The American Mathematical Monthly

ISSN: 0002-9890 (Print) 1930-0972 (Online) Journal homepage: https://www.tandfonline.com/loi/uamm20

p
Simple Proofs of the Uniform Convexity of L and
p
the Riesz Representation Theorem for L

Naoki Shioji

p
To cite this article: Naoki Shioji (2018) Simple Proofs of the Uniform Convexity of L and the
p
Riesz Representation Theorem for L , The American Mathematical Monthly, 125:8, 733-738, DOI:
10.1080/00029890.2018.1496762

To link to this article: https://doi.org/10.1080/00029890.2018.1496762

Published online: 28 Sep 2018.

Submit your article to this journal

Article views: 330

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=uamm20
 NOTES
Edited by Vadim Ponomarenko

Simple Proofs of the Uniform Convexity of Lp

and the Riesz Representation Theorem for Lp
Naoki Shioji

Abstract. We give elementary and self-contained proofs of the facts that the Lp space with
p
1 < p < ∞ is uniformly convex and its dual space is isometrically isomorphic to L p−1 .

1. INTRODUCTION. In 1936, Clarkson [2] defined a Banach space to be uniformly

convex if, for each ε > 0, there exists δ > 0 such that (u + v)/2 ≤ 1 − δ for each
u, v with u = v = 1 and u − v ≥ ε, and he showed that the Lp space with
1 < p < ∞ is uniformly convex. To show it, he obtained the so-called Clarkson’s
inequalities [2, Theorem 2]. Two of them are the following:

u + vp + u − vp ≤ 2p−1 (up + vp ), u, v ∈ Lp with 2 ≤ p < ∞, (1)

  
u + vp + u − vp ≤ 2(up + vp )p −1 , u, v ∈ Lp with 1 < p ≤ 2, (2)

where p  is the Hölder conjugate of p, that is, the unique element in [1, ∞] with 1/p +
1/p  = 1. We can easily prove (1); however, the proof of (2) does not seem to be easy
to understand. In fact, some researchers gave alternative proofs; see [4–8, 10, 11, 13].
Although the uniform convexity is an important concept and the Lp space with 1 <
p < ∞ is a crucial example, the difficulty of the proof of (2) seems to be a barrier to
understanding.
Throughout this note, let X be a nonempty set, let F be a σ -algebra on X and
let μ : F → [0, ∞] be a measure on F . We always assume μ(X) > 0. As usual,
we understand Lp (X) for 1 ≤ p < ∞ (respectively, L∞ (X)) to be the set of all
measurable
 functions from X into K, where K = R or K = C, that satisfy u ≡
( X |u|p dμ)1/p < ∞ (respectively, u ≡ inf{c ≥ 0 : μ({x ∈ X :
|u(x)| > c}) = 0} < ∞). We denote by (Lp (X))∗ the space of all bounded linear
functionals from Lp (X) into K. In this note, instead of Clarkson’s inequalities, we
show that in Lp (X) with 1 < p < ∞, the pth power of the norm is uniformly convex
in the sense of [19] on each bounded ball. As an application, we prove the Riesz
representation theorem for Lp (X) with 1 ≤ p < ∞. Our proofs are elementary and
self-contained.

Theorem 1. Let 1 < p < ∞. Then for each c > 0 and ε > 0 there exists δ > 0 such
that
 
 u + v p up + vp
 
 2  ≤ 2
−δ (3)

doi.org/10.1080/00029890.2018.1496762
MSC: Primary 46B10, Secondary 47J30

October 2018] NOTES 733

for each u, v ∈ Lp (X) with u, v ≤ c and u − v ≥ ε. In particular, Lp (X) is
uniformly convex.

Thanks to Hölder’s inequality, for 1 ≤ p < ∞, we can consider the linear mapping

 : Lp (X) → (Lp (X))∗ defined by


((v))(u) = uv dμ for v ∈ Lp (X) and u ∈ Lp (X). (4)
X

The following are the Riesz representation theorems for Lp (X) with 1 ≤ p < ∞; in

each of them the correspondence from Lp (X) into (Lp (X))∗ is given by (4).

