We are set in $\mathbb{R}^3$. Let $f: \mathbb{R}^3 \rightarrow \mathbb{R}$ be a $C^1_0$ function, i.e. continuously differentiable with compact support. Let $\epsilon > 0$ be small. I need to show that the derivative $$ || \partial_x [\int_{B(x,\epsilon)} \frac{x-y}{|x-y|^3} f(y)dy] ||_{\infty}$$ of this integral function exists and can be bounded from above by a constant times $||f||_{\infty}$ for $\epsilon \rightarrow 0$. We can take any $\infty$ norm we want for the derivative, we're differentiating a function $\mathbb{R}^3 \rightarrow \mathbb{R}^3$ so the derivative is a matrix, but we can take for example supremum over its entries, so $$ \max_{i \in \{1,2,3\}} | \partial_{x_i} [\int_{B(x,\epsilon)} \frac{x_1 - y_1}{|x-y|^3} f(y)dy] |$$ for example (or something else if it's easier, but I guess it shouldn't differ much). I want to show that this quantity can be bounded from above by $C ||f||_{\infty}$ with $\epsilon \rightarrow 0$. I can't just push the derivative in and say that $\partial_x \frac{x-y}{|x-y|^3} \approx |x-y|^{-3}$ because that won't let me conclude - $|x-y|^{-3}$ is not integrable around $0$ in $\mathbb{R}^3$. I was thinking about pushing the derivative inside and computing explicitly using spherical coordiantes but there are two issues: first it seems really cumbersome and I'd rather avoid it and use simpler arguments if possible and second - I'd still have to argue why I can push the derivative inside. Any suggestions are welcome.

edit: when I think about it I imagine an actual difference quotient when I consider two sphere centered at $x$ and $x +h$ and I guess that I could take care of the overlapping part by 'pairing up' the points with factors $f(y)$ and $f(y+h)$, taking supremum of $\nabla f$ and integrating - that would go to $0$ with $\epsilon \rightarrow 0$ and I expect that the parts that I wouldn't get to pair up with anything will have measure suffiently small (so that the integrals are comparable with $|h|$) so that I can use the $||f||_{\infty}$ bound there but I'm having a hard time writing it down

Edit: my more precise although still incomplete try. I show how to bound the derivative assuming it exists:

ok, after all it seems to be not that difficult, which gets me a bit worried so it would be great if someone could point out a mistake I might've made. I guess the only tricky part is showing that the function is indeed differentiable - if we knew that it would suffice to show that $$ \frac{1}{|h|} | \int_{B(x,\epsilon)} \frac{x-y}{|x-y|^3} f(y) dy - \int_{B(x + h,\epsilon)} \frac{x+h-y}{|x+h-y|^3} f(y) dy| \leq C \epsilon$$ for some constant $C$. a change of variables $y = y-h$ in the second integral gives $$ \frac{1}{|h|} | \int_{B(x,\epsilon)} \frac{x-y}{|x-y|^3} f(y) dy - \int_{B(x,\epsilon)} \frac{x-y}{|x-y|^3} f(y+h) dy|$$ i.e. $$ | \int_{B(x,\epsilon)} \frac{x-y}{|x-y|^3} \frac{f(y) - f(y+h)}{|h|} dy | \leq \int_{B(x,\epsilon)} \frac{1}{|x-y|^2} ||\nabla f||_{\infty} dy \leq C ||\nabla f||_{\infty} \epsilon$$ because $\int_{B(x,\epsilon)} \frac{1}{|x-y|^2} = C \epsilon$, hence the whole thing goes to $0$ with $\epsilon$. Nevertheless we still need to know that it is differentiable

however, the paper I'm reading clearly states that we bound it by a constant times $||f||_{\infty}$ which confuses me since if I knew it was differentiable I could say it tends to $0$ with $\epsilon$ and if I don't know whether it's differentiable then I can't really talk about bounding the derivative

edit2: ok, I see where I went wrong - I got the wrong area of integration after my change of variables. I still believe though that if I knew the derivative exists I could replace the second (wrong) integral $$ \int_{B(x,\epsilon)} \frac{x-y}{|x-y|^3} f(y+h) dy| $$ with $$ \int_{B(x,\epsilon - 2|h|)} \frac{x-y}{|x-y|^3} f(y+h) dy|$$ plus some remainder which would then lead to a bound involving $||f||_{\infty}$ but we need differentiability to do that anyway.

  • $\begingroup$ I'm having trouble seeing why the integral exists in general. It seems it would be infinity for every $x$. What am I missing? $\endgroup$ – Alex Schiff Jun 18 '14 at 2:42
  • $\begingroup$ well, $\frac{1}{|x|^2}$ is integrable around $0$ in $\mathbb{R}^3$ so the function is well defined. it should also be differentiable and the way I see it is explained in the last paragraph, although it still isn't a formal argument $\endgroup$ – mm-aops Jun 18 '14 at 12:46
  • $\begingroup$ The only thing I can think of is differentiating under the integral. Not sure if this applies in this situation... $\endgroup$ – Alex Schiff Jun 18 '14 at 17:10
  • $\begingroup$ not really, the function under the integral is pretty singular at $0$ so one would need much general results $\endgroup$ – mm-aops Jun 18 '14 at 17:33
  • $\begingroup$ Sorry, clearly I need to refresh my memory on integration in $\mathbb{R}^n$ for $n\geq 2$. $\endgroup$ – Alex Schiff Jun 18 '14 at 17:35

