prove this inequality related to probability and information theory

107 Views Asked by At

How do I prove this? I'm thinking I should use Jensen's inequality somehow.

$$\sum_K p_k(1-p_k) \le -\sum_K p_k\log p_k$$

The assumption that $\sum_K p_k=1$ holds.

1

There are 1 best solutions below

11
On BEST ANSWER

As explained in the comments, the inequality $$\sum_kp_k(1-p_k)\lt\sum_kp_k\log p_k$$ cannot hold since the LHS is nonnegative and the RHS is nonpositive.

To show that $$\sum_kp_k(1-p_k)\leqslant-\sum_kp_k\log p_k,$$ note that, for every positive $x$, $$1-x\leqslant-\log x.$$