Let A ⊆ Rp and f be a function defined over A. To define R
Af, the integral of f over A, it is reasonable to require that
(a) the content (volume) of A is measurable, e.g. if A = Yp j=1
[aj, bj] is a closed cell in Rp, one can
define its content to be c(A) = Yp j=1
(bj − aj).
(b) the function f is summable, i.e. integrable, over A.
In the first part of this handout, we shall discuss how to define the integral of a function on cells in Rp. In the second part of the handout, we shall extend the definition to a function on more general (measurable) sets in Rp.
Part 1: Integrable Functions on Cells:
Definitions:
(a) K is called a cell in Rp (or a p−cell, or a p−dimensional rectangle) if K = I1× · · · × Ip, where Ij = [aj, bj] ⊂ R for j = 1, . . . , p.
(b) The (p−)content c(K) of K is defined to be c(K) = (b1− a1) × · · · × (bp− ap) = Yp j=1
(bj− aj).
(c) A set Z ⊂ Rp has p−content zero if ∀ ² > 0, ∃ a finite set C = {Kj}mj=1 of p−cells such that (a) Z ⊂
[m j=1
Kj,
(b) Xm
j=1
c(Kj) < ².
Remark 1. Note that the definition implies that if K is a cell (not necessarily closed) in Rp, then the boundary ∂K of K is a set of p−content zero.
Remark 2. Note that the definition of the content for a cell is well defined since it is easy to see that the following properties are satisfied.
(a) Let K be a cell in Rp) and K is a finite disjoint union of cells in Rp), i.e. K = [l i=1
Ki, then
c(K) = Xl
i=1
c(Ki).
(b) Let K1, K2 be cells in Rp). Then c(K1 ∪ K2) = c(K1\ K2) + c(K1∩ K2) + c(K2\ K1).
(c) Let x ∈ Rp, K be a cell in Rp) and x + K = {x + z | z ∈ K}. Then x + K is a cell in Rp) with c(x + K) = c(K), i.e. the definition of content for cells is invariant under translations.
Remark 3. By taking ²/2 > 0, if it is necessary, one may also assume that Z ⊂ Int¡[m
j=1
Kj¢ . Example (1). Let Z = {xj ∈ R | lim
j→∞xj = x}, a (0-dim’l) subset of R. Then (1 − dim’l) c(Z) = 0 since ∀² > 0, ∃ a 1-d cell Kxsuch that x ∈ Int(Kx), c(Kx) < ²/2, and xj ∈ Kx∀j ≥ L. For each j = 1, . . . , L − 1, let Kj be a 1-d cell such that xj ∈ Kj, and c(Kj) ≤ ²/(2L).
Example (2). Let Z = Q ∩ [0, 1], a (0-dim’l) subset of R. Then (1-dim’l) c(Z) 6= 0 since any finite
collection C = {K1, . . . , Km} of 1−dimensional cells that satisfies (a) will have Xm
j=1
c(Kj) ≥ 1.
Example (3). Let Z = {(x, y) | |x| + |y| = 1}, a (1-dim’l) subset of R2. Then the (2-d) content c(Z) = 0.
Definition (4). A collection of sets C = {Kj}mj=1 in Rn is called a partition of a set K in Rn if (a)
[m i=1
Ki = K, and
(b) IntKi∩ IntKj = ∅ holds for each 1 ≤ i 6= j ≤ m.
Example. Let K = [a1, b1] × · · · × [an, bn] = I1× · · · × In, and Pj = {[xji, xji+1] | aj = xj0 < · · · <
xjm(j) = bj}, for j = 1, . . . , n, be a partition of Ij into m(j) (a finite number of) closed cells in R.
Then the set P = {Qn
j=1[xjij, xjij+1] | 0 ≤ ij ≤ m(j) − 1} (induced by Pj’s) partitions K into m(1) × · · · × m(n) (finite number of) parallel closed n-cells.
Example. Let K = [0, 1]×[2, 4] ⊂ R2, and for each n = 1, 2, . . . , let Pn = {[i−1n ,ni]×[2+2(i−1)n , 2+2in] | 1 ≤ i ≤ n} be a partition of K that divides each side of K into n equal length subintervals. One can define the norm of a partition Pn to be the kPnk = max
Kj∈Pn
diam(Kj). In this example, the norm of the partition is kPnk = √n5.
Definition(5). Let P = {Ii}li=1 and Q = {Kj}mj=1 be partitions of (an n-cell) K. We say that P is a refinement of Q, denoted Q ⊂ P, if each cell in P is contained in some cell in Q, i.e. for each Ii ∈ P ∃ Kj ∈ Q such that Ii ⊂ Kj.
