5.3 Fundamental studies
5.3.1 Convergence analysis
Consider equation (5.1) with the convective coefficients p. q and r and the reactive coefficient eas constants. To perform the convergence analysis of the schemes assume a= 1 subject to the initial condition
φ(x, y, z,0) =φ0(x, y, z), (x, y, z)∈Ω∪∂Ω (5.18) and with the boundary condition
φ(x, y, z, t) = constant, (x, y, z)∈∂Ω, t >0, (5.19) where φ0(x, y, z) is a known smooth function and ∂Ω is the boundary of the finite domain Ω.
Thence as discussed in the previous section, our finite difference scheme for this problem is the one given by (5.8) but with constant coefficients.
Lemma 5.3.1. There exists a lower bound C to all the coefficients appearing in equation (5.8) at all grid points, where C is a real number.
Equivalently, there exists a real constant C such that C .V, where V may be any of the coefficients appearing in equation (5.8).
5.3. Fundamental studies 103
Proof. Since all of the coefficients A, B, C, P, Q, R, D, E, F, H, K, L, M, N, O,S, T,U, G and W, appearing in equation (5.8) are real numbers, they have finite values, so the proof of the above lemma follows.
Next, let us introduce some useful notations.
For the sake of our discussions in the current section, let us assume that h =
△x =△y = △z and τ = △t.
Denote the grid points as (xi, yj, zk), where xi = xa + ih, yj = ya + jh, zk
=za +kh and [xa, xb] × [ya,yb] ×[za, zb] is the domain under consideration with the index sets given as
Ix = {i|i = 0,1,2, . . . , I}, Ixo = {i|i = 1,2, . . . , I −1}, where I = (xb − xa)/h+ 1,
Iy ={j|j = 0,1,2, . . . , J},
Iyo ={j|j = 1,2, . . . , J−1}, where J = (yb−ya)/h+ 1, Iz ={k|k= 0,1,2, . . . , K},
Izo ={k|k= 1,2, . . . , K−1}, where K = (zb−za)/h+ 1.
For grid functions of the formφ ={φnijk|(i, j, k)∈Ix×Iy×Iz, n = 0,1,2, . . . , N} (N is the total number of time steps), we have the following finite difference operators:
δxφnijk = 2h1 (φni+1jk−φni−1jk), i6=I;
δx2φnijk = 1
h2(φni+1jk−2φnijk+φni−1jk), i∈Ixo; δyφnijk = 2h1 (φnij+1k−φnij−1k), j 6=J;
δy2φnijk = 1
h2(φnij+1k−2φnijk+φnij−1k), j ∈Iyo; δzφnijk = 2h1(φnijk+1−φnijk−1), k6=K;
δz2φnijk = 1
h2(φnijk+1−2φnijk+φnijk−1), k∈Izo;
∇hφnijk = (δx, δy, δz)φnijk, ∇h2φnijk = (δx2, δ2y, δz2)φnijk, i6=I, j 6=J, k 6=K;
∇2h2φnijk = (δx2δ2y, δy2δ2z, δz2δ2x)φnijk, i6=I, j 6=J, k 6=K;
δt+φnijk = 1τ(φn+1ijk −φnijk), n6=N; δtφnijk = 2τ1 (φn+1ijk −φn−1ijk ), n6= 0, N;
δt2φnijk = 1
τ2(φn+1ijk −2φnijk+φn−1ijk ), n6= 0, N;
Ahφnijk ={1 +h122(δx2−pδx+δy2−qδy+δ2z−rδz)}φnijk; i∈ Ixo, j ∈Iyo, k∈Izo.
104 HOC schemes for the transient 3D convection-diffusion-reaction equations
Further, let Φnijk be the numerical approximation ofφ(xi, yj, zk, tn), then utiliz- ing the average parameterς in equation (5.8), coefficients of the formAδx2φnijk
can be replaced by ςAδx2φn+1ijk + (1−ς)Aδx2φnijk. Likewise, we replace for the other terms as well.
With these replacements, the finite difference scheme can now be written as Ahδt+Φnijk+ (−ςA)δ2xΦn+1ijk − {(1−ς)A}δx2Φnijk+ (−ςB)δy2Φn+1ijk
− {(1−ς)B}δy2Φnijk+ (−ςC)δz2Φn+1ijk − {(1−ς)C}δ2zΦnijk+ (ςP)δxΦn+1ijk
− {(ς −1)P}δxΦnijk+ (ςQ)δyΦn+1ijk − {(ς −1)Q}δyΦnijk+ (ςR)δzΦn+1ijk
− {(ς −1)R}δzΦnijk+ (−ςD)δxδyΦn+1ijk − {(1−ς)D}δxδyΦnijk + (−ςE)δyδzΦn+1ijk − {(1−ς)E}δyδzΦnijk+ (−ςF)δzδxΦn+1ijk
− {(1−ς)F}δzδxΦnijk+ (ςH)δxδy2Φn+1ijk − {(ς −1)H}δxδy2Φnijk + (ςK)δx2δyΦn+1ijk − {(ς−1)K}δx2δyΦnijk+ (−ςL)δx2δy2Φn+1ijk
− {(1−ς)L}δ2xδy2Φnijk+ (ςM)δyδ2zΦn+1ijk − {(ς−1)M}δyδ2zΦnijk + (ςN)δy2δzΦn+1ijk − {(ς −1)N}δy2δzΦnijk+ (−ςO)δy2δz2Φn+1ijk
− {(1−ς)O}δy2δz2Φnijk+ (ςS)δx2δzΦn+1ijk − {(ς −1)S}δx2δzΦnijk
+ (ςT)δxδ2zΦn+1ijk − {(ς −1)T}δxδ2zΦnijk+ (−ςU)δz2δx2Φn+1ijk
− {(1−ς)U}δz2δx2Φnijk+ (ςG)Φn+1ijk − {(ς−1)G}Φnijk =W
(5.20)
Define the initial approximations as
Φ0ijk=φ0(xi, yj, zk), (5.21) where i∈Ix, j∈Iy, k∈Iz and φ0(xi, yj, zk) =φ(xi, yj, zk,0)
Also consider that,
Φ1ijk =φ0(xi, yj, zk) +τ φ1(xi, yj, zk), (i, j, k)∈Ix×Iy×Iz, (5.22) where φnijk =φ(xi, yj, zk, tn) and φ1(xi, yj, zk) = φt(xi, yj, zk,0).
5.3. Fundamental studies 105
Remark 5.3.2. By virtue of lemma 5.3.1, it follows that there exists a real constantKsuch thatK.|ς|ξn+1 andK.|ς−1|ξn, whereξijk can be either of the coefficients A, B, C, P, Q, R, D, E, F, H, K, L, M, N, O, S, T, U orG.
Now, let us consider the space, S := {φ = φijk|(i, j, k) ∈ Ix ×Iy × Iz} ⊆ R(I+1)×(J+1)×(K+1) and the inner product and discrete norms defined overS as hφ1, φ2i=h3
I−1P
i=1 J−1P
j=1 K−1P
k=1
(φ1)ijk(φ2)ijk, where φ1, φ2∈ S,
kφk=hφ, φi12, kδxφk= 8h3
XI−1
i=0
XJ−1
j=1 K−1X
k=1
|δxφijk|2
!12 ,
kδzφk= 8h3 XI−1
i=1
XJ−1
j=1 K−1X
k=0
|δzφijk|2
!12 ,
kδx2φk= h6
I−1
X
i=0
XJ−1
j=1 K−1X
k=1
|δx2φijk|2
!12 ,
kδz2φk= h6
I−1
X
i=1 J−1
X
j=1 K−X1
k=0
|δz2φijk|2
!12 ,
kδx2δy2φk= h12 XI−1
i=0
XJ−1
j=1 K−1X
k=1
|δx2δy2φijk|2
!12 ,
kδz2δx2φk= h12 XI−1
i=1
XJ−1
j=1 K−1X
k=0
|δz2δ2xφijk|2
!12 ,
kφk∞= max
(i,j,k)∈Ix×Iy×Iz|φijk|, kδyφk= 8h3
I−1
X
i=1 J−1
X
j=0 K−X1
k=1
|δyφijk|2
!12 , k∇hφk= kδxφk2+kδyφk2 +kδzφk212
, kδy2φk= h6
XI−1
i=1
XJ−1
j=0 K−1X
k=1
|δ2yφijk|2
!12 , k∇h2φk=
kδx2φk2+kδy2φk2+kδ2zφk212 , kδ2yδz2φk= h12
I−1
X
i=1 J−1
X
j=0 K−X1
k=1
|δy2δz2φijk|2
!12 , k∇2h2φk=
kδx2δy2φk2+kδ2yδz2φk2+kδ2zδx2φk212 .
