We're Only On Coprime Numbers?

Geometry Level 5

Evaluate

P = k sin ( k π 2016 ) \large P=\prod_{k}\sin\left(\dfrac{k\pi}{2016}\right)

where the product is taken over all k k with 1 k 1008 1\leq k\leq 1008 and gcd ( k , 2016 ) = 1 \gcd(k,2016)=1 .

Write your answer in the form P = a 2 b P=\dfrac{a}{2^b} , where a a and b b are integers, and a a is odd. Enter a + b a+b .

Clariifcation : gcd ( ) \gcd(\cdot) denotes the greatest common divisor function.


Inspiration .


The answer is 289.

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.

3 solutions

Otto Bretscher
Apr 25, 2016

We will find the product for 1 k 2016 1\leq k\leq 2016 and gcd ( k , 2016 ) = 1 \gcd(k,2016)=1 . Let n = 2016 n=2016 , an even number.

We will work with moduli so that we don't have to keep track of messy factors i i and e i t e^{it} . Now

2 ϕ ( n ) P 2 = k 2 sin ( k π / n ) = k e i k π / n e i k π / n = k 1 e 2 i k π / n = Φ n ( 1 ) , 2^{\phi(n)}P^2=\prod_{k}2\sin(k\pi/n)=\prod_{k}|e^{ik\pi/n}-e^{-ik\pi/n}|=\prod_{k}|1-e^{2ik\pi/n}|=|\Phi_n(1)|,

where Φ n ( x ) \Phi_n(x) is the cyclotomic polynomial. We can use induction on n n (or fancier methods) to show that Φ n ( 1 ) = p \Phi_n(1)=p if n = p k n=p^k is a prime power, and Φ n ( 1 ) = 1 \Phi_n(1)=1 otherwise.

Thus P = Φ n ( 1 ) 2 ϕ ( n ) = 1 2 576 / 2 = 1 2 288 P=\sqrt{\frac{\Phi_n(1)}{2^{\phi(n)}}}=\frac{1}{2^{576/2}}=\frac{1}{2^{288}}

and the answer is 289 \boxed{289}

Moderator note:

Good usage of cyclotomic polynomials (and indirectly mobius inversion) to simplify this problem. When you see gcd ( n , k ) = 1 \gcd(n,k) = 1 , chances are mobius inversion is involved.

Yeah! Same solution.

Shourya Pandey - 5 years, 1 month ago
Julian Poon
Apr 26, 2016

Relevant wiki: Dirichlet Convolution

My approach to this problem is less elegant.

Lemma: k = 1 n sin ( k π 2 n ) = n 2 n 1 \prod _{k=1}^n\sin \left(\frac{k\pi }{2n}\right)=\frac{\sqrt{n}}{2^{n-1}}


Proof:

( 2 n k = 1 n sin ( k π 2 n ) ) 2 = ( k = 1 n 2 sin ( k π 2 n ) ) 2 = 2 ( k = 1 2 n 1 2 sin ( k π 2 n ) ) \left(2^n\prod _{k=1}^n\sin \left(\frac{k\pi }{2n}\right)\right)^2=\left(\prod _{k=1}^n2\sin \left(\frac{k\pi }{2n}\right)\right)^2=2\left(\prod _{k=1}^{2n-1}2\sin \left(\frac{k\pi }{2n}\right)\right)

Using a method similar to Otto's, the product is

2 ( k = 1 2 n 1 1 e i k π n ) 2\left(\prod _{k=1}^{2n-1}\left|1-e^{\frac{ik\pi }{n}}\right|\right)

Since though roots of unity,

2 ( k = 1 2 n 1 x e i k π n ) = 2 x e 2 i k π k = 1 2 n x e i k π n = 2 x 2 n 1 x 1 = 2 k = 0 2 n 1 x k 2\left(\prod _{k=1}^{2n-1}\left|x-e^{\frac{ik\pi }{n}}\right|\right)=\frac{2}{\left|x-e^{2ik\pi}\right|}\prod _{k=1}^{2n}\left|x-e^{\frac{ik\pi }{n}}\right|=2\left|\frac{x^{2n}-1}{x-1}\right|=2\left|\sum _{k=0}^{2n-1}x^k\right|

2 ( k = 1 n 1 e i k π n ) ( k = n + 1 2 n 1 1 e i k π n ) = 4 n 2\left(\prod _{k=1}^n\left|1-e^{\frac{ik\pi }{n}}\right|\right)\left(\prod _{k=n+1}^{2n-1}\left|1-e^{\frac{ik\pi }{n}}\right|\right)=4n

Putting it all together gives

k = 1 n sin ( k π 2 n ) = n 2 n 1 \prod _{k=1}^n\sin \left(\frac{k\pi }{2n}\right)=\frac{\sqrt{n}}{2^{n-1}}


Now this is where mobius inversion comes in. Let k = 1 n sin ( k π 2 n ) = F ( n ) \prod _{k=1}^n\sin \left(\frac{k\pi }{2n}\right)=F\left(n\right) and 1 k n , ( n , k ) = 1 sin ( k π 2 n ) = f ( n ) \prod _{1\le k\le n, (n,k)=1}\sin \left(\frac{k\pi }{2n}\right)=f(n)

