(Universitext) Emmanuele DiBenedetto - Degenerate Parabolic Equations-Springer (1993)
(Universitext) Emmanuele DiBenedetto - Degenerate Parabolic Equations-Springer (1993)
(Universitext) Emmanuele DiBenedetto - Degenerate Parabolic Equations-Springer (1993)
Degenerate
Parabolic
Equations
,(
•
i Springer-Verlag
Emmanuele DiBenedetto
Degenerate
Parabolic
Equations
With 12 Figures
Springer-Verlag
New York Berlin Heidelberg London Paris
Tokyo Hong Kong Barcelona Budapest
Emmanuele DiBenedetto
Northwestern University
USA
and
University of Rome II
Italy
Editorial Board
(North America):
I.H. Ewing F. W. Gebring
Department of Mathematics Department of Mathematics
Indiana University University of Michigan
Bloomington, IN 47405 Ann Arbor, MI 48109
USA USA
P.R. Halmos
Department of Mathematics
Santa Clara University
Santa Clara, CA 95053
USA
AMS Subject Classifications (1991): 35K65
This reprint has been authorized by Springer-Verlag (Berlin/Heidelberg/New York) for sale in the
Mainland China only and not for export therefrom
987654321
(1.3) {
U E wl!:(n), p> 1
div a(x, u, Du) + b(x, u, Du) = 0, in n,
with structure conditions
a(x,u,Du)· Du ~ 'xIDuI P - tp(x), a.e. nT, p> 1
(1.4) { la(x, u, Du)1 :5 AIDul p - 1 + tp(x),
Ib(x, u, Du)1 :5 AIDul p - 1 + tp(x) .
vi Preface
The modulus of ellipticity of(1.5) is IDuI P- 2. Therefore at points where IDul =0,
the p.d.e. is degenerate if p > 2 and it is singular if 1 < p < 2.
By using the methods of DeGiorgi, Ladyzbenskaja and Ural'tzeva [66] es-
tablished that weak solutions of (1.4) are finder continuous, whereas Serrin [92]
and Trudinger [96], following the methods of Moser, proved that non-negative
solutions satisfy a Harnack principle. The generalisation is twofold, i.e., the prin-
cipal part a(x, u, Du) is pennitted to have a non-linear dependence with respect
to UZi , i = 1,2, ... , N, and a non-linear growth with respect to IDul. The latter
is of particular interest since the equation in (1.5) might be either degenerate or
·singular.
where ai; E L 00 (nT ) satisfy the analog of the ellipticity condition (1.2). As before,
it can be used to prove that weak solutions are locally II)lder continuous in T • n
Since the linearity of (2.3) is immaterial to the proof, one might expect, as in the
elliptic case, an extension of these results to quasilinear equations of the type
part has a linear growth with respect to IDul. This appears in the work of Aron-
son and Serrin [7] and Trudinger [97]. The methods of DeGiorgi also could not be
extended. Ladyzenskaja et al. [67] proved that solutions of (2.4) are R>Ider contin-
uous, provided the principal part has exactly a linear growth with respect to IDul.
Analogous results were established by Kruzkov [60,61,62] and by Nash [84] by
entirely different methods.
Thus it appears that unlike the elliptic case, the degeneracy or singularity of
the principal part plays a peculiar role, and for example, for the non-linear equation
{
u == (Ulo U2, ••. , un), Ui E V1,p(nT), i=l, 2, ... n,
(3.2)
Ut - div IDul p - 2 DUi = 0, in aT.
Besides their intrinsic mathematical interest, this kind of system arises from geom-
etry [99], quasiregularmappings [2,17,55,89] and fluid dynamics [5,8,56,57,74,75].
In particular Ladyzenskaja [65] suggests systems of the type of (3.2) as a model
of motion of non-newtonian fluids. In such a case u is the velocity vector. Non-
newtonian here means that the stress tensor at each point of the fluid is not linearly
proportional to the matrix of the space-gradient of the velocity.
The function w = IDul 2 is formally a subsolution of
(3.3)
where
at,k
_ {fJt,k + (P -
=
2)Ui,Zt Ui ,zlo }
IDul 2 •
4. Main results
In these notes we will discuss these issues and present results obtained during
the past five years or so. These results follow, one way or another, from a sin-
gle unifying idea which we call intrinsic rescaling. The diffusion process in (2.5)
evolves in a time scale determined instant by iQstant by the solution. itself, so that,
loosely speaking, it can be regarded as the heat equation in its own intrinsic time-
configuration. A precise description of this fact as well as its effectiveness is linked
to its technical implementations.
We collect in Chap. I notation and standard material to be used as we proceed.
Degenerate or singular p.d.e. of the type of (2.4) are introduced in Chap. II. We
make precise their functional setting and the meaning of solutions and we derive
truncated energy estimates for them. In Chaps. III and VI, we state and prove
theorems regarding the local and global HOlder continuity of weak solutions of
(2.4) both for p > 2 and 1 < p < 2 and discuss some open problems. In the singular
case 1 < p < 2, we introduce in Chap. IV a novel iteration technique quite different
than that of DeGiorgi [33J or Moser [83].
These theorems assume the solutions to be locally or globally bounded. A
theory of boundedness of solutions is developed in Chap. V and it includes equa-
tions with lower order terms exhibiting the Hadamard natural growth condition.
The sup-estimates we prove appear to be dramatically different than those in the
linear theory. Solutions are locally bounded only if they belong to L ,oc ({}T) for
some r ~l satisfying
We show by counterexamples that the Harnack estimate (2.2) cannot hold for non-
negative solutions of (2.5), in the geometry of (4.2). It does hold however in a
time-scale intrinsic to the solution itself. These Harnack inequalities reduce to (2.2)
when p = 2. In the degenerate case p > 2 we establish a global Harnack type
estimate for non-negative solutions of (1.5) in the whole strip ET == RN X (0, T).
We show that such an estimate is equivalent to a growth condition on the solution
as Ixl - 00. If max{l; J~l} < p < 2, a surprising result is that the Harnack
estimate holds in an elliptic form, i.e., holds over a ball Bp at a given time level.
This is in contrast to the behaviour of non-negative solutions of the heat equation
as pointed out by Moser [83] by a counterexample. These Harnack estimates in
either the degenerate or singular case have been established only for non-negative
solutions of the homogeneous equation (2.5). The proofs rely on some sort of non-
linear versions of 'fundamental solutions'. It is natural to ask whether they hold
for quasilinear equations. This is a challenging open problem and parallels the
Hadamard [50] and Pini [86] approach viafundamental solutions, versus the 'non-
linear' approach of Moser [83].
The number p is required to be larger than 2Nj (N + 1) and such a condition
is sharp for a Harnack estimate to hold. The case 1 < P ~ 2Nj (N + 1) is not
fully understood and it seems to suggest questions similar to those of the limiting
Sobolev exponent for elliptic equations (see Brezis [19]) and questions in differen-
tial geometry. Here we only mention that as
of the type of motion by mean curvature.
p'"1, (2.5) tendsformally to a p.d.e.
HOlder and Harnack estimates as well as precise sup-bounds coalesce in the
theory of the Cauchy problem associated with (2.4). This is presented in Chap. XI
for the degenerate case p > 2 and in Chap. XII for the singular case 1 < p < 2. When
p> 2, we identify the optimal growth of the initial datum as Ix I- 00 for a solution,
local or global in time, to exist. This is the analog of the theory of Tychonov [98],
Tacklind [94] and Widder [105] for the heat equation. When 1 < p < 2 it turns out
that any non-negative initial datum U o E Lfoc(RN) yields a unique solution global
in time. In general
2N
I<P$N+l'
Therefore the main difficulty of the theory is to make precise the meaning of solu-
tion. We introduce in Chap. XII a new notion of non-negative weak solutions and
establish the existence and uniqueness of such solutions. We show by a counterex-
ample that these might be discontinuous. Thus, in view of the possible singulari-
ties, the notion of solution is dramatically different than the notion of 'viscosity'
solution. Issues of solutions of variable sign as well as their local and global be-
haviour are open.
In Chaps. VIII-X, we tum to systems of the type (3.2) and prove that
for (4.3) to hold. Near the lateral boundary of ilT we establish C a estimates/or
all a E (0, 1), provided p > max {I; ~~2}' Estimates in the class c1,a near the
boundary are still lacking even in the elliptic case.
These parts are technically linked but they are conceptually independent, in
the sense that they deal with issues that have developed in independent directions.
We have attempted to present them in such a way that they can be approached
independently.
The motivation in writing these notes, beyond the specific degenerate and
singular p.d.e., is to present a body of ideas and techniques that are surprisingly
flexible and adaptable to a variety of parabolic equations bearing, in one way or
another, a degeneracy or singularity.
Acknowledgments
The book is an outgrowth of my notes for the Lipschitz Vorlesungen that I delivered
in the summer of 1990 at the Institut fUr Angewandte Math. of the University of
Bonn, Germany. I would like to thank the Reinische Friedrich Wilhelm Universilit
and the grantees of the Sonderforschungsbereich 256 for their kind hospitality and
support.
I have used preliminary drafts and portions of the manuscript as a basis for lec-
ture series delivered in the Spring of 1989 at 1st. Naz. Alta Matematica, Rome Italy,
in July 1992 at the Summer course of the Universidad Complutense de Madrid
Spain and in the Winter 1992 at the Korean National Univ. Seoul Korea. My thanks
to all the participants for their critical input and to these institutions for their sup-
port.
I like to thank Y.C. Kwong for a critical reading of a good portion of the
manuscript and for valuable suggestions. I have also benefited from the input of
M. Porzio who read carefully the first draft of the first four Chapters, V. Vespri and
Chen Ya-Zhe who have read various portions of the script and my students J. Park
and M. O'Leary for their input.
Contents
Preface
§1. Elliptic equations: Harnack estimates and HiUder continuity ....... v
§2. Parabolic equations: Harnack estimates and Hi>lder continuity . . . . .. vi
§3. Parabolic equations and systems. . . . . . . . . . . . . . . . . . . . . . . . .. vii
§4. Main results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. viii
1. Some notation
Let fl be a bounded domain in RN of boundary afl and for 0 < T < 00 let flT
denote the cylindrical domain flx (0, TJ. Also let,
denote the lateral boundary and the parabolic boundary of flT respectively.
If fl is a sphere of radius P > 0 centered at some Xo E R N , we denote it by
Bp(xo) == {Ix - xol < p}, and if Xo coincides with the origin, we let Bp(O) == Bp.
The boundary afl will be assumed to satisfy the property of positive geomet-
ric density, i.e.,
there exists 0:* E (0,1) and Po> 0 such that 'r/xo E afl,
(1.1) { for every ball Bp(x o) centered at Xo and radius p::; Po
(1.2)
2 I. Notation and function spaces
Here for a smooth function fjJ defined on a compact subset IC of Rft. for some
positive integer n
if
J(/ I/lqdx)
tl IC
i dr < 00.
(2.1)
(2.2)
(2.2-i) q N=I,
qE[S,OO], QE [O'P+S~_I)];
(2.2-ii) if 1 ~ p < N, a E [0, I] and
oE [0, NP+~:-N»).
COROLLARY 2.1. LetvE W;,"(n), and assume pE [I,N). There exists a con-
stant 'Y depending only upon N, and p, such that
Np
(2.1)' where q=--.
N-p
(2.3)
where
(2.4) f
n
v(x)dx = o.
In such a case the constant C depends upon s, p, q, a, N and the structure of an.
However it does not depend on the size of n. i.e., it does not change under dilations
ofn.
Let k be any real number and for a function v E WI,"(n) consider the trun-
cations of v given by
Remark 2.2. The conclusion of the lemma continues to hold for functions v E
WI,I (n)nC( n) provided n is convex. We will use it in the case a is a hemisphere
or a cube.
Remark 2.3. The continuity is not necessary to the conclusion of Lemma 2.2. The
function v has been assumed to be continuous to give an unambiguous meaning
to the definitions (2.6). If v is only in Wl,l (a). one could fix an arbitrary repre-
sentative out of the equivalence class v say v and define (2.6) accordingly. The
conclusion of the Lemma is independent of the choice of v.
{
0 ~ I.{) ~ 1, "Ix E n,
(2.8)
the sets [I.{) > k] are convex, Vk E (0,1).
Let vE WI,p(a), p~ 1, and assume that the set
e == [v = Ojn[l.{) = 1]
has positive measure. There exists a constant C depending only upon Nand p and
independent of v and I.{), such that
(2.9)
6 I. Notation and fwiction spaces
Iz-zl
Iv(x)1 = Iv(x) - v(z)1 = / J :p v(z+j:::jP) dp/
o
Iz-zl
$ J /VV(Z+RP)/ dp.
o
Multiply this inequality by cp(x) and integrate in dx over 0 and in dz over e. This
gives
Iz-zi
(2.10) lei J cplvldx $ J dz J dx J CP(x)/DV(z+j:::jp)/ dp.
n E n 0
cp(x)/Vv(z+j::=j"p)/ $ cpJDvl(z+j:::jP).
We put this inequality in (2.10) and compute the integral in dx on the right hand
side, in polar coordinates with pole at z and radial variable r = Ix - zl. We let w
be the angular variables and denote by 'R.(w) the polar representation of ao with
pole at z.
diam n ) 1l(... )
lei Jcplvldx $ JdZ J rN- 1dr J J IxcpIDul(x) N1
- z1N-1 r - dr
( dw
n E 0 1... 1=1 0
$ -y(diamO)N Je dz nJI:~~11~1 dx
$ -y(diamOt suPJ
zEn
IX - d~N_1
Z
JcplDvldx.
e n
Next, for all x E O. and for all 6> 0,
dz
Ix - zlN-l +
J dz
Ix - z1N-1
En{lz-zl~6}
(2.11) Jcplvl dx
n
$ -y (~;:N JcplVvl dx,
n
3. Parabolic spaces and embeddings 7
J cplvl P dx <
- P'Y
(diamfl)N
1t:11--k
J cplvlp-1lDvl dx
n n
<
- 2
~jcplvlPdx + 1'(P) [(diam~N1P
1t:1 1-
jCPIDvIPdx.
n n
and
Vom,P(flT) == L oo (0, Tj Lm(fl))nLP (0, Tj W~,p(n)),
both equipped with the nonn. v E Vm,P( flT ).
(3.1) jj1v(x,t)lqdxdt
nT
N+m
q=p--.
N
Moreover
(3.2)
The multiplicative inequality (3.l ) and the embedding (3.2) continue to hold for
n
functions v E V m ,,,( T ) such that
J
n
v(x, t)dx = 0, fora.e. tE(O,T),
provided an is piecewise smooth. In such a case the constant "f depends also on
the structure of an.
PROPOSITION 3.2. Assume that an is piecewise smooth. There exists a constant
"f depending only upon N,p, m and the structure of an, such that for every v E
Vm,"(ilT ) ,
N+m
(3.3) q=p--.
N
PROOF OF PROPOSITION 3.1: Assume first that N(p-m)+mp > O. Write the
embedding inequality (2.1) for the function x -+ v(x, t) for a.e. t E (0, T) and for
the choice of the parameters s =m and
N+m
a=?!.; q=p--- ; N(P - m) + mp>O.
q N
Taking the qthpower in the resulting inequality and then integrating over (0, T)
proves (3.1). If N (p - m) + mp ::5 0, we must have p < N. Therefore applying
Corollary 2.1,
T
JJ
nT
Ivl q dx dt = JJ
0 n
Ivl"lvl mN dx dt
~
$ ! (j Ivl~dz (j
T
) IV1mdz)
N
~
$ ([f IDV1Pdzdt) (~~'f [IV(X,tl1mdz) ~
To prove (3.2), we rewrite (3.1) as
3. Parabolic spaces and embeddings 9
which has zero integral average over n for a.e. t E (0, T). By Remark 2.1, x-+
w(x, t) satisfies the embedding inequality (2.1) for a.e. t E (0, T] and with constant
C depending also upon the structure of an.
Proceeding as before, we arrive at (3.2) for w. For a.e. t E (0, T) ,
IIDwllp,nT = IIDvllp,nT' IIw(" t)lIm,n :::; 21Iv(·, t)lIm,n.
Moreover
(I U
1
IIwll •.a. " II-II •. ",. - 1111 hI. I_1 m dx) ;1; <it) •
Therefore
( Inl N(p~ml+mp)
Nm
9 esssup IIv("
O<t<T
t)lIm,n,
and the proposition follows.
We will use the following Corollaries obtained from the previous Propositions
by taking m = p and by applying the HOlder inequality.
COROLLARY 3.1. Let p > 1. There exists a constant 'Y depending only upon N
and p, such that for every v E Vo"( T ). n
(3.4) IIvll:,nT :::; 'Yllvl > olmllvll~p(nT)'
COROLLARY 3.2. Letp> 1. There exists a constant'Y depending only upon N,p
and the structure of an. such that for every v E V P ( nT ).
(3.5) IIvll:,nT :5 'Y (1 + InlT)'lIIh
N Ilvl > 01 mpIIvll~p(nT)'
The next two Propositions hold in the case m = p.
10 I. Notation and function spaces
PROPOSITION 3.3. There exists a constant 'Y depending only upon N and p such
that for every v E VJ'( (h ),
(3.6)
PROOF: Let v E V!(nT ) and let r ~ 1 to be chosen. From (2.1) with s = pit
follows that
Choose ar=p. Then conditions (2.2)-(2.2-iii) imply (3.7)-(3.8), and the Proposi-
tion follows.
The next Proposition holds for functions v E VP( nT ) not necessarily vanish-
ing on the lateral boundary of nT.
PROPOSITION 3.4. There exists a constant 'Y depending only upon N, p, m and
the structure of 8 n, such that for every v E VP ( nT ),
n
which has zero average in for a.e. t E (0, T). Proceeding as in the proof of Propo-
sition 3.3 we arrive at (3.6) for w where 'Y now also depends upon the structure of
8n. From this
3. Parabolic spaces and embeddings II
3-(i). Steklovaverages
°
Let v be a function in L1 (!1T) and for < h < T introduce the Steklov averages
Vh(-,t) defined for all 0< t < T by
Vh
_ *J v(·,'T)d'T,
= {
t+h
t
t E (O,T - h},
0, t>T-hj
vii.
_{* j
= t-h
v(','T)d'T, t E (h,T],
0, t < h.
LEMMA 3.2. LetvE Lq,r(IlT ). Then, as h-O, Vh converges to v in Lq,r(IlT _t:)
for every e E (0, T).lfv E C (0, Tj Lq(f1)). then as h - 0, Vh(-, t) converges to
v(·, t) in Lq(Il)/or every tE (0, T - e), 'VeE (0, T).
A similar statement holds for vii.. The proof of the lemma is straightforward
from the theory of LP spaces.
12 I. Notation and function spaces
4. Auxiliary lemmas
4-(;). Fast geometric convergence
We state and prove two lemmas concerning the geometric convergence of se-
quences of numbers.
LEMMA 4.1. Let {Yn } , n=O, 1,2, ... , be a sequence of positive numbers, sat-
isfying the recursive inequalities
(4.1)
(4.2)
(4.5)
Consider the tetm in braces in the first of (4.3). If Z!+" :5 Yn• such a term is
majorised by 2M~+a. If Z!+" ~ Yn• then the same term can be majorised by
M n+l <
_
2C 1+K.b(1+K.)nMl+
n '
min {K.,a}
(4.6)
(4.7) 2C
Yo ~ ( bl-~
)±
Remark 4.1. The Lemma turns the qualitative infonnation of equiboundedness
of the sequence {Yn } into a quantitative apriori estimate for Yo.
PROOF OF LEMMA 4.3: From (4.6), by Young's inequality
By iteration
C )~ n-l .
Yo ~ enYn + ( e1- a Q ~ (b±e)'.
Choose b± !
e = so that the sum on the right hand side can be majorised with a
series convergent to 2. Letting n -+ 00 proves the Lemma.
(4.8)
(4.9)
PROOF:
= Isa + (1 - s)bI P- 2 Ia - bl 2 ds
o
!
1
!
1
>
-
la _ b12 ! 1
(Isa + (1- s)bl )P
(2 - s)la - bl 2
2 /2
ds
o
1 )P/2
~~ (
/ Isa + (1- s)bl 2 ds
!
1
+ la - bl /
s.
! (Ial - (1 - s)la - bl)p-1ds
Remark 4.3. The reverse inequality is false, in general, with 'Y independent of
a,b.
5. Bibliographical notes
For the theory of Sobolev spaces we refer to the monographs of Adams [I] and
Mazja [76]. The embedding theorems 2.1 and 2.2 are special cases of more gen-
eral embedding theorems. No attempt has been made to state them in the most
general setting and under the best assumptions on the regularity of afJ. For a
proof of Lemma 2.1 we refer to Mazja [76] and Stampacchia [93]. Lemma 2.2
is due to DeGiorgi [33]. Also the statement in Remark 2.2 follows from the proof
in [33]. The parabolic spaces Vm,P(fJT ) and Vom,P(fJT ) are standard in the the-
ory of parabolic partial differential equations and we refer for example to [67,73].
The embedding theorems 3.2 and 3.3 are a modification of similar statements and
proofs in [67]. The lemmas on rapid geometric convergence are stated in [67]; we
have given a different proof. The interpolation inequality of Lemma 4.3 is taken
from Campanato [22,23]. Lemma 4.4 is taken from [27].
II
Weak solutions and local energy
estimates
where q, r ~ 1 satisfy
1 N .
(As) -=r + -:::
pq
= 1- 1\:1.
and
1. Quasilinear degenerate or singular equations 17
(As-i)
(As-ii)
(A5 -iii)
(1.4) rp ~ o.
The local boundedness of the testing functions rp is required to guarantee the con-
vergence of the integral on the right hand side of (1.3).
A function u that is both a local subsolution and a local supersolution of (1.1)
is a local solution.
the variable T E (t, t + h). Dividing by h and recalling the definition of Steklov
averages we obtain
To recover (1.3), fix a subinterval O<tl < t2 <T,choose h so small thatt2+h ::;; T
and in (1.5) take a testing function as in (1.4). Such a choice is admissible, since
the testing functions in (1.5) are independent of the variable T E (t, t + h) but may
be dependent upon t. Integrating over It I ,t2J and letting h -+ 0 with the aid of
Lemma 3.2 of Chap. I gives (1.3).
We integrate in dt over It I , t2J C (0, T) and let first h -+ 0 and then e -+ 0 to obtain
1. Quasilinear degenerate or singular equations 19
t2
a(x,t,u,11)·11 ~ 0,
(1.8)
-f}diV a(x, t, u, 0) ~
U++e
cp dxdr
tl n
t2
- f"fa(x,t,u,O).Dcp~dxdr.
J
tIn
U++e
°
We let e '\. and discard the non - negative contribution of the first integral. The
sum of the last two terms tends to zero since
(1) The monotonicity assumption (1.7) is natural in the existence theory. It permits one
to apply Minty's Lemma [78] to identify the weak limit of the principal part of the p.d.e.
when (1.1) is approximated by a sequence of regularised problems.
20 n. Weak solutions and local energy estimates
(1.9)
t2
=- jja(x,t,u+,O)'D<PdxdT.
tin
One checks that the assumptions of the lemma are verified for example by
equations with principal part
where 1/Jo is bounded, non-negative and .p;,x; E Ll(nT) and the matrix (ai;) is
only measurable and positive definite.
Remark 1.1. The 'regularity' assumption (1.8) is only needed to justify the limit
in (1.9). It can be dispensed with when working with a sequence of approximating
solutions.
(2.2)
In addition the second of (2.1) holds in the sense that u ~ ~ 9 on an in the sense
of the traces of functions in W1,p(n) for a.e. t e (0, T).
A function u that is both a sub-solution and a super-solution of (2.1) is a
solution of the Dirichlet problem.
The formulation can be rephrased in terms of Steldov averages as in the pre-
vious section, namely
for all 0 < t < T - h and for all Y' e w!,p(n)nLOO(n), Y' ~ o.
Moreover the initial datum is taken in the sense of L 2 (n), i.e.,
(2.6)
where du denotes the surface measure on an. We remark that the testing functions
cp vanish, in the sense on the traces, on the boundary of /C and not on the boundary
of n. The variational datum is reflected in the boundary integral on the right hand
side of (2.10). The formulation in terms of Steklov averages is
timates near the lateral boundary ST as well as near t = 0 will be derived in the
next section.
Let K p denote the N-dimensional cube centered at the origin and wedge 2p.
i.e.•
If Xo E RN. we let [xo + Kp] denote the cube of centre Xo and wedge 2p which is
congruent to Kp. i.e.•
Q (6, p) == Kp x {-6, O} ,
and if (xo, to) ERN+!. we let [(xo, to) + Q (6, p)] denote the cylinder with 'ver-
tex' at (xo, to) congruent to Q (6, p). i.e .•
(3.1) (E [0,1], ID(I < 00, and ((x, t) = 0, for x outside [xo + Kp].
Assume that
(3.2)
and construct the truncated functions (u - k)±. We will choose levels k satisfying
1be statement that a constant "'I depends only upon the data means that it can be
determined a priori only in terms of the numbers N, p, q, r, "', the constants Ci , i =
0,1,2, and the norms
PROOF: After a translation we may assume that (x o, to) coincides with the origin
and it will suffice to prove (3.8) for the cube Q (0, p). In the weak formulation (1.5)
take the testing functions
3. Local integral inequalities 25
Therefore integrating by parts and letting h - 0 with the aid of Lemma 3.2 of
Chap. I,
In estimating the remaining parts we first let h _ 0 and then use the structure
conditions (AI)-(As). To simplify the notation, set
tE(-e,O).
Then
- p I j "PI (u - k)::I::(P-IID(ldxdT,
Q'
Finally
Co
(3.11) esssup I(u - k)±1 :5: 6 ==
0 4C '
Q(9,p) 2
:5 ~o !!ID(U -
Qt
k)±(IPdxdr + Do !!
Qt
C{)2X [(u - k)± > 0] dxdr
:5 I (u - k);(P(x, -6)dx
Kp
By HOlder's inequality
Remark 3.2. The proof shows that the number Do in (3.11) has to be chosen small
according to the constant C2 • If in (1.1) b(x, t, u, Du) =0, then Do can be taken to
be infinite and no restriction is imposed on the levels k.
Remark 3.3. If the lower order terms C{)i are all zero, then in (3.8) the last term can
be discarded.
28 II. Weak solutions and local energy estimates
(3.14) sup
to-9<t<to
!
[zo+K,,]
1/12 (Hr, (u - k)±, c) (x, tKP(x) dx
//!
Qt
1£h [1/J2]' (PdxdT = //!
Qt
1/J2(PdxdT
Therefore letting h -+ 0
//!
Qt
1£h [1/J2]' (PdxdT ----+ /
Kpx{t}
1/1 2 (Ht, (1£ - k)±,c) (Pdx
To estimate the remaining terms we let h -+ 0 first and then make use of the
structure conditions (At}-(As).
/ / a(x,T,1£, D1£)·DIPdxdT
Qt
//(1 +
~ 2Co
Qt
1/J) 1/J,2ID1£I P(PdxdT
-2//(1 +
Qt
1/J) 1/Jf2IPo (x, TKPdxdT
/ /a(x,T,1£, D1£)·DIPdxdT
Qt
//(1 +
~ Co
Qt
1/J)1/J,2ID1£I P(PdxdT
-2//(1 +Qt
1/J) 1/J,2 IPo (x, TKPdxdT
+ //'P21/J1/J'(Pdxdr.
Qt
Next we observe that by virtue of (3.3) and the definition (3.12) and (3.15) of 1/J
where we have used the fact that c < 1. Treating the last integral as before proves
the Proposition.
Remark 3.4. If the constant C2 in (A3) is zero, then we may take 6 =00 and there
is no restriction on the levels k. Also if 'Pi =0, i=O, 1,2, then the last term on the
right hand side of (3.14) can be discarded.
Remark 3.5. In any case, whence the constant 60 has been chosen according to
(3.11). the constant "Y on the right hand side of either (3.8) or (3.14), is independent
o/u.lt is only the levels k that might depend upon the solution u via (3.11).
4. Energy estimates near the boundary 31
+"( jj(U-k)~(P-l(tdxdT+"({lIB~p(T)lidT} r
[(x o ,to)+Q(9,p)]nnT 0- 9
32 II. Weak solutions and local energy estimates
(4.7) sup
to-9<t<to
J!li2 (D~, (u - k)±, c) (x, t)(P(x) dx
[zo+Kp)nn
+ ')' JJ !lil!li
u (D~, (u - k)±,c) 1
2 P
- ID(I P dxdT
[(Zo,to )+Q(9,p»)nnT
~
Remark 4.1. If the Neumann data are homogeneous, i.e., .,pi 0, i = 0, 1, and =
'Pi =0, i=O, 1,2, and b(x, t,u,Du) =0. then we may take 6=00 and the levels
k are not restricted.
Remark 4.2. The proof below is local in nature and it shows that (4.6)-(4.7) hold
true for weak solutions that satisfy the Neumann data on a portion of ST. Accord-
ingly, only such a portion is required to be of class CHA. Also. no reference to
initial data is necessary.
PROOF OF PROPOSITION 4.1: Fix (xo, to) e ST, assume that (xo, to) coincides
. with the origin and work with cubes Kp and cylinders Q(fJ,p). Since an is of
class CHA, for sufficently small p, the portion of an within the cube K p. can be
represented in a local system of coordinates as a portion of the hyperplane x N = 0
and KpnnC {XN >O}. Set
and let h -+ O. All the terms in (2.11) are treated as in the proof of Proposition 3.1,
except for the boundary integral. We arrive at
4. Energy estimates near the boundary 33
(4.9) sup
-8<t<O
I (u - k)~(P(x, t)dx + C2° jr f ID(u - k)±(IPdxdT
J
K: Q+(8,p)
~ I
Kp
(U - k)~(P(x, -9)dx
+1' II k)~ID(IPdxdT II
(u - + l' (u - k)~(p-l(tdXdT
Q+(8,p) Q+(8.p)
o
+1' II
-8 Kp
¢(X, T, U)(U - k)±(PdxdT.
IJI
-8K p
.,(x, T, u)(u - k)±C"dXdTI
~ 11(/a:N
p
(,i(x, T, u)(u - k)±(p) dx N) dXdT
~1' II (I~ZN
Q+(8,p)
I(u - k)±(P + 1~IID(u - k)±(I) dxdT
+1' II
Q+(9,p)
(1,z,I(u - k)±(P-1ID(1 + l~ullD(u - k)±I(u - k)±(p) dxdT.
II I~ZNI(u-
Q+(9,p)
k)±(PdxdT ~ Dt II
Q+(9.p)
¢lX[(u-k)± > OjdxdT,
Finally
~ ~o II ID(u-k)±(IPdxdr+'Y(p,60 ) II(u-k)'±ID(IPdxdr
Q+(9.p) Q+(9.p)
Combining these estimates implies that the boundary integral on the right hand
side of (4.9) can be estimated by
We put this in (4.9) and, to conclude the proof, estimate the integral involving the
functions !Pi, i=O, 1,2, and ?/Ii, i=O, I, as in the proof of Proposition 3.1.
The proof of the logarithmic estimate (4.7) near the lateral boundary ST is
similar to the proof of the interior logarithmic estimate (3.14), modulo the modi-
fications indicated above and we omit the details.
integrating over [(xo, to) + Q (8, p)) and letting h -+ O. Such a choice of testing
functions is admissible if for a.e. t E ( to - 8, to),
(4.11) k ~ sup g.
[(zo ,to )+Q(lI,p »)nST
Analogously the functions -(Uh - k)_(P can be taken as testing functions in (2.5)
if
(4.12)
PROPOSITION 4.2. There exist constants "( and 60 that can be determined a pri-
ori only in terms ofthe data and such that for every (xo, to) E ST .for every cylinder
[(x o, to) + Q (8, p)) such that to-8 > 0 andfor every level k satisfying (4.2)for
6 ~ 60 and in addition (4.11)for the functions (u - k)+ and (4.12) for (u - k)_ ,
the following inequalities hold:
(4.13) sup
to-lI<t<to
I(u - k)~(P(x, t)dx + ,,(-1
J
Jr
f ID(u - k)±(IPdxdT
[zo+Kp)nn [(zo,to)+Q(II,p»)nnT
~ !(U-k)~(P(x,to-(})dX+"( !!(u-k)~ID(IPdxdT
[zo+Kp)nn [(zo,to)+Q(II,p»)nnT
~
+ ~ (1+ In Dn LlIBUT) It dT } ·
where the numbers q, r, It satisfy (3.6)-(3.7).
Local considerations as those in §3 apply to the present case. In particular the
Proposition continues to hold for weak solutions that satisfy the Dirichlet data on
a portion of ST.
Consider a weak solution of (1.1) that takes the initial datum U o in the sense that
k!
h
°
Local energy estimates for u near t = are derived by taking in the weak formu-
lation (1.5) testing functions
Also from the definition (3.12) of the function lli(·). it follows that
lli (Dt, (u - k)±, c) =0 whenever (u - k)± = o.
[(x o ,to)+Q(9,p)]nnT 0
Moreover
Remark 4.3. Local considerations apply to this case along the lines of similar
remarks in the previous sections.
38 D. Weak solutions and local energy estimates
Remark 4.4. The constant 'Yon the right hand sides of either (4.13)-(4.14) or
(4.17)-(4.18) is independent ofu. It is only the levels k that might depend upon
the solution u via (3.11). Moreover if "Pi == 0, i = 0, 1, 2, and C 2 = 0, the levels k
are independent of u.
Remark 4.5. We conclude this section by observing that all the energy estimates
as well as logarithmic estimates for (u - k)+ hold true if merely u is a subsolution
of (1.1) and for (u - k)_ if u is a supersolution of (1.1).
We will make a few remarks on the dependence of the constant 'Y in the energy
and logarithmic estimates and on the restrictions to be placed on the levels k.
(A~)
and q, r satisfy (As) and (As-i)-(As-iii). The structure condition (A3) implies
(A3)' The source term '{J2 is required to be more integrable than the corresponding
source tenn in (A3).
Let us consider local weak solutions u of (1.1) with the structure conditions
(Ad, (A2), (A3), (A4), (A~), (As) and (As-i)-(As-iii). We do not require
that u be locally bounded. To derive local energy and logarithmic estimates for
u we proceed as in the proof of Propositions 3.1 and 3.2. The lower order tenns in
(3.10) are now estimated by repeated use of the Young's inequality as follows.
+ 'Y !!(u -
Qt
k)~ max {ID(I" ; ("} dxdr.