Theorem 2. Let 1 < p < ∞. Then Lp (X) is isometrically isomorphic to (Lp (X))∗ .

As a consequence of Theorem 
2, we have the following. We say μ is σ -finite if there
exists {Xn } ⊂ F0 such that X = n∈N Xn , where F0 = {A ∈ F : μ(A) < ∞}.

Theorem 3. If μ is σ -finite, then L∞ (X) is isometrically isomorphic to (L1 (X))∗ .

We note that if μ is not σ -finite, then L∞ (X) may not be isometrically isomorphic
to (L1 (X))∗ ; see [9, p. 406], [15, Theorem 2]. Regarding the proof of Theorem 2, it is

easy to see that  : Lp (X) → (Lp (X))∗ is linear and norm preserving; see Lemma 4
below. The essential part is the surjectivity of . Its standard proof depends on the
Radon–Nikodym theorem; see, for example, [14, Theorem 6.16]. Some of its proofs
depend on the uniform convexity of Lp (X) for 1 < p < ∞; see [1, Theorem 2.44],
[7, Theorem 9.14], [9, Theorem 2], [17, Section 2], and [18, Theorem 5.4.3]. In the
proof of [7, Theorem 9.14], the Milman–Pettis theorem, the fact that every uniformly
convex Banach space is reflexive [12] was additionally used. The proof in [17, Sec-
tion 2] is essentially the same as that in this note but it deeply depends on Clarkson’s
inequalities.

2. PROOFS OF THEOREMS. Although we can deduce Theorem 1 from the uni-

form convexity of Lp (X) with 1 < p < ∞ and [19, Theorem 4.1 (ii)], we give a direct
and elementary proof by using the argument in [7, Proposition 9.9].

Proof of Theorem 1. We define α : K2 → [0, ∞) by

 
|z1 |p + |z2 |p  z1 + z2 p
α(z) = − for each z = (z1 , z2 ) ∈ K2 .
2 2 

We have α(z) > 0 for each z ∈ K2 with z1  = z2 . In fact, for z ∈ K2 with z1  = z2 , we

have |(z1 + z2 )/2|p ≤ ((|z1 | + |z2 |)/2)p ≤ (|z1 |p + |z2 |p )/2; in the case |z1 | = |z2 |,
the first inequality is strict since |(z1 + z2 )/2| < (|z1 | + |z2 |)/2, and in the case |z1 |  =
|z2 |, the second inequality is strict by the strict convexity of s → |s|p : [0, ∞) →
[0, ∞) for 1 < p < ∞. Let c > 0 and ε > 0. We choose 0 < η < εp /(2cp ). By the
compactness of the set D ≡ {z ∈ K2 : |z1 |p + |z2 |p = 1, |z1 − z2 |p ≥ η}, we have
θ ≡ inf{α(z) : z ∈ D} > 0. We will show that

α(z)
|z1 |p + |z2 |p ≤ for each z ∈ K2 with |z1 − z2 |p ≥ η(|z1 |p + |z2 |p ). (5)
θ

734 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 125
Let z ∈ K2 with |z1 − z2 |p ≥ η(|z1 |p + |z2 |p ). We may assume z  = (0, 0). Setting t =
(|z1 |p + |z2 |p )1/p and noting |z1 /t|p + |z2 /t|p = 1 and |z1 /t − z2 /t|p ≥ η, we have
θ ≤ α(z/t) = α(z)/t p , which yields (5). Let u, v ∈ Lp (X) with u ≤ c, v ≤ c
and u − v ≥ ε. Setting M = {x ∈ X : |u(x) − v(x)|p ≥ η(|u(x)|p + |v(x)|p )} and
using (5) and α(z) ≥ 0 for each z ∈ K2 , we obtain
    
 u − v p
ε ≤
p
|u − v| dμ = p
|u − v| dμ + 2 p p 
 2  dμ
X Mc M
 
|u|p + |v|p
≤η (|u| + |v| ) dμ + 2
p p p
dμ
Mc M 2
  p   
2p−1 |u| + |v|p  u + v p
≤ 2ηc +
p
− dμ,
θ X 2 2 

which yields (3) with δ = (εp − 2ηcp )θ/2p−1 .

It is easy to see that  has the following property; we omit the proof.

Lemma 4. For each 1 ≤ p < ∞ the correspondence  : Lp (X) → (Lp (X))∗

defined by (4) is linear, and satisfies (v) = v for each v ∈ Lp (X).

Now, we give the proof of Theorem 2. It is enough to show that  : Lp (X) →
(L (X))∗ is surjective. First, we consider the case K = R.
p

Proof of Theorem 2 in the case K = R. Fix any F ∈ (Lp (X))∗ . We define


1
Iu = |u|p dμ − F (u) for u ∈ Lp (X) and m= inf I u.
p X u∈Lp (X)

Since 1 < p < ∞ and I u ≥ up /p − F u for each u ∈ Lp (X), we have
m > −∞. The functional I has a unique minimizer as follows.

Lemma 5. There is a unique v ∈ Lp (X) that satisfies I v = m.

We give a proof of this lemma later and proceed now to prove the theorem. Let

v ∈ Lp (X) be the element as in Lemma 5. We have |v|p−2 v ∈ Lp (X). Let u ∈ Lp (X)
and define

1
gu (t) = I (v + tu) = |v + tu|p dμ − F (v + tu), t ∈ R.
p X

We set f (t, x) = (1/p)|v(x) + tu(x)|p for (t, x) ∈ R × X. Since 1 < p < ∞, t →

f (t, x) is differentiable and ∂f/∂t (t, x) = |v(x) + tu(x)|p−2 (v(x) + tu(x))u(x). By
Young’s inequality and the convexity of s → |s|p : R → R, we have
   p
 ∂f  1 1 2 p
+ 
 (t, x) ≤ |v(x) + tu(x)| + |u(x)| = 
p p v(x) tu(x)  + 1 |u(x)|p
 ∂t  p p p   2  p
 
2p−1 2p−1 T p 1
≤ |v(x)|p
+ + |u(x)|p ≡ βT (x)
p p p

October 2018] NOTES 735

for (t, x) ∈ (−T , T ) × X with T > 0. Since βT ∈ L1 (X) for each T > 0, by differen-
tiation under integral sign (see, e.g., [3, Theorem 2.27]), we have
 
∂f
gu (t) = (t, ·) dμ − F (u) = |(v + tu)|p−2 (v + tu)u dμ − F (u)
X ∂t X

for t ∈ R. From Lemma 5, we have gu (0) = mint∈R gu (t), which yields gu (0) = 0.
Since u ∈ Lp (X) is arbitrary, we obtain

F (u) = u|v|p−2 v dμ for each u ∈ Lp (X).
X

Hence, we have shown (|v|p−2 v) = F .

We give the proof of Lemma 5.

Proof of Lemma 5. We define Cn = {u ∈ Lp (X) : I u ≤ m + 1/n} for each n ∈ N.

We can easily see that Cn is nonempty and closed, and C1 ⊃ C2 ⊃ · · · . We will show
diam Cn → 0. Since 1 < p < ∞ and up /p − F u ≤ m + 1 for each u ∈ C1 ,
we have c ≡ sup{u : u ∈ C1 } < ∞. Let ε > 0. From Theorem 1, there is δ > 0
satisfying (3). Let n ∈ N with p/n < δ. For each u, v ∈ Cn , we have
 
1 1 u+v
u, v ≤ c, I u ≤ m + , I v ≤ m + and I ≥ m.
n n 2

From these inequalities, we obtain

 p
up + vp p 
 u + v

u, v ≤ c and − ≤ for each u, v ∈ Cn ,
2 n 2 

which, by (3), yields diam Cn ≤ ε for each n ∈ N with p/n < δ. Thus, we have shown
diam Cn → 0. Hence,
by Cantor’s intersection theorem (see, e.g., [16, Theorem C
in Section 12]), ∞ n=1 n consists of exactly one point, and the point is the unique
C
minimizer of I .

