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

Evans
Chapter 5 Sobolev Spaces

Shih-Hsin Chen∗, Yung-Hsiang Huang†

2017.08.13

Abstract

In these exercises U always denote an open set of Rn with smooth boundary ∂U . As

usual, all given functions are assumed smooth, unless otherwise stated.

1. Suppose k ∈ Z≥0 , 0 < γ ≤ 1. Show that C 0,β (U ) is a Banach space.

Proof. We prove the case k = 0 for simplicity. The general case is almost the same (except we
need the standard convergence theorem between {fn } and {Dfn }).

Given a Cauchy sequence {fn } ⊂ C 0,β (U ), then there is a f ∈ C(U ) such that fn → f uniformly
and a constant M > 0 such that for each n ∈ N and s 6= t,

|fn (s) − fn (t)|

≤ M.
|s − t|β

Then for each s 6= t, there is some N = N (s, t) ∈ N such that

|f (s) − f (t)| ≤ |f (s) − fN (s)| + |fN (s) − fN (t)| + |fN (t) − f (t)|

≤ 2|s − t|β + M |s − t|β .

Therefore, f ∈ C 0,β (U ). It remains to show fn → f in C 0,β (U ). For each s, t ∈ U and  > 0,

by Cauchy’s criteria, we can find a N = N () such that for k, n ≥ N

|fk (s) − fn (s) − fk (t) + fn (t)| ≤ |s − t|β

∗
Department of Math., National Taiwan University. Email: d03221002@ntu.edu.tw
†
Department of Math., National Taiwan University. Email: d04221001@ntu.edu.tw

1
 For each η > 0, we can find a K = K(η, s, t) > N , such that |f (x) − fK (x)| ≤ η for x = s, or
x = t. Therefore, for every n ≥ N

|f (s)−fn (s)−f (t)+fn (t)| ≤ |f (s)−fK (s)|+|fK (s)−fn (s)−fK (t)+fn (t)|+|f (t)−fK (t)| ≤ 2η+|s−t|β .

Letting η → 0, we obtain that |f (s) − fn (s) − f (t) + fn (t)| ≤ |s − t|β for all n ≥ N ().

|u(x) − u(y)|
2. Proof. kukC 0,γ = sup |u(x)| + sup .
|x − y|γ
Set t s.t. (1 − t)β + t = γ. Then
 1−t  t
|u(x) − u(y)| |u(x) − u(y)| |u(x) − u(y)|
= .
|x − y|γ |x − y|β |x − y|
1−t
kukC 0,γ ≤ kuk1−t t t 1−t t 1−t t
∞ kuk∞ + [u]C 0,β [u]C 0,1 := a1 b1 + a2 b2 .

Since y = xt , 0 < t < 1, is a concave function, we have

 
1−t t 1−t t a1  b 1  t a2  b 2  t
a1 b1 + a2 b2 = (a1 + a2 ) +
a1 + a2 a1 a1 + a2 a2
 t
b1 b2
≤ (a1 + a2 ) +
a1 + a2 a1 + a2
= (a1 + a2 )1−t (b1 + b2 ).

Then

1−t
kukC 0,γ ≤ kuk1−t t t
∞ kuk∞ + [u]C 0,β [u]C 0,1

≤ (kuk∞ + [u]C 0,β )1−t (kuk∞ + [u]C 0,1 )t

1−t t
= kukC 0,β kukC 0,1 .

Remark 1. There is a general convexity theorem for Higher-order Hölder norm (due to
Hörmander), see Helms [3, chapter 8].

3. Omit.

4. Proof. (b) follows easilyZfrom (a) and Hölder inequality. For (a), denote the weak derivative of
t
u by u0w . Define v(t) = u0w (s) ds, then v is absolutely continuous. Given φ ∈ Cc∞ ,
0
Z 1 Z 1 Z x Z 1
0
[v − u]φ dx = u0w (t) dtφ0 (x) dx − u(x)φ0 (x) dx
0
Z0 1 Z0 1 Z 01
= φ0 (x) dxu0w (t) dt + u0w (x)φ(x) dx
0
Z 1t Z 1 0
=− φ(t)u0w (t) dt + u0w (x)φ(x) dx = 0.
0 0

2
 (This argument avoids the use of the Lebesgue Differentiation theorem if we only want to show
the 1-d weakly differentiable function F = H a.e., an absolutely continuous function.) We are
done if we use Exercise 11. (There are many other way to prove the second fundamental lemma
of calculus of variations without using mollifications. We also note such statement is true for
distributions.)

