Open
Description
Theoretically it wouldn't be too difficult to implement the method Chemkin uses for sensitivities. Fundamentally in this method you run the simulation and then you discretize the sensitivity equations into finite differences between each time point. These can be solved sequentially time step to time step and the sensitivity to each parameter is independent so it's simply a matter of posing the set of equations for sensitivity to a single parameter for a single time step, solving the equations and then reapplying that for each parameter and time step. Parallelization would probably be handy. Some care probably needed to let users decide what sensitivities to save. See https://aip.scitation.org/doi/pdf/10.1063/1.447938