
2
− |x|
8t
− (a + 1) ln h(t) and v = eφ u. Then, we have
Lv := eφ (∂t u + ∆u) = ∂t v − div(v ⊗ ∇φ) − ∇v∇φ + ∆v + (|∇φ|2 − ∂t φ)v.
229
230APPENDIX A. BACKWARD UNIQUENESS AND UNIQUE CONTINUATION
The main step in the above approach is the decomposition of operator tL

into symmetric and skew symmetric parts, i.e.,
tL = S + A, (A.1.2)
where
1
Sv := t(∆v + (|∇φ|2 − ∂t φ)v) − v (A.1.3)
2
and
1
Av := (∂t (tv) + t∂t v) − t(div(v ⊗ ∇φ) + ∇v∇φ). (A.1.4)
2
Obviously,
R 2 2φ R
t e |∂t u + ∆u|2 dxdt = t2 |Lv|2 dxdt
R R R (A.1.5)
= |Sv|2 dxdt + |Av|2 dxdt + [S, A]v · v dxdt,
where [S, A] = SA − AS is the commutator of S and A. Simple calculations

show that R
I := [S, A]v · v dxdt =
R h i
= 4 t2 φ,ij v,i · v,j + φ,ij φ,i φ,j |v|2 dxdt
(A.1.6)
R 2 2
+ t |v| (∂t2 φ 2 2
− 2∂t |∇φ| − ∆ φ) dxdt
R R
+ t|∇v|2 dxdt − t|v|2 (|∇φ|2 − ∂t φ) dxdt.
Given choice of function φ, we have
Z h 0 0
h0 (t)) i 2
Z
2 h (t) a+1
I = (a + 1) t − − |v| dxdt = t|v|2 dxdt. (A.1.7)
h(t) th(t) 3
By the simple identity
1
|∇v|2 = (∂t + ∆)|v|2 − v · (∂t v + ∆v), (A.1.8)
2
we find
R R R
t2 |∇v|2 dxdt = − t|v|2 dxdt − t2 v · Lv dxdt
R (A.1.9)
+ t2 |v|2 (|∇φ|2 − ∂t φ) dxdt.
A.1. CARLEMAN-TYPE INEQUALITIES 231
In our case,
h0 (t)
|∇φ|2 − ∂t φ = −|∇φ|2 + (a + 1) .
h(t)
The latter relation (together with (A.1.7)) implies the bound
R
t2 (|∇v|2 + |v|2 |∇φ|2 ) dxdt
R R (A.1.10)
≤ 3I − t2 v · Lv dxdt ≤ b1 t2 |Lv|2 dxdt
with an absolute positive constant b1 . Since
eφ |∇u| ≤ |∇v| + |v||∇φ|, (A.1.11)
it follows from (A.1.5)–(A.1.10) that
|u|2
Z |x|2
h−2a (t)(th−1 (t))2 (a + 1) + |∇u|2 e− 4t dxdt
t
Z
|x|2
≤ b2 h−2a (t)(th−1 (t))2 |∂t u + ∆u|2 e− 4t dxdt.
Here, b2 is an absolute positive constant. Inequality (A.1.1) is proved.

The second Carleman-type inequality is, in a sense, an anisotropic one.
Proposition 1.2. Let

φ = φ(1) + φ(2) ,
0 2 2α
where φ(1) (x, t) = − |x8t| and φ(2) (x, t) = a(1 − t) xtnα , x0 = (x1 , x2 , ..., xn−1 )
so that x = (x0 , xn ), and en = (0, 0, ..., 0, 1). Then, for any function u ∈
C0∞ ((Rn+ + en )×]0, 1[; Rm ) and for any number a > a0 (α), the following in-
equality is valid:
2
|∇u|2
t2 e2φ(x,t) a |u|
R
t2 + t
dxdt
(Rn
+ +en )×]0,1[
(A.1.12)
R 2 2φ(x,t) 2
≤ c? te |∂t u + ∆u| dxdt.
(Rn
+ +en )×]0,1[
Here, c? = c? (α) is a positive constant and α ∈]1/2, 1[ is fixed.

Proof Let u ∈ C0∞ (Q1+ ; Rm ), where Q1+ = (Rn+ +en )×]0, 1[. We are going
to use formulae (A.1.2)–(A.1.6) for new functions u, v, and φ. All integrals
in those formulae are taken now over Q1+ .
First, we observe that
∇φ = ∇φ(1) + ∇φ(2)
(A.1.13)
0
(1)
∇φ (x, t) = − x4t , (2)
∇φ (x, t) = 2αa 1−t x2α−1 en .
tα n
Therefore,
∇φ(1) · ∇φ(2) = 0, |∇φ|2 = |∇φ(1) |2 + |∇φ(2) |2 . (A.1.14)
Moreover,
∇2 φ = ∇2 φ(1) + ∇2 φ(2) ,
 δij
 − 4t if 1 ≤ i, j ≤ n − 1
(1)
φ,ij = ,
0 if i = n or j = n (A.1.15)


 0 if i 6= n or j 6= n
(2)
φ,ij = .
2α(2α − 1)a 1−t x2α−2 if i = n and j = n

tα n
In particular, (A.1.15) implies

1 1−t
φ,i φ,j = − |∇φ(1) |2 + 2α(2α − 1)a α x2α−2
n |∇φ(2) |2 ≥
4t t
0 2
φ,ij ≥ − 4t1 |x |
16t3
. (A.1.16)
Using (A.1.14)–(A.1.16), we present integral I in (A.1.6) in the following
way: Z
I = I1 + I2 + t|∇v|2 dxdt, (A.1.17)
where h i
R 2 (s) (s) (s) (s)
Is = 4 t φ,ij v,i · v,j + φ,ij φ,i φ,j |v|2 dxdt
R
+ t |v| ∂t2 φ(s) − 2∂t |∇φ(s) |2 − ∆2 φ(s)
2 2

− 1t |∇φ(s) |2 + 1
∂ φ(s)
t t
dxdt, s = 1, 2.
A.1. CARLEMAN-TYPE INEQUALITIES 233
Direct calculations give us

Z
I1 = − t(|∇v|2 − |v,n |2 ) dxdt
and, therefore, Z
I= t|v,n |2 dxdt + I2 . (A.1.18)
Now, our aim is to estimate I2 from below. Since α ∈]1/2, 1[, we can drop
the fist integral in the expression for I2 . As a result, we have
Z
I2 ≥ t2 |v|2 (A1 + A2 + A3 ) dxdt, (A.1.19)
where
A1 = −∂t |∇φ(2) |2 ,
1
A2 = A1 − ∆2 φ(2) − |∇φ(2) |2 ,
t
1
A3 = ∂t2 φ(2) + ∂t φ(2) .
t
For A2 , we find
1 − t 2α−4 h 4α2 ax2α+2 i

n
A2 ≥ x a(2α − 1) − 2α(2α − 2)(2α − 3) .
tα n tα+1
Since xn ≥ 1 and 0 < t < 1, we see that A2 > 0 for all a ≥ 2. Hence, it
follows from (A.1.18) and (A.1.19) that
Z
I ≥ t2 |v|2 (A1 + A3 ) dxdt. (A.1.20)
It is not difficult to check the following inequality
x2α
n
A3 ≥ a(2α − 1) . (A.1.21)
tα+2
On the other hand,
1 1−t
−∂t |∇φ(2) |2 − |∇φ(2) |2 ≥ (2α − 1) 2α+1 4α2 a2 x2(2α−1)
n ≥0
t t
and thus
1
A1 ≥ |∇φ(2) |2 . (A.1.22)
t
Combining (A.1.20)–(A.1.22), we deduce from (A.1.5) the estimate
R 2
t |Lv|2 dxdt ≥ I
R x2α R
≥ a(2α − 1) n
tα
|v|2 dxdt + t|v|2 |∇φ(2) |2 dxdt (A.1.23)
R R
≥ a(2α − 1) |v|2 dxdt + t|v|2 |∇φ(2) |2 dxdt.
Using (A.1.8), we can find the following analog of (A.1.9)
t|∇v|2 dxdt = − 21 |v|2 dxdt − tv · Lv dxdt
R R R
R (A.1.24)
2 2
+ t|v| (|∇φ| − ∂t φ) dxdt.
Due to special structure of φ, we have
|∇φ|2 − ∂t φ = |∇φ(1) |2 − ∂t φ(1) + |∇φ(2) |2 − ∂t φ(2)
= −|∇φ(1) |2 + |∇φ(2) |2 − ∂t φ(2)
and, therefore, (A.1.24) can be reduced to the form
R
t|∇v|2 + t|v|2 (|∇φ(1) |2 + |∇φ(2) |2 ) dxdt
R
t |∇v|2 + |v|2 |∇φ|2 dxdt = − 21 |v|2 dxdt (A.1.25)
R
=
R R R
− tv · Lv dxdt + 2 t|v|2 |∇φ(2) |2 dxdt − t|v|2 ∂t φ(2) dxdt.
But
x2α
n
−t∂t φ(2) ≤ a
tα
and, by (A.1.11) and (A.1.25),
1
R 2φ R
2
te |∇u|2 ≤ − v · (tLv) dxdt
(A.1.26)
R 2 (2) 2
R x2α
+2 t|v| |∇φ | dxdt + a n
tα
|v|2 dxdt.
The classical Cauchy-Scwartz inequality, (A.1.23), and (A.1.26) yield re-
quired inequality (A.1.12).
A.2. UNIQUE CONTINUATION ACROSS SPATIAL BOUNDARIES 235
A.2 Unique Continuation Across Spatial

Boundaries
We will work with the backward heat operator ∂t + ∆ rather than the more
usual heat operator ∂t − ∆ since this will save us writing some minus signs in
many formulae. In the space-time cylinder Q(R, T ) ≡ B(R)×]0, T [⊂ Rn ×R1 ,
we consider a vector-valued function u = (ui ) = (u1 , u2 , ..., um ), satisfying
three conditions:
u ∈ W22,1 (Q(R, T ); Rm ); (A.2.1)
|∂t u + ∆u| ≤ c1 (|u| + |∇u|) a.e. in Q(R, T ) (A.2.2)
for some positive constant c1 ;
√ k
|u(x, t)| ≤ Ck (|x| + t) (A.2.3)
for all k = 0, 1, ..., for all (x, t) ∈ Q(R, T ), and for some positive constants
Ck . Here,
W22,1 (Q(R, T ); Rm ) ≡ {|u| + |∇ u| + |∇2 u| + |∂t u| ∈ L2 (Q(R, T ))}.
Condition (A.2.3) means that the origin is zero of infinite order for the func-
tion u.
Theorem 2.1. Assume that a function u obeys conditions (A.2.1)–(A.2.3).
Then, u(x, 0) = 0 for all x ∈ B(R).
Without loss of generality, we may assume that T ≤ 1. Theorem 2.1 is
an easy consequence of the following lemma.
Lemma 2.2. Suppose that all conditions of Theorem 2.1 hold. Then, there
exist a constant γ = γ(c1 ) ∈]0, 3/16[ and absolute constants β1 and β2 such
that
|x|2
|u(x, t)| ≤ c2 (c1 )A0 (R, T )e− 4t (A.2.4)
for all (x, t) ∈ Q(R, T ) satisfying the following restrictions:
0 < t ≤ γT, |x| ≤ β1 R, β2 t ≤ |x|2 .
Here, √
A0 ≡ max |u(x, t)| + T |∇u(x, t)|.
(x,t)∈Q( 4 R, 34 T )
3
Remark 2.3. According to the statement of Lemma 2.2, u(x, 0) = 0 if |x| ≤

β1 R.
Remark 2.4. From the regularity theory for parabolic equations (see [33]),
it follows that
Z 21
2
A0 ≤ c3 (c1 , R, T ) |u| dz .
Q(R,T )
√
Proof of Lemma 2.2 We let λ = 2t and % = 2|x|/λ. Suppose that
t ≤ γT ≤ γ, |x| ≤ 83 R, and 8t ≤ |x|2 . Then, as it is easy to verify, we have
% ≥ 4 and
λy ∈ B(3R/4) if y ∈ B(%); λ2 s ∈]0, 3/4[ if s ∈]0, 2[
under the condition 0 < γ ≤ 3/16. Thus the function v(y, s) = u(λy, λ2 s) is
well defined on Q(%, 2) = B(%)×]0, 2[. This function satisfies the conditions:
|∂s v + ∆v| ≤ c1 λ(|v| + |∇ v|) (A.2.5)
in Q(%, 2); √
|v(y, s)| ≤ Ck0 (|y| + s)k (A.2.6)
for all k = 0, 1, ... and for all (y, s) ∈ Q(%, 2). Here, Ck0 = Ck λk .
Given ε > 0, we introduce two smooth cut-off functions such that:

1, (y, s) ∈ Q(% − 1, 3/2)
0 ≤ ϕ(y, s) = ≤ 1,
0, (y, s) ∈
/ B(%)×] − 2, 2[

1, s ∈]2ε, 2[
0 ≤ ϕε (s) = ≤ 1.
0, s ∈]0, ε[
We let w = ϕv and wε = ϕε w. Obviously, (A.2.5) implies the following
inequality:
|∂s wε + ∆wε | ≤ c1 λ(|wε | + |∇ wε |)
(A.2.7)
+c4 (|∇ϕ||∇ v| + |∇ϕ||v| + |∆ ϕ||v| + |∂s ϕ||v|) + c4 |ϕ0ε ||v|.
The crucial point is the application of the following Carleman-type inequality,
see Proposition 1.1, to the function wε
|y|2
h−2a (s)e− 4s (|∇ wε | + |wε |)2 dyds
R
Q(%,2)
(A.2.8)
|y|2
− 4s
h−2a (s)e
R
≤ c5 |∂s wε + ∆ wε |2 dyds.
Q(%,2)
A.2. UNIQUE CONTINUATION ACROSS SPATIAL BOUNDARIES 237
Here, c5 is an absolute positive constant, a is an arbitrary positive number,

1−t
and h(t) = te 3 . We let
A= max |v(y, s)| + |∇ v(y, s)|
(y,s)∈Q(%,2)\Q(%−1, 23 )
and choose γ sufficiently small in order to provide the condition

1
10c5 c21 λ2 ≤ 20c5 c21 γ < . (A.2.9)
2
Condition (A.2.9) makes it possible to hide the strongest term in the right
hand side of (A.2.8) into the left hand side of (A.2.8). So, we derive from
(A.2.7)–(A.2.9) the following relation
|y|2
h−2a (s)e−
R
4s (|∇ wε | + |wε |)2 dyds
Q(%,2)
|y|2
h−2a (s)e−
R
≤ c6 A2 4s χ(y, s) dyds (A.2.10)
Q(%,2)
|y|2
+c6 ε12 h−2a (s)e−
R
4s |v|2 dyds.
Q(%,2ε)
Here, χ is the characteristic function of the set Q(%, 2) \ Q(% − 1, 3/2). We

fix a and take into account (A.2.6). As a result of the passage to the limit
as ε → 0, we find from (A.2.10)
|y|2
h−2a (s)e−
R
D≡ 4s (|∇ v| + |v|)2 dyds
Q(%−1,3/2)
|y|2
h−2a (s)e−
R
≤ c6 A2 4s χ(y, s) dyds (A.2.11)
Q(%,2)
R2 −
(%−1)2

≤ c06 A2 −2a
h (3/2) + ρ n−1
h −2a
(s)e 4s ds .
0
Since % ≥ 4, it follows from (A.2.11) that:

Z2
%2

D ≤ c7 A2 h−2a (3/2) + ρn−1 h−2a (s)e− 8s ds . (A.2.12)
0
In (A.2.12), the constant c7 depends on n and c1 only.

Given positive number β, we can take a number a in the following way
β%2
a= . (A.2.13)
2 ln h(3/2)
This is legal, since h(3/2) > 1. Hence, by (A.2.13), inequality (A.2.12) can
be reduced to the form
Z2 2
2 −βρ2 n−1 −β%2 −2a 2β%2 − %8s
D ≤ c7 A e 1+ρ e h (s)e ds .
0
We fix β ∈]0, 1/64[, say, β = 1/100. Then, the latter relation implies the
estimate
Z2 2
2 %
D≤ c07 (c1 , n)A2 e−β% 1+ h−2a
(s)e − 16s
ds . (A.2.14)
0
ln(3/2)
It is easy to check that β < 12
and therefore g 0 (s) ≥ 0 if s ∈]0, 2[, where
%2
− 16s
g(s) = h−2a (s)e and a and % satisfy condition (A.2.13). So, we have
2
D ≤ c8 (c1 , n)A2 e−β% , (A.2.15)
where β is an absolute positive constant.

By the choice of % and λ, we have B(µ λx , 1) ⊂ B(% − 1) for any µ ∈]0, 1].
e = B(µ x , 1)×]1/2, 1[, we find
Then, setting Q λ
Z
|y|2
D≥ e− 2 |v|2 dyds. (A.2.16)
Q
e
2 √
Observing that |y|2 ≤ 2µ2 |x|
λ2
+ 2 if y ∈ B(µ λx , 1) and letting µ = 2β, we
derive from (A.2.15) and (A.2.16) the following bound
Z
µ2 |x|2 |x|2
|v|2 dyds ≤ c08 A2 e(−2β+ 2 ) t = c08 A2 e−β t . (A.2.17)
Q
e
A.3. BACKWARD UNIQUENESS FOR HEAT OPERATOR IN HALF SPACE239
On the other hand, the regularity theory for linear parabolic equations give
us: Z
|v(µx/λ, 1/2)| ≤ c9 (c1 , n) |v|2 dyds.
2
(A.2.18)
Q
e
Combining (A.2.17) and (A.2.18), we show

p |x|2
|u( 2βx, t)|2 = |u(µx, t)|2 = |v(µx/λ, 1/2)|2 ≤ c09 A2 e−β t .
√
Changing variables x
e = 2βx, we have
p x|2
|e
|u(e
x, t)| ≤ c09 Ae− 4t
√
for |e
x| ≤ β1 R and x|2 ≥ β2 t with β1 = 3/8 2β and β2 = 16β. It remains to
√ |e
note that λ ≤ 2T and
A≤ max |u(x, t)| + λ|∇ u(x, t)|.

(x,t)∈Q( 34 R, 34 T )
Lemma 2.2 is proved.
A.3 Backward Uniqueness for Heat Operator

in Half Space
In this section, we deal with a backward uniqueness problem for the heat
operator. Our approach is due to [12], see also [11].
Let Rn+ = {x = (xi ) ∈ Rn : xn > 0} and Q+ = Rn+ ×]0, 1[. We consider
a vector-valued function u : Q+ → Rm , which is ”sufficiently regular” and
satisfies
|∂t u + ∆u| ≤ c1 (|∇u| + |u|) in Q+ (A.3.1)
for some c1 > 0 and
u(·, 0) = 0 in Rn+ . (A.3.2)
Do (A.3.1) and (A.3.2) imply u ≡ 0 in Q+ ? We prove that the answer is
positive if we impose natural restrictions on the growth of the function u at
infinity. For example, we can assume
2
|u(x, t)| ≤ eM |x| (A.3.3)
for all (x, t) ∈ Q+ and for some M > 0. Natural regularity assumptions,
under which (A.3.1)–(A.3.3) may be considered are, for example, as follows:

u and weak derivatives ∂t u, ∇u, and ∇2 u are square
(A.3.4)
integrable over bounded subdomains of Q+ .
We can formulate the main result result of this section.
Theorem 3.1. Using the above notation introduced, assume that u satisfies
conditions (A.3.1)–(A.3.4). Then u ≡ 0 in Q+ .
This extends the main result of [11] and [12], where an analogue of Theo-
rem 3.1 was proved for (Rn \ B(R))×]0, T [ instead of Q+ . Similarly to those
papers, the proof of Theorem 3.1 is based on two Carleman-type inequalities,
see (A.1.1) and (A.1.12).
We start with proofs of several lemmas. The first of them plays the crucial
role in our approach. It enables us to apply powerful technique of Carleman’s
inequalities.
Lemma 3.2. Suppose that conditions (A.3.1), (A.3.2), and (A.3.4) hold.
There exists an absolute positive constant A0 < 1/32 with the following prop-
erties. If
2
|u(x, t)| ≤ eA|x| (A.3.5)
for all (x, t) ∈ Q+ and for some A ∈ [0, A0 ], then there are constants β(A) >
0, γ(c1 ) ∈]0, 1/12[, and c2 (c1 , A) > 0 such that
0 2 x2
n
|u(x, t)| ≤ c2 e4A|x | e−β t (A.3.6)
for all (x, t) ∈ (Rn+ + 2en )×]0, γ[.

Proof In what follows, we always assume that the function u is extended
by zero to negative values of t.
According to the regularity theory of solutions to parabolic equations, see
[33], we may assume
2
|u(x, t)| + |∇u(x, t)| ≤ c3 e2A|x| (A.3.7)
for all (x, t) ∈ (Rn+ + en )×]0, 1/2[.

We fix xn > 2 and t ∈]0, γ[ and introduce the new function v by usual
parabolic scaling
v(y, s) = u(x + λy, λ2 s − t/2).
The function v is well√defined on the set Qρ = B(ρ)×]0, 2[, where ρ =

(xn − 1)/λ and λ = 3t ∈]0, 1/2[. Then, relations (A.3.1), (A.3.2), and
(A.3.7) take the form:
|∂s v + ∆v| ≤ c1 λ(|∇v| + |v|) a.e. in Qρ ; (A.3.8)

2 2 |y|2
|v(y, s)| + |∇v(y, s)| ≤ c3 e4A|x| e4Aλ (A.3.9)
for (y, s) ∈ Qρ ;
v(y, s) = 0 (A.3.10)
for y ∈ B(ρ) and for s ∈]0, 1/6].
To apply inequality (A.1.1), we pick up two smooth cut-off functions:

0 |y| > ρ − 1/2
φρ (y) = ,
1 |y| < ρ − 1

0 7/4 < s < 2
φt (s) = .
1 0 < s < 3/2
These functions take values in [0, 1]. In addition, function φρ obeys the
inequalities: |∇k φρ | < Ck , k = 1, 2. We let η(y, s) = φρ (y)φt (s) and w = ηv.
It follows from (A.3.8) that
|∂s w + ∆w| ≤ c1 λ(|∇w| + |w|) + χc4 (|∇v| + |v|).
Here, c4 is a positive constant depending on c1 and Ck only, χ(y, s) = 1 if

(y, s) ∈ ω = {ρ − 1 < |y| < ρ, 0 < s < 2} ∪ {|y| ≤ ρ − 1, 3/2 < s < 2}
and χ(y, s) = 0 if (y, s) ∈
/ ω. Obviously, function w has the compact support
n
in R ×]0, 2[ and we may use inequality (A.1.1), see Proposition 1.1. As a
result, we have
|y|2
h−2a (s)e−
R
I≡ 4s (|w|2 + |∇w|2 ) dyds ≤ c0 10(c21 λ2 I + c26 I1 ), (A.3.11)
Qρ
where Z
|y|2
I1 = χ(y, s)h−2a (s)e− 4s (|v|2 + |∇v|2 ) dyds.
Qρ
Choosing γ = γ(c1 ) sufficiently small, we may assume that the inequality

c0 10c21 λ2 ≤ 1/2 holds and then (A.3.11) implies
I ≤ c5 (c1 )I1 .
On the other hand, if A < 1/32, then
1 1
8Aλ2 − <− (A.3.12)
4s 8s
for s ∈]0, 2]. By (A.3.9) and (A.3.12), we have
2 R2 R |y|2
I1 ≤ c23 e8A|x| χ(y, s)h−2a (s)e− 8s dyds
0 B(ρ)
(A.3.13)
2
h R2 (ρ−1)2
i
≤ c6 e8A|x| h−2a (3/2) + h−2a (s)e− 8s ds .
0
Now, taking into account (A.3.13), we deduce the bound
Z Z1 Z Z1
2
D≡ |w| dyds = |v|2 dyds
B(1) 1 B(1) 1
2 2
Z
|y|2
≤ c7 h−2a (s)e− 4s (|w|2 + |∇w|2 ) dyds
Qρ
Z2
ρ2
h i
8A|x|2
≤ c8 (c1 )e h (3/2) + h−2a (s)e− 32s ds
−2a
Z2
ρ2
h i
8A|x|2 −2βρ2 −2a 2βρ2 −2a 2βρ2 − 32s
= c8 e h (3/2)e + h (s)e ds .
0
We can take β = 8A < 1/256 and then choose
a = βρ2 / ln h(3/2).
Since ρ ≥ xn , such a choice leads to the estimate
h Z2 i
8A|x0 |2 −βρ2
D ≤ c8 e e 1 + g(s) ds ,
0
ρ2
where g(s) = h−2a (s)e− 64s . It is easy to check that g 0 (s) ≥ 0 for s ∈]0, 2[ if
1
β < 96 ln h(3/2). So, we have
0 2 2 0 2 βx2
n
D ≤ 2c8 e8A|x | e−βρ ≤ 2c8 e8A|x | e− 12t . (A.3.14)
On the other hand, the regularity theory implies
|v(0, 1/2)|2 = |u(x, t)|2 ≤ c08 D. (A.3.15)
Combining (A.3.14) and (A.3.15), we complete the proof of the lemma.

Next lemma is a consequence of Lemma 3.2 and the second Carleman
inequality (see (A.1.12)).
Lemma 3.3. Suppose that the function u obeys conditions (A.3.1), (A.3.2),
(A.3.4), and (A.3.5). There exists a number γ1 (c1 , c? ) ∈]0, γ/2] such that
u(x, t) = 0 for all x ∈ Rn+ and for all t ∈]0, γ1 [.
Proof As usual, by Lemma 3.2 and by the regularity theory, we may

assume
0 2 x2
n
|u(x, t)| + |∇u(x, t)| ≤ c9 (c1 , A)e8A|x | e−β 2t (A.3.16)
for all x ∈ Rn+ + 3en and for all t ∈]0, γ/2].
By scaling,
√ we define function v(y, s) = u(λy, λ2 s − γ1 ) for (y, s) ∈ Q+
with λ = 2γ1 . This function satisfies the relations:
|∂s v + ∆v| ≤ c1 λ(|∇v| + |v|) a.e. in Q+ ; (A.3.17)
v(y, s) = 0 (A.3.18)
for all y ∈ Rn+ and for all s ∈]0, 1/2[;
βλ2 y 2
n 2
yn
8Aλ2 |y 0 |2 − 2(λ2 s−γ1 ) 2 |y 0 |2
|∇v(y, s)| + |v(y, s)| ≤ c9 e e ≤ c9 e8Aλ e−β 2s (A.3.19)
√
for all 1/2 < s < 1 and for all y ∈ Rn+ + λ3 en . Since A < 1/32 and λ ≤ γ≤
√
1/ 12, (A.3.19) can be reduced to the form
2
|y 0 |2 yn
|∇v(y, s)| + |v(y, s)| ≤ c11 e 48 e−β 2s (A.3.20)
for the same y and s as in (A.3.19).

Let us fix two smooth cut-off functions:

3

0 yn < λ
+1
ψ1 (yn ) = 3 ,
1 yn > λ
+ 32
and
1 r > −1/2
ψ2 (r) = .
0 r < −3/4
We set (see Proposition 1.2 for the definition of φ(1) and φ(2) )
1 (2) yn2α
φB (yn , s) = φ (yn , s) − B = (1 − s) α − B,
a s
where α ∈]1/2, 1[ is fixed, B = a2 φ(2) ( λ3 + 2, 1/2), and
η(yn , s) = ψ1 (yn )ψ2 (φB (yn , s)/B), w(y, s) = η(yn , s)v(y, s).
Although function w is not compactly supported in Q1+ = (R3+ + en )×]0, 1[,

but, by the statement of Lemma 3.2 and by the special structure of the weight
in (A.1.12), we can claim validity of (A.1.12) for w. As a result, we have
Z
(1)
s2 e2φ e2aφB (|w|2 + |∇w|2 ) dyds
Q1+
Z
(1)
≤ c? s2 e2φ e2aφB |∂s w + ∆w|2 dyds.
Q1+
Arguing as in the proof of Lemma 3.2, we can select γ1 (c1 , c? ) so small that
Z
|y 0 |2
I≡ s2 e2aφB (|w|2 + |∇w|2 )e− 4s dyds
Q1+
Z
|y 0 |2
≤ c10 (c1 , c? ) χ(yn , s)(syn )2 e2aφB (|v|2 + |∇v|2 )e− 4s dyds,
3
(Rn
+ +( λ +1)en )×]1/2,1[
where χ(yn , s) = 1 if (yn , s) ∈ ω, χ(yn , s) = 0 if (yn , s) ∈

/ ω, and
ω ≡ {(yn , s) : yn > 1, 1/2 < s < 1, φB (yn , s) < −D/2},

A.4. COMMENTS 245
where D = −2φB ( λ3 + 32 , 12 ) > 0. Now, we wish to estimate the right hand

side of the last inequality with the help of (A.3.20). We find
Z+∞ Z1 2
Z
yn 1 1 0 2
−Da 2 −β
I ≤ c11 e (yn s) e s dyn ds e( 24 − 4s )|y | dy 0 .
3 1/2 Rn−1
λ
+1
Passing to the limit as a → +∞, we see that v(y, s) = 0 if 1/2 ≤ s < 1 and
φB (yn , s) > 0. Using unique continuation across spatial boundaries, we show
that v(y, s) = 0 if y ∈ Rn+ and 0 < s < 1.
Now, Theorem 3.1 follows from Lemmas 3.2 and 3.3 with the help of more
or less standard arguments. We shall demonstrate them just for complete-
ness.
Lemma 3.4. Suppose that the function u meets all conditions of Lemma 3.3.
Then u ≡ 0 in Q+ .
n
Proof By Lemma 3.3, u(x, t) = 0 for x ∈ R √+ and for t ∈]0, γ1 [. By
(1)
scaling, we introduce the function u (y, s) = u( 1 − γ1 y, (1 − γ1 )s + γ1 ).
It easy to check that function u(1) is well-defined in Q+ and satisfies all
conditions of Lemma 3.3 with the same constants c1 and A. Therefore,
u(1) (y, s) = 0 for yn > 0 and for 0 < s < γ1 . The latter means that u(x, t) = 0
for xn > 0 and for 0 < t < γ2 = γ1 + (1 − γ1 )γ1 . Then, we introduce the
function
p
u(2) (y, s) = u( 1 − γ2 y, (1 − γ2 )s + γ2 ), (y, s) ∈ Q+ ,
and apply Lemma 3.3. After k steps we shall see that u(x, t) = 0 for xn > 0
and for 0 < t < γk+1 , where γk+1 = γk + (1 − γk )γ1 → 1.
A0
Proof of Theorem 3.1 Assume that A0 < M . Then λ2 ≡ 2M <
1 2
2
. Introducing function v(y, s) = u(λy, λ s), (y, s) ∈ Q + , we see that this
function satisfies all conditions of Lemma 3.4 with constants c1 and A = 21 A0 .
A0
Therefore, u(x, t) = 0 for xn > 0 and for 0 < t < 2M . Now, we repeat
A0
arguments of Lemma 3.4, replacing γ1 to 2M and A to M , and end up with
the proof of the theorem.
A.4 Comments
The whole chapter is essentially due to a part of the paper [13]
Appendix B
Lemarie-Riesset Local Energy

Solutions
B.1 Introduction
In this Chapter, we deal with the the Cauchy problem for the 3D Navier-
Stokes equations with initial data belonging to the special Morrey class
L2,unif . In contrast to the space L2 , this class does not exclude, for example,
interesting homogeneous functions of order minus one. The main goal is to
show the existence of weak solutions that satisfy the local energy inequality.
Since the global energy inequality, which is the crucial point in the defini-
tion of weak Leray-Hopf solutions, is not valid any more, we are forced to
work with local energy estimates only. However, local estimates involve the
pressure, while, in the global L2 -case, the pressure does not appear at all.
For the Cauchy problem, one would hope to use a nice solution formula for
the pressure in terms of singular integrals. This formula is well-defined for
weak Leray-Hopf solutions, but it should be modified somehow in order to be
useful for functions with very weak decay at the spatial infinity. The problem
of the existence of weak solutions for the initial data from L2,unif has been
essentially solved by P. G. Lemarie-Riesset [35] and our aim is to give our
interpretation of his interesting and important results.
Let us consider the classical Cauchy problem for the Navier-Stokes equa-
tions:
247
248 APPENDIX B. LEMARIE-RIESSET LOCAL ENERGY SOLUTIONS
∂t v(x, t) + div v(x, t) ⊗ v(x, t) − ∆ v(x, t) = g(x, t) − ∇ p(x, t),

(B.1.1)
div v(x, t) = 0
for (x, t) ∈ QT = R3 ×]0, T [ together with the initial condition
v(x, 0) = a(x), x ∈ R3 . (B.1.2)
It is supposed that
◦ ◦
a ∈E 2 , g ∈ G2 (0, T ). (B.1.3)
◦ ◦
Here, spaces E m and Gm (0, T ) with m ≥ 1 are defined as follows:
◦
3
E m = {u ∈ Em : div u = 0 in R },
◦
3
Gm (0, T ) = {u ∈ Gm (0, T ) : div u = 0 in QT = R ×]0, T [ },
Z
Em = {u ∈ Lm,unif : |u(x)|m dx → 0 as |x0 | → +∞},
B(x0 ,1)
ZT Z
Gm (0, T ) = {u ∈ Lm,unif (0, T ) : |u(x, t)|m dxdt → 0
0 B(x0 ,1)
as |x0 | → +∞},
Z 1/m
m
Lm,unif = {u ∈ Lm,loc : kukLm,unif = sup |u(x)| dx < +∞},
x0 ∈R3
B(x0 ,1)
Lm,unif (0, T ) = {u ∈ Lm,loc (QT ) : kukLm,unif (0,T ) =

ZT Z 1/m
sup |u(x, t)|m dxdt < +∞}.
x0 ∈R3
0 B(x0 ,1)
◦
As it has been shown in [35] (see also references there), the space E m is in
fact the closure of the set
◦
∞ 3 ∞ 3 3
C 0 (R ) = {u ∈ C0 (R ) : div u = 0 in R }
B.1. INTRODUCTION 249
with respect to the norm of the space Lm,unif . For the readers’ convenience,
we give the proof of this fact in the last section of this chapter, see Lemma
6.1.
In monograph [35], P. G. Lemarie-Riesset proved that, for g = 0, problem
(B.1.1)–(B.1.3) has at least one weak solution v with the following properties
(see Definition 32.1 in [35]): for any T > 0,
ZT Z
v ∈ L∞ (0, T ; L2,unif ), sup |∇ v|2 dxdt < +∞,
x0 ∈R3
0 B(x0 ,1)
kv(·, t) − a(·)kL2,unif → 0 as t → +0,

v is a suitable weak solution in the sense of Caffarelli-Kohn-Nirenberg
in QT = R3 ×]0, T [.
This definition seems to be a bit weak and admits trivial non-uniqueness.
Indeed, let a smooth vector-valued function c(t) satisfy c(0) = 0. Then
v(x, t) = c(t) and p(x, t) = −c0 (t) · x is also a weak solution for zero initial
data. To avoid such type of uniqueness, one may add more restrictions on
the velocity or on the pressure. Our definition involves the pressure in more
explicit way and is follows.
Definition 1.1. We call a pair of functions v and p defined in the space-time

cylinder QT = R3 ×]0, T [ a local energy weak Leray-Hopf solution or just a
local energy solution to the Cauchy problem (B.1.1)–(B.1.3) if they satisfy
the following conditions:
v ∈ L∞ (0, T ; L2,unif ), ∇v ∈ L2,unif (0, T ), p ∈ L 3 (0, T ; L 3 ,loc (R3 )); (B.1.4)

2 2
v and p meet (B.1.1) in the sense of distributions; (B.1.5)

Z
the f unction t 7→ v(x, t) · w(x) dx is continuous on [0, T ] (B.1.6)
R3
for any compactly supported function w ∈ L2 (R3 );

for any compact K,
kv(·, t) − a(·)kL2 (K) → 0 as t → +0; (B.1.7)

Z Zt Z Zt Z
2 2
ϕ|v(x, t)| dx + 2 ϕ|∇v| dxdt ≤ |v|2 (∂t ϕ + ∆ϕ)
R3 0 R3 0 R3

+v · ∇ϕ(|v|2 + 2p) + 2ϕg · v dxdt (B.1.8)
for a.a. t ∈]0, T [ and for nonnegative smooth functions ϕ vanishing in a

neighborhood of the parabolic boundary of the space-time cylinder R3 ×]0, T [;
for any x0 ∈ R3 , there exists a function cx0 ∈ L 3 (0, T ) such that
2
px0 (x, t) ≡ p(x, t) − cx0 (t) = p1x0 (x, t) + p2x0 (x, t), (B.1.9)
for (x, t) ∈ B(x0 , 3/2)×]0, T [, where

Z
1 1
p1x0 (x, t) = − |v(x, t)|2 + K(x − y) : v(y, t) ⊗ v(y, t) dy,
3 4π
B(x0 ,2)
Z
1
p2x0 (x, t) = (K(x − y) − K(x0 − y)) : v(y, t) ⊗ v(y, t) dy
4π
R3 \B(x0 ,2)
and K(x) = ∇2 (1/|x|).
Remark 1.2. It is easy to see that (B.1.4), (B.1.6)–(B.1.8) imply the fol-
lowing inequality:
Z Zt Z Z
2 2
ϕ(x)|v(x, t)| dx + 2 ϕ|∇ v| dxds ≤ ϕ(x)|v(x, t0 )|2 dx
R3 t0 R3 R3
Zt Z h i
+ |v|2 ∆ ϕ + ∇ ϕ · v |v|2 + 2p + 2ϕg · v dxds. (B.1.10)
t0 R3
It is valid for any t ∈ [0, T ], for a.a. t0 ∈ [0, T ], including t0 = 0, and for
any nonnegative function ϕ ∈ C0∞ (R3 ).
Remark 1.3. In turn, from (B.1.4), (B.1.6), and (B.1.10), it follows that if
v and p are a local energy solution on the set R3 ×]0, T [, then they are a local
energy solution on the set R3 ×]t0 , T [ for a.a. t0 ∈ [0, T ], including t0 = 0.
B.1. INTRODUCTION 251
We are going to prove the followings statements. The first of them shows
that our information about pressure is sufficient to prove decay for both
velocity v and p.
Theorem 1.4. Assume that conditions (B.1.3) hold. Let v and p be a local
energy solution to the Cauchy problem (B.1.1), (B.1.2). Then v and p satisfy
the following additional properties:
◦
v(·, t) ∈E 2 (B.1.11)
for all t ∈ [0, T ];

◦
v(·, t) ∈E 3 (B.1.12)
for a.a. t ∈ [0, T ];
kv(·, t) − a(·)kL2,unif → 0 as t → +0; (B.1.13)
ZT Z
3
sup |p(x, t) − cx0 (t)| 2 dxdt < +∞,
x0 ∈R3
0 B(x0 ,3/2)
ZT Z
3
sup I{|x|>R} |p(x, t) − cx0 (t)| 2 dxdt → 0 (B.1.14)
x0 ∈R3
0 B(x0 ,3/2)
as R → +∞, where I{|x|>R} is the characteristic function of the set {x ∈ R3 :

|x| > R}.
The main theorem of the chapter is Theorem 1.5 below.
Theorem 1.5. Assume that conditions (B.1.3) hold. There exists at least
one local energy solution to the Cauchy problem (B.1.1), (B.1.2).
The substantial counterpart of the proof of Theorem 1.5 is the statement

on the local in time existence of local energy weak solutions.
Proposition 1.6. (local in time solvability) Assume that conditions (B.1.3)

hold. There exist a number T0 ∈]0, T ], depending on kakL2,unif and on
kgkL2,unif (0,T ) only, and two functions v and p, being a local energy solution
to the Cauchy problem:
∂t v(x, t) + div v(x, t) ⊗ v(x, t) − ∆ v(x, t) = g(x, t) − ∇ p(x, t),
(B.1.15)
div v(x, t) = 0
for x ∈ R3 and 0 < t < T0 ,
v(x, 0) = a(x), x ∈ R3 . (B.1.16)
B.2 Proof of Theorem 1.4

Let us introduce the following decomposition:
p2x0 (x, t) = px0 ,R (x, t) + p̄x0 ,R (x, t) (x, t) ∈ B(x0 , 3/2)×]0, T [, (B.2.1)
where
Z
1
p̄x0 ,R (x, t) = (K(x − y) − K(x0 − y)) : v(y, t) ⊗ v(y, t) dy.
4π
R3 \B(x0 ,2R)
Lemma 2.1. For any x0 ∈ R3 , for any t ∈]0, T [, and for any R ≥ 1, the
following estimate is valid:
c
sup |p̄x0 ,R (x, t)| ≤ kv(·, t)k2L2,unif . (B.2.2)
B(x0 ,3/2) R
Proof By our assumptions,

|x − x0 |
|K(x − y) − K(x0 − y)| ≤ c
|x0 − y|4
for x ∈ B(x0 , 3/2) and for y ∈ R3 \ B(x0 , 2R). And then
Z
1
|p̄x0 ,R (x, t)| ≤ c |v(y, t)|2 dy
|x0 − y|4
R3 \B(x0 ,2R)
∞ Z
X 1
=c |v(y, t)|2 dy
i=0
|x0 − y|4
B(x0 ,2i+2 R)\B(x0 ,2i+1 R)
B.2. PROOF OF THEOREM 1.4 253
∞ Z
X 1
≤c |v(y, t)|2 dy
i=0
(2i+1 R)4
B(x0 ,2i+2 R)
∞
X 1
≤c (2i+2 R)3 kv(·, t)k2L2,unif .
i=0
(2i+1 R)4
We let
Zt Z
α(t) = kv(·, t)k2L2,unif , β(t) = sup |∇ v|2 dxds,
x0 ∈R3
0 B(x0 ,1)
Zt Z
γ(t) = sup |v|3 dxds.
x0 ∈R3
0 B(x0 ,1)
By the known multiplicative inequality, we have
Zt 14 Zt 34
3
γ(t) ≤ c α (s) ds β(t) + α(s) ds . (B.2.3)
0 0
From our assumptions and from (B.2.3), it follows that:

2
ess sup α(t) + β(T ) + γ 3 (T ) ≤ A < +∞. (B.2.4)
0<t<T
Next, fix a smooth cut-off function χ so that
χ(x) = 0, x ∈ B(1), χ(x) = 1, x∈

/ B(2),
and then, for χR (x) = χ(x/R), let

Zt Z
αR (t) = kχR v(·, t)k2L2,unif , βR (t) = sup |χR ∇ v|2 dxds,
x0 ∈R3
0 B(x0 ,1)
Zt Z ZT Z
γR (t) = sup |χR v|3 dxds, GR = sup |χR g|2 dxds
x0 ∈R3 x0 ∈R3
0 B(x0 ,1) 0 B(x0 ,1)
Zt Z
3
δR (t) = sup |χR px0 | 2 dxds.
x0 ∈R3
0 B(x0 ,3/2)
An analogue of (B.2.3) is available with the form
Zt 14 Zt
1
Zt 34
3
γR (t) ≤ c αR (s) ds βR (t) + αR (s)ds + 2 α(s)ds . (B.2.5)
R
0 0 0
Lemma 2.2. Assume that v and p are a local energy weak Leray-Hopf solu-
tion to the Cauchy problem (B.1.1)–(B.1.3) on the space-time cylinder QT .
Then we have the estimate
2 4 h
sup αR (t) + βR (T ) + γR (T ) + δR (T ) ≤ C(T, A) kχR ak2L2,unif
3 3
0<t<T
1 i
+GR + . (B.2.6)
R2/3
Proof. To simplify our notation, we let p̂ = px0 .
We fix x0 ∈ R3 and a smooth nonnegative function ϕ such that
ϕ = 1 in B(1), spt ϕ ⊂ B(3/2)
and let ϕx0 (x) = ϕ(x − x0 ). For ψ = χ2R ϕx0 , we find from inequality (B.1.10):
Z Zt Z 5
X
2
L≡ ψ(x)|v(x, t)| dx + 2 ψ|∇ v|2 dxds = Ii , (B.2.7)
0 R3 i=1
R3
where
Z Zt Z
I1 = ψ|a|2 dx, I2 = |v|2 ∆ ψ dxds,
R3 0 R3
Zt Z Zt Z
2
I3 = ∇ ψ · v|v| dxds, I4 = 2 ∇ ψ · v p̂ dxds,
0 R3 0 R3
Zt Z
I5 = 2 ψg · v dxds.
0 R3
Obviously,
I1 ≤ ckχR ak2L2,unif , (B.2.8)
Zt
1
I2 ≤ c αR (s) ds + C(T, A) , (B.2.9)
R
0
Zt
I5 ≤ c αR (s) ds + GR . (B.2.10)
0
The term I3 is evaluated with the help of Hölder inequality in the following
way:
1/3

2/3 c 2/3
I3 ≤ cγ (t) γR (t) + γ (t) .
R
So, by (B.2.4),

2/3 1
I3 ≤ C(T, A) γR (t) + . (B.2.11)
R
Next, we let
I4 = I 0 + I 00 ,
where
Zt Z
00
I =4 χR ϕx0 ∇ χR · v p̂ dxds.
0 B(x0 ,3/2)
The term I 00 can be estimated with the help of Hölder inequality

Zt Z 31 Z t Z 32
00 1 3 3
I ≤c |v| dxds |p̂| 2 dxds .
R
0 B(x0 ,3/2) 0 B(x0 ,3/2)
From Lemma 2.1 and the theory of singular integrals, we find

Zt Z Zt Z Zt
3 3
|p̂| dxds ≤ c
2 |v|3 dxds + c α 2 (s) ds. (B.2.12)
0 B(x0 ,3/2) 0 B(x0 ,2) 0
Now, (B.2.4) and (B.2.12) give us

1
I 00 ≤ C(T, A) . (B.2.13)
R
I 0 can be estimated with the help of Hölder inequality as follows:

1
I 0 ≤ cJγR3 (t), (B.2.14)
where
Zt Z
3
23
J= |χR p̂| 2 dxds .
0 B(x0 ,3/2)
Obviously, J ≤ J1 + J2 + J3 , where
Zt Z
3
23 Zt Z
3
23
J1 = |χR p1x0 | 2 dxds , J2 = |χR px0 ,ρ | 2 dxds ,
0 B(x0 ,3/2) 0 B(x0 ,3/2)
Zt Z
3
23
J3 = |χR p̄x0 ,ρ | 2 dxds ,
0 B(x0 ,3/2)
√
where ρ = R. We start with evaluation of J1 . Letting
χR p1x0 = q1 + q2 ,
where
1
q1 (x, t) = − χR (x)|v(x, t)|2
3
Z
1
+ K(x − y)(χR (x) − χR (y)) : v(y, t) ⊗ v(y, t) dy,
4π
B(x0 ,2)
Z
1
q2 (x, t) = K(x − y)χR (y) : v(y, t) ⊗ v(y, t) dy,
4π
B(x0 ,2)
we use the theory of singular integrals and find the estimate for q2 :
Zt Z Zt Z
3 3
|q2 | dxds ≤ c
2 |χR | 2 |v|3 dxds
0 B(x0 ,3/2) 0 B(x0 ,2)
1/2
≤ C(T, A)γR (t). (B.2.15)
Since
1
q1 (x, t) = − χR (x)|v(x, t)|2
3
Z
1
+ K(x − y)(χR (x) − χR (x0 )) : v(y, t) ⊗ v(y, t) dy
4π
B(x0 ,2)
Z
1
+ K(x − y)(χR (x0 ) − χR (y)) : v(y, t) ⊗ v(y, t) dy,
4π
B(x0 ,2)
the same arguments lead to the estimate
Zt Z Zt Z
3 1/2 c
|q1 | dxds ≤ cγ 1/2 (t)γR (t) + 3/2
2 |v|3 dxds
R
0 B(x0 ,3/2) 0 B(x0 ,2)
Zt Z
3
+c |χR (x0 ) − χR (x))| 2 |v(x, s)|3 dxds
0 B(x0 ,2)
1
1/2
≤ C(T, A)
+ γR (t) .
R3/2
Combining the latter estimate with (B.2.15), we find
1
1/3
J1 ≤ C(T, A) + γR (t) . (B.2.16)
R
Next, we let
χR px0 ,ρ = q3 + q4 ,
where
Z
1
q3 (x, t) = (K(x − y) − K(x0 − y))(χR (x)
4π
B(x0 ,2ρ)\B(x0 ,2)
−χR (y)) : v(y, t) ⊗ v(y, t) dy,

Z
1
q4 (x, t) = (K(x − y) − K(x0 − y))χR (y) : v(y, t) ⊗ v(y, t) dy.
4π
B(x0 ,2ρ)\B(x0 ,2)
For x ∈ B(x0 , 3/2), we have

|x − y|
Z
|q3 (x, t)| ≤ c |K(x − y) − K(x0 − y)| |v(y, t)|2 dy
R
B(x0 ,2ρ)\B(x0 ,2)
Z
ρ
≤c |K(x − y) − K(x0 − y)||v(y, t)|2 dy
R
R3 \B(x0 ,2)
The same arguments as in the proof of Lemma 2.1 show that

ρ 1 1
|q3 (x, t)| ≤ c α(t) = c √ α(t) ≤ C(T, A) √ , (B.2.17)
R R R
√
where we used ρ = R. Similar arguments work for q4 :
1/2 1/2
|q4 (x, t)| ≤ cαR (t)α1/2 (t) ≤ C(T, A)αR (t), x ∈ B(x0 , 3/2). (B.2.18)
From (B.2.17) and (B.2.18), it follows that
h 1 Zt 23 i
3/4
J2 ≤ C(T, A) √ + αR (s) ds
R
0
h 1 Zt 16 i
3
≤ C(T, A) √ + αR (s) ds . (B.2.19)
R
0
The term J3 can be estimated with the help of (B.2.2):

1 1
J3 ≤ c α(t) ≤ C(T, A) √ . (B.2.20)
ρ R
So, by (B.2.14), (B.2.16), (B.2.19), and (B.2.20), we have
h 1 Zt 16 i
1/3 3
J ≤ C(T, A) √ + γR (t) + αR (s) ds (B.2.21)
R
0
and
h1 1 Zt 16 i
1/3
I 0 ≤ C(T, A)γR (t) γR (t) + √ +
3 3
αR (s) ds
R
0
Zt 13 i
h
2/3 1 3
≤ C(T, A) γR (t) + + αR (s) ds . (B.2.22)
R
0
Now, we can derive from (B.2.7) –(B.2.11), (B.2.13), and (B.2.22):

Zt
αR (t) + βR (t) ≤ ckχR ak2L2,unif + cGR + c αR (s) ds
0
h 1 Zt 13 i
2/3 3
+C(T, A) γR (t) + + αR (s) ds
R
0
and, by (B.2.3) and Young inequality, we find the main inequality
αR (t) + βR (t) ≤ ckχR ak2L2,unif + cGR

h1 Zt Zt 13 i
3
+C(T, A) + αR (s) ds + αR (s) ds . (B.2.23)
R
0 0
The important consequence of (B.2.23) is as follows:
h 1 Zt i
3
αR (t) ≤ ckχR ak6L2,unif + cGR 3 + C(T, A) 3 + αR3
(s) ds .
R
0
The latter implies

h 1i
αR (t) ≤ C(T, A) kχR ak2L2,unif + GR + ,
R
which, together with (B.2.23), (B.2.21), and (B.2.3), proves (B.2.6). Lemma
2.2 is proved.
Proof of Theorem 1.4 Now, the proof of Theorem 1.4 is easy. In par-
ticular, (B.1.11) follows from (B.1.4) and (B.1.6), while (B.1.12) is deduced
from (B.2.4) and (B.2.6). In turn, (B.1.7) and (B.2.6) imply (B.1.13).
Regarding the pressure, we see, by known results for singular integrals,
that Z Z
3
1
|px0 (x, t)| 2 dx ≤ c |v(x, t)|3 dx.
B(x0 ,3/2) B(x0 ,2)
So, the first estimate in (B.1.14) follows from (B.2.2) and from (B.2.6). Fi-
nally, the second estimate in (B.1.14) is one of the statements of Lemma
B.2.2, see (B.2.6). Theorem 1.4 is proved.
B.3 Regularized Problem

