M Cauchy Présente

Calculus Level 5

The integral 0 e cos x sin ( sin x ) d x x \int_0^\infty e^{\cos x} \sin(\sin x)\,\frac{dx}{x} can be expressed as 1 P π Q ( R e S ) \frac{1}{P}\pi^Q (Re - S) where P , Q , R , S P,Q,R,S are positive integers where P P , R R and S S are pairwise coprime.

Write the answer as the concatenation of P Q R S PQRS of the integers P , Q , R , S P,Q,R,S . For example, if you think that the integral is equal to 1 3 π 2 ( 4 e 1 ) \tfrac13\pi^2(4e - 1) , give the answer 3241.


The answer is 2111.

This section requires Javascript.
You are seeing this because something didn't load right. We suggest you, (a) try refreshing the page, (b) enabling javascript if it is disabled on your browser and, finally, (c) loading the non-javascript version of this page . We're sorry about the hassle.

5 solutions

Mark Hennings
Jun 17, 2016

For any 0 < ϵ < 1 < R 0 < \epsilon < 1 < R , let Γ ϵ , R \Gamma_{\epsilon,R} be the contour Γ ϵ , R = γ 1 + γ 2 γ 3 γ 4 \Gamma_{\epsilon,R} \,=\, \gamma_1 + \gamma_2 - \gamma_3 - \gamma_4 where

  • γ 1 \gamma_1 is the straight line segment from ϵ \epsilon to R R ,
  • γ 2 \gamma_2 is the semicircular arc z = R e i θ z = Re^{i\theta} for 0 θ π 0 \le \theta \le \pi ,
  • γ 3 \gamma_3 is the straight line segment from ϵ -\epsilon to R -R ,
  • γ 4 \gamma_4 is the semicircular arc z = ϵ e i θ z = \epsilon e^{i\theta} for 0 θ π 0 \le \theta \le \pi .

Since the function f ( z ) = e e i z 1 z f(z) \; = \; \frac{e^{e^{iz}} - 1}{z} is analytic on the punctured disk C \ { 0 } \mathbb{C} \backslash \{0\} , we deduce that Γ ϵ , R f ( z ) d z = 0 \int_{\Gamma_{\epsilon,R}} f(z) \,dz \; = \; 0 for all 0 < ϵ < 1 < R 0 < \epsilon < 1 < R . Now γ 1 f ( z ) d z = ϵ R e e i x 1 x d x γ 3 f ( z ) d z = ϵ R e e i x 1 x d x \int_{\gamma_1} f(z)\,dz \; = \; \int_{\epsilon}^R \frac{e^{e^{ix}} - 1}{x}\,dx \qquad \qquad \int_{\gamma_3} f(z)\,dz \; = \; \int_{\epsilon}^R \frac{e^{e^{-ix}} - 1}{x}\,dx and hence ( γ 1 γ 3 ) f ( z ) d z = ϵ R e e i x e e i x x d x = 2 i ϵ R e cos x sin ( sin x ) d x x \left(\int_{\gamma_1} - \int_{\gamma_3}\right) f(z)\,dz \; =\; \int_\epsilon^R \frac{e^{e^{ix}} - e^{e^{-ix}}}{x}\,dx \; = \; 2i\int_\epsilon^R e^{\cos x}\sin(\sin x)\,\frac{dx}{x} Since f ( z ) f(z) has a simple pole at z = 0 z=0 , we deduce that lim ϵ 0 γ 4 f ( z ) d z = π i R e s z = 0 f ( z ) = π i ( e 1 ) \lim_{\epsilon \to 0} \int_{\gamma_4} f(z)\,dz \; = \; \pi i \mathrm{Res}_{z=0}f(z) \; = \; \pi i (e-1) and so we deduce, letting ϵ 0 \epsilon \to 0 , that 2 i 0 R e cos x sin ( sin x ) d x x + γ 2 f ( z ) d z π i ( e 1 ) = 0 2i\int_0^R e^{\cos x}\sin(\sin x)\,\frac{dx}{x} + \int_{\gamma_2} f(z)\,dz - \pi i(e-1) \; = \; 0 If z γ 2 z \in \gamma_2 and w = e i z w = e^{iz} , then z = R e i θ z = Re^{i\theta} for 0 θ π 0 \le \theta \le \pi , and so w = e R sin θ |w| = e^{-R\sin\theta} . Thus f ( z ) e w 1 z = 1 R n = 1 w n n ! 1 R n = 1 w n n ! w R n = 1 1 n ! e R e R sin θ |f(z)| \; \le \; \displaystyle \left|\frac{e^w - 1}{z}\right| \; = \; \frac{1}{R}\left|\sum_{n=1}^\infty \frac{w^n}{n!}\right| \; \le \; \frac{1}{R}\sum_{n=1}^\infty \frac{|w|^n}{n!} \; \le \; \frac{|w|}{R}\sum_{n=1}^\infty \frac{1}{n!} \; \le \; \frac{e}{R}e^{-R\sin\theta} so that γ 2 f ( z ) d z e 0 π e R sin θ d θ = 2 e 0 1 2 π e R sin θ d θ \left|\int_{\gamma_2} f(z)\,dz\right| \; \le \; e\int_0^\pi e^{-R\sin\theta}\,d\theta \; = \; 2e\int_0^{\frac12\pi}e^{-R\sin\theta}\,d\theta Jordan's Inequality tells us that sin θ 2 π θ \sin\theta \,\ge\, \tfrac{2}{\pi}\theta for 0 θ 1 2 π 0 \le \theta \le \tfrac12\pi , so it follows that γ 2 f ( z ) d z = O ( R 1 ) R \int_{\gamma_2} f(z)\,dz \; = \; O\big(R^{-1}\big) \qquad \qquad R \to \infty which implies, letting R R \to \infty , that 0 e cos x sin ( sin x ) d x x = 1 2 π ( e 1 ) \int_0^\infty e^{\cos x}{\sin(\sin x)} \frac{dx}{x} \; = \; \tfrac12\pi(e-1) so that P = 2 P=2 , Q = 1 Q=1 , R = 1 R=1 and S = 1 S=1 , making the answer 2111 2111 .

