Idea for computing integral (similar to beta function)

90 Views Asked by At

Does anyone have any idea about how to compute the following integral? Or any guess of what it could be?

$$\int_{s<u_1<\cdots <u_p<t} (u_1-s)^{\alpha_1/2}\cdots (u_p-u_{p-1})^{\alpha_p/2}du_1\cdots du_p$$ where $\alpha_1+\cdots+\alpha_p = p$. I was thinking of the Beta function but I do not arrive at anything useful.

Thank you guys

1

There are 1 best solutions below

2
On BEST ANSWER

One can compute it in a "beta-function way". I'll suppose $s=0$ (as it depends only on the difference $t-s$). Substitution $x_1=u_1$, $x_2=u_2-u_1$,... ,$x_p=u_p-u_{p-1}$ brings the integral to $$I(t)=\int_{x_i>0,\sum_i x_i<t}x_1^{\alpha_1/2}\dots x_p^{\alpha_p/2}dx_1\dots dx_p.$$ Clearly $$I(t)=t^{\sum_i(\alpha_i/2+1)}I(1)$$ and thus $$dI(t)/dt=\sum_i(\alpha_i/2+1)t^{\sum_i(\alpha_i/2+1)-1}I(1).$$ Let us now compute $$J:=\int_{x_i>0}x_1^{\alpha_1/2}\dots x_p^{\alpha_p/2}e^{-\sum_i x_i}dx_1\dots dx_p$$ in two ways. On one hand, it is (by definition of $\Gamma$) $$J=\prod_i\Gamma(\alpha_i/2+1).$$ On the other hand (setting $t=\sum_i x_i$) it is $$J=\int_0^\infty \frac{dI(t)}{dt} e^{-t}dt=\sum_i(\alpha_i/2+1)I(1)\int_0^\infty t^{\sum_i(\alpha_i/2+1)-1}e^{-t}dt$$ $$=\sum_i(\alpha_i/2+1)I(1)\Gamma(\sum_i(\alpha_i/2+1))=I(1)\Gamma(\sum_i(\alpha_i/2+1)+1).$$ We thus got (modulo mistakes on the way) $$I(t)=\frac{\prod_i\Gamma(\alpha_i/2+1)}{\Gamma(\sum_i(\alpha_i/2+1)+1)}t^{\sum_i(\alpha_i/2+1)}.$$