6
$\begingroup$

I am looking to simplify the following, without the use of capital Pi notation: $$\prod_{k=3}^{n-1}\cos\left(\frac{\pi}{k}\right)$$ Which is meant to produce the sequence: $\left[1,\ \frac{1}{2},\ \frac{1}{2}\frac{\sqrt{2}}{2},\frac{1}{2}\frac{\sqrt{2}}{2}\frac{1+\sqrt{5}}{4},\frac{1}{2}\frac{\sqrt{2}}{2}\frac{1+\sqrt{5}}{4}\frac{\sqrt{3}}{2}...\right]$.

I have seen identities of a similar structure, such as: $$\prod_{k=1}^{n-1}\sin\left(\frac{k\pi}{n}\right) = \frac{n}{2^{n-1}},\qquad or\qquad \prod_{k=1}^{n-1}\cos\left(\frac{k\pi}{n}\right) = \frac{\sin(\frac{\pi n}{2})}{2^{n-1}}$$

But, I am versed neither in the proofs of these identities, nor the properties of the $\Pi$ notation, so I've had a lot of difficulty trying to simplify this on my own. Dealing with $k$ in the denominator instead of the numerator (like in the two aforementioned identities) is something that I am evidently unequipped to deal with on my own. Thank you to anyone willing to help me out!

$\endgroup$
  • 3
    $\begingroup$ I suspect that having the $k$ in the denominator makes this product fundamentally different from the other examples you cite; in particular, it is unlikely to have a closed form. $\endgroup$ – Greg Martin Apr 16 at 21:41
7
$\begingroup$

As said in comments and answer, a closed form is more than likely improbable.

However, we could make some approximations. Let $a_k=\cos \left(\frac{\pi }{k}\right)$ and consider $$P_n=\prod_{k=3}^{n-1} a_k =\frac{1}{2\sqrt 2}\prod_{k=5}^{n-1} a_k$$ Taking logarithms $$ \log(P_n)=\log\left(\frac{1}{2\sqrt 2}\right)+\sum_{k=5}^{n-1} \log(a_k )$$ Now, using composition of Taylor series $$\log(a_k)=-\frac{\pi ^2}{2 k^2}-\frac{\pi ^4}{12 k^4}-\frac{\pi ^6}{45 k^6}-\frac{17 \pi ^8}{2520 k^8}-\frac{31 \pi ^{10}}{14175 k^{10}}-\frac{691 \pi ^{12}}{935550 k^{12}}+O\left(\frac{1}{k^{14}}\right)$$ Just for your curiosity $$\log(a_k)=-\sum_{p=1}^\infty\frac{2^{2 p-3} \pi ^{2 p} (E_{2 p-1}(1)-E_{2 p-1}(0))}{p\, (2 p-1)!\,k^{2p}}=\sum_{p=1}^\infty \frac {c_p}{k^{2p}}$$ where appear Euler polynomials but I am afraid that we cannot do much for the summation over $k$ beside using the fact that $$\sum_{k=7}^{n-1}k^{-2p}=H_{n-1}^{(2 p)}-1-2^{-2 p}-3^{-2 p}-4^{-2 p}-5^{-2 p}-6^{-2 p}$$

Using the expansion as written above, we should arrive to some ugly expression such as $$\sum_{k=5}^{n-1} \log(a_k )=K-\frac{\pi ^2}{2} H_{n-1}^{(2)}-\frac{\pi ^4 }{12} H_{n-1}^{(4)}-\frac{\pi ^6}{45} H_{n-1}^{(6)}-\frac{17 \pi ^8}{2520} H_{n-1}^{(8)}-\frac{31 \pi ^{10} }{14175} H_{n-1}^{(10)}-\frac{691 \pi ^{12}}{935550} H_{n-1}^{(12)}$$ where appear genralized harmonic numbers and where $$K=\frac{205 \pi ^2}{288}+\frac{22369 \pi ^4}{248832}+\frac{607493 \pi ^6}{26873856}+\frac{7339467953 \pi ^8}{1083553873920}+\frac{76853883151 \pi ^{10}}{35107145515008}+\frac{6162541526621059 \pi ^{12}}{8341457774365900800}$$

Trying for a few values of $n$, the decimal values are $$\left( \begin{array}{ccc} n & \text{approximation} & \text{exact} \\ 3 & 0.99918368 & 1.00000000 \\ 4 & 0.49999441 & 0.50000000 \\ 5 & 0.35355339 & 0.35355339 \\ 6 & 0.28603083 & 0.28603070 \\ 7 & 0.24770997 & 0.24770985 \\ 8 & 0.22317897 & 0.22317887 \\ 9 & 0.20619049 & 0.20619039 \\ 10 & 0.19375568 & 0.19375558 \\ 11 & 0.18427260 & 0.18427251 \\ 12 & 0.17680827 & 0.17680818 \\ 13 & 0.17078367 & 0.17078359 \\ 14 & 0.16582101 & 0.16582093 \\ 15 & 0.16166353 & 0.16166345 \\ 16 & 0.15813079 & 0.15813072 \\ 17 & 0.15509235 & 0.15509228 \\ 18 & 0.15245161 & 0.15245154 \\ 19 & 0.15013553 & 0.15013546 \\ 20 & 0.14808788 & 0.14808780 \end{array} \right)$$

Still using the above expansion and the asymptotics of harmonic numbers, converting to decimals, we should have for large $n$ $$P_n \sim \frac{1}{2 \sqrt{2}} \exp \left(c+\frac{\pi ^2}{2 n} +\frac{\pi ^2}{4 n^2} \right)\qquad \text{with} \qquad c \approx -1.12361$$

Applied to $n=20$, this last formula would give $0.148018$.

