Curious Series

Calculus Level 5

k = 1 sin ( k ) k \large \sum_{k=1}^{\infty}\frac{\sin(k^\circ)}{k}

If the value of the series above is of the form a π b \dfrac{a\pi}{b} , where a a and b b are coprime positive integers, find the value of a + b a+b .


The answer is 539.

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.

4 solutions

By the Dirichlet test for convergence we can anticipate that the series does converge.

We'll look first in general at S ( x ) = k = 1 sin ( k x ) k , S(x) = \displaystyle\sum_{k=1}^{\infty} \dfrac{\sin(kx)}{k}, and then evaluate S ( π 180 ) . S(\frac{\pi}{180}).

Now we have that sin ( y ) = e i y e i y 2 i , \sin(y) = \dfrac{e^{iy} - e^{-iy}}{2i}, so

S ( x ) = k = 1 e i k x e i k x 2 k i = ln ( 1 e i x ) + ln ( 1 e i x ) 2 i , S(x) = \displaystyle\sum_{k=1}^{\infty} \dfrac{e^{ikx} - e^{-ikx}}{2ki} = \dfrac{-\ln(1 - e^{ix}) + \ln(1 - e^{-ix})}{2i},

since in general k = 1 z k k = ln ( 1 z ) . \displaystyle\sum_{k=1}^{\infty} \dfrac{z^{k}}{k} = -\ln(1 - z).

(I'm playing a bit fast and loose with radius of convergence in C \mathbb{C} here; I'll try and deal with that later.)

Continuing, we can then say that

S ( x ) = 1 2 i ln ( 1 e i x 1 e i x ) = 1 2 i ln ( e i x ) = 1 2 i ln ( e i ( π x ) ) = i ( π x ) 2 i = π x 2 , S(x) = \dfrac{1}{2i}\ln\left(\dfrac{1 - e^{-ix}}{1 - e^{ix}}\right) = \dfrac{1}{2i}\ln(-e^{-ix}) = \dfrac{1}{2i}\ln(e^{i(\pi - x)}) = \dfrac{i(\pi - x)}{2i} = \dfrac{\pi - x}{2},

since e i x e i x 1 e i x 1 e i x = e i x ( 1 e i x ) 1 e i x = e i x , \dfrac{-e^{-ix}}{-e^{-ix}} * \dfrac{1 - e^{-ix}}{1 - e^{ix}} = \dfrac{-e^{-ix}(1 - e^{-ix})}{1 - e^{-ix}} = -e^{-ix}, and since

e i x = cos ( x ) + i sin ( x ) = cos ( π x ) + i sin ( π x ) = e i ( π x ) . -e^{-ix} = -\cos(x) + i*\sin(x) = \cos(\pi - x) + i*\sin(\pi - x) = e^{i(\pi - x)}.

Finally, S ( π 180 ) = π π 180 2 = 179 π 360 , S(\frac{\pi}{180}) = \dfrac{\pi - \frac{\pi}{180}}{2} = \dfrac{179 \pi}{360}, and thus a + b = 179 + 360 = 539 . a + b = 179 + 360 = \boxed{539}.

Very nice! This is roughly what I had in mind.

Just to clarify: For which x x does your formula S ( x ) = π x 2 S(x)=\frac{\pi-x}{2} hold?

Otto Bretscher - 5 years, 8 months ago

Log in to reply

Regarding the series convergence issue, I suppose that in place of z z we could look at z e δ ze^{-\delta} as δ 0 + \delta \rightarrow 0^{+} and let that ripple through the steps of the solution. I think it will then filter out at the end. We could choose the l n ln branch to be defined for r > 0 , π < θ < π , r \gt 0, -\pi \lt \theta \lt \pi, in which case the formula would hold for x ( 0 , π ) . x \in (0, \pi). We could throw in π \pi for good measure since then all the terms of the original series would be 0 , 0, making S ( π ) = 0 S(\pi) = 0 as the formula would indicate.

