hp-adaptive finite elements and multi-target goal-oriented ... · first international workshop on...
TRANSCRIPT
First International Workshop on Multiphysics, Multiscale, and Optimization Problems
hp-Adaptive Finite Elements andMulti- Target Goal-Oriented Adaptivity
D. Pardo, I. Gomez-RevueltoBasque Center for Applied Mathematics (BCAM)
and IKERBASQUE
TEAM MEMBERS:D. Pardo (Research Professor)I. Gomez-Revuelto (Visiting Professor)A.-G. Saint-Guirons (Postdoctoral Fellow)M. Grigoroscuta (Postdoctoral Fellow)D. Lasa (Ph.D. Student)
MAIN COLLABORATORS:M. J. Nam, M. Paszynski,
L.E. Garcıa-Castillo, V. Calo,C. Torres-Verdın, L. Demkowicz,
H. Barucq, P. Queralt.
7 June 2010
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
overview
1. Motivation: Oil-Industry Applications.
2. hp Finite Element Methods.
3. hp Adaptivity and Goal-Oriented Adaptivity.
4. hp Goal-Oriented Adaptivity with Multiple Goals.
5. Conclusions.
1
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
motivation and objectives
Seismic Measurements
Figure from the USGS Science Center for Coastal and Marine Geology
2
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
motivation and objectives
Marine Controlled-Source Electromagnetics (CSEM)
Figure from the UCSD Institute of Oceanography
3
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
motivation and objectives
Deviated Wells (Forward Problem)Dip Angle
Invasion
Anisotropy
Different Sources(Triaxial Induction)
Eccentric LoggingInstruments
Laterolog
Through-Casing
Induction-LWD
Induction-Wireline
Goal: To find the EM fields at the receiver antennas.
4
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
motivation and objectives
• Acoustic domain: borehole fluid ( cf , ρf )
iωp + c2fρf∇ · v = 0
iωρfv +∇p = 0
• Elastic tool, casing, formation ( Vp, Vs, ρs)
0 = ∇ · σ + ρsω2u
σ = λI∇ · u + µ(∇u +∇Tu)
λ = ρs(V2
p − 2V 2s ), µ = ρsV
2s
• Coupling:nf · ∇p = ρfω2nf · u
ns · σ = −pns
• Acoustic source:nf · ∇p = 1
5
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
motivation and objectives
Radar Cross Section (RCS)Analysis
EDGE DIFRACTION
MULTIPLEREFLECTIONS
WAVEGUIDEMODES
SURFACE WAVES
Goal: Determine the RCS of a plane.
Waveguide Design
Goal: Determine electric fieldintensity at the ports.
Modeling of a Logging While
Drilling (LWD) electromagnetic
measuring device
SIDE VIEW
T1
T2
R2
R1LAYERSDIFFERENT MATERIAL
EM FIELD
Goal: Determine EM field atthe receiver antennas.
6
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
hp-finite elements
The h-Finite Element Method1. Convergence limited by the polynomial degree, and large material
contrasts.
2. Optimal h-grids do NOT converge exponentially in real applications.
3. They may “lock” (100 % error).
The p-Finite Element Method1. Exponential convergence feasible for analytical (“nice”) solutions.
2. Optimal p-grids do NOT converge exponentially in real applications.
3. If initial h-grid is not adequate, the p-method will fail miserably.
The hp-Finite Element Method1. Exponential convergence feasible for ALL solutions.
2. Optimal hp-grids DO converge exponentially in real applications.
3. If initial hp-grid is not adequate, results will still be great.
7
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
hp-finite elementsWaveguide filled with a dielectric
h-Adaptivity
hp-Adaptivity
10000 10000010
−1
100
101
102
Number of Unknowns
Rel
ativ
e E
nerg
y−N
orm
Err
or (
in %
)
hp−adaptivityp=2, h−adaptivityp=1, h−adaptivity
dd
8
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
hp-finite elements3D waveguide with a meniscus-shaped biological cultive
Final hp-grid
9
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
automatic hp-adaptivity
Fully automatic hp-adaptive strategy
-global hp-refinement
9
-global hp-refinement
9
-hp-adaptivity
10
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
automatic hp-adaptivity
Energy norm based fully automatic hp-adaptive strategyCoarse grids Fine grids
(hp) (h/2, p + 1)
-global hp-refinement
9
-global hp-refinement
9
SOL. METHOD ON FINE GRIDS:A TWO GRID SOLVER
11
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
automatic goal-oriented hp-adaptivity
Motivation (Goal-Oriented Adaptivity)
Test Problem
Resistivity = 1 Ohm mInfinite Domain
Frequency = 2 Mhz
50 meters
Receiver (R)Transmitter (T)
12
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
automatic goal-oriented hp-adaptivity
Motivation (Goal-Oriented Adaptivity)
Resistivity = 1 Ohm mInfinite Domain
Frequency = 2 Mhz
50 meters
Receiver (R)Transmitter (T)
• Solution decays exponentially.
