Abstract Details

Add files

Validation – Applying the Scientific Method in the Era of Simulations

Author: Martin Greenwald
Requested Type: Consider for Invited
Submitted: 2012-11-28 11:37:03

Co-authors:

Contact Info:
MIT
175 Albany St
Cambridge, MA   02139
USA

Abstract Text:
Over the past decade, dramatic progress has been made in the scope and power of plasma simulations. As a result, simulation has become an essential tool for our science. However, as codes embody imperfect models for physical reality, a necessary step towards developing a reliable predictive capability is the demonstration of agreement, without bias, between simulations and experimental results. While comparisons between computer calculations and experimental data are common, there is a compelling need to make these comparisons more systematic and more quantitative. This is nothing more (and nothing less) than an extension of the scientific method to research carried out via simulation. It demands that we carefully account for uncertainties in the experimental measurements, that we assess the quantitative impact of reducing physical equations to discrete mathematics and that we document the sensitivity of simulations to uncertainties in initial or boundary conditions. These assessments are often referred to collectively as “Uncertainty Quantification” or UQ. Tests of models are divided into two phases, usually called Verification and Validation (V&V). Verification is an essentially mathematical demonstration that a chosen physical model, rendered as a set of equations, has been accurately solved by a computer code. Validation is a physical process which attempts to ascertain the extent to which the model used by a code correctly represents reality within some domain of applicability, to some specified level of accuracy. Verification assesses errors from spatial or temporal gridding, algorithms, numerics and convergence, as well as coding errors and bugs in compilers. The arsenal of verification methodology includes formal convergence tests, theory to code comparisons, code to code comparisons, and specialized tools like the method of manufactured solutions. The process of code validation requires an assessment of the critical elements in a physical model and a quantitative approach for testing these elements individually and in combination. Dedicated experiments are typically required along with careful analysis of errors and sensitivity. Taken together, Verification, Validation and Uncertainty Quantification are essential confidence building activities for any science reliant on complex computer simulation. This talk will cover principles and practices for verification and validation including lessons learned from related fields and the opportunities for “exploratory scale” experiments to contribute to code validation.

Characterization: 2.0

Comments: