ORTHOGONALITY IN BANACH SPACES VIA PROJECTIVE

TENSOR PRODUCT

KOUSIK DHARA, NARAYAN RAKSHIT, JAYDEB SARKAR, AND ARYAMAN SENSARMA

Abstract. Let X be a complex Banach space and x, y ∈ X. By definition, we say

that x is Birkhoff-James orthogonal to y if kx + λykX ≥ kxkX for all λ ∈ C. We
prove that x is Birkhoff-James orthogonal to y if and only if there exists a semi-inner
product φ on X such that kφk = 1, φ(x, x) = kxk2 and φ(x, y) = 0. A similar
result holds for C ∗ -algebras. A key point in our approach to orthogonality is the
representations of bounded bilinear maps via projective tensor product spaces.

1. Introduction
This paper deals with the notion of Birkhoff-James orthogonality [2, 8, 9] for vectors
in Banach spaces via projective tensor product of Banach spaces. Let X be a normed
linear space over the scalar field K, where K is either R or C. Suppose x, y ∈ X. We
say that x is Birkhoff-James orthogonal (or simply orthogonal ) to y if

kx + λykX ≥ kxkX ,

for all λ ∈ K. We denote this by x ⊥B y. If X is a Hilbert space, then this notion

coincides with the usual orthogonality, that is, x ⊥B y if and only if

hx, yiX = 0.

The objective of this note is to present an abstract characterization of orthogonality of

vectors in Banach spaces over K. We are motivated by Bhatia and Šemrl’s investigation
[3, Theorem 1.1 and Remark 3.1] of orthogonality in the setting of bounded linear
operators on Hilbert spaces:

Theorem 1.1. (Bhatia and Šemrl) Let S and T be bounded operators on a Hilbert
space H.
(i) Then S ⊥B T if and only if there exists a sequence {hn } of unit vectors such that
kShn k → kSk, and hShn , T hn i → 0 as n → ∞.

2010 Mathematics Subject Classification. 46B28, 47A30, 46B20, 46B22, 47B01, 47L05.
Key words and phrases. Orthogonality, projective tensor product, semi-inner product, bilinear
maps.
1
2 DHARA, RAKSHIT, SARKAR, AND SENSARMA

(ii) If we additionally assume that H is a finite dimensional Hilbert space, then

S ⊥B T if and only if there exists a unit vector h ∈ H such that
kShk = kSk and hSh, T hi = 0.

We treat the above Bhatia-Šemrl result (more specifically, part (ii) of Theorem 1.1)
as a paradigm and examine orthogonality of vectors in Banach spaces in terms of
semi-inner products. Our main result, in the setting of Banach spaces, is the following:

Theorem 1.2. Let X be a Banach space, x, y ∈ X, and x 6= 0. Then x ⊥B y if and

only if there exists a semi-inner product ϕ : X ⊕∞ X → K such that kϕk = 1 and
ϕ(x, x) = kxk2 and ϕ(x, y) = 0.

Actually, this also holds for normed linear spaces, but we will return to it at the end
of this paper.
As an immediate consequence of the above, we also prove an orthogonality result in
the setting of C ∗ -algebras (see Corollary 2.2). Here, for a pair of Banach spaces X and
Y , we define X ⊕∞ Y to be the Banach space
X × Y = {(x, y) : x ∈ X, y ∈ Y },
with the norm
k(x, y)k∞ = max{kxkX , kykY },
for all x ∈ X and y ∈ Y . Also recall that a semi-inner product [6, Page 1] on a vector
space V is scalar-valued function ϕ : V × V → K such that for all x, y, z ∈ V and
α, β ∈ K, we have
(a) ϕ(αx + βy, z) = αϕ(x, z) + βϕ(y, z),
(b) ϕ(x, y) = ϕ(y, x),
(c) ϕ(x, x) ≥ 0.
It is easy to check that, if ϕ is a semi-inner product, then ϕ is an inner product if and
only if ϕ(x, x) = 0 implies x = 0. It is worth noting that the notion of a semi-inner
product is defined somewhat differently by different researchers. In the orthogonality
context, we point out the papers by Giles [7] and Lumer [11].
The main ingredients of our approach to the orthogonality problem are: (1) Bhatia
and Šemrl’s orthogonality of bounded linear operators on infinite-dimensional Hilbert
spaces (part (i) of Theorem 1.1 above), (2) projective tensor product techniques (see
Theorem 1.3 below), and (3) some standard Banach space techniques (like Banach-
Alaoglu theorem).
Before we proceed with the main content of the paper, let us shortly review the
existing literature on orthogonality. The notion of Birkhoff-James orthogonality is an
active research area. In fact, Bhatia and Šemrl’s paper [3] on orthogonality of operators
 ORTHOGONALITY IN BANACH SPACES VIA PROJECTIVE TENSOR PRODUCT 3

