3

I am trying to solve without success the following double integral $$I_1^{(p)}(N)\equiv\frac{1}{2^p}\int_0^1\text{d}x\int_0^1\text{d}y(1+y-x)^{N+p}(1+x-y)^{N-2}B\left(\frac{1}{1+y-x};N,p+1\right)\cdot\theta(y-x)\theta(1-x-y),$$ where $N\in\mathbb{N}$, $p>0$, $\theta(x)$ is the Heaviside step function and $B(x;a,b)$ is the incomplete beta function $$B(x;a,b)=\int_0^xt^{a-1}(1-t)^{b-1}\text{d}t.$$

The product of the two $\theta$ functions can be translated into one of the two following constraints

1) $\quad x\in\left(0,\frac{1}{2}\right)\longrightarrow x<y<1-x$

2) $\quad y\in\left(0,\frac{1}{2}\right)\longrightarrow x<y,\quad y\in\left(\frac{1}{2},1\right)\longrightarrow x<1-y$

so, e.g. in the first case, the integral becomes $$I_1^{(p)}(N)=\frac{1}{2^p}\int_0^{\frac{1}{2}}\text{d}x\int_x^{1-x}\text{d}y(1+y-x)^{N+p}(1+x-y)^{N-2}B\left(\frac{1}{1+y-x};N,p+1\right).$$ At this point I tried some substitutions, such as $t=\frac{y-x}{1-2x}$ in order to get $\int_x^{1-x}\text{d}y\rightarrow\int_0^1\text{d}t$, but the expression remained not tractable for me.

The same happened when I rewrote the incomplete beta function in terms of a hypergeometric one, e.g. by $$B(x;a,b)=\frac{x^a(1-x)^{b-1}}{a}{}_2F_1\left(1,1-b;a+1;\frac{x}{x-1}\right),$$ hoping to be able to use one of the relations that I found here.

Any help would be greatly appreciated.

Edit 1. The above integral is part of a larger expression, containing two other similar terms which can be obtained from $I_1^{(p)}(N)$ with the following substitutions respectively

$I_2^{(p)}(N):\quad B\left(\frac{1}{1+y-x};N,p+1\right)\rightarrow -B\left(\frac{1-y-x}{1+y-x};N,p+1\right)$

$I_3^{(p)}(N):\quad (1+y-x)^{N+p}B\left(\frac{1}{1+y-x};N,p+1\right)\rightarrow 2^p(1-x)^{N+p}B\left(\frac{1-y-x}{1-x};N,p+1\right)$

The structure does not change that much, so I thought that the solution procedure could be similar in the three cases. Nonetheless, probably it is better to report every detail, because the hypothesis that a simplication can occur between different terms cannot be discharged, even if I failed in doing that.


Edit 2. Proceeding as illustrated by @G Cab in his answer below, the result I obtained is \begin{equation}\begin{split} I_1^{(p)}(N)&=2^{2N-1}\left[B(N+p+1,N)B\left(\frac{1}{2};N,p+1\right)-B(N,p+1)B\left(\frac{1}{2};N+p+1,N\right)\right.\\[6pt] &\left.\quad+\int_{\frac{1}{2}}^1t^{N-1}(1-t)^pB\left(\frac{1}{2t};N+p+1,N\right)\text{d}t\right]. \end{split}\end{equation} I am pretty satisfied by the simplication with respect to the starting expression, but now I wonder whether the remaining single integral can be elaborated somehow.

ARWarrior
  • 299

1 Answers1

1

a) the Incomplete Beta function is $$ B \left( {x\;;a,b} \right) = \int_{t\, = \,0}^{\;x} {t^{\,a - 1} \left( {1 - t} \right)^{\,b - 1} dt} $$ the integration variable is different from the upper bound

b) It might be useful to change the Step function with the Iverson bracket.

