LANDAU-GINZBURG TYPE EQUATIONS IN THE SUB CRITICAL CASE
NAKAO HAYASHI, ELENA I. KAIKINA, AND PAVEL I. NAUMKIN
Abstract. We study the Cauchy problem for the nonlinear Landau-Ginzburg equation
∂tu−α∆u+β|u|σu= 0, x∈Rn, t >0, u(0, x) =u0(x), x∈Rn, (0.1)
where α, β ∈ C with dissipation condition α > 0. We are interested in the subcritical case σ ∈ 0,n2
.We assume thatθ =
R
u0(x)dx
= 0 and δ(α, β)>0,where
δ(α, β) = β|α|n−n2σ
(2 +σ)|α|2+σα2
n 2 .
Furthermore we suppose that the initial datau0∈L1are such that (1 +|x|)au0∈ L1, with sufficiently small normε=(1 +|x|)au01, wherea∈(0,1).Also we assume thatσis sufficiently close to n2.Then there exists a unique solution of the Cauchy problem (0.1) such that
u(t, x)∈C((0,∞) ;L∞)∩C [0,∞);L1, satisfying the following time decay estimates for larget >0
u(t)∞≤Cεt−1σ.
Note that in comparison with the corresponding linear case the decay rate of the solutions of (0.1) is more rapid.
1. Introduction
This paper is devoted to the study of global existence and large time asymp- totic behavior of small solutions to the Cauchy problem for the complex Landau - Ginzburg equation
∂tu−α∆u+β|u|σu= 0, x∈Rn, t >0, u(0, x) =u0(x), x∈Rn
(1.1)
in the sub critical case σ∈ 0,n2
, whereα, β ∈C andαsatisfies the dissipation conditionα >0. In the super critical caseσ > n2 the global existence result can be easily obtained for anyαand β,with α >0, since the nonlinear term decays rapidly enough |u|σuL1 ≤ Ct−n2σ, when n2σ >1. In the so-called critical case σ= n2 and sub critical σ∈
0,n2
cases we can not expect a sufficient time decay of the nonlinear term for any α and β with α ≥ 0. Therefore we assume the
Date: April, 17, 2001.
1991Mathematics Subject Classification. 35Q35.
Key words and phrases. Dissipative Nonlinear Evolution Equation, Large Time Asymptotics, Landau-Ginzburg equation.
1
conditionδ(α, β)>0, where
δ(α, β) = β|α|n−n2σ
(2 +σ)|α|2+σα2 n2.
We also suppose that the mean value of the initial dataθ≡ u0(x)dx = 0.The nonlinear term in equation (1.1) have no enough regularity to get smooth solutions in the higher order Sobolev spaces, so using smoothing properties of the linear evolution group we consider the initial value problem under the assumptions of rather small regularity on the initial data u0 ∈ L1. To estimate the remainder terms in the large time asymptotic formulas we assume that the initial data decay at infinity as |x|au0(x) ∈ L1 with somea ∈ (0,1). Existence and uniqueness of local solutions for problem (1.1) were proved (see, e.g., [5], [6], [15]). The blow-up phenomena for positive solutions to the semilinear parabolic equation
ut−∆u=u1+σ (1.2)
(note that equation (1.2) is a particular case of (1.1) if α = 1 and β =−1) was obtained in paper [3] forσ∈
0,n2
, in [8] forσ= n2, n= 1,2, and in paper [14] for σ= n2, for anyn≥1.Global in time existence of small solutions to (1.2) was proved in [3] for the super critical caseσ > n2. Large time behavior of positive solutions was studied extensively for a particular case of (1.1) withα= 1 andβ = 1
ut−∆u=−u1+σ, σ >0 (1.3)
(see paper [12] for the super critical case σ > n2, [4] for the critical case σ = 2n and papers [1], [2], [7], [13] for the sub critical case σ ∈
0,n2
). We describe in more details the results on the asymptotic behavior of solutions to the Cauchy problem for equation (1.3) in the sub critical case σ ∈
0,n2
. In paper [7] it was proved that if the initial data are nonnegative u0 ≥ 0, u0 ∈ L1 and decay slowly at infinity as limx→±∞|x|2σu0(x) = +∞,then the solution of (1.3) has the asymptotic representation u(t, x) = t−1σσ−1σ +o
t−σ1
as t → ∞ uniformly in domains
x∈Rn; |x| ≤C√
t with any C >0. On the other hand in paper [1], there were considered the nonnegative initial data decaying sufficiently rapidly at infinity, i.e. 0≤u0(x)≤Ce−bx2 for all x∈Rn,with some b, C >0. Then it was shown that the main term of the asymptotic behavior of solution has a self-similar character u(t, x) = t−σ1w0
√x t
+o
t−σ1
as t → ∞ uniformly with respect to x∈Rn,wherew0(ξ) is a positive solution of the elliptic equation
−∆w−1
2ξ∇w+w1+σ= 1 σw (1.4)
which decays rapidly at infinity: lim|ξ|→∞|ξ|σ2 w0(ξ) = 0.This result was improved in paper [2], where the intermediate case was considered: if the initial data are such thatu0∈L1, u0 = 0 and lim|x|→∞|x|σ2u0(x) =κ>0,then the solutions have the asymptotic representation u(t, x) =t−1σwκ
√x t
+o
t−σ1
as t → ∞ uniformly with respect to x∈Rn,where wκ(ξ) is a positive solution of equation (1.4) such that lim|ξ|→∞|ξ|σ2wκ(ξ) =κ. Note that in these papers there was no restriction on the size of the initial data. The methods of these papers could not be applied for the case of complexα, β∈C,α >0 and as far as we know there are no results
on the asymptotic behavior of solutions to the Cauchy problem (1.1) in the case of complex coefficientsα, β.
We propose here a different approach, developed in our previous papers [9], [10]
and [11], where we considered the critical caseσ=n2.The main idea is to transform equation (1.1) via a change of the dependent variableu=e−ϕ+iψv to the form
vt−α∆v+βe−σϕ
|v|σ−1 θ
|v|σvdx
v= 0,
where the nonlinear term having zero mean value obtains better decay properties.
However we assume the smallness of the initial data and the restriction onσto be able to apply the contraction mapping principle.
To state our result precisely we give.
Notation and function spaces. The Fourier transformationFφ= ˆφis defined by ˆφ(ξ) = 1
(2π)n/2
e−ixξφ(x)dx andF−1φ(x) or ˇφ(x) is the inverse Fourier transform ofφ, i.e.
φ(x) =ˇ 1 (2π)n/2
eixξφ(ξ)dξ.
We denote by Lp for 1 ≤p≤ ∞, the usual Lebesgue space with a norm φp =
Rn|φ(x)pdx1/p
if 1≤p <∞andφ∞= ess.supx∈Rn|φ(x)|.Define byWp0,a= {φ∈Lp:xaφ∈Lp}weighted Sobolev space with a normφW0,ap =xaφLp, where x = √
1 +x2. By C(I;B) we denote the space of continuous functions from a time interval Ito the Banach space B. Different positive constants could be denoted by the same letterC.
We defineθ= u0(x)dx>0. We suppose thatσ∈ 0,n2
is sufficiently close to n2.
In the present paper we prove the following result.
Theorem 1.1. Let α, β ∈C, α >0, δ(α, β) >0. We assume that the initial data u0 ∈W01,a, a∈(0,1) have a sufficiently small norm ε=xau01 and are such that θ = u0(x)dx ≥ Cε with 0 < C < 1. We assume that n2 −ε3 <
σ < 2n. Then there exists a unique mild solution u(t, x) ∈ C
[0,∞) ;W01,a ∩ C((0,∞) ;L∞) of the Cauchy problem (1.1), satisfying the following time decay estimates
u(t)L∞ ≤Cεt−σ1
for large t >0. Furthermore there exist a number A and a functionV ∈ W10,a∩ W0,a∞ such that the asymptotic formula
u(t, x) =At−σ1V √x
t
eiωlogt+O
t−σ1−γ
, is valid fort→ ∞uniformly with respect tox∈Rn, whereω=−η1
β|V|σV(y)dy, γ=12min
a,1−n2σ
, V (ξ)is the solution of the integral equation V (ξ) = 1
(4πα)n2e−ξ
2
4α − β
η(4πα)n2 1
0
dz z(1−z)n2
e−
(ξ−y√z)2
4α(1−z) F(y)dy, (1.5)
whereη= 1−σn 2σ
β|V (y)|σV(y)dy, F(y) =|V(y)|σV(y)−V(y)
|V(ξ)|σV (ξ)dξ.
We organize our paper as follows. Section 2 is devoted to the proof of preliminary estimates in Lemmas 2.1 - 2.4. In Section 3 we prove Theorem 1.1 and the existence of solution V(ξ) to integral equation (1.5).
2. Preliminaries The solution of the linear Cauchy problem
ut−α∆u=f(t, x), x∈Rn, t >0, u(0, x) =u0(x), x∈Rn
can be written by the Duhamel formulau(t) =G(t)u0+t
0G(t−τ)f(τ)dτ ,where the Green operatorG is given by
G(t)φ=F−1e−αξ2tˆφ(ξ) =
G(t, x−y)φ(y)dy, andG(t, x) = (4παt)−n2 e−x2/4αt. Denoteϑ= (2π)n2 φˆ(0).
We first collect some preliminary estimates.
Lemma 2.1. The following estimates are true, provided that the right-hand sides are finite:
·bG(t)φ
p≤Ct−n2(1q−1p)t2b·bφ
q
and ifφ∈L1
|·|b(G(t)φ−ϑG(t))
p ≤ Ctb−a2 t−n2(1q−1p)·aφq +C|ϑ|tb2−n2(1−1p)
for allt >0, where1≤q≤p≤ ∞, b∈[0, a], a∈(0,1).
Proof. Since xb ≤ x−ybyb, by virtue of the Young inequality with 1p =
1
q +1r−1,we obtain ·bG(t)φ
p ≤ 1
(4πt|α|)n2 xb
e−(x−y)28t (1α+α1)|φ(y)|dy p
≤ Ct−n2
x−ybe−(x−y)28t (α1+α1)yb|φ(y)|dy p
≤ Ct−n2 ·be−Cx2/t
r
·bφ
q
≤ Ct−n2(1q−1p)tb2·bφ
q
for all t > 0, where b ≥ 0,1 ≤ p ≤ ∞. Hence the first estimate of the lemma is true. The second estimate for the case t∈[0,1] follows from the first one and the
inequality |·|bG(t)
p ≤Ctb2−n2(1−1p). We now consider the caset ≥1. We write for any 0≤b≤a≤1
|x|b(G(t)φ−ϑG(t)) = 1 (4πtα)n2
|x|b
|y|a
e−(x−y)24αt −e−4αtx2
|y|aφ(y)dy.
(2.1)
Changing the dependent variablesx=ξ√
t andy=η√
t, applying the inequality
|ξ|b
|η|a
e−(ξ−η)24α −e−ξ
2 4α
≤C|ξ−η|b+|η|b
|η|a e−(ξ−η)24α +C|ξ|b
|η|ae−ξ
2 4α
≤ C(1 +|ξ−η|)e−C(ξ−η)2+C|ξ|e−Cξ2 ≤Ce−C(ξ−η)2+Ce−Cξ2 for all|η| ≥1 and the estimate
e−(ξ−η)24α −e−ξ
2 4α
≤C|η|e−C(ξ−η)2+C|η|e−Cξ2 for all|η| ≤1,we obtain
y∈Rsupn
|x|b
|y|a
e−(x−y)24αt −e−4αtx2
r
= tb−a2 +2rn sup
η∈Rn
|ξ|rb
|η|ra
e−(ξ−η)24α −e−ξ
2 4α
rdξ 1r
≤ Ctb−a2 +2rn sup
η∈Rn e−C(ξ−η)2+e−Cξ2
dξ≤Ctb−a2 +2rn, (2.2)
where r∈[1,∞).The caser=∞is considered in the same way. Substitution of (2.2) into (2.1) via the Young inequality with 1p =1q +1r−1,yields
|·|b(G(t)φ−ϑG(t))
p ≤ Ct−n2 sup
y∈Rn
|x|b
|y|a
e−(x−y)24αt −e−4αtx2
r
|·|aφq
≤ Ctb−a2 −n2(1q−1p)·aφq
for allt≥1 and 1≤p≤ ∞.Therefore the second estimate of the lemma is valid.
Lemma 2.1 is proved.
Lemma 2.2. Let the functionf(t, x)have the zero mean valuefˆ(t,0) = 0and the norm
sup
t>0 sup
1≤s≤ptν+n2(1−1s)t−λ2·λf(t)
s=fF
be finite, where1≤p≤ ∞, λ∈(0,1), ν∈(0,1).We also suppose that the function g(t)is such thatg(t)≥ 12tµ for allt >0,whereµ >0is such thatµ+ν <1 +λ2. Then the following inequality is valid
·b t
0 g−1(τ)G(t−τ)f(τ)dτ p
≤Ctb2t1−ν−µ−n2(1−p1)fF, for allt >0, whereb∈[0, λ].
Proof. Since ˆf(t,0) = 0 we have |x|bG(t)f = |x|b
G(t)f−(2π)n2 fˆ(0)G(t) with anyb∈[0, λ].Applying the second estimate of Lemma 2.1 we obtain
|·|bG(t−τ)f(τ)
p≤Ct−τb−λ2 (t−τ)−n2(1q−1p)·λf(τ) (2.3) q
for all 0< τ < t, where 0≤ b≤ λ≤1, 1 ≤q ≤p≤ ∞. Therefore by virtue of (2.3) we get
|·|b t
0 g−1(τ)G(t−τ)f(τ)dτ p
≤ C t2
0 t−τb−λ2 (t−τ)−n2(1−1p)|·|λf(τ)
1τ−µdτ +C
t
2t
t−τb−λ2 |·|λf(τ)
pτ−µdτ
≤ CfF 2t
0 t−τb−λ2 (t−τ)−n2(1−1p)τ−ντλ2−µdτ +CfF
t
t2
t−τb−λ2 τ−ν−n2(1−1p)τλ2−µdτ
≤ CfFtb−λ2 t−n2(1−1p) t
2
0
τ−ντλ2−µdτ +CfFtλ2 t−ν−n2(1−1p) t
2t
t−τb−λ2 dτ
≤ Ctb2t1−ν−µ−n2(1−1p)fF (2.4)
for allt >0, where 1≤p≤ ∞,0≤b≤λ.Lemma 2.2 is proved.
Denoteζ= θσσ2
n2
(4π)n2σ(1−n2σ)δ(α, β).Since n2−ε3< σ < 2n andθσ≥Cεσ≥Cεn2, we can suppose thatζ= θσσ2
n2
(4π)n2σ(1−n2σ)δ(α, β)≥Cεn2−3≥1.
Lemma 2.3. We assume thatv˜∈W01,awitha∈(0,1)and sufficiently small norm
˜vW0,a
1 =ε and v(0) = (2π)˜ −n2 θ ≥Cε > 0. Let the function v(t, x) satisfy the estimates v(t)1+σ≤Cεt−2(1+σ)nσ andv(t)− G(t) ˜v1+σ≤Cε1+σt−2(1+σ)nσ for all t >0.Then the following inequality is valid
1 + σ θ
t
0 dτ
β|v|σv(τ , x)dx≥ 1
2t1−n2σ (2.5)
for allt >0.
Proof. Applying Lemma 2.1 we get
G(t) ˜v−θG(t)1+σ≤Cεt−2(1+σ)nσ t−a2 , then via the assumption
v(t)− G(t) ˜v1+σ≤Cε1+σt−2(1+σ)nσ
we have by the H¨older inequality |v|σv−θ1+σ|G|σG
1≤ |v|σv− |G(t) ˜v|σG(t) ˜v1 +|G(t) ˜v|σG(t) ˜v−θ1+σ|G|σG
1
≤ C
vσ1+σ+G(t) ˜vσ1+σ
v− G(t) ˜v1+σ +C
G(t) ˜vσ1+σ+θσGσ1+σ
G(t) ˜v−θG1+σ
≤ Cε1+2σt−n2σ+Cε1+σt−n2σt−a2 for allt >0. By a direct computation we obtain
β|G|σG(t, x)dx =
β
(4πt)n2σ+n2 |α|n2σαn2
e−x8t2σ(α1+α1)−4tαx2dx
= 2n2t−n2σ (4π)n2σ
β|α|n−n2σ
(2 +σ)|α|2+σα2 n2
= 2n2t−n2σ
(4π)n2σδ(α, β). Therefore we get
σ θ
β|v|σv(t, x)dx−θσσ2n2t−n2σ
(4π)n2σ δ(α, β)
=
σ θ
β|v|σv(t, x)dx−σ θθ1+σ
β|G|σG(t, x)dx
≤ C
θ |v|σv−θ1+σ|G|σG
1≤Cε2σt−n2σ+Cεσt−n2σt−a2 for allt >0, whence
σ θ
t
0 dτ
β|v|σv(τ , x)dx− θσσ2n2 (4π)n2σ
1−n2σδ(α, β)t1−n2σ
=
σ θ
t
0 dτ
β|v|σv(τ , x)dx−ζt1−n2σ
≤ Cε2σ t
0 τ−n2σdτ+Cεσ t
0 τ−n2στ−a2dτ
≤ Cεσζt1−n2σ+Cεσt1−n2σ−a2 ≤1 2ζt1−n2σ for allt >0. Thus we get
σ θ
t
0 dτ
β|v|σv(τ , x)dx≥1
2ζt1−n2σ. (2.6)
Estimate (2.6) implies (2.5), sinceζ≥1.Lemma 2.3 is proved.
Lemma 2.4. Let the function f(x) have the zero mean value fˆ(0) = 0 and the norm
·afp+·af1
be finite, where a∈(0,1), 1≤p≤ ∞.Then the following inequalities are valid
ξa 1
0
dz z(1−z)n2
dye−
(ξ−y√z)2
4α(1−z) f(y) p
≤C·af1+C·afp
and
ξa 1
0
dz (1−z)n2
1 z − t
tz dye−(ξ−y
√z)2
4α(1−z) f(y) p
≤ Ct−a2
·af1+·afp for allt >0, where1≤p≤ ∞.
Proof. By the Young inequality for convolutions we get ξa
e−
(ξ−y√z)2
4α(1−z) f(y)dy p
≤
ξ−y√ za
e−C(ξ−y
√z)2
1−z |f(y)|dy p
+
e−C(ξ−y
√z)2
1−z ya|f(y)|dy p
≤
ξae−Cξ
2 1−z
1fp+ e−Cξ
2 1−z
1·afp≤(1−z)n2 ·bf
p
for allz∈1
2,1 ,since
φ
· −y√ z
f(y)dy p
= z−2pn
φ(· −y)f y
√z
dy p
≤ Cz−2pn φ1 f
√· z
p
≤Cφ1fp. Therefore
ξa
1
12
1 z(1−z)n2
e−
(ξ−y√z)2
4α(1−z) f(y)dydz p
≤ C 1
12
(1−z)−n2 ξa
e−(ξ−y
√z)2
4α(1−z) f(y)dy p
dz≤C·afp (2.7)
and
ξa 1
12
dz (1−z)n2
1 z − t
tz dye−
(ξ−y√z)2
4α(1−z) f(y) p
≤ Ct−1 1
12
(1−z)−n2 ξa
e−
(ξ−y√ z)2
4α(1−z) f(y)dy p
dz
≤ Ct−1·afp, (2.8)
where 1≤p≤ ∞.
Via the condition
f(y)dy= 0,we write ξa
e−(ξ−y
√z)2
4α(1−z) f(y)dy p
=
ξa e−(ξ−y
√z)2
4α(1−z) −e− ξ
2 4α(1−z)
f(y)dy p
≤
ξa− ξ−y√
za e−(ξ−y
√z)2
4α(1−z)
f(y)dy p
+
ξ−y√ za
e−
(ξ−y√ z)2
4α(1−z) − ξae− ξ
2 4α(1−z)
f(y)dy p
≤ Cza2
e−C(ξ−y√
z)2ya|f(y)|dy p
+Cza2
e−C(ξ−y√
z)2+e−Cξ2
ya|f(y)|dy p
≤ Cza2 ·af1e−Cξ2
p≤Cza2·af1 for allz∈
0,12 , since
φ
· −y√ z
f(y)dy p
= z−2pn
φ(· −y)f y
√z
dy p
≤ Cz−2pn φp f
√· z
1≤Czn2(1−p1)φpf1
≤ Cφpf1.
