Let us denote the Riesz potential in $\mathbb R^d$ by $$ I_\alpha (f)(x) := c_{d, \alpha} \int_{\mathbb R^d} \frac{f(y)}{|x-y|^{d-\alpha}} \, dy.$$

By the classical Hardy-Littlewood-Sobolev theorem on fractional integration we have for $1 < p <d/\alpha$ that $$ \|I_\alpha(f)\|_{L^q} \le C_{d, \alpha, p} \|f\|_{L^p}, \quad\text{where}\ \ \tag 1 q=\frac{dp}{d-p\alpha}. $$

I am looking for a reference (with a proof) for the borderline case $p=d/\alpha$: $$I_\alpha:L^{d/\alpha} \to \rm{BMO}. \tag 2$$

I have checked Grafakos's and Stein's harmonic analysis books, but I don't think the proof is in any of them.

A related result, which would probably be enough for me, is stated as an exercise 8.11. on page 62-63 of


but at least a quick look seems to indicate that the constant in (1) is of the form $C_{d,\alpha} q$ (for $p$ close to $d/\alpha$), so that the corresponding power series in the exercise does not converge after raising the estimate to power $q$ (the factor k^k dominates k! of the denominator). Probably I am missing something.

In any case, the best would be to find a self-contained proof of (2) directly.

  • 1
    $\begingroup$ Hint: K^k is comparable to e^k k! by Stirling. $\endgroup$
    – Fan Zheng
    Apr 27, 2015 at 15:04
  • $\begingroup$ Oh yes, this must be the piece I was missing. Thanks! :) $\endgroup$ Apr 27, 2015 at 16:51

1 Answer 1


Let $a \in \mathbb{R}^d$ and $r > 0$. We have $$ \frac{1}{\vert B_r \vert^2} \int_{B_r} \int_{B_r} \vert I_\alpha (f) (x) - I_\alpha (f) (y) \vert\,\mathrm{d}x\,\mathrm{d}y \le \frac{c_{d, \alpha}}{\vert B_r \vert^2} \int_{\mathbb{R^d}} \int_{B_r} \int_{B_r} \vert f (z)\vert \,\Big\vert \frac{1}{\vert z - x\vert^{d - \alpha}} - \frac{1}{\vert z - y\vert^{d - \alpha}} \Big\vert \,\mathrm{d}x\,\mathrm{d}y\,\mathrm{d}z. $$ We compute $$ \int_{B_r} \int_{B_r}\Big\vert \frac{1}{\vert z - x\vert^{d - \alpha}} - \frac{1}{\vert z - y\vert^{d - \alpha}} \Big\vert \,\mathrm{d}x\,\mathrm{d}y \le \frac{C r^{2 d + 1}}{(r + \vert z \vert)^{d - \alpha + 1}}, $$ and we conclude by the classical Hölder inequality that $$ \int_{B_r} \int_{B_r} \vert I_\alpha (f) (x) - I_\alpha (f) (y) \vert\,\mathrm{d}x\,\mathrm{d}y \le C r^{2 d + 1} \Bigl(\int_{\mathbb{R}^d} \Bigl(\frac{1}{(r + \vert z \vert)^{d - \alpha + 1}}\Bigr)^\frac{d}{d - \alpha} \, \mathrm{d} z\Bigr)^{1 - \frac{\alpha}{d}} \Vert f \Vert_{L^{d/\alpha}}, $$ Since $$ \int_{\mathbb{R}^d} \Bigl(\frac{1}{(r + \vert z \vert)^{d - \alpha + 1}}\Bigr)^\frac{d}{d - \alpha} \, \mathrm{d} z = \frac{1}{r^{d + \frac{d}{d -\alpha}}}\int_{\mathbb{R}^d} \Bigl(\frac{1}{(1 + \vert z \vert/r)^{d - \alpha + 1}}\Bigr)^\frac{d}{d - \alpha} \, \mathrm{d} z = \frac{C}{r^\frac{d}{d - \alpha}}, $$ we conclude that $$\frac{1}{\vert B_r \vert^2} \int_{B_r} \int_{B_r} \vert I_\alpha (f) (x) - I_\alpha (f) (y) \vert\,\mathrm{d}x\,\mathrm{d}y\le C'' \Vert f \Vert_{L^{d/\alpha}} $$

  • $\begingroup$ For some reason I obtain that the constant $C'$ depends on the radius $r$ (I checked the case $\alpha=1$). Are you certain about your second estimate? Maybe I did some mistake somewhere. Looks good in any case. $\endgroup$ Apr 27, 2015 at 14:33
  • $\begingroup$ I have added some details of the computation of the $L^{d/(d - \alpha)}$ norm. I hope that it helps. $\endgroup$ Apr 27, 2015 at 15:25
  • $\begingroup$ Thanks, this is also what I obtained. But in order to show that $I_\alpha(f)$ is in $\rm{BMO}$, don't we want to take supremum over all balls on the left hand side. Now the bound $Cr^{-\frac d{d-\alpha}}$ blows up when $r \to 0$, so how do we know that the supremum stays bounded? $\endgroup$ Apr 27, 2015 at 16:48
  • $\begingroup$ We are taking an $L^{d/(d - \alpha)}$ norm, which blows up like $1/r$ and we divide by $r^{2 d}$ (volume of the ball squared), this compensates exactly the $r^{2 d + 1}$ coming from the second estimate. $\endgroup$ Apr 28, 2015 at 6:24
  • $\begingroup$ Of course, thanks. I forgot the power outside the integral. :) $\endgroup$ Apr 28, 2015 at 10:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.