MA4A2 Advanced PDEs

Definitions and Results

Contents
1 Sobolev spaces
1.1 Holder spaces . . . . . . . . . . . . . .
1.2 Measure theory, functional analysis . .
1.3 Sobolev spaces . . . . . . . . . . . . .
1.4 Approximations with smooth functions
1.5 Extensions . . . . . . . . . . . . . . . .
1.6 Traces . . . . . . . . . . . . . . . . . .
1.7 Sobolev inequalities . . . . . . . . . . .

.
.
.
.
.
.
.

1
1
1
2
3
3
3
4

2 Linear elliptic equations of second order

2.1 Existence of solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2 Regularity theory of elliptic PDEs . . . . . . . . . . . . . . . . . . . . . . . .

5
5
7

3 Parabolic equations
3.1 Existence of weak solutions . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2 Regularity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

9
9
10

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

1
1.1

Sobolev spaces
H
older spaces

Definition 1.1 (Holder spaces). Let Rn open

(i) A function u : Rn is Holder continuous with exponent if |u(x) u(y)|
C|x y| for all x, y , where (0, 1]. We write u C 0, (). We define the
th Holder seminorm
|u(x) u(y)|
|x y|
x,y

[u]C 0, () := sup
x6=y

and the norm

kukC 0, := kuk + [u]C 0,
(ii) The Holder spaces C k, (), k N0 , (0, 1] are the set of functions u C k ()
which satisfy
X
X
kukC k, () :=
k uk +
[ u]C 0, ()
||k

||=k

Theorem 1.2. The space (C k, (), k kC k, ) is a Banach space if Rn

1.2

Measure theory, functional analysis

Definition 1.3. .
(i) A Banach space is called separable if there is a countable dense subset
(ii) A linear map L : X Y is bounded if there exists a C < such that kLxkY
CkxkX
(iii) The dual space X of a Banach space X is the space of all linear bounded functionals x : X R with norm
kx kX := sup |x (x)|
kxk1
xX

(iv) There is a bounded embedding J : X X = (X ) given by

J(x) = (, x)X ,X
:= (x)
A Banach space X is called reflexive if J is a bijection
(v) A Hilbert space is a Banach space whose norm is induced by an inner product.
Note Hilbert spaces are always reflexive by Riesz representation theorem.
1

Definition 1.4 (Lp spaces). Lp (), Rn is the set of all measurable functions u : R
for which kukLp () < , where
Z
1/p

|u| dx
p [1, )
kukLp () :=

ess sup |u(x)|

p=
x

Theorem 1.5. The following hold true

(i) Lp () is a Banach space for all p [1, ]
(ii) Lp () is separable for all p [1, )
(iii) Lp () is reflexive for all p (1, )
Theorem 1.6 (Holders inequality). For p, q [1, ] with 1/p + 1/q = 1 and f Lp (),
g Lq (), we have f g L1 () and kf gkL1 kf kLp kgkLq .
Theorem 1.7 (Dual space of Lp ()). For p [1, ) and q such that 1/p + 1/q = 1, we have
(Lp ())
= Lq (). More precisely, for any L (Lp ()) there is an fL Lq () such
that
Z
fL gdx
L(g) =

and the map L 7 fL is an isometric isomorphism.

1.3

Sobolev spaces

Definition 1.8 (Weak derivatives). Let u, v L1loc () and Nn0 . Then we call v the th
weak partial derivative of u if
Z
Z

||
u dx = (1)
vdx Cc ()

We write u = v in this case.

Lemma 1.9 (Uniqueness). If the weak th partial derivative is defined, it is unique up to a
set of measure 0.
Definition 1.10 (Sobolev spaces). For Rn open, k N, p [1, ] we define
(i) The Sobolev space W k,p () consisting of all functions u L1loc for which the weak
derivatives u exist for all || k and belong to Lp (). We equip the space with
the norm

1/p
X
kukW k,p :=
k ukpLp
||k

(ii) W0k,p () to be the closure of Cc () in W k,p ()

k,p
(iii) Wloc
() to consist of all u L1loc () with u W k,p (K) for all K , and we
k,p
say un u in Wloc
() if un u in W k,p (K) for all K .

Lemma 1.11. W k,p () is a Banach space for Rn open, k N, p [1, ].