5. Proof. For any V ⊂⊂ W ⊂⊂ U , we can find a smooth function u such that u ≡ 1 on V and
1
supp(u) ⊂ W . Take ε = dist(V , W c ) > 0. Let φ (z) be the standard mollifier supported in
3
0
{|z| ≤ }, V := {z : dist(z, V ) < } and
Z Z
u(x) := χV 0 ∗ φε (x) = χV 0 (x − y)φε (y) dy = φε (y) dy.
Rn x−V 0

Then u is smooth (by LDCT), u = 1 on V and supported in {dist(x, V ) ≤ 2} ⊂ W .

6. Apply Exercise 5. We omit it since it can be found in many Differential Geometry textbooks.

7. Proof.
Z Z
p
|u| dS ≤ |u|p α · ν dS
∂U Z∂U
= div (|u|p α) dx
ZU
= |u|p div α + ∇(|u|p ) · α dx
U

8. Proof. Consider un (x) = h1 − n∗ dist(x, ∂U )i+ . Note that un ∈ C(Ū ) , 0 ≤ un (x) ≤ 1 and
{un (x)} is decreasing to 0 for each x ∈ U , MCT tells us kun kp,U → 0. On the other hand,
kun kp,∂U = k1kp,∂U is a positive constant. Hence
kT un kp,U kun kp,∂U
= →∞
kun kp,U kun kp,U
which means T will not be bounded, if T exists.

9. Proof. Given u ∈ H 2 (U ) ∩ H01 (U ), there exists uk ∈ C ∞ (Ū ) with uk → u in H 2 (U ), and

vk ∈ Cc∞ (Ū ) with vk → u in H 1 (U ). Hence {∇uk } and {∆uk } are bounded in L2 . Note
Z Z Z Z
∂uk 2
uk dS = div (uk ∇uk ) dx = |∇uk | dx + uk ∆uk dx.
∂U ∂n U U U
Z Z
2
The right-hand side tends to |∇u| dx + u∆u dx. Note
U U
Z Z
∂uk
uk dS = div ((uk − vk )∇uk ) dx
∂U ∂n U
Z
= ∇(uk − vk )∇uk + (uk − vk )∆uk dx.
U

3
 which tends to 0 by Hölder and (uk − vk ) → 0 in H 1 (U ) . Therefore, for all u ∈ H 2 (U ) ∩ H01 (U ),
Z
|∇u|2 + u∆u dx = 0
U

Then we apply Hölder’s inequality to get the desired inequality.

10. Proof. (Sketch) (b) is similar to (a). To prove (a), we start with the formula in Hint, do
p
integration by parts once, and then apply the generalized Hölder with exponents pair (p, p, p−2 ).

11. Proof. Given K ⊂ U , compact. Then u ∈ L1 (K). Note that the weak derivative D(u ∗ η ) =
Du ∗ η = 0. Since U is connected and u ∗ η is smooth, u ∗ ηε is a constant c which will
converges to u in L1 (K) as  → 0 and hence {c 1 }∞
n=1 forms a Cauchy sequence (note that K
n

has finite measure) that will converges to some constant c. Therefore, u = c a.e. in K and
hence in U .

12. Proof. Consider in R1 , let u(x) = χ(0,1) (x).


13. Proof. Consider U = B(0, 1) − {(x, y)x > 0} in R2 , let u(x, y) = rθ in polar coordinate.
 
1
14. Let n > 1. Show that log log 1 + |x|
∈ W 1,n (B1 (0)).

Proof.
Z  Z 1
1 n
 rn−1 dr, let t = 1
 
|u(x)|n dx = ωn  log log 1 +
U
 r  r
Z0 ∞ n
1 | log log(1 + t)|
= ωn · dt
1 tn t

| log log(1 + t)|n | log log(1 + t)|n

Since lim = 0, there is T > 0 such that < 1 for t ≥ T . Then
t→∞ t t
Z ∞
1 | log log(1 + t)|n
Z
n
|u(x)| dx = ωn · dt
U 1 tn t
Z T Z ∞
1 | log log(1 + t)|n 1
≤ ωn n
· dt + ωn dt < ∞.
1 t t 1 tn

On the other hand, we omit the proof that the first weak derivative of u exists and

1 1 −xi
u xi = 1 · 1 · .
log(1 + |x|
) 1 + |x| |x|3

4
 So
Z  Z 1
n
n
 1 1 1  n−1 1
|uxi | dx ≤ ωn  log(1 + 1 ) · 1 + 1 · r2  r
  dr, let t =
U 0 r r
r
Z ∞
1 1
= ωn n
· n
· tn−1 dt
| log(1 + t)| (1 + t)
Z1 ∞
ωn 1 1
≤ n
· sn · esn ds (s = log(1 + t))
n log2 s e
< ∞.

