Trying to generalize this problem [1] I came up with these two integrals

$$i_{\pm}(\sigma) = \int_0^1 \log^2(x)\frac{ \log(1\pm \sigma x)}{1\pm x}\,dx\tag{1}$$

Here $-1 \le \sigma \le 1$ is a parameter. For $\sigma = \pm 1$ we are led back to the problem [1].

Although these two integrals look like twins they are quite different.

Notice that the essential difference is in the denominator; under the $\log$ the sign change can be accomodated in $\sigma$.

While I was able to solve $i_{-}(\sigma)$ with the methods generalizing the solution to [1] I find $i_{+}(\sigma)$ tough.

Can you find to the closed forms of both integrals?

What I did so far:

The solution is based of the concept of the unified harmonic number $U_n$ I have introduced earlier [2] as a generalization of the harmonic number and the alternating harmonic number.

It is defined as

$$U_n(\sigma) = \sum_{k=1}^n \frac{\sigma^k}{k} = \sigma \int_0^1 \frac{1-(\sigma z)^n}{1-\sigma z}\,dz\tag{2}$$

The generating function is

$$g_{-}(x,\sigma) = \sum_{k=1}^\infty x^n U_n (\sigma) = -\frac{\log(1-\sigma x)}{1-x}\tag{3}$$

Here we recognize a part of the integrand of $i_{-}(\sigma)$.

Hence multiplying $(3)$ with $\log(x)^2$ and integrating from $x=0$ to $x=1$ we have

$$-i_{-}(\sigma) = \sum_{n=1}^\infty U_n(\sigma) \int_0^1 x^n \log(x)^2\,dx \\= \sum_{k=1}^\infty U_n(\sigma) \frac{2}{(1+n)^3} \\=\sigma \int_0^1 \sum_{n=1}^\infty \frac{1-(\sigma z)^n}{1-\sigma z}\frac{2}{(1+n)^3}\,dz\tag{4}$$

$$=2\sigma \int_0^1\frac{ \sigma z \zeta (3)-\text{Li}_3(z \sigma )}{z (1-\sigma z)}\,dz\tag{5}$$

$$-i_{-}(\sigma)= \operatorname{Li}_2(\sigma ){}^2+2 \operatorname{Li}_3(\sigma ) \log (1-\sigma )\\ -2 (\operatorname {Li}_4(\sigma )+\zeta (3) \log (1-\sigma )) \tag{6}$$

Repeating the same for $i_{+}$ we have now

$$g_{+}(x,\sigma) = \sum_{k=1}^\infty (-1)^n x^n U_n (\sigma) = -\frac{\log(1+\sigma x)}{1+x}\tag{3a}$$

Multiplying $(3a)$ with $\log(x)^2$ and integrating from $x=0$ to $x=1$ we have

$$-i_{+}(\sigma) = \sum_{n=1}^\infty (-1)^n U_n(\sigma) \int_0^1 x^n \log(x)^2\,dx \\= \sum_{k=1}^\infty (-1)^n U_n(\sigma) \frac{2}{(1+n)^3} \\=\sigma \int_0^1 \sum_{n=1}^\infty (-1)^n \frac{1-(\sigma z)^n}{1-\sigma z}\frac{2}{(1+n)^3}\,dz\tag{4a}$$

$$=-2\sigma \int_0^1\frac{4 \operatorname{Li}_3(-z \sigma )+3 \sigma z \zeta (3)}{2 z (1-\sigma z)}\,dz\tag{5a}$$

and here I am stuck.

While $(5)$ could be easily solved $(5a)$ is resistant.

My blocking point is this integral

$$\int_0^s \frac{\operatorname{Li}_3(t)}{1+t} \, dt\tag{7}$$

Ali Shather kindly pointed out to me that the integral for $s=1$ is well known, and he provided the useful hint that expanding the denominator gives

$$\int_0^1 \frac{\operatorname{Li}_3(x)}{1+x} \, dt=\sum_{n=1}^{\infty} (-1)^{n-1}\int_0^1 x^{n-1}\operatorname{Li}_3(x)\,dx \\= \sum_{n=1}^{\infty} (-1)^{n-1}\left(\frac{ \zeta(3)}{ n} -\frac{\zeta(2)}{ n^2}+\frac{H_{n}}{n^3}\right)\tag{8}$$

so that, finally borrowing the result for $\sum_{k=1}^{\infty}(-1)^{k+1} \frac{H_{k}}{k^3}$ from Alternating harmonic sum $\sum_{k\geq 1}\frac{(-1)^k}{k^3}H_k$ I find

$$\int_0^1 \frac{\operatorname{Li}_3(x)}{1+x} \, dx= -2 \text{Li}_4\left(\frac{1}{2}\right)-\frac{3}{4} \zeta (3) \log (2)+\frac{\pi ^4}{60}-\frac{1}{12} \log ^4(2)+\frac{1}{12} \pi ^2 \log ^2(2)\simeq 0.339545\tag{9}$$

But, unfortunately, this does not solve the problem in question here which requires the integral $(7)$ as a function of $s$.