I can't give you a complete solution, but maybe the following thoughts are helpful to you:

Using the change of variables $z = x-y$, you can rewrite your original function as follows:

$$ F(x) := \int_{B(x,\varepsilon)} \frac{x-y}{|x-y|^3} \cdot f(y) \, dy = \int_{B(0,\varepsilon)} \frac{z}{|z|^3} \cdot f(x-z) \,dz . $$

Using the fact that $x \mapsto f(x-z)$ is (continuously) differentiable and that $z \mapsto \frac{z}{|z|^3} \chi_{B(0,\varepsilon)}(z) \in L^1$, you can conclude (using standard theorems), that $F$ is indeed differentiable with derivative

$$ \partial_i F(x) = \int_{B(0,\varepsilon)} \frac{z}{|z|^3} (\partial_i f)(x-z) \,dz. $$

Using this, you can also derive $\Vert \partial_i F \Vert_\infty \lesssim \Vert \partial_i f\Vert_\infty$, which is strictly weaker than your target estimate.

But also note that the above estimate even implies

$$ \Vert \partial_i F(x) \Vert_\infty \leq \int_{B(0,\varepsilon)} \frac{1}{|z|^2} \, dz \cdot \Vert \partial_i f\Vert_\infty \xrightarrow[\varepsilon \rightarrow 0]{} 0. $$ Thus, if you are only interested in the case $\varepsilon \rightarrow 0$, this should yield what you want :)

One of the things you could try now (although I could not make it work, but I did not try too hard) is to write

$$ \partial_i F(x) = \lim_{\delta \downarrow 0} \int_{B(0,\varepsilon) \setminus B(0,\delta)} \frac{z}{|z|^3} (\partial_i f)(x-z) \, dz $$

and then use partial integration (i.e. the divergence theorem) to pass on the derivative on $f$ to $\frac{z}{|z|^3}$. Note here, that

  1. There will be boundary terms.
  2. You can not do the same without the limit, because $\frac{z}{|z|^3}$ is not smooth in the origin and (even worse) the derivative is not integrable in a neighborhood of the origin.

Finally, it could be helpful to know where you found the statement. The context could perhaps indicate the method to use.


We have that

$$ \begin{align} \partial_x\int_{B(x,\epsilon)} \frac{x-y}{|x-y|^3}f(y)dy & \approx \partial_x\frac{1}{|B(x,\epsilon)|}\int_{B(x,\epsilon)}(x-y)f(y) \\ &\leq \lim_{\epsilon->0}\frac{1}{\epsilon}\frac{1}{|B(x,\epsilon)|}\int_{B(x,\epsilon)}|\epsilon f(y)|dy\\ &=lim _{\epsilon->0}\frac{1}{|B(x,\epsilon)|}\int_{B(x,\epsilon)}| f(y)|dy \end{align} $$

The first equality $\approx$ holds because: $|B(x,\epsilon)|$ is the volume of the ball centered at $x$ with radius $\epsilon$.
$|B(x,\epsilon)|=4/3\pi \epsilon^3\approx |x-y|^3$ since $y\in B(x,\epsilon)$ definition of derivatives can be found here: http://en.wikipedia.org/wiki/Lebesgue_differentiation_theorem

The ineuqlity $\leq$ holds because by definition of derivative, the derivative with respect to $x$ is defined to be $\partial_x g(x)=lim_{\epsilon->0}g(x)$ for $g:\mathbb{R^3}->\mathbb{R}$, since $\mathbb{R}$ is one dimension. And also $|x-y|\approx \epsilon$ hence two $\epsilon $ are cancelled with each other.

By strong version of Hardy-Littlewood theorem, http://en.wikipedia.org/wiki/Hardy%E2%80%93Littlewood_maximal_function, The result holds.

Is this time better?

  • $\begingroup$ I allowed myself to improve the formatting of your post. One problem that I have with your approach is that even if $\int_{B(x,\varepsilon)} \frac{x-y}{|x-y|^3} f(y) \, dy \approx \frac{1}{|B(x,\varepsilon)|} \int_{B(x,\varepsilon)} (x-y)d(y) \, dy$ were to hold, this would not per se imply that the same holds after taking derivatives. $\endgroup$ – PhoemueX Jun 19 '14 at 7:38
  • $\begingroup$ Thank you for editing. $lim_{\epsilon \rightarrow \infty}1/\epsilon...$ is the definition of derivative $\endgroup$ – claire Jun 19 '14 at 7:43
  • 1
    $\begingroup$ Yes, but for example $1 \approx 1 + \frac{1}{1000} \cdot \cos(\frac{1}{x})$, but the derivative on the left hand side vanishes, whereas the derivative on the right hand side gets quite large for $x \rightarrow 0$. $\endgroup$ – PhoemueX Jun 19 '14 at 8:30
  • 1
    $\begingroup$ I am talking about the FIRST "equality", i.e. the $\approx$, not(!) about the estimate in the second line. $\endgroup$ – PhoemueX Jun 19 '14 at 8:36
  • 1
    $\begingroup$ I don't really get this answer. First of all it seems to me you also need to assume that the derivative exists to proceed with taking the limit but I also agree with @PhoemueX as to his counterexample and doubts. Could you elaborate more on it? $\endgroup$ – mm-aops Jun 19 '14 at 22:17

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.