large-time behavior and numerics for …...large-time behavior and numerics for scalar conservation...
TRANSCRIPT
LARGE-TIME BEHAVIOR AND NUMERICSFOR SCALAR CONSERVATION LAWS
Alejandro PozoJoint work with L. I. Ignat (IMAR) and E. Zuazua (BCAM&Ikerbasque)
BCAM – Basque Center for Applied Mathematics
The Second BCAM Workshop on Computational MathematicsOctober 18, 2013
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Outline
1 Motivation
2 Inviscid vs. Viscous
3 Numerics for 1-D scalar conservation laws
4 Similarity variables
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Motivation
Source: J.J.Alonso-M.R.Colonno (2012) Source: National Oceanic and Atmospheric Administration
Source: European Space Agency
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Burgers equation
Given u0 ∈ L1(R), consider the 1-D conservation law with or without viscosity:ut +
(u2
2
)x
= εuxx , x ∈ R, t > 0,
u(x , 0) = u0(x), x ∈ R.
Then:If ε = 0, u(·, t) ∼ Np,q(·, t) as t →∞.If ε > 0, u(·, t) ∼ uM (·, t) as t →∞.
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
N-wave vs. diffusive wave
N-wave: Np,q is the so-called N-wave, defined as:
Np,q(x , t) :=
{xt , if √pt < x < √qt,0 otherwise,
where
p := −2 miny∈R
∫ y
−∞u0(x)dx , q := 2 max
y∈R
∫ ∞y
u0(x)dx .
Diffusive wave: uM is the solution of the Burgers equation with initial data Mδ0:
uM (x , t) :=
∫R
( x−yt
)e−
FM (x,y,t)
2ε dy∫R e−
−FM (x,y,t)
2ε dy,
where
FM (x , y , t) := M∫ y
−∞δ0(r)dr +
(x − y)2
2t, M :=
∫R
u0(x)dx .
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Intermediate asymptotics
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Conservative schemes
Let us consider now numerical approximation schemes (ε = 0):un+1j = un
j −∆t∆x(
gnj+1/2 − gn
j−1/2), j ∈ Z, n > 0,
u0j = 1
∆x
∫ xj+1/2xj−1/2
u0(x)dx , j ∈ Z.
The approximated solution u∆ is given by
u∆(t, x) = unj , xj−1/2 < x < xj+1/2, tn ≤ t < tn+1,
where tn = n∆t and xj+1/2 = (j + 1/2)∆x .
Which is the behavior of u∆?Is the N-wave preserved?
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Conservative schemes
Let us consider now numerical approximation schemes (ε = 0):un+1j = un
j −∆t∆x(
gnj+1/2 − gn
j−1/2), j ∈ Z, n > 0,
u0j = 1
∆x
∫ xj+1/2xj−1/2
u0(x)dx , j ∈ Z.
The approximated solution u∆ is given by
u∆(t, x) = unj , xj−1/2 < x < xj+1/2, tn ≤ t < tn+1,
where tn = n∆t and xj+1/2 = (j + 1/2)∆x .
Which is the behavior of u∆?Is the N-wave preserved?
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Examples
Three-point conservative schemes for the Burgers equation:1 Lax-Friedrichs
gLF (u, v) =u2 + v2
4−
∆x∆t
( v − u2
),
2 Engquist-Osher
gEO(u, v) =u(u + |u|)
4+
v(v − |v |)4
,
3 Godunov
gG (u, v) =
minw∈[u,v ]
w2
2 , if u ≤ v ,
maxw∈[v,u]
w2
2 , if v ≤ u.
We denote gnj+1/2 = g(un
j , unj+1).
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Numerical viscosity
We can rewrite three-point monotone schemes in the form
un+1j − un
j
∆t+
(unj+1)2 − (un
j−1)2
4∆x= R(un
j , unj+1)− R(un
j−1, unj ).
R is the numerical viscosity:
R(u, v) =1
2∆x
(u2
2+
v2
2− 2g(u, v)
).
