1
$\begingroup$

This is the popular Bonferroni inequality. Does it also hold for characteristic functions of random variables, as in when $P(A_i)$ is replaced by the characteristic function $\chi(A_i)$ and so forth? enter image description here

Given random variables $A_1,A_2...A_n$ (not events like in Theorem 1.2.3), is this satisfied? $$ \sum_{i=1}^n||\chi_{A_i}(t)|| - \sum_{i <j}^n||\chi_{A_i|A_j}(t).\chi_{A_j}(t)|| \le ||\chi_{\sum_{i=1}^n A_i}(t)||$$

|cite|improve this question
$\endgroup$
  • $\begingroup$ What is the definition of your characteristic function? Apparently (?), it is something different than en.wikipedia.org/wiki/… , which everyone in probability knows and loves, because that characteristic function is associated with a random variable, not with an event $A_i$ as you seem to have. Or perhaps you are rather confused. In any event, what is your "proposed" inequality? $\endgroup$ – Mark L. Stone Dec 1 '17 at 1:34
  • $\begingroup$ Given random variables $A_1,A_2...A_n$ (not events like in 1.2.3), and $$ \sum_{i=1}^n||\chi(A_i)|| - \sum_{i <j}^n||\chi(A_i|A_j).\chi(A_j)|| \le ||\chi(\sum_{i=1}^n A_i)||$$ is satisfied? So..the second term has conditional characteristic fn. and $A_1..A_n$ are not necessarily independent. If you want to first check for the case where they are independent as a special case, you are welcome to as well. $\endgroup$ – hearse Dec 1 '17 at 2:11
  • 1
    $\begingroup$ I still have no idea what $\chi(A_i)|$ or $||\chi(A_i)||$ or $\sum_{i=1}^n A_i$ mean. I don't know whether anyone else knows what they mean. $\endgroup$ – Mark L. Stone Dec 1 '17 at 2:14
  • $\begingroup$ The norm of a characteristic function of r.v's is always $\le$ 1. Check in say. en.wikipedia.org/wiki/… bullet no. 4 in properies. It is well-defined. $\endgroup$ – hearse Dec 1 '17 at 2:15
  • 1
    $\begingroup$ Updated........ $\endgroup$ – hearse Dec 6 '17 at 22:29
1
+50
$\begingroup$

No, the stated Bonferroni style inequality does not hold for characteristic functions. Certainly not in the way you've formulated it, and I can't think of any other way of interpreting the Bonferroni inequality in the context of characteristic functions.

Example: $n=2$, take random variables $X=1_A$ and $Y=1_B$ where $A$ and $B$ are independent events. Then $\chi_X(t)=e^{it}P(A)+P(A^c)$, so we can clearly choose $t$ such that $x:=|\chi_X (t)|<1$. Define also $y:=|\chi_Y(t)|$. In this situation, the stated inequality is equivalent to $x+y-xy\leq xy$ i.e. $x+y\leq 2xy$. Combining this with the AM-GM inequality leads to $\sqrt{xy}\leq xy$, which is clearly false when $xy<1$.

|cite|improve this answer
$\endgroup$

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.