| Line 4: | Line 4: | ||
'''Fact''' | '''Fact''' | ||
| − | Let <math>\mathbf{X}</math> and <math>\mathbf{Y}</math> be two jointly-distributed, statistically independent random variables, having pdfs <math>f_{\mathbf{X}}\left(x\right)</math> and <math>f_{\mathbf{Y}}\left(y\right)</math> . Then the pdf of <math>\mathbf{Z}=\mathbf{X}+\mathbf{Y}</math> is | + | Let <math class="inline">\mathbf{X}</math> and <math class="inline">\mathbf{Y}</math> be two jointly-distributed, statistically independent random variables, having pdfs <math class="inline">f_{\mathbf{X}}\left(x\right)</math> and <math class="inline">f_{\mathbf{Y}}\left(y\right)</math> . Then the pdf of <math class="inline">\mathbf{Z}=\mathbf{X}+\mathbf{Y}</math> is |
| − | <math>f_{\mathbf{Z}}\left(z\right)=\left(f_{\mathbf{X}}*f_{\mathbf{Y}}\right)\left(z\right)=\int_{-\infty}^{\infty}f_{\mathbf{X}}\left(x\right)f_{\mathbf{Y}}\left(z-x\right)dx=\int_{-\infty}^{\infty}f_{\mathbf{Y}}\left(y\right)f_{\mathbf{X}}\left(z-y\right)dy</math>. | + | <math class="inline">f_{\mathbf{Z}}\left(z\right)=\left(f_{\mathbf{X}}*f_{\mathbf{Y}}\right)\left(z\right)=\int_{-\infty}^{\infty}f_{\mathbf{X}}\left(x\right)f_{\mathbf{Y}}\left(z-x\right)dx=\int_{-\infty}^{\infty}f_{\mathbf{Y}}\left(y\right)f_{\mathbf{X}}\left(z-y\right)dy</math>. |
The discrete version of above equation is | The discrete version of above equation is | ||
| − | <math>P\left(\mathbf{Z}=z\right)=\sum_{k=-\infty}^{\infty}P\left(\mathbf{X}=k\right)P\left(\mathbf{Y}=z-k\right)=\sum_{k=-\infty}^{\infty}P\left(\mathbf{X}=z-k\right)P\left(\mathbf{Y}=k\right)</math>. | + | <math class="inline">P\left(\mathbf{Z}=z\right)=\sum_{k=-\infty}^{\infty}P\left(\mathbf{X}=k\right)P\left(\mathbf{Y}=z-k\right)=\sum_{k=-\infty}^{\infty}P\left(\mathbf{X}=z-k\right)P\left(\mathbf{Y}=k\right)</math>. |
Example. Sum of two exponential random variables | Example. Sum of two exponential random variables | ||
| − | Let <math>\mathbf{X}</math> and <math>\mathbf{Y}</math> be two jointly-distributed RVs having exponential distributions with mean <math>\mu</math> . Let <math>\mathbf{Z}=\mathbf{X}+\mathbf{Y}</math> . Find <math>f_{\mathbf{Z}}\left(z\right)</math> . | + | Let <math class="inline">\mathbf{X}</math> and <math class="inline">\mathbf{Y}</math> be two jointly-distributed RVs having exponential distributions with mean <math class="inline">\mu</math> . Let <math class="inline">\mathbf{Z}=\mathbf{X}+\mathbf{Y}</math> . Find <math class="inline">f_{\mathbf{Z}}\left(z\right)</math> . |
| − | <math>f_{\mathbf{Z}}\left(z\right)</math> | + | <math class="inline">f_{\mathbf{Z}}\left(z\right)</math> |
Joint Gaussian pdf | Joint Gaussian pdf | ||
| − | <math>f_{\mathbf{XY}}(x,y)=\frac{1}{2\pi\sigma_{\mathbf{X}}\sigma_{\mathbf{Y}}\sqrt{1-r^{2}}}\exp\left\{ \frac{-1}{2\left(1-r^{2}\right)}\left[\frac{\left(x-\mu_{\mathbf{X}}\right)^{2}}{\sigma_{\mathbf{X}}^{2}}-\frac{2r\left(x-\mu_{\mathbf{X}}\right)\left(y-\mu_{\mathbf{Y}}\right)}{\sigma_{\mathbf{X}}\sigma_{\mathbf{Y}}}+\frac{\left(y-\mu_{\mathbf{Y}}\right)^{2}}{\sigma_{\mathbf{Y}}^{2}}\right]\right\}</math> | + | <math class="inline">f_{\mathbf{XY}}(x,y)=\frac{1}{2\pi\sigma_{\mathbf{X}}\sigma_{\mathbf{Y}}\sqrt{1-r^{2}}}\exp\left\{ \frac{-1}{2\left(1-r^{2}\right)}\left[\frac{\left(x-\mu_{\mathbf{X}}\right)^{2}}{\sigma_{\mathbf{X}}^{2}}-\frac{2r\left(x-\mu_{\mathbf{X}}\right)\left(y-\mu_{\mathbf{Y}}\right)}{\sigma_{\mathbf{X}}\sigma_{\mathbf{Y}}}+\frac{\left(y-\mu_{\mathbf{Y}}\right)^{2}}{\sigma_{\mathbf{Y}}^{2}}\right]\right\}</math> |
