LU - Department of Mathematics

Dr H. Gebran

Math 400 Exercises of chapter 4 – Statements and solutions Fall 2022-2023

1. Let E be a normed space and let C ⊂ E be convex.

(a) Show that C is weakly closed if and only if it is strongly closed.

(b) Show that the weak closure of C is equal to the strong closure of C.
(c) Let (xn ) be a sequence of E converging weakly to x. Show that there exists a sequence
(yn ) made of convex combinations of the xn that converges strongly to x.

Solution. (a) It should be clear that a weakly closed set is strongly closed. Suppose now that
C is strongly closed. We prove that E\C is weakly open. Let x0 ∈ E\C. Then {x0 } and C are
disjoint. Since {x0 } is compact and convex and C is closed and convex, by the Hahn-Banach
theorem (second geometric form), there exists f ∈ E ∗ and α ∈ R such that f (x0 ) < α < f (x)
for all x ∈ C. Let U = f −1 (] − ∞, α[). Then U is a weak neighborhood of x0 contained in
E\C. It follows that E\C is a weak neighborhood of x0 . Since x0 was arbitrary, we conclude
that E\C is weakly open.
weak
b) Let C denote the weak closure of C. This set is a weakly closed set that contains C.
Therefore, it is a strongly closed set that contains C. But C is the smallest strongly closed set
weak
that contains C. Therefore C ⊂ C . Now conversely, C is strongly closed and convex. By
weak
(a), it is weakly closed. Since C is the smallest weakly closed set containing C, we conclude
weak
that C ⊂ C.

Remark. If C is not convex, we don’t necessarily have equality. For example, in an infinite
dimensional normed space the weak closure of the unit sphere is the unit ball, whereas its strong
closure is itself.

(c) Let (xn ) be sequence of E converging weakly to some x. Let C = co({x1 , x2 , . . .}). Since
x belongs to the weak closure of {x1 , x2 , . . .}, it belongs to the weak closure of its convex hull,
weak
that is to C . By the previous question this is just C. Since the strong topology is metrizable,
there exists a sequence of C that converges strongly to x.

Remark. We can choose the sequence (yn ) so that yn ∈ co({xn , xn+1 , . . .}). Try to prove this
by modifying slightly the above proof.

2. Let E and F be two Banach spaces and let T : E → F be a linear operator.

(a) Show that the following conditions are equivalent

(i) T : (E, ∥ · ||) → (F, || · ||) is continuous (T is continuous strong-strong).
(ii) T : (E, σ(E, E ∗ )) → (F, σ(F, F ∗ )) is continuous (T is continuous weak-weak).
(iii) T : (E, || · ||) → (F, σ(F, F ∗ )) is continuous (T is continuous strong-weak).
(b) Show that the following two conditions are equivalent.
(iv) T : (E, σ(E, E ∗ )) → (F, ∥ · ∥) is continuous (T is continuous weak-strong).
(v) T is continuous strong-strong and T (E) is finite dimensional (T is a finite rank bounded
linear operator).

1
 Solution. (a) Note that (i) ⇒ (iii) and (ii) ⇒ (iii) even if T is not linear.
(i) ⇒ (ii). Consider the following diagram.
T
(E, σ(E, E ∗ )) - (F, σ(F, F ∗ ))

φ φf = f
f ◦T
- ?
R
By Proposition 4.2, we have to prove that φf ◦ T is continuous for every f ∈ F ∗ . Now,

(φf ◦ T )(x) = ⟨f, T x⟩.

As a function of x ∈ E, this is a linear map which is bounded since

|(φf ◦ T )(x)| ≤ ||f ||||T x|| ≤ ||f ||||T ||||x||

for every x ∈ E.
(iii) ⇒ (i). We show that the graph of T is strongly closed. Let xn → x and T xn → y. Then
xn ⇀ x and T xn ⇀ y. Weak continuity of T implies that T xn ⇀ T x. By uniqueness of weak
limits, T x = y. This proves that G(T ) is strongly closed and so by the closed graph theorem, T
is strongly continuous.