We conclude that, for such solutions, inequalities (3.8)-(3.14) hold true for every
level k, with a constant 'Y independent of u, provided the integral
! j(U - k)~ID(I"dxdr
[(x o ,t o )+Q(8.p)]
in (3.8) is replaced by
[(xo,to )+Q(9,p))
in (3.14) is replaced by
6. Bibliographical notes
Consider solutions u of (1.1) or of Chap. II for the case P > 2. The equation is
degenerate since the modulus of ellipticity vanishes when IDul =O. We will prove
that if u E L~(nT), then it is HOlder continuous within its domain of defini-
tion. It will shown in Chap. V that local weak: solutions of such degenerate equa-
tions are indeed locally bounded. To simplify the presentation we will assume that
u E L 00 (nT ). If u is only locally bounded, it will suffice to work: within a fixed
compact subset of nT. In the theorems below, the statement that a constant 'Y de-
pends upon the data means that it can be determined a priori only in terms of the
norm II ull oo,nT , the constants Ci , i=O, 1,2, and the norms liepa, cpr ,cp2I1ti,T;nT
appearing in the structure conditions (A 1 )-( A3)' We let /C denote a compact sub-
set of nT and let p - dist (/C j r) be the intrinsic parabolic distance from /C to the
parabolic boundary of nT, i.e.,
/or every pair o/points (x}, h), (X2' t2) EK-.I/thelowerordertermsb(x, t, u, Du)
satisfy (A~) 0/§5 o/Chap. II. then 'Y and a are independent o/liulloo.nT.
THEOREM 1.2. Let u be a bounded weak solution 0/ the Dirichlet problem (2.1 )
o/Chap. II and let (D) and (Uo ) hold. The boundary an is assumed to satisfy
the propertyo/positive geometric density (1./) o/Chap.l. Then uEC caT)' and
there exists a continuous positive non-decreasing/unction s -. w( s) : R + -. R + •
such that
/orevery pair o/points (Xl, tt), (X2' t2) E nT.lnparticular if the boundary datum
9 is Holder continuous in ST with exponent say a g • and if the initial datum U o is
Holder continuous in a with exponent say a uo ' then u is Holder continuous in aT
aiad there exist constants 'Y> 1 and a E (0, 1) such that
Assume that an is of class Cl,~ and let (N) and (N - i) hold. Then u is Holder
continuous in n x [Eb T],for all E < El < T, and there exist constants "y and Q
such that
for every pair of points (Xl, tl), (X2, t2) E n x [El, T]. The constants 'Y > 1 and
Q depend only upon E, lIulloo,iix(€,T] and the data, including the structure of an
and the norms IItPl, tPrr IIq,r;UT appearing in (N - i). In addition the constant
'Y depends upon the distance (El - E).
If the Neumann data are homogeneous, i.e., if tPo == tPl == 0, and if in addition
the lower order terms b(x, t, u, Du) satisfy (Aa) of§5 of Chap. II, then 'Y and Q
are independentofllulloo,iix(€,T]'
Remark 1.1. The continuity of u can be claimed up to t = 0 provided (Uo ) of
Chap. II holds. Also, if U o is HOlder continuous in n then u is HOlder continuous
innT.
2. Preliminaries
The HOlder continuity of u, either in the interior of nT or at the parabolic boundary,
will be, heuristically, a consequence of the following fact. The function (x, t) -+
u( x, t) can be modified in a set of measure zero to yield a continuous representative
out of the equivalence class u E Vj!:(nT ), if for every (xo, to) E nT there exist
a family of nested and shrinking cylinders [(xo, to) + Q (6n , Pn)] with the same
vertex such that the essential oscillation Wn of u in [(xo, to) + Q (6n, Pn)] tends
to zero as n -+ 00 in a way quantitatively determined by the structure conditions
(AD!-(A6).
The key idea of the proof is to work with cylinders whose dimensions are
suitably rescaled to reflect the degeneracy exhibited by the equation. To make this
precise, fix (xo, to) E nT and construct the cylinder
44 m. ltilder continuity of solutions of degenerate parabolic equations
where E is a small positive number to be determined later. After a translation we
=
may assume that (xo, to) (0, 0). Set
p.+ = esssup u, p.- = ess inf u, W= essosc =p. + -p. -
Q(RP-',2R) Q(RP-',2R) Q(RP-e,2R)
(2.1) 2. = (~)"-2
ao A
where A is a constant to be determined later only in terms of the data. We will
assume that
(2.2) (W)"-2
A > [lE.
This implies the inclusion
(2.3)
PROPOSITION 3.1. There exist constants Eo, " E (0, 1) and C, A > 1, that can
be determined a priori depending only upon the data, satisfying the following.
Construct the sequences
Ro=R, wo=w
andfor n= 1, 2, ... ,
R". = C-nR,
Construct also the family of cylinders
~ __ (Wn ),,-2, n=0,1,2, ....
an A
Thenfor all n=O, 1,2, ...
Q(n+1) c Q(n) and
LEMMA 3.1. There exist constants 'Y > 1 and 0 E (0, 1) that can be determined a
priori only in terms of the data. such that for all the cylinders
We may assume without loss of generality that eo is so small that" 5: C-Eo. Then
Let now 0 < P 5: R be fixed. There exists a non-negative integer n such that
c-(n+1) R 5: P 5: C- n R.
Therefore
o=mm . { ;"2eo} .
0 1
essosc u5:wn.
Q(aop",p)
Statements of HOlder continuity over a compact set now follow by a standard cov-
ering argument.
46 ill. HOlder continuity of solutions of degenerate parabolic equations
Remark 3.1. The proof of Proposition 3.1 will show that indeed it is sufficient to
work with the number w and the cylinder Q (aoRP, R) linked by
This fact is in general not verifiable, for a given box, since its dimensions would
have to be intrinsically defined in terms of the essential oscillation of u within it.
Therefore the role of having introduced the cylinder Q (RP-E, 2R) and hav-
ing assumed (2.2) is that (3.1) holds true for the constructed box Q (aoRP, R). It
will be part of the proof of Pf<)position 3.1 to show that at each step the cylinders
Q(n) and the essential oscillation of u within them satisfy the intrinsic geometry
dictated by (3.1).
To begin the proof, inside Q (aoRP , R) consider subcylinders of smaller size
constructed as follows. The number w being fixed, let So be the smallest positive
integer such that
w
(3.2) -28 0 <6
_ 0,
Figure 3.1
3. The main proposition 47
These are contained inside Q (aoRP, R) if the number A is chosen larger that
280 and if t ranges over
where 'Y is a constant depending only upon the data and K. is defined in (3.5) of
Chap. II. From the defu\ition of din (3.3) it follows that
for some i ~ o.
These levels are admissible since
for some i ~ o.
These are also admissible since
280 = 8~2I1ulloo.nT.
Having chosen So this way. (3.2) is verified when working within any subdomain of
DT • The a priori knowledge of the norm lIulloo.nT is required through the number
So. If the lower order terms b(x, t, u, Du) in (1.1) satisfy (A~) of §5 of Chap. II,
then. as remarked there. the energy and logarithmic inequalities hold true for the
truncated functions (u - k)± with no restriction on the levels k. Thus in such a
case So can be taken to be one and no a priori knowledge of lIulloo.nT is needed.
The numbers A and Ai introduced in (3.5) will be chosen to be larger than
280 • In the proof below we will choose them of the type
and .-
A . - 2;0+h, , i = 0,1,2, ... ,
where hi ~ 0 will be independent of lIulloo.nT. We have just remarked that if the
lower order terms b(x, t, u, Du) satisfy (A~) of§5 of Chap. II, then So can be taken
to be one. We conclude that for equations with such a structure, the numbers Ai
can be determined a priori only in terms of the data and independent of the norm
lIulloo.nT·
4. The first alternative 49
I(X, t) E [(0, f) + Q (dRP, R)] lu(x, t} < IL- + 2~0 15 volQ (dRP, R) I,
then
PROOF: Fix a cylinder for which the assumption of the lemma holds. Up to a
=
translation we may assume that (0, f) (0, O), and we may work within cylinders
Q(dPP,p) , O<p~R. Let
R R
Rn ="2 + 2n+1' n = 0,1,2, ... ,
construct the family of nested cylinders Q (dR~, Rn) and let (n be a piecewise
smooth cutoff function in Q (dR~, Rn) such that
(4.2)
We will use the energy inequalities of Proposition 3.1 of Chap II. written over the
cylinders Q (d~, Rn), for the functions (u - kn )-. where forn=O, 1,2, ... ,
k = -
n IL +~ W
280+1 + 280+1+n '
In this setting. (3.8) of Chap. II takes the form
57: {JJ(U-kn)~dXd1'+~
Q(dR~,Rn)
!!(U-kn):dXd1'}
Q(dR~,Rn)
,,(1+,,)
+7 { J
-dR~
IAkn,Rn (1')1 i d1'} r
Ilx[(U-knL >O]dxdr-+O.
Q(dR!:.Rn)
and estimate above the first two tenns on the right hand side of (4.3) by
7 ~: (2: )2 (2:
0 0 r- 2
II X[(u - knL > 0] dxdr
Q(d~.Rn)
Q(dR~,R,,)
where we have used the defmition (3.3) of d. Combining these remarks in (4.3)
and dividing through by d, we arrive at
( W ),,-2 d,.
~ 1 I 0 }~
~dR~IAkn.R,,(r)l·dr
1:
{
+7 2 Bo
Since (11 - knL <n vanishes on the lateral boundary of Qn, by Corollary 3.1 of
Chap. I we have
IA 11+~
(4.6) IAn+d ~ -y4np nRP
Using also the fact that. by virtue of Remark 3.2 and (3.5)
RNIC (.!!!...)
2 80
-2 dP(1:IC) <1
- ,
Therefore
From Lemma 4.2 of Chap. I it follows that Yn and Zn tend to zero as n -+ 00,
provided
Suppose the assumptions of Lemma 4.1 are verified for some box of the type
[(0, l) + Q (dRP, R)]. We will exploit the fact that at the time level
-8 =t - (2BoW)2-P (R)P
"2'
the functionx-+u(x, -8) is strictly above the level ",- + 2.::'+1' in the cube K R / 2 •
To simplify the symbolism let us set p = ~ and construct the cylinder
The next lemma asserts that, owing to (4.1), the set where u(·, t) is close to I.e ,
within the smaller cube KR/4' can be made arbitrarily small for all time levels
-8~t~0.
S. The fllSt alternative continued S3
LEMMA 5.1. For every number Vi E (0,1), there exists a positive integer 81.
depending only upon the data and independent of w, R, such that
'lit E (-9,0).
PROOF: Consider the logarithmic estimates of Proposition 3.2 of Chap. II, written
over the cylinder Q (0, p), for
(u-k)_,
As a number c in the definition of W, we take
W
c-~"':"":'"":"
- 2s o+ i + n '
n> 1,
where n is to be chosen. Thus we take
where
Hi; = esssup
Q(fJ.p)
(u - (JL- + 2 W+ i ) )
So _
::; 2 w+ i
80.
.
For t= -0, by virtue of(4.1) we have (u - (JL- + 2.~+dL =0, and therefore
W(x, -0) = 0,
These remarks in (3.14) of Chap. II yield
(5.2) Iw
Kp
2(x, t)(P(x)dx ::; ; II W!Wu !2- Pdxdr
Q(fJ.p)
n(lIA',p(T) dT) · ,
.ell±!!l
W::;ln( ~ ) =nln2
~
and
!WU !2-P = !Hk_- (u - k)_
W n !P-2 ::; (W
+ 2so +l+ 2 80
)P-2
•
Therefore, in also view of (5.1), the first term on the right hand side of (5.2) is
estimated above by
54 m. Itilder continuity of solutions of degenerate parabolic equations
where l' and A are constants depending only upon the data. and A has to be deter-
mined later. The second tenn is estimated by using the conditions (3.6). (3.7) of
Chap. II. linking the parameters r, q, K.. This gives
n
Ei!±cl
where we have used the fact that (== 1 on K p/2. The integral in (5.3) is estimated
below by extending the integration to the smaller set
On such a set
and since the right hand side of this inequality is a decreasing function of Hi; • we
have
1[12 ~ In2( 20:+1 ) = (n _ 1)21n2 2.
2°o+ n
Putting this into (5.3) gives that for all t E (-0,0)
The infonnation in Lemma 5.1 will imply that u is strictly bounded away from p.-
in a smaller cylinder. To make this precise. consider the box
n = 0,1,2, ... ,
construct the cylinders Q ((J, Pn) and let x -+ (n (x) be a piecewise smooth cutoff
function in KPn that equals one on KPn+1 and such that ID(n 1:$ 2n+3 / p. Write the
inequalities (3.8) of Chap. II over Q (0, Pn) for the functions (u - kn )_. where
w w
kn = p.- + 2 81 + 1 + 2s l+ Hn '
(6.1) sup
-8<t<0
j(u - kn)~ '~dx + ,),-1
JJ
r riD (u - knL (niP dxdr
KPn Q(8,Pn)
£lli!!.!.
Q(8,Pn) -8
The first tenn on the left hand side is estimated below. for all t E (-0,0). by
if 81 is chosen so large as to satisfy the conclusion of Lemma 5.1 and the inequality
2- P (281 I A)p-2 ~ 1. We put this in (6.1), divide through by BI pP and introduce
in the cylinders Q (B, Pn), the change of variable z = tPP lB. This maps Q (B, Pn)
into the boxes
Qn == KPn x(-PP,O).
Let us also set v(x, z) = u(x, zBI PP) and
o
IAnl = jIAn(z)ldZ.
-pP
I
Divide through by the coefficient of An+ 11 and set
we arrive at
Yn+1 :5 -y4np {y~+m:;; + Yn;vT,; Z~+/( } .
Proceeding as in the proof of Lemma 4.1, we have
Zn+1 :5 -y4 np {Yn + Z~+/(}.
By Lemma 4.2 of Chap. I it follows that Yn and Zn tend to zero as n -+ 00, provided
(6.2)
Remark 6.1. Let us trace the dependence of "10 and Al upon lIulloo,nT. The num-
bers 110 and III depend only upon the data and are independent of u. The number
81 is given by 81 = 8 0 + n where n is chosen from (5.4). Thus
depends upon lIulloo,nT via 8 0 through (3.2). Also A and Al are of the type Ai =
2so +h ,. where hi, i = 0, 1,2, ... , can be determined a priori only in terms of the
data and are independent of II u II OO,nT. We conclude that if the lower order terms
b(x, t, u, Du) satisfy the structure condition (Aa) of§5 of Chap. II. we have 8 0 = 1
and therefore "10' A, Ai can be determined a priori only in terms of the data and
are independent of lIulloo,nT.
we rewrite this as
i ~ 0.
LEMMA 7.1. Let [(0, t) + Q (dRP, R)] cQ (aom', R) befixedand let (7.1) hold.
There exists a time level
t· E [f - dRP , f - 110 dRP]
2 '
such that
7. Tho second alternative 59
~ f
t-dRP
Ix E KR Iu(x, r) > p.+ - 2~o Idr
> (l-vo ) IQ(dR"',R) I,
contradicting (7.1).
The lemma asserts that at some time level t* the set where u is close to its
supremum occupies only a portion of the cube K R. The next lemma claims that this
indeed occurs for all time levels near the top of the cylinder [(0, t) + Q (dJlP, R)].
LEMMA 7.2. There exists a positive integer 82 > 80 such that
Ix E I
KR u(x,t) > p.+ - 2~21 :5 (1- (~f) IKRI,
fora/ltE [i - ~dJlP,fj.
PROOF: Consider the logarithmic inequalities (3.14) of Chap. II written over the
box K R X (t* , t) for the function (u - k) + for the levels k =p. + - 2":0 . As for the
number c in the definition of the function 1[1, we take
c- -2-
w
-
Bo+n'
n °
> to be chosen.
Thus we take
(7.2)
where
Ht == ess sup (u - ( p.+ - -W ) ) .
[(O,f)+Q(dRP ,R)] 280 +
The cutoff function x-+(x) is taken so that ( = 1 in the cube K(l-O')R, uE (0, I),
and ID(I:5 (UR)-l. With these choices, inequality (3.14) of Chap. II yields for all
tE(t*,t)
60 m. HiUder continuity of solutions of degenerate parabolic equations
t
(7.3) I
K(l-,,)R
p 2(X,t)dx:5 j llf2(x,to<)dx+ «(1~)P j j llf lllful2- PdxdT
KR tOKR
E1.!±.!!.l
Next, from (7.2) it follows that llf vanishes on the set [u<JL+ - 2":0]. Therefore,
using Lemma 7.1, the first integral on the right hand side of (7.3) is estimated
above by
since f - to< :5dRP, and d is given by (3.3). Finally for the last tenn, we have
E1.!±.!!.l
where A2 = 2(B o +n)b and b is defmed in (3.4). By Remark 3.2 and (3.5) we may
assume that nA2W-b RNIt $1. Combining these remarks in (7.3) we conclu(,ie that
for all tE (to<, f)
(7.4) j 22
llf2(x, t) dx :5 n 1n 2 (11_-II:i2) IKRI + ;P nIKRI·
K(1-,,)R
The left hand side of (7.4) is estimated below by integrating over the smaller set
On such a set, since the function llf in (7.2) is a decreasing function of Ht, we
estimate
7. The second alternative 61
for all t E (t* , f). Choose u so small that u N ~ ~ v~ and then n so large that
where ao and d are defined in (2.1) and (3.3) respectively. If the number A is
chosen sufficiently large, we deduce
62 lli. HOlder continuity of solutions of degenerate panboIk equations
From now on we will focus on the cylinder Q (If R", R) and to simplify the
symbolism we set
Remark 8.1. Assume for the moment that the number B. has been chosen. Then
we detennine the length of the cylinder Q (aoR", R) by choosing
(8.1)
PROOF OF LEMMA 8.1: Consider the local eneqy estimates (3.8) of Chap. II
written over the box Q (aoRP, 2R), for the functions (1£ - k)+. The levels k are
given by
w
k=,,+--8
r- 2 '
where 82 :::; 8 :::; B. and B. is to be chosen. We take a cutoff function ( that equals
one on Q (!f RP, R), vanishes on the parabolic boundary of Q (ooR", 2R) and
such that
Neglecting the firsttenn on the left hand side of these energy estimates, and using
the indicated choices, we obtain
8. The second alternative continued 63
Next by virtue of the choice (8.1) of the parameter A, and the defmition (2.1) of
ao ,
(iii) ~ {j~At.2R(T)I\dT} · .
~ ;" (;r IQ (~ R",R) I(A 3w- b RNK )
~;" (;rIQ(~RP,R) I,
where A3 = 2b8 • and b is defined in (3.4). These estimates in (8.2) give
Next we use Lemma 2.2 of Chap. I applied to the function u(·, t) for all times
-!f RP ~ t ~ 0, and for the levels
w
(I - k) = 2 +1 .
8
w
2S+1IAs+1(t)1 :5 v~ IKRI
4-yR N H JIDul dx,
A.(t)\AO+l (t)
for all t E ( - ~ flP , 0). From this, integrating over such a time interval we get
,; ~R ([fIDU1PdxdT) 'IA,\A.+d7'.
Take the ;tr power, estimate the integral on the right hand side by (8.3) and divide
through by (2;+1 ) ~ . This gives
LEMMA 9.1. The number II. (and hence s. and A) can be chosen so that
PROOF: We will apply the local energy estimates of Proposition 3.1 of Chap. II
over the boxes Q (a. R~, Rn) to the function (u - kn ) +, where for all n =
0,1,2, ... ,
and
+~{j~At..''<T)lldT} ·
First by the definition of a.
Next. using again the definition of a., the first two terms on the right hand side of
(9.1) are estimated above by
66 m. HOlder continuity of solutions of degenerate parabolic equations
v(·,z) = u(·,aooz),
and
o
IAnl =
-R~
J IAn(z)ldz,
2-(n+2)p (~)P
28 •
IAn+l I
= (kn+l - knt I(x, z) E Qn+! I v(x, z) > kn+ll
:5 II (v - k n )+ lI~n+l
:5 II (v - kn )+ (nll~n
:5 'YIAnl ~ II (v - k n)+ (nllt"(Qn)
<
-
'"I (~)P
28 •
np
2Rp IA n 11+~
Ei!.±!!l
(9.3)
68 m. UUder continuity of solutions of degenerate parabolic equations
or
The two alternatives just discussed can be combined to prove the main Proposition
3.1. Let us recall that
~ = (~)P-2.
ao A
The concluding statement of the first alternative is that, starting from the cylinder
Q(d(ft,f),
the essential oscillation w decreases by a factor 110 E (0,1), unless w :5 A1R~,
where A 1 is a large constant that can be computed a priori only in tenns of the
data and the number So is introduced in (3.2). Analogously, the conclusion of the
second alternative is that starting from the same cylinder and going down to the
smaller box
Q(!f(ft.f) ,
the number w decreases by a factor 111 E (0, 1), unless w:5 A2R~ , where A2 is a
constant that can be computed a priori in tenns of the data. We combine these two
facts into
LEMMA 10.1. There exist constants
that can be determined a priori only in terms of the data, such that either
w:5.AR~ or
11. Regularity up to t = 0 69
Next we will construct a box for which information of the type of (l0.1) can be
derived. Set
and let us estimate from below the length of the cylinder Q ( d ( i)P ,i) for which
the conclusion of Lemma 10.1 holds. We have
where
and
It follows that, for the cylinder Q (aIRf, R I ), inequality (l0.1) is verified and the
process can now be repeated starting from such a box, thereby proving Proposition
3.1.
As indicated in §3 this implies the interior mlder continuity stated in Theo-
rem 1.1. The constant dependence indicated in the statement of the theorem follows
from the arguments of §3-( I) and Remarks 6.1 and 9.1.
11. Regularity up to t = 0
p.+ = esssup
Qo(RP-~ ,2R)
u, p.- = essinf
Qo(RP-< ,2R)
u, w = essosc
Qo(RP-~ ,2R)
u.
Let So be the smallest positive integer satisfying (3.2) and construct the box
For all R> 0, these boxes are lying on the bottom of ilT.
PROPOSITION 11.1. There exist constants co, ij E (0, 1) and C, A > 1 that can
be determined a priori depending only upon the data. satisfying the following.
Construct the sequences Ro = R, Wo =W and
and there is nothing to prove. Let us assume. for example. that the second of (11.1)
is violated. Then for all 8:::: 8 0 • the levels
_ w
k = Jl. +2 8 '
satisfy the second of (4.16) of Chap. II. Therefore we may derive energy and log-
arithmic estimates for the truncated functions (u - k) _. These take the form
(11.2) sup
O<t<dRP
}(u - k):(x, t}("dx + JfJf ID(u - k)-(rdxdT
- - KR Qo(dRP,R)
.ei.!±.cl
$1' II(U-k)"-'D('''dxdT+1'{jiBk.R(T)'~dT} ,. ,
Qo(dRP,R) 0
Qo(dRP,R)
Changing the sign of (11.5) and adding ess sUPQo( d( ft. f) u to the left hand side
and Jl.+ to the right hand side we obtain
72 ill. ltilder continuity of solutions of degenerate parabolic equations
1
ij = 1 - 2to +1 .
If the frrst of (11.1) is violated, we write the energy and the logarithmic inequalities
for (u - k)+, k=JI.+ - ~ for s~so and proceed as before.
To summarise, going down from Qo (RP-E, 2R) to the smaller box
W <
-
2essoscuo
KR
or
LEMMA 11.1. There exist constants Ao > 1 and ijE (0,1), that can be computed
a priori only in terms of the data, such that either
or
Q(dRP,R) ,
(12.1)
are both true, subtracting the second from the first gives
w ~ 2 osc g,
Q(dRP,R)nST
satisfy (4.11) of Chap. II. and we may derive energy estimates for (u - k) +. Since
(u - k)+ vanishes on Q (dJtP, R)nST. wemayextendittothe wholeQ (dJtP, R)
by setting it to be zero outside nT within the box Q (dRP, R). Also. in (4.13)
of Chap. II we take a cutoff function vanishing on the parabolic boundary of
Q (dRP, R). Taking into account these remarks. we obtain the energy estimates
"'{1.IBt.R(T)11dT} . ,
where Bt,R(r) is defined in (4.4) of Chap. II.
We observe that the conclusion of Lemma 7.2 is automatically verified for
(u - k)+. Indeed the function x -+ (u(x, t) - k)+, vanishes outside nnKR. for
all t E (-dJtP, 0) and an satisfies the property of positive geometric density of
Chap. I. Therefore we may use Lemma 8.1 and its proof to deduce that for all
el E (0,1). there exist positive numbers Al and £1 that can be detennined a priori
only in tenns of the data such that either w < Al RlYf' or
LEMMA 12.1. There exist numbers Al > 1 and ijE (0,1) that can be computed
a priori only in terms of the data such that either
8 .
8T UA - dlvaA (x, t, UA, DUA) = bA (x, t, UA, DUA)
uniformly in A. Then
LEMMA 14.1. {UA} is a family of uniformly Holder continuous functions over
compact subsets of nT.
Results of this kind are referred to in the literature as continuous dependence
of the solution on the operator. Stability results also hold for a family of equations
where also the parameter p ranges over a compact subset of [2, 00).
° °
for given positive constants < 1'1 $ 1'2 and $ f32 $ /31. This behaviour has to
hold only near the degeneracy, i.e., for s near zero. For s > lTo it will suffice that
~(s) be bounded above and below by given positive constants, i.e., for example,
We require that
Let F(.) denote a primitive of ~;6 (.). Then the p.d.e. can be interpreted weakly
by requiring that
F(u) E Lfoc (O,T;W,!::(il»).
If~(s) = 1, 'r:Is >0, then (1.1) is of the p-Iaplacian type. Ifp=2 and ~(s) =sm-1
for some m > 0, then 0.1) exhibits a degeneracy (m > 1), or singularity (0 < m < 1)
of the type of porous medium equation. In the latter case a weak. solution is required
to satisfy
lul m E L~oc (o,TjW,!:(ilT»).
The mlder continuity of solutions of such doubly degenerate equations can be
proved by methods similar to the ones presented here and has been established
independently by Porzio-Vespri [88] and Ivanov [52]. The technique is also flex-
ible enough to handle equations bearing a power-like degeneracy at two values of
the solutions. These arise in the flow of immiscible fluids in a porous medium and
have as a prototype
Ut = Llu(1 - u) = 0, O:5u:51.
Results on continuous dependence appear in [9] in a different context.
IV
Holder continuity of solutions of
singular parabolic equations
THEOREM 1.1. Let 11. be a bounded local weak solution of (1.1) of Chap. 11 and
let (At}-(A5) hold. Then 11. is locally HOlder continuous in nT • and there exists
constants 'Y> 1 and a E (0,1) depending only upon the data. such that VK. c nT.
78 IV. HBlder continuity of solutions of singular parabolic equations
/oreverypairo/points (Xl, td, (X2, t2) E IC. I/the lower order terms b(x, t, u, Du)
satisfy (A~) 0/§5 o/Chap.lI, then 'Y and a are independento/llulloo.or .
THEOREM 1.2. Let u be a bounded weak solution o/the Dirichlet problem (2.1)
o/Chap. II and let (D) and (Uo ) hold. Assume also that the boundary an has
the property 0/ positive geometric density (1.1) 0/ Chap. I. Then u E C (liT) and
there exists a continuous non-decreasingfunction s -+ w(s) : R+ -+ R+, such
that w(O);:O and
THEOREM 1.3. Let u be a bounded weak solution 0/ the Neumann problem (2.7)
o/Chap.1I and let (N) and (N - i) hold. Assume that the boundary an is 0/ class
CI.~. Then u is Holder continuous in nT and there exist constants 'Y and a such
that
The constants 'Y > 1 antfo: only depend upon lIulloo.i'iT and the data, including
the structure ofafl and the norms IItP1, tPf=r IIq.r;,aT appearing in the assumptions
(N) - i.
If the Neumann data are homogeneous. i.e., iftPo=tP1 =0, and ifin addition
the lower order terms b(x, t, u, Du) satisfy (Aa) of§5 of Chap. II. then 'Y and 0:
are independent ofllulloo.i'iT.
The last two Theorems have been stated in a global way. The proof however uses
only local arguments so that they could be stated within any compact portion, say
/(, of fl. Accordingly, the hypotheses on the boundary data need only to hold within
/(,. For example, in the case of Dirichlet data, the boundary datum 9 could be con-
tinuous or HOlder continuous only on a open portion of ST (open in the relative
topology of ST), say E. Then the solution u of the Dirichlet problem would be
continuous (respectively HOlder continuous) up to every compact subset of E.
Analogous considerations can be made for Neumann data satisfying (N)-(N-i) on
relatively open portions of ST.
Similar remarks hold if U o is only locally continuous or locally HOlder con-
tinuous. In particular, to establish the continuity (lli)lder continuity respectively)
of u up to fl x {O}, no reference is needed to any boundary data on ST.
Finally we comment on the assumption that u be locally bounded. It will be
shown in the next Chapter, that when p > 2, solutions of (1.1) are locally bounded.
This is no longer true, in general, if 1 <p< 2. A weak solutions of u of (1.1) is in
L~(flT)' only if
1'+ =Q(RP,RI-C)
esssup u, 1'- = ess inf u,
Q(RP,Rl-<)
W= essosc u==J.L+ - 1'-.
Q(RP,Rl-<)
(2.2)
then we have
Cylinders of the type of (2.1) have the space variables stretched by a factor
(w / A) ~ , which is intrisically detennined by the solution. If p 2 these are the
=
standard parabolic cylinders with the natural homogeneity of the space and time
variables.
PROPOSITION 2.1. There exist constants eo," E (0,1) and C,A,A > 1, that
can be determined a priori depending only upon the data, satisfying the following.
Construct the sequences Ro = R, WO =w
Rn = c-nR, Wn+1 = max{1JWn;.A~t}, n=I,2, ... ,
and the boxes
LEMMA 2.1. There exist constants 'Y> 1 and aE (0, 1) that can be determined a
priori only in terms of the data, such that for all the cylinders
0< p ~ R,
essosc
Q(pP,cop)
U~'Y(w+JtEO)(RP)Q.
This is the analog of Lemma 3.1 of Chap. III. The proof is the same and it
implies the HOlder continuity of u over compact subsets of {}T via a covering
argument.
Remark 2.1. The proof of Proposition 2.1 will show that it would suffice to work
with the number w and the cylinder Q (IV', CoR) linked by
This fact. is in general not verifiable for a given box since its dimensions would
have to be intrinsically dermed in terms of the essential oscillation of u within it.
The reason for introducing the cylinder Q (RP, R1-E) and assuming (2.2) is
that (2.3) holds true for the constructed box Q (RP, coR). It will be part of the proof
of Proposition 2.1 to show that at each step the cylinders Q(n) and the essential
oscillation of u within them satisfy the intrinsic geometry dictated by (2.3).
Remark 2.2. Such a geometry is not the only possible. For example, one could in-
troduce a scaling with different parameters in the space and time variables. Exam-
ples of such mixed scalings will occur along the proof of Proposition 2.1. Here we
mention that the proof could be structured by introducing the boxes Q (RP-E , 2R)
and Q (aoIV', R) formally identical to those of §2 of Chap. III and rephrasing the
Proposition 2.1 in terms of such a geometry.
3. Preliminaries
E.::l
(3.2) [(ii,O) + Q (RP, doR)1 , do = (~)
2
80
P
82 IV. ltiider continuity of solutions of singular parabolic equations
.-
Figure 3.1
These are contained inside Q (RP , coR) if the number A is larger that 260 and if x
ranges over the cube K'R.(w). where
(3.3)
!=J!
= Lo (doR) • where Lo == (:'0)" - 1.
One may view these as boxes moving inside Q (Rf', CoR) as the coordinates x of
dleir vertices range over the cube K'R.(w)' The cylinders [(x. 0) + Q (Rf' , doR) I
can also be viewed as the blocks of a partition of Q (Rf', coR). Indeed we may ar-
range that Lo be an integer and view the cube KeoR as the union, up to a set of mea-
sure zero, of L: disjoint cubes each congruent to KdoR. Analogously Q (Jl1', CoR)
is the disjoint union, up to a set of measure zero of L:
open boxes each congruent
to Q (Jl1', doR). The proof of Theorem 1.1. is based on studying the following two
cases. Let 110 be a small positive number. Then either
there exists a cylinder of the type of [(x, 0) + Q (Jl1', doR)]. making up the parti-
lion of Q (Jl1' , coR), such that
for all cylinders [(x,O) + Q (RP, doR)] making up the partition of Q (Jl1', CoR),
3. Preliminaries 83
(3.5) meas { (x, t) E [(x,O) + Q (R", doR)] 1u(x, t) < JJ- + 2~0 }
~ volQ (R", doR) I·
In either case the conclusion is that the oscillation of u in a smaller cylinder
with vertex at the origin, decreases in a way that can be quantitatively measured.
In the arguments to follow we assume (2.2) holds. Indeed if not,
pE
eo = (2 _ p)
and the fll'St iterative step of Proposition 2.1 would be trivial.
Remark 3.1. Along the proof we will encounter quantities of the type
i=I,2, ... ,IEN,
where Ai are constants that can be detenoined a priori only in tenos of the data
and
for some i ~ O.
84 IV. Holder continuity of solutions of singular parabolic equations
The a priori knowledge of the nonn 1I1.£lloo,nT is required through the number So. If
the lower order tenns b(x, t, 1.£, D1.£) in (1.3) satisfy (A3) of§5 of Chap. II. then, as
remarked there, the energy and logarithmic inequalities hold true for the truncated
functions (1.£ - k)± with no restriction on the levels k. Thus in such a case. So can
be taken to be one and no a priori knowledge of 1I1.£lloo,nT is needed.
The numbers A and Ai introduced in (3.7) will be chosen to be larger than
280 • In the proof below we will choose them of the type
i=1,2, ... ,
where fi ~ 0 will be independent of 1I1.£lloo,nT' We have just remarked that if the
lower order tenns b(x, t, 1.£, D1.£) satisfy (A3) of §5 of Chap. II. then So can be
taken to be one. We conclude that for equations with such a structure the numbers
Ai can be detennined a priori only in tenns of the data and independent of the
nonn 1I1.£lloo,nT'
4. Rescaled iterations
The following rescaled iteration technique applies to any subcylinder of fh and
it is crucial in both alternatives. Let m > 0 be given by
LEMMA 4.1. There exists a number Vo that can be determined a priori only in
terms o/the data and independent o/w, R and mt, m2 such that:
(I).I/u is a super-solution 0/(13) in [(x, l) + QR {m}, m2)] satisfying
essosc u<w
[(f,l)H2R(mlom 2)] -
and
I
meas {(x, t) E [(x, l) + QR (mll m2)] u{x, t) < J.t- + 2: }
:::; vo IQR{mllm2) I,
then either
or _ w
u{x, t) ~ J.t + 2m+1 '
where bo is defined in (3.6) and Ao is a constant depending only upon the data
and the numbers m}, m2. Analogously
(ll).I/u is a sub-solution 0/(13) in [(x, t) + QR (m}, m2)] satisfying
essosc u <w
[(f,l)+QR(ml,m2)] -
and
I
meas { (x, t) E [(x, l) + QR (mb m2)] u(x, t) > J.t+ - 2: }
:::; volQR (mt, m2) I,
then either
or
u(x, t) :::; J.t+ - 2:::+ 1'
w w
kn = J.t- + 2m+! + 2m+}+n' n=O, 1,2, ... ,
Consider (3.8) of Chap. n, written over the boxes Qn for (u - kn) _ and with the
choice of the cutoff functions Cn
0 < Cn(X,t):5 1, V(x,t) E Qn, andCn:l in Qn+1i
{ Cn = 0 on the parabolic boundary of Qni
-2(,,-2)"'2 Rt
Since
sup(u _ k) < ~-:w_
n - - m m 2 l+ 2'
Q"
the first two tenns on the right hand side of (4.1) are estimated above by
!
Qn
!ID (u - knL (niP dxdr ~ !JID
Qn
(u - kn) _ (niPdxdr
y= C:J ~
p x,
which maps Qn into
Setting also
v(y,z)=u (d1y,2(p-2)m2t) I (y,z)=«d 1y,2(p-2)m2t) ,
and
o
An(z)={y E KRnlv(y,z) < kn }, IAnl=! IAn(z)ldz,
-R~
we arrive at
Thus setting
where 80=min{~;K}.
Remark 4.2. The proof shows that the number Vo depends upon p but it is 'stable'
asp/2, i.e.,
as p --+ 2.
Remark 4.3. The conclusion of Lemma 4.1 continues to hold for cylinders of the
type
E=l
(4.2) QR (m,{3)==Kr x (-{3R" , 0) , r=~:) P R, {3 > 0,
(f)" ~ t ~ o.
In a precise way we will prove
5. The flfSt alternative 89
or
or
essosc
Q(pP,cop)
U:::;"'1 w, v P E (0, RI8),
where
"'1 == 1 - T(so+S).
which we may assume is contained in the cube KR1- •. Indeed if not, we would
have
16co > R- E, i.e., w < 16r-; AREo, £ 0 - pc
=--·
2-p
We will be working within the box
[(x,0)+Q(ft,2c o R)] .
This contains Q ( f )P, CoR), regardless of the location of x in the cube K'R.(w).
1- _
I
I
I
I
I
I
I
Figure 5.1
(5.4)
Denoting again with x and t the new variables, the function v satisfies the p.d.e.
(5.5) Vt - div i(x, t, v, Dv) + b(x, t, v, Dv) = 0, in 1Y(Q4),
where i: Q4xRN+1 -+ RN and b: Q4xRN+1 -+ R, satisfy the structure conditions
where '"Y = '"Y(N,p,A,8 0 ,data) is a constant depending only upon the indicated
quantities and bo is defined in (3.6). The numbers" and q and q, f satisfy (3.5)-(3.7)
of Chap. ll. The infonnation (5.1) translates into
(5.11) v(x, t) >i a.e. (x, t) E Q(ho ) == {Ixl < ho } x {-41' < t :s; O},
where
(5.12) doR
ho = 8coR
1
=8 A
(28 )¥ < 1.
0
We regard Q( ho ) as a thin cylinder sitting at the •centre' of Q4. We will prove that
the relative largeness of v in Q(ho ). spreads sidewise(l) over Q2.
4 (0,0)
-------,
2ho
Figure 5.2
(5.13)
or
(5.14)
Since (5.14) holds true for all time levels -2 P 5 t 50, each such box satisfies
meas {(x, t) E [(0, t) + Q2(0, m2)J \ v(x, t) 52-m2} 5110\Q2(O, m2)\.
Therefore by Lemma 4.1 either (5.2) holds or
v(x, t) ;;:: T(m 2 +1) V(x, t) E Ql.
Returning to the original coordinates and redefining the various constants accord-
ingly proves Proposition 5.1.
(6.1)
These will simplify some of the calculations and will be removed later. The weak
formulation of (5.5) is
(6.2) J
K4
Vt'p(x, t}dx + J
K4
sex, t, v, Dv)·Dcpdx = - J
K4
b(x, t, v, Dv}cpdx
Let
(6.4) w == v + G(t),
and rewrite (6.2) in tenns of w. Next, by the parabolic structure(l) of (6.2), the
truncation (k - w)+ is a subsolution of (6.2), i.e., for all testing functions r.p ~ 0
(6.5) !!
K4
(k - w)+ r.p(x, t)dx + !a
K4
(x, t, v, D (k - w)+) ·Dr.pdx
~ -!b
K4
(x, t, v, D (k - w)+) r.pdx - G'(t)! r.p(x, t)dx.
K4
C. -
_. = Ci
24p
(280A )2-P.
Set also
(6.7) ~ (w)
k
_!
-
(k-w)+
ds
[k _ s + 6kjP-1'
o
(6.8) [ k(1 + fJ) ]
!lik(w) = In k(l + 6) - (k - w)+ .
Then we obtain
+p j(~1c(W»(P-I(tdx
K4
By the choice (3.8) of the number 8 0 , the second term involving IDtP1c(WW
is absorbed in the analogous term of the left hand side. The integral involving
IDtP1c(WW-1 is treated by means of Young's inequality and the resulting term
involving 1D!P1c(W)\P is absorbed in the analogous term on the left hand side.
The remaining term is majorised by an absolute constant depending only upon
Ci , i = 0, 1. Next, if we stipulate to take k in the interval (0, 1J, the integral in-
volving (t is majorised by 'Y/(2 - p), where 'Y is an absolute constant depending
only upOn p. Finally the sum of the last two integrals can be majorised by an ab-
solute constant. Indeed
We conclude that there exist constants .:yo and .:y depending only upon N, p, A, 8 0
and the data, such that
7. An auxiliary proposition 95
(6.9) ! j
K4
~1c(w)(Pdx +;Yo jIDtP1c (w)I P(Pdx ~ 2 ~ p.
K4
for some positive number (J depending only upon q and p and some 'Y depending
only upon the data.
The number 6* will be chosen Shortly only in tenns of the data. In view of
(6.11) we may regard the function w introduced in (6.4) as independent of k and
6.
7. An auxiliary proposition
Introduce the quantities
(7.1) Yn ==
-4P<t<O
sup j (P (x, t) dx, n=O, 1,2, ....
K4n[W( ·,t)<6")
PROPOSITION 7.1. The number II E (0,1) being fixed, we may find numbers
6, uE (0,1) depending only upon N,p the data and II, such that/orn=O, 1,2, ... ,
either
(7.2)
or
(7.3)
n=1,2, ....
Since Yo::; IK41, we have only to take n=no so large that
uno-I::; 112- N .
Then the lemma follows with 26· = 6no • Indeed, Yno ::; uno -1 Yo implies
I
meas {x E K2 w(x, t) < 26·} < IIIK21.
Recalling the definition (6.4) of wand the upper bound (6.11), this in tum yields
(5.14) and concludes the proof of the lemma.
The numbers n E N and to E ( -4",0) being fixed, we consider the following two
cases:
either
(7.5) ! / ("
K.
(x, to) 4>6n (w (x, to» dx ~°
or
(7.7)
We minorise this integral by extending the integration over the smaller set
On such a set
Therefore
(In 1 ~ 6) I' J (I' (x, to) dx ~ J (P(x, to)lfIln (w (x, to)) dx.
K.n[w(·,t o )<6 n +1] K.
Yn+l ~c+C
1
( In U +6)-1' .
To prove the proposition in such a case, we choose 6 so small that
1+6)-"
C ( In-U
1/
~2'
Such a choice depends only upon the constants "Y, "Yo and 1/ and therefore it depends
only upon the data.
t. '" sup {t E (-4', t.)1 ~ 1(. (x, t)4'•• (w (x, t» <Ix <' o}.
By the definition oft.,
(8.1) J (P (x, to) ~6n (w (x, to)) dx ~ J (I' (x, t.) ~6n (w (x, t.)) dx.
K. K.
J
K4
("(x, t.)!liln (w (x, t.» dx :5 C
where C is an absolute constant depending only upon N and p and the data. By
the definition (7.1) of Yn • we have for all s E [0, I]
Yn if 0 :5 s < s.
={ C [In 1+1-.]
1 6 -"
if s. :5 s < I,
where s. is the root of the equation
[
Yn=CIn 1 6+ 6 ]-" .
1+ - s.
Solving it we find
e(c/Ynl1/P - 1
(8.3) s. = e(c/Ynl1/p (1 + 6).
Since Yn >11, we have
e(c/lIl l / P - 1
(8.4) s. < e(CM1/p (1 + 6) == 0"0(1 + 6).
Next we estimate the integral on the right hand side of (8.1). By the Fubini theorem
J("(X,t.)~6n (w(x,t.» dx
K4
8. Proof of Proposition 7.1 when (7.6) holds 99
~ /1
n p
)P_l (
c5 (2-
[1 + 15 - s]
/ (P(X,t .. ) dx) ds.
o K"n[(,s,,-w)+>s,s,,)
The last integral on the right hand side of (8.5) is estimated by means of (8.2).
Taking into account the definition of s .. in (8.3), we have
/ (P(x,to)~,s" (w(x,to» dx
K"
Bo
c5n(2-p)
</ y. ds
- [1 + 15 _ sjP 1 n
o
1
c5n(2-p) [ 1 + 15 ] -P
+/ Gin ds
[1 + 15 - sjP-l 1 + 15 - s
So
1
c5n(2-p)
= / y. ds
[1 + 15 _ sjP-l n
o
- /1{y'-Gln[1l+c5-s
+ 15 ] -P}
n [1+c5-sjP-l
ds
c5n(2-p)
Bo
where
1
F(Yn , c5) =/ ds
[1 + 15 - sjP
1
o
F(Y. b) < /
n, - [1 + bds_ sjP-1
o
(8.8) Yn +1 - e ~ Yn (1 - /(b».
We estimate /(b) below. For this let 0"1 ~ 0"0 be defined by
Then integrating the first integral on the right hand side of (8.7) over the smaller
interval [0"1(1 + b), 1]. we derive the estimate
1 - 0"t}2- p
/(6) > -(1 -
( -
2b )2- P .
2 1+6
9. Removing the assumption (6.1) 101
If to ¢ S but
sup {t < to I t E S} = to,
102 IV. ~lder continuity of solutions of singular parabolic equations
by working with a sequence of time levels tn E Sand {tn}-to, we see that (9.1)
continues to hold. If to , S and
T == sup {t < to I t E S} < to,
we derive the two inequalities
f
K,,(T)
(Pl]tf( w)dx :5 'Y,
f {P~k(w)dx:5 f (P~k(w)dx.
K,,(t o ) K4(T)
coR
Figure 10.1
10. The second alternative 103
The cylinders obtained this way are contained in Q (RP , CoR), if the abscissa x of
their 'vertices' ranges over the cube K'R.l(W)' where
!.=.I!
where Ll == (~)
2 o+n
B
p - 1.
We will take A larger than 2B o+n and arrange that Ll is an integer. Then we re-
gard Q (R", CoR) as the union, up to a set of measure zero, of Lf pairwise disjoint
boxes each congruent to Q (RP, d. R). Each of the cylinders [(x, 0) + Q (RP, d. R)]
is the pairwise disjoint union of boxes [(x, 0) + Q (R", doR)] satisfying (10.1).
Therefore we rephrase (3.5) as
LEMMA 10.2. There exists a positive integer n such that/or all t· < t <0.
(10.6)
where
Ht == esssup (1£- (p.+ - ~)) .
K"oRX(tO,O) 2 0 +
1be cutoff function x -+ (( x) is taken so that
1'n
$ uP IKdoRI·
This estimation justifies the choice of the cylinders [(x, 0) + Q (RP, d.R)) over
the boxes [(x, 0) + Q (RP, doR)). Indeed the integrand grows like 2n (2-p) due to
the singularity of the equation. This is balanced by taking a parabolic geometry
!!.1!::.£l
where the space dimensions are stretched by a factor 2 P •
Finally the last tenn on the right hand side of (10.7) is estimated above by
The left hand side of (10.8) is estimated below by integrating over the smaller set
I
{ x E K(l-u)doR u(x, t) > J.I.+ - 2s~+n } .
On such a set, since \li is a decreasing function of H:, we estimate
We carry this in (10.8) and divide through by (n - 1)2 102 2, to obtain for all t· <
t<O,
To prove the lemma we choose u so small that uN $1 v~ and then n so large that
Remark 10.1. Since the number 110 is independent of w and R also n is indepen-
dent of these parameters.
In this process we also determine the number A introduced in §2 which defines the
size of Q (RP, coR). To make this quantitative let us consider the box
meas { (x, t) E I
[(x,O) + Q (,8RP, d.R)] u(x, t) > J.L+ - 2: }
< IIIQ({3RP,d.. R) I·
11. The se<:ond alternative concluded 107
PROOF: After a translation we may assume that (x, 0) == (0, 0). Set 81 =80 + n,
and consider the energy inequality (3.8) of Chap. n written for (u - k)+, where
(11.1) !!ID· (u -
Q(fJRP ,d. R)
(JL+ - ;.)) + dxdr:$11' (d.~)1' (;. )"IQ ({JR1',d.R)1
+; (;,y-1' G:tlQ ({JR1', d.R)! + 1'(d.R)N P
(lt
C
) R1' (1:,,) .
P
We estimate above the various terms on the right hand side of (11.1) as follows.
Since 8 ~ 81 == 80 + n,
( W
28
)2-1' < (~)2-1'
- 2 +
= ..!...
cf.
ao n
:$ (d.~)1'A3W-boRNICIQ({JR1',d.R) I,
where A3 = 2mbo and bo is the number introduced in (3.6). Combining these re-
marks in (11.1) we deduce that there exists a constant l' depending only upon the
data and independent of w and R, such that
(11.2)
!!ID (u -
Q(fJRP,d.R)
(JL+ - ;.)) +1"dxdr:$ (d.~)1' G:tl Q({JRP,d.R)I,
lOS IV. Hi>lder continuity of solutions of singular parabolic equations
for all 8 =81,81 + 1, 81 + 2, ... ,m -1. Next we apply Lemma 2.2 of Chap. lover
the cube K d• R for the functions
v = u(·, t),
and the levels
l- + W
- P. - 2.+1'
By virtue of Lemma 10.2
First integrate both sides in dT over (-fJJlP, 0), then take the p-power and ma-
jorise the right hand side by making use of the IJl)lder inequality and (11.2). We
obtain
From this,
IA.+1I~ ~ 'YIQ(fJRP,d.R) l;!t IA.\A.+1I·
Adding these inequalities for 8=81, 81 +1,81 +2, ... , m - I,
( _'Y
m- 8 1
)~ -<v.
12. Proof of tile main proposition 109
and assume, by taking m even larger if necessary, that m ~ m2. Then determine
A from
(11.5)
With these choices the box Q (PR!', CoR) coincides with the cylinder QR (ml, m2)
introduced in §4. By Lemma 4.1, there exists a constant Ao dependent only upon
the data and independent of w and R such that either
wbo $ Ao RNK.,
or
w
u(x , t) < ,,+ - -
- r- 2m -
+1
where bo is the number introduced in (3.6). We summarise:
PROPOSITION 11.1. Suppose that (3.5) holdsforallcylinders [(x, 0) +Q (R!', doR)]
making up the partition ofQ (R'P, CoR). There exists a constant Ao dependent only
upon the data and indepetUknt of w and R such that either
(11.6)
orforaliO<p $ R/2,
(11.7) essosc u $
Q(fjpP ,cop)
"'0 w, where ~o == 1 - 2-(m+l).
The process can now be repeated and continued as indicated in Proposition 2.1.
Indeed, by Remark 2.1, the process can be continued as long as (2.3) holds.
Let 8 0 be the smallest positive integer satisfying (3.1) and construct the box
where the number m > 1 is to be chosen. Notice that for all R> 0, these boxes are
lying on the bottom of flT. Also withous loss of generality we may assume that
2~ ~ 1 so that there holds
PROPOSITION 13.1. There exist constants '1, eo E (0, 1) and e, m > 1 that can
be determined a priori depending only upon the data satisfying the following. Con-
struct the sequences Ro = R, Wo =W and
1
Rn = en R, wn+1 = max {'1Wni e R:t }, n=I,2, ... ,
n=0,1,2, ....
and· essoscu~max{wni2essOSCUo}.
Q~n) KRn
We indicate how to prove the fIrst iterative step of the Proposition and show,
in the process, how to determine the number m. Set
and there is nothing to prove. Let us assume for example that the second of (13.1)
is violated. Then for all 8 ~ 8 0 , the levels
w
k = IJ- + 28 '
satisfy the second of (4.16) of Chap. II. Therefore we may derive energy and log-
arithmic estimates for the truncated functions (u - k)_. These take the form
:51' /fiu-k)"-'DC'PdxdT+'Y{iiBk.R(T)'~dT} r ,
Qo(dRP,R) 0
where Dt and Bt,R are defined as in (4.2) and (4.4) of Chap. II.
LEMMA 13.1. For every vE (0, I), there exists a numberm>80~ 2 depending
only upon the data and independent of w and R such that either
or
where
p=-
R
and
W
d= ( 2m
)2-P •
2
PROOF: Consider inequalities (13.3) written for k = IJ- 2":0. As a constant c ap-
pearing in the definition of !Ii (see (3.12) of Chap. II), we tak: c = 2~. Thus we
take
13. Boundary regularity 113
where
D; == lI(u - k)-lIoo,Qo(dRP,R)·
on K p ,
vanishes for Ixl = R,
W )1'-2
2 - 1' < ( -
tJi 5 (m-s o }ln2, 1tJi.1
u - 2m '
and
'Y ff tJiltJi I
Qo(dRP,R)
u 2- 1' ID(I 1' dxdr :5 'YmIKpl·
Moreover the last tenn on the right hand side of (13.3) is estimated above by
'Ym\2ffl \2ffl)(2-1')~
CW )-2 CW . RNItIKpl:5 -ym2mbow-bo RNItIKpl,
r
(13.6) f
Kp(t)
tJi2 (D;, (u - k)_,c) dx 5 'YmIKpl, '<It E (0, dRP).
We minorise the left hand side of (13.6) by integrating over the smaller set
I
{ x E Kp u(x, t) < ~- + 2: }, '<It E (0, dJlP).
On such a set w
tJi2 ~ In2 2: 0
= (m - So - 1)21n2 2.
2"'"fT
These remarks in (13.6) give
for a constant 'Y depending only upon the data. To prove the lemma we have only
to choose m so large that
(m - So - 1)2 < v.
114 IV. HOlder continuity of solutions of singular parabolic equations
Remark 13.1. The process described has a double meaning. It defines a level
+;' for the function u and the size of the box
p. -
d-
-
(-2W)2-"
m '
Notice that no shrinking occurs in the t-direction. This is due to the fact that in
(13.1) the cutoff function, can be taken independent of t.
A crucial fact in the proof of Theorem 1.1 is the expansion of positivity of Propo-
sition 5.1. To focus on this phenomenon let us consider homogeneous equations
with measurable coefficients of the type
where the entries (x, t) - tlij(x, t) of the matrix (aij) are only measurable and
satisfy the ellipticity condition
for some A > O. In such a case, the various costants Ai and A appearing in the
proof of Proposition 5.1 are all zero and the function w introduced in (6.4) c0-
incides with v. TIle information (5.3) has been translated into the dimensionless
estimate (5.11 )-(5.12) (see Fig. 5.2).
Let us think of (14.1) as defined weakly in the cylindrical domain K4 x
(-4",0). The information (5.11)-(5.12) is that at the 'centre' of K4 x (-4P,O)
there is a thin cylinder KhD x (-4P, 0) where v > 1. The conclusion of the argu-
ments of §5-9 is that there exists a small positive number "Yo that can be determined
a priori only in terms of N,p and A, such that
Thus the 'positivity' of v over K ho spreads over a full cube K 1. Actually the infor-
mation (5.11)-(5.12) is only used to apply the Poincare inequality of Proposition
2.1 of Chap. I to derive the integral inequality (6.10). Precisely
for all t E (-4P, 0). Now to apply such an inequality it only suffices to have the
information
for some positive numbers ko and Q o and all t E (26,46). Then there exists a
number 'Yo ="10 (N,p, A, 6, Qo, ko ) that can be determined a priori only in terms
of the indicated quantities, such that
v(x, t) ~ 'Yo, V(x, t) E Kl x (36,46).
1. Introduction
(1.1) ~
lIulloo,Q. ,; ( 1 + If lUI'.) .. ,
where the numbers q" i = 0, 1, are detennined a priori in tenns of p and N and the
constant 'Y is detennined a priori in terms of the structure conditions of the p.d.e.
and Ql.
Unlike the elliptic theory, the estimate (1.1) discriminates between the degen-
erate case p > 2 and the singular case 1 < p < 2. To illustrate this point, consider
local weak solutions of the elliptic equation
u E w:1'''(fl)
loe , p>l
{
div IDul,,-2 Du = 0, in fl.
118 V. Boundedness of weak solutions
and the cylindrical domain Q~p == K 2p X (0, tl. Assume fmt that p > 2. Then for
all eE (0, 21 there exists a constant -Y='Y (N,p, e) such that for all tt~s~t
For the singular case 1 < p < 2 a local sup-estimate can be derived only if u is
SJljJiciently integrable. Introduce the numbers
(1.4) A,. == N(P- 2) +rp, r ~ 1,
and assume that u E Lfoe(DT) for some r ~ 1 such that A,. > 0. Then there exist a
constant 'Y='Y(N,p, r) such that for all tt<s=t there holds
When 1 < p < 2 such an order of local integrability is not implicit in the notion of
weak solution and it must be imposed. The counterexample of § 12 of Chap. XII
shows that it is sharp.
N+2
p:$6<p~.
The non-negative functions rpi, i=O,I, 2. are defined in ilT and satisfy
(Bs)
where
-41 = (1-11: 0
p
)--
N +p'
11:0 E (0,1].
For fELl (ilT) and h E (0, T) we let !h denote the Steklov average of f. A
function u is a local weak sub(super)-solution of (2.1) in ilT if
f {!
K:x{t}
uhrp+[a(x, T, u, DU)]h ·Drp- Ibex, T, u, DU)]h rp} dx:$ (?)O
We retain the structure conditions (Bd-(B6). and on the Dirichlet data 9 and 1£0
we assume
(2.6) 9 E L oo (BT),
(2.7) 1£0 E L2(n).
a{x,t,u,Du)· Du ~ CoIDuI P ,
(2.9) {
la{x, t, u, Du)1 ~ C1IDul p - 1,
for two given constants 0 < Co ~ Cl. The lower order terms are zero and the
principal part has the same structure as
3. Sup-hounds
We let u be a non-negative weak subsolution of (2.1) and will state several upper
bounds for it. The assumption that u is non-negative is not essential and is used
here only to deduce that u is locally or globally bounded. If u is a subsolution.
not necessarily bounded below. our results supply a priori bounds above for u.
Analogous statements hold for non-positive local supersolutions and in particular
for solutions.
The estimates of this section hold for P in the range
N+2
(3.2) q=p--,
N
The range of 6 in (B4) is p:$ 6 < q. We will assume that
(3.4) i = 0,1,2.
In this case we may take Ko = 1 in (B6) and K=p/N.
THEOREM 3.1. Let (3.1) hold. Every non-negative. local weak subsolution U 0/
(2.1) in DT is locally bounded in DT . Moreover. if 'Pi E LOO(DT)' i = 0,1,2.
there exists a constant 'Y = 'Y (data) such that V [(x o, to) + Q (PP, p)] CDT and
VUE(O,I).
(3.5) sup U
[(zo.to)+Q(upP ,up»)
Remark 3.1. The weak maximum principle holds for equations with homoge-
neous structure for all p> 1.
As a particular case, Theorems 3.1-3.3 give a priori sup-estimates for non-
negative weak solutions of
These conditions on the lower order terms are optimal for a sup-bound to hold as
it can be seen from the 'linear' case p = 2. Set p = 2 and a2 = 0 in (3.8). For a
local weak solution uE V,~c(flT) to be locally bounded, 6 must not exceed 2Nh2.
Likewise if al = 0, the forcing term II' must satisfy
A N+2
where q > -2-.
These are classical and optimal results for the linear case p=2 (see [67]).
(4.1)
(4.2) sup
[(z .. ,t.. )+Q(a9,ap»)
1.£ :5
'Y";9/PP
(1 _ u)
N(p+l)+p
2
(
sup
t .. -9<T<t.. f 1.£(x, 'T)dx
[z .. +Kp)
)
1\ ( p")~
9 .
Consider a non-negative weak subsolution of the Dirichlet problem (2.5) for equa-
tions with homogeneous structure and let (2.6) hold. If the initial datum 1.£0 is also
bounded above, then the weak maximum principle estimate (3.7) holds true. If
however 1.£t is not bounded, it is of interest to investigate how the supremum of 1.£
behaves when t -+ 0.
THEOREM 4.3. Let 1.£ be a non-negative weak subsolution o/the Dirichlet prob-
lem (2.5) and let (2.6) hold. There exists a constant 'Y = 'Y (data), such that
Vte(O,T),
(4.3) ~ = N(P - 2) + p.
Results of this kind could be used to construct solutions of the Dirichlet prob-
lem with initial data in L 1 (fl) or even finite measures. Indeed the regularity results
of Chap. III supply the necessary compactness to pass to the limit in a sequence of
approximating problems.
(4.4) - . Jp>-Ie
1I1£II{r,t} = O<T<t
sup sup
p~r
1£(x,r)
P
-2) dx, ~= N(P - 2) + p,
Kp
is finite for some r > 0 and for all t e (0, T). The subsolution 1£ at hand is not
necessarily bounded. However it is locally bounded and and as lxi- 00 it grows
no faster than Ixl;;!,. This is the content of the next theorem.
THEOREM 4.4. Let 1£ be a non-negative subsolution 0/(2.8) in L'T, and assume
(4.4) holds. There exist a constant 'Y = 'Y (data), such that/or all te (0, T),
P/2
111£(·, t)lIoo,Kp
Ie
pi' P
-2) ~ "'("Ii ( O<T<tp~r
sup sup JP>-I(
Kp
1£(x, r) )
P
-2) At
_~
p- •
Remark 4.2. The assumption (4.4) is not restrictive. We will show in Chap. XI
that it is necessary and sufficient for a non-negative solution of (2.8) to exist in
L'T·
The right hand side of (4.5) blows up as t '\, 0 at the rate of at least t - ~ . Such
a rate is not optimal. However the advantage of Theorem 4.4 is that it does hold
for all t E (0, T). The purpose of the next theorem is two-fold. It gives an optimal
estimate of how the local sup-bound for 1£ may deteriorate as either Ixl - 00 or
t'\,O.
5. Homogeneous structures. TIle singular case 1 < p < 2 125
f
RN
8(x, t)cpdx --+ M cp(O), as t '\. 0.
°
For t > and for every p> we have °
p > max { 1; : : 2 } .
In this section we will show that weak solutions uEL'oc(nT), r~ I, are bounded
provided
that is unbounded.(I) Thus in the singular range 1 < p < max {I; JZ2}' the
boundedness of a weak solutions is not a purely local fact and, if at all true, it must
be deduced from some global information. One of them is the weak maximum
principle of Theorem 3.3 and Remark 3.1. Another is a sufficiently high order of
integrability.
N+2
q=p--.
N
Therefore ifpis so close to one that ~q ~o, the orderofintegrability in (5.1)-(5.2)
is not implicit in the notion of subsolution and must be imposed.
(1) The notion of solutions that are not in the function class (2.2) is discussed in
Chap. XII
5. Homogeneous structures. The singular case 1 <p<2 127
(5.4)
Fix tE (0, T) and let us rewrite (5.4) for the pair of boxes
'Yt-N/>'r
(5.5) supu(·, t) ~ ¥.E
K"p (1-0') r
°
Then the behaviour of the supremum of u as t '\. is formally the same as that of
solutions of the Dirichlet problem for degenerate equations as in Theorem 4.3.
6. Energy estimates
The proof of the sup-bounds stated in the previous sections is based on local and
global energy estimates similar to those of §3 of Chap. II.
(6.1) sup
to-8<t<to
/ (1.£ - k)! ("(x, t)dx + "1- 1 jrJriD (1£ - k)+ (I"dxdr
[zo+Kp) [(zo,to )+Q(8,p»)
:5 ~o I liD (u - k)+ IP(Pdxdr + 'Y I Iu6 ')( [(u - k)+ > 0] dxdr
Q(9,p) Q(9,p)
+'Y I I u6 ')([(u - k)+ >0] dxdr + 'Y I I~')([(u - k)+ >0] dxdr,
Q(9,p) Q(9,p)
where
..r;.
~ = CPo +;;!:r+
CP1 6
CP2·
By the HOlder inequality and (B5)-(B6),
0 } ~(1+IC)
II~dxdr :511~lIq,nT {
IIA/c,p(r)ldr
Q(9,p) -9
Remark 6.1. Inequality (6.1) for the function (u - k)_ holds true for local super-
solutions of (2.1) and k:5 o.
Remark 6.2. Unlike inequalities (3.8) of Chap. II, the levels k here are not re-
stricted.
130 V. Boundedness of weak solutions
There exists a constant 'Y = 'Y (data), such that for every non-negative function
t--+(t) eCl[O, T] andfor every O<t~T,
and the corresponding cylinders QniEQ ((In, Pn). It follows from the definitions
that
Qo = Q «(J, p), and Qoo = Q (q(J, qp) .
Consider also the family of boxes
k
n
=k-~
2n '
where k is a positive number to be chosen. We will work with the inequalities (6.1)
written for the functions (u - kn+l)+, over the boxes Qn. The cutoff function (
is taken to satisfy
( :ani~hes_on the parabolic boundary of Qn,
{ (= 1 10 Qn,
ID'"I':. < 2"+2
- {1-a)p'
0 < <
- (t -
2"+2
(l-a)B"
With these choices, (6.1) yields
(7.1) sup
-8,,<t<O
j(u - kn+l)! (P(x, t)dx + {{ID (u - kn+l)+
JJ
(I P dxdT
Kp.. Q..
-y2np {{
~ (1- q)ppp JJ (u - kn+tl~ dxdT
Q..
Finally
7. Local iterative inequalities 133
We combine these estimates into (7.1) to derive the following basic iterative in-
equalities
Moreover the last two tenns can be eliminated for equations with homogeneous
structure.
To proceed, construct a non-negative piecewise smooth cutoff function (n in
Qn, which equals one on Qn+l, vanishes on the lateral boundary of Qn and such
that
ID(nl $ 2n+2 /(1 - u}p.
Then the function (u - kn+l)+ (n vanishes on the lateral boundary ofQn and by
the multiplicative inequality of Proposition 3.1 of Chap. I,
(7.7) / /(u -
-
kn+l)~ (!dxdT $ 'Y (_sup
-Bn<t<OK
/ (u _ kn+l)! dx) i
~ ~
x (i/ID(U-kn+d+IPdXdT+ i/(U-kn+l)~ID(nIPdxdT).
Qn Qn
Remark 7.1. The estimates in (7.2)-(7.5) and the inequalities (7.6), (7.7) are
valid for any number
6 ~ max{pj 2}.
The structural restriction 6 = q does not play any role in the derivation of (7.6) and
(7.7).
134 V. Boundedness of weak solutions
(8.1) Yn = H
Qn
(u - kn)~ dxdr, n = 0,1,2, ....
We will derive an iterative inequality for Yn by estimating the right hand side of
(7.7) by (7.6). We assume first
and estimate
We estimate the last integral by (7.7) and in turn estimate the right hand side of
(7.7) by the inequalities (7.6) and (7.3). We arrive at the recursive inequalities
b ( PN(J)~*
n
+ -y k~(q-6)
1...2.
y'l+,N +"Vbn
n,
I.
(pN(J)'" (1
-y.
k6 n
)l+~"
,
where
and
We let u be any non-negative weak subsohition of the Dirichlet problem (2.5) and
assume that (6.2) holds so that u satisfies the energy estimates (6.3). Fix 0 < t ~ T
and introduce the sequence of increasing time levels
iftn+l~T~t
k
kn = sup 9 + k - -2' n = 0, 1,2, ... , k > 0 to be chosen
ST n
and write (6.3) for the functions (u - kn+d + and the cutoff functions (n to obtain
(9.1) sup
t,,+1 <T;St
/(u - kn+l)! dx +
II
/
t
t,,+lll
"n
J" (u - kn+l)+ IPdxdT
t t
The last two terms of (9.1) can be eliminated for equations with homogeneous
structure. Moreover the last term can be eliminated if in the structure conditions
(B 1 )-(B3), !Pi ::0, i=O,I,2. If !Pi E LOO(nT), i=O,l, 2, then It= N'
Proceeding as in (7.2)-(7.5), we estimate
136 V. Boundedness of weak solutions
t t
Remark 9.1. The structure restriction 6 < q does not play any role in the derivation
of (9.2). This inequality holds for all 6 ~ max{p; 2}.
(9.3) Yn == f fa
t"
(u - kn)~ dxdr,
we obtain
(9.4)
10. Homogeneous structures 137
where b = 26(1+ t .. ).
In these. the last two tenos can be eliminated for solutions
of equations with homogeneous structure as in (2.8)-(2.9). Moreover the last teno
can be eliminated if. in the structure conditions (B 1 ) - (B 3 ). 'Pi == 0, i = 0, 1,2.
If 'Pi ELOO(nT), i=O, 1,2. then It=p/N.
Suppose now that the initial datum 1.1.0 in (2.7) is bounded above and let us take in
(6.3)
-ybnlntl*~ 1+*~
(9.5) Yn +1:5 k~(9-6) Yn + -yb
n
Inti K! ( k16 Yn )1+K~
b = 26(1+~~).
For equations with homogeneous structure. all the tenos on the right hand side of
(9.5) are zero.
The numbers Ar have been introduced in (5.1). We also assume that 1.1. can be
constructed as the weak limit in L[oc (nT) of a sequence of bounded subsolutions
of (2.8). By possibly working with such approximations we may assume that 1.1. is
qualitatively locally bounded. Below. we will derive iterative inequalities similar
to (8.3) but involving the L[oc -nonos as well as local sup-bounds of u.
If 1 <p< max { 1 j J~2}' we have q < 2. If (10.1) holds for some r E [1, 2).
then A2
> 0 and p > max { 1 j J~2}. Therefore it suffices to assume that (10.1)
holds for some r > 2. In such a case we have
(10.2) r>q,
138 V. Boundedncss of weak solutions
In (7.6) we discard the last two terms in view of the homogeneous structure of
(2.8) and, owing to Remark 7.1, set also 6 =r. We obtain
(10.3) sup
-9.. <e<o
J (u-k n +1)!(x,t)dx+ ffID(u-kn+d+IPdxdr
11 i
K~.. Q..
Define
Yn = H
Q..
(u - kn)~ dxdr, n= 0,1,2, ... ,
and estimate
Yn +1 ~ lIull:'~(9,p) H
Q..
u 9 dxdr.
We majorise the right hand side by means of (7.7) and in turn estimate the right
hand side of (7.7) by (10.3). We arrive at the recursive inequalities
(10.4) y.
n+1 -
"(bn
< (1 _ u)f(N+p) II u Il oo,Q(9,p)
r- 9 8 AIf y'1+f
n ,
where
i = 0,1,2,
so that we may take K. = N' With these choices, (8.3) yields
for a constant C depending only upon the data. This in tum implies
for a constant C depending only upon the data. This in tum implies that for all
O<t:5T
140 V. Boundedness of weak solutions
The general case of I( E (0, It) is proved by a minor modification of these argu-
ments. The second part of Theorem 3.2 is proved exactly the same way, by starting
from the recursive inequalities (9.5).
