6

$\Large{\text{Introduction:}}$

Here is a link to the Inverse Regularized Incomplete Gamma function used in this problem.

For simplicity, let the unit interval be expressed as $I=[0,1]$: $$\mathfrak{Q}=\int_I Q^{-1}(x,x)dx=.1984234858….$$ Here is a visual representation of this constant. Note that the constant is the area over the domain for $$x\in\Bbb R,(0,0):$$

enter image description here

$\large{\text{Identities}}$:

\begin{gather*} x=Q\bigl(x,Q^{-1}(x,x)\bigr)=Q^{-1}(x,Q(x,x)) \\ \Gamma\bigl(x,Q^{-1}(x)\bigr)=x!=\Gamma(x+1)=x\Gamma(x)=\int_{Q^{-1}(x,x)}^{\infty}t^{x-1}e^{-t}dt=\int_0^{\infty}t^xe^{-t}dt \\ \Gamma\left(x,f(x)\right)=\Gamma(x+1), f(x)=Q^{-1}(x,x). \end{gather*}

$\large{\text{Sum Representation:}}$

Here is the function’s sum representation up to 10 terms. Please look here for patterns as denominator looks easy to find.

$$ Q^{-1}(x,x)=((1-x)x!)^\frac1x+\frac {((1-x)x!)^\frac2x}{x+1}+ \frac {(3x+5)((1-x)x!)^\frac3x}{2(x+1)^2(x+2)}+O\left((x-1)^\frac4x \right)$$

Therefore:$$\int_0^1 Q^{-1}(x,x)dx=\sum_{n=1}^\infty\int_0^1C_{n,x}(1-x)^{n \over x} x!^{n\over x}dx$$

Here is graphical analytic continuation.

I need to find a formula for $C_x$ to solve this problem. $$C_{1,x}=1,C_{2,x}=\frac{1}{x+1},C_{3,x}=\frac{3x+5}{2(x+1)^2(x+2)}$$

An exact solution is preferred. Please correct me with feedback!

I think I found an exact form for the first term…

Тyma Gaidash
  • 12,081
  • 3
    I mean it's a hard question, so there is no point in getting hopes up high. However, from my side, I wish you the best of luck in finding the answer :) – tryst with freedom May 16 '21 at 20:23
  • It would be better to approximate. Must you find a closed form? You could just use numerical methods. – Arbuja May 17 '21 at 00:21
  • 1
    @TymaGaidash I’m nowhere near your level. I have crazy intuition about measures but I have no formal background. I can’t help you, sorry :( But, if a closed form exists, hopefully someone will find it. – Arbuja May 18 '21 at 00:46
  • 1
    Thanks, I wish you could help me, but I guess measure theory isn’t your feild. It’s nice chatting, good luck! – Arbuja May 18 '21 at 00:49
  • 1
    Very confusing. What are the domain and range of $Q^{-1}$? I see $Q^{-1}(x)$, $Q^{-1}(x,x)$, and $Q^{-1}(x,x,x)$. What are the domain and range of $Q$? How is anyone going to find any kind of formula for $\int Q^{-1}$ when there isn't any kind of formula for $Q^{-1}$ or $Q$? – Gerry Myerson May 21 '21 at 11:57
  • 1
    If you are suggesting the upvote was mine, I can assure you that it wasn't. – Gerry Myerson May 22 '21 at 00:43
  • 1
    This is an interesting question for sure and I would enjoy to see the solution. Would you accept to ping me when you get an answer ? Thanks and $\to +1$. Cheers :-) – Claude Leibovici May 22 '21 at 08:06
  • @ClaudeLeibovici that is what the book mark is for! I already have two solutions. One is using the Riemann Sum definition and the other uses the nth derivative of this function for an “Obvious Taylor Series”. I do not have too many other ideas on how to progress.Similar question 1,Similar 2. My solutions are cumbersome and “bad”. – Тyma Gaidash May 22 '21 at 13:25
  • Thanks for your concern ! I was unable to thank you earlier with a reputation of $1$. I am back (and happy to be back). Cheers :-) – Claude Leibovici Jul 09 '21 at 09:52
  • I realized there are many edits, but this post has been overhauled. – Тyma Gaidash Aug 22 '21 at 19:16
  • @ClaudeLeibovici As you requested I may have found a solution to the integral. What do you think and how can I finish? – Тyma Gaidash Aug 24 '21 at 18:40

1 Answers1

2

$\frak Q$:

Note this answer needs help from you to find the full solution.

Amazingly there is another identity found based on this similar question about

Finding $\int_0^{.25991…} Q^{-1}(x,x,x) dx$

which shows it can actually solved for. Here are some gamma functions for context.

It actually is possible to integrate $Q^{-1}(x,x)$ many ways, but one way which does not use a general definite integration formula is harder to find. I suspect a closed form is impossible using currently accepted functions. Let’s start with the series representation of $Q^{-1}(x,x)$. This is the notation I will use called $C_{n,x}$ and its definition:

$$Q^{-1}(x,x)=Q^{-1}(x,x,Q(x,x)-x)=((1-x)x!)^\frac1x+\frac {((1-x)x!)^\frac2x}{x+1}+ \frac {(3x+5)((1-x)x!)^\frac3x}{2(x+1)^2(x+2)}+O\left((x-1)^\frac4x \right)= C_{1,x}(1-x)^{1\over x} x!^{1\over x} + C_{2,x}(1-x)^{2 \over x} x!^{2\over x} + C_{3,x}(1-x)^{3 \over x} x!^{3\over x}+…= \sum_{n=1}^\infty C_{n,x}(1-x)^{n \over x} x!^{n\over x} $$

Therefore:$$\int_0^1 Q^{-1}(x,x)dx=\sum_{n=1}^\infty\int_0^1C_{n,x}((1-x)x!)^{\frac nx} dx$$

The next step is to use a Taylor series of $((1-x)x!)^{\frac nx}$ at x=0 which has the amazing form of the following with Apéry’s constant, the Euler-Mascheroni constant, π, and e. Feel free to simplify it as it appears to have a higher degree polynomial with each term. Notice how $\pi$ and ζ($3$) also exhibit this polynomial behavior with each increasing m:

$$\sum_{m=0}^\infty\frac{\left[\frac{d^m}{dx^m} ((1-x)x!)^{\frac nx}\right]_{x=0}x^m}{m!}= e^{-(1 + γ ) n}\left[1 + \frac{π^2 - 6} {12} n x +\frac { nx^2 ((π^2 - 6)^2 n - 96 (ζ(3) +1)) } {288} + \frac{n x^3 (5 (π^2 - 6)^3 n^2 -1440 (π^2 - 6) n (ζ(3) +1)+ 144 (π^4 - 90))}{51840} +\frac{ n x^4 (5 (π^2 - 6)^4 n^3 -2880 (π^2 - 6)^2 n^2 (ζ(3) +1) + 576 n (-90 π^2 - 6 π^4 + π^6 + 60 (13 +8ζ(3) + 4ζ^2(3))) -497664 (ζ(5) ))}{2488320} + O(x^5)\right] $$

This series converges to $((1-x)x!)^{\frac nx},0\le x\le 1$, so here is the constant so far:

$$\int_0^1 Q^{-1}(x,x)dx= \sum_{n=1}^\infty\int_0^1C_{n,x} \sum_{m=0}^\infty\frac{\left[\frac{d^m}{dx^m} ((1-x)x!)^{\frac nx}\right]_{x=0}x^m}{m!} dx = \sum_{n=1}^\infty\sum_{m=0}^\infty\frac{\left[\frac{d^m}{dx^m} ((1-x)x!)^{\frac nx}\right]_{x=0}}{m!} \int_0^1 C_{n,x} x^m dx$$

Here is where I need to find an expression for $C_{n,x}$. I already know the first few terms using this generalized power series. Here is the constant which should hopefully converge using the n=1,2,3 cases of $C_{n,x}$. Note I used Wolfram Functions and Wolfram Alpha to evaluate and simplify. Note that $H_y$ are the Harmonic numbers and Ф(a,b,c) is the Lerch Transcendent/Zeta function.

Note that I could have integrated up to 4 or more terms, but this becomes tedious and has more interesting complicated functions:

$${\frak Q}=\int_0^1 Q^{-1}(x,x)dx=\int_0^1 P^{-1}(x,x)=\int_0^1 Q^{-1}(x,x,Q(x,x)-x)dx=\int \sum_{m=0}^\infty\frac{\left[\frac{d^m}{dx^m} ((1-x)x!)^{\frac 1x}\right]_{x=0}}{m!} \int_0^1 x^m +\sum_{m=0}^\infty\frac{\left[\frac{d^m}{dx^m} ((1-x)x!)^{\frac 2x}\right]_{x=0}}{m!} \int_0^1\frac{1}{x+1} x^m +\sum_{m=0}^\infty\frac{\left[\frac{d^m}{dx^m} ((1-x)x!)^{\frac 3x}\right]_{x=0}}{m!} \int_0^1\frac {(3x+5)}{2(x+1)^2(x+2)} x^m + \sum_{m=0}^\infty\frac{\left[\frac{d^m}{dx^m} ((1-x)x!)^{\frac 4x}\right]_{x=0}}{m!} \int_0^1\frac{8x^2+33x+31}{3(x+1)^3(x+2)(x+3)}x^m+…=\sum_{m=0}^\infty\frac{\left[\frac{d^m}{dx^m} ((1-x)x!)^{\frac 1x}\right]_{x=0}}{(m+1)!} +\frac12 \sum_{m=0}^\infty\frac{\left[\frac{d^m}{dx^m} ((1-x)x!)^{\frac 2x}\right]_{x=0}}{m!} \left(H_\frac m2-H_\frac{m-1}{2}\right)+ \frac14 \sum_{m=0}^\infty\frac{\left[\frac{d^m}{dx^m} ((1-x)x!)^{\frac 3x}\right]_{x=0}}{m!}\left((2 m - 1) H_\frac m2 - (2 m - 1) H_\frac{m -1}2 -Ф\left(-\frac 12,1,m+1\right)+2\right)+…= \sum_{m=0}^\infty\frac{\left[\frac{d^m}{dx^m} ((1-x)x!)^{\frac 1x}\right]_{x=0}+ \left[\frac{d^m}{dx^m} ((1-x)x!)^{\frac 2x}\right]_{x=0}+ \left[\frac{d^m}{dx^m} ((1-x)x!)^{\frac 3x}\right]_{x=0}+…}{m!} \left[ \frac{1}{m+1}+\frac12 \left(H_\frac m2-H_\frac{m-1}{2}\right)+ \frac14\left((2 m - 1) H_\frac m2 - (2 m - 1) H_\frac{m -1}2 -Ф\left(-\frac 12,1,m+1\right)+2\right)+…\right]=.1984234858… $$

