The rules for combining integrals, the inequalities between them, the mean value theorems for integrals, and the fundamental theorem of the calculus, all proved from the definition.
Every calculation with integrals uses a handful of rules without comment: integrals over adjacent intervals add, constants come out, a larger function has a larger integral. §165 collects these rules and proves them. Two of them carry more weight than the rest. The mean value theorem for integrals says that the average height of a continuous function over an interval is one of its values; its weighted form will produce Cauchy’s remainder in Taylor’s theorem (§167). And the fundamental theorem of the calculus, that dxd∫axf=f(x), ties the integral of §161 to the antiderivatives of Chapter VI.
The idea
Most of the rules are statements about areas that are obvious from a picture. What is less obvious is that they follow from the definition, and the method is always the same: prove the inequality or identity for every approximating sum, then pass to the limit. A sum of rectangles obeys each rule trivially, and limits preserve sums, constant multiples and (non-strict) inequalities.
The mean value theorem is the one rule with a real idea in it. The integral divided by the length, b−a1∫abf, is the average height of f: the height of the rectangle on [a,b] with the same area as the region under the curve. Since that rectangle cannot be taller than the highest point of the curve or lower than the lowest, its height lies between the least value H and the greatest value K; and a continuous function takes every value between H and K.
ExploreThe average height is a value of f
Try this.Move a and b and watch the purple level η (the average height) stay inside the green band from the least to the greatest value. Then drag b down until [a,b] is very short.
f(x)=1.5+sin2x+0.4sin5x on [0.3,2.6]
xy
y = f(x); shaded: the area ∫f, counted negative below the axis
between H (least value) and K (greatest value) on [a, b]
η = (1/(b − a))∫f, the average height
rectangle of height η: same area as under the curve
points ξ with f(ξ) = η
Function
H=0.1351, K=2.595, ∫abf=3.5615, and H(b−a)=0.3107≤3.561≤5.967=K(b−a).
Average height η=b−a1∫abf=1.5485. It is taken at ξ=1.705.
What to notice
The rectangle of height η always has the same area as the region under the curve, so η can never leave the band; and since f is continuous, the curve must cross the level η somewhere in [a,b], at a point ξ (sometimes at several). As b approaches a the band shrinks to the single value f(a), and η is squeezed onto it. That is the fundamental theorem of the calculus in one picture.
Precise statements
The definition of §161 assumed a<b. The first two rules extend it; with the definition by antiderivatives, ∫abf=F(b)−F(a), they are automatic.
DefinitionReversed and empty rangesHardy: §165 (1), (2)
∫abf(x)dx=−∫baf(x)dx(a>b),∫aaf(x)dx=0.
TheoremLinearity and additivityHardy: §165 (3)–(5)
For continuous f, ϕ, any constant k and any a, b, c in an interval where they are continuous:
Let a<b. If f(x)≥0 on [a,b] then ∫abf≥0. If H≤f(x)≤K on [a,b] then
H(b−a)≤∫abf(x)dx≤K(b−a).
TheoremFirst mean value theorem for integralsHardy: §165 (8)
If f is continuous on [a,b], there is a ξ in [a,b] with
∫abf(x)dx=(b−a)f(ξ).
If F is an antiderivative of f this reads F(b)−F(a)=(b−a)F′(ξ): it is the mean value theorem of §126 applied to F.
Now give the points of [a,b] different importance, by a weightϕ(x)≥0. The weighted average of f is ∫fϕ/∫ϕ. The same reasoning places it between H and K.
ExploreA weighted average is still a value of f
Try this.Choose the weight ϕ=x−a, which ignores the left end and stresses the right; then ϕ=(b−x)2, which stresses the left. Watch where the level η and the points ξ go.
f(x)=x3−2x on [−1.4,1.6]
xy
y = f(x); shaded: the area ∫f, counted negative below the axis
between H (least value) and K (greatest value) on [a, b]
η = (1/(b − a))∫f, the average height
rectangle of height η: same area as under the curve
points ξ with f(ξ) = η
Function
Weight
H=−1.089, K=1.089, ∫abf=0.078, and H(b−a)=−3.266≤0.078≤3.266=K(b−a).
