MF 206 A22 Lect 3

Introduction to Computational Fluid Dynamics
Lecture 3
Pressure Correction Methods for the

Navier–Stokes Equations for Incompressible Flows
Marica Pelanti
marica.pelanti@ensta-paris.fr
MF206 (ENSTA Paris) Lecture 3 2021-2022 1 / 34

Outline
1 Finite Difference and Finite Volume Methods
2 Incompressible Navier–Stokes equations
3 Pressure Correction Methods

Projection Method

Finite Difference and Finite Volume Methods
Finite Difference Method

◦ Numerical solution associated to grid points.
un n
i,j,k ≈ u(xi , yj , zk , t )
approximation of pointwise values of solution

◦ Based on differential form of the equations
(assumes smoothness).
Derivatives approximated by Finite Differences.
Finite Volume Method

◦ Domain subdivided into control volumes (cells)
un 1 n
R
i,j,k ≈ |V| V u(xi , yj , zk , t )dV
approximation of solution averaged over CV = V

◦ Based on integral form of the equations.


For simple equations (e.g. linear) and simple Cartesian grids FD and FV approaches
might lead to same discrete equations.
Example: Unsteady heat conduction in a solid (or fluid at rest)

Conservation law: rate of change of energy within the CV = rate of heat addition
Integral form
Z Z Z Z
∂ρe ~H · ~n dS = 0 ∂T ~H · ~n dS = 0
dV + F → ρch dV + F
V ∂t S V ∂t S
ρ = density (Hp. const.), e = specific internal energy, e = ch T

T = temperature, ch = heat coefficient (const.), κ = thermal conductivity coef. (const.)
Heat flux = F~H = −κ∇T~ (Fourier Law of heat conduction)
~H · ~n dS = ~ ·F
~H dV
R R
By using divergence theorem of Gauss: F S
∇ V
Differential form
∂T ~ ·F
~H = 0
ρch +∇
∂t

1D Heat equation
∆x
ρch ∂T
∂t
+ ∂F
∂x
H
=0
xi−2 xi−1 xi xi+1 xi+2
ith CV
xi− 3 xi− 1 xi+ 1 xi+ 3

2 2 2 2
Finite Volume Discretization
R xi+1/2
ith CV = interval [xi−1/2 , xi+1/2 ], numerical solution Ti∗ ≈ ∆x
1
xi−1/2
T (x, t∗ ) dx
Z xi+1/2
∂T
Integral equation over CV: ρch dx + FH (xi+1/2 , t) − FH (xi−1/2 , t) = 0
xi−1/2 ∂t
FH = −κ ∂T
∂x
→ ρch ∆x ∂T
∂t
i
+ (−κ ∂T )
∂x i+1/2
− (−κ ∂T )
∂x i−1/2
=0

Ti+1 −Ti Ti+1 −Ti T −T
( ∂T )
∂x i+1/2
≈ ∆x
⇒ ρch ∂T ∂t
i κ
− ∆x ∆x
− i ∆xi−1 = 0
∂Ti α κ
⇒ ∂t
− (∆x)2
(Ti+1 − 2Ti + Ti−1 ) = 0, α= ch ρ
= thermal diffusivity
2
∂T
Same as Finite Difference discretization in space of PDE ∂t
− α ∂∂xT2 = 0
2
T −2T +T
Ti∗ ≈ T (xi , t∗ ), ∂ T
∂x2
≈ i+1 ∆xi2 i−1
i
Simple uniform Cartesian FD and FV grids

1D
∆x
xi−2 xi−1 xi xi+1 xi+2

ith CV
xi− 3 xi− 1 xi+ 1 xi+ 3

2 2 2 2
2D
j + 23
j+1
j + 21
(xi , yj ) CV (i, j)
j
j − 21
j−1
j − 23
i− 32 i− 21 i+ 21 i+ 32
i−1 i i+1

Incompressible Navier–Stokes equations
Navier–Stokes equations for incompressible flows

Hp. density ρ = const., µ, κ, cv = const., ~fb = 0, qV = 0, ν = µ
ρ
= kinematic viscosity.
~ · ~v = 0 ,
∇ Mass
~
∂~v ~ v = − ∇p + ν ∇2~v ,
+ ~v · ∇~ Momentum
∂t ρ
∂T
ρcv ~
+ ~v · ∇T = κ∇2 T + Φ . Energy
∂t
+ Initial and Boundary Conditions (on solid rigid walls: no-slip condition ~v = 0 )
Unknowns: velocity ~v (~x, t), pressure p(~

x, t), temperature T (~
x, t).
Mixed set of elliptic-parabolic equations.
Hp. constant flow properties ⇒ decoupling of thermal problem ⇒

1) Solve mass & momentum equations for ~v , p ;
2) Given ~v field, solve energy equation for T .

