Xi Fast Rodiak

Aproximación espectral para el operador
linealizado de una ecuación parabólica semilineal.1
Rodiak Nicolai Figueroa López
Departamento de Matemática
IBILCE/UNESP - São José do Rio Preto
XI Fast WorkShop on Applied and Computational Mathematics

04 y 05 de Enero de 2018
1
trabajo junto al Dr. German Lozada-Cruz (IBILCE/UNESP/SJRP)
Rodiak Nicolai Figueroa López Aproximación espectral para el operador linealizado
The semilinear parabolic problem
We consider the problem


ut = Lu + f (u), t > 0, x ∈ Ω

u(t, x) = 0, t > 0, x ∈ ∂Ω (1)

u(0, x) = u 0 (x),

where u 0 ∈ H01 (Ω), Ω ⊂ Rn is a bounded domain with smooth

boundary ∂Ω, n ≥ 2, L is second order elliptic operator given by
n n
X ∂ ∂u X ∂u
Lu = aij (x) + bj (x) + (c(x) + λ)u, (2)
∂xi ∂xj ∂xj
i,j=1 j=1
with coefficients aij , bj , c : Ω → R smooth, λ ∈ R and

f : R → R is a nonlinear function of class C 2 (R) satisfying
|f 0 (s)| ≤ C (1 + |s|γ−1 ), ∀s ∈ R, (C)

n+2
where 1 ≤ γ < n−2 if n ≥ 3, and γ ≥ 1 if n = 2, and
f (u)
lim sup ≤ 0. (D)
|u|→∞ u
We assume that L is uniformly strongly elliptic operator, that is,

there is a constant ϑ > 0 such that
n
X n
X
aij (x)ξi ξj ≥ ϑ ξk2 ,
i,j=1 k=1
for all x ∈ Ω, ξ = (ξ1 , ..., ξn ) ∈ Rn with aij = aji , ∀i, j = 1, ..., n.

Let X = L2 (Ω) be and define the linear operator A : D(A) → X by
Au = −Lu, for all u ∈ D(A) = H 2 (Ω) ∩ H01 (Ω).
Assume that λ is chosen s.t. Re σ(A) > 0. Then, X 1 = D(A),
X 0 = L2 (Ω) and X 1/2 = H01 (Ω).
With this notation, the problem (1) can be written in the abstract
form (
u̇ + Au = F (u)
(AP)
u(0) = u 0 ∈ X 1/2 ,
where F : X 1/2 → X is given by F (u(t))x = f (u(x, t)).

From condition (C), we obtain that F is locally Lipschitz continuous
in X 1/2 and continuously Fréchet differentiable.

As the operator A is sectorial can be associated with a sesquilinear
form σ(·, ·) : X 1/2 × X 1/2 → C such that
σ(u, v ) = hAu, v iX , u ∈ D(A), v ∈ X 1/2 , (3)

1/2
|σ(u, v )| 6 c1 kukX 1/2 kv kX 1/2 , u, v ∈ X (4)
Re σ(u, u) > c2 kuk2X 1/2 , u ∈ X 1/2 , (5)
where the constants c1 , c2 = ϑ/2 are positive.

We see that A(D(A)) = X and A−1 : X → X 1/2 is bounded. Hence
0 ∈ ρ(A) and A has compact resolvent in 0. Thus, A−1 ∈ K0 (X ).
Also, from the fact that A is sectorial, there are constants θ1 ∈
(0, π/2) and M1 > 0 such that
S0,θ1 = {z ∈ C : θ1 6 | arg(z)| 6 π, z 6= 0} ⊂ ρ(A) and (6)

M1
k(z − A)−1 kL (X ) 6 , ∀z ∈ S0,θ1 . (7)
|z|

Discretization of parabolic problems via FEM
In [2]: FUJITA, H.; MIZUTANI, A. On the finite element method
for parabolic equations I: approximation of holomorphic semi-
groups, 1976 (n = 2).
In [2] Figueroa-López, R. N.; Lozada-Cruz, G. J., Some estimates
for resolvent operators under the discretization by finite
element method, Comput. Appl. Math., 2014 (n > 2).
Definition 1
Let Ω ⊂ Rn be a given domain and let {T h }h∈(0,1] be a family of
subdivisions such that max{diam(T ) : T ∈ T h } ≤ h diam(Ω). The
family {T h }h∈(0,1] is said to be quasi-uniform if there exists ρ > 0
such that
min{diam(BT ) : T ∈ T h } ≥ ρh diam(Ω), ∀h ∈ (0, 1], (8)
where BT is the largest ball contained in T such that T is star-

shaped with respect to BT .
Definition 2
Ω is star-shaped with respect to a ball B if, for all x ∈ Ω, the closed
convex hull of {x} ∪ B is a subset of Ω.
Figura: Example of a T ∈ T h star-shaped with respect to ball BT in R2 .

