Pointwise estimates and L p convergence rates to diffusion waves for p-system with damping
Weike Wang
aand Tong Yang
b,*
,1aDepartment of Mathematics, Wuhan University, Wuhan, China
bDepartment of Mathematics, City University of Hong Kong, 83 Tat Chee Ace, Hong Kong, Hong Kong Received March 10, 2001; revised April 17, 2002
Abstract
By introducing a new approximate Green function, we obtain the pointwise estimates on the solutions of Euler equations with linear frictional damping, from which we can deduce the optimalLp ð1pppþNÞconvergence rates to the nonlinear diffusion waves. The pointwise estimates andLp ð1ppo2Þconvergence rates given in this paper are new.
r2002 Elsevier Science (USA). All rights reserved.
MSC:35K; 76N; 35L
Keywords: p-System with damping; Nonlinear diffusion waves; Approximate Green function; Pointwise estimates
1. Introduction
In this paper, we study the time-asymptotic behavior of solutions to thep-system with frictional damping, which in Lagrangian coordinates can be written as
vtux¼0;
utþpðvÞx¼ au; a>0; p0o0;
(
ð1:1Þ with the initial data
ðv;uÞðx;0Þ ¼ ðv0ðxÞ;u0ðxÞÞ:
*Corresponding author.
E-mail address:matyang@math.cityu.edu.hk (T. Yang).
1Supported in part by the RGC Competitive Earmarked Research Grant 9040534.
0022-0396/03/$ - see front matterr2002 Elsevier Science (USA). All rights reserved.
PII: S 0 0 2 2 - 0 3 9 6 ( 0 2 ) 0 0 0 5 6 - 6
Herevðx;tÞ>0 anduðx;tÞrepresent the specific volume and velocity, respectively, anda>0 is the frictional coefficient. The pressurepðvÞis assumed to be a smooth function ofvwithpðvÞ>0; p0ðvÞoC0o0 forvunder consideration.
It was proved in [3] that the solutions to (1.1) time asymptotically behave like those governed by the Darcy’s law inL2 andLNnorms if the solution is away from vacuum. That is, as t tends to infinity, the solution ðvðx;tÞ;uðx;tÞÞ of (1.1) approaches to the solutionðvðx;% tÞ;uðx;% tÞÞgoverned by the following system with the same end states asvðx;0Þat infinity:
%
vt¼ 1apðvÞ%xx; pðvÞ% x¼ au:% (
ð1:2Þ
The convergence rates inLp ð2pppNÞnorms were studied in [9,10] by using the energy method and an approximate Green function.
However, the pointwise estimates and Lp ð1ppo2Þ convergence rate cannot be obtained by the method used in [10]. The reason can be explained as follows.
Since the large time behavior of the solution to the Euler equations with linear damping is governed by a parabolic equation derived by using Darcy’s law, the main idea in [10] is to consider a linear parabolic equation by putting the second-order derivative with respect to time variable to the right-hand side of the equation and treating it as a source. Notice that this second-order derivative term is linear. The advantage of this method is that the obtained linear equation becomes a heat equation with variable coefficient which depends on the diffusion wave.
Therefore, the approximate Green function can be defined easily which is a variation of the heat kernel. However, the second-order derivative with respect to time in the source requires higher-order derivatives on the solutions in the analysis.
Consequently, to close the decay estimates of the solution needs a crude decay estimates on the highest order derivatives of the solution with respect to time obtained by energy method. Since the energy method is crucial there, to our knowledge, this will not give any detailed pointwise estimates. But the estimates for Lp withpX2 can be obtained, cf. [10].
How to overcome the above difficulty is the main purpose of this paper. Here, we find a new way to construct an approximate Green function for the purpose of obtaining the pointwise estimates on the solutions so that a sharper result on the convergence rates in Lp ð1pppNÞ can be established. In particular, the L1 estimate is obtained. Our idea is to keep the second-order derivative with respect to time variable on the left-hand side of the equation. Therefore, the linear equation now is of the second order, and of hyperbolic type with linear damping and variable coefficient. Notice that here the source term is nonlinear where the highest order derivative has a factor of lower-order derivatives. In the following analysis, we first give a detailed pointwise estimates of the Green function to the above linear hyperbolic equation by treating the variable coefficient as a parameter through Fourier analysis. Then we introduce a new method to define an approximate Green function to the original linearized equation. Based on
these analysis, we can estimate the error due to the approximate Green function and the integral of the product of the approximate Green function and the nonlinear source to obtain the desired estimates. We should mention that some of the techniques used here in the Fourier analysis on the Green function come from the study on Navier–Stokes equations and hyperbolic systems with relaxation in [6,12].
Furthermore, as a preparation for the study of planar diffusion wave in multi- dimensional space, we give estimates on higher-order derivatives on the solutions in this one-dimensional setting. For this, we make some interesting improvement on the analysis in closing the a priori estimates in the last section.
There are other interesting problems concerning (1.1) with partial results obtained so far. One of them is the behavior of the solutions when vacuum appears. For this problem, the equivalence between (1.1) and (1.2) is known for some special families of solutions by construction and the LN solutions by compensated compactness.
Another problem is the existence of global weak solutions with finite total variation.
This was proved for the case when the end states of the initial data at infinities coincide. But the problem without this restriction is still open. Interested readers please refer to [1,2,4,5,13] and reference therein. Some interesting works on the convergence to the Darcy’s law from the compressible Euler flow and the study of limiting behavior of nonhomogeneous hyperbolic systems when the relaxed equilibrium is described by parabolic equations are done in [7,8]. There are also some works on the case with boundary and the case for nonisentropic gas which we do not refer here because it is irrelevant to this paper.
The rest of the paper is arranged as follows: The main theorems are stated in Section 2. In Section 3, we will first study the Green function with a parameter in details by Fourier analysis. In Section 4, the new approximate Green function is introduced and the proof of the main theorems is given in the last section.
Throughout this paper,Candbwill be used to denote a generic positive large and small constants, respectively.
2. The main result
As in [3,9,10], we are interested in the large time behavior of the solution of (1.1) with initial data satisfying
ðv;uÞðx;0Þ-ðv7;u7Þ as x-7N ð2:1Þ withvþ not necessarily being equal to v: Denote the self-similar solution, i.e. the diffusion wave, of (1.2) in the form ofjðpffiffiffiffiffiffitþ1x Þbyvðx;% tÞwith the same end states at infinities asvðx;0Þ;i.e.,
%
vðx;tÞ ¼f x ffiffiffiffiffiffiffiffiffiffi tþ1 p
; vð7% N;tÞ ¼v7: ð2:2Þ
Also set
%
uðx;tÞ 1
apðvÞ% x: ð2:3Þ
From [3,10], we know thatvðx;% tÞhas the following exponential decay property intþ1x2: jvðx;% tÞ vþjx2o1þtþ jvðx;tÞ vjx2>1þt
pCjvþvjeCa x
2
1þt;
j@kxvðx;% tÞjpCjvþvjð1þtÞk=2eCa x
2
1þt: ð2:4Þ
Since the u component of the solution is expected to decay exponentially in t at x¼7N;an auxiliary functions ð˜u;vÞ* was introduced in [3] as follows:
uðx;˜ tÞ ¼eat uþ ðuþuÞ Z x
N
m0ðyÞdy
ð2:5Þ and
*
vðx;tÞ ¼uþu
a eatm0ðxÞ; ð2:6Þ
wherem0ðxÞis a smooth function with compact support satisfying
Z N
N
m0ðxÞdx¼1:
It is easy to check thatð˜u;vÞ* satisfies
*
vtu˜x¼0;
˜
ut¼ au:˜ (
Let the initial datav0ðxÞbe a small perturbation of a diffusion wavevðx;% 0Þ:We are going to study how the solution behaves pointwise asttends to infinity. As in [3], there exists a shiftx0 such that the initial data satisfies:
Z N
N
ðv0ðyÞ vðy% þx0;0Þ vðy;* 0ÞÞdy¼0:
By the first equation of (1.1), this implies that the functionVðx;tÞdefined by Vðx;tÞ ¼
Z x N
ðvðy;tÞ vðy% þx0;tÞ vðy;* tÞÞdy; ð2:7Þ
satisfiesVð7N;tÞ ¼0:Herex0 is a constant uniquely determined by
Z N
N
ðvðx;0Þ vðx% þx0;0ÞÞdx¼uþu
a : ð2:8Þ
For later use, denote
Uðx;tÞ ¼uðx;tÞ uðx% þx0;tÞ uðx;˜ tÞ; ð2:9Þ V0ðxÞ ¼Vðx;0Þ and U0ðxÞ ¼Uðx;0Þ ¼Vtðx;0Þ: From (1.1), (1.2), (2.5)–(2.9), we have
VtU¼0;
Utþ ðpðVxþv%þvÞ * pðvÞÞ% xþaU¼1apðvÞ% xt; ðV;UÞjt¼0 ðV0;U0ÞðxÞ-0 as x-7N:
8>
<
>: ð2:10Þ
By linearizing the second equation of (2.10) about v;% we have the following system:
VtU ¼0;
Utþ ðp0ðvÞV% xÞxþaU ¼F1þF2;
ðV;UÞjt¼0 ðV0;U0ÞðxÞ-0 as x-7N; 8>
<
>: ð2:11Þ
whereFjðx;tÞ ¼ ðF˜jðx;tÞÞx ðj¼1;2Þ;and F˜1ðx;tÞ ¼1
apðvÞ%t;
F˜2ðx;tÞ ¼ ðpðVxþv%þvÞ * pðvÞ % p0ðvÞV% xÞ: ð2:12Þ From now on, we will study system (2.11). First, the following theorem is a direct consequence of the a priori estimates obtained in [3,9], and we omit its proof for brevity.
Theorem 2.1. For sufficiently smalle0;if
juþuj þ jvþvj þ jjV0jjHnþ3þ jjU0jjHnþ2pe0
then there exists a global in time solutionðVðx;tÞ;Uðx;tÞÞof(2.10) satisfying jjVðtÞjjHnþ3þ jjUðtÞjjHnþ2pCe0; ð2:13Þ where n is a positive integer.
Our main results of this paper are stated in the following.
Theorem 2.2. For mX3;if V0ðxÞand U0ðxÞsatisfy the condition of Theorem2.1for n¼2m;and for some sufficiently smalle0 and any positive number N>12;
juþuj þ jvþvjpe0; X
hpm
j@xhV0ðxÞj þ X
hpm2
j@hxU0ðxÞjpe0ð1þx2ÞN; ð2:14Þ then there exists a global in time solutionðVðx;tÞ;Uðx;tÞÞof(2.10), which satisfies
j@xkVðx;tÞjpCe0ð1þtÞðkþ1Þ=2BNðx;tÞ;
j@xlUðx;tÞj j@t@xlVðx;tÞjpCe0ð1þtÞðlþ3Þ=2BNðx;tÞ; ð2:15Þ where kpm1;lpm3and
BNðx;tÞ ¼ 1þ x2 1þt
N
:
That is,in the original functionðu;vÞ; we have
j@k1x ðvðx;tÞ vðx% þx0;tÞ vðx;* tÞÞjpCe0ð1þtÞðkþ1Þ=2BNðx;tÞ;
j@lxðuðx;tÞ uðx% þx0;tÞ uðx;˜ tÞÞjpCe0ð1þtÞðlþ3Þ=2BNðx;tÞ:
Remark 2.1. If the initial perturbation decays exponentially in x; then the perturbation decays exponentially inxfor any time. In fact, if
juþuj þ jvþvjpe0; X
hpm
j@xhV0ðxÞj þ X
hpm2
j@hxU0ðxÞjpe0eCx2; ð2:16Þ
then, forkpm1; lpm3;ðVðx;tÞ;Uðx;tÞÞsatisfies:
j@xkVðx;tÞjpCe0ð1þtÞðkþ1Þ=2ebx
2
1þt;
j@xlUðx;tÞjpCe0ð1þtÞðlþ3Þ=2ebx
2
1þt: ð2:17Þ
As a corollary of Theorem 2.2, we have the following theorem.
Theorem 2.3. For pA½1;N and mX2; if V0ðxÞand U0ðxÞ satisfy the conditions of Theorem 2.1 for n¼2m and (2.14), then there exists a global in time solution
ðVðx;tÞ;Uðx;tÞÞof(2.10),which satisfies:
jj@xkVðtÞjjLppCe0ð1þtÞ12ð1þk1pÞ; jj@xlUðtÞjjLppCe0ð1þtÞ
1 2ð3þl1
pÞ
; ð2:18Þ
where kpm1; lpm3:That is,for the functions ðu;vÞ;we have jj@xk1ðvðx;tÞ vðx% þx0;tÞ vðx;* tÞÞjjLppCe0ð1þtÞ12ð1þk1pÞ; jj@xlðuðx;tÞ uðx% þx0;tÞ uðx;˜ tÞÞjjLppCe0ð1þtÞ
1 2ð3þl1
pÞ
:
3. Green function with a parameter
In this section, we first consider a linear partial differential equation with constant coefficient and parameterlas follows:
VttlVxxþaVt¼0: ð3:1Þ IfGnðl;x;tÞis the Green function of (3.1), then it satisfies:
Gttnðl;x;tÞ lðGnðl;x;tÞÞxxþaGntðl;x;tÞ ¼0;
Gnðl;x;0Þ ¼0; Gntðl;x;0Þ ¼dðxÞ; ð3:2Þ wherelis a parameter satisfyingC1>l>C0>0 with constantsC1andC0;anddis the Dirac function.
We now establish estimates of Green function of (3.2) which will be used to obtain the pointwise estimates for the approximate Green function. This can be done by Fourier analysis as in [6] for the Navier–Stokes equations and [12] for hyperbolic system with relaxation.
The Fourier transform offðxÞis fðx;ˆ tÞ ¼
Z
R
fðx;tÞeixxdx;
and the inverse Fourier transform is
fðx;tÞ ðF1fÞðx;ˆ tÞ ¼ ð2pÞ1 Z
R
fðx;ˆ tÞeixxdx:
The symbol of the operator for Eq. (3.1) is
t2þlx2þat: ð3:3Þ
Here,tandxcorrespond to@t@ andDxrespectively, andDx¼1i@@x:It is easy to see that the eigenvalues of (3.3) fortare
t¼m7ðxÞ 12ða7 ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi a24lx2 q
Þ: ð3:4Þ
From (3.2), we have by direct calculation that
Gˆnðl;x;tÞ ¼ ða24lx2Þ1=2ðemþðxÞtemðxÞtÞ: ð3:5Þ For convenience, we sometimes decompose Gˆnðl;x;tÞ ¼Gˆþðl;x;tÞ þGˆðl;x;tÞ;
where
Gˆ7ðl;x;tÞ ¼7m10 em7ðxÞt; m0¼ ða24lx2Þ1=2: Notice that
@lðGˆ7Þðl;x;tÞ ¼ ð2x2m20 8x2tm10 ÞGˆ7ðl;x;tÞ: ð3:6Þ For Gˆnðl;x;tÞ and Gˆ7ðl;x;tÞ; we first have the following lemma from standard consideration.
Lemma 3.1. Gˆnðl;x;tÞis a holomorphic function ofx:Gˆ7ðl;x;tÞalso is a holomorphic function ofxexcept isolated singularitiesx¼72paffiffil:
In the following, we are going to obtain some detailed properties of the Green functionGnðl;x;tÞ:Denote
Gnðl;x;tÞ ¼ Z e
e
Gˆnðl;x;tÞeixxdxþ Z
epjxjpR
Gˆnðl;x;tÞeixxdx þ
Z
RpjxjpN
Gˆnðl;x;tÞeixxdx
¼:Gn1ðl;x;tÞ þGn2ðl;x;tÞ þG3nðl;x;tÞ:
ForG7;the correspondingG7j ðj¼1;2;3Þcan be defined in the same way.
Lemma 3.2. For sufficiently smalle;there exist positive constants C and b;such that j@tl@hlDkxG1þðl;x;tÞjpCð1þtÞð1þ2lþkÞ=2ebx2=ð1þtÞ; ð3:7Þ
j@lt@lhDkxG1ðl;x;tÞjpCeat=2ebx2=ð1þtÞ ð3:8Þ for any non-negative integers l;h and k:
Proof. SinceGˆ7is a holomorphic function ofxat the origin. Thus, we can move the path of integration from½e;etoSðe;e;cÞ;where
Sða;b;cÞ ¼ fxjRex¼a;Imx: 0-cg [fxjImx¼c;Rex:a-bg [fxjRex¼b;Imx:c-0g:
Letx¼zþiZ;zandZare real numbers. Sincejxjpeande>0 is sufficiently small, we have
ReðmþtþixxÞ ¼ ða1lZ2txZÞ a1lz2tþOð1ÞðZ4þz4Þt;
ReðmtþixxÞ ¼ atþ ða1lZ2txZÞ þa1lz2tþOð1ÞðZ4þz4Þt:
We only prove (3.7) in the following since the corresponding estimate onG1can be obtained in the same way. In fact, from the above identities, we have
@tl@lhDkx Z e
e
Gˆþðl;x;tÞeixxdx
¼Oð1Þ Z e
e
eða1lZ2txZÞea1lz2teOð1ÞðZ4þz4ÞtðZ2þz2Þk=2þl ððZ2þz2Þð1þtÞÞhdz
þOð1Þ Z c
0
eða1lZ2txZÞea1lz2teOð1ÞðZ4þz4ÞtðZ2þz2Þk=2þl ððZ2þz2Þð1þtÞÞhdZ:
For ajxj2ltoe2; we let c¼2ltax; then ða1lZ2txZÞjZ¼c¼ax2lt2: Denote the left-hand of (3.7) byRþl;h;k;then
Rþl;h;kpCð1þtÞk=2leax
2
16lt
Z e e
e4a1lz2teOð1ÞðZ4þz4Þt ððZ2þz2Þð1þtÞÞk=2þhþldz
þCð1þtÞk=2le4a1le2t Z 2ltax
0
eð4a1lZ2tjxjZÞeOð1ÞðZ4þz4Þt ððZ2þz2Þð1þtÞÞk=2þlþhdZ
pCð1þtÞðkþ1Þ=2lðebx2=ð1þtÞþebtÞ:
Forajxj2ltXe
2;we letc¼e2signx;then Rþl;h;kpCð1þtÞk=2leejxj=4
Z e e
e4a1lz2teOð1ÞðZ4þz4Þt ððZ2þz2Þð1þtÞÞk=2þhþldz
þCð1þtÞk=2le3a1le2t Z e=2
0
ejxjZeOð1ÞðZ4þz4Þt ððZ2þz2Þð1þtÞÞk=2þlþhdZ
pCð1þtÞðkþ1Þ=2lðebtþeb1jxjÞ pCð1þtÞðkþ1Þ=2lebt:
This completes the proof of the lemma. &
Lemma 3.3. For fixedeand R;there exist positive constants b and C;such that j@lt@lhDkxGn2ðl;x;tÞjpCebt: ð3:9Þ
Proof. For any fixede andR; we can choose b1>0 sufficiently small such that if epjxjpR
jemþðxÞtjpCeb1t: As foremðxÞt;since RemðxÞpa2;we have
jemðxÞtjpCeat=2: Thus ifjm0ðxÞjXe1>0;one has
jGˆn2ðl;x;tÞjpCeb2t:
Notice that emþðxÞtemðxÞt¼Oð1Þeat=2m0ðxÞt if jm0ðxÞjoe1; and e1 is sufficiently small, we also get above inequality. Thus, for any fixedeandR;we deduce that
jGn2ðl;x;tÞjpCeb2t Z
epjxjpR dxpCeb2t:
By applying@tl@lhDkx on both sides of (3.1), we can prove similarly that j@tl@hlDkxG2nðl;x;tÞjpCebt
Z
epjxjpRð1þ jxjÞkþlþhdxpCebt for someb>0;which implies (3.9). &
Now we considerGn3ðl;x;tÞ:By letting
mZ¼ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Z4l=a2
p ;
and then taking the Taylor expansion ofmZ inZ;we have mZ¼i2 ffiffiffi
pl
a þiZ a ffiffiffil p
þOð1ÞðZ2Þ: ð3:10Þ
Since
m7ðxÞ ¼12ða7xa ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x24l=a2 q
Þ;
whenjxjis sufficiently large, we arrive at m7ðxÞ ¼1
2 a72i ffiffiffi pl
x7i a2 ffiffiffil
p xþOðjxj3Þ
:
This implies that
em7ðxÞt ¼eat=2e7ipffiffilxt e7i
a2t 2pffiffilx
ð1þOð1Þðjxj3ÞtÞ: ð3:11Þ Eq. (3.11) will be used in the proofs of the following lemmas.
Lemma 3.4. For R sufficiently large,there exists a constant C>0;such that jGn3ðl;x;tÞjpCeat=4: ð3:12Þ Moreover,there exists a distribution
Kl;k;hn ðl;x;tÞ ¼eat=2lþkþh1X
j¼0
ðqþjþhðtÞdðlþkþhjÞðxþ ffiffiffi pl
tÞ þqjþhðtÞdðlþkþhjÞðx ffiffiffi
pl tÞÞ;
such that when lþkX1;we have
j@tl@lhDkxGn3ðl;x;tÞ Kl;k;hn ðl;x;tÞjpCeat=4; ð3:13Þ where q7jþhðtÞ; j¼0;y;lþk are polynomials of t with degrees not greater than jþh correspondingly.
Proof. Since forjxjlarge enough,
m10 ðxÞ ¼ i 2 ffiffiffi
pl
xþOðjxj2Þ;
Eq. (3.11) gives
Gˆnðl;x;tÞ ¼ i 2 ffiffiffi pl
xeat=2ðeipffiffilxt
eipffiffilxt Þ
þOðjxj2Þð1þtþOðjxj1ÞtÞeat=2: Thus
jGn3ðl;x;tÞjpCeat=2 Z
RpjxjpN
i 2 ffiffiffi pl
xðeipffiffilxt
eipffiffilxt Þeixxdx
þCeat=4 pCeat=2
Z N
R
sinðx ffiffiffi pl
tÞxsinðxþ ffiffiffi pl
tÞx
x dx
þCeat=4
pCeat=4;
which implies (3.12). For the general case, since
@tl@lhxkGˆnðl;x;tÞ
¼Oð1Þeat=2 eipffiffilx lþkþhX
j¼0
qþjþhðtÞjxjjþqþlþkþ2hþ1ðtÞjxjlkh1
!
þeipffiffilx lþkþhX
j¼0
qjþhðtÞjxjjþqlþkþ2hþ1ðtÞjxjlkh1
!!
jxj1þkþlþh if we set
Kˆnl;k;hðl;x;tÞ ¼Oð1Þeat=2
Z N
N
eipffiffilx lþkþh1X
j¼0
qþjþhðtÞjxjj
!
þeipffiffilx lþkþh1X
j¼0
qjþhðtÞjxjj
!!
jxj1þkþhþleixxdx;
where q7mðtÞ is polynomial of t with degree not greater than m; and jqþjþhðtÞj ¼ jqjþhðtÞjthen we can deduce that
j@tlDkx@lhG3nðl;x;tÞ Kl;k;hn ðl;x;tÞj pZ
jxjXR
ð@tlxk@hlGˆnðl;x;tÞ Kˆnl;k;hðl;x;tÞÞeixxdx
þ
Z
jxjpR
Kˆnl;k;hðl;x;tÞdx
pCeat=3 Z
jxjXR
ðeipffiffilx
qþlþkþ2hðtÞx1þeipffiffilx
qlþkþ2hðtÞx1Þeixxdx
þeat=3
Z
jxjXR
eOðjxj1ÞtOðjxj2Þeixxdx
þCeat=4: Similar to the proof of (3.12), we get (3.13). &
Lemma 3.5. For M>0 sufficiently large, there exists a constant C>0; such that whenjxj=ð1þtÞXM;we have
j@tlDkx@lhGnðl;x;tÞ Kl;k;hn ðl;x;tÞjpCeat=4ejxj2=ð1þtÞ: ð3:14Þ
Proof. By Lemma 3.1, we can move the path of integration toSðN;N;2x=ð1þ tÞÞ:Then
Gnðl;x;tÞ ¼C
Z N
N
Gˆnðl;zþ2ix=ð1þtÞ;tÞeixze2x2=ð1þtÞdz:
Sincejxj=ð1þtÞ>M andM is sufficiently large, we know thatjxj ¼ jzþ2ix=ð1þ tÞj is large enough and then (3.11) can be applied to variable zþ2ix=ð1þtÞ:
Combining this observation with the above identity imply that Gnðl;x;tÞ
¼Oð1Þeat=2 eð2x2þpffiffilxtÞ=ð1þtÞZ N N
ðzþ2ix=ð1þtÞÞ1eiðxþpffiffiltÞz dz
þeð2x2pffiffilxtÞ=ð1þtÞZ N N
ðzþ2ix=ð1þtÞÞ1eiðxpffiffiltÞz dz
þOð1Þeat=2e2x2=ð1þtÞ
Z N
N
Oððz2þ ð2x=ð1þtÞÞ2Þ1ð1þteOðz2þð2x=ð1þtÞÞ2ÞtÞÞdz:
Since
1
zþ2ix=ð1þtÞ¼1
z 2ix=ð1þtÞ
z2þ ð2x=ð1þtÞÞ2 ð2x=ð1þtÞÞ2 z2þ ð2x=ð1þtÞÞ2
1 z; we have
Z N
N
ðz72ix=ð1þtÞÞ1eiðxþpffiffiltÞz dz
pC jx7 ffiffiffi pl
tj þ jxj
1þtþ jxj2 ð1þtÞjx7 ffiffiffi
pl tj
! :
In summary, by noticing thatjxj=ð1þtÞ>M;we have jGnðl;x;tÞjpCeat=2ex2=ð1þtÞ
jx7 ffiffiffi pl
tj þ jxj
1þtþ jxj2 ð1þtÞjx7 ffiffiffi
pl
tjþ1þet=M
!
pCeat=2ebx2=ð1þtÞ:
Thus, we have (3.14) forl¼k¼h¼0:The general case can be proved similarly as in Lemma 3.5. &
Combining Lemmas 3.2 with 3.5, we have
Theorem 3.1. For non-negative integers l; k and h; there exists a distribution Kl;k;hn ðl;x;tÞ;such that
j@tlDkx@lhGnðl;x;tÞ Kl;k;hn ðl;x;tÞjpCð1þtÞð1þkþ2lÞ=2ebx2=t: ð3:15Þ
Proof. Ifjxj=ð1þtÞpM;we have tpM2x2
1þt; jx7 ffiffiffi pl
tjpðMþ ffiffiffi pl
Þð1þtÞ:
Thus, (3.15) holds for jxj=ð1þtÞpM by Lemmas 3.2–3.4. For jxj=ð1þtÞ>M;
Lemma 3.5 implies (3.15), and this completes the proof. &
4. Approximate Green function
In this section, we construction the approximate Green function for the unknown functionVðx;tÞ:
Vtt ðaðx;tÞVxÞxþaVt¼F; ð4:1Þ where aðx;tÞ ¼ p0ðvðx;% tÞÞ>C0 >0; F ¼F1þF2: The constructed approximate Green functionGðx;t;y;sÞfor (4.1) satisfies the basic requirement
Gðx;t;y;tÞ ¼0; Gsðx;t;y;tÞ ¼dðyxÞ: ð4:2Þ By multiplying (4.1) whose variables are now changed toðy;sÞbyGand integrating over the regionðy;sÞAR ð0;tÞ;we have
Vðx;tÞ
¼
Z N
N
Gsðx;t;y;0ÞV0ðyÞdyþ
Z N
N
Gðx;t;y;0ÞðaV0ðyÞ þVtðy;0ÞÞdy þ
Z t 0
Z N
N
Gðx;t;y;sÞFðy;sÞdy ds þ
Z t 0
Z N
N
ððGssaGsÞðx;t;y;sÞ
ðaðy;sÞGyðx;t;y;sÞÞyÞVðy;sÞdy ds: ð4:3Þ Ifaðy;sÞis a constant andGis a Green function of (4.1), then we know that the last integral of (4.3) is equal to zero. However, it now depends on the profile of the diffusion wave. Therefore, we can only try to minimize the termGssaGs ðaGyÞy: For this purpose, we compare it with the linear partial differential equation with constant coefficient and a parameter l; i.e. (3.1). From the discussion of the last section, the Green function Gnðl;x;tÞ satisfies (3.2) and Theorem 3.1. Define an approximate Green function for (4.1) as follows:
Gðx;t;y;sÞ ¼Gnðaðy;sðt;sÞÞ;xy;tsÞ; ð4:4Þ wheresðt;sÞAC3ðR2Þand
sðt;sÞ ¼ s; s>t=2þ1;
t=2; spt=2:
(
Moreover, we can choosesðt;sÞto be smooth whensAðt=2;t=2þ1Þ;such that X
1pl1þl2p3
j@tl1@sl2sðt;sÞjpC: ð4:5Þ
Whent>1;we also obtain
sðt;sÞ1p C
1þt: ð4:6Þ
It is clear that the approximate Green function defined in (4.4) satisfies the condition (4.2), and
G33nðaðy;sÞ;xy;tsÞ aðy;sÞG22nðaðy;sÞ;xy;tsÞ
þaGn3ðaðy;sÞ;xy;tsÞ ¼0; ð4:7Þ whereGnj ¼Gnjðaðy;sÞ;xy;tsÞdenotes the partial derivative with respect to the jth ðj¼1;2;3;4Þvariable of Gn:
The functionGðx;t;y;sÞis not symmetric with respect to the variablesx;tandy;s:
Instead, we have the following relations:
@xG¼ @yGþ@aðGnÞay;
@tG¼ @sGþ@aðGnÞðasþatÞ; ð4:8Þ whereay;at andasrepresent the derivatives of awith respect toy;t;s;respectively.
Sinceaðy;sðt;sÞÞ ¼ p0ðvðy;% sðt;sÞÞÞ;it follows from [10] that
j@tl1@sl2@ykaðy;sðt;sÞÞjpCe0ð1þtÞðk=2þl1þl2Þ ð4:9Þ for kþl1þl2X1: Now we give a pointwise estimate to the approximate Green function Gðx;t;y;sÞ: In fact, from Theorem 3.1 and the above discussion, it is straightforward to have the following theorem.
Theorem 4.1. For l¼0;1and a positive integer k;there exists a distribution Kl;kðx;y;t;sÞ
¼eaðtsÞ=2 lþk1X
j¼0
ð˜qþj ðtsÞdðlþkjÞðxyþ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi aðy;sðt;sÞÞ
p Þ
þq˜jðtsÞdðlþkjÞðxy ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi aðy;sðt;sÞÞ
p Þ; ð4:10Þ
such that
X
l1þl2¼l;k1þk2¼k
ð@tl1@ls2@xk1@ky2Gðx;t;y;sÞ Kl;kðx;y;t;sÞÞ
pCðð1þtÞð1þkþ2lÞ=2þ ð1þtsÞð1þkþ2lÞ=2ÞebðxyÞ2=ðtsÞ; ð4:11Þ
where l¼0;1 and ˜q7j ðj¼0;y;lþk1Þ are polynomials of t with degrees not greater than j correspondingly.