(b) (v) ⇒ (iv). Suppose that T is continuous strong-strong and dim T (E) < ∞. Let O be a
strongly open subset of T (E). Since T (E) is finite dimensional, O is weakly open. But we know
from part (a) that T is continuous weak-weak. Therefore T −1 (O) is weakly open. It follows that
T is continuous weak-strong.
(iv) ⇒ (v). Let T be continuous weak-strong. It follows that T is continuous weak-weak. Let
ε > 0 be given. Since T is continuous weak-strong at 0, there exists a weak neighborhood V of 0
such that ||T x|| < ϵ for all x ∈ V . We can assume that V is of the form V = {x ∈ E; |fi (x)| <
δ, i = 1, . . . , k}. Let M = {x ∈ E; fi (x) = 0, ∀i = 1, . . . k}. Then T vanishes on M (since
||T (αx)|| < ε for all α ∈ R and all x ∈ M ). By a remark in the lectures, codim M < ∞. So
there exists a finite dimensional subspace G of F such that F = M ⊕ G.It follows that T (G)
is finite dimensional. But according to what we said, T (E) = T (G). Therefore T (E) is finite
dimensional.

3. Let E be a vector space and let φ, φ1 , . . . , φn : E → R be linear functionals such that for every
x∈E
[∀i ∈ {1, . . . , n} φi (x) = 0] =⇒ [φ(x) = 0].
Show that φ is a linear combination of φ1 , . . . , φn . Hint: Let F : X → Rn+1 be defined by
F (x) = (φ(x), φ1 (x), . . . , φn (x)). The assumption implies that a = (1, 0 . . . , 0) ∈
/ R(F ) where
R(F ) = ImF is the range of F . Equip Rn+1 with any norm (all norms are equivalent). Use the
Hahn-Banach theorem to separate {a} and R(F ).

Solution. Note that R(F ) is closed subspace of Rn+1 since it is finite dimensional. Thus, by
the Hahn-Banach theorem we can separate strictly {(1, 0, . . . , 0)} and R(F ). This means that
there exits a linear f : Rn+1 → R (f is automatically continuous) and α ∈ R such that

f (1, 0, . . . , 0) < α < f (x) ∀x ∈ R(F ).

Since R(F ) is subspace, it follows that f = 0 on R(F ). Now since f is linear it can be written
as
f (x0 , x1 , . . . , xn ) = λ0 x0 + λ1 x1 + · · · + λn xn .

2
 Therefore
λ0 φ(x) + λ1 φ1 (x) + · · · + λn φn (x) = 0 ∀x ∈ E.
and f (1, 0, . . . , 0) = λ0 . Therefore λ0 < α < 0 and so λ0 ̸= 0. It follows that

λ1 λn
φ(x) = − φ1 (x) − · · · − φn (x) ∀x ∈ E.
λ0 λ0
This means that φ is a linear combination of φ1 , . . . , φn .

4. Let E be a normed space and let ξ : E ∗ → R be linear and continuous for the weak* topology
σ(E ∗ , E). Show ξ ∈ J(E) where J : E → E ∗∗ is the canonical injection. Hint: Use Exercise 3.

Solution. Observe that the result we want to show is the converse of the implication
ξ ∈ J(E) ⇒ ξ is continuous for the weak* topology (which follows the definitions).
Now, continuity of ξ implies that there exists a neighborhood V of 0 in the weak* topology such
that |ξ(f )| < 1 whenever f ∈ V . We may assume that V is of the form

V = {f ∈ E ∗ ; |f (xi )| < ε, ∀i = 1, . . . k}.

We claim that
[f (xi ) = 0 ∀i = 1, . . . k] =⇒ [ξ(f ) = 0].
Indeed, suppose that f (xi ) = 0 ∀i = 1, . . . k. Then for every n ∈ N ∗ , (nf )(xi ) = 0 for every
i = 1, . . . , k. It follows that |ξ(nf )| < 1 and so |ξ(f )| < n1 for every n. This implies that
ξ(f ) = 0.

Therefore, we have the implication

[(Jxi )(f ) = 0 ∀i = 1, . . . , k] =⇒ [ξ(f ) = 0] .

