Applied Mathematics
Good solutions of fully nonlinear parabolic equations
?
Tran Duc Van
1and Tran Van Bang
21 Hanoi Institute of Mathematics, P.O. Box 631 Bo Ho, 10.000 Hanoi, Vietnam
e-mail:tdvan@thevinh.ncst.ac.vn
2 Hanoi Pedagogical University, Xuan hoa, Hanoi, Vietnam
Received: April 02, 2002
Summary.
In this paper, we introduce the notion of a "good" so-lution of a fully nonlinear parabolic equation and show that "good" solutions are equivalent to Lp-viscosity solutions of such equations.The results here generalize the ones in 8] about "good" solutions of fully nonlinear elliptic equations. We give here an explicit construc-tion of parabolic equaconstruc-tions withLp-strong solutions that approximate
some nonlinear parabolic equation and itsLp-viscosity solution.
Key words:
Lp-viscosity solutions, good solutions, strongsolu-tions, fully nonlinear parabolic equations
Mathematics Subject Classication (1991): 35J60, 35J65, 49L25
1. Introduction
The notion of viscosity solutions which was introduced in 1981 by M. G. Crandall and P. L. Lions for the general Hamilton { Jacobi equations plays an important role in the theory of partial dieren-tial equations 2]. For convex Hamiltonians, the viscosity solution characterized by a semiconcave stability condition was rst intro-duced by S. N. Kruzkov 9]. There is an enormous activity which is based on these studies. Many fundamental results about viscosity solutions for partial dierential equations of rst and second order have been obtained by M. G. Crandall, L. C. Evans, P. L. Lions,
? This research was supported in part by National Council on Natural Science,
H. Ishii, R. Jensen, P. E. Souganidis, M. Bardi, G. Barles, E. N. Bar-ron, M. Kocan, A. Swiech and so on. There results concern existence and uniqueness theorems, properties of solutions and applications.
We note that a comprehensive treatment of Lp-theory for fully
nonlinear parabolic equations is presented in recent paper by M. G. Crandall, M. Kocan and A. Swiech 5].
Following 8] we introduce the notion of a good solution of a fully nonlinear parabolic equation and also prove some results that are sim-ilar to the ones for good solutions of fully nonlinear elliptic equations. The reader is referred to 1-8] for the modern theory of nonlinear el-liptic and parabolic equations.
In this paper, Rn(n2) is always assumed to be a bounded domain and satisfy an uniform exterior cone condition, Q = (0T], (T >0), S(n) be the set of symmetricnnreal matrices. For
X2S(n) we callX
+,X;respectively the positive, negative parts of
X and also dene kXk= tr(X
+) + tr(X;), where tr(X) is the trace of X.
Let X 2 S(n), 0 < . The Pucci extremal operators are dened as follows P +(X) = ;tr(X +) +tr(X;) P ;(X) = ;tr(X +) +tr(X;): Consider the nonlinear parabolic partial dierential equations of the form
(1:1) ut+G(xtuDuD2u) = 0 in Q where G : QRRnS(n) ! R, Du = (ux
1:::uxn), D 2u = (uxixj) correspond to the spatial gradient and Hessian matrices of u
respectively.
We will always assume that G(xtrpX) is continuous in (rp,
X) with modulus independent of (xt)2Qand is jointly measurable in all variables (xtrpX).
We also require that G of (1.1) satisfy the following structure conditions (1:2) jG(xtrpX);G(xtrqX)j jp;qj (1:3) P ;(X ;Y)G(xtrpX);G(xtrpY)P +(X ;Y) for all (xt)2 Q,r2 R,p, q 2Rn, X,Y 2S(n), and , are positive constants which are xed for all time. Thus (1.3) amounts to (1:4) G(xtrpX)G(xtrpY) whenever XY:
As for the dependence onr, we assume that (1:5) 8 < : r!G(xtrpX) is uniformly continuous, uniformly for (xt)2Q jrj+jpj+kXkR(R >0) (1:6) r!G(xtrpX) is nondecreasing:
We sometimes rewrite (1.1) as follows
(1:7) ut+F(xtuDuD2u) =f(xt) by setting
f(xt) =;G(xt000) F(xtrpX) =G(xtrpX)+f(xt) so that
(1:8) F(xt000)0: Regarding the behavior off, we require that (1:9) f 2Lp(Q) p > p
0
where p0 < n+ 1 is a constant such that the generalized maximum principle for parabolic equations holds forp > p0.
