Concerning L2 (I)

1. L2 (I) is the class of complex-valued functions f defined on the interval I such that
Z
|f (x)|2 dx < ∞. (1)
I

(The integral appearing here and throughout the discussion of L2 (I) is the Lebesgue
integral, as opposed to the Riemann. For any reasonable function, the two integrals
coincide, so we needn’t worry about the distinction.)
a. L2 (I) is a linear, or vector space over the field C of complex scalars. For practical
purposes, this means that for any functions f and g in L2 (I), and any complex numbers
α and β, the linear combination αf + βg lies in L2 (I).
b. The L2 (I) inner product is
Z
hf , gi = f (x)ḡ(x) dx. (2)
I

c. The L2 (I) norm is

Z  12
p 2
kf k2 = hf , f i = |f (x)| dx . (3)
I

The inner product and norm on L2 (I) were clearly inspired by the standard inner
product and norm on Cn .

2. The Cauchy-Schwarz Inequality asserts that for f and g in L2 (I),

| hf , gi | ≤ kf k2 kgk2 . (4)

3. A sequence {fk } of functions in L2 (I) is convergent if there is an f ∈ L2 (I) such that

kfk − f k2 → 0 as k → ∞. (5)

We say that {fk } converges to f and write

lim fk = f,
k→∞

or
fk → f as k → ∞.

4. A sequence {fk } of functions in L2 (I) is Cauchy convergent if

kfm − fn k2 → 0 as m, n → ∞. (6)
 5. If a sequence {fk } in L2 (I) is Cauchy convergent, it is also convergent. Thus L2 (I) is
complete with respect to the norm (3).

6. L2 (I) is a Hibert space, that is, a complete inner product space.

7. A set {fk } of vectors in L2 (I) is orthogonal if

hfi , fj i = 0 for i 6= j.

8. The Kronecker delta is (

0 for i 6= j,
δij =
1 for i = j.

9. A set {ϕk } in L2 (I) is orthonormal if

hϕi , ϕj i = δij .

Thus {ϕk } is orthonormal if it is orthogonal and the ϕk are unit vectors.

10. Note that if {fk } is an orthogonal set of nonzero vectors, then {fk /kfk k2 } is orthonor-
mal.

11. The set {ϕk } is an orthonormal basis of L2 (I) if

a. it is orthonormal and
b. every vector f ∈ L2 (I) can be represented as a linear combination of the ϕk :
X
f= ck ϕk , (7)
k

for scalars ck .

12. If the sum in (7) is infinite, then equality is understood to hold in the sense of norm
convergence:
XN
kf − ck ϕk k2 → 0 as N → ∞.
k=1

13. For g in L2 (I), and O = {ϕk } an orthonormal basis, there are scalars ck such that
X
g= ck ϕk .
k
 By taking inner products, we see that cm = hg , em i . Thus,
X
g= hg , ϕk i ϕk . (8)
k

The sum (8) is called the Fourier series of g with respect to O. The numbers

ĝk = hg , ϕk i

are the Fourier coefficients.

14. Let O = {ϕk } be an orthonormal basis of L2 (I). Then for functions f and g in L2 (I),
X
hf , gi = hf , ϕk i hg , ϕk i. (9)
k

If f = g, this becomes X
kgk22 = | hg , ϕk i |2 . (10)
k

Formulae (9) and (10) are versions of Parseval’s identity; (10) is a generalization of the
Pythagorean theorem.

15. If {ϕk } is an orthonormal set (not necessarily a basis) in L2 (I), then (10) becomes
X
kgk22 ≥ | hg , ϕk i |2 . (11)
k

This is Bessel’s inequality. It can be shown that equality holds in (11) for all g ∈ L2 (I)
if and only if {ϕk } is a basis of L2 (I).