Exercise sheet 6

Problem 6.1
1) Define aj = αj e(j) ∈ A. Then we have
q √
||ak − al ||2 = |αk |2 + |αl |2 ≥ 2 > 1

for any k ̸= l ∈ N. Therefore for any x ∈ ℓ2 \ A we have rj := ||x − aj ||2 > 0 ∀j ∈ N. We now show that
there exists some r > 0 such that inf j∈N rj ≥ r. Indeed if ||x − aj ||2 ≥ 1/2 ∀j ∈ N, we can take r = 1/2.
Now if ∃j ∈ N with ||x − aj ||2 = ε < 1/2, then for any k ̸= j we have with the reversed triangle inequality
1
||x − ak ||2 = ||x − aj − (ak − aj )||2 ≥ ||ak − aj ||2 − ||x − aj ||2 > 1 − ε >
2
(note the uniform bound of 1/2). So we can set r = ε. Therefore Br/2 (x) is an open neighborhood of x
in ℓ2 \ A.

2) We have

dist(0, A) = inf ||x||2 = inf (1 + 2−j ) = 1

x∈A j∈N

and for any x ∈ A we have x = (1 + 2−j )e(j) for some j ∈ N and

||x||2 = 1 + 2−j > 1.

Hence the infimum is not attained in A.

Problem 6.2
(1) Via a mollification argument, we may assume that ϕ ∈ Cc∞ (B1 (0)). As in the hint, we let η ∈
Cc∞ (B1 (0)) with η = 1 on B1/2 (0) and define fj (x) = ϕ(x)(1 − η(jx) ). Then fj converges pointwise to
| {z }
=:ηj (x)
ϕ on U and |fj (x)| ≤ (1 + ∥η∥L∞ )|ϕ(x)|. By dominated convergence, it follows that fj → ϕ in Lp (U ).
Furthermore,
∂i fj − ∂i ϕ = ∂i ϕ(1 − ηj ) − ϕ∂i ηj − ∂i ϕ = −∂i ϕηj − ϕ∂i ηj .
Clearly ηj → 0 in Lp (U ) and as ∂i ηj (x) = j∂i η(jx), we deduce with the substitution y = jx
ˆ ˆ ˆ
p p p p−2
|ϕ(x)|j∂i η(jx)| dx ≤ ∥ϕ∥L∞ j |η(jx)| dx = j |η(y)|p dy → 0 as j → ∞.
U U U

Here we used p < 2. We deduce that

∥fj − ϕ∥W 1,p (U ) = ∥∇fj − ∇ϕ∥Lp (U ) + ∥fj − ϕ∥Lp (U ) → 0 as j → ∞.

As fj ∈ Cc∞ (U ) for j ∈ N, we deduce ϕ ∈ W01,p (U ).

1
 (2) First of all, notice that
ˆ ˆ 1 ˆ 1 ˆ 1 2
1
|f (x)|2 dx = 2π ln(1 − ln r)2 rdr ≤ 2π (1 − ln r)2 rdr ≤ 2π 1− √ rdr < ∞,
B1 (0) 0 0 0 r
√
where we used polar coordinates and ln(r) ≤ min{r, r}. Next, notice
1 1 x
∇f (x) = −
|x| 1 − ln |x| |x|
for all x ∈ B1 (0) \ {0}. In particular, using polar coordinates and the substitution et = r
ˆ ˆ 1 ˆ 0
2 1 1 1
|∇f (x)| dx = 2π 2
dr = 2π 2
dt < ∞.
B1 (0) 0 r (1 − ln r) −∞ (1 − t)

It remains to check that ∇f is actually the weak derivative of f . For that, notice that gk = min{f, k} ∈
W 1,2 (B1 (0)) by Exercise 5.4 (3). Then
∥gk − f ∥L2 + ∥∇gk − ∇f ∥L2 → 0
by dominated convergence. We concude that gk is a Cauchy sequence in W 1,2 (B1 (0)) and by completeness
it converges. As gk → f in L2 , we deduce that f ∈ W 1,2 (U ).

(3) We define fk = k −1 min{f, k} and let ϕk = (1 − fk )ϕ. It holds ϕk ∈ W 1,2 (U ) by Exercise 5.4
(3) with ϕk = 0 in a neighborhood of 0. Hence ηj ∗ ϕk ∈ Cc∞ (U ) for standard mollifiers ηj converge in
W 1,2 (U ) to ϕk . We conclude ϕk ∈ W01,2 (U ). Moreover,
∥ϕk − ϕ∥L2 (U ) = ∥fk ϕ∥L2 (U ) ≤ k −1 ∥f ϕ∥L2 (U ) ≤ k −1 ∥ϕ∥L∞ (U ) ∥f ∥L2 (U ) → 0
and
∇ϕk (x) − ∇ϕ(x) = −∇fk ϕ − fk ∇ϕ
from the product rule Lemma 2.37. We have |∇fk (x)| ≤ k −1 |∇f (x)|, which gives that ∥∇fk ∥L2 (U ) → 0.
As also ∥fk ∥L2 (U ) → 0, we conclude
∥ϕk − ϕ∥W 1,2 (U ) → 0.
We again conclude that ϕk ∈ W01,2 (U ) is a Cauchy sequence, converges in L2 to ϕ and thus ϕ ∈ W01,2 (U ).

Problem 6.3
(1) Observe that for |x| ≤ 1 and p ≥ 2 we have |x|p ≤ |x|2 . Therefore,
!p !p !2 !2
|a| |b| |a| |b|
p + p ≤ p + p = 1,
|a|2 + |b|2 |a|2 + |b|2 |a|2 + |b|2 |a|2 + |b|2
p p
and multiplying both sides by |a|2 + |b|2 yields
p
|a|p + |b|p ≤ (|a|2 + |b|2 ) 2 .
Hence, replacing, as suggested, a = x + y and b = x − y in the inequality above, we get

p
|x + y|p + |x − y|p ≤ |x + y|2 + |x − y|2 2
p
= (2|x|2 + 2|y|2 ) 2
p p
= 2 2 (|x|2 + |y|2 ) 2
1 p
p
1− 2 2
p
h i 
≤ 2 2 (|x|p + |y|p ) p 2 p
p p
= 2 2 2 2 −1 (|x|p + |y|p )
= 2p−1 (|x|p + |y|p ),

2
where the last inequality is just the Hölder inequality in R2 with the functions |x|p and the constant
function equal to 1, and the conjugate exponents satisfying 1p + 1p = 1.
2 p−2

(2) Let f, g ∈ Lp (U ) with 2 ≤ p < ∞ and ||f ||p = ||g||p = 1. Then,

ˆ
||f − g||pp = |f (x) − g(x)|p
ˆU
≤ 2p−1 (|f (x)|p + |g(x)|p ) − |f (x) + g(x)|p
U
= 2p−1 (||f ||pp + ||g||pp ) − ||f + g||pp
!
p f +g p
=2 1− .
2 p

1
ε p p f +g
Let 0 < ε < 2. Then, for δ = 1 − 1 − 2 , the inequality above shows that 2 > 1 − δ implies
p
||f − g||p < ϵ, hence Lp (U ) is uniformly convex for 2 ≤ p < ∞, as desired.

(3) For L1 (U ), consider two measurable subsets A, B ⊆ U such that A ∩ B = ∅ and Ln (A), Ln (B) > 0.
Such sets must exist since U is open, non-empty and Hausdorff, hence it must contain two disjoint balls,
for instance. Now observe that for all δ > 0 we have Lnχ(A)
A
1
= Lnχ(B)
B
1
= 1 and, moreover,
L (U ) L (U )
 
1 χA χB
n
+ n = 1 > 1 − δ,
2 L (A) L (B) L1 (U )

but also, since A ∩ B = ∅, we get

χA χB
n
− n
= 2,
L (A) L (B) L1 (U )

and hence L1 (U ) cannot be uniformly convex.

Similarly, for L∞ (U ), consider A, B ⊆ U with Ln (A ∩ B) > 0 and Ln (A − B) > 0. Such subsets
exist, for instance by taking balls B ⊊ A ⊆ U . Then, we have ||χA ||L∞ (U ) = ||χB ||L∞ (U ) = 1 and also
χA +χB
2 L∞ (U )
= 1 > 1 − δ for all δ > 0, because Ln (A ∩ B) > 0. On the other hand, since B ⊊ A, we
get ||χA − χB ||L∞ (U ) = 1, and hence L∞ (U ) cannot be uniformly convex either.

