the critical exponent log_b a

Every divide-and-conquer recurrence T(n) = a T(n/b) + f(n) has one special growth rate that acts as a dividing line. Compare your combine cost f(n) against this special rate and you instantly know which way the recurrence tips. That dividing line is n raised to the power log_b a — and the exponent log_b a is the heart of the Master Theorem.

Where does it come from? Count the leaves of the recursion tree. The tree has depth log_b n (the number of times you divide n by b before reaching 1), and each node has a children, so the number of leaves is a^(log_b n). By the change-of-base/log rules, a^(log_b n) = n^(log_b a). So n^(log_b a) is exactly the amount of work done at the leaves (each leaf doing Theta(1)). This is why it is the benchmark: if the combine cost f(n) is smaller than this leaf cost, the leaves dominate (Case 1); if equal, every level matches and you get a log factor (Case 2); if larger, the root dominates (Case 3). For merge sort a = b = 2 so log_2 2 = 1; for Strassen a = 7, b = 2 so log_2 7 ≈ 2.807.

Thinking in terms of this exponent makes the whole Master Theorem one comparison: is f(n) below, at, or above n^(log_b a)? It also explains why clever algorithms try to LOWER a (fewer subproblems) more than they worry about f: cutting Strassen's a from 8 to 7 dropped the exponent from log_2 8 = 3 to log_2 7 ≈ 2.81, a real asymptotic win. The honest caveat: 'above' and 'below' must be by a polynomial factor; merely beating n^(log_b a) by a log factor lands you in a Master-Theorem gap, not cleanly in Case 1 or 3.