Professional Documents
Culture Documents
Outline
1 Introduction
2 Problem Statement
5 References
6 Acknowledgements
Outline
1 Introduction
2 Problem Statement
5 References
6 Acknowledgements
Ωf Ωf
Ω Ω
Outline
1 Introduction
2 Problem Statement
5 References
6 Acknowledgements
Problem Statement
µp ∂t p + ∇·u = fp ,
µu ∂t u + ∇p = f u ,
p(x, 0) = 0, u (x, 0) = 0,
Variational Form I.
n o
Vp = q ∈ H 1 (Ω)| q = 0 on Γp ,
n o
Vu = v ∈ H(div, Ω)| γn v = 0 on Γu and γn v ∈ L2 (Γo ) .
Variational Form II.
Vp = L2 (Ω),
n o
Vu = v ∈ H(div, Ω)| γn v = 0 on Γu and γn v ∈ L2 (Γo ) .
Variational Form III.
n o
Vp = q ∈ H 1 (Ω)| q = 0 on Γp ,
d
Vu = L2 (Ω) .
The Galerkin method only controls the L2 (Ω)-norm of the unknowns but
not ∇p or ∇·u. Therefore, stabilization is needed.
Let Vp,h ⊂ Vp and Vu,h ⊂ Vu , Vh = Vp,h × Vu,h .
Stabilized FE methods deal with the following problem: Find a pair
[ph , u h ] ∈ C 1 (Υ; Vh ) satisfying the initial conditions ph (x, 0) = 0,
u h (x, 0) = 0 and such that
for all test functions [qh , v h ] ∈ Vh , where the bilinear form Bs and the
linear form Ls include the Galerkin terms and additional stabilization
terms. Depending on how the stabilization part is designed, a different
stabilization method arises. We analyze two stabilization methods: ASGS
and OSS. The stabilization terms depend on the choice of the stabilization
parameters τp and τu .
H. Espinoza et al. (UPC BarcelonaTech) NRBC Mixed Wave Equation Jul/2014 12 / 32
Introduction Problem Analysis Numerical References Acknowledgements
ASGS:
OSS:
⊥
Bs ([ph , u h ] , [qh , v h ]) = B ([ph , u h ] , [qh , v h ]) + Pp,h (∇·u h ) , τp ∇·v h
⊥
+ Pu,h (∇ph ) , τu ∇qh ,
⊥ ⊥
Ls ([qh , v h ]) = L ([qh , v h ]) + Pp,h (fp ) , τp ∇·v h + Pu,h (f u ) , τu ∇qh .
Stabilization Parameters
s s s
µu `p µp `u
r
τp = Cτ h , τu = Cτ h ,
µp `u µu `p
where Cτ is a dimensionless algorithmic constant and `p , `u are length
scales corresponding to p and u respectively.
The length scales allow to mimic at the discrete level the different
functional settings of the continuous problem.
This is the reason to chose the length scales as shown in the table.
methods.
Variational Form I II III
`p `p = `u L20 /h h
`u `p = `u h L20 /h
τp O(h) O(1) O(h2 )
τu O(h) O(h2 ) O(1)
Outline
1 Introduction
2 Problem Statement
5 References
6 Acknowledgements
Stability
For smooth enough forcing terms:
|||[ph , u h ]|||W ,ASGS,h . |||[fp , f u ]|||W ,ASGS,h ,
|||[ph , u h ]|||W ,OSS,h . |||[fp , f u ]|||W ,OSS,h ,
|||[ph , u h ]|||S,OSS,h . |||[fp , f u ]|||S,OSS,h .
|||[qh , v h ]|||20,h = µp ||qh ||2L∞ (Υ,L2 (Ω)) + (1 + σ)κp ||qh ||2L2 (Υ,L2 (Γo ))
+ µu ||v h ||2L∞ (Υ,L2 (Ω)) + (1 − σ)κu ||γn v h ||2L2 (Υ,L2 (Γo )) ,
|||[qh , v h ]|||2W ,ASGS,h = |||[qh , v h ]|||20,h + τp ||µp ∂t qh + ∇·v h ||2L2 (Υ,L2 (Ω))
+ τu ||µu ∂t v h + ∇qh ||2L2 (Υ,L2 (Ω)) ,
|||[qh , v h ]|||2W ,OSS,h = |||[qh , v h ]|||2W ,ASGS,h ,
|||[qh , v h ]|||2S,OSS,h = |||[qh , v h ]|||20,h + τp ||∇·v h ||2L2 (Υ,L2 (Ω)) + τu ||∇qh ||2L2 (Υ,L2 (Ω)) .
Convergence
For smooth enough exact solutions, we have proved the following a priori
error estimates:
Convergence*
Outline
1 Introduction
2 Problem Statement
5 References
6 Acknowledgements
1.05
0.95
0.85
0.75
0.65
0.55
0.45
0.35
0.25
0.15
0.05
-0.05
-0.15
1.05
0.95
0.85
0.75
0.65
0.55
0.45
0.35
0.25
0.15
0.05
-0.05
-0.15
1.05
0.95
0.85
0.75
0.65
0.55
0.45
0.35
0.25
0.15
0.05
-0.05
-0.15
0.15 -0.01
0.10
0.10
-0.02
0.05 0.05
p (100) -0.03
p (150)
p (50)
0.00
0.00 -0.03
-0.05
-0.04
-0.10
-0.05
-0.15 Exact -0.04
This Work
-0.20 -0.10 -0.04
0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100
x x x
The NRBC works works very well. There is a difference at the final
simulation time, but that difference is in small amplitudes and is low
energy.
1050
950
850
750
650
550
450
350
250
150
50
-50
-150
1050
950
850
750
650
550
450
350
250
150
50
-50
-150
3.5e+6
Big
3.0e+6 Sma
2.5e+6
2.0e+6
E
1.5e+6
1.0e+6
0.5e+6
0
0 5 10 15 20 25
t
Figure : Evolution of total energy E for the big/small domain benchmark problem
in 2D.
55
45
35
25
15
5
-5
-15
-25
-35
-45
-55
-65
55
45
35
25
15
5
-5
-15
-25
-35
-45
-55
-65
3.0e+9
Big
2.5e+9 Sma
2.0e+9
1.5e+9
E
1.0e+9
0.5e+9
0
0 5 10 15 20 25
t
Figure : Evolution of total energy E for the big/small domain benchmark problem
in 3D
References
Acknowledgements