Part II: Finite Difference/Volume Discretisation For CFD

Lehrstuhl Informatik V
Part II: Finite Difference/Volume Discretisation for

CFD
Finite Volume Method of the Advection-Diffusion Equation
A Finite Difference/Volume Method for the Incompressible

Navier-Stokes Equations
Marker-and-Cell Method, Staggered Grid
Spatial Discretisation of the Continuity Equation
Spatial Discretisation of the Momentum Equations
Time Discretisation
Chorin Projection
Implementation Aspects
Tobias Neckel: Scientific Computing I

Module 9: Case Study – Computational Fluid Dynamics, Winter 2013/2014 19
Finite Volume Method –

Advection-Diffusion Equation
• compute tracer concentration q with diffusion β and convection v :
−βqxx + (vq)x = 0 on Ω = (0, 1)
with boundary conditions q(0) = 1 and q(1) = 0.

• equidistant grid points xi = ih, grid cells [xi , xi+1 ]
• back to representation via conservation law (for one grid cell):
Z xi+1 xi+1
∂
F (x) dx = F (x) =0
xi ∂x xi
with F (x) = F (q(x)) = −βqx (x) + vq(x).

• we need to compute the flux F at the boundaries of the grid cells;
however, assume q(x) piecewise constant within the grid cells


Advection-Diffusion Equation (2)
• wanted: compute F (xi ) with F (q(x)) = −βqx (x) + vq(x)
• where q(x) := qi for each Ωi = [xi , xi+1 ]
• computing the diffusive flux is straightforward:
q(xi+1 ) − q(xi )

−βqx x
= −β
i+1 h
• options for advective flux vq:
• symmetric flux:
vq(xi ) + vq(xi+1 )
vq xi+1
=
2
• “upwind” flux:

vq(xi ) if v > 0
vq xi+1
=
vq(xi+1 ) if v < 0


Advection-Diffusion Equation (3)
• system of equations: for all i
xi+1

F (x) = F (xi+1 ) − F (xi ) = 0
xi
• for symmetric flux:

q(xi+1 ) − 2q(xi ) + q(xi−1 ) q(xi+1 ) − q(xi−1 )
−β 2
+ v =0
h 2h
leads to non-physical behaviour as soon as β < vh
2
(observe signs of matrix elements!)
• system of equations for upwind flux (assume v > 0):
q(xi+1 ) − 2q(xi ) + q(xi−1 ) q(xi ) − q(xi−1 )
−β 2
+ v =0
h h
→ stable, but overly diffusive solutions (positive definite matrix)
Marker-and-Cell Method – Staggered Grid

Marker-and-Cell method (Harlow and Welch, 1965):
• discretization scheme: Finite Differences
• can be shown to be equivalent to Finite Volumes, however
• based on a so-called staggered grid:
• Cartesian grid (rectangular grid cells),
with cell centres at xi,j := (ih, jh), e.g.
• pressure located in cell centres
• velocities (those in normal direction) located on cell edges

Spatial Discretisation – Continuity Equation:
• mass conservation: discretise ∇ · ~

u
→ evaluate derivative at cell centres, allows central derivatives:

∂u ∂v ui,j − ui−1,j vi,j − vi,j−1
(∇ · ~u ) i,j = + ≈ +
∂x i,j ∂y i,j h h
remember: ui,j and vi,j located on cell edges

u ) i,j := (∇ · ~u )x
• notation: (∇ · ~
i,j
(evaluate expression at cell centre xi,j )

Spatial Discretisation – Pressure Terms
• note: velocities located on midpoints of cell edges

∂u ∂v
= ... , = ...
∂t i+ 1 ,j ∂t i,j+ 1
2 2
=⇒ all derivatives need to be approximated at midpoints of cell

edges!
• pressure term ∇p: central differences for first derivatives
(as pressure is located in cell centres)

∂p pi+1,j − pi,j ∂p pi,j+1 − pi,j
≈ , ≈
∂x i+ 1 ,j h ∂y i,j+ 1 h
2 2

Spatial Discretisation – Diffusion Term

• for diffusion term ∆~
u : use standard 5- or 7-point stencil
• 2D:
ui−1,j + ui,j−1 − 4ui,j + ui+1,j + ui,j+1

∆u i,j ≈
h2
• 3D:
ui−1,j,k + ui,j−1,k + ui,j,k−1 − 6ui,j,k + ui+1,j,k + ui,j+1,k + ui,j,k+1

∆u ≈
i,j,k h2
Spatial Discretisation – Convection Terms

• treat derivatives of nonlinear terms (~
u · ∇)~u :
• central differences (for momentum equation in x-direction):

∂u ui+1,j − ui−1,j ∂u
ui,j+1 − ui,j−1
u ≈ ui,j , v ≈v x
∂x i+ 1 ,j 2h ∂y i+ 1 ,j i+ 1 ,j
2
2h
2 2
1

with v xi+ 1 ,j
= 4 vi,j + vi,j−1 + vi+1,j + vi+1,j−1
2
• upwind differences (for momentum equation in x-direction):

∂u ui,j − ui−1,j ∂u ui,j − ui,j−1
u ≈ ui,j , v ≈ v
∂x x 2h ∂y x xi+ 1 ,j
2
2h
i+ 1 ,j i+ 1 ,j
2 2

if ui,j > 0 and v x >0
i+ 1 ,j
2
• mix of central and upwind differences possible and used

Time Discretisation
• recall the incompressible Navier-Stokes equations:
∇ · ~u = 0
∂ 1
~u + (~u · ∇)~u = −∇p + ∆~u + f
∂t Re
• note the role of the unknowns:
→ 2 or 3 equations for velocities (x, y , and z component)
resulting from momentum conservation
→ 4th equation (mass conservation) to “ close” the system;
required to determine pressure p
→ however, p does not occur explicitly in mass conservation
• possible approach: Chorin’s projection method
→ p acts as a variable to enforce the mass conservation as
“side condition”

Time Discretisation – Chorin Projection

• continuous NSE:

∂
ρ ~u + (~u · ∇)~u + ∇p − µ∆~u = f
∂t
∇ · ~u = 0
• semi-discrete NSE (after spatial discretisation):
A~u˙ h + D~uh + C(~uh )~uh − M T ph = fh , (3)

M ~uh = 0, (4)
• derive semi-discrete continuity equation (4) w.r.t. t:
M ~u˙ h = 0. (5)
u˙ h and insert into (5)
• formally invert Eq. (3) w.r.t. ~
⇒ discrete pressure Poisson equation (PPE):
(MA−1 M T ) ph = MA−1 (−f + D~uh + C(~uh )~uh ), (6)
| {z }
=:Q
Time Discretisation – Chorin Projection
• updating the velocities:

• discretise time with explicit method (explicit Euler, e.g.):
~uh(n+1) − ~uh(n)
= ...
τ
• update velocity “correctly”: solve momentum equations (3)
• use pressure ph of PPE solution to compute discrete
pressure gradients M T ph :

= ~uh + τ −D~uh − C(~uh )~uh + ~fh + M T ph
(n+1) (n) (n) (n) (n) (n)
⇒ ~uh
• costs: solve 1 linear system of equations in each time step!

Implementation
• geometry representation as a flag field (Marker-and-Cell)
flag field as an array of booleans:

000000000000000000000000
000000000000000000000000
000000001100000000000000
000000000110000000000000
000000000011000000000000
000000000001100000000000
000000000000000000000000
000000000000000000000000
• input data (boundary conditions) and output data (computed

results) as arrays

Implementation (2)
Lab course “Scientific Computing – Computational Fluid Dynamics”:

• modular C/C++ code
• parallelization:
• simple data parallelism, domain decomposition
• straightforward MPI-based parallelization (exchange of
ghost layers)
• target architectures:
• parallel computers with distributed memory
• clusters
• possible extensions:
• free-surface flows (“the falling drop”)
• multigrid solver for the pressure equation
• heat transfer

Part III: The Shallow Water Equations and Finite

Volumes Revisited
The Shallow Water Equations

Modelling Scenario: Tsunami Simulation
Finite Volume Discretisation

Central and Upwind Fluxes
Lax-Friedrichs Flux
Towards Tsunami Simulation

Wave Speed of Tsunamis
Treatment of Bathymetry Data

The Shallow Water Equations

     
h hu hv
∂   ∂  2 1 2 ∂ 
hu + hu + 2 gh + huv  = S(t, x, y )
∂t ∂x ∂y
hv huv hv 2 + 12 gh2
Comments on modelling:
• generalized 2D hyperbolic PDE: q = (h, hu, hv )T
∂ ∂ ∂
q+ F (q) + G(q) = S(t, x, y )
∂t ∂x ∂y
derived from conservations laws for mass and momentum

• may be derived by vertical averaging from the 3D incompressible
Navier-Stokes equations
• compare to Euler equations: density ρ vs. water depth h

The Ocean as “Shallow Water”??

• compare horizontal (∼ 1000 km) to vertical (∼ 5 km) length scale
• wave lengths large compared to water depth
• vertical flow may be neglected; movement of the “entire water
column”


column”


column”


column”

Modelling Scenario: Tsunami Simulation (2)
Tsunami Modelling with the Shallow Water equations:

• source term S(x, y) includes bathymetry data (i.e., elevation of
ocean floor)
• Coriolis forces, friction, etc., as possible further terms
• boundary conditions are difficult: coastal inundation, outflow at
domain boundaries

ocean floor)
domain boundaries

ocean floor)
domain boundaries

ocean floor)
domain boundaries
Finite Volume Discretisation

• discretise system of PDEs
∂ ∂ ∂
q+ F (q) + G(q) = S(t, x, y )
∂t ∂x ∂y
• with      
h hu hv
q := hu  F (q) := hu 2 + 12 gh2  G(q) :=  huv 
hv huv hv 2 + 12 gh2
• basic form of numerical schemes:

(n+1) (n) τ (n) (n)
τ
(n) (n)

Qi,j = Qi,j − Fi+ 1 ,j − Fi− 1 ,j − Gi,j+ 1 − Gi,j− 1
h 2 2 h 2 2
(n) (n)
where Fi+ 1 ,j , Gi,j+ 1 , . . . approximate the flux functions F (q) and
2 2
G(q) at the grid cell boundaries

Central and Upwind Fluxes

(n) (n)
• define fluxes F 1 , G , . . . via 1D numerical flux function F:
i+ ,j i,j+ 1 2 2
(n) (n) (n) (n) (n) (n)

Fi+ 1 =F Qi , Qi+1 Gj− 1 =F Qj−1 , Qj
2 2
• central flux:
(n) (n) (n) 1 (n) (n)

Fi+ 1 = F Qi , Qi+1 := F Qi + F Qi+1
2 2
leads to unstable methods for convective transport
• upwind flux (here, for h-equation, F (h) = hu):
(
hu if u 1 >0
(n) (n) (n) i i+ 2
Fi+ 1 = F hi , hi+1 :=
2 hu i+1 if u i+ 1 < 0
2
stable, but includes artificial diffusion

(Local) Lax-Friedrichs Flux

• classical Lax-Friedrichs method uses as numerical flux:
(n) (n) (n) 1 (n) (n)

h
(n) (n)
Fi+ 1 =F Qi , Qi+1 := F Qi + F Qi+1 − Qi+1 −Qi
2 2 2τ
• can be interpreted as central flux plus diffusion flux:
(n) (n)
h (n) (n) h2 Qi+1 − Qi
Qi+1 − Qi = ·
2τ 2τ h
h2
with diffusion coefficient where c := τh is some kind of
2τ ,
velocity (“one grid cell per time step”)
• idea of local Lax-Friedrichs method: use the “appropriate”
velocity
(n) 1 (n) (n)

ai+ 1
(n) (n)
Fi+ 1 := F Qi + F Qi+1 − 2
Qi+1 − Qi
2 2 2

Wave Speed of Tsunamis

• consider the 1D case

∂ h ∂ hu
+ =0
∂t hu ∂x hu 2 + 12 gh2
• with q = (q1 , q2 )T := (h, hu)T , we obtain

∂ q1 ∂ q2
+ =0
∂t q2 ∂x q22 /q1 + 12 gq12
• write in convective form:

∂ q1 0 ∂ q1
+f =0
∂t q2 ∂x q2
with

∂f1 /∂q1 ∂f1 /∂q2 0 1 0 1
f0 = = =
∂f2 /∂q1 ∂f2 /∂q2 −q22 /q12 + gq1 2q2 /q1 −u 2 + gh 2u

Wave Speed of Tsunamis (2)

• compute eigenvectors and eigenvalues of f 0 :

1/2
p 1/2 1p
λ = u ± gh r =
u± gh
• and then with f 0 = RΛR −1 , where R := (r 1 , r 2 ) and
Λ := diag(λ1 , λ2 ), we can diagonalise the PDE:

∂ w1 ∂ w1
+Λ = 0, w = R −1 q
∂t w2 ∂x w2
• for small changes in h and small velocities, we thus obtain
p that
waves are “advected” (i.e., travel) at speed λ1/2 ≈ ± gh
• recall local Lax-Friedrichs method:
1 ai+ 1
(n) (n) (n) (n) (n)
Fi+ 1 := F Qi + F Qi+1 − 2
Qi − Qi−1
2 2 2
→ choose ai+ 1 = max{λk }
2

Shallow Water Equations with Bathymetry
h
b
       
h hu hv 0
∂   ∂  2 1 2 ∂ 
hu + hu + 2 gh + huv  = −(ghb)x 
∂t ∂x ∂y
hv huv hv 2 + 12 gh2 −(ghb)y
Questions for numerics:

• treat (bh)x and (bh)y as source terms or include these into flux
computations?
• preserve certain properties of solutions – e.g., “lake at rest”
Shallow Water Equations with Bathymetry (2)

Consider “Lake at Rest” Scenario:
• “at rest”: velocities u = 0 and v = 0
• examine local Lax-Friedrichs flux in h equation:
(n) 1 (n) (n)

ai+ 1
(n) (n)
Fi+ 1 = (hu)i + (hu)i+1 − 2
hi+1 − hi =0
2 2 2
(n) (n)
ai+ 1 (n) (n)
ai− 1 (n) (n)
⇒ Fi+ 1 − Fi− 1 =− 2
hi+1 − hi + 2
hi − hi−1 =0
2 2 2 2
p
• note: ai± 1 ≈ gh and if bi−1 6= bi 6= bi+1 then hi−1 6= hi 6= hi+1
2
• thus: “lake at rest” not an equilibrium solution for local

Lax-Friedrichs flux
Additional problems:
• complicated numerics close to the shore
• in particular: “wetting and drying” (inundation of the coast)
References and Literature
Course material is mostly based on:

• R. J. LeVeque: Finite Volume Methods for Hyperbolic Equations,
Cambridge Texts in Applied Mathematics, 2002.
• M. Griebel, T. Dornseifer and T. Neunhoeffer: Numerical
Simulation in Fluid Dynamics: A Practical Introduction,
SIAM Monographs on Mathematical Modeling and Computation,
SIAM, 1997.
Shallow Water Code SWE:

→ http://www5.in.tum.de/SWE/


Part II: Finite Difference/Volume Discretisation For CFD

Diunggah oleh

Informasi Dokumen

Judul Asli

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Part II: Finite Difference/Volume Discretisation For CFD

Diunggah oleh

Hak Cipta:

Format Tersedia

Lehrstuhl Informatik V

Part II: Finite Difference/Volume Discretisation for

Finite Volume Method of the Advection-Diffusion Equation

A Finite Difference/Volume Method for the Incompressible

Tobias Neckel: Scientific Computing I

Finite Volume Method –

−βqxx + (vq)x = 0 on Ω = (0, 1)

with boundary conditions q(0) = 1 and q(1) = 0.

with F (x) = F (q(x)) = −βqx (x) + vq(x).

Tobias Neckel: Scientific Computing I

Finite Volume Method –

Tobias Neckel: Scientific Computing I

Finite Volume Method –

• for symmetric flux:

Marker-and-Cell Method – Staggered Grid

Tobias Neckel: Scientific Computing I

Spatial Discretisation – Continuity Equation:

• mass conservation: discretise ∇ · ~

remember: ui,j and vi,j located on cell edges

Tobias Neckel: Scientific Computing I

Spatial Discretisation – Pressure Terms

• note: velocities located on midpoints of cell edges

=⇒ all derivatives need to be approximated at midpoints of cell

Tobias Neckel: Scientific Computing I

Spatial Discretisation – Diffusion Term

ui−1,j + ui,j−1 − 4ui,j + ui+1,j + ui,j+1

ui−1,j,k + ui,j−1,k + ui,j,k−1 − 6ui,j,k + ui+1,j,k + ui,j+1,k + ui,j,k+1

Spatial Discretisation – Convection Terms

• upwind differences (for momentum equation in x-direction):

• mix of central and upwind differences possible and used

Tobias Neckel: Scientific Computing I

• recall the incompressible Navier-Stokes equations:

Tobias Neckel: Scientific Computing I

Time Discretisation – Chorin Projection

A~u˙ h + D~uh + C(~uh )~uh − M T ph = fh , (3)

Time Discretisation – Chorin Projection

• updating the velocities:

• costs: solve 1 linear system of equations in each time step!

Tobias Neckel: Scientific Computing I

• geometry representation as a flag field (Marker-and-Cell)

flag field as an array of booleans:

• input data (boundary conditions) and output data (computed

Tobias Neckel: Scientific Computing I

Lab course “Scientific Computing – Computational Fluid Dynamics”:

Tobias Neckel: Scientific Computing I

Part III: The Shallow Water Equations and Finite

The Shallow Water Equations

Finite Volume Discretisation

Towards Tsunami Simulation

Tobias Neckel: Scientific Computing I

The Shallow Water Equations

derived from conservations laws for mass and momentum

Tobias Neckel: Scientific Computing I

Modelling Scenario: Tsunami Simulation

The Ocean as “Shallow Water”??

Tobias Neckel: Scientific Computing I

Modelling Scenario: Tsunami Simulation

The Ocean as “Shallow Water”??

Tobias Neckel: Scientific Computing I

Modelling Scenario: Tsunami Simulation

The Ocean as “Shallow Water”??

Tobias Neckel: Scientific Computing I

Modelling Scenario: Tsunami Simulation

(n) (n) (n) 1 (n) (n)

(n) (n) (n) 1 (n) (n)

(n) 1 (n) (n)

(n) 1 (n) (n)