Calculus

Lecture 28: Double Integration: II

Prof. Madhusudan Manjunath

IIT Bombay

Prof. Madhusudan Manjunath, IIT Bombay Calculus: Lecture 28

Today at 8:30 pm, in this hall: Midterm answer sheets.
Fubini Theorem: Examples:
(i) Let R := [0, 1]×[0, 1], and f (x, y ) := (x + y )2 , (x, y ) ∈ R.
Then f is continuous on R. The double integral of f on R is
Z 1Z 1
1 1
 Z
2
1
(x + y ) dx dy = (x + y )3 0 dy
0 0 3 0
1 1
Z
7
(1 + y )3 − y 3 dy = .


=
3 0 6

Prof. Madhusudan Manjunath, IIT Bombay Calculus: Lecture 28

(ii) Let R := [0, 1]×[0, 1], f (0, 0) := 0, and for (x, y ) 6= (0, 0),
let f (x, y ) := xy (x 2 − y 2 )/(x 2 + y 2 )3 . For x ∈ [0, 1], x 6= 0,
1 1
xy (x 2 − y 2 )
Z Z
x
A(x) := f (x, y )dy = 2 2 3
dy = .
0 0 (x + y ) 2(1 + x 2 )2
R1
(Substitute x 2 + y 2 = u.) Also, A(0) = 0
0dy = 0. Hence
Z 1Z 1  Z 1 Z 1
x 1
f (x, y )dy dx = A(x)dx =
2 2
dx = .
0 0 0 0 2(1 + x ) 8
Z 1Z 1 
1
By interchanging x and y , f (x, y )dx dy = − .
0 0 8
Thus the two iterated integrals exist, but they are not equal.
Note that since f (1/n, 1/2n) = 24n2 /125 for all n ∈ N, the
function f is not bounded on R. Thus Fubini’s theorem is not
applicable.
Prof. Madhusudan Manjunath, IIT Bombay Calculus: Lecture 28
Double Integral over a Bounded Set
Let D be a bounded subset of R2 , and let f : D → R be a
bounded function. Consider a rectangle R := [a, b] × [c, d]
such that D ⊂ R, and define f ∗ : R → R by
(
f (x, y ) if (x, y ) ∈ D,
f ∗ (x, y ) :=
0 otherwise.
We say that f is integrable over D if f ∗ is integrable on R,
and in this case, the double integral of f over D is defined
to be the double integral of f ∗ on R, that is,
ZZ ZZ
f (x, y )d(x, y ) := f ∗ (x, y )d(x, y ).
D R
By the domain additivity of double integrals on rectangles, the
integrability of f over D and the value of its double integral
are independent of the choice of a rectangle R containing D
and the corresponding extension f ∗ of f to R.
Prof. Madhusudan Manjunath, IIT Bombay Calculus: Lecture 28
Let D be a bounded subset of R2 , and let f : D → R be
integrable over
RR D. We may also denote the double integral of
f over D by D f .
If f is nonnegative, then the volume of the solid under the
surface given by z = f (x, y ) and above the region D is defined
to be the double integral of f over D. Thus
ZZ
Vol (Ef ) := f (x, y )d(x, y ),
D

where Ef := {(x, y , z) ∈ R3 : (x, y ) ∈ D and 0 ≤ z ≤ f (x, y )}.

Next, if g : D → R is integrable, and 0 ≤ f ≤ g on D, then
ZZ


g (x, y ) − f (x, y ) d(x, y )
D

is defined to be the volume between the surfaces given by

z = f (x, y ) and z = g (x, y ).
Prof. Madhusudan Manjunath, IIT Bombay Calculus: Lecture 28
The double integral over a bounded subset of R2 has algebraic
and order properties analogous to those of the double integral
on a rectangle.
In order to seek conditions under which a bounded function f
defined on a bounded subset D of R2 is integrable over D, we
introduce a new concept.
A bounded subset E of R2 is said to be of (two-dimensional)
content zero if for every  > 0, there are finitely many
rectangles whose union contains E and the sum of whose
areas is less than .
A set of (two-dimensional) content zero is a ‘thin’ subset of R2 .
Examples:
(i) Every finite subset of R2 is of content zero.
(ii) The infinite subset {(1/n, 1/k) : n, k ∈ N} of R2 is of
content zero.
Prof. Madhusudan Manjunath, IIT Bombay Calculus: Lecture 28
(iii) The subset {(x, y ) ∈ [0, 1]×[0, 1] : x, y ∈ Q} of R2 is not
of content zero. (Consider any  less than 1.)
(iv) Let ϕ : [a, b] → R be an integrable function. Then its
graph E := {(x, ϕ(x)) : x ∈ [a, b]} is of content zero. To see
this, let  > 0. By the Riemann condition, there is a partition
P := {x0 , x1S
, . . . , xn } of [a, b] such that U(P, ϕ)−L(P, ϕ) < .
Then E ⊂ ni=1 Ri , where Ri := [xi−1 , xi ]×[mi (ϕ), Mi (ϕ)] and

Area (R1 ) + · · · + Area (Rn ) = U(P, ϕ)−L(P, ϕ) < .

Similarly, if ψ : [c, d] → R is an integrable function, then the

set E := {(ψ(y ), y ) : y ∈ [c, d]} is of content zero.
(v) Let functions x, y : [a, b] → R be continuous on [a, b] and
differentiable on (a, b), and let their derivatives be bounded on
(a, b). Then the set E := {(x(t), y (t)) ∈ R2 : t ∈ [a, b]} is of
two-dimensional content zero. We can use the mean value
theorem to prove this as follows.
Prof. Madhusudan Manjunath, IIT Bombay Calculus: Lecture 28
For n ∈ N, consider the partition of [a, b] into n equal parts,
and let si be the mid-point of the ith subinterval for
i = 1, . . . , n. Let t ∈ [a, b]. Then there is i ∈ {1, . . . , n} such
that |t − si | ≤ (b − a)/2n. By the Mean Value Theorem,
there are cx and cy between t and si such that

x(t)−x(si ) = x 0 (cx )(t −si ) and y (t)−y (si ) = x 0 (cy )(t −si ).

Let α ∈ R with |x 0 |, |y 0 | ≤ α on (a, b), and β := b − a. Then

|x(t) − x(si )|, |y (t) − y (si )| ≤ αβ/2n, and so (x(t), y (t)) lies
in a square Si centered at (x(si ), y (si )) and of area α2 β 2 /n2 .
Thus E is contained in the union of the n squares S1 , . . . , Sn ,
the sum of whose areas is equal to α2 β 2 /n.
Given  > 0, choose n ∈ N such that n > α2 (b − a)2 /. Then
the set E is contained in the union of these n squares, the sum
of whose areas is less that . Hence E is of two-dimensional
content zero.
Prof. Madhusudan Manjunath, IIT Bombay Calculus: Lecture 28
Theorem
Let D be a bounded subset of R2 , and f : D → R be a
bounded function. If the boundary ∂D of D is of
(two-dimensional) content zero and if the set of discontinuities
of f in D is also of (two-dimensional) content zero, then f is
integrable over D.

We omit the proof.

The above theorem says that the two conditions
(i) ‘the boundary ∂D of D is of content zero’ and
(ii) ‘the set of discontinuities of f in D is of content zero’
together are sufficient to guarantee the integrability of f on D.
While neither condition can be dropped from the above
theorem, the integrability of f does not necessarily imply either
condition.
(Find examples to illustrate the above remarks.)
Prof. Madhusudan Manjunath, IIT Bombay Calculus: Lecture 28
Change of Variables in a Double Integral
RR
The calculation of the double integral D f (x, y )d(x, y ) can
often
RR be simplified by reducing it to another double integral
E
g (u, v )d(u, v ), where the pair (u, v ) of new variables is
related to the the pair (x, y ) of the given variables by a
suitable transformation.
In doing so, we look for a function g which is ‘simpler’ than
the given function f such that the domain E of g is also
‘simpler’ than the given domain D of f .
Let x ◦ , y ◦ , a1 , b1 , a2 , b2 ∈ R, and consider the affine
transformation Ψ : R2 → R2 given by Ψ := (ψ1 , ψ2 ), where

x = ψ1 (u, v ) := x ◦ +a1 u+b1 v and y = ψ2 (u, v ) := y ◦ +a2 u+b2 v .