Assumption 3
Let Ω ⊂ Rn be a polyhedral domain with n > 2 which has

{T h }h∈(0,1] a quasi-uniform family of subdivisions with positive
constant ρ and let (K , P, N ) be a reference element of class C 0
satisfying
(i) K is star-shaped with respect to some ball,
(ii) P1 ⊆ P ⊆ W2,∞ (K ) and
0
(iii) N ⊆ C (K ) .
With the Assumption 3 we can define the finite element space

1/2
:= {I h υ : υ ∈ C (Ω), υ ∂Ω = 0} ⊂ X 1/2 ∩ C (Ω),

Xh
1/2
where I h is the global interpolation operator. The space Xh is
finite dimensional.

1/2
We define the orthogonal projection Ph : X → Xh defined by
1/2
hPh g , v iX = hg , v iX , ∀v ∈ Xh . (9)
Hence, kPh g kX s 6 C kg kX s for all g ∈ X s with s = 0, 1/2 (see [2]).

1/2 1/2
The operator of aproximation Ah : Xh → Xh of A is define by
1/2
hAh φh , ψh i = σ(φh , ψh ), ∀φh , ψh ∈ Xh . (10)
1/2
The operator Ah is sectorial operator in Xh (see [3]).
With this in mind, we can establish the discretization of the problem
(AP) and write it in the form
(
u̇h + Ah uh = Fh (uh )
1/2 (APh )
uh (0) = uh0 ∈ Xh ,
1/2 1/2
where Fh := Ph F : Xh → Xh for all h ∈ (0, 1].

Relation between the operators A and Ah
Lemma 4
If (4) and (5) hold, then there is a positive constant C such that
kPh v − v kX s ≤ Ch2(1−s) kAv kX , ∀v ∈ X , s = 0, 1/2. (11)
Theorem 5 (Main result in [2])
If Assumption 3 holds, then there is a constant C > 0 and an acute

angle θ1 such that for any a ∈ X and z ∈ S0,θ1 , we have
k(z −A)−1 a−(z −Ah )−1 Ph akX s 6 Ch2(1−s) kakX , s = 0, 1/2. (12)
Lemma 6
If (4), (5) and Assumption 3 hold, then there is a C > 0 such that
kA−1 −1
h Ph v −Ph A v kX 6 Ch
s
2(1−s)
kv kX , ∀v ∈ X , s = 0, 1/2. (13)
We assume that (AP) has a equilibrium point u ∗ ∈ D(A) ⊂ X 1/2 ,
that is, u ∗ is a solution of
Au ∗ = F (u ∗ ). (14)
Denote by E the set of solutions of (14).

The family of equilibrium points {uh∗ }h∈(0,1] of the discrete problem
1/2
(APh ) is defined by, for every uh ∈ Xh that satisfies
Ah uh∗ = Ph F (uh∗ ), ∀h ∈ (0, 1]. (15)

1/2
As before, denote by Eh the set of solutions of (15) in Xh .
Under the previous assumptions on the nonlinearity F , the set Eh is
bounded in X 1/2 and L∞ (Ω), uniformly for h ∈ [0, 1].

Definition 7
We say that uh∗ ∈ Eh is hyperbolic if the spectrum σ(Ah − Fh0 (uh∗ ))
is disjoint from the imaginary axis, i.e., σ(Ah − Fh0 (uh∗ )) ∩ iR = ∅ for
all h ∈ [0, 1].
Proposition 8 ([1])
/ σ(A0 − F00 (u0∗ )) for all u0∗ ∈ E0 , then E0 is a finite set.

If 0 ∈
Moreover, if all point of E0 are hyperbolic then this have an odd
number of elements.

Linearization of the resolvent operator
Now we begin our discussion of problem (APh ) and its linearization

around uh∗ . We use the ideas in [3].
1/2 1/2
We introduce the finite element approximation Ãh : Xh → Xh
of the operator Ã := A − dI (where d(x) = f 0 (u ∗ (x)) with x ∈ Ω
and d ∈ L∞ (Ω)), defined by
1/2
hÃh φh , ψh iX = σ(φh , ψh ) − hPh dφh , ψh iX , ∀ φh , ψh ∈ Xh . (16)
Let α := sup f 0 (u ∗ (x)), so that the bilinear form

x∈Ω
σ̃(u, v ) = σ(u, v ) + h(α − d)u, v iX (17)
is coercive in X 1/2 , that is, σ̃(·, ·) satisfies the condition (5).

1/2
We also define the Ritz projection operator Rh : X 1/2 → Xh given
by
1/2
σ̃(Rh u − u, φh ) = 0, ∀φh ∈ Xh . (18)
Using the identity (18), (10), (3) and (9), we obtain
A˜h Rh = Ph [Ã − α(Rh − I )]. (19)
Under the Assumption 3, the standard error analysis for elliptic

problems yields
kRh u − ukX + hkRh u − ukX 1/2 6 Ch2 kukX 1 , ∀u ∈ X 1 . (20)
see [2, Theorem 4].

As the solutions belong to different spaces we need a tool that allows
1/2
us to compare the functions defined in the spaces Xh and X 1/2 .
This will be resolved using the projection Rh and we can definition
the R−convergence:
R h→0
uh −→ u0 if, and only if kuh − Rh ukX 1/2 −→ 0.

The important result
Theorem 9 ([3])
If Eh is the set of solutions of (15) for h ∈ (0, 1] and E is the set of

solutions of (14), then for any sequence uh∗ ∈ Eh with h → 0, there
are a subsequence, which also denoted by uh∗ , and a u ∗ ∈ E such
that
kuh∗ − Rh u ∗ kX 1/2 → 0 as h → 0. (21)
Moreover, there exists a positive constant C which is independent
of h such that
kuh∗ − Rh u ∗ kX 1/2 6 Ch. (22)
Also, For any hyperbolic equilibrium point u ∗ ∈ E, there are η > 0
and h0 > 0 s.t. there exists an (hyperbolic) equilibrium uh∗ of (15)
such that
kuh∗ − Rh u ∗ kX 1/2 6 η, ∀h ∈ (0, h0 ]. (23)
Furthermore, (21) and (22) is satisfied.
The main result
Theorem 10
Under the conditions of Theorem 9. Let uh∗ be a sequence of
equilibria of (15) and u ∗ be an equilibrium of (14) satisfying (21)
and (22).
If λj ∈ σ(A − F 0 (u ∗ )) and λhj ∈ σ(Ah − Ph F 0 (uh∗ )) for all j ∈ N,
then
h→0
λhj −→ λj , for j = 1, 2, · · · .
Also, if j ∈ N is such that Re λj < Re λj+1 and define Wj =
span[φ1 , . . . , φj ] and Wjh = span[φh1 , . . . , φhj ] where φj ∈ W (λj , A−
F 0 (u ∗ )) and φhj ∈ W (λhj , Ah − Ph F 0 (uh∗ )), then
h→0
distX 1/2 (Wjh , Rh Wj ) −→ 0. (24)

Theorem 11 (Continuation Theorem 10)
In particular, if u ∗ is an hyperbolic equilibrium point of (14) and uh∗
h→0
is a sequence of equilibria of (15) such that kuh∗ − Rh u ∗ kX 1/2 −→ 0
given by Theorem 9, then for h small enough, uh∗ is hyperbolic also
and its linearized unstable manifold converges, in the sense of (24),
to the linearized unstable manifold of u ∗ .
Furthermore, the dimension of the unstable manifold of uh∗ and u ∗
coincide.
In (24), distX 1/2 is the symmetric Hausdorff distance of V and W
are subspaces of the Banach space X 1/2 , then
distX 1/2 (V , W ) = sup inf kx − y kX 1/2 + sup inf kx − y kX 1/2 ,

x∈BV y ∈BW x∈BW y ∈BV
where BV and BW are the unit balls of V and W , respectively.

The convergence of linearized unstable manifolds is a first step
needed to prove the convergence of attractors (see [1]).

Convergence compact
Let {Xh }h∈[0,1] be a family of Banach spaces and P = {Ph }h∈(0,1]
be a family with Ph ∈ L (X0 , Xh ), for h ∈ (0, 1], with the following
property:
h→0
kPh u0 kXh −→ ku0 kX0 , ∀u0 ∈ X0 . (25)
Definition 12
A family {uh }h∈(0,1] with uh ∈ Xh P−converges to u0 ∈ X0 if
kuh − Ph u0 kXh → 0 as h → 0.
P
We write this as uh −→ u0 .
Similarly, the P−convergence of sequences is defined as: A
n→∞
sequence {uhn }n∈N with uhn ∈ Xhn such that hn −→ 0
P−converges to u0 ∈ X0 if
kuhn − Phn u0 kXhn → 0 as n → ∞.

