mesh adaptation via shape optimization · 2016. 10. 30. · j(uh) = j(uh(th)) = j((th)) min...
TRANSCRIPT
![Page 1: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/1.jpg)
Mesh adaptation via shape optimization
Zoubida Mghazli(1)
(1) Universite Ibn Tofail, Kenitra, Maroc,
collaborators
A. Alla (1), M. Fortin(2), F. Hecht(3), M. Masmoudi(4)
(2) Universite Laval, Quebec , (3) Universite Pierre et Marie Curie, Paris,(4) Universite Paul Sabatier, Toulouse
Wednesday, July 7, 2010
Workshop on A posteriori error estimates and mesh adaptivity for
evolutionary and nonlinear problems
1
![Page 2: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/2.jpg)
Introduction
The shape optimization problems are generally in the form
minΩ∈Oad
j(Ω)
where
• j(Ω) = J(Ω, uΩ) is a fonctional of the domain Ω
• Ω ∈ O, a set of admissible domains
• uΩ solution of a PDE in Ω.
If we consider a functional J(·) which measures the error made
in approaching the PDE by a finite element method built on a
mesh Th, this raises the question of whether one can obtain an
optimal mesh in the sense that minimizes the functional J(·)
2
![Page 3: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/3.jpg)
J(uh) = J(uh(Th)) = j((Th))
minTh∈Tad
j((Th))
The mesh adaptation procedure is based on a number of local
operations on the initial mesh:
• node displacement,
• edge refinement,
• node suppression,
• edge swapping.
This sequence is repeated a number of times until the mesh
stabilizes.
3
![Page 4: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/4.jpg)
The decision whether a given local operation has to be performed
is based on the definition of the error on an element K and on
its gradient. The local operations are used with two objectives:
control the error level on each element so that the absolute error
takes the desired value ed everywhere. The second objective is
to achieve equidistribution of this error by minimizing the error
gradient.
If we fixe the number of nodes of a triangulation Th, we prove
in the first part of this presentation that the optimal position of
the nodes can be obtained by minimizing the approximation error
using a hierarchical estimator. The node displacement procedure
(also called R-adaptation)which is a crucial step in a global mesh
adaptation procedure, is obtained by the derivative of the func-
tional with respect to the triangulation Th.4
![Page 5: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/5.jpg)
The usual techniques for shape optimization exclude changetopological domain and does not allow the creation of holes.
Topological optimization is concerned with the variation of acost function with respect to a topological modification of adomain. The most simple way of modifying the topology consistsin creating a small hole in the domain. In the case of structuralshape optimization, creating a hole means simply removing somematerial.For ε ≥ 0 given let Ωε = Ω \ (x0 + εω) be the subset obtainedby removing a small part (x0 + εω) from Ω, where x0 ∈ Ω andω ⊂ Rn is a fixed open and bounded subset containing the origin.
Oad = Ωε, ε ≥ 0Generally, the functional j(·) is not differentiable functional withrespect to ε at ε = 0,
5
![Page 6: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/6.jpg)
An asymptotic expansion of the function j(·) is obtained in the following form:
(J. Cea, S. Garreau, PH. Guillaume, M. Mamsoudi, 2000)
j(Ωε) = j(Ω) + ρ(ε)g(x0) + o(ρ(ε)).
limε→0
ρ(ε) = 0, ρ(ε) > 0.
The function g(·) is called the topological sensitivity or topological gradient.It can be used as a descent direction of the domain optimization process.Obviously, if we want to minimize j(·), the ”best” place to create an in-finitesimal hole is there where g(x) is the most negative. Starting with thisobservation, topological optimization algorithms can then be constructed.
In the context of mesh adaptation, the edge refinement and node suppressionares local operations that can be seen as a change of mesh topology.
In the second part of this talk, we present some ideas to deal the mesh
refinement as a topological optimization.
6
![Page 7: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/7.jpg)
Outline
• The model problem
• The node displacement as a Shape optimization problem
• Refinement via an asymptotic expansion
• Conclusion and perspective
7
![Page 8: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/8.jpg)
• Model problem
u ∈ V = H10(Ω) such that a(u, v) = L(v) ∀v ∈ V
a(u, v) :=
∫Ω
(∇u · ∇v + αuv) dx, L(v) :=
∫Ωfvdx.
Th triangulation of Ω.
Vkh = vh ∈ V; vh|K ∈ Pk(K), ∀ K ∈ Th ∩H10(Ω)
uh ∈ Vkh tel que a(uh, vh) = L(vh), ∀vh ∈ Vkh.
e = u− uh approximation error
Can we determine the ”optimal” mesh in the sense that
minimizes the approximation error??
8
![Page 9: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/9.jpg)
The first ideas in this direction are due to M. C. Delfour, G. Payre,
J.P. Zolesio, (1986). They considered the energy functional as a
cost functional .
For two triangulations Th and Th′ they showed, with the help
of directional derivatives, that respective approximate solutions
verify
J(uh) = J(u) +1
2a(uh − u, uh − u)
‖u− uh‖2 − ‖u− uh′‖2 = 2 (J(uh) − J(uh′))
so minimizing the error becomes minimizing J .
Il seems that the ”best” representation of the error is the hier-
archical estimator.9
![Page 10: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/10.jpg)
Hierarchical estimator
The hierarchical method is based on a simple idea: given an
approximation of order k, a better approximation, of order k+ 1
could be used to assess the accuracy of the solution. If now Vkh ⊂V denotes the discrete space of piecewise continuous polynomials
of degree k, the approximate solution is obtained by finding ukh ∈Vkh such that:
a(ukh, wkh) = L(wkh), ∀wkh ∈ Vkh.
The error ek = u− ukh is then the solution of:
a(ek, w) = Rh(w), ∀w ∈ V
where Rh(w) = L(w) − a(ukh, w) is the residual.
10
![Page 11: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/11.jpg)
An approximation of this error ek can be obtained by consideringthe space Vk+1
h of piecewise continuous polynomials of degreek + 1 and by solving:
a(ek+1h , wk+1
h ) = R(wk+1h ), ∀wk+1
h ∈ Vk+1h .
This would imply computing a full problem in the space Vk+1h ,
which would obviously be prohibitive.
The idea is then to use a hierarchical finite element basis of thespace Vk+1
h :
Vk+1h = Vkh ⊕ Ek+1
h
so that Vk+1h is the direct sum of Vkh (polynomials of degree k)
and a correction space Ek+1h consisting of polynomials of degree
k + 1.
11
![Page 12: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/12.jpg)
The error can then be approximated by solving the following
problem: Find ek+1h ∈ Ek+1
h such that
a(ek+1h , wk+1
h ) = R(wk+1h ), ∀wk+1
h ∈ Ek+1h .
In the following we consider the case where a piecewise linear
approximation which is corrected by a piecewise quadratic one,
only the mid-edge nodes are computed, leaving unchanged the
values at the vertices.
We drop the index k + 1 and write hencefortheh ∈ Eha(eh, wh) = Rh(wh) ∀ wh ∈ Eh.
12
![Page 13: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/13.jpg)
This method has been analysed for a wide class of problems. The typicalresult is global. Under two principal hypothesis one can show an estimate ofthe following form
(R.E.Bank, R.K. Smith,1993): Under
(1) Saturation assumption
∃β, 0 < β < 1 tel que ‖|u− u(k+1)h ‖| ≤ β ‖|u− u
(k)h ‖|
(2) Cauchy-Binyakowski-Schwarz inegality (ou strong Cauchy-
Schwarz inegality)∃γ < 1 independant of h, such that
∀v ∈ Vkh ∀w ∈ Eh |a(v, w)| < γ ‖|v‖| ‖|w‖|,
We have (1−β2)(1−γ2)‖|u−u(k)h ‖|2 ≤ ‖|eh‖|2 ≤ ‖|u−u(k)
h ‖|2.
