Scribd
Upload
118 views19 pages

CA Cme334 Ch2

This document describes a course on model reduction for parameterized partial differential equations (PDEs). It discusses how PDEs can be discretized and parameterized, leading to high-dimensional dynamical systems. Model reduction is necessary for problems involving multiple queries over various parameter values, such as uncertainty quantification and design optimization. The document proposes approximating parameterized solutions by linear combinations of precomputed "snapshot" solutions, restricting the solutions to a low-dimensional subspace.

Uploaded by

Gabriele Boncoraglio
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
118 views19 pages

CA Cme334 Ch2

This document describes a course on model reduction for parameterized partial differential equations (PDEs). It discusses how PDEs can be discretized and parameterized, leading to high-dimensional dynamical systems. Model reduction is necessary for problems involving multiple queries over various parameter values, such as uncertainty quantification and design optimization. The document proposes approximating parameterized solutions by linear combinations of precomputed "snapshot" solutions, restricting the solutions to a low-dimensional subspace.

Uploaded by

Gabriele Boncoraglio
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 19

CME 345: MODEL REDUCTION - Parameterized PDEs

CME 345: MODEL REDUCTION


Pameterized Partial Differential Equations

David Amsallem & Charbel Farhat


Stanford University
cfarhat@stanford.edu

David Amsallem & Charbel Farhat Stanford University cfarhat@stanford.edu

CME 345: MODEL REDUCTION - Parameterized PDEs

CME 345: MODEL REDUCTION - Parameterized PDEs

Outline

1 Initial Boundary Value Problems


2 Parameters of Interest
3 Semi-discretization Processes and Dynamical Systems
4 The Case for Model Reduction
5 Subspace Approximation

David Amsallem & Charbel Farhat Stanford University cfarhat@stanford.edu

CME 345: MODEL REDUCTION - Parameterized PDEs

CME 345: MODEL REDUCTION - Parameterized PDEs


Initial Boundary Value Problems

Partial differential equation (PDE)


L(W, x, t) = 0
W = W(x, t) Rq : state variable
x Rd , d 3: space variable
t 0: time variable

Examples:
Navier-Stokes equations
Hogkin-Huxley equations
Schrodinger equations

Associated boundary conditions


B(W, xBC , t) = 0
Dirichlet BC
Neumann BC

Initial condition
W0 (x) = WIC (x)
David Amsallem & Charbel Farhat Stanford University cfarhat@stanford.edu

CME 345: MODEL REDUCTION - Parameterized PDEs

CME 345: MODEL REDUCTION - Parameterized PDEs


Initial Boundary Value Problems
Parameterized PDE

Parametrized partial differential equation (PDE)


L(W, x, t; ) = 0
Associated boundary conditions
B(W, xBC , t; ) = 0
Initial condition
W0 (x) = WIC (x, )
D Rp : parameter vector

David Amsallem & Charbel Farhat Stanford University cfarhat@stanford.edu

CME 345: MODEL REDUCTION - Parameterized PDEs

CME 345: MODEL REDUCTION - Parameterized PDEs


Parameters of Interest

Material parameters
Shape parameters

David Amsallem & Charbel Farhat Stanford University cfarhat@stanford.edu

CME 345: MODEL REDUCTION - Parameterized PDEs

CME 345: MODEL REDUCTION - Parameterized PDEs


Parameters of Interest

Initial condition
Boundary conditions

David Amsallem & Charbel Farhat Stanford University cfarhat@stanford.edu

CME 345: MODEL REDUCTION - Parameterized PDEs

CME 345: MODEL REDUCTION - Parameterized PDEs


Semi-discretization Processes and Dynamical Systems
Semi-discretized problem

The PDE is discretized in space by one of the following methods


Finite Differences approximation
Finite Element method
Finite Volumes method
Discontinuous Galerkin method
Spectral methods....

This leads to a system of N = q Nspace ordinary differential


equations (ODEs)
dw
= f(w, t; )
dt
in terms of the discretized state variable
w = w(t; ) RN
with initial condition w(0; ) = w0 ()
This is the high-dimensional model (HDM)
David Amsallem & Charbel Farhat Stanford University cfarhat@stanford.edu

CME 345: MODEL REDUCTION - Parameterized PDEs

CME 345: MODEL REDUCTION - Parameterized PDEs


The Case for Model Reduction

Multi-query context

Routine analysis
Uncertainty quantification
Design optimization
Inverse problems
Optimal control
Model predictive control