(12.1) Yn+l =
'Y bn
~
m Ynl+m ,
-Jb AI:
(1- 0')" k
where Yn are defined in (8.1) and AI: are defined in (8.4) with 6 =p, i.e.,
We stipulate to take k so large that of the two tenns making up AI: the first domi-
nates the second, i.e.,
(12.2) k>
-
(-P")
(J
p!,
so that
(J
AA; ~ 204, 04= pp'
It follows from Lemma 4.1 of Chap. I that Yn - ° as n - 00 if we choose k from
where C is a constant depending only upon 'Y, b, N and p. For such a choice and
(12.2),
(12.3) esssup u
Q("B,,,p)
~ 'YVA
(1 - 0')
~( HUPdxdr) I" (~);6 .
Q(B,p)
This estimate proves the theorem for E = 2. Fix E E (0, 2) and consider the increas-
ing sequences
Po up, =
and for n= 1, 2, ...
12. Proof of Theorem 4.1 141
n n
(12.4) Pn = up+ (1- u)p L2- i, (In = u(J + (1 - u}(J L 2- i,
i=l i=l
Set
(12.6) Mn = esssupu
Q(")
and write (12.3) for the pair of boxes Q(n) and Q(n+1). This gives
<
-
M¥"(2n~ VA ( ffUP-2+Edxdr)! AA~
~1 ~ .
(1 - u) Q(fJ,p)
If1]E (0, I), the right hand side of this inequality is majorised by
~
d= 2 • ,
where
n = 0,1,2, ....
From these, by iteration
n
Mo $1]nMn+1 +BdL(1]d)i, VnEN.
i=O
We choose 1] = fa
so that the sum on the right hand side can be majorised by a
convergent series and let n - 00 to obtain
,,(Af
sup u$ ~
Q(ufJ,up) (1 - u) •
142 V. Boundedness of weak solutions
1\ ( PP)~
0 .
PROOF: We may assume that (x o, to) =(0,0), and having fixed uE (!, 1), con-
sider the increasing sequences {p,,} and {O}" introduced in (12.4) and the corre-
sponding cylinders Q("). Let (x, t) - ,,, be a non-negative piecewise smooth cut-
off function in Q(,,+1) that equals one on Q(") , vanishes on the parabolic boundary
of Q(,,+1) and such that
The constant "Y depends only upon the data and it is independent of p, 0 and n. The
energy estimates for solutions of (2.8) give
H IDu("IPdxdT $ (1 2"P
u)PPP2 H 2
uPdxdT + (1 "Y_ "u)O H u2dxdT
Q .. +l Q"+l Q"+l
otherwise the Proposition becomes trivial. Combining these remarks with (13.2)
and setting
By the interpolation Lemma 4.3 of Chap. I, we conclude that there exists a constant
'Y, depending only upon the data such that
H UlldxdT~
Q(tr9.trp)
'Y N
(1 - u) II
( sup
-9<T<O
K,.
fU(X,T)dx)1I
y.
n+l ~
l tl m y'l+m
'Ybn f1
.!!±.I!......IL
n ,
(ut)N"+2" kWH
t tn.
tn = - - -
2 4 ;=0
2-'
'
L
and apply (14.2) over the expanding domains n x {tn+ 1, t}, with CT taken from
i.e., CT = 1 -"~
LJ,-o
2-(;+1)
~ 2-(n+1).
1 + ,,~+1 2-(i+l)
LJ,=o
Setting also
Mn = sup lIu(·, T)lIoo,a,
t .. <-r<t
we obtain from (14.2)
)..=N(p - 2) + p.
(15.2) sup
Q(ulJ,up)
u ~ 'Yff~ HUPdxdrf'A m~.
(1 - (1)
(
Q(IJ,p) )
If () and SUPK"p u(·,O) satisfy (15.1), it follows from (15.2) and the indicated
choices,
Supu(.,O)
(J N/2p
~ 'Y(PP/ )~ (Supu(.,O~
(2-p)~
N ( )1/2
l1 u p dxdr
K"p (1 - (1) \~"p 'l J(lJ,p)
J
Therefore for p.=N(p - 2) + p2,
P/IJ.
(1) The inequalities (12.1) are written over the cylinders Q(6n , Pn) introduced at the
beginning of §7.
146 V. Boundedness of weak solutions
This inequality holds for all fJ,p,u for which (15.1) is verified. It also holds for
any pair of boxes
[(x o, to) + Q (fJ, p») and [(x o, to) + Q (ufJ, up)] ,
with arbitrary 'vertices' provided they are contained in ET • Fix any te (0, T) and
introduce the boxes
and Kp/2 x Ut, t}.
We rewrite (15.3) and (15.1) in terms of these cylinders, for which u= t.
LEMMA 15.1. Forallte(O,T)andp>Ojorwhich
-(,,-2)
( )
(15.4) t ~ 21'-1", sup u(x, t) ,
K,,/3
tMreholds
ff
t
UPdxdT ~ p~ ft
(lI u(.,T)lI oo
p'tbs
,K,,)
1'-1
f U(X,T) dxdT
pN+'tbs
t/'lK,. t/2 K,.
_ (-
< 2)~ PP-
£:, { sup T
N/A
sup
lIu(.,Tlloo,K,.
.....IL
},,-1
t O<.,.<t p~r pP-'2
X { sup sup
o<.,.<t p~rK p
f ~:'1P-
dxdT}
"
~
= (~). p~ fp-l(t) Ilull{r,t}
16. Proof of Theorems 5.1 and 5.2 147
where the nonn 11·II{r,t} is defined in (4.4). Putting this estimate in (15.5) gives
We divide by (p/2)p/(p-2) and multiply by t N />... Then take the supremum for
p> r and use the fact that t E (0, t·) is arbitrary to deduce
i.e.,
Thus it follows from (15.7) that (15.S) continues to hold for all 0 < t :S t., where
t. = "Y. II u 11 -(p-2)
{r,t"} .
y.
n+l -
< "Y bn A~Ny'l+~N
! JII±,
!( 6) -"'tT n ,
6 = 2,
(1- u)P,-W-k, q-
where
(p-6)~ Ii.}
.Au = { (~) [ sup u] p + (~) p , 6=2,
p'P Q(ulJ,up)
(16.1) sup
Q(fTlJ,fTp)
11.:5
'Y~~
~
(1 - u)"N"('9-Tf
(HuQ(',p)
2dxdT)
~
, 6=2.
We conclude the proof for the case r E [1, 2) by means of an interpolation process
similar to that of Lemma 4.3 of Chap. I; namely, consider the sequences Pn, (In
and the corresponding cylinders Q(n) == Q ((In, Pn), introduced in (12.4)-(12.5).
Define also the numbers Mn as in (12.6), and write (16.1) for the pair of boxes
Q(n) and Q(n+1). This gives
where
(J ) 1/(2-p)
(16.3) Mn< ( -
- pP
and there is nothing to prove. Otherwise, (16.3) fails for all n=O, 1,2, ... and
Mn :5
'Y2np~ ~ (PP)~
~ Mn+1
(1 - u)P ,-
7i (H
Q(n+l)
11.
r
dxdT
)
~
1
k> - sup u.
- 2 Q(fT',fTp)
where
17. Natural growth conditions 149
B. ~ { ( ; ) [Q(:~pA'-2)~ + (~) ~}
With these choices, we obtain from (10.4) the recursive inequalities
y;o = jfurdXdT =
-
C(I- u)N+PB-lk(r-2)~llull(q-r)~
u oo,Q(8,p) '
Q(8,p)
(16.4) sup u
Q(u8,up)
(r 2jfR+pj
Let Q(n) =Q(On, Pn) and Mn be defined as in (12.4)-(12.6). Then from (16.4)
where
of weak solution is that of §2 of Chap. II. Here we stress that if we merely require
that IDul E V(flT), the testing functions must be bounded to account for the
growth of the right hand side.
The problem we address here is that of finding a sup-bound for a solution u.
It is known that weak solutions of (17.1) in general are not bounded, not even in
the elliptic case (see [15]). This is due to the fast growth of the right hand side with
respect to IDul. On the other hand the existence theory is based on constructing
solutions as limits, in some appropriate topology, of bounded solutions of some
sequence of approximating problems. The limiting process is possible if one can
find a uniform upper bound on the approximating solutions. Therefore the main
problem regarding sup-estimates for solutions of (17.1) can be formulated as fol-
lows. Assuming that a weak solution u of (17.1) is qualitatively bounded, find
a quantitative VlO (flT) estimate, depending only upon the data. In such a form,
the problem was fmt formulated by Stampacchia [93] in the context of elliptic
equations.
THEOREM 17.1. Let u be a bounded weak solution of (17.1). Then
07.2) lIulloo,aT :S F =IIflloo,r·
PROOF: By working with u+ and u_ separately, we may assume that u is non-
negative. Set
M = esssupu.
aT
If M > F, in the weak formulation of (17.1) we take the testing functions
(u - k)+, where k =M - E ~ F, for some E>O,
modulo a Steklov averaging process. These are admissible since they vanish on
the parabolic boundary of flT and are bounded. We obtain
(Bs)
where
(B 6 )
1 P
-: = (I-lI:o)-N .. 11:0 E (0,1).
q +p
THEOREM 17.2. Let u be a qualitatively bounded weak solution of(17.3) in nT.
There exists a constant C that can be determined quantitatively a priori only in
terms of the data. such that
(17.4) sup
O<t<T
I
l1x{t}
(u - k)! dx + Co Jr [ ID (u - k)+ IPdxdT
2 J
aT
II
~ C2
aT
ID(u - k)+ IP(u - k)+dxdT
+ II
'Y {<Pox [u > k) + IP2 (u - k)+} dxdT.
aT
152 v. Boundedness of weak solutions
Here and in what follows we denote with 'Y a generic positive constant that can
be detennined a priori only in tenns of the data. Next choose k = M - 2e where
eE (0,1) is so small that M -2e ~ IIflloo,r, and
sup
O<t<T
I
12x{t}
(u - k)! dx + fr [ ID (u - k)+ IPdxdT
1
aT
~ 'Y II
aT
"oX [u > k] dxdT.
sup
O<t<T
I
12x{t}
(u - kn )! dx + 11[[ ID (u -
aT
kn )+ IPdxdT
_ 'Y IAn 14
< (1+11:) ,
for a constant 'Y depending only upon the data. From this and the multiplicative
inequality of Proposition 3.1 of Chap. I,
17. Natural growth conditions 153
-p~ n
IAn+l I _< 'Ybn E IA 11+1<
,
u ~M - e a.e. aT
i.e.,
To prove that lIullp,nT is bounded above only in terms of the data, we may assume,
modulo a shift that involves the supremum of the boundary data, that u is a bounded
non-negative weak solution of (17.3) vanishing on r in the sense of the traces. In
the weak formulation of (17.3), take the testing function
j! aI(j
o 0
(eOS -1)ds) dxdr + Q III DulPeoudxdr
at
:S O2 IltDulPeOUdxdr + 'Y 11(1 + CPo) eoudxdr,
at at
for a constant 'Y = 'Y (data) and for all t E (0, T). We choose =202 and set
Q
to obtain
IIwllt-p(aT) :S 'Yo + 'Yl II (1 + CPo) wPdxdr,
aT
for two constants 'Yi = 'Yi (data) , i=O, 1. Next by (Bs) - (B6 ),
Moreover since w(·, t) vanishes om on for a.e. t E (0, T), by the embedding of
Proposition 3.1 Chap. I,
IIwll:~.aT :S Co +Clln,.I~lIwll:~.aT·
If T is so small that, say
then
IIwll:~.aT :S 2Co •
For arbitrary T> 0, the argument can be repeated up to covering the whole flT in
a finite number of steps.
18. Bibliographical notes 155
The sup-bounds of §3 are essentially due to Porzio [87]. They follow a parabolic
version of DeGiorgi iteration technique (see [67]) and remain valid even in the
'linear' case p = 2. An effort has been made to trace the dependence of the var-
ious constants upon the size of the domains where the estimates are derived. We
have also computed how the various estimates deteriorate when t -+ O. In the
case of homogeneous sbUctures for degenerate equations (see §4), the interpo-
lation estimate (4.1) is of particular interest. It reveals a behaviour dramatically
different from the linear case p= 2. An estimate of this kind (i.e., for small e) had
been proved by Moser [83] for solutions of linear parabolic equations with mea-
surable coefficients. Generalizations to quasilinear equations with 'linear' growth
p = 2 are in [7,97]. The global estimates in ET of §4-(IV) are taken from [41].
We have given a different and simpler proof. For the porous medium equation
with power-type non-linearities, estimates of the same nature have been proved
by BtSnilan-Crandall-Pierre [10]. Analogous estimates for general non-linearities
appear in [4]. Still in the context of the porous medium equation, rather precise
local sup-estimates have been recently obtained by Andreucci [3]. For equation
with singular structure (1 < p < 2), the theory of local and global boundedness
has started only recently in [42] and [43]. Improvements to equations with general
structures are in [87]. The results of Theorems 5.1-5.3 are sharp. They will play
a central role in the Harnack estimates of Chap. VII. The integrability condition
(5.2) is sharp as shown by the counterexample in §13 of Chap. XII. The arguments
of §17 appear in [101].
VI
Harnack estimates: the case p > 2
1. Introduction
We will establish a Harnack-type estimate for non-negative weak solutions of de-
generate parabolic equations of the type
(1.2)
This is not the case, as one can verify for the explicit solution (x, t) - 8(x, t) in-
troduced in (4.7) of Chap. v. Let (xo, to) be a point of the free boundary {t= Ixl>'},
and let p> 1. Then if to is sufficiently large, the ball Bp(x o ) taken at the time level
to - pP intersects the support of x - 8 (x, to - PP) in a open set. Therefore
This reveals a gap between the elliptic theory and the corresponding parabolic
theory. Indeed non-negative weak solutions of
(1.3) sup
Q .. ,,(Xo,to)
u <
-
'Y
(1 - u).;¥
N 2 ( ff u dxdr ) ~ ,
E
Q,,(xo,to)
where Qp(xo, to) is defined by (1.2) with p=2. This local sup-bound of the solu-
tion in tenns of the integral average of a small power of u, is a key fact in Moser's
proof of the Harnack estimate. An estimate of this kind does not hold for solutions
of (1.1) and it is replaced by the more structured inequality (4.1) of Chap. V. A
study of [83) however reveals that (1.3) continues to hold for sufficiently smooth
solutions of
(1.4)
With this in mind one may heuristically regard (1.1) as it were (1.4) written in
a time scale intrinsic to the solution itself and, loosely speaking, of the order of
t [u(x, t)]2- p. Next we observe that (2.2) in the Preface is equivalent to
(1.5)
The Harnack estimate of Krylov and Safonov [64) for non-divergence parabolic
equations is given precisely in this fonn.
This suggests that the number [u(xo, t o)]2-Pis the intrinsic scaling factor and
leads to conjecture that non-negative solutions of (1.1) will satisfy the Harnack
inequality with respect to such an intrinsic time scale.
where
(2.2)
is contained in nT •
t o+ 0
I. I
p
4p
Figure 2.1
Remark 2.1. The values u(xo, to) are well defined since u is locally Wider con-
tinuous in nT.
Remark 2.2. The constants "'( and C tend to infinity as p - 00. However they are
'stable' asp'\.2, i.e.,
(2.4) V (xo, to) E nT. Vp, (J > 0 such that Q4p«(J) c nT,
Remark 2.3. Inequality (2.4) holds for all p E (2,00), but the constant B is not
'stable' as p '\,2, i.e.,
lim B(N,p) = 00.
p'\.2
In (2.4) the positivity of u(xo, to) is not required and (J > 0 is arbitrary so that
Theorems 2.1 and 2.2 may seem markedly different. In fact they are equivalent,
i.e.,
PROPOSITION 2.1. Theorem 2.1 <=> Theorem 2.2.
In view'of Remark 2.3, the equivalence is meant in the sense that (2.1) implies
(2.4) in any case and (2.4) implies (2.1) with a constant 'Y = 'Y( N, p) which may
not be 'stable' as p'\,2. A consequence of Theorem 2.2 is
COROLLARY 2.1. There exists a constant B > 1 depending only upon N and p,
such that
(2.6) V(xo, to) E nT, Vp, (J > 0 such that Q4p«(J) c nT,
2-(i). Generalisations
All the stated results remain valid if the right hand side of (1.1) contains a
forcing term f, provided
(2.7) q> (N +p)/p
and / is non-negative. We will indicate later how to modify the proofs to include
such a case.
~
(3.1) _ kpN { ( Ix - xl ) ;f-r } P-
Blc,p (x, tj x, l) == SN/>'(t) 1 - S1/>'(t) +'
Blc,p (x, t; x, l) ~ k,
and that forf~t~t*, the support of Blc,p (x,t;x, l)
Then u~v in nT .
PROOF: We write the weak form of (1.1) for u and v in terms of the Steklov-
averages, as in (1.5) of Chap. II, against the testing function
Differencing the two equations and integrating over (0, t) giv\ 'S
3. Local comparison functions 161
As h -+ 0 the second tenn on the left hand side tends to zero since (v - u) + E
C (liT). Applying also Lemmas 3.2 and 4.4 of Chap. I we arrive at
(3.3)
(3.4) t ~ t,
where the positive numbers II and ~(II) are linked by
LEMMA 3.2. The number v> 1 can be determined a priori only in terms of N
aTul independent ofpE [2, 5/2], such that Qk,p is a classical subsolution of
Then, by calculation,
1( N(P-l»)]
IIzlIFT ~ 2 1+ N(P-l)+p
--Z..-
£1 == [ ,
we have
:F < p in £1,
- 2[N(P - 1) + p]
and therefore by (3.5)
1( N(P - 1) )]
£2 == [ IIzll ;!y < 2
p- 1 + N(P _ 1) + p ,
we have
:F> P
- 2[N(P-l)+p]
It follows from (3.5) and (3.7) that
p
£* (g/c,p) :5 -v ( 2[N(P _ 1) + p]
);!r + aP - 1 N + ~(v)a
p );!r a -2
:5 -v ( 2[N(P _ 1) + p] + p [NpaP + 1] .
4. Proof of Theorem 2.1 163
Choosing
(hc,p (XI tj X, f) $ k,
U(X, f) ~ k
then
U(X, t) ~ (h:,p (x, tj ft, f) in C*.
Remark 3.1. The same proof shows that gkr is a sub-solution of 0.1) also for
P< 2, providedp is close to 2. Precisely ifpE (4 - p(v),2) ..
X-Xo
x----
P I
(0,0)
I
,
I
•
--------------------~
c/o 1
4 -4
Figure 4.1
We denote again with x and t the new variables. and observe that the rescaled
function
v(x,t)
tpp)
= U (x o,1 t) U ( xo+px, t o+ [U (xo,to )IP-2
0
{
Vt - div (lDvlp-2 Dv) = ° in Q
v(O, 0) = 1.
To prove the Theorem it suffices to find constants 'Yo E (0, 11 and C > 1 depending
only upon N and p such that
Since v is continuous in Q there exists within Q.,." at least one point. say (i, t).
such that
v (i, t) = N.,." = (1 - To)-fl.
The next arguments are intended to establish that within a small ball about i and
at the same time-level f the function v is of the same order of (1- To)-fl. For this
we make use of the R))der continuity of v and more specifically of Lemma 3.1 of
Chap. III.
4. Proof of Theorem 2.1 165
Set
R = 1- 'To
2 '
and consider the cylinder with 'vertex' at (x, t)
It follows that [(x, t) + Q (aoRP, R)] can be taken as the starting box in Lemma
3.1 of Chap. III. We conclude that there exist constants "( > 1 and 0:, Co E (0,1)
such that for all rE (0, R].
We let r =u R and then choose u so small that for all {Ix - x I < u R},
(4.1) v(x, t) ~ v (x, t) - 211+1"((1 - 'To)-pu Ot
= (1 - 2p+1,,(uOt ) (1 - 'To)-p
1
= 2(1 - 'To)-p.
The various constants appearing in Proposition 3.1 and Lemma 3.1 of Chap. III, in
our context, depend only upon N and p and are indePendent of v, (1) therefore the
number u can be determined a priori only in terms of N, p and (J. We summarise:
LEMMA 4.1. There exist a number u E (0,1) depending only upon N,p and P
such that
(4.2)
Remark 4.1. The location of (x, t) and the number 'To (and hence R) are deter-
mined only qualitatively. However in view of (4.1) the number u is quantitatively
determined as soon as P> 1 is quantitatively chosen.
At the time level t=C the support of x-(ik,p (x, Cj x, t) is the ball
where
'Y = 'Y(u, II ) = 21(U)"¢l
'2 .
Choose
II
P = A(II) and
Since Ixl < 1 and t E (-1,0]. these choices imply that the support of x -
(ik,p (x, Cj x, t) contains B2. and by the comparison principle
== 'Yo·
The various constants depend only upon N and p and are 'stable' as p'\. 2.
5. Proof of Theorem 2.2 167
Choosing
3>-
(4.4) {3=N and C = b-yp-2'
we see that the support of Bk,p (-, C; x, l) contains B2, and by the comparison
principle,
~ (2)-(1+'i') Gt {1-
== -Yo·
Gtt'
Remark 4.2. These estimates involving the comparison function Bk,p hold for
all p > 2. However as p'\. 2, the constant -Yo in (4.5) tends to zero. The purpose
of introducing an auxiliary comparison function gk,p for p near 2 is to have the
constants under control as p approaches the non-degenerate case p = 2. We also
remark that gk,p is a subsolution of (1.1) only for p close enough to 2.
(5.1)
Indeed otherwise
B == (2C);f-J ,
168 VI. Harnack estimates: the case p> 2
u* ~ 'Yu(x, t*),
Consider the 'fundamental solution' Bk,p with pole at (0, t*) and with k='Y-1U*.
By the comparison principle, at the level t=(J, we have
(5.2)
where
u«,6);' ;t.; {1- (Sl~~I(t)t r ~
"Ixl < p,
Here .x and b are defined in (2.5) and (3.2) respectively. It follows from (5.1) that
( pp)N!>.
u(x,(J)~u~!>'"9 'Y1, 'Y1 == 'Y1(N,p),
guarantees the local HOlder continuity.(l) Moreover the comparison principle re-
mains applicable since J ~ o.
The assumption that· the cylinder Q 4p( 8) be contained in the domain of definition
of the solution is essential for the Harnack estimates of Theorems 2.1 and 2.2 to
hold. Indeed the function (x, t) - 8(x, t) introduced in (4.11) of Chap. V does
not satisfy (2.4) for Xo = 0 and to arbitrarily close to zero. This is not due to the
pointwise nature of (2.1) and (2.4). A Harnack inequality, with to arbitrarily close
to zero, fails to hold even in the averaged form (2.6). To see this let 1£ be the unique
weak solution of the boundary value problem
1£t - (I1£z IP-21£z)z = 0 in Q:= (O,I)x(O,oo),
('P)
{ 1£(0, t) = 1£(1, t) = 0 for all t ~ 0,
1£(,,0) = 1£0 E C:'(O, 1)
1£o(x) E [0,1], "Ix E (0,1) and 1£o(x) = 1 for x E U, i)·
We claim that
-1 1£
(6.1) 1£t > - - - in V'(Q).
- p-2 t
Let us assume (6.1) for the moment. Since 0 ~ 1£ ~ 1, by the comparison principle
(6.1) implies that
_ 1
- (11£.IP 21£_) < t> O.
• • z - (p - 2)t'
-yx6
v(x,t) = ~, OE ( P-l
-p-,1 ) , (-y0)P-1 (1 - 0)(P - 1) ~ ~2'
ti=I p-
Indeed
1
- (Ivz IP- 2vz)z ~ (p _ 2)t and v(O, t) = 0, v(l, t) > O.
Therefore for every 0 E ( 7' 1) there exists a constant C =C (0), such that
C(O)
1£ (!, t) ~ t 1/(p-l)'
(1) See the structure conditions in §1 of Chap~ II, Theorem 1.1 of Chap. III and Theorem
3.1 of Chap. V.
170 VI. Harnack estimates: the case p> 2
Now assume that (2.6) holds for to=O, x o =!, 9=t and p= 1. Then for t>I
1 ~ canst (cp!-, + c*) -- 0 as t - - 00.
The proof of (6.1) is a particular case of the following
Then ifp>2.
-1 u
(6.3) Ut> - - - in 1)'(11) a.e. t > 0,
- p-2 t
andifI<p<2.
1 u
(6.4) Ut < -- - in 1)'(11) a.e. t > O.
- 2-p t
PROOF: We only prove (6.4). By the homogeneity of the p.d.e., the unique solu-
tion v of (6.2) with initial datum
for some ( e (0, ~).If h <0, and Ihl « I, we have k < 1, v(·,O) ~uo and (6.S)
holds with the inequality sign reversed. Divide by h and in (6.5) take the limit in
'D'(I1) as h-+O.
7. Global Harnack estimates 171
Remark 6.1. In the proof of the proposition, the homogeneity of the operator and
the positivity of the initial datum, are essential.
The averaged Harnack estimate (2.6) holds with to arbitrarily close to zero for
non-negative local solutions of (1.1) in the strip ET == R N X (0, Tj, i.e.,
Inequality (7.2) is more general than (2.6) in that the value u(Xo, to + (J) is
replaced by the infimum of u over the ball Bp(x o) at the time level to + (J.
In (7. 1) no conditions are imposed on x-u(x, t) as /x/- 00 and no reference
is made to possible initial data. The only global information is that the p.d.e. is
solved in the whole strip ET. Nevertheless (7.2) gives some control on the solution
u as /x/- 00, namely,
PROOF: Apply (7.2) with to=TE (0, T-e), divide by p~ and take the supre-
mum of both sides for p ~ rand T E (0, T - e).
172 VI. Harnack estimates: the case p> 2
There exists a constant B=B(N,p) > 1. such that for all 8>0,
(8.4) !
aN
v(x,t)dx = !vo(x)dx,
Br
vt ~O.
+-m
(x, f) == (0,0). For t=O and Ixl <r,
-1 v
(8.5) Vt > ---
-p-~t'
in V' (Eoo) ,
(8.7) -
IIvll r = sup sup
tER+ p?,r
J v(x,t)
p>'-!(
P
-2) dx, A = N(p - 2) + p,
Bp
are finite. Moreover there exists constants 'Y. and 'Y depending only upon N,p.
such that
By the results of Chaps. III and V, the solution u is locally bounded and locally
HOlder continuous in E T • Therefore up to the translation that maps (xo, to) into
the origin, u(·, 0) is continuous in B r . By (8.3) the comparison principle of Lemma
3.1 can be applied over the bounded domain B R(T) X (0, T) to yield v $ u. Then
(7.2) follows from (8.2).
174 VI. Harnack estimates: the case p > 2
I jlDvIP-1(P-1dxdT
o Bap
t
(i (i T-~v2~
c! 1
~ I Tl/PIDVIPV-2/P(PdxdT) P I dxdT) P
o Bap 0 Bap
t
P It;v2~(PdX + P - 2 IIT;IDvIPv-2/P(PdXdT
2(P-l) p
B2p 0 Bap
t
=p I IT;IDvIP-2vl-2/P(P-IDV.D(dxdT
o Bap
t
+ 2(P 1-1) If.1 7"P - 1V2c!
p (PdxdT
.
o B 2p
In the estimates below 'Y denotes a generic positive constant that can be detennined
a priori only in tenns of N and p and that might be different in different contexts.
By Young's inequality and the structure of the cutoff function (
9. Proof of Proposition 8.1 175
P It
:::; "Ypl+~ It
Next
Remark 9.1. The estimates above show that 'Y ='Y( N, p) /00, as p \. 2.
Remark 9.2. The proof is independent of the fact that the initial datum is of com-
pact support and that v is a solution in the whole 1:00 • The lemma continues to hold
for every non-negative solution in 1:T for some T > 0, provided the quantities
lIu(·, T)lIoo,B~
sup sup 2
O<?<r-E p>r ppl(p-)
are finite for all E E (0, T). The conclusion will hold for all times
Remark 9.4. The functional dependence upon t on the right hand side of (9.1) is
optimal, as shown by the following example. The family
IX I ...L..}~
{l-'Y
p-l
.>. -NI.>.
(9.2) Br(x,t)={t+r) p ( 11'>') ,
(t+r'>']
+
solves (8.1) with initial data Brh 0) supported in the ball B r. By calculation we
have, for all p ~ r
!
t
(9.3)
--r-~Eo·
rP
(9.4) for all 0 < t < "Y. EC- 2 ' andfor all p ~ r
t 1
Let ( be the standard cutoff function in B2r that equals one on B r . In the weak
formulation of (8.1) take x -+ (P (x) as a testing function and integrate over B 2r X
(0, to), where
(10.1)
to
f
B2..
v(x, to) dx ~ 2- N
B..
f Vo dx - ~ f
OB2 ..
f IDvl p - 1dx
~ 2- N Eo - "Y (e"Y.)l/~ Eo
-- 2-(N+l) E 0,
We summarise:
178 VI. Harnack estimates: the case p> 2
LEMMA 10.1. There exist a constant c. that can be determined a priori only in
terms of N and p. such that
Next we apply the Harnack estimate of Theorem 2.1. For this, construct the cylin-
der
Q_ B (-) { 4C (6r)" 4C (6r)" }
= 46r X X to - [v(x, t o)],,-2 ' to + [v(x, to)]" 2 '
where C is the constant determined in Theorem 2.1 and 6 > 0 is to be chosen. Such
a box is contained in Loo if
t > 4C(6r)"
o -
v x,to )],,-2·
[(_
(l0.4)
11. Proof of Proposition 8.1 concluded 179
(11.1)
Indeed otherwise
Eo:5 B ( 8);;!-'
rP ,
and (8.2) becomes trivial. We will expand the bound below on v given by (10.3),
up to the time level 8 over the ball B r • Consider the 'fundamental solution' 8/c,p
introduced in (3.1)-(3.2), with pole at (x, t) and
p=6r,
Let us estimate below the right hand side of (11.2) at the time level t=8. First by
(10.4) and (11.1)
- 1
8-t>-8
- 2 '
therefore the support of X-+8co E o .6r (x, 8j x, t) will cover the ball B 4r about the
origin if
(11.3)
Therefore
Eo == f Vo dx :$ 6;>'lp (0 ) NIII [
rP W! v(·, 6}] >'111 ,
Br
1be proof of (7.2) is based on comparing u with the unique solution of the Cauchy
problem
Such a problem plays a role also in the theory of Harnack estimates for non-
negative weak solutions of (1.1) in the singular case 1 < p < 2. The Cauchy
problem for general initial data in Lloc
(RN) and all p > 1 will be studied in
Chaps. XI and XII. To render the theory of Harnack inequalities self-contained
we briefly discuss the unique solvability of (12.1) for all p > 1. First, the notion of
solution is:
(a) For every compact subset JC c RN and for every T > 0, u is a local solutions
of the p.d.e. in JC x (0, T), in the sense of (1.2)-(1.4) of Chap. II.
(b) v(·, t} -+ Vo in L2(RN}.
PROPOSITION 12.1. There exists a unique solution to (12.1) for all p > 1.
PROOF: For n = 1, 2, ... let Bn be the ball of radius n about the origin and con-
sider the boundary value problems
The functions Vn vanish in the sense of the traces on Ixl = n. We regard them as
defined in the whole Eoo by extending them to zero for Ixl > n. The problems
(12.2) can be uniquely solved by a Galerkin(l) procedure and give solutions Vn
satisfying
(12.3)
(12.4) VneN.
Therefore
Vn e LP (R+; W1,P(R N )) uniformly in n.
A subsequence can be selected and relabelled with n such that Vn - v uniformly
on compact subsets of Eoo and weakly in LP (R+; Wl'P(RN)). The limit v is
in the function space specified by (12.1), it is HOlder continuous in Ex (e, 00)
for all e > 0, and it satisfies the p.d.e. weakly in Eoo. To prove this we select a
compact subset IC c RN and some T > O. Then if n is so large that IC C B n , we
write (12.2) weakly against testing functions supported in ICx (0, T). The limiting
process can be carried on the basis of the previous compactness and the non-linear
term is identified by means of Minty's Lemma. (4) It remains to show that v takes
the initial data Vo in the sense of L2(RN). Let 1/ e (0, 1) be arbitrary and let VO,'l
be a mollification of Vo such that
To prove uniqueness we first write the p.d.e. satisfied by the difference w =VI - V2
of two possibly distinct solutions originating from the same initial datum vo , i.e.,
wE C (R+;L2(RN») nLP (R+;Wl,P(RN »),
{
(12.5) Wt - div (IDvIIP-2 DVI - IDV2l p - 2DV2) = 0, in 1:00 ,
w(',O) = 0, in L~oc(RN).
In the weak formulation of (12.5), take the testing function w(, modulo a Steklov
average, where x --+ (x) is a non-negative piecewise smooth cutoff function in
the ball B2R that equals one on BR and such that ID(I ~ 1/R. This gives, for all
t>O,
!
t
!
t
=- !<IDVIIP-2DVI-IDV2I,,-2DV2,D()Wdxd1'.
OB21l
1be second integral on the left hand side is non-negative(l) and it is discarded.
1berefore
(13.2)
(l.1) {
u E C,oc (0, T; L~oc(n») n Lfoc (0, T; W,!:(n)), 1 <p<2
Ut - div IDuIJI-2 Du = 0, in UT.
Weak solutions of (1.1) exhibit an intriguing behaviour. Even though in general
they are not locally bounded,(I) they might become extinct after a finite time. It
turns out however that the Harnack inequality of Theorem 2.1 of Chap. VI contin-
ues to hold provided p satisfies the further restriction
2N
(1.2) N + 1 <p < 2.
We will show that such a range of p is optimal for a Harnack estimate to hold. The
extinction in finite time, the Harnack inequality and the LOO-estimates are linked
by the range (1.2) of the parameter p.
THEOREM 1.1. Let u be a non-negative weak solution of (1.1 ) and let (1.2) hold.
Fix any (x o, to) E nT and assume that u(xo, to) > O. There exist constants 'Y> 1
and cE (0, 1). depending only upon Nand p. such that
(1.3)
where
(1.4)
provided the cylinder
(1.5) Q4p(0) == {Ix - xol < 4p} x {to - 40, to + 40}
is contained in nT.
Remark 1.1. The statement of Theorem 1.1 is the same as that of Theorem 2.1 of
Chap. VI except that now the constant c is 'relatively small'; that is, the positivity
of u(xo, to) spreads over the ball Bp(xo) but is preserved only for the 'relatively
small' time c [u(xo, t o)]2-Ppp.
Remark 1.2. As p '\. J~l' the constant "Y tends to infinity and c tends to zero.
However these constants are 'stable' as p /' 2. i.e.,
lim "Y(N,p), c- 1 (N,p) = "Y(N, 2), c- 1 (N,2) < 00.
p/2
Therefore the classical Harnack inequality for non-negative solutions of the heat
equation can be recovered by letting p /' 2 in (1.3). The limiting process can be
made rigorous by the C,7/:
(nT ) estimates of Chap. IX.
t o+ a
1 I
p
4p
Figure 1.1
Fix (xo, to) E nT, assume that u(xo, to) >0 and construct the truncated
'paraboloid' of two sheets
COROLLARY 1.1. Let u be a non-negative local weak solution of (1.1) and let
p be in the range (1.2). There exist constants c E (0,1) and 'Y > 1 that can be
determined a priori only in terms of N and P. such that
°
V(Xo, to) E nT, v 8 > such that P{s,4} (XO, to) C nT,
(1.3') u(xo, to) ~ 'Yu(x, t), V(x, t) E P{s,l} (xo, to).
