Sei sulla pagina 1di 19

The mortar method with approximate integration:

coupling finite elements on non-matching


unstructured grids in 3D
Silvia Bertoluzza
silvia.bertoluzza@imati.cnr.it

Istituto di Matematica e Tecnologie Informatiche del C.N.R. (Pavia, Italy)

Collaborations with
Silvia Falletta (Milan)
Francesca Rapetti (Nice)

The mortar method with approximate integration: coupling finite elements on non-matching unstructured grids in 3D p.1/19



 



split

domain

Domain decomposition

(disjoint) subdomains

preconditioning

reduce solution of a problem in to solution of set of


smaller problems in the subdomains

parallelism

The mortar method with approximate integration: coupling finite elements on non-matching unstructured grids in 3D p.2/19

Conforming DomDec

same discretization type in each SD

continuity of the discrete solution across the interface


matching FE decomposition

Non Conforming DomDec


no strong continuity required
different spaces in different subdomains allowed
non matching FE grids allowed

The mortar method with approximate integration: coupling finite elements on non-matching unstructured grids in 3D p.3/19

Mortar Method

and





+ Dirichlet B.C.







discretizations of






on





and

to fix ideas:

weak continuity across interface: jump orthogonal to


suitable multiplier space

resp.

multiplier space
trace on slave edge of the
discretization space in the subdomain
(function on the slave edge determined from function on master edge via the
constraint)

The mortar method with approximate integration: coupling finite elements on non-matching unstructured grids in 3D p.4/19

Mortar method

MASTER SIDE

SLAVE SIDE
MULTIPLIER


 





 

constraint:



"

&

,

&

' %

&
)

&
('%

+*#

$
#

problem:

&

constrained space
set of discrete functions
satisfying weak continuity constraint (n.b.

The mortar method with approximate integration: coupling finite elements on non-matching unstructured grids in 3D p.5/19

Computation of the jump constraint


implementation requires computing the constraint

spaces of different type

efficient computation particularly difficult in 3D


problems!

multiplier = suitable dual wavelet space


/

discretization = wavelet

example: wavelet/FEM coupling:


problems in coupling with finite elements

Problem computing integrals of products of functions living on non-matching grids

The mortar method with approximate integration: coupling finite elements on non-matching unstructured grids in 3D p.6/19

Coupling non-matching grids


Problem computing integral of the product of
two functions living on unstructured non matching
grids

standard quadrature is no good!


intersecting triangles very technical
quadrature on master grid: bad for approximation
quadrature on slave grid: bad for consistency

The mortar method with approximate integration: coupling finite elements on non-matching unstructured grids in 3D p.7/19

Approximate Integration

Problem: compute

34
1 0
2

' 

' 

1



1

:
1 0

5 
6

1 0
2


'

1 7


orthogonal projection onto

84
56

1 7

introduce auxiliary space

Idea

The mortar method with approximate integration: coupling finite elements on non-matching unstructured grids in 3D p.8/19





 
1 7







must be easier to compute than

the integrals

must be easy to compute

<=;
?>
@
A $
B

the projector

1 7

the integrals

 '
1 

must be chosen in such a way that three things happen

three fields formulation)

1 0

must be (easily) computable

the mass matrix must be easily invertible

error estimates must be available

The mortar method with approximate integration: coupling finite elements on non-matching unstructured grids in 3D p.9/19

The modified mortar method



&
!

&

&


&

C*#

' %

&

'

s.t. for all

&
('%

&

&

: find

Problem

&

1 7




&

new non conforming approximation space:


set of
discrete functions satisfying the approximate weak
continuity constraint

The mortar method with approximate integration: coupling finite elements on non-matching unstructured grids in 3D p.10/19

Error Estimate (two subdomains)


34

L



J
I
?

M 34
N

? K

'

GFE

M 34
N

? K

'

34
O
P

L

GFE

1 0
D

Assumptions: direct inequality for




norm)
usual mortar error
L
M

P

M  L

&

&

: broken


&
M

Q L
R

then: (

+ usual mortar assumptions

The mortar method with approximate integration: coupling finite elements on non-matching unstructured grids in 3D p.11/19

FE on tensor product grid of mesh-size

=Q

usual mortar error


34
L
V

U

s.t.
W

...

error balancing: choose

M 34
N

M  L

&

&

smooth

error estimate:

choice:

1 0

Coupling non-matching FE

...

The mortar method with approximate integration: coupling finite elements on non-matching unstructured grids in 3D p.12/19

Choosing the auxiliary mesh

]
X

FE mesh-size
X

O
T

a
$

`_^

[Z

P1 FE mortar discretization:

the auxiliary grid can be coarse

The mortar method with approximate integration: coupling finite elements on non-matching unstructured grids in 3D p.13/19

g
ij

e d
f

c
b

function on a tensor product uniform grid

function on the unstructured grid

l

:Q




1 7


Computation of

l

intersection of triangles in
with tensor product
rectangles computed directly
(m+1)

X

each triangle
sees at most rectangles

X

most of the triangles fall inside a rectangle

m
n

(n+1)

The mortar method with approximate integration: coupling finite elements on non-matching unstructured grids in 3D p.14/19

Distribution of triangles
good triangles:
Grid

Q1

Q2

Q3

5x5

20 %

60 %

60 %

10 x 10

20 %

60 %

80 %

20 x 20

40 %

70 %

80 %

40 x 40

45 %

80 %

90 %

triangles interacting with an edge or with a X-point


# El.

Q1

Q2

Q3

# El.

Q1

Q2

Q3

50

40

20

20

50

200

160

80

40

200

25

800

560

240

160

800

49

3200

1760

640

320

3200

121

16

The mortar method with approximate integration: coupling finite elements on non-matching unstructured grids in 3D p.15/19

e d
f
h

 ' 
1

q
@
An$
sr

? p

< ;
o

1 7


<

computing

 ' 
1



: the same arguments apply as for computing

1 7

@
An$
?>

definition of

Computation of


v

M u
tu

tv

tu

nodal basis for

1 0
D


 

inverting the mass-matrix

inverting
best case)

yxw

: tensor product structure


inverting two 1D banded mass matrices (tridiagonal, in the

The mortar method with approximate integration: coupling finite elements on non-matching unstructured grids in 3D p.16/19

Coding the Mortar Method


Requirements for a code: extendibility
Extendibility of a Mortar code: aim at a code s.t.
adding new discretizations is easy
minimal changes needed if the new discretization is
already implemented in the monodomain framework
Difficulty: need to be familiar with all discretizations
already implemented

The mortar method with approximate integration: coupling finite elements on non-matching unstructured grids in 3D p.17/19

Adding a new discretization

, ...,

4 z

{|

require ?

integrals of functions in
multipliers

in the monodomain framework


{|

{|

Case A: Mortar code with discretizations

What does adding a new discretization

times the correspondig

times the multipliers for

times the


z


4 z


z

, ...,

4 z

integrals of functions in
, ...,

{|

integrals of functions in
multipliers

mortar projection

The mortar method with approximate integration: coupling finite elements on non-matching unstructured grids in 3D p.18/19

1 0

and

in the monodomain framework


{|

integrals of functions in
multipliers

{|

1 0

4 z

Case B: Mortar with approx. integration code with


and auxiliary space
discretizations , , . . . ,
.

times the correspondig

{|

times the functions in

1 0
D

integrals of functions in

times the multipliers

integrals of functions in

1 0
}

mortar projection

The code is more easily extended !!

The mortar method with approximate integration: coupling finite elements on non-matching unstructured grids in 3D p.19/19

Potrebbero piacerti anche