Partial Differential Equations, 2nd Edition, L.C.

Evans
Chapter 8 The Calculus of Variations∗

Yung-Hsiang Huang†

2018.03.25

Notation: U denotes a bounded smooth, open subset of Rn . All given functions are assumed
smooth, unless otherwise stated.

1. This is called Rademacher’s functions. Consult Brezis [1, Exercise 4.18].

(a) This is Riemann-Lebesgue lemma, which holds not only for L2 but also for Lp , (p ∈ [1, ∞)),
and weak-* convergence in L∞ ; I put my comment on various proofs below:
(1) I think the easy way to understand is through Bessel’s inequality. But this method does
not work on Lp , except L2 .
(2) The second way is through integration by parts, and one needs the density theorem (simple
functions or Cc∞ functions). This method is adapted to our claim, except for weak convergence
in L1 . But for L1 case, it’s a simple consequence of squeeze theorem in freshman’s Calculus.
It’s also easy to see fn 6→ 0 a.e. or in measure.

Proof. (b)You can apply (2)’s method, start with step functions χB .

2.
1 
−φ(x) 2
L(p, z, x) = e |p| − f (x)z .
2
A good intuition is explained in http://math.stackexchange.com/questions/270110/.

−1
3. The equation can be rewrite as 0 = u
 t
+ 1 ∆x u + utt which is equivalent to

1
0 = (e−t/ ut )t + ∆x (e−t/ u).

∗
Completed: Ex 1,2,3,7,9,10,15.
†
Department of Math., National Taiwan University. Email: [email protected]

1
 So this motivates us to define

1  1 
L = e−t/ u2t + |Dx u|2 ,
2 

and then we calculate the first variation to conclude that, for each v ∈ Cc∞ (U × (0, T ))
Z TZ
0= v · e−t/ (ut − ∆x u − utt ).
0 U

4. Proof. (a) We use the results and notations in p.462-463 without mentions: for k = 1, · · · n,
n
X
− (LPjl (P, Z, x))xj + LZ l (P, Z, x)
j=1
n
X ∂η
=− (η(cofP )kj )xj + det P
j=1
∂Z k
n hX
n
X ∂η ∂Z l k k
i ∂η
=− (cofP )j + η(cofP )j,x + det P.
j=1 l=1
∂Z l ∂xj j
∂Z k

Given u ∈ C ∞ (U ⊆ Rn ; Rn ), set Z = u, P = Du, then the Euler-Lagrange equations becomes

n X n n
X ∂η ∂ul k
X ∂η
− l
(cofDu)j − η (cofDu)kj,xj + k det Du
j=1 l=1
∂u ∂xj j=1
∂u
n n
X ∂η X ∂ul ∂η
=− l
(cofDu)kj + 0 + k det P
l=1
∂u j=1 ∂xj ∂u
n
X ∂η l ∂η
=− l
δk det P + k det P = 0,
l=1
∂u ∂u

where δkl is the Kronecker delta.

(b) Follows from the alternative characterization of null Lagrangians, that is, Theorem 1 in
p.461, and the fact that the above computations are suitable to C 2 functions.

5. Proof.

6. Proof.

7. Proof. Expand L in coordinate form directly, we have

n X
X n
L(P ) = Pki Pik − Pii Pkk
i=1 k=1

Then we see for l = 1, · · · n,

n
X n
X n
X
j
(LPjl (P ))xj = 2 (Pl )xj − 2 (Pii )xl .
j=1 j=1 i=1

2
 For any u ∈ C ∞ (U ⊆ Rn ; Rn ), we plug Plj = (uj )xl into the above identity, and see
n
X n
X n
X
j
(LPjl (Du))xj = 2 ((u )xl )xj − 2 ((ui )xi )xl = 0.
j=1 j=1 i=1

Remark 1. Does this problem have any useful application?

Remark 2. See Giaquinta-Hildebrandt [3, Chapter 1] for more discussions on Null Lagrangians.

8. Explain why the methods in Section 8.2 will not work to prove the existence of a
minimizer of the functional
Z
1
I[w] := (1 + |Dw|2 ) 2 dx
U

over A := {W 1,q (U )|w = g on ∂U }, for any 1 ≤ q < ∞.

Remark 3. This is the minimal surface problem, many great textbooks treat it, e.g. Murrary,
Giusti, or Colding-Minicozzi.
1
(1+|s|2 ) 2
Proof. The best coercive estimate one can get is |I(w)| ≥ kDwkL1 since lims→∞ s
= 1.
However, L1 (or W 1,1 ) are not reflexive spaces. So the boundedness of minimizing sequence
in L1 does not imply existence of a weakly convergent sequence, e.g. approximation of the
identity. If φk (x) = kχ{(0, 1 )} (x) is assumed to be converge weakly in L1 up to a subsequence
k
kj+1 1
φkj and kj
→ ∞ as j → ∞. Then for the bounded test function g(x) = (−1)j if x ∈ ( kj+1 , k1j ),
R R
φkj g → −1 as odd j → ∞ and φkj g → 1 as even j → ∞.

R
9. Proof. (a) Let u : U ⊂ Rn → Rm be the minimizer of I[w] := U
L(Dw, w, x) dx, then for each
v ∈ Cc∞ (U ; Rm )
d2
Z n o
0 ≤ 2 I[u + tv] = Lpki plj Di v k Dj v l + 2Lpki ul v l Di v k + Lzk zl v k v l
dt U

The above identity is true for all Lipschitz continuous Rm -valued function v with compact
support. In particular, we take
x·ξ
v(x) = ρ( )ηζ(x),

where ξ ∈ Rn , η ∈ Rm , ζ ∈ Cc∞ (U ; R) and ρ : R → R is the same tent map in Section 8.1.3.
Thus, |ρ0 | = 1 a.e. and
x·ξ
Di v k = ρ 0 ( )ξi η k ζ + O(), as  → 0.

After substituting this into the first expression and sending  → 0, we obtain
Z
0≤ Lpki plj ξi η k ξj η l ζ 2 dx.
U

3
 Since this holds for all ζ ∈ Cc∞ (U ), we deduce that 0 ≤ Lpki plj ξi η k ξj η l .

(b) Consider L(P ) = det P = p11 p22 − p12 p21 on M 2×2 . This function is not convex since for
t ∈ (0, 1),      
 0 0 1 0  1−t 0
L t  + (1 − t)   = det   > 0. (1)
0 1 1 0 1−t t
But    
0 0 1 0
tL   + (1 − t)L   = 0. (2)
0 1 1 0
Direct computation shows that L satisfies the Legendre-Hadamard condition, that is, for all
P ∈ M 2×2 , ξ ∈ R2 , η ∈ R2 ,
2 X 2
X ∂ 2 L(P )
ηk ηl ξi ξj = 2η1 η2 ξ1 ξ2 − 2η1 η2 ξ2 ξ1 = 0.
i,j=1 k,l=1
∂pki ∂plj

Remark 4. Dacorogna [2, Theorem 5.3] says that for C 2 function L : Rmn → R ∪ {∞}, the
Legendre-Hadamard condition ⇐⇒ rank one convexity, that is,

L(λξ + (1 − λ)η) ≤ λL(ξ) + (1 − λ)L(η),

for every λ ∈ [0, 1] and rank(ξ − η) ≤ 1.

Proof. (⇒) Mean-Value theorem and note that m × n matrix A is of rank 1 iff A = vwT for
some v ∈ Rm and w ∈ Rn .

(⇐) Givenξ ∈ Rm and η ∈ Rn , consider φ(t) = f (P + t · ξη T ) which is C 2 and convex, and

hence φ00 (0) ≥ 0, which is exactly the Legendre-Hadamard condition.

10. Use the methods of §8.4.1 to show the existence of a nontrivial weak solution
n+2
u ∈ H01 (U ), u 6≡ 0, of −∆u = |u|q−1 u in U and u = 0 on ∂Ω for 1 < q < n−2
,n ≥ 3.