Brian Charlesworth - 5 years, 8 months ago

Log in to reply

I tried to deal with the convergence issue in my solution... I'm curious to hear whether you find that approach satisfactory.

If you choose θ \theta between 0 and 2 π 2\pi , then the formula holds in that interval, I believe. S ( x ) S(x) is periodic, of course, with discontinuities at the integer multiples of 2 π 2\pi .

Otto Bretscher - 5 years, 8 months ago

Log in to reply

@Otto Bretscher Ah, yes, the minus sign ...... 0 < θ < 2 π 0 \lt \theta \lt 2\pi and the same for x x it is, then.

Your solution is far more elegant than mine and deftly avoids the convergence issue. Thanks for reacquainting me with Abel's Theorem. :)

Brian Charlesworth - 5 years, 8 months ago

Log in to reply

@Brian Charlesworth Shall we ask the same question about the cosine, maybe in radians this time? If we do, the two of us should probably let somebody else answer first? ;)

Otto Bretscher - 5 years, 8 months ago

Log in to reply

@Otto Bretscher Haha Yes, by all means, that would be a great question for you to post, and I will restrain myself from posting a solution. :)

Perhaps you could frame it as k = 1 cos ( k x ) k \displaystyle\sum_{k=1}^{\infty} \dfrac{\cos(kx)}{k} for, say, x = π 4 . x = \dfrac{\pi}{4}.

Or even n = 1 6 k = 1 cos ( k π n ) k , \displaystyle\sum_{n=1}^{6} \sum_{k=1}^{\infty} \dfrac{\cos\left(\dfrac{k\pi}{n}\right)}{k}, which curiously comes out to within 0.1 0.1 % of a certain special number.

Brian Charlesworth - 5 years, 8 months ago

@Otto Bretscher Sorry for the lack of rigor here; my complex analysis is a bit rusty, so I was glossing over some important details regarding convergence, etc., hoping that it would all pan out regardless, (which it did, I suppose). You may have had a non-complex method in mind, so I'll give some thought to what such a method might look like. Great question, anyway. :)

Brian Charlesworth - 5 years, 8 months ago

Sorry, I'm not familiar with Dirichlet test for convergence so I have to ask: What are the a n a_n and b n b_n for this test?

Pi Han Goh - 5 years, 8 months ago

Log in to reply

a n = 1 n a_n=\frac{1}{n} and b n b_n is the sine. In this particular case, one could certainly also use the Leibniz criterion for 180 terms at a time.

Otto Bretscher - 5 years, 8 months ago

Log in to reply

Oh I was thinking of b n = 1 n b_n= \frac1n and a n = sin ( n ) a_n = \sin(n) . Silly me. ThankYOU. Now lemme read your Abel's theorem solution.

Pi Han Goh - 5 years, 8 months ago

Yeah, did the same. But I can bet that @Otto Bretscher has something brilliant here considering that the answer is too nice even.

Kartik Sharma - 5 years, 8 months ago

Log in to reply

Brian's solution is brilliant... ;)

Otto Bretscher - 5 years, 8 months ago

Correct! Quite an ordinary mathematics but the finding is special. Good choice with Pi/ 180. This made me thought that the setter might have made a mistake. Thanks for this!

Lu Chee Ket - 5 years, 8 months ago

I d i d n t q u i t e g e t h o w y o u w e n t f r o m S ( x ) = k = 1 e i k x e i k x 2 k i t o S ( x ) = ln ( 1 e i x ) + ln ( 1 e i x ) 2 i . \mathrm {I~didn't~quite~get~how~you~went~from~}{S(x) = \displaystyle \sum_{k=1}^\infty \frac {\mathrm{e}^{ikx} - \mathrm{e}^{-ikx}}{2ki}}~ \mathrm {to~}\\{S(x) = \dfrac{-\ln(1 - \mathrm{e}^{ix}) + \ln(1 - \mathrm{e}^{-ix})}{2i}} \mathrm{.}

Ishaan Arora - 5 years, 8 months ago

Log in to reply

Using the general formula k = 1 z k k = ln ( 1 z ) , \displaystyle\sum_{k=1}^{\infty} \dfrac{z^{k}}{k} = -\ln(1 - z), we can in turn set z = e i x z = e^{ix} and z = e i x z = e^{-ix} to find that

S ( x ) = 1 2 i ( k = 1 ( e i x ) k k k = 1 ( e i x ) k k ) = S(x) = \dfrac{1}{2i}\left(\displaystyle\sum_{k=1}^{\infty}\dfrac{(e^{ix})^{k}}{k} - \sum_{k=1}^{\infty} \dfrac{(e^{-ix})^{k}}{k}\right) =

1 2 i ( ln ( 1 e i x ) ( ln ( 1 e i x ) ) = ln ( 1 e i x ) + ln ( 1 e i x ) 2 i . \dfrac{1}{2i}(-\ln(1 - e^{ix}) - (-\ln(1 - e^{-ix})) = \dfrac{-\ln(1 - e^{ix}) + \ln(1 - e^{-ix})}{2i}.

Brian Charlesworth - 5 years, 8 months ago

Log in to reply

T h a n k s a t o n . \mathrm{Thanks~a~ton.}

Ishaan Arora - 5 years, 8 months ago

Log in to reply

@Ishaan Arora A n d m a y I a s k w h y y o u c h o s e t h e v a l u e π 180 ? \mathrm{And~may~I~ask~why~you~chose~the~value~} \dfrac{ \pi}{180}\mathrm{?}

Ishaan Arora - 5 years, 8 months ago

Log in to reply

@Ishaan Arora This is to convert k k^{\circ} into radians, which is the required measure for the formulas used in my solution.

Brian Charlesworth - 5 years, 8 months ago

My first two tries went in vain as i was trying to fluke as i haven't saw such type of questions . But suddenly I got the idea to solve and i forgot the denominator and got answer as 359

Aakash Khandelwal - 5 years, 8 months ago
Otto Bretscher
Oct 2, 2015

I like Brian's solution! However, Brian is rightly concerned about convergence issues; perhaps we can avoid those problems by considering a power series.

Let's start with the geometric series k = 1 e i k a x k = e i a x 1 e i a x \sum_{k=1}^{\infty}e^{ika}x^k=\frac{e^{ia}x}{1-e^{ia}x} where 0 < x < 1 0<x<1 . Now we can divide by x x and integrate to find that k = 1 e i k a k x k = ln ( 1 e i a x ) \sum_{k=1}^{\infty}\frac{e^{ika}}{k}x^k=-\ln(1-e^{ia}x) We can take the limit as x x approaches 1 from below and use Abel's Theorem to see that k = 1 e i k a k = ln ( 1 e i a ) \sum_{k=1}^{\infty}\frac{e^{ika}}{k}=-\ln(1-e^{ia}) Considering the imaginary parts, we end up with k = 1 sin ( k a ) k = arg ( 1 e i a ) \sum_{k=1}^{\infty}\frac{\sin(ka)}{k}=-\arg(1-e^{ia}) If a a is between 0 and 2 π 2\pi , the RHS is π a 2 \frac{\pi-a}{2} ; perhaps the fastest way to see this is to draw the isosceles triangle with its vertices at 0 , 1 0,1 and e i a e^{ia} in the Argand plane. For a = π 180 a=\frac{\pi}{180} the sum is 179 π 360 \frac{179\pi}{360} and the required answer is 539 \boxed{539}

Moderator note:

Abel's theorem is a good way of finding the limit of a power series at the boundary of the radius of convergence, along a straight line from the origin. Note that we do not know if the sequence will converge if we take a different path in the complex plane to that point.

Wait. You have only shown that x approaches 1 from below . Shouldn't you prove that it approaches 1 from above as well?

Pi Han Goh - 5 years, 8 months ago

Log in to reply

No, Abel's convergence theorem is really all about approaching 1 from below, from within the radius of convergence.

Otto Bretscher - 5 years, 8 months ago

Log in to reply

But shouldn't we prove that left hand limit = right hand limit, so only then we know whether the limit exist?

