1
$\begingroup$

$\alpha, \alpha', \beta$ and $\beta'$ are four independent standard normal random variables, I am wondering how to compute the probability of the following two events:

  • $\alpha>\alpha'>0, \ \ \beta<\beta', \ \ |\alpha-\alpha'+\beta-\beta'|\geq c_1,$
  • $\alpha>0>\alpha', \ \ \beta<\beta', \ \ |\alpha-\alpha'+\beta-\beta'|\geq c_1, \ \ |\frac{\alpha}{\alpha'}|>c_2$

where $c_1$ and $c_2$ are positive constants. I know that the sum and difference of normal random variables are still normal random variables, but I'm not sure how to use this with the inequality relation here.

$\endgroup$
5
  • $\begingroup$ These can be calculated with four-dimensional integrals, but they have no simple formulas. $\endgroup$
    – Matt F.
    Sep 3, 2020 at 16:37
  • $\begingroup$ Even the probabilities for simpler events, like $\alpha>0, \ \alpha+\beta>c_1$ or $\alpha-\alpha'>c_1,\ \alpha/\alpha'>c_2$ do not have simple formulas in terms of the usual functions for normal variables. $\endgroup$
    – Matt F.
    Sep 3, 2020 at 17:25
  • $\begingroup$ @MattF. Thank you for your answer! Is it possible to simplify the formula so that we can have a lower bound of the probability? $\endgroup$
    – luw
    Sep 3, 2020 at 22:56
  • $\begingroup$ what are you doing with these probabilities? There is trivially "a lower bound", so the question has to be whether there's a useful lower bound, and that will depend on your use for the probabilities. $\endgroup$
    – Matt F.
    Sep 4, 2020 at 0:55
  • $\begingroup$ @MattF. I am analyzing some machine learning algorithms and I want to say that parameters initialized as standard normal variables have the properties above with certain probability. So I guess any lower bound would be good for me. $\endgroup$
    – luw
    Sep 4, 2020 at 3:52

1 Answer 1

-1
$\begingroup$

This might work, the pdf might have the answer or at least part of the question but could not find the rest. https://www.dartmouth.edu/~chance/teaching_aids/books_articles/probability_book/Chapter4.pdf

$\endgroup$
0

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.