If this formula works, then what is an explicit form for $C_{n,x}$? An explicit formula for $\sum_{m=0}^\infty\frac{\left[\frac{d^m}{dx^m} ((1-x)x!)^{\frac nx}\right]_{x=0}x^m}{m!}$ is also appreciated as its expansion seems to be a double sum of n and x. Any other simplifications are also appreciated.

This leads to the fact that an amazing approximation for $\frak Q$ is the following and proves the series representation of the constant. Note that the rest of the $n\ge 1$ terms will add on smaller values making the approximation better each time:

$${\frak Q}= .19842…≈ \frac{588 + 24 π^2 + π^4 - 96 ζ(3)}{864 e^{γ+1}}=.19289…$$

Note there may be a typo. I hope this shows you that a general Taylor series is all you need to integrate such problems with inverse gamma type functions. Please correct me and give me feedback!

$\mathcal I$:

The real purpose of this question was to find a convergent interval found by @qifeng618:

$$\mathcal I=\int_0^1 \text I^{-1}_x(x,x)=.45697517...$$

which does graphically converge on $0<x\le 2$ from the question.

The area is the area over the function’s domain for $x\in\Bbb R$ Here is the graph of this function:

enter image description here

Here is the definition of the function and its properties along with some various beta functions:

$$x=\text I_{I^{-1}_x(y,z)}(y,z)= \text I_{\text I^{-1}_x(x,x)}(x,x)=\frac{\int_0^{\text I^{-1}_x(x,x)}\left(t-t^2\right)^{x-1}dt}{\text B(x,x)}\implies x \text B(x)= \text B_{\text I^{-1}_x(x,x)}(x,x)= \int_0^{\text I^{-1}_x(x,x)}\left(t-t^2\right)^{x-1}dt$$

See this link to derive more results from the definition of the function. I did find this possible form.

Let’s actually begin solving the integral with this series expansion for the Inverse of the Regularized Incomplete Beta function. See the question where @qifeng618 answers well for the first term approximation. Here appears the definition of our $B_{n,x}$ coefficients unrelated to other functions:

$$\text I^{-1}_x(x,x)\mathop=^{x>0}(x^2 \text B(x,x))^\frac 1 x+\frac{x-1}{x+1}(x^2 \text B(x,x))^\frac2x+\frac{2(x-1)(x^2+x-1)}{(x+1)^2(x+2)} (x^2 \text B(x,x))^\frac 3 x+…=B_{1,x} (x^2 \text B(x,x))^\frac 1 x +B_{2,x} (x^2 \text B(x,x))^\frac 2 x +B_{3,x} (x^2 \text B(x,x))^\frac 1 x +…$$

Therefore:

$$\mathcal I=\int_0^1 \text I^{-1}_x(x,x) =\int_0^1 \sum_{n=0}^\infty B_{n,x} (x^2 \text B(x,x))^\frac n x dx= \int_0^1 \sum_{n=0}^\infty B_{n,x} \frac{x!^\frac{2n}{x}}{Γ(2x)^\frac nx} dx $$

Let’s now use the Taylor series for $x!^\frac{2n}{x}$ at $x=0$ for easier integration. Our choice of integration will have consequences for future steps. I could easily have used a more complex Taylor Series, but it would be too complicated to find a formula for. It would also be impractical. Here is the series converging almost certainly for $0\le x\le 2$: …

Other constant:

Note we also have $$\int_0^.489 \text{erf}^{-1}(x,x)dx$$ as the Inverse of the Generalized Error function. This will be the place for the other main “constants”.

This is a work in progress for fun

Тyma Gaidash
  • 12,081