Note that if P , Q are partitions of K, then P ∩ Q is a (common) refinement of P and Q, and, in general, P ∪ Q is Not a partition,
Example. Let K = [0, 1]×[2, 4] ⊂ R2, and for each n = 1, 2, . . . , let Pn= {[i−1n ,ni]×[2+2(i−1)n , 2+2in] | 1 ≤ i ≤ n}. Then I = [0, 1/2] × [2, 3] ∈ P2 ⊂ P2 ∪ P3, J = [0, 1/3] × [2, 8/3] ∈ P3 ⊂ P2∪ P3, but (0, 1/2) × (2, 3) ∩ (0, 1/3) × (2, 8/3) 6= ∅. Therefore, P2∪ P3 is not a partition of K.
Definition (6). Let f be a bounded function defined on a closed n−cell K with values in R. A Riemann sum SP(f, K) corresponding to a partition P = {Kj}mj=1 of K is given by SP(f, K) =
Xm i=1
f (xi)c(Ki), where xi is any point in Ki, and c(Ki) denotes the (n-dim’l) content of Ki.
Remark. Note that Xm
i=1
mic(Ki) = LP(f ) ≤ SP(f ) = SP(f, K) ≤ UP(f ) = Xm
i=1
Mic(Ki), where mi = inf
Ki
f ≤ f (xi) ≤ Mi = sup
Ki
f, and LP(f ), and UP(f ) are called the lower sum and upper sum, respectively, of f with respect to the partition P of K.
Remark. (Monotonicity of lower and upper sums) If P, Q are partitions of K, and P ⊂ Q i.e.
Q is finer than P, then we have
LP(f ) ≤ LQ(f ) ≤ SQ(f ) ≤ UQ(f ) ≤ UP(f ).
Since the set {LP(f ) | P is a partition of K} is nonempty, and bounded from above by¡
supKf¢ c(K), the L(f, K) = supP LP(f ) = sup{LP(f ) | P is a partition of K} exists.
Analogously, the U(f, K) = infP UP(f ) = inf{UP(f ) | P is a partition of K} exists.
If {Pk} be any sequence of partitions of K such that Pj ⊂ Pj+1 for each j = 1, 2 . . . , and kPjk ≥ kPj+1k → 0, then lim
kPjk→0LPj(f ) = L(f, K), and lim
kPjk→0UPj(f ) = U(f, K).
Definition (of integrability on cells). A bounded function f is called Riemann integrable on K if L(f, K) = U(f, K) and this common value, denoted
Z
K
f, is called the (Riemann) integral of f on K.
Remark. f is integrable on K if and only if there exists a unique number L such that for each partition P of K we have LP(f ) ≤ L ≤ UP(f )
Proof: (⇒) Since LP(f ) ≤ L(f, K) = U(f, K) ≤ UP(f ) holds for each partition P of K, by setting L = L(f, K), the inequality LP(f ) ≤ L ≤ UP(f ) holds for each partition P of K. Suppose that L1 is a number such that the inequality LP(f ) ≤ L1 ≤ UP(f ) holds for each partition P of K.
Then L1 is an upper (resp. a lower) bound of the set {LP(f ) | P is a partition of K} (resp.
{UP(f ) | P is a partition of K}) which implies that L = L(f, K) ≤ L1 (resp. L1 ≤ U(f, K) = L)).
Hence, L1 = L is the unique number such that the inequality LP(f ) ≤ L ≤ UP(f ) holds for each partition P of K.
(⇐) In order to show that L(f, K) = U(f, K), we show that L(f, K) = L, and U(f, K) = L. Since the inequality LP(f ) ≤ L ≤ UP(f ) holds for each partition P of K, we have L(f, K) = supPLP(f ) ≤ L ≤ infP UP(f ) = U(f, K). Suppose that L(f, K) < L (resp. L < U (f, K)), then ∃ ²0 > 0 such that L(f, K) < L − ²0 (resp. L + ²0 < U (f, K)). Thus, for each partition P of K, we have LP(f ) ≤ L(f, K) < L − ²0 < L ≤ UP(f ) (resp. LP(f ) ≤ L < L + ²0 < U(f, K) ≤ UP(f )) which contradicts to the uniqueness of L. This implies that L(f, K) = L (resp. U(f, K) = L), and L(f, K) = L = U(f, K), i.e. f is integrable on K.
Criterion of integrability: Let f be a bounded function defined on K. Then the following are equivalent.
(1) f is integrable on K, i.e. L(f, K) = U(f, K), with integral L =R
Kf = L(f, K)
(2) (Riemann Criterion for Integrability) ∀ ² > 0, ∃ partition P², of K, such that if P is a refinement of P², then |UP(f ) − LP(f )| < ².