(1.3")
Figure 1.2
full domain of defmition of x -+ u(x, to). This is the content of Theorem 14.1
of Chap. IV and holds for non-negative solutions of (1.1) for the whole range
1 < p < 2. When p is in the range (1.2), such a property can be made quantitative
and takes the form of an elliptic Harnack inequality.
THEOREM 1.2. Let u be a non-negative weak solution off1.1) and let (1.2) hold.
Fix any (x o, to) e nT and construct the cylinder
Q4p(O) == {Ix - xol < 4p} x {to - 40, to + 40},
(1.6) {
0= c [u(x o , t o )]2- p pP, c> 0,
where c is the constant of (1.4). There exist a constants.'Y > 1, depending only upon
N and p, such that
(1.7)
1-(U). Generalisations
The theorems generalise to the case when the right hand side of ( 1.1) contains
a forcing term I(x, t, u) provided
(2.4)
By the U>lder inequality and the embedding of Corollary 2.1 of Chap. I, we have
where
By integration
2. Extinction in finite time (bounded domains) 189
(2.6)
and
0< r* :::; 1'''lnl ii lIuoll~:6.
Remark 2.2. The estimate (2.6) is 'stable' as P '\. J~l' i.e., as ~ '\. O. As p /
2, the boundary value problem (2.1) tends to the corresponding boundary value
problem for the heat equation(l) for which there is no extinction in finite time.
Accordingly, letting p / 2 in (2.6) gives
In the weak formulation of (2.1) we select, modulo a Steklov average, the testing
function u s - 1 , where
2-p
(2.9) s=N-- > 1.
P
This gives
where
(1) If u(p) are the solutions of (2.1) and u is the solution of (2.1) with p = 2, the
convergence takes place in the sense
U (p) ,U"'i
(p) --+ U, U"'i .
In Co< [-n X (E,OO )] ,vE > 0 ,
\oJ i = 1,2, ... ,N,
and
uniformly in [0, TI, "IT> o.
Estimates of u~) in Co< (ax (E, 00») uniform in p>2N/(N +2) will be given in Chap. X.
190 Vll. Harnack estimates and extinction profile for singular equations
We conclude that
d
dt lIu(·, t)II.,n + '13I1u(·, t)II::t: ~ 0 in Z>'(R+), '13 == '1-P '12.
u t < u {
II (., )1I.,n - II olls,n - 1
( )
2-P'13 t }
~
lIuolI!1 +
This in tum implies (2.3).
Remark 2.3. These estimates deteriorate as p /' JZl
and are 'stable' as p '\.1.
However we cannot infer the convergence of (2.1) to a boundary value problem,
in some reasonable topology, since the Hinder estimates of Chap. IV deteriorate as
p'\.I and (2.1) only gives IDul EL1(nT) uniformly in p.
Remark 2.4. Proposition 2.1 holds for solutions of variable sign. The only modi-
fication in the proof occurs in the case 1 <p< JZ1'
N ~ 2. For this it suffices to
take the testing functions lul s - 2 u.
Qp(xo, to) == Qt(xo, to) u Q;(xo, to), Q;(xo, to) == Bp(xo) x {to ± PP}·
Indeed if (x o, to) belongs to the extinction profile and p is so small that Qp(x o, to) C
n oo , the solution u of (2.1) is positive in Q;(xo, to) and it vanishes identically in
Qt(xo, to).
1be intrinsic geometry of the Harnack inequality (1.3) implies an estimate of the
rate of extinction of u( ., t) as t /' T*. We let
M = lIull oo ,nco •
LEMMA 2.1. Let u be the unique non-negative weak solution 0/ (2.J) and let p be
in the range (1.2). There exists a constant '1 depending only upon Nand p, such
that/or all (x, t) E n x (~. , TOo)
3. Extinction in finite time (in a,N) 191
PROOF: Fix x En and ~. :S t :S T*, assume that u(x, t) > 0 and set
(2.8)
T*
4p == min { dist{x,8n}; ( 2M2-p
)l/P} .
We apply (1.3) over the ball Bp(x) and the cylinder
PROPOSITION 3.1. Let u be the unique non-negative weak solution ofthe Cauchy
problem
(3.1)
Then, if
2N
(3.2) l<p< N+l, N~2,
there exists a positive number T* depending only upon N, p and U o such that
u(·, t) == 0, "It ~T*.
Moreover
2-p
(3.3) o < T* < 'V·*llu 11 2s,B.,.'
- I 0-p 8=N--,
P
for a constant 'Y** depending only upon N and p.
PROOF: The solution of (3.1) can be constructed as the uniform limit in Eoo of
the sequence {un}nEN of the solutions of the problems in bounded domains(l)
8=N~.
The proof is the same except for making precise in what sense the solutions of
(3.1)n converge to the solution of (3.1). (2)
(3.4) T. C
- to < 4P to,
where c is the constant appearing in (1.4). Now choose p>O so large that
(3.5)
is contained in Loo. If (1.3) were to hold for 1 < p < J~l' N ~ 2, for some
constants c and 'Y independent of p, it would give
(4.1)
Since 1 < p < 2, the number A = N(P - 2) + p might be of either sign. The
proposition can be regarded as a weak form of a Harnack estimate, in that the
Ll-norm of u(·,t) over a ball controls the Ll-norm of U(·,T) over a smaller
ball, for any previous or later time. It could be stated over any pair of balls Bp(xo)
and Bqp(xo) for q E (0, 1). The constant 'Y ='Y(N,p, q) would depend also on q
and 'Y(N, p, q) / 00 as q / 1.
Remark 4.1. The proof shows that the constant 'Y(N,p) deteriorates as p /2.
The proof depends on some local integral estimates of the gradient IDul which we
derive next.
'V(xo, to) E floo, 'Vp > 0 such that B4p(Xo) c fl, 'Vt > to, 'Vv > 0, 'Vq E (0,1),
there.. holds
t
< 'YP
- (1 - q)p
[1 + (t - to) vP-2] (~);.
pP p>'
!tf.::!l
t
(4.3) ~j fiDUIP-ldXdT
toB.. p(zo)
(4.4) ~j jlDulP-1dxdT
toB.. p(zo)
Remark 4.2. The estimates (4.1)-(4.4) have been stated 'locally'. However they
=
continue to hold for to 0, i.e. for cylinders Bp(xo) x (0, t) carrying the 'initial
data'.
<
Fix 0' E (0, 1) and let x -+ (x) be a non-negative piecewise smooth cutoff function
in Bp that equals one on Blip and such that ID(I ~ 1/(1 - O')p. In the weak
formulation of (4.5) take the testing function
v> 0,
We estimate the various tenns on the right hand side in tenns of the quantity
8 == sup ju(x,T)dx.
. O<.,.<t
-: - Bp
(I·) p
2(P-l)"
t.lj(u+v)~ I"Pdx
" ..
Bp
J.
'5:,y t" p "
~ ( -
sup ju(x, T)dx + v pN
O<.,.<t
- Bp
)
~
:5 'Y sup j (u + v) ~
j T" -ldT o<.,.<t
J. " (x, T)dx
o . - - Bp
p "Y(N,p)
(1 - u)p
(.!.-)
pi'
Vp - 2 (.!.-) ; (8 +
p>'
VpN) 2(,;1) •
Combining these estimates in (4.6) proves (4.2). To prove (4.3), write (4.2) with
(x o, to) == (0,0) and select v 2 - p = (t/ PP). Then by th~ HOlder inequality
t
(4.7) IIIDUIP-1dxd.,.
OB"p
t
{f H=1
= JJ 'T'P I' (u+v)-"12=1 IDulp-1.,.-p.12=1 (u+v)"12=1
I' " dxd.,.
I'
OB"p
2=1 .1
~OB"p OB"p
The first integral on the right hand side of (4.7) is estimated by (4.2) with the
indicated choice of v and it is majorised by the same quantity on the right hand side
of (a), apart for a factor (l-U)I- P. Combining these remarks in (4.7) proves (4.3).
Finally (4.4) follows from (4.3) by a further application of Young's inequality.
4. An integral Harnack inequality for all 1 < p < 2 197
and let x -+ 'n (x) be a piecewise smooth non-negative cutoff function in Bn that
equals one on Bn and such that ID'nl ~ 2n+2/ p. In the weak formulation of (4.5)
take 'n as a testing function to obtain
for any two time levels 'Tl and'T2 in [0, tJ. We take as 'T2, a time level in [0, tJ such
that
We also set
Sn == sup ju(x, 'T)dx.
O<T<t
- - Bn
Next we apply (4.3) over the pair of balls Bn and Bn+l for which (1 - u) ~
2-(n+2). This and Young's inequality give
2n;2
t
j jIDuIP-1dxd'T ~ ",bn ( ; )
.1
P (Sn+l) 2(,;1) +",bn (;l) 6
-P
0-
Bft
~eSn+l+",(N,p,e)bn ( t)J!;
pl '
valid for every eE (0, I), for some constant ",(N,p, e) depending only upon N, p
198 VII. Harnack estimates and extinction profile for singular equations
and E. Combining these estimates, we conclude that for every E E (0,1) there exists
a constant 'Y( N, p, E) such that
11le Proposition now follows from the interpolation Lemma 4.3 of Chap. I.
+'YC~to)~ .
This and Proposition 4.1 imply
LEMMA 5.1. Let u be a non-negative local weak solution of (1.1) in noo and let
(1.2) hold. There exists a constant 'Y(N,p) such that
+'Y C~to)~.
The constant 'Y( N, p) tends to infinity as either p '\, J~1 or as p / 2.
Remark 5.1. The lemma continues to hold also for to = 0, i.e., for cylinders
Bp(xo ) x R+ carrying the 'initial' data.
The peculiar feature of this estimate is that the supremum of the solution
over a ball at some time level is bounded above by the Ll-nonn of u over a
larger ball at either the same time level or some 1uture' time. This is in contrast
with the behaviour of non-negative solutions of the heat equation. Accordingly,
the constant 'Y(N, p) deteriorates as p / 2.
6. Local subsolutions 199
Q4R == B4RX{-4, O}
is all contained in the domain of definition of u. Then
6. Local subsolutions
As in the degenerate case, the proof of Theorem 1.1 is based on expanding the
positivity set of the solution u by means of suitable comparison functions. Let
b, k, IJ. be positive parameters satisfying
(6.1)
(6.3)
Figure 6.1
PROOF OF LEMMA 6.1: The function x-!li(x,t} is radial and decreasing with
respect to Ixl. so that writing (6.4) in polar coordinates we have
where
We write
~ (IXI")~
IIzll =kFIb -t ; F = 1 + IIzll
k
W=--l 11 = (1 -lxI 2 );!r.
FF-;
(6.6)
=
We calculate the expressions 1/1' w'v + wv' and 1/1" = w" v + 2w'v' + wv" from
(6.6) and combine them into 'R.(I/I) to obtain
- (p - 1) [ - P -IIzll + 1] + p-1 }
2-p:F :F
+ 2pw(I- p2)~ {(N - 1) _ ~ M + I}.
p-I 2-p:F
Rewrite the first factor in braces on the right-hand side of (6.7) as
P-
{ ... } = ( N -
2- p :F
M).
We will impose on /I z II to be so large that
(6.8) N-_p_M<o.
2 -p :F
This is possible since N(p - 2) + p == A> O. The second term in braces on the
right hand side of (6.7) is negative if we choose IIzll to satisfy (6.8). If N = 1,2,
this is a direct consequence of (6.8). If N ~ 3,
(P_I){N-I _ ~M
p-I 2-p:F
+I} = (N- _P_
2-p :F
M)
+ -p- (3 - 2p)M + (p- 2).
2-p :F
The first term is negative in view of (6.8) and the second is negative since p >
;:~l > ~ if N ~ 3. We drop the last negative term on the right hand side of (6.7)
and estimate
202 VII. Harnack estimates and extinction profile for singular equations
Also
!lit = _1_ vw ~. 1l:.ll
2-p :F t
Therefore
(-!li')2-"!lit :::; ~ (~)2-" vwM~.
2-p p :F t
We combine this with (6.9) into (6.5) and set
w 2-"pP = (
t
IIzll
1 + IIzll
),,-1 1-" <- 11'-1
6 _I_
and
.c*(!li) < ...::L + -p-
- 11'-1 2- p
[-A + E-]
:F
.
We will choose II z II so large that ~ :5 ~,and then select 6 from
"y p A
-----=0
11'-1 2- P 2 .
o < t :::; fJ = ( 2p
A ),,-1 ",.
7. Tune expansion of positivity 203
(7.1) k~
g;(x;t) := Re(t)
{
1-
( IX I,,);2:r}2
R(t) +'
ek,,-lppe 2 2 k"-I~ -L
g;t = - Re+l(t) F + p-l ~+l(t)Fllzllp-l,
2p k~ (IXI);2:r x
Dg; = - p _ 1 ~(t)F R(t) lxi'
IDg;I,,-2 Dg; = _ ( ~ )
,,-1 [ kPPe F ],,-1 -=---
p- 1 Re(t) R(t) ,
Setting
204 VU. Harnack estimates and extinction profile for singular equations
8. Space-time configurations
Locally bounded weak solutions of (1.1) are locally R;lder continuous in the in-
terior of their domain of definition. "ripE (0,1). This is the content of Theorem 1.1
of Chap. IV. The proof consists of controlling the essential oscillation of a local
solution over a family of nested and shrinking cylinders. Such a control is estab-
lished in Proposition 2.1 of Chap. IV. by working with cylinders whose 'space
dimensions' are rescaled in terms of the solution itself. As observed in Remark 2.2
8. Space-time configurations 205
of Chap. IV, such a geometry is not the only possible. A version of Proposition
2.1 holds for an intrinsic parabolic geometry where the scaling occurs in the 'time
dimension'. We restate the proposition for such a geometry in the context of (1.1)
and in a form convenient for the proof of the Harnack inequality. Let 1.£ be a local
weak solution of (1.1). Fix (x, t) E nT and suppose that we can fmd a cylinder of
the type
(8.1) [(x, t) + Q (aoR", R)] == {Ix - xl < R} x {l- aoR"} , ao == (~) 2-",
where A is an absolute constant, R is so small that [(x, t) + Q (aoR", R)] c nT,
and w is any positive number satisfying
PROPOSITION 8.1. There exist constants eo, 1/ E (0,1) and C, A > 1 that can
be determined a priori depending only upon N and p, satisfying the following.
Construct the sequences Ro = R, WO = W
where we assume that p is so small that Q4p(X o, to) c nT . The change of variables
x-xo
x~--,
p
Q+ == B4 x [0,4"),
Denoting again with x and t the new variables, the rescaled function
{
Vt - div IDvl,,-2 Dv =0 in Q,
v(O,O) = 1.
To prove the theorem it suffices to determine constants c and "Yo in (0, 1), depend-
ing only upon N and p such that
M.,. == supv,
Q..
Here 6 E (0, 1) is a small number to be chosen later and has the effect of rendering
'/lat' the boxes Q.,..
9. Proof of the Harnack inequality 207
Remark 9.1. This construction is similar to that in the proof of Theorem 2.1 of
Chap. VI. The cylinders QT however are 'thin' in the t-dimension. Also the ex-
ponent of (1 - T) in the definition of NT is fixed and depends on the singularity
of the p.d.e.
For T = 0, we have Mo = No. Moreover as T / ' 1
and
since v E L~(Q). Therefore the equation MT = NT has a largest root, say To,
which satisfies
Since v is HOlder continuous in Q, it achieves the value MTo at some point (x, f) E
QTo and
LEMMA 9.1. There exist a positive number e that can be determined a priori only
in terms of N and p, such that
OJ -> ~(I-'T.
v(x , f' 2 0
)-~ , 'v'lx - xl < e(1 - To).
Remark 9.2. The proof employs the estimates of Lemma 5.1 in the form (5.3).
Therefore e "\. 0 as p /' 2.
PROOF OF LEMMA 9.1: Construct the box
- I-To
(x, f) + Q4R == {Ix - xl < 4R} x {t - 4, fl, where 4R = -2-'
Apply to such a box the estimate (5.3) with the appropriate change of variables to
obtain
(9.3) vex, t) ~ co(1 - To)-r-;, 'v'lx - xl < e(1 - To), 'v'6 ~ t ~ 26.
p = e(l- To).
(x-i)"-R(t.l)
~ __+-______________-,2
·&to
Figure 9.1
vex, t) ~
~(l-TO)-~
[lie + 1](
{
1-
(ae
ae +
);;!T}2+
1
== co(l - To)-~.
The location of t in the box Q".o is only known qualitatively. However, as (t -l)
ranges over [6, 36], the intervals [t + 6 < t < t + 36] have the common intersection
[6 :5 t :5 26] and the lemma is proved.
Remark 9.3. The number 6"" 0 as p / 2. This follows from Remark 9.2 and the
choice of 6 above.
(9.4)
I/t
( X-x.~)
3'3P'
introduced in (6.3), in the annular cylindrical domain
210 VII. Harnack estimates and extinction profile for singular equations
k = co(l- To)-r-;,
where Co is determined in Lemma 9.2. The parameter JJ here can be chosen by
imposing
11'-1 e1'
i.e.• u<---
r- - 2-1' 31'.
Co
Wechoose
1 11'-1 e1' }
JJ = min { 4; ~-1' 31' '
and pick (J according to the second of (6.4). By further restricting either JJ or the
number 6 of Lemma 9.2 we may assume that (J =6. The function t[I in (9.4) vanishes
for Ix - xl=3 and fort=6. Moreover for Ix - xl =e(l - To) and 6 <t~26.
t[I (
X - x
-3-' 3P
t - 6) 5 Co(l - To)-r-p
-I!.-
5 v(x, t),
by Lemma 9.2. Therefore by the comparison principle. we have for t = 26 and
Vlx-xl<2
IIvlloo,Qo(z,l) ~ 2P (1 - To)-P
Therefore Qo (x, t) satisfies the space-time configuration (8.1)-(8.2). It follows
that
(10.2)
{
Vt - div IDvl p - 2 Dv =0 in Q.
v(O,O) = 1,
We first prove that there exist a quantitative constant C = C (N, p). such that
1 -
(10.3) C ~ v(x, c) ~ C, v Ixl < 1.
By the Harnack inequality (1.3)
(10.4) v(x, c) ~ 'Yo, Vlx/ < 2,
for a quantitative constant 'Yo = 'Yo (N, p). This proves the estimate below in (10.3).
For the estimate above we require the following lemma.
LEMMA 10.1. There exists a quantitative constant" E (0,1) depending only
upon N and p, such that
and in particular v(O, -c} = lho. Since v is HOlder continuous, the set
{x I u(x,-c} <2ho}
is non-empty and contains a ball about the origin. We claim that in particular it
contains the ball B 2 f/' where
(277}P = (~) 2- p •
If not, there would exist some x E B2f/ such that v(x, -c} = 2ho. It follows that
the ball
Ix - xlP < [v(x, _c}]p-2 = (277t
covers the origin, and (10.6) for x=O gives
-2 = v(x,-c) $ 1
-.
~o ~o
We return to the original coordinates and write the estimate above in (10.3) as
On the other hand the estimate above in (10.3) for x = x. gives u(xo, to} $
Cu (x., t.). Combining these last two estimates proves the bound above in (10.1)
and the proposition follows.
to - t. = c[u(x.,t.)]2- p RP,
The defmitions of t. and R give
c [u(xo, t o)]2- p r P = to - t. = c [u (x., t.)]2- p RP.
By Corollary 1.1, u (x., t.) ~ ')'u(xo, to). Therefore
2..=l
R ~ ')' p rP == p/T/.
Applying Proposition 10.1 with such a choice of the point (x., t.) and radius R
proves the theorem.
Theorem 1.1 and its proof is taken from [44]. The form of Theorem 1.2 was con-
ceived by Nash [84], who believed it to be true for solutions of the heat equation.
Moser [83] pointed out that (1.7) is not dilation invariant for solutions of the heat
equation. It becomes scalar invariant in a specific intrinsic geometry. The results
adapt to equations of porous medium -type and its generalisations (see [44n. In the
context of the plasma equations estimates of the rate of extinctions were derived
by Berryman-Holland [13,14]. Proposition 3.1 is due to &nilan and Crandall [9].
The estimates of §§ 4 and 5 are taken from [42]. The subsolution tV of §6 appears
in [44]. The subsolution ~ of §7 is a modification of a subsolution introduced in
[4]. It is natural to ask whether an intrinsic Harnack estimate continues to hold for
non-negative solutions of p.d.e.'s with full quasilinear structure. This is the case if
p = 2 and it remains an open issue for degenerate (p> 2) and singular (1 < p < 2)
equations. A step in this direction is in [29]. It is shown that Theorem 1.1 holds
true for non-negative weak solutions of
Vt - (IDvIP-2aij(X,t)u:J:JI:', = 0,
where (x, t) --+ aij (x, t) are only bounded and measurable and the matrix (aij) is
positive definite.
VIII
Degenerate and singular parabolic
systems
1. Introduction
We turn now to quasilinear systems whose principal part becomes either degener-
ate or singular at points where IDul =0. To present a streamlined cross section of
the theory. we refer to the model system
(1.2) I
n
Uirpi{X,T)dXI
t2
tl
+ II
t2
tin
{-Uic,oi,t + IDuIP-2Dui·Dc,oi} dxdT=O,
for all intervals [tl, t2] C {O, T] and all testing functions cP == (cpt. CP2, ••. , rpm)
satisfying
(1.3) CPi E W,!;; (0, T; L2{n)) n Lfoc (0, T; wJ,p{n)), i = 1,2, ... , m.
For these we derive local sup-bounds on the modulus of the solution lui and its
space gradient IDul and establish the estimate
for some a E (0, 1). This is the focal point of the theory. Weak solutions of elliptic
systems in general are not continuous everywhere within their domain of defini-
tion. We refer to [48J for counterexamples and an account of the theory. Solutions
of (1.1) are regular everywhere in nT because of the special nature of the sys-
tem. If u solves (1.1), then the function IDul 2 is a non-negative subsolution of
a parabolic p.d.e. (1) It is precisely such a property, which for elliptic systems is
called 'quasi-subharmonicity' ,(2) that permits one to prove (1.4) everywhere in
nT.
These estimates can be extended up to t = 0 if the system in (1.1) is associated
with a smooth initial datum 110. They also carry over to the lateral boundary of
nT if (1.1) is associated with homogeneous either Dirichlet or Neumann data on
ST == an x (0, T). If the data are not homogeneous, the theory is fragmented and
incomplete. In the case of non-homogeneous Dirichlet data, we will show that
Ui EC6 (nx (e, T» for arbitrary 6 E (0,1), 'rIe E (0, T),
provided p > max {I; J~2}' However the key estimate (1.4) is not known to
hold in such a case, and it is a major open problem in the theory.
The C 1 ,o regularity (1.4) requires a preliminary estimation ofthe type
(1.5) IIDulloo,K: :5 const, IC a compact subset of nT.
1be degenerate case p> 2 and the singular case pE (1, 2) are rather different with
respect to such an estimate. The function class in (1.1) implies that(3)
(1.6)
If P > 2, such integrability suffices to establish (1.5). If 1 < p < 2, the sup-bound
(1.5) can be derived only if further 'integrability' is assumed on lui. Precisely,
(1.7) lui E L1oc(nT ), where r ~ 2 satisfies Ar==N(p - 2) + rp > O.
This is analogous to the condition imposed in Theorem 5.1 of Chap. V. It implies
(1.4) and in addition
(1.8)
nature of the p.d.e. some global infonnation is needed. This is not related to sys-
tems. Indeed it occurs also in Theorem 5.1 of Chap. V to establish a sup-bound for
solutions of a single equation. Since our estimates involve u and DU the global t
infonnation needed regards both the solution and its space gradient. Let r ~ 2 sat-
isfy (1.7) and let U be a local weak solution of (1.1) for p E (1,2). We assume
that
(1.10) ~1J,'
at - I
div A (i) (x , t , Du) = B(i) (x , t , u , Du) in nT,
i = 1,2, ... ,m,
(S3)
m
(85 ) 2: IB(i)1 ~ CllDul p- l + !P2,
i=l
where Ci , i =0,1. are given positive constants and !Pi, i = 0,1,2, are given non-
negative functions satisfying
N +2
!Po + !PI#r + !P22 E L q,oc (n )
uT, q > -2-'
Remark 1.1. The structure condition (82 ) is somewhat fonnal since there is no
stipulation that Ut.",,,,,,; have meaning at all. More correctly it should be written
with Ut,,,,u:; and DUi''''i replaced by tensors ~t,k,j. Neverthless we prefer the for-
mal but suggestive fonn of (82 ).
We will develop the main points of the theory for the model system (1.1) and
indicate later how to modify the arguments to include (1.10).
(2.1) sup
[(so,t o )+Q(0'9,O'p»)
lui <
-
'Y
(1 -
(Ojpp)lIE
CT)(N+p)/E
(f! lu IP - 2+£dXdT)
(so,to)+Q(9,p»)
liE
A (~);!J .
THEOREM 2.1 (THE CASE l<p<2). Letubealocalweaksolutionof(l.l)
for 1 < p < 2. Assume moreover tlult .
and that (1.9) holds. There exists a constant 'Y depending only upon N, p, m and
r such thatforevery cylinder [(x o, to) + Q (9, p)] c nT andfor every C1E (0, I),
(2.3)
(2.4) lul=w.
PROPOSITION 2.1. Let u be a local weak solution of the system (1.1) in nT,
and let f(·) be a non-negative, bounded, Lipschitz function in R+. There exists a
constant 'Y='Y(N,p, m), such that
(2.5) V(x o, to) E nT Vp, 9> 0 such that [(xo, to) + Q (9, p)] c nT
sup I (1~f(8)d8) ("(x, t)dx
to-9~t~O 0
(zo+Kp)
PROOF: The weak fonnulation (1.2) can be rewritten in tenns of Steklov aver-
ages,as
Without loss of generality we may assume that (x o, to) coincides with the origin.
In (2.6), take the testing function
We add over i = 1,2, ... , m and integrate in dt, over the interval -() $ t $ 0, to
obtain
j! f(t;/(3)ds) ,'dxdT
-8 Kp
t
+ J J [lDulp-2Dudh·Dui,h!(luhl) (PdxdT
-8K p
t
We perform an integration by parts in the rust integral and then let h - t O. The
various limits are justified since IDul E Lfoc(lh) and lui EC,oc (0, T; L~oc(l1T».
This gives
(2.7) sup
-8<t<O
- -
J(
Kp
10f~J(S)dS) (P(x, t)dx
+ J JIDulP !(w)(PdxdT + JJIDuIP-2IDwI2W!'(w)(PdxdT
Q(8,p) Q(8,p)
+ -Y(1]) J J wP !(w)ID(IPdxdT.
Q(8,p)
N m N m
IDwl2 = w- 2 L (Ut Ut.z;)2 $; w- 2 L u~ L L U~.z; == IDuI2.
;=1 l=1 ;=1 t=1
Therefore IDulP ~ IDwI P • Combining these estimates in (2.7) proves the propo-
sition.
COROLLARY 2.1. The integral inequality (2.5) continues to hold/or non-negative.
non-decreasing functions / in R +. satisfying
sup /'(8) <00, /orall k > 0,
O~s~k
provided
(2.8)
PROOF: °
Fix k> and write (2.5) for the truncated functions
/(8)
fk(S) == { f(k)
for ° $; 8 $; k
for S ~ k.
Letting k -+ 00 gives (2.5) for such an f. The limit of the various terms on the left
hand side follows from Fatou's Lemma and the limit of the terms on the right hand
side is justified by virtue of (2.8).
(2.11)
where k is a positive number to be chosen. We will work with the inequalities (2.5)
written for the functions (u - kn+l) +, over the boxes Qn. The cutoff function (n
is taken to satisfy
(n vanis~es ~n the parabolic boundary of Qn
(2.14) { (n == 1 m Qn
2n +2 2n +2
ID(nl ~ (1 _ (1)p' 0 ~ (n,t ~ (1 - (1)6.
Set
if (8 - kn+l) ~ e
(2.15) if 0 < (8 - kn+l) < e
if (8 - kn+d ~ 0,
and as a function few) take f~ [(w - kn+l)+]. We put these choices in (2.5) and
neglect the non-negative term involving IDul p - 2 since f;(8) ~ O. Letting e - 0
we obtain
We estimate the two integrals on the right hand side as in (7.2)-(7.5) of Chap. V.
This gives the inequalities
obtain the analog of the recursive integral inequalities (10.3) of Chap. V. The proof
of Theorem 2.1 for the singular case 1 < p < 2 is now concluded as in the proof of
Theorem 5.1 in §16 of Chap. V.
of the type
ipi = Ui,zj f(lDul),
(3.1) j jIDuIP-2ID2UI2dxdT
[(zo,to)+Q(119,l1p)]
where
m N
ID 2 ul 2 == LL
i=1 j,k=1
u~,z;z.·
Moreover
(3.2) Ui,zj ECloc(O, Tj L1oc(n)) , i=l, 2, ... , m, j= 1, 2, ... , N.
PROPOSITION 3.1 (THE SINGULAR CASE 1 < p < 2). Let u be a local weak
solution of the singular system (1.1) in fiT and let the approximation assumption
(1.9) hold. Then
224 vm. Degenerate and singular parabolic systems
2 r. ~ .
IDul E.jl Ui.x;ELloc\O,TjW,~(il»),
12 .
'=1,2, ... ,m, J=1,2, ... ,N,
(3.3) IIIDuIP-2ID2uI2dxdr
[(xo ,to )+Q( 1711,17 p»)
where
Moreover
Ui,x, E Lfoc (0, Tj w,!;:(n)) ,
and there exists a constant 'Y = 'Y( N, p) such that
Finally
This local regularity pennits to derive local energy estimates for Du. To simplify
the symbolism we set
(3.6) v=IDul·
Given a cylinder [(x o, to) + Q (6, p)] c ilT we let' denote a non-negative piece-
wise smooth cutoff function in [(x o, to) + Q (9, p)] that vanishes on the boundary
ofthe cube [xo + K pl. In particular we are not requiring in general that , vanishes
for t=to-6.
3. Weak differentiability of IDul ~ Du and energy estimates for IDul 225
(3.7) 'v' (xo, to) E nT, 'v' [(Xo, to) + Q (6, p)] c nT
f(
t
sup r:/(S)dS) (2 (X, t)dx
to-9~t~O [zo+K~)Jo t
0-
9
+ (p - 2) t ff vP-3IDv.DuiI2/'(v)('2dxdr
1=1 [(zo,t o )+Q(9,p»)
provided
(3.8)
where we let 1111 be so small that n l'7l is not empty. We also let cF denote the
discrete gradient of F ,i.e.,
226 vm. Degenerate and singular parabolic systems
The discrete derivative of (2.6)', with respect to Xj, takes the fonn
(3.9) ~6
at '·U· h -
I,
div [6, ·IDuI P- 2Du·]
I h
= 0'
i = 1,2, ... , m, a.e. nl'll x (0, T - h).
In transforming the term [6j !Du!P-2Dui] , we only specify the Xj variable for
simplicity of symbolism. We have
o
x (UDUi(X; +,,) + (1- U)DUi(Xj»)}du
!
1
Having fixed the point (xo, to) E nT, if!(xo, to) + Q (9, p)] c nTwe may assume,
up to a translation, that (xo, to) coincides with the origin, and then by choosing 111!
and h sufficiently small we may assume that Q (9, p) c nl'll x (0, T - h). We
multiply (3.9) by the testing function
+2 jJ(f.""';/(B)da)
-9Kp
(C. dztlr.
In this equality we first let h '..... 0, while I'll > 0 remains fixed. The various limits
are justified since IDul eLfoc(nT) and ueC, oc (0, T;Lfoc(n»). Making use also
of (3.10) we obtain
16 1 ) t
(3.11) sup I ( rus!(s)ds (2(x,t)dx
-9<t<0
Kp
Jo -
9
+ l
I I (foiLl(j) (u)I,,-2dtr ) IDDj U 2 !(lc5ul)(2dxdr
Q(9,p)
+(p - 2) I I (foiil(j)(U)I"-4Ll(j)(U)'D6jUil~j)(U)6jUidU)
Q(9,p)
+2 II(foI6~1!(S)dS) "tdxdr.
Q(9,p)
228 VID. Degenerate and singular parabolic systems
First we observe that the sum of the fmt two integrals over Q «(J, p) on the left
hand side, is bounded below by
2! (p - 2) ff (Li
Q(8,p)
J1 (j) (u)IP-2d,q ) ID6;ur f (l6ul) (2dxdr.
ff (l
Q(8,p)
iJ1 (j) (u)IP-2d,q ) ID6;u116ul f (l6ul) (ID(I dxdr
$ eff Q(9,p)
(liJ1(;) (u) IP-2d,q ) ID6;u1 2f (l6ul) (2dxdr
+ ff
'YE ( liJ1(;) (u)I P- 2d,q) 16ul 2f (l6ul) ID(1 2dxdr.
Q(9,p)
These remarks in (3.11) give the integral inequality involving discrete derivatives
3. Weak differentiability of IDul1j! Du and energy estimates for IDul 229
(3.12) sup
-9<t<0
Kp
I(10fI6~f(s) dS) (2(x, t) dx t
-
9
+ II (liil
Q(9,p)
(i) (a)IP-2d,q ) IDl6ufl6ulf' (l6ul) (2dxd-r
+(p - 2) II (liil
Q(9,p)
(j) (a)IP-4 (il(i)(a).D6j u) .1~j)(a)6jUid,q)
x DI6ull' (16ul) (2dxd-r
5')' II (li
Q(9,p)
.1(i) (a)IP-2d,q) 16ul 2f (16ul) ID(1 2dxd-r
+7 JJ (l"'1
Q(9,p)
f (8)d}C, dxd7,
for a constant')' = ')'(P). If p > 2, the inequality (3.1) follows from (3.13) by
choosing (, a cutoff function that equals one on Q (aO, a p) and such that
1 1
ID(/5 (1 - alp' 05 (t 5 (1- a)O·
To prove (3.3) for the singular case, we transfonn the last integral in (3.13) by
means of an integration by parts as follows.
230 vm. Degenerate and singular parabolic systems
lfiD2ulPdxdT = II(vP-2ID2uI2)P/2v~dxdT.
Q(9,p) Q(9,p)
Since v1!j! ID 2uI E L~oc(nT). the energy inequality (3.7) follows from (3.12) by
letting '1--+0.