•
|E(T )|
|E(R)|≈ 1060
• Results using energy-norm adaptivity:
– Energy-norm error: 0.001%
– Relative error in the quantity ofinterest > 1030 %.
Goal-oriented adaptivity is needed. Becker-Rannacher (1995,1996),Rannacher-Stuttmeier (1997), Cirak-Ramm (1998), Paraschivoiu-P atera (1998), Peraire-Patera(1998), Prudhomme-Oden (1999, 2001), Heuveline-Rannacher (2003), Sol in-Demkowicz (2004).
13
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
automatic goal-oriented hp-adaptivity
Motivation (Goal-Oriented Adaptivity)
Resistivity = 1 Ohm mInfinite Domain
Frequency = 2 Mhz
50 meters
Receiver (R)Transmitter (T)
• Solution decays exponentially.
•
|E(T )|
|E(R)|≈ 1060
• Results using energy-norm adaptivity:
– Energy-norm error: 0.001%
– Relative error in the quantity ofinterest > 1030 %.
Goal-oriented adaptivity is needed. Becker-Rannacher (1995,1996),Rannacher-Stuttmeier (1997), Cirak-Ramm (1998), Paraschivoiu-P atera (1998), Peraire-Patera(1998), Prudhomme-Oden (1999, 2001), Heuveline-Rannacher (2003), Sol in-Demkowicz (2004).
14
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
automatic goal-oriented hp-adaptivity
Motivation (Goal-Oriented Adaptivity)
Resistivity = 1 Ohm mInfinite Domain
Frequency = 2 Mhz
50 meters
Receiver (R)Transmitter (T)
10−80
10−60
10−40
10−20
1000
5
10
15
20
25
30
35
40
45
50
55
Amplitude (V/m)
Dis
tan
ce
in
z−
axis
fro
m tra
nsm
itte
r to
re
ce
ive
r (in
m)
−180 −90 0 90 1800
5
10
15
20
25
30
35
40
45
50
55
Phase (degrees)
Dis
tan
ce
in
z−
axis
fro
m tra
nsm
itte
r to
re
ce
ive
r (in
m)
Analytical solutionNumerical solution
Analytical solutionNumerical solution
Goal-oriented adaptivity is needed
15
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
automatic goal-oriented hp-adaptivity
Mathematical Formulation (Goal-Oriented Adaptivity)
We consider the following problem (in variational form):
Find L(u), where u ∈ V = H1(Ω) such that :∫
Ω
σ∇u · ∇v︸ ︷︷ ︸
b(u, v)
=
∫
Ω
f · v︸ ︷︷ ︸
F (v)
∀v ∈ V .
We define residual re(v) := b(e, v). We seek for a functional G ∈ V ′′ ∼ V
relating the residual and the error in the quantity of interest, that is , such thatG(re) := L(e). Functional G is the solution of the so-called dual problem:
Find G ∈ V such that :
b(u, G) = L(u) ∀u ∈ V .
Notice that, in particular, L(e) = b(e, G) ← (Representation Formula) .
16
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
automatic goal-oriented hp-adaptivity
Mathematical Formulation (Goal-Oriented Adaptivity)
DIRECT PROBLEM - u -2D Cross-Section
DUAL PROBLEM - G -2D Cross-Section
Representation Formula for the Error in the Quantity of Interest:
L(u) = b(u, G) =
∫
Ω
σ ∇u · ∇GdV
17
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
automatic goal-oriented hp-adaptivity
Summary on Goal-Oriented Adaptivity
Goal-Oriented adaptivity is a GENERALIZATION of Energy-no rmadaptivity.
• For a number of problems, goal-oriented adaptivity isexactly the same as energy-norm adaptivity.
• For other problems, the use of goal-oriented adaptivitybecomes ESSENTIAL.
To solve a problem with goal-oriented adaptivity is ALWAYSBETTER (or equal) than solving the problem with energy-normadaptivity.