(3) (Cauchy Criterion for Integrability) ∀ ² > 0, ∃ partition P², of K, such that if P and Q are any refinements of P², and SP(f, K) and SQ(f, K) are any corresponding Riemann sums, then |SP(f, K)−
SQ(f, K)| < ².
(4) ∀ ² > 0, ∃ partition P², of K, such that if P is any refinement of P², and SP(f, K) is any corresponding Riemann sum, then |SP(f, K) − L| < ².
Proof Since LP(f ) ≤ L(f, K) ≤ U(f, K) ≤ UP(f ), (by drawing a picture) one notes that
|U(f, K) − L(f, K)| ≤(∗) |UP(f ) − LP(f )| ≤(†) |UP(f ) − L| + |LP(f ) − L|.
(1) ⇒ (2) : Given ² > 0, since L(f, K) = U(f, K) (and by the definitions that L(f, K) being the smallest number that satisfies L(f, K) ≥ LP(f ), and U(f, K) being the largest number that satisfies U(f, K) ≤ UP(f ) for all P ), there exists a partition P² such that
L(f, K) − ²/2 < LP²(f ) ≤ L(f, K), and
L(f, K) = U(f, K) ≤ UP²(f ) < U (f, K) + ²/2 = L(f, K) + ²/2.
Thus, if P is any refinement of P², then
L(f, K) − ²/2 < LP²(f ) ≤ LP(f ) ≤ UP(f ) ≤ UP²(f ) < L(f, K) + ²/2.
Setting L = L(f, K) in the (second) inequality (†), we get that
|UP(f ) − LP(f )| ≤ |UP(f ) − L| + |LP(f ) − L| < ²/2 + ²/2 = ².
Thus, the conclusion of (2) holds.
(2) ⇒ (1) : For each ² > 0, since the (first) inequality (∗), and (2) hold, there exists a partition P² such that if P is any refinement of P², then
|U(f, K) − L(f, K)| ≤ |UP(f ) − LP(f )| < ².
Letting ² → 0, we get 0 ≤ lim
²→0|U(f, K) − L(f, K)| ≤ lim
²→0² = 0 ⇒ U(f, K) = L(f, K)
and f is integrable.
(2) ⇔ (3) : For any refinements P, Q of P², we have
LP²(f ) ≤ LP(f ) ≤ SP(f, K) ≤ UP(f ) ≤ UP²(f ) LP²(f ) ≤ LQ(f ) ≤ SQ(f, K) ≤ UQ(f ) ≤ UP²(f ) Thus, we have
|SP(f, K) − SQ(f, K)| ≤ |UP²(f ) − LP²(f )|
and (2) ⇒ (3).
Conversely, for any refinements P, Q of P², if |SP(f, K) − SQ(f, K)| < ²/2 then, since
|UP(f ) − LQ(f )| = sup |SP(f, K) − SQ(f, K)|,
where the supremum is taken on all possible Riemann sum SP(f, K) and SQ(f, K) corresponding to the given (fixed) partitions P and Q, respectively, we have
|UP(f ) − LQ(f )| ≤ sup |SP(f, K) − SQ(f, K)| ≤ ²/2 and (3) ⇒ (2).
(3) ⇔ (4) : Let {Qj} be a sequence of refinements of P² such that Qj ⊂ Qj+1 and lim
j→∞kQjk = 0.
then
|SP(f, K) − L| = lim
j→∞|SP(f, K) − SQj(f, K)| ≤ ² and (3) ⇒ (4). Conversely, since
|SP(f, K) − SQ(f, K)| ≤ |SP(f, K) − L| + |SQ(f, K) − L|
holds, we have (4) ⇒ (3).
Example (1). For a < b, let f (x) =
(a if x ∈ Q ∩ [0, 1], b if x ∈ [0, 1] \ Q.
Then f is not continuous at each x ∈ [a, b] and f is not integrable on [0, 1] since LP(f, [0, 1]) = a 6=
b = UP(f, [a, b]) for any partition P of [a, b].
Example (2). Let f (x) =
1
n if x = mn ∈ Q ∩ [0, 1], where m, n ∈ N = {1, 2, . . .} and gcd(m, n) = 1, 1 if x = 0,
0 if x ∈ [0, 1] \ Q.
Then f is integrable on [0, 1] and f is continuous at every irrational and discontinuous at every ra- tional.
