Professional Documents
Culture Documents
Eccomas Silvia
Eccomas Silvia
methods
Silvia Bertoluzza
Istituto di Matematica Applicata e Tecnologie Informatiche del CNR, Pavia
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
1 / 44
Motivation
simulation of a mixture of fluid and (rigid) particles
problem arising in a variety of applications
blood flux (particles = eritrocytes)
advection of polluting agents
liquid-gas mixture in reactors (particles = liquid drops)
ECCOMAS, 2016
2 / 44
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
3 / 44
Model problem
Problem: find u H 1 () s.t.
u = f in
u=0
on
Rn : domain with holes
= \ B : (B holes)
= =
: boundary of the box
: boundaries of the
holes
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
4 / 44
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
5 / 44
Fictitious Domain
Idea: consider an equivalent problem on a bigger simple domain = B
solve u = f in
B.C. on
Fast Poisson Solvers
Fast preconditioners (multigrid)
Problems:
extension of the r.h.s. f f
imposing B.C. on
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
6 / 44
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
7 / 44
Observation
u 00 = 1 in (0, a) (b, 1) with homogeneous Dirichlet B.C.
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
8 / 44
h Vh
Assumption on u: u H k+1 ()
No assumption on smoothness of u in
ku uh k1, . hk kukk+1,
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
9 / 44
Several approaches
Optimal control approach [Atamian, Joly 91]
Build compatible extension via wavelets [Mommer 06]
Finite Cell method [Duster, Parvizian, Rank 07]
Cut elements [Burmann, Hansbo 09 ]
Fat boundary method [Maury 01 ]
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
10 / 44
[Maury, 01]
ECCOMAS, 2016
11 / 44
[Maury, 01]
ECCOMAS, 2016
12 / 44
Global problem
u = f + u
,
in
Local problems
u = f ,
S. Bertoluzza (IMATI-CNR)
in ,
u = 0 on ,
u = u on
ECCOMAS, 2016
13 / 44
Schwartz-type iteration
u n un sol. in of
un = f ,
un = u n on
un = 0 on
un U n sol. in of
U n = f + un
U n = 0 on
relaxation: u n+1 = u n + (1 )U n
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
14 / 44
Theoretical result
Convergence of the Schwartz procedure in the continuous [Maury 01] and
in the discrete case (under mild assumptions on the discretization)
[S.B., Ismail, Maury 11]
ECCOMAS, 2016
15 / 44
Z
Z
Z
Z
u
v
v
v=
f v,
v H 1 ()
Z
Z
u
= 0, H 1/2 (),
(u u)
= 0, H 1/2 ()
[Maury 01]
&
ECCOMAS, 2016
16 / 44
Error estimate
Global (non-optimal) estimate: u H s (), s k + 1 (k order of FE
discretization)
ku = uh k1, + ku uk
1, . hs1 kuks,
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
17 / 44
0,01
0,0001
1e-06
0,01
0,1
h (log scale)
Skip Stokes
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
18 / 44
Stokes equation
find u (H 1 ())d , p L20 () s.t.
u + p = f in
u =0
in
u=0
on
Rn : domain with holes
= \ B : (B holes)
ECCOMAS, 2016
19 / 44
in
in
Z
p = 0,
u = 0 on
in
in
u = u on
p = 0
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
20 / 44
Z
Z
p v
v =
f v,
v (H01 ())d
Z
q u = 0,
q L20 ()
Z
Z
Z
Z
Z
u
v p v
v
v=
f v,
v (H 1 ())d
Z
Z
q u pq = 0,
q L2 ()
Z
Z
1/2
d
u
= 0, (H
())
(u u)
= 0, (H 1/2 ())d .
