Scribd
Upload
33 views

13-Convection and Diffusion - B

The document discusses various numerical schemes for solving the convection-diffusion equation, including CDS, UDS, exponential, hybrid, and power-law schemes. It examines the behavior of CDS and UDS schemes using model equations. CDS is found to be dispersive while UDS is dissipative. The Lax-Wendroff scheme aims to counteract the instability of CDS by adding a diffusion term, but it introduces dispersion errors. Explicit UDS damps amplitudes but does not introduce phase errors.

Uploaded by

alagarg137691
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
33 views

13-Convection and Diffusion - B

The document discusses various numerical schemes for solving the convection-diffusion equation, including CDS, UDS, exponential, hybrid, and power-law schemes. It examines the behavior of CDS and UDS schemes using model equations. CDS is found to be dispersive while UDS is dissipative. The Lax-Wendroff scheme aims to counteract the instability of CDS by adding a diffusion term, but it introduces dispersion errors. Explicit UDS damps amplitudes but does not introduce phase errors.

Uploaded by

alagarg137691
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 35

Lecture 13:

Convection and Diffusion (Cont’d)


Last Time …

We
z Looked at CDS/UDS schemes to unstructured meshes
z Look at accuracy of CDS and UDS schemes
z Look at false diffusion in UDS using model equation
This Time…

z We will use model equation to look at behavior of CDS


scheme
z Look at some first-order schemes based on exact
solutions to the convection-diffusion equation
» Exponential scheme
» Hybrid scheme
» Power-law scheme
z Look at unsteady convection to understand the
meaning of dissipation and dispersion
CDS Model Equations

z Pure convection equation:


z Apply CDS:

z Expand in Taylor series

Do same type of
expansion in y
direction
Model Equation (Cont’d)

z Subtract to obtain:

z Do same in y direction:

z Substitute into discrete equation

Dispersion Term
Discussion

z Model equation for CDS has extra third-derivative


(dispersive) term
z This type of odd-derivative term tends to cause spatial
wiggles
z Note that truncation error for CDS is O( ∆x2 )
z Thus, UDS is dissipative and CDS is dispersive
First-Order Schemes Based on Exact
Solutions

z 1D Convection-diffusion equation

-Pe
φ

Pe=0
Pe

What are the limits of this


equation for different Pe?
Exponential Scheme

z Use 1-D exact solution as profile assumption in doing


discretization
z Consider convection-diffusion equation:

z Integrate over control volume:


Exponential Scheme (Cont’d)

z Area vectors

z Flux*Area:

z Use exact solution to write convection and diffusion


terms
Exponential Scheme: Discrete Equations

z Both convection and


diffusion terms estimated
from exact solution
z If S=0, we would get the
exact solution in 1D
problems
z But obviously not exact for
non-zero S, multi-
dimensional problems…
z Discretization has
boundedness, diagonal
dominance
z Only first-order accurate
Hybrid Difference Scheme

z Consider the aE coefficient in exponential scheme

z Limits with respect to Pe:


Approximations to Exponential Scheme

z Exponentials are expensive to compute


z Approximations to the exponential profile assumption
have been used to offset the cost.
» Hybrid difference scheme
» Power-law scheme
z Both these approximations are also only first-order
accurate
Hybrid Difference Scheme (Cont’d)

Instead of using the exact curve for aE/De, use three


tangents

Similar manipulation for other coefficients


Hybrid Difference Scheme (Cont’d)

z Guaranteed bounded
solutions
z Satisfies Scarborough
criterion
z O(∆x) accurate
Power-Law Scheme

z Employs fifth-order polynomial approximation to

z Similar approach to other coefficients


z Scheme is bounded and satisfies the Scarborough
criterion
z Is O(∆x) accurate
Multi-Dimensional Schemes

z Exact solutions have been used as profile assumptions in multi-


dimensional situations
z Control volume-based finite element method of Baliga and
Patankar (1983)

X
φ = A exp( ρUX / Γ) + BY + C U

z This form is the solution to


the 2D convection-diffusion equation
Multi-Dimensional Schemes

z Finite analytic scheme (Chen and Li, 1979)


z Write 2D convection diffusion equation with (i,j+1)
source term for “element”:

(i-1,j) (i,j) (i+1,j)

z Fix coefficient using (i,j) values


(i, j-1)
z Find analytical solution using separation of
variables
z Use exact solution for profiles assumptions
Unsteady Convection Equation

z For simplicity, consider linear wave equation


Convection-diffusion
equation with Γ=0,
ρ=1, u=constant
z Consider 1-D domain of length L=1.
z Initial condition:
z Solution:
Traveling
wave

z How well do our numerical schemes capture this?