So $$ \begin{split} I^{\left( p \right)} (N) &= \frac{1}{2^p} \int\limits_{x = 0}^{1} \int\limits_{y = 0}^{1} ( {1 + y - x} )^{\,N + p}( {1 + {x - y} } )^{\,N - 2} \\ &\qquad\cdot B\left( {{1 \over {1 + y - x}}\;;N,p + 1} \right)[ {0 \le y - x} ][ {y + x \le 1} ]dx\;dy \\ \\ & = \frac{1}{2^p}\int\limits_{x = 0}^{1} \int\limits_{y = 0}^{1} \!\!\int\limits_{t = 0}^{\frac{1}{1 + y - x}} ( {1 + y - x})^{\,N + p}( {1 + {x - y} } )^{\,N - 2} t^{\,N - 1} \left( {1 - t} \right)^{\,p} \\ &\qquad\cdot[ {0 \le y - x} ][ {y + x \le 1} ]dx\;dy\,dt \\ \\ & = \frac{1}{2^p}\iiint\limits_{(x,y,t) \in V} ( {1 + y - x})^{N + p} ( {1 + {x - y} } )^{\,N - 2} t^{N - 1} ( {1 - t} )^{\,p} dx dy dt \end{split} $$ where $$ V = \left\{ {(x,y,t)} \right\}:\;\;\left\{ \matrix{ 0 \le x \le 1 \hfill \cr 0 \le y \le 1 \hfill \cr 0 \le y - x \hfill \cr y + x \le 1 \hfill \cr 0 \le t \le {1 \over {1 + y - x}} \hfill \cr} \right.\quad \Rightarrow \quad \left\{ \matrix{ 0 \le x \le 1/2 \hfill \cr 0 \le x \le y \le 1 - x \hfill \cr 0 \le t \le {1 \over {1 + y - x}} \hfill \cr} \right. $$

Now it remains to change the variables appropriately, so that we can integrate in $t$ at last, after the others, and to reformulate accordingly the bounds on $V$.
Proceeding further the change of variables is $$\begin{cases} v=1+y-x\\ u=1-y-x\end{cases}\quad\Longrightarrow\quad\begin{cases}x=1-\frac{v+u}{2}\\ y=\frac{v-u}{2}\end{cases}$$ and so one obtains $dxdy=\frac{1}{2}dvdu$. It is also easy to check that the domain $V$ splits into two different parts $$V_1=\{(u,v,t)\}:\begin{cases}0<t<\frac{1}{2}\\ 1<v<2\\ 0<u<2-v\end{cases}\qquad\quad V_2=\{(u,v,t)\}:\begin{cases}\frac{1}{2}<t<1\\ 1<v<\frac{1}{t}\\ 0<u<2-v\end{cases} $$ This leads us to $$ \begin{split} I^{\left( p \right)} (N) &= \frac{1}{2^{p+1}}\iiint\limits_{(u,v,t) \in V_1\cup V_2} v^{N + p} u^{N - 2} t^{N - 1} ( {1 - t} )^{p} du dv dt \\ & = \frac{1}{2^{p+1}}\left[\;\int\limits_{t = 0}^{\frac{1}{2}} t^{N - 1} ( {1 - t} )^{p} dt\int\limits_{v = 1}^{2} v^{N + p} ( {2 - v} )^{N - 2} dv \int\limits_{u = 0}^{2 - v} {du} \right.\\ &\quad\left.+\int\limits_{t = \frac{1}{2}}^{1} t^{N - 1} ( {1 - t} )^{p} dt\int\limits_{v = 1}^{\frac{1}{t}} v^{N + p} ( {2 - v} )^{N - 2} dv \int\limits_{u = 0}^{2 - v} {du} \right]\\ & = \frac{1}{2^{p+1}}\left[\;\int\limits_{t = 0}^{\,\frac{1}{2}} t^{\,N - 1} ( {1 - t} )^{p} dt\int\limits_{v = 1}^{2} {v^{N + p} ( {2 - v} )^{N - 1} dv}\right.\\ &\left.\quad+\int\limits_{t = \frac{1}{2}}^{\,1} t^{\,N - 1} ( {1 - t} )^{p} dt\int\limits_{v = 1}^{\frac{1}{t}} {v^{N + p} ( {2 - v} )^{N - 1} dv} \right]\\ & = 2^{2N -1} \left[\;\int\limits_{t = 0}^{\,\frac{1}{2}} {t^{N - 1} ( {1 - t} )^{p} dt \int\limits_{\frac{v}{2} = \frac{1}{2}}^{1} {\left( {{v \over 2}} \right)^{N + p} \left( {1 - {v \over 2}} \right)^{N - 1} d\left( {{v \over 2}} \right)} }\right.\\ &\left.\quad+\int\limits_{t = \frac{1}{2}}^{\,1} {t^{N - 1} ( {1 - t} )^{p} dt \int\limits_{\frac{v}{2} = \frac{1}{2}}^{\frac{1}{2t}} {\left( {{v \over 2}} \right)^{N + p} \left( {1 - {v \over 2}} \right)^{N - 1} d\left( {{v \over 2}} \right)} } \right] = \\ & = 2^{2N -1}\left\{\;\int\limits_{t = 0}^{\frac{1}{2}} {t^{N - 1} \left( {1 - t} \right)^{p} {B\left( {\frac{1}{2};N,N+p+1} \right) } dt}\right.\\ &\left.\quad+\int\limits_{t = \frac{1}{2}}^{1} {t^{N - 1} \left( {1 - t} \right)^{p} \left[ {B\left( {\frac{1}{2t};N + p + 1,N} \right) - B\left( {\frac{1}{2};N + p + 1,N} \right)} \right]dt}\right\} \end{split} $$

