More fun in 2015, Part 18

Algebra Level 5

Let w 1 , . . , w 2014 w_1,..,w_{2014} be the 2015th roots of unity other than 1. Without resorting to calculus, find k = 1 2014 1 ( 1 + w k ) 2 -\sum_{k=1}^{2014}\frac{1}{(1+w_k)^2}


The answer is 1014049.

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.

2 solutions

Otto Bretscher
Sep 13, 2015

Since we are not allowed to use calculus, we will present a somewhat lengthy algebraic solution.

We will at first include the root of unity w 2015 = 1 w_{2015}=1 . At the end we need to remember to add 1 / 4 1/4 to our answer.

Let's make a change of variables, z = 1 1 + w z=\frac{1}{1+w} . Since the w k w_k are the solutions of w 2015 = 1 w^{2015}=1 , the corresponding z k z_k will be the solutions of ( 1 z z ) 2015 = 1 \left(\frac{1-z}{z}\right)^{2015}=1 or z 2015 + ( z 1 ) 2015 = 0 z^{2015}+(z-1)^{2015}=0 or 2 z 2015 2015 z 2014 + 2015 × 2014 2 z 2013 . . . = 0 2z^{2015}-2015z^{2014}+\frac{2015\times{2014}}{2}z^{2013}-...=0 . We seek k = 1 2015 z k 2 = ( k = 1 2015 z k ) 2 2 i < j z i z j \sum_{k=1}^{2015}z_k^2=\left(\sum_{k=1}^{2015}z_k\right)^2-2\sum_{i<j}z_iz_j = ( 2015 2 ) 2 2 ( 2015 × 2014 4 ) = 1014048.75 =\left(\frac{2015}{2}\right)^2-2\left(\frac{2015\times{2014}}{4}\right)=-1014048.75 , by Viete. Changing the sign, as required, and adding the 1 / 4 1/4 stemming form w 2015 = 1 w_{2015}=1 , we find the final answer, 1014049 \boxed{1014049} .

Moderator note:

I don't think this is lengthy. Change of roots is the approach that I used to take before learning Calculus.

did a silly mistake and got the answer -2011.. Nice problem

Dev Sharma - 5 years, 5 months ago

Atlast solved 3rd attempt (phew)!!

Same method!!

One more thing to tell you that using calculus I couldn't get the answer because P ( 1 ) P''(-1) was too lengthy to calculate than this method.

However, it was just hard to find what numberwe have added which is infact 1 2 \frac{1}{2} in our solution.

Ravi Dwivedi - 5 years, 5 months ago

I'm sorry. I resorted to calculus. Nice problem and a nicer solution! Could have never thought of such an approach! You should write a book.

Kartik Sharma - 5 years, 9 months ago

Log in to reply

Thank you for your kind words. Actually, I have published a math text, "Linear Algebra with Applications", with some nice problems (I hope), and I'm writing a calculus text. Thanks for the encouragement!

Otto Bretscher - 5 years, 9 months ago

Log in to reply

I will be looking forward to it. Well, where can I get your Algebra text? Is there a PDF available online? I couldn't get one.

Kartik Sharma - 5 years, 9 months ago

Log in to reply

@Kartik Sharma The publisher (Pearson) is hoping that there are no (free) online files available (although there probably are, "underground"). The text is now in its fifth edition; older editions are available at a reasonable price. There is also an Indian ("sub-continental") and a Chinese edition. The text is meant for an introductory (freshman or sophomore) level linear algebra course, and I doubt that a man of your sophistication could learn much from it ;)

Otto Bretscher - 5 years, 8 months ago

Could you explain how to use calculus to solve this?

