Lecture 15:
Higher-Order Schemes for Steady
Convection
Last Time…
We looked more closely into the behavior of the
convection operator by looking at unsteady convection
z Consider the stability of implicit and explicit time-
differencing schemes for
» CDS
» UDS
» Lax-Wendroff scheme
z Considered the corresponding model equations and
try to explain the observed behavior
This Time…
z Start looking at higher-order schemes for the steady
convection operator
Higher-Order Schemes
z Neither UDS nor CDS are satisfactory
» Lot of research in devising schemes which are at
least second order accurate spatially
» Control of spatial oscillations
z Let’s first look at the basis for creating higher-order
schemes using the Taylor series
z Then, over the next few lectures, we will look at
approaches to controlling spatial oscillations
Taylor Series Basis
z For first-order UDS:
z We may consider this a truncation of a Taylor series:
z What if we truncated to higher order?
z Note that we are still using an upwinded expansion
Second-Order Schemes
z Taylor series about P
z Truncate Taylor series after second term:
Truncation error : O(∆x2 )
z Many ways to write gradient
» Gradient must be written to at least O(∆x)
» Can write gradient at P using forward, backward or
central difference
Basis of Fromm Scheme
z Central difference
Truncation error : O(∆x2 )
z Thus:
z Add and subtract φP/4
Basis of Fromm Scheme
(Actual Fromm scheme
has other terms – more on
this later)
Basis of Beam-Warming Scheme
z Second-order scheme:
z Write gradient as:
Truncation error : O(∆x)
z Combining:
Basis of Beam-Warming
scheme
Third-Order Schemes
z Truncate Taylor series to third-order:
⎛ ∂φ ⎞
z Need to write ⎜ ⎟
⎝ ∂x ⎠ P
to at least second order
⎛ ∂ 2φ ⎞
z Need to write ⎜ 2⎟ to at least first order
⎝ ∂x ⎠ P
Third-Order Scheme: QUICK
z Quadratic Upwind Interpolation for Convective Kinetics
(QUICK):
z Combining:
O(∆x3 ) accurate
QUICK (Cont’d)
z Rearranging
Central
difference Curvature term
(linear)
Curvature factor
C=1/8
Combined Representation
z The second- and third-order schemes we have seen
can be combined into a single expression:
z κ = -1 Beam Warming scheme
z κ=0 Fromm scheme
z κ = 1/2 QUICK
z κ = 1 Central difference scheme
Discussion
z If we used these schemes in steady convection
problems, we would get spatial oscillations
z If used in conjunction with explicit time stepping
schemes, all these schemes are unconditionally unstable
z Can counteract this in a variety of different ways:
» Use implicit schemes
» Multi-stage Runge-Kutta schemes
» Add extra terms from model equation to counteract
negative diffusion coefficient
Beam-Warming Scheme
z Start with
z Use face values based on:
Second-order
artificial
diffusion term
z Rearrange to obtain
Time Marching with Beam-Warming
Scheme
We see the usual dispersive behavior – smooth
profiles are convected relatively well, but square
profile picks up wiggles
Discussion
z First-order UDS is dissipative
z CDS is dispersive
» unconditionally unstable in conjunction with explicit
stepping
» Other symmetric schemes are also dispersive and
unconditionally unstable when used with explicit time
stepping schemes
z Higher-order upwind-weighted schemes must be
stabilized if used with explicit stepping schemes
» Also dispersive
z Must do something to control wiggles
Added Dissipation Schemes
z We saw that the artificial dissipation in the UDS
scheme stabilized it when used with explicit time
stepping
z Some researchers have added an explicit diffusion
term to mimic this effect.
z Want to keep truncation error O(∆x2) if using second-
order schemes.
z Add to original PDE an extra fourth-order diffusion
term:
3 ∂ φ
4 Note that artificial term is O(∆x3 ) –
(constant)∆x preserves O(∆x2 ) error
∂x 4
Added-Dissipation Schemes
z Corresponding face value for CDS is
z Near shocks and discontinuities, need to add more
dissipation. Usually a second-order dissipation term is
necessary:
∂ 2
φ
(constant)∆x 2 2
∂x
Destroys second-order
accuracy of scheme –
reduces to first-order near
shocks
Added-Dissipation Schemes
z Corresponding face value for CDS is:
From second-order
dissipation term – O(∆x) From fourth-order
dissipation term –
O(∆x2)
z How to detect shocks to turn on second-order
dissipation?
z How to choose ε(2) and ε(4) ?
Closure
In this lecture, we
z Started looking at higher-order spatial schemes
» Fromm scheme
» Beam-Warming
» QUICK
» Added-dissipation schemes