15. Proof. Applying Poincaré inequality,

Z  12  Z  21  Z  12
2 2 2
u dx ≤ (u)U dx + |u − (u)U | dx
U U U
Z  12
1
2
≤ |(u)U ||U | + C 2 |Du| dx .
U

Applying Hölider inequality,

Z
1
|(u)U | ≤ |u| dx
|U | {u6=0}
Z  12  Z  21
1 2 2
≤ 1 dx u dx
|U | {u6=0} {u6=0}
Z  21
1 1
2
= (|U | − α) 2 u dx .
|U | U

Therefore, s
Z  12 Z  12 Z  21
2 |U | − α 2 2
u dx ≤ u dx +C |Du| dx ,
U |U | U U
q
|U |−α
Since 0 ≤ |U |
< 1, we proved the desired inequality
 s  Z  12 Z  12
|U | − α 2 2
1− u dx ≤C |Du| dx .
|U | U U

Remark 2. For further discussions on Poincare-Type inequality, see G.Leoni [4, Section 12.2]
and L.Tartar [6, Section 10-11].

16. Proof. For any u ∈ Cc∞ (Rn ),

Z Z  
x 2 x
2
· 2u∇u dx = − u div dx, by Divergence theorem
Rn |x| Rn |x|2
n−2
Z
=− u2 dx.
Rn |x|2

5
 The Hölder’s inequality implies that

n−2
Z
u2  Z u2  12  Z  21
2
2
dx ≤ 2
dx |∇u| dx .
2 Rn |x| Rn |x| Rn

Given u ∈ H 1 (Rn ), there exists a sequence uk ∈ Cc∞ (Rn ) converging to u in H 1 . By picking a

further subsequence, we may assume uk converging to u pointwisely. Fatou’s lemma implies

(n − 2)2 u2k (n − 2)2 u2

Z Z Z Z
2 2
|∇u| dx = liminf |∇uk | dx ≥ liminf 2
dx ≥ 2
dx.
Rn k→∞ Rn k→∞ 4 Rn |x| 4 Rn |x|

Remark 3. This proof is simpler than the one suggested in the hint. Also note that the same
approximation argument should be applied at the end of the proof to Theorem 7 on page 296.

Remark 4. A family of inequalities that interpolate between Hardy and Sobolev inequalities
is given by the Hardy-Sobolev inequality,
! p2
|u|p
Z Z
n− n−2 p
dx ≤ C(p) |∇u|2 dx
Rn |x| 2 Rn

for any u ∈ Cc∞ (Rn ), where 2 ≤ p ≤ 2n/(n − 2), n ≥ 3. One huge extensions and improvements
of the Hardy-Sobolev inequalities is the Caffarelli-Kohn-Nirenberg inequalities established in
their well-know paper [1]. See also a further generalization by C.S.Lin [5].

17. Proof. For 1 ≤ p < ∞, there is a sequence um ∈ W 1,p ∩ C ∞ (U ) such that um → u in W 1,p .
Then Z Z
p 0 p
|F (um ) − F (u)| ≤ sup|F | |um − u|p → 0, as m → ∞.
U U
Z Z Z
0 0 0
p
|F (um )Dum − F (u)Du| ≤ sup|F | |Dum − Du| + p
|F 0 (um ) − F 0 (u)|p |Du|p .
U U U

Up to a subsequence, we know um converges to u a.e. Since F is continuous, F 0 (um ) converge

0

to F 0 (u) a.e. Hence the last integral tends to 0 by the dominate convergence theorem. Conse-
quently, {F (um )} and {F 0 (um )Dum } converge to F (u) and F 0 (u)Du in Lp . By the uniqueness
of weak derivative, D[F (u)] = F 0 (u)Du.

On the other hand, F (u) ∈ Lp since |F (u)| ≤ kF 0 k∞ |u| + |F (0)|.

Remark 5. Note that the hypothesis U is bounded is only used to show F (u) ∈ Lp . Such
restriction can be removed if we know F (0) = 0. (Still need F 0 ∈ L∞ .) A trivial counterexample
for unbounded U and F (0) 6= 0 is F ≡ 1.

18. Proof. (a) and (c) follow from (b), (b) is proved by the method given in the hint easily. Note
that U can be unbounded. (See the remark for problem 17.)