Benson Li - 5 years, 8 months ago
Benson Li
Sep 18, 2015

Let us first consider the following manipulation: 1 ( 1 + e 2 i z ) 2 = e 2 i z ( e i z ) 2 ( 1 + e 2 i z ) 2 \dfrac{1}{(1+e^{2 i z})^2}= \dfrac{e^{-2 i z}}{(e^{-i z})^2(1+e^{2 i z})^2} = e 2 i z ( e i z + e i z ) 2 \hspace{17mm} =\dfrac{e^{-2 i z}}{(e^{-i z}+e^{i z})^2} = e 2 i z 2 c o s 2 ( z ) = 1 4 [ c o s ( 2 z ) c o s 2 ( z ) i s i n ( 2 z ) c o s 2 ( z ) ] \hspace{17mm} =\dfrac{e^{-2 i z}}{2cos^2(z)}=\frac{1}{4} [\frac{cos(2z)}{cos^2(z)}-i \frac{sin(2z)}{cos^2(z)}] = 1 4 ( 2 sec 2 ( z ) 2 i tan ( z ) ) \hspace{17mm}=\frac{1}{4} (2-\sec^2(z)-2 i \tan(z)) = 1 4 ( 1 tan 2 ( z ) 2 i tan ( z ) ) \hspace{17mm}=\frac{1}{4} (1-\tan^2(z)-2 i \tan(z)) Because the roots of unity can be written as w k = e 2 i π k 2015 w_k=e^{\frac{2 i \pi k}{2015}} , our question can be rephrased as the following sum, with z = π k 2015 z=\frac{\pi k}{2015} : 1 4 ( k = 1 2014 1 k = 1 2014 tan 2 ( π k 2015 ) 2 i k = 1 2014 tan ( π k 2015 ) ) \frac{1}{4}(\sum_{k=1}^{2014} 1-\sum_{k=1}^{2014}\tan^2(\frac{\pi k}{2015})-2i\sum_{k=1}^{2014}\tan(\frac{\pi k}{2015})) Now, lets first look at the imaginary part of our problem, namely k = 1 2014 tan ( π k 2015 ) \sum_{k=1}^{2014}\tan(\frac{\pi k}{2015}) . . Using the identity tan ( π k 2015 ) = t a n ( π π k 2015 ) = tan ( π ( 2015 k ) 2015 ) ( 1 ) \tan(\frac{\pi k}{2015})=-tan(\pi-\frac{\pi k}{2015})=-\tan(\frac{\pi(2015-k)}{2015}) \hspace{5mm} (1) , we see that that k = 1 2014 tan ( π k 2015 ) = 0 \sum_{k=1}^{2014} \tan(\frac{\pi k}{2015})=0 , as the terms can be paired off to cancel out

The real part is 1 4 k = 1 2014 ( 1 tan 2 ( π k 2015 ) ) \frac{1}{4}\sum_{k=1}^{2014} (1-\tan^2(\frac{\pi k}{2015})) . Using ( 1 ) (1) we have: 1 4 ( k = 1 2014 1 2 k = 1 1007 tan 2 ( π k 2015 ) ) \frac{1}{4}(\sum_{k=1}^{2014} 1-2\sum_{k=1}^{1007}\tan^2(\frac{\pi k}{2015})) . Using the following identity, with m=1007: k = 1 m tan 2 ( π k 2 m + 1 ) = 2 m 2 + m \sum_{k=1}^{m}\tan^2(\frac{\pi k}{2m+1})=2m^2+m , we finally have: 1 4 ( 2014 2 ( 2 × 100 7 2 + 1007 ) = 100 7 2 = 1014049 \frac{1}{4}(2014-2(2\times 1007^2+1007)=-1007^2=-1014049

Moderator note:

Arguably, the work of this problem is to prove the s u m tan 2 θ i sum \tan^2 \theta_i identity.

One of the easiest ways to find that sum, is to find a polynomial whose roots are tan 2 θ i \tan ^2 \theta_i , by using the fact that the imaginary part of ( cos θ i + i sin θ i ) 2015 = 0 (\cos \theta_i + i \sin \theta_i) ^{2015} = 0 .

It's good to see a different approach! (up-vote) Where did you get the formula k = 1 m tan 2 . . . \sum_{k=1}^m\tan^2... from?

Otto Bretscher - 5 years, 8 months ago

Log in to reply

I initially just took base cases with low values of m, in which both the upper limit of sum and the denominator in t a n 2 tan^2 changed appropriately, and found that they have this recursive structure a n + 1 = a n + 7 + 4 ( n 1 ) , a 1 = 3 a_{n+1}=a_n+7+4*(n-1),a_1=3 , where a n a_n corresponds with the sum k = 1 n t a n 2 ( π k 2 n + 1 ) \sum_{k=1}^{n}tan^2(\frac{\pi k}{2n+1}) and I solved for the recursion, hoping that the structure holds for higher values of m. Then to check, I googled it and confirmed my result with http://math.stackexchange.com/questions/217240/reference-for-a-tangent-squared-sum-identity (page doesn't really provide a proof though)

Benson Li - 5 years, 8 months ago

Log in to reply

Also, can you explain how to do this problem with calculus?

Benson Li - 5 years, 8 months ago

Log in to reply

@Benson Li Write ( z w k ) = z n 1 \prod(z-w_k)=z^n-1 , take the logarithm, differentiate twice, and plug in z = 1 z=-1 .

Otto Bretscher - 5 years, 8 months ago

Log in to reply

@Otto Bretscher Alright thanks!

Benson Li - 5 years, 8 months ago

Wow, so you rediscovered this formula by yourself?! I'm usually being lazy and I'm looking these things up in Jolley's excellent "Summation of Series"...it's available online. His formulas are usually much more general.

Otto Bretscher - 5 years, 8 months ago

Log in to reply

@Otto Bretscher I mean, its not like I proved the recursive structure. That's why I said "hope it holds for higher values" :P.

Benson Li - 5 years, 8 months ago

I guess we have a proof of this formula now, based on this problem ;)

Otto Bretscher - 5 years, 8 months ago

0 pending reports

×

Problem Loading...

Note Loading...

Set Loading...