13
![Page 14: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/14.jpg)
Without saturation assumption (A. Agouzal, 2002)
Vb = Vh ⊕ bT , T ∈ Th
‖u− uh‖ ≤ C0‖εh‖ + C
∑K∈Th
h2K‖f − fK‖2
0,K
1/2
∀K ∈ Th, ‖εh‖ ≤ C1‖u− uh‖1,∆(K).
εh local representation .
14
![Page 15: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/15.jpg)
• The node displacement
(a Shape optimization problem)
A.Alla, M. Fortin, F. Hecht, Z.M.,: ” R-Adaptation de maillage par l’estimateur
d’erreur hierarchique”, ARIMA, Volume 5, p1-10, 2006.
−→ For Th given:
uh = uh(Th), and eh = eh(Th)
Objective: the best position nodes.
−→ minimize error approximation via the error estimator hierar-
chical
15
![Page 16: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/16.jpg)
Let
J(eh(Th)) :=1
2
∫Ωh
|∇eh(Th)|2 =1
2‖|eh(Th)‖|2
the minimization problem isFind the nodes position of Th that minimizes J(eh(Th)),under the constraintsa(eh, vh) = L(vh) − a(uh, vh) ∀ vh ∈ Eha(uh, vh) = L(vh) ∀ vh ∈ Vh.
(1)
16
![Page 17: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/17.jpg)
M: the set of nodes of ThTT the set of triangulations Th(M) generated by M with thesame topological table .mT = M : M ⊂ Ω/ Th(M) ∈ TT ,
and let
j(M) := J(eh(Th(M))
The shape optimization problem is:FindM ∈ mT such that :
j(M) = infM∈mT
j(M), witha(eh(Th(M)), vh) = L(vh) − a(uh(Th(M)), vh) ∀ vh ∈ Eha(uh(Th(M)), vh) = L(vh) ∀ vh ∈ Vh
17
![Page 18: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/18.jpg)
We assume that the last problem admits a unique solution. To
calculate the derivative of the functional j(·) with respect to
shape parameters which are the nodes of the mesh, we intro-
duce the Lagrangian to transform the constrained problem to a
problem without constraints.
The Lagrangian in Eh × Vh × Eh × Vh is given by
L(vh, vh, qh, µh) = J(vh) + [a(vh, qh) − L(qh) + a(vh, qh)]
+[a(vh, µh) − L(µh)].
18
![Page 19: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/19.jpg)
Dual problems are given by:
Find ph ∈ Eh, such that∫Ωh
∇ph ∇vh dΩh = −∫Ωh
∇eh ∇vh dΩh ∀vh ∈ Eh.
Find λh ∈ Vh, such that∫Ωh
∇λh ∇vh dΩh = −∫Ωh
∇ph ∇vh dΩh ∀vh ∈ Vh,
where the notation Ωh means that the domain is meshed with
the triangulation Th.
19
![Page 20: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/20.jpg)
To calculate the shape derivative of j(·) we use the techniquesof Zolezio-Delfour-Sokolowski . The variation of the domain Ωis defined through a regular transformation Tt : Rn −→ Rn givenby a velocity field V and such that
Tt(Ωh) = Ωth, Tt(M) = Mt
we note et := eht = eth T−1h ∈ Eth.
For t > 0:
j(Mt) =1
2
∫Ωth
|∇et|2 dΩth =1
2
∫Ωh
(A(t) ∇eth,∇e
th
)dΩh
where
A(t) = DT−1t Jt
∗DT−1t ,
Jt = det(DTt)
A′(0) = divV (0) I − (DV (0) + ∗(DV (0)))
20
![Page 21: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/21.jpg)
Using this change of variable, the dual problems and the shapederivative of bilinear forms we obtain the explicit formula for thederivative of the fonctional
dj(Ωh;V ) =1
2
∫Ωh
(A′(0) ∇eh,∇eh
)dΩh −
∫Ωh
div (fV (0))
(ph + λh) dΩh +
∫Ωh
A′(0) ∇uh∇λh dΩh
+
∫Ωh
A′(0) ∇(uh + eh) ∇ph dΩh
We use the velocity fields, continuous and piecewise linear (whichtransforms the triangles into triangles) introduced by Delfour-Payre-Zolezio (1985) and defined from φii basis of Vh.
We take successively V1 = (φi, 0) and V2 = (0, φj) to computede partial derivative of J
21
![Page 22: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/22.jpg)
Algorithm
Data: ε, initial mesh T (0)h .
Step 1: computation of the approximate solution uh.
Step 2: computation of the estimator eh and the functional j.
Step 3: computation of the dual solutions ph et λh.
Step 4: For every node do
- computation of dj(M, V1) and dj(M, V2) ou V1 = (φi, 0) et V2 =
(0, φi);
- calcul de ‖dj(M, V )‖ = ((dj(M, V1))2 + (dj(M, V2))
2)1
2.
Etape 5: test := maximum of ‖dj(M, V )‖;if test > ε
- displacement of the node in the direction of dj(M, V ), (new tri-
angulation T (n)h );
-go to step 1.
otherwise STOP.
22
![Page 23: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/23.jpg)
Numerical tests: A manufactured solution given by
u(x, y) = xy(x− 1)(y − 1)e−50((x−1/2)2+(y−1/2)2)
23
![Page 24: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/24.jpg)
50 iterations
24
![Page 25: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/25.jpg)
100 iterations
25
![Page 26: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/26.jpg)
150 iterations
26
![Page 27: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/27.jpg)
200 iterations
27
![Page 28: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/28.jpg)
250 iterations
28
![Page 29: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/29.jpg)
300 iterations
29
![Page 30: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/30.jpg)
350 iterations
30
![Page 31: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/31.jpg)
400 iterations
31
![Page 32: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/32.jpg)
450 iterations
32
![Page 33: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/33.jpg)
500 iterations
33
![Page 34: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/34.jpg)
550 iterations
34
![Page 35: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/35.jpg)
Cost functionnal
0 100 200 300 400 500 6000.6
0.8
1
1.2
1.4
1.6
1.8
2x 10
−3
Itérations
Fon
ctio
nnel
le c
oût
Evolution de la fonctionnelle vs itérations
Fonctionnelle coût
35
![Page 36: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/36.jpg)
• Refinement via an asymptotic expansionwe just want to present some ideas on how to make the er-ror functional minimization, as a shape optimization, when themesh adaptation requires a discontinuity in the mesh topologyby introducing new nodes.
Until now the development is not complete.
General framework of asymptotic expansion
• ε ≥ 0 given, Ωε = Ω \ (x0 + εω), x0 ∈ Ω et ω ⊂ Rn• V Hilbert space,
uε ∈ V, aε(uε, v) = `ε(v), ∀v ∈ V• aε(·, ·) a bilinear form, symmetric and uniformly coercive on V• `ε(·) a linear form uniformly continues on V
36
![Page 37: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/37.jpg)
Hypothesis:
• ‖aε − a0 − f(ε)δa‖ = o(f(ε))
• ‖`ε − `0f(ε)δ`‖ = o(f(ε))
• J0 differentiable with respect to u and
Jε(v) − J0(u) = DJ0(v − u) + f(ε)δJ + o(f(ε))
• p0 solution of the dual problem :
a0(w, p0) = −DJ0(u0)w, ∀w ∈ V
then (M.Masmoudi) j(ε) := J(uε) is such that
j(ε) − j(0) = f(ε)δj + o(f(ε))
δj = δa+ δ`+ δJ
37
![Page 38: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/38.jpg)
Mesh adaptation .
