You are on page 1of 14

Chapter 6

Differentiation and
Integration

1

Z) and the Riemann-Stieltjes integral is the usual Riemann integral. b] → R. zn ). . for f with respect to g using P and Z. .7. If P is a partition of [a. then some work will show that −1 f dg = f (0). P = {a = x0 < x1 < · · · < xn = b}. Given a partition of [a. or briefly. b].3. . then we call an n-tuple of real numbers Z = (z1 . Z) − L| < ε. . In this case. P. z2 . .1. Indeed.7. j=1 We say f is Riemann-Stieltjes integrable with respect to g. say P = {a = x0 < x1 < · · · < xn = b}.6. P. The essential change is to weight the intervals according to a new function. First. there is a partition Pε so that. The partition Pε in the definition differs from the Riemann integral condition: we are not asking for a number δ so that for all partitions P with mesh(P ) < δ. P. notice that for g(x) = x. as the next example shows. called the integrator. 6.8). we introduce a notational convenience and then the crucial definition. as Ig (f. for a Riemann-Stieltjes integrable function. The price to be paid for this increased generality is some careful work with partitions. This weighting gives integrals that combine aspects of both discrete summations and the usual Riemann integral. for all partitions P with P ⊇ Pε and all evaluation sequences Z for P . g : [a. we have |Ig (f. if g(x) is the Heaviside step function and f is R1 continuous at 0. . z2 . n. . Z) − L| < ε. . Z) = n X f (zj )[g(xj ) − g(xj−1 )]. the R-S sum. we have |Ig (f. Definition. b]. we compare this definition to the corresponding condition for the Riemann integral (Condition (4) in Theorem 6. and write a f dg for L. and an evaluation sequence. First off. .7 Riemann-Stieltjes Integration 1 6. for P . Second. 6. if. .2. Notice that zi can equal zi+1 . there is a number L so that for all ε > 0. the Riemann-Stieltjes sum is the usual Riemann sum I(f. we say L is the Riemann-Stieltjes integral of f with respect Rb to g. . there may not be such a δ. Consider bounded functions f. xi ] for i = 1. Z = (z1 . denoted f ∈ R(g). Before developing the general properties of this integral. . Definition. zn ) an evaluation sequence for P if zi ∈ [xi−1 . Riemann-Stieltjes Integration Riemann-Stieltjes integration is a generalization of Riemann integration. .7. P. we define the Riemann-Stieltjes sum.

Since f and g are both constant on [0. for any choice of evaluation sequence Z = (z1 . The key point is that no matter how we choose P .7. 6. . and any evaluation sequence R1 Z for P . no matter how small the mesh size of Q.3. (1) Define f on [0. X). then Ig (f. If f1 ∈ R(g) and f2 ∈ R(g) on [a. xj ] contained in either of these two intervals. Examples. So if zk ≤ 1/2. then if we had a suitable partition Pε . Z) = 1. . 1] containing 1/2. Clearly there is some k ∈ {1. if we choose ε = 1/2. . However. consider any partition P = {a = x0 < x1 < · · · < xn = b} and evaluation sequence Z = (z1 . 1] by f (x) = 0 for x ∈ [0. Thus. Theorem. . For any partition P of [0. To see this. . Since there is no L that satisfies both inequalities. a . f is not Riemann-Stieltjes integrable with respect to g. . then Ig (f. which is f (zk ). . 1]. although we don’t need this). we cannot control the value of the Riemann-Stieltjes sum in terms of the mesh size. Z) = 0 and if zk > 1/2. Then f is not Riemann-Stieltjes integrable with respect to g. n} so that xk−1 ≤ 1/2 < xk . Define g to be the same as f . We leave it as an exercise to show that this “δ-condition” implies R-S integrability. a bit of work shows Ig (f. X) could be either value. save only that g(1/2) = 0.2 Differentiation and Integration Some textbooks define Riemann-Stieltjes integrability using this more restrictive condition. 6. 1/2) and f (x) = 1 for x ∈ [1/2. However. for any number L. zn ). .4. we would have |1 − L| < 1/2 and |0 − L| < 1/2. f (x0j )[g(xj )−g(xj−1 )] = 0 for any j with [xj−1 . b]. we can always choose zj to be either less than 1/2 or at least 1/2 and so Ig (f. Thus. X) = f (zj ). 0 f dg exists (and equals 1. both of these possibilities can occur. Q. where j satisfies xj−1 < 1/2 < xj . . . (2) Change the above example by making f the same as g. zn ) for P . P. b] and Z b Z c1 f1 + c2 f2 dg = c1 a b Z f1 dg + c2 a b f2 dg. for any δ > 0. Thus. and c1 . P. P. Then there is only one nonzero term in Ig (f. then c1 f1 + c2 f2 ∈ R(g) on [a. Z) = 1. . there is always a partition Q with mesh(Q) < δ so that 1/2 ∈ / Q. 1].7. we have Ig (f. 1/2) and (1/2. Thus. . We collect together some basic facts about Riemann-Stieltjes integrals. P. Q. c2 ∈ R.

