Uncertainty quantification patterns for multiscale models

Area of Science:

• Computational Science
• Scientific Computing
• Modeling and Simulation

Background:

• Uncertainty quantification (UQ) is crucial for computational models in decision-making.
• Existing UQ methods may not adequately address complex, coupled multi-scale and multi-domain applications.

Purpose of the Study:

• To present novel uncertainty quantification patterns (UQPs) for UQ in coupled multi-scale and multi-domain applications.
• To provide reusable building blocks for developing tailored UQ procedures for multiscale models.

Main Methods:

• Developed UQPs as generic, customizable, and aggregatable templates.
• Implemented UQPs using the Multiscale Coupling Library and Environment 3.
• Derived potential speed-up factors for UQPs.

Main Results:

• Demonstrated the implementation and utility of UQPs.
• Presented two proof-of-concept examples of multiscale applications utilizing UQPs.
• Quantified potential performance improvements offered by UQPs.

Conclusions:

• UQPs offer a structured approach to UQ in complex multiscale applications.
• The UQPs are flexible and can be adapted for various multiscale modeling scenarios.
• This work contributes to reliability and reproducibility in computational science.