We recall the expression of the fonctionnal to optimize
J(eh) =1
2‖|eh|‖2 where
eh ∈ Eh, such that :a(eh, wh) = R(uh; wh) ∀ wh ∈ Eh.
For v ∈ H10(Ω) we have, by integration by part the following
expression of the residual
R(uh; v) = L(v) − a(uh, v)
=∑K∈Th
∫K
(f − αuh)vdx−∑E∈E0
h
∫E
[∂uh
∂nE
]Evds
38
![Page 39: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/39.jpg)
We introduce a perturbation which is performed in each element
as follow
Rc(uh;wh) =∑K∈Th
∫K
(f − αuh − c1)whdx
−∑E∈Eh
∫E(
[∂uh
∂nE
]E
− c2)whds ∈ R
where, For 1 < p < 2,
c = (c1, c2) ∈ G :=∏K∈Th
Lp(K) ×∏E∈Eh
Lp(E)
39
![Page 40: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/40.jpg)
Find ech ∈ Eh, solution ofa(ech, wh) = Rc(uh;wh) ∀wh ∈ Eh
Cost functional: J (c) = J(ech)
The Lagrangien: L(c, eh, wh) := J(eh)+a(eh, wh)−Rc(uh;wh)
Dual problempch ∈ Eha(ψh; p
ch) = −DJ(ech) · ψ ∀ψ ∈ Eh
J is differentiable and for δc ∈∏K∈Th L
p(K) ×∏E∈Eh L
p(E)
J ′(c) · δc =∑K∈Th
∫Kδc1 · pchdx+
∑E∈Eh
∫Eδc2 · pchds
40
![Page 41: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/41.jpg)
For K ∈ Th and E ∈ E0h, let δcK = (δcK, 0) and δcE = (0, δcE)
where
δcK =
(f − αuh) in K
0 elsewhere, δcE =
[∂uh
∂nE
]E
on E
0 elsewhereso
J ′(0) · δcK =
∫K
(f − αuh)p0hdx
J ′(0) · δcE =
∫E
[∂uh
∂nE
]Ep0hds
41
![Page 42: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/42.jpg)
The perturbation related to adding a new node can be presented
as follows: xεK is the new node in K, ε is the distance to vertex.
M −→ Th −→ uh
M ⊕ xεK −→ T εh −→ uεh = uh + δuh
For K ∈ Th f − αuεh = f − αuh − αδuh
For E ∈ E0h,
[∂uεh∂nE
]E
=[∂uh∂nE
]E
+[∂δuh∂nE
]E
42
![Page 43: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/43.jpg)
j(ε) = J(eh(T εh ))
Th: triangulation associated with the set MT εh : triangulation associated with the set M ⊕ xε.
Question:
j(ε) = j(0) + f(ε)g(xε) + (f(ε))?
Conjecture:
g(xε) :=
∫K(f − αuh)p
0hdx if xε ∈ K∫
E
[∂uh∂n
E
]Ep0hds if xε ∈ E
or
g(xε) :=
∫Kehp
0hdx
We test the three cases. It seems that the last one is the best.43
![Page 44: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/44.jpg)
:Dispalcement 1 (350 iterations)
44
![Page 45: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/45.jpg)
Refinement with the first one
raffinement de maillage: residu
45
![Page 46: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/46.jpg)
Displacement (131 iterations)
46
![Page 47: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/47.jpg)
If we take
g(xε) :=
∫Kehp
0hdx
we have
47
![Page 48: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/48.jpg)
:Dispalcement 1 (350 iterations)
48
![Page 49: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/49.jpg)
First Refinement
raffinement de maillage: eh
49
![Page 50: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/50.jpg)
second Refinement
raffinement de maillage: eh
50
![Page 51: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/51.jpg)
Displacement
51
![Page 52: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/52.jpg)
Iteration 50
52
![Page 53: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/53.jpg)
Iteration 100
53
![Page 54: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/54.jpg)
Iteration 150
54
![Page 55: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/55.jpg)
Iteration 193
55
![Page 56: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/56.jpg)
One dimensionnal Case
Forf ∈ L2(]0, 1[ and α ∈ R let u ∈ H10(]0, 1[) solution of∫ 1
0(u′v′ + αuv)dx =
∫ 1
0fvdx.
