See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/271514205

ORTHOGONAL PROJECTIONS IN A UNIFORMLY

CONVEX BANACH SPACE

Conference Paper · September 2013

CITATIONS READS

0 117

2 authors:

Dominik Mielczarek Ewa Szlachtowska

AGH University of Science and Technology in … AGH University of Science and Technology in …
28 PUBLICATIONS 33 CITATIONS 14 PUBLICATIONS 6 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Geometry of Banach spaces View project

All content following this page was uploaded by Ewa Szlachtowska on 29 January 2015.

The user has requested enhancement of the downloaded file.

 ORTHOGONAL PROJECTIONS IN A UNIFORMLY CONVEX BANACH
SPACE

DOMINIK MIELCZAREK, EWA SZLACHTOWSKA

Abstract. In a uniformly convex Banach space with a continuous generalized semi-inner prod-
uct X we investigate the relation of orthogonality in X and projections acting on X. We prove
that the decomposition theorem holds on a uniformly convex Banach space with a continuous
generalized semi-inner product . This result is presented in detail in Theorem 4. The main
results in this paper is Theorem 8. We prove the uniqueness of orthogonal linear projections.
For more results we refer to [1], [2], [3], [4] and references therein.

1. Generalized Semi-inner Products

Let X be a vector space over C. We call a function [·, ·]ϕ : X ×X → C a generalized semi-inner
product (g.s.i.p.) on a vector space X if it satisfies the following conditions:
(i) Linearity with respect to the first variable:
[αx + βy, z]ϕ = α[x, z]ϕ + β[y, z]ϕ for all α, β ∈ C and x, y, z ∈ X;
(ii) Positivity: [x, x]ϕ > 0 for all x ∈ X \ {0};
(iii) A generalization of the Cauchy-Schwarz inequality: there holds for some ϕ, ψ : R+ → R+
that
|[x, y]ϕ | ≤ ϕ([x, x]ϕ )ψ([y, y]ϕ ), x, y ∈ X
and the equality holds when x = y.
We require that
ϕ(0) = 0, ϕ(t) > 0 for t > 0
and
t
ψ(0) = 0, ψ(t) = for t > 0.
ϕ(t)
The importance of a generalized semi-inner product space (s.i.p.s.) is that every normed vector
space can be represented as a semi-inner product space so that the theory of operators on a
Banach space can be penetrated by Hilbert space type arguments.
Theorem 1. [5] Let [·, ·]ϕ be a g.s.i.p. on a vector space X over C. Then kxkϕ = ϕ([x, x]ϕ )
defines a norm on X.
Moreover let X be a vector space equipped with the norm k · k. If ϕ is surjective onto R+ ,
then there exists a g.s.i.p. on X such that k · kϕ = k · k.
In a normed vector space X we set S = {x ∈ X : kxk = 1} . We introduce additional properties
of a g.s.i.p.
A very convenient property of a g.s.i.p. is continuity with respect to the second variable.
A g.s.i.p. space X is called a continuous g.s.i.p. space when a generalized semi-inner product
satisfies the following additional condition:
For every x, y ∈ S,
(1) Re[y, x + λy]ϕ → Re[y, x]ϕ for all real λ → 0.
Define a relation on a s.i.p. space which may be called an orthogonality relation. Let x, y ∈ X.
We say that x is orthogonal to y and y is transversal to x if [y, x]ϕ = 0.
1991 Mathematics Subject Classification. 41A65, 46B20, 46B25.
Partially supported by the Polish Ministry of Sciences and Higher Education.
 9