Assume that condition (B.1.3) holds. Then according to Lemma B.1.3, we
can consider the following regularised problem:
∂t v ε + Fε (v ε ) · ∇ v ε − ∆ v ε = g ε − ∇ pε ,
(B.3.1)
div v ε = 0
in R3 ×]0, +∞[ with
g ε ∈ L2 (QT ), div g ε = 0, (B.3.2)

◦
v ε |t=0 = aε ∈ C ∞ 3
0 (R ) (B.3.3)
in R3 . Here, Z
Fε (u)(x, t) ≡ %ε (x − x̄)u(x̄, t) dx̄,
R3
%ε is a standard smoothing kernel,
kaε − akL2,unif → 0, kg ε − gkL2,unif (0,T ) → 0 (B.3.4)
as ε → 0. And, we may assume that
kaε kL2,unif ≤ 2kakL2,unif , kg ε kL2,unif (0,T ) ≤ 2kgkL2,unif (0,T ) (B.3.5)
for all ε. Moreover, we may assume also that g ε is a function of class C ∞ in

QT and, for each ε > 0, there exists Rε > 0 such that the support of g ε (·, t)
lies in B(Rε ) for all t ∈]0, T [.
It is known that problem (B.3.1)–(B.3.3) has a unique very smooth solu-
tion v ε which has the finite energy and we can fix the pressure in the following
way:
Z
ε 1 1
ε ε

p (x, t) = ÷ div v (x̄, t) ⊗ Fε (v )(x̄, t) dx̄
4π |x − x̄|
R3
1
= − v ε (x, t) · Fε (v ε )(x, t) (B.3.6)
Z 3
1
+ K(x − x̄) : v ε (x̄, t) ⊗ Fε (v ε )(x̄, t) dx̄,
4π
R3
B.3. REGULARIZED PROBLEM 261
where K(x) = ∇2 (1/|x|).

Our aim is to find estimates of v ε and pε which are uniform with respect
to ε.
In what follows, we shall use the following decomposition of the pressure.
For any x0 ∈ R3 and for any 0 < r ≤ R, we let
p̂εx0 ,r (x, t) ≡ pε (x, t) − pεx0 ,r (t) = p1ε 2ε 3ε

x0 ,r (x, t) + px0 ,r,R (x, t) + px0 ,R (x, t), (B.3.7)
where
Z
1
pεx0 ,r (t) ≡ K(x0 − x̄) : v ε (x̄, t) ⊗ Fε (v ε )(x̄, t) dx̄,
4π
R3 \B(x0 ,r)
1 ε
p1ε ε
x0 ,r (x, t) ≡ − v (x, t) · Fε (v )(x, t)+
3
Z
1
K(x − x̄) : v ε (x̄, t) ⊗ Fε (v ε )(x̄, t) dx̄,
4π
B(x0 ,r)
p2ε
x0 ,r,R (x, t) ≡
Z
1
K(x − x̄) − K(x0 − x̄) : v ε (x̄, t) ⊗ Fε (v ε )(x̄, t) dx̄,
4π
B(x0 ,2R)\B(x0 ,r)
p3ε
x0 ,R (x, t) ≡
Z
1
K(x − x̄) − K(x0 − x̄) : v ε (x̄, t) ⊗ Fε (v ε )(x̄, t)dx̄.
4π
R3 \B(x0 ,2R)
Using the same arguments as in the proof of Lemma B.2.1, we prove
Lemma 3.1. For any x0 ∈ R3 and for any R ≥ 1, we have the following
estimate
cr ε
sup |p3ε
x0 ,R (x, t)| ≤ kv (·, t)kL2,unif kFε (v ε )(·, t)kL2,unif . (B.3.8)
x∈B(x0 ,r) R
Assuming that 0 < ε < 1, we observe that
kFε (v ε )(·, t)kL2,unif ≤ ckv ε (·, t)kL2,unif . (B.3.9)

Taking into account the standard estimates for singular integrals, Lemma
3.1, and inequality (B.3.9), we find:
kp1ε ε 2
x0 ,r (·, t)kL 3 (B(x0 ,r)) ≤ ckv (·, t)kL3 ((B(x0 ,2)) , (B.3.10)
2
sup |p2ε ε 2
x0 ,r,R (x, t)| ≤ C1 (r, R)kv (·, t)kL2,unif , (B.3.11)
x∈B(x0 ,3r/4)
r ε
sup |p3ε 2
x0 ,R (x, t)| ≤ c kv (·, t)kL2,unif . (B.3.12)
x∈B(x0 ,r) R
We let
Zt Z
ε
αε (t) = kv (·, t)k2L2,unif , βε (t) = sup |∇ v ε |2 dxds,
x0 ∈R3
0 B(x0 ,1)
Zt Z ZT Z
γε (t) = sup |v ε |3 dxds, G = sup |g(x, t)|2 dxdt.
x0 ∈R3 x0 ∈R3
0 B(x0 ,1) 0 B(x0 ,1)
By the known multiplicative inequality, we have
Zt 41 Zt 34
γε (t) ≤ c αε3 (s) ds βε (t) + αε (s)ds . (B.3.13)
0 0
Now, we can derive the energy estimate.
Lemma 3.2. For any t ≥ 0, the following energy estimate is valid:
h Zt i
αε (t) + βε (t) ≤ c kak2L2,unif +G+ (αε (s) + αε3 (s)) ds . (B.3.14)
0
Proof We fix x0 ∈ R3 and a smooth nonnegative function ϕ such that
ϕ = 1 in B(1), spt ϕ ⊂ B(3/2)
and let ϕx0 (x) = ϕ(x − x0 ).

B.3. REGULARIZED PROBLEM 263
From the system (B.3.1) and (B.3.2), it is easy to derive the identity
Z Zt Z Z
E≡ ϕ2x0 (x)|v ε (x, t)|2 dx +2 ϕ2x0 |∇ v ε |2 dxds = ϕ2x0 |aε |2 dx
R3 0 R3 R3
Zt Z h
ε 2
+ |v | ∆ ϕ2x0 + ∇ ϕ2x0 ε ε 2 ε
· Fε (v ) |v | + 2p̂x0 ,2 (B.3.15)
0 R3
i
+2ϕ2x0 g ε · v ε dxds.
By Hölder inequality and by (B.3.5), we show

h Zt Zt Z i
3
E≤c kak2L2,unif +G+ αε (s) ds+γε (t)+ |p̂εx0 ,2 | 2 dxds . (B.3.16)
0 0 B(x0 ,3/2)
On the other hand, (B.3.10)–(B.3.12) imply

Zt Z Zt 3
3
|p̂εx0 ,2 | dxds ≤ c γε (t) +
2 αε2 (s) ds . (B.3.17)
0 B(x0 ,3/2) 0
Taking into account (B.3.13), we derive from (B.3.16) and (B.3.17) the fol-
lowing estimate
h Zt Zt 3
αε (t) + βε (t) ≤ c kak2L2,unif +G+ αε (s) ds + αε2 (s) ds
0 0
Zt 14 Zt 34 i
3
+ αε (s) ds βε (t) + αε (s)ds .
0 0
Applying Young’s inequality twice, we complete the proof of the lemma.

A simple consequence of Lemma 3.2 is the following statement.
Lemma 3.3. There exist positive constants A and T0 ≤ T depending on
kakL2,unif and G only such that
2 4
sup αε (t) + βε (T0 ) + γε3 (T0 ) + δε3 (T0 ) ≤ A, (B.3.18)
0<t<T0
where
Zt Z
3
δε (t) = sup |p̂εx0 ,2 | 2 dxds.
x0 ∈R3
0 B(x0 ,3/2)
Indeed, let
n ln 2 o
T0 := min T, . (B.3.19)
c(1 + (2c(kak2L2,unif + G))2 )
We claim that if 0 ≤ t < T0 , then α(t) < 2c(kak2L2,unif + G). Otherwise, there
should exist T 0 < T0 such that
αε (t) < 2c(kak2L2,unif ) + G
for 0 ≤ t < T 0 and
αε (T 0 ) = 2c(kak2L2,unif + G).
The main inequality implies the following estimate
Zt
αε (t) ≤ c kak2L2,unif + G + (1 + (2c(kak2L2,unif + G))2 ) αε (τ )dτ
0
for 0 ≤ t < T 0 . In turn, this inequality implies

αε (t) ≤ c(kak2L2,unif + G) exp{c(1 + (2c(kak2L2,unif + G))2 )t}
for the same t. And thus we find
2c(kak2L2,unif + G) ≤ c(kak2L2,unif + G) exp{c(1 + (2c(kak2L2,unif + G))2 )T 0 }.
But this is possible only if T 0 ≥ T0 and contradicts the above assumption.
B.4 Passing to Limit and Proof of Proposi-

tion 1.6
First, we fix n ∈ N. From Lemma 3.3, it follows that the following estimate
is valid:
Z ZT0 Z
sup |v ε (x, t)|2 dx + |∇v ε |2 dxdt ≤ cn3 A. (B.4.1)
0<t<T0
B(n) 0 B(n)
B.4. PASSING TO LIMIT AND PROOF OF PROPOSITION 1.6 265
Using the known multiplicative inequality, we find from (B.4.1)

ZT0 Z
10
|v ε | 3 dxdt ≤
0 B(n)
Z 23 ZT0 Z 1 ε 2
ε 2 ε 2
c sup |v (x, t)| dx |∇v | + 2 |v | dxdt
0<t<T0 n
B(n) 0 B(n)
and thus
ZT0 Z
10 5
|v ε | 3 dxdt ≤ cn5 A 3 . (B.4.2)
0 B(n)
To estimate the pressure, we use (B.3.10)–(B.3.12) with x0 = 0 and r =

R = 2n. So, we have
pεn (x, t) ≡ p̂ε0,2n (x, t),
where p̂ε0,2n (x, t) ≡ pε (x, t) − pε0,2n (t), and
ZT0 Z
3
|pεn | 2 dxdt ≤ C(n, T0 , A). (B.4.3)
0 B(n)
Next, we are going to use the Navier-Stokes equations in the following

way:
ZT0 Z
∂t v ε · w dxdt
0 B(n)
ZT0 Z
= v ε ⊗ Fε (v ε ) : ∇w − ∇v ε : ∇w + pεn div w + g ε · w dxdt
0 B(n)
ZT0 Z 31 ZT0 Z 31 ZT0 Z 13

≤ |v ε |3 dxdt |Fε (v ε )|3 dxdt |∇w|3 dxdt
0 B(n) 0 B(n) 0 B(n)
ZT0 Z 12 ZT0 Z 12
+ |∇v ε |2 dxdt |∇w|2 dxdt
0 B(n) 0 B(n)
ZT0 Z 3
23 ZT0 Z 13
ε 2 3
+ |pn | dxdt |∇w| dxdt
0 B(n) 0 B(n)
ZT0 Z 12 ZT0 Z 12
ε 2 2
+c |g | dxdt |w| dxdt
0 B(n) 0 B(n)
for any w ∈ C0∞ (B(n)). Since
ZT0 Z ZT0 Z
ε 3
|Fε (v )| dxdt ≤ |v ε |3 dxdt,
0 B(n) 0 B(2n)
we have
ZT0 Z ZT0 Z 31
∂t v ε · w dxdt ≤ C(n, T0 , A, G) |∇w|3 dxdt
0 B(n) 0 B(n)
for any w ∈ C0∞ (B(n)). The latter estimates implies
k∂t v ε kXn ≤ C(n, T0 , A, G), (B.4.4)

◦ ◦
where Xn is the space dual to L3 (0, T0 ; W31 (B(n))) and W31 (B(n)) is the
closure of C0∞ (B(n)) in W31 (B(n)).
Now, we argue by induction in n. Let n = 1. Estimates (B.4.1)–(B.4.4)
make it possible to apply the known compactness arguments and to find
sequences v 1,k and p1,k
1 such that
?
v 1,k * v 1 in L∞ (0, T0 ; L2 (B(1))),
v 1,k * v 1 in L2 (0, T0 ; W21 (B(1))),
v 1,k → v 1 in L3 (0, T0 ; L3 (B(1))),
F1,k (v ) → v 1
1,k
in L3 (0, T0 ; L3 (B(δ))), ∀δ < 1,
p1,k
1 * p1 in L 3 (0, T0 ; L 3 (B(1)))
2 2
as k → +∞. We let v = v 1 and p = p1 in B(1)×]0, T0 [. Obviously, the pair

v and p satisfies the the Navier-Stokes equations in the sense of distributions
and the local energy inequality in B(1)×]0, T0 [. The latter means that
Z Zt Z Zt Z
ϕ(x, t)|v(x, t)|2 dx + 2 ϕ|∇v|2 dxdt ≤ |v|2 (∂t ϕ + ∆ϕ)
B(1) 0 B(1) 0 B(1)

