Let $(\Omega, \mathcal{A}, P)$ be a probability space. The following result is known.

If $X:\Omega\rightarrow\mathbb{R}$ is a random variable, then the CDF $F_X$ is discontinuous at $a\in \mathbb{R}$ iff $a$ is an atom of $X$.

Does this result extend to the $\mathbb{R}^2$, i.e. is the following statement correct?

If $X,Y:\Omega\rightarrow\mathbb{R}$ are random variables, then the joint CDF $F_{X, Y}$ is discontinuous at $(a,b) \in \mathbb{R}^2$ iff $(a,b)$ is an atom of $(X,Y)$.


No, the second highlighted statement is not true. It is possible to find a two-dimensional random variable $(X,Y)$ with no atoms, whose CDF is discontinuous. An example can be given based on item 5.6 (The continuity property of one-dimensional distributions may fail in the multi-dimensional case) on p. 34 of Jordan M. Stoyanov's Counterexamples in Probability, 3rd Edition, Dover, 2013, ISBN-13: 978-0486499987.

Consider the following function $F:\mathbb{R}^2\rightarrow\mathbb{R}$: A discontinuous two dimentional CDF with no atoms whose graph can be visualized using the following diagram: An aid to the visualization of the CDF's graph

It can be verified that:

  • This is a 2-dimensional CDF. (This can be verified by checking the 3 characterizing properties of a multidimensional CDF.)
  • This CDF has no atoms. (This can be verified by checking that for every $x_0, y_0 \in \mathbb{R}^2$ and for every $\epsilon - \epsilon(x_0,y_0) \in (0,\infty)$, there exists some $\delta \in (0,\infty)$ such that whenever $a < x_0 < b$, and $c < y_0 <d$ are such that $b-a, d-c < \delta$, $F(a,c)+F(b,d)-F(a,d)-F(b,c) < \epsilon$.)
  • However, every point with coordinates $(1,y)$, where $y \in (\frac{1}{2},\infty)$, is a discontinuity point of $F$. (This can be verified by checking that, for such a $y$, $\lim_{\substack{x\rightarrow1\\x<1}}F(x,y) = \frac{1}{2} \neq y = \lim_{\substack{x\rightarrow1\\x>1}}F(x,y)$.)

* Note that the example given above is very similar, but no identical, to Stoyanov's example, since Stoyanov's example is wrong, as pointed out by ben in a comment below, since, for instance, at $(1,\frac{3}{4})$ Stoyanov's $F$ is not right-continuous in the first parameter.

  • $\begingroup$ For the example given in book, would you mind explaining how $F(x,y)$ is right continuous in $x$ and $y$? It is presumably supposed to be right continuous in $x$ and $y$, as otherwise it wouldn't count as a cdf, right? My issue is that I've calculated $\lim_{h\rightarrow 0^{+}}F(1+h,3/4)=3/4\neq F(1,3/4)=1/2$, which, if my understanding on the notion of "right continuous in $x$" is correct, would suggest that $F(x,y)$ is not a cdf. $\endgroup$
    – ben
    May 10 '21 at 8:21
  • $\begingroup$ @ben: You are right. Thanks for pointing this out. This is an error in Stoyanov's book. I have corrected my answer. $\endgroup$
    – Evan Aad
    May 16 '21 at 5:32

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.