Chapter 6: The Riemann–Stieltjes Integral

Wissam Raji

Edited by Selena Nadjarian

Contents

1 The Riemann Integral

We begin by recalling the construction of upper and lower sums for the ordinary Riemann
integral.

Definition 1.1 (Partition). Let [a, b] be an interval with a < b. A partition of [a, b] is a finite
set of points
P = {x0 , x1 , . . . , xn }, where a = x0 < x1 < · · · < xn = b.

Definition 1.2 (Upper and Lower Sum). We define the subinterval lengths by:

∆xi = xi − xi−1 , for i = 1, . . . , n.

Let f : [a, b] → R be a bounded function. For each subinterval [xi−1 , xi ] define

Mi = sup{f (x) : xi−1 ≤ x ≤ xi }, mi = inf{f (x) : xi−1 ≤ x ≤ xi }.

Then the upper sum and lower sum of f with respect to P are
n n
U (P, f ) = Mi ∆xi , L(P, f ) = mi ∆xi .
X X

i=1 i=1

Definition 1.3 (Upper and Lower Riemann Integrals). The upper and lower Riemann integrals
of a bounded function f over [a, b] are defined by:
Z b
f (x) dx = inf{U (P, f ) : P is a partition of [a, b]},
a
Z b
f (x) dx = sup{L(P, f ) : P is a partition of [a, b]}.
a

If these two quantities are equal, then f is Riemann integrable on [a, b], written

f ∈R

and we denote the common value by

Z b
f dx.
a

1
 Since f is bounded, there exist constants m and M such that:

m ≤ f (x) ≤ M for all x ∈ [a, b].

It follows that for any partition P ,

m(b − a) ≤ L(P, f ) ≤ U (P, f ) ≤ M (b − a)

Therefore, the upper and lower integrals always exist as real numbers. The key question is
whether they are equal. That is, whether f is Riemann integrable.

f (x)

x
a b

Figure 1: A function on [a, b] together with a partition. Upper sums are formed using rectangles
based on the supremum (top rectangles), while lower sums use the infimum (bottom rectangles)

Example 1.4. Let f (x) = x2 on [0, 1] and consider the partition

P = {0, 12 , 1}.

Since x2 is increasing on [0, 1], we have:

 
M1 = f 1
2 = 14 , M2 = f (1) = 1,
 
m1 = f (0) = 0, m2 = f 1
2 = 41 .
The subinterval lengths are:
1 1
∆x1 = , ∆x2 =
2 2
Hence, the upper and lower sums are:

U (P, f ) = 1
4 · 1
2 +1· 1
2 = 58 ,

L(P, f ) = 0 · 1
2 + 1
4 · 1
2 = 18 .
R1 2
The exact integral is 0 x dx = 3 , which satisfies:
1

1 1 5
L(P, f ) = < < = U (P, f )
8 3 8

2 The Riemann–Stieltjes integral

The Riemann–Stieltjes integral is a generalization of the ordinary Riemann integral.

2
Definition 2.1 (Riemann–Stieltjes Upper and Lower Sums). Let α : [a, b] → R be a monoton-
ically increasing function. For a partition

P = {x0 , x1 . . . , xn }

define
∆αi = α(xi ) − α(xi−1 ) ≥ 0.
Let f : [a, b] → R be bounded. For each subinterval [xi−1 , xi ], define:

Mi = sup{f (x) : xi−1 ≤ x ≤ xi }, Mi = inf{f (x) : xi−1 ≤ x ≤ xi },

Then, the upper and lower sums are given by:

n n
U (P, f, α) = Mi ∆αi , L(P, f, α) = mi ∆αi ,
X X

i=1 i=1

f (x)

x
x0 x1 x2 xn

∆α1 ∆α2 ∆αn

Figure 2: Lower sums for a Riemann-Stieltjes integral. The area of each rectangle is mi ∆αi ,
meaning the "width" is determined by how much α increases over [xi−1 , xi ], not by the length
of the interval.

Definition 2.2 (Riemann–Stieltjes Integral). The upper and lower Riemann − Stieltjes
integrals are defined by:
Z b
f dα = inf{U (P, f, α) : P is a partition of [a, b]},
a
Z b
f dα = sup{L(P, f, α) : P is a partition of [a, b]}.
a

If these two quantities are equal, we say that f is Riemann − Stieltjes integrable with respect
to α on [a, b], written:

f ∈ R(α)
and we denote the common value by
Z b
f dα.
a

3
Remark 2.3. If α(x) = x, then
∆αi = ∆xi
so the Riemann–Stieltjes integral reduces to the ordinary Riemann integral.
Z b Z b
f dα = f (x) dx
a a

Example 2.4. Let α(x) = x2 on [0, 1] and f (x) = 1. Then for any partition P ,
n
U (P, 1, α) = L(P, 1, α) = (α(xi ) − α(xi−1 ))
X

i=1
This sum telescopes:
n
(α(xi ) − α(xi−1 )) = α(xn ) − α(x0 ) = α(1) − α(0) = 1
X

i=1
Hence Z 1
1 d(x2 ) = 1.
0
This example shows that the value of the integral depends on the integrator α not just the
function f

3 Refinements and the Basic Inequality

Definition 3.1 (Refinement). A partition P ∗ is called a refinement of a partiion P if
P ⊂ P∗
If P1 and P2 are two partitions, then their common refinement is
P ∗ = P1 ∪ P2 .

y0 y1 y2

x0 x1 x2 x3 x4

Figure 3: A refinement inserts extra points into a partition.

Theorem 3.2 (Monotonicity under Refinement). If P ∗ is a refinement of P , then

L(P, f, α) ≤ L(P ∗ , f, α), U (P ∗ , f, α) ≤ U (P, f, α).
Proof. It suffices to consider the case where P ∗ is obtained from P by adding a single point
x∗ ∈ (xi−1 , xi )
Define
w1 = inf{f (x) : xi−1 ≤ x ≤ x∗ }, w2 = inf{f (x) : x∗ ≤ x ≤ xi }.
Since w1 ≥ mi and w2 ≥ mi we compute,
L(P ∗ , f, α) − L(P, f, α) = w1 α(x∗ ) − α(xi−1 ) + w2 α(xi ) − α(x∗ )

− mi α(xi ) − α(xi−1 )

≥ 0.
Thus
L(P, f, α) ≤ L(P ∗ , f, α)
The proof for upper sums is analogous.

4
Theorem 3.3 (Basic Inequality). For every bounded function f ,
Z b Z b
f dα ≤ f dα.
a a

Proof. Let P1 , P2 be two partitions and let their common refinement be:

P ∗ = P1 ∪ P2

By the previous theorem,

L(P1 , f, α) ≤ L(P ∗ , f, α) ≤ U (P ∗ , f, α) ≤ U (P2 , f, α).

Since this is true for all partitions P1 and P2 , take the supremum over all lower sums and the
infimum over all upper sums to obtain
Z b Z b
f dα ≤ f dα.
a a

4 Criterion for integrability

Theorem 4.1 (Criterion for Riemann–Stieltjes Integrability). A bounded function f belongs to
R(α) on [a, b] if and only if for every ε > 0 there exists a partition P such that

U (P, f, α) − L(P, f, α) < ε.

Proof. For every partition P ,

Z b Z b
L(P, f, α) ≤ f dα ≤ f dα ≤ U (P, f, α).
a a

(⇒) : If there exists a partition P such that

U (P, f, α) − L(P, f, α) < ε

then Z b Z b
0≤ f dα − f dα < ε.
a a

Since this holds for every ε > 0, the upper and lower integrals are equal. Hence, f ∈ R(α)
(⇐) :Assume f ∈ R(α) and let
Z b
I= f dα
a
Choose partitions P1 and P2 such that
ε ε
U (P1 , f, α) − I < , I − L(P2 , f, α) < .
2 2
Let P = P1 ∪ P2 . By refinement monotonicity,

U (P, f, α) ≤ U (P1 , f, α), L(P2 , f, α) ≤ L(P, f, α).