Let Th = Ki1≤i≤N , et Ki = [xi, xi+1].
Vh = vh ∈ C0([0, 1]/vh|K ∈ P1(K) ∀K ∈ Th, vh(0) = vh(1) = 0
56
![Page 57: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/57.jpg)
eh ∈ Eh:∫ 1
0(e′hw
′h + αehwh)dx = R(uh;wh) ∀wh ∈ Eh
Eh = wh ∈ C0([0, 1]) /wh ∈ P2(K) ∀K ∈ Th,wh(xi) = 0, 1 ≤ i ≤ N
R(uh;wh) :=N∑i=1
∫ xi+1
xi
(f − αuh)whdx+N∑i=1
[u′h]iwh(xi)
=N∑i=1
∫ xi+1
xi
(f − αuh)whdx
For Ki = [xi, xi+1] let xεK = xi0 middle of Ki
g(xεK) = (f − αuh)((xi0))q
0h(x
i0)
57
![Page 58: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/58.jpg)
Adaptation Strategy
Initial meshT 0,0h
Step 1: Adaptation 1:
ηK := hK‖f − fK‖0,K −→ T 0,1h .
Step 2:
− Computation of uh, eh− Displacement 1 −→ T 0,2
h = T 0h .
Step 3: Computation of g(xi0) for i = 1, . . . , N
58
![Page 59: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/59.jpg)
Step 4: In Kn where g(xn0 ) is the most negative, add a node
mεn = xn + hn
2 + ε, and −hn2 < ε < hn
2 .
step 5: local update
(ϕi)1≤i≤N base functions of Vh associated to T 0h = (xi)1≤i≤N
(ϕi)1≤i≤N+1 new base functions:
ϕi = ϕi pour i 6= n− 1, n, n+ 1
59
![Page 60: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/60.jpg)
Computation of the perturbed solution
uεh ∈ C0([0, 1]
∀K 6= Kn uεh|K = uh|K
Sur Kn
uεh(xn) = uh(xn)uεh|[xn,mε
n]∈ P1([xn,m
εn])
uεh|[mεn,xn+1]
∈ P1([mεn, xn+1])
uεh(xn+1) = uh(xn+1).
uεh(mεn) solution of local problem (where ϕmε
nis the base func-
tion associeted to mεn)∫ xn+1
xn
((uεh)
′(ϕmεn)′ + αuεhϕmε
n
)dx =
∫ xn+1
xnfϕmε
ndx
60
![Page 61: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/61.jpg)
Etape 6: globale update by local problems
Computation of uεh(xj) (for j 6= n+ 1)∫ xj+1
xj−1
((uεh)
′(ϕj)′ + uεhϕj
)dx =
∫ xj+1
xj−1
fϕjdx.
Step 7 Computation of the new estimator
a(eεh, wh) = R(uεh; wh), ∀wh ∈ Eh.