It follows from Exercise 3 that

k k
!
X X
ξ= αi Jxi = J αi xi = Jx
i=1 i=1
Pk
where x = i=1 αi xi .

5. Deduce from the Goldstine’s lemma that J(E) is dense in E ∗∗ for the weak* topology.

Solution. Let ξ ∈ E ∗∗ and let V be weak* neighborhood of ξ. If ξ = 0, then ξ ∈ BE ∗∗

and so by the Goldstine lemma, V meets J(BE ) ⊂ J(E). If not, then ξ/||ξ|| ∈ BE ∗∗ . We can
assume that V is of the form

V (ξ, f1 , . . . , fk , ε) = {η ∈ E ∗∗ ; |⟨η − ξ, fi ⟩| < ε, i = 1, . . . k}.

ξ ε ξ
Consider the set U = V ( ||ξ|| , f1 , . . . , fk , ||ξ|| ). Then, U is neighborhood of ||ξ|| in the weak*
topology. By the Goldstine lemma, U meets J(BE ) at a point Jx. Then, it is easy to see that
J(||ξ||x) ∈ J(E) ∩ V .

3
6. Show that if two normed spaces E and F are isomorphic and E is reflexive, then F is reflexive
as well. Hint. Use the Eberlein-Smulyan theorem.

Solution. Let T : E → F be an isomorphism. Let (yn ) be a bounded sequence of F . Let

xn = T −1 yn . Then (xn ) is also bounded. Since E is reflexive, there is a subsequence xnk that
converges weakly to some x. Since T is continuous weak-weak, T xnk converges weakly to T x.
This means that (ynk ) converges weakly to T x. By the Eberlein-Smulyan theorem, F is reflexive.

7. Let E be a normed space.

(a) Let M ⊂ E be a subspace. Show that the topology on M inherited from σ(E, E ∗ ) coincides
with σ(M, M ∗ ).
(b) Show that if E is reflexive and M ⊂ E is a closed subspace, then M is reflexive.
(c) Suppose that E is a Banach space. Show that E is reflexive if and only if E ∗ is reflexive.

Solution. (a) Let U ∈ σ(M, M ∗ ) and let x0 ∈ U . We can assume that U has the form

U = {x ∈ M ; |fi (x − x0 )| < ε, ∀i ∈ I}

where I is finite and fi ∈ M ∗ . By the Hahn-Banach theorem, we can extend each fi to an

element gi ∈ E ∗ . Let
V = {x ∈ E; |gi (x − x0 )| < ε, ∀i ∈ I}.
Then V ∈ σ(E, E ∗ ) and U = V ∩ M . This means that U is open in the topology on M inherited
from σ(E, E ∗ )
Conversely, let U be open in the subspace topology inherited from σ(E, E ∗ ). We may assume
that U = V ∩ M where

V = {x ∈ E; |gi (x − x0 )| < ε, ∀i ∈ I}.

Let fi be the restriction of gi to M . Then fi ∈ M ∗ and so

U = {x ∈ M ; |fi (x − x0 )| < ε, ∀i ∈ I}.

This means that U ∈ σ(M, M ∗ ).

(b) Observe that BM = M ∩BE . By Kakutani’s theorem, E reflexive implies that BE is compact
in σ(E, E ∗ ). Since M is a closed and convex subset of E, it is weakly closed. It follows that
M ∩ BE is weakly closed in the compact set BE and so it is itself compact for σ(E, E ∗ ) and
therefore for the topology σ(M, M ∗ ). By Kakutani’s theorem M is reflexive.

