Some Results by Energy Methods on Large-Time Behavior of Viscous Gas
Akitaka Matsumura (Osaka University)
”Nonlinear Evolution Equations and Related Topics”
Kyoto Universiy, March 9-11, 2015
Introduction
A model system of viscous gas
ρ
t+ ∇ · (ρu) = 0, u
t+ u · ∇ u + 1
ρ ∇ p = µ
ρ ∆u + µ + λ
ρ ∇ ( ∇ · u), p = p(ρ) = aρ
γ,
(1)
where t ≥ 0, x ∈ Rn(n = 1, 2, 3), µ > 0, µ + λ > 0, a > 0, γ ≥ 1.
Consider the Cauchy problem for (1) with the initial data
(ρ, u)(0) = (ρ
0, u
0). (2)
Two papers :
• Nishida-M (1980) , J. Math. Kyoto Univ.,
asymptotic stability of constant states in R3, R2
• Nishihara-M (1985) , Japan J. Appl. Math., asymptotic stability of traveling waves in R1
Topic 1.
(joint work with T. Maeda)• Nishida-M (1980), x ∈ R2 or R3, ρ >¯ 0
(ρ0 − ρ, u¯ 0) ∈ H3, small =⇒ ( ¯ρ, 0) is asymptotically stable.
Since then, Hoff ( ¯ρ > 0), Feireisl, Lions ( ¯ρ = 0), . . . , Z.Xin-J.Li- X.Huang ( ¯ρ ≥ 0), . . .
Our present aim
Consider the asymptotic stability of an unbounded state ρ = P = ρ¯
1 + t, u = U = ( x1
1 + t, 0).
Theorem 1.
Supposex ∈ R
2, p = aρ
, anda − (2µ + λ)
¯
ρ > 0
. Then, there exists a ε0 > 0 such that if ∥ρ0−ρ, u¯ 0−(x1,0)∥H2 ≤ ε0, the Cauchy problem (1),(2) has a unique global solution in time (ρ, u), satisfying (ρ−P, u − U) ∈ C([0, +∞); H2) andsup
x∈R2
| ρ(t, x) − ρ ¯
1 + t | ≤ C (1 + t)
−32, sup
x∈R2
| u(t, x) − ( x
11 + t , 0) | ≤ C (1 + t)
−12.
Remarks
• The proof is given by a combination of changing the variable x1 along a characteristic curve and using a time-weighted energy method .
• For R3 , the similar results hold in H3 with the asymptotics sup
x∈R3
| ρ − P | ≤ C(1 + t)−2, sup
x∈R3
| u − U | ≤ C(1 + t)−1.
•
Open problems
– Isentropic case : p = aργ, (γ > 1)
– Full system case : p = Rρθ, e = γR−1θ ρ = P = ρ¯
1 + t, u = U = ( x1
1 + t, 0), θ = 2µ + λ Rρ¯ .
Topic 2.
(joint work with Yang Wang)Asymptotic stability of traveling wave solutions in R System in Lagrange coordinates :
v
t− u
x= 0,
u
t+ p
x= (µ u
xv )
x, p = p(v ) = av
−γ.
Traveling wave solution (viscous shock wave):
(v, u) = (V, U)(x − st), (V, U)(±∞) = (v±, u±) It exists under the Rankine-Hugoniot and entropy conditions.
Known results (µ : a positive constant)
• Nishihara-M (1985)
∃C(v−, γ) > 0 with C → ∞ as γ → 1 such that if
| v
+− v
−| ≤ C (v
−, γ ),
then (V, U)(x − st) is asymptotically stable for small initial pertur- bations with integral zero, that is,
∫
(v0 − V )(x) dx =
∫
(u0 − U)(x)dx = 0.
– For γ = 1, any large viscous shock wave is OK!.
– For γ > 1, a restriction on the amplitude is imposed.
• Mascia-Zumbrun(2004), Liu-Zeng(2009)
|v+ − v−| : suitably small =⇒ asymptotic stability
for general initial perturbations whose integrals are not necessarily zero.
• Barker-Humpherys-Laffite-Rudd-Zumbrun (2008), Humpherys-Laffite-Zumbrun (2010)
|v+ − v−| : suitably large =⇒ asymptotic stability
They also carried out numerical studies which indicate the asymptotic stability for intermediate amplitude as well.