David Amsallem & Charbel Farhat Stanford University cfarhat@stanford.edu

CME 345: MODEL REDUCTION - Parameterized PDEs

CME 345: MODEL REDUCTION - Parameterized PDEs


The Case for Model Reduction
Multi-query Context

Routine analysis

David Amsallem & Charbel Farhat Stanford University cfarhat@stanford.edu

CME 345: MODEL REDUCTION - Parameterized PDEs

CME 345: MODEL REDUCTION - Parameterized PDEs


The Case for Model Reduction
Multi-query Context

Routine analysis

David Amsallem & Charbel Farhat Stanford University cfarhat@stanford.edu

CME 345: MODEL REDUCTION - Parameterized PDEs

CME 345: MODEL REDUCTION - Parameterized PDEs


The Case for Model Reduction
Multi-query Context

Uncertainty quantification

Monte-Carlo simulations,...
David Amsallem & Charbel Farhat Stanford University cfarhat@stanford.edu

CME 345: MODEL REDUCTION - Parameterized PDEs

CME 345: MODEL REDUCTION - Parameterized PDEs


The Case for Model Reduction
Multi-query Context

Design optimization

David Amsallem & Charbel Farhat Stanford University cfarhat@stanford.edu

CME 345: MODEL REDUCTION - Parameterized PDEs

CME 345: MODEL REDUCTION - Parameterized PDEs


The Case for Model Reduction
Multi-query Context

Model predictive control

David Amsallem & Charbel Farhat Stanford University cfarhat@stanford.edu

CME 345: MODEL REDUCTION - Parameterized PDEs

CME 345: MODEL REDUCTION - Parameterized PDEs


The Case for Model Reduction
Multi-query Context

Model predictive control

David Amsallem & Charbel Farhat Stanford University cfarhat@stanford.edu

CME 345: MODEL REDUCTION - Parameterized PDEs

CME 345: MODEL REDUCTION - Parameterized PDEs


The Case for Model Reduction
Model Parameterized PDE

Advection-diffusion-reaction equation: W = W(x, t; ) solution of


W
+ U W W = fR (W, t, R ) for x
t
with appropriate boundary and initial conditions
W(x, t; ) = WD (x, t; D ) for x D
W(x, t; ) n(x) = 0 for x N
W(x, 0; ) = W0 (x; IC ) for x
Parameters of interest
= [U1 , , Ud , , R , D , IC ]

David Amsallem & Charbel Farhat Stanford University cfarhat@stanford.edu

CME 345: MODEL REDUCTION - Parameterized PDEs

CME 345: MODEL REDUCTION - Parameterized PDEs


The Case for Model Reduction
Parameterized Solutions

Two dimensional advection-diffusion equation


W
+ U W W = 0 for x
t
with boundary and initial conditions
W(x, t; ) = WD (x, t; D ) for x D
W(x, t; ) n(x) = 0 for x N
W(x, 0; ) = W0 (x) for x

N
D

N
David Amsallem & Charbel Farhat Stanford University cfarhat@stanford.edu

CME 345: MODEL REDUCTION - Parameterized PDEs

CME 345: MODEL REDUCTION - Parameterized PDEs


The Case for Model Reduction
Parameterized Solutions

Two dimensional advection-diffusion equation


W
+ U W W = 0 for x
t
with boundary and initial conditions
W(x, t; ) = WD (x, t; D ) for x D
W(x, t; ) n(x) = 0 for x N
W(x, 0; ) = W0 (x) for x
p = 4 parameters of interest
= [U1 , U2 , , D ] R4
w RN with N = 2, 701
David Amsallem & Charbel Farhat Stanford University cfarhat@stanford.edu

CME 345: MODEL REDUCTION - Parameterized PDEs

CME 345: MODEL REDUCTION - Parameterized PDEs


The Case for Model Reduction
Parameterized Solutions

David Amsallem & Charbel Farhat Stanford University cfarhat@stanford.edu

CME 345: MODEL REDUCTION - Parameterized PDEs

CME 345: MODEL REDUCTION - Parameterized PDEs


Subspace Approximation

Can we reuse solutions for different parameter values?


Idea: use linear combinations of pre-computed solutions (snapshots)
w(t; ) q1 (t; )w(t1 ; 1 ) + + qk (t; )w(tk ; k )
w(ti ; i ) RN : pre-computed solution at (ti , i )
qi (t; ) R: coefficient in the expansion

The parameterized solution w(t; ) is then constrained to belonging


to the subspace
S = span {w(t1 ; 1 ), , w(tk ; k )}
Subspace dimension
dim (S) = rank [w(t1 ; 1 ), , w(tk ; k )] k

David Amsallem & Charbel Farhat Stanford University cfarhat@stanford.edu

CME 345: MODEL REDUCTION - Parameterized PDEs

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy