Description
Automated uncertainty propagation of complex computer codes suffers severe complications arising from multiple occurrences of the uncertain variables. This is notoriously referred to as dependence problem. Several methods have been proposed in the literature to tackle this problem, which include constraint propagation, subinterval reconstitution, Taylor models, and more. In this talk, I am going to present a propagation method based on the interval gradient flow, which enables to solve a certain class of problems. I am going to show how this method can be used to find the rigorous best-possible bounds on a single degree-of-freedom dynamical system. Certainty implies subjectivity thus can’t be inferred from empirical data, but can be propagated through the calculations with mathematical rigour. Provided the units, very often the question is not whether the bounds on the physical quantity exist, but rather what are the bounds in which all of the possible values lie for the specific problem under study. Verified propagation with rigorous validated numerics leads to the logic of modus tollens, where models can be discarded if just a single datum falls outside the obtained region of certainty. The opposite generally does not hold. Discarding calibrated models does not necessarily mean bad news, as new models can be investigated and new physics discovered.| Period | 5 Feb 2020 |
|---|---|
| Held at | Institute for Advanced Computational Science - Stoney Brook University |
| Degree of Recognition | International |