c] and Z c Z b Z c f dg = f dg + f dg. . . c] then it is R-S integrable with respect to g on [a. We prove the first result and leave the second and third as exercises. . d2 ∈ R. if a < b < c and f is R-S integrable with respect to g on both [a. Since f1 ∈ R(g). for all partitions P with P ⊇ P1 and all evaluation sequences X for P . .3 6. a a a Finally. X) = n X (c1 f1 (x0j ) + c2 f2 (x0j ))(g(xj ) − g(xj−1 )) j=1 = c1 n X f1 (x0j )(g(xj ) − g(xj−1 )) j=1 + c2 n X f2 (x0j )(g(xj ) − g(xj−1 )) j=1 = c1 Ig (f1 . then f ∈ R(d1 g1 + d2 g2 ) on [a. we have . there is P1 so that. P. We have Ig (c1 f1 + c2 f2 . Pick η > 0 so that |c1 |η + |c2 |η < ε. P. X) + c2 Ig (f2 . a a b Proof. b]. and d1 . P. b] and Z b Z b Z b f d(d1 g1 + d2 g2 ) = d1 f dg1 + d2 f dg2 . Consider any partition P = {a = x0 < x1 < · · · < xn = b} and evaluation sequence X = (x01 . X) Let ε > 0.7 Riemann-Stieltjes Integration If f ∈ R(g1 ) and f ∈ R(g2 ) on [a. x0n ) for P . b] and [b.

.

Z b .

.

.

P.Ig (f1 . X) − f1 dg .

.

. < η.

there is P2 so that. a Similarly. for all partitions P with P ⊇ P1 and all evaluation sequences X for P . we have .

.

Z b .

.

.

X) − f2 dg .Ig (f2 . P.

.

. < η.

a Thus. and for any evaluation sequence X for P . we have . for any partition P with P ⊇ P1 ∪ P2 .

 Z b .

Z b .

.

.

Ig (c1 f1 + c2 f2 . X) − c1 f1 dg + c2 f2 dg . P.

.

.

a a .

.

.

.

Z b Z b .

.

.

.

.

.

.

≤ |c1 | .

Ig (f1 . P. X) − f1 dg .

+ |c2 | .

Ig (f2 . P. X) − f2 dg .

.

a a Since P ⊇ P1 and P ⊇ P2 . it follows from the choices of P1 . . that this last expression is less than ε. P2 .  . and η.

then F ∈ R(G) on [c. . d] and evaluation sequence X = (x01 . . continuous function onto [a. d] and d Z Z F dG = b f dg. IG (F. P. Suppose f. . b]. b]. there is a partition Q so that for all partitions Q of [a. ϕ(P ). b] is a strictly increasing. Theorem. Since ϕ is strictly increasing and onto. we have . j=1 Let  > 0. ϕ(c) = a. d] → [a.5. and ϕ(d) = b. it is one-to-one. . c a Proof. Let F = f ◦ ϕ and G = g ◦ ϕ. x0n ) for P . Consider any partition P = {c = x0 < x1 < · · · < xn = d} of [c. b] → R are bounded and ϕ : [c. As f ∈ R(g). ϕ(x0n )) is an evaluation sequence for ϕ(P ). b] and ϕ(X) = (ϕ(x01 ).6) on page 167) and the integration by parts formula (see the equation at the bottom of page 166). ϕ(X)). ϕ(P ) is a partition of [a. .4. If f ∈ R(g) on [a. . g : [a. . 6. Thus. X) = n X f (ϕ(x0j ))[g(ϕ(xj )) − g(ϕ(xj−1 ))] = Ig (f. b] with Q ⊇ Q and all evaluations sequences Y for Q.7.4 Differentiation and Integration The following two theorems are the analogues of the substitution formula (see Equation (6. Moreover. .

.

Z b .

.

.

Ig (f. Y ) − f dg . Q.

.

. < ε.

If P is a partition of [c. Using the above two displayed equations. for all partitions P with P ⊇ Pε and all evaluation sequences X for P . we have. d] with P ⊇ Pε . . then ϕ(P ) ⊇ Qε . a Let Pε = {ϕ−1 (x) : x ∈ Qε }.

.

Z b .

.

.

IG (F. P. X) − f dg .

.

< ε. .

If f ∈ R(g) on [a.  6. Y ) for a suitable choice of partition Q and evaluation sequence Y . X) and Ig (f. . then g ∈ R(f ) on [a. f dg + a Proof. a By the definition. First. F ∈ R(G) on [c. we need to relate If (g. Theorem. b] → R be bounded functions.6. Let f. P. b] and Z b Z a b g df = g(b)f (b) − g(a)f (a).7. b]. g : [a. Q. d] and Rb a F dG = Rb a f dg.

. we have . x0n ) for P . b] and evaluation sequence X = (x01 . P ∪ X. Then If (g. b] with P ⊇ P and all evaluation sequences X for P . x1 . xn−1 . Q is the evaluation sequence (x0 . we have (6. j=1 Subtracting the first equation from the second.7. . xn ) if x0j ∈ (xj−1 . this last pair of summations is Ig (f. X) = n X g(x0j )(f (xj ) − f (xj−1 )). xj ) for all j and has either an xj−1 or an xj omitted if either x0j = xj−1 or x0j = xj for some j. x1 . . j=1 Deleting any zero terms arising where x0j = xj−1 or x0j = xj . Q).7) − If (g.5 6. . there is a partition P so that for all partitions P of [a. x2 . P. x2 . P. Here. Let  > 0.7 Riemann-Stieltjes Integration Consider any partition P = {a = x0 < x1 < · · · < xn = b} of [a. j=1 We also have the telescoping sum f (b)g(b) − f (a)g(a) = n X f (xj )g(xj ) − j=1 n X f (xj−1 )g(xj−1 ). . xn−1 . As f ∈ R(g). X) + f (b)g(b) − f (a)g(a) = n X f (xj )(g(xj ) − g(x0j )) + j=1 n X f (xj−1 )(g(x0j ) − g(xj−1 )). . . .

.

Z b .

.

.

Ig (f. P. X) − f dg .

.

< ε. .

then P ∪ X ⊇ Pε and so we can conclude that . a The crucial point is that if P ⊇ Pε .

.

Z b .

.

.

Q) − f dg . P ∪ X.Ig (f.

.

< ε .

7. we can conclude that .7). a where Q is constructed as in the last paragraph. Applying (6.

 Z b .

.

.

.

P. X) − − f dg + f (b)g(b) − f (a)g(a) .If (g.

.

. < ε.

the most interesting integrators will be the discontinuous ones. for a smooth function g. b] and that f (a)g(a). a This shows that g ∈ R(f ) on [a. the Riemann-Stieltjes integral with respect to g is really just a Riemann integral.  Rb a g df = − Rb a f dg + f (b)g(b) − Finally. we show that. . In fact.

Let  > 0. it suffices to find P so that for all partitions P with P ⊇ P and evaluation sequences X for P . then f g 0 is Riemann integrable on [a.8. where g : [a. If f : [a. Theorem.I below. b] → R is C 1 .6 Differentiation and Integration 6.7. b] and Z b Z b f dg = f (x)g 0 (x) dx. a a Proof. By Exercise 6.7. b] → R has f ∈ R(g).

.

Z b .

.

.

I(f g 0 . X) − f dg . P.

.

< . .

