Documente Academic
Documente Profesional
Documente Cultură
Zhiqiang Cai
Purdue University
• Introduction
Acknowledgement
• continuous problem
• discrete problem
|||u − uT ||| ≤
=⇒ convergence
• AMR algorithm
|||u − uT ||| ≤ Cr η
• possible avenues
• residual estimator
r = Le = L(u − uT ) = f − LuT
• duality estimator
kα−1/2 (σ T + α ∇uT ) k with ∇ · σ T = f
α(x) = αi > 0 in Ωi
• smoothness
u ∈ H 1+β (Ω)
• exact solution
a(uT , v) = f (v) ∀ v ∈ VT
• residual functional
2 2 1 X
2
ηK = hK kf + div (α∇uT )kK + he k[[ne · (α∇uT )]]k2e
2
e∈∂K
exact solution
2 2 2 1 X
ηK = hK kf + div (α∇uT )kK + he k[[ne · (α∇uT )]]k2e
2
e∈∂K
• ZZ estimators
ξG = kρT − ∇ uT k
where ρT ∈ UT ⊂ C 0 (Ω)d is a recovered gradient
• recovery operators
• Zienkiewicz & Zhu estimator (87, cited 2600 times)
Z
1
ρT (z) = ∇ uT dx ∀ z ∈ N
|ωz | ωz
kρT − ∇ uT k = min kγ − ∇ uT k
γ ∈ UT
• other recovery techniques
Bank-Xu, Carstensen, Schatz-Wahlbin, Z. Zhang, ...
• recovery-based estimators
ξZZ = kρT − ∇ uT k
• theory
ξZZ
k∇ u − ρT )k ≤ βk∇u − ∇ uT k =⇒ 1 − β ≤ ≤1+β
kρT − ∇ uk
+ simple
universal
asymptotically exact
/ C 0 (Ω)d
∇u ∈
• recovery space
ρT ∈ C 0 (Ω)d
σ = −α∇u ∈ H(div, Ω)
• conforming elements
uT ∈ H 1 (Ω) =⇒ ∇uT ∈ H(curl, Ω)
=⇒ σ̂ T ∈ RT0 or BDM1
σ̃ T = −α∇uT ∈
/ H(div, Ω)
+ simple
universal
asymptotically exact
• early work
• Ladevéze-Leguillon (83),
Demkowicz and Swierczek (85),
Oden, Demkowicz, Rachowicz, and Westermann (89)
• recent work
• Vejchodsky (04)
kA1/2 ∇ (u − uT )k ≤ η (τ ) ≡ kA−1/2 (τ − σ̃ T )k ∀ τ ∈ ΣN (f )
σ̂ BS ∈ ΣN (f ) ∩ RT 0
σ̂ BPS ∈ ΣN (f ) ∩ RT p−1
• p-robust efficiency
ηK (σ̂ BPS ) ≤ Ce kA1/2 ∇(u − uT )kωK
where Ce > 0 is a constant independent of p
ï1
10
ï2
10
1 2 3 4 5
10 10 10 10 10
number of nodes
≤ inf kA−1/2 (τ − σ̃ T )k
τ ∈ ΣN (f )
σ̂ CZ ∈ ΣN (f ) ∩ RT 0
σ̂ CZ ∈ ΣN (f ) ∩ RT p−1
1
10
||A1/2(¢ u ï¢ u ) ||
h 0
ï1
10
ï2
10
1 2 3 4 5
10 10 10 10 10
number of nodes
• minimization problem:
J(u) = min
1
J(v)
v∈HD (Ω)
1
where J(v) = (A∇v, ∇v) − f (v) is the energy functional
2
• dual problem:
J ∗ (σ) = max J ∗ (τ )
τ ∈ΣN (f )
∗ 1 −1
where J (τ ) = − (A τ , τ ) is the complimentary functional and
2
ΣN (f ) ≡ {τ ∈ HN (div; Ω) ∇ · τ = f }
kA1/2 ∇h (u − uT )k2
kA1/2 ∇h ek2 ≤ inf kA−1/2 (τ − σ̃ T )k2 + inf kA1/2 (∇v − ρ̃T )k2
τ ∈ΣN (f ) 1
v∈HD (Ω)
σ̂ T ∈ ΣN (f ) ∩ RT p−1
kA1/2 ∇h ek2 ≤ inf kA−1/2 (τ − σ̃ T )k2 + inf kA1/2 (∇v − ρ̃T )k2
τ ∈ΣN (f ) 1
v∈HD (Ω)
ησ,K = kA−1/2 (σ̂ T − σ̃ T )k0,K and ηρ,K = kA1/2 (ρ̂T − ρ̃T )k0,K
0
Energy error
10
−1
10
−2
10
||u−uh||A
N−0.5
eta
−3
10
0 1 2 3 4 5
10 10 10 10 10 10
0
Energy error
10
−1
10
−2
10
||u−uh||A
N−0.5
eta
−3
10
1 2 3 4 5
10 10 10 10 10
• estimators
• explicit residual: reasonable mesh, not efficient error control
• error control
• asymptotic meshes: some smooth problems
• complex systems
• computational difficulties
• oscillations
• interior/boundary layers
• interface singularities
• nonlinearity
• ???