Problem 6.4
(1) Let p ∈ [1, ∞). By definition of the space W01,p (Rn ) we clearly have W01,p (Rn ) ⊆ W 1,p (Rn ). To show
the other inclusion, we consider f ∈ W 1,p (Rn ) and aim to construct a sequence j 7→ fj ∈ W01,p (Rn ) s.t.
fj → f in W 1,p (Rn ); by closedness of W01,p (Rn ) ⊆ W 1,p (Rn ) this implies
 f ∈ W01,p (Rn ). Following the
hint, we fix φ ∈ Cc∞ (Rn ) with φ = 1 on B 1 (0) and set fj (x) := φ x
j f (x). By the product rule (Lemma
2
2.38 together with L integrability of f and ∂i f ) we have fj ∈ W (Rn ) and since supp(fj ) is compact
p 1,p

due to the compact support of φ, Example (i) below Definition 2.45 justifies fj ∈ W01,p (Rn ). It suffices to
show fj → f in W 1,p (Rn ), i.e. fj → f in Lp (Rn ) and ∂i fj → ∂i f in Lp (Rn ). To prove the first assertion,
note
 
x
|fj (x)| = φ |f (x)| ≤ ∥φ∥L∞ (Rn ) |f (x)| ∈ Lp (Rn )
j
 
x 1 x
and fj → f pointwise a.e. since for all x ∈ Rn and j > 2 |x| we have j < 2 and hence φ j = 1. Thus,
dominated convergence implies fj → f in Lp (Rn ).

3
Concerning the derivative, the product rule yields
   
x 1 x
|∂i fj (x)| = ∂i f (x)φ + ∂i φ f (x) ≤ ∥φ∥L∞ (Rn ) |∂i f (x)| + ∥∂i φ∥L∞ (Rn ) |f (x)| ∈ Lp (Rn )
j j j
   
since f, ∂i f ∈ Lp (Rn ). Moreover, ∂i fj = ∂i f (x)φ xj + 1j ∂i φ xj f (x) → ∂i f (x) pointwise a.e. since, by
 
the same argument as above, ∂i f (x)φ xj → ∂i f (x) pointwise a.e. and
 
1 x 1 j→∞
∂i φ f (x) ≤ ∥∂i φ∥L∞ (Rn ) |f (x)| −−−→ 0.
j j j

Another application of dominated convergence yields ∂i fj → ∂i f in Lp (Rn ) and therefore in summary

fj → f in W 1,p (Rn ).

(2) Let f ∈ W 1,2 (Rn ). By part (1), there is a sequence k 7→ φk ∈ Cc∞ (Rn ) s.t. φk → f in W 1,2 (Rn ). Since
F : L2 (Rn ) → L2 (Rn ) is an isometry and therefore continuous, we find Fφk → Ff and F∂j φk → F∂j f
in L2 (Rn ). Hence, it suffices to show (F∂j φ)(ξ) = iξj (Fφ)(ξ) for functions φ ∈ Cc∞ (Rn ). To this end,
proceeding integration by parts yields
ˆ ˆ
1 −iξ·x 1  
(F∂j φ)(ξ) = n ∂j φ(x)e dx = n − φ(x)∂j e−iξ·x dx
(2π) 2 (2π) 2
Rn Rn
ˆ
1
= n iξj φ(x)e−iξ·x dx
(2π) 2
Rn
= iξj (Fφ)(ξ).

(3) ”⇒”: Let f ∈ W 1,2 (Rn ). Then ∂j f ∈ L2 (Rn ). Since F sends L2 (Rn ) to L2 (Rn ) we have F∂j f ∈
L2 (Rn ), so using (2) we find

|F∂j f | = |ξj | |Ff | ∈ L2 (Rn ).

Summing over 1 ≤ j ≤ n and equivalence of norms in Rn yields |ξ| (Ff )(ξ) ∈ L2 (Rn ). More precisely, by
Plancherel we find

∥f ∥L2 (Rn ) = ∥Ff ∥L2 (Rn )

(2)
∥∂j f ∥L2 (Rn ) = ∥F∂j f ∥L2 (Rn ) = ∥ξj Ff ∥L2 (Rn )

and hence
n
X n
X
∥f ∥2W 1,2 (Rn ) = ∥Ff ∥2L2 (Rn ) + ∥∂j f ∥L2 (Rn ) = ∥f ∥2L2 (Rn ) + ∥ξj Ff ∥2L2 (Rn )
j=1 j=1
ˆ Xn
= |(Ff )(ξ)|2 + |ξj |2 |(Ff )(ξ)|2 dξ
Rn j=1
ˆ
= (1 + |ξ|2 ) |(Ff )(ξ)|2 dξ
Rn

as desired.

