JORDAN CONTENT

Definition. Let A ⊂ Rn be a bounded set. Given a rectangle (cartesian

product of compact intervals) R ⊂ Rn containing A, denote by P the set
of finite partitions of R by sub-rectangles (‘intervals’) defined by some grid.
Consider the numbers:
X
J(P, A) = {m(Ik ); Ik an interval of P contained in int(A)}
X
¯ A) =
J(P, {m(Ik ); Ik an interval of P intersecting cl(A)}.
Here int(A) and cl(A) = A ∪ ∂A denote the interior and the closure of A
¯ A), and we set J(P, A) = 0 for any P if
(resp.) Clearly 0 ≤ J(P, A) ≤ J(P,
A has empty interior.
We define the inner and outer Jordan content of A by taking sup (resp
inf):

c(A) = sup{J(P, A); P ∈ P}; ¯ A); P ∈ P}.

c̄(A) = inf{J(P,

A is Jordan-measurable if these two numbers coincide: c(A) = c̄(A) := c(A),

the Jordan content of A.
It is easy to see that the inner and outer Jordan content are independent
of the choice of reference rectangle R containing A.
The following is clear geometrically, and easy to prove. For any partition
P ∈ P of R, and any A ⊂ R:
¯ A) − J(P, A) = J(P,
J(P, ¯ ∂A).

This implies the following.

Proposition 1. Let A ⊂ Rn be a bounded set. Then c̄(A)−c(A) = c̄(∂A).
In particular, A is Jordan-measurable if and only if c̄(∂A) = 0.
Proof. Let R be a rectangle containing A; consider an arbitrary partition
P of R. Then:
¯ ∂A) = J(P,
J(P, ¯ A) − J(P, A) ≥ c̄(A) − c(A);

since this holds for all P , we have c̄(∂A) ≥ c̄(A) − c(A).

For the opposite inequality, given  > 0 consider partitions P1 , P2 of R
so that:
¯ 1 , A) − , c(A) ≤ J(P2 , A) + .
c̄(A) ≥ J(P

1
Let P be a partition of R refining both P1 and P2 . Refinement decreases J¯
and increases J, so:
¯ ∂A) = J(P,
c̄(∂A) ≤ J(P, ¯ A)−J(P, A) ≤ J(P
¯ 1 , A)−J(P2 , A) ≤ c̄(A)−c(A)+2,

and now let  → 0.

Remarks.
(1) The sets of rational numbers and of irrational numbers in an interval
[a, b] ⊂ R both have inner content zero, and outer content b − a (Check this.)
This example also shows that neither the inner nor the outer Jordan content
is additive (over disjoint sets.)
(2) However, Jordan content is finitely additive: if A, B are disjoint
Jordan-measurable sets, then A ∪ B is Jordan-measurable and c(A ∪ B) =
c(A) + c(B)
Problem. (i) Let A, B ⊂ Rn be bounded sets. We have:
c̄(A ∪ B) + c̄(A ∩ B) ≤ c̄(A) + c̄(B).
(Analogously c(A ∪ B) + c(A ∩ B) ≥ c(A) + c(B).)
(ii) If A, B ⊂ R are bounded and Jordan measurable, then A ∪ B and
A ∩ B are Jordan measurable and

c(A ∪ B) + c(A ∩ B) = c(A) + c(B).

(It is also true that If A and B are Jordan measurable and B ⊂ A, then
A \ B is Jordan measurable and c(A \ B) = c(A) − c(B).)
(3) We have the following connection with the Riemann integral: the
outer (resp. inner) Jordan content of a bounded set A equals the upper
(resp. lower) Riemann integral of its characteristic function.

Question 1. Is a countable union of (disjoint) Jordan measurable sets

Jordan measurable? Is Jordan content countably additive?
Question 2. Is every bounded open subset of Rn Jordan-measurable?
Question 3. Let C ⊂ [0, 1] be the standard Cantor set, A = [0, 1] \ C. Is
A Jordan-measurable? What if C has positive outer Lebesgue measure?

LEBESGUE MEASURE. (Following [Fleming, ch.5])

Outer measure. For a bounded set A ⊂ Rn , define:

X [
m∗ (A) = inf{ vol(Rk ); A ⊂ Rk },
k≥1 k≥1

2
where the infimum is taken over all countable coverings of A by open rect-
angles Rk .
Problem 1. For every bounded set A ⊂ Rn we have m∗ (A) ≤ c̄(A).
Hint: Let P be any partition of a rectangle R containing A. Consider
the rectangles Rk of P that intersect A ∪ ∂A. Then, by definition:
X
¯ A) =
J(P, vol(Rk ).
k≥1

Let  > 0 be arbitrary. Show that by enlarging the Rk slightly we obtain a

countable covering {Rk0 }k≥1 of A by open rectangles, satisfying:
X
vol(Rk0 ) = J(P,
¯ A) + .
k≥1

Problem 2. Let A ⊂ Rn be a bounded set, K ⊂ A a compact set. Then

c̄(K) ≤ m∗ (A).
Hint: Let R be a rectangle containing A; let
P  > 0 be given.∗ Take a
countable open covering {Rk }k≥1 of A so that k≥1 vol(Rk ) < m (A) + .
Show that a finite subcover of K determines a partition P of R for which
¯ K) ≤ m∗ (A) + . Conclude.
J(P,
Corollary. If K ⊂ Rn is compact, then c̄(K) = m∗ (K).
We next define Lebesgue measure of bounded sets in Rn . First we define
measure for bounded open sets, and for compact sets. This is based on
approximating from the inside (resp. from the outside) by figures, where
a figure P is a finite union of closed rectangles Rk , k = 1, . . . N . We may
assume the rectangles comprising P have disjoint interior, and then the
volume vol(P ) of P is defined in the obvious geometric way.
Definitions.

(i) Let U ⊂ Rn be a bounded open set. We define its measure by:

m(U ) = sup{vol(P ); P ⊂ U a figure}.

Problem 0.1. Show that, for a bounded open set U : m(U ) = c(U ), the
lower Jordan content.
If K ⊂ Rn is compact, we define:

m(K) = inf{vol(P ), P a figure , K ⊂ int(P )}.

3
Problem 0.2. Show that, for a compact set K: m(K) = c̄(K), the upper
Jordan content.
Consistency check: It is easy to verify that m(P ) = vol(P ) and m(intP ) =
vol(P ), if P is a figure.
(iii) Let A ⊂ Rn be bounded. We define the outer measure of A approx-
imating from the outside by open sets:

m(A) = inf{m(U ); A ⊂ U, U open }.

(iv) Let A ⊂ Rn be bounded. We define the inner measure of A approx-

imating from the inside by compact sets:

m(A) = sup{m(K); K ⊂ A, K compact }.

Remark: Since, for K compact and U open, we always have K ⊂ U ⇒

m(K) ≤ m(U ) (Justify!,) we have, for all bounded A: m(A) ≤ m̄(A).
(v) We say a bounded set A ⊂ Rn is measurable if m(A) = m̄(A), and
then set m(A) to be their common value.
Problem 0.3 We now have potentially two definitions of outer measure
(of a bounded set). Show that these definitions are equivalent: m̄(A) =
m∗ (A), for any bounded set A ⊂ Rn .
Consistency check.
Problem 3. Bounded open sets U are Lebesgue measurable, and m(U ) =
c(U ), the lower Jordan content.
Problem 4. Compact sets K are Lebesgue measurable, and m(K) =
c̄(K), the upper Jordan content.
Question 4. Is every (bounded) Jordan-measurable set Lebesgue-measurable?
Hint: Show that we have, for all bounded A:

c(A) ≤ m(A) ≤ m̄(A) ≤ c̄(A).

Cauchy criterion. A ⊂ Rn (bounded) is measurable if, and only if,

for all  > 0 we may find K ⊂ A ⊂ U (K compact, U open) so that
m(U \ K) < .
This uses the fact that K ⊂ U implies m(U ) = m(K) + m(U \ K).
First properties. (See [Fleming, 5.2] for proofs.)

4
 1. (Finite additivity.) A, B ⊂ Rn bounded, measurable, disjoint ⇒
A t B measurable, and:
m(A t B) = m(A) + m(B).
2. A, B ⊂ Rn measurable ⇒ A \ B, A ∩ B, A ∪ B measurable and:
m(A ∪ B) + m(A ∩ B) = m(A) + m(B).
For unbounded sets A ⊂ Rn , we say A is measurable if A ∪ BR is
measurable, for each R > 0 (open ball of radius B), and set m(A) =
limR→∞ m(A ∩ BR ). We allow the possibility m(A) = +∞.
3. A measurable ⇒ Ac measurable (complement).
4. (Countable
S additivity)
T If (An )n≥1 is a countable family of measurable
sets, then n≥1 An and n≥1 An are measurable, and we have:
[ X
m( An ) ≤ m(An ),
n≥1 n≥1

with equality if the sets An are disjoint.

5. (Limits.) (i) If (An )n≥1 are measurable and An ⊂ An+1 , then
S
m( n≥1 An ) = lim m(An ).
(ii) If (An )n≥1 are measurable and An ⊃ An+1 and m(A1 ) < ∞, then
T
m( n≥1 m(An )) = lim m(An ).

RIEMANN INTEGRAL
Let R ⊂ Rn be a rectangle, f : R → R bounded. Given a finite partition
P = {Ik }N
k=1 of R, define upper and lower Riemann sums by:
N
X N
X
U (P, f ) = Mk m(Ik ), L(P, f ) = mk m(Ik ), Mk = sup f, mk = inf f.
Ik Ik
k=1 k=1

Definition.f is Riemann integrable if for all  > 0 we may find a partition P

of R so that:
U (P, f ) − L(P, f ) < .
Analogously to the Lebesgue theory, we define upper and lower Riemann
integrals by taking inf and sup (resp.) of upper and lower Riemann sums
over the set P of finite partitions of R:
Z¯ Z
f dV = inf{U (P, f ); P ∈ P}; f dV = sup{L(P, f ); P ∈ P}.

5
It is easy to show that f is Riemann integrable if and only if upper and lower
Riemann integrals coincide; the common value is the Riemann integral of f :
Z¯ Z Z
f integrable ⇔ f dV = f dV := f dV.

(We use dm for the Lebesgue integral, dV for the Riemann integral.)
Riemann integrability criteria. For f : R → R (R ⊂ Rn a rectangle)
denote by ωf (x) the oscillation of f at x. Let Df ⊂ R be the discontinuity
set, and denote by D the closed set:
[
D = {x ∈ R; ωf (x) ≥ }, so Df = {x ∈ R; ωf (x) > 0} = D1/n .
n≥1

Theorem. Jordan content criterion. f : R → R (bounded) is Riemann

integrable if, and only if, c̄(D ) = 0, ∀ > 0.
Proof. (i) Assume first c̄(Dδ ) 6= 0 for some δ > 0. Let P be a partition
of R. Then:
N
X X
U (P, f )−L(P, f ) = [Mk −mk ]m(Ik ) ≥ ¯ Dδ ),
[Mk −mk ]m(Ik ) > δ J(P,
k=1 {k;Ik ∩Dδ 6=∅}

¯ Dδ ) equals the
using the facts that (i) Mk − mk ≥ δ if Ik ∩ Dδ 6= ∅; (ii) J(P,
sum of m(Ik ) over all k so that Ik ∩ Dδ 6= ∅, since Dδ is closed.
Thus we have a positive lower bound δc̄(Dδ ) (independent of P ) for the
difference U (P, f ) − L(P, f ); so f can’t be Riemann integrable over R.
(ii) Conversely, assume c̄(D ) > 0 for all  > 0. Given  > 0, let P0
be a partition of R so that J(P¯ 0 , D ) < . In each rectangle I of P not
intersecting D (or contained in the complement Dc ) we have ωf (x) <  for
x ∈ I. It follows by a Lebesgue number argument that there exists a δ > 0
so that if I ⊂ Dc and diam(I) < δ then the oscillation of f in I is smaller
than .
Thus we refine P0 so that all intervals of the refined partition P =
{Ik }nk=1 have diameter smaller than δ. Let A1 = {k; Ik ∩ D 6= ∅}, A2 =
{k; Ik ⊂ Dc }. Then:
X X
U (P , f ) − L(P , f ) = ( + )[Mk − mk ]m(Ik ).
k∈A1 k∈A2

6
The sum over A1 is bounded above by (M −m)J(P ¯  , D ) ≤ (M −m)J(P
¯ 0 , D ),
where M, m are the sup (resp. inf) of f in R. For the terms in the A2 sum,
as seen in the previous paragraph we have Mk − mk < . We conclude the
difference U − L is bounded above by (M − m) + m(R), and hence can be
made as small as desired. This shows f is Riemann-integrable.

Vitali’s theorem. f : R → R (bounded) is Riemann integrable if, and

only if, m∗ (Df ) = 0.
Proof. (i) Assume f is Riemann-integrable. Then m∗ (D ) ≤ c̄(D ) = 0
for all  > 0 (Problem 1 above), hence by subadditivity of m∗ :
X
m∗ (Df ) ≤ m∗ (D1/n ) = 0.
n≥1

(ii) Conversely, if m∗ (Df ) = 0, since D ⊂ Df and D is compact (for all

 > 0), it follows that (Problem 2) c̄(D ) ≤ m∗ (Df ) = 0 for all  > 0. Thus
f is Riemann integrable.
Now if A ⊂ Rn is a bounded set and f : A → R is bounded, we let
R be any rectangle containing R in its interior and consider the extension
fA : R → R of f to R defined by setting fA (x) = 0 for x ∈ R \ A. We say f
is integrable over A if fA is integrable in R. Note that the discontinuity set
of fA is contained in the union Df ∪ ∂A. We conclude:
Corollary. Let A ⊂ Rn be bounded. If m∗ (∂A) = 0 and f : A → R is
bounded, with m∗ (Df ) = 0, then f is integrable over A.
Remark: Note that since ∂A is compact, the condition m∗ (∂A) = 0 is
equivalent to saying A is Jordan measurable.