6
19. Proof. Let φ be a smooth, bounded, nondecreasing function, such that φ0 is bounded and
φ(z) = z, if −1 ≤ z ≤ 1. Let u (x) := φ(u/). Since u ∈ L2 , u is finite a.e. Therefore u → 0
as  → 0 a.e.. Moreover by the mean-value theorem, for a.e. x

u (x)2 = 2 |φ(u(x)/) − φ(0/)|2 ≤ sup(φ0 )2 u(x)2 ∈ L1

Hence u → 0 in L2 by dominated convergence theorem. By exercise 17, {Du } = {φ0 (u/)Du}

is bounded on L2 . Given w ∈ Cc∞ (U ),
Z Z

(Du )w = u Dw → 0, since ku kL2 → 0
U U

Apply the density argument with the help of boundedness of {kDu kL2 }, we know the above
is true for any w ∈ L2 . Take w = (Du)χ{u=0} ∈ L2 , we see
Z Z Z
2 0
|Du| = φ (u/)Du · Du = Du · (Du)χ{u=0} → 0, as  → 0
{u=0} {u=0} U

which is the desired result.

Remark 6. In the next two problems, we define the norm in fractional Sobolev spaces by
Z
2
kukH s (Rn ) = (1 + |ξ|2 )s |û(ξ)|2 dξ, for s ∈ R.
Rn

which is different from Evans’ definition. This definition works for negative power s.

20. (Sobolev Lemma)

ˆ
Proof. Since u ∈ H s ⊂ L2 , Fourier inversion formula on L2 implies that u(−x) = û(x) in L2
sense. However
Z Z h s
i2  21  Z h i2  12
2 2 − 2s
|û(ξ)| dξ ≤ (1 + |ξ| ) |û(ξ)| dξ
2 (1 + |ξ| ) dξ
Rn Rn Rn
Z  12
= kuk2H s (Rn ) (1 + |ξ|2 )−s dξ ≤ Ckuk2H s (Rn ) , since 2s > n,
Rn

we know û ∈ L1 , and the inversion formula holds pointwisely now, this tells us that for each
x ∈ Rn Z
1
|u(x)| ≤ |eix·ξ ||û(ξ)| dξ ≤ Ckuk2H s (Rn )
(2π)n/2 Rn

Remark 7. It’s true that for s = k + α + n2 , 0 < α < 1, H s is continuously embedded in C k,α .

Remark 8. The converse of Sobolev lemma is true: If H s is continuously embedded in C k ,

then s > k + n2 . The proof is sketched in Folland [2, page 195].

7
21. (H s is an algebra if s > n/2)

We need the following convolution identity for Fourier transform on L2 (Rn ) :

cv = (2π)n/2 û ∗ v̂, ∀ u, v ∈ L2 .
Lemma 9. u

Proof. Since s ≥ 0,

s/2
s/2
(1 + |ξ|2 )s/2 ≤ 1 + |ξ − η| + |η|)2 ≤ (1 + 2|ξ − η|2 + 2|η|2

≤ 2s (1 + |ξ − η|2 )s/2 + (1 + |η|2 )s/2 .

 

Hence
Z Z

2
kuvk2H s = 2 s
(1 + |ξ| ) |c 2
uv(ξ)| dξ ≤ (2π) n
(1 + |ξ|2 )s/2 (|u| ∗ |v|)(ξ) dξ
Rn Z Z

2
(1 + |ξ − η|2 )s/2 + (1 + |η|2 )s/2 |û(ξ − η)||v̂(η)| dη dξ
 
≤ C(n, s)
Z Z Z

2
(1 + |ξ − η| ) |û(ξ − η)||v̂(η)| dη + (1 + |η|2 )s/2 |û(η)||v̂(ξ − η)| dη dξ
2 s/2
 
=C
Z
≤ 2C (|h·is û| ∗ |v̂|(ξ))2 + (|û| ∗ |h·is v̂(ξ))2 dξ, where hxi := (1 + |x|2 )1/2 .

≤ C k < · >s ûk2L2 kv̂k2L1 + kûk2L1 k < · >s v̂k2L2 , by Young’s inequality .

≤ Ckuk2H s kvk2H s , since kv̂k ≤ CkvkH s

Acknowledge
The authors would like to thank Tommaso Seneci for his/her reminder that the original proof for
problem 19 is wrong. (2017.08.12)

References
[1] Luis Caffarelli, Robert Kohn, and Louis Nirenberg. First order interpolation inequalities with
weights. Compositio Mathematica, 53(3):259–275, 1984.

[2] Gerald B Folland. Introduction to Partial Differential Equations. Princeton university press,
2nd edition, 1995.

[3] Lester L Helms. Potential Theory. Springer Science & Business Media, 2nd edition, 2014.

8
[4] Giovanni Leoni. A first course in Sobolev spaces, volume 105. American Mathematical Society
Providence, RI, 2009.

[5] Chang Shou Lin. Interpolation inequalities with weights. Communications in Partial Differ-
ential Equations, 11(14):1515–1538, 1986.

[6] Luc Tartar. An introduction to Sobolev spaces and interpolation spaces, volume 3. Springer
Science & Business Media, 2007.