2

1.4

Approximations with smooth functions

Definition 1.12. A mollifier is a function C (Rn ) with

Z
supp() B1 (0),
dx = 1, and 0
Rn

We set for > 0, := n (x/) so that

Z
supp( ) B (0) and

dx = 1
Rn

Given Rn and > 0, we set := {x | dist(x, ) > }, and for u L1loc ()

we set u (x) := (u )
Theorem 1.13 (Interior approximation). Assume u W k,p (), Rn open, p [1, ).
Then
(i) u C ( )
k,p
k,p
(ii) u u in Wloc
(), and u u in Wloc
(Rn ) if = Rn , as 0.

Theorem 1.14 (Global approximation). Let Rn , p [1, ) and u W k,p (). Then
there exists (um ) C () (but not necessarily in C ()) with um u in W k,p ()
Definition 1.15. For Rn , we say is C k if for all x , there exists r > 0 and a
C k function : Rn1 R such that, after a relabelling of coordinates, we have
Br (x ) = {x Br (x ) | xn > (x1 , . . . , xn1 )}
Theorem 1.16 (Approximation up to the boundary). If Rn is open with C 1 boundary,
then C k () is a dense set in W k,p ()

1.5

Extensions

Theorem 1.17. Let Rn with C k boundary. Then there exists and extension operator
for every 0 ,
E : W k,p () W0k,p (0 )

with E(u) = u and kEukW k,p (0 ) CkukW k,p ()

1.6

Traces

Theorem 1.18. Let Rn open with C 1 boundary. Then there exists an operator, called
the trace operator,
T r : W 1,p () Lp ()

such that kT r(u)kLp () CkukW 1,p () and T r(u) = u if u C 1 ()
Theorem 1.19 (Trace zero functions in W 1,p ). Assume is bounded and is C 1 . Suppose
furthermore that u W 1,p (). Then
u W01,p () T r(u) = 0 on
3

1.7

Sobolev inequalities

Definition 1.20. If 1 p < n, the Sobolev conjugate of p is

p :=

np
np

Note that 1/p = 1/p 1/n, p > p.

Theorem 1.21 (Sobolev inequality). Let p [1, n) and let p be its Sobolev conjugate. Then
there exists a C = C(n, p) < such that
kukLp (Rn ) CkukW 1,p (Rn )
for all u Cc1 (Rn ).
Corollary 1.22. Let p [1, n) and p = np/(n p)
(i) If Rn open, then for all u W01,p () we have
kukLp C(n, p)kukLp
(ii) If Rn with C 1 boundary, then for all u W 1,p () we have
kukLp C()kukW 1,p
Theorem 1.23 (General Sobolev embeddings). Let Rn with C 1 boundary
(i) For k, l N0 , p, q [1, ) with ln/q kn/p (?), W k,p () continuously embeds
into W l,q (). If the inequality (?) is strict, then the embedding is compact.
(ii) For k, l N0 , p [1, ] and (0, 1) with l + k n/p (??), the space
W k,p () continuously embeds into C 1, (). If the inequality (??) is strict, then
the embedding is compact.
Theorem 1.24 (Poincare inequality - abstract version). Let Rn be open, connected
with C 1 boundary, and let M W 1,p () be a closed cone (i.e. u M u M for
all > 0) that contains only 0 as a constant function. Then there exists C < such
that for all u M ,
kukLp () CkukLp ()
Theorem 1.25. For every Rn open, there exists C < such that for all u W01,p (),
kukLp () CkukLp ()

Linear elliptic equations of second order

Definition 2.1 (Ellipticity). An operator L given by

Lu = aij ij u + bi i u + cu
is called uniformly elliptic if theres a constant > 0 such that
aij (x)i j ||2
for a.e. x and for all Rn .

2.1

Existence of solutions

Definition 2.2. We say that u H01 () is a weak solution of


i (aij j u) + bi i u + cu = f in
u = 0 on
if and only if
Z

Z
(aij i uj + bi i u + c)dx =

f dx

for all Cc ()
Theorem 2.3 (Riesz representation theorem). Let H be a Hilbert space and F H . Then
there is a unique f H such that for all x H,
F (x) = (f, x)
Theorem 2.4 (First existence theorem). Let Rn open. Then for any f L2 () there
is a weak solution of

i (aij j u) = f in
u = 0 on
(so bi = c = 0)
Theorem 2.5 (Lax-Milgram). Let H be a Hilbert space, B : H H R a bilinear map
which is bounded and coercive (i.e. there exists > 0 such that B[u, u] kuk2 for all
u H). Then for each F H there is a unique u H such that for all v H,
F (v) = B[u, v]
Theorem 2.6 (Second existence theorem). Let L be an elliptic operator in divergence form.
Then there is a R such that there is a weak solution to