Step 8 Computation of the new g(xi0), i = 1, . . . , N + 1
Step 9 Displacement (every k iterations)
61
![Page 62: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/62.jpg)
A Numerical Test
MATLAB
Ω = [0, 1]
u(x, y) = x4(x− 1)
62
![Page 63: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/63.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Approximation initiale (7 noeuds)
sol exactsol approche
63
![Page 64: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/64.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajout d’un noeud
sol exactsol approche
![Page 65: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/65.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajuster les valeurs à gauche du noeud ajouté
sol exactsol approche
![Page 66: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/66.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajout d’un noeud
sol exactsol approche
![Page 67: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/67.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajuster les valeurs à gauche du noeud ajouté
sol exactsol approche
![Page 68: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/68.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajout d’un noeud
sol exactsol approche
![Page 69: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/69.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajuster les valeurs à droite du noeud ajouté
sol exactsol approche
![Page 70: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/70.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajuster les valeurs à gauche du noeud ajouté
sol exactsol approche
![Page 71: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/71.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajout d’un noeud
sol exactsol approche
![Page 72: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/72.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajuster les valeurs à gauche du noeud ajouté
sol exactsol approche
![Page 73: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/73.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajout d’un noeud
sol exactsol approche
![Page 74: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/74.jpg)
And with g(x0) = eh(x0)q0h(x0) we obtain
64
![Page 75: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/75.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajout d’un noeud − indicateur −||eh||2
sol. exactesol. approchée
![Page 76: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/76.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajuster les valeurs à gauche du noeud ajouté
sol. exactesol. approchée
![Page 77: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/77.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajout d’un noeud − indicateur −||eh||2
sol. exactesol. approchée
![Page 78: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/78.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajuster les valeurs à droite du noeud ajouté
sol. exactesol. approchée
![Page 79: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/79.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajuster les valeurs à gauche du noeud ajouté
sol. exactesol. approchée
![Page 80: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/80.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajout d’un noeud − indicateur −||eh||2
sol. exactesol. approchée
![Page 81: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/81.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajuster les valeurs à gauche du noeud ajouté
sol. exactesol. approchée
![Page 82: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/82.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajout d’un noeud − indicateur −||eh||2
sol. exactesol. approchée
![Page 83: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/83.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajuster les valeurs à droite du noeud ajouté
sol. exactesol. approchée
![Page 84: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/84.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajuster les valeurs à gauche du noeud ajouté
sol. exactesol. approchée
![Page 85: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/85.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajout d’un noeud − indicateur −||eh||2
sol. exactesol. approchée
![Page 86: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/86.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajuster les valeurs à droite du noeud ajouté
sol. exactesol. approchée
![Page 87: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/87.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajuster les valeurs à gauche du noeud ajouté
sol. exactesol. approchée
![Page 88: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/88.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajout d’un noeud − indicateur −||eh||2
sol. exactesol. approchée
![Page 89: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/89.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajuster les valeurs à droite du noeud ajouté
sol. exactesol. approchée
![Page 90: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/90.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajuster les valeurs à gauche du noeud ajouté
sol. exactesol. approchée
![Page 91: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/91.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajout d’un noeud − indicateur −||eh||2
sol. exactesol. approchée
![Page 92: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/92.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajuster les valeurs à droite du noeud ajouté
sol. exactesol. approchée
![Page 93: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/93.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajuster les valeurs à gauche du noeud ajouté
sol. exactesol. approchée
![Page 94: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/94.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajout d’un noeud − indicateur −||eh||2
sol. exactesol. approchée
![Page 95: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/95.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
domaine de discrétisation
u(x)
=x4 (x
−1)
Ajuster les valeurs à gauche du noeud ajouté
sol. exactesol. approchée
![Page 96: Mesh adaptation via shape optimization · 2016. 10. 30. · J(uh) = J(uh(Th)) = j((Th)) min Th∈Tad j((Th)) The mesh adaptation procedure is based on a number of local operations](https://reader033.vdocuments.us/reader033/viewer/2022060912/60a64626a8728e2d4c67dc89/html5/thumbnails/96.jpg)
Conclusion and perspective
- Displacement
- minimize the number of nodes
We can say that the node displacement algorithm works quitewell. It remains to find the criterion for swapping edge to com-plete this algorithm.
- up date: solving the linear system only one times
- Hidden error
it remains to determine the exact topological gradient to deter-mine the right places to add or delete node
66