Optimizing a constrained model. The flowchart summarizes the computational steps needed to optimize a kinetic model, subject to parameter and behavioral constraints. Linear parameter constraints are implemented via linear algebra transformations between the model parameters K and the free parameters $X ¯$⁠, whereas behavioral constraints or arbitrary parameter relationships are handled by a penalized cost function $F'(X¯k,αp)$ that measures the overall error of the model relative to the data and the constraints. The $K→X ¯$ and $X ¯ →K$ transformations are detailed in Fig. 3 in the companion paper (Salari et al., 2018). To calculate the cost function, one needs to generate the response of the model (e.g., probability distributions and macroscopic currents) to the same stimulation protocols used to generate the experimental data and formulate the behavioral constraints. The inner computational loop, indexed by k, optimizes the model for a given penalty factor α_p, whereas the outer loop, indexed by p, gradually increases α_p, to more tightly satisfy the behavioral and arbitrary parameter constraints.