Prerequisites
This article requires knowledge of complex numbers and some transforms.
- If you don’t know complex numbers, you can refer to this article.
- If you don’t know transforms, you can refer to this article.
Shift Operator#
We define: D≡dxd. Consider
eD=n=0∑∞n!DnActing on a function f(x), we have:
eDf=n=0∑∞n!Dnf(x)=n=0∑∞n!f(n)(x)1nNoting that this is the Taylor expansion of the function f(x) at x, we clearly have:
eDf=f(x+1)Similarly:
etDf=n=0∑∞n!f(n)(x)tn=f(x+t)Consider computing:
eaD⋅ebD=n=0∑∞n!anDn⋅n=0∑∞n!bnDn=n=0∑∞Dnk=0∑nk!(n−k)!akbn−k=n=0∑∞n!Dn(a+b)nk=0∑n(kn)akbn−k=n=0∑∞n!Dn(a+b)n=e(a+b)DThus the shift operator satisfies exponential operations.
Difference and Shift#
Define the difference operator Δ as:
Δf(x)=f(x+1)−f(x)=eDf(x)−f(x)=(eD−1)f(x)Thus:
Δ=eD−1Taking the inverse on both sides, we have:
Δ−1=eD−11=D1eD−1DThe latter part is exactly the generating function of Bernoulli numbers, so expanding gives:
Δ−1=D1n=0∑∞n!BnDnNote that:
∑=Δ−1,∫=D−1Substituting into the above equation, we get:
∑=∫n=0∑∞n!BnDn=∫+n=1∑∞n!Bn∫Dn=∫+n=1∑∞n!BnDn−1Noting that the odd-indexed Bernoulli numbers (except the first term) are all 0, we have:
∑=∫−21+n=1∑∞(2n)!B2nD2n−1Applying this formula from a to b to the function f(x), we obtain:
a≤n≤b∑f(n)=∫abf(x)dx−2f(b)−f(a)+n=1∑∞(2n)!B2n[f(2n−1)(b)−f(2n−1)(a)]Thus simply deriving the Euler-Maclaurin summation formula.
Ramanujan Master Theorem#
Consider the Laplace transform:
L{xs−1}(eD)=∫0∞xs−1e−xeDdx=Γ(s)e−sDWe substitute
e−xeD=n=0∑∞n!enD(−x)nand let it act on the function ϕ(0), yielding:
∫0∞xs−1n=0∑∞n!enD(−x)nϕ(0)dx=Γ(s)ϕ(−s)e−sD⋅ϕ(0)Let
f(x)=n=0∑∞n!enDϕ(0)(−x)n=n=0∑∞n!ϕ(n)(−x)nThen:
∫0∞xs−1f(x)dx=Γ(s)ϕ(−s)This is the Ramanujan Master Theorem (Ramanujan CF2100 Theorem). We formally state its proposition:
Ramanujan Master Theorem
For a function f(x), its Mellin Transform satisfies:
M{f(x)}(s)=∫0∞xs−1f(x)dx=Γ(s)ϕ(−s)if and only if the function f(x) satisfies:
f(x)=n=0∑∞n!ϕ(n)(−x)n
This theorem describes the general property of the Mellin transform for entire functions.
Applications#
DHL Integral#
First, I will prove the DHL integral mentioned in my blog introduction.
Consider the integral:
I=∫0∞sinxndxLet t=xn,s=n1, then:
I=s∫0∞ts−1sintdt=sI∫0∞ts−1eitdtConsider Taylor expanding eit (I proved in a previous article that this is feasible):
eit=n=0∑∞n!(it)n=n=0∑∞n!(−t)nϕ(n)e−i2nπApplying the Ramanujan Master Theorem, we get:
I=I[sΓ(s)ei2sπ]=n1Γ(n1)sin(2nπ)Similarly:
∫0∞cosxndx=n1Γ(n1)cos(2nπ)An Anomalous Integral#
Consider the integral:
I=∫0∞1+xndxLet t=xn,s=n1, then we have:
∫0∞1+xndx=s∫0∞1+tts−1dtNote that:
1+t1=n=0∑∞(−t)n=n=0∑∞n!(−t)nϕ(n)Γ(1+n)Applying the Ramanujan Master Theorem, we get:
I=sΓ(s)Γ(1−s)=nsinnππFirst, recall the series expansion form of the ψ function:
ψ(n)=−γ+k=0∑∞(k+11−k+n1)Rearranging gives:
Hn=ψ(1+n)+γConsider the integral:
I=∫0∞1+xxs−1ln(1+x)dx,R(s)∈(0,1)Note that:
ln(1+x)=−n=0∑∞n+1(−x)n+1,1+x1=n=0∑∞(−x)nThen:
1+xln(1+x)=−n=0∑∞n+1xn+1⋅n=0∑∞(−x)n=−n=0∑∞(−x)nk=1∑nk1=−n=0∑∞Hn(−x)n=−n=0∑∞[ψ(1+n)+γ](−x)nUsing the same technique as the previous problem, we obtain:
I=−sinπsπ[ψ(1−s)+γ]Zeta Function#
Finally, we give the connection between the Riemann Zeta function and Bernoulli numbers.
First:
ζ(s)Γ(s)=∫0∞ex−1xs−1dxNote that the right side is a Mellin Transform. According to the Ramanujan Master Theorem:
ex−11=n=0∑∞n!ζ(−n)(−x)nNote that this is very similar to the generating function of Bernoulli numbers, so:
n=0∑∞Bnn!xn=n=0∑∞(−1)nn!ζ(−n)xn+1Consider unifying the powers. Note that n→∞limn!n=0, so:
n=0∑∞(−1)nn!ζ(−n)xn+1=n=0∑∞(−1)n+1nζ(1−n)n!xnComparing coefficients, we get:
ζ(1−n)=(−1)n+1nBnNoting that the odd-indexed Bernoulli numbers (except the first term) are all 0, the formula can be rewritten as:
ζ(1−2n)=−2nB2n,n∈Z