18
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Model Problem (Electromagnetism)
40 m
1 Ohm−m2 Mhz1 Tx9 Rx
40 m
ads
19
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Formulations that do NOT work well
• We may define:
L(u) =
∫
RX1
u +
∫
RX2
u + ... +
∫
RXn
u
The above formula will only enable refinements in the proximity o f thereceiver closest to the transmitter, where the solution is larger tha n inthe rest of the receivers.
• We may define:
L(u) = max
∫
RX1
u,
∫
RX2
u, ...,
∫
RXn
u
This is a non-linear objective function, which is INCONSISTENT w ith themathematical theory.
20
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Formulation
We define:
Lhp(u) =
∫
RX1
u
∫
RX1
uhp
+
∫
RX2
u
∫
RX2
uhp
+ ... +
∫
RXn
u
∫
RXn
uhp
=n∑
i=1
∫
RXi
u
∫
RXi
uhp
Note that definition of Lhp depends upon the solution uhp, and therefore, itcannot be defined “a priori”.
Operator Lhp is linear and continous.
21
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Improved Formulation
We define:
Lhp(u) =n∑
i=1
R(eRXihp )
∫
RXi
u
∫
RXi
uhp
,
where R(eRXihp ) is an estimate of the relative error in the i-th quantity of
interest. From a practical point of view, this relative error can b e estimated byapproximating the exact solution with the solution in a globally re finedh/2, p + 1-grid.
The additional term enables save unneeded refinements at the rece iverswhere the error tolerance has already been achieved.
22
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Goal: Approximate Solution at Rx 4.
1000 3000 6000
100
105
1010
Rel
ativ
e E
rror
(in
%)
Number of Unknowns
Rx1
Rx2
Rx3
Rx4
Rx5
Rx6
Rx7
Rx8
Rx9
asdfasdf
23
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Final hp-grid
Goal: Rx 1 Goal: Rx 4
asdfasdf
24
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Goal: Approximate the Sum of the Solution at All Receivers.
1000 4000 10000 30000
100
105
1010
Rel
ativ
e E
rror
(in
%)
Number of Unknowns
Rx1
Rx2
Rx3
Rx4
Rx5
Rx6
Rx7
Rx8
Rx9
asdfasdf
25
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Goal: Approximate Solution at All Receivers.
1000 4000 10000 3000010
−2
100
102
104
106
108
1010
Rel
ativ
e E
rror
(in
%)
Number of Unknowns
Rx1
Rx2
Rx3
Rx4
Rx5
Rx6
Rx7
Rx8
Rx9
asdfasdf
26
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Final hp-grid
Sum of Goals Multi-goal
asdfasdf
27
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Final hp-grid (amplification)
Sum of Goals Multi-goal
asdfasdf
28
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Final solution
Solution ( Ex) Solution (amplification)
asdfasdf
29
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
10
0 c
m5
0 c
m
100 Ohm m
10000 Ohm m
1 Ohm m
100 Ohm m
15
cm
10
0 c
m
0.0
00
00
1 O
hm
m0
.1 O
hm
mDescription of Antennas
10 c
m
10000 Ohm m
Borehole0.1 Ohm m
10000 Rel. Perme.
0.00
0001
Ohm
mM
andr
el
Radius=10.8 cm
Radius 7.6 cm
Magnetic Buffer
Goal: To Study the Effectof Invasion, Anistotropy, andMagnetic Permeability.