Observations. (1) There are finitely many rational numbers pq ∈ [0, 1] such that q < n. In fact, for fixed q < n, the number of pq ∈ [0, 1] is at most q + 1, which is at most n. Moreover, there are less than n positive q such that q < n. Thus, the set An = {pq ∈ [0, 1] : q < n} contains no more than n2 element and note that if mn ∈ (0, 1) is in lowest terms (m and n have no common factors except one), then min{|pq − mn| : pq ∈ An} > n12.
(2) Fix 2 ≤ n ∈ N, and let ˆP = {pq ±n13 : q < n, q ∈ N, p = 0, 1, . . . , q − 1}. Since the set ˆP is finite, it yields a partition P =¡P ∩ [0, 1]ˆ ¢
∪ {1} of [0, 1]. Define a step function sn(x) =
(1 if ∃ q ∈ N, p ∈ {0, 1, . . . , q − 1} with q < n such thatpq − n13 < x < pq + n13, 0 if there do not exist such p and q.
Also, define fn(x) =
(f (x) if ∃ q ∈ N, p ∈ {0, 1, . . . , q − 1} with q < n such thatpq − n13 < x < pq +n13, 0 if there do not exist such p and q.
For each n ≥ 2, since An contains no more than n2 elements, there exist no more than n2 intervals (pq−n13,pq+n13) in the interval [0, 1]. Thus, we have 0 ≤ U(fn, [0, 1]) ≤ UP(fn, [0, 1]) ≤ UP(sn, [0, 1]) = R1
0 sn(x) ≤ n2¡2
n3
¢ = n2 for all n ≥ 2. By letting n go to infinity, we get 0 = U(f, [0, 1]) ≥ L(f, [0, 1]) = 0. This proves that f is integrable on [0, 1].
(3) Let x = mn be a rational number in lowest terms. Note that if y ∈ [0, 1] satisfying that
|y − x| < 4n12 = (2n)1 2 then either y ∈ [0, 1] \ Q or y = pq with q < 2n (since any pq ∈ A2n will have |x − pq| ≥ (2n)1 2). Assume f is continuous at x. Given ² = 2n1 , there exists δ > 0 such that if y ∈ [0, 1] and |y − x| < δ then |f (x) − f (y)| < ². Let d = min{δ,(2n)1 2} and note that if y ∈ [0, 1] and
|y − x| < d then |f (x) − f (y)| > 2n1 (regardless that y is rational or irrational). This contradicts our assumption.
(4) Let α ∈ [0, 1] \ Q. Given ² > 0, choose n ∈ N such that 1n < ². Since there are finitely many rational numbers pq ∈ [0, 1] such that q < n, the minimum distance δ between pq and α for q < n, i.e. δ = min{|α − pq| : q < n}, exists and it is positive since α is irrational. If |x − α| < δ, then either x ∈ [0, 1] \ Q, or x = pq with p and q having no common factors except one and q ≥ n, since
|x − α| < δ. Thus, we have |f (x) − f (α)| =
(0 < ² if x ∈ (α − δ, α + δ) \ Q,
|f (pq)| = 1q ≤ 1n < ² if x = pq ∈ (α − δ, α + δ).
Some basic properties of integrable functions on cells:
(1) Suppose that K, K1, K2 are closed n−cells such that K = K1∪ K2, and Int(K1) ∩ Int(K2) = ∅.
If f is integrable on K, then f is integrable on K1, and K2, and R
Kf =R
K1f +R
K2f.
Proof. Given ² > 0, since f is integrable, there is a partition P²of K such that if P is any refinement of P², then |UP(f, K) − LP(f, K)| < ². For i = 1, 2, let P²,i = P²∩ Ki, then P²,i is a partition of Ki, and P²,1∪ P²,2 is a refinement of P². Thus, we have
² > UP²,1∪P²,2(f, K1∪ K2) − LP²,1∪P²,2(f, K1∪ K2)
= UP²,1(f, K1) − LP²,1(f, K1) + UP²,2(f, K2) − LP²,2(f, K2)
≥ UP²,i(f, Ki) − LP²,i(f, Ki)
≥ 0 Thus, for each i = 1, 2,
² > UP²,i(f, Ki) − LP²,i(f, Ki) ≥ 0 and if Pi is any refinement of P²,i, then
² > UP²,i(f, Ki) − LP²,i(f, Ki) ≥ UPi(f, Ki) − LPi(f, Ki) ≥ 0.
This implies that f is Riemann integrable on Ki and Li =R
Kif exists, for i = 1, 2, and L = L(f, K) = sup{LP(f ) | P is any refinement of P²}
≤ sup{LP1(f ) + LP2(f ) | Piis any refinement of P²,ii = 1, 2}
≤ L1+ L2
≤ inf{UP1(f ) + UP2(f ) | Piis any refinement of P²,ii = 1, 2}
= inf{UP(f ) | P is any refinement of P²,1∪ P²,2}
= U(f, K) = L Thus, we have L = L1+ L2.
