Materials Process Design and Control Laborator Materials Process Design and Control Laborator C C O O R R N N E E L L L L U N I V E R S I T Y Materials Process Design and Control Laboratory Sibley School of Mechanical and Aerospace Engineering 169 Frank H. T. Rhodes Hall Cornell University Ithaca, NY 14853-3801 Email: [email protected], [email protected]URL: http://mpdc.mae.cornell.edu/ NICHOLAS ZABARAS and B. VELAMUR ASOKAN A STOCHASTIC VARIATIONAL MULTISCALE METHOD WITH EXPLICIT SUBGRID MODELING FOR ADVECTION- DIFFUSION SYSTEMS
A STOCHASTIC VARIATIONAL MULTISCALE METHOD WITH EXPLICIT SUBGRID MODELING FOR ADVECTION-DIFFUSION SYSTEMS. NICHOLAS ZABARAS and B. VELAMUR ASOKAN. Materials Process Design and Control Laboratory Sibley School of Mechanical and Aerospace Engineering 169 Frank H. T. Rhodes Hall - PowerPoint PPT Presentation
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
Materials Process Design and Control LaboratorySibley School of Mechanical and Aerospace Engineering
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
MULTISCALE TRANSPORT SYSTEMS
Solidification Fluid-flow Diffusion in composites
Presence of various spatial and temporal length scales
Varied application areas – Engineering, Material science and other
Uncertainty manifests from impreciseness in boundary condition, material property specification and other modeling assumptions
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
IMPORTANCE OF UNCERTAINTY
Component
Meso
micro
Only statistical description of
material properties possible
Imprecise knowledge of
governing model
Imprecise boundary
conditions, initial
perturbations
Propagation and interaction of uncertainties have to be resolved
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
MODELING ASPECTS OF UNCERTAINTY
Probabilistic interpretation – Imprecise knowledge about boundary conditions, governing models, material properties described using stochastic processes
Uncertainty due to codes, algorithms, machine precision are not considered here
Physically, the uncertainty at progressively finer scales is higher [fluctuations]
Ideally, our computation paradigm should reflect above consideration
Computational approach should be orders of magnitude faster than other uncertainty analysis approaches that are sampling based
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
Uncertainty representation techniques
Introduction to spectral stochastic theory [Ghanem, Stochastic finite elements: A spectral approach]
Generalized polynomial chaos expansion [Karniadakis, J. Fluids Engrg., 125, 2001]
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
STOCHASTIC PROCESSES AS FUNCTIONS
A probability space is a triple comprising of collection of the sample space , the -algebra of subsets (events) of and the probability measure on .
FP
A real-valued random variable is a function that maps the probability space to a real line [regions in go to intervals in the real line]
F
X : Random variableX
: ( , , )X F P
A space-time stochastic process is can be represented as
: ( , , )W x t + other regularity conditions
F
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
SERIES REPRESENTATION For special kinds of stochastic processes that have finite variance-covariance function, we have mean-square convergent expansions
Series expansions
Known covariance function
Unknown covariance function
Best approximation in mean-square sense
Useful typically for input uncertainty modeling
Can yield exponentially convergent expansions
Used typically for output uncertainty modeling
Karhunen-Loeve Generalized polynomial chaos
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
SERIES REPRESENTATION [CONTD] Karhunen-Loeve
1
( , , ) ( , ) ( , ) ( )i ii
W x t W x t W x t
Stochastic process
Mean function
ON random variablesDeterministic functions
The deterministic functions are based on the eigen-values and eigenvectors of the covariance function of the stochastic process.
The orthonormal random variables depend on the kind of probability distribution attributed to the stochastic process.
Any function of the stochastic process (typically the solution of PDE system with as input) is of the form( , , )W x t
1( , , ) fn( , , , , )NW x t x t
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
SERIES REPRESENTATION [CONTD] Generalized polynomial chaos expansion is used to represent quantities like
0
( , , ) ( , ) (ξ( ))i ii
W x t W x t
Stochastic process
Askey polynomials in inputDeterministic functions
Stochastic input
1( , , ) fn( , , , , )NW x t x t
The Askey polynomials depend on the kind of joint PDF of the orthonormal random variables
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
NEED FOR SUPPORT-SPACE APPROACH
GPCE and Karhunen-Loeve are Fourier like expansions
Gibb’s effect in describing highly nonlinear, discontinuous uncertainty propagation
Onset of natural convection
[Zabaras JCP 208(1)] – Using support-space method
[Ghanem JCP 197(1)] – Using Wiener-Haar wavelets Finite element representation of stochastic processes [stochastic Galerkin method: Babuska et al]
Incorporation of importance based meshing concept for improving accuracy [support space method]
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
SUPPORT-SPACE [STOCHASTIC GALERKIN]
Let stochastic inputs be represented by ON random variables with a joint PDF
Support space is the region in the span of stochastic input that has a positive PDF
1( , , )N 1( , , )Nf
Example of a 2D input and associated support-space
PDF Grid
1( , , ) : ( ) 0NA f ξ ξ
2
2
1
ˆ
ˆ( ( ) ( )) ( )d
L
A
q
X X
X X f
Ch
ξ ξ ξ ξ
Piecewise polynomials defined on support-space grid
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
VMS – Basic idea
Algebraic subgrid modeling approaches
Illustration of the approach with derivation of subgrid problems for stochastic natural convection equation
Numerical examples
Stochastic advection-diffusion equation
Stochastic fluid-flow
Stochastic natural convection [GPCE approach and capturing unstable equilibrium using support-space method]
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
VMS – ILLUSTRATION [NATURAL CONVECTION]
g
2
0
( ) Pr( ) e
2Pr( ) ( )
1( ) [ ( ) ]
2T
v
vv v Ra
t
vt
pI v
v v v
Continuity
Momentum
Energy
Constitutive laws
hm D Dgm
gtht
.n h
gv v
g 0.n q
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
DEFINITION OF FUNCTION SPACES
1 2 2 22
22 1
( ) : ( ( ) )d ; ( ) : d
( ) : d ; ( ) : d
D D
T T
H D v v v x L D v v x
L T w w t L T w w t
Deterministic function spaces
Stochastic function spaces – The space of all second order random variables is critical to spectral uncertainty modeling
22 ( ) : dPL
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
DERIVED FUNCTION SPACES
Velocity function space
Test
Trial
Pressure function space
Test
Trial
Energy function space
Test
Trial
d12 2: ( ; ( ; ( ))) , g gmV v v L L T H D v v on
d10 2: ( ; ( )) , 0 gmV w w L H D w on
2 1 2: ( ; ( ; ( )))Q p p L L T L D
0 2 2: ( ; ( ))Q q q L L D
12 2: ( ; ( ; ( ))), g gtE L L T H D on
10 2: ( ; ( )), 0 gtE w w L H D w on
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
VARIATIONAL FORMULATION
v 0( , ) ( . , ) ( , ) ( , )htt w v w w q w
v g( , ) ( . , ) ( , ( )) ( , ) ( ( ) Pr( ) e , )
( , ) 0hmtv w v v w v h w Ra w
v q
Energy equation – Find such that, for all , the following holds
E 0w E
Momentum and continuity equation – Find such that, for all , the following holds
[ , ] [ , ]v p V Q0 0[ , ] [ , ]w q V Q
( )Ra ( )Pr Wherein, and are the random Rayleigh number and Prandtl number, respectively. These will be defined separately for each example considered.