(3.15)
These two facts imply that t --+ Ui,:t&; (t) is weakly continuous in L~oc(n). Indeed
let cP E L2 (K p) and let {'Pn} be a sequence of functions in C~ (Kp) such that
Taking CPn as a testing function in (3.14) and integrating over Kp x (tl, t2) gives
4. Boundedness of IDul. Qualitative estimates 231
! !!
t3
I
lim sup f[Ui'z; (t2) - Ui,zj (td] CPdxl
It 3- t d-..o
Kp
$ limsup
I t 3- t lf-<O
Kp
! [Ui,z;(t2) - ui,z;(td] CPn dx
and integrate over Kp x (tl' t2). By calculations similar to those leading to (3.12)
we obtain
If
Kp
[.'(!,) - .'(',)J "dxl
PROOF: Consider first the degenerate case p > 2. Let Q (6, p) c {h and let (
be a standard non-negative cutoff function vanishing on the parabolic boundary of
Q (6, p). Thus, in particular, (., -6) =0. In (3.7), take J(v) =vP, where P?O is
to be chosen. Proceeding fonnally we obtain
- - Kp . Q(9,p)
o
(4.2) IIIDv£¥r dxdr $ 'Y II (1 + vP+ P) dxdT,
-9K p Q(9,p)
where 'Y = 'Y (N,p, p, (t, D(). These are rigorous if the right hand side is finite.
We apply the embedding Theorem 2.1 of Chap. I to the functions
over the cubes Kp. It suffices to consider the case N > 2. Indeed if N = 1, 2, we
may consider u as a vector field defmed in RN N ? 3, up to a localisation, and
deduce inequalities (4.1 )-(4.2) for it. Let 6 be a positive number to be chosen. Then
by Corollary 2.1 of Chap. I and HOlder's inequality
JJ....•..
Q(8.p)
"dzdT $ JJIv.'¥ 'I'dzdT C:~r<o
Q(8.p) ~
J.'If ,'(x, t)JJ
- - Kp
1N
Choosing 6 = 2¥1 and combining this with (4.1)-(4.2) gives the recursive in-
equalities
for a constant 'Y = 'Y(N,p,/3,(t,D(). The right hand side is finite for /3 = o.
Therefore IDul E Lr;4/N ({IT). We may now again apply (4.3) with /3 =4/N and
proceed in this fashion to prove the lemma.
We now tum to the singular case 1 < p < 2. In (3.7) assume that (x o, to) ==
(0,0) and choose a cutoff function ( that vanishes on the parabolic boundary of
Q «(J, p). Take also f (v) = v.8, where /3 ~ 0 is to be chosen. By working with
the approximations claimed by (1.9) we will use the qualitative infonnation that
IDul E L~oc(nT). Our estimates however will be only in tenns of IIDull".Q(8.p).
Proceeding fonnally we obtain from (3.7)
where -y=-y (N,p, {3, (t, D(). Also by a fonnal integration by parts
= IluD(V~DU) v~(dxdr
Q(8.p)
+ II u vP.±f=! Du v~ D( dxdr
Q(8.p)
We combine this with (4.4) and make use of the Schwartz inequality to arrive at
234 vm. Degenerate and singular parabolic systems
for a constant
'Y='Y (N,p, (j, Ct, DC, lI u ll oo ,Q(6,p») .
This inequality is indeed rigorous as long as the right hand side is fmite. We apply it
fIrSt with (j =p-l to deduce that IDul E L~l (nT ). with bounds only dependent on
II Dullp,Q(fI,p) . Then we apply it again with {j =p to deduce that IDul E L~2 (nT ).
Proceeding this way proves the lemma.
,
LEMMA 4.2. Let u be a local weak solution of(1.1). Moreover in the singular
case 1 < p < 2 let the approximation assumption (1.9) be in force. Then
PROOF: Consider first the degenerate case p> 2. Let Q (6, p) c nT and let Qn
and Qn be the family of cylinders introduced in (2.9)-(2.12). Let also k,. and Cn
be respectively the increasing levels defined in (2.13) and the cutoff functions in
Qn introduced in (2.14). We put these choices in the energy estimates (3.7) and as
a function f (v) take
f(v) :: (v - kn+l)r 2 •
By virtue of Lemma 4.1 and Corollary 3.1. such a choice is admissible. The teon
involving D 2 u is estimated below by
2 22
I I v"- ID uI f(v)C2dxd'T ~ (~) P-jIID (v - 12
kn+l)! C!dxd'T.
Qft Qft
1bese choices yield the inequalities
+ kP - 2 !!ID
Qft
[(v - kn+1)l Cnr dxd'T
(4.7) Yn :: !!(V-kn)~dXd'T.
Qft
4. Bc..undedness of IDul. Qualitative estimates 235
Then we have(l)
(if>IJ- - oJ dxdT)
x "-+1)+> .to
where.\2 == N(P - 2) + 2p. These are the key recursive inequalities needed to
derive a quantitative sup-bound for IDul. We will use them first in a qualitative
way as follows. First let A denote a lump constant depending upon (J I (I, P and the
quantities
We have also(1)
(4.10) II
Q..
vPX[(v - kn+d+>O] dxdT ~ -y2np Yn·
Therefore
y;+wh ~ A y1+wh.
1berefore
IIDulloo,Q(a9,ap) ~ max{l; k}
~ 1 + A¥b(N+2)2/p IIDullp,Q(9,p).
We now tum to the singular case 1 < p < 2. The starting point is still the energy es-
timate (3.7) where we choose Qn, Qn, (n and the levels kn as before. As a function
1(·) we take
if r > 2
if r = 2,
where Ie (.) is the Lipschitz approximation to the Heaviside graph, introduced in
(2.15). After we let e -+ 0, the first term on the left hand side is bounded below for
all t E ( -fJ, 0) by the quantity
K
f(l tJ
k"+!
s2- p s (s - kn +1r- 2 dS) (!(x, t)dx
+
p"
r(~dxdr.
Qn
~ r~ IIID (v - kn+d12
Qn
Combining these estimates in (3.7) we arrive at
ff
Q..
l1r+{2- p)X[(v - kn+d+>0] dxdr
Sn+l <Abnk-msl+rtr
_ + n ,
S!+~ ~ A S!+rtr .
The proof is now concluded as in the degenerate case.
(5.1) sup
.
IDul ~ (
'YV(9/p2)
)(N+2)/2
()~
ff IDul dxdrI'
[(zo.to)+Q{...8 ....p») 1- 0-
(zo.to )+Q(8.p»)
"9(p2)i6 .
THEOREM 5.2 (THE SINGULAR CASE 1 < p < 2). Let u be a local weak
solution of the singular system (1.1) and let the approximation assumption (1.9)
be inforce. Moreover let r ~ 2 satisfy
(5.3)
Q..
Therefore (4.9) yields
Yn +1 ~
-ybn
2
(
sup v
)"-2 _~k N+2
l+~ .
Yn
[(1 - u)p] Q(9,p)
It follows from Lemma 4.1 of Chap. I that {Yn } neN - 0 as n - 00, if k is chosen
from
(N+2)/2 (N+2~(p-2)
k>'2/ 2 == ( 'Y
b(N+2)/2)
2
( )
sup v JrJrv"dxdr.
[(1 - u)p] Q(',p) Q(',p)
240 vrn. Degenerate and singular parabolic systems
X
(
IlvPdxdr )
Q(9,p)
If O"E (0,1) is fixed. consider the family of boxes Q(n) :;:: Q (On, Pn). where
Po:;:: O"p
and for n= 1, 2, ...
n n
By construction.
Q(O) :;:: Q (0"0, O"p) and Q(oo) == Q (0, p) .
Set
Mn = esssupv
Q(n)
and write (5.5) for the pair of boxes Q(n) and Q(n+1). This gives
(5.6)
where
B :;:: ,,(>'2/ 4
[(1 - 0")p](N+2)/2
(f!vPdxdr) 1/2
'
d 2(N+2)/2.
=
Q(9,p)
Qn
~ 2 (s~~v
nr
r- Qn
p
Sn.
5. Quantitative sup-bounds of IDul 241
Qn v - - !!...
>
sup 2 p tior all n = 0 , 1 , 2 , ....
p2'
Otherwise there is nothing to prove. Taking this into account, we rewrite (4.13) as
---.l!.±L
sup v< 'Y (
SUP
Q(8,p)
V2-P) 2(,+2-p)
(5.7)
Q(<T8,<Tp) - (1 - 0') r~t!p (J
1!(r+2-p)
( )
X jjvrdXdT
Q(8,p)
The proof is now concluded with an interpolation process as in the degenerate case.
This is possible if the power of the term sUPQ(8,p) von the right hand side of (5.7)
is less than one. Since
(2 - p)(N + 2) =1 _ Vr ,
2(r + 2 - p) 2(r + 2 - p)
this occurs if (5.2) holds. We also remark that the interpolation process applied to
(5.7) generates a dependence of the type of l/vr in the constant 'Y(N,p, r) appear-
ing in (5.3).
sup
Q(<T8,<Tp)
v <
-
'Y ..[(iTiJ)
(1 - 0')(N+2)!2
(sup
Q(8,p)
v) ~ (n Q(8,p)
Vp-2H dXdT)1!2
....1....
A( ~y-3
Such an inequality can be interpolated as long as e E (0, 2] and proves the follow-
ing:
242 vm. Degenerate and singular parabolic systems
THEOREM 5.1' (THE CASE P > 2). Let u be a local weak solution of the
degenerate system (1.1). Then for every E E (0,2]. there exists a constant 'Y =
'Y(N,p,E) such that
V(xo, to) E nT, v [(x o, to) + Q (8, p)] c nT, Vq E (0,1),
)
l/~
(5.8) sup
[(zo,t o)+Q(1J'9,lJ'p»)
IDul $ (
'Y
1-
(8/,r) l/~
q
)(N+2)/~
(
H IDul
,,-2+~
[(zo,to)+Q(9,p»)
dxdT
A ( ,r)~
8 .
Also, (5.3) can be interpolated. We rewrite it for (xo, t o):= (0, 0) and in the fonn
sup
Q(1J'9,lJ'p)
V <
- (1- q)2(N+2)/vr
sup
(Q(9,P) )
V
H
Q(9,p)
vqdxdT
A (!..)"!;
p2 .
This can be interpolated as long as qE (0, r] satisfies 2(::q) < 1. This occurs if
(5.9) IIq :=N(p - 2) + 2q > O.
The interpolation process gives
THEOREM 5.2' (THE SINGULAR CASE 1 < p < 2). Let u be a local weak
solution of the singular system (1.1) and let the approximation assumption (1.9)
be in force. Moreover let r ~ 2 satisfy (5.2). Then for every q E (0, r] satisfying
(5.9) there exists a constant 'Y='Y(N,p, r, q) such that
(5.10)
Remark 5.3. The constant 'Y(N,p, r, q) in (5.10) tends to infinity as IIq -0.
6. General structures 243
Remark 5.4. Estimate (5.10) is fonnally equivalent to (5.3), the only difference
being that q is not required to be larger or equal to 2. The only condition is that
(5.9) be verified. In particular, (5.10) holds for q=p provided
2N
(5.11) p> N +2'
6. General structures
Let U be a local weak solution of the non-linear system (1.10) subject to the struc-
ture conditions (81 )-(86 ), The local boundedness of U can be established as in the
proof of Theorem 2.1. The main modification occurs in the handling of the 'per-
turbation terms' l(Ji, i = 0, 1, 2. These contribute to the energy inequalities (2.5)
with an extra tenn of the type
Given the choice (2.15) of 1(')' these terms are estimated as in the sup-bounds
established in Chap.V for general equations. (1) The weak differentiability of the
tenn IDulp-2 Du follows from the structure conditions (81 )-(82), We proceed as
before by working first with the discrete derivatives. All the tenns involving the
'derivatives' 6j Ui,zc' are dominated by the tenns arising from the right hand side
of (82 ), (2) Following the same process of §3 yields local energy estimates similar
to (3.7) with constants 'Y ='Y( N, p, Co, C1 ) and with the right hand side augmented
by the extra integral
(1) See for example Theorem 3.1 of Chap. V and its proof.
(2) See also Remark 1.1.
244 vm. Degenerate and singular parabolic systems
These energy estimates imply that IDul E L~(nT) be the same iterative tech-
niques of §4. The 'perturbation terms' are dealt with as in Chap. V.
7. Bibliographical notes
In the case of a single equation the estimate (1.5) up to ST has been established
by Lieberman [68]. Estimates in the norm cl,a up to ST for Dirichlet data, are
not known even for elliptic systems. Results for a single elliptic equations are
due to Lieberman [69] and Lin [72]. The general structures of §1-(l1) have been
introduced first by Tolksdorff [95]. The arguments of finite differences to prove
that IDul,-2Ui,Xj is weakly differentiable were introduced by Uhlenbeck [99] in
the context of elliptic systems. The sup-bound of Theorem 2.1 for the degenerate
case p > 2 is new. The same theorem for the singular case 1 < p < 2 is due to
Choe [31]. The qualitative Lemmas 4.1 and 4.2 appear in [36] for all p > ~~2'
and in Choe [31] for all p > 1 provided (1.7) holds. Even though some quantitative
estimates of the gradient appear in a variety of forms in [27,36,37], Chen [25] and
Choe [30], the precise form of Theorems 5.1 and 5.2 as well as their interpolated
version in §5-(lII}, seems to be new.
IX
Parabolic p-systems: Holder continuity
of Du
THEOREM 1.1. Let u be a local weak solution 0/(1.1) o/Chap. VI/l. Moreover
if 1 < p < 2 let the approximation condition (1.9) be in /orce. Then
(x, t) -Ui,zj (x, t) E Ct!c(nT), for some a E (0,1),
for all i = 1, 2, ... ,m and all j = 1, 2, ... ,N. Moreover/or every compact subset
/C OinT. there exist constants a=a(N,p) E (0,1) and-y=-y (N,p, II Dull oo,K:) >
1. such that
Remark 1.2. The functional dependence of'Y upon IIDulloo,K will be given in
§§3 and 4.
there holds
(Du)" = ff Dudxdr.
Q'''R('')
there holds
1. The main theorem 247
THEOREM 1.2. Assume that the cylinder Q R (/J) satisfies (1.2) for some /J > 0.
There exist constants "Y> 1 and a E (0, 1) that can be determined a priori only in
terms of N and p, such that
IIDulloo,QIlC"') = /J.
Such an equation has finite roots since IIDulloo,QIlC",o) > /Jo, and
As /J increases, the cubes {Ixl < /J ~ R} shrink. Therefore there exist some /J for
which (1.2)' holds.
Remark 1.3. The previous propositions could be stated and proved in the geom-
etry of the boxes (1.2)'. Indeed setting /J~ R=r permits one to recast the 'space
scaling' of (1.2)' in terms of the 'time scaling' of (1.2).
248 IX. Parabolic p-systems: HOlder continuity of Du
(2.1) {
1-'0 = 1-', Ro = R and for n = 1,2, ... ,
I'n+l = TJl'n, Rn+l = CoRn,
where TJ and 0' are the numbers claimed by Proposition 1.2 and
(2.2)
Since TJ E (!, 1). we have Co E (0,1) for all p > 1. Suppose the assumption (1.5)
bolds with R = Ro and I' =1'0' Then
From the definitions (2.1) and (2.2) it follows that Rl < Ro and
R21 a 2..,p-2
'r R20 _ (0')2 R20
I-'f-2 = - 4 - TJP-21-'~ 2 = '2 1JP-2 •
This implies that the cylinder QRl (I-'d is contained in QtTRo (1-'0) and
sup IDul ~ 1-'1.
QR1(",d
Therefore QRl (I'd satisfies (1.2) an4 if the assumption (1.5) of Proposition 1.2 is
verified again for such a box we have
2. Estimating the oscillation of Du 249
QRn (I-'n) , n = 0,1,2, ... ,no - 1 for some positive integer no·
Then
One verifies that Co < '1 for all p > 1, and consequently al E (0,1). We rewrite
(2.3) as
Suppose now that the assumption (l.5) of Proposition l.2 fails for no. We call
Rno the switching radius. Then for the box QRno (I-'nJ the assumption (1.3) of
Proposition 1.1 holds and we conclude that
< i 2
_KI-'n . 1 2
o ' t=, , ...
Writing
'Y = 2 (K + cS-(N+2») .
Therefore {(DU)iheN is a Cauchy sequence whose limit we denote with Du( x o , to).
To motivate this terminology we recall that our arguments are carried over an ar-
bitrary cylinder [(xo, to) + QR(I-')] with vertex at (xo, to). Therefore if no is the
switching radius of the box [( x o , to) + QR(1-')], the limit ofthe averages,
HDUdxdr,
[(zo,tO)+Q'iRno (~no)]
250 IX. Parabolic p-systems: Hl)lder continuity of Du
n
is Du(x o• to) for almost all (x o• to) E T • It follows from (2.6) that
2. 2
1Du(xo•to) - (Du), 1 ~ 'Y 1t'lJno ' i = 1.2•....
Fix 0 < p< Rno and denote with (Du)p the integral average of Du over Qp(lJnJ.
Let i be a positive integer such that
(2.7)
and estimate
Therefore
+ 21 (Du)p - (DU),r
::; 'Y(6) It'1J!0'
It follows from (2.7) and (2.9) that
Let 2Qo =min{Ql; Q2}. Then combining (2.9) and (2.4) we conclude
LEMMA 2.1. There exist constants 'Y> 1 and Q o E (0. 1) that can be determined
a priori only in terms 0/ Nand p. such that/or almost all (x o•to) E nT such that
[(xo. to) + QR(IJ)] c nT. and/orallO<p~R. there holds
(2.10)
Moreover
Qp(""o)
Remark 2.1. The lemma holds also in the geometry of the boxes [(xo. to) + QR(IJ)]
introduced in (1.2)'. Indeed we may set 1J2.j! R=r and work within the cylinder
[(xo. to) + Qr(IJ)]. We arrive at a version of (2.10) that reads
(2.10)'
3. RUder continuity of Du (thecasep>2) 251
Returning to the geometry of [(%0. to) + QR(P)] proves the assertion. Analogous
considerations bold for (2.11).
P = II Dull OO,DT •
This is no loss of generality. by possibly working with another compact set
K:,' satisfying K:, C K:,' c nT. and
(3.1)
LEMMA 3.1. Let (3.1) hold. There exist constants 'Y > 1 and Q E (0.1) that can
be determined a priori only in terms 0/ Nand p such that/or all pairs (x o • t.) E
K:" i=O.l.
PROOF: Let Rn, be the switching radii of the cylinders [(xo. ti) + QR(P)] and
introduce the two boxes
Qi = [(xo. t.) + QR", (Pn,)]
= {Ix - x.1 < PRy Rn, } x {t. - R~,. ttl .
Assume first that
(3.3)
252 IX. Parabolic p-systems: H6lder continuity of Du
Set
(Du)c; == H
c;
Dudxdr, i = 0,1,
and estimate
To estimate the last term we add and subtract Du(x, t) where (x, t) E Co n Cl , and
then take the integral average over such intersection, i.e.,
~'YI' ( ~)ao
R
To estimate the second integral. let fJ be a small positive number to be chosen and
assume that
3. .Hl)lder continuity of Du (the case p> 2) 253
(3.5)
H
Co nel
IDu(x, t) - (Du)co Idt
N(P-2)/2H
:5 "I ( ~:: ) IDu(x, t) - (Du)Co Idt
Co
Therefore if (3.3) and (3.5) hold, the assertion (3.2) follows by taking {3 =
Qo/ N (p - 2) and then choosing Q =Qo/2. If (3.5) is violated,
( ~) Q /N(p-2) o
IDu(x o , to) - DU(Xl, tdl :5 21' R 0 ,
Q. == (x o, ttl + QR". (I'n.) == {Ix - xol < I':f Rn. } x{tl - R!.,tl}'
Since we have
2(tt - to) :5 min {R!o j R!.} ,
the box Q .. will now play the same role as the cylinder Ql in the case (3.3). The
proof is now concluded as before, observing that for Q.. the two inequalities (2.10)
and (2.11) hold true.
2S4 IX. Parabolic p-systems: H6lder continuity of Du
LEMMA 3.1'. There exist constants 1> 1 and Q e (0,1) that can be determined a
priori only in terms 0/ N andpsuch that/or every pairo/points (x"to)elC, i=
0,1,
(3.6)
LEMMA 3.2. Let (3.6) hold. Thereexistconstants1> 1 andQE (0, 1) that can be
determined a priori only in terms 0/ Nand p. such that/or all (x" to) E /C, i =0, I.
(3.7) I
IDU(XI, to) - Du(xo, to) $ 11J (Ixo ~ XII) Q •
PROOF: Let R,." be the switching radii corresponding to the two boxes
[(Xi, to) + QR(IJ»). and construct the two cylinders
. Qi == [(x" to) + QR", (IJn.)]
== {Ix - xii < R,.,,} x {to -IJ!;-P R!" to} •
Consider separately the following two cases:
(3.8)
(3.8)'
Set
(Du)c, == if
c,
Dudt,
and estimate
+ I(DU)Cl - (Du)c I· o
The proof now proceeds as for the HOlder continuity in t with minor cbanges.
LEMMA 3.2'. There exist constants "y > 1 and Q E (0, 1) that can be determined a
prior; only ;n terms of N and p such that for every pair ofpoints (Xi, to) E /C, i =
0,1.
(3.7)'
If
( IXl - Xo I )
0/2
2.f!
JJ :5; dist (/C; r) ,
THEOREM 1.1' (THE DEGENERATE CASE p> 2). Let u be a weak solu-
tion in nT of the degenerate system (1.1) of Chap. VIII, and assume that I-' =
IIDulloo.oT < 00. There exist constants 'Y > 1 and Q E (0,1) that can be deter-
mined a priori only in terms of Nand p such that, for every compact subset fC of
nT •
(1.1') IDu(xo, to) - Du(xl. t 1 )1
Vi
Ui
== -, t
.
= 1, 2 , ... ,m, and T = tl-'p-2.
IJ.
[(Xo, ti) + QR(I-')] == {Ix - xol < I-'Ej! R} X {ti - R2,ti}' i = 0,1.
The box [(xo, t l ) + QR(I-')] intersects [(xo, to) + QR(I-')] if (h -to) < R2. More-
over they are contained in flT if
(4.2)'
Arguing as in the proof of lemma 3.2' and bY.possibly redefining the constants 'Y
and a, we obtain
LEMMA 4.1'. There exist constants 'Y > 1 and a E (0,1) that can be determined a
priori only in terms of Nand p such that for every pair of points (xo, ti) E J(" i =
0, I,
[(Xi, to) + QR(I-')] == {Ix - Xii < R} x {to -1-'2-p R2, to}, i = 0, 1,
be two boxes satisfying (1.2). The box [(xo, to) + QR(I-')] intersects Xl if Ixo -
XII < R. Moreover they are contained in flT if
(4.4)
LEMMA 4.2. Let (4.4) hold. Thereexistconstants'Y> 1 andaE (0, I) that can be
determined a priori only in terms of N andp,such thatforall (Xi, to)EK:, i=O, 1.
and the one obtained by taking the derivative with respect to Xj' i.e.,
We also let 'Y='Y(N, p) denote a generic positive constant that can be detennined
a priori only in tenns of the indicated quantities.
5. Some algebraic lemmas 259
LEMMA 5.1. There exists a constant 'Y = 'Y(N,p). such that for every vector
V E RNxm. and/or all p > 1.
LEMMA 5.2. Let 1 < p < 2. There exists a constant 'Y = 'Y(N,p) such that/or
every vector V E R Nxm.
Remark 5.1. These lemmas are algebraic in nature and could be stated for any
pair of vectors U and V, provided (5.3) is replaced by
(5.3)'
Also in (5.4) the number (1'-2)/2 could be replaced by any number and in (5.5)-
(5.7) the number (1'-2) could be replaced by any negative number.
PROOF OF LEMMA 5.1: By calculation,
IIDulY Du -lvIYVIIDu - VI
2: 1(IDuI'i'Du-IVI'i'V, DU-V) I
/ f1d
= \10 ds ISDu + (1 - s)VI
Y (sDu + (1 - slY) ds, Du - V
)
= 10f~IsDu + (1 - s)VI
Y IDu - VI 2ds
1o~IsDu + (I - s)VI
Y ds ~ {lDul + IVI) Y ,
and the lemma follows in this case. If p > 2, assume for example that IDul > IVI.
Then
260 IX. Parabolic p-systems: Hi)lder continuity of Du
These inequalities in (5.8) prove (5.5). If (5.9) is false. its converse gives the two
inequalities
These in (5.8)' imply that the term in braces on the right hand side is bounded
above by an absolute constant. Moreover
IVI,,-2IDu - VI ~ IDul,,-2IDu - VI,,-IIDu - VI 2-".
The two inequalities (5.6) and (5.7) are an immediate consequence of (5.8) and the
assumption (5.3).
5. Some algebraic lemmas 261
We will estimate IHI for all p> 1. For this we first set
(5.13) Wet) == tDu + (1 - t)V, for t E [0,1],
and rewrite (5.12) in the form
1
(5.16)
Remark S.2. The lemma holds for every pair of vectors U and V satisfying (5.3)'.
PROOF OF LEMMA 5.3 (p>2): Assume first thatlVI ~ 21Du - VI. Then
1
(5.18)
t* = IVI
IDu-VI
Then
1
Therefore taking into account (5.19) and (5.3), the lemma follows in this case.
Consider now the case when (5.19) is violated, i.e.,
1 1
1 1
Let V be any vector in RNxm satisfying (5.3). To the system (5.1) we associate
its linearised version
(6.1)
a
at Vi - (IVIP-2 vi ,xI + (p - 2)IVlp-4Vj,k Vj,xt Vi,l) XI '
Let
v == (Vl,V2, ••• ,vm )
and for 0 < p:5 R we let (Dv) p denote the integral average of Dv over Qp (/J).
264 IX. Parabolic p-systems: mlder continuity of Do
8 ( .. )
(6.3) at Vi - a~:~ Vi,z. Zt = 0,
where the coefficients
for two given constants Co < C 1 depending only upon N and p. Therefore it will
suffice to prove Theorem 6.1 for p. = 1. In the remainder of the section we let v be
a solution of (6.3) in QR and let (6.4) hold. Let a denote a multiindex of size lal.
i.e.,
N
a==(al,a2, ... ,aN), ajENU{0},j=l,2, ... ,Nj lal=Laj,
j=1
LEMMA 6.1. There exists a constant 'Y ='Y(N,p) such that for all non-negative
integers m, n and all 0 < p:S R,
6. Linear parabolic systems with constant coefficients 265
PROOF: The system (6.3) is also solved by the vectors W == D';Div. Let (be
a non-negative smooth cutoff function in Qp vanishing on the parabolic boundary
of Qp and such that
Multiply the system (6.3), written for w, by the testing function W(2
and integrate
over Qp. to arrive at (6.5). To prove (6.6). mUltiply the same system by Wt(2
and
integrate over Qp. This gives
The integral involving a~:{ on the left hand side of (6.7) equals
The lemma now follows by applying (6.5) and suitably modifying the scale of the
radii P and p/2.
266 IX. Parabolic p-systems: H61der continuity of Du
LEMMA 6.2. There exists a constant 'Y='Y(N,p) such that/or all 0 < p~ R
(6.8)
PROOF: It suffices to prove the lemma for 0 < p ~ R/2 N +2. Let ( be the standard
cutoff function in QR/2N +1 that equals one on QR/2N +2 and such that
"V("~,QIl/2N+1 ~ 'YR-CN+2>!!lvI2dxdT.
QIl
for a constant 'Y='Y(N,p) and for all 0<p~R/2. To estimate the right hand side
of (6.10) observe that the vectors v z. solve the system
(6.11)
Let W be any constant vector in R Nxm and multiply (6.11) by the testing function
(Wi,z. - Wi,.) (2. where ( is the standard cutoff function iO Q R that equals one
in QR/2. This gives
6. Linear parabolic systems with constant coefficients 267
f
(Dv)p(t) = Dv(x, t)dx,
K,.
VO<p:5R/2, _p2 :5t:50.
Then x - (Dv,(x, t)-(Dv,)p(t» has zero average over Kp. and by the embed-
ding Theorem 2.1 and Remark 2.1 of Chap. I.
Write
LEMMA 7.1. There exists a constant 'Y = 'Y(N,p) such that/or every constant
vector V in RNxm satisfying (5.3),
PROOF: Let ( be a cutoff function in QR(P.) that equals one on QR/2 (p.). and
such that
where
If p > 2. we have
Putting this estimate in (7.2) proves the lemma in the degenerate case. To estimate
J in the singular case 1 <p< 2. we fU'St integrate by parts in the variable Xi. This
gives
7. The perturbation lemma 269
and by (5.6)
12 ~ 'Yp;~2 fflDU - Vl 2dxdr.
QIl(")
Combining these estimates in (7.2) proves the Lemma.
Let 8pQR/2 (p) denote the parabolic boundary of QR/2 (p). Consider the
boundary value problem
LEMMA 7.2. There exists a constant 'Y='Y(N,p), such thatfor all 0< p< R/2
and for every vector V satisfying (5.3),
f flDU - Vl 2dxdr,
QIl(")
where the vectors Hi are introduced in (S.12). From this. subtract (7.4). and in the
weak fonnulation of the system so obtained, take the testing function Ui -Vi. This
is admissible since it vanishes on lJp QR/2 (",). Adding over i= 1, 2, ... , m. gives
where we have taken into account the fact that V satisfies (S.3). Using Schwartz
inequality on the right hand side and then Lemma S.3 to estimate IHI2. we arrive
at
To estimate the right hand side of (7.6) assume first that N ~ 4 so that
a = min{!'
- 2' N
~} = ~
N'
To simplify the symbolism we let r =",2-p R2 /4. We have
By Lemma 7.1
To estimate the last factor in (7.7) we majorise the integrand by means of Lemma
5.1. It gives
N-2
Here in estimating the last term we have used the algebraic inequality
272 IX. Parabolic p-systcms: H5lder continuity of Du
which follows from (5.6) of Lemma 5.2 with p replaced by (p + 2)/2. Therefore
the last factor on the right hand side of (7.7) is estimated by
1V.~~DoI
N-2
j
+ I'p-2 R- 2 jlDU - V 12 dxdT}
QRC",)
QRC",)
where we have also used Lemma 7.1. We now combine these calculations in (7.7)
and then in (7.6) to obtain
QpC",)
~ 1(l~-V'2dzr
x (/ODU + IVI)4(P-2) IDu - V 16 dx) ! d'T
KR/2
'
~ p.2j! sup
-r<t<O
(fIDU
J I
- V12dx) !
- - KR/2
1
By Lemma 7.1
We estimate the last tenn on the right hand side of (7.9) separately for N =3 and
N=2.
K SR./4
1berefore
QR.(")
Combining these estimates in (7.9) and then in (7.6) proves the lemma for N =3.
TbecaseN=2
We apply the embedding Theorem 2.1 of Chap. I with q = 6, Q = 2/3 and
B =1. This gives for a.e. t E (-r, 0)
1berefore
QR(")
We estimate these integrals by means of Lemma 7.1 and combine the calculations
in (7.9) and in (7.6) to conclude that (7.5) holds with a=~.
LEMMA 8.1. There exist constants ~, 6, E E (0, 1) that can be determined a priori
only in terms of N and p. such that ijVo is a constant vector in R Nxm satisfying
(8.1)
Q'R(") QR(")
Vt == HDvdt,
Qu(,,)
+ I fiDV - V l l2 dxdT.
Qu(,,)
By Theorem 6.1
2"Y62 :5 ".
Inequality (8.5) follows from (8.4) and the s1I1Illiness assumption (8.2). To prove
(8.3) write
Vl - Vo =H (Dv - Vo)dxdT
Q,R(")
and
and
LEMMA 8.2. There exist constants K" 6, EE (0,1) that can be determined a priori
only in terms of N and p. such that if V 0 is a constant vector in R Nxm satisfying
(8.1) and (8.2). then there exists a sequence of constant vectors {Vi} el in RNxm.
satisfying
for i = 1, 2, .. "
PROOF: The sequence is constructed inductively by using the procedure of the
previous lemma. To prove that IVil are in the range (8.7), we refer back to (8.6),
H
i.e.
IVi +! - V i l2 ~ 2 (K, + 6-(N+2») IDu - V i l2dxdT.
Q6I a(")
We iterate over i and use again the smallness assumption (8.2) to obtain
where we have used the specific choice of 6 in tenns of It. Choosing now It suffi-
ciently small proves the Lemma.
PROOF: Consider the differentiated equation (S.2) and in its weak fonnulation
take the testing function
Ui,z; (v 2 - k2)+ (2, k = (1- 211)1-',
modulo a Stelclov averaging process. Here ( is a non-negative piecewise smooth
cutoff function in QR(I-') that equals one on QtT R (I-') and such that
1 1-',,-2
ID(I ~ (1- u)R' 0 ~ (t ~ (1- u)W'
After we add over i = 1, 2, ... m and j I =1, 2, ... , N, we arrive at
(9.S) sup, /(,:,2 - k2): (2(x, t) dx
-",3-PR2<t<O
- - Kit
We put this in (9.S) and in the resulting inequality we discard all the non-negative
terms on the left hand side except the integral containing DUi,Zi' This gives
where we have used the structure of (and the intrinsic geometry of QR(J.I.). Also
This is obvious if p > 2. If 1 < p < 2, we observe that the integral is extended
over the set v > (I - 211)J.I.. We estimate below the integral on the left hand side
of (9.6) by extending the integration over the smaller set [v> (I - II )J.I.]. On such
a set, (v 2 - k2)+ ~ IIJ.1.2. These remarks in (9.6) prove (9.4).
Set for all O<p~R and all tE [-J.l. 2- p p2,O]
(Du)p (t) == f
Kp
Du(x, t) dx.
LEMMA 9.2. There ex;sts'(l constant "Y="Y(N,p). such that/or 'all O'E (l,l)
280 IX. Parabolic p-systems: H5lder continuity of Du
x V!IDuI'i'Du-V<tlldz) ~
The last integral is majorised by ''/#£ rn R-Af:r . Therefore integrating this inequal-
ityover [-1l 2- P (uR)2,O] gives
(9.8) f1r [I
Q..R(p)
IDul
¥ 12 'Y 1J2 Vl/(N+l)
Du - VCr) dxdr ~ (1 _ u)2N/(N+l) R
N+2
.
Vet) == Iw(t)l¥w(t),
and observe that
[ [ 2 'Y1J 2 v 1/(N+l)
(9.10) 11 IDu - W(7')1 dxd7' ~ (1- u)2N/(N+l) IQR(IJ)I·
A~1t
Next write
The first integral is estimated in (9.10) and the second is majorised by 21J2 Q R (IJ) vi I.
in view of the 'smallness' condition (9.3). We conclude that
282 IX. Parabolic p-systems: HOlder continuity of Du
for a constant "( = "((N,p). The minimum on the left hand side is achieved for
V == (Du)aR (t). This proves the lemma if p> 2.
The singular case 1 < p < 2
Since Iw(t)1 $ JJ, we have (IDul + Iw(t)l)P-2 ~ 2P- 2JJP- 2. Putting this in
(9.9) and combining it with (9.8) gives
(Du)p == I f DudxdT.
Qp(/J)
LEMMA 10.1. There exists positive constants "(, a, b that can be determined a
priori only in terms of N and P. such that for all u E (i, 1).
(10.1) I f IDu - (Du)aR 12dxdT $ "( JJ2 {(1 :au)b + (1- u)}.
Q"R(/J)
2 "( JJ 2 v1/(N+l)
I f IDu - (Du)aR! dxdT $ (1 _ u)2N/(N+l)
Q"R(/J)
and
$ sup
-"l-P(aR)l<t 8<0
r- - 1 -
lfKtlR
(Du(x,t) - DU(X,8») dx12.
