Professional Documents
Culture Documents
Chap 11
Chap 11
Chap 11
Fig. 11.1. One-dimensional example with two piecewise quadratic functions; the
one on the left is not in H 1 , the one on the right is.
150 Chapter 11. H 1 -, H(curl)-, and H(div)-conforming spaces
Tz := {K ∈ Th | z ∈ ∂K}, (11.4a)
TE := {K ∈ Th | E ⊂ ∂K}, (11.4b)
TF := {K ∈ Th | F ⊂ ∂K}, (11.4c)
K ∈ Tz
Fig. 11.2. Left: vertex z, patch Tz composed of six mesh cells with one cell K ∈ Tz
highlighted in gray; Right: P2 Lagrange nodes in the same connectivity class for a
patch Tz composed of four mesh cells (drawn slightly apart).
11.1.3 Examples
Let (K, i) ∈ Th × N and set a = CK,i . Let (K ′ , i′ ) ∈ a, i.e., (K, i)R(K ′ , i′ ).
Consider a path (Kn , in )n∈{0: N } from (K, i) to (K ′ , i′ ), such that, for all
152 Chapter 11. H 1 -, H(curl)-, and H(div)-conforming spaces
n ∈ {1: N }, there is Fn ∈ Fh◦ and in ∈ NKn ,Fn with F = ∂Kn−1 ∩ ∂Kn and
in = χFn (in−1 ).
Proof. The fact that ϕa ∈ P x,b (Th ) results from (11.5), so that it remains to
verify that [[ϕa ]]xF = 0 for all F ∈ Fh◦ . Let F ∈ Fh◦ with F = ∂Kl ∩ ∂Kr . We
distinguish three cases.
(1) F ⊂ Da , so that both Kl and Kr are in Ta (this case is possible only if
card(a) ≥ 2). There are il ∈ NKl ,F and ir ∈ NKr ,F such that (Kl , il ) ∈ a,
(Kr , ir ) ∈ a, and ir = χlr (il ). We infer that
σKl ,i (ϕa|Kl ) − σKr ,χlr (i) (ϕa|Kr ) = δiil − δχlr (i)χlr (il ) = 0,
for all i ∈ NKl ,F , since χlr is a bijective map. Owing to Assumption 10.11,
we infer that [[ϕa ]]xF = 0.
(2) F ⊂ ∂Da , so that only one of the two adjacent cells is in Da . With-
out loss of generality, we assume that Kl ∈ Ta and Kr 6∈ Ta . Then,
(Kl , il ) ∈ a, but il 6∈ NKl ,F since otherwise Kr would be in Ta . As a re-
sult, σKl ,i (ϕa|Kl ) = σKl ,i (θK,il ) = 0 for all i ∈ NKl ,F . Owing to (10.9), we
x
infer that γK l ,F
(ϕa|Kl ) = 0. Moreover, since Kr 6∈ Ta , ϕa|Kr = 0, and (10.5)
x
implies that γK r ,F
(ϕa|Kr ) = 0. We conclude that [[ϕa ]]xF = 0.
(3) Otherwise, ϕa is zero in Kl ∪ Kr , so that [[ϕa ]]xF = 0 − 0 = 0. ⊓
⊔
Proposition 11.6 (Basis). {ϕa }a∈Ah is a basis of P x (Th ), and {σa }a∈Ah is
a basis of L(P x (Th ); R).
Fig. 11.3. Global shape functions in dimension d = 2: P1 (left) and P2 (center and
right) Lagrange finite elements.
Fig. 11.4. Global shape functions for the lowest-order Nédélec (left) and Raviart–
Thomas (right) elements. element in dimension 2.
for all v ∈ V x (D). This is the key property to analyze the approximation
properties of Ihx since it shows us that we can reason cell-wise exploiting
x
the local approximation properties of IK . It is convenient to introduce the
following spaces (see (10.1)):
for real numbers s > 0 and p ∈ [1, ∞], and to define the following mesh-
weighted semi-norms for t > 0:
! p1
X
t
|ℏ v|W s,p (Th ;Rq ) := htp p
K |v|K |W s,p (K;Rq ) ,
K∈Th
if p ∈ [1, ∞) and |ℏt v|W s,∞ (Th ;Rq ) := maxK∈Th htK |v|K |W s,∞ (K;Rq ) ; the semi-
norm |v|W s,p (Th ;Rq ) is defined as above for t = 0. We will use these mesh-
weighted semi-norms to bound the interpolation error estimates. Since hK ≤ h
for all K ∈ Th , we infer that |ℏt v|W s,p (Th ;Rq ) ≤ ht |v|W s,p (Th ;Rq ) .
Ihg Ihc Id Ib
❄ ❄ ❄h ❄h
∇ ✲ c ∇×✲ d ∇· ✲ b
P g (Th ) P (Th ) P (Th ) P (Th )
Proof. Combine Lemma 8.17 with (11.12). ⊓
⊔
158 Chapter 11. H 1 -, H(curl)-, and H(div)-conforming spaces
|v − IhL v|W m,p (Th ) ≤ c (|ℏk+1−m v|W k+1,p (Th ) + . . . + |ℏl−m v|W l,p (Th ) ),
(11.17)
for all m ∈ {0: k + 1}, and all v ∈ W l,p (Th ) ∩ V g (D).
