Mutual information maximization over input distributions via iterative algorithm
C = — bits | Iterations: — | Input: uniform
0.10
30
Shannon's channel capacity C = max_{p(x)} I(X;Y) is the supreme reliable
communication rate. The Blahut-Arimoto algorithm (1972) computes C via alternating
optimization: q(x|y) ← p(x)·exp(∑_y p(y|x)log q(x|y)) then normalize p(x).
Each iteration increases I(X;Y) monotonically. For the BSC with crossover ε:
C = 1 - H_b(ε). The Z-channel shows how asymmetric channels have non-uniform capacity-achieving distributions.