Definition 13
An operator Ah : Xh → Yh PQ−converges to an operator A0 :
Q P
X0 → Y0 as h → 0 if Ah uh −→ A0 u0 whenever uh −→ u0 . We
PQ
write Ah −→ A0 as h → 0.
Definition 14
n→∞
A sequence {un }n∈N , with un ∈ Xhn and hn −→ 0, is P−relatively
compact if for any subsequence {un0 } de {un }n∈N , there are a
P
subsequence {un00 } of {un0 } and an u0 ∈ X0 such that uhn00 −→ u0 .
Definition 15
An operator Ah ∈ K0 (Xh , Yh ) converges compactly to an operator
A0 ∈ K0 (X0 , Y0 ) if for any {uh }h∈(0,1] with uh ∈ Xh s.t. kuh kXh =
1, for all h ∈ (0, 1], the family {Ah uh }h∈(0,1] is Q−relatively
PQ CC
compact and Ah −→ A0 . We denote Ah −→ A0 as h → 0.
Assume that {Zh }h∈[0,1] is a family of Banach spaces and M =
{Mh }h∈(0,1] a family of bounded linear operators Mh : Z0 → Zh
h→0
satisfying kMh v0 kZh −→ kv0 kZ0 , for all v0 ∈ Z0 .
The diagram shows the relationship between operators Ah and Bh :
A0 B0
X0 / Y0 / Z0
Ph Qh Mh
Ah Bh
Xh / Yh / Zh
The result below is very useful to show the compact convergence

operators composition.
Lemma 16 ([3])
CC QP
If K0 (Yh , Zh ) 3 Bh −→ B0 ∈ K0 (Y0 , Z0 ) and L (Xh , Yh ) 3 Ah −→
CC
A0 ∈ L (X0 , Y0 ), then Bh Ah −→ B0 A0 .
Assume Ah : D(Ah ) ⊂ Xh → Xh and Fh : Xh → Xh for all h ∈ [0, 1],
satisfying
CC
Ah close,compact resolvent, 0 ∈ ρ(Ah ), and A−1 −1
h −→ A0 . (Hh )
Remark 17
If λ0 ∈ σ(A0 ) is an isolated point then we associated the generalized
eigenspace
[ [
W (λ0 , A0 ) := u ∈ X0 : (λ0 −A0 )k u = 0 = N (λ0 −A0 )k .

k>1 k>1
Then, since A0 is a closed operator, we have that dim(W (λ0 , A0 )) <

∞.
CC
Ah satisfy (Hh ) and A−1 −1
h Fh −→ A0 F0 . (Jh )

The following result says that the spectrum of Ah , for h small,
approaches to spectrum of A0 .
Theorem 18
Assume that the family of linear operators {Ah }h∈[0,1] satisfy (Hh ).
Then following statements hold:
(i) If λ0 ∈ σ(A0 ), there exists a sequence {hn }n∈N ⊂ (0, 1] with
n→∞ n→∞
hn −→ 0 and λn ∈ σ(Ahn ), n ∈ N, such that λn −→ λ0 .
n→∞
(ii) If {hn }n∈N ⊂ (0, 1] with hn −→ 0 and λn ∈ σ(Ahn ), n ∈ N
n→∞
and λn −→ λ0 , then λ0 ∈ σ(A0 ).

Theorem 19 (Continuation Theorem 18)
Furthermore, if there h0 ∈ (0, 1] s.t. sup kPh kL (X0 ,Xh ) < ∞,
h∈[0,h0 ]
Then following statements hold:
(iii) There exists h1 ∈ (0, h0 ) such that
dim(W (λ0 , Ah )) = dim(W (λ0 , A0 )), ∀h ∈ [0, h1 ].
(iv) If u ∈ W (λ0 , A0 ), there exists a sequence {hn }n∈N ⊂ (0, 1]

n→∞ P
with hn −→ 0, uhn ∈ W (λ0 , Ahn ), such that uhn −→ u0 as
n → ∞.
n→∞
(v) If {hn }n∈N ⊂ (0, 1] with hn −→ 0, and un ∈ W (λ0 , Ahn ),
kun kXhn = 1, then {un }n∈N has an P−convergent
subsequence and a limit point u0 ∈ W (λ0 , A0 ) with
ku0 kX0 = 1.
Proof. See [1, Theorem 4.10].