Clearly, F of (1.7) will satisfy versions of (1.2){(1.6) whenever G
of (1.1) does, and vice versa.
The notions of solutions which are used in this work are Lp
-viscosity subsolutions,Lp-viscosity supersolutions,Lp-viscosity
solu-tions,Lp-strong solutions, pointwise a. e. solutions (see 5]).
2. Good solutions
Denition 1.
We say that the functionsG1,G2,:::,Gm,:::satisfystructure conditions uniformly inm if (1.2), (1.3), (1.5) are satised uniformly inmwith the same xed,, and if jGm(xt000)j
g(xt) for someg2Lp(Q).
Denition 2.
We say that u 2 C(Q) is a good solution of (1.1) ifthere existGm satisfying structure conditions uniformly inm and an
Lp-strong solution um of ut+Gm = 0 in Q such that um ! u in
C(Q) and Gm(xtrpX)!G(xtrpX) for a. e. (xt)2 Q and
First, we are going to construct anLp-strong solution of the
Cauchy { Dirichlet problem (2:1) ut;u+G(xtuDuD
2u) = 0 in Q u= on @
pQ
whenG is bounded. This will turn out to be an important construc-tion in proving that anLp-viscosity solution of (1.1) is a good solution
of (1.1) as well.
Theorem 1.
LetG:QRRnS(n)!Rbe measurable,bound-ed and satisfy (1.2), (1.4), and (1.5), 2 C(@pQ). Then
the Cauchy { Dirichlet problem (2.1) has an Lp-strong solution u 2
C(Q)\W 21p loc (Q).
Proof. We are going to solve (2.1) by the xed point method. For any given v2C(Q) we consider the Cauchy { Dirichlet problem
(2:2) ut;u+G(xtv(xt)DuD
2u) = 0 inQ u=on @
pQ:
Since the equation in (2.2) is independent ofu, from 3, Theorem 4.5 and Remark 4.8] it follows that (2.2) has an Lp-viscosity solution
in C(Q) which we denote by Tv. By 5, Propositions 3.2 and 4.1],
Tv is parabolically twice pointwise dierentiable a. e., a pointwise a. e. solution and, therefore, anLp-viscosity solution of the Cauchy {
Dirichlet problem
(2:3) ut;u=;g(xt) inQ u= on@pQ whereg(xt) =G(xtv(xt)D(Tv)(xt)D2(Tv)(xt)) (g
2L 1(Q) since G is bounded). However, it is clear that (2.3) has a unique
Lp-strong solution which must coincide with Tv, and it follows that
Tv 2 W 21p
loc (Q). In particular, Tv is a unique Lp-strong solution of (2.3).
On the other hand, sincegis bounded it follows 3, Proposition 4.6 and Remark 4.8] that if Ris large enough then T is a compact map-ping in the ball with the center at 0 and the radiusRinC(Q). Thus, by Schaefer's xed point theorem,T has a xed point or (2.1) has a
Lp-strong solution and the proof is complete. ut
Theorem 2. (Existence of good solutions)
Let 2C(@pQ), letF be measurable and satisfy (1.2), (1.3), (1.5), (1.6), (1.8), andf 2
Lp(Q). Then the Cauchy { Dirichlet problem
(2:4) ut+F(xtuDuD2u) =f(xt) in Q u= on@
pQ
Proof. Noting that under the same conditions we know that (2.4) also has anLp-viscosity solution. We will use Theorem 1 to construct
good solutions of (2.4). For this we rewrite the dierential equation in (2.4) as follows ut;u+G(xtuDuD 2u) = 0 where G(xtrpX) =tr(X) +F(xtrpX);f(xt): Clearly,Gsatises (1.4).