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
VMS HYPTOTHESIS
VMS hypothesis: Exact solution can be written as a sum of coarse scale resolved components [bar quantities] and subgrid scale unresolved components [prime quantities]
Induced function space decomposition [Hughes 1995]: This induces a function space decomposition as follows
The coarse scale function spaces are to be approximated using finite element basis functions, the small scales are to be solved using Green’s functions, element Fourier transform and other
', ', 'v v v p p p
0 0 0
0 0 0 0 0 0
', ', '
', ', '
V V V V V V Q Q Q
Q Q Q E E E E E E
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
Stochastic VMS applied to the energy equation – The algebraic subgrid modeling approach
Scale decomposed variational formulation
Element Fourier transform [Codina, CMAME 191, 2002]
Algebraic subgrid scale model [Stochastic]
Modified coarse scale equation
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
ENERGY EQUATION – SCALE DECOMPOSITION
v 0( ', ) ( . . ', ) ( ', ) ( , )htt t w v v w w q w
v 0( ', ') ( . . ', ') ( ', ') ( , ')htt t w v v w w q w
Energy equation – Find and such that, for all and , the following holds
Coarse scale variational formulation
Subgrid scale variational formulation
These equations can be re-written in the strong form with assumption on regularity as follows
E E
0w E0w E
2 2' . ' ' ( . )t tv v R 2( ) : .L v
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
ASGS APPROACHES In VMS, the central idea is to solve the subgrid scale variational formulation in an approximate manner
Different techniques to generate an algebraic subgrid scale model [approximation] are
Green’s functions and Residual-free bubbles
Element Fourier transform
Two-level finite element methods
Spatial domain discretized
( )eD Spatial domain discretized into Nel disjoint finite element sub-domains
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
ELEMENT FOURIER TRANSFORM
( )eD( )
ˆ ( , ) exp( ) ( , )deD
k xg k i g x x
h
For a random field defined over a coarse element sub-domain, the element Fourier transform is defined as
The spatial derivative can now be represented as
( )
ˆ ˆexp( ) ( , )d ( , ) ( , )e
j jj
D
k kg k xn i g x i g k i g k
x h h h
Term is negligible for large wave numbers
Note – Subgrid solution denotes fluctuations and hence is captured with large wave number terms
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
ASGS MODEL FOR ENERGY EQUATION
2 2' . ' ' ( . )t tv v R
' '( )t n n nL R
2
' '2
1 1ˆ ˆˆn n n
kv ki R
t h h t
122 2
' '1 22
1 1 1,n t n n t
vR c c
t h t h
1 1
1( ), (1 )t n n n n n nf f f f f f
t Time integration
rule
Time-discretized equation
After application of Parseval’s and Mean value theorem
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
MODIFIED COARSE FORMULATION
2v v( ', ) ( ', ), ( ', ) ( ', )v w v w w w
We assume the following strong regularity conditions
Applying the ASGS model, we obtain the following modified coarse scale equation
Time integration choice plays a role in deciding the coarse scale formulation
v 0
2 2
1
' 2
1
( , ) ( , ) ( , ) ( , )
( , ( )
( , /( ) ) ) 0
htt n n n
Nel
t n n n te
Nel
n t te
w v w w q w
v w v w w
w t v w w w
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
VARIATIONAL MULTISCALE METHOD The nonlinear advection term is linearized using Picard assumption [accurate for moderate Reynolds numbers]
Similar to the energy equation, we can derive the subgrid scale strong form of equations as follows
'v v v v v v
g' ' ( ', ') ( ) Pr( ) e ( , )
'
t tv v v v p Ra v v v v p
v v
g( ) Pr( ) e ( , )mom t n n n nR Ra v v v v p
con nR v
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
ALGEBRAIC SUBGRID SCALE MODEL
2
' ' '2
'
1 1ˆˆ ˆ ˆPr( )
ˆ ˆ
n n mom n
ncon
k v k ki v i p R v
t h h h t
k vi R
h
After time discretization, application of element Fourier transform, we have
Application of mean value theorem and Parseval’s theorem1
22 222'
1 1
' 2'
1
, Pr( )
,
n c con c
nn m mom m
c
c v hhp R
c t c
v hv R
t c
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
MODIFIED COARSE FORMULATION
v v( ', ) ( ', ), (Pr( ) ( '), ( )) ( ', Pr( ) ( ))v v w v v w v w v w Assuming strong regularity conditions, we have
