See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/226154720

On the Existence of the Pressure for Solutions of the Variational Navier--

Stokes Equations

Article  in  Journal of Mathematical Fluid Mechanics · January 1999

DOI: 10.1007/s000210050010

CITATIONS READS

50 266

1 author:

Jacques Simon
University of Nice Sophia Antipolis
85 PUBLICATIONS   3,656 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Functional spaces and analysis View project

Domain variations for pde's solutions View project

All content following this page was uploaded by Jacques Simon on 09 February 2015.

The user has requested enhancement of the downloaded file.

 On the existence of the pressure for solutions
of the variational Navier–Stokes equations
Jacques SIMON
Preprint: September 1998

Abstract
We prove that, for a right-hand side valued into V 0 , it is not possi-
ble to associate a pressure to the velocity given by the weak equation
in order to satisfy the Navier–Stokes equations.

1 Introduction
Existence of a solution (u, p) of the Navier–Stokes equations in the space-time
domain Ω × (0, T ), Ω ⊂ IRd , d ≥ 2, has been proved for a right-hand side f in
L2 ((0, T ); (L2 (Ω))d ) by many authors, starting with the fundamental work of
J. Leray [3]; see, e.g., [1], [2], [4]. The methods used are more or less similar
and consist of two steps: first, one determines the existence of a velocity u in
L2 ((0, T ); V ) (weak solution) by solving a weak formulation where the pres-
sure p is eliminated, and, subsequently, one finds the corresponding pressure
by using de Rham’s theorem or similar arguments. The proof easily extends
to the case when f belongs to L1 ((0, T ); (L2 (Ω))d ) or L2 ((0, T ); (H −1 (Ω))d ).
As noticed by J.-L. Lions [5], see also L. Tartar [9] or R. Temam [10],
the existence of a weak solution can be more generally proved for f in
L2 ((0, T ); V 0 ). The question is then wether or not in this situation there
exists a corresponding pressure field. In this note we prove here that, in
general, the answer is negative.
0
CNRS, Laboratoire de Mathématiques Appliquées (UMR 6620),
Université Blaise Pascal (Clermont-Ferrand 2), 63177 Aubière cedex, France.
simon@ucfma.univ-bpclermont.fr

1
 Our argument is based on the following consideration. The left-hand side
of the equation should lie in the space H −1 ((0, T ); (H −1 (Ω))d ) while the right-
hand side is given in L2 ((0, T ); V 0 ). This equality can be meaningful only if
these two spaces could be imbedded in the same Hausdorff space. But this
is not possible since, by Theorem 1, (H −1 (Ω))d and V 0 themselves cannot be
imbedded in the same Hausdorff space.

2 Topological properties of V 0
Let Ω be a bounded open set in IRd which boundary is connected and is
locally the graph of a Lipschitz function.
We denote D(Ω) = {v ∈ C ∞ (Ω) : support v ⊂ Ω}, D0 (Ω) its dual space,
H (Ω) = {v ∈ L2 (Ω) : ∂v/∂xi ∈ L2 (Ω), 1 ≤ i ≤ d}, H01 (Ω) the closure of
1

D(Ω) in H 1 (Ω) and H −1 (Ω) its dual space. Dual always means topological
dual equipped with the strong topology and X 0 denotes the dual of a space
X.
We denote V = {v ∈ (D(Ω))d : ∇ . v = 0}, V and H its closure respec-
tively in (H 1 (Ω))d and (L2 (Ω))d . We denote by . the scalar product in IRd
thus ∇ . v = i ∂vi /∂xi and v . ϕ = i vi ϕi .
P P

Since V ⊂ H (⊂ denotes a topological imbedding) and V is dense in H,

H 0 ⊂ V 0 . Following J.–L. Lions [5], p. 66, H may be identified with H 0 by
Riesz’s theorem, and then
H ⊂ V 0. (1)
On the other hand L2 (Ω) ⊂ H −1 (Ω) ⊂ D0 (Ω) thus

H ⊂ (H −1 (Ω))d . (2)

We are looking for a Hausdorff topological vector space E such that

V 0 ⊂ E, (H −1 (Ω))d ⊂ E. (3)

Remind that a Haussdorff space is a space possessing the separation property.

