Coupling Algorithms for Calculating Sensitivities of Population Balances
We introduce a new class of stochastic algorithms for calculating parametric derivatives of the solution of the space-homogeneous Smoluchowski’s coagulation equation. Currently, it is very difficult to produce low variance estimates of these derivatives in reasonable amounts of computational time through the use of stochastic methods. These new algorithms consider a central difference estimator of the parametric derivative which is calculated by evaluating the coagulation equation at two different parameter values simultaneously, and causing variance reduction by maximising the covariance between these. The two different coupling strategies (‘Single’ and ‘Double’) have been compared to the case when there is no coupling (‘Independent’). Both coupling algorithms converge and the Double coupling is the most ‘efficient’ algorithm. For the numerical example chosen we obtain a factor of about 100 in efficiency in the best case (small system evolution time and small parameter perturbation).
- There are no access options currently available. Please get in touch witClus directly.
Keywords: agglomeration, coagulation, convergence, modelling, Monte Carlo, numerical convergence, sensitivity analysis, simulation, stochastic modeling, stochastic modelling, stochastic simulation, variance reduction,