Linearization
Let us study the behavior of operators of the form Ah + Vh where

Vh : Xh → Xh is a bounded linear operator.
Consider the following hypothesis
CC
(Hh ) holds,Vh ∈ L (Xh ), h ∈ [0, 1] s.t. A−1 −1
h Vh −→ A0 V0 . (Gh )
Furthermore, we assume that
0 6∈ σ(A0 + V0 ). (E0 )
It is clear that A0 + V0 has compact resolvent. Let Āh = Ah + Vh ,

h ∈ [0, 1].

Proposition 20 ([1])
Assume that (Gh ) and (E0 ) are satisfied. Then, there is h0 > 0 such
that 0 6∈ σ(Ah +Vh ) for all h ∈ (0, h0 ], sup k(Ah +Vh )−1 kL (Xh ) <
h∈(0,h0 ]
∞. Moreover,
CC
(Ah + Vh )−1 −→ (A0 + V0 )−1 as h → 0.
In particular, the operators Āh = Ah +Vh , h ∈ [0, 1], satisfy condition

(Hh ).
Corollary 21 ([1])
Under the conditions of Proposition 20, all the results of Theorem

18, remain valid for the family of operators Āh = Ah + Vh , with
h ∈ [0, 1].

Define
wh (uh∗ , v ) := A−1 ∗ −1 ∗ −1 0 ∗
h Fh (uh + v ) − Ah Fh (uh ) − Ah Fh (uh )v .
Consider the following hypothesis
P
(Jh ) holds, if uh∗ ∈ Eh with uh∗ −→ u0∗ , then
CC
A−1 0 ∗ −1 0 ∗
h Fh (uh ) −→ A0 F0 (u0 ) and (Lh )
∗
kwh (uh , vh )kXh
= o(1) as kvh kXh → 0, uniformly in h.
kvh kXh
kwh (uh∗ , vh )kXh

Observe that = o(1) uniformly in h means that
kvh kXh
for each µ > 0, there exists a δ > 0 such that kwh (uh∗ , vh )kXh 6
1/2
µkvh kXh for all vh ∈ Xh with kvh kXh 6 δ.

Theorem 22
Assume (Lh ) holds and let u0∗ ∈ E0 such that 0 ∈ / σ(A0 − F00 (u0∗ )).
Then, there is a δ0 > 0 so that there is a unique uh ∈ Eh ∩Bδ0 (Rh u0∗ ).
∗
Furthermore, if u0∗ ∈ E0 is such that 0 ∈ / σ(A0 − F00 (u0∗ )), then,

from Proposition 8, there are a finite odd number r of elements in
∗ , . . . , u ∗ }. In this case, there is a h > 0 such that E
E0 = {u0,1 0,r 0 h
∗ , . . . , u ∗ , for all h 6 h and u ∗ −→ u ∗ P
has exactly r elements uh,1 h,r 0 h,i 0,i
for 1 6 i 6 r .
If, moreover, u0∗ is hyperbolic then uh∗ is also hyperbolic and we can
apply Corollary 21.
In particular, the linear unstable manifold of uh∗ P−converge to the
linear unstable manifold of u0∗ .
Proof. See [1, Theorem 5.8].

Proof of Theorem 10.
CC
The following result prove that A−1 −1
h −→ A .
Lemma 23
1/2 1/2
The operator A−1h : Xh → Xh , h ∈ (0, 1], compactly converges
to the operator A−1 : X → D(A) ⊂ X 1/2 . Furthermore,
kAh−1 Ph u −Rh A−1 ukX s 6 Ch2(1−s) kukX , ∀u ∈ X , s = 0, 1/2. (26)
Proof. First, the family {A−1h }h∈(0,1] consists of compact linear

operators since all of them have finite rank. Given u ∈ X , from
Lemma 6, [2, Lemmas 2 and 5] and (20), we get
kAh−1 Ph u − Rh A−1 ukX s 6 Ch2(1−s) kukX + kPh (I − Rh )A−1 ukX s
6 Ch2(1−s) kukX + Ch2(1−s) kA−1 ukX 1
6 Ch2(1−s) kukX .
PR
Hence, A−1 −1
h −→ A .
1/2
Let {uh }h∈(0,1] be a sequence with uh ∈ Xh and kuh kX 1/2 = 1.
Since (z − A)−1 is compact for all z ∈ ρ(A) then µ((z − A)−1 v )=0,
for all v ∈ X and where µ(·) is the measure of noncompactness.
From inequality (12), it follows
µ((z − Ah )−1 uh ) 6 µ((z − A)−1 uh ) + µ((z − A)−1 uh − (z − Ah )−1 uh )

6 lim sup k(z − A)−1 uh − (z − Ah )−1 Ph uh kX 1/2
h→0
= 0,
where (z − Ah )−1 uh is bounded. With this, we have shown that

{A−1
h uh }h∈(0,1] is R−relatively compact.
CC
Therefore, A−1 −1
h −→ A .
Corollary 24
1/2 1/2
The operator Ah : Xh → Xh satisfy the condition (Hh ).