(c) Suppose first that E is reflexive. Then σ(E ∗ , E) = σ(E ∗ , E ∗∗ ). By Banach-Alaoglu-Bourbaki

BE ∗ is compact for σ(E ∗ , E), therefore it is compact for σ(E ∗ , E ∗∗ ) which is the weak topology
on E ∗ . By Kakutani’s theorem E ∗ is reflexive.
Suppose now that E ∗ is reflexive. According to what we said, E ∗∗ is reflexive. Since J : E →
J(E) is an isomorphism, J(E) is a Banach space and therefore it is strongly closed in E ∗∗ . Being
convex, it is closed in σ(E ∗∗ , E ∗∗∗ ) = σ(E ∗∗ , E ∗ ). By the Goldstine lemma, J(E) is dense in
E ∗∗ (for the topology σ(E ∗∗ , E ∗ )). Therefore, J(E) = E ∗∗ and so E is reflexive. Alternatively,
we can say that J(E) is reflexive as a closed subspace of a reflexive space. But two isomorphic
spaces are simultaneously reflexive. Therefore E is reflexive. Note that we had to assume that
E is complete because a reflexive space is complete.

4
 8. Let E be a Banach space. Suppose that σ(E ∗ , E) = σ(E ∗ , E ∗∗ ). Show that E is reflexive.

Solution. By Banach-Alaoglu-Bourbaki, the closed unit ball BE ∗ is compact for σ(E ∗ , E).
Therefore, it is compact for σ(E ∗ , E ∗∗ ) (the weak topology of E ∗ ). By Kakutani’s theorem, E ∗
is reflexive. By the previous exercise, E is reflexive,
Alternatively, we can reason as follows. Let ξ ∈ E ∗∗ . Then ξ : E ∗ → R is linear and continuous
for the weak topology. Therefore ξ is continuous for the weak* topology. By exercise 4, ξ ∈ J(E).
Since ξ was arbitrary, we have E ∗∗ = J(E). This means that E is reflexive.

9. Let E be an infinite dimensional normed space satisfying one of the following assumptions.
(i) E ∗ is separable.
(ii) E is reflexive.
Show that there exists a sequence in the unit sphere that converges weakly to 0.

Solution. Suppose first that E ∗ is separable. Then BE is metrizable in the weak topology.
Since E is infinite dimensional, BE is the weak closure of SE . Therefore every x ∈ BE is the
weak limit of a sequence of SE .
Suppose next that E is reflexive. Since E is infinite dimensional, we can choose a countable
linearly independent set {e1 , e2 , . . .}. Let M0 = span{e1 , e2 . . .} and M = M0 . Then M is
infinite dimensional, reflexive and separable. Therefore M ∗ is separable. By the first part, there
exits a sequence in SM ⊂ SE converging weakly to 0.

10. Let E = C([0, 1]) be equipped with the norm ||u||∞ = sup0≤t≤1 |u(t)|.
(a) Set for t ∈ [0, 1], δt (u) = u(t) . Show that δt ∈ E ∗ and compute ||δt ||.
(b) Deduce that if (un ) converges weakly to u in E, then (un ) converges pointwise to u .
(c) Show that the converse is not true by considering the sequence un (t) = n2 t(1 − t)n .
(d) Study the pointwise convergence of the sequence (vn ) defined by
(
−nt + 1 if 0 ≤ t ≤ n1
vn (t) =
0 if n1 < t ≤ 1.

(e) Deduce that C[0, 1] is not reflexive.

(f) Deduce that L∞ [0, 1] is also not reflexive.
Solution. (a) It is clear that δt is linear. Next, observe that |δt (u)| ≤ ||u||∞ . It follows that
δt is bounded and ||δt || ≤ 1. Now, if v is the constant function equal to 1, then ||v||∞ = 1 and
δt (v) = 1. It follows that ||δt || := sup||u||=1 |δt (u)| ≥ |δt (v)| = 1. Thus, ||δt || = 1.
(b) Let t ∈ [0, 1]. If un ⇀ u, then, f (un ) → f (u) for all f ∈ E ∗ . In particular, for δt (un ) → δt (u).
Therefore, un (t) → u(t) (in R). Since t was arbitrary, this means that (un ) converges pointwise
to u.
 n
1 n2 1
(c) A study of the variation of un shows that ||un |||∞ = un ( n+1 ) = n+1 1 − n+1 . Therefore
(un ) is unbounded and so cannot converge weakly.
(d) Draw the graph of vn for some n. Let t ∈ [0, 1]. If t = 0, then vn (0) = 1 and so vn (0) → 1.
If t > 0, choose n0 > 1t . Let n ≥ n0 . Then t > n1 and vn (t) = 0. Therefore vn (t) → 0 as
n → ∞. It follows that (vn ) converges pointwise to the the discontinuous function v defined by
(
1 if t = 0
v(t) =
0 if 0 < t ≤ 1.