Without loss of generality we may replace (2.4) by
ut;u+G(xtuDuD
2u) = 0 inQ u= on @
pQ:
For m= 12:::, consider truncating functions
m(r) = max(min(rm);m) and an approximating Cauchy { Dirichlet problem (2:5) ut;u+ m(G(xtuDuD
2u)) = 0 inQ u=on@
pQ:
Since m(G(xtuDuD2u)) satises the condition of Theorem 1 and G does (1.4), by similar argument, we assert that (2.5) has a unique Lp-strong solution um 2 C(Q) \W
21p
loc (Q). The family of equations satises structure conditions uniformly in m, and by 3, Proposition 4.6 and Remark 4.8] the setfumgis precompact inC(Q). Therefore, passing to a subsequence if necessary, we can assume that
um !u in C(Q). The function u2 C(Q) is a desired good solution since the approximations in (2.5) obviously converge toG a. e. and the proof is complete. ut
Next, we consider the Cauchy { Dirichlet problem (2:6) ut+F(xtDuD2u) =f(xt) inQ u= on@
pQ:
We will prove that every Lp-viscosity solution of (2.6) is a good
solution. The fact that good solutions are Lp-viscosity solutions is
obvious from the denition of good solution. First, we remind the
sup-convolution w" and inf-convolutionw" of a function w:Q!R,
" >0, as w"(xt) = sup y2 w(yt); 1 2"jx;yj 2 w"(xt) = infy 2 w(yt) + 12"jx;yj 2 (xt)2Q: Now for" >0, >0, we put
w"= (w"+)
w" = (w"+)
Theorem 3.
LetF be measurable and satisfy (1.2), (1.3), (1.8), letfsatisfy (1.9),2C(@pQ). Then everyLp-viscosity solution of (2.6) is
a good solution, i. e. there are a sequence of operatorsFm independent
of u, satisfying (1.2), (1.3), and (1.8), a sequencefm 2Lp(Q) and a
sequence um 2C(Q)\W 21p
loc (Q) of Lp-strong solutions of
(um)t+Fm(xtDumD2u
m) =fm(xt) in Q
such that
(2:7) um !u in C(Q)
(2:8) Fm(xtpX)!F(xtpX)
for a. e. (xt)2Q and for all(pX)2RnS(n), and
(2:9) fm !f in L
p(Q):
Proof. We will use the method proposed in 8] in order to prove The-orem 3. Indeed we repeat all steps of the proof of TheThe-orem 3.1 in 8] for elliptic equations. Assume thatu is anLp-viscosity solution of
(2.6).
Step 0.
Firstly we x a countable, dense in RnS(n) sequence (piXi)2RnS(n),i= 12:::.Step 1.
Choose a sequence ~fm 2 C(Q)\L1(Q) such that
m =
kf ;f~mkLp (Q)
!0. By 5, Theorem 2.8], the following problem
Wt+P
;(D2W)
; jDWj=f ;f~m inQ W = 0 on@pQ has a uniqueLp-strong solutionW andkWkL1
(Q)
Cm. Letum =
u;W. Hence um is an Lp-viscosity solution of (2:10) (um)t+F(xtDumD2u m)f~m(xt) in Q and (2:11) ku;umkL1 (Q)= kWkL1 (Q) Cm: Similarly, solving Wt+P +(D2W) + jDWj=f ;f~m inQ W = 0 on@pQ and setting um = u;W, we conclude that um is an Lp-viscosity solution of
(um)t+F(xtDumD2u
and
(2:12) ku;umkL1 (Q)
Cm: Moreover,um=u==um on@pQ:
Step 2.
Letm b be a subdomain ofwith a smooth boundary and such that(2:13) if x2nm then d(x@) 1
m:
Let Qm = m (0T]. It follows from the proof of Lemma 1.2 in 6] and 8] that, for every suciently small " > 0, one can choose a suitable =(") such that u+
m = (um)" 2W
211(Q
m) and it is an
Lp-strong solution of a perturbed version of (2.10), namely, for a. e.
(xt)2Qm, (2:14) (u+ m)t+Fm(xtDu+ m(xt)D2u+ m(xt))fm(xt) where (2:15) Fm(xtpX) =F(T+ mxtpX) fm(xt) = ~fm(T+ mxt) withT+ mx=x+"Du+ m.
Similarly, we conclude that u;
m = (um)"2W 211(Q m), and (2:16) (u; m)t+Fm(xtDu; m(xt)D2u; m(xt))f m(xt) for a. e. (xt)2Qm, where (2:17) Fm(xtpX) =F(T; mxtpX) fm(xt) = ~fm(T; mxt) withT; mx=x;"Du ; m.
Also there is m>0 independent of"such that
(2:18) DT
m mI for a. e. x2m whereI is the (nn) identity matrix.
It follows from (2.18) thatFm andFm are measurable. Moreover,
decreasing"if necessary, we can achieve that (2:19) kfm;f~mkLp (Qm) 1 m kf m;f~mkLp (Qm) 1 m and (2:20) Z Qm jF(T mxtpiXi);F(xtpiXi)jdxdt 1 m i= 12:::m:
Further, without loss of generality we may also assume that (2:21) kum;u + mkL1 (Qm) Cm kum;u ; mkL1 (Qm) Cm: Finally, redening u; m = u; m;3Cm and u + m = u+ m+ 3Cm we see
that (2.14) and (2.16) still hold, while by (2.11), (2.12), and (2.21) (2:22) u;5Cmu
;
m u;Cm u+Cmu +
m u+5CmonQm:
Step 3.
We now establish some limiting properties of the approxi-mations constructed in Step 2, which will be needed later in Step 6.From (2.20), for every (piXi) and Q0
bQ, we have Z Q0 jF(T mxtpiXi);F(xtpiXi)jdxdt!0 asm!1: Thus, there exists a subsequence mk and a null setN Qsuch that
for all (xt)2QnN and i
F(T
mkxtpiXi)!F(xtpiXi) ask!1:
Using structure conditions (1.2) and (1.3) we can generalize this con-vergence, namely, for all (xt)2QnN and (pX)2RnS(n) (2:23) F(T
mkxtpX)!F(xtpX) ask!1:
Step 4.
Next choose a constant Mmm such that for a. e. (xt)2Qm (2:24) jFm(xtDu + m(xt)D2u+ m(xt)) +u+ m(xt)jMm jFm(xtDu ; m(xt)D2u; m(xt)) +u; m(xt)jMm: Let Gm(xtpX) =;tr(X)+ Mm(Fm(xtpX)+tr(X)): It follows from (2.14) and (2.24) that
(u+ m)t+Gm(xtDu+ m(xt)D2u+ m(xt))fm(xt) for a. e. (xt)2Qm:Similarly, (2:25) (u; m)t+Gm(xtDu; m(xt)D2u; m(xt))f m(xt) for a. e. (xt)2Qm, where Gm(xtpX) =;tr(X)+ Mm(Fm(xtpX)+tr(X)):
Let:QmR!01] be a continuous function such that (xts) = 0 if u+ m(xt)s 1 if u; m(xt)s: This can be done due to (2.22). Dene
Hm(xtrpX) =(xtr)Gm(xtpX) +(1;(xtr))Gm(xtpX) =;tr(X) +(xtr) Mm(Fm(xtpX)+tr(X)) +(1;(xtr)) Mm(Fm(xtpX)+tr(X)) (2:26) hm(xtr) =(xtr)fm(xt) + (1;(xtr))f m(xt)
and consider the Cauchy { Dirichlet problem (2:27) vt+Hm(xtvDvD2v) = h
m(xtv)inQm v=uon@pQm:
All assumptions of Theorem 1 are satised and thus (2.27) has an
Lp-strong solution vm2C(Qm)\W 21p
loc (Qm). Now we dene
Gm(xtpX) =Hm(xtvm(xt)pX)
gm(xt) =hm(xtvm(xt))
so thatvm solves
(vm)t+Gm(xtDvm(xt)D2v
m(xt)) =gm(xt)
for a. e. (xt)2Qm,vm =u on@pQm. Observe that by construction (2:28) kgmkL1
(Qm)
kf~mkL1 (Q)
jtr(X)+Gm(xtpX)jMm: Moreover, from (2.19) and (2.26) we have
(2:29) kgm;f~mkLp (Qm) kfm;f~mkLp (Qm)+ kf m;f~mkLp (Qm) 2 m:
Next we claim that
(2:30) vmu + m+ 8Cm onQm: To show (2.30) we considerQ+ m=f(xt)2Qm:vm(xt)> u + m(xt)g. Then (xtvm(xt))0 onQ + m. Therefore, gm(xt) =fm(xt) andGm(xtpX) =Gm(xtpX)
for (xt) 2 Q +
m. It follows that vm is an Lp-strong solution of ut+
Gm =fm on Q+
m. Since by (2.25)u;
m is a supersolution of the same
equation, by the minimum principle and (2.22) we have inf Q+ m(u ; m;vm) inf @pQ+ m(u ; m;vm) min( inf @pQ+ mn@pQm (u; m;u + m) inf@pQm(u; m;u));10Cm: Use of (2.22) again yields (2.30).
