The power law of forgetting, established empirically by Jost (1897) and formalized by Wixted and Ebbesen (1991), states that memory retention decreases as a power function of the retention interval. This function provides a better fit to forgetting data than the exponential function originally proposed by Ebbinghaus, across a remarkably wide range of memory tasks, materials, and time scales.
R = retention (proportion recalled)
t = time since learning
a = initial memory strength
b = rate of forgetting
Why Power Rather Than Exponential?
Anderson and Schooler (1991) showed that the power law of forgetting mirrors the statistical structure of the environment: the probability that information from the past will be needed in the future follows a power function of the time elapsed. This suggests that forgetting is rationally adapted to the statistics of information relevance. The ACT-R cognitive architecture implements this principle directly, using base-level activation decay as a power function of time since last retrieval.