5
 (e) Suppose that C[0, 1] is reflexive. Observe that ||vn || = 1 and so (vn ) is bounded. By
the Eberlein-Smulyan theorem, (vn ) has a subsequence (vnk ) that converges weakly to some
w ∈ C[0, 1]. By (b), (vnk ) converges pointwise to w. By (d), (unk ) converges pointwise to v.
Therefore, w = v. This is a contradiction because w is continuous and v is not.

(f) We know that C[0, 1] is complete. Therefore it is closed in L∞ [0, 1]. If L∞ [0, 1] was reflexive,
then C[0, 1] would have been reflexive, a contradiction.

Remark. Actually, the following condition are equivalent.

(i) un ⇀ u in C[0, 1].

(ii) un converges to u pointwise and ||un ||∞ is bounded.

Proof. (i)⇒ (ii). We already proved that weak convergence implies pointwise convergence.
We also know that a weakly convergent sequence is bounded.
(ii)⇒ (i). Suppose that (un ) converges pointwise to u and that (un ) is bounded. Let f ∈
C[0, 1]∗ . By the
R Riesz representation theorem, there exists a finite Borel measure on [0,1] such
that ⟨f, v⟩ = [0,1] v dµ for all v ∈ C[0, 1]. By the Lebesgue dominated convergence theorem
R R
[0,1] un dµ → [0,1] u dµ. Otherwise stated, ⟨f, un ⟩ → ⟨f, u⟩. Since f was arbitrary, we deduce
that un converges weakly to u.

11. Let (E, || · ||) be a normed space. Prove that the following conditions are equivalent.

(i) (E, || · ||) is uniformly convex.

x+y
(ii) ∀ε > 0 ∃δ > 0 ∀x, y ∈ BE ||x − y|| > ε =⇒ < 1 − δ.
2
x n + yn
(iii) For any two sequences (xn ) and (yn ) in SE , we have → 1 =⇒ ||xn − yn || → 0.
2
xn + yn
(iv) For any two sequences (xn ) and (yn ) in BE , we have → 1 =⇒ ||xn − yn || → 0.
2

x n + yn
Solution. (i) ⇒ (iii). Let (xn ) and (yn ) be two sequences in SE satisfying → 1.
2
Suppose that ||xn −yn || ↛ 0. Then, passing to a subsequence, we may assume that ||xn −yn || > ε
xn + yn
for some ε > 0 and all n ∈ N∗ . By (i), there exists δ > 0 such that < 1 − δ for all
∗
2
n ∈ N . Letting n → ∞, we get 1 ≤ 1 − δ, a contradiction.

(iii) ⇒ (i). Suppose by contradiction that

x+y
∃ε > 0 ∀δ > 0 ∃x, y ∈ SE ||x − y|| > ε and ≥ 1 − δ.
2

Taking δ = 1, 21 , 31 , . . ., we construct two sequences (xn ) and (yn ) in SE such that ||xn −yn || > ε
and xn +y
2
n
≥ 1 − n1 . But xn +y 2
n
≤ 1. It follows that xn +y2
n
→ 1. By (iii), we should
have ||xn − yn || → 0. But this contradicts the inequality ||xn − yn || > ε.

(ii) ⇒ (iv) is similar to (i) ⇒ (iii) and (iv) ⇒ (ii) is similar to (iii) ⇒ (i).

(iv) ⇒ (iii) is trivial since SE ⊂ BE .

6
 x n + yn
(iii) ⇒ (iv). Let (xn ) and (yn ) be two sequences in BE satisfying → 1 (or equivalently
2
||xn + yn || → 2). Since (||xn ||) is a bounded sequence of real numbers, it has a subsequence
(still denoted by the same symbol) converging to some a ≥ 0. By the triangle inequality,
2
2||xn ||2 + 2||yn ||2 − ||xn + yn ||2 ≥ 2||xn ||2 + 2||yn ||2 − ||xn || + ||yn || .