EDIT: After reading the Remarks section , I think my doubt has been resolved. But I still don't understand why that's the case. Either way: Thanks for your help ahah!

Pi Han Goh - 5 years, 7 months ago

Log in to reply

@Pi Han Goh Abel's Theorem, by its very nature, is about a single-sided limit; the power series does not converge on the other side. Consider the most basic example, ln ( 1 + x ) = k = 1 ( 1 ) k + 1 x k k \ln(1+x)=\sum_{k=1}^{\infty}(-1)^{k+1}\frac{x^k}{k} for x < 1 |x|<1 . You take lim x 1 \lim_{x\to 1^{-}} to find the sum of the alternating harmonic series.

Otto Bretscher - 5 years, 7 months ago

Sir i did it by evaluating the imaginary part of the summation of (e^ix)/x (at x=1) by using taylor series and using integration to find the answer.....i arrived at pi - 0.5 as the answer.....now was i wrong or did i do the mistake of not converting into degrees

Arghyadeep Chatterjee - 2 years, 4 months ago

Not that obvious compared to Brian's but I can feel the truth which is correct.

Lu Chee Ket - 5 years, 8 months ago

Let us define, for z R z\in \mathbb{R} , the series A ( z ) = k 1 z k sin ( a k ) k , a = π 180 A(z)=\sum_{k\ge 1}\frac{z^k\sin (ak)}{k},\ a=\frac{\pi}{180} We can see that for z ( 1 , 1 ) z\in (-1,1) the series is absolutely summable and convergent. Also, I think by Monotone convergence theorem, if follows that lim z 1 A ( z ) = k 1 sin ( a k ) k \lim_{z\to 1}A(z)=\sum_{k\ge 1}\frac{\sin(ak)}{k} though it might need some care. Assuming it is correct, now we can proceed to find an expression for A ( z ) A(z) . Note that z ( 1 , 1 ) z\in (-1,1)\implies A ( z ) = k 1 z k exp ( i a k ) k d A d z = k 1 z k 1 exp ( i a k ) = 1 z k 1 z exp ( i a k ) z = sin a ( z cos a ) 2 + sin 2 a A ( z ) = arctan ( z cos a sin a ) + C A(z)=\Im\sum_{k\ge 1}\frac{z^k\exp(iak)}{k}\\\implies \frac{dA}{dz}=\Im\sum_{k\ge 1}z^{k-1}\exp(iak)\\=\frac{1}{z}\Im\sum_{k\ge 1}\frac{z}{\exp(-iak)-z}\\=\frac{\sin a}{(z-\cos a)^2+\sin^2 a}\\\implies A(z)=\arctan\left(\frac{z-\cos a}{\sin a}\right)+C Now, A ( 0 ) = 0 C = arctan cot ( a ) = π 2 a A(0)=0\implies C=\arctan\cot(a)=\frac{\pi}{2}-a . Hence A ( z ) = arctan ( z cos a sin a ) + 88 π 180 lim z 1 A ( z ) = arctan ( 1 cos a sin a ) + 88 π 180 = π 360 + 88 π 180 = 179 π 360 A(z)=\arctan\left(\frac{z-\cos a}{\sin a}\right)+\frac{88\pi}{180}\implies \lim_{z\to 1}A(z)=\arctan\left(\frac{1-\cos a}{\sin a}\right)+\frac{88\pi}{180}\\=\frac{\pi}{360}+\frac{88\pi}{180}=\frac{179\pi}{360} So, 539 \boxed{539} is the answer.

@Otto Bretscher Can you verify whether this answer is correct? Thanks.

Pi Han Goh - 5 years, 7 months ago
Tom Engelsman
Oct 2, 2015

Very nice series problem, Otto! Summing an infinite discretized "sinc" pulse.........nice little midnight signal processing problem for EE's!

This is the power of special mathematics.

Lu Chee Ket - 5 years, 8 months ago

0 pending reports

×

Problem Loading...

Note Loading...

Set Loading...