Notice that

ln ( F ( n ) ) = 1 ln ( f ( n ) ) \ln \left(F\left(n\right)\right)=1 * \ln \left(f\left(n\right)\right)

Where * is the dirichlet convolution. Through mobius inversion,

ln ( f ( n ) ) = μ ln ( F ( n ) ) \ln \left(f\left(n\right)\right)= \mu*\ln \left(F\left(n\right)\right)

Substituting F ( n ) = n 2 n 1 F(n)=\frac{\sqrt{n}}{2^{n-1}} ,

ln ( f ( n ) ) = 1 2 d n μ ( n d ) ln ( d ) ln ( 2 ) ( d n μ ( n d ) d d n μ ( n d ) ) \ln \left(f\left(n\right)\right)=\frac{1}{2}\sum _{d \mid n}^{ }\mu\left(\frac{n}{d}\right)\ln \left(d\right)-\ln \left(2\right)\left(\sum _{d \mid n}^{ }\mu\left(\frac{n}{d}\right)d-\sum _{d \mid n}^{ }\mu\left(\frac{n}{d}\right)\right)

= 1 2 Λ ( n ) ln ( 2 ) ( φ ( n ) ε ) =\frac{1}{2}\Lambda \left(n\right)-\ln \left(2\right)\left(\varphi \left(n\right)-\varepsilon \right)

Where Λ \Lambda is the von Mangoldt function, φ \varphi is euler's totient function and ε \varepsilon is the multiplicative identity.

Substituting n=1008 gives

f ( 1008 ) = 2 288 f\left(1008\right)=\boxed{2^{-288}}

Yes, that works too!(+1) It is good to see different approaches; thanks!

Note that Φ n ( 1 ) = e Λ ( n ) \Phi_n(1)=e^{\Lambda(n)} , meaning that our solutions are not so different after all.

Some small typos in your formula ln ( f ( n ) ) = . . . \ln(f(n))=... : Reverse the rôles of n n and d d in μ \mu (three instances), and you need to subtract ϵ \epsilon , I believe.

Otto Bretscher - 5 years, 1 month ago

Log in to reply

Thanks for telling me, I've edited the solution. I'm not that good at cyclotomic polynomials, though it is a really interesting topic.

Julian Poon - 5 years, 1 month ago

Log in to reply

My solution does not require any "deep" properties of the cyclotomic polynomial. When I came across ( 1 e 2 i k π / n ) \prod(1-e^{2ik\pi/n}) , I noticed that this is Φ n ( 1 ) \Phi_n(1) , by definition.

We all have our favourite techniques; I can tell that one of yours is Möbius inversion ;)

Your Lemma is a well-known fact of basic trigonometry, of course, found in any good table . If you use that , your solution becomes quite short too.

Otto Bretscher - 5 years, 1 month ago

Log in to reply

@Otto Bretscher I know, it just seems that cyclotomic polynomials appear to possess "magical" properties such as having integer coefficients.

Julian Poon - 5 years, 1 month ago

Log in to reply

@Julian Poon Well, things appear magical until you see the "trick" ;) It's just long divison, as you know.

Otto Bretscher - 5 years, 1 month ago

I don't understand how ln(F(n))=1*ln(f(n)) I've tried the algebra a lot of different ways but I can't get the sums to be the same

Aakash Parikh - 5 years, 1 month ago

Log in to reply

Hint: Use the fact that φ 1 = n \varphi*1=n . It's rather late now, so if you still have trouble you can comment again, I'll explain it in greater detail.

Julian Poon - 5 years, 1 month ago
Andreas Wendler
Apr 24, 2016

At the beginning the mention may be allowed that the following solution is obviously not so elegant - but "The end justifies the means!":

First calculate the value for the product via EXCEL: P=2.01076E-87

Now adapt the result to the necessary form: l n ( P a ) l n ( 2 ) = b \frac{ln(\frac{P}{a})}{ln(2)}=-b

Already for a=1 obtain b=288 and therefore solution 289 \boxed{289} .

Genau, "der Zweck heiligt die Mittel" (+1)

Since I don't know how to use Excel, I had to do it the old-fashioned way, the way my Landsmann Euler did it ;)

Otto Bretscher - 5 years, 1 month ago

Log in to reply

Great problem!

Julian Poon - 5 years, 1 month ago

Log in to reply

Why is this so???

Andreas Wendler - 5 years, 1 month ago

Log in to reply

@Andreas Wendler I've never thought of using mobius inversion on a product before.

Julian Poon - 5 years, 1 month ago

Log in to reply

@Julian Poon We need to see your solution, please! I did not use Möbius inversion, although I see how it could be done, in terms of the von Mangoldt function. There are various ways to show that Φ n ( 1 ) = 1 \Phi_n(1)=1 if n n has more than one distinct prime factor.

Otto Bretscher - 5 years, 1 month ago

@Julian Poon Don't think so wide and the problem will reduce suddenly!

Andreas Wendler - 5 years, 1 month ago

0 pending reports

×

Problem Loading...

Note Loading...

Set Loading...