I am wondering what is the $E[\textbf{a}\cdot \textbf{b}]$ and $var[\textbf{a}\cdot \textbf{b}]$ where $\textbf{a}, \textbf{b}$ are independent random vectors. That is as a vector whose elements are random variables. There are n elemetns in the vector. Each element in vector is assumed to be random sample from a normal distribution with mean 0 and variance $\sigma^{2}=1/n$. and $\cdot$ denotes dot product.

I read somewhere that \begin{equation} \begin{aligned} E(\tilde{\textbf{a}}\cdot \tilde{\textbf{b}})&=E(\sum_{i=1}^{n} a_{i} b_{i})\\ &= n E(XY)= 0 \end{aligned} \end{equation}

\begin{equation} \begin{aligned} var(\tilde{\textbf{a}}\cdot \tilde{\textbf{b}})&=var(\sum_{i=1}^{n} a_{i} b_{i})\\ &=n~var(XY) \\ &=\dfrac{1}{n} \end{aligned} \end{equation}

How we can say $var(\sum_{i=1}^{n} a_{i} b_{i}) =n~var(XY)$ or $E(\sum_{i=1}^{n} a_{i} b_{i}) =n~E(XY)$. Does anyone have an idea on this?

  • 1
    $\begingroup$ I think that the Matrix Cookbook will help you out here, specifically the section on General Statistics and Probability $\endgroup$ – ebb-earl-co Dec 11 '17 at 17:13
  • 2
    $\begingroup$ Are you familiar with properties of expectation? The relevant one is linearity. This is not a matrix problem--it requires only the most straightforward application of linearity. $\endgroup$ – whuber Dec 11 '17 at 17:14
  • $\begingroup$ Thank you for your reply @whuber. I am familiar with properties of expectation more or less, but I was not sure that is correct: $var(\sum_{i=1}^{n} a_{i} b_{i})=E((\sum_{i=1}^{n} a_{i} b_{i})^{2})- (E(\sum_{i=1}^{n} a_{i} b_{i}))^{2} =n(E((XY)^{2})-(E(XY))^{2})$. It shouldn't be $n^{2} (E((XY)^{2})-(E(XY))^{2})$? $\endgroup$ – Niki Dec 11 '17 at 17:30

\begin{equation} \begin{aligned} E(\sum_{i=1}^{n} a_{i} b_{i}) &=\sum_{i=1}^{n}E( a_{i} b_{i}) \text{, due to linearity}\\ &= n E(XY) \text{ , due to i.i.d}\\ \end{aligned} \end{equation} Note that variance of sum of independent variables is equal to the sum of their variance. \begin{equation} \begin{aligned} var(\sum_{i=1}^{n} a_{i} b_{i}) &=\sum_{i=1}^{n}var( a_{i} b_{i}) \\ &=n~var(XY) \text{, due to i.i.d}\\ \end{aligned} \end{equation}

Here $X,Y$ are independent are follows distribution $N(0,\sigma^2)$. You will have to use the property that $X$ and $Y$ are independent to evalute $E(XY)$ and $var(XY)$.

  • $\begingroup$ How should be $var(\sum_{x} (\sum_{i} a_{x-i}b_{i} \cdot \sum_{i'} a_{x-i'}c_{i'}))$? this is not correct $n^{3}var(X^{2}YZ)$, am I right? $\endgroup$ – Niki Dec 11 '17 at 19:00
  • $\begingroup$ where does $c$ come from? $\endgroup$ – Siong Thye Goh Dec 11 '17 at 19:09
  • $\begingroup$ Oh,I forgot to write, $\textbf{c}, \textbf{a},\textbf{b} $ are independent random vectors. It is another equation that I want to know its variance. $\endgroup$ – Niki Dec 11 '17 at 19:36
  • $\begingroup$ @SiongThyeGoh hey could you please explain the second step in expectation for which the reason is i.i.d ? Thanks $\endgroup$ – Siddharth Jan 15 at 3:38
  • $\begingroup$ $E(a_1b_1)=E(a_2b_2) = \ldots=E(a_nb_n)=E(XY)$, then we sum them up. $\endgroup$ – Siong Thye Goh Jan 15 at 3:42

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.