Functional Analysis by R.
Vittal Rao
Lecture 8: Convergence (contd.) - June 15, 2012
For any > 0, ∃ positive integer N such that n ≥ N ⇒ ρp (x(n) , x) <
i.e.,
p 1/p
N
8.1 Examples of Metric Spaces
X
n ≥ N ⇒
x(n) − x j <
(8.1)
j
j=1
Example 8.1 (n)
(R, |·|) is a complete metric space. Where |·| is a usual metric. Take any fixed k, 1 ≤ k ≤ N, then {xk }n=1,2,3... is a sequence of a real
numbers. Now
N
Example 8.2 (n) p X (n) p
Let 1 ≤ p < ∞, N is a finite positive integer and consider the xk − xk ≤ x j − x j
p j=1
metric space lN (R) defined as follows,
p p
If x = (x1 , x2 , ...., xN ) ∈ lN (R) and y = (y1 , y2 , ...., yN ) ∈ lN (R), p 1/p
(n) X N
then we define metric ρp as ⇒ xk − xk ≤ x(n) − x j
j
j=1
1/p
XN p
< , ∀n ≥ N By (8.1)
ρp (x, y) =
x j − y j
(n)
j=1 Hence the sequence of real numbers {xk }n=1,2,3... converges to xk
for every k, 1 ≤ k ≤ N i.e., we have coordinatewise convergence.
p
Convergence in lN (R) p
CONCLUSION:A x(n) −→ x in lN (R) ⇒ x(n) −→ x coordi-
What does convergence really mean in lN (R) space?
p natewise.
p p
Let the sequence {x(n) } ∈ lN (R) and x ∈ lN (R), what does {x(n) }
CONCLUSION:B It is easy to see that coordinatewise con-
converge to x mean? p
vergence ⇒ x(n) −→ x in lN (R).
We denote {x(n) } converges to x symbolically as x(n) −→ x.
Notation Example:
CONCLUSION:C Hence from conclusion A and B, conver-
p
(n) (n) (n)
x(n) = (x1 , x2 , .., xk , .., xN )
(n) gence in lN (R) for 1 ≤ p < ∞ is equivalent to coordinatewise
convergence.
p
Suppose x(n) converges to x = (x1 , x2 , .., xk , .., xN ) in lN (R) space. This
means,
1
8.1 Examples of Metric Spaces
Convergence of a sequence of vectors in RN in any one of the Convergence in lp (R)
metric ρp ⇒ convergence in all ρp metrics.
Again, what is meant by convergence?
Exercise: Show that this is also true if we take p = ∞. Let {x(n) } ∈ lp (R), x ∈ lp (R), we say x(n) converges to x if ∀ > 0,
∃ N such that n ≥ N
p
Analogously, we see that {x(n) } is a Cauchy sequence in lN (R) ⇐⇒
p 1/p
∞
it is coordinatewise Cauchy sequence.
X
⇒
x(n) − x j <
j
p j=1
Is lN (R) COMPLETE?
p
For each k, we have
To establish lN (R) is complete, we have to show that for any
p p
Cauchy sequence {x(n) } ∈ lN (R) ∃ a x ∈ lN (R) such that x(n) −→ x in
p 1/p
p ∞
lN (R). (n) p 1/p X (n)
p xk − xk ≤
x j − x j
Let x(n) Cauchy sequence in lN (R)
j=1
(n)
⇒ xk is Cauchy sequence of real numbers, for every k, 1 ≤ k ≤ N. < , ∀n ≥ N
(n)
⇒ ∃ xk ∈ R such that xk −→ xk in R, for every k, 1 ≤ k ≤ N.
(n)
Let x = (x1 , x2 , ..., xk , ..., xN ), clearly x ∈ RN and x(n) −→ x coordi- ⇒ xk −→ xk in R.
natewise. ∴ Convergence in lp (R) ⇒ coordinatewise convergence.
p
⇒ x(n) −→ x in lN (R). However, the converse is NOT true i.e., we may have coordinate-
p
∴ Every Cauchy sequence in lN (R) converges. wise convergence without having lp (R) convergence.
p
∴ lN (R) is complete.
Example 8.4 Converges coordinatewise but NOT in lp (R)
x(1) = (1, 0, 0, 0, .......)
x(2) = (1, 1, 0, 0, ......)
x(3) = (1, 1, 1, 0, ......)
Example 8.3 ...................................
Let 1 ≤ p < ∞, metric space lp (R) defined as ...................................
x(n) = (1, 1, .. 1, 0, 0,..)
∞
p
X
lp (R) = = , , ..., , .....) <
x (x x x : x ∞ x(n) has n number of ones.
1 2 k j
x(n) converges to x = (1, 1, 1, .....) (all one’s) coordinatewise.
j=1
with metric But x(n) does not converge to x = (1, 1, 1, .....) in lp (R), because
∞
1/p x < lp (R).
X p
ρp (x, y) = x j − y j
j=1
CONCLUSION: In lp (R), we have
Convergence in lp (R) ⇒ coordinatewise convergence.
2
8.1 Examples of Metric Spaces
But coordinatewise convergence ; convergence in lp (R). Step 3: To prove (8.3) it is enough to show that
Analogously,
Cauchy sequence in lp (R) ⇒ coordinatewise Cauchy sequence. p 1/p
X N
But coordinatewise Cauchy sequence ; Cauchy sequence in lp (R). x(n) − x j < /2, ∀N
(8.5)
j
j=1
Is lp (R) COMPLETE?
so that
p 1/p
N
(n)
We have to show that, for every Cauchy sequence {x } in
X
lim
x(n) − x j ≤ /2 < , ∀N
lp (R) ∃ x ∈ lp (R) such that x(n) −→ x in lp (R). N→∞
j=1
j
The Idea: Following is an overview of the proof. Given {x(n) } is Cauchy sequence
Step 1: Cauchy sequence in lp (R) ⇒ coordinatewise Cauchy ⇒ ∀ > 0 ∃ N such that n, m ≥ N ⇒ ρp (x(n) , x(m) ) < .
sequence. Step 2: Take = 1
(n)
⇒ For each k, xk is a Cauchy sequence of real numbers, ∃ N1 such that n, m ≥ N1 ⇒ ρp (x(n) , x(m) ) < 1.
(n)
∴ ∃ xk ∈ R such that xk −→ xk in R. ∴ ∃ N1 such that n ≥ N1 ⇒ ρp (x(n) , x(N1 ) ) < 1.
Let x = (x1 , x2 , ..., xk , .....).
p 1/p
Step 2: Make sure that this x ∈ lp (R) X ∞
(n) (N1 )
⇒ x j − x j < 1 (8.6)
i.e., make sure that
j=1
∞
p
X
xj < ∞ (8.2) To get (8.4), consider
1/p
j=1 N
p
X
x j
j=1
Step 3: Make sure that x(n) −→ x in lp (R)
i.e., make sure that ∀ > 0, ∃ N such that n ≥ N p 1/p
XN
=
x j − x(N1 ) + x(N1 )
j j
⇒ ρp (x(n) , x) < (8.3) j=1
Using Minkowski inequality
p 1/p X p 1/p
XN N
Proofs: ≤
x j − x(N1 ) +
x(N1 )
(8.7)
j j
Step 1 is already over.
j=1 j=1
Step 2: To prove (8.2), it is enough to show that ∃ a real number
For the first term in (8.7), consider
K > 0 such that
p 1/p
N N
p
X X
x j ≤ K, ∀N (8.4) x(n) − x(N1 ) < 1, By (8.6)
j j
j=1 j=1
3
8.1 Examples of Metric Spaces
Let n −→ ∞
p 1/p
X N
(N1 )
⇒ x j − x j ≤ 1
j=1
Consider the second term in (8.7),
1/p
N
X (N1 ) p
x j
j=1
1/p
∞
X (N1 ) p
≤ x j
j=1
≤ KN1 ∵ x(N1 ) ∈ lp (R)
Hence put together Equation (8.7) becomes
≤ 1 + KN1
= K
Step 3: A similar idea is used to get (8.5).
p p
lN (R), lN (C), lp (R), lp (C) are all COMPLETE METRIC SPACES.
