8

In many tetration/power tower integrals, one sees a general form of the following. Let this new function be notation used to show the connection between the general result and special cases using types of Incomplete Gamma functions.

The goal is to find an integral representation of the general case or a special case.

If this is not possible, then maybe a special case of it has an integral representation. Note there are ways to put the summand into other functions, but this way is simple. Please note that I will use a made up general “T” function to show how each integral below it is a special case of the following:

$$T_{p,q}^{r,s}\left(_{\ \ a,b,c,d}^{A,B,C,D}\right)=\sum_{n=0}^\infty \frac{(pn+q)^{rn+s}Γ(An+B,Cn+D)}{Γ(an+b,cn+d)}$$

Here is motivation that integral representations are possible. Note that I will use the primitive for simplicity:

$$\int (cx)^{ax^b}dx=\sum_{n=0}^\infty\frac{(-a)^n Q(n+1,-(bn+1)\ln(cx))}{c^{bn+1}(bn+1)^{n+1}}= \frac{1}{-ac^{b-1}}\sum_{n=0}^\infty\frac{ (-ac^bbn-ac^b)^{n+1}Γ(n+1,-bn\ln(cx)-\ln(cx))}{Γ(n+1,0)} = -\frac{1}{ac^{b-1}} T_{-abc^b,-ac^b}^{1,1}\left(_{\ \ 1,1,0,0}^{1,1,-b\ln(cx),-\ln(cx)}\right) $$

$$\int a^{ta^t}dt=t+\frac{1}{\ln(a)}\sum_{n=0}^\infty \frac {(-1)^n Q(n+1,-nt\,\ln(a))}{n^{n+1}}= t-\frac{1}{\ln(a)}\sum_{n=0}^\infty \frac {(-n)^{-n-1} Γ(n+1,-nt\,\ln(a))}{Γ(n+1,0)} =t-\frac{1}{\ln(a)} T_{-1,0}^{-1,-1}\left(_{\ \ 1,1,0,0}^{1,1,-t\,\ln(a),0}\right) $$

$$\int \frac{dx}{xe^x-1}=\sum_{n=0}^\infty \frac{Γ(n+1,-nx)}{n^{n+1}}= \sum_{n=0}^\infty n^{-n-1} Γ(n+1,-nx)= T_{1,0}^{-1,-1}\left(_{0,1,0,0}^{1,1,-x,0}\right) $$

$$\int \text W(\ln(x))dx=\text W(\ln(x))(x-1)+\sum_{n=1}^\infty\frac{(-1)^n Q(n+1,-n\,\text W(\ln(x))}{n^{n+1}}= \text W(\ln(x))(x-1) -\sum_{n=1}^\infty\frac{(-n)^{-n-1}Γ(n+1,-n\,\text W(\ln(x))}{Γ(n+1)} = T_{-1,0}^{-1,-1}\left(_{1,1,0,0}^{1,1, -\,\text W(\ln(x)),0}\right) $$

Miscellaneous sums of interest. Subfactorial: $$\sum_{n=2}^\infty \frac{1}{!n}=e\sum_{n=0}^\infty \frac{1}{Γ(n+2,-1)}=e\,T_{p,q}^{0,0}\left(_{1,2,0,-1}^{0,1,0,0}\right)$$

$$\sum_{n=-\infty}^{-1} Γ(n,n)=\sum_{n=0}^\infty Γ(-n-1,-n-1)= T_{p,q}^{0,0}\left(_{-1,-1,-1,-1}^{\quad 0,1,0,0}\right) $$

I already know about the Abel-Plana formula, but it offers no new insights. There are other theorems that could possibly be used. How can the integral representations for the goal sum be found? If the integral representation is indeed impossible, then what is an integral representation for a special case? This will help us solve similar problems. Please correct me and give me feedback!

Тyma Gaidash
  • 12,081
  • 1
    Did you just randomly spotted the $T$ function? Or is the a motivation? – Rounak Sarkar Sep 09 '21 at 16:00
  • 1
    @RounakSarkar As said in the bounty and question, it is a function that I made up just to show how each example is a special case of the general case. It is a motivation. Do you know how to possibly put it in an integral form? – Тyma Gaidash Sep 09 '21 at 20:44
  • 1
    Nah bro. I am not intelligent enough. But looking at YOUR question history, it seems that you'll find out what you want very quickly. – Rounak Sarkar Sep 10 '21 at 13:55
  • 1
    @RounakSarkar What do you mean by “you’ll find out what you want very quickly”? This website is just a hobby. Thanks for the support. – Тyma Gaidash Sep 10 '21 at 14:48
  • 1
    @RounakSarkar One last thing, there seems to have been an upvote from you associated with this inverse gamma integral question. I have checked part of the expansion and it works. Can you please check if the rest works? Thanks. – Тyma Gaidash Sep 11 '21 at 02:30
  • 1
    Wait how did you figure out that I upvoted that answer? And by the way can you answer this question of mine? – Rounak Sarkar Sep 11 '21 at 04:30
  • 1
    One last thing. If I want to try to find the integral representation of the $T$ function, how do I start? I mean how does someone find an integral representation? – Rounak Sarkar Sep 12 '21 at 12:28
  • 1
    @RounakSarkar If the result is correct and while ignoring conditions for now, $$\int (cx)^{ax^b}dx=\sum_{n=0}^\infty\frac{(-a)^n Q(n+1,-(bn+1)\ln(cx))}{c^{bn+1}(bn+1)^{n+1}}= \frac{1}{-ac^{b-1}}\sum_{n=0}^\infty\frac{ (-ac^bbn-ac^b)^{n+1}Γ(n+1,-bn\ln(cx)-\ln(cx))}{Γ(n+1,0)} = -\frac{1}{ac^{b-1}} T_{-abc^b,-ac^b}^{1,1}\left(_{\ \ 1,1,0,0}^{1,1,-b\ln(cx),-\ln(cx)}\right) $$ already is an integral representation. Just start with an integrand and experiment and you might find a more general answer.. Thanks again. If you cannot find a general solution then try a special case of the general.Thanks – Тyma Gaidash Sep 12 '21 at 12:37
  • 1
    You're right. I still have some time for getting +50 reputation. I don't think I can do it, but goddammit let me just try. – Rounak Sarkar Sep 12 '21 at 12:49
  • 1
    @RounakSarkar Go ahead, just do not stress too much about it. You also have the whole internet to your disposal. Thanks. – Тyma Gaidash Sep 12 '21 at 12:52
  • I just realized that for the second expression in your bounty, basically $T_{p,q}^{r,s}\left({\ \ a,b,0,0}^{A,B,C,D}\right)$ can be easily written as a double integral. You can use the definition of the incomplete gamma function to write $\Gamma(An+B,Cn+D)$ or you can just use $\Gamma(s,x)=x^s\operatorname{E}{1-s}(x)$ and then use the definition of the generalized exponential integral. And for the bottom $\Gamma(an+b)$ we can just treat it as the reciprocal gamma function and use the integral definition of it. – Rounak Sarkar Sep 12 '21 at 15:56
  • 1
    @RounakSarkar That is a nice idea, but how about the $(pn+q)^{rn+s}$ part? I see you are doing a special case, but is there an integral representation for $\frac{1}{Γ(an+b,cn+d)}$, the denominator of the main sum? – Тyma Gaidash Sep 12 '21 at 15:59
  • 1
    I think you don't need to worry about $(pn+q)^{rn+s}$ because that is not really a trouble maker. And for the general case for $\frac{1}{an+b,cn+d}$, I don't know. I don't think it would be easy to derive an integral representation of it. Should I write an answer about the special case? I know that it is not really that interesting because I just plugged in a few identities in the definition of the $T$ function, but i think it can still be considered an integral representation. – Rounak Sarkar Sep 12 '21 at 16:09
  • 1
    @RounakSarkar Notice how I only used this special case for the integrals. I was thinking I could sneak in another case for this question, but you can totally just do the special case. Please try to write the sum in terms of integrals only and no summations. Thanks. – Тyma Gaidash Sep 12 '21 at 16:16
  • 1