(2) If f and g are integrable on K, then, for any c ∈ R, cf + g is integrable on K, andR
K(cf + g) = cR
Kf +R
Kg.
Proof. Given ² > 0, since f, g are integrable on K, there exists a partition P² of K such that if P is any refinement of P² then
|UP(f ) − LP(f )| < ²/2(1 + |c|)
and
|UP(g) − LP(g)| < ²/2.
Thus, we have
|UP(cf + g) − LP(cf + g)| ≤ |c||UP(f ) − LP(f )| + |UP(g) − LP(g)| < |c|²/2(1 + |c|) + ²/2 < ² which implies that cf + g is integrable on K. Let {Pj} be a sequence of partitions of K satisfying that Pj ⊂ Pj+1 for all j = 1, 2, . . ., and lim
j→∞kPjk = 0. Since Z
K
f = lim
j→∞SPj(f, K) and Z
K
g = lim
j→∞SPj(g, K), we have
c Z
K
f + Z
K
g = c lim
j→∞SPj(f, K) + lim
j→∞SPj(g, K)
= lim
j→∞SPj(cf, K) + lim
j→∞SPj(g, K)
= lim
j→∞SPj(cf + g, K)
= Z
K
(cf + g).
(3) Suppose that f and g are integrable on K. If f (x) ≤ g(x) for each x ∈ K, then R
Kf ≤R
Kg.
Proof. Since −f + g ≥ 0 on K and, by (2), it is integrable on K, we have 0 ≤ LP(−f + g) ≤
Z
K
(−f + g) = − Z
K
f + Z
K
g, where P is any partition of K. Since R
Kf ∈ R, by adding R
Kf on both sides of the inequality, we get R
Kf ≤R
Kg.
(4) If f is integrable on K, then |f | is integrable on K, and |R
Kf | ≤ R
K|f |.
Proof. Given ² > 0, since f is integrable on K, there exists a partition P² of K such that if P = {Kj}mj=1 is any refinement of P² then
|UP(|f |)−LP(|f |)| = | Xm
i=1
¡sup
Ki
|f |−inf
Ki
|f |¢
c(Ki)| ≤ | Xm
i=1
¡sup
Ki
f −inf
Ki
f¢
c(Ki)| = |UP(f )−LP(f )| < ².
Thus, |f | is integrable (by Riemann’s Criterion for integrability).
Since ±f, |f | are integrable, and ±f ≤ |f | on K, we have ±R
Kf ≤ R
K|f | ⇒ |R
Kf | ≤R
K|f |.
Examples of integrability.
(1) Let Z ⊂ Rnhave (n-)content zero, and f be a bounded function defined on Z. Then f is integrable on Z, andR
Zf = 0.
Proof. For each ² > 0, since Z has content zero, there exists a collection of cells {Ki}mi=1 such that Z ⊂ ∪mi=1Ki = K, and Pm
i=1c(Ki) < ²/2¡
supZ|f | + 1¢
. Define
f (x) =¯
(f (x) if x ∈ Z, 0 if x ∈ K \ Z.
Then,
| ¯f |(x) =
(|f |(x) ≥ 0 if x ∈ Z,
0 if x ∈ K \ Z,
and note that
| sup
Kj∩Zf − inf
Kj∩Zf | = | sup
Kj∩Z
f − inf¯
Kj∩Z
f |¯
≤ | sup
Kj
f − inf¯
Kj
f |¯
≤ | sup
Kj
| ¯f | − inf
Kj
¡− | ¯f |¢
|
= | sup
Kj
| ¯f | −¡
− sup
Kj
| ¯f |¢
|
= 2 sup
Kj
| ¯f | = 2 sup
Kj∩Z
|f |
≤ 2M.
Thus, if P is any partition of K = ∪mi=1Ki, we have
|UP(f, Z) − LP(f, Z)| ≤ 2 sup
Z |f | Xm
i=1
c(Ki) < ² which implies that f is integrable on Z with R
Zf = 0.
(2) Let I be a closed interval in R, and f be a bounded and monotonic function defined on I = [a, b]
Then f is integrable on I.
Proof. Since f is bounded on I, L(f, I) = sup
P LP(f ) and U(f, I) = inf
P UP(f ) exist.
Let Pn be the partition that divides I into 2n equal length subintervals. Thus,
n→∞lim LPn(f ) = L(f, I) and lim
n→∞UPn(f ) = U(f, I).
Since
n→∞lim |UPn(f ) − LPn(f )| = lim
n→∞
¡f (b) − f (a)¢¡
b − a¢
/2n = 0, we get L(f, I) = U(f, I), i.e. f is integrable on I.