Given a functiongðy;sÞ;andkX1;from (4.8) and (4.9), we have Z b
a
Z N
N
@xhþk@ltGðx;t;y;sÞgðy;sÞdy ds
¼ Z b
a
Z N
N
@hx@tlGðx;t;y;sÞ@ykgðy;sÞdy ds
þOð1Þe0
X
k0ok
Z b a
Z N
N
ðð1þtsÞð1þ2lþhÞ=2 ð1þtÞðkk0ÞÞebðxyÞ2=ðtsÞ
þKl;hðx;y;t;sÞÞ@yk0gðy;sÞdy ds: ð4:12Þ Set the error due to the approximate Green function byRG;
RGGssðx;t;y;sÞ aGsðx;t;y;sÞ ðaðy;sÞGyðx;t;y;sÞÞy: Then (4.7) implies that
RG¼ ðaðy;sÞ aðy;sÞÞGn22þ2G13nasðy;sÞ þGn11ðasðy;sÞÞ2 þG1nassðy;sÞ aðy;sÞð2G12nayðy;sÞ þGn11ðayðy;sÞ2Þ
þG1nayyðy;sÞ aGn1asðy;sÞ: ð4:13Þ Notice that
jaðy;sÞ aðy;sÞj
¼ Oð1Þe0ð1þtsÞ1=2ð1þsÞ1=2e
by2
1þt; sot=2þ1;
0; sXt=2þ1:
8<
: ð4:14Þ
It follows from Theorem 4.1, (4.9) and (4.13) that there exists a distribution KGðx;y;t;sÞ
¼ ðaðy;sÞ aðy;sÞÞK2;0;0n þ2K0;1;1n asðy;sÞ þK0;0;2n ðasðy;sÞÞ2 þK0;0;1n assðy;sÞ aðy;sÞð2K1;0;1n ayðy;sÞ þK0;0;2n ðayðy;sÞ2Þ þK0;0;1n ayyðy;sÞ aK0;0;1n asðy;sÞ;
such that
RG¼Oð1Þe0Yðy;t;sÞðð1þtsÞ1=2ebðxyÞ2=ðtsÞþKGðx;y;t;sÞÞ; ð4:15Þ where
Yðy;t;sÞ ¼ ðð1þtÞ1þ ð1þtsÞ1=2ð1þsÞ1=2Þe
by2 1þt:
Furthermore, it can be shown by straightforward calculation that there exists a distribution
KG;l;kðx;y;t;sÞ
¼eaðtsÞ=2 lþkþ1X
j¼0
ðQþj ðtsÞdðlþkþ2jÞðxyþ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi aðy;sðt;sÞÞ
p Þ
þQj ðtsÞdðlþkþ2jÞðxy ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi aðy;sðt;sÞÞ
p Þ;
such that
X
l1þl2¼l;k1þk2¼k
@lt1@sl2@xk1@yk2RGðx;t;y;sÞ
¼Oð1Þe0Yðy;t;sÞðð1þtsÞ1þkþ2l2 ebðxyÞ2=ðtsÞ
þKG;l;kðx;y;t;sÞÞ: ð4:16Þ
Here l¼0;1 and Q7j ðj¼0;y;lþkþ1Þ are polynomials of t with degrees not greater thanj correspondingly.
In deducing the pointwise estimates, we need to perform integrations by parts for RG and a given functiongðy;sÞ:For later use, we list the following two identities:
Z b a
Z N
N
@xkRGðx;t;y;sÞgðy;sÞdy ds
¼ Z b
a
Z N
N
RGðx;t;y;sÞ@kygðy;sÞdy ds þOð1Þe0
X
lok
Z b a
Z N
N
Yðy;t;sÞ
ðð1þtÞðklÞ=2Þð1þtsÞ1=2ebðxyÞ2=ðtsÞ
þKGðx;y;t;sÞÞ@ylgðy;sÞdy ds ð4:17Þ
and Z b
a
Z N
N
@t@xkRGðx;t;y;sÞgðy;sÞdy ds
¼ Z b
a
Z N
N
RGðx;t;y;sÞ@s@ykgðy;sÞdy ds
Z N
N
@kxRGðx;t;y;sÞgðy;sÞdyjs¼bs¼a
þOð1Þe0
X
lok
Z b a
Z N
N
Yðy;t;sÞðð1þtÞð2þklÞ=2Þð1þtsÞ1=2
ebðxyÞ2=ðtsÞþKGðx;y;t;sÞÞ@ylgðy;sÞdy ds: ð4:18Þ All the above pointwise estimates and identities will be used in the next section to obtain pointwise estimates and theLp convergence rates on the solutions to (4.1).
5. The Proof of Theorem 2.2
From (4.3), we have
@tl@kxVðx;tÞ ¼I1l;kþI2l;kþI3l;k; ð5:1Þ where
I1l;k¼
Z N
N
@tl@xkGsðx;t;y;0ÞVðy;0Þdy
Z N
N
@tl@xkGðx;t;y;0ÞðaVðy;0Þ þVtðy;0ÞÞdy;
I2l;k¼@tl Z t
0
Z N
N
@xkGðx;t;y;sÞFðy;sÞdy ds;
I3l;k¼@tl Z t
0
Z N
N
@xkRGðx;t;y;sÞVðy;sÞdy ds:
We will derive the estimates on Ijl;k ðj¼1;2;3Þ in the following. First, we give a lemma which will be used in the analysis.
