Generation-by-Generation Dissection of the Response Function in Long Memory Epidemic Processes

In a number of natural and social systems, the response to an exogenous shock relaxes back to the average level according to a long-memory kernel $\sim 1/t^{1+\theta}$ with $0 \leq \theta <1$. In the presence of an epidemic-like process of triggered shocks developing in a cascade of generations at or close to criticality, this "bare" kernel is renormalized into an even slower decaying response function $\sim 1/t^{1-\theta}$. Surprisingly, this means that the shorter the memory of the bare kernel (the larger $1+\theta$), the longer the memory of the response function (the smaller $1-\theta$). Here, we present a detailed investigation of this paradoxical behavior based on a generation-by-generation decomposition of the total response function, the use of Laplace transforms and of "anomalous" scaling arguments. The paradox is explained by the fact that the number of triggered generations grows anomalously with time at $\sim t^\theta$ so that the contributions of active generations up to time $t$ more than compensate the shorter memory associated with a larger exponent $\theta$. This anomalous scaling results fundamentally from the property that the expected waiting time is infinite for $0 \leq \theta \leq 1$. The techniques developed here are also applied to the case $\theta >1$ and we find in this case that the total renormalized response is a {\bf constant} for $t < 1/(1-n)$ followed by a cross-over to $\sim 1/t^{1+\theta}$ for $t \gg 1/(1-n)$.