Initial conditions that minimize convergence time in CRN-based NN computation

Identify initial concentration configurations of species in the proposed forward, backward, and update chemical reaction networks that minimize convergence time to steady states under the one-pot computation scheme, and characterize how these optimal initial conditions depend on system parameters.

Background

The authors prove exponential convergence of the forward and backward CRNs but note that convergence speed can be delayed by system components, including reaction rates and initial species amounts.

They explicitly acknowledge uncertainty about which initial conditions minimize convergence time, motivating a precise optimization of initial concentrations for efficient biochemical NN training.