P. then  |g 0 (x) − g 0 (y)| ≤ . j=1 Comparing these two summations cries out for the application of the Mean Value Theorem to g on each interval [xj−1 . there M so that |f (x)| ≤ M for all x ∈ [a. X) = n X j=1 n X f (x0j )g 0 (x0j )[xj − xj−1 ]. Doing this. b]. . d] and evaluation sequence X = (x01 . f (x0j )[g(xj ) − g(xj−1 )]. . there is δ1 > 0 so that. P. X)| ≤ ≤ n X j=1 n X j=1 |f (x0j )[g 0 (x0j ) − g 0 (tj )][xj − xj−1 ]| M ε ε [xj − xj−1 ] = . 2M (b − a) 2 Since f ∈ R(g). for any partition P with P ⊇ P2 and any evaluation sequence X for P . a Consider any partition P = {a = x0 < x1 < · · · < xn = b} of [c. P. X) = Ig (f. X) − Ig (f. P. xj ] to obtain tj so that g(xj ) − g(xj−1 ) = g 0 (tj )[xj − xj−1 ]. X) = n X f (x0j )[g 0 (x0j ) − g(tj )][xj − xj−1 ]. there is P2 so that. y ∈ [a. . if |x − y| < δ1 . P. x0n ) for P . X) − Ig (f. pick a partition P1 with mesh(P1 ) < δ. For any partition P with P ⊇ P1 and any evaluation sequence X for P . P. . Then I(f g 0 . As g 0 is uniformly continuous. b]. we have 0 |I(f g . we have . j=1 Let ε > 0. we have I(f g 0 . Since f is bounded. for all x. 2M (b − a) First.

.

Z b .

.

ε .

X) − f dg . P.Ig (f.

.

. < .

2 a .

7 6. we have . for any partition P with P ⊇ P1 ∪ P2 and for any evaluation sequence X.7 Riemann-Stieltjes Integration Thus.

.

Z b .

.

0 .

P. X) − f dg .I(f g .

.

. < ε.

P )[g(xi ) − g(xi−1 )] i=1 If we assume that g is increasing. P ) = xj−1 ≤x≤xj inf xj−1 ≤x≤xj f (x). P ) = n X i=1 n X Mi (f. R) ≤ Ug (f. (6. Assume that g is increasing.7. R) ≤ Ug (f. P ) = sup f (x) and mj (f. a By the exercise mentioned at the start of the proof.10) Lg (f. g be bounded functions on [a. P ). then it is straightforward to verify that for any evaluation sequence X.  We’ve avoided upper sums and lower sums so far. we’re done. P. b]. let P = {x0 < x1 < · · · < xn } be a partition of [a.7. If the integrator is increasing. then Lg (f. P ) ≤ Ig (f. b] and recall that Mj (f. Only if g(xj ) − g(xj−1 ) ≥ 0 are maximizing f (x) and maximizing f (x)[g(xj ) − g(xj−1 )] equivalent problems.7. The proof is exactly the same as the Refinement Lemma for the Riemann integral. Let f. P ) ≤ Lg (f. 1) and f (1) = 0 and if g(x) = x. then Ug (f. As usual. If R is a refinement of P . It is possible for the above inequalities to be strict. P ) > Ig (f. R partitions of [a. Let f and g be bounded functions on [a. Refinement Lemma. if f (x) = x on [0. P ) = Lg (f. Define the upper sum with respect to g and the lower sum with respect to g to be Ug (f. xj ] ∩ R = . X) ≤ Ug (f. As with the Riemann integral.11. Definition. This is not true if g is not increasing. P ). b]. P )[g(xi ) − g(xi−1 )] mi (f. we have 6. For example. The key points are that if xj−1 .9. b] and P. then upper and lower sums do have properties similar to upper and lower sums for Riemann integrals. X) for all choices of P and X. xj ∈ P and [xj−1 . as they do not behave well for general integrators. 6. P.