Lemma 5.3.3. If φn−1, φn, φn+1 ∈ S, then we have 1
τ(hAhφn+1, φn+1i − hAhφn, φni)≤ hAhδt+φn, δt+φni (5.23)
106 HOC schemes for the transient 3D convection-diffusion-reaction equations
Proof. Since,
hAhδt+φn, δt+φni=hAh
φn+1−φn
τ , φn+1−φn
τ i
= 1
τ2hAh(φn+1−φn), (φn+1−φn)i[∵Ah is linear]
= 1
τ2{hAhφn+1, φn+1i − hAhφn+1, φni
− hAhφn, φn+1i+hAhφn, φni}
Similarly, we have
hAhδt+φn−1, δt+φn−1i = 1
τ2{hAhφn, φni − hAhφn, φn−1i
− hAhφn−1, φni+hAhφn−1, φn−1i}
Therefore, using the definition ofAh, property of symmetry ofhφn, φn+1iand considering the unit vectors as φ
kφk2, the proof of the lemma follows.
In order to analyse convergence of the difference scheme, let us define the local truncation error rn ∈ S as follows
rnijk = Ahδ+t φnijk+ (−ςA)δx2φn+1ijk − {(1−ς)A}δx2φnijk+ (−ςB)δy2φn+1ijk
− {(1−ς)B}δ2yφnijk+ (−ςC)δz2φn+1ijk − {(1−ς)C}δ2zφnijk+ (ςP)δxφn+1ijk (5.24)
− {(ς−1)P}δxφnijk+ (ςQ)δyφn+1ijk − {(ς −1)Q}δyφnijk+ (ςR)δzφn+1ijk
− {(ς−1)R}δzφnijk+ (−ςD)δxδyφn+1ijk − {(1−ς)D}δxδyφnijk+ (−ςE)δyδzφn+1ijk
− {(1−ς)E}δyδzφnijk+ (−ςF)δzδxφn+1ijk − {(1−ς)F}δzδxφnijk+ (ςH)δxδ2yφn+1ijk
− {(ς−1)H}δxδy2φnijk+ (ςK)δx2δyφn+1ijk − {(ς −1)K}δx2δyφnijk+ (−ςL)δx2δy2φn+1ijk
− {(1−ς)L}δx2δy2φnijk+ (ςM)δyδz2φn+1ijk − {(ς−1)M}δyδ2zφnijk+ (ςN)δy2δzφn+1ijk
− {(ς−1)N}δy2δzφnijk+ (−ςO)δy2δ2zφn+1ijk − {(1−ς)O}δy2δ2zφnijk+ (ςS)δx2δzφn+1ijk
− {(ς−1)S}δx2δzφnijk+ (ςT)δxδ2zφn+1ijk − {(ς −1)T}δxδ2zφnijk+ (−ςU)δz2δ2xφn+1ijk
− {(1−ς)U}δz2δ2xφnijk+ (ςG)φn+1ijk − {(ς−1)G}φnijk+W, where (i, j, k)∈Ixo×Iyo×Izo, and 1≤n≤N −1.
5.3. Fundamental studies 107
Using Taylor Series expansion, we have
rijko = φ1ijk−φ0(xi, yj, zk)
τ −φ1(xi, yj, zk), ∀(i, j, k)∈Ixo×Iyo×Izo (5.25) Also we have the following
|δx+rijk0 |.τ2, |δy+rijk0 |.τ2, |δz+r0ijk|.τ2, |rnijk|.τ2, (5.26) where (i, j, k)∈Ixo×Iyo×Izo and 0≤n≤N.
Let the error function en ∈ S be defined as
enijk =φnijk−Φnijk, (i, j, k)∈Ix×Iy ×Iz, 0≤n ≤N. (5.27)
Lemma 5.3.4. For any grid function φ∈ S, we have
|hAhφ, φi| ≤ kφk2 (5.28) Proof. From the defintion of Ahφ and use of diiferential operators, it follows that
|hAhφ, φi| ≤ |hφ, φi|
Now, the proof of the above lemma follows from the definition of inner product and Cauchy-Schwarz inequality.
Theorem 5.3.5. If φ ∈ C6,6,6,3([xa, xb]×[ya, yb]×[za, zb]×[0, T]), and if h and τ are sufficiently small, then the solution of the finite difference problem (5.20)-(5.21)-(5.22) converges to the solution of the initial continuous problem (5.1)-(5.18)-(5.19) with order O(τ2+h4) in the discreteL2 norm for Φn, i.e.
kenk+k∇henk+k∇h2enk+k∇2h2enk.τ2+h4, 0≤n≤N. (5.29) Proof. Let us adopt an induction argument to proof that (5.29) holds for every
108 HOC schemes for the transient 3D convection-diffusion-reaction equations
non-negative integer less thanN. From equation (5.20) and remark (5.3.2), it is obvious that (5.29) holds for n= 0. For (i, j, k)∈Ixo×Iyo×Izo, we have
|e1ijk|=|φ1ijk−Φ1ijk|=|τ rijk0 |.τ3, |δt+e0ijk|=|r0ijk|.τ2, and (5.30a)
|δx+e1ijk|.τ|δx+r0ijk|. τ3 .τ2, (5.30b) Likewise, we have
|δy+e1ijk|.τ2 and |δ+ze1ijk|.τ2 (5.30c) Equation (5.30) clearly implies that (5.29) holds forn = 1. Using the triangle inequality, when τ and h are sufficiently small, we have
|Φ1ijk| ≤ |φ1ijk|+|e1ijk| ≤Co+C(τ2+h2)≤1 +Co, (i, j, k)∈Ix×Iy×Iz (5.31) where Co = max
0≤t≤T kφ(·,·,·, t)kL∞(Ω).
We assume that (5.29) is valid for 0≤n ≤m−1< N−1, then we will show that it is still valid for n = m. Now, adding (5.20) and (5.24), we have the following error equation for the error functionen∈ S
Ahδt+enijk+ (−ςA)δx2en+1ijk − {(1−ς)A}δx2enijk+ (−ςB)δ2yen+1ijk
− {(1−ς)B}δy2enijk+ (−ςC)δz2en+1ijk − {(1−ς)C}δz2enijk+ (ςP)δxen+1ijk
− {(ς−1)P}δxenijk+ (ςQ)δyen+1ijk − {(ς −1)Q}δyenijk+ (ςR)δzen+1ijk
− {(ς−1)R}δzenijk+ (−ςD)δxδyen+1ijk − {(1−ς)D}δxδyenijk+ (−ςE)δyδzen+1ijk
− {(1−ς)E}δyδzenijk+ (−ςF)δzδxen+1ijk − {(1−ς)F}δzδxenijk+ (ςH)δxδy2en+1ijk
− {(ς−1)H}δxδy2enijk+ (ςK)δx2δyen+1ijk − {(ς −1)K}δx2δyenijk+ (−ςL)δ2xδ2yen+1ijk
− {(1−ς)Lnijk}δx2δy2enijk+ (ςM)δyδz2en+1ijk − {(ς −1)M}δyδz2enijk+ (ςN)δ2yδzen+1ijk
− {(ς−1)N}δ2yδzenijk+ (−ςO)δy2δz2en+1ijk − {(1−ς)O}δy2δz2enijk+ (ςS)δx2δzen+1ijk
− {(ς−1)S}δ2xδzenijk+ (ςT)δxδ2zen+1ijk − {(ς −1)T}δxδz2enijk+ (−ςU)δ2zδx2en+1ijk
− {(1−ς)U}δz2δx2enijk+ (ςG)en+1ijk − {(ς −1)G}enijk=rnijk, (i, j, k)∈Ixo×Iyo×Izo
Assuming τ and h to be sufficiently small and considering (5.29) for 1≤n ≤
5.3. Fundamental studies 109
m−1, we have the following estimate
kΦnk∞ ≤ kφnk∞+kenk∞≤Co+ 1, 1≤n≤m−1. (5.32) Let us consider the case when ς = 0.5. Throughout this section, we will con- sider the case only where ς = 0.5.
Now, computing the inner product of (5.32) withδt+enand considering remark (5.3.2) and lemma (5.3.3), we have
1
τ(hAhen+1, en+1i − 1
τhAhen, eni+Kh(δx2+δy2+δ2z+δx+δy+δz+δxδy +δyδz
+δzδx+δxδy2+δx2δy+δ2xδy2+δyδ2z +δy2δz +δy2δ2z+δzδ2x+δz2δx+δz2δx2+ 1)en+1, δt+eni ≤ hrn, δ+t eni+Kh(δ2x+δy2+δz2+δx+δy+δz +δxδy+δyδz+δzδx (5.33) +δxδ2y+δx2δy+δx2δy2+δyδz2+δ2yδz+δ2yδz2+δzδx2+δ2zδx+δ2zδx2+ 1)en, δ+t eni Upon simplifying and transposing, we have
(hAhen+1, en+1i+Kτ2(kδ2xen+1k2 +kδy2en+1k2+kδ2zen+1k2+kδxen+1k2 +kδyen+1k2+kδzen+1k2+kδxδyen+1k2+kδyδzen+1k2+kδzδxen+1k2 +kδxδy2en+1k2+kδx2δyen+1k2+kδ2xδy2en+1k2+kδyδ2zen+1k2+kδ2yδzen+1k2 +kδy2δz2en+1k2+kδzδx2en+1k2+kδz2δxen+1k2+kδz2δx2en+1k2+ken+1k2)
≤ τ2(krnk2− ken+1k2+kenk2) +Kτ2(kδx2enk2+kδy2enk2
+kδz2enk2+kδxenk2+kδyenk2+kδzenk2 +kδxδyenk2+kδyδzenk2+kδzδxenk2 +kδxδy2enk2+kδx2δyenk2+kδx2δy2enk2+kδyδ2zenk2+kδ2yδzenk2+kδy2δz2enk2 +kδzδx2enk2+kδz2δxenk2 +kδz2δx2enk2+kenk2) + (hAhen, eni (5.34) Define Gn as follows:
Gn = (hAhen, eni+τ2kenk2+Kτ2(kδx2enk2 +kδy2enk2+kδ2zenk2+kδxenk2 + kδyenk2+kδzenk2+kδxδyenk2+kδyδzenk2+kδzδxenk2+kδxδy2enk2 + kδ2xδyenk2+kδ2xδy2enk2+kδyδz2enk2+kδy2δzenk2+kδy2δz2enk2+kδzδx2enk2 + kδ2zδxenk2+kδz2δ2xenk2+kenk2), 0≤n≤m−1 (5.35)
110 HOC schemes for the transient 3D convection-diffusion-reaction equations
Then equation (5.34) can be written as
Gn+1−Gn.τ2(krnk2) +Kτ2(Gn+1+Gn), 0≤n < m−1. (5.36) This together with Gronwall’s inequality [22, 38] gives
Gn+1 .[G0+τ2 Xn+1
l=1
(krlk2)]e4CT. (5.37)
Equation (5.37), concurrently with lemma 5.3.4 and equation (5.30) gives the following estimates
kemk+k∇hemk+k∇h2emk+k∇2h2emk+ some other terms .τ2+h4. (5.38) which ultimately shows that, by principle of mathematical induction, we have
kenk+k∇henk+k∇h2enk+k∇2h2enk.τ2+h4, 0≤n≤N. (5.39) This completes the proof.