The precise meaning of imbeddings (1) and (2) is that, to each v ∈ H, we
associate i(v) ∈ V 0 and j(v) ∈ (H −1 (Ω))d by
Z
hi(v), ϕiV 0 ×V = v . ϕ, ∀ϕ ∈ V, (4)
ZΩ
hj(v), ϕi(H −1 (Ω))d ×(H01 (Ω))d = v . ϕ, ∀ϕ ∈ (H01 (Ω))d , (5)
Ω

2
and that respectively i(v) and j(v) are identified to v. These identifications
are allowed since i and j are continuous and injective maps. They are con-
tinuous since, for v 6= 0,
ki(v)kV 0 ≤ kj(v)k(H −1 (Ω))d ≤ kvk(L2 (Ω))d .
The map i is injective since i(v) = 0 implies Ω v . ϕ = 0 for all ϕ ∈ V thus,
R

by continuity, for all ϕ ∈ H and in particular for ϕ = v which gives v = 0.

The map j is injective from the above inequality.
The precise meaning of imbeddings (3) is that there exist linear continu-
ous injective maps I and J respectively from V 0 and (H −1 (Ω))d into E.
These imbeddings cannot be all used at the same time by the following
result.
Theorem 1 There is no Hausdorff topological vector space E for which I ◦
i = J ◦ j. u
t
The condition I ◦ i = J ◦ j means that an element of H is identified to
the same element of E wether (1) or (2) is used. It is obviously necessary to
use all the imbeddings in the same equation.
The proof relies on the following result, which is proved first.
Lemma 2 The set V (or more exactly its image by the imbedding j) is dense
in W = {w ∈ (H −1 (Ω))d : ∇ . w = 0}. u
t
Proof. It suffices to prove that any L ∈ W 0 which vanishes on j(V) is null,
W being provided with the topology of (H −1 (Ω))d .
Since W is closed in this space, Hann–Banach’s theorem provides an
extension Le ∈ ((H −1 (Ω))d )0 . Since (H 1 (Ω))d is reflexive, there exists ` ∈
0
(H0 (Ω)) such that: ∀w ∈ (H −1 (Ω))d ,
1 d

L(w)
e = hw, `i(H −1 (Ω))d ×(H01 (Ω))d .
By assumption this cancels if w = j(v) with v ∈ V that is, by (5), Ω v . ` = 0.
R

Thus, by de Rham’s theorem (see [8] for an easy proof), there exists q ∈ D0 (Ω)
such that ` = ∇q.
On the boundary Γ, ∇q = 0. Since Γ is connected, q is constant on it,
say c, thus q − c ∈ H02 (Ω). Then, for all w ∈ W ,
L(w) = h∇(q − c), wi(H01 (Ω))d ×(H −1 (Ω))d
= −hq − c, ∇ . wiH 2 (Ω)×H −2 (Ω)
0

which cancels since ∇ . w = 0. Thus L = 0. u

t

3
Proof of Theorem 1. At first we extend the definition (5) of j(v) to
v ∈ (L2 (Ω))d , that is we still denote j the canonical continuous injective map
from (L2 (Ω))d into (H −1 (Ω))d .
Let q be a non constant harmonic function in IRd and v = ∇q |Ω . Ob-
viously j(v) ∈ W thus, by Lemma 2, there exists a sequence (vn )n∈IN such
that, as n → ∞,

vn ∈ V, j(vn ) → j(v) in (H −1 (Ω))d .

For all ϕ ∈ V, we have Ω v . ϕ = − Ω q∇ . ϕ = 0 and, by continuity, this

R R

holds for all ϕ ∈ V . Thus the definition (4) yields: ∀ϕ ∈ V ,

Z
hi(vn ), ϕiV 0 ×V = (vn − v) . ϕ
Ω

and therefore
| − v) . ϕ|
R
Ω (vn
ki(vn )kV 0 = sup ≤ kj(vn − v)k(H −1 (Ω))d .
ϕ∈V kϕk(H 1 (Ω))d

This implies i(vn ) → 0 in V 0 and therefore (I ◦ i)(vn ) → 0 in E. On the other

hand, (J ◦ j)(vn ) → (J ◦ j)(v) in E. If J ◦ j = I ◦ i could hold, then we would
get (J ◦ j)(v) = 0 which would contradict the injectivity of J and j. u t

