Memory Capacity Bounds via Spherical Random Sequential Adsorption

Paper 1 derives the first fundamental constraint of adaptive memory systems: a geometric capacity limit. Using sphere-packing arguments and random sequential adsorption on the state manifold, it shows that the number of distinct stable representations is bounded by the geometry of the representational space rather than simply by dimensionality. This result explains why increasing parameter count or embedding width does not indefinitely increase usable representational capacity.