The following diagram shows the relationship between P−convergence
and R−convergence.
F /X A−1 / D(A) ⊂ X 1/2

X 1/2
Rh Ph Rh

Fh A−1
1/2
Xh / X 1/2 h
/ X 1/2
h h

P R
The following results tell us that Fh (uh∗ ) −→ F (u ∗ ) as uh∗ −→ u ∗ .
Corollary 25
Under the conditions of Proposition 9, we have
kFh (uh∗ ) − Ph F (u ∗ )kX 6 C (R)h. (27)
Furthermore, the condition (Jh ) is satisfied.
Proof. From Proposition 9 and inequalities (20), we obtain
kFh (uh∗ ) − Ph F (u ∗ )kX 6 kF (uh∗ ) − F (Rh u ∗ )kX + kF (Rh u ∗ ) − F (u ∗ )kX

6 C (R)(kuh∗ − Rh u ∗ kX 1/2 + kRh u ∗ − u ∗ kX 1/2 )
6 C (R)(Ch + ChkAu ∗ kX )
6 C (R)h.
Finally, the condition (Jh ) holds by Lemma 23 and (27).

Denote Vh = −Ph F 0 (uh∗ ) and V0 = −F 0 (u ∗ ).
P R
The following results tell us that Vh −→ V0 as uh∗ −→ u ∗ .
Corollary 26
Under the conditions of Proposition 9, we have
kVh − Ph V0 kL (X 1/2 ,X ) 6 Ce(R)h. (28)
Proof. Using Proposition 9, we obtain
kVh − Ph V0 kL (X 1/2 ,X )
6 kVh − Ph F 0 (Rh u ∗ )kL (X 1/2 ,X ) + kPh F 0 (Rh u ∗ ) − Ph V0 kL (X 1/2 ,X )
6 Ce(R)(kuh∗ − Rh u ∗ kX 1/2 + kRh u ∗ − u ∗ kX 1/2 ) 6 Ce(R)h.
Therefore, the inequality (28) is shown.

Remark 27
We stated that
CC
A−1 −1
h Vh −→ A V0 .
RP
In fact, we can see that Vh −→ V0 since
kVh φh − Ph V0 φkX
6 kVh (φh − Rh φ)kX + k(Vh − Ph V0 )Rh φkX + kV0 (Rh φ − φ)kX
6 Ce(R)(kφh − Rh φk 1/2 + hkRh φk 1/2 ) + kV0 kL∞ kRh φ − φkX
X X
6 Ce(R)(kφh − Rh φkX 1/2 + ChkφkX 1/2 ) + ChkφkX 1/2 ,
R
where φh −→ φ and we have used Corollary 26 and (20). Now, the
result follow using Lemmas 23 and 16.

Remark 28 (Continuation Remark 27)
Also we affirm that, for each h ∈ [0, 1], we get
1/2
kwh (uh∗ , vh )kX 1/2 6 C (δ)kvh k2X 1/2 , vh ∈ Xh with kvh kX 1/2 6 δ,
where C is a constant independent of h. We note that there exists

a constant C , independent of h, such that
kA−1
h Ph φkX 1/2 6 C kφkX , ∀φ ∈ X .
1/2
Hence, given vh ∈ Xh with kvh kX 1/2 6 δ, we have
kwh (uh∗ , vh )kX 1/2 6 C kF (uh∗ + v ) − F (uh∗ ) − F 0 (uh∗ )v kX

6 C kF 0 (uh∗ + θvh ) − F 0 (uh∗ )kL (X ,X 1/2 ) kvh kX 1/2
6 C Ce(δ)kvh k2X 1/2 ,
where θ : Ω → [0, 1] is a measurable function.

Proof of Theorem 10. From Theorem 9 there is a sequence of
equilibria uh∗ which R−converges to u ∗ satisfying (22).
Defining Vh = −Fh0 (uh∗ ) + αPh and V0 = −F 0 (u ∗ ) + αI , so that
−f 0 (u ∗ (x)) + α > 0, then (Gh ) holds. In fact, it is easy to see that
RP
Vh ∈ L (X 1/2 , X ) and Vh −→ V0 (by Remark 27).
Moreover, since −f 0 (u ∗ (x)) + α > 0 along with the hypothesis that
Re σ(A) > 0, we have that (E0 ) is satisfied.
With this, we are in a position to use Proposition 20 and say that
Ah + Vh and A + V0 satisfy (Hh ) so that by Theorem 18 we obtain
the spectral convergence of Ah + Vh to A + V0 . Using a similar
argument made in [1] where α is just a shift in the spectrum for the
operators above, we show that the results of Theorem 18 are valid
for operators Ah −Ph F 0 (uh∗ ) and A−F 0 (u ∗ ). In particular, we obtain
the convergence of the eigenvalues and the convergence of spectral
projections in X 1/2 .
The condition (Lh ) follows from Corollary 25 and Remark 27. Then
the Theorem 22 holds.

Gracias !!!!!!!!!!!!!

References
Arrieta, J.M.; Carvalho, A. N.; Lozada-Cruz, G., Dynamics in

dumbbell domains. I. Continuity of the set of equilibria.
Journal Differential Equations 231 (2006), n. 2, 551–597.
Arrieta, J.; Carvalho, A.; Rodrı́guez-Bernal, A. Attractors of
parabolic problems with nonlinear boundary conditions.
Uniform bounds. Comm. PDE 25 (2000), n. 1-2, p. 1–37.
Brenner, S. C.; Scott, L. R. The mathematical theory of
finite element methods. New York: Springer-Verlag, 1996.
Burenkov, V. I. Sobolev Spaces on Domains. Leipzig:
Teubner-Texte zur Mathematik, volume 137, 1998.
Carvalho, A.N., Piskarev, S., A general approximations
scheme for attractors of abstract parabolic problems,
Numer. Funct. Anal. Optim. 27 (2006), n. 7-8, p. 785–829.

Cholewa, J. W.; Dlotko, T. Global Attractors in Abstract
Parabolic Problems. Cambridge: University Press, 2000.
Fujita, H.; Mizutani, A. On the finite element method for
parabolic equations I: approximation of holomorphic semi-
groups. Journal of the Mathematical Society of Japan,
Tokyo, v. 28, n. 4, p. 749–771, 1976.
S. Larsson, J. Sanz-Serna. The behavior of finite element
solutions of semilinear parabolic problems near stationary
points, SIAM Journal on Numerical Analysis, v. 31, n. 4, p.
1000–1018, 1994.
Henry, D. Geometric theory of semilinear parabolic
equations, Berlin: Springer-Verlag, 1981.
Kato, T., Perturbation theory for linear operators, Berlin:
Springer-Verlag, 1980.

Figueroa-López, R.; Lozada-Cruz, G., On global attractors
for a class of parabolic problems. Applied Mathematics &
Information Sciences, v. 8, n. 2, p. 493–500, 2014.
Figueroa-López, R. N.; Lozada-Cruz, G. J., Some estimates
for resolvent operators under the discretization by finite
element method, Comput. Appl. Math., v. 34, n.3, p. 1105–
1116, 2014.
Figueroa-López, R. N.; Lozada-Cruz, G. J., Dynamics of
parabolic equations via the finite element method I.
Continuity of the set of equilibria. Journal of Differential
Equations, v. 261, n. 9, p. 5235–5259, 2016.
Figueroa-López, R. N.; Lozada-Cruz, G. J., Dynamics of
parabolic equations via the finite element method II. The
discretized problem. In preparation.

Vainikko, G., Approximative methods for nonlinear equa-
tions (Two approaches to the convergence problem).
Nonlinear Analysis, Theory, Melhods and Applications. v. 2, n.
6, p. 647–687.
Vainikko G., Multidimensional weakly singular integral
equations, Berlin: Springer-Verlag, 1993.
Vainikko G. Funktionalanalysis der Diskretisierungsmetho-
den, Teubner-Texte zur Mathematik, Verlagsgesellschaft,
Leipzig, 1976.
Vainikko, G., Regular convergence of operators and
approximate solution of equations. Translated from Itogi
Nauki i Tekhniki, Seriya Matematicheskii Analiz, Vol. 16, 1979,
pp. 5-53.
Wolf-Jürgen, B.; Piskarev, S., Shadowing for discrete
approximations of abstract parabolic equations. Discrete
Contin. Dyn. Syst. Ser. B 10 (2008), n. 1, pp. 19-42.