(3) Let K be a closed n−cell, and f be a continuous function on K. Then f is integrable on K.
Proof. Since f is continuous on (compact set) K, f is uniformly continuous on K.
Hence, for any given ² > 0 there exists δ > 0 such that
if x, y ∈ K and kx − yk < δ then |f (x) − f (y)| < ²/¡
c(K) + 1¢ . Let P² be a partition of K such that kP²k = max
Kj∈P²
diam(Kj) = max
Kj∈P²
sup{kx − yk : x, y ∈ Kj} < δ.
If P is any refinement of P² then |UP(f ) − LP(f )| < ²c(K)/¡
c(K) + 1¢
< ².
Therefore, f is integrable on K.
(4) Let K be a closed n−cell, and f be a bounded function defined on K. If there exists a (n-)content zero subset Z ⊂ K, such that f is continuous on K \ Z, i.e. f is continuous everywhere on K except at a content zero subset Z of K, then f is integrable on K.
Proof. For each ² > 0, since Z has content zero, there exists a collection of cells {Ii}li=1 such that Z ⊂ Int¡[l
i=1
Ii¢
, IntIi∩ IntIj = ∅, and Xl
i=1
c(Ii) < ²/4¡ sup
K
|f | + 1¢ .
Since K \ Int¡[l
i=1
Ii¢
is compact, f is uniformly continuous there.
Thus, for the given ² > 0, there exists a δ > 0 such that if x, y ∈ Kj \ Int¡[l
i=1
Ii
¢, then |f (x) − f (y)| < ²/2¡
c(K) + 1¢ .
Let P² = {Kj}mj=1 be a partition of K such that {Ii∩K}li=1⊂ P²and kP²k = max
1≤j≤m sup
x,y∈Kj
kx−yk < δ.
If P is any refinement of P² then, by using (1) and (3), we have
|UP(f ) − LP(f )| ≤ Xl
i=1
¡sup
Z∩Ii
f − inf
Z∩Ii
f¢
c(Ii) + |UP(f, K \ Int¡
∪li=1Ii
¢) − LP(f, K \ Int¡
∪li=1Ii
¢)|
≤ 2 sup
Z |f | Xl
i=1
c(Ii) + c(K)²/2¡
c(K) + 1¢
< ²
This implies that f is integrable on K.
Theorems: (1) Suppose f and g are integrable on a closed n−cell K, and f = g everywhere on K except at a content zero subset Z of K, thenR
Kf =R
Kg.
Proof. Since f, g are integrable on K and Z has content zero, f − g is integrable on K and it is continuous with value 0 on K \ Z. Given ² > 0, let {Ii}li=1 be a collection of cells such that
Z ⊂ Int¡[l
i=1
Ii
¢ and Xl
i=1
c(Ii) < ²/4¡ sup
K |f − g| + 1¢ .
Let P²= {Kj}mj=1 such that {Ii∩ K}li=1⊂ P². Thus, if P is any refinement of P² then
|UP(f − g) − LP(f − g)| ≤ |UP(f − g, K \ Int¡[l
i=1
Ii¢
) − LP(f − g, K \ Int¡[l
i=1
Ii¢ )|
+ Xl
i=1
¡sup
Z∩Ii
(f − g) − inf
Z∩Ii
(f − g)¢ c(Ii)
< ².
This implies that R
K(f − g) = 0. Since R
Kg ∈ R, we have R
Kf =R
Kg.
(2) Fundamental Theorem of Calculus Let f be integrable on [a, b]. For each x ∈ [a, b], let F (x) =R
a,x]f =Rx
a f (t)dt. Then F is continuous on [a, b]; moreover, F0(x) exists and equals f (x) at every x at which f is continuous.
Remark. For each x ∈ [a, b], the existence of F (x) is due to [a, x] ⊆ [a, b] and the existence of Rb
a f.
Existence ofRb
a f implies that for each ² > 0 ∃ a partition P² of [a, b] such that if P is any refinement of P², then |UP(f, [a, b]) − LP(f, [a, b])| < ².
Let
P²l = P²∩ [a, x] and P²r = P²∩ [x, b].
Then P²l∪ P²r is a refinement of P² and if Pl is any refinement of P²l, then
|UP²l(f, [a, x]) − LP²l(f, [a, x])| ≤ |UP²l(f, [a, x]) − LP²l(f, [a, x]) + UP²r(f, [x, b]) − LP²r(f, [x, b])|
= |UP²l∪P²r(f, [a, b]) − LP²l∪P²r(f, [a, b])|
< ².
Hence, f is integrable on [a, x], i.e. F (x) exists.