Now the left hand side is less or equal to 4 − ||xn + yn ||2 and the right hand side is equal to
2 2
[||xn || − ||yn ||]2 . Therefore, ||xn ||−||yn || ≤ 4−||xn +yn ||2 . It follows that ||xn ||−||yn || →
 

0 and so ||yn || → a as well. If a = 0, then, xn → 0 and yn → 0 and so || xn +y 2 || → 0,

n

contradicting our assumption. Therefore, a > 0. We already observed that

2
4 − ||xn + yn ||2 ≥ 2||xn ||2 + 2||yn ||2 − ||xn + yn ||2 ≥ ||xn || − ||yn || .


Since the two extreme sides converge to 0, it follows from the sandwich theorem that 2||xn ||2 +
2||yn ||2 − ||xn + yn ||2 → 0 and so ||xn + yn ||2 → 4a2 or, equivalently ||xn + yn || → 2a. It follows
that a = 1. Since ||xn || → a = 1, we have xn ̸= 0 from a certain rank and similarly for yn .
Therefore, the sequences ||xxnn || and ||yynn || are well defined from a certain rank and belong to SE .
yn
Now, on the first hand ||xxnn || + ||yn || ≤ 2. On the other hand, using the second form of the
triangle inequality we get

xn yn xn yn yn yn ||xn + yn || 1 1
+ = + − + ≥ − − ||yn ||.
||xn || ||yn || ||xn || ||xn || ||xn || ||yn || ||xn || ||xn || ||yn ||

xn yn
Since the right hand side converges to 2, the sandwich theorem implies that ||xn || + ||yn || → 2.
xn yn
It follows from (iii) that ||xn || − ||yn || → 0. Now again, by the second form of the triangle
inequality, we have

xn yn xn yn yn yn ||xn − yn || 1 1
− = − + − ≥ − − ||yn ||.
||xn || ||yn || ||xn || ||xn || ||xn || ||yn || ||xn || ||xn || ||yn ||

It follows that
||xn − yn || xn yn 1 1
≤ − + − ||yn || → 0.
||xn || ||xn || ||yn || ||xn || ||yn ||

Since ||xn || → 1, we finally get ||xn − yn || → 0.

12. Let E be a uniformly convex Banach space. Let K be a nonempty closed convex subset of E.
Let x ∈ E and
d = d(x, K) = inf ||x − a||.
a∈K

We would like to show that exists a unique y ∈ K such that ||x − y|| = d(x, K). Note that if
d = 0, then, x ∈ K and there is nothing to prove. So we assume that d > 0.

(a) Show that there exists a sequence (yn ) of K such that ||x − yn || → d.
x−yn 1
(b) Set zn = ||x−yn || and tn = ||x−yn || . Show that ||zn + zm || ≥ (tn + tm )d.
(c) Deduce that (zn ) is a Cauchy sequence. Hint. Reason by contradiction.
(d) Show that there exists a unique y ∈ K such that ||x − y|| = d(x, K).

7
 Solution. (a) By a fundamental property of the inf there exists yn such that
1
d ≤ ||x − yn || < d + .
n

(b) Note that zn = tn (x − yn ), ||zn || = 1 and that tn → d1 . Now, we can write

 
tn tm
zn + zm = tn (x − yn ) + tm (x − ym ) = (tn + tm ) (x − yn ) + (x − ym )
tn + tm tn + tm
  
tn tm
= (tn + tm ) x − yn + ym
tn + tm tn + tm
tn tm
Observe that the tn +tm yn + tn +tm ym ∈ K because K is convex. It follows that

||zn + zm || ≥ (tn + tm )d(x, K) = (tn + tm )d.

(c) Suppose that (zn ) is not a Cauchy sequence. Then, there exists ε > 0 such that ||zn −zm || > ε
for infinitely many couples (n, m). Let δ be as in the definition of uniform convexity. Then, by
the previous question and uniform convexity, we have
tn + tm zn + zm
d≤ <1−δ
2 2
for infinitely many couples (n, m). Letting n, m → ∞ we get 1 ≤ 1 − δ, a contradiction.