Incompressible Navier–Stokes equations Velocity-Pressure formulation

Velocity-Pressure formulation ~v = velocity, p = pressure (primitive variables)
~ · ~v = 0 ,
∇ mass eq.
~
∂~
v ~v=
+ ~v · ∇~ − ∇p + ν ∇2~v . momentum eq.
∂t ρ
• Divergence-free velocity field constraint expresses incompressibility.

• The pressure for incompressible flows is not a thermodynamic variable.
The pressure adjusts instantaneously so that velocity remains divergence-free.
→ pressure projects velocity into subspace of zero-divergence functions.


Velocity-Pressure formulation ~v = velocity, p = pressure (primitive variables)
~ · ~v = 0 ,
∇ mass eq.
~
∂~
v ~v=
+ ~v · ∇~ − ∇p + ν ∇2~v . momentum eq.
∂t ρ
• Divergence-free velocity field constraint expresses incompressibility.

• The pressure for incompressible flows is not a thermodynamic variable.
The pressure adjusts instantaneously so that velocity remains divergence-free.
→ pressure projects velocity into subspace of zero-divergence functions.
Poisson equation for pressure

Take divergence of momentum equation (Hp. appropriate differentiability)
~ · (∇p)
Note: ∇ ~ = ∇2 p, ∇2~v = ∇(
~ ∇~ · ~v ) − ∇
~ ×∇
~ × ~v ⇒ ∇
~ · (∇2~v ) = 0 (div curl ≡ 0)
~ · (~v · ∇~
⇒ ∇2 p = −ρ∇ ~ v ) + Neumann BC
∂p
BC for ∂~
n
derived by normal projection of momentum equation on boundary.
Note: BC for pressure Poisson equation (PPE) complicated issue [Gresho–Sani 87].
Velocity-Pressure formulation
Conservative form
~ · ~v = 0 ,
∇ ~ · ~v = 0 ,
∇
~

∂~v ~ v = − ∇p + ν ∇2~v .
+ ~v · ∇~
∂~v ~ p~~ ~
+ ∇ · ~v ⊗ ~v + I − ν ∇~v = 0 .
∂t ρ ∂t ρ
Integral form V = control volume, S = ∂V, ~n = outward unit normal vector
Z
~v · ~n dS = 0 ,
S
Z Z Z Z
∂~v 1 ~ v · ~n dS = 0 .
dV + ~v (~v · ~n)dS + p~n dS − ν ∇~
V ∂t S ρ S S

2D Velocity-Pressure formulation ~v = (u, v), ~

x = (x, y)
∂u ∂v
+ = 0, (1a)
∂x ∂y
2
∂2u

∂u ∂u ∂u 1 ∂p ∂ u
+u +v =− +ν + , (1b)
∂t ∂x ∂y ρ ∂x ∂x2 ∂y 2
2
∂2v

∂v ∂v ∂v 1 ∂p ∂ v
+u +v =− +ν 2
+ . (1c)
∂t ∂x ∂y ρ ∂y ∂x ∂y 2
Note: By using (1a) can rewrite convective terms in conservative form:
∂u ∂u ∂(u2 ) ∂(uv) ∂v ∂v ∂(uv) ∂(v 2 )

u +v = + , u +v = + .
∂x ∂y ∂x ∂y ∂x ∂y ∂x ∂y
In general different mathematical forms lead to different discrete equations.

Incompressible Navier–Stokes equations Vorticity-Stream function formulation
2D Vorticity-Stream Function formulation ~v = (u, v), ~

x = (x, y)
∂v ∂u
~ × ~v
~ =∇
Vorticity: ω ~ = ω~k , ω =
2D: ω − ~k ⊥ plane (x, y)
∂x ∂y
∂ψ ∂ψ ~ · ~v = 0 identically satisfied
Stream function ψ : u = , v=− ⇒ ∇
∂y ∂x
∂2ω ∂2ω

∂ω ∂ω ∂ω
+u +v =ν + , (2a)
∂t ∂x ∂y ∂x2 ∂y 2
∂2ψ ∂2ψ
2
+ = −ω . (2b)
∂x ∂y 2
(2a) Convection-diffusion equation for the vorticity ω(~

x, t) (parabolic)
∂ ∂
Obtained from momentum eqs., − ∂y (1b)+ ∂x (1c).
(2b) Poisson equation for the stream function ψ(~

x, t) (elliptic)
Obtained by using the definition of ψ into the definition of ω.

