Well known limiters and new ones
Friedemann Kemm BTU Cottbus
kemm@math.tu-cottbus.de
-1 -0.5 0 0.5 1 1.5 2 2.5 3
-4 -2 0 2 4 6
TVD-region Sweby-region linear 3rd order Lax-Wendroff Beam-Warming Fromm
0 0.2 0.4 0.6 0.8 1 1.2
-1 -0.5 0 0.5 1
Ultrabee β=2/3 Superbee
Lax-Wendroff can be written as
a correction of first order upwind.
Lax Wendroff
advection
qn+1
= qn
− ν + 1 2
ν(1 − ν)
1
rn
− 1
Δqn
−1/2
with
r = Δqp Δqdon
ν = distance covered in Δt Δ
Beam-Warming can be written as a correction of first order upwind.
Beam Warming
advection
qn+1
= qn
− ν + 1 2
ν(1 − ν)
rn
rn
− rn
−1
Δqn
−1/2
with
r = Δqp Δqdon
ν = distance covered in Δt Δ
Fromm scheme can be written as a correction of first order upwind.
Fromm
advection
qn+1
= qn
− ν + 1 2
ν(1 − ν)
(1 + rn
)/2 rn
− (1 + rn
−1)/2
Δqn
−1/2
with
r = Δqp Δqdon
ν = distance covered in Δt Δ
TVD-schemes can be written as
a correction of first order upwind.
General Case
advection
qn+1
= qn
− ν + 1 2
ν(1 − ν)
φ(rn
) rn
− φ(rn
−1)
Δqn
−1/2
with
r = Δqp Δqdon
ν = distance covered in Δt Δ
The TVD-region is much larger than the Sweby region.
-1 -0.5 0 0.5 1 1.5 2 2.5 3
-6 -4 -2 0 2 4 6
CFL=0.1 TVD
Sweby
-1 0 1 2 3 4 5
-15 -10 -5 0 5 10 15
CFL=0.5 TVD
Sweby
0 5 10 15 20
-20 -10 0 10 20 30
CFL=0.9 TVD
Sweby
− 2
1 − |ν|≤
φ(r)
r − φ(R) ≤ 2
|ν|
−2≤
φ(r)
r − φ(R) ≤ 2
Third order schemes are upwind biased.
-1 -0.5 0 0.5 1 1.5 2 2.5 3
-6 -4 -2 0 2 4 6
CFL=0.1 3rd order
LW BW Fromm
-1 0 1 2 3 4 5
-15 -10 -5 0 5 10 15
CFL=0.5 3rd order
LW BW Fromm
0 5 10 15 20
-20 -10 0 10 20 30
CFL=0.9 3rd order
LW BW Fromm
φ3(r) =
1 − 1 + |ν| 3
+ 1 + |ν| 3
r φLW(r) = 1 φBW(r) = r
Limiters might be constructed by sticking to third order as long as possible.
-1 -0.5 0 0.5 1 1.5 2 2.5 3
-6 -4 -2 0 2 4 6 CFL=0.1
θ=1.00 θ=0.75 θ=0.50 MC
-1 0 1 2 3 4 5
-15 -10 -5 0 5 10 15
CFL=0.5 θ=1.00
θ=0.75 θ=0.50 MC
0 5 10 15 20
-20 -10 0 10 20 30 CFL=0.9
θ=1.00 θ=0.75 θ=0.50 MC
φθ(r) = minnmxn−(1 − θ) 2
|ν|
, φ3(r)o,
mxn−(1 − θ) 2 1 − |ν|
r, θ 2
|ν|
ro, θ 2 1 − |ν|
o
We introduce an adaption of
Superbee type limiters to third order.
-1 -0.5 0 0.5 1 1.5 2 2.5 3
-6 -4 -2 0 2 4 6 CFL=0.1
Ultrabee β=2/3 Superbee
-1 0 1 2 3 4 5
-15 -10 -5 0 5 10 15
CFL=0.5 Ultrabee
β=2/3 Superbee
0 5 10 15 20
-20 -10 0 10 20 30 CFL=0.9
Ultrabee β=2/3 Superbee
φβ(r) = mxn0,minφUB(r), mx{1 + (φ0
3 − β/2)(r − 1),1 + (φ0
3 + β/2)(r − 1)} o
Superpower is closer to the linear third order scheme than Hyperbee.
-1 -0.5 0 0.5 1 1.5 2 2.5 3
-6 -4 -2 0 2 4 6
CFL=0.1 Hyperbee
Superpower van Leer
-1 0 1 2 3 4 5
-15 -10 -5 0 5 10 15
CFL=0.5 Hyperbee
Superpower van Leer
0 5 10 15 20
-20 -10 0 10 20 30
CFL=0.9 Hyperbee
Superpower van Leer
φsp(r) = mxn0,φ3(r)1 −
1 − |r| 1 + |r|
p(r)o
p(r) = ( 2
|ν| ·2(1 − φ0
3) , r ≤ 1
2
|1−ν| · 2φ0
3 , r ≥ 1
Third order does not guarantee for a good representation of the amplitude.
0 0.5 1
-1 -0.5 0 0.5 1
CFL=0.1
θ=0.50 θ=1.00 MC
0 0.5 1
-1 -0.5 0 0.5 1
CFL=0.9
θ=0.50 θ=1.00 MC
Standard example with 200 cells after 10 full rounds (t = 20)
Adapting Ultrabee to third order reduces squaring without sacrificing the amplitude.
0 0.5 1
-1 -0.5 0 0.5 1
CFL=0.1
Ultrabee β=2/3 Superbee
0 0.5 1
-1 -0.5 0 0.5 1
CFL=0.9
Ultrabee β=2/3 Superbee
Standard example with 200 cells after 10 full rounds (t = 20)
On coarse grids the error is mainly due to the amplitude.
0.1 1
100 200 400
CFL=0.1
Superbee Ultrabee β=2/3 Superpower θ=1
0.1 1
100 200 400
CFL=0.9
Superbee Ultrabee β=2/3 Superpower θ=1
L1-norm of the error
The squaring effect spoils the
convergence of Superbee type limiters.
0.01 0.1 1
100 1000 10000 100000
CFL=0.1
Superbee Ultrabee β=2/3 Superpower θ=1
0.01 0.1 1
100 1000 10000 100000
CFL=0.9
Superbee Ultrabee β=2/3 Superpower θ=1
L1-norm of the error
Smooth limiters are a
good choice for nonlinear waves.
3 3.5 4 4.5
0 0.5 1 1.5 2 2.5 3
reference solution mixed Superpower and Ultrabee
3 3.5 4 4.5
0 0.5 1 1.5 2 2.5 3
reference solution mixed β=2/3 and Ultrabee
Detail of Shu-Osher problem, t = 1.8, 400 cells
In summary, CFL-dependent limiters enhance the quality of TVD-schemes.
0 1 2 3 4 5 6 7
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
Superbee, primitive Minmod, characteristic Superbee, characteristic Ultrabee, characteristic
Detail of Toro’s test case 3 15