Next, we consider the case K = C.

Proof of Theorem 2 in the case K = C. Fix F ∈ (Lp (X))∗ . We define


1
Iu = |u|p dμ − Re F (u) for u ∈ Lp (X) and m = inf I u.
p X u∈Lp (X)

By the obvious analog of Lemma 5, there is a unique v ∈ Lp (X) such that I v = m.


We have |v|p−2 v̄ ∈ Lp (X). Let u ∈ Lp (X) and define gu (t) = I (v + tu) and hu (t) =
I (v − itu) for t ∈ R, where i is the imaginary unit. We note


gu (t) = |v + tu|p−2 (Re(uv̄) + t|u|2 ) dμ − ReF (u),
X

hu (t) = |v − itu|p−2 (Im(uv̄) + t|u|2 ) dμ − ImF (u)
X

736 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 125
 t ∈p−2
for R. From I v = m, we have gu (0) = 0 and hu (0) = 0, which yield F (u) =
X
u|v| v̄ dμ. Since u ∈ Lp (X) is arbitrary, we obtain (|v|p−2 v̄) = F .


Remark 6. In the case K = C, if we define  : Lp (X) → (Lp (X))∗ by (v)(u) =
p
X
uv̄ dμ for v ∈ L (X) and u ∈ Lp (X), then by the proof above, we can see that
the minimizer v of I satisfies (|v|p−2 v) = F . Thus  is norm preserving, surjective,
and conjugate linear, i.e., (av1 + bv2 ) = ā(v1 ) + b̄(v2 ) for a, b ∈ C and v1 , v2 ∈

Lp (X).

Using Theorem 2, we give a proof of Theorem 3. Although the proof of [1,

Theorem 2.45] almost works as well for that of Theorem 3, it is assumed X is an
open set in a Euclidean space. So we prove Theorem 3 briefly by a similar argument
as in [1, Theorem 2.45].

Proof of Theorem 3. We denote by  · p the norm of Lp (X) for 1 ≤ p ≤ ∞. Fix

F ∈ (L1 (X))∗ . First, we consider the case μ(X) < ∞. Let 1 < p ≤ 2. We have

|F (u)| ≤ F u1 ≤ F μ(X)1/p up for each u ∈ Lp (X),

which yields
 F ∈ (Lp (X))∗ . By Theorem 2, there exists vp ∈ Lp (X) such that

F (u) = X uvp dμ for each u ∈ Lp (X), and we have vp p ≤ F μ(X)1/p . Let
1 < p < q ≤ 2. Since X uvp dμ = X uvq  dμ for each u ∈ L2 (X), we have vp =
vq  a.e. We set v = v2 . We will show v ∈ L∞ (X) and v∞ ≤ F . Let c ∈ (0, ∞)

and Ec = {x ∈ X : |v(x)| ≥ c}. For each 1 < p ≤ 2, we have cμ(Ec )1/p ≤ vp ≤

F μ(X)1/p . If μ(Ec ) > 0, then letting p → 1, we obtain
 c ≤ F . Thus, we have
shown v ∈ L∞ (X) and v∞ ≤ F . Since F (u) = X uv dμ for each u ∈ L2 (X)
and L2 (X) is dense in L1 (X), we have F (u) = X uv dμ for each u ∈ L1 (X). By
Lemma 4, we obtain F  = v∞ . Hence, the proof of the case μ(X) < ∞ is com-
plete.
Next, we consider the general case. Since μ is σ -finite, there is a partition {Xn } ⊂
F0 of X. By the case μ(X) < ∞, for each n ∈ N, there is wn ∈ L∞ (X) such that
wn = 0 on X
\ Xn , wn ∞ ≤ F , and F (u1Xn ) = X uwn dμ for each u ∈ L1 (X).
We set w = ∞ ∞
n=1 wn ; we have w ∈ L (X) and w∞ ≤ F . Using the Lebesgue
convergence theorem, we have