Pε ) + η/[n(g(b) − g(a))] and .7. 6. Suppose that Q is any partition with Q ⊇ P and X is any evaluation sequence for Q. which suffices to prove (3). tj ∈ [xj−1 . P ). Assume g(b) − g(a) > 0.8 Differentiation and Integration [tk . Lg (f. P ) ≤ mi (f. P ). . there is a partition P so that Ug (f. The argument that (3) ⇒ (2) is immediate. Pε ) = 0. P ) − Lg (f. Q) ≤ Ug (f. Suppose that Pε = {x0 < x1 . tl ]. g be bounded functions on [a. 6. In keeping with the notation for Riemann integration. Pε ) < 2ε.13. for any i between k + 1 and l. P2 ) > L−ε. P ) ≤ Ug (f. . X) ≤ Ug (f. 6. For each j. P ) ≤ Lg (f. . Definition. P ) and Ug (f ) = infP Ug (f. Let f. Let ε > 0. and (3) for each ε > 0. P ) < L + ε. Riemann-Stieltjes Condition. Suppose that (2) holds and let L = Lg (f ) = Ug (f ). (2) Ug (f ) = Lg (f ). Pε ) − Lg (f. Q) ≤ Ig (f. Proof.12. . b] and assume that g is increasing. If P and Q are any two partitions of [a.7. b]. Moreover. if (2) holds. Let ε > 0 and find a partition Pε as in the definition of Riemann-Stieltjes integrability. L − ε < Lg (f. P ) < L + ε. R Thus. we define Lg (f ) = supP Lg (f. Using (6. R).10) and the Refinement Lemma again.14. Q. By the Refinement Lemma. P ) < ε. suppose that (1) holds.7. b]. P1 ) < L+ε and Lg (f. P ) ≤ Ug (f. R) ≤ Mj (f. g be bounded functions on [a. then common value equals R f dg. choose points sj . (1) holds and L = f dg Next. The following are equivalent: (1) f is Riemann-Stieltjes integrable with respect to g. xj ] so that f (sj ) < mj (f. then g(xj ) − g(xj−1 ) = l X g(ti ) − g(ti−1 ) i=k+1 and. < xn }. Q). as otherwise Ug (f. mj (f. Pε ) = Lg (f. . Corollary. We can find two partitions P1 and P2 so that Ug (f.7. b] and assume that g is increasing on [a. L − ε < Lg (f. Mi (f. We claim that Ug (f. Let P = P1 ∪ P2 be their common refinement. Let η > 0. Let f. .

b]. . A first guess might be that if f is bounded by M on an interval Rb [a. The first step is to define this total change. For example. . fix a partition P and a point c ∈ / P . then a f dg would be bounded by M (g(b) − g(a)).7. Q). P ) = |f (xi ) − f (xi−1 )|. this proves the claim. Since η is arbitrary. it follows that U (f. b] and we need to replace g(b) − g(a) with the total change of g. Remarks. say [c. P ) < 2ε + 2η. Letting S = (s1 . . Pε . b]. tn ). T ) > U (f. d]. . for a function f : [a. i=1 6. Pε ) + n(g(b) − g(a))  [g(xj ) − g(xj−1 )] and using g(xj ) − g(xj−1 ) ≤ g(b) − g(a) (as g is increasing). we have n X η = L(f.7 Riemann-Stieltjes Integration 9 f (tj ) > Mj (f. suppose g is identically zero on [a. c]. c] → R. we then have Ig (f. . xj ].6. if c ∈ [xj−1 . sn ) and T = (t0 . then the variation of f over P is n X V (f. Second. Pε . P ) + V (f. b]. Given a function f : [a. While this is true if g is increasing on [a. In particular. Pε ) + η. then. Pε . Definition. Pε ) − η/[n(g(b) − g(a))]. Pε ) + = L(f. Since |Ig (f. Pε . the integral may not be bounded above by 1(g(b) − g(a)) = 0. P ∪ Q) = V (f. The problem is that g changes over the interval [a.16.7. 6. notice that if P is a partition of [a. If f is continuous at c and d. V (f. P ) − η. . b] → R and a partition of [a. it is not true in general. we have |f (xj ) − f (xj−1 )| ≤ |f (xj ) − f (c)| + |f (c) − f (xj−1 )|. where g is 1. b] and Q is a partition of [b. T ) − Ig (f. . then it is easy to show that Rb a f dg = f (c) − f (d). b] except for a subinterval in the interior. Ig (f. n j=1 Similarly. . say P = {a = x0 < x1 < · · · < xn = b}.15. First. Then. S)| < 2ε. S) = < n X f (sj )[g(xj ) − g(xj−1 )] j=1 n  X j=1 η mj (f. .  An important change from the Riemann integral lies in bounds for integrals. P ) − L(f.