ARWarrior
  • 299
G Cab
  • 35,272
  • a) Silly mistake, I corrected the typo! b) Thank you very much for your suggestions, I'll try to proceed further hoping for a result. I'll let you know! – ARWarrior Jun 10 '19 at 21:33
  • @ARWarrior: good . Yes thanks, I would like to know the result. Let me know if I might still be of help – G Cab Jun 10 '19 at 22:02
  • @G Cab: unfortunately I'm quite busy at the moment, I'll have to wait some days before I can concentrate on these computations. Especially considering that, if I got it well, obtaining a closed form is not guaranteed even with your approach. Maybe I'll need to add the two further contributions cited in the post in order to simplify something during the substitutions. – ARWarrior Jun 10 '19 at 22:59
  • This is surely a great simplication, considering that the starting expression was a double integral. I'll investigate whether the first term above can be integrated somehow, but before I'll check if the other two starting integrals lead to similar expressions when manipulated as you pointed out to me. Again thank you! – ARWarrior Jun 11 '19 at 07:46
  • I just realized that probably you didn't notice my edit on your answer, in which I corrected the typo $(1-y-x)^{N-2}\rightarrow(1+x-y)^{N-2}$, and so you proceeded with the calculations with the old formula. Your substitutions continue to be valid, nothing important changes, excepts for the factor $1/(N-1)$ that disappears in the final result. – ARWarrior Jun 11 '19 at 08:58
  • I just edited (awaiting peer review) your post doing what I explained in my previous comment. Besides this I found another little mistake in the definition of $V$ when you pass from $x,y$ to $u,v$. In fact, together with the interval $;1/2<t<1,;1<v<1/t;$ there should be also $;0<t<1/2,;1<v<2$. Adding this piece, a new contribution emerges, which turns into $2^{2N+p-1}\int_0^{1/2}t^{N-1}(1-t)^p B(1/2;N,N+p+1)\text{d}t$. I checked the result numerically and now it is correct. – ARWarrior Jun 11 '19 at 10:04
  • @ARWarrior: sorry, in fact I didn't catch the typo, and continue with that: then clearly also the variable change and the inequalities are affected. However, the goal I was aiming to remains: the integral splits into (one or two) products of Betas plus a "double beta integral". I don't have presently an idea of how to deal with it, except integrating by parts, maybe reaching to a recursion .. – G Cab Jun 11 '19 at 13:02