• Keine Ergebnisse gefunden

Consistency of the truncated expansion

for allh < h0 andn∈Nh

ke(0)k2a,Ωh+

n−1

X

k=0

(2 + 1

h2)kη(k) +Υ(k)k2Vh≤h21

Te−C0T. (4.63) which allows us to apply Lemma 2 to the inequality (4.58). Hence

ke(n)ka,Ωh≤C

ke(0)ka,Ωh+hκ3−2+hκ1−2

≤C

hκ2+hκ3−2+hκ1−2 (4.64) and the result follows.

Again, we apply the general result of the theorem above to our specific predic-tion funcpredic-tion ˜f,

Theorem 19 (Convergence of f˜ on bounded domains)

Let f be the solution of a given lattice Boltzmann algorithm E3h which fulfills the assumptions of Theorem 15. Assume the prediction function f˜ satisfies Theorem 9 withκb>5/2, Then

kf˜−fkVh =O(hσ˜), σ˜ = min(κ−2, κ0, κb −3

2) =κb−3 2,

and there existh-independent constantsC >0andh0 >0so that for allh < h0 andn∈Nh,

kp(tn)−P(n)kh ≤Chmin(2,κ0−2,κb72), ku(tn)−U(n)kh≤Chmin(2,κ0−1,κb52)

for the moments (4.21) of the lattice Boltzmann solution, where p(tn),u(tn)∈ Gh are the restrictions of the solution p,u of (2.1), (2.3) to the grid.

Proof: Since [κ, κ0, κb+ 1/2]T ∈ordE3( ˜f), applying Theorem 18 to ˜f finally leads to this theorem by takingκ≥6,κ0 ≥2 and 1< κb32 < 32 into account..

4.5 Consistency of the truncated expansion

To assess the consistency order of ˜f to the lattice Boltzmann solution, we have to find the undetermined orders of ˜f inA, which are related to the undetermined orders of ˜f in A0. Recall that the slight difference of A0 and A lies on the additional condition ku(s)k∞,Ωh = O(h). Therefore there is some relation between Ms and M0s, which is described in the following two Lemmas and holds for ˜f.

Lemma 8 Let Emh be a lattice Boltzmann algorithm Emh and s ∈ A0. If k ∈ Ms(Emh) and k≥1, then also k∈ M0s(Emh).

Proof: Since there is a γ ∈ A satisfying 0 ≤ c0 ≤ kγkVh ≤ c1 such that s+hkγ∈ Aand

kE(i)h (s+hkγ)kVh =O(kE(i)h (s)kVh), 1≤i≤m.

this lemma follows if s+hkγ ∈ A0 is proved.

Since s ∈ A0, there is constant C such that ku(s)k∞,Ωh ≤ Ch. Using the triangle rule and ku(γ)kh≤ kγkVh, we find that

ku(s+hkγ)k∞,Ωh≤ ku(s)k∞,Ωh+hkku(γ)k∞,Ωh ≤Ch+hkc1 ≤max(C, c1)h.

In some particular cases, both sets are even equivalent.

Proposition 2 For a given lattice Boltzmann algorithm and some s∈ A0, we have M0s =Ms if infMs≥1.

Proof: This can be easily proved with the help of Lemma 8.

Since a direct checking of the element of Mf˜ is very difficult, we adopt a different strategy. To obtain an upper bound of the consistency order, we explicitly verify that a certain order is undetermined giving rise to a specific element of Mf˜. To achieve a lower bound, we use the estimated convergence order in connection with Theorem 5.

We introduce three numbers for the regular expansion ˜f. κ= sup

κ∈ordEm( ˜f)

κ1, κ0 = sup

κ∈ordEm( ˜f)

κ2, κb = sup

κ∈ordEm( ˜f)

κ3. In case that m= 2,κb is dropped out.

4.5.1 Periodic cases

Proposition 3 (Mf˜ for nonlinear periodic cases)

Provided that the conditions for the lattice Boltzmann scheme E2h and the pre-diction function f˜in Theorem 17 are satisfied, we find Mf˜=M0f˜.

Proof: First, we prove infMf˜ ≥ 1 by contradiction. Assume there is γ satisfying kγkVh≥c0>0 such that for any k <1(k∈R) hold

kE(1)h ( ˜f +hkγ)kVh=O(hκ1), κ1≥κ≥6, kE(2)h ( ˜f +hkγ)kVh=O(hκ2), κ2≥κ0.

4.5.1 Periodic cases 83 We denote E(1)h ( ˜f +hkγ) =η and E(2)h ( ˜f+hkγ) =ζ, write E(1)h in the explicit form and obtain forn, n+ 1∈Nh,

γ(0,j) =h−k(ζ −ˆr0)), (4.65) Sγ(n+ 1) = (I+JL)γ(n) +JQ(γ(n),γ(n))

+JQ(γ(n),2 ˜f(n)) +h−k(η(n)−ˆr(n)), (4.66) in which E(1)h ( ˜f) = ˆr and E(2)h ( ˜f) = ˆr0) have been inserted. Carrying out a similar estimate as in section 4.4.1 (by replacingewith γ), we get a recursion inequality forγ (in the same form of (4.56)),

kγ(n+ 1)k2a,Ωh ≤ kγ(n)k2a,Ωh+h2αkγ(n)k2a,Ωh

βkγ(n)k4a,Ωh+h−2k(2 + 1

h2)kη(n)−r(n)ˆ k2a,Ωh. (4.67) Further, we prove that there exists h0 > 0 so that for all h < h0 and for all n∈Nh

kγ(0)k2a,Ωh+

n−1

X

k=0

h−2k(2 + 1

h2)kη(k)−r(k)ˆ k2a,Ωh ≤h21

Te−C0T. (4.68) Note that

kη−rˆkVh =O(h6+hκ1),

kγ(0)kVh =kζ−ˆr0)ka,Ωh=h−kO(hκ0 +hκ2), the right hand side of (4.68) turns out to be

kγ(0)k2a,Ωh+

n−1

X

k=0

h−2k(2 + 1

h2)kη(k)−ˆr(k)k2a,Ωh

=h2O(h2(κ0−1−k)+h2(κ2−1−k)) +h2O(h6−2k+h1−6−2k) and is thus of sizeh2O(hmin(2(κ0−1−k),6−2k)). Sinceκ0 ≥α0+ 1≥2 and k <1, we can find someh0 >0 to guarantee the inequality (4.68). Now Lemma 2 can be applied to recursion inequality (4.67) ofγ, which yields

kγkVh≤Chκ0−k, nh2 ≤T Taking the limith−→0, we have limh→0kγkVh=0.

Hence we achieve (−∞, 1)∩ Mf˜=∅. According to Proposition 2,Mf˜=M0f˜.

We check the possible elements ofMfˆand find, Lemma 9 If κ0 6=∞, then κ0 ∈ Mf˜(E2h).

Proof: For the sake of convenience, let k=κ0 . Further choose γ=f 6= 0 to be a constant function. Consequently kγkVh is constant too.

We insert ˆf +hkγ into the update rule (2.20) and calculate the residue. Since the equilibrium function has a linear and a quadratic part, it turns out that

E(1)h ( ˆf +hkγ) = E(1)h ( ˆf) +hkA

fL(γ) +fQ(γ,2 ˆf +hkγ)−γ

. But the last term on the right hand side is zero, because fL(γ) = γ and h1,Vγi = 0 which cancels the term connected tofQ. Hence E(1)h ( ˆf +hkγ) = E(1)h ( ˆf).

Now, it follows that E(2)h ( ˜f +hkγ) = E(2)h ( ˜f) +hkγ. Further since

kE(2)h ( ˜f +hkγ)kVh ≤ kE(2)h ( ˜f)kVh+hkkγkVh, (4.69) thusordE2( ˜f)⊆ordE2( ˜f+hkγ). In summary, we have proved the theorem.

At this point, it is not clear whether (k < κ0 ) ∈ Mfˆ(E2h). However, from this theorem, we can conclude that ˆf may be consistent up to orderκ0 to the lattice Boltzmann scheme E2h.

For the linear lattice Boltzmann scheme E2h, we find that a standard conver-gence estimate is admitted in A , and combined with this theorem, we get the estimated convergence order of ˜f, which is a lower bound of the consistency order infMf˜.

For the nonlinear lattice Boltzmann scheme E2h, we obtain a standard conver-gence estimate only in the subspaceA0. The estimated convergence order of ˜f is supposed to be the lower bound of infM0f˜. However, from Proposition 3 it follows that infM0f˜= infMf˜.

Summarizing this discussion, we achieve a conclusion about consistency order in periodic cases.

Theorem 20 (Consistency order of f˜ in periodic domains)

Let the linear part JL of the collision operator in the lattice Boltzmann scheme E2h admit the stability condition (4.14) and the stability estimate (4.15). Fur-ther let the additional condition (4.34) holds for the nonlinear E2h. Assume f˜ satisfies the implications of Theorems 7, 8. Then

κ0 ∈ K0, (κ, κ0 )T∈ordE2( ˜f) (4.70) where

K0= (−∞, α0+ 1]∪ {α0+ 2, α0+ 3, . . . ,5} ∪ {∞}, (4.71) and the consistency order is in the interval,

˜

σ= min(κ−2, κ0 )≤O˜E2( ˜f)≤κ0 . (4.72)

4.5.1 Periodic cases 85 of the moments, the coefficients of ˆr0) are uniformly bounded. This implies that the possible values ofκ0can only be inK0. Henceκ0 ∈ K0 and (κ, κ0 )T∈ ordE2( ˜f) too.

Therefore we have a value ofκ00 . Since the estimated convergence order of f˜, which is min(κ−2, κ0) = min(κ−2, κ0 ) (Theorem 11), cannot be larger than the consistency order according to Theorem 5. On the other hand, Theorem 9 tells us ˜OE2( ˜f) = infMf˜is less or equal to κ0 . Thus (4.72) holds.

In addition, when κ−2 ≥ κ0 , we have min(κ−2, κ0 ) = κ0 . Hence (4.73) holds too.

Let us discuss several cases of (4.73).

In general, κ0 = α0 + 1 (refer to the discussion following equation (3.93)).

Combined withκ−2≥4≥α0+ 1, the convergence order and the consistency order are identical,α0+ 1. This implies that, in general, ˜f precisely represents the lattice Boltzmann solutions up to orderα0+ 1.

In particular cases, for example the case in Remark 7, both κ0 and κ0 are bigger than α0+ 1 when Eα00 = 1,2) is employed. Thus infMf˜ > α0+ 1.

When ψ(x) = 0 and ∂tG(0,x) = 0, ˆr(1)(0) = ˆr(2)(0) = ˆr(3)(0). Hence κ0 = 4 for all α0, and maybe κ0 = 4 since f(4)(0,xj) 6= 0 generally, which implies again the convergence order and the consistency order are identical, and the regular expansion ˜f correctly stands for the lattice Boltzmann solution up to order 4.

In even more particular cases like Poiseuille flow in a horizontal channel, ˆr(3) = 0 and κ0 is ∞ if Eα00 = 3) is applied. Therefore κ0 = ∞. Actually, in this case, the residue from the update rule ˆr(n,j) vanishes, since Poiseuille flow is a stationary, second order polynomial flow. Thereforeκ =∞, too. This implies that ˜f is the exact solution of the lattice Boltzmann method E2h.

4.5.2 Cases with the bounce back rule Lemma 10 If κ0 6=∞, then κ0 ∈ Mf˜(E3h).

Proof: Again we choose γ=f. Moreover,

E(3)h ( ˜f +hkγ) = E(3)h ( ˜f)−hkA(fL(γ)−γ+fQ(γ, hkγ+ 2 ˜f)). (4.75) The remainder of the proof is similar to the case in Theorem 9.

Recalling the relationship between the estimated convergence order and the consistency order, we can discuss the consistency of ˜f more precisely for the linear schemes on bounded domains.

Proposition 4 (Consistency order of f˜on bounded domains)

Let the collision operator JL in the linear lattice Boltzmann scheme E3h ad-mit the stability condition (4.14) and the stability estimate (4.15). Assume f˜ satisfies the implications of Theorem 7, 9. Then O˜E3( ˜f)∈[κb−3/2, κ0 ].

Proof: From Theorem 10, we knowκ0 ∈ Mf˜, thus infMf˜≤κ0 .

In addition ˜σ =κb−3/2≤infMf˜for the linear problem according to Theorem 5.

For the nonlinear problem E3h, we still can get a upper bound from Theorem 10, infMf˜≤κ0 . However,M0f˜=Mf˜is not obtained. Thus we only achieve κb−3/2 ≤infM0f˜ by applying Theorem 5. A lower bound for infMf˜ is not clear.