Optimal and Unstructured High-Order Non-Intrusive Approximations for Uncertain Parameterized Simulations

Abstract

The approximation and prediction of output quantities of interest in large-scale simulation software is an ongoing challenge in scientific computing. This difficulty is compounded when the simulation software contains numerous tunable input parameters that specify modeling scenarios, geometry, and uncertainty. The main goal of this project is robust and efficient prediction of the variability of quantities of interest with respect to these input parameters. This is primarily accomplished via non-intrusive sampling of models. Straightforward and naive sampling methods often (usually) yield suboptimal performance and convergence guarantees. This project aims to develop novel, modern sampling strategies that perform well and are provably convergent, ideally without dependence on dimension. Nearing the end of this project, the efficacy of the developed procedures will be tested on realistic parameterized scientific problems.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Jul 11, 2019
Accession Number
AD1096451

Entities

People

  • Akil C. Narayan

Organizations

  • University of Utah

Tags

Communities of Interest

  • C4I
  • Energy and Power Technologies
  • Human Systems

DTIC Thesaurus Topics

  • Algorithms
  • Chebyshev Polynomials
  • Chemical Reactions
  • Compressed Sensing
  • Differential Equations
  • Equations
  • Gaussian Quadrature
  • Governments
  • High Resolution
  • Machine Learning
  • Neural Networks
  • Numerical Analysis
  • Probability
  • Random Variables
  • Reliability
  • Topology Optimization
  • Two Dimensional

Readers

  • Computational Fluid Dynamics (CFD)
  • Computational Modeling and Simulation
  • Neural Network Machine Learning.