Wave Forms

Sine Wave Square Wave

For u>0, profiles shift to the right by (ut) in


time t. For u=1 and t=0.25, the shift is 0.25
units
Explicit CDS

z Consider 1D uniform mesh


z Integrate over control volume:

Convection terms
at old time

z Can be shown to be O(∆x2, ∆t) accurate


z Unconditionally unstable – not usable
Implicit CDS

z Integrate over control volume

Convective terms
at new time

z Can show to be O(∆x2, ∆t) accurate


z Unconditionally stable
z Boundedness not guaranteed because of mix of
neighbor signs
Explicit UDS

z Integrate over control volume

CFL condition
z Stable for

Sometimes called the


z Courant number ν = u∆t/∆x CFL number after
Courant, Friedrichs and
Lewy
Explicit UDS

z Discrete equation can be re-written as:

z Boundedness guaranteed for ν <1


Explicit UDS (Cont’d)

Start with sine or square wave at t=0. Choose mesh of 50 cells


and time step commensurate with ν =0.5. Take 25 time steps
and plot result.
Discussion

z Computed solution shifts to right as expected


z Profile is smeared
z But no overshoots and undershoots
» “Monotonic” solution
» Solution is bounded by initial conditions for Courant
number <1
Explicit UDS: Error Analysis

z Derive model equation for explicit UDS:

Extra artificial diffusion term


tied to Courant number

z Thus explicit UDS is dissipative


z Note that leading order error terms are O(∆x,∆t)
Explicit and Implicit CDS: Error Analysis

z Model equation for explicit CDS:


Note negative diffusion
coefficient – responsible
for instability

z Model equation for implicit CDS:

Again, false diffusion term


makes time-stepping
diffusive
Lax-Wendroff Scheme

z This is a classic scheme for the linear wave equation


z Recall explicit CDS model equation

z Try to make explicit CDS scheme stable by


counteracting its negative diffusion coefficient
z Lax-Wendroff starts with:
Lax-Wendroff Scheme

z Discretize using explicit CDS

∂ 2φ
Central difference operator for
∂x 2

z Can show that scheme is stable for ν ≤ 1


Wave Advection

Smooth portions of profile are resolved fine, but wiggles near


discontinuities – out of bounds of solution at old time.
Lax-Wendroff Model Equation

z Model equation:

Note dispersion term! No diffusion


term

z In wave transport, dispersion is responsible for altering the


frequency content of a signal.
z Can think of profile as consisting of a Fourier series.
z Numerical scheme transports different frequencies at different
speeds, changing the signal shape – “phase error”
Discussion

z Dispersive schemes work well for wave forms which


have a smooth shape, i.e., not too many frequency
components, eg. sine wave
z Discontinuities imply that there are infinite number of
frequencies
» Phase error causes frequency content to be altered
at any given time
» Square wave is therefore distorted badly
z Explicit UDS, on the other hand, damps amplitude, but
does not introduce phase error, hence “dissipative”
Lax Wendroff Scheme: More Oddities

z Consider discrete equation:

z Re-arrange:

Convective+ (false) diffusive fluxes


on e and w faces at old time
z At steady state:
Notice
anything
odd?
Oddities (Cont’d)

z Discrete equation at steady state

Steady state solution depends on ∆t!


We would get different steady state
solutions depending on our history of
time-stepping
z Note that all our other schemes did not have this type
of undesirable dependence !
Closure

In this lecture, we
z Looked at the model equation for CDS
» Shown dispersive nature of model equation
z Looked at
» differencing schemes based on exact solution to 1D
convection-diffusion equation
» multidimensional schemes based on exact solutions
z Looked at unsteady convection using UDS, CDS and Lax-
Wendroff schemes to better understand the meaning of
dispersion and dissipation

You might also like