�p IS COMPLETE

Let 1 ≤ p ≤ ∞, and recall the definition of the metric space �p :

� ∞
�
�
For 1 ≤ p < ∞, �p = sequences a = (an )∞
n=1 in R such that |an |p < ∞ ;
n=1
∞
whereas � consists of all those sequences a = (an )∞
such that supn∈N |an | < ∞. We
n=1
p
defined the p-norm as the function � · �p : � → [0, ∞), given by
�∞ �1/p
�
�a�p = |an |p , for 1 ≤ p < ∞,
n=1

and �a�∞ = supn∈N |an |. In class, we showed that the function dp : �p × �p → [0, ∞) given
by dp (a, b) = �a − b�p is actually a metric. We now proceed to show that (�p , dp ) is a
complete metric space for 1 ≤ p ≤ ∞. For convenience, we will work with the case p < ∞,
as the case p = ∞ requires slightly different language (although the same ideas apply).
Suppose that a1 , a2 , a3 , . . . is a Cauchy sequence in �p . Note, each term ak in the se­
quence is a point in �p , and so is itself a sequence:
ak = (ak1 , ak2 , ak3 , . . .).
Now, to say that (ak )∞ p
k=1 is a Cauchy sequence in � is precisely to say that

∀� > 0 ∃K ∈ N s.t. ∀k, m ≥ K, �ak − am �p < �.

That is, for given � > 0 and sufficiently large k, m, we have
∞
�
|akn − anm |p = �ak − am �pp < �p .
n=1

Now, the above series has all non-negative terms, and hence is an upper bound for any
fixed term in the series. That is to say, for fixed n0 ∈ N,
∞
�
|akn0 − am
n0 | ≤ |akn − anm |p < �p ,
n=1

and so we see that the sequence (akn0 )∞

k=1 is a Cauchy sequence in R. But we know that R
is a complete metric space, and thus there is a limit an0 ∈ R to this sequence. This holds
for each n0 ∈ N. The following diagram illustrates what’s going on.
a1 = a11 a12 a13 a14 ···
a2 = a21 a22 a23 a24 ···
a3 = a31 a32 a33 a34 ···
a4 = a41 a42 a43 a44 ···
.. .. .. .. ...
. . . .
↓ ↓ ↓ ↓
a1 a2 a3 a4 · · ·
So, we have shown that, in this �p -Cauchy sequence of horizontal sequences, each vertical
sequence actually converges. Hence, there is a sequence a = (a1 , a2 , a3 , a4 , . . .) to which
“ak converges” in a vague sense. The sense is the “point-wise convergence” along vertical
1
2

lines in the above diagram. To be more precise, recall that a sequence a is a function
a : N → R, where we customarily write a(n) = an . What we have shown is that, if
(a1 , a2 , a3 , . . .) is a Cauchy sequence of such �p functions, then there is a function a : N → R
such that ak converges to a point-wise; i.e. ak (n) → a(n) for each n ∈ N.
Now, our goal is to find a point b ∈ �p such that ak → b as k → ∞ in the sense of �p ;
that is, such that �ak − b�p → 0 as k → ∞. The putative choice for this b is the sequence a
given above. In order to show that one works, we need to show first that it is actually an
�p sequence, and second that ak converges to a in �p sense, not just point-wise.
To do this, it is convenient to first pass to a family of subsequences of the (akn ), as fol­
k1
lows. Since (ak1 )∞ 1
k=1 converges to a1 , we can choose k1 so that for k ≥ k1 , |a1 − a1 | < 2 .
Having done so, and knowing that ak2 → a2 , we can choose a larger k2 so that for k ≥ k2 ,
we have |ak1 − a1 | < 14 and |ak2 − a2 | < 14 . Continuing this way iteratively, we can find an
increasing sequence of integers k1 < k2 < k3 < · · · such that
for each j ∈ N, |akn − an | < 2−j for n = 1, 2, . . . , j and k ≥ kj . (1)
k
In particular, we have |anj − an | < 2−j for j ≥ n. That gives us the following.
Lemma 1. The sequence a = (an )∞ k ∞ p
n=1 of point-wise limits of (a )k=1 is in � .

Proof. Fix N ∈ N, and recall that the finite-dimensional versions of the �p -norms,
� N �1/p
�
�(a1 , . . . , aN )�p = |an |p
n=1

also satisfy the triangle inequality (i.e. dp (x, y) = �x − y�p is a metric on RN ). Hence, we
can estimate the initial-segment of N terms of a as follows:
an = (an − aknN ) + aknN ,
and so
� N
�1/p � N
�1/p � N
�1/p
� � �
|an |p ≤ |an − aknN |p + |aknN |p . (2)
n=1 n=1 n=1
p
Now, the last term in Equation 2 is bounded by the actual � -norm of the whole sequence
akN ; that is, we can tack on the infinitely many more terms,
� N �1/p � ∞ �1/p
� �
|aknN |p ≤ |aknN |p = �akN �p .
n=1 n=1

Recall that (ak )∞ p

k=1 is a Cauchy sequence in the metric space � . We have proved that any
Cauchy sequence in a metric space is bounded. Thus, there is a constant R independent of
N such that �akN �p ≤ R. Combining this with Equation 1, we can therefore estimate the
right-hand-side of Equation 2 by
� N �1/p � N �1/p
� � �1/p
|an |p (2−N )p + R = N 2−N p
�
≤ + R.
n=1 n=1
 3
�1/p
Finally, the term N 2−N p = N 1/p 2−N converges to 0 as N → ∞ (remember your
�
calculus!), and hence this sequence is also bounded by some constant S. In total, then,
we have
� N �1/p
�
|an |p ≤ R + S for all N ∈ N.
n=1
�N
In other words, n=1 |an |p ≤ (R + S)p . The constant on the right does not depend on N ;
it is an upper bound for the increasing sequence of partial sums of the series ∞ p
�
n=1 |a n| =
�a�pp . Thus, we have �a�p ≤ R + S, and so a ∈ �p . �

So, we have shown that the putative limit a (the point-wise limit of the sequence (ak )∞k=1
of points in �p ) is actually an element of the metric space �p . But we have yet to show that
it is the limit of the sequence (ak ) in �p . That somewhat involved proof now follows.
Proposition 2. Let (ak )∞ p
k=1 be a Cauchy sequence in � , and let a be its point-wise limit (which is
p k
in � , by Lemma 1). Then �a − a�p → 0 as k → ∞.
Proof. Let � > 0. Lemma 1 shows that a ∈ �p , which means that ∞ p
�
n=1 |an | < ∞. Hence,
by the Cauchy criterion, there is an N1 ∈ N so that
∞
�
|an |p < �p .
n=N1

In addition, we know that (ak )∞ p

k=1 is � -Cauchy, so there is N2 so that, whenever k, m ≥ N2 ,
�ak − am �p < �. Letting N = max{N1 , N2 }, we therefore have
∞
�
|an |p < �p and �aN − ak �p < � ∀k ≥ N. (3)
n=N

Now, the sequence aN is in �p , and so we can apply the Cauchy criterion again: select N �
large enough so that
�∞
|aN p p
n| < � . (4)
n=N �
�
Note, we can always increase N and still maintain this estimate, so we are free to chose
N� ≥ N.
We now use the constant N � we defined above in the bounds we will need later. Since
akn → an for each fixed n, we can choose K1 so that |ak1 − a1 | < �p /N � for k ≥ K1 . Likewise,
we can choose K2 so that |ak2 − a2 | < �p /N � for k ≥ K2 . Continuing this way for N � steps,
we can take K = max{K1 , K2 , . . . , KN � } and then we have
�p
, for k ≥ K and n ≤ N � .
|akn − an | < (5)
N�
For good measure, we will also (increasing it if necessary) make sure that K ≥ N � . Now,
for any k ≥ K, break up b = ak − a as follows:
�
(bn )∞ N −1 ∞
n=1 = (bn )n=1 + (bn )n=N � .
4

(To be a little more pedantic, we are expressing bn = xn + yn where xn = bn when n < N �

and = 0 when n ≥ N � , and yn = 0 when n < N � and = bn when n ≥ N � .) The triangle
inequality for the p-norm then gives
�N � −1 �1/p � ∞ �1/p
� �
�ak − a�p ≤ |akn − an |p + |akn − an |p . (6)
n=1 n=N �
Equation 5 shows that, for k ≥ K, the first term here is
�N � −1 �1/p �N � −1 �1/p � �1/p
�
k p
� �p N� − 1
|an − an | ≤ �
= �
� < �.
n=1 n=1
N N
For the second term in Equation 6, we use the triangle inequality for the �p -norm restricted
to the range n ≥ N � to get
� ∞ �1/p � ∞ �1/p � ∞ �1/p
� � �
|akn − an |p ≤ |akn |p + |an |p .
n=N � n=N � n=N �
Since N � ≥ N , Equation 3 shows that the second term here is < �. So, summing up the
last two estimates, we have
� ∞ �1/p
�
�ak − a�p ≤ 2� + |akn |p , (7)
n=N �
whenever k ≥ K. So we need only show this final term is small. Here we make one more
decomposition: akn = akn − aN N
n + an , and so once again applying the triangle inequality,
� ∞ �1/p � ∞ �1/p � ∞ �1/p
� � �
|akn |p ≤ |ank − anN |p + |anN |p .
n=N � n=N � n=N �
The first of these terms is a sum of non-negative terms over n ≥ N � , and so it is bounded
above by the sum over n ≥ 1 which is equal to �ak − aN �p , which is < � by Equation 3
(since k ≥ K ≥ N � ≥ N ). And the second term is also < �, by Equation 4. Whence, the
last term in Equation 7 is also < 2�, and so we have shown that
∀� > 0, ∃K ∈ N such that ∀k ≥ K �ak − a�p < 4�.
Of course, we should have been more clever and chosen all our constants in terms of �/4
to get a clean � in the end, but such tidying is not really necessary; 4� is also arbitrarily
small, and so we have shown that (ak )∞ p
k=1 does converge to a in � . This concludes the
p
proof that � is complete. Whew! �
Let us conclude by remarking that a very similar (though somewhat simpler) proof
works for p = ∞; the details are left to the reader.
MIT OpenCourseWare
http://ocw.mit.edu

18.100B Analysis I
Fall 2010

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.