Nicholas Dietrich

Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

Existence of Minimizers in the Calculus

of Variations

Nicolas Dietrich

March 7, 2019
1 Existence and uniqueness of
Minimizers
1.1 What we want
In the following U ⊂ Rn is a bounded open set with smooth boundary. The Lagrangian
L is a smooth function
L : Rn × R × U → R
with
(p, z, x) 7→ L(p, z, x).
Further we have Sobolev-functions w ∈ W 1,q (U ) with 1 < q < ∞, where u = g on ∂U .
We consider a functional of the form
Z
I[w] := L(Dw(x), w(x), x) dx.
U

We will now identify some conditions on the Lagrangian L which ensure that the func-
tional I[.] does have a minimizer within an appropriate Sobolev space.

1.2 Coercivity
Example 1.1. Even a smooth function f : R → R which is bounded from below does
1
not need to attain its minimum. Therefore consider the function f (x) := 1+x 2 with
1 1
1+x2
> 0 and inf 1+x2 = 0 but f does not attain its minimum.
x∈R

The upper example suggests that in general we need some hypothesis controlling I[w]
for large functions w. The most effective way to ensure this would be to introduce a
hypothesis like:
I[w] grows rapidly as |w| → ∞.
Definition 1.2. Suppose there exist constants α > 0, β > 0 such that
L(p, z, x) ≥ α|p|q − β (1.1)
for all p ∈ Rn , z ∈ R and x ∈ U . Then we have
I[w] ≥ δkDwkqLq (U ) − γ (1.2)

for γ := β|U | and some constant δ > 0 and we call (1.2) a coercivity condition on I[.].

2
1 Existence and uniqueness of Minimizers

We interpret (1.2) as:


If kDwkLq → ∞ then I[w] → ∞.

Example 1.3. Define the function f : R2 → R as f (x, y) := x2 + y 2 . Then we have

lim f (x, y) = lim x2 + y 2 = lim |(x, y)|2 = ∞.


|(x,y)|→∞ |(x,y)|→∞ |(x,y)|→∞

Therefore f is coercive.

1.3 Problem with the boundary condition


We want to find minimizers in W 1,q (U ) which satisfy u = g on ∂U .
We have a problem in making sense of u = g on ∂U if u ∈ W 1,q (U ). In general
u ∈ W 1,q (U ) is only defined almost everywhere in U . Since |∂U | = 0 there is no
meaning we can give to the expression u |∂U . The solution to this problem is that we
interpret u = g on ∂U as
(u − g) ∈ W01,q (U ).
This also gives us g = u − (u − g) ∈ W 1,q (U ). Now we can define the space in which we
search for minimizers of the functional I[.] in a proper way. We define the space

A := {w ∈ W 1,q (U ) : (w − g) ∈ W01,q (U )}.

1.4 Weak compactness and lower semicontinuity


Remark 1.4. If f : R → R is continuous satisfying a coercivity condition, i.e. there
exist constants α > 0 and β > 0 such that

f (x) ≥ α|x|q − β

then f attains its minimum.

Proof. Because of the coercivity we can find a minimizing sequence (xk )k∈N such that

lim f (xk ) = inf f (x) := m.


k→∞ x∈R

Since the sequence (f (xk ))k∈N converges, it is bounded. So there exists a constant M > 0
such that for all n ∈ N we have

M ≥ f (xn ) ≥ α|xn |q − β

and therefore we have   1q


M +β
≥ |xn | for all n ∈ N.
α

3
1 Existence and uniqueness of Minimizers

This gives us that the sequence (xn )n∈N is bounded in R. Therefore by Bolzano-
Weierstraß-Theorem we can find a subsequence (xnk )k∈N ⊂ (xn )n∈N and x ∈ R such
that xnk → x as k → ∞. Because of the continuity of f we can interchange limit and
function and we get
f (x) = f ( lim xnk ) = lim f (xnk ) = m.
k→∞ k→∞

Therefore x is a minimizer of f .
The problem is that our functional I[.] in general will not have a minimizer as the
following example illustrates
Example 1.5. We define the function f : R → R for q > 1 as
 q
 x , for x > 0
f (x) := 1 , for x = 0
(−x)q , for x < 0

Then obviously f is coercive, but it does not attend its minimum.


Unfortunately we can not use the above proof for general functionals I[.] since our
space is not finite dimensional and I[.] is in general not continuous. But we will use the
idea of the proof and introduce some additional conditions on L. To understand the
problem we set
m := inf I[w]
w∈A

and choose a minimizing sequence (uk )k∈N ⊂ A such that

lim I[uk ] = m.
k→∞

Now we would like to show that some sequence converges to an actual minimizer. For this
we need some kind of compactness argument as in the proof for continuous functions,
but the problem is that the space in which we search for the minimizer is infinitely
dimensional. If we utilize the corecivity argument we only get that the minimizing
sequence is bounded, but nothing more. So it is not a good idea to work with strong
compactness.
Let us work with weak compactness instead. Since we are assuming that 1 < q < ∞ we
have that W 1,q (U ) is reflexive. Because of the coercivity condition on I[.] we know that
the sequence (Duk )k∈N is bounded in Lq (U ) and because of the Poincare-inequality we
further know that (uk )k∈N is bounded in Lq (U ). Therefore (uk )k∈N is bounded in W 1,q (U ).
Because of Theorem 2.2 we conclude that there exists a subsequence (ukj )j∈N ⊂ (uk )k∈N
and u ∈ W 1,q (U ) such that ukj * u weakly in W 1,q (U ), i.e.

ukj * u weakly in Lq (U )
Dukj * Du weakly in Lq (U ; Rn )

Further it is true that u = g on ∂U . We have that (ukj −g) ∈ W01,q (U ) and ukj −g → u−g,
as j → ∞. Since W01,q (U ) is closed we have u − g ∈ W01,q (U ). Therefore we have u ∈ A.

4
1 Existence and uniqueness of Minimizers

The last problem is that the functional I[.] in general is not continuous. So we do not
have I[u] = lim I[uj ] if uj → u. In general weak convergence does not imply convergence
j→∞
almost everywhere.
The good thing is that we do not really need the full strength of continuity. It is sufficient
to assume lower semicontinuity.
Definition 1.6. We say that a function I[.] is sequentially weakly lower semicontinuous
on W 1,q (U ), provided
I[u] ≤ lim inf I[uk ]
k→∞
1,q
whenever uk * u weakly in W (U ) as k → ∞.
In the next chapter we will find a property of the Lagrangian L which together with
the coercivity will give us lower semicontinuity.

1.5 Convexity and lower semicontinuity


Theorem 1.7. Assume that L is smooth, bounded below and in addition the mapping
p 7→ L(p, z, x) is convex or each z ∈ R and x ∈ U . Then I[.] is weakly lower semicon-
tinuous on W 1,q (U ).
Proof. 1. We choose any sequence (uk )k∈N with uk * u weakly in W 1,q (U ) and set

l := lim inf I[uk ]. (1.3)


k→∞

We want to show that I[u] ≤ l. If l = ∞, then we are done. So we assume that l < ∞.
2. Note first that because uk * u weakly in W 1,q (U ) we have that the sequence (uk )k∈N
is bounded in W 1,q (U ). So we have supk∈N kuk kW 1,q (U ) < ∞. Because of (1.3) we can
find a subsequence (uk` )`∈N ⊂ (uk )k∈N such that

l = lim I[uk` ].
`→∞

To simplify the notation we will denote the subsequence also by (uk )k∈N . By the com-
pactness Theorem 2.3 we also know that uk → u strongly in Lq (U ), as k → ∞. Further
we have by Theorem 2.4 that there exists a subsequence (ukr )r∈N ⊂ (uk )k∈N such that

ukr → u (1.4)

almost everywhere in U as r → ∞. Again we call the subsequence (uk )k∈N .


3. We fix ε > 0. Then (1.4) and Theorem 2.1 give us
k→∞
uk −→ u uniformly on Eε , (1.5)

where Eε is a measurable set with

|U \ Eε | ≤ ε. (1.6)

5
1 Existence and uniqueness of Minimizers

0
We may assume Eε ⊂ Eε0 for 0 < ε < ε. We define the set
 
1
Fε := x ∈ U : |u(x)| + |Du(x)| ≤ . (1.7)
ε
For this set we also have
lim |U \ Fε | = 0. (1.8)
ε↓0

We eventually set
Gε := Eε ∩ Fε . (1.9)
Because of (1.6) and (1.8) we further have

|U \ Gε | = |U \ (Eε ∩ Fε )| = |(U \ Eε ) ∪ (U \ Fε )|
ε↓0
≤ |U \ Eε | + |U \ Fε | ≤ ε + |U \ Fε | −→ 0.

4. Because L is bounded from below we can assume without loss of generality that
L ≥ 0 (otherwise apply the following argument to L̃ := L + β ≥ 0 for some constant β).
By using the gradient- inequality for differentiable, convex functions we get
Z Z
I[uk ] = L(Duk , uk , x) dx ≥ L(Duk , uk , x) dx
U Gε
Z Z
≥ L(Du, uk , x) dx + hDp L(Du, uk , x), Duk − Dui dx.
Gε Gε

Because of (1.5), (1.7) and (1.9) we have by using the dominated convergence theorem
and the continuity of z 7→ L(p, z, x):
Z Z
DCT
lim L(Du, uk , x) dx = lim L(Du, uk , x) dx
k→∞ G Gε k→∞
ε
Z
= L(Du, lim uk , x) dx
Gε k→∞
Z
= L(Du, u, x) dx.

Now we argue why we can use the dominated convergence theorem. Because z 7→
L(p, z, x) is smooth, it is Lipschitz-continuous. Therefore there exists a constant C > 0
such that for every uk ∈ W 1,q (U ) and u ∈ W 1,q (U ) we have

|L(Du, uk , x) − L(Du, u, x)| ≤ C1 |uk (x) − u(x)|

for all x ∈ Gε . Further we have because (uk )k∈N is bounded in W 1,q (U ) and (1.7)
Z Z
L(Du, uk , x) − L(Du, u, x) dx ≤ |L(Du, uk , x) − L(Du, u, x)| dx
Gε Gε
Z
≤ C1 |uk (x) − u(x)| dx

6
1 Existence and uniqueness of Minimizers

Z  1q Z
1− 1q q
≤ C1 |U | |uk | dx + |u(x)| dx
U Gε
Z
1− 1q 1
≤ C1 |U | sup kuk kW 1,q (U ) +C1 dx
k∈N Gε ε
| {z }
=:C2
1 C1 |U |
≤ C1 |U |1− q C2 + .
ε
Therefore we found a uniform bound on our integral and therefore we can put the limit
under the integral and get
Z Z
ε↓0
L(Du, uk , x) − L(Du, u, x) dx ≤ |L(Du, uk , x) − L(Du, u, x)| dx −→ 0.
Gε Gε

Now we show that


k→∞
Dp L(Du, uk , x) −→ Dp L(Du, u, x) (1.10)
uniformly on Gε . So we have to show that for every η > 0 there exists an N (ε) ∈ N
such that
|Dp L(Du, uk , x) − Dp L(Du, u, x)| < η.
for all k ≥ N and all x ∈ Gε .
Since z 7→ Dp L(p, z, x) is Lipschitz continuous, there exists a constant C > 0 such that
for every x ∈ Gε we have

|Dp L(Duk , uk , x) − Dp L(Du, u, x)| ≤ C|uk (x) − u(x)|.

Because of the uniform convergence of (uk )k∈N we have that for every η > 0 there exists
an N (η) ∈ N such that for all x ∈ Gε we have
η
|uk − u| <
C
for all k ≥ N . Combining those results we get that for every η > 0 there exists an
N ∈ N such that for every x ∈ Gε
η
|Dp L(Du, uk , x) − Dp L(Du, u, x)| ≤ C|uk (x) − u(x)| < C = η.
C
for all k ≥ N .
Now we show that
Z
lim hDp L(Du, uk , x), Duk − Dui dx = 0. (1.11)
k→∞ Gε

We have
Z
lim hDp L(Du, uk , x), Duk − Dui dx
k→∞ Gε

7
1 Existence and uniqueness of Minimizers
Z
= lim hDp L(Du, uk , x) − Dp L(Du, u, x), Duk − Dui dx
k→∞ Gε
| {z }
=:I1 (k)
Z
+ lim hDp L(Du, u, x), Duk − Dui dx .
k→∞ Gε
| {z }
=:I2 (k)

q
Now we show that Dp L(Du, u, x) ∈ L q−1 (Gε ). For some y ∈ Gε we have
Z Z
q 1 q
|Dp L(Du, u, x)| q−1 dx ≤ 2 q−1 |Dp L(Du(x), u(x), x) − Dp L(Du(y), u(y), y)| q−1 dx
Gε Gε
Z
1 q
+2 q−1 |Dp L(Du(y), u(y), y)| q−1 dx
Z Gε
1 q
≤ 2 q−1 (|Du(x) − Du(y)| + |u(x) − u(y)| + |x − y|) q−1 dx

1 q
+2|U ||Dp L(Du(y), u(y), y)| q−1
q−1

 q
 q−1
1 2
≤ 2 |U |
q−1 + diam(U ) + C3 < ∞.
ε

It follows because we have Duk * Du weakly in Lq (U ; Rn )

lim I2 (k) = 0
k→∞

Further we have because of (1.10)


Z
|I1 (k)| ≤ |Dp L(Du, uk , x) − Dp L(Du, u, x)||Duk − Du| dx
ZGε
≤ |Dp L(Du, u, x) − Dp L(Du, u, x)||Duk | dx

| {z }
=:I3 (k)
Z
+ |Dp L(Du, u, x) − Dp L(Du, u, x)||Du| dx .

| {z }
:=I4 (k)

Moreover we apply Hölders inequality and because of the boundedness of (uk )k∈N in
W 1,q (U ) we get
Z 1− 1q Z  1q
q
q
|I3 (k)| ≤ |Dp L(Du, u, x) − Dp L(Du, u, x)| q−1 dx |Duk | dx
Gε Gε
Z 1− 1q
q
≤C |Dp L(Du, u, x) − Dp L(Du, u, x)| q−1 dx → 0,

8
1 Existence and uniqueness of Minimizers

as k → ∞. Analogous we show that |I4 (k)| → 0, as k → ∞. This gives us (1.11). If we


look at the upper inequality we conclude
Z
l := lim I[uk ] ≥ L(Du, u, x) dx.
k→∞ Gε

This inequality holds for arbitrary ε > 0. So by letting ε ↓ 0 and using the Monotone
Convergence Theorem we get
Z
l ≥ lim L(Du, u, x) dx
ε↓0 G
ε
Z
= lim L(Du, u, x) χGε (x) dx
ε↓0 U | {z } | {z }
≥0 increasing
Z
MTC
= L(Du, u, x) dx = I[u].
U

This gives us the desired result.


With the upper result we can finally prove the existence of a minimizer.

1.6 Existence of a Minimizer


Theorem 1.8 (Existence of a minimizer). Assume that L is a smooth function which
satisfies the coercivity inequality (1.1) and is convex in the variable p. Then there exists
at least one function u ∈ A solving

I[u] = min I[w].


w∈A

Proof. 1. Set m := inf I[w]. If m = +∞, we are done. So we now assume that m < ∞.
w∈A
Because of the coercivity we can find a minimizing sequence (uk )k∈N such that
k→∞
I[uk ] −→ m.

2. Without loss of generality we take β = 0 in the coercivity inequality (1.1), because


we can otherwise just consider L̃ := L + β. Therefore the coercivity inequality simplifies
to L ≥ α|p|q . This gives us Z
I[w] ≥ α |Dw|q dx.
U

Because the sequence (I[uk ])k∈N converges in R, it is bounded in R. Therefore there


exists a constant M > 0 such that for every k ∈ N
Z
M ≥ I[uk ] ≥ α |Duk |q dx.
U

9
1 Existence and uniqueness of Minimizers

This gives us
  1q
M
kDukLq (U ) ≤
α
for all k ∈ N, which implies
sup kDuk kLq (U ) < ∞. (1.12)
k∈N
3. Now we fix w ∈ A. Since uk and w are both equal to g on ∂U we have
uk − w = (uk − g) − (w − g) ∈ W01,q (U ).
The Poincare-Friedrichs inequality gives us
kuk kLq (U ) ≤ kuk − wkLq (U ) + kwkLq (U )
≤ C1 kDuk − DwkLq (U ) + C2
≤ C1 (kDuk kLq (U ) + kDwkLq (U ) ) + C2
| {z }
≤C3

≤ C1 (sup kDuk kLq (U ) +C3 ) + C2 ≤ C.


k∈N
| {z }
≤C4

This gives us
sup kuk kW 1,q (U ) = sup(kuk kLq (U ) + kDuk kLq (U ) )
k∈N k∈N
= sup kuk kLq (U ) + sup kDuk kLq (U ) < ∞.
k∈N k∈N

So (uk )k∈N is bounded in W 1,q (U ).


4. By Theorem 2.2 there exists a subsequence (ukj )j∈N ⊂ (uk )k∈N and a function
u ∈ W 1,q (U ) such that
ukj * u
weakly in W 1,q (U ). We show that u ∈ A. That is to show that u − g ∈ W01,q (U ). We
know that W01,q (U ) is a closed, convex linear subspace of W 1,q (U ). So we can apply
Mazurs Theorem, which says that a convex, closed subset of a reflexive Banach-space is
weakly closed. Now we take the sequence (ukj − g)j∈N ⊂ W01,q (U ) which now, because of
the weak compactness converges to u − g ∈ W01,q (U ). Therefore we have u ∈ A. Because
of the convexity of p 7→ L(p, z, x) we can apply Theorem 1.7 and we get
I[u] ≤ lim inf I[ukj ] = m.
j→∞

Since u ∈ A and m is the infimum we have


I[u] ≥ m.
So in total we get
I[u] = m = min I[w].
w inA
This proves the result.

10
1 Existence and uniqueness of Minimizers

Now we will set the conditions under which our minimizer is unique:

L = L(p, x) does not depend on z (1.13)


There exists θ > 0 such that for all p, q ∈ Rn and x ∈ U and t ∈ [0, 1] we have

(1 − t)L(q, x) + tL(p, x) ≥ L((1 − t)q + tp) + t(1 − t)θ|q − p|2 . (1.14)

The upper condition says that p 7→ L(p, x) is uniformly convex for each x ∈ U . The
next Theorem gives us the uniqueness of the minimizer.

Theorem 1.9 (Uniqueness of the Minimizer). Suppose (1.13) and (1.14) hold. Then a
minimizer u ∈ A is unique.

Proof. 1. Assume u, ũ ∈ A are both minimizers of I[.] over A.


Then we set v := u+ũ
2
∈ A. Indeed we have v ∈ A, because

u + ũ 1 1
− g = (u − g) + (ũ − g) ∈ W01,q (U ).
2 2 2
We claim that
I[u] + I[ũ]
I[v] ≤ (1.15)
2
with is a strict inequality, unless u = ũ almost everywhere.
2. We want to prove (1.15). The uniform convexity (1.14) together with the smoothness
of L gives us
θ
L(p, x) ≥ L(q, x) + hDp L(q, x), p − qi + |p − q|2
2
n Du+Dũ
for all x ∈ U and p, q ∈ R . Now we set q := 2 and p = Du and integrate over U
Z    
Du + Dũ Du − Dũ
I[v] + Dp L ,x , dx
U 2 2
Z
θ
+ |Du − Dũ|2 dx ≤ I[u].
8 U
Du+Dũ
Similarily we set q = 2
and p = Dũ and integrate over U
Z    
Du + Dũ Dũ − Du
I[v] + Dp L ,x , dx
U 2 2
Z
θ
+ |Du − Dũ|2 dx ≤ I[ũ].
8 U
Now we add the upper inequalities and divide by 2 and we get
Z
θ I[u] + I[ũ]
I[v] ≤ I[v] + |Du − Dũ|2 dx ≤ = min I[w].
8 U 2 w∈A

