B1≥ 1
Proof. The boundary condition (4.31) forωimplies that
N−1
The term I2 can be controlled by Cauchy’s inequality directly. First, re-call the definition of hi,0 in (4.31). Then summing by parts gives I2
=
which implies thatI1can be, after summing by parts, expressed as
I1= 25
The first term on the right-hand side of (4.44) is estimated directly, while for the remaining terms we apply Cauchy’s inequality, giving
25
−1
The combination ofI1andI2then gives
N−1
whereBψ was defined in (4.40). Moreover, (4.39) is a direct consequence of (4.48). The treatment of the other three boundary terms is exactly the same.
This completes the proof of Lemma 4.4.
To complete the estimate ofB1we need to control
1+h62D2y this case. The methodology we adopt here is similar to that used in [21], i.e., control the local terms by global terms via elliptic regularity.
Lemma 4.5 For anyψthat vanishes on the boundary, we have
D2xψ21+ D2yψ21≤ 9
Proof. Given the homogeneous boundary conditionψi,j |=0, we perform a Sine transformation of{ψi,j}in both the x and y directions, i.e.,
which in turn implies that
N−1
On the other hand, direct calculation along with the fact that−4
h2 ≤fk, g ≤ 0 shows that
|fk+g+h2
6 fkg|2≥ | 1+h2
6 g
fk|2+ | 1+h2
6fk
g|2, (4.57)
|fk+g+h2
6fkg|2≥
fk2+g2−2 9fkg
≥ 8
9(fk2+g2) . (4.58)
Combining (4.55)–(4.57) gives (4.50). Estimate (4.49) can be argued in a
similar fashion. Lemma 4.5 is proven.
The combination of Lemmas 4.4 and 4.5 results in the estimate B1≥ 1
3h2Bψ −3
4ω21−Ch9. (4.59)
An estimate forB2 can be derived in a similar fashion, which we only briefly outline. Consider first the expression
iDx2ψi,1Dx2ωi,0inB2. Once again the boundary condition (4.31) for ω leads to examining
iD2xψi,1
Dx2ωi,0in two parts,I3andI4, given by I3= 1
h2
N−1 i=1
Dx2ψi,1
8Dx2ψi,1−3Dx2ψi,2+8
9Dx2ψi,3−1 8Dx2ψi,4
,
I4=
N−1 i=1
D2xψi,1Dx2hi,0. (4.60)
The estimate ofI3andI4is similar to that ofI1andI2, respectively. Repeating the arguments in the proof of Lemma 4.4, we arrive at (omitting the details)
B2≥ − 1
144h4D2yDx2ψ21−Ch9. (4.61)
On the other hand, the fact thatD2yDx2ψ1≤ 4
h2Dx2ψ1andDy2Dx2ψ1≤ 4
h2Dy2ψ1implies Dy2Dx2ψ21= 1
2(D2yDx2ψ21+ Dy2Dx2ψ21)≤ 8
h4D2xψ21+ 8
h4Dy2ψ21
≤ 9 h4ω21, (4.62)
where in the last step we applied (4.49) in Lemma 4.5. Substituting (4.62) into (4.61), we arrive at
B2≥ − 1
16ω21−Ch9. (4.63)
Finally,B3can be controlled by applying Cauchy’s inequality (we only consider here the term 1
6ψ1,1ω0,0) 1
6ψ1,1ω0,0≥ − 1 12
ψ1,12
h2 − 1
12h2ω20,0≥ − 1 12
ψ1,12
h2 −Ch10ψe2C8, (4.64)
where in the last step we used the fact that|ω0,0| ≤Ch4ψeC8 by our con-struction of in section 3.1. The first term on the right-hand side of (4.64) can be absorbed into theBψterm, giving
B3≥ − 1
12h2Bψ−Ch9. (4.65)
The combination of (4.63), (4.65), and Lemma 4.4 shows that B ≥
−13
16ω21−h8, whose substitution into (4.37) is exactly (4.36). This
com-pletes the proof of Proposition 4.3.
The estimates for the linearized convection terms in (4.32) are given in the following proposition. The proof is similar to that of Proposition 4.3 and the details are left to interested readers.
Proposition 4.6 Assume a-priori that the error functions for the velocity field and temperature satisfy
uL∞ ≤h2, θL∞ ≤h2. (4.66)
Then we have
1+h2 12h
ψ , L2
1
≤C1∇hψ22+ν
8ω21+h8, (4.67)
where C1= C(1+ ueC0)2
ν +C(2+ ueC1)2+CueC5.
In addition, by Cauchy’s inequality and the boundary condition for the temperature error functionθ in (4.31), we have the estimate of the gravity
term
1+ h2
12h
ψ , Dx
1−h2
12(D2y−D2x) θ
1
≤C(ψ21+ ∇hθ22)+Ch10, (4.68)
In (4.33), the local truncation error term−θ ,g1 can be controlled by Cauchy’s inequality. The technique used in Lemma 4.6 can be applied here for the estimate of the linearized temperature convection term, i.e., the assumed a-priori assumption (4.66) leads to
θ ,L11
≤C2θ21+1
2κ∇hθ22+h8, (4.69)
whereC2= C(1+ ueC0)2
κ .
Next, we apply the technique demonstrated in section 3.1 for the sta-bility analysis of the long-stencil discretization of the one-dimensional heat equation to the temperature diffusion term.
Proposition 4.7 We have
− θ ,
h− h2
12(Dx4+Dy4) θ
1
≥ 1
2∇hθ22−h8. (4.70)
Proof. The proof of (4.70) is just the two-dimensional version of the stability analysis in section 3. Sinceθ vanishes on the boundary, we have
− θ ,
h−h2
12(Dx4+D4y) θ
1
= ∇hθ22+h2
12D2xθ21+h2
12D2yθ21+h2 12B, (4.71)
where Barises from the boundary terms, which after summation by parts, can be written as
B=
N−1 i=1
θi,1(D2yθ )i,0+
N−1 i=1
θi,N−1(Dy2θ )i,N +
N−1
j=1
θ1,j(Dx2θ )0,j
+
N−1 j=1
θN−1,j(Dx2θ )N,j. (4.72)
We focus on the first term appearing on the right-hand side of (4.72); the other three boundary terms can be treated similarly. Applying the boundary condition forθ at the “ghost points” as in (4.31),(Dy2θ )i,0can be written as
(D2yθ )i,0= 1 h2
5
11θi,1− 4
11θi,2+ 3
11θi,3+ei,0
(4.73) ,
which is analogous to (3.7) except for the local error termei,0 (defined in (4.31)), whose product withθi,1 can be controlled by Cauchy’s inequality.
Alternatively, we can rewrite the right-hand side of (4.73), as we did in sec-tion 3, as
(D2yθ )i,0= −2
11(Dy2θ )i,1+ 1
11(Dy2θ )i,2+ei,0
h2 . (4.74)
The aim here is the control of local terms by global terms. Applying Cauchy’s inequality to each term in (4.74) leads to
θi,1(D2yθ )i,0≥ − 22
4·112·h2θi,12 −h2(D2yθi,1)2− 12
4·112·h2θi,12 −h2(Dy2θi,2)2
− 1
4h2θi,12 −e2i,0 h2
≥ − 1
2h2θi,12 −h2(Dy2θi,1)2−h2(Dy2θi,2)2−e2i,0 h2 . (4.75)
Here the arguments in section 4 can be repeated: the first term appearing above can be controlled by∇hθ22, since it will be multiplied by h2
12, resulting in a term greater than 1
2∇hθ22; the second and third terms can be controlled by Dy2θ22; while the last term can be controlled by
h2 12
N−1 i=1
e2i,0
h2 ≤N h2·Ch10θe2C6· 1
h2 ≤Ch9θe2C6, (4.76)