For instance:1 Lax-Friedrichs
RLF (u, v) =v − u
2,
2 Engquist-OsherREO(u, v) =
14∆x
(v |v | − u|u|),
3 Godunov
RG (u, v) =
{ 14∆x sign(|u| − |v |)(v2 − u2), v ≤ 0 ≤ u,
14∆x (v |v | − u|u|), elsewhere.
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Numerical viscosity
We can rewrite three-point monotone schemes in the form
un+1j − un
j
∆t︸ ︷︷ ︸ut
+(un
j+1)2 − (unj−1)2
4∆x︸ ︷︷ ︸u2/2
= R(unj , un
j+1)− R(unj−1, un
j ).
R is the numerical viscosity:
R(u, v) =1
2∆x
(u2
2+
v2
2− 2g(u, v)
).
For instance:1 Lax-Friedrichs
RLF (u, v) =v − u
2,
2 Engquist-OsherREO(u, v) =
14∆x
(v |v | − u|u|),3 Godunov
RG (u, v) =
{ 14∆x sign(|u| − |v |)(v2 − u2), v ≤ 0 ≤ u,
14∆x (v |v | − u|u|), elsewhere.
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Numerical viscosity
We can rewrite three-point monotone schemes in the formun+1
j − unj
∆t+
(unj+1)2 − (un
j−1)2
4∆x= R(un
j , unj+1)− R(un
j−1, unj )︸ ︷︷ ︸
?.
R is the numerical viscosity:
R(u, v) =1
2∆x
(u2
2+
v2
2− 2g(u, v)
).
For instance:1 Lax-Friedrichs
RLF (u, v) =v − u
2,
2 Engquist-OsherREO(u, v) =
14∆x
(v |v | − u|u|),3 Godunov
RG (u, v) =
{ 14∆x sign(|u| − |v |)(v2 − u2), v ≤ 0 ≤ u,
14∆x (v |v | − u|u|), elsewhere.
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Numerical viscosity
We can rewrite three-point monotone schemes in the form
un+1j − un
j
∆t+
(unj+1)2 − (un
j−1)2
4∆x= R(un
j , unj+1)− R(un
j−1, unj ).
R is the numerical viscosity:
R(u, v) =1
2∆x
(u2
2+
v2
2− 2g(u, v)
).
For instance:1 Lax-Friedrichs
RLF (u, v) =v − u
2,
2 Engquist-OsherREO(u, v) =
14∆x
(v |v | − u|u|),
3 Godunov
RG (u, v) =
{ 14∆x sign(|u| − |v |)(v2 − u2), v ≤ 0 ≤ u,
14∆x (v |v | − u|u|), elsewhere.
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Properties
The three schemes ensure:
Monotonicity:
unj ≥ vn
j , ∀j ∈ Z =⇒ un+1j ≥ vn+1
j , ∀j ∈ Z.
Conservation of massEntropyConsistency with OSLCL1 → L∞ decay with a rate O(t−1/2)
BV estimates
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Properties
The three schemes ensure:
MonotonicityConservation of mass: ∑
j∈Z
unj =∑j∈Z
u0j , ∀n ∈ N.
EntropyConsistency with OSLCL1 → L∞ decay with a rate O(t−1/2)
BV estimates
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Properties
The three schemes ensure:
MonotonicityConservation of massEntropy
Consistency with OSLCL1 → L∞ decay with a rate O(t−1/2)
BV estimates
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Properties
The three schemes ensure:
MonotonicityConservation of massEntropyConsistency with OSLC:
unj−1 − un
j+1
2∆x≤
2n∆t
, ∀j ∈ Z.
L1 → L∞ decay with a rate O(t−1/2)
BV estimates
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Properties
The three schemes ensure:
MonotonicityConservation of massEntropyConsistency with OSLCL1 → L∞ decay with a rate O(t−1/2):
‖un‖∞,∆ := maxj∈Z|un
j | ≤ C(n∆t)−1/2‖u0‖1.
BV estimates
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Properties
The three schemes ensure:
MonotonicityConservation of massEntropyConsistency with OSLCL1 → L∞ decay with a rate O(t−1/2)
BV estimates: ∑j∈Z
|unj+k − un
j | =∑j∈Z
|u0j+k − u0
j |, ∀n, k ∈ N.
