Semi-intrusive Uncertainty Quantification for Multiscale models Anna - - PowerPoint PPT Presentation

semi intrusive uncertainty quantification for multiscale
SMART_READER_LITE
LIVE PREVIEW

Semi-intrusive Uncertainty Quantification for Multiscale models Anna - - PowerPoint PPT Presentation

Motivation Uncertainty Quantification Case Study Semi-intrusive Uncertainty Quantification for Multiscale models Anna Nikishova 1 Alfons Hoekstra 1 1 Computational Science Lab Universities of Amsterdam Quantification of Uncertainty: Improving


slide-1
SLIDE 1

Motivation Uncertainty Quantification Case Study

Semi-intrusive Uncertainty Quantification for Multiscale models

Anna Nikishova1 Alfons Hoekstra1

1Computational Science Lab

Universities of Amsterdam

Quantification of Uncertainty: Improving Efficiency and Technology, 2017

  • A. Nikishova, A. Hoekstra

Semi-intrusive UQ for Multiscale models

slide-2
SLIDE 2

Motivation Uncertainty Quantification Case Study Multiscale modeling

Multiscale models

Consider a PDE of the form ∂u(x, t, ξ) ∂t = L(u(x, t, ξ), ξ), where L is an operator acting in the space variable, and ξ denotes n-dimensional space of uncertain input. The analytical solution of this PDE satisfies u(x, t + ∆t, ξ) = e∆tLu(x, t, ξ).

  • A. Nikishova, A. Hoekstra

Semi-intrusive UQ for Multiscale models

slide-3
SLIDE 3

Motivation Uncertainty Quantification Case Study Multiscale modeling

Multiscale models

Consider a PDE of the form ∂u(x, t, ξ) ∂t = L(u(x, t, ξ), ξ), where L is an operator acting in the space variable, and ξ denotes n-dimensional space of uncertain input. The analytical solution of this PDE satisfies u(x, t + ∆t, ξ) = e∆tLu(x, t, ξ).

  • A. Nikishova, A. Hoekstra

Semi-intrusive UQ for Multiscale models

slide-4
SLIDE 4

Motivation Uncertainty Quantification Case Study Multiscale modeling

Multiscale models

Let us assume for L a two-term splitting: L = Lµ + LM, where Lµ and LM are subscale models with micro and macro time scale, respectively. Thus, the equation can be rewritten u(x, t + ∆t, ξ) ≈ e∆tLMe∆tLµu(x, t, ξ)

  • A. Nikishova, A. Hoekstra

Semi-intrusive UQ for Multiscale models

slide-5
SLIDE 5

Motivation Uncertainty Quantification Case Study Multiscale modeling

Multiscale models

The original PDE can be approximated as a sequence of the following two sub-systems ∂u∗(x, t, ξ) ∂t = Lµu∗(x, t, ξ), for tn < t < tn+1, with u∗(x, tn, ξ) ≈ u(x, tn, ξ) ∂u∗∗(x, t, ξ) ∂t = LMu∗∗(x, t, ξ), for tn < t < tn+1, with u∗∗(x, tn, ξ) = u∗(x, tn+1, ξ)

  • A. Nikishova, A. Hoekstra

Semi-intrusive UQ for Multiscale models

slide-6
SLIDE 6

Motivation Uncertainty Quantification Case Study Multiscale modeling

Submodel Execution Loop

In general, a model with two or more different time scales can be illustrated by a Submodel Execution Loop [BFL+13, CFHB11]. uinit are some initial conditions for a sub-scale model, O is the

  • bservation of the current state, S is the solver, and B is the

application of boundary conditions.

  • A. Nikishova, A. Hoekstra

Semi-intrusive UQ for Multiscale models

slide-7
SLIDE 7

Motivation Uncertainty Quantification Case Study

Black Box Monte Carlo

  • A. Nikishova, A. Hoekstra

Semi-intrusive UQ for Multiscale models

slide-8
SLIDE 8

Motivation Uncertainty Quantification Case Study

Semi-intrusive Monte Carlo

  • A. Nikishova, A. Hoekstra

Semi-intrusive UQ for Multiscale models

slide-9
SLIDE 9

Motivation Uncertainty Quantification Case Study

The Gray-Scott reaction diffusion model

∂u(t, x, y, ξ) ∂t = Du(ξ1)∇2u − uv2

  • Macro scale model

+ F(ξ2)(1 − u)

  • Micro scale model

∂v(t, x, y, ξ) ∂t =

  • Dv(ξ3)∇2v + uv2 −
  • (F(ξ2) + K(ξ4))v

where the model reaction and diffusion coefficients contain 10% uncertainty with mean values E(Du(ξ1)) = 2 · 10−5, E(F(ξ2)) = 0.025, E(Dv(ξ3)) = 1 · 10−5, E(K(ξ4)) = 0.053. The model reproduces a compex pattern formation with a transition map studied in [HsQHS16].

  • A. Nikishova, A. Hoekstra

Semi-intrusive UQ for Multiscale models

slide-10
SLIDE 10

Motivation Uncertainty Quantification Case Study

Comparisomn of the performance of the UQ methods

  • A. Nikishova, A. Hoekstra

Semi-intrusive UQ for Multiscale models

slide-11
SLIDE 11

Motivation Uncertainty Quantification Case Study

References I

Joris Borgdorff, Jean-Luc Falcone, Eric Lorenz, Carles Bona-Casas, Bastien Chopard, and Alfons G. Hoekstra, Foundations of distributed multiscale computing: Formalization, specification, and analysis, Journal of Parallel and Distributed Computing 73 (2013), no. 4, 465–483. Bastien Chopard, Jean-Luc Falcone, Alfons G. Hoekstra, and Joris Borgdorff, A framework for multiscale and multiscience modeling and numerical simulations, Lecture Notes in Computer Science, Springer Berlin Heidelberg, 2011, pp. 2–8. Omri Har-shemesh, Rick Quax, Alfons G Hoekstra, and Peter M A Sloot, Information geometric analysis of phase transitions in complex patterns: the case of the Gray-Scott reaction–diffusion model, Journal of Statistical Mechanics: Theory and Experiment (2016), 43301.

  • A. Nikishova, A. Hoekstra

Semi-intrusive UQ for Multiscale models