Remark. It is necessary to be very careful when H is provided with the

topology of V 0 . Indeed, the functions vn ∈ V used in the above proof satisfy

vn → 0 in V 0 , vn → v in (H −1 (Ω))d , v 6= 0. u
t

3 Equations with a right-hand side valued

into V 0
We are now in position to check that, for such a right-hand side, it is not
possible to associate a pressure to the velocity given by the weak equation
in order to satisfy the Navier–Stokes equation.
Let us recall, for d ≤ 4, an existence result of a weak solution given by
J.–L. Lions in [5], see Theorem 6.1 p. 69, Corollary 6.2 and Footnote (2) p. 74

4
(see also [10], Theorem 3.1 p. 282 together with the equivalence of problems
3.1 and 3.2 stated p. 282). Given

f ∈ L2 ((0, T ); V 0 ), u0 ∈ H, (6)

there exists

u ∈ L2 ((0, T ); V ) ∩ L∞ ((0, T ); H) ∩ C([0, T ]; H weak)

such that u(0) = u0 and

h∂t u − ν∆u + (u . ∇)u, ϕi(H −1 (Ω))d ×(H01 (Ω))d = hf, ϕiV 0 ×V , ∀ϕ ∈ V. (7)

The terms in the left-hand side are defined in the following space:

∂t u ∈ H −1 ((0, T ); H), (u . ∇)u ∈ L1 ((0, T ); (L4/3 (Ω))d ),

∆u ∈ L2 ((0, T ); (H −1 (Ω))d ). (8)

They get a sense by using the imbeddings (2) and L4/3 (Ω) ⊂ H −1 (Ω) which
are understated when writing the left-hand side of (7).
The Navier–Stokes equation cannot be satisfied due to the following re-
sult.

Corollary 3 Given (6) and (8), there is no Hausdorff topological vector

space in which
∂t u − ν∆u + (u . ∇)u + ∇p = f (9)
might hold, for some p. u
t

Proof. Would such a space, say E, exist it should contain L2 ((0, T ); V 0 ) as

well as L2 ((0, T ); (H −1 (Ω))d ), in the sense that there should be two linear
continuous injective maps I and J from these spaces into E, but this would
contradict Theorem 1.
Indeed a linear continuous injective map I from V 0 into E would be
defined by I(v) = I(v̂) where v̂(t) = v for all t. Similarly, we would get a
linear continuous injective map J from (H −1 (Ω))d into E. Since an element
of L2 ((0, T ); H) should be identified to the same element of E by I and J ,
we would get I ◦ i = J ◦ j which is forbidden by Theorem 1. u t

5
 In default of the Navier–Stokes equation (9) itself, another strong equa-
tion may be obtained by using a projection P . To each v ∈ (H −1 (Ω))d we
associate a unique P (v) ∈ V 0 by
hP (v), ϕiV 0 ×V = hv, ϕi(H −1 (Ω))d ×(H01 (Ω))d , ∀ϕ ∈ V.
Let us notice that, on H, P coincide with i. More generally, if v ∈ (L2 (Ω))d ,
then P (v) = i ◦ PrH (v) where PrH is the Hilbert projection from (L2 (Ω))d
onto H.
The properties of this projection are summarized in the following result.
Lemma 4 The map P is linear continuous from (H −1 (Ω))d onto V 0 and is
not one to one. u
t
Proof. The continuity holds since kP (v)kV 0 ≤ kvk(H −1 (Ω))d .
The range of P is V 0 since, given w ∈ V 0 , by Hann–Banach’s theorem,
there exists we ∈ (H −1 (Ω))d such that hw, ϕiV 0 ×V = hw,e ϕi(H −1 (Ω))d ×(H 1 (Ω))d
0
for all ϕ ∈ V , whence w = P (w).e
It is not one to one since, if v = ∇q with q ∈ (L2 (Ω))d , then P (v) = 0.
Indeed, hP (v), ϕiV 0 ×V = 0 for all ϕ ∈ V thus, by continuity, for all ϕ ∈ V . u t
Now we are in position to give strong equations which are satisfied by
any solution of the weak equation (7). At first, into V 0 ,
i(∂t u) − νP (∆u) + P ((u . ∇)u) = f in L1 ((0, T ); V 0 ).
This equation is given for d = 2 by L. Tartar in [9], part I, Theorem p. 42
with the notation A = −P ∆ and B(v) = P ((v . ∇)v). For d ≤ 4, it is given
by R. Temam in [10], (3.18) p. 282. It is not totally satisfactory since there
is no pressure term, P is not one to one and V 0 is not a distribution space.
The second possibility is to give an equation into (H −1 (Ω))d : there exist
infinitely many F ∈ L2 ((0, T ); (H −1 (Ω))d ) such that P (F ) = f and p ∈
W −1,∞ ((0, T ); L2 (Ω)) such that
∂t u − ν∆u + (u . ∇)u + ∇p = F in (D0 ((0, T ) × Ω))d .
This is again not satisfactory since F is not unique and may always be choosen
such that p = 0 (it suffices to choose a new couple F 0 = F − ∇p, p0 = 0).
Remark. It is not possible to restore the pressure from an equation valued
into V 0 by a result similar to de Rham’s theorem. Indeed, given g ∈ V 0 , the
assumption hg, ϕiV 0 ×V = 0 for all ϕ ∈ V is nothing else than g = 0. u
t

6
4 Equations with a right-hand side valued
into (H −1(Ω))d
For sake of completness, we recall here a result of existence of a velocity–
pressure solution of the Navier–Stokes equation itself, which was proved in
[7].
Now, we consider

f ∈ L2 ((0, T ); (H −1 (Ω))d ), u0 ∈ H.

Proposition 5 There exists a pair

u ∈ L2 ((0, T ); V ) ∩ L∞ ((0, T ); H) ∩ C([0, T ]; H weak)

p ∈ W −1,∞ ((0, T ); L2 (Ω)) (10)

which satisfy the Navier–Stokes equation (9) and u(0) = u0 . u

t

Remark. Remind that H being a closed subset of (L2 (Ω))d , the weak topology
on H is the weak topology induced by (L2 (Ω))d . Therefore

C([0, T ]; H weak) ⊂ C([0, T]; (L2 (Ω))d weak). u

t

Proof. Weak solution. The method used in [4] provides a solution u ∈

L2 ((0, T ); V ) ∩ L∞ ((0, T ); H) of the following weak equation: ∀ϕ ∈ V ,

h∂t u − ν∆u + (u . ∇)u, ϕi(H −1 (Ω))d ×(H01 (Ω))d = hf, ϕi(H −1 (Ω))d ×(H01 (Ω))d ,
.
Ω u ϕ ∈ C([0, T ]), ( Ω u . ϕ)(0) = Ω u0 . ϕ.
R R R
(11)

Continuity. Given ϕ ∈ H, let ϕn ∈ V satisfy ϕn → ϕ in H as n → ∞. We

have

sup |( u . ϕn − u . ϕm )(t)| ≤ kukL∞ ((0,T );H) kϕn − ϕm kH

R R
Ω Ω
0≤t≤T

( Ω u . ϕn )n∈IN is a Cauchy sequence in C([0, T ]). Moreover Ω u . ϕn →

R R
thus
. ∞
R
Ω u ϕ in L ((0, T )) therefore (11) isR still satisfied.
For any fixed t, the map ϕ 7→ ( Ω u . ϕ)(t) is linear continuous on H
∈ .
R
therefore, by Riesz’s theorem, there exists U (t) H such that ( Ω u ϕ)(t) =
.
Ω U (t) ϕ. It satisfies U ∈ C([0, T ]; H weak), U (0) = u0 , and it coincides
R

with u in D0 ((0, T ); H) due to Fubini’s theorem.

7
Existence of pressure. Let

E = {v ∈ (H −1 (Ω))d : hv, ϕi(H −1 (Ω))d ×(H01 (Ω))d = 0, ∀ϕ ∈ V }

be equipped with the norm of (H −1 (Ω))d . Given v ∈ E, by a Rresult of L.

Tartar, there exists a unique q ∈ L2 (Ω) such that v = ∇q and Ω q = 0. It
satisfies
kqkL2 (Ω) ≤ c(Ω)kvk(H −1 (Ω))d
thus L(v) = q defines a linear continuous map from E into L2 (Ω) such that
∇ ◦ L is the identity on E.
Let w = f −∂t u+ν∆u−(u . ∇)u. It satisfies w ∈ W −1,∞ ((0, T ); (H −1 (Ω))d )
and w(ψ) ∈ E for all ψ ∈ D((0, T )). Since E is closed in H −1 (Ω))d this is
equivalent to w ∈ W −1,∞ ((0, T ); E). Then p = L w satisfies p ∈ W −1,∞ ((0, T ); L2 (Ω))
and ∇p = (∇ ◦ L) w = w which is (9). u t

Remark. If X is reflexive, W −1,∞ ((0, T ); X) is the dual space of W01,1 ((0, T ); X 0 ).

More generally,

W −1,∞ ((0, T ); X) = {v ∈ D0 ((0, T ); X) : v = v0 +∂t v1 , vi ∈ L∞ ((0, T ); X)}. u

t

Remark. If Ω is only assumed to be an open bounded set of IRd , Proposition

5 still holds with the exception of (10) which is replaced by

p ∈ W −1,∞ ((0, T ); L2loc (Ω)), ∇p ∈ W −1,1+2/d ((0, T ) × Ω).

Indeed Ω possesses a countable number of connected components Ωi . For

each i, we choose an open nonempty set ωi Rsuch that ω i ⊂ Ωi . Given v ∈ E,
there exists a unique q ∈ L2loc (Ω) such that ωi q = 0 for all i. For every open
set ω such that ω ⊂ Ω, it satisfies

kqkL2 (ω) ≤ c(Ω, ω)kvk(H −1 (Ω))d

thus L(v) = q defines now a linear continuous map from E into L2loc (Ω),
whence the first property. The second one follows from the equation (9),
since u ∈ L2+4/d ((0, T ) × Ω). u
t

8
5 Remarks on continuity
Another possibility to prove the continuity in Proposition 5 is, see Footnote
(2) p. 74 in [5] or [10], p. 282, to use the following result of J.–L. Lions and
E. Magenes [6], Lemma 8.1 p. 297.

Lemma 6 Let X and Y be two Banach spaces such that X is reflexive and
X ⊂Y.
If g ∈ L∞ ((0, T ); X) and ζ ◦ g ∈ C([0, T ]) for all ζ ∈ Y 0 , then g ∈
C((0, T ); X weak). u
t

This result is stated in [10], Lemma 1.4 p. 263, without assuming that
X is reflexive (but it is used in [10] only for a reflexive space X). This
assumption cannot be removed.
A counter-example is X = L1 ((0, T )), Y = {v ∈ L1loc ((0, T )) : 0T |v(x)| xdx <
R

∞} and (
1
if 0 < x < |t − T /2|,
(g(t))(x) = t−T /2
0 else,
and g(T /2) = 0.
Indeed, g is bounded into L1 ((0, T )) and it is continuous excepted at
t = T /2, thus it is measurable and it lies in L∞ ((0, T ); L1 ((0, T )). Moreover
it is continuous into Y at all t, T /2 included, since kg(t)k
RT Y
= |t/2 − T /4|.
1
However it is not continuous into L ((0, T )) weak since 0 (g(t))(x) dx equals
1 if t > T /2 and −1 if t < T /2.

Remark. For reader’s convenience, let us give a proof of Lemma 6 which

emphizises on the use of reflexivity. At first we consider the case where X is
dense in Y . Then
Y 0 ⊂ X 0 , Y 0 is dense in X 0 . (12)
Given ζ ∈ X 0 , there exists a sequence (ζn )n∈IN of Y 0 such that ζn → ζ in X 0 .
We have

sup |(ζn ◦ g − ζm ◦ g)(t)| ≤ kζn − ζm kX 0 kgkL∞ ((0,T ));X)

0≤t≤T

thus (ζn ◦ g)n∈IN is a Cauchy sequence in C([0, T ]). Moreover ζn ◦ g → ζ ◦ g

in L∞ ((0, T )), therefore ζ ◦ g ∈ C([0, T ]).
For any fixed t, the map ζ 7→ (ζ ◦ g)(t) is linear continuous and therefore
belongs to X 00 . Since X is reflexive, there exists G(t) ∈ X such that (ζ ◦