1 Answers1

3

This will be a very general solution. Notice we can answer any question like this one by:

$$\int f(x) dx=g(x)=\sum_{n=0}^\infty g_n(x)\implies f(x)=\sum_{n=0}^\infty g’_n(x) =g(x)$$

However, we need to find g(x) in closed form. I could do the general summation in the question by finding a general solution family, but it is more beautiful when we use the special case which can be simplified using a function.

This function will be the Mittag-Leffler function:

$$\mathrm E_{a,b}(y)=\sum_{n=0}^\infty \frac{y^n}{Γ(an+b)}$$

Now let’s solve for $y=y(x)$ by making a simple integral equation. Let’s also make each constant a function of x, so $p=p(x),q=q(x),r=r(x),s=s(x),A=A(x),B=B(x),C=C(x),D=D(x)$

$$\int E_{a,b}(y(x))dx=\sum_{n=0}^\infty \int \frac{y^n(x)}{Γ(an+b)}dx=\sum_{n=0}^\infty \frac{(pn+q)^{rn+s}Γ(An+B,Cn+D)}{Γ(an+b)}$$

I could take the nth root, but the index must stay with the summation.

After playing around with more generalized y(x), a solution was found, but two of the arguments must be 1 unfortunately unless you can find a better solution. The integration can be found using gamma identities, but I had machine help:

$$\int \mathrm E_{a,b}\left(\ln^t(ux)x^v\right)dx=\sum_{n=0}^\infty \frac{1}{Γ(an+b)}dx\int \ln^{tn}(ux)x^{vn} dx=C+\sum_{n=0}^\infty \frac{x^{n v} (u x)^{-n v} \ln^{n t}(u x) (-(n v + 1) \ln(u x))^{-n t} Γ(n t + 1, -(n v + 1) \ln(u x)) }{Γ(an+b) (n u v + u)}\quad\mathop=^{\ln^{n t}(u x) \ln^{-n t}(u x)=1}_{x^{nv}x^{-nv}=1}\quad C+\sum_{n=0}^\infty \frac{(-1)^{nt} Γ(tn + 1, -(n v + 1) \ln(u x)) }{Γ(an+b) (n v + 1)^{tn+1} u^{vn+1}} = C+\frac1u\sum_{n=0}^\infty \frac{\left(-u^{\frac vt}vn - u^{\frac vt}\right)^{-tn} Γ(tn + 1, -v\ln(u x) n -\ln(u x))) }{Γ(an+b) (n v + 1)} $$

This expansion has a factor of $(nv+1)$ in the denominator which makes it difficult to put into my “T function form”. Here is a better result using the Productlog/W-Lambert function which also uses a bit of software help:

$$\int \mathrm E_{a,b}\left(\mathrm W^{t}(ux)x^v\right)dx=\int\sum_{n=0}^\infty \frac{\mathrm W^{tn}(ux)x^{un}}{Γ(a,b)}dx=C+\sum_{n=0}^\infty \frac{\left[x^{n v} \mathrm W(u x)^{n t} e^{-n v \mathrm W(u x)} (-(n v + 1) \mathrm W(u x))^{-n (t + v)} ((n v + 1) Γ(n (t + v) + 1, -(n v + 1) \mathrm W(u x)) - Γ(n (t + v) + 2, -(n v + 1) \mathrm W(u x)))\right]}{u (n v + 1)^2 Γ(an+b)}$$