[1] Nice pair of trilog integrals $\int_0^z \frac{\log ^2(x) \log (1\pm x)}{1\mp x} \, dx$

[2] How to calculate the generating function of the unified harmonic sum $U(\sigma,n,p) = \sum_{k=1}^n \frac{\sigma^k}{k^p}$?

  • $\begingroup$ nice problem Wolf (+1) . Regarding $i_{-}(\sigma)$, just use taylor series for $\ln(1-\sigma x)$ and use $\int_0^1\frac{x^n \ln^2x}{1-x}\ dx=2\zeta(3)-2H_n^{(3)}$. $\endgroup$ – Ali Shather Mar 20 at 18:30
  • $\begingroup$ Thanks, Ali. What do you think about the integral $\int \frac{\operatorname{Li}_3(t)}{1+t} \, dt$? $\endgroup$ – Dr. Wolfgang Hintze Mar 21 at 16:57
  • $\begingroup$ You welcome. I know this integral if its definite from zero to one. $\endgroup$ – Ali Shather Mar 21 at 17:47
  • $\begingroup$ I am moving to and fro between the integral and the sum $\sum _{k=1}^{\infty } \frac{(-1)^k H_k}{k^3}$ Could you give me please a hint to $\int_0^1 \frac{\operatorname{Li}_3(t)}{1+t} \, dt$? $\endgroup$ – Dr. Wolfgang Hintze Mar 21 at 18:24
  • $\begingroup$ Sure Wolf just a min $\endgroup$ – Ali Shather Mar 21 at 18:26

Start with writing


$$\int\frac{\text{Li}_3(x)}{1+x}\ dx=\frac14\int\frac{\text{Li}_3(x^2)}{1+x}\ dx-\int\frac{\text{Li}_3(-x)}{1+x}\ dx=\frac14I_1-I_2$$

For $I_2$, apply integration by parts twice we get


For $I_1$

$$I_1=\int\frac{\text{Li}_3(x^2)}{1+x}\ dx=\int\frac{(1-x)\text{Li}_3(x^2)}{1-x^2}\ dx$$

$$=\int\frac{\text{Li}_3(x^2)}{1-x^2}\ dx-\int\frac{x\text{Li}_3(x^2)}{1-x^2}\ dx=I_1'-I_1''$$


$$I_1''=\int\frac{x\text{Li}_3(x^2)}{1-x^2}\ dx\overset{x^2=y}{=}\frac12\int\frac{\text{Li}_3(y)}{1-y}\ dy$$



For $I_1'$ we can simply use the genrating function

$$\sum_{n=1}^\infty H_n^{(a)}x^n=\frac{\text{Li}_a(x)}{1-x}$$

Set $a=3$ and replace $x$ with $x^2$ we get that

$$I_1'=\int\frac{\text{Li}_3(x^2)}{1-x^2}\ dx=\sum_{n=1}^\infty H_n^{(3)}\int x^{2n}\ dx=\sum_{n=1}^\infty \frac{H_n^{(3)}}{2n+1}x^{2n+1}$$

Unfortunately I was not able to find the closed form of the latter generating sum. I hope you find this approach helpful.

I also found that

$$2\int_0^s\frac{\text{Li}_3(t)}{1+t}\ dt=\ln(1+s)\text{Li}_3(s)+\ln(1-s)\text{Li}_3(-s)+\text{Li}_2(-s)\text{Li}_2(s)+\int_{-s}^s\frac{\text{Li}_3(t)}{1+t}\ dt$$

| cite | improve this answer | |
  • $\begingroup$ @ Ali Shather That' nice. I have also found your last line. I have tried this: $\frac{\text{Li}_3(t)}{t+1} =\frac{1-t}{1+t} \sum _{n=1}^{\infty } H_n^{(3)} t^n $ Integrating leads to incomplete Beta functions: $\int_0^z \frac{1-t}{1+t} t^n \, dt = (-1)^{n+1} (B_{-z}(n+1,0)+B_{-z}(n+2,0))$. But finally I get into a circle. $\endgroup$ – Dr. Wolfgang Hintze Mar 23 at 15:23
  • $\begingroup$ @Dr. Wolfgang Hintze Thank you. yes it seems tough. maybe there is no closed form? $\endgroup$ – Ali Shather Mar 24 at 4:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.