Our present aim
Consider the case µ = µ(v) > 0.In the Chapman-Enskog expansion theory in rarefied gas dynamics (cf.
Chapman-Cowling (1970)), the viscosity coefficient is given by a func- tion of the absolute temperature θ.
Typical two examples :
{ µ = ¯µ θ12, Hard sphere Model, µ = ¯µ θ
1
2+ 2
(s−1), Cut-off inverse power force Model, where s (≥ 5) and ¯µ (> 0) are some constants.
The above two models are unified as
µ = ¯ µ θ
β(β ≥ 1
2 ).
Since our model is isentropic,
p = R θ
v = a v
−γ, (R : gas constant)
which implies
θ = a
R v
−(γ−1).
Hence,
µ = µ
0v
−(γ−1)β(β ≥ 1
2 , µ
0= ¯ µ( a
R )
β).
Thus, we assume
(A) µ = µ(v ) = µ
0v
−α(α ≥ 1
2 (γ − 1), µ
0> 0).
Cauchy problem :
v
t− u
x= 0,
u
t+ p
x= µ
0( u
xv
α+1)
x, (α ≥
12(γ − 1)) p = p(v) = av
−γ(3)
with the initial and far field conditions
(v, u)(0, x) = (v
0, u
0)(x),
x→±∞
lim (v, u)(t, x) = (v
±, u
±). (4)
Assumptions on initial data
(v0 − V, u0 − U) ∈ H1 ∩ L1,
xinf∈Rv0(x) > 0,
∫
(v0 − V )(x) dx =
∫
(u0 − U)(x) dx = 0.
Setting
ϕ0(x) =
∫ x
−∞
(v0 − V )(y)dy, ψ0(x) =
∫ x
−∞
(u0 − U)(y) dy, we further assume
(ϕ0, ψ0) ∈ L2. (⇒ (ϕ0, ψ0) ∈ H2)
Theorem 2.
Suppose α ≥ 12(γ − 1). Then, there exists a ε0 > 0 such that if ∥ϕ0, ψ0∥2 ≤ ε0, the Cauchy problem (3),(4) has a unique global solution in time (v, u), satisfying (v − V, u − U) ∈ C([0, ∞); H1) andsup
x∈R
| (v, u)(x, t) − (V, U )(x − st) | → 0 (t → ∞ ).
Remarks
• In the proof, the essetial a priori estimate is given by a technical weighted energy method, “double step weighted energy method”, developed by Mei-M (1997) and Hashimoto-M (2007).
•
Open problem
Full system case :
µ = ¯µ θβ, λ = ¯λ θβ, κ = ¯κ θβ (β ≥ 1 2).
Sketch of the proof of Theorem 1
Write x = (x, y) ∈ R2 and assume ¯ρ = 1, µ + λ = 0 for simplicity.
Cauchy problem :
ρ
t+ (ρu
1)
x+ (ρu
2)
y= 0, u
1t+ (u
1u
1x+ u
2u
1y) + 1
ρ p
x− µ
ρ (u
1xx+ u
1yy) = 0, u
2t+ (u
1u
2x+ u
2u
2y) + 1
ρ p
y− µ
ρ (u
2xx+ u
2yy) = 0, p = aρ
with the initial data
(ρ, u1, u2)(0, x, y) = (ρ0, u1.0, u2.0)(x, y), (x, y) ∈ R2.
Change the unknown variables : (ρ, u1, u2) → (η, ϕ, ψ) ρ = (1 + η)
1 + t , u1 = x
1 + t + ϕ, u2 = ψ.
System for (η, ϕ, ψ) :
ηt + x
1 + tηx + ((1 + η)ϕ)x + ((1 + η)ψ)y = 0, ϕt + x
1 + tϕx + 1
1 + tϕ + ϕϕx + ψϕy + a
1 + ηηx − µ(1 + t)
1 + η (ϕxx + ϕyy) = 0, ψt + x
1 + tψx + ϕψx + ψψy + a
1 + ηηy − µ(1 + t)
1 + η (ψxx + ψyy) = 0, (η, ϕ, ψ)(0, x, y) = (η0, ϕ0, ψ0)(x, y).
Characteristic curve w.r.t. x
dx(t)
dt = x(t) 1 + t, x(0) = ˜x
=⇒ x = x(t) = (1 + t)˜x.
Change of variable x : x = (1 + t)˜x
∂
∂x =⇒ 1 1 + t
∂
∂x˜, ∂
∂t + x 1 + t
∂
∂x =⇒ ∂
∂t
˜
ηt + ((1 + ˜η) ˜ϕ)x˜
1 + t + ((1 + ˜η) ˜ψ)y = 0, ϕ˜t +
ϕ˜
1 + t +
ϕ˜ϕ˜x˜
1 + t + ˜ψϕ˜y + a 1 + t
˜ ηx˜
(1 + ˜η) − µ 1 + ˜η
( ϕ˜x˜x˜
1 + t + (1 + t)ϕ˜yy )
= 0, ψ˜t +
ϕ˜ψ˜x˜
1 + t + ˜ψψ˜y + a
1 + ˜ηη˜y − µ 1 + ˜η
( ψ˜x˜x˜
1 + t + (1 + t)ψ˜yy )
= 0.
Reformulated problem :
η
t+ 1
1 + t ϕ
x+ ψ
y= N
0, ϕ
t+ 1
1 + t ϕ + a
1 + t η
x− µ
( 1
1 + t ϕ
xx+ (1 + t)ϕ
yy)
= N
1, (5) ψ
t+ aη
y− µ
( 1
1 + t ψ
xx+ (1 + t)ψ
yy)
= N
2, (η, ϕ, ψ )(0) = (η
0, ϕ
0, ψ
0) ∈ H
2.
We look for the global solution in time of (5) such that (η, ϕ, ψ) ∈ C([0, ∞); H2), sup
(x,y)∈R2
|(η, ϕ, ψ)(t, x, y)| ≤ C(1 + t)−12.
E(t) := ∥(η, ϕ, ψ)(t)∥22+(1 + t)2∥(η, ϕ, ψ)y(t)∥21+(1 + t)4∥(η, ϕ, ψ)yy(t)∥2. Proposition 3 (a priori estimate). Suppose a − µ > 0. Then there exist positive constants ε0 and C0 such that if (η, ϕ, ψ) ∈ C([0, T]; H2) is the solution of the Cauchy problem (5) for some T > 0 and sup
t∈[0,T]
E(t) ≤ ε0 it holds that
∥(η, ϕ, ψ)(t)∥22 + (1 + t)2∥(η, ϕ, ψ)y(t)∥21 + (1 + t)4∥(η, ϕ, ψ)yy(t)∥2 +
∫ t
0
( 1
1 + τ ∥ϕ(τ)∥2 + 1
1 + τ ∥(ϕ, ψ)x(τ)∥22 + (1 + τ)∥(ϕ, ψ)y(τ)∥22 )
dτ +
∫ t
0
((1 + τ)3∥(ϕ, ψ)yy(τ)∥21 + (1 + τ)5∥(ϕ, ψ)yyy(τ)∥2)) dτ +
∫ t
0
( 1
1 + τ ∥ηx(τ)∥21 + (1 + τ)∥ηy(t)∥21 + (1 + τ)3∥ηyy(τ)∥2 )
dτ
≤ C0∥(η0, ϕ0, ψ0)∥22, t ∈ [0, T].
Decay estimates sup
(x,y)∈R2
|η(t, x, y)|2 ≤
∫∫
|(2ηηx)y| dxdy
≤ C(∥ηx∥∥ηy∥ + ∥η∥∥ηxy∥)
≤ C(1 + t)−1 which implies
sup
(x,y)∈R2
| ρ(t, x, y) − 1
1 + t | ≤ C(1 + t)−32,
and similarly
sup
(x,y)∈R2
| u(t, x, y) − ( x
1 + t,0) | ≤ C(1 + t)−12.
Basic energy estimate
Estimate (5) · (aη, ϕ, ψ) : 1
2 d dt
∫∫
(aη
2+ ϕ
2+ ψ
2)dxdy +
∫∫ ( 1
1 + t ϕ
2+ µ
1 + t (ϕ
2x+ ψ
x2) + µ(1 + t)(ϕ
2y+ ψ
y2) )
dxdy
=
∫∫
(aηN
0+ ϕN
1+ ψN
2) dxdy.
(6)
Estimate (5)
2· η
x+ (5)
3· (1 + t)η
y: d
dt
∫∫ (
ϕη
x+ (1 + t)ψη
y+ 1
2a ψ
2+ µ 2
(
η
x2+ (1 + t)
2η
y2))
dxdy +
∫∫ ( a
1 + t η
x2+ (a − µ)(1 + t)η
y2)
dxdy
−
∫∫ ( 1
1 + t ϕ
2x+ 2ϕ
xψ
y+ (1 + t)ψ
y2)
dxdy +
∫∫ ( µ
a(1 + t) ψ
x2+ µ
a (1 + t)ψ
y2+ 1
1 + t ϕη
x)
dxdy (7)
=
∫∫ (
η
xN
1+ (1 + t)η
yN
2− ϕ
xN
0− (1 + t)ψ
yN
0+ 1
a ψN
2+ µ (
η
xN
0x+ (1 + t)
2η
yN
0y))
dxdy.
By (5), we can estimate
∥(η, ϕ, ψ)(t)∥2 +
∫ t 0
( 1
1 + τ ∥ϕ(τ)∥2 + 1
1 + τ ∥(ϕ, ψ)x(τ)∥2 + (1 + τ)∥(ϕ, ψ)y(τ)∥2 )
dτ.
By combining (5) and (6), we can estimate
∫ t 0
( 1
1 + τ ∥ηx(τ)∥2 + (1 + τ)∥ηy(t)∥2 )
dτ.
Proceed the time-weighted estimates by
∂
y(5) · (1 + t)
2(aη, ϕ, ψ )
y, ∂
y2(5) · (1 + t)
4(aη, ϕ, ψ)
yy,
∂
y(5)
2· (1 + t)
2η
xy+ ∂
y(5)
3· (1 + t)
3η
yy.
Sketch of the proof of Theorem 2
Existence of viscous shock wave
Setting ξ = x − st, and (v, u) = (V, U)(ξ), we have the ODE system
−sVξ − Uξ = 0,
−sUξ + p(V )ξ = µ0( Uξ
V α+1)ξ,
with (V, U)(±∞) = (v±, u±). We only treat the case s > 0.
Integrating the system, we have
sV + U = sv± + u±,
−sU + p(V ) − µ0 Uξ
V α+1 = −su± + p(v±), and the “Rankine-Hugoniot condition”
{−s(v+ − v−) − (u+ − u−) = 0,
−s(u+ − u−) + p(v+) − p(v−) = 0.
Equation of V
V
ξ= V
α+1µ
0s h(V ), ξ ∈ R, V ( ±∞ ) = v
±where
h(V ) := s2(v− − V ) + p(v−) − p(V ) > 0, V ∈ (v∓, v±), and h(v±) = 0. Under the “entropy condition”
v− < v+, (i.e. u− > u+),
the solution V uniquely exists up to the shift of ξ, satisfying Vξ(ξ) > 0, v− < V (ξ) < v+, ξ ∈ R,
and U is given by U = u± + s(v± − V ).
Change of the variables : (x, t) → (ξ = x − st, t)
v
t− sv
ξ− u
ξ= 0,
u
t− su
ξ+ p(v)
ξ= µ
0( u
ξv
α+1)
ξ.
Change of the unknown variables : Define (ϕ, ψ)(t, ξ) by
ϕ(t, ξ) =
∫ ξ
−∞
(v − V )(t, η)dη, ψ(t, ξ) =
∫ ξ
−∞
(u − U)(t, η)dη, that is,
v = V + ϕ
ξ, u = U + ψ
ξ.
Cauchy problem for (ϕ, ψ) :
ϕ
t− sϕ
ξ− ψ
ξ= 0,
ψ
t− sψ
ξ+ p(V + ϕ
ξ) − p(V ) = µ
0( (U + ψ
ξ)
ξ(V + ϕ
ξ)
α+1− U
ξV
α+1) ,
with the initial data
(ϕ, ψ)(0) = (ϕ
0, ψ
0) ∈ H
2.
