Let $X_1$ and $X_2$ be discrete random variables drawn according to probability mass function $p_1$ and $p_2$ over the respective alhabets $X_1={1,2,...m}$ and $X_2={m+1,...,n}$. Let $X=X_1$ with probability $a$ and $X=X_2$ with probability $(1-a)$. Quesiton: Find $H(X)$ in terms of $H(X_1)$ and $H(X_2)$ and $a$.

This problem is from Elements of Information Theory by Thomas M. Cover, Joy A. Thomas.

How to solve the problem ? Actually I am more confused why the answer is not simply $H(X)=-a*log(a) - (1-a)*log(1-a)$.

(The correct answer is $H(X)=-a*log(a) - (1-a)*log(1-a)+aH(X_1)+(1-a)H(X_2)$ according to the solution manual)



Let $Z$ be the random variable such that $Z=0$ if $X=X_1$ and $Z=1$ if $X=X_2$. To write it neatly: $H(X,Z) = H(X) + H(Z|X) = H(X)$ because $H(Z|X) = 0$.

Now write $H(X,Z) = H(Z) + H(X|Z) = H(Z) + p(Z=0) H(X|Z=0)+p(Z=1) H(X|Z=1)$

Now use $H(X|Z=0) = H(X_1)$,$H(X|Z=1) = H(X_2)$ in above equation to get: $H(X,Z) = H(Z) + H(X|Z) = H(Z) + a H(X_1)+(1-a) H(X_2)$


You have $P(X=i)=ap_1(i)$ for $i=1,\ldots,m$ so the contribution to the total entropy from the $X_1$ part equals $$-\sum_i ap_1(i)\log(ap_1(i)) = -a\log{a}\sum_i p_1(i) - a\sum_i p_1(i)\log{p_1(i)} = -a\log{a}+aH(X_1)$$ and then there is a similar term with $(1-a)$ and $X_2$.


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.