problem (4.1.1)-(4.1.3) and optimal order of convergence for the semidiscrete solution in L2(H1) norm is obtained.
4.2 The Continuous time Galerkin Approximation
which, combine with (4.2.3), leads to
A(Phv, φ) = (f∗, φ) = A(v, φ) ∀φ∈Vh, v ∈X∩Ho1(Ω). (4.2.4) Below, we present a proof that shows optimal error bounds for the projection operator Ph. This lemma is very crucial for our later analysis.
Lemma 4.2.1 Let Ph be defined by (4.2.4). Then there exists a positive constant C independent of h such that
kv−Phvk+hkv −Phvk1 ≤Ch2kvkX ∀v ∈X∩H01(Ω).
Proof. By the definition of f∗, we obtain
A(v, φ) = (f∗, φ) ∀φ∈H01(Ω). (4.2.5) For k = 1,2,define fk: Ω→R by
fk=
( f∗|Ωk in Ωk, 0 otherwise.
Clearly, fk ∈L2(Ω) and f∗ =f1+f2 a.e. in Ω. We now consider the following interface problems: Let wk ∈H01(Ω) be the solution of the interface problem
A(wk, φ) = (fk, φ) ∀φ∈H01(Ω). (4.2.6) Then by the coercivity of the operator A it follows immediately that v =w1+w2.
Let whk ∈Vh be the finite element approximation towk defined by
A(wkh, φ) = (fk, φ) ∀φ∈Vh. (4.2.7) Again by the coercivity of the operator A and definition (4.2.4) of the Ph operator, it follows that Phv =w1h+wh2.
Since fk|Ωs = 0, s = 1(2) if k = 2(1), we have for the elliptic interface problem (4.2.6)-(4.2.7) (cf. Theorem 2.3.1)
kwk−whkkH1(Ω) ≤ Ch(kwkkH2(Ω1)+kwkkH2(Ω2))
≤ ChkwkkX. (4.2.8)
Then by the elliptic regularity (cf. Theorem 1.2.1) we have kwkkX ≤ CkfkkL2(Ω)
≤ Ckf∗kL2(Ωk) ≤CkvkH2(Ωk). TH-261_BDEKA
This, together with (4.2.8), yields
kwk−wkhkH1(Ω) ≤ChkvkH2(Ωk). (4.2.9) Then
kv −PhvkH1(Ω) ≤ kw1−wh1kH1(Ω)+kw2−wh2kH1(Ω)
≤ C(hkw1kX +hkw2kX)
≤ Ch kvkH2(Ω1)+kvkH2(Ω2)
≤ ChkvkX ∀v ∈X∩H01(Ω). (4.2.10) For the L2 norm estimate, let us consider the following problem: Find w∈H01(Ω) such that
A(w, φ) = (v−Phv, φ) ∀φ ∈H01(Ω). (4.2.11) By setting φ = v − Phv ∈ H01(Ω) in (4.2.11) and using definition (4.2.4) of the Ph
operator, we have
kv−Phvk2L2(Ω) = A(w, v−Phv)
= A(w−Phw, v−Phv) +A(Phw, v−Phv)
= A(w−Phw, v−Phv)
≤ Ckw−PhwkH1(Ω)kv−PhvkH1(Ω)
≤ Ch2kvkXkwkX,
in the last inequality, we used (4.2.10). Then applying the elliptic regularity estimate (cf. Theorem 1.2.1) for the interface problem (4.2.11), we get
kv−Phvk2L2(Ω) ≤ Ch2kvkXkwkX
≤ Ch2kv−PhvkL2(Ω)kvkX.
This completes the proof of Lemma 4.2.1.
Let Lh :L2(Ω)→Vh be the standard L2 projection defined by
(Lhv, φ) = (v, φ) ∀v ∈L2(Ω), φ∈Vh. (4.2.12) It is well known that Lhv ∈Vh is the best approximation in the L2 norm tov ∈L2(Ω).
The following lemma shows that Lhv is a quasi-best approximation to v ∈ X∩H01(Ω) in the H1 norm.
Lemma 4.2.2 Let Ph and Lh be defined by (4.2.4) and (4.2.12), respectively. Then we have
kLhv−vkH1(Ω) ≤CkPhv−vkH1(Ω) ∀v ∈X∩H01(Ω).
TH-261_BDEKA
Proof. For any v ∈X∩H01(Ω), we know that there exists a unique solution w∈H01(Ω) for the elliptic interface problem
A(w, φ) = (Phv−v, φ) ∀φ∈H01(Ω). (4.2.13) Equation (4.2.13), together with (4.2.4) and Lemma 4.2.1, leads to
kPhv−vk2L2(Ω) = A(w−Phw, Phv−v) +A(Phw, Phv−v)
≤ Ckw−PhwkH1(Ω)kv−PhvkH1(Ω)
≤ ChkwkXkv−PhvkH1(Ω)
≤ Chkv−PhvkL2(Ω)kv−PhvkH1(Ω). (4.2.14) Here, we used the fact that kwkX ≤Ckv−PhvkL2(Ω). Use of the triangle inequality and (4.2.2) yields
kLhv−vkH1(Ω) ≤ kPhv −vkH1(Ω)+kLhv−PhvkH1(Ω)
≤ kPhv −vkH1(Ω)+Ch−1kLhv −PhvkL2(Ω)
≤ kPhv −vkH1(Ω)+Ch−1{kv −PhvkL2(Ω)
+kLhv−vkL2(Ω)}. (4.2.15)
We knowLhv is the best approximation ofv ∈L2(Ω) with respect to theL2 norm. Since v ∈X∩H01(Ω),Phv ∈Vh. Thus, kv−LhvkL2(Ω) ≤Ckv−PhvkL2(Ω). Therefore, (4.2.15) implies
kLhv−vkH1(Ω) ≤ kPhv−vkH1(Ω)+Ch−1kv−PhvkL2(Ω). (4.2.16) The desired estimate now follows from (4.2.14) and (4.2.16).
The following Lemma is crucial for our subsequent error analysis.
Lemma 4.2.3 Let Lh be defined by (4.2.12). Then we have kLhvkH1(Ω) ≤CkvkH1(Ω) ∀v ∈H01(Ω).
Proof. We define a projection Rh :H01(Ω)→Vh by
A(Rhv, vh) = A(v, vh) ∀vh ∈Vh, v ∈H01(Ω). (4.2.17) For anyv ∈H01(Ω), we know that there exists a unique solutionw∈X∩H01(Ω) for the elliptic interface problem
A(w, φ) = (Rhv−v, φ) ∀φ∈H01(Ω). (4.2.18) TH-261_BDEKA
Setting φ=Rhv −v in (4.2.18) and using Lemma 4.2.1, we obtain
kRhv −vk2L2(Ω) = A(w−Phw, Rhv−v) +A(Phw, Rhv −v)
≤ Ckw−PhwkH1(Ω)kv−RhvkH1(Ω)
≤ ChkwkXkv−RhvkH1(Ω)
≤ Chkv−RhvkL2(Ω)kv −RhvkH1(Ω). (4.2.19) Here, we used the fact that kwkX ≤Ckv−RhvkL2(Ω). This, together with (4.2.12), we have
kRhv−Lhvk2L2(Ω) = (Rhv−Lhv, Rhv−Lhv)
= (Rhv−v, Rhv−Lhv) + (v−Lhv, Rhv−Lhv)
≤ Ckv−RhvkL2(Ω)kRhv−LhvkL2(Ω)
≤ Chkv−RhvkH1(Ω)kLhv−RhvkL2(Ω). (4.2.20) Then, inverse inequality and (4.2.20), leads to
kLhvkH1(Ω) ≤ kLhv−RhvkH1(Ω)+kRhvkH1(Ω)
≤ C{h−1kLhv−RhvkL2(Ω)+kvkH1(Ω)}
≤ Ckv −RhvkH1(Ω)+CkvkH1(Ω)≤CkvkH1(Ω) ∀v ∈H01(Ω)
where, we used the fact that kRhvkH1(Ω) ≤ CkvkH1(Ω) ∀v ∈ H01(Ω), and this completes
the rest of the proof.
Now, we are in a position to discuss the continuous time Galerkin finite element approximation to (4.2.1) which may be stated as follows: Find uh(t)∈Vh such that
(uht, vh) +A(uh, vh) = (f, vh) +hg, vhiΓ ∀vh ∈Vh, t∈(0, T], (4.2.21) with uh(0) = Lhu0.