where we used the fact thath= 1
N. (4.70) then follows.
Finally, the combination of (4.32)–(4.37), and (4.67)–(4.70) gives us 1
2 dE
dt +1 2
d
dtθ21≤Ch8+Cf21+Cg21+C1∇hψ22
+C2θ21−κ
2∇hθ22. (4.77)
Integrating in time results in E+ θ21≤C
T 0
(f21+ g21) dt+2C2
T 0
θ21dt +2C1
T 0
∇hψ22dt+CT h8. (4.78)
It can be seen that
∇hψ22≤3(∇hψ22− h2
12hψ21−h2
6DxDyψ21) , (4.79)
sinceψvanishes on the boundary, which along with (4.78) implies that
∇hψ22+ θ21≤C T
0
(f21+ g21) dt +CC1 T
0
∇hψ22dt +CC2
T 0
θ21dt+CT h8. (4.80)
By Gronwall’s inequality we then have
∇hψ22+ θ21≤Cexp
CC1T +CC2T T
0
f(·, s)21+ g(·, s)21ds +CT h8
≤CT h8exp
CC1T +CC2T (4.81)
ue2C7,α(1+ ueC5)2+ θe2C5ue2C5
+θe2C6+CT h8
. Thus, we have proven
u(·, t )−ue(t )L2+ θ (·, t )−θe(t )L2
≤Ch4
ueC7,α(1+ ueC5)+ θeC5ueC5+ θeC6
·exp CT
ν (1+ ueC1)2+CT
κ (1+ ueC0)2 . (4.82)
Using the inverse inequality, we have
uL∞ ≤Ch3. (4.83)
At this point, we can introduce a standard concept which asserts that (4.66) will never be violated ifhis small enough, and Theorem 1.1 is proven.
5 Convergence proof of Theorem 1.2
The numerical scheme with the Neumann boundary condition (1.4), namely (2.7), (2.16), and (2.26), is analyzed in this section. For simplicity of pre-sentation we set θf = 0 in which case the one-sided extrapolation of the temperature at the boundary is given by (2.27).
The consistency analysis of the momentum equation is the same as that presented in section 4. We denote by ψe,ue, andωe the exact solutions of (1.1)–(1.2), and (1.4), and extend ψe smoothly to [−δ,1+δ]2. Then let i,j =ψe(xi, yj)for−2 ≤i, j ≤N +2. The approximated velocity pro-filesU andV, and the vorticity profile are given by (4.1) and (4.2)–(4.5),
respectively. Lemmas 4.1 and 4.2, along with the estimate for the time march-ing term in (4.18), remain valid. The fourth order approximation (4.22) for the constructed vorticity on the boundary is also preserved.
Regarding the temperature variable, instead of substituting the exact solu-tion into the numerical scheme, a careful construcsolu-tion of an approximated temperature profile is performed by adding anO(h4)correction term to θe to satisfy the truncation error fully to fourth order. The reason for this proce-dure is to avoid the loss of accuracy near the boundary which would result from a direct truncation error estimate. To be more precise, we construct the approximate temperature fieldas
=θe+h4θ , (5.1)
in which the correction functionθ satisfies the Poisson equation θ =C1,
(5.2a)
with the Neumann boundary condition
∂yθ (x,0)= 1
80∂y5θe(x,0) , ∂yθ (x,1)= 1
80∂y5θe(x,1) ,
∂xθ (0, y)= 1
80∂x5θe(0, y) , ∂xθ (1, y)= 1
80∂x5θe(1, y) . (5.2b)
The scalarC1(a function of timet) is chosen as
C1dx =
∂
∂θ
∂n dn in order to maintain consistency with the Neumann boundary condition, i.e.,
C1= 1
| | 1
0
− 1
80∂y5θe(x,0)+ 1
80∂y5θe(x,1) dx +
1 0
−1
80∂x5θe(0, y)+ 1
80∂x5θe(1, y) dy (5.3) .
A Schauder estimate applied to the Poisson equation (5.2) gives θCm,α ≤CθeCm+4,α, for m≥2. (5.4)
The reason for taking the boundary condition forθ in (5.2b) will become apparent later. A local Taylor expansion for the exact temperature fieldθeat
points near the boundaryy =0 gives
(θe)i,−1=(θe)i,1−h3
3 ∂y3θe(xi,0)−h5
60∂y5θe(xi,0)+O(h7)θeC7
=(θe)i,1+ h3
3κ (ωe)i,0(∂xθe)i,0−h5
60∂y5θe(xi,0)+O(h7)θeC7, (θe)i,−2=(θe)i,2−8h3
3 ∂y3θe(xi,0)−32h5
60 ∂y5θe(xi,0)+O(h7)θeC7
=(θe)i,2+8h3
3κ(ωe)i,0(∂xθe)i,0−32h5
60 ∂y5θe(xi,0)+O(h7)θeC7, (5.5)
due to the no-flux boundary condition forθe and the derivation for∂y3θe as given in (2.25) by applying the original PDE on the boundary. The insertion of the boundary conditions given by (5.2b) into a Taylor expansion ofθ, along with the Schauder estimateθC3 ≤CθeC7,αgiven by (5.4), gives
θi,−1=θi,1−2h∂yθi,0+O(h3)∂y3θi,0
=θi,1−40h ∂y5θe(xi,0)+O(h3)θeC7,α, θi,−2=θi,2−4h∂yθi,0+O(h3)∂y3θi,0
=θi,2−20h ∂y5θe(xi,0)+O(h3)θeC7,α. (5.6)
The combination of (5.5) and (5.6) results in an estimate for=θe+h4θ given by
i,−1=i,1+ h3
3κ (ωe)i,0(∂xθe)i,0−h5
24∂y5θe(xi,0)+O(h7)θeC7,α, i,−2=i,2+8h3
3κ (ωe)i,0(∂xθe)i,0−7h5
12 ∂y5θe(xi,0)+O(h7)θeC7,α. (5.7)
Similar results can be obtained at the other three boundary segments, namely y =1,x =0, andx=1. Note that theO(h5)coefficients ofi,−1andi,−2 have the ratio 1 : 14. This will be needed for the error analysis of the inner product of the temperature with the diffusion term in the temperature equa-tion. This crucial point is the reason for the choice of the boundary condition forθ in (5.2b).
A direct consequence of the Schauder estimate (5.4) is given by θW2,∞( )≤CθC2,α ≤CθeC6,α,
(5.8)
in which · Wm,∞( ) represents the maximum value, at grids points, of the given function up tom-th order finite-difference, over the domain . As a result, we have
−θeW2,∞( ) =h4θW2,∞( )≤Ch4θeC6,α. (5.9)
The combination of (5.9) and a local Taylor expansion forθe gives the esti-mates
Dx
1+ h2
12(Dy2−Dx2)
=Dx
1+h2
12(D2y−D2x)
θe+O(h4)θeC6,α
= 1+h2
12
∂xθe+O(h4)θeC6,α, (5.10)
Dx
1−h2 6 D2x
=∂xθe+O(h4)θeC6,α, Dy
1−h2
6 D2y
=∂yθe+O(h4)θeC6,α, (5.11)
h−h2
12(Dx4+D4y)
=θe+O(h4)θeC6,α,
UDx
1−h2 6Dx2
=ue∂xθe+O(h4)ueC5θeC6,α, (5.12)
VDx
1−h2 6Dx2
=ve∂yθe+O(h4)ueC5θeC6,α. (5.13)
Moreover, taking the time derivative of (5.2) leads to a Poisson equation for∂tθ, namely
(∂tθ )=∂tC1, (5.14a)
with the Neumann boundary conditions
∂y(∂tθ )(x,0)= 1
80(∂t∂y5θe)(x,0) , ∂y(∂tθ )(x,1)= 1
80(∂t∂y5θe)(x,1) ,
∂x(∂tθ )(0, y)= 1
80(∂t∂x5θe)(0, y) , ∂x(∂tθ )(1, y)= 1
80(∂t∂x5θe)(1, y) . (5.14b)
Similarly, the value of∂tC1is given by
∂tC1= 1
| | −
1 0
1
80(∂t∂y5θe)(x,0)+ 1
80(∂t∂y5θe)(x,1) dx +
1 0
− 1
80(∂t∂x5θe)(0, y)+ 1
80(∂t∂x5θe)(1, y) dy . (5.15)
A Schauder estimate applied to the Poisson equation (5.14) reads ∂tθCm,α ≤C∂tθeCm+4,α ≤C
ueCm+4,αθeCm+5,α+ θeCm+6,α
, for m≥2,
(5.16)
in which the original temperature transport equation∂tθe+(ue·∇)θe =κθe
was used. It can be seen that (5.16) amounts to
∂t=∂tθe+O(h4)
ueC6,αθeC7,α+ θeC8,α
(5.17) .
The combination of (4.10)–(4.13), (4.18)–(4.19), (5.10), and the original momentum equation results in
∂t +Dx
1+h2 6Dy2
(U )+Dy
1+h2 6 D2x
(V )
− h2
12h(UDx +VDy )−gDx
1+h2
12(D2y−D2x)
=ν
h+h2 6Dx2Dy2
+f , (5.18)
where|f| ≤Ch4ueC7,α(1+ ueC5)+Ch4θeC6,α. Similarly, the com-bination of (5.11)–(5.13), (5.17), and the original temperature equation gives
∂t+UDx
1−h2
6Dx2
+VDy
1−h2
6Dy2
=κ
h− h2
12(Dx4+Dy4)
+g, (5.19)
where|g| ≤Ch4(ueC6,αθeC7,α+ θeC8,α).
The error functions at the computational grid points(xi, yj), 0≤i, j ≤N are defined in the same way as in (4.28). Subtracting (5.18)–(5.19) from the
numerical scheme (2.7), (2.16), and (2.27), gives
in which the linearized convection error termsL1andL2are given by (4.31), and the local truncation error terms satisfy
|g| ≤Ch4(ueC6,αθeC7,α+ θeC8,α) , from (5.7), and using the approximations (4.22) and (5.11), that for the tem-perature field
Subtracting (5.23) from (2.27), we arrive at θi,−1=θi,1+ h3
where the ratio 1 : 14. Such a ratio is a crucial point in the error analysis of the temperature diffusion term in (5.20), which will be established in detail in Proposition 5.1.
The estimate of the error functions in the system (5.20)–(5.22) and (5.24) is similar to that in section 4. Multiplying the vorticity error equation by
− 1+ h2
12h
ψat interior grid points 1 ≤ i, j ≤ N −1 gives the same identity as in (4.32)
(5.25)
The energy estimate of the temperature error is different from the Dirichlet boundary condition case since the temperature field is updated at every grid point 0 ≤ i, j ≤ N. Taking the, 3inner product (see the definition in (3.13)) of the temperature error equation withθ gives
1
which is also the same as (4.33) except for the difference of inner product and L2norms. Again, this is due to the fact that the temperature field is updated at every grid points 0≤i, j ≤N.
An estimate for (5.25) is the same as that for (4.32). The identities (4.34)–
(4.35), Propositions 4.3 and 4.6, and (4.68) are still valid. More precisely,
−
provided the a-priori assumption (4.66) is satisfied, along withC1as intro-duced after (4.67).
Similar to (4.69), the linearized temperature convection term can be controlled by
An estimate of the temperature diffusion term in (4.69) is outlined below.
Its proof relies on the stability analysis given in section 3.2 and some error estimates.
Proposition 5.1 We have
−
Proof. Summing by parts under the inner product , 3 and using the boundary extrapolation (5.24) gives
θ ,
hh2
12(D4x+Dy4) θ
3
= −∇hθ22−h2
12D2xθ23−h2
12D2yθ23+B. (5.33)
The boundary termBcan be decomposed as B=B1+B2+B3+B4, (5.34a)
in whichB1, corresponding to the boundary term alongy =0, reads
B1= 1 12
N−1
i=1
θi,1(θi,−1−θi,1)+1 2θi,0
(θi,−2−θi,2)−16(θi,−1−θi,1)
= 1 12
N−1
i=1
θi,1 h3
3κqbi −h5
24rbi +eb1i +1
2θi,0
8h3
3κqbi −7h5
12 rbi +eb2i −16(h3
3κqbi −h5
24rbi +eb1i )
. (5.34b)
It should be noted that the derivation of (5.34b) comes from the formula for
θi,−1andθi,−2 in (5.24a). The definitions ofqbi, rbi, eb1i ,eb2i were given in (5.24b). The boundary terms alongy =1,x =0, andx =1 can be similarly presented. In more detail,B1can be simplified as
B1= h3 36κ
N−1 i=1
qbi(θi,1−4θi,0)+ 1 12
N−1 i=1
eb1i (θi,1−8θi,0)+1 2eb2i θi,0
+ h5 288
N−1 i=1
rbi(θi,0−θi,1)
≡I1b+I2b+I3b. (5.35)
The first termI1bcan be controlled by using the form ofqbi as in (5.24b), namely
h3
N−1 i=1
qbiθi,1=h3
N−1 i=1
θi,1ωi,0Dx(1−h2 6 D2x)θi,0
+h3
N−1
i=1
θi,1 i,0Dx(1−h2 6Dx2)θi,0
≤CθW1,∞∇hψ2θ3+Ch L∞θ3∇hθ2+Ch10 (5.36)
≤C(θeC6,α+1)(∇hψ22+ θ23)+Ch(ueC5 +1) (θ23 + ∇hθ22)+Ch10,
in which the first inequality comes from the boundary formula forωi,0 in (5.22). The second inequality results from the estimate (5.9), (4.6), and the a-priori assumption (4.66). A similar result can be obtained forh3N−1
i=1 qbi θi,0. Then we arrive at
I1b≤C(θeC6,α+1)(∇hψ22+ θ23)+Ch(ueC5+1) (θ23 + ∇hθ22)+Ch10.
(5.37)
The termI2bcan be controlled by Cauchy’s inequality and the estimate (5.24b), giving
I2b≤C N−1
i=1
eb1i θi,1 +C
N−1
i=1
eb1i θi,0
≤Cθ23+Ch10. (5.38)
What remains is the estimate of I3b. As can be seen, the detailed esti-mates forθi,−1andθi,−2in (5.24), which shows that theO(h5)coefficients of θi,−1,θi,−2 have the ratio 1 : 14, allows the term I3b to be written as
h5 288
N−1
i=1 rbi(θi,0−θi,1). That is crucial to implement the error analysis below.
An application of Cauchy’s inequality shows that I3b = h5
288
N−1 i=1
rbi(θi,0−θi,1)≤ 1 288h2
N−1
i=1
(θi,0−θi,1)2 h2 + 1
288h10
N−1 i=1
r2i . (5.39)
It is observed that the first term appearing above can be absorbed into∇hθ22. Meanwhile, we note thatrbi =∂y5θe(xi,0), which is a bounded quantity on y =0. Then we get
I3b= h5 288
N−1 i=1
rbi(θi,0−θi,1)≤ 1
288∇hθ22+Ch9. (5.40)
The combination of (5.37)–(5.38) and (5.40) leads to B1≤C(θeC6,α+1)(∇hψ22+ θ23)+ 1
288∇hθ22+1 8h8. (5.41)
The other three boundary terms B2,B3, B4 can be treated similarly. As a result, we arrive at
B≤C(θeC6,α+1)(∇hψ22+ θ23)+ 1
72∇hθ22+1 2h8. (5.42)
The insertion of (5.42) into (5.33) implies (5.32). Proposition 5.1 is proven.
By the combination of (5.25)–(5.32) we have the following inequality
1 2
dE dt +1
2 d
dtθ23≤Ch8+C(f23+ g23)+C1∇hψ22+C2θ23
−κ
8∇hθ22. (5.43)
The proof of Theorem 1.2 can be carried out by using a similar argument as in (4.78)–(4.83). The details are left to the interested reader.
References
[1] Bathe, K.J. (ed.): First MIT Meeting on Computational Solid and Fluid Mechanics, vols. 1 and 2. MIT Elsevier: New York, 2001
[2] Bell, J.B., Marcus, D.L.: A Second-order projection method for variable-density flows. Journal of Computational Physics 101, 334–348 (1992)
[3] Ben-Artzi, M., Fishelov, D., Trachtenberg, S.: Vorticity dynamics and numerical reso-lution of Navier-Stokes equations. Mathematical Modelling and Numerical Analysis 35, 313–330 (2001)
[4] Briley, W.R.: A Numerical study of laminar separation bubbles using the Navier-Stokes equations. Journal of Fluid Mechanics 47, 713–736 (1971)
[5] Carpenter, M.H., Gottlieb, D., Abarbanel, S.: The stability of numerical boundary treatments for compact high-order finite-difference schemes. Journal of Computa-tional Physics 108, 272–295 (1993)
[6] Christon, M., Gresho, P., Sutton S.B.: Computational predictability of time-depen-dent natural convection flows in enclosures (including a benchmark solution). Inter-national Journal of Numerical Methods in Fluids 40, 953–980 (2001)
[7] E, W., Liu, J.-G.: Vorticity boundary condition for finite difference schemes. Journal of Computational Physics 124, 368–382 (1996)
[8] E, W., Liu, J.-G.: Essentially compact schemes for unsteady viscous incompressible flows. Journal of Computational Physics 126, 122–138 (1996)
[9] Glowinski, R., Pironneau, O.: Numerical methods for the first biharmonic equation and for the two-dimensional Stokes problem. SIAM Review 21, 167–212 (1979) [10] Henshaw, W.D., Kreiss, H.O., Reyna, L.G.M.: A Fourth-order accurate difference
approximation for the incompressible Navier-Stokes equations. Computers and Flu-ids 23, 575–593 (1994)
[11] Hou, T.Y., Wetton, B.: Convergence of a finite difference scheme for the Navier-Stokes equations using vorticity boundary conditions. SIAM Journal of Numerical Analysis 29, 615–639 (1992)
[12] Hou, T.Y., Wetton, B.: Stable fourth order stream-function methods for incompress-ible flows with boundaries. Unpublished (1992)
[13] Johnston, H., Krasny, R.: Fourth order finite difference simulation of a differentially-heated cavity. International Journal of Numerical Methods in Fluids 40, 1031–1037 (2002)
[14] Johnston, H., Liu, J.-G.: Finite difference schemes for incompressible flow based on local pressure boundary conditions. Journal of Computational Physics 80, 120–154 (2002)
[15] Kreiss, H.O.: Comparison of accurate methods for the integration of hyperbolic equations, Tellus, XXIV 3, 199–215 (1972)
[16] Liu, J.-G., Wang, C., Johnston, H.: A Fourth order scheme for incompressible Bous-sinesq equations. Journal of Scientific Computing 18, 253–285 (2003)
[17] Orszag, S.A., Israeli, M.: Numerical simulation of viscous incompressible flows.
Annual Review of Fluid Mechanics 6, 281–318 (1974)
[18] Quartapelle, L.: Numerical solution of the incompressible Navier-Stokes equations.
Birkhauser, Berlin: 1983
[19] Spotz, W.F., Carey, G.F.: High-order compact scheme for the stream-function vortic-ity equations. International Journal for Numerical Methods in Engineering 38 (20), 3497–3512 (1995)
[20] Thom, A.: The flow past circular cylinders at low speeds. Proceedings Royal Society A141, 651–669 (1933)
[21] Wang, C., Liu, J.-G.: Analysis of finite difference schemes for unsteady Navier-Stokes equations in vorticity formulation. Numerische Mathematik 91, 543–576 (2002)