Figure 12.

The dual-integrator model. Only the components that differ from those in Figure 11 are shown; the remaining components are common to the single- and dual-integrator models. The rewarding effect arises from the direct activation of two subpopulations of neurons, each of which projects to a different spatio-temporal integrator; the weighted outputs of these two integrators are pooled. One integrator has a shorter chronaxie than the other, which renders it less sensitive to the reduction of the train duration from 1 to 0.25 s. Note that the pulse-frequency axis is inverted: the frequency decreases from left to right. The translucent blue planes are positioned at F_{Near Max} (F_nm), the pulse frequency beyond which reward intensity approaches asymptote. At both the short and long train durations, the growth of reward intensity at the output of integrator 1 (upper 3D graph) is largely complete at pulse frequencies lower than F_{Near Max}. Although reward growth at the output of integrator 2 is also largely complete at pulse frequencies lower than F_{Near Max} when the train duration is 1 s (lower 3D graph), reward growth has not yet begun at this pulse frequency when the train duration is 0.25 s. Thus, reward fails to grow as a function of pulse frequency when the train duration is short, and the summed output of the two integrators (right-hand 3D graph) is lower at the short train duration than at the long. The parameters used to generate the reward-growth functions in Figure 12 are from the fit of the dual-integrator model to the data from rat C17 (Figures 13, 14).