We look for the global solution in time such that (ϕ, ψ) ∈ C([0, ∞); H2), sup
x∈R |(ϕ, ψ)(t, x)| → 0 (t → ∞).
Proposition 4 (a priori estimate).
Suppose α ≥
12(γ − 1). Then there exist positive constants ε
0and C
0such that if (ϕ, ψ) ∈ C ([0, T ]; H
2) is the solution of the Cauchy problem for some T > 0 and
sup
t∈[0,T]
∥ (ϕ, ψ)(t) ∥
2≤ ε
0, it holds that for t ∈ [0, T ]
∥ (ϕ, ψ)(t) ∥
22+
∫
t0
( ∥ ϕ
ξ(τ ) ∥
21+ ∥ ψ
ξ(τ ) ∥
22) dτ ≤ C
0∥ ϕ
0, ψ
0∥
22.
A priori estimate
Rewrite the system as
ϕ
t− sϕ
ξ− ψ
ξ= 0,
ψ
t− sψ
ξ− K (V )ϕ
ξ− µ
0V
α+1ψ
ξξ= G, (5)
where
K(V ) = −p′(V ) + (α + 1)h(V ) V , and
G = − {p(V + ϕξ) − p(V ) − p′(V )ϕξ} + µ0{ 1
(V + ϕξ)α+1 − 1
V α+1}ψξξ + V α+1h(V ){− 1
(V + ϕξ)α+1 + 1
V α+1 − α + 1
V α+2 ϕξ}.
Recall
s > 0, v− < V < v+, Vξ = V α+1
µ0s h(V ) > 0, p′(V ) = −γp(V ) V , which implies
K(V ) ≥ γ p(v+) v+ . Basic energy estimate
There exists a positive constant C such that it holds that for t ∈ [0, T]
|| (ϕ, ψ)(t) ||2 +
∫ t
0
(|| ψξ(τ) ||2+ || √
Vξψ(τ) ||2)dτ
≤ C{|| ϕ0, ψ0 ||2 +
∫ t 0
∫
|ψ||G| dξdτ}.
Proof
We use a technical weighted energy method, “double step weighted energy method”, developed by Mei-M (1997) and Hashimoto-M (2007).
First-Step : Introduce the functions (transform functions) χ1 = χ1(V ) > 0, χ2 = χ2(V ) > 0,
and transform the unknown variables
ϕ = χ
1(V ) ˜ ϕ, ψ = χ
2(V ) ˜ ψ.
Then we have
(χ1ϕ)˜ t − s(χ1ϕ)˜ ξ − (χ2ψ)˜ ξ = 0,
(χ2ψ)˜ t − s(χ2ψ)˜ ξ − K(χ1ϕ)˜ ξ − µ0
V α+1(χ2ψ)˜ ξξ = G.
Second-Step :
Introduce the another set of functions (weight functions) W1 = W1(V ) > 0, W2 = W2(V ) > 0.
Multiply the first equation by W1ϕ, the second equation by˜ W2ψ˜ and sum them up. Then we have
d dt
∫ 1
2(W1χ1ϕ˜2 + W2χ2ψ˜2) dξ +
∫ s
2{(W1′χ1 − W1χ′1) ˜ϕ2+(W2′χ2 − W2χ′2) ˜ψ2}Vξ dξ +
∫
{(W1′χ2 − KW2χ′1)Vξϕ˜ψ˜ + (W1χ2 − KW2χ1) ˜ϕξψ˜} dξ +
∫
µ0( W2 V α+1
ψ)˜ ξ(χ2ψ)˜ ξ dξ =
∫
W2ψG dξ.˜
In order for the coefficients of the cross terms ˜ϕψ˜ and ˜ϕξψ˜ to banish, we impose that
W1′χ2 − KW2χ′1 = 0, W1χ2 − KW2χ1 = 0.
Choose χ1, χ2, W1 and W2 as
χ1(V ) = W1(V ) = 1, χ2(V ) = K(V )W(V ), W2(V ) = W(V ).
