You are on page 1of 12

# 3.

3

Improper Integrals

In the previous section, we deﬁned Riemann integral for functions deﬁned on closed and
bounded interval [a, b]. In this section our aim is to extend the concept of integration to
the the following cases:
1. The function f (x) deﬁned on unbounded interval [a, ∞) and f ∈ R[a, b] for all
b > a.
2. The function is not deﬁned at some points on the interval [a, b].
We ﬁrst consider
Improper integral of ﬁrst kind: Suppose f is a bounded function deﬁned on [a, ∞)
and f ∈ R[a, b] for all b > a.
Deﬁnition 3.3.1. The improper integral of f on [a, ∞) is deﬁned as  

b

f (x)dx := lim

b→∞

a

f (x)dx.
a

If the limit exists and is ﬁnite, we say that the improper integral converges. If the limit
goes to inﬁnity or does not exist, then we say that the improper integral diverges.
Examples: 
∞ 
b
1. (i) 1 x12 dx = limb→∞ 1 x12 dx = limb→∞ 1 − 1b = 1. 
∞ dx 
b dx
π
b
2. (ii) 0 1+x
2 = limb→∞ 0 1+x2 = limb→∞ arctan x 0 = 2 .
Theorem 3.3.2. Comparison test: Suppose 0 ≤ f (x) ≤ φ(x) for all x ≥ a, then 
∞ 

1. a f (x)dx converges if a φ(x)dx converges. 
∞ 

2. a φ(x)dx diverges if a f (x)dx diverges. 
x 
x
Proof. Deﬁne F (x) = a f (t)dt and G(x) = a g(t)dt. Then by properties of Riemann
integral, 0 ≤ F (x) ≤ G(x) and we are given that lim G(x) exists. So G(x) is bounded.
x→∞

F is monotonically increasing and bounded above. Therefore, lim F (x) exists.
x→∞

Examples: 
∞ dx
1. 1 x2 (1+e
x ) . Note that

1
x2 (1+ex )

<

1
x2

and

15 

1

dx
x2

converges.

then the integral a f (x)dx converges. /// a The converse of the above theorem is not true. x | sin x| dx x (n+1)π | sin x|  nπ π 0 ∞ 21 sin x = . ∞ 0 2 e−x sin x . Easy to see that | sin | ≤ | x13 | and 1 x3 x3 2.3.3. For example take the integral This integral does not converge absolutely. Examples: ∞ x ∞ x 1. Therefore the integral is proper . = 1. Let f ∈ R[a. If the integral a |f (x)|dx converges. b] for all b > a. So the improper integral a f (x) + ∞ |f (x)|dx converges by comparison theorem above.  ∞ π   | sin x| dx = x n=1 ∞ ≥ = nπ ∞  n=1 ∞  n=0 (n+1)π 1 nπ 1 nπ  ∞ π sin x dx.∞ x3 dx. ∞ Deﬁnition 3.  b lim b→∞ 1 sin x dx = lim b→∞ x  b 1 d(1 − cos x) x 1   b (1 − cos b 1 − cos x = lim + dx b→∞ b x2 1 It is not diﬃcult to show that the limits on the right exist. π n=1 n On the other hand. Note that 0 ≤ f (x) + |f (x)| ≤ 2|f (x)|. ∞ ∞ Theorem 3. 1 Note that > on [1. ∞ ∞ ∞ f (x)dx = a f (x) + |f (x)|dx − a |f (x)|dx also converges. 1 sin dx. log(1+x) Here ﬁrst note that limx→0 2 e−x sin x log(1+x) 16 dx x3 converges. Then we say a f (x)dx converges ∞ absolutely if a |f (x)|dx converges. ∞ Proof.4. by integration by parts.3. x+1 x3 x+1 x2 2 ∞ x2 dx diverges. 2. Therefore. Indeed. ∞) and 1 In the following we show that absolutely convergence implies convergence of improper integral. Also a |f (x)| converges.

it is enough to prove the convergence/divergence for x large. 17 . Examples: ∞ ∞ (x) 1 1. ∞).  ∞ dx ∞ f (x) 1 1 2. for any  > 0. a f (x)dx converges ⇐⇒ a g(x) converges. If L = 0. then a f (x)dx converges if a g(x)dx converges. say x ≥ M . we get the result. we have the comparison (L − )g(x) < f (x) < (L + )g(x). In case (2). 1 1+x g(x)dx 2 . x→∞ g(x) ∞ ∞ 1. Then limx→∞ fg(x) = 1 and 1 g(x)dx x+1 ∞ diverges. In case (3) by the deﬁnition. g(x) Thus for x ≥ M .5. there exists. we have f (x) < (L + )g(x). Using the property 3. Proof. then a f (x)dx diverges if a g(x)dx diverges. If L = ∞. Take f (x) = √x+1 and g(x) = √1x .3. So by above theorem. Limit comparison test: Let f (x)..e. we can ﬁnd  > 0 such that L −  > 0. i. there exists M > 0 such that x ≥ M =⇒ L −  < f (x) < L + .e. 1 √dx . i. R such that f (x) > M g(x) for all x > R. For x > 10(say): 2 e−x 2 |f (x)| ≤ < e−x ≤ e−x log(1 + x) Hence the integral ∞ 2 e−x sin x 10 log(1+x) converges by comparison theorem. i. If L ∈ (0. 1 f (x)dx diverges. Now in case (1). From the deﬁnition of limits. Take f (x) = 1+x2 and g(x) = x2 . In this interval. a g(x)dx diverges ∞ =⇒ a f (x)dx diverges. 1 f (x)dx converges. ∞ ∞ ∞ 3. integrate on both sides. Then limx→∞ g(x) = 1 and 1 ∞ converges.e. Again. for every M > 0. since L > 0. g(x) are deﬁned and positive for f (x) all x ≥ a and lim = L. ∞ ∞ ∞ 2. Now integrating this. Theorem 3. we have (L − )g(x) < f (x) < (L + )g(x).at x = 0. then the improper integrals a f (x)dx and a g(x)dx are either both ∞ ∞ convergent or both divergent. Now the result follows similar to (1) and (2).. So by above theorem. a g(x)dx con∞ verges =⇒ a f (x)dx converges.

. (Comparison Theorem:) Suppose 0 ≤ φ(x) ≤ f (x) for all x ∈ [a..6. If L = ∞ and a g(x)dx diverges then a f (x)dx diverges.8. c c 1.. ∞ x dx. a2 .an are ﬁnitely many discontinuities of f (x) in [a. If a φ(x)dx diverges then a f (x)dx diverges. Otherwise. Then we deﬁne  c  c− f (x)dx = lim f (x)dx. Theorem 3.. Example: x →0  x →0 0 Suppose a1 .3.. c c 3. c c 2. ∞). b]. →0 a b a Then a f (x)dx is said to converge if the limit exists and is ﬁnite. cosh x f (x) limx→∞ g(x) 0 x x −x Let f (x) = cosh = e2xe . Then a f (x)dx and a g(x)dx both converge or diverge together.3. Then c c 1. then we say the improper integral of f over [a.. If L = 0 and a g(x)dx converges then a f (x)dx converges. Then   b f (x)dx = a  a1 f (x)dx + a  a2 f (x)dx + a1  a3 b f (x)dx + .  1  1 √ dx dx √ = lim √ = lim 2(1 − ) = 2. Otherwise. If a f (x)dx converges then a φ(x)dx converges. g(x) be continuous in (x) [a. c) and f ∈ R[a. If L ∈ (0. Improper integrals of second kind Deﬁnition 3. c c 2. we say b improper integral a f (x)dx diverges. + a2 f (x)dx an If all the improper integrals on the right hand side converge. we say it diverges. c) and are discontinuous at c.7. b] converges.3. The following comparison and Limit comparison tests can be proved following similar arguments: Theorem 3. . Then 2x +1 ∼ xe x ∞ = 2 and 0 g(x) converges.3. Let f (x) be deﬁned on [a. c − ] for all  > 0. 18 . c) and limx→c fg(x) = L. So choose g(x) = xe−x . (Limit comparison theorem:) Suppose 0 < f (x).

by our familiarity with existence of limits. which is a proper integral.  1 −1 dx = x3  0 −1 dx + x3  −1  1 0 dx x3 dx = lim + lim 1 →0 −1 x3 2 →0 1 1 1 lim ( 2 − 2 ). for each  > 0. ∞). = 2 1 . For example. 2 = n2 . It is easy to see that if one takes 1 = 2 . we are getting diﬀerent limits. Since the domain of integration is (0. But if one takes 1 = (n+1) 2 .Proof. So through diﬀerent sequences. 2 → 0 as n → ∞. Gamma and Beta functions: Consider the Gamma function deﬁned as improper integral for p > 0. the integral is improper 19 .7) does not exist. the transformation 3 − x = u then I = u 3 − u2 0 Remark 3. . From the deﬁnition of limit. we say integral diverges. Transforming improper integrals: Sometimes improper integrals may be transformed into proper integrals.2 →0 1 2  1 2 dx x3 (3. if we choose 1/2 y 3y − 1  √2 2udu 2 √ .  ∞ Γ(p) = xp−1 e−x dx 0 This integral is improper of second kind in the neighbourhood of 0 as xp−1 goes to inﬁnity as x → 0 (when p < 1). It is important to note that the ”symmetric” limit could be convergent but the limit may not exist. By now. for example 3 dx 1 . This is an improper integral of ﬁrst kind. by choosing  < L.6. Instead. c) =⇒ (L − )g(x) < f (x) < (L + )g(x) Rest of the proof follows in the similar lines theorems on ﬁrst kind. there exists δ > 0 such that x ∈ (c − δ. then the limit exits 1 1 and is equal to 0. then the above limit in (3.7) where 1 . Taking the transformation y = consider the improper integral I = √ √ 3−x x 3−x 1  ∞ dy √ we get I = .

 ∞ Γ(α + 1) = 0 xα e−x = −(xα e−x )|∞ 0 +α 20  ∞ 0 xα−1 e−x dx = αΓ(α). we divide the integral into  1 Γ(p) = xp−1 e−x dx +  0 ∞ xp−1 e−x dx 1 =I1 + I2 f (x) =1 To see the convergence of I1 we take f (x) = xp−1 e−x and g(x) = xp−1 . x2 Then lim 1 dx x2 converges. Γ(1) = 0 e−x dx = 1. 2. Some identities of beta and gamma functions: ∞ 1. take f (x) = xp−1 e−x and 0 ∞ f (x) = lim x2+p−1 e−x = 0 and 1 x→∞ g(x) x→∞ (2) of limit comparison theorem. we take f (x) = xp−1 (1 − x)q−1 and g(x) = (1 − x)q−1 . Hence by Next we consider the Beta function deﬁned as improper integral for p > 0. then the integral is deﬁnite integral. then lim x→0 g(x)  1 and xp−1 dx converges. . the integral converges. To prove the convergence. To see the convergence of I1 . To prove the convergence. Integration by parts formula implies. When p < 1 and/or q < 1.of ﬁrst kind. q > 0. Γ(α + 1) = αΓ(α). this integral is improper of second kind at 0 and/or 1. take f (x) = xp−1 (1 − x)q−1 and g(x) = xp−1 . Similarly.  1 β(p. for convergence x→0 g(x) x→0 0 of I2 . we divide as before   1 x 0 p−1 (1 − x) q−1  1/2 dx = x p−1 0 (1 − x) q−1 1 dx + 1/2 xp−1 (1 − x)q−1 dx = I1 + I2 . Then  1/2 f (x) q−1 lim = lim (1 − x) = 1 and xp−1 dx converges. To see the convergence of I2 . q) = 0 xp−1 (1 − x)q−1 dx If p > 1 and q > 1. g(x) = 1 .

take t = x. then the integral becomes t3 (1 − t)1/2 tdt = 2  t4 (1 − t)1/2 dt = 2β(5. m) β(m. n) =  ∞  √ 2/3 − x 1 3 Problem: Evaluate (i) x e dx (ii) x 2 (1 − 0 0 √ For (i). Substituting t = 1 − x in the deﬁnition of β(m. n) = 0  π/2 5.  1 Γ( ) 2 2  ∞  ∞ =4  0 ∞  2 take u = r cos θ. 2 e−r rdrdθ = π 0 4. n) = β(n. v = r sin θ. It is easy to see from the deﬁnition that 0 I = lima→∞ (1 − cos a) which does not exist. n) = 2 sin2m−1 θ cos2n−1 θdθ 0 Taking x = sin2 θ in β(m. Γ(m+n)    ∞ e−t t7/3 dt = 2Γ( 0 √ √ 1 x) 2 dx 10 56 ) = Γ(1/3). Γ( 12 ) = π. 0 π/2 =4 0 2 e−u e−v dudv. β(m. 0 Γ(m)Γ(n) . 3/2) = 2 512 Γ(5)Γ(3/2) = Γ(13/2) 3465 Cauchy Principal Value: ∞ Consider the improper integral I = 0 sin xdx. But  c lim c→∞ sin xdx −c 21 . again take t =  1 2 0 π/2 cos2m−1 θ sin2n−1 θdθ.Therefore. √ 3. we get  1 tn−1 (1 − t)m−1 dt = β(n. n). m). β(m. n). β(m.Similarily. then the given integral becomes ∞ t4/3 e−t 2tdt = 2 0 For (ii). 3 27 x. n) = cos2m−2 θ sin2n−2 θ cos θ sin θdθ = 2 0 6. we get  π β(m. Γ(m + 1) = m! ∀m ∈ IN . −∞ sin xdx does not exist.

which doesn’t make sense. Theorem 3.exists and is equal to 0. . It is sometimes not possible and leads to wrong ∞ assertions. taking tx = y. α)dx a where the integrand is depend on the parameter α. For example.. It is easy to notice with change ∞ = π2 .9... . 2.10. 22 . which explains under which conditions we can do the diﬀerentiation under integral sign. But the the improper integrals 0 x2n+1 and −∞ x2n+1 does not lim a→∞ −a converge. Here we have a theorem. At times we can diﬀerentiate under the integral sign to evaluate the integral. In this case it is easy to check that  a ∞ 0 x2n+1 = 0. Simply evaluate  − x −(2n+1)  1 + −1 x−(2n+1)  to see that the the limit is 0. with c ∈ (a. Now diﬀerentiating of variable formula..3. that I = I(t) = 0 sin(tx) x  ∞ this. This is called Cauchy Principal value of improper integral Deﬁnition 3.. Suppose. 3. for n = 1.3. this symmetric limit exists. b) as point of discontinuity of f (x) as  c−δ  b  b f (x)dx = lim f (x)dx + f (x)dx CPV δ→0 a a c+δ Examples: ∞ First kind: −∞ x2n+1 dx for all n = 1. 3. taking derivative inside integral. Second kind: 1 −1 x−(2n+1) dx. we get I  (t) = 0 cos(tx)dx = 0.. 2. Integrals dependent on a Parameter Consider an integral  b I(α) = f (x. The Cauchy Principal value of improper integral of ﬁrst kind is deﬁned as   ∞ CPV R f (x)dx = lim R→∞ −∞ f (x)dx −R For the improper integral of second kind. we know that I = 0 sinx x = π2 . Though the improper integral does not exist.

α)| ≤ A(x). a(x))a (x) dx sin αx dx. and   b I (α) = a d f (x. . α) +  = f (x. b(x))b (x)−f (x. Evaluate I(α) = |I| ≤ e e−x t2 /2 1 .  2. α)dx Δα→0 a dα dα a dα /// In fact the following holds. α))dx Δα a d d Now by Taylor’s theorem. |f (x. x  0 ∞ By the above formula. t)dt. α)dx. Suppose f. we have dα f (x. α + Δα) − f (x. α + θΔα) = dα f (x. Hence C = 0. dα Proof. Then h (x) = a(x) a(x) Examples:  ∞ df (x. dα f (x. α) = Δα dα f (x. 2 2 . then the improper integrals a Adx. α) + . b are integrable on [a. I  (α) = 0 e−x cos αxdx =  ∞ −x C. α)| ≤ B(x) such that a. Since dα f (x. Test the convergence and evaluate the integral 1. b] and α in an interval of containing α0 . 0  ∞ 0 |e −x2  cos(tx)|dx ≤ C 23 ∞ 0 1 1 2 e−x dx = Γ( ). 1+α2 ∞ 1 e 2 (t Therefore. α) d d is continuous. | dα f (x. where 0 < θ < 1 and  → 0 as Δα → 0.d 1. Also I(0) = 0 e sin 0x = 0. a Bdx converge. d I(α + Δα) − I(α) I(α) = lim Δα→0 dα Δα  b  1 = lim (f (x. α) are continuous functions for x ∈ [a. α + θΔα). f (x. d 2. Thus  b  b d d d I(α) = lim f (x. t)dt+f (x. Newton-Leibnitz Formula:  b(x) Let h(x) =   b(x) f (x. I(α) = arctan α+ 2 −x2 ) cos(tx)dx. Then I is diﬀerentiable. If the b b domain in unbounded. b]. α + Δα) − f (x.

