Fair points. I've made a few refinements already on paper, switching to an ODE model rather than a difference equation and changing the knowledge production rat

Fair points. I've made a few refinements already on paper, switching to an ODE model rather than a difference equation and changing the knowledge production rate to vary. If there are diminishing returns then we would expect c2 to decrease as k increases at large k. As you're aware, perhaps this is offset by increases in population. As you've indicated, heterogeneity in various parts of the problem make this sort of modeling questionable. It's a crude model not meant to make precise predictions, just show that knowledge loss can be a major factor at long times.