I was reading stuffs about Riesz energy which is defined for an open subset $U\in\mathbb{R}^d$ by $I_s(U)=\int_U\int_U|x-y|^{-s}\ dx\ dy$ where $dx$ and $dy$ are Lebesgue measure in $U$. Now if I take for simplicity $U$ to be a ball I am not sure how to evaluate the integration. If I shift the integral to origin i.e like calculating the integral $|x|^{-s}$ over a ball is fine using polar change of coordinate. But here it's not only in one ball over another ball also, I am not sure how to evaluate it or I am missing something. Any help is very much appreciated for my understanding.
-
2$\begingroup$ I doubt very much that there is a neat formula in terms of $\pi$ and $e$ of the coefficient in front of the appropriate power of the ball radius, but a natural question to ask is "why do you want to evaluate it at all? Aren't all positive constants essentially the same?". $\endgroup$– fedjaCommented Aug 3, 2023 at 2:18
-
$\begingroup$ I am not asking about neat formula. I am saying even if the constant is C how to evaluate the integral as it's not essentially $|x|^{-s}$. I mean say $U=B(x_0,r)$ then what is the power of $r$ it's coming when I am doing the integration. It's not over one ball it's over same ball which I am bit confused. $\endgroup$– SarthakCommented Aug 3, 2023 at 9:53
-
$\begingroup$ @fedja: Does $\Gamma(\ldots)$ count as a neat formula in terms of $\pi$ and $e$? :-) $\endgroup$– Mateusz KwaśnickiCommented Aug 3, 2023 at 21:44
-
$\begingroup$ @MateuszKwaśnicki Only at integers and half-integers. Otherwise it is a 1-D integral that is useful and reasonably fast convergent at $\infty$, but the singularity at $0$ is still on the way of efficient numeric evaluation of it. But I'll grant you that for odd $d$ you have it since only the ratio of two values with integer difference matters there in your final formula. I didn't expect that much, so you've got me here. :-) $\endgroup$– fedjaCommented Aug 3, 2023 at 22:26
2 Answers
This is just an extended comment to Iosif Pinelis's answer above, which provides an answer in terms of an unknown constant $C_{d,s}$. Here we evaluate this constant.
Let $B$ be the unit ball. If $f(x)=(1-|x|^2)^p$ if $x\in B$ and $f(x)=0$ otherwise, then $$(-\Delta)^{\alpha/2}f(x)=2^\alpha\frac{\Gamma(p+1)\Gamma(\frac{d+\alpha}2)}{\Gamma(p+1-\frac\alpha2)\Gamma(\frac d2)}{_2F_1}(\tfrac{d+\alpha}2,\tfrac\alpha2-p;\tfrac d2;|x|^2).$$ Here $p > -1$, ${_2F_1}$ is the Gauss's hypergeometric function, and $(-\Delta)^{\alpha/2}$ is the fractional Laplace operator for $\alpha > 0$ and the Riesz potential operator for $\alpha \in (-d, 0)$.
I would not be surprised if the above formula actually appeared in some paper by Boris Rubin, Stefan Samko, or one of the other authors working on Riesz potentials, but as far as I can tell, for $\alpha \in (0, 2)$ this is due to Bartłomiej Dyda DOI:10.2478/s13540-012-0038-8, and extension to an arbitrary $\alpha > -d$ is given in my paper with Bartłomiej Dyda and Alexey Kuznetsov DOI:10.1007/s00365-016-9336-4; for further discussion, see my survey DOI:10.1515/9783110571622-007.
We set $p=0$ and $\alpha=s-d$ to get $$(-\Delta)^{-(d-s)/2}\mathbb 1_B(x)=2^{s-d}\frac{\Gamma(\frac s2)}{\Gamma(1+\frac{d-s}2)\Gamma(\frac d2)}{_2F_1}(\tfrac s2,\tfrac{s-d}2;\tfrac d2;|x|^2).$$ Using the definition of the Riesz potential operator, this translates to $$\int_B|x-y|^{-s}dy=\frac{2\pi^{d/2}}{(d-s)\Gamma(\frac d2)}{_2F_1}(\tfrac s2,\tfrac{s-d}2;\tfrac d2;|x|^2).$$ Integrating this with respect to $x\in B$, we obtain $$\int_B\int_B|x-y|^{-s}dydx=\frac{4\pi^d}{(d-s)(\Gamma(\frac d2))^2}\int_0^1r^{d-1}{_2F_1}(\tfrac s2,\tfrac{s-d}2;\tfrac d2;r^2).$$ Using [http://functions.wolfram.com/07.23.21.0003.01], we find that $$\int_B\int_B|x-y|^{-s}dydx=\frac{4\pi^d{_2F_1}(\tfrac s2,\tfrac{s-d}2;1+\tfrac d2;1)}{d(d-s)(\Gamma(\frac d2))^2},$$ and [http://functions.wolfram.com/07.23.03.0002.01] leads to $$\int_B\int_B|x-y|^{-s}dydx=\frac{4\pi^d\Gamma(1+\tfrac d2)\Gamma(1+d-s)}{d(d-s)\Gamma(1+\frac{d-s}2)\Gamma(1+d-\frac s2)(\Gamma(\frac d2))^2}.$$ After some simplification, we arrive at $$\int_B\int_B|x-y|^{-s}dydx=\frac{2^{d-s+1}\pi^{d-1/2}\Gamma(\frac{1+d-s}2)}{(d-s)(d-\frac s2)\Gamma(\tfrac d2)\Gamma(d-\frac s2)}.$$ (Watch out: this may be full of typos! I only checked the final answer for $d = 1$, and it seems to agree with the result of an explicit calculation.)
-
1$\begingroup$ Wow! I only had an expression for the constant $C_{d,s}$ in the form of an integral over a region in $\mathbb R^3$. $\endgroup$ Commented Aug 3, 2023 at 21:58
Using the substitutions $x=x_0+ru$ and $y=y_0+rv$, for real $r>0$ we get $dx=r^d\,du$ and $dy=r^d\,dv$, and hence $$I_s(B(x_0,r))=\int_{|x-x_0|<r}dx\,\int_{|y-x_0|<r}dy\,|x-y|^{-s} =C_{d,s}r^{2d-s},$$ where $$C_{d,s}:=\int_{|u|<1}du\,\int_{|v|<1}dv\,|u-v|^{-s}>0.$$ Note that $C_{d,s}<\infty$ iff $s<d$.
-
$\begingroup$ If the exact value of $C_{d,s}$ is relevant, one can apply formulae which essentially go to M. Riesz's 1938 seminal paper — see, for example, Table 2 in my survey Fractional Laplace Operator and its Properties DOI:10.1007/s00365-016-9336-4 and the references therein. If you are interested, I can work out the details. $\endgroup$ Commented Aug 3, 2023 at 12:43
-
$\begingroup$ @MateuszKwaśnicki : Thank you for your comment. Yes, this would of course be interesting. $\endgroup$ Commented Aug 3, 2023 at 12:46
-
$\begingroup$ OK, this was way too long for a comment, so I posted this as a separate answer. $\endgroup$ Commented Aug 3, 2023 at 21:42