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Properties
The three schemes ensure:
MonotonicityConservation of massEntropyConsistency with OSLCL1 → L∞ decay with a rate O(t−1/2)
BV estimates
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Conservation of positive and negative masses
Theorem (L. I. Ignat, A.P, E. Zuazua ’13)
Assume that u0 ∈ L1(R) and that the CFL condition λ‖un‖∞,∆ ≤ 1 is fulfilled (whereλ = ∆t/∆x). If the numerical flux of a 3-point monotone conservative schemesatisfies
g(η, ξ) = 0, when − 1/λ ≤ η ≤ 0 ≤ ξ ≤ 1/λ,
andξ − λg(ξ,−ξ) ≥ 0, when 0 ≤ ξ ≤ 1/λ,
then, for any n ∈ N the following holds:
mink∈Z
k∑j=−∞
unj = min
k∈Z
k∑j=−∞
u0j and max
k∈Z
∞∑j=k
unj = max
k∈Z
∞∑j=k
u0j .
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Examples
Engquist-Osher and Godunov schemes satisfy the theorem:
gEO(η, ξ) = gG (η, ξ) = 0, ∀η, ξ s.t. − 1/λ ≤ η ≤ 0 ≤ ξ ≤ 1/λ
andξ − λgEO(ξ,−ξ) = ξ − λξ2 = ξ(1− λξ) ≥ 0, ∀ξ s.t. 0 ≤ ξ ≤ 1/λ
ξ − λgG (ξ,−ξ) = ξ − λξ2
2= ξ(1− λ
ξ
2) ≥ 0, ∀ξ s.t. 0 ≤ ξ ≤ 1/λ
BUT Lax-Friedrichs scheme does not!
gLF (η, ξ) = 0 ⇐⇒ ξ = η = 0, ∀η, ξ s.t. − 1/λ ≤ η ≤ 0 ≤ ξ ≤ 1/λ,
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Examples
Engquist-Osher and Godunov schemes satisfy the theorem:
gEO(η, ξ) = gG (η, ξ) = 0, ∀η, ξ s.t. − 1/λ ≤ η ≤ 0 ≤ ξ ≤ 1/λ
andξ − λgEO(ξ,−ξ) = ξ − λξ2 = ξ(1− λξ) ≥ 0, ∀ξ s.t. 0 ≤ ξ ≤ 1/λ
ξ − λgG (ξ,−ξ) = ξ − λξ2
2= ξ(1− λ
ξ
2) ≥ 0, ∀ξ s.t. 0 ≤ ξ ≤ 1/λ
BUT Lax-Friedrichs scheme does not!
gLF (η, ξ) = 0 ⇐⇒ ξ = η = 0, ∀η, ξ s.t. − 1/λ ≤ η ≤ 0 ≤ ξ ≤ 1/λ,
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Main result
Theorem (L. I. Ignat, A.P, E. Zuazua ’13)
Consider u0 ∈ L1(R) and ∆x and ∆t such that λ‖un‖∞,∆ ≤ 1, λ = ∆t/∆x. Then,for any p ∈ [1,∞), the numerical solution u∆ given by the Lax-Friedrichs schemesatisfies
limt→∞
t12 (1− 1
p )‖u∆(t)− w(t)‖Lp (R) = 0,
where the profile w = wM∆is the unique solution ofwt +
(w2
2
)x
=(∆x)2
2 wxx , x ∈ R, t > 0,
w(0) = M∆δ0,
with M∆ =∫R u0
∆.
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Main result
Theorem (L. I. Ignat, A.P, E. Zuazua ’13)
Consider u0 ∈ L1(R) and ∆x and ∆t such that λ‖un‖∞,∆ ≤ 1, λ = ∆t/∆x. Then,for any p ∈ [1,∞), the numerical solutions u∆ given by Engquist-Osher and Godunovschemes satisfy
limt→∞
t12 (1− 1
p )‖u∆(t)− w(t)‖Lp (R) = 0,
where the profile w = wp∆,q∆ is the unique solution ofwt +
(w2
2
)x
= 0, x ∈ R, t > 0,
w(0) = M∆δ0, limt→0
∫ x
0w(t, z)dz =
0, x < 0,−p∆, x = 0,q∆ − p∆, x > 0,
with M∆ =∫R u0
∆ and
p∆ = −minx∈R
∫ x
−∞u0
∆(z)dz and q∆ = maxx∈R
∫ ∞x
u0∆(z)dz.