30
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Axisymmetric Logging-While-Drilling (LWD) SimulationENERGY-NORM HP-ADAPTIVITY
p=1
p=2
p=3
p=4
p=5
p=6
p=7
p=8
-0.6851852 2.907408 -1.203704
1.018519 2Dhp90: A Fully automatic hp-adaptive Finite Element code
31
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Axisymmetric Logging-While-Drilling (LWD) Simulation
GOAL-ORIENTED HP-ADAPTIVITY (Quadrilateral Elements)
p=1
p=2
p=3
p=4
p=5
p=6
p=7
p=8
-0.7222222 2.870370 -1.137037
1.085185 2Dhp90: A Fully automatic hp-adaptive Finite Element code
32
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Axisymmetric Logging-While-Drilling (LWD) Simulation
GOAL-ORIENTED HP-ADAPTIVITY (Zoom towards first receiver)
p=1
p=2
p=3
p=4
p=5
p=6
p=7
p=8
1.7160494E-02 0.1369136 0.3603704
0.4344445 2Dhp90: A Fully automatic hp-adaptive Finite Element code
33
multiple goals
First Vert. Diff. Hφ for different antennas
10−5
100
−1.5
−1
−0.5
0
0.5
1
1.5
2
2.5
3
Amplitude First Vert. Diff. Magnetic Field (A/m2)
Ver
tical
Pos
ition
of R
ecei
ving
Ant
enna
(m
)No Invasion
Vert. Dipoles (Ring)Toroid
−200 −100 0 100 200−1.5
−1
−0.5
0
0.5
1
1.5
2
2.5
3
Phase (degrees)
Ver
tical
Pos
ition
of R
ecei
ving
Ant
enna
(m
)
Vert. Dipoles (Ring)Toroid
100 Ohm−m
10000 Ohm−m
1 Ohm−m
100 Ohm−m
100 Ohm−m
10000 Ohm−m
1 Ohm−m
100 Ohm−m
In LWD instruments, we obtain similar results using toroids or a ring of vert. dipoles
multiple goals
First Vert. Diff. Ez for a toroid antenna
10−4
10−2
100
102
−1.5
−1
−0.5
0
0.5
1
1.5
2
2.5
3
Amplitude First Vert. Diff. Ez (V/m2)
Ver
tical
Pos
ition
of R
ecei
ving
Ant
enna
(m
)
No Invasion
Toroid
−200 −100 0 100 200−1.5
−1
−0.5
0
0.5
1
1.5
2
2.5
3
Phase (degrees)
Ver
tical
Pos
ition
of R
ecei
ving
Ant
enna
(m
)
Toroid
100 Ohm−m
10000 Ohm−m
1 Ohm−m
100 Ohm−m
100 Ohm−m
10000 Ohm−m
1 Ohm−m
100 Ohm−m
Toroids are adequate for identifying highly resistive layers
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Verification of Sonic Measurements
• Formation thickness: 0.25m.
• Monopole and dipole source, central frequency 8603 Hz .
36
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Monopole source — fast formation
P-wave S-wave DensityVp [m/s] 1/Vp [µs/ft] Vs [m/s] 1/Vs [µs/ft] ρ [kg/m3]
fluid 1524 200.0 − − 1100
fast form. 3048 100.0 1793 170.0 2200
tool 5860 52.0 3130 97.4 7800
37
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Monopole source — fast formation — no tool
38
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Monopole source — fast formation — with tool
39
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Monopole source — fast formation
P-wave S-wave DensityVp [m/s] 1/Vp [µs/ft] Vs [m/s] 1/Vs [µs/ft] ρ [kg/m3]
fluid 1524 200.0 − − 1100
fast form. 3048 100.0 1793 170.0 2200
tool 5860 52.0 3130 97.4 7800
40
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Dipole source — fast formation — no tool
41
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Dipole source — fast formation — with tool
42
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Dipole source — fast formation
P-wave S-wave DensityVp [m/s] 1/Vp [µs/ft] Vs [m/s] 1/Vs [µs/ft] ρ [kg/m3]
fluid 1524 200.0 − − 1100
fast form. 3048 100.0 1793 170.0 2200
tool 5860 52.0 3130 97.4 7800
43
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Direct calculation of dispersion curves
Dispersion curves contain the information about the slowness of theformation.
• dispersion curves are smooth with respect to variations in frequency,
• it is enough to calculate results only for a few frequencies (below 50)
• further reduction of the number of needed frequencies to 10 possible when only Vp
(high frequencies) and Vs (low frequencies) are needed.
⇒ Dispersion curves are obtained directly from frequency domain r esults.
• no need to use in the simulations a Ricker wavelet (neither any other wavel et),
• the added stability of the problem,
• better performance of PML in frequency domain,
• smaller complexity of the problem.
44
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Direct calculation of dispersion curves
45
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Examples with layers
• Formation thickness: 0.25m.
• Monopole/dipole source: Ricker wavelet, central frequency 8603 Hz .
46
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Layers (1) — monopole (waveforms)
47
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
multiple goals
Layers (2) — monopole (waveforms)
48
D. Pardo, I. Gomez-Revuelto For more info, visit: www.bcamath.org/research/mip
conclusions
1. hp-Adaptive Finite Elements provides superior accuracy andcan be used for simulation of multi-physics problems.
2. Goal-oriented adaptivity is useful in engineering probl emswhen a specific goal is pursued.
3. It is possible to design adaptive schemes for the case ofmultiple goals.
49