Therefore
ε ε
U (P, f, α) − L(P, f, α) < + < ε.
2 2

5
Theorem 4.2 (Useful Consequences of the Criterion). Suppose a partition P satisfies

U (P, f, α) − L(P, f, α) < ε.

Then:
(a) every refinement P ∗ of P also satisfies

U (P ∗ , f, α) − L(P ∗ , f, α) < ε;

(b) if P = {x0 , . . . , xn } and ti ∈ [xi−1 , xi ], then for any other choice ci ∈ [xi−1 , xi ],
n
|f (ti ) − f (ci )| ∆αi < ε
X

i=1

(c) if f ∈ R(α) and the hypothesis of (b) holds, then

n Z b
f (ti ) ∆αi −
X
f dα < ε.
i=1 a

Proof. Part (a) follows immediately from monotonicity under refinement. For (b), since

|f (ti ) − f (ci )| ≤ Mi − mi ,

we have
n n
|f (ti ) − f (ci )| ∆αi ≤ (Mi − mi )∆αi = U (P, f, α) − L(P, f, α) < ε.
X X

i=1 i=1

For (c), using

n
f (ti )∆αi ≤ U (P, f, α)
X
L(P, f, α) ≤
i=1
and comparing both sides with the integral, we obtain the desired inequality.

5 Integrability Results
Theorem 5.1 (Continuous Functions are Integrable). If f is continuous on [a, b], then f ∈ R(α)
on [a, b].
Proof. Let ε > 0. Choose η > 0 such that


η α(b) − α(a) < ε.

Since f is continuous on the compact interval [a, b], it is uniformly continuous. Hence, there
exists δ > 0 such that
|x − t| < δ =⇒ |f (x) − f (t)| < η.
Take a partition P with mesh maxi ∆xi < δ. Then on each subinterval [xi−1 , xi ],

Mi − mi < η.

Therefore
n n
U (P, f, α) − L(P, f, α) = (Mi − mi )∆αi < η ∆αi = η α(b) − α(a) < ε.
X X

i=1 i=1

Now apply Theorem ??.

6
Theorem 5.2 (Monotone Functions). If f is monotonic on [a, b] and α is continuous on [a, b],
then f ∈ R(α).
Proof. Assume f is increasing (the decreasing case is similar). Since α is continuous and in-
creasing, for each integer n there exists a partition

Pn = {x0 , x1 . . . , xn }

such that
α(b) − α(a)
∆αi = α(xi ) − α(xi−1 ) = .
n
Since f is increasing,
Mi = f (xi ), mi = f (xi−1 ).
Thus,
n
U (Pn , f, α) − L(Pn , f, α) = (Mi − mi )∆αi
X

i=1
n
α(b) − α(a) X
= f (xi ) − f (xi−1 )


n i=1
α(b) − α(a)
= f (b) − f (a) .


n
This tends to 0 as n → ∞, so the criterion applies.

Theorem 5.3 (Finitely Many Discontinuities). Suppose f is bounded on [a, b] and has only
finitely many points of discontinuity. Assume also that α is continuous at every point where f
is discontinuous. Then f ∈ R(α).
Idea of proof. Let E = {s1 , . . . , sm } be the set of discontinuities of f . Since α is continuous at
each sj , we can choose small intervals (uj , vj ) around each sj such that
m
α(vj ) − α(uj )
X

j=1

is arbitrarily small. Outside these intervals, f is continuous on a compact set, hence uniformly
continuous.
Construct a partition P such that:
• P contains all endpoints uj , vj ,

• P is sufficiently fine on the remaining subintervals.

Then:
• the contribution to U (P, f, α) − L(P, f, α) from the “bad” intervals is small (controlled by
α),

• the contribution from the “good” part is small (controlled by uniform continuity).