Subtracting (4.2.21) from (4.2.1) we have
(ut−uht, vh) +A(u−uh, vh) = 0 ∀vh ∈Vh. (4.2.22) Define the error e(t) as e(t) = u(t)−uh(t). Setting vh = Lhu in (4.2.22) and using (4.2.12), we obtain
1 2
d
dtkek2L2(Ω) + A(e, e) =A(u−uh, u−Lhu) + (ut−uht, u−Lhu)
= A(u−uh, u−Lhu) + (ut−Lhut, u−Lhu) +(Lhut−uht, u−Lhu)
= A(u−uh, u−Lhu) + (ut−(Lhu)t, u−Lhu) +(Lhut−uht, u−Lhu)
= A(u−uh, u−Lhu) + 1 2
d
dt(u−Lhu, u−Lhu). (4.2.23) TH-261_BDEKA
In the last equality we used the fact thatLhut−uht ∈Vhand the definition (4.2.12) of the Lh operator. Integrate the above equation from 0 tot. Then apply the Cauchy-Schwarz inequality and Young’s inequality to obtain
kek2L2(Ω) + Z t
0
kek2H1(Ω)ds
≤C Z t
0
ku−Lhuk2H1(Ω)ds+ku−Lhuk2L2(Ω)+ku0−Lhu0k2L2(Ω)
.
An application of Lemma 4.2.2 leads to kek2L2(Ω) +
Z t 0
kek2H1(Ω)ds
≤C Z t
0
ku−Phuk2H1(Ω)ds+ku−Lhuk2L2(Ω)+ku0−Lhu0k2L2(Ω)
,
which, together with Lemma 4.2.1 and the fact kLhv −vkL2(Ω) ≤ ChkvkH1(Ω) ∀v ∈ H01(Ω), yields
Z t 0
kek2H1(Ω)ds≤Ch2n
ku0k2H1(Ω)+kuk2X +kuk2L2(0,T;X)
o.
Thus, we have proved the following optimal H1 norm estimate.
Theorem 4.2.1 Let u and uh be the solutions of (4.1.1)-(4.1.3) and (4.2.21), respec- tively. Then, for u0 ∈H01(Ω), f ∈H1(0, T;L2(Ω)), and g ∈H1(0, T;H12(Γ)), we have
ku−uhkL2(0,T;H1(Ω)) ≤ Ch
ku0kH1(Ω)+kukX +kukL2(0,T;X) .
Next, for theL2 norm error estimate, we shall use the duality argument. For this purpose, we now consider the following auxiliary problem: Find z ∈H01(Ω) such that
A(z, v) = (u−uh, v) ∀v ∈H01(Ω), t ∈(0, T], (4.2.24) with
A∂∂z
n
= 0 across the interface Γ. Then its finite element approximation is defined to be a function zh ∈Vh satisfying
A(zh, vh) = (u−uh, vh) ∀vh ∈Vh, t∈(0, T]. (4.2.25) Setting v =u−uh in (4.2.24) and using (4.2.22), we obtain
ku−uhk2L2(Ω) = A(z, u−uh)
= A(z−zh, u−uh) +A(zh, u−uh)
= A(z−zh, u−uh)−(ut−uht, zh). (4.2.26) TH-261_BDEKA
Differentiating (4.2.25) with respect to t, we obtain A(zht, vh) = (ut−uht, vh).
Thus, we have
1 2
d
dtA(zh, zh) =A(zht, zh) = (ut−uht, zh), and hence, integrating (4.2.26) from 0 toT we obtain
ku−uhk2L2(0,T;L2(Ω))+1
2A(zh, zh)
≤C Z T
0
kz−zhkH1(Ω)ku−uhkH1(Ω)ds+1
2A(zh(0), zh(0)).
Further, using Theorem 1.2.1 for the elliptic interface problem (4.2.24) and Lemma 4.2.1, we obtain
kz−zhkH1(Ω) ≤ Ckz−PhzkH1(Ω)
≤ ChkzkX
≤ Chku−uhkL2(Ω), and hence
ku−uhk2L2(0,T;L2(Ω))
≤C Z T
0
hku−uhkL2(Ω)ku−uhkH1(Ω)ds+ 1
2A(zh(0), zh(0)). (4.2.27) Taking t→0, it now follows from (4.2.25) that
A(zh(0), zh(0)) = (u0−Lhu0, zh(0)) = 0.
This, together with (4.2.27) and Theorem 4.2.1, leads to ku−uhk2L2(0,T;L2(Ω))
≤Ch Z T
0
ku−uhk2L2(Ω)
12 Z T 0
ku−uhk2H1(Ω)
12
≤Ch2ku−uhkL2(0,T;L2(Ω))
ku0kH1(Ω)+kukX +kukL2(0,T;X) . Thus, we have proved the following optimal L2 norm estimate.
Theorem 4.2.2 Let u and uh be the solutions of (4.1.1)-(4.1.3) and (4.2.21), respec- tively. Then, for u0 ∈H01(Ω), f ∈H1(0, T;L2(Ω)), and g ∈H1(0, T;H12(Γ)), we have
ku−uhkL2(0,T;L2(Ω))≤Ch2
ku0kH1(Ω)+kukX +kukL2(0,T;X) . TH-261_BDEKA