(d) Let z be the limit of (zn ). Then, first, ||z|| = 1. Since yn = x − t1n zn , we see that
yn → y := x − d1 z. Therefore ||x − yn || → ||x − y||. But ||x − yn || → d. So by uniqueness
limits d = ||x − y||. Observe that y ∈ K because K is closed. This proves the existence part.
Suppose that there is y ′ ∈ K, y ′ ̸= y such that d = ||x − y ′ ||. Let ε = ||y − y ′ ||. By a remark in
the lectures, there exists δ > 0 such that
u+v
||u|| = ||v|| = d and ||u − v|| ≥ ε =⇒ <d−δ
2
(this is the uniform roundness of the sphere of radius d). Now, ||(x − y) − (x − y ′ )|| = ε. It
′) ′
follows that (x−y)+(x−y
2 < d − δ. Or equivalently, x − y+y2 < d − δ. Since K is convex,
y+y ′
2 ∈ K. It follows that

y + y′
d = d(x, K) ≤ x − < d − δ,
2
a contradiction. This proves the uniqueness part.

13. Let (E, || · ||) be a uniformly convex Banach space. Let (xn ) be a sequence such that xn ⇀ x
and ||xn || → ||x||. Show that xn → x.

Solution. If x = 0, there is nothing to show. So we assume that x ̸= 0. Then, we can write

x x x
xn − x = ||xn || − ||xn || + ||xn || − x.
||xn || ||x|| ||x||

Taking norms, we get by the triangle inequality

xn x
||xn − x|| ≤ ||xn || − + ||xn || − ||x|| . (∗)
||xn || ||x||

8
 xn x yn +y
Let yn = ||xn || and y = ||x|| . Then ||yn || = ||y|| = 1. Also, 2 ⇀ y (check that). Therefore,

yn + y
1 = ||y|| ≤ lim inf .
2

But on the other hand,

yn + y 1

≤ ||yn || + ||y|| = 1
2 2
and so
yn + y
lim sup ≤ 1.
2
It follows that lim yn2+y = 1. Uniform convexity implies that ||yn − y|| → 0. Since ||xn || is
bounded, the first term of (∗) tends to zero. The second term tends to zero by assumption.

14. Let E be an infinite dimensional normed space and let φ(x) = ||x||.

(a) Find φ−1 (]0, 1[).

(b) Deduce that φ is not weakly continuous.

Solution. (a) φ−1 (]0, 1[) = {x ∈ E; 0 < ||x|| < 1}.

(b) It follows that φ−1 (]0, 1[) is bounded and therefore cannot be weakly open because the space
is infinite dimensional. So we found an open subset of R whose inverse image under φ is not
weakly open. This means that φ is not weakly continuous.

Remarks. 1) Actually more is true: φ is nowhere weakly continuous. Indeed, let x0 ∈ E be an

arbitrary element and let V =]||x0 || − 1, ||x0 || + 1[. Then V is a neighborhood of φ(x0 ) = ||x0 ||.
Now if U is a neighborhood of x0 in the weak topology, then U is unbounded. However, if
φ(U ) ⊂ V , then ||x|| = φ(x) < ||x0 || + 1 whenever x ∈ U . This means that U is contained a
ball, a contradiction. This means that φ is not continuous at x0 .

2) If either E ∗ is separable or E is reflexive, then φ is not even sequentially continuous (see

Exercise 9).

3) The space of sequences ℓ1 has the following surprising property. If a sequence in ℓ1 is weakly
convergent, then it is strongly convergent. This property is called the Schur property. See
Problem 8 in the book of Brézis. It follows that the function φ(x) = ||x||1 is weakly sequentially
continuous everywhere but nowhere weakly continuous.

4) In general, the function φ is only weakly lower semi-continuous, meaning that for any a ∈ R,
the set φ−1 (] − ∞, a]) is weakly closed. Do you see why?