Example 5.4 (Shared Discontinuity Counterexample). To see why α must be continuous at

the discontinuities of f , consider the interval [−1, 1].
Define
0, x ≤ 0, 0, x ≤ 0,
( (
f (x) = α(x) =
1, x > 0, 1, x > 0.

7
 Both functions have a jump discontinuity at x = 0.
For any partition P of [−1, 1] that contains 0, there exists a subinterval [0, xk ] where the
jump occurs. On this subinterval,

Mk = 1, mk = 0, ∆αk = 1.

Thus, the difference between the upper and lower sums contains the term

(Mk − mk ) ∆αk = (1 − 0)(1) = 1.

Hence,
U (P, f, α) − L(P, f, α) ≥ 1
for all such partitions. Therefore, the condition in Theorem 4.1 cannot be satisfied for ε < 1,
and we conclude that
f∈/ R(α).
Theorem 5.5 (Composition with a Continuous Function). Suppose f ∈ R(α) on [a, b], and

m ≤ f (x) ≤ M ∀x ∈ [a, b].

If φ is continuous on [m, M ] and h(x) = φ(f (x)), then

h ∈ R(α)

.
Idea of proof. Since φ is continuous on the compact interval [m, M ], it is uniformly continuous.
Given ε > 0, choose δ > 0 such that

|s − t| < δ =⇒ |φ(s) − φ(t)| < ε.

Take a partition P such that

U (P, f, α) − L(P, f, α) < δ 2 .

Split the indices into two sets:

• those for which Mi − mi < δ,

• those for which Mi − mi ≥ δ.

Then:
• on the first set, the oscillation of h is small by uniform continuity,

• on the second set, the total α-weight is small, since a large oscillation of f can occur only
on a small total α-measure.
Combining these two estimates shows that

U (P, h, α) − L(P, h, α)

can be made arbitrarily small.

Remark 5.6. A deep result from measure theory states that for bounded functions on [a, b],

f ∈R ⇐⇒ f is continuous almost everywhere.

The chapter notes mention this as a later result proved using measure theory.

8
6 Properties of the Integral
Theorem 6.1 (Algebraic Properties). Let f1 , f2 ∈ R(α) on [a, b] and let c ∈ R. Then:

1. f1 + f2 ∈ R(α) and
Z b Z b Z b
(f1 + f2 ) dα = f1 dα + f2 dα;
a a a

2. cf1 ∈ R(α) and

Z b Z b
cf1 dα = c f1 dα;
a a

3. if f1 (x) ≤ f2 (x) for all x ∈ [a, b], then

Z b Z b
f1 dα ≤ f2 dα;
a a

4. if a < c < b and f ∈ R(α) on [a, b], then f ∈ R(α) on both [a, c] and [c, b], and
Z c Z b Z b
f dα + f dα = f dα.
a c a

Theorem 6.2 (Absolute Value Estimate). If f ∈ R(α) on [a, b] and |f (x)| ≤ M on [a, b], then
Z b Z b


f dα ≤ |f | dα ≤ M α(b) − α(a) .
a a

Theorem 6.3 (Changing the Integrator). If f ∈ R(α1 ) and f ∈ R(α2 ), then f ∈ R(α1 + α2 )
and Z b Z b Z b
f d(α1 + α2 ) = f dα1 + f dα2 .
a a a
If c ≥ 0 and f ∈ R(α), then f ∈ R(cα) and
Z b Z b
f d(cα) = c f dα.
a a

Theorem 6.4 (Integration by Parts). If f ∈ R(α) on [a, b], then α ∈ R(f ) on [a, b], and
Z b Z b
f dα + α df = f (b)α(b) − f (a)α(a).
a a

7 The Case α′ ∈ R
Theorem 7.1. Assume the increasing function α has the additional property that α′ ∈ R on
[a, b]. Let f be bounded on [a, b]. Then

f ∈ R(α) ⇐⇒ f α′ ∈ R,

and in this case Z b Z b

f dα = f (x)α′ (x) dx.
a a

9
Idea of proof. Fix ε > 0 and choose a partition P such that

U (P, α′ ) − L(P, α′ ) < ε.