Lemma 5.1. If0oC0plpC1;then for a positive number N>12;we have BNðx7 ffiffiffi
pl
t;tÞpCð1þtÞ2NBNðx;tÞ; ð5:2Þ
Z N
N
ebðxyÞ2=tð1þy2ÞNdypCBNðx;tÞ ð5:3Þ
and
Z N
N
ebðxyÞ2=ðtsÞBNðy;sÞdypC
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð1þtsÞð1þsÞ
1þt r
BNðx;tÞ: ð5:4Þ
Proof. Sinceex2=tpCBNðx;tÞ;(5.2) and(5.3) are direct consequences of Lemmas 3.4 and 5.1 of [11]. For (5.4), we only need to prove that
JN ¼:
Z N
N
BNðxy;tsÞBNðy;sÞdypC
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð1þtsÞð1þsÞ
1þt r
BNðx;tÞ:
Ifx2pt;then 1p2NBNðx;tÞ:Thus,
JNpZ N N
BNðxy;tsÞdypCe0ð1þtsÞ1=2:
Also we can obtain INpCe0ð1þsÞ1=2 similarly. Thus, (5.4) hold when x2pt: For case whenx2>t;since
1þðxyÞ2 1þts
!N
1þ y2 1þs
N
p Cð1þ1þtsx2 ÞNð1þ1þsy2ÞN; jxjXjyj=2;
Cð1þðxyÞ1þts2ÞNð1þ1þsx2ÞN; jxjpjyj=2;
8<
:
we have
JNpC
Z N
N
1þ x2 1þts
N
1þ y2 1þs
N
dy
þC
Z N
N
1þðxyÞ2 1þts
!N
1þ x2 1þs
N
dy
pC 1þts 1þt
N
ð1þsÞ1=2þ 1þs 1þt
N
ð1þtsÞ1=2
! BNðx;tÞ
pC
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð1þtsÞð1þsÞ
1þt r
BNðx;tÞ:
Thus the lemma is proved. &
ForI1l;k;we know from Theorem 4.1 and the conditions in Theorem 2.2 that
I1l;k¼
Z N
N
ð@lt@xkGsðx;t;y;0Þ Klþ1;kðxy;tÞÞVðy;0Þdy
þ
Z N
N
ð@tl@xkGðx;t;y;0Þ Kl;kðxy;tÞÞðaVþVtÞðy;0Þdy
þ
Z N
N
ðKlþ1;kðxy;tÞVðy;0Þ þKl;kðxy;tÞðaVþVtÞðy;0ÞÞdy
pCe0ð1þtÞ1þkþ2l2
RN
NeðxyÞ2=tð1þy2ÞNdy
þeat=3ð1þx2ÞN; 2lþkpm1;
1; 2lþk>m1:
8>
><
>>
:
By Lemma 5.1, we know that
jI1l;kjpCe0ð1þtÞð1þkþ2lÞ=2 BNðx;tÞ; 2lþkpm1;
1; 2lþk>m1:
(
ð5:5Þ
ForI2l;k;we first set
Fk;lðx;tÞ ¼ ð1þtÞnð1þkþ2lÞ ð1þ1þtx2ÞN; 2lþkpm1;
1; 2lþk>m1;
(
and forlp1 and a even integer m;set
MðtÞ ¼ sup
0pspt;xAR;2lþkp2m Fk;lðx;sÞj@tl@xkVðx;sÞj; ð5:6Þ where
nðhÞ ¼1 2
h; hpm;
m2; h¼mþ1;mþ2;
m4; h¼mþ3;mþ4;
^ ^
0; h¼2m1;2m:
8>
>>
>>
><
>>
>>
>>
:
It follows easily from [3,10] that
@lt@xkvðx;% tÞpCe0ð1þtÞðkþ2lÞ=2ebx
2
1þt: ð5:7Þ
Thus,
j@lt@xkF˜1ðx;tÞjpCe0ð1þtÞðkþ2lþ2Þ=2ebx
2
1þt: ð5:8Þ
Since
F˜2¼ ðp0ðvÞ*% vþp00ðv%þyðVxþvÞÞððV* xþvÞ* 2=2Þ;
with 0oyo1;we have
j@tl@kxvðx;* tÞjpCjuþujeat@xkm0ðxÞ:
Noticing thatm0ðxÞis a smooth function with compact support, we have
j@tl@xkvðx;* tÞjpCe0eatex2: ð5:9Þ
By (5.6) and Theorem 2.1, we have forlp1;1þ2lþkp2m j@tl@kx@xðVx2Þj
p
CM2ðtÞð1þtÞ3þkþ2l2 B2Nðx;tÞ; 1þ2lþkpm2;
CM2ðtÞð1þtÞm2BNðx;tÞ; 1þ2lþk
¼m1;m;
CM2ðtÞð1þtÞm22 BNðx;tÞ; 1þ2lþk
¼mþ1;mþ2;
^ ^
CM2ðtÞð1þtÞ1BNðx;tÞ; 1þ2lþk
¼2m3;2m2;
Ce0MðtÞð1þtÞ1BNðx;tÞ; 1þ2lþk
¼2m1;2m:
8>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
><
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
:
ð5:10Þ
Also, (5.7), (5.9) and (5.10) yield
j@tl@xkF2jpCðe0þM2ðtÞÞ
ð1þtÞ3þkþ2l2 B2Nðx;tÞ; 1þ2lþkpm2;
ð1þtÞm2BNðx;tÞ; 1þ2lþk
¼m1;m;
ð1þtÞm22 BNðx;tÞ; 1þ2lþk
¼mþ1;mþ2;
^ ^
ð1þtÞ1BNðx;tÞ; 1þ2lþk
¼2m1;2m:
8>
>>
>>
>>
>>
>>
>>
>>
>>
<
>>
>>
>>
>>
>>
>>
>>
>>
>:
ð5:11Þ
By (4.12), for eachk;we can choose ak%according to (5.10) and (5.11) so that I20;k¼
Z t t=2
Z N
N
ð@kx k˜GK0;kk˜Þðx;y;t;sÞ@yk˜Fðy;sÞdy ds
þ Z t
t=2
Z N
N
K0;kk˜ðx;y;t;sÞ@yk˜Fðy;sÞdy ds
þOð1Þe0
X
hok˜
Z t t=2
Z N
N
ðð1þtsÞð1þkkÞ=2˜ ð1þtÞðkhÞ=2˜