Proof of the Theorem. If x, y ∈ [a, b] ⇒ F (y) − F (x) =Ry
x f (t)dt.
Let c = sup{|f (t)| : t ∈ [a, b]}. (c exists since f is integrable on [a, b] ⇒ f is bounded on [a, b].) Then |F (y) − F (x)| ≤ |Ry
x |f (t)|dt| ≤ c|Ry
x dt| = c|y − x|
⇒ F is (Lipschitz, hence) continuous on [a, b].
Suppose that f is continuous at x; thus ∀ ² > 0, ∃ δ > 0 such that if t ∈ [a, b] and |t − x| < δ then |f (t) − f (x)| < ².
Since f (x) = f (x)y−x1 Ry
x dt = y−x1 Ry
x f (x)dt.
Hence, if y ∈ [a, b] and |y − x| < δ ⇒ t ∈ [a, b] and |t − x| < δ for all t between y and x.
⇒ |f (t) − f (x)| < ² and this implies that
⇒ |F (y) − F (x)
y − x − f (x)| ≤ 1
|y − x||Ry
x |f (t) − f (x)|dt| ≤ 1
|y − x||Ry
x ²dt| = ².
⇒ lim
y→x
F (y) − F (x)
y − x = f (x), i.e. F0(x) exists and equals f (x) at every x at which f is continuous.
(3) Let F be a continuous function on [a, b] that is differentiable except at finitely many points in [a, b], and let f be a function on [a, b] that agrees with F0 at all points where F0 is defined. If f is integrable on [a, b], then Rb
a f (t)dt = F (b) − F (a).
Proof. Let {z1, . . . , zm} ⊂ [a, b] be the set at which F0 does not exist, i.e. F is not differentiable at zi, i = 1, . . . , m.
Let P = {a = x0 < x1 < · · · < xn = b} be a partition of [a, b] with zi, i = 1, . . . , m, being partition point, i.e. each zi ∈ {x0, x1, . . . , xn}.
⇒ F is continuous on each [xj−1, xj], j = 1, . . . , n, and differentiable on each (xj−1, xj).
By the Mean Value Theorem, F (xj) − F (xj−1) = F0(tj) (xj − xj−1) = f (tj) (xj − xj−1) for some tj ∈ (xj−1, xj) and for each j = 1, . . . , n.
Thus, we have F (b) − F (a) = Xn
j=1
F (xj) − F (xj−1) = Xn
j=1
f (tj) (xj − xj−1)
⇒ LP(f, [a, b]) ≤ F (b) − F (a) ≤ UP(f, [a, b])
⇒ sup
P LP(f, [a, b]) ≤ F (b) − F (a) ≤ inf
P UP(f, [a, b]) If f is integrable then Rb
a f (t)dt = supP LP(f, [a, b]) = infP UP(f, [a, b]) = F (b) − F (a).
Part 2: Integrable Functions on General Measurable Sets:
In the following we shall extend the concept of content of a cell in Rn to more general mea- surable subsets of Rn and extend the definition of integrability of a function to general subsets of Rn.
Definition (of integrability on general Euclidaen bounded subsets.) Let A ⊂ Rn be a bounded set and let f : A → R be a bounded function. Let K be a closed cell containing A and define fK : K → R by
fK(x) =
(f (x) if x ∈ A, 0 if x ∈ K \ A.
We say that f is integrable on A if fK is integrable on K, and define R
Af =R
KfK.
Remark. If A = K is a closed cell in Rn, then, since fK = f on K, it is obvious the integrability of f on A agrees with the integrability of fK on K and R
Af is defined to be R
KfK.
Remark. Let I be any closed cell containing A. Then K ∩ I is a closed cell containing A ⇒ fK = fK∩I = fI everywhere in K ∩ I, fK = 0 on K \ (K ∩ I), and fI = 0 on I \ (K ∩ I). Hence, R
KfK =R
K∩IfK∩I =R
IfI ⇒ the definition (of integrability of f ) only depends on f and A (and it is independent of the choice of K ⊇ A).
Basic properties of integrable functions on general sets:
(1). Let f and g be integrable functions defined on a bounded set A ⊂ Rn and let α, β ∈ R. Then the function αf + βg is integrable on A and R
A(αf + βg) = αR
Af + βR
Ag.
Proof. For any partition P of a cell K ⊇ A, since SP(αfK+ βgK, K) = αSP(fK, K) + βSP(gK, K) when the same intermediate points xj are used, the function αf + βg is integrable on A. Thus, by choosing the intermediate points from A whenever it is possible, we obtain that SP(αf + βg, A) = αSP(f, A) + βSP(g, A) which implies thatR
A(αf + βg) = αR
Af + βR
Ag.