9
g)(t) = ζ(G(t)). It satisfies G ∈ C([0, T ]; X weak) and it coincides with g in
D0 ((0, T ); X).
It remains to consider the case where X is not dense in Y . Denoting
Z the closure of X in Y , we bring back to the previous case since, for all
ζ ∈ Z 0 , ζ ◦ g ∈ C([0, T ]). To check this last, we observe that, by Hann–
Banach’s theorem, ζ possesses an extension ζ̃ ∈ Y 0 thus ζ ◦ g = ζ̃ ◦ g which
is continuous. u t

Remark. It is not enough that X be dense in Y to get the density of Y 0

in X 0 . A counter-example, given to us by L. Wu, is X = C([0, 1]) and
Y = L2 ((0, 1)). Indeed X 0 = BV ([0,R1]), Y 0 = L2 ((0, 1)) and the closure of
Y 0 in X 0 is L1 ((0, 1)) since kf kBV = 01 |f | dx for all f ∈ L1 ((0, 1)).
For sake of completness, let us give a proof of (12). Since X is dense in
Y , the map ζ 7→ ζ |X is injective from Y 0 into X 0 , which gives Y 0 ⊂ X 0 .
To check the density, it suffices to prove that, given ξ ∈ X 00 such that
ξ(ζ) = 0 for all ζ ∈ Y 0 , we have ξ = 0. Since X is reflexive, there exists
x ∈ X such that ξ(ζ) = ζ(x) for all ζ ∈ X 0 . This cancels for all ζ ∈ Y 0 thus
x = 0 in Y , and therefore in X. Thus ξ = 0 which proves the density of Y 0
in X 0 . u
t

Remark. The continuity for the topology of V 0 is not enough to get the
continuity in the distribution sense of a function in L2 ((0, T ); V ). Indeed
there exists w such that

w ∈ L2 ((0, T ); V ) ∩ C([0, T ]; V 0 ), w(0) ∈ V, w(t) 6→ w(0) in (D0 (Ω))d .

It suffices to consider a sequence of real numbers tn decreasing to 0 as

n → ∞ with t0 = T , and to define w to be linear in [tn+1 , tn ] with w(tn ) = vn
and w0 = v where v and vn are as in the proof of Theorem 1. Then the
stated properties are satisfied with the exception of w ∈ L2 ((0, T ); V ) which is
obtained by choosing the tn such that tn (kvn k(H 1 (Ω))d +kvn+1 k(H 1 (Ω))d ) ≤ 1/n2 .
u
t

Remark. The boundedness into H is essential to get u(0) = u0 in the distri-

bution sense in Proposition 5 since it allows to pass from the continuity into
V 0 to the continuity into (D0 (Ω))d .

10
 As noticed by J.–L. Lions in [5], Remark 6.5 p. 68, the integrability of
∂t u + ∇p which is given by the Navier–Stokes equation (9) only provides
the continuity into V 0 . The property u ∈ L∞ ((0, T ); H) is the key to go
further. u
t

References
[1] E. Hopf, Über die Anfangswertaufgabe für die hydrodynamischen Gru-
ndgleichungen. Math. Nachr., 4 (1951) 213–231.

[2] A. A. Kiselev and O.. Ladyzhenskaya, On the existence and

uniqueness of the solution of the nonstationary problem for a viscous
incompressible fluid, Izv. Akad. Nauk SSSR, 21 (1957), 655–684.

[3] J. Leray, Sur le mouvement d’un liquide visqueux emplissant l’espace,

Acta Math., 63 (1934), 193–248.

[4] J.-L. Lions, Quelques résultats d’existence dans des équations aux déri-
vées partielles non linéaires, Bull. Soc. Math. France, 87 (1959) 245–273.

[5] J.-L. Lions, Quelques méthodes de résolution des problèmes aux limites
non linéaires. Dunod, 1968.

[6] J.-L. Lions, E. Magenes, Problèmes aux limites non homogènes et

applications, vol. 1. Dunod, 1968.

[7] J. Simon, Equations fortes vérifiées par les solutions faibles des équations
de Navier–Stokes. Habilitation à diriger des recherches, Université Paris 6,
1988.

[8] J. Simon, Démonstration constructive d’un théorème de G. de Rham,

C. R. Acad. Sci. Paris, ser. I, 316 (1993) 1167–1172.

[9] L. Tartar, Topics in nonlinear analysis. Publications mathématiques

d’Orsay, 1978.

[10] R. Temam, Navier–Stokes equations. North-Holland, 1984.

11

View publication stats