The same error estimates hold for Ihg .
Remark 11.12 (Non-affine meshes). The estimates from Corollary 11.11
can also be extended to Lagrange interpolation on non-affine meshes, hinging
on Theorem 6.9 instead of Theorem 5.12, and adapting the notion of shape-
regularity accordingly. ⊓
⊔
Corollary 11.13 (H(curl)-conforming interpolation). Let (Th )h>0 be a
shape-regular sequence of affine matching meshes. Let k ≥ 0 be the degree of
the Nédélec finite element. Let p ∈ [1, ∞]. Let l be the smallest integer such
that l > p2 if p > 1 or l = 2 if p = 1. The following holds (uniformly in p):
(i) If l ≤ k + 1,
for all r ∈ {l: k + 1}, all m ∈ {0: r}, and all v ∈ W r,p (Th ) ∩ V c (D).
(ii) If l > k + 1, i.e., if l = 2, k = 0, and p ∈ [1, 2] (see Remark 8.16),
|v − Ihc v|W m,p (Th ) ≤ c (|ℏ1−m v|W 1,p (Th ) + |ℏ2−m v|W 2,p (Th ) ), (11.19)
|∇×(v − Ihc v)|W m,p (Th ) ≤ c |ℏr−m ∇×v|W r,p (Th ) , (11.20)
for all r ∈ {1: k+1}, all m ∈ {0: r}, and all v ∈ V c (D) with ∇×v ∈ W r,p (Th ).
Part III. Finite Element Spaces 159
for all r ∈ {1: k +1}, all m ∈ {0: r}, and all v ∈ W r,p (Th )∩V d (D). Moreover,
the following holds (uniformly in p):
|∇·(v − Ihd v)|W m,p (Th ) ≤ c |ℏr−m ∇·v|W r,p (Th ) , (11.22)
for all r ∈ {0: k + 1}, all m ∈ {0: r}, and all v ∈ V d (D) with ∇·v ∈ W r,p (Th ).
Since σa (v) = 0 for all v ∈ V0x (D) and all a ∈ A∂h (see Exercise 11.2), we infer
that
x
Ih0 (v) = Ihg (v), ∀v ∈ V0x (D). (11.26)
The same holds for the Lagrange interpolation operator Ih0 L
: V0g (D) −→
g L L g
P0 (Th ), so that Ih0 (v) = Ih (v) for all v ∈ V0 (D).
Let V̌0x (D) := {v ∈ V̌ x (D) | γ x (v) = 0} with V̌ x (D) defined by (11.14).
∇ ∇× ∇·
Then, the sequence V̌0g (D) −→ V̌0c (D) −→ V̌0d (D) −→ L1 (D) is exact, since
the tangential boundary trace of ∇f is zero if γ g (f ) = 0 and the normal
trace of ∇×g is zero if γ c (c) = 0. Since V̌0x (D) ⊂ V0x (Th ), the interpolation
x
operator Ih0 can act on V̌0x (D), and the following diagrams commute (with
g c d
Ih0 of order (k + 1), and Ih0 , Ih0 , Ihb of order k):
∇ ✲ c ∇× ✲ ∇· ✲ 1
V̌0g (D) V̌0 (D) V̌0d (D) L (D)
g c
Ih0 Ih0 Id Ib
❄ ❄ ❄h0 ❄h
∇ ✲ c ∇×✲ d ∇· ✲ b
P0g (Th ) P0 (Th ) P0 (Th ) P (Th )
L g c d
Finally, the interpolation operators Ih0 , Ih0 , Ih0 , and Ih0 enjoy the approx-
imation properties stated in Corollaries 11.11, 11.13, and 11.14, respectively,
provided the function to interpolate satisfies γ g (v) = 0, γ c (v) = 0, and
γ d (v) = 0, respectively.
Remark 11.19 (Enforcement on part of boundary). The above mate-
rial can be readily adapted if one wishes to enforce a zero trace on a part of
the boundary that corresponds to a subset of the boundary faces Fh∂ . ⊓
⊔
Proof. Let vh ∈ P g (Th ) and let K ∈ Th be such that hK ≤ δ2D . Fix x ∈ K. Set
C♯ (x) := {y ∈ C(x) | ky−xkℓ2 (R2 ) ≥ δ2D }. Let y be arbitrary in C♯ (x) with po-
Rr
lar coordinates (r, θ) with respect to x. Since vh (x) = vh (y)− 0 ∂ρ vh (ρ, θ) dρ,
Rh
we infer that |vh (x)|2 ≤ 2|vh (y)|2 + 2(I1 + I2 )2 , with I1 = 0 K ∂ρ vh (ρ, θ) dρ
Rr
and I2 = hK ∂ρ vh (ρ, θ) dρ. Concerning I1 , let B(x, hK ) be the ball of center
x and radius hK and set Tx := {K ′ ∈ Th | K ′ ∩ (B(x, hK ) ∩ C(x)) 6= ∅}.