Average height η=b−a1∫abf=0.026. It is taken at ξ=−0.013,1.421.
What to notice
Weighting moves the average towards the values of f where ϕ is large, but it never pushes it out of the band between H and K, because the weight is never negative. So the weighted average is again a value f(ξ).
TheoremGeneralised mean value theorem for integralsHardy: §165 (9)
Let f and ϕ be continuous on [a,b], with ϕ(x)≥0, and let H, K be the least and greatest values of f there. Then
TheoremThe fundamental theorem of the integral calculusHardy: §165 (10)
If f is continuous on an interval containing a, the function F(x)=∫axf(t)dt has derivative F′(x)=f(x). In particular F is continuous.
Proofs
Hardy suggests proving each rule in two ways: from an antiderivative, and directly from the sums. For (3) the first way is one line: {F(b)−F(a)}+{F(c)−F(b)}=F(c)−F(a). But the direct proof is the one that does not presuppose that an antiderivative exists, and it shows the method.
ProofAdditivity, from the sums
Strategy
Treat a<b<c first: a partition of [a,b] and one of [b,c] together form a partition of [a,c], and its Riemann sum is the sum of theirs. Other orders of a,b,c then follow from the convention (1).
Join two partitions
Let a<b<c. Take partitions of [a,b] and [b,c] with longest piece less than δ, and sums σ1 and σ2 (any sample points). Together, they partition [a,c] with longest piece less than δ, and its sum is exactly σ1+σ2.
Pass to the limit
As δ→0, σ1→∫abf, σ2→∫bcf and σ1+σ2→∫acf, by the existence theorem of §161. Limits of sums are sums of limits, so ∫abf+∫bcf=∫acf.
Any order
If, say, b<a<c, the case just proved gives ∫baf+∫acf=∫bcf. Using ∫baf=−∫abf from (1) and rearranging gives the same identity. Every other order, and equal points (where (2) is used), is handled in the same way.
∎
Rules (4) and (5) go the same way: ∑kfvδv=k∑fvδv and ∑(fv+ϕv)δv=∑fvδv+∑ϕvδv for every partition, with the same sample points; now let the pieces shrink. For (6), every sum ∑fvδv with f≥0 is ≥0, so its limit is ≥0. And (7) is (6) applied to f−H≥0 and K−f≥0, using (4), (5) and ∫ab1=b−a.
ProofFirst mean value theorem
Strategy
Divide the integral by b−a to get a number between the least and greatest values of f; then the intermediate value theorem finds a point where f takes it.
Bounds
f is continuous on the closed interval [a,b], so it attains a least value H and a greatest value K (Chapter V, §§101–105). By (7), H(b−a)≤∫abf≤K(b−a).
The average lies between them
So η=b−a1∫abf satisfies H≤η≤K.
Find ξ
f takes the values H and K at some points of [a,b] and is continuous, so it takes every value between them (§101), in particular η, at some ξ in [a,b]. Then ∫abf=(b−a)f(ξ).
∎
ProofGeneralised mean value theorem
Strategy
Apply positivity (6) to two functions that are non-negative because ϕ≥0; this gives the inequalities. Then the weighted average lies between H and K, and continuity finishes as before.
Two non-negative integrands
On [a,b], {f(x)−H}ϕ(x)≥0 and {K−f(x)}ϕ(x)≥0, because both factors are ≥0. By (6), (4) and (5), ∫fϕ−H∫ϕ≥0 and K∫ϕ−∫fϕ≥0.
If ∫ϕ=0
Then both inequalities force ∫fϕ=0, and any ξ works. (In fact, for continuous ϕ≥0 this happens only when ϕ is identically 0; see the exercise below.)
If ∫ϕ>0
Divide: η=∫fϕ/∫ϕ satisfies H≤η≤K, so η=f(ξ) for some ξ in [a,b] by the intermediate value theorem, and ∫fϕ=f(ξ)∫ϕ.
∎
Notice where ϕ≥0 was used: in the first step. Without it the inequalities can reverse, and the theorem fails (see the quiz).
ProofThe fundamental theorem
Strategy
Write the difference quotient of F as an average of f over a short interval, using additivity; by the mean value theorem that average is a value f(ξ) with ξ squeezed onto x.
The difference quotient is an average
For h=0, additivity gives F(x+h)−F(x)=∫xx+hf(t)dt. So hF(x+h)−F(x) is the average of f over the interval between x and x+h (for h<0 use (1); the quotient is still the average).
Mean value theorem
By (8) it equals f(ξh) for some ξh between x and x+h.
Let h→0
Then ξh→x, and since f is continuous, f(ξh)→f(x). So F′(x)=f(x). A function with a derivative is continuous at each point (if the difference quotient has a limit, the difference itself tends to 0), so F is continuous.
∎
Worked examples and exercises
ExerciseEven and odd integrands●○○Hardy: Examples LXV, 1 (i)
Show that ∫−aaϕ(x2)dx=2∫0aϕ(x2)dx and ∫−aaxϕ(x2)dx=0, for continuous ϕ, using the definition by sums and rules (1)–(5).
Hint
Split [−a,a] at 0 by (3). Reflect a partition of [0,a] in the origin to get one of [−a,0], and use reflected sample points.
Solution
Let g be continuous on [−a,a]. Take a partition 0=x0<⋯<xn=a of [0,a] with sample points ξv; its mirror image −xn<⋯<−x0 partitions [−a,0] into pieces of the same lengths, and we use the sample points −ξv. The two Riemann sums are ∑g(ξv)δv and ∑g(−ξv)δv.
If g(x)=ϕ(x2) then g(−ξ)=g(ξ) and the sums are equal; letting the pieces shrink, ∫−a0g=∫0ag, and (3) gives ∫−aag=2∫0ag. If g(x)=xϕ(x2) then g(−ξ)=−g(ξ), the sums are negatives of each other, ∫−a0g=−∫0ag, and ∫−aag=0. In pictures: an even function has mirror-image areas on the two sides; an odd one has equal areas of opposite sign.
Exercise●●○Hardy: Examples LXV, 3
Prove that ∫0πsinxsinnxdx is π when n is an odd positive integer and 0 when n is even.
Hint
Show that sinnx=sinx{2cos(n−1)x+2cos(n−3)x+…}, where the last term in the bracket is 1 if n is odd and 2cosx if n is even. Use 2sinxcoskx=sin(k+1)x−sin(k−1)x.
Hint
∫0πcoskxdx=0 for every integer k=0.
Solution
By the identity in the first hint, 2sinxcos(n−1)x+2sinxcos(n−3)x+… telescopes: the terms are {sinnx−sin(n−2)x}+{sin(n−2)x−sin(n−4)x}+…. For odd n the last bracket term is 1, contributing sinx, and the sum collapses to sinnx. For even n the last term 2sinxcosx=sin2x, and again the sum collapses to sinnx. So, away from the zeros of sinx,
The right side is continuous on [0,π], and it is the continuous extension of the integrand, so we integrate it. Every term 2coskx with k≥1 integrates to 0 over [0,π]. For odd n the constant 1 remains and gives π; for even n nothing remains, and the integral is 0.
ExerciseEstimating without integrating●○○Hardy: Examples LXV, 8
If n>1, show that 0.5<∫01/21−x2ndx<0.524.
Hint
Use (7) and Examples LXV, 6 (a larger integrand has a larger integral). For 0<x<1 and n>1: x2n<x2.
Solution
On 0<x≤21 we have 0<x2n<x2<1, so 1>1−x2n>1−x2, and therefore
1<1−x2n1<1−x21.
The functions differ except at x=0, so by Examples LXV, 6 and the result of Misc. Ex. 43 the inequalities between the integrals are strict:
ExerciseThe integral of the modulus●○○Hardy: Examples LXV, 12
Prove that ∫abf(x)dx≤∫ab∣f(x)∣dx for a<b.