| − | If <math>r=0</math> (<math>\mathbf{X}</math> and <math>\mathbf{Y}</math> are uncorrelated), | + | If <math class="inline">r=0</math> (<math class="inline">\mathbf{X}</math> and <math class="inline">\mathbf{Y}</math> are uncorrelated), |
| − | <math>f_{\mathbf{XY}}(x,y)</math> | + | <math class="inline">f_{\mathbf{XY}}(x,y)</math> |
| − | <math>\Longrightarrow\mathbf{X}</math> and <math>\mathbf{Y}</math> are independent. | + | <math class="inline">\Longrightarrow\mathbf{X}</math> and <math class="inline">\mathbf{Y}</math> are independent. |
| − | If two RVs <math>\mathbf{X}</math> and <math>\mathbf{Y}</math> are uncorrelated Gaussian, then <math>\mathbf{X}</math> and <math>\mathbf{Y}</math> are independent. This is the exception of “<math>\mathbf{X}</math> and <math>\mathbf{Y}</math> are independent <math>\left(\nLeftarrow\right)\Rightarrow \mathbf{X}</math> and <math>\mathbf{Y}</math> are uncorrelated”. | + | If two RVs <math class="inline">\mathbf{X}</math> and <math class="inline">\mathbf{Y}</math> are uncorrelated Gaussian, then <math class="inline">\mathbf{X}</math> and <math class="inline">\mathbf{Y}</math> are independent. This is the exception of “<math class="inline">\mathbf{X}</math> and <math class="inline">\mathbf{Y}</math> are independent <math class="inline">\left(\nLeftarrow\right)\Rightarrow \mathbf{X}</math> and <math class="inline">\mathbf{Y}</math> are uncorrelated”. |
Definition. Uncorrelation | Definition. Uncorrelation | ||
| Line 34: | Line 34: | ||
Two RVs \mathbf{X} and \mathbf{Y} are uncorrelated if their covariance is equal to zero. This is true if any of the following three equivalent condition is true: | Two RVs \mathbf{X} and \mathbf{Y} are uncorrelated if their covariance is equal to zero. This is true if any of the following three equivalent condition is true: | ||
| − | 1. <math>Cov\left(\mathbf{X},\mathbf{Y}\right)=0</math> | + | 1. <math class="inline">Cov\left(\mathbf{X},\mathbf{Y}\right)=0</math> |
| − | 2. <math>r_{\mathbf{XY}}=0</math> | + | 2. <math class="inline">r_{\mathbf{XY}}=0</math> |
| − | 3. <math>E\left[\mathbf{XY}\right]=E\left[\mathbf{X}\right]E\left[\mathbf{Y}\right]</math> | + | 3. <math class="inline">E\left[\mathbf{XY}\right]=E\left[\mathbf{X}\right]E\left[\mathbf{Y}\right]</math> |
Note | Note | ||
| − | • <math>\mathbf{X}</math> and <math>\mathbf{Y}</math> are uncorrelated <math>\Longleftrightarrow E\left[\mathbf{XY}\right]=E\left[\mathbf{X}\right]\cdot E\left[\mathbf{Y}\right]\Longleftrightarrow Cov\left(\mathbf{X},\mathbf{Y}\right)=0</math> | + | • <math class="inline">\mathbf{X}</math> and <math class="inline">\mathbf{Y}</math> are uncorrelated <math class="inline">\Longleftrightarrow E\left[\mathbf{XY}\right]=E\left[\mathbf{X}\right]\cdot E\left[\mathbf{Y}\right]\Longleftrightarrow Cov\left(\mathbf{X},\mathbf{Y}\right)=0</math> |
| − | • <math>\mathbf{X}</math> and <math>\mathbf{Y}</math> are independent <math>\Longleftrightarrow f_{\mathbf{XY}}(x,y)=f_{\mathbf{X}}(x)\cdot f_{\mathbf{Y}}(y)</math> | + | • <math class="inline">\mathbf{X}</math> and <math class="inline">\mathbf{Y}</math> are independent <math class="inline">\Longleftrightarrow f_{\mathbf{XY}}(x,y)=f_{\mathbf{X}}(x)\cdot f_{\mathbf{Y}}(y)</math> |
| − | • <math>\mathbf{X}</math> and <math>\mathbf{Y}</math> are independent <math>\left(\nLeftarrow\right)\Longrightarrow \mathbf{X}</math> and <math>\mathbf{Y}</math> are uncorrelated | + | • <math class="inline">\mathbf{X}</math> and <math class="inline">\mathbf{Y}</math> are independent <math class="inline">\left(\nLeftarrow\right)\Longrightarrow \mathbf{X}</math> and <math class="inline">\mathbf{Y}</math> are uncorrelated |
Definition. Orthogonality | Definition. Orthogonality | ||
| − | Two RVs are orthogoal if <math>E\left[\mathbf{XY}\right]=0</math> . | + | Two RVs are orthogoal if <math class="inline">E\left[\mathbf{XY}\right]=0</math> . |
Fact | Fact | ||
| − | If <math>E\left[\mathbf{X}^{2}\right]<\infty and E\left[\mathbf{Y}^{2}\right]<\infty , then \left|E\left[\mathbf{XY}\right]\right|\leq\sqrt{E\left[\mathbf{X}\right]^{2}E\left[\mathbf{Y}\right]^{2}}</math> with equality iff <math>\mathbf{Y}=a_{0}\mathbf{X}</math> where <math>a_{0}</math> is constant. | + | If <math class="inline">E\left[\mathbf{X}^{2}\right]<\infty and E\left[\mathbf{Y}^{2}\right]<\infty , then \left|E\left[\mathbf{XY}\right]\right|\leq\sqrt{E\left[\mathbf{X}\right]^{2}E\left[\mathbf{Y}\right]^{2}}</math> with equality iff <math class="inline">\mathbf{Y}=a_{0}\mathbf{X}</math> where <math class="inline">a_{0}</math> is constant. |
Recall | Recall | ||
| − | For a quadratic equation <math>ax^{2}+bx+c=0,\; a\neq0</math> , the discriminant is <math>b^{2}-4ac</math> . | + | For a quadratic equation <math class="inline">ax^{2}+bx+c=0,\; a\neq0</math> , the discriminant is <math class="inline">b^{2}-4ac</math> . |
Proof | Proof | ||
| − | <math>E\left[\left(a\mathbf{X}-\mathbf{Y}\right)^{2}\right]\geq0\Longrightarrow a^{2}E\left[\mathbf{X}^{2}\right]-2aE\left[\mathbf{XY}\right]+E\left[\mathbf{Y}^{2}\right]\geq0.</math> | + | <math class="inline">E\left[\left(a\mathbf{X}-\mathbf{Y}\right)^{2}\right]\geq0\Longrightarrow a^{2}E\left[\mathbf{X}^{2}\right]-2aE\left[\mathbf{XY}\right]+E\left[\mathbf{Y}^{2}\right]\geq0.</math> |
n.b. LHS is a quadratic in a . | n.b. LHS is a quadratic in a . | ||
| − | Let's consider two cases: <math>(i)\; E\left[\left(a\mathbf{X}-\mathbf{Y}\right)^{2}\right]>0 , (ii)\; E\left[\left(a\mathbf{X}-\mathbf{Y}\right)^{2}\right]=0</math> . | + | Let's consider two cases: <math class="inline">(i)\; E\left[\left(a\mathbf{X}-\mathbf{Y}\right)^{2}\right]>0 , (ii)\; E\left[\left(a\mathbf{X}-\mathbf{Y}\right)^{2}\right]=0</math> . |
| − | (i) <math>0<E\left[\left(a\mathbf{X}-\mathbf{Y}\right)^{2}\right]=a^{2}E\left[\mathbf{X}^{2}\right]-2aE\left[\mathbf{XY}\right]+E\left[\mathbf{Y}^{2}\right]</math> | + | (i) <math class="inline">0<E\left[\left(a\mathbf{X}-\mathbf{Y}\right)^{2}\right]=a^{2}E\left[\mathbf{X}^{2}\right]-2aE\left[\mathbf{XY}\right]+E\left[\mathbf{Y}^{2}\right]</math> |