on Hilbert spaces has stimulated extensive research for the past two decades. For in-
stance, orthogonality of a pair of compact operators acting on a reflexive Banach space
to a normed linear space has been studied by Sain, Paul and Mal [16]. Their investiga-
tion involves finer geometric Banach space techniques. Also see [14] on orthogonality
of linear operators on finite dimensional Banach spaces, [15] on norm attainment and
orthogonality, and [12] on approximate orthogonality. More recent advances can be
found, for instance, in the quickly growing literature [1, 4, 5, 10, 18, 19] (also see the
references therein).
Given Banach spaces X and Y , we denote the Banach space of all bounded linear
operators from X to Y by B(X, Y ), and we let X1 denote the closed unit ball in X.
If Y = X, then we write B(X). A bilinear (sesquilinear) map B : X × Y → Z is said
to be bounded if there exists M > 0 such that
kB(x, y)k ≤ M (x ∈ X1 , y ∈ Y1 ).
We denote the Banach space of all bounded bilinear (sesquilinear) maps from X × Y
to Z by Bil(X × Y, Z) (Ses(X × Y, Z)). Here
kBk = sup{kB(x, y)k : x ∈ X1 , y ∈ Y1 },
for all B ∈ Bil(X × Y, Z) (B ∈ Ses(X × Y, Z)). As a tool for the proof of the main
result, we use the notion of projective tensor product. The projective tensor product
X⊗ ˆ π Y of Banach spaces X and Y is the completion of the algebraic tensor product
X ⊗ Y under the projective norm
{∑ n ∑
n }
kukπ = inf kxi kkyi k : u = xi ⊗ y i .
i=1 i=1

Our key point is the following result [13, Theorem 2.9] concerning representations of
bounded bilinear maps via projective tensor product spaces.

Theorem 1.3. Let X, Y and Z be Banach spaces, and let B ∈ Bil(X × Y, Z). Then
there exists a unique B̃ ∈ B(X ⊗
ˆ π Y, Z) such that

B̃(x ⊗ y) = B(x, y) (x ∈ X, y ∈ Y ).
Moreover, the correspondence B ←→ B̃ is an isometric isomorphism between Bil(X ×
Y, Z) and B(X ⊗ˆ π Y, Z).

We fix some more notation that we will use from now on. Given a vector space X
over K, we denote by X the complex conjugate vector space of X. That is, X = X
with the same additive group structures, but with the scalar multiplication ? defined
by
(1.1) α ? x = αx,
4 DHARA, RAKSHIT, SARKAR, AND SENSARMA

for all α ∈ K and x ∈ X. Clearly, if (X, k.k) is normed linear space over K, then there
is an anti-linear isometric isomorphism between (X, k.k) and (X, k.k). If H is a Hilbert
space, then H is identified with the dual (the space of continuous linear functionals) of
H by Riesz representation theorem.
Note that the space Bil(X⊕∞ X, K) is isometrically isomorphic to the space Ses(X⊕∞
X, K). The correspondence is given by

(1.2) ψ(x, y) = ϕ(x, y),

where ψ ∈ Bil(X ⊕∞ X, K), ϕ ∈ Ses(X ⊕∞ X, K), x ∈ X and y ∈ Y .

Note added in proof: After completion of the present paper, Professor Kallol Paul
kindly pointed out to us that our main result, Theorem 1.2 also follows (with a little
more work) from James [9, Theorem 2.1]. Note, however, that our result uses the
technique of projective tensor product and some basic geometric technique of Banach
spaces. Our approach connects two subtle notions, namely, orthogonality in Banach
spaces and projective tensor product of Banach spaces, which is different in the spirit
of the ongoing Birkhoff-James orthogonality program. As we will soon see, the semi-
inner product ϕ in Theorem 1.2 is naturally governed by projective tensor product of
Banach spaces. We believe that the inherited structure of projective tensor product of
Banach spaces in orthogonality will prove useful in future investigations.