Modified coarse scale momentum equation
g1
'
1
'
( , ) ( , ) (2 Pr( ) ( ), ( )) ( , )
( ) Pr( ) e , Pr( ) ( )
( , ) , Pr( ) ( )
( , ) ( , ) 0
hmt n n n n
Nel
t n me
Neln
n n n me
n n c
v v v w p w v w h w
Ra v w v w w
vv v v p w v w w
t
v w v w
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
IMPLEMENTATION ISSUE - GPCE Assume that the input can be represented as a function of orthonormal random variables 1, , N
( )eDnbf
1
( , ) ( ) ( )f x f N x
Spatial random field
Random variablesGalerkin shape function
0
( ) (ξ( ))P
i ii
f f
The random variables are represented in a GPCE
GPCE coefficients
Askey chaos polynomials
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
IMPLEMENTATION ISSUES – SUPPORT-SPACE We use a two level grid approach
Again, we assume that the input can be represented as a function of orthonormal polynomials 1, , N
( )eDEach Gauss point has an underlying support-space grid
Finite element interpolation at spatial and support-space grid
Object oriented structure: every coarse element has nbf basis functions, support-space grid has nbf’ basis functions
nbf nbf ' nbf'
1 1 1
( , ) ( ) ( ) ( ) ( )i i i j j ii j i
f x f N x f N N x
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
CURSE OF DIMENSIONALITY Both GPCE and support-space method are fraught with the curse of dimensionality
As the number of random input orthonormal variables increase, computation time increases exponentially
Support-space grid is usually in a higher-dimensional manifold (if the number of inputs is > 3), we need special tensor product techniques for generation of the support-space
Parallel implementations are currently performed using PETSc (Parallel scientific extensible toolkit )
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
X and Y velocity obtained using a support-space method at Ra = 1870
Comparative X and Y velocity obtained using a deterministic simulation at Ra = 1870 (the upper limit)
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
Explicit subgrid modeling
Stochastic elliptic equation – VMS implementation
Numerical examples
EXPLICIT SUBGRID MODELING
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
MODEL MULTISCALE ELLIPTIC EQUATION
( )K u f
Permeability of Upper Ness formation
DDomain
Boundary
gu uin
on D
Multiple scale variations in K
K is inherently random [property predictions are at best statistical]
Crystal microstructures
Composites
Diffusion processes
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
VMS [VARIATIONAL FORMULATION]
u U[V] Find such that, for all v V
( , ) ( , ); ( , ) : ( , )a u v f v a u v K u v [V] denotes the full variational formulation
U and V denote appropriate function spaces for the multiscale solution u and test function v respectively
VMS hypothesis:
Induced function space decomposition [Hughes 1995]
C Fu u u
C FU U U
Exact = coarse + fine
C FV V V
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
VMS [COARSE AND SUBGRID SCALES]
C Fu u u ,C F C FU U U V V V
Using and the induced function space
decomposition
( , ) ( , ) ( , )C C F C Ca u v a u v f v
C Cu UFind such that, for all C Cv VF Fu Uand
and F Fv V
( , ) ( , ) ( , )C F F F Fa u v a u v f v
Coarse [V]
Subgrid [V]
Solve subgrid [V] using Greens' functions, PU and other
Substitute the subgrid solution in coarse [V]
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
DEFINITIONS AND DISCRETIZATION Assume a finite element discretization of the spatial region D into NelC coarse elements
Each coarse element is further discretized by a subgrid mesh with NelF elements
In each coarse element, the coarse solution uC can be approximated as
span ( ) ( ) : 1, , nbf ; 0, ,C r Cu N x r P
nbf = number of spatial finite element basis functions in each coarse element
PC = number of terms in the GPCE of coarse solution
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
DEFINITIONS AND DISCRETIZATION [CONTD]
( , )
( , )
x y
x y
Coarse meshSubgrid mesh for element EC
All basis function
problems solved here
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
SOLUTION OF HOMOGENEOUS [V] CONTD
max
0
( , ); ( , ) ( ) ( )
max nbf ( 1) 1, ( 1)( 1)
N
C CN N N rN
C C
u u x x N x
N P N P r
