MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let $L^+$ be a set of all real valued functions defined on a real line which are Lebesgue integrable on each $[0,c]$, where $c>0$, and are zero for $x<0$. Let for $a>0$, $f_a(t)=(t-a)^{-3/4}$ for $t>a$ and $0$ for $t \leq a$. Then, for $a,b>0$, convolution $(f_a*f_b)(x):=\int_0^x f_a(x-y)f_b(y)dy=\beta (x-a-b)^{-\frac{1}{2}}$ for $x> a+b$ and $0$ for $x \leq a+b$, where $\beta=B(\frac{1}{4}, \frac{1}{4})$. It shows that convolution of two functions from $L^+$ need not be continuous. Is it possible, maybe by condensation of singularities and above example to show existence of two functions from $L^+$ which convolution is discontinuous everywhere
on $[0, \infty)$?

share|cite|improve this question
You say $\int_0^x f_a(x-y)f_b(y)dy=f_{a+b}(x)$. My calculations disagree. But the convolution $f_a * f_b$ is, indeed, unbounded at $a+b$. – Gerald Edgar Mar 31 '12 at 17:12
Sorry, I did mistake. I just have edited. – arc Mar 31 '12 at 18:21

How about this. If $g \in L^+$, then almost every point is a Lebesgue point for $g$. And if $h \in L^+$, then almost every point is a Lebesgue point for $h$. We should then be able to show that the convolution $g * h$ is differentiable almost everywhere. Or something like that.

share|cite|improve this answer
Certainly $f*g$ is locally $L^1$, so almost all points are also Lebesgue points for $f*g$. For every such point $x$ and every $\varepsilon>0$ there exists a $\delta>0$ with almost all points $y\in B(x,\delta)$ satisfying $|f*g(x)-f*g(y)|<\varepsilon$. (Why? Just assume not...) So there exists a version of $f*g$ continuous at $x$. With a little thought you can now devise a scheme to modify $f*g$ so it really is continuous except at countably many points. – Ollie Margetts Apr 1 '12 at 0:59

You can't have a convolution of two functions in $L_1(0,c)$ that is not in $L_1(0,c)$, else you violate Young's inequality, or the fact that $L_1(0,c)$ (i.e. the space you refer to as $L^{+}$) is a Banach algebra under convolution, so the answer is no. I assume you are taking $a < c$ and $b < c$ (otherwise you can't work in $L_1(\mathbb{R})$ because $f_a$ and $f_b$ don't belong there).

EDIT: Excuse me, Arc. I do apologize. I did not make the relevant distinction between an everywhere discontinuous function and a function defined only on a set of measure zero. I now understand the definition, so my answer is largely unhelpful.

However, if you are trying to construct a function that is discontinuous on a dense set, I do get the feeling that your construction wont do because the output of your convolution is (as you would expect) better behaved than the inputs, it is locally in $L_{2-\epsilon}$ and still has only one point of discontinuity. Perhaps I am missing the point here, though. Can you explain a bit more about what you want to achieve and your approach?

share|cite|improve this answer
An $L^1$ function can be discontinuous everywhere. – Gerald Edgar Mar 31 '12 at 23:36

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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