| − | <math>\Longrightarrow</math> quadratic in a has complex roots <math>\Longrightarrow</math> “discriminant” of this quadratic is negative | + | <math class="inline">\Longrightarrow</math> quadratic in a has complex roots <math class="inline">\Longrightarrow</math> “discriminant” of this quadratic is negative |
| − | <math>4a^{2}E\left[\mathbf{XY}\right]^{2}-4a^{2}E\left[\mathbf{X}^{2}\right]E\left[\mathbf{Y}^{2}\right]</math> | + | <math class="inline">4a^{2}E\left[\mathbf{XY}\right]^{2}-4a^{2}E\left[\mathbf{X}^{2}\right]E\left[\mathbf{Y}^{2}\right]</math> |
| − | (ii) <math>0=E\left[\left(a\mathbf{X}-\mathbf{Y}\right)^{2}\right]=a^{2}E\left[\mathbf{X}^{2}\right]-2aE\left[\mathbf{XY}\right]+E\left[\mathbf{Y}^{2}\right]</math> for some <math>a=a_{0}</math> . | + | (ii) <math class="inline">0=E\left[\left(a\mathbf{X}-\mathbf{Y}\right)^{2}\right]=a^{2}E\left[\mathbf{X}^{2}\right]-2aE\left[\mathbf{XY}\right]+E\left[\mathbf{Y}^{2}\right]</math> for some <math class="inline">a=a_{0}</math> . |
| − | n.b. The discriminant is equal to <math>0</math> when <math>\mathbf{Y}=a_{0}\mathbf{X}</math> . | + | n.b. The discriminant is equal to <math class="inline">0</math> when <math class="inline">\mathbf{Y}=a_{0}\mathbf{X}</math> . |
---- | ---- | ||
[[ECE600|Back to ECE600]] | [[ECE600|Back to ECE600]] | ||
[[ECE 600 Prerequisites|Back to ECE 600 Prerequisites]] | [[ECE 600 Prerequisites|Back to ECE 600 Prerequisites]] | ||
Latest revision as of 11:31, 30 November 2010
1.10 Two Random Variables
From the ECE600 Pre-requisites notes of Sangchun Han, ECE PhD student.
Fact
Let $ \mathbf{X} $ and $ \mathbf{Y} $ be two jointly-distributed, statistically independent random variables, having pdfs $ f_{\mathbf{X}}\left(x\right) $ and $ f_{\mathbf{Y}}\left(y\right) $ . Then the pdf of $ \mathbf{Z}=\mathbf{X}+\mathbf{Y} $ is
$ f_{\mathbf{Z}}\left(z\right)=\left(f_{\mathbf{X}}*f_{\mathbf{Y}}\right)\left(z\right)=\int_{-\infty}^{\infty}f_{\mathbf{X}}\left(x\right)f_{\mathbf{Y}}\left(z-x\right)dx=\int_{-\infty}^{\infty}f_{\mathbf{Y}}\left(y\right)f_{\mathbf{X}}\left(z-y\right)dy $.
The discrete version of above equation is
$ P\left(\mathbf{Z}=z\right)=\sum_{k=-\infty}^{\infty}P\left(\mathbf{X}=k\right)P\left(\mathbf{Y}=z-k\right)=\sum_{k=-\infty}^{\infty}P\left(\mathbf{X}=z-k\right)P\left(\mathbf{Y}=k\right) $.
Example. Sum of two exponential random variables
Let $ \mathbf{X} $ and $ \mathbf{Y} $ be two jointly-distributed RVs having exponential distributions with mean $ \mu $ . Let $ \mathbf{Z}=\mathbf{X}+\mathbf{Y} $ . Find $ f_{\mathbf{Z}}\left(z\right) $ .
$ f_{\mathbf{Z}}\left(z\right) $
Joint Gaussian pdf
$ f_{\mathbf{XY}}(x,y)=\frac{1}{2\pi\sigma_{\mathbf{X}}\sigma_{\mathbf{Y}}\sqrt{1-r^{2}}}\exp\left\{ \frac{-1}{2\left(1-r^{2}\right)}\left[\frac{\left(x-\mu_{\mathbf{X}}\right)^{2}}{\sigma_{\mathbf{X}}^{2}}-\frac{2r\left(x-\mu_{\mathbf{X}}\right)\left(y-\mu_{\mathbf{Y}}\right)}{\sigma_{\mathbf{X}}\sigma_{\mathbf{Y}}}+\frac{\left(y-\mu_{\mathbf{Y}}\right)^{2}}{\sigma_{\mathbf{Y}}^{2}}\right]\right\} $
If $ r=0 $ ($ \mathbf{X} $ and $ \mathbf{Y} $ are uncorrelated),
$ f_{\mathbf{XY}}(x,y) $
$ \Longrightarrow\mathbf{X} $ and $ \mathbf{Y} $ are independent.