Considering the following finite element – GPCE representation for coarse solution uC
The subgrid solution can be represented as followsmax
0
ˆ ( , )N
F CN NN
u u x
Since represents subgrid variations, a nonlinear coarse scale mapping, a higher order GPCE is used [implies more terms in GPCE of ]
ˆFu
ˆFu
ˆFu
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
SOLUTION OF HOMOGENEOUS [V] CONTD Now, we have
Thus, we end up with (Nmax+1) homogeneous subgrid problems in each coarse element D(e)
Following representation is used for approximation
ˆ( , ) ( , ) 0
( , ) ( , ) 0C C
C C
E C F E F F
E N F E N F
a u v a u v
a v a v
span ( ) ( ) : 1, , nbf ; 0, ,N r FN x r P
nbf = number of spatial finite element basis functions in each element defined on the subgrid mesh PF = number of terms in the GPCE. Also, PF >PC
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
SOLUTION OF AFFINE [V]
Affine [V]
This affine correction is unique to the VMS formulation [Arbogast et al] and is not obtained in MsFEM type of formulations
Crucial in case of localized sources and sinks
Again, similar to the homogeneous [V], we have
0( , ) ( , )C CE F F F Ea u v f v
0 span ( ) ( ) : 1, , nbf ; 0, ,F r Fu N x r P This affine correction solution has no dependence on coarse scale behavior
We solve this equation on each coarse element with zero boundary conditions
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
SOLVING REDUCED PROBLEM
Subgrid mesh
Mapping element
edge
sn
Each coarse element edge is mapped to a line grid
Line grid yields coordinates (s – along the line grid, n – normal to line grid)
The reduced problem specified below is solved on the line grid
( ) ( ) 0, ( 1)( 1)
[ ](x ) ( ),
[ ](x ) 0, 1
s s N s s N C
N N r C
N N C F
K K N P r
r P
P r P
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
SOLVING REDUCED PROBLEM
( ) ( ) 0, ( 1)( 1)
[ ](x ) ( ),
[ ](x ) 0, 1
s s N s s N C
N N r C
N N C F
K K N P r
r P
P r P
0
0
1
0
0
0
0
0
0
0
0
0
0
0
0
0
x x
GPCE coefficients of coarse + subgrid BC
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
DESCRIPTION OF NUMERICAL PROBLEMS
1x
2x
3x 4x
Coarse element D(e)
Subgrid mesh
Based on boundary conditions used and subgrid problems solved, we have three studies
MsFEM-L MsFEM-Os VMS-Os Linear boundary conditions are used
No affine correction
Reduced solution as BC
No affine correction
Reduced solution as BC
Affine correction explicitly solved
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
FEM FOR HOMOGENEOUS [V]
( , ) ( , ) 0N F N Fa v a v Thus in each coarse element EC, we can solve for the subgrid basis functions as follows
[ ]{ } 0
[ ] ( ) ( ) ( ) ( )dxdC
r s
E
K
K K N x N x P
Note that we solve for the sum of coarse + subgrid basis functions
The boundary conditions for this equation are obtained as the solution of the reduced problem on coarse element edges
DOF for the problem = (Nno-subgrid)(PF+1)
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
DISTINCTIONS BETWEEN FUNCTION SPACES
{ ( ), 1, , nbf , 0, , }r FN r P
The number of functions spanning the subgrid approximation space is given by the tensor product
The subgrid basis functions however are a subset of these functions and hence do not contain all subgrid information
Is there a contradiction?
h
Subgrid scale
solution
Coarse scale
solution
Actual solution
Idea of VMS is only to solve the subgrid approximately
We just look in an intelligent manner, a subset of the fine function space wherein the solution is best represented
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
STOCHASTIC [CASE I] PSEUDO-PERIODIC MEDIA
0 1
0
11
1
1
, unif[ 1,1]
(2 sin(2 / )) (2 sin(2 / ))2
(2 cos(2 / )) (2 sin(2 / ))
(2 sin(2 / ))
(2 cos(2 / ))
1, 1.9, 0.08, 0.04, 0.1
K K K
P x yK
P y P x
P xK A
P y
f P A
uniformly distributed diffusion coefficient
K0
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
STOCHASTIC [CASE I] RESULTSFEM MsFEM-Os VMS-Os
U0
U1
U2
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
STOCHASTIC [CASE I] ERROR MEASURES
(coarse/fine) grid ratio
L-2
and
L-i
nfe
rro
rs
1 2 3 40
5E-05
0.0001
0.00015
0.0002
0.00025
0.0003
0.00035
MsFEM-OsVMS-OsMsFEM-OsVMS-Os
L-inf error was calculated on the mean value
Again, VMS is consistently better than MsFEM-Os.