Moderator note:

Good clear explanation of how to treat the integral using ideas in complex analysis.

Alternative method [As of now, I still do not know how to solve the issue of convergence. See comments below]

I = 0 e cos x x sin ( sin ( x ) ) d x = 0 [ e cos x x e i sin x ] d x I=\int _0^{\infty}\frac{e^{\cos x}}{x}\sin \left(\sin \left(x\right)\right)dx=\int _0^{\infty}\Im\left[\frac{e^{\cos x}}{x}e^{i\sin x}\right]dx

= 0 [ e cos x + i sin x x ] d x = 0 [ e e i x x ] d x =\int _0^{\infty}\Im\left[\frac{e^{\cos x+i\sin x}}{x}\right]dx=\int _0^{\infty}\Im\left[\frac{e^{e^{ix}}}{x}\right]dx

= 0 [ n = 0 1 n ! e i x n x ] d x = 0 n = 0 1 n ! sin ( n x ) x d x =\int _0^{\infty}\Im\left[\sum _{n=0}^{\infty}\frac{1}{n!}\cdot \frac{e^{ixn}}{x}\right]dx=\int _0^{\infty}\sum _{n=0}^{\infty}\frac{1}{n!}\cdot \frac{\sin \left(nx\right)}{x}dx

= 0 n = 1 1 n ! sin ( n x ) x d x =\int _0^{\infty}\sum _{n=1}^{\infty}\frac{1}{n!}\cdot \frac{\sin \left(nx\right)}{x}dx

Using 0 sin n x x d x = π 2 \displaystyle \int _0^{\infty}\frac{\sin nx}{x}dx=\frac{\pi}{2} for n > 0 n>0 :

I = π 2 n = 1 1 n ! = π 2 ( e 1 ) I=\frac{\pi }{2}\sum _{n=1}^{\infty}\frac{1}{n!}=\frac{\pi}{2}(e-1)

Julian Poon - 4 years, 12 months ago

Log in to reply

This one is not too hard to resolve. We have e cos x sin ( sin x ) x = I ( e e i x x ) = n = 1 1 n ! sin n x x e^{\cos x}\frac{\sin(\sin x)}{x} \; = \; \mathfrak{I}\left(\frac{e^{e^{ix}}}{x}\right) \; =\; \sum_{n=1}^\infty \frac{1}{n!} \frac{\sin nx}{x} for all x 0 x \ge 0 . Moreover, since 1 n ! sin n x x 1 ( n 1 ) ! x 0 , n 1 \left| \frac{1}{n!} \frac{\sin nx}{x}\right| \; \le \; \frac{1}{(n-1)!} \qquad \qquad x \ge 0\;, \; n \ge 1 and since n = 1 1 ( n 1 ) ! < \sum_{n=1}^\infty \frac{1}{(n-1)!} < \infty , we deduce that the above series converges uniformly on ( 0 , ) (0,\infty) . Thus we can reverse the order of integration and summation for a finite interval, and deduce that 0 X e cos x sin ( sin x ) x d x = n = 1 1 n ! 0 X sin n x x d x = n = 1 1 n ! 0 n X sin u u d u \int_0^X e^{\cos x}\frac{\sin(\sin x)}{x}\,dx \; = \; \sum_{n=1}^\infty \frac{1}{n!} \int_0^X \frac{\sin nx}{x}\,dx \; = \; \sum_{n=1}^\infty \frac{1}{n!} \int_0^{nX} \frac{\sin u}{u}\,du for any X > 0 X > 0 , so that 1 2 π ( e 1 ) 0 X e cos x sin ( sin x ) x d x n = 1 1 n ! 1 2 π 0 n X sin u u d u \left| \tfrac12\pi(e-1) - \int_0^X e^{\cos x}\frac{\sin(\sin x)}{x}\,dx \right| \; \le \; \sum_{n=1}^\infty \frac{1}{n!}\left|\tfrac12\pi - \int_0^{nX} \frac{\sin u}{u}\,du \right| for any X > 0 X > 0 .

Given ϵ > 0 \epsilon > 0 , we can find Ξ > 0 \Xi > 0 such that X > Ξ 1 2 π 0 X sin u u d u < ϵ X > \Xi \; \Rightarrow \; \left|\tfrac12\pi - \int_0^X \frac{\sin u}{u}\,du\right| \,<\, \epsilon and hence X > Ξ 1 2 π ( e 1 ) 0 X e cos x sin ( sin x ) x d x n = 1 1 n ! ϵ < e ϵ X > \Xi \; \Rightarrow \; \left| \tfrac12\pi(e-1) - \int_0^X e^{\cos x}\frac{\sin(\sin x)}{x}\,dx \right| \; \le \; \sum_{n=1}^\infty \frac{1}{n!}\epsilon \; < \; e\epsilon By what it means to be an infinite Riemann integral, this implies the result.

Mark Hennings - 4 years, 12 months ago

There are big convergence issues here. Sure, the integrand is given by a uniformly convergent series, but you still need to justify that the term-by-term integral of that series over the infinite interval ( 0 , ) (0,\infty) converges to the integral of the integrand over that interval. You cannot apply either the Monotone Convergence Theorem or the Dominated Convergence Theorem in this case.

Mark Hennings - 4 years, 12 months ago

Log in to reply

Even I did it in this way (yeah, I know I'm technically wrong).

Could you give an example where interchanging the integral and summation gives different results?

Most of the questions on brilliant don't have this issue I guess. (Partly the reason why I don't bother checking much before interchanging.)

A Former Brilliant Member - 4 years, 12 months ago

Log in to reply

@A Former Brilliant Member I have shown an example for a sequence, not a series, but these are equivalent. Consider f n f_n , defined on ( 0 , 1 ] (0,1] by the formula f n ( x ) = n ln ( n + 1 ) ( 1 + n x ) 0 < x 1 f_n(x) \; = \; \frac{n}{\ln (n+1)(1 + nx)} \qquad \qquad 0 < x \le 1 for any positive integer n n . Then 0 1 f n ( x ) d x = [ ln ( 1 + n x ) ln ( n + 1 ) ] 0 1 = 1 \int_0^1 f_n(x)\,dx \; = \; \Big[\frac{\ln(1 + nx)}{\ln(n+1)} \Big]_0^1 \; =\; 1 for all n 1 n \ge 1 , and hence lim n 0 1 f n ( x ) d x = 1 \lim_{n \to \infty} \int_0^1 f_n(x)\,dx \; =\; 1 However, for any 0 < x 1 0 < x \le 1 we have 0 f n ( x ) 1 x ln ( n + 1 ) n 1 0 \;\le\; f_n(x) \; \le \; \frac{1}{x \ln(n+1)} \qquad \qquad n \ge 1 and hence it is clear that lim n f n ( x ) = 0 0 < x 1 \lim_{n \to \infty}f_n(x) \; = \; 0 \qquad \qquad 0 < x \le 1 Thus the sequence of functions f n f_n converges everywhere to the function f ( x ) = 0 f(x) = 0 , and 0 1 f ( x ) d x = 0 \int_0^1 f(x)\,dx \; = \; 0 Thus here is a case where:

  • all the functions f n f_n are integrable,
  • the limit of the integrals 0 1 f n ( x ) d x \int_0^1 f_n(x)\,dx as n n \to \infty exists,
  • the pointwise limit of the sequence of functions f n f_n exists, and defines an integrable function f f

but 0 1 f ( x ) d x lim n 0 1 f n ( x ) d x \int_0^1f(x)\,dx\,\neq\, \lim_{n\to\infty}\int_0^1f_n(x)\,dx .

There are many other examples.

Interchanging limits is always a delicate operation. There are theorems which enable you to interchange a limit and an integral, but when the conditions for these theorems do not hold, as here, you have to be careful. What is tricky is that there are cases (as the one of this problem) where changing the order of the limit and the integral results in the correct answer, although that reversal of order is not justified by a theorem. Life, or mathematics, is like that!

Mark Hennings - 4 years, 12 months ago

Log in to reply

@Mark Hennings Thanks for that example!

Have you come across any problem on brilliant where interchanging gives a different answer?

A Former Brilliant Member - 4 years, 12 months ago

I'm still a beginner on this so mind me if I interpretated it wrongly. Is interchanging the sum and integral not justified? If so, would showing that

n = 1 0 1 n ! sin ( n x ) x d x \sum _{n=1}^{\infty}\int _0^{\infty}\frac{1}{n!}\cdot \frac{\sin \left(nx\right)}{x}dx

converges solve the issue?

Julian Poon - 4 years, 12 months ago

Log in to reply

@Julian Poon It is a nontrivial matter to show that 0 n = 0 f n ( x ) d x = n = 0 0 f n ( x ) d x \int_0^\infty \sum_{n=0}^\infty f_n(x)\,dx \; = \; \sum_{n=0}^\infty \int_0^\infty f_n(x)\,dx even if all the integrals on the RHS exist. The sum on the RHS may fail to converge, or may converge, but not to the integral on the LHS. Even if the sum of integrals on the RHS exists, the integral on the LHS may not exist.

There are two general theorems that make this result true. The Monotone Convergence Theorem tells us that all is well if all the functions f n f_n are positive, and the Dominated Convergence Theorem tells us that all is well if we can find an integrable function g g such that n = 0 N f n ( x ) g ( x ) x , N \left| \sum_{n=0}^N f_n(x) \right| \; \le \; g(x) \qquad \qquad \forall \,x\,,\,N Neither of these conditions are satisfied in this case, so trying to handle this integral via an infinite series is problematic.

Mark Hennings - 4 years, 12 months ago

one more question on contours... ;) it was nice

Aman Rajput - 4 years, 12 months ago

Log in to reply

Thank you.

Mark Hennings - 4 years, 12 months ago
Aman Rajput
Jun 18, 2016

Let C f ( z ) d z = C e e i z z d z \displaystyle \int_Cf(z)dz=\int_C \frac{e^{e^{iz}}}{z} dz

where C C is the contour as shown in figure

Using the Cauchy's theorem :

r R f ( x ) d x + γ f ( z ) d z + R r f ( x ) d x + Γ f ( z ) d z = 0 \displaystyle \int\limits_{r}^{R}f(x)dx + \int_{\gamma}f(z)dz+\int\limits_{-R}^{-r}f(x)dx + \int_{\Gamma}f(z)dz = 0

Now, consider Γ f ( z ) d z = 0 π e e i R e i θ R e i θ R i e i θ d θ \displaystyle \int_{\Gamma}f(z)dz = \int\limits_{0}^{\pi}\frac{e^{e^{iRe^{i\theta}}}}{Re^{i\theta}}Rie^{i\theta} d\theta

As R R \to \infty

0 π i e e i R e i θ d θ i π \displaystyle \int\limits_{0}^{\pi} ie^{e^{iRe^{i\theta}}}d\theta \to i\pi

Similarly, as r 0 r \to 0

π 0 i e e i r e i θ d θ e i π \displaystyle \int\limits_{-\pi}^{0} ie^{e^{ire^{i\theta}}}d\theta \to -ei\pi