P2 ) ≤ Vab f + Vbc f. Definition. then a telescoping sum shows that Vax f = f (x) − f (a). 6. Let P 0 = P ∪ {b}. 6. b] → R is increasing and x ∈ [a. π2n π(2n − 1) 3π 2π As g takes successive values of 0. . 0. xj ].19. P ) ≤ V (f. which diverges. Thus. Pn ) = 2n and V0 f = +∞. Define h : [0. V (f. 1/π We claim that V0 g = +∞. P 0 ) = V (f. More generally. Let P be a partition of [a. c] → R is given. Observe that k has a bounded derivative. Fix n > 1 and let Pn be the partition {0. . b] is Vab f = sup{V (f. . If a < b < c and f : [a. P ) ≤ V (f. P2 is a partition of [b. R). for each interval [xj−1 . P1 = P 0 ∩ [a. then Vac f = Vab f + Vbc f. some work will show that V (f. −1. we define k : [0.7. Using Pnthe partition P above. then V (f. c]. P1 ) + V (f. k(x) = 0 if x = 0. it follows that if R is a refinement of P . |k 0 (x)| = |2x sin(1/x)−cos(1/x)| ≤ 3. .17. Then P1 is a partition of [a. b]. 1. c]. b]. We say f is of bounded variation on [a. c]. The (total) variation of f on [a. Define g : [0. 0. 1/π |g(xj ) − g(xj−1 )| = 1. b]. 1/π] → R by ( x sin(1/x) if x > 0. b]. If a function f : [a. and so k is Lipschitz.10 Differentiation and Integration By induction. Examples. Finally. . and P2 = P 0 ∩ [b. 6. P ) : P a partition of [a. Even this h is not of bounded variation. b]}. Lemma. P ) is a multiple of k=1 1/k. = 1/π}. h(x) = 0 if x = 0.7. V (f. 1/π] → R by ( x2 sin(1/x) if x > 0. then Vab f ≤ C(b − a). . g(x) = 0 if x = 0. 2 2 2 2 . Thus.7. b] if Vab f is finite. 1/π] → R by ( sin(1/x) if x > 0.··· .18. Proof. . an easy estimate shows that if f is Lipschitz with constant C on [a. and P 0 = P1 ∪ P2 .

6. then there are increasing functions g. Define 1 [f (a) + f (x) + Vax f ] 2 1 h(x) = [f (a) − f (x) + Vax f ] 2 Clearly. if P = P1 ∪ P2 . If f : [a. this shows Vac f ≥ Vab f + Vbc f . g : [a. h : [a. If f : [a.7. P ) = |f (y) − f (x)|. There is a partition P1 of [a.20. Theorem. Thus. Proof. b] → R is of bounded variation and f ∈ R(g) on [a.7. Notice that if x < y. 6. let  > 0. we can now bound Riemann-Stieltjes integrals.  The key result about functions of bounded variation is the following characterization.  With these results in hand. In particular. y}. and a partition P2 of [b.21. P1 ) + V (f.7 Riemann-Stieltjes Integration 11 Taking the supremum over all partitions P . c] so that V (f. h is increasing. g(y) − g(x) = Vxy f + f (y) − f (x) ≥ 0 and so g(y) − g(x) ≥ 0. then Vxy f ≥ V (f. b]. b] → R is of bounded variation on [a. P ) = V (f. For the reverse inequality. then the remark above shows that Vac f ≥ V (f. P2 ) ≥ Vbc f − /2. 6. b] → R so that f = g − h. then . b] so that V (f. b] → R is bounded by M . P2 ) ≥ Vab f + Vbc f −  As  is arbitrary. P1 ) ≥ Vab f − /2. we have Vac f ≤ Vab f + Vbc f . Theorem. f (x) = g(x) − h(x). then g(x) = 1 [f (y) − f (x) + Vxy f ] 2 If P = {x. b]. Similarly.

Z b .

.

.

.

f dg .

.

. ≤ M · Vab g.

X)| ≤ n X M |g(xi ) − g(xi−1 )| ≤ M · Vab g.  . a Proof. For any partition P = {a = x0 < x1 < · · · < xn = b} and any evaluation sequence X for P . we have |Ig (f. i=1 Choosing a sequence of partitions Pn and evaluation sequences Xn so that Rb Ig (f. P. we have the result. Xn ) converges to a f dg. Pn .