2
1
22( vms ex) dxd
E[ vms]-E[ ex]
LD
L
E U U P
E U U
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
STOCHASTIC [CASE I] EFFECT OF PC TERMS We have assumed that the fine scale solution has more PC terms in its expansion
While reconstructing the fully resolved solution from the fine scale solution, we can only reconstruct up to the PCC terms. Beyond those terms, the fine scale solution is no longer a one-to-one map, hence, we see abnormalities (still equal in L-2)
L2 e
rror
Linf
err
or
PF-PC
L-2
erro
r
0 1 22.86E-04
2.86E-04
2.87E-04
2.87E-04
2.88E-04
PF-PC
L-2
erro
r
0 1 2
1.80E-05
1.84E-05
1.88E-05
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
Transient multiscale heat conduction
VMS implementation
Formulation 1: no subgrid time evolution
Formulation 2: quasi-static subgrid
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
MODEL MULTISCALE DIFFUSION EQUATION
( )u
K ut
Chip cooling and diffusion heat transfer
DDomain
Boundary
gu uin
on D
Multiple scale variations in K
K is inherently random [property predictions are at best statistical]
Diffusion, heat conduction in
microstructural lengths scale, composites
First step towards a multiscale industrial process simulator
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
VMS [VARIATIONAL FORMULATION]
u U[V] Find such that, for all v V
( , ) ( , ) 0; ( , ) : ( , )tu v a u v a u v K u v [V] denotes the full variational formulation
U and V denote appropriate function spaces for the multiscale solution u and test function v respectively
VMS hypothesis:
Induced function space decomposition [Hughes 1995]
C Fu u u
C FU U U
Exact = coarse + fine
C FV V V
t Ct Ftu u u
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
VMS [COARSE AND SUBGRID SCALES]
C Fu u u ,C F C FU U U V V V
Using and the induced function space
decomposition
( , ) ( , ) ( , ) ( , ) 0Ct C Ft C C C F Cu v u v a u v a u v
C Cu UFind such that, for all C Cv VF Fu Uand
and F Fv V
Coarse [V]
Subgrid [V]
Solve subgrid [V] using Greens' functions, PU and other
Substitute the subgrid solution in coarse [V]
( , ) ( , ) ( , ) ( , ) 0Ct F Ft F C F F Fu v u v a u v a u v
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
ASSUMPTIONS Formulation 1
The projection of time derivative ut on the subgrid function space is zero
This assumption assumes that the phenomena on the subgrid level attains equilibrium at every instant
Formulation 2
Quasistatic subgrid solution
This renders the term (uFt,vF) zero
This assumption is accurate provided the time step size used in the integration of governing equations is small
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
DEFINITIONS
0
( )Nbf
C Cu u x
Considering the following finite element representation for coarse solution uC
The subgrid solution uF can be represented as follows
Note that in case of quasistatic subgrid scale, the subgrid solution is a map of both coarse scale solution and coarse scale time derivative of solution
0
( )Nbf
UF Cu u x
0 0
( ) ( )Nbf Nbf
U UtF C Ctu u x u x
Formulation 1
Formulation 2
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
WHY THE CHOICE?