Substituting these back into Cauchy's theorem, we get

f ( x ) d x = i π ( e 1 ) \displaystyle \int\limits_{-\infty}^{\infty} f(x)dx = i\pi(e-1)

or 0 e e i x x d x = i π 2 ( e 1 ) \displaystyle \int\limits_{0}^{\infty} \frac{e^{e^{ix}}}{x} dx = i\frac{\pi}{2}(e-1)

On expanding left hand side , and comparing imaginary parts, we can easily get

0 e cos x sin ( sin x ) x d x = π 2 ( e 1 ) \displaystyle \int\limits_{0}^{\infty} \frac{e^{\cos x}\sin(\sin x)}{x} dx = \frac{\pi}{2}(e-1)

Hence , the answer is 2111 \boxed{2111}

That the limit of the integral on the radius R R semicircle converges to i π i\pi is not at all obvious - controlling that integral was the largest part of my proof - perhaps some details are necessary...

Mark Hennings - 4 years, 12 months ago

Log in to reply

okay... let me edit again ... i want to save my time in typing :D

Aman Rajput - 4 years, 12 months ago

but let me tell here... i used,.. modulus concept to converge it.... i find it very simpple thats why i did not add while typing... i will add by night..

Aman Rajput - 4 years, 12 months ago

Log in to reply

Your integrand is my f ( z ) + 1 z f(z) + \tfrac{1}{z} . The integral of 1 z \tfrac{1}{z} along the contour gives π i \pi i , and the integral of f ( z ) f(z) along the contour tends to zero, as I proved. The argument is easy, but perhaps not obvious to a general reader, without at least a hint!

Mark Hennings - 4 years, 12 months ago

A solution using the "Feynmann trick":

a > 0 \forall a> 0 , let us define the real valued function f ( a ) = 0 e a x e cos x sin ( sin x ) x d x f(a)=\int_{0}^{\infty}e^{-a x}e^{\cos x}\frac{\sin (\sin x)}{x} dx Then we can write d f d a = 0 e a x e cos x sin ( sin x ) d x \frac{df}{da}=-\int_0^{\infty}e^{-a x}e^{\cos x}{\sin (\sin x)} dx Now note that we can write, d f ( a ) d a = ( 0 e ( a x + e i k x ) d x ) = ( g ( a ) ) \begin{aligned} -\frac{df(a)}{da}=& \Im\left(\int_0^\infty {e^{(-a x +e^{ikx})}} dx\right)=\Im(g(a)) \end{aligned} Furthermore, g ( a ) = 0 k 0 e ( a + i k ) x k ! d x = k 0 1 k ! 0 e ( a + i k ) x \begin{aligned} g(a)=& \int_{0}^{\infty}\sum_{k\ge 0}\frac{e^{(-a+ik)x}}{k!} dx\\ \ =&\sum_{k\ge 0}\frac{1}{k!}\int_0^\infty {e^{(-a+ik)x}} \end{aligned} The second step can be justified using Fubini's theorem . Now, note that h ( a ) : = 0 e ( a + i k ) x d x = 1 a i k \begin{aligned} h(a): &= \int_0^\infty {e^{(-a+ik)x}}dx\\ \ &=\frac{1}{a-ik} \end{aligned} which follows since the integral converges for a > 0 a>0 . Then, it follows, after some algebraic manipulation, and an application of the Monotone Convergence Theorem d f ( a ) d a = k 0 1 k ! ( ( k + 1 ) 2 + a 2 ) f ( a ) = k 0 1 ( k + 1 ) ! arctan ( a k + 1 ) + C \begin{aligned} -\frac{df(a)}{da}= & \sum_{k\ge 0}\frac{1}{k!((k+1)^2+a^2)}\\ \implies f(a)= & -\sum_{k\ge 0}\frac{1}{(k+1)!}\arctan\left(\frac{a}{k+1}\right)+C \end{aligned} where C C is the integration constant. Now note that lim a f ( a ) = 0 = π 2 ( e 1 ) + C C = π ( e 1 ) 2 \lim_{a\to \infty}f(a)=0=-\frac{\pi}{2}(e-1)+C\implies C=\frac{\pi(e-1)}{2} . Finally, note that f ( a ) f(a) is a monotonically decreasing function of a a and is lower bounded by the desired expression. Thus using Bounded Convergence Theorem , we get the evaluation of the expression as π ( e 1 ) 2 \frac{\pi(e-1)}{2} , which yields the answer 2111 \boxed{2111} .

