I wanted to show the derivation of the expression for the Riemann zeta function
at even values, i.e.
ζ(2k)=2(2k)!(−1)k+1(2π)2kβ2k
We start by expressing the function πxsin(πx) in its infinite product form:
πxsin(πx)=∏n=1∞(1−n2x2)
A way of understanding this representation is by realising that πxsin(πx) can be expressed as a product of linear factors, as you would with any other polynomial. Let us consider a second degree polynomial as an example:
f(x)=x2+7x+12
We know that it can be expressed as a product of its linear factors:
f(x)=(x+4)(x+3)
However, it can also be expressed as
f(x)=(1+4x)(1+3x)
This can be verified by seeing that f(−4) and f(−3)=0. However, look into the Weierstrass factorisation theorem for a more rigorous explanation on expressing functions in this form.
πxsin(πx) has roots at x=n for n∈Z and n=0, and so, expressing it as a product of its linear factors in a similar way gives
πxsin(πx)=(1−x)(1+x)(1−2x)(1+2x)(1−3x)(1+3x)⋯
Making the observation that pairing terms gives the difference of two squares we get the infinite product form:
πxsin(πx)=(1−x2)(1−4x2)(1−9x2)⋯
From here we’re going to manipulate the product of πxsin(πx) to a point where we recognise
something of the sort we are trying to derive. Firstly, we take the natural logarithm of both sides:
logπxsin(πx)=log(∏n=1∞(1−n2x2))
Making use of the property that the logarithm of a product is equivalent to the sum of the individual logarithms gives
logπxsin(πx)=∑n=1∞log(1−n2x2)
log(sin(πx))=log(πx)+∑n=1∞log(1−n2x2)
Taking the derivative of both sides:
dxd[log(sin(πx))]=dxd[log(πx)+∑n=1∞log(1−n2x2)]
sin(πx)πcos(πx)=x1−∑n=1∞(n22x)(1−n2x21)
πxcot(πx)=1−2∑n=1∞(n2x2)(1−n2x21)
Expressing 1−n2x21 as its series expansion gives
πxcot(πx)=1−2∑n=1∞∑k=0∞(n2x2)(n2x2)k
πxcot(πx)=1−2∑n=1∞∑k=0∞(n2x2)k+1
The lower limit of the inner summation can be changed to k=1 if we also change the exponent of the term being summed:
πxcot(πx)=1−2∑n=1∞∑k=1∞n2kx2k
πxcot(πx)=1−2∑k=1∞ζ(2k)x2k
And so, we see that we have some sort of expression with ζ(2k) on the right-hand side. Let’s now take a closer look at the cotangent function on the left-hand side.
The cotangent can be expressed as the ratio of cosine to sine. Expressing the cosine and sine in their exponential function form gives
πxcot(πx)=πxsin(πx)cos(πx)=πx2eiπx+e−iπxeiπx−e−iπx2i=iπxeiπx−e−iπxeiπx+e−iπx
Multiplying the right-hand side by eiπxeiπx (just because it
seemed to work):
πxcot(πx)=iπxeiπx−e−iπxeiπx+e−iπx⋅eiπxeiπx=iπxe2iπx−1e2iπx+1
And now, splitting the numerator, we get
πxcot(πx)=iπxe2iπx−1e2iπx−1+2=iπx(1+e2iπx−12)=iπx+e2iπx−12iπx
Those last couple of steps may have seemed counterintuitive, however, it
allows us to make use of the following relation:
ex−1x=∑k=0∞k!βkxk
Substituting this into our second cotangent expression gives
πxcot(πx)=iπx+∑k=0∞k!βk(2iπx)k
You may notice that this looks quite similar in form to the first cotangent equation we derived. Manipulating the above equation into a similar form to the one we derived earlier will allow the x2k coefficients to be equated, which will give the expression for ζ(2k).
We begin by expanding out the first two terms of the summation:
πxcot(πx)=iπx+0!β0+1!β1(2iπx)+∑k=2∞k!βk(2iπx)k
β0=1 and β1=−21, and so we get
πxcot(πx)=iπx+1−21(2iπx)+k=2∑∞k!βk(2iπx)k=1+k=2∑∞k!βk(2iπx)k
Taking out a factor of −2, just because we know what we’re working towards and
the expression is getting close,
πxcot(πx)=1−2∑k=2∞k!βk−2(2iπx)k
We can alter the lower summation limit again by altering the exponent:
πxcot(πx)=1−2k=1∑∞2k!β2k−2(2iπx)2kx2k=1−2k=1∑∞−2(2k)!i2k(2π)2kβ2kx2k=1−2k=1∑∞−2(2k)!(−1)k(2π)2kβ2kx2k
And so, we get
πxcot(πx)=1−2∑k=1∞2(2k)!(−1)k+1(2π)2kβ2kx2k
Now recall the cotangent expression we got to earlier:
πxcot(πx)=1−2∑k=1∞ζ(2k)x2k
Equating the two cotangent expression, we get
ζ(2k)=2(2k)!(−1)k+1(2π)2kβ2k