L u := Lu + u = f in
u = 0 on
for every f L2 () if .
5

Theorem 2.7 (Fredholm alternative). Let V be a normed vector space and K : V V a

compact linear map (i.e. every bounded sequence (n ) V contains a subsequence (xm )
such that K(xm ) converges). Then either
(i) There is a non-trivial solution to the homogeneous problem x + K(x) = 0, or
(ii) For every y V there is a unique solution to x + K(x) = y
The operator (I + K)1 is then a bounded linear operator.
Theorem 2.8 (Third existence theorem). Precisely one of the following statements is true.
Either
(i) The homogeneous problem


Lu = 0 in
u = 0 on

has a unique (weak) solution with u H01 () not 0 a.e, or

(ii) For every f L2 () there is a unique (weak) solution to

Lu = f in
u = 0 on
and the operator L1 which maps to this solution is a bounded linear operator.
Theorem 2.9 (Fourth existence theorem). There is at most a countable set R such that

L u := Lu + u = f in
u = 0 on
has a solution for all f L2 () iff
/ .
Definition 2.10. The following definitions are needed for the next theorem
(i) We say that u 0 on if u+ := max{u, 0} W01,2 ()
(ii) We set sup u = inf{k R | u k on }

Theorem 2.11 (Maximum principle). Let u W 1,2 () satisfy Lu 0 in the weak sense,
i.e.
Z
(aij i uj v + bi i uv + cuv)dx 0
for all v W01,2 () with v 0 almost everywhere. Then
sup u sup u+

Theorem 2.12 (Last existence theorem). If L is an (elliptic) operator in divergence form

and c 0, then

Lu = f in
u = 0 on
is uniquely solvable for all f L2 ().
6

2.2

Regularity theory of elliptic PDEs

Theorem 2.13 (Interior regularity, Fredholms theorem on the interior). Let Rn open
and let L be an elliptic differential operator of second order in divergence form with
aij C 1 (), bi , c L and
X
X
kaij kC 1 () +
kbi kL + kckL K
i,j

and let u H 1 () be a weak solution of

Lu = f in (boundary values not needed)
for f L2 (). Then for arbitrary 0 we get u H 2 (0 ),
kukH 2 (0 ) C(n, K, 0 , )(kf kL2 () + kukL2 () )
and aij (x)ij u(x) + (bi (x) i aij (x))i u(x) + c(x)u(x) = f (x) for a.e. x 0 .
Lemma 2.14 (Classification of W 1,p using finite differences). Let p (1, ) and Rn
open. Define the operators Dih ,
u(x + hei ) u(x)
h

Dih u(x) :=
for x, x + hei , h R, i = 1, . . . , n.

(i) If u W 1,p (), then for 0 and |h| d(0 , )/2 we have
kDih ukLp (0 ) kDi ukLp ()
(ii) If there is a constant K < such that kDih ukLp (0 ) K for all 0 and
h d(0 , )/2, then u is weakly differentiable in direction ei (i u exists weakly)
and ki ukL2 () K
Lemma 2.15 (Cacciopoli inequality). Let L be an elliptic differential operator in divergence
form with bounded coefficients,
X
X
kaij kL +
kbi kL + kckL K
ij

on a domain Rn . Let u be any weak solution of

Lu = f in
with f (H 1 ()) . Then for 0 ,
kukL2 (0 ) C(n, K, 0 , )(kukL2 () + kf kL2 () )
7

Theorem 2.16 (Global regularity). Lets assume that in addition to the hypotheses in the
interior regularity theorem, we have that is of class C 2 and u W01,2 (). Then
u W 2,2 () and
kukW 2,2 () C(kukL2 () + kf kL2 () )
Theorem 2.17 (Cacciopoli inequality at the boundary). Let L be an alliptic differential
operator in divergence form on B1 (0)+ with
aij C 1 (B1 (0)+ ),

c, bi L (B1 (0)+ )

