Index: The Book of Statistical ProofsGeneral Theorems ▷ Information theory ▷ Discrete mutual information ▷ Relation to marginal and joint entropy

Theorem: Let $X$ and $Y$ be discrete random variables with the joint probability $p(x,y)$ for $x \in \mathcal{X}$ and $y \in \mathcal{Y}$. Then, the mutual information of $X$ and $Y$ can be expressed as

$\label{eq:dmi-mje} \mathrm{I}(X,Y) = \mathrm{H}(X) + \mathrm{H}(Y) - \mathrm{H}(X,Y)$

where $\mathrm{H}(X)$ and $\mathrm{H}(Y)$ are the marginal entropies of $X$ and $Y$ and $\mathrm{H}(X,Y)$ is the joint entropy.

Proof: The mutual information of $X$ and $Y$ is defined as

$\label{eq:MI} \mathrm{I}(X,Y) = \sum_{x \in \mathcal{X}} \sum_{y \in \mathcal{Y}} p(x,y) \log \frac{p(x,y)}{p(x)\,p(y)} \; .$

Separating the logarithm, we have:

$\label{eq:MI-s1} \mathrm{I}(X,Y) = \sum_x \sum_y p(x,y) \log p(x,y) - \sum_x \sum_y p(x,y) \log p(x) - \sum_x \sum_y p(x,y) \log p(y) \; .$

$\label{eq:MI-s2} \mathrm{I}(X,Y) = \sum_x \sum_y p(x,y) \log p(x,y) - \sum_x \left( \sum_y p(x,y) \right) \log p(x) - \sum_y \left( \sum_x p(x,y) \right) \log p(y) \; .$

Applying the law of marginal probability, i.e. $p(x) = \sum_y p(x,y)$, we get:

$\label{eq:MI-s3} \mathrm{I}(X,Y) = \sum_x \sum_y p(x,y) \log p(x,y) - \sum_x p(x) \log p(x) - \sum_y p(y) \log p(y) \; .$

Now considering the definitions of marginal and joint entropy

$\label{eq:ME-JE} \begin{split} \mathrm{H}(X) &= - \sum_{x \in \mathcal{X}} p(x) \log p(x) \\ \mathrm{H}(X,Y) &= - \sum_{x \in \mathcal{X}} \sum_{y \in \mathcal{Y}} p(x,y) \log p(x,y) \; , \end{split}$

we can finally show:

$\label{eq:MI-qed} \begin{split} \mathrm{I}(X,Y) &= - \mathrm{H}(X,Y) + \mathrm{H}(X) + \mathrm{H}(Y) \\ &= \mathrm{H}(X) + \mathrm{H}(Y) - \mathrm{H}(X,Y) \; . \end{split}$
Sources:

Metadata: ID: P20 | shortcut: dmi-mje | author: JoramSoch | date: 2020-01-13, 21:53.