Earlier today in my stochastic processes lecture, the prof mentioned that there does not exist a probability measure P(A) defined for all subsets of [0,1] which would satisfy all the 3 Kolmogorov axioms. If I understand correctly, the Kolmogorov axioms are supposed to lay out the basis of any given probability measure so I find the statement a bit unintuitive and can’t fully understand how a probability measure wouldn’t satisfy the axioms it’s built upon. Am I missing something critical? I’d appreciate any tips or further readings.

  • $\begingroup$ The assertion is false. Pick a number $\xi\in[0,1].$ The function $\mathbb{P}:\mathscr{P}([0,1])\to[0,1]$ defined by $\mathscr{P}(\mathcal E) = \mathcal{I}_{\mathcal E}(\xi)$ (trivially) satisfies all three axioms. ($\mathscr{P}$ is the power set and $\mathcal I$ is the indicator function.) $\mathbb{P}$ is called the atom at $\xi.$ It is the probability measure determined by any random variable $X:\Omega\to\mathbb R$ (for any probability space $\Omega$) that is almost surely equal to $\xi.$ $\endgroup$
    – whuber
    Sep 14, 2022 at 16:34

1 Answer 1


the prof mentioned that there does not exist a probability measure P(A) defined for all subsets of [0,1] which would satisfy all the 3 Kolmogorov axioms

At the very outset, please note whuber's comment. The statement is then overreaching (perhaps misinterpreted by OP) and problematic.

Common counter-example: Dirac measure $\delta_\omega$ for a point $\omega \in [0, 1].$

With that stated, define for $a,b\in[0,1],$ $$\mathbf P([a, b])= \mathbf P((a, b])=\mathbf P([a, b)) = \mathbf P((a, b)) = b-a .\tag 1$$

A related question can be addressed: can $\mathbf P(A) $ be plausible? Is it possible for each $A\subseteq [0, 1]? $

The concern stems from the fact that there exists a non-Lebesgue measurable set in $[0,1].$

A brief sketch:

We will use an equivalence relation (check): $$x\sim y\iff y-x~\text{rational}.\tag 2$$

As a result of the equivalence, it partitions $[0, 1].$ Now, the important aspect: invoke Axiom of Choice to create a subset $T$ consisting of exactly one element from each equivalence class. Assume $0\notin T. $

Define $r$-shift of $A\subseteq[0, 1]$ as

$$A\oplus r := \{a+r; ~a\in A, ~a+r\leq 1\}\cup\{a+r-1;~a\in A, ~a+r> 1\}\tag 3.$$ This is the $r$-translate of $A$ modulo $1.$

Then (check) $$\mathbf P(A\oplus r) = \mathbf P(A).\tag 4 $$

Now consider the union of rational shifts in $[0, 1)$ of $T$ i.e. $$S:= \bigcup_{r\in[0, 1), ~r~\text{rational}}(T\oplus r) .\tag 5$$ It is easy to check $(0, 1]$ is contained in $S$ and that all $T \oplus r$ are disjoint.

Countable additivity then allows

$$\mathbf P((0, 1]) =\sum_{r\in[0, 1), ~r~\text{rational}}\mathbf P(T\oplus r) .\tag 6$$

Therefore, by dint of $(4) ,$

$$\underbrace{\mathbf P((0, 1])}_{=1}=\sum_{r\in[0, 1), ~r~\text{rational}}\mathbf P(T) .\tag{ 6.I}$$

What is $\rm (6.I) $ saying? Countably infinite sum of a same quantity is equal to 1. Can it be possible? If $\mathbf P(T) = 0,$ then $0=1$ and if $\mathbf P(T) > 0,$ then $1=\infty.$


A First Look at Rigorous Probability Theory, Jeffrey S. Rosenthal, World Scientific, $2019.$

  • 1
    $\begingroup$ This doesn't appear to address the question that was stated, because it only analyzes a particular measure (Borel measure). The assertion in the question is that there is no measure, Borel or not, defined on the power set $\mathscr{P}([0,1])$ that satisfies all three Kolmogorov axioms. That's problematic because the assertion is wrong. $\endgroup$
    – whuber
    Sep 14, 2022 at 15:43
  • 1
    $\begingroup$ You are correct @whuber. I missed that no measure. Let me leave a message in my post. Thanks for noticing. $\endgroup$ Sep 14, 2022 at 15:48
  • 2
    $\begingroup$ Because you have made a significant contribution, I would suggest adding an introduction that discusses two possible interpretations of the question: as asked and as you think it was intended. You can dismiss the question as asked with a simple counterexample and then move on to what you already wrote. $\endgroup$
    – whuber
    Sep 14, 2022 at 16:09
  • 1
    $\begingroup$ Good suggestion. Done @whuber. $\endgroup$ Sep 14, 2022 at 17:20

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.