This sort of drives how energy is used. We have whatever the energy rules lead to, but nothing like an equilibrium, or a ceiling or floor. This has me thinking about how CAs are computed. To maintain an equilibrium at the system level, we would want to have something like this partially computed t+1 in a CA:
[ 0.1 ] [ 0 ] [ 0 ]
[ 0.1 ] [ ??? ] [ -.5 ]
[ 0 ] [ -.5 ] [ -.4 ]
And to calculate the ??? in a way that the system sum stays at at least zero, we would need to know the t+1 for all of the other cells. But CAs don't know the future states of other cells, just the current state. In other words, we are doing something like this:
For all i from 0 to n:
xi' = f(i, {x1, x2, x3, ...})
Which is a serialized version of this:
{x1', x2', x3', ...} = g({x1, x2, x3, ...})
Which would calculate everything in one fell swoop. I suggested to Yaeger that iterating over f, you can't do things like ensure that the result maintains some sort of equilibrium. But Yaeger pointed out that you certainly can, you just have to add a normalization step at the end:
xi'' = n(i, {x1', x2', x3', ...})
And in a way, this is "knowing the future".