Basel Problem 2.0

Calculus Level 5

Consider the sum S = 1 x 2 S=\displaystyle \sum\dfrac{1}{x^2} which is over all the positive real solutions of the equation tan x x = n \dfrac{\tan{x}}{x}=n .

If S = 1 S=1 , then n = p q n=\dfrac{p}{q} , where p p and q q are positive coprime integers, find p + q p+q .


The answer is 8.

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

Chris Lewis
Jun 24, 2020

Let k = p q k=\frac{p}{q} . We're interested in the roots of tan x x = k \frac{\tan x}{x}=k which we can work with more easily in the form sin x = k x cos x \sin x = kx\cos x

Euler's original approach to the Basel problem was to consider a function's power series as an infinite polynomial, and manipulate it as such. (This has serious flaws, but a) it worked, b) Euler later proved the same result in different ways and c) the rigorous justification for the approach in this case was found later still.)

By Vieta, for a polynomial a 0 + a 1 x + a 2 x 2 + a_0+a_1 x + a_2 x^2 + \cdots with roots x 1 , x 2 , x_1,x_2, \cdots , we have 1 x i 2 = ( a 1 a 0 ) 2 2 a 2 a 0 \sum \frac{1}{x_i^2} = \left( \frac{a_1}{a_0} \right)^2-2\frac{a_2}{a_0}

Rewriting our equation in terms of power series, we have sin x = k x cos x x x 3 6 + = k x ( 1 x 2 2 + ) 1 x 2 6 + = k ( 1 x 2 2 + ) 1 k + ( 3 k 1 ) x 2 6 + = 0 \begin{aligned} \sin x &= kx\cos x \\ x-\frac{x^3}{6}+\cdots &= kx \left( 1-\frac{x^2}{2}+\cdots \right) \\ 1-\frac{x^2}{6}+\cdots &= k \left( 1-\frac{x^2}{2}+\cdots \right) \\ 1-k+(3k-1)\frac{x^2}{6}+\cdots &=0 \end{aligned}

Applying the Vieta formula from above, the sum of the reciprocals of the squares of the roots will be 2 3 k 1 6 ( 1 k ) -2\frac{3k-1}{6(1-k)}

Note that the sum in the problem is restricted to the positive reals. If x x is a root, so is x -x ; so the quantity above is double the required sum, and we get the equation 3 k 1 6 ( 1 k ) = 1 -\frac{3k-1}{6(1-k)}=1 with solution k = 5 3 k=\frac53 , so that p + q = 5 + 3 = 8 p+q=5+3=\boxed8 .

Your approach tells us that, for k = 1 3 k=\dfrac{1}{3} , the sum S = 0 S=0 , which is not really true.

The formula you gave works fine for k > 1 k>1 , but fails for k 1 k\leq 1 .

Can you justify this?

Digvijay Singh - 11 months, 3 weeks ago

Log in to reply

That is a very good point. It's even worse for 1 3 < k < 1 \frac13<k<1 - the formula gives S < 0 S<0 .

Let's say S = 3 k 1 3 ( k 1 ) S'=\frac{3k-1}{3(k-1)} . This formula (as above) should give the sum of the reciprocals of the squares of ALL roots of tan x x = k \frac{\tan x}{x}=k .

I mentioned in the solution that this sum includes contributions from negative roots; luckily we can just halve it to find S S . This works fine if all the roots are positive or negative; but in some cases, they're not: we also get complex roots. In fact, we'll show the extra roots are pure imaginary.

Let's look at the equation over the complex numbers: tan z z = k \frac{\tan z}{z}=k (we'll keep k k real.)

For the ratio of two complex numbers to be real, they must have the same argument (up to a difference of n π n\pi ). We can avoid this difference by taking the tangent of the two arguments; so we need tan ( arg tan z ) = tan ( arg z ) \tan (\arg \tan z)=\tan(\arg z)

Put z = x + i y z=x+iy , with x , y x,y real. Let's assume there's a root that is neither real nor pure imaginary; so x y 0 xy \neq 0 . After a bit of word-processing this equation becomes sinh 2 y sin 2 x = y x \frac{\sinh 2y}{\sin 2x}=\frac{y}{x} or y sinh 2 y = x sin 2 x \frac{y}{\sinh 2y}=\frac{x}{\sin 2x}

However, for real, non-zero x , y x,y , the left-hand side is always less than 1 1 and the right-hand side is always greater than 1 1 ; contradiction. So if any non-real roots exist, they must be of the form z = i y z=iy (ie pure imaginary).

Back to the original equation, this means we have to solve tanh y y = k \frac{\tanh y}{y}=k . Now, this only has solutions for 0 < k < 1 0<k<1 , so in the case k = 5 3 k=\frac53 , we're OK, but when k = 1 3 k=\frac13 , we get a negative contribution to S S' from the pure imaginary roots.

Working out what this contribution actually is seems tricky (to me, anyway) - the roots of tanh y y = k \frac{\tanh y}{y}=k seem to be inextricably linked to those of tan x x = k \frac{\tan x}{x}=k , so the series expansion/Vieta approach doesn't work. However, tanh y y = k \frac{\tanh y}{y}=k is much easier to solve numerically, as there are only ever at most two real roots.

The last question might be whether this contribution is great enough to obscure a solution to the original problem in the range 0 < k < 1 0<k<1 .