If terms can be cancelled out then, the following holds also using Lambert-W function properties:

$$\int \mathrm E_{a,b}\left(\mathrm W^{t}(ux)x^v\right)dx=C+ \sum_{n=0}^\infty \frac{\left[-u^{-vn-1} (-(n v + 1))^{-n (t + v)-1} Γ(n (t + v) + 1, -(n v + 1) \mathrm W(u x)) -u^{-vn-1}(-(n v + 1))^{-n(t+v)-2} Γ(n (t + v) + 2, -(n v + 1) \mathrm W(u x))]\right]}{Γ(an+b)} = C-u^{-\frac{t}{t+v}}\sum_{n=0}^\infty \frac{\left(-u^{\frac{v}{t+v}} n v - u^{\frac{v}{t+v}} \right)^{-n (t + v)-1} Γ(n (t + v) + 1, -n v \mathrm W(u x) -\mathrm W(u x)))}{Γ(an+b)}- u^{\frac{v-t}{t+v}}\sum_{n=0}^\infty \frac{\left(-u^{\frac{v}{t+v}} n v - u^{\frac{v}{t+v}} \right)^{-n (t + v)-1} Γ(n (t + v) + 2, -n v \mathrm W(u x) -\mathrm W(u x)))}{Γ(an+b)} $$

So we mainly found an integral representation for:

$$\sum_{n=0}^\infty \frac{(pn+q)^{rn+\{-1,0,1,2\}}Γ(An+\{1,2\},Bn+C)}{Γ(an+b)}$$

where $\{x_1,x_2,…x_i\}$ represents solved cases of a variable. If you can find a more general solution, then please let me know. Note that the Generalized Exponential Integral function simplifies the problem, but that is not the focus of the problem. Please correct me and give me feedback!

I actually wanted a closed form for this sum and I found an analogous function called the Incomplete Fox-Wright functions where the subscripted variable are constants with respect to $n$. The bolded link has more identities. Also note the Lower Gamma function:

$$\,_pΨ_q^{(\gamma)}\left[\,^{(a_1,A_1,x),(a_2,A_2),…(a_p,A_p)}_{\quad(b_1,B_1),…,(b_p,B_p)}\ t\right]+ \,_pΨ_q^{(\Gamma)}\left[\,^{(a_1,A_1,x),(a_2,A_2),…(a_p,A_p)}_{\quad(b_1,B_1),…,(b_p,B_p)}\ t\right]\mathop=^\text{def}\sum_{n=0}^\infty \frac{\Gamma(a_1+A_1n,x)\prod\limits_{j=2}^pΓ(a_j+A_jn)t^n}{\prod\limits_{j=2}^p \Gamma(b_j+B_jn)n!}+ \sum_{n=0}^\infty \frac{γ(a_1+A_1n,x)\prod\limits_{j=2}^pΓ(a_j+A_jn)t^n}{\prod\limits_{j=2}^p \Gamma(b_j+B_jn)n!} = \sum_{n=0}^\infty \frac{\prod\limits_{j=1}^pΓ(a_j+A_jn)t^n}{\prod\limits_{j=2}^p \Gamma(b_j+B_jn)n!} = \,_pΨ_q\left[\,^{(a_1,A_1,x),(a_2,A_2),…(a_p,A_p)}_{\quad(b_1,B_1),…,(b_p,B_p)}\ t\right] $$

Which is the Fox-Wright function inspired from @Harry Peter’s post and Therefore:

$$\sum_{n=0}^\infty\frac{\Gamma(A+Bn,C)t^n\Gamma(1n+1)}{\Gamma(аn+b)n!}= \sum_{n=0}^\infty\frac{\Gamma(A+Bn,C)t^n}{\Gamma(а+bn)} =\,_2Ψ^{(\Gamma)}_1\left[\,^{(A,B,C),(1,1)}_{\quad(a,b)}\ t\right]$$

which is still not enough to solve general tetration type integrals, but works for some problems. Maybe I will post some later. This function seem like an excuse for any similar sum, but still can be simplified into a few simpler function.

Тyma Gaidash
  • 12,081