Spring 2021 ETH Zürich

Measure and Integration Maran Mohanarangan

Exercise Class 11
Modes of Convergence
If one has a sequence of complex numbers (xn )n∈N , it is unambiguous what it means
for that sequence to converge to a limit x ∈ R. More generally, if we have a sequence
(vn )n∈N of d-dimensional vectors in a real vector space Rn , it is clear what it means for
a sequence to converge to a limit. We usually consider convergence with respect to the
Euclidean norm, but for the purposes of convergence, these norms are all equivalent1 .
If, however, one has a sequence of real-valued functions (fn )n∈N on a common domain
Ω and a perceived limit f , there can now be many different ways how fn may or may
not converge to f . Since the function spaces we consider are infinite dimensional, the
functions fn have an infinite number of degrees of freedom, and this allows them to
approach f in any number of inequivalent ways.
We now introduce different convergence concepts for sequences of measurable func-
tions and then compare them to each other.
Definition 1: Modes of Convergence

Let (fn )n∈N and f : Ω → R be measurable functions. Then we say that (fn )
converges to f

– µ-almost everywhere (µ-a.e.) if there is a measurable set N with µ(N ) = 0

such that
lim fn (x) = f (x) ∀x ∈ N c .
n→∞

We write fn → f µ-a.e.

– in measure µ if for all ε > 0

 
lim µ({x ∈ Ω : f (x) − fn (x) > ε}) = 0.
n→∞

µ
We write fn −
→ f.

– in L1 (Ω, µ) if
Z
lim kfn − f kL1 (Ω,µ) := lim |fn − f | dµ = 0.
n→∞ n→∞ Ω

L1
We write fn −→ f .

If there is no confusion as to what the space (Ω, µ) is, one often just speaks of L1
convergence (instead of L1 (Ω, µ) convergence).
Remark. The L1 mode of convergence is a special case of the Lp mode of convergence,
which is just convergence with respect to the Lp norm.
One particular advantage of L1 convergence is that, in the case when the fn are
µ-summable, it implies convergence of the integrals
Z Z
fn dµ → f dµ.
Ω Ω
1
In fact, on a finite dimensional vector space, any two norms are equivalent.

Questions and Remarks to: [email protected] 1

Spring 2021 ETH Zürich
Measure and Integration Maran Mohanarangan

This follows directly by the triangle inequality, i.e.

Z Z  Z
 
 fn dµ − f dµ ≤ |fn − f | dµ.

Ω Ω Ω

Proposition 2: Simple Implications

Convergence in L1 (Ω, µ) both implies convergence in measure µ. Moreover, if

µ(Ω) < ∞, then convergence µ-a.e. implies convergence in measure µ too.

Proof. By replacing fn with fn − f , we can assume that f ≡ 0 without loss of generality.

1. Recall Chebyshev’s inequality which states that for every µ-summable f : Ω → R,
we have Z
  1
µ(x ∈ Ω : f (x) > a) ≤ |f | dµ ∀a > 0.
a Ω
It immediately follows that for all ε > 0
Z
1 1
µ({x ∈ Ω : |fn | > ε}) ≤ |fn | dµ = kfn kL1 .
ε Ω ε

So L1 (Ω, µ)-convergence implies convergence in measure.

2. From Egorov’s theorem, it follows that for every δ > 0, there exists Fδ ⊂ Ω µ-
measurable with µ(Ω \ Fδ ) < δ such that (fn )n converges uniformly to f on Fδ . In
other words,
 
∀ε > 0 ∃N ≥ 0 : sup fn (x) − f (x) < ε ∀n ≥ N.
x∈Fδ

Therefore, for n ≥ N
 
{x ∈ Ω : fn (x) − f (x) > ε} ⊂ Ω \ Fδ .

Hence  
µ({x ∈ Ω : fn (x) − f (x) > ε}) ≤ µ(Ω \ Fδ ) < δ.
Since δ > 0 was arbitrary, we can conclude.
Alternatively, the latter can be proven by applying the dominated convergence theorem
to the integral of 1{|fn −f |>ε} , which is dominated by 1 on a finite measure space.

All other implications between different convergence concepts are not true in general.
Example 1. As the sequence fn = 1[n,n+1] shows, the finiteness assumption in the
second implication is necessary. The sequence fn converges to 0 pointwise (and thus
µ-a.e.), however it does not converge in measure.
Example 2. Let Ω = [0, 1] and λ be the Lebesgue measure.
– (Convergence µ-a.e. ; L1 convergence)
The sequence fn := n1(0, 1 ) , n ∈ N, converges to 0 pointwise, hence also λ-a.e.
n
and hence in measure (since we are dealing with a finite measure space). But
fn dλ = 1 and thus (fn ) does not converge to 0 in L1 ([0, 1], λ).
R

Questions and Remarks to: [email protected] 2

Spring 2021 ETH Zürich
Measure and Integration Maran Mohanarangan

– (L1 convergence ; convergence µ-a.e.)