I think the easiest way to see that there are no such solutions is to note that for 0 < k < 1 0<k<1 , the smallest positive root is greater than π \pi ; in fact, the n n th positive root is always larger than n π n\pi , so that S 0 < k < 1 < n = 1 1 n 2 π 2 = 1 6 S_{0<k<1}<\sum_{n=1}^{\infty} \frac{1}{n^2 \pi^2} = \frac16

Finally, note that at k = 0 k=0 the positive roots are precisely π , 2 π , 3 π , \pi,2\pi,3\pi,\ldots so S 0 = 1 6 S_0=\frac16 , and this is in agreement again with the formula presented above.

Thanks for the bonus question - I'd be interested in firming up the rigour (I realise there's a lot of hand-waviness here!)

Chris Lewis - 11 months, 3 weeks ago

complex roots may not be considered

Chin-Ming He - 10 months, 2 weeks ago
Mark Hennings
Jun 24, 2020

For any positive real number r > 1 r > 1 , consider the entire function f r ( z ) = r cos z sin z z f_r(z) \; = \; r\cos z - \frac{\sin z}{z} Now f r ( z ) = ( r 1 ) 1 6 ( 3 r 1 ) z 2 + O ( z 4 ) z 0 f_r(z) \; = \; (r - 1) - \tfrac16(3r - 1)z^2 + O(z^4) \hspace{2cm} z \to 0 and so 0 0 is not a root of f r f_r , and thus the roots of f r ( z ) f_r(z) are just the nonzero solutions of tan z = r z \tan z = rz . Since both sin z \sin z and cos z \cos z are entire functions of order 1 1 at infinity, so is f r f_r . Suppose that X r X_r is the set of positive roots of the equation tan z = r z \tan z = rz . Then X r X_r contains one element in the interval ( n π , ( n + 1 2 ) π ) \big(n\pi,(n+\tfrac12)\pi\big) for each integer n 0 n \ge 0 . Thus X r X_r is countable, and we can write X r = { α r , n n N } X_r = \{\alpha_{r,n} \,|\, n \in \mathbb{N}\} . Since it is certainly true that α r n > ( n 1 ) π \alpha_{rn} > (n-1)\pi , we deduce that n = 1 1 α r n 2 < \sum_{n=1}^\infty \frac{1}{\alpha_{rn}^2} \; < \; \infty Using the Hadamard Factorisation Theorem, we deduce that f r ( z ) = e g ( z ) n = 1 ( 1 z 2 α r n 2 ) f_r(z) \; = \; e^{g(z)} \prod_{n=1}^\infty \left(1 - \frac{z^2}{\alpha_{rn}^2}\right) where g ( z ) g(z) is a polynomial of degree at most 1 1 . Since f r ( z ) f_r(z) is even, we deduce that g ( z ) g(z) is contant, and hence f r ( z ) = ( 1 r ) n = 1 ( 1 z 2 α r n 2 ) = ( 1 r ) [ 1 ( n = 1 1 α r n 2 ) z 2 + O ( z 4 ) ] z 0 f_r(z) \; =\; (1 - r)\prod_{n=1}^\infty \left(1 - \frac{z^2}{\alpha_{rn}^2}\right) \; = \; (1 - r)\left[1 - \left(\sum_{n=1}^\infty \frac{1}{\alpha_{rn}^2}\right)z^2 + O(z^4) \right] \hspace{2cm} z \to 0 and hence n = 1 1 α r n 2 = 3 r 1 6 ( r 1 ) ( ) \sum_{n=1}^\infty \frac{1}{\alpha_{rn}^2} \; = \; \frac{3r - 1}{6(r - 1)} \hspace{3cm} (\star) For this sum to be equal to 1 1 , we must have r = 5 3 r = \tfrac53 , and hence the solution is 5 + 3 = 8 5 + 3 = \boxed{8} .

It so happens that the all the roots of tan z = r z \tan z = r z are real if r > 1 r > 1 , while there are imaginary roots if 0 < r < 1 0 < r < 1 , which is why formula ( ) (\star) fails when 0 < r < 1 0 < r < 1 . Well, it does not fail, but the infinite sum n α r n 2 \sum_n \alpha_{rn}^{-2} is over the imaginary roots as well as the real ones.

can we solve it using the fact that 1/(1+c)^2 +1/(2+c)^2-------------------- converges. where c is not an integer. there may be some formula for summing up this series. we can also use cauchy residue formula to sum up this series.

Srikanth Tupurani - 11 months, 1 week ago

Log in to reply

The infinite series is ψ ( 1 ) ( c + 1 ) \psi^{(1)}(c+1) , but it won’t help us much, since the roots of tan x = n x \tan x = nx are not of the form π ( c + n ) \pi(c+n) for integer n n

Mark Hennings - 11 months, 1 week ago
Digvijay Singh
Jun 24, 2020

A similar problem appeared in The American Mathematical Monthly Vol. 93 (1986) which asks for the value of S S when tan x = x \tan{x}=x .

The value of S S in that problem turns out to be 1 10 \dfrac{1}{10} .

The general problem of finding S S is also dealt with, in the solution to the problem, which can be found here , in the section called 'A Rayleigh Popular Problem'.


Check out a similar problem here

0 pending reports

×

Problem Loading...

Note Loading...

Set Loading...