On a normed space we also can study the orthogonality relation (in the sense of Birkhoff)
defined as follows:
A vector x is orthogonal to y in the sense of Birkhoff if
kx + λyk ≥ kxk for all λ ∈ C.
One more piece of notation: throughout the paper we write kxk = ϕ([x, x]ϕ ) and kxk∗ = ψ([x, x]ϕ ).
It is worth noting that orthogonality in the sense of Birkhoff is very close to the concept of an
element of best approximation. In a continuous g.s.i.p. space orthogonality relation is equivalent
to Birkhoff orthogonality relation.
Lemma 2. In a continuous g.s.i.p.s. x is orthogonal to y if and only if x is orthogonal to y in
the sense of Birkhoff.
Proof. Let x be normal to y. Thus,
kx + λykkxk∗ ≥ |[x + λy, x]ϕ | =
|[x, x]ϕ + λ[y, x]ϕ | = kxkkxk∗ .
Therefore, ||x + λy|| ≥ ||x|| for all complex λ.
Let [·, ·]ϕ be a continuous g.s.i.p. If kx + λyk ≥ kxk for all λ ∈ C, then
0 ≤ kx + λyk∗ kx + λyk − kx + λyk∗ kxk ≤
[x + λy, x + λy]ϕ − |[x, x + λy]ϕ | ≤
Re[x, x + λy]ϕ + Re {λ[y, x + λy]ϕ } − Re[x, x + λy]ϕ .
Therefore, Re {λ[y, x + λy]ϕ } ≥ 0.
For real λ we have
Re[y, x + λy]ϕ ≥ 0 for λ ≥ 0;
Re[y, x + λy]ϕ ≤ 0 for λ ≤ 0.
From the continuity condition, for real λ, we have
lim Re[y, x + λy]ϕ = Re[y, x]ϕ
λ→0
through positive values for λ → 0+ and through negative values for λ → 0− . Thus Re[y, x]ϕ = 0.
For imaginary λ, say λ = iλ1 with λ1 real, we obtain Re[iy, x]ϕ = 0, i.e. Im[y, x]ϕ = 0. Therefore,
[y, x]ϕ = 0. 
For the purpose of studing projections, we characterize strict convexity of X in terms of the
g.s.i.p.
Theorem 3. [5] The normed vector space X is striclty convex if and only if whenever
[y, y]ϕ
[x, y]ϕ = ϕ([x, x]ϕ ) , x, y 6= 0
ϕ([y, y]ϕ )
then y = αx for some α > 0.
To extend Hilbert space type argument we prove the decomposition theorem. The desired
statement follows after one impose an additional structure on a g.s.i.p. chiefly to guarantee the
existence of orthogonal vectors to closed subspaces.
Theorem 4. Let X be a uniformly convex Banach space with a continuous generalized semi-
inner product. Let M be a closed subspace of X. Then each x ∈ X can be uniquely decomposed
in the form x = y + z with y ∈ M and z ∈ M ⊥ = {u ∈ X : ∀ v ∈ M [v, u]ϕ = 0 }.
Proof. It is well known that, in a uniformly convex Banach space, for a closed vector subspace
M and a vector x 6∈ M , there exists a unique nonzero vector y ∈ M such that
||x − y|| = d(x, M ) = inf{||x − y 0 || : y 0 ∈ M }.
Let us set z = x − y. Then z is orthogonal to M .
10

In order to prove the uniqueness of the representation x = y + z we assume that x = y1 + z1 =

y2 + z2 where y1 , y2 ∈ M and z1 , z2 ∈ M ⊥ . It follows that z1 − z2 = y1 − y2 ∈ M . If
z1 − z2 ∈ M ∩ M ⊥ , then z1 − z2 = 0 and y1 = y2 . If z1 − z2 6∈ M ⊥ , then
0 = [z1 − z2 , z1 ]ϕ = [z1 , z1 ]ϕ − [z2 , z1 ]ϕ ≥ kz1 kkz1 k∗ − kz2 k kz1 k∗ ,
0 = [z2 − z1 , z2 ]ϕ = [z2 , z2 ]ϕ − [z1 , z2 ]ϕ ≥ kz2 kkz2 k∗ − kz1 k kz2 k∗ .
Therefore,
kz1 k = kz2 k and kz1 k kz2 k∗ = [z1 , z2 ]ϕ .
By the strict convexity of X, we obtain z1 = z2 . This implies that y1 = y2 . 