Edit

For the infinite product

$$\prod_{k=3}^{\infty}\cos\left(\frac{\pi}{k}\right)=0.1149420448532962007010401576568126847536004314847$$ while the given appoximation would lead to $0.11494164$.

Update

Remembering that we know the exact values up to $k=6$, using as an approximation $$\cos(x) \simeq\frac{\pi ^2-4x^2}{\pi ^2+x^2}\qquad (-\frac \pi 2 \leq x\leq\frac \pi 2)\implies \cos\left(\frac{\pi}{k}\right)=\frac{k^2-4}{k^2+1}$$ we arrive to $$P_n=\prod_{k=3}^{n-1}\cos\left(\frac{\pi}{k}\right)=\frac{\sqrt{6}+\sqrt{30}}{32} \prod_{k=7}^{n-1}\cos\left(\frac{\pi}{k}\right)=\frac{40885 \left(\sqrt{6}+\sqrt{30}\right) \pi \text{csch}(\pi )}{774144} \frac{ \Gamma (n-2)\, \Gamma (n+2)}{ \Gamma (n-i) \,\Gamma (n+i)}$$ giving $$P_\infty=\frac{40885 \left(\sqrt{6}+\sqrt{30}\right) \pi \text{csch}(\pi )}{774144}\approx 0.113881$$

Using instead the $[2,2]$ Padé approximant $$\cos\left(\frac{\pi}{k}\right)=\frac{12 k^2-5 \pi ^2}{12 k^2+\pi ^2}$$ we should get $$P_n=\frac{K_1}{K_2} \frac{\Gamma \left(n-\frac{1}{2} \sqrt{\frac{5}{3}} \pi \right) \Gamma \left(n+\frac{1}{2} \sqrt{\frac{5}{3}} \pi \right)}{\Gamma \left(n-\frac{i \pi }{2 \sqrt{3}}\right) \Gamma \left(n+\frac{i \pi }{2 \sqrt{3}}\right)}$$ where $$K_1=\sqrt{5} \left(\sqrt{6}+\sqrt{30}\right) \left(12+\pi ^2\right) \left(48+\pi ^2\right) \left(108+\pi ^2\right) \left(192+\pi ^2\right) \left(300+\pi ^2\right) \left(432+\pi ^2\right)$$ $$ \sin \left(\frac{1}{2} \sqrt{\frac{5}{3}} \pi ^2\right) \text{csch}\left(\frac{\pi ^2}{2 \sqrt{3}}\right)$$ $$K_2=800 \left(\pi ^2-60\right) \left(5 \pi ^2-432\right) \left(5 \pi ^2-192\right) \left(5 \pi ^2-108\right) \left(5 \pi ^2-48\right) \left(5 \pi ^2-12\right)$$ giving $$P_\infty=\frac{K_1}{K_2}\approx 0.114938$$

It is normal that the Padé approximant leads to better results sincd $$\int_0^{\frac \pi 7} \left(\cos (x)-\frac{\pi ^2-4 x^2}{\pi ^2+x^2}\right)^2\,dx\approx 1.028 \times 10^{-7}$$ while $$\int_0^{\frac \pi 7}\left(\cos (x)-\frac{12-5 x^2}{12+x^2}\right)^2\,dx\approx 9.574 \times 10^{-12}$$

$\endgroup$
4
$\begingroup$

Well, you could write $$ \prod_{k=3}^n \cos(\pi/k) = 2^{2-n} \sum_{signs} \cos\left(\pm \frac{\pi}{3} \pm \frac{\pi}{4} \pm \ldots \pm \frac{\pi}{n}\right) $$ where the sum is over all $2^{n-2}$ possible choices of the $\pm$ signs.
If $n$ is moderately large, those $\pm \pi/3 \pm \ldots \pm \pi/n$ will be rather nasty rational multiples of $\pi$. So it's not exactly a "simplification".

For example, if $n=7$ I get $$ 16^{-1} \left(\cos \left( {\frac {11\,\pi}{420}} \right) +\cos \left( {\frac {13\, \pi}{140}} \right) +\cos \left( {\frac {27\,\pi}{140}} \right) +\cos \left( {\frac {109\,\pi}{420}} \right) \\+\cos \left( {\frac {43\,\pi}{ 140}} \right) +\cos \left( {\frac {179\,\pi}{420}} \right) +\cos \left( {\frac {59\,\pi}{420}} \right) +\cos \left( {\frac {83\,\pi}{ 140}} \right) \\+\cos \left( {\frac {57\,\pi}{140}} \right) +\cos \left( {\frac {199\,\pi}{420}} \right) +\cos \left( {\frac {97\,\pi}{ 140}} \right) +\cos \left( {\frac {319\,\pi}{420}} \right)\\ +\cos \left( {\frac {31\,\pi}{420}} \right) +\cos \left( {\frac {113\,\pi}{ 140}} \right) +\cos \left( {\frac {151\,\pi}{420}} \right) +\cos \left( {\frac {153\,\pi}{140}} \right) \right)$$

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.