Home  >  News  >  News



School of Physics makes progress on Model Prediction

Translator: News Agency of BIT  Zhang Andi
Editor: News Center of BIT  Zhao Lin

     Recently, Vice Researcher Guo Wei of the Center of Quantum Control and Its Application of the School of Physics of Beijing Institute of Technology (BIT) has cooperated with three overseas scholars and co-developed a solution to the uncertainty in model prediction, by examining effects of correlated parameters. Overseas scholars includes Professor Vlachos and Doctor Sutton of the Innovation Center for Catalytic Energy of the University of Delaware, and Professor Katsoulakis of the School of Mathematics and Statistics of the University of Massachusetts. Related results have been published on the online issue of Nature Chemistry on February 22, a scientific journal published by Nature Publishing Group (Nature Chemistry8, 331-337, 2016).

     Using parameter-dependent mathematical models to predict and optimize object functions is a method that has been widely adopted in various fields of science and engineering. Applications can be seen in Transportation, Weather Forecasting, Material Science, Chemical Engineering, Biological Technology and so forth. An object function often depends on a parameter space. However, there is usually not an explicit mathematical expression to stimulate their relations. At present, although various models have been successfully adopted in phenomenon expressing and data fitting, those models that are capable of predicting complicated problems still face many limitations, such as selection of parameters of the model and the innate error or uncertainty of the model. Moreover, parameters are usually correlated to one another in reality. To study how much impact the uncertainty of parameters has on our object of interest has always been a tough mission in scientific research.

    Collaborating with his American co-workers, Vice Researcher Guo conducted quantitative analysis in the uncertainty of parameters and how much has the correlations among these parameters affected the accuracy of model predicting, by using kinetic model of ethanol steam reforming for hydrogen production on the surface of Pt/AI2O3. By calculating first principles, they took more than 10 types of exchange-correlation functionals into account, and identified the uncertainty and the correlative effects under different functionals. By kinetic modeling and global sensitivity analysis, they found that although there were multiple parameters in the model, only the relations among a few parameters had a decisive influence on the efficiency of hydrogen production. Therefore, enhanced precision of experiments or accuracy of those influential parameters can lead to a significant increase in the reliability and efficiency of material design. In this example, they also revealed the reason why the calculated hydrogen production speed was slower than measured speed. That was because the estimated activation energy on (111) surface was higher than its actual value, indicating that other active sites and reaction pathways played a more important role compared to commonly used (111) surface.

    These methods can be used to testify the reliability and sensitivity of model prediction. Combining these methods and multiscale method of calculating will improve the prediction ability of models and help reduce the gap between the complexity in actual experiments and the conciseness of theoretical models in material research.

    Link of Published Paper:


Release date:2016-05-24