2. Orthogonal Projections
Let X be a uniformly convex Banach space with a continuous generalized semi-inner product
and let M be a closed subspace of X. Let P : X → M be a linear projection. We say that P is
orthogonal if (ker P )⊥ = M .
Then the following theorem holds.
Theorem 5. Let M be a closed subspace of a uniformly convex Banach space X with a contin-
uous generalized semi-inner product. Let P : X → M be a linear projection. If kP k = 1, then
P is orthogonal.
Proof. We shall show that (ker P )⊥ = M . For x ∈ X we have
(2) kP (x)k ≤ kxk.
Setting x equal to P x + λ(y − P y) in (2) we obtain
kP (P x + λ(y − P y))k ≤ kP x + λ(y − P y)k,
hence
kP xk ≤ kP x + λ(y − P y)k.
By virtue of Theorem 2, it is equivalent to the fact that P x is orthogonal to every z ∈ ker P .
Conversely, suppose that x ∈ (ker P )⊥ . Then [z, x]ϕ = 0 for z ∈ ker P . Hence [x − P x, x]ϕ = 0
and
kxkkxk∗ = [x − P x + P x, x]ϕ = [x − P x, x]ϕ + [P x, x]ϕ ≤ kP xkkxk∗ ≤ kxkkxk∗ .
By assumptions it follows that kxk = kP xk and kP xkkxk∗ = [P x, x]. By the strict convexity of
X, we obtain P x = x, and so x ∈ M . 
Now we can conclude that in a uniformly convex Banach space with a continuous g.s.i.p.
every orthogonal linear projection is a projection of norm one.
Theorem 6. Assume that X is a uniformly convex Banach space with a continuous generalized
semi-inner product and M is a closed subspace of X. Let P : X → M be a linear projection. If
P is orthogonal, then the norm of P is equal to one.
Proof. Let x ∈ X. Then P x − x ∈ ker P and
kP xkkP xk∗ = [P x, P x]ϕ = [P x − x + x, P x]ϕ =
= [P x − x, P x]ϕ + [x, P x]ϕ = [x, P x]ϕ .
Using the generalization of the Cauchy-Schwarz inequality we get
kP xk ≤ kxk,
hence kP k = 1. 
Lewicki and Skrzypek proved that minimal projections onto symmetric subspaces of smooth
Banach spaces are unique (see [2]). Now, we show an analogous theorem in a uniformly convex
Banach space X with a continuous g.s.i.p. In its proof we use the structure of a generalized
semi-inner product.
 11

Lemma 7. Let P : X → M be an ortogonal projection. Then P is a unique orthogonal projec-

tion.
Proof. Let Pi be an orthogonal projection (i = 1, 2). Hence (ker Pi )⊥ = M (i = 1, 2). Then
P1 x − P2 x ∈ M and
kP1 x − P2 xkkP1 x − P2 xk∗ = [P1 x − P2 x, P1 x − P2 x]ϕ =
= [P1 x − x + x − P2 x, P1 x − P2 x]ϕ =
= [P1 x − x, P1 x − P2 x]ϕ + [x − P2 x, P1 x − P2 x]ϕ = 0.
Consequently, we conclude that P1 x = P2 x, which completes the proof. 
We can write the following statement as the result of our previous considerations.
Theorem 8. Let X be a uniformly convex Banach space with continuous semi-inner product.
Let M be a closed subspace of X. If P : X → M is a linear projection such that kP k = 1, then
P is unique.

References
[1] P.A. Cojuhari, M.A Nowak: Projection-iterative methods for a class of difference equations, Integral Equa-
tions and Operator Theory 64 (2009), 155175.
[2] G. Lewicki, L. Skrzypek: Chalmers-Metcalf operator and uniqueness of minimal projections, J. Approx.
Theory 148 (2007), 71-91.
[3] G. Lumer: Semi-inner-product spaces, Trans. Amer. Math. Soc. 100 (1961), 29-43.
[4] D. Mielczarek, E. Szlachtowska, On the uniqueness of minimal projections in Banach spaces, Opusc. Math.
32(3) (2012), 579-590.
[5] H. Zhang and J. Zhang, Generalized Semi-inner Products and Applications to Regularized Learning, Journal
of Mathematical Analysis and Applications 372 (2010), 181-196.
Ewa Szlachtowska
szlachto@agh.edu.pl

AGH University of Science and Technology

Faculty of Applied Mathematics
al. Mickiewicza 30, 30-059 Krakw, Poland

Dominik Mielczarek
dmielcza@wms.mat.agh.edu.pl

AGH University of Science and Technology

Faculty of Applied Mathematics
al. Mickiewicza 30, 30-059 Krakw, Poland

View publication stats