+v · ∇ϕ(|v|2 + 2p) + 2ϕg · v dxdt
for a.a. t ∈]0, T0 [ and for nonnegative smooth functions ϕ vanishing in a

neighbourhood of the parabolic boundary of space-time cylinder B×]0, T0 [.
Now, let n = 2. By the same arguments as above, we assert that there
exists sequences v 2,k , which is a subsequence of v 1,k , and p2,k
2 such that
?
v 2,k * v 2 in L∞ (0, T0 ; L2 (B(2))),
2,k 2
v *v in L2 (0, T0 ; W21 (B(2))),
v 2,k → v 2 in L3 (0, T0 ; L3 (B(2))),
F2,k (v ) → v 2
2,k
in L3 (0, T0 ; L3 (B(δ))), ∀δ < 2,
p2,k
2 * p2 in L 3 (0, T0 ; L 3 (B(2))).
2 2
The functions v 2 and p2 satisfy the Navier-Stokes equations and the local
energy inequality in the space-time cylinder B(2)×]0, T0 [. Obviously, that
v 2 = v on B×]0, T0 [. So, we may extend v by letting v = v 2 on B(2)×]0, T0 [.
As to the function p2 , it follows from the Navier-Stokes equations that ∇p2 =
∇p on B×]0, T0 [. This means that p2 (x, t) − h2 (t) = p(x, t) for x ∈ B and
for t ∈]0, T0 [. Since both p2 and p belong to L 3 (0, T0 ; L 3 (B)), we conclude
2 2
the h2 ∈ L 3 (0, T0 ). This allows to extend the function p to B(2)×]0, T0 [ so
2
that p = p2 − h2 on B(2)×]0, T0 [. Clearly, p ∈ L 3 (0, T0 ; L 3 (B(2))) and the
2 2
functions v and p satisfies the Navier-Stokes equations and the local energy
inequality on the space-time cylinder B(2)×]0, T0 [.
In the case n = 3, we repeat the above arguments choosing a subsequence
of the sequence v 2,k and replacing balls B and B(2) with balls B(2) and B(3),
respectively. Continuing this process, we arrive at the following result. There
exist two functions v and p defined on R3 ×]0, T0 [ such that
v ∈ L∞ (0, T0 ; L2,loc (R3 )) ∩ L2 (0, T0 ; W2,loc

1
(R3 )),
p ∈ L 3 (0, T0 ; L 3 ,loc (R3 )). (B.4.5)
2 2
Next, letting v {k} = v k,k , we observe that v {k} is a subsequence of the

sequence {v n,k }∞ ∞
k=n , i.e., there exists a sequence of natural numbers {rk }k=n
having the following properties:
v {k} = v n,rk , k = n, n + 1, ..., rn = n, rk → ∞
as k → ∞. Then we may let
p{k} n,rk
n = pn .
{k} ∞
Obviously, pn is a subsequence of the sequence {pn,k
n }k=n . For these new
sequences and for any n ∈ N, we have
?
v {k} * v in L∞ (0, T0 ; L2 (B(n))),
v {k} * v in L2 (0, T0 ; W21 (B(n))),
v {k} → v in L3 (0, T0 ; L3 (B(n))), (B.4.6)
{k}
F{k} (v ) → v in L3 (0, T0 ; L3 (B(δ))), ∀δ < n,
p{k}
n * pn in L 3 (0, T0 ; L 3 (B(n)))
2 2
and
k∂t vkXn ≤ C(n, T0 , A, G), (B.4.7)
pn (x, t) = p(x, t) − cn (t), x ∈ B(n), t ∈]0, T0 [ (B.4.8)

for some cn ∈ L 3 (0, T0 ). From (B.3.18) and (B.4.6), it is easy to derive the
2
estimate
ZT0 Z
ess sup kv(·, t)k2L2,unif + sup |∇v|2 dxdt ≤ 2A. (B.4.9)
0<t<T0 x0 ∈R3
0 B(x0 ,1)
Now, by (B.4.7) and (B.4.9), we see that, for each n ∈ N,

Z
the f unction t 7→ v(x, t) · w(x) dx is continuous on [0, T0 ] (B.4.10)
B(n)
for any w ∈ L2 (B(n)).

Next, we note that for the solution of the regularised problem we have
the following identity:
Z Zt Z Z
{k} {k} 2
ϕ(x)|v 2
(x, t)| dx + 2 ϕ|∇ v | dxds = ϕ|a{k} |2 dx
R3 0 R3 R3
Zt Z h
+ |v {k} |2 ∆ ϕ + ∇ ϕ · F{k} (v {k} ) |v {k} |2 + 2p{k}
n (B.4.11)
0 R3
i
{k} {k}
+2ϕg ·v dxds,
which is valid for any function ϕ ∈ C0∞ (R3 ). Taking into account (B.3.4),
(B.4.6)–(B.4.8), and (B.4.10), we deduce from (B.4.11) the inequality
Z Zt Z Z
2 2
ϕ(x)|v(x, t)| dx + 2 ϕ|∇ v| dxds ≤ ϕ|a|2 dx
R3 0 R3 R3
Zt Z h
2 2
+ |v| ∆ ϕ + ∇ ϕ · v |v| + 2p (B.4.12)
0 R3
i
+2ϕf · v dxds.
The latter holds for any t ∈ [0, T0 ] and for any nonnegative function ϕ ∈
C0∞ (R3 ). On the other hand, from (B.4.10) and from (B.4.12) it follows that
Z
ϕ|v(x, t) − a(x)|2 dx → 0 as t → +0 (B.4.13)
R3
for all ϕ ∈ C0∞ (R3 ). So, v meets (B.1.7). The validity of (B.1.8) follows from
(B.4.6). It remains to establish decomposition (B.1.9).
Thanks to (B.3.18), we have
ZT0 Z
3
|p̂εx0 ,2 | 2 dxdt ≤ A. (B.4.14)
0 B(x0 ,3/2)
We would like to emphasize that the constant on the right hand sides of
{k}
(B.4.14) is independent of ε and x0 . Let p̂x0 ,2 be the sequence generated by
{k},x0
v {k} via (B.3.7). For each x0 ∈ R3 , we can find subsequences p̂x0 ,2 and
v {k},x0 such that
{k},x0
p̂x0 ,2 * px0 in L 3 (B(x0 , 3/2)×]0, T0 [).
2
So, it follows from (B.4.14) that

ZT0 Z
3
|px0 | 2 dxdt ≤ A (B.4.15)
0 B(x0 ,3/2)
for each x0 ∈ R3 . Passing to the limit in the Navier-Stokes equations on the

set B(x0 , 3/2)×]0, T0 [, we show that
∇(p − px0 ) = 0
on B(x0 , 3/2)×]0, T0 [ in the sense of distributions. So, we state that, for any
x0 ∈ R3 , there exists a function cx0 ∈ L 3 (0, T0 ) having the property
2
p(x, t) − px0 (x, t) = cx0 (t) (B.4.16)

for x ∈ B(x0 , 3/2) and t ∈]0, T0 [. Now, let us show the validity of (B.1.9).
Using decomposition (B.3.7) and the theory of singular integrals, we observe
1{k},x0
px0 ,2 → p1x0 in L 3 (B(x0 , 3/2)×]0, T0 [).
2
Obviously,
2{k},x
px0 ,2,R0 → p2x0 ,R in L 3 (B(x0 , 3/2)×]0, T0 [),
2
where
Z
1
p2x0 ,R (x, t) = (K(x − y) − K(x0 − y)) : v(y, t) ⊗ v(y, t) dx.
4π
B(x0 ,2R)\B(x0 ,2)
By (B.3.12), we also have

3{k},x 1
sup |px0 ,R 0 (x, t)| ≤ C(A, T0 )
x∈B(x0 ,3/2) R
on ]0, T0 [. On the other hand, the integral in the definition of p2x0 ,R converges
to p2x0 as R → +∞. This follows from the Lemma 3.1 and the inequality
Z
|K(x − y) − K(x0 − y)||v(y, t)|2 dx ≤ ckv(·, t)k2L2,unif
B(x0 ,2R)\B(x0 ,2)
which is valid on B(x0 , 3/2)×]0, T0 [. Passing to the limit as R → +∞, we

show (B.1.11). Proposition 1.6 is proved.
B.5 Proof of Theorem 1.5

The idea of the proof of the main theorem is the same in [35]. It is based on
the theory of solvability of the Cauchy problems for the Stokes and Navier-
◦
Stokes systems with initial data from E 3 . We shall formulate corresponding
results below and their proofs will be given elsewhere. We start with the
most important part: linear theory. Consider the Cauchy problem for the
Stokes system:
∂t v(x, t) − ∆ v(x, t) + ∇ p(x, t) = −div f (x, t) + g(x, t),

(B.5.1)
div v(x, t) = 0
for x ∈ R3 and 0 < t < T ,
v(x, 0) = a(x), x ∈ R3 . (B.5.2)
It is supposed that
◦ ◦
a ∈ E 3, g ∈ G3 (0, T ), f ∈ G 5 (0, T ). (B.5.3)
2
Theorem 5.1. Assume that conditions (B.5.3) hold. There exists a unique
pair of functions v and p having the following properties:
p
v ∈ L∞ (0, T ; L3,unif ), (1 + |v|)|∇v| ∈ L2,unif (0, T ),
p ∈ L 5 (0, T ; L 5 ,loc (R3 ));

2 2
v and p satisf y (B.5.1) in the sense of distributions;

h i
kvkL∞ (0,T ;L3,unif ) ≤ c kf kL 5 ,unif (0,T ) + kgkL3,unif (0,T ) + kakL3,unif ;
2
Z
the t 7→ v(x, t) · u(x) dx is continuous on [0, T ] f or any compactly
R3
supportedf unctions u ∈ L 3 (R3 );

2
◦ ◦
v ∈ G5 (0, T ), v(·, t) ∈E 3 , ∀t ∈ [0, T ];
kv(·, t) − v(·, t0 )kL3,unif → 0 as t → t0 + 0, ∀t0 ∈ [0, T ],
v(·, 0) = a(·);
f or any x0 ∈ R3 , there exists cx0 ∈ L 5 (0, T ) such that

2
p(x, t) − cx0 (t) = p1x0 (x, t) + p2x0 (x, t)

for any x ∈ B(x0 , 3/2) and for any t ∈]0, T [, where
Z
1 1 1
px0 (x, t) = − trf (x, t) + K(x − y) : f (y, t) dy,
3 4π
B(x0 ,2)
Z
1
p2x0 (x, t) = (K(x − y) − K(x0 − y)) : f (y, t) dy
4π
R3 \B(x0 ,2)
and
ZT Z
5
sup |p(x, t) − cx0 (t)| 2 dxdt < +∞,
x0 ∈R3
0 B(x0 ,3/2)
ZT Z
5
sup I{|x|>R} |p(x, t) − cx0 (t)| 2 dxdt → 0
x0 ∈R3
0 B(x0 ,3/2)
as R → +∞.
Using Theorem 5.1 and successive approximations, see, for example, [13]
and [17], we can prove the following theorems about solvability of the Cauchy
problem:
∂t v(x, t) + div v(x, t) ⊗ v(x, t) − ∆ v(x, t) + ∇ p(x, t) = g(x, t),

(B.5.4)
div v(x, t) = 0
for x ∈ R3 and 0 < t < T ,
v(x, 0) = a(x), x ∈ R3 , (B.5.5)
under assumptions that

◦ ◦
a ∈ E 3, g ∈ G3 (0, T ). (B.5.6)
Theorem 5.2. Suppose that conditions (B.5.6) hold. There exists a number
T0 ∈]0, T [ with the following property. Given a and g, there exists a pair of
functions v and p, forming a local energy solutions in the space-time cylinder
QT0 = R3 ×]0, T0 [, such that
◦ ◦
v ∈ C([0, T0 ]; E 3 ) ∩ G5 (0, T0 ),
p
|v||∇v| ∈ L2,unif (0, T0 ), p ∈ L 5 (0, T0 ; L 5 ,loc (R3 ));
2 2
ZT0 Z
5
sup |p(x, t) − cx0 (t)| 2 dxdt < +∞,
x0 ∈R3
0 B(x0 ,3/2)
ZT0 Z
5
x0 ∈R3
0 B(x0 ,3/2)
as R → +∞.
Moreover, assume that a pair u and q is a local energy solution to the
Cauchy problem
∂t u(x, t) + div u(x, t) ⊗ u(x, t) − ∆ u(x, t) + ∇ q(x, t) = g(x, t),

(B.5.7)
div u(x, t) = 0
for x ∈ R3 and 0 < t < T0 ,
u(x, 0) = a(x), x ∈ R3 . (B.5.8)
Then, u = v.
Theorem 5.3. Suppose that conditions (B.5.6) hold. Given T > 0, there
exists a constant ε(T ) with the following property. If
kgkL3,unif (0,T ) + kakL3,unif ≤ ε(T ),
then there exists a pair of functions v and p, forming a local energy solutions
in the space-time cylinder QT , such that:
◦ ◦
v ∈ C([0, T ]; E 3 ) ∩ G5 (0, T ),
p
|v||∇v| ∈ L2,unif (0, T ), p ∈ L 5 (0, T ; L 5 ,loc (R3 ));
2 2
kvkL∞ (0,T ;L3,unif ) ≤ cε(T );