The alternative is to use backward Euler time discretization for the fine scale (after some algebra)
0 0
( ) ( )Nbf Nbf
U UtF C Ctu u x u x
C
prev( , ) ( , ) ( , ) ( , )C F C F F F Fu v ta u v ta u v u v The subgrid homogeneous equation now yields
ˆ( , ) ( , ) ( , ) 0
ˆ( , ) ( , ) ( , ) 0C F C F F F
C F C F F F
u v ta u v ta u v
u v t K u v t K u v
This is a tougher reaction-diffusion equation (notice that the diffusion coefficient is multiplied with a small time step size). In practical experiments, this equation yields spurious solutions
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
SOLUTION OF SUBGRID EQUATION For the assumptions employed for the subgrid model, we have
( , ) ( , ) 0UF Fa v a v
( , ) ( , ) 0
( , ) ( , ) 0
UF F
UtF F
a v a v
v a v
Formulation 1
Formulation 2
Wherein, the index a ranges from 1 to Nbf, the number of coarse basis functions in each coarse spatial sub-domain D(e)
Since the governing model is linear, the basis functions need to be estimated only in the first time step. Hence, we just solve a coarse problem for all remaining time steps!!
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
DETERMINISTIC [CASE I] PERIODIC MEDIA
2 2
(2 sin(2 / )) (2 sin(2 / ))
(2 cos(2 / )) (2 sin(2 / ))
1.9, 0.08
cos(4 ), 0.25(x,0)
0, 0.25
( 0.5) ( 0.5)
P x yK
P y P x
P
r ru
r
r x y
K
Fully resolved solution was obtained using a 128x128 mesh with bilinear quad elements
Time step size = 0.0005, duration of simulation = [0,0.05]
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
RESULTS[ANIMATIONS]
VMS solution FEM solution
L-inf error
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
RESULTS – FORMULATION 1 For the trivial assumption that subgrid attains equilibrium at every instant, we have the following solution error behavior
Error values are small in comparison to the solution maximum error values are L-inf errors that are 8% of solution
Nor
mal
ized
Un-
norm
aliz
ed
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
RESULTS – FORMULATION 2 For the trivial assumption that subgrid attains equilibrium at every instant, we have the following solution error behavior
Error values are consistently lesser than the formulation 1, wherein, the subgrid projection of coarse time derivative is neglected
Nor
mal
ized
Un-
norm
aliz
ed
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
Open issues in VMS
Difficulties in adapting to convection-diffusion
Nonlinearities
Adaptive VMS formulations
Non-quasistatic subgrid phenomena
Open issues in stochastic modeling
Compatibility of statistics across length scales
Transferring information from further lower scales
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
CONVECTION-DIFFUSION [DIFFICULTIES] Need to make transition from diffusion problems to convection-diffusion problems
Advection skew to a mesh shown
Information propagates along the characteristics
Most practical problems have extremely low diffusion. A further stabilized subgrid formulation may be required
Boundary conditions applied to subgrid problems have to be enhanced
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
BOUNDARY CONDITIONS
A typical coarse element
Edge mapped to a line grid
In diffusion type problems, a reduced problem is solved in the line grid
The reduced problem is the set of governing differential equations with derivatives normal to the element edge equated to zero
In convection-diffusion problems, the multidimensional layers that form in the solution should be captured by the basis function [not possible with current oscillatory BC approach]
Internal layers proportional to diffusion
coefficient
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
NONLINEARITY AND ADAPTIVITY Adaptive VMS [Larson, Chalmers preprint]
Nonlinear MsFEM [Hou]
Multiscale regions
Transition regions
Coarse regions
Calculation of basis functions
Very fine subgrid meshCoarser subgrid mesh
Inexact Newton methods for solving subgrid problems calculate coarse
scale solution by iterative methods
calculate error in solution, dual of solution to find regions of varying physics
Materials Process Design and Control LaboratoryMaterials Process Design and Control Laboratory
CCOORRNNEELLLL U N I V E R S I T Y
CCOORRNNEELLLL U N I V E R S I T Y
STATISTICS ACROSS LENGTH SCALES
Subgrid
Coarse
Lower scales
VM
S li
nkag
eLo
wer
ord
er
desc
ripto
rs
Simulations are usually discrete
Need to describe statistics and upscale statistical information
Governing equations not necessarily PDEs
Statistical information received
via MAXENT approach – that PDF that maximizes entropy
VMS then uses series
expansions to characterize uncertainty propagation across the coarse and subgrid scales