By the mean value theorem, on each interval [xi−1 , xi ] there exists ti ∈ [xi−1,xi ] such that

α(xi ) − α(xi−1 ) = α′ (ti )(xi − xi−1 ).

Then the Riemann–Stieltjes sums for f and the Riemann sums for f α′ differ by a quantity
controlled by the oscillation of α′ over the partition. Since f is bounded, this difference is at
most
M ε, where M = sup |f |.
Passing to upper and lower integrals yields the desired formula.

Example 7.2 (Recovering the Ordinary Density). If α(x) = x2 , then α′ (x) = 2x. Hence, for
every bounded function f for which the relevant integrals exist,
Z b Z b
f d(x2 ) = 2xf (x) dx.
a a

In particular,
2
Z 1 Z 1
x d(x2 ) = 2x2 dx = .
0 0 3
Example 7.3 (Importance of Monotonicity). Consider
Z 1
x2 d(x3 )
−1

Here, α(x) = x3 , which is strictly increasing on [−1, 1], so the integral is well-defined. Using
Theorem 7.1, we have α′ (x) = 3x2 , and therefore

6
Z 1 Z 1 Z 1
x2 d(x3 ) = x2 (3x2 ) dx = 3 x4 dx = .
−1 −1 −1 5

8 Applications: Mass and Moment of Inertia

Example 8.1 (Mass of a Wire). Consider a wire of unit length along the interval [0, 1] with
continuous density ρ(x) ≥ 0. The mass of a small segment [xi−1 , xi ] is approximately

ρ(ξi ) ∆xi .

For some ξi ∈ [xi−1 , xi ]. Summing over all subintervals and passing to the limit gives
Z 1
Mass = ρ(x) dx.
0

If we define Z x
α(x) = ρ(t) dt,
0
then α(x) represents the mass of the piece [0, x] and

α′ (x) = ρ(x).

10
Example 8.2 (Moment of Inertia). For the same wire, the moment of inertia about the origin
is obtained by summing each mass element times the square of its distance from the origin. A
small segment near x contributes approximately
x2 ρ(x) dx.
Therefore Z 1
I= x2 ρ(x) dx.
0
Equivalently, if dα denotes the mass distribution, then
Z 1
I= x2 dα.
0
By Theorem ??, these expressions agree because α′ (x) = ρ(x).

9 Integration and Differentiation

Theorem 9.1 (An Integral Defines a Continuous Function). If f ∈ R on [a, b] and
Z x
F (x) = f (t) dt, a ≤ x ≤ b,
a
then F is continuous on [a, b]. Moreover, if f is continuous at x0 ∈ [a, b], then F is differentiable
at x0 and
F ′ (x0 ) = f (x0 ).
Proof. Since f is bounded, there exists M > 0 such that |f (t)| ≤ M on [a, b]. For x, y ∈ [a, b],
Z y Z y
|F (y) − F (x)| = f (t) dt ≤ |f (t)| dt ≤ M |y − x|.
x x
Thus, F is continuous. If f is continuous at x0 , then for t > x0 ,
F (t) − F (x0 ) 1 t Z
− f (x0 ) = (f (u) − f (x0 )) du
t − x0 t − x 0 x0
1
Z t
≤ |f (u) − f (x0 )| du.
t − x 0 x0
The right-hand side tends to 0 by continuity of f at x0 . A similar argument holds for t < x0 ,
so F ′ (x0 ) = f (x0 )
Theorem 9.2 (Fundamental Theorem of Calculus). Let f ∈ R on [a, b]. If there exists a
differentiable function F on [a, b] such that F ′ = f , then
Z b
f (x) dx = F (b) − F (a).
a
Proof. Choose a partition P = {x0 , x1 . . . , xn } such that
U (P, f ) − L(P, f ) < ε.
By the Mean Value Theorem, for each i there exists ti ∈ [xi−1 , xi ] such that
F (xi ) − F (xi−1 ) = F ′ (ti )(xi − xi−1 ) = f (ti )∆xi .
Summing over all subintervals,
n
F (b) − F (a) = f (ti )∆xi .
X

i=1
Rb
Using the estimate for Riemann sums, this differs from a f (x) dx by less than ε. Since ε is
arbitrary, equality follows.

