Let $d\in\mathbb N$ and $0<\alpha<d$. Define the Riesz kernel $K_\alpha(x):=|x|^{\alpha-d}$, and the associated convolution operator $$K_\alpha f(x):=\int\frac{f(y)}{|x-y|^{d-\alpha}}~dy.$$ The classical Hardy-Littlewood-Sobolev inequality states that we have the following boundedness property: for every $q>1$, $$\|K_\alpha f\|_q\leq C\|f\|_p$$ for some universal constant $C>0$ and $1/p=1/q+\alpha/d$.

Now let $0<\alpha_1,\ldots,\alpha_d<1$, and suppose instead that we consider the convolution operator given by the kernel $$K_{(\alpha_1,\ldots,\alpha_d)}(x)=\prod_{i=1}^d|x_i|^{\alpha_i-1},\qquad x=(x_1,\ldots,x_d),$$ which is what I call a "componentwise product of Riesz Kernels."

Question. Given $q>1$, does there exist some constant $C>0$ and a $p>1$ such that $$\|K_{(\alpha_1,\ldots,\alpha_d)} f\|_q\leq C\|f\|_p?\tag{1}$$

By a simple scaling argument, we can see that if (1) holds, then it must be for $1/p=1/q+\sum_i\alpha_i/d$. Indeed, letting $f_c(x):=f(cx)$, then $\|f_c\|_p=c^{-d/p}\|f\|_p$ and $\|K_{(\alpha_1,\ldots,\alpha_d)}f_c\|_q=c^{-\sum_i\alpha_i-d/q}\|K_{(\alpha_1,\ldots,\alpha_d)}f\|_q$.

  • 1
    $\begingroup$ I would write $K_{(\alpha_1,\ldots,\alpha_d)}$ as a composition of operators $K_j$, $j = 1, \ldots, d$, defined by $K_j f(x) = \int_{-\infty}^\infty f(x - y e_j) |y|^{\alpha_j-1} dy$, where $e_j = (0, \ldots, 0, 1, 0, \ldots, 0)$ has $1$ at $j$-th coordinate. Then it seems sufficient to apply the usual H-L-S inequality to each $K_j$. Have you tried that? $\endgroup$ – Mateusz Kwaśnicki Apr 3 '20 at 21:11
  • $\begingroup$ @MateuszKwaśnicki Thanks for pointing out this very simple reduction; I did not think of that. If each $K_j$ does indeed satisfies a HLS inequality of the form $\|K_jf\|_q\leq C\|f\|_p$ with $1/p=1/q+\alpha_j/d$, then we have the result! $\endgroup$ – user78370 Apr 4 '20 at 4:26

I realise I was too optimistic in my comment: no such ineqaulity holds in general. Indeed, consider $f(x) = \prod_j f_j(x_j)$. Then $$ K_{(\alpha_1,\ldots,\alpha_d)} f(x) = \prod_j K_{\alpha_j} f_j(x_j) $$ and so $$ \|f\|_p = \prod_j \|f_j\|_p , \qquad \|K_{(\alpha_1,\ldots,\alpha_d)} f\|_q = \prod_j \|K_{\alpha_j} f_j\|_q $$ So by the usual Hardy–Littlewood—Sobolev inequality (the "only if" part), in order to get the bound we are interested in, we would need $$ \frac{1}{p} = \frac{1}{q} + \frac{\alpha_j}{1} , $$ that is, $$ \alpha_1 = \alpha_2 = \ldots = \alpha_n \qquad \text{and} \qquad \frac{1}{p} = \frac{1}{q} + \frac{\alpha}{d},$$ with $\alpha = \alpha_1 + \alpha_2 + \ldots + \alpha_d$. In this case, the "if part" of the Hardy–Littlewood—Sobolev inequality implies that indeed the desired inequality holds.

  • $\begingroup$ Thanks for pointing out yet another simple observation. The result then only holds when all $\alpha_i$ are equal... $\endgroup$ – user78370 Apr 4 '20 at 9:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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