u v
S. Bertoluzza (IMATI-CNR)
p = 0})
ECCOMAS, 2016
21 / 44
Remedy 1: use different mesh sizes for velocity and pressure on the
local problem
Remedy 2: propose a stabilized formulation
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
22 / 44
Stable discretization
Global problem: stable Stokes elements, meshsize h
Local problem
velocity: meshsize h, h
pressure: meshsize ,
( = h)
ECCOMAS, 2016
23 / 44
a ax
e sin(2y )),
2
1
p = (1 e 2ax )
2
Pk Pk1 elements
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
24 / 44
Errors
ECCOMAS, 2016
25 / 44
Test 2
2D rectangular domain with 70 perforations
inflow on left boundary with parabolic profile
no slip condition otherwise
velocity
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
26 / 44
Test 2
pressure
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
27 / 44
Test 2
streamline
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
28 / 44
Test 2
Similar test in 3D - 10 perforations
velocity
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
29 / 44
Test 2
Similar test in 3D - 10 perforations
streamline
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
30 / 44
Impact of quadrature
Several integrals to compute
Find uh Vh , uh Vh , h h , h h such that
Z
Z
Z
uh vh
h vh =
fvh , vh Vh
Z
Z
Z
Z
uh
vh
h vh
h vh =
f vh ,
vh Vh
Z
Z
Z
uh h = 0, h h ,
uh h
uh h = 0, h h
ECCOMAS, 2016
31 / 44
fvh
Z
Z
Z
Z
uh
vh h vh h vh =
f vh ,
vh Vh
Z
Z
uh h = 0, h h ,
(uh uh )h = 0, h h
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
32 / 44
Integrals on and
Mesh for u, meshsize h
Mesh for u,
meshsize h
Mesh for and inherited by the mesh for u
Finite elements of order k for all functions
Exact computation of the integral: optimal error estimate
Quadrature: which mesh?
Two meshes on and
the mesh of (readily available)
the mesh induced by the mesh on
(the intersection of the two meshes)
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
33 / 44
Quadrature mesh TR
Quadrature formula on reference element approximated element
integral:
Z
g (
x ) d x
K
L
X
Z
i g (
xi ) = I
g (x) IK (g )
K
i=1
S. Bertoluzza (IMATI-CNR)
K TR
g (g ) :=
IK (g )
K TR
ECCOMAS, 2016
34 / 44
u v (v ) =
fv , v H01 ()
Z
Z
Z
Z
u
v
v
v=
f v,
v H 1 ()
Z
Z
u
= 0, H 1/2 (),
(uh ) u
= 0, H 1/2 ()
Remark: all other integrals are calculated exactly by using the quadrature
on the corresponding mesh
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
35 / 44
Which mesh?
Option 1: TR = trace on curve of mesh for
Th fitted to and
Mesh is regular
Simplest option, best for implementation
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
36 / 44
1.5
0.8
0.6
1
0.4
0.5
0.2
0
0
-0.2
-0.5
-0.4
-1
-0.6
-1.5
1
-0.8
0.5
0.9
0.8
0.7
0.6
0.8
0.5
0.6
0.4
0.3
0.2
0.2
0.1
0
S. Bertoluzza (IMATI-CNR)
0.45
1
0.4
0.8
0.6
0.35
0.4
0.4
0.3
0.2
0.25
ECCOMAS, 2016
37 / 44
Local error
Nx
10
20
40
80
Mx
12
24
48
96
S. Bertoluzza (IMATI-CNR)
0 -0
.080844
.022911
.0045992
.0011388
0 -1
.079063
.017887
.0046056
.0011408
2 -1
.080792
.018391
.0047397
.0011738
1 -1
.075926
.019176
.007883
.0045428
ECCOMAS, 2016
38 / 44
Local error
Nx
5
10
20
40
Mx
6
12
24
48
0 -0
0.00060259
.00029501
2.0513e-5
6.0416e-7
S. Bertoluzza (IMATI-CNR)
0 -1
0.00060258
.00029501
2.0513e-5
6.0416e-7
2 -1
0.00060706
.00029509
2.0515e-5
6.042e-7
1 -1
0.00059263
.00029663
2.0602e-5
6.1103e-7
ECCOMAS, 2016
39 / 44
110
P3 slope = 2.9
P4 slope = 3.9
P2 slope = 2
P1 slope = 0.96
-2
110
-4
110
-6
110
-8
110
S. Bertoluzza (IMATI-CNR)
0,04
h (log scale)
ECCOMAS, 2016
40 / 44
ECCOMAS, 2016
41 / 44
What fails?
Consider much simpler example
a(u, v ) = f (v )
ah (uh , vh ) = f (vh ),
uh Uh , vh Vh
ECCOMAS, 2016
42 / 44
Theorem
If u H s (), s k + 1
ku uh k1,\ + ku uh k1, . (hs3/2 + hs3/2 )|u|s,
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
43 / 44
Thank you
S. Bertoluzza (IMATI-CNR)
ECCOMAS, 2016
44 / 44