A Trigonometric Path to Basel |
|
Two of the best known elementary trigonometric identities are |
|
|
|
From these we can deduce the solution of the so-called Basel problem, which is to find the sum of the squared reciprocals of the natural numbers (first solved by Euler by a different method). Dividing through by the squared product of the sine and cosine, the sum of squares identity can be re-written as |
|
|
|
Making use of the fact that cos(x) = sin(π/2 − x), and replacing x with x/2, this gives the identity |
|
|
|
Applying this same identity, the two terms on the right side can be expressed as |
|
|
|
The arguments of the sines on the right sides of these equations are |
|
|
|
These quantities are in arithmetic progression just if x = π/4, in which case the arguments are |
|
|
|
Thus we have the equalities |
|
|
|
Iterating this process, we can expand each of the four terms in the last expression into two terms using the basic identity, to arrive at an equivalent sum of eight terms with the sine arguments kπ/32 where k = 1, 3, 5, , 15. And so on. In general, for any positive integer N, we have |
|
|
|
Dividing through both sides of this equality by (4/π)2 we get |
|
|
|
Each time we increment N, the number of terms on the right side doubles, and the denominator of the sine arguments doubles, so the (arbitrary number of) initial arguments become arbitrarily small, and in this limit the sine approaches the argument. (As the arguments approach π/2 the sines are less than the arguments, so those contributions dont interfere with convergence.) In the limit as N goes to infinity, and noting that the left side equals π2/8, we have |
|
|
|
This is the sum of the squared reciprocals of the odd natural numbers, which we can denote as Sodd, and we seek the sum S of the squared reciprocals of all the natural numbers, which is given by S = Sodd + Seven where Seven is the sum of the reciprocals of the even natural numbers. Its easy to see that Seven = S/4, so we have S = (4/3)Sodd, and hence S = π2/6 as was to be shown. |
|
This approach also leads to a more general identity. Instead of specializing to x = π/4, we can allow x to be an arbitrary quantity, and we have at the third level |
|
|
|
Bringing the leading term on the right side over to the left, and proceeding on to the kth level for N=2k , the arguments of the leading sines become arbitrarily small, so we have the limiting relation as N → ∞ |
|
|
|
When x is any multiple of π, one of the terms in the series is infinite, so the equality applies only for x in the range from 0 to π. In terms of a new parameter u defined by x = πu, we can multiply through by π2 and get the relation (as N goes to infinity) |
|
|
|
Thus we have |
|
|
|
where B2k are the (positive) Bernoulli numbers. With u=0 this gives the Basel sum, whereas if we take (for example) u = √-1 we have |
|
|
|
Writing the preceding equation as |
|
|
|
we can differentiate both sides with respect to u to give |
|
|
|
Differentiating both sides again, we get |
|
|
|
Putting u = 0, this gives the next of Eulers results for even reciprocal powers |
|
|
|
Differentiating the previous expression twice more gives |
|
|
|
and evaluating this at u=0 we get the sum of the reciprocals of the 6th powers |
|
|
|
Thus for the general sum of reciprocals of the (2k)th powers, we have |
|
|
|
As an aside, let us return to the relation |
|
|
|
and recall that this series is valid for u between 0 and 1. Suppose we set u = reiθ for some real magnitude r (with |r|<1) and phase θ, and integrate both sides from θ = 0 to 2π. Thus we are integrating around a circle of radius r centered on the origin of the complex plane. In view of the integral |
|
|
|
we can integrate the left side of the previous equation term-by-term to give π3/3, independent of r. Therefore, the integral of the right side of that equation must also equal this value, independent of r (provided |r| < 1). The integral of the second term in that expression is zero, so we have |
|
|
|
This is slightly non-trivial, because the integrand function has both real and complex parts, and traces out a variety of loci in the complex plane for various values of r, as shown below. |
|
|
|
Despite the variety of these loci, the imaginary part of the integral always vanishes (because they are symmetrical about the real axis) and the real part always equals 2π/3. This plot somewhat disguises things because it doesnt show the variation in density as a function of θ. To clarify, let us define the function |
|
|
|
Thus F(ρ) is the integral of 1/sin(z)2 as z traces out an upper half circle of radius ρ in the complex plane. The plot on the left below shows f(θ,ρ) for three values of ρ in the range from 0 to π. The integrals F(ρ) of these three functions are each equal to π/3. |
|
|
|
In fact, we have F(ρ) = π/3 = 1.04719 for every real ρ in the range from 0 to π. However, for values of ρ in the range from π to 2π we have F(ρ) = 0.410577 as exemplified by the curves in the right hand plot above. For values of ρ in the range from 2π to 3π we have F(ρ) = 0.251422 , and so on. A plot of F(ρ) is shown below. |
|
|
|
The dashed line represents 0.675/ρ, which shows that the discrete values decrease in inverse proportion to ρ. To determine the exact values, let us define the values of the normalized integral for each of the ranges by setting ρ to the mid-point of the range (for convenience), so we have the quantities: |
|
|
|
Thus we have g0 = 1/3, g1 = 0.1306909660 , g2 = 0.080030374 , and so on. Now we notice that the differences (g0 g1), (g1 g2), (g2 g3), are in the proportion 1, 1/4, 1/9, and so on. From this it follows that for any N we have |
|
|
|
In the limit as N approaches infinity the value of gN goes to zero and the summation of reciprocal squares goes to π2/6, so we have |
|
|
|
Therefore, since g0 = 1/3, we have for N > 0 |
|
|
|
Solving this for the partial sums of reciprocal squares and inserting the expression for gN, we have |
|
|
|
Instead of considering just the squared reciprocal sines, we can consider the more general functions involving the nth power of the reciprocal sines (and normalizing the integral for convenience), and we can focus on a value of ρ (such as 1) in the principle range 0 to π . This leads us to define |
|
|
|
which has the rational values (for even n) |
|
|
|
Interestingly, these coefficients have an application in the Gauss-Encke numerical integration formula. Given a sequence of equally-spaced discrete values Vj = V(j), j = −3, −2, , 3, 4, of a continuous analytic function V(t) separated by units of the ordinate t, we can form the first differences aj = Vj Vj−1, and the second differences bj = aj+1 aj, and so on, as depicted in the array below. |
|
|
|
We then define the central means (V) = (V0 + V1)/2, and (b) = (b0 + b1)/2, and so on. Then the Gauss-Encke estimate of the integral of the function V(t) from t = 0 to 1 is given by |
|
|
|
For example, with V(t) = t(t−1)/2 we have the values 0, 1, 3, 6, 10, 15, , with the first differences 1, 2, 3, 4, 5, , and the second differences 1, 1, 1, 1, All higher-order differences are zero. Therefore, since G2/22 = 1/12 and (b) = 1, the integral of V(t) between any two consecutive values is the average of those values minus 1/12. For another example, with the function |
|
|
|
we have the values and differences |
|
|
|
so we have (V) = 10, (b) = 8, and (d) = 1, and all higher order differences are zero. We can use the integration formula to compute |
|
|
|
The usual way of computing the even-indexed Gn coefficients is by the rational integral |
|
|
|
although the equivalence between this integral of rational functions and the previous integral of transcendental functions seems non-trivial. |
|