Thus
ξa 1
2
0
1 z(1−z)n2
e−(ξ−y
√z)2
4α(1−z) f(y)dydz p
≤ 1
2
0
dz z(1−z)n2
ξa
e−(ξ−y
√z)2
4α(1−z) f(y)dy p
≤ C 12
0
dz
z1−a2(1−z)n2 ·af1≤C·af1 (2.9)
and
ξa 12
0
dz (1−z)n2
1 z− t
tz e−
(ξ−y√z)2
4α(1−z) f(y)dy p
≤ 1
2
0
dz (1−z)n2
1 z − t
tz ξa
e−
(ξ−y√ z)2
4α(1−z) f(y)dy p
≤ C 12
0
za2 (1−z)n2
1 z − t
tz
dz·af1≤Ct−a2 ·af1, (2.10)
where 1 ≤ p ≤ ∞. Collecting estimates (2.7) - (2.10), we get the results of the lemma. Lemma 2.4 is proved.
3. Proof of Theorem 1.1
The local existence of solutions for the Cauchy problem (1.1) can be obtained by the standard contraction mapping principle (for the proof see [5], [6], [16]).
Theorem 3.1. Let the initial datau0∈W10,a.Then for someT >0there exists a unique solutionu(t, x)∈C
[0, T);W10,a
∩C((0, T) ;L∞)of the Cauchy problem (1.1). Moreover if the initial data are sufficiently small·au01 ≤ε, then there exists a time T ≥ 1 such that the solution u(t, x) of the Cauchy problem (1.1) satisfy the estimates
1≤p≤∞sup sup
b∈[0,a] sup
t∈(0,T]tσ1−2pn t−2b·bu(t)
p≤2ε.
In order to get desired a-priori estimates of local solutions we make a change of the dependent variableu(t, x) =e−ϕ(t)+iψ(t)v(t, x),then we get from (1.1)
vt−α∆v+βe−σϕ|v|σv−(ϕ −iψ)v= 0.
(3.1)
Now let us demand that the real-valued functionsϕ(t) andψ(t) satisfy the following condition
βe−σϕ|v|σv−(ϕ −iψ)v dx= 0, hence via equation (3.1) we get dtd
v(t, x)dx= 0 for allt >0. Therefore βe−σϕ
|v|σvdx= (ϕ −iψ) (2π)n2 v(0,0).
If we chooseϕ(0) = 0 and ψ(0) = arg ˆu0(0) we have ˆ
v(t,0) = ˆv(0,0) =eϕ(0)−iψ(0)uˆ0(0)≡ |ˆu0(0)| ≡ θ (2π)n2 >0 and we obtain the equations
ϕ (t) = 1θe−σϕ β
|v|σvdx , ψ (t) =−1θe−σϕ
β
|v|σvdx (3.2) .
Thus we need to solve the following system
vt−α∆v+βe−σϕ
|v|σ−1θ
|v|σvdx v= 0, ϕ = 1θe−σϕ
β|v|σvdx , v(0, x) = ˜v(x), ϕ(0) = 0, (3.3)
where the initial data
˜
v(x) =u0(x) exp
−iarg
u0(x)dx
=u0(x) exp (−iarg ˆu0(0)). Multiplying the second equation of system (3.3) by the factor eσϕ(t), then inte- grating with respect to timet >0 and making a change of the dependent variable eσϕ(t)=g(t),we get an integral equation
v=A(v), (3.4)
where
A(v) (t) =G(t) ˜v−β t
0 g−1(τ)G(t−τ)f(τ)dτ and
f(τ) = |v|σv(τ)−v(τ) θ
|v|σv(τ , x)dx, g(τ) = 1 + σ
θ τ
0
β|v|σv(t, x)dxdt.
We define v0 = G(t) ˜v and successive approximations vk+1 = A(vk) for k = 0,1,2, ...We prove thatAis a contraction mapping in the set
X =
v∈C
[0,∞);W01,a
∩C((0,∞);L∞) :
1≤p≤∞sup sup
b∈[0,a]sup
t>0tn2(1−p1)t−b2·bv(t)
p≤2ε
.
We prove that the mapping A transforms the setX into itself, also we prove the estimates
vk ∈X; vk(t)− G(t) ˜v1+σ≤Cε1+σt−2(1+σ)nσ , (3.5)
gk(t)≥1
2t1−n2σ, (3.6)
fk(t, x)dx= 0,
vk(t, x)dx=θ (3.7)
for all t > 0 and k ≥ 0, where fk(t) = |vk|σvk(t)− vkθ(t)
|vk|σvk(t, x)dx and gk