Xi Fast Rodiak

Uploaded by

Copyright:

Available Formats

Xi Fast Rodiak

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Xi Fast Rodiak

Uploaded by

Copyright:

Available Formats

Aproximación espectral para el operador

linealizado de una ecuación parabólica semilineal.1

Rodiak Nicolai Figueroa López

XI Fast WorkShop on Applied and Computational Mathematics

We consider the problem

where u 0 ∈ H01 (Ω), Ω ⊂ Rn is a bounded domain with smooth

with coefficients aij , bj , c : Ω → R smooth, λ ∈ R and

Rodiak Nicolai Figueroa López Aproximación espectral para el operador linealizado

|f 0 (s)| ≤ C (1 + |s|γ−1 ), ∀s ∈ R, (C)

We assume that L is uniformly strongly elliptic operator, that is,

for all x ∈ Ω, ξ = (ξ1 , ..., ξn ) ∈ Rn with aij = aji , ∀i, j = 1, ..., n.

Rodiak Nicolai Figueroa López Aproximación espectral para el operador linealizado

where F : X 1/2 → X is given by F (u(t))x = f (u(x, t)).

Rodiak Nicolai Figueroa López Aproximación espectral para el operador linealizado

σ(u, v ) = hAu, v iX , u ∈ D(A), v ∈ X 1/2 , (3)

where the constants c1 , c2 = ϑ/2 are positive.

S0,θ1 = {z ∈ C : θ1 6 | arg(z)| 6 π, z 6= 0} ⊂ ρ(A) and (6)

Rodiak Nicolai Figueroa López Aproximación espectral para el operador linealizado

min{diam(BT ) : T ∈ T h } ≥ ρh diam(Ω), ∀h ∈ (0, 1], (8)

where BT is the largest ball contained in T such that T is star-

Figura: Example of a T ∈ T h star-shaped with respect to ball BT in R2 .

Rodiak Nicolai Figueroa López Aproximación espectral para el operador linealizado

Let Ω ⊂ Rn be a polyhedral domain with n > 2 which has

With the Assumption 3 we can define the finite element space

Rodiak Nicolai Figueroa López Aproximación espectral para el operador linealizado

Hence, kPh g kX s 6 C kg kX s for all g ∈ X s with s = 0, 1/2 (see [2]).

Rodiak Nicolai Figueroa López Aproximación espectral para el operador linealizado

kPh v − v kX s ≤ Ch2(1−s) kAv kX , ∀v ∈ X , s = 0, 1/2. (11)

Theorem 5 (Main result in [2])

If Assumption 3 holds, then there is a constant C > 0 and an acute

Denote by E the set of solutions of (14).

Ah uh∗ = Ph F (uh∗ ), ∀h ∈ (0, 1]. (15)

Rodiak Nicolai Figueroa López Aproximación espectral para el operador linealizado

/ σ(A0 − F00 (u0∗ )) for all u0∗ ∈ E0 , then E0 is a finite set.

Rodiak Nicolai Figueroa López Aproximación espectral para el operador linealizado

Now we begin our discussion of problem (APh ) and its linearization

Let α := sup f 0 (u ∗ (x)), so that the bilinear form

σ̃(u, v ) = σ(u, v ) + h(α − d)u, v iX (17)

is coercive in X 1/2 , that is, σ̃(·, ·) satisfies the condition (5).

Rodiak Nicolai Figueroa López Aproximación espectral para el operador linealizado

A˜h Rh = Ph [Ã − α(Rh − I )]. (19)

Under the Assumption 3, the standard error analysis for elliptic

kRh u − ukX + hkRh u − ukX 1/2 6 Ch2 kukX 1 , ∀u ∈ X 1 . (20)

see [2, Theorem 4].

Rodiak Nicolai Figueroa López Aproximación espectral para el operador linealizado

If Eh is the set of solutions of (15) for h ∈ (0, 1] and E is the set of

Rodiak Nicolai Figueroa López Aproximación espectral para el operador linealizado

distX 1/2 (V , W ) = sup inf kx − y kX 1/2 + sup inf kx − y kX 1/2 ,

where BV and BW are the unit balls of V and W , respectively.

Rodiak Nicolai Figueroa López Aproximación espectral para el operador linealizado

kuhn − Phn u0 kXhn → 0 as n → ∞.

Rodiak Nicolai Figueroa López Aproximación espectral para el operador linealizado

The result below is very useful to show the compact convergence

Then, since A0 is a closed operator, we have that dim(W (λ0 , A0 )) <

Rodiak Nicolai Figueroa López Aproximación espectral para el operador linealizado

Rodiak Nicolai Figueroa López Aproximación espectral para el operador linealizado

dim(W (λ0 , Ah )) = dim(W (λ0 , A0 )), ∀h ∈ [0, h1 ].

(iv) If u ∈ W (λ0 , A0 ), there exists a sequence {hn }n∈N ⊂ (0, 1]

Proof. See [1, Theorem 4.10].