A symmetric argument shows that
(2:31) vmu
;
m;8Cm onQm: Summarizing (2.30), (2.31), and (2.22) together gives (2:32) u;13Cmvm u+ 13Cm on Qm:
Step 5.
We now extend gm and Gm to the whole Qas followsfm(xt) = gm(xt) for (xt)2Qm 0 for (xt)2QnQm Fm(xtpX) = Gm(xtpX) for (xt)2Qm ;tr(X) for (xt)2QnQm:
Taking into account (2.28), by Theorem 1, the Cauchy { Dirichlet problem
ut+Fm(xtDuD2u) =f
m inQ u=on @pQ
has a unique Lp-strong solution um 2 C(Q)\W 21p
loc (Q). Moreover, it is similar to Remark 4.3 in 3], there exists a modulus of continuity
determined only by the parameters of the cone condition , ,n,
p, ,kfmkLp
(Q), the modulus of continuity of, diam(Q) such that (2:33) jum(xt);(ys)j(jx;yj+jt;sj)
(xt)2Q, (ys)2@pQ.
Step 6.
Denote by u the modulus of continuity of u on Q. From(2.13) and (2.33) we conclude that (2:34) jum(xt);u(xt)j( 1
m) +u( 1m) (xt)2QnQm: Since bothum andvm solve the same equationut+Gm=gm inQm
and vm=u on@pQm, by the maximum principle and (2.34) we have
sup (xt)2Qm jum;vmj sup (xt)2@pQm jum;uj( 1 m) +u( 1m):
By (2.32) this yields sup (xt)2Qm
jum;uj( 1
m) +u( 1m) + 13Cm:
Using (2.34) again we conclude that sup
(xt)2Q
jum;uj( 1
m) +u( 1m) + 13Cm
and therefore (2.7) follows. From (2.29) kfm;fkLp (Q) kfkLp (QnQm)+ kgm;f~mkLp (Qm) +kf;f~mkLp (Q) kfkLp (QnQm)+ 2 m +m
which, together with (2.13), establishes (2.9) along a subsequence. Finally we show that the convergence in (2.8) holds along a sub-sequence mk constructed in Step 3. Since Fm is just a convex
com-bination of truncations ofFmFm dened in (2.15) and (2.17), from
(2.23) it follows easily that for all (xt)2QnN, (pX)2RnS(n),
Fmk(xtpX)!F(xtpX) as k!1 or (2.8) holds and the proof is complete. ut
Acknowledgment.
The rst author thanks Professor A. Swiech for sending him the preprint 8] before its publication in Proceedings of AMS.References
1. Caarelli, L. A. and Cabre, X.(1995):FullyNonlinearEllipticEquations,
Prov-idence, RI: American Mathematical Society.
2. Crandall, M. G. and Lions, P. L. (1983): Viscosity solutions of Hamilton { Jacobi equations, Trans. Amer. Math. Soc.,277, 1{42.
3. Crandall, M. G, Kocan, M., Lions, P. L., and Swiech, A. (1999): Existence re-sults for boundary problems for uniformly elliptic and parabolic fully nonlinear equations, Electronic J. Dierential Equations, 24, 1{20.
4. Crandall, M. G., Ishii, H., and Lions, P. L. (1992): User's guide to viscosity solutions of second order partial dierential equations, Bull. Amer. Math. Soc.,
27, 1{67.
5. Crandall, M. G., Kocan, M., and Swiech, A. (2000):Lp-theory for fully
non-linear uniformly parabolic equations, Commun. in PDEs,25, 1997{2053.
6. Crandall, M. G., Fox, K., Kocan, M., and Swiech, A. (1998): Remarks on nonlinear uniformly parabolic equations, Indi. Univ. Math. Jour., 47, 1293{
7. Evans, L. C. (1998):PartialDierentialEquations, Providence, RI: American
Mathematical Society.
8. Jensen, R., Kocan, M., and Swiech, A. (2002): Good and viscosity solutions of fully nonlinear elliptic equations, Proc. Amer. Math. Soc.,130, 533-542.
9. Kruzkov, S. N. (1964): The Cauchy problem in the large for nonlinear equations and for certain quasilinear systems of the rst order with several variables, Soviet. Math. Dokl.,5, 493{496.