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Example
Let us consider the initial data
u0(x) =
−0.05, x ∈ [−1, 0],
0.15, x ∈ [0, 2],
0, elsewhere.
The parameters that describe the asymptotic N-wave profile are:
M = 0.25 , p = 0.05 and q = 0.3.
We take ∆x = 0.1 as the mesh size for the interval [−350, 800] and ∆t = 0.5.
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Comparison
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Similarity variables
Consider the change of variables:
s = ln(t + 1), ξ = x/√
t + 1, w(ξ, s) =√
t + 1 u(x , t).
This transforms Burgers equation into
ws +
(12
w2 −12ξw)ξ
= εwξξ, ξ ∈ R, s > 0.
The asymptotic profile is a N-wave too:
Np,q(ξ) =
{ξ, −
√2p < ξ <
√2q,
0, elsewhere,
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Similarity variables
Consider the change of variables:
s = ln(t + 1), ξ = x/√
t + 1, w(ξ, s) =√
t + 1 u(x , t).
This transforms Burgers equation into
ws +
(12
w2 −12ξw)ξ
= εwξξ, ξ ∈ R, s > 0.
The asymptotic profile is a N-wave too:
Np,q(ξ) =
{ξ, −
√2p < ξ <
√2q,
0, elsewhere,
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Examples
Engquist-Osher scheme (with ∆ξ = 0.01 and ∆s = 0.0005)
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Examples
Lax-Friedrichs scheme (with ∆ξ = 0.01 and ∆s = 0.0005)
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Physical vs. Similarity variables
Comparison of numerical and exact solutions at t = 1000. We choose ∆ξ such thatthe ‖ · ‖1,∆ error is similar. The time-steps are ∆t = ∆x/2 and ∆s = ∆ξ/20,respectively, enough to satisfy the CFL condition.For ∆x = 0.1:
Nodes Time-steps ‖ · ‖1,∆ ‖ · ‖2,∆ ‖ · ‖∞,∆Physical 1501 19987 0.0867 0.0482 0.0893Similarity 215 4225 0.0897 0.0332 0.0367
For ∆x = 0.01:
Nodes Time-steps ‖ · ‖1,∆ ‖ · ‖2,∆ ‖ · ‖∞,∆Physical 15001 199867 0.0093 0.0118 0.0816Similarity 2000 39459 0.0094 0.0106 0.0233
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Conclusions
Be careful!Numerical viscosity is necessary.BUT too much viscosity can produce undesired large-time dynamics.
To-Do list:Develop numerical schemes preserving large-time asymptotics for nonlinear PDEs.Study the use of self-similarity variables to improve computational cost.Analyze influence on optimal control problems.
Thank you very much! Eskerrik asko!
Acknowledgments:
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Conclusions
Be careful!Numerical viscosity is necessary.BUT too much viscosity can produce undesired large-time dynamics.
To-Do list:Develop numerical schemes preserving large-time asymptotics for nonlinear PDEs.Study the use of self-similarity variables to improve computational cost.Analyze influence on optimal control problems.
Thank you very much! Eskerrik asko!
Acknowledgments:
Alejandro Pozo Large-time behavior and numerics
MotivationInviscid vs. Viscous
Numerics for 1-D scalar conservation lawsSimilarity variables
Conclusions
Be careful!Numerical viscosity is necessary.BUT too much viscosity can produce undesired large-time dynamics.
To-Do list:Develop numerical schemes preserving large-time asymptotics for nonlinear PDEs.Study the use of self-similarity variables to improve computational cost.Analyze influence on optimal control problems.
Thank you very much! Eskerrik asko!
Acknowledgments:
Alejandro Pozo Large-time behavior and numerics