For n ∈ N and k = 1, . . . , 2n , define fnk := 1[(k−1)2−n ,k2−n ] . We then renumber
the
R double sequence f11 , f12 , f21 , . . . to a single sequence (gm )m∈N . Then we have
fnk dλ = 2−n and hence gm → 0 in L1 ([0, 1], λ) for m → ∞. Thus, we also have
gm → 0 in measure. However, lim supm→∞ gm = 1 and lim inf m→∞ gm = 0 show
that (gm ) does not converge to 0 λ-a.e. Intuitively, this is a sequence of indicator
functions of intervals of decreasing length, marching across the unit interval over
and over again. This sequence is also known as the typewriter sequence.

Example 2 shows, in particular, that convergence in measure is a strictly weaker

notion than the other two, as convergence in measure implies neither of them.
Convergence µ-a.e. and convergence in L1 (Ω, µ) do not seem to be related. However,
if one imposes additional assumptions (such as µ(Ω) < ∞) to shut down these “escape
to infinity” scenarios, then one can obtain some additional implications between the
different concepts.
The dominated convergence theorem of Lebesgue states that µ-a.e. convergence to-
gether with the existence of a µ-summable bound for a sequence of measurable functions
imply convergence in L1 (Ω, µ). These conditions are only sufficient, but not necessary.
Thus it is of interest to look for an even sharper result.

Example 3. Let Ω = [0, 1] and consider the Lebesgue measure λ. We define the func-
tions
n
fn := 1 1 ∀n ≥ 1.
log(n) (0, n ]
Then we have fn → 0 pointwise and hence also λ-a.e. Moreover
Z
1
fn dλ = →0
[0,1] log(n)

so that fn → 0 in L1 ([0, 1], λ) since fn ≥ 0. However, there exists no µ-summable

function g with g ≥ fn λ-a.e. for all n.
n
Indeed, such a g would have to satisfy g ≥ log(n) λ-a.e. on (0, n1 ] for all n. But then
Z  
n 1 1 1
g 1( 1 , 1 ] dλ ≥ − =
[0,1] n+1 n log(n) n n + 1 (n + 1) log(n)

and hence Z ∞ Z ∞
1
g 1(
X X
gdλ = 1
,1] dλ ≥ = ∞.
[0,1] n+1 n n log(n)
n=1 [0,1] n=2

Definition 3: Uniform Summability

The family (fn )n∈N is called uniformly µ-summable if for all ε > 0 there exists
δ > 0 such that for all n ∈ N and A ⊂ Ω µ-measurable with µ(A) < δ it holds
Z
|fn | dµ < ε.
A

This allows us to formulate a necessary and sufficient condition for L1 convergence.

Questions and Remarks to: [email protected] 3

Spring 2021 ETH Zürich
Measure and Integration Maran Mohanarangan

Theorem 4: Vitali Convergence Theorem

If µ(Ω) < ∞, the following conditions are equivalent:

i) fn → f in L1 (Ω, µ).
µ
ii) fn −
→ f and (fn )n∈N is uniformly µ-summable.

As we hinted at in the remark, L1 -convergence is just one particular case of a more

general concept called Lp -convergence. We will discuss this concept now without really
introducing Lp -spaces.
Definition 5: Lp convergence

Let p ∈ [1, ∞). For f : Ω → R, we define the Lp (Ω, µ) norm by

Z 1
p
p
kf kLp (Ω,µ) = |f | dµ ≤ ∞.
Ω

For p = ∞, we define the L∞ (Ω, µ) norm by

 
kf kL∞ (Ω,µ) := µ-ess supx∈Ω f (x) = inf{C : |f | ≤ C µ-a.e.}.

For 1 ≤ p ≤ ∞, we say that a sequence of µ-measurable functions (fn )n∈N

converges in Lp (Ω, µ) to a measurable function f if

lim kfn − f kLp (Ω,µ) = 0.

n→∞

For 1 ≤ p < ∞, Lp (Ω, µ) convergence is the same as |fn − f |p → 0 in L1 (Ω, µ); for
p = ∞, it means that there exists a set N with µ(N ) = 0 such that (fn )n∈N converges
to f uniformly on N c .
Proposition 6

If µ(Ω) < ∞, then for 1 ≤ r < s ≤ ∞, we have

Ls (Ω, µ) ⊂ Lr (Ω, µ).

In particular, we have that convergence in Ls (Ω, µ) implies convergence in

Lr (Ω, µ).

Assume that µ(Ω) < ∞ and 1 ≤ r ≤ s ≤ ∞. Let (fn )n∈N be a sequence of measur-
able functions. To summarize, we have the following implications:

r≤s
Ls Lr L1

+ uniformly µ-summable

µ-a.e. in measure µ

Questions and Remarks to: [email protected] 4