probabilities, so that the sum of the decimal fraction parts of all the boxes in each horizontal row add up to 100%, or as a real number, 1.00. Recall that in the example we’ve been using, the binary expansion of the natural number 729, the transition incidence matrix is M; = : and its Markov matrix is top row, 1/4, 3/4 and bottom row 3/5 , 2/5, i.e. Mzp = mn ne Matrix multiplication of Mzp by itself repeatedly is equivalent to tracking the temporal evolution of the transition matrix’s probabilities until the resulting matrices move toward, converge onto, a steady state; each self matrix multiplication step represents what results from the passage of one unit of time. The convergence to equilibrium values is continuous and gradual. When the steady state is reached, both rows become identical. For this example, Mip x Mep oF Min? _ 0.5125 0.4879. Min’ _ 0.4527 09472, Mip® _ 0.4445 0.9994. 0.3900 0.6100 0.4377 0.5622 0.4443 0.5556 Mp '° = as ee which for the first four decimal places remain the same for additional times of self multiplication. Note the convergence of the top and bottom rows to the same asymptotic values. Books discussing the multiplicative and other behavior of these nonnegative matrices are numerous and frequently appear in matrix algebra texts under the rubric of the Frobenius-Perron theorems. Using the entropy formalism of Claude Shannon as developed previously, Hy is computed as the sum across either of the identical rows of each probability times its logarithm, px /og(p,p2x/og(p2)) remembering from above that we are working in base 2 logarithms and to change the minus sign (resulting from taking the logarithms of decimal fractions) to plus: Hy (Mip) = .4444 x log(.4444) + 5555 x log(.5555) = .9911 The nonuniformity of the box occupancy probabilities is reflected in the difference between the topological (maximal estimate) and metric (minimal estimate) entropies and is therefore quantifiable and computable: H; - Hy # 0 = 1.00 - 0.9911 = 0. 0089. I