∞ ∞ 
 
F (u) = F (u1Xn ) = uwn dμ = uw dμ for each u ∈ L1 (X).
n=1 n=1 X X

By Lemma 4, we have F  = w∞ . This completes the proof.

ACKNOWLEDGMENTS. The author is grateful to the anonymous referees and Professor Susan Jane Colley
for their careful reading of the manuscript, valuable comments, and suggestions.

REFERENCES

[1] Adams, R. A., Fournier, J. J. F. (2003). Sobolev Spaces, 2nd ed. Pure and Applied Mathematics, Vol.
140. Amsterdam: Elsevier.
[2] Clarkson, J. A. (1936). Uniformly convex spaces. Trans. Amer. Math. Soc. 40(3): 396–414.
[3] Folland, G. B. (1999). Real analysis, Modern Techniques and Their Applications, 2nd ed. Pure and
Applied Mathematics. New York: John Wiley & Sons.

October 2018] NOTES 737

 [4] Friedrichs, K. O. (1970). On Clarkson’s inequalities. Comm. Pure Appl. Math. 23(4): 603–607.
[5] Hanche-Olsen, H. (2006). On the uniform convexity of Lp . Proc. Amer. Math. Soc. 134(8): 2359–2362.
[6] Hanner, O. (1956). On the uniform convexity of Lp and l p . Ark. Mat. 3(3): 239–244.
[7] Komornik, V. (2016). Lectures on Functional Analysis and the Lebesgue Integral. London: Springer-
Verlag.
[8] Maligranda, L., Persson, L. E. (1992). On Clarkson’s inequalities and interpolation. Math. Nachr.
155(1): 187–197.
[9] McShane, E. J. (1950). Linear functionals on certain Banach spaces, Proc. Amer. Math. Soc. 1(3): 402–
408.
[10] Meir, A. (1984). On the uniform convexity of Lp spaces, 1 < p ≤ 2. Illinois J. Math. 28(3): 420–424.
[11] Ramaswamy, S. (1978). A simple proof of Clarkson’s inequality. Proc. Amer. Math. Soc. 68(2): 249–
250.
[12] Ringrose, J. R. (1959). A note on uniformly convex spaces. J. London Math. Soc. 34(1): 92.
[13] Roselli, P., Willem, M. (2002). A convexity inequality, Amer. Math. Monthly. 109(1): 64–70.
[14] Rudin, W. (1987). Real and Complex Analysis, 3rd ed. New York: McGraw-Hill.
[15] Schwartz, J. (1951). A note on the space L∗p , Proc. Amer. Math. Soc. 2(2): 270–275.
[16] Simmons, G. F. (1963). Introduction to Topology and Modern Analysis. New York: McGraw-Hill.
[17] Takáč, P. (2010). Variational methods and linearization tools towards the spectral analysis of the p-
Laplacian, especially for the Fredholm alternative. Proceedings of the 2007 Conference on Variational
and Topological Methods: Theory, Applications, Numerical Simulations, and Open Problems. Electron.
J. Differ. Equ. Conf. 18: 67–105.
[18] Willem, M. (2013). Functional analysis, Fundamentals and Applications. New York: Birkhäuser.
[19] Zălinescu, C. (1983). On uniformly convex functions. J. Math. Anal. Appl. 95(2): 344–374.

Department of Mathematics, Faculty of Engineering, Yokohama National University, Tokiwadai, Hodogaya,

Yokohama 240-8501, Japan
[email protected]

738 
c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 125