 1 x2 sin(x2 )dx = 0  ∞ 1 0 = ∞  n=0 = ∞  n=0 Hence  1 0 x2 sin(x2 ) = n=0 (−1)n x4n+4 (2n + 1)! (−1)n x4n+5 1 (4n + 5)(2n + 1)! 0 (−1)n (4n + 5)(2n + 1)! 1 1 1 1 − + − 5 54 13 × 5! 17 × 7! Since the 4th term has 4 zeros in the ﬁrst 4 decimal places. Let f (x) = k=0 an t for |t| < R. Then we can approximate the deﬁnite integral using ∞ n the Taylor series. So the approximation has error less than 10−4 . We note that if the nt h term has k zeros in the ﬁrst k decimals.Hence the integral converges. By Newton Leibnitz formula.4 ∞ 0 2 /2 e−x dx = √1 Γ(1/2). b] and has its Taylor series in an interval containing [a. Ia (t)  a =  2 −x2 ) 0 a = 0 =e 1 e 2 (t (t cos tx − x sin tx)dx ∂ 1 (t2 −x2 ) e2 sin txdx ∂x 1 2 (t −a2 ) 2 sin at Therefore. 24 . I  (t) = lima→∞ Ia (t) = 0. 2 Approximation of deﬁnite integrals Suppose we are given a function f : [a. b] → IR that is integrable on [a.  t f (x)dx = 0 ∞   k=0 t 0 an xn dx = ∞  an k=0 tn+1 |t| < R n+1 1 Example: Find the approximate value of the deﬁnite integral 0 x2 sin(x2 )dx with error less than 10−4 . b]. 3. Then we know from the termby-term integration theorem. then the approximation is with error 10k−1 . Now note that I(0) = Hence I(t) = π2 .

Let θ be the angular coordinate. Then the quarter-period time τ /4 is given by L 4g τ = 4  α 0  α θ sin ( ) − sin2 ( ) 2 2 2 By taking the transformation θ .Application: A pendulum of mass m suspended by a inextensible string of length L is released from initial position of angle α (from equilibrium position).

... Then we can still approximate the deﬁnite integral using Riemann sums. When we take equal partitions. that if the function is integrable then deﬁnite integral can be computed as  b f (x)dx = lim a n→∞ n  f (ξk )Δxk k=1 where ξk ∈ [xk−1 . + yn−1 ) a  b f (x)dx ≈ (y1 + y2 + .. When we take ξk = xk−1 Δxk = b−a n and ξk = xk we get the sums ((y0 + y1 + . we get . + yn ) a 25 b−a n b−a . n b a f (x)dx when ... So we write Rectangular formula  b f (x)dx ≈ (y0 + y1 + . + yn−1 ) Δx (y1 + y2 + .. .yn = f (xn ).. Recall from section 3.. y1 = f (x1 ). This integral is an improper integral which is diﬃcult to evaluate. + yn ) Δx Each of these sums is a Riemann sum and converges to deﬁnite integral n → ∞..1. Now suppose the function is not continuous but only integrable. k = sin2 (α/2).... sin(α/2) this integral is transformed 1 (1 − k 2 sin2 φ)− 2 dφ.→ φ deﬁned as sin φ = to  π/2 0 − 12 dθ... We denote y0 = f (x0 ). xk ] and Δxk = xk − xk−1 . sin(θ/2) . We can use Taylor series to ﬁnd approximate value as described in the previous example.

26 . This is also seen as piecewise linear approximation of the function y = f (x). we could take mid-points of the interval for ”better approximation” to obtain Trapezoidal Rule:  b a  y1 + y2 yn−1 + yn y0 + y1 Δx + Δx + . (Explain Geometrically)! We can also take quadratic approximation or with exact values at three points. Instead of taking the end points..... + yn−1 n 2  As the name suggests. Approximations like this and estimation of errors is part of Numerical integration theory which is beyond the scope of this course..These approximations involve the area of rectangles with sides xk−1 xk and height yk−1 or yk . + Δx f (x)dx ≈ 2 2 2   b − a y0 + yn ≈ + y1 + y2 + . the ﬁrst two formulas involve the sum of areas of rectangles and trapezoidal rule consists of areas of trapezoids.