Hint
Hardy’s hint: compare a Riemann sum σ=∑fvδv with the corresponding sum σ′=∑∣fv∣δv for ∣f∣.
Hint
Or without sums: −∣f∣≤f≤∣f∣, and integrate.
Solution
∣f∣ is continuous when f is. For any partition and sample points, the triangle inequality gives ∣σ∣=∣∑fvδv∣≤∑∣fv∣δv=σ′, since every δv>0. Letting the pieces shrink, σ→∫f and σ′→∫∣f∣, and the inequality survives the limit. (Alternatively: −∣f(x)∣≤f(x)≤∣f(x)∣ for every x, so by (6) −∫∣f∣≤∫f≤∫∣f∣.) The same idea, with complex numbers, is the subject of Integrals of complex functions of a real variable.
ExerciseZero integral of a non-negative function●●○Hardy: §165, Misc. Ex. 43
Let f be continuous and f(x)≥0 on [a,b], with ∫abf=0. Prove that f(x)=0 for every x in [a,b].
Hint
Suppose f(ξ)=k>0. Continuity gives an interval around ξ on which f>21k.
Solution
Suppose f(ξ)=k>0 for some ξ. By continuity there is δ>0 with f(x)>21k on J=[ξ−δ,ξ+δ]∩[a,b], an interval of length at least min(δ,b−a)>0. Split [a,b] by (3) into J and the parts outside it. The integrals over the outer parts are ≥0 by (6), and the integral over J is at least 21k times its length by (7). So ∫abf>0, a contradiction. Hence f≡0. Consequently, in (6) the integral of a continuous f≥0 is strictly positive unless f vanishes identically.
Check yourself
Check yourselfEnter a number
Using the convention (1), find ∫20xdx.
Why
∫20xdx=−∫02xdx=−2. The convention makes ∫abf=F(b)−F(a) true for every order of a and b: here F(0)−F(2)=0−2.
Check yourselfEnter a number
The first mean value theorem gives ∫02x2dx=2ξ2 for some ξ in [0,2]. Find ξ.
Why
∫02x2dx=38, so ξ2=34 and ξ=2/3≈1.155. The average height of x2 over [0,2] is 34, which is between H=0 and K=4.
Check yourselfTrue or false?
The generalised mean value theorem ∫abfϕ=f(ξ)∫abϕ holds for every continuous ϕ, positive or not.
Why
False. Take f(x)=ϕ(x)=x on [−1,1]. Then ∫fϕ=∫−11x2dx=32, but ∫ϕ=0, so f(ξ)∫ϕ=0 for every ξ. The proof used ϕ≥0 to keep (f−H)ϕ and (K−f)ϕ non-negative.
Check yourselfChoose one
Let f be continuous on [a,b] with ∫abf=0. What can you conclude?
Why
By the first mean value theorem 0=(b−a)f(ξ), so f vanishes somewhere. It need not vanish everywhere: ∫02πsinxdx=0. If also f≥0, then f≡0 (the last exercise). Oddness has nothing to do with it on a general interval.
Check yourselfChoose one
What is dxd∫x1et2dt?
Why
By (1), ∫x1=−∫1x, and by the fundamental theorem dxd∫1xet2dt=ex2. So the answer is −ex2. No antiderivative of et2 in elementary terms is needed, or exists.
Hardy’s language
The reader will find it an instructive exercise to write out formal proofs of these properties, in each case giving a proof starting from (α) the definition by means of the integral function and (β) the direct definition.
— G. H. Hardy, §165
Summary
Summary
Each rule (reversal, additivity, linearity, positivity, bounds) is proved for every approximating sum and then passes to the limit.
The average height b−a1∫abf lies between the least and greatest values of f, so for continuous f it is a value f(ξ): the first mean value theorem.
With a weight ϕ≥0, ∫fϕ=f(ξ)∫ϕ; the sign condition on ϕ is essential.
The fundamental theorem follows: the difference quotient of ∫axf is an average of f over a shrinking interval, so it tends to f(x).