A weak solution of Lu = f H 1 (B1 (0)+ ) satisfying u = 0 on B1 (0)+ {xn = 0} then

satisfies
kukW 1,2 (B1/2 (0)+ ) C(kf kH 1 (B1/2 (0)+ ) + kukL2 (B1/2 (0)+ ) )
Theorem 2.18 (Higher regularity). By induction we get
(i) Assume that Rn open and u H 1 () is a weak solution to
Lu = f in
with aij C k+1 (), bi , c C k () and f H k () for some k 1. Then u
H k+2 (0 ) for all 0 , and
kukH k+2 (0 ) C(kf kH k () + kukL2 () )
where C = C(n, , , 0 , k), where
K=

n
X

(kaij kC k+1 () + kbi kC k () + kckC k () )

i,j=1

(ii) If in addition to (i) we have C k+2 , and u is a weak solution of


Lu = f in
u = on
where u = on means u H01 () for a H k+2 (), then u H k+2 ()
and
kukH k+2 () C(kf kH k () + kukL2 () + kkH k+2 () )
If in addition the solution is unique, then
kukH k+2 () C(kf kH k () + kkH k+2 () )
Corollary 2.19. If in the situation (ii) of the last theorem, aij C (), c, , f C (()),
is C , then u C ().

Parabolic equations

Definition 3.1 (Notation). For Rn open, T > 0, we set

(i) T := (0, T ]
(ii) l T := (0, T ] (lateral boundary)
(iii) p T := ( (0, T ]) ( {t = 0}) (parabolic boundary)

(iv) C (T ) = smooth functions which vanish near l T = Cc ( [0, T ])

(v) C (T ) = smooth functions which vanish near p T = Cc ( (0, T ])

Definition 3.2 (Weak solution). .
(i) We call u H 1 (T ) a weak solution of
t u + Lu = f in T

(3.1)

for some f L2 and aij , bi , c L iff for every C (T ),

Z
Z
f dx
(t u + aij i uj + bi i u + cu)dx =
T

(ii) We say that u is a weak solution of

t u + Lu = f in T
u = 0 on l T

u = g on {t = 0}
if it is a weak solution of (3.1) and it belongs to V (T ) :=

3.1

H 1 (T )

(T )

Existence of weak solutions

Lemma 3.3 (Characterisation of weak solutions). Let > 0. Then u is a weak solution of
(3.1) if and only if
Z
Z
t
(t ut + aij i uj t + bi i ut + cut )e dx =
f t et dx
T

for all C (T )
Definition 3.4 (Notation). We introduce some more notation

(i) Let H (T ) = V (T ) =

H 1 (T )

(T )

, H (T ) =

H 1 (T )

(T )

so H 1 (T ) is the subspace of functions with u = 0 on l T , and H 1 (T ) is the

subspace of functions which vanish on p T .
e T ) := {v H 1 (T ) | i t v exists and is in L2 }, and equip it with the
(ii) Define X(
norm
!1/2
n
X
kukX(
kuk2H 1 (T ) +
kt i uk2L2 (T )
e T ) :=
i=1

e T ).
Let X(T ) be the closure of C (T ) in X(
Lemma 3.5 (Energy estimates). Define the bounded bilinear functional A : H 1 (T )
X(T ) R by
Z
A(u, ) =
(t ut + aij i uj t + bi i ut + cut )et dx
T

where the coefficients aij , bi , c are all in L . Then for large enough there exist , e > 0
such that
e H 1 ( )
|A(v, v)| kvk
T
when the operator does not depend on time.
Theorem 3.6 (Lions-Lax-Milgram). If A : H X R is a bounded bilinear form, H a
Hilbert space and X a normed space, then the following statements are equivalent:
(i) For all f X there exists u H such that f (v) = A(u, v) for all x X
(ii) There exists > 0 such that inf sup |A(u, v)|
vX uH
kvk=1 kuk=1

Theorem 3.7 (Existence theorem). Given f L2 (T ) and an elliptic operator L in divergence form with bounded coefficients that do not depend on time, and u0 H01 (), there

exists a unique weak solution u H 1 (T ) of

t u + Lu = f in T
u = 0 on l T

u = u0 on {t = 0}

3.2

Regularity

Definition 3.8 (Notation). We let, for i, m {0, 1} and k N,

Wpi,m (T ) = {u Lp (T ) | x u, tm u exist for || i and belong to Lp }
Wp2k,k (T ) = {u Lp (T ) | x tr u exists and belong to Lp if || + 2r 2k}
10

Theorem 3.9 (Regularity). Let Rn

(i) Let u be a weak solution of
t u + Lu = f in T
where f W22k,k (T ), and the time independent coefficients aij , bi , c satisfy aij
C 2k+1 (), bi , c C 2k ().Then for 0 , > 0 we have
u W22k+2,k+1 (0 [, T ])
(ii) If we assume that C 2k+2 , u0 W02k+2,2 (), then we have
u W22k+2,k+1 (T )

11