Then, |I1 | ≤ hK maxK ′ ∈Tx k∇vh kL∞ (K ′ ) . Using the inverse inequality (9.12)
with p = ∞ and q = 2 and the fact that all the mesh cells in Tx have a size
equivalent to that of K owing to mesh regularity (see Proposition 11.1), we
infer that |I1 | ≤ c hK maxK ′ ∈Tx h−1
K ′ k∇vh kL2 (K ′ ) ≤ c k∇vh kL2 (B(x,hK )∩C(x)) .
Concerning I2 , we employ a Cauchy–Schwarz inequality to infer that
Z r 2 Z r
2 − 12 1 r
|I2 | = ρ ρ ∂ρ vh (ρ, θ) dρ ≤ ln
2 |∂ρ vh (ρ, θ)|2 ρ dρ,
hK hK hK
Exercises
Exercise 11.1 (NK,F ). Let (K, P, Σ) be a simplicial finite element satisfying
Assumption 10.10. Let i ∈ N . We say that the degree of freedom σK,i is of
order k ∈ {0: d + 1} if card({F ∈ FK | i ∈ NK,F }) = k.
x
(i) Prove that PK = +F ∈FK ker(γK,F |PK ) (non-direct sum of vector spaces)
if and only if all the degrees of freedom are of order ≤ d. (Hint: use
Exercise 10.4.)
(ii) Determine the order of the degrees of freedom for the canonical hybrid,
Nédélec, and Raviart–Thomas finite elements for d = 3.
for all p ∈ [1, ∞) and all v ∈ Lp (D). (Hint: use a density argument.)
Solution to exercises
Exercise 11.1 (NK,F ).
x
(i) Assume that PK = +F ∈FK ker(γK,F ) (we drop the restriction to PK for
x
simplicity). Let i ∈ N . Owing to Definition 10.8, ker(γK,F ) ⊂ ker(σK,i )
x
for all F ∈ FK such that i ∈ NK,F . Hence, +F ∈FK ,i∈NK,F ker(γK,F )⊂
ker(σK,i ). Reasoning by contradiction, if σK,i is of order (d + 1), then
PK ⊂ ker(σK,i ), which contradicts that σK,i (θK,i ) = 1. Conversely, as-
sume that all the degrees of freedom are of order ≤ d. Then, for all i ∈ N ,
there is F ∈ FK such that i 6∈ NK,F . Using Exercise 10.4, we infer that
x
θK,i ∈ ker(γK,F ). Since any function in PK can be written as a linear com-
x
bination of the functions θK,i , we conclude that PK = +F ∈FK ker(γK,F ).
(ii) The degrees of freedom involving evaluation at a vertex are of order d = 3,
those involving an integral on an edge are of order 2, those involving an
integral on a face are of order 1, and those involving an integral on K
are of order 0.
inf kv−vh kLp (D) ≤ kv−IhL (vǫ )kLp (D) ≤ kv−vǫ kLp (D) +kvǫ −IhL (vǫ )kLp (D) .
vh ∈P g (Th )
Owing to Corollary 11.11 with m = 0, we infer that the second term tends
to zero as h → 0. Hence, lim suph→0 (inf vh ∈P g (Th ) kv − vh kLp (D) ) ≤ ǫ, and the
conclusion follows since ǫ is arbitrary.
164 Chapter 11. H 1 -, H(curl)-, and H(div)-conforming spaces
Exercise 11.4 (Lagrange). For all K ∈ Th , since K is convex and (IhL v)|K
is affine, IhL v reaches its extrema at a vertex of K, where its value coincides
with that of v. This property fails for piecewise quadratic functions since the
basis functions can take negative values.
Exercise 11.5 (Hermite). Let {θbi }1≤i≤4 be the shape functions of the Her-
mite finite element on the reference interval [0, 1], see Exercise 1.2. Then,
b3 x−xi−1
θ if x ∈ [xi−1 , xi ],
x i −x i−1
ϕi,0 (x) = θb1 x−xi
if x ∈ [xi , xi+1 ],
xi+1 −xi
0 otherwise,
and
h i−1
b4 x−xi−1
θ if x ∈ [xi−1 , xi ],
xi −xi−1
ϕi,1 (x) = hi θb2 x−xi
if x ∈ [xi , xi+1 ],
xi+1 −xi
0 otherwise,
with hi = xi+1 − xi for all i ∈ {0: N }. Proceeding as in the proof of Proposi-
tion ?? shows that these global shape functions are linearly independent and
form a spanning set the whole space Hh . Finally, the bi-cubic Hermite rectan-
gular finite element can indeed be used to enforce C 1 -continuity. For instance,
consider p ∈ Q3 in dimension d = 2. On the face {x1 = 1}, the function
∂x1 p|{x1 =1} is in P3,1 as a function of x2 . Owing to the choice of degrees of
freedom, its values and x2 -derivatives are the same on both sides of the face.