Note that the above equations have a unique solution, that is,
the transformation Ψ is one-one ⇐⇒ a1 b2 − a2 b1 6= 0.
Prof. Madhusudan Manjunath, IIT Bombay Calculus: Lecture 28
In this case, the square region E := [0, 1]×[0, 1] is transformed
to the parallelogram D := Ψ (E ) with vertices Ψ (0, 0) = (x ◦ , y ◦ ),
Ψ (1, 0) = (x ◦ + a1 , y ◦ + a2 ), Ψ (0, 1) = (x ◦ + b1 , y ◦ + b2 ) and
Ψ (1, 1) = (x ◦ + a1 + b1 , y ◦ + a2 + b2 ). Note that Area (E ) = 1,
and it gets scaled to the area |a1 b2 − a2 b1 | of D = Ψ (E ).
In general, let E ⊂ R2 , and consider Φ := (φ1 , φ2 ) : E → R2 .
Let Q0 := (u0 , v0 ) be an interior point of E . If x := φ1 (u, v )
and y := φ2 (u, v ) for (u, v ) ∈ E , and the functions x and y
have partial derivatives at (u0 , v0 ), then the Jacobian of the
transformation Φ at (u0 , v0 ) is defined by
∂x ∂x
 

∂(x, y )  ∂u (Q0 ) ∂v
(Q0 ) 
J(Φ)(u0 , v0 ) := (u0 , v0 ) := det  .
 
∂(u, v )  ∂y ∂y 
(Q0 ) (Q0 )
∂u ∂v
In particular, J(Ψ )(u, v ) = a1 b2 − a2 b1 , (u, v ) ∈ R2 , for the
affine transformation Ψ .
Prof. Madhusudan Manjunath, IIT Bombay Calculus: Lecture 28
Now suppose E is a square region about Q0 , and the
transformation Φ := (φ1 , φ2 ) : E → R2 is such that the
functions φ1 and φ2 are differentiable at Q0 .
Let P0 := Φ(u0 , v0 ). For (u, v ) ∈ R2 , let
∂φ1  ∂φ1 
ψ1 (u, v ) := φ1 (u0 , v0 ) + (u − u0 ) + (v − v0 ),
∂u (u0 ,v0 ) ∂v (u0 ,v0 )
 
∂φ2  ∂φ2 
ψ2 (u, v ) := φ2 (u0 , v0 ) + (u − u0 ) + (v − v0 ).
∂u (u0 ,v0 ) ∂v (u0 ,v0 )
 

By the definition of differentiability of φ1 and φ2 at (u0 , v0 ),

φ1 (u, v ) = ψ1 (u, v )+1 (u, v ) and φ2 (u, v ) = ψ2 (u, v )+2 (u, v ),

for all (u, v ) ∈ E , where 1 (u, v )/k(u − u0 , v − v0 )k → 0 and

2 (u, v )/k(u − u0 , v − v0 )k → 0 as (u, v ) → (u0 , v0 ).
Thus the affine transformation Ψ := (ψ1 , ψ2 ) : R2 → R2 maps
Q0 to P0 , and approximates the transformation Φ around Q0 .
Prof. Madhusudan Manjunath, IIT Bombay Calculus: Lecture 28
Also, by the definition of Ψ = (ψ1 , ψ2 ), we obtain
 ∂ψ ∂ψ ∂ψ ∂ψ   ∂φ ∂φ ∂φ ∂φ 
1 2 1 2 1 2 1 2
J(Ψ )= − = − (Q0 )=J(Φ)(Q0 ).
∂u ∂v ∂v ∂u ∂u ∂v ∂v ∂u
Hence Area (Ψ (E )) = |J(Ψ )|Area (E ) = |J(Φ)(Q0 )|Area (E ).
v
E y Φ(E)
Φ
Q0
b

Ψ Ψ(E)
b
P0

u x

To state our next result, we need the following definition.

A subset Ω of R2 is called open if every point of Ω is an
interior point of Ω.
Prof. Madhusudan Manjunath, IIT Bombay Calculus: Lecture 28
Theorem (Change of Variables Formula)
Let D be a closed and bounded subset of R2 which has an
area, and let f : D → R be continuous. Suppose Ω is an open
subset of R2 and Φ : Ω → R2 is a one-one transformation
such that Φ := (φ1 , φ2 ), where φ1 and φ2 have continuous
partial derivatives in Ω and J(Φ)(u, v ) 6= 0 for all (u, v ) ∈ Ω.
Let E ⊂ Ω be such that Φ(E ) = D. Then E is a closed and
bounded subset of Ω, and E has an area. Moreover,
f ◦ Φ : E → R is continuous, and
ZZ ZZ
f (x, y )d(x, y ) = (f ◦ Φ)(u, v )|J(Φ)(u, v )|d(u, v ).
D E

Prof. Madhusudan Manjunath, IIT Bombay Calculus: Lecture 28

Compare the above result on Change of Variables for double
integrals with the result on Integration by Substitution for
Riemann integrals proved in Lecture 12:

(Integration by substitution)
Let f : [a, b] → R be continuous, and let φ : [α, β] → R be a
continuously differentiable function such that φ([α, β]) = [a, b].
If φ0 (t) 6= 0 for all t ∈ (α, β), then
Z b Z β
f (x)dx = f (φ(t))|φ0 (t)|dt.
a α

Prof. Madhusudan Manjunath, IIT Bombay Calculus: Lecture 28

Example
Let D := {(x, y ) ∈ R2 : x ≥ 0, y ≥ 0 and 1 ≤ 2(x + y ) ≤ 2},
and define f : D → R by f (x, y ) := y /(x + y ) for (x, y ) ∈ D.
v y

1 b b 1 b

E b
1/2 D

b b u b b x
1/2 1 1/2 1

Let u := x+y , v := y /(x+y ), that is, x := u(1−v ), y := uv .

We let Ω := {(u, v ) ∈ R
2 : u > 0}, and define Φ : Ω → R2 by
Φ(u, v ) = u(1 − v ), uv . Then the map Φ is one-one from Ω
onto {(x, y ) ∈ R2 : x + y > 0}. If Φ = (φ1 , φ2 ), then the
partial derivatives of φ1 and φ2 exist and are continuous.
Prof. Madhusudan Manjunath, IIT Bombay Calculus: Lecture 28
Also,
 
1−v −u
J(Φ)(u, v ) = det = u 6= 0 for all (u, v ) ∈ Ω.
v u

Further, if E := [1/2, 1] × [0, 1], then Φ(E ) = D and

(f ◦ Φ)(u, v ) = v for all (u, v ) ∈ E . Since f is continuous on
D, the Change of Variables formula gives
ZZ ZZ
f (x, y )d(x, y ) = f (u(1 − v ), uv )|u|d(u, v )
D ZZ E

= uv d(u, v )
E
Z 1  Z 1 
= u du v dv
1/2 0
3
= .
16
Prof. Madhusudan Manjunath, IIT Bombay Calculus: Lecture 28
Polar Transformation Let Φ : R2 → R2 be defined by
Φ(r , θ) := (r cos θ, r sin θ). If Φ := (φ1 , φ2 ), then φ1 and φ2
have continuous partial derivatives in R2 and
 
cos θ −r sin θ
J(Φ)(r , θ) = det = r for all (r , θ) ∈ R2 .
sin θ r cos θ
Thus the Jacobian of Φ is nonzero on {(r , θ) ∈ R2 : r 6= 0}.
Also, the function Φ : {(r , θ) ∈ R2 : r > 0 and − π < θ ≤ π}
→ {(x, y ) ∈ R2 : (x, y ) 6= (0, 0)} is one-one and onto.
In this case, the following result holds.
Proposition
Let D be a closed and bounded subset of R2 which has an
area, and let f : D → R be continuous. Suppose the set
E := {(r , θ) ∈ R2 : r ≥ 0, −π ≤ θ ≤ π, (r cos θ, r sin θ) ∈ D},
has an area. Then the function (r , θ) 7−→ f (r cos θ, r sin θ) is
continuous
RR on E and RR
D
f (x, y )d(x, y ) = E f (r cos θ, r sin θ) r d(r , θ).
Prof. Madhusudan Manjunath, IIT Bombay Calculus: Lecture 28
Examples
Let D := {(x, y ) ∈ R2 : x 2 + y 2 ≤ 1}. Define
E := {(r , θ) ∈ R2 : r ≥ 0, −π ≤ θ ≤ π and (r cos θ, r sin θ) ∈ D}.
Then E := [0, 1] × [−π, π].
p
(i) Let f (x, y ) := 1 − x 2 − y 2 for (x, y ) ∈ D. Then
ZZ ZZ
f (x, y )d(x, y ) = f (r cos θ, r sin θ) r d(r , θ)
D E
Z π Z 1 √ 
= 1 − r 2 r dr dθ
−π 0
Z π Z 1
√

1 2π
= s ds dθ = .
−π 2 0 3
2 2
(ii) Let f (x, y ) := e x +y for (x, y ) ∈ D. Then
Z π Z 1  Z π 
e −1
ZZ
r2
f (x, y )d(x, y )= e r dr dθ= dθ = π(e−1).
D −π 0 −π 2
Prof. Madhusudan Manjunath, IIT Bombay Calculus: Lecture 28