R1 then −1 f dg does not exist. b]. Corollary. b] with mesh(P ) < δ. Hint: Show that for f continuous. As f is continuous and [a. and onto. By the previous theorem. d] → [a. f is uniformly continuous. Using Theorem 6. 1/π].  Exercises for Section 6. k(x) = Vax f is continuous.20. If P = {a = x0 < x1 < · · · < xn = b}.4. then f ∈ R(g) on [a. 1/π] → R by ( x sin(1/x) if x > 0. b] and |x − y| < δ implies  |f (x) − f (y)| ≤ 0 = . Write out carefully the argument above that h is not of bounded variation on [0. Then Mj (f. g(b) − g(a) Let P be a partition of [a. Thus. f ∈ R(c − g) = R(g). P ) and observe that g 0 is bounded. then Ug (f.7. g(b) − g(a) where we have used the fact g is increasing to obtain a telescoping sum. g = c − d where c. Theorem. b]. b] → R is continuous and g : [a. If k(x) = xh(x). B. h are continuous and increasing on [a. then g is of bounded variation. D. Show that G = g ◦ ϕ is of bounded variation and find a formula relating Vab g and Vcd G. b] → R is C 1 . P ) ≤ n X 0 [g(xj ) − g(xj−1 )] = j=1 (g(b) − g(a)) = . Suppose that f (x) = sign(x). 1/π]. E. b] is increasing. as otherwise g is constant and then every function is in R(g).7. then f = g−h where g. By the R-S condition.7 A. b]. We assume g(b) − g(a) > 0. Define h : [0. f ∈ R(c) and f ∈ R(d) so by Theorem 6. continuous. Proof.7. Suppose that g : [a.22. then f ∈ R(g) on [a. If f : [a. f ∈ R(g). y in [a. there is δ > 0 so that x. b] → R is continuous and g : [a. C. h(x) = 0 if x = 0. P ) − Lg (f. Show that if g : [a. . d are increasing functions on [a. show that k is of bounded variation on [0. Show that if f : [a. b] → R is increasing. P ) ≤ 0 .12 Differentiation and Integration 6. Show that if g is the Heaviside step function.  6.7. Proof. b] compact. Hint: Apply the Mean Value Theorem to V (g. Fix  > 0. b]. b] → R is continuous and of bounded variation. If f : [a. b] → R is of bounded variation and that ϕ : [c. b] → R is of bounded variation. P ) − mj (f.23.

b] → R.7 Riemann-Stieltjes Integration 13 F. Suppose that f. there are evaluation sequences X and Y . P. b] → R is Riemann-Stieltjes integrable with respect to g(x) = x. then f ∈ R(h). Find a continuous function f : [0. we have f ∈ R(h)? . 1/π] → R. If H is the ∞ Heaviside step function and g(x) = i=1 ri H(x − xi ). b]. K. find a δ > 0 as in the definition. Suppose i≥1 ri is a convergent series of positive numbers and xi is a sequence of real numbers in (a. 1/π] → R by ( x1/2 sin(1/x) if x > 0. is it true that any continuous function f : [0. J. f dg exists.e. Suppose f P : [a. Thus. RGiven a function g : [a. b] → R. For any partition P and any ε > 0. Show that f ∈ R(g). that f is Riemann integrable. g are bounded on [a. Suppose f : [a. G. b]. a I. M. P H. b] → R is of bounded variation and let h(x) = Vax g for x ∈ [a. i. P.7. For h as in Examples 6. g(x) = 0 if x = 0. P ) + ε. If f : [a. show that g is welldefined and that Z b ∞ X f dg = ri f (xi ). we cannot change ‘g of bounded variation’ to ‘g continuous’.23. L. P ) − ε and Ig (f. b] → R satisfies f ∈ R(g). in Corollary 6. 1/π] → R so that f ∈ / R(g). Define g : [0. Suppose that g : [a. Prove directly from the definition of Riemann integrable. then show that Z  sup f dg : f ∈ C([a. for each continuous f : [a. b]. so that.7. b] → R is continuous except at finitely many points and g is continuous at those points and of bounded variation. kf k∞ ≤ 1 = Vab (g). b]). b] → R is continuous.6. Y ) < Lg (f.18. i=1 Suppose that f : [a. X) > Ug (f. b] and g is increasing on [a. so that Ig (f.