Let (f= (1 +u)/2 and denote with x--+((x) anon-negative smooth cutoff function
in K&R that equals one on KaR and such that
10. Proof of Proposition l.1-(iii) 283
- 2
ID(I ~ (1-i1)R == (l-u)R'
4 1D2-1( ~ (l-u)R'
16
Write
!
K"R
(Du(x, t) - Du(x,s») dx = !
aKR
(Du(x,t) - Du(X,7'»)(2dx
- !(Du(x,t) - Du(x,s»)(2dx.
K.R\K"R
The last integral is estimated above by 'Y( 1 - u )p.RN . To estimate the flJ'St integral
we integrate the differentiated system (5.2) over (7', t). multiply by (and integrate
over KaR. This gives
(10.3) !
K.R
(Ui,Zj (t) - Ui,zj (s) )(2 dx
= Iif ('
aK.R
div ( .,..-' Du;.., + 0:;' Do;) ""dBl·
Thecasep>2
The right hand side of (10.3) is estimated by
+! !IDul~ ID2uldxd7'
S:;R
+ IBill! (!!IDuIp-2ID'U1'''''tt.!!
Qu(,,) )
284 IX. Parabolic p-systems: mlder continuity of Du
The frrst integral is estimated by Lemma 9.1 and the second tenn is estimated
by the 'smallness' condition (9.3) and Lemma 7.1. Combining these estimates in
(10.2) proves the lemma.
The case l<p<2
Estimate the right hand side of (10.3) as follows.
By the sttucture of the cutoff function ( and (5.5) of Lemma 5.2, this is majorised
by
t
We estimate the last integral by Lemma 9.2 and combine it with (10.3) to prove
the lemma.
LEMMA 11.1. Let e E (0, 1) be the number claimed by Lemma B.l. There exists
a number v E (0, i) such that if (9.3) holds, then
(11.1) HIDU -
QR(")
(DU)RI 2dxdr ~ ep.2,
(11.2)
PROOF: Write
II. Proof of Proposition 1.1 concluded 285
The flrst integral is estimated by Lemma 10.1 and the second is bounded above by
")'(1 - q)/J2. To estimate the last integral write
and
H IDu - (DU)RI 2 dxdT :s ")' /J2 { (1 :°O')b + (1 - 0') } .
QR(")
We assume that the smallness condition (9.3) does not hold, i.e.,
(12.1)
(12.2)
such tMt
1-11
(12.3) mess {x E KR I v(x, t.) > (1 - II)",} < 1 _ 11121KR1, v = IDul·
~ (1 - II)/QR("')/,
contradicting (12.1).
We will work with the function w == IDuI 2, which satisfies (1.8) within the
cylinder KR x (t., 0). Introduce the change of variables
r = -tit., (= xlR, w«(,r) = w(Re,-t.r)
and the convex function of w
%
w I}
== max { ",2 i '2 .
Then KR x (t., 0) is mapped into Ql == K 1 X (-1,0) and, denoting again with
(x, t) the transformed variables, % satisfies
(12.4) %t-(At,k%Zt)o;,,:50 in Ql and 0<%:51,
12. Proof of Proposition 1.2-(i) 287
where the matrix (At,Al) is uniformly elliptic with eigenvalues bounded above and
below independent of p. Indeed it follows from (1.9) and the range (12.2) of t.
that
for two constants co(N,p, v) $ Co(N,p, v). The information of Lemma 12.1 in
terms of z implies
I-v
(12.6) meas {x E Kl I z(x, -1) > (1 - v)} $ 1 _ 1I/2IK11.
Without loss of generality we may assume that z satisfies (12.4) in a slightly larger
box, say Q2. This can be achieved by starting for example with Q2R (p). Propo-
sition 1.2 is a consequence of the following:
THEOREM 12.1. Let z E C (-2, OJ L2(K2» nL2 (-2,0; W 1 ,2(K2») be a sub-
solution off12.4)-(12.5), and let (12.6) hold. There exists 11 =lI(N,p, II) E (0, 1),
such that
meas {(x,t) E Q! I z(x,t) > (1-1I)} = O.
In view of (12.4), the proof of the theorem uses techniques typical of a single
equation. Even though these methods have been presented in various forms in
Chapters n and m, we reproduce here the main points, to render the theory self-
contained.
!li(z) = In+ { II }
- II - (z - (1 - 11»+ + 110 •
There exists a constant"( = ,,(N,p, v) such that for all t E (-1,0) andfor all
0<0'< 1,
(12.7) J !li2(z) dx $
K"x{t}
J !li2(z) dx + (I!0")2
KIX{ -I}
JJ
Ql
!li(z) dxdT.
PROOF: Let x - (x) be a cutoff function in Kl that equals one on Ka. and in
the weak formulation of (12.4) take the testing function !lilli' (2 , modulo a Steklov
averaging process. Then (12.7) follows by estimates analogous to those in Propo-
sition 3.2 of Chap. II.
288 IX. Parabolic p-systems: fR)lder continuity of Du
(12.8)
The proof of (12.8) is analogous to the proof of the energy estimates of Propo-
sition 3.1 of Chap. II. The spaces Vm,P( Qp) for m, p ~ 1 are introduced in §3 of
Chap. I.
LEMMA 13.1. There exists a constant 110 E (0, v) depending only upon N,p, v
such that lor all t E ( -1,0)
PROOF: We will use the logarithmic inequality of Lemma 12.2. Since !If(z) van-
ishes on the set [z < (1 - v)], by virtue of (12.6), the first term on the right hand
side of (12.7) is majorised by
I-v 2(1.1)
1- 1.1/2 In ~o IKll·
The second term is majorised by
We estimate below the right hand side by extending the integration to the smaller
set [z(·, t) > (1 - 110)]' On such a set
!If(z) ~ In (1.1/2110)'
Also
13. Proof of Proposition 1.2 concluded 289
Having determined '10. let So be the largest positive integer such that2- So ~
'10' For s ~ So. set
o
A.(t) == {x E KI I z(x, t) > (1 - 2- S ) } , As == jIAs(T)ldT.
-I
(13.3)
PROOF: Apply Lemma 2.2 of Chap. I to the functions x-+z(x, t) fortE (-1,0).
and for the levels
l = 1- 2-(·+1),
Taking into account (13.2). we obtain
We square both sides of this inequality, integrate in dt over ( -I, 0) and estimate
the resulting integral on the right hand side by the energy inequalities (12.8) written
over the pair of cylinders Ql and Q2. This gives
Therefore
y.n+l < ny'1+1i1h
_ "Y4 n , n=O,I,2, ....
It follows from Lemma 4.1 of Chap. I that {Yn } - 0 as n - 00 provided
15. Bibliographical notes 291
(13.4)
To prove the theorem we have only to pick 8. by the procedure of Lemma 13.2 so
that (13.4) is satisfied and then set 'I = 2-(8.+1).
For this, the linear analysis of §6 can be carried with minor changes. The analog of
the 'algebraic' lemmas of §5 are a direct consequence of the structure conditions
(81 )-(~). In the proof of Propositions 1.1 and 1.2, when working within cylinders
[(x o, to) + Q (6, p)], the 'perturbation terms' <Pi, i =0,1,2, contribute terms that
are infinitesimal with p of higher order with respect to those generated by the
principal part. This is due to the integrability condition (86). Further details on the
estimation of the lower order terms can be carried out with an analysis similar to
the estimation of the lower order terms in Chaps. II-V.
1. Introduction
We will establish everywhere regularity up the boundary for weak solutions of the
parabolic system
U::(UlIU2, ..• ,Um), meN,
{ UieC (e,T; L2(O»nV(e,T; Wl,P(O» ,
(1.1)
Ui,t -div IDulp-2 DUi =Bi(X, t, U, Du),
in 0 x (e, T),
ee(O,T), i=I,2,,,.,m, p>max{liJ~2},
associated with Dirichlet boundary data
(1.2)
in the sense of the traces on ao, of functions in Wl,p(n). The basic assumptions
on ao, the boundary data g and the forcing term B
(1.3) -
9i,z; E C (nT),
.\-
9i,t E LOO(nT),
i = 1,2, ... , m , j = 1,2, ... , N.
We set(l)
m N
(1.4) /11/1 == L L {/l9i /loo,nT + /l9i,tlloo,nT + [9i'Z;].\,nT}·
i=1 j=1
for some given constant Bo. We say that a constant 'Y = 'Y (data) depends only
upon the data if it can be determined a priori only in terms of
(I) For a smooth function r/J, the norm [r/Jh,K is defined in (1.3) of Chap. I.
294 X. Parabolic p-systems: boundaJy regularity
where 4> is a function of class C l ,>. in the (N -1) -dimensional ball8'R,. satisfying
(2.1)
The last condition can be realised by taking a smaller Po if necessary. With respect
to the new variables
(2.2)
(2.3)
Figure 2.1
which from now on we assume. The proof of Theorem 1.1 is based on comparing
w in a neighborhood of each point (xo, to) E QX/2 n {x N ~ O}, with the solution
of
V == (111.112"'" 11m) , mEN,
(2.16) { l1i,t-div IDvl,,-2 Dl1i =0, in [(x o, to) + Q (R2+'I, R)] nQ~,
l1i = Wi on 8" [(xo, to) + Q (R2+'I,R)]nQ~,
where 8"Q denotes the parabolic boundary of a cylindrical domain Q. The exis-
tence of a unique weak solution of (2.16) can be established by a Galerkin proce-
dure. (1) Denote by
(2.18) sup
[(zo.to )+Q(p2+'I.p»)
IDvl '5. 'Y (R'7 ff IDvl" dxd1") )
1/2 + R- ~.
[(zo.to )+Q(R2+'I.R)
THEOREM 2.2 (THE SINGULAR CASE max { Ii A~2} < p < 2). Let v be
a weak solution 0/(2.17). There exists a constant'Y = 'Y(N,p) such that/or all
O<p'5.!R
(2.19) sup
[(zo.t o )+Q(p2+'I.p»)
IDvl '5.'Y (R- '7 ff IDvl" dxd1")l/V + Rr-.,
N
P
[(zo.to)+Q(R2+'I.R)
3. An iteration lemma
for given positive constants A, /3,11, K. satisfying in addition /3 > K. and II E (0, 1).
Then for every
there exists a constant'Y depending only upon A, /3, II and 6, such that
298 X. Parabolic p-systems: boundary regularity
P )(j-tf,+6
(3.3) cp(p) ~ 'Y ( R' (cp(R) + 1)
(1 - v)1\:
VO<p~R~ 1, where q = 1 + f3 .
PROOF: Choose Ro < 1 and define the sequence Rn+l = ~, n = 0, 1,2, ....
Then
Now
and
Therefore if A ~ 2
n+1
qno+l InA
-->--.
no+ 1 - pnR~1
It follows from (3.3) that if n ~ no,
If Ra < 1 is fixed, for every p E (0, Raj there exist some n E N such that
Rf+l ~p~ Rtf. Therefore the equation p = R!,q" has a root (J E [I, q]. Starting
the process with Ra replaced by R!, gives
Remark 3.1. The lemma continues to hold for 11=0. The constant 'Y on the right
hand side of (3.3) is 'stable' as 11'\.0.
In the estimates below we integrate over the boxes Q'k, rather than +Q'k. In doing
50 we think of v and w as defined in the whole Q'k through an odd extension as
indicated in §2. By the Schwartz inequality
Since v - w vanishes on the lateral boundary of Q'k, by the Sobolev embedding, (1)
1(') S
z.::.!
+//IDvIPdXdT+'YRN+2+,,-ap~ [:F(O:'''I,R)l~.
Q:
By taking R sufficiently small and by interpolation, the first term on the right hand
side of (4.6) can be eliminated. This is the content of the following lemma:
LEMMA 4.1. There exists a constant "'1 ='Y(data) such that for a/IV 0 < p ~ ! R ~
In
(4.6') //(1 + IDwIP)dxdT ~ 'Y!!IDvIPdxdT
Q: Q;p
+'YRN+2+,,-ap~ [:F(O:'''I,R)l~.
PROOF: It suffices to prove the lemma for R ~ R" where Ro is so small that
302 X. Parabolic p-systems: boundary regularity
'Y (2Ro)'\;!-r ~ ~.
Let 0 < P ~ ~ Ro and consider the sequence of radii
Yn ==/ PI + IDwIP) dt
Q:n
Ir',/2 (u IDvl'do:dr )
~
and
(4.9)
We summarise:
PROPOSITION 4.1. Let a > 0 and 71 = a(p - 2). There exists a constant'Y =
'Y (data), independent 0/ a, 71, p, R, such that
/orall (xo,to)EQ:k/2 and/orall O<p~R~'R./2,
~ 'Yg(a)(~)N+2+'I /f (1 + IDwlP)dxdr
[( Zo ,to )+Q7tJ
+ 'Yg(a)RN+2+'I-ap~ + 'YpN+2+'I R-aP.
PROOF: The previous arguments prove the proposition for those points
(xo, to) EQ!Rn{XN =O}.
The estimate is obvious for boxes [(x o, to) + Q7tl c Q~, by interior estimates. If
[( xo, to) + Qkl intersects {x N =O}, then either
!
In the first case we may establish (4.10) with R replaced by R. The general case
follows by suitably modifying the constant 'Y. If the second of (4.11) holds, we let
(4.12) X. == (Xo,l' Xo,2, . .. ,Xo,(N-l), 0)
and observe that
[(X., to) + QiR] C I(xo, to) + Qkl·
We carry on the process leading to (4.10) for such a new box, for all 2X o ,N :5 p <
! !
R. This implies that (4.10) holds for all Xo,N < p:5 R. If p:5 Xo,N, we consider
the cylinder [(Xo, to) + Qt.N]' which satisfies the inclusion
LEMMA 5.1. For every 0: E (0,1) there exists a constant 'Y = 'Y (0:, data), such
thot
jorall (xo,to) E Q:k/2 and/orall 0<p:5R:5'R/2,
(5.2)
Since IDwIELP(nT),
Therefore
6. Estimating the local averages of w (the case p > 2) 305
Suppose the lemma holds for Q n and let us show that it continues to hold for Q n +1.
If F(Qn):'5 1'(Qn), the quantity Q(Q n ) introduced in (4.9) is bounded and we may
use (4.10) with Q = Q n and TI = TIn. We apply the iterative Lemma 3.1 to the
function
tp(p) = JJ(1 + IDwI P) dxdr,
[(zo,to)+Q~l
with the choice of parameters
p-2 -
(5.3) v = --,
p-l
6 = 6n Q np.
We obtain
and partition the cylinder [(x o, to) + Q~n+l] into s = p'r/n+l-'r/n adjacent boxes
with 'vertices', say (0, td, (0, t2), ... , (0, t s ). Then
We may treat analogously the other points of Q'R./2 and the inductive inequality
(5.2) follows. To prove the lemma it suffices to prove that {Q n } -. 0 as n -. 00.
The sequence {Q n } is deacreasing. We claim that {Qn} -. O. Indeed if not,
lim
n--+oo
Q n = Q o > 0,
Since w vanishes for XN =0, we regard it as defined in the whole QR, by an odd
extension across {x N =o}. Let
(w)o,p == HW(X,t)dxdr
[(zo,to)+QpJ
denote the integral average of w over I(xo , to) + Qp]. If (x o, to) coincides with
the origin, we let (w)o,p==(w)p' Also let
We observe that if Xo E {XN =O}, we have (w)o,p (t) =0 for all t E (to - p2, to).
since w is odd across {XN =O}, and in particular (w)o,p =0.
LEMMA 6.1. For every a E (0,1) there exists a constant "( = "( (a, data), such
lhat
By a translation we may assume that (xo, to) coincides with the origin. We have
6. Estimating the local averages of w (the case p > 2) 3C17
Next
We estimate the integrand on the right hand side of (6.5) by making use of the
equation (2.5), over the cylinder Q P+tT p. Let x -+ ({ x) be a non-negative piecewise
smooth cutoff function in K p+tTP that equals one on Kp and such that ID(I :51/up.
In the weak formulation of (2.5), take ( as a testing function and integrate over
K p+tTP x [T, t] to obtain
/!
Kp+"p
([w{x, t) - w{x, r)] dx/
:5 t !!IAi{X,w, Dw).D(+At(xt+Bi(ldxdT
1=1 Qp+"p
:5 u'Yp! /(1 +
Qp+"p
IDwIP-1) dxdT
+ / /[w{x,t) - w{x,T)]dxl.
Kp+"p\Kp
Let {w)P+tTP denote the integral average ofw over the Qp+tTP, i.e.,
308 X. Parabolic p-systems: boundary regularity
(W)P+ITP == H
Qp+"p
w(x, r)dxdr.
Then
r}
Kp+"p
.1
We conclude that for every a E (0, 1) there exists a constant 'Y = 'Y( a) such that for
every uE (0,1) and for every pE (0, fR)
(6.6) H
Qp
Iw - (w)pI P dxdr ~ 'Y~~) p(l-a)p
+ 'Y(a)uP- 1 H
Qp+ .. p
Iw - (w)p+ITPI Pdxdr.
This implies the lemma, in the case (6.4) holds, by the interpolation process of
Lemma 4.3 of Chap. I. This process yields the choice of u E (0, !).
Finally, having
fixed u E (0, !), consider the case when
H Iw - (w)o,p IPdxdr
[(zo,to)+Qp)
~ 'Y H IwlPdxdr
[(z.,to)+Q3p)
~ 'Y(a)pp(l-a).
7. Comparing w and v 309
(7.2) //(11ID (sv + (1- s)w) IP- 2dS) IDw - Dvl2 dxdr
+Qk
In carrying the estimates below, we think of v and w as defined in the whole Q'k
by an odd reflexion across {XN =o}. By the Poincare inequality and (2.7)
2::!
+ ! ! IDuIP-IIDw - Dvldxdr}
£2
~
+ { [[ IDw - Dvldxd-r + [/ IDw - DvldxdT }
2::!
In t~l~" inequality we absorb the integrals of IDw - vi extended over £1 into the
analogous term on the left hand side by means of Young's inequalitj. Using also
the definition of £2 we arrive at
7. Comparing w and v 311
[(Zo.to)+Q~l
1
[(zo.to)+Q~l
Therefore
Rewrite the integrand in the second integral on the left hand side of (7.4) as
LEMMA 7.1. There exits a constant 'Y = 'Y (data), such that
forall (xo,to) E Q:k/2 andforall 0<p5,R5,'R/2,
To estimate the last integral on the right hand side of (7 .5) we make use of Theorem
2.2. the defmition of F(Q) and (7.6). Assuming that (xo, to) coincides with the
¥l
origin, we have for all 0 < p 5, 5, 'R/4
Then the first term on the right hand side of (7.7) is estimated above by
'Y pN +2+'1 R-oP.
Setting also
LEMMA 8.1. For every a E (0,1) there exists a constant -y = -y (a,data), such
that
forall (xo,to) E Q:k/2 andforall 0<p~R~n/2,
(8.1) H + IDwI
(1 P) dxdr ~ -y(a,data)p-ap •
[(xo,to)+Q~J
Therefore
8-(1). PROOF OF THEOREM 1.1 (THE CASE max {I; ~~2} < p < 2)
By a cube decomposition technique similar to the one outlined, Lemma 8.1
can be rephrased in tenns of the parabolic cylinders Qp=Q(p2, p).
LEMMA 8.1'. For every a E (0, 1) there exists a constant 'Y = 'Y (a, data), such
that
With this lemma at hand the proof is now concluded as in the degenerate case.
First we may establish a version of Lemma 6.1 and then the HOlder continuity of u
follows from the arguments of [22,23,32,79] or by those in §§2 and 3 of Chap. IX.
9. Bibliographical notes
The proof of Theorem 1.1 is in [27]. The iteration Lemma 3.1 is in the same spirit of
similar results of Campanato [22,23]. The technique is indeed a degenerate version
of [22,23]. Techniques of this type near the boundary appear in Giaquinta-Giusti
[49]. The boundary behaviour of solutions of (1.1) is essentially not understood. In
the case of a single equation some results appear in Liebennan [69] and Lin [72].
XI
Non-negative solutions in ET-
The case p>2
1. Introduction
Non-negative solutions of the heat equation in a strip ET == RN X (0, T) are some-
what special in the sense that they grow no faster than
RN
THEOREM 2.1. There exists a constant 'Y='Y(N,p) such that/or all eE (0, T),
;!, }>./P
(2.1) IIlullr,T-~ $ 'Ye-p!-, { 1 + (~) p- u(O,T-e)
pp/(p-2) pI>'
(2.2) lIu(·, t)lIoo,K p $ 'Y tNt>. IIlullr,T-~'
t
p2/(p-2) 2/ >.
(2.4) IIDu(', t)lIoo,K p ~ 'Y t(N+l)/>' Ilullr,T_~'
(2.5)
for a constant 'Y. ='Y.(N,p). In view of (2.1) they can be considered valid for all
tE (0, T-e). Indeed working within ET • we may state them for every substrip
for all r E (it, t). Estimating the right hand side by (2.3) we obtain
IIDu(·,r)lIoo,Kp <
p2/(p-2) - "Y
(till UnUlp-2
r,T-£:
)1/>'111 ullIII r,T-£: +C~ •
Next we will turn such infonnation into the quantitative estimate (2.4).
3. Proof of (2.4)
Let t>O and p>r be fixed and consider the box Qo==K2pX [it, tjcE T • Let the
radii {Pn} and time levels {t n } be defined by
n=0,1,2, ...
and introduce the corresponding family of nested shrinking cylinders
Qn == K pft x{tn,t},
with vertex at (0, t). We will estimate the quantity IIDu(·, t)lIoo,K p ' by using the
techniques developed in Chap. VIII. The starting point is the the iterative inequality
(5.4) in that chapter, which we rewrite here in the context of the cubes Qn as
We conclude that there exists a constant "Y = "Y( N, p) such that for all it:5 r:5 t,
t/2 K2p
320 XI. Non-negative solutions in E T • The case p>2
By Lemma 2.1 and (2.2) this quantity is well dermed. In the estimates below we
write ~ == ~(t) if the dependence upon t is unambiguous. We estimate the quantity
1t introduced in (3.1) by
and
(3.4)
In estimating G2 (t) we use the estimation (2.2) and the range (2.5) oft.
Combining these estimates in (3.4) gives for all 0 < t ~ 'Y.lllulll ~,7-E'
322 XI. Non-negative solutions in Er. The case p>2
2/>.}-1/(P-2)
{1- 'Y (t l~ull~:T2_E) ~ 2,
we will have
t¥ IIDu(·, t)lIoo,K < 2 IlluI1 2/>'
p
p2/(p-2) - 'Y r,T-E
for all such t and all p> r.
4. Initial traces
(4.1) lJ$ f
RN
u(x, t)cpdx = f
RN
cpdIL, 'v'CPEC~(RN).
(4.2) sup
p>r
f
Kp
dIL
/( -2)
pP P
< 00, 'v'r > O.
PROOF: The existence of a Radon measure IL satisfying (4.1 )-(4.2) follows from
the global Harnack estimates of §7 of Chap. VI. Indeed by Corollary 7.1 of that
Chapter, for every cube [x o + K pj C RN and all cp E Cr;' (Kp),
If
_ Kp
u(x, t)cp(X)dXI ~ 'Y (N,p, p, T, U(Xo, T-e» IIcplloo,K p '
5. Estimating lDur- 1 in Er 323
for all 0 < t :$ T -E and all E E (0, T). Therefore {u(·, t)}O<t<T-€ is a net of
equibounded linear operators in Cgo(RN), and for a subnet, indexed with t',
for a Radon measure IJ. The uniqueness of such a measure is a consequence of the
following:
LEMMA 4.1. Let u be a non-negative local weak solution of (1.5) in ET. Then
PROOF: Fix 0 < r < t and 0' E (0, I), and let x ...... (x) be a non-negative piecewise
smooth cutoff function in K(1+cr)p that equals one on Kp and such that ID(I :$
2p/0'. In the weak fonnulation of (I.S) take ( as a testing function. Integrating
over (r, t) gives
t
f u(x, t) dx
K(1+")p
~ f
Kp
u(x, r) dx - 0'2p f
T
PDuI P- 1 dxds.
K(1+")p
To prove (4.3) we estimate the right hand side of this inequality by (2.3).
We now prove the uniqueness part of Theorem 4.1. Suppose that out of the
net {u(x, t)}O<t<T-€ we may select two subnets indexed with r' and t' such that
for all 'P E ego (RN) and IJ :/= II. Then we let r ...... 0 along r' in (4.2) and then let
t ...... 0 along t'. This gives
Interchanging the role of IJ and II proves the Theorem since 0' E (0, I) is arbitrary.
Local integral estimates of IDul p - 1 are crucial both in the global Harnack estimate
of §7 Chap. VI and in the theory of initial traces. The inequality (2.3) of Theorem
324 XI. Non-negative solutions in Er. The case p>2
2.1 is local but holds for all p > r. Therefore it implies some control on the be-
haviour of IDul as Ixl -+ 00. This behaviour can be given an integral form, by
means of the weights
(5.1)
.x
(5.2) ap= --2 +u, for some u > o.
p-
THEOREM 5.1. Letu bea non-negative local weak solution of(1.5) in ET. Then
for every u > 0, there exists a constant 'Y = 'Y( N, p, u) such that for all r > 0 and
all EE (0, T),
(5.3) sup
O<t<T_
ju(x, t)Aa(x) dx ~ 'Ylllullr,T-E'
- aN
t
(5.4) j jlDulP-l Aa(x) dxdr ~ 'Ytl/>'l~ull~;'~.
oaN
Remark 5.1. The constant 'Y(N,p, u) /00 as u '\,0.
PROOF OF (5.3): Without loss of generality we may assume that r = 1. Then
foralIO<t~T-E,
~ mU~lr.T-E + 2~ L 2- an mullr,T-E.
n=O
PROOF OF (5.4): It will suffice to establish the estimate for t in the interval (2.5).
We will use this fact with no further mention. First we observe that the inequality
(5.5)
holds for all x E RN and all 0 < t ~ T-E. This is obvious if Ixl ~ r with the
constant 'Y depending also upon r. If Ixl > r, we apply (2.2) to the cube K 21zl.
Let T/ E (0, T -£ ) and in the weak formulation of (1.5), take the testing function
where x -+ «(x) is the usual cutoff function in Kp. After a Steklov averaging
process and standard calculations, we obtain
5. Estimating IDulp - t in Er 325
t
IP A 1. (Pdxdr
(5.6) f (r - "l)l/pjIDU
u 2 / p "'+,.
" Kp
t
~ 'Y j(r - "l)I/P j u¥up-IID (A~:*()IP dxdr
" Kp
t
+ 'Y pr - "l)~-lj u¥ A1/puA",dxdr = J~l) + J~2).
" Kp
t £=l
J(2) < "'f(r_'TI)t- 1jr N <:A 2 ) lu(x,r)I" u(x r)A (x)dxdr
p - /./ (1 + Ixl p)1/p ,'" ,
" Kp
~ 'Y(t - "l)!-,Alllull!;'~.
As for J~I,l), since ID(I ~ 2/ p, again by (5.5) and (5.3)
(5.7)
Ilullr.T-~ + Ilvllr.T~ == A.
Then and u and v satisfy all the estimates of Theorem 2.1 with the quantities
llu. vllr.T~ replaced by A within the strip ETa' where
0< To = min{T;-y.A-(p-2)}.
and -y. is the constant appearing in (2.5). It will suffice to prove uniqueness within
the strip ETo' The difference w=u-v satisfies
6. Uniqueness for data in Lloc(RN ) 327
(6.1)
where
+ (p - 2) j ID(su + (1 - s)v)IP-4
o
X (su + (1 - s)v)x. (su + (1 - s)v)xjds.
The matrix (ai,j) is positive semi-definite and for all eE RN and (x, t) E ETo
ao(x, t)lel 2 S ai,j (x, t)eiej S (p - l)a o(x, t)leI 2 ,
{ l
(6.2)
ao(x, t) = flD(su + (1 - s)v)IP- 2ds, (x, t) E ETo.
o
Let Ao(x) be the weight introduced in (5.1) with Q satisfying (5.2). In the argu-
ments below, 'Y denotes a positive constant that can be determined a priori only in
terms of N, p, (T and A.
In the last integral.IDAal $'YAa+l/p and in the first integral. since IDcl =0 on
Kp/2 we have AalDCI $'YAa+l/p for p > 1. Therefore letting p-+oo
t
jlw(x.t)IAa(X)dX $ 'Y jjODvl + IDvI)P-l Aa+l/pdxdr,
RN ORN
PROOF: Let fiE (0, *,) be fixed. Then 'v'tE (0, To)
+11 If
6Kp
IDwI2 (1)2
ao(x,r)(W+6)1-'7 A dxdr Q (
~ 1 ~ 11 !
Kpx{6}
(w + 6)1+'7AQ(2dx
t
!
+ '1(11) !ao(x, r)(w + 6)1+'7 (AQID(1 2 + IDA! 12) dxdr.
6K p
We absorb the integral involving IDwl2 on the left-hand side of (6.3) and discard
the resulting non-negative term. Finally, we observe that by the definition of AQ
and the structure of ( we have
+'Y!!ao(X,r)A;(x)(w + 6)1+'7AQ(x)dxdr.
6Kp
t - j1w(x,tW+f/Ao(X)dx == 0,
RN
t - /lw(x,tW+f/Ao(X)dX E VlO(O,To)'
RN
Now the parameter Q in the calculations above is arbitrary and only restricted by
(5.2). If Q is replaced by Q+,,/(p-2). then Lemma 6.2 and its proof ensure the
Loo(O, To) requirement and the theorem follows.
Remark 6.1. For non-negative solutions u and v of (1.5) in ET. the quantities
(7.3)
- f
Iluolir = :~~
luo(x)1
P"/(p-2) dx<oo, for some r > O.
Kp
Since U o ELloc(RN) if Iluollir is finite for some r >0, it is finite for all r>O.
THEOREM 7.1. Let U o satisfy (7.3) for some r > O. There exists a constant 'Y. =
'Y.(N,p) such that defining
(7.4)
there exists a unique solution u to (7.1) in ET. Moreover u satisfies (7.2) for all
eE(O, T) and the estimates (2.2)-(2.4) of Theorem 2.1.
Remark 7.1. This is an existence theorem local in time and the largest existence
time is estimated by (7.4). The functional dependence in (7.4) is optimal as shown
by the following explicit solution.
( ) = {max{-n;
uo,n x - 0,
min{uo(x);n}}, for Ixl < n
for Ixl ~ n.
It is apparent that for all n= 1, 2, ... ,
(7.5)
are finite for all r, t > o. It follows that the sequence {Un} satisfies (2.2)-(2.4) of
1beorem 2.1. We will tum such n-dependent information into a quantitative sup-
estimate of {un} independent of n. Let x --+ '(x) be the standard cutoff function
in K 2p • Then (7.l)n implies
We divide by p>'/ (p- 2) and take the supremum over all p> r. Taking into account
(7.5) and (2.3) this gives
P-2) 1/>. 1
'Yl ( tn IIlunllr,t = 2·
Then from (7.6) for all t E (0, t n )
We summarise:
LEMMA 7.1. Let {Un} be the sequence of the approximating solutions (7.1)n.
There exists a constants 'Y = 'Y(N,p) and 'Y. ='Y.(N,p) independent ofn. such
that
(7.7)
where
(7.8)
8. Bibliographical notes 333
Given such an estimate, the Cauchy problem (7.1) can be solved by a standard
limiting process. Indeed by Theorem 2.1 the sequences
{ -f)
Un } , i = 1,2, ... ,N,
f)xi nEN
are locally equibounded and equi-HOlder continuous in RN x (O, Tr). This gives
the existence of a unique solution in ETr • The largest time of existence can be
calculated from (7.8) by letting r -+ 00. In particular the solution to (7.1) is global
in time if
.
lim sup
p>r
f'-tOO
j uo{x)
>./( -2) dx = O.
P P
Kp
8. Bibliographical notes
Theorem 2.1 is taken from [41]. A weaker version of (2.2) in I-space dimension
is due to Kalashnikov [58]. It is remarkable that in (2.4) one can also control the
behaviour of the space-gradient IDul as Ixl-+ 00. Since IDul 2 is a non-negative
subsolution of a porous medium-type equation (see (1.8) of Chap. IX) the same
techniques yield a version of (2.2) for such degenerate p.d.e. The analog of (2.2)
for the porous medium equation is due to Benilan-Crandall-Pierre [10] in the
context of an existence theorem. A rather general version is in [4]. Perhaps the
most relevant estimate of Theorem 2.1 is the integral gradient bound (2.3) proved
in [41]. A version of such a local bound, for the porous medium equation is in [4]
and reads
IIIulllr,T-E
- sup sup j
= u(x,t)
,./(m-l)dx.
O<tST~p>r p
Kp
The estimate holds for small time intervals and for general non-linearities. We
refer to [4] for details. There is no analog of (2.4) for the porous medium equation.
Theorems 4.1 is taken from [41]. The analog for the porous medium equations is
in [6] and for general non-linearities [4]. It would be desirable to have a version
of the uniqueness Theorem 6.1 for initial data measures. This would parallel the
analogous theory for the heat equation.
XII
Non-negative solutions in E T .
The case 1<p<2
1. Introduction
A striking feature of these singular equations is that, unlike the degenerate case
p>2, non-negative solutions of (1.1) are not restricted by any 'growth condition'
as Ixl- 00. Nevertheless they have initial traces that are Radon measures. More-
over they are unique whenever the initial traces are in Lloc{RN ). Accordingly, the
Cauchy problem for (1.1) associated with an initial datum-
(1.2) U o ~O,
2N
(1.3) and P>-N
+r '
1. Inttoduction 335
then the solution '1£ of (1.1)-(1.2) belongs to Lroc{ST) , "It> O. This is the content
of Theorem 5.1 of Chap. V. In §13 we will give a counterexample that shows that
if '1£0 violates (1.3), then '1£ ¢ L~c{ET). The basic formal energy estimate for (1.1)
is
VO<8<t~T, VKp
t
(1.4) sup ju 2 {x, r) dx + f flDulPdxdr
s<r<t
- - Kp
11
sKp
Thus ifu e L~oc{ET), the left hand side of (1.4) is finite and IDul e Lfoc{ET)'
However if '1£0 e Ltoc{RN), there is no a priori information to guarantee that
(1.5)
We have spoken oholutions of (1.1); however if (1.5) fails, one of the main prob-
lems is to make precise what it is meant by solution. Thus the starting point of the
theory is to give a precise meaning to Du to make sense out of (1.1). The previous
remarks suggest that IDul might fail to be in Lfoc{ET ), roughly speaking at those
points where '1£ is unbounded. Motivated by these remarks, we have given a novel
formulation of non-negative weak solutions. Such solutions are 'regular' in the
sense that the truncations
(1.6) Vk > 0, Uk = min{u, k},
satisfy
(1.7)
Then (1.1) can be interpreted weakly against testing functions that vanish 'when-
ever '1£ is large'. A suitable choice of such testing functions is
(~- '1£)+ == max{(~ - u);O}, ~ e C~{ET); ET'
The notion is introduced and discussed §§2 and 3. We prove that these solutions
coincide with the distributional ones if (1.5) holds and that the truncations Uk are
distributional super-SOlutions of (1.1) Vk > O. We derive a spectrum of properties
of such local weak solutions, regardless of their initial datum. In particular we
investigate the behaviour of DUk as k -+ 00. A relevant fact is the estimate
VO<8<t~T, VK2p,
336 XII. Non-negative solutions in E r . The case I <p<2
for a constant 'Y = 'Y( N, p, q). The theorem follows by letting p -+ 00 after we
choose q so large that N(p - 2) + pq>O.