11
Theorem 9.3 (Integration by Parts). Let F, G be differentiable on [a, b], and write F ′ = f ,
G′ = g, where f, g ∈ R on [a, b]. Then
Z b Z b
F (x)g(x) dx = F (b)G(b) − F (a)G(a) − f (x)G(x) dx.
a a

Proof. Let H(x) = F (x)G(x). Then

H ′ (x) = F (x)g(x) + f (x)G(x).

Apply the Fundamental Theorem of Calculus to H ′ .

Example 9.4. Take F (x) = x and G(x) = ex on [0, 1]. Then

Z 1 Z 1
xe dx =
x
[xex ]10 − ex dx = e − (e − 1) = 1.
0 0

10 Step-function Integrators and Point Masses

Theorem 10.1 (A Single Jump). Let a < s < b, let f be bounded on [a, b], and suppose f is
continuous at s. Define the unit step integrator

0,
(
x ≤ s,
α(x) = I(x − s) =
1, x > s.

Then Z b
f dα = f (s).
a

Idea of the Proof. Use partitions of the form Pn = {a, s, s + εn , b} with εn ↓ 0. Since f is
continuous at s, the oscillation of f on [s, s + εn ] tends to 0. The upper and lower sums both
converge to f (s).

α(x)

x
s

Figure 4: The unit step function I(x − s). The integral with respect to this integrator picks out
the value at the jump.

P∞
Theorem 10.2 (Countably Many Jumps). Suppose cn ≥ 0, the series n=1 cn converges, and
{sn } is a sequence of distinct points in (a, b). Define
∞
α(x) = cn I(x − sn ).
X

n=1

If f is continuous on [a, b], then

Z b ∞
f dα = cn f (sn ).
X
a n=1

12
Idea of the Proof. Split
N ∞
α = α1 + α2 , α1 (x) = cn I(x − sn ), α2 (x) = cn I(x − sn ).
X X

n=1 n=N +1

For the finite sum, repeated application of the single-jump theorem gives
Z b N
f dα1 = cn f (sn ).
X
a n=1

The tail α2 has total variation at most ∞

n=N +1 cn , so by the absolute value estimate its integral
P

can be made arbitrarily small. Let N → ∞.

11 Change of Variables
Theorem 11.1 (Change of Variables). Suppose

• ψ : [A, B] → [a, b] is strictly increasing and continuous,

• α is monotonically increasing on [a, b],

• f ∈ R(α) on [a, b].

Define
β(y) = α(ψ(y)), g(y) = f (ψ(y)).
Then g ∈ R(β) on [A, B] and
Z B Z b
g dβ = f dα.
A a

Sketch. Let Q = {y0 , y1 . . . , yn } be a partition of [A, B], and define xi = ψ(yi ). Then P =
{x0 , x1 . . . , xn } is a partition of [a, b]. Since ψ is strictly increasing, it preserves order. Therefore,

U (Q, g, β) = U (P, f, α), L(Q, g, β) = L(P, f, α).

Hence, the criterion for integrability transfers from (f, α) to (g, β), and the integrals are equal.

Corollary 11.2 (Usual Substitution Formula). If α(x) = x and ψ ′ ∈ R on [A, B], then
Z b Z B
f (x) dx = f (ψ(y))ψ ′ (y) dy.
a A

Example 11.3. To compute Z 1p

1 − x2 dx,
0
use the substitution x = sin y with y ∈ [0, π2 ]. Then dx = cos y dy and
Z 1p Z π/2 q Z π/2
π
1− x2 dx = 1− sin2 y cos y dy = cos2 y dy = .
0 0 0 4

13