Last month, my friend at https://ceti.kaush.com/ sent me a particularly challenging integral. Back then, I had no idea how to approach the question — I barely knew what an integral even was.
But I’ve done a lot since a month ago (mostly through Apostol Calc 1, recommended by the same friend), so I decided to revisit this and give it a shot. Ceti wrote a post on this himself, so if you’re looking for something more coherent and organized, make sure to check that out!
The problem statement is as follows:
I=∫01ln(1−xx)x−21 dx
This integral was a lot of fun to solve, so I recommend anybody who’s reading this to try it yourself before reading the solution. :)
1. Substitution
My first thought is to use substitution to simplify the problem. We can write u as the following and rewrite each part in terms of u:
u=1−xx,x=1+uu
We can also differentiate with respect to u using the quotient rule:
dudx=dud1+uu=(1+u)2(1)(1+u)−(u)(1)=(1+u)21
We can now rewrite the entire integral, and with simple algebraic manipulation:
I=21∫0∞(1+u)3lnuu−1du
(Just to note, when x=1, u=01=∞)
We successfully took apart our integral into something a little simpler. For ease, we set I=21J.
2. Make the integral J more symmetrical
We can transform the integral J over [0,∞] to [0,1] to make it more manageable. We split the integral:
∫0∞=∫01+∫1∞
To map the interval [1,∞) to [0,1], we set u=v1. When u=1, v=1 and when u=∞, v=0:
∫1∞(1+u)3lnuu−1du=−∫01(1+v)3lnv1−vdv
Now both integrals are over [0,1]. Since v is a dummy variable we rewrite it as u, and combine:
∫01(1+u)3lnuu−1du−∫01(1+u)3lnu1−udu=2∫01(1+u)3lnuu−1du
Since J=2I, we divide by 2 to work with I from now on:
I=∫01(1+u)3lnuu−1du
3. Power Series
We split our integral:
∫01(1+u)3u−1lnu1du
The first part has a rational denominator without a convenient antiderivative. We can use the binomial theorem to get rid of it:
(1+u)31=n=0∑∞(−1)n(2n+2)un(∣u∣<1)
Multiplying by u−1:
(1+u)3u−1=(u−1)n=0∑∞(−1)n(2n+2)un
Setting an=(−1)n(2n+2) and distributing:
(u−1)n=0∑∞anun=n=0∑∞anun+1−n=0∑∞anun
Using an index shift m=n+1 on the first sum and combining (setting a0=(22)=1):
n=0∑∞(−1)n(2n+2)(un+1−un)
Since this converges uniformly, we can take the summation outside the integral:
n=0∑∞(−1)n(2n+2)∫01(un+1−un)lnu1du
4. Integral identity
It is easily proved that:
∫01lnyym−1dy=ln(m+1),m≥0
Therefore:
∫01lnuun+1−undu=ln(n+2)−ln(n+1)=lnn+1n+2
So:
I=n=0∑∞(−1)n(2n+2)lnn+1n+2
5. Series + Zeta Function
Because of the nature of the sum, it’s easy to spot that we want to rewrite our equation in terms of ζ.
We first write each log as:
S=n=0∑∞an(ln(n+2)−ln(n+1))=n=0∑∞anln(n+2)−n=0∑∞anln(n+1)
Rewriting in terms of k=n+2 and k=n+1 and combining:
k=1∑∞(ak−2−ak−1)lnk
After careful algebra:
I=k=1∑∞(−1)kk2lnk
The structure of this sum corresponds to the derivative of the Dirichlet eta function:
η(s)=k=1∑∞ks(−1)k−1,η′(s)=−k=1∑∞(−1)k−1k−slnk
Setting s=−2:
η′(−2)=−k=1∑∞(−1)k−1k2lnk=k=1∑∞(−1)kk2lnk
This observation was made through a lot of trial and error — you aren’t expected to recognize this immediately!
We relate η′(−2) to ζ′(−2). Using η(s)=(1−21−s)ζ(s) and differentiating (the 8ζ(−2) term vanishes since ζ(−2)=0):
η′(−2)=−7ζ′(−2)
Using the known value ζ′(−2)=−4π2ζ(3):
η′(−2)=−7ζ′(−2)=4π27ζ(3)
Quick numerical check: 4π27ζ(3)≈0.2131
This is the first integral I’ve ever actually sat down and solved, so I’m rather proud of my solution. For questions, comments, concerns, or other proofs, email me at kikisadventures37@gmail.com.