If, in (1.3), r = 1 and p > J~ l' the existence and uniqueness theory remains
valid if U o E Lloc (RN) with no sign restriction. Indeed in such a case the sequences
2. Weak solutions
A measurable function U : ET -+ R + is a local weak solution of (1.1) in ET if
(2.3)
PROOF: Let IC c IC' be compact subsets of ET such that dist (8IC, 8IC') = d > 0
and let (E C:'(IC') be such that 0 ~ (:51 and (== Ion IC. Choose 1/J E Xloc(ET)
and in (2.2) take
where
a.e. in IC.
338 XII. Non-negative solutions in ET. The case I <p<2
Conversely. if t/J E C~ (I:T ). we may write (2.7) for s < t such that supp{ t/J} C
RN x (s,t). By taking (so that p>2diam(supp{t/J}). we obtain (2.2). We con-
clude that the fonnulations (2.2), (2.5) and (2.7) are equivalent.
LEMMA 2.2. Let UES satisfy
Then
Ut - div IDul p- 2Du = 0 in 1>'(I:T).
PROOF: In (2.5) take t/J=un + 1 E X'oc(I:T). nEN. We obtain 'VTJE C~(I:T)
j j{UtTJ + IDulp-2 DuDTJ}(un - u + l)+dxdT = j j IDuI P'1 dxdT .
I:T I:Tn[n<u<n+1!
Since IDul E Lfoc(I:T ). the right-hand side tends to zero as n- 00. The left-hand
side converges to
j j {U'Tlj!"(k)(k - u)dk
I:T U
Since
00
the assertion follows for f e 0 2(0,00). TIle general case is proved by approxima-
tion.
340 XU. Non-negative solutions in Er. The case 1<p<2
3. Estimating IDul
LEMMA 3.1. There exists a constant 'Y = 'Y(N, p) such that
Yk> 0, Yp > 0, YO<s<t::5T, Yu E S
t; s) .
t
PROOF: Let ( be the standard cutoff function in K2p. Then from (2.7) with 1/J =k
t t
+~ JJ:T (k - u)!(PdxdT
BRN
p; J
t
::5 1 jIDUkIP(PdXdT
SRN
J
t
+ ~ fik - u)!(Pdx.
RNX{t}
I
IDuP-~-Q e Lroc(I:T ),
tmd there exists a constant 'Y='Y(N,p) such thatYO< s<t::5T and/or all p>O
PROOF: Fix k > 0 and e e (0,1). and in (2.8) take '1 = (P1/J-Q. where ( is the
standard cutoff function in K 2p and
3. Estimating IDul 341
u>e
u ~e.
We obtain
t
o /JIDuIPu-Q-l(Px[e<u<kj dxdr
BRN
t
~ P //IDuIP-lu-Q(P-IID(lx[e<u<kjdXdT
sRN
(3.3)
t
+ p / /IDu~IP-le-Q(P-IID(1 dxdr
sRN
By virtue of Lemma 3.1 the second integral tends to zero as e -+ 0 at the rate of
eP - 1 - Q • Combining these calculations we deduce
t
o j/IDuIPu-Q-lx[e<u<kl dxdr
BKp
~ O(eP-1-Q) + oP
"(_1 { (sup
'TE(s,t)
ju(x, T)dx) l-Q (2p)QN
K2p
(3.4)
+ (~) (sup
pi' 'TE(s,t)
ju(x,r)dx)P-I-Q (2P)N(2-P+Q)}
K2p
+ O(eP - 1- Q ).
342 XII. Non-negative solutions in ET. The case I <p<2
If
c S)
~ ~ [Ms ,t(2p)]2-",
the quantity in braces on the rightmost side of (3.4) is majorised by [Ms ,t(2p)j1-0<.
Otherwise it is majorised by
J
sKp
f IDu¥ I"x.[n < u< n + 1] dxdT
~ I~2) ~ 1pN In (1 + ~) ,
.1
SK2p
(~ ft
SK2p
Therefore
Then IDzl E Lfoc(ET) and there exists "Y="Y(N,p) such that'v'O < s< t ~ T and
'v'p>O,
PROOF: Divide both sides of the inequality of Lemma 3.3 by lnl+E n, and add
over all n=2,3, ...
The estimate (3.7) deteriorates as E-O. The following corollary gives some
information in the case E = O.
COROLLARY 3.2. Let uES. Then 'v'O<s<t$T. andforall C> 1.
t
lim
k .....
ffIDuIP-I-l-
oo}} U nu
X[k<u<Ckjdxdr = O.
s Kp
PROOF: Without loss of generality we may assume that k and Ck are positive
integers. Divide both sides of the inequality of Lemma 3.3 by In n and add for
n=k, k + 1, ... ,Ck. This gives
t
= "YIn ( 1 + InC)
Ink [ M s •t (2p) + (tIi'S) ~l .
THEOREM 4.1. Let U E S. There exists "I = "I(N,p), such that "10 < s < t $. T
andVp>O
(4.1) sup
TE(s,t)
!
Kp
u(x, r)dx $. "I
K2p
j u(x, t)dx + "I ( t-S)~
P
-.>.- ,
A = N(p - 2) + p.
The uniqueness of the initial trace J.I. relies on the next gradient estimates.
LEMMA 4.1. Let uES. There exists a constant "I="I(N,p) such that
VO<s<t$.T, Vp> 0, "10' E (0,1), "Iv> 0,
(4.2) h
1 IDulp-1dxdr
p}} $. "I (t-S)~
7-
8 Kp
:!iE..=..ll
+ "I (t -/) *{
P
sup ju(x, r)
S<T<t
dx} P
K2p
Moreover
(4.3) -
p
Ih'lt IDulp-1dxdr $. "I sup
S<T<t
j u(x, r) dx + "I (t-S)~
P
-.>.- .
S Kp K2p
PROOF: The proof is the same as that of Propositions 4.1 and 4.2 of Chap. VII.
The only difference is that instead of working with the solution u we work with
the truncations Uk and use the fact that these are supersolutions. In (2.8) we take
the testing functions
{ jU(r)'1dx}
RN TE(O,t)
is equibounded, with bound depending only upon I '1 II oo,RN • A subnet indexed
with {r'} converges to a Radon measure 1', in the sense of the measures, i.e.,
346 XII. Non-negative solutions in ET. The case 1<p<2
Suppose now that there exist another subnet, indexed with {r"} and a Radon mea-
sure jJ., such that
We will prove that J.I. == jJ.. Let u E (0,1) and write (2.8) with 1/J == 1 and ( the
standard cutoff function in K(l+u)p. Letting k -+ 00, standard calculations give
VO<s<tST
t
We estimate the last tenn by using (4.2) and let s '\. 0 along r' while t> 0 remains
fixed. Then we let t '\. 0 along the net r" to get
Since uE (0;1) is arbitrary, interchanging the role of J.I. and jJ. proves the theorem.
(5.1)
for some "'( = ",((N,p, t), Vk E R+,
for every compact subset K:. C ET and for all C> 1. In section §§8-12 we will
construct solutions of the Cauchy problem (1.1)-(1.2) that satisfy both (5.1) and
(5.2); therefore S· is not empty. Corollary 3.2 suggests that (5.2) is almost satisfied
s. The uniqueness theorem 347
5-(i). Preliminaries
LEMMA 5.1. LetuEs*. Then/oraIlO<s<t'5,T, Vp>O, VC> 1,
t
PROOF: Consider (2.8) written for Uk replaced by UCk. against testing functions
7J = ( In (k/2w k,C)
It follows from these definitions that 7J '5, 0 a.e. in ET and 7J = 0 a.e. on the set
[O<u'5, !kJ. By calculation from (2.8) we obtain
t t
The fll'St integral on the right hand side of (5.3) tends to zero as k - 00 by virtue
of (5.2). We estimate the second integral. formally. by
348 XII. Non-negative solutions in ~T. The case I <p<2
=In-2C
p
- ljlD
t
UCk
,
IP-l U- (<>+I)(p-l)
P U(<>+I)(p-l)
P [
XU> k/2]dxdr
SK2p
cl
SK2p
1be last integral is estimated by means of (5.1) and the lemma follows.
Remark 5.1. The assertion of the lemma is trivial if Ut E LJoc{ET)'
We give next a weak fonnulation for the difference of two solutions U1, U2.
First we recall that, by Lemma 2.3, the truncated function
ifO<U2<k
if u2~k
where ( is a non-negative piecewise smooth cutoff function in K(1+a)p, 0' E (0, 1),
such that
In view of the definition of X 10c (ET) and the regularity properties (2.1) of Ui, i =
1,2, such a choice of testing function is admissible, modulo a density argument.
On the other hand the weak formulation (2.7) of Ul holds against the same testing
functions. Therefore setting
kER+,
where
Jk == IDuIlp-2Dul -IDu2,klp-2Du2,k
1 .
+ (p - 2) ( /ID(~Ul + (1 - ~)U2,k)IP-4
o
XD(~Ul + (1 - ~)U2,k)(~Ul + (1- ~)U2,k)zjd{ )W(k),Zj'
Set also
1
Ao == /ID(~Ul + (1 - ~)U2,k)IP-2d{.
o
LEMMA 5.2. Ao:5P~1IDw(k)IP-2.
PROOF: If IDu2,kl ~ IDw(k)l, we have
350 xu. Non-negative solutions in ET. The case I <p<2
ID(eUI + (1 - e)U2,1c)1 = IDu2,1c + eDW(1c) I
~ IIDu2,1cI- eI DW(1c)11
~ (1 - e)IDw(1c) I·
1berefore
eo =_ ID
IDu2,1c I ( )
W(1c)
leo, 1 .
JlcDw(lc) ~ (p - I)AoIDw(Ic)12,
(5.6) {
IJlcl S AoIDw(1c)1 s p~IIDw(1c)IP-I.
In what follows we will use these inequalities without specific mention.
6. An auxiliary proposition
PROPOSITION 6.1. Let Ui E S· ,i = 1, 2, satisfy
{:/i)
ifW(k)
(6.2) W/i).h" (UI - ", •• )t - ifw(k) < h
ifw(k) ~ h
and in (5.5) consider the testing function
where
eE(O,I), a,b>O, n,mENj n>m+1.
We obtain
In using 1/1 as a testing function in (6.4) we keep in mind that the truncated functions
Ui,h, i = 1,2, Vh > 0, are regular in the sense of (2.1). In particular the first two
integrals on the left hand side of (6:4) are well defined V0 < 8 < t ~ T. We will
eliminate the parameters e,k,8,n,m by letting e-O, k-oo, 8-0, n,m-oo
in the indicated order.
352 XII. Non-negative solutions in I:r. The case I <p<2
(6.5) f
RNx{r}
W(k)(t/J - ud+(?dx -+ f W(k)
RNx{r}
(w~).nf (w~).m)b (Pdx.
This determines the limit for the first two terms on the left hand side of (6.4). To
examine the remaining terms we let 'iii, i = 1, 2, be arbitrarily selected but fixed
representatives out of the equivalence classes Ui, define iii, iii(k) accordingly, and
let
Next
t
ff
LE == -E
sRN
W(k) :r (t/J - ud+(Pdxdr
-ff BRN
w(k)(1 - EUlhx(FE)(Pdxdr
IL~3)1::; JJeIW(k)II!UIIX(Fe)dxdr.
aK2p
On the set Fe we have
1
-
e
::; Ul ::; -
1
e
1
+ -(n
e
+ 1)a+b == -,
'Y
e
eIW(k)/ ::; 'Y a.e. Fe·
Therefore
(6.6)
-b! 1 !
JJ (w~),n) a (wtk),m) b+l (?dxdr.
aRN
Since n>m + 1, .
+ )a Or{) (+
(w(k),n w(k),m )b+l = (wtk),m )a Or
{) (wtk),m )b+l
-- a +b+b +1 1 Or
{) (
w+(k),m )a+b+l ' a.e.ET.
b
- (a + l)(a + b + 1)
J( w~),m )a+b+ 1 (Pdx
RNX{t} .
We combine this with (6.5) and conclude that the sum of the first three tenns on
the left hand side of (6.4) has a limit. as e - 0, that is minorised by
(6.8) 1
a + b+ 1
I(w~).m )4+/1+1
("dx
aNx{t}
- a + b+ 1
1 I (w~) w~).n
)4+/1
("dx.
aNx{s}
We tum to estimate below the lim-inf as e - 0 of the last integral on the left hand
side of (6.4):
t
e IIJIcD(1/J - ud+("dxdr
saN
t
x(w~).m + e) /I ("x(g,Jdxdr
(6.9)
X ( w~).m + e) ("X(g,;}dxdr
t
- p ~ 1 I I AoIDw(k)IIDull("x(F~)dxdr
saN
== H~I) + H~2) .
By weak lower semicontinuity
(6.10) lim in! H~I)
~-o
t
t
IH~2)1 ::; €C(N,p) jfiDUl - DU2,IcIP-1IDullx(FE)dxdr
aK2p
t t
::; C€ j jIDU1IPx(FE)dXdr + C€ j jIDu2, Ic IPx(FE) dxdr
aK2p SK2p
=
-
H(2) +H(2)
E,l E,2·
Since IDu2,1c1 E LfoAET) the second tenD tends to zero as € - o. As for H!~l
write
t
H!~l::;c(P,n,m)jjIDu~IPx(~ ::;Ul::; ~)dxdr-o as €-o.
aK2p .
j AoIDw(lc) I (w~),n +
::; "1 j €) a
aaN
t
+ 'Y j j AoIDw(lc) IWIX(FE)(P-l ID(ldxdr.
saN
t
(6.12) 'Y !fiDW(k)IP-IX(:Fe)dxdr
SK2p
t
for a constant 'Y = 'Y(P, n, m, a, b). The second integral on the right hand side of
(6.12) tends to zero as e ..... O. since Ul E Lloc(ET). As for the frrst integral. let
oE (O,p - 1) be so small that (0 + 1)(p - 1) < 1. Then
<~IIDu P-~-Q II P- 1
- 1 p,K2px(s,t)
(rJJ(u(O:+l)(P-l)X[Ul >
1 •
)P
1 1dxdr
SK2p
---+ 0 as e ..... O.
We examine the lim-sup as e ..... 0 of the first integral on the right-hand side of
(6.11). The numbers k E R + , n E N being fixed. if e is small enough. we have the
inclusion
We write
6. An auxiliary proposition 357
t
K~l) -+ f f AoIDw~),nl(w~),nt(w~),m)b(P-lID(ldxdT.
BRN
The operation Dw~) coincides with the weak derivative of w~) only on those sets
Ai where w~) is bounded by a positive constant i, i.e.,
Dw~)x (At) == Dw~),t.
Since Dw~) is not well defined a.e. in the whole strip ET we estimate K~2) as
follows:
t
b t
+ 'Y
(m+l)
Up Jff a
J IDU2,kI P- 1 UIX[UI > n + U2,k]X(Qe) dxdT.
aK2p
358 XII. Non-negative solutions in Er. The case I<p<2
t
=JJIDUIIP-IU~ (Q+l~p-l) u~Q+1)('P-ll+"P X[UI > njx(ge)dxdr
aK2p
c.! !
~ 'Y (jjIDu;- :-° 1, /hd) •rjj .\0+1 )(,-1 )+«, xl'l > n]dzdT Y
BK2p ') ~BK2P ')
Analogously
t
c.!
x (jj,:'u!,o+I)(,-I)xl'l
BK2p
> n + ....]dzdT I·
)
~ (jfiDu~IPdxdr) ~
'Y
BK2p
6. An auxiliary proposition 359
We conclude that
provided 0: and a> 0 are chosen so that (6.14) holds. Combining these estimates
and limiting processes as parts of (6.4) we obtain
a+ b1 + 1 J(w(k),m+ )a+b+l
(P dx
RNx{t}
t
+ a(p - 1) JJ AoIDw~),~J (w~),n) a-I (w~),m)" (Pdxdr
BRN
where C is the constant appearing in the last integral on the right hand side of
(6.16). This integral is estimated as follows:
360 XII. Non-negative solutions in ~T. The case I <p<2
t
sRN
t
SRN
$
a(p-l)
2
It/ AoIDw~),nl
2(w~),n )4-1 (w~),m )b (Pdxdr
sRN
+ 4PC P
aP-1(p _ l)p(crp)p
1/( t
w+
(k),n
)P-l+4 (w+ (k),m
)b dxdr
•
SRN
We carry this estimate in (6.16), move the integral involving IDw~),nI2 on the
left hand side and discard the resulting non-negative tenn to obtain
(6.17)
-y(p)
+ (crp)P It I( w~),n )P-l+4 (+ w(k),m
)b dxdr
s K(l+")p
(6.18) 1 jW+(W+)4+b(Pdx.
a+b+l n
RNx{s}
Now letting 8-0, the integral in (6.18) tends to zero since it can be majorised by
as B - 0.
6. An auxiliary proposition 361
(6.20) p - 1 + a E (0,1),
This inequality holds true "1m E N, Vb ~ 0, "10- E (0,1), Vp > O. The positive
number a is fixed, satisfying the restrictions (6.14) and (6.20). The sequence {w~}
increases to w+ a.e. in ET . Therefore as m -+ 00, we may pass to the limit under
the integrals in (6.21) for those b ~ 0 for which
(7.1)
Pn = (~
~
2-i) K= K
p, n P.. , q
n -- 2-(n+1) ,
,=0
(7.2)
By the interpolation Lemma 4.3 of Chap. I we conclude that for every q E [1,00)
there exists a constant -y=-y(N,p, q), independent of p, such that for all tE (0, T)
(7.3)
N(P - 2) + pq > 0
q
and then, such a q being fixed. we let p-oo in (7.3).
(8.1)n
I Un
Un
E C (O,Tj
( )_
x,O -
L10c(RN»)nV
/:rUn - div IDun l
U on=
'0
p - 2 DUn
_ {min{uojn}
(0, Tj W,!;:(RN»)
= 0, in ET
for Ixl < n
II
for x ~n.
The initial data are bounded and compactly supported in RN. Therefore the unique
solvability of (8.1)n can be established as indicated in §12 of Chap. VI. Since the
initial data {uo,n}f1S\l form an increasing sequence of functions in Lloc<RN) we
have
(8.2) 't/p>O.
The solution of (1.1 )-( 1.2) will be constructed as the limit of the sequence {u n }f1S\l
in a suitable topology. For this we establish flJ'St some basic compactness of
{un}f1S\l.
LEMMA 8.1. There exists a constant "'( = "'( N, p) independent of n such that for
all t,p>O
MoreoverforallaE(O,p-l),
PROOF: The Lloc-estimate follows from (4.1) with s = 0, and the gradient es-
timate (8.4) is a consequence of (4.2) with s = O. Finally (8.5) is the content of
Lemma 3.2.
< 'Vk
_ I
_(l-<O+lHP-l»)
p P
No2=.!
p
(t )*{/
-
pP u 0 dx +
(- )~}P_Q~P_l)
t
p>'
K3p
u~2 == {:
ifO<u~k
if k < u < Ck
Ck if u? Ck
(k»)
( In U~k ((x),
-j/ (i
OKb
:T
k
In min{~jCk} de)
+
((x)dxdT == G~l) + G~2).
Let 0 be any positive number satisfying
G{l)
k ~-
2')' jjlD IP-l -
t
U U
(o+l)(p-l)
pup
(o+l)(p-l)
xu>kdxdT
[ j
p
OK2p
')'(a,p) jilDu
=--
t
PUP I
.-1-0 P- 1 (0+1)(,-1) [
XU> kjdxdT
p
OK2p
(0+1)(,-1)
X{ sup
O<T'<t
jU(X, T)dx} P
K2p
1-(0+1)(,-1)
X { sup
O<T'<t
K2p
J X[U > kj dx} P
~')'(a,p)(~);{Juodx+(:~)~} P
K4p
1-(o+l)(p-l)
SUP
{ O<T'<t
jX[U> kj dx} P
K2p
366 XII. Non-negative solutions in E1. The case 1<p<2
1berefore
LEMMA 10.1. Let a E (O,p - 1). There exists a constant "{ = ,,{(N,p, a) such
that
VO<s<t~T, VB ~ a + 1, Vn = 1,2, ... ,
,,{p-aN
+ -s -
{f ( )~}l-a
uodx+ ~
P
t
K2p
QoEKpx(s,t), Ql EBtpX(i,t),
= -t I I !IDuIP(u + 1)-8(2dxdr
Ql
f
+ (J IIDuIP(u + 1)-8-1Ut(2dxdT
Ql
~ ~ (p - 1) IIIDuIP{u + 1)-8-1Ut(2dxdr
Ql
+ ~ IIIDuIP(U + 1)-8((Tdxdr
Ql
By Young's inequality
~ ffiDuIPu-(O+1)(u + 1)-[6-(o+1)ldxdr
Ql
and the lemma follows by fonnal calculations. The calculations are fonnal since
Un,t (un + 1) -6 (2. need not be an admissible testing functions in (8.1)n. The ar-
guments would be rigorous if
(10.3)
Indeed. if so. we may take in the weak fonnulation (8.1)n the testing function
Un(t + h) - un(t) ( h (0 T _ )
h un + 1)-6/"2
.. , E, Is, Is~t<T-h.
where B j is the ball of radius j about the origin and {uo,n,; };:n+1' is a sequence
offunctions in C':' (B n +1). such that .
and
10. Compactness in the t ¥ariable 369
In the remarks below we drop the subscript n, j and write v = vn ,;. We write
(10.4) for the time levels t+h and t and set
By difference
where
Jh =IDv(t + hW- Dv(t + h) -IDv(t)IP- Dv(t).
2 2
In the weak formulation of (10.7) take the testing function W (t- V+ which van-
ishes on Ixl =j and for t::; ~. This gives
T-h T-h t+h
(10.8) / /(t -
! Bj
i) + Ao,;IDwI dxdr ::; 'Y /
2 /1 f
0 Bj
:r
t
vex, r)drr dxdr,
where
1
The last integral is finite by virtue of (10.5) and the lemma follows.
1berefore
The fust integral on the right hand side is estimated by (10.2). i.e .•
t-s{f
s-
$ 'Y- uodx + (t),!p}
p>' •
K4p
(12.2)
p-l-..
By Lemma 8.1 the sequence {'Un J> } is equibounded in
'Un
p-l-..
J> -+ U
p-l-..
J> weakly in LP(O, Ti W1,P(K p », 'rip> 0.
This implies that the sequences
Un,A: = Un 1\ k = min{un,k}
are equibounded in £P (0, Ti W1,P(Kp )) , 'rip> 0. and
(12.3) Un,k -+ U1\ k weakly in LP (0, T; W1,P(Kp ») , 'rip> 0, 'rIk > O.
to obtain
1_ ~ riD
II I < fJ Un 11'-1 Un-(01+1)7 Un(a+1)7(Uk - Un,k )cp .l-dT «=
ET
1
(flu
.1
--+ 0 as n-oo.
10 = IIIDun,kIP-1IDUkl cp dxdT
ET
~ p; 1 IllDun,klPCPdxdT + ~ IIIDUkIPcpdxdT.
ET ET
Combining these calculations in (12.4) gives
2-.} nEN
{ ata (Un + 1)2"
is equibounded in L~oAET) for all (J~ 0: + 1 and for all o:E (O,p - 1). Therefore
a (Un + 1)2"
at 2-' -+
a (u + 1)"'-
at 2-' weakly in L
2{ s, t; L 2(Kp) ) ,
for all 0 < s < t :5 T and all p> O. This implies that
(12.5)
o
Choose 1/J EXloc (ET) and in (S.I)n consider the testing function r.p= (1/J - u)+.
Fix O<s<t:5T and let
Then
(1/J - u)+ = (1/J - U" k)+ o
EXloc (ET),
so that r.p is an admissible testing function. It gives
t
Analogously,
374 XII. Non-negative solutions in l::r. The case 1<p<2
Therefore taking into account Lemma 12.1 and letting n-+oo in (12.6) gives
o
for all '1/1 EXloc (ET)' It remains to prove that u takes the initial datum U o in the
sense of Lloc(RN) and that ueS*.
Such a family can be constructed by first defining a function that coincides with
U o in K 3p and zero otherwise and then by mollifying the function so obtained.
Let also Ue be the unique solution of (1.1) with initial datum uo,e' We take the
difference of (8.1)n and the equation satisfied by Ue. In the p.d.e. so obtained take
the testing function
tp = [(un - u e)+ + 6]",(
where (1,6 e (0,1) and x-((x) is the usual cutoff function in K 2p that equals one
on Kp. We perfonn an integration by parts and let 6-+0, 8-0, (1-0. to obtain
We use (8.4), interchange the role of Un and Ue and, for t > 0 fixed, let n -+ 00.
This gives
12. The limiting process 375
From this
Letting t '\. 0
12-(ii). uES·
-
o(U I\k } < -1- -Un.
at n - 2-p t
As n-+oo
1 u
(12.7) (u 1\ k}t ~ -- - a.e. in ET.
2-pt
The limit is flISt taken in 1)'(0, T) and then (12.7) holds almost everywhere in ET
in view of (12.5). Next from Lemma 9.1 it follows that 'VC> 1
t
Here we have used the fact that Un / u implies [un < Ckl ~ Iu < Ckl. Letting
n -+ 00 for k > 0 and C> 1 fixed yields by lower semicontinuity
t
jjIDuIP~X[k<U<CkldxdT = 0 (~).
sKp
We show next that if(l3.1) is violated, then initial data in Ltoc(RN) might produce
unbounded solutions.
13-(i). A counterexample
Let a E (0, 1) be a positive constant and let Be denote the ball of radius a in
RN centered at the origin. Consider the functions
(a 2-lxI2)2
(13.2) z= + and v = (1 - ht)+ z,
Ixl N lin Ix121f3
where {j, h > 1 are to be chosen. One verifies that
where
2/3 41X12}
F = { N + In Ixl2 + a2 _ Ix l2 .
We choose a =e -k and k> 1 so large that F> O. Compute .
_ zp-IFp-1
IDzlP 2Dz =- x
. Ixl p
zp-2 FP-l zp-l FP-l
div(lDzlp-2 Dz) = -(p - 1) Ixl p Dz . x + P Ixl1>+1 Dlxl· X
Zp-l FP-l zp-l FP-2
-N Ixl p -(P-I) Ixl p DF·x.
We calculate the expression in braces on the right-hand side using the definition
of F and the fact that N(p - 2) + p=O, to obtain
1t >
-
8(2 _!!.)
k
_Nf3.
2k
378 XII. Non-negative solutions in I:T. The case I <p<2
zp-2 'Y(N,P)]
.c(v) ~ z [-h - Ixl p In Ixl2 .
By calculation on £~2) •
II{
1:1
vtCP+IDvI P- 2Dv.Dcp} dxdr
= lim
~\,O
jrJ{{vtcp+IDvlp-2 Dv.Dcp} dxdr
Q.
I
~ ~~ II
o{lzl=G-~}
IDvl p- 2Dv· 1:1 tpdtrdr
I
- lim
~ .... "'O
J"J{IDvIP-2 Dv . -ixix tpdtrdr,
I
O{lzl=~}
14. Bibliographical notes 319
where du denotes the surface measure on {Ixl =e} and on {Ixl =e}. The limits
on the right hand side are zero. In particular we have
One also verifies by direct calculation that v satisfies (5.1) and (5.2) and therefore
is a subsolution of (13.3) in the class S·.
Next we return to (13.3). This problem has a unique solution U E S·, by
the construction of §§8-12 and the uniqueness theorem 7.1. By the comparison
principle U ~ v and therefore U is not bounded. The comparison principle here
is applied as follows. By the definition of weak solution the truncated functions
Uk == mint Ui k} are, for all k > 0 distributional subsolutions of (13.3). Setting
w == v - U and W(k) == v - Uk
and using (13.5) we find
j (w+(t»)q dx ~ _'Y_
(up)p
11
t
(w+)P-2+ q dxdr.
Kp OK(1+a)p
were fmt investigated by Btizis and Friedman [18J. The notion of weak solution
introduced in §2 is taken from [42J. B6nilan has infonned us of a more general no-
tion of solution. introduced in [II J. that would include solutions of variable sign.
The remainder of the chapter is essentially taken from [42J. It would be of interest
to investigate questions of existence/Uniqueness for (1.1) in ET when the initial
datum is of variable sign or is a measure. Singular equations are little understood.
mostly if p violates (13.1). Preliminary investigations seem to indicate questions
of limiting Sobolev exponent (see [19]) and differential geometry.
Bibliography
[29J Y.Z. Chen and E. DiBenedetto, On the Harnack inequality for non-negative
solutions of singular parabolic equations, Proc. of Conf. Non-linear diffu-
sion, in Honour of J. Serrin, Minneapolis May 1990.
[30J H. Choe, HOlder regularity for the gradient of solutions of certain singu-
lar parabolic equations, Comm. Part. Diff. Equations 16 II 11 (1991), pp.
1709-1732.
[31] H. Choe, HOlder continuity of solutions of certain degenerate parabolic sys-
tems, Non-linear Anal. 8 II 3 (1992), pp. 235-243.
[32] O. DaPrato, Spazi C(p,9)(fl, 6) e loro proprietA, Ann. Math. Pura Appl. 69
(1965), pp. 383-392.
[33] E. DeGiorgi, Sulla differenziabilita' e l'analiticita' delle estremali degli in-
tegrali multipli regolari, Mem. Acc. Sci. Torino, Cl. Sc. Fis. Mat. Nat. (3) 3
(1957), pp. 25-43.
[34] E. DiBenedetto, Continuity of weak solutions to certain singular parabolic
equations, Ann.Mat. Puta Appl. 4 II 130 (1982), pp.131-176.
[35] E. DiBenedetto, Continuity of weak solutions to a general porous medium
equation, Indiana Univ. Math. J. 32 II 1 (1983), pp. 83-118.
[36] E. DiBenedetto and A. Friedman, Regularity of solutions of non-linear de-
generate parabolic systems, J. Reine Angew. Math. 349 (1984), pp. 83-128.
[37] E. DiBenedetto and A. Friedman, HOlder estimates for non-linear degener-
ate parabolic systems, J. Reine Angew. Math. 357 (1985), pp. 1-22.
[38] E. DiBenedetto, A boundary modulus of continuity for a class of singular
parabolic equations, J. Diff. Equations 6 II 3 (1986), pp. 418-447.
[39] E. DiBenedetto, On the local behaviour of solutions of degenerate parabolic
equations with measurable coefficients, Ann. Sc. Norm. Sup. Pisa Cl. Sc.
Serie IV, XIII 3 (1986),487-535.
[40J E. DiBenedetto, Intrinsic Harnack type inequalities for solutions of certain
degenerate parabolic equations, Arch. Rational Mech. Anal. 100 II 2 (1988),
pp.129-147.
[41] E. DiBenedetto and M.A. Herrero, On the Cauchy problem and initial traces
for a degenerate parabolic equation Trans. AMS 314 (1989), pp. 187 - 224.
[42] E. DiBenedetto and M.A. Herrero, Non negative solutions of the evolution
p-Laplacian equation. Initial traces and Cauchy problem when 1 < p < 2,
Arch. Rational Mech. Anal. 111 II 3 (1990), pp. 225 - 290.
[43] E. DiBenedetto, Y. C. Kwong and V. Vespri, Local space analiticity of so-
lutions of certain singular parabolic equations, Indiana Univ. Math. J. 40 II
2 (1991), pp. 741-765.
[44] E. DiBenedetto and Y.C. Kwong, Intrinsic Harnack estimates and extinction
profile for certain singUlar parabolic equar.i6ns, Trans AMS 330 II 2 (1992),
pp. 783-811.
384 Bibliography
[75] L.K. Martinson and K.B. Paplov, Unsteady shear flows of a conducting
fluid with a rheological power law, Magnit. Gidrodinarnika # 2 (1970), pp.
50-58.
[76] V.G. Mazja, Sobolev Spaces. Springer-Verlag, New York, (1985).
[77] M. Meier, Boundedness and integrability properties of weak solutions of
quasilinearelliptic systems. J. Reine Angew. Math. 333 (1982), pp. 191-220.
[78] G. Minty, Monotone (non-linear) operators in Hilbert spaces, Duke Math.
J. 29 (1962), pp. 341-346.
[791 C.B. Morrey, Multiple integrals in the calculus of variations. Springer-
Verlag, New York, (1966).
[80] C.B. Morrey, Partial regularity results for non-linear elliptic systems. J.
Math. Mech. 17 (1968), pp. 649-670.
[81] J. Moser, A new proof of DeGiorgi's theorem concerning the regularity
problem for elliptic differential equations, Comm. Pure Appl. Math. 13
(1960), pp. 457 -468.
[82] J. Moser, On Harnack's theorem for elliptic differential equations, Comm.
Pure Appl. Math. 14 (1961), pp. 577-591.
[83] J. Moser, A Harnack inequality for parabolic differential equations, Comm.
Pure Appl. Math. 17 (1964), pp. 101-134.
[841 J. Nash, Continuity of solutions of parabolic and elliptic equations, Amer.
J. Math. 80 (1958), pp. 931-954.
[85] M. Pierre, Uniqueness of the solutions of Ut - Llt/>( u) = 0 with initial datum
a measure, Nonlin. Anal. 6, # 2 (1987), pp. 175-187.
[86] B. Pini, Sulla soluzione generalizzata di Wiener per il primo problema di
valori al contorno nel caso parabolico, Rend. Sem. Math. Univ. Padova 23
(1954), pp. 422-434.
[87] M. Porzio, A priori bounds for weak solutions of certain degenerate parabolic
equations, Nonlin. Anal. 20, # II (1991), pp. 1093, 1107.
[881 M. Porzio and V. Vespri, HOlder estimates for local solutions of some doubly
non-linear degenerate parabolic equations, J. Diff. Equ. (to appear).
[89] Y.G. Resetniak, Mappings with bounded distorsion in space. 'Nauka' Novosi-
birski, Moscow (1982) (Russian).
[901 P.E. Sacks, Continuity of solutions of a singular parabolic equation, Non-
linear Anal. 7 (1983), pp. 387-409.
[91] M.V. Safonov, The Harnack inequality for elliptic equations and the Holder
continuity of their solutions, Boundary problems of mathematical physics
and adjacent questions in the theory offunctions, Zap. Nauchn. Sem. Leningrad
Otdel. Math. Inst. Steklov (LOMI) 96 (1980), pp. 272-287 (Engl. transl.:
J. Soviet Math. (LOMI) 20 (1983), pp. 851-863).
[92] J. Serrin, Local behaviour of solutions of quasilinear elliptic equations, Acta
Math. III (1964), pp. 101-134.
Bibliography 387