Our method is not a direct application of theorems in Section 8.4.1. Since the corresponding g
does not satisfy the growth condition.

Proof. First, we are going to show the existence of minimizer for the energy functional E on
the admissible class A where
Z Z
1 2 1
E(u) = |∇u| dx − |u|q+1
2 U q+1 U

and
A = {u ∈ H01 (U ) : kukLq+1 = 1}.

4
 Then E is coercive on A since E A = 1 1
R
2 U
|∇u|2 dx − q+1 . Let um be the minimizing sequence,
which is bounded in H01 (Ω) by coercivity. Hence weak compactness theorem and Rellich’s
compactness theorem imply that, up to a subsequence, um * u in H01 (Ω) and um → u in
Lq+1 (Ω). Hence kukLq+1 = 1, that is, u ∈ A . Therefore u is a minimizer since

lim inf E(um ) ≥ E(u) ≥ inf E = lim E(um )

m→∞ A m→∞

Finally, a almost identically argument to Lagrange mulitiplier theorem to show this minimizer
solves −∆v = λ|v|q−1 v for some λ ∈ R with zero Dirichlet boundary condition weakly. If λ = 0,
by uniqueness theorem for laplace equation, we see u ≡ 0. But this contradicts to kukLq+1 = 1.
1
Hence we see a nontrivial function ũ = λ q−1 u solves the given boundary value problem weakly.
We remark two details in the beginning of the proof of Lagrange multiplier theorem:
(1.)Since kukLq+1 = 1 6= 0, |u|q−1 u is not identical zero a.e.
R
(2.)For any v ∈ H01 (U ), the integral U |u|q−1 uv dx is well-defined by Sobolev embedding the-
orem, the assumption that U is a bounded domain, and Hölder’s inequality to exponent pair
( q+1
q
, q + 1).

11. If u ∈ C ∞ (U ) and Multiply the equation with v ∈ C ∞ (U ), then we see

Z Z Z Z
∂u
(f, v)L2 = ∇u∇v dx − v ds = ∇u∇v dx + vβ(u) ds
U ∂U ∂ν U ∂U

For u ∈ H 1 (U ), since for any z, w ∈ R, |β(z) − β(w)| ≤ b|z − w|, trace theorem and standard
approximation argument implies that we can define u ∈ H 1 (U ) is a weak solution to our
nonlinear boundary-value problem provided for each v ∈ H 1 (U ),
Z Z
(f, v)L2 = ∇u∇v dx + T v · β(T u) ds
U ∂U

where T : H 1 (U ) → L2 (∂U ) is the trace operator.

Proof.

12. Proof.

13. Proof.

14. Proof.

15. (Pointwise gradient constraint)

5
 (a) Show tere exists a unique minimizer u ∈ A of
Z
1
I[w] := |Dw|2 − f w dx,
U 2
where f ∈ L2 (U ) and
A := {w ∈ H01 (U ) : |Dw| ≤ 1 a.e.}.

(b) Prove Z Z
Du · D(w − u) dx ≥ f (w − u)
U U
for all w ∈ A

Proof. The proof is almost identical to the proof of Theorem 3 and 4 in Section 8.4.2, except
we have to check A is weakly closed by Mazur’s theorem, a corollary of the Hahn-Banach
Theorem:

Theorem 5. [1, Section 3.3] Let C be a convex set in a Banach space E, then C is weakly
closed if and only if it is closed.

One can see the convexity of A easily. The closedness of A is proved by the same argument as
Step 1 of Theorem 3 I referred in the beginning. So we complete the proof.

16. Proof.

17. Proof.

18. Proof.

19. Proof.

20. Proof.

References
[1] Haim Brezis. Functional Analysis, Sobolev Spaces and Partial Differential Equations. Springer
Science & Business Media, 2010.

[2] Bernard Dacorogna. Direct methods in the calculus of variations, volume 78. Springer Science
& Business Media, 2007.

[3] Mariano Giaquinta and Stefan Hildebrandt. Calculus of Variations I, volume 310. Springer
Science & Business Media, 2004.