If two RVs $ \mathbf{X} $ and $ \mathbf{Y} $ are uncorrelated Gaussian, then $ \mathbf{X} $ and $ \mathbf{Y} $ are independent. This is the exception of “$ \mathbf{X} $ and $ \mathbf{Y} $ are independent $ \left(\nLeftarrow\right)\Rightarrow \mathbf{X} $ and $ \mathbf{Y} $ are uncorrelated”.
Definition. Uncorrelation
Two RVs \mathbf{X} and \mathbf{Y} are uncorrelated if their covariance is equal to zero. This is true if any of the following three equivalent condition is true:
1. $ Cov\left(\mathbf{X},\mathbf{Y}\right)=0 $
2. $ r_{\mathbf{XY}}=0 $
3. $ E\left[\mathbf{XY}\right]=E\left[\mathbf{X}\right]E\left[\mathbf{Y}\right] $
Note
• $ \mathbf{X} $ and $ \mathbf{Y} $ are uncorrelated $ \Longleftrightarrow E\left[\mathbf{XY}\right]=E\left[\mathbf{X}\right]\cdot E\left[\mathbf{Y}\right]\Longleftrightarrow Cov\left(\mathbf{X},\mathbf{Y}\right)=0 $
• $ \mathbf{X} $ and $ \mathbf{Y} $ are independent $ \Longleftrightarrow f_{\mathbf{XY}}(x,y)=f_{\mathbf{X}}(x)\cdot f_{\mathbf{Y}}(y) $
• $ \mathbf{X} $ and $ \mathbf{Y} $ are independent $ \left(\nLeftarrow\right)\Longrightarrow \mathbf{X} $ and $ \mathbf{Y} $ are uncorrelated Definition. Orthogonality
Two RVs are orthogoal if $ E\left[\mathbf{XY}\right]=0 $ .
Fact
If $ E\left[\mathbf{X}^{2}\right]<\infty and E\left[\mathbf{Y}^{2}\right]<\infty , then \left|E\left[\mathbf{XY}\right]\right|\leq\sqrt{E\left[\mathbf{X}\right]^{2}E\left[\mathbf{Y}\right]^{2}} $ with equality iff $ \mathbf{Y}=a_{0}\mathbf{X} $ where $ a_{0} $ is constant.
Recall
For a quadratic equation $ ax^{2}+bx+c=0,\; a\neq0 $ , the discriminant is $ b^{2}-4ac $ .
Proof
$ E\left[\left(a\mathbf{X}-\mathbf{Y}\right)^{2}\right]\geq0\Longrightarrow a^{2}E\left[\mathbf{X}^{2}\right]-2aE\left[\mathbf{XY}\right]+E\left[\mathbf{Y}^{2}\right]\geq0. $
n.b. LHS is a quadratic in a .
Let's consider two cases: $ (i)\; E\left[\left(a\mathbf{X}-\mathbf{Y}\right)^{2}\right]>0 , (ii)\; E\left[\left(a\mathbf{X}-\mathbf{Y}\right)^{2}\right]=0 $ .
(i) $ 0<E\left[\left(a\mathbf{X}-\mathbf{Y}\right)^{2}\right]=a^{2}E\left[\mathbf{X}^{2}\right]-2aE\left[\mathbf{XY}\right]+E\left[\mathbf{Y}^{2}\right] $ $ \Longrightarrow $ quadratic in a has complex roots $ \Longrightarrow $ “discriminant” of this quadratic is negative
$ 4a^{2}E\left[\mathbf{XY}\right]^{2}-4a^{2}E\left[\mathbf{X}^{2}\right]E\left[\mathbf{Y}^{2}\right] $
(ii) $ 0=E\left[\left(a\mathbf{X}-\mathbf{Y}\right)^{2}\right]=a^{2}E\left[\mathbf{X}^{2}\right]-2aE\left[\mathbf{XY}\right]+E\left[\mathbf{Y}^{2}\right] $ for some $ a=a_{0} $ .
n.b. The discriminant is equal to $ 0 $ when $ \mathbf{Y}=a_{0}\mathbf{X} $ .