11
1 Existence and uniqueness of Minimizers

Therefore we have
I[u] + I[ũ]
I[v] = ,
2
which gives us Z
θ
|Du − Dũ|2 dx = 0.
8 U
This implies Du = Dũ almost everywhere in U . We have that

u − ũ = (u − g) − (ũ − g) ∈ W01,q (U ).

Therefore we can use Poincare Friedrichs inequality and get


Z Z
q
|u − ũ| dx ≤ C |Du − Dũ|q dx = 0.
U U

This finally gives us u = ũ almost everywhere in U .

12
2 Appendix
Theorem 2.1 (Egoroff). Let (X, M, µ) be a measure space µ(X) < ∞. Let (fn )n∈N
be a sequence of measurable functions on X and let f be a measurable function on X.
Assume that fn → f µ-almost everywhere as n → ∞. Then for every ε > 0 there exists
a measurable subset D of X such that µ(D) < ε and fn → f uniformly on X \ D.
Proof. Let ε > 0 be arbitrary. By definition of convergence almost everywhere there
exists a set E ∈ M such that µ(E) = 0 and lim fn (x) = f (x) for all x ∈ A, where
n→∞
A := X \ E. Now for all n, k ∈ N we define An,k by
 
1
An,k := x ∈ A : |fn (x) − f (x)| ≥ .
k
Further we define Bn,k by

[
Bn,k := Ai,k .
i=n
Since fn converges pointwise to f on A it follows by the definition of convergence that
for each x ∈ A and k ∈ N we have
1
|fi (x) − f (x)| <
k
for all i ∈ N sufficiently large.
Thus, when k is fixed, no element of A belongs to An,k for infinitely many n. Therefore
we have
lim sup An,k = ∅.
n→∞

So we have since µ(X) < ∞ and using the continuity of the measure
0 = µ(∅) = µ(lim sup An,k )
n→∞
∞ ∞
! ∞
!
\[ \
=µ Ai,k = µ Bn,k
n=0 i=n n=0
= µ( lim Bn,k ) = lim µ(Bn,k ).
n→∞ n→∞
ε
k ∈ N was arbitrary we can choose nk ∈ N such that µ(Bnk ,k ) < 2k+1
Since S . We now set
B := k∈N Bnk ,k and get by using the subadditivity of the measure
!
[ X X ε
µ(B) = µ Bnk ,k ≤ µ(Bnk ,k ) < = ε.
k∈N k∈N k∈N
2k+1

13
2 Appendix

Now we set D := B ∪ E. Then we have since µ(E) = 0

µ(B ∪ E) ≤ µ(B) + µ(E) = µ(B) < ε.

Also given any k ∈ N we have that x ∈ X \ D implies x 6∈ Bnk ,k . So we have

1
|fi (x) − f (x)| <
k
or all i ≥ nk . Since this is true or all x ∈ X \ D, it follows that fn converges uniformly
to f as n → ∞ on X \ D.

Theorem 2.2. If X is a reflexive Banach-space and (xn )n∈N is a bounded sequence in


X. Then there exists a subsequence (xnk )k∈N of (xn )n∈N and x ∈ X such that xnk * x
weakly as k → ∞.

Theorem 2.3. If 1 < q < ∞ and uk * u weakly in W 1,q (U ). Then we have uk → u


strongly in Lq (U ).

Theorem 2.4. If the sequence (uk )k∈N ⊂ Lq (U ) converges to some u ∈ Lq (U ), i.e.


uk → u in Lq (U ). Then there exists a subsequence (ukj )j∈N ⊂ (uk )k∈N such that ukj → u
almost everywhere, as k → ∞.

14

You might also like