CENTER FOR COMPUTATION & TECHNOLOGY

We use generalized polynomial chaos (gPC) to construct a surrogate model of the response surface. By employing compressive sensing method, we are able to compute the gPC coefficients based on relatively few simulations. Then we apply Bayesian inference with the gPC surrogate model to search the suitable sets of parameters. In many cases, this method will help to achieve model reduction as the correlation between parameters can be revealed, hence parametric compression is possible. We present an example of using dissipative particle dynamics (DPD) model to study a polymer melt system. In this example, we set six parameters originally; and our method helps to determine four of them. The correlation between the remaining two parameters are also discovered. Thus, the number of parameters can be reduced.