I am afraid there are big problems with this. The first that I can see comes from your introduction of the "function(s)" g ( a ) g(a) and h ( a ) h(a) , where you write: g ( a ) = 0 e a x + e i k x x d x g(a) \; = \; \int_0^\infty \frac{e^{-ax + e^{ikx}}}{x}\,dx h ( a ) = 0 e a x + i k x x d x h(a) \; = \; \int_0^\infty \frac{e^{--ax + ikx}}{x}\,dx None of these integrals exist, since their integrands are all O ( x 1 ) O(x^{-1}) near x = 0 x=0 . Given the nonexistence of the constituent integrals, you cannot appeal to Fubini's Theorem.

As a minor point, your definition of f f omits a factor of x 1 x^{-1} .

Mark Hennings - 4 years, 10 months ago

Log in to reply

@Mark Hennings , thanks for pointing out those major defects in the proof, for which I should have been more careful. I have modified the proof a bit hoping to make it error-free. I hope the subtle points are also taken care of. Please let me know if any point still needs to be given special care.

Samrat Mukhopadhyay - 4 years, 10 months ago

Log in to reply

That's much better. This is now a good proof. (+1)

Mark Hennings - 4 years, 10 months ago

Log in to reply

@Mark Hennings @Mark Hennings Thanks.

Samrat Mukhopadhyay - 4 years, 10 months ago
Abhi Kumbale
Jul 12, 2016

This is a good go. You need to look at my discussion with Julian Poon, who came up with much the same solution, since your argument needs work to get precise.

Mark Hennings - 4 years, 11 months ago
Hasan Kassim
Jul 8, 2016

Maybe the best way of solving this integral is by complex analysis; I chose the elementary method: Let:

I = 0 e cos x sin ( sin x ) d x x \displaystyle I= \int_0^{\infty} e^{\cos x} \sin (\sin x) \frac{dx}{x}

Recalling the definitions and properties of hyperbolic sine and hyperbolic cosine :

cosh x + sinh x = e x = > cosh ( cos x ) + sinh ( cos x ) = e cos x \displaystyle \cosh x + \sinh x = e^x \;\;\;\; => \;\; \cosh (\cos x ) + \sinh (\cos x) = e^{\cos x}

Therefore:

I = 0 cosh ( cos x ) sin ( sin x ) d x x + 0 sinh ( cos x ) sin ( sin x ) d x x \displaystyle I = \int_0^{\infty} \cosh (\cos x ) \sin (\sin x) \frac{dx}{x} + \int_0^{\infty} \sinh (\cos x) \sin (\sin x) \frac{dx}{x}

Now consider ( For real numbers a a and b b ) :

sin ( a + i b ) = sin a cos i b + cos a sin i b = sin ( a ) cosh ( b ) + i cos ( a ) sinh ( b ) \displaystyle \sin (a+ib) = \sin a \cos ib + \cos a \sin ib = \sin (a) \cosh (b) + i\cos (a) \sinh (b)

cos ( a + i b ) = cos a cos i b sin a sin i b = cos ( a ) cosh ( b ) i sin ( a ) sinh ( b ) \displaystyle \cos (a+ib) = \cos a \cos ib - \sin a \sin ib = \cos (a) \cosh (b) - i\sin (a) \sinh (b)

Therefore:

sin ( a ) cosh ( b ) = sin ( a + i b ) and sin ( a ) sinh ( b ) = cos ( a + i b ) \displaystyle \sin (a) \cosh (b) = \Re \sin (a+ib) \;\;\;\;\;\text{and}\;\;\;\;\;\; \sin (a) \sinh (b) = - \Im \cos (a+ib)

Hence :

sin ( sin x ) cosh ( cos x ) = sin ( sin x + i cos x ) = sin ( i e i x ) \displaystyle \sin (\sin x) \cosh (\cos x) = \Re \sin (\sin x+i\cos x) = \Re \sin ( i e^{-ix} )

And,

sin ( sin x ) sinh ( cos x ) = cos ( sin x + i cos x ) = cos ( i e i x ) \displaystyle \sin (\sin x) \sinh (\cos x) = \Im \cos (\sin x+i\cos x) = \Im \cos ( i e^{-ix} )

Now substitute these in the expression of I I :