2. Main Results
We begin with the proof of Theorem 1.2. But before we do so, let us recall the
classical Banach–Alaoglu theorem: Let X be a Banach space over K. Then the closed
unit ball (X ∗ )1 of the dual X ∗ is compact with respect to the weak∗ topology on X ∗ .
Proof of Theorem 1.2: Suppose x ⊥B y. Note that the Banach space X is isometrically
isomorphic to a closed subspace of C((X ∗ )1 ), where (X ∗ )1 endowed with the weak∗
topology is a compact set. Here the correspondence is given by the formula X 3 u 7→ û,
where
û(f ) = f (u) (f ∈ (X ∗ )1 ).
Next we note that the commutative Banach algebra C((X ∗ )1 ) is isometrically isomor-
phic to a closed subspace of B(L2 (µ)) for some σ-finite measure µ. Here the corre-
spondence is given by the formula C((X ∗ )1 ) 3 g 7→ Mg , where Mg : L2 (µ) → L2 (µ) is
the multiplication operator defined by

Mg (h) = gh (h ∈ L2 (µ)).

Using the above identifications, we have Mx̂ , Mŷ ∈ B(L2 (µ)) and

Mx̂ ⊥B Mŷ .
 ORTHOGONALITY IN BANACH SPACES VIA PROJECTIVE TENSOR PRODUCT 5

By the infinite dimensional part of Theorem 1.1, there exists a sequence {hn } of unit
vectors in L2 (µ) such that
kMx̂ (hn )kL2 (µ) → kMx̂ kB(L2 (µ)) ,
and
hMx̂ (hn ), Mŷ (hn )iL2 (µ) → 0 as n → ∞.
For each n ≥ 1, define ψn : X ⊕∞ X → K by
ψn (z, w) = hMẑ (hn ), Mŵ (hn )iL2 (µ) (z ∈ X, w ∈ X).
Now we prove that ψn is bilinear. Clearly, ψn is linear in its first variable. Suppose
α1 , α2 ∈ K and z, w1 , w2 ∈ X. Then
ψn (z, (α1 ? w1 + α2 ? w2 )) = hMẑ (hn ), Mα1 ⋆w\
1 +α2 ⋆w2
(hn )i
= hMẑ (hn ), Mα¯1 w\
1 +α¯2 w2
(hn )i
= hẑhn , (α¯1 ŵ1 + α¯2 ŵ2 )hn i
= α1 hẑhn , ŵ1 hn i + α2 hẑhn , ŵ2 hn i,
that is
ψn (z, (α1 ? w1 + α2 ? w2 )) = α1 ψn (z, w1 ) + α2 ψn (z, w2 ),
and hence ψn is bilinear for all n ≥ 1. To prove that ψn is bounded, suppose (z, w) ∈
X ⊕∞ X and k(z, w)k∞ ≤ 1. Then
|ψn (z, w)| ≤ kMẑ kB(L2 (µ)) kMŵ kB(L2 (µ)) khn k2L2 (µ)
= kẑkC((X ∗ )1 ) kŵkC((X ∗ )1 )
= kzkX kwkX ,

which implies that kψn k ≤ 1 for all n. Consequently {ψn }n≥1 ⊆ Bil(X ⊕∞ X, K). Note,
by Theorem 1.3, that
Bil(X ⊕∞ X, K) ' B(X ⊗ ˆ π X, K).
Here the correspondence is given by Bil(X ⊕∞ X, K) 3 ψ 7→ ψ̃, where
ψ̃(z ⊗ w) = ψ(z, w) (z, w ∈ X).
Since kψn k ≤ 1, we have kψ˜n k ≤ 1 for each n. By applying the Banach-Alaoglu
theorem, one can find a subsequence {ψ̃nk } of {ψ˜n } and a map ψ̃ ∈ B(X ⊗
ˆ π X, K) with
kψ̃k ≤ 1 such that
w∗
ψ̃nk → ψ̃.
This yields ψnk → ψ in the pointwise topology, that is
ψnk (z, w) → ψ(z, w) (z, w ∈ X).
6 DHARA, RAKSHIT, SARKAR, AND SENSARMA

By isometry, kψk ≤ 1. Observe that

ψnk (x, x) = hMx̂ (hnk ), Mx̂ (hnk )iL2 (µ) = kMx̂ (hnk )k2L2 (µ) .
But since
kMx̂ (hnk )k2L2 (µ) → kMx̂ k2B(L2 (µ)) = kx̂k2 ,
it follows that
ψnk (x, x) → kxk2 ,
for all x ∈ X. On the other hand, ψnk (x, x) → ψ(x, x). Hence
ψ(x, x) = kxk2 (x ∈ X).
Since
ψnk (x, y) = hMx̂ (hnk ), Mx̂ (hnk )iL2 (µ) → 0,
we have ψ(x, y) = 0. Note that if kψk < 1, then ψ(x, x) < kxk2 . Hence kψk = 1. Now
observe that the identification in (1.2) provides a sesquilinear map ϕ ∈ Ses(X ⊕∞ X, K),
with
ϕ(x, y) = ψ(x, y) (x, y ∈ X).
This gives ϕ(z, z) ≥ 0 for all z ∈ X, and ϕ(z, w) = ϕ(w, z) for all z, w ∈ X. Hence ϕ
is a semi-inner product on X such that kϕk = 1, ϕ(x, x) = kxk2 and ϕ(x, y) = 0.
Conversely, suppose that such a ϕ exists. Then for all λ ∈ K,
kxk2 = |ϕ(x, x + λy)| ≤ kϕkkxkkx + λyk,
which implies that kxk ≤ kx + λyk, that is, x ⊥B y. This completes the proof of the
theorem.

Remark 2.1. It is worth pointing out that besides Theorem 1.1 and Theorem 1.3, the
inclusion
X ,→ C((X ∗ )1 ) ,→ B(L2 (µ)),
also plays an important role in our proof.

Now we turn to orthogonality of elements in C ∗ -algebras. The following result follows

from Theorem 1.2.

Corollary 2.2. Let A be a C ∗ -algebra and a, b ∈ A and a 6= 0. Then a ⊥B b if and

only if there exists a bilinear map ψ : A ⊕∞ A → C such that kψk = 1, ψ(a, a∗ ) = kak2
and ψ(a, b∗ ) = 0.

Proof. Suppose a ⊥B b. Then by Theorem 1.2, there exists a semi-inner product

ϕ : A ⊕∞ A → C such that kϕk = 1, ϕ(a, a) = kak2 and ϕ(a, b) = 0. Define
ψ : A ⊕∞ A → C by
ψ(a, b) = ϕ(a, b∗ ).
 ORTHOGONALITY IN BANACH SPACES VIA PROJECTIVE TENSOR PRODUCT 7

Then ψ is the required map. To prove the converse, suppose such a ψ exists. Then
kak2 = |ψ(a, (a + λb)∗ )|
≤ kψkkakk(a + λb)∗ k
= kakka + λbk,
that is, ka + λbk ≥ kak for all λ ∈ C. 
We refer the reader to [4, Proposition 4.1] for a more natural version of orthogonality
in the setting of C ∗ -algebras.
It is worth pointing out that Theorem 1.2 is also applicable to finite dimensional
Banach spaces. Clearly, Theorem 1.2 and Corollary 2.2 are analogous to Bhatia and
Šemrl classifications (see part (ii) of Theorem 1.1) of orthogonality of matrices on finite
dimensional Hilbert spaces. On the other hand, our results uses the Bhatia and Šemrl
classifications of orthogonality in the setting of infinite dimensional Hilbert spaces (see
part (i) of Theorem 1.1). In addition, it is not completely clear if our results recovers
the Bhatia and Šemrl classifications of orthogonality in the setting of finite matrices.
All in all, on one hand our results are valid for general Banach spaces and rather
abstract, and on the other hand our approach is intimately related to the delicate
structure of projective tensor product of Banach spaces (see Chapter 2 in [13]). We
also believe that our approach to orthogonality via projective tensor product is of
independent interest and may have other applications.
Finally, we remark that our main result Theorem 1.2 is also valid in full generality
in the setting of normed linear spaces. The proof works verbatim. In this case, [17,
page 443, Proposition 43.12. (b)] is the normed linear space counterpart to our key
tool Theorem 1.3.
Acknowledgement: The research of the second named author is supported by the
NBHM postdoctoral fellowship, File No: 0204/12/2019/ R&D-II/10895. The research
of the third named author is supported in part by NBHM grant NBHM/R.P.64/2014,
and the Mathematical Research Impact Centric Support (MATRICS) grant, File No:
MTR/2017/000522 and Core Research Grant, File No: CRG/2019/000908, by the
Science and Engineering Research Board (SERB), Department of Science & Technology
(DST), Government of India. The research of the fourth named author is supported
by the NBHM postdoctoral fellowship, File No: 0204/3/2020/R&D-II/2445.

References
[1] L. Arambašić and R. Rajić, The Birkhoff-James orthogonality in Hilbert C ∗ -modules, Linear
Algebra Appl. 437 (2012) 1913–1929.
[2] G. Birkhoff, Orthogonality in linear metric spaces, Duke Math. J. 1 (1935), 169–172.
[3] R. Bhatia and P. Šemrl, Orthogonality of matrices and some distance problems, Linear Algebra
Appl. 287 (1999), 77–85.
8 DHARA, RAKSHIT, SARKAR, AND SENSARMA

[4] T. Bhattacharyya and P. Grover, Characterization of Birkhoff-James orthogonality, J. Math.

Anal. Appl. 407 (2013), 350–358.
[5] J. Chmieliński, T. Stypula and P. Wójcik, Approximate orthogonality in normed spaces and its
applications, Linear Algebra Appl. 531 (2017), 305–317.
[6] J. Conway, A course in functional analysis, Second edition. Graduate Texts in Mathematics, 96.
Springer-Verlag, New York, 1990.
[7] J.R. Giles, Classes of semi-inner-product spaces, Trans. Amer. Math. Soc. 129 (1967) 436–446.
[8] R. C. James, Orthogonality in normed linear spaces, Duke Math. J. 12 (1945), 291–302.
[9] R. C. James, Orthogonality and linear functionals in normed linear spaces, Trans. Amer. Math.
Soc. 61 (1947), 265–292.
[10] N. Komuro, K. Saito and R. Tanaka, On symmetry of Birkhoff orthogonality in the positive cones
of C ∗ -algebras with applications, J. Math. Anal. Appl. 474 (2019), 1488–1497.
[11] G. Lumer, Semi-inner-product spaces, Trans. Amer. Math. Soc. 100 (1961) 29–43.
[12] A. Mal, K. Paul, T.S.S.R.K. Rao and D. Sain, Approximate Birkhoff-James orthogonality and
smoothness in the space of bounded linear operators, Monatsh. Math. 190 (2019), 549–558.
[13] R. Ryan, Introduction to tensor products of Banach spaces, Springer Monographs in Mathematics
(2002).
[14] D. Sain, Birkhoff–James orthogonality of linear operators on finite dimensional Banach spaces,
J. Math. Anal. Appl. 447 (2017), 860–866.
[15] D. Sain and K. Paul, Operator norm attainment and inner product spaces, Linear Algebra Appl.
439 (2013), 2448–2452.
[16] D. Sain, K. Paul and A. Mal, A complete characterization of Birkhoff-James orthogonality in
infinite dimensional normed space, J. Operator Theory. 80 (2018), 399–413.
[17] F. Trèves, Topological vector spaces, distributions and kernels, Unabridged republication of the
1967 original. Dover Publications, Inc., Mineola, NY, 2006.
[18] A. Turnšek, A remark on orthogonality and symmetry of operators in B(H), Linear Algebra Appl.
535 (2017), 141–150.
[19] P. Wójcik, The Birkhoff orthogonality in pre-Hilbert C ∗ -modules, Oper. Matrices 10 (3) (2016)
713–729.

Kousik Dhara, Indian Statistical Institute, Statistics and Mathematics Unit, 8th
Mile, Mysore Road, Bangalore, 560 059, India.
Email address: kousik.dhara1@gmail.com

Narayan Rakshit, Indian Statistical Institute, Statistics and Mathematics Unit, 8th
Mile, Mysore Road, Bangalore, 560 059, India.
Email address: narayan753@gmail.com

Jaydeb Sarkar, Indian Statistical Institute, Statistics and Mathematics Unit, 8th
Mile, Mysore Road, Bangalore, 560 059, India.
Email address: jaydeb@gmail.com, jay@isibang.ac.in

Aryaman Sensarma, Indian Statistical Institute, Statistics and Mathematics Unit,

8th Mile, Mysore Road, Bangalore, 560 059, India.
Email address: aryamansensarma@gmail.com