(2) Let A1 and A2 be bounded sets with no pints in common, and let f be a bounded function. If f is integrable on A1 and on A2, then f is integrable on A1∪ A2 and R
A1∪A2f = R
A1f +R
A2f.
Proof. Let K be a closed cell containing both A1and A2, and let fK(x) = (
f (x) if x ∈ A1 ∪ A2, 0 if x ∈ K \ (A1 ∪ A2) and fKi (x) =
(f (x) if x ∈ Ai,
0 if x ∈ K \ Ai for i = 1, 2. Since f is integrable on Ai, i = 1, 2, fKi is integrable on K and, since fK = fK1 + fK2, and f is integrable on A1∪ A2. Also, for any partition P of K, note that SP(fK, K) = SP(fK1, K) + SP(fK2, K) when the same intermediate points xj are used. Thus, we haveR
A1∪A2f = R
A1f +R
A2f.
(3) If f : A → R is integrable on (bounded set) A and f (x) ≥ 0 for x ∈ A, then R
Af ≥ 0.
Proof. For any closed cell K ⊇ A and any partition P of K, note that SP(fK, K) ≥ 0 for any Riemann sum. Thus, R
Af ≥ 0.
Remark. This implies that if f and g are integrable on A and f (x) ≤ g(x) for x ∈ A, then (a) R
Af ≤R
Ag, and (b) |f | is integable on A, and |R
Af | ≤ R
A|f |.
(4) Let f : A → R be a bounded function and suppose that A has content zero. Then f is integrable on A andR
Af = 0.
Proof. LetK ⊇ A be a closed cell. If ² > 0 is given, let P² be a partition of K such that those cells in P² which contain points of A have total content less than ². Now if P is a refinement of P², then those cells in P containing points of A will also have total content less than ². Hence if |f (x)| ≤ M for x ∈ A, we have |SP(fK, K) − 0| ≤ M² for any Riemann sum corresponding to P. Since ² is arbitrary, this implies that R
Af = 0.
(5) Let f, g : A → R be bounded functions and suppose that f is integrable on (bounded set) A. Let Z ⊆ A have content zero and suppose that f (x) = g(x) for all x ∈ A \ Z. Then g is integrable on A and R
Af =R
Ag.
Proof. Extend f and g to functions fK, gK defined on a closed cell K ⊇ A. Thus, the function hK = fK− gK is bounded and equals 0 except on Z. Hence, hK is integrable on K and R
KhK = 0.
Since fK is also integrable on K, we have R
Af =R
KfK =R
K(fK− hK) =R
KgK =R
Ag.
(6) Let U be a connected, open subset of Rn and let f : U ⊂ Rn→ Rn+1 be a C1 map on U. If K is any convex, compact subset of U, then f (K) has measure (or content) zero.
Definition. Let A ⊂ Rn be a bounded set. The characteristic function of A is the function χA defined by χA(x) =
(1 if x ∈ A,
0 otherwise. Now, suppose A is a bounded subset of Rnand f is a bounded function on Rn. Let K be a closed cell that contains A. We say that f is integrable on A if f χA is integrable on K, and define R
Af =R
Kf χA. (Note that f χA= 0 on K \ A, so it is independent of the choice of K ⊇ A.)
Question: Let f ≡ 1 on A ⊂ Rn. What does it mean when we say that f is integrable on A?
Definition. A set A ⊂ Rn is said to have content (or it is said to be (Jordan) measurable) if it is bounded and its boundary ∂A has content zero. Let D(Rn) (= {A ⊂ Rn | A has content } = {A ⊂ Rn| A is measurable }) denote the set of all measurable subsets of Rn.
Remark. If A ∈ D(Rn) and if K is a closed cell containing A, then the function gK defined by gK(x) =
(
1 if x ∈ A
0 if x ∈ K \ A is continuous on K except possibly at points of ∂A (which has content zero). Thus, gK is integrable on K and we define the content of A, denoted c(A), by c(A) =R
KgK =R
A1.
Remark. ∀ ² > 0, since c(A) =R
KgK, ∃ a partition P² = {Ij}mj=1of K such that |SP²(gk, K)−c(A)| <
² for any Riemann sum SP²(gk, K). By choosing the intermediate points in SP²(gk, K) to belong to A whenever possible, we have Pm
j=1c(Ij) + ² ≥ SP²(gk, K) + ² > c(A) > SP²(gk, K) − ², where the first inequality holds since A ⊂ ∪mj=1Ij.
Thus, we have: A set A ⊂ Rn has content zero if and only if A has content and R
A1 = c(A) = 0.