ZT Z
5
sup |p(x, t) − cx0 (t)| 2 dxdt < +∞,
x0 ∈R3
0 B(x0 ,3/2)
ZT Z
5
x0 ∈R3
0 B(x0 ,3/2)
as R → +∞.
Moreover, assume that a pair u and q is a local energy solution to the
Cauchy problem (B.5.4), (B.5.5). Then, u = v.
Now, let us outline the proof of Theorem 1.5.
So, according to Proposition 1.6, we can find a number T0 ∈]0, T ] and a
pair of functions v and p that are a local energy solution in the space-time
cylinder QT0 . If T0 = T , then we are done. Assume that it is not. By
Theorem 1.4, we can find t0 ∈]0, T0 [ so that
◦
v(·, t0 ) ∈E 3 ,
v and p are a local energy solution in R3 ×]t0 , T0 [.
Next, there exist T1 and a pair of functions u and q which is a local
◦
energy solution in R3 ×]t0 , T1 [ and u ∈ C([t0 , T1 ]; E 3 ) with u(·, t0 ) = v(·, t0 ).
However, we know that there must be v = u in R3 ×]t0 , T1 [. Without loss of
generality, we may assume that T1 < T . Using density of smooth functions,
let us decompose v(·, t0 ) = a1 + a2 and g = g1 + g2 so that
kg1 kL3,unif (0,T ) + ka1 kL3,unif ≤ ε(T − t0 ),
◦
a2 ∈ C ∞ 3
0 (R ),
and g2 is a function of class C ∞ in QT and there exists R2 > 0 such that the
support of g2 (·, t) lies in B(R2 ) for all t ∈]t0 , T [. According to Theorem 5.3,
there exists a pair u1 and q1 which is a local energy solutions to the Cauchy
problem:
∂t u1 (x, t) + div u1 (x, t) ⊗ u1 (x, t) − ∆ u1 (x, t) + ∇ q1 (x, t) = g1 (x, t),
div u1 (x, t) = 0
for x ∈ R3 and t0 < t < T ,
u1 (x, t0 ) = a1 (x), x ∈ R3 .
Moreover,
ku1 kL∞ (t0 ,T ;L3,unif ) ≤ cε(T − t0 ). (B.5.9)
We seek functions u2 and q2 , solving the following Cauchy problem:
∂t u2 (x, t) + div (u2 (x, t) ⊗ u2 (x, t) + u1 (x, t) ⊗ u2 (x, t)+
u2 (x, t) ⊗ u1 (x, t)) − ∆ u2 (x, t) + ∇ q2 (x, t) = g2 (x, t), div u2 (x, t) = 0
u2 (x, t0 ) = a2 (x), x ∈ R3 .
We state that this problem has a weak Leray-Hopf solution with the finite
global energy satisfying the local energy inequality. To see that it is really
possible, let us comment the crucial term in proving a priori global energy
estimate. This term has the form
Zt Z
I0 = (u2 ⊗ u1 + u1 ⊗ u2 ) : ∇u2 dxds.
t0 R3
So, we need to estimate the integral

Zt Z
I= |u1 |2 |u2 |2 dxds.
t0 R3
To this end, we fix x0 ∈ R3 and apply successively Hölder’s and Gagliardo’s

inequalities, and estimate (B.5.9):
Zt Z Zt Z 32 Z 13
2 2 3 6
|u1 | |u2 | dxds ≤ |u1 | dx |u2 | dx ds
t0 B(x0 ,1) t0 B(x0 ,1) B(x0 ,1)
Zt Z
≤ cku1 k2L∞ (t0 ,T ;L3,unif ) |∇u2 |2 + |u2 |2 dxds
t0 B(x0 ,1)
Zt Z
2 2 2
≤ cε (T − t0 ) |∇u2 | + |u2 | dxds.
t0 B(x0 ,1)
Using Besicovitch covering lemma, we can easily show

Zt Z
I ≤ cε2 (T − t0 ) |∇u2 |2 + |u2 |2 dxds
t0 R3
and therefore
Zt Z
I0 ≤ cε(T − t0 ) |∇u2 |2 + |u2 |2 dxds.
t0 R3
This latter allows us to hide I0 into the left hand side of the global energy
inequality by choosing ε(T − t0 ) sufficiently small and to find
Z Zt Z Z Zt Z
|u2 (x, t)|2 dx + |∇u2 |2 dxds ≤ |a2 (x)|2 dx + 2 g2 · u2 dxds.
R3 t0 R3 R3 t0 R3
Using this estimate and suitable approximations, we can easily prove our
statements about u2 . In addition, all above arguments show that pressure q2
may be taken in the form
1
q2 = K ∗ (u2 ⊗ u2 + u1 ⊗ u2 + u2 ⊗ u1 )
4π
and, moreover,
q2 ∈ L 3 (R3 ×]t1 , T [)
2
and we have representation:

1 2
q2 (x, t) − c2x0 (t) = q2x 0
(x, t) + q2x0
(x, t)
for x0 ∈ B(x0 , 3/2) and t ∈]t0 , T [, where
1 1
q2x 0
= − (|u2 (x, t)|2 + 2u1 (x, t) · u2 (x, t))+
3
Z
1
+ K(x − y) : (u2 ⊗ u2 + u1 ⊗ u2 + u2 ⊗ u1 )(y, t)dy,
4π
B(x0 ,2)
B.6. DENSITY 277
Z
2 1
q2x = (K(x−y)−K(x0 −y)) : (u2 ⊗u2 +u1 ⊗u2 +u2 ⊗u1 )(y, t)dy,
0
4π
R3 \B(x0 ,2)
Z
1
c2x0 (t) = K(x0 − y) : (u2 ⊗ u2 + u1 ⊗ u2 + u2 ⊗ u1 )(y, t)dy.
4π
R3 \B(x0 ,2)
Now, we let
u = u1 + u2 , q = q1 + q2 .
Our task is to verify that this new pair forms a local energy solution to the
Cauchy problem:
∂t u(x, t) + div u(x, t) ⊗ u(x, t) − ∆ u(x, t) + ∇ q(x, t) = g(x, t),
div u(x, t) = 0
u(x, t0 ) = a(x), x ∈ R3 .
The most difficult part of this task is to show that u and q satisfy the local
energy inequality. It can be done essentially in the same way as the corre-
◦
sponding part of the proof of the uniqueness for C([0, T ]; E 3 )-solutions. And
this immediately implies that u = v in the R3 ×]t0 , T1 [. Since p(x, t)−q(x, t) =
c(t) ∈ L 3 (t0 , T1 ), we can change function cx0 in a suitable way and assume
2
that q = p in the R3 ×]t0 , T1 [. So, the pair u and q can be regarded as a
required extension of v and p to the whole space-time cylinder QT . Theorem
1.5 is proved.
B.6 Density
◦ ◦
Lemma 6.1. For any f ∈E m and for any ε > 0, there exists fε ∈ C ∞ 3
0 (R )
such that
kf − fε kLm,unif < ε. (B.6.1)
Proof Let
Bk = B(xk , 2), xk ∈ Z3 .
Bk = R3 . Moreover, there is a universal constant N2 with the

S
Clearly
k
following property:
for any x0 ∈ R3 , the ball B(x0 , 1) is covered by not more
than N2 balls Bk ,
For this covering, we can find a partition of unity such that
X
ϕk ∈ C0∞ (R3 ), spt ϕk ⊂ Bk , ϕk ≥ 0, ϕk = 1.
k
Now, given R > 0, we introduce two smooth cut-off functions
χ(x) = 1, x ∈ B(1), spt χ ⊂ B(2), χR (x) = χ(x/R).

◦
We fix a ball Bk . There exists a functions v k ∈W 12 (Bk ; R3 ) that is a solution
of the equation
Z
k 1
div v = f · ∇ χR ϕk − f · ∇ χR ϕk dx
|B(2)|
Bk
and satisfies the estimate

c c
kv k kBk ≤ kf kBk ≤ kf kLm,unif
R R
with a universal constant c. Extending v k by zero to the whole R3 , we set
X
vR = vk
k
and observe that, for each R > 0, the function v R has a compact support
and, moreover,
div v R = f · ∇ χR in R3 ,
X cN2
kv R kB(x0 ,1) ≤ kv k kB(x0 ,1) ≤ kf kLm,unif , ∀x0 ∈ R3 .
k
R
Next, we let
uR = f χR − v R .
B.7. COMMENTS 279
Obviously, we have
div uR = 0 in R3 ,
and uR has a compact support. Since f ∈ E2 , we see that, for an arbitrary
ε > 0, we can find R > 0 such that
kf − uR kLm,unif ≤ kf − f χR kLm,unif + kv R kLm,unif
cN2
kf kLm,unif < ε.
≤ kf − f χR kLm,unif +
R
To complete the proof of the lemma, it is enough to smooth uR which is
easy. Lemma 6.1 is proved.
B.7 Comments
The main source for the content of Appendix B is the monograph of P.-
G. Lemarie-Riesset [35]. Our interpretation of his results is given in the
paper [25] and we follow it here. We wish to emphasize that the Lemarie-
Riesset conception of local energy weak Leray-Hopf solution1 is heavily used
in Chapter 7.
1
In fact, G.-P. Lemarie-Riesset himself calls them simply local Leray solutions
Bibliography
[1] Bogovskii, M. E. Solution of some problems of vector analysis with the

operators Div and Grad , in: Trudy S. L. Sobolevs Seminar, 80 (1980),
pp. 540.
[2] Caffarelli, L., Kohn, R.-V., Nirenberg, L., Partial regularity of suit-
able weak solutions of the Navier-Stokes equations, Comm. Pure Appl.
Math., Vol. XXXV (1982), pp. 771–831.
[3] Cannone, M., Harmonic analysis tools for solving the incompressible
Navier-Stokes equations, Handbook of Mathematical Fluid Dynamics,
vol. 3, 2002.
[4] Chen, C.-C., Strain, R. M., Tsai, T.-P., Yau, H.-T., Lower bound on
the blow-up rate of the axisymmetric Navier-Stokes equations, Int.
Math. Res. Not., (2008) Vol. 2008 : article ID rnn016, 31 pages,
doi:10.1093/imrn/rnn016.
[5] Chen, C.-C., Strain, R. M., Tsai, T.-P., Yau, H.-T., Lower bounds on
the blow-up rate of the axisymmetric Navier-Stokes equations II, Com-
munications in Partial Differential Equations, 34: 203–232, 2009.
[6] Chen Z.-M., Price W. G., Blow-up rate estimates for weak solutions of
the Navier–Stokes equations, R. Soc. Lond. Proc. Ser. A Math. Phys.
Eng. Sci. 457 (2001), 2625–2642.
[7] Choe, H. L., Lewis, J. L., On the singular set in the Navier-Stokes
equations, J. Functional Anal., 175(2000), pp. 348–369.
[8] Constantin, P., Foias, C., Navier-Stokes Equations, Chicago Lecture in

Mathematics, 1988, 200p.
281
282 BIBLIOGRAPHY
[9] Escauriaza, L., Carleman inequalities and the heat operator,Duke Math-
ematical Journal, 104(2000), pp. 113 – 126.
[10] L. Escauriaza, c. E. Kenig, G. Ponce, L. Vega, Decay at infinity of caloric
functions within characteristic hyperplanes, Math. Res. Lett., 13 (2006),
441-453.
[11] Escauriaza,L., Seregin, G., Šverák, V., On backward uniqueness for
parabolic equations, Zap. Nauchn. Seminar. POMI, 288(2002), 100–103.
[12] Escauriaza,L., Seregin, G., Šverák, V., On backward uniqueness for
parabolic equations, Arch. Rational Mech. Anal., 169(2003)2, 147–157.
[13] Escauriaza, L., Seregin, G., Šverák, V., L3,∞ -solutions to the Navier-
Stokes equations and backward uniqueness, Russian Mathematical Sur-
veys, 58(2003)2, pp. 211-250.
[14] Escauriaza, L., Vega, L., Carleman inequalities and the heat operator
II, Indiana university Mathematics Journal, 50(2001), pp. 1149 – 1169.
[15] Fefferman, C., http://www.claymath.org/millennium/Navier-Stokes
equations.
[16] Galdi, G. P., An Introduction to the Mathematical Theory of the
Navier-Stokes Equations: Steady-State Problems, Springer Monographs
in Mathematics, Second edition, 2011, 1032p.
[17] Galdi, G.P., An introduction to the Navier-Stokes initial-boundary
value problem, Fundamental directions in mathematical fluid mechan-
ics/Giovanni P. Galdi...ed.,-Basel; Boston; Berlin: Birkhäuser 2000, pp.
1-70.
[18] Giga, Y., Solutions for semilinear parabolic equations in Lp and regu-
larity of weak solutions of the Navier-Stokes equations, J. of Diff. Equa-
tions, 62(1986), pp. 186–212.
[19] Gustafson, S., Kang, K., Tsai, T.-P., Interior regularity criteria for suit-
able weak solutions of the Navier-Stokes equations, Commun. Math.
Phys., 273(2007), 161–176.
[20] Hopf, E., Über die Anfangswertaufgabe für die hydrodynamischen
Grundgleichungen, Math. Nachrichten, 4 (1950-51), 213–231.
BIBLIOGRAPHY 283
[21] Hörmander, L., Linear Partial Differential Operators, Springer Verlag,

1963.
[22] Kang, K., Unbounded normal derivative for the Stokes system near
boundary, Math. Ann. 331(2005), pp. 87–109.
[23] Kapitanskii, L. V. and Pileckas, K. I, Some problems of vector analysis,

Zap. Nauchn. Semin. LOMI, 138, 6585 (1984).
[24] Kato, T., Strong Lp -solutions of the NavierStokes equation in Rm , with

applications to weak solutions. Math. Z., 187 (1984), 471-480.
[25] Kikuchi, N., Seregin, G., Weak solutions to the Cauchy problem for
the Navier-Stokes equations satisfying the local energy inequality. AMS
translations, Series 2, 220, 141-164.
[26] Kim H., Kozono H., Interior regularity criteria in weak spaces for the
Navier-Stokes equations, manuscripta math., 115 (2004), 85–100.
[27] Koch, G., Nadirashvili, N., Seregin, G., Sverak, V., Liouville theorems
for the Navier-Stokes equations and applications, Acta Mathematica,
203 (2009), 83–105.
[28] Koch, H. Tataru, D., Well-posedness for the NavierStokes equations.