Then we have d
dt
∫ 1
2( ˜ϕ2 + KW2ψ˜2)dξ +
∫
(−s
2K′W2Vξψ˜2 + µ0{( W
V α+1)′(KW)′Vξ2ψ˜2 + (KW2
V α+1 )′Vξψ˜ψ˜ξ + KW2
V α+1 ( ˜ψξ)2}) dξ =
∫
WψG dξ.˜
It further deduces d
dt
∫ 1
2( ˜ϕ2 + KW2ψ˜2) dξ+
∫
(A(V )Vξψ˜2 + µ0KW2
V α+1 ( ˜ψξ)2)dξ
=
∫
WψG dξ,˜ where
A(V ) = −s
2K′W2 + 1
s( W
V α+1)′(KW)′V α+1h − 1
2s((KW2
V α+1 )′V α+1h)′.
We show the uniform positiveness of A(V ) for V ∈ [v
−, v
+].
Recall
p′(V ) = −γp(V )
V , h′(V ) = −s2 + γp(V ) V . Representations of K, K′ and K′′ in terms of h and p :
K(V ) = γp(V )
V + (α + 1)h(V ) V , K′(V ) = −γ(γ − α)p(V )
V 2 − (α + 1)s2
V − (α + 1)h(V ) V 2 , K′′(V ) = γ((γ − α)(γ + 2) − (α + 1))p(V )
V 3 + 2(α + 1) s2
V 2 + 2(α + 1)h(V ) V 3 .
Reformulate A(V ) as a quadratic form of h : A(V ) = W
2s(A0(V ) + A1(V )h(V ) + A2(V )h2(V )), where
A0(V ) =γ2(γ + 1)p2
V 3 W − 2(γ2p2
V 2 − γs2 p
V )W′, A1(V ) =(γ(γ + 2)(2α + 1 − γ) p
V 3 − 2s2(α + 1) V 2 )W
− 2(γ(2α + 1 − γ) p
V 2 − 2s2(α + 1)
V )W′ − 2γ p
V W′′, A2(V ) =2(α + 1)(−W
V 3 + W′
V 2 − W′′
V ).
Choice of W(V ) :
W(V ) = V. ( =⇒ A2(V ) = 0) Final form of A(V ) :
A(V ) = 1
2s{2s2γp(V ) + γ2(γ − 1)p2(V ) V +2γ2(α − 1
2(γ − 1))p(V )
V h(V ) + 2(α + 1)s2h(V )}.
By the physical assumption α ≥ 12(γ − 1),
A(V ) ≥ sγp(v
+), V ∈ [v
−, v
+].
Thus, integrating the inequality, we have
∥( ˜ϕ,ψ)(t)˜ ∥2 +
∫ t
0
(∥ψ˜ξ(τ)∥2 + ∥√
Vξψ(τ˜ )∥2) dτ
≤ C{∥ϕ˜0, ψ˜0∥2 +
∫ t
0
∫
|ψ˜||G| dξdτ}.
Recalling the transformations ϕ = ˜ϕ, ψ = V K(V ) ˜ψ, the proof is thus completed.
Thank You !
A remark
We give a remark that our arguments in the previous sections can be extended to the models with more general viscosity coefficient. In the book of Chapman-Cowling, the viscosity coefficient is given in more general form as
µ = ¯µ θ12F (θ).
It is noted that F(θ) = 1 for the Hard sphere Model, and F(θ) = θ
2 (s−1)
for the Cut-off inverse power force Model. As a typical model whose F is not a power function of θ, the Sutherland’s model is well known where
F(θ) = 1 1 + s0
θ
, (s0 > 0 : Sutherland’s constant).
Even for the above general case, if we also take
χ1(V ) = W1(V ) = 1, χ2(V ) = K(V )W(V ), W2(V ) = W(V ) = V
as in the proof of the Lemma 3, then after long calculations we can have the following :
A(V ) = 1
2s{2s2γp(V ) + γ2(γ − 1)p2(V ) V +2s2h(V )
(γ + 1
2 + (γ − 1)θF ′
F − (γ − 1)2θ(θF ′
F )′)
θ=(a/R)V1−γ
+2γ(γ − 1)h(V )p(V ) V
(
γθF ′
F + (γ − 1)θ(θF ′
F )′)
θ=(a/R)V1−γ
−(γ − 1)2h2(V ) V
(
γθ(θF′
F )′ + (γ − 1)θ2(θF′
F )′′)
θ=(a/R)V1−