”⇐”: Let f ∈ L2 (Rn ) and |ξ| (Ff )(ξ) ∈ L2 (Rn ). We need to show f is weakly differentiable with

4
weak derivative ∂j f ∈ L2 (Rn ). Part (2) suggests to set ∂j f := F̃(iξj (Ff )(ξ)) ∈ L2 (Rn ) and show that
∂j f is indeed the weak derivative of f : To this end, let φ ∈ Cc∞ (Rn ). Then
ˆ ˆ ˆ
1
∂j f (x)φ(x) dx = φ(x) n iξj (Ff )(ξ)eiξ·x dξ dx
(2π) 2
Rn Rn Rn
ˆ ˆ ˆ
1 1
= n φ(x) iξ j n f (y)e−iξ·y dyeiξ·x dξ dx
(2π) 2 (2π) 2
Rn n Rn
ˆ ˆ ˆR
1
= iξj φ(x)eiξ·x dxf (y)e−iξ·y dy dξ
(2π)n
Rn Rn Rn
ˆ ˆ ˆ
1  
= φ(x)∂ x j e iξ·x
dxf (y)e−iξ·y dy dξ.
(2π)n
Rn Rn Rn

Now in the inner integral we can proceed integration by parts and obtain
ˆ ˆ ˆ ˆ
1
∂j f (x)φ(x) dx = − ∂i φ(x)eiξ·x dxf (y)e−iξ·y dy dξ
(2π)n
Rn Rn Rn Rn
ˆ ˆ ˆ
1
=− f (y)e−iξ·y dy∂i φ(x)eiξ·x dx dξ
(2π)n
Rn Rn Rn
ˆ ˆ
1
=− n (Ff )(ξ)∂i φ(x)eiξ·x dx dξ
(2π) 2
Rn Rn
ˆ
= − (F̃Ff )(x)∂i φ(x) dx
Rn
ˆ
=− f (x)∂i φ(x) dx
Rn

as desired. Hence, f ∈ W 1,2 (Rn ) with ∂j f = F̃(iξj (Ff )(ξ)).

Alternative: Since F : L2 (Rn ) → L2 (Rn ) is a linear isometry, also F̃ : L2 (Rn ) → L2 (Rn ) is a linear
isometry since
F isom
F̃f = F F̃f = ∥f ∥L2 .
L2

Thus, by (the proof of) Problem 3.1 we know (Ff, Fg)L2 = (f, g)L2 = (F̃f, F̃g)L2 for all f, g ∈ L2 (Rn ),
i.e. the Fourier transform and its inverse preserve the L2 inner product. Now let φ ∈ Cc∞ (Rn ). Then
clearly φ ∈ W 1,2 (Rn ), so (2) implies

(F∂j φ)(ξ) = iξj (Fφ)(ξ) =⇒ ∂j φ(x) = (F̃Fφ)(x) = F̃[iξj (Fφ)(ξ)](x).

5
In particular, ξ 7→ iξj (Fφ)(ξ) ∈ L2 (Rn ) (what is also clear using direction ”⇒”). Choosing ∂j f :=
F̃(iξj (Ff )(ξ)) ∈ L2 (Rn ) we therefore find
ˆ
F isom
∂j f (x)φ(x) dx = (∂j f, φ)L2 = (F̃(iξj (Ff )(ξ)), φ)L2 = (F F̃(iξj (Ff )(ξ)), Fφ)L2
Rn
= (iξj (Ff )(ξ), Fφ)L2
ˆ
= iξj (Ff )(ξ)(Fφ)(ξ) dξ
Rn
ˆ
=− (Ff )(ξ)iξj (Fφ)(ξ) dξ
Rn
= −(Ff, iξj (Fφ)(ξ))L2
F̃ isom
= −(F̃Ff, F̃[iξj (Fφ)(ξ)])L2
= −(f, ∂j φ)L2
ˆ
= − f (x)∂j φ(x) dx
Rn

showing ∂j f = F̃(iξj (Ff )(ξ)) rally is the weak derivative of f .