Adv. Math., 157 (2001), 2235.
[29] Ladyzhenskaya, O. A., Mathematical problems of the dynamics of vis-

cous incompressible fluids, 2nd edition, Nauka, Moscow 1970.
[30] Ladyzhenskaya, O. A., On unique solvability of the three-dimensional

Cauchy problem for the Navier-Stokes equations under the axial sym-
metry, Zap. Nauchn. Sem. LOMI 7(1968), 155-177.
[31] Ladyzhenskaya, O. A., Seregin, G. A., On partial regularity of suit-

able weak solutions to the three-dimensional Navier-Stokes equations,
J. math. fluid mech., 1(1999), pp. 356-387.
[32] Ladyzhenskaya, O. A. and Solonnikov, V.A. Some problems of vector

analysis and generalized settings of boundary-value problems for the
NavierStokes equations, Zap. Nauchn. Semin. LOMI, 59, 81116 (1976).
284 BIBLIOGRAPHY
[33] Ladyzhenskaya, O. A., Solonnikov, V. A., Uralt’seva, N. N., Linear and

quasi-linear equations of parabolic type, Moscow, 1967; English trans-
lation, American Math. Soc., Providence 1968.
[34] Ladyzhenskaya, O. A., Uraltseva,N. N., Linear and quasilinear equations

of elliptic type, ”Nauka”, Moscow, 1973.
[35] Lemarie-Riesset, P. G., Recent developemnets in the Navier-Stokes prob-

lem, Chapman&Hall/CRC reseacrh notes in mathematics series, 431.
[36] Leray, J., Sur le mouvement d’un liquide visqueux emplissant l’espace,
Acta Math. 63(1934), pp. 193–248.
[37] Lions, P. L., Mathematical Topics in Fluid Mechanics: Volume 1: In-

compressible Models, Oxford University Press, 1996.
[38] Lin, F.-H., A new proof of the Caffarelly-Kohn-Nirenberg theorem,

Comm. Pure Appl. Math., 51(1998), no.3, pp. 241–257.
(k)
[39] Necas, J. Sur les normes equivalentes dans Wp (Ω) et sur la coercitivite
des formes formellement positive, in ”Equations aux derivees partielles,
Presses de l’Universite de Montreal, 1966.
[40] J. Necas, M. Ruzicka, V. Šverák, On Leray’s self-similar solutions of

the Navier-Stokes equations. Acta Math. 176 (1996), 283–294.
[41] Rusin, W., Sverak, V., Miminimal initial data for potential Navier-
Stokes singularities, arXiv:0911.0500.
[42] Scheffer, V., Partial regularity of solutions to the Navier-Stokes equa-

tions, Pacific J. Math., 66(1976), 535–552.
[43] Scheffer, V., Hausdorff measure and the Navier-Stokes equations, Com-
mun. Math. Phys., 55(1977), pp. 97–112.
[44] Scheffer, V., The Navier-Stokes equations in a bounded domain, Com-

mun. Math. Phys., 73(1980), pp. 1–42.
[45] Scheffer, V., Boundary regularity for the Navier-Stokes equations in a

half-space, Commun. Math. Phys., 85(1982), pp. 275–299.
BIBLIOGRAPHY 285
[46] Seregin, G.A., Interior regularity for solutions to the modified Navier-
Stokes equations, J. of math. fluid mech., 1(1999), no.3, pp. 235–281.
[47] Seregin, G., Local regularity theory of the Navier-Stokes equations,
Handbook of Mathematical Fluid Mechanics, Vol. 4, Edited by Fried-
lander, D. Serre, pp, 159-200.
[48] Seregin, G. A., Estimates of suitable weak solutions to the Navier-Stokes
equations in critical Morrey spaces, J. Math. Sci. 143:2 (2007), 2961–
2968.
[49] Seregin, G., A note on local boundary regularity for the Stokes system,
Zapiski Nauchn. Seminar. POMI., v. 370(2009), pp. 151-159.
[50] Seregin, G., Local regularity for suitable weak solutions of the Navier-
Stokes equations, Russian Math. Surveys 62:3 595–614.
[51] Seregin, G. A. On the number of singular points of weak solutions to
the Navier-Stokes equations, Comm. Pure Appl. Math., 54(2001), issue
8, pp. 1019-1028.
[52] Seregin, G.A., Local regularity of suitable weak solutions to the Navier-
Stokes equations near the boundary, J. math. fluid mech., 4(2002),
no.1,1–29.
[53] Seregin, G.A., Differentiability properties of weak solutions to the
Navier-Stokes equations, Algebra and Analysis, 14(2002), No. 1, pp.
193-237.
[54] Seregin, G.A., Navier-Stokes equations: almost L3,∞ -cases, Journal of
mathematical fluid mechanics, 9(2007), pp. 34-43.
[55] Seregin, G., A note on necessary conditions for blow-up of energy solu-
tions to the Navier-Stokes equations, Progress in Nonlinear Differential
Equations and Their Applications, 2011 Springer Basel AG, Vol. 60,
631-645. 98.
[56] Seregin, G., Necessary conditions of potential blow up for the Navier-
Stokes equations, Zapiski Nauchn.Seminar. POMI, 385(2010), 187-199.
[57] Seregin, G., A certain necessary condition of potential blow up for
Navier-Stokes equations, Commun. Math. Phys. 312, 833-845.
286 BIBLIOGRAPHY
[58] Seregin, G., Selected topics of local regularity theory for the Navier-
Stokes equations. Topics in mathematical fluid mechanics, 239313, Lec-
ture Notes in Math., 2073, Springer, Heidelberg, 2013.
[59] Seregin, G., Šverák, V., On Type I singularities of the local axi-
symmetric solutions of the Navier-Stokes equations, Communications
in PDE’s, 34(2009), pp. 171-201.
[60] Seregin, G., Šverák, V., On a bounded shear flow in half space, Zapiski
Nauchn. Seminar POMI, 385(2010), pp. 200-205.
[61] Seregin, G., Zajaczkowski, W., A sufficient condition of local regular-

ity for the Navier-Stokes equations, Zapiski Nauchn. Seminar, POMI,
336(2006), pp. 46-54.
[62] Seregin, G., Zajaczkowski, W., A sufficient condition of regularity for

axially symmetric solutions to the Navier-Stokes equations, SIMA J.
Math. Anal., (39)2007, pp. 669–685.
[63] Serrin, J., On the interior regularity of weak solutions of the Navier-
Stokes equations, Arch. Ration. Mech. Anal., 9(1962), pp. 187–195.
[64] Stein, E., Singular Integrals and Differentiabilty Properties of Functions,

Princeton University Press, Princeton, New Jersey, 1970.
[65] Struwe M., On partial regularity results for the Navier-Stokes equations,
Comm. Pure Appl. Math., 41 (1988), pp. 437–458.
[66] Takahashi S., On interior regularity criteria for weak solutions of the
Navier-Stokes equations, Manuscripta Math., 69 (1990), 237-254.
[67] Tataru, D., Carleman estimates, unique continuation, and applications,

notes downloadable from http://math.berkeley.edu/ tataru/ucp.html
[68] Temam, R., Navier-Stokes Equations: Theory and Numerical Analysis,

American Mathematical Soc., 2001, 408 p.
[69] Ukhovskij, M. R., Yudovich, V. L., Axially symmetric motions of ideal

and viscous fluids filling all space, Prikl. Mat. Mech. 32 (1968), 59-69.

Lecture Notes 156.01556

Uploaded by

Copyright:

Available Formats

Lecture Notes 156.01556

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Lecture Notes 156.01556

Uploaded by

Copyright:

Available Formats

Lecture Notes OxPDE-14/01

Lecture Notes on Regularity Theory for the

Oxford Centre for Nonlinear PDE

5.3. Weak Leray-Hopf solutions

B.5. Proof of Theorem 1.5

and $% is a standard mollifier. It is easy to check that l is a bounded

for all ψ ∈ L1 (Ω0 ). Next, for any ϕ ∈ C0∞ (Ω0 ) we have

On the other hand, by known properties of mollification we find

The latter means that

ku%0 ks,Ω0 ≤ c(n, s, Ω0 )kg % ks,Ω0 ≤ c(n, s, Ω0 )kgks,Ω .

Without loss of generality, we may assume that

in Ls (Ω0 ). And thus

Now, let us described a required function u. We let u = u0 in Ω0 . Then,

is NOT true. To see that, let us consider a standard cut-off function ϕ ∈

ϕ(x) = 1 x ∈ B(1), ϕ(x) = 0 x ∈

u1 ∼ u2 ⇔ ku1 − u2 kLks (Ω) = 0.

Proposition 1.4. For bounded domains Ω, the Friedrichs inequality is valid:

kukWsk (Ω) ≤ ckukLks (Ω)

for any u ∈ C0∞ (Ω) with a positive constant independent of u, i.e.,

1.2 Newtonian Potential

if n = 2, where ωn is the volume of unit ball in Rn .

Proposition 2.1. Let f ∈ Lp (Rn ) with 1 < p < ∞ and u = E ? f . The

as R → ∞ for each fixed m.

k∇2 u − ∇2 um kp,Rn ≤ ckf − fm kp,Rn → 0

as R → ∞. This implies (ii).

u(x0 , xn ) = −u(x0 , −xn )

1.3 Equation div u = b

(ii) k∇uks,Rn ≤ c(s, n)kbks,Rn .

k∇uks,Rn = k∇2 hks,Rn ≤ c(s, n)kbks,Rn

Proposition 3.2. Let 1 < s < ∞. Given b ∈ Ls (Rn+ ), there exists u ∈

(ii) k∇uks,Rn+ ≤ c(s, n)kbks,Rn+ .

Remark 3.3. The above vector-valued function u satisfies the homogeneous

Proof of Proposition 3.2 To show the essence of the matter, let us

4h = b̂ = b in R3+ , h,3 |x3 =0 = 0 (1.3.1)

where A is unknown vector field. Obviously,

So, the main equations for A = (A1 , A2 , 0) are:

Aα,3 (x0 , 0) = Bα (x0 ) α = 1, 2,

where a function K ∈ C0∞ (R2 ) is supposed to obey the following conditions:

Now, our aim to show that functions

By the choice of K, this immediately implies Aα,3 (x0 , 0) = Bα (x0 ), α = 1, 2.

k∇2 Aα k2,R3+ ≤ ckbk2,R3+ , α = 1, 2, (1.3.4)

for some universal constant c.

with some universal constant c.

Proof By the shift in variables,

Applying the triangle inequality, we find

|f (x0 + z 0 , 0) − f (x0 , 0)| ≤ |f (x0 + z 0 , |z 0 |) − f (x0 , |z 0 |)|+

+|f (x0 + z 0 , |z 0 |) − f (x0 + z 0 , 0)| + |f (x0 , |z 0 |) − f (x0 , 0)|. (1.3.6)

with 1 < p < ∞. Then

Now, let us treat I2 , using the same arguments. As a result, we have

To estimate the third term, we exploit the following simple inequality

It remains to make the change of variables z 0 = (% cos ϕ, % sin ϕ)

and complete our proof.

|fR (x0 , 0) − fR (y 0 , 0)|2 0 0

Passing to the limit as R → ∞ and using Fatou’s lemma, we establish (1.3.5).

k∇2 Aα k2,R3+ ≤ ck∇h(·, 0)k 1 . (1.3.8)

k∇h(·, 0)k 1 ≤ ck∇2 hk2,R3+ ≤ ckbk2,R3+ .

and, by Hölder inequality,

Introducing polar coordinates y 0 = %(cos ϕ, sin ϕ), we find

If we set z 0 = x3 (cos ϕ, sin ϕ), then