I = 0 sin ( i e i x ) d x x + 0 cos ( i e i x ) d x x \displaystyle I = \Re \int_0^{\infty} \sin ( i e^{-ix} ) \frac{dx}{x} + \Im \int_0^{\infty} \cos ( i e^{-ix} ) \frac{dx}{x}

Using Taylor series expansion of the sine and the cosine functions:

I = 0 n = 0 ( 1 ) n ( i e i x ) 2 n + 1 ( 2 n + 1 ) ! d x x + 0 m = 0 ( 1 ) m ( i e i x ) 2 m ( 2 m ) ! d x x \displaystyle I = \Re \int_0^{\infty} \sum_{n=0}^{\infty} (-1)^n \frac{(i e^{-ix})^{2n+1}}{(2n+1)!} \frac{dx}{x} + \Im \int_0^{\infty} \sum_{m=0}^{\infty} (-1)^m \frac{(i e^{-ix})^{2m}}{(2m)!} \frac{dx}{x}

Note that i 2 m = ( 1 ) m i^{2m} = (-1)^m and i = e i π 2 i=e^{i\frac{\pi}{2}} :

I = 0 n = 0 e i ( π 2 x ( 2 n + 1 ) ) ( 2 n + 1 ) ! d x x + 0 m = 0 e i x ( 2 m ) ( 2 m ) ! d x x \displaystyle I = \Re \int_0^{\infty} \sum_{n=0}^{\infty} \frac{e^{i(\frac{\pi}{2} -x(2n+1)})}{(2n+1)!} \frac{dx}{x} + \Im \int_0^{\infty} \sum_{m=0}^{\infty} \frac{e^{-ix(2m)}}{(2m)!} \frac{dx}{x}

taking the real and the imaginary parts :

I = 0 n = 0 sin ( ( 2 n + 1 ) x ) ( 2 n + 1 ) ! d x x + 0 m = 0 sin ( 2 m x ) ( 2 m ) ! d x x \displaystyle I = \int_0^{\infty} \sum_{n=0}^{\infty} \frac{\sin ((2n+1)x)}{(2n+1)!} \frac{dx}{x} + \int_0^{\infty} \sum_{m=0}^{\infty} \frac{\sin (2mx)}{(2m)!} \frac{dx}{x}

Now using Fubini's Theorem:

I = n = 0 1 ( 2 n + 1 ) ! 0 sin ( ( 2 n + 1 ) x ) x d x + m = 0 1 ( 2 m ) ! 0 sin ( 2 m x ) x d x \displaystyle I = \sum_{n=0}^{\infty} \frac{1}{(2n+1)!} \int_0^{\infty} \frac{\sin ((2n+1)x)}{x} dx + \sum_{m=0}^{\infty} \frac{1}{(2m)!} \int_0^{\infty} \frac{\sin (2mx)}{x} dx

And use the well-known Dirichlet Integral :

I = n = 0 1 ( 2 n + 1 ) ! ( π 2 ) + m = 1 1 ( 2 m ) ! ( π 2 ) \displaystyle I = \sum_{n=0}^{\infty} \frac{1}{(2n+1)!} (\frac{\pi}{2} ) + \sum_{m=1}^{\infty} \frac{1}{(2m)!} (\frac{\pi}{2} )

Note that we omitted the m = 0 m=0 term of the second summation , because the term will disappear before performing the Dirichlet Integral. Now it is only the Taylor series of hyperbolic sine and hyperbolic cosine:

I = π 2 ( sinh ( 1 ) + cosh ( 1 ) 1 ) = 1 2 π 1 ( 1 e 1 ) \displaystyle I= \frac{\pi}{2} (\sinh (1) + \cosh (1) - 1) = \boxed{\frac{1}{2} \pi ^1 (1e -1) }

Before you can use Fubini's Theorem to reverse the order of summation and integration, you need to know that the integrand is doubly Lebesgue integrable. It is not. We are talking about infinite Riemann integral here, and Fubini's Theorem does not apply. Your argument is formally interesting, but needs work to make it precise.

Mark Hennings - 4 years, 11 months ago

In essence, what you are doing is like what Julian Poon did, and his and my discussion showed how his approach could be made to work.

Mark Hennings - 4 years, 11 months ago

Log in to reply

Oh I didn't see the comments below your solution. But still I don't know the exact condition for which one can interchange the sum and integral. Can you tell or explain a little more?

Hasan Kassim - 4 years, 11 months ago

0 pending reports

×

Problem Loading...

Note Loading...

Set Loading...