Incompressible Navier–Stokes equations Vorticity-Stream function formulation

3D extension of the 2D Vorticity-Stream Function formulation
Vorticity-Vector Potential formulation ~v = (u, v, w), ~

x = (x, y, z)
~ × ~v
~ =∇
Vorticity: ω
~ = (ψx , ψy , ψz ):
Express ~v as the curl of a vector potential ψ
~v = ∇ ~
~ ×ψ ~ · ~v = 0 identically satisfied
⇒ ∇
~ into definition of ω
Insert definition of ψ ~ × (∇
~ ⇒ ∇ ~ =ω
~ × ψ) ~
~ × (∇
Recall identity ∇ ~ = ∇(
~ × ψ) ~ ∇ ~ − ∇2 ψ
~ · ψ) ~
~ can be arbitraily chosen to satisfy ∇
Since vector potential ψ ~=0
~ ·ψ
~ = −~
⇒ ∇2 ψ ω ~
vector Poisson equation for ψ
+ vector convection-diffusion equation for ω
~ : ∂~
ω ~ × (~
+∇ ω × ~v ) = ν∇2 ω
~
∂t
Need to solve 3 elliptic equations + 3 parabolic equations.

∂ψz ∂ψy ∂ψy
~ componentwise: u =
~ ×ψ
(Note ~v = ∇ ∂y
− ∂z
, v = − ∂ψ
∂x
z
+ ∂ψx
∂z
, w= ∂x
− ∂ψx
∂y
.)

FD/FV Methods for Incompressible Navier–Stokes Equations
FD/FV Methods for the Incompressible Navier–Stokes Equations

Methods based on 2D Vorticity-Stream function formulation
Advantages:
- Pressure does not appear explicitly (can be obtained later).
- Incompressibility automatically satisfied (by definition of stream function).
Drawback: Extension to 3D complicated.
Methods based on Velocity-Pressure Formulation

• Artificial compressibility [Chorin 1967]
• Pressure Correction Methods
- MAC (Marker-And-Cell) [Harlow and Welch 1965]
- Projection Method [Chorin 1968, Temam 1969]
- SIMPLE (Semi-Implicit Method for Pressure-Linked Equations)
[Caretto et. al. 1972, Patankar and Spalding 1972)
- SIMPLER (SIMPLE Revised) [Patankar 1981]
- SIMPLEC [van Doormal and Raithby 1984]
- PISO (Pressure-Implicit with Splitting of Operators) [Issa 1985]

FD/FV Methods for Incompressible Navier–Stokes Equations Pressure Correction Methods
Pressure Correction Methods
Velocity-Pressure formulation ~v = velocity, p = pressure
~ · ~v = 0 ,
∇ mass (continuity) eq.
~
∂~v ~ v = − ∇p + ν ∇2~v .
+ ~v · ∇~ momentum eq.
∂t ρ
General idea of the algorithm

At each time level:
I Solve momentum equation with estimate of pressure field

(projection method: with no pressure field)
⇒ Obtain temporary velocity field ~v ∗ that does not satisfy zero-divergence constraint
I Find pressure field that ensures divergence-free velocity field

→ solution of pressure Poisson equation
~ · ~v = 0
I Correct velocity ~v ∗ to obtain new velocity field satisfying ∇

FD/FV Methods for Incompressible Navier–Stokes Equations Staggered vs. Collocated grids
Staggered vs. Collocated grids
Collocated grid ui , pi ∆x
i−2 i−1 i i+1 i+2
Staggered grid ui− 1 ui+ 1

2 2
(staggered because of a stability problem)
pi ∆x
i−2 i−1 i i+1 i+2
i− 32 i− 21 i+ 21 i+ 32
Velocities and pressure not defined in the same grid points.

Pressure p defined at cell centers, velocities u defined at cell edges.
(Here we assume uniform grids)

FD/FV Methods for Incompressible Navier–Stokes Equations Staggered vs. Collocated grids
2D staggered uniform grid ~v = (u, v), ~

x = (x, y) MAC Method Grid (1965)
j + 32
j+1 p
vi,j+ 1 u
j + 12 2
v
pi,j ui+ 1 ,j
2
j
j − 12
j−1
j − 32
i− 32 i− 21 i+ 12 i+ 23
i−1 i i+1
Pressure p defined at cell centers;

Velocities u and v defined in centers of vertical & horizontal cell faces, respectively.

Pressure Correction Methods Projection Method
Projection Method [Chorin 1967, Temam 1968] (Originally on colocated grid)
Space-time discretization of Navier–Stokes equations

Here assume 1st-order time-integration, explicit for convective and diffusive terms.
~ h · ~v n+1 = 0 ,
∇
~v n+1 − ~v n 1~
= −Ah (~v n ) − ∇ hp
n+1
+ ν∇2h ~v n .
∆t ρ
~v
Ah = discrete spatial differential operator for convective term ~v · ∇~
~ h , ∇2h = discrete spatial differential operators for ∇
∇ ~ and ∇2
Fractional Step Method

I. Integrate in time momentum equation with no pressure term
v ∗ −~
vn
~
∆t
= −Ah (~v n ) + ν∇2h ~v n ~v n → ~v ∗
II. Project ~v into subspace of divergence-free (soleinodal) functions
v∗
v n+1 −~ ~ h p n+1 ~ h · ~v n+1 = 0
~
∆t
= − ρ1 ∇ ~v ∗ → ~v n+1 with ∇

Projection Method
I. Time-integration with no pressure term

v ∗ −~
vn
~
∆t
= −Ah (~v n ) + ν∇2h ~v n ⇒ ~v ∗ = ~v n + ∆t(−Ah (~v n ) + ν∇2h ~v n )
Velocity ~v ∗ does not satisfy zero-divergence constraint (mass eq. = incompressibility).
II. Projection
v∗
v n+1 −~ ~ h p n+1 ~ h pn+1
~
∆t
= − ρ1 ∇ ⇒ ~v n+1 = ~v ∗ − ∆t
ρ
∇
~ h · ~v n+1 = ∇
Take divergence: ∇ ~ h · ~v ∗ − ∆t ~ h · (∇
∇ ~ h pn+1 )
ρ
~ h · ~v n+1 = 0
Mass equation: ∇
ρ ~
⇒ ∇2h pn+1 = ∇
∆t h
· ~v ∗ Pressure Poisson Equation
Pressure field that projects velocity into subspace of soleinodal functions.

Projection Method for the Incompressible Navier–Stokes equations
Algorithm
1. Find temporary velocity ~v ∗ by integrating convective and diffusive terms only
~v ∗ = ~v n + ∆t(−Ah (~v n ) + ν∇2h ~v n )
2. Find pressure field needed to make velocity field divergence-free
ρ ~
∇2h pn+1 = ∇h · ~v ∗
∆t this is a splitted operator resolution
3. Correct velocity by adding the pressure gradient (projection)
∆t ~ n+1
~v n+1 = ~v ∗ − ∇h p
ρ

Pressure Correction Methods MAC Method
Marker-And-Cell (MAC) Method [Harlow and Welch 1965]
Originally derived for free surface flows on staggered grid.

Space-time discretization of Navier–Stokes equations
~ h · ~v n+1 = 0 ,
∇
~v n+1 − ~v n 1~
= −Ah (~v n ) − ∇ hp
n+1
+ ν∇2h ~v n .
∆t ρ
Take divergence of momentum equation:

∇ v n+1 −∇
~ h ·~ vn
~ h ·~
~ h · Ah (~v n ) − 1 ∇
~ · (∇
~ h p n+1 ) + ν ∇
~ h · (∇2h ~v n ).
∆t
= −∇ ρ h
~ h · ~v n = 0 and ∇
Using ∇ ~ h · ~v n+1 = 0
~ h · Ah (~v n )
⇒ ∇2h pn+1 = −ρ∇ Pressure Poisson Equation
~ h · (∇
Note: ∇ ~ h p n+1 ) = ∇2h pn+1 and
~ h · (∇2h ~v n ) = ∇
∇ ~ h · (∇
~ h (∇
~ h · ~v n ) − ∇
~h×∇
~ h × ~v n ) = ∇2h · (∇
~ h · ~v n ) = 0.

Pressure Correction Methods MAC Method
MAC vs. Projection Method

Using explicit time-integration for convective and diffusive terms and staggered grid:
MAC and Projection methods can be equivalent although written in different form.
MAC vs. Projection

the choice of one of those methods
1. depends on the problem we study 1. Integration without pressure
mac was created for a free surface
flow
~v n = ~v n
~v ∗ = ~v n + ∆t(−Ah (~v n ) + ν∇2h ~v n )
2. Pressure Poisson equation
~ h · Ah (~v n )
∇2h pn+1 = −ρ∇ ∇2h pn+1 = ρ ~
∇ · ~v ∗
∆t h
3. Projection into incompressible field
~v n+1 = ~v n +∆t(−Ah (~v n ) + ν∇2h ~v n )− ∆t ~ h pn+1

∇ ~v n+1 = ~v ∗ − ∆t ~ h pn+1
∇
ρ ρ

Projection Method Space discretization on staggered grid
Spatial discretization on a 2D staggered uniform grid

~v = (u, v) Hp. uniform grid ∆x = ∆y ≡ h (grid spacing)
Finite Volume approach: integral form of Navier–Stokes eqs. over Control Volume
Cell averages: ui+ 1 ,j ≈ |V1u | Vu u dV, vi,j+ 1 ≈ |V1v | Vv v dV, pi,j ≈ |V1p | Vp p dV
R R R
2 2
j+1 p
vi,j+ 1 u
2
j + 12 v
pi,j ui+ 1 ,j
2
j
CV for mass equation = Vp

j − 12
CV for x-momentum equation = Vu
j−1 CV for y-momentum equation = Vv
i−1 i i+1
i− 12 i+ 12
Finite Difference approach: use differential Navier–Stokes equations
pointwise approx.: ui+ 1 ,j ≈ u(xi+ 1 , yj ), vi,j+ 1 ≈ v(xi , yj+ 1 ), pi,j ≈ p(xi , yj )
2 2 2 2

Mass equation
j +1 p
Mass conservation law vi,j+ 1 u
R j + 12 2
v
Sp
~v · ~n dS = 0 pi,j ui+ 1 ,j
2
j
hui+ 1 ,j −hui− 1 ,j +hvi,j+ 1 −hvi,j− 1 = 0
2 2 2 2
j − 12
⇒ ui+ 1 ,j − ui− 1 ,j + vi,j+ 1 − vi,j− 1 = 0
2 2 2 2 j −1
i−1 i i+1
n = outward unit normal
~
i− 12 i+ 21
Note: same result with Finite Difference approach
~ ·~
PDE: ∇ u= ∂u
+ ∂v
=0
∂x ∂y
u −u v −v
∂u
i+ 1 ,j i− 1 ,j ∂v
i,j+ 1 i,j− 1
Take centered differences: ∂x i,j
≈ 2
h
2
, ∂x i,j
≈ 2
h
2
u −u v −v
i+ 1 ,j i− 1 ,j i,j+ 1 i,j− 1
~ ·~
∇ u = 0 at (i, j) ⇒ 2 2
+ 2 2
=0
h h

x-momentum equation
∂u 1 ~ · ~n dS = 0
R R R R
Vu ∂t
dV + u(~v · ~n)dS +
Su ρ Su
p nx dS − ν Su
∇u
j +1 p
Unsteady term
vi,j+ 1 u
∂u
= h2 ∂t
∂ 1
j + 12
R R 2
Vu ∂t
dV h2 Vu
udV v
pi,j ui+ 1 ,j
∂ui+1/2,j j 2
→ h2 ∂t
Pressure term j − 12
R
Su
p nx dS → h(pi+1,j − pi,j ) j −1
i−1 i i+1
i− 21 i+ 12
Convective term
u(~v · ~n)dS → h u2i+1,j − u2i,j + (uv)i+ 1 ,j+ 1 − (uv)i+ 1 ,j− 1
R
Su 2 2 2 2
Diffusive term

~ · ~n dS → h ∂u ∂u ∂u ∂u
R
Su
∇u ∂x i+1,j
− ∂x i,j
+ ∂y
− ∂y
i+ 1
2
1
,j+ 2 i+ 1
2
,j− 1
2
~
nEast = (1, 0), ~
nNorth = (0, 1), ~ nSouth = (0, −1)
nWest = (−1, 0), ~
x-momentum equation

u(~v · ~n)dS → h u2i+1,j − u2i,j + (uv)i+ 1 ,j+ 1 − (uv)i+ 1 ,j− 1
R
Convective term Su 2 2 2 2
2
Interpolate the velocities: (u2 )i,j = 21 (ui+ 1 ,j + ui− 1 ,j )
2 2

(uv)i+ 1 ,j+ 1 = 12 (ui+ 1 ,j+1 + ui+ 1 ,j ) 1
2
(vi,j+ 1 + vi+1,j+ 1 ) .
2 2 2 2 2 2

~ · ~ndS → h ∂u ∂u ∂u ∂u
R
Diffusive term S ∇u ∂x i+1,j
− ∂x i,j
+ ∂y 1 1
− ∂y
u 1 i+ 2 ,j+ 2 i+ 2 ,j− 1
2
→ ui+ 3 ,j + ui− 1 ,j + ui+ 1 ,j+1 + ui+ 1 ,j−1 − 4ui+ 1 ,j

2 2 2 2 2
j +1 p
Same as FD 5-point (∇2 u)i+ 1 ,j vi,j+ 1 u
2
j + 12 2
v
u −u pi,j ui+ 1 ,j
∂u
i+ 1 ,j i− 1 ,j 2
∂x i,j
= 2
h
2 j
u −u
i+ 1 ,j+1 i+ 1 ,j j − 12

∂u 2 2
∂y
= h
i+ 1
2
,j+ 1
2
j −1
i−1 i− 1 i i+ 12 i+1
2
y-momentum equation
∂v 1 ~ · ~n dS = 0
R R R R
Vv ∂t
dV + v(~v · ~n)dS +
Sv ρ Sv
p ny dS − ν Sv
∇v
j +1 p
vi,j+ 1 u
j + 12 2
v
∂v
∂v
h2 i,j+1/2
R
dV → pi,j ui+ 1 ,j
Vv ∂t ∂t 2
j
j − 12
R
Sv
p ny dS → h(pi,j+1 − pi,j )
j −1
i−1 i i+1
i− 21 i+ 12

2 2
R
v(~v · ~n)dS → h vi,j+1
Sv
− vi,j + (uv)i+ 1 ,j+ 1 − (uv)i− 1 ,j+ 1
2 2 2 2
2
2 1
(v )i,j = 2 (vi,j+ 1 + vi,j− 1 )
2 2
~ · ~n dS → v
R
Sv
∇v i,j+ 3 + vi,j− 1 + vi+1,j+ 1 + vi−1,j+ 1 − 4vi,j+ 1
2 2 2 2 2

Space discretization of Navier–Stokes equations on staggered grid

Mass ui+ 1 ,j − ui− 1 ,j + vi,j+ 1 − vi,j− 1 = 0
2 2 2 2
x−momentum
∂u
i+ 1 ,j

∂t
2
= − h1 (u2 )i+1,j − (u2 )i,j + (uv)i+ 1 ,j+ 1 − (uv)i+ 1 ,j− 1
2 2 2 2
1
− hρ (pi+1,j − pi,j )

ν
+ h2 ui+ 3 ,j + ui− 1 ,j + ui+ 1 ,j+1 + ui+ 1 ,j−1 − 4ui+ 1 ,j
2 2 2 2 2
y-momentum
∂v
i,j+ 1

2
∂t
= − h1 (v 2 )i,j+1 − (v 2 )i,j + (uv)i+ 1 ,j+ 1 − (uv)i− 1 ,j+ 1
2 2 2 2
1
− hρ (pi,j+1 − pi,j )

ν
+ h2 vi,j+ 3 + vi,j− 1 + vi+1,j+ 1 + vi−1,j+ 1 − 4vi,j+ 1 .
2 2 2 2 2
2 2
(u2 )i,j = 21 (ui+ 1 ,j + ui− 1 ,j ) (v 2 )i,j = 12 (vi,j+ 1 + vi,j− 1 )
2 2 2 2
(uv)i+ 1 ,j+ 1 = 12 (ui+ 1 ,j+1 + ui+ 1 ,j ) 1
(v
2 i,j+ 1
+ vi+1,j+ 1 )
2 2 2 2 2 2

Projection Method Space-time discretization
Space-time discretization of Navier–Stokes equations on staggered grid

1st-order time-integration, explicit for convective and diffusive terms.
~ h · ~v n+1 = 0,
∇
v n+1 −~
vn ~ h p n+1 +ν∇2h ~v n .
~
∆t
= −Ah (~v n )− ρ1 ∇
un+1
i+ 1 ,j
− un+1
i− 1 ,j
n+1
+ vi,j+ n+1
1 − vi,j− 1 = 0 ,
2 2 2 2
un+1 −un 1
i+ 1 ,j i+ ,j

2
∆t
2
= − h1 (u2 )n 2 n n n
i+1,j − (u )i,j + (uv)i+ 1 ,j+ 1 − (uv)i+ 1 ,j− 1
2 2 2 2

1
− hρ (pn+1
i+1,j − pn+1
i,j ) + ν
h2
u n
i+ 3 ,j
+ u n
i− 1 ,j
+ u n
i+ 1 ,j+1
+ u n
i+ 1 ,j−1
− 4uni+ 1 ,j
,
2 2 2 2 2
v n+1 1 −v n
i,j+ 1
i,j+

2
∆t
= − h1 (v 2 )n
2 2 n n n
i,j+1 − (v )i,j + (uv)i+ 1 ,j+ 1 − (uv)i− 1 ,j+ 1
2 2 2 2

1
− hρ (pn+1
i,j+1 − p n+1
i,j ) + ν
h2
v n
i,j+ 3 + v i,j− 1 + v n
i+1,j+ 1 + v n
i−1,j+ 1 − n
4vi,j+ 1 .
2 2 2 2 2
(u2 )n 2 n n
i,j , (v )i,j , (uv)i+ 1 ,j+ 1 interpolated
2 2

Projection Method Pressure Poisson Equation
Projection Method - Pressure Poisson equation

Recall derivation of Pressure Poisson Equation
Projection Step ~v n+1 = ~v ∗ − ∆t ~ h pn+1 (*)
∇
ρ
~ h · (*) and use continuity equation ∇

Take divergence ∇ ~ h · ~v n+1 = 0
ρ ~
⇒ ∇2h pn+1 = ∇
∆t h
· ~v ∗
For the considered space discretization:
un+1
i+ 1 ,j
= u∗i+ 1 ,j − ∆t
hρ
(pn+1
i+1,j − pn+1
i,j )
2 2
n+1 ∗
vi,j+ 1 = vi,j+ 1 −
∆t
hρ
(pn+1
i,j+1
n+1
− pi,j )
2 2
Insert in the continuity equation un+1

i+ 1 ,j
− un+1
i− 1 ,j
n+1
+ vi,j+ n+1
1 − vi,j− 1 = 0
2 2 2 2
⇒ Pressure Poisson Equation

pn+1 n+1 n+1 n+1 n+1
i+1,j + pi−1,j + pi,j+1 + pi,j−1 − 4pi,j =
ρ
h∆t
u∗i+ 1 ,j − u∗i− 1 ,j + vi,j+
∗ ∗
1 − vi,j− 1
2 2 2 2
Same result with centered FD:

pn+1 n+1 n+1 n+1 n+1
i+1,j +pi−1,j +pi,j+1 +pi,j−1 −4pi,j ρ
h2
= (u∗ 1 − u∗ + v∗ − v∗ )
h∆t i+ ,j 2
i− 1
2
,j i,j+ 1
2
i,j− 1
2
Projection Method Staggered vs. Collocated grids
Pressure Correction Methods - Staggered vs. Collocated grids

Centered Finite Difference approximations for incompressible Navier–Stokes
equations may lead to odd-even decoupling → pressure oscillations.
1D Projection method
ui , pi ∆x = h
Collocated grid
i−2 i−1 i i+1 i+2
∂u un+1 n+1
i+1 −ui−1 ∂u ∂u2 ∂p 2
Mass eq.: ∂x
=0 → 2h
=0 Momentum eq.: ∂t
+ ∂x
= − ρ1 ∂x + ν ∂∂xu2
u∗ n
i −ui
(un 2 n
i+1 ) −(ui−1 )
2
un n n
i+1 −2ui +ui−1
Temporary velocity u∗i : ∆t
+ 2h
=ν h2
Projection: un+1
i = u∗i − ∆t
2hρ
(pn+1
i+1 − pn+1
i−1 )
pn+1 n+1
i+2 −2pi +pn+1
PPE: 4h2
i−2
= ρ
2h∆t
(u∗i+1 − u∗i−1 )
Stencil of Laplacian of p is (i + 2, i, i − 2) → only odd or even indices

Can lead to decoupling of pressure field resulting in high frequency oscillations.
∂ρ
Note: Related to absence of ∂t
in mass equation. Compressible flow equations:
decoupling does not occur thanks to density-velocity coupling in mass equation.
Projection Method Staggered vs. Collocated grids
Pressure Correction Methods - Staggered vs. Collocated grids

1D Projection method ui− 1 ui+ 1
2 2
Staggered grid
pi ∆x = h
[Harlow-Welch 65]
i−2 i−1 i i+1 i+2
i− 32 i− 21 i+ 21 i+ 32
un+1
1 −u
n+1
1
i+ i−
Mass eq.: ∂u
∂x
=0 → 2
h
2
=0 Momentum eq. → u∗i+ 1
2
Projection: un+1
i+ 1
= u∗i+ 1 − ∆t
hρ
(pn+1
i+1 − pn+1
i )
2 2
pn+1 n+1
i+1 −2pi +pn+1
PPE: h2
i−1
= ρ
h∆t
(u∗i+ 1 − u∗i− 1 )
2 2
In the discrete PPE pressure and velocity fully coupled

→ problem of odd-even decoupling eliminated
Note: Collocated grids may be preferable for e.g. body-fitted grids, AMR,...
Various ways to overcome decoupling problem (e.g. upwind for convective terms).
Projection Method Lid-driven cavity flow test
Lid-driven cavity flow problem

UL
Lid
Staggered grid
Ny p
u
v
ghost p
ghost u
ghost v
Green line = physical domain

Ghost values used for BC
1 Nx

Projection Method Lid-driven cavity flow test
LUL
Lid-driven cavity flow problem Projection method, Re = ν = 100, N = 50
Ghia–Ghia–Shin 1982

Projection Method Stability conditions
Stability conditions
Explicit forward in time centered in space projection scheme
For stability necessary that prediction step (no pressure term included) is stable,
numerical evidence indicates that this also sufficient.
Stability condition (FTCS method 2D convection-diffusion equation)
h2

2ν
∆t ≤ min , .
max(u2 + v 2 ) 4ν
1
Nondimensional variables: ν = Re
, Re = Reynolds number
∆t → 0 for Re → 0 and Re → ∞
Improvements of method:
• Better stability for convective terms for high Reynolds numbers.
• Implicit viscous terms for low Reynolds numbers.

MF 206 A22 Lect 3

Uploaded by

Copyright:

Available Formats

MF 206 A22 Lect 3

Uploaded by

Document Information

Original Title

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

MF 206 A22 Lect 3

Uploaded by

Copyright:

Available Formats

Introduction to Computational Fluid Dynamics

Pressure Correction Methods for the

MF206 (ENSTA Paris) Lecture 3 2021-2022 1 / 34

1 Finite Difference and Finite Volume Methods

2 Incompressible Navier–Stokes equations

3 Pressure Correction Methods

MF206 (ENSTA Paris) Lecture 3 2021-2022 2 / 34

Finite Difference and Finite Volume Methods

Finite Difference Method

approximation of pointwise values of solution

Finite Volume Method

approximation of solution averaged over CV = V

MF206 (ENSTA Paris) Lecture 3 2021-2022 3 / 34

Finite Difference and Finite Volume Methods

Example: Unsteady heat conduction in a solid (or fluid at rest)

ρ = density (Hp. const.), e = specific internal energy, e = ch T

MF206 (ENSTA Paris) Lecture 3 2021-2022 4 / 34

xi− 3 xi− 1 xi+ 1 xi+ 3

Simple uniform Cartesian FD and FV grids

xi−2 xi−1 xi xi+1 xi+2

xi− 3 xi− 1 xi+ 1 xi+ 3

MF206 (ENSTA Paris) Lecture 3 2021-2022 6 / 34

Navier–Stokes equations for incompressible flows

Unknowns: velocity ~v (~x, t), pressure p(~

Mixed set of elliptic-parabolic equations.

Hp. constant flow properties ⇒ decoupling of thermal problem ⇒

MF206 (ENSTA Paris) Lecture 3 2021-2022 7 / 34

Incompressible Navier–Stokes equations

• Divergence-free velocity field constraint expresses incompressibility.

MF206 (ENSTA Paris) Lecture 3 2021-2022 8 / 34

Incompressible Navier–Stokes equations

• Divergence-free velocity field constraint expresses incompressibility.

Poisson equation for pressure

Incompressible Navier–Stokes equations

Integral form V = control volume, S = ∂V, ~n = outward unit normal vector

MF206 (ENSTA Paris) Lecture 3 2021-2022 9 / 34

Incompressible Navier–Stokes equations

2D Velocity-Pressure formulation ~v = (u, v), ~

Note: By using (1a) can rewrite convective terms in conservative form:

∂u ∂u ∂(u2 ) ∂(uv) ∂v ∂v ∂(uv) ∂(v 2 )

In general different mathematical forms lead to different discrete equations.

MF206 (ENSTA Paris) Lecture 3 2021-2022 10 / 34

Incompressible Navier–Stokes equations

2D Vorticity-Stream Function formulation ~v = (u, v), ~

(2a) Convection-diffusion equation for the vorticity ω(~

(2b) Poisson equation for the stream function ψ(~

MF206 (ENSTA Paris) Lecture 3 2021-2022 11 / 34

Incompressible Navier–Stokes equations

Vorticity-Vector Potential formulation ~v = (u, v, w), ~

Need to solve 3 elliptic equations + 3 parabolic equations.

MF206 (ENSTA Paris) Lecture 3 2021-2022 12 / 34

FD/FV Methods for the Incompressible Navier–Stokes Equations

Methods based on Velocity-Pressure Formulation

MF206 (ENSTA Paris) Lecture 3 2021-2022 13 / 34

Pressure Correction Methods

Velocity-Pressure formulation ~v = velocity, p = pressure

General idea of the algorithm

I Solve momentum equation with estimate of pressure field

I Find pressure field that ensures divergence-free velocity field