1
$\begingroup$

Problem Let $\alpha_1,...,\alpha_n,\beta_1,...,\beta_n \in \mathbb{R}$, $\sum \alpha_i^2=\alpha$,$\sum \beta_i^2=\beta$, and $\sum \alpha_i\beta_i=\gamma$. If $\Delta=\alpha\beta-\gamma^2>0$, find the density of the joint distribution of $\alpha_1 X_1+...+\alpha_{n}X_n$ and $\beta_1 X_1+...+\beta_{n}X_n$ where $X_i$ are independent identically distributed Bernoulli $\pm 1$ random variables which take value $1$ with probability $0.5$ and value $-1$ with probability $0.5$.

Background The case where the $X_i$ are $N(0,1)$ appears in Lemma 4 of the paper On The Average Number of Real Roots of a Random Algebraic Equation.

Approach Let $\chi:=\alpha_1 X_1+...+\alpha_{n}X_n$ and $\mathcal{Y}:=\beta_1 X_1+...+\beta_{n}X_n$.

First, we must find $\mathbb{P}[\chi=x]$ and $\mathbb{P}[\mathcal{Y}=y]$. We could use convolutions inductively?

Then we need $\mathbb{P}[\chi = x \ and \ \mathcal{Y}=y]=\mathbb{P}[\chi=x | \mathcal{Y}=y]*\mathbb{P}[\mathcal{Y}=y]$


Any help with this problem is immensely appreciated.

$\endgroup$
1
$\begingroup$

Let $a=(\alpha_1,\alpha_2,\ldots,\alpha_n)^{T}$ and $b = (\beta_1, \beta_2, \ldots, \beta_n)^T$.

Sample space of $\chi$ is $\{v^{T}a\text{ : } v \text{ is a binary n-vector with 0's replaced with -1's}\}$.

Sample space of $\mathcal{Y}$ is $\{v^{T}b\text{ : } v \text{ is a binary n-vector with 0's replaced with -1's}\}$.

Sample space of $(\chi,\mathcal{Y})$ is $\{(v^{T}a,v^{T}b)\text{ : } v \text{ is a binary n-vector with 0's replaced with -1's}\}$.

$$P(\chi=v_{0}^Ta, \mathcal{Y}=v_{0}^Tb) = P(X_1=v_{01}, X_2=v_{02}, \ldots, X_n=v_{0n}) = \prod_{i=1}^{n}P(X_i=v_{0i}) = \left(\frac{1}{2}\right)^n$$

For a fixed $x$ and $y$, let $V = \{v_{k} \text{ : } v_k^Ta = x \text{ and } v_k^Tb = y\}$, then,

$$P(\chi=x, \mathcal{Y}=y) = \sum_{v_{k}\in V} P(\chi=v_k^Ta,\mathcal{Y}=v_k^tb) = \sum_{v_k\in V}\left(\frac{1}{2}\right)^n = \frac{|V|}{2^n}$$

I am not sure if a better formulation of joint pmf is possible based on the condition OP wrote in the question.

$\endgroup$

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.