Conditional probability $P(A|B)$ is defined as: $P(A|B) = \frac{P(A\cap B )}{P(B)}$ when $P(B) > 0$, where $A$ and $B$ are events in the sample space.

Is $P(A|B)$ not defined for $P(B)=0$?

Consider the following problem.

The experiment is an infinite sequence of coin tosses. Suppose the coin lands $Heads$ at each toss with a probability $p \in (0,1)$.

Suppose $B = \{Heads, Heads, Heads,..., Heads,...\}$ i.e, coin lands $Heads$ after every toss,


$A$ is the event in which the 1st coin toss is Heads i.e., $A=\{Heads, H/T, H/T,...\}$.

We have, $P(A)=p$ and $P(B)=0$

I am interested in $P(A|B)$.

Consider the following attempt.

Suppose we consider the first $n$ coin toss.

Let $B_n = \{Heads, Heads,...,Heads\}$.

Let $A_n$ be the event in which the first coin toss is Heads.

We have: $ A_n \cap B_n = B_n$

$ \displaystyle P(A_n|B_n) = \frac{P(A_n \cap B_n)}{P(B_n)} $

$ \implies P(A_n|B_n) = \frac{P(B_n)}{P(B_n)} $

$ \implies P(A_n|B_n) =1. $

$ \implies \displaystyle \lim_{n\to\infty} P(A_n|B_n) = \lim_{n\to\infty} 1 $

$ \implies P(A|B) = 1 $

This result matches our intuition. But, $P(A|B)$ is not defined by the definition when $P(B)=0$.

Where am I getting it wrong?


The equality: $$P(A\mid B)P(B)=P(A\cap B)\tag1$$ characterizes conditional probability.

If $P(B)=0$ then also $P(A\cap B)=0$ and we get:$$P(A\mid B)0=0$$

This is true for every value given to $P(A\mid B)$ and gives us the liberty to fill in a value that agrees with our intuition.

I admit that then a rigorous definition of $P(A\mid B)$ lacks and care is needed to avoid accidents.

Nevertheless, personally I prefer $(1)$ and nothing more than $(1)$ above the "handicapped" expression $P(A\cap B)/P(B)$.

Another example (next to yours): let it be that $X$ has binomial distribution, that $Y$ has normal distribution and that $X$ and $Y$ are independent.

Then how about expressions like $P(X=4\mid Y=2)$?

Following my intuition I choose for $P(X=4\mid Y=2)=P(X=4)$ on base of the independence.

  • $\begingroup$ Well, in the case of $P(X=4|Y=2) $, its hybrid joint distributions (X: discrete r.v and Y: continuous r.v). $P(X=4|Y=2)$ is defined as $P(X=4|Y=2) = \frac{p_Y(2|X=4) P(X=4)}{p_{Y}(y)}$. But, I understand your point $\endgroup$ – Suhan Shetty Aug 19 '18 at 16:13
  • $\begingroup$ Any ways, $P(A|B)P(B) = P(A \cap B)$ still does not define $P(A|B)$ in the case $P(B)=0$ $\endgroup$ – Suhan Shetty Aug 19 '18 at 16:20

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.